From mboxrd@z Thu Jan 1 00:00:00 1970 From: maxg@mellanox.com (Max Gurtovoy) Date: Sun, 15 Apr 2018 16:50:58 +0300 Subject: [PATCH 2/2 v3] nvme-rdma: Fix command completion race at error recovery In-Reply-To: <20180412133730.GB5791@lst.de> References: <1523462824-25643-1-git-send-email-israelr@mellanox.com> <1523462824-25643-3-git-send-email-israelr@mellanox.com> <20180412133730.GB5791@lst.de> Message-ID: On 4/12/2018 4:37 PM, Christoph Hellwig wrote: > We really need to keep this code in sync with the various transports, > chances are any race you see in RDMA will happen elsewhere as well. > I saw that the pci driver calls nvme_pci_disable so it will not get any interrupts/completion from the drive. so the nvme_cancel_request loop is safe. Regarding FC we need James here. If you're talking about pushing it to common code in the core layer, then I guess it's a good idea with/without this patchset.