[PATCH 0/2] Fix up qcom reset controller

[PATCH 0/2] Fix up qcom reset controller

[Konrad Dybcio]

Let the toggle include a bigger delay and make sure it's using the
correct function to achieve that.

Signed-off-by: Konrad Dybcio <konrad.dybcio@linaro.org>
---
Konrad Dybcio (2):
      clk: qcom: reset: Increase max reset delay
      clk: qcom: reset: Use the correct type of sleep/delay based on length

 drivers/clk/qcom/reset.c | 8 +++++++-
 drivers/clk/qcom/reset.h | 2 +-
 2 files changed, 8 insertions(+), 2 deletions(-)
---
base-commit: 0ba5d07205771c50789fd9063950aa75e7f1183f
change-id: 20230726-topic-qcom_reset-44879c0387c6

Best regards,
-- 
Konrad Dybcio <konrad.dybcio@linaro.org>

[PATCH 1/2] clk: qcom: reset: Increase max reset delay

[Konrad Dybcio]

u8 limits us to 255 microseconds of delay. Promote the delay variable to
u16 to hold bigger values.

Signed-off-by: Konrad Dybcio <konrad.dybcio@linaro.org>
---
 drivers/clk/qcom/reset.h | 2 +-
 1 file changed, 1 insertion(+), 1 deletion(-)

diff --git a/drivers/clk/qcom/reset.h b/drivers/clk/qcom/reset.h
index 9a47c838d9b1..fe0561bf53d4 100644
--- a/drivers/clk/qcom/reset.h
+++ b/drivers/clk/qcom/reset.h
@@ -11,7 +11,7 @@
 struct qcom_reset_map {
 	unsigned int reg;
 	u8 bit;
-	u8 udelay;
+	u16 udelay;
 	u32 bitmask;
 };
 

-- 
2.41.0

[PATCH 2/2] clk: qcom: reset: Use the correct type of sleep/delay based on length

[Konrad Dybcio]

Based on the length of the delay (see: [1]), use the correct sleep/delay
functions.

[1] https://www.kernel.org/doc/Documentation/timers/timers-howto.txt
Fixes: b36ba30c8ac6 ("clk: qcom: Add reset controller support")
Signed-off-by: Konrad Dybcio <konrad.dybcio@linaro.org>
---
 drivers/clk/qcom/reset.c | 8 +++++++-
 1 file changed, 7 insertions(+), 1 deletion(-)

diff --git a/drivers/clk/qcom/reset.c b/drivers/clk/qcom/reset.c
index 0e914ec7aeae..928995c3f369 100644
--- a/drivers/clk/qcom/reset.c
+++ b/drivers/clk/qcom/reset.c
@@ -14,9 +14,15 @@
 static int qcom_reset(struct reset_controller_dev *rcdev, unsigned long id)
 {
 	struct qcom_reset_controller *rst = to_qcom_reset_controller(rcdev);
+	u16 delay_us = rst->reset_map[id].udelay ?: 1;
 
 	rcdev->ops->assert(rcdev, id);
-	udelay(rst->reset_map[id].udelay ?: 1); /* use 1 us as default */
+
+	if (delay_us < 10)
+		udelay(delay_us);
+	else
+		usleep_range(delay_us, delay_us + 5);
+
 	rcdev->ops->deassert(rcdev, id);
 	return 0;
 }

-- 
2.41.0

Re: [PATCH 1/2] clk: qcom: reset: Increase max reset delay

[Stephan Gerhold]

On Wed, Jul 26, 2023 at 03:26:19PM +0200, Konrad Dybcio wrote:
> u8 limits us to 255 microseconds of delay. Promote the delay variable to
> u16 to hold bigger values.
> 
> Signed-off-by: Konrad Dybcio <konrad.dybcio@linaro.org>

It would be clearer to change this together with an actual user that
needs > 255us. AFAICT atm MSM8909 is the only user of this and it has
just 15us.

Thanks,
Stephan

> ---
>  drivers/clk/qcom/reset.h | 2 +-
>  1 file changed, 1 insertion(+), 1 deletion(-)
> 
> diff --git a/drivers/clk/qcom/reset.h b/drivers/clk/qcom/reset.h
> index 9a47c838d9b1..fe0561bf53d4 100644
> --- a/drivers/clk/qcom/reset.h
> +++ b/drivers/clk/qcom/reset.h
> @@ -11,7 +11,7 @@
>  struct qcom_reset_map {
>  	unsigned int reg;
>  	u8 bit;
> -	u8 udelay;
> +	u16 udelay;
>  	u32 bitmask;
>  };
>  
> 
> -- 
> 2.41.0
> 

Re: [PATCH 1/2] clk: qcom: reset: Increase max reset delay

[Konrad Dybcio]

On 26.07.2023 20:21, Stephan Gerhold wrote:
> On Wed, Jul 26, 2023 at 03:26:19PM +0200, Konrad Dybcio wrote:
>> u8 limits us to 255 microseconds of delay. Promote the delay variable to
>> u16 to hold bigger values.
>>
>> Signed-off-by: Konrad Dybcio <konrad.dybcio@linaro.org>
> 
> It would be clearer to change this together with an actual user that
> needs > 255us. AFAICT atm MSM8909 is the only user of this and it has
> just 15us.
Some LPASS resets ask for 500, but I'm still working on that driver.

Konrad

Re: [PATCH 2/2] clk: qcom: reset: Use the correct type of sleep/delay based on length

[Stephan Gerhold]

On Wed, Jul 26, 2023 at 03:26:20PM +0200, Konrad Dybcio wrote:
> Based on the length of the delay (see: [1]), use the correct sleep/delay
> functions.
> 
> [1] https://www.kernel.org/doc/Documentation/timers/timers-howto.txt
> Fixes: b36ba30c8ac6 ("clk: qcom: Add reset controller support")

If anything this would fix 2cb8a39b6781 ("clk: qcom: reset: Allow
specifying custom reset delay") since before there was udelay(1)
hardcoded (which is correct).

But given that we just allowed small delays <= 255us so far it was
probably okay-ish, so I think it's not worth bringing this to the
attention of backporters with the Fixes tag.

> Signed-off-by: Konrad Dybcio <konrad.dybcio@linaro.org>
> ---
>  drivers/clk/qcom/reset.c | 8 +++++++-
>  1 file changed, 7 insertions(+), 1 deletion(-)
> 
> diff --git a/drivers/clk/qcom/reset.c b/drivers/clk/qcom/reset.c
> index 0e914ec7aeae..928995c3f369 100644
> --- a/drivers/clk/qcom/reset.c
> +++ b/drivers/clk/qcom/reset.c
> @@ -14,9 +14,15 @@
>  static int qcom_reset(struct reset_controller_dev *rcdev, unsigned long id)
>  {
>  	struct qcom_reset_controller *rst = to_qcom_reset_controller(rcdev);
> +	u16 delay_us = rst->reset_map[id].udelay ?: 1;
>  
>  	rcdev->ops->assert(rcdev, id);
> -	udelay(rst->reset_map[id].udelay ?: 1); /* use 1 us as default */
> +
> +	if (delay_us < 10)
> +		udelay(delay_us);
> +	else
> +		usleep_range(delay_us, delay_us + 5);
> +

There is fsleep() that implements this logic in a similar way already:

	fsleep(rst->reset_map[id].udelay ?: 1); /* use 1 us as default */

Thanks,
Stephan

Re: [PATCH 1/2] clk: qcom: reset: Increase max reset delay

[Stephan Gerhold]

On Wed, Jul 26, 2023 at 08:24:33PM +0200, Konrad Dybcio wrote:
> On 26.07.2023 20:21, Stephan Gerhold wrote:
> > On Wed, Jul 26, 2023 at 03:26:19PM +0200, Konrad Dybcio wrote:
> >> u8 limits us to 255 microseconds of delay. Promote the delay variable to
> >> u16 to hold bigger values.
> >>
> >> Signed-off-by: Konrad Dybcio <konrad.dybcio@linaro.org>
> > 
> > It would be clearer to change this together with an actual user that
> > needs > 255us. AFAICT atm MSM8909 is the only user of this and it has
> > just 15us.
> Some LPASS resets ask for 500, but I'm still working on that driver.
> 

Maybe send it together with that driver then? This feels close to the
typical "no API changes without also adding the user" rule.

Thanks,
Stephan