* [PATCH v3 2/3] percpu: Do not trust hint starts when they are not set
2026-04-10 17:44 [PATCH v3 1/3] percpu: Fix wrong chunk hints update Joonwon Kang
@ 2026-04-10 17:44 ` Joonwon Kang
2026-04-21 23:28 ` Dennis Zhou
2026-04-10 17:44 ` [PATCH v3 3/3] percpu: Fix hint invariant breakage Joonwon Kang
2026-04-21 23:26 ` [PATCH v3 1/3] percpu: Fix wrong chunk hints update Dennis Zhou
2 siblings, 1 reply; 6+ messages in thread
From: Joonwon Kang @ 2026-04-10 17:44 UTC (permalink / raw)
To: dennis, tj, cl; +Cc: akpm, linux-mm, linux-kernel, dodam, Joonwon Kang
contig_hint_start can be trusted outside the hint update function since
it will be updated everytime contig_hint is broken. On the other hand,
scan_hint_start might still be invalid anywhere in the code due to the
broken scan_hint not being updated promptly. If those starts are trusted
when they are not set, it could lead to false invalidation or update of
the hints.
Signed-off-by: Joonwon Kang <joonwonkang@google.com>
---
v3: Initial version.
mm/percpu.c | 15 ++++++++++++---
1 file changed, 12 insertions(+), 3 deletions(-)
diff --git a/mm/percpu.c b/mm/percpu.c
index 3ecd86096641..f16533ed4a49 100644
--- a/mm/percpu.c
+++ b/mm/percpu.c
@@ -638,6 +638,13 @@ static void pcpu_block_update(struct pcpu_block_md *block, int start, int end)
if (end == block->nr_bits)
block->right_free = contig;
+ if (block->contig_hint == 0) {
+ block->contig_hint = contig;
+ block->contig_hint_start = start;
+ block->scan_hint = 0;
+ return;
+ }
+
if (contig > block->contig_hint) {
/* promote the old contig_hint to be the new scan_hint */
if (start > block->contig_hint_start) {
@@ -845,7 +852,8 @@ static void pcpu_block_update_hint_alloc(struct pcpu_chunk *chunk, int bit_off,
PCPU_BITMAP_BLOCK_BITS,
s_off + bits);
- if (pcpu_region_overlap(s_block->scan_hint_start,
+ if (s_block->scan_hint &&
+ pcpu_region_overlap(s_block->scan_hint_start,
s_block->scan_hint_start + s_block->scan_hint,
s_off,
s_off + bits))
@@ -889,7 +897,7 @@ static void pcpu_block_update_hint_alloc(struct pcpu_chunk *chunk, int bit_off,
/* reset the block */
e_block++;
} else {
- if (e_off > e_block->scan_hint_start)
+ if (e_block->scan_hint && e_off > e_block->scan_hint_start)
e_block->scan_hint = 0;
e_block->left_free = 0;
@@ -922,7 +930,8 @@ static void pcpu_block_update_hint_alloc(struct pcpu_chunk *chunk, int bit_off,
if (nr_empty_pages)
pcpu_update_empty_pages(chunk, -nr_empty_pages);
- if (pcpu_region_overlap(chunk_md->scan_hint_start,
+ if (chunk_md->scan_hint &&
+ pcpu_region_overlap(chunk_md->scan_hint_start,
chunk_md->scan_hint_start +
chunk_md->scan_hint,
bit_off,
--
2.53.0.1213.gd9a14994de-goog
^ permalink raw reply [flat|nested] 6+ messages in thread* Re: [PATCH v3 2/3] percpu: Do not trust hint starts when they are not set
2026-04-10 17:44 ` [PATCH v3 2/3] percpu: Do not trust hint starts when they are not set Joonwon Kang
@ 2026-04-21 23:28 ` Dennis Zhou
0 siblings, 0 replies; 6+ messages in thread
From: Dennis Zhou @ 2026-04-21 23:28 UTC (permalink / raw)
To: Joonwon Kang; +Cc: tj, cl, akpm, linux-mm, linux-kernel, dodam
On Fri, Apr 10, 2026 at 05:44:16PM +0000, Joonwon Kang wrote:
> contig_hint_start can be trusted outside the hint update function since
> it will be updated everytime contig_hint is broken. On the other hand,
> scan_hint_start might still be invalid anywhere in the code due to the
> broken scan_hint not being updated promptly. If those starts are trusted
> when they are not set, it could lead to false invalidation or update of
> the hints.
>
> Signed-off-by: Joonwon Kang <joonwonkang@google.com>
> ---
> v3: Initial version.
>
> mm/percpu.c | 15 ++++++++++++---
> 1 file changed, 12 insertions(+), 3 deletions(-)
>
> diff --git a/mm/percpu.c b/mm/percpu.c
> index 3ecd86096641..f16533ed4a49 100644
> --- a/mm/percpu.c
> +++ b/mm/percpu.c
> @@ -638,6 +638,13 @@ static void pcpu_block_update(struct pcpu_block_md *block, int start, int end)
> if (end == block->nr_bits)
> block->right_free = contig;
>
> + if (block->contig_hint == 0) {
> + block->contig_hint = contig;
> + block->contig_hint_start = start;
> + block->scan_hint = 0;
> + return;
> + }
> +
This change isn't described in the commit log. I'd prefer to clean up
the logic below than adding this additional if empty case.
> if (contig > block->contig_hint) {
> /* promote the old contig_hint to be the new scan_hint */
> if (start > block->contig_hint_start) {
> @@ -845,7 +852,8 @@ static void pcpu_block_update_hint_alloc(struct pcpu_chunk *chunk, int bit_off,
> PCPU_BITMAP_BLOCK_BITS,
> s_off + bits);
>
> - if (pcpu_region_overlap(s_block->scan_hint_start,
> + if (s_block->scan_hint &&
> + pcpu_region_overlap(s_block->scan_hint_start,
> s_block->scan_hint_start + s_block->scan_hint,
> s_off,
> s_off + bits))
> @@ -889,7 +897,7 @@ static void pcpu_block_update_hint_alloc(struct pcpu_chunk *chunk, int bit_off,
> /* reset the block */
> e_block++;
> } else {
> - if (e_off > e_block->scan_hint_start)
> + if (e_block->scan_hint && e_off > e_block->scan_hint_start)
> e_block->scan_hint = 0;
>
> e_block->left_free = 0;
> @@ -922,7 +930,8 @@ static void pcpu_block_update_hint_alloc(struct pcpu_chunk *chunk, int bit_off,
> if (nr_empty_pages)
> pcpu_update_empty_pages(chunk, -nr_empty_pages);
>
> - if (pcpu_region_overlap(chunk_md->scan_hint_start,
> + if (chunk_md->scan_hint &&
> + pcpu_region_overlap(chunk_md->scan_hint_start,
> chunk_md->scan_hint_start +
> chunk_md->scan_hint,
> bit_off,
> --
> 2.53.0.1213.gd9a14994de-goog
>
^ permalink raw reply [flat|nested] 6+ messages in thread
* [PATCH v3 3/3] percpu: Fix hint invariant breakage
2026-04-10 17:44 [PATCH v3 1/3] percpu: Fix wrong chunk hints update Joonwon Kang
2026-04-10 17:44 ` [PATCH v3 2/3] percpu: Do not trust hint starts when they are not set Joonwon Kang
@ 2026-04-10 17:44 ` Joonwon Kang
2026-04-22 0:04 ` Dennis Zhou
2026-04-21 23:26 ` [PATCH v3 1/3] percpu: Fix wrong chunk hints update Dennis Zhou
2 siblings, 1 reply; 6+ messages in thread
From: Joonwon Kang @ 2026-04-10 17:44 UTC (permalink / raw)
To: dennis, tj, cl; +Cc: akpm, linux-mm, linux-kernel, dodam, Joonwon Kang
The invariant "scan_hint_start > contig_hint_start if and only if
scan_hint == contig_hint" should be kept for hint management. However,
it could be broken in some cases:
- if (new contig == contig_hint == scan_hint) && (contig_hint_start <
scan_hint_start < new contig start) && the new contig is to become a
new contig_hint due to its better alignment, then scan_hint should
be invalidated instead of keeping the old value.
- if (new contig == contig_hint > scan_hint) && (new contig start <
contig_hint_start) && the new contig is not to become a new
contig_hint, then scan_hint should be not updated to the new contig.
This commit mainly fixes this invariant breakage and includes more:
- Refactor the percpu block update code to make it more visible on
what to consider, e.g. when the new contig overlaps with the old
contig_hint or scan_hint.
- Merge the new contig with other hints when it overlaps with them and
treat it as a whole free region instead of a separate small region.
- Fix the invariant breakage and also optimizes scan_hint further.
Some of the optimization cases when no overlap occurs are:
- if (new contig > contig_hint > scan_hint) && (scan_hint_start < new
contig start < contig_hint_start), then keep scan_hint instead of
invalidating it.
- if (new contig > contig_hint == scan_hint) && (contig_hint_start <
new contig start < scan_hint_start), then update scan_hint to the
old contig_hint instead of invalidating it.
- if (new contig == contig_hint > scan_hint) && (new contig start <
contig_hint_start) && the new contig is to become a new contig_hint
due to its better alignment, then update scan_hint to the old
contig_hint instead of invalidating or keeping it.
Signed-off-by: Joonwon Kang <joonwonkang@google.com>
---
v2 -> v3: Merge the new contig with other hints when it overlaps with
them and treat it as a whole free region instead of a separate small
region.
v1 -> v2: Consider cases where the new contig overlaps with the existing
contig_hint or scan_hint.
mm/percpu.c | 130 ++++++++++++++++++++++++++++++++++++----------------
1 file changed, 90 insertions(+), 40 deletions(-)
diff --git a/mm/percpu.c b/mm/percpu.c
index f16533ed4a49..d5b0b4863ffe 100644
--- a/mm/percpu.c
+++ b/mm/percpu.c
@@ -629,7 +629,27 @@ static inline bool pcpu_region_overlap(int a, int b, int x, int y)
*/
static void pcpu_block_update(struct pcpu_block_md *block, int start, int end)
{
- int contig = end - start;
+ int contig;
+ int scan_hint_cand_1 = 0;
+ int scan_hint_cand_1_start = 0;
+ int scan_hint_cand_2 = 0;
+ int scan_hint_cand_2_start = 0;
+ bool overlap_with_contig_hint = pcpu_region_overlap(start, end,
+ block->contig_hint_start,
+ block->contig_hint_start + block->contig_hint);
+ bool overlap_with_scan_hint = pcpu_region_overlap(start, end,
+ block->scan_hint_start,
+ block->scan_hint_start + block->scan_hint);
+
+ if (block->contig_hint && overlap_with_contig_hint) {
+ start = min(start, block->contig_hint_start);
+ end = max(end, block->contig_hint_start + block->contig_hint);
+ }
+ if (block->scan_hint && overlap_with_scan_hint) {
+ start = min(start, block->scan_hint_start);
+ end = max(end, block->scan_hint_start + block->scan_hint);
+ }
+ contig = end - start;
block->first_free = min(block->first_free, start);
if (start == 0)
@@ -646,56 +666,86 @@ static void pcpu_block_update(struct pcpu_block_md *block, int start, int end)
}
if (contig > block->contig_hint) {
- /* promote the old contig_hint to be the new scan_hint */
- if (start > block->contig_hint_start) {
- if (block->contig_hint > block->scan_hint) {
- block->scan_hint_start =
- block->contig_hint_start;
- block->scan_hint = block->contig_hint;
- } else if (start < block->scan_hint_start) {
- /*
- * The old contig_hint == scan_hint. But, the
- * new contig is larger so hold the invariant
- * scan_hint_start < contig_hint_start.
- */
- block->scan_hint = 0;
- }
- } else {
- block->scan_hint = 0;
+ if (!overlap_with_contig_hint) {
+ scan_hint_cand_1 = block->contig_hint;
+ scan_hint_cand_1_start = block->contig_hint_start;
}
- block->contig_hint_start = start;
+
block->contig_hint = contig;
+ block->contig_hint_start = start;
} else if (contig == block->contig_hint) {
if (block->contig_hint_start &&
(!start ||
__ffs(start) > __ffs(block->contig_hint_start))) {
- /* start has a better alignment so use it */
+ scan_hint_cand_1 = block->contig_hint;
+ scan_hint_cand_1_start = block->contig_hint_start;
+
+ /* Start has a better alignment so use it. */
block->contig_hint_start = start;
- if (start < block->scan_hint_start &&
- block->contig_hint > block->scan_hint)
- block->scan_hint = 0;
- } else if (start > block->scan_hint_start ||
- block->contig_hint > block->scan_hint) {
- /*
- * Knowing contig == contig_hint, update the scan_hint
- * if it is farther than or larger than the current
- * scan_hint.
- */
- block->scan_hint_start = start;
- block->scan_hint = contig;
+ } else {
+ if (!overlap_with_contig_hint) {
+ scan_hint_cand_1 = contig;
+ scan_hint_cand_1_start = start;
+ }
}
} else {
/*
- * The region is smaller than the contig_hint. So only update
- * the scan_hint if it is larger than or equal and farther than
- * the current scan_hint.
+ * Consider only when the new contig is larger than or equal to
+ * the old scan hint.
*/
- if ((start < block->contig_hint_start &&
- (contig > block->scan_hint ||
- (contig == block->scan_hint &&
- start > block->scan_hint_start)))) {
- block->scan_hint_start = start;
- block->scan_hint = contig;
+ if (contig >= block->scan_hint) {
+ scan_hint_cand_1 = contig;
+ scan_hint_cand_1_start = start;
+ }
+ }
+
+ if (block->scan_hint &&
+ !pcpu_region_overlap(start, end, block->scan_hint_start,
+ block->scan_hint_start + block->scan_hint)) {
+ scan_hint_cand_2 = block->scan_hint;
+ scan_hint_cand_2_start = block->scan_hint_start;
+ }
+
+ /* Make scan_hint_cand_1 be the best candidate for the new scan hint. */
+ if ((scan_hint_cand_2 > scan_hint_cand_1) ||
+ (scan_hint_cand_2 == scan_hint_cand_1 &&
+ scan_hint_cand_2_start > scan_hint_cand_1_start)) {
+ int tmp_hint = scan_hint_cand_1;
+ int tmp_hint_start = scan_hint_cand_1_start;
+
+ scan_hint_cand_1 = scan_hint_cand_2;
+ scan_hint_cand_1_start = scan_hint_cand_2_start;
+ scan_hint_cand_2 = tmp_hint;
+ scan_hint_cand_2_start = tmp_hint_start;
+ }
+
+ /*
+ * At this point, it is guaranteed that none of the scan hint
+ * candidates overlaps with the new contig hint while they may overlap
+ * with the old scan hint, and that the first candidate is larger in
+ * size or, it equal, farther than the second one.
+ */
+
+ if (block->contig_hint > scan_hint_cand_1) {
+ if (scan_hint_cand_1_start < block->contig_hint_start) {
+ block->scan_hint = scan_hint_cand_1;
+ block->scan_hint_start = scan_hint_cand_1_start;
+ } else if (scan_hint_cand_2_start < block->contig_hint_start) {
+ block->scan_hint = scan_hint_cand_2;
+ block->scan_hint_start = scan_hint_cand_2_start;
+ } else {
+ block->scan_hint = 0;
+ }
+ } else if (block->contig_hint == scan_hint_cand_1) {
+ if (scan_hint_cand_1_start > block->contig_hint_start) {
+ block->scan_hint = scan_hint_cand_1;
+ block->scan_hint_start = scan_hint_cand_1_start;
+ } else if (scan_hint_cand_2 < block->contig_hint &&
+ scan_hint_cand_2_start < scan_hint_cand_1_start) {
+ block->scan_hint = scan_hint_cand_2;
+ block->scan_hint_start = scan_hint_cand_2_start;
+ } else {
+ block->scan_hint = 0;
}
}
}
--
2.53.0.1213.gd9a14994de-goog
^ permalink raw reply [flat|nested] 6+ messages in thread* Re: [PATCH v3 3/3] percpu: Fix hint invariant breakage
2026-04-10 17:44 ` [PATCH v3 3/3] percpu: Fix hint invariant breakage Joonwon Kang
@ 2026-04-22 0:04 ` Dennis Zhou
0 siblings, 0 replies; 6+ messages in thread
From: Dennis Zhou @ 2026-04-22 0:04 UTC (permalink / raw)
To: Joonwon Kang; +Cc: tj, cl, akpm, linux-mm, linux-kernel, dodam
On Fri, Apr 10, 2026 at 05:44:17PM +0000, Joonwon Kang wrote:
> The invariant "scan_hint_start > contig_hint_start if and only if
> scan_hint == contig_hint" should be kept for hint management. However,
> it could be broken in some cases:
>
> - if (new contig == contig_hint == scan_hint) && (contig_hint_start <
> scan_hint_start < new contig start) && the new contig is to become a
> new contig_hint due to its better alignment, then scan_hint should
> be invalidated instead of keeping the old value.
>
> - if (new contig == contig_hint > scan_hint) && (new contig start <
> contig_hint_start) && the new contig is not to become a new
> contig_hint, then scan_hint should be not updated to the new contig.
>
> This commit mainly fixes this invariant breakage and includes more:
>
> - Refactor the percpu block update code to make it more visible on
> what to consider, e.g. when the new contig overlaps with the old
> contig_hint or scan_hint.
>
Could you please split this up between the fixing the scan_hint issues
and then refactoring. That way this commit isn't doing too much.
> - Merge the new contig with other hints when it overlaps with them and
> treat it as a whole free region instead of a separate small region.
>
> - Fix the invariant breakage and also optimizes scan_hint further.
> Some of the optimization cases when no overlap occurs are:
>
> - if (new contig > contig_hint > scan_hint) && (scan_hint_start < new
> contig start < contig_hint_start), then keep scan_hint instead of
> invalidating it.
>
> - if (new contig > contig_hint == scan_hint) && (contig_hint_start <
> new contig start < scan_hint_start), then update scan_hint to the
> old contig_hint instead of invalidating it.
>
> - if (new contig == contig_hint > scan_hint) && (new contig start <
> contig_hint_start) && the new contig is to become a new contig_hint
> due to its better alignment, then update scan_hint to the old
> contig_hint instead of invalidating or keeping it.
>
> Signed-off-by: Joonwon Kang <joonwonkang@google.com>
> ---
> v2 -> v3: Merge the new contig with other hints when it overlaps with
> them and treat it as a whole free region instead of a separate small
> region.
>
> v1 -> v2: Consider cases where the new contig overlaps with the existing
> contig_hint or scan_hint.
>
> mm/percpu.c | 130 ++++++++++++++++++++++++++++++++++++----------------
> 1 file changed, 90 insertions(+), 40 deletions(-)
>
> diff --git a/mm/percpu.c b/mm/percpu.c
> index f16533ed4a49..d5b0b4863ffe 100644
> --- a/mm/percpu.c
> +++ b/mm/percpu.c
> @@ -629,7 +629,27 @@ static inline bool pcpu_region_overlap(int a, int b, int x, int y)
> */
> static void pcpu_block_update(struct pcpu_block_md *block, int start, int end)
> {
> - int contig = end - start;
> + int contig;
> + int scan_hint_cand_1 = 0;
> + int scan_hint_cand_1_start = 0;
> + int scan_hint_cand_2 = 0;
> + int scan_hint_cand_2_start = 0;
I'm not really a fan of this cand_1 and cand_2. Re-reading all this
code. I think I really should have introduced something like:
struct pcpu_region {
int start;
int size;
}
It's really easy to mix contig_hint and contig_hint_start. If you don't
mind, it would be ideal if we could introduce it but it might be a
non-trivial amount of refactor.
I'd say the shortest path forward would be to do just to fix the
scan_hint issues.
> + bool overlap_with_contig_hint = pcpu_region_overlap(start, end,
> + block->contig_hint_start,
> + block->contig_hint_start + block->contig_hint);
> + bool overlap_with_scan_hint = pcpu_region_overlap(start, end,
> + block->scan_hint_start,
> + block->scan_hint_start + block->scan_hint);
> +
This one isn't used again so we should probably just inline it.
> + if (block->contig_hint && overlap_with_contig_hint) {
> + start = min(start, block->contig_hint_start);
> + end = max(end, block->contig_hint_start + block->contig_hint);
> + }
> + if (block->scan_hint && overlap_with_scan_hint) {
> + start = min(start, block->scan_hint_start);
> + end = max(end, block->scan_hint_start + block->scan_hint);
> + }
> + contig = end - start;
>
> block->first_free = min(block->first_free, start);
> if (start == 0)
> @@ -646,56 +666,86 @@ static void pcpu_block_update(struct pcpu_block_md *block, int start, int end)
> }
>
> if (contig > block->contig_hint) {
> - /* promote the old contig_hint to be the new scan_hint */
> - if (start > block->contig_hint_start) {
> - if (block->contig_hint > block->scan_hint) {
> - block->scan_hint_start =
> - block->contig_hint_start;
> - block->scan_hint = block->contig_hint;
> - } else if (start < block->scan_hint_start) {
> - /*
> - * The old contig_hint == scan_hint. But, the
> - * new contig is larger so hold the invariant
> - * scan_hint_start < contig_hint_start.
> - */
> - block->scan_hint = 0;
> - }
> - } else {
> - block->scan_hint = 0;
> + if (!overlap_with_contig_hint) {
> + scan_hint_cand_1 = block->contig_hint;
> + scan_hint_cand_1_start = block->contig_hint_start;
> }
> - block->contig_hint_start = start;
> +
> block->contig_hint = contig;
> + block->contig_hint_start = start;
> } else if (contig == block->contig_hint) {
> if (block->contig_hint_start &&
> (!start ||
> __ffs(start) > __ffs(block->contig_hint_start))) {
> - /* start has a better alignment so use it */
> + scan_hint_cand_1 = block->contig_hint;
> + scan_hint_cand_1_start = block->contig_hint_start;
> +
> + /* Start has a better alignment so use it. */
> block->contig_hint_start = start;
> - if (start < block->scan_hint_start &&
> - block->contig_hint > block->scan_hint)
> - block->scan_hint = 0;
> - } else if (start > block->scan_hint_start ||
> - block->contig_hint > block->scan_hint) {
> - /*
> - * Knowing contig == contig_hint, update the scan_hint
> - * if it is farther than or larger than the current
> - * scan_hint.
> - */
> - block->scan_hint_start = start;
> - block->scan_hint = contig;
> + } else {
> + if (!overlap_with_contig_hint) {
> + scan_hint_cand_1 = contig;
> + scan_hint_cand_1_start = start;
> + }
> }
> } else {
> /*
> - * The region is smaller than the contig_hint. So only update
> - * the scan_hint if it is larger than or equal and farther than
> - * the current scan_hint.
> + * Consider only when the new contig is larger than or equal to
> + * the old scan hint.
> */
> - if ((start < block->contig_hint_start &&
> - (contig > block->scan_hint ||
> - (contig == block->scan_hint &&
> - start > block->scan_hint_start)))) {
> - block->scan_hint_start = start;
> - block->scan_hint = contig;
> + if (contig >= block->scan_hint) {
> + scan_hint_cand_1 = contig;
> + scan_hint_cand_1_start = start;
> + }
> + }
> +
> + if (block->scan_hint &&
> + !pcpu_region_overlap(start, end, block->scan_hint_start,
> + block->scan_hint_start + block->scan_hint)) {
> + scan_hint_cand_2 = block->scan_hint;
> + scan_hint_cand_2_start = block->scan_hint_start;
> + }
> +
> + /* Make scan_hint_cand_1 be the best candidate for the new scan hint. */
> + if ((scan_hint_cand_2 > scan_hint_cand_1) ||
> + (scan_hint_cand_2 == scan_hint_cand_1 &&
> + scan_hint_cand_2_start > scan_hint_cand_1_start)) {
> + int tmp_hint = scan_hint_cand_1;
> + int tmp_hint_start = scan_hint_cand_1_start;
> +
> + scan_hint_cand_1 = scan_hint_cand_2;
> + scan_hint_cand_1_start = scan_hint_cand_2_start;
> + scan_hint_cand_2 = tmp_hint;
> + scan_hint_cand_2_start = tmp_hint_start;
> + }
> +
> + /*
> + * At this point, it is guaranteed that none of the scan hint
> + * candidates overlaps with the new contig hint while they may overlap
> + * with the old scan hint, and that the first candidate is larger in
> + * size or, it equal, farther than the second one.
> + */
> +
> + if (block->contig_hint > scan_hint_cand_1) {
> + if (scan_hint_cand_1_start < block->contig_hint_start) {
> + block->scan_hint = scan_hint_cand_1;
> + block->scan_hint_start = scan_hint_cand_1_start;
> + } else if (scan_hint_cand_2_start < block->contig_hint_start) {
> + block->scan_hint = scan_hint_cand_2;
> + block->scan_hint_start = scan_hint_cand_2_start;
> + } else {
> + block->scan_hint = 0;
> + }
> + } else if (block->contig_hint == scan_hint_cand_1) {
> + if (scan_hint_cand_1_start > block->contig_hint_start) {
> + block->scan_hint = scan_hint_cand_1;
> + block->scan_hint_start = scan_hint_cand_1_start;
> + } else if (scan_hint_cand_2 < block->contig_hint &&
> + scan_hint_cand_2_start < scan_hint_cand_1_start) {
> + block->scan_hint = scan_hint_cand_2;
> + block->scan_hint_start = scan_hint_cand_2_start;
> + } else {
> + block->scan_hint = 0;
> }
This seems easier to invert and do something like this:
Probably something like:
bool overlap_contig_hint = pcpu_region_overlap(new_region, scan_hint);
bool overlap_scan_hint = pcpu_region_overlap(new_region, scan_hint);
if (overlap_scan_hint) {
scan_hint = 0;
}
if (overlap_contig_hint) {
contig_hint = new_region;
return;
}
// Now the new_region is distinct from the contig_hint and then can do
// the standard logic here.
if (new_region > contig_hint) { }
else if (new_region == contig_hint) {}
else {}
What do you think? I think makes cand_1 and cand_2 not necessary.
> }
> }
> --
> 2.53.0.1213.gd9a14994de-goog
>
Thanks,
Dennis
^ permalink raw reply [flat|nested] 6+ messages in thread
* Re: [PATCH v3 1/3] percpu: Fix wrong chunk hints update
2026-04-10 17:44 [PATCH v3 1/3] percpu: Fix wrong chunk hints update Joonwon Kang
2026-04-10 17:44 ` [PATCH v3 2/3] percpu: Do not trust hint starts when they are not set Joonwon Kang
2026-04-10 17:44 ` [PATCH v3 3/3] percpu: Fix hint invariant breakage Joonwon Kang
@ 2026-04-21 23:26 ` Dennis Zhou
2 siblings, 0 replies; 6+ messages in thread
From: Dennis Zhou @ 2026-04-21 23:26 UTC (permalink / raw)
To: Joonwon Kang; +Cc: dennis, tj, cl, akpm, linux-mm, linux-kernel, dodam
Hello,
On Fri, Apr 10, 2026 at 05:44:15PM +0000, Joonwon Kang wrote:
> Chunk end offset was set to a block end offset, which could prevent
> chunk hints from being updated correctly. It was observed that the chunk
> free size gets minus or shorter than the actual free size due to this.
> This commit fixes it.
>
> Signed-off-by: Joonwon Kang <joonwonkang@google.com>
> ---
> v3: Initial version.
>
> mm/percpu.c | 2 +-
> 1 file changed, 1 insertion(+), 1 deletion(-)
>
> diff --git a/mm/percpu.c b/mm/percpu.c
> index 81462ce5866e..3ecd86096641 100644
> --- a/mm/percpu.c
> +++ b/mm/percpu.c
> @@ -1054,7 +1054,7 @@ static void pcpu_block_update_hint_free(struct pcpu_chunk *chunk, int bit_off,
> else
> pcpu_block_update(&chunk->chunk_md,
> pcpu_block_off_to_off(s_index, start),
> - end);
> + pcpu_block_off_to_off(e_index, end));
> }
This needs a fixes tag for:
Fixes: 92c14cab4326 ("percpu: convert chunk hints to be based on pcpu_block_md")
Reviewed-by: Dennis Zhou <dennis@kernel.org>
Thanks,
Dennis
>
> /**
> --
> 2.53.0.1213.gd9a14994de-goog
>
^ permalink raw reply [flat|nested] 6+ messages in thread