Re: [PATCH 1/1] clk: aspeed: modify some default clks are critical

[Joel Stanley]

Thanks for the response Stephen. Sorry it's taken me a while to get back to you.

On Wed, 14 Oct 2020 at 17:16, Stephen Boyd <sboyd@xxxxxxxxxx> wrote:
>
> Quoting Joel Stanley (2020-10-13 22:28:00)
> > On Wed, 14 Oct 2020 at 02:50, Stephen Boyd <sboyd@xxxxxxxxxx> wrote:
> > >
> > > Quoting Ryan Chen (2020-09-28 00:01:08)
> > > > In ASPEED SoC LCLK is LPC clock for all SuperIO device, UART1/UART2 are
> > > > default for Host SuperIO UART device, eSPI clk for Host eSPI bus access
> > > > eSPI slave channel, those clks can't be disable should keep default,
> > > > otherwise will affect Host side access SuperIO and SPI slave device.
> > > >
> > > > Signed-off-by: Ryan Chen <ryan_chen@xxxxxxxxxxxxxx>
> > > > ---
> > >
> > > Is there resolution on this thread?
> >
> > Not yet.
> >
> > We have a system where the BMC (management controller) controls some
> > clocks, but the peripherals that it's clocking are outside the BMC's
> > control. In this case, the host processor us using some UARTs and what
> > not independent of any code running on the BMC.
> >
> > Ryan wants to have them marked as critical so the BMC never powers them down.
> >
> > However, there are systems that don't use this part of the soc, so for
> > those implementations they are not critical and Linux on the BMC can
> > turn them off.
> >
> > Do you have any thoughts? Has anyone solved a similar problem already?
> >
>
> Is this critical clocks in DT? Where we want to have different DT for
> different device configurations to indicate that some clks should be
> marked critical so they're never turned off and other times they aren't
> so they're turned off?

Spot on.

> It also sounds sort of like the protected-clocks binding. Where you
> don't want to touch certain clks depending on the usage configuration of
> the SoC. There is a patch to make that generic that I haven't applied
> because it looks wrong at first glance[1].

That binding is exactly what I had in mind. I wasn't aware of it.

The drawbacks outlined in the commit message do sound concerning. I
take it we could avoid those drawbacks by having a driver-specific
implementation of protected-clocks, like qcom does?

> Maybe not registering those
> clks to the framework on the configuration that Ryan has is good enough?

I didn't quite follow here. Did you mean with protected-clocks, or
using a different mechanism?

Cheers,

Joel

>
> [1] https://lore.kernel.org/r/20200903040015.5627-2-samuel@xxxxxxxxxxxx