[PATCH 1/2] drivers/of: Add of_detach_node_no_notify()

[PATCH 1/2] drivers/of: Add of_detach_node_no_notify()

[Haren Myneni]

of_changeset_apply() interface adds nodes for action
OF_RECONFIG_ATTACH_NODE and calls notifiers for each attached node.
Then the caller uses of_detach_node() to release each node and
also calls the corresponding notifier function. In the case of
failure from the add notifier for any node, of_changeset_apply()
returns failure after adding nodes. The current implementation
does not the provide any interface to the caller to remove nodes
from the DT without calling notifiers. It may end up having nodes
in DT even though ADD is not successful.

This patch introduces of_detach_node_no_notify() interface for the
caller to only remove node if it is not attached.

Signed-off-by: Haren Myneni <haren@linux.ibm.com>
---
 drivers/of/dynamic.c | 16 ++++++++++++++++
 include/linux/of.h   |  1 +
 2 files changed, 17 insertions(+)

diff --git a/drivers/of/dynamic.c b/drivers/of/dynamic.c
index 744ce0e1eb24..fc6f22371a97 100644
--- a/drivers/of/dynamic.c
+++ b/drivers/of/dynamic.c
@@ -318,6 +318,22 @@ int of_detach_node(struct device_node *np)
 }
 EXPORT_SYMBOL_GPL(of_detach_node);
 
+/**
+ * of_detach_node_no_notify() - "Unplug" a node from the device
+ * and return without running notifiers.
+ * @np:	Pointer to the caller's Device Node
+ */
+int of_detach_node_no_notify(struct device_node *np)
+{
+	mutex_lock(&of_mutex);
+	if (!of_node_check_flag(np, OF_DETACHED))
+		__of_detach_node(np);
+	mutex_unlock(&of_mutex);
+
+	return 0;
+}
+EXPORT_SYMBOL_GPL(of_detach_node_no_notify);
+
 void __of_prop_free(struct property *prop)
 {
 	kfree(prop->name);
diff --git a/include/linux/of.h b/include/linux/of.h
index b920aac6b975..7a509341e998 100644
--- a/include/linux/of.h
+++ b/include/linux/of.h
@@ -442,6 +442,7 @@ extern int of_update_property(struct device_node *np, struct property *newprop);
 
 extern int of_attach_node(struct device_node *);
 extern int of_detach_node(struct device_node *);
+extern int of_detach_node_no_notify(struct device_node *);
 
 #define of_match_ptr(_ptr)	(_ptr)
 
-- 
2.55.0

[PATCH 2/2] powerpc/pseries/dlpar: Remove DT entries if failure from CPU ADD notifier

[Haren Myneni]

For CPU ADD, the device tree entries are retrieved with
configure-connector RTAS call and attached to the device tree.
Then the CPU is added as part of DT node notification. If the CPU
ADD notifier returns failure, the corresponding CPU node entries
should be deleted from the device-tree. See pseries_add_processor()
for the possible failure cases.

The current code does not remove CPU node entries during CPU ADD
notifier failure and causing the following issues:
- Can not add this CPU later since already present in the
  device-tree.
- Trying to free memory allocated to CPU node without detaching
  the node and it causes freeing its sibling node (existing CPU
  nodes).

This patch fixes this issue by calling of_detach_node_no_notify()
for the failure from CPU ADD notifier which removes CPU node
entries without calling CPU REMOVE notifier.

Signed-off-by: Haren Myneni <haren@linux.ibm.com>
---
 arch/powerpc/platforms/pseries/dlpar.c       | 9 ++++++---
 arch/powerpc/platforms/pseries/hotplug-cpu.c | 4 ++--
 arch/powerpc/platforms/pseries/mobility.c    | 2 +-
 arch/powerpc/platforms/pseries/pmem.c        | 2 +-
 arch/powerpc/platforms/pseries/pseries.h     | 2 +-
 5 files changed, 11 insertions(+), 8 deletions(-)

diff --git a/arch/powerpc/platforms/pseries/dlpar.c b/arch/powerpc/platforms/pseries/dlpar.c
index f4d33b8dffd8..01a9849773cd 100644
--- a/arch/powerpc/platforms/pseries/dlpar.c
+++ b/arch/powerpc/platforms/pseries/dlpar.c
@@ -247,15 +247,18 @@ int dlpar_attach_node(struct device_node *dn, struct device_node *parent)
 	return 0;
 }
 
-int dlpar_detach_node(struct device_node *dn)
+int dlpar_detach_node(struct device_node *dn, bool notify)
 {
 	struct device_node *child;
 	int rc;
 
 	for_each_child_of_node(dn, child)
-		dlpar_detach_node(child);
+		dlpar_detach_node(child, notify);
 
-	rc = of_detach_node(dn);
+	if (notify)
+		rc = of_detach_node(dn);
+	else
+		rc = of_detach_node_no_notify(dn);
 	if (rc)
 		return rc;
 
diff --git a/arch/powerpc/platforms/pseries/hotplug-cpu.c b/arch/powerpc/platforms/pseries/hotplug-cpu.c
index bc6926dbf148..db359fa3f922 100644
--- a/arch/powerpc/platforms/pseries/hotplug-cpu.c
+++ b/arch/powerpc/platforms/pseries/hotplug-cpu.c
@@ -603,7 +603,7 @@ static ssize_t dlpar_cpu_add(u32 drc_index)
 
 		rc = dlpar_release_drc(drc_index);
 		if (!rc)
-			dlpar_free_cc_nodes(dn);
+			dlpar_detach_node(dn, false);
 
 		return saved_rc;
 	}
@@ -616,7 +616,7 @@ static ssize_t dlpar_cpu_add(u32 drc_index)
 		pr_warn("Failed to online cpu %pOFn, rc: %d, drc index: %x\n",
 			dn, rc, drc_index);
 
-		rc = dlpar_detach_node(dn);
+		rc = dlpar_detach_node(dn, true);
 		if (!rc)
 			dlpar_release_drc(drc_index);
 
diff --git a/arch/powerpc/platforms/pseries/mobility.c b/arch/powerpc/platforms/pseries/mobility.c
index b5c2abd12432..e0e4ef04c679 100644
--- a/arch/powerpc/platforms/pseries/mobility.c
+++ b/arch/powerpc/platforms/pseries/mobility.c
@@ -112,7 +112,7 @@ static int delete_dt_node(struct device_node *dn)
 	}
 
 	pr_debug("removing node %pOFfp\n", dn);
-	dlpar_detach_node(dn);
+	dlpar_detach_node(dn, true);
 	return 0;
 }
 
diff --git a/arch/powerpc/platforms/pseries/pmem.c b/arch/powerpc/platforms/pseries/pmem.c
index 0f1d45f32e4a..b60a01745137 100644
--- a/arch/powerpc/platforms/pseries/pmem.c
+++ b/arch/powerpc/platforms/pseries/pmem.c
@@ -85,7 +85,7 @@ static ssize_t pmem_drc_remove_node(u32 drc_index)
 	pr_debug("Attempting to remove %pOF, drc index: %x\n", dn, drc_index);
 
 	/* * NB: tears down the ibm,pmemory device as a side-effect */
-	rc = dlpar_detach_node(dn);
+	rc = dlpar_detach_node(dn, true);
 	if (rc)
 		return rc;
 
diff --git a/arch/powerpc/platforms/pseries/pseries.h b/arch/powerpc/platforms/pseries/pseries.h
index 3968a6970fa8..6f80e58e6186 100644
--- a/arch/powerpc/platforms/pseries/pseries.h
+++ b/arch/powerpc/platforms/pseries/pseries.h
@@ -50,7 +50,7 @@ extern void dlpar_free_cc_property(struct property *);
 extern struct device_node *dlpar_configure_connector(__be32,
 						struct device_node *);
 extern int dlpar_attach_node(struct device_node *, struct device_node *);
-extern int dlpar_detach_node(struct device_node *);
+extern int dlpar_detach_node(struct device_node *, bool);
 extern int dlpar_acquire_drc(u32 drc_index);
 extern int dlpar_release_drc(u32 drc_index);
 extern int dlpar_unisolate_drc(u32 drc_index);
-- 
2.55.0

Re: [PATCH 2/2] powerpc/pseries/dlpar: Remove DT entries if failure from CPU ADD notifier

[Rob Herring]

On Tue, Sep 15, 2026 at 09:32:49PM -0700, Haren Myneni wrote:
> For CPU ADD, the device tree entries are retrieved with
> configure-connector RTAS call and attached to the device tree.
> Then the CPU is added as part of DT node notification. If the CPU
> ADD notifier returns failure, the corresponding CPU node entries
> should be deleted from the device-tree. See pseries_add_processor()
> for the possible failure cases.
> 
> The current code does not remove CPU node entries during CPU ADD
> notifier failure and causing the following issues:
> - Can not add this CPU later since already present in the
>   device-tree.
> - Trying to free memory allocated to CPU node without detaching
>   the node and it causes freeing its sibling node (existing CPU
>   nodes).
> 
> This patch fixes this issue by calling of_detach_node_no_notify()
> for the failure from CPU ADD notifier which removes CPU node
> entries without calling CPU REMOVE notifier.

Would using the of_changeset_ API directly solve your issue? It's better 
designed for handling reverting a changeset. The of_attach_node/ 
of_detach_node() APIs have limited users and I'd really like to remove 
them. Or make them PPC specific perhaps. Looks like there is 1 non-PPC 
user that snuck in.

Rob

Re: [PATCH 2/2] powerpc/pseries/dlpar: Remove DT entries if failure from CPU ADD notifier

[Haren Myneni]

On Wed, 2026-09-16 at 16:33 -0500, Rob Herring wrote:
> On Tue, Sep 15, 2026 at 09:32:49PM -0700, Haren Myneni wrote:
> > For CPU ADD, the device tree entries are retrieved with
> > configure-connector RTAS call and attached to the device tree.
> > Then the CPU is added as part of DT node notification. If the CPU
> > ADD notifier returns failure, the corresponding CPU node entries
> > should be deleted from the device-tree. See pseries_add_processor()
> > for the possible failure cases.
> > 
> > The current code does not remove CPU node entries during CPU ADD
> > notifier failure and causing the following issues:
> > - Can not add this CPU later since already present in the
> >   device-tree.
> > - Trying to free memory allocated to CPU node without detaching
> >   the node and it causes freeing its sibling node (existing CPU
> >   nodes).
> > 
> > This patch fixes this issue by calling of_detach_node_no_notify()
> > for the failure from CPU ADD notifier which removes CPU node
> > entries without calling CPU REMOVE notifier.
> 
> Would using the of_changeset_ API directly solve your issue? It's
> better 
> designed for handling reverting a changeset. The of_attach_node/ 
> of_detach_node() APIs have limited users and I'd really like to
> remove 
> them. Or make them PPC specific perhaps. Looks like there is 1 non-
> PPC 
> user that snuck in.

Thanks for your comments.

of_changeset_apply(): After adding all entries, calls notifiers for
each node and continue even after one notifier  failure and returns
failure. The caller does not know whether the failure is from attach
node or notifier and from which node. So
__of_changeset_revert_entries() can be used to revert entries if
stopped on the first failure.

of_changeset_revert(): always calls notifier for each node. I can use
__of_changeset_revert_entries() since CPU add has just one notifier but
not a standard interface.

pseries_cpuhp_attach_nodes():

if (!of_node_check_flag(np, OF_DETACHED))
	__of_changeset_revert_entries(&cs, *ret);

The current code calls dlpar_detach_node() if online CPU failed after
success from of_changeset_apply(). Hence added
of_detach_node_no_notify(). PPC code calls of_changeset_apply() for
multiple nodes and of_attach_node/of_detach_node for single node. 

 Or defining some other node flag OF_NOTIFIED to call the notifier for
revert / detach in of_changeset_*. 
 
Please let me know if you prefer to change of_changeset_*() instead of
of_detach_node_no_notify(0.

- Haren

> 
> 
> Rob