5/20 NetBSD Xen HVM 3.0.4 crashed

Hi,

FYI, 5/20 NetBSD as DOM0 Xen 3.0.4 with HVM crashed
whenever
I tried to start HVM (actually trying to start WinXP). 
4/30
NetBSD worked fine.  Both are using the same xenkernel,
xentools, xentools-hvm.

Last time, I reported the crash in "mmapbatch: remap
error
14! ???" mail, but mmapbatch is not related to the
problem.
With 4/30 NetBSD, I'm still having DomU locks-up problem.

Is there any good-to-try version?  hehe.  I'll try to find
it out.

Regards,
-- Jam
Love is an ideal thing, marriage a real thing; a confusion
of the real with
the ideal never goes unpunished.
		-- Johann Wolfgang von Goethe

On Sat, May 26, 2007 at 05:04:02PM +0900, Kazushi Marukawa
wrote:
> Hi,
> 
> FYI, 5/20 NetBSD as DOM0 Xen 3.0.4 with HVM crashed
whenever
> I tried to start HVM (actually trying to start WinXP). 
4/30
> NetBSD worked fine.  Both are using the same
xenkernel,
> xentools, xentools-hvm.
> 
> Last time, I reported the crash in "mmapbatch:
remap error
> 14! ???" mail, but mmapbatch is not related to the
problem.
> With 4/30 NetBSD, I'm still having DomU locks-up
problem.
> 
> Is there any good-to-try version?  hehe.  I'll try to
find
> it out.

Maybe 4.0_BETA2. I also get the mmapbatch: remap error but
it
doesn't prevent my DomUs from running.

-- 
Manuel Bouyer <bouyerantioche.eu.org>
     NetBSD: 26 ans d'experience feront toujours la
difference
--

On Sat, May 26, 2007 at 10:46:33PM +0200, Manuel Bouyer
wrote:
> On Sat, May 26, 2007 at 05:04:02PM +0900, Kazushi
Marukawa wrote:
> > Hi,
> > 
> > FYI, 5/20 NetBSD as DOM0 Xen 3.0.4 with HVM
crashed whenever
> > I tried to start HVM (actually trying to start
WinXP).  
> 

I'm seeing the same thing, though I don't know when it
started - for a
variety of reasons it had been a while since I'd last used
HVM..  I
can run normal xen domU's just fine, but some moments after
starting a
HVM domU I get a panic.

It's tripping the KASSERT(read_psl == 0) at pmap.c:2189. 
Call path is
sys_select -> copyout -> pmap_load with qemu-dm as the
calling
process.

> I also get the mmapbatch: remap error but it
> doesn't prevent my DomUs from running.

Me too, and I have done for a long time, regardless of this
panic.

--
Dan.

On Wed, Jun 13, 2007 at 02:33:37PM +1000, Daniel Carosone
wrote:
> I'm guessing there's some mismatching spl or lock
handling somewhere

Found in the syslog from that run:

  cpu_setmcontext error: uc EFL: 0x00210006 tf EFL:
0x00200246 uc CS: 0x10017

which comes from xen/i386/machdep.c:2290; the comments right
above
there strongly suggest the processor has been left in the
wrong state
and the original KASSERT was doing its job to notice that. 


--
Dan.

On Wed, Jun 13, 2007 at 02:10:13PM +1000, Daniel Carosone
wrote:
> It's tripping the KASSERT(read_psl == 0) at
pmap.c:2189.  Call path is
> sys_select -> copyout -> pmap_load with qemu-dm
as the calling
> process.

Observing that nothing else in the file seems to look at
read_psl, I
decided to try the "brute force and ignorance"
experiment of just
commenting out that line and seeing what happens.

The HVM now starts, and seems to pretty much run as normal.
I can see
evidence of activity from the disk light and so on as the
domU starts
up and runs, but clearly something else is not right because
the
Xserver gets very sluggish to respond to the mouse, and
quickly
freezes.

I'm guessing there's some mismatching spl or lock handling
somewhere
in the path, potentially introduced with the merge of
idlelwp?

--
Dan.

With the latest interrupt changes, my HVM domU is now
working properly
again, thanks heaps!

--
Dan.