On systems where many CPUs share one LLC, unbound workqueues using
WQ_AFFN_CACHE collapse to a single worker pool, causing heavy spinlock
contention on pool->lock. For example, Chuck Lever measured 39% of
cycles lost to native_queued_spin_lock_slowpath on a 12-core shared-L3
NFS-over-RDMA system.
The existing affinity hierarchy (cpu, smt, cache, numa, system) offers
no intermediate option between per-LLC and per-SMT-core granularity.
Add WQ_AFFN_CACHE_SHARD, which subdivides each LLC into groups of at
most wq_cache_shard_size cores (default 8, tunable via boot parameter).
Shards are always split on core (SMT group) boundaries so that
Hyper-Threading siblings are never placed in different pods. Cores are
distributed across shards as evenly as possible -- for example, 36 cores
in a single LLC with max shard size 8 produces 5 shards of 8+7+7+7+7
cores.
The implementation follows the same comparator pattern as other affinity
scopes: precompute_cache_shard_ids() pre-fills the cpu_shard_id[] array
from the already-initialized WQ_AFFN_CACHE and WQ_AFFN_SMT topology,
and cpus_share_cache_shard() is passed to init_pod_type().
Benchmark on NVIDIA Grace (72 CPUs, single LLC, 50k items/thread), show
cache_shard delivers ~5x the throughput and ~6.5x lower p50 latency
compared to cache scope on this 72-core single-LLC system.
Suggested-by: Tejun Heo <tj@kernel.org>
Signed-off-by: Breno Leitao <leitao@debian.org>
Signed-off-by: Tejun Heo <tj@kernel.org>
Add a Resource-managed version of alloc_workqueue() to fix common
problem of drivers mixing devm() calls with destroy_workqueue. Such
naive and discouraged driver approach leads to difficult to debug bugs
when the driver:
1. Allocates workqueue in standard way and destroys it in driver
remove() callback,
2. Sets work struct with devm_work_autocancel(),
3. Registers interrupt handler with devm_request_threaded_irq().
Which leads to following unbind/removal path:
1. destroy_workqueue() via driver remove(),
Any interrupt coming now would still execute the interrupt handler,
which queues work on destroyed workqueue.
2. devm_irq_release(),
3. devm_work_drop() -> cancel_work_sync() on destroyed workqueue.
devm_alloc_workqueue() has two benefits:
1. Solves above problem of mix-and-match devres and non-devres code in
driver,
2. Simplify any sane drivers which were correctly using
alloc_workqueue() + devm_add_action_or_reset().
Signed-off-by: Krzysztof Kozlowski <krzysztof.kozlowski@oss.qualcomm.com>
Acked-by: Tejun Heo <tj@kernel.org>
Reviewed-by: Andy Shevchenko <andriy.shevchenko@linux.intel.com>
Signed-off-by: Tejun Heo <tj@kernel.org>
Currently there are users of queue_delayed_work() who specify
system_long_wq, the per-cpu workqueue. This workqueue should
be used for long per-cpu works, but queue_delayed_work()
queue the work using:
queue_delayed_work_on(WORK_CPU_UNBOUND, ...);
This would end up calling __queue_delayed_work() that does:
if (housekeeping_enabled(HK_TYPE_TIMER)) {
// [....]
} else {
if (likely(cpu == WORK_CPU_UNBOUND))
add_timer_global(timer);
else
add_timer_on(timer, cpu);
}
So when cpu == WORK_CPU_UNBOUND the timer is global and is
not using a specific CPU. Later, when __queue_work() is called:
if (req_cpu == WORK_CPU_UNBOUND) {
if (wq->flags & WQ_UNBOUND)
cpu = wq_select_unbound_cpu(raw_smp_processor_id());
else
cpu = raw_smp_processor_id();
}
Because the wq is not unbound, it takes the CPU where the timer
fired and enqueue the work on that CPU.
The consequence of all of this is that the work can run anywhere,
depending on where the timer fired.
Introduce system_dfl_long_wq in order to change, in a future step,
users that are still calling:
queue_delayed_work(system_long_wq, ...);
with the new system_dfl_long_wq instead, so that the work may
benefit from scheduler task placement.
Signed-off-by: Marco Crivellari <marco.crivellari@suse.com>
Signed-off-by: Tejun Heo <tj@kernel.org>
Update documentation to use "per-CPU workqueue" instead of
"global workqueue" to match the system_wq to system_percpu_wq
rename. The workqueue behavior remains unchanged; this just
aligns terminology with the clearer naming.
Fixes: a2be943b46 ("workqueue: replace use of system_wq with system_percpu_wq")
Signed-off-by: Mallesh Koujalagi <mallesh.koujalagi@intel.com>
Signed-off-by: Tejun Heo <tj@kernel.org>
Sphinx emitted a warning during make texinfodocs:
WARNING: Inline literal start-string without end-string.
This was caused by the trailing '*' in "%WQ_*" being parsed as
reStructuredText markup in the kernel-doc comment.
Escape the '*' in the comment so that Sphinx treats it as a literal
character, resolving the warning.
Signed-off-by: Kriish Sharma <kriish.sharma2006@gmail.com>
Signed-off-by: Tejun Heo <tj@kernel.org>
Currently if a user enqueue a work item using schedule_delayed_work() the
used wq is "system_wq" (per-cpu wq) while queue_delayed_work() use
WORK_CPU_UNBOUND (used when a cpu is not specified). The same applies to
schedule_work() that is using system_wq and queue_work(), that makes use
again of WORK_CPU_UNBOUND.
This lack of consistentcy cannot be addressed without refactoring the API.
alloc_workqueue() treats all queues as per-CPU by default, while unbound
workqueues must opt-in via WQ_UNBOUND.
This default is suboptimal: most workloads benefit from unbound queues,
allowing the scheduler to place worker threads where they’re needed and
reducing noise when CPUs are isolated.
This patch adds a new WQ_PERCPU flag to explicitly request the use of
the per-CPU behavior. Both flags coexist for one release cycle to allow
callers to transition their calls.
Once migration is complete, WQ_UNBOUND can be removed and unbound will
become the implicit default.
With the introduction of the WQ_PERCPU flag (equivalent to !WQ_UNBOUND),
any alloc_workqueue() caller that doesn’t explicitly specify WQ_UNBOUND
must now use WQ_PERCPU.
All existing users have been updated accordingly.
Suggested-by: Tejun Heo <tj@kernel.org>
Signed-off-by: Marco Crivellari <marco.crivellari@suse.com>
Signed-off-by: Tejun Heo <tj@kernel.org>
Currently if a user enqueue a work item using schedule_delayed_work() the
used wq is "system_wq" (per-cpu wq) while queue_delayed_work() use
WORK_CPU_UNBOUND (used when a cpu is not specified). The same applies to
schedule_work() that is using system_wq and queue_work(), that makes use
again of WORK_CPU_UNBOUND.
This lack of consistentcy cannot be addressed without refactoring the API.
system_wq is a per-CPU worqueue, yet nothing in its name tells about that
CPU affinity constraint, which is very often not required by users. Make
it clear by adding a system_percpu_wq.
queue_work() / queue_delayed_work() mod_delayed_work() will now use the
new per-cpu wq: whether the user still stick on the old name a warn will
be printed along a wq redirect to the new one.
This patch add the new system_percpu_wq except for mm, fs and net
subsystem, whom are handled in separated patches.
The old wq will be kept for a few release cylces.
Suggested-by: Tejun Heo <tj@kernel.org>
Signed-off-by: Marco Crivellari <marco.crivellari@suse.com>
Signed-off-by: Tejun Heo <tj@kernel.org>
Currently if a user enqueue a work item using schedule_delayed_work() the
used wq is "system_wq" (per-cpu wq) while queue_delayed_work() use
WORK_CPU_UNBOUND (used when a cpu is not specified). The same applies to
schedule_work() that is using system_wq and queue_work(), that makes use
again of WORK_CPU_UNBOUND.
This lack of consistentcy cannot be addressed without refactoring the API.
system_unbound_wq should be the default workqueue so as not to enforce
locality constraints for random work whenever it's not required.
Adding system_dfl_wq to encourage its use when unbound work should be used.
queue_work() / queue_delayed_work() / mod_delayed_work() will now use the
new unbound wq: whether the user still use the old wq a warn will be
printed along with a wq redirect to the new one.
The old system_unbound_wq will be kept for a few release cycles.
Suggested-by: Tejun Heo <tj@kernel.org>
Signed-off-by: Marco Crivellari <marco.crivellari@suse.com>
Signed-off-by: Tejun Heo <tj@kernel.org>
The last use of the work_on_cpu_safe() macro was removed recently by
commit 9cda46babd ("crypto: n2 - remove Niagara2 SPU driver")
Remove it, and the work_on_cpu_safe_key() function it calls.
Signed-off-by: Dr. David Alan Gilbert <linux@treblig.org>
Signed-off-by: Tejun Heo <tj@kernel.org>
Currently if a user enqueue a work item using schedule_delayed_work() the
used wq is "system_wq" (per-cpu wq) while queue_delayed_work() use
WORK_CPU_UNBOUND (used when a cpu is not specified). The same applies to
schedule_work() that is using system_wq and queue_work(), that makes use
again of WORK_CPU_UNBOUND.
This lack of consistentcy cannot be addressed without refactoring the API.
This patch adds a new WQ_PERCPU flag to explicitly request the use of
the per-CPU behavior. Both flags coexist for one release cycle to allow
callers to transition their calls.
Once migration is complete, WQ_UNBOUND can be removed and unbound will
become the implicit default.
tj: Merged doc patch.
Suggested-by: Tejun Heo <tj@kernel.org>
Signed-off-by: Marco Crivellari <marco.crivellari@suse.com>
Signed-off-by: Tejun Heo <tj@kernel.org>
Currently, if a user enqueue a work item using schedule_delayed_work() the
used wq is "system_wq" (per-cpu wq) while queue_delayed_work() use
WORK_CPU_UNBOUND (used when a cpu is not specified). The same applies to
schedule_work() that is using system_wq and queue_work(), that makes use
again of WORK_CPU_UNBOUND.
This lack of consistentcy cannot be addressed without refactoring the API.
system_wq is a per-CPU worqueue, yet nothing in its name tells about that
CPU affinity constraint, which is very often not required by users. Make it
clear by adding a system_percpu_wq.
system_unbound_wq should be the default workqueue so as not to enforce
locality constraints for random work whenever it's not required.
Adding system_dfl_wq to encourage its use when unbound work should be used.
Suggested-by: Tejun Heo <tj@kernel.org>
Signed-off-by: Marco Crivellari <marco.crivellari@suse.com>
Signed-off-by: Tejun Heo <tj@kernel.org>
Hook alloc_workqueue and alloc_workqueue_attrs() so that they're
accounted to the callsite. Since we're doing allocations on behalf of
another subsystem, this helps when using memory allocation profiling to
check for leaks.
Cc: Tejun Heo <tj@kernel.org>
Cc: Lai Jiangshan <jiangshanlai@gmail.com>
Signed-off-by: Kent Overstreet <kent.overstreet@linux.dev>
Signed-off-by: Tejun Heo <tj@kernel.org>
Pull workqueue updates from Tejun Heo:
"Fix statistic update race condition and a couple documentation
updates"
* tag 'wq-for-6.16' of git://git.kernel.org/pub/scm/linux/kernel/git/tj/wq:
workqueue: fix typo in comment
workqueue: Fix race condition in wq->stats incrementation
workqueue: Better document teardown for delayed_work
Fix a duplicated word "that that" in the comment describing the
@max_active behavior for unbound workqueues.
Signed-off-by: Guan-Chun Wu <409411716@gms.tku.edu.tw>
Signed-off-by: Tejun Heo <tj@kernel.org>
WQ_MAX_ACTIVE is currently set to 512, which was established approximately
15 yeas ago. However, with the significant increase in machine sizes and
capabilities, the previous limit of 256 concurrent tasks is no longer
sufficient. Therefore, we propose to increase WQ_MAX_ACTIVE to 2048.
and WQ_DFL_ACTIVE is 1024 now.
Signed-off-by: Chen Ridong <chenridong@huawei.com>
Signed-off-by: Tejun Heo <tj@kernel.org>
Calling va_start / va_end multiple times is undefined and causes
problems with certain compiler / platforms.
Change alloc_ordered_workqueue_lockdep_map to a macro and updated
__alloc_workqueue to take a va_list argument.
Cc: Sergey Senozhatsky <senozhatsky@chromium.org>
Cc: Tejun Heo <tj@kernel.org>
Cc: Lai Jiangshan <jiangshanlai@gmail.com>
Signed-off-by: Matthew Brost <matthew.brost@intel.com>
Signed-off-by: Tejun Heo <tj@kernel.org>
Add an interface for a user-defined workqueue lockdep map, which is
helpful when multiple workqueues are created for the same purpose. This
also helps avoid leaking lockdep maps on each workqueue creation.
v2:
- Add alloc_workqueue_lockdep_map (Tejun)
v3:
- Drop __WQ_USER_OWNED_LOCKDEP (Tejun)
- static inline alloc_ordered_workqueue_lockdep_map (Tejun)
Cc: Tejun Heo <tj@kernel.org>
Cc: Lai Jiangshan <jiangshanlai@gmail.com>
Signed-off-by: Matthew Brost <matthew.brost@intel.com>
Signed-off-by: Tejun Heo <tj@kernel.org>
Breno Leitao
Tejun Heo
Krzysztof Kozlowski
Marco Crivellari
Mallesh Koujalagi
Frederic Weisbecker
Kriish Sharma
Dr. David Alan Gilbert
Kent Overstreet
Linus Torvalds
Ingo Molnar
Guan-Chun Wu
Chen Ridong
Matthew Brost