sched/pelt: Use rq_clock_task() for hw_pressure
PATCH v3 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. This commit uses the sched_clock_task() in sched_tick() while it 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.
Fix this by using rq_clock_task() inside update_hw_load_avg(), because:
1. hw_pressure doesn't care about invariance.
2. avoid any inconsistence in the future.
---
v2>v3:
call rq_clock_task() inside update_hw_load_avg() to avoid any inconsistence in the future. (Qais Yousef)
Added comments around update_hw_load_avg(). (Qais Yousef)v1->v2:
Added Hongyan's Reviewed-by tag.
Removed the Reported-by/Closes tags because they are not related to this fix.(Hongyan Xia)
--- kernel/sched/core.c | 2 +- kernel/sched/fair.c | 2 +- kernel/sched/pelt.c | 5 +++-- kernel/sched/pelt.h | 6 +++--- 4 files changed, 8 insertions(+), 7 deletions(-)
kernel/sched/core.c +1 -1
@@ -5462,7 +5462,7 @@ void sched_tick(void)update_rq_clock (rq); hw_pressure =arch_scale_hw_pressure (cpu_of (rq)); -update_hw_load_avg (rq_clock_task (rq), rq, hw_pressure); +update_hw_load_avg (rq, hw_pressure); curr->sched_class->task_tick (rq, curr,0 );if (sched_feat (LATENCY_WARN )) resched_latency =cpu_resched_latency (rq);
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, hw_pressure) |update_irq_load_avg (rq,0 );if (others_have_blocked (rq))
kernel/sched/pelt.c +3 -2
@@ -400,9 +400,10 @@ int update_dl_rq_load_avg(u64 now, struct rq *rq, int running)* capped capacity a cpu due to a HW event. */ -int update_hw_load_avg (u64 now,struct rq *rq,u64 capacity) +int update_hw_load_avg (struct rq *rq,u64 capacity) { -if (___update_load_sum (now, &rq->avg_hw, +/* hw_pressure doesn't care about invariance */ +if (___update_load_sum (rq_clock_task (rq), &rq->avg_hw, capacity, capacity, capacity)) {
kernel/sched/pelt.h +3 -3
@@ -8,7 +8,7 @@ int update_rt_rq_load_avg(u64 now, struct rq *rq, int running);int update_dl_rq_load_avg (u64 now,struct rq *rq,int running);#ifdef CONFIG_SCHED_HW_PRESSURE -int update_hw_load_avg (u64 now,struct rq *rq,u64 capacity); +int update_hw_load_avg (struct rq *rq,u64 capacity);static inline u64 hw_load_avg (struct rq *rq) {
@@ -16,7 +16,7 @@ static inline u64 hw_load_avg(struct rq *rq) }#else static inline int -update_hw_load_avg (u64 now,struct rq *rq,u64 capacity) +update_hw_load_avg (struct rq *rq,u64 capacity) {return 0 ; }
@@ -202,7 +202,7 @@ update_dl_rq_load_avg(u64 now, struct rq *rq, int running) }static inline int -update_hw_load_avg (u64 now,struct rq *rq,u64 capacity) +update_hw_load_avg (struct rq *rq,u64 capacity) {return 0 ; }
Signature · 1 line
2.25.1
Vincent Guittot
maintainerSorry for the late reply on this
Quoted 5 lines · Chen Yu
commit
97450eb90965("sched/pelt: Remove shift of thermal clock") removed thedecay_shiftforhw_pressure. This commit uses thesched_clock_task()insched_tick()while it replaces thesched_clock_task()withrq_clock_pelt()in__update_blocked_others().
Good catch, it should be sched_clock_task() everywhere
Quoted 5 lines
This could bring inconsistence. One possible scenario I can think of is in
___update_load_sum():u64 delta = now -
sa->last_update_time
Quoted 8 lines
'now' could be calculated by
rq_clock_pelt()from__update_blocked_others(), andlast_update_timewas calculated byrq_clock_task()previously fromsched_tick(). Usually the former chases after the latter, it cause a very large 'delta' and brings unexpected behavior.Fix this by using
rq_clock_task()insideupdate_hw_load_avg(), because:
No, don't call the rq_clock_task() inside update_hw_load_avg(), keep it outside. V2 was the correct solution
Nack for this v3. Please come back on v2
Quoted 100 lines
1.
hw_pressuredoesn't care about invariance.
2. avoid any inconsistence in the future.Fixes 97450eb90965 ("sched/pelt: Remove shift of thermal clock")Suggested-by Qais Yousef <[email protected]>Reviewed-by Hongyan Xia <[email protected]>Signed-off-by Chen Yu <[email protected]>---
v2>v3:
callrq_clock_task()insideupdate_hw_load_avg()to avoid any inconsistence in the future. (Qais Yousef)
Added comments aroundupdate_hw_load_avg(). (Qais Yousef)v1->v2:
Added Hongyan's Reviewed-by tag.
Removed the Reported-by/Closes tags because they are not related to this fix.(Hongyan Xia)--- kernel/sched/core.c | 2 +- kernel/sched/fair.c | 2 +- kernel/sched/pelt.c | 5 +++-- kernel/sched/pelt.h | 6 +++--- 4 files changed, 8 insertions(+), 7 deletions(-)kernel/sched/core.c +1
-1@@ -5462,7 +5462,7 @@ void sched_tick(void)update_rq_clock (rq); hw_pressure =arch_scale_hw_pressure (cpu_of (rq)); -update_hw_load_avg (rq_clock_task (rq), rq, hw_pressure); +update_hw_load_avg (rq, hw_pressure); curr->sched_class->task_tick (rq, curr,0 );if (sched_feat (LATENCY_WARN )) resched_latency =cpu_resched_latency (rq);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, hw_pressure) |update_irq_load_avg (rq,0 );if (others_have_blocked (rq))kernel/sched/pelt.c +3
-2@@ -400,9 +400,10 @@ int update_dl_rq_load_avg(u64 now, struct rq *rq, int running)* capped capacity a cpu due to a HW event. */ -int update_hw_load_avg (u64 now,struct rq *rq,u64 capacity) +int update_hw_load_avg (struct rq *rq,u64 capacity) { -if (___update_load_sum (now, &rq->avg_hw, +/* hw_pressure doesn't care about invariance */ +if (___update_load_sum (rq_clock_task (rq), &rq->avg_hw, capacity, capacity, capacity)) {kernel/sched/pelt.h +3
-4@@ -8,7 +8,7 @@ int update_rt_rq_load_avg(u64 now, struct rq *rq, int running);int update_dl_rq_load_avg (u64 now,struct rq *rq,int running);#ifdef CONFIG_SCHED_HW_PRESSURE -int update_hw_load_avg (u64 now,struct rq *rq,u64 capacity); +int update_hw_load_avg (struct rq *rq,u64 capacity);static inline u64 hw_load_avg (struct rq *rq) {@@ -16,7 +16,7 @@ static inline u64 hw_load_avg(struct rq *rq) }#else static inline int -update_hw_load_avg (u64 now,struct rq *rq,u64 capacity) +update_hw_load_avg (struct rq *rq,u64 capacity) {return 0 ; }@@ -202,7 +202,7 @@ update_dl_rq_load_avg(u64 now, struct rq *rq, int running) }static inline int -update_hw_load_avg (u64 now,struct rq *rq,u64 capacity) +update_hw_load_avg (struct rq *rq,u64 capacity) {return 0 ; } --2.25.1
Quoted 22 lines · Vincent Guittot
Sorry for the late reply on this
Quoted 5 lines · Chen Yu
commit
97450eb90965("sched/pelt: Remove shift of thermal clock") removed thedecay_shiftforhw_pressure. This commit uses thesched_clock_task()insched_tick()while it replaces thesched_clock_task()withrq_clock_pelt()in__update_blocked_others().Good catch, it should be
sched_clock_task()everywhereQuoted 11 lines
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(), andlast_update_timewas calculated byrq_clock_task()previously fromsched_tick(). Usually the former chases after the latter, it cause a very large 'delta' and brings unexpected behavior.
Quoted 6 lines
Quoted 2 lines
Fix this by using
rq_clock_task()insideupdate_hw_load_avg(), because:No, don't call the
rq_clock_task()insideupdate_hw_load_avg(), keep it outside. V2 was the correct solution
Nack for this v3. Please come back on v2
OK, thanks for taking a look, I'll refine v2 and send v4 out.
thanks,
Chenyu
Quoted 22 lines · Vincent Guittot
Sorry for the late reply on this
Quoted 5 lines · Chen Yu
commit
97450eb90965("sched/pelt: Remove shift of thermal clock") removed thedecay_shiftforhw_pressure. This commit uses thesched_clock_task()insched_tick()while it replaces thesched_clock_task()withrq_clock_pelt()in__update_blocked_others().Good catch, it should be
sched_clock_task()everywhereQuoted 11 lines
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(), andlast_update_timewas calculated byrq_clock_task()previously fromsched_tick(). Usually the former chases after the latter, it cause a very large 'delta' and brings unexpected behavior.
Quoted 5 lines
Quoted 2 lines
Fix this by using
rq_clock_task()insideupdate_hw_load_avg(), because:No, don't call the
rq_clock_task()insideupdate_hw_load_avg(), keep it outside. V2 was the correct solution
Curious, why this is not correct?
Quoted 102 lines
Nack for this v3. Please come back on v2
Quoted 100 lines
1.
hw_pressuredoesn't care about invariance.
2. avoid any inconsistence in the future.Fixes 97450eb90965 ("sched/pelt: Remove shift of thermal clock")Suggested-by Qais Yousef <[email protected]>Reviewed-by Hongyan Xia <[email protected]>Signed-off-by Chen Yu <[email protected]>---
v2>v3:
callrq_clock_task()insideupdate_hw_load_avg()to avoid any inconsistence in the future. (Qais Yousef)
Added comments aroundupdate_hw_load_avg(). (Qais Yousef)v1->v2:
Added Hongyan's Reviewed-by tag.
Removed the Reported-by/Closes tags because they are not related to this fix.(Hongyan Xia)--- kernel/sched/core.c | 2 +- kernel/sched/fair.c | 2 +- kernel/sched/pelt.c | 5 +++-- kernel/sched/pelt.h | 6 +++--- 4 files changed, 8 insertions(+), 7 deletions(-)kernel/sched/core.c +1
-1@@ -5462,7 +5462,7 @@ void sched_tick(void)update_rq_clock (rq); hw_pressure =arch_scale_hw_pressure (cpu_of (rq)); -update_hw_load_avg (rq_clock_task (rq), rq, hw_pressure); +update_hw_load_avg (rq, hw_pressure); curr->sched_class->task_tick (rq, curr,0 );if (sched_feat (LATENCY_WARN )) resched_latency =cpu_resched_latency (rq);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, hw_pressure) |update_irq_load_avg (rq,0 );if (others_have_blocked (rq))kernel/sched/pelt.c +3
-2@@ -400,9 +400,10 @@ int update_dl_rq_load_avg(u64 now, struct rq *rq, int running)* capped capacity a cpu due to a HW event. */ -int update_hw_load_avg (u64 now,struct rq *rq,u64 capacity) +int update_hw_load_avg (struct rq *rq,u64 capacity) { -if (___update_load_sum (now, &rq->avg_hw, +/* hw_pressure doesn't care about invariance */ +if (___update_load_sum (rq_clock_task (rq), &rq->avg_hw, capacity, capacity, capacity)) {kernel/sched/pelt.h +3
-4@@ -8,7 +8,7 @@ int update_rt_rq_load_avg(u64 now, struct rq *rq, int running);int update_dl_rq_load_avg (u64 now,struct rq *rq,int running);#ifdef CONFIG_SCHED_HW_PRESSURE -int update_hw_load_avg (u64 now,struct rq *rq,u64 capacity); +int update_hw_load_avg (struct rq *rq,u64 capacity);static inline u64 hw_load_avg (struct rq *rq) {@@ -16,7 +16,7 @@ static inline u64 hw_load_avg(struct rq *rq) }#else static inline int -update_hw_load_avg (u64 now,struct rq *rq,u64 capacity) +update_hw_load_avg (struct rq *rq,u64 capacity) {return 0 ; }@@ -202,7 +202,7 @@ update_dl_rq_load_avg(u64 now, struct rq *rq, int running) }static inline int -update_hw_load_avg (u64 now,struct rq *rq,u64 capacity) +update_hw_load_avg (struct rq *rq,u64 capacity) {return 0 ; } --2.25.1
Vincent Guittot
maintainerQuoted 24 lines · Qais Yousef
Quoted 22 lines · Vincent Guittot
Sorry for the late reply on this
Quoted 5 lines · Chen Yu
commit
97450eb90965("sched/pelt: Remove shift of thermal clock") removed thedecay_shiftforhw_pressure. This commit uses thesched_clock_task()insched_tick()while it replaces thesched_clock_task()withrq_clock_pelt()in__update_blocked_others().Good catch, it should be
sched_clock_task()everywhereQuoted 11 lines
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(), andlast_update_timewas calculated byrq_clock_task()previously fromsched_tick(). Usually the former chases after the latter, it cause a very large 'delta' and brings unexpected behavior.
Quoted 7 lines
Quoted 5 lines
Quoted 2 lines
Fix this by using
rq_clock_task()insideupdate_hw_load_avg(), because:No, don't call the
rq_clock_task()insideupdate_hw_load_avg(), keep it outside. V2 was the correct solutionCurious, why this is not correct?
Not correct in the sense that don't embed the clock in the pelt function but keep it outside like others. The only one doing this is update_irq_load_avg() which is one my todo list for a while to be aligned with others (just missing time)
Quoted 103 lines
Quoted 103 lines
Nack for this v3. Please come back on v2
Quoted 100 lines
1.
hw_pressuredoesn't care about invariance.
2. avoid any inconsistence in the future.Fixes 97450eb90965 ("sched/pelt: Remove shift of thermal clock")Suggested-by Qais Yousef <[email protected]>Reviewed-by Hongyan Xia <[email protected]>Signed-off-by Chen Yu <[email protected]>---
v2>v3:
callrq_clock_task()insideupdate_hw_load_avg()to avoid any inconsistence in the future. (Qais Yousef)
Added comments aroundupdate_hw_load_avg(). (Qais Yousef)v1->v2:
Added Hongyan's Reviewed-by tag.
Removed the Reported-by/Closes tags because they are not related to this fix.(Hongyan Xia)--- kernel/sched/core.c | 2 +- kernel/sched/fair.c | 2 +- kernel/sched/pelt.c | 5 +++-- kernel/sched/pelt.h | 6 +++--- 4 files changed, 8 insertions(+), 7 deletions(-)kernel/sched/core.c +1
-1@@ -5462,7 +5462,7 @@ void sched_tick(void)update_rq_clock (rq); hw_pressure =arch_scale_hw_pressure (cpu_of (rq)); -update_hw_load_avg (rq_clock_task (rq), rq, hw_pressure); +update_hw_load_avg (rq, hw_pressure); curr->sched_class->task_tick (rq, curr,0 );if (sched_feat (LATENCY_WARN )) resched_latency =cpu_resched_latency (rq);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, hw_pressure) |update_irq_load_avg (rq,0 );if (others_have_blocked (rq))kernel/sched/pelt.c +3
-2@@ -400,9 +400,10 @@ int update_dl_rq_load_avg(u64 now, struct rq *rq, int running)* capped capacity a cpu due to a HW event. */ -int update_hw_load_avg (u64 now,struct rq *rq,u64 capacity) +int update_hw_load_avg (struct rq *rq,u64 capacity) { -if (___update_load_sum (now, &rq->avg_hw, +/* hw_pressure doesn't care about invariance */ +if (___update_load_sum (rq_clock_task (rq), &rq->avg_hw, capacity, capacity, capacity)) {kernel/sched/pelt.h +3
-4@@ -8,7 +8,7 @@ int update_rt_rq_load_avg(u64 now, struct rq *rq, int running);int update_dl_rq_load_avg (u64 now,struct rq *rq,int running);#ifdef CONFIG_SCHED_HW_PRESSURE -int update_hw_load_avg (u64 now,struct rq *rq,u64 capacity); +int update_hw_load_avg (struct rq *rq,u64 capacity);static inline u64 hw_load_avg (struct rq *rq) {@@ -16,7 +16,7 @@ static inline u64 hw_load_avg(struct rq *rq) }#else static inline int -update_hw_load_avg (u64 now,struct rq *rq,u64 capacity) +update_hw_load_avg (struct rq *rq,u64 capacity) {return 0 ; }@@ -202,7 +202,7 @@ update_dl_rq_load_avg(u64 now, struct rq *rq, int running) }static inline int -update_hw_load_avg (u64 now,struct rq *rq,u64 capacity) +update_hw_load_avg (struct rq *rq,u64 capacity) {return 0 ; } --2.25.1