This is the start of the `swap migration' patch series.
Swap migration allows the moving of the physical location of pages between
nodes in a numa system while the process is running. This means that the
virtual addresses that the process sees do not change. However, the system
rearranges the physical location of those pages.
The main intent of page migration patches here is to reduce the latency of
memory access by moving pages near to the processor where the process
accessing that memory is running.
The patchset allows a process to manually relocate the node on which its
pages are located through the MF_MOVE and MF_MOVE_ALL options while
setting a new memory policy.
The pages of process can also be relocated from another process using the
sys_migrate_pages() function call. Requires CAP_SYS_ADMIN. The migrate_pages
function call takes two sets of nodes and moves pages of a process that are
located on the from nodes to the destination nodes.
Manual migration is very useful if for example the scheduler has relocated a
process to a processor on a distant node. A batch scheduler or an
administrator can detect the situation and move the pages of the process
nearer to the new processor.
sys_migrate_pages() could be used on non-numa machines as well, to force all
of a particualr process's pages out to swap, if someone thinks that's useful.
Larger installations usually partition the system using cpusets into sections
of nodes. Paul has equipped cpusets with the ability to move pages when a
task is moved to another cpuset. This allows automatic control over locality
of a process. If a task is moved to a new cpuset then also all its pages are
moved with it so that the performance of the process does not sink
dramatically (as is the case today).
Swap migration works by simply evicting the page. The pages must be faulted
back in. The pages are then typically reallocated by the system near the node
where the process is executing.
For swap migration the destination of the move is controlled by the allocation
policy. Cpusets set the allocation policy before calling sys_migrate_pages()
in order to move the pages as intended.
No allocation policy changes are performed for sys_migrate_pages(). This
means that the pages may not faulted in to the specified nodes if no
allocation policy was set by other means. The pages will just end up near the
node where the fault occurred.
There's another patch series in the pipeline which implements "direct
migration".
The direct migration patchset extends the migration functionality to avoid
going through swap. The destination node of the relation is controllable
during the actual moving of pages. The crutch of using the allocation policy
to relocate is not necessary and the pages are moved directly to the target.
Its also faster since swap is not used.
And sys_migrate_pages() can then move pages directly to the specified node.
Implement functions to isolate pages from the LRU and put them back later.
This patch:
An earlier implementation was provided by Hirokazu Takahashi
<taka@valinux.co.jp> and IWAMOTO Toshihiro <iwamoto@valinux.co.jp> for the
memory hotplug project.
From: Magnus
This breaks out isolate_lru_page() and putpack_lru_page(). Needed for swap
migration.
Signed-off-by: Magnus Damm <magnus.damm@gmail.com>
Signed-off-by: Christoph Lameter <clameter@sgi.com>
Signed-off-by: Andrew Morton <akpm@osdl.org>
Signed-off-by: Linus Torvalds <torvalds@osdl.org>
Add /proc/sys/vm/drop_caches. When written to, this will cause the kernel to
discard as much pagecache and/or reclaimable slab objects as it can. THis
operation requires root permissions.
It won't drop dirty data, so the user should run `sync' first.
Caveats:
a) Holds inode_lock for exorbitant amounts of time.
b) Needs to be taught about NUMA nodes: propagate these all the way through
so the discarding can be controlled on a per-node basis.
This is a debugging feature: useful for getting consistent results between
filesystem benchmarks. We could possibly put it under a config option, but
it's less than 300 bytes.
Signed-off-by: Andrew Morton <akpm@osdl.org>
Signed-off-by: Linus Torvalds <torvalds@osdl.org>
Optimise page_state manipulations by introducing interrupt unsafe accessors
to page_state fields. Callers must provide their own locking (either
disable interrupts or not update from interrupt context).
Switch over the hot callsites that can easily be moved under interrupts off
sections.
Signed-off-by: Nick Piggin <npiggin@suse.de>
Cc: Hugh Dickins <hugh@veritas.com>
Signed-off-by: Andrew Morton <akpm@osdl.org>
Signed-off-by: Linus Torvalds <torvalds@osdl.org>
There are numerous places we check whether a zone is populated or not.
Provide a helper function to check for populated zones and convert all
checks for zone->present_pages.
Signed-off-by: Con Kolivas <kernel@kolivas.org>
Signed-off-by: Andrew Morton <akpm@osdl.org>
Signed-off-by: Linus Torvalds <torvalds@osdl.org>
Revert a patch which went into 2.6.8-rc1. The changelog for that patch was:
The shrink_zone() logic can, under some circumstances, cause far too many
pages to be reclaimed. Say, we're scanning at high priority and suddenly
hit a large number of reclaimable pages on the LRU.
Change things so we bale out when SWAP_CLUSTER_MAX pages have been
reclaimed.
Problem is, this change caused significant imbalance in inter-zone scan
balancing by truncating scans of larger zones.
Suppose, for example, ZONE_HIGHMEM is 10x the size of ZONE_NORMAL. The zone
balancing algorithm would require that if we're scanning 100 pages of
ZONE_HIGHMEM, we should scan 10 pages of ZONE_NORMAL. But this logic will
cause the scanning of ZONE_HIGHMEM to bale out after only 32 pages are
reclaimed. Thus effectively causing smaller zones to be scanned relatively
harder than large ones.
Now I need to remember what the workload was which caused me to write this
patch originally, then fix it up in a different way...
Signed-off-by: Andrew Morton <akpm@osdl.org>
Signed-off-by: Linus Torvalds <torvalds@osdl.org>
Remove the last bits of Martin's ill-fated sys_set_zone_reclaim().
Cc: Martin Hicks <mort@wildopensource.com>
Signed-off-by: Andrew Morton <akpm@osdl.org>
Signed-off-by: Linus Torvalds <torvalds@osdl.org>
readpage(), prepare_write(), and commit_write() callers are updated to
understand the special return code AOP_TRUNCATED_PAGE in the style of
writepage() and WRITEPAGE_ACTIVATE. AOP_TRUNCATED_PAGE tells the caller that
the callee has unlocked the page and that the operation should be tried again
with a new page. OCFS2 uses this to detect and work around a lock inversion in
its aop methods. There should be no change in behaviour for methods that don't
return AOP_TRUNCATED_PAGE.
WRITEPAGE_ACTIVATE is also prepended with AOP_ for consistency and they are
made enums so that kerneldoc can be used to document their semantics.
Signed-off-by: Zach Brown <zach.brown@oracle.com>
With Andrew Morton <akpm@osdl.org>
The slab scanning code tries to balance the scanning rate of slabs versus the
scanning rate of LRU pages. To do this, it retains state concerning how many
slabs have been scanned - if a particular slab shrinker didn't scan enough
objects, we remember that for next time, and scan more objects on the next
pass.
The problem with this is that with (say) a huge number of GFP_NOIO
direct-reclaim attempts, the number of objects which are to be scanned when we
finally get a GFP_KERNEL request can be huge. Because some shrinker handlers
just bail out if !__GFP_FS.
So the patch clamps the number of objects-to-be-scanned to 2* the total number
of objects in the slab cache.
Signed-off-by: Andrea Arcangeli <andrea@suse.de>
Signed-off-by: Andrew Morton <akpm@osdl.org>
Signed-off-by: Linus Torvalds <torvalds@osdl.org>
Some users (hi Zwane) have seen a problem when running a workload that
eats nearly all of physical memory - th system does an OOM kill, even
when there is still a lot of swap free.
The problem appears to be a very big task that is holding the swap
token, and the VM has a very hard time finding any other page in the
system that is swappable.
Instead of ignoring the swap token when sc->priority reaches 0, we could
simply take the swap token away from the memory hog and make sure we
don't give it back to the memory hog for a few seconds.
This patch resolves the problem Zwane ran into.
Signed-off-by: Rik van Riel <riel@redhat.com>
Signed-off-by: Andrew Morton <akpm@osdl.org>
Signed-off-by: Linus Torvalds <torvalds@osdl.org>
Clean up of __alloc_pages.
Restoration of previous behaviour, plus further cleanups by introducing an
'alloc_flags', removing the last of should_reclaim_zone.
Signed-off-by: Rohit Seth <rohit.seth@intel.com>
Signed-off-by: Nick Piggin <npiggin@suse.de>
Signed-off-by: Andrew Morton <akpm@osdl.org>
Signed-off-by: Linus Torvalds <torvalds@osdl.org>
Christoph Lameter demonstrated very poor scalability on the SGI 512-way, with
a many-threaded application which concurrently initializes different parts of
a large anonymous area.
This patch corrects that, by using a separate spinlock per page table page, to
guard the page table entries in that page, instead of using the mm's single
page_table_lock. (But even then, page_table_lock is still used to guard page
table allocation, and anon_vma allocation.)
In this implementation, the spinlock is tucked inside the struct page of the
page table page: with a BUILD_BUG_ON in case it overflows - which it would in
the case of 32-bit PA-RISC with spinlock debugging enabled.
Splitting the lock is not quite for free: another cacheline access. Ideally,
I suppose we would use split ptlock only for multi-threaded processes on
multi-cpu machines; but deciding that dynamically would have its own costs.
So for now enable it by config, at some number of cpus - since the Kconfig
language doesn't support inequalities, let preprocessor compare that with
NR_CPUS. But I don't think it's worth being user-configurable: for good
testing of both split and unsplit configs, split now at 4 cpus, and perhaps
change that to 8 later.
There is a benefit even for singly threaded processes: kswapd can be attacking
one part of the mm while another part is busy faulting.
Signed-off-by: Hugh Dickins <hugh@veritas.com>
Signed-off-by: Andrew Morton <akpm@osdl.org>
Signed-off-by: Linus Torvalds <torvalds@osdl.org>
Martin Hicks' page cache reclaim patch added the 'may_swap' flag to the
scan_control struct; and modified shrink_list() not to add anon pages to
the swap cache if may_swap is not asserted.
Ref: http://marc.theaimsgroup.com/?l=linux-mm&m=111461480725322&w=4
However, further down, if the page is mapped, shrink_list() calls
try_to_unmap() which will call try_to_unmap_one() via try_to_unmap_anon ().
try_to_unmap_one() will BUG_ON() an anon page that is NOT in the swap
cache. Martin says he never encountered this path in his testing, but
agrees that it might happen.
This patch modifies shrink_list() to skip anon pages that are not already
in the swap cache when !may_swap, rather than just not adding them to the
cache.
Signed-off-by: Lee Schermerhorn <lee.schermerhorn@hp.com>
Cc: Hugh Dickins <hugh@veritas.com>
Signed-off-by: Andrew Morton <akpm@osdl.org>
Signed-off-by: Linus Torvalds <torvalds@osdl.org>
As noticed by Nick Piggin, we need to make sure that we check the page
count before we check for PageDirty, since the dirty check is only valid
if the count implies that we're the only possible ones holding the page.
We always did do this, but the code needs a read-memory-barrier to make
sure that the orderign is also honored by the CPU.
(The writer side is ordered due to the atomic decrement and test on the
page count, see the discussion on linux-kernel)
Signed-off-by: Linus Torvalds <torvalds@osdl.org>
Use the pgdat pointer we've already defined in wakeup_kswapd
Signed-off-by: Con Kolivas <kernel@kolivas.org>
Signed-off-by: Andrew Morton <akpm@osdl.org>
Signed-off-by: Linus Torvalds <torvalds@osdl.org>
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>
Christoph Lameter and Marcelo Tosatti asked to get rid of the
atomic_inc_and_test() to cleanup the atomic ops in the zone reclaim code.
Signed-off-by: Martin Hicks <mort@sgi.com>
Signed-off-by: Christoph Lameter <clameter@sgi.com>
Signed-off-by: Andrew Morton <akpm@osdl.org>
Signed-off-by: Linus Torvalds <torvalds@osdl.org>
Add a capability check to sys_set_zone_reclaim(). This syscall is not
something that should be available to a user.
Signed-off-by: Martin Hicks <mort@sgi.com>
Signed-off-by: Andrew Morton <akpm@osdl.org>
Signed-off-by: Linus Torvalds <torvalds@osdl.org>
1. Establish a simple API for process freezing defined in linux/include/sched.h:
frozen(process) Check for frozen process
freezing(process) Check if a process is being frozen
freeze(process) Tell a process to freeze (go to refrigerator)
thaw_process(process) Restart process
frozen_process(process) Process is frozen now
2. Remove all references to PF_FREEZE and PF_FROZEN from all
kernel sources except sched.h
3. Fix numerous locations where try_to_freeze is manually done by a driver
4. Remove the argument that is no longer necessary from two function calls.
5. Some whitespace cleanup
6. Clear potential race in refrigerator (provides an open window of PF_FREEZE
cleared before setting PF_FROZEN, recalc_sigpending does not check
PF_FROZEN).
This patch does not address the problem of freeze_processes() violating the rule
that a task may only modify its own flags by setting PF_FREEZE. This is not clean
in an SMP environment. freeze(process) is therefore not SMP safe!
Signed-off-by: Christoph Lameter <christoph@lameter.com>
Signed-off-by: Linus Torvalds <torvalds@osdl.org>
try_to_free_pages accepts a third argument, order, but hasn't used it since
before 2.6.0. The following patch removes the argument and updates all the
calls to try_to_free_pages.
Signed-off-by: Darren Hart <dvhltc@us.ibm.com>
Signed-off-by: Andrew Morton <akpm@osdl.org>
Signed-off-by: Linus Torvalds <torvalds@osdl.org>
When early zone reclaim is turned on the LRU is scanned more frequently when a
zone is low on memory. This limits when the zone reclaim can be called by
skipping the scan if another thread (either via kswapd or sync reclaim) is
already reclaiming from the zone.
Signed-off-by: Martin Hicks <mort@sgi.com>
Signed-off-by: Andrew Morton <akpm@osdl.org>
Signed-off-by: Linus Torvalds <torvalds@osdl.org>
This is the core of the (much simplified) early reclaim. The goal of this
patch is to reclaim some easily-freed pages from a zone before falling back
onto another zone.
One of the major uses of this is NUMA machines. With the default allocator
behavior the allocator would look for memory in another zone, which might be
off-node, before trying to reclaim from the current zone.
This adds a zone tuneable to enable early zone reclaim. It is selected on a
per-zone basis and is turned on/off via syscall.
Adding some extra throttling on the reclaim was also required (patch
4/4). Without the machine would grind to a crawl when doing a "make -j"
kernel build. Even with this patch the System Time is higher on
average, but it seems tolerable. Here are some numbers for kernbench
runs on a 2-node, 4cpu, 8Gig RAM Altix in the "make -j" run:
wall user sys %cpu ctx sw. sleeps
---- ---- --- ---- ------ ------
No patch 1009 1384 847 258 298170 504402
w/patch, no reclaim 880 1376 667 288 254064 396745
w/patch & reclaim 1079 1385 926 252 291625 548873
These numbers are the average of 2 runs of 3 "make -j" runs done right
after system boot. Run-to-run variability for "make -j" is huge, so
these numbers aren't terribly useful except to seee that with reclaim
the benchmark still finishes in a reasonable amount of time.
I also looked at the NUMA hit/miss stats for the "make -j" runs and the
reclaim doesn't make any difference when the machine is thrashing away.
Doing a "make -j8" on a single node that is filled with page cache pages
takes 700 seconds with reclaim turned on and 735 seconds without reclaim
(due to remote memory accesses).
The simple zone_reclaim syscall program is at
http://www.bork.org/~mort/sgi/zone_reclaim.c
Signed-off-by: Martin Hicks <mort@sgi.com>
Signed-off-by: Andrew Morton <akpm@osdl.org>
Signed-off-by: Linus Torvalds <torvalds@osdl.org>
Here's the next round of these patches. These are totally different in
an attempt to meet the "simpler" request after the last patches. For
reference the earlier threads are:
http://marc.theaimsgroup.com/?l=linux-kernel&m=110839604924587&w=2http://marc.theaimsgroup.com/?l=linux-mm&m=111461480721249&w=2
This set of patches replaces my other vm- patches that are currently in
-mm. So they're against 2.6.12-rc5-mm1 about half way through the -mm
patchset.
As I said already this patch is a lot simpler. The reclaim is turned on
or off on a per-zone basis using a syscall. I haven't tested the x86
syscall, so it might be wrong. It uses the existing reclaim/pageout
code with the small addition of a may_swap flag to scan_control
(patch 1/4).
I also added __GFP_NORECLAIM (patch 3/4) so that certain allocation
types can be flagged to never cause reclaim. This was a deficiency
that was in all of my earlier patch sets. Previously, doing a big
buffered read would fill one zone with page cache and then start to
reclaim from that same zone, leaving the other zones untouched.
Adding some extra throttling on the reclaim was also required (patch
4/4). Without the machine would grind to a crawl when doing a "make -j"
kernel build. Even with this patch the System Time is higher on
average, but it seems tolerable. Here are some numbers for kernbench
runs on a 2-node, 4cpu, 8Gig RAM Altix in the "make -j" run:
wall user sys %cpu ctx sw. sleeps
---- ---- --- ---- ------ ------
No patch 1009 1384 847 258 298170 504402
w/patch, no reclaim 880 1376 667 288 254064 396745
w/patch & reclaim 1079 1385 926 252 291625 548873
These numbers are the average of 2 runs of 3 "make -j" runs done right
after system boot. Run-to-run variability for "make -j" is huge, so
these numbers aren't terribly useful except to seee that with reclaim
the benchmark still finishes in a reasonable amount of time.
I also looked at the NUMA hit/miss stats for the "make -j" runs and the
reclaim doesn't make any difference when the machine is thrashing away.
Doing a "make -j8" on a single node that is filled with page cache pages
takes 700 seconds with reclaim turned on and 735 seconds without reclaim
(due to remote memory accesses).
The simple zone_reclaim syscall program is at
http://www.bork.org/~mort/sgi/zone_reclaim.c
This patch:
This adds an extra switch to the scan_control struct. It simply lets the
reclaim code know if its allowed to swap pages out.
This was required for a simple per-zone reclaimer. Without this addition
pages would be swapped out as soon as a zone ran out of memory and the early
reclaim kicked in.
Signed-off-by: Martin Hicks <mort@sgi.com>
Signed-off-by: Andrew Morton <akpm@osdl.org>
Signed-off-by: Linus Torvalds <torvalds@osdl.org>
Fix a problem identified by Andrea Arcangeli <andrea@suse.de>
kswapd will set a zone into all_unreclaimable state if it sees that we're not
successfully reclaiming LRU pages. But that fails to notice that we're
successfully reclaiming slab obects, so we can set all_unreclaimable too soon.
So change shrink_slab() to return a success indication if it actually
reclaimed some objects, and don't assume that the zone is all_unreclaimable if
that is true. This means that we won't enter all_unreclaimable state if we
are successfully freeing slab objects but we're not yet actually freeing slab
pages, due to internal fragmentation.
(hm, this has a shortcoming. We could be successfully freeing ZONE_NORMAL
slab objects while being really oom on ZONE_DMA. If that happens then kswapd
might burn a lot of CPU. But given that there might be some slab objects in
ZONE_DMA, perhaps that is appropriate.)
Signed-off-by: Andrew Morton <akpm@osdl.org>
Signed-off-by: Linus Torvalds <torvalds@osdl.org>
)
We only call pageout() for dirty pages, so this test is redundant.
Signed-off-by: Andrew Morton <akpm@osdl.org>
Signed-off-by: Linus Torvalds <torvalds@osdl.org>
Initial git repository build. I'm not bothering with the full history,
even though we have it. We can create a separate "historical" git
archive of that later if we want to, and in the meantime it's about
3.2GB when imported into git - space that would just make the early
git days unnecessarily complicated, when we don't have a lot of good
infrastructure for it.
Let it rip!
Christoph Lameter