From mboxrd@z Thu Jan 1 00:00:00 1970 From: Andi Kleen Subject: Re: [PATCH] [9/20] Add blk_kmalloc/blk_alloc_pages Date: Mon, 17 Mar 2008 11:03:29 +0100 Message-ID: <20080317100329.GC12405@basil.nowhere.org> References: <20080307653.720459648@firstfloor.org> <20080307175408.E290E1B41AE@basil.firstfloor.org> <1205445980.2893.85.camel@localhost.localdomain> <20080314134814.GS17940@kernel.dk> <20080314135945.GO2522@one.firstfloor.org> <20080317082711.GB17940@kernel.dk> <20080317083603.GJ27015@one.firstfloor.org> <20080317083832.GE17940@kernel.dk> <20080317085341.GK27015@one.firstfloor.org> <47DE3778.2090600@panasas.com> Mime-Version: 1.0 Content-Type: text/plain; charset=us-ascii Return-path: Received: from smtp-out01.alice-dsl.net ([88.44.60.11]:40744 "EHLO smtp-out01.alice-dsl.net" rhost-flags-OK-OK-OK-OK) by vger.kernel.org with ESMTP id S1751396AbYCQKDb (ORCPT ); Mon, 17 Mar 2008 06:03:31 -0400 Content-Disposition: inline In-Reply-To: <47DE3778.2090600@panasas.com> Sender: linux-scsi-owner@vger.kernel.org List-Id: linux-scsi@vger.kernel.org To: Boaz Harrosh Cc: Andi Kleen , Jens Axboe , James Bottomley , linux-scsi@vger.kernel.org > I would like if we could do a regular allocation and let the That is what v1 of the patchkit did. I only introduced blk_* based on feedback from James who was concerned about the additional bounces. It would be fine for me to go back to that state. Please Jens and James work out what way is the preferred one. Right now there are conflicting requirements from different maintainers. > this code wants to change to use proper BIO and blk_ API as was > proposed by Mike Christie. So there is only that one user at scsi_scan It's actually not only scsi-scan, but a couple of users (see the whole patchkit) > On the other hand it is not only the ISA cases right? Even in an > x86_64 system with a PCI-32 host, we are going to bounce. Perhaps if > there is a convenient API for allocating un-bounced memory for a > block device there will be more users for it, like file-systems. > > The API, I would like that if failing to allocate un--bounced memory > would then allocate any memory and bounce later. pci dma api/ swiotlb is that API (which either bounces or uses an iommu), but it doesn't work for ISA DMA -Andi