1 // SPDX-License-Identifier: GPL-2.0-or-later 2 /* 3 * Copyright (C) 2001 Anton Blanchard <anton@au.ibm.com>, IBM 4 * 5 * Communication to userspace based on kernel/printk.c 6 */ 7 8 #include <linux/types.h> 9 #include <linux/errno.h> 10 #include <linux/sched.h> 11 #include <linux/kernel.h> 12 #include <linux/poll.h> 13 #include <linux/proc_fs.h> 14 #include <linux/init.h> 15 #include <linux/vmalloc.h> 16 #include <linux/spinlock.h> 17 #include <linux/cpu.h> 18 #include <linux/workqueue.h> 19 #include <linux/slab.h> 20 #include <linux/topology.h> 21 22 #include <linux/uaccess.h> 23 #include <asm/io.h> 24 #include <asm/rtas.h> 25 #include <asm/nvram.h> 26 #include <linux/atomic.h> 27 #include <asm/machdep.h> 28 #include <asm/topology.h> 29 30 31 static DEFINE_SPINLOCK(rtasd_log_lock); 32 33 static DECLARE_WAIT_QUEUE_HEAD(rtas_log_wait); 34 35 static char *rtas_log_buf; 36 static unsigned long rtas_log_start; 37 static unsigned long rtas_log_size; 38 39 static int surveillance_timeout = -1; 40 41 static unsigned int rtas_error_log_max; 42 static unsigned int rtas_error_log_buffer_max; 43 44 /* RTAS service tokens */ 45 static unsigned int event_scan; 46 static unsigned int rtas_event_scan_rate; 47 48 static bool full_rtas_msgs; 49 50 /* Stop logging to nvram after first fatal error */ 51 static int logging_enabled; /* Until we initialize everything, 52 * make sure we don't try logging 53 * anything */ 54 static int error_log_cnt; 55 56 /* 57 * Since we use 32 bit RTAS, the physical address of this must be below 58 * 4G or else bad things happen. Allocate this in the kernel data and 59 * make it big enough. 60 */ 61 static unsigned char logdata[RTAS_ERROR_LOG_MAX]; 62 63 static char *rtas_type[] = { 64 "Unknown", "Retry", "TCE Error", "Internal Device Failure", 65 "Timeout", "Data Parity", "Address Parity", "Cache Parity", 66 "Address Invalid", "ECC Uncorrected", "ECC Corrupted", 67 }; 68 69 static char *rtas_event_type(int type) 70 { 71 if ((type > 0) && (type < 11)) 72 return rtas_type[type]; 73 74 switch (type) { 75 case RTAS_TYPE_EPOW: 76 return "EPOW"; 77 case RTAS_TYPE_PLATFORM: 78 return "Platform Error"; 79 case RTAS_TYPE_IO: 80 return "I/O Event"; 81 case RTAS_TYPE_INFO: 82 return "Platform Information Event"; 83 case RTAS_TYPE_DEALLOC: 84 return "Resource Deallocation Event"; 85 case RTAS_TYPE_DUMP: 86 return "Dump Notification Event"; 87 case RTAS_TYPE_PRRN: 88 return "Platform Resource Reassignment Event"; 89 case RTAS_TYPE_HOTPLUG: 90 return "Hotplug Event"; 91 } 92 93 return rtas_type[0]; 94 } 95 96 /* To see this info, grep RTAS /var/log/messages and each entry 97 * will be collected together with obvious begin/end. 98 * There will be a unique identifier on the begin and end lines. 99 * This will persist across reboots. 100 * 101 * format of error logs returned from RTAS: 102 * bytes (size) : contents 103 * -------------------------------------------------------- 104 * 0-7 (8) : rtas_error_log 105 * 8-47 (40) : extended info 106 * 48-51 (4) : vendor id 107 * 52-1023 (vendor specific) : location code and debug data 108 */ 109 static void printk_log_rtas(char *buf, int len) 110 { 111 112 int i,j,n = 0; 113 int perline = 16; 114 char buffer[64]; 115 char * str = "RTAS event"; 116 117 if (full_rtas_msgs) { 118 printk(RTAS_DEBUG "%d -------- %s begin --------\n", 119 error_log_cnt, str); 120 121 /* 122 * Print perline bytes on each line, each line will start 123 * with RTAS and a changing number, so syslogd will 124 * print lines that are otherwise the same. Separate every 125 * 4 bytes with a space. 126 */ 127 for (i = 0; i < len; i++) { 128 j = i % perline; 129 if (j == 0) { 130 memset(buffer, 0, sizeof(buffer)); 131 n = sprintf(buffer, "RTAS %d:", i/perline); 132 } 133 134 if ((i % 4) == 0) 135 n += sprintf(buffer+n, " "); 136 137 n += sprintf(buffer+n, "%02x", (unsigned char)buf[i]); 138 139 if (j == (perline-1)) 140 printk(KERN_DEBUG "%s\n", buffer); 141 } 142 if ((i % perline) != 0) 143 printk(KERN_DEBUG "%s\n", buffer); 144 145 printk(RTAS_DEBUG "%d -------- %s end ----------\n", 146 error_log_cnt, str); 147 } else { 148 struct rtas_error_log *errlog = (struct rtas_error_log *)buf; 149 150 printk(RTAS_DEBUG "event: %d, Type: %s (%d), Severity: %d\n", 151 error_log_cnt, 152 rtas_event_type(rtas_error_type(errlog)), 153 rtas_error_type(errlog), 154 rtas_error_severity(errlog)); 155 } 156 } 157 158 static int log_rtas_len(char * buf) 159 { 160 int len; 161 struct rtas_error_log *err; 162 uint32_t extended_log_length; 163 164 /* rtas fixed header */ 165 len = 8; 166 err = (struct rtas_error_log *)buf; 167 extended_log_length = rtas_error_extended_log_length(err); 168 if (rtas_error_extended(err) && extended_log_length) { 169 170 /* extended header */ 171 len += extended_log_length; 172 } 173 174 if (rtas_error_log_max == 0) 175 rtas_error_log_max = rtas_get_error_log_max(); 176 177 if (len > rtas_error_log_max) 178 len = rtas_error_log_max; 179 180 return len; 181 } 182 183 /* 184 * First write to nvram, if fatal error, that is the only 185 * place we log the info. The error will be picked up 186 * on the next reboot by rtasd. If not fatal, run the 187 * method for the type of error. Currently, only RTAS 188 * errors have methods implemented, but in the future 189 * there might be a need to store data in nvram before a 190 * call to panic(). 191 * 192 * XXX We write to nvram periodically, to indicate error has 193 * been written and sync'd, but there is a possibility 194 * that if we don't shutdown correctly, a duplicate error 195 * record will be created on next reboot. 196 */ 197 void pSeries_log_error(char *buf, unsigned int err_type, int fatal) 198 { 199 unsigned long offset; 200 unsigned long s; 201 int len = 0; 202 203 pr_debug("rtasd: logging event\n"); 204 if (buf == NULL) 205 return; 206 207 spin_lock_irqsave(&rtasd_log_lock, s); 208 209 /* get length and increase count */ 210 switch (err_type & ERR_TYPE_MASK) { 211 case ERR_TYPE_RTAS_LOG: 212 len = log_rtas_len(buf); 213 if (!(err_type & ERR_FLAG_BOOT)) 214 error_log_cnt++; 215 break; 216 case ERR_TYPE_KERNEL_PANIC: 217 default: 218 WARN_ON_ONCE(!irqs_disabled()); /* @@@ DEBUG @@@ */ 219 spin_unlock_irqrestore(&rtasd_log_lock, s); 220 return; 221 } 222 223 #ifdef CONFIG_PPC64 224 /* Write error to NVRAM */ 225 if (logging_enabled && !(err_type & ERR_FLAG_BOOT)) 226 nvram_write_error_log(buf, len, err_type, error_log_cnt); 227 #endif /* CONFIG_PPC64 */ 228 229 /* 230 * rtas errors can occur during boot, and we do want to capture 231 * those somewhere, even if nvram isn't ready (why not?), and even 232 * if rtasd isn't ready. Put them into the boot log, at least. 233 */ 234 if ((err_type & ERR_TYPE_MASK) == ERR_TYPE_RTAS_LOG) 235 printk_log_rtas(buf, len); 236 237 /* Check to see if we need to or have stopped logging */ 238 if (fatal || !logging_enabled) { 239 logging_enabled = 0; 240 WARN_ON_ONCE(!irqs_disabled()); /* @@@ DEBUG @@@ */ 241 spin_unlock_irqrestore(&rtasd_log_lock, s); 242 return; 243 } 244 245 /* call type specific method for error */ 246 switch (err_type & ERR_TYPE_MASK) { 247 case ERR_TYPE_RTAS_LOG: 248 offset = rtas_error_log_buffer_max * 249 ((rtas_log_start+rtas_log_size) & LOG_NUMBER_MASK); 250 251 /* First copy over sequence number */ 252 memcpy(&rtas_log_buf[offset], (void *) &error_log_cnt, sizeof(int)); 253 254 /* Second copy over error log data */ 255 offset += sizeof(int); 256 memcpy(&rtas_log_buf[offset], buf, len); 257 258 if (rtas_log_size < LOG_NUMBER) 259 rtas_log_size += 1; 260 else 261 rtas_log_start += 1; 262 263 WARN_ON_ONCE(!irqs_disabled()); /* @@@ DEBUG @@@ */ 264 spin_unlock_irqrestore(&rtasd_log_lock, s); 265 wake_up_interruptible(&rtas_log_wait); 266 break; 267 case ERR_TYPE_KERNEL_PANIC: 268 default: 269 WARN_ON_ONCE(!irqs_disabled()); /* @@@ DEBUG @@@ */ 270 spin_unlock_irqrestore(&rtasd_log_lock, s); 271 return; 272 } 273 } 274 275 static void handle_rtas_event(const struct rtas_error_log *log) 276 { 277 if (!machine_is(pseries)) 278 return; 279 280 if (rtas_error_type(log) == RTAS_TYPE_PRRN) 281 pr_info_ratelimited("Platform resource reassignment ignored.\n"); 282 } 283 284 static int rtas_log_open(struct inode * inode, struct file * file) 285 { 286 return 0; 287 } 288 289 static int rtas_log_release(struct inode * inode, struct file * file) 290 { 291 return 0; 292 } 293 294 /* This will check if all events are logged, if they are then, we 295 * know that we can safely clear the events in NVRAM. 296 * Next we'll sit and wait for something else to log. 297 */ 298 static ssize_t rtas_log_read(struct file * file, char __user * buf, 299 size_t count, loff_t *ppos) 300 { 301 int error; 302 char *tmp; 303 unsigned long s; 304 unsigned long offset; 305 306 if (!buf || count < rtas_error_log_buffer_max) 307 return -EINVAL; 308 309 count = rtas_error_log_buffer_max; 310 311 if (!access_ok(buf, count)) 312 return -EFAULT; 313 314 tmp = kmalloc(count, GFP_KERNEL); 315 if (!tmp) 316 return -ENOMEM; 317 318 spin_lock_irqsave(&rtasd_log_lock, s); 319 320 /* if it's 0, then we know we got the last one (the one in NVRAM) */ 321 while (rtas_log_size == 0) { 322 if (file->f_flags & O_NONBLOCK) { 323 spin_unlock_irqrestore(&rtasd_log_lock, s); 324 error = -EAGAIN; 325 goto out; 326 } 327 328 if (!logging_enabled) { 329 spin_unlock_irqrestore(&rtasd_log_lock, s); 330 error = -ENODATA; 331 goto out; 332 } 333 #ifdef CONFIG_PPC64 334 nvram_clear_error_log(); 335 #endif /* CONFIG_PPC64 */ 336 337 spin_unlock_irqrestore(&rtasd_log_lock, s); 338 error = wait_event_interruptible(rtas_log_wait, rtas_log_size); 339 if (error) 340 goto out; 341 spin_lock_irqsave(&rtasd_log_lock, s); 342 } 343 344 offset = rtas_error_log_buffer_max * (rtas_log_start & LOG_NUMBER_MASK); 345 memcpy(tmp, &rtas_log_buf[offset], count); 346 347 rtas_log_start += 1; 348 rtas_log_size -= 1; 349 spin_unlock_irqrestore(&rtasd_log_lock, s); 350 351 error = copy_to_user(buf, tmp, count) ? -EFAULT : count; 352 out: 353 kfree(tmp); 354 return error; 355 } 356 357 static __poll_t rtas_log_poll(struct file *file, poll_table * wait) 358 { 359 poll_wait(file, &rtas_log_wait, wait); 360 if (rtas_log_size) 361 return EPOLLIN | EPOLLRDNORM; 362 return 0; 363 } 364 365 static const struct proc_ops rtas_log_proc_ops = { 366 .proc_read = rtas_log_read, 367 .proc_poll = rtas_log_poll, 368 .proc_open = rtas_log_open, 369 .proc_release = rtas_log_release, 370 .proc_lseek = noop_llseek, 371 }; 372 373 static int enable_surveillance(int timeout) 374 { 375 int error; 376 377 error = rtas_set_indicator(SURVEILLANCE_TOKEN, 0, timeout); 378 379 if (error == 0) 380 return 0; 381 382 if (error == -EINVAL) { 383 printk(KERN_DEBUG "rtasd: surveillance not supported\n"); 384 return 0; 385 } 386 387 printk(KERN_ERR "rtasd: could not update surveillance\n"); 388 return -1; 389 } 390 391 static void do_event_scan(void) 392 { 393 int error; 394 do { 395 memset(logdata, 0, rtas_error_log_max); 396 error = rtas_call(event_scan, 4, 1, NULL, 397 RTAS_EVENT_SCAN_ALL_EVENTS, 0, 398 __pa(logdata), rtas_error_log_max); 399 if (error == -1) { 400 printk(KERN_ERR "event-scan failed\n"); 401 break; 402 } 403 404 if (error == 0) { 405 if (rtas_error_type((struct rtas_error_log *)logdata) != 406 RTAS_TYPE_PRRN) 407 pSeries_log_error(logdata, ERR_TYPE_RTAS_LOG, 408 0); 409 handle_rtas_event((struct rtas_error_log *)logdata); 410 } 411 412 } while(error == 0); 413 } 414 415 static void rtas_event_scan(struct work_struct *w); 416 static DECLARE_DELAYED_WORK(event_scan_work, rtas_event_scan); 417 418 /* 419 * Delay should be at least one second since some machines have problems if 420 * we call event-scan too quickly. 421 */ 422 static unsigned long event_scan_delay = 1*HZ; 423 static int first_pass = 1; 424 425 static void rtas_event_scan(struct work_struct *w) 426 { 427 unsigned int cpu; 428 429 do_event_scan(); 430 431 cpus_read_lock(); 432 433 /* raw_ OK because just using CPU as starting point. */ 434 cpu = cpumask_next(raw_smp_processor_id(), cpu_online_mask); 435 if (cpu >= nr_cpu_ids) { 436 cpu = cpumask_first(cpu_online_mask); 437 438 if (first_pass) { 439 first_pass = 0; 440 event_scan_delay = 30*HZ/rtas_event_scan_rate; 441 442 if (surveillance_timeout != -1) { 443 pr_debug("rtasd: enabling surveillance\n"); 444 enable_surveillance(surveillance_timeout); 445 pr_debug("rtasd: surveillance enabled\n"); 446 } 447 } 448 } 449 450 schedule_delayed_work_on(cpu, &event_scan_work, 451 __round_jiffies_relative(event_scan_delay, cpu)); 452 453 cpus_read_unlock(); 454 } 455 456 #ifdef CONFIG_PPC64 457 static void __init retrieve_nvram_error_log(void) 458 { 459 unsigned int err_type ; 460 int rc ; 461 462 /* See if we have any error stored in NVRAM */ 463 memset(logdata, 0, rtas_error_log_max); 464 rc = nvram_read_error_log(logdata, rtas_error_log_max, 465 &err_type, &error_log_cnt); 466 /* We can use rtas_log_buf now */ 467 logging_enabled = 1; 468 if (!rc) { 469 if (err_type != ERR_FLAG_ALREADY_LOGGED) { 470 pSeries_log_error(logdata, err_type | ERR_FLAG_BOOT, 0); 471 } 472 } 473 } 474 #else /* CONFIG_PPC64 */ 475 static void __init retrieve_nvram_error_log(void) 476 { 477 } 478 #endif /* CONFIG_PPC64 */ 479 480 static void __init start_event_scan(void) 481 { 482 printk(KERN_DEBUG "RTAS daemon started\n"); 483 pr_debug("rtasd: will sleep for %d milliseconds\n", 484 (30000 / rtas_event_scan_rate)); 485 486 /* Retrieve errors from nvram if any */ 487 retrieve_nvram_error_log(); 488 489 schedule_delayed_work_on(cpumask_first(cpu_online_mask), 490 &event_scan_work, event_scan_delay); 491 } 492 493 /* Cancel the rtas event scan work */ 494 void rtas_cancel_event_scan(void) 495 { 496 cancel_delayed_work_sync(&event_scan_work); 497 } 498 EXPORT_SYMBOL_GPL(rtas_cancel_event_scan); 499 500 static int __init rtas_event_scan_init(void) 501 { 502 if (!machine_is(pseries) && !machine_is(chrp)) 503 return 0; 504 505 /* No RTAS */ 506 event_scan = rtas_token("event-scan"); 507 if (event_scan == RTAS_UNKNOWN_SERVICE) { 508 printk(KERN_INFO "rtasd: No event-scan on system\n"); 509 return -ENODEV; 510 } 511 512 rtas_event_scan_rate = rtas_token("rtas-event-scan-rate"); 513 if (rtas_event_scan_rate == RTAS_UNKNOWN_SERVICE) { 514 printk(KERN_ERR "rtasd: no rtas-event-scan-rate on system\n"); 515 return -ENODEV; 516 } 517 518 if (!rtas_event_scan_rate) { 519 /* Broken firmware: take a rate of zero to mean don't scan */ 520 printk(KERN_DEBUG "rtasd: scan rate is 0, not scanning\n"); 521 return 0; 522 } 523 524 /* Make room for the sequence number */ 525 rtas_error_log_max = rtas_get_error_log_max(); 526 rtas_error_log_buffer_max = rtas_error_log_max + sizeof(int); 527 528 rtas_log_buf = vmalloc(array_size(LOG_NUMBER, 529 rtas_error_log_buffer_max)); 530 if (!rtas_log_buf) { 531 printk(KERN_ERR "rtasd: no memory\n"); 532 return -ENOMEM; 533 } 534 535 start_event_scan(); 536 537 return 0; 538 } 539 arch_initcall(rtas_event_scan_init); 540 541 static int __init rtas_init(void) 542 { 543 struct proc_dir_entry *entry; 544 545 if (!machine_is(pseries) && !machine_is(chrp)) 546 return 0; 547 548 if (!rtas_log_buf) 549 return -ENODEV; 550 551 entry = proc_create("powerpc/rtas/error_log", 0400, NULL, 552 &rtas_log_proc_ops); 553 if (!entry) 554 printk(KERN_ERR "Failed to create error_log proc entry\n"); 555 556 return 0; 557 } 558 __initcall(rtas_init); 559 560 static int __init surveillance_setup(char *str) 561 { 562 int i; 563 564 /* We only do surveillance on pseries */ 565 if (!machine_is(pseries)) 566 return 0; 567 568 if (get_option(&str,&i)) { 569 if (i >= 0 && i <= 255) 570 surveillance_timeout = i; 571 } 572 573 return 1; 574 } 575 __setup("surveillance=", surveillance_setup); 576 577 static int __init rtasmsgs_setup(char *str) 578 { 579 return (kstrtobool(str, &full_rtas_msgs) == 0); 580 } 581 __setup("rtasmsgs=", rtasmsgs_setup); 582