Re: [PATCH 1/5] clk: berlin: add support for berlin plls

[Sebastian Hesselbarth <sebastian.hesselbarth@gmail.com>]

On 03/21/2014 04:45 PM, Alexandre Belloni wrote:
> [all commentis I agree on are snipped]

:)

> On 21/03/2014 at 13:49:32 +0100, Sebastian Hesselbarth wrote :
>> On 03/21/2014 12:43 PM, Alexandre Belloni wrote:
>> obj-y += pll.o
>> obj-$(CONFIG_MACH_BERLIN_BG2)	+= pll-berlin2.o
>> obj-$(CONFIG_MACH_BERLIN_BG2CD)	+= pll-berlin2.o
>> obj-$(CONFIG_MACH_BERLIN_BG2Q)	+= pll-berlin2q.o
>>
>> Which reminds me, that we forgot to add MACH_BERLIN_BG2Q to
>> arch/arm/mach-berlin/Kconfig. Can you spin a patch?
>>
>
> I will do that.
>
>>> +static const u8 vcodiv_berlin2[] = {10, 15, 20, 25, 30, 40, 50, 60, 80};
>>> +
>>> +static struct berlin_pllmap berlin_pll_map = {
>>> +	.vcodiv = vcodiv_berlin2,
>>> +	.fbdiv_mask = 0x1FF,
>>> +	.rfdiv_mask = 0x1F,
>>> +	.divsel_mask = 0xF,
>>
>> divsel_mask allows 16 possible values, vcodiv_berlin2[] only provides
>> 9, and common pll driver below does not know how many valid vcodiv
>> values are passed. That can become dangerous..
>>
>> I'd rather extend vcodiv_berlin2 to full divsel range and provide
>> safe (=1) divisiors. This way wrong/new register values will only
>> break clock frequency derived.
>>
>
> Good catch ! Then, what about simply shrinking the mask so that we don't
> overflow the table. We'll put it back to its supposed real value whant
> we know what are the remaining divisors (my guess is that they are already
> all listed here). I would say that we are getting the divisor wrong if
> divsel > 8 anyway.

Hmm, maybe I should look up valid vcodiv myself, but your vcodiv_berlin2
has 9 values and I guess they are all valid, aren't they?

The next possible, larger mask where 0-8 fits in, is 0xf. You used that
above and that reveals 16 possible indices.

The only option for shrinking the table that I see, would be min/max
allowed indices, but that is as useful as having a slightly larger
table.

>>> +	.fbdiv_shift = 6,
>>> +	.rfdiv_shift = 1,
>>> +	.divsel_shift = 7,
>>
>> Have .foo_mask and .foo_shift together?
>>
>
> This will make the struct larger but I don't really have an opinion.

Maybe, I wasn't clear enough. Just assign .foo_mask and .foo_shift in
subsequent lines of code, i.e.

static struct berlin_pllmap berlin_pll_map = {
	.vcodiv = vcodiv_berlin2,
	.fbdiv_mask = 0x1FF,
	.fbdiv_shift = 6,
	.rfdiv_mask = 0x1F,
	.rfdiv_shift = 1,
	.divsel_mask = 0xF,
	.divsel_shift = 7,
};

Sebastian