Re: [PATCH] mm: do not print backtraces on GFP_ATOMIC failures

[Andrew Morton]

On Tue, 21 Sep 2010 12:18:18 -0400 Rik van Riel <riel@xxxxxxxxxx> wrote:

> Atomic allocations cannot fall back to the page eviction code
> and are expected to fail.  In fact, in some network intensive
> workloads, it is common to experience hundreds of GFP_ATOMIC
> allocation failures.
> 
> Printing out a backtrace for every one of those expected
> allocation failures accomplishes nothing good. At multi-gigabit
> network speeds with jumbo frames, a burst of allocation failure
> backtraces could even slow down the system.
> 
> We're better off not printing out backtraces on GFP_ATOMIC
> allocation failures.
> 
> Signed-off-by: Rik van Riel <riel@xxxxxxxxxx>
> 
> diff --git a/include/linux/gfp.h b/include/linux/gfp.h
> index 975609c..5a0bddb 100644
> --- a/include/linux/gfp.h
> +++ b/include/linux/gfp.h
> @@ -72,7 +72,7 @@ struct vm_area_struct;
>  /* This equals 0, but use constants in case they ever change */
>  #define GFP_NOWAIT	(GFP_ATOMIC & ~__GFP_HIGH)
>  /* GFP_ATOMIC means both !wait (__GFP_WAIT not set) and use emergency pool */
> -#define GFP_ATOMIC	(__GFP_HIGH)
> +#define GFP_ATOMIC	(__GFP_HIGH | __GFP_NOWARN)
>  #define GFP_NOIO	(__GFP_WAIT)
>  #define GFP_NOFS	(__GFP_WAIT | __GFP_IO)
>  #define GFP_KERNEL	(__GFP_WAIT | __GFP_IO | __GFP_FS)

A much finer-tuned implementation would be to add __GFP_NOWARN just to
the networking call sites.  I asked about this in June and it got
nixed:

http://www.spinics.net/lists/netdev/msg131965.html
--
To unsubscribe from this list: send the line "unsubscribe linux-kernel" in
the body of a message to majordomo@xxxxxxxxxxxxxxx
More majordomo info at  http://vger.kernel.org/majordomo-info.html
Please read the FAQ at  http://www.tux.org/lkml/