Re: 07/26 GIT - Build breakage

[Parag Warudkar <parag.warudkar@gmail.com>]

Rusty Russell wrote:
> On Thu, 2007-07-26 at 16:07 -0700, Linus Torvalds wrote: 
>   
>> On Thu, 26 Jul 2007, Parag Warudkar wrote:
>>     
>>> arch/i386/kernel/paravirt.o: In function `init_IRQ':
>>> (.text+0xb20): multiple definition of `init_IRQ'
>>> arch/i386/kernel/i8259.o:(.init.text+0x30): first defined here
>>> ld: Warning: size of symbol `init_IRQ' changed from 66 in
>>> arch/i386/kernel/i8259.o to 14 in arch/i386/kernel/paravirt.o
>>> make[1]: *** [arch/i386/kernel/built-in.o] Error 1
>>> make: *** [arch/i386/kernel] Error 2
>>>       
>> Ok, it seems that some versions of gcc/binutils have problems with the 
>> games that paravirt introduced with the "weak" thing. 
>>
>> I'm not entirely surprised. That stuff was trying to be too clever by 
>> half.
>>
>> Rusty?
>>     
>
> Sure, but this was in .21 and .22.  This config is from 2.6.21.5, in
> fact.
>
> Parag, does your arch/i386/kernel/i8259.c look like this around line
> 384?  If not, perhaps it's a mis-merge for you? 
>         
>         /* Overridden in paravirt.c */
>         void init_IRQ(void) __attribute__((weak, alias("native_init_IRQ")));
>         
>         void __init native_init_IRQ(void)
>         {
>         
>   
Never mind - I was compiling using an account which was not in use for 
long and I had totally forgotten that llvm gcc comes before the normal 
gcc version in the $PATH on that account. Correcting the path fixes the 
problem.

Sorry for the trouble.
Parag