Re: [PATCH v2] ubifs: fix kconfig dependency warning

[Randy Dunlap <randy.dunlap@oracle.com>]

On 3/30/2011 1:12 AM, Artem Bityutskiy wrote:
> On Tue, 2011-03-29 at 08:48 -0700, Randy Dunlap wrote:
>> On Tue, 29 Mar 2011 18:04:15 +0300 Artem Bityutskiy wrote:
>>
>>> On Tue, 2011-03-29 at 08:01 -0700, Randy Dunlap wrote:
>>>> On 03/29/11 00:02, Artem Bityutskiy wrote:
>>>>> On Mon, 2011-03-28 at 13:40 -0700, Randy Dunlap wrote:
>>>>>> From: Randy Dunlap<randy.dunlap@oracle.com>
>>>>>>
>>>>>> Fix another kconfig dependency warning, this time in ubifs.
>>>>>>
>>>>>> warning: (UBIFS_FS_DEBUG&&  LOCKDEP&&  LATENCYTOP) selects KALLSYMS_ALL which has unmet direct dependencies (DEBUG_KERNEL&&  KALLSYMS)
>>>>>>
>>>>>> Without this patch, we can have:
>>>>>> # CONFIG_KALLSYMS is not set
>>>>>> CONFIG_KALLSYMS_ALL=y
>>>>>> which is useless (does nothing unless KALLSYMS is enabled).
>>>>>>
>>>>>> However, ubifs builds successfully with or without this patch,
>>>>>> and it builds with this line completely deleted,
>>>>>> so what was this 'select' for?  Just developer convenience?
>>>>>
>>>>> Well, here is the idea. You can compile UBIFS with debugging and without
>>>>> debugging. Without debugging the resulting ubifs.ko is much smaller, so
>>>>> some embedded people prefer it this way.
>>>>>
>>>>> If you select debugging support, then we'll compile it a lot of
>>>>> assertions, self-checks, test-modes, extra error messages with detailed
>>>>> dumps. And we want to see stackdumps when errors or problems happen,
>>>>> this is why we select KALLSYMS_ALL.
>>>>>
>>>>> So I guess instead we should do:
>>>>>
>>>>> select KALLSYMS
>>>>> select KALLSYMS_ALL
>>>>
>>>> Yes, that should do it.  Thanks for the explanation.
>>>
>>> Will you submit a patch? Alternatively, I can make it myself. What is
>>> your preference?
>>
>> Here's an updated patch.  But since KALLSYMS_ALL depends on DEBUG_KERNEL,
>> the lines above aren't quite sufficient and I don't care to select
>> DEBUG_KERNEL.
>
> What is the real difference between KALLSYMS_ALL and KALLSYMS? It looks
> like for stack dumps KALLSYMS is enough. The Kconfig help text is not
> very helpful. And when I look at the help text of
> CONFIG_KALLSYMS_EXTRA_PASS I get feeling that this area needs some
> clean-up work.

Yes, KALLSYMS is enough for most of us.

You can see what KALLSYMS_ALL does by looking at scripts/kallsyms.c,
the --all-symbols option:

>         /* if --all-symbols is not specified, then symbols outside the text
>          * and inittext sections are discarded */



> Anyway, any idea why we wouldn't just kill KALLSYMS_ALL by merging it
> with KALLSYMS?

KALLSYMS_ALL probably generates noise for most use cases.

---