linux-kernel - Re: r8169: disable ASPM during NAPI poll locking changes --> BUG: Invalid wait context --> PREEMPT

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 for Android: free password hash cracker in your pocket

[<prev] [next>] [<thread-prev] [thread-next>] [day] [month] [year] [list]

Message-ID: <d7fa9ffe0e6cd546b8ff3826379c2cf78e5342b6.camel@gmx.de>
Date:   Mon, 22 May 2023 12:41:21 +0200
From:   Mike Galbraith <efault@....de>
To:     Sebastian Andrzej Siewior <bigeasy@...utronix.de>
Cc:     Heiner Kallweit <hkallweit1@...il.com>,
        lkml <linux-kernel@...r.kernel.org>,
        Thomas Gleixner <tglx@...utronix.de>
Subject: Re: r8169: disable ASPM during NAPI poll locking changes --> BUG:
 Invalid wait context --> PREEMPT_RT pain

On Mon, 2023-05-22 at 10:06 +0200, Sebastian Andrzej Siewior wrote:
> On 2023-05-21 16:27:10 [+0200], Mike Galbraith wrote:
> > On Sun, 2023-05-21 at 08:03 +0200, Mike Galbraith wrote:
> > > Greetings,
> > >
> > > The locks added to prep for $subject induce invalid context moaning due
> > > to not being raw locks, but if I do that, the hold time does very bad
> > > things to RT.
> >
> > The locks aren't really 160us horrible. The GPU was enabled (oops),
> > nouveau then makes box horrible for RT (and graphics:).  With GPU
> > turned back off, the largest trace I saw was 77us, which jibes pretty
> > well with new max of low to mid 50s without ftrace running, or up a tad
> > over 20us for this now pretty old commodity i7 desktop box.
>
> So you are saying it is all good?

I wouldn't say good, but not the sky is falling 160us bad.

> I've been looking at the r8169 the other day and it seemed all good
> since it was only scheduling workqueues and NAPI. But now I see this
> mac_ocp_lock + config25_lock which are acquire in hardirq context but
> can't. This needs to be taken care.

I did a patch converting the locks, but didn't post it hoping maybe
those locks weren't really really needed.  It doesn't matter for my
box, as it doesn't do anything but give RT chances to trip over bugs.

	-Mike