[<prev] [next>] [<thread-prev] [thread-next>] [day] [month] [year] [list]
Message-ID: <55D6F580.4030906@mailbox.org>
Date: Fri, 21 Aug 2015 11:55:12 +0200
From: Frank Dittrich <frank.dittrich@...lbox.org>
To: john-dev@...ts.openwall.com
Subject: Re: Unaligned access (was: JtR on ARM)
On 08/21/2015 11:44 AM, Lei Zhang wrote:
> I just update a bunch of formats to avoid unaligned accesses when ARCH_ALLOWS_UNALIGNED = 0. All unaligned accesses observed so far happen in set_key(char *key, int), where 'key' might be cast to an int-type pointer. To avoid unaligned access, I create a buffer that is properly aligned, strcpy key to that buffer and then use this buffer as replacement. It is questionable which kind of overhead is higher: unaligned access or redundant 'strcpy's. I tested it on Power, and here's the result:
So far, when ./configure --enable-ubsan is used, we add
-fno-sanitize=alignment, because on x86 we would get tons of false
positives.
You might just remove that -fno-sanitize=alignment for your ARM tests.
If you want to see warnings and habe libubsan installed, you can also
drop -fsanitize-undefined-trap-on-error.
If that helps, you might want to improve the autoconfig, so that for
archs with ARCH_ALLOWS_UNALIGNED = 0 unaligned access is treated as
undefined behavior in --enable-ubsan builds.
Frank
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.
