Re: [PATCH v4 16/23] ARM: OMAP2+: clock data: Merge utmi_px_gfclk into usb_host_hs_utmi_px_clk

[Benoit Cousson <b-cousson@ti.com>]

On 12/17/2012 05:13 PM, Roger Quadros wrote:
> On 12/17/2012 10:13 AM, Benoit Cousson wrote:
>> Hi,
>>
>> On 12/14/2012 07:44 PM, Paul Walmsley wrote:
>>> Hi
>>>
>>> On Fri, 14 Dec 2012, Tony Lindgren wrote:
>>>
>>>> Paul, what about this patch? Looks like you've acked the other clock 
>>>> patches in this series but not this one?
>>>
>>> I commented on it briefly here:
>>>
>>> https://patchwork.kernel.org/patch/1838111/
>>>
>>> Maybe Benoît could comment here, but it looks to me (based on a 
>>> superficial look at the hardware clock tree data) that these clock nodes 
>>> should exist.  In an ideal world, we'd be able to get back to the 
>>> autogeneration of this clock data.
>>
>> I'm not sure to understand either the rational for that patch. What the
>> point of merging the two nodes?
>> I mean, we can do it, but AFAIR, we have always decided to use atomic
>> node instead of big nodes that handle everything.
>>
> 
> I can see a similar thing done for mcbsp clocks (e.g. /* Merged
> func_mcbsp1_gfclk into mcbsp1 */), mmc clocks, timer clocks, mcasp
> clock, and sgx clock. i.e. The clock sel (mux) is combined with clock
> gate. I don't see why USB host has to be done differently.

Hehe, well, in fact USB is using the right approach, the others are the
exceptions :-)

It was done for legacy reason but should disappear once the modulemode
will be be removed from the clock nodes.

> Were exceptions made for the above clocks in the auto generation code?
> 
> The problem from driver point of view is that it has to manage an
> additional clock per port. Not a big deal, but I thought it could be
> avoided.

In theory, the driver should just managed the mux. The modulemode being
managed already by hwmod.

Regards,
Benoit