arch/powerpc/platforms/pseries/ras.c

   1 /*
   2  * Copyright (C) 2001 Dave Engebretsen IBM Corporation
   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
  19 /* Change Activity:
  20  * 2001/09/21 : engebret : Created with minimal EPOW and HW exception support.
  21  * End Change Activity
  22  */
  23
  24 #include <linux/errno.h>
  25 #include <linux/threads.h>
  26 #include <linux/kernel_stat.h>
  27 #include <linux/signal.h>
  28 #include <linux/sched.h>
  29 #include <linux/ioport.h>
  30 #include <linux/interrupt.h>
  31 #include <linux/timex.h>
  32 #include <linux/init.h>
  33 #include <linux/delay.h>
  34 #include <linux/irq.h>
  35 #include <linux/random.h>
  36 #include <linux/sysrq.h>
  37 #include <linux/bitops.h>
  38
  39 #include <asm/uaccess.h>
  40 #include <asm/system.h>
  41 #include <asm/io.h>
  42 #include <asm/pgtable.h>
  43 #include <asm/irq.h>
  44 #include <asm/cache.h>
  45 #include <asm/prom.h>
  46 #include <asm/ptrace.h>
  47 #include <asm/machdep.h>
  48 #include <asm/rtas.h>
  49 #include <asm/udbg.h>
  50 #include <asm/firmware.h>
  51
  52 #include "pseries.h"
  53
  54 static unsigned char ras_log_buf[RTAS_ERROR_LOG_MAX];
  55 static DEFINE_SPINLOCK(ras_log_buf_lock);
  56
  57 static char mce_data_buf[RTAS_ERROR_LOG_MAX];
  58
  59 static int ras_get_sensor_state_token;
  60 static int ras_check_exception_token;
  61
  62 #define EPOW_SENSOR_TOKEN       9
  63 #define EPOW_SENSOR_INDEX       0
  64 #define RAS_VECTOR_OFFSET       0x500
  65
  66 static irqreturn_t ras_epow_interrupt(int irq, void *dev_id);
  67 static irqreturn_t ras_error_interrupt(int irq, void *dev_id);
  68
  69
  70 static void request_ras_irqs(struct device_node *np,
  71                         irq_handler_t handler,
  72                         const char *name)
  73 {
  74         int i, index, count = 0;
  75         struct of_irq oirq;
  76         const u32 *opicprop;
  77         unsigned int opicplen;
  78         unsigned int virqs[16];
  79
  80         /* Check for obsolete "open-pic-interrupt" property. If present, then
  81          * map those interrupts using the default interrupt host and default
  82          * trigger
  83          */
  84         opicprop = of_get_property(np, "open-pic-interrupt", &opicplen);
  85         if (opicprop) {
  86                 opicplen /= sizeof(u32);
  87                 for (i = 0; i < opicplen; i++) {
  88                         if (count > 15)
  89                                 break;
  90                         virqs[count] = irq_create_mapping(NULL, *(opicprop++));
  91                         if (virqs[count] == NO_IRQ)
  92                                 printk(KERN_ERR "Unable to allocate interrupt "
  93                                        "number for %s\n", np->full_name);
  94                         else
  95                                 count++;
  96
  97                 }
  98         }
  99         /* Else use normal interrupt tree parsing */
 100         else {
 101                 /* First try to do a proper OF tree parsing */
 102                 for (index = 0; of_irq_map_one(np, index, &oirq) == 0;
 103                      index++) {
 104                         if (count > 15)
 105                                 break;
 106                         virqs[count] = irq_create_of_mapping(oirq.controller,
 107                                                             oirq.specifier,
 108                                                             oirq.size);
 109                         if (virqs[count] == NO_IRQ)
 110                                 printk(KERN_ERR "Unable to allocate interrupt "
 111                                        "number for %s\n", np->full_name);
 112                         else
 113                                 count++;
 114                 }
 115         }
 116
 117         /* Now request them */
 118         for (i = 0; i < count; i++) {
 119                 if (request_irq(virqs[i], handler, 0, name, NULL)) {
 120                         printk(KERN_ERR "Unable to request interrupt %d for "
 121                                "%s\n", virqs[i], np->full_name);
 122                         return;
 123                 }
 124         }
 125 }
 126
 127 /*
 128  * Initialize handlers for the set of interrupts caused by hardware errors
 129  * and power system events.
 130  */
 131 static int __init init_ras_IRQ(void)
 132 {
 133         struct device_node *np;
 134
 135         ras_get_sensor_state_token = rtas_token("get-sensor-state");
 136         ras_check_exception_token = rtas_token("check-exception");
 137
 138         /* Internal Errors */
 139         np = of_find_node_by_path("/event-sources/internal-errors");
 140         if (np != NULL) {
 141                 request_ras_irqs(np, ras_error_interrupt, "RAS_ERROR");
 142                 of_node_put(np);
 143         }
 144
 145         /* EPOW Events */
 146         np = of_find_node_by_path("/event-sources/epow-events");
 147         if (np != NULL) {
 148                 request_ras_irqs(np, ras_epow_interrupt, "RAS_EPOW");
 149                 of_node_put(np);
 150         }
 151
 152         return 0;
 153 }
 154 __initcall(init_ras_IRQ);
 155
 156 /*
 157  * Handle power subsystem events (EPOW).
 158  *
 159  * Presently we just log the event has occurred.  This should be fixed
 160  * to examine the type of power failure and take appropriate action where
 161  * the time horizon permits something useful to be done.
 162  */
 163 static irqreturn_t ras_epow_interrupt(int irq, void *dev_id)
 164 {
 165         int status = 0xdeadbeef;
 166         int state = 0;
 167         int critical;
 168
 169         status = rtas_call(ras_get_sensor_state_token, 2, 2, &state,
 170                            EPOW_SENSOR_TOKEN, EPOW_SENSOR_INDEX);
 171
 172         if (state > 3)
 173                 critical = 1;  /* Time Critical */
 174         else
 175                 critical = 0;
 176
 177         spin_lock(&ras_log_buf_lock);
 178
 179         status = rtas_call(ras_check_exception_token, 6, 1, NULL,
 180                            RAS_VECTOR_OFFSET,
 181                            irq_map[irq].hwirq,
 182                            RTAS_EPOW_WARNING | RTAS_POWERMGM_EVENTS,
 183                            critical, __pa(&ras_log_buf),
 184                                 rtas_get_error_log_max());
 185
 186         udbg_printf("EPOW <0x%lx 0x%x 0x%x>\n",
 187                     *((unsigned long *)&ras_log_buf), status, state);
 188         printk(KERN_WARNING "EPOW <0x%lx 0x%x 0x%x>\n",
 189                *((unsigned long *)&ras_log_buf), status, state);
 190
 191         /* format and print the extended information */
 192         log_error(ras_log_buf, ERR_TYPE_RTAS_LOG, 0);
 193
 194         spin_unlock(&ras_log_buf_lock);
 195         return IRQ_HANDLED;
 196 }
 197
 198 /*
 199  * Handle hardware error interrupts.
 200  *
 201  * RTAS check-exception is called to collect data on the exception.  If
 202  * the error is deemed recoverable, we log a warning and return.
 203  * For nonrecoverable errors, an error is logged and we stop all processing
 204  * as quickly as possible in order to prevent propagation of the failure.
 205  */
 206 static irqreturn_t ras_error_interrupt(int irq, void *dev_id)
 207 {
 208         struct rtas_error_log *rtas_elog;
 209         int status = 0xdeadbeef;
 210         int fatal;
 211
 212         spin_lock(&ras_log_buf_lock);
 213
 214         status = rtas_call(ras_check_exception_token, 6, 1, NULL,
 215                            RAS_VECTOR_OFFSET,
 216                            irq_map[irq].hwirq,
 217                            RTAS_INTERNAL_ERROR, 1 /*Time Critical */,
 218                            __pa(&ras_log_buf),
 219                                 rtas_get_error_log_max());
 220
 221         rtas_elog = (struct rtas_error_log *)ras_log_buf;
 222
 223         if ((status == 0) && (rtas_elog->severity >= RTAS_SEVERITY_ERROR_SYNC))
 224                 fatal = 1;
 225         else
 226                 fatal = 0;
 227
 228         /* format and print the extended information */
 229         log_error(ras_log_buf, ERR_TYPE_RTAS_LOG, fatal);
 230
 231         if (fatal) {
 232                 udbg_printf("Fatal HW Error <0x%lx 0x%x>\n",
 233                             *((unsigned long *)&ras_log_buf), status);
 234                 printk(KERN_EMERG "Error: Fatal hardware error <0x%lx 0x%x>\n",
 235                        *((unsigned long *)&ras_log_buf), status);
 236
 237 #ifndef DEBUG_RTAS_POWER_OFF
 238                 /* Don't actually power off when debugging so we can test
 239                  * without actually failing while injecting errors.
 240                  * Error data will not be logged to syslog.
 241                  */
 242                 ppc_md.power_off();
 243 #endif
 244         } else {
 245                 udbg_printf("Recoverable HW Error <0x%lx 0x%x>\n",
 246                             *((unsigned long *)&ras_log_buf), status);
 247                 printk(KERN_WARNING
 248                        "Warning: Recoverable hardware error <0x%lx 0x%x>\n",
 249                        *((unsigned long *)&ras_log_buf), status);
 250         }
 251
 252         spin_unlock(&ras_log_buf_lock);
 253         return IRQ_HANDLED;
 254 }
 255
 256 /* Get the error information for errors coming through the
 257  * FWNMI vectors.  The pt_regs' r3 will be updated to reflect
 258  * the actual r3 if possible, and a ptr to the error log entry
 259  * will be returned if found.
 260  *
 261  * The mce_data_buf does not have any locks or protection around it,
 262  * if a second machine check comes in, or a system reset is done
 263  * before we have logged the error, then we will get corruption in the
 264  * error log.  This is preferable over holding off on calling
 265  * ibm,nmi-interlock which would result in us checkstopping if a
 266  * second machine check did come in.
 267  */
 268 static struct rtas_error_log *fwnmi_get_errinfo(struct pt_regs *regs)
 269 {
 270         unsigned long errdata = regs->gpr[3];
 271         struct rtas_error_log *errhdr = NULL;
 272         unsigned long *savep;
 273
 274         if ((errdata >= 0x7000 && errdata < 0x7fff0) ||
 275             (errdata >= rtas.base && errdata < rtas.base + rtas.size - 16)) {
 276                 savep = __va(errdata);
 277                 regs->gpr[3] = savep[0];        /* restore original r3 */
 278                 memset(mce_data_buf, 0, RTAS_ERROR_LOG_MAX);
 279                 memcpy(mce_data_buf, (char *)(savep + 1), RTAS_ERROR_LOG_MAX);
 280                 errhdr = (struct rtas_error_log *)mce_data_buf;
 281         } else {
 282                 printk("FWNMI: corrupt r3\n");
 283         }
 284         return errhdr;
 285 }
 286
 287 /* Call this when done with the data returned by FWNMI_get_errinfo.
 288  * It will release the saved data area for other CPUs in the
 289  * partition to receive FWNMI errors.
 290  */
 291 static void fwnmi_release_errinfo(void)
 292 {
 293         int ret = rtas_call(rtas_token("ibm,nmi-interlock"), 0, 1, NULL);
 294         if (ret != 0)
 295                 printk("FWNMI: nmi-interlock failed: %d\n", ret);
 296 }
 297
 298 int pSeries_system_reset_exception(struct pt_regs *regs)
 299 {
 300         if (fwnmi_active) {
 301                 struct rtas_error_log *errhdr = fwnmi_get_errinfo(regs);
 302                 if (errhdr) {
 303                         /* XXX Should look at FWNMI information */
 304                 }
 305                 fwnmi_release_errinfo();
 306         }
 307         return 0; /* need to perform reset */
 308 }
 309
 310 /*
 311  * See if we can recover from a machine check exception.
 312  * This is only called on power4 (or above) and only via
 313  * the Firmware Non-Maskable Interrupts (fwnmi) handler
 314  * which provides the error analysis for us.
 315  *
 316  * Return 1 if corrected (or delivered a signal).
 317  * Return 0 if there is nothing we can do.
 318  */
 319 static int recover_mce(struct pt_regs *regs, struct rtas_error_log * err)
 320 {
 321         int nonfatal = 0;
 322
 323         if (err->disposition == RTAS_DISP_FULLY_RECOVERED) {
 324                 /* Platform corrected itself */
 325                 nonfatal = 1;
 326         } else if ((regs->msr & MSR_RI) &&
 327                    user_mode(regs) &&
 328                    err->severity == RTAS_SEVERITY_ERROR_SYNC &&
 329                    err->disposition == RTAS_DISP_NOT_RECOVERED &&
 330                    err->target == RTAS_TARGET_MEMORY &&
 331                    err->type == RTAS_TYPE_ECC_UNCORR &&
 332                    !(current->pid == 0 || is_global_init(current))) {
 333                 /* Kill off a user process with an ECC error */
 334                 printk(KERN_ERR "MCE: uncorrectable ecc error for pid %d\n",
 335                        current->pid);
 336                 /* XXX something better for ECC error? */
 337                 _exception(SIGBUS, regs, BUS_ADRERR, regs->nip);
 338                 nonfatal = 1;
 339         }
 340
 341         log_error((char *)err, ERR_TYPE_RTAS_LOG, !nonfatal);
 342
 343         return nonfatal;
 344 }
 345
 346 /*
 347  * Handle a machine check.
 348  *
 349  * Note that on Power 4 and beyond Firmware Non-Maskable Interrupts (fwnmi)
 350  * should be present.  If so the handler which called us tells us if the
 351  * error was recovered (never true if RI=0).
 352  *
 353  * On hardware prior to Power 4 these exceptions were asynchronous which
 354  * means we can't tell exactly where it occurred and so we can't recover.
 355  */
 356 int pSeries_machine_check_exception(struct pt_regs *regs)
 357 {
 358         struct rtas_error_log *errp;
 359
 360         if (fwnmi_active) {
 361                 errp = fwnmi_get_errinfo(regs);
 362                 fwnmi_release_errinfo();
 363                 if (errp && recover_mce(regs, errp))
 364                         return 1;
 365         }
 366
 367         return 0;
 368 }