linux-kernel - Re: [PATCH 0/6] Per-processor private data areas for i386

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: Windows password security audit tool. GUI, reports in PDF.

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

Date:	Wed, 27 Sep 2006 19:46:00 +0000
From:	Pavel Machek <pavel@....cz>
To:	jeremy@...p.org
Cc:	akpm@...l.org, linux-kernel@...r.kernel.org
Subject: Re: [PATCH 0/6] Per-processor private data areas for i386

Hi!

> [ Changes since previous post:
>   - roll a new set of patches with all updates, based on 2.6.18-mm1 ]
> 
> Implement per-processor data areas for i386.
> 
> This patch implements per-processor data areas by using %gs as the
> base segment of the per-processor memory.  This has two principle
> advantages:
> 
> - It allows very simple direct access to per-processor data by
>   effectively using an effective address of the form %gs:offset, where
>   offset is the offset into struct i386_pda.  These sequences are faster
>   and smaller than the current mechanism using current_thread_info().
> 
> - It also allows per-CPU data to be allocated as each CPU is brought
>   up, rather than statically allocating it based on the maximum number
>   of CPUs which could be brought up.
> 
> Performance:
> 
> I've done some simple performance tests on an Intel Core Duo running
> at 1GHz (to emphisize any performance delta).  The results for the
> lmbench null syscall latency test, which should show the most negative
> effect from this change, show a ~9ns decline (.237uS -> .245uS).
> This corresponds to around 9 CPU cycles, and correlates well with
> the addition of the push/load/pop %gs into the hot path.

So we have 4% slowdown...

> I have not yet measured the effect on other typees of processor or
> more complex syscalls (though I would expect the push/pop overhead
> would be drowned by longer times spent in the kernel, and mitigated by
> actual use of the PDA).
> 
> The size improvements on the kernel text are nice as well: 
>     2889361 -> 2883936 = 5425 bytes saved

...and 0.2% smaller kernel. I guess you should demonstrate speedup at
complex syscalls before wedecide it is worth it...?

-- 
Thanks for all the (sleeping) penguins.
-
To unsubscribe from this list: send the line "unsubscribe linux-kernel" in
the body of a message to majordomo@...r.kernel.org
More majordomo info at  http://vger.kernel.org/majordomo-info.html
Please read the FAQ at  http://www.tux.org/lkml/