arch/powerpc/kernel/mce.c

   1 /*
   2  * Machine check exception handling.
   3  *
   4  * This program is free software; you can redistribute it and/or modify
   5  * it under the terms of the GNU General Public License as published by
   6  * the Free Software Foundation; either version 2 of the License, or
   7  * (at your option) any later version.
   8  *
   9  * This program is distributed in the hope that it will be useful,
  10  * but WITHOUT ANY WARRANTY; without even the implied warranty of
  11  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
  12  * GNU General Public License for more details.
  13  *
  14  * You should have received a copy of the GNU General Public License
  15  * along with this program; if not, write to the Free Software
  16  * Foundation, Inc., 59 Temple Place - Suite 330, Boston, MA 02111-1307, USA.
  17  *
  18  * Copyright 2013 IBM Corporation
  19  * Author: Mahesh Salgaonkar <mahesh@linux.vnet.ibm.com>
  20  */
  21
  22 #undef DEBUG
  23 #define pr_fmt(fmt) "mce: " fmt
  24
  25 #include <linux/types.h>
  26 #include <linux/ptrace.h>
  27 #include <linux/percpu.h>
  28 #include <linux/export.h>
  29 #include <linux/irq_work.h>
  30 #include <asm/mce.h>
  31
  32 static DEFINE_PER_CPU(int, mce_nest_count);
  33 static DEFINE_PER_CPU(struct machine_check_event[MAX_MC_EVT], mce_event);
  34
  35 /* Queue for delayed MCE events. */
  36 static DEFINE_PER_CPU(int, mce_queue_count);
  37 static DEFINE_PER_CPU(struct machine_check_event[MAX_MC_EVT], mce_event_queue);
  38
  39 static void machine_check_process_queued_event(struct irq_work *work);
  40 struct irq_work mce_event_process_work = {
  41         .func = machine_check_process_queued_event,
  42 };
  43
  44 static void mce_set_error_info(struct machine_check_event *mce,
  45                                struct mce_error_info *mce_err)
  46 {
  47         mce->error_type = mce_err->error_type;
  48         switch (mce_err->error_type) {
  49         case MCE_ERROR_TYPE_UE:
  50                 mce->u.ue_error.ue_error_type = mce_err->u.ue_error_type;
  51                 break;
  52         case MCE_ERROR_TYPE_SLB:
  53                 mce->u.slb_error.slb_error_type = mce_err->u.slb_error_type;
  54                 break;
  55         case MCE_ERROR_TYPE_ERAT:
  56                 mce->u.erat_error.erat_error_type = mce_err->u.erat_error_type;
  57                 break;
  58         case MCE_ERROR_TYPE_TLB:
  59                 mce->u.tlb_error.tlb_error_type = mce_err->u.tlb_error_type;
  60                 break;
  61         case MCE_ERROR_TYPE_UNKNOWN:
  62         default:
  63                 break;
  64         }
  65 }
  66
  67 /*
  68  * Decode and save high level MCE information into per cpu buffer which
  69  * is an array of machine_check_event structure.
  70  */
  71 void save_mce_event(struct pt_regs *regs, long handled,
  72                     struct mce_error_info *mce_err,
  73                     uint64_t addr)
  74 {
  75         uint64_t srr1;
  76         int index = __get_cpu_var(mce_nest_count)++;
  77         struct machine_check_event *mce = &__get_cpu_var(mce_event[index]);
  78
  79         /*
  80          * Return if we don't have enough space to log mce event.
  81          * mce_nest_count may go beyond MAX_MC_EVT but that's ok,
  82          * the check below will stop buffer overrun.
  83          */
  84         if (index >= MAX_MC_EVT)
  85                 return;
  86
  87         /* Populate generic machine check info */
  88         mce->version = MCE_V1;
  89         mce->srr0 = regs->nip;
  90         mce->srr1 = regs->msr;
  91         mce->gpr3 = regs->gpr[3];
  92         mce->in_use = 1;
  93
  94         mce->initiator = MCE_INITIATOR_CPU;
  95         if (handled)
  96                 mce->disposition = MCE_DISPOSITION_RECOVERED;
  97         else
  98                 mce->disposition = MCE_DISPOSITION_NOT_RECOVERED;
  99         mce->severity = MCE_SEV_ERROR_SYNC;
 100
 101         srr1 = regs->msr;
 102
 103         /*
 104          * Populate the mce error_type and type-specific error_type.
 105          */
 106         mce_set_error_info(mce, mce_err);
 107
 108         if (!addr)
 109                 return;
 110
 111         if (mce->error_type == MCE_ERROR_TYPE_TLB) {
 112                 mce->u.tlb_error.effective_address_provided = true;
 113                 mce->u.tlb_error.effective_address = addr;
 114         } else if (mce->error_type == MCE_ERROR_TYPE_SLB) {
 115                 mce->u.slb_error.effective_address_provided = true;
 116                 mce->u.slb_error.effective_address = addr;
 117         } else if (mce->error_type == MCE_ERROR_TYPE_ERAT) {
 118                 mce->u.erat_error.effective_address_provided = true;
 119                 mce->u.erat_error.effective_address = addr;
 120         } else if (mce->error_type == MCE_ERROR_TYPE_UE) {
 121                 mce->u.ue_error.effective_address_provided = true;
 122                 mce->u.ue_error.effective_address = addr;
 123         }
 124         return;
 125 }
 126
 127 /*
 128  * get_mce_event:
 129  *      mce     Pointer to machine_check_event structure to be filled.
 130  *      release Flag to indicate whether to free the event slot or not.
 131  *              0 <= do not release the mce event. Caller will invoke
 132  *                   release_mce_event() once event has been consumed.
 133  *              1 <= release the slot.
 134  *
 135  *      return  1 = success
 136  *              0 = failure
 137  *
 138  * get_mce_event() will be called by platform specific machine check
 139  * handle routine and in KVM.
 140  * When we call get_mce_event(), we are still in interrupt context and
 141  * preemption will not be scheduled until ret_from_expect() routine
 142  * is called.
 143  */
 144 int get_mce_event(struct machine_check_event *mce, bool release)
 145 {
 146         int index = __get_cpu_var(mce_nest_count) - 1;
 147         struct machine_check_event *mc_evt;
 148         int ret = 0;
 149
 150         /* Sanity check */
 151         if (index < 0)
 152                 return ret;
 153
 154         /* Check if we have MCE info to process. */
 155         if (index < MAX_MC_EVT) {
 156                 mc_evt = &__get_cpu_var(mce_event[index]);
 157                 /* Copy the event structure and release the original */
 158                 if (mce)
 159                         *mce = *mc_evt;
 160                 if (release)
 161                         mc_evt->in_use = 0;
 162                 ret = 1;
 163         }
 164         /* Decrement the count to free the slot. */
 165         if (release)
 166                 __get_cpu_var(mce_nest_count)--;
 167
 168         return ret;
 169 }
 170
 171 void release_mce_event(void)
 172 {
 173         get_mce_event(NULL, true);
 174 }
 175
 176 /*
 177  * Queue up the MCE event which then can be handled later.
 178  */
 179 void machine_check_queue_event(void)
 180 {
 181         int index;
 182         struct machine_check_event evt;
 183
 184         if (!get_mce_event(&evt, MCE_EVENT_RELEASE))
 185                 return;
 186
 187         index = __get_cpu_var(mce_queue_count)++;
 188         /* If queue is full, just return for now. */
 189         if (index >= MAX_MC_EVT) {
 190                 __get_cpu_var(mce_queue_count)--;
 191                 return;
 192         }
 193         __get_cpu_var(mce_event_queue[index]) = evt;
 194
 195         /* Queue irq work to process this event later. */
 196         irq_work_queue(&mce_event_process_work);
 197 }
 198
 199 /*
 200  * process pending MCE event from the mce event queue. This function will be
 201  * called during syscall exit.
 202  */
 203 static void machine_check_process_queued_event(struct irq_work *work)
 204 {
 205         int index;
 206
 207         /*
 208          * For now just print it to console.
 209          * TODO: log this error event to FSP or nvram.
 210          */
 211         while (__get_cpu_var(mce_queue_count) > 0) {
 212                 index = __get_cpu_var(mce_queue_count) - 1;
 213                 machine_check_print_event_info(
 214                                 &__get_cpu_var(mce_event_queue[index]));
 215                 __get_cpu_var(mce_queue_count)--;
 216         }
 217 }
 218
 219 void machine_check_print_event_info(struct machine_check_event *evt)
 220 {
 221         const char *level, *sevstr, *subtype;
 222         static const char *mc_ue_types[] = {
 223                 "Indeterminate",
 224                 "Instruction fetch",
 225                 "Page table walk ifetch",
 226                 "Load/Store",
 227                 "Page table walk Load/Store",
 228         };
 229         static const char *mc_slb_types[] = {
 230                 "Indeterminate",
 231                 "Parity",
 232                 "Multihit",
 233         };
 234         static const char *mc_erat_types[] = {
 235                 "Indeterminate",
 236                 "Parity",
 237                 "Multihit",
 238         };
 239         static const char *mc_tlb_types[] = {
 240                 "Indeterminate",
 241                 "Parity",
 242                 "Multihit",
 243         };
 244
 245         /* Print things out */
 246         if (evt->version != MCE_V1) {
 247                 pr_err("Machine Check Exception, Unknown event version %d !\n",
 248                        evt->version);
 249                 return;
 250         }
 251         switch (evt->severity) {
 252         case MCE_SEV_NO_ERROR:
 253                 level = KERN_INFO;
 254                 sevstr = "Harmless";
 255                 break;
 256         case MCE_SEV_WARNING:
 257                 level = KERN_WARNING;
 258                 sevstr = "";
 259                 break;
 260         case MCE_SEV_ERROR_SYNC:
 261                 level = KERN_ERR;
 262                 sevstr = "Severe";
 263                 break;
 264         case MCE_SEV_FATAL:
 265         default:
 266                 level = KERN_ERR;
 267                 sevstr = "Fatal";
 268                 break;
 269         }
 270
 271         printk("%s%s Machine check interrupt [%s]\n", level, sevstr,
 272                evt->disposition == MCE_DISPOSITION_RECOVERED ?
 273                "Recovered" : "[Not recovered");
 274         printk("%s  Initiator: %s\n", level,
 275                evt->initiator == MCE_INITIATOR_CPU ? "CPU" : "Unknown");
 276         switch (evt->error_type) {
 277         case MCE_ERROR_TYPE_UE:
 278                 subtype = evt->u.ue_error.ue_error_type <
 279                         ARRAY_SIZE(mc_ue_types) ?
 280                         mc_ue_types[evt->u.ue_error.ue_error_type]
 281                         : "Unknown";
 282                 printk("%s  Error type: UE [%s]\n", level, subtype);
 283                 if (evt->u.ue_error.effective_address_provided)
 284                         printk("%s    Effective address: %016llx\n",
 285                                level, evt->u.ue_error.effective_address);
 286                 if (evt->u.ue_error.physical_address_provided)
 287                         printk("%s      Physial address: %016llx\n",
 288                                level, evt->u.ue_error.physical_address);
 289                 break;
 290         case MCE_ERROR_TYPE_SLB:
 291                 subtype = evt->u.slb_error.slb_error_type <
 292                         ARRAY_SIZE(mc_slb_types) ?
 293                         mc_slb_types[evt->u.slb_error.slb_error_type]
 294                         : "Unknown";
 295                 printk("%s  Error type: SLB [%s]\n", level, subtype);
 296                 if (evt->u.slb_error.effective_address_provided)
 297                         printk("%s    Effective address: %016llx\n",
 298                                level, evt->u.slb_error.effective_address);
 299                 break;
 300         case MCE_ERROR_TYPE_ERAT:
 301                 subtype = evt->u.erat_error.erat_error_type <
 302                         ARRAY_SIZE(mc_erat_types) ?
 303                         mc_erat_types[evt->u.erat_error.erat_error_type]
 304                         : "Unknown";
 305                 printk("%s  Error type: ERAT [%s]\n", level, subtype);
 306                 if (evt->u.erat_error.effective_address_provided)
 307                         printk("%s    Effective address: %016llx\n",
 308                                level, evt->u.erat_error.effective_address);
 309                 break;
 310         case MCE_ERROR_TYPE_TLB:
 311                 subtype = evt->u.tlb_error.tlb_error_type <
 312                         ARRAY_SIZE(mc_tlb_types) ?
 313                         mc_tlb_types[evt->u.tlb_error.tlb_error_type]
 314                         : "Unknown";
 315                 printk("%s  Error type: TLB [%s]\n", level, subtype);
 316                 if (evt->u.tlb_error.effective_address_provided)
 317                         printk("%s    Effective address: %016llx\n",
 318                                level, evt->u.tlb_error.effective_address);
 319                 break;
 320         default:
 321         case MCE_ERROR_TYPE_UNKNOWN:
 322                 printk("%s  Error type: Unknown\n", level);
 323                 break;
 324         }
 325 }
 326
 327 uint64_t get_mce_fault_addr(struct machine_check_event *evt)
 328 {
 329         switch (evt->error_type) {
 330         case MCE_ERROR_TYPE_UE:
 331                 if (evt->u.ue_error.effective_address_provided)
 332                         return evt->u.ue_error.effective_address;
 333                 break;
 334         case MCE_ERROR_TYPE_SLB:
 335                 if (evt->u.slb_error.effective_address_provided)
 336                         return evt->u.slb_error.effective_address;
 337                 break;
 338         case MCE_ERROR_TYPE_ERAT:
 339                 if (evt->u.erat_error.effective_address_provided)
 340                         return evt->u.erat_error.effective_address;
 341                 break;
 342         case MCE_ERROR_TYPE_TLB:
 343                 if (evt->u.tlb_error.effective_address_provided)
 344                         return evt->u.tlb_error.effective_address;
 345                 break;
 346         default:
 347         case MCE_ERROR_TYPE_UNKNOWN:
 348                 break;
 349         }
 350         return 0;
 351 }
 352 EXPORT_SYMBOL(get_mce_fault_addr);