Re: [PATCH] powerpc: enable CPU_FTR_CI_LARGE_PAGE for cell

Re: [PATCH] powerpc: enable CPU_FTR_CI_LARGE_PAGE for cell

[Benjamin Herrenschmidt]

On Sat, 2006-06-17 at 17:59 +0000, Linux Kernel Mailing List wrote:
> commit ce221982e0bef039d7047b0f667bb414efece5af
> tree fa01b712522338d3f19ee5a6fedace7b7149c430
> parent 19242b240793ac769f5b91b68a5e43dd39f0c530
> author Arnd Bergmann <arnd.bergmann@de.ibm.com> Thu, 15 Jun 2006 15:09:16 +0200
> committer Linus Torvalds <torvalds@g5.osdl.org> Sun, 18 Jun 2006 00:56:24 -0700
> 
> [PATCH] powerpc: enable CPU_FTR_CI_LARGE_PAGE for cell
> 
> Reflect the fact that the Cell Broadband Engine supports 64k
> pages by adding the bit to the CPU features.

Are you sure you want that in ? The SPU code upstream isn't ready for
64k pages yet... I spotted at least:

__spu_trap_data_seg() and get_kernel_slb()

Those need to encode the proper page size. I think you have patches for
that already. I wouldn't enable 64k pages with the abvoe without these
as you may end up with infinite hash fault loops due to the mismatch
between page size encoding in the hash table an in the SLB.

Ben.

Re: [PATCH] powerpc: enable CPU_FTR_CI_LARGE_PAGE for cell

[Arnd Bergmann]

Am Sunday 18 June 2006 00:29 schrieb Benjamin Herrenschmidt:
> Are you sure you want that in ? The SPU code upstream isn't ready for
> 64k pages yet... I spotted at least:
>
> __spu_trap_data_seg() and get_kernel_slb()
>
> Those need to encode the proper page size. I think you have patches for
> that already. I wouldn't enable 64k pages with the abvoe without these
> as you may end up with infinite hash fault loops due to the mismatch
> between page size encoding in the hash table an in the SLB.

It doesn't make much of a difference. The upstream spufs is broken
for 64k pages regardless of whether they are HW or compound pages.

The other patch was for making HW 64k pages work at all, as is this
one.

	Arnd <><