[PATCH v2 0/2] Hi3660: enable sp804 timer

[PATCH v2 0/2] Hi3660: enable sp804 timer

[Leo Yan]

This patch set is to enable sp804 timer on Hi3660 for Hikey960 platform.

On Hi3660, the sp804 timer co-exists with CPUs' architecture timer; but
sp804 timer is located in SoC level but CPU's architecture timer is in
CPU power domain. sp804 timer is used as broadcast timer when CPU enters
idle states and the CPU (includes architecture timer) power domain is
powered off. So sp804 timer enabling is prerequisite before we enable
CPUIdle on Hi3660.

This patch set is to enable sp804 timer, the first patch is to adjust
crgctrl clock regiseration so ensure clock is avaiable for timer. The
second patch is to add description for timer in dts. Have rebased this
patch set on Guodong's dts patch set [1] and verified this patch set on
Hikey960.

[1] http://archive.armlinux.org.uk/lurker/message/20170517.083733.8207a50e.en.html

Changes from v1:
* Follow Steven's suggestion on IRC, Change to only use
  CLK_OF_DECLARE_DRIVER for crgctrl clock.

Leo Yan (2):
  clk: Hi3660: change to register crgctrl clock with
    CLK_OF_DECLARE_DRIVER
  arm64: dts: add sp804 timer node for Hi3660

 arch/arm64/boot/dts/hisilicon/hi3660.dtsi | 11 +++++++++++
 drivers/clk/hisilicon/clk-hi3660.c        |  4 ++--
 2 files changed, 13 insertions(+), 2 deletions(-)

-- 
1.9.1

[PATCH v2 1/2] clk: Hi3660: change to register crgctrl clock with CLK_OF_DECLARE_DRIVER

[Leo Yan]

The timer will register into system at very early phase at kernel boot;
if timer needs to use clock, the clock should be get ready in function
of_clk_init() so later the timer driver probe can retrieve clock
successfully. This is finished in below flow on arm64:

  start_kernel()
    `-> time_init()
          `-> of_clk_init(NULL)    => register timer's clock
          `-> clocksource_probe()  => register timer

On Hi3660 the sp804 timer uses clock "osc32k", this clock is registered
as platform driver rather than CLK_OF_DECLARE_DRIVER method. As result,
sp804 timer probe returns failure due if cannot bind clock properly.
To fix the failure, this patch is to change crgctrl clock registration
from platform driver to CLK_OF_DECLARE_DRIVER method so the clocks can
be registered ahead with function of_clk_init() and then timer driver
can use it.

Signed-off-by: Leo Yan <leo.yan@linaro.org>
---
 drivers/clk/hisilicon/clk-hi3660.c | 4 ++--
 1 file changed, 2 insertions(+), 2 deletions(-)

diff --git a/drivers/clk/hisilicon/clk-hi3660.c b/drivers/clk/hisilicon/clk-hi3660.c
index fd5ce7f..4e5713b 100644
--- a/drivers/clk/hisilicon/clk-hi3660.c
+++ b/drivers/clk/hisilicon/clk-hi3660.c
@@ -547,10 +547,10 @@ static void hi3660_clk_crgctrl_init(struct device_node *np)
 				  ARRAY_SIZE(hi3660_crgctrl_divider_clks),
 				  clk_data);
 }
+CLK_OF_DECLARE_DRIVER(hi3660_clk_crgctrl, "hisilicon,hi3660-crgctrl",
+		      hi3660_clk_crgctrl_init);
 
 static const struct of_device_id hi3660_clk_match_table[] = {
-	{ .compatible = "hisilicon,hi3660-crgctrl",
-	  .data = hi3660_clk_crgctrl_init },
 	{ .compatible = "hisilicon,hi3660-pctrl",
 	  .data = hi3660_clk_pctrl_init },
 	{ .compatible = "hisilicon,hi3660-pmuctrl",
-- 
1.9.1

[PATCH v2 2/2] arm64: dts: add sp804 timer node for Hi3660

[Leo Yan]

The Hi3660 SoC comes with the sp804 timer in addition to the
architecture timers. These ones are shutdown when reaching a deep idle
states and a backup timer is needed. The sp804 belongs to another power
domain and can fulfill the purpose of replacing temporarily an
architecture timer when the CPU is idle.

Describe it in the device tree, so it can be enabled at boot time.

Suggested-by: Daniel Lezcano <daniel.lezcano@linaro.org>
Acked-by: Daniel Lezcano <daniel.lezcano@linaro.org>
Signed-off-by: Leo Yan <leo.yan@linaro.org>
---
 arch/arm64/boot/dts/hisilicon/hi3660.dtsi | 11 +++++++++++
 1 file changed, 11 insertions(+)

diff --git a/arch/arm64/boot/dts/hisilicon/hi3660.dtsi b/arch/arm64/boot/dts/hisilicon/hi3660.dtsi
index 138fcba..f75c792 100644
--- a/arch/arm64/boot/dts/hisilicon/hi3660.dtsi
+++ b/arch/arm64/boot/dts/hisilicon/hi3660.dtsi
@@ -173,6 +173,17 @@
 			#clock-cells = <1>;
 		};
 
+		dual_timer0: timer@fff14000 {
+			compatible = "arm,sp804", "arm,primecell";
+			reg = <0x0 0xfff14000 0x0 0x1000>;
+			interrupts = <GIC_SPI 48 IRQ_TYPE_LEVEL_HIGH>,
+				     <GIC_SPI 49 IRQ_TYPE_LEVEL_HIGH>;
+			clocks = <&crg_ctrl HI3660_OSC32K>,
+				 <&crg_ctrl HI3660_OSC32K>,
+				 <&crg_ctrl HI3660_OSC32K>;
+			clock-names = "timer1", "timer2", "apb_pclk";
+		};
+
 		ufs: ufs@ff3b0000 {
 			compatible = "jedec,ufs-1.1", "hisilicon,hi3660-ufs";
 			reg = <0x0 0xff3b0000 0x0 0x1000>, /* 0: HCI standard */
-- 
1.9.1

Re: [PATCH v2 1/2] clk: Hi3660: change to register crgctrl clock with CLK_OF_DECLARE_DRIVER

[Daniel Lezcano]

On Thu, May 18, 2017 at 09:50:46AM +0800, Leo Yan wrote:
> The timer will register into system at very early phase at kernel boot;
> if timer needs to use clock, the clock should be get ready in function
> of_clk_init() so later the timer driver probe can retrieve clock
> successfully. This is finished in below flow on arm64:
> 
>   start_kernel()
>     `-> time_init()
>           `-> of_clk_init(NULL)    => register timer's clock
>           `-> clocksource_probe()  => register timer
> 
> On Hi3660 the sp804 timer uses clock "osc32k", this clock is registered
> as platform driver rather than CLK_OF_DECLARE_DRIVER method. As result,
> sp804 timer probe returns failure due if cannot bind clock properly.
> To fix the failure, this patch is to change crgctrl clock registration
> from platform driver to CLK_OF_DECLARE_DRIVER method so the clocks can
> be registered ahead with function of_clk_init() and then timer driver
> can use it.
> 
> Signed-off-by: Leo Yan <leo.yan@linaro.org>
> ---
>  drivers/clk/hisilicon/clk-hi3660.c | 4 ++--
>  1 file changed, 2 insertions(+), 2 deletions(-)
> 
> diff --git a/drivers/clk/hisilicon/clk-hi3660.c b/drivers/clk/hisilicon/clk-hi3660.c
> index fd5ce7f..4e5713b 100644
> --- a/drivers/clk/hisilicon/clk-hi3660.c
> +++ b/drivers/clk/hisilicon/clk-hi3660.c
> @@ -547,10 +547,10 @@ static void hi3660_clk_crgctrl_init(struct device_node *np)
>  				  ARRAY_SIZE(hi3660_crgctrl_divider_clks),
>  				  clk_data);
>  }
> +CLK_OF_DECLARE_DRIVER(hi3660_clk_crgctrl, "hisilicon,hi3660-crgctrl",
> +		      hi3660_clk_crgctrl_init);
>  
>  static const struct of_device_id hi3660_clk_match_table[] = {
> -	{ .compatible = "hisilicon,hi3660-crgctrl",
> -	  .data = hi3660_clk_crgctrl_init },
>  	{ .compatible = "hisilicon,hi3660-pctrl",
>  	  .data = hi3660_clk_pctrl_init },
>  	{ .compatible = "hisilicon,hi3660-pmuctrl",

Why not do the change for all the clocks and remove the
hi3660_clk_match_table[]? and all the related platform driver code?

That will be consistent with the other clk-hi* drivers.

-- 

 <http://www.linaro.org/> Linaro.org │ Open source software for ARM SoCs

Follow Linaro:  <http://www.facebook.com/pages/Linaro> Facebook |
<http://twitter.com/#!/linaroorg> Twitter |
<http://www.linaro.org/linaro-blog/> Blog

Re: [PATCH v2 1/2] clk: Hi3660: change to register crgctrl clock with CLK_OF_DECLARE_DRIVER

[Leo Yan]

Hi Daniel,

On Thu, May 18, 2017 at 11:32:26AM +0200, Daniel Lezcano wrote:
> On Thu, May 18, 2017 at 09:50:46AM +0800, Leo Yan wrote:
> > The timer will register into system at very early phase at kernel boot;
> > if timer needs to use clock, the clock should be get ready in function
> > of_clk_init() so later the timer driver probe can retrieve clock
> > successfully. This is finished in below flow on arm64:
> > 
> >   start_kernel()
> >     `-> time_init()
> >           `-> of_clk_init(NULL)    => register timer's clock
> >           `-> clocksource_probe()  => register timer
> > 
> > On Hi3660 the sp804 timer uses clock "osc32k", this clock is registered
> > as platform driver rather than CLK_OF_DECLARE_DRIVER method. As result,
> > sp804 timer probe returns failure due if cannot bind clock properly.
> > To fix the failure, this patch is to change crgctrl clock registration
> > from platform driver to CLK_OF_DECLARE_DRIVER method so the clocks can
> > be registered ahead with function of_clk_init() and then timer driver
> > can use it.
> > 
> > Signed-off-by: Leo Yan <leo.yan@linaro.org>
> > ---
> >  drivers/clk/hisilicon/clk-hi3660.c | 4 ++--
> >  1 file changed, 2 insertions(+), 2 deletions(-)
> > 
> > diff --git a/drivers/clk/hisilicon/clk-hi3660.c b/drivers/clk/hisilicon/clk-hi3660.c
> > index fd5ce7f..4e5713b 100644
> > --- a/drivers/clk/hisilicon/clk-hi3660.c
> > +++ b/drivers/clk/hisilicon/clk-hi3660.c
> > @@ -547,10 +547,10 @@ static void hi3660_clk_crgctrl_init(struct device_node *np)
> >  				  ARRAY_SIZE(hi3660_crgctrl_divider_clks),
> >  				  clk_data);
> >  }
> > +CLK_OF_DECLARE_DRIVER(hi3660_clk_crgctrl, "hisilicon,hi3660-crgctrl",
> > +		      hi3660_clk_crgctrl_init);
> >  
> >  static const struct of_device_id hi3660_clk_match_table[] = {
> > -	{ .compatible = "hisilicon,hi3660-crgctrl",
> > -	  .data = hi3660_clk_crgctrl_init },
> >  	{ .compatible = "hisilicon,hi3660-pctrl",
> >  	  .data = hi3660_clk_pctrl_init },
> >  	{ .compatible = "hisilicon,hi3660-pmuctrl",
> 
> Why not do the change for all the clocks and remove the
> hi3660_clk_match_table[]? and all the related platform driver code?
> 
> That will be consistent with the other clk-hi* drivers.

I checked a bit with Stephen on IRC, follow the suggestion to 'split
the clks are registsered between CLK_OF_DECLARE_DRIVER() and driver
probe so that the ones you need really early are done through the
CLK_OF path and the other ones are done through regular driver probe'.

Also checked for mainline kernel, now there have some examples,
like: drivers/clk/mediatek/clk-mt2701.c.

Thanks,
Leo Yan

Re: [PATCH v2 1/2] clk: Hi3660: change to register crgctrl clock with CLK_OF_DECLARE_DRIVER

[Stephen Boyd]

On 05/18, Leo Yan wrote:
> diff --git a/drivers/clk/hisilicon/clk-hi3660.c b/drivers/clk/hisilicon/clk-hi3660.c
> index fd5ce7f..4e5713b 100644
> --- a/drivers/clk/hisilicon/clk-hi3660.c
> +++ b/drivers/clk/hisilicon/clk-hi3660.c
> @@ -547,10 +547,10 @@ static void hi3660_clk_crgctrl_init(struct device_node *np)
>  				  ARRAY_SIZE(hi3660_crgctrl_divider_clks),
>  				  clk_data);
>  }
> +CLK_OF_DECLARE_DRIVER(hi3660_clk_crgctrl, "hisilicon,hi3660-crgctrl",
> +		      hi3660_clk_crgctrl_init);
>  
>  static const struct of_device_id hi3660_clk_match_table[] = {
> -	{ .compatible = "hisilicon,hi3660-crgctrl",
> -	  .data = hi3660_clk_crgctrl_init },

I'm slightly confused. CLK_OF_DECLARE_DRIVER is for when a
platform driver _and_ CLK_OF_DELARE need to "probe" the same
device node. But in this patch, the compatible string is being
deleted from the driver's match table. Also, are all the
registered in hi3660_clk_crgctrl_init() actually needed for the
timer? If only a subset, then I would expect to split the code
and have the CLK_OF_DECLARE_DRIVER() path register just the clks
that matter for the timer, and then leave the rest of the clks to
be registered by the probe of the platform driver.

-- 
Qualcomm Innovation Center, Inc. is a member of Code Aurora Forum,
a Linux Foundation Collaborative Project

Re: [PATCH v2 1/2] clk: Hi3660: change to register crgctrl clock with CLK_OF_DECLARE_DRIVER

[Leo Yan]

On Thu, May 18, 2017 at 06:51:37PM -0700, Stephen Boyd wrote:
> On 05/18, Leo Yan wrote:
> > diff --git a/drivers/clk/hisilicon/clk-hi3660.c b/drivers/clk/hisilicon/clk-hi3660.c
> > index fd5ce7f..4e5713b 100644
> > --- a/drivers/clk/hisilicon/clk-hi3660.c
> > +++ b/drivers/clk/hisilicon/clk-hi3660.c
> > @@ -547,10 +547,10 @@ static void hi3660_clk_crgctrl_init(struct device_node *np)
> >  				  ARRAY_SIZE(hi3660_crgctrl_divider_clks),
> >  				  clk_data);
> >  }
> > +CLK_OF_DECLARE_DRIVER(hi3660_clk_crgctrl, "hisilicon,hi3660-crgctrl",
> > +		      hi3660_clk_crgctrl_init);
> >  
> >  static const struct of_device_id hi3660_clk_match_table[] = {
> > -	{ .compatible = "hisilicon,hi3660-crgctrl",
> > -	  .data = hi3660_clk_crgctrl_init },
> 
> I'm slightly confused. CLK_OF_DECLARE_DRIVER is for when a
> platform driver _and_ CLK_OF_DELARE need to "probe" the same
> device node. But in this patch, the compatible string is being
> deleted from the driver's match table. Also, are all the
> registered in hi3660_clk_crgctrl_init() actually needed for the
> timer?

No.

> If only a subset, then I would expect to split the code
> and have the CLK_OF_DECLARE_DRIVER() path register just the clks
> that matter for the timer, and then leave the rest of the clks to
> be registered by the probe of the platform driver.

I'm also puzzled if need divide crgctrl clock into two parts. Thanks
for suggestion, will do this.

Thanks,
Leo Yan