On Wed, May 30, 2018 at 4:36 PM, jianchao.wang
[off-list ref] wrote:

Hi ming

Thanks for your kindly response.

On 05/30/2018 04:22 PM, Ming Lei wrote:

quoted

quoted

quoted

you could keep the software queues as-is but add our own version of
flush_busy_ctxs() that only removes requests of the domain that we want.
If one domain gets backed up, that might get messy with long iterations,
though.

Yes, I also considered this approach :)
But the long iterations on every ctx->rq_list looks really inefficient.

Right, this list can be quite long if dispatch token is used up.

You might try to introduce per-domain list into ctx directly, then 'none'
may benefit from this change too since bio merge should be done
on the per-domain list actually.

Yes, it maybe good for merging of 'none', because the rq_list is split into 3
lists, and not need to iterate the whole rq_list any more.
But what's about the dispatch when there is no io scheduler.

blk_mq_flush_busy_ctxs() and blk_mq_dequeue_from_ctx() should work
fine in case of 'none' if per-domain list is added to ctx. Then we can make
none to be a bit fair on READ/WRITE.

We will dispatch request from ctx one by one at the moment.
If we have per-domain list in ctx, we have to introduce some policies to determine
which domain to dispatch, and these policies should be in io scheduler actually.

The policy is done by IO scheduler, and you can just pick up request
from ctx/domain list easily by introducing one blk-mq core API.

Thanks,
Ming Lei