[PATCH v2] mm, slub: emit the "free" trace report before freeing memory in kmem_cache_free()

[PATCH v2] mm, slub: emit the "free" trace report before freeing memory in kmem_cache_free()

[Yunfeng Ye]

After the memory is freed, it can be immediately allocated by other
CPUs, before the "free" trace report has been emitted. This causes
inaccurate traces.

For example, if the following sequence of events occurs:

    CPU 0                 CPU 1

  (1) alloc xxxxxx
  (2) free  xxxxxx
                         (3) alloc xxxxxx
                         (4) free  xxxxxx

Then they will be inaccurately reported via tracing, so that they appear
to have happened in this order:

    CPU 0                 CPU 1

  (1) alloc xxxxxx
                         (2) alloc xxxxxx
  (3) free  xxxxxx
                         (4) free  xxxxxx

This makes it look like CPU 1 somehow managed to allocate mmemory that
CPU 0 still had allocated for itself.

In order to avoid this, emit the "free xxxxxx" tracing report just
before the actual call to free the memory, instead of just after it.

Signed-off-by: Yunfeng Ye <yeyunfeng@huawei.com>
Reviewed-by: Vlastimil Babka <vbabka@suse.cz>
---
v1 -> v2:
 - Modify the description
 - Add "Reviewed-by"

 mm/slub.c | 2 +-
 1 file changed, 1 insertion(+), 1 deletion(-)

diff --git a/mm/slub.c b/mm/slub.c
index 432145d7b4ec..427e62034c3f 100644
--- a/mm/slub.c
+++ b/mm/slub.c
@@ -3526,8 +3526,8 @@ void kmem_cache_free(struct kmem_cache *s, void *x)
 	s = cache_from_obj(s, x);
 	if (!s)
 		return;
-	slab_free(s, virt_to_head_page(x), x, NULL, 1, _RET_IP_);
 	trace_kmem_cache_free(_RET_IP_, x, s->name);
+	slab_free(s, virt_to_head_page(x), x, NULL, 1, _RET_IP_);
 }
 EXPORT_SYMBOL(kmem_cache_free);

-- 
2.27.0

Re: [PATCH v2] mm, slub: emit the "free" trace report before freeing memory in kmem_cache_free()

[Tang Yizhou]

On 2021/11/2 19:43, Yunfeng Ye wrote:
> After the memory is freed, it can be immediately allocated by other
> CPUs, before the "free" trace report has been emitted. This causes
> inaccurate traces.
> 
> For example, if the following sequence of events occurs:
> 
>     CPU 0                 CPU 1
> 
>   (1) alloc xxxxxx
>   (2) free  xxxxxx
>                          (3) alloc xxxxxx
>                          (4) free  xxxxxx
> 
> Then they will be inaccurately reported via tracing, so that they appear
> to have happened in this order:
> 
>     CPU 0                 CPU 1
> 
>   (1) alloc xxxxxx
>                          (2) alloc xxxxxx
>   (3) free  xxxxxx
>                          (4) free  xxxxxx
> 
> This makes it look like CPU 1 somehow managed to allocate mmemory that
> CPU 0 still had allocated for itself.
> 
> In order to avoid this, emit the "free xxxxxx" tracing report just
> before the actual call to free the memory, instead of just after it.
> 
> Signed-off-by: Yunfeng Ye <yeyunfeng@huawei.com>
> Reviewed-by: Vlastimil Babka <vbabka@suse.cz>
> ---
> v1 -> v2:
>  - Modify the description
>  - Add "Reviewed-by"
> 
>  mm/slub.c | 2 +-
>  1 file changed, 1 insertion(+), 1 deletion(-)
> 
> diff --git a/mm/slub.c b/mm/slub.c
> index 432145d7b4ec..427e62034c3f 100644
> --- a/mm/slub.c
> +++ b/mm/slub.c
> @@ -3526,8 +3526,8 @@ void kmem_cache_free(struct kmem_cache *s, void *x)
>  	s = cache_from_obj(s, x);
>  	if (!s)
>  		return;
> -	slab_free(s, virt_to_head_page(x), x, NULL, 1, _RET_IP_);
>  	trace_kmem_cache_free(_RET_IP_, x, s->name);
> +	slab_free(s, virt_to_head_page(x), x, NULL, 1, _RET_IP_);
>  }

It seems that kmem_cache_free() in mm/slab.c has the same problem.
We can fix it. Thanks.

>  EXPORT_SYMBOL(kmem_cache_free);
> 

Re: [PATCH v2] mm, slub: emit the "free" trace report before freeing memory in kmem_cache_free()

[Vlastimil Babka]

On 11/2/21 14:53, Tang Yizhou wrote:
> On 2021/11/2 19:43, Yunfeng Ye wrote:
>> --- a/mm/slub.c
>> +++ b/mm/slub.c
>> @@ -3526,8 +3526,8 @@ void kmem_cache_free(struct kmem_cache *s, void *x)
>>  	s = cache_from_obj(s, x);
>>  	if (!s)
>>  		return;
>> -	slab_free(s, virt_to_head_page(x), x, NULL, 1, _RET_IP_);
>>  	trace_kmem_cache_free(_RET_IP_, x, s->name);
>> +	slab_free(s, virt_to_head_page(x), x, NULL, 1, _RET_IP_);
>>  }
> 
> It seems that kmem_cache_free() in mm/slab.c has the same problem.
> We can fix it. Thanks.

Doh, true. Should go best before the local_irq_save() there.
And also kmem_cache_free() in mm/slob.c.

Interestingly kfree() is already OK in all 3 implementations.

>>  EXPORT_SYMBOL(kmem_cache_free);
>> 
> 

Re: [PATCH v2] mm, slub: emit the "free" trace report before freeing memory in kmem_cache_free()

[John Hubbard]

On 11/2/21 04:43, Yunfeng Ye wrote:
> After the memory is freed, it can be immediately allocated by other
> CPUs, before the "free" trace report has been emitted. This causes
> inaccurate traces.
> 
> For example, if the following sequence of events occurs:
> 
>      CPU 0                 CPU 1
> 
>    (1) alloc xxxxxx
>    (2) free  xxxxxx
>                           (3) alloc xxxxxx
>                           (4) free  xxxxxx
> 
> Then they will be inaccurately reported via tracing, so that they appear
> to have happened in this order:
> 
>      CPU 0                 CPU 1
> 
>    (1) alloc xxxxxx
>                           (2) alloc xxxxxx
>    (3) free  xxxxxx
>                           (4) free  xxxxxx
> 
> This makes it look like CPU 1 somehow managed to allocate mmemory that


I see I created a typo for you, sorry about that: s/mmemory/memory/

But anyway, the wording looks good now. Please feel free to add:

Reviewed-by: John Hubbard <jhubbard@nvidia.com>


thanks,
-- 
John Hubbard
NVIDIA

> CPU 0 still had allocated for itself.
> 
> In order to avoid this, emit the "free xxxxxx" tracing report just
> before the actual call to free the memory, instead of just after it.
> 
> Signed-off-by: Yunfeng Ye <yeyunfeng@huawei.com>
> Reviewed-by: Vlastimil Babka <vbabka@suse.cz>
> ---
> v1 -> v2:
>   - Modify the description
>   - Add "Reviewed-by"
> 
>   mm/slub.c | 2 +-
>   1 file changed, 1 insertion(+), 1 deletion(-)
> 
> diff --git a/mm/slub.c b/mm/slub.c
> index 432145d7b4ec..427e62034c3f 100644
> --- a/mm/slub.c
> +++ b/mm/slub.c
> @@ -3526,8 +3526,8 @@ void kmem_cache_free(struct kmem_cache *s, void *x)
>   	s = cache_from_obj(s, x);
>   	if (!s)
>   		return;
> -	slab_free(s, virt_to_head_page(x), x, NULL, 1, _RET_IP_);
>   	trace_kmem_cache_free(_RET_IP_, x, s->name);
> +	slab_free(s, virt_to_head_page(x), x, NULL, 1, _RET_IP_);
>   }
>   EXPORT_SYMBOL(kmem_cache_free);
> 

Re: [PATCH v2] mm, slub: emit the "free" trace report before freeing memory in kmem_cache_free()

[Yunfeng Ye]

On 2021/11/2 22:39, Vlastimil Babka wrote:
> On 11/2/21 14:53, Tang Yizhou wrote:
>> On 2021/11/2 19:43, Yunfeng Ye wrote:
>>> --- a/mm/slub.c
>>> +++ b/mm/slub.c
>>> @@ -3526,8 +3526,8 @@ void kmem_cache_free(struct kmem_cache *s, void *x)
>>>  	s = cache_from_obj(s, x);
>>>  	if (!s)
>>>  		return;
>>> -	slab_free(s, virt_to_head_page(x), x, NULL, 1, _RET_IP_);
>>>  	trace_kmem_cache_free(_RET_IP_, x, s->name);
>>> +	slab_free(s, virt_to_head_page(x), x, NULL, 1, _RET_IP_);
>>>  }
>>
>> It seems that kmem_cache_free() in mm/slab.c has the same problem.
>> We can fix it. Thanks.
> 
> Doh, true. Should go best before the local_irq_save() there.
> And also kmem_cache_free() in mm/slob.c.
> 
Yes, I will fix the same problem together in the v3 patch.

Thanks.


> Interestingly kfree() is already OK in all 3 implementations.
> 
>>>  EXPORT_SYMBOL(kmem_cache_free);
>>>
>>
> 
> .
> 

Re: [PATCH v2] mm, slub: emit the "free" trace report before freeing memory in kmem_cache_free()

[Yunfeng Ye]

On 2021/11/3 2:37, John Hubbard wrote:
> On 11/2/21 04:43, Yunfeng Ye wrote:
>> After the memory is freed, it can be immediately allocated by other
>> CPUs, before the "free" trace report has been emitted. This causes
>> inaccurate traces.
>>
>> For example, if the following sequence of events occurs:
>>
>>      CPU 0                 CPU 1
>>
>>    (1) alloc xxxxxx
>>    (2) free  xxxxxx
>>                           (3) alloc xxxxxx
>>                           (4) free  xxxxxx
>>
>> Then they will be inaccurately reported via tracing, so that they appear
>> to have happened in this order:
>>
>>      CPU 0                 CPU 1
>>
>>    (1) alloc xxxxxx
>>                           (2) alloc xxxxxx
>>    (3) free  xxxxxx
>>                           (4) free  xxxxxx
>>
>> This makes it look like CPU 1 somehow managed to allocate mmemory that
> 
> 
> I see I created a typo for you, sorry about that: s/mmemory/memory/
> 
> But anyway, the wording looks good now. Please feel free to add:
> 
> Reviewed-by: John Hubbard <jhubbard@nvidia.com>
> 
Ok, I will fix the typo in the v3 patch.

Thanks.

> 
> thanks,