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/pci.h>
  35 #include <linux/delay.h>
  36 #include <linux/irq.h>
  37 #include <linux/random.h>
  38 #include <linux/sysrq.h>
  39 #include <linux/bitops.h>
  40
  41 #include <asm/uaccess.h>
  42 #include <asm/system.h>
  43 #include <asm/io.h>
  44 #include <asm/pgtable.h>
  45 #include <asm/irq.h>
  46 #include <asm/cache.h>
  47 #include <asm/prom.h>
  48 #include <asm/ptrace.h>
  49 #include <asm/machdep.h>
  50 #include <asm/rtas.h>
  51 #include <asm/udbg.h>
  52 #include <asm/firmware.h>
  53
  54 #include "ras.h"
  55
  56 static unsigned char ras_log_buf[RTAS_ERROR_LOG_MAX];
  57 static DEFINE_SPINLOCK(ras_log_buf_lock);
  58
  59 char mce_data_buf[RTAS_ERROR_LOG_MAX];
  60
  61 static int ras_get_sensor_state_token;
  62 static int ras_check_exception_token;
  63
  64 #define EPOW_SENSOR_TOKEN       9
  65 #define EPOW_SENSOR_INDEX       0
  66 #define RAS_VECTOR_OFFSET       0x500
  67
  68 static irqreturn_t ras_epow_interrupt(int irq, void *dev_id);
  69 static irqreturn_t ras_error_interrupt(int irq, void *dev_id);
  70
  71 /* #define DEBUG */
  72
  73
  74 static void request_ras_irqs(struct device_node *np,
  75                         irq_handler_t handler,
  76                         const char *name)
  77 {
  78         int i, index, count = 0;
  79         struct of_irq oirq;
  80         const u32 *opicprop;
  81         unsigned int opicplen;
  82         unsigned int virqs[16];
  83
  84         /* Check for obsolete "open-pic-interrupt" property. If present, then
  85          * map those interrupts using the default interrupt host and default
  86          * trigger
  87          */
  88         opicprop = get_property(np, "open-pic-interrupt", &opicplen);
  89         if (opicprop) {
  90                 opicplen /= sizeof(u32);
  91                 for (i = 0; i < opicplen; i++) {
  92                         if (count > 15)
  93                                 break;
  94                         virqs[count] = irq_create_mapping(NULL, *(opicprop++));
  95                         if (virqs[count] == NO_IRQ)
  96                                 printk(KERN_ERR "Unable to allocate interrupt "
  97                                        "number for %s\n", np->full_name);
  98                         else
  99                                 count++;
 100
 101                 }
 102         }
 103         /* Else use normal interrupt tree parsing */
 104         else {
 105                 /* First try to do a proper OF tree parsing */
 106                 for (index = 0; of_irq_map_one(np, index, &oirq) == 0;
 107                      index++) {
 108                         if (count > 15)
 109                                 break;
 110                         virqs[count] = irq_create_of_mapping(oirq.controller,
 111                                                             oirq.specifier,
 112                                                             oirq.size);
 113                         if (virqs[count] == NO_IRQ)
 114                                 printk(KERN_ERR "Unable to allocate interrupt "
 115                                        "number for %s\n", np->full_name);
 116                         else
 117                                 count++;
 118                 }
 119         }
 120
 121         /* Now request them */
 122         for (i = 0; i < count; i++) {
 123                 if (request_irq(virqs[i], handler, 0, name, NULL)) {
 124                         printk(KERN_ERR "Unable to request interrupt %d for "
 125                                "%s\n", virqs[i], np->full_name);
 126                         return;
 127                 }
 128         }
 129 }
 130
 131 /*
 132  * Initialize handlers for the set of interrupts caused by hardware errors
 133  * and power system events.
 134  */
 135 static int __init init_ras_IRQ(void)
 136 {
 137         struct device_node *np;
 138
 139         ras_get_sensor_state_token = rtas_token("get-sensor-state");
 140         ras_check_exception_token = rtas_token("check-exception");
 141
 142         /* Internal Errors */
 143         np = of_find_node_by_path("/event-sources/internal-errors");
 144         if (np != NULL) {
 145                 request_ras_irqs(np, ras_error_interrupt, "RAS_ERROR");
 146                 of_node_put(np);
 147         }
 148
 149         /* EPOW Events */
 150         np = of_find_node_by_path("/event-sources/epow-events");
 151         if (np != NULL) {
 152                 request_ras_irqs(np, ras_epow_interrupt, "RAS_EPOW");
 153                 of_node_put(np);
 154         }
 155
 156         return 0;
 157 }
 158 __initcall(init_ras_IRQ);
 159
 160 /*
 161  * Handle power subsystem events (EPOW).
 162  *
 163  * Presently we just log the event has occurred.  This should be fixed
 164  * to examine the type of power failure and take appropriate action where
 165  * the time horizon permits something useful to be done.
 166  */
 167 static irqreturn_t ras_epow_interrupt(int irq, void *dev_id)
 168 {
 169         int status = 0xdeadbeef;
 170         int state = 0;
 171         int critical;
 172
 173         status = rtas_call(ras_get_sensor_state_token, 2, 2, &state,
 174                            EPOW_SENSOR_TOKEN, EPOW_SENSOR_INDEX);
 175
 176         if (state > 3)
 177                 critical = 1;  /* Time Critical */
 178         else
 179                 critical = 0;
 180
 181         spin_lock(&ras_log_buf_lock);
 182
 183         status = rtas_call(ras_check_exception_token, 6, 1, NULL,
 184                            RAS_VECTOR_OFFSET,
 185                            irq_map[irq].hwirq,
 186                            RTAS_EPOW_WARNING | RTAS_POWERMGM_EVENTS,
 187                            critical, __pa(&ras_log_buf),
 188                                 rtas_get_error_log_max());
 189
 190         udbg_printf("EPOW <0x%lx 0x%x 0x%x>\n",
 191                     *((unsigned long *)&ras_log_buf), status, state);
 192         printk(KERN_WARNING "EPOW <0x%lx 0x%x 0x%x>\n",
 193                *((unsigned long *)&ras_log_buf), status, state);
 194
 195         /* format and print the extended information */
 196         log_error(ras_log_buf, ERR_TYPE_RTAS_LOG, 0);
 197
 198         spin_unlock(&ras_log_buf_lock);
 199         return IRQ_HANDLED;
 200 }
 201
 202 /*
 203  * Handle hardware error interrupts.
 204  *
 205  * RTAS check-exception is called to collect data on the exception.  If
 206  * the error is deemed recoverable, we log a warning and return.
 207  * For nonrecoverable errors, an error is logged and we stop all processing
 208  * as quickly as possible in order to prevent propagation of the failure.
 209  */
 210 static irqreturn_t ras_error_interrupt(int irq, void *dev_id)
 211 {
 212         struct rtas_error_log *rtas_elog;
 213         int status = 0xdeadbeef;
 214         int fatal;
 215
 216         spin_lock(&ras_log_buf_lock);
 217
 218         status = rtas_call(ras_check_exception_token, 6, 1, NULL,
 219                            RAS_VECTOR_OFFSET,
 220                            irq_map[irq].hwirq,
 221                            RTAS_INTERNAL_ERROR, 1 /*Time Critical */,
 222                            __pa(&ras_log_buf),
 223                                 rtas_get_error_log_max());
 224
 225         rtas_elog = (struct rtas_error_log *)ras_log_buf;
 226
 227         if ((status == 0) && (rtas_elog->severity >= RTAS_SEVERITY_ERROR_SYNC))
 228                 fatal = 1;
 229         else
 230                 fatal = 0;
 231
 232         /* format and print the extended information */
 233         log_error(ras_log_buf, ERR_TYPE_RTAS_LOG, fatal);
 234
 235         if (fatal) {
 236                 udbg_printf("Fatal HW Error <0x%lx 0x%x>\n",
 237                             *((unsigned long *)&ras_log_buf), status);
 238                 printk(KERN_EMERG "Error: Fatal hardware error <0x%lx 0x%x>\n",
 239                        *((unsigned long *)&ras_log_buf), status);
 240
 241 #ifndef DEBUG
 242                 /* Don't actually power off when debugging so we can test
 243                  * without actually failing while injecting errors.
 244                  * Error data will not be logged to syslog.
 245                  */
 246                 ppc_md.power_off();
 247 #endif
 248         } else {
 249                 udbg_printf("Recoverable HW Error <0x%lx 0x%x>\n",
 250                             *((unsigned long *)&ras_log_buf), status);
 251                 printk(KERN_WARNING
 252                        "Warning: Recoverable hardware error <0x%lx 0x%x>\n",
 253                        *((unsigned long *)&ras_log_buf), status);
 254         }
 255
 256         spin_unlock(&ras_log_buf_lock);
 257         return IRQ_HANDLED;
 258 }
 259
 260 /* Get the error information for errors coming through the
 261  * FWNMI vectors.  The pt_regs' r3 will be updated to reflect
 262  * the actual r3 if possible, and a ptr to the error log entry
 263  * will be returned if found.
 264  *
 265  * The mce_data_buf does not have any locks or protection around it,
 266  * if a second machine check comes in, or a system reset is done
 267  * before we have logged the error, then we will get corruption in the
 268  * error log.  This is preferable over holding off on calling
 269  * ibm,nmi-interlock which would result in us checkstopping if a
 270  * second machine check did come in.
 271  */
 272 static struct rtas_error_log *fwnmi_get_errinfo(struct pt_regs *regs)
 273 {
 274         unsigned long errdata = regs->gpr[3];
 275         struct rtas_error_log *errhdr = NULL;
 276         unsigned long *savep;
 277
 278         if ((errdata >= 0x7000 && errdata < 0x7fff0) ||
 279             (errdata >= rtas.base && errdata < rtas.base + rtas.size - 16)) {
 280                 savep = __va(errdata);
 281                 regs->gpr[3] = savep[0];        /* restore original r3 */
 282                 memset(mce_data_buf, 0, RTAS_ERROR_LOG_MAX);
 283                 memcpy(mce_data_buf, (char *)(savep + 1), RTAS_ERROR_LOG_MAX);
 284                 errhdr = (struct rtas_error_log *)mce_data_buf;
 285         } else {
 286                 printk("FWNMI: corrupt r3\n");
 287         }
 288         return errhdr;
 289 }
 290
 291 /* Call this when done with the data returned by FWNMI_get_errinfo.
 292  * It will release the saved data area for other CPUs in the
 293  * partition to receive FWNMI errors.
 294  */
 295 static void fwnmi_release_errinfo(void)
 296 {
 297         int ret = rtas_call(rtas_token("ibm,nmi-interlock"), 0, 1, NULL);
 298         if (ret != 0)
 299                 printk("FWNMI: nmi-interlock failed: %d\n", ret);
 300 }
 301
 302 int pSeries_system_reset_exception(struct pt_regs *regs)
 303 {
 304         if (fwnmi_active) {
 305                 struct rtas_error_log *errhdr = fwnmi_get_errinfo(regs);
 306                 if (errhdr) {
 307                         /* XXX Should look at FWNMI information */
 308                 }
 309                 fwnmi_release_errinfo();
 310         }
 311         return 0; /* need to perform reset */
 312 }
 313
 314 /*
 315  * See if we can recover from a machine check exception.
 316  * This is only called on power4 (or above) and only via
 317  * the Firmware Non-Maskable Interrupts (fwnmi) handler
 318  * which provides the error analysis for us.
 319  *
 320  * Return 1 if corrected (or delivered a signal).
 321  * Return 0 if there is nothing we can do.
 322  */
 323 static int recover_mce(struct pt_regs *regs, struct rtas_error_log * err)
 324 {
 325         int nonfatal = 0;
 326
 327         if (err->disposition == RTAS_DISP_FULLY_RECOVERED) {
 328                 /* Platform corrected itself */
 329                 nonfatal = 1;
 330         } else if ((regs->msr & MSR_RI) &&
 331                    user_mode(regs) &&
 332                    err->severity == RTAS_SEVERITY_ERROR_SYNC &&
 333                    err->disposition == RTAS_DISP_NOT_RECOVERED &&
 334                    err->target == RTAS_TARGET_MEMORY &&
 335                    err->type == RTAS_TYPE_ECC_UNCORR &&
 336                    !(current->pid == 0 || is_init(current))) {
 337                 /* Kill off a user process with an ECC error */
 338                 printk(KERN_ERR "MCE: uncorrectable ecc error for pid %d\n",
 339                        current->pid);
 340                 /* XXX something better for ECC error? */
 341                 _exception(SIGBUS, regs, BUS_ADRERR, regs->nip);
 342                 nonfatal = 1;
 343         }
 344
 345         log_error((char *)err, ERR_TYPE_RTAS_LOG, !nonfatal);
 346
 347         return nonfatal;
 348 }
 349
 350 /*
 351  * Handle a machine check.
 352  *
 353  * Note that on Power 4 and beyond Firmware Non-Maskable Interrupts (fwnmi)
 354  * should be present.  If so the handler which called us tells us if the
 355  * error was recovered (never true if RI=0).
 356  *
 357  * On hardware prior to Power 4 these exceptions were asynchronous which
 358  * means we can't tell exactly where it occurred and so we can't recover.
 359  */
 360 int pSeries_machine_check_exception(struct pt_regs *regs)
 361 {
 362         struct rtas_error_log *errp;
 363
 364         if (fwnmi_active) {
 365                 errp = fwnmi_get_errinfo(regs);
 366                 fwnmi_release_errinfo();
 367                 if (errp && recover_mce(regs, errp))
 368                         return 1;
 369         }
 370
 371         return 0;
 372 }