From: Nicholas Bellinger <nab@xxxxxxxxxxxxxxx>
Hi folks,
This series attempts to reduce se_device->execute_task_lock contention
between qla2xxx (fabric) wq context for task execution of READs with
target_submit_cmd() vs. transport_processing_thread() context usage for
pending WRITEs with __transport_execute_tasks(). It also no longer attempts
to drain ->execute_task_list queue while processing each se_cmd descriptor
in transport_processing_thread() context, which also appears to have a
minor positive effect on small I/O block performance with qla2xxx LUNs
With qla2xxx fabric ports and raw block flash backends + manual IRQ
affinity, it appears to save a few extra percent of cpu per qla2xxx wq
process context vs. transport_processing_thread() with mixed random 4k
workloads from what I can tell with perf and fio testing.
So far this has been running with qla2xxx + loopback LUNs, but still
needs to test with other fabrics and heavy active I/O shutdown.
Thanks,
Cc: Christoph Hellwig <hch@xxxxxx>
Cc: Roland Dreier <roland@xxxxxxxxxxxxxxx>
Cc: Joern Engel <joern@xxxxxxxxx>
Nicholas Bellinger (3):
target: Remove extra se_device->execute_task_lock access in fast path
target: Remove __transport_execute_tasks() for each processing
context
target: Remove legacy device status check from
transport_execute_tasks
drivers/target/target_core_transport.c | 51 +++++++++++++++-----------------
1 files changed, 24 insertions(+), 27 deletions(-)
--
1.7.2.5
--
To unsubscribe from this list: send the line "unsubscribe target-devel" in
the body of a message to majordomo@xxxxxxxxxxxxxxx
More majordomo info at http://vger.kernel.org/majordomo-info.html
