[PATCH v2 0/2] improve nvme quiesce time for large amount of namespaces

[Chao Leng]

On 2022/10/20 14:34, Christoph Hellwig wrote:
> I though the conclusion from last round was to move the srcu_struct to
> the tagset?
If move the srcu_struct to the tagset, may loss the flexibility.
I am not sure that it is good for other modules currently using blk_mq_quiesce_queue.
Another, I am not sure if there will be a future scenario where blk_mq_quiesce_queue
will have to be used, and if it is good for such scenario.
It is a acceptable cost to allocate a temporary array for SRCU, the max memory size
is actually a few hundred KB, and most of the time it's less than a few KB.
So I did not move the srcu_struct to the tagset in patch V3.

It sounds like a good idea that explore moving the srcu_struct to the tag_set,
But we may need more time to analysis it.
I suggest that do the optimization in a separate patch set.
> 
> .
>