Follow @Openwall on Twitter for new release announcements and other news
[<prev] [next>] [<thread-prev] [thread-next>] [day] [month] [year] [list]
Date: Mon, 15 Jul 2013 20:53:39 +0200
From: Katja Malvoni <>
Subject: Re: Parallella: bcrypt

Hi Alexander,

On Mon, Jul 15, 2013 at 7:21 PM, Solar Designer <> wrote:

> Hi Katja,
> On Mon, Jul 15, 2013 at 06:22:51PM +0200, Katja Malvoni wrote:
> > [...] it can't be hardware problem because those cores sometimes
> > return correct results
> I am not so sure it's not a hardware problem.  How soon do the faults
> start happening?  Could it be overheating?

I moved code out of JtR, took test vectors and looked at outputs. It
varies, sometimes first incorrect value is after 4th iteration, sometimes
it's already in 1st iteration. I ran it four times, first incorrect value
appeared in iteration 5, 1, 1, and 5. On the other board in 4 runs first
incorrect value appeared in iteration 2, 4, 1 and 1.
When integrated in JtR and it fails self test (when return value from core
that returned wrong value is tested), usually it's get_hash[0](13) or
get_hash[0](15) (rarely get_hash[0](10)). I tried it again now, passed self
test but after first run there were 12 non cracked passwords although they
should be cracked. When I use 8 cores either all are cracked or 2 - 3
remain after first run.

>  Can you try on the other
> board?  Also, I may try adding some homebrew cooling for the Epiphany
> chip to see if it makes any difference.  (There's no heatsink on this
> chip normally, and it's getting pretty hot.)

I tried other board, passed self test (Raw:    795 c/s real, 795 c/s
virtual) but also didn't crack all the passwords from (25
remained, all cores used). I ran self test again after cracking and it
failed on get_hash[0](15).


Content of type "text/html" skipped

Powered by blists - more mailing lists

Confused about mailing lists and their use? Read about mailing lists on Wikipedia and check out these guidelines on proper formatting of your messages.