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