| Message ID | 20241022064109.3303704-1-ming.lei@redhat.com (mailing list archive) |
|---|---|
| State | New, archived |
| Headers | show |
| Series | block/027: setup scsi_debug with MQ | expand |
On 10/21/24 23:41, Ming Lei wrote: > block/027 focuses on race condition between blkg association and > request_queue shutdown. Unfortunately it is a bit easy to trigger > lockup by setting scsi_debug in the following way: > > - single queue > - 21 LUNs > - small queue depth(110) > - 10us completion delay > - fio: 4 jobs, with iodepth 2048 > > The above setting creates big contention on tag allocation of blk-mq > code path, especially scsi_debug takes memcpy to simulate IO, which > doesn't match device in reality. > > So setup scsi_debug with MQ and avoid to trigger lockup which doesn't > exist in real storage device usually. > > Signed-off-by: Ming Lei<ming.lei@redhat.com> Looks good. Reviewed-by: Chaitanya Kulkarni <kch@nvidia.com> -ck
On Oct 22, 2024 / 14:41, Ming Lei wrote: > block/027 focuses on race condition between blkg association and > request_queue shutdown. Unfortunately it is a bit easy to trigger > lockup by setting scsi_debug in the following way: > > - single queue > - 21 LUNs > - small queue depth(110) > - 10us completion delay > - fio: 4 jobs, with iodepth 2048 > > The above setting creates big contention on tag allocation of blk-mq > code path, especially scsi_debug takes memcpy to simulate IO, which > doesn't match device in reality. > > So setup scsi_debug with MQ and avoid to trigger lockup which doesn't > exist in real storage device usually. > > Signed-off-by: Ming Lei <ming.lei@redhat.com> I applied it along with a shellcheck warning fix. Thanks!
diff --git a/tests/block/027 b/tests/block/027 index a79a115..cf4ff16 100755 --- a/tests/block/027 +++ b/tests/block/027 @@ -84,7 +84,7 @@ test() { echo "Running ${TEST_NAME}" scsi_debug_stress_remove virtual_gb=128 max_luns=21 ndelay=10000 \ - max_queue=110 + max_queue=110 submit_queues=$(nproc) echo "Test complete" }
block/027 focuses on race condition between blkg association and request_queue shutdown. Unfortunately it is a bit easy to trigger lockup by setting scsi_debug in the following way: - single queue - 21 LUNs - small queue depth(110) - 10us completion delay - fio: 4 jobs, with iodepth 2048 The above setting creates big contention on tag allocation of blk-mq code path, especially scsi_debug takes memcpy to simulate IO, which doesn't match device in reality. So setup scsi_debug with MQ and avoid to trigger lockup which doesn't exist in real storage device usually. Signed-off-by: Ming Lei <ming.lei@redhat.com> --- tests/block/027 | 2 +- 1 file changed, 1 insertion(+), 1 deletion(-)