linux-kernel - Re: atc260x has broken locking

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

Message-ID: <2559722.y8k8t1CynJ@mobilepool36.emlix.com>
Date:   Fri, 25 Feb 2022 08:36:55 +0100
From:   Rolf Eike Beer <eb@...ix.com>
To:     Cristian Ciocaltea <cristian.ciocaltea@...il.com>
Cc:     Manivannan Sadhasivam <mani@...nel.org>,
        Lee Jones <lee.jones@...aro.org>,
        linux-actions@...ts.infradead.org, linux-kernel@...r.kernel.org
Subject: Re: atc260x has broken locking

Am Freitag, 25. Februar 2022, 00:14:03 CET schrieb Cristian Ciocaltea:
> Hi Eike,
> 
> On Wed, Feb 23, 2022 at 12:07:48PM +0100, Rolf Eike Beer wrote:
> > When looking at this functions I found the locking to be broken for the
> > atomic case (comments stripped):
> > 
> > static void regmap_lock_mutex(void *__mutex)
> > {
> > 
> > 	struct mutex *mutex = __mutex;
> > 	
> > 	if (system_state > SYSTEM_RUNNING && irqs_disabled())
> > 	
> > 		mutex_trylock(mutex);
> > 	
> > 	else
> > 	
> > 		mutex_lock(mutex);
> > 
> > }
> > 
> > static void regmap_unlock_mutex(void *__mutex)
> > {
> > 
> > 	struct mutex *mutex = __mutex;
> > 	
> > 	mutex_unlock(mutex);
> > 
> > }
> > 
> > When the mutex is already locked and the atomic context is hit then the
> > lock will not be acquired, this is never noticed, and it afterwards is
> > unlocked anyway.
> > 
> > The comment says this is inspired from i2c_in_atomic_xfer_mode() to detect
> > the atomic case, but the important caller __i2c_lock_bus_helper()
> > actually does check and pass on the return value of mutex_trylock(),
> > which is missing here.
> This is indeed a limitation of the current implementation because the
> main goal was to offer initial support for SBC poweroff and reboot
> use cases, which were the only cases where the atomic context kicks in.
> 
> Hence it was more important to make sure those operations are triggered
> rather than failing due to an error condition which is hard to be
> handled properly - e.g. getting a behaviour similar with the '-EGAIN'
> scenario in the I2C implementation.

Which makes sense as the unlock is in fact never reached then I guess because 
the system reboots or shuts down before. Maybe this should end up as comment 
somewhere in the code.

> Out of pure curiosity, on which hardware do you plan to use this, if you
> haven't already?

On no hardware. I annotated mutex_trylock() as __must_check and this was the 
only place that broke the build afterwards.

Eike
-- 
Rolf Eike Beer, emlix GmbH, https://www.emlix.com
Fon +49 551 30664-0, Fax +49 551 30664-11
Gothaer Platz 3, 37083 Göttingen, Germany
Sitz der Gesellschaft: Göttingen, Amtsgericht Göttingen HR B 3160
Geschäftsführung: Heike Jordan, Dr. Uwe Kracke – Ust-IdNr.: DE 205 198 055

emlix - smart embedded open source
Download attachment "signature.asc" of type "application/pgp-signature" (314 bytes)