x86服务器MCE（Machine Check Exception）问题

MCE现象

Intel在Pentium 4、Xenon和P6系列处理器中实现了机器检查（Machinecheck）架构，提供能够检测和报告硬件（机器）的错误机制，如系统总线错误、ECC错误、奇偶校验错误、缓存错误、TLB错误等。它包括一直MSR（Model-Specific Registers）寄存器，用来设置机器检查和额外的bank MSR记录错误。

当机器检查到不可纠正的machine-check错误时，就触发一个machine-check异常。machine-check架构不允许在出现MCE后处理器重启，但MCE处理程序可以从MSR寄存器收集相关信息。

CPU 7: Machine Check Exception: 5 Bank 0: b200004010000400

RIP !INEXACT! 10:<ffffffff8010f16e> {mwait_idle+0x5e/0x90}

TSC 1952dbeebcc8

Kernel panic: Machine check

Reconfiguring memory bank information….

This may take a while….

done waiting: 3 cpus not responding

Warning: Non-empty request queue

I/O requests in flight at dump time

CPU 7: Machine Check Exception: 4 Bank 0: f200004040000400

RIP !INEXACT! 10:<ffffffff8011ef69>

 
 

MCE错误判断原则

凡是内核死机打印“Machine Check Exception“或内核栈信息中打印有do_machine_check()函数，均为MCE问题。

 
 

MCE错误来源

• PCI-E设备信号质量/时钟
• CPU芯片损坏/设计BUG

  CPU Cache损坏或其它故障

• CPU可能的缺陷

  如CPU生产制造过程中带来的缺陷

• 内存坏/接触不良
• BIOS配置不当
• OS/MCE中断程序Bug
• 环境因素，如温度/湿度

 
 

MCE错误码解析

以上面MCE错误为例，Machine Check Exception和Bank 0(5)的值分别对应IA32_MCG_STATUS MSR、IA32_MCi_STATUS寄存器。

则对应的寄存器值为：

IA32_MCG_STATUS MSR寄存器的值为0000000000000004

IA32_MC0_STATUS MSR的值为f200000410000800

IA32_MC5_STATUS MSR的值为f200001044100e0f

 
 

根据MSR的值，对照Intel编程手册和Intel其他资料，就可以比较容易找出MCE原因。

 

 

dmesg显示

1 2 3 4 5 6 7 8

... sbridge: HANDLING MCE MEMORY ERROR CPU 0: Machine Check Exception: 0 Bank 5: 8c00004000010093 TSC 0 ADDR 67081b300 MISC 2140040486 PROCESSOR 0:206d7 TIME 1441181676 SOCKET 0 APIC 0 EDAC MC0: CE row 2, channel 0, label "CPU_SrcID#0_Channel#3_DIMM#0": 1 Unknown error(s): memory read on FATAL area : cpu=0 Err=0001:0093 (ch=3), addr= 0x67081b300 => socket=0, Channel=3(mask=8), rank=0 ...

保存4行log为mlog

1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21

# mcelog --ascii < /tmp/mlog WARNING: with --dmi mcelog --ascii must run on the same machine with the same BIOS/memory configuration as where the machine check occurred. sbridge: HANDLING MCE MEMORY ERROR CPU 0: Machine Check Exception: 0 Bank 5: 8c00004000010093 HARDWARE ERROR. This is *NOT* a software problem! Please contact your hardware vendor Wed Sep 2 16:14:36 2015 CPU 0 BANK 5 MISC 2140040486 ADDR 67081b300 STATUS 8c00004000010093 MCGSTATUS 0 CPUID Vendor Intel Family 6 Model 45 WARNING: SMBIOS data is often unreliable. Take with a grain of salt! <24> DIMM 1333 Mhz Res13 Width 72 Data Width 64 Size 16 GB Device Locator: Node0_Channel2_Dimm0 Bank Locator: Node0_Bank0 Manufacturer: Hynix Semiconducto Serial Number: 40743B5A Asset Tag: Dimm2_AssetTag Part Number: HMT42GR7BFR4A-PB TSC 0 ADDR 67081b300 MISC 2140040486 PROCESSOR 0:206d7 TIME 1441181676 SOCKET 0 APIC 0 EDAC MC0: CE row 2, channel 0, label "CPU_SrcID#0_Channel#3_DIMM#0": 1 Unknown error(s): memory read on FATAL area : cpu=0 Err=0001:0093 (ch=3), addr = 0x67081b300 => socket=0, Channel=3(mask=8), rank=0

根据
Part Number: HMT42GR7BFR4A-PB
Serial Number: 40743B5A

在lshw中找相应硬件

1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36

... *-memory:0 description: System Memory physical id: 2d slot: System board or motherboard *-bank:0 description: DIMM 1333 MHz (0.8 ns) product: HMT42GR7BFR4A-PB vendor: Hynix Semiconducto physical id: 0 serial: 905D21AE slot: Node0_Channel1_Dimm0 size: 16GiB width: 64 bits clock: 1333MHz (0.8ns) *-bank:1 description: DIMM Synchronous [empty] product: A1_Dimm1_PartNumber vendor: Dimm1_Manufacturer physical id: 1 serial: Dimm1_SerNum slot: Node0_Channel1_Dimm1 width: 64 bits *-bank:2 description: DIMM 1333 MHz (0.8 ns) product: HMT42GR7BFR4A-PB vendor: Hynix Semiconducto physical id: 2 serial: 40743B5A slot: Node0_Channel2_Dimm0 size: 16GiB width: 64 bits clock: 1333MHz (0.8ns) ...