From mboxrd@z Thu Jan 1 00:00:00 1970 Return-Path: Received: from ns.suse.de ([195.135.220.2]:33043 "EHLO mx1.suse.de" rhost-flags-OK-OK-OK-OK) by vger.kernel.org with ESMTP id S1754440AbXHXLYO (ORCPT ); Fri, 24 Aug 2007 07:24:14 -0400 Date: Fri, 24 Aug 2007 13:24:13 +0200 From: Nick Piggin Subject: Re: [rfc] vmalloc: lazy, batched vunmapping Message-ID: <20070824112413.GB12412@wotan.suse.de> References: <20070824032646.GE6989@wotan.suse.de> <14030.1187945965@redhat.com> Mime-Version: 1.0 Content-Type: text/plain; charset=us-ascii Content-Disposition: inline In-Reply-To: <14030.1187945965@redhat.com> Sender: linux-arch-owner@vger.kernel.org To: David Howells Cc: linux-arch@vger.kernel.org List-ID: On Fri, Aug 24, 2007 at 09:59:25AM +0100, David Howells wrote: > Nick Piggin wrote: > > > I'm working on a bit of code that vmaps/vunmaps a bit more often than > > I'd like. I've implemented a frontend to cache commonly used mappings, > > which solves most of the problem, but in looking various other ways to > > get the last bit of performance, I thought might be generally helpful > > to batch up vunmap driven TLB flushes. So I'll just throw the idea > > out there (is anyone else doing a lot of vmapping? I'd like to hear > > from you!). > > Can the OOM handler force a purge? You could, it would be just one function call from a process context. I didn't really bother at this stage, because it's just a couple of K of physical memory really -- the virtual space purge is obviously the critical thing.