ESXI PSOD #PF Exception 14 occurs on Intel Sapphire Rapids and Emerald Rapids Generation CPU

Products

• VMware vSphere ESXi

Issue/Introduction

• PSOD(Purple Screen Of Death), backtrace with keywords such as  WorldArchAddAMXSaveStateInt and XMap_Unmap or XMap_MapWithRA64 as below, occurs repeatedly.

#PF Exception 14 in world 58776186:vmm5:IP###### addr ######
cpu22:58776186)Code start: 0x420027600000 VMK uptime: 88:13:57:51.038
cpu22:58776186)0x453989e1bb60:[0x42002777bb94]World_Switch@vmkernel#nover+0x1abc stack: 0x46b20e32532fdc
cpu22:58776186)0x453989e1bbb0:[0x420027a2954e]CpuSchedDispatch@vmkernel#nover+0xa53 stack: 0x453900000001
cpu22:58776186)0x453989e1bdf0:[0x4200276d76fb]BH_Check@vmkernel#nover+0xb8 stack: 0x43142960f838
cpu22:58776186)0x453989e1be60:[0x420027754854]XMap_MapWithRA64@vmkernel#nover+0xf9 stack: 0x2  -----------!!! 
cpu22:58776186)0x453989e1bea0:[0x42002777c59d]WorldArchAddAMXSaveStateInt@vmkernel#nover+0x15a stack: 0x19d96206  -----------!!! 
cpu22:58776186)0x453989e1bf20:[0x420027776422]VMMVMKCall_Call@vmkernel#nover+0x10b stack: 0x0
cpu22:58776186)0x453989e1bfe0:[0x420027773379]VMKVMM_ArchEnterVMKernel@vmkernel#nover+0xe stack: 0x42002777336c
cpu22:58776186)0xfffffffffc407d88:[0xfffffffffc067ae2] (vmm64)

Environment

VMware vSphere ESXi 8.0U1, U2

Cause

Memory corruption occurs during the execution of functions that support Advanced Matrix Extensions (AMX).

### The issue has been identified in Sapphire Rapids and Emerald Rapids Generation CPUs, but it is expected to potentially occur in all generations that support the AMX feature. 

Resolution

The issue has been fixed on vSphere ESXi 8.0U3.


Workaround:

• Disable AMX Feature in BIOS settings. 
• Enable Ice Lake or below generation EVC mode.