Re: Section conflicts with percpu variables

Re: Section conflicts with percpu variables

[Tejun Heo]

On 10/27/2010 09:05 PM, Jeremy Fitzhardinge wrote:
>  Hi Tejun,
> 
> I wonder if you could have a look at this.  I have someone reporting
> compilation failures when using the stock Centos 5 compiler:
> 
> arch/x86/xen/mmu.c:163: error: __pcpu_scope_xen_cr3 causes a section type conflict
> arch/x86/xen/mmu.c:164: error: __pcpu_scope_xen_current_cr3 causes a section type conflict
> arch/x86/xen/mmu.c:163: error: __pcpu_unique_xen_cr3 causes a section type conflict
> arch/x86/xen/mmu.c:164: error: __pcpu_unique_xen_current_cr3 causes a section type conflict
> 
> Looking at mmu.i, I can't see why it is picking on these particular
> per-cpu variables.  Do you have any insight into this.

Hmmm... me neither.  section type conflict?  Does it make any
different if you move the definitions near the top of the file or use
a different compiler version?

Thanks.

-- 
tejun

Re: Section conflicts with percpu variables

[Jeremy Fitzhardinge]

 On 10/28/2010 02:21 AM, Tejun Heo wrote:
> On 10/27/2010 09:05 PM, Jeremy Fitzhardinge wrote:
>>  Hi Tejun,
>>
>> I wonder if you could have a look at this.  I have someone reporting
>> compilation failures when using the stock Centos 5 compiler:
>>
>> arch/x86/xen/mmu.c:163: error: __pcpu_scope_xen_cr3 causes a section type conflict
>> arch/x86/xen/mmu.c:164: error: __pcpu_scope_xen_current_cr3 causes a section type conflict
>> arch/x86/xen/mmu.c:163: error: __pcpu_unique_xen_cr3 causes a section type conflict
>> arch/x86/xen/mmu.c:164: error: __pcpu_unique_xen_current_cr3 causes a section type conflict
>>
>> Looking at mmu.i, I can't see why it is picking on these particular
>> per-cpu variables.  Do you have any insight into this.
> Hmmm... me neither.  section type conflict?  Does it make any
> different if you move the definitions near the top of the file or use
> a different compiler version?

This is the first report of this kind I've seen, so it mostly works. 
The compiler in question is the stock Centos 5 (.1?) compiler, so it
would be nice to make sure it works as-is.

I've seen section type conflicts with .discard before when discarding a
function and a data type with the same section, which is why I added
general support for .discard.* so they can all get their own sections. 
But that doesn't seem to be the case here.

    J