[PATCH v4 0/6] USB keyboard improvements for asahi / desktop systems

[Marek Vasut]

On 4/8/24 9:46 AM, Janne Grunau wrote:
> On Sun, Apr 07, 2024 at 03:05:59AM +0200, Marek Vasut wrote:
>> On 4/6/24 10:04 PM, Janne Grunau wrote:
>>> On Sat, Apr 06, 2024 at 08:52:17PM +0200, Marek Vasut wrote:
>>>> On 4/5/24 9:05 PM, Janne Grunau wrote:
>>>>> On Fri, Apr 05, 2024 at 04:52:32PM +0200, Marek Vasut wrote:
>>>>>> On 4/4/24 8:25 AM, Janne Grunau via B4 Relay wrote:
>>>>>>> Apple USB Keyboards from 2021 need quirks to be useable. The boot HID
>>>>>>> keyboard protocol is unfortunately not described in the first interface
>>>>>>> descriptor but the second. This needs several changes. The USB keyboard
>>>>>>> driver has to look at all (2) interface descriptors during probing.
>>>>>>> Since I didn't want to rebuild the USB driver probe code the Apple
>>>>>>> keyboards are bound to the keyboard driver via USB vendor and product
>>>>>>> IDs.
>>>>>>> To make the keyboards useable on Apple silicon devices the xhci driver
>>>>>>> needs to initializes rings for the endpoints of the first two interface
>>>>>>> descriptors. If this is causes concerns regarding regressions or memory
>>>>>>> use the USB_MAX_ACTIVE_INTERFACES define could be turned into a CONFIG
>>>>>>> option.
>>>>>>> Even after this changes the keyboards still do not probe successfully
>>>>>>> since they apparently do not behave HID standard compliant. They only
>>>>>>> generate reports on key events. This leads the final check whether the
>>>>>>> keyboard is operational to fail unless the user presses keys during the
>>>>>>> probe. Skip this check for known keyboards.
>>>>>>> Keychron seems to emulate Apple keyboards (some models even "re-use"
>>>>>>> Apple's USB vendor ID) so apply this quirk as well.
>>>>>>>
>>>>>>> Some devices like Yubikeys emulate a keyboard. since u-boot only binds a
>>>>>>> single keyboard block this kind of devices from the USB keyboard driver.
>>>>>>>
>>>>>>> Signed-off-by: Janne Grunau <j at jannau.net>
>>>>>>
>>>>>> I picked the series, but CI indicates build errors, can you have a look ?
>>>>>>
>>>>>> https://source.denx.de/u-boot/custodians/u-boot-usb/-/pipelines/20215
>>>>>
>>>>> The issue seems to be that the field dev in struct usb_device exists
>>>>> only for DM_USB. That means we can't use dev_dbg.
>>>>> Either take the following fixup patch or I can resend the series.
>>>>
>>>> I squashed the extra patch in, but I think the CI still complains:
>>>>
>>>> Pipeline #20236 (
>>>> https://source.denx.de/u-boot/custodians/u-boot-usb/-/pipelines/20236 )
>>>> triggered by Marek Vasut ( https://source.denx.de/marex )
>>>> had 1 failed job.
>>>>
>>>> Job #812215 (
>>>> https://source.denx.de/u-boot/custodians/u-boot-usb/-/jobs/812215/raw )
>>>
>>> env_get() missing without CONFIG_ENV_SUPPORT. I'm too accustomed to the
>>> kernel's stub functions. Best option seems to to just #if the
>>> functionality out in this case. See attached fixup patch.
>>>
>>> Sorry,
>>
>> No worries.
>>
>> Does it work if you do this instead ?
>>
>>    static int usb_device_is_ignored(u16 id_vendor, u16 id_product)
>>    {
>>           ulong vid, pid;
>> +       /* No env support, nothing can be ignored */
>> +       if (CONFIG_IS_ENABLED(ENV_SUPPORT))
>> +           return 0;
>>
>> That way, the function is always compiled and if it is unreachable, then
>> compiler throws it out. This should improve code compile coverage.
> 
> Seems to work here with a broken imx8 config from the CI. Is it ok to
> rely on dead code elimination? Apparently it is, build with KCFLAGS=-O0
> has already several other missing symbols.
> 
> See attached fixup

Thanks, squashed, let's see how CI likes this.