r13 is non-volatile?

r13 is non-volatile?

[Kumar Gala]

Was looking at merging some code between ppc32 & ppc64 and noticed  
that in ppc32 we consider r13 to be part of the non-volatile register  
set.  Is this really correct?

r13 is suppose to be sdata pointer in ppc32 and system thread id in  
ppc64.  So I'm wondering should we really be considering it NV on ppc32?


PPC32:

#define SAVE_NVGPRS(base)       SAVE_GPR(13, base); SAVE_8GPRS(14,  
base); \
                                 SAVE_10GPRS(22, base)
#define REST_NVGPRS(base)       REST_GPR(13, base); REST_8GPRS(14,  
base); \
                                 REST_10GPRS(22, base)

PPC64:
#define SAVE_NVGPRS(base)       SAVE_8GPRS(14, base); SAVE_10GPRS(22,  
base)
#define REST_NVGPRS(base)       REST_8GPRS(14, base); REST_10GPRS(22,  
base)

- kumar

Re: r13 is non-volatile?

[Segher Boessenkool]

> Was looking at merging some code between ppc32 & ppc64 and noticed 
> that in ppc32 we consider r13 to be part of the non-volatile register 
> set.  Is this really correct?

I believe so, yes.

> r13 is suppose to be sdata pointer in ppc32

sdata pointer is non-volatile by definition...

> and system thread id in ppc64.

The 64-bit kernel uses GPR13 as the PACA pointer, so it handles
it specially -- it treats the userland values of the register
as non-volatile, though.

> So I'm wondering should we really be considering it NV on ppc32?

Yeah.  I'm sure there is some way to make the macro's identical
between 32- and 64-bit, though.


Segher

Re: r13 is non-volatile?

[linas]

On Thu, Sep 08, 2005 at 07:01:18PM +0200, Segher Boessenkool was heard to remark:
> >Was looking at merging some code between ppc32 & ppc64 and noticed 
> >that in ppc32 we consider r13 to be part of the non-volatile register 
> >set.  Is this really correct?
> 
> >r13 is suppose to be sdata pointer in ppc32
> 
> sdata pointer is non-volatile by definition...
> 
> >and system thread id in ppc64.
> 
> The 64-bit kernel uses GPR13 as the PACA pointer, so it handles
> it specially 

There's a different value in R13 depending on which cpu you're on,
so one mustn't save R13 on one CPU and attempt to restore it on
another.

Maybe, for ppc32, you can pretend that r13 "is like a paca"? So
that instead of saving/restoring r13 like this, you could set it up
at the same time/instead of setting up the paca r13?

--linas

Re: r13 is non-volatile?

[Benjamin Herrenschmidt]

> There's a different value in R13 depending on which cpu you're on,
> so one mustn't save R13 on one CPU and attempt to restore it on
> another.
> 
> Maybe, for ppc32, you can pretend that r13 "is like a paca"? So
> that instead of saving/restoring r13 like this, you could set it up
> at the same time/instead of setting up the paca r13?

ppc32 uses r2 for that, it contains "current".

I don't think we need to bother merging those low level CPU specific
things. We need to have a cpu32 and a cpu64 subdirs in there for things
like these, memory management, etc... imho.

Ben.

Re: r13 is non-volatile?

[Kumar Gala]

On Sep 8, 2005, at 4:53 PM, Benjamin Herrenschmidt wrote:

>
>
>> There's a different value in R13 depending on which cpu you're on,
>> so one mustn't save R13 on one CPU and attempt to restore it on
>> another.
>>
>> Maybe, for ppc32, you can pretend that r13 "is like a paca"? So
>> that instead of saving/restoring r13 like this, you could set it up
>> at the same time/instead of setting up the paca r13?
>>
>
> ppc32 uses r2 for that, it contains "current".
>
> I don't think we need to bother merging those low level CPU specific
> things. We need to have a cpu32 and a cpu64 subdirs in there for  
> things
> like these, memory management, etc... imho.

My intent is to see if we can make a change so that there isn't an  
ifdef.

- kumar