Skip to contentSkip to conversation

mm: slab: comment __GFP_ZERO case for kmem_cache_alloc

PATCH In review Alexander Aring 2 participants 5 messages Oct 11, 2022 – Nov 10, 2022 Memory management

lore
Share & tools
lore
#0 lore raw

PATCH v2 1/1 mm: slab: comment __GFP_ZERO case for kmem_cache_alloc

This patch will add a comment for the __GFP_ZERO flag case for kmem_cache_alloc(). As the current comment mentioned that the flags only matters if the cache has no available objects it's different for the __GFP_ZERO flag which will ensure that the returned object is always zeroed in any case.

I have the feeling I run into this question already two times if the user need to zero the object or not, but the user does not need to zero the object afterwards. However another use of __GFP_ZERO and only zero the object if the cache has no available objects would also make no sense.

---
changes since v2:
- don't make a separate sentence for except

 mm/slab.c | 3 ++-
 1 file changed, 2 insertions(+), 1 deletion(-)
mm/slab.c +2 -1
index 10e96137b44f..a86879f42f2e 100644
@@ -3482,7 +3482,8 @@ void *__kmem_cache_alloc_lru(struct kmem_cache *cachep, struct list_lru *lru,
  * @flags: See kmalloc().
  *
  * Allocate an object from this cache.  The flags are only relevant
- * if the cache has no available objects.
+ * if the cache has no available objects, except flag __GFP_ZERO which
+ * will zero the returned object.
  *
  * Return: pointer to the new object or %NULL in case of error
  */
Signature · 1 line
2.31.1
#1 lore raw
Quoted 11 lines · Alexander Aring

This patch will add a comment for the __GFP_ZERO flag case for kmem_cache_alloc(). As the current comment mentioned that the flags only matters if the cache has no available objects it's different for the __GFP_ZERO flag which will ensure that the returned object is always zeroed in any case.

I have the feeling I run into this question already two times if the user need to zero the object or not, but the user does not need to zero the object afterwards. However another use of __GFP_ZERO and only zero the object if the cache has no available objects would also make no sense.

Hmm, but even with the update, the comment is still rather misleading, no?
- can the caller know if the cache has available objects and thus the flags are irrelevant, in order to pass flags that are potentially wrong (if there were no objects)? Not really.
- even if cache has available objects, we'll always end up in slab_pre_alloc_hook doing might_alloc(flags) which will trigger warnings (given CONFIG_DEBUG_ATOMIC_SLEEP etc.) if the flags are inappropriate for given context. So they are still "relevant"

So maybe just delete the whole comment? slub.c doesn't have it, and if any such comment should exist for kmem_cache_alloc() and contain anything useful and not misleading, it should be probably in include/linux/slab.h anyway?

Quoted 24 lines

---
changes since v2:
- don't make a separate sentence for except

 mm/slab.c | 3 ++-
 1 file changed, 2 insertions(+), 1 deletion(-)
mm/slab.c +2 -1
index 10e96137b44f..a86879f42f2e 100644
@@ -3482,7 +3482,8 @@ void *__kmem_cache_alloc_lru(struct kmem_cache *cachep, struct list_lru *lru,
  * @flags: See kmalloc().
  *
  * Allocate an object from this cache.  The flags are only relevant
- * if the cache has no available objects.
+ * if the cache has no available objects, except flag __GFP_ZERO which
+ * will zero the returned object.
  *
  * Return: pointer to the new object or %NULL in case of error
  */
#2 lore raw

Hi,

Quoted 14 lines · Vlastimil Babka
Quoted 11 lines · Alexander Aring

This patch will add a comment for the __GFP_ZERO flag case for kmem_cache_alloc(). As the current comment mentioned that the flags only matters if the cache has no available objects it's different for the __GFP_ZERO flag which will ensure that the returned object is always zeroed in any case.

I have the feeling I run into this question already two times if the user need to zero the object or not, but the user does not need to zero the object afterwards. However another use of __GFP_ZERO and only zero the object if the cache has no available objects would also make no sense.

Quoted 4 lines

Hmm, but even with the update, the comment is still rather misleading, no?
- can the caller know if the cache has available objects and thus the flags are irrelevant, in order to pass flags that are potentially wrong (if there were no objects)? Not really.

No, the caller cannot know it and that's why __GFP_ZERO makes no sense if they matter only if the cache has no available objects.

Quoted 5 lines

- even if cache has available objects, we'll always end up in slab_pre_alloc_hook doing might_alloc(flags) which will trigger warnings (given CONFIG_DEBUG_ATOMIC_SLEEP etc.) if the flags are inappropriate for given context. So they are still "relevant"

yes, so they are _always_ relevant in the current implementation. Also as you said the user doesn't know when they become relevant or not..

Quoted 4 lines

So maybe just delete the whole comment? slub.c doesn't have it, and if any such comment should exist for kmem_cache_alloc() and contain anything useful and not misleading, it should be probably in include/linux/slab.h anyway?

ctags brought me there, but this isn't a real argument why it should not be in the header file...

I am not sure about deleting the whole comment as people have an vague idea about how kmem_cache works and still need to know for __GFP_ZERO that it will always zero the memory, but thinking again somebody will make the conclusion it does not make sense as the user doesn't know when objects are reused or allocated. Having that in mind and reading the current comment was making me do more investigations into the internal behaviour to figure out how it works regarding __GFP_ZERO.

- Alex

#3 lore raw
Quoted 42 lines · Alexander Aring

Hi,

Quoted 14 lines · Vlastimil Babka
Quoted 11 lines · Alexander Aring

This patch will add a comment for the __GFP_ZERO flag case for kmem_cache_alloc(). As the current comment mentioned that the flags only matters if the cache has no available objects it's different for the __GFP_ZERO flag which will ensure that the returned object is always zeroed in any case.

I have the feeling I run into this question already two times if the user need to zero the object or not, but the user does not need to zero the object afterwards. However another use of __GFP_ZERO and only zero the object if the cache has no available objects would also make no sense.

Quoted 4 lines

Hmm, but even with the update, the comment is still rather misleading, no?
- can the caller know if the cache has available objects and thus the flags are irrelevant, in order to pass flags that are potentially wrong (if there were no objects)? Not really.

No, the caller cannot know it and that's why __GFP_ZERO makes no sense if they matter only if the cache has no available objects.

Quoted 5 lines

- even if cache has available objects, we'll always end up in slab_pre_alloc_hook doing might_alloc(flags) which will trigger warnings (given CONFIG_DEBUG_ATOMIC_SLEEP etc.) if the flags are inappropriate for given context. So they are still "relevant"

yes, so they are _always_ relevant in the current implementation. Also as you said the user doesn't know when they become relevant or not..

Quoted 4 lines

So maybe just delete the whole comment? slub.c doesn't have it, and if any such comment should exist for kmem_cache_alloc() and contain anything useful and not misleading, it should be probably in include/linux/slab.h anyway?

ctags brought me there, but this isn't a real argument why it should not be in the header file...

Quoted 7 lines

I am not sure about deleting the whole comment as people have an vague idea about how kmem_cache works and still need to know for __GFP_ZERO that it will always zero the memory, but thinking again somebody will make the conclusion it does not make sense as the user doesn't know when objects are reused or allocated. Having that in mind and reading the current comment was making me do more investigations into the internal behaviour to figure out how it works regarding __GFP_ZERO.

So, I did the following, which IMHO resolves the misleading parts and also mentions __GFP_ZERO. Sounds OK?

https://git.kernel.org/pub/scm/linux/kernel/git/vbabka/slab.git/commit/?h=slab/for-6.2/cleanups&id=d6a3a7c3f65dfebcbc4872d5912d3465c8e8b051