Re: Query regarding hid-multitouch.c driver in 4.14/4.19

From: Neeraj Upadhyay
Date: Sun Dec 08 2019 - 22:51:11 EST

Next message: Masahiro Yamada: "[PATCH] scripts/kallsyms: fix offset overflow of kallsyms_relative_base"
Previous message: Yan Zhao: "Re: [RFC PATCH 1/9] vfio/pci: introduce mediate ops to intercept vfio-pci ops"
Messages sorted by: [ date ] [ thread ] [ subject ] [ author ]

Hi Benjamin,

Thanks for taking a look at it and providing your inputs. We will check
in the direction suggested by you. I was off from work for some time, so
couldn't reply earlier.

Thanks
Neeraj

On 11/22/2019 2:55 PM, Benjamin Tissoires wrote:

Hi Neeraj,

On Fri, Nov 22, 2019 at 5:42 AM Neeraj Upadhyay <neeraju@xxxxxxxxxxxxxx> wrote:

Hi Benjamin,

Can you please provide guidance on how to workaround this problem? Is it
possible to support the 4.14 kernel behaviour again, so that existing
clients are not impacted?

Hmm, hard to say. Both evdev outputs from 4.14 and 4.19 are
technically valid. So I guess the issue is more that the android
version you are using is not interpreting correctly the 4.19 event
stream and probably requires some more events (like pressure or
width/height).

Also, the report descriptor is crap:

0x05, 0x0d, // Usage Page (Digitizers) 0
0x09, 0x02, // Usage (Pen) 2
0xa1, 0x01, // Collection (Application) 4
0x85, 0x01, // Report ID (1) 6
0x09, 0x22, // Usage (Finger) 8
0xa1, 0x02, // Collection (Logical) 10
0x09, 0x42, // Usage (Tip Switch) 12
0x15, 0x00, // Logical Minimum (0) 14
0x25, 0x01, // Logical Maximum (1) 16
0x75, 0x01, // Report Size (1) 18
0x95, 0x01, // Report Count (1) 20
0x81, 0x02, // Input (Data,Var,Abs) 22
0x09, 0x32, // Usage (In Range) 24
0x81, 0x02, // Input (Data,Var,Abs) 26
0x95, 0x06, // Report Count (6) 28
0x81, 0x03, // Input (Cnst,Var,Abs) 30
0x75, 0x08, // Report Size (8) 32
0x09, 0x51, // Usage (Contact Id) 34
0x95, 0x01, // Report Count (1) 36
0x81, 0x02, // Input (Data,Var,Abs) 38
0x05, 0x01, // Usage Page (Generic Desktop) 40
0x26, 0xff, 0x0f, // Logical Maximum (4095) 42
0x75, 0x10, // Report Size (16) 45
0x55, 0x0e, // Unit Exponent (-2) 47
0x65, 0x33, // Unit (Inch^3,EngLinear) 49
0x09, 0x30, // Usage (X) 51
0x35, 0x00, // Physical Minimum (0) 53
0x46, 0xb5, 0x04, // Physical Maximum (1205) 55
0x81, 0x02, // Input (Data,Var,Abs) 58
0x46, 0x8a, 0x03, // Physical Maximum (906) 60
0x09, 0x31, // Usage (Y) 63
0x81, 0x02, // Input (Data,Var,Abs) 65
0xc0, // End Collection 67
0x05, 0x0d, // Usage Page (Digitizers) 68
0x09, 0x54, // Usage (Contact Count) 70
0x95, 0x01, // Report Count (1) 72
0x75, 0x08, // Report Size (8) 74
0x81, 0x02, // Input (Data,Var,Abs) 76
0x85, 0x08, // Report ID (8) 78
0x09, 0x55, // Usage (Contact Max) 80
0x25, 0x05, // Logical Maximum (5) 82
0xb1, 0x02, // Feature (Data,Var,Abs) 84
0xc0, // End Collection 86
0x05, 0x0c, // Usage Page (Consumer Devices) 87
0x09, 0x01, // Usage (Consumer Control) 89
0xa1, 0x01, // Collection (Application) 91
0x85, 0x02, // Report ID (2) 93
0x09, 0xe9, // Usage (Volume Up) 95
0x09, 0xea, // Usage (Volume Down) 97
0x0a, 0xae, 0x01, // Usage (AL Keyboard Layout) 99
0x09, 0xe2, // Usage (Mute) 102
0x09, 0x30, // Usage (Power) 104
0x15, 0x01, // Logical Minimum (1) 106
0x25, 0x0c, // Logical Maximum (12) 108
0x75, 0x10, // Report Size (16) 110
0x95, 0x01, // Report Count (1) 112
0x81, 0x00, // Input (Data,Arr,Abs) 114
0xc0, // End Collection 116

It declares right at the beginning a Pen application collection then a
Finger logical application. This should never happen as you can not
use a pen with a bare finger.

Anyway, there are a few things you could do to solve the issue,
depending on what you are allowed to do:
- if you can tweak the kernel, then maybe adding an extra kernel
module that just binds to the device without changing anything in
probe or in events might do the trick. For extra safety you should
probably fix the report descriptor in this new kernel module
- if you can tweak bluez or the device itself, at least change the
ContactID usage to 00 (byte offset 35 above: change 0x51 to 0x00). It
would be best to actually change the whole report descriptor with the
correct applications, but, meh, this seems to be working.
- if you can tweak android, then ensure it can understand this
particular event stream

Anyway, I don't think this is an upstream problem, as the device
behaves somehow correctly by 5.3, at least it sends me the events I
expect. So I can't really spend a lot of time providing a workaround
for that.

My best bet would be to contact the device maker, and ask them to fix
their device to be working with 4.19 kernel.

Cheers,
Benjamin

Thanks
Neeraj

On 11/14/2019 10:01 AM, Neeraj Upadhyay wrote:

Hi Benjamin,

Sorry for the delay, was waiting for the required information from our
team.

On 11/13/2019 3:00 PM, Benjamin Tissoires wrote:

Hi Neeraj,

On Wed, Nov 13, 2019 at 4:11 AM Neeraj Upadhyay
<neeraju@xxxxxxxxxxxxxx> wrote:

Hi,

I have one query regarding hid-multitouch.c driver and need your
guidance on
how hid-multitouchc can restore/support the original behaviour,
where, for
devices, for which application is not
HID_DG_TOUCHSCREEN/HID_DG_TOUCHPAD, and has
HID_DG_CONTACTID usage in its report, can still use generic input
mappings.

We are using kernel versions 4.14 , 4.19 respectively in 2 different
projects:

4.14:
https://git.kernel.org/pub/scm/linux/kernel/git/stable/linux.git/tree/drivers/hid/hid-multitouch.c?h=v4.14.153

4.19:
https://git.kernel.org/pub/scm/linux/kernel/git/stable/linux.git/tree/drivers/hid/hid-multitouch.c?h=v4.19.83

I checked the application for our hid device, it's HID_DG_PEN, device
also has a HID_DG_CONTACTID usage defined in

its report.

In 4.19, is_mt_collection is set to 'true'. All multitouch code paths or
input mapping is configured

mt_allocate_report_data()
...
for (n = 0; n < field->report_count; n++) {
if (field->usage[n].hid == HID_DG_CONTACTID)
rdata->is_mt_collection = true; //
is_mt_collection is set to 'true'
}
}

mt_input_mapping()
...
if (rdata->is_mt_collection)
return mt_touch_input_mapping(...) //
mt_touch_input_mapping() is called

mt_event()
if (rdata && rdata->is_mt_collection)
return mt_touch_event(); // mt_touch_event() is called

However, in 4.14, the behaviour was different, mt input mapping was done
only
for HID_DG_TOUCHSCREEN/HID_DG_TOUCHPAD , and because our hid device is
HID_DG_PEN, generic mappings were applied for it; with these settings,
device
responds to events.

static int mt_input_mapping()
if (field->application == HID_DG_TOUCHSCREEN ||
field->application == HID_DG_TOUCHPAD)
return mt_touch_input_mapping(); // This is not called.

mt_touch_input_mapping()
case HID_DG_CONTACTID:
mt_store_field(usage, td, hi);
td->touches_by_report++;
td->mt_report_id = field->report->id; //
mt_report_id is not set.
return 1;

Looks like this behaviour changed, with below commits:

https://git.kernel.org/pub/scm/linux/kernel/git/stable/linux.git/commit/drivers/hid/hid-multitouch.c?h=v4.19.83&id=8dfe14b3b47ff832cb638731f9fc696a3a84f804

8dfe14b3b47f HID: multitouch: ditch mt_report_id
https://git.kernel.org/pub/scm/linux/kernel/git/stable/linux.git/commit/drivers/hid/hid-multitouch.c?h=v4.19.83&id=ba6b055e0f3b4ff4942e4ab273260affcfad9bff

ba6b055e0f3b HID: input: enable Totem on the Dell Canvas 27

Can you please suggest on how we can support/preserve the original
behaviour?

Hmm, I would initially say that a firmware that exports Contact ID for
a Pen is definitely wrong. The Contact ID usage has been introduced in
https://www.usb.org/sites/default/files/hutrr34.pdf and is
specifically for multi-touch, not multi pen.

Anyway, couple of questions:
- does the device supports multi-pen?

Actually the device is a selfie stick device:
https://item.jd.com/33082497741.html

It does not support multi-pen.

- can you share the report descriptor and a few events when triggering
this particular report (ideally with hid-recorder from
https://gitlab.freedesktop.org/libevdev/hid-tools/

Report descriptor is below:

05 0d 09 02 a1 01 85 01 09 22 a1 02 09 42 15 00 25 01 75 01 95 01 81 02
09 32 81 02 95 06 81 03 75 08 09 51 95 01 81 02 05 01 26 ff 0f 75 10 55
0e 65 33 09 30 35 00 46 b5 04 81 02 46 8a 03 09 31 81 02 c0 05 0d 09 54
95 01 75 08 81 02 85 08 09 55 25 05 b1 02 c0 05 0c 09 01 a1 01 85 02 09
e9 09 ea 0a ae 01 09 e2 09 30 15 01 25 0c 75 10 95 01 81 00 c0

Events were collected using getevent call in adb shell in android:

On 4.19

# getevent -l

add device 7: /dev/input/event6
name: "BLE-M1 UNKNOWN"

/dev/input/event6: EV_ABS ABS_MT_TRACKING_ID 00000000
/dev/input/event6: EV_ABS ABS_MT_POSITION_X 00000800
/dev/input/event6: EV_ABS ABS_MT_POSITION_Y 00000d60
/dev/input/event6: EV_KEY BTN_TOUCH DOWN
/dev/input/event6: EV_SYN SYN_REPORT 00000000
/dev/input/event6: EV_ABS ABS_MT_TRACKING_ID ffffffff
/dev/input/event6: EV_KEY BTN_TOUCH UP
/dev/input/event6: EV_SYN SYN_REPORT 00000000
/dev/input/event6: EV_ABS ABS_MT_TRACKING_ID 00000001
/dev/input/event6: EV_KEY BTN_TOUCH DOWN
/dev/input/event6: EV_SYN SYN_REPORT 00000000
/dev/input/event6: EV_ABS ABS_MT_TRACKING_ID ffffffff
/dev/input/event6: EV_KEY BTN_TOUCH UP
/dev/input/event6: EV_SYN SYN_REPORT 00000000

On 4.14

add device 2: /dev/input/event5
name: "BLE-M1 UNKNOWN"

/dev/input/event5: EV_MSC MSC_SCAN 000d0042
/dev/input/event5: EV_KEY BTN_TOUCH DOWN
/dev/input/event5: EV_KEY BTN_DIGI DOWN
/dev/input/event5: EV_ABS ABS_MISC 00000001
/dev/input/event5: EV_ABS ABS_X 00000800
/dev/input/event5: EV_ABS ABS_Y 00000d60
/dev/input/event5: EV_ABS 0029 00000001
/dev/input/event5: EV_SYN SYN_REPORT 00000000

/dev/input/event5: EV_MSC MSC_SCAN 000d0042
/dev/input/event5: EV_KEY BTN_TOUCH UP
/dev/input/event5: EV_KEY BTN_DIGI UP
/dev/input/event5: EV_ABS 0029 00000000
/dev/input/event5: EV_SYN SYN_REPORT 00000000

/dev/input/event5: EV_MSC MSC_SCAN 000d0042
/dev/input/event5: EV_KEY BTN_TOUCH DOWN
/dev/input/event5: EV_KEY BTN_DIGI DOWN
/dev/input/event5: EV_ABS 0029 00000001
/dev/input/event5: EV_SYN SYN_REPORT 00000000

/dev/input/event5: EV_MSC MSC_SCAN 000d0042
/dev/input/event5: EV_KEY BTN_TOUCH UP
/dev/input/event5: EV_KEY BTN_DIGI UP
/dev/input/event5: EV_ABS 0029 00000000
/dev/input/event5: EV_SYN SYN_REPORT 00000000

As I have little understanding of the framework, use cases and of the
flow, I am sorry, if the information provided above is

incomplete (w.r.t. what you were expecting).

Thanks

Neeraj

Cheers,
Benjamin

Thanks
Neeraj

--
QUALCOMM INDIA, on behalf of Qualcomm Innovation Center, Inc. is a
member of the Code Aurora Forum, hosted by The Linux Foundation

--
QUALCOMM INDIA, on behalf of Qualcomm Innovation Center, Inc. is a
member of the Code Aurora Forum, hosted by The Linux Foundation

--
QUALCOMM INDIA, on behalf of Qualcomm Innovation Center, Inc. is a member of the Code Aurora Forum, hosted by The Linux Foundation

Next message: Masahiro Yamada: "[PATCH] scripts/kallsyms: fix offset overflow of kallsyms_relative_base"
Previous message: Yan Zhao: "Re: [RFC PATCH 1/9] vfio/pci: introduce mediate ops to intercept vfio-pci ops"
Messages sorted by: [ date ] [ thread ] [ subject ] [ author ]