From mboxrd@z Thu Jan 1 00:00:00 1970 From: "Martin K. Petersen" Subject: Re: [dm-devel] QUEUE_FLAG_CLUSTER? Date: Thu, 25 Nov 2010 10:21:43 -0500 Message-ID: References: <85ABFDC4641C4847BA6088F2982EFD8F049D4EB8@nonamew.ptu.promise.com> Mime-Version: 1.0 Content-Type: text/plain; charset=us-ascii Return-path: In-Reply-To: <85ABFDC4641C4847BA6088F2982EFD8F049D4EB8@nonamew.ptu.promise.com> (Ed Lin's message of "Fri, 19 Nov 2010 17:24:14 -0800") Sender: linux-scsi-owner@vger.kernel.org To: device-mapper development Cc: linux-scsi@vger.kernel.org, Promise_Linux List-Id: dm-devel.ids >>>>> "Ed" == "Ed Lin <- PTU" > writes: Ed, Ed> The dm already has function to adjust limits based on underlying Ed> device limits. So, is it good to add the no_cluster setting to the Ed> queue limits at the scsi side? For example, some code change like Ed> the following? We should never issue a command that does not adhere to the limits set by the device driver. Upon inspection I agree this is busted and it's my fault. DM does not have a request_queue when stacking and as a result we ended up with two entities tracking whether a device supports clustering or not. And as it turns out the queue flag and the queue limits are not always in agreement. I've coded up a proper fix for this and will post a patch shortly. I'm just running a bunch of tests to ensure we do the right thing... -- Martin K. Petersen Oracle Linux Engineering