netdev - Re: [6.12.15][be2net?] Voluntary context switch within RCU read-side critical section!

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: Windows password security audit tool. GUI, reports in PDF.

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

Message-ID: <a6753983-df29-4d79-a25c-e1339816bd02@blackwall.org>
Date: Wed, 26 Feb 2025 12:33:23 +0200
From: Nikolay Aleksandrov <razor@...ckwall.org>
To: Ian Kumlien <ian.kumlien@...il.com>, Jakub Kicinski <kuba@...nel.org>
Cc: Linux Kernel Network Developers <netdev@...r.kernel.org>
Subject: Re: [6.12.15][be2net?] Voluntary context switch within RCU read-side
 critical section!

On 2/26/25 11:55, Ian Kumlien wrote:
> On Wed, Feb 26, 2025 at 10:24 AM Ian Kumlien <ian.kumlien@...il.com> wrote:
>>
>> On Wed, Feb 26, 2025 at 2:05 AM Jakub Kicinski <kuba@...nel.org> wrote:
>>>
>>> On Tue, 25 Feb 2025 11:13:47 +0100 Ian Kumlien wrote:
>>>> Same thing happens in 6.13.4, FYI
>>>
>>> Could you do a minor bisection? Does it not happen with 6.11?
>>> Nothing jumps out at quick look.
>>
>> I have to admint that i haven't been tracking it too closely until it
>> turned out to be an issue
>> (makes network traffic over wireguard, through that node very slow)
>>
>> But i'm pretty sure it was ok in early 6.12.x - I'll try to do a bisect though
>> (it's a gw to reach a internal server network in the basement, so not
>> the best setup for this)
> 
> Since i'm at work i decided to check if i could find all the boot
> logs, which is actually done nicely by systemd
> first known bad: 6.11.7-300.fc41.x86_64
> last known ok: 6.11.6-200.fc40.x86_64
> 
> Narrows the field for a bisect at least, =)
> 

Saw bridge, took a look. :)

I think there are multiple issues with benet's be_ndo_bridge_getlink()
because it calls be_cmd_get_hsw_config() which can sleep in multiple
places, e.g. the most obvious is the mutex_lock() in the beginning of
be_cmd_get_hsw_config(), then we have the call trace here which is:
be_cmd_get_hsw_config -> be_mcc_notify_wait -> be_mcc_wait_compl -> usleep_range()

Maybe you updated some tool that calls down that path along with the kernel and system
so you started seeing it in Fedora 41?

IMO this has been problematic for a very long time, but obviously it depends on the
chip type. Could you share your benet chip type to confirm the path?

For the blamed commit I'd go with:
 commit b71724147e73
 Author: Sathya Perla <sathya.perla@...adcom.com>
 Date:   Wed Jul 27 05:26:18 2016 -0400

     be2net: replace polling with sleeping in the FW completion path

This one changed the udelay() (which is safe) to usleep_range() and the spinlock
to a mutex.

Cheers,
 Nik