PM-runtime: Switch accounting over to ktime_get_mono_fast_ns()
authorVincent Guittot <vincent.guittot@linaro.org>
Mon, 4 Feb 2019 16:25:52 +0000 (17:25 +0100)
committerRafael J. Wysocki <rafael.j.wysocki@intel.com>
Tue, 5 Feb 2019 11:04:39 +0000 (12:04 +0100)
Similar to what happened whith autosuspend, a deadlock has been
reported with PM-runtime accounting in the call path:

change_clocksource
    ...
    write_seqcount_begin
    ...
    timekeeping_update
        ...
        sh_cmt_clocksource_enable
            ...
            rpm_resume
                update_pm_runtime_accounting
                    ktime_get
                        do
                            read_seqcount_begin
                        while read_seqcount_retry
    ....
    write_seqcount_end

Make PM-runtime accounting use ktime_get_mono_fast_ns() to avoid this
problem.

With ktime_get_mono_fast_ns(), the timestamp is not guaranteed to be
monotonic across an update of timekeeping and as a result time can go
backward. Add a test to skip accounting for such situation which should
stay exceptional.

Fixes: a08c2a5a3194 ("PM-runtime: Replace jiffies-based accounting with ktime-based accounting")
Reported-by: Biju Das <biju.das@bp.renesas.com>
Signed-off-by: Vincent Guittot <vincent.guittot@linaro.org>
Reviewed-by: Ulf Hansson <ulf.hansson@linaro.org>
[ rjw: Subject, changelog, comment cleanup ]
Signed-off-by: Rafael J. Wysocki <rafael.j.wysocki@intel.com>
drivers/base/power/runtime.c

index 1caa3944898f201cec39a4da14896c61595c1069..6a58bb77a0181672b022f5197bcedecc6cf50597 100644 (file)
@@ -66,13 +66,22 @@ static int rpm_suspend(struct device *dev, int rpmflags);
  */
 void update_pm_runtime_accounting(struct device *dev)
 {
-       u64 now = ktime_to_ns(ktime_get());
+       u64 now = ktime_get_mono_fast_ns();
+       u64 last = dev->power.accounting_timestamp;
        u64 delta;
 
-       delta = now - dev->power.accounting_timestamp;
-
        dev->power.accounting_timestamp = now;
 
+       /*
+        * Because ktime_get_mono_fast_ns() is not monotonic during
+        * timekeeping updates, ensure that 'now' is after the last saved
+        * timesptamp.
+        */
+       if (now < last)
+               return;
+
+       delta = now - last;
+
        if (dev->power.disable_depth > 0)
                return;
 
@@ -1312,7 +1321,7 @@ void pm_runtime_enable(struct device *dev)
 
                /* About to enable runtime pm, set accounting_timestamp to now */
                if (!dev->power.disable_depth)
-                       dev->power.accounting_timestamp = ktime_to_ns(ktime_get());
+                       dev->power.accounting_timestamp = ktime_get_mono_fast_ns();
        } else {
                dev_warn(dev, "Unbalanced %s!\n", __func__);
        }