port-amd64/57092: nvmm_init_vcpu failed: Cannot allocate memory

[gson%gson.org@localhost (Andreas Gustafsson)]

>Number:         57092
>Category:       port-amd64
>Synopsis:       nvmm_init_vcpu failed: Cannot allocate memory
>Confidential:   no
>Severity:       serious
>Priority:       medium
>Responsible:    port-amd64-maintainer
>State:          open
>Class:          sw-bug
>Submitter-Id:   net
>Arrival-Date:   Tue Nov 29 10:10:00 +0000 2022
>Originator:     Andreas Gustafsson
>Release:        NetBSD 9.0_STABLE
>Organization:
>Environment:
System: NetBSD babylon5.netbsd.org 9.0_STABLE NetBSD 9.0_STABLE (BABYLON5-EPYC) #0: Wed Sep  9 05:05:23 UTC 2020
Architecture: x86_64
Machine: amd64
>Description:

On the testbed on babylon5.netbsd.org, a small percentage of the test
runs fail when attempting to start a qemu/NVMM virtual machine, with
error messages like these:

  NetBSD Virtual Machine Monitor accelerator is operational
  qemu-system-i386: NVMM: Failed to create a virtual processor, error=12
  nvmm_init_vcpu failed: Cannot allocate memory

The full log output from the most recent such failure is here:

  http://releng.netbsd.org/b5reports/i386/2022/2022.11.28.19.32.45/install.log

The host in case has plenty of memory:

  babylon5.netbsd.org$ sysctl hw.physmem64
  hw.physmem64 = 137300484096

but it's hard to tell after the fact exactly how much of it was in use
at the time of the failure.

Perhaps this is a case of kernel memory fragmentation similar to the
one affecting umass(4) in PR 54977.

>How-To-Repeat:

>Fix: