Re: [RFC PATCH 2/6] mm/damon/core: support probe_hits_wsum damos core filter

[sashiko-bot@kernel.org]

Thank you for your contribution! Sashiko AI review found 1 potential issue(s) to consider:
- [Medium] DAMOS_FILTER_TYPE_PROBE_HITS_WSUM uses raw, unscaled probe hits instead of a moving sum, leading to inaccurate filter evaluation when apply_interval is smaller than aggr_interval.
--

commit 75c1f2781e8de8f015329fce3e8a8d40e521f7de
Author: SJ Park <sj@kernel.org>

mm/damon/core: support probe_hits_wsum damos core filter

This commit implements the probe_hits_wsum DAMOS core filter support. It
updates damos_filter_for_ops, handles the range parameters during filter
commits, and calculates the weighted sum of probe hits for filter matching.

> diff --git a/mm/damon/core.c b/mm/damon/core.c
> index ce8c6f99106ed..ec1a1666aa9cc 100644
> --- a/mm/damon/core.c
> +++ b/mm/damon/core.c

[ ... ]

> @@ -2538,6 +2543,11 @@ static bool damos_filter_match(struct damon_ctx *ctx, struct damon_target *t,
>  		damon_split_region_at(t, r, end - r->ar.start);
>  		matched = true;
>  		break;
> +	case DAMOS_FILTER_TYPE_PROBE_HITS_WSUM:
> +		wsum = damon_probe_hits_wsum(r, false, ctx);
> +		matched = filter->range_min <= wsum &&
> +			wsum <= filter->range_max;

[Severity: Medium]
When a scheme's apply_interval_us is smaller than its aggr_interval_us, does
this evaluate incomplete probe hits?

Because the probe_hits array accumulates from zero over the aggregation
interval, evaluating the raw probe hits mid-interval yields artificially low
values. This differs from access pattern evaluations, which scale the count
based on the elapsed fraction of the aggregation interval (for example, using
damon_nr_accesses_mvsum()).

Could this cause regions to incorrectly fail or falsely match the filter
depending on when the scheme is evaluated within the aggregation interval?

> +		break;
>  	default:
>  		return false;
>  	}

-- 
Sashiko AI review · https://sashiko.dev/#/patchset/20260906210513.106895-1-sj@kernel.org?part=2