Re: [PATCH] Input: atkbd - Skip ATKBD_CMD_GETID in translated mode

[Hans de Goede <hdegoede@redhat.com>]

Hi Dmitry,

Thank you for the review.

On 11/14/23 16:59, Dmitry Torokhov wrote:
> Hi Hans,
> 
> On Mon, Nov 06, 2023 at 04:54:29PM +0100, Hans de Goede wrote:
>> There have been multiple reports of keyboard issues on recent laptop models
>> which can be worked around by setting i8042.dumbkbd, with the downside
>> being this breaks the capslock LED.
>>
>> It seems that these issues are caused by recent laptops getting confused by
>> ATKBD_CMD_GETID. Rather then adding and endless growing list of quirks for
>> this, lets just skip ATKBD_CMD_GETID alltogether when in translated mode.
>>
>> The main goal of sending ATKBD_CMD_GETID is to skip binding to ps/2
>> mice/touchpads and those are never used in translated mode.
>>
>> Examples of laptop models which benefit from skipping ATKBD_CMD_GETID:
>>
>> * "HP Laptop 15s-fq2xxx", "HP laptop 15s-fq4xxx" and "HP Laptop 15-dy2xxx"
>>   models the kbd stops working for the first 2 - 5 minutes after boot
>>   (waiting for EC watchdog reset?)
>>
>> * On "HP Spectre x360 13-aw2xxx" atkbd fails to probe the keyboard
>>
>> * At least 9 different Lenovo models have issues with ATKBD_CMD_GETID, see:
>>   https://github.com/yaescallop/atkbd-nogetid
>>
>> Note this also removes the "NCD terminal keyboards are only supported on
>> non-translating controllers." warning since that code is now unreachable.
>>
>> This has been tested on:
>>
>> 1. A MSI B550M PRO-VDH WIFI desktop, where the i8042 controller is not
>>    in translated mode when no keyboard is plugged in and with a ps/2 kbd
>>    a "AT Translated Set 2 keyboard" /dev/input/event# node shows up
>>
>> 2. A Dell Latitude 9420 (always has a "AT Translated Set 2 keyboard")
>>
>> 3. A Lenovo ThinkPad X1 Yoga gen 8 (idem)
> 
> I agree that the mice/touchpads are not going to work if the controller
> is in translated mode, however I wonder if on a device with external
> PS/2 ports we could not end up with a port in translated mode with
> "wrong" device plugged in.
> 
> Can we consider not skipping the check completely, but rather use DMI to
> check the chassis type (we already have a similar check in 8042)
> and skip ATKBD_CMD_GETID on mobile devices, but still try
> ATKBD_CMD_SETLEDS on them?

I think that should work. At least the patches from:

https://github.com/yescallop/atkbd-nogetid

for the affected Lenovo models to exactly that and from my (remote)
debugging of the issue on one of the HP laptops I expect it to work
fine there too.

I've prepared a v2 implementing this and I'll send v2 out
right after this email.

Regards,

Hans