| 1 | // SPDX-License-Identifier: GPL-2.0-only |
| 2 | /* |
| 3 | * linux/kernel/panic.c |
| 4 | * |
| 5 | * Copyright (C) 1991, 1992 Linus Torvalds |
| 6 | */ |
| 7 | |
| 8 | /* |
| 9 | * This function is used through-out the kernel (including mm and fs) |
| 10 | * to indicate a major problem. |
| 11 | */ |
| 12 | #include <linux/debug_locks.h> |
| 13 | #include <linux/sched/debug.h> |
| 14 | #include <linux/interrupt.h> |
| 15 | #include <linux/kgdb.h> |
| 16 | #include <linux/kmsg_dump.h> |
| 17 | #include <linux/kallsyms.h> |
| 18 | #include <linux/notifier.h> |
| 19 | #include <linux/vt_kern.h> |
| 20 | #include <linux/module.h> |
| 21 | #include <linux/random.h> |
| 22 | #include <linux/ftrace.h> |
| 23 | #include <linux/reboot.h> |
| 24 | #include <linux/delay.h> |
| 25 | #include <linux/kexec.h> |
| 26 | #include <linux/panic_notifier.h> |
| 27 | #include <linux/sched.h> |
| 28 | #include <linux/string_helpers.h> |
| 29 | #include <linux/sysrq.h> |
| 30 | #include <linux/init.h> |
| 31 | #include <linux/nmi.h> |
| 32 | #include <linux/console.h> |
| 33 | #include <linux/bug.h> |
| 34 | #include <linux/ratelimit.h> |
| 35 | #include <linux/debugfs.h> |
| 36 | #include <linux/sysfs.h> |
| 37 | #include <linux/context_tracking.h> |
| 38 | #include <linux/seq_buf.h> |
| 39 | #include <linux/sys_info.h> |
| 40 | #include <trace/events/error_report.h> |
| 41 | #include <asm/sections.h> |
| 42 | |
| 43 | #define PANIC_TIMER_STEP 100 |
| 44 | #define PANIC_BLINK_SPD 18 |
| 45 | |
| 46 | #ifdef CONFIG_SMP |
| 47 | /* |
| 48 | * Should we dump all CPUs backtraces in an oops event? |
| 49 | * Defaults to 0, can be changed via sysctl. |
| 50 | */ |
| 51 | static unsigned int __read_mostly sysctl_oops_all_cpu_backtrace; |
| 52 | #else |
| 53 | #define sysctl_oops_all_cpu_backtrace 0 |
| 54 | #endif /* CONFIG_SMP */ |
| 55 | |
| 56 | int panic_on_oops = IS_ENABLED(CONFIG_PANIC_ON_OOPS); |
| 57 | static unsigned long tainted_mask = |
| 58 | IS_ENABLED(CONFIG_RANDSTRUCT) ? (1 << TAINT_RANDSTRUCT) : 0; |
| 59 | static int pause_on_oops; |
| 60 | static int pause_on_oops_flag; |
| 61 | static DEFINE_SPINLOCK(pause_on_oops_lock); |
| 62 | bool crash_kexec_post_notifiers; |
| 63 | int panic_on_warn __read_mostly; |
| 64 | unsigned long panic_on_taint; |
| 65 | bool panic_on_taint_nousertaint = false; |
| 66 | static unsigned int warn_limit __read_mostly; |
| 67 | static bool panic_console_replay; |
| 68 | |
| 69 | bool panic_triggering_all_cpu_backtrace; |
| 70 | static bool panic_this_cpu_backtrace_printed; |
| 71 | |
| 72 | int panic_timeout = CONFIG_PANIC_TIMEOUT; |
| 73 | EXPORT_SYMBOL_GPL(panic_timeout); |
| 74 | |
| 75 | unsigned long panic_print; |
| 76 | |
| 77 | ATOMIC_NOTIFIER_HEAD(panic_notifier_list); |
| 78 | |
| 79 | EXPORT_SYMBOL(panic_notifier_list); |
| 80 | |
| 81 | static void panic_print_deprecated(void) |
| 82 | { |
| 83 | pr_info_once("Kernel: The 'panic_print' parameter is now deprecated. Please use 'panic_sys_info' and 'panic_console_replay' instead.\n" ); |
| 84 | } |
| 85 | |
| 86 | #ifdef CONFIG_SYSCTL |
| 87 | |
| 88 | /* |
| 89 | * Taint values can only be increased |
| 90 | * This means we can safely use a temporary. |
| 91 | */ |
| 92 | static int proc_taint(const struct ctl_table *table, int write, |
| 93 | void *buffer, size_t *lenp, loff_t *ppos) |
| 94 | { |
| 95 | struct ctl_table t; |
| 96 | unsigned long tmptaint = get_taint(); |
| 97 | int err; |
| 98 | |
| 99 | if (write && !capable(CAP_SYS_ADMIN)) |
| 100 | return -EPERM; |
| 101 | |
| 102 | t = *table; |
| 103 | t.data = &tmptaint; |
| 104 | err = proc_doulongvec_minmax(&t, write, buffer, lenp, ppos); |
| 105 | if (err < 0) |
| 106 | return err; |
| 107 | |
| 108 | if (write) { |
| 109 | int i; |
| 110 | |
| 111 | /* |
| 112 | * If we are relying on panic_on_taint not producing |
| 113 | * false positives due to userspace input, bail out |
| 114 | * before setting the requested taint flags. |
| 115 | */ |
| 116 | if (panic_on_taint_nousertaint && (tmptaint & panic_on_taint)) |
| 117 | return -EINVAL; |
| 118 | |
| 119 | /* |
| 120 | * Poor man's atomic or. Not worth adding a primitive |
| 121 | * to everyone's atomic.h for this |
| 122 | */ |
| 123 | for (i = 0; i < TAINT_FLAGS_COUNT; i++) |
| 124 | if ((1UL << i) & tmptaint) |
| 125 | add_taint(flag: i, LOCKDEP_STILL_OK); |
| 126 | } |
| 127 | |
| 128 | return err; |
| 129 | } |
| 130 | |
| 131 | static int sysctl_panic_print_handler(const struct ctl_table *table, int write, |
| 132 | void *buffer, size_t *lenp, loff_t *ppos) |
| 133 | { |
| 134 | panic_print_deprecated(); |
| 135 | return proc_doulongvec_minmax(table, write, buffer, lenp, ppos); |
| 136 | } |
| 137 | |
| 138 | static const struct ctl_table kern_panic_table[] = { |
| 139 | #ifdef CONFIG_SMP |
| 140 | { |
| 141 | .procname = "oops_all_cpu_backtrace" , |
| 142 | .data = &sysctl_oops_all_cpu_backtrace, |
| 143 | .maxlen = sizeof(int), |
| 144 | .mode = 0644, |
| 145 | .proc_handler = proc_dointvec_minmax, |
| 146 | .extra1 = SYSCTL_ZERO, |
| 147 | .extra2 = SYSCTL_ONE, |
| 148 | }, |
| 149 | #endif |
| 150 | { |
| 151 | .procname = "tainted" , |
| 152 | .maxlen = sizeof(long), |
| 153 | .mode = 0644, |
| 154 | .proc_handler = proc_taint, |
| 155 | }, |
| 156 | { |
| 157 | .procname = "panic" , |
| 158 | .data = &panic_timeout, |
| 159 | .maxlen = sizeof(int), |
| 160 | .mode = 0644, |
| 161 | .proc_handler = proc_dointvec, |
| 162 | }, |
| 163 | { |
| 164 | .procname = "panic_on_oops" , |
| 165 | .data = &panic_on_oops, |
| 166 | .maxlen = sizeof(int), |
| 167 | .mode = 0644, |
| 168 | .proc_handler = proc_dointvec, |
| 169 | }, |
| 170 | { |
| 171 | .procname = "panic_print" , |
| 172 | .data = &panic_print, |
| 173 | .maxlen = sizeof(unsigned long), |
| 174 | .mode = 0644, |
| 175 | .proc_handler = sysctl_panic_print_handler, |
| 176 | }, |
| 177 | { |
| 178 | .procname = "panic_on_warn" , |
| 179 | .data = &panic_on_warn, |
| 180 | .maxlen = sizeof(int), |
| 181 | .mode = 0644, |
| 182 | .proc_handler = proc_dointvec_minmax, |
| 183 | .extra1 = SYSCTL_ZERO, |
| 184 | .extra2 = SYSCTL_ONE, |
| 185 | }, |
| 186 | { |
| 187 | .procname = "warn_limit" , |
| 188 | .data = &warn_limit, |
| 189 | .maxlen = sizeof(warn_limit), |
| 190 | .mode = 0644, |
| 191 | .proc_handler = proc_douintvec, |
| 192 | }, |
| 193 | #if (defined(CONFIG_X86_32) || defined(CONFIG_PARISC)) && \ |
| 194 | defined(CONFIG_DEBUG_STACKOVERFLOW) |
| 195 | { |
| 196 | .procname = "panic_on_stackoverflow" , |
| 197 | .data = &sysctl_panic_on_stackoverflow, |
| 198 | .maxlen = sizeof(int), |
| 199 | .mode = 0644, |
| 200 | .proc_handler = proc_dointvec, |
| 201 | }, |
| 202 | #endif |
| 203 | { |
| 204 | .procname = "panic_sys_info" , |
| 205 | .data = &panic_print, |
| 206 | .maxlen = sizeof(panic_print), |
| 207 | .mode = 0644, |
| 208 | .proc_handler = sysctl_sys_info_handler, |
| 209 | }, |
| 210 | }; |
| 211 | |
| 212 | static __init int kernel_panic_sysctls_init(void) |
| 213 | { |
| 214 | register_sysctl_init("kernel" , kern_panic_table); |
| 215 | return 0; |
| 216 | } |
| 217 | late_initcall(kernel_panic_sysctls_init); |
| 218 | #endif |
| 219 | |
| 220 | /* The format is "panic_sys_info=tasks,mem,locks,ftrace,..." */ |
| 221 | static int __init setup_panic_sys_info(char *buf) |
| 222 | { |
| 223 | /* There is no risk of race in kernel boot phase */ |
| 224 | panic_print = sys_info_parse_param(str: buf); |
| 225 | return 1; |
| 226 | } |
| 227 | __setup("panic_sys_info=" , setup_panic_sys_info); |
| 228 | |
| 229 | static atomic_t warn_count = ATOMIC_INIT(0); |
| 230 | |
| 231 | #ifdef CONFIG_SYSFS |
| 232 | static ssize_t warn_count_show(struct kobject *kobj, struct kobj_attribute *attr, |
| 233 | char *page) |
| 234 | { |
| 235 | return sysfs_emit(buf: page, fmt: "%d\n" , atomic_read(v: &warn_count)); |
| 236 | } |
| 237 | |
| 238 | static struct kobj_attribute warn_count_attr = __ATTR_RO(warn_count); |
| 239 | |
| 240 | static __init int kernel_panic_sysfs_init(void) |
| 241 | { |
| 242 | sysfs_add_file_to_group(kobj: kernel_kobj, attr: &warn_count_attr.attr, NULL); |
| 243 | return 0; |
| 244 | } |
| 245 | late_initcall(kernel_panic_sysfs_init); |
| 246 | #endif |
| 247 | |
| 248 | static long no_blink(int state) |
| 249 | { |
| 250 | return 0; |
| 251 | } |
| 252 | |
| 253 | /* Returns how long it waited in ms */ |
| 254 | long (*panic_blink)(int state); |
| 255 | EXPORT_SYMBOL(panic_blink); |
| 256 | |
| 257 | /* |
| 258 | * Stop ourself in panic -- architecture code may override this |
| 259 | */ |
| 260 | void __weak __noreturn panic_smp_self_stop(void) |
| 261 | { |
| 262 | while (1) |
| 263 | cpu_relax(); |
| 264 | } |
| 265 | |
| 266 | /* |
| 267 | * Stop ourselves in NMI context if another CPU has already panicked. Arch code |
| 268 | * may override this to prepare for crash dumping, e.g. save regs info. |
| 269 | */ |
| 270 | void __weak __noreturn nmi_panic_self_stop(struct pt_regs *regs) |
| 271 | { |
| 272 | panic_smp_self_stop(); |
| 273 | } |
| 274 | |
| 275 | /* |
| 276 | * Stop other CPUs in panic. Architecture dependent code may override this |
| 277 | * with more suitable version. For example, if the architecture supports |
| 278 | * crash dump, it should save registers of each stopped CPU and disable |
| 279 | * per-CPU features such as virtualization extensions. |
| 280 | */ |
| 281 | void __weak crash_smp_send_stop(void) |
| 282 | { |
| 283 | static int cpus_stopped; |
| 284 | |
| 285 | /* |
| 286 | * This function can be called twice in panic path, but obviously |
| 287 | * we execute this only once. |
| 288 | */ |
| 289 | if (cpus_stopped) |
| 290 | return; |
| 291 | |
| 292 | /* |
| 293 | * Note smp_send_stop is the usual smp shutdown function, which |
| 294 | * unfortunately means it may not be hardened to work in a panic |
| 295 | * situation. |
| 296 | */ |
| 297 | smp_send_stop(); |
| 298 | cpus_stopped = 1; |
| 299 | } |
| 300 | |
| 301 | atomic_t panic_cpu = ATOMIC_INIT(PANIC_CPU_INVALID); |
| 302 | |
| 303 | bool panic_try_start(void) |
| 304 | { |
| 305 | int old_cpu, this_cpu; |
| 306 | |
| 307 | /* |
| 308 | * Only one CPU is allowed to execute the crash_kexec() code as with |
| 309 | * panic(). Otherwise parallel calls of panic() and crash_kexec() |
| 310 | * may stop each other. To exclude them, we use panic_cpu here too. |
| 311 | */ |
| 312 | old_cpu = PANIC_CPU_INVALID; |
| 313 | this_cpu = raw_smp_processor_id(); |
| 314 | |
| 315 | return atomic_try_cmpxchg(v: &panic_cpu, old: &old_cpu, new: this_cpu); |
| 316 | } |
| 317 | EXPORT_SYMBOL(panic_try_start); |
| 318 | |
| 319 | void panic_reset(void) |
| 320 | { |
| 321 | atomic_set(v: &panic_cpu, PANIC_CPU_INVALID); |
| 322 | } |
| 323 | EXPORT_SYMBOL(panic_reset); |
| 324 | |
| 325 | bool panic_in_progress(void) |
| 326 | { |
| 327 | return unlikely(atomic_read(&panic_cpu) != PANIC_CPU_INVALID); |
| 328 | } |
| 329 | EXPORT_SYMBOL(panic_in_progress); |
| 330 | |
| 331 | /* Return true if a panic is in progress on the current CPU. */ |
| 332 | bool panic_on_this_cpu(void) |
| 333 | { |
| 334 | /* |
| 335 | * We can use raw_smp_processor_id() here because it is impossible for |
| 336 | * the task to be migrated to the panic_cpu, or away from it. If |
| 337 | * panic_cpu has already been set, and we're not currently executing on |
| 338 | * that CPU, then we never will be. |
| 339 | */ |
| 340 | return unlikely(atomic_read(&panic_cpu) == raw_smp_processor_id()); |
| 341 | } |
| 342 | EXPORT_SYMBOL(panic_on_this_cpu); |
| 343 | |
| 344 | /* |
| 345 | * Return true if a panic is in progress on a remote CPU. |
| 346 | * |
| 347 | * On true, the local CPU should immediately release any printing resources |
| 348 | * that may be needed by the panic CPU. |
| 349 | */ |
| 350 | bool panic_on_other_cpu(void) |
| 351 | { |
| 352 | return (panic_in_progress() && !panic_on_this_cpu()); |
| 353 | } |
| 354 | EXPORT_SYMBOL(panic_on_other_cpu); |
| 355 | |
| 356 | /* |
| 357 | * A variant of panic() called from NMI context. We return if we've already |
| 358 | * panicked on this CPU. If another CPU already panicked, loop in |
| 359 | * nmi_panic_self_stop() which can provide architecture dependent code such |
| 360 | * as saving register state for crash dump. |
| 361 | */ |
| 362 | void nmi_panic(struct pt_regs *regs, const char *msg) |
| 363 | { |
| 364 | if (panic_try_start()) |
| 365 | panic(fmt: "%s" , msg); |
| 366 | else if (panic_on_other_cpu()) |
| 367 | nmi_panic_self_stop(regs); |
| 368 | } |
| 369 | EXPORT_SYMBOL(nmi_panic); |
| 370 | |
| 371 | void check_panic_on_warn(const char *origin) |
| 372 | { |
| 373 | unsigned int limit; |
| 374 | |
| 375 | if (panic_on_warn) |
| 376 | panic(fmt: "%s: panic_on_warn set ...\n" , origin); |
| 377 | |
| 378 | limit = READ_ONCE(warn_limit); |
| 379 | if (atomic_inc_return(v: &warn_count) >= limit && limit) |
| 380 | panic(fmt: "%s: system warned too often (kernel.warn_limit is %d)" , |
| 381 | origin, limit); |
| 382 | } |
| 383 | |
| 384 | static void panic_trigger_all_cpu_backtrace(void) |
| 385 | { |
| 386 | /* Temporary allow non-panic CPUs to write their backtraces. */ |
| 387 | panic_triggering_all_cpu_backtrace = true; |
| 388 | |
| 389 | if (panic_this_cpu_backtrace_printed) |
| 390 | trigger_allbutcpu_cpu_backtrace(raw_smp_processor_id()); |
| 391 | else |
| 392 | trigger_all_cpu_backtrace(); |
| 393 | |
| 394 | panic_triggering_all_cpu_backtrace = false; |
| 395 | } |
| 396 | |
| 397 | /* |
| 398 | * Helper that triggers the NMI backtrace (if set in panic_print) |
| 399 | * and then performs the secondary CPUs shutdown - we cannot have |
| 400 | * the NMI backtrace after the CPUs are off! |
| 401 | */ |
| 402 | static void panic_other_cpus_shutdown(bool crash_kexec) |
| 403 | { |
| 404 | if (panic_print & SYS_INFO_ALL_CPU_BT) |
| 405 | panic_trigger_all_cpu_backtrace(); |
| 406 | |
| 407 | /* |
| 408 | * Note that smp_send_stop() is the usual SMP shutdown function, |
| 409 | * which unfortunately may not be hardened to work in a panic |
| 410 | * situation. If we want to do crash dump after notifier calls |
| 411 | * and kmsg_dump, we will need architecture dependent extra |
| 412 | * bits in addition to stopping other CPUs, hence we rely on |
| 413 | * crash_smp_send_stop() for that. |
| 414 | */ |
| 415 | if (!crash_kexec) |
| 416 | smp_send_stop(); |
| 417 | else |
| 418 | crash_smp_send_stop(); |
| 419 | } |
| 420 | |
| 421 | /** |
| 422 | * vpanic - halt the system |
| 423 | * @fmt: The text string to print |
| 424 | * @args: Arguments for the format string |
| 425 | * |
| 426 | * Display a message, then perform cleanups. This function never returns. |
| 427 | */ |
| 428 | void vpanic(const char *fmt, va_list args) |
| 429 | { |
| 430 | static char buf[1024]; |
| 431 | long i, i_next = 0, len; |
| 432 | int state = 0; |
| 433 | bool _crash_kexec_post_notifiers = crash_kexec_post_notifiers; |
| 434 | |
| 435 | if (panic_on_warn) { |
| 436 | /* |
| 437 | * This thread may hit another WARN() in the panic path. |
| 438 | * Resetting this prevents additional WARN() from panicking the |
| 439 | * system on this thread. Other threads are blocked by the |
| 440 | * panic_mutex in panic(). |
| 441 | */ |
| 442 | panic_on_warn = 0; |
| 443 | } |
| 444 | |
| 445 | /* |
| 446 | * Disable local interrupts. This will prevent panic_smp_self_stop |
| 447 | * from deadlocking the first cpu that invokes the panic, since |
| 448 | * there is nothing to prevent an interrupt handler (that runs |
| 449 | * after setting panic_cpu) from invoking panic() again. |
| 450 | */ |
| 451 | local_irq_disable(); |
| 452 | preempt_disable_notrace(); |
| 453 | |
| 454 | /* |
| 455 | * It's possible to come here directly from a panic-assertion and |
| 456 | * not have preempt disabled. Some functions called from here want |
| 457 | * preempt to be disabled. No point enabling it later though... |
| 458 | * |
| 459 | * Only one CPU is allowed to execute the panic code from here. For |
| 460 | * multiple parallel invocations of panic, all other CPUs either |
| 461 | * stop themself or will wait until they are stopped by the 1st CPU |
| 462 | * with smp_send_stop(). |
| 463 | * |
| 464 | * cmpxchg success means this is the 1st CPU which comes here, |
| 465 | * so go ahead. |
| 466 | * `old_cpu == this_cpu' means we came from nmi_panic() which sets |
| 467 | * panic_cpu to this CPU. In this case, this is also the 1st CPU. |
| 468 | */ |
| 469 | /* atomic_try_cmpxchg updates old_cpu on failure */ |
| 470 | if (panic_try_start()) { |
| 471 | /* go ahead */ |
| 472 | } else if (panic_on_other_cpu()) |
| 473 | panic_smp_self_stop(); |
| 474 | |
| 475 | console_verbose(); |
| 476 | bust_spinlocks(yes: 1); |
| 477 | len = vscnprintf(buf, size: sizeof(buf), fmt, args); |
| 478 | |
| 479 | if (len && buf[len - 1] == '\n') |
| 480 | buf[len - 1] = '\0'; |
| 481 | |
| 482 | pr_emerg("Kernel panic - not syncing: %s\n" , buf); |
| 483 | /* |
| 484 | * Avoid nested stack-dumping if a panic occurs during oops processing |
| 485 | */ |
| 486 | if (test_taint(TAINT_DIE) || oops_in_progress > 1) { |
| 487 | panic_this_cpu_backtrace_printed = true; |
| 488 | } else if (IS_ENABLED(CONFIG_DEBUG_BUGVERBOSE)) { |
| 489 | dump_stack(); |
| 490 | panic_this_cpu_backtrace_printed = true; |
| 491 | } |
| 492 | |
| 493 | /* |
| 494 | * If kgdb is enabled, give it a chance to run before we stop all |
| 495 | * the other CPUs or else we won't be able to debug processes left |
| 496 | * running on them. |
| 497 | */ |
| 498 | kgdb_panic(msg: buf); |
| 499 | |
| 500 | /* |
| 501 | * If we have crashed and we have a crash kernel loaded let it handle |
| 502 | * everything else. |
| 503 | * If we want to run this after calling panic_notifiers, pass |
| 504 | * the "crash_kexec_post_notifiers" option to the kernel. |
| 505 | * |
| 506 | * Bypass the panic_cpu check and call __crash_kexec directly. |
| 507 | */ |
| 508 | if (!_crash_kexec_post_notifiers) |
| 509 | __crash_kexec(NULL); |
| 510 | |
| 511 | panic_other_cpus_shutdown(crash_kexec: _crash_kexec_post_notifiers); |
| 512 | |
| 513 | printk_legacy_allow_panic_sync(); |
| 514 | |
| 515 | /* |
| 516 | * Run any panic handlers, including those that might need to |
| 517 | * add information to the kmsg dump output. |
| 518 | */ |
| 519 | atomic_notifier_call_chain(nh: &panic_notifier_list, val: 0, v: buf); |
| 520 | |
| 521 | sys_info(si_mask: panic_print); |
| 522 | |
| 523 | kmsg_dump_desc(reason: KMSG_DUMP_PANIC, desc: buf); |
| 524 | |
| 525 | /* |
| 526 | * If you doubt kdump always works fine in any situation, |
| 527 | * "crash_kexec_post_notifiers" offers you a chance to run |
| 528 | * panic_notifiers and dumping kmsg before kdump. |
| 529 | * Note: since some panic_notifiers can make crashed kernel |
| 530 | * more unstable, it can increase risks of the kdump failure too. |
| 531 | * |
| 532 | * Bypass the panic_cpu check and call __crash_kexec directly. |
| 533 | */ |
| 534 | if (_crash_kexec_post_notifiers) |
| 535 | __crash_kexec(NULL); |
| 536 | |
| 537 | console_unblank(); |
| 538 | |
| 539 | /* |
| 540 | * We may have ended up stopping the CPU holding the lock (in |
| 541 | * smp_send_stop()) while still having some valuable data in the console |
| 542 | * buffer. Try to acquire the lock then release it regardless of the |
| 543 | * result. The release will also print the buffers out. Locks debug |
| 544 | * should be disabled to avoid reporting bad unlock balance when |
| 545 | * panic() is not being callled from OOPS. |
| 546 | */ |
| 547 | debug_locks_off(); |
| 548 | console_flush_on_panic(mode: CONSOLE_FLUSH_PENDING); |
| 549 | |
| 550 | if ((panic_print & SYS_INFO_PANIC_CONSOLE_REPLAY) || |
| 551 | panic_console_replay) |
| 552 | console_flush_on_panic(mode: CONSOLE_REPLAY_ALL); |
| 553 | |
| 554 | if (!panic_blink) |
| 555 | panic_blink = no_blink; |
| 556 | |
| 557 | if (panic_timeout > 0) { |
| 558 | /* |
| 559 | * Delay timeout seconds before rebooting the machine. |
| 560 | * We can't use the "normal" timers since we just panicked. |
| 561 | */ |
| 562 | pr_emerg("Rebooting in %d seconds..\n" , panic_timeout); |
| 563 | |
| 564 | for (i = 0; i < panic_timeout * 1000; i += PANIC_TIMER_STEP) { |
| 565 | touch_nmi_watchdog(); |
| 566 | if (i >= i_next) { |
| 567 | i += panic_blink(state ^= 1); |
| 568 | i_next = i + 3600 / PANIC_BLINK_SPD; |
| 569 | } |
| 570 | mdelay(PANIC_TIMER_STEP); |
| 571 | } |
| 572 | } |
| 573 | if (panic_timeout != 0) { |
| 574 | /* |
| 575 | * This will not be a clean reboot, with everything |
| 576 | * shutting down. But if there is a chance of |
| 577 | * rebooting the system it will be rebooted. |
| 578 | */ |
| 579 | if (panic_reboot_mode != REBOOT_UNDEFINED) |
| 580 | reboot_mode = panic_reboot_mode; |
| 581 | emergency_restart(); |
| 582 | } |
| 583 | #ifdef __sparc__ |
| 584 | { |
| 585 | extern int stop_a_enabled; |
| 586 | /* Make sure the user can actually press Stop-A (L1-A) */ |
| 587 | stop_a_enabled = 1; |
| 588 | pr_emerg("Press Stop-A (L1-A) from sun keyboard or send break\n" |
| 589 | "twice on console to return to the boot prom\n" ); |
| 590 | } |
| 591 | #endif |
| 592 | #if defined(CONFIG_S390) |
| 593 | disabled_wait(); |
| 594 | #endif |
| 595 | pr_emerg("---[ end Kernel panic - not syncing: %s ]---\n" , buf); |
| 596 | |
| 597 | /* Do not scroll important messages printed above */ |
| 598 | suppress_printk = 1; |
| 599 | |
| 600 | /* |
| 601 | * The final messages may not have been printed if in a context that |
| 602 | * defers printing (such as NMI) and irq_work is not available. |
| 603 | * Explicitly flush the kernel log buffer one last time. |
| 604 | */ |
| 605 | console_flush_on_panic(mode: CONSOLE_FLUSH_PENDING); |
| 606 | nbcon_atomic_flush_unsafe(); |
| 607 | |
| 608 | local_irq_enable(); |
| 609 | for (i = 0; ; i += PANIC_TIMER_STEP) { |
| 610 | touch_softlockup_watchdog(); |
| 611 | if (i >= i_next) { |
| 612 | i += panic_blink(state ^= 1); |
| 613 | i_next = i + 3600 / PANIC_BLINK_SPD; |
| 614 | } |
| 615 | mdelay(PANIC_TIMER_STEP); |
| 616 | } |
| 617 | } |
| 618 | EXPORT_SYMBOL(vpanic); |
| 619 | |
| 620 | /* Identical to vpanic(), except it takes variadic arguments instead of va_list */ |
| 621 | void panic(const char *fmt, ...) |
| 622 | { |
| 623 | va_list args; |
| 624 | |
| 625 | va_start(args, fmt); |
| 626 | vpanic(fmt, args); |
| 627 | va_end(args); |
| 628 | } |
| 629 | EXPORT_SYMBOL(panic); |
| 630 | |
| 631 | #define TAINT_FLAG(taint, _c_true, _c_false, _module) \ |
| 632 | [ TAINT_##taint ] = { \ |
| 633 | .c_true = _c_true, .c_false = _c_false, \ |
| 634 | .module = _module, \ |
| 635 | .desc = #taint, \ |
| 636 | } |
| 637 | |
| 638 | /* |
| 639 | * TAINT_FORCED_RMMOD could be a per-module flag but the module |
| 640 | * is being removed anyway. |
| 641 | */ |
| 642 | const struct taint_flag taint_flags[TAINT_FLAGS_COUNT] = { |
| 643 | TAINT_FLAG(PROPRIETARY_MODULE, 'P', 'G', true), |
| 644 | TAINT_FLAG(FORCED_MODULE, 'F', ' ', true), |
| 645 | TAINT_FLAG(CPU_OUT_OF_SPEC, 'S', ' ', false), |
| 646 | TAINT_FLAG(FORCED_RMMOD, 'R', ' ', false), |
| 647 | TAINT_FLAG(MACHINE_CHECK, 'M', ' ', false), |
| 648 | TAINT_FLAG(BAD_PAGE, 'B', ' ', false), |
| 649 | TAINT_FLAG(USER, 'U', ' ', false), |
| 650 | TAINT_FLAG(DIE, 'D', ' ', false), |
| 651 | TAINT_FLAG(OVERRIDDEN_ACPI_TABLE, 'A', ' ', false), |
| 652 | TAINT_FLAG(WARN, 'W', ' ', false), |
| 653 | TAINT_FLAG(CRAP, 'C', ' ', true), |
| 654 | TAINT_FLAG(FIRMWARE_WORKAROUND, 'I', ' ', false), |
| 655 | TAINT_FLAG(OOT_MODULE, 'O', ' ', true), |
| 656 | TAINT_FLAG(UNSIGNED_MODULE, 'E', ' ', true), |
| 657 | TAINT_FLAG(SOFTLOCKUP, 'L', ' ', false), |
| 658 | TAINT_FLAG(LIVEPATCH, 'K', ' ', true), |
| 659 | TAINT_FLAG(AUX, 'X', ' ', true), |
| 660 | TAINT_FLAG(RANDSTRUCT, 'T', ' ', true), |
| 661 | TAINT_FLAG(TEST, 'N', ' ', true), |
| 662 | TAINT_FLAG(FWCTL, 'J', ' ', true), |
| 663 | }; |
| 664 | |
| 665 | #undef TAINT_FLAG |
| 666 | |
| 667 | static void print_tainted_seq(struct seq_buf *s, bool verbose) |
| 668 | { |
| 669 | const char *sep = "" ; |
| 670 | int i; |
| 671 | |
| 672 | if (!tainted_mask) { |
| 673 | seq_buf_puts(s, str: "Not tainted" ); |
| 674 | return; |
| 675 | } |
| 676 | |
| 677 | seq_buf_printf(s, fmt: "Tainted: " ); |
| 678 | for (i = 0; i < TAINT_FLAGS_COUNT; i++) { |
| 679 | const struct taint_flag *t = &taint_flags[i]; |
| 680 | bool is_set = test_bit(i, &tainted_mask); |
| 681 | char c = is_set ? t->c_true : t->c_false; |
| 682 | |
| 683 | if (verbose) { |
| 684 | if (is_set) { |
| 685 | seq_buf_printf(s, fmt: "%s[%c]=%s" , sep, c, t->desc); |
| 686 | sep = ", " ; |
| 687 | } |
| 688 | } else { |
| 689 | seq_buf_putc(s, c); |
| 690 | } |
| 691 | } |
| 692 | } |
| 693 | |
| 694 | static const char *_print_tainted(bool verbose) |
| 695 | { |
| 696 | /* FIXME: what should the size be? */ |
| 697 | static char buf[sizeof(taint_flags)]; |
| 698 | struct seq_buf s; |
| 699 | |
| 700 | BUILD_BUG_ON(ARRAY_SIZE(taint_flags) != TAINT_FLAGS_COUNT); |
| 701 | |
| 702 | seq_buf_init(s: &s, buf, size: sizeof(buf)); |
| 703 | |
| 704 | print_tainted_seq(s: &s, verbose); |
| 705 | |
| 706 | return seq_buf_str(s: &s); |
| 707 | } |
| 708 | |
| 709 | /** |
| 710 | * print_tainted - return a string to represent the kernel taint state. |
| 711 | * |
| 712 | * For individual taint flag meanings, see Documentation/admin-guide/sysctl/kernel.rst |
| 713 | * |
| 714 | * The string is overwritten by the next call to print_tainted(), |
| 715 | * but is always NULL terminated. |
| 716 | */ |
| 717 | const char *print_tainted(void) |
| 718 | { |
| 719 | return _print_tainted(verbose: false); |
| 720 | } |
| 721 | |
| 722 | /** |
| 723 | * print_tainted_verbose - A more verbose version of print_tainted() |
| 724 | */ |
| 725 | const char *print_tainted_verbose(void) |
| 726 | { |
| 727 | return _print_tainted(verbose: true); |
| 728 | } |
| 729 | |
| 730 | int test_taint(unsigned flag) |
| 731 | { |
| 732 | return test_bit(flag, &tainted_mask); |
| 733 | } |
| 734 | EXPORT_SYMBOL(test_taint); |
| 735 | |
| 736 | unsigned long get_taint(void) |
| 737 | { |
| 738 | return tainted_mask; |
| 739 | } |
| 740 | |
| 741 | /** |
| 742 | * add_taint: add a taint flag if not already set. |
| 743 | * @flag: one of the TAINT_* constants. |
| 744 | * @lockdep_ok: whether lock debugging is still OK. |
| 745 | * |
| 746 | * If something bad has gone wrong, you'll want @lockdebug_ok = false, but for |
| 747 | * some notewortht-but-not-corrupting cases, it can be set to true. |
| 748 | */ |
| 749 | void add_taint(unsigned flag, enum lockdep_ok lockdep_ok) |
| 750 | { |
| 751 | if (lockdep_ok == LOCKDEP_NOW_UNRELIABLE && __debug_locks_off()) |
| 752 | pr_warn("Disabling lock debugging due to kernel taint\n" ); |
| 753 | |
| 754 | set_bit(nr: flag, addr: &tainted_mask); |
| 755 | |
| 756 | if (tainted_mask & panic_on_taint) { |
| 757 | panic_on_taint = 0; |
| 758 | panic("panic_on_taint set ..." ); |
| 759 | } |
| 760 | } |
| 761 | EXPORT_SYMBOL(add_taint); |
| 762 | |
| 763 | static void spin_msec(int msecs) |
| 764 | { |
| 765 | int i; |
| 766 | |
| 767 | for (i = 0; i < msecs; i++) { |
| 768 | touch_nmi_watchdog(); |
| 769 | mdelay(1); |
| 770 | } |
| 771 | } |
| 772 | |
| 773 | /* |
| 774 | * It just happens that oops_enter() and oops_exit() are identically |
| 775 | * implemented... |
| 776 | */ |
| 777 | static void do_oops_enter_exit(void) |
| 778 | { |
| 779 | unsigned long flags; |
| 780 | static int spin_counter; |
| 781 | |
| 782 | if (!pause_on_oops) |
| 783 | return; |
| 784 | |
| 785 | spin_lock_irqsave(&pause_on_oops_lock, flags); |
| 786 | if (pause_on_oops_flag == 0) { |
| 787 | /* This CPU may now print the oops message */ |
| 788 | pause_on_oops_flag = 1; |
| 789 | } else { |
| 790 | /* We need to stall this CPU */ |
| 791 | if (!spin_counter) { |
| 792 | /* This CPU gets to do the counting */ |
| 793 | spin_counter = pause_on_oops; |
| 794 | do { |
| 795 | spin_unlock(lock: &pause_on_oops_lock); |
| 796 | spin_msec(MSEC_PER_SEC); |
| 797 | spin_lock(lock: &pause_on_oops_lock); |
| 798 | } while (--spin_counter); |
| 799 | pause_on_oops_flag = 0; |
| 800 | } else { |
| 801 | /* This CPU waits for a different one */ |
| 802 | while (spin_counter) { |
| 803 | spin_unlock(lock: &pause_on_oops_lock); |
| 804 | spin_msec(msecs: 1); |
| 805 | spin_lock(lock: &pause_on_oops_lock); |
| 806 | } |
| 807 | } |
| 808 | } |
| 809 | spin_unlock_irqrestore(lock: &pause_on_oops_lock, flags); |
| 810 | } |
| 811 | |
| 812 | /* |
| 813 | * Return true if the calling CPU is allowed to print oops-related info. |
| 814 | * This is a bit racy.. |
| 815 | */ |
| 816 | bool oops_may_print(void) |
| 817 | { |
| 818 | return pause_on_oops_flag == 0; |
| 819 | } |
| 820 | |
| 821 | /* |
| 822 | * Called when the architecture enters its oops handler, before it prints |
| 823 | * anything. If this is the first CPU to oops, and it's oopsing the first |
| 824 | * time then let it proceed. |
| 825 | * |
| 826 | * This is all enabled by the pause_on_oops kernel boot option. We do all |
| 827 | * this to ensure that oopses don't scroll off the screen. It has the |
| 828 | * side-effect of preventing later-oopsing CPUs from mucking up the display, |
| 829 | * too. |
| 830 | * |
| 831 | * It turns out that the CPU which is allowed to print ends up pausing for |
| 832 | * the right duration, whereas all the other CPUs pause for twice as long: |
| 833 | * once in oops_enter(), once in oops_exit(). |
| 834 | */ |
| 835 | void oops_enter(void) |
| 836 | { |
| 837 | nbcon_cpu_emergency_enter(); |
| 838 | tracing_off(); |
| 839 | /* can't trust the integrity of the kernel anymore: */ |
| 840 | debug_locks_off(); |
| 841 | do_oops_enter_exit(); |
| 842 | |
| 843 | if (sysctl_oops_all_cpu_backtrace) |
| 844 | trigger_all_cpu_backtrace(); |
| 845 | } |
| 846 | |
| 847 | static void print_oops_end_marker(void) |
| 848 | { |
| 849 | pr_warn("---[ end trace %016llx ]---\n" , 0ULL); |
| 850 | } |
| 851 | |
| 852 | /* |
| 853 | * Called when the architecture exits its oops handler, after printing |
| 854 | * everything. |
| 855 | */ |
| 856 | void oops_exit(void) |
| 857 | { |
| 858 | do_oops_enter_exit(); |
| 859 | print_oops_end_marker(); |
| 860 | nbcon_cpu_emergency_exit(); |
| 861 | kmsg_dump(reason: KMSG_DUMP_OOPS); |
| 862 | } |
| 863 | |
| 864 | struct warn_args { |
| 865 | const char *fmt; |
| 866 | va_list args; |
| 867 | }; |
| 868 | |
| 869 | void __warn(const char *file, int line, void *caller, unsigned taint, |
| 870 | struct pt_regs *regs, struct warn_args *args) |
| 871 | { |
| 872 | nbcon_cpu_emergency_enter(); |
| 873 | |
| 874 | disable_trace_on_warning(); |
| 875 | |
| 876 | if (file) |
| 877 | pr_warn("WARNING: CPU: %d PID: %d at %s:%d %pS\n" , |
| 878 | raw_smp_processor_id(), current->pid, file, line, |
| 879 | caller); |
| 880 | else |
| 881 | pr_warn("WARNING: CPU: %d PID: %d at %pS\n" , |
| 882 | raw_smp_processor_id(), current->pid, caller); |
| 883 | |
| 884 | #pragma GCC diagnostic push |
| 885 | #ifndef __clang__ |
| 886 | #pragma GCC diagnostic ignored "-Wsuggest-attribute=format" |
| 887 | #endif |
| 888 | if (args) |
| 889 | vprintk(fmt: args->fmt, args: args->args); |
| 890 | #pragma GCC diagnostic pop |
| 891 | |
| 892 | print_modules(); |
| 893 | |
| 894 | if (regs) |
| 895 | show_regs(regs); |
| 896 | |
| 897 | check_panic_on_warn(origin: "kernel" ); |
| 898 | |
| 899 | if (!regs) |
| 900 | dump_stack(); |
| 901 | |
| 902 | print_irqtrace_events(current); |
| 903 | |
| 904 | print_oops_end_marker(); |
| 905 | trace_error_report_end(error_detector: ERROR_DETECTOR_WARN, id: (unsigned long)caller); |
| 906 | |
| 907 | /* Just a warning, don't kill lockdep. */ |
| 908 | add_taint(taint, LOCKDEP_STILL_OK); |
| 909 | |
| 910 | nbcon_cpu_emergency_exit(); |
| 911 | } |
| 912 | |
| 913 | #ifdef CONFIG_BUG |
| 914 | #ifndef __WARN_FLAGS |
| 915 | void warn_slowpath_fmt(const char *file, int line, unsigned taint, |
| 916 | const char *fmt, ...) |
| 917 | { |
| 918 | bool rcu = warn_rcu_enter(); |
| 919 | struct warn_args args; |
| 920 | |
| 921 | pr_warn(CUT_HERE); |
| 922 | |
| 923 | if (!fmt) { |
| 924 | __warn(file, line, __builtin_return_address(0), taint, |
| 925 | NULL, NULL); |
| 926 | warn_rcu_exit(rcu); |
| 927 | return; |
| 928 | } |
| 929 | |
| 930 | args.fmt = fmt; |
| 931 | va_start(args.args, fmt); |
| 932 | __warn(file, line, __builtin_return_address(0), taint, NULL, &args); |
| 933 | va_end(args.args); |
| 934 | warn_rcu_exit(rcu); |
| 935 | } |
| 936 | EXPORT_SYMBOL(warn_slowpath_fmt); |
| 937 | #else |
| 938 | void __warn_printk(const char *fmt, ...) |
| 939 | { |
| 940 | bool rcu = warn_rcu_enter(); |
| 941 | va_list args; |
| 942 | |
| 943 | pr_warn(CUT_HERE); |
| 944 | |
| 945 | va_start(args, fmt); |
| 946 | vprintk(fmt, args); |
| 947 | va_end(args); |
| 948 | warn_rcu_exit(rcu); |
| 949 | } |
| 950 | EXPORT_SYMBOL(__warn_printk); |
| 951 | #endif |
| 952 | |
| 953 | /* Support resetting WARN*_ONCE state */ |
| 954 | |
| 955 | static int clear_warn_once_set(void *data, u64 val) |
| 956 | { |
| 957 | generic_bug_clear_once(); |
| 958 | memset(s: __start_once, c: 0, n: __end_once - __start_once); |
| 959 | return 0; |
| 960 | } |
| 961 | |
| 962 | DEFINE_DEBUGFS_ATTRIBUTE(clear_warn_once_fops, NULL, clear_warn_once_set, |
| 963 | "%lld\n" ); |
| 964 | |
| 965 | static __init int register_warn_debugfs(void) |
| 966 | { |
| 967 | /* Don't care about failure */ |
| 968 | debugfs_create_file_unsafe(name: "clear_warn_once" , mode: 0200, NULL, NULL, |
| 969 | fops: &clear_warn_once_fops); |
| 970 | return 0; |
| 971 | } |
| 972 | |
| 973 | device_initcall(register_warn_debugfs); |
| 974 | #endif |
| 975 | |
| 976 | #ifdef CONFIG_STACKPROTECTOR |
| 977 | |
| 978 | /* |
| 979 | * Called when gcc's -fstack-protector feature is used, and |
| 980 | * gcc detects corruption of the on-stack canary value |
| 981 | */ |
| 982 | __visible noinstr void __stack_chk_fail(void) |
| 983 | { |
| 984 | unsigned long flags; |
| 985 | |
| 986 | instrumentation_begin(); |
| 987 | flags = user_access_save(); |
| 988 | |
| 989 | panic("stack-protector: Kernel stack is corrupted in: %pB" , |
| 990 | __builtin_return_address(0)); |
| 991 | |
| 992 | user_access_restore(flags); |
| 993 | instrumentation_end(); |
| 994 | } |
| 995 | EXPORT_SYMBOL(__stack_chk_fail); |
| 996 | |
| 997 | #endif |
| 998 | |
| 999 | core_param(panic, panic_timeout, int, 0644); |
| 1000 | core_param(pause_on_oops, pause_on_oops, int, 0644); |
| 1001 | core_param(panic_on_warn, panic_on_warn, int, 0644); |
| 1002 | core_param(crash_kexec_post_notifiers, crash_kexec_post_notifiers, bool, 0644); |
| 1003 | core_param(panic_console_replay, panic_console_replay, bool, 0644); |
| 1004 | |
| 1005 | static int panic_print_set(const char *val, const struct kernel_param *kp) |
| 1006 | { |
| 1007 | panic_print_deprecated(); |
| 1008 | return param_set_ulong(val, kp); |
| 1009 | } |
| 1010 | |
| 1011 | static int panic_print_get(char *val, const struct kernel_param *kp) |
| 1012 | { |
| 1013 | panic_print_deprecated(); |
| 1014 | return param_get_ulong(buffer: val, kp); |
| 1015 | } |
| 1016 | |
| 1017 | static const struct kernel_param_ops panic_print_ops = { |
| 1018 | .set = panic_print_set, |
| 1019 | .get = panic_print_get, |
| 1020 | }; |
| 1021 | __core_param_cb(panic_print, &panic_print_ops, &panic_print, 0644); |
| 1022 | |
| 1023 | static int __init oops_setup(char *s) |
| 1024 | { |
| 1025 | if (!s) |
| 1026 | return -EINVAL; |
| 1027 | if (!strcmp(s, "panic" )) |
| 1028 | panic_on_oops = 1; |
| 1029 | return 0; |
| 1030 | } |
| 1031 | early_param("oops" , oops_setup); |
| 1032 | |
| 1033 | static int __init panic_on_taint_setup(char *s) |
| 1034 | { |
| 1035 | char *taint_str; |
| 1036 | |
| 1037 | if (!s) |
| 1038 | return -EINVAL; |
| 1039 | |
| 1040 | taint_str = strsep(&s, "," ); |
| 1041 | if (kstrtoul(s: taint_str, base: 16, res: &panic_on_taint)) |
| 1042 | return -EINVAL; |
| 1043 | |
| 1044 | /* make sure panic_on_taint doesn't hold out-of-range TAINT flags */ |
| 1045 | panic_on_taint &= TAINT_FLAGS_MAX; |
| 1046 | |
| 1047 | if (!panic_on_taint) |
| 1048 | return -EINVAL; |
| 1049 | |
| 1050 | if (s && !strcmp(s, "nousertaint" )) |
| 1051 | panic_on_taint_nousertaint = true; |
| 1052 | |
| 1053 | pr_info("panic_on_taint: bitmask=0x%lx nousertaint_mode=%s\n" , |
| 1054 | panic_on_taint, str_enabled_disabled(panic_on_taint_nousertaint)); |
| 1055 | |
| 1056 | return 0; |
| 1057 | } |
| 1058 | early_param("panic_on_taint" , panic_on_taint_setup); |
| 1059 | |