GdkDisplayX11: Properly translate server timestamps from _NET_WM_FRAME_* messages

When using frame times from _NET_WM_FRAME_DRAWN and _NET_WM_FRAME_TIMINGS, we
were treating them as local monotonic times, but they are actually extended-precision
versions of the server time, and need to be translated to monotonic times in the
case where the X server and client aren't running on the same system.

This fixes rendering stalls when using X over a remote ssh connection.
https://bugzilla.gnome.org/show_bug.cgi?id=741800
This commit is contained in:
Owen W. Taylor 2015-07-15 12:38:38 -04:00 committed by Matthias Clasen
parent 24e1323eb3
commit 6504b2e534
2 changed files with 48 additions and 1 deletions

View File

@ -1108,6 +1108,47 @@ find_frame_timings (GdkFrameClock *clock,
return NULL;
}
/* _NET_WM_FRAME_DRAWN and _NET_WM_FRAME_TIMINGS messages represent time
* as a "high resolution server time" - this is the server time interpolated
* to microsecond resolution. The advantage of this time representation
* is that if X server is running on the same computer as a client, and
* the Xserver uses 'clock_gettime(CLOCK_MONOTONIC, ...)' for the server
* time, the client can detect this, and all such clients will share a
* a time representation with high accuracy. If there is not a common
* time source, then the time synchronization will be less accurate.
*/
gint64
server_time_to_monotonic_time (GdkX11Display *display_x11,
gint64 server_time)
{
if (display_x11->server_time_query_time == 0 ||
(!display_x11->server_time_is_monotonic_time &&
server_time > display_x11->server_time_query_time + 10*1000*1000)) /* 10 seconds */
{
gint64 current_server_time = gdk_x11_get_server_time (display_x11->leader_gdk_window);
gint64 current_server_time_usec = (gint64)current_server_time * 1000;
gint64 current_monotonic_time = g_get_monotonic_time ();
display_x11->server_time_query_time = current_monotonic_time;
/* If the server time is within a second of the monotonic time,
* we assume that they are identical. This seems like a big margin,
* but we want to be as robust as possible even if the system
* is under load and our processing of the server response is
* delayed.
*/
if (current_server_time_usec > current_monotonic_time - 1000*1000 &&
current_server_time_usec < current_monotonic_time + 1000*1000)
display_x11->server_time_is_monotonic_time = TRUE;
display_x11->server_time_offset = current_server_time_usec - current_monotonic_time;
}
if (display_x11->server_time_is_monotonic_time)
return server_time;
else
return server_time - display_x11->server_time_offset;
}
GdkFilterReturn
_gdk_wm_protocols_filter (GdkXEvent *xev,
GdkEvent *event,
@ -1140,7 +1181,7 @@ _gdk_wm_protocols_filter (GdkXEvent *xev,
guint32 d3 = xevent->xclient.data.l[3];
guint64 serial = ((guint64)d1 << 32) | d0;
gint64 frame_drawn_time = ((guint64)d3 << 32) | d2;
gint64 frame_drawn_time = server_time_to_monotonic_time (GDK_X11_DISPLAY (display), ((guint64)d3 << 32) | d2);
gint64 refresh_interval, presentation_time;
GdkFrameClock *clock = gdk_window_get_frame_clock (win);

View File

@ -131,6 +131,12 @@ struct _GdkX11Display
gint glx_error_base;
gint glx_event_base;
/* Translation between X server time and system-local monotonic time */
gint64 server_time_query_time;
gint64 server_time_offset;
guint server_time_is_monotonic_time : 1;
guint have_glx : 1;
/* GLX extensions we check */