On Mon, Nov 18, 2019 at 04:05:56PM -0800, Sagi Grimberg wrote: > > I'm starting to think we maybe need to get the connect out of the block > layer execution if its such a big problem... Its a real shame if that is > the case... We still need timeout handling for connect commands, so bypassing the block layer will need to figure out some other way to handle that. This patch proposal doesn't really handle the timeouts very well either, though: nvme_rdma_timeout() is going to end up referncing the wrong queue rather than the one the request was submitted on. It doesn't appear to really matter in the current code since it just resets the entire controller, but if it ever wanted to do something queue specific...
