Re: [PATCH v2 00/18] Entry stack switching

[Ingo Molnar <mingo@kernel.org>]

* Andy Lutomirski <luto@kernel.org> wrote:

> On Tue, Nov 21, 2017 at 10:22 PM, Ingo Molnar <mingo@kernel.org> wrote:
> >
> > * Andy Lutomirski <luto@kernel.org> wrote:
> >
> >> This sets up stack switching, including for SYSCALL.  I think it's
> >> in decent shape.
> >>
> >> Known issues:
> >>  - I think we're going to want a way to turn the stack switching on and
> >>    off either at boot time or at runtime.  It should be fairly straightforward
> >>    to make it work.
> >>
> >>  - I think the ORC unwinder isn't so good at dealing with stack overflows.
> >>    It bails too early (I think), resulting in lots of ? entries.  This
> >>    isn't a regression with this series -- it's just something that could
> >>    be improved.
> >>
> >> Ingo, patch 1 may be tip/urgent material.  It fixes what I think is
> >> a bug in Xen.  I'm having a hard time testing because it's being
> >> masked by a bigger unrelated bug that's keeping Xen from booting
> >> when configured to hit the bug I'm fixing.  (The latter bug goes at
> >> least back to v4.13, I think I know roughtly what's wrong, and I've
> >> reported it to the maintainers.)
> >
> > Hm, with this series the previous IRQ vector bug appears again:
> >
> > [   51.156370] do_IRQ: 16.34 No irq handler for vector
> > [   57.511030] do_IRQ: 16.34 No irq handler for vector
> > [   57.528335] do_IRQ: 16.34 No irq handler for vector
> > [   57.533256] do_IRQ: 16.34 No irq handler for vector
> > [   63.991913] do_IRQ: 16.34 No irq handler for vector
> > [   63.996810] do_IRQ: 16.34 No irq handler for vector
> >
> > I've attached the reproducer config. Note that the system appears to be working to
> > a certain extent (I could ssh to it and extract its config), but produces these
> > warnings sporadically.
> 
> I'll try to reproduce this, but this is weird.  This is vector 34,
> which is, or could be, a genuine IRQ vector.  The only way I can think
> of that my series would have caused this is if I very severely broke
> common_interrupt, but I don't see how that could have happened without
> breaking everything.  It's also weird that you're seeing this only on
> CPU 16.  Maybe it's worth adding a WARN_ON to that warning to get a
> stack trace just in case.
> 
> Thomas, any insight here?
> 
> > but don't get the IRQ vector warnings.
> 
> Ingo, are you saying that you only get the IRQ vector warnings with
> the SYSCALL hwframe fix applied?  That's bizarre.

Correct. I assume it's because lockdep is working fine with that fix applied,
but that also means that different irq-tracing code paths are taken.

The lockdep error disables lockdep globally and immediately.

> Anyway, I booted your config (more or less -- I munged it through
> virtme-configkernel --update first) with 17 vCPUs and it seems fine.
> Is the issue reliable enough to bisect?

Ok, it should be bisectable, will try to bisect it.

I think it's a key aspect that the CPU is AMD - a similar config on Intel seems to 
be working fine (modulo the unwinder warning).

Thanks,

	Ingo