Skip to contentSkip to conversation

mm/damon/core: add short circuit for temporal-goal schemes in damos_adjust_quota()

RFC PATCH In review Liew Rui Yan 3 participants 3 messages Sep 12

Subsystems (2)
lore
Share & tools
lore
#0 lore raw

RFC PATCH 1/1 mm/damon/core: add short circuit for temporal-goal schemes in damos_adjust_quota()

When using the temporal auto-tuning algorithm, the effective size quota becomes zero once the goal is [over-]achieved.

In this situation, damos_adjust_quota() still calculates quota->min_score for it. However, this min_score will not be used in this window, because in damon_do_apply_schemes(), damos_quota_is_full() will always returns true, preventing the scheme from being applied to any region.

Therefore, add a short circuit for temporal-goal algorithm schemes to early return from damos_adjust_quota() before calculating min_score.

---

I tested with virtme-ng + perf on a Proactive Memory Reclaim workload; within measurement noise (~3%), no measurable difference was observed.

The purpose of this patch is to reduce unnecessary operations (calculating min_score). But, I'd like to know if this patch needs to demonstrate that it provides better performance before it's merged.

---
 mm/damon/core.c | 2 ++
 1 file changed, 2 insertions(+)
mm/damon/core.c +2 -0
index ce8c6f99106e..fdcea989f0e0 100644
@@ -3320,6 +3320,8 @@ static void damos_adjust_quota(struct damon_ctx *c, struct damos *s)
 
 	if (!c->ops.get_scheme_score)
 		return;
+	if (quota->esz == 0)
+		return;
 
 	/* Fill up the score histogram */
 	memset(c->regions_score_histogram, 0,
Signature · 1 line
2.55.0
#1 lore raw
Sashiko has reviewed this patch and found no issues. It looks great!
Quoted 9 lines

When using the temporal auto-tuning algorithm, the effective size quota becomes zero once the goal is [over-]achieved.

In this situation, damos_adjust_quota() still calculates quota->min_score for it. However, this min_score will not be used in this window, because in damon_do_apply_schemes(), damos_quota_is_full() will always returns true, preventing the scheme from being applied to any region.

Quoted 4 lines

Therefore, add a short circuit for temporal-goal algorithm schemes to early return from damos_adjust_quota() before calculating min_score.

Sashiko has reviewed this patch and found no issues. It looks great!

Signature · 1 line
Sashiko AI review · https://sashiko.dev/#/patchset/[email protected]?part=1
maintainer
#2 lore raw
In reply to Liew Rui Yan
Quoted 11 lines · On Sat, 12 Sep 2026 20:43:12 +0800 Liew Rui Yan <[email protected]>…

When using the temporal auto-tuning algorithm, the effective size quota becomes zero once the goal is [over-]achieved.

In this situation, damos_adjust_quota() still calculates quota->min_score for it. However, this min_score will not be used in this window, because in damon_do_apply_schemes(), damos_quota_is_full() will always returns true, preventing the scheme from being applied to any region.

Therefore, add a short circuit for temporal-goal algorithm schemes to early return from damos_adjust_quota() before calculating min_score.

Nice catch. Makes sense to me.

Quoted 10 lines

---

I tested with virtme-ng + perf on a Proactive Memory Reclaim workload; within measurement noise (~3%), no measurable difference was observed.

The purpose of this patch is to reduce unnecessary operations (calculating min_score). But, I'd like to know if this patch needs to demonstrate that it provides better performance before it's merged.

This function is supposed to be not performance critical. I expect performance difference would be shown only in some setups that I didn't imagine.

But the change is small and makes sense. I wouldn't mind having no performance measurement for this small change.

Quoted 10 lines
---
 mm/damon/core.c | 2 ++
 1 file changed, 2 insertions(+)
mm/damon/core.c +0 -0
index ce8c6f99106e..fdcea989f0e0 100644

mm/damon/core.c static void damos_adjust_quota(struct damon_ctx *c, struct damos *s) · line 3324

Quoted 5 lines
mm/damon/core.c +2 -0
@@ -3320,6 +3320,8 @@ static void damos_adjust_quota(struct damon_ctx *c, struct damos *s)
 
 	if (!c->ops.get_scheme_score)
 		return;
+	if (quota->esz == 0)
+		return;

Direct zero esz comparison looks redundant and incomplete. It doesn't catch the case that <min_region_sz remaining quota case. Let's use damos_quota_is_full() instead. Also, let's do the check before the get_scheme_score check. If quota is already full, get_scheme_score check also makes no sense.

Quoted 5 lines
mm/damon/core.c +0 -0
 	/* Fill up the score histogram */
 	memset(c->regions_score_histogram, 0,
Signature · 1 line
2.55.0

Thanks,
SJ