Messages in this thread | | | Date | Mon, 20 Apr 2015 09:58:48 -0600 | From | Jens Axboe <> | Subject | Re: BLK_MQ_F_DEFER_ISSUE |
| |
On 04/20/2015 09:09 AM, Jeff Moyer wrote: > Hi, Jens, > > This flag was introduced under the auspices that the nvme blk-mq > conversion would make use of it. That never materialized. Are there > still plans to make use of it? If not, I'd like to get rid of it. If > so, then let's document it better.
Yep I still want to use it, and I do have a patch that uses it as well that I've run through testing. I thought I had pushed it somewhere, but it doesn't seem like I have. I just wanted to get the lower hanging optimizations out there first, like the kmalloc/kfree and queue-lock-for-completions removal. Former is in, latter is not yet.
It's a useful optimization for lots of cases and I'm sure it'll see other uses, once the nvme patch is in. That will give people a reference point.
-- Jens Axboe
| |