[PATCH] clk: ti: Simplify ti_find_clock_provider()

[PATCH] clk: ti: Simplify ti_find_clock_provider()

[Rob Herring (Arm)]

Remove using for_each_of_allnodes_from() which is not safe to use
without holding the DT spinlock. In reality that probably doesn't
matter here. This is the only user in the whole tree, so it can be
made private once removed here. The "from" argument is always NULL, so
it can be dropped as well.

There's a slight change in behavior in matching the "clock-output-names"
value as the prior code would match if the node name matched the
beginning of the value and the comparision was case insensitive. Now
it must be an exact match.

Signed-off-by: Rob Herring (Arm) <robh@kernel.org>
---
Compiled only. I'm not sure if the the change in behavior is going to 
matter. 

 drivers/clk/ti/clk.c | 27 ++++++---------------------
 1 file changed, 6 insertions(+), 21 deletions(-)

diff --git a/drivers/clk/ti/clk.c b/drivers/clk/ti/clk.c
index 9c75dcc9a534..693a4459a01b 100644
--- a/drivers/clk/ti/clk.c
+++ b/drivers/clk/ti/clk.c
@@ -118,13 +118,10 @@ int ti_clk_setup_ll_ops(struct ti_clk_ll_ops *ops)
  * Eventually we could standardize to using '_' for clk-*.c files to follow the
  * TRM naming.
  */
-static struct device_node *ti_find_clock_provider(struct device_node *from,
-						  const char *name)
+static struct device_node *ti_find_clock_provider(const char *name)
 {
 	char *tmp __free(kfree) = NULL;
 	struct device_node *np;
-	bool found = false;
-	const char *n;
 	char *p;
 
 	tmp = kstrdup_and_replace(name, '-', '_', GFP_KERNEL);
@@ -137,25 +134,13 @@ static struct device_node *ti_find_clock_provider(struct device_node *from,
 		*p = '\0';
 
 	/* Node named "clock" with "clock-output-names" */
-	for_each_of_allnodes_from(from, np) {
-		if (of_property_read_string_index(np, "clock-output-names",
-						  0, &n))
-			continue;
-
-		if (!strncmp(n, tmp, strlen(tmp))) {
-			of_node_get(np);
-			found = true;
-			break;
-		}
-	}
-
-	if (found) {
-		of_node_put(from);
-		return np;
+	for_each_node_with_property(np, "clock-output-names") {
+		if (of_property_match_string(np, "clock-output-names", tmp) == 0)
+			return np;
 	}
 
 	/* Fall back to using old node name base provider name */
-	return of_find_node_by_name(from, tmp);
+	return of_find_node_by_name(NULL, tmp);
 }
 
 /**
@@ -208,7 +193,7 @@ void __init ti_dt_clocks_register(struct ti_dt_clk oclks[])
 		if (num_args && clkctrl_nodes_missing)
 			continue;
 
-		node = ti_find_clock_provider(NULL, buf);
+		node = ti_find_clock_provider(buf);
 		if (num_args && compat_mode) {
 			parent = node;
 			child = of_get_child_by_name(parent, "clock");
-- 
2.47.2

Re: [PATCH] clk: ti: Simplify ti_find_clock_provider()

[Andreas Kemnade]

Am Wed, 12 Mar 2025 11:33:30 -0500
schrieb "Rob Herring (Arm)" <robh@kernel.org>:

> Remove using for_each_of_allnodes_from() which is not safe to use
> without holding the DT spinlock. In reality that probably doesn't
> matter here. This is the only user in the whole tree, so it can be
> made private once removed here. The "from" argument is always NULL, so
> it can be dropped as well.
> 
> There's a slight change in behavior in matching the "clock-output-names"
> value as the prior code would match if the node name matched the
> beginning of the value and the comparision was case insensitive. Now
> it must be an exact match.
> 
> Signed-off-by: Rob Herring (Arm) <robh@kernel.org>
> ---
> Compiled only. I'm not sure if the the change in behavior is going to 
> matter. 
> 
did some testing. Nothing odd seen at the platforms I tested but I feel
a bit uncomfortable with
https://lore.kernel.org/linux-clk/20240925100603.4cba9176@akair/
in combination with this.

Regards,
Andreas

Re: [PATCH] clk: ti: Simplify ti_find_clock_provider()

[Stephen Boyd]

Quoting Rob Herring (Arm) (2025-03-12 09:33:30)
> Remove using for_each_of_allnodes_from() which is not safe to use
> without holding the DT spinlock. In reality that probably doesn't
> matter here. This is the only user in the whole tree, so it can be
> made private once removed here. The "from" argument is always NULL, so
> it can be dropped as well.
> 
> There's a slight change in behavior in matching the "clock-output-names"
> value as the prior code would match if the node name matched the
> beginning of the value and the comparision was case insensitive. Now
> it must be an exact match.
> 
> Signed-off-by: Rob Herring (Arm) <robh@kernel.org>
> ---

Applied to clk-next