[PATCH 2/2] nvme: don't do scan work if io queue count is zero

[liruozhu]

It seems that transport driver will only release the tagset when the 
connection is completely deleted.
In reconnection, if IO tagset is created, transport driver is not 
allowed to reconnect without any IO queues.
See patch("nvme-rdma: fix possible hang when failing to set io queues").
But it didn’t work as expected, the first patch try to fix it.

This patch try to prevent similar logical problems in the transport driver.

Thanks,
Ruozhu
在 2021/8/3 23:55, Keith Busch 写道:
> On Tue, Aug 03, 2021 at 05:06:30PM +0800, Ruozhu Li wrote:
>> kernel panic when try to rescan ns when io queue count is zero.Because
>> kernel BUG_ON when hctx_idx is greater than ctrl->queue_count.
>> --
>> nvme_rdma_init_hctx+0x58/0x60 [nvme_rdma]
>> blk_mq_realloc_hw_ctxs+0x140/0x4c0
>> blk_mq_init_allocated_queue+0x130/0x410
>> blk_mq_init_queue+0x40/0x88
>> nvme_validate_ns+0xb8/0x740
>> nvme_scan_work+0x29c/0x460
>> process_one_work+0x1f8/0x490
>> worker_thread+0x50/0x4b8
>> kthread+0x134/0x138
>> ret_from_fork+0x10/0x18
>> --
>> Defence it by not allowing rescan ns when io queue count is zero.
>>
>> Signed-off-by: Ruozhu Li <liruozhu at huawei.com>
>> ---
>>   drivers/nvme/host/core.c | 6 ++++--
>>   1 file changed, 4 insertions(+), 2 deletions(-)
>>
>> diff --git a/drivers/nvme/host/core.c b/drivers/nvme/host/core.c
>> index dfd9dec0c1f6..d9f837eb3e26 100644
>> --- a/drivers/nvme/host/core.c
>> +++ b/drivers/nvme/host/core.c
>> @@ -141,7 +141,8 @@ void nvme_queue_scan(struct nvme_ctrl *ctrl)
>>   	/*
>>   	 * Only new queue scan work when admin and IO queues are both alive
>>   	 */
>> -	if (ctrl->state == NVME_CTRL_LIVE && ctrl->tagset)
>> +	if (ctrl->state == NVME_CTRL_LIVE &&
>> +	    ctrl->tagset && ctrl->queue_count > 1)
> Why is the transport driver leaving an unusable tagset allocated when
> there are no IO queues to support it? I think it should free the tagset
> in that case. At least that's what pci does.
> .