[GIT,PULL,3/4] ARM: imx: device tree changes for 4.2

Commit Message

Shawn Guo June 2, 2015, 3:33 a.m. UTC

On Mon, Jun 01, 2015 at 01:52:53PM -0300, Fabio Estevam wrote:
> > Also, could you fold this change into the patch that moves the gpt driver?
> >
> > diff --git a/drivers/clocksource/Kconfig b/drivers/clocksource/Kconfig
> > index 3c15a49fdac9..fd43dbb5e47b 100644
> > --- a/drivers/clocksource/Kconfig
> > +++ b/drivers/clocksource/Kconfig
> > @@ -278,7 +278,7 @@ config CLKSRC_PXA
> >
> >  config CLKSRC_IMX_GPT
> >         bool "Clocksource using i.MX GPT" if COMPILE_TEST
> > -       depends on OF
> > +       depends on OF && COMMON_CLK
> >         select CLKSRC_MMIO
> >
> >  config H8300_TMR16
> >
> > Without this, we get build errors for COMPILE_TEST.
> 
> Shouldn't it be like this instead?
> 
> --- a/drivers/clocksource/Kconfig
> +++ b/drivers/clocksource/Kconfig
> @@ -278,7 +278,7 @@ config CLKSRC_PXA
> 
>  config CLKSRC_IMX_GPT
>         bool "Clocksource using i.MX GPT" if COMPILE_TEST
> -       depends on OF
> +       select CLKSRC_OF if OF
>         select CLKSRC_MMIO

I just received a message [1] from Stephen which reports the build
failures on powerpc allyesconfig.  Basically, the errors are around the
following 3 things.

 1) delay_timer
 2) clk_get_sys
 3) SZ_4K

The 1) is an ARM specific implementation, so we need to depend on ARM.
For 2), clk_get_sys() is only available with CLKDEV_LOOKUP.  I can
understand that Arnd's change can help fix it, as COMMON_CLK indirectly
selects CLKDEV_LOOKUP.  But I'm not sure how CLKSRC_OF selecting is
going to help.  From compile test point of view, CLKSRC_OF is not
mandatory, since there is a stub clocksource_of_init() in
include/linux/clocksource.h for !CONFIG_CLKSRC_OF.  For 3), we need
the header include/linux/sizes.h.  It doesn't expose on ARM, and should
be fixed by a separate patch anyway.

[1] https://lkml.org/lkml/2015/6/1/328

So here is my changes to fix all the errors reported by Stephen, except
'SZ_4K' one.  Let me know if it's preferred or we'd like to drop
COMPILE_TEST and simply depend on ARCH_MXC.

------8<--------------

Comments

Fabio Estevam June 2, 2015, 1:46 p.m. UTC | #1

On Tue, Jun 2, 2015 at 12:33 AM, Shawn Guo <[e-mail]> wrote:

> So here is my changes to fix all the errors reported by Stephen, except
> 'SZ_4K' one.  Let me know if it's preferred or we'd like to drop
> COMPILE_TEST and simply depend on ARCH_MXC.

I would say that we could drop COMPILE_TEST and use 'depends on
ARCH_MXC' instead.

Regards,

Fabio Estevam

Arnd Bergmann June 2, 2015, 3:57 p.m. UTC | #2

On Tuesday 02 June 2015, Fabio Estevam wrote:
> Spam Status: Spamassassin 0% probability of being spam.
> 
> Full report:
> No, score=-2.7 required=5.0 tests=BAYES_00,DKIM_SIGNED, DKIM_VALID,DKIM_VALID_AU,FREEMAIL_FROM,RCVD_IN_DNSWL_LOW,UNPARSEABLE_RELAY autolearn=unavailable version=3.3.1  
> On Tue, Jun 2, 2015 at 12:33 AM, Shawn Guo <[e-mail]> wrote:
> 
> > So here is my changes to fix all the errors reported by Stephen, except
> > 'SZ_4K' one.  Let me know if it's preferred or we'd like to drop
> > COMPILE_TEST and simply depend on ARCH_MXC.
> 
> I would say that we could drop COMPILE_TEST and use 'depends on
> ARCH_MXC' instead.

Including COMPILE_TEST is generally a good idea, because it enables automatic
checking by build bots, e.g. the coverity bot that runs x86-64 allmodconfig.

	Arnd

Shawn Guo June 3, 2015, 6:23 a.m. UTC | #3

On Tue, Jun 02, 2015 at 05:57:53PM +0200, Arnd Bergmann wrote:
> On Tuesday 02 June 2015, Fabio Estevam wrote:
> > Spam Status: Spamassassin 0% probability of being spam.
> > 
> > Full report:
> > No, score=-2.7 required=5.0 tests=BAYES_00,DKIM_SIGNED, DKIM_VALID,DKIM_VALID_AU,FREEMAIL_FROM,RCVD_IN_DNSWL_LOW,UNPARSEABLE_RELAY autolearn=unavailable version=3.3.1  
> > On Tue, Jun 2, 2015 at 12:33 AM, Shawn Guo <[e-mail]> wrote:
> > 
> > > So here is my changes to fix all the errors reported by Stephen, except
> > > 'SZ_4K' one.  Let me know if it's preferred or we'd like to drop
> > > COMPILE_TEST and simply depend on ARCH_MXC.
> > 
> > I would say that we could drop COMPILE_TEST and use 'depends on
> > ARCH_MXC' instead.
> 
> Including COMPILE_TEST is generally a good idea, because it enables automatic
> checking by build bots, e.g. the coverity bot that runs x86-64 allmodconfig.

For this particular case, it will not be covered by x86-64 test, as we
already need to depend on ARM due to delay_timer support.  But I agree
that having COMPILE_TEST is good in general, so will keep it.

Shawn

Patch

diff --git a/drivers/clocksource/Kconfig b/drivers/clocksource/Kconfig
index 139a20795073..618102e5aa2a 100644
--- a/drivers/clocksource/Kconfig
+++ b/drivers/clocksource/Kconfig
@@ -261,7 +261,7 @@  config CLKSRC_PXA
 
 config CLKSRC_IMX_GPT
        bool "Clocksource using i.MX GPT" if COMPILE_TEST
-       depends on OF
+       depends on ARM && CLKDEV_LOOKUP
        select CLKSRC_MMIO
 
 endmenu