On Thursday 05 November 2015 10:05:44 Hans Verkuil wrote: > On 11/05/15 09:36, Arnd Bergmann wrote: > > On Thursday 05 November 2015 08:41:11 Hans Verkuil wrote: > >> Hi Arnd, > >> > >> We're redesigning the timestamp handling in the video4linux subsystem moving away > >> from struct timeval to a single timestamp in ns (what ktime_get_ns() gives us). > >> But I was wondering: ktime_get_ns() gives a s64, so should we use s64 as well as > >> the timestamp type we'll eventually be returning to the user, or should we use u64? > >> > >> The current patch series we made uses a u64, but I am now beginning to doubt that > >> decision. > > > > I would lean towards u64, but I don't think it really matters either way, > > especially since all the drivers should be using monotonic timestamps now. > > One thing that might be easier if it is a s64 is when adding/subtracting offsets > from the timestamp. And the other reason is that a u64 gives a false view of the > underlying type. With a s64 it is clear that a timestamp will wrap around after > 292 years instead of double that. Admittedly, not our problem, but if we ever send > a space probe to Alpha Centauri, then it might be nice to know as application > developer that you need to take special measures if the mission takes longer than > 292 years There is another problem here if you are worried about the overflow: Unsigned overflow is well-defined in C, i.e. (U64_MAX + 1) is known to be 0. However, signed overflow is not defined in C for historic reasons, so taking the result of (S64_MAX + 1) can have arbitrary results [1], including a kernel oops, or random other things not related to the variable that carries the result, if gcc decides that 'this cannot happen'. Arnd [1] http://stackoverflow.com/questions/3948479/integer-overflow-and-undefined-behavior -- To unsubscribe from this list: send the line "unsubscribe linux-media" in the body of a message to majordomo@xxxxxxxxxxxxxxx More majordomo info at http://vger.kernel.org/majordomo-info.html
