[<prev] [next>] [<thread-prev] [thread-next>] [day] [month] [year] [list]
Message-ID: <dd9d62d5-54fc-4e7e-8508-1b8e22ac28d5@molgen.mpg.de>
Date: Fri, 14 Mar 2025 12:54:22 +0100
From: Paul Menzel <pmenzel@...gen.mpg.de>
To: Andy Shevchenko <andriy.shevchenko@...ux.intel.com>,
Bartosz Golaszewski <bartosz.golaszewski@...aro.org>
Cc: Linus Walleij <linus.walleij@...aro.org>, brgl@...ev.pl,
linux-gpio@...r.kernel.org, LKML <linux-kernel@...r.kernel.org>,
linux-pci@...r.kernel.org, regressions@...ts.linux.dev
Subject: Re: Linux logs new warning `gpio gpiochip0:
gpiochip_add_data_with_key: get_direction failed: -22`
Dear Andy, dear Bartosz,
Am 26.02.25 um 15:22 schrieb Andy Shevchenko:
> On Wed, Feb 26, 2025 at 03:14:24PM +0100, Bartosz Golaszewski wrote:
>> On Wed, 26 Feb 2025 at 14:47, Andy Shevchenko wrote:
>>> On Wed, Feb 26, 2025 at 03:37:47PM +0200, Andy Shevchenko wrote:
>>>> On Tue, Feb 25, 2025 at 10:25:00PM +0100, Linus Walleij wrote:
>>>>> On Mon, Feb 24, 2025 at 9:51 AM <brgl@...ev.pl> wrote:
>>>>>
>>>>>> In any case: Linus: what should be our policy here? There are some pinctrl
>>>>>> drivers which return EINVAL if the pin in question is not in GPIO mode. I don't
>>>>>> think this is an error. Returning errors should be reserved for read failures
>>>>>> and so on. Are you fine with changing the logic here to explicitly default to
>>>>>> INPUT as until recently all errors would be interpreted as such anyway?
>>>>>
>>>>> Oh hm I guess. There was no defined semantic until now anyway. Maybe
>>>>> Andy has something to say about it though, it's very much his pin controller.
>>>>
>>>> Driver is doing correct things. If you want to be pedantic, we need to return
>>>> all possible pin states (which are currently absent from GPIO get_direction()
>>>> perspective) and even though it's not possible to tell from the pin muxer
>>>> p.o.v. If function is I2C, it's open-drain, if some other, it may be completely
>>>> different, but pin muxer might only guesstimate the state of the particular
>>>> function is and I do not think guesstimation is a right approach.
>>>>
>>>> We may use the specific error code, though. and document that semantics.
>>>
>>> Brief looking at the error descriptions and the practical use the best (and
>>> unique enough) choice may be EBADSLT.
>>
>> In any case, I proposed to revert to the previous behavior in
>> gpiochip_add_data() in my follow-up series so the issue should soon go
>> away.
>
> Yes, I noted. The above is a material to discuss. We can make that semantics
> documented and strict and then one may filter out those errors if/when
> required.
I am still seeing this with 6.14.0-rc6-00022-gb7f94fcf5546. Do you know,
if the reverts are going to be in the final 6.14 release?
Kind regards,
Paul
Powered by blists - more mailing lists