2024-02-26 19:46:59

by Levi Yun

Subject: [PATCH] psi: Fix avg trigger being fired unexpectedly.

commit 915a087e4c473("psi: Fix trigger being fired unexpectedly at initial")
fixes unexpected event fired when group->total accumlated for PSI_POLL.
But, for PSI_AVGS, win->value should be initialized with group->total[PSI_AVGS].
Moreover, to get exact initial value for win->value, it should be set
under each trigger lock to synchronize each rtpoll/avgs_works.

Signed-off-by: Levi Yun <[email protected]>
---
kernel/sched/psi.c | 8 +++++---
1 file changed, 5 insertions(+), 3 deletions(-)

diff --git a/kernel/sched/psi.c b/kernel/sched/psi.c
index 7b4aa5809c0f..e7f66ab2ad3e 100644
--- a/kernel/sched/psi.c
+++ b/kernel/sched/psi.c
@@ -1323,9 +1323,6 @@ struct psi_trigger *psi_trigger_create(struct psi_group *group, char *buf,
t->state = state;
t->threshold = threshold_us * NSEC_PER_USEC;
t->win.size = window_us * NSEC_PER_USEC;
- window_reset(&t->win, sched_clock(),
- group->total[PSI_POLL][t->state], 0);
-
t->event = 0;
t->last_event_time = 0;
t->of = of;
@@ -1336,6 +1333,8 @@ struct psi_trigger *psi_trigger_create(struct psi_group *group, char *buf,

if (privileged) {
mutex_lock(&group->rtpoll_trigger_lock);
+ window_reset(&t->win, sched_clock(),
+ group->total[PSI_POLL][t->state], 0);

if (!rcu_access_pointer(group->rtpoll_task)) {
struct task_struct *task;
@@ -1361,6 +1360,9 @@ struct psi_trigger *psi_trigger_create(struct psi_group *group, char *buf,
} else {
mutex_lock(&group->avgs_lock);

+ window_reset(&t->win, sched_clock(),
+ group->total[PSI_AVGS][t->state], 0);
+
list_add(&t->node, &group->avg_triggers);
group->avg_nr_triggers[t->state]++;

--
2.39.2

2024-02-26 20:08:04

by Suren Baghdasaryan

Subject: Re: [PATCH] psi: Fix avg trigger being fired unexpectedly.

On Mon, Feb 26, 2024 at 11:46 AM Levi Yun <[email protected]> wrote:
>
> commit 915a087e4c473("psi: Fix trigger being fired unexpectedly at initial")
> fixes unexpected event fired when group->total accumlated for PSI_POLL.

s/accumlated/accumulated

> But, for PSI_AVGS, win->value should be initialized with group->total[PSI_AVGS].
> Moreover, to get exact initial value for win->value, it should be set
> under each trigger lock to synchronize each rtpoll/avgs_works.

That last "to synchronize each rtpoll/avgs_works." part I would
replace with "to avoid concurrent changes to group->total[]."

Other than these nits LGTM. After fixing them feel free to add:

Acked-by: Suren Baghdasaryan <[email protected]>

>
> Signed-off-by: Levi Yun <[email protected]>
> ---
> kernel/sched/psi.c | 8 +++++---
> 1 file changed, 5 insertions(+), 3 deletions(-)
>
> diff --git a/kernel/sched/psi.c b/kernel/sched/psi.c
> index 7b4aa5809c0f..e7f66ab2ad3e 100644
> --- a/kernel/sched/psi.c
> +++ b/kernel/sched/psi.c
> @@ -1323,9 +1323,6 @@ struct psi_trigger *psi_trigger_create(struct psi_group *group, char *buf,
> t->state = state;
> t->threshold = threshold_us * NSEC_PER_USEC;
> t->win.size = window_us * NSEC_PER_USEC;
> - window_reset(&t->win, sched_clock(),
> - group->total[PSI_POLL][t->state], 0);
> -
> t->event = 0;
> t->last_event_time = 0;
> t->of = of;
> @@ -1336,6 +1333,8 @@ struct psi_trigger *psi_trigger_create(struct psi_group *group, char *buf,
>
> if (privileged) {
> mutex_lock(&group->rtpoll_trigger_lock);
> + window_reset(&t->win, sched_clock(),
> + group->total[PSI_POLL][t->state], 0);
>
> if (!rcu_access_pointer(group->rtpoll_task)) {
> struct task_struct *task;
> @@ -1361,6 +1360,9 @@ struct psi_trigger *psi_trigger_create(struct psi_group *group, char *buf,
> } else {
> mutex_lock(&group->avgs_lock);
>
> + window_reset(&t->win, sched_clock(),
> + group->total[PSI_AVGS][t->state], 0);
> +
> list_add(&t->node, &group->avg_triggers);
> group->avg_nr_triggers[t->state]++;
>
> --
> 2.39.2

2024-02-26 21:08:10

by Levi Yun

Subject: Re: [PATCH] psi: Fix avg trigger being fired unexpectedly.

Hi Suren.

> s/accumlated/accumulated
>
> > But, for PSI_AVGS, win->value should be initialized with group->total[PSI_AVGS].
> > Moreover, to get exact initial value for win->value, it should be set
> > under each trigger lock to synchronize each rtpoll/avgs_works.
>
> That last "to synchronize each rtpoll/avgs_works." part I would
> replace with "to avoid concurrent changes to group->total[]."
>
> Other than these nits LGTM. After fixing them feel free to add:
>
> Acked-by: Suren Baghdasaryan <[email protected]>

Thanks. I'll fix my commit message ;)