Re: [PATCH v3 04/11] ARM: Exynos4: Migrate clock support to common clock framework

[Tomasz Figa <t.figa@samsung.com>]

On Thursday 15 of November 2012 14:43:21 Thomas Abraham wrote:
> On 15 November 2012 05:01, Tomasz Figa <tomasz.figa@gmail.com> wrote:
> > On Thursday 15 of November 2012 03:37:26 Thomas Abraham wrote:
> >> Remove Samsung specific clock support in Exynos4 and migrate to use
> >> common clock framework.
> >> 
> >> Cc: Kukjin Kim <kgene.kim@samsung.com>
> >> Signed-off-by: Thomas Abraham <thomas.abraham@linaro.org>
> >> ---
> >> 
> >>  arch/arm/mach-exynos/Kconfig               |    1 +
> >>  arch/arm/mach-exynos/Makefile              |    3 -
> >>  arch/arm/mach-exynos/clock-exynos4.c       | 1602
> >> 
> >> ---------------------------- arch/arm/mach-exynos/clock-exynos4.h
> >> 
> >> |   35 -
> >>  
> >>  arch/arm/mach-exynos/clock-exynos4210.c    |  188 ----
> >>  arch/arm/mach-exynos/clock-exynos4212.c    |  192 ----
> >>  arch/arm/mach-exynos/common.c              |   22 +-
> >>  arch/arm/mach-exynos/common.h              |    3 +
> >>  arch/arm/mach-exynos/mach-armlex4210.c     |    1 -
> >>  arch/arm/mach-exynos/mach-exynos4-dt.c     |    1 -
> >>  arch/arm/mach-exynos/mach-nuri.c           |    1 -
> >>  arch/arm/mach-exynos/mach-origen.c         |    1 -
> >>  arch/arm/mach-exynos/mach-smdk4x12.c       |    1 -
> >>  arch/arm/mach-exynos/mach-smdkv310.c       |    1 -
> >>  arch/arm/mach-exynos/mach-universal_c210.c |    1 -
> >>  arch/arm/mach-exynos/mct.c                 |   19 +
> >>  arch/arm/plat-samsung/Kconfig              |    4 +-
> >>  17 files changed, 27 insertions(+), 2049 deletions(-)
> >>  delete mode 100644 arch/arm/mach-exynos/clock-exynos4.c
> >>  delete mode 100644 arch/arm/mach-exynos/clock-exynos4.h
> >>  delete mode 100644 arch/arm/mach-exynos/clock-exynos4210.c
> >>  delete mode 100644 arch/arm/mach-exynos/clock-exynos4212.c
> > 
> > [snip]
> > 
> >> diff --git a/arch/arm/mach-exynos/mct.c b/arch/arm/mach-exynos/mct.c
> >> index f7792b8..c2e806c 100644
> >> --- a/arch/arm/mach-exynos/mct.c
> >> +++ b/arch/arm/mach-exynos/mct.c
> >> @@ -31,6 +31,7 @@
> >> 
> >>  #include <mach/map.h>
> >>  #include <mach/irqs.h>
> >>  #include <asm/mach/time.h>
> >> 
> >> +#include "common.h"
> >> 
> >>  #define EXYNOS4_MCTREG(x)            (x)
> >>  #define EXYNOS4_MCT_G_CNT_L          EXYNOS4_MCTREG(0x100)
> >> 
> >> @@ -517,6 +518,24 @@ static void __init exynos4_timer_init(void)
> >> 
> >>       struct device_node *np;
> >>       u32 nr_irqs, i;
> >> 
> >> +#ifdef CONFIG_COMMON_CLK
> >> +     /*
> >> +      * Clock lookup should be functional now since the MCT
> >> controller
> >> driver +       * looks up clocks. So the clock initialization is
> >> initiated here. +        */
> >> +     if (of_have_populated_dt()) {
> >> +             if (of_machine_is_compatible("samsung,exynos4210") ||
> >> +                      of_machine_is_compatible("samsung,exynos4212")
> >> || +                     
> >> of_machine_is_compatible("samsung,exynos4412")) +                   
> >>  exynos4_clk_init();
> >> +     } else {
> >> +             if (soc_is_exynos4210() || soc_is_exynos4212() ||
> >> +                             soc_is_exynos4412()) {
> >> +                     exynos4_clk_init();
> >> +             }
> >> +     }
> >> +#endif
> >> +
> > 
> > I don't like the idea of initializing the clocks from timer
> > initialization. What about some platforms where MCT isn't used? It is
> > also far from being elegant.
> 
> Very true, I did also prefer not do this. But, clock lookup should be
> functional atleast by the time mct initialization begins. So I tried
> few options such as adding .early_init_call callback in MACHINE_DESC
> which then can call exynos4_clk_init, but that did not help since
> mem_init isn't complete by then and memory allocation failed during
> clock registration. Other methods also did not help much. If you know
> of a solution to get around this, could you please let me know.

This is somehow similar to the problem with mapping SYSRAM_NS for Exynos 
secure firmware. Currently it must be mapped statically using io_table, 
because ioremap does not work early enough. I guess this kind of approach 
might be also enough for clocks, until io mem mapping becomes accessible 
early enough, but someone else should also comment this.

Best regards,
-- 
Tomasz Figa
Samsung Poland R&D Center
SW Solution Development, Linux Platform