Message-ID: <CAJZ5v0jwkwawYHRe893Fj7EuQLGCVzNSgq2YQN08Ycf0bqrhOQ@mail.gmail.com>
Date: Wed, 12 Feb 2025 12:14:29 +0100
From: "Rafael J. Wysocki" <rafael@...nel.org>
To: "Rafael J. Wysocki" <rjw@...ysocki.net>
Cc: Linux PM <linux-pm@...r.kernel.org>, LKML <linux-kernel@...r.kernel.org>, 
	Alan Stern <stern@...land.harvard.edu>, Ulf Hansson <ulf.hansson@...aro.org>, 
	Johan Hovold <johan@...nel.org>, Manivannan Sadhasivam <manivannan.sadhasivam@...aro.org>, 
	Jon Hunter <jonathanh@...dia.com>
Subject: Re: [PATCH v1 10/10] PM: runtime: Discover the lack of runtime PM support

On Tue, Feb 11, 2025 at 10:25 PM Rafael J. Wysocki <rjw@...ysocki.net> wrote:
>
> From: Rafael J. Wysocki <rafael.j.wysocki@...el.com>
>
> Previous changes have updated the PM core to special-case devices that
> have never had runtime PM enabled in some places, but what if a device
> had had runtime PM enabled at one point, but then it was permanently
> disabled?  Arguably, there is not much of a difference between such
> devices and the devices that have never had runtime PM enabled as far
> as system-wide suspend and resume is concerned, so they should be
> handled in the same way.
>
> For this reason, add a mechanism for discovering "lost" runtime PM
> support in devices with the help of the power.last_status field used
> for saving the last runtime PM status of the device known at the time
> when runtime PM was disabled for it.
>
> That field is set to RPM_INVALID initially and whenever runtime PM is
> enabled for a device (that is, when its power.disable_depth counter
> drops down to zero) and it is set to the current runtime PM status of
> the device when runtime PM is disabled (that is, the power.disable_depth
> counter becomes nonzero).  Therefore, if power.last_status is equal to
> RPM_INVALID for a device with runtime PM disabled, it means that
> runtime PM has never been enabled for that device.
>
> The PM core will now change the power.last_status value to RPM_UNKNOWN
> for devices having runtime PM disabled and power.last_status different
> from RPM_INVALID during the "prepare" phase of system suspend.  Then,
> __pm_runtime_disable() called subsequently on the device will set
> power.last_status to RPM_INVALID unless it changes from RPM_UNKNOWN
> to some other value in the meantime which requires enabling runtime PM
> for the device.  When power.last_status becomes RPM_INVALID and runtime
> PM is still disabled, the device will be handled as a "no runtime PM
> support" one from that point on until runtime PM is enabled for it
> again.

So the interim RPM_UNKNOWN value of last_status isn't really
necessary, it may as well be changed directly to RPM_INVALID in
device_prepare().

Scratch this one and I'll replace it with a different patch.

> Signed-off-by: Rafael J. Wysocki <rafael.j.wysocki@...el.com>
> ---
>  drivers/base/power/main.c    |    6 ++++++
>  drivers/base/power/runtime.c |   25 +++++++++++++++++++++++++
>  include/linux/pm.h           |    1 +
>  include/linux/pm_runtime.h   |    2 ++
>  4 files changed, 34 insertions(+)
>
> --- a/drivers/base/power/main.c
> +++ b/drivers/base/power/main.c
> @@ -1817,6 +1817,12 @@
>          * it again during the complete phase.
>          */
>         pm_runtime_get_noresume(dev);
> +       /*
> +        * Devices that have had runtime PM disabled recently may need to be
> +        * handled as though they have never supported it, so arrange for
> +        * detecting that situation.
> +        */
> +       pm_runtime_kick_last_status(dev);
>
>         if (dev->power.syscore)
>                 return 0;
> --- a/drivers/base/power/runtime.c
> +++ b/drivers/base/power/runtime.c
> @@ -1480,6 +1480,9 @@
>
>         if (dev->power.disable_depth > 0) {
>                 dev->power.disable_depth++;
> +               if (dev->power.last_status == RPM_UNKNOWN)
> +                       dev->power.last_status = RPM_INVALID;
> +
>                 goto out;
>         }
>
> @@ -1568,6 +1571,28 @@
>  EXPORT_SYMBOL_GPL(devm_pm_runtime_enable);
>
>  /**
> + * pm_runtime_kick_last_status - Start runtime PM support verification.
> + * @dev: Target device.
> + *
> + * If runtime PM is currently disabled for @dev, but it has been enabled at one
> + * point, change power.last_status for it to RPM_UNKNOWN, and if it is still
> + * RPM_UNKNOWN when __pm_runtime_disabled() is called for @dev next time, it
> + * will be changed to RPM_INVALID indicating no runtime PM support going
> + * forward until pm_runtime_enable() is called for @dev.
> + *
> + * This function is used by the PM core.
> + */
> +void pm_runtime_kick_last_status(struct device *dev)
> +{
> +       spin_lock_irq(&dev->power.lock);
> +
> +       if (dev->power.disable_depth && dev->power.last_status != RPM_INVALID)
> +               dev->power.last_status = RPM_UNKNOWN;
> +
> +       spin_unlock_irq(&dev->power.lock);
> +}
> +
> +/**
>   * pm_runtime_forbid - Block runtime PM of a device.
>   * @dev: Device to handle.
>   *
> --- a/include/linux/pm.h
> +++ b/include/linux/pm.h
> @@ -597,6 +597,7 @@
>         RPM_RESUMING,
>         RPM_SUSPENDED,
>         RPM_SUSPENDING,
> +       RPM_UNKNOWN,
>  };
>
>  /*
> --- a/include/linux/pm_runtime.h
> +++ b/include/linux/pm_runtime.h
> @@ -80,6 +80,7 @@
>  extern int pm_runtime_barrier(struct device *dev);
>  extern void pm_runtime_enable(struct device *dev);
>  extern void __pm_runtime_disable(struct device *dev, bool check_resume);
> +extern void pm_runtime_kick_last_status(struct device *dev);
>  extern void pm_runtime_allow(struct device *dev);
>  extern void pm_runtime_forbid(struct device *dev);
>  extern void pm_runtime_no_callbacks(struct device *dev);
> @@ -288,6 +289,7 @@
>  static inline int pm_runtime_barrier(struct device *dev) { return 0; }
>  static inline void pm_runtime_enable(struct device *dev) {}
>  static inline void __pm_runtime_disable(struct device *dev, bool c) {}
> +static inline void pm_runtime_kick_last_status(struct device *dev) {}
>  static inline void pm_runtime_allow(struct device *dev) {}
>  static inline void pm_runtime_forbid(struct device *dev) {}
>
>
>
>
>

Powered by blists - more mailing lists