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

...

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

# 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

...

	 *-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)

		...

巴特西

x86服务器MCE（Machine Check Exception）问题

MCE现象

MCE错误判断原则

MCE错误来源

MCE错误码解析

最新文章

热门文章