From mboxrd@z Thu Jan 1 00:00:00 1970 From: sagi@grimberg.me (Sagi Grimberg) Date: Sat, 10 Aug 2019 22:28:39 -0700 Subject: [PATCH v3 7/7] nvme: don't remove namespace if revalidate failed because of a transport error In-Reply-To: <16b63247-f020-5522-ffab-9404ba7b3408@suse.de> References: <20190808205325.24036-1-sagi@grimberg.me> <20190808205325.24036-8-sagi@grimberg.me> <16b63247-f020-5522-ffab-9404ba7b3408@suse.de> Message-ID: > This is a neat solution, yet I'm not convinced that it'll solve the > problem entirely. I'll give it a spin. Thanks Hannes, Note that with tcp/rdma, there is another patch needed to pass this test. See: [PATCH rfc] nvme: make all fabrics command run on a separate request queue The reason is, that because we have admin commands making their way to the target between we disable the controller and (re)enable it as part of the reset process. Because of that, we have the target failing some of the scan commands that are wrongly made it through (which cause the scan work to wrongly remove the ns and hang). So I would recommend running it with that patch as well (and also get your feedback on it ;))