Re: [PATCH 3/7] soc: apple: Add driver for Apple PMGR power state controls

[Philipp Zabel]

Hi Hector,

On Wed, 2021-10-06 at 00:59 +0900, Hector Martin wrote:
> Implements genpd and reset providers for downstream devices. Each
> instance of the driver binds to a single register and represents a
> single SoC power domain.
> 
> The driver does not currently implement all features (auto-pm,
> clockgate-only state), but we declare the respective registers for
> documentation purposes. These features will be added as they become
> useful for downstream devices.
> 
> This also creates the apple/soc tree and Kconfig submenu.
> 
> Signed-off-by: Hector Martin <marcan@xxxxxxxxx>
> ---
>  MAINTAINERS                             |   1 +
>  drivers/soc/Kconfig                     |   1 +
>  drivers/soc/Makefile                    |   1 +
>  drivers/soc/apple/Kconfig               |  21 ++
>  drivers/soc/apple/Makefile              |   2 +
>  drivers/soc/apple/apple-pmgr-pwrstate.c | 281 ++++++++++++++++++++++++
>  6 files changed, 307 insertions(+)
>  create mode 100644 drivers/soc/apple/Kconfig
>  create mode 100644 drivers/soc/apple/Makefile
>  create mode 100644 drivers/soc/apple/apple-pmgr-pwrstate.c
> 
[...]
> diff --git a/drivers/soc/apple/apple-pmgr-pwrstate.c b/drivers/soc/apple/apple-pmgr-pwrstate.c
> new file mode 100644
> index 000000000000..a0338dbb29b8
> --- /dev/null
> +++ b/drivers/soc/apple/apple-pmgr-pwrstate.c
> @@ -0,0 +1,281 @@
[...]
> +static int apple_pmgr_reset_assert(struct reset_controller_dev *rcdev, unsigned long id)
> +{
> +	struct apple_pmgr_ps *ps = rcdev_to_apple_pmgr_ps(rcdev);
> +
> +	mutex_lock(&ps->genpd.mlock);
> +
> +	if (ps->genpd.status == GENPD_STATE_OFF)
> +		dev_err(ps->dev, "PS 0x%x: asserting RESET while powered down\n", ps->offset);
> +
> +	dev_dbg(ps->dev, "PS 0x%x: assert reset\n", ps->offset);
> +	/* Quiesce device before asserting reset */
> +	regmap_set_bits(ps->regmap, ps->offset, APPLE_PMGR_DEV_DISABLE);
> +	regmap_set_bits(ps->regmap, ps->offset, APPLE_PMGR_RESET);
> +
> +	mutex_unlock(&ps->genpd.mlock);
> +
> +	return 0;
> +}
> +
> +static int apple_pmgr_reset_deassert(struct reset_controller_dev *rcdev, unsigned long id)
> +{
> +	struct apple_pmgr_ps *ps = rcdev_to_apple_pmgr_ps(rcdev);
> +
> +	mutex_lock(&ps->genpd.mlock);
> +
> +	dev_dbg(ps->dev, "PS 0x%x: deassert reset\n", ps->offset);
> +	regmap_clear_bits(ps->regmap, ps->offset, APPLE_PMGR_RESET);
> +	regmap_clear_bits(ps->regmap, ps->offset, APPLE_PMGR_DEV_DISABLE);
> +
> +	if (ps->genpd.status == GENPD_STATE_OFF)
> +		dev_err(ps->dev, "PS 0x%x: RESET was deasserted while powered down\n", ps->offset);
> +
> +	mutex_unlock(&ps->genpd.mlock);
> +
> +	return 0;
> +}
> +
> +static int apple_pmgr_reset_reset(struct reset_controller_dev *rcdev, unsigned long id)
> +{
> +	int ret;
> +
> +	ret = apple_pmgr_reset_assert(rcdev, id);
> +	if (ret)
> +		return ret;
> +
> +	usleep_range(APPLE_PMGR_RESET_TIME, 2 * APPLE_PMGR_RESET_TIME);

Is this delay known to be long enough for all consumers using the
reset_control_reset() functionality? Are there any users at all?

Is it ok for a genpd transition to happen during this sleep?

> +	return apple_pmgr_reset_deassert(rcdev, id);
> +}

regards
Philipp