arch/powerpc/kernel/rtasd.c

   1 /*
   2  * Copyright (C) 2001 Anton Blanchard <anton@au.ibm.com>, IBM
   3  *
   4  * This program is free software; you can redistribute it and/or
   5  * modify it under the terms of the GNU General Public License
   6  * as published by the Free Software Foundation; either version
   7  * 2 of the License, or (at your option) any later version.
   8  *
   9  * Communication to userspace based on kernel/printk.c
  10  */
  11
  12 #include <linux/types.h>
  13 #include <linux/errno.h>
  14 #include <linux/sched.h>
  15 #include <linux/kernel.h>
  16 #include <linux/poll.h>
  17 #include <linux/proc_fs.h>
  18 #include <linux/init.h>
  19 #include <linux/vmalloc.h>
  20 #include <linux/spinlock.h>
  21 #include <linux/cpu.h>
  22 #include <linux/workqueue.h>
  23 #include <linux/slab.h>
  24 #include <linux/topology.h>
  25
  26 #include <linux/uaccess.h>
  27 #include <asm/io.h>
  28 #include <asm/rtas.h>
  29 #include <asm/prom.h>
  30 #include <asm/nvram.h>
  31 #include <linux/atomic.h>
  32 #include <asm/machdep.h>
  33 #include <asm/topology.h>
  34
  35
  36 static DEFINE_SPINLOCK(rtasd_log_lock);
  37
  38 static DECLARE_WAIT_QUEUE_HEAD(rtas_log_wait);
  39
  40 static char *rtas_log_buf;
  41 static unsigned long rtas_log_start;
  42 static unsigned long rtas_log_size;
  43
  44 static int surveillance_timeout = -1;
  45
  46 static unsigned int rtas_error_log_max;
  47 static unsigned int rtas_error_log_buffer_max;
  48
  49 /* RTAS service tokens */
  50 static unsigned int event_scan;
  51 static unsigned int rtas_event_scan_rate;
  52
  53 static bool full_rtas_msgs;
  54
  55 /* Stop logging to nvram after first fatal error */
  56 static int logging_enabled; /* Until we initialize everything,
  57                              * make sure we don't try logging
  58                              * anything */
  59 static int error_log_cnt;
  60
  61 /*
  62  * Since we use 32 bit RTAS, the physical address of this must be below
  63  * 4G or else bad things happen. Allocate this in the kernel data and
  64  * make it big enough.
  65  */
  66 static unsigned char logdata[RTAS_ERROR_LOG_MAX];
  67
  68 static char *rtas_type[] = {
  69         "Unknown", "Retry", "TCE Error", "Internal Device Failure",
  70         "Timeout", "Data Parity", "Address Parity", "Cache Parity",
  71         "Address Invalid", "ECC Uncorrected", "ECC Corrupted",
  72 };
  73
  74 static char *rtas_event_type(int type)
  75 {
  76         if ((type > 0) && (type < 11))
  77                 return rtas_type[type];
  78
  79         switch (type) {
  80                 case RTAS_TYPE_EPOW:
  81                         return "EPOW";
  82                 case RTAS_TYPE_PLATFORM:
  83                         return "Platform Error";
  84                 case RTAS_TYPE_IO:
  85                         return "I/O Event";
  86                 case RTAS_TYPE_INFO:
  87                         return "Platform Information Event";
  88                 case RTAS_TYPE_DEALLOC:
  89                         return "Resource Deallocation Event";
  90                 case RTAS_TYPE_DUMP:
  91                         return "Dump Notification Event";
  92                 case RTAS_TYPE_PRRN:
  93                         return "Platform Resource Reassignment Event";
  94         }
  95
  96         return rtas_type[0];
  97 }
  98
  99 /* To see this info, grep RTAS /var/log/messages and each entry
 100  * will be collected together with obvious begin/end.
 101  * There will be a unique identifier on the begin and end lines.
 102  * This will persist across reboots.
 103  *
 104  * format of error logs returned from RTAS:
 105  * bytes        (size)  : contents
 106  * --------------------------------------------------------
 107  * 0-7          (8)     : rtas_error_log
 108  * 8-47         (40)    : extended info
 109  * 48-51        (4)     : vendor id
 110  * 52-1023 (vendor specific) : location code and debug data
 111  */
 112 static void printk_log_rtas(char *buf, int len)
 113 {
 114
 115         int i,j,n = 0;
 116         int perline = 16;
 117         char buffer[64];
 118         char * str = "RTAS event";
 119
 120         if (full_rtas_msgs) {
 121                 printk(RTAS_DEBUG "%d -------- %s begin --------\n",
 122                        error_log_cnt, str);
 123
 124                 /*
 125                  * Print perline bytes on each line, each line will start
 126                  * with RTAS and a changing number, so syslogd will
 127                  * print lines that are otherwise the same.  Separate every
 128                  * 4 bytes with a space.
 129                  */
 130                 for (i = 0; i < len; i++) {
 131                         j = i % perline;
 132                         if (j == 0) {
 133                                 memset(buffer, 0, sizeof(buffer));
 134                                 n = sprintf(buffer, "RTAS %d:", i/perline);
 135                         }
 136
 137                         if ((i % 4) == 0)
 138                                 n += sprintf(buffer+n, " ");
 139
 140                         n += sprintf(buffer+n, "%02x", (unsigned char)buf[i]);
 141
 142                         if (j == (perline-1))
 143                                 printk(KERN_DEBUG "%s\n", buffer);
 144                 }
 145                 if ((i % perline) != 0)
 146                         printk(KERN_DEBUG "%s\n", buffer);
 147
 148                 printk(RTAS_DEBUG "%d -------- %s end ----------\n",
 149                        error_log_cnt, str);
 150         } else {
 151                 struct rtas_error_log *errlog = (struct rtas_error_log *)buf;
 152
 153                 printk(RTAS_DEBUG "event: %d, Type: %s, Severity: %d\n",
 154                        error_log_cnt, rtas_event_type(rtas_error_type(errlog)),
 155                        rtas_error_severity(errlog));
 156         }
 157 }
 158
 159 static int log_rtas_len(char * buf)
 160 {
 161         int len;
 162         struct rtas_error_log *err;
 163         uint32_t extended_log_length;
 164
 165         /* rtas fixed header */
 166         len = 8;
 167         err = (struct rtas_error_log *)buf;
 168         extended_log_length = rtas_error_extended_log_length(err);
 169         if (rtas_error_extended(err) && extended_log_length) {
 170
 171                 /* extended header */
 172                 len += extended_log_length;
 173         }
 174
 175         if (rtas_error_log_max == 0)
 176                 rtas_error_log_max = rtas_get_error_log_max();
 177
 178         if (len > rtas_error_log_max)
 179                 len = rtas_error_log_max;
 180
 181         return len;
 182 }
 183
 184 /*
 185  * First write to nvram, if fatal error, that is the only
 186  * place we log the info.  The error will be picked up
 187  * on the next reboot by rtasd.  If not fatal, run the
 188  * method for the type of error.  Currently, only RTAS
 189  * errors have methods implemented, but in the future
 190  * there might be a need to store data in nvram before a
 191  * call to panic().
 192  *
 193  * XXX We write to nvram periodically, to indicate error has
 194  * been written and sync'd, but there is a possibility
 195  * that if we don't shutdown correctly, a duplicate error
 196  * record will be created on next reboot.
 197  */
 198 void pSeries_log_error(char *buf, unsigned int err_type, int fatal)
 199 {
 200         unsigned long offset;
 201         unsigned long s;
 202         int len = 0;
 203
 204         pr_debug("rtasd: logging event\n");
 205         if (buf == NULL)
 206                 return;
 207
 208         spin_lock_irqsave(&rtasd_log_lock, s);
 209
 210         /* get length and increase count */
 211         switch (err_type & ERR_TYPE_MASK) {
 212         case ERR_TYPE_RTAS_LOG:
 213                 len = log_rtas_len(buf);
 214                 if (!(err_type & ERR_FLAG_BOOT))
 215                         error_log_cnt++;
 216                 break;
 217         case ERR_TYPE_KERNEL_PANIC:
 218         default:
 219                 WARN_ON_ONCE(!irqs_disabled()); /* @@@ DEBUG @@@ */
 220                 spin_unlock_irqrestore(&rtasd_log_lock, s);
 221                 return;
 222         }
 223
 224 #ifdef CONFIG_PPC64
 225         /* Write error to NVRAM */
 226         if (logging_enabled && !(err_type & ERR_FLAG_BOOT))
 227                 nvram_write_error_log(buf, len, err_type, error_log_cnt);
 228 #endif /* CONFIG_PPC64 */
 229
 230         /*
 231          * rtas errors can occur during boot, and we do want to capture
 232          * those somewhere, even if nvram isn't ready (why not?), and even
 233          * if rtasd isn't ready. Put them into the boot log, at least.
 234          */
 235         if ((err_type & ERR_TYPE_MASK) == ERR_TYPE_RTAS_LOG)
 236                 printk_log_rtas(buf, len);
 237
 238         /* Check to see if we need to or have stopped logging */
 239         if (fatal || !logging_enabled) {
 240                 logging_enabled = 0;
 241                 WARN_ON_ONCE(!irqs_disabled()); /* @@@ DEBUG @@@ */
 242                 spin_unlock_irqrestore(&rtasd_log_lock, s);
 243                 return;
 244         }
 245
 246         /* call type specific method for error */
 247         switch (err_type & ERR_TYPE_MASK) {
 248         case ERR_TYPE_RTAS_LOG:
 249                 offset = rtas_error_log_buffer_max *
 250                         ((rtas_log_start+rtas_log_size) & LOG_NUMBER_MASK);
 251
 252                 /* First copy over sequence number */
 253                 memcpy(&rtas_log_buf[offset], (void *) &error_log_cnt, sizeof(int));
 254
 255                 /* Second copy over error log data */
 256                 offset += sizeof(int);
 257                 memcpy(&rtas_log_buf[offset], buf, len);
 258
 259                 if (rtas_log_size < LOG_NUMBER)
 260                         rtas_log_size += 1;
 261                 else
 262                         rtas_log_start += 1;
 263
 264                 WARN_ON_ONCE(!irqs_disabled()); /* @@@ DEBUG @@@ */
 265                 spin_unlock_irqrestore(&rtasd_log_lock, s);
 266                 wake_up_interruptible(&rtas_log_wait);
 267                 break;
 268         case ERR_TYPE_KERNEL_PANIC:
 269         default:
 270                 WARN_ON_ONCE(!irqs_disabled()); /* @@@ DEBUG @@@ */
 271                 spin_unlock_irqrestore(&rtasd_log_lock, s);
 272                 return;
 273         }
 274 }
 275
 276 #ifdef CONFIG_PPC_PSERIES
 277 static s32 prrn_update_scope;
 278
 279 static void prrn_work_fn(struct work_struct *work)
 280 {
 281         /*
 282          * For PRRN, we must pass the negative of the scope value in
 283          * the RTAS event.
 284          */
 285         pseries_devicetree_update(-prrn_update_scope);
 286         numa_update_cpu_topology(false);
 287 }
 288
 289 static DECLARE_WORK(prrn_work, prrn_work_fn);
 290
 291 static void prrn_schedule_update(u32 scope)
 292 {
 293         flush_work(&prrn_work);
 294         prrn_update_scope = scope;
 295         schedule_work(&prrn_work);
 296 }
 297
 298 static void handle_rtas_event(const struct rtas_error_log *log)
 299 {
 300         if (rtas_error_type(log) != RTAS_TYPE_PRRN || !prrn_is_enabled())
 301                 return;
 302
 303         /* For PRRN Events the extended log length is used to denote
 304          * the scope for calling rtas update-nodes.
 305          */
 306         prrn_schedule_update(rtas_error_extended_log_length(log));
 307 }
 308
 309 #else
 310
 311 static void handle_rtas_event(const struct rtas_error_log *log)
 312 {
 313         return;
 314 }
 315
 316 #endif
 317
 318 static int rtas_log_open(struct inode * inode, struct file * file)
 319 {
 320         return 0;
 321 }
 322
 323 static int rtas_log_release(struct inode * inode, struct file * file)
 324 {
 325         return 0;
 326 }
 327
 328 /* This will check if all events are logged, if they are then, we
 329  * know that we can safely clear the events in NVRAM.
 330  * Next we'll sit and wait for something else to log.
 331  */
 332 static ssize_t rtas_log_read(struct file * file, char __user * buf,
 333                          size_t count, loff_t *ppos)
 334 {
 335         int error;
 336         char *tmp;
 337         unsigned long s;
 338         unsigned long offset;
 339
 340         if (!buf || count < rtas_error_log_buffer_max)
 341                 return -EINVAL;
 342
 343         count = rtas_error_log_buffer_max;
 344
 345         if (!access_ok(VERIFY_WRITE, buf, count))
 346                 return -EFAULT;
 347
 348         tmp = kmalloc(count, GFP_KERNEL);
 349         if (!tmp)
 350                 return -ENOMEM;
 351
 352         spin_lock_irqsave(&rtasd_log_lock, s);
 353
 354         /* if it's 0, then we know we got the last one (the one in NVRAM) */
 355         while (rtas_log_size == 0) {
 356                 if (file->f_flags & O_NONBLOCK) {
 357                         spin_unlock_irqrestore(&rtasd_log_lock, s);
 358                         error = -EAGAIN;
 359                         goto out;
 360                 }
 361
 362                 if (!logging_enabled) {
 363                         spin_unlock_irqrestore(&rtasd_log_lock, s);
 364                         error = -ENODATA;
 365                         goto out;
 366                 }
 367 #ifdef CONFIG_PPC64
 368                 nvram_clear_error_log();
 369 #endif /* CONFIG_PPC64 */
 370
 371                 spin_unlock_irqrestore(&rtasd_log_lock, s);
 372                 error = wait_event_interruptible(rtas_log_wait, rtas_log_size);
 373                 if (error)
 374                         goto out;
 375                 spin_lock_irqsave(&rtasd_log_lock, s);
 376         }
 377
 378         offset = rtas_error_log_buffer_max * (rtas_log_start & LOG_NUMBER_MASK);
 379         memcpy(tmp, &rtas_log_buf[offset], count);
 380
 381         rtas_log_start += 1;
 382         rtas_log_size -= 1;
 383         spin_unlock_irqrestore(&rtasd_log_lock, s);
 384
 385         error = copy_to_user(buf, tmp, count) ? -EFAULT : count;
 386 out:
 387         kfree(tmp);
 388         return error;
 389 }
 390
 391 static __poll_t rtas_log_poll(struct file *file, poll_table * wait)
 392 {
 393         poll_wait(file, &rtas_log_wait, wait);
 394         if (rtas_log_size)
 395                 return EPOLLIN | EPOLLRDNORM;
 396         return 0;
 397 }
 398
 399 static const struct file_operations proc_rtas_log_operations = {
 400         .read =         rtas_log_read,
 401         .poll =         rtas_log_poll,
 402         .open =         rtas_log_open,
 403         .release =      rtas_log_release,
 404         .llseek =       noop_llseek,
 405 };
 406
 407 static int enable_surveillance(int timeout)
 408 {
 409         int error;
 410
 411         error = rtas_set_indicator(SURVEILLANCE_TOKEN, 0, timeout);
 412
 413         if (error == 0)
 414                 return 0;
 415
 416         if (error == -EINVAL) {
 417                 printk(KERN_DEBUG "rtasd: surveillance not supported\n");
 418                 return 0;
 419         }
 420
 421         printk(KERN_ERR "rtasd: could not update surveillance\n");
 422         return -1;
 423 }
 424
 425 static void do_event_scan(void)
 426 {
 427         int error;
 428         do {
 429                 memset(logdata, 0, rtas_error_log_max);
 430                 error = rtas_call(event_scan, 4, 1, NULL,
 431                                   RTAS_EVENT_SCAN_ALL_EVENTS, 0,
 432                                   __pa(logdata), rtas_error_log_max);
 433                 if (error == -1) {
 434                         printk(KERN_ERR "event-scan failed\n");
 435                         break;
 436                 }
 437
 438                 if (error == 0) {
 439                         if (rtas_error_type((struct rtas_error_log *)logdata) !=
 440                             RTAS_TYPE_PRRN)
 441                                 pSeries_log_error(logdata, ERR_TYPE_RTAS_LOG,
 442                                                   0);
 443                         handle_rtas_event((struct rtas_error_log *)logdata);
 444                 }
 445
 446         } while(error == 0);
 447 }
 448
 449 static void rtas_event_scan(struct work_struct *w);
 450 static DECLARE_DELAYED_WORK(event_scan_work, rtas_event_scan);
 451
 452 /*
 453  * Delay should be at least one second since some machines have problems if
 454  * we call event-scan too quickly.
 455  */
 456 static unsigned long event_scan_delay = 1*HZ;
 457 static int first_pass = 1;
 458
 459 static void rtas_event_scan(struct work_struct *w)
 460 {
 461         unsigned int cpu;
 462
 463         do_event_scan();
 464
 465         get_online_cpus();
 466
 467         /* raw_ OK because just using CPU as starting point. */
 468         cpu = cpumask_next(raw_smp_processor_id(), cpu_online_mask);
 469         if (cpu >= nr_cpu_ids) {
 470                 cpu = cpumask_first(cpu_online_mask);
 471
 472                 if (first_pass) {
 473                         first_pass = 0;
 474                         event_scan_delay = 30*HZ/rtas_event_scan_rate;
 475
 476                         if (surveillance_timeout != -1) {
 477                                 pr_debug("rtasd: enabling surveillance\n");
 478                                 enable_surveillance(surveillance_timeout);
 479                                 pr_debug("rtasd: surveillance enabled\n");
 480                         }
 481                 }
 482         }
 483
 484         schedule_delayed_work_on(cpu, &event_scan_work,
 485                 __round_jiffies_relative(event_scan_delay, cpu));
 486
 487         put_online_cpus();
 488 }
 489
 490 #ifdef CONFIG_PPC64
 491 static void retrieve_nvram_error_log(void)
 492 {
 493         unsigned int err_type ;
 494         int rc ;
 495
 496         /* See if we have any error stored in NVRAM */
 497         memset(logdata, 0, rtas_error_log_max);
 498         rc = nvram_read_error_log(logdata, rtas_error_log_max,
 499                                   &err_type, &error_log_cnt);
 500         /* We can use rtas_log_buf now */
 501         logging_enabled = 1;
 502         if (!rc) {
 503                 if (err_type != ERR_FLAG_ALREADY_LOGGED) {
 504                         pSeries_log_error(logdata, err_type | ERR_FLAG_BOOT, 0);
 505                 }
 506         }
 507 }
 508 #else /* CONFIG_PPC64 */
 509 static void retrieve_nvram_error_log(void)
 510 {
 511 }
 512 #endif /* CONFIG_PPC64 */
 513
 514 static void start_event_scan(void)
 515 {
 516         printk(KERN_DEBUG "RTAS daemon started\n");
 517         pr_debug("rtasd: will sleep for %d milliseconds\n",
 518                  (30000 / rtas_event_scan_rate));
 519
 520         /* Retrieve errors from nvram if any */
 521         retrieve_nvram_error_log();
 522
 523         schedule_delayed_work_on(cpumask_first(cpu_online_mask),
 524                                  &event_scan_work, event_scan_delay);
 525 }
 526
 527 /* Cancel the rtas event scan work */
 528 void rtas_cancel_event_scan(void)
 529 {
 530         cancel_delayed_work_sync(&event_scan_work);
 531 }
 532 EXPORT_SYMBOL_GPL(rtas_cancel_event_scan);
 533
 534 static int __init rtas_event_scan_init(void)
 535 {
 536         if (!machine_is(pseries) && !machine_is(chrp))
 537                 return 0;
 538
 539         /* No RTAS */
 540         event_scan = rtas_token("event-scan");
 541         if (event_scan == RTAS_UNKNOWN_SERVICE) {
 542                 printk(KERN_INFO "rtasd: No event-scan on system\n");
 543                 return -ENODEV;
 544         }
 545
 546         rtas_event_scan_rate = rtas_token("rtas-event-scan-rate");
 547         if (rtas_event_scan_rate == RTAS_UNKNOWN_SERVICE) {
 548                 printk(KERN_ERR "rtasd: no rtas-event-scan-rate on system\n");
 549                 return -ENODEV;
 550         }
 551
 552         if (!rtas_event_scan_rate) {
 553                 /* Broken firmware: take a rate of zero to mean don't scan */
 554                 printk(KERN_DEBUG "rtasd: scan rate is 0, not scanning\n");
 555                 return 0;
 556         }
 557
 558         /* Make room for the sequence number */
 559         rtas_error_log_max = rtas_get_error_log_max();
 560         rtas_error_log_buffer_max = rtas_error_log_max + sizeof(int);
 561
 562         rtas_log_buf = vmalloc(rtas_error_log_buffer_max*LOG_NUMBER);
 563         if (!rtas_log_buf) {
 564                 printk(KERN_ERR "rtasd: no memory\n");
 565                 return -ENOMEM;
 566         }
 567
 568         start_event_scan();
 569
 570         return 0;
 571 }
 572 arch_initcall(rtas_event_scan_init);
 573
 574 static int __init rtas_init(void)
 575 {
 576         struct proc_dir_entry *entry;
 577
 578         if (!machine_is(pseries) && !machine_is(chrp))
 579                 return 0;
 580
 581         if (!rtas_log_buf)
 582                 return -ENODEV;
 583
 584         entry = proc_create("powerpc/rtas/error_log", 0400, NULL,
 585                             &proc_rtas_log_operations);
 586         if (!entry)
 587                 printk(KERN_ERR "Failed to create error_log proc entry\n");
 588
 589         return 0;
 590 }
 591 __initcall(rtas_init);
 592
 593 static int __init surveillance_setup(char *str)
 594 {
 595         int i;
 596
 597         /* We only do surveillance on pseries */
 598         if (!machine_is(pseries))
 599                 return 0;
 600
 601         if (get_option(&str,&i)) {
 602                 if (i >= 0 && i <= 255)
 603                         surveillance_timeout = i;
 604         }
 605
 606         return 1;
 607 }
 608 __setup("surveillance=", surveillance_setup);
 609
 610 static int __init rtasmsgs_setup(char *str)
 611 {
 612         return (kstrtobool(str, &full_rtas_msgs) == 0);
 613 }
 614 __setup("rtasmsgs=", rtasmsgs_setup);