| Message ID | 1409051215-16788-10-git-send-email-vincent.guittot@linaro.org |
|---|---|
| State | New |
| Headers | show |
On Tue, Aug 26, 2014 at 01:06:52PM +0200, Vincent Guittot wrote: > index 5c2c885..7dfd584 100644 > --- a/include/linux/sched.h > +++ b/include/linux/sched.h > @@ -1073,10 +1073,10 @@ struct sched_avg { > * above by 1024/(1-y). Thus we only need a u32 to store them for all > * choices of y < 1-2^(-32)*1024. > */ > - u32 runnable_avg_sum, runnable_avg_period; > + u32 runnable_avg_sum, runnable_avg_period, running_avg_sum; Seeing how we use runnable_avg_period for both runnable and running, does it make sense to remove the runnable part of it from the name? Also, 4 byte hole here, not sure we've got anything useful to stuff in it though. > u64 last_runnable_update; > s64 decay_count; > - unsigned long load_avg_contrib; > + unsigned long load_avg_contrib, usage_avg_contrib; > }; > Man, I should go look at Yuyang's rewrite of this all again. I just tried to figure out the decay stuff and my head hurts ;-) -- To unsubscribe from this list: send the line "unsubscribe linux-kernel" in the body of a message to majordomo@vger.kernel.org More majordomo info at http://vger.kernel.org/majordomo-info.html Please read the FAQ at http://www.tux.org/lkml/
On 11 September 2014 13:17, Peter Zijlstra <peterz@infradead.org> wrote: > On Tue, Aug 26, 2014 at 01:06:52PM +0200, Vincent Guittot wrote: >> index 5c2c885..7dfd584 100644 >> --- a/include/linux/sched.h >> +++ b/include/linux/sched.h >> @@ -1073,10 +1073,10 @@ struct sched_avg { >> * above by 1024/(1-y). Thus we only need a u32 to store them for all >> * choices of y < 1-2^(-32)*1024. >> */ >> - u32 runnable_avg_sum, runnable_avg_period; >> + u32 runnable_avg_sum, runnable_avg_period, running_avg_sum; > > Seeing how we use runnable_avg_period for both runnable and running, > does it make sense to remove the runnable part of it from the name? It's right > > Also, 4 byte hole here, not sure we've got anything useful to stuff in > it though. I can move all u32 declaration at the end of the struct unless it has been put before any u64 for good reason > >> u64 last_runnable_update; >> s64 decay_count; >> - unsigned long load_avg_contrib; >> + unsigned long load_avg_contrib, usage_avg_contrib; >> }; >> > > Man, I should go look at Yuyang's rewrite of this all again. I just > tried to figure out the decay stuff and my head hurts ;-) -- To unsubscribe from this list: send the line "unsubscribe linux-kernel" in the body of a message to majordomo@vger.kernel.org More majordomo info at http://vger.kernel.org/majordomo-info.html Please read the FAQ at http://www.tux.org/lkml/
On 11 September 2014 13:17, Peter Zijlstra <peterz@infradead.org> wrote: >> }; >> > > Man, I should go look at Yuyang's rewrite of this all again. I just > tried to figure out the decay stuff and my head hurts ;-) Regarding Yuyang's rewrite, i had a patch above his patches to add the running figure that i haven't sent yet -- To unsubscribe from this list: send the line "unsubscribe linux-kernel" in the body of a message to majordomo@vger.kernel.org More majordomo info at http://vger.kernel.org/majordomo-info.html Please read the FAQ at http://www.tux.org/lkml/
On Tue, Aug 26, 2014 at 12:06:52PM +0100, Vincent Guittot wrote: > Add new statistics which reflect the average time a task is running on the > CPU and the sum of the tasks' running on a runqueue. The latter is named > usage_avg_contrib. > > This patch is based on the usage metric that was proposed in the 1st > versions of the per-entity load tracking patchset but that has be removed > afterward. This version differs from the original one in the sense that it's > not linked to task_group. > > The rq's usage_avg_contrib will be used to check if a rq is overloaded or not > instead of trying to compute how many task a group of CPUs can handle > > Signed-off-by: Vincent Guittot <vincent.guittot@linaro.org> I should have read this patch before I did almost the same patch for as part a series to introduce scale-invariance which I am about to post :( The only difference I see is slightly different naming and that, AFAICT, task group usage is not accounted for in this patch. Can we add the support for task groups as well? I can provide a patch based on this one if you want. Also, since more than half this patch comes directly from PJT's original patch I would add "by Paul Turner <pjt@google.com>" somewhere in the text above. [...] > +static inline void __update_task_entity_usage(struct sched_entity *se) > +{ > + u32 contrib; > + > + /* avoid overflowing a 32-bit type w/ SCHED_LOAD_SCALE */ > + contrib = se->avg.running_avg_sum * scale_load_down(SCHED_LOAD_SCALE); > + contrib /= (se->avg.runnable_avg_period + 1); > + se->avg.usage_avg_contrib = scale_load(contrib); > +} > + > +static long __update_entity_usage_avg_contrib(struct sched_entity *se) > +{ > + long old_contrib = se->avg.usage_avg_contrib; > + > + if (entity_is_task(se)) > + __update_task_entity_usage(se); Groups are never updated? As said above. I have code that does it. Morten > + > + return se->avg.usage_avg_contrib - old_contrib; > +} > + -- To unsubscribe from this list: send the line "unsubscribe linux-kernel" in the body of a message to majordomo@vger.kernel.org More majordomo info at http://vger.kernel.org/majordomo-info.html Please read the FAQ at http://www.tux.org/lkml/
On 15 September 2014 21:15, Morten Rasmussen <morten.rasmussen@arm.com> wrote: > On Tue, Aug 26, 2014 at 12:06:52PM +0100, Vincent Guittot wrote: >> Add new statistics which reflect the average time a task is running on the >> CPU and the sum of the tasks' running on a runqueue. The latter is named >> usage_avg_contrib. >> >> This patch is based on the usage metric that was proposed in the 1st >> versions of the per-entity load tracking patchset but that has be removed >> afterward. This version differs from the original one in the sense that it's >> not linked to task_group. >> >> The rq's usage_avg_contrib will be used to check if a rq is overloaded or not >> instead of trying to compute how many task a group of CPUs can handle >> >> Signed-off-by: Vincent Guittot <vincent.guittot@linaro.org> > > I should have read this patch before I did almost the same patch for as > part a series to introduce scale-invariance which I am about to post :( > > The only difference I see is slightly different naming and that, AFAICT, > task group usage is not accounted for in this patch. Can we add the > support for task groups as well? I can provide a patch based on this one > if you want. > > Also, since more than half this patch comes directly from PJT's original > patch I would add "by Paul Turner <pjt@google.com>" somewhere in the > text above. Sorry, It was obvious for me that per-entity load tracking patchset were done by pjt but i can surely replace "the per-entity load tracking patchset" by "the per-entity load tracking patchset by Paul Turner <pjt@google.com>" > > [...] > >> +static inline void __update_task_entity_usage(struct sched_entity *se) >> +{ >> + u32 contrib; >> + >> + /* avoid overflowing a 32-bit type w/ SCHED_LOAD_SCALE */ >> + contrib = se->avg.running_avg_sum * scale_load_down(SCHED_LOAD_SCALE); >> + contrib /= (se->avg.runnable_avg_period + 1); >> + se->avg.usage_avg_contrib = scale_load(contrib); >> +} >> + >> +static long __update_entity_usage_avg_contrib(struct sched_entity *se) >> +{ >> + long old_contrib = se->avg.usage_avg_contrib; >> + >> + if (entity_is_task(se)) >> + __update_task_entity_usage(se); > > Groups are never updated? > > As said above. I have code that does it. Yes, feel free to send a patch above this one that add group Vincent > > Morten > >> + >> + return se->avg.usage_avg_contrib - old_contrib; >> +} >> + > -- To unsubscribe from this list: send the line "unsubscribe linux-kernel" in the body of a message to majordomo@vger.kernel.org More majordomo info at http://vger.kernel.org/majordomo-info.html Please read the FAQ at http://www.tux.org/lkml/
diff --git a/include/linux/sched.h b/include/linux/sched.h index 5c2c885..7dfd584 100644 --- a/include/linux/sched.h +++ b/include/linux/sched.h @@ -1073,10 +1073,10 @@ struct sched_avg { * above by 1024/(1-y). Thus we only need a u32 to store them for all * choices of y < 1-2^(-32)*1024. */ - u32 runnable_avg_sum, runnable_avg_period; + u32 runnable_avg_sum, runnable_avg_period, running_avg_sum; u64 last_runnable_update; s64 decay_count; - unsigned long load_avg_contrib; + unsigned long load_avg_contrib, usage_avg_contrib; }; #ifdef CONFIG_SCHEDSTATS diff --git a/kernel/sched/fair.c b/kernel/sched/fair.c index 60ae1ce..1fd2131 100644 --- a/kernel/sched/fair.c +++ b/kernel/sched/fair.c @@ -676,7 +676,7 @@ void init_task_runnable_average(struct task_struct *p) p->se.avg.decay_count = 0; slice = sched_slice(task_cfs_rq(p), &p->se) >> 10; - p->se.avg.runnable_avg_sum = slice; + p->se.avg.runnable_avg_sum = p->se.avg.running_avg_sum = slice; p->se.avg.runnable_avg_period = slice; __update_task_entity_contrib(&p->se); } @@ -2289,7 +2289,8 @@ static u32 __compute_runnable_contrib(u64 n) */ static __always_inline int __update_entity_runnable_avg(u64 now, struct sched_avg *sa, - int runnable) + int runnable, + int running) { u64 delta, periods; u32 runnable_contrib; @@ -2328,6 +2329,8 @@ static __always_inline int __update_entity_runnable_avg(u64 now, delta_w = 1024 - delta_w; if (runnable) sa->runnable_avg_sum += delta_w; + if (running) + sa->running_avg_sum += delta_w; sa->runnable_avg_period += delta_w; delta -= delta_w; @@ -2338,6 +2341,8 @@ static __always_inline int __update_entity_runnable_avg(u64 now, sa->runnable_avg_sum = decay_load(sa->runnable_avg_sum, periods + 1); + sa->running_avg_sum = decay_load(sa->running_avg_sum, + periods + 1); sa->runnable_avg_period = decay_load(sa->runnable_avg_period, periods + 1); @@ -2345,12 +2350,16 @@ static __always_inline int __update_entity_runnable_avg(u64 now, runnable_contrib = __compute_runnable_contrib(periods); if (runnable) sa->runnable_avg_sum += runnable_contrib; + if (running) + sa->running_avg_sum += runnable_contrib; sa->runnable_avg_period += runnable_contrib; } /* Remainder of delta accrued against u_0` */ if (runnable) sa->runnable_avg_sum += delta; + if (running) + sa->running_avg_sum += delta; sa->runnable_avg_period += delta; return decayed; @@ -2490,6 +2499,27 @@ static long __update_entity_load_avg_contrib(struct sched_entity *se) return se->avg.load_avg_contrib - old_contrib; } + +static inline void __update_task_entity_usage(struct sched_entity *se) +{ + u32 contrib; + + /* avoid overflowing a 32-bit type w/ SCHED_LOAD_SCALE */ + contrib = se->avg.running_avg_sum * scale_load_down(SCHED_LOAD_SCALE); + contrib /= (se->avg.runnable_avg_period + 1); + se->avg.usage_avg_contrib = scale_load(contrib); +} + +static long __update_entity_usage_avg_contrib(struct sched_entity *se) +{ + long old_contrib = se->avg.usage_avg_contrib; + + if (entity_is_task(se)) + __update_task_entity_usage(se); + + return se->avg.usage_avg_contrib - old_contrib; +} + static inline void subtract_blocked_load_contrib(struct cfs_rq *cfs_rq, long load_contrib) { @@ -2506,7 +2536,7 @@ static inline void update_entity_load_avg(struct sched_entity *se, int update_cfs_rq) { struct cfs_rq *cfs_rq = cfs_rq_of(se); - long contrib_delta; + long contrib_delta, usage_delta; u64 now; /* @@ -2518,16 +2548,20 @@ static inline void update_entity_load_avg(struct sched_entity *se, else now = cfs_rq_clock_task(group_cfs_rq(se)); - if (!__update_entity_runnable_avg(now, &se->avg, se->on_rq)) + if (!__update_entity_runnable_avg(now, &se->avg, se->on_rq, + cfs_rq->curr == se)) return; contrib_delta = __update_entity_load_avg_contrib(se); + usage_delta = __update_entity_usage_avg_contrib(se); if (!update_cfs_rq) return; - if (se->on_rq) + if (se->on_rq) { cfs_rq->runnable_load_avg += contrib_delta; + cfs_rq->usage_load_avg += usage_delta; + } else subtract_blocked_load_contrib(cfs_rq, -contrib_delta); } @@ -2604,6 +2638,7 @@ static inline void enqueue_entity_load_avg(struct cfs_rq *cfs_rq, } cfs_rq->runnable_load_avg += se->avg.load_avg_contrib; + cfs_rq->usage_load_avg += se->avg.usage_avg_contrib; /* we force update consideration on load-balancer moves */ update_cfs_rq_blocked_load(cfs_rq, !wakeup); } @@ -2622,6 +2657,7 @@ static inline void dequeue_entity_load_avg(struct cfs_rq *cfs_rq, update_cfs_rq_blocked_load(cfs_rq, !sleep); cfs_rq->runnable_load_avg -= se->avg.load_avg_contrib; + cfs_rq->usage_load_avg -= se->avg.usage_avg_contrib; if (sleep) { cfs_rq->blocked_load_avg += se->avg.load_avg_contrib; se->avg.decay_count = atomic64_read(&cfs_rq->decay_counter); @@ -2959,6 +2995,7 @@ set_next_entity(struct cfs_rq *cfs_rq, struct sched_entity *se) */ update_stats_wait_end(cfs_rq, se); __dequeue_entity(cfs_rq, se); + update_entity_load_avg(se, 1); } update_stats_curr_start(cfs_rq, se); diff --git a/kernel/sched/sched.h b/kernel/sched/sched.h index 7c0a74e..d625fbb 100644 --- a/kernel/sched/sched.h +++ b/kernel/sched/sched.h @@ -340,7 +340,7 @@ struct cfs_rq { * This allows for the description of both thread and group usage (in * the FAIR_GROUP_SCHED case). */ - unsigned long runnable_load_avg, blocked_load_avg; + unsigned long runnable_load_avg, blocked_load_avg, usage_load_avg; atomic64_t decay_counter; u64 last_decay; atomic_long_t removed_load;
Add new statistics which reflect the average time a task is running on the CPU and the sum of the tasks' running on a runqueue. The latter is named usage_avg_contrib. This patch is based on the usage metric that was proposed in the 1st versions of the per-entity load tracking patchset but that has be removed afterward. This version differs from the original one in the sense that it's not linked to task_group. The rq's usage_avg_contrib will be used to check if a rq is overloaded or not instead of trying to compute how many task a group of CPUs can handle Signed-off-by: Vincent Guittot <vincent.guittot@linaro.org> --- include/linux/sched.h | 4 ++-- kernel/sched/fair.c | 47 ++++++++++++++++++++++++++++++++++++++++++----- kernel/sched/sched.h | 2 +- 3 files changed, 45 insertions(+), 8 deletions(-)