Re: [PATCH 1/1] pinctrl: stmfx: fix valid_mask init sequence

From: Linus Walleij <linus.walleij@linaro.org>
To: Amelie DELAUNAY <amelie.delaunay@st.com>
Cc: Alexandre TORGUE <alexandre.torgue@st.com>,
	Maxime Coquelin <mcoquelin.stm32@gmail.com>,
	"open list:GPIO SUBSYSTEM" <linux-gpio@vger.kernel.org>,
	"linux-kernel@vger.kernel.org" <linux-kernel@vger.kernel.org>,
	Linux ARM <linux-arm-kernel@lists.infradead.org>,
	"linux-stm32@st-md-mailman.stormreply.com" 
	<linux-stm32@st-md-mailman.stormreply.com>
Subject: Re: [PATCH 1/1] pinctrl: stmfx: fix valid_mask init sequence
Date: Thu, 7 Nov 2019 10:09:38 +0100	[thread overview]
Message-ID: <CACRpkdbK=aB84hW0FtuMBtzqi3ftKBxedXEpqmnJhbN_vJWx2Q@mail.gmail.com> (raw)
In-Reply-To: <64f8096f-cec6-fef1-5a4e-ddca3bf8c73d@st.com>

On Tue, Nov 5, 2019 at 4:14 PM Amelie DELAUNAY <amelie.delaunay@st.com> wrote:
> On 11/5/19 3:32 PM, Linus Walleij wrote:
> > On Mon, Nov 4, 2019 at 11:09 AM Amelie Delaunay <amelie.delaunay@st.com>
> > wrote:
> >
> >> With stmfx_pinctrl_gpio_init_valid_mask callback, gpio_valid_mask was used
> >> to initialize gpiochip valid_mask for gpiolib. But gpio_valid_mask was not
> >> yet initialized. gpio_valid_mask required gpio-ranges to be registered,
> >> this is the case after gpiochip_add_data call. But init_valid_mask
> >> callback is also called under gpiochip_add_data. gpio_valid_mask
> >> initialization cannot be moved before gpiochip_add_data because
> >> gpio-ranges are not registered.
> >
> > Sorry but this doesn't add up, look at this call graph:
> >
> > gpiochip_add_data()
> >    gpiochip_add_data_with_key()
> >      gpiochip_alloc_valid_mask()
> >      of_gpiochip_add()
> >      of_gpiochip_add_pin_range()
> >      gpiochip_init_valid_mask()
> >
> > So the .initi_valid_mask() is clearly called *after*
> > of_gpiochip_add_pin_range() so this cannot be the real reason,
> > provided that the ranges come from the device tree. AFAICT that
> > is the case with the stmfx.
> >
> > Can you check and see if the problem is something else?
> >
>
> stmfx_pinctrl_gpio_init_valid_mask uses pctl->gpio_valid_mask to
> initialize gpiochip valid_mask.
>
> pctl->gpio_valid_mask is initialized in
> stmfx_pinctrl_gpio_function_enable depending on gpio ranges.
>
> stmfx_pinctrl_gpio_function_enable is called after gpiochip_add_data
> because it requires gpio ranges to be registered.
>
> So, in stmfx driver the call graph is
>
> stmfx_pinctrl_probe
>    gpiochip_add_data()
>      gpiochip_add_data_with_key()
>        gpiochip_alloc_valid_mask()
>        of_gpiochip_add()
>        of_gpiochip_add_pin_range()
>        gpiochip_init_valid_mask()
>          stmfx_pinctrl_gpio_init_valid_mask (but pctl->gpio_valid_mask
> is not yet initialized so gpiochip valid_mask is wrong)
>    stmfx_pinctrl_gpio_function_enable (pctl->gpio_valid_mask is going to
> be initialized thanks to gpio ranges)
>
> When consumer tries to take a pin (it is the case for the joystick on
> stm32mp157c-ev1), it gets the following issue:
> [    3.347391] irq: :soc:i2c@40013000:stmfx@42:stmfx-pin-controller
> didn't like hwirq-0x0 to VIRQ92 mapping (rc=-6)
> [    3.356418] irq: :soc:i2c@40013000:stmfx@42:stmfx-pin-controller
> didn't like hwirq-0x1 to VIRQ92 mapping (rc=-6)
> [    3.366512] irq: :soc:i2c@40013000:stmfx@42:stmfx-pin-controller
> didn't like hwirq-0x2 to VIRQ92 mapping (rc=-6)
> [    3.376671] irq: :soc:i2c@40013000:stmfx@42:stmfx-pin-controller
> didn't like hwirq-0x3 to VIRQ92 mapping (rc=-6)
> [    3.387169] irq: :soc:i2c@40013000:stmfx@42:stmfx-pin-controller
> didn't like hwirq-0x4 to VIRQ92 mapping (rc=-6)
> [    3.397065] gpio-keys joystick: Found button without gpio or irq
> [    3.403041] gpio-keys: probe of joystick failed with error -22
>
> I can reword the commit message to make it clearer.

No need I understand it now, thanks for explaining!

We need to populate the valid mask some other way if you
want to safeguard this, I don't know if the existing
gpio-reserved-ranges would work? But it feels a bit unsafe
if you actually determine this some other way.

Yours,
Linus Walleij