linux-kernel - Re: Linux 5.17-rc8

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:   Mon, 14 Mar 2022 21:45:05 -0300
From:   Marcelo Roberto Jimenez <marcelo.jimenez@...il.com>
To:     Linus Torvalds <torvalds@...ux-foundation.org>
Cc:     Guenter Roeck <linux@...ck-us.net>,
        Thierry Reding <treding@...dia.com>,
        Linus Walleij <linus.walleij@...aro.org>,
        Vidya Sagar <vidyas@...dia.com>,
        Edmond Chung <edmondchung@...gle.com>,
        Andrew Chant <achant@...gle.com>,
        Will McVicker <willmcvicker@...gle.com>,
        Linux Kernel Mailing List <linux-kernel@...r.kernel.org>,
        Bartosz Golaszewski <brgl@...ev.pl>
Subject: Re: Linux 5.17-rc8

Hi Linus,

I am the author of the "gpio: Revert regression..." patch.

On Mon, Mar 14, 2022 at 5:14 PM Linus Torvalds
<torvalds@...ux-foundation.org> wrote:
>
> [ Adding more people to the cc, since this last change was triggered
> by earlier changes.
>
> On Mon, Mar 14, 2022 at 12:25 PM Guenter Roeck <linux@...ck-us.net> wrote:
> >
> > Build results:
> >         total: 155 pass: 155 fail: 0
> > Qemu test results:
> >         total: 488 pass: 484 fail: 4
>
> Uhhuh. We got all the previous problems sorted out, but a new one instead.
>
> > This is a new problem. It bisects to commit fc328a7d1fcc ("gpio: Revert
> > regression in sysfs-gpio (gpiolib.c)"). The network connection fails
> > in the affected tests. Reverting the offending commit (ie reverting the
> > revert) fixes the problem.
>
> Hmm. Looking at the changes since 5.16, that commit fc328a7d1fcc looks
> somewhat suspicious.
>
> It claims to "revert" things, but the behavior it reverts goes
> basically all the way back to v5.7 (with one of the patches going into
> 5.10).
>
> And it clearly breaks things that used to work much more recently (ie
> this worked in rc7, but it was also the state in every release since
> 5.10).
>
> So unless somebody can find the _real_ issue here, I suspect very
> strongly that that "fix" that came in last week was just wrong.
>
> It is also very non-specific "Some GPIO lines have stopped working"
> with no pointer to actual reports.

The original message in which I posted the patch also had a small
report. I listed the board in which the problem appeared and a small
test script to show the error, which I have used to bisect the issue.

The whole thread is here, the test is in the first message:
https://lore.kernel.org/all/a7fbb773-eb85-ccc7-8bfb-0bfab062ffe1@leemhuis.info/t/

> LinusW? Thierry? Bartoz? Anybody?
>
> Yes, there;s something bad going on here, but we can't randomly "fix"
> things in an rc8 that have worked for several releases by now.

The original patch just reverted the patch that introduced the problem
I found. But if the reversion introduces problems at this point, then
the sane thing to do is to revert the reversion.

At a certain point, I tried Thorsten's suggestion to add a gpio-ranges
property in a way similar to another patch, but the kernel went into
an EPROBE_DEFER deadlock. Thierry Reding made some comments about this
in the sequence.

Following Linus Walleij's suggestion, we are moving the code from the
sysfs interface to the character device. But in the meantime, we are
using this "revert patch" in a 5.10.80 kernel, so maybe someone could
point me to details of the network misbehaviour so that I can also
check it?

>
>                Linus

Regards,
Marcelo.