Re: [PATCH 2/4] PCI: qcom: Use clk_bulk_ API for 1.0.0 clocks handling

[Johan Hovold <johan@kernel.org>]

On Thu, Oct 20, 2022 at 02:22:47PM +0300, Dmitry Baryshkov wrote:
> On 20/10/2022 14:08, Johan Hovold wrote:
> > On Thu, Oct 20, 2022 at 01:31:18PM +0300, Dmitry Baryshkov wrote:

> >> +	res->clks[0].id = "aux";
> >> +	res->clks[1].id = "iface";
> >> +	res->clks[2].id = "master_bus";
> >> +	res->clks[3].id = "slave_bus";
> >>   
> >> -	res->slave_bus = devm_clk_get(dev, "slave_bus");
> >> -	if (IS_ERR(res->slave_bus))
> >> -		return PTR_ERR(res->slave_bus);
> >> +	ret = devm_clk_bulk_get(dev, ARRAY_SIZE(res->clks), res->clks);
> >> +	if (ret < 0)
> >> +		return ret;
> > 
> > Are you sure there are no dependencies between these clocks and that
> > they can be enabled and disabled in any order?
> 
> The order is enforced by the bulk API. Forward to enable, backward to 
> disable.

Right you are. (I had it mixed up with a different API which had no such
guarantees and now I can't seem to remember which it was, maybe I dreamt
it.)

> > Are you also convinced that they will always be enabled and disabled
> > together (e.g. not controlled individually during suspend)?
> 
>  From what I see downstream, yes. They separate host and pipe clocks, 
> but for each of these groups all clocks are disabled and enabled in 
> sequence.
> 
> For the newer platforms the only exceptions are refgen (handled by the 
> PHY in our kernels) and ddrss_sf_tbu (only on some platforms), which is 
> not touched by these patches.

Sounds good.

Johan