On Mon, 2019-01-07 at 13:21 -0800, Bart Van Assche wrote:
> When I tried to run the SRP tests from the blktests repository against the
> v5.0 kernel several issues were reported. The same tests pass against kernel
> v4.20.
A correction: this was triggered by a new test ("Run sg_reset while I/O is
ongoing").
Anyway, I don't think that such a test should trigger the following statement in
blk_mq_start_request():
WARN_ON_ONCE(blk_mq_rq_state(rq) != MQ_RQ_IDLE);
A few kernel versions back I ran sg_reset manually while I/O was ongoing and
that
test passed. So this test result may indicate an issue in the blk-mq request
state
rework.
Bart.
