linux-kernel - [regression] commit d298b03506d3 ("x86/fpu: Restore the masking out of reserved MXCSR bits")

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-next>] [day] [month] [year] [list]

Message-ID: <YWgYIYXLriayyezv@intel.com>
Date:   Thu, 14 Oct 2021 14:44:33 +0300
From:   Ville Syrjälä <ville.syrjala@...ux.intel.com>
To:     Borislav Petkov <bp@...e.de>
Cc:     Ser Olmy <ser.olmy@...tonmail.com>,
        Thomas Gleixner <tglx@...utronix.de>,
        Ingo Molnar <mingo@...hat.com>,
        "H. Peter Anvin" <hpa@...or.com>, x86@...nel.org,
        linux-kernel@...r.kernel.org
Subject: [regression] commit d298b03506d3 ("x86/fpu: Restore the masking out
 of reserved MXCSR bits")

Hi,

I have a 32bit installation here that stopped working. Bisected it
to commit d298b03506d3 ("x86/fpu: Restore the masking out of reserved
MXCSR bits").

dhcpcd was the first thing I notice being affected on account of
network not coming up, and after trying to look at it with gdb also
gdb turned out to be broken.

strace of dhcpcd shows a SIGFPE getting delivered, after which it gets
stuck (seem to be sitting in poll but not responding to even ^C).
And gdb seems to be stuck in a perpetual SIGFPE loop and won't even
get to the prompt.

The crucial bit here seems to be that most of the software is built
with -mfpmath=sse. After rebuilding dhcpcd without that it started
to work on the broken kernel. Rebuilding gdb didn't help so I whatever
SSE usage is causing the issue is presumably happening in a library.
Had to do the rebuilds on a working kernel as well because otherwise
the build itself would die to a SIGFPE somewhere.

Tested the same disk on on both a 64bit capable Pentium D
and a 32bit only Pentium 4 just to rule out the specific CPU.
Busted on both.

-- 
Ville Syrjälä
Intel