RE: Pistachio: Code for parsing the ACPI tables doesn't map everyphysical pointer

Frank,

There is a (not-yet-checked-in) patch that solves the problem by remapping
the ACPI tables.  Maybe Jan could add that to the tree...

- Volkmar

> -----Original Message-----
> From: Frank Mehnert
> Sent: Monday, April 10, 2006 11:58 AM
> 
> Hi,
> 
> when searching the ACPI tables for, pointers to physical 
> pointers are not mapped in every case. Look at 
> acpi__std_t::find(): This code assumes that ptrs[] contains 
> pointers to physical pages which are already mapped. This is 
> not necessary guaranteed. My test box fails to boot when APIC 
> support is enabled:
> 
>   kmem_init (f0132000, f0152000) [128K]
>   remap area f0000000->00000000 len=400000, w=1, k=1, size=1
>   remap area 00000000->00000000 len=400000, w=1, k=1, size=1
>   remap area 00000000->00000000 len=400000, w=1, k=1, size=1
>   activate_gdt
>   remap area fe000000->0e000000 len=1000000, w=1, k=1, size=1
>   RSDP is at 000feb00
>   RSDT is at 000fd211
>   XSDT is at 000fd265
> 
>   headers.len = 0x5C => 7 ptrs
>   ptrs[0] = 000fd35d
>   ptrs[1] = fffcaf60
>   ptrs[2] = 000fd451
>   ptrs[3] = 000fd4c3
>   ptrs[4] = 000fd4eb
>   ptrs[5] = 000fd552
>   ptrs[6] = 000fd590
>   ptrs[7] = 00000000
> 
> The value of ptrs[2] points to a physical memory page which 
> is still not mapped. This examples shows furthermore that it 
> is dangerous to map the ACPI tables one-by-one (conflicts 
> with kernel region).
> 
> Frank
> --
> ## InnoTek Systemberatung GmbH, http://www.innotek.de ##
>