[PATCH/RFC 2/2] Cache AT_PLATFORM and AT_L1D_CACHESHAPE from auxvec

[Carmelo AMOROSO]

On 4/17/2011 6:57 PM, Kevin Cernekee wrote:
> On Sun, Apr 17, 2011 at 1:56 AM, Carmelo Amoroso <carmelo73 at gmail.com> wrote:
>> I'm really interested into such a thing. I've posted months ago a patch for SH4 to retrieve Cache info through auxvect
>> as well (also pagesize).
> 
> I do remember seeing your patch on the list.  Didn't realize it was
> never merged.
> 

my fault.... too overloaded.

>> I think we should do this in a dl_aux_init as in my patch instead of polluting
>> all these stuff in __uClibc_main.
> 
> I agree conceptually with reading all auxvect variables in a dedicated
> function instead of putting an ever-growing list in __uClibc_main().
> And I did originally consider _dl_aux_init(), until I saw that
> dl-support.c is only used for static builds of libc.  Maybe there is
> another place to put the code?
> 

I've modified the build system to use dl-support.c for shared libs too
indeed.

> One thing to consider is how much of this code can (and should) be
> shared between ld.so and libc.  There were a couple of different
> places where the auxvect information was getting grabbed/parsed and
> maybe it does make sense to consolidate it somehow.
> 

> In the case of memcpy(), one would want to be able to retrieve the
> auxvect values on both shared and static versions of libc.  This might
> wind up looking a little different from what needs to be done to
> extract your SHMLBA value for dynamic linking purposes.
> 


I made dl_aux_init usable in both contexts.

> Regarding this part of your patch:
> 
> +/* Starting with version 2.6.25-rc1, dcache shape info are passed in auxvect */
> +#if __LINUX_KERNEL_VERSION >= KERNEL_VERSION(2,6,25) && defined __sh__
> +#define __ASSUME_DCACHESHAPE_AUXVECT	1
> +#endif
> 

this is needed specifically for choosing the proper implementation of
SHMLBA.

> Is that something that could be figured out at runtime instead, based
> on whether there is an auxvect entry with the AT_L1D_CACHESHAPE tag?
> 

this is an option that could be investigated

> I ask, because my architecture doesn't have AT_L1D_CACHESHAPE
> implemented in mainline yet...

I see. I'll review my code and re-post. I'd ask you to see if it works
for you.

Thanks,
Carmelo

> _______________________________________________

> uClibc mailing list
> uClibc at uclibc.org
> http://lists.busybox.net/mailman/listinfo/uclibc
>