Linux kernel mirror (for testing)
git.kernel.org/pub/scm/linux/kernel/git/torvalds/linux.git
kernel
os
linux
sched_ext: Defer scx_hardlockup() out of NMI
scx_hardlockup() runs from NMI and eventually calls scx_claim_exit(),
which takes scx_sched_lock. scx_sched_lock isn't NMI-safe and grabbing
it from NMI context can lead to deadlocks.
The hardlockup handler is best-effort recovery and the disable path it
triggers runs off of irq_work anyway. Move the handle_lockup() call into
an irq_work so it runs in IRQ context.
Fixes: ebeca1f930ea ("sched_ext: Introduce cgroup sub-sched support")
Signed-off-by: Tejun Heo <tj@kernel.org>
Reviewed-by: Andrea Righi <arighi@nvidia.com>
+27
-6
+27
-6
kernel/sched/ext.c
+27
-6
kernel/sched/ext.c
···
4940
4940
smp_processor_id(), dur_s);
4941
4941
}
4942
4942
4943
+
/*
4944
+
* scx_hardlockup() runs from NMI and eventually calls scx_claim_exit(),
4945
+
* which takes scx_sched_lock. scx_sched_lock isn't NMI-safe and grabbing
4946
+
* it from NMI context can lead to deadlocks. Defer via irq_work; the
4947
+
* disable path runs off irq_work anyway.
4948
+
*/
4949
+
static atomic_t scx_hardlockup_cpu = ATOMIC_INIT(-1);
4950
+
4951
+
static void scx_hardlockup_irq_workfn(struct irq_work *work)
4952
+
{
4953
+
int cpu = atomic_xchg(&scx_hardlockup_cpu, -1);
4954
+
4955
+
if (cpu >= 0 && handle_lockup("hard lockup - CPU %d", cpu))
4956
+
printk_deferred(KERN_ERR "sched_ext: Hard lockup - CPU %d, disabling BPF scheduler\n",
4957
+
cpu);
4958
+
}
4959
+
4960
+
static DEFINE_IRQ_WORK(scx_hardlockup_irq_work, scx_hardlockup_irq_workfn);
4961
+
4943
4962
/**
4944
4963
* scx_hardlockup - sched_ext hardlockup handler
4945
4964
*
···
4967
4948
* Try kicking out the current scheduler in an attempt to recover the system to
4968
4949
* a good state before taking more drastic actions.
4969
4950
*
4970
-
* Returns %true if sched_ext is enabled and abort was initiated, which may
4971
-
* resolve the reported hardlockup. %false if sched_ext is not enabled or
4972
-
* someone else already initiated abort.
4951
+
* Queues an irq_work; the handle_lockup() call happens in IRQ context (see
4952
+
* scx_hardlockup_irq_workfn).
4953
+
*
4954
+
* Returns %true if sched_ext is enabled and the work was queued, %false
4955
+
* otherwise.
4973
4956
*/
4974
4957
bool scx_hardlockup(int cpu)
4975
4958
{
4976
-
if (!handle_lockup("hard lockup - CPU %d", cpu))
4959
+
if (!rcu_access_pointer(scx_root))
4977
4960
return false;
4978
4961
4979
-
printk_deferred(KERN_ERR "sched_ext: Hard lockup - CPU %d, disabling BPF scheduler\n",
4980
-
cpu);
4962
+
atomic_cmpxchg(&scx_hardlockup_cpu, -1, cpu);
4963
+
irq_work_queue(&scx_hardlockup_irq_work);
4981
4964
return true;
4982
4965
}
4983
4966