Products Openwall GNU/*/Linux   server OS Linux Kernel Runtime Guard John the Ripper   password cracker Free & Open Source for any platform in the cloud Pro for Linux Pro for macOS Wordlists   for password cracking passwdqc   policy enforcement Free & Open Source for Unix Pro for Windows (Active Directory) yescrypt   KDF & password hashing yespower   Proof-of-Work (PoW) crypt_blowfish   password hashing phpass   ditto in PHP tcb   better password shadowing Pluggable Authentication Modules scanlogd   port scan detector popa3d   tiny POP3 daemon blists   web interface to mailing lists msulogin   single user mode login php_mt_seed   mt_rand() cracker Services Publications Articles Presentations Resources Mailing lists Community wiki Source code repositories (GitHub) Source code repositories (CVSweb) File archive & mirrors How to verify digital signatures OVE IDs What's new

Date: Fri, 21 Aug 2015 17:44:56 +0800
From: Lei Zhang <zhanglei.april@...il.com>
To: john-dev@...ts.openwall.com
Subject: Unaligned access (was: JtR on ARM)

Hi,

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:

[ARCH_ALLOWS_UNALIGNED = 1]
Benchmarking: Raw-MD4 [MD4 128/128 AltiVec 4x]... DONE
Raw:	14005K c/s real, 14146K c/s virtual

Benchmarking: Raw-MD5 [MD5 128/128 AltiVec 4x]... DONE
Raw:	13777K c/s real, 13777K c/s virtual

Benchmarking: Raw-SHA1 [SHA1 128/128 AltiVec 4x]... DONE
Raw:	7733K c/s real, 7733K c/s virtual

Benchmarking: Raw-SHA256 [SHA256 128/128 AltiVec 4x]... DONE
Raw:	3130K c/s real, 3161K c/s virtual

Benchmarking: Raw-SHA512 [SHA512 128/128 AltiVec 2x]... DONE
Raw:	1320K c/s real, 1333K c/s virtual

[ARCH_ALLOWS_UNALIGNED = 0]
Benchmarking: Raw-MD4 [MD4 128/128 AltiVec 4x]... DONE
Raw:	12504K c/s real, 12631K c/s virtual

Benchmarking: Raw-MD5 [MD5 128/128 AltiVec 4x]... DONE
Raw:	3898K c/s real, 3937K c/s virtual

Benchmarking: Raw-SHA1 [SHA1 128/128 AltiVec 4x]... DONE
Raw:	5215K c/s real, 5268K c/s virtual

Benchmarking: Raw-SHA256 [SHA256 128/128 AltiVec 4x]... DONE
Raw:	3924K c/s real, 3963K c/s virtual

Benchmarking: Raw-SHA512 [SHA512 128/128 AltiVec 2x]... DONE
Raw:	1178K c/s real, 1178K c/s virtual

Only raw-sha256 gets faster after forcing aligned accesses. raw-md5 even gets a lot slower. The situation might be different on other archs though.


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.