On Mon, Aug 20, 2007 at 07:34:18PM +0200, Bernd Ernesti wrote:
> On Mon, Aug 20, 2007 at 04:36:09PM +0200, Bernd Ernesti wrote:
> > On Mon, Aug 20, 2007 at 04:26:45PM +0200, Tobias Nygren wrote:
> > > On Mon, 20 Aug 2007 12:54:57 +0200
> > > Bernd Ernesti <netbsd@lists.veego.de> wrote:
> >
> > [spinning gcc process, sucking 100% cpu cycles]

I had the same issue yesterday while building pkgsrc/sysutils/xosview.
It got struck in the configure script, when "checking for suffix of
object files".

> > > Does this happen if using -O2 instead of -O1? It could be a g++
> > > optimizer bug. (I guess O1 doesn't receive much real-world testing.)

No options were changed.

> And I got a panic while pressing ctrl-c:
> 
> uvm_fault(0xc07f9820, 0xffeec000, 2) -> 0xe
> kernel: supervisor trap page fault, code=0
> Stopped in pid 3380.1 (cc1plus) at      netbsd:pmap_page_remove+0x10e:  xchgl   %edi,0(%eax)
> db> bt
> pmap_page_remove(c1a06240,d4ae97d8,ffffffff,c03704a7,dda1c034) at netbsd:pmap_page_remove+0x10e
> uvm_anfree(d4a7d814,d57df848,3a5a3c,bb0c2000,bb0c2000) at netbsd:uvm_anfree+0x9f
> [...]

I didn't wait that long and terminated it with CTRL-C.

Shortly after that the machine crashed. The process mentioned in the panic
was syslogd.  bt showed one of the x86 cas primitives and then a hole bunch
of (uvm|pmap)_(unmap|destroy|free)() calls.

This machine has no serial console and I hadn't the time to write them down.

It is an Athlon X2 64 running i386/GENERIC.MP from 21.8.2007 and an
userland from 24.7.2007.

    Bernhard  //
            \X/