Re: [PATCH] media: ipu3: add a module to probe sensors via ACPI

[Dan Scally <djrscally@gmail.com>]

Hi Heikki

On 26/05/2020 15:31, Heikki Krogerus wrote:
> On Fri, May 22, 2020 at 11:57:36AM +0200, Mauro Carvalho Chehab wrote:
>> Em Thu, 21 May 2020 11:00:19 +0300
>> Andy Shevchenko <andy.shevchenko@gmail.com> escreveu:
>>
>>> +Cc: Heikki (swnode expert)
>>>
>>> On Wed, May 20, 2020 at 2:19 PM Mauro Carvalho Chehab
>>> <mchehab+huawei@kernel.org> wrote:
>>>> Em Wed, 20 May 2020 11:26:08 +0300
>>>> Sakari Ailus <sakari.ailus@linux.intel.com> escreveu:
>>>
>>> ...
>>>
>>>> As I said, the problem is not probing the sensor via ACPI, but, instead,
>>>> to be able receive platform-specific data.
>>>
>>> There is no problem with swnodes, except missing parts (*).
>>> I have Skylake laptop with IPU3 and with half-baked ACPI tables, but
>>> since we have drivers in place with fwnode support, we only need to
>>> recreate fwnode graph in some board file to compensate the gap in
>>> ACPI.
>>>
>>> *) Missing part is graph support for swnodes. With that done it will
>>> be feasible to achieve the rest.
>>> I forgot if we have anything for this already done. Heikki?
>>
>> Hmm... I guess I should try this approach. I never heard about swnodes
>> before. Do you have already some patch with the needed swnodes setup,
>> and the missing parts to recreate the fwnode graph?
> 
> Here you go. I tested it with this code:
> 
>          static const struct software_node nodes[];
> 
>          static const struct property_entry ep0_props[] = {
>                 PROPERTY_ENTRY_REF("remote-endpoint", &nodes[5]),
>                 { }
>          };
> 
>          static const struct property_entry ep1_props[] = {
>                 PROPERTY_ENTRY_REF("remote-endpoint", &nodes[2]),
>                 { }
>          };
> 
>          static const struct software_node nodes[] = {
>                 { "dev0" },
>                 { "port0", &nodes[0] },
>                 { "endpoint", &nodes[1], ep0_props },
>                 { "dev1" },
>                 { "port0", &nodes[3] },
>                 { "endpoint", &nodes[4], ep1_props },
>                 { }
>          };
> 
>          void test(void)
>          {
>                  const struct software_node *swnode;
>                  struct fwnode_handle *fwnode;
> 
>                  software_node_register_nodes(nodes);
> 
>                  fwnode = fwnode_graph_get_remote_port_parent(software_node_fwnode(&nodes[5]));
>                  swnode = to_software_node(fwnode);
>                  printk("first parent: %s\n", swnode->name);
> 
>                  fwnode = fwnode_graph_get_remote_port_parent(software_node_fwnode(&nodes[2]));
>                  swnode = to_software_node(fwnode);
>                  printk("second parent: %s\n", swnode->name);
> 
>                  software_node_unregister_nodes(nodes);
>          }
> 
> thanks,
> 

One of the problems we're having trying to build (using the changes you 
attached here) a module to connect sensors to the cio2 infrastructure is 
that we can't unload it cleanly. There seems to be a couple of reasons 
for that; but one of them is that cio2_parse_firmware() in ipu3-cio2.c 
ticks up the refcount for fwnode_handles of the ports for the CIO2 
device by calling software_node_graph_get_next_endpoint() once per 
_possible_ cio2 port; each time that happens it gets a reference to the 
port's fwnode_handle but doesn't release it.

This isn't really a patch as such, since I don't think the changes you 
attached are actually applied either upstream or in the media_tree git 
(what are the plans in that regard, by the way? Will that patch be sent 
upstream at some point?) so there's nowhere to apply it to, but I think 
something like the below fixes it.

What do you think?

Regards,
Dan

---
diff --git a/drivers/base/swnode.c b/drivers/base/swnode.c
index 3667467196f0..62a1e3de8cb3 100644
--- a/drivers/base/swnode.c
+++ b/drivers/base/swnode.c
@@ -584,7 +584,9 @@ software_node_graph_get_next_endpoint(const struct 
fwnode_handle *fwnode,
                 endpoint = software_node_get_next_child(port, old);
                 fwnode_handle_put(old);
                 if (endpoint)
-                       break;
+                       break;
+               else
+                       fwnode_handle_put(port);
         }

         fwnode_handle_put(port);