Re: [PATCH 0/3 -v3] GFP_NOFAIL cleanups

[Michal Hocko]

On Thu 05-01-17 19:50:23, Tetsuo Handa wrote:
[...]
> Anyway, I suggest merging description update shown below into this series and
> getting confirmation from all existing __GFP_NOFAIL users. If all existing
> __GFP_NOFAIL users are OK with this series (in other words, informed the risk
> caused by this series), I'm also OK with this series.
> 
> --- a/include/linux/gfp.h
> +++ b/include/linux/gfp.h
> @@ -135,16 +135,24 @@
>   * __GFP_REPEAT: Try hard to allocate the memory, but the allocation attempt
>   *   _might_ fail.  This depends upon the particular VM implementation.
>   *
> - * __GFP_NOFAIL: The VM implementation _must_ retry infinitely: the caller
> - *   cannot handle allocation failures. New users should be evaluated carefully
> - *   (and the flag should be used only when there is no reasonable failure
> - *   policy) but it is definitely preferable to use the flag rather than
> - *   opencode endless loop around allocator.
> - *
> - * __GFP_NORETRY: The VM implementation must not retry indefinitely and will
> - *   return NULL when direct reclaim and memory compaction have failed to allow
> - *   the allocation to succeed.  The OOM killer is not called with the current
> - *   implementation.
> + * __GFP_NOFAIL: The VM implementation must not give up even after direct
> + *   reclaim and memory compaction have failed to allow the allocation to
> + *   succeed. Note that since the OOM killer is not called with the current
> + *   implementation when direct reclaim and memory compaction have failed to
> + *   allow the allocation to succeed unless __GFP_FS is also used (and some
> + *   other conditions are met), e.g. GFP_NOFS | __GFP_NOFAIL allocation has
> + *   possibility of lockup. To reduce the possibility of lockup, __GFP_HIGH is
> + *   implicitly granted by the current implementation if __GFP_NOFAIL is used.
> + *   New users of __GFP_NOFAIL should be evaluated carefully (and __GFP_NOFAIL
> + *   should be used only when there is no reasonable failure policy) but it is
> + *   definitely preferable to use __GFP_NOFAIL rather than opencode endless
> + *   loop around allocator, for a stall detection check inside allocator will
> + *   likely be able to emit possible lockup warnings unless __GFP_NOWARN is
> + *   also used.

This is both wrong and unnecessarily describing implementation details.
Non-failing allocation which must not give up can lockup pretty much by
definition. IMHO the current description is sufficient.

-- 
Michal Hocko
SUSE Labs