On 5/3/21 10:42 AM, Jens Axboe wrote: > On 5/3/21 9:47 AM, Bart Van Assche wrote: >> On 4/17/21 8:29 AM, Jens Axboe wrote: >>> There's currently no way to experiment with polled IO with null_blk, >>> which seems like an oversight. This patch adds support for polled IO. >>> We keep a list of issued IOs on submit, and then process that list >>> when mq_ops->poll() is invoked. >>> >>> A new parameter is added, poll_queues. It defaults to 1 like the >>> submit queues, meaning we'll have 1 poll queue available. >> >> Has anyone run blktests against blk-for-next since this patch got >> queued? The following appears in the kernel log if I run blktests: > > Do you know what parameters the module was loaded with? I'll drop this > one from the 5.13 queue for now. Hi Jens, The call trace was reported about 35 seconds after the message indicating the start of the test so I assume that the null_blk driver was loaded by the following shell code from tests/block/010: if ! _init_null_blk queue_mode=2 submit_queues=16 nr_devices=32 shared_tags=1; then return 1 fi Bart.
