arch/x86/kernel/cpu/mcheck/non-fatal.c

   1 /*
   2  * Non Fatal Machine Check Exception Reporting
   3  *
   4  * (C) Copyright 2002 Dave Jones. <davej@redhat.com>
   5  *
   6  * This file contains routines to check for non-fatal MCEs every 15s
   7  *
   8  */
   9
  10 #include <linux/init.h>
  11 #include <linux/types.h>
  12 #include <linux/kernel.h>
  13 #include <linux/jiffies.h>
  14 #include <linux/workqueue.h>
  15 #include <linux/interrupt.h>
  16 #include <linux/smp.h>
  17 #include <linux/module.h>
  18
  19 #include <asm/processor.h>
  20 #include <asm/system.h>
  21 #include <asm/msr.h>
  22
  23 #include "mce.h"
  24
  25 static int firstbank;
  26
  27 #define MCE_RATE        15*HZ   /* timer rate is 15s */
  28
  29 static void mce_checkregs(void *info)
  30 {
  31         u32 low, high;
  32         int i;
  33
  34         for (i = firstbank; i < nr_mce_banks; i++) {
  35                 rdmsr(MSR_IA32_MC0_STATUS+i*4, low, high);
  36
  37                 if (high & (1<<31)) {
  38                         printk(KERN_INFO "MCE: The hardware reports a non "
  39                                 "fatal, correctable incident occurred on "
  40                                 "CPU %d.\n",
  41                                 smp_processor_id());
  42                         printk(KERN_INFO "Bank %d: %08x%08x\n", i, high, low);
  43
  44                         /*
  45                          * Scrub the error so we don't pick it up in MCE_RATE
  46                          * seconds time.
  47                          */
  48                         wrmsr(MSR_IA32_MC0_STATUS+i*4, 0UL, 0UL);
  49
  50                         /* Serialize */
  51                         wmb();
  52                         add_taint(TAINT_MACHINE_CHECK);
  53                 }
  54         }
  55 }
  56
  57 static void mce_work_fn(struct work_struct *work);
  58 static DECLARE_DELAYED_WORK(mce_work, mce_work_fn);
  59
  60 static void mce_work_fn(struct work_struct *work)
  61 {
  62         on_each_cpu(mce_checkregs, NULL, 1);
  63         schedule_delayed_work(&mce_work, round_jiffies_relative(MCE_RATE));
  64 }
  65
  66 static int __init init_nonfatal_mce_checker(void)
  67 {
  68         struct cpuinfo_x86 *c = &boot_cpu_data;
  69
  70         /* Check for MCE support */
  71         if (!cpu_has(c, X86_FEATURE_MCE))
  72                 return -ENODEV;
  73
  74         /* Check for PPro style MCA */
  75         if (!cpu_has(c, X86_FEATURE_MCA))
  76                 return -ENODEV;
  77
  78         /* Some Athlons misbehave when we frob bank 0 */
  79         if (boot_cpu_data.x86_vendor == X86_VENDOR_AMD &&
  80                 boot_cpu_data.x86 == 6)
  81                         firstbank = 1;
  82         else
  83                         firstbank = 0;
  84
  85         /*
  86          * Check for non-fatal errors every MCE_RATE s
  87          */
  88         schedule_delayed_work(&mce_work, round_jiffies_relative(MCE_RATE));
  89         printk(KERN_INFO "Machine check exception polling timer started.\n");
  90         return 0;
  91 }
  92 module_init(init_nonfatal_mce_checker);
  93
  94 MODULE_LICENSE("GPL");