Re: g_hid and hid-multitouch compatibility?

From: Benjamin Tissoires <hidden>
Date: 2011-05-31 16:55:19

On Tue, May 31, 2011 at 18:03, Pablo Cases [off-list ref] wrote:

Hi Benjamin,

I've attached some logs (rdesc.txt, events.txt and evtest.txt) showing a simulated 2-touch sequence spanning over four frames (last frame is a "untouch" event). Looking at the logs I'd say that I have a problem with the MIN/MAX limit decoding as evtest says limits are -127/127 but that is not what we're reporting according to rdesc.

Indeed, there are troubles with ranges of X and Y. The good point is
that besides that, hid-multitouch seems to work well.
However, what strikes me is that the values are not clamped by the input layer.

The problems comes from your mouse emulation, here:

  INPUT(4)[INPUT]
    Field(0)
      ....
    Field(1)
      Physical(GenericDesktop.Pointer)
      Usage(3)
        GenericDesktop.X
        GenericDesktop.Y
        GenericDesktop.Wheel
      Logical Minimum(-127)            <---------
      Logical Maximum(127)            <---------
      Report Size(8)
      Report Count(3)
      Report Offset(8)
      Flags( Variable Relative )

the values are not coherent with the one reported above. Maybe we
should patch hid-multitouch to ignore non mt inputs... ;-) but I
really don"t have the time to do it right now.
Maybe you don't need to introduce this mouse emulation at all in your
device (it depends if your "firmware" will support it or not).

There is also a problem with the feature
Digitizers.ContactMaximumNumber. Apparently, I would say that you put
the initial value is at 2 instead of 10 as your maximum value
suggests. If it's intentional, it's fine, but you won't be able to
send more than two touches, even by changing this number later...
(just saw the question after)

While we're at it I have some more questions:
Q: Do we need to acknowledge the Device Configuration Feature Report sent from the host to the device (instructing the device to report "multi-touch" events)?

If you are talking about the "Digitizers.InputMode" setting, then, we
just send the signal and we don't wait for an answer. My memory is
quite confusing here, but again, if you don't intend to support mouse
emulation, dropping this input in the reports will ease things.

Q: Do we need to send the ContactCountMaximum Feature Report as described in the Windows 7 specification? Because I can see that these values are hardcoded (or just initialized?) in hid-multitouch for each supported device.

So, the first implementation of hid-multitouch did not care about this
feature. Now, I tried to use it for devices that send it correctly. As
I did not asked all the owners of the devices to redo the tests, some
device still got .maxcontact
I'd rather not have this thing hardcoded in order to be able to reuse
the different classes with a lot of devices.

Cheers,
Benjamin

Regards,
Pablo

quoted

-----Original Message-----
From: Benjamin Tissoires [mailto:benjamin.tissoires@gmail.com]
Sent: den 31 maj 2011 13:19
To: Pablo Cases
Cc: Stéphane Chatty; Henrik Rydberg; Jiri Kosina; linux-
input@vger.kernel.org
Subject: Re: g_hid and hid-multitouch compatibility?

Hi Pablo,

On Mon, May 30, 2011 at 18:11, Pablo Cases [off-list ref]
wrote:

quoted

Hi Benjamin,

Now things seem to work a lot better thanks to your input :) I'm

still on the 2.6.38 version of the multitouch code, but static
registering works after adding our faked USB VIP/PID to hid.ids.c, hid-
core.c and hid-multitouch.c:

quoted

Digitizers.TipSwitch ---> Key.Touch
Digitizers.InRange ---> Sync.Report
Digitizers.ContactID ---> Sync.Report
GenericDesktop.X ---> Absolute.MTPositionX
GenericDesktop.Y ---> Absolute.MTPositionY
Digitizers.TipSwitch ---> Key.Touch
Digitizers.InRange ---> Sync.Report
Digitizers.ContactID ---> Sync.Report
GenericDesktop.X ---> Absolute.MTPositionX
GenericDesktop.Y ---> Absolute.MTPositionY
Digitizers.ContactCount ---> Sync.Report
Button.0001 ---> Key.LeftBtn
Button.0002 ---> Key.RightBtn
Button.0003 ---> Key.MiddleBtn
GenericDesktop.X ---> Absolute.MTPositionX
GenericDesktop.Y ---> Absolute.MTPositionY
GenericDesktop.Wheel ---> Relative.Wheel

Running some diagnostics:
workstation$ sudo python eviocg.py /dev/input/event3
the device '/dev/input/event3' is multitouch enabled with the

protocol B.

Can you send me the output of the command "evtest /dev/input/event3"
while moving/releasing fingers? I'm particularly interested in the
headers of this output (with the maximum of the ABS_MT_SLOTS).

Can you also send the full reports descriptor from
/sys/kernel/debug/hid/0003:xxxx:yyyy.0002/rdesc

quoted

An unmodified version of mtdiag now discovers our multitouch device,

but I cannot get any valid touch output in the "drawing surface" of the
application. You mentioned that I could optimize things by using some
of the quirk defined, but is this necessary or optional?

purely optional. (normally)

quoted

If I'm reporting as the Windows 7 specification describes do I still

need to use any of the quirks?

the point is that many hardware makers interpreted differently this
documentation (that's why we have those quirks). I'm not sure how
Microsoft handles this mess, but it may be even more complicated than
our hid-multitouch... So I can not guarantee that it will work out of
the box.
The big point lies in the understanding of the different
significations of the fields during a touch life time.

quoted

Specially I'm wondering about the quirks that relate to contact

id/contact number and slots. Do I really need to think about slots at
all if I have valid ContactIDs in my reports?

Well it's optimization, so it's optional (though I prefer, because it
avoids finding elements in lists).

Cheers,
Benjamin

quoted

Regards,
Pablo

quoted

-----Original Message-----
From: Benjamin Tissoires [mailto:benjamin.tissoires@gmail.com]
Sent: den 28 maj 2011 00:42
To: Pablo Cases
Cc: Stéphane Chatty; Henrik Rydberg; Jiri Kosina; linux-
input@vger.kernel.org
Subject: Re: g_hid and hid-multitouch compatibility?

Hi Pablo

On Sat, May 28, 2011 at 00:08, Pablo Cases

[off-list ref]

quoted

wrote:

quoted

Hi Benjamin,

Thanks for the response. It is my ignorance that should be excused

as

quoted

I have shown/will show that I am the novice with regards to USB HID

and

quoted

multitouch in this forum :)

quoted

I´ll continue interleaving additional comments below.

-Pablo

quoted

-----Original Message-----
From: Benjamin Tissoires [mailto:benjamin.tissoires@gmail.com]
Sent: den 27 maj 2011 18:10
To: Pablo Cases
Cc: Stéphane Chatty; Henrik Rydberg; Jiri Kosina; linux-
input@vger.kernel.org
Subject: Re: g_hid and hid-multitouch compatibility?

Hi Pablo,

Adding Stéphane, Henrik and Jiri in CC.

On Fri, May 27, 2011 at 17:33, Pablo Cases

[off-list ref]

quoted

wrote:

quoted

Hi all,

I have searched the net and the archives of linux-input for

information on how to use the Linux g_hid kernel module on a

device

quoted

and

quoted

the hid-multitouch kernel module on a host to communicate

multitouch

quoted

events. But unfortunately I can only get single touch working.

Excuse my ignorance, I didn't know g_hid existed before your

mail.

quoted

If

quoted

I understand it well enough, it's an hid emulation from user-

space.

quoted

Are you trying to inject events from a device in user-space to

the

quoted

kernel? If so, I think it would be easier to use uinput. And the

funny

quoted

think was that I'm currently writing a small lib to inject

mulitouch

quoted

events in a uinput device in an easier way.

Maybe I misunderstand your explanation above but we´re not running

g_hid and hid-multitouch on the same system. We have a separate
hardware USB device running Linux with the g_hid kernel module

loaded.

quoted

On this device we then have (thanks to g_hid) a userspace file

called

quoted

/dev/hidg0 that we can write USB HID input reports to, which then

will

quoted

be sent through the device's gadget usb framework, through an

physical

quoted

USB cable, to a separate PC host running hid-multitouch.

Thanks for the explaination. I didn't knew that it was possible to

do

quoted

such things. Linux is really a wonderful world ;-)

quoted

Q: Has anyone successfully completed a g_hid <-> hid-multitouch

setup? And is there a description somewhere of such setup?

I don't think so, hid-multitouch is really new (since 2.6.38, and

was

quoted

not very generic at this time).

quoted

Q: Is the Windows 7 multitouch USB HID descriptor the correct

one

quoted

to

quoted

use also for Linux hid-multitouch? Or is some tweaking necessary?

Normally, any valid Windows 7 multitouch USB HID descriptor can

be

quoted

handled by hid-multitouch.
But it's more the events that are sent to hid-multitouch that

will

quoted

tell if your device is compatible or not.

quoted

DETAILED DESCRIPTION BELOW
The device is a Linux device that uses g_hid (currently 2.6.37

kernel) and a USB HID Report Descriptor for multitouch using two-

touch

quoted

parallel mode according to Microsoft document
http://msdn.microsoft.com/en-us/windows/hardware/gg487437.

quoted

On the host (ubuntu 11.04, 2.6.38 kernel) I register

dynamically

quoted

according to the description at http://lii-
enac.fr/en/architecture/linux-input/multitouch-ubuntu-howto.html.

Using

quoted

the events/rdesc files in debugfs on Ubuntu 11.04 I have

validated

quoted

the

quoted

parsing of the report descriptor and the input reports.

This is really a bad idea to use these howto with ubuntu 11.04

ATM.

quoted

didn't found the time to update this stuff, and this branch can

not

quoted

be

quoted

used in ubuntu 11.04 (it contains code for 2.6.35 only, and there

were

quoted

changes in the hid layer after). If you only used the part "How

to

quoted

add

quoted

your device to hid-multitouch from the user space" without

updating

quoted

your hid module, then it's as if you have'nt done anything (the
generic hid module can not autodetect multitouch devices).

Currently I am using the hid-multitouch code as existing in the

source distribution of a 2.6.38 Ubuntu 11.04 fetched through apt-

get.

quoted

In the page referred to above I have only used the "How to add your
device to hid-multitouch from the user space", and not the 2.6.35
related parts.

So there are no autodetection in your kernel.

quoted

Ok, so for autodetection of multitouch devices to work I need to

update the hid-multitouch module to a newer version (than was

available

quoted

in 2.6.38), preferably head on a git master branch somewhere?

Another

quoted

option is to add a static configuration field in hid-multitouch and
hid-ids.h for my USB device's VID/PID. But then again, maybe it's
better to do that on the latest available code as well?

Autodetection is not available upstream too, sorry. The latest
available code is on Jiri's tree. I'll try to maintain a bunch of
patches to backport the work done on hid-multitouch since 2.6.35 on
http://lii-enac.fr/en/architecture/linux-input/multitouch-
howto.html#hid-multitouch
but I'm a little late for the backports right now.

Jiri's tree is at
http://git.kernel.org/?p=linux/kernel/git/jikos/hid.git;a=summary

quoted

Example multitouch input report from

/sys/kernel/debug/hid/0003:xxxx:yyyy.0002/events looks correct:

quoted

report (size 14) (numbered) =  05 03 00 a5 12 32 0e 03 01 99 14

quoted

0e

quoted

Digitizers.TipSwitch = 1
Digitizers.InRange = 1
Digitizers.ContactID = 0
GenericDesktop.X = 4773
GenericDesktop.Y = 3634
Digitizers.TipSwitch = 1
Digitizers.InRange = 1
Digitizers.ContactID = 1
GenericDesktop.X = 5273
GenericDesktop.Y = 3634
Digitizers.ContactCount = 2

Most of the time, the problematic part comes from the releases.

But

quoted

it's a good start.

With "releases" I assume you mean the finger removal from the

touch

quoted

surface, like a "touch-up" event? I have not implemented that part

yet,

quoted

but our Microsoft contact informed us that for Windows 7

compatibility

quoted

it should be reported as a final touch report with

quoted

TipSwitch = 0
InRange = 1
Is this applicable to hid-multitouch too?

Exactly, I meant touch-up. Your behavior is applicable to hid-
multitouch.

Once registering your device with hid-multitouch, you can choose a
class that controls how optimized your driver will be. Normally, the
MT_CLS_DEFAULT class will work with any Win 7 compliant device
(crossing fingers).

Then, you can optimize it by looking for the classes that have
MT_QUIRK_VALID_IS_INRANGE (or adding a new one). In the same time,

you

quoted

can also optimize the way your device actually does the match

between

quoted

slots (from 0 to N-1 with N the maximum contact count of your

device)

quoted

and the tracking_id reported by the device (or the position in the

hid

quoted

report).
For dual touches only devices, the differences are not big, but we
prefer to optimize things all the time ;-)

quoted

The first touch in the report is correctly presented both on

Windows

quoted

and on Ubuntu 11.04 (using ENAC's mtdiag tool with a minor tweak

to

quoted

display data from all devices not just multitouch ones). I can

see

quoted

in

quoted

the debugfs files that the data for the second touch data is

correctly

quoted

transferred and interpreted (see above), but the mapping to the

linux

quoted

input system seems a bit strange though (see below). My guess is

that I

quoted

have not been correctly registered as a multitouch device. I'm

assuming

quoted

this also because I cannot see any activity from the hid-

multitouch

quoted

module other than it being initialized when I connect my device
(probably another reason for the needed tweak in mtdiag above).

quoted

Example mapping from

/sys/kernel/debug/hid/0003:xxxx:yyyy.0002/rdesc

quoted

that does not look entirely correct for the second touch:

quoted

Digitizers.TipSwitch ---> Key.Touch
Digitizers.InRange ---> Key.ToolPen
Digitizers.ContactID ---> Absolute.Misc
GenericDesktop.X ---> Absolute.X
GenericDesktop.Y ---> Absolute.Y
Digitizers.TipSwitch ---> Key.Touch
Digitizers.InRange ---> Key.ToolRubber
Digitizers.ContactID ---> Absolute.?
GenericDesktop.X ---> Absolute.Z
GenericDesktop.Y ---> Absolute.Rx

This means that your device is handled by the generic usbhid and

not

quoted

hid-multitouch.

That is what troubles me. I can see in the host's dmesg that

"generic-usb" seems to be the driver chosen for my device when it is
connected to the host. And according to your comment above I might

not

quoted

have correctly registered my device in hid-multitouch which would be
the source of the problem...

quoted

As a side note; I have sprinkled printk() here and there hid-

multitouch.c and hid-core.c. While running the code I can only see

that

quoted

mt_init() and mt_exit() are called during module loading/unloading;
never do I see any evidence of that the probing or mapping functions

in

quoted

hid-multitouch are used, only the generic usb parts are used for
parsing the report descriptors and input reports.

Yep, hid-multitouch does not handle your device. There is currently

no

quoted

user-space way to tell usbhid (the generic) not to handle a device.
You'll have to add your VID-PID to the hid_have_special_driver list
(in hid-core.c) to disable generic usb hid handling for your device.
You can check the latest patches adding a specific device to
hid-multitouch to have some examples.

quoted

Could you point me to the location in the code where the decision

is

quoted

taken to use the hid-multitouch driver for a hid-multitouch

supported

quoted

device? I'm guessing it is somewhere in hid-core.c but I have not

been

quoted

able to pinpoint the location exactly. I could use this location as

quoted

start point to backtrace why we're not considered as a hid-

multitouch

quoted

supported device.

it's just above ;-) hid_have_special_driver in hid-core.c

Cheers,
Benjamin

quoted

If you are trying to inject events from the user space, I

strongly

quoted

recommend to use uinput instead. Let me know if you want to have

my

quoted

little uinput-multitouch lib.

Cheers,
Benjamin

quoted

Digitizers.ContactCount ---> Absolute.?
Button.0001 ---> Key.LeftBtn
Button.0002 ---> Key.RightBtn
Button.0003 ---> Key.MiddleBtn
GenericDesktop.X ---> Relative.X
GenericDesktop.Y ---> Relative.Y
GenericDesktop.Wheel ---> Relative.Wheel


Regards,
Pablo Cases

-------------------------
Pablo Cases, M.Sc.
Development Engineer Software
FlatFrog Laboratories AB
Magistratsvägen 10
22643 Lund
Sweden
Tel: +46 708 393816
Mail: pc@flatfrog.com
Web: www.flatfrog.com
--
To unsubscribe from this list: send the line "unsubscribe

linux-

quoted

input" in

quoted

the body of a message to majordomo@vger.kernel.org
More majordomo info at  http://vger.kernel.org/majordomo-

info.html

quoted

quoted
quoted
quoted
quoted

--
To unsubscribe from this list: send the line "unsubscribe linux-input" in
the body of a message to majordomo@vger.kernel.org
More majordomo info at  http://vger.kernel.org/majordomo-info.html

`h`	back out one level
`j`	next message in thread
`k`	previous message in thread
`l`	drill in
`Esc`	close help / fold thread tree
`?`	toggle this help