Re: [PATCH] PAE36 further diagnosis: Fix for 2.3.42

[Manfred Spraul]

"Stephen C. Tweedie" wrote:
> 
> On Wed, 09 Feb 2000 21:19:57 +0100, Manfred Spraul
> <[e-mail]> said:
> 
> 
> > Could you enable ACPI?
> 
> Yes, I have no problems with that enabled, even using the full 8GB.
> 
I think I found the problem:
acpi tries to ioremap() memory that's just behind the end of the normal
memory:

I have 192 MB ram, and the last few bytes are used for the ACPI tables.

acpi calls ioremap(0xbff3000,36,flags).
high_memory is bff3000 --> remap_area_pages() gets called, but the page
is still mapped by a 2MB pte.

it seems that remap_area_pages() can't handle PSE pages -->
pte_alloc_kernel() calls pmd_bad(), and that function complains.

A few seconds later, the kernel runs into a BUG().

Could someone check what these remap_area_xy() functions should do? The
modifications to phys_address are interesting:

	phys_address -= address;
	[..., no access to phys_address]
	sub_func(,phys_address+address,,);

Without PAE36, the kernel runs smoothly.
--
	Manfred