[PATCH] [makedumpfile] Fix ELF output with overlapping sections

Ken'ichi Ohmichi oomichi at mxs.nes.nec.co.jp
Fri Jun 1 05:08:27 EDT 2007


Hi Bernhard,

2007/05/31 17:18:42 +0200, Bernhard Walle <bwalle at suse.de> wrote:
>* Ken'ichi Ohmichi <oomichi at mxs.nes.nec.co.jp> [2007-05-31 04:26]:
>> 2007/05/18 12:47:15 +0200, Bernhard Walle <bwalle at suse.de> wrote:
>> >I have a core dump with following program headers on IA64:
>> >
>> >  LOAD           0x0000000000001f4c 0xa000000100000000 0x0000000004000000
>> >                 0x000000000055c4a0 0x000000000055c4a0  RWE    0
>> >...
>> >  LOAD           0x00000000044fc3ec 0xe000000004000000 0x0000000004000000
>> >                 0x0000000000c92000 0x0000000000c92000  RWE    0
>> >
>> >The interesting thing is the overlap in the physical address space.
>> >write_elf_pages() assumes that there's no overlap because it looks only
>> >for the file offset according to the physical start address and silently
>> >assumes that the length matches.
>> 
>> Thank you for the patch.
>> I'm worry that the overlapping spaces may have the same p_filesz.
>> I think it is certain to check it by p_offset, and created the
>> attached patch. Please let me know your opinion.
>> 
>> 
>> BTW, I don't know the reason why your system's /proc/vmcore has the
>> overlapping physical address. If you know it, please let me know.
>
>thanks for the patch. It looks more safe than my attempt to fix it.
>I tested your patch on the machine on which I originally had my
>problem. It worked without any problems.
>
>So I vote for applying the patch to your code base. :-)

Thank you for the test !
I will merge it.


Thanks
Ken'ichi Ohmichi



More information about the kexec mailing list