Re: [PATCH v5 5/7] sched/numa: scan PID-inactive VMAs for promotion

[sashiko-bot@kernel.org]

Thank you for your contribution! Sashiko AI review found 1 potential
issue to consider:
- [Medium] sched/numa: skipped VMA scan completion update in
  task_numa_work()
--

commit f65331bdcd950c59eb449c304cff02b0db3d6e66
Author: Gregory Price (Meta) <gourry@gourry.net>

sched/numa: scan PID-inactive VMAs for promotion

This commit allows memory tiering to scan PID-inactive VMAs with
promotion-only scans when placement sampling is not due, and tracks the
last completed placement scan sequence separately from promotion-only
scans.

> diff --git a/kernel/sched/fair.c b/kernel/sched/fair.c
> index 778ba0ed34fe8..279bf12b74b64 100644
> --- a/kernel/sched/fair.c
> +++ b/kernel/sched/fair.c
[ ... ]
> @@ -4506,8 +4519,14 @@ static void task_numa_work(struct callback_head *work)
>  			cond_resched();
>  		} while (end != vma->vm_end);
>  
> -		/* VMA scan is complete, do not scan until next sequence. */
> +		/*
> +		 * VMA scan is complete, do not scan until next sequence.
> +		 * A promotion-only scan did not cover top-tier folios, so it
> +		 * does not count towards the placement-scan starvation check.
> +		 */
>  		vma->numab_state->prev_scan_seq = mm->numa_scan_seq;
> +		if (placement_scan)
> +			vma->numab_state->prev_placement_scan_seq = mm->numa_scan_seq;
>  		vma->numab_state->promo_only = false;

[Severity: Medium]
Can task_numa_work() skip updating prev_placement_scan_seq, prev_scan_seq
and promo_only when the final chunk of a VMA exhausts the scan budget?

Inside the loop right above this block, task_numa_work() jumps to out when
pages <= 0 || virtpages <= 0 before checking while (end != vma->vm_end):

kernel/sched/fair.c:task_numa_work() {
	...
			if (nr_pte_updates)
				pages -= (end - start) >> PAGE_SHIFT;
			virtpages -= (end - start) >> PAGE_SHIFT;

			start = end;
			if (pages <= 0 || virtpages <= 0)
				goto out;

			cond_resched();
		} while (end != vma->vm_end);
	...
out:
	if (vma)
		mm->numa_scan_offset = start;
	...
}

When a VMA's size (or remaining unscanned range) is an exact multiple of
sysctl_numa_balancing_scan_size, the last chunk reaches end == vma->vm_end
with pages <= 0 and jumps to out.  That sets mm->numa_scan_offset to
vma->vm_end, so the next task_numa_work() call moves on to the next VMA
without ever running the completion block for the finished VMA.

Because prev_placement_scan_seq is never updated on such a VMA,
vma_needs_placement_scan() permanently returns true once mm->numa_scan_seq
passes prev_placement_scan_seq + get_nr_threads(current), preventing
promotion-only scans on PID-inactive VMAs.

-- 
Sashiko AI review · https://sashiko.dev/#/patchset/20261110150446.2452975-1-gourry@gourry.net?part=5