Kirkwood issues with 3.8-rc1 - Ooops and hang on reboot

[Josh Coombs]

I did a test build with the __init tag removed (which is not the
correct answer) and everything works.  GPIO is functional (I didn't
catch the power LED wasn't working prior), I don't get a panic, and I
can now reboot cleanly as well.  So that confirms the basic diagnosis.

I've been trying to wrap my head around include/linux/init.h to see
what the correct tagging should be but I'll admit, I'm in over my head
so far.  If anyone has suggested reading on the topic I'd be very
appreciative.

Josh C

On Thu, Jan 3, 2013 at 3:22 AM, Andrew Lunn <andrew at lunn.ch> wrote:
> On 03/01/13 05:50, Josh Coombs wrote:
>>
>> On Sat, Dec 29, 2012 at 7:23 PM, Russell King - ARM Linux
>> <linux at arm.linux.org.uk>  wrote:
>>>
>>> On Sat, Dec 29, 2012 at 06:31:37PM -0500, Josh Coombs wrote:
>>>>
>>>> Bah, didn't send to everyone last time...
>>>>
>>>> Building with CONFIG_DEBUG_SECTION_MISMATCH enabled, the only warning
>>>> I get is for drivers/w1/masters/w1-gpio.o.
>>>>
>>>> WARNING: drivers/w1/masters/w1-gpio.o(.data+0x0): Section mismatch in
>>>> reference from the variable w1_gpio_driver to the function
>>>> .init.text:w1_gpio_probe()
>>>> The variable w1_gpio_driver references
>>>> the function __init w1_gpio_probe()
>>>> If the reference is valid then annotate the
>>>> variable with __init* or __refdata (see linux/init.h) or name the
>>>> variable:
>>>> *_template, *_timer, *_sht, *_ops, *_probe, *_probe_one, *_console
>>>>
>>>> Shouldn't I be seeing a warning for
>>>> drivers/clk/mvebu/clk-gating-ctrl.o at this point as well?
>>>
>>>
>>> Unfortunately not.  mvebu_clk_gating_get_src() is referenced by another
>>> __init function, which is registering the pointer to
>>> mvebu_clk_gating_get_src() into other code.  That reference (obviously
>>> from the oops dump) persists past the point where the __init sections
>>> are freed.
>>>
>>> Hence why no section mismatch warning issued from the static tools;
>>> they're not infallible.
>>
>>
>> Ok, so a workaround would be to remove the __init tag so persists
>> properly, but it sounds like what I really should do is to find out
>> where that other reference is and why it's persisting to make sure
>> that's the right behavior in the first place?
>
>
> Hi Josh
>
> If you look in drivers/clk/mvebu/clk-gating-ctrl.c,
> mvebu_clk_gating_get_src() is passed as a parameter to
> of_clk_add_provider(). of_clk_add_provider() allocates a structure with
> kzalloc() and then assigns the function pointer to a member of the
> structure.
>
> Because mvebu_clk_gateing_src() is passed as a parameter, the section
> missmatch checks cannot detect a problem. As far as i understand, its the
> linker which is checking for section miss matches. However, the linker is
> not involved here, its not a function calling a function, its a function
> passed to a function, and later a function called via a function pointer.
>
> As far as i can see, the only error here is the wrong __init tag.
>
>     Andrew