-- On 1/8/07, Ralf Baechle <ralf@xxxxxxxxxxxxxx> wrote:
On Mon, Jan 08, 2007 at 06:44:51PM +0100, Franck Bui-Huu wrote:
> This patch makes a better usage of these two globals.
> 'min_low_pfn' is now correctly setup for all configs, which
> allow us to rely on it in boot memory code init.
[snip]
Call Trace:
[<80109484>] dump_stack+0x8/0x34
[<80159fbc>] bad_page+0x68/0xa8
[<8015a804>] free_hot_cold_page+0x3a4/0x3b4
[<80414150>] free_all_bootmem_core+0x150/0x258
[<8040aefc>] mem_init+0x40/0x1c0
[<80400940>] start_kernel+0x2b4/0x3c8
[snip]
From a working kernel:
Determined physical RAM map:
memory: 00001000 @ 00000000 (reserved)
memory: 000ef000 @ 00001000 (ROM data)
memory: 0035a000 @ 000f0000 (reserved)
memory: 03bb6000 @ 0044a000 (usable)
So it's basically 64MB starting at physical 0.
Oh and Malta does not use CONFIG_HIGHMEM.
Yeah I just noticed that, Malta supports highmem but actually doesn't
use it.
Regarding to your mapping, usable ram starts at 0x44a000. So
min_low_pfn is now set to the corresponding pfn and the bootmem alloc
assumes it's the begining of your memory. This is done by
bootmem_init():
bootmap_size = init_bootmem_node(NODE_DATA(0), mapstart,
min_low_pfn, max_low_pfn);
However, old code always assumed that the begining of the memory was
0. That was done by:
bootmap_size = init_bootmem(mapstart, highest);
After a while paging_init() is called which in turn calls:
free_area_init(zones_size);
which expands to:
free_area_init_node(0, NODE_DATA(0), zones_size,
__pa(PAGE_OFFSET) >> PAGE_SHIFT, NULL);
In your case "__pa(PAGE_OFFSET) >> PAGE_SHIFT" results in 0 since you
haven't applied patch #2 and setup PHYS_OFFSET. So now, we're assuming
that the start of physical mem is 0 and I think it's wrong since we
previously assumed a different value.
Could you try to apply patch #2 on top of patch #1 and give it a try ?
I expect your kernel to panic in a different way. This time you should
be shooted by the following sanity check added by patch #2:
if (min_low_pfn != ARCH_PFN_OFFSET)
panic("PHYS_OFFSET(%lx) and your mem start(%lx) don't match",
PHYS_OFFSET, PFN_PHYS(min_low_pfn));
If so, could you give another try with PHYS_OFFSET defined as follow:
#define PHYS_OFFSET 0x44a000
BTW mem_map[] size should be 32Kb smaller.
thanks,
Franck
