Coming back from resume may leave us with a file descriptor that can be opened
but fails on the first read (ENODEV).
In this case, try to open the device until it becomes available or until the
predefined count expires. To be safe, we cache the information from the device
and compare against it when we re-open. This way we ensure that if the
topology changes under us, we don't open a completely different device. If a
device has changed, we disable it.
Adds option "ReopenAttempts" <int>
Conflicts:
man/evdev.man
src/evdev.c
src/evdev.h
Keycodes over 255 are silently ignored in the server. The least we can do is
put a warning in the logs.
Signed-off-by: Peter Hutterer <peter.hutterer@who-t.net>
(cherry picked from commit c1f7f8c3d2)
With this fix, on my PowerBook HAL hotplugging correctly detects my USB mouse,
and no longer thinks keyboards have random numbers of mouse buttons. :)
The LONG_BITS and NBITS macro definitions are stolen from xf86-input-synaptics.
Signed-off-by: Michel Dänzer <michel@tungstengraphics.com>
Signed-off-by: Peter Hutterer <peter.hutterer@who-t.net>
[cherry-picked from master and fixed the trivial conflict -- jcristau]
After suspend/resume, sometimes the device doesn't come back up on the same
node. Since we do not call PreInit for the device (which would detect this
situation), we continue to try to read a nonexisting file, spamming the log
file with "Read Error".
(cherry picked from commit bf0d81011e)
This ensures that the middle button emulation is re-enabled after VT switch,
otherwise the block handler that deals with the timeouts would not get
re-registered.
Signed-off-by: Keith Packard <keithp@keithp.com>
(cherry picked from commit 76800bfa75)
Default setting is still "on" until middle button is pressed. If the options
is however explicitly stated in the config file, it takes the value from the
config file, no matter if a middle button is present.
Devices may report middle mouse buttons even if they don't have one (PS/2
devices just don't know any better), so we can't be sure until we see the
event.
The commit b4a5a204 fixed an issue, where we can't move the pointer to
other screens and this happens in current master branch again. This commit
ports the old commit to the current master branch.
Signed-off-by: Sven Wegener <swegener@gentoo.org>
Signed-off-by: Peter Hutterer <peter@cs.unisa.edu.au>
GetMotionEvents() doesn't exist, led to compile errors with servers pre-MPX
merge. Thanks to Sven Wegener for pointing this out.
This reverts commit 42422d8f69.
If the grab fails, this is most likely a sign that the device has been grabbed
already (probably by a device specified in xorg.conf). So let's not add the
device to the server's input device list, since it won't generate events
anyway.
Exception: keyboards and kernel 2.4 are not affected.
In the mouse driver, these options are only used if XFree86LOADER is
undefined. configure.ac in the xserver forces said define to 1 if we're
building the xfree86 DDX, so I don't see the point of having them around.
Especially since they weren't used in evdev anyway.
This is a bit of a mess. The MS Optical Desktop 2000 registers both relative
and absolute axes on the same device (the mouse). The absolute axes have a
valid min/max range for x/y and thus overwrite the x/y relative axes in the
server (no, this is not a server bug). And I wouldn't be surprised if other
devices have similar issues.
Since the device only sends relative events after that, the mouse is
essentially restricted to the min..max range of 0..255. The server simply
doesn't do unrestricted relative axis and restricted absolute axis on the same
device (not for the same axis numbers anyway).
Not many mice do this, but some do, Apple Mighty Mouse in particular, and
it makes click-and-drag pretty much impossible to use.
Arguably we should filter _all_ repeat events from the kernel and handle
synthesizing them in the server.
Don't do this in the button map. That's writeable by clients, which means
they have the chance to get it wrong. Just swap right and middle around
in event dispatch before you get to the user's map.
This resets to before the ad-hoc input hotplug state, ports that to
current server ABI, adds input hotplug integration, and re-adds support
for absolute coordinate events and XKB. Several other small fixes were
cherry-picked from the new 1.2 branch.
