netdev - Re: [PATCH net] net_sched: limit try_bulk_dequeue

lists.openwall.net		lists / announce owl-users owl-dev john-users john-dev passwdqc-users yescrypt popa3d-users / oss-security kernel-hardening musl sabotage tlsify passwords / crypt-dev xvendor / Bugtraq Full-Disclosure linux-kernel linux-netdev linux-ext4 linux-hardening linux-cve-announce PHC
Open Source and information security mailing list archives

Hash Suite for Android: free password hash cracker in your pocket

[<prev] [next>] [<thread-prev] [thread-next>] [day] [month] [year] [list]

Message-ID: <CANn89iLUYB7UEvHVaa9w0M=kyDkbOHWQGyC3fh=Nq7yNTaV=og@mail.gmail.com>
Date: Mon, 10 Nov 2025 07:13:31 -0800
From: Eric Dumazet <edumazet@...gle.com>
To: Jesper Dangaard Brouer <hawk@...nel.org>
Cc: "David S . Miller" <davem@...emloft.net>, Jakub Kicinski <kuba@...nel.org>, 
	Paolo Abeni <pabeni@...hat.com>, Simon Horman <horms@...nel.org>, 
	Jamal Hadi Salim <jhs@...atatu.com>, Cong Wang <xiyou.wangcong@...il.com>, 
	Jiri Pirko <jiri@...nulli.us>, Kuniyuki Iwashima <kuniyu@...gle.com>, 
	Willem de Bruijn <willemb@...gle.com>, netdev@...r.kernel.org, eric.dumazet@...il.com, 
	Toke Høiland-Jørgensen <toke@...hat.com>, 
	kernel-team <kernel-team@...udflare.com>, Jesse Brandeburg <jbrandeburg@...udflare.com>
Subject: Re: [PATCH net] net_sched: limit try_bulk_dequeue_skb() batches

On Mon, Nov 10, 2025 at 7:04 AM Jesper Dangaard Brouer <hawk@...nel.org> wrote:
>
> Thanks for sharing.
>
> With these numbers it makes sense that BQL bytelimit isn't limiting this
> much code much.
>
> e.g. 1595568 bytes / 1500 MTU = 1063 packets.
>
> Our prod also have large numbers:
>
> $ grep -H . /sys/class/net/REDACT0/queues/tx-*/byte_queue_limits/limit |
> sort -k2rn -t: | head -n 10
> /sys/class/net/ext0/queues/tx-38/byte_queue_limits/limit:819432
> /sys/class/net/ext0/queues/tx-95/byte_queue_limits/limit:766227
> /sys/class/net/ext0/queues/tx-2/byte_queue_limits/limit:715412
> /sys/class/net/ext0/queues/tx-66/byte_queue_limits/limit:692073
> /sys/class/net/ext0/queues/tx-20/byte_queue_limits/limit:679817
> /sys/class/net/ext0/queues/tx-61/byte_queue_limits/limit:647638
> /sys/class/net/ext0/queues/tx-11/byte_queue_limits/limit:642212
> /sys/class/net/ext0/queues/tx-10/byte_queue_limits/limit:615188
> /sys/class/net/ext0/queues/tx-48/byte_queue_limits/limit:613745
> /sys/class/net/ext0/queues/tx-80/byte_queue_limits/limit:584850
>

Exactly :/

BQL was very nice when we had a single queue and/or relatively slow NIC.

For 200Gbit NIC, assuming being able to run a napi poll every 100 usec,
this means the napi poll has to tx-complete 2.5 MB per run, if a single flow
is pushing 200Gbit on a single tx-queue.

In reality, interrupt mitigations tend to space the napi poll delays, and BQL
reflects this by inflating /limit to a point BQL is not anymore effective.