Thor Lancelot Simon a écrit :
> On Sat, Jan 28, 2006 at 01:58:53AM +0000, segv@netctl.net wrote:
> 
>>Hi, I have a script that sets up a chroot environment under /opt/sandbox by
>>calling 'mount_null' to mount /bin, /sbin, /lib, /usr/bin, /usr/sbin, /usr/lib,
>>etc under /opt/sandbox. When some of the directories are exported via NFS and I
>>call this scripts on the server that exports those directories, I get a kernel
>>panic. This is on a pretty recent NetBSD-current.
>>
>>uvm_fault(0xcace89a0, 0, 0, 1) -> 0xe
>>kernel: supervisor trap page fault, code = 0
>>Stopped in pid 543.1 (mountd) at 0: invalid address
> 
> 
> I've seen a huge number of these on multiprocessor machines -- you mention
> in a later message that your system has two CPUs.  Our quad-CPU build
> systems see them very frequently, while our dual-CPU systems see them
> hardly at all.
> 
> It's been suggested that these are memory errors -- that is, hardware
> failures -- but I can't see any other evidence that the memory in these
> systems is bad.  I think we must have some kind of race in the VM system.
> 

I've had many of these on a mono proc system. That was a GENERIC kernel
(3.0 RC5). However, no nfs mounts were used (noir server, nor client).

The faults disappeared after I rebuilt a custom kernel. so the faults
were caused by one of the removed options, but there are too many
possibilities to test (and the faults occured unpredictably, so I can't
just rebuild a new kernel and test...).