[PATCH] hrtimer: Retrigger local timers after clock changes

[PATCH] hrtimer: Retrigger local timers after clock changes

[Avi Weiss]

clock_was_set() selects CPUs whose hrtimer deadlines move earlier
after a clock change. smp_call_function_many() excludes the calling
CPU even when it is in the mask, leaving its clock event programmed
for the old deadline. An overdue realtime timer can then expire
only when a later event triggers timer handling.

Use on_each_cpu_mask() to retrigger the local CPU when selected,
without interrupting CPUs with unaffected deadlines.

Fixes: 81d741d3460c ("hrtimer: Avoid unnecessary SMP function calls in clock_was_set()")
Signed-off-by: Avi Weiss <thnkslprpt@gmail.com>
---
 kernel/time/hrtimer.c | 2 +-
 1 file changed, 1 insertion(+), 1 deletion(-)

diff --git a/kernel/time/hrtimer.c b/kernel/time/hrtimer.c
index 17dd38a6cee7..3133d7403a74 100644
--- a/kernel/time/hrtimer.c
+++ b/kernel/time/hrtimer.c
@@ -1000,7 +1000,7 @@ void clock_was_set(unsigned int bases)
 				cpumask_set_cpu(cpu, mask);
 		}
 		scoped_guard(preempt)
-			smp_call_function_many(mask, retrigger_next_event, NULL, 1);
+			on_each_cpu_mask(mask, retrigger_next_event, NULL, 1);
 	}
 	free_cpumask_var(mask);
 

base-commit: d135ab453a7400a34930bdda1ef4bdc8acfac80e
-- 
2.43.0

Re: [PATCH] hrtimer: Retrigger local timers after clock changes

[Frederic Weisbecker]

Le Sun, Nov 01, 2026 at 04:15:53PM +0200, Avi Weiss a écrit :
> clock_was_set() selects CPUs whose hrtimer deadlines move earlier
> after a clock change. smp_call_function_many() excludes the calling
> CPU even when it is in the mask, leaving its clock event programmed
> for the old deadline. An overdue realtime timer can then expire
> only when a later event triggers timer handling.
> 
> Use on_each_cpu_mask() to retrigger the local CPU when selected,
> without interrupting CPUs with unaffected deadlines.
> 
> Fixes: 81d741d3460c ("hrtimer: Avoid unnecessary SMP function calls in clock_was_set()")
> Signed-off-by: Avi Weiss <thnkslprpt@gmail.com>
> ---
>  kernel/time/hrtimer.c | 2 +-
>  1 file changed, 1 insertion(+), 1 deletion(-)
> 
> diff --git a/kernel/time/hrtimer.c b/kernel/time/hrtimer.c
> index 17dd38a6cee7..3133d7403a74 100644
> --- a/kernel/time/hrtimer.c
> +++ b/kernel/time/hrtimer.c
> @@ -1000,7 +1000,7 @@ void clock_was_set(unsigned int bases)
>  				cpumask_set_cpu(cpu, mask);
>  		}
>  		scoped_guard(preempt)
> -			smp_call_function_many(mask, retrigger_next_event, NULL,
> 1);

Heh, this API is so error prone... If it really doesn't want to call on the
local CPU (for some undescribed reason), it should at least be renamed to
smp_call_function_many_butself().

> +			on_each_cpu_mask(mask, retrigger_next_event, NULL, 1);

Reviewed-by: Frederic Weisbecker <frederic@kernel.org>

-- 
Frederic Weisbecker
SUSE Labs