netdev - Re: netpoll: raspberrypi [4 5] driver locking woes

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: <8e60d336-9cab-4003-8972-bda0b041d8cf@linux.dev>
Date: Thu, 28 Aug 2025 14:56:00 -0400
From: Sean Anderson <sean.anderson@...ux.dev>
To: Mike Galbraith <efault@....de>, Robert Hancock <robert.hancock@...ian.com>
Cc: Breno Leitao <leitao@...ian.org>,
 "netdev@...r.kernel.org" <netdev@...r.kernel.org>
Subject: Re: netpoll: raspberrypi [4 5] driver locking woes

Hi Mike,

On 8/28/25 13:26, Mike Galbraith wrote:
> On Thu, 2025-08-28 at 10:57 -0400, Sean Anderson wrote:
>> Hi Mike,
>> 
>> On 8/27/25 12:02, Mike Galbraith wrote:
>> > Unexpected addendum to done deal datapoint, so off list.
>> > 
>> > On Tue, 2025-08-26 at 11:49 +0200, Mike Galbraith wrote:
>> > > 
>> > > The pi5 gripe fix is equally trivial, but submitting that is pointless
>> > > given there's something else amiss in fingered commit.  This is all of
>> > > the crash info that escapes the box w/wo gripes silenced.
>> > > 
>> > > [   51.688868] sysrq: Trigger a crash
>> > > [   51.688892] Kernel panic - not syncing: sysrq triggered crash
>> > > [   51.698066] CPU: 0 UID: 0 PID: 0 Comm: swapper/0 Not tainted 6.17.0-rc2-v8-lockdep #533 PREEMPTLAZY
>> > > [   51.707234] Hardware name: Raspberry Pi 5 Model B Rev 1.0 (DT)
>> > > [   51.713085] Call trace:
>> > > [   51.715532]  show_stack+0x20/0x38 (C)
>> > > [   51.719206]  dump_stack_lvl+0x38/0xd0
>> > > [   51.722878]  dump_stack+0x18/0x28
>> > > 
>> > > That aspect is a punt and run atm (time.. and dash of laziness:).
>> > 
>> > Plan was to end datapoint thread, but after booting pi5's 6.12 kernel,
>> > for some reason I fired up netconsole.. and box promptly exhibited the
>> > netpoll locking bug warning, indicating presence of 138badbc21a0. 
>> > Instead of saying to self "nope, just walk away", I poked SysRq-C.. and
>> > the bloody damn monitoring box received a 100% complete death rattle. 
>> > Well bugger.
>> 
>> Did you get a backtrace for this?
> 
> Yes, logs for 6.12.41 and 6.17.0-rc2 attached.
> 
> Since a patch has meanwhile landed, also a log of patched 6.17.0-rc3
> now gripe free (yay) but with aforementioned broken output, followed by
> addition of the e6a532185daa revert to confirm it still cures that.
> 
>> And to be clear, the steps to reproduce this are to boot a kernel with
>> lockdep enabled with netconsole on macb and then hit sysrq?
> 
> Yup.

Looks like the tx completion path can also be called from netpoll. Can
you try the attached patch?

--Sean
View attachment "0001-net-macb-Fix-tx_ptr_lock-locking.patch" of type "text/x-patch" (4585 bytes)