Re: i2c designware change broke touchpad of a thinkpad

[johannes.goede@oss.qualcomm.com]

[-- Attachment #1: Type: text/plain, Size: 2206 bytes --]

Hi Hardik,

On 14-Aug-26 12:09 PM, Hardik Prakash wrote:
> On Wed, 13 Aug 2026, Thorsten Leemhuis wrote:
>> Hardik, could you submit a revert to the lists? And then I'll provide my
>> Tested-by and ask Linus to directly pick this up?
> 
> Sent: https://lore.kernel.org/all/20260814100719.9548-1-hardikprakash.official@gmail.com/
> 
> Verified on my end that it builds clean, boots clean, and correctly
> restores the original Yoga 7 race (expected tradeoff) without
> introducing anything else.

Thank you for submitting a revert for this. Note it looks like Torvalds himself
beat you to it and already reverted this for 7.2 final :)

I hit another problem caused by this now reverted change, where the PMIC
i2c bus would not show up on many Intel BYT/CHT devices. I wrote a fix for
this (attached) on top of the now reverted commit.

I think my fix might also have helped for the broken touchpad issue from this
thread, but I believe there is a better way to fix all this, so IMHO it is
good that this was reverted.

If I understand things correctly the problem the reverted fix was trying
to fix is i2c-transfer errors happening before the GPIO controller driver
is ready.

The i2c-core will not initiate transfers itself, so the problem is that
the i2c-hid driver is initiating transfers before the GPIO controller
is setup.

The i2c-hid driver does actually request the interrupt itself, so we can
simply wait for that to succeed inside the i2c-hid code. ATM the i2c-hid
code starts with an i2c-connectivity check because some ACPI tables list
non existing I2C-HID devices.

But we could make the i2c-hid driver start with first requesting its IRQ
and if that fails bail (with -EPROBE_DEFER) before doing any i2c-transfers.

And then do the i2c-connectivity test after requesting the IRQ. So basically
swap the order of i2c-connectivity test vs IRQ requesting.

The IRQ requesting should then use the IRQF_NO_AUTOEN flag to keep the IRQ
disabled at first and explicitly enable it later when the rest of the driver
setup is done.

Hardik, can you take a shot at coding up the suggested i2c-hid(-core) changes
and test to see if this fixes the original Yoga 7 race in a cleaner manner ?

Regards,

Hans


[-- Attachment #2: 0001-i2c-designware-Make-check_child_gpioint-skip-devices.patch --]
[-- Type: text/x-patch, Size: 2339 bytes --]

From e12a71c64b996560d521ea9dda31a8c629b504e4 Mon Sep 17 00:00:00 2001
From: Hans de Goede <johannes.goede@oss.qualcomm.com>
Date: Sun, 16 Aug 2026 23:06:07 +0200
Subject: [PATCH] i2c: designware: Make check_child_gpioint() skip devices
 which are not present

Commit 0a4bb2abc3e5 ("i2c: designware: defer probe if child GpioInt
controllers are not bound") makes i2c-designware-platdrv delay binding
until all GpioInt resources of children of the i2c-controller are
available.

This causes the driver to sometimes never bound in case of bogus, or
not supported by Linux GpioInt resources on some of the i2c-clients of
the controller, which causes *all* of the clients to not work!

ACPI tables typically contain descriptions of many possible i2c-clients
under an i2c-controller (e.g. second sources of devices) but only
a few are actually marked as being present. The kernel will only
instantiate i2c-clients for those which are present, make the GpioInt
"ready" check only check actually present i2c-clients.

Intel BYT/CHT devices have a special INT33FE i2c-client which is
a duplicate of the exisiting PMIC i2c-client, with non supported
GpioInt resources (pointing back to the PMIC). Linux already ignores
these, so make the GpioInt check skip these too.

This fixes i2c-designware-platdrv not binding to the PMIC I2C bus on many
Intel Bay Trail and Cherry Trail devices like e.g. the MPman Convertor9.

Fixes: 0a4bb2abc3e5 ("i2c: designware: defer probe if child GpioInt controllers are not bound")
Cc: stable@vger.kernel.org
Signed-off-by: Hans de Goede <johannes.goede@oss.qualcomm.com>
---
 drivers/i2c/busses/i2c-designware-platdrv.c | 7 +++++++
 1 file changed, 7 insertions(+)

diff --git a/drivers/i2c/busses/i2c-designware-platdrv.c b/drivers/i2c/busses/i2c-designware-platdrv.c
index c8a203fff4d1..f80817e6221d 100644
--- a/drivers/i2c/busses/i2c-designware-platdrv.c
+++ b/drivers/i2c/busses/i2c-designware-platdrv.c
@@ -180,6 +180,13 @@ static int check_child_gpioint(struct acpi_device *adev, void *data)
 	LIST_HEAD(res_list);
 	int ret;
 
+	if (!adev->status.present)
+		return 0;
+
+	/* Skip broken Intel BYT/CHT INT33FE battery devices */
+	if (acpi_dev_hid_match(adev, "INT33FE"))
+		return 0;
+
 	ret = acpi_dev_get_resources(adev, &res_list, check_gpioint_resource, NULL);
 	if (ret < 0)
 		return ret;
-- 
2.55.0