Re: [PATCH v2 6/7] iio: light: rpr0521 triggered buffer added

["Koivunen, Mikko" <Mikko.Koivunen@fi.rohmeurope.com>]

On 8.4.2017 18:28, Jonathan Cameron wrote:
> On 07/04/17 13:07, Mikko Koivunen wrote:
>> Driver sets up and uses triggered buffer if there is irq defined for device in
>> device tree. Trigger producer triggers from rpr0521 drdy interrupt line.
>> Trigger consumer reads rpr0521 data to scan buffer.
>> Depends on previous commits of _scale and _offset.
>>
>> Signed-off-by: Mikko Koivunen <mikko.koivunen@fi.rohmeurope.com>
>> ---
> A few comments inline...
>
> Jonathan
>> Tested on LeMaker HiKey with AOSP7.1, IIO_HAL and kernel 4.4.
>> Builds OK with torvalds/linux 4.9
>> checkpatch.pl OK.
>>
>> Patch v1->v2 changes:
>> removed flagging of the new feature
>> removed unnecesssary rpr0521_update_scan_mode()
>> removed unnecessary comments.
>> added IIO_LE to channel spec
>>
>>  drivers/iio/light/rpr0521.c |  287 ++++++++++++++++++++++++++++++++++++++++++-
>>  1 file changed, 282 insertions(+), 5 deletions(-)
>>
>> diff --git a/drivers/iio/light/rpr0521.c b/drivers/iio/light/rpr0521.c
>> index dc522fd..72f608d 100644
>> --- a/drivers/iio/light/rpr0521.c
>> +++ b/drivers/iio/light/rpr0521.c
>> @@ -9,9 +9,10 @@
>>   *
>>   * IIO driver for RPR-0521RS (7-bit I2C slave address 0x38).
>>   *
>> - * TODO: illuminance channel, PM support, buffer
>> + * TODO: illuminance channel, PM support
>>   */
>>  
>> +
> Be careful to clear out stray whitespace changes like this one.
> Just add noise to the patch and may mean it takes longer to be accepted...

Ack. Need to double check next time.

>>  #include <linux/module.h>
>>  #include <linux/init.h>
>>  #include <linux/i2c.h>
>> @@ -20,6 +21,10 @@
>>  #include <linux/acpi.h>
>>  
>>  #include <linux/iio/iio.h>
>> +#include <linux/iio/buffer.h>
>> +#include <linux/iio/trigger.h>
>> +#include <linux/iio/trigger_consumer.h>
>> +#include <linux/iio/triggered_buffer.h>
>>  #include <linux/iio/sysfs.h>
>>  #include <linux/pm_runtime.h>
>>  
>> @@ -30,6 +35,7 @@
>>  #define RPR0521_REG_PXS_DATA		0x44 /* 16-bit, little endian */
>>  #define RPR0521_REG_ALS_DATA0		0x46 /* 16-bit, little endian */
>>  #define RPR0521_REG_ALS_DATA1		0x48 /* 16-bit, little endian */
>> +#define RPR0521_REG_INTERRUPT		0x4A
>>  #define RPR0521_PS_OFFSET_LSB		0x53
>>  #define RPR0521_PS_OFFSET_MSB		0x54
>>  
>> @@ -44,16 +50,33 @@
>>  #define RPR0521_ALS_DATA1_GAIN_SHIFT	2
>>  #define RPR0521_PXS_GAIN_MASK		GENMASK(5, 4)
>>  #define RPR0521_PXS_GAIN_SHIFT		4
>> +#define RPR0521_PXS_PERSISTENCE_MASK	GENMASK(3, 0)
>> +#define RPR0521_INTERRUPT_INT_TRIG_PS_MASK	BIT(0)
>> +#define RPR0521_INTERRUPT_INT_TRIG_ALS_MASK	BIT(1)
>> +#define RPR0521_INTERRUPT_INT_LATCH_MASK	BIT(2)
>> +#define RPR0521_INTERRUPT_INT_REASSERT_MASK	BIT(3)
>> +#define RPR0521_INTERRUPT_INT_MODE_MASK		GENMASK(5, 4)
>>  
>>  #define RPR0521_MODE_ALS_ENABLE		BIT(7)
>>  #define RPR0521_MODE_ALS_DISABLE	0x00
>>  #define RPR0521_MODE_PXS_ENABLE		BIT(6)
>>  #define RPR0521_MODE_PXS_DISABLE	0x00
>> +#define RPR0521_PXS_PERSISTENCE_DRDY	0x00
>> +#define RPR0521_INTERRUPT_INT_TRIG_PS_ENABLE	BIT(0)
>> +#define RPR0521_INTERRUPT_INT_TRIG_PS_DISABLE	0x00
>> +#define RPR0521_INTERRUPT_INT_TRIG_ALS_ENABLE	BIT(1)
>> +#define RPR0521_INTERRUPT_INT_TRIG_ALS_DISABLE	0x00
>> +#define RPR0521_INTERRUPT_INT_LATCH_DISABLE	BIT(2)
>> +#define RPR0521_INTERRUPT_INT_LATCH_ENABLE	0x00
>> +#define RPR0521_INTERRUPT_INT_REASSERT_ENABLE	BIT(3)
>> +#define RPR0521_INTERRUPT_INT_REASSERT_DISABLE	0x00
>> +#define RPR0521_INTERRUPT_INT_PS_OUTSIDE_DETECT	BIT(5)
>>  
>>  #define RPR0521_MANUFACT_ID		0xE0
>>  #define RPR0521_DEFAULT_MEAS_TIME	0x06 /* ALS - 100ms, PXS - 100ms */
>>  
>>  #define RPR0521_DRV_NAME		"RPR0521"
>> +#define RPR0521_IRQ_NAME		"rpr0521_event"
>>  #define RPR0521_REGMAP_NAME		"rpr0521_regmap"
>>  
>>  #define RPR0521_SLEEP_DELAY_MS	2000
>> @@ -169,6 +192,9 @@ struct rpr0521_data {
>>  	bool als_dev_en;
>>  	bool pxs_dev_en;
>>  
>> +	struct iio_trigger *drdy_trigger0;
>> +	bool drdy_trigger_on;
>> +
>>  	/* optimize runtime pm ops - enable device only if needed */
>>  	bool als_ps_need_en;
>>  	bool pxs_ps_need_en;
>> @@ -195,6 +221,13 @@ static const struct attribute_group rpr0521_attribute_group = {
>>  	.attrs = rpr0521_attributes,
>>  };
>>  
>> +/* Order of the channel data in buffer */
>> +enum rpr0521_scan_index_order {
>> +	RPR0521_CHAN_INDEX_PXS,
>> +	RPR0521_CHAN_INDEX_BOTH,
>> +	RPR0521_CHAN_INDEX_IR,
>> +};
>> +
>>  static const struct iio_chan_spec rpr0521_channels[] = {
>>  	{
>>  		.type = IIO_PROXIMITY,
>> @@ -203,6 +236,14 @@ static const struct iio_chan_spec rpr0521_channels[] = {
>>  			BIT(IIO_CHAN_INFO_OFFSET) |
>>  			BIT(IIO_CHAN_INFO_SCALE),
>>  		.info_mask_shared_by_all = BIT(IIO_CHAN_INFO_SAMP_FREQ),
>> +		.scan_index = RPR0521_CHAN_INDEX_PXS,
>> +		.scan_type = {
>> +			.sign = 'u',
>> +			.realbits = 16,
>> +			.storagebits = 16,
>> +			.shift = 0,
>> +			.endianness = IIO_LE,
>> +		},
>>  	},
>>  	{
>>  		.type = IIO_INTENSITY,
>> @@ -212,6 +253,14 @@ static const struct iio_chan_spec rpr0521_channels[] = {
>>  		.info_mask_separate = BIT(IIO_CHAN_INFO_RAW) |
>>  			BIT(IIO_CHAN_INFO_SCALE),
>>  		.info_mask_shared_by_all = BIT(IIO_CHAN_INFO_SAMP_FREQ),
>> +		.scan_index = RPR0521_CHAN_INDEX_BOTH,
>> +		.scan_type = {
>> +			.sign = 'u',
>> +			.realbits = 16,
>> +			.storagebits = 16,
>> +			.shift = 0,
>> +			.endianness = IIO_LE,
>> +		},
>>  	},
>>  	{
>>  		.type = IIO_INTENSITY,
>> @@ -221,9 +270,154 @@ static const struct iio_chan_spec rpr0521_channels[] = {
>>  		.info_mask_separate = BIT(IIO_CHAN_INFO_RAW) |
>>  			BIT(IIO_CHAN_INFO_SCALE),
>>  		.info_mask_shared_by_all = BIT(IIO_CHAN_INFO_SAMP_FREQ),
>> +		.scan_index = RPR0521_CHAN_INDEX_IR,
>> +		.scan_type = {
>> +			.sign = 'u',
>> +			.realbits = 16,
>> +			.storagebits = 16,
>> +			.shift = 0,
> shift default makes sense as 0 so don't bother specifying it
> (will be set to 0 anyway)

Ack.
>> +			.endianness = IIO_LE,
>> +		},
>>  	},
>>  };
>>  
>> +static int rpr0521_set_power_state(struct rpr0521_data *data, bool on,
>> +				   u8 device_mask);
>> +static int rpr0521_als_enable(struct rpr0521_data *data, u8 status);
>> +static int rpr0521_pxs_enable(struct rpr0521_data *data, u8 status);
> These forward definitions rather look like they could be removed with a bit
> of reorganization fo the file.  Do that in preference to having these in
> the middle of the code.

Ok.

>> +
>> +/* IRQ to trigger -handler */
> why the -?
>> +static irqreturn_t rpr0521_drdy_irq_handler(int irq, void *private)
>> +{
>> +	struct iio_dev *indio_dev = private;
>> +	struct rpr0521_data *data = iio_priv(indio_dev);
>> +
>> +	/* Notify trigger0 consumers/subscribers */
>> +	if (data->drdy_trigger_on)
>> +		iio_trigger_poll(data->drdy_trigger0);
> How else are you getting here? Also the core will be fine
> if nothing cares and the trigger is polled.
> If there is another cause, we aren't handling it so should claim
> to be doing so.

Sometimes there is static on line and interrupt is generated even though
sensor interrupt output is in high impedance mode. What I've seen it's
max 2 interrupts in second, usually none. This doesn't matter much for
stability if sensor is switched on. It just causes extra reads. But if
extra interrupt happens when sensor power is switched off it isn't there
to respond to i2c data reading.
I don't know enough to know if all systems can handle this gracefully or
not. That's why safe playing here.
I'm keeping it unless someone can confirm that 2Hz regmap_bulk_read in
trigger_consumer_handler will not be causing ill effects on system when
there is no answer from sensor.

>> +
>> +	return IRQ_HANDLED;
>> +}
>> +
>> +static irqreturn_t rpr0521_trigger_consumer_handler(int irq, void *p)
>> +{
>> +	struct iio_poll_func *pf = p;
>> +	struct iio_dev *indio_dev = pf->indio_dev;
>> +	struct rpr0521_data *data = iio_priv(indio_dev);
>> +	int ret;
>> +
>> +	u8 buffer[16]; /* 3 16-bit channels + padding + ts */
>> +
>> +	ret = regmap_bulk_read(data->regmap,
>> +		RPR0521_REG_PXS_DATA,
>> +		&buffer,
>> +		(3*2)+1);	/* 3 * 16-bit + (discarded) int clear reg. */
> kernel style is spaces around the * and +
Ack.
>> +	if (ret < 0)
>> +		goto done;
>> +
>> +	iio_push_to_buffers_with_timestamp(indio_dev,
>> +		buffer,
> Dont wrap more than you have two.  So make this two lines perhaps.
>> +		pf->timestamp);
>> +done:
Refactoring to v3 to get rid of done:
>> +	iio_trigger_notify_done(indio_dev->trig);
>> +
>> +	return IRQ_HANDLED;
>> +}
>> +
>> +static int rpr0521_write_int_enable(struct rpr0521_data *data)
>> +{
>> +	int err;
>> +
>> +	/* Interrupt after each measurement */
>> +	err = regmap_update_bits(data->regmap, RPR0521_REG_PXS_CTRL,
>> +		RPR0521_PXS_PERSISTENCE_MASK,
>> +		RPR0521_PXS_PERSISTENCE_DRDY);
>> +	err = err || regmap_update_bits(data->regmap, RPR0521_REG_INTERRUPT,
> umm. really?  All of them? Are we looking at a clear the whole register
> case?  If so we should probably only clear relevant ones..

Hmm. .  Valid point. In drdy case we can write what ever to latch and
mode. This will be cleaner with write instead up write_bits.
==> change to v3
> I don't like these stacked error handlers (err = err | *).
> Just handle each err as and when.  It's more code, but easier
> to review / follow.

It's (err = err || *), bail out on first error.
Matter of taste, ack.

>> +		RPR0521_INTERRUPT_INT_REASSERT_MASK |
>> +		RPR0521_INTERRUPT_INT_LATCH_MASK |
>> +		RPR0521_INTERRUPT_INT_TRIG_ALS_MASK |
>> +		RPR0521_INTERRUPT_INT_TRIG_PS_MASK,
>> +		RPR0521_INTERRUPT_INT_REASSERT_DISABLE |
>> +		RPR0521_INTERRUPT_INT_LATCH_ENABLE |
>> +		RPR0521_INTERRUPT_INT_TRIG_ALS_DISABLE |
>> +		RPR0521_INTERRUPT_INT_TRIG_PS_ENABLE
>> +		);
>> +	err = err || regmap_update_bits(data->regmap, RPR0521_REG_INTERRUPT,
>> +		RPR0521_INTERRUPT_INT_MODE_MASK,
>> +		RPR0521_INTERRUPT_INT_PS_OUTSIDE_DETECT);
>> +	if (err) {
>> +		dev_err(&data->client->dev, "Interrupt setup write fail.\n");
>> +		return -EBUSY;
>> +		}
>> +	return 0;
>> +}
>> +
>> +static int rpr0521_write_int_disable(struct rpr0521_data *data)
>> +{
>> +	return regmap_update_bits(data->regmap, RPR0521_REG_INTERRUPT,
>> +				RPR0521_INTERRUPT_INT_TRIG_ALS_MASK |
>> +				RPR0521_INTERRUPT_INT_TRIG_PS_MASK,
>> +				RPR0521_INTERRUPT_INT_TRIG_ALS_DISABLE |
>> +				RPR0521_INTERRUPT_INT_TRIG_PS_DISABLE
>> +				);
>> +}
>> +
>> +/* Trigger -enable/disable */
>> +static int rpr0521_pxs_drdy_set_state(struct iio_trigger *trigger,
>> +	bool enable_drdy)
>> +{
> This may be called from the moment iio_trigger_register is called until
> the trigger is unregistered on remove.
>
> Could that potentially be a problem as it might occur after the device
> is powered off in remove?  Hard to be sure without diving into the data
> sheet.  If it if safe as it is, then use the devm functions as described
> below, but perhaps put a note here about the fact it is safe to have
> this potential race.

Sensor itself doesn't care much, but since measurement is enabled we are
draining power for nothing.
I didn't like to touch pm_ -stuff, since I didn't know much of it. But
now since I did, I think I found couple of bugs both in the current code
and new code.

There is another thing for the order. Int pin keeps state on power down.
So shutting down sensor before clearing interrupt might cause power
drain also on interrupt line.

==>
Reordering the _probe() and _remove() for v3.
Adding int clear to poweroff.
Adding couple commits for pm bugfixes.

One question though. Is it ok to register device first, and then
trigger/buffer stuff? For now it seems like correct order.

>> +	struct iio_dev *indio_dev = iio_trigger_get_drvdata(trigger);
>> +	struct rpr0521_data *data = iio_priv(indio_dev);
>> +	int err;
>> +
>> +	if (enable_drdy) {
>> +		/* ENABLE DRDY */
>> +		mutex_lock(&data->lock);
>> +		err = rpr0521_set_power_state(data, true,
>> +			(RPR0521_MODE_PXS_MASK & RPR0521_MODE_ALS_MASK));
>> +		mutex_unlock(&data->lock);
>> +		if (err)
>> +			goto rpr0521_set_state_err;
>> +		err = rpr0521_pxs_enable(data, RPR0521_MODE_PXS_ENABLE);

Removing rpr0521_[pxs/als]_[en/dis]able(), since they are by default
enabled from rpr0521_init and _set_power_state sets them on if pm has
disabled them.
>> +		if (err)
>> +			goto rpr0521_set_state_err;
>> +		err = rpr0521_als_enable(data, RPR0521_MODE_ALS_ENABLE);
>> +		if (err)
>> +			goto rpr0521_set_state_err;
>> +		err = rpr0521_write_int_enable(data);
>> +		if (err)
>> +			goto rpr0521_set_state_err;
>> +	} else {
>> +		/* DISABLE DRDY */
>> +		mutex_lock(&data->lock);
>> +		err = rpr0521_set_power_state(data, false,
>> +			(RPR0521_MODE_PXS_MASK & RPR0521_MODE_ALS_MASK));
>> +		mutex_unlock(&data->lock);
>> +		if (err)
>> +			goto rpr0521_set_state_err;
>> +		err = rpr0521_als_enable(data, RPR0521_MODE_ALS_DISABLE);
>> +		if (err)
>> +			goto rpr0521_set_state_err;
>> +		err = rpr0521_pxs_enable(data, RPR0521_MODE_PXS_DISABLE);
>> +		if (err)
>> +			goto rpr0521_set_state_err;
>> +		err = rpr0521_write_int_disable(data);
>> +		if (err)
>> +			goto rpr0521_set_state_err;
>> +	}
>> +	data->drdy_trigger_on = enable_drdy;
>> +	return 0;
>> +
>> +rpr0521_set_state_err:
>> +	dev_err(&data->client->dev, "rpr0521_pxs_drdy_set_state failed\n");
>> +	return err;
>> +}
>> +
>> +static const struct iio_trigger_ops rpr0521_trigger_ops = {
>> +	.set_trigger_state = rpr0521_pxs_drdy_set_state,
>> +	.owner = THIS_MODULE,
>> +	};
>> +
>>  static int rpr0521_als_enable(struct rpr0521_data *data, u8 status)
>>  {
>>  	int ret;
>> @@ -454,6 +648,9 @@ static int rpr0521_read_raw(struct iio_dev *indio_dev,
>>  		if (chan->type != IIO_INTENSITY && chan->type != IIO_PROXIMITY)
>>  			return -EINVAL;
>>  
>> +		if (iio_buffer_enabled(indio_dev))
>> +			return -EBUSY;
>> +
>>  		device_mask = rpr0521_data_reg[chan->address].device_mask;
>>  
>>  		mutex_lock(&data->lock);
>> @@ -664,20 +861,90 @@ static int rpr0521_probe(struct i2c_client *client,
>>  		return ret;
>>  	}
>>  
>> -	ret = pm_runtime_set_active(&client->dev);
>> -	if (ret < 0)
>> -		return ret;
>> +	/* IRQ to trigger setup */
>> +	if (client->irq) {
>> +		/* Trigger0 producer setup */
>> +		data->drdy_trigger0 = devm_iio_trigger_alloc(
>> +			indio_dev->dev.parent,
>> +			"%s-dev%d",
>> +			indio_dev->name,
>> +			indio_dev->id);
>> +		if (!data->drdy_trigger0) {
>> +			ret = -ENOMEM;
>> +			return ret;
>> +		}
>> +		data->drdy_trigger0->dev.parent = indio_dev->dev.parent;
>> +		data->drdy_trigger0->ops = &rpr0521_trigger_ops;
>> +		iio_trigger_set_drvdata(data->drdy_trigger0, indio_dev);
>> +
>> +		ret = iio_trigger_register(data->drdy_trigger0);
> devm_iio_trigger_register should be fine
See below.
>> +		if (ret) {
>> +			dev_err(&client->dev, "iio trigger register failed\n");
>> +			return ret;
>> +		}
>>  
>> +		/* Set trigger0 as default trigger (and inc refcount) */
>> +		indio_dev->trig = iio_trigger_get(data->drdy_trigger0);
>> +
>> +		/* Ties irq to trigger producer handler. */
>> +		ret = devm_request_threaded_irq(&client->dev, client->irq,
>> +			rpr0521_drdy_irq_handler,
>> +			NULL,
>> +			IRQF_TRIGGER_FALLING | IRQF_ONESHOT,
>> +			RPR0521_IRQ_NAME,
>> +			indio_dev);
>> +		if (ret < 0) {
>> +			dev_err(&client->dev, "request irq %d for trigger0 failed\n",
>> +				client->irq);
>> +			goto err_iio_trigger_put_unregister;
>> +			}
>> +		/*
>> +		 * Now whole pipe from physical interrupt (irq defined by
>> +		 * devicetree to device) to trigger0 output is set up.
>> +		 */
>> +
>> +		/* Trigger consumer setup */
>> +		ret = iio_triggered_buffer_setup(indio_dev,
>> +			&iio_pollfunc_store_time,
>> +			rpr0521_trigger_consumer_handler,
>> +			NULL);		//Use default _ops
>> +		if (ret < 0) {
>> +			dev_err(&client->dev, "iio triggered buffer setup failed\n");
>> +			/* Count on devm to free_irq */
>> +			goto err_iio_trigger_put_unregister;
>> +		}
>> +	}
>> +
>> +	ret = pm_runtime_set_active(&client->dev);
>> +	if (ret < 0) {
>> +		dev_err(&client->dev, "set_active failed\n");
> I'm not particularly bothered either way by the change, but technically
> adding this error message is unconnected to the subject of this patch so
> should not be here.
>
>> +		goto err_iio_unregister_trigger_consumer;
>> +		}
>>  	pm_runtime_enable(&client->dev);
>>  	pm_runtime_set_autosuspend_delay(&client->dev, RPR0521_SLEEP_DELAY_MS);
>>  	pm_runtime_use_autosuspend(&client->dev);
>>  
>>  	return iio_device_register(indio_dev);
>> +
> One line is enough.
Ok.
>> +
> Use devm versions and this lot goes away.
>> +err_iio_unregister_trigger_consumer:
>> +	iio_triggered_buffer_cleanup(indio_dev);
>> +err_iio_trigger_put_unregister:
>> +	if (indio_dev->trig) {
>> +		iio_trigger_put(indio_dev->trig);
>> +		indio_dev->trig = NULL;
>> +		}
>> +	if (data->drdy_trigger0) {
>> +		iio_trigger_unregister(data->drdy_trigger0);
>> +		data->drdy_trigger0 = NULL;
>> +		}
>> +	return ret;
>>  }
>>  
>>  static int rpr0521_remove(struct i2c_client *client)
>>  {
>>  	struct iio_dev *indio_dev = i2c_get_clientdata(client);
>> +	struct rpr0521_data *data = iio_priv(indio_dev);
>>  
>>  	iio_device_unregister(indio_dev);
>>  
>> @@ -685,8 +952,18 @@ static int rpr0521_remove(struct i2c_client *client)
>>  	pm_runtime_set_suspended(&client->dev);
>>  	pm_runtime_put_noidle(&client->dev);
>>  
>> -	rpr0521_poweroff(iio_priv(indio_dev));
>> +	rpr0521_poweroff(data);
>>  
>> +	iio_triggered_buffer_cleanup(indio_dev);
>> +	if (indio_dev->trig) {
>> +		iio_trigger_put(indio_dev->trig);
>> +		indio_dev->trig = NULL;
>> +		}
>> +	if (data->drdy_trigger0) {
>> +		iio_trigger_unregister(data->drdy_trigger0);
>> +		data->drdy_trigger0 = NULL;
>> +		}
>> +	/* Rest of the cleanup is done by devm. */
> Givem where you have them, I don't think anything stops you using devm to
> handle the trigger registration and buffer cleanup.  No need to set
> either of hte magic flags to NULL as far as I can see...

For a second let's think we removed this line

>>+	if (data->drdy_trigger_on)

from irq_handler().
If irq handler is not yet unregistered by devm_ and thus interrupt can
be generated (f.ex. by static), then what does
iio_trigger_poll(data->drdy_trigger0) do with this unregistered trigger
pointer? If call is ok, then no need for NULL I guess.

>
> Use the devm versions and remove should be unchanged...
Copy-paste from another post:

Can't, since no devm_ on kernel 4.4.
I could make another commit to change to devm_, but I don't have target
to test it.

> OK. We can leave it for someone else to clean up in future.

> However - look above - there is a race in how you have it currently ordered
> that means you should order it differently and not use the devm versions.
Ack.

>>  	return 0;
>>  }
>>  
>>