linux-kernel - Re: [PATCH v2 0/7] Implement inline static calls on PPC32

lists.openwall.net		lists / announce owl-users owl-dev john-users john-dev passwdqc-users yescrypt popa3d-users / oss-security kernel-hardening musl sabotage tlsify passwords / crypt-dev xvendor / Bugtraq Full-Disclosure linux-kernel linux-netdev linux-ext4 linux-hardening linux-cve-announce PHC
Open Source and information security mailing list archives

Hash Suite for Android: free password hash cracker in your pocket

[<prev] [next>] [<thread-prev] [day] [month] [year] [list]

Message-ID: <9e4be6b5-d9c3-8968-8edd-320708909347@csgroup.eu>
Date:   Thu, 8 Sep 2022 06:11:52 +0000
From:   Christophe Leroy <christophe.leroy@...roup.eu>
To:     Benjamin Gray <bgray@...ux.ibm.com>,
        Ard Biesheuvel <ardb@...nel.org>
CC:     X86 ML <x86@...nel.org>, Peter Zijlstra <peterz@...radead.org>,
        Chen Zhongjin <chenzhongjin@...wei.com>,
        Dave Hansen <dave.hansen@...ux.intel.com>,
        Linux Kernel Mailing List <linux-kernel@...r.kernel.org>,
        Nicholas Piggin <npiggin@...il.com>,
        Jason Baron <jbaron@...mai.com>,
        Ingo Molnar <mingo@...hat.com>,
        "sv@...ux.ibm.com" <sv@...ux.ibm.com>,
        "Steven Rostedt (VMware)" <rostedt@...dmis.org>,
        "H. Peter Anvin" <hpa@...or.com>, Borislav Petkov <bp@...en8.de>,
        Thomas Gleixner <tglx@...utronix.de>,
        "agust@...x.de" <agust@...x.de>,
        "open list:LINUX FOR POWERPC (32-BIT AND 64-BIT)" 
        <linuxppc-dev@...ts.ozlabs.org>,
        Josh Poimboeuf <jpoimboe@...nel.org>
Subject: Re: [PATCH v2 0/7] Implement inline static calls on PPC32 - v2



Le 08/09/2022 à 02:13, Benjamin Gray a écrit :
> On Thu, 2022-09-01 at 16:46 +0000, Christophe Leroy wrote:
>> Surprisingly, I get worst performance with inline static call than
>> with
>> out of line static call:
> 
> I'm not sure what hackbench is doing, but when microbenchmarking 64 bit
> out-of-line calls in a loop I saw a similar thing where adding more
> indirection improved the performance despite doing more work. The cause
> seemed to be a combination of using older hardware and the target being
> too short (just an integer increment). Moving to a newer machine and
> adding a lot of NOPs to the target made the performance make sense.

Yes might be.

I think I'll first do new tests with CONFIG_DEBUG_FORCE_FUNCTION_ALIGN_64B

Christophe