[<prev] [next>] [<thread-prev] [thread-next>] [day] [month] [year] [list]
Date: Fri, 3 Mar 2023 11:31:35 +0100
From: Jesper Dangaard Brouer <jbrouer@...hat.com>
To: Yunsheng Lin <linyunsheng@...wei.com>,
Alexander Lobakin <aleksander.lobakin@...el.com>,
Alexei Starovoitov <ast@...nel.org>,
Daniel Borkmann <daniel@...earbox.net>,
Andrii Nakryiko <andrii@...nel.org>,
Martin KaFai Lau <martin.lau@...ux.dev>
Cc: brouer@...hat.com,
Maciej Fijalkowski <maciej.fijalkowski@...el.com>,
Larysa Zaremba <larysa.zaremba@...el.com>,
Toke Høiland-Jørgensen <toke@...hat.com>,
Song Liu <song@...nel.org>,
Jesper Dangaard Brouer <hawk@...nel.org>,
Jakub Kicinski <kuba@...nel.org>, bpf@...r.kernel.org,
netdev@...r.kernel.org, linux-kernel@...r.kernel.org
Subject: Re: [PATCH bpf-next v1 1/2] xdp: recycle Page Pool backed skbs built
from XDP frames
On 02/03/2023 03.30, Yunsheng Lin wrote:
> On 2023/3/2 0:03, Alexander Lobakin wrote:
>> __xdp_build_skb_from_frame() state(d):
>>
>> /* Until page_pool get SKB return path, release DMA here */
>>
>> Page Pool got skb pages recycling in April 2021, but missed this
>> function.
>>
>> xdp_release_frame() is relevant only for Page Pool backed frames and it
>> detaches the page from the corresponding Pool in order to make it
>> freeable via page_frag_free(). It can instead just mark the output skb
>> as eligible for recycling if the frame is backed by a PP. No change for
>> other memory model types (the same condition check as before).
>> cpumap redirect and veth on Page Pool drivers now become zero-alloc (or
>> almost).
>>
>> Signed-off-by: Alexander Lobakin <aleksander.lobakin@...el.com>
>> ---
>> net/core/xdp.c | 4 ++--
>> 1 file changed, 2 insertions(+), 2 deletions(-)
>>
>> diff --git a/net/core/xdp.c b/net/core/xdp.c
>> index 8c92fc553317..a2237cfca8e9 100644
>> --- a/net/core/xdp.c
>> +++ b/net/core/xdp.c
>> @@ -658,8 +658,8 @@ struct sk_buff *__xdp_build_skb_from_frame(struct xdp_frame *xdpf,
>> * - RX ring dev queue index (skb_record_rx_queue)
>> */
>>
>> - /* Until page_pool get SKB return path, release DMA here */
>> - xdp_release_frame(xdpf);
>> + if (xdpf->mem.type == MEM_TYPE_PAGE_POOL)
>> + skb_mark_for_recycle(skb);
>
>
> We both rely on both skb->pp_recycle and page->pp_magic to decide
> the page is really from page pool. So there was a few corner case
> problem when we are sharing a page for different skb in the driver
> level or calling skb_clone() or skb_try_coalesce().
> see:
> https://github.com/torvalds/linux/commit/2cc3aeb5ecccec0d266813172fcd82b4b5fa5803
> https://lore.kernel.org/netdev/MW5PR15MB51214C0513DB08A3607FBC1FBDE19@MW5PR15MB5121.namprd15.prod.outlook.com/t/
> https://lore.kernel.org/netdev/167475990764.1934330.11960904198087757911.stgit@localhost.localdomain/
>
> As the 'struct xdp_frame' also use 'struct skb_shared_info' which is
> sharable, see xdp_get_shared_info_from_frame().
>
> For now xdpf_clone() does not seems to handling frag page yet,
> so it should be fine for now.
>
> IMHO we should find a way to use per-page marker, instead of both
> per-skb and per-page markers, in order to avoid the above problem
> for xdp if xdp has a similar processing as skb, as suggested by Eric.
>
Moving to a per-page marker can be *more* expensive if the struct-page
memory isn't cache-hot. So, if struct-page is accessed anyhow then sure
we can move it to a per-page marker.
> https://lore.kernel.org/netdev/CANn89iKgZU4Q+THXupzZi4hETuKuCOvOB=iHpp5JzQTNv_Fg_A@mail.gmail.com/
>
>>
>> /* Allow SKB to reuse area used by xdp_frame */
>> xdp_scrub_frame(xdpf);
>>
>
Powered by blists - more mailing lists