[PATCH v2 0/6] Migrate PXA27x platforms to clock framework

Robert Jarzmik robert.jarzmik at free.fr
Sat Jul 26 01:52:49 PDT 2014

Mike Turquette <mturquette at linaro.org> writes:

> Quoting Robert Jarzmik (2014-07-21 11:23:51)
>> Robert Jarzmik <robert.jarzmik at free.fr> writes:
>> > This is the second opus of this serie.
>> Ping ?
> Hi Robert,
> Overall the series looks good to me. The only problem I see is that
> somewhere in the use of the PARENTS() macro there is a section mismatch
> introduced. I spent about 5 minutes looking at it and gave up. This
> problem introduces 26 section mismatch warnings building
> colibri_pxa270_defconfig.
OK Mike, I'll fix this. This is probably because pxa27x_clocks array is not
__init, and it references parents which are __init. I was lazy letting
pxa27x_clocks not __init because I reuse the internal fields "hw" and "gate->hw"
later (hence I didn't leave the __init).
This is a bit dirty and I'll launder it.

Moreover, I have one fix which I had fixed wrt boot ordering : the
postcore_init() must die. The reason behind is that I have no guarantee of
ordering between clocks init and clocksource init, and clocksource relies on

The right way will be AFAIK to call the core clocks init :
 - within the device-tree initialization function : pxa27x_dt_clocks_init()
 calls pxa27x_clocks_init9)
 - directly from pxa27x_init() or pxa_timer_init() in non-device tree setup

> Once those are fixed up then I will be happy to take this series. Do you
> want me to take all six patches through the clk tree, including the
> arm/arm stuff?
I would feel more comfortable if :
 - you take the 4 first ones (3 clock ones + 1 arm pxa header bit definitions)
 - Haojian takes the last 2 ones (pxa ones)

I will check that taking only the first 4 ones doesn't break anything, in case
there is a merge window hipcup.



More information about the linux-arm-kernel mailing list