VirtualBox

Opened 9 years ago

Last modified 9 years ago

#13599 new defect

Windows 7 guest renders Linux host unresponsive; uncompleted fence warnings

Reported by: Philip Boulain (SW) Owned by:
Component: other Version: VirtualBox 4.3.18
Keywords: Cc:
Guest type: other Host type: other

Description

Host: Ubuntu 12.04.5 LTS, running Virtualbox 4.3.18-96516~Ubuntu~precise from your repo.

Guest: Windows 7 64-bit, 3D accelleration on, WDDM driver, using Aero. Guest additions are up to date.

After some time the host system will almost completely stop responding to UI input---screen repaints and mouse moves by tiny amounts every few seconds or so (think: forkbombs or paging hell). SSHing in and the text console remain responsive. The host has not run out of memory, the only notable CPU usage is the VirtualBox process for that guest maxing a single core, there is no substantial disk activity. Trying to close the VM via the UI didn't managed to finish painting the power-off dialogue in a timely fashion. The VM does not appear to be able to complete processing VBoxManage commands either although the log does show they are getting through.

The log contains:

75:18:45.743604 Guest Log: VBoxMP::vboxCmdVbvaCheckCompletedIrqCb: WARNING! :no vbva
75:18:45.743733 Guest Log: VBoxMP::VBoxVBVAExAllocContiguous: WARNING! :Host did not clean up the buffer!
75:18:45.743863 Guest Log: VBoxMP::VBoxCmdVbvaSubmitLock: WARNING! :failed to allocate contiguous buffer, trying nopping the tail
92:07:13.526281 Guest Log: VBoxMP::DxgkDdiQueryCurrentFenceNew: WARNING! :=>DxgkDdiQueryCurrentFenceNew
92:07:13.526395 Guest Log: VBoxMP::DxgkDdiQueryCurrentFenceNew: WARNING! :uncompleted fences, u32FenceSubmitted(2961375), u32FenceCompleted(2961374) u32FenceProcessed(2961374)

The last line repeats. (Will attach full logs.)

Host binary nVidia drivers recently upgraded to 304.123-0ubuntu0.0.1, but have been running them all week; problem has only occured overnight Thursday, and repeated again Friday morning. Disabling 3D accelleration seems to have stopped the problem for now but this may be a red herring. Guest has not had any other changes recently, although has experienced graphical flickering since VirtualBox (I think) 4.3.14. Otherwise has been working for months.

A Kubuntu 14.04.1LTS guest with 3D accelleration running at the same time (all week) has not been affected. No occurences of "WARNING" in its current log.

Ticket #13510 is about the same log error message but with different symptoms.

Attachments (4)

VBox-overnight-lockup.log.bz2 (66.1 KB ) - added by Philip Boulain (SW) 9 years ago.
Overnight failure. Last few lines are attempting to stop VM gracefully; had to kill -9 in the end. (Bzipped for size limit.)
VBox-repeated-lockup.log (217.6 KB ) - added by Philip Boulain (SW) 9 years ago.
Second occurence; didn't attempt VBoxManage this time. Log doesn't have some of the same warnings before the repeated line.
VBox.log.1 (109.2 KB ) - added by amazombie 9 years ago.
Sorry, wrong log file. This is the one that contains the last entries prior to crashing.
VBox.log (102.7 KB ) - added by amazombie 9 years ago.
Replace

Download all attachments as: .zip

Change History (5)

by Philip Boulain (SW), 9 years ago

Overnight failure. Last few lines are attempting to stop VM gracefully; had to kill -9 in the end. (Bzipped for size limit.)

by Philip Boulain (SW), 9 years ago

Attachment: VBox-repeated-lockup.log added

Second occurence; didn't attempt VBoxManage this time. Log doesn't have some of the same warnings before the repeated line.

comment:1 by Philip Boulain (SW), 9 years ago

Apologies, this is looking like a host system error. Host syslog has many instances of:

Nov  7 09:45:22 philb-desktop kernel: [344925.092919] NVRM: Xid (PCI:0000:01:00): 8, Channel 00000001

And occasionally:

Nov  7 12:50:51 philb-desktop kernel: [356022.402432] NVRM: os_schedule: Attempted to yield the CPU while in atomic or interrupt context

Also managed to reproduce on the Kubuntu guest; window compositing alone isn't doing it, but glxgears did. Getting *recovering* sub-second freezes in a host OpenGL application (Minetest). Don't know why it's hitting VirtualBox so much harder, but something's wrong with the host.

by amazombie, 9 years ago

Attachment: VBox.log.1 added

Sorry, wrong log file. This is the one that contains the last entries prior to crashing.

by amazombie, 9 years ago

Attachment: VBox.log added

Replace

Note: See TracTickets for help on using tickets.

© 2023 Oracle
ContactPrivacy policyTerms of Use