20.0-20.2 Only: VMware vSphere ESXi host might crash when rebooted while vGPU VMs are running#

Description#

If a VMware vSphere ESXi host is rebooted while one or more vGPU VMs are running, the host might crash with a purple diagnostic screen.

The diagnostic information reports a page fault in the devlayer world and identifies the nvidia-gpu module. The stack trace might contain CrashCat teardown functions such as crashcatEngineUnmapBufferDescriptor_IMPL and crashcatEngineUnmapCrashBuffer_IMPL.

Workaround#

Before rebooting the host, shut down all vGPU VMs.

Status#

Resolved in NVIDIA vGPU software 20.3

Ref. ##

6343496