Skip to contentSkip to conversation

clk: Make clock framework user selectable

PATCH In review Mark Brown 3 participants 6 messages Aug 11, 2016 – Aug 16, 2016 Common clk

lore
Share & tools
lore

Mark Brown

author
#0 lore raw

PATCH 1/1 clk: Make clock framework user selectable

Some architectures like x86 don't use the clock API in architecture code but one of the advantages of the common clock API is that it can be used by regular devices to represent their clocking structures so we should allow users to enable the clock API themselves even if the architecture doesn't need it. This is more straightforward than going round every architecture and convincing their maintainers to add the select.

Unfortunately some architectures do still provide a custom implementation of the clock API which is incompatible with common clock. Provide a config option HAS_CUSTOM_CLK which these architectures can select in order to prevent COMMON_CLK being enabled.

---
 drivers/clk/Kconfig | 6 +++++-
 1 file changed, 5 insertions(+), 1 deletion(-)
drivers/clk/Kconfig +5 -1
index e2d9bd760c84..08a0b2cd49c6 100644
@@ -7,7 +7,8 @@ config HAVE_CLK_PREPARE
 	bool
 
 config COMMON_CLK
-	bool
+	bool "Common Clock Framework Support"
+	depends on !HAVE_CUSTOM_CLK
 	select HAVE_CLK_PREPARE
 	select CLKDEV_LOOKUP
 	select SRCU
@@ -19,6 +20,9 @@ config COMMON_CLK
 	  Architectures utilizing the common struct clk should select
 	  this option.
 
+config HAVE_CUSTOM_CLK
+	bool
+
 menu "Common Clock Framework"
 	depends on COMMON_CLK
 
Signature · 1 line
2.8.1
#1 lore raw
Quoted 11 lines · Mark Brown

Some architectures like x86 don't use the clock API in architecture code but one of the advantages of the common clock API is that it can be used by regular devices to represent their clocking structures so we should allow users to enable the clock API themselves even if the architecture doesn't need it. This is more straightforward than going round every architecture and convincing their maintainers to add the select.

Unfortunately some architectures do still provide a custom implementation of the clock API which is incompatible with common clock. Provide a config option HAS_CUSTOM_CLK which these architectures can select in order to prevent COMMON_CLK being enabled.

The list is not that long anymore, so it looks fixable with not too much effort:

$ git grep '\*clk_get\>'
arch/avr32/mach-at32ap/clock.c:struct clk *clk_get(struct device *dev,
const char *id)
arch/m68k/coldfire/clk.c:struct clk *clk_get(struct device *dev, const char *id)
arch/mips/ar7/clock.c:struct clk *clk_get(struct device *dev, const char *id)
arch/mips/bcm63xx/clk.c:struct clk *clk_get(struct device *dev, const char *id)
arch/mips/loongson64/lemote-2f/clock.c:struct clk *clk_get(struct
device *dev, const char *id)
arch/mips/txx9/generic/setup.c:struct clk *clk_get(struct device *dev,
const char *id)
arch/unicore32/kernel/clock.c:struct clk *clk_get(struct device *dev,
const char *id)
drivers/clk/clkdev.c:struct clk *clk_get(struct device *dev, const char *con_id)
Quoted 13 lines
drivers/clk/Kconfig +2 -1
@@ -7,7 +7,8 @@ config HAVE_CLK_PREPARE
        bool
 
 config COMMON_CLK
-       bool
+       bool "Common Clock Framework Support"
+       depends on !HAVE_CUSTOM_CLK
        select HAVE_CLK_PREPARE
        select CLKDEV_LOOKUP
        select SRCU

drivers/clk/Kconfig config COMMON_CLK · line 23

Quoted 4 lines
drivers/clk/Kconfig +1 -0
@@ -19,6 +20,9 @@ config COMMON_CLK
          Architectures utilizing the common struct clk should select
          this option.
 
+config HAVE_CUSTOM_CLK

HAVE_LEGACY_CLK, to give a better clue?

drivers/clk/Kconfig config COMMON_CLK · line 27

Quoted 4 lines
drivers/clk/Kconfig +2 -0
+       bool
+
 menu "Common Clock Framework"
        depends on COMMON_CLK

Gr{oetje,eeting}s,

Geert

--
Geert Uytterhoeven -- There's lots of Linux beyond ia32 -- [email protected]

In personal conversations with technical people, I call myself a hacker. But when I'm talking to journalists I just say "programmer" or something like that.
-- Linus Torvalds

#2 lore raw
Quoted 16 lines · Geert Uytterhoeven
Quoted 11 lines · Mark Brown

Some architectures like x86 don't use the clock API in architecture code but one of the advantages of the common clock API is that it can be used by regular devices to represent their clocking structures so we should allow users to enable the clock API themselves even if the architecture doesn't need it. This is more straightforward than going round every architecture and convincing their maintainers to add the select.

Unfortunately some architectures do still provide a custom implementation of the clock API which is incompatible with common clock. Provide a config option HAS_CUSTOM_CLK which these architectures can select in order to prevent COMMON_CLK being enabled.

The list is not that long anymore, so it looks fixable with not too much effort:

Quoted 13 lines
$ git grep '\*clk_get\>'
arch/avr32/mach-at32ap/clock.c:struct clk *clk_get(struct device *dev,
const char *id)
arch/m68k/coldfire/clk.c:struct clk *clk_get(struct device *dev, const char *id)
arch/mips/ar7/clock.c:struct clk *clk_get(struct device *dev, const char *id)
arch/mips/bcm63xx/clk.c:struct clk *clk_get(struct device *dev, const char *id)
arch/mips/loongson64/lemote-2f/clock.c:struct clk *clk_get(struct
device *dev, const char *id)
arch/mips/txx9/generic/setup.c:struct clk *clk_get(struct device *dev,
const char *id)
arch/unicore32/kernel/clock.c:struct clk *clk_get(struct device *dev,
const char *id)
drivers/clk/clkdev.c:struct clk *clk_get(struct device *dev, const char *con_id)

Having your own clk_get() doesn't seem to be sufficient. Found a few more:

$ git grep 'long clk_get_rate\>'arch/arm/mach-davinci/clock.c:unsigned
long clk_get_rate(struct clk *clk)
arch/arm/mach-ep93xx/clock.c:unsigned long clk_get_rate(struct clk *clk)
arch/arm/mach-mmp/clock.c:unsigned long clk_get_rate(struct clk *clk)
arch/arm/mach-omap1/clock.c:unsigned long clk_get_rate(struct clk *clk)
arch/arm/mach-sa1100/clock.c:unsigned long clk_get_rate(struct clk *clk)
arch/arm/mach-w90x900/clock.c:unsigned long clk_get_rate(struct clk *clk)
arch/avr32/mach-at32ap/clock.c:unsigned long clk_get_rate(struct clk *clk)
arch/blackfin/mach-bf609/clock.c:unsigned long clk_get_rate(struct clk *clk)
arch/c6x/platforms/pll.c:unsigned long clk_get_rate(struct clk *clk)
arch/m68k/coldfire/clk.c:unsigned long clk_get_rate(struct clk *clk)
arch/mips/ar7/clock.c:unsigned long clk_get_rate(struct clk *clk)
arch/mips/bcm63xx/clk.c:unsigned long clk_get_rate(struct clk *clk)
arch/mips/lantiq/clk.c:unsigned long clk_get_rate(struct clk *clk)
arch/mips/loongson64/lemote-2f/clock.c:unsigned long clk_get_rate(struct clk *cl
arch/mips/ralink/clk.c:unsigned long clk_get_rate(struct clk *clk)
arch/mips/txx9/generic/setup.c:unsigned long clk_get_rate(struct clk *clk)
arch/unicore32/kernel/clock.c:unsigned long clk_get_rate(struct clk *clk)
drivers/clk/clk.c:unsigned long clk_get_rate(struct clk *clk)
drivers/sh/clk/core.c:unsigned long clk_get_rate(struct clk *clk)

Gr{oetje,eeting}s,

Geert

--
Geert Uytterhoeven -- There's lots of Linux beyond ia32 -- [email protected]

In personal conversations with technical people, I call myself a hacker. But when I'm talking to journalists I just say "programmer" or something like that.
-- Linus Torvalds

author
#3 lore raw
In reply to Geert Uytterhoeven
Quoted 9 lines · Geert Uytterhoeven
Quoted 4 lines · Mark Brown

Unfortunately some architectures do still provide a custom implementation of the clock API which is incompatible with common clock. Provide a config option HAS_CUSTOM_CLK which these architectures can select in order to prevent COMMON_CLK being enabled.

The list is not that long anymore, so it looks fixable with not too much effort:

Yeah, it's probably a good project for someone. TBH my main goal here was to stop x86 blocking use of the clock API.

#4 lore raw
In reply to Mark Brown
Quoted 10 lines · Mark Brown

Some architectures like x86 don't use the clock API in architecture code but one of the advantages of the common clock API is that it can be used by regular devices to represent their clocking structures so we should allow users to enable the clock API themselves even if the architecture doesn't need it. This is more straightforward than going round every architecture and convincing their maintainers to add the select.

Unfortunately some architectures do still provide a custom implementation of the clock API which is incompatible with common clock. Provide a config option HAS_CUSTOM_CLK which these architectures can select in order to

s/HAS/HAVE/

Quoted 3 lines

prevent COMMON_CLK being enabled.

Concept seems fine.

Quoted 9 lines
---
 drivers/clk/Kconfig | 6 +++++-
 1 file changed, 5 insertions(+), 1 deletion(-)
drivers/clk/Kconfig +0 -0
index e2d9bd760c84..08a0b2cd49c6 100644

drivers/clk/Kconfig config HAVE_CLK_PREPARE · line 11

Quoted 6 lines
drivers/clk/Kconfig +2 -1
@@ -7,7 +7,8 @@ config HAVE_CLK_PREPARE
 	bool
 
 config COMMON_CLK
-	bool
+	bool "Common Clock Framework Support"
+	depends on !HAVE_CUSTOM_CLK

Is the plan that architectures will keep selecting this when they want to enforce common clk framework on their arch? We could move enabling this config to the defconfigs if we make of_clk_init() a stub when this config is disabled. Interestingly arm already has that ifdefed out so nothing fails to compile there, but not every arch is following that design.

Also, are there patches for all the users to select HAVE_CUSTOM_CLK? Otherwise we can get randconfig failures when two places define clk API functions.

Signature · 2 lines
Qualcomm Innovation Center, Inc. is a member of Code Aurora Forum,
a Linux Foundation Collaborative Project
author
#5 lore raw
Quoted 9 lines · Stephen Boyd

drivers/clk/Kconfig config HAVE_CLK_PREPARE · line 11

Quoted 4 lines · Mark Brown

config COMMON_CLK

drivers/clk/Kconfig +2 -1
-	bool
+	bool "Common Clock Framework Support"
+	depends on !HAVE_CUSTOM_CLK

Is the plan that architectures will keep selecting this when they want to enforce common clk framework on their arch? We could move

I think that's a per-architecture thing - if the architecture doesn't need control of clocks then there's no reason for it to force the API on, but if the architecture is unlikely to do anything useful without then it's a different story.

Quoted 4 lines

enabling this config to the defconfigs if we make of_clk_init() a stub when this config is disabled. Interestingly arm already has that ifdefed out so nothing fails to compile there, but not every arch is following that design.

That's probably a good idea anyway.

Quoted 3 lines

Also, are there patches for all the users to select HAVE_CUSTOM_CLK? Otherwise we can get randconfig failures when two places define clk API functions.

They're trivial to make if you apply this.