QEMU crashes with assertion when running macOS guest with Intel Hypervisor

[conath]

When attempting to run a macOS Monterey Beta VM in UTM on an Intel Mac, QEMU crashes with the message: Assertion failed: (vector < nr_vectors), function msi_notify, file ../hw/pci/msi.c, line 317.

After searching online I can not find anyone with this error before. I have tried to change my configuration to use only one core, at no avail (maybe vectors has to do with the number of cores).

Configuration

• UTM Version: 2.1.2 (1) Compiled myself from 77d6ca1
• OS Version: macOS 11.3.1 (20E241)
• Intel or Apple Silicon? Intel

Debug log debug.log

VM Config config.plist.zip

More info:
As you can see, the VM uses custom EFI BIOS and OpenCore to get macOS to boot. Here's a screenshot of the verbose macOS boot right before the crash (only QEMU crashes, not UTM):

[conath]

Update: I have discovered a workaround for this issue. By changing the configuration to disable the HD Audio and USB Redirect features, the error goes away.

[thenickdude]

The issue seems to be the USB 3 controller (added by using USB passthrough of a USB 3 device) and its MSI setting.

I added this to my QEMU command line to disable MSI and the assert went away, and my USB 3 card reader functions correctly:

-global nec-usb-xhci.msi=off

[conath]

@thenickdude Thanks! Do you know if it is safe to add this MSI setting as a default in UTM?

[thenickdude]

I think for non-macOS guests this will cause a performance reduction for USB 3 devices (as non-MSI interrupts have higher overhead)

For macOS guests this should be perfect

[osy]

I think this is a bug in QEMU and should be fixed in QEMU rather than hacking in a config to disable MSI. Has this been reported to QEMU?

In the meantime, -global nec-usb-xhci.msi=off can be added to custom arguments as a workaround right?

[conath]

It could, just with a performance hit to USB passthrough, according to @thenickdude. What do you think about a "quirks" tab instead of adding this as a default? https://github.com/utmapp/UTM/pull/3031#discussion_r696593486

[thenickdude]

Often virtual machine software rolls quirks like this into a set that is implicitly activated by telling it what guest OS you're running (e.g. Workstation, VirtualBox and Proxmox all do this).

[osy]

@conath So for a while we’ve used qemu-xhci which has MSI=0 by default. However I think that caused issues with Windows (#3194) so I changed it back to nec-usb-xhci. However I’m wondering if maybe QEMU 6.2.0 has fixed this issue so can you re-test after 8fb5b5d2fa7c04f137eff7a4ad42d879772b7cc3?

[lwabish]

-global nec-usb-xhci.msi=off

I think this is a bug in QEMU and should be fixed in QEMU rather than hacking in a config to disable MSI. Has this been reported to QEMU?

In the meantime, -global nec-usb-xhci.msi=off can be added to custom arguments as a workaround right?

Thanks.This tweak works in pve 8.1.3 with hacintosh sonoma