From: Nick Piggin <npiggin@suse.de>
To: Andi Kleen <ak@suse.de>
Cc: Jeremy Fitzhardinge <jeremy@goop.org>,
David Chinner <dgc@sgi.com>,
Linux Memory Management List <linux-mm@kvack.org>
Subject: Re: [rfc][patch] mm: scalable vmaps
Date: Tue, 19 Feb 2008 02:46:00 +0100 [thread overview]
Message-ID: <20080219014600.GC21165@wotan.suse.de> (raw)
In-Reply-To: <200802181104.45898.ak@suse.de>
On Mon, Feb 18, 2008 at 11:04:45AM +0100, Andi Kleen wrote:
>
> > One thing that will be common to any high performance vmap implementation,
> > however, will be the use of lazy TLB flushing. So I'm mainly interested
> > in comments about this. AFAIK, Xen must be able to eliminate these aliases
> > on demand, and CPA also doesn't want aliases around even if they don't
> > get explicitly referenced by software
>
> It's not really a requirement by CPA, but one by the hardware. Alias
> mappings always need to have the same caching attributes.
Right, yes.
> > (because the hardware may do a
> > random speculative operation through the TLB).
> >
> > So I just wonder if it is enough to provide a (quite heavyweight) function
> > to flush aliases? (vm_unmap_aliases)
>
> For CPA that would work currently (calling that function there
> if the caching attributes are changed), although when CPA use is more wide
> spread than it currently is it might be a problem at some point if it is very slow.
I guess CPA is pretty slow anyway because it does a global tlb flush.
vm_unmap_aliases is not going to be terribly slow by comparison (the
global TLB flush is one of its more expensive aspects).
> > I ripped the not-very-good vunmap batching code out of XFS, and implemented
> > the large buffer mapping with vm_map_ram and vm_unmap_ram... along with
> > a couple of other tricks, I was able to speed up a large directory workload
> > by 20x on a 64 CPU system. Basically I believe vmap/vunmap is actually
> > sped up a lot more than 20x on such a system, but I'm running into other
> > locks now. vmap is pretty well blown off the profiles.
>
> Cool. Gratulations.
Thanks! I'm not sure how "interesting" the workload is ;) but at least it
shows the new vmap is scalable and working properly
--
To unsubscribe, send a message with 'unsubscribe linux-mm' in
the body to majordomo@kvack.org. For more info on Linux MM,
see: http://www.linux-mm.org/ .
Don't email: <a href=mailto:"dont@kvack.org"> email@kvack.org </a>
prev parent reply other threads:[~2008-02-19 1:46 UTC|newest]
Thread overview: 6+ messages / expand[flat|nested] mbox.gz Atom feed top
2008-02-18 8:22 [rfc][patch] mm: scalable vmaps Nick Piggin
2008-02-18 9:29 ` Jeremy Fitzhardinge
2008-02-18 10:20 ` Andi Kleen
2008-02-19 1:42 ` Nick Piggin
2008-02-18 10:04 ` Andi Kleen
2008-02-19 1:46 ` Nick Piggin [this message]
Reply instructions:
You may reply publicly to this message via plain-text email
using any one of the following methods:
* Save the following mbox file, import it into your mail client,
and reply-to-all from there: mbox
Avoid top-posting and favor interleaved quoting:
https://en.wikipedia.org/wiki/Posting_style#Interleaved_style
* Reply using the --to, --cc, and --in-reply-to
switches of git-send-email(1):
git send-email \
--in-reply-to=20080219014600.GC21165@wotan.suse.de \
--to=npiggin@suse.de \
--cc=ak@suse.de \
--cc=dgc@sgi.com \
--cc=jeremy@goop.org \
--cc=linux-mm@kvack.org \
/path/to/YOUR_REPLY
https://kernel.org/pub/software/scm/git/docs/git-send-email.html
* If your mail client supports setting the In-Reply-To header
via mailto: links, try the mailto: link
Be sure your reply has a Subject: header at the top and a blank line
before the message body.
This is a public inbox, see mirroring instructions
for how to clone and mirror all data and code used for this inbox;
as well as URLs for NNTP newsgroup(s).