UTM 4.7.5 MacOS guest - memory exhausted

[janmalecek]

On my MacOS guest VM, there is almost whole memory marked as occupied and the system is barely responsible. Problem occured when I have tried to update the MacOS VM system from 26.x to 26.7 and restarted the VM (the update failed). The problem is independent on a memory amount assigned to a VM (occupied 15,5/16GB, 23,5/24GB etc) and persists after reboot.

Configuration

• UTM Version: 4.7.5
• macOS Version: host 27.0 (the problem was also present on 26.7), VM 26.6.2
• Mac Chip (Intel, M1, ...): M5 Pro

Anyone notices the same issue? Is there anything I can attach for better investigation? Let me know, please. Thanks in advance!

EDIT: Tried to reproduce, a new VM botstrapped with 26.6.2 seems to be okay, upgrade attempt to 26.7 leads to the issue described.

[smilberg]

Same issue here, with diagnostics suggesting the memory is withheld before the
guest kernel gets it (not a guest-side leak), and a reliable reproduction on a
fresh VM.

To reproduce on a 26.7 M4 24GB host (this happened to me on my old VM, and a
freshly created new one):

• Create new VM (was 26.6.2). Activity monitor shows ~2/8GB used, no pressure.
• Attempt to upgrade to 26.7 on the guest, give up and shut down after ~45 minutes.
• Restart again under 26.6.2:
  • On host activity monitor shows UTM service using 2.2GB of RAM.
  • On guest activity monitor shows 7.5/8GB used.

The VM trashes the swap like crazy with ~6GB of memory used but completely unaccounted for.

Configuration

• UTM version: 4.7.5 (118)
• Host macOS: 26.7
• Guest macOS: 26.6.2, build 25G83 (update target: 25G229)
• Chip: M4 24GB
• VM RAM: 8 GB (Apple Virtualization backend)

Guest diagnostics (original VM)

sysctl hw.memsize hw.memsize_usable:

hw.memsize: 8589934592
hw.memsize_usable: 1879031808

Activity Monitor (nothing open): Memory Used 7.54 GB, but App 286.5 MB + Wired 815.9 MB + Compressed 221.0 MB ≈ 1.3 GB. Swap used: 408 MB so over 6GB just lost somewhere.

vm_stat (16 KB pages): free 2147, active 22862, inactive 21337, speculative 1005, wired 53141, occupied by compressor 14041. Total ≈ 114.5k of 524,288 pages (~1.75 GB). The remaining ~410k pages (~6.25 GB) are in no VM queue at all.

zprint -s: no unusually large kernel zones (largest ~22 MB), so not a kernel zone leak.

ioreg: AppleVirtIOBalloon = 0 instances, so no balloon device is attached. There is no balloon option in the UTM settings for this VM.

NVRAM after the failed update (relevant keys)

ota-original-os-version    25G83
ota-os-version             25G229
ota-result                 success
ota-outcome                success
ota-updateType             incremental
ota-snapshot-update        true
IASCurrentInstallPhase     Boot 1
boot-volume                ...:51D161C5-E372-4122-B4CF-629ADAFC99EF
upgrade-boot-volume        ...:5F9234CF-C7F4-496F-B5F4-646CA2537157
boot-args                  (empty)

The update was staged successfully, but the install is stuck at phase "Boot 1", and a separate upgrade boot volume is still pending. Full nvram -p output available on request.

Attempted workarounds

Failed: Deleted the pending upgrade with sudo nvram -d upgrade-boot-volume, then did a full shutdown and cold start from UTM. hw.memsize_usable was unchanged, so clearing that key alone does not restore the memory.

Works: Create a new VM with 26.6.2 and do not try to upgrade. Works with a 26.7 host.

[mrtnd]

After updating to macOS 27, I'm experiencing a similar issue: the VM becomes completely unresponsive after some time, but only when the screen is off. When I wake the VM, it remains frozen and requires a hard reboot. I can't verify whether memory is exhausted within the guest in this scenario. This occurs on an M3 Pro; the host is still running macOS 26.7.