Re: kvm splat in mmu_spte_clear_track_bits

[Adam Borowski <kilobyte@angband.pl>]

On Mon, Aug 21, 2017 at 09:58:34PM +0200, Radim Krčmář wrote:
> 2017-08-21 21:12+0200, Adam Borowski:
> > On Mon, Aug 21, 2017 at 09:26:57AM +0800, Wanpeng Li wrote:
> > > 2017-08-21 7:13 GMT+08:00 Adam Borowski <kilobyte@angband.pl>:
> > > > I'm afraid I keep getting a quite reliable, but random, splat when running
> > > > KVM:
> > > 
> > > I reported something similar before. https://lkml.org/lkml/2017/6/29/64
> > 
> > Your problem seems to require OOM; I don't have any memory pressure at all:
> > running a single 2GB guest while there's nothing big on the host (bloatfox,
> > xfce, xorg, terminals + some minor junk); 8GB + (untouched) swap.  There's
> > no memory pressure inside the guest either -- none was Linux (I wanted to
> > test something on hurd, kfreebsd) and I doubt they even got to use all of
> > their frames.
> 
> I even tried hurd, but couldn't reproduce ...

Also happens with a win10 guest, and with multiple Linuxes.

> what is your qemu command
> line and the output of host's `grep . /sys/module/kvm*/parameters/*`?

qemu-system-x86_64 -enable-kvm -m 2048 -vga qxl -usbdevice tablet \
 -net bridge -net nic \
 -drive file="$DISK",cache=writeback,index=0,media=disk,discard=on

qemu-system-x86_64 -enable-kvm -m 2048 -vga qxl -usbdevice tablet \
 -net bridge -net nic \
 -drive file="$DISK",cache=unsafe,index=0,media=disk,discard=on,if=virtio,format=raw

/sys/module/kvm/parameters/halt_poll_ns:200000
/sys/module/kvm/parameters/halt_poll_ns_grow:2
/sys/module/kvm/parameters/halt_poll_ns_shrink:0
/sys/module/kvm/parameters/ignore_msrs:N
/sys/module/kvm/parameters/kvmclock_periodic_sync:Y
/sys/module/kvm/parameters/lapic_timer_advance_ns:0
/sys/module/kvm/parameters/min_timer_period_us:500
/sys/module/kvm/parameters/tsc_tolerance_ppm:250
/sys/module/kvm/parameters/vector_hashing:Y
/sys/module/kvm_amd/parameters/avic:0
/sys/module/kvm_amd/parameters/nested:1
/sys/module/kvm_amd/parameters/npt:1
/sys/module/kvm_amd/parameters/vls:0

> > Also, it doesn't reproduce for me on 4.12.
> 
> Great info ... the most suspicious between v4.12 and v4.13-rc5 is the
> series with dcdca5fed5f6 ("x86: kvm: mmu: make spte mmio mask more
> explicit"), does reverting it help?
> 
> `git revert ce00053b1cfca312c22e2a6465451f1862561eab~1..995f00a619584e65e53eff372d9b73b121a7bad5`

Alas, doesn't seem to help.

I've first installed a Debian stretch guest, the host survived both the
installation and subsequent fooling around.  But then I started a win10
guest which splatted as soon as the initial screen.


Meow!
-- 
⢀⣴⠾⠻⢶⣦⠀ 
⣾⠁⢰⠒⠀⣿⡁ Vat kind uf sufficiently advanced technology iz dis!?
⢿⡄⠘⠷⠚⠋⠀                                        -- Genghis Ht'rok'din
⠈⠳⣄⠀⠀⠀⠀