Re: [PATCH] bootmem.c clean up bad pfn convertions

[Dave Hansen <haveblue@us.ibm.com>]

On Fri, 2005-04-08 at 15:32 +0200, Franck Bui-Huu wrote:
> As I described in my previous email, bootmem.c does improper
> pfn convertions into phys addr. This simple patch fixes that.
...
> -       bdata->node_bootmem_map = phys_to_virt(mapstart << PAGE_SHIFT);
> -       bdata->node_boot_start = (start << PAGE_SHIFT);
> +       bdata->node_bootmem_map = phys_to_virt(pfn_to_phys(mapstart));
> +       bdata->node_boot_start = pfn_to_phys(start);

The only arch with phys_to_pfn() defined is UML, so the patch simply
won't compile anything but UML on current kernels (unless I'm missing
something).

Could you try to give us a more complete description of your problem?  I
know your memory doesn't start at 0x0, but what problems does that
cause?  Does the mem_map[] allocation blow up, etc...  

If it's just mem_map[], That calculation could be fixed pretty easily.
Something like

+#ifdef CONFIG_CRAZY_MIPS_FOO_MEM_MAP_START... 
+extern unsigned long mem_map_start_pfn
+#else
+#define mem_map_start_pfn 0UL
+#endif
-#define pfn_to_page(pfn)        (mem_map + (pfn))
+#define pfn_to_page(pfn)        (mem_map + (pfn) - mem_map_start_pfn)

(those names are horrid, please improve them, if you plan to do this)

All of the zone (and allocator) calculations should be just fine,
because it already has a zone_start_pfn.

-- Dave