2021-11-12 05:14:43

by Zqiang

Subject: [PATCH] watchdog: ignore nohz_full cores in new cpumask

If the nohz_full is enabled, when update watchdog_mask, the
nohz_full cores should be ignored.

Signed-off-by: Zqiang <[email protected]>
---
kernel/watchdog.c | 4 +++-
1 file changed, 3 insertions(+), 1 deletion(-)

diff --git a/kernel/watchdog.c b/kernel/watchdog.c
index ad912511a0c0..3ef11a94783c 100644
--- a/kernel/watchdog.c
+++ b/kernel/watchdog.c
@@ -628,7 +628,9 @@ void lockup_detector_soft_poweroff(void)
static void proc_watchdog_update(void)
{
/* Remove impossible cpus to keep sysctl output clean. */
- cpumask_and(&watchdog_cpumask, &watchdog_cpumask, cpu_possible_mask);
+ cpumask_and(&watchdog_cpumask, &watchdog_cpumask, housekeeping_cpumask(HK_FLAG_TIMER));
+ if (cpumask_empty(&watchdog_cpumask))
+ return;
lockup_detector_reconfigure();
}

--
2.17.1

2021-11-15 15:32:05

by Petr Mladek

Subject: Re: [PATCH] watchdog: ignore nohz_full cores in new cpumask

On Fri 2021-11-12 13:14:34, Zqiang wrote:
> If the nohz_full is enabled, when update watchdog_mask, the
> nohz_full cores should be ignored.
>
> Signed-off-by: Zqiang <[email protected]>
> ---
> kernel/watchdog.c | 4 +++-
> 1 file changed, 3 insertions(+), 1 deletion(-)
>
> diff --git a/kernel/watchdog.c b/kernel/watchdog.c
> index ad912511a0c0..3ef11a94783c 100644
> --- a/kernel/watchdog.c
> +++ b/kernel/watchdog.c
> @@ -628,7 +628,9 @@ void lockup_detector_soft_poweroff(void)
> static void proc_watchdog_update(void)
> {
> /* Remove impossible cpus to keep sysctl output clean. */
> - cpumask_and(&watchdog_cpumask, &watchdog_cpumask, cpu_possible_mask);
> + cpumask_and(&watchdog_cpumask, &watchdog_cpumask,
> housekeeping_cpumask(HK_FLAG_TIMER));

I am not familiar with nozh_full code but this looks fine.


> + if (cpumask_empty(&watchdog_cpumask))
> + return;

But this looks looks wrong. Is there any reason for this?

We need to stop the watchdog when it was running before.

I mean that lockup_detector_reconfigure() must be called anytime
when the mask has changed even when it became empty.


> lockup_detector_reconfigure();
> }

Best Regards,
Petr

2021-11-16 06:06:16

by Zqiang

Subject: Re: [PATCH] watchdog: ignore nohz_full cores in new cpumask

On 2021/11/15 下午11:31, Petr Mladek wrote:
> On Fri 2021-11-12 13:14:34, Zqiang wrote:
>> If the nohz_full is enabled, when update watchdog_mask, the
>> nohz_full cores should be ignored.
>>
>> Signed-off-by: Zqiang <[email protected]>
>> ---
>> kernel/watchdog.c | 4 +++-
>> 1 file changed, 3 insertions(+), 1 deletion(-)
>>
>> diff --git a/kernel/watchdog.c b/kernel/watchdog.c
>> index ad912511a0c0..3ef11a94783c 100644
>> --- a/kernel/watchdog.c
>> +++ b/kernel/watchdog.c
>> @@ -628,7 +628,9 @@ void lockup_detector_soft_poweroff(void)
>> static void proc_watchdog_update(void)
>> {
>> /* Remove impossible cpus to keep sysctl output clean. */
>> - cpumask_and(&watchdog_cpumask, &watchdog_cpumask, cpu_possible_mask);
>> + cpumask_and(&watchdog_cpumask, &watchdog_cpumask,
>> housekeeping_cpumask(HK_FLAG_TIMER));
> I am not familiar with nozh_full code but this looks fine.
>
>
>> + if (cpumask_empty(&watchdog_cpumask))
>> + return;
> But this looks looks wrong. Is there any reason for this?
>
> We need to stop the watchdog when it was running before.
>
> I mean that lockup_detector_reconfigure() must be called anytime
> when the mask has changed even when it became empty.

Thanks Petr

I will resend patch v2.

>
>
>> lockup_detector_reconfigure();
>> }
> Best Regards,
> Petr