Hi Benjamin,
I think I've found the decisive evidence showing that there's
something wrong with the GPIO chip's interrupt setting of this laptop
model of Lenovo Legion-5 15ARH05 but there's nothing wrong with the
touchpad like scanning frequency, report rate, etc.
This GPIO interrupt controller has been assigned with IRQ#7 and
the touchpad's interrupt line is connected to the GPIO chip's
pin#130.
Device (GPIO)
{
Name (_HID, "AMDI0030") // _HID: Hardware ID
Name (_CID, "AMDI0030") // _CID: Compatible ID
Name (_UID, Zero) // _UID: Unique ID
Method (_CRS, 0, NotSerialized) // _CRS: Current Resource Settings
{
Name (RBUF, ResourceTemplate ()
{
Interrupt (ResourceConsumer, Level, ActiveLow, Shared, ,, )
{
0x00000007,
}
})
Return (RBUF) /* \_SB_.GPIO._CRS.RBUF */
}
}
Scope (_SB.I2CD)
{
Device (TPDD)
{
Name (_HID, "XXXX0000") // _HID: Hardware ID
Name (_CID, "PNP0C50" /* HID Protocol Device (I2C bus) */) // _CID: Compatible ID
Name (_SUB, "XXXX0000") // _SUB: Subsystem ID
Method (_INI, 0, NotSerialized) // _INI: Initialize
{
If ((TPTY == One))
{
_HID = "MSFT0001"
_SUB = "ELAN0001"
}
If ((TPTY == 0x02))
{
_HID = "MSFT0001"
_SUB = "SYNA0001"
}
Return (Zero)
}
Method (_CRS, 0, NotSerialized) // _CRS: Current Resource Settings
{
If ((TPTY == One))
{
Name (SBFB, ResourceTemplate ()
{
I2cSerialBusV2 (0x0015, ControllerInitiated, 0x00061A80,
AddressingMode7Bit, "\\_SB.I2CD",
0x00, ResourceConsumer, , Exclusive,
)
})
}
If ((TPTY == 0x02))
{
Name (SBFS, ResourceTemplate ()
{
I2cSerialBusV2 (0x002C, ControllerInitiated, 0x00061A80,
AddressingMode7Bit, "\\_SB.I2CD",
0x00, ResourceConsumer, , Exclusive,
)
})
}
Name (SBFI, ResourceTemplate ()
{
GpioInt (Level, ActiveLow, ExclusiveAndWake, PullUp, 0x0000,
"\\_SB.GPIO", 0x00, ResourceConsumer, ,
)
{ // Pin list
0x0082
}
When there's new data from the touchpad, IRQ#7 will be activated. The
handler amd_gpio_irq_handler will iterate over the pins and find
pin#130 is the source of the interrupt. Then i2c_hid_irq will be
woken up to read HID reports. Here's the code execution path,
<IRQ>
dump_stack+0x64/0x88
__irq_wake_thread.cold+0x9/0x12
__handle_irq_event_percpu+0x80/0x1c0
handle_irq_event+0x58/0xb0
handle_level_irq+0xb7/0x1a0
generic_handle_irq+0x4a/0x60
amd_gpio_irq_handler+0x15f/0x1b0 [pinctrl_amd]
__handle_irq_event_percpu+0x45/0x1c0
handle_irq_event+0x58/0xb0
handle_fasteoi_irq+0xa2/0x210
do_IRQ+0x70/0x120
common_interrupt+0xf/0xf
</IRQ>
So I add some debugging code to track when each IRQ#7 will be handled
and what's the status of pin#130 as I move my finger on the surface of
the touchpad for about 0.5s. It turns out i2c_hid_irq can handle ~500
interrupts per second but IRQ#7 is only fired ~7 times per second.
During the time of I move the finger, pin#130 has the input of low
("active low", meaning touchpad has asserted its interrupt line to
indicate new data) most of the time. So my next step is to find out
why IRQ#7 fails to fire when the touchpad has received new data.
Please check the attachments for the details,
- handler_irq72: timing of hanlder of IRQ#7
- touchpad_interrupt: the status of touchpad interrupt line
- handler_irq7_touchpad_interrupt.pdf: the plotting
Btw, I've added a polling mode for rivers/hid/i2c-hid/i2c-hid-core.c
when trying to fix the touchpad issue. With the polling mode, the
cursor is able to follow the movement of my finger.
Patches of debugging code
========================#
When the kernel receives IRQ#7
$ diff kernel/irq/chip.{ori,c} -u
--- kernel/irq/chip.ori 2020-09-22 22:56:33.684265452 +0800
+++ kernel/irq/chip.c 2020-09-22 23:25:40.435297613 +0800
@@ -699,6 +699,12 @@
void handle_fasteoi_irq(struct irq_desc *desc)
{
struct irq_chip *chip = desc->irq_data.chip;
+ struct timespec64 ts1, ts2;
+ unsigned int irq = irq_desc_get_irq(desc);
+
+ if (irq == 7) {
+ ktime_get_ts64(&ts1);
+ }
raw_spin_lock(&desc->lock);
@@ -727,6 +733,11 @@
cond_unmask_eoi_irq(desc, chip);
raw_spin_unlock(&desc->lock);
+ if (irq == 7) {
+ ktime_get_ts64(&ts2);
+ pr_alert("handle_fasteoi_irq: %d, %llu %llu, %llu %llu", irq, ts1.tv_sec, ts1.tv_nsec, ts2.tv_sec-ts1.tv_sec, ts2.tv_nsec-ts1.tv_nsec);
+ dump_stack();
+ }
return;
out:
if (!(chip->flags & IRQCHIP_EOI_IF_HANDLED))
When the kernel finished processing IRQ#7
$ diff drivers/pinctrl/pinctrl-amd.{c,ori} -u -p
--- drivers/pinctrl/pinctrl-amd.c 2020-09-24 14:51:50.815843952 +0800
+++ drivers/pinctrl/pinctrl-amd.ori 2020-09-20 11:57:40.867390865 +0800
@@ -389,7 +389,6 @@ static void amd_gpio_irq_unmask(struct i
{
u32 pin_reg;
unsigned long flags;
- struct timespec64 ts1;
struct gpio_chip *gc = irq_data_get_irq_chip_data(d);
struct amd_gpio *gpio_dev = gpiochip_get_data(gc);
@@ -397,8 +396,6 @@ static void amd_gpio_irq_unmask(struct i
pin_reg = readl(gpio_dev->base + (d->hwirq)*4);
pin_reg |= BIT(INTERRUPT_MASK_OFF);
writel(pin_reg, gpio_dev->base + (d->hwirq)*4);
- ktime_get_ts64(&ts1);
- pr_alert("gpio_unmask %u: %llu %llu", d->hwirq, ts1.tv_sec, ts1.tv_nsec);
raw_spin_unlock_irqrestore(&gpio_dev->lock, flags);
}
Polling the status of pin#130 every 1ms for 1 second
$ diff drivers/hid/i2c-hid/i2c-hid-core.{ori,c} -u -p
--- drivers/hid/i2c-hid/i2c-hid-core.ori 2020-09-10 12:35:36.160989731 +0800
+++ drivers/hid/i2c-hid/i2c-hid-core.c 2020-09-24 14:54:17.481705797 +0800
@@ -522,10 +522,54 @@ static void i2c_hid_get_input(struct i2c
return;
}
+static struct task_struct *thread_hid;
+
+#include <linux/gpio/driver.h>
+#include <linux/kthread.h>
+#define times 1000
+int get_pin_state_thread(void *irq_d)
+{
+ struct irq_desc *irq_desc = irq_d;
+ struct gpio_chip *gc = irq_data_get_irq_chip_data(&irq_desc->irq_data);
+ int i;
+ int *stat;
+ struct timespec64 *ts;
+
+ stat = kmalloc(sizeof(int) * times, GFP_KERNEL);
+ ts = kmalloc(sizeof(struct timespec64) * times, GFP_KERNEL);
+
+ for (i = 0; i < times; i++) {
+ usleep_range(1000, 1000);
+ stat[i] = gc->get(gc, 130);
+ ktime_get_ts64(&ts[i]);
+ }
+
+ for (i = 0; i < times; i++) {
+ pr_alert("pin130_state: %d %lld %ld %d", i, ts[i].tv_sec, ts[i].tv_nsec, stat[i]);
+ }
+ kfree(stat);
+ kfree(ts);
+ thread_hid = NULL;
+ do_exit(0);
+ return 0;
+ /** struct amd_gpio *gpio_dev = gpiochip_get_data(gc); */
+
+}
+
static irqreturn_t i2c_hid_irq(int irq, void *dev_id)
{
struct i2c_hid *ihid = dev_id;
+ struct irq_desc *irq_desc;
+ irq_desc = irq_to_desc(irq);
+ if (!thread_hid) {
+ thread_hid = kthread_create(get_pin_state_thread, irq_desc, "mythread_ihid");
+ if (thread_hid) {
+ ktime_get_ts64(&ts1);
+ pr_alert("hid_thread created: %lld %ld", ts1.tv_sec, ts1.tv_nsec);
+ wake_up_process(thread_hid);
+ }
+ }
if (test_bit(I2C_HID_READ_PENDING, &ihid->flags))
return IRQ_HANDLED;
On Tue, Sep 15, 2020 at 06:14:51PM +0800, Coiby Xu wrote:
On Fri, Sep 11, 2020 at 02:29:03PM +0200, Benjamin Tissoires wrote:
Hi,
On Thu, Sep 10, 2020 at 6:37 PM Coiby Xu <coiby.xu@xxxxxxxxx> wrote:
On Fri, Sep 04, 2020 at 06:47:42PM +0200, Benjamin Tissoires wrote:
On Fri, Sep 4, 2020 at 4:59 PM Coiby Xu <coiby.xu@xxxxxxxxx> wrote:
On Fri, Sep 04, 2020 at 10:16:51AM +0200, Benjamin Tissoires wrote:
Hi,
On Tue, Aug 11, 2020 at 11:15 AM Coiby Xu <coiby.xu@xxxxxxxxx> wrote:
Hi,
I'm working on a touchpad device issue as reported on
https://bugs.launchpad.net/ubuntu/+source/linux/+bug/1887190.
This touchpad device MSFT0001:00 04F3:Touchpad should be handled by
hid_rmi. But currently hid-core.c chooses hid_multitouch by mistake,
1. When scanning this device's report descriptor, HID_DG_CONTACTID
usage is found. Thus group HID_GROUP_MULTITOUCH is assigned to
the device.
2. The flag HID_SCAN_FLAG_MT_WIN_8 is also found. Thus group
HID_GROUP_MULTITOUCH_WIN_8 is assigned to the device.
3. hid-multitouch.c claims handling devices with the group of
HID_GROUP_MULTITOUCH_WIN_8
static const struct hid_device_id mt_devices[] = {
/* Generic MT device */
{ HID_DEVICE(HID_BUS_ANY, HID_GROUP_MULTITOUCH, HID_ANY_ID, HID_ANY_ID) },
/* Generic Win 8 certified MT device */
{ .driver_data = MT_CLS_WIN_8,
HID_DEVICE(HID_BUS_ANY, HID_GROUP_MULTITOUCH_WIN_8,
HID_ANY_ID, HID_ANY_ID) },
{ }
};
There are several potential solutions,
- Let the device vendor fix this problem since this device's report
descriptor shouldn't have the HID_DG_CONTACTID usage.
- Make it a special case by setting the device's group to
HID_GROUP_RMI in hid_scan_report when vendor id and product ID
are matched.
- hid-quirks.c seems to be designed to handle special cases, is it
suitable for this case?
AFAIU, the touchpad doesn't work at all with hid-multitouch. So I
guess the best is to add the VID/PID to hid-quirks.c in
hid_have_special_driver[], and add it to the hid-rmi driver too.
This way, you will ensure hid-rmi will pick up the device all the time.
Cheers,
Benjamin
Thank you for the advice! I have exactly adopted this approach by looking
at commit e9287099ba6539bccb20cd791269186f3ae28b85
("HID: rmi: Add support for the touchpad in the Razer Blade 14 laptop")
as an example.
My previous email is a bit misleading because 0x04F3 is the vendor code
of ELAN while hid-rmi is for the Synaptics touchpad. And actually this
laptop model of Lenovo Legion-5 15ARH05 is shipped with both kinds of
touchpads,
- for the Synaptics touchpad, hid-rmi could almost handle it perfectly
except the clicking is not sensitive enough. I need to let my finger
linger on the touchpad for a while. I notice when I click on the touchpad,
an HID report would be received by hid-recorder. But evtest couldn't receive
any EV_ event. If hid-multitouch is handling the device, the cursor
won't move but 2-4 finger touching events could still be received by
evtest.
- for the ELAN touchpad, only HID reporters can be read and parsed by
hid-core then two input devices (mouse and touchpad) could created by
hid-multitouch as seen from /proc/bus/input/devices. But hid-recorder
could never get any HID report.
huh. So in both cases you have a buggy touchpad with hid-multitouch :(
Do both touchpads share the same VID/PID? If so, this is going to be
interesting to decide how any of those touchpad should be handled.
No, they have different VID/PIDs,
- 04F3:3140 (Elan)
- 06CB:7F28 (Synaptics)
That's good to know :)
AFAIU, isn't hid-multitouch supposed to be the implementation of Windows
Precision Touchpad?
yes, it works for most of those. The only ones that are not working
are usually because OEM or device makers tend to do "fun" things.
And since Precision touchpad is mandatory for
Windows 10 notebooks and this laptop model of Lenovo Legion-5 15ARH05
seem to have been certificated by Windows 10, does it mean by theory
hid-multiouch could handle these two touchpad devices?
Well, it should, yes, but it clearly can not. You can try to give a
try at hid-recorder from
https://gitlab.freedesktop.org/libevdev/hid-tools. This will show what
is actually exported by the hardware before any processing by the
kernel. Maybe there is a new simple thing to do in hid-multitouch to
handle those devices.
Also, last time I heard, Synaptics dropped the use of hid-rmi in favor
of hid-multitouch. Any hid-rmi touchpad should be able to use
hid-multitouch, as this is the preferred way on Windows. But sometimes
the various teams decide to change the rules.
Anyway, it seems I need to install Windows Driver Kit to capture&analyze
HID reports to see what's happening. Or do you have any suggestion?
Unless it changed recently (I think I have seen something like that
recently), I had to write a I2C man-in-the-middle to dump the logs
from Windows. Project is at
https://github.com/bentiss/SimplePeripheralBusProbe and requires a
little bit of manual work to be able to start capturing data :(
Cheers,
Benjamin
--
Best regards,
Coiby
Thank you for developing hid-tools and SimplePeripheralBusProbe!
No worries.
Wow, I am really impressed in how quickly you managed to get all that
data. I think it's the first time I ever point somebody at
SimplePeripheralBusProbe, and I don't have to do anything to get this
up and running, and analyzed :)
Thank you! The instructions of setting up SimplePeripheralBusProbe
are well-documented so it's quite a straightforward process:)
The problem with the Synaptics touchpad when it's handled by hid-multiouch
is the pointer won't move. A few days ago, I took another look at the
results produced by hid-recorder and evtest and noticed something abnormal
for the Synaptics touchpad. For a straight continuous motion of one finger,
the value of ABS_MT_TRACKING_ID kept changing,
$ sudo evtest
...
Event: time 1599651409.969002, type 3 (EV_ABS), code 57 (ABS_MT_TRACKING_ID), value 68
Event: time 1599651409.969002, type 3 (EV_ABS), code 53 (ABS_MT_POSITION_X), value 401
Event: time 1599651409.969002, type 3 (EV_ABS), code 54 (ABS_MT_POSITION_Y), value 201
Event: time 1599651409.969002, type 1 (EV_KEY), code 330 (BTN_TOUCH), value 1
Event: time 1599651409.969002, type 1 (EV_KEY), code 325 (BTN_TOOL_FINGER), value 1
Event: time 1599651409.969002, type 3 (EV_ABS), code 0 (ABS_X), value 401
Event: time 1599651409.969002, type 3 (EV_ABS), code 1 (ABS_Y), value 201
Event: time 1599651409.969002, type 4 (EV_MSC), code 5 (MSC_TIMESTAMP), value 0
Event: time 1599651409.969002, -------------- SYN_REPORT ------------
Event: time 1599651410.070105, type 3 (EV_ABS), code 57 (ABS_MT_TRACKING_ID), value -1
Event: time 1599651410.070105, type 1 (EV_KEY), code 330 (BTN_TOUCH), value 0
Event: time 1599651410.070105, type 1 (EV_KEY), code 325 (BTN_TOOL_FINGER), value 0
Event: time 1599651410.070105, -------------- SYN_REPORT ------------
Event: time 1599651410.093740, type 3 (EV_ABS), code 57 (ABS_MT_TRACKING_ID), value 69
Event: time 1599651410.093740, type 1 (EV_KEY), code 330 (BTN_TOUCH), value 1
Event: time 1599651410.093740, type 1 (EV_KEY), code 325 (BTN_TOOL_FINGER), value 1
Event: time 1599651410.093740, type 4 (EV_MSC), code 5 (MSC_TIMESTAMP), value 7200
Event: time 1599651410.093740, -------------- SYN_REPORT ------------
Event: time 1599651410.196576, type 3 (EV_ABS), code 57 (ABS_MT_TRACKING_ID), value -1
Event: time 1599651410.196576, type 1 (EV_KEY), code 330 (BTN_TOUCH), value 0
Event: time 1599651410.196576, type 1 (EV_KEY), code 325 (BTN_TOOL_FINGER), value 0
Event: time 1599651410.196576, -------------- SYN_REPORT ------------
Event: time 1599651410.219020, type 3 (EV_ABS), code 57 (ABS_MT_TRACKING_ID), value 70
Event: time 1599651410.219020, type 3 (EV_ABS), code 53 (ABS_MT_POSITION_X), value 406
Event: time 1599651410.219020, type 3 (EV_ABS), code 54 (ABS_MT_POSITION_Y), value 204
Event: time 1599651410.219020, type 1 (EV_KEY), code 330 (BTN_TOUCH), value 1
Event: time 1599651410.219020, type 1 (EV_KEY), code 325 (BTN_TOOL_FINGER), value 1
Event: time 1599651410.219020, type 3 (EV_ABS), code 0 (ABS_X), value 406
Event: time 1599651410.219020, type 3 (EV_ABS), code 1 (ABS_Y), value 204
Event: time 1599651410.219020, type 4 (EV_MSC), code 5 (MSC_TIMESTAMP), value 129000
Event: time 1599651410.219020, -------------- SYN_REPORT ------------
Event: time 1599651410.320097, type 3 (EV_ABS), code 57 (ABS_MT_TRACKING_ID), value -1
Event: time 1599651410.320097, type 1 (EV_KEY), code 330 (BTN_TOUCH), value 0
Event: time 1599651410.320097, type 1 (EV_KEY), code 325 (BTN_TOOL_FINGER), value 0
Event: time 1599651410.320097, -------------- SYN_REPORT ------------
Event: time 1599651410.343867, type 3 (EV_ABS), code 57 (ABS_MT_TRACKING_ID), value 71
Event: time 1599651410.343867, type 3 (EV_ABS), code 53 (ABS_MT_POSITION_X), value 424
Event: time 1599651410.343867, type 3 (EV_ABS), code 54 (ABS_MT_POSITION_Y), value 218
Event: time 1599651410.343867, type 1 (EV_KEY), code 330 (BTN_TOUCH), value 1
Event: time 1599651410.343867, type 1 (EV_KEY), code 325 (BTN_TOOL_FINGER), value 1
Event: time 1599651410.343867, type 3 (EV_ABS), code 0 (ABS_X), value 424
Event: time 1599651410.343867, type 3 (EV_ABS), code 1 (ABS_Y), value 218
Event: time 1599651410.343867, type 4 (EV_MSC), code 5 (MSC_TIMESTAMP), value 250900
...
while hid-recorder showed "Tip Switch" remained 1 and the "contact Id"
remained 0 for this set of reports,
$ sudo hid-recorder
# ReportID: 3 / Confidence: 1 | Tip Switch: 1 | Contact Id: 0 | # | X: 401 | Y: 201
# | Confidence: 0 | Tip Switch: 0 | Contact Id: 0 | # | X: 0 | Y: 0
# | Confidence: 0 | Tip Switch: 0 | Contact Id: 0 | # | X: 0 | Y: 0
# | Confidence: 0 | Tip Switch: 0 | Contact Id: 0 | # | X: 0 | Y: 0
# | Confidence: 0 | Tip Switch: 0 | Contact Id: 0 | # | X: 0 | Y: 0 | Scan Time: 14579 | Contact Count: 1 | Button: 0 | #
E: 000000.000000 30 03 03 91 01 c9 00 00 00 00 00 00 00 00 00 00 00 00 00 00 00 00 00 00 00 00 00 f3 38 01 00
# ReportID: 3 / Confidence: 1 | Tip Switch: 1 | Contact Id: 0 | # | X: 401 | Y: 201
# | Confidence: 0 | Tip Switch: 0 | Contact Id: 0 | # | X: 0 | Y: 0
# | Confidence: 0 | Tip Switch: 0 | Contact Id: 0 | # | X: 0 | Y: 0
# | Confidence: 0 | Tip Switch: 0 | Contact Id: 0 | # | X: 0 | Y: 0
# | Confidence: 0 | Tip Switch: 0 | Contact Id: 0 | # | X: 0 | Y: 0 | Scan Time: 14651 | Contact Count: 1 | Button: 0 | #
E: 000000.124738 30 03 03 91 01 c9 00 00 00 00 00 00 00 00 00 00 00 00 00 00 00 00 00 00 00 00 00 3b 39 01 00
# ReportID: 3 / Confidence: 1 | Tip Switch: 1 | Contact Id: 0 | # | X: 406 | Y: 204
# | Confidence: 0 | Tip Switch: 0 | Contact Id: 0 | # | X: 0 | Y: 0
# | Confidence: 0 | Tip Switch: 0 | Contact Id: 0 | # | X: 0 | Y: 0
# | Confidence: 0 | Tip Switch: 0 | Contact Id: 0 | # | X: 0 | Y: 0
# | Confidence: 0 | Tip Switch: 0 | Contact Id: 0 | # | X: 0 | Y: 0 | Scan Time: 15869 | Contact Count: 1 | Button: 0 | #
E: 000000.249932 30 03 03 96 01 cc 00 00 00 00 00 00 00 00 00 00 00 00 00 00 00 00 00 00 00 00 00 fd 3d 01 00
# ReportID: 3 / Confidence: 1 | Tip Switch: 1 | Contact Id: 0 | # | X: 424 | Y: 218
# | Confidence: 0 | Tip Switch: 0 | Contact Id: 0 | # | X: 0 | Y: 0
# | Confidence: 0 | Tip Switch: 0 | Contact Id: 0 | # | X: 0 | Y: 0
# | Confidence: 0 | Tip Switch: 0 | Contact Id: 0 | # | X: 0 | Y: 0
# | Confidence: 0 | Tip Switch: 0 | Contact Id: 0 | # | X: 0 | Y: 0 | Scan Time: 17088 | Contact Count: 1 | Button: 0 | #
E: 000000.374781 30 03 03 a8 01 da 00 00 00 00 00 00 00 00 00 00 00 00 00 00 00 00 00 00 00 00 00 c0 42 01 00
# ReportID: 3 / Confidence: 1 | Tip Switch: 1 | Contact Id: 0 | # | X: 449 | Y: 251
# | Confidence: 0 | Tip Switch: 0 | Contact Id: 0 | # | X: 0 | Y: 0
# | Confidence: 0 | Tip Switch: 0 | Contact Id: 0 | # | X: 0 | Y: 0
# | Confidence: 0 | Tip Switch: 0 | Contact Id: 0 | # | X: 0 | Y: 0
# | Confidence: 0 | Tip Switch: 0 | Contact Id: 0 | # | X: 0 | Y: 0 | Scan Time: 18379 | Contact Count: 1 | Button: 0 | #
E: 000000.499704 30 03 03 c1 01 fb 00 00 00 00 00 00 00 00 00 00 00 00 00 00 00 00 00 00 00 00 00 cb 47 01 00
# ReportID: 3 / Confidence: 1 | Tip Switch: 1 | Contact Id: 0 | # | X: 475 | Y: 281
# | Confidence: 0 | Tip Switch: 0 | Contact Id: 0 | # | X: 0 | Y: 0
# | Confidence: 0 | Tip Switch: 0 | Contact Id: 0 | # | X: 0 | Y: 0
# | Confidence: 0 | Tip Switch: 0 | Contact Id: 0 | # | X: 0 | Y: 0
# | Confidence: 0 | Tip Switch: 0 | Contact Id: 0 | # | X: 0 | Y: 0 | Scan Time: 19598 | Contact Count: 1 | Button: 0 | #
I managed to let hid-multitouch to send the set of input events with the
same ABS_MT_TRACKING_ID by disabling the release_timer via commenting out
the following code,
// drivers/hid/hid-multitouch.c
static void mt_touch_report(struct hid_device *hid,
struct mt_report_data *rdata)
{
...
* Windows 8 specs says 2 things:
* - once a contact has been reported, it has to be reported in each
* subsequent report
* - the report rate when fingers are present has to be at least
* the refresh rate of the screen, 60 or 120 Hz
*
* I interprete this that the specification forces a report rate of
* at least 60 Hz for a touchscreen to be certified.
* Which means that if we do not get a report whithin 16 ms, either
* something wrong happens, either the touchscreen forgets to send
* a release. Taking a reasonable margin allows to remove issues
* with USB communication or the load of the machine.
*
* Given that Win 8 devices are forced to send a release, this will
* only affect laggish machines and the ones that have a firmware
* defect.
*/
/* if (app->quirks & MT_QUIRK_STICKY_FINGERS) {
if (test_bit(MT_IO_FLAGS_PENDING_SLOTS, &td->mt_io_flags))
mod_timer(&td->release_timer,
jiffies + msecs_to_jiffies(100));
else
del_timer(&td->release_timer);
} */
...
}
Now the pointer can move but there is a noticeable lag as I move my
finger. evhz shows the report rate is only ~7Hz. While under Windows,
SimplePeripheralBusProbe shows there are 25+ reports for one finger
sliding event. This also explains why under Linux release_timer keeping
inactivating report slots.
Yep, you found the root cause. We don't have enough reports to ensure
a proper reporting mechanism.
Comparing the hid commands sent between Windows and Linux, they are
almost identical. SimplePeripheralBusProbe doesn't record hid commands
like fetching HID descriptor, report descriptor, resetting the device.
I haven't yet found a clue about why this Synaptics touchpad operates
with such a report rate under Linux.
Under Linux, the follow hid commands are sent to the touchpad device.
Sep 10 06:54:13 Rk kernel: __i2c_hid_command: cmd=20 00
Sep 10 06:54:13 Rk kernel: __i2c_hid_command: cmd=22 00 00 08
Sep 10 06:54:13 Rk kernel: __i2c_hid_command: cmd=22 00 00 01
Sep 10 06:54:13 Rk kernel: __i2c_hid_command: cmd=22 00 00 08
Sep 10 06:54:13 Rk kernel: __i2c_hid_command: cmd=21 00
Sep 10 06:54:13 Rk kernel: __i2c_hid_command: cmd=22 00 38 02 23 00
Sep 10 06:54:13 Rk kernel: __i2c_hid_command: cmd=22 00 38 02 23 00
Sep 10 06:54:13 Rk kernel: __i2c_hid_command: cmd=22 00 37 02 23 00
Sep 10 06:54:13 Rk kernel: __i2c_hid_command: cmd=22 00 3d 03 23 00 04 00 0d 00
Sep 10 06:54:13 Rk kernel: __i2c_hid_command: cmd=22 00 34 03 23 00 04 00 04 03
Sep 10 06:54:13 Rk kernel: __i2c_hid_command: cmd=22 00 36 03 23 00 04 00 06 03
whereas under Windows,
[0]0000.0000::09/09/2020-14:41:47.732 [SimplePeripheralBusProbe]device 1: ##00 write 6 - 0000: 22 00 38 02 23 00
[0]0000.0000::09/09/2020-14:41:47.740 [SimplePeripheralBusProbe]device 1: ##00 write 6 - 0000: 22 00 37 02 23 00
[0]0000.0000::09/09/2020-14:41:47.741 [SimplePeripheralBusProbe]device 1: ##00 write 10 - 0000: 22 00 34 03 23 00 04 00 04 03
[0]0000.0000::09/09/2020-14:41:47.742 [SimplePeripheralBusProbe]device 1: ##00 write 10 - 0000: 22 00 36 03 23 00 04 00 06 03
[0]0000.0000::09/09/2020-14:41:47.742 [SimplePeripheralBusProbe]device 1: ##00 write 10 - 0000: 22 00 34 03 23 00 04 00 04 03
[0]0000.0000::09/09/2020-14:41:47.742 [SimplePeripheralBusProbe]device 1: ##00 write 10 - 0000: 22 00 3d 03 23 00 04 00 0d 00
[0]1AB4.3074::09/09/2020-14:41:47.743 [SimplePeripheralBusProbe]device 1: ##00 write 10 - 0000: 22 00 36 03 23 00 04 00 06 03
Thanks for comparing the 2. As you noticed, there doesn't seem to be
any differences.
My initial thoughts were that we were wrongly setting the "Latency
mode" [0]. But this should have been shown in the differences. I
*think* we set it in hid-multitouch, but it can't hurt to do a couple
of additional tests on this side.
I think "latency mode" has been correctly set based on the following
tests,
- When HID_LATENCY_NORMAL is replaced with HID_LATENCY_HIGH when calling
mt_set_modes, the cursor would never move even with release_timer
disabled.
- hid-generic could take over this device if hid-multitouch is
blacklisted. evhz shows the report rate is still only ~7Hz.
- hid-rmi could also handle the device and the report rate is also
~7Hz. According to "Synaptics RMI4 Specification", the control
registers has the ReportRate (F01_RMI_Ctrl0, bit 6),
> This field sets the report rate for the device. It applies in common to all functions on the
device that have a natural report rate.
When this bit is set, there is no change observed.
I also did some tests on power setting because according to
"HID Over I2C Protocol Specification" [1], the device's sensing
frequency could drop to 10Hz in sleep mode,
DEVICE reduces its sensing frequency: The DEVICE reduces the frequency
at which it scans for data (e.g. if the digitizer does not sense a human
finger is present, it samples at 10Hz as compared to 100Hz. However once
user interaction is detected, it increases its sensing interval).
- Not power managing the device doesn't work when hid-multitouch
handles the device,
$ echo "on" | sudo tee /sys/bus/i2c/devices/i2c-MSFT0001:00/power/control
- Set NoSleep (F01_RMI_Ctrl0, bit 2) when hid-rmi handles this device
Unfortunately the above changes didn't bring any improvement. The report rate stays at
~7Hz.
The other thing that might explain the difference, is at the I2C
level. I see on the dmesg that the touchpad is attached to
AMDI0010:03. I wonder if there is not a misconfiguration on the AMD
I2C adapter driver that prevents the touchpad from using the correct
report rate.
Unfortunately, I don't know much about this driver, but maybe given
all you showed to me you could have a look too? :)
Cheers,
Benjamin
[0] https://docs.microsoft.com/en-us/windows-hardware/design/component-guidelines/windows-precision-touchpad-required-hid-top-level-collections#latency-mode-feature-report
Thank you for pointing me to another direction!
AMDI0010 is handled by i2c-designware (drivers/i2c/busses/i2c-designware-platdrv.c),
$ sudo i2cdetect -l
i2c-0 i2c Synopsys DesignWare I2C adapter I2C adapter
The crucial set parameters when configuring the I2C adapter seems to
be the I2C related timing parameters. Firmware doesn't provide the
values (i2c_parse_fw_timings doesn't find any device properties). The
I2C adapter uses the following set of values,
clock-frequency: 400000
i2c-scl-rising-time-ns: 0
i2c-scl-falling-time-ns: 0
i2c-scl-internal-delay-ns: 0
i2c-sda-falling-time-ns: 0
i2c-sda-hold-time-ns: 0
i2c-digital-filter-width-ns: 0
i2c-analog-filter-cutoff-frequency: 0
For now I don't have sufficent knowledge to tell what may be wrong.
So I studied several cases of I2C adapter configuration causing touchpad
issues,
- i2c: designware: Do not use parameters from ACPI on Dell Inspiron 7348 - Patchwork [2]
- [BUG] i2c-hid: ELAN Touchpad does not work on ASUS X580GD - Patchwork [3]
- LKML: Chris Chiu: Tweak I2C SDA hold time on GemniLake to make touchpad work
But I couldn't find a similar case to touchpad having low report rate
due to a misconfiguration of I2C adapter.
Some unsuccessful attempts are
- Set acpi_osi to Windows like "Windows 2015"
- Set I2C bus speed to the standard mode (clock-frequency: 100000)
doesn't change anything and set it to high speed mode make cause
touchpad to be not recognized.
I'll try another expedient way of finding out what are the values of
I2C related timing parameters under Windows and setting the same values
under Linux. If this fails, it seems I need to try the hard way to
understand how I2C bus work by reading materials like NXP
Semiconductors's I2C manual.
I'm not sure if it's worth looking at yet another direction, i.e., to check
if there's something wrong with the touchpad device's interrupt setting
up. Here's the ACPI dump of the I2C controller and the touchpad device,
Device (I2CD)
{
Name (_HID, "AMDI0010") // _HID: Hardware ID
Name (_UID, 0x03) // _UID: Unique ID
Name (_CRS, ResourceTemplate () // _CRS: Current Resource Settings
{
IRQ (Edge, ActiveHigh, Exclusive, )
{6}
Memory32Fixed (ReadWrite,
0xFEDC5000, // Address Base
0x00001000, // Address Length
)
})
}
Scope (_SB.I2CD)
{
Device (TPDD)
{
Name (_HID, "XXXX0000") // _HID: Hardware ID
Name (_CID, "PNP0C50" /* HID Protocol Device (I2C bus) */) // _CID: Compatible ID
Name (_SUB, "XXXX0000") // _SUB: Subsystem ID
Method (_INI, 0, NotSerialized) // _INI: Initialize
{
If ((TPTY == One))
{
_HID = "MSFT0001"
_SUB = "ELAN0001"
}
If ((TPTY == 0x02))
{
_HID = "MSFT0001"
_SUB = "SYNA0001"
}
Return (Zero)
}
Method (_CRS, 0, NotSerialized) // _CRS: Current Resource Settings
{
If ((TPTY == One))
{
Name (SBFB, ResourceTemplate ()
{
I2cSerialBusV2 (0x0015, ControllerInitiated, 0x00061A80,
AddressingMode7Bit, "\\_SB.I2CD",
0x00, ResourceConsumer, , Exclusive,
)
})
}
If ((TPTY == 0x02))
{
Name (SBFS, ResourceTemplate ()
{
I2cSerialBusV2 (0x002C, ControllerInitiated, 0x00061A80,
AddressingMode7Bit, "\\_SB.I2CD",
0x00, ResourceConsumer, , Exclusive,
)
})
}
Name (SBFI, ResourceTemplate ()
{
GpioInt (Level, ActiveLow, ExclusiveAndWake, PullUp, 0x0000,
"\\_SB.GPIO", 0x00, ResourceConsumer, ,
)
{ // Pin list
0x0082
}
})
If ((TPTY == One))
{
Return (ConcatenateResTemplate (SBFB, SBFI))
}
If ((TPTY == 0x02))
{
Return (ConcatenateResTemplate (SBFS, SBFI))
}
}
}
}
Best regards,
Coiby
[1] https://docs.microsoft.com/en-us/previous-versions/windows/hardware/design/dn642101(v=vs.85)?redirectedfrom=MSDN
[2] https://lore.kernel.org/patchwork/patch/601043/
[3] https://lore.kernel.org/patchwork/patch/903152/
[4] https://lkml.org/lkml/2019/9/3/191
--
Best regards,
Coiby
