On Sat, Apr 13, 2019 at 10:00 AM Arnd Bergmann <arnd@xxxxxxxx> wrote:
>
> On Sat, Apr 13, 2019 at 12:07 AM Kirtika Ruchandani <kirtika@xxxxxxxxxx> wrote:
> >
> > On Fri, Apr 12, 2019 at 2:49 PM Ben Greear <greearb@xxxxxxxxxxxxxxx> wrote:
> > >
> > > On 4/12/19 2:43 PM, Kirtika Ruchandani wrote:
> > > > On Fri, Apr 12, 2019 at 2:40 PM <greearb@xxxxxxxxxxxxxxx> wrote:
> > > >>
> > > >> From: Ben Greear <greearb@xxxxxxxxxxxxxxx>
> > > >>
> > > >> This lets us more precisely calculate the absolute timestamp
> > > >> of last-rix (ie, now - idle).
> > > >
> > > > Can you use 64-bit timestamps? struct timeval suffers from the
> > > > overflow after 2038 problem.
> > >
> > > What is the preferred API to do this? Whatever it is, it would need
> > > to compile on old crufty systems as well.
> >
> > I am not sure what the guidance for userspace is. The kernel uses
> > 'struct timespec64' I think.
> > Arnd (cc-ed) who has mostly led the 2038 problem in the kernel might
> > have more input on the
> > "old crufty systems" part.
>
> I'm not sure what you are trying to do, and there are different
> answers depending on the usecase.
>
> For getting the time in the kernel, see Documentation/core-api/timekeeping.rst
> do_gettimeofday() is going away for many reasons, so don't use that.
> It sounds like you want "ktime_to_ms(ktime_get())" here.
>
> In userspace interfaces, you should pass 64-bit nanoseconds as returned
> by ktime_get_ns().
>
> If you want to pretty-print the current wall-clock, use the %pt format string
> on a 'struct rtc_time'.
Ah, I see now this was just userspace code. In that case, using gettimeofday()
works fine, it will end up using a 64-bit version of 'timeval', and converting
that to 64-bit milliseconds is safe.
Using clock_gettime() is generally preferred over gettimeofday() since it
avoids the conversion from nanoseconds to microseconds (which you then
convert to milliseconds).
Arnd
