Re: [RFC][PATCH] Dynamic Tick: Allow 32-bit machines to sleep formorethan2.15 seconds

[Jon Hunter <jon-hunter@ti.com>]

John Stultz wrote:
>>>> One final question, I noticed in clocksource.h that the definition of 
>>>> function cyc2ns returns a type of s64, however, in the function itself a 
>>>> variable of type u64 is used and returned. Should this function be 
>>>> modified as follows?
>>>>
>>>>   static inline s64 cyc2ns(struct clocksource *cs, cycle_t cycles)
>>>>   {
>>>> -       u64 ret = (u64)cycles;
>>>> +       s64 ret = (s64)cycles;
>>>>          ret = (ret * cs->mult) >> cs->shift;
>>>>          return ret;
>>>>   }
>>> Damn. So this brings up an issue I had missed prior.
>> Any comments on whether this should be u64 versus s64?
> 
> I'd leave it alone for now. I'm concerns that in large multiplies, if
> its a s64 the sign might get extended down by the shift. I need to look
> at it in more detail though.

I have been thinking about this some more and I do agree that there is a 
chance that the multiply could overflow if the "cycles" and "mult" are 
large. From the perspective of the timekeeping_max_deferment() function 
this would be very likely for 64-bit clocksources when the mask will be 
equal to (2^64)-1. Therefore, how about modifying the function as 
follows in order to catch any occurrences of overflow?

Let me know if this is aligned with your thinking or if I am barking up 
the wrong tree here.

Cheers
Jon


diff --git a/include/linux/clocksource.h b/include/linux/clocksource.h
index 507235a..8204373 100644
--- a/include/linux/clocksource.h
+++ b/include/linux/clocksource.h
@@ -316,8 +316,32 @@ static inline void clocksource_disable(struct 
clocksource *cs)
   */
  static inline s64 cyc2ns(struct clocksource *cs, cycle_t cycles)
  {
-	s64 ret = (s64)cycles;
-	ret = (ret * cs->mult) >> cs->shift;
+	s64 ret;
+	u64 upper, lower, overflow;
+
+	/*
+	 * Split the calculation into two halves to ensure
+	 * that we can catch any overflow that may occur.
+	 */
+	upper = ((cycles >> 32) * cs->mult) >> cs->shift;
+	lower = ((cycles & 0xFFFFFFFF) * cs->mult) >> cs->shift;
+
+	/*
+	 * Check to see if the result will overflow. If
+	 * overflow is non-zero then the result is greater
+	 * than 63-bits which is the max positive value
+	 * for a signed result.
+	 */
+	overflow = (upper + (lower >> 32)) >> 31;
+
+	/*
+	 * If the result overflows, return the max value we can.
+	 */
+	if (overflow)
+		ret = LONG_MAX;
+	else
+		ret = (s64)((upper << 32) + lower);
+
  	return ret;
  }

-- 
1.6.1