Failure to crashdump under Centos-6.4 x86_64 with kernel 3.12.30

[Dave Anderson]

----- Original Message -----

> My kernel crashdumps don't work with vanilla 3.12.30 on a 256GB x86_64 intel box.
> Centos-6.4. With the standard makedumpfile, makedumpfile segfaults, and no output is written to disk.
> with makedumpfile 1.5.7, kexec-tools-2.0.8, and crash 7.0.2 I actually get vmcore files,
> but then crash rejects them with:
> 
> crash vmcore /boot/vmlinux-3.12.30
> 
> crash 7.0.2
> ...
> GNU gdb (GDB) 7.6
> ...
> 
> crash: seek error: kernel virtual address: ffff88207fffcb60  type: "pglist node_id"

With respect to the crash error, since it's a seek error, it means that the page for
physical address 0x207fffcb60 (~130GB) is not included in the dumpfile.  Are you getting any 
other error messages prior to that one?  Getting some debug output might show something
else of interest, for example by entering:

  $ crash -d4 vmcore /boot/vmlinux-3.12.30

Also, for a sanity check, what happens when you do this: 

  $ crash --minimal vmcore /boot/vmlinux-3.12.30
  ...
  crash> rd -a linux_banner 

It should show something like this (taken from the only 3.12 dumpfile I have):

  crash> rd -a linux_banner
  ffffffff818000a0:  Linux version 3.12.0-rc2+ (root at vm9-f19) (gcc version 4.8.1 
  ffffffff818000dc:  20130603 (Red Hat 4.8.1-1) (GCC) ) #2 SMP Wed Nov 6 13:25:13
  ffffffff81800118:   EST 2013
  crash> 

Dave 

 
> 
> the grub entry for this kernel is:
>      kernel /boot/vmlinuz-3.12.30 ro root=LABEL=ROOT
> numa_balancing=disable intel_idle.max_cstate=0 processor.max_cstate=0
> mce=ignore_mce clocksource=tsc nohz=off nomodeset nodmraid rd_NO_LUKS
> rd_NO_LVM rd_NO_MD rd_NO_DM LANG=en_US.UTF-8 SYSFONT=latarcyrheb-sun16
> KEYBOARDTYPE=pc KEYTABLE=us crashkernel=256M at 256M  console=ttyS1,115200n8
> 
> 
> CONFIG_CRASH_DUMP=y
> CONFIG_KEXEC_JUMP=y
> CONFIG_PHYSICAL_START=0x2000000
> CONFIG_RELOCATABLE=y
> CONFIG_PHYSICAL_ALIGN=0x1000000
> 
> A similar issue appears with a grub crashkernel=128M at 32M
> 
> Turning off all the makedumpfile exclusions (dump all pages) results in a much larger
> vmcore, but it still gets the same failures.
> 
> Where do I start looking for the problems here?
> kernel 3.10.65 won't dump either, but an old 3.0.29 kernel dumps just fine.
> Is this a problem with some .config setting?
> 
> Berkley