Re: [PATCH] soc: fsl: qe: properly scan GPIO nodes at startup

Hi Christophe,

On Thu, 30 Jul 2026 10:11:29 +0200
"Christophe Leroy (CS GROUP)" [off-list ref] wrote:

Hi Hervé,

Le 29/07/2026 à 17:21, Herve Codina a écrit :

quoted

Hi Christophe,

On Wed, 29 Jul 2026 16:10:36 +0200
"Christophe Leroy (CS GROUP)" [off-list ref] wrote:
  

quoted

Before commit 156460811def ("soc: fsl: qe: Change GPIO driver to a
proper platform driver") qe_add_gpiochips() was walking the device
tree to find all nodes with compatible "fsl,mpc8323-qe-pario-bank".

After that commit the discovery is handled by the platform core,
therefore it is necessary to call of_platform_default_populate() on
the par_io node.

Fixes: 156460811def ("soc: fsl: qe: Change GPIO driver to a proper platform driver")
Signed-off-by: Christophe Leroy (CS GROUP) <chleroy@kernel.org>
---
  drivers/soc/fsl/qe/qe_io.c | 15 +++++++++++++++
  1 file changed, 15 insertions(+)

diff --git a/drivers/soc/fsl/qe/qe_io.c b/drivers/soc/fsl/qe/qe_io.c
index a5e2d0e5ab51..02ca556c8db0 100644
--- a/drivers/soc/fsl/qe/qe_io.c
+++ b/drivers/soc/fsl/qe/qe_io.c

@@ -15,6 +15,7 @@
  #include <linux/errno.h>
  #include <linux/module.h>
  #include <linux/ioport.h>
+#include <linux/of_platform.h>
  
  #include <asm/io.h>
  #include <soc/fsl/qe/qe.h>

@@ -184,3 +185,17 @@ int par_io_of_config(struct device_node *np)
  	return 0;
  }
  EXPORT_SYMBOL(par_io_of_config);
+
+static int __init par_io_populate(void)
+{
+	struct device_node *np = of_find_node_by_name(NULL, "par_io");
+
+	if (!np)
+		return 0;
+
+	of_platform_default_populate(np, NULL, NULL);
+	of_node_put(np);
+
+	return 0;
+}
+arch_initcall(par_io_populate);  

The par_io node is a sub-node of the soc@xxx node.
The soc@xxx node is simple-bus compatible. This child node should be
populated.

Why do we need this additional of_platform_default_populate?

Also par_io node has a compatible. Maybe the related driver should call
of_platform_populate() to populate children of par_io if needed. If this
driver is missing, maybe it should be created.

Do I miss something?  

As discussed together off-line, at the time being the par_io driver is 
not a platform driver. par_io_init() is called from setup_arch() by some 
board setups.

In order to fix the regression, I keep it minimal with an 
arch_initcall() function which is equivalent to what we had before 
changing this driver to a platform driver. It will also ease applying 
the fix to stable without too much churn.

It will still be possible to rework par_io driver in a follow-up patch 
and change it to a propoer platform driver.

Yes, indeed, this makes sense.

Thanks for the clarification.

Best regards,
Hervé