Follow @Openwall on Twitter for new release announcements and other news
[<prev] [next>] [<thread-prev] [thread-next>] [day] [month] [year] [list]
Message-Id: <E0AE951F-38C2-43A8-B2CA-A4C8AEC59824@gmail.com>
Date: Wed, 25 Mar 2015 18:52:31 +0800
From: Lei Zhang <zhanglei.april@...il.com>
To: john-dev@...ts.openwall.com
Subject: Re: [GSoC] building JtR for MIC

Hi,

I tried to tweak OpenSSL's build options for MIC, but this issue still exists, even using the same options as target x86_64 (except for adding "-mmic" and disabling assembly code).

Out of curiosity, I gave LibreSSL a try. I managed to build LibreSSL (the latest stable version) for MIC, and linked it with john-jumbo in place of OpenSSL. Surprisingly, the issue was gone. I reran the benchmark on MIC with the relinked john-jumbo, and it PASSED all the tests!

So far LibreSSL seems more friendly to MIC. Maybe we could use LibreSSL instead of OpenSSL when building for MIC? Either way, the SSL library need to be built manually.


Lei

> On Mar 24, 2015, at 9:14 PM, Lei Zhang <zhanglei.april@...il.com> wrote:
> 
> 
>> On Mar 24, 2015, at 8:55 PM, Solar Designer <solar@...nwall.com> wrote:
>> 
>> This certainly looks like a miscompile of OpenSSL.  Can you please
>> provide the numbers printed right after SHA1_Init()?  Are they the same
>> or also different for the two builds?
> 
> Yes, the numbers are the same after SHA1_Init(), which is 1732584193.
> 
> 
>> I'd guess that OpenSSL somehow got configured for a wrong size of "long"
>> or/and endianness.  Since you were cross-compiling, you couldn't even
>> "make check" it easily.
> 
> That makes sense. I'll check the configuration options and see if there's anything wrong. I guess all the failing tests have something to do with OpenSSL.
> 
> 
> Lei
> 

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.