[PATCH] PCI/pwrctrl: Power off child devices on power-on failure

[PATCH] PCI/pwrctrl: Power off child devices on power-on failure

[Lorenzo Bianconi]

pci_pwrctrl_power_on_device() recursively powers on the child nodes
of a device before powering on the device itself. If powering on a
child fails (e.g. returning -EPROBE_DEFER), the function returned
immediately, leaving the sibling children that were already powered
on in the same loop iteration enabled.

Since probe deferral is a standard boot path, this leads to unbounded
regulator and clock enable increments across retries. The error path
in pci_pwrctrl_power_on_devices() only powers off the top-level
children processed before the failing one and explicitly stops at it,
so it can never clean up the siblings leaked deeper in the failed
subtree.

Power off the sibling children that were successfully powered on
before a child fails, and power off all children if the device's own
power_on() fails after its children were already enabled. Switch to a
manual child iterator so the failing node can be tracked and its
reference released in the error path.

Fixes: b35cf3b6aa1e ("PCI/pwrctrl: Add APIs to power on/off pwrctrl devices")
Signed-off-by: Lorenzo Bianconi <lorenzo.bianconi@oss.qualcomm.com>
---
 drivers/pci/pwrctrl/core.c | 19 +++++++++++++++++--
 1 file changed, 17 insertions(+), 2 deletions(-)

diff --git a/drivers/pci/pwrctrl/core.c b/drivers/pci/pwrctrl/core.c
index 10d817aefb09..d29cd506972d 100644
--- a/drivers/pci/pwrctrl/core.c
+++ b/drivers/pci/pwrctrl/core.c
@@ -252,14 +252,16 @@ static int __pci_pwrctrl_power_on_device(struct device *dev)
 static int pci_pwrctrl_power_on_device(struct device_node *np)
 {
 	struct platform_device *pdev;
+	struct device_node *child;
 	int ret = 0;
 
-	for_each_available_child_of_node_scoped(np, child) {
+	for_each_available_child_of_node(np, child) {
 		ret = pci_pwrctrl_power_on_device(child);
 		if (ret)
-			return ret;
+			goto err_power_off;
 	}
 
+	child = NULL;
 	if (!pci_pwrctrl_is_required(np))
 		return 0;
 
@@ -279,6 +281,19 @@ static int pci_pwrctrl_power_on_device(struct device_node *np)
 
 	platform_device_put(pdev);
 
+	if (ret)
+		goto err_power_off;
+
+	return 0;
+
+err_power_off:
+	for_each_available_child_of_node_scoped(np, tmp) {
+		if (tmp == child)
+			break;
+		pci_pwrctrl_power_off_device(tmp);
+	}
+	of_node_put(child);
+
 	return ret;
 }
 

---
base-commit: 028f86457e822ab9de5b0d6d23efddc1b8b54ae1
change-id: 20260908-pwrtctl-power-on-rallback-5c6364b080ce

Best regards,
-- 
Lorenzo Bianconi <lorenzo.bianconi@oss.qualcomm.com>

Re: [PATCH] PCI/pwrctrl: Power off child devices on power-on failure

[sashiko-bot]

> pci_pwrctrl_power_on_device() recursively powers on the child nodes
> of a device before powering on the device itself. If powering on a
> child fails (e.g. returning -EPROBE_DEFER), the function returned
> immediately, leaving the sibling children that were already powered
> on in the same loop iteration enabled.
> 
> Since probe deferral is a standard boot path, this leads to unbounded
> regulator and clock enable increments across retries. The error path
> in pci_pwrctrl_power_on_devices() only powers off the top-level
> children processed before the failing one and explicitly stops at it,
> so it can never clean up the siblings leaked deeper in the failed
> subtree.
> 
> Power off the sibling children that were successfully powered on
> before a child fails, and power off all children if the device's own
> power_on() fails after its children were already enabled. Switch to a
> manual child iterator so the failing node can be tracked and its
> reference released in the error path.
> 
> Fixes: b35cf3b6aa1e ("PCI/pwrctrl: Add APIs to power on/off pwrctrl devices")
> Signed-off-by: Lorenzo Bianconi <lorenzo.bianconi@oss.qualcomm.com>

Sashiko has reviewed this patch and found no issues. It looks great!

-- 
Sashiko AI review · https://sashiko.dev/#/patchset/20260909-pwrtctl-power-on-rallback-v1-1-412ef996ba27@oss.qualcomm.com?part=1

Re: [PATCH] PCI/pwrctrl: Power off child devices on power-on failure

[Manivannan Sadhasivam]

On Wed, Sep 09, 2026 at 05:28:29PM +0200, Lorenzo Bianconi wrote:
> pci_pwrctrl_power_on_device() recursively powers on the child nodes
> of a device before powering on the device itself. If powering on a
> child fails (e.g. returning -EPROBE_DEFER), the function returned
> immediately, leaving the sibling children that were already powered
> on in the same loop iteration enabled.
> 
> Since probe deferral is a standard boot path, this leads to unbounded
> regulator and clock enable increments across retries. The error path
> in pci_pwrctrl_power_on_devices() only powers off the top-level
> children processed before the failing one and explicitly stops at it,
> so it can never clean up the siblings leaked deeper in the failed
> subtree.
> 
> Power off the sibling children that were successfully powered on
> before a child fails, and power off all children if the device's own
> power_on() fails after its children were already enabled. Switch to a
> manual child iterator so the failing node can be tracked and its
> reference released in the error path.
> 
> Fixes: b35cf3b6aa1e ("PCI/pwrctrl: Add APIs to power on/off pwrctrl devices")
> Signed-off-by: Lorenzo Bianconi <lorenzo.bianconi@oss.qualcomm.com>

Reviewed-by: Manivannan Sadhasivam <manivannan.sadhasivam@oss.qualcomm.com>

- Mani

-- 
மணிவண்ணன் சதாசிவம்

Re: [PATCH] PCI/pwrctrl: Power off child devices on power-on failure

[Bartosz Golaszewski]

On Wed, 9 Sep 2026 17:28:29 +0200, Lorenzo Bianconi
<lorenzo.bianconi@oss.qualcomm.com> said:
> pci_pwrctrl_power_on_device() recursively powers on the child nodes
> of a device before powering on the device itself. If powering on a
> child fails (e.g. returning -EPROBE_DEFER), the function returned
> immediately, leaving the sibling children that were already powered
> on in the same loop iteration enabled.
>
> Since probe deferral is a standard boot path, this leads to unbounded
> regulator and clock enable increments across retries. The error path
> in pci_pwrctrl_power_on_devices() only powers off the top-level
> children processed before the failing one and explicitly stops at it,
> so it can never clean up the siblings leaked deeper in the failed
> subtree.
>
> Power off the sibling children that were successfully powered on
> before a child fails, and power off all children if the device's own
> power_on() fails after its children were already enabled. Switch to a
> manual child iterator so the failing node can be tracked and its
> reference released in the error path.
>
> Fixes: b35cf3b6aa1e ("PCI/pwrctrl: Add APIs to power on/off pwrctrl devices")
> Signed-off-by: Lorenzo Bianconi <lorenzo.bianconi@oss.qualcomm.com>
> ---

Reviewed-by: Bartosz Golaszewski <bartosz.golaszewski@oss.qualcomm.com>