Subject: Re: savecore(8) doesn't find dump
To: Jukka Salmi <j+nbsd@2008.salmi.ch>
From: Steven M. Bellovin <smb@cs.columbia.edu>
List: current-users
Date: 01/04/2008 01:39:47
On Fri, 4 Jan 2008 01:28:40 +0100
Jukka Salmi <j+nbsd@2008.salmi.ch> wrote:

> Steven M. Bellovin --> current-users (2008-01-03 11:15:22 -0500):
> > On Thu, 3 Jan 2008 16:48:42 +0100
> > Jukka Salmi <j+nbsd@2008.salmi.ch> wrote:
> > 
> > > Hello,
> > > 
> > > on a 4.99.47 i386 system, savecore(8) is unable to save a dump
> > > even though the dump seems to have been created successfully:
> > > 
> > > 	$ shutdown -d now
> > > 	Shutdown NOW!
> > > 	[...]
> > > 	syncing disks... done
> > > 	unmounting file systems... done
> > > 	
> > > 	dumping to dev 0,1 offset 1141767
> > > 	dump 1014 1013 1012 [...] 3 2 1 succeeded
> > > 	
> > > 	
> > > 	rebooting...
> > > 
> > > ...but during the next boot savecore(8) doesn't find any dump.
> > > 
> > Try running 'savecore -v'; it may say what the problem is.
> > Possibly, adding -f will let you get your dump.
> 
> $ savecore -n
> $ echo $?
> 1
> 
> $ savecore -v /a/crash
> dumplo = 584584704 (1141767 * 512)
> savecore: kvm_read: invalid translation (invalid PTE)
> 
> $ savecore -vf /a/crash
> dumplo = 584584704 (1141767 * 512)
> savecore: kvm_read: invalid translation (invalid PTE)
> savecore: warning: /netbsd version mismatch:
> 	NetBSD 4.99.48 (GENERIC) #0: Thu Jan  3 22:49:59 CET 2008
> 	build@moray.salmi.ch:/b/build/nbsd/c/i386/sys/arch/i386/compile/GENERIC
> 
> and	7
> 
> savecore: kvm_read: invalid translation (invalid PTE)
> savecore: kvm_read: invalid translation (invalid PTE)
> savecore: reboot
> savecore: kvm_read: invalid translation (invalid PTE) (and
> _time_second is not defined also) savecore: writing core
> to /a/crash/netbsd.4.core savecore: writing kernel
> to /a/crash/netbsd.4 savecore: /dev/wd0b: kvm_clear_dump: invalid
> translation (invalid PTE)
> 
> 
> This seems to be the same problem Antti mentioned. I'll file a PR.

And the same problem I've been having.  Since my crash was due to a
uvm_fault, I assumed that the bad PTE was a legitimate issue.  

		--Steve Bellovin, http://www.cs.columbia.edu/~smb