gpio-omap: wakeup mask

gpio-omap: wakeup mask

[Pascal Huerst]

Hi everyone,

we have a device with an am335x and are using some gpios on bank0 to
wake up the device from suspend to ram.

We have some user buttons which are configured in the devicetree as
gpio-keys and one power-key which should wake up the device:

&buttons {
        power {
                label = "Power";
                gpios = <&gpio0 6 1>;
                linux,code = <116>;
                gpio-key,wakeup;
        };

        one {
                label = "One";
                gpios = <&gpio0 11 1>;
                linux,code = <2>;
        };

        : : :
}

The problem is, that the device wakes up on any trigger on bank 0. No
matter which button was pressed. "gpio-key,wakeup" seems not to have any
influence.

Now, if I comment the following lines out in drivers/gpio/gpio-omap.c :
set_gpio_trigger(...).

321
322 if (likely(!(bank->non_wakeup_gpios & gpio_bit))) {
323    _gpio_rmw(base, bank->regs->wkup_en, gpio_bit, trigger != 0);
324      bank->context.wake_en =
325        readl_relaxed(bank->base + bank->regs->wkup_en);
326 }
327

Everything works as expected. But I don't really understand why? Is this
a bug, or does this break something else I have not discovered yet?

Thanks,
Pascal

Re: gpio-omap: wakeup mask

[Tony Lindgren]

* Pascal Huerst <pascal.huerst@gmail.com> [140702 05:12]:
> Hi everyone,
> 
> we have a device with an am335x and are using some gpios on bank0 to
> wake up the device from suspend to ram.
> 
> We have some user buttons which are configured in the devicetree as
> gpio-keys and one power-key which should wake up the device:
> 
> &buttons {
>         power {
>                 label = "Power";
>                 gpios = <&gpio0 6 1>;
>                 linux,code = <116>;
>                 gpio-key,wakeup;
>         };
> 
>         one {
>                 label = "One";
>                 gpios = <&gpio0 11 1>;
>                 linux,code = <2>;
>         };
> 
>         : : :
> }
> 
> The problem is, that the device wakes up on any trigger on bank 0. No
> matter which button was pressed. "gpio-key,wakeup" seems not to have any
> influence.
> 
> Now, if I comment the following lines out in drivers/gpio/gpio-omap.c :
> set_gpio_trigger(...).
> 
> 321
> 322 if (likely(!(bank->non_wakeup_gpios & gpio_bit))) {
> 323    _gpio_rmw(base, bank->regs->wkup_en, gpio_bit, trigger != 0);
> 324      bank->context.wake_en =
> 325        readl_relaxed(bank->base + bank->regs->wkup_en);
> 326 }
> 327
> 
> Everything works as expected. But I don't really understand why? Is this
> a bug, or does this break something else I have not discovered yet?

The GPIO wake-up events are working from off-idle for omap3
the last time I checked, so sounds like this is some am335x
related issue.

What does your bank->context.wake_en mask get set to with this code?

Regards,

Tony

Re: gpio-omap: wakeup mask

[Pascal Huerst]

Hey Tony, all

On 04.07.2014 09:27, Tony Lindgren wrote:
> * Pascal Huerst <pascal.huerst@gmail.com> [140702 05:12]:
>> Hi everyone,
>>
>> we have a device with an am335x and are using some gpios on bank0 to
>> wake up the device from suspend to ram.
>>
>> We have some user buttons which are configured in the devicetree as
>> gpio-keys and one power-key which should wake up the device:
>>
>> &buttons {
>>         power {
>>                 label = "Power";
>>                 gpios = <&gpio0 6 1>;
>>                 linux,code = <116>;
>>                 gpio-key,wakeup;
>>         };
>>
>>         one {
>>                 label = "One";
>>                 gpios = <&gpio0 11 1>;
>>                 linux,code = <2>;
>>         };
>>
>>         : : :
>> }
>>
>> The problem is, that the device wakes up on any trigger on bank 0. No
>> matter which button was pressed. "gpio-key,wakeup" seems not to have any
>> influence.
>>
>> Now, if I comment the following lines out in drivers/gpio/gpio-omap.c :
>> set_gpio_trigger(...).
>>
>> 321
>> 322 if (likely(!(bank->non_wakeup_gpios & gpio_bit))) {
>> 323    _gpio_rmw(base, bank->regs->wkup_en, gpio_bit, trigger != 0);
>> 324      bank->context.wake_en =
>> 325        readl_relaxed(bank->base + bank->regs->wkup_en);
>> 326 }
>> 327
>>
>> Everything works as expected. But I don't really understand why? Is this
>> a bug, or does this break something else I have not discovered yet?
> 
> The GPIO wake-up events are working from off-idle for omap3
> the last time I checked, so sounds like this is some am335x
> related issue.

Ok, that sounds reasonable.

> What does your bank->context.wake_en mask get set to with this code?

by calling 'echo mem > /sys/power/state' followed by a wakeup, I get the
following:

[   28.738416] PM: Syncing filesystems ... done.
[   28.779153] Freezing user space processes ... (elapsed 0.003 seconds)
done.
[   28.790236] Freezing remaining freezable tasks ... (elapsed 0.002
seconds) done.
[   28.801386] PM: Sending message for entering DeepSleep mode
[   28.935412] bank->context->wake_en = 00080840
[   28.945935] PM: suspend of devices complete after 133.156 msecs
[   28.961008] PM: late suspend of devices complete after 8.834 msecs
[   28.978874] omap_hwmod: cpgmac0: _wait_target_disable failed
[   28.985940] PM: noirq suspend of devices complete after 18.324 msecs
[   28.992697] PM: GFX domain entered low power state
[   28.992697] PM: Could not transition all powerdomains to target state
[   28.992697] PM: Wakeup source GPIO0
[   28.992866] bank->context->wake_en = 00080040
[   29.020974] PM: noirq resume of devices complete after 22.439 msecs
[   29.028257] bank->context->wake_en = 00080040
[   29.032840] bank->context->wake_en = 00080840
[   29.039207] bank->context->wake_en = 00080840
[   29.043961] bank->context->wake_en = 0000F200
[   29.048534] bank->context->wake_en = 0000F200
[   29.053093] bank->context->wake_en = 0000F200
[   29.057704] bank->context->wake_en = 0000F200
[   29.062266] bank->context->wake_en = 0000F200
[   29.066992] bank->context->wake_en = 00060000
[   29.071554] bank->context->wake_en = 00060000
[   29.082499] PM: early resume of devices complete after 6.136 msecs
[   29.091048] net eth0: initializing cpsw version 1.12 (0)
[   29.173888] net eth0: phy found : id is : 0x4dd076
[   29.229388] mmc0: error -110 during resume (card was removed?)
[   29.235609] dpm_run_callback(): mmc_bus_resume+0x0/0x74 returns -110
[   29.242272] PM: Device mmc0:0001 failed to resume: error -110
[   29.293992] PM: resume of devices complete after 204.891 msecs
[   29.306801] PM: Sending message for resetting M3 state machine
[   29.313693] Restarting tasks ... done.

Thanks
Pascal

Re: gpio-omap: wakeup mask

[Tony Lindgren]

* Pascal Huerst <pascal.huerst@gmail.com> [140709 03:18]:
> Hey Tony, all
> 
> On 04.07.2014 09:27, Tony Lindgren wrote:
> > * Pascal Huerst <pascal.huerst@gmail.com> [140702 05:12]:
> >> Hi everyone,
> >>
> >> we have a device with an am335x and are using some gpios on bank0 to
> >> wake up the device from suspend to ram.
> >>
> >> We have some user buttons which are configured in the devicetree as
> >> gpio-keys and one power-key which should wake up the device:
> >>
> >> &buttons {
> >>         power {
> >>                 label = "Power";
> >>                 gpios = <&gpio0 6 1>;
> >>                 linux,code = <116>;
> >>                 gpio-key,wakeup;
> >>         };
> >>
> >>         one {
> >>                 label = "One";
> >>                 gpios = <&gpio0 11 1>;
> >>                 linux,code = <2>;
> >>         };
> >>
> >>         : : :
> >> }
> >>
> >> The problem is, that the device wakes up on any trigger on bank 0. No
> >> matter which button was pressed. "gpio-key,wakeup" seems not to have any
> >> influence.
> >>
> >> Now, if I comment the following lines out in drivers/gpio/gpio-omap.c :
> >> set_gpio_trigger(...).
> >>
> >> 321
> >> 322 if (likely(!(bank->non_wakeup_gpios & gpio_bit))) {
> >> 323    _gpio_rmw(base, bank->regs->wkup_en, gpio_bit, trigger != 0);
> >> 324      bank->context.wake_en =
> >> 325        readl_relaxed(bank->base + bank->regs->wkup_en);
> >> 326 }
> >> 327
> >>
> >> Everything works as expected. But I don't really understand why? Is this
> >> a bug, or does this break something else I have not discovered yet?
> > 
> > The GPIO wake-up events are working from off-idle for omap3
> > the last time I checked, so sounds like this is some am335x
> > related issue.
> 
> Ok, that sounds reasonable.
> 
> > What does your bank->context.wake_en mask get set to with this code?
> 
> by calling 'echo mem > /sys/power/state' followed by a wakeup, I get the
> following:
> 
> [   28.992866] bank->context->wake_en = 00080040
> [   29.020974] PM: noirq resume of devices complete after 22.439 msecs

I guess that's for gpio bank 0 with no bit 11 as your dts has it
configured. If you just comment out the _gpio_rmw part above do
things work as expected?

Then if that works as expected, maybe write only some unused bits
into wkup_en register and see if it still wakes to all events while
it should not?

It might be worth comparing the GPIO registers between am37xx and
am335x for the wkup_en register.

Regards,

Tony

Re: gpio-omap: wakeup mask

[Pascal Huerst]

On 09.07.2014 12:41, Tony Lindgren wrote:
> * Pascal Huerst <pascal.huerst@gmail.com> [140709 03:18]:
>> Hey Tony, all
>>
>> On 04.07.2014 09:27, Tony Lindgren wrote:
>>> * Pascal Huerst <pascal.huerst@gmail.com> [140702 05:12]:
>>>> Hi everyone,
>>>>
>>>> we have a device with an am335x and are using some gpios on bank0 to
>>>> wake up the device from suspend to ram.
>>>>
>>>> We have some user buttons which are configured in the devicetree as
>>>> gpio-keys and one power-key which should wake up the device:
>>>>
>>>> &buttons {
>>>>         power {
>>>>                 label = "Power";
>>>>                 gpios = <&gpio0 6 1>;
>>>>                 linux,code = <116>;
>>>>                 gpio-key,wakeup;
>>>>         };
>>>>
>>>>         one {
>>>>                 label = "One";
>>>>                 gpios = <&gpio0 11 1>;
>>>>                 linux,code = <2>;
>>>>         };
>>>>
>>>>         : : :
>>>> }
>>>>
>>>> The problem is, that the device wakes up on any trigger on bank 0. No
>>>> matter which button was pressed. "gpio-key,wakeup" seems not to have any
>>>> influence.
>>>>
>>>> Now, if I comment the following lines out in drivers/gpio/gpio-omap.c :
>>>> set_gpio_trigger(...).
>>>>
>>>> 321
>>>> 322 if (likely(!(bank->non_wakeup_gpios & gpio_bit))) {
>>>> 323    _gpio_rmw(base, bank->regs->wkup_en, gpio_bit, trigger != 0);
>>>> 324      bank->context.wake_en =
>>>> 325        readl_relaxed(bank->base + bank->regs->wkup_en);
>>>> 326 }
>>>> 327
>>>>
>>>> Everything works as expected. But I don't really understand why? Is this
>>>> a bug, or does this break something else I have not discovered yet?
>>>
>>> The GPIO wake-up events are working from off-idle for omap3
>>> the last time I checked, so sounds like this is some am335x
>>> related issue.
>>
>> Ok, that sounds reasonable.
>>
>>> What does your bank->context.wake_en mask get set to with this code?
>>
>> by calling 'echo mem > /sys/power/state' followed by a wakeup, I get the
>> following:
>>
>> [   28.992866] bank->context->wake_en = 00080040
>> [   29.020974] PM: noirq resume of devices complete after 22.439 msecs
> 
> I guess that's for gpio bank 0 with no bit 11 as your dts has it
> configured. 

Yes that's right. I forgot to print the context, too. If I do so, it's:

[   31.034972] bank->base = F9E07000 gpio = 6 bank->context->wake_en =
00080040

respectively:

[   31.846992] bank->base = F9E07000 gpio = 11 bank->context->wake_en =
00080040

> If you just comment out the _gpio_rmw part above do
> things work as expected?

Yes. It only wakes up on gpio 6 not on gpio 11 anymore.

> Then if that works as expected, maybe write only some unused bits
> into wkup_en register and see if it still wakes to all events while
> it should not?

If I use the following testcode, everything works as expected:

//WARN(true, "WARN: bank->base = %08X\n", bank->base);

if (likely(!(bank->non_wakeup_gpios & gpio_bit))) {
  //_gpio_rmw(base, bank->regs->wkup_en, gpio_bit, trigger != 0);
  writel(0xABC00000, base + bank->regs->wkup_en);

  bank->context.wake_en = readl_relaxed(bank->base +
    bank->regs->wkup_en);
  printk("bank->base = %08X gpio = %i bank->context->wake_en = %08X\n",
    bank->base, gpio, bank->context.wake_en);
}

This produces the following output:

[   35.319228] PM: Syncing filesystems ... done.
[   35.358702] Freezing user space processes ... (elapsed 0.003 seconds)
done.
[   35.369754] Freezing remaining freezable tasks ... (elapsed 0.002
seconds) done.
[   35.380734] PM: Sending message for entering DeepSleep mode
[   35.515932] bank->base = F9E07000 gpio = 19 bank->context->wake_en =
ABC00000
[   35.529319] PM: suspend of devices complete after 137.218 msecs
[   35.544410] PM: late suspend of devices complete after 8.715 msecs
[   35.562247] omap_hwmod: cpgmac0: _wait_target_disable failed
[   35.569325] PM: noirq suspend of devices complete after 18.289 msecs
[   35.576156] PM: GFX domain entered low power state
[   35.576156] PM: Could not transition all powerdomains to target state
[   35.576156] PM: Wakeup source GPIO0
[   35.576156] PM Wakeup source extra: 00000840
[   35.576329] bank->base = F9E07000 gpio = 6 bank->context->wake_en =
ABC00000
[   35.583710] bank->base = F9E07000 gpio = 11 bank->context->wake_en =
ABC00000
[   35.614846] PM: noirq resume of devices complete after 22.582 msecs
[   35.622009] bank->base = F9E07000 gpio = 6 bank->context->wake_en =
ABC00000
[   35.631798] bank->base = F9E07000 gpio = 11 bank->context->wake_en =
ABC00000
[   35.639511] bank->base = F9E07000 gpio = 19 bank->context->wake_en =
ABC00000
[   35.647116] bank->base = FA04C000 gpio = 9 bank->context->wake_en =
ABC00000
[   35.654547] bank->base = FA04C000 gpio = 12 bank->context->wake_en =
ABC00000
[   35.662015] bank->base = FA04C000 gpio = 13 bank->context->wake_en =
ABC00000
[   35.669526] bank->base = FA04C000 gpio = 14 bank->context->wake_en =
ABC00000
[   35.677039] bank->base = FA04C000 gpio = 15 bank->context->wake_en =
ABC00000
[   35.685035] bank->base = FA1AC000 gpio = 17 bank->context->wake_en =
ABC00000
[   35.692507] bank->base = FA1AC000 gpio = 18 bank->context->wake_en =
ABC00000
[   35.706466] PM: early resume of devices complete after 6.235 msecs
[   35.715010] net eth0: initializing cpsw version 1.12 (0)
[   35.794388] net eth0: phy found : id is : 0x4dd076
[   35.849941] mmc0: error -110 during resume (card was removed?)
[   35.856164] dpm_run_callback(): mmc_bus_resume+0x0/0x74 returns -110
[   35.862828] PM: Device mmc0:0001 failed to resume: error -110
[   35.914573] PM: resume of devices complete after 201.604 msecs
[   35.927336] PM: Sending message for resetting M3 state machine
[   35.934236] Restarting tasks ... done.

> It might be worth comparing the GPIO registers between am37xx and
> am335x for the wkup_en register.

I had a quick look, but did not find anything. May be I will have to
look into that again.

Thanks
Pascal

Re: gpio-omap: wakeup mask

[Tony Lindgren]

* Pascal Huerst <pascal.huerst@gmail.com> [140709 05:47]:
> On 09.07.2014 12:41, Tony Lindgren wrote:
> 
> > If you just comment out the _gpio_rmw part above do
> > things work as expected?
> 
> Yes. It only wakes up on gpio 6 not on gpio 11 anymore.
> 
> > Then if that works as expected, maybe write only some unused bits
> > into wkup_en register and see if it still wakes to all events while
> > it should not?
> 
> If I use the following testcode, everything works as expected:
> 
> //WARN(true, "WARN: bank->base = %08X\n", bank->base);
> 
> if (likely(!(bank->non_wakeup_gpios & gpio_bit))) {
>   //_gpio_rmw(base, bank->regs->wkup_en, gpio_bit, trigger != 0);
>   writel(0xABC00000, base + bank->regs->wkup_en);
> 
>   bank->context.wake_en = readl_relaxed(bank->base +
>     bank->regs->wkup_en);
>   printk("bank->base = %08X gpio = %i bank->context->wake_en = %08X\n",
>     bank->base, gpio, bank->context.wake_en);
> }

Hmm weird. It sounds like something like the following is happening:

1. The first GPIO bank is always powered, and does not need to set
   wake-up events

2. When setting the GPIO wake-up events it seems that enabling any
   wake-up event for the first (16?) bits wakes up the system

You might want to check this with some spare GPIOs not in the
first bank and see if you need the wake-up events and if enabling
some bits enables more than one GPIO for wake-up events.

Regards,

Tony

Re: gpio-omap: wakeup mask

[Pascal Huerst]

-----BEGIN PGP SIGNED MESSAGE-----
Hash: SHA1



On 09.07.2014 17:24, Tony Lindgren wrote:
> * Pascal Huerst <pascal.huerst@gmail.com> [140709 05:47]:
>> On 09.07.2014 12:41, Tony Lindgren wrote:
>> 
>>> If you just comment out the _gpio_rmw part above do things work
>>> as expected?
>> 
>> Yes. It only wakes up on gpio 6 not on gpio 11 anymore.
>> 
>>> Then if that works as expected, maybe write only some unused
>>> bits into wkup_en register and see if it still wakes to all
>>> events while it should not?
>> 
>> If I use the following testcode, everything works as expected:
>> 
>> //WARN(true, "WARN: bank->base = %08X\n", bank->base);
>> 
>> if (likely(!(bank->non_wakeup_gpios & gpio_bit))) { 
>> //_gpio_rmw(base, bank->regs->wkup_en, gpio_bit, trigger != 0); 
>> writel(0xABC00000, base + bank->regs->wkup_en);
>> 
>> bank->context.wake_en = readl_relaxed(bank->base + 
>> bank->regs->wkup_en); printk("bank->base = %08X gpio = %i
>> bank->context->wake_en = %08X\n", bank->base, gpio,
>> bank->context.wake_en); }
> 
> Hmm weird. It sounds like something like the following is
> happening:
> 
> 1. The first GPIO bank is always powered, and does not need to set 
> wake-up events

I dont think so. If I apply my hack (commenting the _gpio_rmw() call)
and don't have a wakeup source in the dt, I have to reboot the system,
sonce im not able to wake it up anymore. Or did I miss something?

> 2. When setting the GPIO wake-up events it seems that enabling any 
> wake-up event for the first (16?) bits wakes up the system
> 
> You might want to check this with some spare GPIOs not in the first
> bank and see if you need the wake-up events and if enabling some
> bits enables more than one GPIO for wake-up events.

Im not sure if this makes sense, since only the first gpio bank can
act as a wakeup source. So it should not be possible.


thanks!
pascal
-----BEGIN PGP SIGNATURE-----
Version: GnuPG v1
Comment: Using GnuPG with Thunderbird - http://www.enigmail.net/

iEYEARECAAYFAlPFLEYACgkQo7eFcXQQ8U1NcACbBzQiKoU5MO/MLSdrjeg5CaAc
YJcAn3bSthD4ByFOQ31jjOFI1F/HP4Lr
=o4r2
-----END PGP SIGNATURE-----

Re: gpio-omap: wakeup mask

[Tony Lindgren]

* Pascal Huerst <pascal.huerst@gmail.com> [140715 06:29]:
> 
> 
> On 09.07.2014 17:24, Tony Lindgren wrote:
> > * Pascal Huerst <pascal.huerst@gmail.com> [140709 05:47]:
> >> On 09.07.2014 12:41, Tony Lindgren wrote:
> >> 
> >>> If you just comment out the _gpio_rmw part above do things work
> >>> as expected?
> >> 
> >> Yes. It only wakes up on gpio 6 not on gpio 11 anymore.
> >> 
> >>> Then if that works as expected, maybe write only some unused
> >>> bits into wkup_en register and see if it still wakes to all
> >>> events while it should not?
> >> 
> >> If I use the following testcode, everything works as expected:
> >> 
> >> //WARN(true, "WARN: bank->base = %08X\n", bank->base);
> >> 
> >> if (likely(!(bank->non_wakeup_gpios & gpio_bit))) { 
> >> //_gpio_rmw(base, bank->regs->wkup_en, gpio_bit, trigger != 0); 
> >> writel(0xABC00000, base + bank->regs->wkup_en);
> >> 
> >> bank->context.wake_en = readl_relaxed(bank->base + 
> >> bank->regs->wkup_en); printk("bank->base = %08X gpio = %i
> >> bank->context->wake_en = %08X\n", bank->base, gpio,
> >> bank->context.wake_en); }
> > 
> > Hmm weird. It sounds like something like the following is
> > happening:
> > 
> > 1. The first GPIO bank is always powered, and does not need to set 
> > wake-up events
> 
> I dont think so. If I apply my hack (commenting the _gpio_rmw() call)
> and don't have a wakeup source in the dt, I have to reboot the system,
> sonce im not able to wake it up anymore. Or did I miss something?

Well this just from my head without looking at the code, so please
check..

When you comment out the _gpio_rmw() call you are not configuring
any GPIO bank wake-up events.

And the .dts entry just manages the keypad wake-up event for
suspend and resume and does not affect the GPIO bank after you
commented out the _gpio_rmw().
 
> > 2. When setting the GPIO wake-up events it seems that enabling any 
> > wake-up event for the first (16?) bits wakes up the system
> > 
> > You might want to check this with some spare GPIOs not in the first
> > bank and see if you need the wake-up events and if enabling some
> > bits enables more than one GPIO for wake-up events.
> 
> Im not sure if this makes sense, since only the first gpio bank can
> act as a wakeup source. So it should not be possible.

Usually on omaps the first GPIO bank is always powered but also
the other banks can provide wake-up events for most GPIOs
via the IO chain path. I think on am335x, there is no IO chain
path, maybe because of the support for dual voltage IO? So yeah
it could be that only the first GPIO bank can wake up the system
on am335x.

Regards,

Tony