mm/zsmalloc: fix class per-fullness zspage counts

We always record_obj() to make handle points to object after obj_malloc(), so simplify the code by moving record_obj() into obj_malloc(). There should be no functional change.

Signed-off-by Chengming Zhou <[email protected]>

We always use insert_zspage() and remove_zspage() to update zspage's fullness location, which will account correctly.

But this special async free path use "splice" instead of remove_zspage(), so the per-fullness zspage count for ZS_INUSE_RATIO_0 won't decrease.

Fix it by decreasing when iterate over the zspage free list.

...

Signed-off-by Chengming Zhou <[email protected]>

+++ b/mm/zsmalloc.c

Quoted 14 lines · On Thu, 27 Jun 2024 15:59:58 +0800 Chengming Zhou <[email protected]>…

We always use insert_zspage() and remove_zspage() to update zspage's fullness location, which will account correctly.

But this special async free path use "splice" instead of remove_zspage(), so the per-fullness zspage count for ZS_INUSE_RATIO_0 won't decrease.

Fix it by decreasing when iterate over the zspage free list.

...

Signed-off-by Chengming Zhou <[email protected]>

+++ b/mm/zsmalloc.c

(quoted diff) +0 -0

@@ -1883,6 +1883,7 @@ static void async_free_zspage(struct work_struct *work)
 

mm/zsmalloc.c static void async_free_zspage(struct work_struct *work) · line 1889

Quoted 6 lines

mm/zsmalloc.c +1 -0

 		class = zspage_class(pool, zspage);
 		spin_lock(&class->lock);
+		class_stat_dec(class, ZS_INUSE_RATIO_0, 1);
 		__free_zspage(pool, class, zspage);
 		spin_unlock(&class->lock);
 	}

What are the runtime effects of this bug? Should we backport the fix into earlier kernels? And are we able to identify the appropriate
Fixes: target?

We always use insert_zspage() and remove_zspage() to update zspage's fullness location, which will account correctly.

But this special async free path use "splice" instead of remove_zspage(), so the per-fullness zspage count for ZS_INUSE_RATIO_0 won't decrease.

Fix it by decreasing when iterate over the zspage free list.

We can land this patch, if you insist/prefer, but it's some sort of a NOOP, essentially (in a sense that it has no visible effects).

Quoted 2 lines · Sergey Senozhatsky

We can land this patch, if you insist/prefer, but it's some sort of a NOOP, essentially (in a sense that it has no visible effects).

Forgot to mention, sorry. If we land this then I'll ask for a different commit message. "Fix" sounds alarming. It would be better to say something like "keep that stat counter accurate, but RATIO_O is never used anywhere".

Andrew, could you please help to change the subject as Sergey asked? Sorry, I should have noted these details in the changelog when I wrote this subject.

Quoted 3 lines · Chengming Zhou

Andrew, could you please help to change the subject as Sergey asked? Sorry, I should have noted these details in the changelog when I wrote this subject.

Chengming, can I ask you to resend these patches with a proper commit message?

Quoted 7 lines · Sergey Senozhatsky

Quoted 3 lines · Chengming Zhou

Andrew, could you please help to change the subject as Sergey asked? Sorry, I should have noted these details in the changelog when I wrote this subject.

Chengming, can I ask you to resend these patches with a proper commit message?

Of course, will update and send later.

Quoted 6 lines · Chengming Zhou

Quoted 5 lines · Sergey Senozhatsky

Quoted 3 lines · Chengming Zhou

Andrew, could you please help to change the subject as Sergey asked? Sorry, I should have noted these details in the changelog when I wrote this subject.

Quoted 4 lines

Quoted 2 lines

Chengming, can I ask you to resend these patches with a proper commit message?

Of course, will update and send later.

I just pulled mm/mm-unstable and ready to update, but find Andrew has already helped to change the subject and commit message as below, which is great enough! Thanks!