In our test of iocost, we encountered some list add/del corruptions of
inner_walk list in ioc_timer_fn.
The reason can be described as follows:
cpu 0 cpu 1
ioc_qos_write ioc_qos_write
ioc = q_to_ioc(queue);
if (!ioc) {
ioc = kzalloc();
ioc = q_to_ioc(queue);
if (!ioc) {
ioc = kzalloc();
...
rq_qos_add(q, rqos);
}
...
rq_qos_add(q, rqos);
...
}
When the io.cost.qos file is written by two cpus concurrently, rq_qos may
be added to one disk twice. In that case, there will be two iocs enabled
and running on one disk. They own different iocgs on their active list. In
the ioc_timer_fn function, because of the iocgs from two iocs have the
same root iocg, the root iocg's walk_list may be overwritten by each other
and this leads to list add/del corruptions in building or destroying the
inner_walk list.
And so far, the blk-rq-qos framework works in case that one instance for
one type rq_qos per queue by default. This patch make this explicit and
also fix the crash above.
Signed-off-by: Jinke Han <hanjinke.666@bytedance.com>
Reviewed-by: Muchun Song <songmuchun@bytedance.com>
Acked-by: Tejun Heo <tj@kernel.org>
Cc: <stable@vger.kernel.org>
Link: https://lore.kernel.org/r/20220720093616.70584-1-hanjinke.666@bytedance.com
Signed-off-by: Jens Axboe <axboe@kernel.dk>
-----BEGIN PGP SIGNATURE-----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=0Rw1
-----END PGP SIGNATURE-----
Merge tag 'for-5.19/block-2022-05-22' of git://git.kernel.dk/linux-block
Pull block updates from Jens Axboe:
"Here are the core block changes for 5.19. This contains:
- blk-throttle accounting fix (Laibin)
- Series removing redundant assignments (Michal)
- Expose bio cache via the bio_set, so that DM can use it (Mike)
- Finish off the bio allocation interface cleanups by dealing with
the weirdest member of the family. bio_kmalloc combines a kmalloc
for the bio and bio_vecs with a hidden bio_init call and magic
cleanup semantics (Christoph)
- Clean up the block layer API so that APIs consumed by file systems
are (almost) only struct block_device based, so that file systems
don't have to poke into block layer internals like the
request_queue (Christoph)
- Clean up the blk_execute_rq* API (Christoph)
- Clean up various lose end in the blk-cgroup code to make it easier
to follow in preparation of reworking the blkcg assignment for bios
(Christoph)
- Fix use-after-free issues in BFQ when processes with merged queues
get moved to different cgroups (Jan)
- BFQ fixes (Jan)
- Various fixes and cleanups (Bart, Chengming, Fanjun, Julia, Ming,
Wolfgang, me)"
* tag 'for-5.19/block-2022-05-22' of git://git.kernel.dk/linux-block: (83 commits)
blk-mq: fix typo in comment
bfq: Remove bfq_requeue_request_body()
bfq: Remove superfluous conversion from RQ_BIC()
bfq: Allow current waker to defend against a tentative one
bfq: Relax waker detection for shared queues
blk-cgroup: delete rcu_read_lock_held() WARN_ON_ONCE()
blk-throttle: Set BIO_THROTTLED when bio has been throttled
blk-cgroup: Remove unnecessary rcu_read_lock/unlock()
blk-cgroup: always terminate io.stat lines
block, bfq: make bfq_has_work() more accurate
block, bfq: protect 'bfqd->queued' by 'bfqd->lock'
block: cleanup the VM accounting in submit_bio
block: Fix the bio.bi_opf comment
block: reorder the REQ_ flags
blk-iocost: combine local_stat and desc_stat to stat
block: improve the error message from bio_check_eod
block: allow passing a NULL bdev to bio_alloc_clone/bio_init_clone
block: remove superfluous calls to blkcg_bio_issue_init
kthread: unexport kthread_blkcg
blk-cgroup: cleanup blkcg_maybe_throttle_current
...
With the removal of seq_get_buf in blkcg_print_one_stat, we
cannot make adding the newline conditional on there being
relevant stats because the name was already written out
unconditionally.
Otherwise we may end up with multiple device names in one
line which is confusing and doesn't follow the nested-keyed
file format.
Signed-off-by: Wolfgang Bumiller <w.bumiller@proxmox.com>
Fixes: 252c651a4c ("blk-cgroup: stop using seq_get_buf")
Acked-by: Tejun Heo <tj@kernel.org>
Link: https://lore.kernel.org/r/20220111083159.42340-1-w.bumiller@proxmox.com
Signed-off-by: Jens Axboe <axboe@kernel.dk>
When we flush usage, wait, indebt stat in iocg_flush_stat(), we use
local_stat and desc_stat, which has no point since the leaf iocg
only has local_stat and the inner iocg only has desc_stat. Also
we don't need to flush percpu abs_vusage for these inner iocgs.
This patch combine local_stat and desc_stat to stat, only flush
percpu abs_vusage for active leaf iocgs, then build inner walk
list to propagate.
Signed-off-by: Chengming Zhou <zhouchengming@bytedance.com>
Acked-by: Tejun Heo <tj@kernel.org>
Link: https://lore.kernel.org/r/20220510034757.21761-1-zhouchengming@bytedance.com
Signed-off-by: Jens Axboe <axboe@kernel.dk>
When an iocg is in debt, its inuse weight is owned by debt handling and
should stay at 1. This invariant was broken when determining the amount of
surpluses at the beginning of donation calculation - when an iocg's
hierarchical weight is too low, the iocg is excluded from donation
calculation and its inuse is reset to its active regardless of its
indebtedness, triggering warnings like the following:
WARNING: CPU: 5 PID: 0 at block/blk-iocost.c:1416 iocg_kick_waitq+0x392/0x3a0
...
RIP: 0010:iocg_kick_waitq+0x392/0x3a0
Code: 00 00 be ff ff ff ff 48 89 4d a8 e8 98 b2 70 00 48 8b 4d a8 85 c0 0f 85 4a fe ff ff 0f 0b e9 43 fe ff ff 0f 0b e9 4d fe ff ff <0f> 0b e9 50 fe ff ff e8 a2 ae 70 00 66 90 0f 1f 44 00 00 55 48 89
RSP: 0018:ffffc90000200d08 EFLAGS: 00010016
...
<IRQ>
ioc_timer_fn+0x2e0/0x1470
call_timer_fn+0xa1/0x2c0
...
As this happens only when an iocg's hierarchical weight is negligible, its
impact likely is limited to triggering the warnings. Fix it by skipping
resetting inuse of under-weighted debtors.
Signed-off-by: Tejun Heo <tj@kernel.org>
Reported-by: Rik van Riel <riel@surriel.com>
Fixes: c421a3eb2e ("blk-iocost: revamp debt handling")
Cc: stable@vger.kernel.org # v5.10+
Link: https://lore.kernel.org/r/YmjODd4aif9BzFuO@slm.duckdns.org
Signed-off-by: Jens Axboe <axboe@kernel.dk>
Partition include/linux/blk-cgroup.h into two parts: one is public part,
the other is block layer private part.
Suggested by Christoph Hellwig.
Signed-off-by: Ming Lei <ming.lei@redhat.com>
Reviewed-by: Christoph Hellwig <hch@lst.de>
Link: https://lore.kernel.org/r/20220211101149.2368042-4-ming.lei@redhat.com
Signed-off-by: Jens Axboe <axboe@kernel.dk>
The donation calculation logic assumes that the donor has non-zero
after-donation hweight, so the lowest active hweight a donating cgroup can
have is 2 so that it can donate 1 while keeping the other 1 for itself.
Earlier, we only donated from cgroups with sizable surpluses so this
condition was always true. However, with the precise donation algorithm
implemented, f1de2439ec ("blk-iocost: revamp donation amount
determination") made the donation amount calculation exact enabling even low
hweight cgroups to donate.
This means that in rare occasions, a cgroup with active hweight of 1 can
enter donation calculation triggering the following warning and then a
divide-by-zero oops.
WARNING: CPU: 4 PID: 0 at block/blk-iocost.c:1928 transfer_surpluses.cold+0x0/0x53 [884/94867]
...
RIP: 0010:transfer_surpluses.cold+0x0/0x53
Code: 92 ff 48 c7 c7 28 d1 ab b5 65 48 8b 34 25 00 ae 01 00 48 81 c6 90 06 00 00 e8 8b 3f fe ff 48 c7 c0 ea ff ff ff e9 95 ff 92 ff <0f> 0b 48 c7 c7 30 da ab b5 e8 71 3f fe ff 4c 89 e8 4d 85 ed 74 0
4
...
Call Trace:
<IRQ>
ioc_timer_fn+0x1043/0x1390
call_timer_fn+0xa1/0x2c0
__run_timers.part.0+0x1ec/0x2e0
run_timer_softirq+0x35/0x70
...
iocg: invalid donation weights in /a/b: active=1 donating=1 after=0
Fix it by excluding cgroups w/ active hweight < 2 from donating. Excluding
these extreme low hweight donations shouldn't affect work conservation in
any meaningful way.
Signed-off-by: Tejun Heo <tj@kernel.org>
Fixes: f1de2439ec ("blk-iocost: revamp donation amount determination")
Cc: stable@vger.kernel.org # v5.10+
Link: https://lore.kernel.org/r/Ybfh86iSvpWKxhVM@slm.duckdns.org
Signed-off-by: Jens Axboe <axboe@kernel.dk>
seq_get_buf is a crutch that undoes all the memory safety of the
seq_file interface. Use the normal seq_printf interfaces instead.
Signed-off-by: Christoph Hellwig <hch@lst.de>
Acked-by: Tejun Heo <tj@kernel.org>
Link: https://lore.kernel.org/r/20210810152623.1796144-2-hch@lst.de
Signed-off-by: Jens Axboe <axboe@kernel.dk>
blkcg->lock depends on q->queue_lock which may depend on another driver
lock required in irq context, one example is dm-thin:
Chain exists of:
&pool->lock#3 --> &q->queue_lock --> &blkcg->lock
Possible interrupt unsafe locking scenario:
CPU0 CPU1
---- ----
lock(&blkcg->lock);
local_irq_disable();
lock(&pool->lock#3);
lock(&q->queue_lock);
<Interrupt>
lock(&pool->lock#3);
Fix the issue by using spin_lock_irq(&blkcg->lock) in ioc_weight_write().
Cc: Tejun Heo <tj@kernel.org>
Reported-by: Bruno Goncalves <bgoncalv@redhat.com>
Link: https://lore.kernel.org/linux-block/CA+QYu4rzz6079ighEanS3Qq_Dmnczcf45ZoJoHKVLVATTo1e4Q@mail.gmail.com/T/#u
Signed-off-by: Ming Lei <ming.lei@redhat.com>
Acked-by: Tejun Heo <tj@kernel.org>
Link: https://lore.kernel.org/r/20210803070608.1766400-1-ming.lei@redhat.com
Signed-off-by: Jens Axboe <axboe@kernel.dk>
iocg_wake_fn() open-codes wait_queue_entry removal and wakeup because it
wants the wq_entry to be always removed whether it ended up waking the
task or not. finish_wait() tests whether wq_entry needs removal without
grabbing the wait_queue lock and expects the waker to use
list_del_init_careful() after all waking operations are complete, which
iocg_wake_fn() didn't do. The operation order was wrong and the regular
list_del_init() was used.
The result is that if a waiter wakes up racing the waker, it can free pop
the wq_entry off stack before the waker is still looking at it, which can
lead to a backtrace like the following.
[7312084.588951] general protection fault, probably for non-canonical address 0x586bf4005b2b88: 0000 [#1] SMP
...
[7312084.647079] RIP: 0010:queued_spin_lock_slowpath+0x171/0x1b0
...
[7312084.858314] Call Trace:
[7312084.863548] _raw_spin_lock_irqsave+0x22/0x30
[7312084.872605] try_to_wake_up+0x4c/0x4f0
[7312084.880444] iocg_wake_fn+0x71/0x80
[7312084.887763] __wake_up_common+0x71/0x140
[7312084.895951] iocg_kick_waitq+0xe8/0x2b0
[7312084.903964] ioc_rqos_throttle+0x275/0x650
[7312084.922423] __rq_qos_throttle+0x20/0x30
[7312084.930608] blk_mq_make_request+0x120/0x650
[7312084.939490] generic_make_request+0xca/0x310
[7312084.957600] submit_bio+0x173/0x200
[7312084.981806] swap_readpage+0x15c/0x240
[7312084.989646] read_swap_cache_async+0x58/0x60
[7312084.998527] swap_cluster_readahead+0x201/0x320
[7312085.023432] swapin_readahead+0x2df/0x450
[7312085.040672] do_swap_page+0x52f/0x820
[7312085.058259] handle_mm_fault+0xa16/0x1420
[7312085.066620] do_page_fault+0x2c6/0x5c0
[7312085.074459] page_fault+0x2f/0x40
Fix it by switching to list_del_init_careful() and putting it at the end.
Signed-off-by: Tejun Heo <tj@kernel.org>
Reported-by: Rik van Riel <riel@surriel.com>
Fixes: 7caa47151a ("blkcg: implement blk-iocost")
Cc: stable@vger.kernel.org # v5.4+
Signed-off-by: Jens Axboe <axboe@kernel.dk>
When the weight of an active iocg is updated, weight_updated() is called
which in turn calls __propagate_weights() to update the active and inuse
weights so that the effective hierarchical weights are update accordingly.
The current implementation is incorrect for inner active nodes. For an
active leaf iocg, inuse can be any value between 1 and active and the
difference represents how much the iocg is donating. When weight is updated,
as long as inuse is clamped between 1 and the new weight, we're alright and
this is what __propagate_weights() currently implements.
However, that's not how an active inner node's inuse is set. An inner node's
inuse is solely determined by the ratio between the sums of inuse's and
active's of its children - ie. they're results of propagating the leaves'
active and inuse weights upwards. __propagate_weights() incorrectly applies
the same clamping as for a leaf when an active inner node's weight is
updated. Consider a hierarchy which looks like the following with saturating
workloads in AA and BB.
R
/ \
A B
| |
AA BB
1. For both A and B, active=100, inuse=100, hwa=0.5, hwi=0.5.
2. echo 200 > A/io.weight
3. __propagate_weights() update A's active to 200 and leave inuse at 100 as
it's already between 1 and the new active, making A:active=200,
A:inuse=100. As R's active_sum is updated along with A's active,
A:hwa=2/3, B:hwa=1/3. However, because the inuses didn't change, the
hwi's remain unchanged at 0.5.
4. The weight of A is now twice that of B but AA and BB still have the same
hwi of 0.5 and thus are doing the same amount of IOs.
Fix it by making __propgate_weights() always calculate the inuse of an
active inner iocg based on the ratio of child_inuse_sum to child_active_sum.
Signed-off-by: Tejun Heo <tj@kernel.org>
Reported-by: Dan Schatzberg <dschatzberg@fb.com>
Fixes: 7caa47151a ("blkcg: implement blk-iocost")
Cc: stable@vger.kernel.org # v5.4+
Link: https://lore.kernel.org/r/YJsxnLZV1MnBcqjj@slm.duckdns.org
Signed-off-by: Jens Axboe <axboe@kernel.dk>
ioc_adjust_base_vrate() ignored vrate_min when rq_wait_pct indicates that
there is QD contention. The reasoning was that QD depletion always reliably
indicates device saturation and thus it's safe to override user specified
vrate_min. However, this sometimes leads to unnecessary throttling,
especially on really fast devices, because vrate adjustments have delays and
inertia. It also confuses users because the behavior violates the explicitly
specified configuration.
This patch drops the special case handling so that vrate_min is always
applied.
Signed-off-by: Tejun Heo <tj@kernel.org>
Link: https://lore.kernel.org/r/YIIo1HuyNmhDeiNx@slm.duckdns.org
Signed-off-by: Jens Axboe <axboe@kernel.dk>
When initializing iocost for a queue, its rqos should be registered before
the blkcg policy is activated to allow policy data initiailization to lookup
the associated ioc. This unfortunately means that the rqos methods can be
called on bios before iocgs are attached to all existing blkgs.
While the race is theoretically possible on ioc_rqos_throttle(), it mostly
happened in ioc_rqos_merge() due to the difference in how they lookup ioc.
The former determines it from the passed in @rqos and then bails before
dereferencing iocg if the looked up ioc is disabled, which most likely is
the case if initialization is still in progress. The latter looked up ioc by
dereferencing the possibly NULL iocg making it a lot more prone to actually
triggering the bug.
* Make ioc_rqos_merge() use the same method as ioc_rqos_throttle() to look
up ioc for consistency.
* Make ioc_rqos_throttle() and ioc_rqos_merge() test for NULL iocg before
dereferencing it.
* Explain the danger of NULL iocgs in blk_iocost_init().
Signed-off-by: Tejun Heo <tj@kernel.org>
Reported-by: Jonathan Lemon <bsd@fb.com>
Cc: stable@vger.kernel.org # v5.4+
Signed-off-by: Jens Axboe <axboe@kernel.dk>
It will be helpful to trace the iocg's whole state, including active and
idle state. And we can easily expand the original iocost_iocg_activate
trace event to support a state trace class, including active and idle
state tracing.
Acked-by: Tejun Heo <tj@kernel.org>
Signed-off-by: Baolin Wang <baolin.wang@linux.alibaba.com>
Signed-off-by: Jens Axboe <axboe@kernel.dk>
Factor out the base vrate change code into a separate function
to fimplify the ioc_timer_fn().
No functional change.
Signed-off-by: Baolin Wang <baolin.wang@linux.alibaba.com>
Acked-by: Tejun Heo <tj@kernel.org>
Signed-off-by: Jens Axboe <axboe@kernel.dk>
Factor out the iocgs' state check into a separate function to
simplify the ioc_timer_fn().
No functional change.
Signed-off-by: Baolin Wang <baolin.wang@linux.alibaba.com>
Acked-by: Tejun Heo <tj@kernel.org>
Signed-off-by: Jens Axboe <axboe@kernel.dk>
We only use the hweight based usage ratio to calculate the new
hweight_inuse of the iocg to decide if this iocg can donate some
surplus vtime.
Thus move the usage ratio calculation to the correct place to
avoid unnecessary calculation for some vtime shortage iocgs.
Signed-off-by: Baolin Wang <baolin.wang@linux.alibaba.com>
Acked-by: Tejun Heo <tj@kernel.org>
Signed-off-by: Jens Axboe <axboe@kernel.dk>
To simplify block device lookup and a few other upcoming areas, make sure
that we always have a struct block_device available for each disk and
each partition, and only find existing block devices in bdget. The only
downside of this is that each device and partition uses a little more
memory. The upside will be that a lot of code can be simplified.
With that all we need to look up the block device is to lookup the inode
and do a few sanity checks on the gendisk, instead of the separate lookup
for the gendisk. For blk-cgroup which wants to access a gendisk without
opening it, a new blkdev_{get,put}_no_open low-level interface is added
to replace the previous get_gendisk use.
Note that the change to look up block device directly instead of the two
step lookup using struct gendisk causes a subtile change in behavior:
accessing a non-existing partition on an existing block device can now
cause a call to request_module. That call is harmless, and in practice
no recent system will access these nodes as they aren't created by udev
and static /dev/ setups are unusual.
Signed-off-by: Christoph Hellwig <hch@lst.de>
Reviewed-by: Jan Kara <jack@suse.cz>
Reviewed-by: Hannes Reinecke <hare@suse.de>
Signed-off-by: Jens Axboe <axboe@kernel.dk>
We have defined common interface blk_queue_registered() to
test QUEUE_FLAG_REGISTERED. Just use it.
Signed-off-by: Yufen Yu <yuyufen@huawei.com>
Signed-off-by: Jens Axboe <axboe@kernel.dk>
An iocg may have 0 debt but non-zero delay. The current debt forgiveness
logic doesn't act on such iocgs. This can lead to unexpected behaviors - an
iocg with a little bit of debt will have its delay canceled through debt
forgiveness but one w/o any debt but active delay will have to wait out
until its delay decays out.
This patch updates the debt handling logic so that it treats delays the same
as debts. If either debt or delay is active, debt forgiveness logic kicks in
and acts on both the same way.
Also, avoid turning the debt and delay directly to zero as that can confuse
state transitions.
Signed-off-by: Tejun Heo <tj@kernel.org>
Signed-off-by: Jens Axboe <axboe@kernel.dk>
Debt forgiveness logic was counting the number of consecutive !busy periods
as the trigger condition. While this usually works, it can easily be thrown
off by temporary fluctuations especially on configurations w/ short periods.
This patch reimplements debt forgiveness so that:
* Use the average usage over the forgiveness period instead of counting
consecutive periods.
* Debt is reduced at around the target rate (1/2 every 100ms) regardless of
ioc period duration.
* Usage threshold is raised to 50%. Combined with the preceding changes and
the switch to average usage, this makes debt forgivness a lot more
effective at reducing the amount of unnecessary idleness.
* Constants are renamed with DFGV_ prefix.
Signed-off-by: Tejun Heo <tj@kernel.org>
Signed-off-by: Jens Axboe <axboe@kernel.dk>
Debt sets the initial delay duration which is decayed over time. The current
debt reduction halved the debt but didn't change the delay. It prevented
future debts from increasing delay but didn't do anything to lower the
existing delay, limiting the mechanism's ability to reduce unnecessary
idling.
Reset iocg->delay to 0 after debt reduction so that iocg_kick_waitq()
recalculates new delay value based on the reduced debt amount.
Signed-off-by: Tejun Heo <tj@kernel.org>
Signed-off-by: Jens Axboe <axboe@kernel.dk>
Debt reduction was blocked if any iocg was short on budget in the past
period to avoid reducing debts while some iocgs are saturated. However, this
ends up unnecessarily blocking debt reduction due to temporary local
imbalances when the device is generally being underutilized, while also
failing to block when the underlying device is overwhelmed and the usage
becomes low from high latency.
Given that debt accumulation mostly happens with swapout bursts which can
significantly deteriorate the underlying device's latency response, the
current logic is not great.
Let's replace it with ioc->busy_level based condition so that we block debt
reduction when the underlying device is being saturated. ioc_forgive_debts()
call is moved after busy_level determination.
Signed-off-by: Tejun Heo <tj@kernel.org>
Signed-off-by: Jens Axboe <axboe@kernel.dk>
Debt reduction logic is going to be improved and expanded. Factor it out
into ioc_forgive_debts() and generalize the comment a bit. No functional
change.
Signed-off-by: Tejun Heo <tj@kernel.org>
Signed-off-by: Jens Axboe <axboe@kernel.dk>
adjust_inuse_and_calc_cost() is responsible for reducing the amount of
donated weights dynamically in period as the budget runs low. Because we
don't want to do full donation calculation in period, we keep latching up
inuse by INUSE_ADJ_STEP_PCT of the active weight of the cgroup until the
resulting hweight_inuse is satisfactory.
Unfortunately, the adj_step calculation was reading the active weight before
acquiring ioc->lock. Because the current thread could have lost race to
activate the iocg to another thread before entering this function, it may
read the active weight as zero before acquiring ioc->lock. When this
happens, the adj_step is calculated as zero and the incremental adjustment
loop becomes an infinite one.
Fix it by fetching the active weight after acquiring ioc->lock.
Fixes: b0853ab4a2 ("blk-iocost: revamp in-period donation snapbacks")
Signed-off-by: Tejun Heo <tj@kernel.org>
Signed-off-by: Jens Axboe <axboe@kernel.dk>
Conceptually, root_iocg->hweight_donating must be less than WEIGHT_ONE but
all hweight calculations round up and thus it may end up >= WEIGHT_ONE
triggering divide-by-zero and other issues. Bound the value to avoid
surprises.
Fixes: e08d02aa5f ("blk-iocost: implement Andy's method for donation weight updates")
Signed-off-by: Tejun Heo <tj@kernel.org>
Signed-off-by: Jens Axboe <axboe@kernel.dk>
These are really cheap to collect and can be useful in debugging iocost
behavior. Add them as debug stats for now.
Signed-off-by: Tejun Heo <tj@kernel.org>
Signed-off-by: Jens Axboe <axboe@kernel.dk>
When an iocg accumulates too much vtime or gets deactivated, we throw away
some vtime, which lowers the overall device utilization. As the exact amount
which is being thrown away is known, we can compensate by accelerating the
vrate accordingly so that the extra vtime generated in the current period
matches what got lost.
This significantly improves work conservation when involving high weight
cgroups with intermittent and bursty IO patterns.
Signed-off-by: Tejun Heo <tj@kernel.org>
Signed-off-by: Jens Axboe <axboe@kernel.dk>
A low weight iocg can amass a large amount of debt, for example, when
anonymous memory gets reclaimed aggressively. If the system has a lot of
memory paired with a slow IO device, the debt can span multiple seconds or
more. If there are no other subsequent IO issuers, the in-debt iocg may end
up blocked paying its debt while the IO device is idle.
This patch implements a mechanism to protect against such pathological
cases. If the device has been sufficiently idle for a substantial amount of
time, the debts are halved. The criteria are on the conservative side as we
want to resolve the rare extreme cases without impacting regular operation
by forgiving debts too readily.
Signed-off-by: Tejun Heo <tj@kernel.org>
Signed-off-by: Jens Axboe <axboe@kernel.dk>
Curently, iocost syncs the delay duration to the outstanding debt amount,
which seemed enough to protect the system from anon memory hogs. However,
that was mostly because the delay calcuation was using hweight_inuse which
quickly converges towards zero under debt for delay duration calculation,
often pusnishing debtors overly harshly for longer than deserved.
The previous patch fixed the delay calcuation and now the protection against
anonymous memory hogs isn't enough because the effect of delay is indirect
and non-linear and a huge amount of future debt can accumulate abruptly
while unthrottled.
This patch implements delay hysteresis so that delay is decayed
exponentially over time instead of getting cleared immediately as debt is
paid off. While the overall behavior is similar to the blk-cgroup
implementation used by blk-iolatency, a lot of the details are different and
due to the empirical nature of the mechanism, it's challenging to adapt the
mechanism for one controller without negatively impacting the other.
As the delay is gradually decayed now, there's no point in running it from
its own hrtimer. Periodic updates are now performed from ioc_timer_fn() and
the dedicated hrtimer is removed.
Signed-off-by: Tejun Heo <tj@kernel.org>
Cc: Josef Bacik <josef@toxicpanda.com>
Signed-off-by: Jens Axboe <axboe@kernel.dk>
Debt handling had several issues.
* How much inuse a debtor carries wasn't clearly defined. inuse would be
driven down over time from not issuing IOs but it'd be better to clamp it
to minimum immediately once in debt.
* How much can be paid off was determined by hweight_inuse. As inuse was
driven down, the payment amount would fall together regardless of the
debtor's active weight. This means that the debtors were punished harshly.
* ioc_rqos_merge() wasn't calling blkcg_schedule_throttle() after
iocg_kick_delay().
This patch revamps debt handling so that
* Debt handling owns inuse for iocgs in debt and keeps them at zero.
* Payment amount is determined by hweight_active. This is more deterministic
and safer than hweight_inuse but still far from ideal in that it doesn't
factor in possible donations from other iocgs for debt payments. This
likely needs further improvements in the future.
* iocg_rqos_merge() now calls blkcg_schedule_throttle() as necessary.
Signed-off-by: Tejun Heo <tj@kernel.org>
Cc: Andy Newell <newella@fb.com>
Signed-off-by: Jens Axboe <axboe@kernel.dk>
When the margin drops below the minimum on a donating iocg, donation is
immediately canceled in full. There are a couple shortcomings with the
current behavior.
* It's abrupt. A small temporary budget deficit can lead to a wide swing in
weight allocation and a large surplus.
* It's open coded in the issue path but not implemented for the merge path.
A series of merges at a low inuse can make the iocg incur debts and stall
incorrectly.
This patch reimplements in-period donation snapbacks so that
* inuse adjustment and cost calculations are factored into
adjust_inuse_and_calc_cost() which is called from both the issue and merge
paths.
* Snapbacks are more gradual. It occurs in quarter steps.
* A snapback triggers if the margin goes below the low threshold and is
lower than the budget at the time of the last adjustment.
* For the above, __propagate_weights() stores the margin in
iocg->saved_margin. Move iocg->last_inuse storing together into
__propagate_weights() for consistency.
* Full snapback is guaranteed when there are waiters.
* With precise donation and gradual snapbacks, inuse adjustments are now a
lot more effective and the value of scaling inuse on weight changes isn't
clear. Removed inuse scaling from weight_update().
Signed-off-by: Tejun Heo <tj@kernel.org>
Signed-off-by: Jens Axboe <axboe@kernel.dk>
iocost has various safety nets to combat inuse adjustment calculation
inaccuracies. With Andy's method implemented in transfer_surpluses(), inuse
adjustment calculations are now accurate and we can make donation amount
determinations accurate too.
* Stop keeping track of past usage history and using the maximum. Act on the
immediate usage information.
* Remove donation constraints defined by SURPLUS_* constants. Donate
whatever isn't used.
* Determine the donation amount so that the iocg will end up with
MARGIN_TARGET_PCT budget at the end of the coming period assuming the same
usage as the previous period. TARGET is set at 50% of period, which is the
previous maximum. This provides smooth convergence for most repetitive IO
patterns.
* Apply donation logic early at 20% budget. There's no risk in doing so as
the calculation is based on the delta between the current budget and the
target budget at the end of the coming period.
* Remove preemptive iocg activation for zero cost IOs. As donation can reach
near zero now, the mere activation doesn't provide any protection anymore.
In the unlikely case that this becomes a problem, the right solution is
assigning appropriate costs for such IOs.
This significantly improves the donation determination logic while also
simplifying it. Now all donations are immediate, exact and smooth.
Signed-off-by: Tejun Heo <tj@kernel.org>
Cc: Andy Newell <newella@fb.com>
Signed-off-by: Jens Axboe <axboe@kernel.dk>
iocost implements work conservation by reducing iocg->inuse and propagating
the adjustment upwards proportionally. However, while I knew the target
absolute hierarchical proportion - adjusted hweight_inuse, I couldn't figure
out how to determine the iocg->inuse adjustment to achieve that and
approximated the adjustment by scaling iocg->inuse using the proportion of
the needed hweight_inuse changes.
When nested, these scalings aren't accurate even when adjusting a single
node as the donating node also receives the benefit of the donated portion.
When multiple nodes are donating as they often do, they can be wildly wrong.
iocost employed various safety nets to combat the inaccuracies. There are
ample buffers in determining how much to donate, the adjustments are
conservative and gradual. While it can achieve a reasonable level of work
conservation in simple scenarios, the inaccuracies can easily add up leading
to significant loss of total work. This in turn makes it difficult to
closely cap vrate as vrate adjustment is needed to compensate for the loss
of work. The combination of inaccurate donation calculations and vrate
adjustments can lead to wide fluctuations and clunky overall behaviors.
Andy Newell devised a method to calculate the needed ->inuse updates to
achieve the target hweight_inuse's. The method is compatible with the
proportional inuse adjustment propagation which allows all hot path
operations to be local to each iocg.
To roughly summarize, Andy's method divides the tree into donating and
non-donating parts, calculates global donation rate which is used to
determine the target hweight_inuse for each node, and then derives per-level
proportions. There's non-trivial amount of math involved. Please refer to
the following pdfs for detailed descriptions.
https://drive.google.com/file/d/1PsJwxPFtjUnwOY1QJ5AeICCcsL7BM3bohttps://drive.google.com/file/d/1vONz1-fzVO7oY5DXXsLjSxEtYYQbOvsEhttps://drive.google.com/file/d/1WcrltBOSPN0qXVdBgnKm4mdp9FhuEFQN
This patch implements Andy's method in transfer_surpluses(). This makes the
donation calculations accurate per cycle and enables further improvements in
other parts of the donation logic.
Signed-off-by: Tejun Heo <tj@kernel.org>
Cc: Andy Newell <newella@fb.com>
Signed-off-by: Jens Axboe <axboe@kernel.dk>
The way the surplus donation logic is structured isn't great. There are two
separate paths for starting/increasing donations and decreasing them making
the logic harder to follow and is prone to unnecessary behavior differences.
In preparation for improved donation handling, this patch restructures the
code so that
* All donors - new, increasing and decreasing - are funneled through the
same code path.
* The target donation calculation is factored into hweight_after_donation()
which is called once from the same spot for all possible donors.
* Actual inuse adjustment is factored into trasnfer_surpluses().
This change introduces a few behavior differences - e.g. donation amount
reduction now uses the max usage of the recent three periods just like new
and increasing donations, and inuse now gets adjusted upwards the same way
it gets downwards. These differences are unlikely to have severely negative
implications and the whole logic will be revamped soon.
This patch also removes two tracepoints. The existing TPs don't quite fit
the new implementation. A later patch will update and reinstate them.
Signed-off-by: Tejun Heo <tj@kernel.org>
Signed-off-by: Jens Axboe <axboe@kernel.dk>
Budget donations are inaccurate and could take multiple periods to converge.
To prevent triggering vrate adjustments while surplus transfers were
catching up, vrate adjustment was suppressed if donations were increasing,
which was indicated by non-zero nr_surpluses.
This entangling won't be necessary with the scheduled rewrite of donation
mechanism which will make it precise and immediate. Let's decouple the two
in preparation.
Signed-off-by: Tejun Heo <tj@kernel.org>
Signed-off-by: Jens Axboe <axboe@kernel.dk>
Instead of marking iocgs with surplus with a flag and filtering for them
while walking all active iocgs, build a surpluses list. This doesn't make
much difference now but will help implementing improved donation logic which
will iterate iocgs with surplus multiple times.
Signed-off-by: Tejun Heo <tj@kernel.org>
Signed-off-by: Jens Axboe <axboe@kernel.dk>
Currently, iocg->usages[] which are used to guide inuse adjustments are
calculated from vtime deltas. This, however, assumes that the hierarchical
inuse weight at the time of calculation held for the entire period, which
often isn't true and can lead to significant errors.
Now that we have absolute usage information collected, we can derive
iocg->usages[] from iocg->local_stat.usage_us so that inuse adjustment
decisions are made based on actual absolute usage. The calculated usage is
clamped between 1 and WEIGHT_ONE and WEIGHT_ONE is also used to signal
saturation regardless of the current hierarchical inuse weight.
Signed-off-by: Tejun Heo <tj@kernel.org>
Signed-off-by: Jens Axboe <axboe@kernel.dk>
Currently, iocost doesn't collect or expose any statistics punting off all
monitoring duties to drgn based iocost_monitor.py. While it works for some
scenarios, there are some usability and data availability challenges. For
example, accurate per-cgroup usage information can't be tracked by vtime
progression at all and the number available in iocg->usages[] are really
short-term snapshots used for control heuristics with possibly significant
errors.
This patch implements per-cgroup absolute usage stat counter and exposes it
through io.stat along with the current vrate. Usage stat collection and
flushing employ the same method as cgroup rstat on the active iocg's and the
only hot path overhead is preemption toggling and adding to a percpu
counter.
Signed-off-by: Tejun Heo <tj@kernel.org>
Signed-off-by: Jens Axboe <axboe@kernel.dk>
Currently, debt handling requires only iocg->waitq.lock. In the future, we
want to adjust and propagate inuse changes depending on debt status. Let's
grab ioc->lock in debt handling paths in preparation.
* Because ioc->lock nests outside iocg->waitq.lock, the decision to grab
ioc->lock needs to be made before entering the critical sections.
* Add and use iocg_[un]lock() which handles the conditional double locking.
* Add @pay_debt to iocg_kick_waitq() so that debt payment happens only when
the caller grabbed both locks.
This patch is prepatory and the comments contain references to future
changes.
Signed-off-by: Tejun Heo <tj@kernel.org>
Signed-off-by: Jens Axboe <axboe@kernel.dk>
The margin handling was pretty inconsistent.
* ioc->margin_us and ioc->inuse_margin_vtime were used as vtime margin
thresholds. However, the two are in different units with the former
requiring conversion to vtime on use.
* iocg_kick_waitq() was using a quarter of WAITQ_TIMER_MARGIN_PCT of
period_us as the timer slack - ~1.2%. While iocg_kick_delay() was using a
quarter of ioc->margin_us - ~12.5%. There aren't strong reasons to use
different values for the two.
This patch cleans up margin and timer slack handling:
* vtime margins are now recorded in ioc->margins.{min, max} on period
duration changes and used consistently.
* Timer slack is now 1% of period_us and recorded in ioc->timer_slack_ns and
used consistently for iocg_kick_waitq() and iocg_kick_delay().
The only functional change is shortening of timer slack. No meaningful
visible change is expected.
Signed-off-by: Tejun Heo <tj@kernel.org>
Signed-off-by: Jens Axboe <axboe@kernel.dk>
They are in microseconds and wrap in around 1.2 hours with u32. While
unlikely, confusions from wraparounds are still possible. We aren't saving
anything meaningful by keeping these u32. Let's make them u64.
Signed-off-by: Tejun Heo <tj@kernel.org>
Signed-off-by: Jens Axboe <axboe@kernel.dk>
Jinke Han