[<prev] [next>] [<thread-prev] [day] [month] [year] [list]
Message-ID: <bfb11627-64d5-42a0-911e-8be99e222396@blackwall.org>
Date: Fri, 29 Aug 2025 19:23:24 +0300
From: Nikolay Aleksandrov <razor@...ckwall.org>
To: Jakub Kicinski <kuba@...nel.org>, Linus Lüssing
<linus.luessing@...3.blue>
Cc: bridge@...ts.linux.dev, netdev@...r.kernel.org,
linux-kernel@...r.kernel.org, Ido Schimmel <idosch@...dia.com>,
Andrew Lunn <andrew+netdev@...n.ch>, Simon Horman <horms@...nel.org>,
Paolo Abeni <pabeni@...hat.com>, Eric Dumazet <edumazet@...gle.com>,
"David S . Miller" <davem@...emloft.net>,
Kuniyuki Iwashima <kuniyu@...gle.com>, Stanislav Fomichev <sdf@...ichev.me>,
Xiao Liang <shaw.leon@...il.com>
Subject: Re: [PATCH 0/9] net: bridge: reduce multicast checks in fast path
On 8/29/25 18:47, Jakub Kicinski wrote:
> On Fri, 29 Aug 2025 10:53:41 +0200 Linus Lüssing wrote:
>> This patchset introduces new state variables to combine and reduce the
>> number of checks we would otherwise perform on every multicast packet
>> in fast/data path.
>>
>> The second reason for introducing these new, internal multicast active
>> variables is to later propagate a safety mechanism which was introduced
>> in b00589af3b04 ("bridge: disable snooping if there is no querier") to
>> switchdev/DSA, too. That is to notify switchdev/DSA if multicast
>> snooping can safely be applied without potential packet loss.
>
> Please leave the git-generated diff stat in the cover letter.
> Please include tree designation in the subject, per:
> https://www.kernel.org/doc/html/next/process/maintainer-netdev.html
>
> I'll leave the real review to the experts but this series appears
> to make kselftests unhappy:
>
just fyi my email wasn't working for 2 days and unfortunately I missed this set
I took a look now on patchwork, I do have comments but it's difficult to reply as
I don't have the emails and have to do it manually to each, so I'd rather wait
for v2.
a few notes for v2:
- please use READ/WRTE_ONCE() for variables that are used without locking
- please make locking symmetric, I saw that br_multicast_open() expects the lock to be already held, while
__br_multicast_stop() takes it itself
- target net-next
- is the mcast lock really necessary, would atomic ops do for this tracking?
- can you provide the full view somewhere, how would this tracking be used? I fear
there might still be races.
- please add more details exactly what we save on the fast-path, I know but it'd be nice
to have it in the commit message as well, all commits just say "reduce checks, save cycles"
but there are no details what we save
I will try to give more detailed comments in v2.
Thank you,
Nik
Powered by blists - more mailing lists