On 13.02.2018 03:59, Chen-Yu Tsai wrote:

On Tue, Feb 13, 2018 at 9:25 AM, Philipp Rossak [off-list ref] wrote:

quoted

On 12.02.2018 19:21, Philipp Rossak wrote:

quoted

Hey,

When I try to boot my A31s (Bananapi M2) u-boot is showing only Starting
kernel ... . After enabling the earlyprintk I could capture this log: [1].

After reverting those 5 commits from Chen-Yu I was able to boot again:


clk: sunxi-ng: Support fixed post-dividers on NM style clocks
7d333ef1cc1b8c8951f3a2c41f6406e2295d8be9

clk: sunxi-ng: sun8i: a83t: Add /2 fixed post divider to audio PLL
10e6eb4f2c5b35ae71c9bc0db83d74238719b453

clk: sunxi-ng: sun8i: a83t: Use sigma-delta modulation for audio PLL
e952ca3c6b2ffdfbf9618e4bd3e9aad1ff3f5eb4

clk: sunxi-ng: Support fixed post-dividers on MP style clocks
946797aa3f08e2f6f5992f3ec2be44791e9b9260

clk: sunxi-ng: sun50i: a64: Add 2x fixed post-divider to MMC module clocks
83fe3be4d1974f5f50c5e2039a1609f4960e8579


I allready tried to fix it with making them save against zero:

if (cmp->common.features & CCU_FEATURE_FIXED_POSTDIV && \
          cmp->fixed_post_div with)
                 rate *= cmp->fixed_post_div;

But that didn't help.

Any ideas?

Regards,
Philipp

[1]: https://pastebin.com/64Fzzqvg

It took me some time, but I have now a few more infos:

Right now the code breaks at this point here [1], with this clock [2].
If we have a look now at the clock config [3], we see here a table which is
an u8 array and also a fixed_predivs struct.

The u8 array is for mapping the parents from the index in the parents
array to the actual register value you listed below.

How are you figuring out which clock is triggering this? Because that
is not even the right type of clock. The backtrace you posted shows
the error occurring in a DIV or M type clock, not the MP type you
are pointing to.

Could you add some noisy printk calls to the sunxi_ccu_probe()
function in drivers/clk/sunxi-ng/ccu_common.c so it's much clearer
which clock is failing?

Thats what I basically did to find out which clock is failing. This here 
are the changes I'm doing [1]  and thats the dirty log [2]. It fails at 
clock NR 155 which is this one [3] mentioned before.

quoted

If we have a look at the function call where it breaks [4], shouldn't the
table be a clk_div_table struct instead of an u8?

The table argument is an option. Did you go through how the sunxi-ng driver
calls this function? As mentioned above, you are looking at the wrong thing.

Thanks
ChenYu

I followed the failing call to this function (through the clock driver). 
As you can see I added an additional printk statement to see where it fails.

Regards,
Philipp

quoted

The a31s is the only board where we have this combination of a fixed_predivs
and a table.

Philipp


Related Clock source register A31s:

0000: OSC24MHz/750=32KHz
0001: LOSC
0010: OSC24MHz
0011: /
0100: /
0101: /
0110: /
0111: /
1000: /
1001: /
1010: /
1011: AXICLK/4
1100: /
1101: AHB1CLK/4
1110: /
1111: /


[1]:
http://lxr.bootlin.com/linux/v4.16-rc1/source/drivers/clk/clk-divider.c#L89

[2]:
http://lxr.bootlin.com/linux/v4.16-rc1/source/drivers/clk/sunxi-ng/ccu-sun6i-a31.c#L1137

[3]:
http://lxr.bootlin.com/linux/v4.16-rc1/source/drivers/clk/sunxi-ng/ccu-sun6i-a31.c#L749

[4]:
http://lxr.bootlin.com/linux/v4.16-rc1/source/drivers/clk/clk-divider.c#L93

[1]: https://pastebin.com/tvDunAWq
[2]: https://pastebin.com/SBn2VQLu
[3]: 
http://lxr.bootlin.com/linux/v4.16-rc1/source/include/dt-bindings/clock/sun6i-a31-ccu.h#L187