commit c6e80201e0 upstream.
to_ratio() computes BW_SHIFT-scaled bandwidth ratios from u64 period and
runtime values, but it returns unsigned long. tg_rt_schedulable() also
stores the current group limit and the accumulated child sum in unsigned
long.
On 32-bit builds, large bandwidth ratios can be truncated and the RT
group sum can wrap when enough siblings are present. That can let an
overcommitted RT hierarchy pass the schedulability check, and it also
narrows the helper result for other callers.
Return u64 from to_ratio() and use u64 for the RT group totals so
bandwidth ratios are preserved and compared at full width on both 32-bit
and 64-bit builds.
Fixes: b40b2e8eb5 ("sched: rt: multi level group constraints")
Assisted-by: Codex:GPT-5
Signed-off-by: Joseph Salisbury <joseph.salisbury@oracle.com>
Signed-off-by: Peter Zijlstra (Intel) <peterz@infradead.org>
Cc: stable@vger.kernel.org
Link: https://patch.msgid.link/20260403210014.2713404-1-joseph.salisbury@oracle.com
Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
[ Upstream commit 14a8570564 ]
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
Signed-off-by: Sasha Levin <sashal@kernel.org>
[ Upstream commit 7e0ffb72de ]
@p->scx.ddsp_dsq_id can be left set (non-SCX_DSQ_INVALID) triggering a
spurious warning in mark_direct_dispatch() when the next wakeup's
ops.select_cpu() calls scx_bpf_dsq_insert(), such as:
WARNING: kernel/sched/ext.c:1273 at scx_dsq_insert_commit+0xcd/0x140
The root cause is that ddsp_dsq_id was only cleared in dispatch_enqueue(),
which is not reached in all paths that consume or cancel a direct dispatch
verdict.
Fix it by clearing it at the right places:
- direct_dispatch(): cache the direct dispatch state in local variables
and clear it before dispatch_enqueue() on the synchronous path. For
the deferred path, the direct dispatch state must remain set until
process_ddsp_deferred_locals() consumes them.
- process_ddsp_deferred_locals(): cache the dispatch state in local
variables and clear it before calling dispatch_to_local_dsq(), which
may migrate the task to another rq.
- do_enqueue_task(): clear the dispatch state on the enqueue path
(local/global/bypass fallbacks), where the direct dispatch verdict is
ignored.
- dequeue_task_scx(): clear the dispatch state after dispatch_dequeue()
to handle both the deferred dispatch cancellation and the holding_cpu
race, covering all cases where a pending direct dispatch is
cancelled.
- scx_disable_task(): clear the direct dispatch state when
transitioning a task out of the current scheduler. Waking tasks may
have had the direct dispatch state set by the outgoing scheduler's
ops.select_cpu() and then been queued on a wake_list via
ttwu_queue_wakelist(), when SCX_OPS_ALLOW_QUEUED_WAKEUP is set. Such
tasks are not on the runqueue and are not iterated by scx_bypass(),
so their direct dispatch state won't be cleared. Without this clear,
any subsequent SCX scheduler that tries to direct dispatch the task
will trigger the WARN_ON_ONCE() in mark_direct_dispatch().
Fixes: 5b26f7b920 ("sched_ext: Allow SCX_DSQ_LOCAL_ON for direct dispatches")
Cc: stable@vger.kernel.org # v6.12+
Cc: Daniel Hodges <hodgesd@meta.com>
Cc: Patrick Somaru <patsomaru@meta.com>
Signed-off-by: Andrea Righi <arighi@nvidia.com>
Signed-off-by: Tejun Heo <tj@kernel.org>
[ adapted function signatures and code paths ]
Signed-off-by: Sasha Levin <sashal@kernel.org>
Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
[ Upstream commit 1319ea5752 ]
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
Signed-off-by: Sasha Levin <sashal@kernel.org>
[ Upstream commit 7a8464555d ]
bpf_iter_scx_dsq_new() reads dsq->seq via READ_ONCE() without holding
any lock, making dsq->seq a lock-free concurrently accessed variable.
However, dispatch_enqueue(), the sole writer of dsq->seq, uses a plain
increment without the matching WRITE_ONCE() on the write side:
dsq->seq++;
^^^^^^^^^^^
plain write -- KCSAN data race
The KCSAN documentation requires that if one accessor uses READ_ONCE()
or WRITE_ONCE() on a variable to annotate lock-free access, all other
accesses must also use the appropriate accessor. A plain write leaves
the pair incomplete and will trigger KCSAN warnings.
Fix by using WRITE_ONCE() for the write side of the update:
WRITE_ONCE(dsq->seq, dsq->seq + 1);
This is consistent with bpf_iter_scx_dsq_new() and makes the
concurrent access annotation complete and KCSAN-clean.
Signed-off-by: zhidao su <suzhidao@xiaomi.com>
Signed-off-by: Tejun Heo <tj@kernel.org>
Signed-off-by: Sasha Levin <sashal@kernel.org>
[ Upstream commit f4c31b07b1 ]
There are two special cases in the idle loop that are handled
inconsistently even though they are analogous.
The first one is when a cpuidle driver is absent and the default CPU
idle time power management implemented by the architecture code is used.
In that case, the scheduler tick is stopped every time before invoking
default_idle_call().
The second one is when a cpuidle driver is present, but there is only
one idle state in its table. In that case, the scheduler tick is never
stopped at all.
Since each of these approaches has its drawbacks, reconcile them with
the help of one simple heuristic. Namely, stop the tick if the CPU has
been woken up by it in the previous iteration of the idle loop, or let
it tick otherwise.
Signed-off-by: Rafael J. Wysocki <rafael.j.wysocki@intel.com>
Reviewed-by: Christian Loehle <christian.loehle@arm.com>
Reviewed-by: Frederic Weisbecker <frederic@kernel.org>
Reviewed-by: Qais Yousef <qyousef@layalina.io>
Reviewed-by: Aboorva Devarajan <aboorvad@linux.ibm.com>
Fixes: ed98c34919 ("sched: idle: Do not stop the tick before cpuidle_idle_call()")
[ rjw: Added Fixes tag, changelog edits ]
Link: https://patch.msgid.link/4741364.LvFx2qVVIh@rafael.j.wysocki
Signed-off-by: Rafael J. Wysocki <rafael.j.wysocki@intel.com>
Signed-off-by: Sasha Levin <sashal@kernel.org>
commit b3d99f43c7 upstream.
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
Tested-by: Eric Hagberg <ehagberg@janestreet.com>
Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
commit 1336b579f6 upstream.
The iterator css_for_each_descendant_pre() walks the cgroup hierarchy
under cgroup_lock(). It does not increment the reference counts on
yielded css structs.
According to the cgroup documentation, css_put() should only be used
to release a reference obtained via css_get() or css_tryget_online().
Since the iterator does not use either of these to acquire a reference,
calling css_put() in the error path of scx_cgroup_init() causes a
refcount underflow.
Remove the unbalanced css_put() to prevent a potential Use-After-Free
(UAF) vulnerability.
Fixes: 8195136669 ("sched_ext: Add cgroup support")
Cc: stable@vger.kernel.org # v6.12+
Signed-off-by: Cheng-Yang Chou <yphbchou0911@gmail.com>
Reviewed-by: Andrea Righi <arighi@nvidia.com>
Signed-off-by: Tejun Heo <tj@kernel.org>
Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
[ Upstream commit b06ccbabe2 ]
During scx_enable(), the READY -> ENABLED task switching loop changes the
calling thread's sched_class from fair to ext. Since fair has higher
priority than ext, saturating fair-class workloads can indefinitely starve
the enable thread, hanging the system. This was introduced when the enable
path switched from preempt_disable() to scx_bypass() which doesn't protect
against fair-class starvation. Note that the original preempt_disable()
protection wasn't complete either - in partial switch modes, the calling
thread could still be starved after preempt_enable() as it may have been
switched to ext class.
Fix it by offloading the enable body to a dedicated system-wide RT
(SCHED_FIFO) kthread which cannot be starved by either fair or ext class
tasks. scx_enable() lazily creates the kthread on first use and passes the
ops pointer through a struct scx_enable_cmd containing the kthread_work,
then synchronously waits for completion.
The workfn runs on a different kthread from sch->helper (which runs
disable_work), so it can safely flush disable_work on the error path
without deadlock.
Fixes: 8c2090c504 ("sched_ext: Initialize in bypass mode")
Cc: stable@vger.kernel.org # v6.12+
Signed-off-by: Tejun Heo <tj@kernel.org>
[ adapted per-scheduler scx_sched struct references to globals ]
Signed-off-by: Sasha Levin <sashal@kernel.org>
Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
[ Upstream commit 83236b2e43 ]
scx_claim_exit() atomically sets exit_kind, which prevents scx_error() from
triggering further error handling. After claiming exit, the caller must kick
the helper kthread work which initiates bypass mode and teardown.
If the calling task gets preempted between claiming exit and kicking the
helper work, and the BPF scheduler fails to schedule it back (since error
handling is now disabled), the helper work is never queued, bypass mode
never activates, tasks stop being dispatched, and the system wedges.
Disable preemption across scx_claim_exit() and the subsequent work kicking
in all callers - scx_disable() and scx_vexit(). Add
lockdep_assert_preemption_disabled() to scx_claim_exit() to enforce the
requirement.
Fixes: f0e1a0643a ("sched_ext: Implement BPF extensible scheduler class")
Cc: stable@vger.kernel.org # v6.12+
Reviewed-by: Andrea Righi <arighi@nvidia.com>
Signed-off-by: Tejun Heo <tj@kernel.org>
[ adapted per-scheduler struct (sch->exit_kind, scx_disable, scx_vexit) to global variables (scx_exit_kind, scx_ops_disable, scx_ops_exit_kind) ]
Signed-off-by: Sasha Levin <sashal@kernel.org>
Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
[ Upstream commit 6080fb2116 ]
Updating "ppos" on error conditions does not make much sense. The pattern
is to return the error code directly without modifying the position, or
modify the position on success and return the number of bytes written.
Since on success, the return value of apply is 0, there is no point in
modifying ppos either. Fix it by removing all this and just returning
error code or number of bytes written on success.
Signed-off-by: Joel Fernandes <joelagnelf@nvidia.com>
Signed-off-by: Peter Zijlstra (Intel) <peterz@infradead.org>
Reviewed-by: Juri Lelli <juri.lelli@redhat.com>
Reviewed-by: Andrea Righi <arighi@nvidia.com>
Acked-by: Tejun Heo <tj@kernel.org>
Tested-by: Christian Loehle <christian.loehle@arm.com>
Link: https://patch.msgid.link/20260126100050.3854740-3-arighi@nvidia.com
Signed-off-by: Sasha Levin <sashal@kernel.org>
[ Upstream commit 94894c9c47 ]
CPU0 becomes overloaded when hosting a CPU-bound RT task, a non-CPU-bound
RT task, and a CFS task stuck in kernel space. When other CPUs switch from
RT to non-RT tasks, RT load balancing (LB) is triggered; with
HAVE_RT_PUSH_IPI enabled, they send IPIs to CPU0 to drive the execution
of rto_push_irq_work_func. During push_rt_task on CPU0,
if next_task->prio < rq->donor->prio, resched_curr() sets NEED_RESCHED
and after the push operation completes, CPU0 calls rto_next_cpu().
Since only CPU0 is overloaded in this scenario, rto_next_cpu() should
ideally return -1 (no further IPI needed).
However, multiple CPUs invoking tell_cpu_to_push() during LB increments
rd->rto_loop_next. Even when rd->rto_cpu is set to -1, the mismatch between
rd->rto_loop and rd->rto_loop_next forces rto_next_cpu() to restart its
search from -1. With CPU0 remaining overloaded (satisfying rt_nr_migratory
&& rt_nr_total > 1), it gets reselected, causing CPU0 to queue irq_work to
itself and send self-IPIs repeatedly. As long as CPU0 stays overloaded and
other CPUs run pull_rt_tasks(), it falls into an infinite self-IPI loop,
which triggers a CPU hardlockup due to continuous self-interrupts.
The trigging scenario is as follows:
cpu0 cpu1 cpu2
pull_rt_task
tell_cpu_to_push
<------------irq_work_queue_on
rto_push_irq_work_func
push_rt_task
resched_curr(rq) pull_rt_task
rto_next_cpu tell_cpu_to_push
<-------------------------- atomic_inc(rto_loop_next)
rd->rto_loop != next
rto_next_cpu
irq_work_queue_on
rto_push_irq_work_func
Fix redundant self-IPI by filtering the initiating CPU in rto_next_cpu().
This solution has been verified to effectively eliminate spurious self-IPIs
and prevent CPU hardlockup scenarios.
Fixes: 4bdced5c9a ("sched/rt: Simplify the IPI based RT balancing logic")
Suggested-by: Steven Rostedt (Google) <rostedt@goodmis.org>
Suggested-by: K Prateek Nayak <kprateek.nayak@amd.com>
Signed-off-by: Chen Jinghuang <chenjinghuang2@huawei.com>
Signed-off-by: Peter Zijlstra (Intel) <peterz@infradead.org>
Reviewed-by: Steven Rostedt (Google) <rostedt@goodmis.org>
Reviewed-by: Valentin Schneider <vschneid@redhat.com>
Link: https://patch.msgid.link/20260122012533.673768-1-chenjinghuang2@huawei.com
Signed-off-by: Sasha Levin <sashal@kernel.org>
[ Upstream commit 2614069c59 ]
Place the notes that resulted from going through the dl_server code in a
comment.
Signed-off-by: Peter Zijlstra (Intel) <peterz@infradead.org>
Stable-dep-of: 1151354225 ("sched/deadline: Fix 'stuck' dl_server")
Signed-off-by: Sasha Levin <sashal@kernel.org>
[ Upstream commit a257e97421 ]
For PREEMPT_RT=y kernels, the deferred_irq_workfn() is executed in
the per-cpu irq_work/* task context and not disable-irq, if the rq
returned by container_of() is current CPU's rq, the following scenarios
may occur:
lock(&rq->__lock);
<Interrupt>
lock(&rq->__lock);
This commit use IRQ_WORK_INIT_HARD() to replace init_irq_work() to
initialize rq->scx.deferred_irq_work, make the deferred_irq_workfn()
is always invoked in hard-irq context.
Signed-off-by: Zqiang <qiang.zhang@linux.dev>
Signed-off-by: Tejun Heo <tj@kernel.org>
Signed-off-by: Chen Yu <xnguchen@sina.cn>
Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
[ Upstream commit f5e1e5ec20 ]
move_local_task_to_local_dsq() is used when moving a task from a non-local
DSQ to a local DSQ on the same CPU. It directly manipulates the local DSQ
without going through dispatch_enqueue() and was missing the post-enqueue
handling that triggers preemption when SCX_ENQ_PREEMPT is set or the idle
task is running.
The function is used by move_task_between_dsqs() which backs
scx_bpf_dsq_move() and may be called while the CPU is busy.
Add local_dsq_post_enq() call to move_local_task_to_local_dsq(). As the
dispatch path doesn't need post-enqueue handling, add SCX_RQ_IN_BALANCE
early exit to keep consume_dispatch_q() behavior unchanged and avoid
triggering unnecessary resched when scx_bpf_dsq_move() is used from the
dispatch path.
Fixes: 4c30f5ce4f ("sched_ext: Implement scx_bpf_dispatch[_vtime]_from_dsq()")
Cc: stable@vger.kernel.org # v6.12+
Reviewed-by: Andrea Righi <arighi@nvidia.com>
Reviewed-by: Emil Tsalapatis <emil@etsalapatis.com>
Signed-off-by: Tejun Heo <tj@kernel.org>
Signed-off-by: Sasha Levin <sashal@kernel.org>
Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
[ Upstream commit 530b6637c7 ]
Factor out local_dsq_post_enq() which performs post-enqueue handling for
local DSQs - triggering resched_curr() if SCX_ENQ_PREEMPT is specified or if
the current CPU is idle. No functional change.
This will be used by the next patch to fix move_local_task_to_local_dsq().
Cc: stable@vger.kernel.org # v6.12+
Reviewed-by: Andrea Righi <arighi@nvidia.com>
Reviewed-by: Emil Tsalapatis <emil@etsalapatis.com>
Signed-off-by: Tejun Heo <tj@kernel.org>
Signed-off-by: Sasha Levin <sashal@kernel.org>
Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
[ Upstream commit 1dd6c84f1c ]
When loading the ebpf scheduler, the tasks in the scx_tasks list will
be traversed and invoke __setscheduler_class() to get new sched_class.
however, this would also incorrectly set the per-cpu migration
task's->sched_class to rt_sched_class, even after unload, the per-cpu
migration task's->sched_class remains sched_rt_class.
The log for this issue is as follows:
./scx_rustland --stats 1
[ 199.245639][ T630] sched_ext: "rustland" does not implement cgroup cpu.weight
[ 199.269213][ T630] sched_ext: BPF scheduler "rustland" enabled
04:25:09 [INFO] RustLand scheduler attached
bpftrace -e 'iter:task /strcontains(ctx->task->comm, "migration")/
{ printf("%s:%d->%pS\n", ctx->task->comm, ctx->task->pid, ctx->task->sched_class); }'
Attaching 1 probe...
migration/0:24->rt_sched_class+0x0/0xe0
migration/1:27->rt_sched_class+0x0/0xe0
migration/2:33->rt_sched_class+0x0/0xe0
migration/3:39->rt_sched_class+0x0/0xe0
migration/4:45->rt_sched_class+0x0/0xe0
migration/5:52->rt_sched_class+0x0/0xe0
migration/6:58->rt_sched_class+0x0/0xe0
migration/7:64->rt_sched_class+0x0/0xe0
sched_ext: BPF scheduler "rustland" disabled (unregistered from user space)
EXIT: unregistered from user space
04:25:21 [INFO] Unregister RustLand scheduler
bpftrace -e 'iter:task /strcontains(ctx->task->comm, "migration")/
{ printf("%s:%d->%pS\n", ctx->task->comm, ctx->task->pid, ctx->task->sched_class); }'
Attaching 1 probe...
migration/0:24->rt_sched_class+0x0/0xe0
migration/1:27->rt_sched_class+0x0/0xe0
migration/2:33->rt_sched_class+0x0/0xe0
migration/3:39->rt_sched_class+0x0/0xe0
migration/4:45->rt_sched_class+0x0/0xe0
migration/5:52->rt_sched_class+0x0/0xe0
migration/6:58->rt_sched_class+0x0/0xe0
migration/7:64->rt_sched_class+0x0/0xe0
This commit therefore generate a new scx_setscheduler_class() and
add check for stop_sched_class to replace __setscheduler_class().
Fixes: f0e1a0643a ("sched_ext: Implement BPF extensible scheduler class")
Cc: stable@vger.kernel.org # v6.12+
Signed-off-by: Zqiang <qiang.zhang@linux.dev>
Reviewed-by: Andrea Righi <arighi@nvidia.com>
Signed-off-by: Tejun Heo <tj@kernel.org>
[ Adjust context ]
Signed-off-by: Sasha Levin <sashal@kernel.org>
Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
[ Upstream commit 79f3f9bedd ]
Basically, from the constraint that the sum of lag is zero, you can
infer that the 0-lag point is the weighted average of the individual
vruntime, which is what we're trying to compute:
\Sum w_i * v_i
avg = --------------
\Sum w_i
Now, since vruntime takes the whole u64 (worse, it wraps), this
multiplication term in the numerator is not something we can compute;
instead we do the min_vruntime (v0 henceforth) thing like:
v_i = (v_i - v0) + v0
This does two things:
- it keeps the key: (v_i - v0) 'small';
- it creates a relative 0-point in the modular space.
If you do that subtitution and work it all out, you end up with:
\Sum w_i * (v_i - v0)
avg = --------------------- + v0
\Sum w_i
Since you cannot very well track a ratio like that (and not suffer
terrible numerical problems) we simpy track the numerator and
denominator individually and only perform the division when strictly
needed.
Notably, the numerator lives in cfs_rq->avg_vruntime and the denominator
lives in cfs_rq->avg_load.
The one extra 'funny' is that these numbers track the entities in the
tree, and current is typically outside of the tree, so avg_vruntime()
adds current when needed before doing the division.
(vruntime_eligible() elides the division by cross-wise multiplication)
Anyway, as mentioned above, we currently use the CFS era min_vruntime
for this purpose. However, this thing can only move forward, while the
above avg can in fact move backward (when a non-eligible task leaves,
the average becomes smaller), this can cause trouble when through
happenstance (or construction) these values drift far enough apart to
wreck the game.
Replace cfs_rq::min_vruntime with cfs_rq::zero_vruntime which is kept
near/at avg_vruntime, following its motion.
The down-side is that this requires computing the avg more often.
Fixes: 147f3efaa2 ("sched/fair: Implement an EEVDF-like scheduling policy")
Reported-by: Zicheng Qu <quzicheng@huawei.com>
Signed-off-by: Peter Zijlstra (Intel) <peterz@infradead.org>
Link: https://patch.msgid.link/20251106111741.GC4068168@noisy.programming.kicks-ass.net
Cc: stable@vger.kernel.org
[ Adjust context in comments + init_cfs_rq ]
Signed-off-by: Sasha Levin <sashal@kernel.org>
Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
commit 690e47d140 upstream.
Overview
========
When a CPU chooses to call push_rt_task and picks a task to push to
another CPU's runqueue then it will call find_lock_lowest_rq method
which would take a double lock on both CPUs' runqueues. If one of the
locks aren't readily available, it may lead to dropping the current
runqueue lock and reacquiring both the locks at once. During this window
it is possible that the task is already migrated and is running on some
other CPU. These cases are already handled. However, if the task is
migrated and has already been executed and another CPU is now trying to
wake it up (ttwu) such that it is queued again on the runqeue
(on_rq is 1) and also if the task was run by the same CPU, then the
current checks will pass even though the task was migrated out and is no
longer in the pushable tasks list.
Crashes
=======
This bug resulted in quite a few flavors of crashes triggering kernel
panics with various crash signatures such as assert failures, page
faults, null pointer dereferences, and queue corruption errors all
coming from scheduler itself.
Some of the crashes:
-> kernel BUG at kernel/sched/rt.c:1616! BUG_ON(idx >= MAX_RT_PRIO)
Call Trace:
? __die_body+0x1a/0x60
? die+0x2a/0x50
? do_trap+0x85/0x100
? pick_next_task_rt+0x6e/0x1d0
? do_error_trap+0x64/0xa0
? pick_next_task_rt+0x6e/0x1d0
? exc_invalid_op+0x4c/0x60
? pick_next_task_rt+0x6e/0x1d0
? asm_exc_invalid_op+0x12/0x20
? pick_next_task_rt+0x6e/0x1d0
__schedule+0x5cb/0x790
? update_ts_time_stats+0x55/0x70
schedule_idle+0x1e/0x40
do_idle+0x15e/0x200
cpu_startup_entry+0x19/0x20
start_secondary+0x117/0x160
secondary_startup_64_no_verify+0xb0/0xbb
-> BUG: kernel NULL pointer dereference, address: 00000000000000c0
Call Trace:
? __die_body+0x1a/0x60
? no_context+0x183/0x350
? __warn+0x8a/0xe0
? exc_page_fault+0x3d6/0x520
? asm_exc_page_fault+0x1e/0x30
? pick_next_task_rt+0xb5/0x1d0
? pick_next_task_rt+0x8c/0x1d0
__schedule+0x583/0x7e0
? update_ts_time_stats+0x55/0x70
schedule_idle+0x1e/0x40
do_idle+0x15e/0x200
cpu_startup_entry+0x19/0x20
start_secondary+0x117/0x160
secondary_startup_64_no_verify+0xb0/0xbb
-> BUG: unable to handle page fault for address: ffff9464daea5900
kernel BUG at kernel/sched/rt.c:1861! BUG_ON(rq->cpu != task_cpu(p))
-> kernel BUG at kernel/sched/rt.c:1055! BUG_ON(!rq->nr_running)
Call Trace:
? __die_body+0x1a/0x60
? die+0x2a/0x50
? do_trap+0x85/0x100
? dequeue_top_rt_rq+0xa2/0xb0
? do_error_trap+0x64/0xa0
? dequeue_top_rt_rq+0xa2/0xb0
? exc_invalid_op+0x4c/0x60
? dequeue_top_rt_rq+0xa2/0xb0
? asm_exc_invalid_op+0x12/0x20
? dequeue_top_rt_rq+0xa2/0xb0
dequeue_rt_entity+0x1f/0x70
dequeue_task_rt+0x2d/0x70
__schedule+0x1a8/0x7e0
? blk_finish_plug+0x25/0x40
schedule+0x3c/0xb0
futex_wait_queue_me+0xb6/0x120
futex_wait+0xd9/0x240
do_futex+0x344/0xa90
? get_mm_exe_file+0x30/0x60
? audit_exe_compare+0x58/0x70
? audit_filter_rules.constprop.26+0x65e/0x1220
__x64_sys_futex+0x148/0x1f0
do_syscall_64+0x30/0x80
entry_SYSCALL_64_after_hwframe+0x62/0xc7
-> BUG: unable to handle page fault for address: ffff8cf3608bc2c0
Call Trace:
? __die_body+0x1a/0x60
? no_context+0x183/0x350
? spurious_kernel_fault+0x171/0x1c0
? exc_page_fault+0x3b6/0x520
? plist_check_list+0x15/0x40
? plist_check_list+0x2e/0x40
? asm_exc_page_fault+0x1e/0x30
? _cond_resched+0x15/0x30
? futex_wait_queue_me+0xc8/0x120
? futex_wait+0xd9/0x240
? try_to_wake_up+0x1b8/0x490
? futex_wake+0x78/0x160
? do_futex+0xcd/0xa90
? plist_check_list+0x15/0x40
? plist_check_list+0x2e/0x40
? plist_del+0x6a/0xd0
? plist_check_list+0x15/0x40
? plist_check_list+0x2e/0x40
? dequeue_pushable_task+0x20/0x70
? __schedule+0x382/0x7e0
? asm_sysvec_reschedule_ipi+0xa/0x20
? schedule+0x3c/0xb0
? exit_to_user_mode_prepare+0x9e/0x150
? irqentry_exit_to_user_mode+0x5/0x30
? asm_sysvec_reschedule_ipi+0x12/0x20
Above are some of the common examples of the crashes that were observed
due to this issue.
Details
=======
Let's look at the following scenario to understand this race.
1) CPU A enters push_rt_task
a) CPU A has chosen next_task = task p.
b) CPU A calls find_lock_lowest_rq(Task p, CPU Z’s rq).
c) CPU A identifies CPU X as a destination CPU (X < Z).
d) CPU A enters double_lock_balance(CPU Z’s rq, CPU X’s rq).
e) Since X is lower than Z, CPU A unlocks CPU Z’s rq. Someone else has
locked CPU X’s rq, and thus, CPU A must wait.
2) At CPU Z
a) Previous task has completed execution and thus, CPU Z enters
schedule, locks its own rq after CPU A releases it.
b) CPU Z dequeues previous task and begins executing task p.
c) CPU Z unlocks its rq.
d) Task p yields the CPU (ex. by doing IO or waiting to acquire a
lock) which triggers the schedule function on CPU Z.
e) CPU Z enters schedule again, locks its own rq, and dequeues task p.
f) As part of dequeue, it sets p.on_rq = 0 and unlocks its rq.
3) At CPU B
a) CPU B enters try_to_wake_up with input task p.
b) Since CPU Z dequeued task p, p.on_rq = 0, and CPU B updates
B.state = WAKING.
c) CPU B via select_task_rq determines CPU Y as the target CPU.
4) The race
a) CPU A acquires CPU X’s lock and relocks CPU Z.
b) CPU A reads task p.cpu = Z and incorrectly concludes task p is
still on CPU Z.
c) CPU A failed to notice task p had been dequeued from CPU Z while
CPU A was waiting for locks in double_lock_balance. If CPU A knew
that task p had been dequeued, it would return NULL forcing
push_rt_task to give up the task p's migration.
d) CPU B updates task p.cpu = Y and calls ttwu_queue.
e) CPU B locks Ys rq. CPU B enqueues task p onto Y and sets task
p.on_rq = 1.
f) CPU B unlocks CPU Y, triggering memory synchronization.
g) CPU A reads task p.on_rq = 1, cementing its assumption that task p
has not migrated.
h) CPU A decides to migrate p to CPU X.
This leads to A dequeuing p from Y's queue and various crashes down the
line.
Solution
========
The solution here is fairly simple. After obtaining the lock (at 4a),
the check is enhanced to make sure that the task is still at the head of
the pushable tasks list. If not, then it is anyway not suitable for
being pushed out.
Testing
=======
The fix is tested on a cluster of 3 nodes, where the panics due to this
are hit every couple of days. A fix similar to this was deployed on such
cluster and was stable for more than 30 days.
Co-developed-by: Jon Kohler <jon@nutanix.com>
Signed-off-by: Jon Kohler <jon@nutanix.com>
Co-developed-by: Gauri Patwardhan <gauri.patwardhan@nutanix.com>
Signed-off-by: Gauri Patwardhan <gauri.patwardhan@nutanix.com>
Co-developed-by: Rahul Chunduru <rahul.chunduru@nutanix.com>
Signed-off-by: Rahul Chunduru <rahul.chunduru@nutanix.com>
Signed-off-by: Harshit Agarwal <harshit@nutanix.com>
Signed-off-by: Peter Zijlstra (Intel) <peterz@infradead.org>
Reviewed-by: "Steven Rostedt (Google)" <rostedt@goodmis.org>
Reviewed-by: Phil Auld <pauld@redhat.com>
Tested-by: Will Ton <william.ton@nutanix.com>
Cc: stable@vger.kernel.org
Link: https://lore.kernel.org/r/20250225180553.167995-1-harshit@nutanix.com
Signed-off-by: Rajani Kantha <681739313@139.com>
Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
[ Upstream commit 382748c05e ]
Commit 16b269436b ("sched/deadline: Modify cpudl::free_cpus
to reflect rd->online") introduced the cpudl_set/clear_freecpu
functions to allow the cpu_dl::free_cpus mask to be manipulated
by the deadline scheduler class rq_on/offline callbacks so the
mask would also reflect this state.
Commit 9659e1eeee ("sched/deadline: Remove cpu_active_mask
from cpudl_find()") removed the check of the cpu_active_mask to
save some processing on the premise that the cpudl::free_cpus
mask already reflected the runqueue online state.
Unfortunately, there are cases where it is possible for the
cpudl_clear function to set the free_cpus bit for a CPU when the
deadline runqueue is offline. When this occurs while a CPU is
connected to the default root domain the flag may retain the bad
state after the CPU has been unplugged. Later, a different CPU
that is transitioning through the default root domain may push a
deadline task to the powered down CPU when cpudl_find sees its
free_cpus bit is set. If this happens the task will not have the
opportunity to run.
One example is outlined here:
https://lore.kernel.org/lkml/20250110233010.2339521-1-opendmb@gmail.com
Another occurs when the last deadline task is migrated from a
CPU that has an offlined runqueue. The dequeue_task member of
the deadline scheduler class will eventually call cpudl_clear
and set the free_cpus bit for the CPU.
This commit modifies the cpudl_clear function to be aware of the
online state of the deadline runqueue so that the free_cpus mask
can be updated appropriately.
It is no longer necessary to manage the mask outside of the
cpudl_set/clear functions so the cpudl_set/clear_freecpu
functions are removed. In addition, since the free_cpus mask is
now only updated under the cpudl lock the code was changed to
use the non-atomic __cpumask functions.
Signed-off-by: Doug Berger <opendmb@gmail.com>
Signed-off-by: Peter Zijlstra (Intel) <peterz@infradead.org>
Signed-off-by: Sasha Levin <sashal@kernel.org>
[ Upstream commit ca125231dd ]
When a task is migrated out, there is a probability that the tg->load_avg
value will become abnormal. The reason is as follows:
1. Due to the 1ms update period limitation in update_tg_load_avg(), there
is a possibility that the reduced load_avg is not updated to tg->load_avg
when a task migrates out.
2. Even though __update_blocked_fair() traverses the leaf_cfs_rq_list and
calls update_tg_load_avg() for cfs_rqs that are not fully decayed, the key
function cfs_rq_is_decayed() does not check whether
cfs->tg_load_avg_contrib is null. Consequently, in some cases,
__update_blocked_fair() removes cfs_rqs whose avg.load_avg has not been
updated to tg->load_avg.
Add a check of cfs_rq->tg_load_avg_contrib in cfs_rq_is_decayed(),
which fixes the case (2.) mentioned above.
Fixes: 1528c661c2 ("sched/fair: Ratelimit update to tg->load_avg")
Signed-off-by: xupengbo <xupengbo@oppo.com>
Signed-off-by: Peter Zijlstra (Intel) <peterz@infradead.org>
Signed-off-by: Ingo Molnar <mingo@kernel.org>
Reviewed-by: Aaron Lu <ziqianlu@bytedance.com>
Reviewed-by: Vincent Guittot <vincent.guittot@linaro.org>
Tested-by: Aaron Lu <ziqianlu@bytedance.com>
Link: https://patch.msgid.link/20250827022208.14487-1-xupengbo@oppo.com
Signed-off-by: Sasha Levin <sashal@kernel.org>
[ Upstream commit 79104becf4 ]
If a task yields, the scheduler may decide to pick it again. The task in
turn may decide to yield immediately or shortly after, leading to a tight
loop of yields.
If there's another runnable task as this point, the deadline will be
increased by the slice at each loop. This can cause the deadline to runaway
pretty quickly, and subsequent elevated run delays later on as the task
doesn't get picked again. The reason the scheduler can pick the same task
again and again despite its deadline increasing is because it may be the
only eligible task at that point.
Fix this by making the task forfeiting its remaining vruntime and pushing
the deadline one slice ahead. This implements yield behavior more
authentically.
We limit the forfeiting to eligible tasks. This is because core scheduling
prefers running ineligible tasks rather than force idling. As such, without
the condition, we can end up on a yield loop which makes the vruntime
increase rapidly, leading to anomalous run delays later down the line.
Fixes: 147f3efaa2 ("sched/fair: Implement an EEVDF-like scheduling policy")
Signed-off-by: Fernand Sieber <sieberf@amazon.com>
Signed-off-by: Peter Zijlstra (Intel) <peterz@infradead.org>
Link: https://lore.kernel.org/r/20250401123622.584018-1-sieberf@amazon.com
Link: https://lore.kernel.org/r/20250911095113.203439-1-sieberf@amazon.com
Link: https://lore.kernel.org/r/20250916140228.452231-1-sieberf@amazon.com
Signed-off-by: Sasha Levin <sashal@kernel.org>
[ Upstream commit 54e96258a6 ]
scx_bpf_dsq_move_set_slice() and scx_bpf_dsq_move_set_vtime() take a DSQ
iterator argument which has to be valid. Mark them with KF_RCU.
Fixes: 4c30f5ce4f ("sched_ext: Implement scx_bpf_dispatch[_vtime]_from_dsq()")
Cc: stable@vger.kernel.org # v6.12+
Acked-by: Andrea Righi <arighi@nvidia.com>
Signed-off-by: Tejun Heo <tj@kernel.org>
[ scx_bpf_dsq_move_set_* => scx_bpf_dispatch_from_dsq_set_* ]
Signed-off-by: Sasha Levin <sashal@kernel.org>
Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
[ Upstream commit 53abe3e1c1 ]
Clang is not happy with set but unused variable (this is visible
with `make W=1` build:
kernel/sched/sched.h:3744:18: error: variable 'cpumask' set but not used [-Werror,-Wunused-but-set-variable]
It seems like the variable was never used along with the assignment
that does not have side effects as far as I can see. Remove those
altogether.
Fixes: 223baf9d17 ("sched: Fix performance regression introduced by mm_cid")
Signed-off-by: Andy Shevchenko <andriy.shevchenko@linux.intel.com>
Tested-by: Eric Biggers <ebiggers@kernel.org>
Reviewed-by: Breno Leitao <leitao@debian.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
Signed-off-by: Sasha Levin <sashal@kernel.org>
[ Upstream commit 17e3e88ed0 ]
The check for some lost idle pelt time should be always done when
pick_next_task_fair() fails to pick a task and not only when we call it
from the fair fast-path.
The case happens when the last running task on rq is a RT or DL task. When
the latter goes to sleep and the /Sum of util_sum of the rq is at the max
value, we don't account the lost of idle time whereas we should.
Fixes: 67692435c4 ("sched: Rework pick_next_task() slow-path")
Signed-off-by: Vincent Guittot <vincent.guittot@linaro.org>
Signed-off-by: Peter Zijlstra (Intel) <peterz@infradead.org>
Signed-off-by: Sasha Levin <sashal@kernel.org>
Dequeuing a fair task on a throttled hierarchy returns early on
encountering a throttled cfs_rq since the throttle path has already
dequeued the hierarchy above and has adjusted the h_nr_* accounting till
the root cfs_rq.
dequeue_entities() crucially misses calling __block_task() for delayed
tasks being dequeued on the throttled hierarchies, but this was mostly
harmless until commit b7ca5743a2 ("sched/core: Tweak
wait_task_inactive() to force dequeue sched_delayed tasks") since all
existing cases would re-enqueue the task if task_on_rq_queued() returned
true and the task would eventually be blocked at pick after the
hierarchy was unthrottled.
wait_task_inactive() is special as it expects the delayed task on
throttled hierarchy to reach the blocked state on dequeue but since
__block_task() is never called, task_on_rq_queued() continues to return
true. Furthermore, since the task is now off the hierarchy, the pick
never reaches it to fully block the task even after unthrottle leading
to wait_task_inactive() looping endlessly.
Remedy this by calling __block_task() if a delayed task is being
dequeued on a throttled hierarchy.
This fix is only required for stabled kernels implementing delay dequeue
(>= v6.12) before v6.18 since upstream commit e1fad12dcb ("sched/fair:
Switch to task based throttle model") indirectly fixes this by removing
the early return conditions in dequeue_entities() as part of the per-task
throttle feature.
Cc: stable@vger.kernel.org
Reported-by: Matt Fleming <matt@readmodwrite.com>
Closes: https://lore.kernel.org/all/20250925133310.1843863-1-matt@readmodwrite.com/
Fixes: b7ca5743a2 ("sched/core: Tweak wait_task_inactive() to force dequeue sched_delayed tasks")
Tested-by: Matt Fleming <mfleming@cloudflare.com>
Signed-off-by: K Prateek Nayak <kprateek.nayak@amd.com>
Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
commit 8fd5485fb4 upstream.
When a CPU chooses to call push_dl_task and picks a task to push to
another CPU's runqueue then it will call find_lock_later_rq method
which would take a double lock on both CPUs' runqueues. If one of the
locks aren't readily available, it may lead to dropping the current
runqueue lock and reacquiring both the locks at once. During this window
it is possible that the task is already migrated and is running on some
other CPU. These cases are already handled. However, if the task is
migrated and has already been executed and another CPU is now trying to
wake it up (ttwu) such that it is queued again on the runqeue
(on_rq is 1) and also if the task was run by the same CPU, then the
current checks will pass even though the task was migrated out and is no
longer in the pushable tasks list.
Please go through the original rt change for more details on the issue.
To fix this, after the lock is obtained inside the find_lock_later_rq,
it ensures that the task is still at the head of pushable tasks list.
Also removed some checks that are no longer needed with the addition of
this new check.
However, the new check of pushable tasks list only applies when
find_lock_later_rq is called by push_dl_task. For the other caller i.e.
dl_task_offline_migration, existing checks are used.
Signed-off-by: Harshit Agarwal <harshit@nutanix.com>
Signed-off-by: Peter Zijlstra (Intel) <peterz@infradead.org>
Acked-by: Juri Lelli <juri.lelli@redhat.com>
Cc: stable@vger.kernel.org
Link: https://lore.kernel.org/r/20250408045021.3283624-1-harshit@nutanix.com
Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
commit f0c6eab5e4 upstream.
A BPF scheduler may want to use the built-in idle cpumasks in ops.init()
before the scheduler is fully initialized, either directly or through a
BPF timer for example.
However, this would result in an error, since the idle state has not
been properly initialized yet.
This can be easily verified by modifying scx_simple to call
scx_bpf_get_idle_cpumask() in ops.init():
$ sudo scx_simple
DEBUG DUMP
===========================================================================
scx_simple[121] triggered exit kind 1024:
runtime error (built-in idle tracking is disabled)
...
Fix this by properly initializing the idle state before ops.init() is
called. With this change applied:
$ sudo scx_simple
local=2 global=0
local=19 global=11
local=23 global=11
...
Fixes: d73249f887 ("sched_ext: idle: Make idle static keys private")
Signed-off-by: Andrea Righi <arighi@nvidia.com>
Reviewed-by: Joel Fernandes <joelagnelf@nvidia.com>
Signed-off-by: Tejun Heo <tj@kernel.org>
[ Backport to 6.12:
- Original commit doesn't apply cleanly to 6.12 since d73249f887 is
not present.
- This backport applies the same logical fix to prevent BPF scheduler
failures while accessing idle cpumasks from ops.init(). ]
Signed-off-by: Andrea Righi <arighi@nvidia.com>
Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
commit ddf7233fca upstream.
When enabling a sched_ext scheduler, we may trigger invalid task state
transitions, resulting in warnings like the following (which can be
easily reproduced by running the hotplug selftest in a loop):
sched_ext: Invalid task state transition 0 -> 3 for fish[770]
WARNING: CPU: 18 PID: 787 at kernel/sched/ext.c:3862 scx_set_task_state+0x7c/0xc0
...
RIP: 0010:scx_set_task_state+0x7c/0xc0
...
Call Trace:
<TASK>
scx_enable_task+0x11f/0x2e0
switching_to_scx+0x24/0x110
scx_enable.isra.0+0xd14/0x13d0
bpf_struct_ops_link_create+0x136/0x1a0
__sys_bpf+0x1edd/0x2c30
__x64_sys_bpf+0x21/0x30
do_syscall_64+0xbb/0x370
entry_SYSCALL_64_after_hwframe+0x77/0x7f
This happens because we skip initialization for tasks that are already
dead (with their usage counter set to zero), but we don't exclude them
during the scheduling class transition phase.
Fix this by also skipping dead tasks during class swiching, preventing
invalid task state transitions.
Fixes: a8532fac7b ("sched_ext: TASK_DEAD tasks must be switched into SCX on ops_enable")
Cc: stable@vger.kernel.org # v6.12+
Signed-off-by: Andrea Righi <arighi@nvidia.com>
Signed-off-by: Tejun Heo <tj@kernel.org>
Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
[ Upstream commit 155213a2ae ]
schbench (https://github.com/masoncl/schbench.git) is showing a
regression from previous production kernels that bisected down to:
sched/fair: Remove sysctl_sched_migration_cost condition (c5b0a7eefc)
The schbench command line was:
schbench -L -m 4 -M auto -t 256 -n 0 -r 0 -s 0
This creates 4 message threads pinned to CPUs 0-3, and 256x4 worker
threads spread across the rest of the CPUs. Neither the worker threads
or the message threads do any work, they just wake each other up and go
back to sleep as soon as possible.
The end result is the first 4 CPUs are pegged waking up those 1024
workers, and the rest of the CPUs are constantly banging in and out of
idle. If I take a v6.9 Linus kernel and revert that one commit,
performance goes from 3.4M RPS to 5.4M RPS.
schedstat shows there are ~100x more new idle balance operations, and
profiling shows the worker threads are spending ~20% of their CPU time
on new idle balance. schedstats also shows that almost all of these new
idle balance attemps are failing to find busy groups.
The fix used here is to crank up the cost of the newidle balance whenever it
fails. Since we don't want sd->max_newidle_lb_cost to grow out of
control, this also changes update_newidle_cost() to use
sysctl_sched_migration_cost as the upper limit on max_newidle_lb_cost.
Signed-off-by: Chris Mason <clm@fb.com>
Signed-off-by: Peter Zijlstra (Intel) <peterz@infradead.org>
Acked-by: Vincent Guittot <vincent.guittot@linaro.org>
Link: https://lkml.kernel.org/r/20250626144017.1510594-2-clm@fb.com
Signed-off-by: Sasha Levin <sashal@kernel.org>
[ Upstream commit 99b773d720 ]
With the seqcount moved out of the group into a global psi_seq,
re-initializing the seqcount on group creation is causing seqcount
corruption.
Fixes: 570c8efd5e ("sched/psi: Optimize psi_group_change() cpu_clock() usage")
Reported-by: Chris Mason <clm@meta.com>
Suggested-by: Beata Michalska <beata.michalska@arm.com>
Signed-off-by: Peter Zijlstra (Intel) <peterz@infradead.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
Signed-off-by: Sasha Levin <sashal@kernel.org>
[ Upstream commit 570c8efd5e ]
Dietmar reported that commit 3840cbe24c ("sched: psi: fix bogus
pressure spikes from aggregation race") caused a regression for him on
a high context switch rate benchmark (schbench) due to the now
repeating cpu_clock() calls.
In particular the problem is that get_recent_times() will extrapolate
the current state to 'now'. But if an update uses a timestamp from
before the start of the update, it is possible to get two reads
with inconsistent results. It is effectively back-dating an update.
(note that this all hard-relies on the clock being synchronized across
CPUs -- if this is not the case, all bets are off).
Combine this problem with the fact that there are per-group-per-cpu
seqcounts, the commit in question pushed the clock read into the group
iteration, causing tree-depth cpu_clock() calls. On architectures
where cpu_clock() has appreciable overhead, this hurts.
Instead move to a per-cpu seqcount, which allows us to have a single
clock read for all group updates, increasing internal consistency and
lowering update overhead. This comes at the cost of a longer update
side (proportional to the tree depth) which can cause the read side to
retry more often.
Fixes: 3840cbe24c ("sched: psi: fix bogus pressure spikes from aggregation race")
Reported-by: Dietmar Eggemann <dietmar.eggemann@arm.com>
Signed-off-by: Peter Zijlstra (Intel) <peterz@infradead.org>
Acked-by: Johannes Weiner <hannes@cmpxchg.org>
Tested-by: Dietmar Eggemann <dietmar.eggemann@arm.com>,
Link: https://lkml.kernel.org/20250522084844.GC31726@noisy.programming.kicks-ass.net
Signed-off-by: Sasha Levin <sashal@kernel.org>
commit 36569780b0 upstream.
The commit e6fe3f422b ("sched: Make multiple runqueue task counters
32-bit") changed nr_uninterruptible to an unsigned int. But the
nr_uninterruptible values for each of the CPU runqueues can grow to
large numbers, sometimes exceeding INT_MAX. This is valid, if, over
time, a large number of tasks are migrated off of one CPU after going
into an uninterruptible state. Only the sum of all nr_interruptible
values across all CPUs yields the correct result, as explained in a
comment in kernel/sched/loadavg.c.
Change the type of nr_uninterruptible back to unsigned long to prevent
overflows, and thus the miscalculation of load average.
Fixes: e6fe3f422b ("sched: Make multiple runqueue task counters 32-bit")
Signed-off-by: Aruna Ramakrishna <aruna.ramakrishna@oracle.com>
Signed-off-by: Peter Zijlstra (Intel) <peterz@infradead.org>
Link: https://lkml.kernel.org/r/20250709173328.606794-1-aruna.ramakrishna@oracle.com
Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
[ Upstream commit fc975cfb36 ]
In our testing with 6.12 based kernel on a big.LITTLE system, we were
seeing instances of RT tasks being blocked from running on the LITTLE
cpus for multiple seconds of time, apparently by the dl_server. This
far exceeds the default configured 50ms per second runtime.
This is due to the fair dl_server runtime calculation being scaled
for frequency & capacity of the cpu.
Consider the following case under a Big.LITTLE architecture:
Assume the runtime is: 50,000,000 ns, and Frequency/capacity
scale-invariance defined as below:
Frequency scale-invariance: 100
Capacity scale-invariance: 50
First by Frequency scale-invariance,
the runtime is scaled to 50,000,000 * 100 >> 10 = 4,882,812
Then by capacity scale-invariance,
it is further scaled to 4,882,812 * 50 >> 10 = 238,418.
So it will scaled to 238,418 ns.
This smaller "accounted runtime" value is what ends up being
subtracted against the fair-server's runtime for the current period.
Thus after 50ms of real time, we've only accounted ~238us against the
fair servers runtime. This 209:1 ratio in this example means that on
the smaller cpu the fair server is allowed to continue running,
blocking RT tasks, for over 10 seconds before it exhausts its supposed
50ms of runtime. And on other hardware configurations it can be even
worse.
For the fair deadline_server, to prevent realtime tasks from being
unexpectedly delayed, we really do want to use fixed time, and not
scaled time for smaller capacity/frequency cpus. So remove the scaling
from the fair server's accounting to fix this.
Fixes: a110a81c52 ("sched/deadline: Deferrable dl server")
Suggested-by: Peter Zijlstra <peterz@infradead.org>
Suggested-by: John Stultz <jstultz@google.com>
Signed-off-by: kuyo chang <kuyo.chang@mediatek.com>
Signed-off-by: Peter Zijlstra (Intel) <peterz@infradead.org>
Acked-by: Juri Lelli <juri.lelli@redhat.com>
Acked-by: John Stultz <jstultz@google.com>
Tested-by: John Stultz <jstultz@google.com>
Link: https://lore.kernel.org/r/20250702021440.2594736-1-kuyo.chang@mediatek.com
Signed-off-by: Sasha Levin <sashal@kernel.org>
[ Upstream commit 009836b4fa ]
On Mon, Jun 02, 2025 at 03:22:13PM +0800, Kuyo Chang wrote:
> So, the potential race scenario is:
>
> CPU0 CPU1
> // doing migrate_swap(cpu0/cpu1)
> stop_two_cpus()
> ...
> // doing _cpu_down()
> sched_cpu_deactivate()
> set_cpu_active(cpu, false);
> balance_push_set(cpu, true);
> cpu_stop_queue_two_works
> __cpu_stop_queue_work(stopper1,...);
> __cpu_stop_queue_work(stopper2,..);
> stop_cpus_in_progress -> true
> preempt_enable();
> ...
> 1st balance_push
> stop_one_cpu_nowait
> cpu_stop_queue_work
> __cpu_stop_queue_work
> list_add_tail -> 1st add push_work
> wake_up_q(&wakeq); -> "wakeq is empty.
> This implies that the stopper is at wakeq@migrate_swap."
> preempt_disable
> wake_up_q(&wakeq);
> wake_up_process // wakeup migrate/0
> try_to_wake_up
> ttwu_queue
> ttwu_queue_cond ->meet below case
> if (cpu == smp_processor_id())
> return false;
> ttwu_do_activate
> //migrate/0 wakeup done
> wake_up_process // wakeup migrate/1
> try_to_wake_up
> ttwu_queue
> ttwu_queue_cond
> ttwu_queue_wakelist
> __ttwu_queue_wakelist
> __smp_call_single_queue
> preempt_enable();
>
> 2nd balance_push
> stop_one_cpu_nowait
> cpu_stop_queue_work
> __cpu_stop_queue_work
> list_add_tail -> 2nd add push_work, so the double list add is detected
> ...
> ...
> cpu1 get ipi, do sched_ttwu_pending, wakeup migrate/1
>
So this balance_push() is part of schedule(), and schedule() is supposed
to switch to stopper task, but because of this race condition, stopper
task is stuck in WAKING state and not actually visible to be picked.
Therefore CPU1 can do another schedule() and end up doing another
balance_push() even though the last one hasn't been done yet.
This is a confluence of fail, where both wake_q and ttwu_wakelist can
cause crucial wakeups to be delayed, resulting in the malfunction of
balance_push.
Since there is only a single stopper thread to be woken, the wake_q
doesn't really add anything here, and can be removed in favour of
direct wakeups of the stopper thread.
Then add a clause to ttwu_queue_cond() to ensure the stopper threads
are never queued / delayed.
Of all 3 moving parts, the last addition was the balance_push()
machinery, so pick that as the point the bug was introduced.
Fixes: 2558aacff8 ("sched/hotplug: Ensure only per-cpu kthreads run during hotplug")
Reported-by: Kuyo Chang <kuyo.chang@mediatek.com>
Signed-off-by: Peter Zijlstra (Intel) <peterz@infradead.org>
Tested-by: Kuyo Chang <kuyo.chang@mediatek.com>
Link: https://lkml.kernel.org/r/20250605100009.GO39944@noisy.programming.kicks-ass.net
Signed-off-by: Sasha Levin <sashal@kernel.org>
[ Upstream commit c50784e99f ]
Otherwise, tg->scx.weight can go out of sync while scx_cgroup is not enabled
and ops.cgroup_init() may be called with a stale weight value.
Signed-off-by: Tejun Heo <tj@kernel.org>
Fixes: 8195136669 ("sched_ext: Add cgroup support")
Cc: stable@vger.kernel.org # v6.12+
Signed-off-by: Sasha Levin <sashal@kernel.org>
[ Upstream commit aa3ee4f0b7 ]
Delayed dequeued feature keeps a sleeping task enqueued until its
lag has elapsed. As a result, it stays also visible in rq->nr_running.
So when in wake_affine_idle(), we should use the real running-tasks
in rq to check whether we should place the wake-up task to
current cpu.
On the other hand, add a helper function to return the nr-delayed.
Fixes: 152e11f6df ("sched/fair: Implement delayed dequeue")
Signed-off-by: Xuewen Yan <xuewen.yan@unisoc.com>
Reviewed-and-tested-by: Tianchen Ding <dtcccc@linux.alibaba.com>
Signed-off-by: Peter Zijlstra (Intel) <peterz@infradead.org>
Reviewed-by: Vincent Guittot <vincent.guittot@linaro.org>
Link: https://lore.kernel.org/r/20250303105241.17251-2-xuewen.yan@unisoc.com
Signed-off-by: Sasha Levin <sashal@kernel.org>
[ Upstream commit c2a295bffe ]
With delayed dequeued feature, a sleeping sched_entity remains queued in
the rq until its lag has elapsed. As a result, it stays also visible
in the statistics that are used to balance the system and in particular
the field cfs.h_nr_queued when the sched_entity is associated to a task.
Create a new h_nr_runnable that tracks only queued and runnable tasks.
Signed-off-by: Vincent Guittot <vincent.guittot@linaro.org>
Signed-off-by: Peter Zijlstra (Intel) <peterz@infradead.org>
Reviewed-by: Dietmar Eggemann <dietmar.eggemann@arm.com>
Link: https://lore.kernel.org/r/20241202174606.4074512-5-vincent.guittot@linaro.org
Stable-dep-of: aa3ee4f0b7 ("sched/fair: Fixup wake_up_sync() vs DELAYED_DEQUEUE")
Signed-off-by: Sasha Levin <sashal@kernel.org>
Joseph Salisbury
Peter Zijlstra
Andrea Righi
Vincent Guittot
zhidao su
Rafael J. Wysocki
Cheng-Yang Chou
Tejun Heo
Joel Fernandes
Chen Jinghuang
Zqiang
Harshit Agarwal
Doug Berger
xupengbo
Fernand Sieber
Andy Shevchenko
K Prateek Nayak
Christian Loehle
Chris Mason
Juri Lelli
Aruna Ramakrishna
kuyo chang
Xuewen Yan