From mboxrd@z Thu Jan 1 00:00:00 1970 From: Paolo Bonzini Subject: Re: virtio(-scsi) vs. chained sg_lists (was Re: [PATCH] scsi: virtio-scsi: Fix address translation failure of HighMem pages used by sg list) Date: Wed, 25 Jul 2012 19:43:31 +0200 Message-ID: <50103043.5050508@redhat.com> References: <1343204966-23560-1-git-send-email-senwang@linux.vnet.ibm.com> <500FB1DE.1000100@redhat.com> <500FBAE8.2050107@panasas.com> <500FBF37.50603@redhat.com> <500FE7D2.7070101@panasas.com> <500FEB63.3000709@redhat.com> <500FF412.3090600@panasas.com> <50100014.2010109@redhat.com> <50101091.5090909@panasas.com> Mime-Version: 1.0 Content-Type: text/plain; charset=UTF-8 Content-Transfer-Encoding: 7bit Return-path: In-Reply-To: <50101091.5090909@panasas.com> Sender: linux-kernel-owner@vger.kernel.org To: Boaz Harrosh Cc: Wang Sen , linux-scsi@vger.kernel.org, JBottomley@parallels.com, stefanha@linux.vnet.ibm.com, mc@linux.vnet.ibm.com, linux-kernel@vger.kernel.org, "kvm@vger.kernel.org" , Rusty Russell List-Id: linux-scsi@vger.kernel.org Il 25/07/2012 17:28, Boaz Harrosh ha scritto: >> 1) what I get is a scsi_cmnd which contains an N-element scatterlist. >> >> 2) virtio-scsi has to build the "packet" that is passed to the hardware >> (it does not matter that the hardware is virtual). This packet (per >> virtio-scsi spec) has an N+1-element scatterlist, where the first >> element is a request descriptor (struct virtio_scsi_cmd_req), and the >> others describe the written data. > > Then "virtio-scsi spec" is crap. It overloads the meaning of > "struct scatterlist" of the first element in an array. to be a > "struct virtio_scsi_cmd_req". What the holy fuck? The first element simply _points_ to the "struct virtio_scsi_cmd_req", just like subsequent elements point to the data. And the protocol of the device is _not_ a struct scatterlist[]. The virtio _API_ takes that array and converts to a series of physical address + offset pairs. > Since you need to change the standard to support chaining then > it is a good time to fix this. Perhaps it is a good time for you to read the virtio spec. You are making a huge confusion between the LLD->virtio interface and the virtio->hardware interface. I'm talking only of the former. >> 3) virtio takes care of converting the "packet" from a scatterlist >> (which currently must be a flat one) to the hardware representation. >> Here a walk is inevitable, so we don't care about this walk. > > "hardware representation" you mean aio or biovec, what ever the > IO submission path uses at the host end? No, I mean the way the virtio spec encodes the physical address + offset pairs. I stopped reading here. Paolo