Re: [PATCH] clock: Fix smp_processor_id() in preemptible bug
From: Peter Zijlstra
Date: Thu Mar 09 2017 - 10:25:38 EST
On Wed, Mar 08, 2017 at 01:53:06PM -0800, Paul E. McKenney wrote:
> The v4.11-rc1 kernel emits the following splat in some configurations:
>
> [ 43.681891] BUG: using smp_processor_id() in preemptible [00000000] code: kworker/3:1/49
> [ 43.682511] caller is debug_smp_processor_id+0x17/0x20
> [ 43.682893] CPU: 0 PID: 49 Comm: kworker/3:1 Not tainted 4.11.0-rc1+ #1
> [ 43.683382] Hardware name: QEMU Standard PC (i440FX + PIIX, 1996), BIOS Bochs 01/01/2011
> [ 43.683497] Workqueue: events __clear_sched_clock_stable
> [ 43.683497] Call Trace:
> [ 43.683497] dump_stack+0x4f/0x69
> [ 43.683497] check_preemption_disabled+0xd9/0xf0
> [ 43.683497] debug_smp_processor_id+0x17/0x20
> [ 43.683497] __clear_sched_clock_stable+0x11/0x60
> [ 43.683497] process_one_work+0x146/0x430
> [ 43.683497] worker_thread+0x126/0x490
> [ 43.683497] kthread+0xfc/0x130
> [ 43.683497] ? process_one_work+0x430/0x430
> [ 43.683497] ? kthread_create_on_node+0x40/0x40
> [ 43.683497] ? umh_complete+0x30/0x30
> [ 43.683497] ? call_usermodehelper_exec_async+0x12a/0x130
> [ 43.683497] ret_from_fork+0x29/0x40
> [ 43.689244] sched_clock: Marking unstable (43688244724, 179505618)<-(43867750342, 0)
>
> This happens because workqueue handlers run with preemption enabled
> by default and the new this_scd() function accesses per-CPU variables.
> This commit therefore disables preemption across this call to this_scd()
> and to the uses of the pointer that it returns. Lightly tested
> successfully on x86.
Does this also work?
---
kernel/sched/clock.c | 2 +-
1 file changed, 1 insertion(+), 1 deletion(-)
diff --git a/kernel/sched/clock.c b/kernel/sched/clock.c
index a08795e21628..c63042253b65 100644
--- a/kernel/sched/clock.c
+++ b/kernel/sched/clock.c
@@ -173,7 +173,7 @@ void clear_sched_clock_stable(void)
smp_mb(); /* matches sched_clock_init_late() */
if (sched_clock_running == 2)
- schedule_work(&sched_clock_work);
+ schedule_work_on(smp_processor_id(), &sched_clock_work);
}
void sched_clock_init_late(void)