[elrepo-devel] why it's not working for me?

[Phil Perry]

Farkas Levente wrote:
> On 09/01/2009 05:56 PM, Phil Perry wrote:
>> Correct - and you've worked out the answer for yourself :)
>>
>> The explanation:
>>
>> In July 2007 these 3 particular drivers were converted to platform
>> drivers (using functions defined in<linux/platform_device.h>) in the
>> mainline kernel, from where the source is backported. Thus, when built
>> as a kABI-tracking package, they Require ksym(platform_device_add_data)
>> as you note. However, the current upstream el5.3 kernel doesn't have
>> ksym(platform_device_add_data) defined within it's ABI whitelist
>> (presumably because no modules within the kernel currently use this
>> function).
>>
>> As the el5.3 kernel is unable to meet the requirement from within it's
>> ABI whitelist, there's no point building the package with that
>> requirement as it will never be met, so we don't.
>>
>> As you've correctly deduced, that behaviour is determined by
>> find-requires.ksyms, which we modify and build against. We don't use
>> mock to build elrepo mods, but rather build on native architecture
>> machines (doesn't matter whether virtual or real). If you wanted to
>> build in mock then I guess you'd need to patch redhat-rpm-config
>> installed in your local buildroot.
> 
> that's what i already did. but anyway creating none reproducable packages...
> you wouldn't like to diverge too far from kmodtool from 
> find-requires.ksyms etc. but it's not working there are many place where 
> you should have to. anyway if there is already a redhat-rpm-config in 
> elrepo why not make it working with this kmods.
> the best would be to comes together and fix this problems at once with 
> redhat Jon or Dag or anybody. imho the current situation is not good for 
> anyone:-(
> 


Ideally the platform_device_add_data symbol needs adding to the 
[rhel5_drivers_base_ga] whitelist grouping.

The Red Hat whitepaper states that the whitelist is created by looking 
at all kernel modules and 3rd party drivers they are aware of, with the 
intention of providing a minimal whitelist that can grow over time as 
needed.

So in this situation it would appear the correct course of action would 
be for us to file a bug report requesting the addition of 
ksym(platform_device_add_data) to the [rhel5_drivers_base_ga] whitelist 
thus fixing the issue for this (these) driver(s).

However, doing so will not fix the issue for past kernels, only those 
going forward.