x86/irq: Optimize free vector check in the CPU offline path
authorChen Yu <yu.c.chen@intel.com>
Sun, 16 Apr 2017 15:43:30 +0000 (23:43 +0800)
committerThomas Gleixner <tglx@linutronix.de>
Thu, 20 Apr 2017 13:25:09 +0000 (15:25 +0200)
Before offlining a CPU its required to check whether there are enough free
irq vectors available so interrupts can be migrated away from the CPU.

This check is executed whether its required or not and neither stops
searching when the number of required free vectors are reached.

Bypass the free vector check if the current CPU has no irq to migrate and
leave the for_each_online_cpu() loop when the free vector count reaches the
number of required vectors.

Suggested-by: Thomas Gleixner <tglx@linutronix.de>
Signed-off-by: Chen Yu <yu.c.chen@intel.com>
Cc: Prarit Bhargava <prarit@redhat.com>
Cc: Len Brown <lenb@kernel.orq>
Cc: "Rafael J. Wysocki" <rjw@rjwysocki.net>
Link: http://lkml.kernel.org/r/1492357410-510-1-git-send-email-yu.c.chen@intel.com
Signed-off-by: Thomas Gleixner <tglx@linutronix.de>
arch/x86/kernel/irq.c

index 4d8183b5f11323789f133de97bd32a6c8c403524..f34fe7444836ae51a2307973428db94d18846f84 100644 (file)
@@ -394,6 +394,9 @@ int check_irq_vectors_for_cpu_disable(void)
                    !cpumask_subset(&affinity_new, &online_new))
                        this_count++;
        }
+       /* No need to check any further. */
+       if (!this_count)
+               return 0;
 
        count = 0;
        for_each_online_cpu(cpu) {
@@ -411,8 +414,10 @@ int check_irq_vectors_for_cpu_disable(void)
                for (vector = FIRST_EXTERNAL_VECTOR;
                     vector < first_system_vector; vector++) {
                        if (!test_bit(vector, used_vectors) &&
-                           IS_ERR_OR_NULL(per_cpu(vector_irq, cpu)[vector]))
-                           count++;
+                           IS_ERR_OR_NULL(per_cpu(vector_irq, cpu)[vector])) {
+                               if (++count == this_count)
+                                       return 0;
+                       }
                }
        }