Re: [PATCH] spi: tegra114: add spi driver

[Stephen Warren <swarren-3lzwWm7+Weoh9ZMKESR00Q@public.gmane.org>]

On 02/20/2013 06:26 AM, Laxman Dewangan wrote:
> On Wednesday 20 February 2013 06:41 PM, Mark Brown wrote:
>> * PGP Signed by an unknown key
>>
>> On Wed, Feb 20, 2013 at 05:59:03PM +0530, Laxman Dewangan wrote:
>>>>> +    tspi->clk = devm_clk_get(&pdev->dev, "spi");
>>>> Does this HW block use multiple clocks? If not, I think s/"spi"/NULL/
>>>> there, just like the Tegra20 driver.
>>> No, spi controller uses the only one clock. I will change to NULL.
>> I'm never convinced that NULL is a helpful clock name to pick, it's not
>> awesome if you ever acquire a second clock.
> 
> I had other idea of okeeping clock name here for power optimization.
> Spi controller can take the power source ffrom different clock source
> say clk_m (crystal) and pllp.
> 
> I want to set the parent clock dynamically based on required speed so
> that if the desire speed can be meet from clk_m, no need to increase pll
> count and possible we may endup with siwtchng off pllp which can save
> power.
> 
> So for this I may require
> 
> spi_clk = devm_clk_get(&pdev->dev, "spi");
> spi_clkm = devm_clk_get(&pdev->dev, "clmkm");
> spi_pllp = devm_clk_get(&pdev->dev, "pllp");
> 
> and call clk_set_parent(spi_clk, spi_clkm) or clk_set_parent(spi_clk,
> spi_pllp).

OK, so that's certainly an argument for requesting a specific clock name
rather than NULL.

But, please do think this approach through fully. The DT binding needs
to define which clock-names the driver requires to be present, and any
optional clock names. DT bindings are supposed to be immutable, or
perhaps extendible in a completely backwards-compatible fashion. This
implies that you need to have thought through the entire list of clocks
that the driver might want in the DT clock-names property when you first
write the DT binding documentation...