Re: [PATCH v2 16/19] clk: tegra: pll: Add Set_default logic

[Rhyland Klein <rklein-DDmLM1+adcrQT0dZR+AlfA@public.gmane.org>]

On 5/11/2015 7:50 AM, Peter De Schrijver wrote:
> On Thu, Apr 30, 2015 at 11:31:22AM -0400, Rhyland Klein wrote:
>> On 4/30/2015 6:12 AM, Peter De Schrijver wrote:
>>> On Wed, Apr 29, 2015 at 01:21:46PM -0400, Rhyland Klein wrote:
>>>> From: Bill Huang <bilhuang-DDmLM1+adcrQT0dZR+AlfA@public.gmane.org>
>>>>
>>>> Add logic which (if specified for a pll) can verify that a PLL is set
>>>> to the proper default value and if not can set it. This can be
>>>> specified per PLL as each will have different default values.
>>>>
>>>
>>> Why can't we just set the default values at init time?
>>
>> Sorry, I did some investigation into this and wrote up a nice response
>> ... and forgot to hit send ...
>>
>> The reason this can't be run only once at init time is the following. In
>> reality, we want to have the defined default values written as early as
>> possible. Idealy, the bootloader could write these, so the kernel need
>> only check, see they are right, and not touch them. However, since we
>> can't rely on the bootloader to do so, the kernel needs the support to
>> be able to write these default values. At init time, some pll's will be
>> enabled (from bootloader) and because they are enabled (and the rest of
>> the clk framework isn't done being setup yet) we can't disable them to
>> write the full register values. Therefore, the set_defaults logic uses a
>> 2-pass system.
>>
>> first pass: Try to set defaults at init/registration time. If pll is
>> disabled, this works fine. If it is enabled, then we update a subset of
>> the register as a "best effort" setting of the defaults.
>>
>> second pass: Should only run the first time we go through set_rate for a
>> pll. If the first pass already wrote default value, then it skips this
>> step. Otherwise it will go in, once the pll is disabled in the set_rate
>> path, and write the full register default.
>>
>> This is required because some registers need to be reset to the default
>> values we have to ensure locking works correctly. Does that make sense?
> 
> Ok. I see... Should we print a warning (pr_warn()) the bootloader isn't
> initializing the hw correctly if the second pass needs to write the default
> values?

All the set default routines use the inline function
"_pll_misc_chk_default" (used to be a MACRO). Inside, it compares
register values vs expected defaults and warns:

        if (boot_val != default_val) {
                pr_warn("boot misc%d 0x%x: expected 0x%x\n",
                        misc_num, boot_val, default_val);
                pr_warn(" (comparison mask = 0x%x)\n", mask);
                params->defaults_set = false;
        }

so this is already done. I suppose the only other place we could add a
warning is if the first past set-defaults can't be fully run since the
clock is on, that might make things a little clearer, so I guess I'll go
ahead and add that.

-rhyland

> 
> Thanks,
> 
> Peter.
> 


-- 
nvpublic