From mboxrd@z Thu Jan 1 00:00:00 1970 From: willy@linux.intel.com (Matthew Wilcox) Date: Wed, 18 Jun 2014 11:45:32 -0400 Subject: [PATCH] NVMe: Fail SCSI->NVMe translation for UNMAP when anchor is set In-Reply-To: References: <1403082188-4852-1-git-send-email-daniel.mcleran@intel.com> <20140618152708.GH12025@linux.intel.com> Message-ID: <20140618154532.GI12025@linux.intel.com> On Wed, Jun 18, 2014@09:30:24AM -0600, Dan McLeran wrote: > Good point. So, is Anchor == 1 still illegal for non-resource-provisioned > devices? The spec seems to imply that must be the case. The current revision of the spec seems contradictory. On the one hand, it says that ANC_SUP shall be set to 0. SBC says you're not allowed to set the ANCHOR bit if ANC_SUP is 0. And then the NVMe-SCSI spec says that you can set ANCHOR if the underlying namespace is "resource provisioned" which isn't something that actually exists in NVMe. Since there's no way to actually pass the ANCHOR bit through, and we always set ANC_SUP to 0, I think we should go with your original patch, failing the command if ANCHOR is set.