[BUG] keyboard doesn't work on Olimex teres-I notebook

[BUG] keyboard doesn't work on Olimex teres-I notebook

[Milan P. Stanić]

[ Please CC to me, I'm not subscribed to mailing list ]

Hi,

On u-boot release 2022.04 keyboard doesn't work in u-boot on Olimex
TERES-I notebook.

With git bisect I found that commit 35ae126c16a6a9149edc6638faaa247f67b8a400
is introduced this. Reverting this commit solved problem and in my test
keyboard now works.

-- 
Kind regards

Re: [BUG] keyboard doesn't work on Olimex teres-I notebook

[Andre Przywara]

On Tue, 19 Apr 2022 14:27:24 +0200
Milan P. Stanić <mps@arvanta.net> wrote:

Hi Milan,

> [ Please CC to me, I'm not subscribed to mailing list ]
> 
> Hi,
> 
> On u-boot release 2022.04 keyboard doesn't work in u-boot on Olimex
> TERES-I notebook.

Thanks for the report, and sorry for the delay!

> With git bisect I found that commit 35ae126c16a6a9149edc6638faaa247f67b8a400
> is introduced this. Reverting this commit solved problem and in my test
> keyboard now works.

Unfortunately just reverting this commit is not really the way forward,
but I had a closer look and few printf's later found the problem:
The USB PHY driver uses the legacy GPIO interface, which is now wrapped
in gpio-uclass.c, to call the actual DM functions.
The commit that you bisected to implements the recommended
.set_flags ops member, which relies on the GPIO being properly
configured before. Unfortunately those compat wrappers seem to be
broken, and not using .set_flags just papered over this problem before:
- A call to gpio_direction_output() fills the DM struct gpio_desc
  *locally*, on the stack, so any setup is lost when the function
  returns. The GPIO hardware is still configured as an output, but
  U-Boot doesn't remember this.
- A call to gpio_direction_output() in the legacy interface is supposed
  to also set the output value, but the compat wrapper does not
  implement this.
- When .set_flags is implemented in a GPIO driver implementation, the
  DM GPIO interface *requires* GPIOs to be configured before, but
  since this information is lot, a call to gpio_set_value() does
  nothing.

So while those issues looks like they demand fixing, in the generic
compat GPIO wrappers, I decided to upgrade the sunxi USB PHY driver to
just use the DM GPIO interface directly. Eventually this driver should
drop the U-Boot configured GPIOs, and use proper DT methods, but this
is a bigger change I'd rather not introduce that late in the
development cycle.
So for now I sent this patch, that should fix the issue:
https://lore.kernel.org/u-boot/20220608000604.3357-1-andre.przywara@arm.com/

If you could please test this patch, then reply to that patch email,
for instance with a Tested-by:, this would be greatly appreciated!

Cheers,
Andre

Re: [BUG] keyboard doesn't work on Olimex teres-I notebook

[Andre Przywara]

On Wed, 8 Jun 2022 01:30:42 +0100
Andre Przywara <andre.przywara@arm.com> wrote:

Hi,

> On Tue, 19 Apr 2022 14:27:24 +0200
> Milan P. Stanić <mps@arvanta.net> wrote:
> 
> Hi Milan,
> 
> > [ Please CC to me, I'm not subscribed to mailing list ]
> > 
> > Hi,
> > 
> > On u-boot release 2022.04 keyboard doesn't work in u-boot on Olimex
> > TERES-I notebook.  
> 
> Thanks for the report, and sorry for the delay!
> 
> > With git bisect I found that commit 35ae126c16a6a9149edc6638faaa247f67b8a400
> > is introduced this. Reverting this commit solved problem and in my test
> > keyboard now works.  
> 
> Unfortunately just reverting this commit is not really the way forward,
> but I had a closer look and few printf's later found the problem:
> The USB PHY driver uses the legacy GPIO interface, which is now wrapped
> in gpio-uclass.c, to call the actual DM functions.
> The commit that you bisected to implements the recommended
> .set_flags ops member, which relies on the GPIO being properly
> configured before. Unfortunately those compat wrappers seem to be
> broken, and not using .set_flags just papered over this problem before:
> - A call to gpio_direction_output() fills the DM struct gpio_desc
>   *locally*, on the stack, so any setup is lost when the function
>   returns. The GPIO hardware is still configured as an output, but
>   U-Boot doesn't remember this.
> - A call to gpio_direction_output() in the legacy interface is supposed
>   to also set the output value, but the compat wrapper does not
>   implement this.

Scratch that part: that works, as the h/w routines are called inside
dm_gpio_clrset_flags(), where we set the correct flags in our (temporary)
struct gpio_desc, before calling the .set_flags implementation. Got
tricked by the rather innocent function name.

But still calling gpio_direction_output(), then later gpio_set_value() is
broken. I guess we should fix it, maybe by forcing GPIOD_IS_OUT in the
desc, before calling dm_gpio_set_value()?

Cheers,
Andre

> - When .set_flags is implemented in a GPIO driver implementation, the
>   DM GPIO interface *requires* GPIOs to be configured before, but
>   since this information is lot, a call to gpio_set_value() does
>   nothing.
> 
> So while those issues looks like they demand fixing, in the generic
> compat GPIO wrappers, I decided to upgrade the sunxi USB PHY driver to
> just use the DM GPIO interface directly. Eventually this driver should
> drop the U-Boot configured GPIOs, and use proper DT methods, but this
> is a bigger change I'd rather not introduce that late in the
> development cycle.
> So for now I sent this patch, that should fix the issue:
> https://lore.kernel.org/u-boot/20220608000604.3357-1-andre.przywara@arm.com/
> 
> If you could please test this patch, then reply to that patch email,
> for instance with a Tested-by:, this would be greatly appreciated!
> 
> Cheers,
> Andre