Skip to contentSkip to conversation

time: Fix sleeptime injection for non-stop clocksource & persistent clock

PATCH RESEND Superseded Mukesh Ojha 2 participants 3 messages May 29, 2018 – May 30, 2018 Timekeeping, clocksource core, NTP, alarmtimer

v1 of 2 Superseded Latest is v2 →

lore
Share & tools
lore
#0 lore raw

PATCH RESEND 1/1 time: Fix sleeptime injection for non-stop clocksource & persistent clock

Currently, for both non-stop clocksource and persistent clock there is a corner case, when a driver failed to go suspend mode rtc_resume() injects the sleeptime as timekeeping_rtc_skipresume() returned 'false' due to which we can see mismatch in time between system clock and other timers.

Success case:

                                         {sleeptime_injected=true}

rtc_suspend() => timekeeping_suspend() => timekeeping_resume() =>
rtc_resume()

Failure case:

             {failure in sleep path} {sleeptime_injected=false}

rtc_suspend() => rtc_resume()

---

Changes from Resend:
* updated commit text.

 kernel/time/timekeeping.c | 14 +++++++++-----
 1 file changed, 9 insertions(+), 5 deletions(-)
kernel/time/timekeeping.c +9 -5
index 427e33d..c022d82 100644
@@ -1511,9 +1511,6 @@ void __weak read_boot_clock64(struct timespec64 *ts)
 	ts->tv_nsec = 0;
 }
 
-/* Flag for if timekeeping_resume() has injected sleeptime */
-static bool sleeptime_injected;
-
 /* Flag for if there is a persistent clock on this platform */
 static bool persistent_clock_exists;
 
@@ -1611,7 +1608,14 @@ static void __timekeeping_inject_sleeptime(struct timekeeper *tk,
  */
 bool timekeeping_rtc_skipresume(void)
 {
-	return sleeptime_injected;
+	struct timekeeper *tk = &tk_core.timekeeper;
+	bool skip_rtc_resume = false;
+
+	skip_rtc_resume = ((tk->tkr_mono.clock->flags &
+			CLOCK_SOURCE_SUSPEND_NONSTOP) ||
+			(persistent_clock_exists)) ? true : false;
+
+	return skip_rtc_resume;
 }
 
 /**
@@ -1671,8 +1675,8 @@ void timekeeping_resume(void)
 	unsigned long flags;
 	struct timespec64 ts_new, ts_delta;
 	cycle_t cycle_now, cycle_delta;
+	bool sleeptime_injected = false;
 
-	sleeptime_injected = false;
 	read_persistent_clock64(&ts_new);
 
 	clockevents_resume();
Signature · 2 lines
Qualcomm India Private Limited, on behalf of Qualcomm Innovation Center,
Inc. is a member of the Code Aurora Forum, a Linux Foundation Collaborative Project
#1 lore raw
Quoted 8 lines · Mukesh Ojha

Currently, for both non-stop clocksource and persistent clock there is a corner case, when a driver failed to go suspend mode rtc_resume() injects the sleeptime as timekeeping_rtc_skipresume() returned 'false' due to which we can see mismatch in time between system clock and other timers.

Success case:

                                         {sleeptime_injected=true}
Quoted 8 lines

rtc_suspend() => timekeeping_suspend() => timekeeping_resume() =>
rtc_resume()

Failure case:

             {failure in sleep path} {sleeptime_injected=false}

rtc_suspend() => rtc_resume()

I'm not sure this patch makes sense yet (since I don't really see how its used - mind cc'ing me on the patch that makes use of this?)

And more problematic, the patch doesn't seem to apply to mainline. Could you respin and resend?

thanks
-john

#2 lore raw

Hi John,

Quoted 12 lines · John Stultz
Quoted 11 lines · Mukesh Ojha

Currently, for both non-stop clocksource and persistent clock there is a corner case, when a driver failed to go suspend mode rtc_resume() injects the sleeptime as timekeeping_rtc_skipresume() returned 'false' due to which we can see mismatch in time between system clock and other timers.

Success case:

                                          {sleeptime_injected=true}

rtc_suspend() => timekeeping_suspend() => timekeeping_resume() =>
rtc_resume()

Quoted 7 lines
Quoted 5 lines

Failure case:

              {failure in sleep path} {sleeptime_injected=false}

rtc_suspend() => rtc_resume()

I'm not sure this patch makes sense yet (since I don't really see how its used

In a system where both non-stop clocksource and rtc exist

Timestamp mismatch seen between a (Gyro)timer based on non-stop clocksource and system clock after attempting a sleep.
Here, System clock timestamps increase by some millisecond as compare to other timer.

And we see this is possible only when timekeeping_resume() has not been called which is keeping sleeptime_injected to its default value(false). So, the call to timekeeping_rtc_skipresume() in rtc_resume() is false and thats ends up in sleeptime injection via rtc_resume().

When timekeeping_resume() has not been called means one of the driver failed during suspend in syscore_suspend() in kernel/power/suspend.c .

So if we let rtc inject the sleeptime here, we would see the timestamp increase in case of system clock.

Quoted 4 lines

- mind cc'ing me on the patch that makes use of this?)

And more problematic, the patch doesn't seem to apply to mainline. Could you respin and resend?

Will spin the patch again with some minor code change.

Thanks,
Mukesh

Quoted 3 lines

thanks
-john