Problems writing ELF dumps with makedumpfile 1.2.9

[Worth, Kevin]

My "kdump" script (the script that runs as part of the initrd when a crash occurs) tries makedumpfile first, and if that fails, does a simple "cp /proc/vmcore /var/crash/vmcore". That copy always seems to succeed and is 4GB in size. I can run a check tomorrow to see how big /proc/vmcore appears to be, if this information is not sufficient.

-Kevin
________________________________________
From: Ken'ichi Ohmichi [oomichi at mxs.nes.nec.co.jp]
Sent: Tuesday, September 30, 2008 7:18 PM
To: Worth, Kevin
Cc: Masaki Tachibana; kexec-ml
Subject: Re: Problems writing ELF dumps with makedumpfile 1.2.9

Hi Kevin,

Worth, Kevin wrote:
> I've attached the log files from running strace.
> Hopefully they contain some clue as to where this failure is occurring.
> I tried specifying -d30 and it still failed. Both traces attached.

Thank you for some log files.

vmcore-d30.log:
  open("/proc/vmcore", O_RDONLY)          = 4
  [snip]
  lseek(4, 2885288664, SEEK_SET)          = -1 EINVAL (Invalid argument)

Your system has 4GB memory, and /proc/vmcore size is almost 4GB, right ?
If yes, the above lseek() should *not* fail because the offset (almost 2.7GB)
is in /proc/vmcore. So I guess that this problem is due to a kernel.

Please try copying /proc/vmcore by `cp` command, and check whether the
size of copied file is the same as /proc/vmcore.


Thanks
Ken'ichi Ohmichi