Re: [PATCH] mm, slub: place the trace before freeing memory in kmem_cache_free()

From: Vlastimil Babka
Date: Tue Nov 02 2021 - 04:41:54 EST


On 11/2/21 08:03, John Hubbard wrote:
> On 10/30/21 03:11, Yunfeng Ye wrote:
>> After the memory is freed, it may be allocated by other CPUs and has
>> been recorded by trace. So the timing sequence of the memory tracing is
>> inaccurate.
>>
>> For example, we expect the following timing sequeuce:
>>
>>      CPU 0                 CPU 1
>>
>>    (1) alloc xxxxxx
>>    (2) free  xxxxxx
>>                           (3) alloc xxxxxx
>>                           (4) free  xxxxxx
>>
>> However, the following timing sequence may occur:
>>
>>      CPU 0                 CPU 1
>>
>>    (1) alloc xxxxxx
>>                           (2) alloc xxxxxx
>>    (3) free  xxxxxx
>>                           (4) free  xxxxxx
>>
>> So place the trace before freeing memory in kmem_cache_free().
>
> Hi Yunfeng,
>
> Like Muchun, I had some difficulty with the above description, but
> now I think I get it. :)
>
> In order to make it easier for others, how about this wording and subject
> line, instead:
>
>
> mm, slub: emit the "free" trace report before freeing memory in
> kmem_cache_free()
>
> 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. This makes it look like CPU 1
> somehow managed to allocate mmemory that CPU 0 still had allocated for
> itself:
>
>     CPU 0                 CPU 1
>
>   (1) alloc xxxxxx
>                          (2) alloc xxxxxx
>   (3) free  xxxxxx
>                          (4) free  xxxxxx
>
> 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.

Agree, this wording is better.
IIRC the same problem was fixed for mmap_lock trace ordering just recently.

>> Signed-off-by: Yunfeng Ye <yeyunfeng@xxxxxxxxxx>

Reviewed-by: Vlastimil Babka <vbabka@xxxxxxx>