[04/14] mux: mmio: Add resume support

From: Théo Lebrun <theo.lebrun@bootlin.com>

Implement resume support

Signed-off-by: Théo Lebrun <theo.lebrun@bootlin.com>
Signed-off-by: Thomas Richard <thomas.richard@bootlin.com>
---
 drivers/mux/mmio.c | 34 ++++++++++++++++++++++++++++++++++
 1 file changed, 34 insertions(+)

Hi!

2024-01-15 at 17:14, Thomas Richard wrote:
> From: Théo Lebrun <theo.lebrun@bootlin.com>
> 
> Implement resume support

What Andy said, and please don't omit punctuation. Try to make it a
pleasure to read your patches!

> 
> Signed-off-by: Théo Lebrun <theo.lebrun@bootlin.com>
> Signed-off-by: Thomas Richard <thomas.richard@bootlin.com>
> ---
>  drivers/mux/mmio.c | 34 ++++++++++++++++++++++++++++++++++
>  1 file changed, 34 insertions(+)
> 
> diff --git a/drivers/mux/mmio.c b/drivers/mux/mmio.c
> index fd1d121a584b..ab4ef195fc0d 100644
> --- a/drivers/mux/mmio.c
> +++ b/drivers/mux/mmio.c
> @@ -125,13 +125,47 @@ static int mux_mmio_probe(struct platform_device *pdev)
>  
>  	mux_chip->ops = &mux_mmio_ops;
>  
> +	dev_set_drvdata(dev, mux_chip);
> +
>  	return devm_mux_chip_register(dev, mux_chip);
>  }
>  
> +#ifdef CONFIG_PM
> +static int mux_mmio_resume_noirq(struct device *dev)
> +{
> +	struct mux_chip *mux_chip = dev_get_drvdata(dev);
> +	int global_ret = 0;
> +	unsigned int i;
> +
> +	for (i = 0; i < mux_chip->controllers; i++) {
> +		struct mux_control *mux = &mux_chip->mux[i];
> +		int val = mux->cached_state;

You are not supposed to look at (or change) cached_state outside the
mux core.

> +		int ret;
> +
> +		if (val == MUX_IDLE_AS_IS)

The cached_state can never be MUX_IDLE_AS_IS. Sure, it happens to have
the same actual value as the correct MUX_CACHE_UNKNOWN, but abusing
that is all kinds of wrong.

> +			continue;
> +
> +		ret = mux_mmio_set(mux, val);
> +		if (ret) {

If mux_mmio_set() fails, cached_state ends up wrong as it should be set
to MUX_CACHE_UNKNOWN on failure. Low-level stuff like this needs to be
done by the mux core, or things becomes a maintenance hazard...

So, the meat of this function belongs in the mux core since none of
it looks mmio specific. It could probably be named mux_chip_resume()
or something such. That makes it simple to use the correct constant,
and the mux_control_set() helper makes it easy to get the handling of
cached_state right.

Cheers,
Peter

> +			dev_err(dev, "control %u: error restoring mux: %d\n", i, ret);
> +			if (!global_ret)
> +				global_ret = ret;
> +		}
> +	}
> +
> +	return global_ret;
> +}
> +#endif
> +
> +static const struct dev_pm_ops mux_mmio_pm_ops = {
> +	SET_NOIRQ_SYSTEM_SLEEP_PM_OPS(NULL, mux_mmio_resume_noirq)
> +};
> +
>  static struct platform_driver mux_mmio_driver = {
>  	.driver = {
>  		.name = "mmio-mux",
>  		.of_match_table	= mux_mmio_dt_ids,
> +		.pm = &mux_mmio_pm_ops,
>  	},
>  	.probe = mux_mmio_probe,
>  };
>

Hello Peter,

Thanks for the review.

On 1/15/24 23:31, Peter Rosin wrote:
> Hi!
> 
> 2024-01-15 at 17:14, Thomas Richard wrote:
>> From: Théo Lebrun <theo.lebrun@bootlin.com>
>>
>> Implement resume support
> 
> What Andy said, and please don't omit punctuation. Try to make it a
> pleasure to read your patches!

Yes my commit message needs to be more verbose, sorry.

> 
>>
>> Signed-off-by: Théo Lebrun <theo.lebrun@bootlin.com>
>> Signed-off-by: Thomas Richard <thomas.richard@bootlin.com>
>> ---
>>  drivers/mux/mmio.c | 34 ++++++++++++++++++++++++++++++++++
>>  1 file changed, 34 insertions(+)
>>
>> diff --git a/drivers/mux/mmio.c b/drivers/mux/mmio.c
>> index fd1d121a584b..ab4ef195fc0d 100644
>> --- a/drivers/mux/mmio.c
>> +++ b/drivers/mux/mmio.c
>> @@ -125,13 +125,47 @@ static int mux_mmio_probe(struct platform_device *pdev)
>>  
>>  	mux_chip->ops = &mux_mmio_ops;
>>  
>> +	dev_set_drvdata(dev, mux_chip);
>> +
>>  	return devm_mux_chip_register(dev, mux_chip);
>>  }
>>  
>> +#ifdef CONFIG_PM
>> +static int mux_mmio_resume_noirq(struct device *dev)
>> +{
>> +	struct mux_chip *mux_chip = dev_get_drvdata(dev);
>> +	int global_ret = 0;
>> +	unsigned int i;
>> +
>> +	for (i = 0; i < mux_chip->controllers; i++) {
>> +		struct mux_control *mux = &mux_chip->mux[i];
>> +		int val = mux->cached_state;
> 
> You are not supposed to look at (or change) cached_state outside the
> mux core.
> 
>> +		int ret;
>> +
>> +		if (val == MUX_IDLE_AS_IS)
> 
> The cached_state can never be MUX_IDLE_AS_IS. Sure, it happens to have
> the same actual value as the correct MUX_CACHE_UNKNOWN, but abusing
> that is all kinds of wrong.
> 
>> +			continue;
>> +
>> +		ret = mux_mmio_set(mux, val);
>> +		if (ret) {
> 
> If mux_mmio_set() fails, cached_state ends up wrong as it should be set
> to MUX_CACHE_UNKNOWN on failure. Low-level stuff like this needs to be
> done by the mux core, or things becomes a maintenance hazard...
> 
> So, the meat of this function belongs in the mux core since none of
> it looks mmio specific. It could probably be named mux_chip_resume()
> or something such. That makes it simple to use the correct constant,
> and the mux_control_set() helper makes it easy to get the handling of
> cached_state right.
> 

Thanks for the explanations.

So I created a mux_chip_resume function in the mux core.
This function restores each mux using mux_control_set.
The restored state is the cached state.
The muxes with a MUX_CACHE_UNKNOWN cache state are ignored.

So this patch will be splitted, one patch for the core, one for the mmio
driver.

Regards,