osfmk/i386/machine_check.c

   1 /*
   2  * Copyright (c) 2007 Apple Inc. All rights reserved.
   3  *
   4  * @APPLE_OSREFERENCE_LICENSE_HEADER_START@
   5  *
   6  * This file contains Original Code and/or Modifications of Original Code
   7  * as defined in and that are subject to the Apple Public Source License
   8  * Version 2.0 (the 'License'). You may not use this file except in
   9  * compliance with the License. The rights granted to you under the License
  10  * may not be used to create, or enable the creation or redistribution of,
  11  * unlawful or unlicensed copies of an Apple operating system, or to
  12  * circumvent, violate, or enable the circumvention or violation of, any
  13  * terms of an Apple operating system software license agreement.
  14  *
  15  * Please obtain a copy of the License at
  16  * http://www.opensource.apple.com/apsl/ and read it before using this file.
  17  *
  18  * The Original Code and all software distributed under the License are
  19  * distributed on an 'AS IS' basis, WITHOUT WARRANTY OF ANY KIND, EITHER
  20  * EXPRESS OR IMPLIED, AND APPLE HEREBY DISCLAIMS ALL SUCH WARRANTIES,
  21  * INCLUDING WITHOUT LIMITATION, ANY WARRANTIES OF MERCHANTABILITY,
  22  * FITNESS FOR A PARTICULAR PURPOSE, QUIET ENJOYMENT OR NON-INFRINGEMENT.
  23  * Please see the License for the specific language governing rights and
  24  * limitations under the License.
  25  *
  26  * @APPLE_OSREFERENCE_LICENSE_HEADER_END@
  27  */
  28
  29 #include <kern/kalloc.h>
  30 #include <i386/cpu_data.h>
  31 #include <i386/cpuid.h>
  32 #include <i386/machine_check.h>
  33 #include <i386/proc_reg.h>
  34
  35 #define IF(bool,str)    ((bool) ? (str) : "")
  36
  37 static boolean_t        mca_initialized = FALSE;
  38 static boolean_t        mca_MCE_present = FALSE;
  39 static boolean_t        mca_MCA_present = FALSE;
  40 static uint32_t         mca_family = 0;
  41 static unsigned int     mca_error_bank_count = 0;
  42 static boolean_t        mca_control_MSR_present = FALSE;
  43 static boolean_t        mca_threshold_status_present = FALSE;
  44 static boolean_t        mca_extended_MSRs_present = FALSE;
  45 static unsigned int     mca_extended_MSRs_count = 0;
  46 static ia32_mcg_cap_t   ia32_mcg_cap;
  47 static boolean_t        mca_exception_taken = FALSE;
  48
  49 decl_simple_lock_data(static, mca_lock);
  50
  51 typedef struct {
  52         ia32_mci_ctl_t          mca_mci_ctl;
  53         ia32_mci_status_t       mca_mci_status;
  54         ia32_mci_misc_t         mca_mci_misc;
  55         ia32_mci_addr_t         mca_mci_addr;
  56 } mca_mci_bank_t;
  57
  58 typedef struct mca_state {
  59         ia32_mcg_ctl_t          mca_mcg_ctl;
  60         ia32_mcg_status_t       mca_mcg_status;
  61         mca_mci_bank_t          mca_error_bank[0];
  62 } mca_state_t;
  63
  64 static void
  65 mca_get_availability(void)
  66 {
  67         uint64_t        features = cpuid_info()->cpuid_features;
  68         uint32_t        family =  cpuid_info()->cpuid_family;
  69
  70         mca_MCE_present = (features & CPUID_FEATURE_MCE) != 0;
  71         mca_MCA_present = (features & CPUID_FEATURE_MCA) != 0;
  72         mca_family = family;
  73
  74         /*
  75          * If MCA, the number of banks etc is reported by the IA32_MCG_CAP MSR.
  76          */
  77         if (mca_MCA_present) {
  78                 ia32_mcg_cap.u64 = rdmsr64(IA32_MCG_CAP);
  79                 mca_error_bank_count = ia32_mcg_cap.bits.count;
  80                 mca_control_MSR_present = ia32_mcg_cap.bits.mcg_ctl_p;
  81                 mca_threshold_status_present = ia32_mcg_cap.bits.mcg_tes_p;
  82                 if (family == 0x0F) {
  83                         mca_extended_MSRs_present = ia32_mcg_cap.bits.mcg_ext_p;
  84                         mca_extended_MSRs_count = ia32_mcg_cap.bits.mcg_ext_cnt;
  85                 }
  86         }
  87 }
  88
  89 void
  90 mca_cpu_init(void)
  91 {
  92         unsigned int    i;
  93
  94         /*
  95          * The first (boot) processor is responsible for discovering the
  96          * machine check architecture present on this machine.
  97          */
  98         if (!mca_initialized) {
  99                 mca_get_availability();
 100                 mca_initialized = TRUE;
 101                 simple_lock_init(&mca_lock, 0);
 102         }
 103
 104         if (mca_MCA_present) {
 105
 106                 /* Enable all MCA features */
 107                 if (mca_control_MSR_present)
 108                         wrmsr64(IA32_MCG_CTL, IA32_MCG_CTL_ENABLE);
 109
 110                 switch (mca_family) {
 111                 case 0x06:
 112                         /* Enable all but mc0 */
 113                         for (i = 1; i < mca_error_bank_count; i++)
 114                                 wrmsr64(IA32_MCi_CTL(i),0xFFFFFFFFFFFFFFFFULL);
 115
 116                         /* Clear all errors */
 117                         for (i = 0; i < mca_error_bank_count; i++)
 118                                 wrmsr64(IA32_MCi_STATUS(i), 0ULL);
 119                         break;
 120                 case 0x0F:
 121                         /* Enable all banks */
 122                         for (i = 0; i < mca_error_bank_count; i++)
 123                                 wrmsr64(IA32_MCi_CTL(i),0xFFFFFFFFFFFFFFFFULL);
 124
 125                         /* Clear all errors */
 126                         for (i = 0; i < mca_error_bank_count; i++)
 127                                 wrmsr64(IA32_MCi_STATUS(i), 0ULL);
 128                         break;
 129                 }
 130         }
 131
 132         /* Enable machine check exception handling if available */
 133         if (mca_MCE_present) {
 134                 set_cr4(get_cr4()|CR4_MCE);
 135         }
 136 }
 137
 138 void
 139 mca_cpu_alloc(cpu_data_t        *cdp)
 140 {
 141         vm_size_t       mca_state_size;
 142
 143         /*
 144          * Allocate space for an array of error banks.
 145          */
 146         mca_state_size = sizeof(mca_state_t) +
 147                                 sizeof(mca_mci_bank_t) * mca_error_bank_count;
 148         cdp->cpu_mca_state = kalloc(mca_state_size);
 149         if (cdp->cpu_mca_state == NULL) {
 150                 printf("mca_cpu_alloc() failed for cpu %d\n", cdp->cpu_number);
 151                 return;
 152         }
 153         bzero((void *) cdp->cpu_mca_state, mca_state_size);
 154
 155         /*
 156          * If the boot processor is yet have its allocation made,
 157          * do this now.
 158          */
 159         if (cpu_datap(master_cpu)->cpu_mca_state == NULL)
 160                 mca_cpu_alloc(cpu_datap(master_cpu));
 161 }
 162
 163 static void
 164 mca_save_state(void)
 165 {
 166         mca_state_t     *mca_state;
 167         mca_mci_bank_t  *bank;
 168         unsigned int    i;
 169
 170         assert(!ml_get_interrupts_enabled() || get_preemption_level() > 0);
 171
 172         mca_state = (mca_state_t *) current_cpu_datap()->cpu_mca_state;
 173         if  (mca_state == NULL)
 174                 return;
 175
 176         mca_state->mca_mcg_ctl = mca_control_MSR_present ?
 177                                         rdmsr64(IA32_MCG_CTL) : 0ULL;
 178         mca_state->mca_mcg_status.u64 = rdmsr64(IA32_MCG_STATUS);
 179
 180         bank = (mca_mci_bank_t *) &mca_state->mca_error_bank[0];
 181         for (i = 0; i < mca_error_bank_count; i++, bank++) {
 182                 bank->mca_mci_ctl        = rdmsr64(IA32_MCi_CTL(i));
 183                 bank->mca_mci_status.u64 = rdmsr64(IA32_MCi_STATUS(i));
 184                 if (!bank->mca_mci_status.bits.val)
 185                         continue;
 186                 bank->mca_mci_misc = (bank->mca_mci_status.bits.miscv)?
 187                                         rdmsr64(IA32_MCi_MISC(i)) : 0ULL;
 188                 bank->mca_mci_addr = (bank->mca_mci_status.bits.addrv)?
 189                                         rdmsr64(IA32_MCi_ADDR(i)) : 0ULL;
 190         }
 191 }
 192
 193 void
 194 mca_check_save(void)
 195 {
 196         if (mca_exception_taken)
 197                 mca_save_state();
 198 }
 199
 200 static void mca_dump_64bit_state(void)
 201 {
 202         kdb_printf("Extended Machine Check State:\n");
 203         kdb_printf("  IA32_MCG_RAX:    0x%016qx\n", rdmsr64(IA32_MCG_RAX));
 204         kdb_printf("  IA32_MCG_RBX:    0x%016qx\n", rdmsr64(IA32_MCG_RBX));
 205         kdb_printf("  IA32_MCG_RCX:    0x%016qx\n", rdmsr64(IA32_MCG_RCX));
 206         kdb_printf("  IA32_MCG_RDX:    0x%016qx\n", rdmsr64(IA32_MCG_RDX));
 207         kdb_printf("  IA32_MCG_RSI:    0x%016qx\n", rdmsr64(IA32_MCG_RSI));
 208         kdb_printf("  IA32_MCG_RDI:    0x%016qx\n", rdmsr64(IA32_MCG_RDI));
 209         kdb_printf("  IA32_MCG_RBP:    0x%016qx\n", rdmsr64(IA32_MCG_RBP));
 210         kdb_printf("  IA32_MCG_RSP:    0x%016qx\n", rdmsr64(IA32_MCG_RSP));
 211         kdb_printf("  IA32_MCG_RFLAGS: 0x%016qx\n", rdmsr64(IA32_MCG_RFLAGS));
 212         kdb_printf("  IA32_MCG_RIP:    0x%016qx\n", rdmsr64(IA32_MCG_RIP));
 213         kdb_printf("  IA32_MCG_MISC:   0x%016qx\n", rdmsr64(IA32_MCG_MISC));
 214         kdb_printf("  IA32_MCG_R8:     0x%016qx\n", rdmsr64(IA32_MCG_R8));
 215         kdb_printf("  IA32_MCG_R9:     0x%016qx\n", rdmsr64(IA32_MCG_R9));
 216         kdb_printf("  IA32_MCG_R10:    0x%016qx\n", rdmsr64(IA32_MCG_R10));
 217         kdb_printf("  IA32_MCG_R11:    0x%016qx\n", rdmsr64(IA32_MCG_R11));
 218         kdb_printf("  IA32_MCG_R12:    0x%016qx\n", rdmsr64(IA32_MCG_R12));
 219         kdb_printf("  IA32_MCG_R13:    0x%016qx\n", rdmsr64(IA32_MCG_R13));
 220         kdb_printf("  IA32_MCG_R14:    0x%016qx\n", rdmsr64(IA32_MCG_R14));
 221         kdb_printf("  IA32_MCG_R15:    0x%016qx\n", rdmsr64(IA32_MCG_R15));
 222 }
 223
 224 static uint32_t rdmsr32(uint32_t msr)
 225 {
 226         return (uint32_t) rdmsr64(msr);
 227 }
 228
 229 static void mca_dump_32bit_state(void)
 230 {
 231         kdb_printf("Extended Machine Check State:\n");
 232         kdb_printf("  IA32_MCG_EAX:    0x%08x\n", rdmsr32(IA32_MCG_EAX));
 233         kdb_printf("  IA32_MCG_EBX:    0x%08x\n", rdmsr32(IA32_MCG_EBX));
 234         kdb_printf("  IA32_MCG_ECX:    0x%08x\n", rdmsr32(IA32_MCG_ECX));
 235         kdb_printf("  IA32_MCG_EDX:    0x%08x\n", rdmsr32(IA32_MCG_EDX));
 236         kdb_printf("  IA32_MCG_ESI:    0x%08x\n", rdmsr32(IA32_MCG_ESI));
 237         kdb_printf("  IA32_MCG_EDI:    0x%08x\n", rdmsr32(IA32_MCG_EDI));
 238         kdb_printf("  IA32_MCG_EBP:    0x%08x\n", rdmsr32(IA32_MCG_EBP));
 239         kdb_printf("  IA32_MCG_ESP:    0x%08x\n", rdmsr32(IA32_MCG_ESP));
 240         kdb_printf("  IA32_MCG_EFLAGS: 0x%08x\n", rdmsr32(IA32_MCG_EFLAGS));
 241         kdb_printf("  IA32_MCG_EIP:    0x%08x\n", rdmsr32(IA32_MCG_EIP));
 242         kdb_printf("  IA32_MCG_MISC:   0x%08x\n", rdmsr32(IA32_MCG_MISC));
 243 }
 244
 245 static void
 246 mca_report_cpu_info(void)
 247 {
 248         uint64_t        microcode;
 249         i386_cpu_info_t *infop = cpuid_info();
 250
 251         // microcode revision is top 32 bits of MSR_IA32_UCODE_REV
 252         microcode = rdmsr64(MSR_IA32_UCODE_REV) >> 32;
 253         kdb_printf("family: %d model: %d stepping: %d microcode revision %d\n",
 254                 infop->cpuid_family,
 255                 infop->cpuid_model,
 256                 infop->cpuid_stepping,
 257                 (uint32_t) microcode);
 258         kdb_printf("%s\n", infop->cpuid_brand_string);
 259 }
 260
 261
 262 static const char *mca_threshold_status[] = {
 263         [THRESHOLD_STATUS_NO_TRACKING]  "No tracking",
 264         [THRESHOLD_STATUS_GREEN]        "Green",
 265         [THRESHOLD_STATUS_YELLOW]       "Yellow",
 266         [THRESHOLD_STATUS_RESERVED]     "Reserved"
 267 };
 268
 269 static void
 270 mca_dump_error_banks(void)
 271 {
 272         unsigned int            i;
 273         ia32_mci_status_t       status;
 274
 275         kdb_printf("MCA error-reporting registers:\n");
 276         for (i = 0; i < mca_error_bank_count; i++ ) {
 277                 status.u64 = rdmsr64(IA32_MCi_STATUS(i));
 278                 kdb_printf(
 279                         " IA32_MC%d_STATUS(0x%x): 0x%016qx %svalid\n",
 280                         i, IA32_MCi_STATUS(i), status.u64,
 281                         IF(!status.bits.val, "in"));
 282                 if (!status.bits.val)
 283                         continue;
 284                 kdb_printf(
 285                         "  MCA error code           : 0x%04x\n",
 286                         status.bits.mca_error);
 287                 kdb_printf(
 288                         "  Model specific error code: 0x%04x\n",
 289                         status.bits.model_specific_error);
 290                 if (!mca_threshold_status_present) {
 291                         kdb_printf(
 292                                 "  Other information        : 0x%08x\n",
 293                                 status.bits.other_information);
 294                 } else {
 295                         int     threshold = status.bits_tes_p.threshold;
 296                         kdb_printf(
 297                                 "  Other information        : 0x%08x\n"
 298                                 "  Threshold-based status   : %s\n",
 299                                 status.bits_tes_p.other_information,
 300                                 (status.bits_tes_p.uc == 0) ?
 301                                         mca_threshold_status[threshold] :
 302                                         "Undefined");
 303                 }
 304                 kdb_printf(
 305                         "  Status bits:\n%s%s%s%s%s%s",
 306                         IF(status.bits.pcc,   "   Processor context corrupt\n"),
 307                         IF(status.bits.addrv, "   ADDR register valid\n"),
 308                         IF(status.bits.miscv, "   MISC register valid\n"),
 309                         IF(status.bits.en,    "   Error enabled\n"),
 310                         IF(status.bits.uc,    "   Uncorrected error\n"),
 311                         IF(status.bits.over,  "   Error overflow\n"));
 312                 if (status.bits.addrv)
 313                         kdb_printf(
 314                                 "  IA32_MC%d_ADDR(0x%x): 0x%016qx\n",
 315                                 i, IA32_MCi_ADDR(i), rdmsr64(IA32_MCi_ADDR(i)));
 316                 if (status.bits.miscv)
 317                         kdb_printf(
 318                                 "  IA32_MC%d_MISC(0x%x): 0x%016qx\n",
 319                                 i, IA32_MCi_MISC(i), rdmsr64(IA32_MCi_MISC(i)));
 320         }
 321 }
 322
 323 void
 324 mca_dump(void)
 325 {
 326         ia32_mcg_status_t       status;
 327
 328         mca_save_state();
 329
 330         /*
 331          * Serialize in case of multiple simultaneous machine-checks.
 332          * Only the first caller is allowed to print MCA registers.
 333          */
 334         simple_lock(&mca_lock);
 335         if (mca_exception_taken) {
 336                 simple_unlock(&mca_lock);
 337                 return;
 338         }
 339         mca_exception_taken = TRUE;
 340
 341         /*
 342          * Report machine-check capabilities:
 343          */
 344         kdb_printf(
 345                 "Machine-check capabilities (cpu %d) 0x%016qx:\n",
 346                 cpu_number(), ia32_mcg_cap.u64);
 347
 348         mca_report_cpu_info();
 349
 350         kdb_printf(
 351                 " %d error-reporting banks\n%s%s", mca_error_bank_count,
 352                 IF(mca_control_MSR_present,
 353                    " control MSR present\n"),
 354                 IF(mca_threshold_status_present,
 355                    " threshold-based error status present\n"));
 356         if (mca_extended_MSRs_present)
 357                 kdb_printf(
 358                         " %d extended MSRs present\n", mca_extended_MSRs_count);
 359
 360         /*
 361          * Report machine-check status:
 362          */
 363         status.u64 = rdmsr64(IA32_MCG_STATUS);
 364         kdb_printf(
 365                 "Machine-check status 0x%016qx\n%s%s%s", status.u64,
 366                 IF(status.bits.ripv, " restart IP valid\n"),
 367                 IF(status.bits.eipv, " error IP valid\n"),
 368                 IF(status.bits.mcip, " machine-check in progress\n"));
 369
 370         /*
 371          * Dump error-reporting registers:
 372          */
 373         mca_dump_error_banks();
 374
 375         /*
 376          * Dump any extended machine state:
 377          */
 378         if (mca_extended_MSRs_present) {
 379                 if (cpu_mode_is64bit())
 380                         mca_dump_64bit_state();
 381                 else
 382                         mca_dump_32bit_state();
 383         }
 384
 385         simple_unlock(&mca_lock);
 386 }