Lee Jia Jie reported that since event::addr_filter_ranges is used
under RCU, it should be RCU freed.
Reported-by: Lee Jia Jie <jiajie.lee@starlabs.sg>
Signed-off-by: Peter Zijlstra (Intel) <peterz@infradead.org>
Breno reported significant c2c HITM in a futex hash heavy workload.
It turns out that the hash bucket to private hash table reverse pointer
(futex_hash_bucket::priv) was to blame. Notably when the hash buckets are
heavily contended, the: 'fph = bh->priv;' load in futex_hash() will typically
miss and consequently become quite expensive.
Since this load in particular is quite superfluous, removing it is fairly
straight forward. However, removing it does not in fact achieve anything much.
The pain moves to the next user, notably: futex_hash_put().
Therefore rework the whole private hash refcounting to avoid needing this back
pointer (and removing it). Instead of passing around 'struct futex_hash_bucket
*hb', pass around a new structure that contains it and the related 'struct
futex_private_hash *fph' pointer in tandem.
Funnily this turns out to remove more code than it adds and significantly
improves futex hash performance (as measured by 'perf bench futex hash'):
SKL dual socket 112 threads:
Baseline Patched
shared (16k) 1571857 1641435 + 4.4%
autosize (512) 646390 903371 +39.7%
-b 256 464395 587014 +26.4%
-b 512 715687 995943 +39.2%
-b 1024 995085 1396328 +40.3%
-b 2048 1293114 1668395 +29.0%
-b 4096 2124438 2240228 + 5.5%
Zen3 dual socket 256 threads:
Baseline Patched
shared (16k) 1275840 1381279 + 8.2%
autosize (512) 1252745 1482179 +18.3%
-b 256 856274 955455 +11.5%
-b 512 1267490 1544010 +21.8%
-b 1024 1424013 1625424 +14.1%
-b 2048 1505181 1669342 +10.9%
-b 4096 1465993 1688932 +15.2%
AMD EPYC 9D64 (Zen4, single socket) 176 threads:
Baseline Patched Delta
shared (16k) 1,230,599 1,368,655 +11.2%
autosize (1024) 1,285,440 1,556,946 +21.1%
-b 256 1,341,471 1,520,303 +13.3%
-b 512 1,438,330 1,599,319 +11.2%
-b 1024 1,443,772 1,622,493 +12.4%
-b 2048 1,472,108 1,643,975 +11.7%
-b 4096 1,333,098 1,570,897 +17.8%
Reported-by: Breno Leitao <leitao@debian.org>
Signed-off-by: Peter Zijlstra (Intel) <peterz@infradead.org>
Tested-by: Breno Leitao <leitao@debian.org>
Tested-by: Thomas Gleixner <tglx@kernel.org>
Link: https://patch.msgid.link/20260610135510.GB1430057@noisy.programming.kicks-ass.net
assign_cfs_rq_runtime() during update_curr() sets the resched indicator
and relies on check_cfs_rq_runtime() during pick_next_task() /
put_prev_entity() to throttle the hierarchy once current task is
preempted / blocks.
Per-task throttle, on the other hand, uses throttle_cfs_rq() to simply
propagate the throttle signals, and then relies on task work to
individually throttle the runnable tasks on their way out to the
userspace.
Remove check_cfs_rq_runtime() and unify throttling into
account_cfs_rq_runtime() which only sets the cfs_rq->throttled,
cfs_rq->throttle_count indicators via throttle_cfs_rq() and optionally
adds the task work to the current task (donor) it is on the throttled
hierarchy.
throttle_cfs_rq() requests for sched_cfs_bandwidth_slice() worth of
bandwidth for the current hierarchy that enable it to continue running
uninterrupted when selected. For the rest, it requests a bare minimum of
"1" to ensure some bandwidth is available and pass the
"runtime_remaining > 0" checks once selected.
For SCHED_PROXY_EXEC, a mutex holder cannot exit to userspace without
dropping it first and the mutex_unlock() ensures proxy is stopped before
the mutex handoff which preserves the current semantics for running a
throttled task until it exits to the userspace even if it acts as a
donor.
[ prateek: rebased on tip, comments, commit message. ]
Reviewed-By: Benjamin Segall <bsegall@google.com>
Signed-off-by: Peter Zijlstra (Intel) <peterz@infradead.org>
Signed-off-by: K Prateek Nayak <kprateek.nayak@amd.com>
Signed-off-by: Peter Zijlstra (Intel) <peterz@infradead.org>
Tested-by: Aaron Lu <ziqianlu@bytedance.com>
Link: https://patch.msgid.link/20260602071005.11942-1-kprateek.nayak@amd.com
Rather than gate the proxy paths with p->blocked_on, use p->is_blocked.
This opens up the state: '->is_blocked && !->blocked_on' for future use.
Notably, only proxy and delayed tasks can be ->on_rq && ->is_blocked, and it is
guaranteed that sched_class::pick_task() will never return a delayed task.
Therefore any task returned from pick_next_task() that has ->is_blocked set,
must be a proxy task.
Suggested-by: K Prateek Nayak <kprateek.nayak@amd.com>
Signed-off-by: Peter Zijlstra (Intel) <peterz@infradead.org>
Link: https://patch.msgid.link/20260526113322.477954312%40infradead.org
Current code will 'unconditionally' return migrate on PROXY_WAKING, even if the
task is (still) on the original CPU.
Check task_cpu(p) against p->waking_cpu, which per proxy_set_task_cpu()
preserves the original CPU the task was on. If they do not mis-match, there is
no need to go through the more expensive wakeup path.
Signed-off-by: Peter Zijlstra (Intel) <peterz@infradead.org>
Link: https://patch.msgid.link/20260527082916.GP3126523%40noisy.programming.kicks-ass.net
The reason for the clause in try_to_wake_up() is, per its comment, that
find_proxy_task()'s proxy_deactivate() is not always called with a cleared
p->blocked_on.
However, that seems silly and easily cured. Make sure to always call
proxy_deactivate() with a cleared p->blocked_on such that we might remove this
clause from the common wake-up path.
Signed-off-by: Peter Zijlstra (Intel) <peterz@infradead.org>
Acked-by: John Stultz <jstultz@google.com>
Link: https://patch.msgid.link/20260526113322.244729903%40infradead.org
The reason for pick_next_task_fair() is the put/set optimization that
avoids touching the common ancestors. However, it is possible to
implement this in the put_prev_task() and set_next_task() calls as
used in put_prev_set_next_task().
Notably, put_prev_set_next_task() is the only site that:
- calls put_prev_task() with a .next argument;
- calls set_next_task() with .first = true.
This means that put_prev_task() can determine the common hierarchy and
stop there, and then set_next_task() can terminate where put_prev_task
stopped.
Signed-off-by: Peter Zijlstra (Intel) <peterz@infradead.org>
Reviewed-by: Vincent Guittot <vincent.guittot@linaro.org>
Link: https://patch.msgid.link/20260511120628.057634261@infradead.org
With commit 50653216e4 ("sched: Add support to pick functions to
take rf") removing the balance callback, the pick_task() callback is
in charge of newidle balancing.
This means pick_task_fair() should do so too. This hasn't been a
problem in practise because pick_next_task_fair() is used. However,
since we'll be removing that one shortly, make sure pick_next_task()
is up to scratch.
Signed-off-by: Peter Zijlstra (Intel) <peterz@infradead.org>
Reviewed-by: Vincent Guittot <vincent.guittot@linaro.org>
Link: https://patch.msgid.link/20260511120627.944705718@infradead.org
Ian noted that commit 77de62ad3d ("perf/core: Fix refcount bug and
potential UAF in perf_mmap") would cause a deadlock due to
event->mmap_mutex recursion.
This happens because we're now calling perf_mmap_close() under
mmap_mutex, while that function itself can also take mmap_mutex.
Solve this by noting that perf_mmap_close() is far more complicated
than we need at this particular point, since it deals with scenarios
that cannot happen in this particular case.
Replace the call to perf_mmap_close() with a very narrow undo for the
case of first-exposure. If this is not the first mmap(), there is no
race and it is fine to drop the lock and call perf_mmap_close() to
handle to more complicated scenarios.
Note: move the rb->mmap_user (namespace) handling into the rb
init/free code such that it does not complicate the mmap handling.
Fixes: 77de62ad3d ("perf/core: Fix refcount bug and potential UAF in perf_mmap")
Reported-by: Ian Rogers <irogers@google.com>
Closes: https://patch.msgid.link/CAP-5%3DfVJyVMZw%3DDqP53Kxg58nUmJ_0bxoaeOKAbC03BVc11HaA%40mail.gmail.com
Signed-off-by: Peter Zijlstra (Intel) <peterz@infradead.org>
Link: https://patch.msgid.link/20260326112821.GK3738786@noisy.programming.kicks-ass.net
Vincent reported that my rework of his original patch lost a little
something.
Specifically it got the return value wrong; it should not compare
against the old se->vlag, but rather against the current value. Since
the thing that matters is if the effective vruntime of an entity is
affected and the thing needs repositioning or not.
Fixes: 059258b0d4 ("sched/fair: Prevent negative lag increase during delayed dequeue")
Reported-by: Vincent Guittot <vincent.guittot@linaro.org>
Signed-off-by: Peter Zijlstra (Intel) <peterz@infradead.org>
Tested-by: Vincent Guittot <vincent.guittot@linaro.org>
Link: https://patch.msgid.link/20260423094107.GT3102624%40noisy.programming.kicks-ass.net
Adds infrastructure to enable cache-aware load balancing,
which improves cache locality by grouping tasks that share resources
within the same cache domain. This reduces cache misses and improves
overall data access efficiency.
In this initial implementation, threads belonging to the same process
are treated as entities that likely share working sets. The mechanism
tracks per-process CPU occupancy across cache domains and attempts to
migrate threads toward cache-hot domains where their process already
has active threads, thereby enhancing locality.
This provides a basic model for cache affinity. While the current code
targets the last-level cache (LLC), the approach could be extended to
other domain types such as clusters (L2) or node-internal groupings.
At present, the mechanism selects the CPU within an LLC that has the
highest recent runtime. Subsequent patches in this series will use this
information in the load-balancing path to guide task placement toward
preferred LLCs.
In the future, more advanced policies could be integrated through NUMA
balancing-for example, migrating a task to its preferred LLC when spare
capacity exists, or swapping tasks across LLCs to improve cache affinity.
Grouping of tasks could also be generalized from that of a process
to be that of a NUMA group, or be user configurable.
Signed-off-by: Peter Zijlstra (Intel) <peterz@infradead.org>
Signed-off-by: Chen Yu <yu.c.chen@intel.com>
Signed-off-by: Tim Chen <tim.c.chen@linux.intel.com>
Signed-off-by: Peter Zijlstra (Intel) <peterz@infradead.org>
Link: https://patch.msgid.link/6269a53221b9439b9ca00d18a9d1946fb64d8cff.1775065312.git.tim.c.chen@linux.intel.com
Add a (debug) option to virtually split the LLC, no CAT involved, just fake
topology. Used to test code that depends (either in behaviour or directly) on
there being multiple LLC domains in a node.
Signed-off-by: Peter Zijlstra (Intel) <peterz@infradead.org>
John noted that commit 1151354225 ("sched/deadline: Fix 'stuck' dl_server")
unfixed the issue from commit a3a70caf79 ("sched/deadline: Fix dl_server
behaviour").
The issue in commit 1151354225 was for wakeups of the server after the
deadline; in which case you *have* to start a new period. The case for
a3a70caf79 is wakeups before the deadline.
Now, because the server is effectively running a least-laxity policy, it means
that any wakeup during the runnable phase means dl_entity_overflow() will be
true. This means we need to adjust the runtime to allow it to still run until
the existing deadline expires.
Use the revised wakeup rule for dl_defer entities.
Fixes: 1151354225 ("sched/deadline: Fix 'stuck' dl_server")
Reported-by: John Stultz <jstultz@google.com>
Signed-off-by: Peter Zijlstra (Intel) <peterz@infradead.org>
Acked-by: Juri Lelli <juri.lelli@redhat.com>
Tested-by: John Stultz <jstultz@google.com>
Link: https://patch.msgid.link/20260404102244.GB22575@noisy.programming.kicks-ass.net
John reported that stress-ng-yield could make his machine unhappy and
managed to bisect it to commit b3d99f43c7 ("sched/fair: Fix
zero_vruntime tracking").
The commit in question changes avg_vruntime() from a function that is
a pure reader, to a function that updates variables. This turns an
unlocked sched/debug usage of this function from a minor mistake into
a data corruptor.
Fixes: af4cf40470 ("sched/fair: Add cfs_rq::avg_vruntime")
Fixes: b3d99f43c7 ("sched/fair: Fix zero_vruntime tracking")
Reported-by: John Stultz <jstultz@google.com>
Signed-off-by: Peter Zijlstra (Intel) <peterz@infradead.org>
Reviewed-by: Vincent Guittot <vincent.guittot@linaro.org>
Tested-by: K Prateek Nayak <kprateek.nayak@amd.com>
Tested-by: John Stultz <jstultz@google.com>
Link: https://patch.msgid.link/20260401132355.196370805@infradead.org
John reported that stress-ng-yield could make his machine unhappy and
managed to bisect it to commit b3d99f43c7 ("sched/fair: Fix
zero_vruntime tracking").
The combination of yield and that commit was specific enough to
hypothesize the following scenario:
Suppose we have 2 runnable tasks, both doing yield. Then one will be
eligible and one will not be, because the average position must be in
between these two entities.
Therefore, the runnable task will be eligible, and be promoted a full
slice (all the tasks do is yield after all). This causes it to jump over
the other task and now the other task is eligible and current is no
longer. So we schedule.
Since we are runnable, there is no {de,en}queue. All we have is the
__{en,de}queue_entity() from {put_prev,set_next}_task(). But per the
fingered commit, those two no longer move zero_vruntime.
All that moves zero_vruntime are tick and full {de,en}queue.
This means, that if the two tasks playing leapfrog can reach the
critical speed to reach the overflow point inside one tick's worth of
time, we're up a creek.
Additionally, when multiple cgroups are involved, there is no guarantee
the tick will in fact hit every cgroup in a timely manner. Statistically
speaking it will, but that same statistics does not rule out the
possibility of one cgroup not getting a tick for a significant amount of
time -- however unlikely.
Therefore, just like with the yield() case, force an update at the end
of every slice. This ensures the update is never more than a single
slice behind and the whole thing is within 2 lag bounds as per the
comment on entity_key().
Fixes: b3d99f43c7 ("sched/fair: Fix zero_vruntime tracking")
Reported-by: John Stultz <jstultz@google.com>
Signed-off-by: Peter Zijlstra (Intel) <peterz@infradead.org>
Reviewed-by: Vincent Guittot <vincent.guittot@linaro.org>
Tested-by: K Prateek Nayak <kprateek.nayak@amd.com>
Tested-by: John Stultz <jstultz@google.com>
Link: https://patch.msgid.link/20260401132355.081530332@infradead.org
Nicholas reported that his LLM found it was possible to create a UaF
when sys_futex_requeue() is used with different flags. The initial
motivation for allowing different flags was the variable sized futex,
but since that hasn't been merged (yet), simply mandate the flags are
identical, as is the case for the old style sys_futex() requeue
operations.
Fixes: 0f4b5f9722 ("futex: Add sys_futex_requeue()")
Reported-by: Nicholas Carlini <npc@anthropic.com>
Signed-off-by: Peter Zijlstra (Intel) <peterz@infradead.org>
Commit 8e8e23dea4 ("sched/topology: Compute sd_weight considering
cpuset partitions") ends up relying on the fact that structure
initialization should not touch the flexible array.
However, the official GCC specification for "Arrays of Length Zero"
[*] says:
Although the size of a zero-length array is zero, an array member of
this kind may increase the size of the enclosing type as a result of
tail padding.
Additionally, structure initialization will zero tail padding. With
the end result that since offsetof(*type, member) < sizeof(*type),
array initialization will clobber the flex array.
Luckily, the way flexible array sizes are calculated is:
sizeof(*type) + count * sizeof(*type->member)
This means we have the complete size of the flex array *outside* of
sizeof(*type), so use that instead of relying on the broken flex array
definition.
[*] https://gcc.gnu.org/onlinedocs/gcc/Zero-Length.html
Fixes: 8e8e23dea4 ("sched/topology: Compute sd_weight considering cpuset partitions")
Reported-by: Nathan Chancellor <nathan@kernel.org>
Debugged-by: K Prateek Nayak <kprateek.nayak@amd.com>
Signed-off-by: Peter Zijlstra (Intel) <peterz@infradead.org>
Tested-by: Jon Hunter <jonathanh@nvidia.com>
Tested-by: Chen Yu <yu.c.chen@intel.com>
Tested-by: K Prateek Nayak <kprateek.nayak@amd.com>
Tested-by: Nathan Chancellor <nathan@kernel.org>
Link: https://patch.msgid.link/20260323093627.GY3738010@noisy.programming.kicks-ass.net
Andrew reported that a guard() conversion of zone_lock increased the
code size unnecessarily.
It turns out the unconditional __GUARD_IS_ERR() is to blame. As
explored earlier [1], __GUARD_IS_ERR(), similar to IS_ERR_OR_NULL(),
generates somewhat sub-optimal code.
However, looking at things again, it is possible to avoid doing the
__GUARD_IS_ERR() unconditionally. Revert the normal destructors to a
simple NULL test and only add the IS_ERR bit to COND guards.
This cures the reported overhead; as compiled by GCC-16:
page_alloc.o:
pre: Total: Before=45299, After=45371, chg +0.16%
post: Total: Before=45299, After=45026, chg -0.60%
[1] https://lkml.kernel.org/r/20250513085001.GC25891@noisy.programming.kicks-ass.net
Reported-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Peter Zijlstra (Intel) <peterz@infradead.org>
Tested-by: Dan Williams <dan.j.williams@intel.com>
Link: https://patch.msgid.link/20260309164516.GE606826@noisy.programming.kicks-ass.net
Oliver reported that x86_pmu_del() ended up doing an out-of-bound memory access
when group_sched_in() fails and needs to roll back.
This *should* be handled by the transaction callbacks, but he found that when
the group leader is a software event, the transaction handlers of the wrong PMU
are used. Despite the move_group case in perf_event_open() and group_sched_in()
using pmu_ctx->pmu.
Turns out, inherit uses event->pmu to clone the events, effectively undoing the
move_group case for all inherited contexts. Fix this by also making inherit use
pmu_ctx->pmu, ensuring all inherited counters end up in the same pmu context.
Similarly, __perf_event_read() should use equally use pmu_ctx->pmu for the
group case.
Fixes: bd27568117 ("perf: Rewrite core context handling")
Reported-by: Oliver Rosenberg <olrose55@gmail.com>
Signed-off-by: Peter Zijlstra (Intel) <peterz@infradead.org>
Reviewed-by: Ian Rogers <irogers@google.com>
Link: https://patch.msgid.link/20260309133713.GB606826@noisy.programming.kicks-ass.net
Both Mi Dapeng and Ian Rogers noted that not everything that sets HES_STOPPED
is required to EF_UPDATE. Specifically the 'step 1' loop of rescheduling
explicitly does EF_UPDATE to ensure the counter value is read.
However, then 'step 2' simply leaves the new counter uninitialized when
HES_STOPPED, even though, as noted above, the thing that stopped them might not
be aware it needs to EF_RELOAD -- since it didn't EF_UPDATE on stop.
One such location that is affected is throttling, throttle does pmu->stop(, 0);
and unthrottle does pmu->start(, 0); possibly restarting an uninitialized counter.
Fixes: a4eaf7f146 ("perf: Rework the PMU methods")
Reported-by: Dapeng Mi <dapeng1.mi@linux.intel.com>
Reported-by: Ian Rogers <irogers@google.com>
Signed-off-by: Peter Zijlstra (Intel) <peterz@infradead.org>
Reviewed-by: Dapeng Mi <dapeng1.mi@linux.intel.com>
Link: https://patch.msgid.link/20260311204035.GX606826@noisy.programming.kicks-ass.net
When the hrtimer_interrupt needs to restart more than 3 times and still has
expired timers, the interrupt is considered hung. To give the system a
little time to recover, the hardware timer is programmed a little into the
future.
Prior to commit 2889243848 ("hrtimer: Re-arrange hrtimer_interrupt()"),
this was relative to the amount of time spend serving the interrupt with a
max of 100 msec.
However, in order to simplify, and because this condition 'should' not
happen, the timeout was unconditionally set to 100 msec.
'Obviously' there is a benchmark that hits this hard, by programming a
ton of very short timers :-/
Since reprogramming is decoupled from the interrupt handling, the actual
execution time is lost, however the code does track max_hang_time. Using
that, rather than the 100 ms max restores performance.
stress-ng --timeout 60 --times --verify --metrics --no-rand-seed --timermix 64
bogo ops/s
288924384856^1: 23715979.93
2889243848: 11550049.77
patched: 23361116.78
Additionally, Thomas noted that cpu_base->hang_detected should not be
cleared until the next interrupt, such that __hrtimer_reprogram() won't
undo the extra delay.
Fixes: 2889243848 ("hrtimer: Re-arrange hrtimer_interrupt()")
Reported-by: kernel test robot <oliver.sang@intel.com>
Signed-off-by: Peter Zijlstra (Intel) <peterz@infradead.org>
Signed-off-by: Thomas Gleixner <tglx@kernel.org>
Link: https://patch.msgid.link/20260311121500.GF652779@noisy.programming.kicks-ass.net
Closes: https://lore.kernel.org/oe-lkp/202603102229.74b9dee4-lkp@intel.com
Use the MADT and SRAT table data to compute __num_nodes_per_package.
Specifically, SRAT has already been parsed in x86_numa_init(), which is called
before acpi_boot_init() which parses MADT. So both are available in
topology_init_possible_cpus().
This number is useful to divinate the various Intel CoD/SNC and AMD NPS modes,
since the platforms are failing to provide this otherwise.
Doing it this way is independent of the number of online CPUs and
other such shenanigans.
Signed-off-by: Peter Zijlstra (Intel) <peterz@infradead.org>
Reviewed-by: Ingo Molnar <mingo@kernel.org>
Tested-by: Tony Luck <tony.luck@intel.com>
Tested-by: K Prateek Nayak <kprateek.nayak@amd.com>
Tested-by: Zhang Rui <rui.zhang@intel.com>
Tested-by: Chen Yu <yu.c.chen@intel.com>
Tested-by: Kyle Meyer <kyle.meyer@hpe.com>
Link: https://patch.msgid.link/20260303110100.004091624@infradead.org
The deferred rearm of the clock event device after an interrupt and and
other hrtimer optimizations allow now to enable HRTICK for generic entry
architectures.
This decouples preemption from CONFIG_HZ, leaving only the periodic
load-balancer and various accounting things relying on the tick.
Signed-off-by: Peter Zijlstra (Intel) <peterz@infradead.org>
Signed-off-by: Thomas Gleixner <tglx@kernel.org>
Signed-off-by: Peter Zijlstra (Intel) <peterz@infradead.org>
Link: https://patch.msgid.link/20260224163431.937531564@kernel.org
Currently hrtimer_interrupt() runs expired timers, which can re-arm
themselves, after which it computes the next expiration time and
re-programs the hardware.
However, things like HRTICK, a highres timer driving preemption, cannot
re-arm itself at the point of running, since the next task has not been
determined yet. The schedule() in the interrupt return path will switch to
the next task, which then causes a new hrtimer to be programmed.
This then results in reprogramming the hardware at least twice, once after
running the timers, and once upon selecting the new task.
Notably, *both* events happen in the interrupt.
By pushing the hrtimer reprogram all the way into the interrupt return
path, it runs after schedule() picks the new task and the double reprogram
can be avoided.
Signed-off-by: Peter Zijlstra (Intel) <peterz@infradead.org>
Signed-off-by: Thomas Gleixner <tglx@kernel.org>
Signed-off-by: Peter Zijlstra (Intel) <peterz@infradead.org>
Link: https://patch.msgid.link/20260224163431.273488269@kernel.org
The hrtimer interrupt expires timers and at the end of the interrupt it
rearms the clockevent device for the next expiring timer.
That's obviously correct, but in the case that a expired timer sets
NEED_RESCHED the return from interrupt ends up in schedule(). If HRTICK is
enabled then schedule() will modify the hrtick timer, which causes another
reprogramming of the hardware.
That can be avoided by deferring the rearming to the return from interrupt
path and if the return results in a immediate schedule() invocation then it
can be deferred until the end of schedule(), which avoids multiple rearms
and re-evaluation of the timer wheel.
Add the rearm checks to the existing sched_hrtick_enter/exit() functions,
which already handle the batched rearm of the hrtick timer.
For now this is just placing empty stubs at the right places which are all
optimized out by the compiler until the guard condition becomes true.
Signed-off-by: Peter Zijlstra (Intel) <peterz@infradead.org>
Signed-off-by: Thomas Gleixner <tglx@kernel.org>
Signed-off-by: Peter Zijlstra (Intel) <peterz@infradead.org>
Link: https://patch.msgid.link/20260224163431.208580085@kernel.org
The hrtimer interrupt expires timers and at the end of the interrupt it
rearms the clockevent device for the next expiring timer.
That's obviously correct, but in the case that a expired timer sets
NEED_RESCHED the return from interrupt ends up in schedule(). If HRTICK is
enabled then schedule() will modify the hrtick timer, which causes another
reprogramming of the hardware.
That can be avoided by deferring the rearming to the return from interrupt
path and if the return results in a immediate schedule() invocation then it
can be deferred until the end of schedule(), which avoids multiple rearms
and re-evaluation of the timer wheel.
In case that the return from interrupt ends up handling softirqs before
reaching the rearm conditions in the return to user entry code functions, a
deferred rearm has to be handled before softirq handling enables interrupts
as soft interrupt handling can be long and would therefore introduce hard
to diagnose latencies to the timer interrupt.
Place the for now empty stub call right before invoking the softirq
handling routine.
Signed-off-by: Peter Zijlstra (Intel) <peterz@infradead.org>
Signed-off-by: Thomas Gleixner <tglx@kernel.org>
Signed-off-by: Peter Zijlstra (Intel) <peterz@infradead.org>
Link: https://patch.msgid.link/20260224163431.142854488@kernel.org
The hrtimer interrupt expires timers and at the end of the interrupt it
rearms the clockevent device for the next expiring timer.
That's obviously correct, but in the case that a expired timer sets
NEED_RESCHED the return from interrupt ends up in schedule(). If HRTICK is
enabled then schedule() will modify the hrtick timer, which causes another
reprogramming of the hardware.
That can be avoided by deferring the rearming to the return from interrupt
path and if the return results in a immediate schedule() invocation then it
can be deferred until the end of schedule(), which avoids multiple rearms
and re-evaluation of the timer wheel.
As this is only relevant for interrupt to user return split the work masks
up and hand them in as arguments from the relevant exit to user functions,
which allows the compiler to optimize the deferred handling out for the
syscall exit to user case.
Add the rearm checks to the approritate places in the exit to user loop and
the interrupt return to kernel path, so that the rearming is always
guaranteed.
In the return to user space path this is handled in the same way as
TIF_RSEQ to avoid extra instructions in the fast path, which are truly
hurtful for device interrupt heavy work loads as the extra instructions and
conditionals while benign at first sight accumulate quickly into measurable
regressions. The return from syscall path is completely unaffected due to
the above mentioned split so syscall heavy workloads wont have any extra
burden.
For now this is just placing empty stubs at the right places which are all
optimized out by the compiler until the actual functionality is in place.
Signed-off-by: Peter Zijlstra (Intel) <peterz@infradead.org>
Signed-off-by: Thomas Gleixner <tglx@kernel.org>
Signed-off-by: Peter Zijlstra (Intel) <peterz@infradead.org>
Link: https://patch.msgid.link/20260224163431.066469985@kernel.org
The hrtimer interrupt expires timers and at the end of the interrupt it
rearms the clockevent device for the next expiring timer.
That's obviously correct, but in the case that a expired timer set
NEED_RESCHED the return from interrupt ends up in schedule(). If HRTICK is
enabled then schedule() will modify the hrtick timer, which causes another
reprogramming of the hardware.
That can be avoided by deferring the rearming to the return from interrupt
path and if the return results in a immediate schedule() invocation then it
can be deferred until the end of schedule().
To make this correct the affected code parts need to be made aware of this.
Provide empty stubs for the deferred rearming mechanism, so that the
relevant code changes for entry, softirq and scheduler can be split up into
separate changes independent of the actual enablement in the hrtimer code.
Signed-off-by: Peter Zijlstra (Intel) <peterz@infradead.org>
Signed-off-by: Thomas Gleixner <tglx@kernel.org>
Signed-off-by: Peter Zijlstra (Intel) <peterz@infradead.org>
Link: https://patch.msgid.link/20260224163431.000891171@kernel.org
Rework hrtimer_interrupt() such that reprogramming is split out into an
independent function at the end of the interrupt.
This prepares for reprogramming getting delayed beyond the end of
hrtimer_interrupt().
Notably, this changes the hang handling to always wait 100ms instead of
trying to keep it proportional to the actual delay. This simplifies the
state, also this really shouldn't be happening.
Signed-off-by: Peter Zijlstra (Intel) <peterz@infradead.org>
Signed-off-by: Thomas Gleixner <tglx@kernel.org>
Signed-off-by: Peter Zijlstra (Intel) <peterz@infradead.org>
Link: https://patch.msgid.link/20260224163430.870639266@kernel.org
The hrtick timer is frequently rearmed before expiry and most of the time
the new expiry is past the armed one. As this happens on every context
switch it becomes expensive with scheduling heavy work loads especially in
virtual machines as the "hardware" reprogamming implies a VM exit.
hrtimer now provide a lazy rearm mode flag which skips the reprogamming if:
1) The timer was the first expiring timer before the rearm
2) The new expiry time is farther out than the armed time
This avoids a massive amount of reprogramming operations of the hrtick
timer for the price of eventually taking the alredy armed interrupt for
nothing.
Mark the hrtick timer accordingly.
Signed-off-by: Peter Zijlstra (Intel) <peterz@infradead.org>
Signed-off-by: Thomas Gleixner <tglx@kernel.org>
Signed-off-by: Peter Zijlstra (Intel) <peterz@infradead.org>
Link: https://patch.msgid.link/20260224163429.475409346@kernel.org
The hrtick timer is frequently rearmed before expiry and most of the time
the new expiry is past the armed one. As this happens on every context
switch it becomes expensive with scheduling heavy work loads especially in
virtual machines as the "hardware" reprogamming implies a VM exit.
Add a lazy rearm mode flag which skips the reprogamming if:
1) The timer was the first expiring timer before the rearm
2) The new expiry time is farther out than the armed time
This avoids a massive amount of reprogramming operations of the hrtick
timer for the price of eventually taking the alredy armed interrupt for
nothing.
Signed-off-by: Peter Zijlstra (Intel) <peterz@infradead.org>
Signed-off-by: Thomas Gleixner <tglx@kernel.org>
Signed-off-by: Peter Zijlstra (Intel) <peterz@infradead.org>
Link: https://patch.msgid.link/20260224163429.408524456@kernel.org
The nominal duration for an EEVDF task to run is until its deadline. At
which point the deadline is moved ahead and a new task selection is done.
Try and predict the time 'lost' to higher scheduling classes. Since this is
an estimate, the timer can be both early or late. In case it is early
task_tick_fair() will take the !need_resched() path and restarts the timer.
Signed-off-by: Peter Zijlstra (Intel) <peterz@infradead.org>
Signed-off-by: Thomas Gleixner <tglx@kernel.org>
Signed-off-by: Peter Zijlstra (Intel) <peterz@infradead.org>
Reviewed-by: Juri Lelli <juri.lelli@redhat.com>
Link: https://patch.msgid.link/20260224163428.798198874@kernel.org
Make sure that __perf_event_overflow() runs with IRQs disabled for all
possible callchains. Specifically the software events can end up running
it with only preemption disabled.
This opens up a race vs perf_event_exit_event() and friends that will go
and free various things the overflow path expects to be present, like
the BPF program.
Fixes: 592903cdcb ("perf_counter: add an event_list")
Reported-by: Simond Hu <cmdhh1767@gmail.com>
Signed-off-by: Peter Zijlstra (Intel) <peterz@infradead.org>
Tested-by: Simond Hu <cmdhh1767@gmail.com>
Link: https://patch.msgid.link/20260224122909.GV1395416@noisy.programming.kicks-ass.net
Zicheng Qu reported that, because avg_vruntime() always includes
cfs_rq->curr, when ->on_rq, place_entity() doesn't work right.
Specifically, the lag scaling in place_entity() relies on
avg_vruntime() being the state *before* placement of the new entity.
However in this case avg_vruntime() will actually already include the
entity, which breaks things.
Also, Zicheng Qu argues that avg_vruntime should be invariant under
reweight. IOW commit 6d71a9c616 ("sched/fair: Fix EEVDF entity
placement bug causing scheduling lag") was wrong!
The issue reported in 6d71a9c616 could possibly be explained by
rounding artifacts -- notably the extreme weight '2' is outside of the
range of avg_vruntime/sum_w_vruntime, since that uses
scale_load_down(). By scaling vruntime by the real weight, but
accounting it in vruntime with a factor 1024 more, the average moves
significantly. However, that is now cured.
Tested by reverting 66951e4860 ("sched/fair: Fix update_cfs_group()
vs DELAY_DEQUEUE") and tracing vruntime and vlag figures again.
Reported-by: Zicheng Qu <quzicheng@huawei.com>
Signed-off-by: Peter Zijlstra (Intel) <peterz@infradead.org>
Reviewed-by: Vincent Guittot <vincent.guittot@linaro.org>
Tested-by: K Prateek Nayak <kprateek.nayak@amd.com>
Tested-by: Shubhang Kaushik <shubhang@os.amperecomputing.com>
Link: https://patch.msgid.link/20260219080625.066102672%40infradead.org
Due to the zero_vruntime patch, the deltas are now a lot smaller and
measurement with kernel-build and hackbench runs show about 45 bits
used.
This ensures avg_vruntime() tracks the full weight range, reducing
numerical artifacts in reweight and the like.
Also, lets keep the paranoid debug code around fow now.
Signed-off-by: Peter Zijlstra (Intel) <peterz@infradead.org>
Tested-by: K Prateek Nayak <kprateek.nayak@amd.com>
Tested-by: Shubhang Kaushik <shubhang@os.amperecomputing.com>
Link: https://patch.msgid.link/20260219080624.942813440%40infradead.org
It turns out that a few workloads (easyWave, fio) have a fairly low
success rate on newidle balance, but still benefit greatly from having
it anyway.
Luckliky these workloads have a faily low newidle rate, so the cost if
doing the newidle is relatively low, even if unsuccessfull.
Add a simple rate based part to the newidle ratio compute, such that
low rate newidle will still have a high newidle ratio.
This cures the easyWave and fio workloads while not affecting the
schbench numbers either (which have a very high newidle rate).
Reported-by: Mario Roy <marioeroy@gmail.com>
Reported-by: "Mohamed Abuelfotoh, Hazem" <abuehaze@amazon.com>
Signed-off-by: Peter Zijlstra (Intel) <peterz@infradead.org>
Tested-by: Mario Roy <marioeroy@gmail.com>
Tested-by: "Mohamed Abuelfotoh, Hazem" <abuehaze@amazon.com>
Link: https://patch.msgid.link/20260127151748.GA1079264@noisy.programming.kicks-ass.net
Kernel test robot reported that
tools/testing/selftests/kvm/hardware_disable_test was failing due to
commit 704069649b ("sched/core: Rework sched_class::wakeup_preempt()
and rq_modified_*()")
It turns out there were two related problems that could lead to a
missed preemption:
- when hitting newidle balance from the idle thread, it would elevate
rb->next_class from &idle_sched_class to &fair_sched_class, causing
later wakeup_preempt() calls to not hit the sched_class_above()
case, and not issue resched_curr().
Notably, this modification pattern should only lower the
next_class, and never raise it. Create two new helper functions to
wrap this.
- when doing schedule_idle(), it was possible to miss (re)setting
rq->next_class to &idle_sched_class, leading to the very same
problem.
Cc: Sean Christopherson <seanjc@google.com>
Fixes: 704069649b ("sched/core: Rework sched_class::wakeup_preempt() and rq_modified_*()")
Reported-by: kernel test robot <oliver.sang@intel.com>
Closes: https://lore.kernel.org/oe-lkp/202602122157.4e861298-lkp@intel.com
Signed-off-by: Peter Zijlstra (Intel) <peterz@infradead.org>
Link: https://patch.msgid.link/20260218163329.GQ1395416@noisy.programming.kicks-ass.net
It turns out that zero_vruntime tracking is broken when there is but a single
task running. Current update paths are through __{en,de}queue_entity(), and
when there is but a single task, pick_next_task() will always return that one
task, and put_prev_set_next_task() will end up in neither function.
This can cause entity_key() to grow indefinitely large and cause overflows,
leading to much pain and suffering.
Furtermore, doing update_zero_vruntime() from __{de,en}queue_entity(), which
are called from {set_next,put_prev}_entity() has problems because:
- set_next_entity() calls __dequeue_entity() before it does cfs_rq->curr = se.
This means the avg_vruntime() will see the removal but not current, missing
the entity for accounting.
- put_prev_entity() calls __enqueue_entity() before it does cfs_rq->curr =
NULL. This means the avg_vruntime() will see the addition *and* current,
leading to double accounting.
Both cases are incorrect/inconsistent.
Noting that avg_vruntime is already called on each {en,de}queue, remove the
explicit avg_vruntime() calls (which removes an extra 64bit division for each
{en,de}queue) and have avg_vruntime() update zero_vruntime itself.
Additionally, have the tick call avg_vruntime() -- discarding the result, but
for the side-effect of updating zero_vruntime.
While there, optimize avg_vruntime() by noting that the average of one value is
rather trivial to compute.
Test case:
# taskset -c -p 1 $$
# taskset -c 2 bash -c 'while :; do :; done&'
# cat /sys/kernel/debug/sched/debug | awk '/^cpu#/ {P=0} /^cpu#2,/ {P=1} {if (P) print $0}' | grep -e zero_vruntime -e "^>"
PRE:
.zero_vruntime : 31316.407903
>R bash 487 50787.345112 E 50789.145972 2.800000 50780.298364 16 120 0.000000 0.000000 0.000000 /
.zero_vruntime : 382548.253179
>R bash 487 427275.204288 E 427276.003584 2.800000 427268.157540 23 120 0.000000 0.000000 0.000000 /
POST:
.zero_vruntime : 17259.709467
>R bash 526 17259.709467 E 17262.509467 2.800000 16915.031624 9 120 0.000000 0.000000 0.000000 /
.zero_vruntime : 18702.723356
>R bash 526 18702.723356 E 18705.523356 2.800000 18358.045513 9 120 0.000000 0.000000 0.000000 /
Fixes: 79f3f9bedd ("sched/eevdf: Fix min_vruntime vs avg_vruntime")
Reported-by: K Prateek Nayak <kprateek.nayak@amd.com>
Signed-off-by: Peter Zijlstra (Intel) <peterz@infradead.org>
Tested-by: K Prateek Nayak <kprateek.nayak@amd.com>
Tested-by: Shubhang Kaushik <shubhang@os.amperecomputing.com>
Link: https://patch.msgid.link/20260219080624.438854780%40infradead.org
Rustam reported his clang builds did not boot properly; turns out his
.config has: CONFIG_DEBUG_FORCE_FUNCTION_ALIGN_64B=y set.
Fix up the FineIBT code to deal with this unusual alignment.
Fixes: 931ab63664 ("x86/ibt: Implement FineIBT")
Reported-by: Rustam Kovhaev <rkovhaev@gmail.com>
Signed-off-by: Peter Zijlstra (Intel) <peterz@infradead.org>
Tested-by: Rustam Kovhaev <rkovhaev@gmail.com>
There are two problems with sched_server_write_common() that can cause the
dl_server to malfunction upon attempting to change the parameters:
1) when, after having disabled the dl_server by setting runtime=0, it is
enabled again while tasks are already enqueued. In this case is_active would
still be 0 and dl_server_start() would not be called.
2) when dl_server_apply_params() would fail, runtime is not applied and does
not reflect the new state.
Instead have dl_server_start() check its actual dl_runtime, and have
sched_server_write_common() unconditionally (re)start the dl_server. It will
automatically stop if there isn't anything to do, so spurious activation is
harmless -- while failing to start it is a problem.
While there, move the printk out of the locked region and make it symmetric,
also printing on enable.
Signed-off-by: Peter Zijlstra (Intel) <peterz@infradead.org>
Link: https://patch.msgid.link/20260203103407.GK1282955@noisy.programming.kicks-ass.net
Andrea reported the dl_server getting stuck for him. He tracked it
down to a state where dl_server_start() saw dl_defer_running==1, but
the dl_server's job is no longer valid at the time of
dl_server_start().
In the state diagram this corresponds to [4] D->A (or dl_server_stop()
due to no more runnable tasks) followed by [1], which in case of a
lapsed deadline must then be A->B.
Now our A has dl_defer_running==1, while B demands
dl_defer_running==0, therefore it must get cleared when the CBS wakeup
rules demand a replenish.
Fixes: a110a81c52 ("sched/deadline: Deferrable dl server")
Reported-by: Andrea Righi arighi@nvidia.com
Signed-off-by: Peter Zijlstra (Intel) <peterz@infradead.org>
Acked-by: Juri Lelli <juri.lelli@redhat.com>
Tested-by: Andrea Righi arighi@nvidia.com
Link: https://lkml.kernel.org/r/20260123161645.2181752-1-arighi@nvidia.com
Link: https://patch.msgid.link/20260130124100.GC1079264@noisy.programming.kicks-ass.net
Change the minimum slice extension to 5 usec.
Since slice_test selftest reaches a staggering ~350 nsec extension:
Task: slice_test Mean: 350.266 ns
Latency (us) | Count
------------------------------
EXPIRED | 238
0 us | 143189
1 us | 167
2 us | 26
3 us | 11
4 us | 28
5 us | 31
6 us | 22
7 us | 23
8 us | 32
9 us | 16
10 us | 35
Lower the minimal (and default) value to 5 usecs -- which is still massive.
Signed-off-by: Peter Zijlstra (Intel) <peterz@infradead.org>
Link: https://patch.msgid.link/20260121143208.073200729@infradead.org
The recent changes to get_unaligned() resulted in a new sparse warning:
net/rds/ib_cm.c:96:35: sparse: sparse: incorrect type in argument 1 (different modifiers) @@ expected void * @@ got restricted __be64 const * @@
net/rds/ib_cm.c:96:35: sparse: expected void *
net/rds/ib_cm.c:96:35: sparse: got restricted __be64 const *
The updated get_unaligned_t() uses __unqual_scalar_typeof() to get an
unqualified type. This works correctly for the compilers, but fails for
sparse when the data type is __be64 (or any other __beNN variant).
On sparse runs (C=[12]) __beNN types are annotated with
__attribute__((bitwise)).
That annotation allows sparse to detect incompatible operations on __beNN
variables, but it also prevents sparse from evaluating the _Generic() in
__unqual_scalar_typeof() and map __beNN to a unqualified scalar type, so it
ends up with the default, i.e. the original qualified type of a 'const
__beNN' pointer. That then ends up as the first pointer argument to
builtin_memcpy(), which obviously causes the above sparse warnings.
The sparse git tree supports typeof_unqual() now, which allows to use it
instead of the _Generic() based __unqual_scalar_typeof(). With that sparse
correctly evaluates the unqualified type and keeps the __beNN logic intact.
The downside is that this requires a top of tree sparse build and an old
sparse version will emit a metric ton of incomprehensible error messages
before it dies with a segfault.
Therefore implement a sanity check which validates that the checker is
available and capable of handling typeof_unqual(). Emit a warning if not so
the user can take informed action.
[ tglx: Move the evaluation of USE_TYPEOF_UNQUAL to compiler_types.h so it is
set before use and implement the sanity checker ]
Reported-by: kernel test robot <lkp@intel.com>
Signed-off-by: Peter Zijlstra <peterz@infradead.org>
Signed-off-by: Thomas Gleixner <tglx@kernel.org>
Acked-by: Ian Rogers <irogers@google.com>
Link: https://patch.msgid.link/87ecnp2zh3.ffs@tglx
Closes: https://lore.kernel.org/oe-kbuild-all/202601150001.sKSN644a-lkp@intel.com/
Pierre reported hitting balance callback warnings for deadline tasks
after commit 6455ad5346 ("sched: Move sched_class::prio_changed()
into the change pattern").
It turns out that DEQUEUE_SAVE+ENQUEUE_RESTORE does not preserve DL
priority and subsequently trips a balance pass -- where one was not
expected.
From discussion with Juri and Luca, the purpose of this clause was to
deal with tasks new to DL and all those sites will have MOVE set (as
well as CLASS, but MOVE is move conservative at this point).
Per the previous patches MOVE is audited to always run the balance
callbacks, so switch enqueue_dl_entity() to use MOVE for this case.
Fixes: 6455ad5346 ("sched: Move sched_class::prio_changed() into the change pattern")
Reported-by: Pierre Gondois <pierre.gondois@arm.com>
Signed-off-by: Peter Zijlstra (Intel) <peterz@infradead.org>
Tested-by: Pierre Gondois <pierre.gondois@arm.com>
Tested-by: Juri Lelli <juri.lelli@redhat.com>
Link: https://patch.msgid.link/20260114130528.GB831285@noisy.programming.kicks-ass.net
Pratheek tripped a WARN and noted the following issue:
> Inspecting the set of events that led to the warning being triggered
> showed the following:
>
> systemd-1 [008] dN.31 ...: do_set_cpus_allowed: set_cpus_allowed begin!
>
> systemd-1 [008] dN.31 ...: sched_change_begin: Begin!
> systemd-1 [008] dN.31 ...: sched_change_begin: Before dequeue_task()!
> systemd-1 [008] dN.31 ...: update_curr_dl_se: update_curr_dl_se: ENQUEUE_REPLENISH
> systemd-1 [008] dN.31 ...: enqueue_dl_entity: enqueue_dl_entity: ENQUEUE_REPLENISH
> systemd-1 [008] dN.31 ...: replenish_dl_entity: Replenish before: 14815760217
> systemd-1 [008] dN.31 ...: replenish_dl_entity: Replenish after: 14816960047
> systemd-1 [008] dN.31 ...: sched_change_begin: Before put_prev_task()!
>
> systemd-1 [008] dN.31 ...: sched_change_end: Before enqueue_task()!
> systemd-1 [008] dN.31 ...: sched_change_end: Before put_prev_task()!
> systemd-1 [008] dN.31 ...: prio_changed_dl: Queuing pull task on prio change: 14815760217 -> 14816960047
> systemd-1 [008] dN.31 ...: prio_changed_dl: Queuing balance callback!
> systemd-1 [008] dN.31 ...: sched_change_end: End!
>
> systemd-1 [008] dN.31 ...: do_set_cpus_allowed: set_cpus_allowed end!
> systemd-1 [008] dN.21 ...: __schedule: Woops! Balance callback found!
>
> 1. sched_change_begin() from guard(sched_change) in
> do_set_cpus_allowed() stashes the priority, which for the deadline
> task, is "p->dl.deadline".
> 2. The dequeue of the deadline task replenishes the deadline.
> 3. The task is enqueued back after guard's scope ends and since there is
> no *_CLASS flags set, sched_change_end() calls
> dl_sched_class->prio_changed() which compares the deadline.
> 4. Since deadline was moved on dequeue, prio_changed_dl() sees the value
> differ from the stashed value and queues a balance pull callback.
> 5. do_set_cpus_allowed() finishes and drops the rq_lock without doing a
> do_balance_callbacks().
> 6. Grabbing the rq_lock() at subsequent __schedule() triggers the
> warning since the balance pull callback was never executed before
> dropping the lock.
Meaning get_prio_dl() ought to update current and return an up-to-date
value.
Fixes: 6455ad5346 ("sched: Move sched_class::prio_changed() into the change pattern")
Reported-by: K Prateek Nayak <kprateek.nayak@amd.com>
Signed-off-by: Peter Zijlstra (Intel) <peterz@infradead.org>
Reviewed-by: K Prateek Nayak <kprateek.nayak@amd.com>
Tested-by: K Prateek Nayak <kprateek.nayak@amd.com>
Link: https://patch.msgid.link/20260106104113.GX3707891@noisy.programming.kicks-ass.net
A fix for the dl_server 'requires' idle_cpu() usage, which made me
note that it and available_idle_cpu() are extern function calls.
And while idle_cpu() is used outside of kernel/sched/,
available_idle_cpu() is not.
This makes it hard to make idle_cpu() an inline helper, so provide
idle_rq() and implement idle_cpu() and available_idle_cpu() using
that.
Signed-off-by: Peter Zijlstra (Intel) <peterz@infradead.org>
The introduction of PREEMPT_LAZY was for multiple reasons:
- PREEMPT_RT suffered from over-scheduling, hurting performance compared to
!PREEMPT_RT.
- the introduction of (more) features that rely on preemption; like
folio_zero_user() which can do large memset() without preemption checks.
(Xen already had a horrible hack to deal with long running hypercalls)
- the endless and uncontrolled sprinkling of cond_resched() -- mostly cargo
cult or in response to poor to replicate workloads.
By moving to a model that is fundamentally preemptable these things become
managable and avoid needing to introduce more horrible hacks.
Since this is a requirement; limit PREEMPT_NONE to architectures that do not
support preemption at all. Further limit PREEMPT_VOLUNTARY to those
architectures that do not yet have PREEMPT_LAZY support (with the eventual goal
to make this the empty set and completely remove voluntary preemption and
cond_resched() -- notably VOLUNTARY is already limited to !ARCH_NO_PREEMPT.)
This leaves up-to-date architectures (arm64, loongarch, powerpc, riscv, s390,
x86) with only two preemption models: full and lazy.
While Lazy has been the recommended setting for a while, not all distributions
have managed to make the switch yet. Force things along. Keep the patch minimal
in case of hard to address regressions that might pop up.
Signed-off-by: Peter Zijlstra (Intel) <peterz@infradead.org>
Reviewed-by: Valentin Schneider <vschneid@redhat.com>
Link: https://patch.msgid.link/20251219101502.GB1132199@noisy.programming.kicks-ass.net
With the introduction of compiler context analysis (LLVM
ThreadSafetyAnalysis) the struct definition of various locks get
wrapped in a macro. This hides them from tags based navigation,
although clangd/LSP sees right through it and works as expected.
Add a regex to the tags script to help it along.
Requested-by: Bart Van Assche <bvanassche@acm.org>
Signed-off-by: Peter Zijlstra (Intel) <peterz@infradead.org>
Link: https://patch.msgid.link/20251220133307.GR3707891@noisy.programming.kicks-ass.net
With the change to hrtimer_try_to_cancel() in
perf_swevent_cancel_hrtimer() it appears possible for the hrtimer to
still be active by the time the event gets freed.
Make sure the event does a full hrtimer_cancel() on the free path by
installing a perf_event::destroy handler.
Fixes: eb3182ef04 ("perf/core: Fix system hang caused by cpu-clock usage")
Reported-by: CyberUnicorns <a101e_iotvul@163.com>
Tested-by: CyberUnicorns <a101e_iotvul@163.com>
Debugged-by: Thomas Gleixner <tglx@linutronix.de>
Signed-off-by: Peter Zijlstra (Intel) <peterz@infradead.org>
For some mysterious reasons the GCC 8 and 9 preprocessor manages to
sporadically fumble _ASM_BYTES(0x0f, 0x0b):
$ grep ".byte[ ]*0x0f" defconfig-build/drivers/net/wireless/realtek/rtlwifi/base.s
1: .byte0x0f,0x0b ;
1: .byte 0x0f,0x0b ;
which makes the assembler upset and all that. While there are more
_ASM_BYTES() users (notably the NOP instructions), those don't seem
affected. Therefore replace the offending ASM_UD2 with one using the
ud2 mnemonic.
Reported-by: Jean Delvare <jdelvare@suse.de>
Suggested-by: Uros Bizjak <ubizjak@gmail.com>
Fixes: 85a2d4a890 ("x86,ibt: Use UDB instead of 0xEA")
Cc: stable@kernel.org
Signed-off-by: Peter Zijlstra (Intel) <peterz@infradead.org>
Link: https://patch.msgid.link/20251218104659.GT3911114@noisy.programming.kicks-ass.net
The mediated_pmu_account_event() and perf_create_mediated_pmu()
functions implement the exclusion between '!exclude_guest' counters
and mediated vPMUs. Their implementation is basically identical,
except mirrored in what they count/check.
Make sure the actual implementations reflect this similarity.
Notably:
- while perf_release_mediated_pmu() has an underflow check;
mediated_pmu_unaccount_event() did not.
- while perf_create_mediated_pmu() has an inc_not_zero() path;
mediated_pmu_account_event() did not.
Also, the inc_not_zero() path can be outsite of
perf_mediated_pmu_mutex. The mutex must guard the 0->1 (of either
nr_include_guest_events or nr_mediated_pmu_vms) transition, but once a
counter is already non-zero, it can safely be incremented further.
Signed-off-by: Peter Zijlstra (Intel) <peterz@infradead.org>
Link: https://patch.msgid.link/20251208115156.GE3707891@noisy.programming.kicks-ass.net
Commit 47efe2ddcc ("sched/core: Add assertions to QUEUE_CLASS") added an
assert to sched_change_end() verifying that a class demotion would result in a
reschedule.
As it turns out; rt_mutex_setprio() does not force a resched on class
demontion. Furthermore, this is only relevant to running tasks.
Change the warning into a reschedule and make sure to only do so for running
tasks.
Fixes: 47efe2ddcc ("sched/core: Add assertions to QUEUE_CLASS")
Reported-by: Naresh Kamboju <naresh.kamboju@linaro.org>
Tested-by: Linux Kernel Functional Testing <lkft@linaro.org>
Signed-off-by: Peter Zijlstra (Intel) <peterz@infradead.org>
Link: https://patch.msgid.link/20251216141725.GW3707837@noisy.programming.kicks-ass.net
Change sched_class::wakeup_preempt() to also get called for
cross-class wakeups, specifically those where the woken task
is of a higher class than the previous highest class.
In order to do this, track the current highest class of the runqueue
in rq::next_class and have wakeup_preempt() track this upwards for
each new wakeup. Additionally have schedule() re-set the value on
pick.
Signed-off-by: Peter Zijlstra (Intel) <peterz@infradead.org>
Signed-off-by: Ingo Molnar <mingo@kernel.org>
Link: https://patch.msgid.link/20251127154725.901391274@infradead.org
The task_tick_fair() function does:
- update the hierarchical runtimes
- drive NUMA-balancing
- update load-balance statistics
- drive force-idle preemption
All but the very first can be limited to the periodic tick. Let hrtick
only update accounting and drive preemption, not load-balancing and
other bits.
Signed-off-by: Peter Zijlstra (Intel) <peterz@infradead.org>
Signed-off-by: Ingo Molnar <mingo@kernel.org>
Link: https://patch.msgid.link/20250918080205.563385766@infradead.org
With the {rcu,sched,bh} RCU flavours being unified, it doesn't really
make sense to check for just the rcu one. Switch to the _all family of
verification which includes all 3 of the listed flavours.
Notably, this will enable us to remove some superfluous
rcu_read_lock() regions when we know they are inside preempt/IRQ
disabled regions.
Signed-off-by: Peter Zijlstra (Intel) <peterz@infradead.org>
Signed-off-by: Ingo Molnar <mingo@kernel.org>
Remove check from the name for being surplus to requirements.
Signed-off-by: Peter Zijlstra (Intel) <peterz@infradead.org>
Signed-off-by: Ingo Molnar <mingo@kernel.org>
Peter Zijlstra