[Skiboot] [PATCH 03/13] cpu.h: implement this_cpu for clang

[Joel Stanley]

On 3 May 2018 at 17:11, Joel Stanley <joel at jms.id.au> wrote:
> On 3 May 2018 at 12:43, Joel Stanley <joel at jms.id.au> wrote:
>> On 3 May 2018 at 07:19, Balbir Singh <bsingharora at gmail.com> wrote:
>>> On Wed, 2018-05-02 at 18:07 +0930, Joel Stanley wrote:
>>>> Clang can't use the global register variable, so instead load struct
>>>> cpu_thread from r13 and return it.
>>>>
>>>
>>> How do we know clang is not using r13 for anything else? Does -ffixed-reg
>>> work with r13?
>>
>> That is a really good question. I was really surprised (excited!) that
>> it booted once I had fixed all of the issues I knew about, I forgot to
>> go back and ensure clang was reserving r13.
>>
>
> Here's the answer:
>
> https://github.com/llvm-mirror/llvm/blob/ff51e42075425aa13ef7d6f5adbe408ace59b8d6/lib/Target/PowerPC/PPCRegisterInfo.cpp#L282
>
> // On PPC64, r13 is the thread pointer. Never allocate this register.
> if (TM.isPPC64())
>     markSuperRegs(Reserved, PPC::R13);
>
> So there we go. I will add this information to the commit message.
>

Even better, this patch and the other one to cpu.h can be dropped as
clang supports the global register variable.

I think clang was warning about this back before I was properly
telling it to build for ppc64, where r13 is allowed to be used by a
global register variable (as clang reserves, as we discover above).

Cheers,

Joel