2005-12-08 04:15:17

by Vivek Goyal

Subject: [PATCH] x86_64: acpi map table fix

o Memory till end_pfn_map has been directly mapped. So all the memory
references to the last page (represented by end_pfn_map) should be
valid.

o I run into problem with kdump when I use memmap=exactmap option and also
pass memmap=X#Y to directly map acpi tables. ACPI initialization in second
kernel fails because some of the valid ACPI memory is not accessible.

o /proc/iomem shows ACPI tables at c7fcb940-c7fcf7ff : ACPI Tables. Here
end_pfn_map is set to c7fcf000. But c7fcf700 should also be a valid access.

Signed-off-by: Vivek Goyal <[email protected]>
---


diff -puN arch/i386/kernel/acpi/boot.c~acpi-map-address-fix arch/i386/kernel/acpi/boot.c
--- linux-2.6.15-rc5-mm1-16M/arch/i386/kernel/acpi/boot.c~acpi-map-address-fix 2005-12-07 15:56:33.000000000 +0530
+++ linux-2.6.15-rc5-mm1-16M-root/arch/i386/kernel/acpi/boot.c 2005-12-07 15:58:04.000000000 +0530
@@ -108,7 +108,7 @@ char *__acpi_map_table(unsigned long phy
if (!phys_addr || !size)
return NULL;

- if (phys_addr < (end_pfn_map << PAGE_SHIFT))
+ if (phys_addr < ((end_pfn_map << PAGE_SHIFT) + PAGE_SIZE))
return __va(phys_addr);

return NULL;
_

2005-12-08 08:37:16

by Andi Kleen

Subject: Re: [PATCH] x86_64: acpi map table fix

On Thu, Dec 08, 2005 at 09:45:09AM +0530, Vivek Goyal wrote:
>
>
> o Memory till end_pfn_map has been directly mapped. So all the memory
> references to the last page (represented by end_pfn_map) should be
> valid.
>
I think the correct test is

if (phys_addr+size <= (end_pfn_map << PAGE_SHIFT) + PAGE_SIZE)
return __va(phys_addr);


I changed it to that. Thanks.

-Andi