Re: [PATCH 03/15] extcon: cht-wc: Add Intel Cherry Trail Whiskey Cove PMIC extcon driver

[Hans de Goede]

Hi,

On 17-03-17 18:18, Andy Shevchenko wrote:
> On Fri, 2017-03-17 at 10:55 +0100, Hans de Goede wrote:
> > Add a driver for charger detection / control on the Intel Cherrytrail
> > Whiskey Cove PMIC.
>
> +Cc: Felipe for some question(s) below.
>
> >  drivers/extcon/extcon-cht-wc.c | 356
>
> I would use same pattern across drivers, i.e. "chtwc" (same for the rest
> of the drivers in this series).

I already answered this in another part of the thread,
but for the archives sake let me copy and paste my answer:

"One thing I disagree with is the cht_wc > chtwc rename you are
proposing for one Cherry Trail and Whiskey Cove are 2 different
words (4 if you look at spelling but 2 if you look at pronunciation,
so IMHO cht_wc is more readable other then that I see the suggested
rename as a lot of extra churn without any tangible benefits."

> > +#define CHT_WC_PWRSRC_IRQ		0x6e03
> > +#define CHT_WC_PWRSRC_IRQ_MASK		0x6e0f
> > +#define CHT_WC_PWRSRC_STS		0x6e1e
> > +#define CHT_WC_PWRSRC_VBUS		BIT(0)
> > +#define CHT_WC_PWRSRC_DC		BIT(1)
> > +#define CHT_WC_PWRSRC_BAT		BIT(2)
> > +#define CHT_WC_PWRSRC_ID_GND		BIT(3)
> > +#define CHT_WC_PWRSRC_ID_FLOAT		BIT(4)
>
> Not obvious for which register those bit definitions are.

They are for all 3 CHT_WC_PWRSRC_* registers, this is the
more or less usual irq setup for some i2c devices where
there is a status register, an irq register where the hardware
sets bits to 1 (and we write 1 to clear) when the corresponding
status bits changes and a mask register to select which irq
register bits actually will raise the interrupt line.

> Also, keep them ordered by offset.

Will fix.

> > +
> > +#define CHT_WC_PHYCTRL			0x5e07
> > +
>
> > +#define CHT_WC_CHGRCTRL0		0x5e16
>
> Dup!

Good catch, will fix.


> > +
> > +#define CHT_WC_CHGRCTRL0		0x5e16
>
> > +static int cht_wc_extcon_get_charger(struct cht_wc_extcon_data *ext)
> > +{
> > +	int ret, usbsrc, status, retries = 5;
> > +
> > +	do {
> > +		ret = regmap_read(ext->regmap, CHT_WC_USBSRC,
> > &usbsrc);
> > +		if (ret) {
> > +			dev_err(ext->dev, "Error reading usbsrc:
> > %d\n", ret);
> > +			return ret;
> > +		}
> > +		status = usbsrc & CHT_WC_USBSRC_STS_MASK;
> > +		if (status == CHT_WC_USBSRC_STS_SUCCESS ||
> > +		    status == CHT_WC_USBSRC_STS_FAIL)
> > +			break;
> > +
>
> > +		msleep(200);
>
> Comment why and why so long?

Fixed for v2 (actually switched to using jiffies +
time_before for a more accurate timeout).

> > +	} while (retries--);
>
> > +static void cht_wc_extcon_det_event(struct cht_wc_extcon_data *ext)
>
> det -> detect ?

Renamed to more accurate cht_wc_extcon_pwrsrc_event

> +static irqreturn_t cht_wc_extcon_isr(int irq, void *data)
> > +{
> > +	struct cht_wc_extcon_data *ext = data;
> > +	int ret, irqs;
> > +
> > +	ret = regmap_read(ext->regmap, CHT_WC_PWRSRC_IRQ, &irqs);
> > +	if (ret)
> > +		dev_err(ext->dev, "Error reading irqs: %d\n", ret);
>
> Shouldn't we return IRQ_NONE here?

I was wondering the same myself here, there is no good
answer here, this should simply never fail ... which does
make returning IRQ_NONE a good idea, I was worried that
would lead to a "nobody cared, disabling irq", but as said this
should never happen, so when it does, disabling the irq is
probably for the best. Will fix.

> > +
> > +	cht_wc_extcon_det_event(ext);
> > +
> > +	ret = regmap_write(ext->regmap, CHT_WC_PWRSRC_IRQ, irqs);
> > +	if (ret)
> > +		dev_err(ext->dev, "Error writing irqs: %d\n", ret);
> > +
> > +	return IRQ_HANDLED;
> > +}
> > +
>
> > +/* usb_id sysfs attribute for debug / testing purposes */
>
> Hmm... I would use debugfs for debug, otherwise it looks like it should
> be framework (extcon) wide.

Unfortunately these kinda sysfs files are somewhat normal when
dealing with USB-OTG. For example my board does not have the
id-pin hooked up to the connector, so to test host mode
I need to echo "gnd" to the sysfs attr. But also if I actually
want to use host-mode (or anyone else with the same or a similar
board).

This definitely is not something which belongs in the extcon-core.

> Perhaps Felipe can advise something here.
>
> > +static int cht_wc_extcon_probe(struct platform_device *pdev)
> > +{
>
> > +	struct cht_wc_extcon_data *ext;
> > +	struct intel_soc_pmic *pmic = dev_get_drvdata(pdev-
> > > dev.parent);
>
> Exchange them (assignment first).

Fixed.

> > +	int irq, ret;
> > +
>
>
> > +	ret = devm_request_threaded_irq(ext->dev, irq, NULL,
> > cht_wc_extcon_isr,
> > +					IRQF_ONESHOT, pdev->name,
> > ext);
> > +	if (ret) {
> > +		dev_err(ext->dev, "Failed to request interrupt\n");
> > +		return ret;
> > +	}
> > +
> > +	/* Unmask irqs */
> > +	ret = regmap_write(ext->regmap, CHT_WC_PWRSRC_IRQ_MASK,
> > +			   (int)~(CHT_WC_PWRSRC_VBUS |
>
> Hmm... Do you need explicit casting here?

Yes because the BIT(x) macros are unsigned longs and the ~ sets
the MSB so then the compiler complaints about truncating the
variable without the cast.

> > CHT_WC_PWRSRC_ID_GND |
> > +				  CHT_WC_PWRSRC_ID_FLOAT));
> > +	if (ret) {
> > +		dev_err(ext->dev, "Error writing irq-mask: %d\n",
> > ret);
> > +		return ret;
> > +	}

Regards,

Hans