Hi,
SCSI's per-LUN queue depth is usually for improving IO merge and
balancing IO load among LUNs. blk-mq has provides fair driver tag
allocation and managed IRQ balances interrupt load among queues,
meantime IO merge doesn't play a big role for SSD, and NVMe doesn't
apply per-namespace queue depth.
This patchset tries to don't apply per-LUN queue depth for SSD, then
we can avoid the expensive atomic operation on sdev->device_busy.
We do understand that this shared counter affects IOPS a lot.
The 1st two patches replaces .device_busy with two private counter for
megaraid_sas and mpt3sas to track inflight per-LUN commands since the
two drivers need the counter for balancing load, even though it is
probably no much use.
The last two patches avoid to operate on sdev->device_busy in IO path
for SSD.
Ming Lei (4):
scsi: megaraid_sas: use private counter for tracking inflight per-LUN
commands
scsi: mpt3sas: use private counter for tracking inflight per-LUN
commands
scsi: sd: register request queue after sd_revalidate_disk is done
scsi: core: don't limit per-LUN queue depth for SSD
block/blk-sysfs.c | 14 +++++++++++-
drivers/scsi/megaraid/megaraid_sas.h | 1 +
drivers/scsi/megaraid/megaraid_sas_base.c | 15 +++++++++++--
drivers/scsi/megaraid/megaraid_sas_fusion.c | 13 +++++++----
drivers/scsi/mpt3sas/mpt3sas_base.c | 3 ++-
drivers/scsi/mpt3sas/mpt3sas_base.h | 1 +
drivers/scsi/mpt3sas/mpt3sas_scsih.c | 15 +++++++++++--
drivers/scsi/scsi_lib.c | 24 +++++++++++++++------
drivers/scsi/sd.c | 8 ++++++-
9 files changed, 77 insertions(+), 17 deletions(-)
Cc: Sathya Prakash <sathya.prakash@xxxxxxxxxxxx>
Cc: Chaitra P B <chaitra.basappa@xxxxxxxxxxxx>
Cc: Suganath Prabu Subramani <suganath-prabu.subramani@xxxxxxxxxxxx>
Cc: Kashyap Desai <kashyap.desai@xxxxxxxxxxxx>
Cc: Sumit Saxena <sumit.saxena@xxxxxxxxxxxx>
Cc: Shivasharan S <shivasharan.srikanteshwara@xxxxxxxxxxxx>
Cc: Ewan D. Milne <emilne@xxxxxxxxxx>
Cc: Christoph Hellwig <hch@xxxxxx>,
Cc: Hannes Reinecke <hare@xxxxxxx>
Cc: Bart Van Assche <bart.vanassche@xxxxxxx>
--
2.20.1
