linux-kernel - Re: Bug in vhost-net with RT kernels with SMP

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: <20211217103209.1122679-1-bigeasy@linutronix.de>
Date:   Fri, 17 Dec 2021 11:32:07 +0100
From:   Sebastian Andrzej Siewior <bigeasy@...utronix.de>
To:     fcarli@...il.com
Cc:     linux-kernel@...r.kernel.org,
        "Luis Claudio R . Goncalves" <lgoncalv@...hat.com>,
        stable-rt@...r.kernel.org, Thomas Gleixner <tglx@...utronix.de>
Subject: Re: Bug in vhost-net with RT kernels with SMP

On 2021-12-15 16:44:12 [+0100], Florent Carli wrote:
> Hello all,
Hi,
…

I tried to reproduce this but nothing. However…

> If we try the experiment once again, the tap will fail but we won't
> get a new backtrace.

The backtrace is meant to appear only once.

> A first analyzis was done on the #linux-rt irc channel:
> "huh, looks like the underlying issue is hitting the WARN_ON_ONCE()
> condition in eventfd_signal(), but in the bug handler (which
> apparently is preemptible), we get into print_stop_info() and try to
> use smp_processor_id() in that context, which is wrong because thigns
> are preemptible.
> I can't see the smp_processor_id() usage in mainlin (I don't have an
> RT tree to hand), but someone with more RT knowledge might want to
> look at that"

So that smp_processor_id() thingy is sad because the code changed before
it hit upstream and I didn't notice it.
Patch #1 contains the missing bits and avoids the backtrace from
WARN_ON_ONCE().
Patch #2 is a backport from upstream avoiding trigger WARN_ON_ONCE in
the first place.
Could you please try these two if they help?

> Thanks.

Sebastian