[PATCH 0/3] improve nvme quiesce time for large amount of namespaces

[Sagi Grimberg]

>> On Sun, Jul 31, 2022 at 01:23:36PM +0300, Sagi Grimberg wrote:
>>> But maybe we can avoid that, and because we allocate
>>> the connect_q ourselves, and fully know that it should
>>> not be apart of the tagset quiesce, perhaps we can introduce
>>> a new interface like:
>>> -- 
>>> static inline int nvme_ctrl_init_connect_q(struct nvme_ctrl *ctrl)
>>> {
>>>     ctrl->connect_q = blk_mq_init_queue_self_quiesce(ctrl->tagset);
>>>     if (IS_ERR(ctrl->connect_q))
>>>         return PTR_ERR(ctrl->connect_q);
>>>     return 0;
>>> }
>>> -- 
>>>
>>> And then blk_mq_quiesce_tagset can simply look into a per request-queue
>>> self_quiesce flag and skip as needed.
>>
>> I'd just make that a queue flag set after allocation to keep the
>> interface simple, but otherwise this seems like the right thing
>> to do.
> Now the code used NVME_NS_STOPPED to avoid unpaired stop/start.
> If we use blk_mq_quiesce_tagset, It will cause the above mechanism to fail.
> I review the code, only pci can not ensure secure stop/start pairing.
> So there is a choice, We only use blk_mq_quiesce_tagset on fabrics, not 
> PCI.
> Do you think that's acceptable?
> If that's acceptable, I will try to send a patch set.

I don't think that this is acceptable. But I don't understand how
NVME_NS_STOPPED would change anything in the behavior of tagset-wide
quiesce?