Re: [PATCH] arm64: dts: imx8mp: Improve VPU clock configuration

[Marek Vasut <marex@denx.de>]

On 8/10/23 14:46, Adam Ford wrote:
> On Thu, Aug 10, 2023 at 2:42 AM Marco Felsch <m.felsch@pengutronix.de> wrote:
>>
>> Hi Marek,
>>
>> On 23-08-10, Marek Vasut wrote:
>>> Update VPU clock according to reference manual to improve performance.
>>> i.MX 8M Plus Applications Processor Reference Manual, Rev. 1, 06/2021
>>>
>>> Table 5-1. Clock Root Table (continued) reads as follows:
>>> Clock Root ............... Max Freq (MHz)
>>> VPU_BUS_CLK_ROOT ......... 800
>>> VPU_G1_CLK_ROOT .......... 800
>>> VPU_G2_CLK_ROOT .......... 700
>>>
>>> 5.1.5.1 Input Clocks
>>> Input Clock .............. Frequency (MHz)
>>> VPU_PLL_CLK .............. 800
>>
>> Acoording the datasheet not the reference manual, this would be the
>> overdrive mode now. Can we ensure that VDD_SOC is in overdrive state?
> 
> Shoot, I explicitly searched the TRM for overdrive, because I remember
> this coming before with either the 8MQ or 8MM, but I couldn't
> remember.  I forgot all about the actual datasheet.

Same here, I vaguely recall there might've been a reason why those lower 
freqs, but I couldn't find it either.

> The operating-points for the various voltage rails would have to
> increase to get this into overdrive mode.  I think this would have to
> be done on a per-board basis if memory serves when this came up
> before.  :-(

I don't mind doing it per board. Actually, I don't mind dropping this 
patch, since the decoding performance increase is not even an issue in 
any of the use cases I care about.

I guess the general agreement here is to go with the "nominal mode" by 
default and enable "overdrive mode" per-board , while the NXP downstream 
goes full "overdrive mode" all the time ?