HVM stub-domain occasionally runs out of memory

[eug48]

Qubes OS version:

R4.0

Affected component(s):

HVM stub domains

---

Steps to reproduce the behavior:

Run Windows 10 HVM, giving it 16GB of RAM (out of 32GB).
The VM occasionally fails when there's relatively high memory/cpu usage.

Expected behavior:

VM runs without interruption

Actual behavior:

The Windows 10 VM fails due to the stub-domain OOM killer killing qemu.
Logs here: https://gist.github.com/eug48/472ede7f275bd1ccf4fb9bf1f6157b8d

General notes:

---

Related issues:

[eug48]

It hope that increasing the stubdom's memory could mitigate this but I can't find how to do this even though these two patches suggest it is possible:

https://github.com/QubesOS/qubes-core-libvirt/blob/385f8f3f432fcdf6aacdbd8c11d62635cea10c2f/0020-libxl-add-support-for-stubdom_mem-option.patch
https://github.com/QubesOS/qubes-vmm-xen/blob/ab932b916883893f5c105b5338686c3856858fd7/patch-stubdom-linux-config-stubdom-mem.patch

Any help would be much appreciated.

[eug48]

I've figured it out: to set the stubdom's memory to ~500MB run qvm-prefs Win10 stubdom_mem 500. The setting can be verified through xl list.

This seems to have helped as I can now install the latest Windows 10 Feature Update.

[andrewdavidwong]

Reopening because it's still listed as known bug on #1861.