One new property, and the existing accel profile gets extended to keep one
extra value. The new property libinput Accel Curve Points is a list of pairs
of points to be added to the acceleration curve.
libinput only supports adding points to the curve so we simply declare the
behavior as undefined when the curve is set multiple times. Also helps to
identify those that bother to read the man page before playing with random
driver values.
Signed-off-by: Peter Hutterer <peter.hutterer@who-t.net>
Properties are initialized on the correct devices only but on resume we'd just
blindly apply the config from our device. Depending on the resume order, this
would mean we'd apply a previously set config with a default config.
Example:
* pointer device with keyboard subdevice
* pointer device exports natural scrolling, keyboard device does not and
remains at default (off)
* client enables natural scrolling on the pointer device
* VT switch away, VT switch back
* pointer device gets enabled first, enables natural scrolling on the
libinput device
* keyboard device gets enabled second, resets to the default value
Reported-by: Yuxuan Shui <yshuiv7@gmail.com>
Signed-off-by: Peter Hutterer <peter.hutterer@who-t.net>
Tested-by: Yuxuan Shui <yshuiv7@gmail.com>
87f9fe3a6fafe60134c6's intention was to not create properties that a subdevice
doesn't have configuration options for (i.e. if you have a pointer+keyboard
device, don't expose tapping configuration on the keyboard subdevice).
The result was messy, the checker function had a confusing triple-negation and
some properties weren't checked - e.g. left-handed was allowed for touch/tablet
but not for pointer, dwt was allowed for any device.
Fix this by moving the check into the property init function directly and
inverting the helper function to be easier to read.
Signed-off-by: Peter Hutterer <peter.hutterer@who-t.net>
Not all clients update the pointer position correctly from the button events
(for historical reasons) so we need to send a motion event before the button
event that represents a tip state change.
https://bugs.freedesktop.org/show_bug.cgi?id=101588
Signed-off-by: Peter Hutterer <peter.hutterer@who-t.net>
If a device is split into multiple subdevices, usually pointer+keyboard, we
initialized properties matching the libinput device on both devices. This
results in the keyboard having e.g. a Accel Speed or Left Handed settings even
though it cannot send any events of that type.
Filter by capabilities on the subdevice so we only get those properties that
match the subdevice's capabilities.
https://bugs.freedesktop.org/show_bug.cgi?id=100900
Signed-off-by: Peter Hutterer <peter.hutterer@who-t.net>
This patch splits the meat of xf86libinput_handle_tablet_axis into a helper
function xf86libinput_post_tablet_motion(), to be called right after we send
the proximity in event.
Clients that don't handle proximity (e.g. all XI2 clients) don't see the
coordinates we send along with the proximity events. And, for historical
reasons, they may not look at the coordinates in button events. So a device
that comes into proximity and immediately sends a tip down button event
doesn't send a motion event, causing the client to think the tip down was at
whatever the last known position was (before previous prox-out).
The practical effect is that when a user tries to draw a few dots, they end up
being connected to each other.
https://bugzilla.redhat.com/show_bug.cgi?id=1433755
Signed-off-by: Peter Hutterer <peter.hutterer@who-t.net>
Only use-case here are pad mode LEDs that now live in /sys/class/leds. Asking
the server to open them is pointless, the server only knows how to open Option
"Device". And since the LEDs are in sysfs we should have access to them
anyway, so no need for jumping through or hula-ing hoops.
xf86CloseSerial() works as intended as it's a slim wrapper around close(), so
we only have to worry about the open() path here.
Signed-off-by: Peter Hutterer <peter.hutterer@who-t.net>
Updating the property directly causes us to send events from the input thread
which has some "interesting" side effects like messing up the reply order or
just crashing the server.
Schedule a work proc instead and update it whenever the server is back in the
main thread.
Signed-off-by: Peter Hutterer <peter.hutterer@who-t.net>
By default, the X server maps the tablet axes to the available screen area.
When a tablet is mapped to the screen but has a different aspect ratio than
the screen, input data is skewed. Expose an area ratio property to map the
a subsection of the available tablet area into the desired ratio.
Differences to the wacom driver: there the x/y min/max values must be
specified manually and in device coordinates. For this driver we merely
provide the area ratio (e.g. 4:3) and let the driver work out the rest.
Signed-off-by: Peter Hutterer <peter.hutterer@who-t.net>
Reviewed-by: Jason Gerecke <jason.gerecke@wacom.com>
Takes a 4-point cubic bezier curve as input and maps the pressure coordinates
to the values outlined by this curve. This is an extension of the current
implementation in the xf86-input-wacom driver which only allows the two center
control points to be modified.
Over the years a few users have noted that the wacom driver's pressure curve
makes it impossible to cap the pressure at a given value. Given our bezier
implementation here, it's effectively a freebie to add configurability of the
first and last control points. We do require all control points' x coordinates
to be in ascending order.
Signed-off-by: Peter Hutterer <peter.hutterer@who-t.net>
For a mouse with a click angle of 15 degrees things are unchanged. For devices
with angles less than 10, the current code scrolled way too fast. Because the
angle wasn't used anywhere, each tick would count as full scroll wheel event,
a slight movement of the wheel would thus scroll as much as a large movement
on a normal mouse.
Fix this by taking the actual click angle of the device into account. We
calculate some multiple of the angle that's close enough to the default 15
degrees of the wheel and then require that many click events to hit the full
scroll distance. For example, a mouse with a click angle of 3 degrees now
requires 5 clicks to trigger a full legacy scroll button event.
XI2.1 clients get the intermediate events (i.e. in this case five times
one-fifth of the scroll distance) and can thus scroll smoothly, or more
specifically in smaller events than usual.
https://bugs.freedesktop.org/show_bug.cgi?id=92772
Signed-off-by: Peter Hutterer <peter.hutterer@who-t.net>
Reviewed-by: Hans de Goede <hdegoede@redhat.com>
Changed this during development because I forgot that the value actually
matters (for touchpads anyway).
Signed-off-by: Peter Hutterer <peter.hutterer@who-t.net>
Reviewed-by: Hans de Goede <hdegoede@redhat.com>
If an XKB AccessX timeout is set and a VT switch is triggered, the
AccessXTimeoutExpire function may be called after the device has already been
disabled. This can cause a null-pointer dereference as our shared libinput
device may have been released by then.
In the legacy drivers this would've simply caused a write to an invalid fd
(-1), not a crash. Here we need to be more careful.
https://bugs.freedesktop.org/show_bug.cgi?id=98464
Signed-off-by: Peter Hutterer <peter.hutterer@who-t.net>
Reviewed-by: Hans de Goede <hdegoede@redhat.com>
The parent device ref's the libinput device during pre_init and unref's it
during DEVICE_INIT, so the copy is lost. During DEVICE_ON, the libinput device
is re-added and ref'd, this one stays around now. But the takeaway is: unless
the device is enabled, no libinput device reference is available.
If a device is a mixed pointer + keyboard device, a subdevice is created
during a WorkProc. The subdevice relied on the parent's libinput_device being
available and didn't even check for it. This WorkProc usually runs after
the parent's DEVICE_ON, so in most cases all is well.
But when running without logind and the server is vt-switched away, the parent
device only runs PreInit and DEVICE_INIT but never DEVICE_ON, causing the
subdevice to burn, crash, and generally fail horribly when it dereferences the
parent's libinput device.
Fix this because we have global warming already and don't need to burn more
things and also because it's considered bad user experience to have the
server crash. The simple fix is to check the parent device first and if it is
unavailable, create a new one because it will end up disabled as well anyway,
so the ref goes away as well. The use-case where the parent somehow gets
disabled but the subdevice doesn't is a bit too niche to worry about.
This doesn't happen with logind because in that case we don't get a usable fd
while VT-switched away, so we can't even run PreInit and never get this far
(see the paused fd handling in the xfree86 code for that). It can be
reproduced by setting AutoEnableDevices off, but why would you do that,
seriously.
https://bugs.freedesktop.org/show_bug.cgi?id=97117
Signed-off-by: Peter Hutterer <peter.hutterer@who-t.net>
Reviewed-by: Hans de Goede <hdegoede@redhat.com>
The property is tablet-wide, not just per tool. So when one tool is updated,
run through all other devices that share the same underlying device.
Signed-off-by: Peter Hutterer <peter.hutterer@who-t.net>
Reviewed-by: Hans de Goede <hdegoede@redhat.com>
is_libinput_device(next) causes a dereference of next anyway, so this cannot
ever be NULL.
Besides, if next ends up as NULL that means we have lost count of how many
remaining devices use libinput, so we have other issues.
Found by coverity.
Signed-off-by: Peter Hutterer <peter.hutterer@who-t.net>
If we don't swap out the pInfo previously passed to xf86AddEnabledDevice(),
the thread eventually calls read_input on a struct that has been deleted.
Avoid this by swapping out the to-be-destroyed pInfo with the first one we
find.
Reproducer: sudo udevadm trigger --type=devices --action=add
Signed-off-by: Peter Hutterer <peter.hutterer@who-t.net>
Reviewed-by: Hans de Goede <hdegoede@redhat.com>
This looks like a cut&paste coding error to me, and it generated a
compiler warning about possibly uninitialized value.
Signed-off-by: Keith Packard <keithp@keithp.com>
Signed-off-by: Peter Hutterer <peter.hutterer@who-t.net>
Avoid creating new devices from within the input thread which was the case for
tablet tools. It requires a lot more care about locking and has a potential to
mess up things.
Instead, schedule a WorkProc and buffer all events until we have the device
created. Once that's done, replay the event sequence so far. If the device
comes into proximity and out again before we manage to create the new device
we just ditch the whole sequence and wait for the next proximity in.
Signed-off-by: Peter Hutterer <peter.hutterer@who-t.net>
There is not good wire protocol for pad modes so instead we just export the
information via properties. One property to tell us how many groups and how
many modes each group has. One property for the current mode of each group.
And three properties to tell us which group each button, ring and strip is in.
Signed-off-by: Peter Hutterer <peter.hutterer@who-t.net>
The list returned by xf86FirstLocalDevice() includes our own device. If the
parent device is removed before the hotplug callback is invoked, the first
match with the same shared-device ID is our own device (or potentially another
subdevice on the same already-removed parent). Avoid this by making sure the
matched device is actually a parent device.
Signed-off-by: Peter Hutterer <peter.hutterer@who-t.net>
Reviewed-by: Hans de Goede <hdegoede@redhat.com>
Tested-by: Keith Packard <keithp@keithp.com>
If the parent device is removed before the WorkProc is called, the private
data is NULL.
Signed-off-by: Peter Hutterer <peter.hutterer@who-t.net>
Reviewed-by: Hans de Goede <hdegoede@redhat.com>
libinput can't run at SIGIO time, so it has been using
AddEnabledDevice to run in non-signal context.
Threaded input runs all input in non-signal context, so we want to use
xf86AddEnabledDevice at last.
v2: use XINPUT ABI version check instead of testing for presence of
AddEnabledDevice, which can't get removed from the server until
a few more patches past the threaded input change are merged.
v3: remove reference to XI86_SIGNAL_IO, which was presumably
a planned change to the xf86AddEnabledDevice path to make that
not use SIGIO.
Signed-off-by: Keith Packard <keithp@keithp.com>
Tested-by: Michel Dänzer <michel.daenzer@amd.com>
Signed-off-by: Peter Hutterer <peter.hutterer@who-t.net>
Two bugs caused proximity events to be discarded. First, on proximity out
posting through pDev would be discarded because pDev is the parent device that
we use as a base for hotplugging the real devices for each tool from. That
device never sends events though, doing so will see the event discarded in the
server.
Second, if the tool already exists don't just exit, send the proximity event
first. To unify the three paths where we do send the events simply move them
down to the exit phase of the function.
https://bugs.freedesktop.org/show_bug.cgi?id=95484
Signed-off-by: Peter Hutterer <peter.hutterer@who-t.net>
Modelled to be mostly compatible to the xf86-input-wacom driver behavior. The
pad gets 7 axes, the first three of which are mute and the others are always
available but obviously only send events when the axis is there.
The strip axes are incompatible, the wacom driver merely forwards the device
events (which are a bitshifted value), libinput normalizes it and we just
expand this back into an integer range. Let's see how we go with this.
Signed-off-by: Peter Hutterer <peter.hutterer@who-t.net>
