On 15-07-2026 18:44, Bartosz Golaszewski wrote:

On Tue, 14 Jul 2026 19:16:37 +0200, Shivendra Pratap
[off-list ref] said:

quoted

The PSCI "reboot-mode" node does not define a compatible because it is a
configuration of boot-states provided by the underlying firmware. With
the new firmware-node based cells in mfd-core, this node can now be
exposed as a proper child cell.

Add the psci-reboot-mode child cell to the psci-mfd driver with a
named_fwnode. Add psci-cpuidle-domain cell first to isolate it from
reboot-mode failures.

Suggested-by: Bartosz Golaszewski <redacted>
Signed-off-by: Shivendra Pratap <redacted>
---
  drivers/mfd/psci-mfd.c      | 30 +++++++++++++++++++++++++++++-
  drivers/power/reset/Kconfig |  2 +-
  2 files changed, 30 insertions(+), 2 deletions(-)

diff --git a/drivers/mfd/psci-mfd.c b/drivers/mfd/psci-mfd.c
index 7affd6bb09dd..8c6b78ac471b 100644
--- a/drivers/mfd/psci-mfd.c
+++ b/drivers/mfd/psci-mfd.c

@@ -7,6 +7,7 @@
  #include <linux/module.h>
  #include <linux/of.h>
  #include <linux/platform_device.h>
+#include <linux/property.h>

  static const struct mfd_cell psci_cells[] = {
  	{

@@ -14,10 +15,37 @@ static const struct mfd_cell psci_cells[] = {
  	},
  };

+static const struct mfd_cell psci_reboot_mode_cell[] = {
+	{
+		.name = "psci-reboot-mode",
+		.named_fwnode = "reboot-mode",
+	},
+};
+
  static int psci_mfd_probe(struct platform_device *pdev)
  {
-	return devm_mfd_add_devices(&pdev->dev, PLATFORM_DEVID_AUTO, psci_cells,
+	struct fwnode_handle *fwnode;
+	int ret;
+
+	ret = devm_mfd_add_devices(&pdev->dev, PLATFORM_DEVID_AUTO, psci_cells,
  				   ARRAY_SIZE(psci_cells), NULL, 0, NULL);
+	if (ret)
+		return ret;
+
+	fwnode = device_get_named_child_node(&pdev->dev, "reboot-mode");
+	if (!fwnode)
+		return 0;
+
+	fwnode_handle_put(fwnode);
+
+	ret = devm_mfd_add_devices(&pdev->dev, PLATFORM_DEVID_AUTO,
+				   psci_reboot_mode_cell,
+				   ARRAY_SIZE(psci_reboot_mode_cell),
+				   NULL, 0, NULL);
+	if (ret)
+		dev_warn(&pdev->dev, "reboot-mode child cell failed to add: %d\n", ret);

Why do we have to look up the named node here anyway? Shouldn't MFD core handle
that already in a generic way?

yes. MFD is handling this. Will remove this explicit check.

thanks,
Shivendra