On Thu, 2 Sept 2021 at 00:06, Saravana Kannan [off-list ref] wrote:

On Wed, Sep 1, 2021 at 2:27 PM Ulf Hansson [off-list ref] wrote:

quoted

On Wed, 1 Sept 2021 at 22:56, Saravana Kannan [off-list ref] wrote:

quoted

On Wed, Sep 1, 2021 at 12:45 AM Ulf Hansson [off-list ref] wrote:

quoted

On Tue, 31 Aug 2021 at 19:31, Saravana Kannan [off-list ref] wrote:

quoted

On Tue, Aug 31, 2021 at 3:21 AM Ulf Hansson [off-list ref] wrote:

quoted

In the struct supplier_bindings the member 'node_not_dev' is described as
"The consumer node containing the property is never a device.", but that
doesn't match the behaviour of the code in of_link_property().

To make the behaviour consistent with the description, let's rename the
member to "optional_con_dev" and clarify the corresponding comment.

Signed-off-by: Ulf Hansson <redacted>
---
 drivers/of/property.c | 9 +++++----
 1 file changed, 5 insertions(+), 4 deletions(-)

diff --git a/drivers/of/property.c b/drivers/of/property.c
index 6c028632f425..2babb1807228 100644
--- a/drivers/of/property.c
+++ b/drivers/of/property.c

@@ -1249,7 +1249,8 @@ static struct device_node *parse_##fname(struct device_node *np,       \
  * @parse_prop.index: For properties holding a list of phandles, this is the
  *                   index into the list
  * @optional: Describes whether a supplier is mandatory or not
- * @node_not_dev: The consumer node containing the property is never a device.
+ * @optional_con_dev: The consumer node containing the property may not be a
+ *                   device, then try finding one from an ancestor node.

Nak. This flag is not about "may not be". This is explicitly for
"never a device". It has to do with stuff like remote-endpoint which
is never listed under the root node of the device node. Your
documentation change is changing the meaning of the flag.

Okay, fair enough.

Although, as stated in the commit message this isn't the way code
behaves. Shouldn't we at least make the behaviour consistent with the
description of the 'node_not_dev' flag?

I know what you mean, but if you use the flag correctly (where the
phandle pointed to will never be a device with compatible property),
the existing code would work correctly. And since the flag is relevant
only in this file, it's easy to keep it correct. I'd just leave it as
is.

Sorry, but that just sounds lazy to me, I am sure we can do better.
The current code and the name of the flag is confusing, at least to me
(and I bet to others as well).

Moreover, I don't quite understand your objections to changing this.
Why leave this to be inconsistent when it can be easily fixed?

If you feel so strong about it, go for it. No strong objections. Just
double check the refcounts are done correctly.

The refcounts should be okay, I think.

I am fine with either of the two suggestions I have made. But another
option could be to come up with an alternative name (and a
description) for the flag, instead of "optional_con_dev", if you
perhaps have a better suggestion?

-Saravana

Kind regards
Uffe

_______________________________________________
linux-arm-kernel mailing list
linux-arm-kernel@lists.infradead.org
http://lists.infradead.org/mailman/listinfo/linux-arm-kernel