arch/ppc64/kernel/ras.c

   1 /*
   2  * ras.c
   3  * Copyright (C) 2001 Dave Engebretsen IBM Corporation
   4  *
   5  * This program is free software; you can redistribute it and/or modify
   6  * it under the terms of the GNU General Public License as published by
   7  * the Free Software Foundation; either version 2 of the License, or
   8  * (at your option) any later version.
   9  *
  10  * This program is distributed in the hope that it will be useful,
  11  * but WITHOUT ANY WARRANTY; without even the implied warranty of
  12  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
  13  * GNU General Public License for more details.
  14  *
  15  * You should have received a copy of the GNU General Public License
  16  * along with this program; if not, write to the Free Software
  17  * Foundation, Inc., 59 Temple Place, Suite 330, Boston, MA  02111-1307 USA
  18  */
  19
  20 /* Change Activity:
  21  * 2001/09/21 : engebret : Created with minimal EPOW and HW exception support.
  22  * End Change Activity
  23  */
  24
  25 #include <linux/errno.h>
  26 #include <linux/threads.h>
  27 #include <linux/kernel_stat.h>
  28 #include <linux/signal.h>
  29 #include <linux/sched.h>
  30 #include <linux/ioport.h>
  31 #include <linux/interrupt.h>
  32 #include <linux/timex.h>
  33 #include <linux/init.h>
  34 #include <linux/slab.h>
  35 #include <linux/pci.h>
  36 #include <linux/delay.h>
  37 #include <linux/irq.h>
  38 #include <linux/random.h>
  39 #include <linux/sysrq.h>
  40 #include <linux/bitops.h>
  41
  42 #include <asm/uaccess.h>
  43 #include <asm/system.h>
  44 #include <asm/io.h>
  45 #include <asm/pgtable.h>
  46 #include <asm/irq.h>
  47 #include <asm/cache.h>
  48 #include <asm/prom.h>
  49 #include <asm/ptrace.h>
  50 #include <asm/machdep.h>
  51 #include <asm/rtas.h>
  52 #include <asm/ppcdebug.h>
  53
  54 static unsigned char ras_log_buf[RTAS_ERROR_LOG_MAX];
  55 static DEFINE_SPINLOCK(ras_log_buf_lock);
  56
  57 char mce_data_buf[RTAS_ERROR_LOG_MAX]
  58 ;
  59 /* This is true if we are using the firmware NMI handler (typically LPAR) */
  60 extern int fwnmi_active;
  61
  62 extern void _exception(int signr, struct pt_regs *regs, int code, unsigned long addr);
  63
  64 static int ras_get_sensor_state_token;
  65 static int ras_check_exception_token;
  66
  67 #define EPOW_SENSOR_TOKEN       9
  68 #define EPOW_SENSOR_INDEX       0
  69 #define RAS_VECTOR_OFFSET       0x500
  70
  71 static irqreturn_t ras_epow_interrupt(int irq, void *dev_id,
  72                                         struct pt_regs * regs);
  73 static irqreturn_t ras_error_interrupt(int irq, void *dev_id,
  74                                         struct pt_regs * regs);
  75
  76 /* #define DEBUG */
  77
  78 static void request_ras_irqs(struct device_node *np, char *propname,
  79                         irqreturn_t (*handler)(int, void *, struct pt_regs *),
  80                         const char *name)
  81 {
  82         unsigned int *ireg, len, i;
  83         int virq, n_intr;
  84
  85         ireg = (unsigned int *)get_property(np, propname, &len);
  86         if (ireg == NULL)
  87                 return;
  88         n_intr = prom_n_intr_cells(np);
  89         len /= n_intr * sizeof(*ireg);
  90
  91         for (i = 0; i < len; i++) {
  92                 virq = virt_irq_create_mapping(*ireg);
  93                 if (virq == NO_IRQ) {
  94                         printk(KERN_ERR "Unable to allocate interrupt "
  95                                "number for %s\n", np->full_name);
  96                         return;
  97                 }
  98                 if (request_irq(irq_offset_up(virq), handler, 0, name, NULL)) {
  99                         printk(KERN_ERR "Unable to request interrupt %d for "
 100                                "%s\n", irq_offset_up(virq), np->full_name);
 101                         return;
 102                 }
 103                 ireg += n_intr;
 104         }
 105 }
 106
 107 /*
 108  * Initialize handlers for the set of interrupts caused by hardware errors
 109  * and power system events.
 110  */
 111 static int __init init_ras_IRQ(void)
 112 {
 113         struct device_node *np;
 114
 115         ras_get_sensor_state_token = rtas_token("get-sensor-state");
 116         ras_check_exception_token = rtas_token("check-exception");
 117
 118         /* Internal Errors */
 119         np = of_find_node_by_path("/event-sources/internal-errors");
 120         if (np != NULL) {
 121                 request_ras_irqs(np, "open-pic-interrupt", ras_error_interrupt,
 122                                  "RAS_ERROR");
 123                 request_ras_irqs(np, "interrupts", ras_error_interrupt,
 124                                  "RAS_ERROR");
 125                 of_node_put(np);
 126         }
 127
 128         /* EPOW Events */
 129         np = of_find_node_by_path("/event-sources/epow-events");
 130         if (np != NULL) {
 131                 request_ras_irqs(np, "open-pic-interrupt", ras_epow_interrupt,
 132                                  "RAS_EPOW");
 133                 request_ras_irqs(np, "interrupts", ras_epow_interrupt,
 134                                  "RAS_EPOW");
 135                 of_node_put(np);
 136         }
 137
 138         return 1;
 139 }
 140 __initcall(init_ras_IRQ);
 141
 142 /*
 143  * Handle power subsystem events (EPOW).
 144  *
 145  * Presently we just log the event has occurred.  This should be fixed
 146  * to examine the type of power failure and take appropriate action where
 147  * the time horizon permits something useful to be done.
 148  */
 149 static irqreturn_t
 150 ras_epow_interrupt(int irq, void *dev_id, struct pt_regs * regs)
 151 {
 152         int status = 0xdeadbeef;
 153         int state = 0;
 154         int critical;
 155
 156         status = rtas_call(ras_get_sensor_state_token, 2, 2, &state,
 157                            EPOW_SENSOR_TOKEN, EPOW_SENSOR_INDEX);
 158
 159         if (state > 3)
 160                 critical = 1;  /* Time Critical */
 161         else
 162                 critical = 0;
 163
 164         spin_lock(&ras_log_buf_lock);
 165
 166         status = rtas_call(ras_check_exception_token, 6, 1, NULL,
 167                            RAS_VECTOR_OFFSET,
 168                            virt_irq_to_real(irq_offset_down(irq)),
 169                            RTAS_EPOW_WARNING | RTAS_POWERMGM_EVENTS,
 170                            critical, __pa(&ras_log_buf),
 171                                 rtas_get_error_log_max());
 172
 173         udbg_printf("EPOW <0x%lx 0x%x 0x%x>\n",
 174                     *((unsigned long *)&ras_log_buf), status, state);
 175         printk(KERN_WARNING "EPOW <0x%lx 0x%x 0x%x>\n",
 176                *((unsigned long *)&ras_log_buf), status, state);
 177
 178         /* format and print the extended information */
 179         log_error(ras_log_buf, ERR_TYPE_RTAS_LOG, 0);
 180
 181         spin_unlock(&ras_log_buf_lock);
 182         return IRQ_HANDLED;
 183 }
 184
 185 /*
 186  * Handle hardware error interrupts.
 187  *
 188  * RTAS check-exception is called to collect data on the exception.  If
 189  * the error is deemed recoverable, we log a warning and return.
 190  * For nonrecoverable errors, an error is logged and we stop all processing
 191  * as quickly as possible in order to prevent propagation of the failure.
 192  */
 193 static irqreturn_t
 194 ras_error_interrupt(int irq, void *dev_id, struct pt_regs * regs)
 195 {
 196         struct rtas_error_log *rtas_elog;
 197         int status = 0xdeadbeef;
 198         int fatal;
 199
 200         spin_lock(&ras_log_buf_lock);
 201
 202         status = rtas_call(ras_check_exception_token, 6, 1, NULL,
 203                            RAS_VECTOR_OFFSET,
 204                            virt_irq_to_real(irq_offset_down(irq)),
 205                            RTAS_INTERNAL_ERROR, 1 /*Time Critical */,
 206                            __pa(&ras_log_buf),
 207                                 rtas_get_error_log_max());
 208
 209         rtas_elog = (struct rtas_error_log *)ras_log_buf;
 210
 211         if ((status == 0) && (rtas_elog->severity >= RTAS_SEVERITY_ERROR_SYNC))
 212                 fatal = 1;
 213         else
 214                 fatal = 0;
 215
 216         /* format and print the extended information */
 217         log_error(ras_log_buf, ERR_TYPE_RTAS_LOG, fatal);
 218
 219         if (fatal) {
 220                 udbg_printf("Fatal HW Error <0x%lx 0x%x>\n",
 221                             *((unsigned long *)&ras_log_buf), status);
 222                 printk(KERN_EMERG "Error: Fatal hardware error <0x%lx 0x%x>\n",
 223                        *((unsigned long *)&ras_log_buf), status);
 224
 225 #ifndef DEBUG
 226                 /* Don't actually power off when debugging so we can test
 227                  * without actually failing while injecting errors.
 228                  * Error data will not be logged to syslog.
 229                  */
 230                 ppc_md.power_off();
 231 #endif
 232         } else {
 233                 udbg_printf("Recoverable HW Error <0x%lx 0x%x>\n",
 234                             *((unsigned long *)&ras_log_buf), status);
 235                 printk(KERN_WARNING
 236                        "Warning: Recoverable hardware error <0x%lx 0x%x>\n",
 237                        *((unsigned long *)&ras_log_buf), status);
 238         }
 239
 240         spin_unlock(&ras_log_buf_lock);
 241         return IRQ_HANDLED;
 242 }
 243
 244 /* Get the error information for errors coming through the
 245  * FWNMI vectors.  The pt_regs' r3 will be updated to reflect
 246  * the actual r3 if possible, and a ptr to the error log entry
 247  * will be returned if found.
 248  *
 249  * The mce_data_buf does not have any locks or protection around it,
 250  * if a second machine check comes in, or a system reset is done
 251  * before we have logged the error, then we will get corruption in the
 252  * error log.  This is preferable over holding off on calling
 253  * ibm,nmi-interlock which would result in us checkstopping if a
 254  * second machine check did come in.
 255  */
 256 static struct rtas_error_log *fwnmi_get_errinfo(struct pt_regs *regs)
 257 {
 258         unsigned long errdata = regs->gpr[3];
 259         struct rtas_error_log *errhdr = NULL;
 260         unsigned long *savep;
 261
 262         if ((errdata >= 0x7000 && errdata < 0x7fff0) ||
 263             (errdata >= rtas.base && errdata < rtas.base + rtas.size - 16)) {
 264                 savep = __va(errdata);
 265                 regs->gpr[3] = savep[0];        /* restore original r3 */
 266                 memset(mce_data_buf, 0, RTAS_ERROR_LOG_MAX);
 267                 memcpy(mce_data_buf, (char *)(savep + 1), RTAS_ERROR_LOG_MAX);
 268                 errhdr = (struct rtas_error_log *)mce_data_buf;
 269         } else {
 270                 printk("FWNMI: corrupt r3\n");
 271         }
 272         return errhdr;
 273 }
 274
 275 /* Call this when done with the data returned by FWNMI_get_errinfo.
 276  * It will release the saved data area for other CPUs in the
 277  * partition to receive FWNMI errors.
 278  */
 279 static void fwnmi_release_errinfo(void)
 280 {
 281         int ret = rtas_call(rtas_token("ibm,nmi-interlock"), 0, 1, NULL);
 282         if (ret != 0)
 283                 printk("FWNMI: nmi-interlock failed: %d\n", ret);
 284 }
 285
 286 void pSeries_system_reset_exception(struct pt_regs *regs)
 287 {
 288         if (fwnmi_active) {
 289                 struct rtas_error_log *errhdr = fwnmi_get_errinfo(regs);
 290                 if (errhdr) {
 291                         /* XXX Should look at FWNMI information */
 292                 }
 293                 fwnmi_release_errinfo();
 294         }
 295 }
 296
 297 /*
 298  * See if we can recover from a machine check exception.
 299  * This is only called on power4 (or above) and only via
 300  * the Firmware Non-Maskable Interrupts (fwnmi) handler
 301  * which provides the error analysis for us.
 302  *
 303  * Return 1 if corrected (or delivered a signal).
 304  * Return 0 if there is nothing we can do.
 305  */
 306 static int recover_mce(struct pt_regs *regs, struct rtas_error_log * err)
 307 {
 308         int nonfatal = 0;
 309
 310         if (err->disposition == RTAS_DISP_FULLY_RECOVERED) {
 311                 /* Platform corrected itself */
 312                 nonfatal = 1;
 313         } else if ((regs->msr & MSR_RI) &&
 314                    user_mode(regs) &&
 315                    err->severity == RTAS_SEVERITY_ERROR_SYNC &&
 316                    err->disposition == RTAS_DISP_NOT_RECOVERED &&
 317                    err->target == RTAS_TARGET_MEMORY &&
 318                    err->type == RTAS_TYPE_ECC_UNCORR &&
 319                    !(current->pid == 0 || current->pid == 1)) {
 320                 /* Kill off a user process with an ECC error */
 321                 printk(KERN_ERR "MCE: uncorrectable ecc error for pid %d\n",
 322                        current->pid);
 323                 /* XXX something better for ECC error? */
 324                 _exception(SIGBUS, regs, BUS_ADRERR, regs->nip);
 325                 nonfatal = 1;
 326         }
 327
 328         log_error((char *)err, ERR_TYPE_RTAS_LOG, !nonfatal);
 329
 330         return nonfatal;
 331 }
 332
 333 /*
 334  * Handle a machine check.
 335  *
 336  * Note that on Power 4 and beyond Firmware Non-Maskable Interrupts (fwnmi)
 337  * should be present.  If so the handler which called us tells us if the
 338  * error was recovered (never true if RI=0).
 339  *
 340  * On hardware prior to Power 4 these exceptions were asynchronous which
 341  * means we can't tell exactly where it occurred and so we can't recover.
 342  */
 343 int pSeries_machine_check_exception(struct pt_regs *regs)
 344 {
 345         struct rtas_error_log *errp;
 346
 347         if (fwnmi_active) {
 348                 errp = fwnmi_get_errinfo(regs);
 349                 fwnmi_release_errinfo();
 350                 if (errp && recover_mce(regs, errp))
 351                         return 1;
 352         }
 353
 354         return 0;
 355 }