When closing down all devices, we manually unset master for all attached
devices, but the device's sprite info still points to the master's sprite
info. This leaves us a window where the master is freed already but the
device isn't yet. A signal during that window causes dereference of the
already freed spriteInfo in mieqEnqueue's EnqueueScreen macro.
Simply block signals when removing all devices. It's not like we're really
worrying about high-responsive input at this stage.
https://bugzilla.redhat.com/show_bug.cgi?id=737031
Signed-off-by: Peter Hutterer <peter.hutterer@who-t.net>
Reviewed-by: Julien Cristau <jcristau@debian.org>
(cherry picked from commit d7c44a7c97)
https://bugs.freedesktop.org/show_bug.cgi?id=38420
Exit with fatal error message, not segfault.
Signed-off-by: Alan Coopersmith <alan.coopersmith@oracle.com>
Reviewed-by: Jeremy Huddleston <jeremyhu@apple.com>
Signed-off-by: Keith Packard <keithp@keithp.com>
(cherry picked from commit 7d50211ab5)
If you started an X server with no connected outputs, we pick a default
1024x768 mode, however if you then ran an xvidmode using app against that
server it would segfault the server due to not finding any valid modes.
This was due to the no output mode set code, only adding the modes to the
scrn->modes once, when something called randr 1.2 xf86SetScrnInfoModes would
get called and remove all the modes and we'd end up with 0.
This change fixes xf86SetScrnInfoModes to always report a scrn mode of at
least 1024x768, and pushes the initial configuration to just call it instead
of setting up the mode itself.
Fixes: https://bugzilla.redhat.com/show_bug.cgi?id=746926
I've seen other bugs like this on other distros so it might also actually fix them.
Signed-off-by: Dave Airlie <airlied@redhat.com>
Reviewed-by: Keith Packard <keithp@keithp.com>
Signed-off-by: Keith Packard <keithp@keithp.com>
(cherry picked from commit 17416e88dc)
Even though it's only valid when local, it is possible for a local
client and the server to not match endianness, such as when running
a ppc application under Rosetta.
Signed-off-by: Jeremy Huddleston <jeremyhu@apple.com>
(cherry picked from commit 1c8bda798b)
Even though it's only valid when local, it is possible for a local
client and the server to not match endianness, such as when running
a ppc application under Rosetta.
Signed-off-by: Jeremy Huddleston <jeremyhu@apple.com>
Reviewed-by: Jamey Sharp <jamey@minilop.net>
(cherry picked from commit 14205ade0c)
Conflicts:
hw/xquartz/xpr/appledri.c
Signed-off-by: Jeremy Huddleston <jeremyhu@apple.com>
This was a regression.
Introduced by: 08363c5830 and
32db27a7f8
Masked by: 1e69fd4a60
Signed-off-by: Jeremy Huddleston <jeremyhu@apple.com>
Reviewed-by: Jamey Sharp <jamey@minilop.net>
(cherry picked from commit 83fef4235d)
Bug introduced by 9dca441670
xfree86: add a hook to replace the new console handler.
console_handler was not being set, making the server eat up CPU spinning
in WaitForSomething selecting consoleFd over and over again, every time
trying to unregister drain_console without success due to
console_handler being NULL.
Let's just fix the unregistration in xf86SetConsoleHandler() and use that.
But wait, there could be a catch: If some driver replaced the handler using
xf86SetConsoleHandler(), the unregistration in xf86CloseConsole will unregister
that one. I don't understand Xorg well enough to know whether this poses a
problem (could mess up driver deinit somehow or something like that). As it is,
xf86SetConsoleHandler() doesn't offer any way to prevent this (i.e. check which
handler is currently registered).
I had been using it for two days on my machine that previously hit 100% CPU
several times a day. That has now gone away without any new problems appearing.
Signed-off-by: Tomas Trnka <tomastrnka@gmx.com>
Reviewed-by: Peter Hutterer <peter.hutterer@who-t.net>
Signed-off-by: Peter Hutterer <peter.hutterer@who-t.net>
(cherry picked from commit 323869f329)
It's fairly common to have multiple, identical monitors plugged in. In
that case, it's preferable to run the monitor's preferred mode on each
output, rather than just matching the width & height and end up with
different timings or refresh rates.
Signed-off-by: Jesse Barnes <jbarnes@virtuousgeek.org>
Reviewed-by: Keith Packard <keithp@keithp.com>
Signed-off-by: Keith Packard <keithp@keithp.com>
(cherry picked from commit 3e145d3d67)
Previously the parent constrained borderClip was copied over
when compRedirectWindow() is called. That is insufficient eg. in
case the window was already redirected, but not yet realized. So
copy the borderClip over in compAllocPixmap() instead.
Example:
Window 1 is below an automatically redirect window 2. Window 2 is
unmapped and moved outside the extents of window 1. Window 2 is
then mapped again, and MarkOverlappedWindows() uses the up to
date borderSize of window 2 to mark windows, which leaves
window 1 unmarked. Then exposures are calculated using the stale
borderClip of window 2, which causes the window below window 2,
to be exposed through an apparent hole in window 1.
Fixes: https://bugs.freedesktop.org/show_bug.cgi?id=22566
Signed-off-by: Ville Syrjala <syrjala@sci.fi>
Reviewed-by: Keith Packard <keithp@keithp.com>
Signed-off-by: Keith Packard <keithp@keithp.com>
(cherry picked from commit a5266dcb3a)
Use fchmod() to change permissions of the lock file instead
of chmod(), thus avoid the race that can be exploited to set
a symbolic link to any file or directory in the system.
Signed-off-by: Matthieu Herrb <matthieu.herrb@laas.fr>
Reviewed-by: Alan Coopersmith <alan.coopersmith@oracle.com>
(cherry picked from commit b67581cf82)
use O_NOFOLLOW to open the existing lock file, so symbolic links
aren't followed, thus avoid revealing if it point to an existing
file.
Signed-off-by: Matthieu Herrb <matthieu.herrb@laas.fr>
Reviewed-by: Alan Coopersmith <alan.coopersmith@oracle.com>
(cherry picked from commit 6ba44b91e3)
This line was introduced pre-1.6 to fix Bug 19297. The effect of warping
through the VCP then was that if a device had custom valuator ranges, the
warp position would be wrong. The better device for this effect is the the
XTest device.
This fixes a server crash where the lastSlave is a pointer device without
valuators (Bug 38313#0).
And while we're at it, make sure the Xinerama code-path does the same.
X.Org Bug 38313 <http://bugs.freedesktop.org/show_bug.cgi?id=38313>
Signed-off-by: Peter Hutterer <peter.hutterer@who-t.net>
Reviewed-by: Daniel Stone <daniel@fooishbar.org>
(cherry picked from commit 2bfb802839)
Renesas SH is not aligned at size of double.
When structure has double value, It is aligned in 4byte (long).
Signed-off-by: Nobuhiro Iwamatsu <iwamatsu@nigauri.org>
Reviewed-by: Peter Hutterer <peter.hutterer@who-t.net>
Signed-off-by: Peter Hutterer <peter.hutterer@who-t.net>
(cherry picked from commit b29ce0726d)
Modifier device is always the keyboard.
Signed-off-by: Peter Hutterer <peter.hutterer@who-t.net>
Reviewed-by: Daniel Stone <daniel@fooishbar.org>
(cherry picked from commit 8c5a4d6fbe)
Passive core grabs are mostly device-independent. In an MPX scenario, they
may change to reflect whichever master pair activated the grab last. For
adding new grabs to the list, ignore the device for core grabs to return
failures when trying to set the same grab combo twice on a window.
X.Org Bug 39545 <http://bugs.freedesktop.org/show_bug.cgi?id=39545>
Signed-off-by: Peter Hutterer <peter.hutterer@who-t.net>
Reviewed-by: Daniel Stone <daniel@fooishbar.org>
(cherry picked from commit 09496996ac)
Core grabs may change device when they're activated to reflect the master
they apply to. If the device is a keyboard, modifierDevice is erroneously
set to the Virtual Core Pointer.
Signed-off-by: Peter Hutterer <peter.hutterer@who-t.net>
Reviewed-by: Daniel Stone <daniel@fooishbar.org>
(cherry picked from commit 80c3704853)
PRODUCT was taken from the parent, hence ppath.
Signed-off-by: Peter Hutterer <peter.hutterer@who-t.net>
Reviewed-by: Daniel Stone <daniel@fooishbar.org>
(cherry picked from commit e684e816ac)
Caught during review of e095369bf.
Signed-off-by: Peter Harris <pharris@opentext.com>
Reviewed-by-by: Matt Turner <mattst88@gmail.com>
Signed-off-by: Matt Turner <mattst88@gmail.com>
(cherry picked from commit c90903b4f7)
Conflicts:
xkb/xkb.c
Signed-off-by: Peter Hutterer <peter.hutterer@who-t.net>
Some extension libraries may set this bit before converting the event to
wire protocol and as such range checking the event will cause an invalid
BadValue error to result. As the documentation suggests the the bit
should be "forced on", remove it before doing range checks and continue
to force it on in the server.
Reviewed-by: Jamey Sharp <jamey@minilop.net>
Signed-off-by: Peter Hutterer <peter.hutterer@who-t.net>
(cherry picked from commit 2d2dce558d)
The corresponding DeviceIntPtr wasn't being gotten properly,
resulting in BadDevice from dixLookupDevice().
Signed-off-by: Carlos Garnacho <carlosg@gnome.org>
Reviewed-by: Peter Hutterer <peter.hutterer@who-t.net>
Signed-off-by: Peter Hutterer <peter.hutterer@who-t.net>
(cherry picked from commit f52d5cd374)
If AddResource fails, it will automatically free the object that was
passed to it by calling the appropriate deleteFunc; and of course
FreeResource also calls the deleteFunc. In both cases it's wrong to call
the destroy hook manually.
Commit by Jamey Sharp and Josh Triplett.
Signed-off-by: Jamey Sharp <jamey@minilop.net>
Signed-off-by: Josh Triplett <josh@joshtriplett.org>
Reviewed-by: Peter Hutterer <peter.hutterer@who-t.net>
Reviewed-by: Aaron Plattner <aplattner@nvidia.com>
(cherry picked from commit 0f380a5005)
These symbols were not renamed when they were added to libfb:
# nm -D libwfb.so | grep ' fb'
0000000000028d00 T fbTrapezoids
0000000000028d60 T fbTriangles
This causes corruption and/or crashes on wfb-ful drivers like nvidia:
Program received signal SIGABRT, Aborted.
0x00007fd67f3a0405 in *__GI_raise (sig=<optimized out>) at ../nptl/sysdeps/unix/sysv/linux/raise.c:64
64 ../nptl/sysdeps/unix/sysv/linux/raise.c: No such file or directory.
in ../nptl/sysdeps/unix/sysv/linux/raise.c
(gdb) bt
#0 0x00007fd67f3a0405 in *__GI_raise (sig=<optimized out>) at ../nptl/sysdeps/unix/sysv/linux/raise.c:64
#1 0x00007fd67f3a3680 in *__GI_abort () at abort.c:92
#2 0x00007fd67f3995b1 in *__GI___assert_fail (assertion=0x7fd679ecb804 "key->initialized", file=<optimized out>, line=116, function=0x7fd679ecbbc0 "dixGetPrivateAddr")
at assert.c:81
#3 0x00007fd679ec55b6 in ?? () from /usr/lib/xorg/modules/libfb.so
#4 0x00007fd679eca9ef in ?? () from /usr/lib/xorg/modules/libfb.so
#5 0x00007fd679ecae20 in fbTriangles () from /usr/lib/xorg/modules/libfb.so
#6 0x00007fd67a58fc55 in ?? () from /usr/lib/xorg/modules/drivers/nvidia_drv.so
#7 0x00000000004f38d1 in ?? ()
#8 0x0000000000437ae9 in ?? ()
#9 0x0000000000426eaa in ?? ()
#10 0x00007fd67f38cead in __libc_start_main (main=<optimized out>, argc=<optimized out>, ubp_av=<optimized out>, init=<optimized out>, fini=<optimized out>,
rtld_fini=<optimized out>, stack_end=0x7fff99860d78) at libc-start.c:228
#11 0x000000000042719d in _start ()
Signed-off-by: Aaron Plattner <aplattner@nvidia.com>
Reviewed-by: Jeremy Huddleston <jeremyhu@apple.com>
Signed-off-by: Keith Packard <keithp@keithp.com>
(cherry picked from commit 03d032991d)
The video driver ABI was bumped to 11.0 in commit
0de7cec907 because of a change to the
size of ATOM in commit 51f353d0a0. This
also affects extension modules, so the extension ABI version should
have been bumped too.
Signed-off-by: Aaron Plattner <aplattner@nvidia.com>
Reviewed-by: Peter Hutterer <peter.hutterer@who-t.net>
Signed-off-by: Keith Packard <keithp@keithp.com>
The memcpy fast path implicitly assumes that the copy walks
left-to-right. That's not something memcpy guarantees, and newer glibc
on some processors will indeed break that assumption. Since we walk a
line at a time, check the source and destination against the width of
the blit to determine whether we can be sloppy enough to allow memcpy.
(Having done this, we can remove the check for !reverse as well.)
On an Intel Core i7-2630QM with an NVIDIA GeForce GTX 460M running in
NoAccel, the broken code and various fixes for -copywinwin{10,100,500}
gives (edited to fit in 80 columns):
1: Disable the fastpath entirely
2: Replace memcpy with memmove
3: This fix
4: The code before this fix
1 2 3 4 Operation
------ --------------- --------------- --------------- ------------
258000 269000 ( 1.04) 544000 ( 2.11) 552000 ( 2.14) Copy 10x10
21300 23000 ( 1.08) 43700 ( 2.05) 47100 ( 2.21) Copy 100x100
960 962 ( 1.00) 1990 ( 2.09) 1990 ( 2.07) Copy 500x500
So it's a modest performance hit, but correctness demands it, and it's
probably worth keeping the 2x speedup from having the fast path in the
first place.
Signed-off-by: Adam Jackson <ajax@redhat.com>
Signed-off-by: Keith Packard <keithp@keithp.com>
Otherwise sys_pitch will be stale when a system memory copy is allocated.
Fixes https://bugs.freedesktop.org/show_bug.cgi?id=38322 and a crash when
unlocking the screen with xscreensaver, reported by Janne Huttunen.
Signed-off-by: Michel Dänzer <michel.daenzer@amd.com>
Tested-by: Janne Huttunen <jahuttun@gmail.com>
Tested-by: Jan Kriho <Erbureth@gmail.com>
Signed-off-by: Keith Packard <keithp@keithp.com>
Fixes assertion failure when calling dixSetPrivate
Debian bug#632549 <http://bugs.debian.org/632549>
Reported-and-tested-by: Mohammed Sameer <msameer@foolab.org>
Signed-off-by: Julien Cristau <jcristau@debian.org>
Reviewed-by: Daniel Stone <daniel@fooishbar.org>
Signed-off-by: Keith Packard <keithp@keithp.com>
The composite extension spec says that window background painting
should be inhibited when the subwindow redirection mode is set to
manual.
This eliminates the ugly flashing effect when compiz unredirects a
fullscreen window.
Signed-off-by: Ville Syrjälä <ville.syrjala@nokia.com>
Reviewed-by: Owen Taylor <otaylor@fishsoup.net>
Reviewed-by: Keith Packard <keithp@keithp.com>
Signed-off-by: Keith Packard <keithp@keithp.com>
This reverts commit db8840600e.
It was an optimization for the resize case, but 193ecc8b45 made
it so that no backfilling takes place on resize if left in.
Signed-off-by: Pierre-Loup A. Griffais <pgriffais@nvidia.com>
Reviewed-by: Adam Jackson <ajax@redhat.com>
Conflicts:
composite/compalloc.c
Signed-off-by: Keith Packard <keithp@keithp.com>
RRPostPendingProperties tries to compare the pending and current
property values to decide whether they're actually changing. However,
it does this using a memcmp that passes in pending_value->size as the
number of bytes. This is actually the number of elements, where each
element is (pending_value->format / 8) bytes long. This causes the
pending value to not be propagated if the first pending_value->size
bytes are the same and only the end of it is changing.
Fix this by computing the total number of bytes to compare in the
memcmp.
Signed-off-by: Aaron Plattner <aplattner@nvidia.com>
Reviewed-by: Keith Packard <keithp@keithp.com>
Signed-off-by: Keith Packard <keithp@keithp.com>
