Re: [PATCH] Input: elants_i2c - Do not bind to i2c-hid compatible ACPI instantiated devices
From: Benjamin Tissoires <hidden>
Date: 2021-03-03 05:38:02
Hi Hans, On Tue, Mar 2, 2021 at 3:51 PM Hans de Goede [off-list ref] wrote:
Several users have been reporting that elants_i2c gives several errors during probe and that their touchscreen does not work on their Lenovo AMD based laptops with a touchscreen with a ELAN0001 ACPI hardware-id: [ 0.550596] elants_i2c i2c-ELAN0001:00: i2c-ELAN0001:00 supply vcc33 not found, using dummy regulator [ 0.551836] elants_i2c i2c-ELAN0001:00: i2c-ELAN0001:00 supply vccio not found, using dummy regulator [ 0.560932] elants_i2c i2c-ELAN0001:00: elants_i2c_send failed (77 77 77 77): -121 [ 0.562427] elants_i2c i2c-ELAN0001:00: software reset failed: -121 [ 0.595925] elants_i2c i2c-ELAN0001:00: elants_i2c_send failed (77 77 77 77): -121 [ 0.597974] elants_i2c i2c-ELAN0001:00: software reset failed: -121 [ 0.621893] elants_i2c i2c-ELAN0001:00: elants_i2c_send failed (77 77 77 77): -121 [ 0.622504] elants_i2c i2c-ELAN0001:00: software reset failed: -121 [ 0.632650] elants_i2c i2c-ELAN0001:00: elants_i2c_send failed (4d 61 69 6e): -121 [ 0.634256] elants_i2c i2c-ELAN0001:00: boot failed: -121 [ 0.699212] elants_i2c i2c-ELAN0001:00: invalid 'hello' packet: 00 00 ff ff [ 1.630506] elants_i2c i2c-ELAN0001:00: Failed to read fw id: -121 [ 1.645508] elants_i2c i2c-ELAN0001:00: unknown packet 00 00 ff ff Despite these errors, the elants_i2c driver stays bound to the device (it returns 0 from its probe method despite the errors), blocking the i2c-hid driver from binding. Manually unbinding the elants_i2c driver and binding the i2c-hid driver makes the touchscreen work. Check if the ACPI-fwnode for the touchscreen contains one of the i2c-hid compatiblity-id strings and if it has the I2C-HID spec's DSM to get the HID descriptor address, If it has both then make elants_i2c not bind, so that the i2c-hid driver can bind. This assumes that non of the (older) elan touchscreens which actually need the elants_i2c driver falsely advertise an i2c-hid compatiblity-id + DSM in their ACPI-fwnodes. If some of them actually do have this false advertising, then this change may lead to regressions.
I like the v2 much better, and I believe the regressions should be fairly small now. One nitpick below:
quoted hunk ↗ jump to hunk
Cc: Benjamin Tissoires <redacted> BugLink: https://bugzilla.kernel.org/show_bug.cgi?id=207759 Signed-off-by: Hans de Goede <redacted> --- drivers/input/touchscreen/elants_i2c.c | 25 +++++++++++++++++++++++++ 1 file changed, 25 insertions(+)diff --git a/drivers/input/touchscreen/elants_i2c.c b/drivers/input/touchscreen/elants_i2c.c index 4c2b579f6c8b..510638e5ba5a 100644 --- a/drivers/input/touchscreen/elants_i2c.c +++ b/drivers/input/touchscreen/elants_i2c.c@@ -1334,6 +1334,12 @@ static void elants_i2c_power_off(void *_data) } } +static const struct acpi_device_id i2c_hid_ids[] = { + {"ACPI0C50", 0 }, + {"PNP0C50", 0 }, + { }, +}; + static int elants_i2c_probe(struct i2c_client *client, const struct i2c_device_id *id) {@@ -1342,6 +1348,25 @@ static int elants_i2c_probe(struct i2c_client *client, unsigned long irqflags; int error; +#ifdef CONFIG_ACPI + /* Don't bind to i2c-hid compatible devices, these are handled by the i2c-hid drv. */ + if (acpi_match_device_ids(ACPI_COMPANION(&client->dev), i2c_hid_ids) == 0) { + static const guid_t i2c_hid_guid = + GUID_INIT(0x3CDFF6F7, 0x4267, 0x4555, + 0xAD, 0x05, 0xB3, 0x0A, 0x3D, 0x89, 0x38, 0xDE);
Andy in https://patchwork.kernel.org/project/linux-input/patch/20210226193225.47129-3-andriy.shevchenko@linux.intel.com/ proposed a few cleanups for i2c-hid. I believe a similar pacth should be squashed here (the include is the one I am more worried about). Besides that: Acked-by: Benjamin Tissoires <redacted> Cheers, Benjamin
+ acpi_handle handle = ACPI_HANDLE(&client->dev);
+ union acpi_object *obj;
+
+ obj = acpi_evaluate_dsm_typed(handle, &i2c_hid_guid, 1, 1, NULL,
+ ACPI_TYPE_INTEGER);
+ if (obj) {
+ dev_warn(&client->dev, "elants_i2c: This device appears to be an I2C-HID device, not binding\n");
+ ACPI_FREE(obj);
+ return -ENODEV;
+ }
+ }
+#endif
+
if (!i2c_check_functionality(client->adapter, I2C_FUNC_I2C)) {
dev_err(&client->dev,
"%s: i2c check functionality error\n", DEVICE_NAME);
--
2.30.1