[<prev] [next>] [<thread-prev] [thread-next>] [day] [month] [year] [list]
Message-ID: <2d630a13-ca07-40df-8544-7b0e9373eaf1@intel.com>
Date: Mon, 8 Sep 2025 14:02:05 +0200
From: Alexander Lobakin <aleksander.lobakin@...el.com>
To: Jakub Kicinski <kuba@...nel.org>
CC: <davem@...emloft.net>, <netdev@...r.kernel.org>, <edumazet@...gle.com>,
<pabeni@...hat.com>, <andrew+netdev@...n.ch>, <horms@...nel.org>,
<daniel@...earbox.net>, <hawk@...nel.org>, <john.fastabend@...il.com>,
<sdf@...ichev.me>, <michael.chan@...adcom.com>, <anthony.l.nguyen@...el.com>,
<marcin.s.wojtas@...il.com>, <tariqt@...dia.com>, <mbloch@...dia.com>,
<jasowang@...hat.com>, <bpf@...r.kernel.org>, <pavan.chebbi@...adcom.com>,
<przemyslaw.kitszel@...el.com>
Subject: Re: [PATCH net-next 0/2] net: xdp: handle frags with unreadable
memory
From: Jakub Kicinski <kuba@...nel.org>
Date: Fri, 5 Sep 2025 15:15:37 -0700
> Make XDP helpers compatible with unreadable memory. This is very
> similar to how we handle pfmemalloc frags today. Record the info
> in xdp_buf flags as frags get added and then update the skb once
> allocated.
>
> This series adds the unreadable memory metadata tracking to drivers
> using xdp_build_skb_from*() with no changes on the driver side - hence
> the only driver changes here are refactoring. Obviously, unreadable memory
> is incompatible with XDP today, but thanks to xdp_build_skb_from_buf()
> increasing number of drivers have a unified datapath, whether XDP is
> enabled or not.
>
> RFC: https://lore.kernel.org/20250812161528.835855-1-kuba@kernel.org
Reviewed-by: Alexander Lobakin <aleksander.lobakin@...el.com>
I have plans to add possibility to run XDP progs on devmem/io_uring
queues when it's known that the program doesn't heed to access the
payload, only the headers. Currently no free time slots for this, but
eventually we'll get there. This is a good prereq for this.
Thanks,
Olek
Powered by blists - more mailing lists