arch/x86/include/asm/mce.h

   1 #ifndef _ASM_X86_MCE_H
   2 #define _ASM_X86_MCE_H
   3
   4 #include <linux/types.h>
   5 #include <asm/ioctls.h>
   6
   7 /*
   8  * Machine Check support for x86
   9  */
  10
  11 #define MCG_BANKCNT_MASK        0xff         /* Number of Banks */
  12 #define MCG_CTL_P               (1ULL<<8)    /* MCG_CTL register available */
  13 #define MCG_EXT_P               (1ULL<<9)    /* Extended registers available */
  14 #define MCG_CMCI_P              (1ULL<<10)   /* CMCI supported */
  15 #define MCG_EXT_CNT_MASK        0xff0000     /* Number of Extended registers */
  16 #define MCG_EXT_CNT_SHIFT       16
  17 #define MCG_EXT_CNT(c)          (((c) & MCG_EXT_CNT_MASK) >> MCG_EXT_CNT_SHIFT)
  18 #define MCG_SER_P               (1ULL<<24)   /* MCA recovery/new status bits */
  19
  20 #define MCG_STATUS_RIPV  (1ULL<<0)   /* restart ip valid */
  21 #define MCG_STATUS_EIPV  (1ULL<<1)   /* ip points to correct instruction */
  22 #define MCG_STATUS_MCIP  (1ULL<<2)   /* machine check in progress */
  23
  24 #define MCI_STATUS_VAL   (1ULL<<63)  /* valid error */
  25 #define MCI_STATUS_OVER  (1ULL<<62)  /* previous errors lost */
  26 #define MCI_STATUS_UC    (1ULL<<61)  /* uncorrected error */
  27 #define MCI_STATUS_EN    (1ULL<<60)  /* error enabled */
  28 #define MCI_STATUS_MISCV (1ULL<<59)  /* misc error reg. valid */
  29 #define MCI_STATUS_ADDRV (1ULL<<58)  /* addr reg. valid */
  30 #define MCI_STATUS_PCC   (1ULL<<57)  /* processor context corrupt */
  31 #define MCI_STATUS_S     (1ULL<<56)  /* Signaled machine check */
  32 #define MCI_STATUS_AR    (1ULL<<55)  /* Action required */
  33
  34 /* MISC register defines */
  35 #define MCM_ADDR_SEGOFF  0      /* segment offset */
  36 #define MCM_ADDR_LINEAR  1      /* linear address */
  37 #define MCM_ADDR_PHYS    2      /* physical address */
  38 #define MCM_ADDR_MEM     3      /* memory address */
  39 #define MCM_ADDR_GENERIC 7      /* generic */
  40
  41 #define MCJ_CTX_MASK            3
  42 #define MCJ_CTX(flags)          ((flags) & MCJ_CTX_MASK)
  43 #define MCJ_CTX_RANDOM          0    /* inject context: random */
  44 #define MCJ_CTX_PROCESS         1    /* inject context: process */
  45 #define MCJ_CTX_IRQ             2    /* inject context: IRQ */
  46 #define MCJ_NMI_BROADCAST       4    /* do NMI broadcasting */
  47 #define MCJ_EXCEPTION           8    /* raise as exception */
  48
  49 /* Fields are zero when not available */
  50 struct mce {
  51         __u64 status;
  52         __u64 misc;
  53         __u64 addr;
  54         __u64 mcgstatus;
  55         __u64 ip;
  56         __u64 tsc;      /* cpu time stamp counter */
  57         __u64 time;     /* wall time_t when error was detected */
  58         __u8  cpuvendor;        /* cpu vendor as encoded in system.h */
  59         __u8  inject_flags;     /* software inject flags */
  60         __u16  pad;
  61         __u32 cpuid;    /* CPUID 1 EAX */
  62         __u8  cs;               /* code segment */
  63         __u8  bank;     /* machine check bank */
  64         __u8  cpu;      /* cpu number; obsolete; use extcpu now */
  65         __u8  finished;   /* entry is valid */
  66         __u32 extcpu;   /* linux cpu number that detected the error */
  67         __u32 socketid; /* CPU socket ID */
  68         __u32 apicid;   /* CPU initial apic ID */
  69         __u64 mcgcap;   /* MCGCAP MSR: machine check capabilities of CPU */
  70 };
  71
  72 /*
  73  * This structure contains all data related to the MCE log.  Also
  74  * carries a signature to make it easier to find from external
  75  * debugging tools.  Each entry is only valid when its finished flag
  76  * is set.
  77  */
  78
  79 #define MCE_LOG_LEN 32
  80
  81 struct mce_log {
  82         char signature[12]; /* "MACHINECHECK" */
  83         unsigned len;       /* = MCE_LOG_LEN */
  84         unsigned next;
  85         unsigned flags;
  86         unsigned recordlen;     /* length of struct mce */
  87         struct mce entry[MCE_LOG_LEN];
  88 };
  89
  90 #define MCE_OVERFLOW 0          /* bit 0 in flags means overflow */
  91
  92 #define MCE_LOG_SIGNATURE       "MACHINECHECK"
  93
  94 #define MCE_GET_RECORD_LEN   _IOR('M', 1, int)
  95 #define MCE_GET_LOG_LEN      _IOR('M', 2, int)
  96 #define MCE_GETCLEAR_FLAGS   _IOR('M', 3, int)
  97
  98 /* Software defined banks */
  99 #define MCE_EXTENDED_BANK       128
 100 #define MCE_THERMAL_BANK        MCE_EXTENDED_BANK + 0
 101
 102 #define K8_MCE_THRESHOLD_BASE      (MCE_EXTENDED_BANK + 1)      /* MCE_AMD */
 103 #define K8_MCE_THRESHOLD_BANK_0    (MCE_THRESHOLD_BASE + 0 * 9)
 104 #define K8_MCE_THRESHOLD_BANK_1    (MCE_THRESHOLD_BASE + 1 * 9)
 105 #define K8_MCE_THRESHOLD_BANK_2    (MCE_THRESHOLD_BASE + 2 * 9)
 106 #define K8_MCE_THRESHOLD_BANK_3    (MCE_THRESHOLD_BASE + 3 * 9)
 107 #define K8_MCE_THRESHOLD_BANK_4    (MCE_THRESHOLD_BASE + 4 * 9)
 108 #define K8_MCE_THRESHOLD_BANK_5    (MCE_THRESHOLD_BASE + 5 * 9)
 109 #define K8_MCE_THRESHOLD_DRAM_ECC  (MCE_THRESHOLD_BANK_4 + 0)
 110
 111 #ifdef __KERNEL__
 112
 113 #include <linux/percpu.h>
 114 #include <linux/init.h>
 115 #include <asm/atomic.h>
 116
 117 extern int mce_disabled;
 118 extern int mce_p5_enabled;
 119
 120 #ifdef CONFIG_X86_MCE
 121 void mcheck_init(struct cpuinfo_x86 *c);
 122 #else
 123 static inline void mcheck_init(struct cpuinfo_x86 *c) {}
 124 #endif
 125
 126 #ifdef CONFIG_X86_ANCIENT_MCE
 127 void intel_p5_mcheck_init(struct cpuinfo_x86 *c);
 128 void winchip_mcheck_init(struct cpuinfo_x86 *c);
 129 static inline void enable_p5_mce(void) { mce_p5_enabled = 1; }
 130 #else
 131 static inline void intel_p5_mcheck_init(struct cpuinfo_x86 *c) {}
 132 static inline void winchip_mcheck_init(struct cpuinfo_x86 *c) {}
 133 static inline void enable_p5_mce(void) {}
 134 #endif
 135
 136 void mce_setup(struct mce *m);
 137 void mce_log(struct mce *m);
 138 DECLARE_PER_CPU(struct sys_device, mce_dev);
 139
 140 /*
 141  * Maximum banks number.
 142  * This is the limit of the current register layout on
 143  * Intel CPUs.
 144  */
 145 #define MAX_NR_BANKS 32
 146
 147 #ifdef CONFIG_X86_MCE_INTEL
 148 extern int mce_cmci_disabled;
 149 extern int mce_ignore_ce;
 150 void mce_intel_feature_init(struct cpuinfo_x86 *c);
 151 void cmci_clear(void);
 152 void cmci_reenable(void);
 153 void cmci_rediscover(int dying);
 154 void cmci_recheck(void);
 155 #else
 156 static inline void mce_intel_feature_init(struct cpuinfo_x86 *c) { }
 157 static inline void cmci_clear(void) {}
 158 static inline void cmci_reenable(void) {}
 159 static inline void cmci_rediscover(int dying) {}
 160 static inline void cmci_recheck(void) {}
 161 #endif
 162
 163 #ifdef CONFIG_X86_MCE_AMD
 164 void mce_amd_feature_init(struct cpuinfo_x86 *c);
 165 #else
 166 static inline void mce_amd_feature_init(struct cpuinfo_x86 *c) { }
 167 #endif
 168
 169 int mce_available(struct cpuinfo_x86 *c);
 170
 171 DECLARE_PER_CPU(unsigned, mce_exception_count);
 172 DECLARE_PER_CPU(unsigned, mce_poll_count);
 173
 174 extern atomic_t mce_entry;
 175
 176 typedef DECLARE_BITMAP(mce_banks_t, MAX_NR_BANKS);
 177 DECLARE_PER_CPU(mce_banks_t, mce_poll_banks);
 178
 179 enum mcp_flags {
 180         MCP_TIMESTAMP = (1 << 0),       /* log time stamp */
 181         MCP_UC = (1 << 1),              /* log uncorrected errors */
 182         MCP_DONTLOG = (1 << 2),         /* only clear, don't log */
 183 };
 184 void machine_check_poll(enum mcp_flags flags, mce_banks_t *b);
 185
 186 int mce_notify_irq(void);
 187 void mce_notify_process(void);
 188
 189 DECLARE_PER_CPU(struct mce, injectm);
 190 extern struct file_operations mce_chrdev_ops;
 191
 192 /*
 193  * Exception handler
 194  */
 195
 196 /* Call the installed machine check handler for this CPU setup. */
 197 extern void (*machine_check_vector)(struct pt_regs *, long error_code);
 198 void do_machine_check(struct pt_regs *, long);
 199
 200 /*
 201  * Threshold handler
 202  */
 203
 204 extern void (*mce_threshold_vector)(void);
 205 extern void (*threshold_cpu_callback)(unsigned long action, unsigned int cpu);
 206
 207 /*
 208  * Thermal handler
 209  */
 210
 211 void intel_init_thermal(struct cpuinfo_x86 *c);
 212
 213 void mce_log_therm_throt_event(__u64 status);
 214
 215 #endif /* __KERNEL__ */
 216 #endif /* _ASM_X86_MCE_H */