Re: [PATCH 1/4] kernel: constructor support

[Peter Oberparleiter <oberpar@linux.vnet.ibm.com>]

Ingo Molnar wrote:
> * Peter Oberparleiter <oberpar@linux.vnet.ibm.com> wrote:
>> Ingo Molnar wrote:
>>> * Peter Oberparleiter <oberpar@linux.vnet.ibm.com> wrote:
>>>
>>>> Disable constructor support for usermode Linux to prevent conflicts
>>>> with host glibc.
>>>> +++ linux-2.6.30-rc4/init/Kconfig
>>>> @@ -16,6 +16,11 @@ config DEFCONFIG_LIST
>>>>  	default "$ARCH_DEFCONFIG"
>>>>  	default "arch/$ARCH/defconfig"
>>>>  +config CONSTRUCTORS
>>>> +	bool
>>>> +	depends on !UML
>>>> +	default y
>>>> +
>>>>  menu "General setup"
>>> Hm, excluding UML like that is sad. Is there no better solution?
>> UML is excluded because in that environment constructors are 
>> called by the host glibc, so there is no need for kernel support 
>> on UML (in fact it would break things).
>>
>> Or were you referring to the actual way the exclusion is 
>> implemented?
> 
> the way it's done is OK (there's really just UML in this situation), 
> but the question is really, shouldnt it be possible to coverage-test 
> UML instances 'from the inside'?

 From a mere gcov perspective, coverage-testing from the outside is 
superior because that is the way it was meant to be run in the first place.

> 
> Plus, if any other kernel facility grows out of this or makes use of 
> it, UML will be left out in the cold.

I'm afraid that trying to over-engineer the gcov-kernel mechanism at 
this time might serve neither the gcov-kernel users, nor potential new 
users. Once the base is established, it will be far easier to decide 
which other purposes the infrastructure can serve (without completely 
bending it).