Re: Port amd64 - kernel crashing

[Andrew Doran <ad%netbsd.org@localhost>]

On Sat, Nov 15, 2008 at 12:10:22PM -0800, Paul Goyette wrote:

> I've repeated this three times, and it panics every time, although at
> different points in the pkg_rolling-replace process.

Are you confident that the hardware is stable? Is the BIOS recent?

> I'm suspecting this might be yet another NFS issue, but I have no clue 
> for debugging it further.
... 
> Console output transcribed manually, as I have no serial console.
> 
> panic: pmap_removes_ptes: unmanaged page marked PG_VLIST, 
> va=0x7f7ffd996000, pa=0x415eb66000
...
> panic() at netbsd:panic+0x260
> pmap_do_remove() at netbsd:pmap_do_remove+0x561
> uvm_unmap_remove() at netbsd:uvm_unmap_remove+0x217
> uvm_space_free() at netbsd:uvm_space_free+0x10b
> exit1() at netbsd:exit1+0x1b1
> sys_exit() at netbsd:sys_exit+0x53
> syscall() at netbsd:syscall+0xb6

If you want to debug it I would suggest beginning by finding out what
uvm_unmap_remove() is trying to unmap when the panic occurs. There are two
calls to pmap_remove(), ignore the call to pmap_kremove(). The map entry
will be either an amap (anonymous memory, entry->aref.ar_amap) or a
uvm_object (UVM_ET_ISOBJ). If it is a uvm_object, what does the uvm_object
describe? Is it a vnode (uvm_vnodeops) or a device (uvm_deviceops) or what?

You could probably get this from gdb on a crashdump by compiling the kernel
with -O0.

Cheers,
Andrew