On 07/22/2015 01:16 AM, Vaibhav Hiremath wrote:

On Wednesday 22 July 2015 12:16 PM, Krzysztof Kozlowski wrote:

quoted

I am really busy now so I am not following closely other discussions. I
assume you are referring to clk-s2mps11.c. The of_node_put() matches
of_get_child_by_name() when parsing DT.

So why not of_node_put() just after parsing DT? Well, the result of
of_get_child_by_name() is stored in state container for entire device
life-cycle so we can use it in of_clk_del_provider().

That was the idea behind it. If it looks incorrect I would be happy to
see a patch :) .

About to respond, I digged more on kobject stuff and sequence in
of/dynamic.c and

I think you are right, we need of_node_put, as a result of
of_get_child_by_name().

Stephen,
Please let me know if you think otherwise.

Yes, sounds fine. I was thinking that we grab the reference to the node 
in of_clk_add_provider() so dropping it here was to undo that, but that 
isn't true. It probably can be dropped after we register the provider 
because adding the provider will keep it pinned, but this way is more 
symmetric so it's fine.

Either way, the error path on probe doesn't call of_node_put(), so 
that's a leak.

-- 
Qualcomm Innovation Center, Inc. is a member of Code Aurora Forum,
a Linux Foundation Collaborative Project