Re: Question about kimage_alloc_page in kexec.c, bug?

[Américo Wang <xiyou.wangcong@gmail.com>]

On Mon, Oct 11, 2010 at 10:46:59PM +0800, wilbur.chan wrote:
>Hi all !
>
>I have a question on kimage_alloc_page:
>
...

You can Cc me on kexec/kdump issues. :)

>
>
>If  a page frame is right at physical address 0x20000000 ,  then when calling
>
>machine_kexec , the following code:
>
>*ptr = (unsigned long) phys_to_virt(*ptr);
>
>will generate an virtual address of 0xa0000000 = 0x20000000+0x80000000 , which
>
>results in physical address 0.
>
>
>
>should it be like this?  :
>
>--- kexec.c     2010-08-27 07:47:12.000000000 +0800
>+++ kexec.changed.c     2010-10-11 22:04:08.000000000 +0800
>@@ -723,7 +723,7 @@
>                if (!page)
>                        return NULL;
>                /* If the page cannot be used file it away */
>-               if (page_to_pfn(page) >
>+               if (page_to_pfn(page) >=
>                                (KEXEC_SOURCE_MEMORY_LIMIT >> PAGE_SHIFT)) {
>                        list_add(&page->lru, &image->unuseable_pages);
>                        continue;
>

No, this could break x86_64.

IMHO, what needs to fix is the definition of KEXEC_SOURCE_MEMORY_LIMIT
on MIPS, something like:

- #define KEXEC_SOURCE_MEMORY_LIMIT (0x20000000)
+ #define KEXEC_SOURCE_MEMORY_LIMIT (0x1FFFFFFF)


Thanks.