Crash during vmcore_init

[Américo Wang]

On Tue, Nov 15, 2011 at 4:14 PM, Dave Young <dyoung at redhat.com> wrote:
>> On Mon, 2011-11-14 at 13:39 +0000, WANG Cong wrote:
>>> On Tue, 11 Oct 2011 16:39:05 -0700, Tim Hartrick wrote:
>>>
>>>> Kexec,
>>>>
>>>> I have been experiencing the crash below on Ubuntu 10.04 running
>>>> 2.6.32-34-server and 2.6.38-8-server as the crashkernel on X86_64. The
>>>> tools are:
>>>>
>>>> kexec-tools 1:2.0.2-1ubuntu3
>>>> makedumpfile        1.3.7-2
>>>> kdump-tools 1.3.7-2
>>>>
>>>> I would be interested to know if this is a known problem and if so
>>>> whether or not there is a patch in the pipeline to correct the problem.
>>>>
>>>> I will be happy to provide any other details that are required including
>>>> debug builds if necessary.
>>> ....
>>>>
>>>> [    1.322100] ioremap: invalid physical address db74000000000000 [
>
>
> Searching db74000000000000 got several similar cases of this, all are
> about per cpu invalid crash_notes address, is this one more?
>

Why it is per cpu crash_note?

I bet Tim is using i386 machine, and on i386 we have,

kexec/arch/i386/crashdump-x86.c

static int get_crash_notes(int cpu, uint64_t *addr, uint64_t *len)
{
        char crash_notes[PATH_MAX];
        char line[MAX_LINE];
        FILE *fp;
        unsigned long vaddr;
        int count;

        sprintf(crash_notes, "/sys/kernel/crash_notes");
        fp = fopen(crash_notes, "r");
        if (fp) {
               <...>
        } else
                return get_crash_notes_per_cpu(cpu, addr, len);
}

As long as we have .notes section, /sys/kernel/crash_notes exists.
So I see no reason that is per cpu notes stuff.