[Qemu-devel] Runtime-modified DIMMs and live migration issue

[Andrey Korolyov]

Hello Igor,

the current hotplug code for dimms effectively prohibiting a
successful migration for VM if memory was added after startup:

- start a VM with certain amount of empty memory slots,
- add some dimms and online them in guest (I am transitioning from 2
to 16G with 512Mb DIMMs),
- migrate a VM and observe guest null pointer dereference (or BSOD
with reboot, for Windows).

Issue is currently touching all stable versions and assumingly master,
as there are no related fixes/RFCs since 2.3 I`m currently using for
testing. The issue is related to an incorrect population of the
regions during runtime hotplugging, hopefully 2.4 will get the fix.

You may run some workload in guest to achieve one hundred percent
certainty of hitting the issue, for example, fio against
http://xdel.ru/downloads/fio.txt . QEMU args are simular to '... -m
512,slots=31,maxmem=16384M -object
memory-backend-ram,id=mem0,size=512M -device
pc-dimm,id=dimm0,node=0,memdev=mem0 -object
memory-backend-ram,id=mem1,size=512M -device
pc-dimm,id=dimm1,node=0,memdev=mem1 -object
memory-backend-ram,id=mem2,size=512M -device
pc-dimm,id=dimm2,node=0,memdev=mem2...'

Thanks for looking into this!

fio-migrate-hotplug.trace.txt
Description: Text document