sched: don't account throttle time for empty groups
PATCH v2 1/2 sched: don't account throttle time for empty groups
It is easy for a cfs_rq to become throttled even when it has no enqueued entities (for example, if we have just put_prev()'d the last runnable task of the cfs_rq, and the cfs_rq is out of quota).
Avoid accounting this time towards total throttle time, since it otherwise falsely inflates the stats.
Note that the dequeue path is special, since we normally disallow migrations when a task is in a throttled hierarchy (see throttled_lb_pair()).
--- kernel/sched/fair.c | 17 ++++++++++++++--- 1 file changed, 14 insertions(+), 3 deletions(-)
kernel/sched/fair.c +14 -3
@@ -4873,8 +4873,14 @@ enqueue_entity(struct cfs_rq *cfs_rq, struct sched_entity *se, int flags)if (cfs_rq->nr_running ==1 ) {check_enqueue_throttle (cfs_rq); -if (!throttled_hierarchy (cfs_rq)) +if (!throttled_hierarchy (cfs_rq)) {list_add_leaf_cfs_rq (cfs_rq); + }else { +#ifdef CONFIG_CFS_BANDWIDTH +if (!cfs_rq->throttled_clock) + cfs_rq->throttled_clock =rq_clock (rq_of (cfs_rq)); +#endif + } } }
@@ -5480,7 +5486,9 @@ static bool throttle_cfs_rq(struct cfs_rq *cfs_rq)* throttled-list. rq->lock protects completion. */ cfs_rq->throttled =1 ; - cfs_rq->throttled_clock =rq_clock (rq); +SCHED_WARN_ON (cfs_rq->throttled_clock); +if (cfs_rq->nr_running) + cfs_rq->throttled_clock =rq_clock (rq);return true; }
@@ -5498,7 +5506,10 @@ void unthrottle_cfs_rq(struct cfs_rq *cfs_rq)update_rq_clock (rq);raw_spin_lock (&cfs_b->lock); - cfs_b->throttled_time +=rq_clock (rq) - cfs_rq->throttled_clock; +if (cfs_rq->throttled_clock) { + cfs_b->throttled_time +=rq_clock (rq) - cfs_rq->throttled_clock; + cfs_rq->throttled_clock =0 ; + }list_del_rcu (&cfs_rq->throttled_list);raw_spin_unlock (&cfs_b->lock);
Signature · 1 line
2.41.0.162.gfafddb0af9-goog
PATCH v2 2/2 sched: add throttled time stat for throttled children
We currently export the total throttled time for cgroups that are given a bandwidth limit. This patch extends this accounting to also account the total time that each children cgroup has been throttled.
This is useful to understand the degree to which children have been affected by the throttling control. Children which are not runnable during the entire throttled period, for example, will not show any self-throttling time during this period.
Expose this in a new interface, 'cpu.stat.local', which is similar to how non-hierarchical events are accounted in 'memory.events.local'.
---
v2:
- moved export to new cpu.stat.local file, per Tejun's recommendation
include/linux/cgroup-defs.h | 2 ++ kernel/cgroup/cgroup.c | 34 ++++++++++++++++++++++++++++ kernel/sched/core.c | 44 +++++++++++++++++++++++++++++++++++++ kernel/sched/fair.c | 21 +++++++++++++++++- kernel/sched/sched.h | 2 ++ 5 files changed, 102 insertions(+), 1 deletion(-)
include/linux/cgroup-defs.h +2 -0
@@ -661,6 +661,8 @@ struct cgroup_subsys {void (*css_rstat_flush)(struct cgroup_subsys_state *css,int cpu);int (*css_extra_stat_show)(struct seq_file *seq,struct cgroup_subsys_state *css); +int (*css_local_stat_show)(struct seq_file *seq, +struct cgroup_subsys_state *css);int (*can_attach)(struct cgroup_taskset *tset);void (*cancel_attach)(struct cgroup_taskset *tset);
kernel/cgroup/cgroup.c +34 -0
@@ -3726,6 +3726,36 @@ static int cpu_stat_show(struct seq_file *seq, void *v)return ret; } +static int __maybe_unused cgroup_local_stat_show (struct seq_file *seq, +struct cgroup *cgrp,int ssid) +{ +struct cgroup_subsys *ss = cgroup_subsys[ssid]; +struct cgroup_subsys_state *css; +int ret; + +if (!ss->css_local_stat_show) +return 0 ; + + css =cgroup_tryget_css (cgrp, ss); +if (!css) +return 0 ; + + ret = ss->css_local_stat_show (seq, css); +css_put (css); +return ret; +} + +static int cpu_local_stat_show (struct seq_file *seq,void *v) +{ +struct cgroup __maybe_unused *cgrp =seq_css (seq)->cgroup; +int ret =0 ; + +#ifdef CONFIG_CGROUP_SCHED + ret =cgroup_local_stat_show (seq, cgrp, cpu_cgrp_id); +#endif +return ret; +} +#ifdef CONFIG_PSI static int cgroup_io_pressure_show (struct seq_file *seq,void *v) {
@@ -5276,6 +5306,10 @@ static struct cftype cgroup_base_files[] = { .name = "cpu.stat", .seq_show = cpu_stat_show, }, + { + .name = "cpu.stat.local", + .seq_show = cpu_local_stat_show, + }, { }/* terminate */ };
kernel/sched/core.c +44 -0
68 lines not on this page · raw message
kernel/sched/fair.c +20 -1
@@ -4877,8 +4877,12 @@ enqueue_entity(struct cfs_rq *cfs_rq, struct sched_entity *se, int flags)list_add_leaf_cfs_rq (cfs_rq); }else {#ifdef CONFIG_CFS_BANDWIDTH +struct rq *rq =rq_of (cfs_rq); +if (!cfs_rq->throttled_clock) - cfs_rq->throttled_clock =rq_clock (rq_of (cfs_rq)); + cfs_rq->throttled_clock =rq_clock (rq); +if (!cfs_rq->throttled_clock_self) + cfs_rq->throttled_clock_self =rq_clock (rq);#endif } }
@@ -5385,6 +5389,17 @@ static int tg_unthrottle_up(struct task_group *tg, void *data)list_add_leaf_cfs_rq (cfs_rq); } +if (cfs_rq->throttled_clock_self) { +u64 delta =rq_clock (rq) - cfs_rq->throttled_clock_self; + + cfs_rq->throttled_clock_self =0 ; + +if (SCHED_WARN_ON ((s64 )delta <0 )) + delta =0 ; + + cfs_rq->throttled_clock_self_time += delta; + } +return 0 ; }
@@ -5400,6 +5415,10 @@ static int tg_throttle_down(struct task_group *tg, void *data) } cfs_rq->throttle_count++; +SCHED_WARN_ON (cfs_rq->throttled_clock_self); +if (cfs_rq->nr_running) + cfs_rq->throttled_clock_self =rq_clock (rq); +return 0 ; }
kernel/sched/sched.h +2 -0
@@ -642,6 +642,8 @@ struct cfs_rq {u64 throttled_clock;u64 throttled_clock_pelt;u64 throttled_clock_pelt_time; +u64 throttled_clock_self; +u64 throttled_clock_self_time;int throttled;int throttle_count;struct list_head throttled_list;
Signature · 1 line
2.41.0.162.gfafddb0af9-goog
[tip: sched/core] sched: Add throttled time stat for throttled children
The following commit has been merged into the sched/core branch of tip:
The following commit has been merged into the sched/core branch of tip:
Commit-ID: 06f5576bf513824b40d535d6b872c0265bb44c0d Gitweb: https://git.kernel.org/tip/06f5576bf513824b40d535d6b872c0265bb44c0d Author: Josh Don <[email protected]> AuthorDate: Mon, 12 Jun 2023 16:27:48 -07:00 Committer: Peter Zijlstra <[email protected]> CommitterDate: Thu, 15 Jun 2023 13:28:20 +02:00
sched: Add throttled time stat for throttled children
We currently export the total throttled time for cgroups that are given a bandwidth limit. This patch extends this accounting to also account the total time that each children cgroup has been throttled.
This is useful to understand the degree to which children have been affected by the throttling control. Children which are not runnable during the entire throttled period, for example, will not show any self-throttling time during this period.
Expose this in a new interface, 'cpu.stat.local', which is similar to how non-hierarchical events are accounted in 'memory.events.local'.
--- include/linux/cgroup-defs.h | 2 ++- kernel/cgroup/cgroup.c | 34 ++++++++++++++++++++++++++++- kernel/sched/core.c | 44 ++++++++++++++++++++++++++++++++++++- kernel/sched/fair.c | 21 ++++++++++++++++- kernel/sched/sched.h | 2 ++- 5 files changed, 102 insertions(+), 1 deletion(-)
include/linux/cgroup-defs.h +2 -0
@@ -661,6 +661,8 @@ struct cgroup_subsys {void (*css_rstat_flush)(struct cgroup_subsys_state *css,int cpu);int (*css_extra_stat_show)(struct seq_file *seq,struct cgroup_subsys_state *css); +int (*css_local_stat_show)(struct seq_file *seq, +struct cgroup_subsys_state *css);int (*can_attach)(struct cgroup_taskset *tset);void (*cancel_attach)(struct cgroup_taskset *tset);
kernel/cgroup/cgroup.c +34 -0
@@ -3726,6 +3726,36 @@ static int cpu_stat_show(struct seq_file *seq, void *v)return ret; } +static int __maybe_unused cgroup_local_stat_show (struct seq_file *seq, +struct cgroup *cgrp,int ssid) +{ +struct cgroup_subsys *ss = cgroup_subsys[ssid]; +struct cgroup_subsys_state *css; +int ret; + +if (!ss->css_local_stat_show) +return 0 ; + + css =cgroup_tryget_css (cgrp, ss); +if (!css) +return 0 ; + + ret = ss->css_local_stat_show (seq, css); +css_put (css); +return ret; +} + +static int cpu_local_stat_show (struct seq_file *seq,void *v) +{ +struct cgroup __maybe_unused *cgrp =seq_css (seq)->cgroup; +int ret =0 ; + +#ifdef CONFIG_CGROUP_SCHED + ret =cgroup_local_stat_show (seq, cgrp, cpu_cgrp_id); +#endif +return ret; +} +#ifdef CONFIG_PSI static int cgroup_io_pressure_show (struct seq_file *seq,void *v) {
@@ -5276,6 +5306,10 @@ static struct cftype cgroup_base_files[] = { .name = "cpu.stat", .seq_show = cpu_stat_show, }, + { + .name = "cpu.stat.local", + .seq_show = cpu_local_stat_show, + }, { }/* terminate */ };
kernel/sched/core.c +44 -0
68 lines not on this page · raw message
kernel/sched/fair.c +20 -1
@@ -4877,8 +4877,12 @@ enqueue_entity(struct cfs_rq *cfs_rq, struct sched_entity *se, int flags)list_add_leaf_cfs_rq (cfs_rq); }else {#ifdef CONFIG_CFS_BANDWIDTH +struct rq *rq =rq_of (cfs_rq); +if (!cfs_rq->throttled_clock) - cfs_rq->throttled_clock =rq_clock (rq_of (cfs_rq)); + cfs_rq->throttled_clock =rq_clock (rq); +if (!cfs_rq->throttled_clock_self) + cfs_rq->throttled_clock_self =rq_clock (rq);#endif } }
@@ -5385,6 +5389,17 @@ static int tg_unthrottle_up(struct task_group *tg, void *data)list_add_leaf_cfs_rq (cfs_rq); } +if (cfs_rq->throttled_clock_self) { +u64 delta =rq_clock (rq) - cfs_rq->throttled_clock_self; + + cfs_rq->throttled_clock_self =0 ; + +if (SCHED_WARN_ON ((s64 )delta <0 )) + delta =0 ; + + cfs_rq->throttled_clock_self_time += delta; + } +return 0 ; }
@@ -5400,6 +5415,10 @@ static int tg_throttle_down(struct task_group *tg, void *data) } cfs_rq->throttle_count++; +SCHED_WARN_ON (cfs_rq->throttled_clock_self); +if (cfs_rq->nr_running) + cfs_rq->throttled_clock_self =rq_clock (rq); +return 0 ; }
kernel/sched/sched.h +2 -0
@@ -636,6 +636,8 @@ struct cfs_rq {u64 throttled_clock;u64 throttled_clock_pelt;u64 throttled_clock_pelt_time; +u64 throttled_clock_self; +u64 throttled_clock_self_time;int throttled;int throttle_count;struct list_head throttled_list;
Michal Koutný
maintainerRe: [PATCH v2 2/2] sched: add throttled time stat for throttled children
In reply to Josh DonQuoted 2 lines · Josh Don
We currently export the total throttled time for cgroups that are given a bandwidth limit.
I assume you refer to cpu.stat:throttled_usec (from struct cfs_bandwidth) -- notice that the value is not properly hierarchical despite v2 filename.
Quoted 2 lines
This patch extends this accounting to also account the total time that each children cgroup has been throttled.
IIUC, this is visible on inner-nodes cpu cgroups (i.e. with no tasks)?
IOW, wouldn't you get the intended information if hierarchical summing was added/fixed for cpu.stat:throttled_usec?
Thanks,
Michal
Re: [PATCH v2 2/2] sched: add throttled time stat for throttled children
Hi Michal,
Quoted 9 lines · Michal Koutný
Quoted 2 lines · Josh Don
We currently export the total throttled time for cgroups that are given a bandwidth limit.
I assume you refer to cpu.stat:throttled_usec (from struct
cfs_bandwidth) -- notice that the value is not properly hierarchical despite v2 filename.
Quoted 7 lines
Quoted 2 lines
This patch extends this accounting to also account the total time that each children cgroup has been throttled.
IIUC, this is visible on inner-nodes cpu cgroups (i.e. with no tasks)?
IOW, wouldn't you get the intended information if hierarchical summing was added/fixed for cpu.stat:throttled_usec?
It isn't currently hierarchical in the sense that the inner-nodes don't themselves account their throttled time, but the summation at the top is still correct. This patch is intended to close the gap. I suppose your question here is why not simply make the existing throttled_usec export properly hierarchical, and avoid the extra stat export here. I think it might be useful to still expose a non-hierarchical metric indicating the throttled time due to the group's own configured limit; the accounting can look strange with nested bandwidth limits. Not strongly opposed to the idea, but your hierarchical accounting proposal is essentially what this patch adds.
[tip: sched/core] sched: Don't account throttle time for empty groups
In reply to Josh DonThe following commit has been merged into the sched/core branch of tip:
The following commit has been merged into the sched/core branch of tip:
Commit-ID: eed673d4f701e17b8f887dbcbfd3b4bcc53503ec Gitweb: https://git.kernel.org/tip/eed673d4f701e17b8f887dbcbfd3b4bcc53503ec Author: Josh Don <[email protected]> AuthorDate: Mon, 12 Jun 2023 16:27:47 -07:00 Committer: Peter Zijlstra <[email protected]> CommitterDate: Thu, 15 Jun 2023 13:28:19 +02:00
sched: Don't account throttle time for empty groups
It is easy for a cfs_rq to become throttled even when it has no enqueued entities (for example, if we have just put_prev()'d the last runnable task of the cfs_rq, and the cfs_rq is out of quota).
Avoid accounting this time towards total throttle time, since it otherwise falsely inflates the stats.
Note that the dequeue path is special, since we normally disallow migrations when a task is in a throttled hierarchy (see throttled_lb_pair()).
--- kernel/sched/fair.c | 17 ++++++++++++++--- 1 file changed, 14 insertions(+), 3 deletions(-)
kernel/sched/fair.c +14 -3
@@ -4873,8 +4873,14 @@ enqueue_entity(struct cfs_rq *cfs_rq, struct sched_entity *se, int flags)if (cfs_rq->nr_running ==1 ) {check_enqueue_throttle (cfs_rq); -if (!throttled_hierarchy (cfs_rq)) +if (!throttled_hierarchy (cfs_rq)) {list_add_leaf_cfs_rq (cfs_rq); + }else { +#ifdef CONFIG_CFS_BANDWIDTH +if (!cfs_rq->throttled_clock) + cfs_rq->throttled_clock =rq_clock (rq_of (cfs_rq)); +#endif + } } }
@@ -5480,7 +5486,9 @@ done:* throttled-list. rq->lock protects completion. */ cfs_rq->throttled =1 ; - cfs_rq->throttled_clock =rq_clock (rq); +SCHED_WARN_ON (cfs_rq->throttled_clock); +if (cfs_rq->nr_running) + cfs_rq->throttled_clock =rq_clock (rq);return true; }
@@ -5498,7 +5506,10 @@ void unthrottle_cfs_rq(struct cfs_rq *cfs_rq)update_rq_clock (rq);raw_spin_lock (&cfs_b->lock); - cfs_b->throttled_time +=rq_clock (rq) - cfs_rq->throttled_clock; +if (cfs_rq->throttled_clock) { + cfs_b->throttled_time +=rq_clock (rq) - cfs_rq->throttled_clock; + cfs_rq->throttled_clock =0 ; + }list_del_rcu (&cfs_rq->throttled_list);raw_spin_unlock (&cfs_b->lock);