Sat, Jun 08, 2024 at 05:16:33PM +0300, Laurent Pinchart kirjoitti:
> From: Clark Wang <xiaoning.wang@xxxxxxx>
>
> The ADP5585 is a 10/11 input/output port expander with a built in keypad
> matrix decoder, programmable logic, reset generator, and PWM generator.
> This driver supports the PWM function using the platform device
> registered by the core MFD driver.
>
> The driver is derived from an initial implementation from NXP, available
> in commit 113113742208 ("MLK-25922-1 pwm: adp5585: add adp5585 PWM
> support") in their BSP kernel tree. It has been extensively rewritten.
...
> +#define ADP5585_PWM_OSC_FREQ_HZ 1000000U
(1 * HZ_PER_MHZ) ?
Variant to use MEGA. Or even #define MHz in units.h if you wish.
Putting a few 0:s in a row is error prone.
...
> + ret = regmap_write(regmap, ADP5585_PWM_OFFT_LOW,
> + off & 0xff);
> + if (ret)
> + return ret;
> + ret = regmap_write(regmap, ADP5585_PWM_OFFT_HIGH,
> + (off >> 8) & 0xff);
> + if (ret)
> + return ret;
This is regular I²C regmap, why do you avoid using regmap bulk APIs?
> + ret = regmap_write(regmap, ADP5585_PWM_ONT_LOW,
> + on & 0xff);
> + if (ret)
> + return ret;
> + ret = regmap_write(regmap, ADP5585_PWM_ONT_HIGH,
> + (on >> 8) & 0xff);
> + if (ret)
> + return ret;
Ditto.
...
> +static int pwm_adp5585_get_state(struct pwm_chip *chip,
> + struct pwm_device *pwm,
> + struct pwm_state *state)
> +{
> + struct regmap *regmap = pwmchip_get_drvdata(chip);
> + unsigned int on, off;
> + unsigned int val;
> +
> + regmap_read(regmap, ADP5585_PWM_OFFT_LOW, &off);
> + regmap_read(regmap, ADP5585_PWM_OFFT_HIGH, &val);
> + off |= val << 8;
Ditto.
> + regmap_read(regmap, ADP5585_PWM_ONT_LOW, &on);
> + regmap_read(regmap, ADP5585_PWM_ONT_HIGH, &val);
> + on |= val << 8;
Ditto.
> + state->duty_cycle = on * (NSEC_PER_SEC / ADP5585_PWM_OSC_FREQ_HZ);
> + state->period = (on + off) * (NSEC_PER_SEC / ADP5585_PWM_OSC_FREQ_HZ);
> +
> + state->polarity = PWM_POLARITY_NORMAL;
> +
> + regmap_read(regmap, ADP5585_PWM_CFG, &val);
> + state->enabled = !!(val & ADP5585_PWM_EN);
> +
> + return 0;
Besides that, how do you guarantee that no IO may happen in between of those
calls? Probably you want a driver explicit lock? In that case, would you still
want to have a regmap internal lock?
> +}
...
> +static int adp5585_pwm_probe(struct platform_device *pdev)
> +{
> + struct device *dev = &pdev->dev;
> + struct adp5585_dev *adp5585 = dev_get_drvdata(dev->parent);
> + struct pwm_chip *chip;
> + int ret;
> +
> + chip = devm_pwmchip_alloc(dev, ADP5585_PWM_CHAN_NUM, 0);
> + if (IS_ERR(chip))
> + return PTR_ERR(chip);
> + device_set_of_node_from_dev(dev, dev->parent);
Still unclear to me why only few drivers use this.
> + pwmchip_set_drvdata(chip, adp5585->regmap);
> + chip->ops = &adp5585_pwm_ops;
> +
> + ret = devm_pwmchip_add(dev, chip);
> + if (ret)
> + return dev_err_probe(dev, ret, "failed to add PWM chip\n");
> +
> + return 0;
> +}
...
> +static const struct platform_device_id adp5585_pwm_id_table[] = {
> + { "adp5585-pwm" },
> + { /* Sentinel */ },
Drop comma. Otherwise it's not a sentinel strictly speaking.
> +};
--
With Best Regards,
Andy Shevchenko
