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