Re: [PATCH net-next 1/2] net: veth: add page_pool for page recycling

[Lorenzo Bianconi <lorenzo@kernel.org>]

[-- Attachment #1: Type: text/plain, Size: 1826 bytes --]

> On Thu, 20 Apr 2023 13:16:21 +0200 Lorenzo Bianconi wrote:
> > @@ -727,17 +729,21 @@ static int veth_convert_skb_to_xdp_buff(struct veth_rq *rq,
> >  			goto drop;
> >  
> >  		/* Allocate skb head */
> > -		page = alloc_page(GFP_ATOMIC | __GFP_NOWARN);
> > +		if (rq->page_pool)
> 
> There's some condition under which we can get to XDP enabled but no
> pool?

ack, since we destroy the page_pool after disabling the rq napi we can get rid
of 'if (rq->page_pool)' checks here.

> 
> > +			page = page_pool_dev_alloc_pages(rq->page_pool);
> >  		if (!page)
> >  			goto drop;
> >  
> >  		nskb = build_skb(page_address(page), PAGE_SIZE);
> >  		if (!nskb) {
> > -			put_page(page);
> > +			page_pool_put_full_page(rq->page_pool, page, false);
> 
> You can recycle direct, AFAIU the basic rule of thumb is that it's
> always safe to recycle direct from the context which allocates from
> the pool.

ack, I will fix it.

> 
> >  			goto drop;
> >  		}
> >  
> >  		skb_reserve(nskb, VETH_XDP_HEADROOM);
> > +		skb_copy_header(nskb, skb);
> > +		skb_mark_for_recycle(nskb);
> > +
> >  		size = min_t(u32, skb->len, max_head_size);
> >  		if (skb_copy_bits(skb, 0, nskb->data, size)) {
> >  			consume_skb(nskb);
> > @@ -745,16 +751,17 @@ static int veth_convert_skb_to_xdp_buff(struct veth_rq *rq,
> >  		}
> >  		skb_put(nskb, size);
> >  
> > -		skb_copy_header(nskb, skb);
> >  		head_off = skb_headroom(nskb) - skb_headroom(skb);
> >  		skb_headers_offset_update(nskb, head_off);
> >  
> >  		/* Allocate paged area of new skb */
> >  		off = size;
> >  		len = skb->len - off;
> > +		page = NULL;
> 
> Why do you clear the page pointer?

since we do not need 'if (rq->page_pool)' checks anymore we can get rid of it.

Regards,
Lorenzo

> 
> -- 
> pw-bot: cr

[-- Attachment #2: signature.asc --]
[-- Type: application/pgp-signature, Size: 228 bytes --]