Re: [PATCH 4/5] mm: Introduce GFP_NOFAIL with the inclusion of __GFP_RECLAIM

[Michal Hocko <mhocko@suse.com>]

On Wed 24-07-24 20:55:43, Barry Song wrote:
> From: Barry Song <v-songbaohua@oppo.com>
> 
> Introducing GFP_NOFAIL and gradually increasing enforcement to prevent
> direct use of __GFP_NOFAIL which might be misused within non-sleepable
> contexts with GFP_ATOMIC and GFP_NOWAIT.

I do not think this makes sense without removing __GFP_NOFAIL and
changing existing users because a new flag will very likely not be used
and therefore it will not achieve the ultimate goal to remove a
potential abuse.

> Signed-off-by: Barry Song <v-songbaohua@oppo.com>
> ---
>  include/linux/gfp_types.h | 5 +++++
>  1 file changed, 5 insertions(+)
> 
> diff --git a/include/linux/gfp_types.h b/include/linux/gfp_types.h
> index 0dad2c7914be..1666db74f25c 100644
> --- a/include/linux/gfp_types.h
> +++ b/include/linux/gfp_types.h
> @@ -339,6 +339,10 @@ enum {
>   * recurse into the FS layer with a short explanation why. All allocation
>   * requests will inherit GFP_NOFS implicitly.
>   *
> + * %GFP_NOFAIL employs direct memory reclaim and continuously retries until
> + * successful memory allocation. It should never be used in contexts where
> + * sleeping is not allowed.
> + *
>   * %GFP_USER is for userspace allocations that also need to be directly
>   * accessibly by the kernel or hardware. It is typically used by hardware
>   * for buffers that are mapped to userspace (e.g. graphics) that hardware
> @@ -378,6 +382,7 @@ enum {
>  #define GFP_NOWAIT	(__GFP_KSWAPD_RECLAIM | __GFP_NOWARN)
>  #define GFP_NOIO	(__GFP_RECLAIM)
>  #define GFP_NOFS	(__GFP_RECLAIM | __GFP_IO)
> +#define GFP_NOFAIL	(__GFP_RECLAIM | __GFP_NOFAIL)
>  #define GFP_USER	(__GFP_RECLAIM | __GFP_IO | __GFP_FS | __GFP_HARDWALL)
>  #define GFP_DMA		__GFP_DMA
>  #define GFP_DMA32	__GFP_DMA32
> -- 
> 2.34.1

-- 
Michal Hocko
SUSE Labs