arch/x86/include/asm/mce.h

   1 #ifndef _ASM_X86_MCE_H
   2 #define _ASM_X86_MCE_H
   3
   4 #include <uapi/asm/mce.h>
   5
   6 /*
   7  * Machine Check support for x86
   8  */
   9
  10 /* MCG_CAP register defines */
  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 #define MCG_ELOG_P              (1ULL<<26)   /* Extended error log supported */
  20 #define MCG_LMCE_P              (1ULL<<27)   /* Local machine check supported */
  21
  22 /* MCG_STATUS register defines */
  23 #define MCG_STATUS_RIPV  (1ULL<<0)   /* restart ip valid */
  24 #define MCG_STATUS_EIPV  (1ULL<<1)   /* ip points to correct instruction */
  25 #define MCG_STATUS_MCIP  (1ULL<<2)   /* machine check in progress */
  26 #define MCG_STATUS_LMCES (1ULL<<3)   /* LMCE signaled */
  27
  28 /* MCG_EXT_CTL register defines */
  29 #define MCG_EXT_CTL_LMCE_EN (1ULL<<0) /* Enable LMCE */
  30
  31 /* MCi_STATUS register defines */
  32 #define MCI_STATUS_VAL   (1ULL<<63)  /* valid error */
  33 #define MCI_STATUS_OVER  (1ULL<<62)  /* previous errors lost */
  34 #define MCI_STATUS_UC    (1ULL<<61)  /* uncorrected error */
  35 #define MCI_STATUS_EN    (1ULL<<60)  /* error enabled */
  36 #define MCI_STATUS_MISCV (1ULL<<59)  /* misc error reg. valid */
  37 #define MCI_STATUS_ADDRV (1ULL<<58)  /* addr reg. valid */
  38 #define MCI_STATUS_PCC   (1ULL<<57)  /* processor context corrupt */
  39 #define MCI_STATUS_S     (1ULL<<56)  /* Signaled machine check */
  40 #define MCI_STATUS_AR    (1ULL<<55)  /* Action required */
  41
  42 /* AMD-specific bits */
  43 #define MCI_STATUS_TCC          (1ULL<<55)  /* Task context corrupt */
  44 #define MCI_STATUS_SYNDV        (1ULL<<53)  /* synd reg. valid */
  45 #define MCI_STATUS_DEFERRED     (1ULL<<44)  /* uncorrected error, deferred exception */
  46 #define MCI_STATUS_POISON       (1ULL<<43)  /* access poisonous data */
  47
  48 /*
  49  * McaX field if set indicates a given bank supports MCA extensions:
  50  *  - Deferred error interrupt type is specifiable by bank.
  51  *  - MCx_MISC0[BlkPtr] field indicates presence of extended MISC registers,
  52  *    But should not be used to determine MSR numbers.
  53  *  - TCC bit is present in MCx_STATUS.
  54  */
  55 #define MCI_CONFIG_MCAX         0x1
  56 #define MCI_IPID_MCATYPE        0xFFFF0000
  57 #define MCI_IPID_HWID           0xFFF
  58
  59 /*
  60  * Note that the full MCACOD field of IA32_MCi_STATUS MSR is
  61  * bits 15:0.  But bit 12 is the 'F' bit, defined for corrected
  62  * errors to indicate that errors are being filtered by hardware.
  63  * We should mask out bit 12 when looking for specific signatures
  64  * of uncorrected errors - so the F bit is deliberately skipped
  65  * in this #define.
  66  */
  67 #define MCACOD            0xefff     /* MCA Error Code */
  68
  69 /* Architecturally defined codes from SDM Vol. 3B Chapter 15 */
  70 #define MCACOD_SCRUB    0x00C0  /* 0xC0-0xCF Memory Scrubbing */
  71 #define MCACOD_SCRUBMSK 0xeff0  /* Skip bit 12 ('F' bit) */
  72 #define MCACOD_L3WB     0x017A  /* L3 Explicit Writeback */
  73 #define MCACOD_DATA     0x0134  /* Data Load */
  74 #define MCACOD_INSTR    0x0150  /* Instruction Fetch */
  75
  76 /* MCi_MISC register defines */
  77 #define MCI_MISC_ADDR_LSB(m)    ((m) & 0x3f)
  78 #define MCI_MISC_ADDR_MODE(m)   (((m) >> 6) & 7)
  79 #define  MCI_MISC_ADDR_SEGOFF   0       /* segment offset */
  80 #define  MCI_MISC_ADDR_LINEAR   1       /* linear address */
  81 #define  MCI_MISC_ADDR_PHYS     2       /* physical address */
  82 #define  MCI_MISC_ADDR_MEM      3       /* memory address */
  83 #define  MCI_MISC_ADDR_GENERIC  7       /* generic */
  84
  85 /* CTL2 register defines */
  86 #define MCI_CTL2_CMCI_EN                (1ULL << 30)
  87 #define MCI_CTL2_CMCI_THRESHOLD_MASK    0x7fffULL
  88
  89 #define MCJ_CTX_MASK            3
  90 #define MCJ_CTX(flags)          ((flags) & MCJ_CTX_MASK)
  91 #define MCJ_CTX_RANDOM          0    /* inject context: random */
  92 #define MCJ_CTX_PROCESS         0x1  /* inject context: process */
  93 #define MCJ_CTX_IRQ             0x2  /* inject context: IRQ */
  94 #define MCJ_NMI_BROADCAST       0x4  /* do NMI broadcasting */
  95 #define MCJ_EXCEPTION           0x8  /* raise as exception */
  96 #define MCJ_IRQ_BROADCAST       0x10 /* do IRQ broadcasting */
  97
  98 #define MCE_OVERFLOW 0          /* bit 0 in flags means overflow */
  99
 100 #define MCE_LOG_LEN 32
 101 #define MCE_LOG_SIGNATURE       "MACHINECHECK"
 102
 103 /* AMD Scalable MCA */
 104 #define MSR_AMD64_SMCA_MC0_CTL          0xc0002000
 105 #define MSR_AMD64_SMCA_MC0_STATUS       0xc0002001
 106 #define MSR_AMD64_SMCA_MC0_ADDR         0xc0002002
 107 #define MSR_AMD64_SMCA_MC0_MISC0        0xc0002003
 108 #define MSR_AMD64_SMCA_MC0_CONFIG       0xc0002004
 109 #define MSR_AMD64_SMCA_MC0_IPID         0xc0002005
 110 #define MSR_AMD64_SMCA_MC0_SYND         0xc0002006
 111 #define MSR_AMD64_SMCA_MC0_DESTAT       0xc0002008
 112 #define MSR_AMD64_SMCA_MC0_DEADDR       0xc0002009
 113 #define MSR_AMD64_SMCA_MC0_MISC1        0xc000200a
 114 #define MSR_AMD64_SMCA_MCx_CTL(x)       (MSR_AMD64_SMCA_MC0_CTL + 0x10*(x))
 115 #define MSR_AMD64_SMCA_MCx_STATUS(x)    (MSR_AMD64_SMCA_MC0_STATUS + 0x10*(x))
 116 #define MSR_AMD64_SMCA_MCx_ADDR(x)      (MSR_AMD64_SMCA_MC0_ADDR + 0x10*(x))
 117 #define MSR_AMD64_SMCA_MCx_MISC(x)      (MSR_AMD64_SMCA_MC0_MISC0 + 0x10*(x))
 118 #define MSR_AMD64_SMCA_MCx_CONFIG(x)    (MSR_AMD64_SMCA_MC0_CONFIG + 0x10*(x))
 119 #define MSR_AMD64_SMCA_MCx_IPID(x)      (MSR_AMD64_SMCA_MC0_IPID + 0x10*(x))
 120 #define MSR_AMD64_SMCA_MCx_SYND(x)      (MSR_AMD64_SMCA_MC0_SYND + 0x10*(x))
 121 #define MSR_AMD64_SMCA_MCx_DESTAT(x)    (MSR_AMD64_SMCA_MC0_DESTAT + 0x10*(x))
 122 #define MSR_AMD64_SMCA_MCx_DEADDR(x)    (MSR_AMD64_SMCA_MC0_DEADDR + 0x10*(x))
 123 #define MSR_AMD64_SMCA_MCx_MISCy(x, y)  ((MSR_AMD64_SMCA_MC0_MISC1 + y) + (0x10*(x)))
 124
 125 /*
 126  * This structure contains all data related to the MCE log.  Also
 127  * carries a signature to make it easier to find from external
 128  * debugging tools.  Each entry is only valid when its finished flag
 129  * is set.
 130  */
 131 struct mce_log {
 132         char signature[12]; /* "MACHINECHECK" */
 133         unsigned len;       /* = MCE_LOG_LEN */
 134         unsigned next;
 135         unsigned flags;
 136         unsigned recordlen;     /* length of struct mce */
 137         struct mce entry[MCE_LOG_LEN];
 138 };
 139
 140 struct mca_config {
 141         bool dont_log_ce;
 142         bool cmci_disabled;
 143         bool lmce_disabled;
 144         bool ignore_ce;
 145         bool disabled;
 146         bool ser;
 147         bool recovery;
 148         bool bios_cmci_threshold;
 149         u8 banks;
 150         s8 bootlog;
 151         int tolerant;
 152         int monarch_timeout;
 153         int panic_timeout;
 154         u32 rip_msr;
 155 };
 156
 157 struct mce_vendor_flags {
 158         /*
 159          * Indicates that overflow conditions are not fatal, when set.
 160          */
 161         __u64 overflow_recov    : 1,
 162
 163         /*
 164          * (AMD) SUCCOR stands for S/W UnCorrectable error COntainment and
 165          * Recovery. It indicates support for data poisoning in HW and deferred
 166          * error interrupts.
 167          */
 168               succor            : 1,
 169
 170         /*
 171          * (AMD) SMCA: This bit indicates support for Scalable MCA which expands
 172          * the register space for each MCA bank and also increases number of
 173          * banks. Also, to accommodate the new banks and registers, the MCA
 174          * register space is moved to a new MSR range.
 175          */
 176               smca              : 1,
 177
 178               __reserved_0      : 61;
 179 };
 180
 181 struct mca_msr_regs {
 182         u32 (*ctl)      (int bank);
 183         u32 (*status)   (int bank);
 184         u32 (*addr)     (int bank);
 185         u32 (*misc)     (int bank);
 186 };
 187
 188 extern struct mce_vendor_flags mce_flags;
 189
 190 extern struct mca_config mca_cfg;
 191 extern struct mca_msr_regs msr_ops;
 192
 193 enum mce_notifier_prios {
 194         MCE_PRIO_SRAO           = INT_MAX,
 195         MCE_PRIO_EXTLOG         = INT_MAX - 1,
 196         MCE_PRIO_NFIT           = INT_MAX - 2,
 197         MCE_PRIO_EDAC           = INT_MAX - 3,
 198         MCE_PRIO_LOWEST         = 0,
 199 };
 200
 201 extern void mce_register_decode_chain(struct notifier_block *nb);
 202 extern void mce_unregister_decode_chain(struct notifier_block *nb);
 203
 204 #include <linux/percpu.h>
 205 #include <linux/atomic.h>
 206
 207 extern int mce_p5_enabled;
 208
 209 #ifdef CONFIG_X86_MCE
 210 int mcheck_init(void);
 211 void mcheck_cpu_init(struct cpuinfo_x86 *c);
 212 void mcheck_cpu_clear(struct cpuinfo_x86 *c);
 213 void mcheck_vendor_init_severity(void);
 214 #else
 215 static inline int mcheck_init(void) { return 0; }
 216 static inline void mcheck_cpu_init(struct cpuinfo_x86 *c) {}
 217 static inline void mcheck_cpu_clear(struct cpuinfo_x86 *c) {}
 218 static inline void mcheck_vendor_init_severity(void) {}
 219 #endif
 220
 221 #ifdef CONFIG_X86_ANCIENT_MCE
 222 void intel_p5_mcheck_init(struct cpuinfo_x86 *c);
 223 void winchip_mcheck_init(struct cpuinfo_x86 *c);
 224 static inline void enable_p5_mce(void) { mce_p5_enabled = 1; }
 225 #else
 226 static inline void intel_p5_mcheck_init(struct cpuinfo_x86 *c) {}
 227 static inline void winchip_mcheck_init(struct cpuinfo_x86 *c) {}
 228 static inline void enable_p5_mce(void) {}
 229 #endif
 230
 231 void mce_setup(struct mce *m);
 232 void mce_log(struct mce *m);
 233 DECLARE_PER_CPU(struct device *, mce_device);
 234
 235 /*
 236  * Maximum banks number.
 237  * This is the limit of the current register layout on
 238  * Intel CPUs.
 239  */
 240 #define MAX_NR_BANKS 32
 241
 242 #ifdef CONFIG_X86_MCE_INTEL
 243 void mce_intel_feature_init(struct cpuinfo_x86 *c);
 244 void mce_intel_feature_clear(struct cpuinfo_x86 *c);
 245 void cmci_clear(void);
 246 void cmci_reenable(void);
 247 void cmci_rediscover(void);
 248 void cmci_recheck(void);
 249 #else
 250 static inline void mce_intel_feature_init(struct cpuinfo_x86 *c) { }
 251 static inline void mce_intel_feature_clear(struct cpuinfo_x86 *c) { }
 252 static inline void cmci_clear(void) {}
 253 static inline void cmci_reenable(void) {}
 254 static inline void cmci_rediscover(void) {}
 255 static inline void cmci_recheck(void) {}
 256 #endif
 257
 258 #ifdef CONFIG_X86_MCE_AMD
 259 void mce_amd_feature_init(struct cpuinfo_x86 *c);
 260 int umc_normaddr_to_sysaddr(u64 norm_addr, u16 nid, u8 umc, u64 *sys_addr);
 261 #else
 262 static inline void mce_amd_feature_init(struct cpuinfo_x86 *c) { }
 263 static inline int umc_normaddr_to_sysaddr(u64 norm_addr, u16 nid, u8 umc, u64 *sys_addr) { return -EINVAL; };
 264 #endif
 265
 266 int mce_available(struct cpuinfo_x86 *c);
 267 bool mce_is_memory_error(struct mce *m);
 268
 269 DECLARE_PER_CPU(unsigned, mce_exception_count);
 270 DECLARE_PER_CPU(unsigned, mce_poll_count);
 271
 272 typedef DECLARE_BITMAP(mce_banks_t, MAX_NR_BANKS);
 273 DECLARE_PER_CPU(mce_banks_t, mce_poll_banks);
 274
 275 enum mcp_flags {
 276         MCP_TIMESTAMP   = BIT(0),       /* log time stamp */
 277         MCP_UC          = BIT(1),       /* log uncorrected errors */
 278         MCP_DONTLOG     = BIT(2),       /* only clear, don't log */
 279 };
 280 bool machine_check_poll(enum mcp_flags flags, mce_banks_t *b);
 281
 282 int mce_notify_irq(void);
 283
 284 DECLARE_PER_CPU(struct mce, injectm);
 285
 286 extern void register_mce_write_callback(ssize_t (*)(struct file *filp,
 287                                     const char __user *ubuf,
 288                                     size_t usize, loff_t *off));
 289
 290 /* Disable CMCI/polling for MCA bank claimed by firmware */
 291 extern void mce_disable_bank(int bank);
 292
 293 /*
 294  * Exception handler
 295  */
 296
 297 /* Call the installed machine check handler for this CPU setup. */
 298 extern void (*machine_check_vector)(struct pt_regs *, long error_code);
 299 void do_machine_check(struct pt_regs *, long);
 300
 301 /*
 302  * Threshold handler
 303  */
 304 extern void (*mce_threshold_vector)(void);
 305
 306 /* Deferred error interrupt handler */
 307 extern void (*deferred_error_int_vector)(void);
 308
 309 /*
 310  * Thermal handler
 311  */
 312
 313 void intel_init_thermal(struct cpuinfo_x86 *c);
 314
 315 /* Interrupt Handler for core thermal thresholds */
 316 extern int (*platform_thermal_notify)(__u64 msr_val);
 317
 318 /* Interrupt Handler for package thermal thresholds */
 319 extern int (*platform_thermal_package_notify)(__u64 msr_val);
 320
 321 /* Callback support of rate control, return true, if
 322  * callback has rate control */
 323 extern bool (*platform_thermal_package_rate_control)(void);
 324
 325 #ifdef CONFIG_X86_THERMAL_VECTOR
 326 extern void mcheck_intel_therm_init(void);
 327 #else
 328 static inline void mcheck_intel_therm_init(void) { }
 329 #endif
 330
 331 /*
 332  * Used by APEI to report memory error via /dev/mcelog
 333  */
 334
 335 struct cper_sec_mem_err;
 336 extern void apei_mce_report_mem_error(int corrected,
 337                                       struct cper_sec_mem_err *mem_err);
 338
 339 /*
 340  * Enumerate new IP types and HWID values in AMD processors which support
 341  * Scalable MCA.
 342  */
 343 #ifdef CONFIG_X86_MCE_AMD
 344
 345 /* These may be used by multiple smca_hwid_mcatypes */
 346 enum smca_bank_types {
 347         SMCA_LS = 0,    /* Load Store */
 348         SMCA_IF,        /* Instruction Fetch */
 349         SMCA_L2_CACHE,  /* L2 Cache */
 350         SMCA_DE,        /* Decoder Unit */
 351         SMCA_EX,        /* Execution Unit */
 352         SMCA_FP,        /* Floating Point */
 353         SMCA_L3_CACHE,  /* L3 Cache */
 354         SMCA_CS,        /* Coherent Slave */
 355         SMCA_PIE,       /* Power, Interrupts, etc. */
 356         SMCA_UMC,       /* Unified Memory Controller */
 357         SMCA_PB,        /* Parameter Block */
 358         SMCA_PSP,       /* Platform Security Processor */
 359         SMCA_SMU,       /* System Management Unit */
 360         N_SMCA_BANK_TYPES
 361 };
 362
 363 #define HWID_MCATYPE(hwid, mcatype) (((hwid) << 16) | (mcatype))
 364
 365 struct smca_hwid {
 366         unsigned int bank_type; /* Use with smca_bank_types for easy indexing. */
 367         u32 hwid_mcatype;       /* (hwid,mcatype) tuple */
 368         u32 xec_bitmap;         /* Bitmap of valid ExtErrorCodes; current max is 21. */
 369         u8 count;               /* Number of instances. */
 370 };
 371
 372 struct smca_bank {
 373         struct smca_hwid *hwid;
 374         u32 id;                 /* Value of MCA_IPID[InstanceId]. */
 375         u8 sysfs_id;            /* Value used for sysfs name. */
 376 };
 377
 378 extern struct smca_bank smca_banks[MAX_NR_BANKS];
 379
 380 extern const char *smca_get_long_name(enum smca_bank_types t);
 381
 382 extern int mce_threshold_create_device(unsigned int cpu);
 383 extern int mce_threshold_remove_device(unsigned int cpu);
 384
 385 #else
 386
 387 static inline int mce_threshold_create_device(unsigned int cpu) { return 0; };
 388 static inline int mce_threshold_remove_device(unsigned int cpu) { return 0; };
 389
 390 #endif
 391
 392 #endif /* _ASM_X86_MCE_H */