Openwall GNU/*/Linux - a small security-enhanced Linux distro for servers
[<prev] [next>] [<thread-prev] [thread-next>] [day] [month] [year] [list]
Date: Sun, 19 Oct 2014 08:07:02 -0800
From: Royce Williams <royce@...ho.org>
To: john-dev <john-dev@...ts.openwall.com>
Subject: Re: "Failed copy data to gpu" when using fork with descrypt-opencl

On Fri, Oct 10, 2014 at 12:18 AM, Royce Williams <royce@...ho.org> wrote:
> On Fri, Oct 10, 2014 at 12:05 AM, Solar Designer <solar@...nwall.com> wrote:
>
>> Thank you for reporting this.  On further occasions, please direct such
>> postings to john-users rather than john-dev.  This is essentially user
>> experience, and it's useful for other users to know that they may/should
>> use --fork for this purpose, and that it failed in this version for this
>> GPU type.
>
> Ah, understood -- makes sense.  Will do so in the future.
>
>> On Thu, Oct 09, 2014 at 11:58:26PM -0800, Royce Williams wrote:
>>> When running this version of bleeding-jumbo:
>>>
>>> commit 3d8c4ea17f918130e62108be91309883c7023541
>>> Merge: 379881a 53cbf8b
>>> Author: Dhiru Kholia <dhiru@...nwall.com>
>>> Date:   Thu Oct 9 20:28:03 2014 +0200
>>>
>>> ... I'm getting the following error when I try to use fork to get more
>>> CPU cores participating in feeding my GPU:
>>
>> Did this work for you before, on this same GPU, with older
>> bleeding-jumbo?  Can you identify the newer version that worked?
>
> Unfortunately, it's the first time that I had a CPU in this system
> that had multiple cores, so I hadn't tried it before.  It also fails
> to work on a snapshot from September 16th using the same GPU.

I'm still experiencing this symptom, after pulling latest
bleeding-jumbo last night.

$ ./john --format=descrypt-opencl --fork=4 ./badcrypt.test
Device 0: GeForce GTX 970
Local worksize (LWS) 64, Global worksize (GWS) 16384
Loaded 1 password hash (descrypt-opencl, traditional crypt(3) [DES OpenCL])
Node numbers 1-4 of 4 (fork)
OpenCL error (CL_INVALID_COMMAND_QUEUE) in file
(opencl_DES_bs_b_plug.c) at line (637) - (Failed Copy data to gpu)
OpenCL error (CL_OUT_OF_RESOURCES) in file (opencl_DES_bs_b_plug.c) at
line (637) - (Failed Copy data to gpu)
OpenCL error (CL_OUT_OF_RESOURCES) in file (opencl_DES_bs_b_plug.c) at
line (637) - (Failed Copy data to gpu)
OpenCL error (CL_INVALID_COMMAND_QUEUE) in file
(opencl_DES_bs_b_plug.c) at line (631) - (Failed Copy data to gpu)


I'm trying to use 'git-bisect' to track down this issue (and an
opportunity to learn).  Following the two-level test-script model in
the 'git-bisect' manpage, I've created a test script that runs
./configure and make, tries a descrypt-opencl session, and exits with
0 if it exits successfuly, exits with 1 if 'Failed Copy data to gpu'
appears, and exits with 125 otherwise.

# git bisect start ffef50cb45b49f160218ac5189cb1878615bd0ce
48f1d4d4c245c3f6734c63b2371ab10f3d70dc0f --
# git bisect run /home/royce/projects/yestercrack/bisect-test.sh

The contents of bisect-test.sh are:

(cd /usr/local/src/sec/crack/john/JohnTheRipper/src && ./configure &&
make -j8) || exit 125
~/projects/yestercrack/bisect-check_test_case.sh

It appears that some stub Makefiles are colliding with the shipped one.

[snip]
Make process completed.
error: Your local changes to the following files would be overwritten
by checkout:
        src/Makefile
Please, commit your changes or stash them before you can switch branches.
Aborting
Bisecting: 3098 revisions left to test after this (roughly 12 steps)
bisect run failed:
'bisect_state skip' exited with error code 1

This also happens for makefiles deeper in the tree, like
src/escrypt/Makefile, etc.

What is the recommended way to  tell 'git bisect' to ignore Makefiles
generated by ./configure?

Royce

Powered by blists - more mailing lists

Your e-mail address:

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