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