Scheduling deadlock on alpha?

[Bernd Walter]

On Sun, Feb 15, 2004 at 10:36:51PM -0800, Kris Kennaway wrote:
> On Mon, Feb 16, 2004 at 07:18:22AM +0100, Bernd Walter wrote:
> > On Sun, Feb 15, 2004 at 09:58:40PM -0800, Kris Kennaway wrote:
> > > One of the alpha package machines seems to have deadlocked; no
> > > processes are running, but the kernel itself responds to pings, and I
> > > can break to DDB:
> > 
> > How long have you waited?
> > It seems that the system was busy waiting for page faults and swap.
> > I've seen conditions lately where this can block the system for a very
> > long time in case there is not enough memory free.
> > The number of processes point to memory stress.
> 
> It's been about 1/2 hour now...this machine doesn't seem to have been
> under exceptional load compared to the other equivalent alpha package
> machines

OK - it was not _that_ long, but it was long enough for me to to really
think about a hanging machine once.

I'm no VM expert, but those two look very strange to me:
    7 fffffc0003ca4b40 fffffe0004f38000    0     0     0 0000204 [SLP]VMWait 0xfffffc000068c468] pagedaemon
    0 fffffc0000693178 fffffc0000796000    0     0     0 0000200 [SLP]vmwait 0xfffffc00006e4028] swapper

At least I think that all others are awaiting them to do their job
first.
Not to mention about the different writing of the same word.

-- 
B.Walter                   BWCT                http://www.bwct.de
ticso at bwct.de                                  info at bwct.de