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/delay.h>
  34 #include <linux/irq.h>
  35 #include <linux/random.h>
  36 #include <linux/sysrq.h>
  37 #include <linux/bitops.h>
  38
  39 #include <asm/uaccess.h>
  40 #include <asm/system.h>
  41 #include <asm/io.h>
  42 #include <asm/pgtable.h>
  43 #include <asm/irq.h>
  44 #include <asm/cache.h>
  45 #include <asm/prom.h>
  46 #include <asm/ptrace.h>
  47 #include <asm/machdep.h>
  48 #include <asm/rtas.h>
  49 #include <asm/udbg.h>
  50 #include <asm/firmware.h>
  51
  52 #include "pseries.h"
  53
  54 static unsigned char ras_log_buf[RTAS_ERROR_LOG_MAX];
  55 static DEFINE_SPINLOCK(ras_log_buf_lock);
  56
  57 static char global_mce_data_buf[RTAS_ERROR_LOG_MAX];
  58 static DEFINE_PER_CPU(__u64, mce_data_buf);
  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
  66 static irqreturn_t ras_epow_interrupt(int irq, void *dev_id);
  67 static irqreturn_t ras_error_interrupt(int irq, void *dev_id);
  68
  69
  70 /*
  71  * Initialize handlers for the set of interrupts caused by hardware errors
  72  * and power system events.
  73  */
  74 static int __init init_ras_IRQ(void)
  75 {
  76         struct device_node *np;
  77
  78         ras_get_sensor_state_token = rtas_token("get-sensor-state");
  79         ras_check_exception_token = rtas_token("check-exception");
  80
  81         /* Internal Errors */
  82         np = of_find_node_by_path("/event-sources/internal-errors");
  83         if (np != NULL) {
  84                 request_event_sources_irqs(np, ras_error_interrupt,
  85                                            "RAS_ERROR");
  86                 of_node_put(np);
  87         }
  88
  89         /* EPOW Events */
  90         np = of_find_node_by_path("/event-sources/epow-events");
  91         if (np != NULL) {
  92                 request_event_sources_irqs(np, ras_epow_interrupt, "RAS_EPOW");
  93                 of_node_put(np);
  94         }
  95
  96         return 0;
  97 }
  98 __initcall(init_ras_IRQ);
  99
 100 /*
 101  * Handle power subsystem events (EPOW).
 102  *
 103  * Presently we just log the event has occurred.  This should be fixed
 104  * to examine the type of power failure and take appropriate action where
 105  * the time horizon permits something useful to be done.
 106  */
 107 static irqreturn_t ras_epow_interrupt(int irq, void *dev_id)
 108 {
 109         int status = 0xdeadbeef;
 110         int state = 0;
 111         int critical;
 112
 113         status = rtas_call(ras_get_sensor_state_token, 2, 2, &state,
 114                            EPOW_SENSOR_TOKEN, EPOW_SENSOR_INDEX);
 115
 116         if (state > 3)
 117                 critical = 1;  /* Time Critical */
 118         else
 119                 critical = 0;
 120
 121         spin_lock(&ras_log_buf_lock);
 122
 123         status = rtas_call(ras_check_exception_token, 6, 1, NULL,
 124                            RTAS_VECTOR_EXTERNAL_INTERRUPT,
 125                            virq_to_hw(irq),
 126                            RTAS_EPOW_WARNING | RTAS_POWERMGM_EVENTS,
 127                            critical, __pa(&ras_log_buf),
 128                                 rtas_get_error_log_max());
 129
 130         udbg_printf("EPOW <0x%lx 0x%x 0x%x>\n",
 131                     *((unsigned long *)&ras_log_buf), status, state);
 132         printk(KERN_WARNING "EPOW <0x%lx 0x%x 0x%x>\n",
 133                *((unsigned long *)&ras_log_buf), status, state);
 134
 135         /* format and print the extended information */
 136         log_error(ras_log_buf, ERR_TYPE_RTAS_LOG, 0);
 137
 138         spin_unlock(&ras_log_buf_lock);
 139         return IRQ_HANDLED;
 140 }
 141
 142 /*
 143  * Handle hardware error interrupts.
 144  *
 145  * RTAS check-exception is called to collect data on the exception.  If
 146  * the error is deemed recoverable, we log a warning and return.
 147  * For nonrecoverable errors, an error is logged and we stop all processing
 148  * as quickly as possible in order to prevent propagation of the failure.
 149  */
 150 static irqreturn_t ras_error_interrupt(int irq, void *dev_id)
 151 {
 152         struct rtas_error_log *rtas_elog;
 153         int status = 0xdeadbeef;
 154         int fatal;
 155
 156         spin_lock(&ras_log_buf_lock);
 157
 158         status = rtas_call(ras_check_exception_token, 6, 1, NULL,
 159                            RTAS_VECTOR_EXTERNAL_INTERRUPT,
 160                            virq_to_hw(irq),
 161                            RTAS_INTERNAL_ERROR, 1 /*Time Critical */,
 162                            __pa(&ras_log_buf),
 163                                 rtas_get_error_log_max());
 164
 165         rtas_elog = (struct rtas_error_log *)ras_log_buf;
 166
 167         if ((status == 0) && (rtas_elog->severity >= RTAS_SEVERITY_ERROR_SYNC))
 168                 fatal = 1;
 169         else
 170                 fatal = 0;
 171
 172         /* format and print the extended information */
 173         log_error(ras_log_buf, ERR_TYPE_RTAS_LOG, fatal);
 174
 175         if (fatal) {
 176                 udbg_printf("Fatal HW Error <0x%lx 0x%x>\n",
 177                             *((unsigned long *)&ras_log_buf), status);
 178                 printk(KERN_EMERG "Error: Fatal hardware error <0x%lx 0x%x>\n",
 179                        *((unsigned long *)&ras_log_buf), status);
 180
 181 #ifndef DEBUG_RTAS_POWER_OFF
 182                 /* Don't actually power off when debugging so we can test
 183                  * without actually failing while injecting errors.
 184                  * Error data will not be logged to syslog.
 185                  */
 186                 ppc_md.power_off();
 187 #endif
 188         } else {
 189                 udbg_printf("Recoverable HW Error <0x%lx 0x%x>\n",
 190                             *((unsigned long *)&ras_log_buf), status);
 191                 printk(KERN_WARNING
 192                        "Warning: Recoverable hardware error <0x%lx 0x%x>\n",
 193                        *((unsigned long *)&ras_log_buf), status);
 194         }
 195
 196         spin_unlock(&ras_log_buf_lock);
 197         return IRQ_HANDLED;
 198 }
 199
 200 /*
 201  * Some versions of FWNMI place the buffer inside the 4kB page starting at
 202  * 0x7000. Other versions place it inside the rtas buffer. We check both.
 203  */
 204 #define VALID_FWNMI_BUFFER(A) \
 205         ((((A) >= 0x7000) && ((A) < 0x7ff0)) || \
 206         (((A) >= rtas.base) && ((A) < (rtas.base + rtas.size - 16))))
 207
 208 /*
 209  * Get the error information for errors coming through the
 210  * FWNMI vectors.  The pt_regs' r3 will be updated to reflect
 211  * the actual r3 if possible, and a ptr to the error log entry
 212  * will be returned if found.
 213  *
 214  * If the RTAS error is not of the extended type, then we put it in a per
 215  * cpu 64bit buffer. If it is the extended type we use global_mce_data_buf.
 216  *
 217  * The global_mce_data_buf does not have any locks or protection around it,
 218  * if a second machine check comes in, or a system reset is done
 219  * before we have logged the error, then we will get corruption in the
 220  * error log.  This is preferable over holding off on calling
 221  * ibm,nmi-interlock which would result in us checkstopping if a
 222  * second machine check did come in.
 223  */
 224 static struct rtas_error_log *fwnmi_get_errinfo(struct pt_regs *regs)
 225 {
 226         unsigned long *savep;
 227         struct rtas_error_log *h, *errhdr = NULL;
 228
 229         if (!VALID_FWNMI_BUFFER(regs->gpr[3])) {
 230                 printk(KERN_ERR "FWNMI: corrupt r3 0x%016lx\n", regs->gpr[3]);
 231                 return NULL;
 232         }
 233
 234         savep = __va(regs->gpr[3]);
 235         regs->gpr[3] = savep[0];        /* restore original r3 */
 236
 237         /* If it isn't an extended log we can use the per cpu 64bit buffer */
 238         h = (struct rtas_error_log *)&savep[1];
 239         if (!h->extended) {
 240                 memcpy(&__get_cpu_var(mce_data_buf), h, sizeof(__u64));
 241                 errhdr = (struct rtas_error_log *)&__get_cpu_var(mce_data_buf);
 242         } else {
 243                 int len;
 244
 245                 len = max_t(int, 8+h->extended_log_length, RTAS_ERROR_LOG_MAX);
 246                 memset(global_mce_data_buf, 0, RTAS_ERROR_LOG_MAX);
 247                 memcpy(global_mce_data_buf, h, len);
 248                 errhdr = (struct rtas_error_log *)global_mce_data_buf;
 249         }
 250
 251         return errhdr;
 252 }
 253
 254 /* Call this when done with the data returned by FWNMI_get_errinfo.
 255  * It will release the saved data area for other CPUs in the
 256  * partition to receive FWNMI errors.
 257  */
 258 static void fwnmi_release_errinfo(void)
 259 {
 260         int ret = rtas_call(rtas_token("ibm,nmi-interlock"), 0, 1, NULL);
 261         if (ret != 0)
 262                 printk(KERN_ERR "FWNMI: nmi-interlock failed: %d\n", ret);
 263 }
 264
 265 int pSeries_system_reset_exception(struct pt_regs *regs)
 266 {
 267         if (fwnmi_active) {
 268                 struct rtas_error_log *errhdr = fwnmi_get_errinfo(regs);
 269                 if (errhdr) {
 270                         /* XXX Should look at FWNMI information */
 271                 }
 272                 fwnmi_release_errinfo();
 273         }
 274         return 0; /* need to perform reset */
 275 }
 276
 277 /*
 278  * See if we can recover from a machine check exception.
 279  * This is only called on power4 (or above) and only via
 280  * the Firmware Non-Maskable Interrupts (fwnmi) handler
 281  * which provides the error analysis for us.
 282  *
 283  * Return 1 if corrected (or delivered a signal).
 284  * Return 0 if there is nothing we can do.
 285  */
 286 static int recover_mce(struct pt_regs *regs, struct rtas_error_log *err)
 287 {
 288         int recovered = 0;
 289
 290         if (!(regs->msr & MSR_RI)) {
 291                 /* If MSR_RI isn't set, we cannot recover */
 292                 recovered = 0;
 293
 294         } else if (err->disposition == RTAS_DISP_FULLY_RECOVERED) {
 295                 /* Platform corrected itself */
 296                 recovered = 1;
 297
 298         } else if (err->disposition == RTAS_DISP_LIMITED_RECOVERY) {
 299                 /* Platform corrected itself but could be degraded */
 300                 printk(KERN_ERR "MCE: limited recovery, system may "
 301                        "be degraded\n");
 302                 recovered = 1;
 303
 304         } else if (user_mode(regs) && !is_global_init(current) &&
 305                    err->severity == RTAS_SEVERITY_ERROR_SYNC) {
 306
 307                 /*
 308                  * If we received a synchronous error when in userspace
 309                  * kill the task. Firmware may report details of the fail
 310                  * asynchronously, so we can't rely on the target and type
 311                  * fields being valid here.
 312                  */
 313                 printk(KERN_ERR "MCE: uncorrectable error, killing task "
 314                        "%s:%d\n", current->comm, current->pid);
 315
 316                 _exception(SIGBUS, regs, BUS_MCEERR_AR, regs->nip);
 317                 recovered = 1;
 318         }
 319
 320         log_error((char *)err, ERR_TYPE_RTAS_LOG, 0);
 321
 322         return recovered;
 323 }
 324
 325 /*
 326  * Handle a machine check.
 327  *
 328  * Note that on Power 4 and beyond Firmware Non-Maskable Interrupts (fwnmi)
 329  * should be present.  If so the handler which called us tells us if the
 330  * error was recovered (never true if RI=0).
 331  *
 332  * On hardware prior to Power 4 these exceptions were asynchronous which
 333  * means we can't tell exactly where it occurred and so we can't recover.
 334  */
 335 int pSeries_machine_check_exception(struct pt_regs *regs)
 336 {
 337         struct rtas_error_log *errp;
 338
 339         if (fwnmi_active) {
 340                 errp = fwnmi_get_errinfo(regs);
 341                 fwnmi_release_errinfo();
 342                 if (errp && recover_mce(regs, errp))
 343                         return 1;
 344         }
 345
 346         return 0;
 347 }