[PATCH 2/2] pinctrl: pinctrl-imx: do not assume mux 0 is gpio

Stefan Agner stefan at agner.ch
Tue May 23 12:55:05 PDT 2017 

On 2017-05-23 03:23, A.S. Dong wrote:
> Hi Stefan,
> 
>> -----Original Message-----
>> From: A.S. Dong
>> Sent: Thursday, May 18, 2017 3:00 PM
>> To: 'Stefan Agner'
>> Cc: linux-gpio at vger.kernel.org; linux-arm-kernel at lists.infradead.org;
>> linus.walleij at linaro.org; shawnguo at kernel.org; Jacky Bai; Andy Duan;
>> kernel at pengutronix.de; Alexandre Courbot
>> Subject: RE: [PATCH 2/2] pinctrl: pinctrl-imx: do not assume mux 0 is gpio
>>
>> > -----Original Message-----
>> > From: Stefan Agner [mailto:stefan at agner.ch]
>> > Sent: Thursday, May 18, 2017 2:16 AM
>> > To: A.S. Dong
>> > Cc: linux-gpio at vger.kernel.org; linux-arm-kernel at lists.infradead.org;
>> > linus.walleij at linaro.org; shawnguo at kernel.org; Jacky Bai; Andy Duan;
>> > kernel at pengutronix.de; Alexandre Courbot
>> > Subject: Re: [PATCH 2/2] pinctrl: pinctrl-imx: do not assume mux 0 is
>> > gpio
>> >
>> > On 2017-05-17 00:18, A.S. Dong wrote:
>> > >> -----Original Message-----
>> > >> From: Stefan Agner [mailto:stefan at agner.ch]
>> > >> Sent: Tuesday, May 16, 2017 1:27 AM
>> > >> To: A.S. Dong
>> > >> Cc: linux-gpio at vger.kernel.org;
>> > >> linux-arm-kernel at lists.infradead.org;
>> > >> linus.walleij at linaro.org; shawnguo at kernel.org; Jacky Bai; Andy
>> > >> Duan; kernel at pengutronix.de; Alexandre Courbot
>> > >> Subject: Re: [PATCH 2/2] pinctrl: pinctrl-imx: do not assume mux 0
>> > >> is gpio
>> > >>
>> > >> On 2017-05-14 23:48, Dong Aisheng wrote:
>> > >> > Do not assume MUX 0 is GPIO function in core driver as a
>> > >> > different SoC may have different value. e.g. MX7ULP Mux 1 is GPIO.
>> > >> >
>> > >> > Cc: Linus Walleij <linus.walleij at linaro.org>
>> > >> > Cc: Alexandre Courbot <gnurou at gmail.com>
>> > >> > Cc: Shawn Guo <shawnguo at kernel.org>
>> > >> > Cc: Stefan Agner <stefan at agner.ch>
>> > >> > Cc: Fugang Duan <fugang.duan at nxp.com>
>> > >> > Cc: Bai Ping <ping.bai at nxp.com>
>> > >> > Signed-off-by: Dong Aisheng <aisheng.dong at nxp.com>
>> > >> > ---
>> > >> >  drivers/pinctrl/freescale/pinctrl-imx.c | 3 ++-
>> > >> >  1 file changed, 2 insertions(+), 1 deletion(-)
>> > >> >
>> > >> > diff --git a/drivers/pinctrl/freescale/pinctrl-imx.c
>> > >> > b/drivers/pinctrl/freescale/pinctrl-imx.c
>> > >> > index 0d6aaca..ed8ea32 100644
>> > >> > --- a/drivers/pinctrl/freescale/pinctrl-imx.c
>> > >> > +++ b/drivers/pinctrl/freescale/pinctrl-imx.c
>> > >> > @@ -281,7 +281,7 @@ static int imx_pmx_gpio_request_enable(struct
>> > >> > pinctrl_dev *pctldev,
>> > >> >  			continue;
>> > >> >  		for (pin = 0; pin < grp->num_pins; pin++) {
>> > >> >  			imx_pin = &((struct imx_pin *)(grp->data))[pin];
>> > >> > -			if (imx_pin->pin == offset && !imx_pin->mux_mode)
>> > >> > +			if (imx_pin->pin == offset)
>> > >> >  				goto mux_pin;
>> > >>
>> > >> The reason I added that check was to make sure we pick a mux option
>> > >> which is GPIO... With this change, any pinmux might be picked...
>> > >>
>> > >
>> > > First of all, this seems to be wrong to me that GPIO mux mode is SoC
>> > > Dependant and should not be put in pinctrl-imx core driver.
>> >
>> > Hm, agree, we should consider to move
>> > imx_pmx_gpio_request_enable/disable_free and
>> > imx_pmx_gpio_set_direction into pinctrl-vf610.c
>> >
>>
>> IMX7ULP may want to use imx_pmx_gpio_set_direction as well to support
>> dynamically change GPIO from output to input.
>>
>> > >
>> > > Secondly, I think we may be over worried and it's not quite
>> > > necessary As we did not do the sanity check for both raw config and
>> > > mux data read From Device tree, why only do it for GPIO?
>> > >
>> > > We may trust the data in device tree.
>> >
>> > In Vybrid, there is no need to explicitly assign a pinmux to use a pin
>> > as GPIO. So the pinmux could be anything... The implemented semantics
>> > for Vyrbid is really different than i.MX, see below.
>> >
>>
>> Strange, I do see Vybrid assigning pinmux to GPIO in device tree.
>> e.g.
>> arch/arm/boot/dts/vf-colibri.dtsi
>> pinctrl_esdhc1: esdhc1grp {
>>         fsl,pins = <
>>                 VF610_PAD_PTA24__ESDHC1_CLK     0x31ef
>>                 VF610_PAD_PTA25__ESDHC1_CMD     0x31ef
>>                 VF610_PAD_PTA26__ESDHC1_DAT0    0x31ef
>>                 VF610_PAD_PTA27__ESDHC1_DAT1    0x31ef
>>                 VF610_PAD_PTA28__ESDHC1_DATA2   0x31ef
>>                 VF610_PAD_PTA29__ESDHC1_DAT3    0x31ef
>>                 VF610_PAD_PTB20__GPIO_42        0x219d
>>         >;
>> };
>>
>> > >
>> > >> >  		}
>> > >> >  	}
>> > >> > @@ -292,6 +292,7 @@ static int imx_pmx_gpio_request_enable(struct
>> > >> > pinctrl_dev *pctldev,
>> > >> >  	reg = readl(ipctl->base + pin_reg->mux_reg);
>> > >> >  	reg &= ~info->mux_mask;
>> > >> >  	reg |= imx_pin->config;
>> > >> > +	reg |= imx_pin->mux_mode << info->mux_shift;
>> > >>
>> > >> ... and muxed...
>> > >>
>> > >> Not sure if we want that.
>> > >>
>> > >> I had to control GPIO output/input through pinctrl since Vybrid
>> > >> does not allow to control that from the GPIO block.
>> > >>
>> > >> However, according to your GPIO patchset, the i.MX 7ULP has a new
>> > >> register GPIO_PDDR to control output from the GPIO block. Is
>> > >> controlling the output driver from IOMUXC still required?
>> > >
>> > > Yes, it's still required.
>> > >
>> >
>> > That sounds weird, what is the GPIO_PDDR for then? Sure I  need to
>> > enable the output driver to drive the pin, but can I disable output
>> > just using GPIO_PDDR?
>>
>> No, to fully disable a output, you must disable OBE as well.
>>
>> >
>> > Maybe we have a miss understanding here:
>> >
>> > Lets assume we want to switch a GPIO between output and input:
>> >
>> > echo "output" > /sys/class/gpio/gpioN/direction ..
>> > echo "input" > /sys/class/gpio/gpioN/direction
>> >
>> > Do I need to disable the output driver in the IOMUXC or can we just
>> > disable GPIO_PDDR and use the pin as input?
>> >
>>
>> OBE should also be disabled. Otherwise the input may not function well.
>>
>> > If we can disable the output driver just using GPIO_PDDR, we can avoid
>> > the gpio_set_direction cross call.
>> >
>> >
>> > >> If not, I really would just not use all that "find pinctrl config"
>> > >> hackery... e.g. add a new flag, USE_IOMUXC_FOR_GPIO_OUTPUT, and set
>> > >> that only for Vybrid.
>> > >>
>> > >> This would also align much better with the other i.MX devices,
>> > >> where pinmuxing and GPIO is completely orthogonal.
>> > >>
>> > >
>> > > Actually this patch came only because to make the exist code not
>> > > break MX7ULP.
>> > >
>> > > Actually I'm wondering why we need implement
>> > > imx_pmx_gpio_request_enable function?
>> > >
>> > > Per my understanding, IMX binding already set GPIO mux by Parsing
>> > > MUX mode from device tree, so why need gpio_request_enable which
>> > > looks like is duplicated.
>> > >
>> > > Can you help explain it?
>> >
>> > I suggest to read this clarification email wrt to pinctrl/gpio from
>> > Linus
>> > Walleij:
>> > https://lkml.org/lkml/2016/10/10/87
>> >
>> > To summarize: We have different semantics in Vybrid: The GPIO
>> > subsystem automatically mux the GPIO for you. So in Vybrid, you do not
>> > have to mux a GPIO (but a valid entry in your device tree is needed,
>> > but does not assigned to any node).
>>
>> Okay, Clearer now.
>>
>> But I do see the users of GPIO pads in Vybrid dts.
>> Above is an example which make me confuse at first.
>>
>> >
>> > Is what the driver is doing for Vybrid wrong? It is different from
>> > i.MX, but afaik, it is not really wrong... Its a valid implementation
>> > according to the currently defined semantics...  Due to the *need* to
>> > touch pinctrl for direction, I had to implement cross calls anyway, so
>> > I thought I might as well go full mile and also mux the GPIO on
>> request...
>> >
>>
>> It's not strickly wrong.
>> Just a bit confuse that gpio_request_enable seems not quite necessary As
>> we actually already and must define GPIO mux in device tree according To
>> standard IMX binding format.
>> e.g. VF610_PAD_PTB20__GPIO_42 in above sd pad group.
>> That means pinctrl already does the GPIO mux when enable sd function.
>>
>> > So the question is, what semantic do we want for i.MX 7ULP? Since it
>> > is a i.MX device, we probably want the same semantics as i.MX 6/7 is
>> > already using for the sake of consistency. So in this case the
>> > gpio_request_enable/disable callbacks are not needed...
>> >
>> > This is how I hope we can do the implementation for i.MX 7ULP:
>> > - Do not use gpio_request_enable/disable
>>
>> Yes, we do want that.
>>
>> > - Do not use gpio_set_direction either
>>
>> Not, ULP needs it to support GPIO direction switch.
>>
>> > - Users using a GPIO should enable output driver in IOMUXC (just use a
>> > pin configuration where output driver is enabled)
>>
>> Users still need configure OBE/IBE in devicetree for statically assignment.
>>
>> > - The GPIO driver only enables/disables the output driver using its
>> > GPIO_PDDR register depending on GPIO direction
>>
>> No, same reason as the second question.
>>
>>
>> So, finnaly, I think the solution may be:
>> 1. If Vybrid does not use gpio_request_enable/disable, we can simply
>> remove it. Both driver keeps using pinctrl gpio_set_direction.
>>
>> Or.
>>
>> 2. Make gpio_request_enable/disable and gpio_set_direction As pinctrl-imx
>> core driver callbacks. And only assign gpio_set_direction For IMX7ULP
>> platform driver while assign both for Vybrid.
>>
>> Which one would you prefer?

1 would mean a semantic change.

For all GPIO I checked in upstream device trees we assign a pinctrl to
the same node, so in all cases gpio_request_enable/disable is really
unnecessary.

And since the current implementation has adversarial effects for I2C
recovery (https://patchwork.kernel.org/patch/9351401/), we use the
orthogonal semantic in the closely related i.MX SoCs and it even
simplifies the driver, I am ok to change the semantic.

Can you prepare such a patchset so that we can further test that on
Vybrid?

--
Stefan

More information about the linux-arm-kernel mailing list