[bug report] NVMe/IB: reset_controller need more than 1min

[Sagi Grimberg]

> Hi
> I found this issue on 5.13-rc2 with NVMe/IB environment, could anyone
> help check it?
> Thanks.
> 
> $ time echo 1 >/sys/block/nvme0n1/device/reset_controller
> real 0m10.678s
> user 0m0.000s
> sys 0m0.000s
> $ time echo 1 >/sys/block/nvme0n1/device/reset_controller
> real 1m11.530s
> user 0m0.000s
> sys 0m0.000s
> 
> target:
> $ dmesg | grep nvme
> [  276.891454] nvmet: creating controller 1 for subsystem testnqn for
> NQN nqn.2014-08.org.nvmexpress:uuid:4c4c4544-0056-4c10-8058-b7c04f383432.
> [  287.374412] nvmet: ctrl 1 keep-alive timer (5 seconds) expired!
> [  287.399317] nvmet: ctrl 1 fatal error occurred!
> [  348.412672] nvmet: creating controller 1 for subsystem testnqn for
> NQN nqn.2014-08.org.nvmexpress:uuid:4c4c4544-0056-4c10-8058-b7c04f383432.
> 
> client:
> $ dmesg | grep nvme
> [  281.704475] nvme nvme0: creating 40 I/O queues.
> [  285.557759] nvme nvme0: mapped 40/0/0 default/read/poll queues.
> [  353.187809] nvme nvme0: I/O 8 QID 0 timeout
> [  353.193100] nvme nvme0: Property Set error: 881, offset 0x14
> [  353.226082] nvme nvme0: creating 40 I/O queues.
> [  357.088266] nvme nvme0: mapped 40/0/0 default/read/poll queues.

It appears that there is an admin timeout that is either triggered
by the reset or unrelated.

Can you run nvme reset /dev/nvme0 instead so we can see the "resetting
controller" print?