Pull workqueue updates from Tejun Heo:
- New default WQ_AFFN_CACHE_SHARD affinity scope subdivides LLCs into
smaller shards to improve scalability on machines with many CPUs per
LLC
- Misc:
- system_dfl_long_wq for long unbound works
- devm_alloc_workqueue() for device-managed allocation
- sysfs exposure for ordered workqueues and the EFI workqueue
- removal of HK_TYPE_WQ from wq_unbound_cpumask
- various small fixes
* tag 'wq-for-7.1' of git://git.kernel.org/pub/scm/linux/kernel/git/tj/wq: (21 commits)
workqueue: validate cpumask_first() result in llc_populate_cpu_shard_id()
workqueue: use NR_STD_WORKER_POOLS instead of hardcoded value
workqueue: avoid unguarded 64-bit division
docs: workqueue: document WQ_AFFN_CACHE_SHARD affinity scope
workqueue: add test_workqueue benchmark module
tools/workqueue: add CACHE_SHARD support to wq_dump.py
workqueue: set WQ_AFFN_CACHE_SHARD as the default affinity scope
workqueue: add WQ_AFFN_CACHE_SHARD affinity scope
workqueue: fix typo in WQ_AFFN_SMT comment
workqueue: Remove HK_TYPE_WQ from affecting wq_unbound_cpumask
workqueue: unlink pwqs from wq->pwqs list in alloc_and_link_pwqs() error path
workqueue: Remove NULL wq WARN in __queue_delayed_work()
workqueue: fix parse_affn_scope() prefix matching bug
workqueue: devres: Add device-managed allocate workqueue
workqueue: Add system_dfl_long_wq for long unbound works
tools/workqueue/wq_dump.py: add NODE prefix to all node columns
tools/workqueue/wq_dump.py: fix column alignment in node_nr/max_active section
tools/workqueue/wq_dump.py: remove backslash separator from node_nr/max_active header
efi: Allow to expose the workqueue via sysfs
workqueue: Allow to expose ordered workqueues via sysfs
...
On uniprocessor (UP) configs such as nios2, NR_CPUS is 1, so
cpu_shard_id[] is a single-element array (int[1]). In
llc_populate_cpu_shard_id(), cpumask_first(sibling_cpus) returns an
unsigned int that the compiler cannot prove is always 0, triggering
a -Warray-bounds warning when the result is used to index
cpu_shard_id[]:
kernel/workqueue.c:8321:55: warning: array subscript 1 is above
array bounds of 'int[1]' [-Warray-bounds]
8321 | cpu_shard_id[c] = cpu_shard_id[cpumask_first(sibling_cpus)];
| ~~~~~~~~~~~~^~~~~~~~~~~~~~~~~~~~~~~~~~~~~
This is a false positive: sibling_cpus can never be empty here because
'c' itself is always set in it, so cpumask_first() will always return a
valid CPU. However, the compiler cannot prove this statically, and the
warning only manifests on UP configs where the array size is 1.
Add a bounds check with WARN_ON_ONCE to silence the warning, and store
the result in a local variable to make the code clearer and avoid calling
cpumask_first() twice.
Fixes: 5920d046f7 ("workqueue: add WQ_AFFN_CACHE_SHARD affinity scope")
Reported-by: kernel test robot <lkp@intel.com>
Closes: https://lore.kernel.org/oe-kbuild-all/202604022343.GQtkF2vO-lkp@intel.com/
Signed-off-by: Breno Leitao <leitao@debian.org>
Signed-off-by: Tejun Heo <tj@kernel.org>
use NR_STD_WORKER_POOLS for irq_work_fns[] array definition.
NR_STD_WORKER_POOLS is also 2, but better to use MACRO.
Initialization loop for_each_bh_worker_pool() also uses same MACRO.
Signed-off-by: Maninder Singh <maninder1.s@samsung.com>
Signed-off-by: Tejun Heo <tj@kernel.org>
Set WQ_AFFN_CACHE_SHARD as the default affinity scope for unbound
workqueues. On systems where many CPUs share one LLC, the previous
default (WQ_AFFN_CACHE) collapses all CPUs to a single worker pool,
causing heavy spinlock contention on pool->lock.
WQ_AFFN_CACHE_SHARD subdivides each LLC into smaller groups, providing
a better balance between locality and contention. Users can revert to
the previous behavior with workqueue.default_affinity_scope=cache.
On systems with 8 or fewer cores per LLC, CACHE_SHARD produces a single
shard covering the entire LLC, making it functionally identical to the
previous CACHE default. The sharding only activates when an LLC has more
than 8 cores.
Signed-off-by: Breno Leitao <leitao@debian.org>
Signed-off-by: Tejun Heo <tj@kernel.org>
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>
In unplug_oldest_pwq(), the first inactive work item on the
pool_workqueue is activated correctly. However, if multiple inactive
works exist on the same pool_workqueue, subsequent works fail to
activate because wq_node_nr_active.pending_pwqs is empty — the list
insertion is skipped when the pool_workqueue is plugged.
Fix this by checking for additional inactive works in
unplug_oldest_pwq() and updating wq_node_nr_active.pending_pwqs
accordingly.
Fixes: 4c065dbce1 ("workqueue: Enable unbound cpumask update on ordered workqueues")
Cc: stable@vger.kernel.org
Cc: Carlos Santa <carlos.santa@intel.com>
Cc: Ryan Neph <ryanneph@google.com>
Cc: Lai Jiangshan <jiangshanlai@gmail.com>
Cc: Waiman Long <longman@redhat.com>
Cc: linux-kernel@vger.kernel.org
Signed-off-by: Matthew Brost <matthew.brost@intel.com>
Signed-off-by: Tejun Heo <tj@kernel.org>
Acked-by: Waiman Long <longman@redhat.com>
For historical reason, wq_unbound_cpumask is initially set as
intersection of HK_TYPE_DOMAIN, HK_TYPE_WQ and workqueue.unbound_cpus
boot command line option.
At run time, users can update the unbound cpumask via the
/sys/devices/virtual/workqueue/cpumask sysfs file. Creation
and modification of cpuset isolated partitions will also update
wq_unbound_cpumask based on the latest HK_TYPE_DOMAIN cpumask.
The HK_TYPE_WQ cpumask is out of the picture with these runtime updates.
Complete the transition by taking HK_TYPE_WQ out from the workqueue code
and make it depends on HK_TYPE_DOMAIN only from the housekeeping side.
The final goal is to eliminate HK_TYPE_WQ as a housekeeping cpumask type.
Signed-off-by: Waiman Long <longman@redhat.com>
Acked-by: Frederic Weisbecker <frederic@kernel.org>
Signed-off-by: Tejun Heo <tj@kernel.org>
When alloc_and_link_pwqs() fails partway through the per-cpu allocation
loop, some pool_workqueues may have already been linked into wq->pwqs
via link_pwq(). The error path frees these pwqs with kmem_cache_free()
but never removes them from the wq->pwqs list, leaving dangling pointers
in the list.
Currently this is not exploitable because the workqueue was never added
to the global workqueues list and the caller frees the wq immediately
after. However, this makes sure that alloc_and_link_pwqs() doesn't leave
any half-baked structure, which may have side effects if not properly
cleaned up.
Fix this by unlinking each pwq from wq->pwqs before freeing it. No
locking is needed as the workqueue has not been published yet, thus
no concurrency is possible.
Signed-off-by: Breno Leitao <leitao@debian.org>
Signed-off-by: Tejun Heo <tj@kernel.org>
Try to be more explicit why the workqueue watchdog does not take
pool->lock by default. Spin locks are full memory barriers which
delay anything. Obviously, they would primary delay operations
on the related worker pools.
Explain why it is enough to prevent the false positive by re-checking
the timestamp under the pool->lock.
Finally, make it clear what would be the alternative solution in
__queue_work() which is a hotter path.
Signed-off-by: Petr Mladek <pmladek@suse.com>
Acked-by: Song Liu <song@kernel.org>
Signed-off-by: Tejun Heo <tj@kernel.org>
On weakly ordered architectures (e.g., arm64), the lockless check in
wq_watchdog_timer_fn() can observe a reordering between the worklist
insertion and the last_progress_ts update. Specifically, the watchdog
can see a non-empty worklist (from a list_add) while reading a stale
last_progress_ts value, causing a false positive stall report.
This was confirmed by reading pool->last_progress_ts again after holding
pool->lock in wq_watchdog_timer_fn():
workqueue watchdog: pool 7 false positive detected!
lockless_ts=4784580465 locked_ts=4785033728
diff=453263ms worklist_empty=0
To avoid slowing down the hot path (queue_work, etc.), recheck
last_progress_ts with pool->lock held. This will eliminate the false
positive with minimal overhead.
Remove two extra empty lines in wq_watchdog_timer_fn() as we are on it.
Fixes: 82607adcf9 ("workqueue: implement lockup detector")
Cc: stable@vger.kernel.org # v4.5+
Assisted-by: claude-code:claude-opus-4-6
Signed-off-by: Song Liu <song@kernel.org>
Signed-off-by: Tejun Heo <tj@kernel.org>
parse_affn_scope() uses strncasecmp() with the length of the candidate
name, which means it only checks if the input *starts with* a known
scope name.
Given that the upcoming diff will create "cache_shard" affinity scope,
writing "cache_shard" to a workqueue's affinity_scope sysfs attribute
always matches "cache" first, making it impossible to select
"cache_shard" via sysfs, so, this fix enable it to distinguish "cache"
and "cache_shard"
Fix by replacing the hand-rolled prefix matching loop with
sysfs_match_string(), which uses sysfs_streq() for exact matching
(modulo trailing newlines). Also add the missing const qualifier to
the wq_affn_names[] array declaration.
Note that sysfs_streq() is case-sensitive, unlike the previous
strncasecmp() approach. This is intentional and consistent with
how other sysfs attributes handle string matching in the kernel.
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>
show_cpu_pool_hog() and show_cpu_pools_hogs() no longer only dump CPU
hogs — since commit 8823eaef45 ("workqueue: Show all busy workers in
stall diagnostics"), they dump every in-flight worker in the pool's
busy_hash.
Rename them to show_cpu_pool_busy_workers() and
show_cpu_pools_busy_workers() to accurately describe what they do.
Also fix the pr_info() message to say "stalled worker pools" instead of
"stalled CPU-bound worker pools", since sleeping/blocked workers are now
included.
No functional change.
Suggested-by: Tejun Heo <tj@kernel.org>
Signed-off-by: Breno Leitao <leitao@debian.org>
Signed-off-by: Tejun Heo <tj@kernel.org>
show_cpu_pool_hog() only prints workers whose task is currently running
on the CPU (task_is_running()). This misses workers that are busy
processing a work item but are sleeping or blocked — for example, a
worker that clears PF_WQ_WORKER and enters wait_event_idle(). Such a
worker still occupies a pool slot and prevents progress, yet produces
an empty backtrace section in the watchdog output.
This is happening on real arm64 systems, where
toggle_allocation_gate() IPIs every single CPU in the machine (which
lacks NMI), causing workqueue stalls that show empty backtraces because
toggle_allocation_gate() is sleeping in wait_event_idle().
Remove the task_is_running() filter so every in-flight worker in the
pool's busy_hash is dumped. The busy_hash is protected by pool->lock,
which is already held.
Signed-off-by: Breno Leitao <leitao@debian.org>
Acked-by: Song Liu <song@kernel.org>
Signed-off-by: Tejun Heo <tj@kernel.org>
When diagnosing workqueue stalls, knowing how long each in-flight work
item has been executing is valuable. Add a current_start timestamp
(jiffies) to struct worker, set it when a work item begins execution in
process_one_work(), and print the elapsed wall-clock time in show_pwq().
Unlike current_at (which tracks CPU runtime and resets on wakeup for
CPU-intensive detection), current_start is never reset because the
diagnostic cares about total wall-clock time including sleeps.
Before: in-flight: 165:stall_work_fn [wq_stall]
After: in-flight: 165:stall_work_fn [wq_stall] for 100s
Signed-off-by: Breno Leitao <leitao@debian.org>
Acked-by: Song Liu <song@kernel.org>
Signed-off-by: Tejun Heo <tj@kernel.org>
The watchdog_ts name doesn't convey what the timestamp actually tracks.
This field tracks the last time a workqueue got progress.
Rename it to last_progress_ts to make it clear that it records when the
pool last made forward progress (started processing new work items).
No functional change.
Signed-off-by: Breno Leitao <leitao@debian.org>
Acked-by: Song Liu <song@kernel.org>
Signed-off-by: Tejun Heo <tj@kernel.org>
pr_cont_worker_id() checks pool->flags against WQ_BH, which is a
workqueue-level flag (defined in workqueue.h). Pool flags use a
separate namespace with POOL_* constants (defined in workqueue.c).
The correct constant is POOL_BH. Both WQ_BH and POOL_BH are defined
as (1 << 0) so this has no behavioral impact, but it is semantically
wrong and inconsistent with every other pool-level BH check in the
file.
Fixes: 4cb1ef6460 ("workqueue: Implement BH workqueues to eventually replace tasklets")
Signed-off-by: Breno Leitao <leitao@debian.org>
Acked-by: Song Liu <song@kernel.org>
Signed-off-by: Tejun Heo <tj@kernel.org>
Ordered workqueues are not exposed via sysfs because the 'max_active'
attribute changes the number actives worker. More than one active worker
can break ordering guarantees.
This can be avoided by forbidding writes the file for ordered
workqueues. Exposing it via sysfs allows to alter other attributes such
as the cpumask on which CPU the worker can run.
The 'max_active' value shouldn't be changed for BH worker because the
core never spawns additional worker and the worker itself can not be
preempted. So this make no sense.
Allow to expose ordered workqueues via sysfs if requested and forbid
changing 'max_active' value for ordered and BH worker.
Signed-off-by: Sebastian Andrzej Siewior <bigeasy@linutronix.de>
Acked-by: Tejun Heo <tj@kernel.org>
Acked-by: Ard Biesheuvel <ardb@kernel.org>
Signed-off-by: Tejun Heo <tj@kernel.org>
This is the exact same thing as the 'alloc_obj()' version, only much
smaller because there are a lot fewer users of the *alloc_flex()
interface.
As with alloc_obj() version, this was done entirely with mindless brute
force, using the same script, except using 'flex' in the pattern rather
than 'objs*'.
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
This was done entirely with mindless brute force, using
git grep -l '\<k[vmz]*alloc_objs*(.*, GFP_KERNEL)' |
xargs sed -i 's/\(alloc_objs*(.*\), GFP_KERNEL)/\1)/'
to convert the new alloc_obj() users that had a simple GFP_KERNEL
argument to just drop that argument.
Note that due to the extreme simplicity of the scripting, any slightly
more complex cases spread over multiple lines would not be triggered:
they definitely exist, but this covers the vast bulk of the cases, and
the resulting diff is also then easier to check automatically.
For the same reason the 'flex' versions will be done as a separate
conversion.
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
This is the result of running the Coccinelle script from
scripts/coccinelle/api/kmalloc_objs.cocci. The script is designed to
avoid scalar types (which need careful case-by-case checking), and
instead replace kmalloc-family calls that allocate struct or union
object instances:
Single allocations: kmalloc(sizeof(TYPE), ...)
are replaced with: kmalloc_obj(TYPE, ...)
Array allocations: kmalloc_array(COUNT, sizeof(TYPE), ...)
are replaced with: kmalloc_objs(TYPE, COUNT, ...)
Flex array allocations: kmalloc(struct_size(PTR, FAM, COUNT), ...)
are replaced with: kmalloc_flex(*PTR, FAM, COUNT, ...)
(where TYPE may also be *VAR)
The resulting allocations no longer return "void *", instead returning
"TYPE *".
Signed-off-by: Kees Cook <kees@kernel.org>
Pull workqueue updates from Tejun Heo:
- Rework the rescuer to process work items one-by-one instead of
slurping all pending work items in a single pass.
As there is only one rescuer per workqueue, a single long-blocking
work item could cause high latency for all tasks queued behind it,
even after memory pressure is relieved and regular kworkers become
available to service them.
- Add CONFIG_BOOTPARAM_WQ_STALL_PANIC build-time option and
workqueue.panic_on_stall_time parameter for time-based stall panic,
giving systems more control over workqueue stall handling.
- Replace BUG_ON() with panic() in the stall panic path for clearer
intent and more informative output.
* tag 'wq-for-6.20' of git://git.kernel.org/pub/scm/linux/kernel/git/tj/wq:
workqueue: replace BUG_ON with panic in panic_on_wq_watchdog
workqueue: add time-based panic for stalls
workqueue: add CONFIG_BOOTPARAM_WQ_STALL_PANIC option
workqueue: Process extra works in rescuer on memory pressure
workqueue: Process rescuer work items one-by-one using a cursor
workqueue: Make send_mayday() take a PWQ argument directly
Linus Torvalds
Breno Leitao
Maninder Singh
Matthew Brost
Waiman Long
Petr Mladek
Song Liu
Tejun Heo
Krzysztof Kozlowski
Marco Crivellari
Sebastian Andrzej Siewior
Kees Cook