netdev - Re: BUG: bcmgenet transmit queue timeout lockup 6.15+

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] [day] [month] [year] [list]

Message-ID: <C87FDFC5-FD31-46EA-B95B-E6BE03530B17@broadcom.com>
Date: Fri, 27 Jun 2025 07:40:06 -0700
From: Florian Fainelli <florian.fainelli@...adcom.com>
To: Zak Kemble <zakkemble@...il.com>, Doug Berger <opendmb@...il.com>
CC: netdev@...r.kernel.org
Subject: Re: BUG: bcmgenet transmit queue timeout lockup 6.15+

On June 27, 2025 7:31:23 AM PDT, Zak Kemble <zakkemble@...il.com> wrote:
>Hi, I have a Raspberry Pi CM4 setup as a NAT router with the bcmgenet
>ethernet on the LAN side and an RTL8111 on the WAN (although any
>adapter will work fine for this). I've found that downloading
>something from the internet to a PC on the LAN while also running
>iperf3 from the Pi to the same or another PC at the same time will
>cause transmit queue timeouts and queue 1 locks up. Network
>connectivity is lost and a reboot is needed to fix it. Doing only one
>at a time is fine.
>
>6.14.11-v8+ bcm2711_build
>https://github.com/raspberrypi/linux/actions/runs/15676052461 is fine,
>no timeouts.
>6.15.3-v8+ bcm2711_build
>https://github.com/raspberrypi/linux/actions/runs/15845884292 is
>affected.
>
>The same occurs when I backport the latest driver code (just
>bcmgenet.c and bcmgenet.h IIRC) from kernel 6.16 to 6.12.

There have been a fair amount of changes in the 6.15 cycle, would you be able to run a bisection on the driver alone so we could come up with a more targeted fix? Of particular interest would be these changes:

<https://git.kernel.org/pub/scm/linux/kernel/git/torvalds/linux.git/commit/drivers/net/ethernet/broadcom/genet?id=3b5d4f5a820d362dd46472542b2e961fb1f93515>

We will see about reproducing this here as well. The multi queue implementation has always been a tad peculiar given the unequal sizing of the number of descriptors between the priority queues and non priority queue.

Thanks
Hi Zak,
Florian