ESX 9.x fails to boot with a PSOD when Sub-NUMA Clustering (SNC) is enabled

Products

• VMware vSphere ESXi

Issue/Introduction

• After installing ESX 9.x or upgrading to ESX 9.x on a system with Sub-NUMA Clustering (SNC) enabled, the host crashes with a Purple Screen of Death (PSOD) during early system boot.
  
  
• The backtrace on the PSOD display includes calls similar to the following:

#PF Exception 14 in world #######:bootstrap IP 0x############ addr 0x############

CpuSchedIdleBalanceOnReady
CpuSchedQueueAddInt
CpuSchedQueueAdd
CpuSchedVcpuMakeReady
CpuSched_Add
Sched_Add
World_NewSysWorldSched
CpuSched_LateInit
Sched_LateInit
SysInitTable_Start
InitVMKernel
LegacyBootVMKernel
BootVMKernel

Environment

VMware ESX 9.x

Cause

This issue occurs on certain host platforms when Sub-NUMA Clustering (SNC) is enabled and the CPU topology data provided by the ACPI BIOS is not presented in the sequence expected by ESX 9.x.

When ESX 9.x processes this data during early boot initialization, a memory mapping error occurs within the CPU scheduler's topology handling logic, resulting in an unmapped memory access and a subsequent Page Fault (#PF) crash.

Resolution

This is a known issue affecting ESX 9.x. Broadcom Engineering is currently working on a resolution. This KB article will be updated once a fix is available.

Workaround:

Disable Sub-NUMA Clustering (SNC)

1. 1. Access the system BIOS / UEFI setup menu.
   2. Disable Sub-NUMA Clustering (SNC) (or set the NUMA clustering configuration to standard/disabled).
      Note: Please refer to your hardware vendor's BIOS/UEFI manual for detailed instructions on navigating and modifying BIOS settings.
   3. Reboot the server to allow ESX 9.x to complete the boot process.

 

Additional Information

Sub-NUMA Clustering (SNC) が有効になっていると、ESX 9.x が PSOD で起動に失敗する