Jens Axboe <axboe@xxxxxxxxx> writes: > I did make make sqthread_poll == true set registerfiles as well, and added > the man page bits for the option. If you have further updates, please do > send them. Also interested in how the testing went ;-) OK, I had a look and the patch looks good to me. It tested out fine as well with the attached jobfile, run against both a block device and file system. -Jeff [global] bs=4k time_based=1 runtime=10s ioengine=io_uring rw=randread iodepth=8 size=8g [defaults-buffered] hipri=0 fixedbufs=0 [defaults-direct] stonewall direct=1 hipri=0 fixedbufs=0 [hipri] stonewall hipri=1 fixedbufs=0 direct=1 [fixedbufs-buffered] stonewall hipri=0 fixedbufs=1 direct=0 [fixedbufs-direct] stonewall hipri=0 fixedbufs=1 direct=1 [hipri_fixed] stonewall hipri=1 fixedbufs=1 direct=1 [sqthread_poll] stonewall hipri=0 fixedbufs=0 direct=0 [sqthread_poll-buffered] stonewall hipri=0 fixedbufs=0 direct=0 [sqthread_poll-hipri] stonewall hipri=1 fixedbufs=0 direct=1 [sqthread_poll-fixedbufs] stonewall hipri=0 fixedbufs=1 direct=0
