scx_p2dq: Refactor vtime handling of migrating tasks #1833
Conversation
| */ | ||
| if ((p->flags & PF_KTHREAD) && !taskc->all_cpus && | ||
| if ((p->flags & PF_KTHREAD) && p->cpus_ptr == &p->cpus_mask && p->nr_cpus_allowed == nr_cpus && |
There was a problem hiding this comment.
This is purely to avoid looking up the task_ctx for affinitized kthreads.
| return; | ||
| } | ||
|
|
||
| // Give a slight boost to high priority tasks migrating across LLCs. |
There was a problem hiding this comment.
So in case of a cross-LLC migration we reset all the accumulated vtime budget? I'm wondering if we're penalizing too much tasks that are migrating.
Have you tried for example to change just the vtime_min with this new logic?
There was a problem hiding this comment.
Let me do some more testing on this. I was trying to avoid extra lookups of the previous llc_ctx and use some heuristic to give the right tasks enough of a bump. Perhaps in this case is is worth doing the check though.
| } | ||
|
|
||
| // Give a slight boost to high priority tasks migrating across LLCs. | ||
| if (p->scx.weight < 100) { |
There was a problem hiding this comment.
If I recall correctly, higher scx.weight means higher priority. If so, it should be p->scx.weight > 100.
There was a problem hiding this comment.
Also, it will be a good idea to document the idea behind this if-else block more in detail.
There was a problem hiding this comment.
If I recall correctly, higher
scx.weightmeans higher priority. If so, it should bep->scx.weight > 100.
Minor note: scx.weight > 100 in terms of scheduling weight.
There was a problem hiding this comment.
Going to go with the simple approach and just set the set the vtime to the current llc vtime.
hodgesds
force-pushed
the
p2dq-vtime-clamp
branch
from
5e32e10 to
e8fe6b5
Compare
Refactor vtime updating into a helper method and update the enqueue method to properly update vtime after selecting a CPU. Signed-off-by: Daniel Hodges <[email protected]>
hodgesds
force-pushed
the
p2dq-vtime-clamp
branch
from
e8fe6b5 to
778f1e5
Compare
When updating the vtime of a task moving across LLCs give tasks with a lower weight a boost in vtime. Refactor vtime updating into a helper method and update the enqueue method to properly update vtime after selecting a CPU.