On 15/05/2014 at 17:43:03 +0200, Sebastian Hesselbarth wrote :

On 05/15/2014 10:09 AM, Alexandre Belloni wrote:

quoted

On 14/05/2014 at 22:15:17 +0200, Sebastian Hesselbarth wrote :

quoted

+	/* clock divider cells */
+	parent_names[1] = avpllb_names[CH4];
+	parent_names[2] = avpllb_names[CH5];
+	parent_names[3] = avpllb_names[CH6];
+	parent_names[4] = avpllb_names[CH7];
+
+	parent_names[0] = refclk_names[SYSPLL];

It should actually be:

parent_names[0] = avpllb_names[CH4];
parent_names[1] = avpllb_names[CH5];
parent_names[2] = avpllb_names[CH6];
parent_names[3] = avpllb_names[CH7];
parent_names[4] = refclk_names[SYSPLL];

Given the comment to remove index 0 in the last patch, I translate that
into: "the input mux bypass is there, but {cannot,should not,we do not
want it to} be used". *sigh*

Actually, almost all of this is based on Chromecast mirrored BSP code
and I though about leaving the bypass mux in - even if it is not used
at all.

The reason is that I am _very_ tired of reading through the BSP code
and have all the things in mind where the BSP code is unclear.

Ok, I made a mistake, I also have a hard time to picture everything
myself ;)

So, the mux is there iand can be used, hence, the parents are actually:

parent_names[0] = refclk_names[SYSPLL];
parent_names[1] = avpllb_names[CH4];
parent_names[2] = avpllb_names[CH5];
parent_names[3] = avpllb_names[CH6];
parent_names[4] = avpllb_names[CH7];
parent_names[5] = refclk_names[SYSPLL];

and disregard my comment on the previous patch.

quoted

quoted

+	data = &bg2_divs[CLKID_SYS];
+	clks[CLKID_SYS] = berlin2_div_register(&data->map, base, data->name,
+		       data->div_flags, parent_names, 5, data->flags, &lock);
+
+	parent_names[0] = refclk_names[CPUPLL];
+	parent_names[5] = refclk_names[MEMPLL];

The only valid choice here should be (remember, we are not adding 1 to
the index anymore):
parent_names[4] = refclk_names[MEMPLL];

There, you actually had it right, maybe we could set parent_names[1] to
parent_names[4] to something bogus or all to refclk_names[MEMPLL].

Funny to see that there ought to be a CPUPLL which isn't used by the
CPU at all. This also implies to remove CPUPLL, right?

quoted

quoted

+	data = &bg2_divs[CLKID_CPU];
+	clks[CLKID_CPU] = berlin2_div_register(&data->map, base, data->name,
+			 data->div_flags, parent_names, 6, data->flags, &lock);
+

This is where it gets tricky, now we should have:
parent_names[0] = avpllb_names[CH4];
parent_names[1] = avplla_names[CH5];
parent_names[2] = avpllb_names[CH6];
parent_names[3] = avpllb_names[CH7];
parent_names[4] = refclk_names[SYSPLL];

First I thought that it is just the default input mux clocks again..
but then I noticed that it is actually AVPLL_A5 not B5.

Here it becomes:
parent_names[0] = refclk_names[SYSPLL];
parent_names[1] = avpllb_names[CH4];
parent_names[2] = avplla_names[CH5];
parent_names[3] = avpllb_names[CH6];
parent_names[4] = avpllb_names[CH7];
parent_names[5] = refclk_names[SYSPLL];

Ok, I admit having confirmed information is maybe better. So, you agree
that we can remove the input mux bypass on the complex divider, too?
(Including all the consequences: remove it from the divmap, driver, ...)

No, let's keep the mux, sorry about that confusion.


-- 
Alexandre Belloni, Free Electrons
Embedded Linux, Kernel and Android engineering
http://free-electrons.com