Re: Tweak I2C SDA hold time on GemniLake to make touchpad work

From: Jarkko Nikula
Date: Tue Sep 03 2019 - 08:03:54 EST


This is a multi-part message in MIME format. Hi Chris

On 9/3/19 11:18 AM, Mika Westerberg wrote:
+Jarkko

On Tue, Sep 03, 2019 at 04:10:27PM +0800, Chris Chiu wrote:
Hi,

We're working on the acer Gemnilake laptop TravelMate B118-M for
touchpad not working issue. The touchpad fails to bring up and the
i2c-hid ouput the message as follows
[ 8.317293] i2c_hid i2c-ELAN0502:00: hid_descr_cmd failed
We tried on latest linux kernel 5.3.0-rc6 and it reports the same.

We then look into I2C signal level measurement to find out why.
The following is the signal output from LA for the SCL/SDA.
https://imgur.com/sKcpvdo
The SCL frequency is ~400kHz from the SCL period, but the SDA
transition is quite weird. Per the I2C spec, the data on the SDA line
must be stable during the high period of the clock. The HIGH or LOW
state of the data line can only change when the clock signal on the
SCL line is LOW. The SDA period span across 2 SCL high, I think
that's the reason why the I2C read the wrong data and fail to initialize.

Thus, we treak the SDA hold time by the following modification.

--- a/drivers/mfd/intel-lpss-pci.c
+++ b/drivers/mfd/intel-lpss-pci.c
@@ -97,7 +97,8 @@ static const struct intel_lpss_platform_info bxt_uart_info = {
};

static struct property_entry bxt_i2c_properties[] = {
- PROPERTY_ENTRY_U32("i2c-sda-hold-time-ns", 42),
+ PROPERTY_ENTRY_U32("i2c-sda-hold-time-ns", 230),
PROPERTY_ENTRY_U32("i2c-sda-falling-time-ns", 171),
PROPERTY_ENTRY_U32("i2c-scl-falling-time-ns", 208),
{ },

The reason why I choose sda hold time is by the Table 10 of
https://www.nxp.com/docs/en/user-guide/UM10204.pdf, the device
must provide a hold time at lease 300ns and and 42 here is relatively
too small. The signal measurement result for the same pin on Windows
is as follows.
https://imgur.com/BtKUIZB
Comparing to the same result running Linux
https://imgur.com/N4fPTYN

After applying the sda hold time tweak patch above, the touchpad can
be correctly initialized and work. The LA signal is shown as down below.
https://imgur.com/B3PmnIp

Could you try does attached patch work for you?

It's from last year for another related issue but there platform was actually Apollo Lake instead of Gemini Lake but anyway it was found out that Windows uses different timing parameters than Linux on Gemini Lake.

I didn't take patch forward back then due known Gemini Lake machines were working with the Broxton I2C timing parameters but now it's time if attached patch fixes the issue on your machine.

Patch is from top of v5.3-rc7 but should probably apply also to older kernels.

--
Jarkko