RE: [PATCH v1 1/3] mfd: Add new mfd device TPS68470

["Mani, Rajmohan" <rajmohan.mani@intel.com>]

Hi Heikki,

Thanks for the reviews and patience.

> -----Original Message-----
> From: Heikki Krogerus [mailto:heikki.krogerus@linux.intel.com]
> Sent: Tuesday, June 06, 2017 5:49 AM
> To: Mani, Rajmohan <rajmohan.mani@intel.com>
> Cc: linux-kernel@vger.kernel.org; linux-gpio@vger.kernel.org; linux-
> acpi@vger.kernel.org; Lee Jones <lee.jones@linaro.org>; Linus Walleij
> <linus.walleij@linaro.org>; Alexandre Courbot <gnurou@gmail.com>; Rafael J.
> Wysocki <rjw@rjwysocki.net>; Len Brown <lenb@kernel.org>
> Subject: Re: [PATCH v1 1/3] mfd: Add new mfd device TPS68470
> 
> Hi Rajmohan,
> 
> On Tue, Jun 06, 2017 at 04:55:16AM -0700, Rajmohan Mani wrote:
> > +/*
> > + * tps68470_reg_read: Read a single tps68470 register.
> > + *
> > + * @tps: Device to read from.
> > + * @reg: Register to read.
> > + * @val: Contains the value
> > + */
> > +int tps68470_reg_read(struct tps68470 *tps, unsigned int reg,
> > +			unsigned int *val)
> > +{
> > +	int ret;
> > +
> > +	mutex_lock(&tps->lock);
> > +	ret = regmap_read(tps->regmap, reg, val);
> > +	mutex_unlock(&tps->lock);
> > +	return ret;
> > +}
> > +EXPORT_SYMBOL_GPL(tps68470_reg_read);
> > +
> > +/*
> > + * tps68470_reg_write: Write a single tps68470 register.
> > + *
> > + * @tps68470: Device to write to.
> > + * @reg: Register to write to.
> > + * @val: Value to write.
> > + */
> > +int tps68470_reg_write(struct tps68470 *tps, unsigned int reg,
> > +			unsigned int val)
> > +{
> > +	int ret;
> > +
> > +	mutex_lock(&tps->lock);
> > +	ret = regmap_write(tps->regmap, reg, val);
> > +	mutex_unlock(&tps->lock);
> > +	return ret;
> > +}
> > +EXPORT_SYMBOL_GPL(tps68470_reg_write);
> > +
> > +/*
> > + * tps68470_update_bits: Modify bits w.r.t mask and val.
> > + *
> > + * @tps68470: Device to write to.
> > + * @reg: Register to read-write to.
> > + * @mask: Mask.
> > + * @val: Value to write.
> > + */
> > +int tps68470_update_bits(struct tps68470 *tps, unsigned int reg,
> > +				unsigned int mask, unsigned int val) {
> > +	int ret;
> > +
> > +	mutex_lock(&tps->lock);
> > +	ret = regmap_update_bits(tps->regmap, reg, mask, val);
> > +	mutex_unlock(&tps->lock);
> > +	return ret;
> > +}
> > +EXPORT_SYMBOL_GPL(tps68470_update_bits);
> 
> I'm not sure you need those above wrappers at all, regmap is handling locking in
> any case.
> 

I had this following question from Alan Cox on the original code without these wrappers.

"What is the model for insuring that no interrupt or thread of a driver is not in parallel issuing a tps68470_ operation when the device goes away (eg if I down the i2c controller) ?"

To address the above concerns, I got extra cautious and implemented locks around the regmap_* calls.
Now, I have been asked from more than one reviewer about the necessity of the same.
With the use of devm_* calls, tps68470_remove() goes away and leaves the driver just with regmap_* calls.
Unless I hear from Alan or other reviewers otherwise, I will drop these wrappers around regmap_* calls.

> > +static const struct regmap_config tps68470_regmap_config = {
> > +	.reg_bits = 8,
> > +	.val_bits = 8,
> > +	.max_register = TPS68470_REG_MAX,
> > +};
> > +
> > +static int tps68470_chip_init(struct tps68470 *tps) {
> > +	unsigned int version;
> > +	int ret;
> > +
> > +	ret = tps68470_reg_read(tps, TPS68470_REG_REVID, &version);
> > +	if (ret < 0) {
> > +		dev_err(tps->dev,
> > +			"Failed to read revision register: %d\n", ret);
> > +		return ret;
> > +	}
> > +
> > +	dev_info(tps->dev, "TPS68470 REVID: 0x%x\n", version);
> > +
> > +	ret = tps68470_reg_write(tps, TPS68470_REG_RESET, 0xff);
> > +	if (ret < 0)
> > +		return ret;
> > +
> > +	/* FIXME: configure these dynamically */
> > +	/* Enable Daisy Chain LDO and configure relevant GPIOs as output */
> > +	ret = tps68470_reg_write(tps, TPS68470_REG_S_I2C_CTL, 2);
> > +	if (ret < 0)
> > +		return ret;
> > +
> > +	ret = tps68470_reg_write(tps, TPS68470_REG_GPCTL4A, 2);
> > +	if (ret < 0)
> > +		return ret;
> > +
> > +	ret = tps68470_reg_write(tps, TPS68470_REG_GPCTL5A, 2);
> > +	if (ret < 0)
> > +		return ret;
> > +
> > +	ret = tps68470_reg_write(tps, TPS68470_REG_GPCTL6A, 2);
> > +	if (ret < 0)
> > +		return ret;
> > +
> > +	/*
> > +	 * When SDA and SCL are routed to GPIO1 and GPIO2, the mode
> > +	 * for these GPIOs must be configured using their respective
> > +	 * GPCTLxA registers as inputs with no pull-ups.
> > +	 */
> > +	ret = tps68470_reg_write(tps, TPS68470_REG_GPCTL1A, 0);
> > +	if (ret < 0)
> > +		return ret;
> > +
> > +	ret = tps68470_reg_write(tps, TPS68470_REG_GPCTL2A, 0);
> > +	if (ret < 0)
> > +		return ret;
> > +
> > +	/* Enable daisy chain */
> > +	ret = tps68470_update_bits(tps, TPS68470_REG_S_I2C_CTL, 1, 1);
> > +	if (ret < 0)
> > +		return ret;
> > +
> > +	usleep_range(TPS68470_DAISY_CHAIN_DELAY_US,
> > +			TPS68470_DAISY_CHAIN_DELAY_US + 10);
> > +	return 0;
> > +}
> > +
> > +static int tps68470_probe(struct i2c_client *client) {
> > +	struct tps68470 *tps;
> > +	int ret;
> > +
> > +	tps = devm_kzalloc(&client->dev, sizeof(*tps), GFP_KERNEL);
> > +	if (!tps)
> > +		return -ENOMEM;
> > +
> > +	mutex_init(&tps->lock);
> > +	i2c_set_clientdata(client, tps);
> > +	tps->dev = &client->dev;
> > +
> > +	tps->regmap = devm_regmap_init_i2c(client,
> &tps68470_regmap_config);
> > +	if (IS_ERR(tps->regmap)) {
> > +		dev_err(tps->dev, "devm_regmap_init_i2c Error %d\n", ret);
> > +		return PTR_ERR(tps->regmap);
> > +	}
> > +
> > +	ret = mfd_add_devices(tps->dev, -1, tps68470s,
> > +			      ARRAY_SIZE(tps68470s), NULL, 0, NULL);
> > +	if (ret < 0) {
> > +		dev_err(tps->dev, "mfd_add_devices failed: %d\n", ret);
> > +		return ret;
> > +	}
> 
> devm_mfd_add_devices()?
> 

Ack

> > +	ret = tps68470_chip_init(tps);
> > +	if (ret < 0) {
> > +		dev_err(tps->dev, "TPS68470 Init Error %d\n", ret);
> > +		goto fail;
> > +	}
> > +
> > +	return 0;
> > +fail:
> > +	mutex_lock(&tps->lock);
> 
> Why do you need to lock here?
> 

Same as explained above (to address Alan's comments)

> > +	mfd_remove_devices(tps->dev);
> > +	mutex_unlock(&tps->lock);
> > +
> > +	return ret;
> > +}
> > +
> > +static int tps68470_remove(struct i2c_client *client) {
> > +	struct tps68470 *tps = i2c_get_clientdata(client);
> > +
> > +	mutex_lock(&tps->lock);
> > +	mfd_remove_devices(tps->dev);
> > +	mutex_unlock(&tps->lock);
> > +
> > +	return 0;
> > +}
> > +
> > +static const struct acpi_device_id tps68470_acpi_ids[] = {
> > +	{"INT3472"},
> > +	{},
> > +};
> > +
> > +MODULE_DEVICE_TABLE(acpi, tps68470_acpi_ids);
> > +
> > +static struct i2c_driver tps68470_driver = {
> > +	.driver = {
> > +		   .name = "tps68470",
> > +		   .acpi_match_table = ACPI_PTR(tps68470_acpi_ids),
> > +	},
> > +	.probe_new = tps68470_probe,
> > +	.remove = tps68470_remove,
> > +};
> 
> <snip>
> 
> > +/**
> > + * struct tps68470 - tps68470 sub-driver chip access routines
> > + *
> > + * Device data may be used to access the TPS68470 chip  */
> > +
> > +struct tps68470 {
> > +	struct device *dev;
> > +	struct regmap *regmap;
> > +	/*
> > +	 * Used to synchronize access to tps68470_ operations
> > +	 * and addition and removal of mfd devices
> > +	 */
> > +	struct mutex lock;
> 
> Is this lock really necessary at all? Actually, you probable don't even need this
> structure at all if you just rely on regmap functions in the drivers.
> 

Ack
I am looking into this and will get back with v2.

> 
> Thanks,
> 
> --
> heikki