[PATCH 1/3] nvme-rdma: handle cpu unplug when re-establishing the controller

[Sagi Grimberg]

>> diff --git a/drivers/nvme/host/rdma.c b/drivers/nvme/host/rdma.c
>> index 779f516e7a4e..47a479f26e5d 100644
>> --- a/drivers/nvme/host/rdma.c
>> +++ b/drivers/nvme/host/rdma.c
>> @@ -343,8 +343,6 @@ static int __nvme_rdma_init_request(struct nvme_rdma_ctrl *ctrl,
>>  	struct ib_device *ibdev = dev->dev;
>>  	int ret;
>>
>> -	BUG_ON(queue_idx >= ctrl->queue_count);
>
> This makes me a bit nervous, shouldn't we ensure that queue_count
> is accurate even with the fix?

I don't think this is true anymore given that we allocate the
tagset with X nr_hw_queues and if we unplug a cpu when we reset
or reconnect a controller we take the minimum of initial nr_io_queues
(X) and online cpus.

I can change this to be:

BUG_ON(queue_idx >= ctrl->opts->nr_io_queues);

instead if we want to keep the check...