Raudi
Expert
Expert

PSOD - "Failed to ack TLB invalidate" or host freezes when power on VM with PCI passthrough

Jump to solution

Hello,

since a few weeks i have the problem that my ESXi host freezes sometimes when powering on a Windows 10 VM with a NVidia Quadro 2200 in PCI passthrough and today i got a PSOD.

psod-04.04.2016.JPG

I think it happens first after installing the patch released at 02/23/2016.

Is it possible that a update is responsible for my problem? I'm thinking about a new clean install with 6.0U1b to test if a update is causing my problem.

In the vmkernel.log this line must be the next line when the VM is starting:

PCI: 717: Device 0000:84:00.0: restoring TV/VC Map (0x1) to offset 114

The line before this line is the last line in the vmkernel.log:

Net: 2441: connected wks10 eth0 to VM Network, portID 0x200000b

When the VM is running i have no problems...

Kind regards

Stefan

0 Kudos
1 Solution

Accepted Solutions
Raudi
Expert
Expert

O.k. i replaced my server hardware and reused all expansion cards (NVidia Quadro, QLogic FC card and Teradici Apex).

The problem is gone, so something on the motherboard or cpu's was broken...

View solution in original post

0 Kudos
2 Replies
Raudi
Expert
Expert

Today i made some tests and made a clean install with vSphere 6.0 U2 and also 6.0 U1b everytime i got the problem with the VM.

And i noticed that the problem is when turning off the VM off too. Sometimes i can power on the VM but then i got a problem when shutdown the VM.

So it must be a hardware issue of one of the CPU's or the motherboard. I don't think that this is a problem of the NVidia card, because i had the same problem with a different VM with a different IO card too.

And it happens with cards in slots from CPU 1 and CPU 2.

How does the PCI passthrough work? Is there a issue with the motherboard or with one of the CPU's?

Kind regards

Stefan

0 Kudos
Raudi
Expert
Expert

O.k. i replaced my server hardware and reused all expansion cards (NVidia Quadro, QLogic FC card and Teradici Apex).

The problem is gone, so something on the motherboard or cpu's was broken...

View solution in original post

0 Kudos