[<prev] [next>] [<thread-prev] [thread-next>] [day] [month] [year] [list]
Message-ID: <0141b4a7b45605db66d96d5dd636dd6e@smtp.hushmail.com>
Date: Sat, 24 Oct 2015 00:34:37 +0200
From: magnum <john.magnum@...hmail.com>
To: john-dev@...ts.openwall.com
Subject: Re: Re: "Illegal instruction" error on
On 2015-10-24 00:09, magnum wrote:
> On 2015-10-23 22:00, Solar Designer wrote:
>> The core JtR tree checks for AVX at runtime and fails gracefully if
>> it's built for AVX but the system (the CPU and/or the kernel) lacks
>> AVX support. Ditto for XOP (and even for SSE2 in 32-bit builds).
>> Or have we broken that in jumbo, maybe along with autoconf'ing it?
>> This is worth checking.
>
> I tried these on well:
> a) make -sj8 -f Makefile.legacy linux-x86-64-xop
> b) ./configure CFLAGS="-g -O2 -mxop" --disable-native-tests && make -sj8
>
> Regardless whether autoconf'ed or using legacy Makefile, Jumbo produces
> the "Sorry, XOP is required for this build". I also tried reverting to
> the very same version as Kali 4.0.0 seem to have used and this still
> works. But we've got reports from several places that it doesn't work
> for AVX (with Jumbo, at least).
>
> Could it be that the XOP check somehow works while the AVX one doesn't?
> I have no way to test that. Do we have a machine somewhere that lacks AVX?
Found the problem. When I supplied CFLAGS as above, the reciepe for
john.o got them too but with a pure autoconf with nothing given, we did
not supply any -mavx for that very target. The intentions were good but
not the result: We did not get any CPU_DETECT from arch.h...
I will fix.
magnum
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.
