2005-04-16 16:20:36 -06:00
|
|
|
/*
|
|
|
|
* kernel/cpuset.c
|
|
|
|
*
|
|
|
|
* Processor and Memory placement constraints for sets of tasks.
|
|
|
|
*
|
|
|
|
* Copyright (C) 2003 BULL SA.
|
[PATCH] cpuset memory spread basic implementation
This patch provides the implementation and cpuset interface for an alternative
memory allocation policy that can be applied to certain kinds of memory
allocations, such as the page cache (file system buffers) and some slab caches
(such as inode caches).
The policy is called "memory spreading." If enabled, it spreads out these
kinds of memory allocations over all the nodes allowed to a task, instead of
preferring to place them on the node where the task is executing.
All other kinds of allocations, including anonymous pages for a tasks stack
and data regions, are not affected by this policy choice, and continue to be
allocated preferring the node local to execution, as modified by the NUMA
mempolicy.
There are two boolean flag files per cpuset that control where the kernel
allocates pages for the file system buffers and related in kernel data
structures. They are called 'memory_spread_page' and 'memory_spread_slab'.
If the per-cpuset boolean flag file 'memory_spread_page' is set, then the
kernel will spread the file system buffers (page cache) evenly over all the
nodes that the faulting task is allowed to use, instead of preferring to put
those pages on the node where the task is running.
If the per-cpuset boolean flag file 'memory_spread_slab' is set, then the
kernel will spread some file system related slab caches, such as for inodes
and dentries evenly over all the nodes that the faulting task is allowed to
use, instead of preferring to put those pages on the node where the task is
running.
The implementation is simple. Setting the cpuset flags 'memory_spread_page'
or 'memory_spread_cache' turns on the per-process flags PF_SPREAD_PAGE or
PF_SPREAD_SLAB, respectively, for each task that is in the cpuset or
subsequently joins that cpuset. In subsequent patches, the page allocation
calls for the affected page cache and slab caches are modified to perform an
inline check for these flags, and if set, a call to a new routine
cpuset_mem_spread_node() returns the node to prefer for the allocation.
The cpuset_mem_spread_node() routine is also simple. It uses the value of a
per-task rotor cpuset_mem_spread_rotor to select the next node in the current
tasks mems_allowed to prefer for the allocation.
This policy can provide substantial improvements for jobs that need to place
thread local data on the corresponding node, but that need to access large
file system data sets that need to be spread across the several nodes in the
jobs cpuset in order to fit. Without this patch, especially for jobs that
might have one thread reading in the data set, the memory allocation across
the nodes in the jobs cpuset can become very uneven.
A couple of Copyright year ranges are updated as well. And a couple of email
addresses that can be found in the MAINTAINERS file are removed.
Signed-off-by: Paul Jackson <pj@sgi.com>
Signed-off-by: Andrew Morton <akpm@osdl.org>
Signed-off-by: Linus Torvalds <torvalds@osdl.org>
2006-03-24 04:16:03 -07:00
|
|
|
* Copyright (C) 2004-2006 Silicon Graphics, Inc.
|
2005-04-16 16:20:36 -06:00
|
|
|
*
|
|
|
|
* Portions derived from Patrick Mochel's sysfs code.
|
|
|
|
* sysfs is Copyright (c) 2001-3 Patrick Mochel
|
|
|
|
*
|
[PATCH] cpuset memory spread basic implementation
This patch provides the implementation and cpuset interface for an alternative
memory allocation policy that can be applied to certain kinds of memory
allocations, such as the page cache (file system buffers) and some slab caches
(such as inode caches).
The policy is called "memory spreading." If enabled, it spreads out these
kinds of memory allocations over all the nodes allowed to a task, instead of
preferring to place them on the node where the task is executing.
All other kinds of allocations, including anonymous pages for a tasks stack
and data regions, are not affected by this policy choice, and continue to be
allocated preferring the node local to execution, as modified by the NUMA
mempolicy.
There are two boolean flag files per cpuset that control where the kernel
allocates pages for the file system buffers and related in kernel data
structures. They are called 'memory_spread_page' and 'memory_spread_slab'.
If the per-cpuset boolean flag file 'memory_spread_page' is set, then the
kernel will spread the file system buffers (page cache) evenly over all the
nodes that the faulting task is allowed to use, instead of preferring to put
those pages on the node where the task is running.
If the per-cpuset boolean flag file 'memory_spread_slab' is set, then the
kernel will spread some file system related slab caches, such as for inodes
and dentries evenly over all the nodes that the faulting task is allowed to
use, instead of preferring to put those pages on the node where the task is
running.
The implementation is simple. Setting the cpuset flags 'memory_spread_page'
or 'memory_spread_cache' turns on the per-process flags PF_SPREAD_PAGE or
PF_SPREAD_SLAB, respectively, for each task that is in the cpuset or
subsequently joins that cpuset. In subsequent patches, the page allocation
calls for the affected page cache and slab caches are modified to perform an
inline check for these flags, and if set, a call to a new routine
cpuset_mem_spread_node() returns the node to prefer for the allocation.
The cpuset_mem_spread_node() routine is also simple. It uses the value of a
per-task rotor cpuset_mem_spread_rotor to select the next node in the current
tasks mems_allowed to prefer for the allocation.
This policy can provide substantial improvements for jobs that need to place
thread local data on the corresponding node, but that need to access large
file system data sets that need to be spread across the several nodes in the
jobs cpuset in order to fit. Without this patch, especially for jobs that
might have one thread reading in the data set, the memory allocation across
the nodes in the jobs cpuset can become very uneven.
A couple of Copyright year ranges are updated as well. And a couple of email
addresses that can be found in the MAINTAINERS file are removed.
Signed-off-by: Paul Jackson <pj@sgi.com>
Signed-off-by: Andrew Morton <akpm@osdl.org>
Signed-off-by: Linus Torvalds <torvalds@osdl.org>
2006-03-24 04:16:03 -07:00
|
|
|
* 2003-10-10 Written by Simon Derr.
|
2005-04-16 16:20:36 -06:00
|
|
|
* 2003-10-22 Updates by Stephen Hemminger.
|
[PATCH] cpuset memory spread basic implementation
This patch provides the implementation and cpuset interface for an alternative
memory allocation policy that can be applied to certain kinds of memory
allocations, such as the page cache (file system buffers) and some slab caches
(such as inode caches).
The policy is called "memory spreading." If enabled, it spreads out these
kinds of memory allocations over all the nodes allowed to a task, instead of
preferring to place them on the node where the task is executing.
All other kinds of allocations, including anonymous pages for a tasks stack
and data regions, are not affected by this policy choice, and continue to be
allocated preferring the node local to execution, as modified by the NUMA
mempolicy.
There are two boolean flag files per cpuset that control where the kernel
allocates pages for the file system buffers and related in kernel data
structures. They are called 'memory_spread_page' and 'memory_spread_slab'.
If the per-cpuset boolean flag file 'memory_spread_page' is set, then the
kernel will spread the file system buffers (page cache) evenly over all the
nodes that the faulting task is allowed to use, instead of preferring to put
those pages on the node where the task is running.
If the per-cpuset boolean flag file 'memory_spread_slab' is set, then the
kernel will spread some file system related slab caches, such as for inodes
and dentries evenly over all the nodes that the faulting task is allowed to
use, instead of preferring to put those pages on the node where the task is
running.
The implementation is simple. Setting the cpuset flags 'memory_spread_page'
or 'memory_spread_cache' turns on the per-process flags PF_SPREAD_PAGE or
PF_SPREAD_SLAB, respectively, for each task that is in the cpuset or
subsequently joins that cpuset. In subsequent patches, the page allocation
calls for the affected page cache and slab caches are modified to perform an
inline check for these flags, and if set, a call to a new routine
cpuset_mem_spread_node() returns the node to prefer for the allocation.
The cpuset_mem_spread_node() routine is also simple. It uses the value of a
per-task rotor cpuset_mem_spread_rotor to select the next node in the current
tasks mems_allowed to prefer for the allocation.
This policy can provide substantial improvements for jobs that need to place
thread local data on the corresponding node, but that need to access large
file system data sets that need to be spread across the several nodes in the
jobs cpuset in order to fit. Without this patch, especially for jobs that
might have one thread reading in the data set, the memory allocation across
the nodes in the jobs cpuset can become very uneven.
A couple of Copyright year ranges are updated as well. And a couple of email
addresses that can be found in the MAINTAINERS file are removed.
Signed-off-by: Paul Jackson <pj@sgi.com>
Signed-off-by: Andrew Morton <akpm@osdl.org>
Signed-off-by: Linus Torvalds <torvalds@osdl.org>
2006-03-24 04:16:03 -07:00
|
|
|
* 2004 May-July Rework by Paul Jackson.
|
2005-04-16 16:20:36 -06:00
|
|
|
*
|
|
|
|
* This file is subject to the terms and conditions of the GNU General Public
|
|
|
|
* License. See the file COPYING in the main directory of the Linux
|
|
|
|
* distribution for more details.
|
|
|
|
*/
|
|
|
|
|
|
|
|
#include <linux/cpu.h>
|
|
|
|
#include <linux/cpumask.h>
|
|
|
|
#include <linux/cpuset.h>
|
|
|
|
#include <linux/err.h>
|
|
|
|
#include <linux/errno.h>
|
|
|
|
#include <linux/file.h>
|
|
|
|
#include <linux/fs.h>
|
|
|
|
#include <linux/init.h>
|
|
|
|
#include <linux/interrupt.h>
|
|
|
|
#include <linux/kernel.h>
|
|
|
|
#include <linux/kmod.h>
|
|
|
|
#include <linux/list.h>
|
[PATCH] cpusets: automatic numa mempolicy rebinding
This patch automatically updates a tasks NUMA mempolicy when its cpuset
memory placement changes. It does so within the context of the task,
without any need to support low level external mempolicy manipulation.
If a system is not using cpusets, or if running on a system with just the
root (all-encompassing) cpuset, then this remap is a no-op. Only when a
task is moved between cpusets, or a cpusets memory placement is changed
does the following apply. Otherwise, the main routine below,
rebind_policy() is not even called.
When mixing cpusets, scheduler affinity, and NUMA mempolicies, the
essential role of cpusets is to place jobs (several related tasks) on a set
of CPUs and Memory Nodes, the essential role of sched_setaffinity is to
manage a jobs processor placement within its allowed cpuset, and the
essential role of NUMA mempolicy (mbind, set_mempolicy) is to manage a jobs
memory placement within its allowed cpuset.
However, CPU affinity and NUMA memory placement are managed within the
kernel using absolute system wide numbering, not cpuset relative numbering.
This is ok until a job is migrated to a different cpuset, or what's the
same, a jobs cpuset is moved to different CPUs and Memory Nodes.
Then the CPU affinity and NUMA memory placement of the tasks in the job
need to be updated, to preserve their cpuset-relative position. This can
be done for CPU affinity using sched_setaffinity() from user code, as one
task can modify anothers CPU affinity. This cannot be done from an
external task for NUMA memory placement, as that can only be modified in
the context of the task using it.
However, it easy enough to remap a tasks NUMA mempolicy automatically when
a task is migrated, using the existing cpuset mechanism to trigger a
refresh of a tasks memory placement after its cpuset has changed. All that
is needed is the old and new nodemask, and notice to the task that it needs
to rebind its mempolicy. The tasks mems_allowed has the old mask, the
tasks cpuset has the new mask, and the existing
cpuset_update_current_mems_allowed() mechanism provides the notice. The
bitmap/cpumask/nodemask remap operators provide the cpuset relative
calculations.
This patch leaves open a couple of issues:
1) Updating vma and shmfs/tmpfs/hugetlbfs memory policies:
These mempolicies may reference nodes outside of those allowed to
the current task by its cpuset. Tasks are migrated as part of jobs,
which reside on what might be several cpusets in a subtree. When such
a job is migrated, all NUMA memory policy references to nodes within
that cpuset subtree should be translated, and references to any nodes
outside that subtree should be left untouched. A future patch will
provide the cpuset mechanism needed to mark such subtrees. With that
patch, we will be able to correctly migrate these other memory policies
across a job migration.
2) Updating cpuset, affinity and memory policies in user space:
This is harder. Any placement state stored in user space using
system-wide numbering will be invalidated across a migration. More
work will be required to provide user code with a migration-safe means
to manage its cpuset relative placement, while preserving the current
API's that pass system wide numbers, not cpuset relative numbers across
the kernel-user boundary.
Signed-off-by: Paul Jackson <pj@sgi.com>
Signed-off-by: Andrew Morton <akpm@osdl.org>
Signed-off-by: Linus Torvalds <torvalds@osdl.org>
2005-10-30 16:02:36 -07:00
|
|
|
#include <linux/mempolicy.h>
|
2005-04-16 16:20:36 -06:00
|
|
|
#include <linux/mm.h>
|
|
|
|
#include <linux/module.h>
|
|
|
|
#include <linux/mount.h>
|
|
|
|
#include <linux/namei.h>
|
|
|
|
#include <linux/pagemap.h>
|
|
|
|
#include <linux/proc_fs.h>
|
[PATCH] cpuset: use rcu directly optimization
Optimize the cpuset impact on page allocation, the most performance critical
cpuset hook in the kernel.
On each page allocation, the cpuset hook needs to check for a possible change
in the current tasks cpuset. It can now handle the common case, of no change,
without taking any spinlock or semaphore, thanks to RCU.
Convert a spinlock on the current task to an rcu_read_lock(), saving
approximately a memory barrier and an atomic op, depending on architecture.
This is done by adding rcu_assign_pointer() and synchronize_rcu() calls to the
write side of the task->cpuset pointer, in cpuset.c:attach_task(), to delay
freeing up a detached cpuset until after any critical sections referencing
that pointer.
Thanks to Andi Kleen, Nick Piggin and Eric Dumazet for ideas.
Signed-off-by: Paul Jackson <pj@sgi.com>
Signed-off-by: Andrew Morton <akpm@osdl.org>
Signed-off-by: Linus Torvalds <torvalds@osdl.org>
2006-01-08 02:02:02 -07:00
|
|
|
#include <linux/rcupdate.h>
|
2005-04-16 16:20:36 -06:00
|
|
|
#include <linux/sched.h>
|
|
|
|
#include <linux/seq_file.h>
|
2006-06-23 03:04:00 -06:00
|
|
|
#include <linux/security.h>
|
2005-04-16 16:20:36 -06:00
|
|
|
#include <linux/slab.h>
|
|
|
|
#include <linux/smp_lock.h>
|
|
|
|
#include <linux/spinlock.h>
|
|
|
|
#include <linux/stat.h>
|
|
|
|
#include <linux/string.h>
|
|
|
|
#include <linux/time.h>
|
|
|
|
#include <linux/backing-dev.h>
|
|
|
|
#include <linux/sort.h>
|
|
|
|
|
|
|
|
#include <asm/uaccess.h>
|
|
|
|
#include <asm/atomic.h>
|
2006-03-23 04:00:18 -07:00
|
|
|
#include <linux/mutex.h>
|
2005-04-16 16:20:36 -06:00
|
|
|
|
2006-01-08 02:01:51 -07:00
|
|
|
#define CPUSET_SUPER_MAGIC 0x27e0eb
|
2005-04-16 16:20:36 -06:00
|
|
|
|
2006-01-08 02:01:57 -07:00
|
|
|
/*
|
|
|
|
* Tracks how many cpusets are currently defined in system.
|
|
|
|
* When there is only one cpuset (the root cpuset) we can
|
|
|
|
* short circuit some hooks.
|
|
|
|
*/
|
2006-01-08 02:02:03 -07:00
|
|
|
int number_of_cpusets __read_mostly;
|
2006-01-08 02:01:57 -07:00
|
|
|
|
[PATCH] cpuset: memory pressure meter
Provide a simple per-cpuset metric of memory pressure, tracking the -rate-
that the tasks in a cpuset call try_to_free_pages(), the synchronous
(direct) memory reclaim code.
This enables batch managers monitoring jobs running in dedicated cpusets to
efficiently detect what level of memory pressure that job is causing.
This is useful both on tightly managed systems running a wide mix of
submitted jobs, which may choose to terminate or reprioritize jobs that are
trying to use more memory than allowed on the nodes assigned them, and with
tightly coupled, long running, massively parallel scientific computing jobs
that will dramatically fail to meet required performance goals if they
start to use more memory than allowed to them.
This patch just provides a very economical way for the batch manager to
monitor a cpuset for signs of memory pressure. It's up to the batch
manager or other user code to decide what to do about it and take action.
==> Unless this feature is enabled by writing "1" to the special file
/dev/cpuset/memory_pressure_enabled, the hook in the rebalance
code of __alloc_pages() for this metric reduces to simply noticing
that the cpuset_memory_pressure_enabled flag is zero. So only
systems that enable this feature will compute the metric.
Why a per-cpuset, running average:
Because this meter is per-cpuset, rather than per-task or mm, the
system load imposed by a batch scheduler monitoring this metric is
sharply reduced on large systems, because a scan of the tasklist can be
avoided on each set of queries.
Because this meter is a running average, instead of an accumulating
counter, a batch scheduler can detect memory pressure with a single
read, instead of having to read and accumulate results for a period of
time.
Because this meter is per-cpuset rather than per-task or mm, the
batch scheduler can obtain the key information, memory pressure in a
cpuset, with a single read, rather than having to query and accumulate
results over all the (dynamically changing) set of tasks in the cpuset.
A per-cpuset simple digital filter (requires a spinlock and 3 words of data
per-cpuset) is kept, and updated by any task attached to that cpuset, if it
enters the synchronous (direct) page reclaim code.
A per-cpuset file provides an integer number representing the recent
(half-life of 10 seconds) rate of direct page reclaims caused by the tasks
in the cpuset, in units of reclaims attempted per second, times 1000.
Signed-off-by: Paul Jackson <pj@sgi.com>
Signed-off-by: Andrew Morton <akpm@osdl.org>
Signed-off-by: Linus Torvalds <torvalds@osdl.org>
2006-01-08 02:01:49 -07:00
|
|
|
/* See "Frequency meter" comments, below. */
|
|
|
|
|
|
|
|
struct fmeter {
|
|
|
|
int cnt; /* unprocessed events count */
|
|
|
|
int val; /* most recent output value */
|
|
|
|
time_t time; /* clock (secs) when val computed */
|
|
|
|
spinlock_t lock; /* guards read or write of above */
|
|
|
|
};
|
|
|
|
|
2005-04-16 16:20:36 -06:00
|
|
|
struct cpuset {
|
|
|
|
unsigned long flags; /* "unsigned long" so bitops work */
|
|
|
|
cpumask_t cpus_allowed; /* CPUs allowed to tasks in cpuset */
|
|
|
|
nodemask_t mems_allowed; /* Memory Nodes allowed to tasks */
|
|
|
|
|
[PATCH] cpusets: dual semaphore locking overhaul
Overhaul cpuset locking. Replace single semaphore with two semaphores.
The suggestion to use two locks was made by Roman Zippel.
Both locks are global. Code that wants to modify cpusets must first
acquire the exclusive manage_sem, which allows them read-only access to
cpusets, and holds off other would-be modifiers. Before making actual
changes, the second semaphore, callback_sem must be acquired as well. Code
that needs only to query cpusets must acquire callback_sem, which is also a
global exclusive lock.
The earlier problems with double tripping are avoided, because it is
allowed for holders of manage_sem to nest the second callback_sem lock, and
only callback_sem is needed by code called from within __alloc_pages(),
where the double tripping had been possible.
This is not quite the same as a normal read/write semaphore, because
obtaining read-only access with intent to change must hold off other such
attempts, while allowing read-only access w/o such intention. Changing
cpusets involves several related checks and changes, which must be done
while allowing read-only queries (to avoid the double trip), but while
ensuring nothing changes (holding off other would be modifiers.)
This overhaul of cpuset locking also makes careful use of task_lock() to
guard access to the task->cpuset pointer, closing a couple of race
conditions noticed while reading this code (thanks, Roman). I've never
seen these races fail in any use or test.
See further the comments in the code.
Signed-off-by: Paul Jackson <pj@sgi.com>
Signed-off-by: Andrew Morton <akpm@osdl.org>
Signed-off-by: Linus Torvalds <torvalds@osdl.org>
2005-10-30 16:02:30 -07:00
|
|
|
/*
|
|
|
|
* Count is atomic so can incr (fork) or decr (exit) without a lock.
|
|
|
|
*/
|
2005-04-16 16:20:36 -06:00
|
|
|
atomic_t count; /* count tasks using this cpuset */
|
|
|
|
|
|
|
|
/*
|
|
|
|
* We link our 'sibling' struct into our parents 'children'.
|
|
|
|
* Our children link their 'sibling' into our 'children'.
|
|
|
|
*/
|
|
|
|
struct list_head sibling; /* my parents children */
|
|
|
|
struct list_head children; /* my children */
|
|
|
|
|
|
|
|
struct cpuset *parent; /* my parent */
|
|
|
|
struct dentry *dentry; /* cpuset fs entry */
|
|
|
|
|
|
|
|
/*
|
|
|
|
* Copy of global cpuset_mems_generation as of the most
|
|
|
|
* recent time this cpuset changed its mems_allowed.
|
|
|
|
*/
|
[PATCH] cpuset: memory pressure meter
Provide a simple per-cpuset metric of memory pressure, tracking the -rate-
that the tasks in a cpuset call try_to_free_pages(), the synchronous
(direct) memory reclaim code.
This enables batch managers monitoring jobs running in dedicated cpusets to
efficiently detect what level of memory pressure that job is causing.
This is useful both on tightly managed systems running a wide mix of
submitted jobs, which may choose to terminate or reprioritize jobs that are
trying to use more memory than allowed on the nodes assigned them, and with
tightly coupled, long running, massively parallel scientific computing jobs
that will dramatically fail to meet required performance goals if they
start to use more memory than allowed to them.
This patch just provides a very economical way for the batch manager to
monitor a cpuset for signs of memory pressure. It's up to the batch
manager or other user code to decide what to do about it and take action.
==> Unless this feature is enabled by writing "1" to the special file
/dev/cpuset/memory_pressure_enabled, the hook in the rebalance
code of __alloc_pages() for this metric reduces to simply noticing
that the cpuset_memory_pressure_enabled flag is zero. So only
systems that enable this feature will compute the metric.
Why a per-cpuset, running average:
Because this meter is per-cpuset, rather than per-task or mm, the
system load imposed by a batch scheduler monitoring this metric is
sharply reduced on large systems, because a scan of the tasklist can be
avoided on each set of queries.
Because this meter is a running average, instead of an accumulating
counter, a batch scheduler can detect memory pressure with a single
read, instead of having to read and accumulate results for a period of
time.
Because this meter is per-cpuset rather than per-task or mm, the
batch scheduler can obtain the key information, memory pressure in a
cpuset, with a single read, rather than having to query and accumulate
results over all the (dynamically changing) set of tasks in the cpuset.
A per-cpuset simple digital filter (requires a spinlock and 3 words of data
per-cpuset) is kept, and updated by any task attached to that cpuset, if it
enters the synchronous (direct) page reclaim code.
A per-cpuset file provides an integer number representing the recent
(half-life of 10 seconds) rate of direct page reclaims caused by the tasks
in the cpuset, in units of reclaims attempted per second, times 1000.
Signed-off-by: Paul Jackson <pj@sgi.com>
Signed-off-by: Andrew Morton <akpm@osdl.org>
Signed-off-by: Linus Torvalds <torvalds@osdl.org>
2006-01-08 02:01:49 -07:00
|
|
|
int mems_generation;
|
|
|
|
|
|
|
|
struct fmeter fmeter; /* memory_pressure filter */
|
2005-04-16 16:20:36 -06:00
|
|
|
};
|
|
|
|
|
|
|
|
/* bits in struct cpuset flags field */
|
|
|
|
typedef enum {
|
|
|
|
CS_CPU_EXCLUSIVE,
|
|
|
|
CS_MEM_EXCLUSIVE,
|
[PATCH] cpusets: swap migration interface
Add a boolean "memory_migrate" to each cpuset, represented by a file
containing "0" or "1" in each directory below /dev/cpuset.
It defaults to false (file contains "0"). It can be set true by writing
"1" to the file.
If true, then anytime that a task is attached to the cpuset so marked, the
pages of that task will be moved to that cpuset, preserving, to the extent
practical, the cpuset-relative placement of the pages.
Also anytime that a cpuset so marked has its memory placement changed (by
writing to its "mems" file), the tasks in that cpuset will have their pages
moved to the cpusets new nodes, preserving, to the extent practical, the
cpuset-relative placement of the moved pages.
Signed-off-by: Paul Jackson <pj@sgi.com>
Cc: Christoph Lameter <christoph@lameter.com>
Signed-off-by: Andrew Morton <akpm@osdl.org>
Signed-off-by: Linus Torvalds <torvalds@osdl.org>
2006-01-08 02:00:56 -07:00
|
|
|
CS_MEMORY_MIGRATE,
|
2005-04-16 16:20:36 -06:00
|
|
|
CS_REMOVED,
|
[PATCH] cpuset memory spread basic implementation
This patch provides the implementation and cpuset interface for an alternative
memory allocation policy that can be applied to certain kinds of memory
allocations, such as the page cache (file system buffers) and some slab caches
(such as inode caches).
The policy is called "memory spreading." If enabled, it spreads out these
kinds of memory allocations over all the nodes allowed to a task, instead of
preferring to place them on the node where the task is executing.
All other kinds of allocations, including anonymous pages for a tasks stack
and data regions, are not affected by this policy choice, and continue to be
allocated preferring the node local to execution, as modified by the NUMA
mempolicy.
There are two boolean flag files per cpuset that control where the kernel
allocates pages for the file system buffers and related in kernel data
structures. They are called 'memory_spread_page' and 'memory_spread_slab'.
If the per-cpuset boolean flag file 'memory_spread_page' is set, then the
kernel will spread the file system buffers (page cache) evenly over all the
nodes that the faulting task is allowed to use, instead of preferring to put
those pages on the node where the task is running.
If the per-cpuset boolean flag file 'memory_spread_slab' is set, then the
kernel will spread some file system related slab caches, such as for inodes
and dentries evenly over all the nodes that the faulting task is allowed to
use, instead of preferring to put those pages on the node where the task is
running.
The implementation is simple. Setting the cpuset flags 'memory_spread_page'
or 'memory_spread_cache' turns on the per-process flags PF_SPREAD_PAGE or
PF_SPREAD_SLAB, respectively, for each task that is in the cpuset or
subsequently joins that cpuset. In subsequent patches, the page allocation
calls for the affected page cache and slab caches are modified to perform an
inline check for these flags, and if set, a call to a new routine
cpuset_mem_spread_node() returns the node to prefer for the allocation.
The cpuset_mem_spread_node() routine is also simple. It uses the value of a
per-task rotor cpuset_mem_spread_rotor to select the next node in the current
tasks mems_allowed to prefer for the allocation.
This policy can provide substantial improvements for jobs that need to place
thread local data on the corresponding node, but that need to access large
file system data sets that need to be spread across the several nodes in the
jobs cpuset in order to fit. Without this patch, especially for jobs that
might have one thread reading in the data set, the memory allocation across
the nodes in the jobs cpuset can become very uneven.
A couple of Copyright year ranges are updated as well. And a couple of email
addresses that can be found in the MAINTAINERS file are removed.
Signed-off-by: Paul Jackson <pj@sgi.com>
Signed-off-by: Andrew Morton <akpm@osdl.org>
Signed-off-by: Linus Torvalds <torvalds@osdl.org>
2006-03-24 04:16:03 -07:00
|
|
|
CS_NOTIFY_ON_RELEASE,
|
|
|
|
CS_SPREAD_PAGE,
|
|
|
|
CS_SPREAD_SLAB,
|
2005-04-16 16:20:36 -06:00
|
|
|
} cpuset_flagbits_t;
|
|
|
|
|
|
|
|
/* convenient tests for these bits */
|
|
|
|
static inline int is_cpu_exclusive(const struct cpuset *cs)
|
|
|
|
{
|
2006-03-24 04:16:00 -07:00
|
|
|
return test_bit(CS_CPU_EXCLUSIVE, &cs->flags);
|
2005-04-16 16:20:36 -06:00
|
|
|
}
|
|
|
|
|
|
|
|
static inline int is_mem_exclusive(const struct cpuset *cs)
|
|
|
|
{
|
2006-03-24 04:16:00 -07:00
|
|
|
return test_bit(CS_MEM_EXCLUSIVE, &cs->flags);
|
2005-04-16 16:20:36 -06:00
|
|
|
}
|
|
|
|
|
|
|
|
static inline int is_removed(const struct cpuset *cs)
|
|
|
|
{
|
2006-03-24 04:16:00 -07:00
|
|
|
return test_bit(CS_REMOVED, &cs->flags);
|
2005-04-16 16:20:36 -06:00
|
|
|
}
|
|
|
|
|
|
|
|
static inline int notify_on_release(const struct cpuset *cs)
|
|
|
|
{
|
2006-03-24 04:16:00 -07:00
|
|
|
return test_bit(CS_NOTIFY_ON_RELEASE, &cs->flags);
|
2005-04-16 16:20:36 -06:00
|
|
|
}
|
|
|
|
|
[PATCH] cpusets: swap migration interface
Add a boolean "memory_migrate" to each cpuset, represented by a file
containing "0" or "1" in each directory below /dev/cpuset.
It defaults to false (file contains "0"). It can be set true by writing
"1" to the file.
If true, then anytime that a task is attached to the cpuset so marked, the
pages of that task will be moved to that cpuset, preserving, to the extent
practical, the cpuset-relative placement of the pages.
Also anytime that a cpuset so marked has its memory placement changed (by
writing to its "mems" file), the tasks in that cpuset will have their pages
moved to the cpusets new nodes, preserving, to the extent practical, the
cpuset-relative placement of the moved pages.
Signed-off-by: Paul Jackson <pj@sgi.com>
Cc: Christoph Lameter <christoph@lameter.com>
Signed-off-by: Andrew Morton <akpm@osdl.org>
Signed-off-by: Linus Torvalds <torvalds@osdl.org>
2006-01-08 02:00:56 -07:00
|
|
|
static inline int is_memory_migrate(const struct cpuset *cs)
|
|
|
|
{
|
2006-03-24 04:16:00 -07:00
|
|
|
return test_bit(CS_MEMORY_MIGRATE, &cs->flags);
|
[PATCH] cpusets: swap migration interface
Add a boolean "memory_migrate" to each cpuset, represented by a file
containing "0" or "1" in each directory below /dev/cpuset.
It defaults to false (file contains "0"). It can be set true by writing
"1" to the file.
If true, then anytime that a task is attached to the cpuset so marked, the
pages of that task will be moved to that cpuset, preserving, to the extent
practical, the cpuset-relative placement of the pages.
Also anytime that a cpuset so marked has its memory placement changed (by
writing to its "mems" file), the tasks in that cpuset will have their pages
moved to the cpusets new nodes, preserving, to the extent practical, the
cpuset-relative placement of the moved pages.
Signed-off-by: Paul Jackson <pj@sgi.com>
Cc: Christoph Lameter <christoph@lameter.com>
Signed-off-by: Andrew Morton <akpm@osdl.org>
Signed-off-by: Linus Torvalds <torvalds@osdl.org>
2006-01-08 02:00:56 -07:00
|
|
|
}
|
|
|
|
|
[PATCH] cpuset memory spread basic implementation
This patch provides the implementation and cpuset interface for an alternative
memory allocation policy that can be applied to certain kinds of memory
allocations, such as the page cache (file system buffers) and some slab caches
(such as inode caches).
The policy is called "memory spreading." If enabled, it spreads out these
kinds of memory allocations over all the nodes allowed to a task, instead of
preferring to place them on the node where the task is executing.
All other kinds of allocations, including anonymous pages for a tasks stack
and data regions, are not affected by this policy choice, and continue to be
allocated preferring the node local to execution, as modified by the NUMA
mempolicy.
There are two boolean flag files per cpuset that control where the kernel
allocates pages for the file system buffers and related in kernel data
structures. They are called 'memory_spread_page' and 'memory_spread_slab'.
If the per-cpuset boolean flag file 'memory_spread_page' is set, then the
kernel will spread the file system buffers (page cache) evenly over all the
nodes that the faulting task is allowed to use, instead of preferring to put
those pages on the node where the task is running.
If the per-cpuset boolean flag file 'memory_spread_slab' is set, then the
kernel will spread some file system related slab caches, such as for inodes
and dentries evenly over all the nodes that the faulting task is allowed to
use, instead of preferring to put those pages on the node where the task is
running.
The implementation is simple. Setting the cpuset flags 'memory_spread_page'
or 'memory_spread_cache' turns on the per-process flags PF_SPREAD_PAGE or
PF_SPREAD_SLAB, respectively, for each task that is in the cpuset or
subsequently joins that cpuset. In subsequent patches, the page allocation
calls for the affected page cache and slab caches are modified to perform an
inline check for these flags, and if set, a call to a new routine
cpuset_mem_spread_node() returns the node to prefer for the allocation.
The cpuset_mem_spread_node() routine is also simple. It uses the value of a
per-task rotor cpuset_mem_spread_rotor to select the next node in the current
tasks mems_allowed to prefer for the allocation.
This policy can provide substantial improvements for jobs that need to place
thread local data on the corresponding node, but that need to access large
file system data sets that need to be spread across the several nodes in the
jobs cpuset in order to fit. Without this patch, especially for jobs that
might have one thread reading in the data set, the memory allocation across
the nodes in the jobs cpuset can become very uneven.
A couple of Copyright year ranges are updated as well. And a couple of email
addresses that can be found in the MAINTAINERS file are removed.
Signed-off-by: Paul Jackson <pj@sgi.com>
Signed-off-by: Andrew Morton <akpm@osdl.org>
Signed-off-by: Linus Torvalds <torvalds@osdl.org>
2006-03-24 04:16:03 -07:00
|
|
|
static inline int is_spread_page(const struct cpuset *cs)
|
|
|
|
{
|
|
|
|
return test_bit(CS_SPREAD_PAGE, &cs->flags);
|
|
|
|
}
|
|
|
|
|
|
|
|
static inline int is_spread_slab(const struct cpuset *cs)
|
|
|
|
{
|
|
|
|
return test_bit(CS_SPREAD_SLAB, &cs->flags);
|
|
|
|
}
|
|
|
|
|
2005-04-16 16:20:36 -06:00
|
|
|
/*
|
2006-03-24 04:16:11 -07:00
|
|
|
* Increment this integer everytime any cpuset changes its
|
2005-04-16 16:20:36 -06:00
|
|
|
* mems_allowed value. Users of cpusets can track this generation
|
|
|
|
* number, and avoid having to lock and reload mems_allowed unless
|
|
|
|
* the cpuset they're using changes generation.
|
|
|
|
*
|
|
|
|
* A single, global generation is needed because attach_task() could
|
|
|
|
* reattach a task to a different cpuset, which must not have its
|
|
|
|
* generation numbers aliased with those of that tasks previous cpuset.
|
|
|
|
*
|
|
|
|
* Generations are needed for mems_allowed because one task cannot
|
|
|
|
* modify anothers memory placement. So we must enable every task,
|
|
|
|
* on every visit to __alloc_pages(), to efficiently check whether
|
|
|
|
* its current->cpuset->mems_allowed has changed, requiring an update
|
|
|
|
* of its current->mems_allowed.
|
2006-03-24 04:16:11 -07:00
|
|
|
*
|
|
|
|
* Since cpuset_mems_generation is guarded by manage_mutex,
|
|
|
|
* there is no need to mark it atomic.
|
2005-04-16 16:20:36 -06:00
|
|
|
*/
|
2006-03-24 04:16:11 -07:00
|
|
|
static int cpuset_mems_generation;
|
2005-04-16 16:20:36 -06:00
|
|
|
|
|
|
|
static struct cpuset top_cpuset = {
|
|
|
|
.flags = ((1 << CS_CPU_EXCLUSIVE) | (1 << CS_MEM_EXCLUSIVE)),
|
|
|
|
.cpus_allowed = CPU_MASK_ALL,
|
|
|
|
.mems_allowed = NODE_MASK_ALL,
|
|
|
|
.count = ATOMIC_INIT(0),
|
|
|
|
.sibling = LIST_HEAD_INIT(top_cpuset.sibling),
|
|
|
|
.children = LIST_HEAD_INIT(top_cpuset.children),
|
|
|
|
};
|
|
|
|
|
|
|
|
static struct vfsmount *cpuset_mount;
|
[PATCH] cpuset: memory pressure meter
Provide a simple per-cpuset metric of memory pressure, tracking the -rate-
that the tasks in a cpuset call try_to_free_pages(), the synchronous
(direct) memory reclaim code.
This enables batch managers monitoring jobs running in dedicated cpusets to
efficiently detect what level of memory pressure that job is causing.
This is useful both on tightly managed systems running a wide mix of
submitted jobs, which may choose to terminate or reprioritize jobs that are
trying to use more memory than allowed on the nodes assigned them, and with
tightly coupled, long running, massively parallel scientific computing jobs
that will dramatically fail to meet required performance goals if they
start to use more memory than allowed to them.
This patch just provides a very economical way for the batch manager to
monitor a cpuset for signs of memory pressure. It's up to the batch
manager or other user code to decide what to do about it and take action.
==> Unless this feature is enabled by writing "1" to the special file
/dev/cpuset/memory_pressure_enabled, the hook in the rebalance
code of __alloc_pages() for this metric reduces to simply noticing
that the cpuset_memory_pressure_enabled flag is zero. So only
systems that enable this feature will compute the metric.
Why a per-cpuset, running average:
Because this meter is per-cpuset, rather than per-task or mm, the
system load imposed by a batch scheduler monitoring this metric is
sharply reduced on large systems, because a scan of the tasklist can be
avoided on each set of queries.
Because this meter is a running average, instead of an accumulating
counter, a batch scheduler can detect memory pressure with a single
read, instead of having to read and accumulate results for a period of
time.
Because this meter is per-cpuset rather than per-task or mm, the
batch scheduler can obtain the key information, memory pressure in a
cpuset, with a single read, rather than having to query and accumulate
results over all the (dynamically changing) set of tasks in the cpuset.
A per-cpuset simple digital filter (requires a spinlock and 3 words of data
per-cpuset) is kept, and updated by any task attached to that cpuset, if it
enters the synchronous (direct) page reclaim code.
A per-cpuset file provides an integer number representing the recent
(half-life of 10 seconds) rate of direct page reclaims caused by the tasks
in the cpuset, in units of reclaims attempted per second, times 1000.
Signed-off-by: Paul Jackson <pj@sgi.com>
Signed-off-by: Andrew Morton <akpm@osdl.org>
Signed-off-by: Linus Torvalds <torvalds@osdl.org>
2006-01-08 02:01:49 -07:00
|
|
|
static struct super_block *cpuset_sb;
|
2005-04-16 16:20:36 -06:00
|
|
|
|
|
|
|
/*
|
2006-03-23 04:00:18 -07:00
|
|
|
* We have two global cpuset mutexes below. They can nest.
|
|
|
|
* It is ok to first take manage_mutex, then nest callback_mutex. We also
|
[PATCH] cpusets: dual semaphore locking overhaul
Overhaul cpuset locking. Replace single semaphore with two semaphores.
The suggestion to use two locks was made by Roman Zippel.
Both locks are global. Code that wants to modify cpusets must first
acquire the exclusive manage_sem, which allows them read-only access to
cpusets, and holds off other would-be modifiers. Before making actual
changes, the second semaphore, callback_sem must be acquired as well. Code
that needs only to query cpusets must acquire callback_sem, which is also a
global exclusive lock.
The earlier problems with double tripping are avoided, because it is
allowed for holders of manage_sem to nest the second callback_sem lock, and
only callback_sem is needed by code called from within __alloc_pages(),
where the double tripping had been possible.
This is not quite the same as a normal read/write semaphore, because
obtaining read-only access with intent to change must hold off other such
attempts, while allowing read-only access w/o such intention. Changing
cpusets involves several related checks and changes, which must be done
while allowing read-only queries (to avoid the double trip), but while
ensuring nothing changes (holding off other would be modifiers.)
This overhaul of cpuset locking also makes careful use of task_lock() to
guard access to the task->cpuset pointer, closing a couple of race
conditions noticed while reading this code (thanks, Roman). I've never
seen these races fail in any use or test.
See further the comments in the code.
Signed-off-by: Paul Jackson <pj@sgi.com>
Signed-off-by: Andrew Morton <akpm@osdl.org>
Signed-off-by: Linus Torvalds <torvalds@osdl.org>
2005-10-30 16:02:30 -07:00
|
|
|
* require taking task_lock() when dereferencing a tasks cpuset pointer.
|
|
|
|
* See "The task_lock() exception", at the end of this comment.
|
|
|
|
*
|
2006-03-23 04:00:18 -07:00
|
|
|
* A task must hold both mutexes to modify cpusets. If a task
|
|
|
|
* holds manage_mutex, then it blocks others wanting that mutex,
|
|
|
|
* ensuring that it is the only task able to also acquire callback_mutex
|
[PATCH] cpusets: dual semaphore locking overhaul
Overhaul cpuset locking. Replace single semaphore with two semaphores.
The suggestion to use two locks was made by Roman Zippel.
Both locks are global. Code that wants to modify cpusets must first
acquire the exclusive manage_sem, which allows them read-only access to
cpusets, and holds off other would-be modifiers. Before making actual
changes, the second semaphore, callback_sem must be acquired as well. Code
that needs only to query cpusets must acquire callback_sem, which is also a
global exclusive lock.
The earlier problems with double tripping are avoided, because it is
allowed for holders of manage_sem to nest the second callback_sem lock, and
only callback_sem is needed by code called from within __alloc_pages(),
where the double tripping had been possible.
This is not quite the same as a normal read/write semaphore, because
obtaining read-only access with intent to change must hold off other such
attempts, while allowing read-only access w/o such intention. Changing
cpusets involves several related checks and changes, which must be done
while allowing read-only queries (to avoid the double trip), but while
ensuring nothing changes (holding off other would be modifiers.)
This overhaul of cpuset locking also makes careful use of task_lock() to
guard access to the task->cpuset pointer, closing a couple of race
conditions noticed while reading this code (thanks, Roman). I've never
seen these races fail in any use or test.
See further the comments in the code.
Signed-off-by: Paul Jackson <pj@sgi.com>
Signed-off-by: Andrew Morton <akpm@osdl.org>
Signed-off-by: Linus Torvalds <torvalds@osdl.org>
2005-10-30 16:02:30 -07:00
|
|
|
* and be able to modify cpusets. It can perform various checks on
|
|
|
|
* the cpuset structure first, knowing nothing will change. It can
|
2006-03-23 04:00:18 -07:00
|
|
|
* also allocate memory while just holding manage_mutex. While it is
|
[PATCH] cpusets: dual semaphore locking overhaul
Overhaul cpuset locking. Replace single semaphore with two semaphores.
The suggestion to use two locks was made by Roman Zippel.
Both locks are global. Code that wants to modify cpusets must first
acquire the exclusive manage_sem, which allows them read-only access to
cpusets, and holds off other would-be modifiers. Before making actual
changes, the second semaphore, callback_sem must be acquired as well. Code
that needs only to query cpusets must acquire callback_sem, which is also a
global exclusive lock.
The earlier problems with double tripping are avoided, because it is
allowed for holders of manage_sem to nest the second callback_sem lock, and
only callback_sem is needed by code called from within __alloc_pages(),
where the double tripping had been possible.
This is not quite the same as a normal read/write semaphore, because
obtaining read-only access with intent to change must hold off other such
attempts, while allowing read-only access w/o such intention. Changing
cpusets involves several related checks and changes, which must be done
while allowing read-only queries (to avoid the double trip), but while
ensuring nothing changes (holding off other would be modifiers.)
This overhaul of cpuset locking also makes careful use of task_lock() to
guard access to the task->cpuset pointer, closing a couple of race
conditions noticed while reading this code (thanks, Roman). I've never
seen these races fail in any use or test.
See further the comments in the code.
Signed-off-by: Paul Jackson <pj@sgi.com>
Signed-off-by: Andrew Morton <akpm@osdl.org>
Signed-off-by: Linus Torvalds <torvalds@osdl.org>
2005-10-30 16:02:30 -07:00
|
|
|
* performing these checks, various callback routines can briefly
|
2006-03-23 04:00:18 -07:00
|
|
|
* acquire callback_mutex to query cpusets. Once it is ready to make
|
|
|
|
* the changes, it takes callback_mutex, blocking everyone else.
|
[PATCH] cpusets: dual semaphore locking overhaul
Overhaul cpuset locking. Replace single semaphore with two semaphores.
The suggestion to use two locks was made by Roman Zippel.
Both locks are global. Code that wants to modify cpusets must first
acquire the exclusive manage_sem, which allows them read-only access to
cpusets, and holds off other would-be modifiers. Before making actual
changes, the second semaphore, callback_sem must be acquired as well. Code
that needs only to query cpusets must acquire callback_sem, which is also a
global exclusive lock.
The earlier problems with double tripping are avoided, because it is
allowed for holders of manage_sem to nest the second callback_sem lock, and
only callback_sem is needed by code called from within __alloc_pages(),
where the double tripping had been possible.
This is not quite the same as a normal read/write semaphore, because
obtaining read-only access with intent to change must hold off other such
attempts, while allowing read-only access w/o such intention. Changing
cpusets involves several related checks and changes, which must be done
while allowing read-only queries (to avoid the double trip), but while
ensuring nothing changes (holding off other would be modifiers.)
This overhaul of cpuset locking also makes careful use of task_lock() to
guard access to the task->cpuset pointer, closing a couple of race
conditions noticed while reading this code (thanks, Roman). I've never
seen these races fail in any use or test.
See further the comments in the code.
Signed-off-by: Paul Jackson <pj@sgi.com>
Signed-off-by: Andrew Morton <akpm@osdl.org>
Signed-off-by: Linus Torvalds <torvalds@osdl.org>
2005-10-30 16:02:30 -07:00
|
|
|
*
|
|
|
|
* Calls to the kernel memory allocator can not be made while holding
|
2006-03-23 04:00:18 -07:00
|
|
|
* callback_mutex, as that would risk double tripping on callback_mutex
|
[PATCH] cpusets: dual semaphore locking overhaul
Overhaul cpuset locking. Replace single semaphore with two semaphores.
The suggestion to use two locks was made by Roman Zippel.
Both locks are global. Code that wants to modify cpusets must first
acquire the exclusive manage_sem, which allows them read-only access to
cpusets, and holds off other would-be modifiers. Before making actual
changes, the second semaphore, callback_sem must be acquired as well. Code
that needs only to query cpusets must acquire callback_sem, which is also a
global exclusive lock.
The earlier problems with double tripping are avoided, because it is
allowed for holders of manage_sem to nest the second callback_sem lock, and
only callback_sem is needed by code called from within __alloc_pages(),
where the double tripping had been possible.
This is not quite the same as a normal read/write semaphore, because
obtaining read-only access with intent to change must hold off other such
attempts, while allowing read-only access w/o such intention. Changing
cpusets involves several related checks and changes, which must be done
while allowing read-only queries (to avoid the double trip), but while
ensuring nothing changes (holding off other would be modifiers.)
This overhaul of cpuset locking also makes careful use of task_lock() to
guard access to the task->cpuset pointer, closing a couple of race
conditions noticed while reading this code (thanks, Roman). I've never
seen these races fail in any use or test.
See further the comments in the code.
Signed-off-by: Paul Jackson <pj@sgi.com>
Signed-off-by: Andrew Morton <akpm@osdl.org>
Signed-off-by: Linus Torvalds <torvalds@osdl.org>
2005-10-30 16:02:30 -07:00
|
|
|
* from one of the callbacks into the cpuset code from within
|
|
|
|
* __alloc_pages().
|
|
|
|
*
|
2006-03-23 04:00:18 -07:00
|
|
|
* If a task is only holding callback_mutex, then it has read-only
|
[PATCH] cpusets: dual semaphore locking overhaul
Overhaul cpuset locking. Replace single semaphore with two semaphores.
The suggestion to use two locks was made by Roman Zippel.
Both locks are global. Code that wants to modify cpusets must first
acquire the exclusive manage_sem, which allows them read-only access to
cpusets, and holds off other would-be modifiers. Before making actual
changes, the second semaphore, callback_sem must be acquired as well. Code
that needs only to query cpusets must acquire callback_sem, which is also a
global exclusive lock.
The earlier problems with double tripping are avoided, because it is
allowed for holders of manage_sem to nest the second callback_sem lock, and
only callback_sem is needed by code called from within __alloc_pages(),
where the double tripping had been possible.
This is not quite the same as a normal read/write semaphore, because
obtaining read-only access with intent to change must hold off other such
attempts, while allowing read-only access w/o such intention. Changing
cpusets involves several related checks and changes, which must be done
while allowing read-only queries (to avoid the double trip), but while
ensuring nothing changes (holding off other would be modifiers.)
This overhaul of cpuset locking also makes careful use of task_lock() to
guard access to the task->cpuset pointer, closing a couple of race
conditions noticed while reading this code (thanks, Roman). I've never
seen these races fail in any use or test.
See further the comments in the code.
Signed-off-by: Paul Jackson <pj@sgi.com>
Signed-off-by: Andrew Morton <akpm@osdl.org>
Signed-off-by: Linus Torvalds <torvalds@osdl.org>
2005-10-30 16:02:30 -07:00
|
|
|
* access to cpusets.
|
|
|
|
*
|
|
|
|
* The task_struct fields mems_allowed and mems_generation may only
|
|
|
|
* be accessed in the context of that task, so require no locks.
|
|
|
|
*
|
|
|
|
* Any task can increment and decrement the count field without lock.
|
2006-03-23 04:00:18 -07:00
|
|
|
* So in general, code holding manage_mutex or callback_mutex can't rely
|
[PATCH] cpusets: dual semaphore locking overhaul
Overhaul cpuset locking. Replace single semaphore with two semaphores.
The suggestion to use two locks was made by Roman Zippel.
Both locks are global. Code that wants to modify cpusets must first
acquire the exclusive manage_sem, which allows them read-only access to
cpusets, and holds off other would-be modifiers. Before making actual
changes, the second semaphore, callback_sem must be acquired as well. Code
that needs only to query cpusets must acquire callback_sem, which is also a
global exclusive lock.
The earlier problems with double tripping are avoided, because it is
allowed for holders of manage_sem to nest the second callback_sem lock, and
only callback_sem is needed by code called from within __alloc_pages(),
where the double tripping had been possible.
This is not quite the same as a normal read/write semaphore, because
obtaining read-only access with intent to change must hold off other such
attempts, while allowing read-only access w/o such intention. Changing
cpusets involves several related checks and changes, which must be done
while allowing read-only queries (to avoid the double trip), but while
ensuring nothing changes (holding off other would be modifiers.)
This overhaul of cpuset locking also makes careful use of task_lock() to
guard access to the task->cpuset pointer, closing a couple of race
conditions noticed while reading this code (thanks, Roman). I've never
seen these races fail in any use or test.
See further the comments in the code.
Signed-off-by: Paul Jackson <pj@sgi.com>
Signed-off-by: Andrew Morton <akpm@osdl.org>
Signed-off-by: Linus Torvalds <torvalds@osdl.org>
2005-10-30 16:02:30 -07:00
|
|
|
* on the count field not changing. However, if the count goes to
|
2006-03-23 04:00:18 -07:00
|
|
|
* zero, then only attach_task(), which holds both mutexes, can
|
[PATCH] cpusets: dual semaphore locking overhaul
Overhaul cpuset locking. Replace single semaphore with two semaphores.
The suggestion to use two locks was made by Roman Zippel.
Both locks are global. Code that wants to modify cpusets must first
acquire the exclusive manage_sem, which allows them read-only access to
cpusets, and holds off other would-be modifiers. Before making actual
changes, the second semaphore, callback_sem must be acquired as well. Code
that needs only to query cpusets must acquire callback_sem, which is also a
global exclusive lock.
The earlier problems with double tripping are avoided, because it is
allowed for holders of manage_sem to nest the second callback_sem lock, and
only callback_sem is needed by code called from within __alloc_pages(),
where the double tripping had been possible.
This is not quite the same as a normal read/write semaphore, because
obtaining read-only access with intent to change must hold off other such
attempts, while allowing read-only access w/o such intention. Changing
cpusets involves several related checks and changes, which must be done
while allowing read-only queries (to avoid the double trip), but while
ensuring nothing changes (holding off other would be modifiers.)
This overhaul of cpuset locking also makes careful use of task_lock() to
guard access to the task->cpuset pointer, closing a couple of race
conditions noticed while reading this code (thanks, Roman). I've never
seen these races fail in any use or test.
See further the comments in the code.
Signed-off-by: Paul Jackson <pj@sgi.com>
Signed-off-by: Andrew Morton <akpm@osdl.org>
Signed-off-by: Linus Torvalds <torvalds@osdl.org>
2005-10-30 16:02:30 -07:00
|
|
|
* increment it again. Because a count of zero means that no tasks
|
|
|
|
* are currently attached, therefore there is no way a task attached
|
|
|
|
* to that cpuset can fork (the other way to increment the count).
|
2006-03-23 04:00:18 -07:00
|
|
|
* So code holding manage_mutex or callback_mutex can safely assume that
|
[PATCH] cpusets: dual semaphore locking overhaul
Overhaul cpuset locking. Replace single semaphore with two semaphores.
The suggestion to use two locks was made by Roman Zippel.
Both locks are global. Code that wants to modify cpusets must first
acquire the exclusive manage_sem, which allows them read-only access to
cpusets, and holds off other would-be modifiers. Before making actual
changes, the second semaphore, callback_sem must be acquired as well. Code
that needs only to query cpusets must acquire callback_sem, which is also a
global exclusive lock.
The earlier problems with double tripping are avoided, because it is
allowed for holders of manage_sem to nest the second callback_sem lock, and
only callback_sem is needed by code called from within __alloc_pages(),
where the double tripping had been possible.
This is not quite the same as a normal read/write semaphore, because
obtaining read-only access with intent to change must hold off other such
attempts, while allowing read-only access w/o such intention. Changing
cpusets involves several related checks and changes, which must be done
while allowing read-only queries (to avoid the double trip), but while
ensuring nothing changes (holding off other would be modifiers.)
This overhaul of cpuset locking also makes careful use of task_lock() to
guard access to the task->cpuset pointer, closing a couple of race
conditions noticed while reading this code (thanks, Roman). I've never
seen these races fail in any use or test.
See further the comments in the code.
Signed-off-by: Paul Jackson <pj@sgi.com>
Signed-off-by: Andrew Morton <akpm@osdl.org>
Signed-off-by: Linus Torvalds <torvalds@osdl.org>
2005-10-30 16:02:30 -07:00
|
|
|
* if the count is zero, it will stay zero. Similarly, if a task
|
2006-03-23 04:00:18 -07:00
|
|
|
* holds manage_mutex or callback_mutex on a cpuset with zero count, it
|
[PATCH] cpusets: dual semaphore locking overhaul
Overhaul cpuset locking. Replace single semaphore with two semaphores.
The suggestion to use two locks was made by Roman Zippel.
Both locks are global. Code that wants to modify cpusets must first
acquire the exclusive manage_sem, which allows them read-only access to
cpusets, and holds off other would-be modifiers. Before making actual
changes, the second semaphore, callback_sem must be acquired as well. Code
that needs only to query cpusets must acquire callback_sem, which is also a
global exclusive lock.
The earlier problems with double tripping are avoided, because it is
allowed for holders of manage_sem to nest the second callback_sem lock, and
only callback_sem is needed by code called from within __alloc_pages(),
where the double tripping had been possible.
This is not quite the same as a normal read/write semaphore, because
obtaining read-only access with intent to change must hold off other such
attempts, while allowing read-only access w/o such intention. Changing
cpusets involves several related checks and changes, which must be done
while allowing read-only queries (to avoid the double trip), but while
ensuring nothing changes (holding off other would be modifiers.)
This overhaul of cpuset locking also makes careful use of task_lock() to
guard access to the task->cpuset pointer, closing a couple of race
conditions noticed while reading this code (thanks, Roman). I've never
seen these races fail in any use or test.
See further the comments in the code.
Signed-off-by: Paul Jackson <pj@sgi.com>
Signed-off-by: Andrew Morton <akpm@osdl.org>
Signed-off-by: Linus Torvalds <torvalds@osdl.org>
2005-10-30 16:02:30 -07:00
|
|
|
* knows that the cpuset won't be removed, as cpuset_rmdir() needs
|
2006-03-23 04:00:18 -07:00
|
|
|
* both of those mutexes.
|
[PATCH] cpusets: dual semaphore locking overhaul
Overhaul cpuset locking. Replace single semaphore with two semaphores.
The suggestion to use two locks was made by Roman Zippel.
Both locks are global. Code that wants to modify cpusets must first
acquire the exclusive manage_sem, which allows them read-only access to
cpusets, and holds off other would-be modifiers. Before making actual
changes, the second semaphore, callback_sem must be acquired as well. Code
that needs only to query cpusets must acquire callback_sem, which is also a
global exclusive lock.
The earlier problems with double tripping are avoided, because it is
allowed for holders of manage_sem to nest the second callback_sem lock, and
only callback_sem is needed by code called from within __alloc_pages(),
where the double tripping had been possible.
This is not quite the same as a normal read/write semaphore, because
obtaining read-only access with intent to change must hold off other such
attempts, while allowing read-only access w/o such intention. Changing
cpusets involves several related checks and changes, which must be done
while allowing read-only queries (to avoid the double trip), but while
ensuring nothing changes (holding off other would be modifiers.)
This overhaul of cpuset locking also makes careful use of task_lock() to
guard access to the task->cpuset pointer, closing a couple of race
conditions noticed while reading this code (thanks, Roman). I've never
seen these races fail in any use or test.
See further the comments in the code.
Signed-off-by: Paul Jackson <pj@sgi.com>
Signed-off-by: Andrew Morton <akpm@osdl.org>
Signed-off-by: Linus Torvalds <torvalds@osdl.org>
2005-10-30 16:02:30 -07:00
|
|
|
*
|
|
|
|
* The cpuset_common_file_write handler for operations that modify
|
2006-03-23 04:00:18 -07:00
|
|
|
* the cpuset hierarchy holds manage_mutex across the entire operation,
|
[PATCH] cpusets: dual semaphore locking overhaul
Overhaul cpuset locking. Replace single semaphore with two semaphores.
The suggestion to use two locks was made by Roman Zippel.
Both locks are global. Code that wants to modify cpusets must first
acquire the exclusive manage_sem, which allows them read-only access to
cpusets, and holds off other would-be modifiers. Before making actual
changes, the second semaphore, callback_sem must be acquired as well. Code
that needs only to query cpusets must acquire callback_sem, which is also a
global exclusive lock.
The earlier problems with double tripping are avoided, because it is
allowed for holders of manage_sem to nest the second callback_sem lock, and
only callback_sem is needed by code called from within __alloc_pages(),
where the double tripping had been possible.
This is not quite the same as a normal read/write semaphore, because
obtaining read-only access with intent to change must hold off other such
attempts, while allowing read-only access w/o such intention. Changing
cpusets involves several related checks and changes, which must be done
while allowing read-only queries (to avoid the double trip), but while
ensuring nothing changes (holding off other would be modifiers.)
This overhaul of cpuset locking also makes careful use of task_lock() to
guard access to the task->cpuset pointer, closing a couple of race
conditions noticed while reading this code (thanks, Roman). I've never
seen these races fail in any use or test.
See further the comments in the code.
Signed-off-by: Paul Jackson <pj@sgi.com>
Signed-off-by: Andrew Morton <akpm@osdl.org>
Signed-off-by: Linus Torvalds <torvalds@osdl.org>
2005-10-30 16:02:30 -07:00
|
|
|
* single threading all such cpuset modifications across the system.
|
|
|
|
*
|
2006-03-23 04:00:18 -07:00
|
|
|
* The cpuset_common_file_read() handlers only hold callback_mutex across
|
[PATCH] cpusets: dual semaphore locking overhaul
Overhaul cpuset locking. Replace single semaphore with two semaphores.
The suggestion to use two locks was made by Roman Zippel.
Both locks are global. Code that wants to modify cpusets must first
acquire the exclusive manage_sem, which allows them read-only access to
cpusets, and holds off other would-be modifiers. Before making actual
changes, the second semaphore, callback_sem must be acquired as well. Code
that needs only to query cpusets must acquire callback_sem, which is also a
global exclusive lock.
The earlier problems with double tripping are avoided, because it is
allowed for holders of manage_sem to nest the second callback_sem lock, and
only callback_sem is needed by code called from within __alloc_pages(),
where the double tripping had been possible.
This is not quite the same as a normal read/write semaphore, because
obtaining read-only access with intent to change must hold off other such
attempts, while allowing read-only access w/o such intention. Changing
cpusets involves several related checks and changes, which must be done
while allowing read-only queries (to avoid the double trip), but while
ensuring nothing changes (holding off other would be modifiers.)
This overhaul of cpuset locking also makes careful use of task_lock() to
guard access to the task->cpuset pointer, closing a couple of race
conditions noticed while reading this code (thanks, Roman). I've never
seen these races fail in any use or test.
See further the comments in the code.
Signed-off-by: Paul Jackson <pj@sgi.com>
Signed-off-by: Andrew Morton <akpm@osdl.org>
Signed-off-by: Linus Torvalds <torvalds@osdl.org>
2005-10-30 16:02:30 -07:00
|
|
|
* small pieces of code, such as when reading out possibly multi-word
|
|
|
|
* cpumasks and nodemasks.
|
|
|
|
*
|
|
|
|
* The fork and exit callbacks cpuset_fork() and cpuset_exit(), don't
|
2006-03-23 04:00:18 -07:00
|
|
|
* (usually) take either mutex. These are the two most performance
|
[PATCH] cpusets: dual semaphore locking overhaul
Overhaul cpuset locking. Replace single semaphore with two semaphores.
The suggestion to use two locks was made by Roman Zippel.
Both locks are global. Code that wants to modify cpusets must first
acquire the exclusive manage_sem, which allows them read-only access to
cpusets, and holds off other would-be modifiers. Before making actual
changes, the second semaphore, callback_sem must be acquired as well. Code
that needs only to query cpusets must acquire callback_sem, which is also a
global exclusive lock.
The earlier problems with double tripping are avoided, because it is
allowed for holders of manage_sem to nest the second callback_sem lock, and
only callback_sem is needed by code called from within __alloc_pages(),
where the double tripping had been possible.
This is not quite the same as a normal read/write semaphore, because
obtaining read-only access with intent to change must hold off other such
attempts, while allowing read-only access w/o such intention. Changing
cpusets involves several related checks and changes, which must be done
while allowing read-only queries (to avoid the double trip), but while
ensuring nothing changes (holding off other would be modifiers.)
This overhaul of cpuset locking also makes careful use of task_lock() to
guard access to the task->cpuset pointer, closing a couple of race
conditions noticed while reading this code (thanks, Roman). I've never
seen these races fail in any use or test.
See further the comments in the code.
Signed-off-by: Paul Jackson <pj@sgi.com>
Signed-off-by: Andrew Morton <akpm@osdl.org>
Signed-off-by: Linus Torvalds <torvalds@osdl.org>
2005-10-30 16:02:30 -07:00
|
|
|
* critical pieces of code here. The exception occurs on cpuset_exit(),
|
2006-03-23 04:00:18 -07:00
|
|
|
* when a task in a notify_on_release cpuset exits. Then manage_mutex
|
[PATCH] cpuset exit NULL dereference fix
There is a race in the kernel cpuset code, between the code
to handle notify_on_release, and the code to remove a cpuset.
The notify_on_release code can end up trying to access a
cpuset that has been removed. In the most common case, this
causes a NULL pointer dereference from the routine cpuset_path.
However all manner of bad things are possible, in theory at least.
The existing code decrements the cpuset use count, and if the
count goes to zero, processes the notify_on_release request,
if appropriate. However, once the count goes to zero, unless we
are holding the global cpuset_sem semaphore, there is nothing to
stop another task from immediately removing the cpuset entirely,
and recycling its memory.
The obvious fix would be to always hold the cpuset_sem
semaphore while decrementing the use count and dealing with
notify_on_release. However we don't want to force a global
semaphore into the mainline task exit path, as that might create
a scaling problem.
The actual fix is almost as easy - since this is only an issue
for cpusets using notify_on_release, which the top level big
cpusets don't normally need to use, only take the cpuset_sem
for cpusets using notify_on_release.
This code has been run for hours without a hiccup, while running
a cpuset create/destroy stress test that could crash the existing
kernel in seconds. This patch applies to the current -linus
git kernel.
Signed-off-by: Paul Jackson <pj@sgi.com>
Acked-by: Simon Derr <simon.derr@bull.net>
Acked-by: Dinakar Guniguntala <dino@in.ibm.com>
Signed-off-by: Linus Torvalds <torvalds@osdl.org>
2005-05-27 03:02:43 -06:00
|
|
|
* is taken, and if the cpuset count is zero, a usermode call made
|
2005-04-16 16:20:36 -06:00
|
|
|
* to /sbin/cpuset_release_agent with the name of the cpuset (path
|
|
|
|
* relative to the root of cpuset file system) as the argument.
|
|
|
|
*
|
[PATCH] cpusets: dual semaphore locking overhaul
Overhaul cpuset locking. Replace single semaphore with two semaphores.
The suggestion to use two locks was made by Roman Zippel.
Both locks are global. Code that wants to modify cpusets must first
acquire the exclusive manage_sem, which allows them read-only access to
cpusets, and holds off other would-be modifiers. Before making actual
changes, the second semaphore, callback_sem must be acquired as well. Code
that needs only to query cpusets must acquire callback_sem, which is also a
global exclusive lock.
The earlier problems with double tripping are avoided, because it is
allowed for holders of manage_sem to nest the second callback_sem lock, and
only callback_sem is needed by code called from within __alloc_pages(),
where the double tripping had been possible.
This is not quite the same as a normal read/write semaphore, because
obtaining read-only access with intent to change must hold off other such
attempts, while allowing read-only access w/o such intention. Changing
cpusets involves several related checks and changes, which must be done
while allowing read-only queries (to avoid the double trip), but while
ensuring nothing changes (holding off other would be modifiers.)
This overhaul of cpuset locking also makes careful use of task_lock() to
guard access to the task->cpuset pointer, closing a couple of race
conditions noticed while reading this code (thanks, Roman). I've never
seen these races fail in any use or test.
See further the comments in the code.
Signed-off-by: Paul Jackson <pj@sgi.com>
Signed-off-by: Andrew Morton <akpm@osdl.org>
Signed-off-by: Linus Torvalds <torvalds@osdl.org>
2005-10-30 16:02:30 -07:00
|
|
|
* A cpuset can only be deleted if both its 'count' of using tasks
|
|
|
|
* is zero, and its list of 'children' cpusets is empty. Since all
|
|
|
|
* tasks in the system use _some_ cpuset, and since there is always at
|
|
|
|
* least one task in the system (init, pid == 1), therefore, top_cpuset
|
|
|
|
* always has either children cpusets and/or using tasks. So we don't
|
|
|
|
* need a special hack to ensure that top_cpuset cannot be deleted.
|
|
|
|
*
|
|
|
|
* The above "Tale of Two Semaphores" would be complete, but for:
|
|
|
|
*
|
|
|
|
* The task_lock() exception
|
|
|
|
*
|
|
|
|
* The need for this exception arises from the action of attach_task(),
|
|
|
|
* which overwrites one tasks cpuset pointer with another. It does
|
2006-03-23 04:00:18 -07:00
|
|
|
* so using both mutexes, however there are several performance
|
[PATCH] cpusets: dual semaphore locking overhaul
Overhaul cpuset locking. Replace single semaphore with two semaphores.
The suggestion to use two locks was made by Roman Zippel.
Both locks are global. Code that wants to modify cpusets must first
acquire the exclusive manage_sem, which allows them read-only access to
cpusets, and holds off other would-be modifiers. Before making actual
changes, the second semaphore, callback_sem must be acquired as well. Code
that needs only to query cpusets must acquire callback_sem, which is also a
global exclusive lock.
The earlier problems with double tripping are avoided, because it is
allowed for holders of manage_sem to nest the second callback_sem lock, and
only callback_sem is needed by code called from within __alloc_pages(),
where the double tripping had been possible.
This is not quite the same as a normal read/write semaphore, because
obtaining read-only access with intent to change must hold off other such
attempts, while allowing read-only access w/o such intention. Changing
cpusets involves several related checks and changes, which must be done
while allowing read-only queries (to avoid the double trip), but while
ensuring nothing changes (holding off other would be modifiers.)
This overhaul of cpuset locking also makes careful use of task_lock() to
guard access to the task->cpuset pointer, closing a couple of race
conditions noticed while reading this code (thanks, Roman). I've never
seen these races fail in any use or test.
See further the comments in the code.
Signed-off-by: Paul Jackson <pj@sgi.com>
Signed-off-by: Andrew Morton <akpm@osdl.org>
Signed-off-by: Linus Torvalds <torvalds@osdl.org>
2005-10-30 16:02:30 -07:00
|
|
|
* critical places that need to reference task->cpuset without the
|
2006-03-23 04:00:18 -07:00
|
|
|
* expense of grabbing a system global mutex. Therefore except as
|
[PATCH] cpusets: dual semaphore locking overhaul
Overhaul cpuset locking. Replace single semaphore with two semaphores.
The suggestion to use two locks was made by Roman Zippel.
Both locks are global. Code that wants to modify cpusets must first
acquire the exclusive manage_sem, which allows them read-only access to
cpusets, and holds off other would-be modifiers. Before making actual
changes, the second semaphore, callback_sem must be acquired as well. Code
that needs only to query cpusets must acquire callback_sem, which is also a
global exclusive lock.
The earlier problems with double tripping are avoided, because it is
allowed for holders of manage_sem to nest the second callback_sem lock, and
only callback_sem is needed by code called from within __alloc_pages(),
where the double tripping had been possible.
This is not quite the same as a normal read/write semaphore, because
obtaining read-only access with intent to change must hold off other such
attempts, while allowing read-only access w/o such intention. Changing
cpusets involves several related checks and changes, which must be done
while allowing read-only queries (to avoid the double trip), but while
ensuring nothing changes (holding off other would be modifiers.)
This overhaul of cpuset locking also makes careful use of task_lock() to
guard access to the task->cpuset pointer, closing a couple of race
conditions noticed while reading this code (thanks, Roman). I've never
seen these races fail in any use or test.
See further the comments in the code.
Signed-off-by: Paul Jackson <pj@sgi.com>
Signed-off-by: Andrew Morton <akpm@osdl.org>
Signed-off-by: Linus Torvalds <torvalds@osdl.org>
2005-10-30 16:02:30 -07:00
|
|
|
* noted below, when dereferencing or, as in attach_task(), modifying
|
|
|
|
* a tasks cpuset pointer we use task_lock(), which acts on a spinlock
|
|
|
|
* (task->alloc_lock) already in the task_struct routinely used for
|
|
|
|
* such matters.
|
[PATCH] cpuset: use rcu directly optimization
Optimize the cpuset impact on page allocation, the most performance critical
cpuset hook in the kernel.
On each page allocation, the cpuset hook needs to check for a possible change
in the current tasks cpuset. It can now handle the common case, of no change,
without taking any spinlock or semaphore, thanks to RCU.
Convert a spinlock on the current task to an rcu_read_lock(), saving
approximately a memory barrier and an atomic op, depending on architecture.
This is done by adding rcu_assign_pointer() and synchronize_rcu() calls to the
write side of the task->cpuset pointer, in cpuset.c:attach_task(), to delay
freeing up a detached cpuset until after any critical sections referencing
that pointer.
Thanks to Andi Kleen, Nick Piggin and Eric Dumazet for ideas.
Signed-off-by: Paul Jackson <pj@sgi.com>
Signed-off-by: Andrew Morton <akpm@osdl.org>
Signed-off-by: Linus Torvalds <torvalds@osdl.org>
2006-01-08 02:02:02 -07:00
|
|
|
*
|
|
|
|
* P.S. One more locking exception. RCU is used to guard the
|
|
|
|
* update of a tasks cpuset pointer by attach_task() and the
|
|
|
|
* access of task->cpuset->mems_generation via that pointer in
|
|
|
|
* the routine cpuset_update_task_memory_state().
|
2005-04-16 16:20:36 -06:00
|
|
|
*/
|
|
|
|
|
2006-03-23 04:00:18 -07:00
|
|
|
static DEFINE_MUTEX(manage_mutex);
|
|
|
|
static DEFINE_MUTEX(callback_mutex);
|
[PATCH] cpuset semaphore depth check deadlock fix
The cpusets-formalize-intermediate-gfp_kernel-containment patch
has a deadlock problem.
This patch was part of a set of four patches to make more
extensive use of the cpuset 'mem_exclusive' attribute to
manage kernel GFP_KERNEL memory allocations and to constrain
the out-of-memory (oom) killer.
A task that is changing cpusets in particular ways on a system
when it is very short of free memory could double trip over
the global cpuset_sem semaphore (get the lock and then deadlock
trying to get it again).
The second attempt to get cpuset_sem would be in the routine
cpuset_zone_allowed(). This was discovered by code inspection.
I can not reproduce the problem except with an artifically
hacked kernel and a specialized stress test.
In real life you cannot hit this unless you are manipulating
cpusets, and are very unlikely to hit it unless you are rapidly
modifying cpusets on a memory tight system. Even then it would
be a rare occurence.
If you did hit it, the task double tripping over cpuset_sem
would deadlock in the kernel, and any other task also trying
to manipulate cpusets would deadlock there too, on cpuset_sem.
Your batch manager would be wedged solid (if it was cpuset
savvy), but classic Unix shells and utilities would work well
enough to reboot the system.
The unusual condition that led to this bug is that unlike most
semaphores, cpuset_sem _can_ be acquired while in the page
allocation code, when __alloc_pages() calls cpuset_zone_allowed.
So it easy to mistakenly perform the following sequence:
1) task makes system call to alter a cpuset
2) take cpuset_sem
3) try to allocate memory
4) memory allocator, via cpuset_zone_allowed, trys to take cpuset_sem
5) deadlock
The reason that this is not a serious bug for most users
is that almost all calls to allocate memory don't require
taking cpuset_sem. Only some code paths off the beaten
track require taking cpuset_sem -- which is good. Taking
a global semaphore on the main code path for allocating
memory would not scale well.
This patch fixes this deadlock by wrapping the up() and down()
calls on cpuset_sem in kernel/cpuset.c with code that tracks
the nesting depth of the current task on that semaphore, and
only does the real down() if the task doesn't hold the lock
already, and only does the real up() if the nesting depth
(number of unmatched downs) is exactly one.
The previous required use of refresh_mems(), anytime that
the cpuset_sem semaphore was acquired and the code executed
while holding that semaphore might try to allocate memory, is
no longer required. Two refresh_mems() calls were removed
thanks to this. This is a good change, as failing to get
all the necessary refresh_mems() calls placed was a primary
source of bugs in this cpuset code. The only remaining call
to refresh_mems() is made while doing a memory allocation,
if certain task memory placement data needs to be updated
from its cpuset, due to the cpuset having been changed behind
the tasks back.
Signed-off-by: Paul Jackson <pj@sgi.com>
Signed-off-by: Andrew Morton <akpm@osdl.org>
Signed-off-by: Linus Torvalds <torvalds@osdl.org>
2005-09-10 01:26:06 -06:00
|
|
|
|
2005-04-16 16:20:36 -06:00
|
|
|
/*
|
|
|
|
* A couple of forward declarations required, due to cyclic reference loop:
|
|
|
|
* cpuset_mkdir -> cpuset_create -> cpuset_populate_dir -> cpuset_add_file
|
|
|
|
* -> cpuset_create_file -> cpuset_dir_inode_operations -> cpuset_mkdir.
|
|
|
|
*/
|
|
|
|
|
|
|
|
static int cpuset_mkdir(struct inode *dir, struct dentry *dentry, int mode);
|
|
|
|
static int cpuset_rmdir(struct inode *unused_dir, struct dentry *dentry);
|
|
|
|
|
|
|
|
static struct backing_dev_info cpuset_backing_dev_info = {
|
|
|
|
.ra_pages = 0, /* No readahead */
|
|
|
|
.capabilities = BDI_CAP_NO_ACCT_DIRTY | BDI_CAP_NO_WRITEBACK,
|
|
|
|
};
|
|
|
|
|
|
|
|
static struct inode *cpuset_new_inode(mode_t mode)
|
|
|
|
{
|
|
|
|
struct inode *inode = new_inode(cpuset_sb);
|
|
|
|
|
|
|
|
if (inode) {
|
|
|
|
inode->i_mode = mode;
|
|
|
|
inode->i_uid = current->fsuid;
|
|
|
|
inode->i_gid = current->fsgid;
|
|
|
|
inode->i_blocks = 0;
|
|
|
|
inode->i_atime = inode->i_mtime = inode->i_ctime = CURRENT_TIME;
|
|
|
|
inode->i_mapping->backing_dev_info = &cpuset_backing_dev_info;
|
|
|
|
}
|
|
|
|
return inode;
|
|
|
|
}
|
|
|
|
|
|
|
|
static void cpuset_diput(struct dentry *dentry, struct inode *inode)
|
|
|
|
{
|
|
|
|
/* is dentry a directory ? if so, kfree() associated cpuset */
|
|
|
|
if (S_ISDIR(inode->i_mode)) {
|
|
|
|
struct cpuset *cs = dentry->d_fsdata;
|
|
|
|
BUG_ON(!(is_removed(cs)));
|
|
|
|
kfree(cs);
|
|
|
|
}
|
|
|
|
iput(inode);
|
|
|
|
}
|
|
|
|
|
|
|
|
static struct dentry_operations cpuset_dops = {
|
|
|
|
.d_iput = cpuset_diput,
|
|
|
|
};
|
|
|
|
|
|
|
|
static struct dentry *cpuset_get_dentry(struct dentry *parent, const char *name)
|
|
|
|
{
|
2005-06-23 01:09:12 -06:00
|
|
|
struct dentry *d = lookup_one_len(name, parent, strlen(name));
|
2005-04-16 16:20:36 -06:00
|
|
|
if (!IS_ERR(d))
|
|
|
|
d->d_op = &cpuset_dops;
|
|
|
|
return d;
|
|
|
|
}
|
|
|
|
|
|
|
|
static void remove_dir(struct dentry *d)
|
|
|
|
{
|
|
|
|
struct dentry *parent = dget(d->d_parent);
|
|
|
|
|
|
|
|
d_delete(d);
|
|
|
|
simple_rmdir(parent->d_inode, d);
|
|
|
|
dput(parent);
|
|
|
|
}
|
|
|
|
|
|
|
|
/*
|
|
|
|
* NOTE : the dentry must have been dget()'ed
|
|
|
|
*/
|
|
|
|
static void cpuset_d_remove_dir(struct dentry *dentry)
|
|
|
|
{
|
|
|
|
struct list_head *node;
|
|
|
|
|
|
|
|
spin_lock(&dcache_lock);
|
|
|
|
node = dentry->d_subdirs.next;
|
|
|
|
while (node != &dentry->d_subdirs) {
|
[PATCH] shrink dentry struct
Some long time ago, dentry struct was carefully tuned so that on 32 bits
UP, sizeof(struct dentry) was exactly 128, ie a power of 2, and a multiple
of memory cache lines.
Then RCU was added and dentry struct enlarged by two pointers, with nice
results for SMP, but not so good on UP, because breaking the above tuning
(128 + 8 = 136 bytes)
This patch reverts this unwanted side effect, by using an union (d_u),
where d_rcu and d_child are placed so that these two fields can share their
memory needs.
At the time d_free() is called (and d_rcu is really used), d_child is known
to be empty and not touched by the dentry freeing.
Lockless lookups only access d_name, d_parent, d_lock, d_op, d_flags (so
the previous content of d_child is not needed if said dentry was unhashed
but still accessed by a CPU because of RCU constraints)
As dentry cache easily contains millions of entries, a size reduction is
worth the extra complexity of the ugly C union.
Signed-off-by: Eric Dumazet <dada1@cosmosbay.com>
Cc: Dipankar Sarma <dipankar@in.ibm.com>
Cc: Maneesh Soni <maneesh@in.ibm.com>
Cc: Miklos Szeredi <miklos@szeredi.hu>
Cc: "Paul E. McKenney" <paulmck@us.ibm.com>
Cc: Ian Kent <raven@themaw.net>
Cc: Paul Jackson <pj@sgi.com>
Cc: Al Viro <viro@ftp.linux.org.uk>
Cc: Christoph Hellwig <hch@lst.de>
Cc: Trond Myklebust <trond.myklebust@fys.uio.no>
Cc: Neil Brown <neilb@cse.unsw.edu.au>
Cc: James Morris <jmorris@namei.org>
Cc: Stephen Smalley <sds@epoch.ncsc.mil>
Signed-off-by: Andrew Morton <akpm@osdl.org>
Signed-off-by: Linus Torvalds <torvalds@osdl.org>
2006-01-08 02:03:32 -07:00
|
|
|
struct dentry *d = list_entry(node, struct dentry, d_u.d_child);
|
2005-04-16 16:20:36 -06:00
|
|
|
list_del_init(node);
|
|
|
|
if (d->d_inode) {
|
|
|
|
d = dget_locked(d);
|
|
|
|
spin_unlock(&dcache_lock);
|
|
|
|
d_delete(d);
|
|
|
|
simple_unlink(dentry->d_inode, d);
|
|
|
|
dput(d);
|
|
|
|
spin_lock(&dcache_lock);
|
|
|
|
}
|
|
|
|
node = dentry->d_subdirs.next;
|
|
|
|
}
|
[PATCH] shrink dentry struct
Some long time ago, dentry struct was carefully tuned so that on 32 bits
UP, sizeof(struct dentry) was exactly 128, ie a power of 2, and a multiple
of memory cache lines.
Then RCU was added and dentry struct enlarged by two pointers, with nice
results for SMP, but not so good on UP, because breaking the above tuning
(128 + 8 = 136 bytes)
This patch reverts this unwanted side effect, by using an union (d_u),
where d_rcu and d_child are placed so that these two fields can share their
memory needs.
At the time d_free() is called (and d_rcu is really used), d_child is known
to be empty and not touched by the dentry freeing.
Lockless lookups only access d_name, d_parent, d_lock, d_op, d_flags (so
the previous content of d_child is not needed if said dentry was unhashed
but still accessed by a CPU because of RCU constraints)
As dentry cache easily contains millions of entries, a size reduction is
worth the extra complexity of the ugly C union.
Signed-off-by: Eric Dumazet <dada1@cosmosbay.com>
Cc: Dipankar Sarma <dipankar@in.ibm.com>
Cc: Maneesh Soni <maneesh@in.ibm.com>
Cc: Miklos Szeredi <miklos@szeredi.hu>
Cc: "Paul E. McKenney" <paulmck@us.ibm.com>
Cc: Ian Kent <raven@themaw.net>
Cc: Paul Jackson <pj@sgi.com>
Cc: Al Viro <viro@ftp.linux.org.uk>
Cc: Christoph Hellwig <hch@lst.de>
Cc: Trond Myklebust <trond.myklebust@fys.uio.no>
Cc: Neil Brown <neilb@cse.unsw.edu.au>
Cc: James Morris <jmorris@namei.org>
Cc: Stephen Smalley <sds@epoch.ncsc.mil>
Signed-off-by: Andrew Morton <akpm@osdl.org>
Signed-off-by: Linus Torvalds <torvalds@osdl.org>
2006-01-08 02:03:32 -07:00
|
|
|
list_del_init(&dentry->d_u.d_child);
|
2005-04-16 16:20:36 -06:00
|
|
|
spin_unlock(&dcache_lock);
|
|
|
|
remove_dir(dentry);
|
|
|
|
}
|
|
|
|
|
|
|
|
static struct super_operations cpuset_ops = {
|
|
|
|
.statfs = simple_statfs,
|
|
|
|
.drop_inode = generic_delete_inode,
|
|
|
|
};
|
|
|
|
|
|
|
|
static int cpuset_fill_super(struct super_block *sb, void *unused_data,
|
|
|
|
int unused_silent)
|
|
|
|
{
|
|
|
|
struct inode *inode;
|
|
|
|
struct dentry *root;
|
|
|
|
|
|
|
|
sb->s_blocksize = PAGE_CACHE_SIZE;
|
|
|
|
sb->s_blocksize_bits = PAGE_CACHE_SHIFT;
|
|
|
|
sb->s_magic = CPUSET_SUPER_MAGIC;
|
|
|
|
sb->s_op = &cpuset_ops;
|
|
|
|
cpuset_sb = sb;
|
|
|
|
|
|
|
|
inode = cpuset_new_inode(S_IFDIR | S_IRUGO | S_IXUGO | S_IWUSR);
|
|
|
|
if (inode) {
|
|
|
|
inode->i_op = &simple_dir_inode_operations;
|
|
|
|
inode->i_fop = &simple_dir_operations;
|
|
|
|
/* directories start off with i_nlink == 2 (for "." entry) */
|
|
|
|
inode->i_nlink++;
|
|
|
|
} else {
|
|
|
|
return -ENOMEM;
|
|
|
|
}
|
|
|
|
|
|
|
|
root = d_alloc_root(inode);
|
|
|
|
if (!root) {
|
|
|
|
iput(inode);
|
|
|
|
return -ENOMEM;
|
|
|
|
}
|
|
|
|
sb->s_root = root;
|
|
|
|
return 0;
|
|
|
|
}
|
|
|
|
|
[PATCH] VFS: Permit filesystem to override root dentry on mount
Extend the get_sb() filesystem operation to take an extra argument that
permits the VFS to pass in the target vfsmount that defines the mountpoint.
The filesystem is then required to manually set the superblock and root dentry
pointers. For most filesystems, this should be done with simple_set_mnt()
which will set the superblock pointer and then set the root dentry to the
superblock's s_root (as per the old default behaviour).
The get_sb() op now returns an integer as there's now no need to return the
superblock pointer.
This patch permits a superblock to be implicitly shared amongst several mount
points, such as can be done with NFS to avoid potential inode aliasing. In
such a case, simple_set_mnt() would not be called, and instead the mnt_root
and mnt_sb would be set directly.
The patch also makes the following changes:
(*) the get_sb_*() convenience functions in the core kernel now take a vfsmount
pointer argument and return an integer, so most filesystems have to change
very little.
(*) If one of the convenience function is not used, then get_sb() should
normally call simple_set_mnt() to instantiate the vfsmount. This will
always return 0, and so can be tail-called from get_sb().
(*) generic_shutdown_super() now calls shrink_dcache_sb() to clean up the
dcache upon superblock destruction rather than shrink_dcache_anon().
This is required because the superblock may now have multiple trees that
aren't actually bound to s_root, but that still need to be cleaned up. The
currently called functions assume that the whole tree is rooted at s_root,
and that anonymous dentries are not the roots of trees which results in
dentries being left unculled.
However, with the way NFS superblock sharing are currently set to be
implemented, these assumptions are violated: the root of the filesystem is
simply a dummy dentry and inode (the real inode for '/' may well be
inaccessible), and all the vfsmounts are rooted on anonymous[*] dentries
with child trees.
[*] Anonymous until discovered from another tree.
(*) The documentation has been adjusted, including the additional bit of
changing ext2_* into foo_* in the documentation.
[akpm@osdl.org: convert ipath_fs, do other stuff]
Signed-off-by: David Howells <dhowells@redhat.com>
Acked-by: Al Viro <viro@zeniv.linux.org.uk>
Cc: Nathan Scott <nathans@sgi.com>
Cc: Roland Dreier <rolandd@cisco.com>
Signed-off-by: Andrew Morton <akpm@osdl.org>
Signed-off-by: Linus Torvalds <torvalds@osdl.org>
2006-06-23 03:02:57 -06:00
|
|
|
static int cpuset_get_sb(struct file_system_type *fs_type,
|
|
|
|
int flags, const char *unused_dev_name,
|
|
|
|
void *data, struct vfsmount *mnt)
|
2005-04-16 16:20:36 -06:00
|
|
|
{
|
[PATCH] VFS: Permit filesystem to override root dentry on mount
Extend the get_sb() filesystem operation to take an extra argument that
permits the VFS to pass in the target vfsmount that defines the mountpoint.
The filesystem is then required to manually set the superblock and root dentry
pointers. For most filesystems, this should be done with simple_set_mnt()
which will set the superblock pointer and then set the root dentry to the
superblock's s_root (as per the old default behaviour).
The get_sb() op now returns an integer as there's now no need to return the
superblock pointer.
This patch permits a superblock to be implicitly shared amongst several mount
points, such as can be done with NFS to avoid potential inode aliasing. In
such a case, simple_set_mnt() would not be called, and instead the mnt_root
and mnt_sb would be set directly.
The patch also makes the following changes:
(*) the get_sb_*() convenience functions in the core kernel now take a vfsmount
pointer argument and return an integer, so most filesystems have to change
very little.
(*) If one of the convenience function is not used, then get_sb() should
normally call simple_set_mnt() to instantiate the vfsmount. This will
always return 0, and so can be tail-called from get_sb().
(*) generic_shutdown_super() now calls shrink_dcache_sb() to clean up the
dcache upon superblock destruction rather than shrink_dcache_anon().
This is required because the superblock may now have multiple trees that
aren't actually bound to s_root, but that still need to be cleaned up. The
currently called functions assume that the whole tree is rooted at s_root,
and that anonymous dentries are not the roots of trees which results in
dentries being left unculled.
However, with the way NFS superblock sharing are currently set to be
implemented, these assumptions are violated: the root of the filesystem is
simply a dummy dentry and inode (the real inode for '/' may well be
inaccessible), and all the vfsmounts are rooted on anonymous[*] dentries
with child trees.
[*] Anonymous until discovered from another tree.
(*) The documentation has been adjusted, including the additional bit of
changing ext2_* into foo_* in the documentation.
[akpm@osdl.org: convert ipath_fs, do other stuff]
Signed-off-by: David Howells <dhowells@redhat.com>
Acked-by: Al Viro <viro@zeniv.linux.org.uk>
Cc: Nathan Scott <nathans@sgi.com>
Cc: Roland Dreier <rolandd@cisco.com>
Signed-off-by: Andrew Morton <akpm@osdl.org>
Signed-off-by: Linus Torvalds <torvalds@osdl.org>
2006-06-23 03:02:57 -06:00
|
|
|
return get_sb_single(fs_type, flags, data, cpuset_fill_super, mnt);
|
2005-04-16 16:20:36 -06:00
|
|
|
}
|
|
|
|
|
|
|
|
static struct file_system_type cpuset_fs_type = {
|
|
|
|
.name = "cpuset",
|
|
|
|
.get_sb = cpuset_get_sb,
|
|
|
|
.kill_sb = kill_litter_super,
|
|
|
|
};
|
|
|
|
|
|
|
|
/* struct cftype:
|
|
|
|
*
|
|
|
|
* The files in the cpuset filesystem mostly have a very simple read/write
|
|
|
|
* handling, some common function will take care of it. Nevertheless some cases
|
|
|
|
* (read tasks) are special and therefore I define this structure for every
|
|
|
|
* kind of file.
|
|
|
|
*
|
|
|
|
*
|
|
|
|
* When reading/writing to a file:
|
|
|
|
* - the cpuset to use in file->f_dentry->d_parent->d_fsdata
|
|
|
|
* - the 'cftype' of the file is file->f_dentry->d_fsdata
|
|
|
|
*/
|
|
|
|
|
|
|
|
struct cftype {
|
|
|
|
char *name;
|
|
|
|
int private;
|
|
|
|
int (*open) (struct inode *inode, struct file *file);
|
|
|
|
ssize_t (*read) (struct file *file, char __user *buf, size_t nbytes,
|
|
|
|
loff_t *ppos);
|
|
|
|
int (*write) (struct file *file, const char __user *buf, size_t nbytes,
|
|
|
|
loff_t *ppos);
|
|
|
|
int (*release) (struct inode *inode, struct file *file);
|
|
|
|
};
|
|
|
|
|
|
|
|
static inline struct cpuset *__d_cs(struct dentry *dentry)
|
|
|
|
{
|
|
|
|
return dentry->d_fsdata;
|
|
|
|
}
|
|
|
|
|
|
|
|
static inline struct cftype *__d_cft(struct dentry *dentry)
|
|
|
|
{
|
|
|
|
return dentry->d_fsdata;
|
|
|
|
}
|
|
|
|
|
|
|
|
/*
|
2006-03-23 04:00:18 -07:00
|
|
|
* Call with manage_mutex held. Writes path of cpuset into buf.
|
2005-04-16 16:20:36 -06:00
|
|
|
* Returns 0 on success, -errno on error.
|
|
|
|
*/
|
|
|
|
|
|
|
|
static int cpuset_path(const struct cpuset *cs, char *buf, int buflen)
|
|
|
|
{
|
|
|
|
char *start;
|
|
|
|
|
|
|
|
start = buf + buflen;
|
|
|
|
|
|
|
|
*--start = '\0';
|
|
|
|
for (;;) {
|
|
|
|
int len = cs->dentry->d_name.len;
|
|
|
|
if ((start -= len) < buf)
|
|
|
|
return -ENAMETOOLONG;
|
|
|
|
memcpy(start, cs->dentry->d_name.name, len);
|
|
|
|
cs = cs->parent;
|
|
|
|
if (!cs)
|
|
|
|
break;
|
|
|
|
if (!cs->parent)
|
|
|
|
continue;
|
|
|
|
if (--start < buf)
|
|
|
|
return -ENAMETOOLONG;
|
|
|
|
*start = '/';
|
|
|
|
}
|
|
|
|
memmove(buf, start, buf + buflen - start);
|
|
|
|
return 0;
|
|
|
|
}
|
|
|
|
|
|
|
|
/*
|
|
|
|
* Notify userspace when a cpuset is released, by running
|
|
|
|
* /sbin/cpuset_release_agent with the name of the cpuset (path
|
|
|
|
* relative to the root of cpuset file system) as the argument.
|
|
|
|
*
|
|
|
|
* Most likely, this user command will try to rmdir this cpuset.
|
|
|
|
*
|
|
|
|
* This races with the possibility that some other task will be
|
|
|
|
* attached to this cpuset before it is removed, or that some other
|
|
|
|
* user task will 'mkdir' a child cpuset of this cpuset. That's ok.
|
|
|
|
* The presumed 'rmdir' will fail quietly if this cpuset is no longer
|
|
|
|
* unused, and this cpuset will be reprieved from its death sentence,
|
|
|
|
* to continue to serve a useful existence. Next time it's released,
|
|
|
|
* we will get notified again, if it still has 'notify_on_release' set.
|
|
|
|
*
|
[PATCH] cpuset release ABBA deadlock fix
Fix possible cpuset_sem ABBA deadlock if 'notify_on_release' set.
For a particular usage pattern, creating and destroying cpusets fairly
frequently using notify_on_release, on a very large system, this deadlock
can be seen every few days. If you are not using the cpuset
notify_on_release feature, you will never see this deadlock.
The existing code, on task exit (or cpuset deletion) did:
get cpuset_sem
if cpuset marked notify_on_release and is ready to release:
compute cpuset path relative to /dev/cpuset mount point
call_usermodehelper() forks /sbin/cpuset_release_agent with path
drop cpuset_sem
Unfortunately, the fork in call_usermodehelper can allocate memory, and
allocating memory can require cpuset_sem, if the mems_generation values
changed in the interim. This results in an ABBA deadlock, trying to obtain
cpuset_sem when it is already held by the current task.
To fix this, I put the cpuset path (which must be computed while holding
cpuset_sem) in a temporary buffer, to be used in the call_usermodehelper
call of /sbin/cpuset_release_agent only _after_ dropping cpuset_sem.
So the new logic is:
get cpuset_sem
if cpuset marked notify_on_release and is ready to release:
compute cpuset path relative to /dev/cpuset mount point
stash path in kmalloc'd buffer
drop cpuset_sem
call_usermodehelper() forks /sbin/cpuset_release_agent with path
free path
The sharp eyed reader might notice that this patch does not contain any
calls to kmalloc. The existing code in the check_for_release() routine was
already kmalloc'ing a buffer to hold the cpuset path. In the old code, it
just held the buffer for a few lines, over the cpuset_release_agent() call
that in turn invoked call_usermodehelper(). In the new code, with the
application of this patch, it returns that buffer via the new char
**ppathbuf parameter, for later use and freeing in cpuset_release_agent(),
which is called after cpuset_sem is dropped. Whereas the old code has just
one call to cpuset_release_agent(), right in the check_for_release()
routine, the new code has three calls to cpuset_release_agent(), from the
various places that a cpuset can be released.
This patch has been build and booted on SN2, and passed a stress test that
previously hit the deadlock within a few seconds.
Signed-off-by: Paul Jackson <pj@sgi.com>
Signed-off-by: Andrew Morton <akpm@osdl.org>
Signed-off-by: Linus Torvalds <torvalds@osdl.org>
2005-08-09 11:07:59 -06:00
|
|
|
* The final arg to call_usermodehelper() is 0, which means don't
|
|
|
|
* wait. The separate /sbin/cpuset_release_agent task is forked by
|
|
|
|
* call_usermodehelper(), then control in this thread returns here,
|
|
|
|
* without waiting for the release agent task. We don't bother to
|
|
|
|
* wait because the caller of this routine has no use for the exit
|
|
|
|
* status of the /sbin/cpuset_release_agent task, so no sense holding
|
|
|
|
* our caller up for that.
|
|
|
|
*
|
2006-03-23 04:00:18 -07:00
|
|
|
* When we had only one cpuset mutex, we had to call this
|
[PATCH] cpusets: dual semaphore locking overhaul
Overhaul cpuset locking. Replace single semaphore with two semaphores.
The suggestion to use two locks was made by Roman Zippel.
Both locks are global. Code that wants to modify cpusets must first
acquire the exclusive manage_sem, which allows them read-only access to
cpusets, and holds off other would-be modifiers. Before making actual
changes, the second semaphore, callback_sem must be acquired as well. Code
that needs only to query cpusets must acquire callback_sem, which is also a
global exclusive lock.
The earlier problems with double tripping are avoided, because it is
allowed for holders of manage_sem to nest the second callback_sem lock, and
only callback_sem is needed by code called from within __alloc_pages(),
where the double tripping had been possible.
This is not quite the same as a normal read/write semaphore, because
obtaining read-only access with intent to change must hold off other such
attempts, while allowing read-only access w/o such intention. Changing
cpusets involves several related checks and changes, which must be done
while allowing read-only queries (to avoid the double trip), but while
ensuring nothing changes (holding off other would be modifiers.)
This overhaul of cpuset locking also makes careful use of task_lock() to
guard access to the task->cpuset pointer, closing a couple of race
conditions noticed while reading this code (thanks, Roman). I've never
seen these races fail in any use or test.
See further the comments in the code.
Signed-off-by: Paul Jackson <pj@sgi.com>
Signed-off-by: Andrew Morton <akpm@osdl.org>
Signed-off-by: Linus Torvalds <torvalds@osdl.org>
2005-10-30 16:02:30 -07:00
|
|
|
* without holding it, to avoid deadlock when call_usermodehelper()
|
|
|
|
* allocated memory. With two locks, we could now call this while
|
2006-03-23 04:00:18 -07:00
|
|
|
* holding manage_mutex, but we still don't, so as to minimize
|
|
|
|
* the time manage_mutex is held.
|
2005-04-16 16:20:36 -06:00
|
|
|
*/
|
|
|
|
|
[PATCH] cpuset release ABBA deadlock fix
Fix possible cpuset_sem ABBA deadlock if 'notify_on_release' set.
For a particular usage pattern, creating and destroying cpusets fairly
frequently using notify_on_release, on a very large system, this deadlock
can be seen every few days. If you are not using the cpuset
notify_on_release feature, you will never see this deadlock.
The existing code, on task exit (or cpuset deletion) did:
get cpuset_sem
if cpuset marked notify_on_release and is ready to release:
compute cpuset path relative to /dev/cpuset mount point
call_usermodehelper() forks /sbin/cpuset_release_agent with path
drop cpuset_sem
Unfortunately, the fork in call_usermodehelper can allocate memory, and
allocating memory can require cpuset_sem, if the mems_generation values
changed in the interim. This results in an ABBA deadlock, trying to obtain
cpuset_sem when it is already held by the current task.
To fix this, I put the cpuset path (which must be computed while holding
cpuset_sem) in a temporary buffer, to be used in the call_usermodehelper
call of /sbin/cpuset_release_agent only _after_ dropping cpuset_sem.
So the new logic is:
get cpuset_sem
if cpuset marked notify_on_release and is ready to release:
compute cpuset path relative to /dev/cpuset mount point
stash path in kmalloc'd buffer
drop cpuset_sem
call_usermodehelper() forks /sbin/cpuset_release_agent with path
free path
The sharp eyed reader might notice that this patch does not contain any
calls to kmalloc. The existing code in the check_for_release() routine was
already kmalloc'ing a buffer to hold the cpuset path. In the old code, it
just held the buffer for a few lines, over the cpuset_release_agent() call
that in turn invoked call_usermodehelper(). In the new code, with the
application of this patch, it returns that buffer via the new char
**ppathbuf parameter, for later use and freeing in cpuset_release_agent(),
which is called after cpuset_sem is dropped. Whereas the old code has just
one call to cpuset_release_agent(), right in the check_for_release()
routine, the new code has three calls to cpuset_release_agent(), from the
various places that a cpuset can be released.
This patch has been build and booted on SN2, and passed a stress test that
previously hit the deadlock within a few seconds.
Signed-off-by: Paul Jackson <pj@sgi.com>
Signed-off-by: Andrew Morton <akpm@osdl.org>
Signed-off-by: Linus Torvalds <torvalds@osdl.org>
2005-08-09 11:07:59 -06:00
|
|
|
static void cpuset_release_agent(const char *pathbuf)
|
2005-04-16 16:20:36 -06:00
|
|
|
{
|
|
|
|
char *argv[3], *envp[3];
|
|
|
|
int i;
|
|
|
|
|
[PATCH] cpuset release ABBA deadlock fix
Fix possible cpuset_sem ABBA deadlock if 'notify_on_release' set.
For a particular usage pattern, creating and destroying cpusets fairly
frequently using notify_on_release, on a very large system, this deadlock
can be seen every few days. If you are not using the cpuset
notify_on_release feature, you will never see this deadlock.
The existing code, on task exit (or cpuset deletion) did:
get cpuset_sem
if cpuset marked notify_on_release and is ready to release:
compute cpuset path relative to /dev/cpuset mount point
call_usermodehelper() forks /sbin/cpuset_release_agent with path
drop cpuset_sem
Unfortunately, the fork in call_usermodehelper can allocate memory, and
allocating memory can require cpuset_sem, if the mems_generation values
changed in the interim. This results in an ABBA deadlock, trying to obtain
cpuset_sem when it is already held by the current task.
To fix this, I put the cpuset path (which must be computed while holding
cpuset_sem) in a temporary buffer, to be used in the call_usermodehelper
call of /sbin/cpuset_release_agent only _after_ dropping cpuset_sem.
So the new logic is:
get cpuset_sem
if cpuset marked notify_on_release and is ready to release:
compute cpuset path relative to /dev/cpuset mount point
stash path in kmalloc'd buffer
drop cpuset_sem
call_usermodehelper() forks /sbin/cpuset_release_agent with path
free path
The sharp eyed reader might notice that this patch does not contain any
calls to kmalloc. The existing code in the check_for_release() routine was
already kmalloc'ing a buffer to hold the cpuset path. In the old code, it
just held the buffer for a few lines, over the cpuset_release_agent() call
that in turn invoked call_usermodehelper(). In the new code, with the
application of this patch, it returns that buffer via the new char
**ppathbuf parameter, for later use and freeing in cpuset_release_agent(),
which is called after cpuset_sem is dropped. Whereas the old code has just
one call to cpuset_release_agent(), right in the check_for_release()
routine, the new code has three calls to cpuset_release_agent(), from the
various places that a cpuset can be released.
This patch has been build and booted on SN2, and passed a stress test that
previously hit the deadlock within a few seconds.
Signed-off-by: Paul Jackson <pj@sgi.com>
Signed-off-by: Andrew Morton <akpm@osdl.org>
Signed-off-by: Linus Torvalds <torvalds@osdl.org>
2005-08-09 11:07:59 -06:00
|
|
|
if (!pathbuf)
|
|
|
|
return;
|
|
|
|
|
2005-04-16 16:20:36 -06:00
|
|
|
i = 0;
|
|
|
|
argv[i++] = "/sbin/cpuset_release_agent";
|
[PATCH] cpuset release ABBA deadlock fix
Fix possible cpuset_sem ABBA deadlock if 'notify_on_release' set.
For a particular usage pattern, creating and destroying cpusets fairly
frequently using notify_on_release, on a very large system, this deadlock
can be seen every few days. If you are not using the cpuset
notify_on_release feature, you will never see this deadlock.
The existing code, on task exit (or cpuset deletion) did:
get cpuset_sem
if cpuset marked notify_on_release and is ready to release:
compute cpuset path relative to /dev/cpuset mount point
call_usermodehelper() forks /sbin/cpuset_release_agent with path
drop cpuset_sem
Unfortunately, the fork in call_usermodehelper can allocate memory, and
allocating memory can require cpuset_sem, if the mems_generation values
changed in the interim. This results in an ABBA deadlock, trying to obtain
cpuset_sem when it is already held by the current task.
To fix this, I put the cpuset path (which must be computed while holding
cpuset_sem) in a temporary buffer, to be used in the call_usermodehelper
call of /sbin/cpuset_release_agent only _after_ dropping cpuset_sem.
So the new logic is:
get cpuset_sem
if cpuset marked notify_on_release and is ready to release:
compute cpuset path relative to /dev/cpuset mount point
stash path in kmalloc'd buffer
drop cpuset_sem
call_usermodehelper() forks /sbin/cpuset_release_agent with path
free path
The sharp eyed reader might notice that this patch does not contain any
calls to kmalloc. The existing code in the check_for_release() routine was
already kmalloc'ing a buffer to hold the cpuset path. In the old code, it
just held the buffer for a few lines, over the cpuset_release_agent() call
that in turn invoked call_usermodehelper(). In the new code, with the
application of this patch, it returns that buffer via the new char
**ppathbuf parameter, for later use and freeing in cpuset_release_agent(),
which is called after cpuset_sem is dropped. Whereas the old code has just
one call to cpuset_release_agent(), right in the check_for_release()
routine, the new code has three calls to cpuset_release_agent(), from the
various places that a cpuset can be released.
This patch has been build and booted on SN2, and passed a stress test that
previously hit the deadlock within a few seconds.
Signed-off-by: Paul Jackson <pj@sgi.com>
Signed-off-by: Andrew Morton <akpm@osdl.org>
Signed-off-by: Linus Torvalds <torvalds@osdl.org>
2005-08-09 11:07:59 -06:00
|
|
|
argv[i++] = (char *)pathbuf;
|
2005-04-16 16:20:36 -06:00
|
|
|
argv[i] = NULL;
|
|
|
|
|
|
|
|
i = 0;
|
|
|
|
/* minimal command environment */
|
|
|
|
envp[i++] = "HOME=/";
|
|
|
|
envp[i++] = "PATH=/sbin:/bin:/usr/sbin:/usr/bin";
|
|
|
|
envp[i] = NULL;
|
|
|
|
|
[PATCH] cpuset release ABBA deadlock fix
Fix possible cpuset_sem ABBA deadlock if 'notify_on_release' set.
For a particular usage pattern, creating and destroying cpusets fairly
frequently using notify_on_release, on a very large system, this deadlock
can be seen every few days. If you are not using the cpuset
notify_on_release feature, you will never see this deadlock.
The existing code, on task exit (or cpuset deletion) did:
get cpuset_sem
if cpuset marked notify_on_release and is ready to release:
compute cpuset path relative to /dev/cpuset mount point
call_usermodehelper() forks /sbin/cpuset_release_agent with path
drop cpuset_sem
Unfortunately, the fork in call_usermodehelper can allocate memory, and
allocating memory can require cpuset_sem, if the mems_generation values
changed in the interim. This results in an ABBA deadlock, trying to obtain
cpuset_sem when it is already held by the current task.
To fix this, I put the cpuset path (which must be computed while holding
cpuset_sem) in a temporary buffer, to be used in the call_usermodehelper
call of /sbin/cpuset_release_agent only _after_ dropping cpuset_sem.
So the new logic is:
get cpuset_sem
if cpuset marked notify_on_release and is ready to release:
compute cpuset path relative to /dev/cpuset mount point
stash path in kmalloc'd buffer
drop cpuset_sem
call_usermodehelper() forks /sbin/cpuset_release_agent with path
free path
The sharp eyed reader might notice that this patch does not contain any
calls to kmalloc. The existing code in the check_for_release() routine was
already kmalloc'ing a buffer to hold the cpuset path. In the old code, it
just held the buffer for a few lines, over the cpuset_release_agent() call
that in turn invoked call_usermodehelper(). In the new code, with the
application of this patch, it returns that buffer via the new char
**ppathbuf parameter, for later use and freeing in cpuset_release_agent(),
which is called after cpuset_sem is dropped. Whereas the old code has just
one call to cpuset_release_agent(), right in the check_for_release()
routine, the new code has three calls to cpuset_release_agent(), from the
various places that a cpuset can be released.
This patch has been build and booted on SN2, and passed a stress test that
previously hit the deadlock within a few seconds.
Signed-off-by: Paul Jackson <pj@sgi.com>
Signed-off-by: Andrew Morton <akpm@osdl.org>
Signed-off-by: Linus Torvalds <torvalds@osdl.org>
2005-08-09 11:07:59 -06:00
|
|
|
call_usermodehelper(argv[0], argv, envp, 0);
|
|
|
|
kfree(pathbuf);
|
2005-04-16 16:20:36 -06:00
|
|
|
}
|
|
|
|
|
|
|
|
/*
|
|
|
|
* Either cs->count of using tasks transitioned to zero, or the
|
|
|
|
* cs->children list of child cpusets just became empty. If this
|
|
|
|
* cs is notify_on_release() and now both the user count is zero and
|
[PATCH] cpuset release ABBA deadlock fix
Fix possible cpuset_sem ABBA deadlock if 'notify_on_release' set.
For a particular usage pattern, creating and destroying cpusets fairly
frequently using notify_on_release, on a very large system, this deadlock
can be seen every few days. If you are not using the cpuset
notify_on_release feature, you will never see this deadlock.
The existing code, on task exit (or cpuset deletion) did:
get cpuset_sem
if cpuset marked notify_on_release and is ready to release:
compute cpuset path relative to /dev/cpuset mount point
call_usermodehelper() forks /sbin/cpuset_release_agent with path
drop cpuset_sem
Unfortunately, the fork in call_usermodehelper can allocate memory, and
allocating memory can require cpuset_sem, if the mems_generation values
changed in the interim. This results in an ABBA deadlock, trying to obtain
cpuset_sem when it is already held by the current task.
To fix this, I put the cpuset path (which must be computed while holding
cpuset_sem) in a temporary buffer, to be used in the call_usermodehelper
call of /sbin/cpuset_release_agent only _after_ dropping cpuset_sem.
So the new logic is:
get cpuset_sem
if cpuset marked notify_on_release and is ready to release:
compute cpuset path relative to /dev/cpuset mount point
stash path in kmalloc'd buffer
drop cpuset_sem
call_usermodehelper() forks /sbin/cpuset_release_agent with path
free path
The sharp eyed reader might notice that this patch does not contain any
calls to kmalloc. The existing code in the check_for_release() routine was
already kmalloc'ing a buffer to hold the cpuset path. In the old code, it
just held the buffer for a few lines, over the cpuset_release_agent() call
that in turn invoked call_usermodehelper(). In the new code, with the
application of this patch, it returns that buffer via the new char
**ppathbuf parameter, for later use and freeing in cpuset_release_agent(),
which is called after cpuset_sem is dropped. Whereas the old code has just
one call to cpuset_release_agent(), right in the check_for_release()
routine, the new code has three calls to cpuset_release_agent(), from the
various places that a cpuset can be released.
This patch has been build and booted on SN2, and passed a stress test that
previously hit the deadlock within a few seconds.
Signed-off-by: Paul Jackson <pj@sgi.com>
Signed-off-by: Andrew Morton <akpm@osdl.org>
Signed-off-by: Linus Torvalds <torvalds@osdl.org>
2005-08-09 11:07:59 -06:00
|
|
|
* the list of children is empty, prepare cpuset path in a kmalloc'd
|
|
|
|
* buffer, to be returned via ppathbuf, so that the caller can invoke
|
2006-03-23 04:00:18 -07:00
|
|
|
* cpuset_release_agent() with it later on, once manage_mutex is dropped.
|
|
|
|
* Call here with manage_mutex held.
|
[PATCH] cpuset release ABBA deadlock fix
Fix possible cpuset_sem ABBA deadlock if 'notify_on_release' set.
For a particular usage pattern, creating and destroying cpusets fairly
frequently using notify_on_release, on a very large system, this deadlock
can be seen every few days. If you are not using the cpuset
notify_on_release feature, you will never see this deadlock.
The existing code, on task exit (or cpuset deletion) did:
get cpuset_sem
if cpuset marked notify_on_release and is ready to release:
compute cpuset path relative to /dev/cpuset mount point
call_usermodehelper() forks /sbin/cpuset_release_agent with path
drop cpuset_sem
Unfortunately, the fork in call_usermodehelper can allocate memory, and
allocating memory can require cpuset_sem, if the mems_generation values
changed in the interim. This results in an ABBA deadlock, trying to obtain
cpuset_sem when it is already held by the current task.
To fix this, I put the cpuset path (which must be computed while holding
cpuset_sem) in a temporary buffer, to be used in the call_usermodehelper
call of /sbin/cpuset_release_agent only _after_ dropping cpuset_sem.
So the new logic is:
get cpuset_sem
if cpuset marked notify_on_release and is ready to release:
compute cpuset path relative to /dev/cpuset mount point
stash path in kmalloc'd buffer
drop cpuset_sem
call_usermodehelper() forks /sbin/cpuset_release_agent with path
free path
The sharp eyed reader might notice that this patch does not contain any
calls to kmalloc. The existing code in the check_for_release() routine was
already kmalloc'ing a buffer to hold the cpuset path. In the old code, it
just held the buffer for a few lines, over the cpuset_release_agent() call
that in turn invoked call_usermodehelper(). In the new code, with the
application of this patch, it returns that buffer via the new char
**ppathbuf parameter, for later use and freeing in cpuset_release_agent(),
which is called after cpuset_sem is dropped. Whereas the old code has just
one call to cpuset_release_agent(), right in the check_for_release()
routine, the new code has three calls to cpuset_release_agent(), from the
various places that a cpuset can be released.
This patch has been build and booted on SN2, and passed a stress test that
previously hit the deadlock within a few seconds.
Signed-off-by: Paul Jackson <pj@sgi.com>
Signed-off-by: Andrew Morton <akpm@osdl.org>
Signed-off-by: Linus Torvalds <torvalds@osdl.org>
2005-08-09 11:07:59 -06:00
|
|
|
*
|
|
|
|
* This check_for_release() routine is responsible for kmalloc'ing
|
|
|
|
* pathbuf. The above cpuset_release_agent() is responsible for
|
|
|
|
* kfree'ing pathbuf. The caller of these routines is responsible
|
|
|
|
* for providing a pathbuf pointer, initialized to NULL, then
|
2006-03-23 04:00:18 -07:00
|
|
|
* calling check_for_release() with manage_mutex held and the address
|
|
|
|
* of the pathbuf pointer, then dropping manage_mutex, then calling
|
[PATCH] cpuset release ABBA deadlock fix
Fix possible cpuset_sem ABBA deadlock if 'notify_on_release' set.
For a particular usage pattern, creating and destroying cpusets fairly
frequently using notify_on_release, on a very large system, this deadlock
can be seen every few days. If you are not using the cpuset
notify_on_release feature, you will never see this deadlock.
The existing code, on task exit (or cpuset deletion) did:
get cpuset_sem
if cpuset marked notify_on_release and is ready to release:
compute cpuset path relative to /dev/cpuset mount point
call_usermodehelper() forks /sbin/cpuset_release_agent with path
drop cpuset_sem
Unfortunately, the fork in call_usermodehelper can allocate memory, and
allocating memory can require cpuset_sem, if the mems_generation values
changed in the interim. This results in an ABBA deadlock, trying to obtain
cpuset_sem when it is already held by the current task.
To fix this, I put the cpuset path (which must be computed while holding
cpuset_sem) in a temporary buffer, to be used in the call_usermodehelper
call of /sbin/cpuset_release_agent only _after_ dropping cpuset_sem.
So the new logic is:
get cpuset_sem
if cpuset marked notify_on_release and is ready to release:
compute cpuset path relative to /dev/cpuset mount point
stash path in kmalloc'd buffer
drop cpuset_sem
call_usermodehelper() forks /sbin/cpuset_release_agent with path
free path
The sharp eyed reader might notice that this patch does not contain any
calls to kmalloc. The existing code in the check_for_release() routine was
already kmalloc'ing a buffer to hold the cpuset path. In the old code, it
just held the buffer for a few lines, over the cpuset_release_agent() call
that in turn invoked call_usermodehelper(). In the new code, with the
application of this patch, it returns that buffer via the new char
**ppathbuf parameter, for later use and freeing in cpuset_release_agent(),
which is called after cpuset_sem is dropped. Whereas the old code has just
one call to cpuset_release_agent(), right in the check_for_release()
routine, the new code has three calls to cpuset_release_agent(), from the
various places that a cpuset can be released.
This patch has been build and booted on SN2, and passed a stress test that
previously hit the deadlock within a few seconds.
Signed-off-by: Paul Jackson <pj@sgi.com>
Signed-off-by: Andrew Morton <akpm@osdl.org>
Signed-off-by: Linus Torvalds <torvalds@osdl.org>
2005-08-09 11:07:59 -06:00
|
|
|
* cpuset_release_agent() with pathbuf, as set by check_for_release().
|
2005-04-16 16:20:36 -06:00
|
|
|
*/
|
|
|
|
|
[PATCH] cpuset release ABBA deadlock fix
Fix possible cpuset_sem ABBA deadlock if 'notify_on_release' set.
For a particular usage pattern, creating and destroying cpusets fairly
frequently using notify_on_release, on a very large system, this deadlock
can be seen every few days. If you are not using the cpuset
notify_on_release feature, you will never see this deadlock.
The existing code, on task exit (or cpuset deletion) did:
get cpuset_sem
if cpuset marked notify_on_release and is ready to release:
compute cpuset path relative to /dev/cpuset mount point
call_usermodehelper() forks /sbin/cpuset_release_agent with path
drop cpuset_sem
Unfortunately, the fork in call_usermodehelper can allocate memory, and
allocating memory can require cpuset_sem, if the mems_generation values
changed in the interim. This results in an ABBA deadlock, trying to obtain
cpuset_sem when it is already held by the current task.
To fix this, I put the cpuset path (which must be computed while holding
cpuset_sem) in a temporary buffer, to be used in the call_usermodehelper
call of /sbin/cpuset_release_agent only _after_ dropping cpuset_sem.
So the new logic is:
get cpuset_sem
if cpuset marked notify_on_release and is ready to release:
compute cpuset path relative to /dev/cpuset mount point
stash path in kmalloc'd buffer
drop cpuset_sem
call_usermodehelper() forks /sbin/cpuset_release_agent with path
free path
The sharp eyed reader might notice that this patch does not contain any
calls to kmalloc. The existing code in the check_for_release() routine was
already kmalloc'ing a buffer to hold the cpuset path. In the old code, it
just held the buffer for a few lines, over the cpuset_release_agent() call
that in turn invoked call_usermodehelper(). In the new code, with the
application of this patch, it returns that buffer via the new char
**ppathbuf parameter, for later use and freeing in cpuset_release_agent(),
which is called after cpuset_sem is dropped. Whereas the old code has just
one call to cpuset_release_agent(), right in the check_for_release()
routine, the new code has three calls to cpuset_release_agent(), from the
various places that a cpuset can be released.
This patch has been build and booted on SN2, and passed a stress test that
previously hit the deadlock within a few seconds.
Signed-off-by: Paul Jackson <pj@sgi.com>
Signed-off-by: Andrew Morton <akpm@osdl.org>
Signed-off-by: Linus Torvalds <torvalds@osdl.org>
2005-08-09 11:07:59 -06:00
|
|
|
static void check_for_release(struct cpuset *cs, char **ppathbuf)
|
2005-04-16 16:20:36 -06:00
|
|
|
{
|
|
|
|
if (notify_on_release(cs) && atomic_read(&cs->count) == 0 &&
|
|
|
|
list_empty(&cs->children)) {
|
|
|
|
char *buf;
|
|
|
|
|
|
|
|
buf = kmalloc(PAGE_SIZE, GFP_KERNEL);
|
|
|
|
if (!buf)
|
|
|
|
return;
|
|
|
|
if (cpuset_path(cs, buf, PAGE_SIZE) < 0)
|
[PATCH] cpuset release ABBA deadlock fix
Fix possible cpuset_sem ABBA deadlock if 'notify_on_release' set.
For a particular usage pattern, creating and destroying cpusets fairly
frequently using notify_on_release, on a very large system, this deadlock
can be seen every few days. If you are not using the cpuset
notify_on_release feature, you will never see this deadlock.
The existing code, on task exit (or cpuset deletion) did:
get cpuset_sem
if cpuset marked notify_on_release and is ready to release:
compute cpuset path relative to /dev/cpuset mount point
call_usermodehelper() forks /sbin/cpuset_release_agent with path
drop cpuset_sem
Unfortunately, the fork in call_usermodehelper can allocate memory, and
allocating memory can require cpuset_sem, if the mems_generation values
changed in the interim. This results in an ABBA deadlock, trying to obtain
cpuset_sem when it is already held by the current task.
To fix this, I put the cpuset path (which must be computed while holding
cpuset_sem) in a temporary buffer, to be used in the call_usermodehelper
call of /sbin/cpuset_release_agent only _after_ dropping cpuset_sem.
So the new logic is:
get cpuset_sem
if cpuset marked notify_on_release and is ready to release:
compute cpuset path relative to /dev/cpuset mount point
stash path in kmalloc'd buffer
drop cpuset_sem
call_usermodehelper() forks /sbin/cpuset_release_agent with path
free path
The sharp eyed reader might notice that this patch does not contain any
calls to kmalloc. The existing code in the check_for_release() routine was
already kmalloc'ing a buffer to hold the cpuset path. In the old code, it
just held the buffer for a few lines, over the cpuset_release_agent() call
that in turn invoked call_usermodehelper(). In the new code, with the
application of this patch, it returns that buffer via the new char
**ppathbuf parameter, for later use and freeing in cpuset_release_agent(),
which is called after cpuset_sem is dropped. Whereas the old code has just
one call to cpuset_release_agent(), right in the check_for_release()
routine, the new code has three calls to cpuset_release_agent(), from the
various places that a cpuset can be released.
This patch has been build and booted on SN2, and passed a stress test that
previously hit the deadlock within a few seconds.
Signed-off-by: Paul Jackson <pj@sgi.com>
Signed-off-by: Andrew Morton <akpm@osdl.org>
Signed-off-by: Linus Torvalds <torvalds@osdl.org>
2005-08-09 11:07:59 -06:00
|
|
|
kfree(buf);
|
|
|
|
else
|
|
|
|
*ppathbuf = buf;
|
2005-04-16 16:20:36 -06:00
|
|
|
}
|
|
|
|
}
|
|
|
|
|
|
|
|
/*
|
|
|
|
* Return in *pmask the portion of a cpusets's cpus_allowed that
|
|
|
|
* are online. If none are online, walk up the cpuset hierarchy
|
|
|
|
* until we find one that does have some online cpus. If we get
|
|
|
|
* all the way to the top and still haven't found any online cpus,
|
|
|
|
* return cpu_online_map. Or if passed a NULL cs from an exit'ing
|
|
|
|
* task, return cpu_online_map.
|
|
|
|
*
|
|
|
|
* One way or another, we guarantee to return some non-empty subset
|
|
|
|
* of cpu_online_map.
|
|
|
|
*
|
2006-03-23 04:00:18 -07:00
|
|
|
* Call with callback_mutex held.
|
2005-04-16 16:20:36 -06:00
|
|
|
*/
|
|
|
|
|
|
|
|
static void guarantee_online_cpus(const struct cpuset *cs, cpumask_t *pmask)
|
|
|
|
{
|
|
|
|
while (cs && !cpus_intersects(cs->cpus_allowed, cpu_online_map))
|
|
|
|
cs = cs->parent;
|
|
|
|
if (cs)
|
|
|
|
cpus_and(*pmask, cs->cpus_allowed, cpu_online_map);
|
|
|
|
else
|
|
|
|
*pmask = cpu_online_map;
|
|
|
|
BUG_ON(!cpus_intersects(*pmask, cpu_online_map));
|
|
|
|
}
|
|
|
|
|
|
|
|
/*
|
|
|
|
* Return in *pmask the portion of a cpusets's mems_allowed that
|
|
|
|
* are online. If none are online, walk up the cpuset hierarchy
|
|
|
|
* until we find one that does have some online mems. If we get
|
|
|
|
* all the way to the top and still haven't found any online mems,
|
|
|
|
* return node_online_map.
|
|
|
|
*
|
|
|
|
* One way or another, we guarantee to return some non-empty subset
|
|
|
|
* of node_online_map.
|
|
|
|
*
|
2006-03-23 04:00:18 -07:00
|
|
|
* Call with callback_mutex held.
|
2005-04-16 16:20:36 -06:00
|
|
|
*/
|
|
|
|
|
|
|
|
static void guarantee_online_mems(const struct cpuset *cs, nodemask_t *pmask)
|
|
|
|
{
|
|
|
|
while (cs && !nodes_intersects(cs->mems_allowed, node_online_map))
|
|
|
|
cs = cs->parent;
|
|
|
|
if (cs)
|
|
|
|
nodes_and(*pmask, cs->mems_allowed, node_online_map);
|
|
|
|
else
|
|
|
|
*pmask = node_online_map;
|
|
|
|
BUG_ON(!nodes_intersects(*pmask, node_online_map));
|
|
|
|
}
|
|
|
|
|
2006-01-08 02:01:54 -07:00
|
|
|
/**
|
|
|
|
* cpuset_update_task_memory_state - update task memory placement
|
|
|
|
*
|
|
|
|
* If the current tasks cpusets mems_allowed changed behind our
|
|
|
|
* backs, update current->mems_allowed, mems_generation and task NUMA
|
|
|
|
* mempolicy to the new value.
|
[PATCH] cpusets: dual semaphore locking overhaul
Overhaul cpuset locking. Replace single semaphore with two semaphores.
The suggestion to use two locks was made by Roman Zippel.
Both locks are global. Code that wants to modify cpusets must first
acquire the exclusive manage_sem, which allows them read-only access to
cpusets, and holds off other would-be modifiers. Before making actual
changes, the second semaphore, callback_sem must be acquired as well. Code
that needs only to query cpusets must acquire callback_sem, which is also a
global exclusive lock.
The earlier problems with double tripping are avoided, because it is
allowed for holders of manage_sem to nest the second callback_sem lock, and
only callback_sem is needed by code called from within __alloc_pages(),
where the double tripping had been possible.
This is not quite the same as a normal read/write semaphore, because
obtaining read-only access with intent to change must hold off other such
attempts, while allowing read-only access w/o such intention. Changing
cpusets involves several related checks and changes, which must be done
while allowing read-only queries (to avoid the double trip), but while
ensuring nothing changes (holding off other would be modifiers.)
This overhaul of cpuset locking also makes careful use of task_lock() to
guard access to the task->cpuset pointer, closing a couple of race
conditions noticed while reading this code (thanks, Roman). I've never
seen these races fail in any use or test.
See further the comments in the code.
Signed-off-by: Paul Jackson <pj@sgi.com>
Signed-off-by: Andrew Morton <akpm@osdl.org>
Signed-off-by: Linus Torvalds <torvalds@osdl.org>
2005-10-30 16:02:30 -07:00
|
|
|
*
|
2006-01-08 02:01:54 -07:00
|
|
|
* Task mempolicy is updated by rebinding it relative to the
|
|
|
|
* current->cpuset if a task has its memory placement changed.
|
|
|
|
* Do not call this routine if in_interrupt().
|
|
|
|
*
|
2006-03-31 03:30:50 -07:00
|
|
|
* Call without callback_mutex or task_lock() held. May be
|
|
|
|
* called with or without manage_mutex held. Thanks in part to
|
|
|
|
* 'the_top_cpuset_hack', the tasks cpuset pointer will never
|
|
|
|
* be NULL. This routine also might acquire callback_mutex and
|
2006-01-08 02:01:54 -07:00
|
|
|
* current->mm->mmap_sem during call.
|
[PATCH] cpusets: dual semaphore locking overhaul
Overhaul cpuset locking. Replace single semaphore with two semaphores.
The suggestion to use two locks was made by Roman Zippel.
Both locks are global. Code that wants to modify cpusets must first
acquire the exclusive manage_sem, which allows them read-only access to
cpusets, and holds off other would-be modifiers. Before making actual
changes, the second semaphore, callback_sem must be acquired as well. Code
that needs only to query cpusets must acquire callback_sem, which is also a
global exclusive lock.
The earlier problems with double tripping are avoided, because it is
allowed for holders of manage_sem to nest the second callback_sem lock, and
only callback_sem is needed by code called from within __alloc_pages(),
where the double tripping had been possible.
This is not quite the same as a normal read/write semaphore, because
obtaining read-only access with intent to change must hold off other such
attempts, while allowing read-only access w/o such intention. Changing
cpusets involves several related checks and changes, which must be done
while allowing read-only queries (to avoid the double trip), but while
ensuring nothing changes (holding off other would be modifiers.)
This overhaul of cpuset locking also makes careful use of task_lock() to
guard access to the task->cpuset pointer, closing a couple of race
conditions noticed while reading this code (thanks, Roman). I've never
seen these races fail in any use or test.
See further the comments in the code.
Signed-off-by: Paul Jackson <pj@sgi.com>
Signed-off-by: Andrew Morton <akpm@osdl.org>
Signed-off-by: Linus Torvalds <torvalds@osdl.org>
2005-10-30 16:02:30 -07:00
|
|
|
*
|
[PATCH] cpuset: use rcu directly optimization
Optimize the cpuset impact on page allocation, the most performance critical
cpuset hook in the kernel.
On each page allocation, the cpuset hook needs to check for a possible change
in the current tasks cpuset. It can now handle the common case, of no change,
without taking any spinlock or semaphore, thanks to RCU.
Convert a spinlock on the current task to an rcu_read_lock(), saving
approximately a memory barrier and an atomic op, depending on architecture.
This is done by adding rcu_assign_pointer() and synchronize_rcu() calls to the
write side of the task->cpuset pointer, in cpuset.c:attach_task(), to delay
freeing up a detached cpuset until after any critical sections referencing
that pointer.
Thanks to Andi Kleen, Nick Piggin and Eric Dumazet for ideas.
Signed-off-by: Paul Jackson <pj@sgi.com>
Signed-off-by: Andrew Morton <akpm@osdl.org>
Signed-off-by: Linus Torvalds <torvalds@osdl.org>
2006-01-08 02:02:02 -07:00
|
|
|
* Reading current->cpuset->mems_generation doesn't need task_lock
|
|
|
|
* to guard the current->cpuset derefence, because it is guarded
|
|
|
|
* from concurrent freeing of current->cpuset by attach_task(),
|
|
|
|
* using RCU.
|
|
|
|
*
|
|
|
|
* The rcu_dereference() is technically probably not needed,
|
|
|
|
* as I don't actually mind if I see a new cpuset pointer but
|
|
|
|
* an old value of mems_generation. However this really only
|
|
|
|
* matters on alpha systems using cpusets heavily. If I dropped
|
|
|
|
* that rcu_dereference(), it would save them a memory barrier.
|
|
|
|
* For all other arch's, rcu_dereference is a no-op anyway, and for
|
|
|
|
* alpha systems not using cpusets, another planned optimization,
|
|
|
|
* avoiding the rcu critical section for tasks in the root cpuset
|
|
|
|
* which is statically allocated, so can't vanish, will make this
|
|
|
|
* irrelevant. Better to use RCU as intended, than to engage in
|
|
|
|
* some cute trick to save a memory barrier that is impossible to
|
|
|
|
* test, for alpha systems using cpusets heavily, which might not
|
|
|
|
* even exist.
|
[PATCH] cpusets: dual semaphore locking overhaul
Overhaul cpuset locking. Replace single semaphore with two semaphores.
The suggestion to use two locks was made by Roman Zippel.
Both locks are global. Code that wants to modify cpusets must first
acquire the exclusive manage_sem, which allows them read-only access to
cpusets, and holds off other would-be modifiers. Before making actual
changes, the second semaphore, callback_sem must be acquired as well. Code
that needs only to query cpusets must acquire callback_sem, which is also a
global exclusive lock.
The earlier problems with double tripping are avoided, because it is
allowed for holders of manage_sem to nest the second callback_sem lock, and
only callback_sem is needed by code called from within __alloc_pages(),
where the double tripping had been possible.
This is not quite the same as a normal read/write semaphore, because
obtaining read-only access with intent to change must hold off other such
attempts, while allowing read-only access w/o such intention. Changing
cpusets involves several related checks and changes, which must be done
while allowing read-only queries (to avoid the double trip), but while
ensuring nothing changes (holding off other would be modifiers.)
This overhaul of cpuset locking also makes careful use of task_lock() to
guard access to the task->cpuset pointer, closing a couple of race
conditions noticed while reading this code (thanks, Roman). I've never
seen these races fail in any use or test.
See further the comments in the code.
Signed-off-by: Paul Jackson <pj@sgi.com>
Signed-off-by: Andrew Morton <akpm@osdl.org>
Signed-off-by: Linus Torvalds <torvalds@osdl.org>
2005-10-30 16:02:30 -07:00
|
|
|
*
|
|
|
|
* This routine is needed to update the per-task mems_allowed data,
|
|
|
|
* within the tasks context, when it is trying to allocate memory
|
|
|
|
* (in various mm/mempolicy.c routines) and notices that some other
|
|
|
|
* task has been modifying its cpuset.
|
2005-04-16 16:20:36 -06:00
|
|
|
*/
|
|
|
|
|
2006-02-03 04:04:23 -07:00
|
|
|
void cpuset_update_task_memory_state(void)
|
2005-04-16 16:20:36 -06:00
|
|
|
{
|
[PATCH] cpusets: dual semaphore locking overhaul
Overhaul cpuset locking. Replace single semaphore with two semaphores.
The suggestion to use two locks was made by Roman Zippel.
Both locks are global. Code that wants to modify cpusets must first
acquire the exclusive manage_sem, which allows them read-only access to
cpusets, and holds off other would-be modifiers. Before making actual
changes, the second semaphore, callback_sem must be acquired as well. Code
that needs only to query cpusets must acquire callback_sem, which is also a
global exclusive lock.
The earlier problems with double tripping are avoided, because it is
allowed for holders of manage_sem to nest the second callback_sem lock, and
only callback_sem is needed by code called from within __alloc_pages(),
where the double tripping had been possible.
This is not quite the same as a normal read/write semaphore, because
obtaining read-only access with intent to change must hold off other such
attempts, while allowing read-only access w/o such intention. Changing
cpusets involves several related checks and changes, which must be done
while allowing read-only queries (to avoid the double trip), but while
ensuring nothing changes (holding off other would be modifiers.)
This overhaul of cpuset locking also makes careful use of task_lock() to
guard access to the task->cpuset pointer, closing a couple of race
conditions noticed while reading this code (thanks, Roman). I've never
seen these races fail in any use or test.
See further the comments in the code.
Signed-off-by: Paul Jackson <pj@sgi.com>
Signed-off-by: Andrew Morton <akpm@osdl.org>
Signed-off-by: Linus Torvalds <torvalds@osdl.org>
2005-10-30 16:02:30 -07:00
|
|
|
int my_cpusets_mem_gen;
|
2006-01-08 02:01:54 -07:00
|
|
|
struct task_struct *tsk = current;
|
[PATCH] cpuset: use rcu directly optimization
Optimize the cpuset impact on page allocation, the most performance critical
cpuset hook in the kernel.
On each page allocation, the cpuset hook needs to check for a possible change
in the current tasks cpuset. It can now handle the common case, of no change,
without taking any spinlock or semaphore, thanks to RCU.
Convert a spinlock on the current task to an rcu_read_lock(), saving
approximately a memory barrier and an atomic op, depending on architecture.
This is done by adding rcu_assign_pointer() and synchronize_rcu() calls to the
write side of the task->cpuset pointer, in cpuset.c:attach_task(), to delay
freeing up a detached cpuset until after any critical sections referencing
that pointer.
Thanks to Andi Kleen, Nick Piggin and Eric Dumazet for ideas.
Signed-off-by: Paul Jackson <pj@sgi.com>
Signed-off-by: Andrew Morton <akpm@osdl.org>
Signed-off-by: Linus Torvalds <torvalds@osdl.org>
2006-01-08 02:02:02 -07:00
|
|
|
struct cpuset *cs;
|
[PATCH] cpusets: dual semaphore locking overhaul
Overhaul cpuset locking. Replace single semaphore with two semaphores.
The suggestion to use two locks was made by Roman Zippel.
Both locks are global. Code that wants to modify cpusets must first
acquire the exclusive manage_sem, which allows them read-only access to
cpusets, and holds off other would-be modifiers. Before making actual
changes, the second semaphore, callback_sem must be acquired as well. Code
that needs only to query cpusets must acquire callback_sem, which is also a
global exclusive lock.
The earlier problems with double tripping are avoided, because it is
allowed for holders of manage_sem to nest the second callback_sem lock, and
only callback_sem is needed by code called from within __alloc_pages(),
where the double tripping had been possible.
This is not quite the same as a normal read/write semaphore, because
obtaining read-only access with intent to change must hold off other such
attempts, while allowing read-only access w/o such intention. Changing
cpusets involves several related checks and changes, which must be done
while allowing read-only queries (to avoid the double trip), but while
ensuring nothing changes (holding off other would be modifiers.)
This overhaul of cpuset locking also makes careful use of task_lock() to
guard access to the task->cpuset pointer, closing a couple of race
conditions noticed while reading this code (thanks, Roman). I've never
seen these races fail in any use or test.
See further the comments in the code.
Signed-off-by: Paul Jackson <pj@sgi.com>
Signed-off-by: Andrew Morton <akpm@osdl.org>
Signed-off-by: Linus Torvalds <torvalds@osdl.org>
2005-10-30 16:02:30 -07:00
|
|
|
|
2006-01-08 02:02:04 -07:00
|
|
|
if (tsk->cpuset == &top_cpuset) {
|
|
|
|
/* Don't need rcu for top_cpuset. It's never freed. */
|
|
|
|
my_cpusets_mem_gen = top_cpuset.mems_generation;
|
|
|
|
} else {
|
|
|
|
rcu_read_lock();
|
|
|
|
cs = rcu_dereference(tsk->cpuset);
|
|
|
|
my_cpusets_mem_gen = cs->mems_generation;
|
|
|
|
rcu_read_unlock();
|
|
|
|
}
|
2005-04-16 16:20:36 -06:00
|
|
|
|
2006-01-08 02:01:54 -07:00
|
|
|
if (my_cpusets_mem_gen != tsk->cpuset_mems_generation) {
|
2006-03-23 04:00:18 -07:00
|
|
|
mutex_lock(&callback_mutex);
|
2006-01-08 02:01:54 -07:00
|
|
|
task_lock(tsk);
|
|
|
|
cs = tsk->cpuset; /* Maybe changed when task not locked */
|
|
|
|
guarantee_online_mems(cs, &tsk->mems_allowed);
|
|
|
|
tsk->cpuset_mems_generation = cs->mems_generation;
|
[PATCH] cpuset memory spread basic implementation
This patch provides the implementation and cpuset interface for an alternative
memory allocation policy that can be applied to certain kinds of memory
allocations, such as the page cache (file system buffers) and some slab caches
(such as inode caches).
The policy is called "memory spreading." If enabled, it spreads out these
kinds of memory allocations over all the nodes allowed to a task, instead of
preferring to place them on the node where the task is executing.
All other kinds of allocations, including anonymous pages for a tasks stack
and data regions, are not affected by this policy choice, and continue to be
allocated preferring the node local to execution, as modified by the NUMA
mempolicy.
There are two boolean flag files per cpuset that control where the kernel
allocates pages for the file system buffers and related in kernel data
structures. They are called 'memory_spread_page' and 'memory_spread_slab'.
If the per-cpuset boolean flag file 'memory_spread_page' is set, then the
kernel will spread the file system buffers (page cache) evenly over all the
nodes that the faulting task is allowed to use, instead of preferring to put
those pages on the node where the task is running.
If the per-cpuset boolean flag file 'memory_spread_slab' is set, then the
kernel will spread some file system related slab caches, such as for inodes
and dentries evenly over all the nodes that the faulting task is allowed to
use, instead of preferring to put those pages on the node where the task is
running.
The implementation is simple. Setting the cpuset flags 'memory_spread_page'
or 'memory_spread_cache' turns on the per-process flags PF_SPREAD_PAGE or
PF_SPREAD_SLAB, respectively, for each task that is in the cpuset or
subsequently joins that cpuset. In subsequent patches, the page allocation
calls for the affected page cache and slab caches are modified to perform an
inline check for these flags, and if set, a call to a new routine
cpuset_mem_spread_node() returns the node to prefer for the allocation.
The cpuset_mem_spread_node() routine is also simple. It uses the value of a
per-task rotor cpuset_mem_spread_rotor to select the next node in the current
tasks mems_allowed to prefer for the allocation.
This policy can provide substantial improvements for jobs that need to place
thread local data on the corresponding node, but that need to access large
file system data sets that need to be spread across the several nodes in the
jobs cpuset in order to fit. Without this patch, especially for jobs that
might have one thread reading in the data set, the memory allocation across
the nodes in the jobs cpuset can become very uneven.
A couple of Copyright year ranges are updated as well. And a couple of email
addresses that can be found in the MAINTAINERS file are removed.
Signed-off-by: Paul Jackson <pj@sgi.com>
Signed-off-by: Andrew Morton <akpm@osdl.org>
Signed-off-by: Linus Torvalds <torvalds@osdl.org>
2006-03-24 04:16:03 -07:00
|
|
|
if (is_spread_page(cs))
|
|
|
|
tsk->flags |= PF_SPREAD_PAGE;
|
|
|
|
else
|
|
|
|
tsk->flags &= ~PF_SPREAD_PAGE;
|
|
|
|
if (is_spread_slab(cs))
|
|
|
|
tsk->flags |= PF_SPREAD_SLAB;
|
|
|
|
else
|
|
|
|
tsk->flags &= ~PF_SPREAD_SLAB;
|
2006-01-08 02:01:54 -07:00
|
|
|
task_unlock(tsk);
|
2006-03-23 04:00:18 -07:00
|
|
|
mutex_unlock(&callback_mutex);
|
[PATCH] cpuset: numa_policy_rebind cleanup
Cleanup, reorganize and make more robust the mempolicy.c code to rebind
mempolicies relative to the containing cpuset after a tasks memory placement
changes.
The real motivator for this cleanup patch is to lay more groundwork for the
upcoming patch to correctly rebind NUMA mempolicies that are attached to vma's
after the containing cpuset memory placement changes.
NUMA mempolicies are constrained by the cpuset their task is a member of.
When either (1) a task is moved to a different cpuset, or (2) the 'mems'
mems_allowed of a cpuset is changed, then the NUMA mempolicies have embedded
node numbers (for MPOL_BIND, MPOL_INTERLEAVE and MPOL_PREFERRED) that need to
be recalculated, relative to their new cpuset placement.
The old code used an unreliable method of determining what was the old
mems_allowed constraining the mempolicy. It just looked at the tasks
mems_allowed value. This sort of worked with the present code, that just
rebinds the -task- mempolicy, and leaves any -vma- mempolicies broken,
referring to the old nodes. But in an upcoming patch, the vma mempolicies
will be rebound as well. Then the order in which the various task and vma
mempolicies are updated will no longer be deterministic, and one can no longer
count on the task->mems_allowed holding the old value for as long as needed.
It's not even clear if the current code was guaranteed to work reliably for
task mempolicies.
So I added a mems_allowed field to each mempolicy, stating exactly what
mems_allowed the policy is relative to, and updated synchronously and reliably
anytime that the mempolicy is rebound.
Also removed a useless wrapper routine, numa_policy_rebind(), and had its
caller, cpuset_update_task_memory_state(), call directly to the rewritten
policy_rebind() routine, and made that rebind routine extern instead of
static, and added a "mpol_" prefix to its name, making it
mpol_rebind_policy().
Signed-off-by: Paul Jackson <pj@sgi.com>
Signed-off-by: Andrew Morton <akpm@osdl.org>
Signed-off-by: Linus Torvalds <torvalds@osdl.org>
2006-01-08 02:01:56 -07:00
|
|
|
mpol_rebind_task(tsk, &tsk->mems_allowed);
|
2005-04-16 16:20:36 -06:00
|
|
|
}
|
|
|
|
}
|
|
|
|
|
|
|
|
/*
|
|
|
|
* is_cpuset_subset(p, q) - Is cpuset p a subset of cpuset q?
|
|
|
|
*
|
|
|
|
* One cpuset is a subset of another if all its allowed CPUs and
|
|
|
|
* Memory Nodes are a subset of the other, and its exclusive flags
|
2006-03-23 04:00:18 -07:00
|
|
|
* are only set if the other's are set. Call holding manage_mutex.
|
2005-04-16 16:20:36 -06:00
|
|
|
*/
|
|
|
|
|
|
|
|
static int is_cpuset_subset(const struct cpuset *p, const struct cpuset *q)
|
|
|
|
{
|
|
|
|
return cpus_subset(p->cpus_allowed, q->cpus_allowed) &&
|
|
|
|
nodes_subset(p->mems_allowed, q->mems_allowed) &&
|
|
|
|
is_cpu_exclusive(p) <= is_cpu_exclusive(q) &&
|
|
|
|
is_mem_exclusive(p) <= is_mem_exclusive(q);
|
|
|
|
}
|
|
|
|
|
|
|
|
/*
|
|
|
|
* validate_change() - Used to validate that any proposed cpuset change
|
|
|
|
* follows the structural rules for cpusets.
|
|
|
|
*
|
|
|
|
* If we replaced the flag and mask values of the current cpuset
|
|
|
|
* (cur) with those values in the trial cpuset (trial), would
|
|
|
|
* our various subset and exclusive rules still be valid? Presumes
|
2006-03-23 04:00:18 -07:00
|
|
|
* manage_mutex held.
|
2005-04-16 16:20:36 -06:00
|
|
|
*
|
|
|
|
* 'cur' is the address of an actual, in-use cpuset. Operations
|
|
|
|
* such as list traversal that depend on the actual address of the
|
|
|
|
* cpuset in the list must use cur below, not trial.
|
|
|
|
*
|
|
|
|
* 'trial' is the address of bulk structure copy of cur, with
|
|
|
|
* perhaps one or more of the fields cpus_allowed, mems_allowed,
|
|
|
|
* or flags changed to new, trial values.
|
|
|
|
*
|
|
|
|
* Return 0 if valid, -errno if not.
|
|
|
|
*/
|
|
|
|
|
|
|
|
static int validate_change(const struct cpuset *cur, const struct cpuset *trial)
|
|
|
|
{
|
|
|
|
struct cpuset *c, *par;
|
|
|
|
|
|
|
|
/* Each of our child cpusets must be a subset of us */
|
|
|
|
list_for_each_entry(c, &cur->children, sibling) {
|
|
|
|
if (!is_cpuset_subset(c, trial))
|
|
|
|
return -EBUSY;
|
|
|
|
}
|
|
|
|
|
|
|
|
/* Remaining checks don't apply to root cpuset */
|
|
|
|
if ((par = cur->parent) == NULL)
|
|
|
|
return 0;
|
|
|
|
|
|
|
|
/* We must be a subset of our parent cpuset */
|
|
|
|
if (!is_cpuset_subset(trial, par))
|
|
|
|
return -EACCES;
|
|
|
|
|
|
|
|
/* If either I or some sibling (!= me) is exclusive, we can't overlap */
|
|
|
|
list_for_each_entry(c, &par->children, sibling) {
|
|
|
|
if ((is_cpu_exclusive(trial) || is_cpu_exclusive(c)) &&
|
|
|
|
c != cur &&
|
|
|
|
cpus_intersects(trial->cpus_allowed, c->cpus_allowed))
|
|
|
|
return -EINVAL;
|
|
|
|
if ((is_mem_exclusive(trial) || is_mem_exclusive(c)) &&
|
|
|
|
c != cur &&
|
|
|
|
nodes_intersects(trial->mems_allowed, c->mems_allowed))
|
|
|
|
return -EINVAL;
|
|
|
|
}
|
|
|
|
|
|
|
|
return 0;
|
|
|
|
}
|
|
|
|
|
2005-06-25 15:57:34 -06:00
|
|
|
/*
|
|
|
|
* For a given cpuset cur, partition the system as follows
|
|
|
|
* a. All cpus in the parent cpuset's cpus_allowed that are not part of any
|
|
|
|
* exclusive child cpusets
|
|
|
|
* b. All cpus in the current cpuset's cpus_allowed that are not part of any
|
|
|
|
* exclusive child cpusets
|
|
|
|
* Build these two partitions by calling partition_sched_domains
|
|
|
|
*
|
2006-03-23 04:00:18 -07:00
|
|
|
* Call with manage_mutex held. May nest a call to the
|
2005-06-25 15:57:34 -06:00
|
|
|
* lock_cpu_hotplug()/unlock_cpu_hotplug() pair.
|
2006-07-23 12:36:08 -06:00
|
|
|
* Must not be called holding callback_mutex, because we must
|
|
|
|
* not call lock_cpu_hotplug() while holding callback_mutex.
|
2005-06-25 15:57:34 -06:00
|
|
|
*/
|
2005-08-25 13:47:56 -06:00
|
|
|
|
2005-06-25 15:57:34 -06:00
|
|
|
static void update_cpu_domains(struct cpuset *cur)
|
|
|
|
{
|
|
|
|
struct cpuset *c, *par = cur->parent;
|
|
|
|
cpumask_t pspan, cspan;
|
|
|
|
|
|
|
|
if (par == NULL || cpus_empty(cur->cpus_allowed))
|
|
|
|
return;
|
|
|
|
|
|
|
|
/*
|
|
|
|
* Get all cpus from parent's cpus_allowed not part of exclusive
|
|
|
|
* children
|
|
|
|
*/
|
|
|
|
pspan = par->cpus_allowed;
|
|
|
|
list_for_each_entry(c, &par->children, sibling) {
|
|
|
|
if (is_cpu_exclusive(c))
|
|
|
|
cpus_andnot(pspan, pspan, c->cpus_allowed);
|
|
|
|
}
|
2006-07-23 12:36:08 -06:00
|
|
|
if (!is_cpu_exclusive(cur)) {
|
2005-06-25 15:57:34 -06:00
|
|
|
cpus_or(pspan, pspan, cur->cpus_allowed);
|
|
|
|
if (cpus_equal(pspan, cur->cpus_allowed))
|
|
|
|
return;
|
|
|
|
cspan = CPU_MASK_NONE;
|
|
|
|
} else {
|
|
|
|
if (cpus_empty(pspan))
|
|
|
|
return;
|
|
|
|
cspan = cur->cpus_allowed;
|
|
|
|
/*
|
|
|
|
* Get all cpus from current cpuset's cpus_allowed not part
|
|
|
|
* of exclusive children
|
|
|
|
*/
|
|
|
|
list_for_each_entry(c, &cur->children, sibling) {
|
|
|
|
if (is_cpu_exclusive(c))
|
|
|
|
cpus_andnot(cspan, cspan, c->cpus_allowed);
|
|
|
|
}
|
|
|
|
}
|
|
|
|
|
|
|
|
lock_cpu_hotplug();
|
|
|
|
partition_sched_domains(&pspan, &cspan);
|
|
|
|
unlock_cpu_hotplug();
|
|
|
|
}
|
|
|
|
|
[PATCH] cpusets: dual semaphore locking overhaul
Overhaul cpuset locking. Replace single semaphore with two semaphores.
The suggestion to use two locks was made by Roman Zippel.
Both locks are global. Code that wants to modify cpusets must first
acquire the exclusive manage_sem, which allows them read-only access to
cpusets, and holds off other would-be modifiers. Before making actual
changes, the second semaphore, callback_sem must be acquired as well. Code
that needs only to query cpusets must acquire callback_sem, which is also a
global exclusive lock.
The earlier problems with double tripping are avoided, because it is
allowed for holders of manage_sem to nest the second callback_sem lock, and
only callback_sem is needed by code called from within __alloc_pages(),
where the double tripping had been possible.
This is not quite the same as a normal read/write semaphore, because
obtaining read-only access with intent to change must hold off other such
attempts, while allowing read-only access w/o such intention. Changing
cpusets involves several related checks and changes, which must be done
while allowing read-only queries (to avoid the double trip), but while
ensuring nothing changes (holding off other would be modifiers.)
This overhaul of cpuset locking also makes careful use of task_lock() to
guard access to the task->cpuset pointer, closing a couple of race
conditions noticed while reading this code (thanks, Roman). I've never
seen these races fail in any use or test.
See further the comments in the code.
Signed-off-by: Paul Jackson <pj@sgi.com>
Signed-off-by: Andrew Morton <akpm@osdl.org>
Signed-off-by: Linus Torvalds <torvalds@osdl.org>
2005-10-30 16:02:30 -07:00
|
|
|
/*
|
2006-03-23 04:00:18 -07:00
|
|
|
* Call with manage_mutex held. May take callback_mutex during call.
|
[PATCH] cpusets: dual semaphore locking overhaul
Overhaul cpuset locking. Replace single semaphore with two semaphores.
The suggestion to use two locks was made by Roman Zippel.
Both locks are global. Code that wants to modify cpusets must first
acquire the exclusive manage_sem, which allows them read-only access to
cpusets, and holds off other would-be modifiers. Before making actual
changes, the second semaphore, callback_sem must be acquired as well. Code
that needs only to query cpusets must acquire callback_sem, which is also a
global exclusive lock.
The earlier problems with double tripping are avoided, because it is
allowed for holders of manage_sem to nest the second callback_sem lock, and
only callback_sem is needed by code called from within __alloc_pages(),
where the double tripping had been possible.
This is not quite the same as a normal read/write semaphore, because
obtaining read-only access with intent to change must hold off other such
attempts, while allowing read-only access w/o such intention. Changing
cpusets involves several related checks and changes, which must be done
while allowing read-only queries (to avoid the double trip), but while
ensuring nothing changes (holding off other would be modifiers.)
This overhaul of cpuset locking also makes careful use of task_lock() to
guard access to the task->cpuset pointer, closing a couple of race
conditions noticed while reading this code (thanks, Roman). I've never
seen these races fail in any use or test.
See further the comments in the code.
Signed-off-by: Paul Jackson <pj@sgi.com>
Signed-off-by: Andrew Morton <akpm@osdl.org>
Signed-off-by: Linus Torvalds <torvalds@osdl.org>
2005-10-30 16:02:30 -07:00
|
|
|
*/
|
|
|
|
|
2005-04-16 16:20:36 -06:00
|
|
|
static int update_cpumask(struct cpuset *cs, char *buf)
|
|
|
|
{
|
|
|
|
struct cpuset trialcs;
|
2005-06-25 15:57:34 -06:00
|
|
|
int retval, cpus_unchanged;
|
2005-04-16 16:20:36 -06:00
|
|
|
|
[PATCH] cpuset: top_cpuset tracks hotplug changes to cpu_online_map
Change the list of cpus allowed to tasks in the top (root) cpuset to
dynamically track what cpus are online, using a CPU hotplug notifier. Make
this top cpus file read-only.
On systems that have cpusets configured in their kernel, but that aren't
actively using cpusets (for some distros, this covers the majority of
systems) all tasks end up in the top cpuset.
If that system does support CPU hotplug, then these tasks cannot make use
of CPUs that are added after system boot, because the CPUs are not allowed
in the top cpuset. This is a surprising regression over earlier kernels
that didn't have cpusets enabled.
In order to keep the behaviour of cpusets consistent between systems
actively making use of them and systems not using them, this patch changes
the behaviour of the 'cpus' file in the top (root) cpuset, making it read
only, and making it automatically track the value of cpu_online_map. Thus
tasks in the top cpuset will have automatic use of hot plugged CPUs allowed
by their cpuset.
Thanks to Anton Blanchard and Nathan Lynch for reporting this problem,
driving the fix, and earlier versions of this patch.
Signed-off-by: Paul Jackson <pj@sgi.com>
Cc: Nathan Lynch <ntl@pobox.com>
Cc: Anton Blanchard <anton@samba.org>
Signed-off-by: Andrew Morton <akpm@osdl.org>
Signed-off-by: Linus Torvalds <torvalds@osdl.org>
2006-08-27 02:23:51 -06:00
|
|
|
/* top_cpuset.cpus_allowed tracks cpu_online_map; it's read-only */
|
|
|
|
if (cs == &top_cpuset)
|
|
|
|
return -EACCES;
|
|
|
|
|
2005-04-16 16:20:36 -06:00
|
|
|
trialcs = *cs;
|
|
|
|
retval = cpulist_parse(buf, trialcs.cpus_allowed);
|
|
|
|
if (retval < 0)
|
|
|
|
return retval;
|
|
|
|
cpus_and(trialcs.cpus_allowed, trialcs.cpus_allowed, cpu_online_map);
|
|
|
|
if (cpus_empty(trialcs.cpus_allowed))
|
|
|
|
return -ENOSPC;
|
|
|
|
retval = validate_change(cs, &trialcs);
|
2005-06-25 15:57:34 -06:00
|
|
|
if (retval < 0)
|
|
|
|
return retval;
|
|
|
|
cpus_unchanged = cpus_equal(cs->cpus_allowed, trialcs.cpus_allowed);
|
2006-03-23 04:00:18 -07:00
|
|
|
mutex_lock(&callback_mutex);
|
2005-06-25 15:57:34 -06:00
|
|
|
cs->cpus_allowed = trialcs.cpus_allowed;
|
2006-03-23 04:00:18 -07:00
|
|
|
mutex_unlock(&callback_mutex);
|
2005-06-25 15:57:34 -06:00
|
|
|
if (is_cpu_exclusive(cs) && !cpus_unchanged)
|
|
|
|
update_cpu_domains(cs);
|
|
|
|
return 0;
|
2005-04-16 16:20:36 -06:00
|
|
|
}
|
|
|
|
|
2006-03-31 03:30:52 -07:00
|
|
|
/*
|
|
|
|
* cpuset_migrate_mm
|
|
|
|
*
|
|
|
|
* Migrate memory region from one set of nodes to another.
|
|
|
|
*
|
|
|
|
* Temporarilly set tasks mems_allowed to target nodes of migration,
|
|
|
|
* so that the migration code can allocate pages on these nodes.
|
|
|
|
*
|
|
|
|
* Call holding manage_mutex, so our current->cpuset won't change
|
|
|
|
* during this call, as manage_mutex holds off any attach_task()
|
|
|
|
* calls. Therefore we don't need to take task_lock around the
|
|
|
|
* call to guarantee_online_mems(), as we know no one is changing
|
|
|
|
* our tasks cpuset.
|
|
|
|
*
|
|
|
|
* Hold callback_mutex around the two modifications of our tasks
|
|
|
|
* mems_allowed to synchronize with cpuset_mems_allowed().
|
|
|
|
*
|
|
|
|
* While the mm_struct we are migrating is typically from some
|
|
|
|
* other task, the task_struct mems_allowed that we are hacking
|
|
|
|
* is for our current task, which must allocate new pages for that
|
|
|
|
* migrating memory region.
|
|
|
|
*
|
|
|
|
* We call cpuset_update_task_memory_state() before hacking
|
|
|
|
* our tasks mems_allowed, so that we are assured of being in
|
|
|
|
* sync with our tasks cpuset, and in particular, callbacks to
|
|
|
|
* cpuset_update_task_memory_state() from nested page allocations
|
|
|
|
* won't see any mismatch of our cpuset and task mems_generation
|
|
|
|
* values, so won't overwrite our hacked tasks mems_allowed
|
|
|
|
* nodemask.
|
|
|
|
*/
|
|
|
|
|
|
|
|
static void cpuset_migrate_mm(struct mm_struct *mm, const nodemask_t *from,
|
|
|
|
const nodemask_t *to)
|
|
|
|
{
|
|
|
|
struct task_struct *tsk = current;
|
|
|
|
|
|
|
|
cpuset_update_task_memory_state();
|
|
|
|
|
|
|
|
mutex_lock(&callback_mutex);
|
|
|
|
tsk->mems_allowed = *to;
|
|
|
|
mutex_unlock(&callback_mutex);
|
|
|
|
|
|
|
|
do_migrate_pages(mm, from, to, MPOL_MF_MOVE_ALL);
|
|
|
|
|
|
|
|
mutex_lock(&callback_mutex);
|
|
|
|
guarantee_online_mems(tsk->cpuset, &tsk->mems_allowed);
|
|
|
|
mutex_unlock(&callback_mutex);
|
|
|
|
}
|
|
|
|
|
[PATCH] cpusets: dual semaphore locking overhaul
Overhaul cpuset locking. Replace single semaphore with two semaphores.
The suggestion to use two locks was made by Roman Zippel.
Both locks are global. Code that wants to modify cpusets must first
acquire the exclusive manage_sem, which allows them read-only access to
cpusets, and holds off other would-be modifiers. Before making actual
changes, the second semaphore, callback_sem must be acquired as well. Code
that needs only to query cpusets must acquire callback_sem, which is also a
global exclusive lock.
The earlier problems with double tripping are avoided, because it is
allowed for holders of manage_sem to nest the second callback_sem lock, and
only callback_sem is needed by code called from within __alloc_pages(),
where the double tripping had been possible.
This is not quite the same as a normal read/write semaphore, because
obtaining read-only access with intent to change must hold off other such
attempts, while allowing read-only access w/o such intention. Changing
cpusets involves several related checks and changes, which must be done
while allowing read-only queries (to avoid the double trip), but while
ensuring nothing changes (holding off other would be modifiers.)
This overhaul of cpuset locking also makes careful use of task_lock() to
guard access to the task->cpuset pointer, closing a couple of race
conditions noticed while reading this code (thanks, Roman). I've never
seen these races fail in any use or test.
See further the comments in the code.
Signed-off-by: Paul Jackson <pj@sgi.com>
Signed-off-by: Andrew Morton <akpm@osdl.org>
Signed-off-by: Linus Torvalds <torvalds@osdl.org>
2005-10-30 16:02:30 -07:00
|
|
|
/*
|
[PATCH] cpuset: rebind vma mempolicies fix
Fix more of longstanding bug in cpuset/mempolicy interaction.
NUMA mempolicies (mm/mempolicy.c) are constrained by the current tasks cpuset
to just the Memory Nodes allowed by that cpuset. The kernel maintains
internal state for each mempolicy, tracking what nodes are used for the
MPOL_INTERLEAVE, MPOL_BIND or MPOL_PREFERRED policies.
When a tasks cpuset memory placement changes, whether because the cpuset
changed, or because the task was attached to a different cpuset, then the
tasks mempolicies have to be rebound to the new cpuset placement, so as to
preserve the cpuset-relative numbering of the nodes in that policy.
An earlier fix handled such mempolicy rebinding for mempolicies attached to a
task.
This fix rebinds mempolicies attached to vma's (address ranges in a tasks
address space.) Due to the need to hold the task->mm->mmap_sem semaphore while
updating vma's, the rebinding of vma mempolicies has to be done when the
cpuset memory placement is changed, at which time mmap_sem can be safely
acquired. The tasks mempolicy is rebound later, when the task next attempts
to allocate memory and notices that its task->cpuset_mems_generation is
out-of-date with its cpusets mems_generation.
Because walking the tasklist to find all tasks attached to a changing cpuset
requires holding tasklist_lock, a spinlock, one cannot update the vma's of the
affected tasks while doing the tasklist scan. In general, one cannot acquire
a semaphore (which can sleep) while already holding a spinlock (such as
tasklist_lock). So a list of mm references has to be built up during the
tasklist scan, then the tasklist lock dropped, then for each mm, its mmap_sem
acquired, and the vma's in that mm rebound.
Once the tasklist lock is dropped, affected tasks may fork new tasks, before
their mm's are rebound. A kernel global 'cpuset_being_rebound' is set to
point to the cpuset being rebound (there can only be one; cpuset modifications
are done under a global 'manage_sem' semaphore), and the mpol_copy code that
is used to copy a tasks mempolicies during fork catches such forking tasks,
and ensures their children are also rebound.
When a task is moved to a different cpuset, it is easier, as there is only one
task involved. It's mm->vma's are scanned, using the same
mpol_rebind_policy() as used above.
It may happen that both the mpol_copy hook and the update done via the
tasklist scan update the same mm twice. This is ok, as the mempolicies of
each vma in an mm keep track of what mems_allowed they are relative to, and
safely no-op a second request to rebind to the same nodes.
Signed-off-by: Paul Jackson <pj@sgi.com>
Signed-off-by: Andrew Morton <akpm@osdl.org>
Signed-off-by: Linus Torvalds <torvalds@osdl.org>
2006-01-08 02:01:59 -07:00
|
|
|
* Handle user request to change the 'mems' memory placement
|
|
|
|
* of a cpuset. Needs to validate the request, update the
|
|
|
|
* cpusets mems_allowed and mems_generation, and for each
|
2006-01-08 02:02:00 -07:00
|
|
|
* task in the cpuset, rebind any vma mempolicies and if
|
|
|
|
* the cpuset is marked 'memory_migrate', migrate the tasks
|
|
|
|
* pages to the new memory.
|
[PATCH] cpuset: rebind vma mempolicies fix
Fix more of longstanding bug in cpuset/mempolicy interaction.
NUMA mempolicies (mm/mempolicy.c) are constrained by the current tasks cpuset
to just the Memory Nodes allowed by that cpuset. The kernel maintains
internal state for each mempolicy, tracking what nodes are used for the
MPOL_INTERLEAVE, MPOL_BIND or MPOL_PREFERRED policies.
When a tasks cpuset memory placement changes, whether because the cpuset
changed, or because the task was attached to a different cpuset, then the
tasks mempolicies have to be rebound to the new cpuset placement, so as to
preserve the cpuset-relative numbering of the nodes in that policy.
An earlier fix handled such mempolicy rebinding for mempolicies attached to a
task.
This fix rebinds mempolicies attached to vma's (address ranges in a tasks
address space.) Due to the need to hold the task->mm->mmap_sem semaphore while
updating vma's, the rebinding of vma mempolicies has to be done when the
cpuset memory placement is changed, at which time mmap_sem can be safely
acquired. The tasks mempolicy is rebound later, when the task next attempts
to allocate memory and notices that its task->cpuset_mems_generation is
out-of-date with its cpusets mems_generation.
Because walking the tasklist to find all tasks attached to a changing cpuset
requires holding tasklist_lock, a spinlock, one cannot update the vma's of the
affected tasks while doing the tasklist scan. In general, one cannot acquire
a semaphore (which can sleep) while already holding a spinlock (such as
tasklist_lock). So a list of mm references has to be built up during the
tasklist scan, then the tasklist lock dropped, then for each mm, its mmap_sem
acquired, and the vma's in that mm rebound.
Once the tasklist lock is dropped, affected tasks may fork new tasks, before
their mm's are rebound. A kernel global 'cpuset_being_rebound' is set to
point to the cpuset being rebound (there can only be one; cpuset modifications
are done under a global 'manage_sem' semaphore), and the mpol_copy code that
is used to copy a tasks mempolicies during fork catches such forking tasks,
and ensures their children are also rebound.
When a task is moved to a different cpuset, it is easier, as there is only one
task involved. It's mm->vma's are scanned, using the same
mpol_rebind_policy() as used above.
It may happen that both the mpol_copy hook and the update done via the
tasklist scan update the same mm twice. This is ok, as the mempolicies of
each vma in an mm keep track of what mems_allowed they are relative to, and
safely no-op a second request to rebind to the same nodes.
Signed-off-by: Paul Jackson <pj@sgi.com>
Signed-off-by: Andrew Morton <akpm@osdl.org>
Signed-off-by: Linus Torvalds <torvalds@osdl.org>
2006-01-08 02:01:59 -07:00
|
|
|
*
|
2006-03-23 04:00:18 -07:00
|
|
|
* Call with manage_mutex held. May take callback_mutex during call.
|
[PATCH] cpuset: rebind vma mempolicies fix
Fix more of longstanding bug in cpuset/mempolicy interaction.
NUMA mempolicies (mm/mempolicy.c) are constrained by the current tasks cpuset
to just the Memory Nodes allowed by that cpuset. The kernel maintains
internal state for each mempolicy, tracking what nodes are used for the
MPOL_INTERLEAVE, MPOL_BIND or MPOL_PREFERRED policies.
When a tasks cpuset memory placement changes, whether because the cpuset
changed, or because the task was attached to a different cpuset, then the
tasks mempolicies have to be rebound to the new cpuset placement, so as to
preserve the cpuset-relative numbering of the nodes in that policy.
An earlier fix handled such mempolicy rebinding for mempolicies attached to a
task.
This fix rebinds mempolicies attached to vma's (address ranges in a tasks
address space.) Due to the need to hold the task->mm->mmap_sem semaphore while
updating vma's, the rebinding of vma mempolicies has to be done when the
cpuset memory placement is changed, at which time mmap_sem can be safely
acquired. The tasks mempolicy is rebound later, when the task next attempts
to allocate memory and notices that its task->cpuset_mems_generation is
out-of-date with its cpusets mems_generation.
Because walking the tasklist to find all tasks attached to a changing cpuset
requires holding tasklist_lock, a spinlock, one cannot update the vma's of the
affected tasks while doing the tasklist scan. In general, one cannot acquire
a semaphore (which can sleep) while already holding a spinlock (such as
tasklist_lock). So a list of mm references has to be built up during the
tasklist scan, then the tasklist lock dropped, then for each mm, its mmap_sem
acquired, and the vma's in that mm rebound.
Once the tasklist lock is dropped, affected tasks may fork new tasks, before
their mm's are rebound. A kernel global 'cpuset_being_rebound' is set to
point to the cpuset being rebound (there can only be one; cpuset modifications
are done under a global 'manage_sem' semaphore), and the mpol_copy code that
is used to copy a tasks mempolicies during fork catches such forking tasks,
and ensures their children are also rebound.
When a task is moved to a different cpuset, it is easier, as there is only one
task involved. It's mm->vma's are scanned, using the same
mpol_rebind_policy() as used above.
It may happen that both the mpol_copy hook and the update done via the
tasklist scan update the same mm twice. This is ok, as the mempolicies of
each vma in an mm keep track of what mems_allowed they are relative to, and
safely no-op a second request to rebind to the same nodes.
Signed-off-by: Paul Jackson <pj@sgi.com>
Signed-off-by: Andrew Morton <akpm@osdl.org>
Signed-off-by: Linus Torvalds <torvalds@osdl.org>
2006-01-08 02:01:59 -07:00
|
|
|
* Will take tasklist_lock, scan tasklist for tasks in cpuset cs,
|
|
|
|
* lock each such tasks mm->mmap_sem, scan its vma's and rebind
|
|
|
|
* their mempolicies to the cpusets new mems_allowed.
|
[PATCH] cpusets: dual semaphore locking overhaul
Overhaul cpuset locking. Replace single semaphore with two semaphores.
The suggestion to use two locks was made by Roman Zippel.
Both locks are global. Code that wants to modify cpusets must first
acquire the exclusive manage_sem, which allows them read-only access to
cpusets, and holds off other would-be modifiers. Before making actual
changes, the second semaphore, callback_sem must be acquired as well. Code
that needs only to query cpusets must acquire callback_sem, which is also a
global exclusive lock.
The earlier problems with double tripping are avoided, because it is
allowed for holders of manage_sem to nest the second callback_sem lock, and
only callback_sem is needed by code called from within __alloc_pages(),
where the double tripping had been possible.
This is not quite the same as a normal read/write semaphore, because
obtaining read-only access with intent to change must hold off other such
attempts, while allowing read-only access w/o such intention. Changing
cpusets involves several related checks and changes, which must be done
while allowing read-only queries (to avoid the double trip), but while
ensuring nothing changes (holding off other would be modifiers.)
This overhaul of cpuset locking also makes careful use of task_lock() to
guard access to the task->cpuset pointer, closing a couple of race
conditions noticed while reading this code (thanks, Roman). I've never
seen these races fail in any use or test.
See further the comments in the code.
Signed-off-by: Paul Jackson <pj@sgi.com>
Signed-off-by: Andrew Morton <akpm@osdl.org>
Signed-off-by: Linus Torvalds <torvalds@osdl.org>
2005-10-30 16:02:30 -07:00
|
|
|
*/
|
|
|
|
|
2005-04-16 16:20:36 -06:00
|
|
|
static int update_nodemask(struct cpuset *cs, char *buf)
|
|
|
|
{
|
|
|
|
struct cpuset trialcs;
|
2006-01-08 02:02:00 -07:00
|
|
|
nodemask_t oldmem;
|
[PATCH] cpuset: rebind vma mempolicies fix
Fix more of longstanding bug in cpuset/mempolicy interaction.
NUMA mempolicies (mm/mempolicy.c) are constrained by the current tasks cpuset
to just the Memory Nodes allowed by that cpuset. The kernel maintains
internal state for each mempolicy, tracking what nodes are used for the
MPOL_INTERLEAVE, MPOL_BIND or MPOL_PREFERRED policies.
When a tasks cpuset memory placement changes, whether because the cpuset
changed, or because the task was attached to a different cpuset, then the
tasks mempolicies have to be rebound to the new cpuset placement, so as to
preserve the cpuset-relative numbering of the nodes in that policy.
An earlier fix handled such mempolicy rebinding for mempolicies attached to a
task.
This fix rebinds mempolicies attached to vma's (address ranges in a tasks
address space.) Due to the need to hold the task->mm->mmap_sem semaphore while
updating vma's, the rebinding of vma mempolicies has to be done when the
cpuset memory placement is changed, at which time mmap_sem can be safely
acquired. The tasks mempolicy is rebound later, when the task next attempts
to allocate memory and notices that its task->cpuset_mems_generation is
out-of-date with its cpusets mems_generation.
Because walking the tasklist to find all tasks attached to a changing cpuset
requires holding tasklist_lock, a spinlock, one cannot update the vma's of the
affected tasks while doing the tasklist scan. In general, one cannot acquire
a semaphore (which can sleep) while already holding a spinlock (such as
tasklist_lock). So a list of mm references has to be built up during the
tasklist scan, then the tasklist lock dropped, then for each mm, its mmap_sem
acquired, and the vma's in that mm rebound.
Once the tasklist lock is dropped, affected tasks may fork new tasks, before
their mm's are rebound. A kernel global 'cpuset_being_rebound' is set to
point to the cpuset being rebound (there can only be one; cpuset modifications
are done under a global 'manage_sem' semaphore), and the mpol_copy code that
is used to copy a tasks mempolicies during fork catches such forking tasks,
and ensures their children are also rebound.
When a task is moved to a different cpuset, it is easier, as there is only one
task involved. It's mm->vma's are scanned, using the same
mpol_rebind_policy() as used above.
It may happen that both the mpol_copy hook and the update done via the
tasklist scan update the same mm twice. This is ok, as the mempolicies of
each vma in an mm keep track of what mems_allowed they are relative to, and
safely no-op a second request to rebind to the same nodes.
Signed-off-by: Paul Jackson <pj@sgi.com>
Signed-off-by: Andrew Morton <akpm@osdl.org>
Signed-off-by: Linus Torvalds <torvalds@osdl.org>
2006-01-08 02:01:59 -07:00
|
|
|
struct task_struct *g, *p;
|
|
|
|
struct mm_struct **mmarray;
|
|
|
|
int i, n, ntasks;
|
2006-01-08 02:02:00 -07:00
|
|
|
int migrate;
|
[PATCH] cpuset: rebind vma mempolicies fix
Fix more of longstanding bug in cpuset/mempolicy interaction.
NUMA mempolicies (mm/mempolicy.c) are constrained by the current tasks cpuset
to just the Memory Nodes allowed by that cpuset. The kernel maintains
internal state for each mempolicy, tracking what nodes are used for the
MPOL_INTERLEAVE, MPOL_BIND or MPOL_PREFERRED policies.
When a tasks cpuset memory placement changes, whether because the cpuset
changed, or because the task was attached to a different cpuset, then the
tasks mempolicies have to be rebound to the new cpuset placement, so as to
preserve the cpuset-relative numbering of the nodes in that policy.
An earlier fix handled such mempolicy rebinding for mempolicies attached to a
task.
This fix rebinds mempolicies attached to vma's (address ranges in a tasks
address space.) Due to the need to hold the task->mm->mmap_sem semaphore while
updating vma's, the rebinding of vma mempolicies has to be done when the
cpuset memory placement is changed, at which time mmap_sem can be safely
acquired. The tasks mempolicy is rebound later, when the task next attempts
to allocate memory and notices that its task->cpuset_mems_generation is
out-of-date with its cpusets mems_generation.
Because walking the tasklist to find all tasks attached to a changing cpuset
requires holding tasklist_lock, a spinlock, one cannot update the vma's of the
affected tasks while doing the tasklist scan. In general, one cannot acquire
a semaphore (which can sleep) while already holding a spinlock (such as
tasklist_lock). So a list of mm references has to be built up during the
tasklist scan, then the tasklist lock dropped, then for each mm, its mmap_sem
acquired, and the vma's in that mm rebound.
Once the tasklist lock is dropped, affected tasks may fork new tasks, before
their mm's are rebound. A kernel global 'cpuset_being_rebound' is set to
point to the cpuset being rebound (there can only be one; cpuset modifications
are done under a global 'manage_sem' semaphore), and the mpol_copy code that
is used to copy a tasks mempolicies during fork catches such forking tasks,
and ensures their children are also rebound.
When a task is moved to a different cpuset, it is easier, as there is only one
task involved. It's mm->vma's are scanned, using the same
mpol_rebind_policy() as used above.
It may happen that both the mpol_copy hook and the update done via the
tasklist scan update the same mm twice. This is ok, as the mempolicies of
each vma in an mm keep track of what mems_allowed they are relative to, and
safely no-op a second request to rebind to the same nodes.
Signed-off-by: Paul Jackson <pj@sgi.com>
Signed-off-by: Andrew Morton <akpm@osdl.org>
Signed-off-by: Linus Torvalds <torvalds@osdl.org>
2006-01-08 02:01:59 -07:00
|
|
|
int fudge;
|
2005-04-16 16:20:36 -06:00
|
|
|
int retval;
|
|
|
|
|
|
|
|
trialcs = *cs;
|
|
|
|
retval = nodelist_parse(buf, trialcs.mems_allowed);
|
|
|
|
if (retval < 0)
|
2006-01-08 02:01:52 -07:00
|
|
|
goto done;
|
2005-04-16 16:20:36 -06:00
|
|
|
nodes_and(trialcs.mems_allowed, trialcs.mems_allowed, node_online_map);
|
2006-01-08 02:02:00 -07:00
|
|
|
oldmem = cs->mems_allowed;
|
|
|
|
if (nodes_equal(oldmem, trialcs.mems_allowed)) {
|
|
|
|
retval = 0; /* Too easy - nothing to do */
|
|
|
|
goto done;
|
|
|
|
}
|
2006-01-08 02:01:52 -07:00
|
|
|
if (nodes_empty(trialcs.mems_allowed)) {
|
|
|
|
retval = -ENOSPC;
|
|
|
|
goto done;
|
2005-04-16 16:20:36 -06:00
|
|
|
}
|
2006-01-08 02:01:52 -07:00
|
|
|
retval = validate_change(cs, &trialcs);
|
|
|
|
if (retval < 0)
|
|
|
|
goto done;
|
|
|
|
|
2006-03-23 04:00:18 -07:00
|
|
|
mutex_lock(&callback_mutex);
|
2006-01-08 02:01:52 -07:00
|
|
|
cs->mems_allowed = trialcs.mems_allowed;
|
2006-03-24 04:16:11 -07:00
|
|
|
cs->mems_generation = cpuset_mems_generation++;
|
2006-03-23 04:00:18 -07:00
|
|
|
mutex_unlock(&callback_mutex);
|
2006-01-08 02:01:52 -07:00
|
|
|
|
[PATCH] cpuset: rebind vma mempolicies fix
Fix more of longstanding bug in cpuset/mempolicy interaction.
NUMA mempolicies (mm/mempolicy.c) are constrained by the current tasks cpuset
to just the Memory Nodes allowed by that cpuset. The kernel maintains
internal state for each mempolicy, tracking what nodes are used for the
MPOL_INTERLEAVE, MPOL_BIND or MPOL_PREFERRED policies.
When a tasks cpuset memory placement changes, whether because the cpuset
changed, or because the task was attached to a different cpuset, then the
tasks mempolicies have to be rebound to the new cpuset placement, so as to
preserve the cpuset-relative numbering of the nodes in that policy.
An earlier fix handled such mempolicy rebinding for mempolicies attached to a
task.
This fix rebinds mempolicies attached to vma's (address ranges in a tasks
address space.) Due to the need to hold the task->mm->mmap_sem semaphore while
updating vma's, the rebinding of vma mempolicies has to be done when the
cpuset memory placement is changed, at which time mmap_sem can be safely
acquired. The tasks mempolicy is rebound later, when the task next attempts
to allocate memory and notices that its task->cpuset_mems_generation is
out-of-date with its cpusets mems_generation.
Because walking the tasklist to find all tasks attached to a changing cpuset
requires holding tasklist_lock, a spinlock, one cannot update the vma's of the
affected tasks while doing the tasklist scan. In general, one cannot acquire
a semaphore (which can sleep) while already holding a spinlock (such as
tasklist_lock). So a list of mm references has to be built up during the
tasklist scan, then the tasklist lock dropped, then for each mm, its mmap_sem
acquired, and the vma's in that mm rebound.
Once the tasklist lock is dropped, affected tasks may fork new tasks, before
their mm's are rebound. A kernel global 'cpuset_being_rebound' is set to
point to the cpuset being rebound (there can only be one; cpuset modifications
are done under a global 'manage_sem' semaphore), and the mpol_copy code that
is used to copy a tasks mempolicies during fork catches such forking tasks,
and ensures their children are also rebound.
When a task is moved to a different cpuset, it is easier, as there is only one
task involved. It's mm->vma's are scanned, using the same
mpol_rebind_policy() as used above.
It may happen that both the mpol_copy hook and the update done via the
tasklist scan update the same mm twice. This is ok, as the mempolicies of
each vma in an mm keep track of what mems_allowed they are relative to, and
safely no-op a second request to rebind to the same nodes.
Signed-off-by: Paul Jackson <pj@sgi.com>
Signed-off-by: Andrew Morton <akpm@osdl.org>
Signed-off-by: Linus Torvalds <torvalds@osdl.org>
2006-01-08 02:01:59 -07:00
|
|
|
set_cpuset_being_rebound(cs); /* causes mpol_copy() rebind */
|
|
|
|
|
|
|
|
fudge = 10; /* spare mmarray[] slots */
|
|
|
|
fudge += cpus_weight(cs->cpus_allowed); /* imagine one fork-bomb/cpu */
|
|
|
|
retval = -ENOMEM;
|
|
|
|
|
|
|
|
/*
|
|
|
|
* Allocate mmarray[] to hold mm reference for each task
|
|
|
|
* in cpuset cs. Can't kmalloc GFP_KERNEL while holding
|
|
|
|
* tasklist_lock. We could use GFP_ATOMIC, but with a
|
|
|
|
* few more lines of code, we can retry until we get a big
|
|
|
|
* enough mmarray[] w/o using GFP_ATOMIC.
|
|
|
|
*/
|
|
|
|
while (1) {
|
|
|
|
ntasks = atomic_read(&cs->count); /* guess */
|
|
|
|
ntasks += fudge;
|
|
|
|
mmarray = kmalloc(ntasks * sizeof(*mmarray), GFP_KERNEL);
|
|
|
|
if (!mmarray)
|
|
|
|
goto done;
|
|
|
|
write_lock_irq(&tasklist_lock); /* block fork */
|
|
|
|
if (atomic_read(&cs->count) <= ntasks)
|
|
|
|
break; /* got enough */
|
|
|
|
write_unlock_irq(&tasklist_lock); /* try again */
|
|
|
|
kfree(mmarray);
|
|
|
|
}
|
|
|
|
|
|
|
|
n = 0;
|
|
|
|
|
|
|
|
/* Load up mmarray[] with mm reference for each task in cpuset. */
|
|
|
|
do_each_thread(g, p) {
|
|
|
|
struct mm_struct *mm;
|
|
|
|
|
|
|
|
if (n >= ntasks) {
|
|
|
|
printk(KERN_WARNING
|
|
|
|
"Cpuset mempolicy rebind incomplete.\n");
|
|
|
|
continue;
|
|
|
|
}
|
|
|
|
if (p->cpuset != cs)
|
|
|
|
continue;
|
|
|
|
mm = get_task_mm(p);
|
|
|
|
if (!mm)
|
|
|
|
continue;
|
|
|
|
mmarray[n++] = mm;
|
|
|
|
} while_each_thread(g, p);
|
|
|
|
write_unlock_irq(&tasklist_lock);
|
|
|
|
|
|
|
|
/*
|
|
|
|
* Now that we've dropped the tasklist spinlock, we can
|
|
|
|
* rebind the vma mempolicies of each mm in mmarray[] to their
|
|
|
|
* new cpuset, and release that mm. The mpol_rebind_mm()
|
|
|
|
* call takes mmap_sem, which we couldn't take while holding
|
|
|
|
* tasklist_lock. Forks can happen again now - the mpol_copy()
|
|
|
|
* cpuset_being_rebound check will catch such forks, and rebind
|
|
|
|
* their vma mempolicies too. Because we still hold the global
|
2006-03-23 04:00:18 -07:00
|
|
|
* cpuset manage_mutex, we know that no other rebind effort will
|
[PATCH] cpuset: rebind vma mempolicies fix
Fix more of longstanding bug in cpuset/mempolicy interaction.
NUMA mempolicies (mm/mempolicy.c) are constrained by the current tasks cpuset
to just the Memory Nodes allowed by that cpuset. The kernel maintains
internal state for each mempolicy, tracking what nodes are used for the
MPOL_INTERLEAVE, MPOL_BIND or MPOL_PREFERRED policies.
When a tasks cpuset memory placement changes, whether because the cpuset
changed, or because the task was attached to a different cpuset, then the
tasks mempolicies have to be rebound to the new cpuset placement, so as to
preserve the cpuset-relative numbering of the nodes in that policy.
An earlier fix handled such mempolicy rebinding for mempolicies attached to a
task.
This fix rebinds mempolicies attached to vma's (address ranges in a tasks
address space.) Due to the need to hold the task->mm->mmap_sem semaphore while
updating vma's, the rebinding of vma mempolicies has to be done when the
cpuset memory placement is changed, at which time mmap_sem can be safely
acquired. The tasks mempolicy is rebound later, when the task next attempts
to allocate memory and notices that its task->cpuset_mems_generation is
out-of-date with its cpusets mems_generation.
Because walking the tasklist to find all tasks attached to a changing cpuset
requires holding tasklist_lock, a spinlock, one cannot update the vma's of the
affected tasks while doing the tasklist scan. In general, one cannot acquire
a semaphore (which can sleep) while already holding a spinlock (such as
tasklist_lock). So a list of mm references has to be built up during the
tasklist scan, then the tasklist lock dropped, then for each mm, its mmap_sem
acquired, and the vma's in that mm rebound.
Once the tasklist lock is dropped, affected tasks may fork new tasks, before
their mm's are rebound. A kernel global 'cpuset_being_rebound' is set to
point to the cpuset being rebound (there can only be one; cpuset modifications
are done under a global 'manage_sem' semaphore), and the mpol_copy code that
is used to copy a tasks mempolicies during fork catches such forking tasks,
and ensures their children are also rebound.
When a task is moved to a different cpuset, it is easier, as there is only one
task involved. It's mm->vma's are scanned, using the same
mpol_rebind_policy() as used above.
It may happen that both the mpol_copy hook and the update done via the
tasklist scan update the same mm twice. This is ok, as the mempolicies of
each vma in an mm keep track of what mems_allowed they are relative to, and
safely no-op a second request to rebind to the same nodes.
Signed-off-by: Paul Jackson <pj@sgi.com>
Signed-off-by: Andrew Morton <akpm@osdl.org>
Signed-off-by: Linus Torvalds <torvalds@osdl.org>
2006-01-08 02:01:59 -07:00
|
|
|
* be contending for the global variable cpuset_being_rebound.
|
|
|
|
* It's ok if we rebind the same mm twice; mpol_rebind_mm()
|
2006-01-08 02:02:00 -07:00
|
|
|
* is idempotent. Also migrate pages in each mm to new nodes.
|
[PATCH] cpuset: rebind vma mempolicies fix
Fix more of longstanding bug in cpuset/mempolicy interaction.
NUMA mempolicies (mm/mempolicy.c) are constrained by the current tasks cpuset
to just the Memory Nodes allowed by that cpuset. The kernel maintains
internal state for each mempolicy, tracking what nodes are used for the
MPOL_INTERLEAVE, MPOL_BIND or MPOL_PREFERRED policies.
When a tasks cpuset memory placement changes, whether because the cpuset
changed, or because the task was attached to a different cpuset, then the
tasks mempolicies have to be rebound to the new cpuset placement, so as to
preserve the cpuset-relative numbering of the nodes in that policy.
An earlier fix handled such mempolicy rebinding for mempolicies attached to a
task.
This fix rebinds mempolicies attached to vma's (address ranges in a tasks
address space.) Due to the need to hold the task->mm->mmap_sem semaphore while
updating vma's, the rebinding of vma mempolicies has to be done when the
cpuset memory placement is changed, at which time mmap_sem can be safely
acquired. The tasks mempolicy is rebound later, when the task next attempts
to allocate memory and notices that its task->cpuset_mems_generation is
out-of-date with its cpusets mems_generation.
Because walking the tasklist to find all tasks attached to a changing cpuset
requires holding tasklist_lock, a spinlock, one cannot update the vma's of the
affected tasks while doing the tasklist scan. In general, one cannot acquire
a semaphore (which can sleep) while already holding a spinlock (such as
tasklist_lock). So a list of mm references has to be built up during the
tasklist scan, then the tasklist lock dropped, then for each mm, its mmap_sem
acquired, and the vma's in that mm rebound.
Once the tasklist lock is dropped, affected tasks may fork new tasks, before
their mm's are rebound. A kernel global 'cpuset_being_rebound' is set to
point to the cpuset being rebound (there can only be one; cpuset modifications
are done under a global 'manage_sem' semaphore), and the mpol_copy code that
is used to copy a tasks mempolicies during fork catches such forking tasks,
and ensures their children are also rebound.
When a task is moved to a different cpuset, it is easier, as there is only one
task involved. It's mm->vma's are scanned, using the same
mpol_rebind_policy() as used above.
It may happen that both the mpol_copy hook and the update done via the
tasklist scan update the same mm twice. This is ok, as the mempolicies of
each vma in an mm keep track of what mems_allowed they are relative to, and
safely no-op a second request to rebind to the same nodes.
Signed-off-by: Paul Jackson <pj@sgi.com>
Signed-off-by: Andrew Morton <akpm@osdl.org>
Signed-off-by: Linus Torvalds <torvalds@osdl.org>
2006-01-08 02:01:59 -07:00
|
|
|
*/
|
2006-01-08 02:02:00 -07:00
|
|
|
migrate = is_memory_migrate(cs);
|
[PATCH] cpuset: rebind vma mempolicies fix
Fix more of longstanding bug in cpuset/mempolicy interaction.
NUMA mempolicies (mm/mempolicy.c) are constrained by the current tasks cpuset
to just the Memory Nodes allowed by that cpuset. The kernel maintains
internal state for each mempolicy, tracking what nodes are used for the
MPOL_INTERLEAVE, MPOL_BIND or MPOL_PREFERRED policies.
When a tasks cpuset memory placement changes, whether because the cpuset
changed, or because the task was attached to a different cpuset, then the
tasks mempolicies have to be rebound to the new cpuset placement, so as to
preserve the cpuset-relative numbering of the nodes in that policy.
An earlier fix handled such mempolicy rebinding for mempolicies attached to a
task.
This fix rebinds mempolicies attached to vma's (address ranges in a tasks
address space.) Due to the need to hold the task->mm->mmap_sem semaphore while
updating vma's, the rebinding of vma mempolicies has to be done when the
cpuset memory placement is changed, at which time mmap_sem can be safely
acquired. The tasks mempolicy is rebound later, when the task next attempts
to allocate memory and notices that its task->cpuset_mems_generation is
out-of-date with its cpusets mems_generation.
Because walking the tasklist to find all tasks attached to a changing cpuset
requires holding tasklist_lock, a spinlock, one cannot update the vma's of the
affected tasks while doing the tasklist scan. In general, one cannot acquire
a semaphore (which can sleep) while already holding a spinlock (such as
tasklist_lock). So a list of mm references has to be built up during the
tasklist scan, then the tasklist lock dropped, then for each mm, its mmap_sem
acquired, and the vma's in that mm rebound.
Once the tasklist lock is dropped, affected tasks may fork new tasks, before
their mm's are rebound. A kernel global 'cpuset_being_rebound' is set to
point to the cpuset being rebound (there can only be one; cpuset modifications
are done under a global 'manage_sem' semaphore), and the mpol_copy code that
is used to copy a tasks mempolicies during fork catches such forking tasks,
and ensures their children are also rebound.
When a task is moved to a different cpuset, it is easier, as there is only one
task involved. It's mm->vma's are scanned, using the same
mpol_rebind_policy() as used above.
It may happen that both the mpol_copy hook and the update done via the
tasklist scan update the same mm twice. This is ok, as the mempolicies of
each vma in an mm keep track of what mems_allowed they are relative to, and
safely no-op a second request to rebind to the same nodes.
Signed-off-by: Paul Jackson <pj@sgi.com>
Signed-off-by: Andrew Morton <akpm@osdl.org>
Signed-off-by: Linus Torvalds <torvalds@osdl.org>
2006-01-08 02:01:59 -07:00
|
|
|
for (i = 0; i < n; i++) {
|
|
|
|
struct mm_struct *mm = mmarray[i];
|
|
|
|
|
|
|
|
mpol_rebind_mm(mm, &cs->mems_allowed);
|
2006-03-31 03:30:52 -07:00
|
|
|
if (migrate)
|
|
|
|
cpuset_migrate_mm(mm, &oldmem, &cs->mems_allowed);
|
[PATCH] cpuset: rebind vma mempolicies fix
Fix more of longstanding bug in cpuset/mempolicy interaction.
NUMA mempolicies (mm/mempolicy.c) are constrained by the current tasks cpuset
to just the Memory Nodes allowed by that cpuset. The kernel maintains
internal state for each mempolicy, tracking what nodes are used for the
MPOL_INTERLEAVE, MPOL_BIND or MPOL_PREFERRED policies.
When a tasks cpuset memory placement changes, whether because the cpuset
changed, or because the task was attached to a different cpuset, then the
tasks mempolicies have to be rebound to the new cpuset placement, so as to
preserve the cpuset-relative numbering of the nodes in that policy.
An earlier fix handled such mempolicy rebinding for mempolicies attached to a
task.
This fix rebinds mempolicies attached to vma's (address ranges in a tasks
address space.) Due to the need to hold the task->mm->mmap_sem semaphore while
updating vma's, the rebinding of vma mempolicies has to be done when the
cpuset memory placement is changed, at which time mmap_sem can be safely
acquired. The tasks mempolicy is rebound later, when the task next attempts
to allocate memory and notices that its task->cpuset_mems_generation is
out-of-date with its cpusets mems_generation.
Because walking the tasklist to find all tasks attached to a changing cpuset
requires holding tasklist_lock, a spinlock, one cannot update the vma's of the
affected tasks while doing the tasklist scan. In general, one cannot acquire
a semaphore (which can sleep) while already holding a spinlock (such as
tasklist_lock). So a list of mm references has to be built up during the
tasklist scan, then the tasklist lock dropped, then for each mm, its mmap_sem
acquired, and the vma's in that mm rebound.
Once the tasklist lock is dropped, affected tasks may fork new tasks, before
their mm's are rebound. A kernel global 'cpuset_being_rebound' is set to
point to the cpuset being rebound (there can only be one; cpuset modifications
are done under a global 'manage_sem' semaphore), and the mpol_copy code that
is used to copy a tasks mempolicies during fork catches such forking tasks,
and ensures their children are also rebound.
When a task is moved to a different cpuset, it is easier, as there is only one
task involved. It's mm->vma's are scanned, using the same
mpol_rebind_policy() as used above.
It may happen that both the mpol_copy hook and the update done via the
tasklist scan update the same mm twice. This is ok, as the mempolicies of
each vma in an mm keep track of what mems_allowed they are relative to, and
safely no-op a second request to rebind to the same nodes.
Signed-off-by: Paul Jackson <pj@sgi.com>
Signed-off-by: Andrew Morton <akpm@osdl.org>
Signed-off-by: Linus Torvalds <torvalds@osdl.org>
2006-01-08 02:01:59 -07:00
|
|
|
mmput(mm);
|
|
|
|
}
|
|
|
|
|
|
|
|
/* We're done rebinding vma's to this cpusets new mems_allowed. */
|
|
|
|
kfree(mmarray);
|
|
|
|
set_cpuset_being_rebound(NULL);
|
|
|
|
retval = 0;
|
2006-01-08 02:01:52 -07:00
|
|
|
done:
|
2005-04-16 16:20:36 -06:00
|
|
|
return retval;
|
|
|
|
}
|
|
|
|
|
[PATCH] cpuset: memory pressure meter
Provide a simple per-cpuset metric of memory pressure, tracking the -rate-
that the tasks in a cpuset call try_to_free_pages(), the synchronous
(direct) memory reclaim code.
This enables batch managers monitoring jobs running in dedicated cpusets to
efficiently detect what level of memory pressure that job is causing.
This is useful both on tightly managed systems running a wide mix of
submitted jobs, which may choose to terminate or reprioritize jobs that are
trying to use more memory than allowed on the nodes assigned them, and with
tightly coupled, long running, massively parallel scientific computing jobs
that will dramatically fail to meet required performance goals if they
start to use more memory than allowed to them.
This patch just provides a very economical way for the batch manager to
monitor a cpuset for signs of memory pressure. It's up to the batch
manager or other user code to decide what to do about it and take action.
==> Unless this feature is enabled by writing "1" to the special file
/dev/cpuset/memory_pressure_enabled, the hook in the rebalance
code of __alloc_pages() for this metric reduces to simply noticing
that the cpuset_memory_pressure_enabled flag is zero. So only
systems that enable this feature will compute the metric.
Why a per-cpuset, running average:
Because this meter is per-cpuset, rather than per-task or mm, the
system load imposed by a batch scheduler monitoring this metric is
sharply reduced on large systems, because a scan of the tasklist can be
avoided on each set of queries.
Because this meter is a running average, instead of an accumulating
counter, a batch scheduler can detect memory pressure with a single
read, instead of having to read and accumulate results for a period of
time.
Because this meter is per-cpuset rather than per-task or mm, the
batch scheduler can obtain the key information, memory pressure in a
cpuset, with a single read, rather than having to query and accumulate
results over all the (dynamically changing) set of tasks in the cpuset.
A per-cpuset simple digital filter (requires a spinlock and 3 words of data
per-cpuset) is kept, and updated by any task attached to that cpuset, if it
enters the synchronous (direct) page reclaim code.
A per-cpuset file provides an integer number representing the recent
(half-life of 10 seconds) rate of direct page reclaims caused by the tasks
in the cpuset, in units of reclaims attempted per second, times 1000.
Signed-off-by: Paul Jackson <pj@sgi.com>
Signed-off-by: Andrew Morton <akpm@osdl.org>
Signed-off-by: Linus Torvalds <torvalds@osdl.org>
2006-01-08 02:01:49 -07:00
|
|
|
/*
|
2006-03-23 04:00:18 -07:00
|
|
|
* Call with manage_mutex held.
|
[PATCH] cpuset: memory pressure meter
Provide a simple per-cpuset metric of memory pressure, tracking the -rate-
that the tasks in a cpuset call try_to_free_pages(), the synchronous
(direct) memory reclaim code.
This enables batch managers monitoring jobs running in dedicated cpusets to
efficiently detect what level of memory pressure that job is causing.
This is useful both on tightly managed systems running a wide mix of
submitted jobs, which may choose to terminate or reprioritize jobs that are
trying to use more memory than allowed on the nodes assigned them, and with
tightly coupled, long running, massively parallel scientific computing jobs
that will dramatically fail to meet required performance goals if they
start to use more memory than allowed to them.
This patch just provides a very economical way for the batch manager to
monitor a cpuset for signs of memory pressure. It's up to the batch
manager or other user code to decide what to do about it and take action.
==> Unless this feature is enabled by writing "1" to the special file
/dev/cpuset/memory_pressure_enabled, the hook in the rebalance
code of __alloc_pages() for this metric reduces to simply noticing
that the cpuset_memory_pressure_enabled flag is zero. So only
systems that enable this feature will compute the metric.
Why a per-cpuset, running average:
Because this meter is per-cpuset, rather than per-task or mm, the
system load imposed by a batch scheduler monitoring this metric is
sharply reduced on large systems, because a scan of the tasklist can be
avoided on each set of queries.
Because this meter is a running average, instead of an accumulating
counter, a batch scheduler can detect memory pressure with a single
read, instead of having to read and accumulate results for a period of
time.
Because this meter is per-cpuset rather than per-task or mm, the
batch scheduler can obtain the key information, memory pressure in a
cpuset, with a single read, rather than having to query and accumulate
results over all the (dynamically changing) set of tasks in the cpuset.
A per-cpuset simple digital filter (requires a spinlock and 3 words of data
per-cpuset) is kept, and updated by any task attached to that cpuset, if it
enters the synchronous (direct) page reclaim code.
A per-cpuset file provides an integer number representing the recent
(half-life of 10 seconds) rate of direct page reclaims caused by the tasks
in the cpuset, in units of reclaims attempted per second, times 1000.
Signed-off-by: Paul Jackson <pj@sgi.com>
Signed-off-by: Andrew Morton <akpm@osdl.org>
Signed-off-by: Linus Torvalds <torvalds@osdl.org>
2006-01-08 02:01:49 -07:00
|
|
|
*/
|
|
|
|
|
|
|
|
static int update_memory_pressure_enabled(struct cpuset *cs, char *buf)
|
|
|
|
{
|
|
|
|
if (simple_strtoul(buf, NULL, 10) != 0)
|
|
|
|
cpuset_memory_pressure_enabled = 1;
|
|
|
|
else
|
|
|
|
cpuset_memory_pressure_enabled = 0;
|
|
|
|
return 0;
|
|
|
|
}
|
|
|
|
|
2005-04-16 16:20:36 -06:00
|
|
|
/*
|
|
|
|
* update_flag - read a 0 or a 1 in a file and update associated flag
|
|
|
|
* bit: the bit to update (CS_CPU_EXCLUSIVE, CS_MEM_EXCLUSIVE,
|
[PATCH] cpuset memory spread basic implementation
This patch provides the implementation and cpuset interface for an alternative
memory allocation policy that can be applied to certain kinds of memory
allocations, such as the page cache (file system buffers) and some slab caches
(such as inode caches).
The policy is called "memory spreading." If enabled, it spreads out these
kinds of memory allocations over all the nodes allowed to a task, instead of
preferring to place them on the node where the task is executing.
All other kinds of allocations, including anonymous pages for a tasks stack
and data regions, are not affected by this policy choice, and continue to be
allocated preferring the node local to execution, as modified by the NUMA
mempolicy.
There are two boolean flag files per cpuset that control where the kernel
allocates pages for the file system buffers and related in kernel data
structures. They are called 'memory_spread_page' and 'memory_spread_slab'.
If the per-cpuset boolean flag file 'memory_spread_page' is set, then the
kernel will spread the file system buffers (page cache) evenly over all the
nodes that the faulting task is allowed to use, instead of preferring to put
those pages on the node where the task is running.
If the per-cpuset boolean flag file 'memory_spread_slab' is set, then the
kernel will spread some file system related slab caches, such as for inodes
and dentries evenly over all the nodes that the faulting task is allowed to
use, instead of preferring to put those pages on the node where the task is
running.
The implementation is simple. Setting the cpuset flags 'memory_spread_page'
or 'memory_spread_cache' turns on the per-process flags PF_SPREAD_PAGE or
PF_SPREAD_SLAB, respectively, for each task that is in the cpuset or
subsequently joins that cpuset. In subsequent patches, the page allocation
calls for the affected page cache and slab caches are modified to perform an
inline check for these flags, and if set, a call to a new routine
cpuset_mem_spread_node() returns the node to prefer for the allocation.
The cpuset_mem_spread_node() routine is also simple. It uses the value of a
per-task rotor cpuset_mem_spread_rotor to select the next node in the current
tasks mems_allowed to prefer for the allocation.
This policy can provide substantial improvements for jobs that need to place
thread local data on the corresponding node, but that need to access large
file system data sets that need to be spread across the several nodes in the
jobs cpuset in order to fit. Without this patch, especially for jobs that
might have one thread reading in the data set, the memory allocation across
the nodes in the jobs cpuset can become very uneven.
A couple of Copyright year ranges are updated as well. And a couple of email
addresses that can be found in the MAINTAINERS file are removed.
Signed-off-by: Paul Jackson <pj@sgi.com>
Signed-off-by: Andrew Morton <akpm@osdl.org>
Signed-off-by: Linus Torvalds <torvalds@osdl.org>
2006-03-24 04:16:03 -07:00
|
|
|
* CS_NOTIFY_ON_RELEASE, CS_MEMORY_MIGRATE,
|
|
|
|
* CS_SPREAD_PAGE, CS_SPREAD_SLAB)
|
2005-04-16 16:20:36 -06:00
|
|
|
* cs: the cpuset to update
|
|
|
|
* buf: the buffer where we read the 0 or 1
|
[PATCH] cpusets: dual semaphore locking overhaul
Overhaul cpuset locking. Replace single semaphore with two semaphores.
The suggestion to use two locks was made by Roman Zippel.
Both locks are global. Code that wants to modify cpusets must first
acquire the exclusive manage_sem, which allows them read-only access to
cpusets, and holds off other would-be modifiers. Before making actual
changes, the second semaphore, callback_sem must be acquired as well. Code
that needs only to query cpusets must acquire callback_sem, which is also a
global exclusive lock.
The earlier problems with double tripping are avoided, because it is
allowed for holders of manage_sem to nest the second callback_sem lock, and
only callback_sem is needed by code called from within __alloc_pages(),
where the double tripping had been possible.
This is not quite the same as a normal read/write semaphore, because
obtaining read-only access with intent to change must hold off other such
attempts, while allowing read-only access w/o such intention. Changing
cpusets involves several related checks and changes, which must be done
while allowing read-only queries (to avoid the double trip), but while
ensuring nothing changes (holding off other would be modifiers.)
This overhaul of cpuset locking also makes careful use of task_lock() to
guard access to the task->cpuset pointer, closing a couple of race
conditions noticed while reading this code (thanks, Roman). I've never
seen these races fail in any use or test.
See further the comments in the code.
Signed-off-by: Paul Jackson <pj@sgi.com>
Signed-off-by: Andrew Morton <akpm@osdl.org>
Signed-off-by: Linus Torvalds <torvalds@osdl.org>
2005-10-30 16:02:30 -07:00
|
|
|
*
|
2006-03-23 04:00:18 -07:00
|
|
|
* Call with manage_mutex held.
|
2005-04-16 16:20:36 -06:00
|
|
|
*/
|
|
|
|
|
|
|
|
static int update_flag(cpuset_flagbits_t bit, struct cpuset *cs, char *buf)
|
|
|
|
{
|
|
|
|
int turning_on;
|
|
|
|
struct cpuset trialcs;
|
2005-06-25 15:57:34 -06:00
|
|
|
int err, cpu_exclusive_changed;
|
2005-04-16 16:20:36 -06:00
|
|
|
|
|
|
|
turning_on = (simple_strtoul(buf, NULL, 10) != 0);
|
|
|
|
|
|
|
|
trialcs = *cs;
|
|
|
|
if (turning_on)
|
|
|
|
set_bit(bit, &trialcs.flags);
|
|
|
|
else
|
|
|
|
clear_bit(bit, &trialcs.flags);
|
|
|
|
|
|
|
|
err = validate_change(cs, &trialcs);
|
2005-06-25 15:57:34 -06:00
|
|
|
if (err < 0)
|
|
|
|
return err;
|
|
|
|
cpu_exclusive_changed =
|
|
|
|
(is_cpu_exclusive(cs) != is_cpu_exclusive(&trialcs));
|
2006-03-23 04:00:18 -07:00
|
|
|
mutex_lock(&callback_mutex);
|
2005-06-25 15:57:34 -06:00
|
|
|
if (turning_on)
|
|
|
|
set_bit(bit, &cs->flags);
|
|
|
|
else
|
|
|
|
clear_bit(bit, &cs->flags);
|
2006-03-23 04:00:18 -07:00
|
|
|
mutex_unlock(&callback_mutex);
|
2005-06-25 15:57:34 -06:00
|
|
|
|
|
|
|
if (cpu_exclusive_changed)
|
|
|
|
update_cpu_domains(cs);
|
|
|
|
return 0;
|
2005-04-16 16:20:36 -06:00
|
|
|
}
|
|
|
|
|
[PATCH] cpuset: memory pressure meter
Provide a simple per-cpuset metric of memory pressure, tracking the -rate-
that the tasks in a cpuset call try_to_free_pages(), the synchronous
(direct) memory reclaim code.
This enables batch managers monitoring jobs running in dedicated cpusets to
efficiently detect what level of memory pressure that job is causing.
This is useful both on tightly managed systems running a wide mix of
submitted jobs, which may choose to terminate or reprioritize jobs that are
trying to use more memory than allowed on the nodes assigned them, and with
tightly coupled, long running, massively parallel scientific computing jobs
that will dramatically fail to meet required performance goals if they
start to use more memory than allowed to them.
This patch just provides a very economical way for the batch manager to
monitor a cpuset for signs of memory pressure. It's up to the batch
manager or other user code to decide what to do about it and take action.
==> Unless this feature is enabled by writing "1" to the special file
/dev/cpuset/memory_pressure_enabled, the hook in the rebalance
code of __alloc_pages() for this metric reduces to simply noticing
that the cpuset_memory_pressure_enabled flag is zero. So only
systems that enable this feature will compute the metric.
Why a per-cpuset, running average:
Because this meter is per-cpuset, rather than per-task or mm, the
system load imposed by a batch scheduler monitoring this metric is
sharply reduced on large systems, because a scan of the tasklist can be
avoided on each set of queries.
Because this meter is a running average, instead of an accumulating
counter, a batch scheduler can detect memory pressure with a single
read, instead of having to read and accumulate results for a period of
time.
Because this meter is per-cpuset rather than per-task or mm, the
batch scheduler can obtain the key information, memory pressure in a
cpuset, with a single read, rather than having to query and accumulate
results over all the (dynamically changing) set of tasks in the cpuset.
A per-cpuset simple digital filter (requires a spinlock and 3 words of data
per-cpuset) is kept, and updated by any task attached to that cpuset, if it
enters the synchronous (direct) page reclaim code.
A per-cpuset file provides an integer number representing the recent
(half-life of 10 seconds) rate of direct page reclaims caused by the tasks
in the cpuset, in units of reclaims attempted per second, times 1000.
Signed-off-by: Paul Jackson <pj@sgi.com>
Signed-off-by: Andrew Morton <akpm@osdl.org>
Signed-off-by: Linus Torvalds <torvalds@osdl.org>
2006-01-08 02:01:49 -07:00
|
|
|
/*
|
2006-06-30 10:27:16 -06:00
|
|
|
* Frequency meter - How fast is some event occurring?
|
[PATCH] cpuset: memory pressure meter
Provide a simple per-cpuset metric of memory pressure, tracking the -rate-
that the tasks in a cpuset call try_to_free_pages(), the synchronous
(direct) memory reclaim code.
This enables batch managers monitoring jobs running in dedicated cpusets to
efficiently detect what level of memory pressure that job is causing.
This is useful both on tightly managed systems running a wide mix of
submitted jobs, which may choose to terminate or reprioritize jobs that are
trying to use more memory than allowed on the nodes assigned them, and with
tightly coupled, long running, massively parallel scientific computing jobs
that will dramatically fail to meet required performance goals if they
start to use more memory than allowed to them.
This patch just provides a very economical way for the batch manager to
monitor a cpuset for signs of memory pressure. It's up to the batch
manager or other user code to decide what to do about it and take action.
==> Unless this feature is enabled by writing "1" to the special file
/dev/cpuset/memory_pressure_enabled, the hook in the rebalance
code of __alloc_pages() for this metric reduces to simply noticing
that the cpuset_memory_pressure_enabled flag is zero. So only
systems that enable this feature will compute the metric.
Why a per-cpuset, running average:
Because this meter is per-cpuset, rather than per-task or mm, the
system load imposed by a batch scheduler monitoring this metric is
sharply reduced on large systems, because a scan of the tasklist can be
avoided on each set of queries.
Because this meter is a running average, instead of an accumulating
counter, a batch scheduler can detect memory pressure with a single
read, instead of having to read and accumulate results for a period of
time.
Because this meter is per-cpuset rather than per-task or mm, the
batch scheduler can obtain the key information, memory pressure in a
cpuset, with a single read, rather than having to query and accumulate
results over all the (dynamically changing) set of tasks in the cpuset.
A per-cpuset simple digital filter (requires a spinlock and 3 words of data
per-cpuset) is kept, and updated by any task attached to that cpuset, if it
enters the synchronous (direct) page reclaim code.
A per-cpuset file provides an integer number representing the recent
(half-life of 10 seconds) rate of direct page reclaims caused by the tasks
in the cpuset, in units of reclaims attempted per second, times 1000.
Signed-off-by: Paul Jackson <pj@sgi.com>
Signed-off-by: Andrew Morton <akpm@osdl.org>
Signed-off-by: Linus Torvalds <torvalds@osdl.org>
2006-01-08 02:01:49 -07:00
|
|
|
*
|
|
|
|
* These routines manage a digitally filtered, constant time based,
|
|
|
|
* event frequency meter. There are four routines:
|
|
|
|
* fmeter_init() - initialize a frequency meter.
|
|
|
|
* fmeter_markevent() - called each time the event happens.
|
|
|
|
* fmeter_getrate() - returns the recent rate of such events.
|
|
|
|
* fmeter_update() - internal routine used to update fmeter.
|
|
|
|
*
|
|
|
|
* A common data structure is passed to each of these routines,
|
|
|
|
* which is used to keep track of the state required to manage the
|
|
|
|
* frequency meter and its digital filter.
|
|
|
|
*
|
|
|
|
* The filter works on the number of events marked per unit time.
|
|
|
|
* The filter is single-pole low-pass recursive (IIR). The time unit
|
|
|
|
* is 1 second. Arithmetic is done using 32-bit integers scaled to
|
|
|
|
* simulate 3 decimal digits of precision (multiplied by 1000).
|
|
|
|
*
|
|
|
|
* With an FM_COEF of 933, and a time base of 1 second, the filter
|
|
|
|
* has a half-life of 10 seconds, meaning that if the events quit
|
|
|
|
* happening, then the rate returned from the fmeter_getrate()
|
|
|
|
* will be cut in half each 10 seconds, until it converges to zero.
|
|
|
|
*
|
|
|
|
* It is not worth doing a real infinitely recursive filter. If more
|
|
|
|
* than FM_MAXTICKS ticks have elapsed since the last filter event,
|
|
|
|
* just compute FM_MAXTICKS ticks worth, by which point the level
|
|
|
|
* will be stable.
|
|
|
|
*
|
|
|
|
* Limit the count of unprocessed events to FM_MAXCNT, so as to avoid
|
|
|
|
* arithmetic overflow in the fmeter_update() routine.
|
|
|
|
*
|
|
|
|
* Given the simple 32 bit integer arithmetic used, this meter works
|
|
|
|
* best for reporting rates between one per millisecond (msec) and
|
|
|
|
* one per 32 (approx) seconds. At constant rates faster than one
|
|
|
|
* per msec it maxes out at values just under 1,000,000. At constant
|
|
|
|
* rates between one per msec, and one per second it will stabilize
|
|
|
|
* to a value N*1000, where N is the rate of events per second.
|
|
|
|
* At constant rates between one per second and one per 32 seconds,
|
|
|
|
* it will be choppy, moving up on the seconds that have an event,
|
|
|
|
* and then decaying until the next event. At rates slower than
|
|
|
|
* about one in 32 seconds, it decays all the way back to zero between
|
|
|
|
* each event.
|
|
|
|
*/
|
|
|
|
|
|
|
|
#define FM_COEF 933 /* coefficient for half-life of 10 secs */
|
|
|
|
#define FM_MAXTICKS ((time_t)99) /* useless computing more ticks than this */
|
|
|
|
#define FM_MAXCNT 1000000 /* limit cnt to avoid overflow */
|
|
|
|
#define FM_SCALE 1000 /* faux fixed point scale */
|
|
|
|
|
|
|
|
/* Initialize a frequency meter */
|
|
|
|
static void fmeter_init(struct fmeter *fmp)
|
|
|
|
{
|
|
|
|
fmp->cnt = 0;
|
|
|
|
fmp->val = 0;
|
|
|
|
fmp->time = 0;
|
|
|
|
spin_lock_init(&fmp->lock);
|
|
|
|
}
|
|
|
|
|
|
|
|
/* Internal meter update - process cnt events and update value */
|
|
|
|
static void fmeter_update(struct fmeter *fmp)
|
|
|
|
{
|
|
|
|
time_t now = get_seconds();
|
|
|
|
time_t ticks = now - fmp->time;
|
|
|
|
|
|
|
|
if (ticks == 0)
|
|
|
|
return;
|
|
|
|
|
|
|
|
ticks = min(FM_MAXTICKS, ticks);
|
|
|
|
while (ticks-- > 0)
|
|
|
|
fmp->val = (FM_COEF * fmp->val) / FM_SCALE;
|
|
|
|
fmp->time = now;
|
|
|
|
|
|
|
|
fmp->val += ((FM_SCALE - FM_COEF) * fmp->cnt) / FM_SCALE;
|
|
|
|
fmp->cnt = 0;
|
|
|
|
}
|
|
|
|
|
|
|
|
/* Process any previous ticks, then bump cnt by one (times scale). */
|
|
|
|
static void fmeter_markevent(struct fmeter *fmp)
|
|
|
|
{
|
|
|
|
spin_lock(&fmp->lock);
|
|
|
|
fmeter_update(fmp);
|
|
|
|
fmp->cnt = min(FM_MAXCNT, fmp->cnt + FM_SCALE);
|
|
|
|
spin_unlock(&fmp->lock);
|
|
|
|
}
|
|
|
|
|
|
|
|
/* Process any previous ticks, then return current value. */
|
|
|
|
static int fmeter_getrate(struct fmeter *fmp)
|
|
|
|
{
|
|
|
|
int val;
|
|
|
|
|
|
|
|
spin_lock(&fmp->lock);
|
|
|
|
fmeter_update(fmp);
|
|
|
|
val = fmp->val;
|
|
|
|
spin_unlock(&fmp->lock);
|
|
|
|
return val;
|
|
|
|
}
|
|
|
|
|
[PATCH] cpusets: dual semaphore locking overhaul
Overhaul cpuset locking. Replace single semaphore with two semaphores.
The suggestion to use two locks was made by Roman Zippel.
Both locks are global. Code that wants to modify cpusets must first
acquire the exclusive manage_sem, which allows them read-only access to
cpusets, and holds off other would-be modifiers. Before making actual
changes, the second semaphore, callback_sem must be acquired as well. Code
that needs only to query cpusets must acquire callback_sem, which is also a
global exclusive lock.
The earlier problems with double tripping are avoided, because it is
allowed for holders of manage_sem to nest the second callback_sem lock, and
only callback_sem is needed by code called from within __alloc_pages(),
where the double tripping had been possible.
This is not quite the same as a normal read/write semaphore, because
obtaining read-only access with intent to change must hold off other such
attempts, while allowing read-only access w/o such intention. Changing
cpusets involves several related checks and changes, which must be done
while allowing read-only queries (to avoid the double trip), but while
ensuring nothing changes (holding off other would be modifiers.)
This overhaul of cpuset locking also makes careful use of task_lock() to
guard access to the task->cpuset pointer, closing a couple of race
conditions noticed while reading this code (thanks, Roman). I've never
seen these races fail in any use or test.
See further the comments in the code.
Signed-off-by: Paul Jackson <pj@sgi.com>
Signed-off-by: Andrew Morton <akpm@osdl.org>
Signed-off-by: Linus Torvalds <torvalds@osdl.org>
2005-10-30 16:02:30 -07:00
|
|
|
/*
|
|
|
|
* Attack task specified by pid in 'pidbuf' to cpuset 'cs', possibly
|
|
|
|
* writing the path of the old cpuset in 'ppathbuf' if it needs to be
|
|
|
|
* notified on release.
|
|
|
|
*
|
2006-03-23 04:00:18 -07:00
|
|
|
* Call holding manage_mutex. May take callback_mutex and task_lock of
|
[PATCH] cpusets: dual semaphore locking overhaul
Overhaul cpuset locking. Replace single semaphore with two semaphores.
The suggestion to use two locks was made by Roman Zippel.
Both locks are global. Code that wants to modify cpusets must first
acquire the exclusive manage_sem, which allows them read-only access to
cpusets, and holds off other would-be modifiers. Before making actual
changes, the second semaphore, callback_sem must be acquired as well. Code
that needs only to query cpusets must acquire callback_sem, which is also a
global exclusive lock.
The earlier problems with double tripping are avoided, because it is
allowed for holders of manage_sem to nest the second callback_sem lock, and
only callback_sem is needed by code called from within __alloc_pages(),
where the double tripping had been possible.
This is not quite the same as a normal read/write semaphore, because
obtaining read-only access with intent to change must hold off other such
attempts, while allowing read-only access w/o such intention. Changing
cpusets involves several related checks and changes, which must be done
while allowing read-only queries (to avoid the double trip), but while
ensuring nothing changes (holding off other would be modifiers.)
This overhaul of cpuset locking also makes careful use of task_lock() to
guard access to the task->cpuset pointer, closing a couple of race
conditions noticed while reading this code (thanks, Roman). I've never
seen these races fail in any use or test.
See further the comments in the code.
Signed-off-by: Paul Jackson <pj@sgi.com>
Signed-off-by: Andrew Morton <akpm@osdl.org>
Signed-off-by: Linus Torvalds <torvalds@osdl.org>
2005-10-30 16:02:30 -07:00
|
|
|
* the task 'pid' during call.
|
|
|
|
*/
|
|
|
|
|
[PATCH] cpuset release ABBA deadlock fix
Fix possible cpuset_sem ABBA deadlock if 'notify_on_release' set.
For a particular usage pattern, creating and destroying cpusets fairly
frequently using notify_on_release, on a very large system, this deadlock
can be seen every few days. If you are not using the cpuset
notify_on_release feature, you will never see this deadlock.
The existing code, on task exit (or cpuset deletion) did:
get cpuset_sem
if cpuset marked notify_on_release and is ready to release:
compute cpuset path relative to /dev/cpuset mount point
call_usermodehelper() forks /sbin/cpuset_release_agent with path
drop cpuset_sem
Unfortunately, the fork in call_usermodehelper can allocate memory, and
allocating memory can require cpuset_sem, if the mems_generation values
changed in the interim. This results in an ABBA deadlock, trying to obtain
cpuset_sem when it is already held by the current task.
To fix this, I put the cpuset path (which must be computed while holding
cpuset_sem) in a temporary buffer, to be used in the call_usermodehelper
call of /sbin/cpuset_release_agent only _after_ dropping cpuset_sem.
So the new logic is:
get cpuset_sem
if cpuset marked notify_on_release and is ready to release:
compute cpuset path relative to /dev/cpuset mount point
stash path in kmalloc'd buffer
drop cpuset_sem
call_usermodehelper() forks /sbin/cpuset_release_agent with path
free path
The sharp eyed reader might notice that this patch does not contain any
calls to kmalloc. The existing code in the check_for_release() routine was
already kmalloc'ing a buffer to hold the cpuset path. In the old code, it
just held the buffer for a few lines, over the cpuset_release_agent() call
that in turn invoked call_usermodehelper(). In the new code, with the
application of this patch, it returns that buffer via the new char
**ppathbuf parameter, for later use and freeing in cpuset_release_agent(),
which is called after cpuset_sem is dropped. Whereas the old code has just
one call to cpuset_release_agent(), right in the check_for_release()
routine, the new code has three calls to cpuset_release_agent(), from the
various places that a cpuset can be released.
This patch has been build and booted on SN2, and passed a stress test that
previously hit the deadlock within a few seconds.
Signed-off-by: Paul Jackson <pj@sgi.com>
Signed-off-by: Andrew Morton <akpm@osdl.org>
Signed-off-by: Linus Torvalds <torvalds@osdl.org>
2005-08-09 11:07:59 -06:00
|
|
|
static int attach_task(struct cpuset *cs, char *pidbuf, char **ppathbuf)
|
2005-04-16 16:20:36 -06:00
|
|
|
{
|
|
|
|
pid_t pid;
|
|
|
|
struct task_struct *tsk;
|
|
|
|
struct cpuset *oldcs;
|
|
|
|
cpumask_t cpus;
|
[PATCH] cpusets: swap migration interface
Add a boolean "memory_migrate" to each cpuset, represented by a file
containing "0" or "1" in each directory below /dev/cpuset.
It defaults to false (file contains "0"). It can be set true by writing
"1" to the file.
If true, then anytime that a task is attached to the cpuset so marked, the
pages of that task will be moved to that cpuset, preserving, to the extent
practical, the cpuset-relative placement of the pages.
Also anytime that a cpuset so marked has its memory placement changed (by
writing to its "mems" file), the tasks in that cpuset will have their pages
moved to the cpusets new nodes, preserving, to the extent practical, the
cpuset-relative placement of the moved pages.
Signed-off-by: Paul Jackson <pj@sgi.com>
Cc: Christoph Lameter <christoph@lameter.com>
Signed-off-by: Andrew Morton <akpm@osdl.org>
Signed-off-by: Linus Torvalds <torvalds@osdl.org>
2006-01-08 02:00:56 -07:00
|
|
|
nodemask_t from, to;
|
[PATCH] cpuset: rebind vma mempolicies fix
Fix more of longstanding bug in cpuset/mempolicy interaction.
NUMA mempolicies (mm/mempolicy.c) are constrained by the current tasks cpuset
to just the Memory Nodes allowed by that cpuset. The kernel maintains
internal state for each mempolicy, tracking what nodes are used for the
MPOL_INTERLEAVE, MPOL_BIND or MPOL_PREFERRED policies.
When a tasks cpuset memory placement changes, whether because the cpuset
changed, or because the task was attached to a different cpuset, then the
tasks mempolicies have to be rebound to the new cpuset placement, so as to
preserve the cpuset-relative numbering of the nodes in that policy.
An earlier fix handled such mempolicy rebinding for mempolicies attached to a
task.
This fix rebinds mempolicies attached to vma's (address ranges in a tasks
address space.) Due to the need to hold the task->mm->mmap_sem semaphore while
updating vma's, the rebinding of vma mempolicies has to be done when the
cpuset memory placement is changed, at which time mmap_sem can be safely
acquired. The tasks mempolicy is rebound later, when the task next attempts
to allocate memory and notices that its task->cpuset_mems_generation is
out-of-date with its cpusets mems_generation.
Because walking the tasklist to find all tasks attached to a changing cpuset
requires holding tasklist_lock, a spinlock, one cannot update the vma's of the
affected tasks while doing the tasklist scan. In general, one cannot acquire
a semaphore (which can sleep) while already holding a spinlock (such as
tasklist_lock). So a list of mm references has to be built up during the
tasklist scan, then the tasklist lock dropped, then for each mm, its mmap_sem
acquired, and the vma's in that mm rebound.
Once the tasklist lock is dropped, affected tasks may fork new tasks, before
their mm's are rebound. A kernel global 'cpuset_being_rebound' is set to
point to the cpuset being rebound (there can only be one; cpuset modifications
are done under a global 'manage_sem' semaphore), and the mpol_copy code that
is used to copy a tasks mempolicies during fork catches such forking tasks,
and ensures their children are also rebound.
When a task is moved to a different cpuset, it is easier, as there is only one
task involved. It's mm->vma's are scanned, using the same
mpol_rebind_policy() as used above.
It may happen that both the mpol_copy hook and the update done via the
tasklist scan update the same mm twice. This is ok, as the mempolicies of
each vma in an mm keep track of what mems_allowed they are relative to, and
safely no-op a second request to rebind to the same nodes.
Signed-off-by: Paul Jackson <pj@sgi.com>
Signed-off-by: Andrew Morton <akpm@osdl.org>
Signed-off-by: Linus Torvalds <torvalds@osdl.org>
2006-01-08 02:01:59 -07:00
|
|
|
struct mm_struct *mm;
|
2006-06-23 03:04:00 -06:00
|
|
|
int retval;
|
2005-04-16 16:20:36 -06:00
|
|
|
|
[PATCH] cpuset release ABBA deadlock fix
Fix possible cpuset_sem ABBA deadlock if 'notify_on_release' set.
For a particular usage pattern, creating and destroying cpusets fairly
frequently using notify_on_release, on a very large system, this deadlock
can be seen every few days. If you are not using the cpuset
notify_on_release feature, you will never see this deadlock.
The existing code, on task exit (or cpuset deletion) did:
get cpuset_sem
if cpuset marked notify_on_release and is ready to release:
compute cpuset path relative to /dev/cpuset mount point
call_usermodehelper() forks /sbin/cpuset_release_agent with path
drop cpuset_sem
Unfortunately, the fork in call_usermodehelper can allocate memory, and
allocating memory can require cpuset_sem, if the mems_generation values
changed in the interim. This results in an ABBA deadlock, trying to obtain
cpuset_sem when it is already held by the current task.
To fix this, I put the cpuset path (which must be computed while holding
cpuset_sem) in a temporary buffer, to be used in the call_usermodehelper
call of /sbin/cpuset_release_agent only _after_ dropping cpuset_sem.
So the new logic is:
get cpuset_sem
if cpuset marked notify_on_release and is ready to release:
compute cpuset path relative to /dev/cpuset mount point
stash path in kmalloc'd buffer
drop cpuset_sem
call_usermodehelper() forks /sbin/cpuset_release_agent with path
free path
The sharp eyed reader might notice that this patch does not contain any
calls to kmalloc. The existing code in the check_for_release() routine was
already kmalloc'ing a buffer to hold the cpuset path. In the old code, it
just held the buffer for a few lines, over the cpuset_release_agent() call
that in turn invoked call_usermodehelper(). In the new code, with the
application of this patch, it returns that buffer via the new char
**ppathbuf parameter, for later use and freeing in cpuset_release_agent(),
which is called after cpuset_sem is dropped. Whereas the old code has just
one call to cpuset_release_agent(), right in the check_for_release()
routine, the new code has three calls to cpuset_release_agent(), from the
various places that a cpuset can be released.
This patch has been build and booted on SN2, and passed a stress test that
previously hit the deadlock within a few seconds.
Signed-off-by: Paul Jackson <pj@sgi.com>
Signed-off-by: Andrew Morton <akpm@osdl.org>
Signed-off-by: Linus Torvalds <torvalds@osdl.org>
2005-08-09 11:07:59 -06:00
|
|
|
if (sscanf(pidbuf, "%d", &pid) != 1)
|
2005-04-16 16:20:36 -06:00
|
|
|
return -EIO;
|
|
|
|
if (cpus_empty(cs->cpus_allowed) || nodes_empty(cs->mems_allowed))
|
|
|
|
return -ENOSPC;
|
|
|
|
|
|
|
|
if (pid) {
|
|
|
|
read_lock(&tasklist_lock);
|
|
|
|
|
|
|
|
tsk = find_task_by_pid(pid);
|
[PATCH] cpusets: dual semaphore locking overhaul
Overhaul cpuset locking. Replace single semaphore with two semaphores.
The suggestion to use two locks was made by Roman Zippel.
Both locks are global. Code that wants to modify cpusets must first
acquire the exclusive manage_sem, which allows them read-only access to
cpusets, and holds off other would-be modifiers. Before making actual
changes, the second semaphore, callback_sem must be acquired as well. Code
that needs only to query cpusets must acquire callback_sem, which is also a
global exclusive lock.
The earlier problems with double tripping are avoided, because it is
allowed for holders of manage_sem to nest the second callback_sem lock, and
only callback_sem is needed by code called from within __alloc_pages(),
where the double tripping had been possible.
This is not quite the same as a normal read/write semaphore, because
obtaining read-only access with intent to change must hold off other such
attempts, while allowing read-only access w/o such intention. Changing
cpusets involves several related checks and changes, which must be done
while allowing read-only queries (to avoid the double trip), but while
ensuring nothing changes (holding off other would be modifiers.)
This overhaul of cpuset locking also makes careful use of task_lock() to
guard access to the task->cpuset pointer, closing a couple of race
conditions noticed while reading this code (thanks, Roman). I've never
seen these races fail in any use or test.
See further the comments in the code.
Signed-off-by: Paul Jackson <pj@sgi.com>
Signed-off-by: Andrew Morton <akpm@osdl.org>
Signed-off-by: Linus Torvalds <torvalds@osdl.org>
2005-10-30 16:02:30 -07:00
|
|
|
if (!tsk || tsk->flags & PF_EXITING) {
|
2005-04-16 16:20:36 -06:00
|
|
|
read_unlock(&tasklist_lock);
|
|
|
|
return -ESRCH;
|
|
|
|
}
|
|
|
|
|
|
|
|
get_task_struct(tsk);
|
|
|
|
read_unlock(&tasklist_lock);
|
|
|
|
|
|
|
|
if ((current->euid) && (current->euid != tsk->uid)
|
|
|
|
&& (current->euid != tsk->suid)) {
|
|
|
|
put_task_struct(tsk);
|
|
|
|
return -EACCES;
|
|
|
|
}
|
|
|
|
} else {
|
|
|
|
tsk = current;
|
|
|
|
get_task_struct(tsk);
|
|
|
|
}
|
|
|
|
|
2006-06-23 03:04:00 -06:00
|
|
|
retval = security_task_setscheduler(tsk, 0, NULL);
|
|
|
|
if (retval) {
|
|
|
|
put_task_struct(tsk);
|
|
|
|
return retval;
|
|
|
|
}
|
|
|
|
|
2006-03-23 04:00:18 -07:00
|
|
|
mutex_lock(&callback_mutex);
|
[PATCH] cpusets: dual semaphore locking overhaul
Overhaul cpuset locking. Replace single semaphore with two semaphores.
The suggestion to use two locks was made by Roman Zippel.
Both locks are global. Code that wants to modify cpusets must first
acquire the exclusive manage_sem, which allows them read-only access to
cpusets, and holds off other would-be modifiers. Before making actual
changes, the second semaphore, callback_sem must be acquired as well. Code
that needs only to query cpusets must acquire callback_sem, which is also a
global exclusive lock.
The earlier problems with double tripping are avoided, because it is
allowed for holders of manage_sem to nest the second callback_sem lock, and
only callback_sem is needed by code called from within __alloc_pages(),
where the double tripping had been possible.
This is not quite the same as a normal read/write semaphore, because
obtaining read-only access with intent to change must hold off other such
attempts, while allowing read-only access w/o such intention. Changing
cpusets involves several related checks and changes, which must be done
while allowing read-only queries (to avoid the double trip), but while
ensuring nothing changes (holding off other would be modifiers.)
This overhaul of cpuset locking also makes careful use of task_lock() to
guard access to the task->cpuset pointer, closing a couple of race
conditions noticed while reading this code (thanks, Roman). I've never
seen these races fail in any use or test.
See further the comments in the code.
Signed-off-by: Paul Jackson <pj@sgi.com>
Signed-off-by: Andrew Morton <akpm@osdl.org>
Signed-off-by: Linus Torvalds <torvalds@osdl.org>
2005-10-30 16:02:30 -07:00
|
|
|
|
2005-04-16 16:20:36 -06:00
|
|
|
task_lock(tsk);
|
|
|
|
oldcs = tsk->cpuset;
|
|
|
|
if (!oldcs) {
|
|
|
|
task_unlock(tsk);
|
2006-03-23 04:00:18 -07:00
|
|
|
mutex_unlock(&callback_mutex);
|
2005-04-16 16:20:36 -06:00
|
|
|
put_task_struct(tsk);
|
|
|
|
return -ESRCH;
|
|
|
|
}
|
|
|
|
atomic_inc(&cs->count);
|
[PATCH] cpuset: use rcu directly optimization
Optimize the cpuset impact on page allocation, the most performance critical
cpuset hook in the kernel.
On each page allocation, the cpuset hook needs to check for a possible change
in the current tasks cpuset. It can now handle the common case, of no change,
without taking any spinlock or semaphore, thanks to RCU.
Convert a spinlock on the current task to an rcu_read_lock(), saving
approximately a memory barrier and an atomic op, depending on architecture.
This is done by adding rcu_assign_pointer() and synchronize_rcu() calls to the
write side of the task->cpuset pointer, in cpuset.c:attach_task(), to delay
freeing up a detached cpuset until after any critical sections referencing
that pointer.
Thanks to Andi Kleen, Nick Piggin and Eric Dumazet for ideas.
Signed-off-by: Paul Jackson <pj@sgi.com>
Signed-off-by: Andrew Morton <akpm@osdl.org>
Signed-off-by: Linus Torvalds <torvalds@osdl.org>
2006-01-08 02:02:02 -07:00
|
|
|
rcu_assign_pointer(tsk->cpuset, cs);
|
2005-04-16 16:20:36 -06:00
|
|
|
task_unlock(tsk);
|
|
|
|
|
|
|
|
guarantee_online_cpus(cs, &cpus);
|
|
|
|
set_cpus_allowed(tsk, cpus);
|
|
|
|
|
[PATCH] cpusets: swap migration interface
Add a boolean "memory_migrate" to each cpuset, represented by a file
containing "0" or "1" in each directory below /dev/cpuset.
It defaults to false (file contains "0"). It can be set true by writing
"1" to the file.
If true, then anytime that a task is attached to the cpuset so marked, the
pages of that task will be moved to that cpuset, preserving, to the extent
practical, the cpuset-relative placement of the pages.
Also anytime that a cpuset so marked has its memory placement changed (by
writing to its "mems" file), the tasks in that cpuset will have their pages
moved to the cpusets new nodes, preserving, to the extent practical, the
cpuset-relative placement of the moved pages.
Signed-off-by: Paul Jackson <pj@sgi.com>
Cc: Christoph Lameter <christoph@lameter.com>
Signed-off-by: Andrew Morton <akpm@osdl.org>
Signed-off-by: Linus Torvalds <torvalds@osdl.org>
2006-01-08 02:00:56 -07:00
|
|
|
from = oldcs->mems_allowed;
|
|
|
|
to = cs->mems_allowed;
|
|
|
|
|
2006-03-23 04:00:18 -07:00
|
|
|
mutex_unlock(&callback_mutex);
|
[PATCH] cpuset: rebind vma mempolicies fix
Fix more of longstanding bug in cpuset/mempolicy interaction.
NUMA mempolicies (mm/mempolicy.c) are constrained by the current tasks cpuset
to just the Memory Nodes allowed by that cpuset. The kernel maintains
internal state for each mempolicy, tracking what nodes are used for the
MPOL_INTERLEAVE, MPOL_BIND or MPOL_PREFERRED policies.
When a tasks cpuset memory placement changes, whether because the cpuset
changed, or because the task was attached to a different cpuset, then the
tasks mempolicies have to be rebound to the new cpuset placement, so as to
preserve the cpuset-relative numbering of the nodes in that policy.
An earlier fix handled such mempolicy rebinding for mempolicies attached to a
task.
This fix rebinds mempolicies attached to vma's (address ranges in a tasks
address space.) Due to the need to hold the task->mm->mmap_sem semaphore while
updating vma's, the rebinding of vma mempolicies has to be done when the
cpuset memory placement is changed, at which time mmap_sem can be safely
acquired. The tasks mempolicy is rebound later, when the task next attempts
to allocate memory and notices that its task->cpuset_mems_generation is
out-of-date with its cpusets mems_generation.
Because walking the tasklist to find all tasks attached to a changing cpuset
requires holding tasklist_lock, a spinlock, one cannot update the vma's of the
affected tasks while doing the tasklist scan. In general, one cannot acquire
a semaphore (which can sleep) while already holding a spinlock (such as
tasklist_lock). So a list of mm references has to be built up during the
tasklist scan, then the tasklist lock dropped, then for each mm, its mmap_sem
acquired, and the vma's in that mm rebound.
Once the tasklist lock is dropped, affected tasks may fork new tasks, before
their mm's are rebound. A kernel global 'cpuset_being_rebound' is set to
point to the cpuset being rebound (there can only be one; cpuset modifications
are done under a global 'manage_sem' semaphore), and the mpol_copy code that
is used to copy a tasks mempolicies during fork catches such forking tasks,
and ensures their children are also rebound.
When a task is moved to a different cpuset, it is easier, as there is only one
task involved. It's mm->vma's are scanned, using the same
mpol_rebind_policy() as used above.
It may happen that both the mpol_copy hook and the update done via the
tasklist scan update the same mm twice. This is ok, as the mempolicies of
each vma in an mm keep track of what mems_allowed they are relative to, and
safely no-op a second request to rebind to the same nodes.
Signed-off-by: Paul Jackson <pj@sgi.com>
Signed-off-by: Andrew Morton <akpm@osdl.org>
Signed-off-by: Linus Torvalds <torvalds@osdl.org>
2006-01-08 02:01:59 -07:00
|
|
|
|
|
|
|
mm = get_task_mm(tsk);
|
|
|
|
if (mm) {
|
|
|
|
mpol_rebind_mm(mm, &to);
|
2006-03-31 03:30:51 -07:00
|
|
|
if (is_memory_migrate(cs))
|
2006-03-31 03:30:52 -07:00
|
|
|
cpuset_migrate_mm(mm, &from, &to);
|
[PATCH] cpuset: rebind vma mempolicies fix
Fix more of longstanding bug in cpuset/mempolicy interaction.
NUMA mempolicies (mm/mempolicy.c) are constrained by the current tasks cpuset
to just the Memory Nodes allowed by that cpuset. The kernel maintains
internal state for each mempolicy, tracking what nodes are used for the
MPOL_INTERLEAVE, MPOL_BIND or MPOL_PREFERRED policies.
When a tasks cpuset memory placement changes, whether because the cpuset
changed, or because the task was attached to a different cpuset, then the
tasks mempolicies have to be rebound to the new cpuset placement, so as to
preserve the cpuset-relative numbering of the nodes in that policy.
An earlier fix handled such mempolicy rebinding for mempolicies attached to a
task.
This fix rebinds mempolicies attached to vma's (address ranges in a tasks
address space.) Due to the need to hold the task->mm->mmap_sem semaphore while
updating vma's, the rebinding of vma mempolicies has to be done when the
cpuset memory placement is changed, at which time mmap_sem can be safely
acquired. The tasks mempolicy is rebound later, when the task next attempts
to allocate memory and notices that its task->cpuset_mems_generation is
out-of-date with its cpusets mems_generation.
Because walking the tasklist to find all tasks attached to a changing cpuset
requires holding tasklist_lock, a spinlock, one cannot update the vma's of the
affected tasks while doing the tasklist scan. In general, one cannot acquire
a semaphore (which can sleep) while already holding a spinlock (such as
tasklist_lock). So a list of mm references has to be built up during the
tasklist scan, then the tasklist lock dropped, then for each mm, its mmap_sem
acquired, and the vma's in that mm rebound.
Once the tasklist lock is dropped, affected tasks may fork new tasks, before
their mm's are rebound. A kernel global 'cpuset_being_rebound' is set to
point to the cpuset being rebound (there can only be one; cpuset modifications
are done under a global 'manage_sem' semaphore), and the mpol_copy code that
is used to copy a tasks mempolicies during fork catches such forking tasks,
and ensures their children are also rebound.
When a task is moved to a different cpuset, it is easier, as there is only one
task involved. It's mm->vma's are scanned, using the same
mpol_rebind_policy() as used above.
It may happen that both the mpol_copy hook and the update done via the
tasklist scan update the same mm twice. This is ok, as the mempolicies of
each vma in an mm keep track of what mems_allowed they are relative to, and
safely no-op a second request to rebind to the same nodes.
Signed-off-by: Paul Jackson <pj@sgi.com>
Signed-off-by: Andrew Morton <akpm@osdl.org>
Signed-off-by: Linus Torvalds <torvalds@osdl.org>
2006-01-08 02:01:59 -07:00
|
|
|
mmput(mm);
|
|
|
|
}
|
|
|
|
|
2005-04-16 16:20:36 -06:00
|
|
|
put_task_struct(tsk);
|
[PATCH] cpuset: use rcu directly optimization
Optimize the cpuset impact on page allocation, the most performance critical
cpuset hook in the kernel.
On each page allocation, the cpuset hook needs to check for a possible change
in the current tasks cpuset. It can now handle the common case, of no change,
without taking any spinlock or semaphore, thanks to RCU.
Convert a spinlock on the current task to an rcu_read_lock(), saving
approximately a memory barrier and an atomic op, depending on architecture.
This is done by adding rcu_assign_pointer() and synchronize_rcu() calls to the
write side of the task->cpuset pointer, in cpuset.c:attach_task(), to delay
freeing up a detached cpuset until after any critical sections referencing
that pointer.
Thanks to Andi Kleen, Nick Piggin and Eric Dumazet for ideas.
Signed-off-by: Paul Jackson <pj@sgi.com>
Signed-off-by: Andrew Morton <akpm@osdl.org>
Signed-off-by: Linus Torvalds <torvalds@osdl.org>
2006-01-08 02:02:02 -07:00
|
|
|
synchronize_rcu();
|
2005-04-16 16:20:36 -06:00
|
|
|
if (atomic_dec_and_test(&oldcs->count))
|
[PATCH] cpuset release ABBA deadlock fix
Fix possible cpuset_sem ABBA deadlock if 'notify_on_release' set.
For a particular usage pattern, creating and destroying cpusets fairly
frequently using notify_on_release, on a very large system, this deadlock
can be seen every few days. If you are not using the cpuset
notify_on_release feature, you will never see this deadlock.
The existing code, on task exit (or cpuset deletion) did:
get cpuset_sem
if cpuset marked notify_on_release and is ready to release:
compute cpuset path relative to /dev/cpuset mount point
call_usermodehelper() forks /sbin/cpuset_release_agent with path
drop cpuset_sem
Unfortunately, the fork in call_usermodehelper can allocate memory, and
allocating memory can require cpuset_sem, if the mems_generation values
changed in the interim. This results in an ABBA deadlock, trying to obtain
cpuset_sem when it is already held by the current task.
To fix this, I put the cpuset path (which must be computed while holding
cpuset_sem) in a temporary buffer, to be used in the call_usermodehelper
call of /sbin/cpuset_release_agent only _after_ dropping cpuset_sem.
So the new logic is:
get cpuset_sem
if cpuset marked notify_on_release and is ready to release:
compute cpuset path relative to /dev/cpuset mount point
stash path in kmalloc'd buffer
drop cpuset_sem
call_usermodehelper() forks /sbin/cpuset_release_agent with path
free path
The sharp eyed reader might notice that this patch does not contain any
calls to kmalloc. The existing code in the check_for_release() routine was
already kmalloc'ing a buffer to hold the cpuset path. In the old code, it
just held the buffer for a few lines, over the cpuset_release_agent() call
that in turn invoked call_usermodehelper(). In the new code, with the
application of this patch, it returns that buffer via the new char
**ppathbuf parameter, for later use and freeing in cpuset_release_agent(),
which is called after cpuset_sem is dropped. Whereas the old code has just
one call to cpuset_release_agent(), right in the check_for_release()
routine, the new code has three calls to cpuset_release_agent(), from the
various places that a cpuset can be released.
This patch has been build and booted on SN2, and passed a stress test that
previously hit the deadlock within a few seconds.
Signed-off-by: Paul Jackson <pj@sgi.com>
Signed-off-by: Andrew Morton <akpm@osdl.org>
Signed-off-by: Linus Torvalds <torvalds@osdl.org>
2005-08-09 11:07:59 -06:00
|
|
|
check_for_release(oldcs, ppathbuf);
|
2005-04-16 16:20:36 -06:00
|
|
|
return 0;
|
|
|
|
}
|
|
|
|
|
|
|
|
/* The various types of files and directories in a cpuset file system */
|
|
|
|
|
|
|
|
typedef enum {
|
|
|
|
FILE_ROOT,
|
|
|
|
FILE_DIR,
|
[PATCH] cpusets: swap migration interface
Add a boolean "memory_migrate" to each cpuset, represented by a file
containing "0" or "1" in each directory below /dev/cpuset.
It defaults to false (file contains "0"). It can be set true by writing
"1" to the file.
If true, then anytime that a task is attached to the cpuset so marked, the
pages of that task will be moved to that cpuset, preserving, to the extent
practical, the cpuset-relative placement of the pages.
Also anytime that a cpuset so marked has its memory placement changed (by
writing to its "mems" file), the tasks in that cpuset will have their pages
moved to the cpusets new nodes, preserving, to the extent practical, the
cpuset-relative placement of the moved pages.
Signed-off-by: Paul Jackson <pj@sgi.com>
Cc: Christoph Lameter <christoph@lameter.com>
Signed-off-by: Andrew Morton <akpm@osdl.org>
Signed-off-by: Linus Torvalds <torvalds@osdl.org>
2006-01-08 02:00:56 -07:00
|
|
|
FILE_MEMORY_MIGRATE,
|
2005-04-16 16:20:36 -06:00
|
|
|
FILE_CPULIST,
|
|
|
|
FILE_MEMLIST,
|
|
|
|
FILE_CPU_EXCLUSIVE,
|
|
|
|
FILE_MEM_EXCLUSIVE,
|
|
|
|
FILE_NOTIFY_ON_RELEASE,
|
[PATCH] cpuset: memory pressure meter
Provide a simple per-cpuset metric of memory pressure, tracking the -rate-
that the tasks in a cpuset call try_to_free_pages(), the synchronous
(direct) memory reclaim code.
This enables batch managers monitoring jobs running in dedicated cpusets to
efficiently detect what level of memory pressure that job is causing.
This is useful both on tightly managed systems running a wide mix of
submitted jobs, which may choose to terminate or reprioritize jobs that are
trying to use more memory than allowed on the nodes assigned them, and with
tightly coupled, long running, massively parallel scientific computing jobs
that will dramatically fail to meet required performance goals if they
start to use more memory than allowed to them.
This patch just provides a very economical way for the batch manager to
monitor a cpuset for signs of memory pressure. It's up to the batch
manager or other user code to decide what to do about it and take action.
==> Unless this feature is enabled by writing "1" to the special file
/dev/cpuset/memory_pressure_enabled, the hook in the rebalance
code of __alloc_pages() for this metric reduces to simply noticing
that the cpuset_memory_pressure_enabled flag is zero. So only
systems that enable this feature will compute the metric.
Why a per-cpuset, running average:
Because this meter is per-cpuset, rather than per-task or mm, the
system load imposed by a batch scheduler monitoring this metric is
sharply reduced on large systems, because a scan of the tasklist can be
avoided on each set of queries.
Because this meter is a running average, instead of an accumulating
counter, a batch scheduler can detect memory pressure with a single
read, instead of having to read and accumulate results for a period of
time.
Because this meter is per-cpuset rather than per-task or mm, the
batch scheduler can obtain the key information, memory pressure in a
cpuset, with a single read, rather than having to query and accumulate
results over all the (dynamically changing) set of tasks in the cpuset.
A per-cpuset simple digital filter (requires a spinlock and 3 words of data
per-cpuset) is kept, and updated by any task attached to that cpuset, if it
enters the synchronous (direct) page reclaim code.
A per-cpuset file provides an integer number representing the recent
(half-life of 10 seconds) rate of direct page reclaims caused by the tasks
in the cpuset, in units of reclaims attempted per second, times 1000.
Signed-off-by: Paul Jackson <pj@sgi.com>
Signed-off-by: Andrew Morton <akpm@osdl.org>
Signed-off-by: Linus Torvalds <torvalds@osdl.org>
2006-01-08 02:01:49 -07:00
|
|
|
FILE_MEMORY_PRESSURE_ENABLED,
|
|
|
|
FILE_MEMORY_PRESSURE,
|
[PATCH] cpuset memory spread basic implementation
This patch provides the implementation and cpuset interface for an alternative
memory allocation policy that can be applied to certain kinds of memory
allocations, such as the page cache (file system buffers) and some slab caches
(such as inode caches).
The policy is called "memory spreading." If enabled, it spreads out these
kinds of memory allocations over all the nodes allowed to a task, instead of
preferring to place them on the node where the task is executing.
All other kinds of allocations, including anonymous pages for a tasks stack
and data regions, are not affected by this policy choice, and continue to be
allocated preferring the node local to execution, as modified by the NUMA
mempolicy.
There are two boolean flag files per cpuset that control where the kernel
allocates pages for the file system buffers and related in kernel data
structures. They are called 'memory_spread_page' and 'memory_spread_slab'.
If the per-cpuset boolean flag file 'memory_spread_page' is set, then the
kernel will spread the file system buffers (page cache) evenly over all the
nodes that the faulting task is allowed to use, instead of preferring to put
those pages on the node where the task is running.
If the per-cpuset boolean flag file 'memory_spread_slab' is set, then the
kernel will spread some file system related slab caches, such as for inodes
and dentries evenly over all the nodes that the faulting task is allowed to
use, instead of preferring to put those pages on the node where the task is
running.
The implementation is simple. Setting the cpuset flags 'memory_spread_page'
or 'memory_spread_cache' turns on the per-process flags PF_SPREAD_PAGE or
PF_SPREAD_SLAB, respectively, for each task that is in the cpuset or
subsequently joins that cpuset. In subsequent patches, the page allocation
calls for the affected page cache and slab caches are modified to perform an
inline check for these flags, and if set, a call to a new routine
cpuset_mem_spread_node() returns the node to prefer for the allocation.
The cpuset_mem_spread_node() routine is also simple. It uses the value of a
per-task rotor cpuset_mem_spread_rotor to select the next node in the current
tasks mems_allowed to prefer for the allocation.
This policy can provide substantial improvements for jobs that need to place
thread local data on the corresponding node, but that need to access large
file system data sets that need to be spread across the several nodes in the
jobs cpuset in order to fit. Without this patch, especially for jobs that
might have one thread reading in the data set, the memory allocation across
the nodes in the jobs cpuset can become very uneven.
A couple of Copyright year ranges are updated as well. And a couple of email
addresses that can be found in the MAINTAINERS file are removed.
Signed-off-by: Paul Jackson <pj@sgi.com>
Signed-off-by: Andrew Morton <akpm@osdl.org>
Signed-off-by: Linus Torvalds <torvalds@osdl.org>
2006-03-24 04:16:03 -07:00
|
|
|
FILE_SPREAD_PAGE,
|
|
|
|
FILE_SPREAD_SLAB,
|
2005-04-16 16:20:36 -06:00
|
|
|
FILE_TASKLIST,
|
|
|
|
} cpuset_filetype_t;
|
|
|
|
|
|
|
|
static ssize_t cpuset_common_file_write(struct file *file, const char __user *userbuf,
|
|
|
|
size_t nbytes, loff_t *unused_ppos)
|
|
|
|
{
|
|
|
|
struct cpuset *cs = __d_cs(file->f_dentry->d_parent);
|
|
|
|
struct cftype *cft = __d_cft(file->f_dentry);
|
|
|
|
cpuset_filetype_t type = cft->private;
|
|
|
|
char *buffer;
|
[PATCH] cpuset release ABBA deadlock fix
Fix possible cpuset_sem ABBA deadlock if 'notify_on_release' set.
For a particular usage pattern, creating and destroying cpusets fairly
frequently using notify_on_release, on a very large system, this deadlock
can be seen every few days. If you are not using the cpuset
notify_on_release feature, you will never see this deadlock.
The existing code, on task exit (or cpuset deletion) did:
get cpuset_sem
if cpuset marked notify_on_release and is ready to release:
compute cpuset path relative to /dev/cpuset mount point
call_usermodehelper() forks /sbin/cpuset_release_agent with path
drop cpuset_sem
Unfortunately, the fork in call_usermodehelper can allocate memory, and
allocating memory can require cpuset_sem, if the mems_generation values
changed in the interim. This results in an ABBA deadlock, trying to obtain
cpuset_sem when it is already held by the current task.
To fix this, I put the cpuset path (which must be computed while holding
cpuset_sem) in a temporary buffer, to be used in the call_usermodehelper
call of /sbin/cpuset_release_agent only _after_ dropping cpuset_sem.
So the new logic is:
get cpuset_sem
if cpuset marked notify_on_release and is ready to release:
compute cpuset path relative to /dev/cpuset mount point
stash path in kmalloc'd buffer
drop cpuset_sem
call_usermodehelper() forks /sbin/cpuset_release_agent with path
free path
The sharp eyed reader might notice that this patch does not contain any
calls to kmalloc. The existing code in the check_for_release() routine was
already kmalloc'ing a buffer to hold the cpuset path. In the old code, it
just held the buffer for a few lines, over the cpuset_release_agent() call
that in turn invoked call_usermodehelper(). In the new code, with the
application of this patch, it returns that buffer via the new char
**ppathbuf parameter, for later use and freeing in cpuset_release_agent(),
which is called after cpuset_sem is dropped. Whereas the old code has just
one call to cpuset_release_agent(), right in the check_for_release()
routine, the new code has three calls to cpuset_release_agent(), from the
various places that a cpuset can be released.
This patch has been build and booted on SN2, and passed a stress test that
previously hit the deadlock within a few seconds.
Signed-off-by: Paul Jackson <pj@sgi.com>
Signed-off-by: Andrew Morton <akpm@osdl.org>
Signed-off-by: Linus Torvalds <torvalds@osdl.org>
2005-08-09 11:07:59 -06:00
|
|
|
char *pathbuf = NULL;
|
2005-04-16 16:20:36 -06:00
|
|
|
int retval = 0;
|
|
|
|
|
|
|
|
/* Crude upper limit on largest legitimate cpulist user might write. */
|
|
|
|
if (nbytes > 100 + 6 * NR_CPUS)
|
|
|
|
return -E2BIG;
|
|
|
|
|
|
|
|
/* +1 for nul-terminator */
|
|
|
|
if ((buffer = kmalloc(nbytes + 1, GFP_KERNEL)) == 0)
|
|
|
|
return -ENOMEM;
|
|
|
|
|
|
|
|
if (copy_from_user(buffer, userbuf, nbytes)) {
|
|
|
|
retval = -EFAULT;
|
|
|
|
goto out1;
|
|
|
|
}
|
|
|
|
buffer[nbytes] = 0; /* nul-terminate */
|
|
|
|
|
2006-03-23 04:00:18 -07:00
|
|
|
mutex_lock(&manage_mutex);
|
2005-04-16 16:20:36 -06:00
|
|
|
|
|
|
|
if (is_removed(cs)) {
|
|
|
|
retval = -ENODEV;
|
|
|
|
goto out2;
|
|
|
|
}
|
|
|
|
|
|
|
|
switch (type) {
|
|
|
|
case FILE_CPULIST:
|
|
|
|
retval = update_cpumask(cs, buffer);
|
|
|
|
break;
|
|
|
|
case FILE_MEMLIST:
|
|
|
|
retval = update_nodemask(cs, buffer);
|
|
|
|
break;
|
|
|
|
case FILE_CPU_EXCLUSIVE:
|
|
|
|
retval = update_flag(CS_CPU_EXCLUSIVE, cs, buffer);
|
|
|
|
break;
|
|
|
|
case FILE_MEM_EXCLUSIVE:
|
|
|
|
retval = update_flag(CS_MEM_EXCLUSIVE, cs, buffer);
|
|
|
|
break;
|
|
|
|
case FILE_NOTIFY_ON_RELEASE:
|
|
|
|
retval = update_flag(CS_NOTIFY_ON_RELEASE, cs, buffer);
|
|
|
|
break;
|
[PATCH] cpusets: swap migration interface
Add a boolean "memory_migrate" to each cpuset, represented by a file
containing "0" or "1" in each directory below /dev/cpuset.
It defaults to false (file contains "0"). It can be set true by writing
"1" to the file.
If true, then anytime that a task is attached to the cpuset so marked, the
pages of that task will be moved to that cpuset, preserving, to the extent
practical, the cpuset-relative placement of the pages.
Also anytime that a cpuset so marked has its memory placement changed (by
writing to its "mems" file), the tasks in that cpuset will have their pages
moved to the cpusets new nodes, preserving, to the extent practical, the
cpuset-relative placement of the moved pages.
Signed-off-by: Paul Jackson <pj@sgi.com>
Cc: Christoph Lameter <christoph@lameter.com>
Signed-off-by: Andrew Morton <akpm@osdl.org>
Signed-off-by: Linus Torvalds <torvalds@osdl.org>
2006-01-08 02:00:56 -07:00
|
|
|
case FILE_MEMORY_MIGRATE:
|
|
|
|
retval = update_flag(CS_MEMORY_MIGRATE, cs, buffer);
|
|
|
|
break;
|
[PATCH] cpuset: memory pressure meter
Provide a simple per-cpuset metric of memory pressure, tracking the -rate-
that the tasks in a cpuset call try_to_free_pages(), the synchronous
(direct) memory reclaim code.
This enables batch managers monitoring jobs running in dedicated cpusets to
efficiently detect what level of memory pressure that job is causing.
This is useful both on tightly managed systems running a wide mix of
submitted jobs, which may choose to terminate or reprioritize jobs that are
trying to use more memory than allowed on the nodes assigned them, and with
tightly coupled, long running, massively parallel scientific computing jobs
that will dramatically fail to meet required performance goals if they
start to use more memory than allowed to them.
This patch just provides a very economical way for the batch manager to
monitor a cpuset for signs of memory pressure. It's up to the batch
manager or other user code to decide what to do about it and take action.
==> Unless this feature is enabled by writing "1" to the special file
/dev/cpuset/memory_pressure_enabled, the hook in the rebalance
code of __alloc_pages() for this metric reduces to simply noticing
that the cpuset_memory_pressure_enabled flag is zero. So only
systems that enable this feature will compute the metric.
Why a per-cpuset, running average:
Because this meter is per-cpuset, rather than per-task or mm, the
system load imposed by a batch scheduler monitoring this metric is
sharply reduced on large systems, because a scan of the tasklist can be
avoided on each set of queries.
Because this meter is a running average, instead of an accumulating
counter, a batch scheduler can detect memory pressure with a single
read, instead of having to read and accumulate results for a period of
time.
Because this meter is per-cpuset rather than per-task or mm, the
batch scheduler can obtain the key information, memory pressure in a
cpuset, with a single read, rather than having to query and accumulate
results over all the (dynamically changing) set of tasks in the cpuset.
A per-cpuset simple digital filter (requires a spinlock and 3 words of data
per-cpuset) is kept, and updated by any task attached to that cpuset, if it
enters the synchronous (direct) page reclaim code.
A per-cpuset file provides an integer number representing the recent
(half-life of 10 seconds) rate of direct page reclaims caused by the tasks
in the cpuset, in units of reclaims attempted per second, times 1000.
Signed-off-by: Paul Jackson <pj@sgi.com>
Signed-off-by: Andrew Morton <akpm@osdl.org>
Signed-off-by: Linus Torvalds <torvalds@osdl.org>
2006-01-08 02:01:49 -07:00
|
|
|
case FILE_MEMORY_PRESSURE_ENABLED:
|
|
|
|
retval = update_memory_pressure_enabled(cs, buffer);
|
|
|
|
break;
|
|
|
|
case FILE_MEMORY_PRESSURE:
|
|
|
|
retval = -EACCES;
|
|
|
|
break;
|
[PATCH] cpuset memory spread basic implementation
This patch provides the implementation and cpuset interface for an alternative
memory allocation policy that can be applied to certain kinds of memory
allocations, such as the page cache (file system buffers) and some slab caches
(such as inode caches).
The policy is called "memory spreading." If enabled, it spreads out these
kinds of memory allocations over all the nodes allowed to a task, instead of
preferring to place them on the node where the task is executing.
All other kinds of allocations, including anonymous pages for a tasks stack
and data regions, are not affected by this policy choice, and continue to be
allocated preferring the node local to execution, as modified by the NUMA
mempolicy.
There are two boolean flag files per cpuset that control where the kernel
allocates pages for the file system buffers and related in kernel data
structures. They are called 'memory_spread_page' and 'memory_spread_slab'.
If the per-cpuset boolean flag file 'memory_spread_page' is set, then the
kernel will spread the file system buffers (page cache) evenly over all the
nodes that the faulting task is allowed to use, instead of preferring to put
those pages on the node where the task is running.
If the per-cpuset boolean flag file 'memory_spread_slab' is set, then the
kernel will spread some file system related slab caches, such as for inodes
and dentries evenly over all the nodes that the faulting task is allowed to
use, instead of preferring to put those pages on the node where the task is
running.
The implementation is simple. Setting the cpuset flags 'memory_spread_page'
or 'memory_spread_cache' turns on the per-process flags PF_SPREAD_PAGE or
PF_SPREAD_SLAB, respectively, for each task that is in the cpuset or
subsequently joins that cpuset. In subsequent patches, the page allocation
calls for the affected page cache and slab caches are modified to perform an
inline check for these flags, and if set, a call to a new routine
cpuset_mem_spread_node() returns the node to prefer for the allocation.
The cpuset_mem_spread_node() routine is also simple. It uses the value of a
per-task rotor cpuset_mem_spread_rotor to select the next node in the current
tasks mems_allowed to prefer for the allocation.
This policy can provide substantial improvements for jobs that need to place
thread local data on the corresponding node, but that need to access large
file system data sets that need to be spread across the several nodes in the
jobs cpuset in order to fit. Without this patch, especially for jobs that
might have one thread reading in the data set, the memory allocation across
the nodes in the jobs cpuset can become very uneven.
A couple of Copyright year ranges are updated as well. And a couple of email
addresses that can be found in the MAINTAINERS file are removed.
Signed-off-by: Paul Jackson <pj@sgi.com>
Signed-off-by: Andrew Morton <akpm@osdl.org>
Signed-off-by: Linus Torvalds <torvalds@osdl.org>
2006-03-24 04:16:03 -07:00
|
|
|
case FILE_SPREAD_PAGE:
|
|
|
|
retval = update_flag(CS_SPREAD_PAGE, cs, buffer);
|
2006-03-24 04:16:11 -07:00
|
|
|
cs->mems_generation = cpuset_mems_generation++;
|
[PATCH] cpuset memory spread basic implementation
This patch provides the implementation and cpuset interface for an alternative
memory allocation policy that can be applied to certain kinds of memory
allocations, such as the page cache (file system buffers) and some slab caches
(such as inode caches).
The policy is called "memory spreading." If enabled, it spreads out these
kinds of memory allocations over all the nodes allowed to a task, instead of
preferring to place them on the node where the task is executing.
All other kinds of allocations, including anonymous pages for a tasks stack
and data regions, are not affected by this policy choice, and continue to be
allocated preferring the node local to execution, as modified by the NUMA
mempolicy.
There are two boolean flag files per cpuset that control where the kernel
allocates pages for the file system buffers and related in kernel data
structures. They are called 'memory_spread_page' and 'memory_spread_slab'.
If the per-cpuset boolean flag file 'memory_spread_page' is set, then the
kernel will spread the file system buffers (page cache) evenly over all the
nodes that the faulting task is allowed to use, instead of preferring to put
those pages on the node where the task is running.
If the per-cpuset boolean flag file 'memory_spread_slab' is set, then the
kernel will spread some file system related slab caches, such as for inodes
and dentries evenly over all the nodes that the faulting task is allowed to
use, instead of preferring to put those pages on the node where the task is
running.
The implementation is simple. Setting the cpuset flags 'memory_spread_page'
or 'memory_spread_cache' turns on the per-process flags PF_SPREAD_PAGE or
PF_SPREAD_SLAB, respectively, for each task that is in the cpuset or
subsequently joins that cpuset. In subsequent patches, the page allocation
calls for the affected page cache and slab caches are modified to perform an
inline check for these flags, and if set, a call to a new routine
cpuset_mem_spread_node() returns the node to prefer for the allocation.
The cpuset_mem_spread_node() routine is also simple. It uses the value of a
per-task rotor cpuset_mem_spread_rotor to select the next node in the current
tasks mems_allowed to prefer for the allocation.
This policy can provide substantial improvements for jobs that need to place
thread local data on the corresponding node, but that need to access large
file system data sets that need to be spread across the several nodes in the
jobs cpuset in order to fit. Without this patch, especially for jobs that
might have one thread reading in the data set, the memory allocation across
the nodes in the jobs cpuset can become very uneven.
A couple of Copyright year ranges are updated as well. And a couple of email
addresses that can be found in the MAINTAINERS file are removed.
Signed-off-by: Paul Jackson <pj@sgi.com>
Signed-off-by: Andrew Morton <akpm@osdl.org>
Signed-off-by: Linus Torvalds <torvalds@osdl.org>
2006-03-24 04:16:03 -07:00
|
|
|
break;
|
|
|
|
case FILE_SPREAD_SLAB:
|
|
|
|
retval = update_flag(CS_SPREAD_SLAB, cs, buffer);
|
2006-03-24 04:16:11 -07:00
|
|
|
cs->mems_generation = cpuset_mems_generation++;
|
[PATCH] cpuset memory spread basic implementation
This patch provides the implementation and cpuset interface for an alternative
memory allocation policy that can be applied to certain kinds of memory
allocations, such as the page cache (file system buffers) and some slab caches
(such as inode caches).
The policy is called "memory spreading." If enabled, it spreads out these
kinds of memory allocations over all the nodes allowed to a task, instead of
preferring to place them on the node where the task is executing.
All other kinds of allocations, including anonymous pages for a tasks stack
and data regions, are not affected by this policy choice, and continue to be
allocated preferring the node local to execution, as modified by the NUMA
mempolicy.
There are two boolean flag files per cpuset that control where the kernel
allocates pages for the file system buffers and related in kernel data
structures. They are called 'memory_spread_page' and 'memory_spread_slab'.
If the per-cpuset boolean flag file 'memory_spread_page' is set, then the
kernel will spread the file system buffers (page cache) evenly over all the
nodes that the faulting task is allowed to use, instead of preferring to put
those pages on the node where the task is running.
If the per-cpuset boolean flag file 'memory_spread_slab' is set, then the
kernel will spread some file system related slab caches, such as for inodes
and dentries evenly over all the nodes that the faulting task is allowed to
use, instead of preferring to put those pages on the node where the task is
running.
The implementation is simple. Setting the cpuset flags 'memory_spread_page'
or 'memory_spread_cache' turns on the per-process flags PF_SPREAD_PAGE or
PF_SPREAD_SLAB, respectively, for each task that is in the cpuset or
subsequently joins that cpuset. In subsequent patches, the page allocation
calls for the affected page cache and slab caches are modified to perform an
inline check for these flags, and if set, a call to a new routine
cpuset_mem_spread_node() returns the node to prefer for the allocation.
The cpuset_mem_spread_node() routine is also simple. It uses the value of a
per-task rotor cpuset_mem_spread_rotor to select the next node in the current
tasks mems_allowed to prefer for the allocation.
This policy can provide substantial improvements for jobs that need to place
thread local data on the corresponding node, but that need to access large
file system data sets that need to be spread across the several nodes in the
jobs cpuset in order to fit. Without this patch, especially for jobs that
might have one thread reading in the data set, the memory allocation across
the nodes in the jobs cpuset can become very uneven.
A couple of Copyright year ranges are updated as well. And a couple of email
addresses that can be found in the MAINTAINERS file are removed.
Signed-off-by: Paul Jackson <pj@sgi.com>
Signed-off-by: Andrew Morton <akpm@osdl.org>
Signed-off-by: Linus Torvalds <torvalds@osdl.org>
2006-03-24 04:16:03 -07:00
|
|
|
break;
|
2005-04-16 16:20:36 -06:00
|
|
|
case FILE_TASKLIST:
|
[PATCH] cpuset release ABBA deadlock fix
Fix possible cpuset_sem ABBA deadlock if 'notify_on_release' set.
For a particular usage pattern, creating and destroying cpusets fairly
frequently using notify_on_release, on a very large system, this deadlock
can be seen every few days. If you are not using the cpuset
notify_on_release feature, you will never see this deadlock.
The existing code, on task exit (or cpuset deletion) did:
get cpuset_sem
if cpuset marked notify_on_release and is ready to release:
compute cpuset path relative to /dev/cpuset mount point
call_usermodehelper() forks /sbin/cpuset_release_agent with path
drop cpuset_sem
Unfortunately, the fork in call_usermodehelper can allocate memory, and
allocating memory can require cpuset_sem, if the mems_generation values
changed in the interim. This results in an ABBA deadlock, trying to obtain
cpuset_sem when it is already held by the current task.
To fix this, I put the cpuset path (which must be computed while holding
cpuset_sem) in a temporary buffer, to be used in the call_usermodehelper
call of /sbin/cpuset_release_agent only _after_ dropping cpuset_sem.
So the new logic is:
get cpuset_sem
if cpuset marked notify_on_release and is ready to release:
compute cpuset path relative to /dev/cpuset mount point
stash path in kmalloc'd buffer
drop cpuset_sem
call_usermodehelper() forks /sbin/cpuset_release_agent with path
free path
The sharp eyed reader might notice that this patch does not contain any
calls to kmalloc. The existing code in the check_for_release() routine was
already kmalloc'ing a buffer to hold the cpuset path. In the old code, it
just held the buffer for a few lines, over the cpuset_release_agent() call
that in turn invoked call_usermodehelper(). In the new code, with the
application of this patch, it returns that buffer via the new char
**ppathbuf parameter, for later use and freeing in cpuset_release_agent(),
which is called after cpuset_sem is dropped. Whereas the old code has just
one call to cpuset_release_agent(), right in the check_for_release()
routine, the new code has three calls to cpuset_release_agent(), from the
various places that a cpuset can be released.
This patch has been build and booted on SN2, and passed a stress test that
previously hit the deadlock within a few seconds.
Signed-off-by: Paul Jackson <pj@sgi.com>
Signed-off-by: Andrew Morton <akpm@osdl.org>
Signed-off-by: Linus Torvalds <torvalds@osdl.org>
2005-08-09 11:07:59 -06:00
|
|
|
retval = attach_task(cs, buffer, &pathbuf);
|
2005-04-16 16:20:36 -06:00
|
|
|
break;
|
|
|
|
default:
|
|
|
|
retval = -EINVAL;
|
|
|
|
goto out2;
|
|
|
|
}
|
|
|
|
|
|
|
|
if (retval == 0)
|
|
|
|
retval = nbytes;
|
|
|
|
out2:
|
2006-03-23 04:00:18 -07:00
|
|
|
mutex_unlock(&manage_mutex);
|
[PATCH] cpuset release ABBA deadlock fix
Fix possible cpuset_sem ABBA deadlock if 'notify_on_release' set.
For a particular usage pattern, creating and destroying cpusets fairly
frequently using notify_on_release, on a very large system, this deadlock
can be seen every few days. If you are not using the cpuset
notify_on_release feature, you will never see this deadlock.
The existing code, on task exit (or cpuset deletion) did:
get cpuset_sem
if cpuset marked notify_on_release and is ready to release:
compute cpuset path relative to /dev/cpuset mount point
call_usermodehelper() forks /sbin/cpuset_release_agent with path
drop cpuset_sem
Unfortunately, the fork in call_usermodehelper can allocate memory, and
allocating memory can require cpuset_sem, if the mems_generation values
changed in the interim. This results in an ABBA deadlock, trying to obtain
cpuset_sem when it is already held by the current task.
To fix this, I put the cpuset path (which must be computed while holding
cpuset_sem) in a temporary buffer, to be used in the call_usermodehelper
call of /sbin/cpuset_release_agent only _after_ dropping cpuset_sem.
So the new logic is:
get cpuset_sem
if cpuset marked notify_on_release and is ready to release:
compute cpuset path relative to /dev/cpuset mount point
stash path in kmalloc'd buffer
drop cpuset_sem
call_usermodehelper() forks /sbin/cpuset_release_agent with path
free path
The sharp eyed reader might notice that this patch does not contain any
calls to kmalloc. The existing code in the check_for_release() routine was
already kmalloc'ing a buffer to hold the cpuset path. In the old code, it
just held the buffer for a few lines, over the cpuset_release_agent() call
that in turn invoked call_usermodehelper(). In the new code, with the
application of this patch, it returns that buffer via the new char
**ppathbuf parameter, for later use and freeing in cpuset_release_agent(),
which is called after cpuset_sem is dropped. Whereas the old code has just
one call to cpuset_release_agent(), right in the check_for_release()
routine, the new code has three calls to cpuset_release_agent(), from the
various places that a cpuset can be released.
This patch has been build and booted on SN2, and passed a stress test that
previously hit the deadlock within a few seconds.
Signed-off-by: Paul Jackson <pj@sgi.com>
Signed-off-by: Andrew Morton <akpm@osdl.org>
Signed-off-by: Linus Torvalds <torvalds@osdl.org>
2005-08-09 11:07:59 -06:00
|
|
|
cpuset_release_agent(pathbuf);
|
2005-04-16 16:20:36 -06:00
|
|
|
out1:
|
|
|
|
kfree(buffer);
|
|
|
|
return retval;
|
|
|
|
}
|
|
|
|
|
|
|
|
static ssize_t cpuset_file_write(struct file *file, const char __user *buf,
|
|
|
|
size_t nbytes, loff_t *ppos)
|
|
|
|
{
|
|
|
|
ssize_t retval = 0;
|
|
|
|
struct cftype *cft = __d_cft(file->f_dentry);
|
|
|
|
if (!cft)
|
|
|
|
return -ENODEV;
|
|
|
|
|
|
|
|
/* special function ? */
|
|
|
|
if (cft->write)
|
|
|
|
retval = cft->write(file, buf, nbytes, ppos);
|
|
|
|
else
|
|
|
|
retval = cpuset_common_file_write(file, buf, nbytes, ppos);
|
|
|
|
|
|
|
|
return retval;
|
|
|
|
}
|
|
|
|
|
|
|
|
/*
|
|
|
|
* These ascii lists should be read in a single call, by using a user
|
|
|
|
* buffer large enough to hold the entire map. If read in smaller
|
|
|
|
* chunks, there is no guarantee of atomicity. Since the display format
|
|
|
|
* used, list of ranges of sequential numbers, is variable length,
|
|
|
|
* and since these maps can change value dynamically, one could read
|
|
|
|
* gibberish by doing partial reads while a list was changing.
|
|
|
|
* A single large read to a buffer that crosses a page boundary is
|
|
|
|
* ok, because the result being copied to user land is not recomputed
|
|
|
|
* across a page fault.
|
|
|
|
*/
|
|
|
|
|
|
|
|
static int cpuset_sprintf_cpulist(char *page, struct cpuset *cs)
|
|
|
|
{
|
|
|
|
cpumask_t mask;
|
|
|
|
|
2006-03-23 04:00:18 -07:00
|
|
|
mutex_lock(&callback_mutex);
|
2005-04-16 16:20:36 -06:00
|
|
|
mask = cs->cpus_allowed;
|
2006-03-23 04:00:18 -07:00
|
|
|
mutex_unlock(&callback_mutex);
|
2005-04-16 16:20:36 -06:00
|
|
|
|
|
|
|
return cpulist_scnprintf(page, PAGE_SIZE, mask);
|
|
|
|
}
|
|
|
|
|
|
|
|
static int cpuset_sprintf_memlist(char *page, struct cpuset *cs)
|
|
|
|
{
|
|
|
|
nodemask_t mask;
|
|
|
|
|
2006-03-23 04:00:18 -07:00
|
|
|
mutex_lock(&callback_mutex);
|
2005-04-16 16:20:36 -06:00
|
|
|
mask = cs->mems_allowed;
|
2006-03-23 04:00:18 -07:00
|
|
|
mutex_unlock(&callback_mutex);
|
2005-04-16 16:20:36 -06:00
|
|
|
|
|
|
|
return nodelist_scnprintf(page, PAGE_SIZE, mask);
|
|
|
|
}
|
|
|
|
|
|
|
|
static ssize_t cpuset_common_file_read(struct file *file, char __user *buf,
|
|
|
|
size_t nbytes, loff_t *ppos)
|
|
|
|
{
|
|
|
|
struct cftype *cft = __d_cft(file->f_dentry);
|
|
|
|
struct cpuset *cs = __d_cs(file->f_dentry->d_parent);
|
|
|
|
cpuset_filetype_t type = cft->private;
|
|
|
|
char *page;
|
|
|
|
ssize_t retval = 0;
|
|
|
|
char *s;
|
|
|
|
|
|
|
|
if (!(page = (char *)__get_free_page(GFP_KERNEL)))
|
|
|
|
return -ENOMEM;
|
|
|
|
|
|
|
|
s = page;
|
|
|
|
|
|
|
|
switch (type) {
|
|
|
|
case FILE_CPULIST:
|
|
|
|
s += cpuset_sprintf_cpulist(s, cs);
|
|
|
|
break;
|
|
|
|
case FILE_MEMLIST:
|
|
|
|
s += cpuset_sprintf_memlist(s, cs);
|
|
|
|
break;
|
|
|
|
case FILE_CPU_EXCLUSIVE:
|
|
|
|
*s++ = is_cpu_exclusive(cs) ? '1' : '0';
|
|
|
|
break;
|
|
|
|
case FILE_MEM_EXCLUSIVE:
|
|
|
|
*s++ = is_mem_exclusive(cs) ? '1' : '0';
|
|
|
|
break;
|
|
|
|
case FILE_NOTIFY_ON_RELEASE:
|
|
|
|
*s++ = notify_on_release(cs) ? '1' : '0';
|
|
|
|
break;
|
[PATCH] cpusets: swap migration interface
Add a boolean "memory_migrate" to each cpuset, represented by a file
containing "0" or "1" in each directory below /dev/cpuset.
It defaults to false (file contains "0"). It can be set true by writing
"1" to the file.
If true, then anytime that a task is attached to the cpuset so marked, the
pages of that task will be moved to that cpuset, preserving, to the extent
practical, the cpuset-relative placement of the pages.
Also anytime that a cpuset so marked has its memory placement changed (by
writing to its "mems" file), the tasks in that cpuset will have their pages
moved to the cpusets new nodes, preserving, to the extent practical, the
cpuset-relative placement of the moved pages.
Signed-off-by: Paul Jackson <pj@sgi.com>
Cc: Christoph Lameter <christoph@lameter.com>
Signed-off-by: Andrew Morton <akpm@osdl.org>
Signed-off-by: Linus Torvalds <torvalds@osdl.org>
2006-01-08 02:00:56 -07:00
|
|
|
case FILE_MEMORY_MIGRATE:
|
|
|
|
*s++ = is_memory_migrate(cs) ? '1' : '0';
|
|
|
|
break;
|
[PATCH] cpuset: memory pressure meter
Provide a simple per-cpuset metric of memory pressure, tracking the -rate-
that the tasks in a cpuset call try_to_free_pages(), the synchronous
(direct) memory reclaim code.
This enables batch managers monitoring jobs running in dedicated cpusets to
efficiently detect what level of memory pressure that job is causing.
This is useful both on tightly managed systems running a wide mix of
submitted jobs, which may choose to terminate or reprioritize jobs that are
trying to use more memory than allowed on the nodes assigned them, and with
tightly coupled, long running, massively parallel scientific computing jobs
that will dramatically fail to meet required performance goals if they
start to use more memory than allowed to them.
This patch just provides a very economical way for the batch manager to
monitor a cpuset for signs of memory pressure. It's up to the batch
manager or other user code to decide what to do about it and take action.
==> Unless this feature is enabled by writing "1" to the special file
/dev/cpuset/memory_pressure_enabled, the hook in the rebalance
code of __alloc_pages() for this metric reduces to simply noticing
that the cpuset_memory_pressure_enabled flag is zero. So only
systems that enable this feature will compute the metric.
Why a per-cpuset, running average:
Because this meter is per-cpuset, rather than per-task or mm, the
system load imposed by a batch scheduler monitoring this metric is
sharply reduced on large systems, because a scan of the tasklist can be
avoided on each set of queries.
Because this meter is a running average, instead of an accumulating
counter, a batch scheduler can detect memory pressure with a single
read, instead of having to read and accumulate results for a period of
time.
Because this meter is per-cpuset rather than per-task or mm, the
batch scheduler can obtain the key information, memory pressure in a
cpuset, with a single read, rather than having to query and accumulate
results over all the (dynamically changing) set of tasks in the cpuset.
A per-cpuset simple digital filter (requires a spinlock and 3 words of data
per-cpuset) is kept, and updated by any task attached to that cpuset, if it
enters the synchronous (direct) page reclaim code.
A per-cpuset file provides an integer number representing the recent
(half-life of 10 seconds) rate of direct page reclaims caused by the tasks
in the cpuset, in units of reclaims attempted per second, times 1000.
Signed-off-by: Paul Jackson <pj@sgi.com>
Signed-off-by: Andrew Morton <akpm@osdl.org>
Signed-off-by: Linus Torvalds <torvalds@osdl.org>
2006-01-08 02:01:49 -07:00
|
|
|
case FILE_MEMORY_PRESSURE_ENABLED:
|
|
|
|
*s++ = cpuset_memory_pressure_enabled ? '1' : '0';
|
|
|
|
break;
|
|
|
|
case FILE_MEMORY_PRESSURE:
|
|
|
|
s += sprintf(s, "%d", fmeter_getrate(&cs->fmeter));
|
|
|
|
break;
|
[PATCH] cpuset memory spread basic implementation
This patch provides the implementation and cpuset interface for an alternative
memory allocation policy that can be applied to certain kinds of memory
allocations, such as the page cache (file system buffers) and some slab caches
(such as inode caches).
The policy is called "memory spreading." If enabled, it spreads out these
kinds of memory allocations over all the nodes allowed to a task, instead of
preferring to place them on the node where the task is executing.
All other kinds of allocations, including anonymous pages for a tasks stack
and data regions, are not affected by this policy choice, and continue to be
allocated preferring the node local to execution, as modified by the NUMA
mempolicy.
There are two boolean flag files per cpuset that control where the kernel
allocates pages for the file system buffers and related in kernel data
structures. They are called 'memory_spread_page' and 'memory_spread_slab'.
If the per-cpuset boolean flag file 'memory_spread_page' is set, then the
kernel will spread the file system buffers (page cache) evenly over all the
nodes that the faulting task is allowed to use, instead of preferring to put
those pages on the node where the task is running.
If the per-cpuset boolean flag file 'memory_spread_slab' is set, then the
kernel will spread some file system related slab caches, such as for inodes
and dentries evenly over all the nodes that the faulting task is allowed to
use, instead of preferring to put those pages on the node where the task is
running.
The implementation is simple. Setting the cpuset flags 'memory_spread_page'
or 'memory_spread_cache' turns on the per-process flags PF_SPREAD_PAGE or
PF_SPREAD_SLAB, respectively, for each task that is in the cpuset or
subsequently joins that cpuset. In subsequent patches, the page allocation
calls for the affected page cache and slab caches are modified to perform an
inline check for these flags, and if set, a call to a new routine
cpuset_mem_spread_node() returns the node to prefer for the allocation.
The cpuset_mem_spread_node() routine is also simple. It uses the value of a
per-task rotor cpuset_mem_spread_rotor to select the next node in the current
tasks mems_allowed to prefer for the allocation.
This policy can provide substantial improvements for jobs that need to place
thread local data on the corresponding node, but that need to access large
file system data sets that need to be spread across the several nodes in the
jobs cpuset in order to fit. Without this patch, especially for jobs that
might have one thread reading in the data set, the memory allocation across
the nodes in the jobs cpuset can become very uneven.
A couple of Copyright year ranges are updated as well. And a couple of email
addresses that can be found in the MAINTAINERS file are removed.
Signed-off-by: Paul Jackson <pj@sgi.com>
Signed-off-by: Andrew Morton <akpm@osdl.org>
Signed-off-by: Linus Torvalds <torvalds@osdl.org>
2006-03-24 04:16:03 -07:00
|
|
|
case FILE_SPREAD_PAGE:
|
|
|
|
*s++ = is_spread_page(cs) ? '1' : '0';
|
|
|
|
break;
|
|
|
|
case FILE_SPREAD_SLAB:
|
|
|
|
*s++ = is_spread_slab(cs) ? '1' : '0';
|
|
|
|
break;
|
2005-04-16 16:20:36 -06:00
|
|
|
default:
|
|
|
|
retval = -EINVAL;
|
|
|
|
goto out;
|
|
|
|
}
|
|
|
|
*s++ = '\n';
|
|
|
|
|
2005-09-29 20:26:43 -06:00
|
|
|
retval = simple_read_from_buffer(buf, nbytes, ppos, page, s - page);
|
2005-04-16 16:20:36 -06:00
|
|
|
out:
|
|
|
|
free_page((unsigned long)page);
|
|
|
|
return retval;
|
|
|
|
}
|
|
|
|
|
|
|
|
static ssize_t cpuset_file_read(struct file *file, char __user *buf, size_t nbytes,
|
|
|
|
loff_t *ppos)
|
|
|
|
{
|
|
|
|
ssize_t retval = 0;
|
|
|
|
struct cftype *cft = __d_cft(file->f_dentry);
|
|
|
|
if (!cft)
|
|
|
|
return -ENODEV;
|
|
|
|
|
|
|
|
/* special function ? */
|
|
|
|
if (cft->read)
|
|
|
|
retval = cft->read(file, buf, nbytes, ppos);
|
|
|
|
else
|
|
|
|
retval = cpuset_common_file_read(file, buf, nbytes, ppos);
|
|
|
|
|
|
|
|
return retval;
|
|
|
|
}
|
|
|
|
|
|
|
|
static int cpuset_file_open(struct inode *inode, struct file *file)
|
|
|
|
{
|
|
|
|
int err;
|
|
|
|
struct cftype *cft;
|
|
|
|
|
|
|
|
err = generic_file_open(inode, file);
|
|
|
|
if (err)
|
|
|
|
return err;
|
|
|
|
|
|
|
|
cft = __d_cft(file->f_dentry);
|
|
|
|
if (!cft)
|
|
|
|
return -ENODEV;
|
|
|
|
if (cft->open)
|
|
|
|
err = cft->open(inode, file);
|
|
|
|
else
|
|
|
|
err = 0;
|
|
|
|
|
|
|
|
return err;
|
|
|
|
}
|
|
|
|
|
|
|
|
static int cpuset_file_release(struct inode *inode, struct file *file)
|
|
|
|
{
|
|
|
|
struct cftype *cft = __d_cft(file->f_dentry);
|
|
|
|
if (cft->release)
|
|
|
|
return cft->release(inode, file);
|
|
|
|
return 0;
|
|
|
|
}
|
|
|
|
|
2005-10-30 16:02:31 -07:00
|
|
|
/*
|
|
|
|
* cpuset_rename - Only allow simple rename of directories in place.
|
|
|
|
*/
|
|
|
|
static int cpuset_rename(struct inode *old_dir, struct dentry *old_dentry,
|
|
|
|
struct inode *new_dir, struct dentry *new_dentry)
|
|
|
|
{
|
|
|
|
if (!S_ISDIR(old_dentry->d_inode->i_mode))
|
|
|
|
return -ENOTDIR;
|
|
|
|
if (new_dentry->d_inode)
|
|
|
|
return -EEXIST;
|
|
|
|
if (old_dir != new_dir)
|
|
|
|
return -EIO;
|
|
|
|
return simple_rename(old_dir, old_dentry, new_dir, new_dentry);
|
|
|
|
}
|
|
|
|
|
2005-04-16 16:20:36 -06:00
|
|
|
static struct file_operations cpuset_file_operations = {
|
|
|
|
.read = cpuset_file_read,
|
|
|
|
.write = cpuset_file_write,
|
|
|
|
.llseek = generic_file_llseek,
|
|
|
|
.open = cpuset_file_open,
|
|
|
|
.release = cpuset_file_release,
|
|
|
|
};
|
|
|
|
|
|
|
|
static struct inode_operations cpuset_dir_inode_operations = {
|
|
|
|
.lookup = simple_lookup,
|
|
|
|
.mkdir = cpuset_mkdir,
|
|
|
|
.rmdir = cpuset_rmdir,
|
2005-10-30 16:02:31 -07:00
|
|
|
.rename = cpuset_rename,
|
2005-04-16 16:20:36 -06:00
|
|
|
};
|
|
|
|
|
|
|
|
static int cpuset_create_file(struct dentry *dentry, int mode)
|
|
|
|
{
|
|
|
|
struct inode *inode;
|
|
|
|
|
|
|
|
if (!dentry)
|
|
|
|
return -ENOENT;
|
|
|
|
if (dentry->d_inode)
|
|
|
|
return -EEXIST;
|
|
|
|
|
|
|
|
inode = cpuset_new_inode(mode);
|
|
|
|
if (!inode)
|
|
|
|
return -ENOMEM;
|
|
|
|
|
|
|
|
if (S_ISDIR(mode)) {
|
|
|
|
inode->i_op = &cpuset_dir_inode_operations;
|
|
|
|
inode->i_fop = &simple_dir_operations;
|
|
|
|
|
|
|
|
/* start off with i_nlink == 2 (for "." entry) */
|
|
|
|
inode->i_nlink++;
|
|
|
|
} else if (S_ISREG(mode)) {
|
|
|
|
inode->i_size = 0;
|
|
|
|
inode->i_fop = &cpuset_file_operations;
|
|
|
|
}
|
|
|
|
|
|
|
|
d_instantiate(dentry, inode);
|
|
|
|
dget(dentry); /* Extra count - pin the dentry in core */
|
|
|
|
return 0;
|
|
|
|
}
|
|
|
|
|
|
|
|
/*
|
|
|
|
* cpuset_create_dir - create a directory for an object.
|
2006-01-08 02:01:51 -07:00
|
|
|
* cs: the cpuset we create the directory for.
|
2005-04-16 16:20:36 -06:00
|
|
|
* It must have a valid ->parent field
|
|
|
|
* And we are going to fill its ->dentry field.
|
|
|
|
* name: The name to give to the cpuset directory. Will be copied.
|
|
|
|
* mode: mode to set on new directory.
|
|
|
|
*/
|
|
|
|
|
|
|
|
static int cpuset_create_dir(struct cpuset *cs, const char *name, int mode)
|
|
|
|
{
|
|
|
|
struct dentry *dentry = NULL;
|
|
|
|
struct dentry *parent;
|
|
|
|
int error = 0;
|
|
|
|
|
|
|
|
parent = cs->parent->dentry;
|
|
|
|
dentry = cpuset_get_dentry(parent, name);
|
|
|
|
if (IS_ERR(dentry))
|
|
|
|
return PTR_ERR(dentry);
|
|
|
|
error = cpuset_create_file(dentry, S_IFDIR | mode);
|
|
|
|
if (!error) {
|
|
|
|
dentry->d_fsdata = cs;
|
|
|
|
parent->d_inode->i_nlink++;
|
|
|
|
cs->dentry = dentry;
|
|
|
|
}
|
|
|
|
dput(dentry);
|
|
|
|
|
|
|
|
return error;
|
|
|
|
}
|
|
|
|
|
|
|
|
static int cpuset_add_file(struct dentry *dir, const struct cftype *cft)
|
|
|
|
{
|
|
|
|
struct dentry *dentry;
|
|
|
|
int error;
|
|
|
|
|
2006-01-09 16:59:24 -07:00
|
|
|
mutex_lock(&dir->d_inode->i_mutex);
|
2005-04-16 16:20:36 -06:00
|
|
|
dentry = cpuset_get_dentry(dir, cft->name);
|
|
|
|
if (!IS_ERR(dentry)) {
|
|
|
|
error = cpuset_create_file(dentry, 0644 | S_IFREG);
|
|
|
|
if (!error)
|
|
|
|
dentry->d_fsdata = (void *)cft;
|
|
|
|
dput(dentry);
|
|
|
|
} else
|
|
|
|
error = PTR_ERR(dentry);
|
2006-01-09 16:59:24 -07:00
|
|
|
mutex_unlock(&dir->d_inode->i_mutex);
|
2005-04-16 16:20:36 -06:00
|
|
|
return error;
|
|
|
|
}
|
|
|
|
|
|
|
|
/*
|
|
|
|
* Stuff for reading the 'tasks' file.
|
|
|
|
*
|
|
|
|
* Reading this file can return large amounts of data if a cpuset has
|
|
|
|
* *lots* of attached tasks. So it may need several calls to read(),
|
|
|
|
* but we cannot guarantee that the information we produce is correct
|
|
|
|
* unless we produce it entirely atomically.
|
|
|
|
*
|
|
|
|
* Upon tasks file open(), a struct ctr_struct is allocated, that
|
|
|
|
* will have a pointer to an array (also allocated here). The struct
|
|
|
|
* ctr_struct * is stored in file->private_data. Its resources will
|
|
|
|
* be freed by release() when the file is closed. The array is used
|
|
|
|
* to sprintf the PIDs and then used by read().
|
|
|
|
*/
|
|
|
|
|
|
|
|
/* cpusets_tasks_read array */
|
|
|
|
|
|
|
|
struct ctr_struct {
|
|
|
|
char *buf;
|
|
|
|
int bufsz;
|
|
|
|
};
|
|
|
|
|
|
|
|
/*
|
|
|
|
* Load into 'pidarray' up to 'npids' of the tasks using cpuset 'cs'.
|
[PATCH] cpusets: dual semaphore locking overhaul
Overhaul cpuset locking. Replace single semaphore with two semaphores.
The suggestion to use two locks was made by Roman Zippel.
Both locks are global. Code that wants to modify cpusets must first
acquire the exclusive manage_sem, which allows them read-only access to
cpusets, and holds off other would-be modifiers. Before making actual
changes, the second semaphore, callback_sem must be acquired as well. Code
that needs only to query cpusets must acquire callback_sem, which is also a
global exclusive lock.
The earlier problems with double tripping are avoided, because it is
allowed for holders of manage_sem to nest the second callback_sem lock, and
only callback_sem is needed by code called from within __alloc_pages(),
where the double tripping had been possible.
This is not quite the same as a normal read/write semaphore, because
obtaining read-only access with intent to change must hold off other such
attempts, while allowing read-only access w/o such intention. Changing
cpusets involves several related checks and changes, which must be done
while allowing read-only queries (to avoid the double trip), but while
ensuring nothing changes (holding off other would be modifiers.)
This overhaul of cpuset locking also makes careful use of task_lock() to
guard access to the task->cpuset pointer, closing a couple of race
conditions noticed while reading this code (thanks, Roman). I've never
seen these races fail in any use or test.
See further the comments in the code.
Signed-off-by: Paul Jackson <pj@sgi.com>
Signed-off-by: Andrew Morton <akpm@osdl.org>
Signed-off-by: Linus Torvalds <torvalds@osdl.org>
2005-10-30 16:02:30 -07:00
|
|
|
* Return actual number of pids loaded. No need to task_lock(p)
|
|
|
|
* when reading out p->cpuset, as we don't really care if it changes
|
|
|
|
* on the next cycle, and we are not going to try to dereference it.
|
2005-04-16 16:20:36 -06:00
|
|
|
*/
|
2006-01-14 14:20:43 -07:00
|
|
|
static int pid_array_load(pid_t *pidarray, int npids, struct cpuset *cs)
|
2005-04-16 16:20:36 -06:00
|
|
|
{
|
|
|
|
int n = 0;
|
|
|
|
struct task_struct *g, *p;
|
|
|
|
|
|
|
|
read_lock(&tasklist_lock);
|
|
|
|
|
|
|
|
do_each_thread(g, p) {
|
|
|
|
if (p->cpuset == cs) {
|
|
|
|
pidarray[n++] = p->pid;
|
|
|
|
if (unlikely(n == npids))
|
|
|
|
goto array_full;
|
|
|
|
}
|
|
|
|
} while_each_thread(g, p);
|
|
|
|
|
|
|
|
array_full:
|
|
|
|
read_unlock(&tasklist_lock);
|
|
|
|
return n;
|
|
|
|
}
|
|
|
|
|
|
|
|
static int cmppid(const void *a, const void *b)
|
|
|
|
{
|
|
|
|
return *(pid_t *)a - *(pid_t *)b;
|
|
|
|
}
|
|
|
|
|
|
|
|
/*
|
|
|
|
* Convert array 'a' of 'npids' pid_t's to a string of newline separated
|
|
|
|
* decimal pids in 'buf'. Don't write more than 'sz' chars, but return
|
|
|
|
* count 'cnt' of how many chars would be written if buf were large enough.
|
|
|
|
*/
|
|
|
|
static int pid_array_to_buf(char *buf, int sz, pid_t *a, int npids)
|
|
|
|
{
|
|
|
|
int cnt = 0;
|
|
|
|
int i;
|
|
|
|
|
|
|
|
for (i = 0; i < npids; i++)
|
|
|
|
cnt += snprintf(buf + cnt, max(sz - cnt, 0), "%d\n", a[i]);
|
|
|
|
return cnt;
|
|
|
|
}
|
|
|
|
|
[PATCH] cpusets: dual semaphore locking overhaul
Overhaul cpuset locking. Replace single semaphore with two semaphores.
The suggestion to use two locks was made by Roman Zippel.
Both locks are global. Code that wants to modify cpusets must first
acquire the exclusive manage_sem, which allows them read-only access to
cpusets, and holds off other would-be modifiers. Before making actual
changes, the second semaphore, callback_sem must be acquired as well. Code
that needs only to query cpusets must acquire callback_sem, which is also a
global exclusive lock.
The earlier problems with double tripping are avoided, because it is
allowed for holders of manage_sem to nest the second callback_sem lock, and
only callback_sem is needed by code called from within __alloc_pages(),
where the double tripping had been possible.
This is not quite the same as a normal read/write semaphore, because
obtaining read-only access with intent to change must hold off other such
attempts, while allowing read-only access w/o such intention. Changing
cpusets involves several related checks and changes, which must be done
while allowing read-only queries (to avoid the double trip), but while
ensuring nothing changes (holding off other would be modifiers.)
This overhaul of cpuset locking also makes careful use of task_lock() to
guard access to the task->cpuset pointer, closing a couple of race
conditions noticed while reading this code (thanks, Roman). I've never
seen these races fail in any use or test.
See further the comments in the code.
Signed-off-by: Paul Jackson <pj@sgi.com>
Signed-off-by: Andrew Morton <akpm@osdl.org>
Signed-off-by: Linus Torvalds <torvalds@osdl.org>
2005-10-30 16:02:30 -07:00
|
|
|
/*
|
|
|
|
* Handle an open on 'tasks' file. Prepare a buffer listing the
|
|
|
|
* process id's of tasks currently attached to the cpuset being opened.
|
|
|
|
*
|
2006-03-23 04:00:18 -07:00
|
|
|
* Does not require any specific cpuset mutexes, and does not take any.
|
[PATCH] cpusets: dual semaphore locking overhaul
Overhaul cpuset locking. Replace single semaphore with two semaphores.
The suggestion to use two locks was made by Roman Zippel.
Both locks are global. Code that wants to modify cpusets must first
acquire the exclusive manage_sem, which allows them read-only access to
cpusets, and holds off other would-be modifiers. Before making actual
changes, the second semaphore, callback_sem must be acquired as well. Code
that needs only to query cpusets must acquire callback_sem, which is also a
global exclusive lock.
The earlier problems with double tripping are avoided, because it is
allowed for holders of manage_sem to nest the second callback_sem lock, and
only callback_sem is needed by code called from within __alloc_pages(),
where the double tripping had been possible.
This is not quite the same as a normal read/write semaphore, because
obtaining read-only access with intent to change must hold off other such
attempts, while allowing read-only access w/o such intention. Changing
cpusets involves several related checks and changes, which must be done
while allowing read-only queries (to avoid the double trip), but while
ensuring nothing changes (holding off other would be modifiers.)
This overhaul of cpuset locking also makes careful use of task_lock() to
guard access to the task->cpuset pointer, closing a couple of race
conditions noticed while reading this code (thanks, Roman). I've never
seen these races fail in any use or test.
See further the comments in the code.
Signed-off-by: Paul Jackson <pj@sgi.com>
Signed-off-by: Andrew Morton <akpm@osdl.org>
Signed-off-by: Linus Torvalds <torvalds@osdl.org>
2005-10-30 16:02:30 -07:00
|
|
|
*/
|
2005-04-16 16:20:36 -06:00
|
|
|
static int cpuset_tasks_open(struct inode *unused, struct file *file)
|
|
|
|
{
|
|
|
|
struct cpuset *cs = __d_cs(file->f_dentry->d_parent);
|
|
|
|
struct ctr_struct *ctr;
|
|
|
|
pid_t *pidarray;
|
|
|
|
int npids;
|
|
|
|
char c;
|
|
|
|
|
|
|
|
if (!(file->f_mode & FMODE_READ))
|
|
|
|
return 0;
|
|
|
|
|
|
|
|
ctr = kmalloc(sizeof(*ctr), GFP_KERNEL);
|
|
|
|
if (!ctr)
|
|
|
|
goto err0;
|
|
|
|
|
|
|
|
/*
|
|
|
|
* If cpuset gets more users after we read count, we won't have
|
|
|
|
* enough space - tough. This race is indistinguishable to the
|
|
|
|
* caller from the case that the additional cpuset users didn't
|
|
|
|
* show up until sometime later on.
|
|
|
|
*/
|
|
|
|
npids = atomic_read(&cs->count);
|
|
|
|
pidarray = kmalloc(npids * sizeof(pid_t), GFP_KERNEL);
|
|
|
|
if (!pidarray)
|
|
|
|
goto err1;
|
|
|
|
|
|
|
|
npids = pid_array_load(pidarray, npids, cs);
|
|
|
|
sort(pidarray, npids, sizeof(pid_t), cmppid, NULL);
|
|
|
|
|
|
|
|
/* Call pid_array_to_buf() twice, first just to get bufsz */
|
|
|
|
ctr->bufsz = pid_array_to_buf(&c, sizeof(c), pidarray, npids) + 1;
|
|
|
|
ctr->buf = kmalloc(ctr->bufsz, GFP_KERNEL);
|
|
|
|
if (!ctr->buf)
|
|
|
|
goto err2;
|
|
|
|
ctr->bufsz = pid_array_to_buf(ctr->buf, ctr->bufsz, pidarray, npids);
|
|
|
|
|
|
|
|
kfree(pidarray);
|
|
|
|
file->private_data = ctr;
|
|
|
|
return 0;
|
|
|
|
|
|
|
|
err2:
|
|
|
|
kfree(pidarray);
|
|
|
|
err1:
|
|
|
|
kfree(ctr);
|
|
|
|
err0:
|
|
|
|
return -ENOMEM;
|
|
|
|
}
|
|
|
|
|
|
|
|
static ssize_t cpuset_tasks_read(struct file *file, char __user *buf,
|
|
|
|
size_t nbytes, loff_t *ppos)
|
|
|
|
{
|
|
|
|
struct ctr_struct *ctr = file->private_data;
|
|
|
|
|
|
|
|
if (*ppos + nbytes > ctr->bufsz)
|
|
|
|
nbytes = ctr->bufsz - *ppos;
|
|
|
|
if (copy_to_user(buf, ctr->buf + *ppos, nbytes))
|
|
|
|
return -EFAULT;
|
|
|
|
*ppos += nbytes;
|
|
|
|
return nbytes;
|
|
|
|
}
|
|
|
|
|
|
|
|
static int cpuset_tasks_release(struct inode *unused_inode, struct file *file)
|
|
|
|
{
|
|
|
|
struct ctr_struct *ctr;
|
|
|
|
|
|
|
|
if (file->f_mode & FMODE_READ) {
|
|
|
|
ctr = file->private_data;
|
|
|
|
kfree(ctr->buf);
|
|
|
|
kfree(ctr);
|
|
|
|
}
|
|
|
|
return 0;
|
|
|
|
}
|
|
|
|
|
|
|
|
/*
|
|
|
|
* for the common functions, 'private' gives the type of file
|
|
|
|
*/
|
|
|
|
|
|
|
|
static struct cftype cft_tasks = {
|
|
|
|
.name = "tasks",
|
|
|
|
.open = cpuset_tasks_open,
|
|
|
|
.read = cpuset_tasks_read,
|
|
|
|
.release = cpuset_tasks_release,
|
|
|
|
.private = FILE_TASKLIST,
|
|
|
|
};
|
|
|
|
|
|
|
|
static struct cftype cft_cpus = {
|
|
|
|
.name = "cpus",
|
|
|
|
.private = FILE_CPULIST,
|
|
|
|
};
|
|
|
|
|
|
|
|
static struct cftype cft_mems = {
|
|
|
|
.name = "mems",
|
|
|
|
.private = FILE_MEMLIST,
|
|
|
|
};
|
|
|
|
|
|
|
|
static struct cftype cft_cpu_exclusive = {
|
|
|
|
.name = "cpu_exclusive",
|
|
|
|
.private = FILE_CPU_EXCLUSIVE,
|
|
|
|
};
|
|
|
|
|
|
|
|
static struct cftype cft_mem_exclusive = {
|
|
|
|
.name = "mem_exclusive",
|
|
|
|
.private = FILE_MEM_EXCLUSIVE,
|
|
|
|
};
|
|
|
|
|
|
|
|
static struct cftype cft_notify_on_release = {
|
|
|
|
.name = "notify_on_release",
|
|
|
|
.private = FILE_NOTIFY_ON_RELEASE,
|
|
|
|
};
|
|
|
|
|
[PATCH] cpusets: swap migration interface
Add a boolean "memory_migrate" to each cpuset, represented by a file
containing "0" or "1" in each directory below /dev/cpuset.
It defaults to false (file contains "0"). It can be set true by writing
"1" to the file.
If true, then anytime that a task is attached to the cpuset so marked, the
pages of that task will be moved to that cpuset, preserving, to the extent
practical, the cpuset-relative placement of the pages.
Also anytime that a cpuset so marked has its memory placement changed (by
writing to its "mems" file), the tasks in that cpuset will have their pages
moved to the cpusets new nodes, preserving, to the extent practical, the
cpuset-relative placement of the moved pages.
Signed-off-by: Paul Jackson <pj@sgi.com>
Cc: Christoph Lameter <christoph@lameter.com>
Signed-off-by: Andrew Morton <akpm@osdl.org>
Signed-off-by: Linus Torvalds <torvalds@osdl.org>
2006-01-08 02:00:56 -07:00
|
|
|
static struct cftype cft_memory_migrate = {
|
|
|
|
.name = "memory_migrate",
|
|
|
|
.private = FILE_MEMORY_MIGRATE,
|
|
|
|
};
|
|
|
|
|
[PATCH] cpuset: memory pressure meter
Provide a simple per-cpuset metric of memory pressure, tracking the -rate-
that the tasks in a cpuset call try_to_free_pages(), the synchronous
(direct) memory reclaim code.
This enables batch managers monitoring jobs running in dedicated cpusets to
efficiently detect what level of memory pressure that job is causing.
This is useful both on tightly managed systems running a wide mix of
submitted jobs, which may choose to terminate or reprioritize jobs that are
trying to use more memory than allowed on the nodes assigned them, and with
tightly coupled, long running, massively parallel scientific computing jobs
that will dramatically fail to meet required performance goals if they
start to use more memory than allowed to them.
This patch just provides a very economical way for the batch manager to
monitor a cpuset for signs of memory pressure. It's up to the batch
manager or other user code to decide what to do about it and take action.
==> Unless this feature is enabled by writing "1" to the special file
/dev/cpuset/memory_pressure_enabled, the hook in the rebalance
code of __alloc_pages() for this metric reduces to simply noticing
that the cpuset_memory_pressure_enabled flag is zero. So only
systems that enable this feature will compute the metric.
Why a per-cpuset, running average:
Because this meter is per-cpuset, rather than per-task or mm, the
system load imposed by a batch scheduler monitoring this metric is
sharply reduced on large systems, because a scan of the tasklist can be
avoided on each set of queries.
Because this meter is a running average, instead of an accumulating
counter, a batch scheduler can detect memory pressure with a single
read, instead of having to read and accumulate results for a period of
time.
Because this meter is per-cpuset rather than per-task or mm, the
batch scheduler can obtain the key information, memory pressure in a
cpuset, with a single read, rather than having to query and accumulate
results over all the (dynamically changing) set of tasks in the cpuset.
A per-cpuset simple digital filter (requires a spinlock and 3 words of data
per-cpuset) is kept, and updated by any task attached to that cpuset, if it
enters the synchronous (direct) page reclaim code.
A per-cpuset file provides an integer number representing the recent
(half-life of 10 seconds) rate of direct page reclaims caused by the tasks
in the cpuset, in units of reclaims attempted per second, times 1000.
Signed-off-by: Paul Jackson <pj@sgi.com>
Signed-off-by: Andrew Morton <akpm@osdl.org>
Signed-off-by: Linus Torvalds <torvalds@osdl.org>
2006-01-08 02:01:49 -07:00
|
|
|
static struct cftype cft_memory_pressure_enabled = {
|
|
|
|
.name = "memory_pressure_enabled",
|
|
|
|
.private = FILE_MEMORY_PRESSURE_ENABLED,
|
|
|
|
};
|
|
|
|
|
|
|
|
static struct cftype cft_memory_pressure = {
|
|
|
|
.name = "memory_pressure",
|
|
|
|
.private = FILE_MEMORY_PRESSURE,
|
|
|
|
};
|
|
|
|
|
[PATCH] cpuset memory spread basic implementation
This patch provides the implementation and cpuset interface for an alternative
memory allocation policy that can be applied to certain kinds of memory
allocations, such as the page cache (file system buffers) and some slab caches
(such as inode caches).
The policy is called "memory spreading." If enabled, it spreads out these
kinds of memory allocations over all the nodes allowed to a task, instead of
preferring to place them on the node where the task is executing.
All other kinds of allocations, including anonymous pages for a tasks stack
and data regions, are not affected by this policy choice, and continue to be
allocated preferring the node local to execution, as modified by the NUMA
mempolicy.
There are two boolean flag files per cpuset that control where the kernel
allocates pages for the file system buffers and related in kernel data
structures. They are called 'memory_spread_page' and 'memory_spread_slab'.
If the per-cpuset boolean flag file 'memory_spread_page' is set, then the
kernel will spread the file system buffers (page cache) evenly over all the
nodes that the faulting task is allowed to use, instead of preferring to put
those pages on the node where the task is running.
If the per-cpuset boolean flag file 'memory_spread_slab' is set, then the
kernel will spread some file system related slab caches, such as for inodes
and dentries evenly over all the nodes that the faulting task is allowed to
use, instead of preferring to put those pages on the node where the task is
running.
The implementation is simple. Setting the cpuset flags 'memory_spread_page'
or 'memory_spread_cache' turns on the per-process flags PF_SPREAD_PAGE or
PF_SPREAD_SLAB, respectively, for each task that is in the cpuset or
subsequently joins that cpuset. In subsequent patches, the page allocation
calls for the affected page cache and slab caches are modified to perform an
inline check for these flags, and if set, a call to a new routine
cpuset_mem_spread_node() returns the node to prefer for the allocation.
The cpuset_mem_spread_node() routine is also simple. It uses the value of a
per-task rotor cpuset_mem_spread_rotor to select the next node in the current
tasks mems_allowed to prefer for the allocation.
This policy can provide substantial improvements for jobs that need to place
thread local data on the corresponding node, but that need to access large
file system data sets that need to be spread across the several nodes in the
jobs cpuset in order to fit. Without this patch, especially for jobs that
might have one thread reading in the data set, the memory allocation across
the nodes in the jobs cpuset can become very uneven.
A couple of Copyright year ranges are updated as well. And a couple of email
addresses that can be found in the MAINTAINERS file are removed.
Signed-off-by: Paul Jackson <pj@sgi.com>
Signed-off-by: Andrew Morton <akpm@osdl.org>
Signed-off-by: Linus Torvalds <torvalds@osdl.org>
2006-03-24 04:16:03 -07:00
|
|
|
static struct cftype cft_spread_page = {
|
|
|
|
.name = "memory_spread_page",
|
|
|
|
.private = FILE_SPREAD_PAGE,
|
|
|
|
};
|
|
|
|
|
|
|
|
static struct cftype cft_spread_slab = {
|
|
|
|
.name = "memory_spread_slab",
|
|
|
|
.private = FILE_SPREAD_SLAB,
|
|
|
|
};
|
|
|
|
|
2005-04-16 16:20:36 -06:00
|
|
|
static int cpuset_populate_dir(struct dentry *cs_dentry)
|
|
|
|
{
|
|
|
|
int err;
|
|
|
|
|
|
|
|
if ((err = cpuset_add_file(cs_dentry, &cft_cpus)) < 0)
|
|
|
|
return err;
|
|
|
|
if ((err = cpuset_add_file(cs_dentry, &cft_mems)) < 0)
|
|
|
|
return err;
|
|
|
|
if ((err = cpuset_add_file(cs_dentry, &cft_cpu_exclusive)) < 0)
|
|
|
|
return err;
|
|
|
|
if ((err = cpuset_add_file(cs_dentry, &cft_mem_exclusive)) < 0)
|
|
|
|
return err;
|
|
|
|
if ((err = cpuset_add_file(cs_dentry, &cft_notify_on_release)) < 0)
|
|
|
|
return err;
|
[PATCH] cpusets: swap migration interface
Add a boolean "memory_migrate" to each cpuset, represented by a file
containing "0" or "1" in each directory below /dev/cpuset.
It defaults to false (file contains "0"). It can be set true by writing
"1" to the file.
If true, then anytime that a task is attached to the cpuset so marked, the
pages of that task will be moved to that cpuset, preserving, to the extent
practical, the cpuset-relative placement of the pages.
Also anytime that a cpuset so marked has its memory placement changed (by
writing to its "mems" file), the tasks in that cpuset will have their pages
moved to the cpusets new nodes, preserving, to the extent practical, the
cpuset-relative placement of the moved pages.
Signed-off-by: Paul Jackson <pj@sgi.com>
Cc: Christoph Lameter <christoph@lameter.com>
Signed-off-by: Andrew Morton <akpm@osdl.org>
Signed-off-by: Linus Torvalds <torvalds@osdl.org>
2006-01-08 02:00:56 -07:00
|
|
|
if ((err = cpuset_add_file(cs_dentry, &cft_memory_migrate)) < 0)
|
|
|
|
return err;
|
[PATCH] cpuset: memory pressure meter
Provide a simple per-cpuset metric of memory pressure, tracking the -rate-
that the tasks in a cpuset call try_to_free_pages(), the synchronous
(direct) memory reclaim code.
This enables batch managers monitoring jobs running in dedicated cpusets to
efficiently detect what level of memory pressure that job is causing.
This is useful both on tightly managed systems running a wide mix of
submitted jobs, which may choose to terminate or reprioritize jobs that are
trying to use more memory than allowed on the nodes assigned them, and with
tightly coupled, long running, massively parallel scientific computing jobs
that will dramatically fail to meet required performance goals if they
start to use more memory than allowed to them.
This patch just provides a very economical way for the batch manager to
monitor a cpuset for signs of memory pressure. It's up to the batch
manager or other user code to decide what to do about it and take action.
==> Unless this feature is enabled by writing "1" to the special file
/dev/cpuset/memory_pressure_enabled, the hook in the rebalance
code of __alloc_pages() for this metric reduces to simply noticing
that the cpuset_memory_pressure_enabled flag is zero. So only
systems that enable this feature will compute the metric.
Why a per-cpuset, running average:
Because this meter is per-cpuset, rather than per-task or mm, the
system load imposed by a batch scheduler monitoring this metric is
sharply reduced on large systems, because a scan of the tasklist can be
avoided on each set of queries.
Because this meter is a running average, instead of an accumulating
counter, a batch scheduler can detect memory pressure with a single
read, instead of having to read and accumulate results for a period of
time.
Because this meter is per-cpuset rather than per-task or mm, the
batch scheduler can obtain the key information, memory pressure in a
cpuset, with a single read, rather than having to query and accumulate
results over all the (dynamically changing) set of tasks in the cpuset.
A per-cpuset simple digital filter (requires a spinlock and 3 words of data
per-cpuset) is kept, and updated by any task attached to that cpuset, if it
enters the synchronous (direct) page reclaim code.
A per-cpuset file provides an integer number representing the recent
(half-life of 10 seconds) rate of direct page reclaims caused by the tasks
in the cpuset, in units of reclaims attempted per second, times 1000.
Signed-off-by: Paul Jackson <pj@sgi.com>
Signed-off-by: Andrew Morton <akpm@osdl.org>
Signed-off-by: Linus Torvalds <torvalds@osdl.org>
2006-01-08 02:01:49 -07:00
|
|
|
if ((err = cpuset_add_file(cs_dentry, &cft_memory_pressure)) < 0)
|
|
|
|
return err;
|
[PATCH] cpuset memory spread basic implementation
This patch provides the implementation and cpuset interface for an alternative
memory allocation policy that can be applied to certain kinds of memory
allocations, such as the page cache (file system buffers) and some slab caches
(such as inode caches).
The policy is called "memory spreading." If enabled, it spreads out these
kinds of memory allocations over all the nodes allowed to a task, instead of
preferring to place them on the node where the task is executing.
All other kinds of allocations, including anonymous pages for a tasks stack
and data regions, are not affected by this policy choice, and continue to be
allocated preferring the node local to execution, as modified by the NUMA
mempolicy.
There are two boolean flag files per cpuset that control where the kernel
allocates pages for the file system buffers and related in kernel data
structures. They are called 'memory_spread_page' and 'memory_spread_slab'.
If the per-cpuset boolean flag file 'memory_spread_page' is set, then the
kernel will spread the file system buffers (page cache) evenly over all the
nodes that the faulting task is allowed to use, instead of preferring to put
those pages on the node where the task is running.
If the per-cpuset boolean flag file 'memory_spread_slab' is set, then the
kernel will spread some file system related slab caches, such as for inodes
and dentries evenly over all the nodes that the faulting task is allowed to
use, instead of preferring to put those pages on the node where the task is
running.
The implementation is simple. Setting the cpuset flags 'memory_spread_page'
or 'memory_spread_cache' turns on the per-process flags PF_SPREAD_PAGE or
PF_SPREAD_SLAB, respectively, for each task that is in the cpuset or
subsequently joins that cpuset. In subsequent patches, the page allocation
calls for the affected page cache and slab caches are modified to perform an
inline check for these flags, and if set, a call to a new routine
cpuset_mem_spread_node() returns the node to prefer for the allocation.
The cpuset_mem_spread_node() routine is also simple. It uses the value of a
per-task rotor cpuset_mem_spread_rotor to select the next node in the current
tasks mems_allowed to prefer for the allocation.
This policy can provide substantial improvements for jobs that need to place
thread local data on the corresponding node, but that need to access large
file system data sets that need to be spread across the several nodes in the
jobs cpuset in order to fit. Without this patch, especially for jobs that
might have one thread reading in the data set, the memory allocation across
the nodes in the jobs cpuset can become very uneven.
A couple of Copyright year ranges are updated as well. And a couple of email
addresses that can be found in the MAINTAINERS file are removed.
Signed-off-by: Paul Jackson <pj@sgi.com>
Signed-off-by: Andrew Morton <akpm@osdl.org>
Signed-off-by: Linus Torvalds <torvalds@osdl.org>
2006-03-24 04:16:03 -07:00
|
|
|
if ((err = cpuset_add_file(cs_dentry, &cft_spread_page)) < 0)
|
|
|
|
return err;
|
|
|
|
if ((err = cpuset_add_file(cs_dentry, &cft_spread_slab)) < 0)
|
|
|
|
return err;
|
2005-04-16 16:20:36 -06:00
|
|
|
if ((err = cpuset_add_file(cs_dentry, &cft_tasks)) < 0)
|
|
|
|
return err;
|
|
|
|
return 0;
|
|
|
|
}
|
|
|
|
|
|
|
|
/*
|
|
|
|
* cpuset_create - create a cpuset
|
|
|
|
* parent: cpuset that will be parent of the new cpuset.
|
|
|
|
* name: name of the new cpuset. Will be strcpy'ed.
|
|
|
|
* mode: mode to set on new inode
|
|
|
|
*
|
2006-03-23 04:00:18 -07:00
|
|
|
* Must be called with the mutex on the parent inode held
|
2005-04-16 16:20:36 -06:00
|
|
|
*/
|
|
|
|
|
|
|
|
static long cpuset_create(struct cpuset *parent, const char *name, int mode)
|
|
|
|
{
|
|
|
|
struct cpuset *cs;
|
|
|
|
int err;
|
|
|
|
|
|
|
|
cs = kmalloc(sizeof(*cs), GFP_KERNEL);
|
|
|
|
if (!cs)
|
|
|
|
return -ENOMEM;
|
|
|
|
|
2006-03-23 04:00:18 -07:00
|
|
|
mutex_lock(&manage_mutex);
|
2006-01-08 02:01:54 -07:00
|
|
|
cpuset_update_task_memory_state();
|
2005-04-16 16:20:36 -06:00
|
|
|
cs->flags = 0;
|
|
|
|
if (notify_on_release(parent))
|
|
|
|
set_bit(CS_NOTIFY_ON_RELEASE, &cs->flags);
|
[PATCH] cpuset memory spread basic implementation
This patch provides the implementation and cpuset interface for an alternative
memory allocation policy that can be applied to certain kinds of memory
allocations, such as the page cache (file system buffers) and some slab caches
(such as inode caches).
The policy is called "memory spreading." If enabled, it spreads out these
kinds of memory allocations over all the nodes allowed to a task, instead of
preferring to place them on the node where the task is executing.
All other kinds of allocations, including anonymous pages for a tasks stack
and data regions, are not affected by this policy choice, and continue to be
allocated preferring the node local to execution, as modified by the NUMA
mempolicy.
There are two boolean flag files per cpuset that control where the kernel
allocates pages for the file system buffers and related in kernel data
structures. They are called 'memory_spread_page' and 'memory_spread_slab'.
If the per-cpuset boolean flag file 'memory_spread_page' is set, then the
kernel will spread the file system buffers (page cache) evenly over all the
nodes that the faulting task is allowed to use, instead of preferring to put
those pages on the node where the task is running.
If the per-cpuset boolean flag file 'memory_spread_slab' is set, then the
kernel will spread some file system related slab caches, such as for inodes
and dentries evenly over all the nodes that the faulting task is allowed to
use, instead of preferring to put those pages on the node where the task is
running.
The implementation is simple. Setting the cpuset flags 'memory_spread_page'
or 'memory_spread_cache' turns on the per-process flags PF_SPREAD_PAGE or
PF_SPREAD_SLAB, respectively, for each task that is in the cpuset or
subsequently joins that cpuset. In subsequent patches, the page allocation
calls for the affected page cache and slab caches are modified to perform an
inline check for these flags, and if set, a call to a new routine
cpuset_mem_spread_node() returns the node to prefer for the allocation.
The cpuset_mem_spread_node() routine is also simple. It uses the value of a
per-task rotor cpuset_mem_spread_rotor to select the next node in the current
tasks mems_allowed to prefer for the allocation.
This policy can provide substantial improvements for jobs that need to place
thread local data on the corresponding node, but that need to access large
file system data sets that need to be spread across the several nodes in the
jobs cpuset in order to fit. Without this patch, especially for jobs that
might have one thread reading in the data set, the memory allocation across
the nodes in the jobs cpuset can become very uneven.
A couple of Copyright year ranges are updated as well. And a couple of email
addresses that can be found in the MAINTAINERS file are removed.
Signed-off-by: Paul Jackson <pj@sgi.com>
Signed-off-by: Andrew Morton <akpm@osdl.org>
Signed-off-by: Linus Torvalds <torvalds@osdl.org>
2006-03-24 04:16:03 -07:00
|
|
|
if (is_spread_page(parent))
|
|
|
|
set_bit(CS_SPREAD_PAGE, &cs->flags);
|
|
|
|
if (is_spread_slab(parent))
|
|
|
|
set_bit(CS_SPREAD_SLAB, &cs->flags);
|
2005-04-16 16:20:36 -06:00
|
|
|
cs->cpus_allowed = CPU_MASK_NONE;
|
|
|
|
cs->mems_allowed = NODE_MASK_NONE;
|
|
|
|
atomic_set(&cs->count, 0);
|
|
|
|
INIT_LIST_HEAD(&cs->sibling);
|
|
|
|
INIT_LIST_HEAD(&cs->children);
|
2006-03-24 04:16:11 -07:00
|
|
|
cs->mems_generation = cpuset_mems_generation++;
|
[PATCH] cpuset: memory pressure meter
Provide a simple per-cpuset metric of memory pressure, tracking the -rate-
that the tasks in a cpuset call try_to_free_pages(), the synchronous
(direct) memory reclaim code.
This enables batch managers monitoring jobs running in dedicated cpusets to
efficiently detect what level of memory pressure that job is causing.
This is useful both on tightly managed systems running a wide mix of
submitted jobs, which may choose to terminate or reprioritize jobs that are
trying to use more memory than allowed on the nodes assigned them, and with
tightly coupled, long running, massively parallel scientific computing jobs
that will dramatically fail to meet required performance goals if they
start to use more memory than allowed to them.
This patch just provides a very economical way for the batch manager to
monitor a cpuset for signs of memory pressure. It's up to the batch
manager or other user code to decide what to do about it and take action.
==> Unless this feature is enabled by writing "1" to the special file
/dev/cpuset/memory_pressure_enabled, the hook in the rebalance
code of __alloc_pages() for this metric reduces to simply noticing
that the cpuset_memory_pressure_enabled flag is zero. So only
systems that enable this feature will compute the metric.
Why a per-cpuset, running average:
Because this meter is per-cpuset, rather than per-task or mm, the
system load imposed by a batch scheduler monitoring this metric is
sharply reduced on large systems, because a scan of the tasklist can be
avoided on each set of queries.
Because this meter is a running average, instead of an accumulating
counter, a batch scheduler can detect memory pressure with a single
read, instead of having to read and accumulate results for a period of
time.
Because this meter is per-cpuset rather than per-task or mm, the
batch scheduler can obtain the key information, memory pressure in a
cpuset, with a single read, rather than having to query and accumulate
results over all the (dynamically changing) set of tasks in the cpuset.
A per-cpuset simple digital filter (requires a spinlock and 3 words of data
per-cpuset) is kept, and updated by any task attached to that cpuset, if it
enters the synchronous (direct) page reclaim code.
A per-cpuset file provides an integer number representing the recent
(half-life of 10 seconds) rate of direct page reclaims caused by the tasks
in the cpuset, in units of reclaims attempted per second, times 1000.
Signed-off-by: Paul Jackson <pj@sgi.com>
Signed-off-by: Andrew Morton <akpm@osdl.org>
Signed-off-by: Linus Torvalds <torvalds@osdl.org>
2006-01-08 02:01:49 -07:00
|
|
|
fmeter_init(&cs->fmeter);
|
2005-04-16 16:20:36 -06:00
|
|
|
|
|
|
|
cs->parent = parent;
|
|
|
|
|
2006-03-23 04:00:18 -07:00
|
|
|
mutex_lock(&callback_mutex);
|
2005-04-16 16:20:36 -06:00
|
|
|
list_add(&cs->sibling, &cs->parent->children);
|
2006-01-08 02:01:57 -07:00
|
|
|
number_of_cpusets++;
|
2006-03-23 04:00:18 -07:00
|
|
|
mutex_unlock(&callback_mutex);
|
2005-04-16 16:20:36 -06:00
|
|
|
|
|
|
|
err = cpuset_create_dir(cs, name, mode);
|
|
|
|
if (err < 0)
|
|
|
|
goto err;
|
|
|
|
|
|
|
|
/*
|
2006-03-23 04:00:18 -07:00
|
|
|
* Release manage_mutex before cpuset_populate_dir() because it
|
2006-01-09 16:59:24 -07:00
|
|
|
* will down() this new directory's i_mutex and if we race with
|
2005-04-16 16:20:36 -06:00
|
|
|
* another mkdir, we might deadlock.
|
|
|
|
*/
|
2006-03-23 04:00:18 -07:00
|
|
|
mutex_unlock(&manage_mutex);
|
2005-04-16 16:20:36 -06:00
|
|
|
|
|
|
|
err = cpuset_populate_dir(cs->dentry);
|
|
|
|
/* If err < 0, we have a half-filled directory - oh well ;) */
|
|
|
|
return 0;
|
|
|
|
err:
|
|
|
|
list_del(&cs->sibling);
|
2006-03-23 04:00:18 -07:00
|
|
|
mutex_unlock(&manage_mutex);
|
2005-04-16 16:20:36 -06:00
|
|
|
kfree(cs);
|
|
|
|
return err;
|
|
|
|
}
|
|
|
|
|
|
|
|
static int cpuset_mkdir(struct inode *dir, struct dentry *dentry, int mode)
|
|
|
|
{
|
|
|
|
struct cpuset *c_parent = dentry->d_parent->d_fsdata;
|
|
|
|
|
2006-01-09 16:59:24 -07:00
|
|
|
/* the vfs holds inode->i_mutex already */
|
2005-04-16 16:20:36 -06:00
|
|
|
return cpuset_create(c_parent, dentry->d_name.name, mode | S_IFDIR);
|
|
|
|
}
|
|
|
|
|
2006-07-23 12:36:08 -06:00
|
|
|
/*
|
|
|
|
* Locking note on the strange update_flag() call below:
|
|
|
|
*
|
|
|
|
* If the cpuset being removed is marked cpu_exclusive, then simulate
|
|
|
|
* turning cpu_exclusive off, which will call update_cpu_domains().
|
|
|
|
* The lock_cpu_hotplug() call in update_cpu_domains() must not be
|
|
|
|
* made while holding callback_mutex. Elsewhere the kernel nests
|
|
|
|
* callback_mutex inside lock_cpu_hotplug() calls. So the reverse
|
|
|
|
* nesting would risk an ABBA deadlock.
|
|
|
|
*/
|
|
|
|
|
2005-04-16 16:20:36 -06:00
|
|
|
static int cpuset_rmdir(struct inode *unused_dir, struct dentry *dentry)
|
|
|
|
{
|
|
|
|
struct cpuset *cs = dentry->d_fsdata;
|
|
|
|
struct dentry *d;
|
|
|
|
struct cpuset *parent;
|
[PATCH] cpuset release ABBA deadlock fix
Fix possible cpuset_sem ABBA deadlock if 'notify_on_release' set.
For a particular usage pattern, creating and destroying cpusets fairly
frequently using notify_on_release, on a very large system, this deadlock
can be seen every few days. If you are not using the cpuset
notify_on_release feature, you will never see this deadlock.
The existing code, on task exit (or cpuset deletion) did:
get cpuset_sem
if cpuset marked notify_on_release and is ready to release:
compute cpuset path relative to /dev/cpuset mount point
call_usermodehelper() forks /sbin/cpuset_release_agent with path
drop cpuset_sem
Unfortunately, the fork in call_usermodehelper can allocate memory, and
allocating memory can require cpuset_sem, if the mems_generation values
changed in the interim. This results in an ABBA deadlock, trying to obtain
cpuset_sem when it is already held by the current task.
To fix this, I put the cpuset path (which must be computed while holding
cpuset_sem) in a temporary buffer, to be used in the call_usermodehelper
call of /sbin/cpuset_release_agent only _after_ dropping cpuset_sem.
So the new logic is:
get cpuset_sem
if cpuset marked notify_on_release and is ready to release:
compute cpuset path relative to /dev/cpuset mount point
stash path in kmalloc'd buffer
drop cpuset_sem
call_usermodehelper() forks /sbin/cpuset_release_agent with path
free path
The sharp eyed reader might notice that this patch does not contain any
calls to kmalloc. The existing code in the check_for_release() routine was
already kmalloc'ing a buffer to hold the cpuset path. In the old code, it
just held the buffer for a few lines, over the cpuset_release_agent() call
that in turn invoked call_usermodehelper(). In the new code, with the
application of this patch, it returns that buffer via the new char
**ppathbuf parameter, for later use and freeing in cpuset_release_agent(),
which is called after cpuset_sem is dropped. Whereas the old code has just
one call to cpuset_release_agent(), right in the check_for_release()
routine, the new code has three calls to cpuset_release_agent(), from the
various places that a cpuset can be released.
This patch has been build and booted on SN2, and passed a stress test that
previously hit the deadlock within a few seconds.
Signed-off-by: Paul Jackson <pj@sgi.com>
Signed-off-by: Andrew Morton <akpm@osdl.org>
Signed-off-by: Linus Torvalds <torvalds@osdl.org>
2005-08-09 11:07:59 -06:00
|
|
|
char *pathbuf = NULL;
|
2005-04-16 16:20:36 -06:00
|
|
|
|
2006-01-09 16:59:24 -07:00
|
|
|
/* the vfs holds both inode->i_mutex already */
|
2005-04-16 16:20:36 -06:00
|
|
|
|
2006-03-23 04:00:18 -07:00
|
|
|
mutex_lock(&manage_mutex);
|
2006-01-08 02:01:54 -07:00
|
|
|
cpuset_update_task_memory_state();
|
2005-04-16 16:20:36 -06:00
|
|
|
if (atomic_read(&cs->count) > 0) {
|
2006-03-23 04:00:18 -07:00
|
|
|
mutex_unlock(&manage_mutex);
|
2005-04-16 16:20:36 -06:00
|
|
|
return -EBUSY;
|
|
|
|
}
|
|
|
|
if (!list_empty(&cs->children)) {
|
2006-03-23 04:00:18 -07:00
|
|
|
mutex_unlock(&manage_mutex);
|
2005-04-16 16:20:36 -06:00
|
|
|
return -EBUSY;
|
|
|
|
}
|
2006-07-23 12:36:08 -06:00
|
|
|
if (is_cpu_exclusive(cs)) {
|
|
|
|
int retval = update_flag(CS_CPU_EXCLUSIVE, cs, "0");
|
|
|
|
if (retval < 0) {
|
|
|
|
mutex_unlock(&manage_mutex);
|
|
|
|
return retval;
|
|
|
|
}
|
|
|
|
}
|
2005-04-16 16:20:36 -06:00
|
|
|
parent = cs->parent;
|
2006-03-23 04:00:18 -07:00
|
|
|
mutex_lock(&callback_mutex);
|
2005-04-16 16:20:36 -06:00
|
|
|
set_bit(CS_REMOVED, &cs->flags);
|
|
|
|
list_del(&cs->sibling); /* delete my sibling from parent->children */
|
2005-06-25 15:57:34 -06:00
|
|
|
spin_lock(&cs->dentry->d_lock);
|
2005-04-16 16:20:36 -06:00
|
|
|
d = dget(cs->dentry);
|
|
|
|
cs->dentry = NULL;
|
|
|
|
spin_unlock(&d->d_lock);
|
|
|
|
cpuset_d_remove_dir(d);
|
|
|
|
dput(d);
|
2006-01-08 02:01:57 -07:00
|
|
|
number_of_cpusets--;
|
2006-03-23 04:00:18 -07:00
|
|
|
mutex_unlock(&callback_mutex);
|
[PATCH] cpusets: dual semaphore locking overhaul
Overhaul cpuset locking. Replace single semaphore with two semaphores.
The suggestion to use two locks was made by Roman Zippel.
Both locks are global. Code that wants to modify cpusets must first
acquire the exclusive manage_sem, which allows them read-only access to
cpusets, and holds off other would-be modifiers. Before making actual
changes, the second semaphore, callback_sem must be acquired as well. Code
that needs only to query cpusets must acquire callback_sem, which is also a
global exclusive lock.
The earlier problems with double tripping are avoided, because it is
allowed for holders of manage_sem to nest the second callback_sem lock, and
only callback_sem is needed by code called from within __alloc_pages(),
where the double tripping had been possible.
This is not quite the same as a normal read/write semaphore, because
obtaining read-only access with intent to change must hold off other such
attempts, while allowing read-only access w/o such intention. Changing
cpusets involves several related checks and changes, which must be done
while allowing read-only queries (to avoid the double trip), but while
ensuring nothing changes (holding off other would be modifiers.)
This overhaul of cpuset locking also makes careful use of task_lock() to
guard access to the task->cpuset pointer, closing a couple of race
conditions noticed while reading this code (thanks, Roman). I've never
seen these races fail in any use or test.
See further the comments in the code.
Signed-off-by: Paul Jackson <pj@sgi.com>
Signed-off-by: Andrew Morton <akpm@osdl.org>
Signed-off-by: Linus Torvalds <torvalds@osdl.org>
2005-10-30 16:02:30 -07:00
|
|
|
if (list_empty(&parent->children))
|
|
|
|
check_for_release(parent, &pathbuf);
|
2006-03-23 04:00:18 -07:00
|
|
|
mutex_unlock(&manage_mutex);
|
[PATCH] cpuset release ABBA deadlock fix
Fix possible cpuset_sem ABBA deadlock if 'notify_on_release' set.
For a particular usage pattern, creating and destroying cpusets fairly
frequently using notify_on_release, on a very large system, this deadlock
can be seen every few days. If you are not using the cpuset
notify_on_release feature, you will never see this deadlock.
The existing code, on task exit (or cpuset deletion) did:
get cpuset_sem
if cpuset marked notify_on_release and is ready to release:
compute cpuset path relative to /dev/cpuset mount point
call_usermodehelper() forks /sbin/cpuset_release_agent with path
drop cpuset_sem
Unfortunately, the fork in call_usermodehelper can allocate memory, and
allocating memory can require cpuset_sem, if the mems_generation values
changed in the interim. This results in an ABBA deadlock, trying to obtain
cpuset_sem when it is already held by the current task.
To fix this, I put the cpuset path (which must be computed while holding
cpuset_sem) in a temporary buffer, to be used in the call_usermodehelper
call of /sbin/cpuset_release_agent only _after_ dropping cpuset_sem.
So the new logic is:
get cpuset_sem
if cpuset marked notify_on_release and is ready to release:
compute cpuset path relative to /dev/cpuset mount point
stash path in kmalloc'd buffer
drop cpuset_sem
call_usermodehelper() forks /sbin/cpuset_release_agent with path
free path
The sharp eyed reader might notice that this patch does not contain any
calls to kmalloc. The existing code in the check_for_release() routine was
already kmalloc'ing a buffer to hold the cpuset path. In the old code, it
just held the buffer for a few lines, over the cpuset_release_agent() call
that in turn invoked call_usermodehelper(). In the new code, with the
application of this patch, it returns that buffer via the new char
**ppathbuf parameter, for later use and freeing in cpuset_release_agent(),
which is called after cpuset_sem is dropped. Whereas the old code has just
one call to cpuset_release_agent(), right in the check_for_release()
routine, the new code has three calls to cpuset_release_agent(), from the
various places that a cpuset can be released.
This patch has been build and booted on SN2, and passed a stress test that
previously hit the deadlock within a few seconds.
Signed-off-by: Paul Jackson <pj@sgi.com>
Signed-off-by: Andrew Morton <akpm@osdl.org>
Signed-off-by: Linus Torvalds <torvalds@osdl.org>
2005-08-09 11:07:59 -06:00
|
|
|
cpuset_release_agent(pathbuf);
|
2005-04-16 16:20:36 -06:00
|
|
|
return 0;
|
|
|
|
}
|
|
|
|
|
2006-01-08 02:02:01 -07:00
|
|
|
/*
|
|
|
|
* cpuset_init_early - just enough so that the calls to
|
|
|
|
* cpuset_update_task_memory_state() in early init code
|
|
|
|
* are harmless.
|
|
|
|
*/
|
|
|
|
|
|
|
|
int __init cpuset_init_early(void)
|
|
|
|
{
|
|
|
|
struct task_struct *tsk = current;
|
|
|
|
|
|
|
|
tsk->cpuset = &top_cpuset;
|
2006-03-24 04:16:11 -07:00
|
|
|
tsk->cpuset->mems_generation = cpuset_mems_generation++;
|
2006-01-08 02:02:01 -07:00
|
|
|
return 0;
|
|
|
|
}
|
|
|
|
|
2005-04-16 16:20:36 -06:00
|
|
|
/**
|
|
|
|
* cpuset_init - initialize cpusets at system boot
|
|
|
|
*
|
|
|
|
* Description: Initialize top_cpuset and the cpuset internal file system,
|
|
|
|
**/
|
|
|
|
|
|
|
|
int __init cpuset_init(void)
|
|
|
|
{
|
|
|
|
struct dentry *root;
|
|
|
|
int err;
|
|
|
|
|
|
|
|
top_cpuset.cpus_allowed = CPU_MASK_ALL;
|
|
|
|
top_cpuset.mems_allowed = NODE_MASK_ALL;
|
|
|
|
|
[PATCH] cpuset: memory pressure meter
Provide a simple per-cpuset metric of memory pressure, tracking the -rate-
that the tasks in a cpuset call try_to_free_pages(), the synchronous
(direct) memory reclaim code.
This enables batch managers monitoring jobs running in dedicated cpusets to
efficiently detect what level of memory pressure that job is causing.
This is useful both on tightly managed systems running a wide mix of
submitted jobs, which may choose to terminate or reprioritize jobs that are
trying to use more memory than allowed on the nodes assigned them, and with
tightly coupled, long running, massively parallel scientific computing jobs
that will dramatically fail to meet required performance goals if they
start to use more memory than allowed to them.
This patch just provides a very economical way for the batch manager to
monitor a cpuset for signs of memory pressure. It's up to the batch
manager or other user code to decide what to do about it and take action.
==> Unless this feature is enabled by writing "1" to the special file
/dev/cpuset/memory_pressure_enabled, the hook in the rebalance
code of __alloc_pages() for this metric reduces to simply noticing
that the cpuset_memory_pressure_enabled flag is zero. So only
systems that enable this feature will compute the metric.
Why a per-cpuset, running average:
Because this meter is per-cpuset, rather than per-task or mm, the
system load imposed by a batch scheduler monitoring this metric is
sharply reduced on large systems, because a scan of the tasklist can be
avoided on each set of queries.
Because this meter is a running average, instead of an accumulating
counter, a batch scheduler can detect memory pressure with a single
read, instead of having to read and accumulate results for a period of
time.
Because this meter is per-cpuset rather than per-task or mm, the
batch scheduler can obtain the key information, memory pressure in a
cpuset, with a single read, rather than having to query and accumulate
results over all the (dynamically changing) set of tasks in the cpuset.
A per-cpuset simple digital filter (requires a spinlock and 3 words of data
per-cpuset) is kept, and updated by any task attached to that cpuset, if it
enters the synchronous (direct) page reclaim code.
A per-cpuset file provides an integer number representing the recent
(half-life of 10 seconds) rate of direct page reclaims caused by the tasks
in the cpuset, in units of reclaims attempted per second, times 1000.
Signed-off-by: Paul Jackson <pj@sgi.com>
Signed-off-by: Andrew Morton <akpm@osdl.org>
Signed-off-by: Linus Torvalds <torvalds@osdl.org>
2006-01-08 02:01:49 -07:00
|
|
|
fmeter_init(&top_cpuset.fmeter);
|
2006-03-24 04:16:11 -07:00
|
|
|
top_cpuset.mems_generation = cpuset_mems_generation++;
|
2005-04-16 16:20:36 -06:00
|
|
|
|
|
|
|
init_task.cpuset = &top_cpuset;
|
|
|
|
|
|
|
|
err = register_filesystem(&cpuset_fs_type);
|
|
|
|
if (err < 0)
|
|
|
|
goto out;
|
|
|
|
cpuset_mount = kern_mount(&cpuset_fs_type);
|
|
|
|
if (IS_ERR(cpuset_mount)) {
|
|
|
|
printk(KERN_ERR "cpuset: could not mount!\n");
|
|
|
|
err = PTR_ERR(cpuset_mount);
|
|
|
|
cpuset_mount = NULL;
|
|
|
|
goto out;
|
|
|
|
}
|
|
|
|
root = cpuset_mount->mnt_sb->s_root;
|
|
|
|
root->d_fsdata = &top_cpuset;
|
|
|
|
root->d_inode->i_nlink++;
|
|
|
|
top_cpuset.dentry = root;
|
|
|
|
root->d_inode->i_op = &cpuset_dir_inode_operations;
|
2006-01-08 02:01:57 -07:00
|
|
|
number_of_cpusets = 1;
|
2005-04-16 16:20:36 -06:00
|
|
|
err = cpuset_populate_dir(root);
|
[PATCH] cpuset: memory pressure meter
Provide a simple per-cpuset metric of memory pressure, tracking the -rate-
that the tasks in a cpuset call try_to_free_pages(), the synchronous
(direct) memory reclaim code.
This enables batch managers monitoring jobs running in dedicated cpusets to
efficiently detect what level of memory pressure that job is causing.
This is useful both on tightly managed systems running a wide mix of
submitted jobs, which may choose to terminate or reprioritize jobs that are
trying to use more memory than allowed on the nodes assigned them, and with
tightly coupled, long running, massively parallel scientific computing jobs
that will dramatically fail to meet required performance goals if they
start to use more memory than allowed to them.
This patch just provides a very economical way for the batch manager to
monitor a cpuset for signs of memory pressure. It's up to the batch
manager or other user code to decide what to do about it and take action.
==> Unless this feature is enabled by writing "1" to the special file
/dev/cpuset/memory_pressure_enabled, the hook in the rebalance
code of __alloc_pages() for this metric reduces to simply noticing
that the cpuset_memory_pressure_enabled flag is zero. So only
systems that enable this feature will compute the metric.
Why a per-cpuset, running average:
Because this meter is per-cpuset, rather than per-task or mm, the
system load imposed by a batch scheduler monitoring this metric is
sharply reduced on large systems, because a scan of the tasklist can be
avoided on each set of queries.
Because this meter is a running average, instead of an accumulating
counter, a batch scheduler can detect memory pressure with a single
read, instead of having to read and accumulate results for a period of
time.
Because this meter is per-cpuset rather than per-task or mm, the
batch scheduler can obtain the key information, memory pressure in a
cpuset, with a single read, rather than having to query and accumulate
results over all the (dynamically changing) set of tasks in the cpuset.
A per-cpuset simple digital filter (requires a spinlock and 3 words of data
per-cpuset) is kept, and updated by any task attached to that cpuset, if it
enters the synchronous (direct) page reclaim code.
A per-cpuset file provides an integer number representing the recent
(half-life of 10 seconds) rate of direct page reclaims caused by the tasks
in the cpuset, in units of reclaims attempted per second, times 1000.
Signed-off-by: Paul Jackson <pj@sgi.com>
Signed-off-by: Andrew Morton <akpm@osdl.org>
Signed-off-by: Linus Torvalds <torvalds@osdl.org>
2006-01-08 02:01:49 -07:00
|
|
|
/* memory_pressure_enabled is in root cpuset only */
|
|
|
|
if (err == 0)
|
|
|
|
err = cpuset_add_file(root, &cft_memory_pressure_enabled);
|
2005-04-16 16:20:36 -06:00
|
|
|
out:
|
|
|
|
return err;
|
|
|
|
}
|
|
|
|
|
[PATCH] cpuset: top_cpuset tracks hotplug changes to cpu_online_map
Change the list of cpus allowed to tasks in the top (root) cpuset to
dynamically track what cpus are online, using a CPU hotplug notifier. Make
this top cpus file read-only.
On systems that have cpusets configured in their kernel, but that aren't
actively using cpusets (for some distros, this covers the majority of
systems) all tasks end up in the top cpuset.
If that system does support CPU hotplug, then these tasks cannot make use
of CPUs that are added after system boot, because the CPUs are not allowed
in the top cpuset. This is a surprising regression over earlier kernels
that didn't have cpusets enabled.
In order to keep the behaviour of cpusets consistent between systems
actively making use of them and systems not using them, this patch changes
the behaviour of the 'cpus' file in the top (root) cpuset, making it read
only, and making it automatically track the value of cpu_online_map. Thus
tasks in the top cpuset will have automatic use of hot plugged CPUs allowed
by their cpuset.
Thanks to Anton Blanchard and Nathan Lynch for reporting this problem,
driving the fix, and earlier versions of this patch.
Signed-off-by: Paul Jackson <pj@sgi.com>
Cc: Nathan Lynch <ntl@pobox.com>
Cc: Anton Blanchard <anton@samba.org>
Signed-off-by: Andrew Morton <akpm@osdl.org>
Signed-off-by: Linus Torvalds <torvalds@osdl.org>
2006-08-27 02:23:51 -06:00
|
|
|
/*
|
|
|
|
* The top_cpuset tracks what CPUs and Memory Nodes are online,
|
|
|
|
* period. This is necessary in order to make cpusets transparent
|
|
|
|
* (of no affect) on systems that are actively using CPU hotplug
|
|
|
|
* but making no active use of cpusets.
|
|
|
|
*
|
|
|
|
* This handles CPU hotplug (cpuhp) events. If someday Memory
|
|
|
|
* Nodes can be hotplugged (dynamically changing node_online_map)
|
|
|
|
* then we should handle that too, perhaps in a similar way.
|
|
|
|
*/
|
|
|
|
|
|
|
|
#ifdef CONFIG_HOTPLUG_CPU
|
|
|
|
static int cpuset_handle_cpuhp(struct notifier_block *nb,
|
|
|
|
unsigned long phase, void *cpu)
|
|
|
|
{
|
|
|
|
mutex_lock(&manage_mutex);
|
|
|
|
mutex_lock(&callback_mutex);
|
|
|
|
|
|
|
|
top_cpuset.cpus_allowed = cpu_online_map;
|
|
|
|
|
|
|
|
mutex_unlock(&callback_mutex);
|
|
|
|
mutex_unlock(&manage_mutex);
|
|
|
|
|
|
|
|
return 0;
|
|
|
|
}
|
|
|
|
#endif
|
|
|
|
|
2005-04-16 16:20:36 -06:00
|
|
|
/**
|
|
|
|
* cpuset_init_smp - initialize cpus_allowed
|
|
|
|
*
|
|
|
|
* Description: Finish top cpuset after cpu, node maps are initialized
|
|
|
|
**/
|
|
|
|
|
|
|
|
void __init cpuset_init_smp(void)
|
|
|
|
{
|
|
|
|
top_cpuset.cpus_allowed = cpu_online_map;
|
|
|
|
top_cpuset.mems_allowed = node_online_map;
|
[PATCH] cpuset: top_cpuset tracks hotplug changes to cpu_online_map
Change the list of cpus allowed to tasks in the top (root) cpuset to
dynamically track what cpus are online, using a CPU hotplug notifier. Make
this top cpus file read-only.
On systems that have cpusets configured in their kernel, but that aren't
actively using cpusets (for some distros, this covers the majority of
systems) all tasks end up in the top cpuset.
If that system does support CPU hotplug, then these tasks cannot make use
of CPUs that are added after system boot, because the CPUs are not allowed
in the top cpuset. This is a surprising regression over earlier kernels
that didn't have cpusets enabled.
In order to keep the behaviour of cpusets consistent between systems
actively making use of them and systems not using them, this patch changes
the behaviour of the 'cpus' file in the top (root) cpuset, making it read
only, and making it automatically track the value of cpu_online_map. Thus
tasks in the top cpuset will have automatic use of hot plugged CPUs allowed
by their cpuset.
Thanks to Anton Blanchard and Nathan Lynch for reporting this problem,
driving the fix, and earlier versions of this patch.
Signed-off-by: Paul Jackson <pj@sgi.com>
Cc: Nathan Lynch <ntl@pobox.com>
Cc: Anton Blanchard <anton@samba.org>
Signed-off-by: Andrew Morton <akpm@osdl.org>
Signed-off-by: Linus Torvalds <torvalds@osdl.org>
2006-08-27 02:23:51 -06:00
|
|
|
|
|
|
|
hotcpu_notifier(cpuset_handle_cpuhp, 0);
|
2005-04-16 16:20:36 -06:00
|
|
|
}
|
|
|
|
|
|
|
|
/**
|
|
|
|
* cpuset_fork - attach newly forked task to its parents cpuset.
|
2005-07-27 12:45:11 -06:00
|
|
|
* @tsk: pointer to task_struct of forking parent process.
|
2005-04-16 16:20:36 -06:00
|
|
|
*
|
[PATCH] cpusets: dual semaphore locking overhaul
Overhaul cpuset locking. Replace single semaphore with two semaphores.
The suggestion to use two locks was made by Roman Zippel.
Both locks are global. Code that wants to modify cpusets must first
acquire the exclusive manage_sem, which allows them read-only access to
cpusets, and holds off other would-be modifiers. Before making actual
changes, the second semaphore, callback_sem must be acquired as well. Code
that needs only to query cpusets must acquire callback_sem, which is also a
global exclusive lock.
The earlier problems with double tripping are avoided, because it is
allowed for holders of manage_sem to nest the second callback_sem lock, and
only callback_sem is needed by code called from within __alloc_pages(),
where the double tripping had been possible.
This is not quite the same as a normal read/write semaphore, because
obtaining read-only access with intent to change must hold off other such
attempts, while allowing read-only access w/o such intention. Changing
cpusets involves several related checks and changes, which must be done
while allowing read-only queries (to avoid the double trip), but while
ensuring nothing changes (holding off other would be modifiers.)
This overhaul of cpuset locking also makes careful use of task_lock() to
guard access to the task->cpuset pointer, closing a couple of race
conditions noticed while reading this code (thanks, Roman). I've never
seen these races fail in any use or test.
See further the comments in the code.
Signed-off-by: Paul Jackson <pj@sgi.com>
Signed-off-by: Andrew Morton <akpm@osdl.org>
Signed-off-by: Linus Torvalds <torvalds@osdl.org>
2005-10-30 16:02:30 -07:00
|
|
|
* Description: A task inherits its parent's cpuset at fork().
|
|
|
|
*
|
|
|
|
* A pointer to the shared cpuset was automatically copied in fork.c
|
|
|
|
* by dup_task_struct(). However, we ignore that copy, since it was
|
|
|
|
* not made under the protection of task_lock(), so might no longer be
|
|
|
|
* a valid cpuset pointer. attach_task() might have already changed
|
|
|
|
* current->cpuset, allowing the previously referenced cpuset to
|
|
|
|
* be removed and freed. Instead, we task_lock(current) and copy
|
|
|
|
* its present value of current->cpuset for our freshly forked child.
|
|
|
|
*
|
|
|
|
* At the point that cpuset_fork() is called, 'current' is the parent
|
|
|
|
* task, and the passed argument 'child' points to the child task.
|
2005-04-16 16:20:36 -06:00
|
|
|
**/
|
|
|
|
|
[PATCH] cpusets: dual semaphore locking overhaul
Overhaul cpuset locking. Replace single semaphore with two semaphores.
The suggestion to use two locks was made by Roman Zippel.
Both locks are global. Code that wants to modify cpusets must first
acquire the exclusive manage_sem, which allows them read-only access to
cpusets, and holds off other would-be modifiers. Before making actual
changes, the second semaphore, callback_sem must be acquired as well. Code
that needs only to query cpusets must acquire callback_sem, which is also a
global exclusive lock.
The earlier problems with double tripping are avoided, because it is
allowed for holders of manage_sem to nest the second callback_sem lock, and
only callback_sem is needed by code called from within __alloc_pages(),
where the double tripping had been possible.
This is not quite the same as a normal read/write semaphore, because
obtaining read-only access with intent to change must hold off other such
attempts, while allowing read-only access w/o such intention. Changing
cpusets involves several related checks and changes, which must be done
while allowing read-only queries (to avoid the double trip), but while
ensuring nothing changes (holding off other would be modifiers.)
This overhaul of cpuset locking also makes careful use of task_lock() to
guard access to the task->cpuset pointer, closing a couple of race
conditions noticed while reading this code (thanks, Roman). I've never
seen these races fail in any use or test.
See further the comments in the code.
Signed-off-by: Paul Jackson <pj@sgi.com>
Signed-off-by: Andrew Morton <akpm@osdl.org>
Signed-off-by: Linus Torvalds <torvalds@osdl.org>
2005-10-30 16:02:30 -07:00
|
|
|
void cpuset_fork(struct task_struct *child)
|
2005-04-16 16:20:36 -06:00
|
|
|
{
|
[PATCH] cpusets: dual semaphore locking overhaul
Overhaul cpuset locking. Replace single semaphore with two semaphores.
The suggestion to use two locks was made by Roman Zippel.
Both locks are global. Code that wants to modify cpusets must first
acquire the exclusive manage_sem, which allows them read-only access to
cpusets, and holds off other would-be modifiers. Before making actual
changes, the second semaphore, callback_sem must be acquired as well. Code
that needs only to query cpusets must acquire callback_sem, which is also a
global exclusive lock.
The earlier problems with double tripping are avoided, because it is
allowed for holders of manage_sem to nest the second callback_sem lock, and
only callback_sem is needed by code called from within __alloc_pages(),
where the double tripping had been possible.
This is not quite the same as a normal read/write semaphore, because
obtaining read-only access with intent to change must hold off other such
attempts, while allowing read-only access w/o such intention. Changing
cpusets involves several related checks and changes, which must be done
while allowing read-only queries (to avoid the double trip), but while
ensuring nothing changes (holding off other would be modifiers.)
This overhaul of cpuset locking also makes careful use of task_lock() to
guard access to the task->cpuset pointer, closing a couple of race
conditions noticed while reading this code (thanks, Roman). I've never
seen these races fail in any use or test.
See further the comments in the code.
Signed-off-by: Paul Jackson <pj@sgi.com>
Signed-off-by: Andrew Morton <akpm@osdl.org>
Signed-off-by: Linus Torvalds <torvalds@osdl.org>
2005-10-30 16:02:30 -07:00
|
|
|
task_lock(current);
|
|
|
|
child->cpuset = current->cpuset;
|
|
|
|
atomic_inc(&child->cpuset->count);
|
|
|
|
task_unlock(current);
|
2005-04-16 16:20:36 -06:00
|
|
|
}
|
|
|
|
|
|
|
|
/**
|
|
|
|
* cpuset_exit - detach cpuset from exiting task
|
|
|
|
* @tsk: pointer to task_struct of exiting process
|
|
|
|
*
|
|
|
|
* Description: Detach cpuset from @tsk and release it.
|
|
|
|
*
|
[PATCH] cpusets: dual semaphore locking overhaul
Overhaul cpuset locking. Replace single semaphore with two semaphores.
The suggestion to use two locks was made by Roman Zippel.
Both locks are global. Code that wants to modify cpusets must first
acquire the exclusive manage_sem, which allows them read-only access to
cpusets, and holds off other would-be modifiers. Before making actual
changes, the second semaphore, callback_sem must be acquired as well. Code
that needs only to query cpusets must acquire callback_sem, which is also a
global exclusive lock.
The earlier problems with double tripping are avoided, because it is
allowed for holders of manage_sem to nest the second callback_sem lock, and
only callback_sem is needed by code called from within __alloc_pages(),
where the double tripping had been possible.
This is not quite the same as a normal read/write semaphore, because
obtaining read-only access with intent to change must hold off other such
attempts, while allowing read-only access w/o such intention. Changing
cpusets involves several related checks and changes, which must be done
while allowing read-only queries (to avoid the double trip), but while
ensuring nothing changes (holding off other would be modifiers.)
This overhaul of cpuset locking also makes careful use of task_lock() to
guard access to the task->cpuset pointer, closing a couple of race
conditions noticed while reading this code (thanks, Roman). I've never
seen these races fail in any use or test.
See further the comments in the code.
Signed-off-by: Paul Jackson <pj@sgi.com>
Signed-off-by: Andrew Morton <akpm@osdl.org>
Signed-off-by: Linus Torvalds <torvalds@osdl.org>
2005-10-30 16:02:30 -07:00
|
|
|
* Note that cpusets marked notify_on_release force every task in
|
2006-03-23 04:00:18 -07:00
|
|
|
* them to take the global manage_mutex mutex when exiting.
|
[PATCH] cpusets: dual semaphore locking overhaul
Overhaul cpuset locking. Replace single semaphore with two semaphores.
The suggestion to use two locks was made by Roman Zippel.
Both locks are global. Code that wants to modify cpusets must first
acquire the exclusive manage_sem, which allows them read-only access to
cpusets, and holds off other would-be modifiers. Before making actual
changes, the second semaphore, callback_sem must be acquired as well. Code
that needs only to query cpusets must acquire callback_sem, which is also a
global exclusive lock.
The earlier problems with double tripping are avoided, because it is
allowed for holders of manage_sem to nest the second callback_sem lock, and
only callback_sem is needed by code called from within __alloc_pages(),
where the double tripping had been possible.
This is not quite the same as a normal read/write semaphore, because
obtaining read-only access with intent to change must hold off other such
attempts, while allowing read-only access w/o such intention. Changing
cpusets involves several related checks and changes, which must be done
while allowing read-only queries (to avoid the double trip), but while
ensuring nothing changes (holding off other would be modifiers.)
This overhaul of cpuset locking also makes careful use of task_lock() to
guard access to the task->cpuset pointer, closing a couple of race
conditions noticed while reading this code (thanks, Roman). I've never
seen these races fail in any use or test.
See further the comments in the code.
Signed-off-by: Paul Jackson <pj@sgi.com>
Signed-off-by: Andrew Morton <akpm@osdl.org>
Signed-off-by: Linus Torvalds <torvalds@osdl.org>
2005-10-30 16:02:30 -07:00
|
|
|
* This could impact scaling on very large systems. Be reluctant to
|
|
|
|
* use notify_on_release cpusets where very high task exit scaling
|
|
|
|
* is required on large systems.
|
|
|
|
*
|
|
|
|
* Don't even think about derefencing 'cs' after the cpuset use count
|
2006-03-23 04:00:18 -07:00
|
|
|
* goes to zero, except inside a critical section guarded by manage_mutex
|
|
|
|
* or callback_mutex. Otherwise a zero cpuset use count is a license to
|
[PATCH] cpusets: dual semaphore locking overhaul
Overhaul cpuset locking. Replace single semaphore with two semaphores.
The suggestion to use two locks was made by Roman Zippel.
Both locks are global. Code that wants to modify cpusets must first
acquire the exclusive manage_sem, which allows them read-only access to
cpusets, and holds off other would-be modifiers. Before making actual
changes, the second semaphore, callback_sem must be acquired as well. Code
that needs only to query cpusets must acquire callback_sem, which is also a
global exclusive lock.
The earlier problems with double tripping are avoided, because it is
allowed for holders of manage_sem to nest the second callback_sem lock, and
only callback_sem is needed by code called from within __alloc_pages(),
where the double tripping had been possible.
This is not quite the same as a normal read/write semaphore, because
obtaining read-only access with intent to change must hold off other such
attempts, while allowing read-only access w/o such intention. Changing
cpusets involves several related checks and changes, which must be done
while allowing read-only queries (to avoid the double trip), but while
ensuring nothing changes (holding off other would be modifiers.)
This overhaul of cpuset locking also makes careful use of task_lock() to
guard access to the task->cpuset pointer, closing a couple of race
conditions noticed while reading this code (thanks, Roman). I've never
seen these races fail in any use or test.
See further the comments in the code.
Signed-off-by: Paul Jackson <pj@sgi.com>
Signed-off-by: Andrew Morton <akpm@osdl.org>
Signed-off-by: Linus Torvalds <torvalds@osdl.org>
2005-10-30 16:02:30 -07:00
|
|
|
* any other task to nuke the cpuset immediately, via cpuset_rmdir().
|
|
|
|
*
|
2006-03-23 04:00:18 -07:00
|
|
|
* This routine has to take manage_mutex, not callback_mutex, because
|
|
|
|
* it is holding that mutex while calling check_for_release(),
|
|
|
|
* which calls kmalloc(), so can't be called holding callback_mutex().
|
[PATCH] cpusets: dual semaphore locking overhaul
Overhaul cpuset locking. Replace single semaphore with two semaphores.
The suggestion to use two locks was made by Roman Zippel.
Both locks are global. Code that wants to modify cpusets must first
acquire the exclusive manage_sem, which allows them read-only access to
cpusets, and holds off other would-be modifiers. Before making actual
changes, the second semaphore, callback_sem must be acquired as well. Code
that needs only to query cpusets must acquire callback_sem, which is also a
global exclusive lock.
The earlier problems with double tripping are avoided, because it is
allowed for holders of manage_sem to nest the second callback_sem lock, and
only callback_sem is needed by code called from within __alloc_pages(),
where the double tripping had been possible.
This is not quite the same as a normal read/write semaphore, because
obtaining read-only access with intent to change must hold off other such
attempts, while allowing read-only access w/o such intention. Changing
cpusets involves several related checks and changes, which must be done
while allowing read-only queries (to avoid the double trip), but while
ensuring nothing changes (holding off other would be modifiers.)
This overhaul of cpuset locking also makes careful use of task_lock() to
guard access to the task->cpuset pointer, closing a couple of race
conditions noticed while reading this code (thanks, Roman). I've never
seen these races fail in any use or test.
See further the comments in the code.
Signed-off-by: Paul Jackson <pj@sgi.com>
Signed-off-by: Andrew Morton <akpm@osdl.org>
Signed-off-by: Linus Torvalds <torvalds@osdl.org>
2005-10-30 16:02:30 -07:00
|
|
|
*
|
|
|
|
* We don't need to task_lock() this reference to tsk->cpuset,
|
|
|
|
* because tsk is already marked PF_EXITING, so attach_task() won't
|
[PATCH] cpuset: fork hook fix
Fix obscure, never seen in real life, cpuset fork race. The cpuset_fork()
call in fork.c was setting up the correct task->cpuset pointer after the
tasklist_lock was dropped, which briefly exposed the newly forked process with
an unsafe (copied from parent without locks or usage counter increment) cpuset
pointer.
In theory, that exposed cpuset pointer could have been pointing at a cpuset
that was already freed and removed, and in theory another task that had been
sitting on the tasklist_lock waiting to scan the task list could have raced
down the entire tasklist, found our new child at the far end, and dereferenced
that bogus cpuset pointer.
To fix, setup up the correct cpuset pointer in the new child by calling
cpuset_fork() before the new task is linked into the tasklist, and with that,
add a fork failure case, to dereference that cpuset, if the fork fails along
the way, after cpuset_fork() was called.
Had to remove a BUG_ON() from cpuset_exit(), because it was no longer valid -
the call to cpuset_exit() from a failed fork would not have PF_EXITING set.
Signed-off-by: Paul Jackson <pj@sgi.com>
Signed-off-by: Andrew Morton <akpm@osdl.org>
Signed-off-by: Linus Torvalds <torvalds@osdl.org>
2006-01-08 02:01:53 -07:00
|
|
|
* mess with it, or task is a failed fork, never visible to attach_task.
|
2006-02-15 16:17:38 -07:00
|
|
|
*
|
2006-03-24 04:16:10 -07:00
|
|
|
* the_top_cpuset_hack:
|
2006-02-15 16:17:38 -07:00
|
|
|
*
|
|
|
|
* Set the exiting tasks cpuset to the root cpuset (top_cpuset).
|
|
|
|
*
|
|
|
|
* Don't leave a task unable to allocate memory, as that is an
|
|
|
|
* accident waiting to happen should someone add a callout in
|
|
|
|
* do_exit() after the cpuset_exit() call that might allocate.
|
|
|
|
* If a task tries to allocate memory with an invalid cpuset,
|
|
|
|
* it will oops in cpuset_update_task_memory_state().
|
|
|
|
*
|
|
|
|
* We call cpuset_exit() while the task is still competent to
|
|
|
|
* handle notify_on_release(), then leave the task attached to
|
|
|
|
* the root cpuset (top_cpuset) for the remainder of its exit.
|
|
|
|
*
|
|
|
|
* To do this properly, we would increment the reference count on
|
|
|
|
* top_cpuset, and near the very end of the kernel/exit.c do_exit()
|
|
|
|
* code we would add a second cpuset function call, to drop that
|
|
|
|
* reference. This would just create an unnecessary hot spot on
|
|
|
|
* the top_cpuset reference count, to no avail.
|
|
|
|
*
|
|
|
|
* Normally, holding a reference to a cpuset without bumping its
|
|
|
|
* count is unsafe. The cpuset could go away, or someone could
|
|
|
|
* attach us to a different cpuset, decrementing the count on
|
|
|
|
* the first cpuset that we never incremented. But in this case,
|
|
|
|
* top_cpuset isn't going away, and either task has PF_EXITING set,
|
|
|
|
* which wards off any attach_task() attempts, or task is a failed
|
|
|
|
* fork, never visible to attach_task.
|
|
|
|
*
|
|
|
|
* Another way to do this would be to set the cpuset pointer
|
|
|
|
* to NULL here, and check in cpuset_update_task_memory_state()
|
|
|
|
* for a NULL pointer. This hack avoids that NULL check, for no
|
|
|
|
* cost (other than this way too long comment ;).
|
2005-04-16 16:20:36 -06:00
|
|
|
**/
|
|
|
|
|
|
|
|
void cpuset_exit(struct task_struct *tsk)
|
|
|
|
{
|
|
|
|
struct cpuset *cs;
|
|
|
|
|
|
|
|
cs = tsk->cpuset;
|
2006-03-24 04:16:10 -07:00
|
|
|
tsk->cpuset = &top_cpuset; /* the_top_cpuset_hack - see above */
|
2005-04-16 16:20:36 -06:00
|
|
|
|
[PATCH] cpuset exit NULL dereference fix
There is a race in the kernel cpuset code, between the code
to handle notify_on_release, and the code to remove a cpuset.
The notify_on_release code can end up trying to access a
cpuset that has been removed. In the most common case, this
causes a NULL pointer dereference from the routine cpuset_path.
However all manner of bad things are possible, in theory at least.
The existing code decrements the cpuset use count, and if the
count goes to zero, processes the notify_on_release request,
if appropriate. However, once the count goes to zero, unless we
are holding the global cpuset_sem semaphore, there is nothing to
stop another task from immediately removing the cpuset entirely,
and recycling its memory.
The obvious fix would be to always hold the cpuset_sem
semaphore while decrementing the use count and dealing with
notify_on_release. However we don't want to force a global
semaphore into the mainline task exit path, as that might create
a scaling problem.
The actual fix is almost as easy - since this is only an issue
for cpusets using notify_on_release, which the top level big
cpusets don't normally need to use, only take the cpuset_sem
for cpusets using notify_on_release.
This code has been run for hours without a hiccup, while running
a cpuset create/destroy stress test that could crash the existing
kernel in seconds. This patch applies to the current -linus
git kernel.
Signed-off-by: Paul Jackson <pj@sgi.com>
Acked-by: Simon Derr <simon.derr@bull.net>
Acked-by: Dinakar Guniguntala <dino@in.ibm.com>
Signed-off-by: Linus Torvalds <torvalds@osdl.org>
2005-05-27 03:02:43 -06:00
|
|
|
if (notify_on_release(cs)) {
|
[PATCH] cpuset release ABBA deadlock fix
Fix possible cpuset_sem ABBA deadlock if 'notify_on_release' set.
For a particular usage pattern, creating and destroying cpusets fairly
frequently using notify_on_release, on a very large system, this deadlock
can be seen every few days. If you are not using the cpuset
notify_on_release feature, you will never see this deadlock.
The existing code, on task exit (or cpuset deletion) did:
get cpuset_sem
if cpuset marked notify_on_release and is ready to release:
compute cpuset path relative to /dev/cpuset mount point
call_usermodehelper() forks /sbin/cpuset_release_agent with path
drop cpuset_sem
Unfortunately, the fork in call_usermodehelper can allocate memory, and
allocating memory can require cpuset_sem, if the mems_generation values
changed in the interim. This results in an ABBA deadlock, trying to obtain
cpuset_sem when it is already held by the current task.
To fix this, I put the cpuset path (which must be computed while holding
cpuset_sem) in a temporary buffer, to be used in the call_usermodehelper
call of /sbin/cpuset_release_agent only _after_ dropping cpuset_sem.
So the new logic is:
get cpuset_sem
if cpuset marked notify_on_release and is ready to release:
compute cpuset path relative to /dev/cpuset mount point
stash path in kmalloc'd buffer
drop cpuset_sem
call_usermodehelper() forks /sbin/cpuset_release_agent with path
free path
The sharp eyed reader might notice that this patch does not contain any
calls to kmalloc. The existing code in the check_for_release() routine was
already kmalloc'ing a buffer to hold the cpuset path. In the old code, it
just held the buffer for a few lines, over the cpuset_release_agent() call
that in turn invoked call_usermodehelper(). In the new code, with the
application of this patch, it returns that buffer via the new char
**ppathbuf parameter, for later use and freeing in cpuset_release_agent(),
which is called after cpuset_sem is dropped. Whereas the old code has just
one call to cpuset_release_agent(), right in the check_for_release()
routine, the new code has three calls to cpuset_release_agent(), from the
various places that a cpuset can be released.
This patch has been build and booted on SN2, and passed a stress test that
previously hit the deadlock within a few seconds.
Signed-off-by: Paul Jackson <pj@sgi.com>
Signed-off-by: Andrew Morton <akpm@osdl.org>
Signed-off-by: Linus Torvalds <torvalds@osdl.org>
2005-08-09 11:07:59 -06:00
|
|
|
char *pathbuf = NULL;
|
|
|
|
|
2006-03-23 04:00:18 -07:00
|
|
|
mutex_lock(&manage_mutex);
|
[PATCH] cpuset exit NULL dereference fix
There is a race in the kernel cpuset code, between the code
to handle notify_on_release, and the code to remove a cpuset.
The notify_on_release code can end up trying to access a
cpuset that has been removed. In the most common case, this
causes a NULL pointer dereference from the routine cpuset_path.
However all manner of bad things are possible, in theory at least.
The existing code decrements the cpuset use count, and if the
count goes to zero, processes the notify_on_release request,
if appropriate. However, once the count goes to zero, unless we
are holding the global cpuset_sem semaphore, there is nothing to
stop another task from immediately removing the cpuset entirely,
and recycling its memory.
The obvious fix would be to always hold the cpuset_sem
semaphore while decrementing the use count and dealing with
notify_on_release. However we don't want to force a global
semaphore into the mainline task exit path, as that might create
a scaling problem.
The actual fix is almost as easy - since this is only an issue
for cpusets using notify_on_release, which the top level big
cpusets don't normally need to use, only take the cpuset_sem
for cpusets using notify_on_release.
This code has been run for hours without a hiccup, while running
a cpuset create/destroy stress test that could crash the existing
kernel in seconds. This patch applies to the current -linus
git kernel.
Signed-off-by: Paul Jackson <pj@sgi.com>
Acked-by: Simon Derr <simon.derr@bull.net>
Acked-by: Dinakar Guniguntala <dino@in.ibm.com>
Signed-off-by: Linus Torvalds <torvalds@osdl.org>
2005-05-27 03:02:43 -06:00
|
|
|
if (atomic_dec_and_test(&cs->count))
|
[PATCH] cpuset release ABBA deadlock fix
Fix possible cpuset_sem ABBA deadlock if 'notify_on_release' set.
For a particular usage pattern, creating and destroying cpusets fairly
frequently using notify_on_release, on a very large system, this deadlock
can be seen every few days. If you are not using the cpuset
notify_on_release feature, you will never see this deadlock.
The existing code, on task exit (or cpuset deletion) did:
get cpuset_sem
if cpuset marked notify_on_release and is ready to release:
compute cpuset path relative to /dev/cpuset mount point
call_usermodehelper() forks /sbin/cpuset_release_agent with path
drop cpuset_sem
Unfortunately, the fork in call_usermodehelper can allocate memory, and
allocating memory can require cpuset_sem, if the mems_generation values
changed in the interim. This results in an ABBA deadlock, trying to obtain
cpuset_sem when it is already held by the current task.
To fix this, I put the cpuset path (which must be computed while holding
cpuset_sem) in a temporary buffer, to be used in the call_usermodehelper
call of /sbin/cpuset_release_agent only _after_ dropping cpuset_sem.
So the new logic is:
get cpuset_sem
if cpuset marked notify_on_release and is ready to release:
compute cpuset path relative to /dev/cpuset mount point
stash path in kmalloc'd buffer
drop cpuset_sem
call_usermodehelper() forks /sbin/cpuset_release_agent with path
free path
The sharp eyed reader might notice that this patch does not contain any
calls to kmalloc. The existing code in the check_for_release() routine was
already kmalloc'ing a buffer to hold the cpuset path. In the old code, it
just held the buffer for a few lines, over the cpuset_release_agent() call
that in turn invoked call_usermodehelper(). In the new code, with the
application of this patch, it returns that buffer via the new char
**ppathbuf parameter, for later use and freeing in cpuset_release_agent(),
which is called after cpuset_sem is dropped. Whereas the old code has just
one call to cpuset_release_agent(), right in the check_for_release()
routine, the new code has three calls to cpuset_release_agent(), from the
various places that a cpuset can be released.
This patch has been build and booted on SN2, and passed a stress test that
previously hit the deadlock within a few seconds.
Signed-off-by: Paul Jackson <pj@sgi.com>
Signed-off-by: Andrew Morton <akpm@osdl.org>
Signed-off-by: Linus Torvalds <torvalds@osdl.org>
2005-08-09 11:07:59 -06:00
|
|
|
check_for_release(cs, &pathbuf);
|
2006-03-23 04:00:18 -07:00
|
|
|
mutex_unlock(&manage_mutex);
|
[PATCH] cpuset release ABBA deadlock fix
Fix possible cpuset_sem ABBA deadlock if 'notify_on_release' set.
For a particular usage pattern, creating and destroying cpusets fairly
frequently using notify_on_release, on a very large system, this deadlock
can be seen every few days. If you are not using the cpuset
notify_on_release feature, you will never see this deadlock.
The existing code, on task exit (or cpuset deletion) did:
get cpuset_sem
if cpuset marked notify_on_release and is ready to release:
compute cpuset path relative to /dev/cpuset mount point
call_usermodehelper() forks /sbin/cpuset_release_agent with path
drop cpuset_sem
Unfortunately, the fork in call_usermodehelper can allocate memory, and
allocating memory can require cpuset_sem, if the mems_generation values
changed in the interim. This results in an ABBA deadlock, trying to obtain
cpuset_sem when it is already held by the current task.
To fix this, I put the cpuset path (which must be computed while holding
cpuset_sem) in a temporary buffer, to be used in the call_usermodehelper
call of /sbin/cpuset_release_agent only _after_ dropping cpuset_sem.
So the new logic is:
get cpuset_sem
if cpuset marked notify_on_release and is ready to release:
compute cpuset path relative to /dev/cpuset mount point
stash path in kmalloc'd buffer
drop cpuset_sem
call_usermodehelper() forks /sbin/cpuset_release_agent with path
free path
The sharp eyed reader might notice that this patch does not contain any
calls to kmalloc. The existing code in the check_for_release() routine was
already kmalloc'ing a buffer to hold the cpuset path. In the old code, it
just held the buffer for a few lines, over the cpuset_release_agent() call
that in turn invoked call_usermodehelper(). In the new code, with the
application of this patch, it returns that buffer via the new char
**ppathbuf parameter, for later use and freeing in cpuset_release_agent(),
which is called after cpuset_sem is dropped. Whereas the old code has just
one call to cpuset_release_agent(), right in the check_for_release()
routine, the new code has three calls to cpuset_release_agent(), from the
various places that a cpuset can be released.
This patch has been build and booted on SN2, and passed a stress test that
previously hit the deadlock within a few seconds.
Signed-off-by: Paul Jackson <pj@sgi.com>
Signed-off-by: Andrew Morton <akpm@osdl.org>
Signed-off-by: Linus Torvalds <torvalds@osdl.org>
2005-08-09 11:07:59 -06:00
|
|
|
cpuset_release_agent(pathbuf);
|
[PATCH] cpuset exit NULL dereference fix
There is a race in the kernel cpuset code, between the code
to handle notify_on_release, and the code to remove a cpuset.
The notify_on_release code can end up trying to access a
cpuset that has been removed. In the most common case, this
causes a NULL pointer dereference from the routine cpuset_path.
However all manner of bad things are possible, in theory at least.
The existing code decrements the cpuset use count, and if the
count goes to zero, processes the notify_on_release request,
if appropriate. However, once the count goes to zero, unless we
are holding the global cpuset_sem semaphore, there is nothing to
stop another task from immediately removing the cpuset entirely,
and recycling its memory.
The obvious fix would be to always hold the cpuset_sem
semaphore while decrementing the use count and dealing with
notify_on_release. However we don't want to force a global
semaphore into the mainline task exit path, as that might create
a scaling problem.
The actual fix is almost as easy - since this is only an issue
for cpusets using notify_on_release, which the top level big
cpusets don't normally need to use, only take the cpuset_sem
for cpusets using notify_on_release.
This code has been run for hours without a hiccup, while running
a cpuset create/destroy stress test that could crash the existing
kernel in seconds. This patch applies to the current -linus
git kernel.
Signed-off-by: Paul Jackson <pj@sgi.com>
Acked-by: Simon Derr <simon.derr@bull.net>
Acked-by: Dinakar Guniguntala <dino@in.ibm.com>
Signed-off-by: Linus Torvalds <torvalds@osdl.org>
2005-05-27 03:02:43 -06:00
|
|
|
} else {
|
|
|
|
atomic_dec(&cs->count);
|
2005-04-16 16:20:36 -06:00
|
|
|
}
|
|
|
|
}
|
|
|
|
|
|
|
|
/**
|
|
|
|
* cpuset_cpus_allowed - return cpus_allowed mask from a tasks cpuset.
|
|
|
|
* @tsk: pointer to task_struct from which to obtain cpuset->cpus_allowed.
|
|
|
|
*
|
|
|
|
* Description: Returns the cpumask_t cpus_allowed of the cpuset
|
|
|
|
* attached to the specified @tsk. Guaranteed to return some non-empty
|
|
|
|
* subset of cpu_online_map, even if this means going outside the
|
|
|
|
* tasks cpuset.
|
|
|
|
**/
|
|
|
|
|
2006-01-08 02:01:55 -07:00
|
|
|
cpumask_t cpuset_cpus_allowed(struct task_struct *tsk)
|
2005-04-16 16:20:36 -06:00
|
|
|
{
|
|
|
|
cpumask_t mask;
|
|
|
|
|
2006-03-23 04:00:18 -07:00
|
|
|
mutex_lock(&callback_mutex);
|
2006-01-08 02:01:55 -07:00
|
|
|
task_lock(tsk);
|
2005-04-16 16:20:36 -06:00
|
|
|
guarantee_online_cpus(tsk->cpuset, &mask);
|
2006-01-08 02:01:55 -07:00
|
|
|
task_unlock(tsk);
|
2006-03-23 04:00:18 -07:00
|
|
|
mutex_unlock(&callback_mutex);
|
2005-04-16 16:20:36 -06:00
|
|
|
|
|
|
|
return mask;
|
|
|
|
}
|
|
|
|
|
|
|
|
void cpuset_init_current_mems_allowed(void)
|
|
|
|
{
|
|
|
|
current->mems_allowed = NODE_MASK_ALL;
|
|
|
|
}
|
|
|
|
|
2006-01-08 02:01:55 -07:00
|
|
|
/**
|
|
|
|
* cpuset_mems_allowed - return mems_allowed mask from a tasks cpuset.
|
|
|
|
* @tsk: pointer to task_struct from which to obtain cpuset->mems_allowed.
|
|
|
|
*
|
|
|
|
* Description: Returns the nodemask_t mems_allowed of the cpuset
|
|
|
|
* attached to the specified @tsk. Guaranteed to return some non-empty
|
|
|
|
* subset of node_online_map, even if this means going outside the
|
|
|
|
* tasks cpuset.
|
|
|
|
**/
|
|
|
|
|
|
|
|
nodemask_t cpuset_mems_allowed(struct task_struct *tsk)
|
|
|
|
{
|
|
|
|
nodemask_t mask;
|
|
|
|
|
2006-03-23 04:00:18 -07:00
|
|
|
mutex_lock(&callback_mutex);
|
2006-01-08 02:01:55 -07:00
|
|
|
task_lock(tsk);
|
|
|
|
guarantee_online_mems(tsk->cpuset, &mask);
|
|
|
|
task_unlock(tsk);
|
2006-03-23 04:00:18 -07:00
|
|
|
mutex_unlock(&callback_mutex);
|
2006-01-08 02:01:55 -07:00
|
|
|
|
|
|
|
return mask;
|
|
|
|
}
|
|
|
|
|
2005-07-27 12:45:11 -06:00
|
|
|
/**
|
|
|
|
* cpuset_zonelist_valid_mems_allowed - check zonelist vs. curremt mems_allowed
|
|
|
|
* @zl: the zonelist to be checked
|
|
|
|
*
|
2005-04-16 16:20:36 -06:00
|
|
|
* Are any of the nodes on zonelist zl allowed in current->mems_allowed?
|
|
|
|
*/
|
|
|
|
int cpuset_zonelist_valid_mems_allowed(struct zonelist *zl)
|
|
|
|
{
|
|
|
|
int i;
|
|
|
|
|
|
|
|
for (i = 0; zl->zones[i]; i++) {
|
2006-09-26 00:31:55 -06:00
|
|
|
int nid = zone_to_nid(zl->zones[i]);
|
2005-04-16 16:20:36 -06:00
|
|
|
|
|
|
|
if (node_isset(nid, current->mems_allowed))
|
|
|
|
return 1;
|
|
|
|
}
|
|
|
|
return 0;
|
|
|
|
}
|
|
|
|
|
[PATCH] cpusets: formalize intermediate GFP_KERNEL containment
This patch makes use of the previously underutilized cpuset flag
'mem_exclusive' to provide what amounts to another layer of memory placement
resolution. With this patch, there are now the following four layers of
memory placement available:
1) The whole system (interrupt and GFP_ATOMIC allocations can use this),
2) The nearest enclosing mem_exclusive cpuset (GFP_KERNEL allocations can use),
3) The current tasks cpuset (GFP_USER allocations constrained to here), and
4) Specific node placement, using mbind and set_mempolicy.
These nest - each layer is a subset (same or within) of the previous.
Layer (2) above is new, with this patch. The call used to check whether a
zone (its node, actually) is in a cpuset (in its mems_allowed, actually) is
extended to take a gfp_mask argument, and its logic is extended, in the case
that __GFP_HARDWALL is not set in the flag bits, to look up the cpuset
hierarchy for the nearest enclosing mem_exclusive cpuset, to determine if
placement is allowed. The definition of GFP_USER, which used to be identical
to GFP_KERNEL, is changed to also set the __GFP_HARDWALL bit, in the previous
cpuset_gfp_hardwall_flag patch.
GFP_ATOMIC and GFP_KERNEL allocations will stay within the current tasks
cpuset, so long as any node therein is not too tight on memory, but will
escape to the larger layer, if need be.
The intended use is to allow something like a batch manager to handle several
jobs, each job in its own cpuset, but using common kernel memory for caches
and such. Swapper and oom_kill activity is also constrained to Layer (2). A
task in or below one mem_exclusive cpuset should not cause swapping on nodes
in another non-overlapping mem_exclusive cpuset, nor provoke oom_killing of a
task in another such cpuset. Heavy use of kernel memory for i/o caching and
such by one job should not impact the memory available to jobs in other
non-overlapping mem_exclusive cpusets.
This patch enables providing hardwall, inescapable cpusets for memory
allocations of each job, while sharing kernel memory allocations between
several jobs, in an enclosing mem_exclusive cpuset.
Like Dinakar's patch earlier to enable administering sched domains using the
cpu_exclusive flag, this patch also provides a useful meaning to a cpuset flag
that had previously done nothing much useful other than restrict what cpuset
configurations were allowed.
Signed-off-by: Paul Jackson <pj@sgi.com>
Signed-off-by: Andrew Morton <akpm@osdl.org>
Signed-off-by: Linus Torvalds <torvalds@osdl.org>
2005-09-06 16:18:12 -06:00
|
|
|
/*
|
|
|
|
* nearest_exclusive_ancestor() - Returns the nearest mem_exclusive
|
2006-03-23 04:00:18 -07:00
|
|
|
* ancestor to the specified cpuset. Call holding callback_mutex.
|
[PATCH] cpusets: formalize intermediate GFP_KERNEL containment
This patch makes use of the previously underutilized cpuset flag
'mem_exclusive' to provide what amounts to another layer of memory placement
resolution. With this patch, there are now the following four layers of
memory placement available:
1) The whole system (interrupt and GFP_ATOMIC allocations can use this),
2) The nearest enclosing mem_exclusive cpuset (GFP_KERNEL allocations can use),
3) The current tasks cpuset (GFP_USER allocations constrained to here), and
4) Specific node placement, using mbind and set_mempolicy.
These nest - each layer is a subset (same or within) of the previous.
Layer (2) above is new, with this patch. The call used to check whether a
zone (its node, actually) is in a cpuset (in its mems_allowed, actually) is
extended to take a gfp_mask argument, and its logic is extended, in the case
that __GFP_HARDWALL is not set in the flag bits, to look up the cpuset
hierarchy for the nearest enclosing mem_exclusive cpuset, to determine if
placement is allowed. The definition of GFP_USER, which used to be identical
to GFP_KERNEL, is changed to also set the __GFP_HARDWALL bit, in the previous
cpuset_gfp_hardwall_flag patch.
GFP_ATOMIC and GFP_KERNEL allocations will stay within the current tasks
cpuset, so long as any node therein is not too tight on memory, but will
escape to the larger layer, if need be.
The intended use is to allow something like a batch manager to handle several
jobs, each job in its own cpuset, but using common kernel memory for caches
and such. Swapper and oom_kill activity is also constrained to Layer (2). A
task in or below one mem_exclusive cpuset should not cause swapping on nodes
in another non-overlapping mem_exclusive cpuset, nor provoke oom_killing of a
task in another such cpuset. Heavy use of kernel memory for i/o caching and
such by one job should not impact the memory available to jobs in other
non-overlapping mem_exclusive cpusets.
This patch enables providing hardwall, inescapable cpusets for memory
allocations of each job, while sharing kernel memory allocations between
several jobs, in an enclosing mem_exclusive cpuset.
Like Dinakar's patch earlier to enable administering sched domains using the
cpu_exclusive flag, this patch also provides a useful meaning to a cpuset flag
that had previously done nothing much useful other than restrict what cpuset
configurations were allowed.
Signed-off-by: Paul Jackson <pj@sgi.com>
Signed-off-by: Andrew Morton <akpm@osdl.org>
Signed-off-by: Linus Torvalds <torvalds@osdl.org>
2005-09-06 16:18:12 -06:00
|
|
|
* If no ancestor is mem_exclusive (an unusual configuration), then
|
|
|
|
* returns the root cpuset.
|
|
|
|
*/
|
|
|
|
static const struct cpuset *nearest_exclusive_ancestor(const struct cpuset *cs)
|
|
|
|
{
|
|
|
|
while (!is_mem_exclusive(cs) && cs->parent)
|
|
|
|
cs = cs->parent;
|
|
|
|
return cs;
|
|
|
|
}
|
|
|
|
|
2005-07-27 12:45:11 -06:00
|
|
|
/**
|
[PATCH] cpusets: formalize intermediate GFP_KERNEL containment
This patch makes use of the previously underutilized cpuset flag
'mem_exclusive' to provide what amounts to another layer of memory placement
resolution. With this patch, there are now the following four layers of
memory placement available:
1) The whole system (interrupt and GFP_ATOMIC allocations can use this),
2) The nearest enclosing mem_exclusive cpuset (GFP_KERNEL allocations can use),
3) The current tasks cpuset (GFP_USER allocations constrained to here), and
4) Specific node placement, using mbind and set_mempolicy.
These nest - each layer is a subset (same or within) of the previous.
Layer (2) above is new, with this patch. The call used to check whether a
zone (its node, actually) is in a cpuset (in its mems_allowed, actually) is
extended to take a gfp_mask argument, and its logic is extended, in the case
that __GFP_HARDWALL is not set in the flag bits, to look up the cpuset
hierarchy for the nearest enclosing mem_exclusive cpuset, to determine if
placement is allowed. The definition of GFP_USER, which used to be identical
to GFP_KERNEL, is changed to also set the __GFP_HARDWALL bit, in the previous
cpuset_gfp_hardwall_flag patch.
GFP_ATOMIC and GFP_KERNEL allocations will stay within the current tasks
cpuset, so long as any node therein is not too tight on memory, but will
escape to the larger layer, if need be.
The intended use is to allow something like a batch manager to handle several
jobs, each job in its own cpuset, but using common kernel memory for caches
and such. Swapper and oom_kill activity is also constrained to Layer (2). A
task in or below one mem_exclusive cpuset should not cause swapping on nodes
in another non-overlapping mem_exclusive cpuset, nor provoke oom_killing of a
task in another such cpuset. Heavy use of kernel memory for i/o caching and
such by one job should not impact the memory available to jobs in other
non-overlapping mem_exclusive cpusets.
This patch enables providing hardwall, inescapable cpusets for memory
allocations of each job, while sharing kernel memory allocations between
several jobs, in an enclosing mem_exclusive cpuset.
Like Dinakar's patch earlier to enable administering sched domains using the
cpu_exclusive flag, this patch also provides a useful meaning to a cpuset flag
that had previously done nothing much useful other than restrict what cpuset
configurations were allowed.
Signed-off-by: Paul Jackson <pj@sgi.com>
Signed-off-by: Andrew Morton <akpm@osdl.org>
Signed-off-by: Linus Torvalds <torvalds@osdl.org>
2005-09-06 16:18:12 -06:00
|
|
|
* cpuset_zone_allowed - Can we allocate memory on zone z's memory node?
|
|
|
|
* @z: is this zone on an allowed node?
|
|
|
|
* @gfp_mask: memory allocation flags (we use __GFP_HARDWALL)
|
2005-07-27 12:45:11 -06:00
|
|
|
*
|
[PATCH] cpusets: formalize intermediate GFP_KERNEL containment
This patch makes use of the previously underutilized cpuset flag
'mem_exclusive' to provide what amounts to another layer of memory placement
resolution. With this patch, there are now the following four layers of
memory placement available:
1) The whole system (interrupt and GFP_ATOMIC allocations can use this),
2) The nearest enclosing mem_exclusive cpuset (GFP_KERNEL allocations can use),
3) The current tasks cpuset (GFP_USER allocations constrained to here), and
4) Specific node placement, using mbind and set_mempolicy.
These nest - each layer is a subset (same or within) of the previous.
Layer (2) above is new, with this patch. The call used to check whether a
zone (its node, actually) is in a cpuset (in its mems_allowed, actually) is
extended to take a gfp_mask argument, and its logic is extended, in the case
that __GFP_HARDWALL is not set in the flag bits, to look up the cpuset
hierarchy for the nearest enclosing mem_exclusive cpuset, to determine if
placement is allowed. The definition of GFP_USER, which used to be identical
to GFP_KERNEL, is changed to also set the __GFP_HARDWALL bit, in the previous
cpuset_gfp_hardwall_flag patch.
GFP_ATOMIC and GFP_KERNEL allocations will stay within the current tasks
cpuset, so long as any node therein is not too tight on memory, but will
escape to the larger layer, if need be.
The intended use is to allow something like a batch manager to handle several
jobs, each job in its own cpuset, but using common kernel memory for caches
and such. Swapper and oom_kill activity is also constrained to Layer (2). A
task in or below one mem_exclusive cpuset should not cause swapping on nodes
in another non-overlapping mem_exclusive cpuset, nor provoke oom_killing of a
task in another such cpuset. Heavy use of kernel memory for i/o caching and
such by one job should not impact the memory available to jobs in other
non-overlapping mem_exclusive cpusets.
This patch enables providing hardwall, inescapable cpusets for memory
allocations of each job, while sharing kernel memory allocations between
several jobs, in an enclosing mem_exclusive cpuset.
Like Dinakar's patch earlier to enable administering sched domains using the
cpu_exclusive flag, this patch also provides a useful meaning to a cpuset flag
that had previously done nothing much useful other than restrict what cpuset
configurations were allowed.
Signed-off-by: Paul Jackson <pj@sgi.com>
Signed-off-by: Andrew Morton <akpm@osdl.org>
Signed-off-by: Linus Torvalds <torvalds@osdl.org>
2005-09-06 16:18:12 -06:00
|
|
|
* If we're in interrupt, yes, we can always allocate. If zone
|
|
|
|
* z's node is in our tasks mems_allowed, yes. If it's not a
|
|
|
|
* __GFP_HARDWALL request and this zone's nodes is in the nearest
|
|
|
|
* mem_exclusive cpuset ancestor to this tasks cpuset, yes.
|
|
|
|
* Otherwise, no.
|
|
|
|
*
|
|
|
|
* GFP_USER allocations are marked with the __GFP_HARDWALL bit,
|
|
|
|
* and do not allow allocations outside the current tasks cpuset.
|
|
|
|
* GFP_KERNEL allocations are not so marked, so can escape to the
|
|
|
|
* nearest mem_exclusive ancestor cpuset.
|
|
|
|
*
|
2006-03-23 04:00:18 -07:00
|
|
|
* Scanning up parent cpusets requires callback_mutex. The __alloc_pages()
|
[PATCH] cpusets: formalize intermediate GFP_KERNEL containment
This patch makes use of the previously underutilized cpuset flag
'mem_exclusive' to provide what amounts to another layer of memory placement
resolution. With this patch, there are now the following four layers of
memory placement available:
1) The whole system (interrupt and GFP_ATOMIC allocations can use this),
2) The nearest enclosing mem_exclusive cpuset (GFP_KERNEL allocations can use),
3) The current tasks cpuset (GFP_USER allocations constrained to here), and
4) Specific node placement, using mbind and set_mempolicy.
These nest - each layer is a subset (same or within) of the previous.
Layer (2) above is new, with this patch. The call used to check whether a
zone (its node, actually) is in a cpuset (in its mems_allowed, actually) is
extended to take a gfp_mask argument, and its logic is extended, in the case
that __GFP_HARDWALL is not set in the flag bits, to look up the cpuset
hierarchy for the nearest enclosing mem_exclusive cpuset, to determine if
placement is allowed. The definition of GFP_USER, which used to be identical
to GFP_KERNEL, is changed to also set the __GFP_HARDWALL bit, in the previous
cpuset_gfp_hardwall_flag patch.
GFP_ATOMIC and GFP_KERNEL allocations will stay within the current tasks
cpuset, so long as any node therein is not too tight on memory, but will
escape to the larger layer, if need be.
The intended use is to allow something like a batch manager to handle several
jobs, each job in its own cpuset, but using common kernel memory for caches
and such. Swapper and oom_kill activity is also constrained to Layer (2). A
task in or below one mem_exclusive cpuset should not cause swapping on nodes
in another non-overlapping mem_exclusive cpuset, nor provoke oom_killing of a
task in another such cpuset. Heavy use of kernel memory for i/o caching and
such by one job should not impact the memory available to jobs in other
non-overlapping mem_exclusive cpusets.
This patch enables providing hardwall, inescapable cpusets for memory
allocations of each job, while sharing kernel memory allocations between
several jobs, in an enclosing mem_exclusive cpuset.
Like Dinakar's patch earlier to enable administering sched domains using the
cpu_exclusive flag, this patch also provides a useful meaning to a cpuset flag
that had previously done nothing much useful other than restrict what cpuset
configurations were allowed.
Signed-off-by: Paul Jackson <pj@sgi.com>
Signed-off-by: Andrew Morton <akpm@osdl.org>
Signed-off-by: Linus Torvalds <torvalds@osdl.org>
2005-09-06 16:18:12 -06:00
|
|
|
* routine only calls here with __GFP_HARDWALL bit _not_ set if
|
|
|
|
* it's a GFP_KERNEL allocation, and all nodes in the current tasks
|
|
|
|
* mems_allowed came up empty on the first pass over the zonelist.
|
|
|
|
* So only GFP_KERNEL allocations, if all nodes in the cpuset are
|
2006-03-23 04:00:18 -07:00
|
|
|
* short of memory, might require taking the callback_mutex mutex.
|
[PATCH] cpusets: formalize intermediate GFP_KERNEL containment
This patch makes use of the previously underutilized cpuset flag
'mem_exclusive' to provide what amounts to another layer of memory placement
resolution. With this patch, there are now the following four layers of
memory placement available:
1) The whole system (interrupt and GFP_ATOMIC allocations can use this),
2) The nearest enclosing mem_exclusive cpuset (GFP_KERNEL allocations can use),
3) The current tasks cpuset (GFP_USER allocations constrained to here), and
4) Specific node placement, using mbind and set_mempolicy.
These nest - each layer is a subset (same or within) of the previous.
Layer (2) above is new, with this patch. The call used to check whether a
zone (its node, actually) is in a cpuset (in its mems_allowed, actually) is
extended to take a gfp_mask argument, and its logic is extended, in the case
that __GFP_HARDWALL is not set in the flag bits, to look up the cpuset
hierarchy for the nearest enclosing mem_exclusive cpuset, to determine if
placement is allowed. The definition of GFP_USER, which used to be identical
to GFP_KERNEL, is changed to also set the __GFP_HARDWALL bit, in the previous
cpuset_gfp_hardwall_flag patch.
GFP_ATOMIC and GFP_KERNEL allocations will stay within the current tasks
cpuset, so long as any node therein is not too tight on memory, but will
escape to the larger layer, if need be.
The intended use is to allow something like a batch manager to handle several
jobs, each job in its own cpuset, but using common kernel memory for caches
and such. Swapper and oom_kill activity is also constrained to Layer (2). A
task in or below one mem_exclusive cpuset should not cause swapping on nodes
in another non-overlapping mem_exclusive cpuset, nor provoke oom_killing of a
task in another such cpuset. Heavy use of kernel memory for i/o caching and
such by one job should not impact the memory available to jobs in other
non-overlapping mem_exclusive cpusets.
This patch enables providing hardwall, inescapable cpusets for memory
allocations of each job, while sharing kernel memory allocations between
several jobs, in an enclosing mem_exclusive cpuset.
Like Dinakar's patch earlier to enable administering sched domains using the
cpu_exclusive flag, this patch also provides a useful meaning to a cpuset flag
that had previously done nothing much useful other than restrict what cpuset
configurations were allowed.
Signed-off-by: Paul Jackson <pj@sgi.com>
Signed-off-by: Andrew Morton <akpm@osdl.org>
Signed-off-by: Linus Torvalds <torvalds@osdl.org>
2005-09-06 16:18:12 -06:00
|
|
|
*
|
2006-05-20 16:00:10 -06:00
|
|
|
* The first call here from mm/page_alloc:get_page_from_freelist()
|
|
|
|
* has __GFP_HARDWALL set in gfp_mask, enforcing hardwall cpusets, so
|
|
|
|
* no allocation on a node outside the cpuset is allowed (unless in
|
|
|
|
* interrupt, of course).
|
|
|
|
*
|
|
|
|
* The second pass through get_page_from_freelist() doesn't even call
|
|
|
|
* here for GFP_ATOMIC calls. For those calls, the __alloc_pages()
|
|
|
|
* variable 'wait' is not set, and the bit ALLOC_CPUSET is not set
|
|
|
|
* in alloc_flags. That logic and the checks below have the combined
|
|
|
|
* affect that:
|
[PATCH] cpusets: formalize intermediate GFP_KERNEL containment
This patch makes use of the previously underutilized cpuset flag
'mem_exclusive' to provide what amounts to another layer of memory placement
resolution. With this patch, there are now the following four layers of
memory placement available:
1) The whole system (interrupt and GFP_ATOMIC allocations can use this),
2) The nearest enclosing mem_exclusive cpuset (GFP_KERNEL allocations can use),
3) The current tasks cpuset (GFP_USER allocations constrained to here), and
4) Specific node placement, using mbind and set_mempolicy.
These nest - each layer is a subset (same or within) of the previous.
Layer (2) above is new, with this patch. The call used to check whether a
zone (its node, actually) is in a cpuset (in its mems_allowed, actually) is
extended to take a gfp_mask argument, and its logic is extended, in the case
that __GFP_HARDWALL is not set in the flag bits, to look up the cpuset
hierarchy for the nearest enclosing mem_exclusive cpuset, to determine if
placement is allowed. The definition of GFP_USER, which used to be identical
to GFP_KERNEL, is changed to also set the __GFP_HARDWALL bit, in the previous
cpuset_gfp_hardwall_flag patch.
GFP_ATOMIC and GFP_KERNEL allocations will stay within the current tasks
cpuset, so long as any node therein is not too tight on memory, but will
escape to the larger layer, if need be.
The intended use is to allow something like a batch manager to handle several
jobs, each job in its own cpuset, but using common kernel memory for caches
and such. Swapper and oom_kill activity is also constrained to Layer (2). A
task in or below one mem_exclusive cpuset should not cause swapping on nodes
in another non-overlapping mem_exclusive cpuset, nor provoke oom_killing of a
task in another such cpuset. Heavy use of kernel memory for i/o caching and
such by one job should not impact the memory available to jobs in other
non-overlapping mem_exclusive cpusets.
This patch enables providing hardwall, inescapable cpusets for memory
allocations of each job, while sharing kernel memory allocations between
several jobs, in an enclosing mem_exclusive cpuset.
Like Dinakar's patch earlier to enable administering sched domains using the
cpu_exclusive flag, this patch also provides a useful meaning to a cpuset flag
that had previously done nothing much useful other than restrict what cpuset
configurations were allowed.
Signed-off-by: Paul Jackson <pj@sgi.com>
Signed-off-by: Andrew Morton <akpm@osdl.org>
Signed-off-by: Linus Torvalds <torvalds@osdl.org>
2005-09-06 16:18:12 -06:00
|
|
|
* in_interrupt - any node ok (current task context irrelevant)
|
|
|
|
* GFP_ATOMIC - any node ok
|
|
|
|
* GFP_KERNEL - any node in enclosing mem_exclusive cpuset ok
|
|
|
|
* GFP_USER - only nodes in current tasks mems allowed ok.
|
2006-05-20 16:00:10 -06:00
|
|
|
*
|
|
|
|
* Rule:
|
|
|
|
* Don't call cpuset_zone_allowed() if you can't sleep, unless you
|
|
|
|
* pass in the __GFP_HARDWALL flag set in gfp_flag, which disables
|
|
|
|
* the code that might scan up ancestor cpusets and sleep.
|
[PATCH] cpusets: formalize intermediate GFP_KERNEL containment
This patch makes use of the previously underutilized cpuset flag
'mem_exclusive' to provide what amounts to another layer of memory placement
resolution. With this patch, there are now the following four layers of
memory placement available:
1) The whole system (interrupt and GFP_ATOMIC allocations can use this),
2) The nearest enclosing mem_exclusive cpuset (GFP_KERNEL allocations can use),
3) The current tasks cpuset (GFP_USER allocations constrained to here), and
4) Specific node placement, using mbind and set_mempolicy.
These nest - each layer is a subset (same or within) of the previous.
Layer (2) above is new, with this patch. The call used to check whether a
zone (its node, actually) is in a cpuset (in its mems_allowed, actually) is
extended to take a gfp_mask argument, and its logic is extended, in the case
that __GFP_HARDWALL is not set in the flag bits, to look up the cpuset
hierarchy for the nearest enclosing mem_exclusive cpuset, to determine if
placement is allowed. The definition of GFP_USER, which used to be identical
to GFP_KERNEL, is changed to also set the __GFP_HARDWALL bit, in the previous
cpuset_gfp_hardwall_flag patch.
GFP_ATOMIC and GFP_KERNEL allocations will stay within the current tasks
cpuset, so long as any node therein is not too tight on memory, but will
escape to the larger layer, if need be.
The intended use is to allow something like a batch manager to handle several
jobs, each job in its own cpuset, but using common kernel memory for caches
and such. Swapper and oom_kill activity is also constrained to Layer (2). A
task in or below one mem_exclusive cpuset should not cause swapping on nodes
in another non-overlapping mem_exclusive cpuset, nor provoke oom_killing of a
task in another such cpuset. Heavy use of kernel memory for i/o caching and
such by one job should not impact the memory available to jobs in other
non-overlapping mem_exclusive cpusets.
This patch enables providing hardwall, inescapable cpusets for memory
allocations of each job, while sharing kernel memory allocations between
several jobs, in an enclosing mem_exclusive cpuset.
Like Dinakar's patch earlier to enable administering sched domains using the
cpu_exclusive flag, this patch also provides a useful meaning to a cpuset flag
that had previously done nothing much useful other than restrict what cpuset
configurations were allowed.
Signed-off-by: Paul Jackson <pj@sgi.com>
Signed-off-by: Andrew Morton <akpm@osdl.org>
Signed-off-by: Linus Torvalds <torvalds@osdl.org>
2005-09-06 16:18:12 -06:00
|
|
|
**/
|
|
|
|
|
2006-01-08 02:01:57 -07:00
|
|
|
int __cpuset_zone_allowed(struct zone *z, gfp_t gfp_mask)
|
2005-04-16 16:20:36 -06:00
|
|
|
{
|
[PATCH] cpusets: formalize intermediate GFP_KERNEL containment
This patch makes use of the previously underutilized cpuset flag
'mem_exclusive' to provide what amounts to another layer of memory placement
resolution. With this patch, there are now the following four layers of
memory placement available:
1) The whole system (interrupt and GFP_ATOMIC allocations can use this),
2) The nearest enclosing mem_exclusive cpuset (GFP_KERNEL allocations can use),
3) The current tasks cpuset (GFP_USER allocations constrained to here), and
4) Specific node placement, using mbind and set_mempolicy.
These nest - each layer is a subset (same or within) of the previous.
Layer (2) above is new, with this patch. The call used to check whether a
zone (its node, actually) is in a cpuset (in its mems_allowed, actually) is
extended to take a gfp_mask argument, and its logic is extended, in the case
that __GFP_HARDWALL is not set in the flag bits, to look up the cpuset
hierarchy for the nearest enclosing mem_exclusive cpuset, to determine if
placement is allowed. The definition of GFP_USER, which used to be identical
to GFP_KERNEL, is changed to also set the __GFP_HARDWALL bit, in the previous
cpuset_gfp_hardwall_flag patch.
GFP_ATOMIC and GFP_KERNEL allocations will stay within the current tasks
cpuset, so long as any node therein is not too tight on memory, but will
escape to the larger layer, if need be.
The intended use is to allow something like a batch manager to handle several
jobs, each job in its own cpuset, but using common kernel memory for caches
and such. Swapper and oom_kill activity is also constrained to Layer (2). A
task in or below one mem_exclusive cpuset should not cause swapping on nodes
in another non-overlapping mem_exclusive cpuset, nor provoke oom_killing of a
task in another such cpuset. Heavy use of kernel memory for i/o caching and
such by one job should not impact the memory available to jobs in other
non-overlapping mem_exclusive cpusets.
This patch enables providing hardwall, inescapable cpusets for memory
allocations of each job, while sharing kernel memory allocations between
several jobs, in an enclosing mem_exclusive cpuset.
Like Dinakar's patch earlier to enable administering sched domains using the
cpu_exclusive flag, this patch also provides a useful meaning to a cpuset flag
that had previously done nothing much useful other than restrict what cpuset
configurations were allowed.
Signed-off-by: Paul Jackson <pj@sgi.com>
Signed-off-by: Andrew Morton <akpm@osdl.org>
Signed-off-by: Linus Torvalds <torvalds@osdl.org>
2005-09-06 16:18:12 -06:00
|
|
|
int node; /* node that zone z is on */
|
|
|
|
const struct cpuset *cs; /* current cpuset ancestors */
|
2006-03-24 04:16:12 -07:00
|
|
|
int allowed; /* is allocation in zone z allowed? */
|
[PATCH] cpusets: formalize intermediate GFP_KERNEL containment
This patch makes use of the previously underutilized cpuset flag
'mem_exclusive' to provide what amounts to another layer of memory placement
resolution. With this patch, there are now the following four layers of
memory placement available:
1) The whole system (interrupt and GFP_ATOMIC allocations can use this),
2) The nearest enclosing mem_exclusive cpuset (GFP_KERNEL allocations can use),
3) The current tasks cpuset (GFP_USER allocations constrained to here), and
4) Specific node placement, using mbind and set_mempolicy.
These nest - each layer is a subset (same or within) of the previous.
Layer (2) above is new, with this patch. The call used to check whether a
zone (its node, actually) is in a cpuset (in its mems_allowed, actually) is
extended to take a gfp_mask argument, and its logic is extended, in the case
that __GFP_HARDWALL is not set in the flag bits, to look up the cpuset
hierarchy for the nearest enclosing mem_exclusive cpuset, to determine if
placement is allowed. The definition of GFP_USER, which used to be identical
to GFP_KERNEL, is changed to also set the __GFP_HARDWALL bit, in the previous
cpuset_gfp_hardwall_flag patch.
GFP_ATOMIC and GFP_KERNEL allocations will stay within the current tasks
cpuset, so long as any node therein is not too tight on memory, but will
escape to the larger layer, if need be.
The intended use is to allow something like a batch manager to handle several
jobs, each job in its own cpuset, but using common kernel memory for caches
and such. Swapper and oom_kill activity is also constrained to Layer (2). A
task in or below one mem_exclusive cpuset should not cause swapping on nodes
in another non-overlapping mem_exclusive cpuset, nor provoke oom_killing of a
task in another such cpuset. Heavy use of kernel memory for i/o caching and
such by one job should not impact the memory available to jobs in other
non-overlapping mem_exclusive cpusets.
This patch enables providing hardwall, inescapable cpusets for memory
allocations of each job, while sharing kernel memory allocations between
several jobs, in an enclosing mem_exclusive cpuset.
Like Dinakar's patch earlier to enable administering sched domains using the
cpu_exclusive flag, this patch also provides a useful meaning to a cpuset flag
that had previously done nothing much useful other than restrict what cpuset
configurations were allowed.
Signed-off-by: Paul Jackson <pj@sgi.com>
Signed-off-by: Andrew Morton <akpm@osdl.org>
Signed-off-by: Linus Torvalds <torvalds@osdl.org>
2005-09-06 16:18:12 -06:00
|
|
|
|
2006-09-26 00:31:40 -06:00
|
|
|
if (in_interrupt() || (gfp_mask & __GFP_THISNODE))
|
[PATCH] cpusets: formalize intermediate GFP_KERNEL containment
This patch makes use of the previously underutilized cpuset flag
'mem_exclusive' to provide what amounts to another layer of memory placement
resolution. With this patch, there are now the following four layers of
memory placement available:
1) The whole system (interrupt and GFP_ATOMIC allocations can use this),
2) The nearest enclosing mem_exclusive cpuset (GFP_KERNEL allocations can use),
3) The current tasks cpuset (GFP_USER allocations constrained to here), and
4) Specific node placement, using mbind and set_mempolicy.
These nest - each layer is a subset (same or within) of the previous.
Layer (2) above is new, with this patch. The call used to check whether a
zone (its node, actually) is in a cpuset (in its mems_allowed, actually) is
extended to take a gfp_mask argument, and its logic is extended, in the case
that __GFP_HARDWALL is not set in the flag bits, to look up the cpuset
hierarchy for the nearest enclosing mem_exclusive cpuset, to determine if
placement is allowed. The definition of GFP_USER, which used to be identical
to GFP_KERNEL, is changed to also set the __GFP_HARDWALL bit, in the previous
cpuset_gfp_hardwall_flag patch.
GFP_ATOMIC and GFP_KERNEL allocations will stay within the current tasks
cpuset, so long as any node therein is not too tight on memory, but will
escape to the larger layer, if need be.
The intended use is to allow something like a batch manager to handle several
jobs, each job in its own cpuset, but using common kernel memory for caches
and such. Swapper and oom_kill activity is also constrained to Layer (2). A
task in or below one mem_exclusive cpuset should not cause swapping on nodes
in another non-overlapping mem_exclusive cpuset, nor provoke oom_killing of a
task in another such cpuset. Heavy use of kernel memory for i/o caching and
such by one job should not impact the memory available to jobs in other
non-overlapping mem_exclusive cpusets.
This patch enables providing hardwall, inescapable cpusets for memory
allocations of each job, while sharing kernel memory allocations between
several jobs, in an enclosing mem_exclusive cpuset.
Like Dinakar's patch earlier to enable administering sched domains using the
cpu_exclusive flag, this patch also provides a useful meaning to a cpuset flag
that had previously done nothing much useful other than restrict what cpuset
configurations were allowed.
Signed-off-by: Paul Jackson <pj@sgi.com>
Signed-off-by: Andrew Morton <akpm@osdl.org>
Signed-off-by: Linus Torvalds <torvalds@osdl.org>
2005-09-06 16:18:12 -06:00
|
|
|
return 1;
|
2006-09-26 00:31:55 -06:00
|
|
|
node = zone_to_nid(z);
|
2006-05-20 16:00:11 -06:00
|
|
|
might_sleep_if(!(gfp_mask & __GFP_HARDWALL));
|
[PATCH] cpusets: formalize intermediate GFP_KERNEL containment
This patch makes use of the previously underutilized cpuset flag
'mem_exclusive' to provide what amounts to another layer of memory placement
resolution. With this patch, there are now the following four layers of
memory placement available:
1) The whole system (interrupt and GFP_ATOMIC allocations can use this),
2) The nearest enclosing mem_exclusive cpuset (GFP_KERNEL allocations can use),
3) The current tasks cpuset (GFP_USER allocations constrained to here), and
4) Specific node placement, using mbind and set_mempolicy.
These nest - each layer is a subset (same or within) of the previous.
Layer (2) above is new, with this patch. The call used to check whether a
zone (its node, actually) is in a cpuset (in its mems_allowed, actually) is
extended to take a gfp_mask argument, and its logic is extended, in the case
that __GFP_HARDWALL is not set in the flag bits, to look up the cpuset
hierarchy for the nearest enclosing mem_exclusive cpuset, to determine if
placement is allowed. The definition of GFP_USER, which used to be identical
to GFP_KERNEL, is changed to also set the __GFP_HARDWALL bit, in the previous
cpuset_gfp_hardwall_flag patch.
GFP_ATOMIC and GFP_KERNEL allocations will stay within the current tasks
cpuset, so long as any node therein is not too tight on memory, but will
escape to the larger layer, if need be.
The intended use is to allow something like a batch manager to handle several
jobs, each job in its own cpuset, but using common kernel memory for caches
and such. Swapper and oom_kill activity is also constrained to Layer (2). A
task in or below one mem_exclusive cpuset should not cause swapping on nodes
in another non-overlapping mem_exclusive cpuset, nor provoke oom_killing of a
task in another such cpuset. Heavy use of kernel memory for i/o caching and
such by one job should not impact the memory available to jobs in other
non-overlapping mem_exclusive cpusets.
This patch enables providing hardwall, inescapable cpusets for memory
allocations of each job, while sharing kernel memory allocations between
several jobs, in an enclosing mem_exclusive cpuset.
Like Dinakar's patch earlier to enable administering sched domains using the
cpu_exclusive flag, this patch also provides a useful meaning to a cpuset flag
that had previously done nothing much useful other than restrict what cpuset
configurations were allowed.
Signed-off-by: Paul Jackson <pj@sgi.com>
Signed-off-by: Andrew Morton <akpm@osdl.org>
Signed-off-by: Linus Torvalds <torvalds@osdl.org>
2005-09-06 16:18:12 -06:00
|
|
|
if (node_isset(node, current->mems_allowed))
|
|
|
|
return 1;
|
|
|
|
if (gfp_mask & __GFP_HARDWALL) /* If hardwall request, stop here */
|
|
|
|
return 0;
|
|
|
|
|
2005-11-13 17:06:35 -07:00
|
|
|
if (current->flags & PF_EXITING) /* Let dying task have memory */
|
|
|
|
return 1;
|
|
|
|
|
[PATCH] cpusets: formalize intermediate GFP_KERNEL containment
This patch makes use of the previously underutilized cpuset flag
'mem_exclusive' to provide what amounts to another layer of memory placement
resolution. With this patch, there are now the following four layers of
memory placement available:
1) The whole system (interrupt and GFP_ATOMIC allocations can use this),
2) The nearest enclosing mem_exclusive cpuset (GFP_KERNEL allocations can use),
3) The current tasks cpuset (GFP_USER allocations constrained to here), and
4) Specific node placement, using mbind and set_mempolicy.
These nest - each layer is a subset (same or within) of the previous.
Layer (2) above is new, with this patch. The call used to check whether a
zone (its node, actually) is in a cpuset (in its mems_allowed, actually) is
extended to take a gfp_mask argument, and its logic is extended, in the case
that __GFP_HARDWALL is not set in the flag bits, to look up the cpuset
hierarchy for the nearest enclosing mem_exclusive cpuset, to determine if
placement is allowed. The definition of GFP_USER, which used to be identical
to GFP_KERNEL, is changed to also set the __GFP_HARDWALL bit, in the previous
cpuset_gfp_hardwall_flag patch.
GFP_ATOMIC and GFP_KERNEL allocations will stay within the current tasks
cpuset, so long as any node therein is not too tight on memory, but will
escape to the larger layer, if need be.
The intended use is to allow something like a batch manager to handle several
jobs, each job in its own cpuset, but using common kernel memory for caches
and such. Swapper and oom_kill activity is also constrained to Layer (2). A
task in or below one mem_exclusive cpuset should not cause swapping on nodes
in another non-overlapping mem_exclusive cpuset, nor provoke oom_killing of a
task in another such cpuset. Heavy use of kernel memory for i/o caching and
such by one job should not impact the memory available to jobs in other
non-overlapping mem_exclusive cpusets.
This patch enables providing hardwall, inescapable cpusets for memory
allocations of each job, while sharing kernel memory allocations between
several jobs, in an enclosing mem_exclusive cpuset.
Like Dinakar's patch earlier to enable administering sched domains using the
cpu_exclusive flag, this patch also provides a useful meaning to a cpuset flag
that had previously done nothing much useful other than restrict what cpuset
configurations were allowed.
Signed-off-by: Paul Jackson <pj@sgi.com>
Signed-off-by: Andrew Morton <akpm@osdl.org>
Signed-off-by: Linus Torvalds <torvalds@osdl.org>
2005-09-06 16:18:12 -06:00
|
|
|
/* Not hardwall and node outside mems_allowed: scan up cpusets */
|
2006-03-23 04:00:18 -07:00
|
|
|
mutex_lock(&callback_mutex);
|
[PATCH] cpusets: dual semaphore locking overhaul
Overhaul cpuset locking. Replace single semaphore with two semaphores.
The suggestion to use two locks was made by Roman Zippel.
Both locks are global. Code that wants to modify cpusets must first
acquire the exclusive manage_sem, which allows them read-only access to
cpusets, and holds off other would-be modifiers. Before making actual
changes, the second semaphore, callback_sem must be acquired as well. Code
that needs only to query cpusets must acquire callback_sem, which is also a
global exclusive lock.
The earlier problems with double tripping are avoided, because it is
allowed for holders of manage_sem to nest the second callback_sem lock, and
only callback_sem is needed by code called from within __alloc_pages(),
where the double tripping had been possible.
This is not quite the same as a normal read/write semaphore, because
obtaining read-only access with intent to change must hold off other such
attempts, while allowing read-only access w/o such intention. Changing
cpusets involves several related checks and changes, which must be done
while allowing read-only queries (to avoid the double trip), but while
ensuring nothing changes (holding off other would be modifiers.)
This overhaul of cpuset locking also makes careful use of task_lock() to
guard access to the task->cpuset pointer, closing a couple of race
conditions noticed while reading this code (thanks, Roman). I've never
seen these races fail in any use or test.
See further the comments in the code.
Signed-off-by: Paul Jackson <pj@sgi.com>
Signed-off-by: Andrew Morton <akpm@osdl.org>
Signed-off-by: Linus Torvalds <torvalds@osdl.org>
2005-10-30 16:02:30 -07:00
|
|
|
|
|
|
|
task_lock(current);
|
|
|
|
cs = nearest_exclusive_ancestor(current->cpuset);
|
|
|
|
task_unlock(current);
|
|
|
|
|
[PATCH] cpusets: formalize intermediate GFP_KERNEL containment
This patch makes use of the previously underutilized cpuset flag
'mem_exclusive' to provide what amounts to another layer of memory placement
resolution. With this patch, there are now the following four layers of
memory placement available:
1) The whole system (interrupt and GFP_ATOMIC allocations can use this),
2) The nearest enclosing mem_exclusive cpuset (GFP_KERNEL allocations can use),
3) The current tasks cpuset (GFP_USER allocations constrained to here), and
4) Specific node placement, using mbind and set_mempolicy.
These nest - each layer is a subset (same or within) of the previous.
Layer (2) above is new, with this patch. The call used to check whether a
zone (its node, actually) is in a cpuset (in its mems_allowed, actually) is
extended to take a gfp_mask argument, and its logic is extended, in the case
that __GFP_HARDWALL is not set in the flag bits, to look up the cpuset
hierarchy for the nearest enclosing mem_exclusive cpuset, to determine if
placement is allowed. The definition of GFP_USER, which used to be identical
to GFP_KERNEL, is changed to also set the __GFP_HARDWALL bit, in the previous
cpuset_gfp_hardwall_flag patch.
GFP_ATOMIC and GFP_KERNEL allocations will stay within the current tasks
cpuset, so long as any node therein is not too tight on memory, but will
escape to the larger layer, if need be.
The intended use is to allow something like a batch manager to handle several
jobs, each job in its own cpuset, but using common kernel memory for caches
and such. Swapper and oom_kill activity is also constrained to Layer (2). A
task in or below one mem_exclusive cpuset should not cause swapping on nodes
in another non-overlapping mem_exclusive cpuset, nor provoke oom_killing of a
task in another such cpuset. Heavy use of kernel memory for i/o caching and
such by one job should not impact the memory available to jobs in other
non-overlapping mem_exclusive cpusets.
This patch enables providing hardwall, inescapable cpusets for memory
allocations of each job, while sharing kernel memory allocations between
several jobs, in an enclosing mem_exclusive cpuset.
Like Dinakar's patch earlier to enable administering sched domains using the
cpu_exclusive flag, this patch also provides a useful meaning to a cpuset flag
that had previously done nothing much useful other than restrict what cpuset
configurations were allowed.
Signed-off-by: Paul Jackson <pj@sgi.com>
Signed-off-by: Andrew Morton <akpm@osdl.org>
Signed-off-by: Linus Torvalds <torvalds@osdl.org>
2005-09-06 16:18:12 -06:00
|
|
|
allowed = node_isset(node, cs->mems_allowed);
|
2006-03-23 04:00:18 -07:00
|
|
|
mutex_unlock(&callback_mutex);
|
[PATCH] cpusets: formalize intermediate GFP_KERNEL containment
This patch makes use of the previously underutilized cpuset flag
'mem_exclusive' to provide what amounts to another layer of memory placement
resolution. With this patch, there are now the following four layers of
memory placement available:
1) The whole system (interrupt and GFP_ATOMIC allocations can use this),
2) The nearest enclosing mem_exclusive cpuset (GFP_KERNEL allocations can use),
3) The current tasks cpuset (GFP_USER allocations constrained to here), and
4) Specific node placement, using mbind and set_mempolicy.
These nest - each layer is a subset (same or within) of the previous.
Layer (2) above is new, with this patch. The call used to check whether a
zone (its node, actually) is in a cpuset (in its mems_allowed, actually) is
extended to take a gfp_mask argument, and its logic is extended, in the case
that __GFP_HARDWALL is not set in the flag bits, to look up the cpuset
hierarchy for the nearest enclosing mem_exclusive cpuset, to determine if
placement is allowed. The definition of GFP_USER, which used to be identical
to GFP_KERNEL, is changed to also set the __GFP_HARDWALL bit, in the previous
cpuset_gfp_hardwall_flag patch.
GFP_ATOMIC and GFP_KERNEL allocations will stay within the current tasks
cpuset, so long as any node therein is not too tight on memory, but will
escape to the larger layer, if need be.
The intended use is to allow something like a batch manager to handle several
jobs, each job in its own cpuset, but using common kernel memory for caches
and such. Swapper and oom_kill activity is also constrained to Layer (2). A
task in or below one mem_exclusive cpuset should not cause swapping on nodes
in another non-overlapping mem_exclusive cpuset, nor provoke oom_killing of a
task in another such cpuset. Heavy use of kernel memory for i/o caching and
such by one job should not impact the memory available to jobs in other
non-overlapping mem_exclusive cpusets.
This patch enables providing hardwall, inescapable cpusets for memory
allocations of each job, while sharing kernel memory allocations between
several jobs, in an enclosing mem_exclusive cpuset.
Like Dinakar's patch earlier to enable administering sched domains using the
cpu_exclusive flag, this patch also provides a useful meaning to a cpuset flag
that had previously done nothing much useful other than restrict what cpuset
configurations were allowed.
Signed-off-by: Paul Jackson <pj@sgi.com>
Signed-off-by: Andrew Morton <akpm@osdl.org>
Signed-off-by: Linus Torvalds <torvalds@osdl.org>
2005-09-06 16:18:12 -06:00
|
|
|
return allowed;
|
2005-04-16 16:20:36 -06:00
|
|
|
}
|
|
|
|
|
2006-01-14 14:21:06 -07:00
|
|
|
/**
|
|
|
|
* cpuset_lock - lock out any changes to cpuset structures
|
|
|
|
*
|
2006-03-23 04:00:18 -07:00
|
|
|
* The out of memory (oom) code needs to mutex_lock cpusets
|
2006-01-14 14:21:06 -07:00
|
|
|
* from being changed while it scans the tasklist looking for a
|
2006-03-23 04:00:18 -07:00
|
|
|
* task in an overlapping cpuset. Expose callback_mutex via this
|
2006-01-14 14:21:06 -07:00
|
|
|
* cpuset_lock() routine, so the oom code can lock it, before
|
|
|
|
* locking the task list. The tasklist_lock is a spinlock, so
|
2006-03-23 04:00:18 -07:00
|
|
|
* must be taken inside callback_mutex.
|
2006-01-14 14:21:06 -07:00
|
|
|
*/
|
|
|
|
|
|
|
|
void cpuset_lock(void)
|
|
|
|
{
|
2006-03-23 04:00:18 -07:00
|
|
|
mutex_lock(&callback_mutex);
|
2006-01-14 14:21:06 -07:00
|
|
|
}
|
|
|
|
|
|
|
|
/**
|
|
|
|
* cpuset_unlock - release lock on cpuset changes
|
|
|
|
*
|
|
|
|
* Undo the lock taken in a previous cpuset_lock() call.
|
|
|
|
*/
|
|
|
|
|
|
|
|
void cpuset_unlock(void)
|
|
|
|
{
|
2006-03-23 04:00:18 -07:00
|
|
|
mutex_unlock(&callback_mutex);
|
2006-01-14 14:21:06 -07:00
|
|
|
}
|
|
|
|
|
[PATCH] cpuset memory spread basic implementation
This patch provides the implementation and cpuset interface for an alternative
memory allocation policy that can be applied to certain kinds of memory
allocations, such as the page cache (file system buffers) and some slab caches
(such as inode caches).
The policy is called "memory spreading." If enabled, it spreads out these
kinds of memory allocations over all the nodes allowed to a task, instead of
preferring to place them on the node where the task is executing.
All other kinds of allocations, including anonymous pages for a tasks stack
and data regions, are not affected by this policy choice, and continue to be
allocated preferring the node local to execution, as modified by the NUMA
mempolicy.
There are two boolean flag files per cpuset that control where the kernel
allocates pages for the file system buffers and related in kernel data
structures. They are called 'memory_spread_page' and 'memory_spread_slab'.
If the per-cpuset boolean flag file 'memory_spread_page' is set, then the
kernel will spread the file system buffers (page cache) evenly over all the
nodes that the faulting task is allowed to use, instead of preferring to put
those pages on the node where the task is running.
If the per-cpuset boolean flag file 'memory_spread_slab' is set, then the
kernel will spread some file system related slab caches, such as for inodes
and dentries evenly over all the nodes that the faulting task is allowed to
use, instead of preferring to put those pages on the node where the task is
running.
The implementation is simple. Setting the cpuset flags 'memory_spread_page'
or 'memory_spread_cache' turns on the per-process flags PF_SPREAD_PAGE or
PF_SPREAD_SLAB, respectively, for each task that is in the cpuset or
subsequently joins that cpuset. In subsequent patches, the page allocation
calls for the affected page cache and slab caches are modified to perform an
inline check for these flags, and if set, a call to a new routine
cpuset_mem_spread_node() returns the node to prefer for the allocation.
The cpuset_mem_spread_node() routine is also simple. It uses the value of a
per-task rotor cpuset_mem_spread_rotor to select the next node in the current
tasks mems_allowed to prefer for the allocation.
This policy can provide substantial improvements for jobs that need to place
thread local data on the corresponding node, but that need to access large
file system data sets that need to be spread across the several nodes in the
jobs cpuset in order to fit. Without this patch, especially for jobs that
might have one thread reading in the data set, the memory allocation across
the nodes in the jobs cpuset can become very uneven.
A couple of Copyright year ranges are updated as well. And a couple of email
addresses that can be found in the MAINTAINERS file are removed.
Signed-off-by: Paul Jackson <pj@sgi.com>
Signed-off-by: Andrew Morton <akpm@osdl.org>
Signed-off-by: Linus Torvalds <torvalds@osdl.org>
2006-03-24 04:16:03 -07:00
|
|
|
/**
|
|
|
|
* cpuset_mem_spread_node() - On which node to begin search for a page
|
|
|
|
*
|
|
|
|
* If a task is marked PF_SPREAD_PAGE or PF_SPREAD_SLAB (as for
|
|
|
|
* tasks in a cpuset with is_spread_page or is_spread_slab set),
|
|
|
|
* and if the memory allocation used cpuset_mem_spread_node()
|
|
|
|
* to determine on which node to start looking, as it will for
|
|
|
|
* certain page cache or slab cache pages such as used for file
|
|
|
|
* system buffers and inode caches, then instead of starting on the
|
|
|
|
* local node to look for a free page, rather spread the starting
|
|
|
|
* node around the tasks mems_allowed nodes.
|
|
|
|
*
|
|
|
|
* We don't have to worry about the returned node being offline
|
|
|
|
* because "it can't happen", and even if it did, it would be ok.
|
|
|
|
*
|
|
|
|
* The routines calling guarantee_online_mems() are careful to
|
|
|
|
* only set nodes in task->mems_allowed that are online. So it
|
|
|
|
* should not be possible for the following code to return an
|
|
|
|
* offline node. But if it did, that would be ok, as this routine
|
|
|
|
* is not returning the node where the allocation must be, only
|
|
|
|
* the node where the search should start. The zonelist passed to
|
|
|
|
* __alloc_pages() will include all nodes. If the slab allocator
|
|
|
|
* is passed an offline node, it will fall back to the local node.
|
|
|
|
* See kmem_cache_alloc_node().
|
|
|
|
*/
|
|
|
|
|
|
|
|
int cpuset_mem_spread_node(void)
|
|
|
|
{
|
|
|
|
int node;
|
|
|
|
|
|
|
|
node = next_node(current->cpuset_mem_spread_rotor, current->mems_allowed);
|
|
|
|
if (node == MAX_NUMNODES)
|
|
|
|
node = first_node(current->mems_allowed);
|
|
|
|
current->cpuset_mem_spread_rotor = node;
|
|
|
|
return node;
|
|
|
|
}
|
|
|
|
EXPORT_SYMBOL_GPL(cpuset_mem_spread_node);
|
|
|
|
|
2005-09-06 16:18:13 -06:00
|
|
|
/**
|
|
|
|
* cpuset_excl_nodes_overlap - Do we overlap @p's mem_exclusive ancestors?
|
|
|
|
* @p: pointer to task_struct of some other task.
|
|
|
|
*
|
|
|
|
* Description: Return true if the nearest mem_exclusive ancestor
|
|
|
|
* cpusets of tasks @p and current overlap. Used by oom killer to
|
|
|
|
* determine if task @p's memory usage might impact the memory
|
|
|
|
* available to the current task.
|
|
|
|
*
|
2006-03-23 04:00:18 -07:00
|
|
|
* Call while holding callback_mutex.
|
2005-09-06 16:18:13 -06:00
|
|
|
**/
|
|
|
|
|
|
|
|
int cpuset_excl_nodes_overlap(const struct task_struct *p)
|
|
|
|
{
|
|
|
|
const struct cpuset *cs1, *cs2; /* my and p's cpuset ancestors */
|
2006-08-27 02:23:54 -06:00
|
|
|
int overlap = 1; /* do cpusets overlap? */
|
2005-09-06 16:18:13 -06:00
|
|
|
|
[PATCH] cpusets: dual semaphore locking overhaul
Overhaul cpuset locking. Replace single semaphore with two semaphores.
The suggestion to use two locks was made by Roman Zippel.
Both locks are global. Code that wants to modify cpusets must first
acquire the exclusive manage_sem, which allows them read-only access to
cpusets, and holds off other would-be modifiers. Before making actual
changes, the second semaphore, callback_sem must be acquired as well. Code
that needs only to query cpusets must acquire callback_sem, which is also a
global exclusive lock.
The earlier problems with double tripping are avoided, because it is
allowed for holders of manage_sem to nest the second callback_sem lock, and
only callback_sem is needed by code called from within __alloc_pages(),
where the double tripping had been possible.
This is not quite the same as a normal read/write semaphore, because
obtaining read-only access with intent to change must hold off other such
attempts, while allowing read-only access w/o such intention. Changing
cpusets involves several related checks and changes, which must be done
while allowing read-only queries (to avoid the double trip), but while
ensuring nothing changes (holding off other would be modifiers.)
This overhaul of cpuset locking also makes careful use of task_lock() to
guard access to the task->cpuset pointer, closing a couple of race
conditions noticed while reading this code (thanks, Roman). I've never
seen these races fail in any use or test.
See further the comments in the code.
Signed-off-by: Paul Jackson <pj@sgi.com>
Signed-off-by: Andrew Morton <akpm@osdl.org>
Signed-off-by: Linus Torvalds <torvalds@osdl.org>
2005-10-30 16:02:30 -07:00
|
|
|
task_lock(current);
|
|
|
|
if (current->flags & PF_EXITING) {
|
|
|
|
task_unlock(current);
|
|
|
|
goto done;
|
|
|
|
}
|
|
|
|
cs1 = nearest_exclusive_ancestor(current->cpuset);
|
|
|
|
task_unlock(current);
|
|
|
|
|
|
|
|
task_lock((struct task_struct *)p);
|
|
|
|
if (p->flags & PF_EXITING) {
|
|
|
|
task_unlock((struct task_struct *)p);
|
|
|
|
goto done;
|
|
|
|
}
|
|
|
|
cs2 = nearest_exclusive_ancestor(p->cpuset);
|
|
|
|
task_unlock((struct task_struct *)p);
|
|
|
|
|
2005-09-06 16:18:13 -06:00
|
|
|
overlap = nodes_intersects(cs1->mems_allowed, cs2->mems_allowed);
|
|
|
|
done:
|
|
|
|
return overlap;
|
|
|
|
}
|
|
|
|
|
[PATCH] cpuset: memory pressure meter
Provide a simple per-cpuset metric of memory pressure, tracking the -rate-
that the tasks in a cpuset call try_to_free_pages(), the synchronous
(direct) memory reclaim code.
This enables batch managers monitoring jobs running in dedicated cpusets to
efficiently detect what level of memory pressure that job is causing.
This is useful both on tightly managed systems running a wide mix of
submitted jobs, which may choose to terminate or reprioritize jobs that are
trying to use more memory than allowed on the nodes assigned them, and with
tightly coupled, long running, massively parallel scientific computing jobs
that will dramatically fail to meet required performance goals if they
start to use more memory than allowed to them.
This patch just provides a very economical way for the batch manager to
monitor a cpuset for signs of memory pressure. It's up to the batch
manager or other user code to decide what to do about it and take action.
==> Unless this feature is enabled by writing "1" to the special file
/dev/cpuset/memory_pressure_enabled, the hook in the rebalance
code of __alloc_pages() for this metric reduces to simply noticing
that the cpuset_memory_pressure_enabled flag is zero. So only
systems that enable this feature will compute the metric.
Why a per-cpuset, running average:
Because this meter is per-cpuset, rather than per-task or mm, the
system load imposed by a batch scheduler monitoring this metric is
sharply reduced on large systems, because a scan of the tasklist can be
avoided on each set of queries.
Because this meter is a running average, instead of an accumulating
counter, a batch scheduler can detect memory pressure with a single
read, instead of having to read and accumulate results for a period of
time.
Because this meter is per-cpuset rather than per-task or mm, the
batch scheduler can obtain the key information, memory pressure in a
cpuset, with a single read, rather than having to query and accumulate
results over all the (dynamically changing) set of tasks in the cpuset.
A per-cpuset simple digital filter (requires a spinlock and 3 words of data
per-cpuset) is kept, and updated by any task attached to that cpuset, if it
enters the synchronous (direct) page reclaim code.
A per-cpuset file provides an integer number representing the recent
(half-life of 10 seconds) rate of direct page reclaims caused by the tasks
in the cpuset, in units of reclaims attempted per second, times 1000.
Signed-off-by: Paul Jackson <pj@sgi.com>
Signed-off-by: Andrew Morton <akpm@osdl.org>
Signed-off-by: Linus Torvalds <torvalds@osdl.org>
2006-01-08 02:01:49 -07:00
|
|
|
/*
|
|
|
|
* Collection of memory_pressure is suppressed unless
|
|
|
|
* this flag is enabled by writing "1" to the special
|
|
|
|
* cpuset file 'memory_pressure_enabled' in the root cpuset.
|
|
|
|
*/
|
|
|
|
|
2006-01-08 02:01:51 -07:00
|
|
|
int cpuset_memory_pressure_enabled __read_mostly;
|
[PATCH] cpuset: memory pressure meter
Provide a simple per-cpuset metric of memory pressure, tracking the -rate-
that the tasks in a cpuset call try_to_free_pages(), the synchronous
(direct) memory reclaim code.
This enables batch managers monitoring jobs running in dedicated cpusets to
efficiently detect what level of memory pressure that job is causing.
This is useful both on tightly managed systems running a wide mix of
submitted jobs, which may choose to terminate or reprioritize jobs that are
trying to use more memory than allowed on the nodes assigned them, and with
tightly coupled, long running, massively parallel scientific computing jobs
that will dramatically fail to meet required performance goals if they
start to use more memory than allowed to them.
This patch just provides a very economical way for the batch manager to
monitor a cpuset for signs of memory pressure. It's up to the batch
manager or other user code to decide what to do about it and take action.
==> Unless this feature is enabled by writing "1" to the special file
/dev/cpuset/memory_pressure_enabled, the hook in the rebalance
code of __alloc_pages() for this metric reduces to simply noticing
that the cpuset_memory_pressure_enabled flag is zero. So only
systems that enable this feature will compute the metric.
Why a per-cpuset, running average:
Because this meter is per-cpuset, rather than per-task or mm, the
system load imposed by a batch scheduler monitoring this metric is
sharply reduced on large systems, because a scan of the tasklist can be
avoided on each set of queries.
Because this meter is a running average, instead of an accumulating
counter, a batch scheduler can detect memory pressure with a single
read, instead of having to read and accumulate results for a period of
time.
Because this meter is per-cpuset rather than per-task or mm, the
batch scheduler can obtain the key information, memory pressure in a
cpuset, with a single read, rather than having to query and accumulate
results over all the (dynamically changing) set of tasks in the cpuset.
A per-cpuset simple digital filter (requires a spinlock and 3 words of data
per-cpuset) is kept, and updated by any task attached to that cpuset, if it
enters the synchronous (direct) page reclaim code.
A per-cpuset file provides an integer number representing the recent
(half-life of 10 seconds) rate of direct page reclaims caused by the tasks
in the cpuset, in units of reclaims attempted per second, times 1000.
Signed-off-by: Paul Jackson <pj@sgi.com>
Signed-off-by: Andrew Morton <akpm@osdl.org>
Signed-off-by: Linus Torvalds <torvalds@osdl.org>
2006-01-08 02:01:49 -07:00
|
|
|
|
|
|
|
/**
|
|
|
|
* cpuset_memory_pressure_bump - keep stats of per-cpuset reclaims.
|
|
|
|
*
|
|
|
|
* Keep a running average of the rate of synchronous (direct)
|
|
|
|
* page reclaim efforts initiated by tasks in each cpuset.
|
|
|
|
*
|
|
|
|
* This represents the rate at which some task in the cpuset
|
|
|
|
* ran low on memory on all nodes it was allowed to use, and
|
|
|
|
* had to enter the kernels page reclaim code in an effort to
|
|
|
|
* create more free memory by tossing clean pages or swapping
|
|
|
|
* or writing dirty pages.
|
|
|
|
*
|
|
|
|
* Display to user space in the per-cpuset read-only file
|
|
|
|
* "memory_pressure". Value displayed is an integer
|
|
|
|
* representing the recent rate of entry into the synchronous
|
|
|
|
* (direct) page reclaim by any task attached to the cpuset.
|
|
|
|
**/
|
|
|
|
|
|
|
|
void __cpuset_memory_pressure_bump(void)
|
|
|
|
{
|
|
|
|
struct cpuset *cs;
|
|
|
|
|
|
|
|
task_lock(current);
|
|
|
|
cs = current->cpuset;
|
|
|
|
fmeter_markevent(&cs->fmeter);
|
|
|
|
task_unlock(current);
|
|
|
|
}
|
|
|
|
|
2005-04-16 16:20:36 -06:00
|
|
|
/*
|
|
|
|
* proc_cpuset_show()
|
|
|
|
* - Print tasks cpuset path into seq_file.
|
|
|
|
* - Used for /proc/<pid>/cpuset.
|
[PATCH] cpusets: dual semaphore locking overhaul
Overhaul cpuset locking. Replace single semaphore with two semaphores.
The suggestion to use two locks was made by Roman Zippel.
Both locks are global. Code that wants to modify cpusets must first
acquire the exclusive manage_sem, which allows them read-only access to
cpusets, and holds off other would-be modifiers. Before making actual
changes, the second semaphore, callback_sem must be acquired as well. Code
that needs only to query cpusets must acquire callback_sem, which is also a
global exclusive lock.
The earlier problems with double tripping are avoided, because it is
allowed for holders of manage_sem to nest the second callback_sem lock, and
only callback_sem is needed by code called from within __alloc_pages(),
where the double tripping had been possible.
This is not quite the same as a normal read/write semaphore, because
obtaining read-only access with intent to change must hold off other such
attempts, while allowing read-only access w/o such intention. Changing
cpusets involves several related checks and changes, which must be done
while allowing read-only queries (to avoid the double trip), but while
ensuring nothing changes (holding off other would be modifiers.)
This overhaul of cpuset locking also makes careful use of task_lock() to
guard access to the task->cpuset pointer, closing a couple of race
conditions noticed while reading this code (thanks, Roman). I've never
seen these races fail in any use or test.
See further the comments in the code.
Signed-off-by: Paul Jackson <pj@sgi.com>
Signed-off-by: Andrew Morton <akpm@osdl.org>
Signed-off-by: Linus Torvalds <torvalds@osdl.org>
2005-10-30 16:02:30 -07:00
|
|
|
* - No need to task_lock(tsk) on this tsk->cpuset reference, as it
|
|
|
|
* doesn't really matter if tsk->cpuset changes after we read it,
|
2006-03-23 04:00:18 -07:00
|
|
|
* and we take manage_mutex, keeping attach_task() from changing it
|
2006-03-24 04:16:10 -07:00
|
|
|
* anyway. No need to check that tsk->cpuset != NULL, thanks to
|
|
|
|
* the_top_cpuset_hack in cpuset_exit(), which sets an exiting tasks
|
|
|
|
* cpuset to top_cpuset.
|
2005-04-16 16:20:36 -06:00
|
|
|
*/
|
|
|
|
static int proc_cpuset_show(struct seq_file *m, void *v)
|
|
|
|
{
|
2006-06-26 01:25:56 -06:00
|
|
|
struct pid *pid;
|
2005-04-16 16:20:36 -06:00
|
|
|
struct task_struct *tsk;
|
|
|
|
char *buf;
|
2006-06-26 01:25:55 -06:00
|
|
|
int retval;
|
2005-04-16 16:20:36 -06:00
|
|
|
|
2006-06-26 01:25:55 -06:00
|
|
|
retval = -ENOMEM;
|
2005-04-16 16:20:36 -06:00
|
|
|
buf = kmalloc(PAGE_SIZE, GFP_KERNEL);
|
|
|
|
if (!buf)
|
2006-06-26 01:25:55 -06:00
|
|
|
goto out;
|
|
|
|
|
|
|
|
retval = -ESRCH;
|
2006-06-26 01:25:56 -06:00
|
|
|
pid = m->private;
|
|
|
|
tsk = get_pid_task(pid, PIDTYPE_PID);
|
2006-06-26 01:25:55 -06:00
|
|
|
if (!tsk)
|
|
|
|
goto out_free;
|
2005-04-16 16:20:36 -06:00
|
|
|
|
2006-06-26 01:25:55 -06:00
|
|
|
retval = -EINVAL;
|
2006-03-23 04:00:18 -07:00
|
|
|
mutex_lock(&manage_mutex);
|
2006-06-26 01:25:55 -06:00
|
|
|
|
2006-03-24 04:16:10 -07:00
|
|
|
retval = cpuset_path(tsk->cpuset, buf, PAGE_SIZE);
|
2005-04-16 16:20:36 -06:00
|
|
|
if (retval < 0)
|
2006-06-26 01:25:55 -06:00
|
|
|
goto out_unlock;
|
2005-04-16 16:20:36 -06:00
|
|
|
seq_puts(m, buf);
|
|
|
|
seq_putc(m, '\n');
|
2006-06-26 01:25:55 -06:00
|
|
|
out_unlock:
|
2006-03-23 04:00:18 -07:00
|
|
|
mutex_unlock(&manage_mutex);
|
2006-06-26 01:25:55 -06:00
|
|
|
put_task_struct(tsk);
|
|
|
|
out_free:
|
2005-04-16 16:20:36 -06:00
|
|
|
kfree(buf);
|
2006-06-26 01:25:55 -06:00
|
|
|
out:
|
2005-04-16 16:20:36 -06:00
|
|
|
return retval;
|
|
|
|
}
|
|
|
|
|
|
|
|
static int cpuset_open(struct inode *inode, struct file *file)
|
|
|
|
{
|
2006-06-26 01:25:56 -06:00
|
|
|
struct pid *pid = PROC_I(inode)->pid;
|
|
|
|
return single_open(file, proc_cpuset_show, pid);
|
2005-04-16 16:20:36 -06:00
|
|
|
}
|
|
|
|
|
|
|
|
struct file_operations proc_cpuset_operations = {
|
|
|
|
.open = cpuset_open,
|
|
|
|
.read = seq_read,
|
|
|
|
.llseek = seq_lseek,
|
|
|
|
.release = single_release,
|
|
|
|
};
|
|
|
|
|
|
|
|
/* Display task cpus_allowed, mems_allowed in /proc/<pid>/status file. */
|
|
|
|
char *cpuset_task_status_allowed(struct task_struct *task, char *buffer)
|
|
|
|
{
|
|
|
|
buffer += sprintf(buffer, "Cpus_allowed:\t");
|
|
|
|
buffer += cpumask_scnprintf(buffer, PAGE_SIZE, task->cpus_allowed);
|
|
|
|
buffer += sprintf(buffer, "\n");
|
|
|
|
buffer += sprintf(buffer, "Mems_allowed:\t");
|
|
|
|
buffer += nodemask_scnprintf(buffer, PAGE_SIZE, task->mems_allowed);
|
|
|
|
buffer += sprintf(buffer, "\n");
|
|
|
|
return buffer;
|
|
|
|
}
|