Re: [PATCH v9 08/12] clk: meson: g12a: make VCLK2 and ENCL clock path configurable by CCF

[Date Prev][Date Next][Thread Prev][Thread Next][Date Index][Thread Index]

 



>> 
>>>
>>> I suspect mipi_dsi_pxclk_div was added to achieve fractional vclk/bitclk ratios,
>>> since it doesn't exist on AXG. Not sure we would ever need it... and none
>>> of the other upstream DSI drivers supports such setups.
>>>
>>> The main reasons I set only mipi_dsi_pxclk in DT is because :
>>> 1) the DSI controller requires a bitclk to respond, pclk is not enough
>>> 2) GP0 is disabled with an invalid config at cold boot, thus we cannot
>>> rely on a default/safe rate on an initial prepare_enable().
>>> This permits setting initial valid state for the DSI controller, while
>>> the actual bitclk and vclk are calculated dynamically with panel/bridge
>>> runtime parameters.
>> Nothing against setting rate in DT when it is static. Setting it then
>> overriding it is not easy to follow.
>
> Yup, would be simpler to only have parenting set in DT, since it must
> stay static, I'm fine trying to move rate setup to code.
>
>> To work around GP0 not being set, assuming you want to keep rate
>> propagation as it is, you could call clk_set_rate() on cts_encl (possibly w/o
>> enabling it) to force a setup on gp0 then clk_prepare_enable() on
>> pxclk. You'd get a your safe rate on GP0 and the clock you need on pxclk.
>> It is a bit hackish. Might be better to claim gp0 in your driver to
>> manage it directly, cutting rate propagation above it to control each
>> branch of the subtree as you need. It seems you need to have control over
>> that anyway and it would be clear GP0 is expected to belong to DSI.
>
> Controlling the PLL from the DSI controller seems violating too much layers,
> DSI controller driver is not feed directly by the PLL so it's a non-sense
> regarding DT properties.

Not sure what you mean by this. You have shown in your the commit
message that the DSI clocks make significant subtree. I don't see a
problem if you need to manage the root of that subtree. I'd be great if
you didn't need to, but it is what it is ...

>
> Setting a safe clock from the DSI controller probe is an idea, but again I
> don't know which value I should use...

You mentionned that the problem comes DSI bridges that needs to change
at runtime. I don't know much about those TBH, but is there
anyway you can come up with a static GP0 rate that would then be able to
divide to serve all the rates bridge would need in your use case ?

GP0 can go a lot higher than ~100MHz and there are dividers unused in the
tree it seems.

I suppose there is a finite number of required rate for each use case ?
If there are not too many and there is a common divider that allows a
common rate GP0 can do, it would solve your problem. It's a lot of if
but It is worth checking.

This is how audio works and DT assigned rate is a good match in this case.

>
> I'll review the clk parenting flags and try to hack something.
>
> Thanks,
> Neil
>
>




[Index of Archives]     [Device Tree Compilter]     [Device Tree Spec]     [Linux Driver Backports]     [Video for Linux]     [Linux USB Devel]     [Linux PCI Devel]     [Linux Audio Users]     [Linux Kernel]     [Linux SCSI]     [XFree86]     [Yosemite Backpacking]


  Powered by Linux