Re: [PATCH 1/2] pinctrl: qcom: msm8226: Add MPM pin mappings

[Luca Weiss <luca@z3ntu.xyz>]

On Samstag, 23. September 2023 13:35:25 CEST Stephan Gerhold wrote:
> On Sat, Sep 23, 2023 at 01:19:46PM +0200, Luca Weiss wrote:
> > On Samstag, 23. September 2023 12:00:52 CEST Stephan Gerhold wrote:
> > > On Sat, Sep 23, 2023 at 11:32:47AM +0200, Luca Weiss wrote:
> > > > Hi Matti,
> > > > 
> > > > On Samstag, 23. September 2023 00:40:26 CEST Matti Lehtimäki wrote:
> > > > > Add pin <-> wakeirq mappings to allow for waking up the AP from
> > > > > sleep
> > > > > through MPM-connected pins.
> > > > > 
> > > > > Signed-off-by: Matti Lehtimäki <matti.lehtimaki@gmail.com>
> > > > > ---
> > > > > 
> > > > >  drivers/pinctrl/qcom/pinctrl-msm8226.c | 12 ++++++++++++
> > > > >  1 file changed, 12 insertions(+)
> > > > > 
> > > > > diff --git a/drivers/pinctrl/qcom/pinctrl-msm8226.c
> > > > > b/drivers/pinctrl/qcom/pinctrl-msm8226.c index
> > > > > 994619840a70..1e46a9ab382f
> > > > > 100644
> > > > > --- a/drivers/pinctrl/qcom/pinctrl-msm8226.c
> > > > > +++ b/drivers/pinctrl/qcom/pinctrl-msm8226.c
> > > > > @@ -612,6 +612,16 @@ static const struct msm_pingroup
> > > > > msm8226_groups[] =
> > > > > {
> > > > > 
> > > > >  #define NUM_GPIO_PINGROUPS 117
> > > > > 
> > > > > +static const struct msm_gpio_wakeirq_map msm8226_mpm_map[] = {
> > > > > +	{ 1, 3 }, { 4, 4 }, { 5, 5 }, { 9, 6 }, { 13, 7 }, { 17, 8 
},
> > > > 
> > > > I'm not really convinced this is the correct order of values...
> > > > 
> > > > Let's look at downstream:
> > > >   qcom,gpio-map = <3  1>,
> > > >   
> > > >                   <4  4 >,
> > > >                   <5  5 >,
> > > >                   <6  9 >,
> > > >                   [...]
> > > > 
> > > > From Documentation/devicetree/bindings/arm/msm/mpm.txt downstream:
> > > >   Each tuple represents a MPM pin and which GIC interrupt is routed to
> > > >   it.
> > > > 
> > > > So first is pin number, second is interrupt number.
> > > > 
> > > > And check mainline:
> > > >   /**
> > > >   
> > > >    * struct msm_gpio_wakeirq_map - Map of GPIOs and their wakeup pins
> > > >    * @gpio:          The GPIOs that are wakeup capable
> > > >    * @wakeirq:       The interrupt at the always-on interrupt
> > > >    controller
> > > >    */
> > > >   
> > > >   struct msm_gpio_wakeirq_map {
> > > >   
> > > >   	unsigned int gpio;
> > > >   	unsigned int wakeirq;
> > > >   
> > > >   };
> > > > 
> > > > So here we also have the order pin-interrupt, not the reverse order.
> > > > 
> > > > Therefore I believe the order in this patch is incorrect, and it
> > > > should
> > > > rather>
> > > > 
> > > > be:
> > > >   { 3, 1 }, { 4, 4 }, { 5, 5 }, { 6, 9 }, { 7, 13 }, { 8, 17 },
> > > >   [...]
> > > > 
> > > > Or do you think I'm missing something?
> > > 
> > > Yes :)
> > > 
> > > Let's look at the later entries:
> > > > > +	{ 21, 9 }, { 27, 10 }, { 29, 11 }, { 31, 12 }, { 33, 13 }, 
{ 35,
> > > > > 14
> > > > 
> > > > },
> > > > 
> > > > > +	{ 37, 15 }, { 38, 16 }, { 39, 17 }, { 41, 18 }, { 46, 19 
}, { 48,
> > > > > 20
> > > > 
> > > > },
> > > > 
> > > > > +	{ 49, 21 }, { 50, 22 }, { 51, 23 }, { 52, 24 }, { 54, 25 
}, { 62,
> > > > > 26
> > > > 
> > > > },
> > > > 
> > > > > +	{ 63, 27 }, { 64, 28 }, { 65, 29 }, { 66, 30 }, { 67, 31 
}, { 68,
> > > > > 32
> > > > 
> > > > },
> > > > 
> > > > > +	{ 69, 33 }, { 71, 34 }, { 72, 35 }, { 106, 36 }, { 107, 37 
},
> > > > > +	{ 108, 38 }, { 109, 39 }, { 110, 40 }, { 111, 54 }, { 113, 
55 },
> > > > > +};
> > > > > +
> > > 
> > > For example: { 113, 55 }, i.e. { .gpio = 113, .wakeirq = 55 }.
> > > 
> > > MSM8226 has GPIOs 0-116 and 64 MPM pins/interrupts. The order in this
> > > patch is the only one that can be correct because the definition would
> > > be invalid the other way around. 113 must be the GPIO number because it
> > > is larger than the 64 available MPM interrupt pins. :)
> > 
> > So basically you're saying downstream is wrong / buggy?
> 
> "Misleading" or "confusing" would be the words I would use. :-)

;)

> 
> > From qcom,gpio-map = [...], <55 113>; it's taking the properties like this
> > 
> > (drivers/soc/qcom/mpm-of.c):
> >   unsigned long pin = be32_to_cpup(list++);
> >   irq_hw_number_t hwirq = be32_to_cpup(list++);
> > 
> > Your explanation does make sense I guess but somewhere the link downstream
> > -> mainline must be broken, no?
> 
> After staring at mpm-of.c for a while I would say that there:
>  - downstream "pin" = MPM pin = mainline "wakeirq"
>    - because this is used as index to msm_mpm_irqs_m2a, which has a size
>      of MSM_MPM_NR_MPM_IRQS (64)
>  - downstream "hwirq" = GPIO / GIC IRQ = mainline "gpio"
> 
> This means for <55 113>: pin = wakeirq = 55 and hwirq = gpio = 113.
> Which matches the definition in this patch:
>   { .gpio = 113, .wakeirq = 55 } = { 113, 55 }

Fun, thanks for digging into it!

@Matti: I think I see one missing entry here "<41  115>," on downstream, so
{ 115, 41 } appears to be missing in this patch? Or is there a reason you 
omitted that one? The rest looks correct :)

Regards
Luca


> 
> Stephan