Bug#1014394: linux kernel 5.10.0-15 on virtualbox host causes random process crashes in guests

[Michael <ml@hemathor.de>]

package: linux-image-5.10.0-15-arm64
version: 5.10.120-1

i am running virtualbox 6.1.34 from the virtualbox.org repo on a debian 
11.3 host. the guest also runs debian 11.3.

when both host and guest run the latest stable kernel 5.10.0-15 
(5.10.120-1) i get random process crashes in the guest when having 
significant i/o for longer than a few seconds on the host.

i.e., when i do a

 # mv <large file> <destination>

or even a

 # md5sum <large file>

or anything that causes significant i/o traffic on the host, the guest 
first starts with complaining:

Jul 01 20:44:49 vmguest kernel: hrtimer: interrupt took 21312191 ns
Jul 01 20:45:20 vmguest kernel: clocksource: timekeeping watchdog on CPU0: 
kvm-clock wd-wd read-back delay of 816732ns
Jul 01 20:45:20 vmguest kernel: clocksource: wd-tsc-wd read-back delay of 
2540783ns, clock-skew test skipped!

and eventually guest processes start to randomly crash, e.g.:

Jun 30 20:17:01 vmguest kernel: traps: sh[250946] general protection fault 
ip:7fb0c341708e sp:7ffec3154378 error:0 in 
libc-2.31.so[7fb0c33a6000+14b000]
Jul 01 00:00:02 vmguest kernel: traps: hostname[253617] general protection 
fault ip:7f905f2b24a6 sp:7fff44a30e30 error:0 in 
libc-2.31.so[7f905f299000+14b000]
Jul 01 00:53:01 vmguest kernel: traps: wget[254290] general protection 
fault ip:7f934bc23fda sp:7ffd716954d0 error:0 in 
libtasn1.so.6.6.0[7f934bc1a000+c000]

if i switch to kernel 5.10.0-14 (5.10.113-1) on the host (the guest kernel 
remains 5.10.0-15), then the random process crashes in the guest disappear, 
although the complaints from hrtimer and clocksource still remain, but 
significantly less often.

i started a thread in the 'debian-user' mailing list:
https://lists.debian.org/debian-user/2022/07/msg00043.html

thank you for looking into this!

greetings...