linux-kernel - Re: [V2 1/3] firmware: arm_sdei: fix possible deadlock

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]

Date:   Fri, 14 Feb 2020 18:32:54 +0000
From:   James Morse <james.morse@....com>
To:     luanshi <zhangliguang@...ux.alibaba.com>
Cc:     linux-arm-kernel@...ts.infradead.org, linux-kernel@...r.kernel.org
Subject: Re: [V2 1/3] firmware: arm_sdei: fix possible deadlock

Hi Luanshi,

On 16/01/2020 03:28, luanshi wrote:
> We call sdei_reregister_event() with sdei_list_lock held but
> _sdei_event_register() and sdei_event_destroy() also acquires
> sdei_list_lock thus creating A-A deadlock.
> 
> Fixes: da351827240e ("firmware: arm_sdei: Add support for CPU and system
> power states")
> 

(Nit: stray whitespace in the fixes tag, the backport tools may choke on this)

(Please include 'PATCH' in the [] section of the subject when posting, its part of the
'canonical patch format', and my scripts for pulling a series of the list depend on it!)

> ---

Thanks for picking up my suggestion, ... it was what I think should have been done in the
first place to avoid this bug.
Looking at your patch, we'd need to take the per-event lock around the reads of reregister
and reenable in sdei_cpuhp_up() too, and sdei_reregister_shared(), ... and this quickly
becomes much noisier than a patch for stable should be. (Sorry, I should have tried it
before suggesting it!)

I've picked up your first version, but instead of duplicating the contents of the
function, I've added '_llocked' wrappers to account for that lock already being held. This
isn't great as we have _locked too, but lockdep should keep us honest.
Because I started with your patch, git has kept you as author.
This ended up as patch 2, because it was also necessary to move those reregister updates
into their callers to fix hibernate.

I'll posted what I have next week, sorry for the hiatus.

Thanks,

James