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