Skip to contentSkip to conversation

sched/pelt: Use rq_clock_task() for hw_pressure

PATCH Superseded Chen Yu 2 participants 5 messages Jul 25, 2024 Scheduler

v1 of 5 Superseded Latest is v5 →

lore
Share & tools
lore

Chen Yu

author
#0 lore raw

PATCH 1/1 sched/pelt: Use rq_clock_task() for hw_pressure

commit 97450eb90965 ("sched/pelt: Remove shift of thermal clock") removed the decay_shift for hw_pressure. While looking at a related bug report, it is found that this commit uses the sched_clock_task() in sched_tick() while replaces the sched_clock_task() with rq_clock_pelt() in __update_blocked_others(). This could bring inconsistence. One possible scenario I can think of is in ___update_load_sum():

u64 delta = now - sa->last_update_time

'now' could be calculated by rq_clock_pelt() from __update_blocked_others(), and last_update_time was calculated by rq_clock_task() previously from sched_tick(). Usually the former chases after the latter, it cause a very large 'delta' and brings unexpected behavior. Although this should not impact x86 platform in the bug report, it should be fixed for other platforms.

Fixes 97450eb90965 ("sched/pelt: Remove shift of thermal clock")
---
 kernel/sched/fair.c | 2 +-
 1 file changed, 1 insertion(+), 1 deletion(-)
kernel/sched/fair.c +1 -1
index 9057584ec06d..cfd4755954fd 100644
@@ -9362,7 +9362,7 @@ static bool __update_blocked_others(struct rq *rq, bool *done)
 
 	decayed = update_rt_rq_load_avg(now, rq, curr_class == &rt_sched_class) |
 		  update_dl_rq_load_avg(now, rq, curr_class == &dl_sched_class) |
-		  update_hw_load_avg(now, rq, hw_pressure) |
+		  update_hw_load_avg(rq_clock_task(rq), rq, hw_pressure) |
 		  update_irq_load_avg(rq, 0);
 
 	if (others_have_blocked(rq))
Signature · 1 line
2.25.1
Reviewed-by on v1 1/1
#1 lore raw
Quoted 7 lines · Chen Yu

commit 97450eb90965 ("sched/pelt: Remove shift of thermal clock") removed the decay_shift for hw_pressure. While looking at a related bug report, it is found that this commit uses the sched_clock_task() in sched_tick() while replaces the sched_clock_task() with rq_clock_pelt() in __update_blocked_others(). This could bring inconsistence. One possible scenario I can think of is in ___update_load_sum():

Quoted 8 lines

u64 delta = now - sa->last_update_time

'now' could be calculated by rq_clock_pelt() from __update_blocked_others(), and last_update_time was calculated by rq_clock_task() previously from sched_tick(). Usually the former chases after the latter, it cause a very large 'delta' and brings unexpected behavior. Although this should not impact x86 platform in the bug report, it should be fixed for other platforms.

I agree with this patch but I'm a bit confused here. May I know what you mean by 'should not impact x86 platform in the bug report'? But it closes a bug report on qemu x86_64, so it does have an impact?

Quoted 14 lines
Fixes 97450eb90965 ("sched/pelt: Remove shift of thermal clock")
---
  kernel/sched/fair.c | 2 +-
  1 file changed, 1 insertion(+), 1 deletion(-)
kernel/sched/fair.c +0 -0
index 9057584ec06d..cfd4755954fd 100644

kernel/sched/fair.c static bool __update_blocked_others(struct rq *rq, bool *done) · line 9368

Quoted 8 lines
kernel/sched/fair.c +1 -1
@@ -9362,7 +9362,7 @@ static bool __update_blocked_others(struct rq *rq, bool *done)
 
  	decayed = update_rt_rq_load_avg(now, rq, curr_class == &rt_sched_class) |
  		  update_dl_rq_load_avg(now, rq, curr_class == &dl_sched_class) |
-		  update_hw_load_avg(now, rq, hw_pressure) |
+		  update_hw_load_avg(rq_clock_task(rq), rq, hw_pressure) |
  		  update_irq_load_avg(rq, 0);
 
  	if (others_have_blocked(rq))
author
#2 lore raw

Hi Hongyan,

Quoted 17 lines · Hongyan Xia
Quoted 7 lines · Chen Yu

commit 97450eb90965 ("sched/pelt: Remove shift of thermal clock") removed the decay_shift for hw_pressure. While looking at a related bug report, it is found that this commit uses the sched_clock_task() in sched_tick() while replaces the sched_clock_task() with rq_clock_pelt() in __update_blocked_others(). This could bring inconsistence. One possible scenario I can think of is in ___update_load_sum():

Quoted 8 lines

u64 delta = now - sa->last_update_time

'now' could be calculated by rq_clock_pelt() from __update_blocked_others(), and last_update_time was calculated by rq_clock_task() previously from sched_tick(). Usually the former chases after the latter, it cause a very large 'delta' and brings unexpected behavior. Although this should not impact x86 platform in the bug report, it should be fixed for other platforms.

Quoted 4 lines

I agree with this patch but I'm a bit confused here. May I know what you mean by 'should not impact x86 platform in the bug report'? But it closes a bug report on qemu x86_64, so it does have an impact?

It should not have any impact on x86_64. I added the bug link here because I checked the code while looking at that report. But that report might be false positve, or at least not caused by this logic introduced by this commit, because CONFIG_SCHED_HW_PRESSURE was not even set in the kernel config[1]. Maybe I should remove the 'reported-by' and 'closes' tags?

[1] https://download.01.org/0day-ci/archive/20240709/[email protected]/config-6.9.0-rc1-00051-g97450eb90965

Quoted 6 lines

kernel/sched/fair.c static bool __update_blocked_others(struct rq *rq, bool *done) · line 9368

Quoted 4 lines
kernel/sched/fair.c +1 -1
-		  update_hw_load_avg(now, rq, hw_pressure) |
+		  update_hw_load_avg(rq_clock_task(rq), rq, hw_pressure) |
  		  update_irq_load_avg(rq, 0);
  	if (others_have_blocked(rq))

Thanks!

Best,
Chenyu

#3 lore raw
Quoted 25 lines · Chen Yu

Hi Hongyan,

Quoted 17 lines · Hongyan Xia
Quoted 7 lines · Chen Yu

commit 97450eb90965 ("sched/pelt: Remove shift of thermal clock") removed the decay_shift for hw_pressure. While looking at a related bug report, it is found that this commit uses the sched_clock_task() in sched_tick() while replaces the sched_clock_task() with rq_clock_pelt() in __update_blocked_others(). This could bring inconsistence. One possible scenario I can think of is in ___update_load_sum():

Quoted 8 lines

u64 delta = now - sa->last_update_time

'now' could be calculated by rq_clock_pelt() from __update_blocked_others(), and last_update_time was calculated by rq_clock_task() previously from sched_tick(). Usually the former chases after the latter, it cause a very large 'delta' and brings unexpected behavior. Although this should not impact x86 platform in the bug report, it should be fixed for other platforms.

Quoted 4 lines

I agree with this patch but I'm a bit confused here. May I know what you mean by 'should not impact x86 platform in the bug report'? But it closes a bug report on qemu x86_64, so it does have an impact?

Quoted 8 lines

It should not have any impact on x86_64. I added the bug link here because I checked the code while looking at that report. But that report might be false positve, or at least not caused by this logic introduced by this commit, because CONFIG_SCHED_HW_PRESSURE was not even set in the kernel config[1]. Maybe I should remove the 'reported-by' and 'closes' tags?

[1] https://download.01.org/0day-ci/archive/20240709/[email protected]/config-6.9.0-rc1-00051-g97450eb90965

Yeah, it might be a good idea to remove the link to avoid confusion, like you said HW pressure is not compiled in.

Even if there is pressure support, before your patch the big 'delta' should only result in a HW pressure value that decays more than it should, and should not be able to block tasks like in that bug report, so it's very likely that it's unrelated.

author
#4 lore raw
Quoted 32 lines · Hongyan Xia
Quoted 31 lines · Chen Yu

Hi Hongyan,

Quoted 17 lines · Hongyan Xia
Quoted 7 lines · Chen Yu

commit 97450eb90965 ("sched/pelt: Remove shift of thermal clock") removed the decay_shift for hw_pressure. While looking at a related bug report, it is found that this commit uses the sched_clock_task() in sched_tick() while replaces the sched_clock_task() with rq_clock_pelt() in __update_blocked_others(). This could bring inconsistence. One possible scenario I can think of is in ___update_load_sum():

Quoted 8 lines

u64 delta = now - sa->last_update_time

'now' could be calculated by rq_clock_pelt() from __update_blocked_others(), and last_update_time was calculated by rq_clock_task() previously from sched_tick(). Usually the former chases after the latter, it cause a very large 'delta' and brings unexpected behavior. Although this should not impact x86 platform in the bug report, it should be fixed for other platforms.

Quoted 4 lines

I agree with this patch but I'm a bit confused here. May I know what you mean by 'should not impact x86 platform in the bug report'? But it closes a bug report on qemu x86_64, so it does have an impact?

It should not have any impact on x86_64. I added the bug link here because I checked the code while looking at that report. But that report might be false positve, or at least not caused by this logic introduced by this commit, because CONFIG_SCHED_HW_PRESSURE was not even set in the kernel config[1]. Maybe I should remove the 'reported-by' and 'closes' tags?

Quoted 6 lines
Quoted 2 lines

[1] https://download.01.org/0day-ci/archive/20240709/[email protected]/config-6.9.0-rc1-00051-g97450eb90965

Yeah, it might be a good idea to remove the link to avoid confusion, like you said HW pressure is not compiled in.

OK, will do and send a new version.

Quoted 4 lines

Even if there is pressure support, before your patch the big 'delta' should only result in a HW pressure value that decays more than it should, and should not be able to block tasks like in that bug report, so it's very likely that it's unrelated.

Yes, for x86_64, the rq->avg_hw.load_avg is even 0 as it does not have a chance to get accumulated.

thanks,
Chenyu