vmstat: get rid of the ugly cpu_stat_off variable
authorChristoph Lameter <cl@linux.com>
Fri, 20 May 2016 23:58:21 +0000 (16:58 -0700)
committerLinus Torvalds <torvalds@linux-foundation.org>
Sat, 21 May 2016 00:58:30 +0000 (17:58 -0700)
The cpu_stat_off variable is unecessary since we can check if a
workqueue request is pending otherwise.  Removal of cpu_stat_off makes
it pretty easy for the vmstat shepherd to ensure that the proper things
happen.

Removing the state also removes all races related to it.  Should a
workqueue not be scheduled as needed for vmstat_update then the shepherd
will notice and schedule it as needed.  Should a workqueue be
unecessarily scheduled then the vmstat updater will disable it.

[akpm@linux-foundation.org: fix indentation, per Michal]
Link: http://lkml.kernel.org/r/alpine.DEB.2.20.1605061306460.17934@east.gentwo.org
Signed-off-by: Christoph Lameter <cl@linux.com>
Cc: Tejun Heo <htejun@gmail.com>
Acked-by: Michal Hocko <mhocko@suse.com>
Cc: Tetsuo Handa <penguin-kernel@I-love.SAKURA.ne.jp>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
mm/vmstat.c

index 5b72a8ad281353223fe678e0ae8df354453303ca..77e42ef388c2a832169d4bcace8a505c45d95516 100644 (file)
@@ -1352,7 +1352,6 @@ static const struct file_operations proc_vmstat_file_operations = {
 static struct workqueue_struct *vmstat_wq;
 static DEFINE_PER_CPU(struct delayed_work, vmstat_work);
 int sysctl_stat_interval __read_mostly = HZ;
-static cpumask_var_t cpu_stat_off;
 
 #ifdef CONFIG_PROC_FS
 static void refresh_vm_stats(struct work_struct *work)
@@ -1421,24 +1420,10 @@ static void vmstat_update(struct work_struct *w)
                 * Counters were updated so we expect more updates
                 * to occur in the future. Keep on running the
                 * update worker thread.
-                * If we were marked on cpu_stat_off clear the flag
-                * so that vmstat_shepherd doesn't schedule us again.
                 */
-               if (!cpumask_test_and_clear_cpu(smp_processor_id(),
-                                               cpu_stat_off)) {
-                       queue_delayed_work_on(smp_processor_id(), vmstat_wq,
+               queue_delayed_work_on(smp_processor_id(), vmstat_wq,
                                this_cpu_ptr(&vmstat_work),
                                round_jiffies_relative(sysctl_stat_interval));
-               }
-       } else {
-               /*
-                * We did not update any counters so the app may be in
-                * a mode where it does not cause counter updates.
-                * We may be uselessly running vmstat_update.
-                * Defer the checking for differentials to the
-                * shepherd thread on a different processor.
-                */
-               cpumask_set_cpu(smp_processor_id(), cpu_stat_off);
        }
 }
 
@@ -1470,16 +1455,17 @@ static bool need_update(int cpu)
        return false;
 }
 
+/*
+ * Switch off vmstat processing and then fold all the remaining differentials
+ * until the diffs stay at zero. The function is used by NOHZ and can only be
+ * invoked when tick processing is not active.
+ */
 void quiet_vmstat(void)
 {
        if (system_state != SYSTEM_RUNNING)
                return;
 
-       /*
-        * If we are already in hands of the shepherd then there
-        * is nothing for us to do here.
-        */
-       if (cpumask_test_and_set_cpu(smp_processor_id(), cpu_stat_off))
+       if (!delayed_work_pending(this_cpu_ptr(&vmstat_work)))
                return;
 
        if (!need_update(smp_processor_id()))
@@ -1494,7 +1480,6 @@ void quiet_vmstat(void)
        refresh_cpu_vm_stats(false);
 }
 
-
 /*
  * Shepherd worker thread that checks the
  * differentials of processors that have their worker
@@ -1511,20 +1496,11 @@ static void vmstat_shepherd(struct work_struct *w)
 
        get_online_cpus();
        /* Check processors whose vmstat worker threads have been disabled */
-       for_each_cpu(cpu, cpu_stat_off) {
+       for_each_online_cpu(cpu) {
                struct delayed_work *dw = &per_cpu(vmstat_work, cpu);
 
-               if (need_update(cpu)) {
-                       if (cpumask_test_and_clear_cpu(cpu, cpu_stat_off))
-                               queue_delayed_work_on(cpu, vmstat_wq, dw, 0);
-               } else {
-                       /*
-                        * Cancel the work if quiet_vmstat has put this
-                        * cpu on cpu_stat_off because the work item might
-                        * be still scheduled
-                        */
-                       cancel_delayed_work(dw);
-               }
+               if (!delayed_work_pending(dw) && need_update(cpu))
+                       queue_delayed_work_on(cpu, vmstat_wq, dw, 0);
        }
        put_online_cpus();
 
@@ -1540,10 +1516,6 @@ static void __init start_shepherd_timer(void)
                INIT_DEFERRABLE_WORK(per_cpu_ptr(&vmstat_work, cpu),
                        vmstat_update);
 
-       if (!alloc_cpumask_var(&cpu_stat_off, GFP_KERNEL))
-               BUG();
-       cpumask_copy(cpu_stat_off, cpu_online_mask);
-
        vmstat_wq = alloc_workqueue("vmstat", WQ_FREEZABLE|WQ_MEM_RECLAIM, 0);
        schedule_delayed_work(&shepherd,
                round_jiffies_relative(sysctl_stat_interval));
@@ -1578,16 +1550,13 @@ static int vmstat_cpuup_callback(struct notifier_block *nfb,
        case CPU_ONLINE_FROZEN:
                refresh_zone_stat_thresholds();
                node_set_state(cpu_to_node(cpu), N_CPU);
-               cpumask_set_cpu(cpu, cpu_stat_off);
                break;
        case CPU_DOWN_PREPARE:
        case CPU_DOWN_PREPARE_FROZEN:
                cancel_delayed_work_sync(&per_cpu(vmstat_work, cpu));
-               cpumask_clear_cpu(cpu, cpu_stat_off);
                break;
        case CPU_DOWN_FAILED:
        case CPU_DOWN_FAILED_FROZEN:
-               cpumask_set_cpu(cpu, cpu_stat_off);
                break;
        case CPU_DEAD:
        case CPU_DEAD_FROZEN: