In a similar spirit to
commit d75e8146c4
Author: Keith Packard <keithp@keithp.com>
Date: Mon Jul 12 16:01:34 2010 -0700
Unwrap/rewrap EnterVT/LeaveVT completely, Fixes 28998
we need to unwrap our pScrn->EnterVT/LeaveVT hooks around server
regeneration or else we cause an infinite recursion on the next VT
switch afterwards.
Bugzilla: https://bugs.launchpad.net/ubuntu/+source/xserver-xorg-video-intel/+bug/1235516
Signed-off-by: Chris Wilson <chris@chris-wilson.co.uk>
Reviewed-by: Keith Packard <keithp@keithp.com>
Signed-off-by: Keith Packard <keithp@keithp.com>
(I found an amended version of this patch and applied the difference
here - keithp)
v3: Don't call Xsync before restoring error handler as any errors
generated by XGetImage() should be processed when this call
returns as suggested by Jamey Sharp <jamey@minilop.net>
Signed-off-by: Egbert Eich <eich@freedesktop.org>
Reviewed-by: Jamey Sharp <jamey@minilop.net>
unflip happens after the clip lists have been updated, so instead of
smashing the whole screen and drawing over other windows, just draw to
the original flip window; it'll have the right clip list and so the
copy will work just fine.
Signed-off-by: Keith Packard <keithp@keithp.com>
Reviewed-by: Adam Jackson <ajax@redhat.com>
Pend presentation until wait_fence is also triggered by having the
SyncFence trigger invoke present_execute once triggered.
Signed-off-by: Keith Packard <keithp@keithp.com>
Reviewed-by: Adam Jackson <ajax@redhat.com>
This eliminates dereferencing freed window pointers when there is a
flip for that window in progress. The flip will complete, and then
immediately get undone (as we can't stop an in-progress flip).
Remove the vblank->window_destroyed field as we can signal this with
vblank->window == NULL instead.
Change check to vblank->window == NULL in:
present_flip_notify
Add check for vblank->window == NULL in:
present_vblank_notify
present_execute
present_flip_notify was also using vblank->window->drawable.pScreen,
so stop doing that and use vblank->screen instead.
Signed-off-by: Keith Packard <keithp@keithp.com>
Reviewed-by: Adam Jackson <ajax@redhat.com>
We use event_id 0 to mean 'no such event'; if a driver sends us that
event_id, make sure we don't accidentally match it.
Signed-off-by: Keith Packard <keithp@keithp.com>
Reviewed-by: Adam Jackson <ajax@redhat.com>
If the timer fired too early, we'd sometimes mis-compute the MSC for
fake vblanks. Rounding the computation to the nearest MSC fixes this nicely.
Signed-off-by: Keith Packard <keithp@keithp.com>
Reviewed-by: Adam Jackson <ajax@redhat.com>
Signed-off-by: Peter Hutterer <peter.hutterer@who-t.net>
Reviewed-by: Keith Packard <keithp@keithp.com>
Signed-off-by: Keith Packard <keithp@keithp.com>
Based on some bugzilla scraping I did around November 2012. Of xserver
bugs in Red Hat bugzilla with an EQ size message in the log, the
distribution looked like:
String | Matches
-------------------------------------
Increasing EQ size to 512 | 460
Increasing EQ size to 1024 | 52
Increasing EQ size to 2048 | 6
Increasing EQ size to 4096 | 0
Most of the "512" ones appear to be mostly harmless, some relatively
expensive path in either rendering or resource destruction simply taking
too long due to external pressures like paging or CPU contention. So
let's raise the initial queue size, both to reduce the number of
spurious abrt reports and to drop fewer events in all but the most
pathological cases.
Signed-off-by: Adam Jackson <ajax@redhat.com>
Reviewed-by: Jasper St. Pierre <jstpierre@mecheye.net>
Signed-off-by: Keith Packard <keithp@keithp.com>
With outputless GPUs showing up we crash here if there are not outputs
try and recover with a bit of grace.
Reviewed-by: Adam Jackson <ajax@redhat.com>
Signed-off-by: Dave Airlie <airlied@redhat.com>
Signed-off-by: Keith Packard <keithp@keithp.com>
VNC needs key_is_down to check if a key is processed as down before it
simulates various key releases. Make it available, because I seriously can't
be bothered thinking about how to rewrite VNC to not need that.
Signed-off-by: Peter Hutterer <peter.hutterer@who-t.net>
Acked-by: Keith Packard <keithp@keithp.com>
Since all the inb/outb/etc. use in the X server itself (except for
xf86SlowBcopy) has been replaced by calls to libpciaccess, we no
longer need to pass inline assembly files to replace the gcc inline
assembly from hw/xfree86/common/compiler.h when building Xorg itself.
The .il files are still generated and installed in the SDK for the
benefit of drivers who may use them.
Binary diff of before and after showed that xf86SlowBcopy was the
only function changed across the Xorg binary and all modules built
in the Xserver build, it just calls the outb() function now instead
of having the outb instructions inlined, making it a slightly slower
bcopy.
Signed-off-by: Alan Coopersmith <alan.coopersmith@oracle.com>
Reviewed-by: Keith Packard <keithp@keithp.com>
Signed-off-by: Keith Packard <keithp@keithp.com>
This patch fixes cursor jumps when there is a grab on the Xephyr window and
the pointer moves outside the window.
So on two side-by-side 640x480 screens, a coordinate of 0/481
triggers KdCursorOffscreen.
If the delta between two screens is 0, they share the same offset for
that dimension. When searching for the new screen, the loop always rules out
the current screen. So we get to the second screen, trigger the conditions
where dy <= 0 and decide that this new screen is the correct one. The result
is that whenever KdCursorOffScreen is called, the pointer jumps to the other
screen.
Change to check for dy < 0 etc. so that the cursor stays on the same screen if
there is no other screen at the target location.
Signed-off-by: Peter Hutterer <peter.hutterer@who-t.net>
Reviewed-by: Keith Packard <keithp@keithp.com>
A multi-head Xephyr instance has the pointer stuck on one screen
because of bad coordinate calculation. The coordinates passed to
GetPointerEvents are per-screen, so the cursor gets stuck on the left-most
screen by default.
Adjust and mark the events as POINTER_DESKTOP, so the DIX
can adjust them accordingly.
Signed-off-by: Peter Hutterer <peter.hutterer@who-t.net>
Reviewed-by: Keith Packard <keithp@keithp.com>
If a screen size was specified as WxH, the loop returned early and kdOrigin
was never advanced. Thus, screen->origin was always 0 (or whatever was given
at the -origin commandline flag).
If a screen size was given with a bit depth (WxHxD), kdOrigin would always
advance by the current screen, offsetting the next screen.
Signed-off-by: Keith Packard <keithp@keithp.com>
Reviewed-by: Peter Hutterer <peter.hutterer@who-t.net>
Signed-off-by: Peter Hutterer <peter.hutterer@who-t.net>
shmint.h is part of sdk_HEADERS, and so can't use anything not
included in sdk_HEADERS.
busfault.h includes dix-config.h which is not. Leave the use of
struct busfault in shmint.h and move the include of busfault.h to
shm.c.
protocol-versions.h is not part of sdk_HEADERS, so instead of using
that, just use XTRANS_SEND_FDS to choose whether to expose the fd
passing requests directly.
Reported-by: Julien Cristau <jcristau@debian.org>
Signed-off-by: Keith Packard <keithp@keithp.com>
Tested-by: Knut Petersen <Knut_Petersen@t-online.de>
Reviewed-by: Julien Cristau <jcristau@debian.org>
v2: also avoid using protocol-versions.h
Requires passing through the __EXTENSIONS__ and _XOPEN_SOURCE defines
in order to expose the msg_control members in struct msghdr.
Signed-off-by: Alan Coopersmith <alan.coopersmith@oracle.com>
Reviewed-by: Adam Jackson <ajax@redhat.com>
When building on Solaris with _XOPEN_SOURCE set to a recent XPG release,
<stdlib.h> and other core headers start including <sys/regset.h>, which
has a bunch of unfortunately named macros such as "CS", "ES", etc. for
x86 & x64 registers which clash with existing variable & struct member
names in Xorg - so #undef these so they don't interfere with our use.
(Yes, have filed a bug against the system headers for exposing these,
but this solves the problem for building on existing releases.)
Signed-off-by: Alan Coopersmith <alan.coopersmith@oracle.com>
Reviewed-by: Adam Jackson <ajax@redhat.com>
The former doesn't exist on BSD and the latter is available everywhere
AFAIK (checked Solaris and Linux).
You also might want to wrap that line ;).
Reported-by: Mark Kettenis <mark.kettenis@xs4all.nl>
Signed-off-by: Keith Packard <keithp@keithp.com>
req_fds and SetReqFds in include/dixstruct.h
ReadFdFromClient, WriteFdToClient and the FD flushing in os/io.c
Signed-off-by: Keith Packard <keithp@keithp.com>
There's a --disable-present, so it'd be nice if it worked.
Signed-off-by: Adam Jackson <ajax@redhat.com>
Reviewed-by: Jasper St. Pierre <jstpierre@mecheye.net>
Reviewed-by: Mark Kettenis <kettenis@openbsd.org>
Signed-off-by: Keith Packard <keithp@keithp.com>
A client destroying objects in the middle of an unflip can end up
having the screen flip window or fence set to NULL in the unflip
notify path. Check for these and don't try to use those objects.
Signed-off-by: Keith Packard <keithp@keithp.com>
Reviewed-by: Adam Jackson <ajax@redhat.com>
If a client passes a section of memory via file descriptor and then
subsequently truncates that file, the underlying pages will be freed
and the addresses invalidated. Subsequent accesses to the page will
fail with a SIGBUS error.
Trap that SIGBUS, figure out which segment was causing the error and
then allocate new pages to fill in for that region. Mark the offending
shared segment as invalid and free the resource ID so that the client
will be able to tell when subsequently attempting to use the segment.
Signed-off-by: Keith Packard <keithp@keithp.com>
v2: Use MAP_FIXED to simplify the recovery logic (Mark Kettenis)
v3: Also catch errors in ShmCreateSegment
Conflicts:
include/dix-config.h.in
include/xorg-config.h.in
Check to see if xtrans FD passing is available and use that to
advertise the appropriate version of the SHM extension
Signed-off-by: Keith Packard <keithp@keithp.com>
Reviewed-by: Alan Coopersmith <alan.coopersmith@oracle.com>
This includes the MIT-SHM FD passing requests
Signed-off-by: Keith Packard <keithp@keithp.com>
Reviewed-by: Alan Coopersmith <alan.coopersmith@oracle.com>
Until other operating systems have a libXtrans port for FD passing,
disable this on non-Linux systems.
Note that this define affects how libXtrans gets built into the X
server, which is why it need only define the symbol
Signed-off-by: Keith Packard <keithp@keithp.com>
The GLX section of configura.ac checks the state of DRI2, so it
needs to be after DRI2=auto is resolved.
Also reset libgl requirement to 7.1.0 in non-dri2 case.
Signed-off-by: Jeremy Huddleston Sequoia <jeremyhu@apple.com>
Reviewed-by: Adam Jackson <ajax@redhat.com>
Signed-off-by: Andreas Schwab <schwab@suse.de>
Reviewed-by: Mark Kettenis <kettenis@openbsd.org>
Reviewed-by: Adam Jackson <ajax@redhat.com>
Signed-off-by: Keith Packard <keithp@keithp.com>
This gets the server to link with xshmfence again, and also ensures
that the miSyncShm code is linked into the server with the reference
from sdksyms.
Signed-off-by: Keith Packard <keithp@keithp.com>
The selection of which clock to use for this function was not actually
getting used when fetching the final clock value.
Reported-by: Julien Cristau <jcristau@debian.org>
Signed-off-by: Keith Packard <keithp@keithp.com>
APM support in the Xserver was used to restore the console mode
prior to a power management event. This was to ensure the mode
upon suspend/resume was one that the system firmware or kernel
could deal with.
APM support is now largely obsolete, KMS drivers don't require a
mode restoration anyhow. Therefore it should be possible to disable
this feature.
(small modification by keithp - move test for XF86PM flag after check
for APM, then move XF86PM flag to xorg-config.h.in)
Signed-off-by: Egbert Eich <eich@freedesktop.org>
Tested-by: Keith Packard <keithp@keithp.com>
Signed-off-by: Keith Packard <keithp@keithp.com>
ProcessInputEvent() resets the device idle times. If idle time was higher than
the lower bracket, this should trigger an event in the idle time wakeup
handler.
If processing is slow, the idle time may advance past the lower bracket
between the reset and the time the BlockHandler is called. In that case, we'd
never schedule a wakeup to handle the event, causing us to randomly miss
events.
Ran tests with a neg transition trigger on 5ms with 200 repeats of the test
and it succeeded. Anything below that gets a bit tricky to make sure the
server sees the same idle time as the client usleeps for.
Signed-off-by: Peter Hutterer <peter.hutterer@who-t.net>
Reviewed-by: Keith Packard <keithp@keithp.com>
ACPI support in the Xserver was used to restore the console mode
prior to a power management event. This was to ensure the mode
upon suspend/resume was one that the system firmware or kernel
could deal with.
The feature depended on acpid to be running. Most of this functionality
is now take over by systemd, KMS drivers don't require a mode restoration
anyhow. Therefore it should be possible to disable this feature under
some circumstances.
Tested-by: Keith Packard <keithp@keithp.com>
Signed-off-by: Keith Packard <keithp@keithp.com>