Re: [PATCH 1/1] block: System crashes when cpu hotplug + bouncing port

[Daniel Wagner]

On Tue, Jun 29, 2021 at 05:01:16PM +0800, Ming Lei wrote:
> No, this way fails the request allocation, which isn't expected from
> NVMe fc/rdma/tcp/loop, since io queue can't be connected in this way.

But __nvme_submit_sync_cmd() can return errors too and they need to be
handled in the connect path. So why is this so special? Not that I am
against your patch, I just like to understand the reasoning.