NVIDIA: SAUCE: sched/deadline: Skip DL server initialization on nohz_full CPUs#400
Closed
kobak2026 wants to merge 1 commit into
Closed
Conversation
kobak2026
changed the title
Collaborator
|
@kobak2026 What are the upstream plans for this patch? In general we should be pursuing upstream-first and then backporting to our kernels. |
Contributor
PR Validation ReportPatchscan ✅ No Missing FixesAll cherry-picked commits checked — no missing upstream fixes found. PR Lint ❌ Errors foundDetailsChecking 1 commits... Cherry-pick digest: ┌──────────────┬───────────────────────────────────────────────┬────────────┬─────────┬───────────────────────────┐ │ Local │ Referenced upstream / Patch subject │ Patch-ID │ Subject │ SoB chain │ ├──────────────┼───────────────────────────────────────────────┼────────────┼─────────┼───────────────────────────┤ │ f7c99286f2b7 │ [SAUCE] sched/deadline: skip dl server initia │ N/A │ N/A │ kobak │ └──────────────┴───────────────────────────────────────────────┴────────────┴─────────┴───────────────────────────┘ Lint: all checks passed. PR metadata: W: PR title missing [<branch>] prefix: "NVIDIA: SAUCE: sched/deadline: Skip DL server initialization on nohz_full CPUs" E: PR targets 24.04_linux-nvidia-6.17-next but body has no 'BugLink:' or 'LP:' https://bugs.launchpad.net/... line |
Collaborator
Author
@nvmochs okay, I will upstream today |
…full CPUs The DL scheduler server (fair_server) is initialized on every online CPU with a default runtime of 50ms/1000ms. On nohz_full isolated CPUs, the DL server's dl_task_timer hrtimer fires periodically, which triggers hrtimer_interrupt(). This re-activates the tick's sched_timer (via tick_nohz_handler returning HRTIMER_RESTART) before the IRQ-exit path can suppress it through tick_nohz_full_update_tick(), causing the scheduler tick to run at full HZ rate on isolated CPUs. This is a regression from v6.8 where the DL server did not exist and nohz_full CPUs had no spurious hrtimer activity. Measured with osnoise tracer on ARM64 (nohz_full=4-64): Before: IRQ=~1000/s SIRQ=~30/s on isolated CPUs After: IRQ=0 SIRQ=0 on isolated CPUs (100.00000%) Verified across 6 reboot cycles with the osnoise method (CPUs 0-60, 10s runtime) - all isolated CPUs consistently at 100.00000% CPU availability. Skip DL server initialization for nohz_full CPUs since they are isolated and do not need CFS bandwidth protection. Fixes: bd9bbc9 ("sched: Rework dl_server") Cc: Peter Zijlstra <peterz@infradead.org> Cc: Daniel Bristot de Oliveira <bristot@kernel.org> Cc: Juri Lelli <juri.lelli@redhat.com> Cc: Frederic Weisbecker <frederic@kernel.org> Signed-off-by: Koba Ko <kobak@nvidia.com>
kobak2026
force-pushed
the
fix/nohz-full-disable-dl-server
branch
from
cfd7e6c to
f7c9928
Compare
Collaborator
|
Discussed with Koba, this is still WIP, moving PR to draft mode for now. |
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
2 participants
Add this suggestion to a batch that can be applied as a single commit.This suggestion is invalid because no changes were made to the code.Suggestions cannot be applied while the pull request is closed.Suggestions cannot be applied while viewing a subset of changes.Only one suggestion per line can be applied in a batch.Add this suggestion to a batch that can be applied as a single commit.Applying suggestions on deleted lines is not supported.You must change the existing code in this line in order to create a valid suggestion.Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be applied on multi-line comments.Suggestions cannot be applied while the pull request is queued to merge.Suggestion cannot be applied right now. Please check back later.
[issue description]
The DL scheduler server (
fair_server) is initialized on every online CPU with the default 50ms/1000ms bandwidth. Onnohz_fullisolated CPUs, itsdl_task_timerhrtimer fires periodically and reactivates the scheduler tick before the IRQ-exit path can suppress it. The result is that isolated CPUs can run the scheduler tick at full HZ rate.This regresses from v6.8, where the DL server did not exist and
nohz_fullCPUs had no matching spurious hrtimer activity.[how to reproduce]
Use the osnoise testing procedure:
Observed before this fix:
Observed after this fix:
The validation was repeated across 6 reboot cycles.
[how to fix]
Skip DL server initialization for CPUs covered by
tick_nohz_full_cpu(cpu). These CPUs are isolated and do not need CFS bandwidth protection from the DL server.Fixes: bd9bbc9 ("sched: Rework dl_server")