Re: [PATCH 7/8] ublk: rewrite ublk_ctrl_get_queue_affinity to not rely on hctx->cpumask

[Ming Lei <ming.lei@redhat.com>]

On Thu, Jul 21, 2022 at 07:16:31AM +0200, Christoph Hellwig wrote:
> Looking at the hctxs and cpumap is not safe without at very last a RCU
> reference.  It also requires the queue to be set up before starting the
> device, which leads to rather awkware life time rules.
> 
> Instead rewrite ublk_ctrl_get_queue_affinity to call blk_mq_map_queues
> on an on-stack blk_mq_queue_map and build the cpumask from that.
> 
> Note: given that ublk has not made it into a released kernel it might
> make sense to change the ABI for this command to instead copy the
> qmap.mq_map array (a nr_cpu_ids sized array of unsigned integer
> values  where the CPU index directly points to the queue) to userspace.

qmap.mq_map is too big.

tag_set is embedded into ublk_device, and can be allocated once during
adding device, then ublk_ctrl_get_queue_affinity() can retrieve the info from
tag_set's map directly, then it is simplified a lot.

Also all info for building tag_set is setup during adding device, and
these info won't be changed after adding device, so it is reasonable
to allocate tagset just once.


Thanks,
Ming