Skip to contentSkip to conversation

mm/damon: fix a minor bug, clean code, and add a test

PATCH Applied SJ Park 2 participants 7 messages Oct 6

Subsystems (3)
lore
Share & tools
lore

SJ Park

author maintainer
#0 lore raw

cover mm/damon: fix a minor bug, clean code, and add a test

Patch 1 from Jaeyeon Lee fixes a minor overflow issue in DAMOS. Patch 2 from Gregory Price (Meta) cleans up the code for handling of hugetlb pages. Patch 3 from Liew Rui Yan adds a kunit test for a recently fixed bug.

Note to Andrew. This is a batched reposting of DAMON patches that I reviewed and asking for merge into mm.git. Since there are only two weeks before the expected opening of the next merge window, this will be the last mm.git merge requesting unurgent DAMON reposting until the next rc1.

Gregory Price (1):
mm/damon: use damon_get_monitor_folio() for hugetlb entries

Jaeyeon Lee (1):
mm/damon/ops-common: fix age_in_sec overflow on 32-bit

Liew Rui Yan (1):
mm/damon/tests/core-kunit: add test for unconditionally skipping the
last region

 mm/damon/ops-common.c       | 11 ++++++----
 mm/damon/tests/core-kunit.h | 42 +++++++++++++++++++++++++++++++++++++
 mm/damon/vaddr.c            | 10 +++++----
 3 files changed, 55 insertions(+), 8 deletions(-)

base-commit: 1b1d9187f46db1efb64533dab8b8d5f4f2af0eb5

Signature · 1 line
2.47.3
author maintainer
#1 lore raw

PATCH 1/3 mm/damon/ops-common: fix age_in_sec overflow on 32-bit

From: Jaeyeon Lee <[email protected]>

damon_hot_score() calculates the age of a region by multiplying r->age and the aggregation interval. On 32-bit architectures, unsigned long is 32 bits, so the multiplication overflows once the age exceeds about 4295 seconds. Then old regions are scored as if they were young.

Fix it by doing the calculation in 64-bit arithmetic using div_u64(). Also make age_in_sec u64, so the result is not truncated back to 32 bits.

The issue was discovered [1] by Sashiko.

[1] https://lore.kernel.org/[email protected]

Fixes 198f0f4c58b9 ("mm/damon/vaddr,paddr: support pageout prioritization")
Cc Andrew Morton <[email protected]>
Assisted-by LLM

---
Changes from v1
- v1: https://lore.kernel.org/[email protected]
- Collect R-b: from SJ.
- Rebase to the latest mm-new.

 mm/damon/ops-common.c | 5 +++--
 1 file changed, 3 insertions(+), 2 deletions(-)
mm/damon/ops-common.c +3 -2
index 77366f42b3e5..7a7429359dc8 100644
@@ -164,7 +164,7 @@ int damon_hot_score(struct damon_ctx *c, struct damon_region *r,
 			struct damos *s)
 {
 	int freq_subscore;
-	unsigned int age_in_sec;
+	u64 age_in_sec;
 	int age_in_log, age_subscore;
 	unsigned int freq_weight = s->quota.weight_nr_accesses;
 	unsigned int age_weight = s->quota.weight_age;
@@ -174,7 +174,8 @@ int damon_hot_score(struct damon_ctx *c, struct damon_region *r,
 			DAMON_MAX_SUBSCORE,
 			damon_nr_samples_per_aggr(&c->attrs));
 
-	age_in_sec = (unsigned long)r->age * c->attrs.aggr_interval / 1000000;
+	age_in_sec = div_u64((u64)r->age * c->attrs.aggr_interval,
+			     USEC_PER_SEC);
 	if (age_in_sec)
 		age_in_log = min_t(int, ilog2(age_in_sec) + 1,
 				DAMON_MAX_AGE_IN_LOG);
Signature · 1 line
2.47.3
#2 lore raw

Re: [PATCH 1/3] mm/damon/ops-common: fix age_in_sec overflow on 32-bit

Sashiko has reviewed this patch and found no issues. It looks great!
Quoted 13 lines

From: Jaeyeon Lee <[email protected]>

damon_hot_score() calculates the age of a region by multiplying r->age and the aggregation interval. On 32-bit architectures, unsigned long is 32 bits, so the multiplication overflows once the age exceeds about 4295 seconds. Then old regions are scored as if they were young.

Fix it by doing the calculation in 64-bit arithmetic using div_u64(). Also make age_in_sec u64, so the result is not truncated back to 32 bits.

The issue was discovered [1] by Sashiko.

Quoted 8 lines

[1] https://lore.kernel.org/[email protected]

Fixes 198f0f4c58b9 ("mm/damon/vaddr,paddr: support pageout prioritization")
Cc Andrew Morton <[email protected]>
Assisted-by LLM

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
author maintainer
#3 lore raw

PATCH 2/3 mm/damon: use damon_get_monitor_folio() for hugetlb entries

In reply to SJ Park

From: Gregory Price <[email protected]>

Hugetlb folio acquisition is open-coded in three places:
- damon_hugetlb_mkold()
- damon_young_hugetlb_entry()
- damon_va_probe_hugetlb_entry()

DAMON already has damon_get_monitor_folio(), which paddr uses to get monitored LRU and hugetlb folios and to filter them through damon_folio_acceptable(). These paths bypass that check.

Use that helper in these paths too, so that DAMON acquires hugetlb folios and applies its acceptance policy in one place.

No functional change intended. All callers reach these paths with a present hugetlb PTE under the page table lock, so the folio is always a mapped hugetlb folio and the helper always returns it.

---
Changes from v1
- v1: https://lore.kernel.org/[email protected]
- Collect R-b: from SJ.
- Rebase to the latest mm-new.

 mm/damon/ops-common.c |  6 ++++--
 mm/damon/vaddr.c      | 10 ++++++----
 2 files changed, 10 insertions(+), 6 deletions(-)
mm/damon/ops-common.c +4 -2
index 7a7429359dc8..2f3bf86a221b 100644
@@ -140,9 +140,11 @@ void damon_hugetlb_mkold(pte_t *pte, struct mm_struct *mm,
 {
 	bool referenced = false;
 	pte_t entry = huge_ptep_get(mm, addr, pte);
-	struct folio *folio = pfn_folio(pte_pfn(entry));
+	struct folio *folio;
 
-	folio_get(folio);
+	folio = damon_get_monitor_folio(pte_pfn(entry));
+	if (!folio)
+		return;
 
 	referenced = damon_hugetlb_ptep_mkold(pte, mm, vma, addr, &entry);
 	if (mmu_notifier_clear_young(mm, addr,
mm/damon/vaddr.c +6 -4
index d5dde97b3cd0..b549496ea8e2 100644
@@ -411,8 +411,9 @@ static int damon_young_hugetlb_entry(pte_t *pte, unsigned long hmask,
 	if (!pte_present(entry))
 		goto out;
 
-	folio = pfn_folio(pte_pfn(entry));
-	folio_get(folio);
+	folio = damon_get_monitor_folio(pte_pfn(entry));
+	if (!folio)
+		goto out;
 
 	if (pte_young(entry) || !folio_test_idle(folio) ||
 	    mmu_notifier_test_young(walk->mm, addr))
@@ -660,8 +661,9 @@ static int damon_va_probe_hugetlb_entry(pte_t *pte, unsigned long hmask,
 	if (!pte_present(entry))
 		goto out;
 
-	folio = pfn_folio(pte_pfn(entry));
-	folio_get(folio);
+	folio = damon_get_monitor_folio(pte_pfn(entry));
+	if (!folio)
+		goto out;
 	damon_va_probe_folio(priv->ctx, priv->r, folio, &entry, NULL,
 			walk->vma->vm_mm);
 	folio_put(folio);
Signature · 1 line
2.47.3
#4 lore raw

Re: [PATCH 2/3] mm/damon: use damon_get_monitor_folio() for hugetlb entries

Sashiko has reviewed this patch and found no issues. It looks great!
Quoted 14 lines

From: Gregory Price <[email protected]>

Hugetlb folio acquisition is open-coded in three places:
- damon_hugetlb_mkold()
- damon_young_hugetlb_entry()
- damon_va_probe_hugetlb_entry()

DAMON already has damon_get_monitor_folio(), which paddr uses to get monitored LRU and hugetlb folios and to filter them through damon_folio_acceptable(). These paths bypass that check.

Use that helper in these paths too, so that DAMON acquires hugetlb folios and applies its acceptance policy in one place.

Quoted 8 lines

No functional change intended. All callers reach these paths with a present hugetlb PTE under the page table lock, so the folio is always a mapped hugetlb folio and the helper always returns it.

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=2
author maintainer
#5 lore raw

PATCH 3/3 mm/damon/tests/core-kunit: add test for unconditionally skipping the last region

In reply to SJ Park

From: Liew Rui Yan <[email protected]>

Commit b3723b596b54 ("mm/damon/core: fix unconditionally skip last region") fixed a minor bug.

Add a KUnit test case to cover this behavior and prevent regressions in the future.

---
Changes from v3
- v3: https://lore.kernel.org/[email protected]
- Collect R-b: from SJ.
- Rebase to the latest mm-new.
Changes from v2:
- v2: https://lore.kernel.org/damon/[email protected]
- Change the commit message, mention the fix rather than the original
bug.
Changes from v1 RESEND:
- v1 RESEND: https://lore.kernel.org/damon/[email protected]
- Drop braces for single line, as suggested [1] in coding-style.
- Set min_region_sz initial value to power-of-2 value, '1'.
Changes from v1:
- v1: https://lore.kernel.org/damon/[email protected]
- Resend since the dependent patch is merged into mm-new, that would
help us getting AI review help.

 mm/damon/tests/core-kunit.h | 42 +++++++++++++++++++++++++++++++++++++
 1 file changed, 42 insertions(+)
mm/damon/tests/core-kunit.h +42 -0
index ef146ca2ae8a..3fbb4e4e36fa 100644
@@ -2439,6 +2439,47 @@ static void damos_test_esz_goal_temporal(struct kunit *test)
 	damon_destroy_ctx(ctx);
 }
 
+static void damon_test_last_region_quota_reset(struct kunit *test)
+{
+	struct damos s = {0};
+	struct damon_target *t;
+	struct damon_region *r1, *r2;
+	unsigned long min_region_sz = 1;
+	bool skipped;
+
+	t = damon_new_target();
+	if (!t)
+		kunit_skip(test, "target alloc fail");
+
+	r1 = damon_new_region(0, 100);
+	if (!r1) {
+		damon_free_target(t);
+		kunit_skip(test, "region 1 alloc fail");
+	}
+	damon_add_region(r1, t);
+
+	r2 = damon_new_region(100, 200);
+	if (!r2) {
+		damon_free_target(t);
+		kunit_skip(test, "region 2 alloc fail");
+	}
+	damon_add_region(r2, t);
+
+	s.quota.charge_target_from = t;
+	s.quota.charge_addr_from = r1->ar.end;
+
+	skipped = damos_skip_charged_region(t, r2, &s, min_region_sz);
+
+	/* 'r2' is not processed, it should not skip */
+	KUNIT_EXPECT_EQ(test, skipped, false);
+
+	/* 'r2' is last region, it should reset charge_{target,addr}_from */
+	KUNIT_EXPECT_PTR_EQ(test, s.quota.charge_target_from, NULL);
+	KUNIT_EXPECT_EQ(test, s.quota.charge_addr_from, 0);
+
+	damon_free_target(t);
+}
+
 static struct kunit_case damon_test_cases[] = {
 	KUNIT_CASE(damon_test_target),
 	KUNIT_CASE(damon_test_regions),
@@ -2482,6 +2523,7 @@ static struct kunit_case damon_test_cases[] = {
 	KUNIT_CASE(damon_test_walk_control_obsolete),
 	KUNIT_CASE(damon_test_rand),
 	KUNIT_CASE(damos_test_esz_goal_temporal),
+	KUNIT_CASE(damon_test_last_region_quota_reset),
 	{},
 };
 
Signature · 1 line
2.47.3
#6 lore raw

Re: [PATCH 3/3] mm/damon/tests/core-kunit: add test for unconditionally skipping the last region

Sashiko has reviewed this patch and found no issues. It looks great!
Quoted 11 lines

From: Liew Rui Yan <[email protected]>

Commit b3723b596b54 ("mm/damon/core: fix unconditionally skip last region") fixed a minor bug.

Add a KUnit test case to cover this behavior and prevent regressions in the future.

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=3