Re: [PATCH RESEND v3 2/2] Input: atmel_mxt_ts - support poweroff in suspend

[Stefan Eichenberger]

Hi Dmitry,

Thanks for the feedback, I had a first look at the changes and I'm not
sure if we would break some use cases. Therfore, here some questions.

On Fri, Mar 15, 2024 at 11:51:19AM -0700, Dmitry Torokhov wrote:
> > diff --git a/drivers/input/touchscreen/atmel_mxt_ts.c b/drivers/input/touchscreen/atmel_mxt_ts.c
> > index 542a31448c8f..2d5655385702 100644
> > --- a/drivers/input/touchscreen/atmel_mxt_ts.c
> > +++ b/drivers/input/touchscreen/atmel_mxt_ts.c
> > @@ -317,6 +317,7 @@ struct mxt_data {
> >  	struct gpio_desc *reset_gpio;
> >  	struct gpio_desc *wake_gpio;
> >  	bool use_retrigen_workaround;
> > +	bool poweroff_sleep;
> >  
> >  	/* Cached parameters from object table */
> >  	u16 T5_address;
> > @@ -2799,15 +2800,18 @@ static int mxt_configure_objects(struct mxt_data *data,
> >  			dev_warn(dev, "Error %d updating config\n", error);
> >  	}
> >  
> > -	if (data->multitouch) {
> > -		error = mxt_initialize_input_device(data);
> > -		if (error)
> > -			return error;
> > -	} else {
> > -		dev_warn(dev, "No touch object detected\n");
> > -	}
> > +	/* If input device is not already registered */
> > +	if (!data->input_dev) {
> > +		if (data->multitouch) {
> > +			error = mxt_initialize_input_device(data);
> > +			if (error)
> > +				return error;
> > +		} else {
> > +			dev_warn(dev, "No touch object detected\n");
> > +		}
> >  
> > -	mxt_debug_init(data);
> > +		mxt_debug_init(data);
> > +	}
> >  
> >  	return 0;
> >  }
> > @@ -3325,6 +3329,8 @@ static int mxt_probe(struct i2c_client *client)
> >  		msleep(MXT_RESET_INVALID_CHG);
> >  	}
> >  
> > +	data->poweroff_sleep = device_property_read_bool(&client->dev,
> > +							 "atmel,poweroff-sleep");
> >  	/*
> >  	 * Controllers like mXT1386 have a dedicated WAKE line that could be
> >  	 * connected to a GPIO or to I2C SCL pin, or permanently asserted low.
> > @@ -3387,12 +3393,21 @@ static int mxt_suspend(struct device *dev)
> >  	if (!input_dev)
> >  		return 0;
> >  
> > -	mutex_lock(&input_dev->mutex);
> > +	if (!device_may_wakeup(dev) && data->poweroff_sleep) {
> > +		if (data->reset_gpio)
> > +			gpiod_set_value(data->reset_gpio, 1);
> >  
> > -	if (input_device_enabled(input_dev))
> > -		mxt_stop(data);
> > +		regulator_bulk_disable(ARRAY_SIZE(data->regulators),
> > +				data->regulators);
> > +		data->T44_address = 0;
> > +	} else {
> > +		mutex_lock(&input_dev->mutex);
> > +
> > +		if (input_device_enabled(input_dev))
> > +			mxt_stop(data);
> >  
> > -	mutex_unlock(&input_dev->mutex);
> > +		mutex_unlock(&input_dev->mutex);
> > +	}
> 
> This all should go into mxt_stop(), so that if device is closed, or
> inhibited, you power it down as well (if you can).

We would then have to power it up during probe to see if the device is
threre, read the configuration and power it down again afterwards until
the device is opened. If the device is in bootloader mode we would most
likely have to keep the power on all the time and never turn it off,
right?

> 
> >  
> >  	disable_irq(data->irq);
> >  
> > @@ -3408,14 +3423,37 @@ static int mxt_resume(struct device *dev)
> >  	if (!input_dev)
> >  		return 0;
> >  
> > -	enable_irq(data->irq);
> > +	if (!device_may_wakeup(dev) && data->poweroff_sleep) {
> > +		int ret;
> >  
> > -	mutex_lock(&input_dev->mutex);
> > +		ret = regulator_bulk_enable(ARRAY_SIZE(data->regulators),
> > +				data->regulators);
> > +		if (ret) {
> > +			dev_err(dev, "failed to enable regulators: %d\n",
> > +					ret);
> > +			return ret;
> > +		}
> > +		msleep(MXT_BACKUP_TIME);
> >  
> > -	if (input_device_enabled(input_dev))
> > -		mxt_start(data);
> > +		if (data->reset_gpio) {
> > +			/* Wait a while and then de-assert the RESET GPIO line */
> > +			msleep(MXT_RESET_GPIO_TIME);
> > +			gpiod_set_value(data->reset_gpio, 0);
> > +			msleep(MXT_RESET_INVALID_CHG);
> > +		}
> >  
> > -	mutex_unlock(&input_dev->mutex);
> > +		/* This also enables the irq again */
> > +		mxt_initialize(data);
> 
> And this needs to go into mxt_start(). Also, we should make sure that
> once resume operation completes the device is fully operational. That
> means you should not request the firmware asynchronously in
> mxt_initialize() in case you are in the resume path. I think you should
> also unwind mxt_initialize() and mxt_configure_objects() to make it
> clear what is the part of initial initialization and what is part of
> re-initializing during resume. The configuration that is exposed to
> userspace (resolution, number of objects, other properties) should stay
> the same, the configuration of the chip itself (power mode, etc) should
> be restored.

Here we would most likely have to load the firmware (configuration)
synchronously all the time if the poweroff_sleep flag is set. Ths makes
sure that the device is ready when we open the device. Would this delay
be acceptable when opening the input device? Normally the configuration
is not that big and should load quite fast. 

Regards,
Stefan