Re: [PATCH 0/4] clk: mvebu: fix clk init order

[Gregory CLEMENT <gregory.clement@free-electrons.com>]

On 04/02/2014 00:36, Sebastian Hesselbarth wrote:
> On 02/04/2014 12:16 AM, Willy Tarreau wrote:
>> On Thu, Jan 30, 2014 at 11:31:32AM +0100, Sebastian Hesselbarth wrote:
>>> On 01/30/14 11:24, Gregory CLEMENT wrote:
>>>> On 25/01/2014 19:19, Sebastian Hesselbarth wrote:
>>>>> This patch set fixes clk init order that went upside-down with
>>>>> v3.14. I haven't really investigated what caused this, but I assume
>>>>> it is related with DT node reordering by addresses.
>>>>
>>>> Can you explain what kind of issue do you observe?
>>>
>>> Sure. When probing CLK_OF_DECLAREed clock drivers, clock-gating driver
>>> gets registered before core-clocks. It therefore cannot resolve it's
>>> parent clock name for tclk and all clock gates will have no parent
>>> clock.
>>>
>>> Usually, you'll see in some drivers (e.g. v643xx_eth) div_by_zero errors
>>> poping up, when they calculate a frequency division factors based on
>>> clock gate frequency, which should have been tclk but is 0 now.
>>
>> Well, to be honnest, I have no idea whether your patch is the right way
>> to fix the problem or not, but what I can say is that it fixes such oopses
>> that appear in 3.14-rc1 when booting on mirabox :
>>
>> Division by zero in kernel.
> 
> Willy,
> 
> you have hit exactly the reason for this patch.
> 
> [...]
>> By the way, seeing how often a trick related to the DT is nedeed to solve an
>> oops or a panic, I'm really scared that this whole DT mess is just becoming
>> the exact copy of the ACPI mess (but 15 years later) and we'll experience the
>> same horrible things :-( Sometimes I'm wondering whether there are not too
>> many structural things put in there...
> 
> To be precise, it is not a DT-related trick at all. You would have the
> same issues, if you'd register those "low-level" (i.e. early) drivers
> without DT. It is more about missing init ordering, here.
> 
> There could be different ways to work this out, even elevating clock
> devices to "normal" probed devices could be possible. I am sure, in the
> long run, it will work out, but now this is a fix for v3.14-rc1.
> 
> @Jason, Andrew, Gregory, Thomas:
> Now that v3.14 is out, anything against taking this in as fixes for rc1?

Hi Sebastian,

I am not found of this solution I still think it should be done
at framework level. However we still have this very annoying issue,
and this fix is better than nothing. So I am not against taking this
for rc1 with the hope that it will be later revert with a better
solution.


Thanks,

Gregory


> 
> Sebastian
> 


-- 
Gregory Clement, Free Electrons
Kernel, drivers, real-time and embedded Linux
development, consulting, training and support.
http://free-electrons.com