Message-ID: <afb13ea8-fc67-14a2-f53f-0dfbd7a0eb8b@kernel.org>
Date:   Sat, 28 Jan 2017 14:02:45 +0000
From:   Jonathan Cameron <jic23@...nel.org>
To:     Fabrice Gasnier <fabrice.gasnier@...com>, linux@...linux.org.uk,
        robh+dt@...nel.org, linux-arm-kernel@...ts.infradead.org,
        devicetree@...r.kernel.org, linux-kernel@...r.kernel.org
Cc:     linux-iio@...r.kernel.org, mark.rutland@....com,
        mcoquelin.stm32@...il.com, alexandre.torgue@...com,
        lars@...afoo.de, knaack.h@....de, pmeerw@...erw.net,
        benjamin.gaignard@...aro.org, benjamin.gaignard@...com
Subject: Re: [PATCH 1/7] iio: adc: stm32: add support for triggered buffer
 mode

On 24/01/17 14:35, Fabrice Gasnier wrote:
> On 01/22/2017 01:53 PM, Jonathan Cameron wrote:
>> On 19/01/17 13:34, Fabrice Gasnier wrote:
>>> STM32 ADC conversions can be launched using hardware triggers.
>>> It can be used to start conversion sequences (group of channels).
>>> Selected channels are select via sequence registers.
>>> Trigger source is selected via 'extsel' (external trigger mux).
>>> Trigger polarity is set to rising edge by default.
>>>
>>> Signed-off-by: Fabrice Gasnier <fabrice.gasnier@...com>
>> A few small bits and pieces inline. The only significant element is about
>> making the data flow conform a little better to our poorly documented :(
>> typical data flow.  All about what the trigger is responsible for rather than
>> the device.
>>
>> Looks pretty good to me!
> 
> Hi Jonathan,
> 
> Many thanks for reviewing.
> Please find bellow some answers and a few questions.
> 
>> Jonathan
>>> ---
>>>  drivers/iio/adc/Kconfig     |   2 +
>>>  drivers/iio/adc/stm32-adc.c | 349 +++++++++++++++++++++++++++++++++++++++++++-
>>>  2 files changed, 348 insertions(+), 3 deletions(-)
>>>
>>> diff --git a/drivers/iio/adc/Kconfig b/drivers/iio/adc/Kconfig
>>> index 9c8b558..33341f4 100644
>>> --- a/drivers/iio/adc/Kconfig
>>> +++ b/drivers/iio/adc/Kconfig
>>> @@ -446,6 +446,8 @@ config STM32_ADC_CORE
>>>  	depends on ARCH_STM32 || COMPILE_TEST
>>>  	depends on OF
>>>  	depends on REGULATOR
>>> +	select IIO_BUFFER
>>> +	select IIO_TRIGGERED_BUFFER
>>>  	help
>>>  	  Select this option to enable the core driver for STMicroelectronics
>>>  	  STM32 analog-to-digital converter (ADC).
>>> diff --git a/drivers/iio/adc/stm32-adc.c b/drivers/iio/adc/stm32-adc.c
>>> index 5715e79..8d0b74b 100644
>>> --- a/drivers/iio/adc/stm32-adc.c
>>> +++ b/drivers/iio/adc/stm32-adc.c
>>> @@ -22,11 +22,16 @@
>>>  #include <linux/clk.h>
>>>  #include <linux/delay.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/interrupt.h>
>>>  #include <linux/io.h>
>>>  #include <linux/module.h>
>>>  #include <linux/platform_device.h>
>>>  #include <linux/of.h>
>>> +#include <linux/slab.h>
>>>  
>>>  #include "stm32-adc-core.h"
>>>  
>>> @@ -58,21 +63,66 @@
>>>  
>>>  /* STM32F4_ADC_CR2 - bit fields */
>>>  #define STM32F4_SWSTART			BIT(30)
>>> +#define STM32F4_EXTEN_SHIFT		28
>>>  #define STM32F4_EXTEN_MASK		GENMASK(29, 28)
>>> +#define STM32F4_EXTSEL_SHIFT		24
>>> +#define STM32F4_EXTSEL_MASK		GENMASK(27, 24)
>>>  #define STM32F4_EOCS			BIT(10)
>>>  #define STM32F4_ADON			BIT(0)
>>>  
>>>  /* STM32F4_ADC_SQR1 - bit fields */
>>>  #define STM32F4_L_SHIFT			20
>>>  #define STM32F4_L_MASK			GENMASK(23, 20)
>>> +#define STM32F4_SQ16_SHIFT		15
>>> +#define STM32F4_SQ16_MASK		GENMASK(19, 15)
>>> +#define STM32F4_SQ15_SHIFT		10
>>> +#define STM32F4_SQ15_MASK		GENMASK(14, 10)
>>> +#define STM32F4_SQ14_SHIFT		5
>>> +#define STM32F4_SQ14_MASK		GENMASK(9, 5)
>>> +#define STM32F4_SQ13_SHIFT		0
>>> +#define STM32F4_SQ13_MASK		GENMASK(4, 0)
>>> +
>>> +/* STM32F4_ADC_SQR2 - bit fields */
>>> +#define STM32F4_SQ12_SHIFT		25
>>> +#define STM32F4_SQ12_MASK		GENMASK(29, 25)
>>> +#define STM32F4_SQ11_SHIFT		20
>>> +#define STM32F4_SQ11_MASK		GENMASK(24, 20)
>>> +#define STM32F4_SQ10_SHIFT		15
>>> +#define STM32F4_SQ10_MASK		GENMASK(19, 15)
>>> +#define STM32F4_SQ9_SHIFT		10
>>> +#define STM32F4_SQ9_MASK		GENMASK(14, 10)
>>> +#define STM32F4_SQ8_SHIFT		5
>>> +#define STM32F4_SQ8_MASK		GENMASK(9, 5)
>>> +#define STM32F4_SQ7_SHIFT		0
>>> +#define STM32F4_SQ7_MASK		GENMASK(4, 0)
>>>  
>>>  /* STM32F4_ADC_SQR3 - bit fields */
>>> +#define STM32F4_SQ6_SHIFT		25
>>> +#define STM32F4_SQ6_MASK		GENMASK(29, 25)
>>> +#define STM32F4_SQ5_SHIFT		20
>>> +#define STM32F4_SQ5_MASK		GENMASK(24, 20)
>>> +#define STM32F4_SQ4_SHIFT		15
>>> +#define STM32F4_SQ4_MASK		GENMASK(19, 15)
>>> +#define STM32F4_SQ3_SHIFT		10
>>> +#define STM32F4_SQ3_MASK		GENMASK(14, 10)
>>> +#define STM32F4_SQ2_SHIFT		5
>>> +#define STM32F4_SQ2_MASK		GENMASK(9, 5)
>>>  #define STM32F4_SQ1_SHIFT		0
>>>  #define STM32F4_SQ1_MASK		GENMASK(4, 0)
>>>
>> Given all the above are only used in a big table which basically says what
>> the are, I wonder in this particular case if there is any real benefit to
>> using the defines?  Might actually be clearer just to have the numbers
>> inline.  What do you think?  I'm not fussed either way.
> 
> As you're suggesting it, I'll send an updated version without all these definitions.
> 
>>> +#define STM32_ADC_MAX_SQ		16	/* SQ1..SQ16 */
>>>  #define STM32_ADC_TIMEOUT_US		100000
>>>  #define STM32_ADC_TIMEOUT	(msecs_to_jiffies(STM32_ADC_TIMEOUT_US / 1000))
>>>  
>>> +/* External trigger enable */
>>> +enum stm32_adc_exten {
>>> +	STM32_EXTEN_SWTRIG,
>>> +	STM32_EXTEN_HWTRIG_RISING_EDGE,
>>> +	STM32_EXTEN_HWTRIG_FALLING_EDGE,
>>> +	STM32_EXTEN_HWTRIG_BOTH_EDGES,
>>> +};
>>> +
>>> +
>>>  /**
>>>   * struct stm32_adc - private data of each ADC IIO instance
>>>   * @common:		reference to ADC block common data
>>> @@ -82,6 +132,8 @@
>>>   * @clk:		clock for this adc instance
>>>   * @irq:		interrupt for this adc instance
>>>   * @lock:		spinlock
>>> + * @bufi:		data buffer index
>>> + * @num_conv:		expected number of scan conversions
>>>   */
>>>  struct stm32_adc {
>>>  	struct stm32_adc_common	*common;
>>> @@ -91,6 +143,8 @@ struct stm32_adc {
>>>  	struct clk		*clk;
>>>  	int			irq;
>>>  	spinlock_t		lock;		/* interrupt lock */
>>> +	int			bufi;
>>> +	int			num_conv;
>>>  };
>>>  
>>>  /**
>>> @@ -105,6 +159,18 @@ struct stm32_adc_chan_spec {
>>>  	const char		*name;
>>>  };
>>>  
>>> +/**
>>> + * stm32_adc_regs - stm32 ADC misc registers & bitfield desc
>>> + * @reg:		register offset
>>> + * @mask:		bitfield mask
>>> + * @shift:		left shift
>>> + */
>>> +struct stm32_adc_regs {
>>> +	int reg;
>>> +	int mask;
>>> +	int shift;
>> You could use the FIELD_PREP macro to avoid having to store both the mask
>> and shift.  Up to you though as definitely a matter for personal taste!
> 
> Thanks for pointing this out. I haven't noticed these macros before that.
> But this applies only to compile-time constant. This cannot be used with bellow
> array.
> 
>>> +};
>>> +
>>>  /* Input definitions common for all STM32F4 instances */
>>>  static const struct stm32_adc_chan_spec stm32f4_adc123_channels[] = {
>>>  	{ IIO_VOLTAGE, 0, "in0" },
>>> @@ -126,6 +192,33 @@ struct stm32_adc_chan_spec {
>>>  };
>>>  
>>>  /**
>>> + * stm32f4_sqr_regs - describe regular sequence registers
>>> + * - L: sequence len (register & bit field)
>>> + * - SQ1..SQ16: sequence entries (register & bit field)
>>> + */
>>> +static const struct stm32_adc_regs stm32f4_sqr_regs[STM32_ADC_MAX_SQ + 1] = {
>>> +	/* L: len bit field description to be kept as first element */
>>> +	{ STM32F4_ADC_SQR1, STM32F4_L_MASK, STM32F4_L_SHIFT },
>>> +	/* SQ1..SQ16 registers & bit fields */
>>> +	{ STM32F4_ADC_SQR3, STM32F4_SQ1_MASK, STM32F4_SQ1_SHIFT },
>>> +	{ STM32F4_ADC_SQR3, STM32F4_SQ2_MASK, STM32F4_SQ2_SHIFT },
>>> +	{ STM32F4_ADC_SQR3, STM32F4_SQ3_MASK, STM32F4_SQ3_SHIFT },
>>> +	{ STM32F4_ADC_SQR3, STM32F4_SQ4_MASK, STM32F4_SQ4_SHIFT },
>>> +	{ STM32F4_ADC_SQR3, STM32F4_SQ5_MASK, STM32F4_SQ5_SHIFT },
>>> +	{ STM32F4_ADC_SQR3, STM32F4_SQ6_MASK, STM32F4_SQ6_SHIFT },
>>> +	{ STM32F4_ADC_SQR2, STM32F4_SQ7_MASK, STM32F4_SQ7_SHIFT },
>>> +	{ STM32F4_ADC_SQR2, STM32F4_SQ8_MASK, STM32F4_SQ8_SHIFT },
>>> +	{ STM32F4_ADC_SQR2, STM32F4_SQ9_MASK, STM32F4_SQ9_SHIFT },
>>> +	{ STM32F4_ADC_SQR2, STM32F4_SQ10_MASK, STM32F4_SQ10_SHIFT },
>>> +	{ STM32F4_ADC_SQR2, STM32F4_SQ11_MASK, STM32F4_SQ11_SHIFT },
>>> +	{ STM32F4_ADC_SQR2, STM32F4_SQ12_MASK, STM32F4_SQ12_SHIFT },
>>> +	{ STM32F4_ADC_SQR1, STM32F4_SQ13_MASK, STM32F4_SQ13_SHIFT },
>>> +	{ STM32F4_ADC_SQR1, STM32F4_SQ14_MASK, STM32F4_SQ14_SHIFT },
>>> +	{ STM32F4_ADC_SQR1, STM32F4_SQ15_MASK, STM32F4_SQ15_SHIFT },
>>> +	{ STM32F4_ADC_SQR1, STM32F4_SQ16_MASK, STM32F4_SQ16_SHIFT },
>>> +};
>>> +
>>> +/**
>>>   * STM32 ADC registers access routines
>>>   * @adc: stm32 adc instance
>>>   * @reg: reg offset in adc instance
>>> @@ -211,6 +304,106 @@ static void stm32_adc_stop_conv(struct stm32_adc *adc)
>>>  }
>>>  
>>>  /**
>>> + * stm32_adc_conf_scan_seq() - Build regular channels scan sequence
>>> + * @indio_dev: IIO device
>>> + * @scan_mask: channels to be converted
>>> + *
>>> + * Conversion sequence :
>>> + * Configure ADC scan sequence based on selected channels in scan_mask.
>>> + * Add channels to SQR registers, from scan_mask LSB to MSB, then
>>> + * program sequence len.
>>> + */
>>> +static int stm32_adc_conf_scan_seq(struct iio_dev *indio_dev,
>>> +				   const unsigned long *scan_mask)
>>> +{
>>> +	struct stm32_adc *adc = iio_priv(indio_dev);
>>> +	const struct iio_chan_spec *chan;
>>> +	u32 val, bit;
>>> +	int i = 0;
>>> +
>>> +	for_each_set_bit(bit, scan_mask, indio_dev->masklength) {
>>> +		chan = indio_dev->channels + bit;
>>> +		/*
>>> +		 * Assign one channel per SQ entry in regular
>>> +		 * sequence, starting with SQ1.
>>> +		 */
>>> +		i++;
>>> +		if (i > STM32_ADC_MAX_SQ)
>>> +			return -EINVAL;
>>> +
>>> +		dev_dbg(&indio_dev->dev, "%s chan %d to SQ%d\n",
>>> +			__func__, chan->channel, i);
>>> +
>>> +		val = stm32_adc_readl(adc, stm32f4_sqr_regs[i].reg);
>>> +		val &= ~stm32f4_sqr_regs[i].mask;
>>> +		val |= chan->channel << stm32f4_sqr_regs[i].shift;
>>> +		stm32_adc_writel(adc, stm32f4_sqr_regs[i].reg, val);
>>> +	}
>>> +
>>> +	if (!i)
>>> +		return -EINVAL;
>>> +
>>> +	/* Sequence len */
>>> +	val = stm32_adc_readl(adc, stm32f4_sqr_regs[0].reg);
>>> +	val &= ~stm32f4_sqr_regs[0].mask;
>>> +	val |= ((i - 1) << stm32f4_sqr_regs[0].shift);
>>> +	stm32_adc_writel(adc, stm32f4_sqr_regs[0].reg, val);
>>> +
>>> +	return 0;
>>> +}
>>> +
>>> +/**
>>> + * stm32_adc_get_trig_extsel() - Get external trigger selection
>>> + * @indio_dev: IIO device
>>> + * @trig: trigger
>>> + *
>>> + * Returns trigger extsel value, if trig matches, -EINVAL otherwise.
>>> + */
>>> +static int stm32_adc_get_trig_extsel(struct iio_dev *indio_dev,
>>> +				     struct iio_trigger *trig)
>>> +{
>>> +	return -EINVAL;
>> ?????
> 
> Sorry about that... I've split into separate patches to ease the review.
> I agree this is miss-leading on first sight.
> Would you like me to squash patch 2 into this ?
> Or I can put a temporary comment here, removed then in patch 2...
> Or keep it as it is... please let me know if this matters.
This is fine. Perhaps a comment in the patch description?
> 
>>> +}
>>> +
>>> +/**
>>> + * stm32_adc_set_trig() - Set a regular trigger
>>> + * @indio_dev: IIO device
>>> + * @trig: IIO trigger
>>> + *
>>> + * Set trigger source/polarity (e.g. SW, or HW with polarity) :
>>> + * - if HW trigger disabled (e.g. trig == NULL, conversion launched by sw)
>>> + * - if HW trigger enabled, set source & polarity
>>> + */
>>> +static int stm32_adc_set_trig(struct iio_dev *indio_dev,
>>> +			      struct iio_trigger *trig)
>>> +{
>>> +	struct stm32_adc *adc = iio_priv(indio_dev);
>>> +	u32 val, extsel = 0, exten = STM32_EXTEN_SWTRIG;
>>> +	unsigned long flags;
>>> +	int ret;
>>> +
>>> +	if (trig) {
>>> +		ret = stm32_adc_get_trig_extsel(indio_dev, trig);
>>> +		if (ret < 0)
>>> +			return ret;
>>> +
>>> +		/* set trigger source, default to rising edge */
>> Interesting.  The complexity of this device kicks in again.
>>
>> If we are talking true exposed hardware pin (which I think it can be?) then
>> this ought to come from DT.  For the case of trigging on either edge of the
>> pwm style signals, we probably want to figure some way of allowing userspace
>> to control this.  For now in either case this is fine though! Got to start
>> somewhere.
> 
> You're completely right. I haven't included support for "exti" trigger (e.g. exti gpio)
> in this patchset (only pwm/timers).
> But this would make sense to set polarity in dt in this case. I'll think about this.
> I'm opened to suggestion, but if you're okay, i'll keep this for now.
Fine to start with a default, but later ideally make the polarity of this trigger
at least configurable in DT.
> 
>>> +		extsel = ret;
>>> +		exten = STM32_EXTEN_HWTRIG_RISING_EDGE;
>>> +	}
>>> +
>>> +	spin_lock_irqsave(&adc->lock, flags);
>>> +	val = stm32_adc_readl(adc, STM32F4_ADC_CR2);
>>> +	val &= ~(STM32F4_EXTEN_MASK | STM32F4_EXTSEL_MASK);
>>> +	val |= exten << STM32F4_EXTEN_SHIFT;
>>> +	val |= extsel << STM32F4_EXTSEL_SHIFT;
>>> +	stm32_adc_writel(adc, STM32F4_ADC_CR2, val);
>>> +	spin_unlock_irqrestore(&adc->lock, flags);
>>> +
>>> +	return 0;
>>> +}
>>> +
>>> +/**
>>>   * stm32_adc_single_conv() - Performs a single conversion
>>>   * @indio_dev: IIO device
>>>   * @chan: IIO channel
>>> @@ -234,6 +427,7 @@ static int stm32_adc_single_conv(struct iio_dev *indio_dev,
>>>  	reinit_completion(&adc->completion);
>>>  
>>>  	adc->buffer = &result;
>>> +	adc->bufi = 0;
>>>  
>>>  	/* Program chan number in regular sequence */
>>>  	val = stm32_adc_readl(adc, STM32F4_ADC_SQR3);
>>> @@ -301,17 +495,58 @@ static int stm32_adc_read_raw(struct iio_dev *indio_dev,
>>>  static irqreturn_t stm32_adc_isr(int irq, void *data)
>>>  {
>>>  	struct stm32_adc *adc = data;
>>> +	struct iio_dev *indio_dev = iio_priv_to_dev(adc);
>>>  	u32 status = stm32_adc_readl(adc, STM32F4_ADC_SR);
>>>  
>>>  	if (status & STM32F4_EOC) {
>>> -		*adc->buffer = stm32_adc_readw(adc, STM32F4_ADC_DR);
>>> -		complete(&adc->completion);
>> Normally I'd have put the separation between trigger and device a bit
>> earlier and had the actual reads in the function called by iio_trigger_poll
>> (and here after the wait for completion for the sysfs reads).
>>
>> I think it will make little practical difference, but it will conform better
>> to the data flow model (that kicks around in my head and has been in various
>> presentations others have done) of the trigger indicating either:
>> 1. Data ready
>> 2. Data should be ready - i.e. grab it now if on demand.
>>
>> The device side is responsible for then actually doing the read.
>>
>> Code wise I think this means moving a few lines but I may have missed some
>> details.
> 
> 1 - yes, for sysfs read, as ADC is converting a single value, Data Register
> (DR) read can be moved after the wait for completion. I haven't thought about it
> in the original patchset.
> Then, only slight change would be to mask EOC interrupt here, or clear EOC flag,
> because reading DR from ISR plays a dual role:
> - actual data read
> - clear EOC status flag.
> I'll add a comment to mention this.
EOC should only be cleared after the read - even if that is sometime later.
> 
> I don't see improvement in doing this, but I can update this for single (raw) conversion.
Yeah, not improvement when looking at this one driver - it's more about
conforming to a general structure shared by many drivers.
> 
> 2 - But, regarding buffer mode, when converting sequence of two or more channels,
> I'm not sure this is achievable: there is no internal fifo in ADC hardware IP.
> Then it needs one interrupt per channel (or use DMA).
Sadly this is quite common.  Usually result is that we only support this sort
of buffered reading of one channel at a time.

Presumably such a sequence of channels needs a separate trigger for each channel
capture as well?
> Once channel group has been converted, iio_trigger_poll() can be called.
> So I need to fill buffer from the ISR (actual read), for each channel.
hmm. Fair enough I suppose.

The classic way around this on simpler devices is to just not have the trigger
at all. Here however, we need it because the part itself supports a range
of different triggers.
> 
> In the end, it's advised to use DMA, especially when converting at 'high' rates
> or with two or more channels, to free up some CPU load, and avoid possible
> overun.
> 
> Shall I add comment to describe this here ?
Yes a comment might make this clear when we see it again in the future.
> Do you agree to keep this part of ISR as it is now ?
Ok. Leave it as it is, but a few comments explaining the logic will hopefully
allow us to follow it later without having to find this email thread!
> 
>>> +		adc->buffer[adc->bufi] = stm32_adc_readw(adc, STM32F4_ADC_DR);
>>> +		if (iio_buffer_enabled(indio_dev)) {
>>> +			adc->bufi++;
>>> +			if (adc->bufi >= adc->num_conv) {
>>> +				stm32_adc_conv_irq_disable(adc);
>>> +				iio_trigger_poll(indio_dev->trig);
>>> +			}
>>> +		} else {
>>> +			complete(&adc->completion);
>>> +		}
>>>  		return IRQ_HANDLED;
>>>  	}
>>>  
>>>  	return IRQ_NONE;
>>>  }
>>>  
>>> +/**
>>> + * stm32_adc_validate_trigger() - validate trigger for stm32 adc
>>> + * @indio_dev: IIO device
>>> + * @trig: new trigger
>>> + *
>>> + * Returns: 0 if trig matches one of the triggers registered by stm32 adc
>>> + * driver, -EINVAL otherwise.
>>> + */
>>> +static int stm32_adc_validate_trigger(struct iio_dev *indio_dev,
>>> +				      struct iio_trigger *trig)
>>> +{
>>> +	return stm32_adc_get_trig_extsel(indio_dev, trig) < 0 ? -EINVAL : 0;
>>> +}
>>> +
>>> +static int stm32_adc_update_scan_mode(struct iio_dev *indio_dev,
>>> +				      const unsigned long *scan_mask)
>>> +{
>>> +	struct stm32_adc *adc = iio_priv(indio_dev);
>>> +	int ret;
>>> +	u32 bit;
>>> +
>>> +	adc->num_conv = 0;
>>> +	for_each_set_bit(bit, scan_mask, indio_dev->masklength)
>>> +		adc->num_conv++;
>> Isn't there a count bits function? bitmap_weight I think which using the
>> hamming weight.
> I'll fix this.
> 
>>> +
>>> +	ret = stm32_adc_conf_scan_seq(indio_dev, scan_mask);
>>> +	if (ret)
>>> +		return ret;
>>> +
>>> +	return 0;
>>> +}
>>> +
>>>  static int stm32_adc_of_xlate(struct iio_dev *indio_dev,
>>>  			      const struct of_phandle_args *iiospec)
>>>  {
>>> @@ -350,11 +585,106 @@ static int stm32_adc_debugfs_reg_access(struct iio_dev *indio_dev,
>>>  
>>>  static const struct iio_info stm32_adc_iio_info = {
>>>  	.read_raw = stm32_adc_read_raw,
>>> +	.validate_trigger = stm32_adc_validate_trigger,
>>> +	.update_scan_mode = stm32_adc_update_scan_mode,
>>>  	.debugfs_reg_access = stm32_adc_debugfs_reg_access,
>>>  	.of_xlate = stm32_adc_of_xlate,
>>>  	.driver_module = THIS_MODULE,
>>>  };
>>>  
>>> +static int stm32_adc_buffer_preenable(struct iio_dev *indio_dev)
>>> +{
>>> +	struct stm32_adc *adc = iio_priv(indio_dev);
>>> +
>>> +	/* Reset adc buffer index */
>>> +	adc->bufi = 0;
>>> +
>>> +	/* Allocate adc buffer */
>>> +	adc->buffer = kzalloc(indio_dev->scan_bytes, GFP_KERNEL);
>> Is this big enough to justify a separate allocation?  I'd be tempted to
>> just allocate the maximum possible size as part of the stm32_adc structure..
> 
> So, you advise probe-time allocation for this buffer, why not ? :-)
> This would allow to simplify handling here, possibly remove preenable/postdisable
> routines.
> I'll do this in next patchset.
Quite.  Funilly enough this is one of the most common things to come up
in review.  People tend to forget that typically when you ask for a small allocation
you get a larger one anyway so the saving of making it exactly the right size
is very small if anything.
> 
>>> +	if (!adc->buffer)
>>> +		return -ENOMEM;
>>> +
>>> +	return 0;
>>> +}
>>> +
>>> +static int stm32_adc_buffer_postenable(struct iio_dev *indio_dev)
>>> +{
>>> +	struct stm32_adc *adc = iio_priv(indio_dev);
>>> +	int ret;
>>> +
>>> +	ret = stm32_adc_set_trig(indio_dev, indio_dev->trig);
>>> +	if (ret) {
>>> +		dev_err(&indio_dev->dev, "Can't set trigger\n");
>>> +		return ret;
>>> +	}
>>> +
>>> +	ret = iio_triggered_buffer_postenable(indio_dev);
>>> +	if (ret < 0)
>> Handle unwinding of previous calls if this fails.
> I'll fix this.
> 
> Thanks again, Regards,
> Fabrice
> 
>>> +		return ret;
>>> +
>>> +	stm32_adc_conv_irq_enable(adc);
>>> +	stm32_adc_start_conv(adc);
>>> +
>>> +	return 0;
>>> +}
>>> +
>>> +static int stm32_adc_buffer_predisable(struct iio_dev *indio_dev)
>>> +{
>>> +	struct stm32_adc *adc = iio_priv(indio_dev);
>>> +	int ret;
>>> +
>>> +	stm32_adc_stop_conv(adc);
>>> +	stm32_adc_conv_irq_disable(adc);
>>> +
>>> +	ret = iio_triggered_buffer_predisable(indio_dev);
>>> +	if (ret < 0)
>>> +		return ret;
>>> +
>>> +	ret = stm32_adc_set_trig(indio_dev, NULL);
>>> +	if (ret)
>>> +		dev_err(&indio_dev->dev, "Can't clear trigger\n");
>>> +
>>> +	return ret;
>>> +}
>>> +
>>> +static int stm32_adc_buffer_postdisable(struct iio_dev *indio_dev)
>>> +{
>>> +	struct stm32_adc *adc = iio_priv(indio_dev);
>>> +
>>> +	kfree(adc->buffer);
>>> +	adc->buffer = NULL;
>>> +
>>> +	return 0;
>>> +}
>>> +
>>> +static const struct iio_buffer_setup_ops stm32_adc_buffer_setup_ops = {
>>> +	.preenable = &stm32_adc_buffer_preenable,
>>> +	.postenable = &stm32_adc_buffer_postenable,
>>> +	.predisable = &stm32_adc_buffer_predisable,
>>> +	.postdisable = &stm32_adc_buffer_postdisable,
>>> +};
>>> +
>>> +static irqreturn_t stm32_adc_trigger_handler(int irq, void *p)
>>> +{
>>> +	struct iio_poll_func *pf = p;
>>> +	struct iio_dev *indio_dev = pf->indio_dev;
>>> +	struct stm32_adc *adc = iio_priv(indio_dev);
>>> +
>>> +	dev_dbg(&indio_dev->dev, "%s bufi=%d\n", __func__, adc->bufi);
>>> +
>>> +	/* reset buffer index */
>>> +	adc->bufi = 0;
>>> +	iio_push_to_buffers_with_timestamp(indio_dev, adc->buffer,
>>> +					   pf->timestamp);
>>> +
>>> +	iio_trigger_notify_done(indio_dev->trig);
>>> +
>>> +	/* re-enable eoc irq */
>>> +	stm32_adc_conv_irq_enable(adc);
>>> +
>>> +	return IRQ_HANDLED;
>>> +}
>>> +
>>>  static void stm32_adc_chan_init_one(struct iio_dev *indio_dev,
>>>  				    struct iio_chan_spec *chan,
>>>  				    const struct stm32_adc_chan_spec *channel,
>>> @@ -471,14 +801,26 @@ static int stm32_adc_probe(struct platform_device *pdev)
>>>  	if (ret < 0)
>>>  		goto err_clk_disable;
>>>  
>>> +	ret = iio_triggered_buffer_setup(indio_dev,
>>> +					 &iio_pollfunc_store_time,
>>> +					 &stm32_adc_trigger_handler,
>>> +					 &stm32_adc_buffer_setup_ops);
>>> +	if (ret) {
>>> +		dev_err(&pdev->dev, "buffer setup failed\n");
>>> +		goto err_clk_disable;
>>> +	}
>>> +
>>>  	ret = iio_device_register(indio_dev);
>>>  	if (ret) {
>>>  		dev_err(&pdev->dev, "iio dev register failed\n");
>>> -		goto err_clk_disable;
>>> +		goto err_buffer_cleanup;
>>>  	}
>>>  
>>>  	return 0;
>>>  
>>> +err_buffer_cleanup:
>>> +	iio_triggered_buffer_cleanup(indio_dev);
>>> +
>>>  err_clk_disable:
>>>  	clk_disable_unprepare(adc->clk);
>>>  
>>> @@ -491,6 +833,7 @@ static int stm32_adc_remove(struct platform_device *pdev)
>>>  	struct iio_dev *indio_dev = iio_priv_to_dev(adc);
>>>  
>>>  	iio_device_unregister(indio_dev);
>>> +	iio_triggered_buffer_cleanup(indio_dev);
>>>  	clk_disable_unprepare(adc->clk);
>>>  
>>>  	return 0;
>>>
> 

Powered by blists - more mailing lists