[PATCH v1] PM: sleep: Unblock runtime PM when device prepare fails

[PATCH v1] PM: sleep: Unblock runtime PM when device prepare fails

[Shibo Zhu]

device_prepare() blocks runtime PM for a device with runtime PM disabled
before invoking its system-sleep ->prepare() callback. For a device that
has never enabled runtime PM, this changes dev->power.last_status from
RPM_INVALID to RPM_BLOCKED.

If the callback returns an error, dpm_prepare() does not move the device
to dpm_prepared_list. Consequently, the recovery path through
dpm_complete() never calls device_complete() for the failing device.

The error path drops the runtime PM usage reference, but does not clear
RPM_BLOCKED. A later legitimate pm_runtime_enable() then reports:

  Attempt to enable runtime PM when it is blocked

before clearing the stale state.

Call pm_runtime_unblock() on the prepare error path before dropping the
runtime PM reference, matching the cleanup performed by device_complete().

The issue was reproduced with a platform test device whose ->prepare()
callback returns -EIO while runtime PM has never been enabled. Before
the fix, last_status remained RPM_BLOCKED after the failed suspend and
the first pm_runtime_enable() produced the warning above. With the fix,
last_status is restored to RPM_INVALID and the warning is absent.

Fixes: 3e5eee147b7b ("PM: Block enabling of runtime PM during system suspend")
Cc: stable@vger.kernel.org
Assisted-by: Codex:gpt-5.6-sol
Assisted-by: ChatGPT:GPT-5.6 Sol
Signed-off-by: Shibo Zhu <3499129952@qq.com>
---
 drivers/base/power/main.c | 1 +
 1 file changed, 1 insertion(+)

diff --git a/drivers/base/power/main.c b/drivers/base/power/main.c
index f71467f6ada4..4ed2e9b7e99b 100644
--- a/drivers/base/power/main.c
+++ b/drivers/base/power/main.c
@@ -2234,6 +2234,7 @@ static int device_prepare(struct device *dev, pm_message_t state)
 
 	if (ret < 0) {
 		suspend_report_result(dev, callback, ret);
+		pm_runtime_unblock(dev);
 		pm_runtime_put(dev);
 		return ret;
 	}

base-commit: 0f23d56f17fdfc7db69d51f64c8b91bbab947aa9
-- 
2.43.0

Re: [PATCH v1] PM: sleep: Unblock runtime PM when device prepare fails

[Pengjie Zhang]

On 8/19/2026 12:18 AM, Shibo Zhu wrote:
> device_prepare() blocks runtime PM for a device with runtime PM disabled
> before invoking its system-sleep ->prepare() callback. For a device that
> has never enabled runtime PM, this changes dev->power.last_status from
> RPM_INVALID to RPM_BLOCKED.
>
> If the callback returns an error, dpm_prepare() does not move the device
> to dpm_prepared_list. Consequently, the recovery path through
> dpm_complete() never calls device_complete() for the failing device.
>
> The error path drops the runtime PM usage reference, but does not clear
> RPM_BLOCKED. A later legitimate pm_runtime_enable() then reports:
>
>    Attempt to enable runtime PM when it is blocked
>
> before clearing the stale state.
>
> Call pm_runtime_unblock() on the prepare error path before dropping the
> runtime PM reference, matching the cleanup performed by device_complete().
>
> The issue was reproduced with a platform test device whose ->prepare()
> callback returns -EIO while runtime PM has never been enabled. Before
> the fix, last_status remained RPM_BLOCKED after the failed suspend and
> the first pm_runtime_enable() produced the warning above. With the fix,
> last_status is restored to RPM_INVALID and the warning is absent.
>
> Fixes: 3e5eee147b7b ("PM: Block enabling of runtime PM during system suspend")
> Cc: stable@vger.kernel.org
> Assisted-by: Codex:gpt-5.6-sol
> Assisted-by: ChatGPT:GPT-5.6 Sol
> Signed-off-by: Shibo Zhu <3499129952@qq.com>
> ---
>   drivers/base/power/main.c | 1 +
>   1 file changed, 1 insertion(+)
>
> diff --git a/drivers/base/power/main.c b/drivers/base/power/main.c
> index f71467f6ada4..4ed2e9b7e99b 100644
> --- a/drivers/base/power/main.c
> +++ b/drivers/base/power/main.c
> @@ -2234,6 +2234,7 @@ static int device_prepare(struct device *dev, pm_message_t state)
>   
>   	if (ret < 0) {
>   		suspend_report_result(dev, callback, ret);
> +		pm_runtime_unblock(dev);
>   		pm_runtime_put(dev);
>   		return ret;
>   	}

LGTM.

Reviewed-by: Pengjie Zhang <zhangpengjie2@huawei.com>

> base-commit: 0f23d56f17fdfc7db69d51f64c8b91bbab947aa9