From mboxrd@z Thu Jan 1 00:00:00 1970 Return-Path: Received: (majordomo@vger.kernel.org) by vger.kernel.org via listexpand id ; Sun, 16 Sep 2001 05:24:17 -0400 Received: (majordomo@vger.kernel.org) by vger.kernel.org id ; Sun, 16 Sep 2001 05:24:08 -0400 Received: from ns.virtualhost.dk ([195.184.98.160]:42254 "EHLO virtualhost.dk") by vger.kernel.org with ESMTP id ; Sun, 16 Sep 2001 05:23:53 -0400 Date: Sun, 16 Sep 2001 11:24:14 +0200 From: Jens Axboe To: Daniel Phillips Cc: kelley eicher , J , linux-kernel@vger.kernel.org Subject: Re: 0-order allocation failed in 2.4.10-pre8 Message-ID: <20010916112414.A7402@suse.de> In-Reply-To: <3BA24EB0.5000402@i2net.com> <20010916015917Z16125-2757+260@humbolt.nl.linux.org> <20010916100325.B1045@suse.de> <20010916091402Z16065-2757+289@humbolt.nl.linux.org> Mime-Version: 1.0 Content-Type: text/plain; charset=us-ascii Content-Disposition: inline In-Reply-To: <20010916091402Z16065-2757+289@humbolt.nl.linux.org> Sender: linux-kernel-owner@vger.kernel.org X-Mailing-List: linux-kernel@vger.kernel.org On Sun, Sep 16 2001, Daniel Phillips wrote: > On September 16, 2001 10:03 am, Jens Axboe wrote: > > On Sun, Sep 16 2001, Daniel Phillips wrote: > > > > Use the > > > > > > > > > *.kernel.org/pub/linux/kernel/people/axboe/patches/2.4.9/block-highmem-all > > > > > > > > patch and you can use highmem without having to worry about failed > > > > 0-order bounce pages allocations. > > > > > > Right, by using 64 bit DMA instead of bounce buffers. But aren't there > cases > > > where the 64 bit capable hardware isn't there but somebody still wants to > use > > > highmem? > > > > Yes of course. The common case is not 64-bit dma here though, it's just > > being able to DMA to highmem pages (just full 32-bit dma instead of low > > memory dma). And that should cover most systems out there. > > Right, but that does not mean we can forget about bounce buffers, does it. > Most users will probably be able to use full 32-bit dma and users with more > than 4 GB of memory really should go to the effort of making sure their > hardware supports 64 bit dma. But there will still be a few people who have > to use bounce buffers. Of course. My point was merely what with the block-highmem patch, most users will never need bounce -> it would therefore solve the posters issue. > I'm just confirming that we really do have to push on and get bounce buffers > working reliably, even if most people will be able to use your far nicer > alternative. Agreed. It will be much less important, but there will still be a need for it. -- Jens Axboe