In record_times() we use 'now' and groupc->state_start to calculate the
delta as bellow,
delta = now - groupc->state_start;
But note that groupc->state_start may be not initialized yet, IOW, the
state_start may be 0 currently. If state_start is 0, this calculation is
same with assigning the lower 32-bit of 'now' to delta, that is a random
value. To fix this value, we should initialize groupc->state_start before.
After we calculate the delta, we will assign 'now' to
groupc->state_start then,
groupc->state_start = now;
This will cause the same issue if groupc->state_start will not be used in a
long period. Let's take an example. We create a cgroup foo and run tasks
in it. Some of these tasks enter into memstall and state_start is set.
Then we move all of these tasks out of cgroup foo for more than (1 << 32)
nsecs, and then move them in. That will cause the same issue as above.
The root cause of these issues is that we don't initialize the state_start
properly. To fix it, we should record how many tasks in this per cpu
psi_group. If there's no task in it, we just set state_start and don't
calculate the delta, that means it is the begin of the pressure. To
avoid redundant calculating the total number of tasks in this per cpu
psi_group, a new member 'total_tasks' is introduced in struct
psi_group_cpu, which is the sum of array members in tasks[].
Fixes: eb414681d5a0 ("psi: pressure stall information for CPU, memory, and IO")
Cc: Johannes Weiner <[email protected]>
Cc: Peter Zijlstra (Intel) <[email protected]>
Cc: Daniel Drake <[email protected]>
Cc: Suren Baghdasaryan <[email protected]>
Signed-off-by: Yafang Shao <[email protected]>
---
include/linux/psi_types.h | 2 ++
kernel/sched/psi.c | 13 ++++++++++---
2 files changed, 12 insertions(+), 3 deletions(-)
diff --git a/include/linux/psi_types.h b/include/linux/psi_types.h
index 4b7258495a04..b42cbfdb15e9 100644
--- a/include/linux/psi_types.h
+++ b/include/linux/psi_types.h
@@ -69,6 +69,8 @@ struct psi_group_cpu {
/* States of the tasks belonging to this group */
unsigned int tasks[NR_PSI_TASK_COUNTS];
+ /* Sum of above array members */
+ unsigned int total_tasks;
/* Aggregate pressure state derived from the tasks */
u32 state_mask;
diff --git a/kernel/sched/psi.c b/kernel/sched/psi.c
index 8f45cdb6463b..7061529dc406 100644
--- a/kernel/sched/psi.c
+++ b/kernel/sched/psi.c
@@ -690,7 +690,10 @@ static void psi_group_change(struct psi_group *group, int cpu,
*/
write_seqcount_begin(&groupc->seq);
- record_times(groupc, cpu, false);
+ if (groupc->total_tasks)
+ record_times(groupc, cpu, false);
+ else
+ groupc->state_start = cpu_clock(cpu);
for (t = 0, m = clear; m; m &= ~(1 << t), t++) {
if (!(m & (1 << t)))
@@ -703,11 +706,15 @@ static void psi_group_change(struct psi_group *group, int cpu,
psi_bug = 1;
}
groupc->tasks[t]--;
+ groupc->total_tasks--;
}
- for (t = 0; set; set &= ~(1 << t), t++)
- if (set & (1 << t))
+ for (t = 0; set; set &= ~(1 << t), t++) {
+ if (set & (1 << t)) {
groupc->tasks[t]++;
+ groupc->total_tasks++;
+ }
+ }
/* Calculate state mask representing active states */
for (s = 0; s < NR_PSI_STATES; s++) {
--
2.18.2
On Fri, Apr 03, 2020 at 04:55:25AM -0400, Yafang Shao wrote:
> @@ -690,7 +690,10 @@ static void psi_group_change(struct psi_group *group, int cpu,
> */
> write_seqcount_begin(&groupc->seq);
>
> - record_times(groupc, cpu, false);
> + if (groupc->total_tasks)
> + record_times(groupc, cpu, false);
> + else
> + groupc->state_start = cpu_clock(cpu);
This change appears is a no-op. If there are no tasks,
groupc->state_mask is 0, and the only thing record_times() does is
groupc->state_start = cpu_clock(cpu);
Did you encounter actual problems, or are you just reading the code?
On Fri, Apr 3, 2020 at 11:20 PM Johannes Weiner <[email protected]> wrote:
>
> On Fri, Apr 03, 2020 at 04:55:25AM -0400, Yafang Shao wrote:
> > @@ -690,7 +690,10 @@ static void psi_group_change(struct psi_group *group, int cpu,
> > */
> > write_seqcount_begin(&groupc->seq);
> >
> > - record_times(groupc, cpu, false);
> > + if (groupc->total_tasks)
> > + record_times(groupc, cpu, false);
> > + else
> > + groupc->state_start = cpu_clock(cpu);
>
> This change appears is a no-op. If there are no tasks,
> groupc->state_mask is 0, and the only thing record_times() does is
>
> groupc->state_start = cpu_clock(cpu);
>
> Did you encounter actual problems, or are you just reading the code?
No real issues. Just reading the code and found it looks like a little weird.
But as you explained that state_mask can also guarantee it then we
don't need to make this change.
Thanks for the explanation.
Thanks
Yafang