Asynchronous I/O can easily starve synchronous I/O (both sync reads
and sync writes), by consuming all request tags. Similarly, storms of
synchronous writes, such as those that sync(2) may trigger, can starve
synchronous reads. In their turn, these two problems may also cause
BFQ to loose control on latency for interactive and soft real-time
applications. For example, on a PLEXTOR PX-256M5S SSD, LibreOffice
Writer takes 0.6 seconds to start if the device is idle, but it takes
more than 45 seconds (!) if there are sequential writes in the
background.
This commit addresses this issue by limiting the maximum percentage of
tags that asynchronous I/O requests and synchronous write requests can
consume. In particular, this commit grants a higher threshold to
synchronous writes, to prevent the latter from being starved by
asynchronous I/O.
According to the above test, LibreOffice Writer now starts in about
1.2 seconds on average, regardless of the background workload, and
apart from some rare outlier. To check this improvement, run, e.g.,
sudo ./comm_startup_lat.sh bfq 5 5 seq 10 "lowriter --terminate_after_init"
for the comm_startup_lat benchmark in the S suite [1].
[1] https://github.com/Algodev-github/S
Tested-by: Oleksandr Natalenko <oleksandr@natalenko.name>
Tested-by: Holger Hoffstätte <holger@applied-asynchrony.com>
Signed-off-by: Paolo Valente <paolo.valente@linaro.org>
Signed-off-by: Jens Axboe <axboe@kernel.dk>
If we run into blk_mq_request_direct_issue(), when queue is busy, we
don't want to dispatch this request into hctx->dispatch_list, and
what we need to do is to return the queue busy info to caller, so
that caller can deal with it well.
Fixes: 396eaf21ee ("blk-mq: improve DM's blk-mq IO merging via blk_insert_cloned_request feedback")
Reported-by: Laurence Oberman <loberman@redhat.com>
Reviewed-by: Mike Snitzer <snitzer@redhat.com>
Signed-off-by: Ming Lei <ming.lei@redhat.com>
Signed-off-by: Jens Axboe <axboe@kernel.dk>
Fixes: 4246a0b63b ("block: add a bi_error field to struct bio")
Reviewed-by: Martin K. Petersen <martin.petersen@oracle.com>
Signed-off-by: Bart Van Assche <bart.vanassche@wdc.com>
Signed-off-by: Jens Axboe <axboe@kernel.dk>
After commit:
923218f616 ("blk-mq: don't allocate driver tag upfront for flush rq")
we no longer use the 'can_block' argument in
blk_mq_sched_insert_request(). Kill it.
Signed-off-by: Mike Snitzer <snitzer@redhat.com>
Added actual commit message as to why it's being removed.
Signed-off-by: Jens Axboe <axboe@kernel.dk>
blk_insert_cloned_request() is called in the fast path of a dm-rq driver
(e.g. blk-mq request-based DM mpath). blk_insert_cloned_request() uses
blk_mq_request_bypass_insert() to directly append the request to the
blk-mq hctx->dispatch_list of the underlying queue.
1) This way isn't efficient enough because the hctx spinlock is always
used.
2) With blk_insert_cloned_request(), we completely bypass underlying
queue's elevator and depend on the upper-level dm-rq driver's elevator
to schedule IO. But dm-rq currently can't get the underlying queue's
dispatch feedback at all. Without knowing whether a request was issued
or not (e.g. due to underlying queue being busy) the dm-rq elevator will
not be able to provide effective IO merging (as a side-effect of dm-rq
currently blindly destaging a request from its elevator only to requeue
it after a delay, which kills any opportunity for merging). This
obviously causes very bad sequential IO performance.
Fix this by updating blk_insert_cloned_request() to use
blk_mq_request_direct_issue(). blk_mq_request_direct_issue() allows a
request to be issued directly to the underlying queue and returns the
dispatch feedback (blk_status_t). If blk_mq_request_direct_issue()
returns BLK_SYS_RESOURCE the dm-rq driver will now use DM_MAPIO_REQUEUE
to _not_ destage the request. Whereby preserving the opportunity to
merge IO.
With this, request-based DM's blk-mq sequential IO performance is vastly
improved (as much as 3X in mpath/virtio-scsi testing).
Signed-off-by: Ming Lei <ming.lei@redhat.com>
[blk-mq.c changes heavily influenced by Ming Lei's initial solution, but
they were refactored to make them less fragile and easier to read/review]
Signed-off-by: Mike Snitzer <snitzer@redhat.com>
Signed-off-by: Jens Axboe <axboe@kernel.dk>
No functional change. Just makes code flow more logically.
In following commit, __blk_mq_try_issue_directly() will be used to
return the dispatch result (blk_status_t) to DM. DM needs this
information to improve IO merging.
Signed-off-by: Mike Snitzer <snitzer@redhat.com>
Signed-off-by: Jens Axboe <axboe@kernel.dk>
We know this WARN_ON is harmless and in reality it may be trigged,
so convert it to printk() and dump_stack() to avoid to confusing
people.
Also add comment about two releated races here.
Cc: Christian Borntraeger <borntraeger@de.ibm.com>
Cc: Stefan Haberland <sth@linux.vnet.ibm.com>
Cc: Christoph Hellwig <hch@lst.de>
Cc: Thomas Gleixner <tglx@linutronix.de>
Cc: "jianchao.wang" <jianchao.w.wang@oracle.com>
Signed-off-by: Ming Lei <ming.lei@redhat.com>
Signed-off-by: Jens Axboe <axboe@kernel.dk>
When hctx->next_cpu is set from possible online CPUs, there is one
race in which hctx->next_cpu may be set as >= nr_cpu_ids, and finally
break workqueue.
The race can be triggered in the following two sitations:
1) when one CPU is becoming DEAD, blk_mq_hctx_notify_dead() is called
to dispatch requests from the DEAD cpu context, but at that
time, this DEAD CPU has been cleared from 'cpu_online_mask', so all
CPUs in hctx->cpumask may become offline, and cause hctx->next_cpu set
a bad value.
2) blk_mq_delay_run_hw_queue() is called from CPU B, and found the queue
should be run on the other CPU A, then CPU A may become offline at the
same time and all CPUs in hctx->cpumask become offline.
This patch deals with this issue by re-selecting next CPU, and making
sure it is set correctly.
Cc: Christian Borntraeger <borntraeger@de.ibm.com>
Cc: Stefan Haberland <sth@linux.vnet.ibm.com>
Cc: Christoph Hellwig <hch@lst.de>
Cc: Thomas Gleixner <tglx@linutronix.de>
Reported-by: "jianchao.wang" <jianchao.w.wang@oracle.com>
Tested-by: "jianchao.wang" <jianchao.w.wang@oracle.com>
Fixes: 20e4d81393 ("blk-mq: simplify queue mapping & schedule with each possisble CPU")
Signed-off-by: Ming Lei <ming.lei@redhat.com>
Signed-off-by: Jens Axboe <axboe@kernel.dk>
During stress tests by syzkaller on the sg driver the block layer
infrequently returns EINVAL. Closer inspection shows the block
layer was trying to return ENOMEM (which is much more
understandable) but for some reason overroad that useful error.
Patch below does not show this (unchanged) line:
ret =__blk_rq_map_user_iov(rq, map_data, &i, gfp_mask, copy);
That 'ret' was being overridden when that function failed.
Signed-off-by: Douglas Gilbert <dgilbert@interlog.com>
Signed-off-by: Jens Axboe <axboe@kernel.dk>
Since I can remember DM has forced the block layer to allow the
allocation and initialization of the request_queue to be distinct
operations. Reason for this is block/genhd.c:add_disk() has requires
that the request_queue (and associated bdi) be tied to the gendisk
before add_disk() is called -- because add_disk() also deals with
exposing the request_queue via blk_register_queue().
DM's dynamic creation of arbitrary device types (and associated
request_queue types) requires the DM device's gendisk be available so
that DM table loads can establish a master/slave relationship with
subordinate devices that are referenced by loaded DM tables -- using
bd_link_disk_holder(). But until these DM tables, and their associated
subordinate devices, are known DM cannot know what type of request_queue
it needs -- nor what its queue_limits should be.
This chicken and egg scenario has created all manner of problems for DM
and, at times, the block layer.
Summary of changes:
- Add device_add_disk_no_queue_reg() and add_disk_no_queue_reg() variant
that drivers may use to add a disk without also calling
blk_register_queue(). Driver must call blk_register_queue() once its
request_queue is fully initialized.
- Return early from blk_unregister_queue() if QUEUE_FLAG_REGISTERED
is not set. It won't be set if driver used add_disk_no_queue_reg()
but driver encounters an error and must del_gendisk() before calling
blk_register_queue().
- Export blk_register_queue().
These changes allow DM to use add_disk_no_queue_reg() to anchor its
gendisk as the "master" for master/slave relationships DM must establish
with subordinate devices referenced in DM tables that get loaded. Once
all "slave" devices for a DM device are known its request_queue can be
properly initialized and then advertised via sysfs -- important
improvement being that no request_queue resource initialization
performed by blk_register_queue() is missed for DM devices anymore.
Signed-off-by: Mike Snitzer <snitzer@redhat.com>
Reviewed-by: Ming Lei <ming.lei@redhat.com>
Signed-off-by: Jens Axboe <axboe@kernel.dk>
The original commit e9a823fb34 (block: fix warning when I/O elevator
is changed as request_queue is being removed) is pretty conflated.
"conflated" because the resource being protected by q->sysfs_lock isn't
the queue_flags (it is the 'queue' kobj).
q->sysfs_lock serializes __elevator_change() (via elv_iosched_store)
from racing with blk_unregister_queue():
1) By holding q->sysfs_lock first, __elevator_change() can complete
before a racing blk_unregister_queue().
2) Conversely, __elevator_change() is testing for QUEUE_FLAG_REGISTERED
in case elv_iosched_store() loses the race with blk_unregister_queue(),
it needs a way to know the 'queue' kobj isn't there.
Expand the scope of blk_unregister_queue()'s q->sysfs_lock use so it is
held until after the 'queue' kobj is removed.
To do so blk_mq_unregister_dev() must not also take q->sysfs_lock. So
rename __blk_mq_unregister_dev() to blk_mq_unregister_dev().
Also, blk_unregister_queue() should use q->queue_lock to protect against
any concurrent writes to q->queue_flags -- even though chances are the
queue is being cleaned up so no concurrent writes are likely.
Fixes: e9a823fb34 ("block: fix warning when I/O elevator is changed as request_queue is being removed")
Signed-off-by: Mike Snitzer <snitzer@redhat.com>
Reviewed-by: Ming Lei <ming.lei@redhat.com>
Signed-off-by: Jens Axboe <axboe@kernel.dk>
device_add_disk() will only call bdi_register_owner() if
!GENHD_FL_HIDDEN, so it follows that del_gendisk() should only call
bdi_unregister() if !GENHD_FL_HIDDEN.
Found with code inspection. bdi_unregister() won't do any harm if
bdi_register_owner() wasn't used but best to avoid the unnecessary
call to bdi_unregister().
Fixes: 8ddcd65325 ("block: introduce GENHD_FL_HIDDEN")
Signed-off-by: Mike Snitzer <snitzer@redhat.com>
Reviewed-by: Ming Lei <ming.lei@redhat.com>
Reviewed-by: Hannes Reinecke <hare@suse.com>
Signed-off-by: Jens Axboe <axboe@kernel.dk>
A previous commit moved the clearing of rq->rq_flags later,
but we may have already set RQF_MQ_INFLIGHT when that happens.
Ensure that we correctly initialize rq->rq_flags to the
right value.
This is based on an original fix by Ming, just rewritten to not
require a conditional.
Fixes: 7c3fb70f03 ("block: rearrange a few request fields for better cache layout")
Reviewed-by: Mike Snitzer <snitzer@redhat.com>
Signed-off-by: Jens Axboe <axboe@kernel.dk>
Looking at debug output, we see:
./000000009ddfa913/requeue_list:000000009646711c {.op=READ, .state=idle, gen=0x1
18, abort_gen=0x0, .cmd_flags=, .rq_flags=SORTED|1|SOFTBARRIER|IO_STAT, complete
=0, .tag=-1, .internal_tag=217}
Note the '1' between SORTED and SOFTBARRIER - that's because no name
as defined for RQF_STARTED. Fixed that.
Signed-off-by: Jens Axboe <axboe@kernel.dk>
The previous patch assigns interrupt vectors to all possible CPUs, so
now hctx can be mapped to possible CPUs, this patch applies this fact
to simplify queue mapping & schedule so that we don't need to handle
CPU hotplug for dealing with physical CPU plug & unplug. With this
simplication, we can work well on physical CPU plug & unplug, which
is a normal use case for VM at least.
Make sure we allocate blk_mq_ctx structures for all possible CPUs, and
set hctx->numa_node for possible CPUs which are mapped to this hctx. And
only choose the online CPUs for schedule.
Reported-by: Christian Borntraeger <borntraeger@de.ibm.com>
Tested-by: Christian Borntraeger <borntraeger@de.ibm.com>
Tested-by: Stefan Haberland <sth@linux.vnet.ibm.com>
Cc: Thomas Gleixner <tglx@linutronix.de>
Signed-off-by: Christoph Hellwig <hch@lst.de>
Fixes: 4b855ad371 ("blk-mq: Create hctx for each present CPU")
(merged the three into one because any single one may not work, and fix
selecting online CPUs for scheduler)
Signed-off-by: Ming Lei <ming.lei@redhat.com>
Signed-off-by: Jens Axboe <axboe@kernel.dk>
This patch does not change any functionality but makes the
blk_mq_mark_tag_wait() code slightly easier to read.
Signed-off-by: Bart Van Assche <bart.vanassche@wdc.com>
Cc: Christoph Hellwig <hch@lst.de>
Cc: Omar Sandoval <osandov@fb.com>
Cc: Hannes Reinecke <hare@suse.de>
Cc: Johannes Thumshirn <jthumshirn@suse.de>
Signed-off-by: Jens Axboe <axboe@kernel.dk>
After the first few months, the message has not led to many bug reports.
It's been almost five years now, and in practice the main source of
it seems to be MTIOCGET that someone is using to detect tape devices.
While we could whitelist it just like CDROM_GET_CAPABILITY, this patch
just removes the message altogether.
The patch also removes the "safe but not very useful" ioctl whitelist,
as suggested by Christoph. I doubt anything is using most of those
ioctls _in general_, let alone on a partition.
Reviewed-by: Christoph Hellwig <hch@lst.de>
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
Signed-off-by: Jens Axboe <axboe@kernel.dk>
Move completion related items (like the call single data) near the
end of the struct, instead of mixing them in with the initial
queueing related fields.
Move queuelist below the bio structures. Then we have all
queueing related bits in the first cache line.
This yields a 1.5-2% increase in IOPS for a null_blk test, both for
sync and for high thread count access. Sync test goes form 975K to
992K, 32-thread case from 20.8M to 21.2M IOPS.
Reviewed-by: Bart Van Assche <bart.vanassche@wdc.com>
Reviewed-by: Omar Sandoval <osandov@fb.com>
Signed-off-by: Jens Axboe <axboe@kernel.dk>
We only have one atomic flag left. Instead of using an entire
unsigned long for that, steal the bottom bit of the deadline
field that we already reserved.
Remove ->atomic_flags, since it's now unused.
Reviewed-by: Bart Van Assche <bart.vanassche@wdc.com>
Reviewed-by: Omar Sandoval <osandov@fb.com>
Signed-off-by: Jens Axboe <axboe@kernel.dk>
We reduce the resolution of request expiry, but since we're already
using jiffies for this where resolution depends on the kernel
configuration and since the timeout resolution is coarse anyway,
that should be fine.
Reviewed-by: Bart Van Assche <bart.vanassche@wdc.com>
Reviewed-by: Omar Sandoval <osandov@fb.com>
Signed-off-by: Jens Axboe <axboe@kernel.dk>
We don't need this to be an atomic flag, it can be a regular
flag. We either end up on the same CPU for the polling, in which
case the state is sane, or we did the sleep which would imply
the needed barrier to ensure we see the right state.
Reviewed-by: Bart Van Assche <bart.vanassche@wdc.com>
Reviewed-by: Omar Sandoval <osandov@fb.com>
Signed-off-by: Jens Axboe <axboe@kernel.dk>
It is nontrivial to derive from the blk-mq source code when
blk_mq_tags.active_queues is decremented. Hence add a comment that
explains this.
Signed-off-by: Bart Van Assche <bart.vanassche@wdc.com>
Cc: Christoph Hellwig <hch@lst.de>
Cc: Hannes Reinecke <hare@suse.de>
Cc: Johannes Thumshirn <jthumshirn@suse.de>
Signed-off-by: Jens Axboe <axboe@kernel.dk>
UFS partitions from newer versions of FreeBSD 10 and 11 use relative
addressing for their subpartitions. But older versions of FreeBSD still
use absolute addressing just like OpenBSD and NetBSD.
Instead of simply testing for a FreeBSD partition, the code needs to
also test if the starting offset of the C subpartition is zero.
https://bugzilla.kernel.org/show_bug.cgi?id=197733
Signed-off-by: Richard Narron <comet.berkeley@gmail.com>
Signed-off-by: Jens Axboe <axboe@kernel.dk>
Commit '7b9e93616399' ("blk-mq-sched: unify request finished methods")
changed the old name of current bfq_finish_request method, but left it
unchanged elsewhere in the code (related comments, part of function
name bfq_put_rq_priv_body).
This commit fixes all occurrences of the old name of this method by
changing them into the current name.
Fixes: 7b9e936163 ("blk-mq-sched: unify request finished methods")
Reviewed-by: Paolo Valente <paolo.valente@linaro.org>
Signed-off-by: Federico Motta <federico@willer.it>
Signed-off-by: Chiara Bruschi <bruschi.chiara@outlook.it>
Signed-off-by: Jens Axboe <axboe@kernel.dk>
This reverts commit a2d37968d7.
If max segment size isn't 512-aligned, this patch won't work well.
Also once multipage bvec is enabled, adjacent bvecs won't be physically
contiguous if page is added via bio_add_page(), so we don't need this
kind of complicated logic.
Reported-by: Dmitry Osipenko <digetx@gmail.com>
Signed-off-by: Ming Lei <ming.lei@redhat.com>
Signed-off-by: Jens Axboe <axboe@kernel.dk>
It's not available if we don't have group io scheduling set, and
there's no need to call it.
Fixes: 0d52af5905 ("block, bfq: release oom-queue ref to root group on exit")
Signed-off-by: Jens Axboe <axboe@kernel.dk>
Commit 3a025e1d1c ("Add optional check for bad kernel-doc comments")
causes W=1 the kernel-doc script to be run and thereby causes several
new warnings to appear when building the kernel with W=1. Fix the
block layer kernel-doc headers such that the block layer again builds
cleanly with W=1.
Signed-off-by: Bart Van Assche <bart.vanassche@wdc.com>
Cc: Martin K. Petersen <martin.petersen@oracle.com>
Cc: Christoph Hellwig <hch@lst.de>
Cc: Hannes Reinecke <hare@suse.de>
Cc: Johannes Thumshirn <jthumshirn@suse.de>
Signed-off-by: Jens Axboe <axboe@kernel.dk>
Change "nedeing" into "needing" and "caes" into "cases".
Fixes: commit f906a6a0f4 ("blk-mq: improve tag waiting setup for non-shared tags")
Signed-off-by: Bart Van Assche <bart.vanassche@wdc.com>
Cc: Christoph Hellwig <hch@lst.de>
Cc: Omar Sandoval <osandov@fb.com>
Cc: Hannes Reinecke <hare@suse.de>
Cc: Johannes Thumshirn <jthumshirn@suse.de>
Signed-off-by: Jens Axboe <axboe@kernel.dk>
In some stupider versions of gcc, it complains:
block/blk-mq.c: In function ‘blk_mq_complete_request’:
./include/linux/srcu.h:175:2: warning: ‘srcu_idx’ may be used uninitialized in this function [-Wmaybe-uninitialized]
__srcu_read_unlock(sp, idx);
^
block/blk-mq.c:620:6: note: ‘srcu_idx’ was declared here
int srcu_idx;
^
which is completely bogus, since we only use srcu_idx when
hctx->flags & BLK_MQ_F_BLOCKING is set, and that's the case where
hctx_lock() has initialized it.
Just set it to '0' in the normal path in hctx_lock() to silence
this annoying warning.
Fixes: 04ced159ce ("blk-mq: move hctx lock/unlock into a helper")
Fixes: 5197c05e16 ("blk-mq: protect completion path with RCU")
Signed-off-by: Jens Axboe <axboe@kernel.dk>
The RCU protection has been expanded to cover both queueing and
completion paths making ->queue_rq_srcu a misnomer. Rename it to
->srcu as suggested by Bart.
Signed-off-by: Tejun Heo <tj@kernel.org>
Cc: Bart Van Assche <Bart.VanAssche@wdc.com>
Signed-off-by: Jens Axboe <axboe@kernel.dk>
After the recent updates to use generation number and state based
synchronization, we can easily replace REQ_ATOM_STARTED usages by
adding an extra state to distinguish completed but not yet freed
state.
Add MQ_RQ_COMPLETE and replace REQ_ATOM_STARTED usages with
blk_mq_rq_state() tests. REQ_ATOM_STARTED no longer has any users
left and is removed.
Signed-off-by: Tejun Heo <tj@kernel.org>
Signed-off-by: Jens Axboe <axboe@kernel.dk>
After the recent updates to use generation number and state based
synchronization, blk-mq no longer depends on REQ_ATOM_COMPLETE except
to avoid firing the same timeout multiple times.
Remove all REQ_ATOM_COMPLETE usages and use a new rq_flags flag
RQF_MQ_TIMEOUT_EXPIRED to avoid firing the same timeout multiple
times. This removes atomic bitops from hot paths too.
v2: Removed blk_clear_rq_complete() from blk_mq_rq_timed_out().
v3: Added RQF_MQ_TIMEOUT_EXPIRED flag.
Signed-off-by: Tejun Heo <tj@kernel.org>
Cc: "jianchao.wang" <jianchao.w.wang@oracle.com>
Signed-off-by: Jens Axboe <axboe@kernel.dk>
With issue/complete and timeout paths now using the generation number
and state based synchronization, blk_abort_request() is the only one
which depends on REQ_ATOM_COMPLETE for arbitrating completion.
There's no reason for blk_abort_request() to be a completely separate
path. This patch makes blk_abort_request() piggyback on the timeout
path instead of trying to terminate the request directly.
This removes the last dependency on REQ_ATOM_COMPLETE in blk-mq.
Note that this makes blk_abort_request() asynchronous - it initiates
abortion but the actual termination will happen after a short while,
even when the caller owns the request. AFAICS, SCSI and ATA should be
fine with that and I think mtip32xx and dasd should be safe but not
completely sure. It'd be great if people who know the drivers take a
look.
v2: - Add comment explaining the lack of synchronization around
->deadline update as requested by Bart.
Signed-off-by: Tejun Heo <tj@kernel.org>
Cc: Asai Thambi SP <asamymuthupa@micron.com>
Cc: Stefan Haberland <sth@linux.vnet.ibm.com>
Cc: Jan Hoeppner <hoeppner@linux.vnet.ibm.com>
Cc: Bart Van Assche <Bart.VanAssche@wdc.com>
Signed-off-by: Jens Axboe <axboe@kernel.dk>
blk_mq_check_inflight() and blk_mq_poll_hybrid_sleep() test
REQ_ATOM_COMPLETE to determine the request state. Both uses are
speculative and we can test REQ_ATOM_STARTED and blk_mq_rq_state() for
equivalent results. Replace the tests. This will allow removing
REQ_ATOM_COMPLETE usages from blk-mq.
Signed-off-by: Tejun Heo <tj@kernel.org>
Signed-off-by: Jens Axboe <axboe@kernel.dk>
Currently, blk-mq timeout path synchronizes against the usual
issue/completion path using a complex scheme involving atomic
bitflags, REQ_ATOM_*, memory barriers and subtle memory coherence
rules. Unfortunately, it contains quite a few holes.
There's a complex dancing around REQ_ATOM_STARTED and
REQ_ATOM_COMPLETE between issue/completion and timeout paths; however,
they don't have a synchronization point across request recycle
instances and it isn't clear what the barriers add.
blk_mq_check_expired() can easily read STARTED from N-2'th iteration,
deadline from N-1'th, blk_mark_rq_complete() against Nth instance.
In fact, it's pretty easy to make blk_mq_check_expired() terminate a
later instance of a request. If we induce 5 sec delay before
time_after_eq() test in blk_mq_check_expired(), shorten the timeout to
2s, and issue back-to-back large IOs, blk-mq starts timing out
requests spuriously pretty quickly. Nothing actually timed out. It
just made the call on a recycle instance of a request and then
terminated a later instance long after the original instance finished.
The scenario isn't theoretical either.
This patch replaces the broken synchronization mechanism with a RCU
and generation number based one.
1. Each request has a u64 generation + state value, which can be
updated only by the request owner. Whenever a request becomes
in-flight, the generation number gets bumped up too. This provides
the basis for the timeout path to distinguish different recycle
instances of the request.
Also, marking a request in-flight and setting its deadline are
protected with a seqcount so that the timeout path can fetch both
values coherently.
2. The timeout path fetches the generation, state and deadline. If
the verdict is timeout, it records the generation into a dedicated
request abortion field and does RCU wait.
3. The completion path is also protected by RCU (from the previous
patch) and checks whether the current generation number and state
match the abortion field. If so, it skips completion.
4. The timeout path, after RCU wait, scans requests again and
terminates the ones whose generation and state still match the ones
requested for abortion.
By now, the timeout path knows that either the generation number
and state changed if it lost the race or the completion will yield
to it and can safely timeout the request.
While it's more lines of code, it's conceptually simpler, doesn't
depend on direct use of subtle memory ordering or coherence, and
hopefully doesn't terminate the wrong instance.
While this change makes REQ_ATOM_COMPLETE synchronization unnecessary
between issue/complete and timeout paths, REQ_ATOM_COMPLETE isn't
removed yet as it's still used in other places. Future patches will
move all state tracking to the new mechanism and remove all bitops in
the hot paths.
Note that this patch adds a comment explaining a race condition in
BLK_EH_RESET_TIMER path. The race has always been there and this
patch doesn't change it. It's just documenting the existing race.
v2: - Fixed BLK_EH_RESET_TIMER handling as pointed out by Jianchao.
- s/request->gstate_seqc/request->gstate_seq/ as suggested by Peter.
- READ_ONCE() added in blk_mq_rq_update_state() as suggested by Peter.
v3: - Fixed possible extended seqcount / u64_stats_sync read looping
spotted by Peter.
- MQ_RQ_IDLE was incorrectly being set in complete_request instead
of free_request. Fixed.
v4: - Rebased on top of hctx_lock() refactoring patch.
- Added comment explaining the use of hctx_lock() in completion path.
v5: - Added comments requested by Bart.
- Note the addition of BLK_EH_RESET_TIMER race condition in the
commit message.
Signed-off-by: Tejun Heo <tj@kernel.org>
Cc: "jianchao.wang" <jianchao.w.wang@oracle.com>
Cc: Peter Zijlstra <peterz@infradead.org>
Cc: Christoph Hellwig <hch@lst.de>
Cc: Bart Van Assche <Bart.VanAssche@wdc.com>
Signed-off-by: Jens Axboe <axboe@kernel.dk>
Currently, blk-mq protects only the issue path with RCU. This patch
puts the completion path under the same RCU protection. This will be
used to synchronize issue/completion against timeout by later patches,
which will also add the comments.
Signed-off-by: Tejun Heo <tj@kernel.org>
Signed-off-by: Jens Axboe <axboe@kernel.dk>
Move the RCU vs SRCU logic into lock/unlock helpers, which makes
the actual functional bits within the locked region much easier
to read.
tj: Reordered in front of timeout revamp patches and added the missing
blk_mq_run_hw_queue() conversion.
Signed-off-by: Jens Axboe <axboe@kernel.dk>
Signed-off-by: Tejun Heo <tj@kernel.org>
Signed-off-by: Jens Axboe <axboe@kernel.dk>
On scheduler init, a reference to the root group, and a reference to
its corresponding blkg are taken for the oom queue. Yet these
references are not released on scheduler exit, which prevents these
objects from be freed. This commit adds the missing reference
releases.
Reported-by: Davide Ferrari <davideferrari8@gmail.com>
Tested-by: Holger Hoffstätte <holger@applied-asynchrony.com>
Signed-off-by: Paolo Valente <paolo.valente@linaro.org>
Signed-off-by: Jens Axboe <axboe@kernel.dk>
For each pair [device for which bfq is selected as I/O scheduler,
group in blkio/io], bfq maintains a corresponding bfq group. Each such
bfq group contains a set of async queues, with each async queue
created on demand, i.e., when some I/O request arrives for it. On
creation, an async queue gets an extra reference, to make sure that
the queue is not freed as long as its bfq group exists. Accordingly,
to allow the queue to be freed after the group exited, this extra
reference must released on group exit.
The above holds also for a bfq root group, i.e., for the bfq group
corresponding to the root blkio/io root for a given device. Yet, by
mistake, the references to the existing async queues of a root group
are not released when the latter exits. This causes a memory leak when
the instance of bfq for a given device exits. In a similar vein,
bfqg_stats_xfer_dead is not executed for a root group.
This commit fixes bfq_pd_offline so that the latter executes the above
missing operations for a root group too.
Reported-by: Holger Hoffstätte <holger@applied-asynchrony.com>
Reported-by: Guoqing Jiang <gqjiang@suse.com>
Tested-by: Holger Hoffstätte <holger@applied-asynchrony.com>
Signed-off-by: Davide Ferrari <davideferrari8@gmail.com>
Signed-off-by: Paolo Valente <paolo.valente@linaro.org>
Signed-off-by: Jens Axboe <axboe@kernel.dk>
In both elevator_switch_mq() and blk_mq_update_nr_hw_queues(), sched tags
can be allocated, and q->nr_hw_queue is used, and race is inevitable, for
example: blk_mq_init_sched() may trigger use-after-free on hctx, which is
freed in blk_mq_realloc_hw_ctxs() when nr_hw_queues is decreased.
This patch fixes the race be holding q->sysfs_lock.
Reviewed-by: Christoph Hellwig <hch@lst.de>
Reported-by: Yi Zhang <yi.zhang@redhat.com>
Tested-by: Yi Zhang <yi.zhang@redhat.com>
Signed-off-by: Ming Lei <ming.lei@redhat.com>
Signed-off-by: Jens Axboe <axboe@kernel.dk>
Dispatch may still be in-progress after queue is frozen, so we have to
quiesce queue before switching IO scheduler and updating nr_requests.
Also when switching io schedulers, blk_mq_run_hw_queue() may still be
called somewhere(such as from nvme_reset_work()), and io scheduler's
per-hctx data may not be setup yet, so cause oops even inside
blk_mq_hctx_has_pending(), such as it can be run just between:
ret = e->ops.mq.init_sched(q, e);
AND
ret = e->ops.mq.init_hctx(hctx, i)
inside blk_mq_init_sched().
This reverts commit 7a148c2fcff8330(block: don't call blk_mq_quiesce_queue()
after queue is frozen) basically, and makes sure blk_mq_hctx_has_pending
won't be called if queue is quiesced.
Reviewed-by: Christoph Hellwig <hch@lst.de>
Fixes: 7a148c2fcff83309(block: don't call blk_mq_quiesce_queue() after queue is frozen)
Reported-by: Yi Zhang <yi.zhang@redhat.com>
Tested-by: Yi Zhang <yi.zhang@redhat.com>
Signed-off-by: Ming Lei <ming.lei@redhat.com>
Signed-off-by: Jens Axboe <axboe@kernel.dk>
Don't pass in the hardware queue to __dd_dispatch_request(), since it
leads the reader to believe that we are returning a request for that
specific hardware queue. That's not how mq-deadline works, the state
for determining which request to serve next is shared across all
hardware queues for a device.
Reviewed-by: Omar Sandoval <osandov@fb.com>
Reviewed-by: Ming Lei <ming.lei@redhat.com>
Signed-off-by: Jens Axboe <axboe@kernel.dk>
In this case, 'sectors' can't be zero at all, so remove the check
and let the bio be split.
Reviewed-by: Christoph Hellwig <hch@lst.de>
Signed-off-by: Ming Lei <ming.lei@redhat.com>
Signed-off-by: Jens Axboe <axboe@kernel.dk>
When merging one bvec into segment, if the bvec is too big
to merge, current policy is to move the whole bvec into another
new segment.
This patchset changes the policy into trying to maximize size of
front segments, that means in above situation, part of bvec
is merged into current segment, and the remainder is put
into next segment.
This patch prepares for support multipage bvec because
it can be quite common to see this case and we should try
to make front segments in full size.
Signed-off-by: Ming Lei <ming.lei@redhat.com>
Signed-off-by: Jens Axboe <axboe@kernel.dk>
It is enough to check and compute bio->bi_seg_front_size just
after the 1st segment is found, but current code checks that
for each bvec, which is inefficient.
This patch follows the way in __blk_recalc_rq_segments()
for computing bio->bi_seg_front_size, and it is more efficient
and code becomes more readable too.
Signed-off-by: Ming Lei <ming.lei@redhat.com>
Signed-off-by: Jens Axboe <axboe@kernel.dk>
Paolo Valente