Re: [PATCH] ACPI: PM: Use nearest power-manageable ancestor
From: Rafael J. Wysocki
Date: Thu Jul 24 2025 - 06:01:13 EST
On Thu, Jul 24, 2025 at 11:21 AM Guan-Chun.Wu <409411716@xxxxxxxxxxxxxx> wrote:
>
> When a device’s power_manageable flag is false, we currently only
> fall back to the direct parent’s power state. In a deep hierarchy
> there may be a more distant ancestor that does support power
> management.
>
> Walk up the parent chain until we find the closest power_manageable
> ancestor and use its power state. If none is found, default to
> ACPI_STATE_D0 (fully on).
Is there a specific use case in the field in which this change is needed?
If there's none, let's not make it.
> Signed-off-by: Guan-Chun.Wu <409411716@xxxxxxxxxxxxxx>
> ---
> drivers/acpi/device_pm.c | 19 +++++++++++++++++--
> 1 file changed, 17 insertions(+), 2 deletions(-)
>
> diff --git a/drivers/acpi/device_pm.c b/drivers/acpi/device_pm.c
> index dbd4446025ec..81b47fb00e80 100644
> --- a/drivers/acpi/device_pm.c
> +++ b/drivers/acpi/device_pm.c
> @@ -84,8 +84,23 @@ int acpi_device_get_power(struct acpi_device *device, int *state)
> parent = acpi_dev_parent(device);
>
> if (!device->flags.power_manageable) {
> - /* TBD: Non-recursive algorithm for walking up hierarchy. */
> - *state = parent ? parent->power.state : ACPI_STATE_D0;
> + /*
> + * If the device itself is not power-manageable,
> + * walk up the parent hierarchy to find the closest
> + * ancestor that is power-manageable.
> + * Use that ancestor's power state as an estimate
> + * for this device. If no such ancestor exists,
> + * default to D0 (Fully On).
> + */
> + struct acpi_device *ancestor = parent;
> + /*
> + * Keep traversing up until a power-manageable ancestor
> + * is found or the root is reached
> + */
> + while (ancestor && !ancestor->flags.power_manageable)
> + ancestor = acpi_dev_parent(ancestor);
> + /* Use the found ancestor's power state, or D0 if none is found */
> + *state = ancestor ? ancestor->power.state : ACPI_STATE_D0;
> goto out;
> }
>
> --
> 2.34.1
>