On Thu, Aug 17, 2023 at 04:30:20PM +0900, Shin'ichiro Kawasaki wrote: > The helper function _nvme_connect_subsys() creates a nvme device. It may > take some time after the function call until the device gets ready for > I/O. So it is expected that the test cases call _find_nvme_dev() after > _nvme_connect_subsys() before I/O. _find_nvme_dev() returns the path of > the created device, and it also waits for uuid and wwid sysfs attributes > of the created device get ready. This wait works as the wait for the > device I/O readiness. > > However, this wait by _find_nvme_dev() has two problems. The first > problem is missing call of _find_nvme_dev(). The test case nvme/047 > calls _nvme_connect_subsys() twice, but _find_nvme_dev() is called only > for the first _nvme_connect_subsys() call. This causes too early I/O to > the device with tcp transport [1]. Fix this by moving the wait for the > device readiness from _find_nvme_dev() to _nvme_connect_subsys(). Also > add --wait-for option to _nvme_connect_subsys(). It allows to skip the > wait in _nvmet_passthru_target_connect() which has its own wait for > device readiness. > > The second problem is wrong paths for the sysfs attributes. The paths > do not include namespace index, so the check for the attributes always > fail. Still _find_nvme_dev() does 1 second wait and allows the device > get ready for I/O in most cases, but this is not intended behavior. > Fix the paths by adding the namespace index. > > On top of the checks for sysfs attributes, add 'udevadm settle' and a > check for the created device file. These ensures that the create device > is ready for I/O. > > [1] https://lore.kernel.org/linux-block/CAHj4cs9GNohGUjohNw93jrr8JGNcRYC-ienAZz+sa7az1RK77w@xxxxxxxxxxxxxx/ > > Fixes: c766fccf3aff ("Make the NVMe tests more reliable") > Signed-off-by: Shin'ichiro Kawasaki <shinichiro.kawasaki@xxxxxxx> Thanks! Reviewed-by: Daniel Wagner <dwagner@xxxxxxx>
