On 12/11/06, Ralf Baechle <ralf@xxxxxxxxxxxxxx> wrote:
On Wed, Dec 06, 2006 at 04:48:29PM +0100, Franck Bui-Huu wrote:
> The old code was assuming that min_low_pfn was always 0. This
> means that we can't handle platforms with a big hole at start
> of memory since mem_map[] size would blew up.
It was - and IP22 was paying a high price for that. It's currently 1MB
on 32-bit and 1.75MB on 64-bit kernels - but used to be much higher in
the past.
Btw, I think you're confusing the terms here; your patch description reads
wrong while the patch looks fine. PFN is page frame number, not physical
frame number. To avoid wasting dead mem_map[] entries idealy the pfn for
the first allocatable page should be zero.
So ignoring NUMA and discontig memory for a moment (those make everything
more complicated ...) PFN 0 is the first entry in mem_map[] - which usually
but not necessarily is physical address 0x0.
well if you look at the generic definition of pfn_to_page(), you find:
#define __pfn_to_page(pfn) (mem_map + ((pfn) - ARCH_PFN_OFFSET))
With the current implementation, PFN 0 is something that does not
exist. PFN ARCH_PFN_OFFSET is the first entry in mem_map[]...
I think I'm missing your point....
> This patch does not relax this constraint but it's a first
> step to achieve that.
> diff --git a/arch/mips/kernel/setup.c b/arch/mips/kernel/setup.c
> index 89440a0..8e58d7f 100644
> --- a/arch/mips/kernel/setup.c
> +++ b/arch/mips/kernel/setup.c
> @@ -271,7 +271,6 @@ static void __init bootmem_init(void)
> static void __init bootmem_init(void)
> {
> unsigned long reserved_end;
> - unsigned long highest = 0;
> unsigned long mapstart = -1UL;
Assigning a negative number to an unsigned long variable ...
Well, it's a convenient way to set mapstart to 0xffffffff on both
32bits and 64bits kernels as you guessed. I think it's quite readable
thanks to the 'UL' suffix. But I'll change the whole line with:
unsigned long mapstart = ~(0UL);
> unsigned long bootmap_size;
> int i;
> @@ -284,6 +283,13 @@ static void __init bootmem_init(void)
> reserved_end = max(init_initrd(), PFN_UP(__pa_symbol(&_end)));
>
> /*
> + * max_low_pfn is not a number of pages. The number of pages
> + * of the system is given by 'max_low_pfn - min_low_pfn'.
> + */
> + min_low_pfn = -1UL;
Assigning a negative number to an unsigned long variable ...
ditto
thanks
--
Franck
