linux-kernel - Re: [rfc] increase struct page size?!

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]

Message-ID: <20070520092552.GA7318@wotan.suse.de>
Date:	Sun, 20 May 2007 11:25:52 +0200
From:	Nick Piggin <npiggin@...e.de>
To:	William Lee Irwin III <wli@...omorphy.com>
Cc:	Christoph Lameter <clameter@....com>,
	Linux Kernel Mailing List <linux-kernel@...r.kernel.org>,
	Linux Memory Management List <linux-mm@...ck.org>,
	linux-arch@...r.kernel.org
Subject: Re: [rfc] increase struct page size?!

On Sun, May 20, 2007 at 01:46:47AM -0700, William Lee Irwin III wrote:
> On Sat, May 19, 2007 at 11:15:01AM -0700, William Lee Irwin III wrote:
> >> The cache cost argument is specious. Even misaligned, smaller is
> >> smaller.
> 
> On Sun, May 20, 2007 at 07:22:29AM +0200, Nick Piggin wrote:
> > Of course smaller is smaller ;) Why would that make the cache cost
> > argument specious?
> 
> It's not possible to ignore aggregation. For instance, for a subset
> of mem_map whose size ignoring alignment would otherwise fit in the
> cache to completely avoid sharing any cachelines between page
> structures requires page structures to be separated by at least one
> mem_map index. This is highly unlikely in uniform distributions.

But that wasn't my argument. I _know_ there are cases where the smaller
struct would be better, and I'm sure they would even arise in a running
kernel.
 

> On Sat, May 19, 2007 at 11:15:01AM -0700, William Lee Irwin III wrote:
> >> The cache footprint reduction is merely amortized,
> >> probabilistic, etc.
> 
> On Sun, May 20, 2007 at 07:22:29AM +0200, Nick Piggin wrote:
> > I don't really know what you mean by this, or what part of my cache cost
> > argument you disagree with...
> > I think it is that you could construct mem_map access patterns, without
> > specifically looking at alignment, where a 56 byte struct page would suffer
> > about 75% more cache misses than a 64 byte aligned one (and you could also
> > get about 12% fewer cache misses with other access patterns).
> > I also think the kernel's mem_map access patterns would be more on the
> > random side, so overall would result in significantly fewer cache misses
> > with 64 byte aligned pages.
> > Which part do you disagree with?
> 
> The lack of consideration of the average case. I'll see what I can smoke
> out there.

I _am_ considering the average case, and I consider the aligned structure
is likely to win on average :) I just don't have numbers for it yet.


-
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/