Re: [External] Re: [PATCH 2/3] include/linux/gfp.h: use unsigned int in gfp_zone

[David Sterba]

On Mon, May 07, 2018 at 05:25:47PM -0700, Matthew Wilcox wrote:
> On Mon, May 07, 2018 at 11:25:01PM +0200, David Sterba wrote:
> > On Mon, May 07, 2018 at 11:44:10AM -0700, Matthew Wilcox wrote:
> > > But something like btrfs should almost certainly be using ~GFP_ZONEMASK.
> > 
> > Agreed, the direct use of __GFP_DMA32 was added in 3ba7ab220e8918176c6f
> > to substitute GFP_NOFS, so the allocation flags are less restrictive but
> > still acceptable for allocation from slab.
> > 
> > The requirement from btrfs is to avoid highmem, the 'must be acceptable
> > for slab' requirement is more MM internal and should have been hidden
> > under some opaque flag mask. There was no strong need for that at the
> > time.
> 
> The GFP flags encode a multiple of different requirements.  There's
> "What can the allocator do to free memory" and "what area of memory
> can the allocation come from".  btrfs doesn't actually want to
> allocate memory from ZONE_MOVABLE or ZONE_DMA either.  It's probably never
> been called with those particular flags set, but in the spirit of
> future-proofing btrfs, perhaps a patch like this is in order?
> 
> ---- >8 ----
> 
> Subject: btrfs: Allocate extents from ZONE_NORMAL
> From: Matthew Wilcox <mawilcox@xxxxxxxxxxxxx>
> 
> If anyone ever passes a GFP_DMA or GFP_MOVABLE allocation flag to
> allocate_extent_state, it will try to allocate memory from the wrong zone.
> We just want to allocate memory from ZONE_NORMAL, so use GFP_RECLAIM_MASK
> to get what we want.

Looks good to me.

> Signed-off-by: Matthew Wilcox <mawilcox@xxxxxxxxxxxxx>
> 
> diff --git a/fs/btrfs/extent_io.c b/fs/btrfs/extent_io.c
> index e99b329002cf..4e4a67b7b29d 100644
> --- a/fs/btrfs/extent_io.c
> +++ b/fs/btrfs/extent_io.c
> @@ -216,12 +216,7 @@ static struct extent_state *alloc_extent_state(gfp_t mask)
>  {
>  	struct extent_state *state;
>  
> -	/*
> -	 * The given mask might be not appropriate for the slab allocator,
> -	 * drop the unsupported bits
> -	 */
> -	mask &= ~(__GFP_DMA32|__GFP_HIGHMEM);

I've noticed there's GFP_SLAB_BUG_MASK that's basically open coded here,
but this would not filter out the placement flags.

> -	state = kmem_cache_alloc(extent_state_cache, mask);

I'd prefer some comment here, it's not obvious why the mask is used.

> +	state = kmem_cache_alloc(extent_state_cache, mask & GFP_RECLAIM_MASK);
>  	if (!state)
>  		return state;
>  	state->state = 0;