At least in the DBus API we can't just say "doesn't exist", so let's allow for
a zero-sized list.
Signed-off-by: Peter Hutterer <peter.hutterer@who-t.net>
To be able to query which modes are available.
Exposed via ratbagctl as
./build/ratbagctl.devel led 0 capabilities event4
Signed-off-by: Peter Hutterer <peter.hutterer@who-t.net>
Previously, we could get a min/max value but there was no guarantee that it
any value in between would actually work on the device. Some devices have a
discrete number of resolutions (e.g. 500, 1000, 2000, 2500), most others have
at least a minimum step size (so 1234 isn't a valid dpi value).
Change the min/max to a discrete list of which resolutions are supported. This
allows a UI to adjust its interface dynamically - e.g. a combobox if there are
only 3 resolutions or a slider if there are a million of them.
This is a DBus API break
Signed-off-by: Peter Hutterer <peter.hutterer@who-t.net>
Previously, probe() would call ratbag_init_profiles() which in turn would call
the various function pointers in the driver to fill in profiles, buttons and
leds (the latter while filling in profiles). This made the probe() function
harder to understand than necessary and unreliable because it wasn't always
clear which bits were already set up while some driver function was called and
which ones weren't yet.
This patch switches to a flat probe(). The driver still needs to call
ratbag_init_profiles(), libratbag sets up everything (to default values). Then
the driver can fill in the rest as it wants.
This patch still makes use of the current infrastructure, the drivers
just call the read_profile and read_button functions manually now.
Fixes#370
Signed-off-by: Peter Hutterer <peter.hutterer@who-t.net>
Resolutions were still stored in an array over a linked list like the
buttons/leds/profiles and initializing them from a driver was different than
buttons or leds. This brings the resolutions in line with the buttons and
leds.
Signed-off-by: Peter Hutterer <peter.hutterer@who-t.net>
At least on HID++ 2.0, the LED types appear to be random and don't match the
protocol specs. Other protocols don't even have those bits, so let's make this
a generic entry.
This is pre-filled for all LEDs but drivers can override it where required.
Fixes#377
Signed-off-by: Peter Hutterer <peter.hutterer@who-t.net>
A driver is supposed to know what type an LED is, unknown doesn't help anyone
here. So replace this with a private #define and abort when a driver doesn't
set the type. This also adds a requirement that any driver initializing a LED
must have the read_led() function, otherwise we can't fill it in correctly.
And bump the enum up by one, the first valid value should start at 1 (or some
other non-boolean value).
Related #365Fixes#364
Signed-off-by: Peter Hutterer <peter.hutterer@who-t.net>
1-bit RGB is needed for the Logitech G300, let's rename the main RGB to
indicate how many bits we actually have.
See #335
Signed-off-by: Peter Hutterer <peter.hutterer@who-t.net>
The initial value of 400 in the enum is chosen so it looks weird when someone
tries to interpret the value as integer. Values of 1, 2, 4, etc. could be
confused with something realistic.
Signed-off-by: Peter Hutterer <peter.hutterer@who-t.net>
Some devices can have a custom name, so let's export that capability.
In ratbagd, libratbag/ratbag_profile_set_name() returns an error, we
put it under silence, the client has to check for the capabilities
before asking ratbagd to change the profile name.
Fixes: #266
Signed-off-by: Benjamin Tissoires <benjamin.tissoires@gmail.com>
The hwdb was a quick and easy way to add entries for new devices but we need
more and more info about the various types of devices. It's better to move
this to a proper database.
Side-effect: this also gives us an easy way to point to the list of devices
supported and makes checking for e.g. missing svgs easier.
Signed-off-by: Peter Hutterer <peter.hutterer@who-t.net>
If a driver succeeded but then fails sanity checks, leave the driver set and
let the cleanup code handle the rest. Otherwise, memory allocated by the
driver's probe() function is never released.
Fixes a valgrind memory leak complaint when running tests for invalid devices
(e.g. device_profiles_num_0).
Signed-off-by: Peter Hutterer <peter.hutterer@who-t.net>
Any specific capability like 'switchable resolution' requires the basic
'resolution' capabilty to be set.
Signed-off-by: Peter Hutterer <peter.hutterer@who-t.net>
Make sure we keep the defaults as such by duplicating them in
driver-test.c (and mark the input data as const).
Add a dbus method to reset the test device from the caller.
This is much useful when running tests given we have no guarantees on the
ordering.
Signed-off-by: Benjamin Tissoires <benjamin.tissoires@gmail.com>
In case the udev property is not present, we might want to query
the driver for the SVG name.
This is useful for driver-test as there is no udev device attached to
it, and for hidpp20 as we have auto-detection. We should then be able
to implement this in hidpp20 also so that the name gets automatically
created.
Signed-off-by: Benjamin Tissoires <benjamin.tissoires@gmail.com>
This shuts up a scan-build warning about a memleak that looked like a false
positive but I couldn't get rid of.
Signed-off-by: Peter Hutterer <peter.hutterer@who-t.net>
A semi false-positive from Coverity, because action.macro is unset. Doesn't
matter because we never use it, but having it zeroed is the nice thing to do
(and what we do in the other functions).
Signed-off-by: Peter Hutterer <peter.hutterer@who-t.net>
Supports the common functionality exposed by libratbag:
- selecting profile 1-3 (profiles cannot be disabled)
- selecting resolution 1-4 (can also not be disabled)
- report frequency
- default resolution per profile
- button actions: mouse button numbers or key+modifier
(modifiers are just a bit field and can easly be set
in the driver, but not included as libratbag does not
support it yet)
- LEDs are not yet hooked up but are just a table of 7
different predifined colors.
- Seting active profile and resolution are both implemented
but only profile activation is in use. The resolution
activation code is commented until supported from libratbag
itself.
The G300 has an unusual way of numbering buttons:
Left:1, Right:2, Middle: 3
The purpose of report type 0xF2 is not known.
This is especially important for the mx780, so we can get rid of the
rude workaround we currently have in the driver for it.
Signed-off-by: Lyude <lyude@redhat.com>
Changes since v1:
- Just have profile enable/disable in ratbag-command
- Use bool instead of int
Changes since v2:
- Remove leftover function, whoops
- Fix leftover ints to bools
- Lots of style fixes
Make sure that there can only be one resolution marked
as default.
The previous code allowed to set multiple resolutions
as default. To clean up any bad state we allways loop
all other resolutions and unset the default if neccesary.
Almost all of the devices we've worked with here are the same: they have
some sort of profile on them, and if they don't (which is quite rare)
it's rather easy to simulate one in the driver. In fact basically all of
the device drivers in ratbag just update the profile and rewrite it on
calls to the write_resolution_dpi() and write_button() calls. This is
annoying since profile writes can take a good bit of time, and we're
wasting a lot of time with redundant profile writes that way.
During XDC this year, we came up with an idea that makes more sense: a
profile oriented API. Instead of having multiple entrypoints in the API
to writing and updating the profile we just cache all of the profile
changes until ratbag_device_commit_profiles() is called, at which point
we do a single set of writes for each profile and button that changed.
This has a lot of advantages:
- Where the profile is and is not written out to the device is now
explicit.
- Changing attributes of each profile won't result in any sort of delay
up until the library user commits the profiles, allowing us to avoid
hanging the application every time a simple attribute is changed.
- We can cut down on workarounds in drivers to cater to the previous
API.
Since this commit doesn't handle converting all of the device drivers
over to the new API, we also add a temporary wrapper for devices that
haven't yet been converted to this new API so that they're still usable.
Once all of the drivers have been converted to the new API, the wrapper
will be removed along with the internal driver API calls.
Signed-off-by: Lyude <thatslyude@gmail.com>
Changes since v1:
- has_changed→dirty
- Change namespace to LIBRATBAG_0.5.0
Changes since v2:
- Remove unneeded note about not setting dirty bit
- commit_profiles->commit
- Make FIXMEs more obvious
- Minor styling nitpicks
- Split up capability check fixes into seperate patch
This commit adds complete (with the exception of LEDs and a few other
small things) support for configuring the G.Skill Ripjaws MX780. So far
this has been tested on firmware versions 19 and 21, and both of them
seem to work perfectly with this.
Changes since v1:
- Lots of style fixes
- Remove reset hook, comment out with #if 0 in case of usage in the
future
- Add ret = -EAGAIN check in the loop in gskill_general_cmd()
- Explain why we wait longer then the spec says to in
gskill_general_cmd()
- Check that buf[0] is set to GSKILL_GENERAL_CMD in gskill_general_cmd()
- Add some notes as to why we don't use gskill_general_cmd() in
gskill_select_profile()
- Cleanup return codes, change gskill_general_cmd() so it only returns
valid error codes. In places not using gskill_general_cmd(), do
return rc < 0 ? rc : -EPROTO;
- Split command for selecting the macro to read/write from into it's own
function, just like we do with the profile
- Move profile count workaround into it's own function, make explicit
note of why it's needed
- Add proper support for reading/writing profile/macro names, ensure
they aren't left blank so as to not confuse G.Skill's software
- Use HID kbd modifier codes for translating macro kbd modifier codes
- Use AS_MASK instead of (1 << foo) (apparently we already had this :)
Signed-off-by: Lyude <cpaul@redhat.com>
If a device is not specifically assigned a driver, try hidpp20 and then
hidpp10 hoping that one of them will work.
Signed-off-by: Peter Hutterer <peter.hutterer@who-t.net>
Prep work for being able to probe multiple drivers on a device. The function
that tests the actual driver simply takes a const char * with the driver name,
so in the future we can loop through multiple driver names easily.
Signed-off-by: Peter Hutterer <peter.hutterer@who-t.net>
No reason for the drivers to have their own printf each, just return the
negative errno and let the common code print it.
This now also means that we only return -ENODEV on something that's not really
an error, for anything else we print the strerror.
Signed-off-by: Peter Hutterer <peter.hutterer@who-t.net>
We now get an error message for all devices that have duplicate event nodes
because we use -ENODEV for 'device not found', 'device is a duplicates' and
'device doesn't have the hidraw report'.
The original commit was triggered by a misunderstanding of the error message,
will be fixed in a follow-up.
This reverts commit 855a6bd831.
Signed-off-by: Peter Hutterer <peter.hutterer@who-t.net>
Otherwise, as soon as we set an action for a button, we get a nice
invalid free on cleanup.
Signed-off-by: Benjamin Tissoires <benjamin.tissoires@gmail.com>
The previous API was inconsistent and different to the rest of libratbag's
API. Replace with one where a macro is a separate struct, can be initialized
and then written to the device with a set_macro() call. The device copies the
contents, the macro is otherwise unmodified and can be re-used by the client
if need be.
Signed-off-by: Peter Hutterer <peter.hutterer@who-t.net>
We don't do anything with the return value anyway and we do assign the driver
in this function. Rename it and return a boolean only.
Signed-off-by: Peter Hutterer <peter.hutterer@who-t.net>
Whenever we do something that may fail, return an enum of possible errors.
This also changes the API for ratbag_device_new_from_udev_device() to return
an error code now too instead of just NULL on failure.
Error numbers start at -1000 to avoid overlap with errnos, error_code() should
be used internally to avoid returning an errno.
Signed-off-by: Peter Hutterer <peter.hutterer@who-t.net>
