Re: [RFC] i386: Remove page sized slabs for pgds and pmds

[William Lee Irwin III <wli@holomorphy.com>]

* Christoph Lameter (clameter@sgi.com) wrote:
>> +#ifdef CONFIG_HIGHMEM64G
>> +#define __pgd_alloc()		kmem_cache_alloc(pgd_cache, GFP_KERNEL|__GFP_REPEAT)
>> +#define __pgd_free(pgd)		kmem_cache_free(pgd_cache, pgd)

On Wed, Mar 28, 2007 at 03:26:56PM -0700, Chris Wright wrote:
> I must've glazed over something, I thought this was removal of slabs?

The pgd slab is not fully removable in the PAE case because a dedicated
slab is the only way to enforce alignment for allocations as small as
PAE PGD's.


On Wed, Mar 28, 2007 at 03:26:56PM -0700, Chris Wright wrote:
> BTW, this will interact shared_kernel_pmd patch that Jeremy's posted a
> few times (I know at least wli has looked over that one).  We need to
> make sure that PAE under at least Xen hypervisor has a page-sized pgd,
> although the mmlist chaining looks nice to me.

That, not to mention the total lack of verification of the pageattr.c
code, are among the reasons I didn't want it posted.


* Christoph Lameter (clameter@sgi.com) wrote:
>> +	memcpy(&pgd[USER_PTRS_PER_PGD], &swapper_pg_dir[USER_PTRS_PER_PGD],
>> +						KERNEL_PGD_PTRS*sizeof(pgd_t));

On Wed, Mar 28, 2007 at 03:26:56PM -0700, Chris Wright wrote:
> clone_pgd_range() for consistency?  and it seems we lost a
> paravirt_alloc_pd_clone() in there somewhere.

Yes, another reason why it shouldn't have been posted as-is. It was not
intended to for anything more than comparative benchmarking on systems
without graphics running on the bare metal as opposed to Xen/etc. guests.


-- wli