Lars Ellenberg [Wed, 24 Nov 2010 09:11:14 +0000 (10:11 +0100)]
drbd: add packet_type 27 (return_code_only) to netlink api
In case we ever should add an other packet type,
we must not reuse 27, as that currently used for
"empty" return code only replies.
Document it as such.
Signed-off-by: Philipp Reisner <philipp.reisner@linbit.com> Signed-off-by: Lars Ellenberg <lars.ellenberg@linbit.com>
Lars Ellenberg [Wed, 24 Nov 2010 09:37:35 +0000 (10:37 +0100)]
drbd: remove /proc/drbd before unregistering from netlink
There still exists a (theoretical) race on module unload, where
/proc/drbd may still exist, but the netlink callback has been
unregistered already, allowing drbdsetup to shout without listeners,
and get no reply.
Reorder remove_proc_entry and unregister of netlink callback.
drbdsetup first checks for existence of the proc entry,
and if that is missing, won't even try to contact the module.
Signed-off-by: Philipp Reisner <philipp.reisner@linbit.com> Signed-off-by: Lars Ellenberg <lars.ellenberg@linbit.com>
Lars Ellenberg [Wed, 24 Nov 2010 09:33:02 +0000 (10:33 +0100)]
drbd: increase module count on /proc/drbd access
If someone holds /proc/drbd open, previously rmmod would
"succeed" in starting the unload, but then block on remove_proc_entry,
leading to a situation where the lsmod does not show drbd anymore,
but /proc/drbd being still there (but no longer accessible).
I'd rather have rmmod fail up front in this case.
Signed-off-by: Philipp Reisner <philipp.reisner@linbit.com> Signed-off-by: Lars Ellenberg <lars.ellenberg@linbit.com>
Philipp Reisner [Mon, 22 Nov 2010 14:49:17 +0000 (15:49 +0100)]
drbd: Removed 20 seconds upper bound for side-stepping
Given low-enough network bandwidth combined with a IO
pattern that hammers onto a single RS-extent, side-stepping
might be necessary for much longer times.
Changed the code to print a single informal message after
20 seconds, but it keeps on stepping aside forever.
Signed-off-by: Philipp Reisner <philipp.reisner@linbit.com> Signed-off-by: Lars Ellenberg <lars.ellenberg@linbit.com>
Philipp Reisner [Wed, 10 Nov 2010 11:08:37 +0000 (12:08 +0100)]
drbd: Starting with protocol 96 we can allow app-IO while receiving the bitmap
* C_STARTING_SYNC_S, C_STARTING_SYNC_T In these states the bitmap gets
written to disk. Locking out of app-IO is done by using the
drbd_queue_bitmap_io() and drbd_bitmap_io() functions these days.
It is no longer necessary to lock out app-IO based on the connection
state.
App-IO that may come in after the BITMAP_IO flag got cleared before the
state transition to C_SYNC_(SOURCE|TARGET) does not get mirrored, sets
a bit in the local bitmap, that is already set, therefore changes nothing.
* C_WF_BITMAP_S In this state we send updates (P_OUT_OF_SYNC packets).
With that we make sure they have the same number of bits when going
into the C_SYNC_(SOURCE|TARGET) connection state.
* C_UNCONNECTED: The receiver starts, no need to lock out IO.
* C_DISCONNECTING: in drbd_disconnect() we had a wait_event()
to wait until ap_bio_cnt reaches 0. Removed that.
* C_TIMEOUT, C_BROKEN_PIPE, C_NETWORK_FAILURE
C_PROTOCOL_ERROR, C_TEAR_DOWN: Same as C_DISCONNECTING
* C_WF_REPORT_PARAMS: IO still possible since that is still
like C_WF_CONNECTION.
And we do not need to send barriers in C_WF_BITMAP_S connection state.
Allow concurrent accesses to the bitmap when receiving the bitmap.
Everything gets ORed anyways.
A drbd_free_tl_hash() is in after_state_chg_work(). At that point
all the work items of the last connections must have been processed.
Introduced a call to drbd_free_tl_hash() into drbd_free_mdev()
for paranoia reasons.
Signed-off-by: Philipp Reisner <philipp.reisner@linbit.com> Signed-off-by: Lars Ellenberg <lars.ellenberg@linbit.com>
Philipp Reisner [Wed, 17 Nov 2010 15:54:36 +0000 (16:54 +0100)]
drbd: Improvements in sanitize_state()
The relevant change is that the state change to C_FW_BITMAP_S should
implicitly change pdsk to C_CONSISTENT. (Think of it as C_OUTDATED, only
without the guarantee that the peer has the outdated written to its
meta data)
At that opportunity I restructured the switch statement so that it
gets evaluated every time. (Has declarative character)
Signed-off-by: Philipp Reisner <philipp.reisner@linbit.com> Signed-off-by: Lars Ellenberg <lars.ellenberg@linbit.com>
Philipp Reisner [Tue, 9 Nov 2010 12:59:41 +0000 (13:59 +0100)]
drbd: Implemented side-stepping in drbd_res_begin_io()
Before:
drbd_rs_begin_io() locked app-IO out of an RS extent, and
waited then until all previous app-IO in that area finished.
(But not only until the disk-IO was finished but until the
barrier/epoch ack came in for that == round trip time latency ++)
After:
As soon as a new app-IO waits wants to start new IO on that
RS extent, drbd_rs_begin_io() steps aside (clearing the
BME_NO_WRITES flag again). It retries after 100ms.
Signed-off-by: Philipp Reisner <philipp.reisner@linbit.com> Signed-off-by: Lars Ellenberg <lars.ellenberg@linbit.com>
Philipp Reisner [Sun, 7 Nov 2010 14:56:29 +0000 (15:56 +0100)]
drbd: Implemented priority inheritance for resync requests
We only issue resync requests if there is no significant application IO
going on. = Application IO has higher priority than resnyc IO.
If application IO can not be started because the resync process locked
an resync_lru entry, start the IO operations necessary to release the
lock ASAP.
Signed-off-by: Philipp Reisner <philipp.reisner@linbit.com> Signed-off-by: Lars Ellenberg <lars.ellenberg@linbit.com>
Lars Ellenberg [Thu, 11 Nov 2010 21:41:04 +0000 (22:41 +0100)]
drbd: restore compatibility with 32bit kernels
With commit
drbd: further converge progress display of resync and online-verify
accidentally an u64/u64 div was introduced, causing an unresolvable
symbol __udivdi3 to be reference. Actually for that division, 32bit are
still suficient for now, so we can revert to unsigned long instead.
Signed-off-by: Philipp Reisner <philipp.reisner@linbit.com> Signed-off-by: Lars Ellenberg <lars.ellenberg@linbit.com>
Lars Ellenberg [Thu, 11 Nov 2010 14:19:07 +0000 (15:19 +0100)]
drbd: properly use max_hw_sectors to limit the our bio size
To ease tracking of bios in some hash tables, we want it to
not cross certain boundaries (128k, used to be 32k).
We limit the maximum bio size using queue parameters.
Historically some defines and variables we use there have been named
max_segment_size, which was misguided. Rename them to max_bio_size,
and use [blk_]queue_max_hw_sectors where appropriate.
Signed-off-by: Philipp Reisner <philipp.reisner@linbit.com> Signed-off-by: Lars Ellenberg <lars.ellenberg@linbit.com>
Lars Ellenberg [Wed, 10 Nov 2010 09:36:52 +0000 (10:36 +0100)]
drbd: detect modification of in-flight buffers
With data-integrity digest enabled, double-check on the sending side
for modifications by upper layers of buffers under write back,
so we can tell it appart from corruption on the "wire".
Signed-off-by: Philipp Reisner <philipp.reisner@linbit.com> Signed-off-by: Lars Ellenberg <lars.ellenberg@linbit.com>
Lars Ellenberg [Tue, 9 Nov 2010 13:12:10 +0000 (14:12 +0100)]
drbd: fix potential wrap of 32bit oos:%lu display in /proc/drbd
When converting bits (4k resolution, still) to kB, we shift left. If it
was a large number of bits on a 32bit box (>= 4 TiB storage), we may
wrap the 32bit unsigned long base type, resulting in incorrect display.
Signed-off-by: Philipp Reisner <philipp.reisner@linbit.com> Signed-off-by: Lars Ellenberg <lars.ellenberg@linbit.com>
Lars Ellenberg [Fri, 5 Nov 2010 08:39:06 +0000 (09:39 +0100)]
drbd: improve online-verify progress tracking
For a partial (resumed) online-verify, initialize rs_total not to total
bits, but to number of bits to check in this run, to match the meaning
rs_total has for actual resync.
Signed-off-by: Philipp Reisner <philipp.reisner@linbit.com> Signed-off-by: Lars Ellenberg <lars.ellenberg@linbit.com>
Lars Ellenberg [Fri, 5 Nov 2010 08:56:33 +0000 (09:56 +0100)]
drbd: only reset online-verify start sector if verify completed
For network hickups during online-verify, on the next verify
triggered, we by default want to resume where it left off.
After any replication link interruption, there will be a (possibly
empty) resync. Do not reset online-verify start sector if some resync
completed, that would defeats the purpose.
Only reset the start sector once a verify run is completed.
Signed-off-by: Philipp Reisner <philipp.reisner@linbit.com> Signed-off-by: Lars Ellenberg <lars.ellenberg@linbit.com>
Jens Axboe [Wed, 9 Mar 2011 10:56:30 +0000 (11:56 +0100)]
block: kill off REQ_UNPLUG
With the plugging now being explicitly controlled by the
submitter, callers need not pass down unplugging hints
to the block layer. If they want to unplug, it's because they
manually plugged on their own - in which case, they should just
unplug at will.
Jens Axboe [Thu, 10 Mar 2011 07:52:07 +0000 (08:52 +0100)]
block: remove per-queue plugging
Code has been converted over to the new explicit on-stack plugging,
and delay users have been converted to use the new API for that.
So lets kill off the old plugging along with aops->sync_page().
Jens Axboe [Tue, 8 Mar 2011 12:19:51 +0000 (13:19 +0100)]
block: initial patch for on-stack per-task plugging
This patch adds support for creating a queuing context outside
of the queue itself. This enables us to batch up pieces of IO
before grabbing the block device queue lock and submitting them to
the IO scheduler.
The context is created on the stack of the process and assigned in
the task structure, so that we can auto-unplug it if we hit a schedule
event.
The current queue plugging happens implicitly if IO is submitted to
an empty device, yet callers have to remember to unplug that IO when
they are going to wait for it. This is an ugly API and has caused bugs
in the past. Additionally, it requires hacks in the vm (->sync_page()
callback) to handle that logic. By switching to an explicit plugging
scheme we make the API a lot nicer and can get rid of the ->sync_page()
hack in the vm.
It was always abuse to reuse the plugging infrastructure for this,
convert it to the (new) real API for delaying queueing a bit. A
default delay of 3 msec is defined, to match the previous
behaviour.
Tejun Heo [Wed, 9 Mar 2011 18:54:29 +0000 (19:54 +0100)]
staging: Convert to bdops->check_events()
Convert two staging drivers - blkvsc_drv and cyasblkdev_block - from
->media_changed() to ->check_events(). The former always indicated
media changed while the latter always indicated media not changed.
Not sure what the drivers are trying to achieve but keep the original
behavior.
Tejun Heo [Wed, 9 Mar 2011 18:54:28 +0000 (19:54 +0100)]
pktcdvd: Convert to bdops->check_events()
Convert from ->media_changed() to ->check_events().
pktcdvd needs to forward all event related operations to the
underlying device. Forward ->check_events() instead of
->media_changed() and inherit disk->[async_]events.
Signed-off-by: Tejun Heo <tj@kernel.org> Cc: Jens Axboe <axboe@kernel.dk> Cc: Kay Sievers <kay.sievers@vrfy.org> Cc: Peter Osterlund <petero2@telia.com>
Tejun Heo [Wed, 9 Mar 2011 18:54:28 +0000 (19:54 +0100)]
dac960: Convert to bdops->check_events()
Convert from ->media_changed() to ->check_events().
DAC960 media change notification seems to be one way (once set, never
cleared) and will generate spurious events when polled once the
condition triggers.
Tejun Heo [Wed, 9 Mar 2011 18:54:27 +0000 (19:54 +0100)]
floppy,{ami|ata}flop: Convert to bdops->check_events()
Convert the floppy drivers from ->media_changed() to ->check_events().
Both floppy and ataflop buffer media changed state bit and clear them
on revalidation and will behave correctly with kernel event polling.
I can't tell how amiflop clears its event and it's possible that it
may generate spurious events when polled.
Tejun Heo [Wed, 9 Mar 2011 18:54:27 +0000 (19:54 +0100)]
ide: Convert to bdops->check_events()
Convert ->media_changed() to the new ->check_events() method. The
conversion is mostly mechanical. The only notable change is that
cdrom now doesn't generate any event if @slot_nr isn't CDSL_CURRENT.
It used to return -EINVAL which would be treated as media changed. As
media changer isn't supported anyway, this doesn't make any
difference.
This makes ide emit the standard disk events and allows kernel event
polling. Currently, only MEDIA_CHANGE event is implemented. Adding
support for EJECT_REQUEST shouldn't be difficult; however, given that
ide driver is already deprecated, it probably is best to leave it
alone.
Signed-off-by: Tejun Heo <tj@kernel.org> Acked-by: Jens Axboe <axboe@kernel.dk> Cc: Kay Sievers <kay.sievers@vrfy.org> Cc: "David S. Miller" <davem@davemloft.net> Cc: linux-ide@vger.kernel.org
Tejun Heo [Wed, 9 Mar 2011 18:54:27 +0000 (19:54 +0100)]
block: Don't check events while open is in progress
Not all block drivers clear events immediately after reporting. Some
do so in ->revalidate_disk() or other steps during ->open(). There is
a slim chance event poll may happen between the clearing event check
from check_disk_change() and the actual clearing of the events which
would result in spurious events.
Block event checks while block device open is in progress. There is
no need to kick explicit event check afterwards as events are always
checked during open.
-v2: The original patch could have called disk_unblock_events() with
an already released or %NULL @disk causing oops. Fixed by making
sure references are put after disk_unblock_events() is called.
It also makes the error path of __blkdev_get() a bit simpler.
This problem was reported by Jens.
Tejun Heo [Wed, 9 Mar 2011 18:54:27 +0000 (19:54 +0100)]
block: Don't check events on close unless it was blocked
The block event mechanism currently always checks events when the
device is being closed regardless of the open mode. The intention was
to allow detection of EJECT_REQUEST when a device is closed whether
disk event polling is enabled or not.
This is unnecessary as, for devices of interest, events are checked
from either userland or kernel and in the former case ->check_events()
is performed on open of each poll attempt anyway. Furthermore, this
unconditional event check on close makes the code susceptible to event
loop if the block driver doesn't clear reported events correctly - an
event triggers userland to open and close the device which in turn
causes another event, rinse and repeat.
Check events on close only if it was blocked by excl write open.
Tejun Heo [Wed, 9 Mar 2011 18:54:27 +0000 (19:54 +0100)]
block: Don't implicitly trigger event check on disk_unblock_events()
Currently, disk_unblock_events() implicitly kick event check if the
block count reaches zero. This behavior is not described in the
comment and hinders with future changes. Make the unblocker
explicitly check events by calling disk_check_events() as necessary.
Justin TerAvest [Tue, 8 Mar 2011 18:45:00 +0000 (19:45 +0100)]
blk-cgroup: Lower minimum weight from 100 to 10.
We've found that we still get good, useful isolation at weights this
low. I'd like to adjust the minimum so that any other changes can take
these values into account.
Vivek Goyal [Mon, 7 Mar 2011 20:09:32 +0000 (21:09 +0100)]
blk-throttle: Some cleanups and race fixes in limit update code
When throttle group limits are updated through cgroups, a thread is
woken up to process these updates. While reviewing that code, oleg noted
couple of race conditions existed in the code and he also suggested that
code can be simplified.
This patch fixes the races simplifies the code based on Oleg's suggestions:
- Use xchg().
- Introduced a common function throtl_update_blkio_group_common()
which is shared now by all iops/bps update functions.
Reviewed-by: Oleg Nesterov <oleg@redhat.com> Reviewed-by: Paul E. McKenney <paulmck@linux.vnet.ibm.com> Signed-off-by: Vivek Goyal <vgoyal@redhat.com>
Fixed a merge issue, throtl_schedule_delayed_work() takes throtl_data
as the argument now, not the queue.
Vivek Goyal [Mon, 7 Mar 2011 20:05:14 +0000 (21:05 +0100)]
blk-throttle: process limit change only through one function
With the help of cgroup interface one can go and upate the bps/iops
limits of existing group. Once the limits are udpated, a thread is
woken up to see if some blocked group needs recalculation based on new
limits and needs to be requeued.
There was also a piece of code where I was checking for group limit
update when a fresh bio comes in. This patch gets rid of that piece of
code and keeps processing the limit change at one place
throtl_process_limit_change(). It just keeps the code simple and easy
to understand.
Shaohua Li [Mon, 7 Mar 2011 08:26:29 +0000 (09:26 +0100)]
cfq-iosched: give busy sync queue no dispatch limit
If there are a sync and an async queue and the sync queue's think time
is small, we can ignore the sync queue's dispatch quantum. Because the
sync queue will always preempt the async queue, we don't need to care
about async's latency. This can fix a performance regression of
aiostress test, which is introduced by commit f8ae6e3eb825. The issue
should exist even without the commit, but the commit amplifies the
impact.
The initial post does the same optimization for RT queue too, but since
I have no real workload for it, Vivek suggests to drop it.
Tejun Heo [Fri, 4 Mar 2011 18:09:02 +0000 (19:09 +0100)]
Merge branch 'for-linus' of ../linux-2.6-block into block-for-2.6.39/core
This merge creates two set of conflicts. One is simple context
conflicts caused by removal of throtl_scheduled_delayed_work() in
for-linus and removal of throtl_shutdown_timer_wq() in
for-2.6.39/core.
The other is caused by commit 255bb490c8 (block: blk-flush shouldn't
call directly into q->request_fn() __blk_run_queue()) in for-linus
crashing with FLUSH reimplementation in for-2.6.39/core. The conflict
isn't trivial but the resolution is straight-forward.
* __blk_run_queue() calls in flush_end_io() and flush_data_end_io()
should be called with @force_kblockd set to %true.
* elv_insert() in blk_kick_flush() should use
%ELEVATOR_INSERT_REQUEUE.
Both changes are to avoid invoking ->request_fn() directly from
request completion path and closely match the changes in the commit 255bb490c8.
Tao Ma [Thu, 3 Mar 2011 15:53:20 +0000 (10:53 -0500)]
blktrace: Remove blk_fill_rwbs_rq.
If we enable trace events to trace block actions, We use
blk_fill_rwbs_rq to analyze the corresponding actions
in request's cmd_flags, but we only choose the minor 2 bits
from it, so most of other flags(e.g, REQ_SYNC) are missing.
For example, with a sync write we get:
write_test-2409 [001] 160.013869: block_rq_insert: 3,64 W 0 () 258135 + =
8 [write_test]
Since now we have integrated the flags of both bio and request,
it is safe to pass rq->cmd_flags directly to blk_fill_rwbs and
blk_fill_rwbs_rq isn't needed any more.
With this patch, after a sync write we get:
write_test-2417 [000] 226.603878: block_rq_insert: 3,64 WS 0 () 258135 +=
8 [write_test]
Signed-off-by: Tao Ma <boyu.mt@taobao.com> Acked-by: Jeff Moyer <jmoyer@redhat.com> Signed-off-by: Jens Axboe <jaxboe@fusionio.com>
Vivek Goyal [Thu, 3 Mar 2011 00:05:33 +0000 (19:05 -0500)]
block: Move blk_throtl_exit() call to blk_cleanup_queue()
Move blk_throtl_exit() in blk_cleanup_queue() as blk_throtl_exit() is
written in such a way that it needs queue lock. In blk_release_queue()
there is no gurantee that ->queue_lock is still around.
Initially blk_throtl_exit() was in blk_cleanup_queue() but Ingo reported
one problem.
https://lkml.org/lkml/2010/10/23/86
And a quick fix moved blk_throtl_exit() to blk_release_queue().
block: fix use-after-free bug in blk throttle code
This patch reverts above change and does not try to shutdown the
throtl work in blk_sync_queue(). By avoiding call to
throtl_shutdown_timer_wq() from blk_sync_queue(), we should also avoid
the problem reported by Ingo.
blk_sync_queue() seems to be used only by md driver and it seems to be
using it to make sure q->unplug_fn is not called as md registers its
own unplug functions and it is about to free up the data structures
used by unplug_fn(). Block throttle does not call back into unplug_fn()
or into md. So there is no need to cancel blk throttle work.
In fact I think cancelling block throttle work is bad because it might
happen that some bios are throttled and scheduled to be dispatched later
with the help of pending work and if work is cancelled, these bios might
never be dispatched.
Block layer also uses blk_sync_queue() during blk_cleanup_queue() and
blk_release_queue() time. That should be safe as we are also calling
blk_throtl_exit() which should make sure all the throttling related
data structures are cleaned up.
Vivek Goyal [Thu, 3 Mar 2011 00:04:42 +0000 (19:04 -0500)]
block: Initialize ->queue_lock to internal lock at queue allocation time
There does not seem to be a clear convention whether q->queue_lock is
initialized or not when blk_cleanup_queue() is called. In the past it
was not necessary but now blk_throtl_exit() takes up queue lock by
default and needs queue lock to be available.
In fact elevator_exit() code also has similar requirement just that it
is less stringent in the sense that elevator_exit() is called only if
elevator is initialized.
Two problems have been noticed because of ambiguity about spin lock
status.
- If a driver calls blk_alloc_queue() and then soon calls
blk_cleanup_queue() almost immediately, (because some other
driver structure allocation failed or some other error happened)
then blk_throtl_exit() will run into issues as queue lock is not
initialized. Loop driver ran into this issue recently and I
noticed error paths in md driver too. Similar error paths should
exist in other drivers too.
- If some driver provided external spin lock and zapped the lock
before blk_cleanup_queue(), then it can lead to issues.
So this patch initializes the default queue lock at queue allocation time.
block throttling code is one of the users of queue lock and it is
initialized at the queue allocation time, so it makes sense to
initialize ->queue_lock also to internal lock. A driver can overide that
lock later. This will take care of the issue where a driver does not have
to worry about initializing the queue lock to default before calling
blk_cleanup_queue()
Tejun Heo [Wed, 2 Mar 2011 13:48:06 +0000 (08:48 -0500)]
block: blk-flush shouldn't call directly into q->request_fn() __blk_run_queue()
blk-flush decomposes a flush into sequence of multiple requests. On
completion of a request, the next one is queued; however, block layer
must not implicitly call into q->request_fn() directly from completion
path. This makes the queue behave unexpectedly when seen from the
drivers and violates the assumption that q->request_fn() is called
with process context + queue_lock.
This patch makes blk-flush the following two changes to make sure
q->request_fn() is not called directly from request completion path.
- blk_flush_complete_seq_end_io() now asks __blk_run_queue() to always
use kblockd instead of calling directly into q->request_fn().
- queue_next_fseq() uses ELEVATOR_INSERT_REQUEUE instead of
ELEVATOR_INSERT_FRONT so that elv_insert() doesn't try to unplug the
request queue directly.
Tejun Heo [Wed, 2 Mar 2011 13:48:05 +0000 (08:48 -0500)]
block: add @force_kblockd to __blk_run_queue()
__blk_run_queue() automatically either calls q->request_fn() directly
or schedules kblockd depending on whether the function is recursed.
blk-flush implementation needs to be able to explicitly choose
kblockd. Add @force_kblockd.
All the current users are converted to specify %false for the
parameter and this patch doesn't introduce any behavior change.
stable: This is prerequisite for fixing ide oops caused by the new
blk-flush implementation.
Signed-off-by: Tejun Heo <tj@kernel.org> Cc: Jan Beulich <JBeulich@novell.com> Cc: James Bottomley <James.Bottomley@HansenPartnership.com> Cc: stable@kernel.org Signed-off-by: Jens Axboe <jaxboe@fusionio.com>
Justin TerAvest [Tue, 1 Mar 2011 20:05:08 +0000 (15:05 -0500)]
cfq-iosched: Always provide group isolation.
Effectively, make group_isolation=1 the default and remove the tunable.
The setting group_isolation=0 was because by default we idle on
sync-noidle tree and on fast devices, this can be very harmful for
throughput.
However, this problem can also be addressed by tuning slice_idle and
possibly group_idle on faster storage devices.
This change simplifies the CFQ code by removing the feature entirely.
Vivek Goyal [Tue, 1 Mar 2011 18:40:54 +0000 (13:40 -0500)]
blk-throttle: Do not use kblockd workqueue for throtl work
o Dominik Klein reported a system hang issue while doing some blkio
throttling testing.
https://lkml.org/lkml/2011/2/24/173
o Some tracing revealed that CFQ was not dispatching any more jobs as
queue unplug was not happening. And queue unplug was not happening
because unplug work was not being called as there was one throttling
work on same cpu which as not finished yet. And throttling work had not
finished as it was tyring to dispatch a bio to CFQ but all the request
descriptors were consume to it was put to sleep.
o So basically it is a cyclic dependecny between CFQ unplug work and
throtl dispatch work. Tejun suggested that use separate workqueue for
such cases.
o This patch uses a separate workqueue for throttle related work and
does not rely on kblockd workqueue anymore.
projects / linux-beck.git / log