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/slab.h>
  34 #include <linux/delay.h>
  35 #include <linux/irq.h>
  36 #include <linux/random.h>
  37 #include <linux/sysrq.h>
  38 #include <linux/bitops.h>
  39
  40 #include <asm/uaccess.h>
  41 #include <asm/system.h>
  42 #include <asm/io.h>
  43 #include <asm/pgtable.h>
  44 #include <asm/irq.h>
  45 #include <asm/cache.h>
  46 #include <asm/prom.h>
  47 #include <asm/ptrace.h>
  48 #include <asm/machdep.h>
  49 #include <asm/rtas.h>
  50 #include <asm/udbg.h>
  51 #include <asm/firmware.h>
  52
  53 #include "pseries.h"
  54
  55 static unsigned char ras_log_buf[RTAS_ERROR_LOG_MAX];
  56 static DEFINE_SPINLOCK(ras_log_buf_lock);
  57
  58 char mce_data_buf[RTAS_ERROR_LOG_MAX];
  59
  60 static int ras_get_sensor_state_token;
  61 static int ras_check_exception_token;
  62
  63 #define EPOW_SENSOR_TOKEN       9
  64 #define EPOW_SENSOR_INDEX       0
  65 #define RAS_VECTOR_OFFSET       0x500
  66
  67 static irqreturn_t ras_epow_interrupt(int irq, void *dev_id);
  68 static irqreturn_t ras_error_interrupt(int irq, void *dev_id);
  69
  70 /* #define DEBUG */
  71
  72
  73 static void request_ras_irqs(struct device_node *np,
  74                         irq_handler_t handler,
  75                         const char *name)
  76 {
  77         int i, index, count = 0;
  78         struct of_irq oirq;
  79         const u32 *opicprop;
  80         unsigned int opicplen;
  81         unsigned int virqs[16];
  82
  83         /* Check for obsolete "open-pic-interrupt" property. If present, then
  84          * map those interrupts using the default interrupt host and default
  85          * trigger
  86          */
  87         opicprop = of_get_property(np, "open-pic-interrupt", &opicplen);
  88         if (opicprop) {
  89                 opicplen /= sizeof(u32);
  90                 for (i = 0; i < opicplen; i++) {
  91                         if (count > 15)
  92                                 break;
  93                         virqs[count] = irq_create_mapping(NULL, *(opicprop++));
  94                         if (virqs[count] == NO_IRQ)
  95                                 printk(KERN_ERR "Unable to allocate interrupt "
  96                                        "number for %s\n", np->full_name);
  97                         else
  98                                 count++;
  99
 100                 }
 101         }
 102         /* Else use normal interrupt tree parsing */
 103         else {
 104                 /* First try to do a proper OF tree parsing */
 105                 for (index = 0; of_irq_map_one(np, index, &oirq) == 0;
 106                      index++) {
 107                         if (count > 15)
 108                                 break;
 109                         virqs[count] = irq_create_of_mapping(oirq.controller,
 110                                                             oirq.specifier,
 111                                                             oirq.size);
 112                         if (virqs[count] == NO_IRQ)
 113                                 printk(KERN_ERR "Unable to allocate interrupt "
 114                                        "number for %s\n", np->full_name);
 115                         else
 116                                 count++;
 117                 }
 118         }
 119
 120         /* Now request them */
 121         for (i = 0; i < count; i++) {
 122                 if (request_irq(virqs[i], handler, 0, name, NULL)) {
 123                         printk(KERN_ERR "Unable to request interrupt %d for "
 124                                "%s\n", virqs[i], np->full_name);
 125                         return;
 126                 }
 127         }
 128 }
 129
 130 /*
 131  * Initialize handlers for the set of interrupts caused by hardware errors
 132  * and power system events.
 133  */
 134 static int __init init_ras_IRQ(void)
 135 {
 136         struct device_node *np;
 137
 138         ras_get_sensor_state_token = rtas_token("get-sensor-state");
 139         ras_check_exception_token = rtas_token("check-exception");
 140
 141         /* Internal Errors */
 142         np = of_find_node_by_path("/event-sources/internal-errors");
 143         if (np != NULL) {
 144                 request_ras_irqs(np, ras_error_interrupt, "RAS_ERROR");
 145                 of_node_put(np);
 146         }
 147
 148         /* EPOW Events */
 149         np = of_find_node_by_path("/event-sources/epow-events");
 150         if (np != NULL) {
 151                 request_ras_irqs(np, ras_epow_interrupt, "RAS_EPOW");
 152                 of_node_put(np);
 153         }
 154
 155         return 0;
 156 }
 157 __initcall(init_ras_IRQ);
 158
 159 /*
 160  * Handle power subsystem events (EPOW).
 161  *
 162  * Presently we just log the event has occurred.  This should be fixed
 163  * to examine the type of power failure and take appropriate action where
 164  * the time horizon permits something useful to be done.
 165  */
 166 static irqreturn_t ras_epow_interrupt(int irq, void *dev_id)
 167 {
 168         int status = 0xdeadbeef;
 169         int state = 0;
 170         int critical;
 171
 172         status = rtas_call(ras_get_sensor_state_token, 2, 2, &state,
 173                            EPOW_SENSOR_TOKEN, EPOW_SENSOR_INDEX);
 174
 175         if (state > 3)
 176                 critical = 1;  /* Time Critical */
 177         else
 178                 critical = 0;
 179
 180         spin_lock(&ras_log_buf_lock);
 181
 182         status = rtas_call(ras_check_exception_token, 6, 1, NULL,
 183                            RAS_VECTOR_OFFSET,
 184                            irq_map[irq].hwirq,
 185                            RTAS_EPOW_WARNING | RTAS_POWERMGM_EVENTS,
 186                            critical, __pa(&ras_log_buf),
 187                                 rtas_get_error_log_max());
 188
 189         udbg_printf("EPOW <0x%lx 0x%x 0x%x>\n",
 190                     *((unsigned long *)&ras_log_buf), status, state);
 191         printk(KERN_WARNING "EPOW <0x%lx 0x%x 0x%x>\n",
 192                *((unsigned long *)&ras_log_buf), status, state);
 193
 194         /* format and print the extended information */
 195         log_error(ras_log_buf, ERR_TYPE_RTAS_LOG, 0);
 196
 197         spin_unlock(&ras_log_buf_lock);
 198         return IRQ_HANDLED;
 199 }
 200
 201 /*
 202  * Handle hardware error interrupts.
 203  *
 204  * RTAS check-exception is called to collect data on the exception.  If
 205  * the error is deemed recoverable, we log a warning and return.
 206  * For nonrecoverable errors, an error is logged and we stop all processing
 207  * as quickly as possible in order to prevent propagation of the failure.
 208  */
 209 static irqreturn_t ras_error_interrupt(int irq, void *dev_id)
 210 {
 211         struct rtas_error_log *rtas_elog;
 212         int status = 0xdeadbeef;
 213         int fatal;
 214
 215         spin_lock(&ras_log_buf_lock);
 216
 217         status = rtas_call(ras_check_exception_token, 6, 1, NULL,
 218                            RAS_VECTOR_OFFSET,
 219                            irq_map[irq].hwirq,
 220                            RTAS_INTERNAL_ERROR, 1 /*Time Critical */,
 221                            __pa(&ras_log_buf),
 222                                 rtas_get_error_log_max());
 223
 224         rtas_elog = (struct rtas_error_log *)ras_log_buf;
 225
 226         if ((status == 0) && (rtas_elog->severity >= RTAS_SEVERITY_ERROR_SYNC))
 227                 fatal = 1;
 228         else
 229                 fatal = 0;
 230
 231         /* format and print the extended information */
 232         log_error(ras_log_buf, ERR_TYPE_RTAS_LOG, fatal);
 233
 234         if (fatal) {
 235                 udbg_printf("Fatal HW Error <0x%lx 0x%x>\n",
 236                             *((unsigned long *)&ras_log_buf), status);
 237                 printk(KERN_EMERG "Error: Fatal hardware error <0x%lx 0x%x>\n",
 238                        *((unsigned long *)&ras_log_buf), status);
 239
 240 #ifndef DEBUG
 241                 /* Don't actually power off when debugging so we can test
 242                  * without actually failing while injecting errors.
 243                  * Error data will not be logged to syslog.
 244                  */
 245                 ppc_md.power_off();
 246 #endif
 247         } else {
 248                 udbg_printf("Recoverable HW Error <0x%lx 0x%x>\n",
 249                             *((unsigned long *)&ras_log_buf), status);
 250                 printk(KERN_WARNING
 251                        "Warning: Recoverable hardware error <0x%lx 0x%x>\n",
 252                        *((unsigned long *)&ras_log_buf), status);
 253         }
 254
 255         spin_unlock(&ras_log_buf_lock);
 256         return IRQ_HANDLED;
 257 }
 258
 259 /* Get the error information for errors coming through the
 260  * FWNMI vectors.  The pt_regs' r3 will be updated to reflect
 261  * the actual r3 if possible, and a ptr to the error log entry
 262  * will be returned if found.
 263  *
 264  * The mce_data_buf does not have any locks or protection around it,
 265  * if a second machine check comes in, or a system reset is done
 266  * before we have logged the error, then we will get corruption in the
 267  * error log.  This is preferable over holding off on calling
 268  * ibm,nmi-interlock which would result in us checkstopping if a
 269  * second machine check did come in.
 270  */
 271 static struct rtas_error_log *fwnmi_get_errinfo(struct pt_regs *regs)
 272 {
 273         unsigned long errdata = regs->gpr[3];
 274         struct rtas_error_log *errhdr = NULL;
 275         unsigned long *savep;
 276
 277         if ((errdata >= 0x7000 && errdata < 0x7fff0) ||
 278             (errdata >= rtas.base && errdata < rtas.base + rtas.size - 16)) {
 279                 savep = __va(errdata);
 280                 regs->gpr[3] = savep[0];        /* restore original r3 */
 281                 memset(mce_data_buf, 0, RTAS_ERROR_LOG_MAX);
 282                 memcpy(mce_data_buf, (char *)(savep + 1), RTAS_ERROR_LOG_MAX);
 283                 errhdr = (struct rtas_error_log *)mce_data_buf;
 284         } else {
 285                 printk("FWNMI: corrupt r3\n");
 286         }
 287         return errhdr;
 288 }
 289
 290 /* Call this when done with the data returned by FWNMI_get_errinfo.
 291  * It will release the saved data area for other CPUs in the
 292  * partition to receive FWNMI errors.
 293  */
 294 static void fwnmi_release_errinfo(void)
 295 {
 296         int ret = rtas_call(rtas_token("ibm,nmi-interlock"), 0, 1, NULL);
 297         if (ret != 0)
 298                 printk("FWNMI: nmi-interlock failed: %d\n", ret);
 299 }
 300
 301 int pSeries_system_reset_exception(struct pt_regs *regs)
 302 {
 303         if (fwnmi_active) {
 304                 struct rtas_error_log *errhdr = fwnmi_get_errinfo(regs);
 305                 if (errhdr) {
 306                         /* XXX Should look at FWNMI information */
 307                 }
 308                 fwnmi_release_errinfo();
 309         }
 310         return 0; /* need to perform reset */
 311 }
 312
 313 /*
 314  * See if we can recover from a machine check exception.
 315  * This is only called on power4 (or above) and only via
 316  * the Firmware Non-Maskable Interrupts (fwnmi) handler
 317  * which provides the error analysis for us.
 318  *
 319  * Return 1 if corrected (or delivered a signal).
 320  * Return 0 if there is nothing we can do.
 321  */
 322 static int recover_mce(struct pt_regs *regs, struct rtas_error_log * err)
 323 {
 324         int nonfatal = 0;
 325
 326         if (err->disposition == RTAS_DISP_FULLY_RECOVERED) {
 327                 /* Platform corrected itself */
 328                 nonfatal = 1;
 329         } else if ((regs->msr & MSR_RI) &&
 330                    user_mode(regs) &&
 331                    err->severity == RTAS_SEVERITY_ERROR_SYNC &&
 332                    err->disposition == RTAS_DISP_NOT_RECOVERED &&
 333                    err->target == RTAS_TARGET_MEMORY &&
 334                    err->type == RTAS_TYPE_ECC_UNCORR &&
 335                    !(current->pid == 0 || is_global_init(current))) {
 336                 /* Kill off a user process with an ECC error */
 337                 printk(KERN_ERR "MCE: uncorrectable ecc error for pid %d\n",
 338                        current->pid);
 339                 /* XXX something better for ECC error? */
 340                 _exception(SIGBUS, regs, BUS_ADRERR, regs->nip);
 341                 nonfatal = 1;
 342         }
 343
 344         log_error((char *)err, ERR_TYPE_RTAS_LOG, !nonfatal);
 345
 346         return nonfatal;
 347 }
 348
 349 /*
 350  * Handle a machine check.
 351  *
 352  * Note that on Power 4 and beyond Firmware Non-Maskable Interrupts (fwnmi)
 353  * should be present.  If so the handler which called us tells us if the
 354  * error was recovered (never true if RI=0).
 355  *
 356  * On hardware prior to Power 4 these exceptions were asynchronous which
 357  * means we can't tell exactly where it occurred and so we can't recover.
 358  */
 359 int pSeries_machine_check_exception(struct pt_regs *regs)
 360 {
 361         struct rtas_error_log *errp;
 362
 363         if (fwnmi_active) {
 364                 errp = fwnmi_get_errinfo(regs);
 365                 fwnmi_release_errinfo();
 366                 if (errp && recover_mce(regs, errp))
 367                         return 1;
 368         }
 369
 370         return 0;
 371 }