Re: [PATCH v2 1/2] v2: more sampling fun 1

speck for Andi Kleen <[email protected]> writes:

>> > > I implemented this in the way I did to avoid modifying x86_cpu_id as
>> > > that structure is an exported ABI and any change would impact user mode
>> > > code using the structure.
>> > 
>> > Exported ABI? Which usermode code uses this? The module loading tools?
>> 
>> Yeah, Andi pointed it out to me on an internal review.  I don't know what tool
>> is using it.
>
> Yes it's modprobe to find which module to load.
>
> All of mod_devicetable.h is ABI

That's simply not true.

mod_devicetable.h is a kernel internal header which gets analyzed by the
kernel internal tool modpost which uses this header to generate the
ALIAS entries in the .modinfo section of the .ko elf file.

The ALIAS entries are user space ABI but not the header file.

If the header file changes then file2alias.c has to be updated as
clearly stated in the comment at the top of mod_devicetable.h

And if you look at the driver_data member of x86_cpu_id then you'll
notice that file2alias does not care about it at all, neither would it
care about a stepping entry or whatever.

Thanks,

        tglx