Re: [PATCH 04/15] MIPS: Add sched_clock support

Re: [PATCH 04/15] MIPS: Add sched_clock support

[Maciej W. Rozycki]

On Wed, 4 Mar 2015, Deng-Cheng Zhu wrote:

> This will provide sched_clock interface to implement individual
> read_sched_clock(). Not for CAVIUM_OCTEON_SOC as it defines its own
> sched_clock() directly (not using the sched_clock_register interface).
> 
> Signed-off-by: Deng-Cheng Zhu <dengcheng.zhu@imgtec.com>
> ---
>  arch/mips/Kconfig | 1 +
>  1 file changed, 1 insertion(+)
> 
> diff --git a/arch/mips/Kconfig b/arch/mips/Kconfig
> index 068592a..09405dc 100644
> --- a/arch/mips/Kconfig
> +++ b/arch/mips/Kconfig
> @@ -43,6 +43,7 @@ config MIPS
>  	select GENERIC_SMP_IDLE_THREAD
>  	select BUILDTIME_EXTABLE_SORT
>  	select GENERIC_CLOCKEVENTS
> +	select GENERIC_SCHED_CLOCK if !CAVIUM_OCTEON_SOC
>  	select GENERIC_CMOS_UPDATE
>  	select HAVE_MOD_ARCH_SPECIFIC
>  	select VIRT_TO_BUS

 Why does this change add this question:

ARM Versatile (Express) reference platforms clock source (CLKSRC_VERSATILE) [N/y/?] (NEW) ?

This option enables clock source based on free running
counter available in the "System Registers" block of
ARM Versatile, RealView and Versatile Express reference
platforms.

Symbol: CLKSRC_VERSATILE [=n]
Type  : boolean
Prompt: ARM Versatile (Express) reference platforms clock source
  Location:
    -> Device Drivers
      -> Clock Source drivers
  Defined at drivers/clocksource/Kconfig:216
  Depends on: GENERIC_SCHED_CLOCK [=y] && !ARCH_USES_GETTIMEOFFSET [=n]
  Selects: CLKSRC_OF [=n]

to a MIPS configuration?  I find it silly, this appears a platform 
device to me (use reverse dependencies?).

  Maciej

Re: [PATCH 04/15] MIPS: Add sched_clock support

[Deng-Cheng Zhu]

On 03/06/2015 03:58 AM, Maciej W. Rozycki wrote:
> On Wed, 4 Mar 2015, Deng-Cheng Zhu wrote:
>
>> This will provide sched_clock interface to implement individual
>> read_sched_clock(). Not for CAVIUM_OCTEON_SOC as it defines its own
>> sched_clock() directly (not using the sched_clock_register interface).
>>
>> Signed-off-by: Deng-Cheng Zhu <dengcheng.zhu@imgtec.com>
>> ---
>>   arch/mips/Kconfig | 1 +
>>   1 file changed, 1 insertion(+)
>>
>> diff --git a/arch/mips/Kconfig b/arch/mips/Kconfig
>> index 068592a..09405dc 100644
>> --- a/arch/mips/Kconfig
>> +++ b/arch/mips/Kconfig
>> @@ -43,6 +43,7 @@ config MIPS
>>   	select GENERIC_SMP_IDLE_THREAD
>>   	select BUILDTIME_EXTABLE_SORT
>>   	select GENERIC_CLOCKEVENTS
>> +	select GENERIC_SCHED_CLOCK if !CAVIUM_OCTEON_SOC
>>   	select GENERIC_CMOS_UPDATE
>>   	select HAVE_MOD_ARCH_SPECIFIC
>>   	select VIRT_TO_BUS
>   Why does this change add this question:
>
> ARM Versatile (Express) reference platforms clock source (CLKSRC_VERSATILE) [N/y/?] (NEW) ?

Good catch.

>
> This option enables clock source based on free running
> counter available in the "System Registers" block of
> ARM Versatile, RealView and Versatile Express reference
> platforms.
>
> Symbol: CLKSRC_VERSATILE [=n]
> Type  : boolean
> Prompt: ARM Versatile (Express) reference platforms clock source
>    Location:
>      -> Device Drivers
>        -> Clock Source drivers
>    Defined at drivers/clocksource/Kconfig:216
>    Depends on: GENERIC_SCHED_CLOCK [=y] && !ARCH_USES_GETTIMEOFFSET [=n]
>    Selects: CLKSRC_OF [=n]
>
> to a MIPS configuration?  I find it silly, this appears a platform
> device to me (use reverse dependencies?).

GENERIC_SCHED_CLOCK is generic, by default it uses jiffy_sched_clock_read() 
for read_sched_clock(). Instead of using reverse dependencies, I suggest 
the following:

diff --git a/drivers/clocksource/Kconfig b/drivers/clocksource/Kconfig
index 1c2506f..22e0ee1 100644
--- a/drivers/clocksource/Kconfig
+++ b/drivers/clocksource/Kconfig
@@ -225,7 +225,7 @@ config CLKSRC_QCOM

  config CLKSRC_VERSATILE
         bool "ARM Versatile (Express) reference platforms clock source"
-       depends on GENERIC_SCHED_CLOCK && !ARCH_USES_GETTIMEOFFSET
+       depends on PLAT_VERSATILE && GENERIC_SCHED_CLOCK && 
!ARCH_USES_GETTIMEOFFSET
         select CLKSRC_OF
         default y if MFD_VEXPRESS_SYSREG
         help


Deng-Cheng