linux-kernel - Re: [PATCH 1/2] workqueue: skip lockdep wq dependency in cancel_work

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: <1534872621.25523.39.camel@sipsolutions.net>
Date:   Tue, 21 Aug 2018 19:30:21 +0200
From:   Johannes Berg <johannes@...solutions.net>
To:     Tejun Heo <tj@...nel.org>
Cc:     Lai Jiangshan <jiangshanlai@...il.com>,
        linux-kernel@...r.kernel.org, linux-wireless@...r.kernel.org
Subject: Re: [PATCH 1/2] workqueue: skip lockdep wq dependency in
 cancel_work_sync()

Hi,

> On Tue, Aug 21, 2018 at 07:18:14PM +0200, Johannes Berg wrote:
> > > But this can lead to a deadlock.  I'd much rather err on the side of
> > > discouraging complex lock dancing around ordered workqueues, no?
> > 
> > What can lead to a deadlock?
> 
> Oh not this particular case, but I was wondering whether we'd be
> missing legitimate possible deadlock cases by skipping lockdep for all
> cancel_work_sync()'s as they can actually flush.

I don't see how? This is only relevant in ordered/single-threaded WQs,
but even there it doesn't matter doesn't matter as explained?

I'm actually seeing a false positive report from lockdep, because it
*is* flushing, i.e. I'm running into the case of the work actually
running, i.e. the "_sync" part of "cancel_work_sync()" is kicking in,
but in that case a single-threaded WQ can't have anything executing
*before* it, so we don't need to generate a lockdep dependency - and in
fact don't *want* to create one to avoid the false positive.

I'm not really sure what you think we might be missing? Am I missing
some case where cancel_work_sync() can possibly deadlock? Apart from the
issue I addressed in the second patch, obviously.

johannes