scx_p2dq: Acccount for non scx usage #1844
Conversation
hodgesds
requested review from
JakeHillion,
arighi,
etsal,
htejun,
likewhatevs and
multics69
hodgesds
changed the title
hodgesds
force-pushed
the
p2dq-non-scx-accounting
branch
from
87a73f8 to
bf60129
Compare
|
This assumes the CPU was running full throttle when not in scx, but IIUC that includes time spent in the idle scheduling class, so isn't this going to skew accounting? |
Hmmm good observation... let me think about this some more may be able to test if |
hodgesds
force-pushed
the
p2dq-non-scx-accounting
branch
2 times, most recently
from
45a0279 to
c98beb6
Compare
|
|
||
| UEI_DEFINE(uei); | ||
|
|
||
|
|
| if (!(cpuc = lookup_cpu_ctx(cpu))) | ||
| return; | ||
|
|
||
| if (is_idle_task(args->task)) { |
There was a problem hiding this comment.
Hm... I don't think ops.cpu_release() is called if we're switching to a scheduling class below SCHED_EXT. See: https://web.git.kernel.org/pub/scm/linux/kernel/git/torvalds/linux.git/tree/kernel/sched/ext.c#n3171
So this condition should be always false, unless I'm missing something...
| if (!(cpuc = lookup_cpu_ctx(cpu))) | ||
| return; | ||
|
|
||
| if (is_idle_task(args->task)) { |
There was a problem hiding this comment.
A better approach would be to use ops.update_idle() to track the idle transition.
There was a problem hiding this comment.
I completely missed that, let me try that!
Add accounting for non scx usage for LLC utilization. This should make load balancing decisions more accurate. Signed-off-by: Daniel Hodges <hodgesd@meta.com>
hodgesds
force-pushed
the
p2dq-non-scx-accounting
branch
from
c98beb6 to
039760e
Compare
Nit: s/Acccount/Account |
5 participants
Add accounting for non scx usage for LLC utilization. This should make load balancing decisions more accurate.