Intel MCA-CMCI初探

RAS（三）Intel MCA-CMCI

Corrected machine-check error interrupt (CMCI)是MCA的增强特性，它提供了一种threshold-based的错误上报方式。这种模式下，软件可以配置硬件corrected MC errors的阈值，硬件发生CE（Corrected Error）次数达到阈值后，会产生一个中断通知到软件处理。

值得一提的是，CMCI是随MCA加入的特性，最开始只能通过软件轮询方式获取CE信息。CMCI中断通知方式的优点是每个CE都会经过IRQ Handle处理，不会丢失任一CE；而轮询方式可能因为轮询频率低、存储空间有限等原因，导致丢失CE。但是并不是说CMCI最优，CMCI的缺点是大量CE会产生中断风暴，影响机器的性能。不幸的是在云服务器场景，CE风暴是比较常见的，那么当下Intel服务器是如何解决这个问题的呢？下面会讲到。

CMCI机制

CMCI默认是关闭的，软件需要通过配置IA32_MCG_CAP[10] = 1打开。

软件通过IA32_MCi_CTL2 MSR来控制对应Bank使能/关闭CMCI功能。

通过IA32_MCi_CTL2 Bit 14:0设置阈值，如果设置非0，则使用配置的阈值；如果CMCI不支持，则全0；

CMCI机制如下图 

硬件通过比较IA32_MCi_CTL2 Bit 14:0和IA32_MCi_STATUS Bit 52:38，如果数值相等，那么overflow event发送到APIC的CMCI LVT entry。如果MC error涉及多个processors，那么CMCI中断会同时发送到这些processors，比如2个cpu共享的cache发生CE，那么这两个cpu都会收到CMCI。

CMCI初始化

以Linux v6.3分支为例，内核使能CMCI代码

C++ arch/x86/kernel/cpu/mce/intel.c void intel_init_cmci(void) {         int banks;         if (!cmci_supported(&banks))                 return;         mce_threshold_vector = intel_threshold_interrupt;         cmci_discover(banks);         /*          * For CPU #0 this runs with still disabled APIC, but that's          * ok because only the vector is set up. We still do another          * check for the banks later for CPU #0 just to make sure          * to not miss any events.          */         apic_write(APIC_LVTCMCI, THRESHOLD_APIC_VECTOR|APIC_DM_FIXED);         cmci_recheck(); }

1.cmci_supported()函数主要事项包括

•根据内核启动参数"mce=no_cmci,ignore_ce"判断是否打开cmci和ce上报功能

•检查硬件是否支持cmci

•通过MCG_CMCI_P bit判断硬件是否使能cmci功能

2.mce_threshold_vector = intel_threshold_interrupt; 声明cmci的中断处理函数为intel_threshold_interrupt();

3.cmci_discover()函数主要完成

•遍历所有banks，通过配置IA32_MCi_CTL2寄存器使能所有bank的cmci功能；

C++                 rdmsrl(MSR_IA32_MCx_CTL2(i), val);                 ...                                  val |= MCI_CTL2_CMCI_EN;                 wrmsrl(MSR_IA32_MCx_CTL2(i), val);                 rdmsrl(MSR_IA32_MCx_CTL2(i), val);

•设置cmci threshold值，代码如下

C++ #define CMCI_THRESHOLD          1                 if (!mca_cfg.bios_cmci_threshold) {                         val &= ~MCI_CTL2_CMCI_THRESHOLD_MASK;                         val |= CMCI_THRESHOLD;                 } else if (!(val & MCI_CTL2_CMCI_THRESHOLD_MASK)) {                         /*                          * If bios_cmci_threshold boot option was specified                          * but the threshold is zero, we'll try to initialize                          * it to 1.                          */                         bios_zero_thresh = 1;                         val |= CMCI_THRESHOLD;                 }

如果用户未通过启动参数"mce=bios_cmci_threshold"配置值，则val = CMCI_THRESHOLD，为1；

如果启动参数"mce=bios_cmci_threshold"配置，那么表示bios已配置threshold值，即val & MCI_CTL2_CMCI_THRESHOLD_MASK不为0，跳过else if判断，采用bios配置值；如果bios未配置值，val & MCI_CTL2_CMCI_THRESHOLD_MASK为0，那么驱动初始化threshold为1。

4.cmci_recheck()

cmci_recheck函数通过调用machine_check_poll()，检查CPU #0是否有遗漏的CE&UCE events。

CMCI处理

cmci中断处理函数为intel_threshold_interrupt()，定义在arch/x86/kernel/cpu/mce/intel.c

C++ /*  * The interrupt handler. This is called on every event.  * Just call the poller directly to log any events.  * This could in theory increase the threshold under high load,  * but doesn't for now.  */ static void intel_threshold_interrupt(void) {         if (cmci_storm_detect())                 return;         machine_check_poll(MCP_TIMESTAMP, this_cpu_ptr(&mce_banks_owned)); }

1.cmci_storm_detect()函数主要是对cmci storm的处理，代码如下

C++ static bool cmci_storm_detect(void) {         unsigned int cnt = __this_cpu_read(cmci_storm_cnt);         unsigned long ts = __this_cpu_read(cmci_time_stamp);         unsigned long now = jiffies;         int r;         if (__this_cpu_read(cmci_storm_state) != CMCI_STORM_NONE)                 return true;         if (time_before_eq(now, ts + CMCI_STORM_INTERVAL)) {                 cnt++;         } else {                 cnt = 1;                 __this_cpu_write(cmci_time_stamp, now);         }         __this_cpu_write(cmci_storm_cnt, cnt);         if (cnt <= CMCI_STORM_THRESHOLD)                 return false;         cmci_toggle_interrupt_mode(false);         __this_cpu_write(cmci_storm_state, CMCI_STORM_ACTIVE);         r = atomic_add_return(1, &cmci_storm_on_cpus);         mce_timer_kick(CMCI_STORM_INTERVAL);         this_cpu_write(cmci_backoff_cnt, INITIAL_CHECK_INTERVAL);         if (r == 1)                 pr_notice("CMCI storm detected: switching to poll mode ");         return true; }

该函数通过jiffies，判断固定时间内发生的cmci次数是否大于CMCI_STORM_THRESHOLD（15），如果否则return，反之说明发生cmci storm，则执行cmci_toggle_interrupt_mode()关闭cmci功能， 切换为poll mode，通过轮询方式获取event；

2.非cmci storm情况下，通过machine_check_poll(MCP_TIMESTAMP, this_cpu_ptr(&mce_banks_owned))函数获取并记录故障信息

参数1定义如下，MCP_TIMESTAMP表示会记录当前TSC

C++ enum mcp_flags {         MCP_TIMESTAMP   = BIT(0),       /* log time stamp */         MCP_UC          = BIT(1),       /* log uncorrected errors */         MCP_DONTLOG     = BIT(2),       /* only clear, don't log */ };

machine_check_poll函数主要功能是通过读取IA32_MCG_STATUS、IA32_MCi_STATUS寄存器信息和CPU的ip、cs等相关信息，然后故障分类，将CE event或其他故障类型event记录到/dev/mcelog。用户可以通过读取/dev/mcelog获取错误记录。

执行流程如下，过程说明在代码注释中

C++ bool machine_check_poll(enum mcp_flags flags, mce_banks_t *b) {         if (flags & MCP_TIMESTAMP)                 m.tsc = rdtsc(); // 记录当前TSC /*CE Error记录*/                 /* If this entry is not valid, ignore it */                 if (!(m.status & MCI_STATUS_VAL))                         continue;                 /*                  * If we are logging everything (at CPU online) or this                  * is a corrected error, then we must log it.                  */                 if ((flags & MCP_UC) || !(m.status & MCI_STATUS_UC))                         goto log_it; /*UCNA Error记录*/                 /*                  * Log UCNA (SDM: 15.6.3 "UCR Error Classification")                  * UC == 1 && PCC == 0 && S == 0                  */                 if (!(m.status & MCI_STATUS_PCC) && !(m.status & MCI_STATUS_S))                         goto log_it; /*通过mce_log记录故障信息*/ log_it:                          /*                  * Don't get the IP here because it's unlikely to                  * have anything to do with the actual error location.                  */                 if (!(flags & MCP_DONTLOG) && !mca_cfg.dont_log_ce)                         mce_log(&m);                 else if (mce_usable_address(&m)) {                         /*                          * Although we skipped logging this, we still want                          * to take action. Add to the pool so the registered                          * notifiers will see it.                          */                         if (!mce_gen_pool_add(&m))                                 mce_schedule_work();                 }         }

总结一下，CMCI是MCA的一个增强特性，主要用于将硬件CE、UCNA等类型故障通过中断方式上报到软件，软件收到中断后，执行中断处理函数intel_threshold_interrupt()采取irq mode或poll mode记录错误信息到/dev/mcelog，用户态可以通过/dev/mcelog获取硬件故障信息。

　　审核编辑：汤梓红