Re: [PATCH 12/13] target, vhost-scsi: don't switch cpus on completion

On 2/10/21 2:44 AM, Christoph Hellwig wrote:

quoted

 	struct se_device *se_dev = se_cmd->se_dev;
-	int cpu = se_cmd->cpuid;
+	int cpu;
+
+	if (se_cmd->se_cmd_flags & SCF_IGNORE_CPUID_COMPL)
+		cpu = smp_processor_id();
+	else
+		cpu = se_cmd->cpuid;
 
 	target_queue_cmd_work(&se_dev->queues[cpu].cq, se_cmd, cpu,
 			      target_completion_wq);

Can't we just use se_cmd->cpuid == WORK_CPU_UNBOUND as the no affinity
indicator, which would remove all branches here.

We can't right now because the workqueue_struct does
not have the WQ_UNBOUND bit set. __queue_work will then do:

        /* pwq which will be used unless @work is executing elsewhere */
        if (wq->flags & WQ_UNBOUND) {
                if (req_cpu == WORK_CPU_UNBOUND)
                        cpu = wq_select_unbound_cpu(raw_smp_processor_id());
                pwq = unbound_pwq_by_node(wq, cpu_to_node(cpu));
        } else {
                if (req_cpu == WORK_CPU_UNBOUND)
                        cpu = raw_smp_processor_id();
                pwq = per_cpu_ptr(wq->cpu_pwqs, cpu);
        }

so even if you pass in WORK_CPU_UNBOUND, it will do raw_smp_processor_id
and add the work to the cpu's worker pool.

I think if I add in a new tunable to make the workqueue bound or unbound like I
mentioned in the other thread then I think it will do what you want for both
review comments.