On 2/6/18 19:53 , Timur Kristóf wrote:
On Sat, 2018-06-02 at 14:14 +1000, Peter Hutterer wrote:
> On 1/6/18 22:50 , Timur Kristóf wrote:
>> On Thu, 2018-05-31 at 13:44 -0700, Laura Abbott wrote:
>>> On 05/24/2018 04:13 PM, Timur Kristóf wrote:
>>>> Hi,
>>>>
>>>> On the XPS 13 9360 and 9370 xinput sees two touchpads instead
>>>> of
>>>> one:
>>>> ⎡ Virtual core pointer id=2 [master
>>>> pointer (3)]
>>>> ⎜ ↳ Virtual core XTEST
>>>> pointer id=4 [slave pointer (2)]
>>>> ⎜ ↳ DELL07E6:00 06CB:76AF
>>>> Touchpad id=12 [slave pointer (2)]
>>>> ⎜ ↳ SynPS/2 Synaptics
>>>> TouchPad id=17 [slave pointer (2)]
>>>>
>>>> dmesg also gives me the following:
>>>>
>>>> [ 1.429811] psmouse serio1: synaptics: Your touchpad (PNP:
>>>> DLL07e6
>>>> PNP0f13) says it can support a different bus. If i2c-hid and
>>>> hid-
>>>> rmi
>>>> are not used, you might want to try setting
>>>> psmouse.synaptics_intertouch to 1 and report this to
>>>> linux-input(a)vger.kernel.org.
>>>>
>>>> The Dell one is the real touchpad, the other one is an artifact
>>>> of
>>>> the
>>>> psmouse driver. Weird touchpad freezes and jittering issues can
>>>> be
>>>> observed when both of these devices are there. The general
>>>> advice
>>>> on
>>>> the web is to blacklist the psmouse driver. (This is also
>>>> published
>>>> by
>>>> Dell as a .deb package which contains a config file doing just
>>>> this.)
>>>>
>>>> However, on Fedora I cannot blacklist the psmouse driver
>>>> because it
>>>> is
>>>> compiled built-in instead of as a module. Could you guys change
>>>> the
>>>> Fedora kernel config to compile it as a module instead?
>>>>
>>>> Thanks & best regards,
>>>> Timur
>>>
>>> So assuming nobody else has objections, I think it's okay to at
>>> least try this on rawhide and see if it uncovers any other
>>> problems.
>>>
>>> Thanks,
>>> Laura
>>>
>>
>> Looking into this a bit more, the root cause of this issue is that
>> the
>> touchpad is wired up in such a way that it is accessible on both
>> PS/2
>> and I2C. I2C is preferred and used by i2c-hid but at the same time
>> the
>> PS/2 interface is also picked up by psmouse. Thus, Xorg sees the
>> two
>> interfaces as two different devices.
>
> That is fairly standard and doesn't matter. This was also the case
> for
> earlier SMBus implementation but the common thing was that the PS/2
> device never actually sent events. So it was only an issue for some
> tools that expected only one touchpad to exist (e.g. synclient).
> Check with evemu-record whether the device sends events. if not,
> then
> Xorg is definitely not confused by it. If it does and both event
> nodes
> send events then yes, we have a kernel bug there.
I've talked to Mario, a developer at Dell, and he pointed me to this
freedesktop bug report:
https://bugs.freedesktop.org/show_bug.cgi?id=101470
According to his description, there is actually no input coming from
PS/2 but it is still (wrongly) picked up by syndaemon. To be honest,
I'm not sure if Fedora uses syndaemon (I assume libinput is used for
everything now?), but I can tell that the touchpad does jitter when the
PS/2 device is there.
syndaemon works in that it monitors the keyboard state (or attaches to
XRECORD events) and then sets a synaptics-specific property on the
device it attaches to. This means that unless the synaptics driver is
used, it cannot do anything. Do you have xorg-x11-drv-synaptics-legacy
installed? If not, then libinput is being used (also: xinput list-props
<device name> shows the driver name in the property names).
Even if you do have the synaptics driver installed, disabling a kernel
device that doesn't send events ... well, you can't get less than "no
events" :)
I can also confirm that evemu-record doesn't see any input coming
from
the PS/2 device.
>> Solution would be to somehow remove the psmouse-detected device
>> node
>> from the system when i2c-hid detects the same device. Is this
>> possible?
>
> This is more an upstream discussion, CC-ing benjamin for details.
>
> I do recommend filing a bug with an evemu recording of a
> freezing/jittering interaction. Lots of touchpad processing happens
> in
> userspace and this could be the cause of pressure thresholds set too
> high. CC me on that bug and I can have a look. sudo libinput
> debug-events --verbose will also show extra information including
> things
> when a touch ended based on its pressure information.
I can try to reproduce the problem and see if I can make an evemu
recording. What should I file the bug against (and where)?
Unfortunately I'm not familiar with the components involved.
For upstream, this link has everything you need:
https://wayland.freedesktop.org/libinput/doc/latest/reporting_bugs.html
In the Fedora bugzilla it's the libinput component. Either bugzilla will
work, it's the same person answering you anyway (i.e. me :)
Searching for libinput in the journal reveals a bunch of touch jump
messages, which seem to be related to the problem:
https://paste.fedoraproject.org/paste/MoZGqf25VpHjCLSYzTxdfw
there are a few issues here. The cursor jump complaint could be the
cause of the jerkiness but we only trigger that when you move >20mm
within one event - not something that can be triggered easily (or at
all) in real life. All the timer offset negative bugs indicate that your
compositor isn't rendering fast enough and libinput is starved for
attention. That is currently relatively common under Wayland but I
haven't seen this under Xorg yet.
IOW I think we may have a multitude of bugs here that need untangling
before we can pin down psmouse as real culprit.
Cheers,
Peter