sched_ext: pass the initial cpu.idle state in scx_cgroup_init_args
Subsystems (3)
v3 cover sched_ext: pass the initial cpu.idle state in scx_cgroup_init_args
From: Tao Cui <[email protected]>
This carries the initial cpu.idle state of a cgroup through ops.cgroup_init(), which was the only cpu controller knob whose initial value never reached the BPF scheduler. Patch 1 is the fix; patch 2 renames tg->scx.idle to tg->scx.sched_idle, suggested in review, as a separate cleanup.
Changes since v2:
- rename the new field to sched_idle, as a bare "idle" reads as CPU idle state in sched_ext (Tejun)
- split the tg->scx.idle rename into 2/2 so 1/2 stays minimal for stable backport
- add the Fixes: tag (Andrea)
- regenerate on top of current linux-next, which also resolves the CI conflict reported against v2
v2: https://lore.kernel.org/r/[email protected]
v1: https://lore.kernel.org/r/[email protected]
Tao Cui (2):
sched_ext: pass the initial cpu.idle state in scx_cgroup_init_args
sched_ext: rename tg->scx.idle to tg->scx.sched_idle
include/linux/sched/ext.h | 2 +- kernel/sched/ext/ext.c | 8 +++++--- kernel/sched/ext/internal.h | 3 +++ kernel/sched/ext/sub.c | 2 ++ 4 files changed, 11 insertions(+), 4 deletions(-)
Signature · 1 line
2.43.0
PATCH v3 1/2 sched_ext: pass the initial cpu.idle state in scx_cgroup_init_args
From: Tao Cui <[email protected]>
scx_cgroup_init_args carries the initial weight and bandwidth control parameters of a cgroup to ops.cgroup_init(), but not its cpu.idle state. A cgroup that was already configured idle before the scheduler was loaded (or before it was onlined under it) is presented as non-idle, and the BPF scheduler only learns about it if cpu.idle is written again later.
Add the sched_idle state to scx_cgroup_init_args and fill it in all four places that build the args: scx_tg_online() for cgroups onlined under the scheduler, scx_cgroup_init() for cgroups that already exist when the scheduler is loaded, and the sub-scheduler handover paths scx_cgroup_claim_subtree() and scx_cgroup_return_subtree().
Verified in a VM with a probe scheduler printing the init args: a cgroup configured cpu.idle=1 before loading shows sched_idle=1 in ops.cgroup_init(), the default shows 0, and later cpu.idle writes still come through ops.cgroup_set_idle(). The sub-scheduler paths are compile-tested only.
---
v2 -> v3: Rename the new field to sched_idle, as a bare "idle" reads as CPU idle state in sched_ext, per Tejun. Add the Fixes: tag suggested by Andrea. Regenerated on top of current linux-next, which also resolves the CI conflict reported for v2. The tg->scx.idle rename now lives in 2/2.
v2: https://lore.kernel.org/r/[email protected]
kernel/sched/ext/ext.c | 4 +++- kernel/sched/ext/internal.h | 3 +++ kernel/sched/ext/sub.c | 2 ++ 3 files changed, 8 insertions(+), 1 deletion(-)
kernel/sched/ext/ext.c +3 -1
@@ -4764,7 +4764,8 @@ int scx_tg_online(struct task_group *tg) { .weight = tg->scx.weight, .bw_period_us = tg->scx.bw_period_us, .bw_quota_us = tg->scx.bw_quota_us, - .bw_burst_us = tg->scx.bw_burst_us }; + .bw_burst_us = tg->scx.bw_burst_us, + .sched_idle = tg->scx.idle }; ret =SCX_CALL_OP_RET (sch, cgroup_init,NULL , tg->css.cgroup, &args);
@@ -5185,6 +5186,7 @@ static int scx_cgroup_init(struct scx_sched *sch) .bw_period_us = tg->scx.bw_period_us, .bw_quota_us = tg->scx.bw_quota_us, .bw_burst_us = tg->scx.bw_burst_us, + .sched_idle = tg->scx.idle, }; ret =SCX_CALL_OP_RET (sch, cgroup_init,NULL , css->cgroup, &args);
kernel/sched/ext/internal.h +3 -0
@@ -259,6 +259,9 @@ struct scx_cgroup_init_args {u64 bw_period_us;u64 bw_quota_us;u64 bw_burst_us; + +/* whether the cgroup is configured SCHED_IDLE via cpu.idle */ +bool sched_idle; };enum scx_cpu_preempt_reason {
kernel/sched/ext/sub.c +2 -0
@@ -1361,6 +1361,7 @@ static s32 scx_cgroup_claim_subtree(struct scx_sched *sch) .bw_period_us = tg->scx.bw_period_us, .bw_quota_us = tg->scx.bw_quota_us, .bw_burst_us = tg->scx.bw_burst_us, + .sched_idle = tg->scx.idle, }; if (tg->scx.sched != parent ||
@@ -1464,6 +1465,7 @@ static void scx_cgroup_return_subtree(struct scx_sched *sch) .bw_period_us = tg->scx.bw_period_us, .bw_quota_us = tg->scx.bw_quota_us, .bw_burst_us = tg->scx.bw_burst_us, + .sched_idle = tg->scx.idle, };/* the first pass must have transferred everything */
Signature · 1 line
2.43.0
PATCH v3 2/2 sched_ext: rename tg->scx.idle to tg->scx.sched_idle
In reply to Tao CuiFrom: Tao Cui <[email protected]>
A bare "idle" reads as CPU idle state in sched_ext (ops.update_idle(), idle cpumasks, scx_bpf_pick_idle_cpu()). cpu.idle is the cgroup analog of the SCHED_IDLE policy, so name the task_group field sched_idle to match scx_cgroup_init_args.sched_idle.
Pure rename, no behavior change.
--- include/linux/sched/ext.h | 2 +- kernel/sched/ext/ext.c | 8 ++++---- kernel/sched/ext/sub.c | 4 ++-- 3 files changed, 7 insertions(+), 7 deletions(-)
include/linux/sched/ext.h +1 -1
@@ -323,7 +323,7 @@ struct scx_task_group {u64 bw_period_us;u64 bw_quota_us;u64 bw_burst_us; -bool idle; +bool sched_idle;#endif };
kernel/sched/ext/ext.c +4 -4
@@ -4682,7 +4682,7 @@ void scx_tg_init(struct task_group *tg) tg->scx.weight =CGROUP_WEIGHT_DFL ; tg->scx.bw_period_us =default_bw_period_us (); tg->scx.bw_quota_us =RUNTIME_INF ; - tg->scx.idle = false; + tg->scx.sched_idle = false; }/**
@@ -4765,7 +4765,7 @@ int scx_tg_online(struct task_group *tg) .bw_period_us = tg->scx.bw_period_us, .bw_quota_us = tg->scx.bw_quota_us, .bw_burst_us = tg->scx.bw_burst_us, - .sched_idle = tg->scx.idle }; + .sched_idle = tg->scx.sched_idle }; ret =SCX_CALL_OP_RET (sch, cgroup_init,NULL , tg->css.cgroup, &args);
@@ -4935,7 +4935,7 @@ void scx_group_set_idle(struct task_group *tg, bool idle)SCX_CALL_OP (sch, cgroup_set_idle,NULL ,tg_cgrp (tg), idle);/* Update the task group's idle state */ - tg->scx.idle = idle; + tg->scx.sched_idle = idle;percpu_up_read (&scx_cgroup_ops_rwsem); }
@@ -5186,7 +5186,7 @@ static int scx_cgroup_init(struct scx_sched *sch) .bw_period_us = tg->scx.bw_period_us, .bw_quota_us = tg->scx.bw_quota_us, .bw_burst_us = tg->scx.bw_burst_us, - .sched_idle = tg->scx.idle, + .sched_idle = tg->scx.sched_idle, }; ret =SCX_CALL_OP_RET (sch, cgroup_init,NULL , css->cgroup, &args);
kernel/sched/ext/sub.c +2 -2
@@ -1361,7 +1361,7 @@ static s32 scx_cgroup_claim_subtree(struct scx_sched *sch) .bw_period_us = tg->scx.bw_period_us, .bw_quota_us = tg->scx.bw_quota_us, .bw_burst_us = tg->scx.bw_burst_us, - .sched_idle = tg->scx.idle, + .sched_idle = tg->scx.sched_idle, }; if (tg->scx.sched != parent ||
@@ -1465,7 +1465,7 @@ static void scx_cgroup_return_subtree(struct scx_sched *sch) .bw_period_us = tg->scx.bw_period_us, .bw_quota_us = tg->scx.bw_quota_us, .bw_burst_us = tg->scx.bw_burst_us, - .sched_idle = tg->scx.idle, + .sched_idle = tg->scx.sched_idle, };/* the first pass must have transferred everything */
Signature · 1 line
2.43.0
Re: [PATCH v3 2/2] sched_ext: rename tg->scx.idle to tg->scx.sched_idle
Hi,
Quoted 9 lines · Tao Cui
From: Tao Cui <[email protected]>
A bare "idle" reads as CPU idle state in
sched_ext(ops.update_idle(), idle cpumasks,scx_bpf_pick_idle_cpu()). cpu.idle is the cgroup analog of theSCHED_IDLEpolicy, so name thetask_groupfieldsched_idleto match scx_cgroup_init_args.sched_idle.Pure rename, no behavior change.
While testing this series, I noticed that scx_group_set_idle() delivers ops.cgroup_set_idle() on every successful cpu.idle write, even rewrites of the current value, although the kerneldoc says the callback is invoked on transitions. scx_group_set_weight() and scx_group_set_bandwidth() both skip unchanged values, so this seems like an oversight.
The fix is a one-line guard, but it reads the field renamed here, so I'll post it on top of this series once it lands.
Thanks,
Tao
Quoted 82 lines
Signed-off-by Tao Cui <[email protected]>--- include/linux/sched/ext.h | 2 +- kernel/sched/ext/ext.c | 8 ++++---- kernel/sched/ext/sub.c | 4 ++-- 3 files changed, 7 insertions(+), 7 deletions(-)include/linux/sched/ext.h +1
-1@@ -323,7 +323,7 @@ struct scx_task_group {u64 bw_period_us;u64 bw_quota_us;u64 bw_burst_us; -bool idle; +bool sched_idle;#endif };kernel/sched/ext/ext.c +4
-4@@ -4682,7 +4682,7 @@ void scx_tg_init(struct task_group *tg) tg->scx.weight =CGROUP_WEIGHT_DFL ; tg->scx.bw_period_us =default_bw_period_us (); tg->scx.bw_quota_us =RUNTIME_INF ; - tg->scx.idle = false; + tg->scx.sched_idle = false; }/** @@ -4765,7 +4765,7 @@ int scx_tg_online(struct task_group *tg) .bw_period_us = tg->scx.bw_period_us, .bw_quota_us = tg->scx.bw_quota_us, .bw_burst_us = tg->scx.bw_burst_us, - .sched_idle = tg->scx.idle }; + .sched_idle = tg->scx.sched_idle }; ret =SCX_CALL_OP_RET (sch, cgroup_init,NULL , tg->css.cgroup, &args);@@ -4935,7 +4935,7 @@ void scx_group_set_idle(struct task_group *tg, bool idle)SCX_CALL_OP (sch, cgroup_set_idle,NULL ,tg_cgrp (tg), idle);/* Update the task group's idle state */ - tg->scx.idle = idle; + tg->scx.sched_idle = idle;percpu_up_read (&scx_cgroup_ops_rwsem); }@@ -5186,7 +5186,7 @@ static int scx_cgroup_init(struct scx_sched *sch) .bw_period_us = tg->scx.bw_period_us, .bw_quota_us = tg->scx.bw_quota_us, .bw_burst_us = tg->scx.bw_burst_us, - .sched_idle = tg->scx.idle, + .sched_idle = tg->scx.sched_idle, }; ret =SCX_CALL_OP_RET (sch, cgroup_init,NULL , css->cgroup, &args);kernel/sched/ext/sub.c +2
-2@@ -1361,7 +1361,7 @@ static s32 scx_cgroup_claim_subtree(struct scx_sched *sch) .bw_period_us = tg->scx.bw_period_us, .bw_quota_us = tg->scx.bw_quota_us, .bw_burst_us = tg->scx.bw_burst_us, - .sched_idle = tg->scx.idle, + .sched_idle = tg->scx.sched_idle, }; if (tg->scx.sched != parent ||@@ -1465,7 +1465,7 @@ static void scx_cgroup_return_subtree(struct scx_sched *sch) .bw_period_us = tg->scx.bw_period_us, .bw_quota_us = tg->scx.bw_quota_us, .bw_burst_us = tg->scx.bw_burst_us, - .sched_idle = tg->scx.idle, + .sched_idle = tg->scx.sched_idle, };/* the first pass must have transferred everything */
Quoted 7 lines · Tao Cui
From: Tao Cui <[email protected]>
This carries the initial cpu.idle state of a cgroup through
ops.cgroup_init(), which was the only cpu controller knob whose initial value never reached the BPF scheduler. Patch 1 is the fix; patch 2 renamestg->scx.idletotg->scx.sched_idle, suggested in review, as a separate cleanup.
This looks good to me.
Thanks,
-Andrea
Quoted 27 lines
Changes since v2:
- rename the new field to
sched_idle, as a bare "idle" reads as CPU idle state insched_ext(Tejun)
- split thetg->scx.idlerename into 2/2 so 1/2 stays minimal for stable backport
- add the Fixes: tag (Andrea)
- regenerate on top of current linux-next, which also resolves the CI conflict reported against v2v2: https://lore.kernel.org/r/[email protected]
v1: https://lore.kernel.org/r/[email protected]Tao Cui (2):
sched_ext: pass the initial cpu.idle state inscx_cgroup_init_args
sched_ext: renametg->scx.idletotg->scx.sched_idleinclude/linux/sched/ext.h | 2 +- kernel/sched/ext/ext.c | 8 +++++--- kernel/sched/ext/internal.h | 3 +++ kernel/sched/ext/sub.c | 2 ++ 4 files changed, 11 insertions(+), 4 deletions(-)Signature · 1 line
2.43.0
Tejun Heo
reviewerQuoted 3 lines
Tao Cui (2):
sched_ext: pass the initial cpu.idle state inscx_cgroup_init_args
sched_ext: renametg->scx.idletotg->scx.sched_idle
Applied 1 to sched_ext/for-7.3-fixes and 2 to sched_ext/for-7.4, with Andrea's Reviewed-by from the cover added to 2 and the subjects capitalized.
Thanks.
--
tejun