kern/85931: panic: "vm_fault: fault on nofault entry" when using md

[Nate Eldredge]

>Number:         85931
>Category:       kern
>Synopsis:       panic: "vm_fault: fault on nofault entry" when using md
>Confidential:   no
>Severity:       serious
>Priority:       medium
>Responsible:    freebsd-bugs
>State:          open
>Quarter:        
>Keywords:       
>Date-Required:
>Class:          sw-bug
>Submitter-Id:   current-users
>Arrival-Date:   Sat Sep 10 00:10:15 GMT 2005
>Closed-Date:
>Last-Modified:
>Originator:     Nate Eldredge
>Release:        FreeBSD 5.4-RELEASE-p6 amd64
>Organization:
>Environment:
System: FreeBSD vulcan.lan 5.4-RELEASE-p6 FreeBSD 5.4-RELEASE-p6 #0: Sun Sep 4 02:52:11 PDT 2005 nate at vulcan.lan:/usr/obj/usr/src/sys/VULCAN amd64


Opteron, 1 GB ram.

>Description:

I encountered a kernel panic when compiling the gcc40 port on a memory
disk.

As a benchmark test, I wanted to compile gcc on a ramdisk.  
"mdconfig -t malloc" cannot create such a large ramdisk, but I
discovered after some research that you can use a preloaded ramdisk to
accomplish this.  So I created a 600MB sparse file in / (it had to be
sparse because that partition is only 400MB large), rebooted and issued

load -t md_image /sparse.img
boot

at the loader prompt.  Then did along the lines of

newfs /dev/md0
mount /dev/md0 /mnt/md0
mkdir /mnt/md0/nate  # so as not to have to compile as root
chown nate:nate /mnt/md0/nate
su nate
cd /mnt/md0/nate
cp -r /usr/ports/lang/gcc40 .
cd gcc40
time make

After about 20 minutes the kernel panicked with the message

vm_fault: fault on nofault entry, addr: ffffffffb36d8000

I have a vmcore file and can debug further if necessary.  I can also
test patches if I get them reasonably soon (this machine has to go
into production eventually).

On a previous attempt to do this, the machine rebooted after about 20
minutes but I did not get there in time to see if there was a panic
message.  So on the second attempt I watched and also enabled dumping.
Thus it seems this may be reproducible.

This is a new machine, so hardware problems are conceivable.  But I
had previously compiled gcc on disk several times as a burn-in test,
without any problems.  I also ran memtest86 overnight.  The memory is
ECC as well.

>How-To-Repeat:

See above.

>Fix:

I don't know.

Thanks.

>Release-Note:
>Audit-Trail:
>Unformatted: