The memory barriers in list_del_init_careful() and list_empty_careful()
in pairs already handle the proper ordering between data.got_token
and data.wq.entry. So remove the redundant explicit barriers. And also
change a "break" statement to "return" to avoid redundant calling of
finish_wait().
Signed-off-by: Muchun Song <songmuchun@bytedance.com>
Reviewed-by: Chengming Zhou <chengming.zhou@linux.dev>
Reviewed-by: Omar Sandoval <osandov@fb.com>
Link: https://lore.kernel.org/r/20241021085251.73353-1-songmuchun@bytedance.com
Signed-off-by: Jens Axboe <axboe@kernel.dk>
We're seeing crashes from rq_qos_wake_function that look like this:
BUG: unable to handle page fault for address: ffffafe180a40084
#PF: supervisor write access in kernel mode
#PF: error_code(0x0002) - not-present page
PGD 100000067 P4D 100000067 PUD 10027c067 PMD 10115d067 PTE 0
Oops: Oops: 0002 [#1] PREEMPT SMP PTI
CPU: 17 UID: 0 PID: 0 Comm: swapper/17 Not tainted 6.12.0-rc3-00013-geca631b8fe80 #11
Hardware name: QEMU Standard PC (i440FX + PIIX, 1996), BIOS rel-1.16.0-0-gd239552ce722-prebuilt.qemu.org 04/01/2014
RIP: 0010:_raw_spin_lock_irqsave+0x1d/0x40
Code: 90 90 90 90 90 90 90 90 90 90 90 90 90 f3 0f 1e fa 0f 1f 44 00 00 41 54 9c 41 5c fa 65 ff 05 62 97 30 4c 31 c0 ba 01 00 00 00 <f0> 0f b1 17 75 0a 4c 89 e0 41 5c c3 cc cc cc cc 89 c6 e8 2c 0b 00
RSP: 0018:ffffafe180580ca0 EFLAGS: 00010046
RAX: 0000000000000000 RBX: ffffafe180a3f7a8 RCX: 0000000000000011
RDX: 0000000000000001 RSI: 0000000000000003 RDI: ffffafe180a40084
RBP: 0000000000000000 R08: 00000000001e7240 R09: 0000000000000011
R10: 0000000000000028 R11: 0000000000000888 R12: 0000000000000002
R13: ffffafe180a40084 R14: 0000000000000000 R15: 0000000000000003
FS: 0000000000000000(0000) GS:ffff9aaf1f280000(0000) knlGS:0000000000000000
CS: 0010 DS: 0000 ES: 0000 CR0: 0000000080050033
CR2: ffffafe180a40084 CR3: 000000010e428002 CR4: 0000000000770ef0
DR0: 0000000000000000 DR1: 0000000000000000 DR2: 0000000000000000
DR3: 0000000000000000 DR6: 00000000fffe0ff0 DR7: 0000000000000400
PKRU: 55555554
Call Trace:
<IRQ>
try_to_wake_up+0x5a/0x6a0
rq_qos_wake_function+0x71/0x80
__wake_up_common+0x75/0xa0
__wake_up+0x36/0x60
scale_up.part.0+0x50/0x110
wb_timer_fn+0x227/0x450
...
So rq_qos_wake_function() calls wake_up_process(data->task), which calls
try_to_wake_up(), which faults in raw_spin_lock_irqsave(&p->pi_lock).
p comes from data->task, and data comes from the waitqueue entry, which
is stored on the waiter's stack in rq_qos_wait(). Analyzing the core
dump with drgn, I found that the waiter had already woken up and moved
on to a completely unrelated code path, clobbering what was previously
data->task. Meanwhile, the waker was passing the clobbered garbage in
data->task to wake_up_process(), leading to the crash.
What's happening is that in between rq_qos_wake_function() deleting the
waitqueue entry and calling wake_up_process(), rq_qos_wait() is finding
that it already got a token and returning. The race looks like this:
rq_qos_wait() rq_qos_wake_function()
==============================================================
prepare_to_wait_exclusive()
data->got_token = true;
list_del_init(&curr->entry);
if (data.got_token)
break;
finish_wait(&rqw->wait, &data.wq);
^- returns immediately because
list_empty_careful(&wq_entry->entry)
is true
... return, go do something else ...
wake_up_process(data->task)
(NO LONGER VALID!)-^
Normally, finish_wait() is supposed to synchronize against the waker.
But, as noted above, it is returning immediately because the waitqueue
entry has already been removed from the waitqueue.
The bug is that rq_qos_wake_function() is accessing the waitqueue entry
AFTER deleting it. Note that autoremove_wake_function() wakes the waiter
and THEN deletes the waitqueue entry, which is the proper order.
Fix it by swapping the order. We also need to use
list_del_init_careful() to match the list_empty_careful() in
finish_wait().
Fixes: 38cfb5a45e ("blk-wbt: improve waking of tasks")
Cc: stable@vger.kernel.org
Signed-off-by: Omar Sandoval <osandov@fb.com>
Acked-by: Tejun Heo <tj@kernel.org>
Reviewed-by: Johannes Thumshirn <johannes.thumshirn@wdc.com>
Link: https://lore.kernel.org/r/d3bee2463a67b1ee597211823bf7ad3721c26e41.1729014591.git.osandov@fb.com
Signed-off-by: Jens Axboe <axboe@kernel.dk>
The explanatory comment used `set_task_state` instead of
`set_current_state` which is the function actually used in the code.
Signed-off-by: Alvaro Parker <alparkerdf@gmail.com>
Link: https://lore.kernel.org/r/20240903172214.520086-1-alparkerdf@gmail.com
Signed-off-by: Jens Axboe <axboe@kernel.dk>
The rq_qos_wait calls common wake-up function rq_qos_wake_function to get
token. Just replace stale wbt_wake_function with rq_qos_wake_function in
comment.
Signed-off-by: Kemeng Shi <shikemeng@huaweicloud.com>
Acked-by: Tejun Heo <tj@kernel.org>
Link: https://lore.kernel.org/r/20230914091508.36232-1-shikemeng@huaweicloud.com
Signed-off-by: Jens Axboe <axboe@kernel.dk>
commit 50e34d7881 ("block: disable the elevator int del_gendisk")
move rq_qos_exit() from disk_release() to del_gendisk(), this will
introduce some problems:
1) If rq_qos_add() is triggered by enabling iocost/iolatency through
cgroupfs, then it can concurrent with del_gendisk(), it's not safe to
write 'q->rq_qos' concurrently.
2) Activate cgroup policy that is relied on rq_qos will call
rq_qos_add() and blkcg_activate_policy(), and if rq_qos_exit() is
called in the middle, null-ptr-dereference will be triggered in
blkcg_activate_policy().
3) blkg_conf_open_bdev() can call blkdev_get_no_open() first to find the
disk, then if rq_qos_exit() from del_gendisk() is done before
rq_qos_add(), then memory will be leaked.
This patch add a new disk level mutex 'rq_qos_mutex':
1) The lock will protect rq_qos_exit() directly.
2) For wbt that doesn't relied on blk-cgroup, rq_qos_add() can only be
called from disk initialization for now because wbt can't be
destructed until rq_qos_exit(), so it's safe not to protect wbt for
now. Hoever, in case that rq_qos dynamically destruction is supported
in the furture, this patch also protect rq_qos_add() from wbt_init()
directly, this is enough because blk-sysfs already synchronize
writers with disk removal.
3) For iocost and iolatency, in order to synchronize disk removal and
cgroup configuration, the lock is held after blkdev_get_no_open()
from blkg_conf_open_bdev(), and is released in blkg_conf_exit().
In order to fix the above memory leak, disk_live() is checked after
holding the new lock.
Fixes: 50e34d7881 ("block: disable the elevator int del_gendisk")
Signed-off-by: Yu Kuai <yukuai3@huawei.com>
Acked-by: Tejun Heo <tj@kernel.org>
Link: https://lore.kernel.org/r/20230414084008.2085155-1-yukuai1@huaweicloud.com
Signed-off-by: Jens Axboe <axboe@kernel.dk>
This is what about half of the users already want, and it's only going to
grow more.
Signed-off-by: Christoph Hellwig <hch@lst.de>
Reviewed-by: Andreas Herrmann <aherrmann@suse.de>
Acked-by: Tejun Heo <tj@kernel.org>
Link: https://lore.kernel.org/r/20230203150400.3199230-16-hch@lst.de
Signed-off-by: Jens Axboe <axboe@kernel.dk>
These op vectors are constant, so mark them const.
Signed-off-by: Christoph Hellwig <hch@lst.de>
Reviewed-by: Andreas Herrmann <aherrmann@suse.de>
Acked-by: Tejun Heo <tj@kernel.org>
Link: https://lore.kernel.org/r/20230203150400.3199230-15-hch@lst.de
Signed-off-by: Jens Axboe <axboe@kernel.dk>
Switch to passing a gendisk, and make rq_qos_add initialize all required
fields and drop the not required q argument from rq_qos_del.
Signed-off-by: Christoph Hellwig <hch@lst.de>
Reviewed-by: Andreas Herrmann <aherrmann@suse.de>
Acked-by: Tejun Heo <tj@kernel.org>
Link: https://lore.kernel.org/r/20230203150400.3199230-14-hch@lst.de
Signed-off-by: Jens Axboe <axboe@kernel.dk>
These two functions are rather larger and not in a fast path, so move
them out of line.
Signed-off-by: Christoph Hellwig <hch@lst.de>
Acked-by: Tejun Heo <tj@kernel.org>
Link: https://lore.kernel.org/r/20230203150400.3199230-13-hch@lst.de
Signed-off-by: Jens Axboe <axboe@kernel.dk>
Use atomic_try_cmpxchg instead of atomic_cmpxchg (*ptr, old, new) == old in
atomic_inc_below. x86 CMPXCHG instruction returns success in ZF flag,
so this change saves a compare after cmpxchg (and related move instruction
in front of cmpxchg).
Also, atomic_try_cmpxchg implicitly assigns old *ptr value to "old" when
cmpxchg fails, enabling further code simplifications.
No functional change intended.
Signed-off-by: Uros Bizjak <ubizjak@gmail.com>
Cc: Jens Axboe <axboe@kernel.dk>
Link: https://lore.kernel.org/r/20220712150547.5786-1-ubizjak@gmail.com
Signed-off-by: Jens Axboe <axboe@kernel.dk>
The block debugfs files are created in blk_register_queue, which is
called by add_disk and use a naming scheme based on the disk_name.
After del_gendisk returns that name can be reused and thus we must not
leave these debugfs files around, otherwise the kernel is unhappy
and spews messages like:
Directory XXXXX with parent 'block' already present!
and the newly created devices will not have working debugfs files.
Move the unregistration to blk_unregister_queue instead (which matches
the sysfs unregistration) to make sure the debugfs life time rules match
those of the disk name.
As part of the move also make sure the whole debugfs unregistration is
inside a single debugfs_mutex critical section.
Note that this breaks blktests block/002, which checks that the debugfs
directory has not been removed while blktests is running, but that
particular check should simply be removed from the test case.
Signed-off-by: Christoph Hellwig <hch@lst.de>
Link: https://lore.kernel.org/r/20220614074827.458955-4-hch@lst.de
Signed-off-by: Jens Axboe <axboe@kernel.dk>
Various places like I/O schedulers or the QOS infrastructure try to
register debugfs files on demans, which can race with creating and
removing the main queue debugfs directory. Use the existing
debugfs_mutex to serialize all debugfs operations that rely on
q->debugfs_dir or the directories hanging off it.
To make the teardown code a little simpler declare all debugfs dentry
pointers and not just the main one uncoditionally in blkdev.h.
Move debugfs_mutex next to the dentries that it protects and document
what it is used for.
Signed-off-by: Christoph Hellwig <hch@lst.de>
Link: https://lore.kernel.org/r/20220614074827.458955-3-hch@lst.de
Signed-off-by: Jens Axboe <axboe@kernel.dk>
Commit 545fbd0775 ("rq-qos: fix missed wake-ups in rq_qos_throttle")
tried to fix a problem that a process could be sleeping in rq_qos_wait()
without anyone to wake it up. However the fix is not complete and the
following can still happen:
CPU1 (waiter1) CPU2 (waiter2) CPU3 (waker)
rq_qos_wait() rq_qos_wait()
acquire_inflight_cb() -> fails
acquire_inflight_cb() -> fails
completes IOs, inflight
decreased
prepare_to_wait_exclusive()
prepare_to_wait_exclusive()
has_sleeper = !wq_has_single_sleeper() -> true as there are two sleepers
has_sleeper = !wq_has_single_sleeper() -> true
io_schedule() io_schedule()
Deadlock as now there's nobody to wakeup the two waiters. The logic
automatically blocking when there are already sleepers is really subtle
and the only way to make it work reliably is that we check whether there
are some waiters in the queue when adding ourselves there. That way, we
are guaranteed that at least the first process to enter the wait queue
will recheck the waiting condition before going to sleep and thus
guarantee forward progress.
Fixes: 545fbd0775 ("rq-qos: fix missed wake-ups in rq_qos_throttle")
CC: stable@vger.kernel.org
Signed-off-by: Jan Kara <jack@suse.cz>
Link: https://lore.kernel.org/r/20210607112613.25344-1-jack@suse.cz
Signed-off-by: Jens Axboe <axboe@kernel.dk>
scale_up wakes up waiters after scaling up. But after scaling max, it
should not wake up more waiters as waiters will not have anything to
do. This patch fixes this by making scale_up (and also scale_down)
return when threshold is reached.
This bug causes increased fdatasync latency when fdatasync and dd
conv=sync are performed in parallel on 4.19 compared to 4.14. This
bug was introduced during refactoring of blk-wbt code.
Fixes: a79050434b ("blk-rq-qos: refactor out common elements of blk-wbt")
Cc: stable@vger.kernel.org
Cc: Josef Bacik <jbacik@fb.com>
Signed-off-by: Harshad Shirwadkar <harshadshirwadkar@gmail.com>
Signed-off-by: Jens Axboe <axboe@kernel.dk>
wbt already gets queue depth changed notification through
wbt_set_queue_depth(). Generalize it into
rq_qos_ops->queue_depth_changed() so that other rq_qos policies can
easily hook into the events too.
Signed-off-by: Tejun Heo <tj@kernel.org>
Signed-off-by: Jens Axboe <axboe@kernel.dk>
Oleg noticed that our checking of data.got_token is unsafe in the
cleanup case, and should really use a memory barrier. Use a wmb on the
write side, and a rmb() on the read side. We don't need one in the main
loop since we're saved by set_current_state().
Reviewed-by: Oleg Nesterov <oleg@redhat.com>
Signed-off-by: Josef Bacik <josef@toxicpanda.com>
Signed-off-by: Jens Axboe <axboe@kernel.dk>
In case we get a spurious wakeup we need to make sure to re-set
ourselves to TASK_UNINTERRUPTIBLE so we don't busy wait.
Reviewed-by: Oleg Nesterov <oleg@redhat.com>
Signed-off-by: Josef Bacik <josef@toxicpanda.com>
Signed-off-by: Jens Axboe <axboe@kernel.dk>
If we raced with somebody else getting an inflight counter we could fail
to get an inflight counter with no sleepers on the list, and thus need
to go to sleep. In this case has_sleepers should be true because we are
now relying on the waker to get our inflight counter for us. And in the
case of spurious wakeups we'd still want this to be the case. So set
has_sleepers to true if we went to sleep to make sure we're woken up the
proper way.
Reviewed-by: Oleg Nesterov <oleg@redhat.com>
Signed-off-by: Josef Bacik <josef@toxicpanda.com>
Signed-off-by: Jens Axboe <axboe@kernel.dk>
We saw a hang in production with WBT where there was only one waiter in
the throttle path and no outstanding IO. This is because of the
has_sleepers optimization that is used to make sure we don't steal an
inflight counter for new submitters when there are people already on the
list.
We can race with our check to see if the waitqueue has any waiters (this
is done locklessly) and the time we actually add ourselves to the
waitqueue. If this happens we'll go to sleep and never be woken up
because nobody is doing IO to wake us up.
Fix this by checking if the waitqueue has a single sleeper on the list
after we add ourselves, that way we have an uptodate view of the list.
Reviewed-by: Oleg Nesterov <oleg@redhat.com>
Signed-off-by: Josef Bacik <josef@toxicpanda.com>
Signed-off-by: Jens Axboe <axboe@kernel.dk>
Various block layer files do not have any licensing information at all.
Add SPDX tags for the default kernel GPLv2 license to those.
Reviewed-by: Chaitanya Kulkarni <chaitanya.kulkarni@wdc.com>
Signed-off-by: Christoph Hellwig <hch@lst.de>
Signed-off-by: Jens Axboe <axboe@kernel.dk>
blk-mq-debugfs has been proved as very helpful for debug some
tough issues, such as IO hang.
We have seen blk-wbt related IO hang several times, even inside
Red Hat BZ, there is such report not sovled yet, so this patch
adds support debugfs on rq_qos.
Cc: Bart Van Assche <bart.vanassche@wdc.com>
Cc: Omar Sandoval <osandov@fb.com>
Cc: Christoph Hellwig <hch@lst.de>
Cc: Josef Bacik <josef@toxicpanda.com>
Signed-off-by: Ming Lei <ming.lei@redhat.com>
Signed-off-by: Jens Axboe <axboe@kernel.dk>
Originally when I split out the common code from blk-wbt into rq_qos I
left the wbt_wait() where it was and simply copied and modified it
slightly to work for io-latency. However they are both basically the
same thing, and as time has gone on wbt_wait() has ended up much smarter
and kinder than it was when I copied it into io-latency, which means
io-latency has lost out on these improvements.
Since they are the same thing essentially except for a few minor things,
create rq_qos_wait() that replicates what wbt_wait() currently does with
callbacks that can be passed in for the snowflakes to do their own thing
as appropriate.
Signed-off-by: Josef Bacik <josef@toxicpanda.com>
Signed-off-by: Jens Axboe <axboe@kernel.dk>
Put the short code in the fast path, where we don't have any
functions attached to the queue. This minimizes the impact on
the hot path in the core code.
Cc: Josef Bacik <josef@toxicpanda.com>
Reviewed-by: Christoph Hellwig <hch@lst.de>
Signed-off-by: Jens Axboe <axboe@kernel.dk>
Unused now that the legacy request path is gone.
Reviewed-by: Hannes Reinecke <hare@suse.com>
Signed-off-by: Christoph Hellwig <hch@lst.de>
Signed-off-by: Jens Axboe <axboe@kernel.dk>
With the change to use UINT_MAX I broke the depth check as any value of
inflight (ie 0) would be less than (int)UINT_MAX. Fix this by changing
everything to unsigned int to match the depth.
Signed-off-by: Josef Bacik <josef@toxicpanda.com>
Signed-off-by: Jens Axboe <axboe@kernel.dk>
wbt cares only about request completion time, but controllers may need
information that is on the bio itself, so add a done_bio callback for
rq-qos so things like blk-iolatency can use it to have the bio when it
completes.
Signed-off-by: Josef Bacik <jbacik@fb.com>
Signed-off-by: Jens Axboe <axboe@kernel.dk>
We don't really need to save this stuff in the core block code, we can
just pass the bio back into the helpers later on to derive the same
flags and update the rq->wbt_flags appropriately.
Signed-off-by: Josef Bacik <jbacik@fb.com>
Signed-off-by: Jens Axboe <axboe@kernel.dk>
blkcg-qos is going to do essentially what wbt does, only on a cgroup
basis. Break out the common code that will be shared between blkcg-qos
and wbt into blk-rq-qos.* so they can both utilize the same
infrastructure.
Signed-off-by: Josef Bacik <jbacik@fb.com>
Signed-off-by: Jens Axboe <axboe@kernel.dk>
