Openwall GNU/*/Linux - a small security-enhanced Linux distro for servers
[<prev] [next>] [<thread-prev] [thread-next>] [day] [month] [year] [list]
Date: Tue, 20 Nov 2012 22:19:46 +0100
From: magnum <john.magnum@...hmail.com>
To: john-dev@...ts.openwall.com
Subject: Re: Problems on recent AMD driver

On 20 Nov, 2012, at 16:28 , Claudio André <claudioandre.br@...il.com> wrote:
> As reported on gitHub repository, some people have seen errors using AMD Catalyst 12.10 (and also 12.08). I'm sure at least some of these formats were working on Catalyst 12.06 (using my Radeon HD 6770).
> 
> On CPU they are working fine.
> 
> Test i did and the results:
> 
> ../run/john -t -fo:md5crypt-opencl       => fail (segmentation fault)
> ../run/john -t -fo:rar                    => fail (segmentation fault)
> ../run/john -t -fo:office2007-opencl      => fail (segmentation fault)
> ../run/john -t -fo:office2010-opencl      => fail (segmentation fault)
> ../run/john -t -fo:office2013-opencl      => fail (segmentation fault)
> ../run/john -t -fo:bf-opencl              => ASIC hang (twice)
> ../run/john -t -fo:ssha-opencl            => ok, very slow (probing LWS/GWS)
> ../run/john -t -fo:phpass-opencl          => ok
> ../run/john -t -fo:raw-md4-opencl         => ok
> ../run/john -t -fo:raw-md5-opencl         => ok
> ../run/john -t -fo:raw-sha1-opencl        => ok
> ../run/john -t -fo:nt-opencl              => ok
> ../run/john -t -fo:mysql-sha1-opencl      => ok
> ../run/john -t -fo:mscash2-opencl         => ok
> ../run/john -t -fo:raw-sha256-opencl      => ok
> ../run/john -t -fo:raw-sha512-opencl      => ok
> ../run/john -t -fo:raw-sha512-ng-opencl   => ok
> ../run/john -t -fo:sha256crypt-opencl     => ok
> ../run/john -t -fo:sha512crypt-opencl     => ok
> ../run/john -t -fo:xsha512-opencl         => ok
> ../run/john -t -fo:pwsafe-opencl          => ok

The sad part of the story is I suspect md5crypt-opencl worked just fine right up to commit 1bee9a4, where I "improved" it. And all other segfaulting formats are mine. It seems the AMD drivers are allergic to me. While I do suspect driver bugs (all these formats work fine on several other versions of the Catalyst driver, on nvidia and even under OSX with the exception of office2013) I can't help thinking there's something I do wrong, or should do in some other way. A systematic error, like releasing buffers before unmapping them, something like that. Portions of the code in all the problematic formats are very similar. But I have checked, double checked and triple checked and I just can't see anything wrong with the code.

If someone could review commit 1bee9a4 *very* closely, looking for violations of the specifications, problematic casts, misalignment or whatever, there's a chance you'll find something that solves this for all of the formats. But first, of course, confirm my assumption that md5crypt-opencl worked prior to that commit.

magnum

Powered by blists - more mailing lists

Your e-mail address:

Powered by Openwall GNU/*/Linux - Powered by OpenVZ