Re: BUG: soft lockup in __kmalloc_node() with KFENCE enabled

[Marco Elver]

On Mon, 11 Oct 2021 at 16:42, Andrea Righi <andrea.righi@xxxxxxxxxxxxx> wrote:
> On Mon, Oct 11, 2021 at 12:03:52PM +0200, Marco Elver wrote:
> > On Mon, 11 Oct 2021 at 11:53, Andrea Righi <andrea.righi@xxxxxxxxxxxxx> wrote:
> > > On Mon, Oct 11, 2021 at 11:23:32AM +0200, Andrea Righi wrote:
> > > ...
> > > > > You seem to use the default 20s stall timeout. FWIW syzbot uses 160
> > > > > secs timeout for TCG emulation to avoid false positive warnings:
> > > > > https://github.com/google/syzkaller/blob/838e7e2cd9228583ca33c49a39aea4d863d3e36d/dashboard/config/linux/upstream-arm64-kasan.config#L509
> > > > > There are a number of other timeouts raised as well, some as high as
> > > > > 420 seconds.
> > > >
> > > > I see, I'll try with these settings and see if I can still hit the soft
> > > > lockup messages.
> > >
> > > Still getting soft lockup messages even with the new timeout settings:
> > >
> > > [  462.663766] watchdog: BUG: soft lockup - CPU#2 stuck for 430s! [systemd-udevd:168]
> > > [  462.755758] watchdog: BUG: soft lockup - CPU#3 stuck for 430s! [systemd-udevd:171]
> > > [  924.663765] watchdog: BUG: soft lockup - CPU#2 stuck for 861s! [systemd-udevd:168]
> > > [  924.755767] watchdog: BUG: soft lockup - CPU#3 stuck for 861s! [systemd-udevd:171]
> >
> > The lockups are expected if you're hitting the TCG bug I linked. Try
> > to pass '-enable-kvm' to the inner qemu instance (my bad if you
> > already have), assuming that's somehow easy to do.
>
> If I add '-enable-kvm' I can triggering other random panics (almost
> immediately), like this one for example:

Just FYI: https://lkml.kernel.org/r/20211019102524.2807208-2-elver@xxxxxxxxxx

But you can already flip that switch in your config
(CONFIG_KFENCE_STATIC_KEYS=n), which we recommend as a default now.

As a side-effect it'd also make your QEMU TCG tests pass.