From mboxrd@z Thu Jan 1 00:00:00 1970 Return-Path: Received: (majordomo@vger.kernel.org) by vger.kernel.org via listexpand id ; Tue, 18 Jun 2002 08:19:29 -0400 Received: (majordomo@vger.kernel.org) by vger.kernel.org id ; Tue, 18 Jun 2002 08:19:27 -0400 Received: from ns.virtualhost.dk ([195.184.98.160]:30625 "EHLO virtualhost.dk") by vger.kernel.org with ESMTP id ; Tue, 18 Jun 2002 08:18:46 -0400 Date: Tue, 18 Jun 2002 14:18:46 +0200 From: Jens Axboe To: Linux Kernel Subject: [PATCH] block-highmem-all-19 Message-ID: <20020618121846.GC814@suse.de> Mime-Version: 1.0 Content-Type: text/plain; charset=us-ascii Content-Disposition: inline Sender: linux-kernel-owner@vger.kernel.org X-Mailing-List: linux-kernel@vger.kernel.org Hi, In preparation of 2.4.20-pre, here's an updated version of the block-highme patch. This enables block drivers to do I/O to high memory pages instead of reverting to low memory bounce buffers. Changes: - (scsi) Fix critical SCpnt->request_buffer -> SCpnt->request.buffer error - (scsi) Backport __init_io() changes from 2.5, that enable us to politely back off on queueing more I/O with any failure. We use this now instead of reverting to single segment requests on sgtable allocation failures. For a highmem host this got nasty, because we had to map the buffer_head and use a bounce buffer for that single segment anyway. That is now eliminated. Even for non-highmem I/O the single segment fall back is slower because of the over head associated with command setup etc. *.kernel.org/pub/linux/kernel/people/axboe/patches/v2.4/2.4.19-pre10/block-highmem-all-19.bz2 Enjoy, -- Jens Axboe