scsi: storvsc: Spread interrupts when picking a channel for I/O requests
authorMichael Kelley (EOSG) <Michael.H.Kelley@microsoft.com>
Wed, 24 Jan 2018 22:14:08 +0000 (22:14 +0000)
committerMartin K. Petersen <martin.petersen@oracle.com>
Thu, 22 Feb 2018 03:30:40 +0000 (22:30 -0500)
Update the algorithm in storvsc_do_io to look for a channel
starting with the current CPU + 1 and wrap around (within the
current NUMA node). This spreads VMbus interrupts more evenly
across CPUs. Previous code always started with first CPU in
the current NUMA node, skewing the interrupt load to that CPU.

Signed-off-by: Michael Kelley <mikelley@microsoft.com>
Reviewed-by: Long Li <longli@microsoft.com>
Signed-off-by: Martin K. Petersen <martin.petersen@oracle.com>
drivers/scsi/storvsc_drv.c

index 8eadb30115aaad1d3bcda70e63abe44e44794e34..620510787763919a5b115aa1eef46f8e84aa4839 100644 (file)
@@ -1310,7 +1310,8 @@ static int storvsc_do_io(struct hv_device *device,
                         */
                        cpumask_and(&alloced_mask, &stor_device->alloced_cpus,
                                    cpumask_of_node(cpu_to_node(q_num)));
-                       for_each_cpu(tgt_cpu, &alloced_mask) {
+                       for_each_cpu_wrap(tgt_cpu, &alloced_mask,
+                                       outgoing_channel->target_cpu + 1) {
                                if (tgt_cpu != outgoing_channel->target_cpu) {
                                        outgoing_channel =
                                        stor_device->stor_chns[tgt_cpu];