From mboxrd@z Thu Jan 1 00:00:00 1970 From: Mike Snitzer Subject: Re: block: Deprecate QUEUE_FLAG_CLUSTER and use queue_limits instead Date: Thu, 25 Nov 2010 21:28:05 -0500 Message-ID: <20101126022804.GA15662@redhat.com> References: <85ABFDC4641C4847BA6088F2982EFD8F049D4EB8@nonamew.ptu.promise.com> <4CEEAE35.5060805@fusionio.com> Mime-Version: 1.0 Content-Type: text/plain; charset=us-ascii Return-path: Content-Disposition: inline In-Reply-To: Sender: linux-scsi-owner@vger.kernel.org To: "Martin K. Petersen" Cc: Jens Axboe , dm-devel@redhat.com, Promise_Linux , "linux-scsi@vger.kernel.org" List-Id: dm-devel.ids On Thu, Nov 25 2010 at 7:37pm -0500, Martin K. Petersen wrote: > New version below, fixing the things you and Matthew pointed out... > > > > block: Deprecate QUEUE_FLAG_CLUSTER and use queue_limits instead > > When stacking devices, a request_queue is not always available. This > forced us to have a no_cluster flag in the queue_limits that could be > used as a carrier until the request_queue had been set up for a > metadevice. > > There were several problems with that approach. First of all it was up > to the stacking device to remember to set queue flag after stacking had > completed. But that was already done properly, so that wasn't a problem that needed fixing (it just had potential to be overlooked if/when there is a new stacking driver). > Also, the queue flag and the queue limits had to be kept in > sync at all times. We got that wrong, which could lead to us issuing > commands that went beyond the max scatterlist limit set by the driver. It took me a bit to see exactly where we got it wrong. Looks like __scsi_alloc_queue was only concerned with the queue flag. So a minimalist fix would've been to also set no_cluster = 1 in __scsi_alloc_queue? (OK, I just reviewed Ed's initial report and that is exactly what his proposed patch did! :) This patch's header could stand to be a bit more explicit about where the real problem was ... But I agree that this patch cleans things up nicely. Acked-by: Mike Snitzer Thanks, Mike