[PATCH xserver] xwayland: Make it possible to force CLOCK_MONOTONIC

[Daniel Stone]

Hi,

On 9 March 2016 at 04:27, Jonas Ådahl <jadahl at gmail.com> wrote:
> By default the X server will try CLOCK_MONOTONIC_COARSE before
> CLOCK_MONOTONIC. This causes various issues for Wayland display
> servers who may get their internal timestamps from the CLOCK_MONOTONIC,
> since it may very well happen that a timestamp from CLOCK_MONOTONIC
> retrieved before a sending an X request will still be "later" than the
> timestamp the X server than gets after receiving the request, due to
> the fact that CLOCK_MONOTONIC_COARSE has a lower resolution.

For reference, this isn't anything to do with Wayland core protocol or
similar, but that evdev devices can provide events with
CLOCK_MONOTONIC timestamps, which we try to match.

The original reason for using CLOCK_MONOTONIC_COARSE over
CLOCK_MONOTONIC is that the former had a runtime power impact by
battering HPET. I don't know if that still holds up on modern systems.

> FWIW, I would be just as fine with always forcing CLOCK_MONOTONIC for Xwayland
> if the command line argument seems unnecessary.

This one is :
Reviewed-by: Daniel Stone <daniels at collabora.com>

and just forcing Xwayland to do it unconditionally is:
Acked-by: Daniel Stone <daniels at collabora.com>

> @@ -86,6 +88,10 @@ ddxProcessArgument(int argc, char *argv[], int i)
>      else if (strcmp(argv[i], "-shm") == 0) {
>          return 1;
>      }
> +    else if (strcmp(argv[i], "-forceMonotonic") == 0) {
> +        ForceClockId(CLOCK_MONOTONIC);
> +        return 1;
> +    }

Needs #ifdef MONOTONIC_CLOCK.

> +#ifdef MONOTONIC_CLOCK
> +static clockid_t clockid;
> +#endif
> +
>  OsSigHandlerPtr
>  OsSignal(int sig, OsSigHandlerPtr handler)
>  {
> @@ -427,6 +431,26 @@ GiveUp(int sig)
>      errno = olderrno;
>  }
>
> +#ifdef MONOTONIC_CLOCK
> +void
> +ForceClockId(clockid_t forced_clockid)
> +{
> +    struct timespec tp;
> +
> +    if (clockid) {
> +        ErrorF("Warning: clock id was forced after it was initialized.\n");

Turn this into a BUG(), especially as the one below is a FatalError ...

Cheers,
Daniel