soc: sunxi: select CONFIG_PM

  From: Arnd Bergmann <arnd@arndb.de>

Selecting CONFIG_PM_GENERIC_DOMAINS without CONFIG_PM leads to a
build failure:

WARNING: unmet direct dependencies detected for PM_GENERIC_DOMAINS
  Depends on [n]: PM [=n]
  Selected by [y]:
  - SUN20I_PPU [=y] && (ARCH_SUNXI [=n] || COMPILE_TEST [=y])

drivers/base/power/domain_governor.c: In function 'default_suspend_ok':
drivers/base/power/domain_governor.c:85:24: error: 'struct dev_pm_info' has no member named 'ignore_children'
   85 |         if (!dev->power.ignore_children)
      |                        ^
drivers/base/power/domain.c: In function 'genpd_queue_power_off_work':
drivers/base/power/domain.c:657:20: error: 'pm_wq' undeclared (first use in this function)
  657 |         queue_work(pm_wq, &genpd->power_off_work);
      |                    ^~~~~

Unfortunately platforms are inconsistent between using 'select PM'
and 'depends on PM' here. CONFIG_PM is a user-visible symbol, so
in principle we should be using 'depends on', but on the other hand
using 'select' here is more common among drivers/soc. Go with the
majority for now, as this has a smaller risk of introducing circular
dependencies. We may need to clean this up for consistency later.

Fixes: 0e30ca5ab0a8 ("soc: sunxi: Add Allwinner D1 PPU driver")
Signed-off-by: Arnd Bergmann <arnd@arndb.de>
---
 drivers/soc/sunxi/Kconfig | 1 +
 1 file changed, 1 insertion(+)
  

Dne ponedeljek, 30. januar 2023 ob 14:04:47 CET je Arnd Bergmann napisal(a):
> From: Arnd Bergmann <arnd@arndb.de>
> 
> Selecting CONFIG_PM_GENERIC_DOMAINS without CONFIG_PM leads to a
> build failure:
> 
> WARNING: unmet direct dependencies detected for PM_GENERIC_DOMAINS
>   Depends on [n]: PM [=n]
>   Selected by [y]:
>   - SUN20I_PPU [=y] && (ARCH_SUNXI [=n] || COMPILE_TEST [=y])
> 
> drivers/base/power/domain_governor.c: In function 'default_suspend_ok':
> drivers/base/power/domain_governor.c:85:24: error: 'struct dev_pm_info' has
> no member named 'ignore_children' 85 |         if
> (!dev->power.ignore_children)
> 
>       |                        ^
> 
> drivers/base/power/domain.c: In function 'genpd_queue_power_off_work':
> drivers/base/power/domain.c:657:20: error: 'pm_wq' undeclared (first use in
> this function) 657 |         queue_work(pm_wq, &genpd->power_off_work);
> 
>       |                    ^~~~~
> 
> Unfortunately platforms are inconsistent between using 'select PM'
> and 'depends on PM' here. CONFIG_PM is a user-visible symbol, so
> in principle we should be using 'depends on', but on the other hand
> using 'select' here is more common among drivers/soc. Go with the
> majority for now, as this has a smaller risk of introducing circular
> dependencies. We may need to clean this up for consistency later.
> 
> Fixes: 0e30ca5ab0a8 ("soc: sunxi: Add Allwinner D1 PPU driver")
> Signed-off-by: Arnd Bergmann <arnd@arndb.de>

Acked-by: Jernej Skrabec <jernej.skrabec@gmail.com>

Although, would it be better if commit in question is fixed and another PR is 
issued? That way we can avoid issue with broken config.

Best regards,
Jernej
  

On Mon, Jan 30, 2023, at 22:18, Jernej Škrabec wrote:
> Dne ponedeljek, 30. januar 2023 ob 14:04:47 CET je Arnd Bergmann napisal(a):
>> From: Arnd Bergmann <arnd@arndb.de>
>> 
>> Selecting CONFIG_PM_GENERIC_DOMAINS without CONFIG_PM leads to a
>> build failure:
>> 
>> WARNING: unmet direct dependencies detected for PM_GENERIC_DOMAINS
>>   Depends on [n]: PM [=n]
>>   Selected by [y]:
>>   - SUN20I_PPU [=y] && (ARCH_SUNXI [=n] || COMPILE_TEST [=y])
>> 
>> drivers/base/power/domain_governor.c: In function 'default_suspend_ok':
>> drivers/base/power/domain_governor.c:85:24: error: 'struct dev_pm_info' has
>> no member named 'ignore_children' 85 |         if
>> (!dev->power.ignore_children)
>> 
>>       |                        ^
>> 
>> drivers/base/power/domain.c: In function 'genpd_queue_power_off_work':
>> drivers/base/power/domain.c:657:20: error: 'pm_wq' undeclared (first use in
>> this function) 657 |         queue_work(pm_wq, &genpd->power_off_work);
>> 
>>       |                    ^~~~~
>> 
>> Unfortunately platforms are inconsistent between using 'select PM'
>> and 'depends on PM' here. CONFIG_PM is a user-visible symbol, so
>> in principle we should be using 'depends on', but on the other hand
>> using 'select' here is more common among drivers/soc. Go with the
>> majority for now, as this has a smaller risk of introducing circular
>> dependencies. We may need to clean this up for consistency later.
>> 
>> Fixes: 0e30ca5ab0a8 ("soc: sunxi: Add Allwinner D1 PPU driver")
>> Signed-off-by: Arnd Bergmann <arnd@arndb.de>
>
> Acked-by: Jernej Skrabec <jernej.skrabec@gmail.com>
>
> Although, would it be better if commit in question is fixed and another PR is 
> issued? That way we can avoid issue with broken config.

Agreed, merging it into the original commit would have been ideal.
unfortunately the half of my brain that does the pull request
didn't talk to the half that wrote the bugfix, so I already pulled
it in after I sent the fix for linux-next...

I'll just add the fix with your Ack on top then.

    Arnd
  

Hi--

On 1/30/23 05:04, Arnd Bergmann wrote:
> From: Arnd Bergmann <arnd@arndb.de>
> 
> Selecting CONFIG_PM_GENERIC_DOMAINS without CONFIG_PM leads to a
> build failure:
> 
> WARNING: unmet direct dependencies detected for PM_GENERIC_DOMAINS
>   Depends on [n]: PM [=n]
>   Selected by [y]:
>   - SUN20I_PPU [=y] && (ARCH_SUNXI [=n] || COMPILE_TEST [=y])
> 
> drivers/base/power/domain_governor.c: In function 'default_suspend_ok':
> drivers/base/power/domain_governor.c:85:24: error: 'struct dev_pm_info' has no member named 'ignore_children'
>    85 |         if (!dev->power.ignore_children)
>       |                        ^
> drivers/base/power/domain.c: In function 'genpd_queue_power_off_work':
> drivers/base/power/domain.c:657:20: error: 'pm_wq' undeclared (first use in this function)
>   657 |         queue_work(pm_wq, &genpd->power_off_work);
>       |                    ^~~~~
> 
> Unfortunately platforms are inconsistent between using 'select PM'
> and 'depends on PM' here. CONFIG_PM is a user-visible symbol, so
> in principle we should be using 'depends on', but on the other hand
> using 'select' here is more common among drivers/soc. Go with the
> majority for now, as this has a smaller risk of introducing circular
> dependencies. We may need to clean this up for consistency later.
> 
> Fixes: 0e30ca5ab0a8 ("soc: sunxi: Add Allwinner D1 PPU driver")
> Signed-off-by: Arnd Bergmann <arnd@arndb.de>

With this patch (in linux-next-20230203), building on sparc32 with
COMPILE_TEST=y and ARCH_SUNXI not set:

WARNING: unmet direct dependencies detected for PM
  Depends on [n]: SPARC64 [=n]
  Selected by [y]:
  - SUN20I_PPU [=y] && (ARCH_SUNXI || COMPILE_TEST [=y])

WARNING: unmet direct dependencies detected for PM
  Depends on [n]: SPARC64 [=n]
  Selected by [y]:
  - SUN20I_PPU [=y] && (ARCH_SUNXI || COMPILE_TEST [=y])

WARNING: unmet direct dependencies detected for PM_GENERIC_DOMAINS
  Depends on [n]: SPARC64 [=n] && PM [=y]
  Selected by [y]:
  - QCOM_GDSC [=y] && COMMON_CLK [=y] && PM [=y]
  - SUN20I_PPU [=y] && (ARCH_SUNXI || COMPILE_TEST [=y])
  - MESON_GX_PM_DOMAINS [=y] && (ARCH_MESON || COMPILE_TEST [=y]) && PM [=y] && OF [=y]
  - BCM2835_POWER [=y] && (ARCH_BCM2835 || COMPILE_TEST [=y] && OF [=y]) && PM [=y]
  - BCM_PMB [=y] && (ARCH_BCMBCA || COMPILE_TEST [=y] && OF [=y]) && PM [=y]
  - ROCKCHIP_PM_DOMAINS [=y] && (ARCH_ROCKCHIP || COMPILE_TEST [=y]) && PM [=y]
  Selected by [m]:
  - ARM_SCPI_POWER_DOMAIN [=m] && (ARM_SCPI_PROTOCOL [=m] || COMPILE_TEST [=y] && OF [=y]) && PM [=y]
  - MESON_EE_PM_DOMAINS [=m] && (ARCH_MESON || COMPILE_TEST [=y]) && PM [=y] && OF [=y]
  - QCOM_AOSS_QMP [=m] && (ARCH_QCOM || COMPILE_TEST [=y]) && MAILBOX [=y] && COMMON_CLK [=y] && PM [=y]

WARNING: unmet direct dependencies detected for PM_GENERIC_DOMAINS_OF
  Depends on [n]: SPARC64 [=n] && PM_GENERIC_DOMAINS [=y] && OF [=y]
  Selected by [y]:
  - MESON_GX_PM_DOMAINS [=y] && (ARCH_MESON || COMPILE_TEST [=y]) && PM [=y] && OF [=y]
  Selected by [m]:
  - MESON_EE_PM_DOMAINS [=m] && (ARCH_MESON || COMPILE_TEST [=y]) && PM [=y] && OF [=y]

Apparently sparc32 does not support PM (arch/sparc/Kconfig):

if SPARC64
source "kernel/power/Kconfig"
endif

so I think that SUN20I_PPU should also depend on !SPARC32.
Does that make sense?

Thanks.

> ---
>  drivers/soc/sunxi/Kconfig | 1 +
>  1 file changed, 1 insertion(+)
> 
> diff --git a/drivers/soc/sunxi/Kconfig b/drivers/soc/sunxi/Kconfig
> index 29e9ba2d520d..02d0b338feb3 100644
> --- a/drivers/soc/sunxi/Kconfig
> +++ b/drivers/soc/sunxi/Kconfig
> @@ -23,6 +23,7 @@ config SUNXI_SRAM
>  config SUN20I_PPU
>  	bool "Allwinner D1 PPU power domain driver"
>  	depends on ARCH_SUNXI || COMPILE_TEST
> +	select PM
>  	select PM_GENERIC_DOMAINS
>  	help
>  	  Say y to enable the PPU power domain driver. This saves power
  

On Sat, Feb 4, 2023, at 08:47, Randy Dunlap wrote:
> On 1/30/23 05:04, Arnd Bergmann wrote:
>
> Apparently sparc32 does not support PM (arch/sparc/Kconfig):
>
> if SPARC64
> source "kernel/power/Kconfig"
> endif
>
> so I think that SUN20I_PPU should also depend on !SPARC32.
> Does that make sense?

I would suggest working around this in arch/sparc/ instead
of every driver that uses 'select PM', perhaps something like

--- a/arch/sparc/Kconfig
+++ b/arch/sparc/Kconfig
@@ -283,7 +283,7 @@ config ARCH_FORCE_MAX_ORDER
          This config option is actually maximum order plus one. For example,
          a value of 13 means that the largest free memory block is 2^12 pages.
 
-if SPARC64
+if SPARC64 || COMPILE_TEST
 source "kernel/power/Kconfig"
 endif
 
The issue does not happen anywhere else, as it's fine if
kernel/power/Kconfig is not included at all. I'm also not too
worried about random sparc32 configs since it's pretty much impossible
to build a sparc32 allmodconfig or randconfig, with all the other
bugs that runs into.

     Arnd
  

On 2/4/23 02:38, Arnd Bergmann wrote:
> On Sat, Feb 4, 2023, at 08:47, Randy Dunlap wrote:
>> On 1/30/23 05:04, Arnd Bergmann wrote:
>>
>> Apparently sparc32 does not support PM (arch/sparc/Kconfig):
>>
>> if SPARC64
>> source "kernel/power/Kconfig"
>> endif
>>
>> so I think that SUN20I_PPU should also depend on !SPARC32.
>> Does that make sense?
> 
> I would suggest working around this in arch/sparc/ instead
> of every driver that uses 'select PM', perhaps something like
> 
> --- a/arch/sparc/Kconfig
> +++ b/arch/sparc/Kconfig
> @@ -283,7 +283,7 @@ config ARCH_FORCE_MAX_ORDER
>           This config option is actually maximum order plus one. For example,
>           a value of 13 means that the largest free memory block is 2^12 pages.
>  
> -if SPARC64
> +if SPARC64 || COMPILE_TEST
>  source "kernel/power/Kconfig"
>  endif
>

That looks good. Thanks.

> The issue does not happen anywhere else, as it's fine if
> kernel/power/Kconfig is not included at all. I'm also not too
> worried about random sparc32 configs since it's pretty much impossible
> to build a sparc32 allmodconfig or randconfig, with all the other
> bugs that runs into.

OK/ack.