Re: [PATCH 1/3] clk: Add new function of_clk_is_provider()

Hi Stephen

When the device tree is populated or when an overlay is added, all its
nodes have the flag OF_POPULATED set. The flag is enabled recursively
in
of_platform_bus_create->of_platform_device_create_pdata()
So we cannot use that flag to mark what is enabled and what is not.

The other issue that I see is of_clk_mutex. Whatever final
implementation that we decide to do, it should take into consideration
that mutex, otherwise it will not be thread-safe.
of_clk_is_provider() is already taking care of it.

Another advantage of of_clk_is_provider() is that it opens the door to
implement something like:  CLK_OF_DECLARE_EARLY_PLATFORM(probe,remove)
 That allows a driver to implement early clk and platform clk at the
same time and automatically, following a logic similar to what I have
done in fixed-clk.

I will send a v2 with the changes you proposed to fixed-clk

Thanks and best regards!

On Thu, Jun 16, 2016 at 2:44 AM, Stephen Boyd [off-list ref] wrote:

On 06/08, Ricardo Ribalda Delgado wrote:

quoted

of_clk_is_provider() checks if a device_node has already been added to
the clk provider list. This can be used to avoid adding the same clock
provider twice.

Signed-off-by: Ricardo Ribalda Delgado <redacted>

While I don't disagree with the concept, I'd like to do this
outside of the clk framework checking for nodes, because the
problem doesn't seem clk specific. From digging in the OF
platform layer I see of_node_test_and_set_flag(OF_POPULATED) may
be what we should be using. It looks like this can be used to
make sure that any clk provider nodes aren't populated as
platform devices when we've initialized them early.

The only problem now is that we have drivers using a hybrid
approach with of_clk_init(). Sometimes drivers need to get clks
up early for timers, so they have CLK_OF_DECLARE() in their
driver, but then they also use a platform driver to handle the
non-timer related clks. If we mark all nodes as populated in
of_clk_init() we'll preclude these drivers from working. The
solution there is to make those drivers specifically clear the
populated flag in the clk init callback. Or we can automatically
do that with some new CLK_OF_DECLARE_EARLY() macro that hides
this clearing from them. Either way, the drivers will need to
indicate they're using this hybrid style so that we still
populate platform devices.

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

-- 
Ricardo Ribalda