Re: [PATCH net-next v4 2/5] page_frag: unify gfp bits for order 3 page allocation

[Paolo Abeni <pabeni@redhat.com>]

On Tue, 2024-01-30 at 19:37 +0800, Yunsheng Lin wrote:
> Currently there seems to be three page frag implementions
> which all try to allocate order 3 page, if that fails, it
> then fail back to allocate order 0 page, and each of them
> all allow order 3 page allocation to fail under certain
> condition by using specific gfp bits.
> 
> The gfp bits for order 3 page allocation are different
> between different implementation, __GFP_NOMEMALLOC is
> or'd to forbid access to emergency reserves memory for
> __page_frag_cache_refill(), but it is not or'd in other
> implementions, __GFP_DIRECT_RECLAIM is masked off to avoid
> direct reclaim in skb_page_frag_refill(), but it is not
> masked off in __page_frag_cache_refill().
> 
> This patch unifies the gfp bits used between different
> implementions by or'ing __GFP_NOMEMALLOC and masking off
> __GFP_DIRECT_RECLAIM for order 3 page allocation to avoid
> possible pressure for mm.
> 
> Signed-off-by: Yunsheng Lin <linyunsheng@huawei.com>
> Reviewed-by: Alexander Duyck <alexanderduyck@fb.com>
> CC: Alexander Duyck <alexander.duyck@gmail.com>
> ---
>  drivers/vhost/net.c | 2 +-
>  mm/page_alloc.c     | 4 ++--
>  net/core/sock.c     | 2 +-
>  3 files changed, 4 insertions(+), 4 deletions(-)
> 
> diff --git a/drivers/vhost/net.c b/drivers/vhost/net.c
> index f2ed7167c848..e574e21cc0ca 100644
> --- a/drivers/vhost/net.c
> +++ b/drivers/vhost/net.c
> @@ -670,7 +670,7 @@ static bool vhost_net_page_frag_refill(struct vhost_net *net, unsigned int sz,
>  		/* Avoid direct reclaim but allow kswapd to wake */
>  		pfrag->page = alloc_pages((gfp & ~__GFP_DIRECT_RECLAIM) |
>  					  __GFP_COMP | __GFP_NOWARN |
> -					  __GFP_NORETRY,
> +					  __GFP_NORETRY | __GFP_NOMEMALLOC,
>  					  SKB_FRAG_PAGE_ORDER);

>  		if (likely(pfrag->page)) {
>  			pfrag->size = PAGE_SIZE << SKB_FRAG_PAGE_ORDER;
> diff --git a/mm/page_alloc.c b/mm/page_alloc.c
> index c0f7e67c4250..636145c29f70 100644
> --- a/mm/page_alloc.c
> +++ b/mm/page_alloc.c
> @@ -4685,8 +4685,8 @@ static struct page *__page_frag_cache_refill(struct page_frag_cache *nc,
>  	gfp_t gfp = gfp_mask;
>  
>  #if (PAGE_SIZE < PAGE_FRAG_CACHE_MAX_SIZE)
> -	gfp_mask |= __GFP_COMP | __GFP_NOWARN | __GFP_NORETRY |
> -		    __GFP_NOMEMALLOC;
> +	gfp_mask = (gfp_mask & ~__GFP_DIRECT_RECLAIM) |  __GFP_COMP |
> +		   __GFP_NOWARN | __GFP_NORETRY | __GFP_NOMEMALLOC;
>  	page = alloc_pages_node(NUMA_NO_NODE, gfp_mask,
>  				PAGE_FRAG_CACHE_MAX_ORDER);
>  	nc->size = page ? PAGE_FRAG_CACHE_MAX_SIZE : PAGE_SIZE;
> diff --git a/net/core/sock.c b/net/core/sock.c
> index 88bf810394a5..8289a3d8c375 100644
> --- a/net/core/sock.c
> +++ b/net/core/sock.c
> @@ -2919,7 +2919,7 @@ bool skb_page_frag_refill(unsigned int sz, struct page_frag *pfrag, gfp_t gfp)
>  		/* Avoid direct reclaim but allow kswapd to wake */
>  		pfrag->page = alloc_pages((gfp & ~__GFP_DIRECT_RECLAIM) |
>  					  __GFP_COMP | __GFP_NOWARN |
> -					  __GFP_NORETRY,
> +					  __GFP_NORETRY | __GFP_NOMEMALLOC,
>  					  SKB_FRAG_PAGE_ORDER);

This will prevent memory reserve usage when allocating order 3 pages,
but not when allocating a single page as a fallback. Still different
from the __page_frag_cache_refill() allocator - which never accesses
the memory reserves.

I'm unsure we want to propagate the __page_frag_cache_refill behavior
here, the current behavior could be required by some systems.

It looks like this series still leave the skb_page_frag_refill()
allocator alone, what about dropping this chunk, too? 

Thanks!

Paolo