Linux Audio

Check our new training course

Loading...
v6.13.7
   1// SPDX-License-Identifier: GPL-2.0-only
   2/*
   3 *  linux/kernel/printk.c
   4 *
   5 *  Copyright (C) 1991, 1992  Linus Torvalds
   6 *
   7 * Modified to make sys_syslog() more flexible: added commands to
   8 * return the last 4k of kernel messages, regardless of whether
   9 * they've been read or not.  Added option to suppress kernel printk's
  10 * to the console.  Added hook for sending the console messages
  11 * elsewhere, in preparation for a serial line console (someday).
  12 * Ted Ts'o, 2/11/93.
  13 * Modified for sysctl support, 1/8/97, Chris Horn.
  14 * Fixed SMP synchronization, 08/08/99, Manfred Spraul
  15 *     manfred@colorfullife.com
  16 * Rewrote bits to get rid of console_lock
  17 *	01Mar01 Andrew Morton
  18 */
  19
  20#define pr_fmt(fmt) KBUILD_MODNAME ": " fmt
  21
  22#include <linux/kernel.h>
  23#include <linux/mm.h>
  24#include <linux/tty.h>
  25#include <linux/tty_driver.h>
  26#include <linux/console.h>
  27#include <linux/init.h>
  28#include <linux/jiffies.h>
  29#include <linux/nmi.h>
  30#include <linux/module.h>
  31#include <linux/moduleparam.h>
  32#include <linux/delay.h>
  33#include <linux/smp.h>
  34#include <linux/security.h>
  35#include <linux/memblock.h>
  36#include <linux/syscalls.h>
  37#include <linux/syscore_ops.h>
  38#include <linux/vmcore_info.h>
  39#include <linux/ratelimit.h>
  40#include <linux/kmsg_dump.h>
  41#include <linux/syslog.h>
  42#include <linux/cpu.h>
  43#include <linux/rculist.h>
  44#include <linux/poll.h>
  45#include <linux/irq_work.h>
  46#include <linux/ctype.h>
  47#include <linux/uio.h>
  48#include <linux/sched/clock.h>
  49#include <linux/sched/debug.h>
  50#include <linux/sched/task_stack.h>
  51
  52#include <linux/uaccess.h>
  53#include <asm/sections.h>
  54
  55#include <trace/events/initcall.h>
  56#define CREATE_TRACE_POINTS
  57#include <trace/events/printk.h>
  58
  59#include "printk_ringbuffer.h"
  60#include "console_cmdline.h"
  61#include "braille.h"
  62#include "internal.h"
  63
  64int console_printk[4] = {
  65	CONSOLE_LOGLEVEL_DEFAULT,	/* console_loglevel */
  66	MESSAGE_LOGLEVEL_DEFAULT,	/* default_message_loglevel */
  67	CONSOLE_LOGLEVEL_MIN,		/* minimum_console_loglevel */
  68	CONSOLE_LOGLEVEL_DEFAULT,	/* default_console_loglevel */
  69};
  70EXPORT_SYMBOL_GPL(console_printk);
  71
  72atomic_t ignore_console_lock_warning __read_mostly = ATOMIC_INIT(0);
  73EXPORT_SYMBOL(ignore_console_lock_warning);
  74
  75EXPORT_TRACEPOINT_SYMBOL_GPL(console);
  76
  77/*
  78 * Low level drivers may need that to know if they can schedule in
  79 * their unblank() callback or not. So let's export it.
  80 */
  81int oops_in_progress;
  82EXPORT_SYMBOL(oops_in_progress);
  83
  84/*
  85 * console_mutex protects console_list updates and console->flags updates.
  86 * The flags are synchronized only for consoles that are registered, i.e.
  87 * accessible via the console list.
  88 */
  89static DEFINE_MUTEX(console_mutex);
  90
  91/*
  92 * console_sem protects updates to console->seq
  93 * and also provides serialization for console printing.
  94 */
  95static DEFINE_SEMAPHORE(console_sem, 1);
  96HLIST_HEAD(console_list);
  97EXPORT_SYMBOL_GPL(console_list);
  98DEFINE_STATIC_SRCU(console_srcu);
  99
 100/*
 101 * System may need to suppress printk message under certain
 102 * circumstances, like after kernel panic happens.
 103 */
 104int __read_mostly suppress_printk;
 105
 
 
 
 
 
 
 106#ifdef CONFIG_LOCKDEP
 107static struct lockdep_map console_lock_dep_map = {
 108	.name = "console_lock"
 109};
 110
 111void lockdep_assert_console_list_lock_held(void)
 112{
 113	lockdep_assert_held(&console_mutex);
 114}
 115EXPORT_SYMBOL(lockdep_assert_console_list_lock_held);
 116#endif
 117
 118#ifdef CONFIG_DEBUG_LOCK_ALLOC
 119bool console_srcu_read_lock_is_held(void)
 120{
 121	return srcu_read_lock_held(&console_srcu);
 122}
 123EXPORT_SYMBOL(console_srcu_read_lock_is_held);
 124#endif
 125
 126enum devkmsg_log_bits {
 127	__DEVKMSG_LOG_BIT_ON = 0,
 128	__DEVKMSG_LOG_BIT_OFF,
 129	__DEVKMSG_LOG_BIT_LOCK,
 130};
 131
 132enum devkmsg_log_masks {
 133	DEVKMSG_LOG_MASK_ON             = BIT(__DEVKMSG_LOG_BIT_ON),
 134	DEVKMSG_LOG_MASK_OFF            = BIT(__DEVKMSG_LOG_BIT_OFF),
 135	DEVKMSG_LOG_MASK_LOCK           = BIT(__DEVKMSG_LOG_BIT_LOCK),
 136};
 137
 138/* Keep both the 'on' and 'off' bits clear, i.e. ratelimit by default: */
 139#define DEVKMSG_LOG_MASK_DEFAULT	0
 140
 141static unsigned int __read_mostly devkmsg_log = DEVKMSG_LOG_MASK_DEFAULT;
 142
 143static int __control_devkmsg(char *str)
 144{
 145	size_t len;
 146
 147	if (!str)
 148		return -EINVAL;
 149
 150	len = str_has_prefix(str, "on");
 151	if (len) {
 152		devkmsg_log = DEVKMSG_LOG_MASK_ON;
 153		return len;
 154	}
 155
 156	len = str_has_prefix(str, "off");
 157	if (len) {
 158		devkmsg_log = DEVKMSG_LOG_MASK_OFF;
 159		return len;
 160	}
 161
 162	len = str_has_prefix(str, "ratelimit");
 163	if (len) {
 164		devkmsg_log = DEVKMSG_LOG_MASK_DEFAULT;
 165		return len;
 166	}
 167
 168	return -EINVAL;
 169}
 170
 171static int __init control_devkmsg(char *str)
 172{
 173	if (__control_devkmsg(str) < 0) {
 174		pr_warn("printk.devkmsg: bad option string '%s'\n", str);
 175		return 1;
 176	}
 177
 178	/*
 179	 * Set sysctl string accordingly:
 180	 */
 181	if (devkmsg_log == DEVKMSG_LOG_MASK_ON)
 182		strscpy(devkmsg_log_str, "on");
 183	else if (devkmsg_log == DEVKMSG_LOG_MASK_OFF)
 184		strscpy(devkmsg_log_str, "off");
 185	/* else "ratelimit" which is set by default. */
 186
 187	/*
 188	 * Sysctl cannot change it anymore. The kernel command line setting of
 189	 * this parameter is to force the setting to be permanent throughout the
 190	 * runtime of the system. This is a precation measure against userspace
 191	 * trying to be a smarta** and attempting to change it up on us.
 192	 */
 193	devkmsg_log |= DEVKMSG_LOG_MASK_LOCK;
 194
 195	return 1;
 196}
 197__setup("printk.devkmsg=", control_devkmsg);
 198
 199char devkmsg_log_str[DEVKMSG_STR_MAX_SIZE] = "ratelimit";
 200#if defined(CONFIG_PRINTK) && defined(CONFIG_SYSCTL)
 201int devkmsg_sysctl_set_loglvl(const struct ctl_table *table, int write,
 202			      void *buffer, size_t *lenp, loff_t *ppos)
 203{
 204	char old_str[DEVKMSG_STR_MAX_SIZE];
 205	unsigned int old;
 206	int err;
 207
 208	if (write) {
 209		if (devkmsg_log & DEVKMSG_LOG_MASK_LOCK)
 210			return -EINVAL;
 211
 212		old = devkmsg_log;
 213		strscpy(old_str, devkmsg_log_str);
 214	}
 215
 216	err = proc_dostring(table, write, buffer, lenp, ppos);
 217	if (err)
 218		return err;
 219
 220	if (write) {
 221		err = __control_devkmsg(devkmsg_log_str);
 222
 223		/*
 224		 * Do not accept an unknown string OR a known string with
 225		 * trailing crap...
 226		 */
 227		if (err < 0 || (err + 1 != *lenp)) {
 228
 229			/* ... and restore old setting. */
 230			devkmsg_log = old;
 231			strscpy(devkmsg_log_str, old_str);
 232
 233			return -EINVAL;
 234		}
 235	}
 236
 237	return 0;
 238}
 239#endif /* CONFIG_PRINTK && CONFIG_SYSCTL */
 240
 241/**
 242 * console_list_lock - Lock the console list
 243 *
 244 * For console list or console->flags updates
 245 */
 246void console_list_lock(void)
 247{
 248	/*
 249	 * In unregister_console() and console_force_preferred_locked(),
 250	 * synchronize_srcu() is called with the console_list_lock held.
 251	 * Therefore it is not allowed that the console_list_lock is taken
 252	 * with the srcu_lock held.
 253	 *
 254	 * Detecting if this context is really in the read-side critical
 255	 * section is only possible if the appropriate debug options are
 256	 * enabled.
 257	 */
 258	WARN_ON_ONCE(debug_lockdep_rcu_enabled() &&
 259		     srcu_read_lock_held(&console_srcu));
 260
 261	mutex_lock(&console_mutex);
 262}
 263EXPORT_SYMBOL(console_list_lock);
 264
 265/**
 266 * console_list_unlock - Unlock the console list
 267 *
 268 * Counterpart to console_list_lock()
 269 */
 270void console_list_unlock(void)
 271{
 272	mutex_unlock(&console_mutex);
 273}
 274EXPORT_SYMBOL(console_list_unlock);
 275
 276/**
 277 * console_srcu_read_lock - Register a new reader for the
 278 *	SRCU-protected console list
 279 *
 280 * Use for_each_console_srcu() to iterate the console list
 281 *
 282 * Context: Any context.
 283 * Return: A cookie to pass to console_srcu_read_unlock().
 284 */
 285int console_srcu_read_lock(void)
 286	__acquires(&console_srcu)
 287{
 288	return srcu_read_lock_nmisafe(&console_srcu);
 289}
 290EXPORT_SYMBOL(console_srcu_read_lock);
 291
 292/**
 293 * console_srcu_read_unlock - Unregister an old reader from
 294 *	the SRCU-protected console list
 295 * @cookie: cookie returned from console_srcu_read_lock()
 296 *
 297 * Counterpart to console_srcu_read_lock()
 298 */
 299void console_srcu_read_unlock(int cookie)
 300	__releases(&console_srcu)
 301{
 302	srcu_read_unlock_nmisafe(&console_srcu, cookie);
 303}
 304EXPORT_SYMBOL(console_srcu_read_unlock);
 305
 306/*
 307 * Helper macros to handle lockdep when locking/unlocking console_sem. We use
 308 * macros instead of functions so that _RET_IP_ contains useful information.
 309 */
 310#define down_console_sem() do { \
 311	down(&console_sem);\
 312	mutex_acquire(&console_lock_dep_map, 0, 0, _RET_IP_);\
 313} while (0)
 314
 315static int __down_trylock_console_sem(unsigned long ip)
 316{
 317	int lock_failed;
 318	unsigned long flags;
 319
 320	/*
 321	 * Here and in __up_console_sem() we need to be in safe mode,
 322	 * because spindump/WARN/etc from under console ->lock will
 323	 * deadlock in printk()->down_trylock_console_sem() otherwise.
 324	 */
 325	printk_safe_enter_irqsave(flags);
 326	lock_failed = down_trylock(&console_sem);
 327	printk_safe_exit_irqrestore(flags);
 328
 329	if (lock_failed)
 330		return 1;
 331	mutex_acquire(&console_lock_dep_map, 0, 1, ip);
 332	return 0;
 333}
 334#define down_trylock_console_sem() __down_trylock_console_sem(_RET_IP_)
 335
 336static void __up_console_sem(unsigned long ip)
 337{
 338	unsigned long flags;
 339
 340	mutex_release(&console_lock_dep_map, ip);
 341
 342	printk_safe_enter_irqsave(flags);
 343	up(&console_sem);
 344	printk_safe_exit_irqrestore(flags);
 345}
 346#define up_console_sem() __up_console_sem(_RET_IP_)
 347
 348static bool panic_in_progress(void)
 349{
 350	return unlikely(atomic_read(&panic_cpu) != PANIC_CPU_INVALID);
 351}
 352
 353/* Return true if a panic is in progress on the current CPU. */
 354bool this_cpu_in_panic(void)
 355{
 356	/*
 357	 * We can use raw_smp_processor_id() here because it is impossible for
 358	 * the task to be migrated to the panic_cpu, or away from it. If
 359	 * panic_cpu has already been set, and we're not currently executing on
 360	 * that CPU, then we never will be.
 361	 */
 362	return unlikely(atomic_read(&panic_cpu) == raw_smp_processor_id());
 363}
 364
 365/*
 366 * Return true if a panic is in progress on a remote CPU.
 367 *
 368 * On true, the local CPU should immediately release any printing resources
 369 * that may be needed by the panic CPU.
 370 */
 371bool other_cpu_in_panic(void)
 372{
 373	return (panic_in_progress() && !this_cpu_in_panic());
 374}
 375
 376/*
 377 * This is used for debugging the mess that is the VT code by
 378 * keeping track if we have the console semaphore held. It's
 379 * definitely not the perfect debug tool (we don't know if _WE_
 380 * hold it and are racing, but it helps tracking those weird code
 381 * paths in the console code where we end up in places I want
 382 * locked without the console semaphore held).
 383 */
 384static int console_locked;
 385
 386/*
 387 *	Array of consoles built from command line options (console=)
 388 */
 389
 390#define MAX_CMDLINECONSOLES 8
 391
 392static struct console_cmdline console_cmdline[MAX_CMDLINECONSOLES];
 393
 394static int preferred_console = -1;
 395int console_set_on_cmdline;
 396EXPORT_SYMBOL(console_set_on_cmdline);
 397
 398/* Flag: console code may call schedule() */
 399static int console_may_schedule;
 400
 401enum con_msg_format_flags {
 402	MSG_FORMAT_DEFAULT	= 0,
 403	MSG_FORMAT_SYSLOG	= (1 << 0),
 404};
 405
 406static int console_msg_format = MSG_FORMAT_DEFAULT;
 407
 408/*
 409 * The printk log buffer consists of a sequenced collection of records, each
 410 * containing variable length message text. Every record also contains its
 411 * own meta-data (@info).
 412 *
 413 * Every record meta-data carries the timestamp in microseconds, as well as
 414 * the standard userspace syslog level and syslog facility. The usual kernel
 415 * messages use LOG_KERN; userspace-injected messages always carry a matching
 416 * syslog facility, by default LOG_USER. The origin of every message can be
 417 * reliably determined that way.
 418 *
 419 * The human readable log message of a record is available in @text, the
 420 * length of the message text in @text_len. The stored message is not
 421 * terminated.
 422 *
 423 * Optionally, a record can carry a dictionary of properties (key/value
 424 * pairs), to provide userspace with a machine-readable message context.
 425 *
 426 * Examples for well-defined, commonly used property names are:
 427 *   DEVICE=b12:8               device identifier
 428 *                                b12:8         block dev_t
 429 *                                c127:3        char dev_t
 430 *                                n8            netdev ifindex
 431 *                                +sound:card0  subsystem:devname
 432 *   SUBSYSTEM=pci              driver-core subsystem name
 433 *
 434 * Valid characters in property names are [a-zA-Z0-9.-_]. Property names
 435 * and values are terminated by a '\0' character.
 436 *
 437 * Example of record values:
 438 *   record.text_buf                = "it's a line" (unterminated)
 439 *   record.info.seq                = 56
 440 *   record.info.ts_nsec            = 36863
 441 *   record.info.text_len           = 11
 442 *   record.info.facility           = 0 (LOG_KERN)
 443 *   record.info.flags              = 0
 444 *   record.info.level              = 3 (LOG_ERR)
 445 *   record.info.caller_id          = 299 (task 299)
 446 *   record.info.dev_info.subsystem = "pci" (terminated)
 447 *   record.info.dev_info.device    = "+pci:0000:00:01.0" (terminated)
 448 *
 449 * The 'struct printk_info' buffer must never be directly exported to
 450 * userspace, it is a kernel-private implementation detail that might
 451 * need to be changed in the future, when the requirements change.
 452 *
 453 * /dev/kmsg exports the structured data in the following line format:
 454 *   "<level>,<sequnum>,<timestamp>,<contflag>[,additional_values, ... ];<message text>\n"
 455 *
 456 * Users of the export format should ignore possible additional values
 457 * separated by ',', and find the message after the ';' character.
 458 *
 459 * The optional key/value pairs are attached as continuation lines starting
 460 * with a space character and terminated by a newline. All possible
 461 * non-prinatable characters are escaped in the "\xff" notation.
 462 */
 463
 464/* syslog_lock protects syslog_* variables and write access to clear_seq. */
 465static DEFINE_MUTEX(syslog_lock);
 466
 467/*
 468 * Specifies if a legacy console is registered. If legacy consoles are
 469 * present, it is necessary to perform the console lock/unlock dance
 470 * whenever console flushing should occur.
 471 */
 472bool have_legacy_console;
 473
 474/*
 475 * Specifies if an nbcon console is registered. If nbcon consoles are present,
 476 * synchronous printing of legacy consoles will not occur during panic until
 477 * the backtrace has been stored to the ringbuffer.
 478 */
 479bool have_nbcon_console;
 480
 481/*
 482 * Specifies if a boot console is registered. If boot consoles are present,
 483 * nbcon consoles cannot print simultaneously and must be synchronized by
 484 * the console lock. This is because boot consoles and nbcon consoles may
 485 * have mapped the same hardware.
 486 */
 487bool have_boot_console;
 488
 489/* See printk_legacy_allow_panic_sync() for details. */
 490bool legacy_allow_panic_sync;
 491
 492#ifdef CONFIG_PRINTK
 493DECLARE_WAIT_QUEUE_HEAD(log_wait);
 494static DECLARE_WAIT_QUEUE_HEAD(legacy_wait);
 495/* All 3 protected by @syslog_lock. */
 496/* the next printk record to read by syslog(READ) or /proc/kmsg */
 497static u64 syslog_seq;
 498static size_t syslog_partial;
 499static bool syslog_time;
 500
 501/* True when _all_ printer threads are available for printing. */
 502bool printk_kthreads_running;
 503
 504struct latched_seq {
 505	seqcount_latch_t	latch;
 506	u64			val[2];
 507};
 508
 509/*
 510 * The next printk record to read after the last 'clear' command. There are
 511 * two copies (updated with seqcount_latch) so that reads can locklessly
 512 * access a valid value. Writers are synchronized by @syslog_lock.
 513 */
 514static struct latched_seq clear_seq = {
 515	.latch		= SEQCNT_LATCH_ZERO(clear_seq.latch),
 516	.val[0]		= 0,
 517	.val[1]		= 0,
 518};
 519
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 520#define LOG_LEVEL(v)		((v) & 0x07)
 521#define LOG_FACILITY(v)		((v) >> 3 & 0xff)
 522
 523/* record buffer */
 524#define LOG_ALIGN __alignof__(unsigned long)
 525#define __LOG_BUF_LEN (1 << CONFIG_LOG_BUF_SHIFT)
 526#define LOG_BUF_LEN_MAX ((u32)1 << 31)
 527static char __log_buf[__LOG_BUF_LEN] __aligned(LOG_ALIGN);
 528static char *log_buf = __log_buf;
 529static u32 log_buf_len = __LOG_BUF_LEN;
 530
 531/*
 532 * Define the average message size. This only affects the number of
 533 * descriptors that will be available. Underestimating is better than
 534 * overestimating (too many available descriptors is better than not enough).
 535 */
 536#define PRB_AVGBITS 5	/* 32 character average length */
 537
 538#if CONFIG_LOG_BUF_SHIFT <= PRB_AVGBITS
 539#error CONFIG_LOG_BUF_SHIFT value too small.
 540#endif
 541_DEFINE_PRINTKRB(printk_rb_static, CONFIG_LOG_BUF_SHIFT - PRB_AVGBITS,
 542		 PRB_AVGBITS, &__log_buf[0]);
 543
 544static struct printk_ringbuffer printk_rb_dynamic;
 545
 546struct printk_ringbuffer *prb = &printk_rb_static;
 547
 548/*
 549 * We cannot access per-CPU data (e.g. per-CPU flush irq_work) before
 550 * per_cpu_areas are initialised. This variable is set to true when
 551 * it's safe to access per-CPU data.
 552 */
 553static bool __printk_percpu_data_ready __ro_after_init;
 554
 555bool printk_percpu_data_ready(void)
 556{
 557	return __printk_percpu_data_ready;
 558}
 559
 560/* Must be called under syslog_lock. */
 561static void latched_seq_write(struct latched_seq *ls, u64 val)
 562{
 563	write_seqcount_latch_begin(&ls->latch);
 564	ls->val[0] = val;
 565	write_seqcount_latch(&ls->latch);
 566	ls->val[1] = val;
 567	write_seqcount_latch_end(&ls->latch);
 568}
 569
 570/* Can be called from any context. */
 571static u64 latched_seq_read_nolock(struct latched_seq *ls)
 572{
 573	unsigned int seq;
 574	unsigned int idx;
 575	u64 val;
 576
 577	do {
 578		seq = read_seqcount_latch(&ls->latch);
 579		idx = seq & 0x1;
 580		val = ls->val[idx];
 581	} while (read_seqcount_latch_retry(&ls->latch, seq));
 582
 583	return val;
 584}
 585
 586/* Return log buffer address */
 587char *log_buf_addr_get(void)
 588{
 589	return log_buf;
 590}
 591
 592/* Return log buffer size */
 593u32 log_buf_len_get(void)
 594{
 595	return log_buf_len;
 596}
 597
 598/*
 599 * Define how much of the log buffer we could take at maximum. The value
 600 * must be greater than two. Note that only half of the buffer is available
 601 * when the index points to the middle.
 602 */
 603#define MAX_LOG_TAKE_PART 4
 604static const char trunc_msg[] = "<truncated>";
 605
 606static void truncate_msg(u16 *text_len, u16 *trunc_msg_len)
 607{
 608	/*
 609	 * The message should not take the whole buffer. Otherwise, it might
 610	 * get removed too soon.
 611	 */
 612	u32 max_text_len = log_buf_len / MAX_LOG_TAKE_PART;
 613
 614	if (*text_len > max_text_len)
 615		*text_len = max_text_len;
 616
 617	/* enable the warning message (if there is room) */
 618	*trunc_msg_len = strlen(trunc_msg);
 619	if (*text_len >= *trunc_msg_len)
 620		*text_len -= *trunc_msg_len;
 621	else
 622		*trunc_msg_len = 0;
 623}
 624
 625int dmesg_restrict = IS_ENABLED(CONFIG_SECURITY_DMESG_RESTRICT);
 626
 627static int syslog_action_restricted(int type)
 628{
 629	if (dmesg_restrict)
 630		return 1;
 631	/*
 632	 * Unless restricted, we allow "read all" and "get buffer size"
 633	 * for everybody.
 634	 */
 635	return type != SYSLOG_ACTION_READ_ALL &&
 636	       type != SYSLOG_ACTION_SIZE_BUFFER;
 637}
 638
 639static int check_syslog_permissions(int type, int source)
 640{
 641	/*
 642	 * If this is from /proc/kmsg and we've already opened it, then we've
 643	 * already done the capabilities checks at open time.
 644	 */
 645	if (source == SYSLOG_FROM_PROC && type != SYSLOG_ACTION_OPEN)
 646		goto ok;
 647
 648	if (syslog_action_restricted(type)) {
 649		if (capable(CAP_SYSLOG))
 650			goto ok;
 
 
 
 
 
 
 
 
 
 
 
 651		return -EPERM;
 652	}
 653ok:
 654	return security_syslog(type);
 655}
 656
 657static void append_char(char **pp, char *e, char c)
 658{
 659	if (*pp < e)
 660		*(*pp)++ = c;
 661}
 662
 663static ssize_t info_print_ext_header(char *buf, size_t size,
 664				     struct printk_info *info)
 665{
 666	u64 ts_usec = info->ts_nsec;
 667	char caller[20];
 668#ifdef CONFIG_PRINTK_CALLER
 669	u32 id = info->caller_id;
 670
 671	snprintf(caller, sizeof(caller), ",caller=%c%u",
 672		 id & 0x80000000 ? 'C' : 'T', id & ~0x80000000);
 673#else
 674	caller[0] = '\0';
 675#endif
 676
 677	do_div(ts_usec, 1000);
 678
 679	return scnprintf(buf, size, "%u,%llu,%llu,%c%s;",
 680			 (info->facility << 3) | info->level, info->seq,
 681			 ts_usec, info->flags & LOG_CONT ? 'c' : '-', caller);
 682}
 683
 684static ssize_t msg_add_ext_text(char *buf, size_t size,
 685				const char *text, size_t text_len,
 686				unsigned char endc)
 687{
 688	char *p = buf, *e = buf + size;
 689	size_t i;
 690
 691	/* escape non-printable characters */
 692	for (i = 0; i < text_len; i++) {
 693		unsigned char c = text[i];
 694
 695		if (c < ' ' || c >= 127 || c == '\\')
 696			p += scnprintf(p, e - p, "\\x%02x", c);
 697		else
 698			append_char(&p, e, c);
 699	}
 700	append_char(&p, e, endc);
 701
 702	return p - buf;
 703}
 704
 705static ssize_t msg_add_dict_text(char *buf, size_t size,
 706				 const char *key, const char *val)
 707{
 708	size_t val_len = strlen(val);
 709	ssize_t len;
 710
 711	if (!val_len)
 712		return 0;
 713
 714	len = msg_add_ext_text(buf, size, "", 0, ' ');	/* dict prefix */
 715	len += msg_add_ext_text(buf + len, size - len, key, strlen(key), '=');
 716	len += msg_add_ext_text(buf + len, size - len, val, val_len, '\n');
 717
 718	return len;
 719}
 720
 721static ssize_t msg_print_ext_body(char *buf, size_t size,
 722				  char *text, size_t text_len,
 723				  struct dev_printk_info *dev_info)
 724{
 725	ssize_t len;
 726
 727	len = msg_add_ext_text(buf, size, text, text_len, '\n');
 728
 729	if (!dev_info)
 730		goto out;
 731
 732	len += msg_add_dict_text(buf + len, size - len, "SUBSYSTEM",
 733				 dev_info->subsystem);
 734	len += msg_add_dict_text(buf + len, size - len, "DEVICE",
 735				 dev_info->device);
 736out:
 737	return len;
 738}
 739
 740/* /dev/kmsg - userspace message inject/listen interface */
 741struct devkmsg_user {
 742	atomic64_t seq;
 743	struct ratelimit_state rs;
 744	struct mutex lock;
 745	struct printk_buffers pbufs;
 
 
 
 
 746};
 747
 748static __printf(3, 4) __cold
 749int devkmsg_emit(int facility, int level, const char *fmt, ...)
 750{
 751	va_list args;
 752	int r;
 753
 754	va_start(args, fmt);
 755	r = vprintk_emit(facility, level, NULL, fmt, args);
 756	va_end(args);
 757
 758	return r;
 759}
 760
 761static ssize_t devkmsg_write(struct kiocb *iocb, struct iov_iter *from)
 762{
 763	char *buf, *line;
 764	int level = default_message_loglevel;
 765	int facility = 1;	/* LOG_USER */
 766	struct file *file = iocb->ki_filp;
 767	struct devkmsg_user *user = file->private_data;
 768	size_t len = iov_iter_count(from);
 769	ssize_t ret = len;
 770
 771	if (len > PRINTKRB_RECORD_MAX)
 772		return -EINVAL;
 773
 774	/* Ignore when user logging is disabled. */
 775	if (devkmsg_log & DEVKMSG_LOG_MASK_OFF)
 776		return len;
 777
 778	/* Ratelimit when not explicitly enabled. */
 779	if (!(devkmsg_log & DEVKMSG_LOG_MASK_ON)) {
 780		if (!___ratelimit(&user->rs, current->comm))
 781			return ret;
 782	}
 783
 784	buf = kmalloc(len+1, GFP_KERNEL);
 785	if (buf == NULL)
 786		return -ENOMEM;
 787
 788	buf[len] = '\0';
 789	if (!copy_from_iter_full(buf, len, from)) {
 790		kfree(buf);
 791		return -EFAULT;
 792	}
 793
 794	/*
 795	 * Extract and skip the syslog prefix <[0-9]*>. Coming from userspace
 796	 * the decimal value represents 32bit, the lower 3 bit are the log
 797	 * level, the rest are the log facility.
 798	 *
 799	 * If no prefix or no userspace facility is specified, we
 800	 * enforce LOG_USER, to be able to reliably distinguish
 801	 * kernel-generated messages from userspace-injected ones.
 802	 */
 803	line = buf;
 804	if (line[0] == '<') {
 805		char *endp = NULL;
 806		unsigned int u;
 807
 808		u = simple_strtoul(line + 1, &endp, 10);
 809		if (endp && endp[0] == '>') {
 810			level = LOG_LEVEL(u);
 811			if (LOG_FACILITY(u) != 0)
 812				facility = LOG_FACILITY(u);
 813			endp++;
 814			line = endp;
 815		}
 816	}
 817
 818	devkmsg_emit(facility, level, "%s", line);
 819	kfree(buf);
 820	return ret;
 821}
 822
 823static ssize_t devkmsg_read(struct file *file, char __user *buf,
 824			    size_t count, loff_t *ppos)
 825{
 826	struct devkmsg_user *user = file->private_data;
 827	char *outbuf = &user->pbufs.outbuf[0];
 828	struct printk_message pmsg = {
 829		.pbufs = &user->pbufs,
 830	};
 831	ssize_t ret;
 832
 
 
 
 833	ret = mutex_lock_interruptible(&user->lock);
 834	if (ret)
 835		return ret;
 836
 837	if (!printk_get_next_message(&pmsg, atomic64_read(&user->seq), true, false)) {
 838		if (file->f_flags & O_NONBLOCK) {
 839			ret = -EAGAIN;
 840			goto out;
 841		}
 842
 843		/*
 844		 * Guarantee this task is visible on the waitqueue before
 845		 * checking the wake condition.
 846		 *
 847		 * The full memory barrier within set_current_state() of
 848		 * prepare_to_wait_event() pairs with the full memory barrier
 849		 * within wq_has_sleeper().
 850		 *
 851		 * This pairs with __wake_up_klogd:A.
 852		 */
 853		ret = wait_event_interruptible(log_wait,
 854				printk_get_next_message(&pmsg, atomic64_read(&user->seq), true,
 855							false)); /* LMM(devkmsg_read:A) */
 856		if (ret)
 857			goto out;
 858	}
 859
 860	if (pmsg.dropped) {
 861		/* our last seen message is gone, return error and reset */
 862		atomic64_set(&user->seq, pmsg.seq);
 863		ret = -EPIPE;
 864		goto out;
 865	}
 866
 867	atomic64_set(&user->seq, pmsg.seq + 1);
 
 
 
 
 
 868
 869	if (pmsg.outbuf_len > count) {
 870		ret = -EINVAL;
 871		goto out;
 872	}
 873
 874	if (copy_to_user(buf, outbuf, pmsg.outbuf_len)) {
 875		ret = -EFAULT;
 876		goto out;
 877	}
 878	ret = pmsg.outbuf_len;
 879out:
 880	mutex_unlock(&user->lock);
 881	return ret;
 882}
 883
 884/*
 885 * Be careful when modifying this function!!!
 886 *
 887 * Only few operations are supported because the device works only with the
 888 * entire variable length messages (records). Non-standard values are
 889 * returned in the other cases and has been this way for quite some time.
 890 * User space applications might depend on this behavior.
 891 */
 892static loff_t devkmsg_llseek(struct file *file, loff_t offset, int whence)
 893{
 894	struct devkmsg_user *user = file->private_data;
 895	loff_t ret = 0;
 896
 
 
 897	if (offset)
 898		return -ESPIPE;
 899
 900	switch (whence) {
 901	case SEEK_SET:
 902		/* the first record */
 903		atomic64_set(&user->seq, prb_first_valid_seq(prb));
 904		break;
 905	case SEEK_DATA:
 906		/*
 907		 * The first record after the last SYSLOG_ACTION_CLEAR,
 908		 * like issued by 'dmesg -c'. Reading /dev/kmsg itself
 909		 * changes no global state, and does not clear anything.
 910		 */
 911		atomic64_set(&user->seq, latched_seq_read_nolock(&clear_seq));
 912		break;
 913	case SEEK_END:
 914		/* after the last record */
 915		atomic64_set(&user->seq, prb_next_seq(prb));
 916		break;
 917	default:
 918		ret = -EINVAL;
 919	}
 920	return ret;
 921}
 922
 923static __poll_t devkmsg_poll(struct file *file, poll_table *wait)
 924{
 925	struct devkmsg_user *user = file->private_data;
 926	struct printk_info info;
 927	__poll_t ret = 0;
 928
 
 
 
 929	poll_wait(file, &log_wait, wait);
 930
 931	if (prb_read_valid_info(prb, atomic64_read(&user->seq), &info, NULL)) {
 932		/* return error when data has vanished underneath us */
 933		if (info.seq != atomic64_read(&user->seq))
 934			ret = EPOLLIN|EPOLLRDNORM|EPOLLERR|EPOLLPRI;
 935		else
 936			ret = EPOLLIN|EPOLLRDNORM;
 937	}
 938
 939	return ret;
 940}
 941
 942static int devkmsg_open(struct inode *inode, struct file *file)
 943{
 944	struct devkmsg_user *user;
 945	int err;
 946
 947	if (devkmsg_log & DEVKMSG_LOG_MASK_OFF)
 948		return -EPERM;
 949
 950	/* write-only does not need any file context */
 951	if ((file->f_flags & O_ACCMODE) != O_WRONLY) {
 952		err = check_syslog_permissions(SYSLOG_ACTION_READ_ALL,
 953					       SYSLOG_FROM_READER);
 954		if (err)
 955			return err;
 956	}
 957
 958	user = kvmalloc(sizeof(struct devkmsg_user), GFP_KERNEL);
 959	if (!user)
 960		return -ENOMEM;
 961
 962	ratelimit_default_init(&user->rs);
 963	ratelimit_set_flags(&user->rs, RATELIMIT_MSG_ON_RELEASE);
 964
 965	mutex_init(&user->lock);
 966
 
 
 
 967	atomic64_set(&user->seq, prb_first_valid_seq(prb));
 968
 969	file->private_data = user;
 970	return 0;
 971}
 972
 973static int devkmsg_release(struct inode *inode, struct file *file)
 974{
 975	struct devkmsg_user *user = file->private_data;
 976
 
 
 
 977	ratelimit_state_exit(&user->rs);
 978
 979	mutex_destroy(&user->lock);
 980	kvfree(user);
 981	return 0;
 982}
 983
 984const struct file_operations kmsg_fops = {
 985	.open = devkmsg_open,
 986	.read = devkmsg_read,
 987	.write_iter = devkmsg_write,
 988	.llseek = devkmsg_llseek,
 989	.poll = devkmsg_poll,
 990	.release = devkmsg_release,
 991};
 992
 993#ifdef CONFIG_VMCORE_INFO
 994/*
 995 * This appends the listed symbols to /proc/vmcore
 996 *
 997 * /proc/vmcore is used by various utilities, like crash and makedumpfile to
 998 * obtain access to symbols that are otherwise very difficult to locate.  These
 999 * symbols are specifically used so that utilities can access and extract the
1000 * dmesg log from a vmcore file after a crash.
1001 */
1002void log_buf_vmcoreinfo_setup(void)
1003{
1004	struct dev_printk_info *dev_info = NULL;
1005
1006	VMCOREINFO_SYMBOL(prb);
1007	VMCOREINFO_SYMBOL(printk_rb_static);
1008	VMCOREINFO_SYMBOL(clear_seq);
1009
1010	/*
1011	 * Export struct size and field offsets. User space tools can
1012	 * parse it and detect any changes to structure down the line.
1013	 */
1014
1015	VMCOREINFO_STRUCT_SIZE(printk_ringbuffer);
1016	VMCOREINFO_OFFSET(printk_ringbuffer, desc_ring);
1017	VMCOREINFO_OFFSET(printk_ringbuffer, text_data_ring);
1018	VMCOREINFO_OFFSET(printk_ringbuffer, fail);
1019
1020	VMCOREINFO_STRUCT_SIZE(prb_desc_ring);
1021	VMCOREINFO_OFFSET(prb_desc_ring, count_bits);
1022	VMCOREINFO_OFFSET(prb_desc_ring, descs);
1023	VMCOREINFO_OFFSET(prb_desc_ring, infos);
1024	VMCOREINFO_OFFSET(prb_desc_ring, head_id);
1025	VMCOREINFO_OFFSET(prb_desc_ring, tail_id);
1026
1027	VMCOREINFO_STRUCT_SIZE(prb_desc);
1028	VMCOREINFO_OFFSET(prb_desc, state_var);
1029	VMCOREINFO_OFFSET(prb_desc, text_blk_lpos);
1030
1031	VMCOREINFO_STRUCT_SIZE(prb_data_blk_lpos);
1032	VMCOREINFO_OFFSET(prb_data_blk_lpos, begin);
1033	VMCOREINFO_OFFSET(prb_data_blk_lpos, next);
1034
1035	VMCOREINFO_STRUCT_SIZE(printk_info);
1036	VMCOREINFO_OFFSET(printk_info, seq);
1037	VMCOREINFO_OFFSET(printk_info, ts_nsec);
1038	VMCOREINFO_OFFSET(printk_info, text_len);
1039	VMCOREINFO_OFFSET(printk_info, caller_id);
1040	VMCOREINFO_OFFSET(printk_info, dev_info);
1041
1042	VMCOREINFO_STRUCT_SIZE(dev_printk_info);
1043	VMCOREINFO_OFFSET(dev_printk_info, subsystem);
1044	VMCOREINFO_LENGTH(printk_info_subsystem, sizeof(dev_info->subsystem));
1045	VMCOREINFO_OFFSET(dev_printk_info, device);
1046	VMCOREINFO_LENGTH(printk_info_device, sizeof(dev_info->device));
1047
1048	VMCOREINFO_STRUCT_SIZE(prb_data_ring);
1049	VMCOREINFO_OFFSET(prb_data_ring, size_bits);
1050	VMCOREINFO_OFFSET(prb_data_ring, data);
1051	VMCOREINFO_OFFSET(prb_data_ring, head_lpos);
1052	VMCOREINFO_OFFSET(prb_data_ring, tail_lpos);
1053
1054	VMCOREINFO_SIZE(atomic_long_t);
1055	VMCOREINFO_TYPE_OFFSET(atomic_long_t, counter);
1056
1057	VMCOREINFO_STRUCT_SIZE(latched_seq);
1058	VMCOREINFO_OFFSET(latched_seq, val);
1059}
1060#endif
1061
1062/* requested log_buf_len from kernel cmdline */
1063static unsigned long __initdata new_log_buf_len;
1064
1065/* we practice scaling the ring buffer by powers of 2 */
1066static void __init log_buf_len_update(u64 size)
1067{
1068	if (size > (u64)LOG_BUF_LEN_MAX) {
1069		size = (u64)LOG_BUF_LEN_MAX;
1070		pr_err("log_buf over 2G is not supported.\n");
1071	}
1072
1073	if (size)
1074		size = roundup_pow_of_two(size);
1075	if (size > log_buf_len)
1076		new_log_buf_len = (unsigned long)size;
1077}
1078
1079/* save requested log_buf_len since it's too early to process it */
1080static int __init log_buf_len_setup(char *str)
1081{
1082	u64 size;
1083
1084	if (!str)
1085		return -EINVAL;
1086
1087	size = memparse(str, &str);
1088
1089	log_buf_len_update(size);
1090
1091	return 0;
1092}
1093early_param("log_buf_len", log_buf_len_setup);
1094
1095#ifdef CONFIG_SMP
1096#define __LOG_CPU_MAX_BUF_LEN (1 << CONFIG_LOG_CPU_MAX_BUF_SHIFT)
1097
1098static void __init log_buf_add_cpu(void)
1099{
1100	unsigned int cpu_extra;
1101
1102	/*
1103	 * archs should set up cpu_possible_bits properly with
1104	 * set_cpu_possible() after setup_arch() but just in
1105	 * case lets ensure this is valid.
1106	 */
1107	if (num_possible_cpus() == 1)
1108		return;
1109
1110	cpu_extra = (num_possible_cpus() - 1) * __LOG_CPU_MAX_BUF_LEN;
1111
1112	/* by default this will only continue through for large > 64 CPUs */
1113	if (cpu_extra <= __LOG_BUF_LEN / 2)
1114		return;
1115
1116	pr_info("log_buf_len individual max cpu contribution: %d bytes\n",
1117		__LOG_CPU_MAX_BUF_LEN);
1118	pr_info("log_buf_len total cpu_extra contributions: %d bytes\n",
1119		cpu_extra);
1120	pr_info("log_buf_len min size: %d bytes\n", __LOG_BUF_LEN);
1121
1122	log_buf_len_update(cpu_extra + __LOG_BUF_LEN);
1123}
1124#else /* !CONFIG_SMP */
1125static inline void log_buf_add_cpu(void) {}
1126#endif /* CONFIG_SMP */
1127
1128static void __init set_percpu_data_ready(void)
1129{
1130	__printk_percpu_data_ready = true;
1131}
1132
1133static unsigned int __init add_to_rb(struct printk_ringbuffer *rb,
1134				     struct printk_record *r)
1135{
1136	struct prb_reserved_entry e;
1137	struct printk_record dest_r;
1138
1139	prb_rec_init_wr(&dest_r, r->info->text_len);
1140
1141	if (!prb_reserve(&e, rb, &dest_r))
1142		return 0;
1143
1144	memcpy(&dest_r.text_buf[0], &r->text_buf[0], r->info->text_len);
1145	dest_r.info->text_len = r->info->text_len;
1146	dest_r.info->facility = r->info->facility;
1147	dest_r.info->level = r->info->level;
1148	dest_r.info->flags = r->info->flags;
1149	dest_r.info->ts_nsec = r->info->ts_nsec;
1150	dest_r.info->caller_id = r->info->caller_id;
1151	memcpy(&dest_r.info->dev_info, &r->info->dev_info, sizeof(dest_r.info->dev_info));
1152
1153	prb_final_commit(&e);
1154
1155	return prb_record_text_space(&e);
1156}
1157
1158static char setup_text_buf[PRINTKRB_RECORD_MAX] __initdata;
1159
1160static void print_log_buf_usage_stats(void)
1161{
1162	unsigned int descs_count = log_buf_len >> PRB_AVGBITS;
1163	size_t meta_data_size;
1164
1165	meta_data_size = descs_count * (sizeof(struct prb_desc) + sizeof(struct printk_info));
1166
1167	pr_info("log buffer data + meta data: %u + %zu = %zu bytes\n",
1168		log_buf_len, meta_data_size, log_buf_len + meta_data_size);
1169}
1170
1171void __init setup_log_buf(int early)
1172{
1173	struct printk_info *new_infos;
1174	unsigned int new_descs_count;
1175	struct prb_desc *new_descs;
1176	struct printk_info info;
1177	struct printk_record r;
1178	unsigned int text_size;
1179	size_t new_descs_size;
1180	size_t new_infos_size;
1181	unsigned long flags;
1182	char *new_log_buf;
1183	unsigned int free;
1184	u64 seq;
1185
1186	/*
1187	 * Some archs call setup_log_buf() multiple times - first is very
1188	 * early, e.g. from setup_arch(), and second - when percpu_areas
1189	 * are initialised.
1190	 */
1191	if (!early)
1192		set_percpu_data_ready();
1193
1194	if (log_buf != __log_buf)
1195		return;
1196
1197	if (!early && !new_log_buf_len)
1198		log_buf_add_cpu();
1199
1200	if (!new_log_buf_len) {
1201		/* Show the memory stats only once. */
1202		if (!early)
1203			goto out;
1204
1205		return;
1206	}
1207
1208	new_descs_count = new_log_buf_len >> PRB_AVGBITS;
1209	if (new_descs_count == 0) {
1210		pr_err("new_log_buf_len: %lu too small\n", new_log_buf_len);
1211		goto out;
1212	}
1213
1214	new_log_buf = memblock_alloc(new_log_buf_len, LOG_ALIGN);
1215	if (unlikely(!new_log_buf)) {
1216		pr_err("log_buf_len: %lu text bytes not available\n",
1217		       new_log_buf_len);
1218		goto out;
1219	}
1220
1221	new_descs_size = new_descs_count * sizeof(struct prb_desc);
1222	new_descs = memblock_alloc(new_descs_size, LOG_ALIGN);
1223	if (unlikely(!new_descs)) {
1224		pr_err("log_buf_len: %zu desc bytes not available\n",
1225		       new_descs_size);
1226		goto err_free_log_buf;
1227	}
1228
1229	new_infos_size = new_descs_count * sizeof(struct printk_info);
1230	new_infos = memblock_alloc(new_infos_size, LOG_ALIGN);
1231	if (unlikely(!new_infos)) {
1232		pr_err("log_buf_len: %zu info bytes not available\n",
1233		       new_infos_size);
1234		goto err_free_descs;
1235	}
1236
1237	prb_rec_init_rd(&r, &info, &setup_text_buf[0], sizeof(setup_text_buf));
1238
1239	prb_init(&printk_rb_dynamic,
1240		 new_log_buf, ilog2(new_log_buf_len),
1241		 new_descs, ilog2(new_descs_count),
1242		 new_infos);
1243
1244	local_irq_save(flags);
1245
1246	log_buf_len = new_log_buf_len;
1247	log_buf = new_log_buf;
1248	new_log_buf_len = 0;
1249
1250	free = __LOG_BUF_LEN;
1251	prb_for_each_record(0, &printk_rb_static, seq, &r) {
1252		text_size = add_to_rb(&printk_rb_dynamic, &r);
1253		if (text_size > free)
1254			free = 0;
1255		else
1256			free -= text_size;
1257	}
1258
1259	prb = &printk_rb_dynamic;
1260
1261	local_irq_restore(flags);
1262
1263	/*
1264	 * Copy any remaining messages that might have appeared from
1265	 * NMI context after copying but before switching to the
1266	 * dynamic buffer.
1267	 */
1268	prb_for_each_record(seq, &printk_rb_static, seq, &r) {
1269		text_size = add_to_rb(&printk_rb_dynamic, &r);
1270		if (text_size > free)
1271			free = 0;
1272		else
1273			free -= text_size;
1274	}
1275
1276	if (seq != prb_next_seq(&printk_rb_static)) {
1277		pr_err("dropped %llu messages\n",
1278		       prb_next_seq(&printk_rb_static) - seq);
1279	}
1280
1281	print_log_buf_usage_stats();
1282	pr_info("early log buf free: %u(%u%%)\n",
1283		free, (free * 100) / __LOG_BUF_LEN);
1284	return;
1285
1286err_free_descs:
1287	memblock_free(new_descs, new_descs_size);
1288err_free_log_buf:
1289	memblock_free(new_log_buf, new_log_buf_len);
1290out:
1291	print_log_buf_usage_stats();
1292}
1293
1294static bool __read_mostly ignore_loglevel;
1295
1296static int __init ignore_loglevel_setup(char *str)
1297{
1298	ignore_loglevel = true;
1299	pr_info("debug: ignoring loglevel setting.\n");
1300
1301	return 0;
1302}
1303
1304early_param("ignore_loglevel", ignore_loglevel_setup);
1305module_param(ignore_loglevel, bool, S_IRUGO | S_IWUSR);
1306MODULE_PARM_DESC(ignore_loglevel,
1307		 "ignore loglevel setting (prints all kernel messages to the console)");
1308
1309static bool suppress_message_printing(int level)
1310{
1311	return (level >= console_loglevel && !ignore_loglevel);
1312}
1313
1314#ifdef CONFIG_BOOT_PRINTK_DELAY
1315
1316static int boot_delay; /* msecs delay after each printk during bootup */
1317static unsigned long long loops_per_msec;	/* based on boot_delay */
1318
1319static int __init boot_delay_setup(char *str)
1320{
1321	unsigned long lpj;
1322
1323	lpj = preset_lpj ? preset_lpj : 1000000;	/* some guess */
1324	loops_per_msec = (unsigned long long)lpj / 1000 * HZ;
1325
1326	get_option(&str, &boot_delay);
1327	if (boot_delay > 10 * 1000)
1328		boot_delay = 0;
1329
1330	pr_debug("boot_delay: %u, preset_lpj: %ld, lpj: %lu, "
1331		"HZ: %d, loops_per_msec: %llu\n",
1332		boot_delay, preset_lpj, lpj, HZ, loops_per_msec);
1333	return 0;
1334}
1335early_param("boot_delay", boot_delay_setup);
1336
1337static void boot_delay_msec(int level)
1338{
1339	unsigned long long k;
1340	unsigned long timeout;
1341	bool suppress = !is_printk_force_console() &&
1342			suppress_message_printing(level);
1343
1344	if ((boot_delay == 0 || system_state >= SYSTEM_RUNNING) || suppress)
 
1345		return;
 
1346
1347	k = (unsigned long long)loops_per_msec * boot_delay;
1348
1349	timeout = jiffies + msecs_to_jiffies(boot_delay);
1350	while (k) {
1351		k--;
1352		cpu_relax();
1353		/*
1354		 * use (volatile) jiffies to prevent
1355		 * compiler reduction; loop termination via jiffies
1356		 * is secondary and may or may not happen.
1357		 */
1358		if (time_after(jiffies, timeout))
1359			break;
1360		touch_nmi_watchdog();
1361	}
1362}
1363#else
1364static inline void boot_delay_msec(int level)
1365{
1366}
1367#endif
1368
1369static bool printk_time = IS_ENABLED(CONFIG_PRINTK_TIME);
1370module_param_named(time, printk_time, bool, S_IRUGO | S_IWUSR);
1371
1372static size_t print_syslog(unsigned int level, char *buf)
1373{
1374	return sprintf(buf, "<%u>", level);
1375}
1376
1377static size_t print_time(u64 ts, char *buf)
1378{
1379	unsigned long rem_nsec = do_div(ts, 1000000000);
1380
1381	return sprintf(buf, "[%5lu.%06lu]",
1382		       (unsigned long)ts, rem_nsec / 1000);
1383}
1384
1385#ifdef CONFIG_PRINTK_CALLER
1386static size_t print_caller(u32 id, char *buf)
1387{
1388	char caller[12];
1389
1390	snprintf(caller, sizeof(caller), "%c%u",
1391		 id & 0x80000000 ? 'C' : 'T', id & ~0x80000000);
1392	return sprintf(buf, "[%6s]", caller);
1393}
1394#else
1395#define print_caller(id, buf) 0
1396#endif
1397
1398static size_t info_print_prefix(const struct printk_info  *info, bool syslog,
1399				bool time, char *buf)
1400{
1401	size_t len = 0;
1402
1403	if (syslog)
1404		len = print_syslog((info->facility << 3) | info->level, buf);
1405
1406	if (time)
1407		len += print_time(info->ts_nsec, buf + len);
1408
1409	len += print_caller(info->caller_id, buf + len);
1410
1411	if (IS_ENABLED(CONFIG_PRINTK_CALLER) || time) {
1412		buf[len++] = ' ';
1413		buf[len] = '\0';
1414	}
1415
1416	return len;
1417}
1418
1419/*
1420 * Prepare the record for printing. The text is shifted within the given
1421 * buffer to avoid a need for another one. The following operations are
1422 * done:
1423 *
1424 *   - Add prefix for each line.
1425 *   - Drop truncated lines that no longer fit into the buffer.
1426 *   - Add the trailing newline that has been removed in vprintk_store().
1427 *   - Add a string terminator.
1428 *
1429 * Since the produced string is always terminated, the maximum possible
1430 * return value is @r->text_buf_size - 1;
1431 *
1432 * Return: The length of the updated/prepared text, including the added
1433 * prefixes and the newline. The terminator is not counted. The dropped
1434 * line(s) are not counted.
1435 */
1436static size_t record_print_text(struct printk_record *r, bool syslog,
1437				bool time)
1438{
1439	size_t text_len = r->info->text_len;
1440	size_t buf_size = r->text_buf_size;
1441	char *text = r->text_buf;
1442	char prefix[PRINTK_PREFIX_MAX];
1443	bool truncated = false;
1444	size_t prefix_len;
1445	size_t line_len;
1446	size_t len = 0;
1447	char *next;
1448
1449	/*
1450	 * If the message was truncated because the buffer was not large
1451	 * enough, treat the available text as if it were the full text.
1452	 */
1453	if (text_len > buf_size)
1454		text_len = buf_size;
1455
1456	prefix_len = info_print_prefix(r->info, syslog, time, prefix);
1457
1458	/*
1459	 * @text_len: bytes of unprocessed text
1460	 * @line_len: bytes of current line _without_ newline
1461	 * @text:     pointer to beginning of current line
1462	 * @len:      number of bytes prepared in r->text_buf
1463	 */
1464	for (;;) {
1465		next = memchr(text, '\n', text_len);
1466		if (next) {
1467			line_len = next - text;
1468		} else {
1469			/* Drop truncated line(s). */
1470			if (truncated)
1471				break;
1472			line_len = text_len;
1473		}
1474
1475		/*
1476		 * Truncate the text if there is not enough space to add the
1477		 * prefix and a trailing newline and a terminator.
1478		 */
1479		if (len + prefix_len + text_len + 1 + 1 > buf_size) {
1480			/* Drop even the current line if no space. */
1481			if (len + prefix_len + line_len + 1 + 1 > buf_size)
1482				break;
1483
1484			text_len = buf_size - len - prefix_len - 1 - 1;
1485			truncated = true;
1486		}
1487
1488		memmove(text + prefix_len, text, text_len);
1489		memcpy(text, prefix, prefix_len);
1490
1491		/*
1492		 * Increment the prepared length to include the text and
1493		 * prefix that were just moved+copied. Also increment for the
1494		 * newline at the end of this line. If this is the last line,
1495		 * there is no newline, but it will be added immediately below.
1496		 */
1497		len += prefix_len + line_len + 1;
1498		if (text_len == line_len) {
1499			/*
1500			 * This is the last line. Add the trailing newline
1501			 * removed in vprintk_store().
1502			 */
1503			text[prefix_len + line_len] = '\n';
1504			break;
1505		}
1506
1507		/*
1508		 * Advance beyond the added prefix and the related line with
1509		 * its newline.
1510		 */
1511		text += prefix_len + line_len + 1;
1512
1513		/*
1514		 * The remaining text has only decreased by the line with its
1515		 * newline.
1516		 *
1517		 * Note that @text_len can become zero. It happens when @text
1518		 * ended with a newline (either due to truncation or the
1519		 * original string ending with "\n\n"). The loop is correctly
1520		 * repeated and (if not truncated) an empty line with a prefix
1521		 * will be prepared.
1522		 */
1523		text_len -= line_len + 1;
1524	}
1525
1526	/*
1527	 * If a buffer was provided, it will be terminated. Space for the
1528	 * string terminator is guaranteed to be available. The terminator is
1529	 * not counted in the return value.
1530	 */
1531	if (buf_size > 0)
1532		r->text_buf[len] = 0;
1533
1534	return len;
1535}
1536
1537static size_t get_record_print_text_size(struct printk_info *info,
1538					 unsigned int line_count,
1539					 bool syslog, bool time)
1540{
1541	char prefix[PRINTK_PREFIX_MAX];
1542	size_t prefix_len;
1543
1544	prefix_len = info_print_prefix(info, syslog, time, prefix);
1545
1546	/*
1547	 * Each line will be preceded with a prefix. The intermediate
1548	 * newlines are already within the text, but a final trailing
1549	 * newline will be added.
1550	 */
1551	return ((prefix_len * line_count) + info->text_len + 1);
1552}
1553
1554/*
1555 * Beginning with @start_seq, find the first record where it and all following
1556 * records up to (but not including) @max_seq fit into @size.
1557 *
1558 * @max_seq is simply an upper bound and does not need to exist. If the caller
1559 * does not require an upper bound, -1 can be used for @max_seq.
1560 */
1561static u64 find_first_fitting_seq(u64 start_seq, u64 max_seq, size_t size,
1562				  bool syslog, bool time)
1563{
1564	struct printk_info info;
1565	unsigned int line_count;
1566	size_t len = 0;
1567	u64 seq;
1568
1569	/* Determine the size of the records up to @max_seq. */
1570	prb_for_each_info(start_seq, prb, seq, &info, &line_count) {
1571		if (info.seq >= max_seq)
1572			break;
1573		len += get_record_print_text_size(&info, line_count, syslog, time);
1574	}
1575
1576	/*
1577	 * Adjust the upper bound for the next loop to avoid subtracting
1578	 * lengths that were never added.
1579	 */
1580	if (seq < max_seq)
1581		max_seq = seq;
1582
1583	/*
1584	 * Move first record forward until length fits into the buffer. Ignore
1585	 * newest messages that were not counted in the above cycle. Messages
1586	 * might appear and get lost in the meantime. This is a best effort
1587	 * that prevents an infinite loop that could occur with a retry.
1588	 */
1589	prb_for_each_info(start_seq, prb, seq, &info, &line_count) {
1590		if (len <= size || info.seq >= max_seq)
1591			break;
1592		len -= get_record_print_text_size(&info, line_count, syslog, time);
1593	}
1594
1595	return seq;
1596}
1597
1598/* The caller is responsible for making sure @size is greater than 0. */
1599static int syslog_print(char __user *buf, int size)
1600{
1601	struct printk_info info;
1602	struct printk_record r;
1603	char *text;
1604	int len = 0;
1605	u64 seq;
1606
1607	text = kmalloc(PRINTK_MESSAGE_MAX, GFP_KERNEL);
1608	if (!text)
1609		return -ENOMEM;
1610
1611	prb_rec_init_rd(&r, &info, text, PRINTK_MESSAGE_MAX);
1612
1613	mutex_lock(&syslog_lock);
1614
1615	/*
1616	 * Wait for the @syslog_seq record to be available. @syslog_seq may
1617	 * change while waiting.
1618	 */
1619	do {
1620		seq = syslog_seq;
1621
1622		mutex_unlock(&syslog_lock);
1623		/*
1624		 * Guarantee this task is visible on the waitqueue before
1625		 * checking the wake condition.
1626		 *
1627		 * The full memory barrier within set_current_state() of
1628		 * prepare_to_wait_event() pairs with the full memory barrier
1629		 * within wq_has_sleeper().
1630		 *
1631		 * This pairs with __wake_up_klogd:A.
1632		 */
1633		len = wait_event_interruptible(log_wait,
1634				prb_read_valid(prb, seq, NULL)); /* LMM(syslog_print:A) */
1635		mutex_lock(&syslog_lock);
1636
1637		if (len)
1638			goto out;
1639	} while (syslog_seq != seq);
1640
1641	/*
1642	 * Copy records that fit into the buffer. The above cycle makes sure
1643	 * that the first record is always available.
1644	 */
1645	do {
1646		size_t n;
1647		size_t skip;
1648		int err;
1649
1650		if (!prb_read_valid(prb, syslog_seq, &r))
1651			break;
1652
1653		if (r.info->seq != syslog_seq) {
1654			/* message is gone, move to next valid one */
1655			syslog_seq = r.info->seq;
1656			syslog_partial = 0;
1657		}
1658
1659		/*
1660		 * To keep reading/counting partial line consistent,
1661		 * use printk_time value as of the beginning of a line.
1662		 */
1663		if (!syslog_partial)
1664			syslog_time = printk_time;
1665
1666		skip = syslog_partial;
1667		n = record_print_text(&r, true, syslog_time);
1668		if (n - syslog_partial <= size) {
1669			/* message fits into buffer, move forward */
1670			syslog_seq = r.info->seq + 1;
1671			n -= syslog_partial;
1672			syslog_partial = 0;
1673		} else if (!len){
1674			/* partial read(), remember position */
1675			n = size;
1676			syslog_partial += n;
1677		} else
1678			n = 0;
1679
1680		if (!n)
1681			break;
1682
1683		mutex_unlock(&syslog_lock);
1684		err = copy_to_user(buf, text + skip, n);
1685		mutex_lock(&syslog_lock);
1686
1687		if (err) {
1688			if (!len)
1689				len = -EFAULT;
1690			break;
1691		}
1692
1693		len += n;
1694		size -= n;
1695		buf += n;
1696	} while (size);
1697out:
1698	mutex_unlock(&syslog_lock);
1699	kfree(text);
1700	return len;
1701}
1702
1703static int syslog_print_all(char __user *buf, int size, bool clear)
1704{
1705	struct printk_info info;
1706	struct printk_record r;
1707	char *text;
1708	int len = 0;
1709	u64 seq;
1710	bool time;
1711
1712	text = kmalloc(PRINTK_MESSAGE_MAX, GFP_KERNEL);
1713	if (!text)
1714		return -ENOMEM;
1715
1716	time = printk_time;
1717	/*
1718	 * Find first record that fits, including all following records,
1719	 * into the user-provided buffer for this dump.
1720	 */
1721	seq = find_first_fitting_seq(latched_seq_read_nolock(&clear_seq), -1,
1722				     size, true, time);
1723
1724	prb_rec_init_rd(&r, &info, text, PRINTK_MESSAGE_MAX);
1725
 
1726	prb_for_each_record(seq, prb, seq, &r) {
1727		int textlen;
1728
1729		textlen = record_print_text(&r, true, time);
1730
1731		if (len + textlen > size) {
1732			seq--;
1733			break;
1734		}
1735
1736		if (copy_to_user(buf + len, text, textlen))
1737			len = -EFAULT;
1738		else
1739			len += textlen;
1740
1741		if (len < 0)
1742			break;
1743	}
1744
1745	if (clear) {
1746		mutex_lock(&syslog_lock);
1747		latched_seq_write(&clear_seq, seq);
1748		mutex_unlock(&syslog_lock);
1749	}
1750
1751	kfree(text);
1752	return len;
1753}
1754
1755static void syslog_clear(void)
1756{
1757	mutex_lock(&syslog_lock);
1758	latched_seq_write(&clear_seq, prb_next_seq(prb));
1759	mutex_unlock(&syslog_lock);
1760}
1761
1762int do_syslog(int type, char __user *buf, int len, int source)
1763{
1764	struct printk_info info;
1765	bool clear = false;
1766	static int saved_console_loglevel = LOGLEVEL_DEFAULT;
1767	int error;
1768
1769	error = check_syslog_permissions(type, source);
1770	if (error)
1771		return error;
1772
1773	switch (type) {
1774	case SYSLOG_ACTION_CLOSE:	/* Close log */
1775		break;
1776	case SYSLOG_ACTION_OPEN:	/* Open log */
1777		break;
1778	case SYSLOG_ACTION_READ:	/* Read from log */
1779		if (!buf || len < 0)
1780			return -EINVAL;
1781		if (!len)
1782			return 0;
1783		if (!access_ok(buf, len))
1784			return -EFAULT;
1785		error = syslog_print(buf, len);
1786		break;
1787	/* Read/clear last kernel messages */
1788	case SYSLOG_ACTION_READ_CLEAR:
1789		clear = true;
1790		fallthrough;
1791	/* Read last kernel messages */
1792	case SYSLOG_ACTION_READ_ALL:
1793		if (!buf || len < 0)
1794			return -EINVAL;
1795		if (!len)
1796			return 0;
1797		if (!access_ok(buf, len))
1798			return -EFAULT;
1799		error = syslog_print_all(buf, len, clear);
1800		break;
1801	/* Clear ring buffer */
1802	case SYSLOG_ACTION_CLEAR:
1803		syslog_clear();
1804		break;
1805	/* Disable logging to console */
1806	case SYSLOG_ACTION_CONSOLE_OFF:
1807		if (saved_console_loglevel == LOGLEVEL_DEFAULT)
1808			saved_console_loglevel = console_loglevel;
1809		console_loglevel = minimum_console_loglevel;
1810		break;
1811	/* Enable logging to console */
1812	case SYSLOG_ACTION_CONSOLE_ON:
1813		if (saved_console_loglevel != LOGLEVEL_DEFAULT) {
1814			console_loglevel = saved_console_loglevel;
1815			saved_console_loglevel = LOGLEVEL_DEFAULT;
1816		}
1817		break;
1818	/* Set level of messages printed to console */
1819	case SYSLOG_ACTION_CONSOLE_LEVEL:
1820		if (len < 1 || len > 8)
1821			return -EINVAL;
1822		if (len < minimum_console_loglevel)
1823			len = minimum_console_loglevel;
1824		console_loglevel = len;
1825		/* Implicitly re-enable logging to console */
1826		saved_console_loglevel = LOGLEVEL_DEFAULT;
1827		break;
1828	/* Number of chars in the log buffer */
1829	case SYSLOG_ACTION_SIZE_UNREAD:
1830		mutex_lock(&syslog_lock);
1831		if (!prb_read_valid_info(prb, syslog_seq, &info, NULL)) {
1832			/* No unread messages. */
1833			mutex_unlock(&syslog_lock);
1834			return 0;
1835		}
1836		if (info.seq != syslog_seq) {
1837			/* messages are gone, move to first one */
1838			syslog_seq = info.seq;
1839			syslog_partial = 0;
1840		}
1841		if (source == SYSLOG_FROM_PROC) {
1842			/*
1843			 * Short-cut for poll(/"proc/kmsg") which simply checks
1844			 * for pending data, not the size; return the count of
1845			 * records, not the length.
1846			 */
1847			error = prb_next_seq(prb) - syslog_seq;
1848		} else {
1849			bool time = syslog_partial ? syslog_time : printk_time;
1850			unsigned int line_count;
1851			u64 seq;
1852
1853			prb_for_each_info(syslog_seq, prb, seq, &info,
1854					  &line_count) {
1855				error += get_record_print_text_size(&info, line_count,
1856								    true, time);
1857				time = printk_time;
1858			}
1859			error -= syslog_partial;
1860		}
1861		mutex_unlock(&syslog_lock);
1862		break;
1863	/* Size of the log buffer */
1864	case SYSLOG_ACTION_SIZE_BUFFER:
1865		error = log_buf_len;
1866		break;
1867	default:
1868		error = -EINVAL;
1869		break;
1870	}
1871
1872	return error;
1873}
1874
1875SYSCALL_DEFINE3(syslog, int, type, char __user *, buf, int, len)
1876{
1877	return do_syslog(type, buf, len, SYSLOG_FROM_READER);
1878}
1879
1880/*
1881 * Special console_lock variants that help to reduce the risk of soft-lockups.
1882 * They allow to pass console_lock to another printk() call using a busy wait.
1883 */
1884
1885#ifdef CONFIG_LOCKDEP
1886static struct lockdep_map console_owner_dep_map = {
1887	.name = "console_owner"
1888};
1889#endif
1890
1891static DEFINE_RAW_SPINLOCK(console_owner_lock);
1892static struct task_struct *console_owner;
1893static bool console_waiter;
1894
1895/**
1896 * console_lock_spinning_enable - mark beginning of code where another
1897 *	thread might safely busy wait
1898 *
1899 * This basically converts console_lock into a spinlock. This marks
1900 * the section where the console_lock owner can not sleep, because
1901 * there may be a waiter spinning (like a spinlock). Also it must be
1902 * ready to hand over the lock at the end of the section.
1903 */
1904void console_lock_spinning_enable(void)
1905{
1906	/*
1907	 * Do not use spinning in panic(). The panic CPU wants to keep the lock.
1908	 * Non-panic CPUs abandon the flush anyway.
1909	 *
1910	 * Just keep the lockdep annotation. The panic-CPU should avoid
1911	 * taking console_owner_lock because it might cause a deadlock.
1912	 * This looks like the easiest way how to prevent false lockdep
1913	 * reports without handling races a lockless way.
1914	 */
1915	if (panic_in_progress())
1916		goto lockdep;
1917
1918	raw_spin_lock(&console_owner_lock);
1919	console_owner = current;
1920	raw_spin_unlock(&console_owner_lock);
1921
1922lockdep:
1923	/* The waiter may spin on us after setting console_owner */
1924	spin_acquire(&console_owner_dep_map, 0, 0, _THIS_IP_);
1925}
1926
1927/**
1928 * console_lock_spinning_disable_and_check - mark end of code where another
1929 *	thread was able to busy wait and check if there is a waiter
1930 * @cookie: cookie returned from console_srcu_read_lock()
1931 *
1932 * This is called at the end of the section where spinning is allowed.
1933 * It has two functions. First, it is a signal that it is no longer
1934 * safe to start busy waiting for the lock. Second, it checks if
1935 * there is a busy waiter and passes the lock rights to her.
1936 *
1937 * Important: Callers lose both the console_lock and the SRCU read lock if
1938 *	there was a busy waiter. They must not touch items synchronized by
1939 *	console_lock or SRCU read lock in this case.
1940 *
1941 * Return: 1 if the lock rights were passed, 0 otherwise.
1942 */
1943int console_lock_spinning_disable_and_check(int cookie)
1944{
1945	int waiter;
1946
1947	/*
1948	 * Ignore spinning waiters during panic() because they might get stopped
1949	 * or blocked at any time,
1950	 *
1951	 * It is safe because nobody is allowed to start spinning during panic
1952	 * in the first place. If there has been a waiter then non panic CPUs
1953	 * might stay spinning. They would get stopped anyway. The panic context
1954	 * will never start spinning and an interrupted spin on panic CPU will
1955	 * never continue.
1956	 */
1957	if (panic_in_progress()) {
1958		/* Keep lockdep happy. */
1959		spin_release(&console_owner_dep_map, _THIS_IP_);
1960		return 0;
1961	}
1962
1963	raw_spin_lock(&console_owner_lock);
1964	waiter = READ_ONCE(console_waiter);
1965	console_owner = NULL;
1966	raw_spin_unlock(&console_owner_lock);
1967
1968	if (!waiter) {
1969		spin_release(&console_owner_dep_map, _THIS_IP_);
1970		return 0;
1971	}
1972
1973	/* The waiter is now free to continue */
1974	WRITE_ONCE(console_waiter, false);
1975
1976	spin_release(&console_owner_dep_map, _THIS_IP_);
1977
1978	/*
1979	 * Preserve lockdep lock ordering. Release the SRCU read lock before
1980	 * releasing the console_lock.
1981	 */
1982	console_srcu_read_unlock(cookie);
1983
1984	/*
1985	 * Hand off console_lock to waiter. The waiter will perform
1986	 * the up(). After this, the waiter is the console_lock owner.
1987	 */
1988	mutex_release(&console_lock_dep_map, _THIS_IP_);
1989	return 1;
1990}
1991
1992/**
1993 * console_trylock_spinning - try to get console_lock by busy waiting
1994 *
1995 * This allows to busy wait for the console_lock when the current
1996 * owner is running in specially marked sections. It means that
1997 * the current owner is running and cannot reschedule until it
1998 * is ready to lose the lock.
1999 *
2000 * Return: 1 if we got the lock, 0 othrewise
2001 */
2002static int console_trylock_spinning(void)
2003{
2004	struct task_struct *owner = NULL;
2005	bool waiter;
2006	bool spin = false;
2007	unsigned long flags;
2008
2009	if (console_trylock())
2010		return 1;
2011
2012	/*
2013	 * It's unsafe to spin once a panic has begun. If we are the
2014	 * panic CPU, we may have already halted the owner of the
2015	 * console_sem. If we are not the panic CPU, then we should
2016	 * avoid taking console_sem, so the panic CPU has a better
2017	 * chance of cleanly acquiring it later.
2018	 */
2019	if (panic_in_progress())
2020		return 0;
2021
2022	printk_safe_enter_irqsave(flags);
2023
2024	raw_spin_lock(&console_owner_lock);
2025	owner = READ_ONCE(console_owner);
2026	waiter = READ_ONCE(console_waiter);
2027	if (!waiter && owner && owner != current) {
2028		WRITE_ONCE(console_waiter, true);
2029		spin = true;
2030	}
2031	raw_spin_unlock(&console_owner_lock);
2032
2033	/*
2034	 * If there is an active printk() writing to the
2035	 * consoles, instead of having it write our data too,
2036	 * see if we can offload that load from the active
2037	 * printer, and do some printing ourselves.
2038	 * Go into a spin only if there isn't already a waiter
2039	 * spinning, and there is an active printer, and
2040	 * that active printer isn't us (recursive printk?).
2041	 */
2042	if (!spin) {
2043		printk_safe_exit_irqrestore(flags);
2044		return 0;
2045	}
2046
2047	/* We spin waiting for the owner to release us */
2048	spin_acquire(&console_owner_dep_map, 0, 0, _THIS_IP_);
2049	/* Owner will clear console_waiter on hand off */
2050	while (READ_ONCE(console_waiter))
2051		cpu_relax();
2052	spin_release(&console_owner_dep_map, _THIS_IP_);
2053
2054	printk_safe_exit_irqrestore(flags);
2055	/*
2056	 * The owner passed the console lock to us.
2057	 * Since we did not spin on console lock, annotate
2058	 * this as a trylock. Otherwise lockdep will
2059	 * complain.
2060	 */
2061	mutex_acquire(&console_lock_dep_map, 0, 1, _THIS_IP_);
2062
2063	/*
2064	 * Update @console_may_schedule for trylock because the previous
2065	 * owner may have been schedulable.
2066	 */
2067	console_may_schedule = 0;
2068
2069	return 1;
2070}
2071
2072/*
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
2073 * Recursion is tracked separately on each CPU. If NMIs are supported, an
2074 * additional NMI context per CPU is also separately tracked. Until per-CPU
2075 * is available, a separate "early tracking" is performed.
2076 */
2077static DEFINE_PER_CPU(u8, printk_count);
2078static u8 printk_count_early;
2079#ifdef CONFIG_HAVE_NMI
2080static DEFINE_PER_CPU(u8, printk_count_nmi);
2081static u8 printk_count_nmi_early;
2082#endif
2083
2084/*
2085 * Recursion is limited to keep the output sane. printk() should not require
2086 * more than 1 level of recursion (allowing, for example, printk() to trigger
2087 * a WARN), but a higher value is used in case some printk-internal errors
2088 * exist, such as the ringbuffer validation checks failing.
2089 */
2090#define PRINTK_MAX_RECURSION 3
2091
2092/*
2093 * Return a pointer to the dedicated counter for the CPU+context of the
2094 * caller.
2095 */
2096static u8 *__printk_recursion_counter(void)
2097{
2098#ifdef CONFIG_HAVE_NMI
2099	if (in_nmi()) {
2100		if (printk_percpu_data_ready())
2101			return this_cpu_ptr(&printk_count_nmi);
2102		return &printk_count_nmi_early;
2103	}
2104#endif
2105	if (printk_percpu_data_ready())
2106		return this_cpu_ptr(&printk_count);
2107	return &printk_count_early;
2108}
2109
2110/*
2111 * Enter recursion tracking. Interrupts are disabled to simplify tracking.
2112 * The caller must check the boolean return value to see if the recursion is
2113 * allowed. On failure, interrupts are not disabled.
2114 *
2115 * @recursion_ptr must be a variable of type (u8 *) and is the same variable
2116 * that is passed to printk_exit_irqrestore().
2117 */
2118#define printk_enter_irqsave(recursion_ptr, flags)	\
2119({							\
2120	bool success = true;				\
2121							\
2122	typecheck(u8 *, recursion_ptr);			\
2123	local_irq_save(flags);				\
2124	(recursion_ptr) = __printk_recursion_counter();	\
2125	if (*(recursion_ptr) > PRINTK_MAX_RECURSION) {	\
2126		local_irq_restore(flags);		\
2127		success = false;			\
2128	} else {					\
2129		(*(recursion_ptr))++;			\
2130	}						\
2131	success;					\
2132})
2133
2134/* Exit recursion tracking, restoring interrupts. */
2135#define printk_exit_irqrestore(recursion_ptr, flags)	\
2136	do {						\
2137		typecheck(u8 *, recursion_ptr);		\
2138		(*(recursion_ptr))--;			\
2139		local_irq_restore(flags);		\
2140	} while (0)
2141
2142int printk_delay_msec __read_mostly;
2143
2144static inline void printk_delay(int level)
2145{
2146	boot_delay_msec(level);
2147
2148	if (unlikely(printk_delay_msec)) {
2149		int m = printk_delay_msec;
2150
2151		while (m--) {
2152			mdelay(1);
2153			touch_nmi_watchdog();
2154		}
2155	}
2156}
2157
2158static inline u32 printk_caller_id(void)
2159{
2160	return in_task() ? task_pid_nr(current) :
2161		0x80000000 + smp_processor_id();
2162}
2163
2164/**
2165 * printk_parse_prefix - Parse level and control flags.
2166 *
2167 * @text:     The terminated text message.
2168 * @level:    A pointer to the current level value, will be updated.
2169 * @flags:    A pointer to the current printk_info flags, will be updated.
2170 *
2171 * @level may be NULL if the caller is not interested in the parsed value.
2172 * Otherwise the variable pointed to by @level must be set to
2173 * LOGLEVEL_DEFAULT in order to be updated with the parsed value.
2174 *
2175 * @flags may be NULL if the caller is not interested in the parsed value.
2176 * Otherwise the variable pointed to by @flags will be OR'd with the parsed
2177 * value.
2178 *
2179 * Return: The length of the parsed level and control flags.
2180 */
2181u16 printk_parse_prefix(const char *text, int *level,
2182			enum printk_info_flags *flags)
2183{
2184	u16 prefix_len = 0;
2185	int kern_level;
2186
2187	while (*text) {
2188		kern_level = printk_get_level(text);
2189		if (!kern_level)
2190			break;
2191
2192		switch (kern_level) {
2193		case '0' ... '7':
2194			if (level && *level == LOGLEVEL_DEFAULT)
2195				*level = kern_level - '0';
2196			break;
2197		case 'c':	/* KERN_CONT */
2198			if (flags)
2199				*flags |= LOG_CONT;
2200		}
2201
2202		prefix_len += 2;
2203		text += 2;
2204	}
2205
2206	return prefix_len;
2207}
2208
2209__printf(5, 0)
2210static u16 printk_sprint(char *text, u16 size, int facility,
2211			 enum printk_info_flags *flags, const char *fmt,
2212			 va_list args)
2213{
2214	u16 text_len;
2215
2216	text_len = vscnprintf(text, size, fmt, args);
2217
2218	/* Mark and strip a trailing newline. */
2219	if (text_len && text[text_len - 1] == '\n') {
2220		text_len--;
2221		*flags |= LOG_NEWLINE;
2222	}
2223
2224	/* Strip log level and control flags. */
2225	if (facility == 0) {
2226		u16 prefix_len;
2227
2228		prefix_len = printk_parse_prefix(text, NULL, NULL);
2229		if (prefix_len) {
2230			text_len -= prefix_len;
2231			memmove(text, text + prefix_len, text_len);
2232		}
2233	}
2234
2235	trace_console(text, text_len);
2236
2237	return text_len;
2238}
2239
2240__printf(4, 0)
2241int vprintk_store(int facility, int level,
2242		  const struct dev_printk_info *dev_info,
2243		  const char *fmt, va_list args)
2244{
2245	struct prb_reserved_entry e;
2246	enum printk_info_flags flags = 0;
2247	struct printk_record r;
2248	unsigned long irqflags;
2249	u16 trunc_msg_len = 0;
2250	char prefix_buf[8];
2251	u8 *recursion_ptr;
2252	u16 reserve_size;
2253	va_list args2;
2254	u32 caller_id;
2255	u16 text_len;
2256	int ret = 0;
2257	u64 ts_nsec;
2258
2259	if (!printk_enter_irqsave(recursion_ptr, irqflags))
2260		return 0;
2261
2262	/*
2263	 * Since the duration of printk() can vary depending on the message
2264	 * and state of the ringbuffer, grab the timestamp now so that it is
2265	 * close to the call of printk(). This provides a more deterministic
2266	 * timestamp with respect to the caller.
2267	 */
2268	ts_nsec = local_clock();
2269
2270	caller_id = printk_caller_id();
2271
2272	/*
2273	 * The sprintf needs to come first since the syslog prefix might be
2274	 * passed in as a parameter. An extra byte must be reserved so that
2275	 * later the vscnprintf() into the reserved buffer has room for the
2276	 * terminating '\0', which is not counted by vsnprintf().
2277	 */
2278	va_copy(args2, args);
2279	reserve_size = vsnprintf(&prefix_buf[0], sizeof(prefix_buf), fmt, args2) + 1;
2280	va_end(args2);
2281
2282	if (reserve_size > PRINTKRB_RECORD_MAX)
2283		reserve_size = PRINTKRB_RECORD_MAX;
2284
2285	/* Extract log level or control flags. */
2286	if (facility == 0)
2287		printk_parse_prefix(&prefix_buf[0], &level, &flags);
2288
2289	if (level == LOGLEVEL_DEFAULT)
2290		level = default_message_loglevel;
2291
2292	if (dev_info)
2293		flags |= LOG_NEWLINE;
2294
2295	if (is_printk_force_console())
2296		flags |= LOG_FORCE_CON;
2297
2298	if (flags & LOG_CONT) {
2299		prb_rec_init_wr(&r, reserve_size);
2300		if (prb_reserve_in_last(&e, prb, &r, caller_id, PRINTKRB_RECORD_MAX)) {
2301			text_len = printk_sprint(&r.text_buf[r.info->text_len], reserve_size,
2302						 facility, &flags, fmt, args);
2303			r.info->text_len += text_len;
2304
2305			if (flags & LOG_FORCE_CON)
2306				r.info->flags |= LOG_FORCE_CON;
2307
2308			if (flags & LOG_NEWLINE) {
2309				r.info->flags |= LOG_NEWLINE;
2310				prb_final_commit(&e);
2311			} else {
2312				prb_commit(&e);
2313			}
2314
2315			ret = text_len;
2316			goto out;
2317		}
2318	}
2319
2320	/*
2321	 * Explicitly initialize the record before every prb_reserve() call.
2322	 * prb_reserve_in_last() and prb_reserve() purposely invalidate the
2323	 * structure when they fail.
2324	 */
2325	prb_rec_init_wr(&r, reserve_size);
2326	if (!prb_reserve(&e, prb, &r)) {
2327		/* truncate the message if it is too long for empty buffer */
2328		truncate_msg(&reserve_size, &trunc_msg_len);
2329
2330		prb_rec_init_wr(&r, reserve_size + trunc_msg_len);
2331		if (!prb_reserve(&e, prb, &r))
2332			goto out;
2333	}
2334
2335	/* fill message */
2336	text_len = printk_sprint(&r.text_buf[0], reserve_size, facility, &flags, fmt, args);
2337	if (trunc_msg_len)
2338		memcpy(&r.text_buf[text_len], trunc_msg, trunc_msg_len);
2339	r.info->text_len = text_len + trunc_msg_len;
2340	r.info->facility = facility;
2341	r.info->level = level & 7;
2342	r.info->flags = flags & 0x1f;
2343	r.info->ts_nsec = ts_nsec;
2344	r.info->caller_id = caller_id;
2345	if (dev_info)
2346		memcpy(&r.info->dev_info, dev_info, sizeof(r.info->dev_info));
2347
2348	/* A message without a trailing newline can be continued. */
2349	if (!(flags & LOG_NEWLINE))
2350		prb_commit(&e);
2351	else
2352		prb_final_commit(&e);
2353
2354	ret = text_len + trunc_msg_len;
2355out:
2356	printk_exit_irqrestore(recursion_ptr, irqflags);
2357	return ret;
2358}
2359
2360/*
2361 * This acts as a one-way switch to allow legacy consoles to print from
2362 * the printk() caller context on a panic CPU. It also attempts to flush
2363 * the legacy consoles in this context.
2364 */
2365void printk_legacy_allow_panic_sync(void)
2366{
2367	struct console_flush_type ft;
2368
2369	legacy_allow_panic_sync = true;
2370
2371	printk_get_console_flush_type(&ft);
2372	if (ft.legacy_direct) {
2373		if (console_trylock())
2374			console_unlock();
2375	}
2376}
2377
2378asmlinkage int vprintk_emit(int facility, int level,
2379			    const struct dev_printk_info *dev_info,
2380			    const char *fmt, va_list args)
2381{
2382	struct console_flush_type ft;
2383	int printed_len;
 
2384
2385	/* Suppress unimportant messages after panic happens */
2386	if (unlikely(suppress_printk))
2387		return 0;
2388
2389	/*
2390	 * The messages on the panic CPU are the most important. If
2391	 * non-panic CPUs are generating any messages, they will be
2392	 * silently dropped.
2393	 */
2394	if (other_cpu_in_panic() && !panic_triggering_all_cpu_backtrace)
2395		return 0;
2396
2397	printk_get_console_flush_type(&ft);
2398
2399	/* If called from the scheduler, we can not call up(). */
2400	if (level == LOGLEVEL_SCHED) {
2401		level = LOGLEVEL_DEFAULT;
2402		ft.legacy_offload |= ft.legacy_direct;
2403		ft.legacy_direct = false;
2404	}
2405
2406	printk_delay(level);
2407
2408	printed_len = vprintk_store(facility, level, dev_info, fmt, args);
2409
2410	if (ft.nbcon_atomic)
2411		nbcon_atomic_flush_pending();
2412
2413	if (ft.nbcon_offload)
2414		nbcon_kthreads_wake();
2415
2416	if (ft.legacy_direct) {
2417		/*
2418		 * The caller may be holding system-critical or
2419		 * timing-sensitive locks. Disable preemption during
2420		 * printing of all remaining records to all consoles so that
2421		 * this context can return as soon as possible. Hopefully
2422		 * another printk() caller will take over the printing.
2423		 */
2424		preempt_disable();
2425		/*
2426		 * Try to acquire and then immediately release the console
2427		 * semaphore. The release will print out buffers. With the
2428		 * spinning variant, this context tries to take over the
2429		 * printing from another printing context.
2430		 */
2431		if (console_trylock_spinning())
2432			console_unlock();
2433		preempt_enable();
2434	}
2435
2436	if (ft.legacy_offload)
2437		defer_console_output();
2438	else
2439		wake_up_klogd();
2440
2441	return printed_len;
2442}
2443EXPORT_SYMBOL(vprintk_emit);
2444
2445int vprintk_default(const char *fmt, va_list args)
2446{
2447	return vprintk_emit(0, LOGLEVEL_DEFAULT, NULL, fmt, args);
2448}
2449EXPORT_SYMBOL_GPL(vprintk_default);
2450
2451asmlinkage __visible int _printk(const char *fmt, ...)
2452{
2453	va_list args;
2454	int r;
2455
2456	va_start(args, fmt);
2457	r = vprintk(fmt, args);
2458	va_end(args);
2459
2460	return r;
2461}
2462EXPORT_SYMBOL(_printk);
2463
2464static bool pr_flush(int timeout_ms, bool reset_on_progress);
2465static bool __pr_flush(struct console *con, int timeout_ms, bool reset_on_progress);
2466
2467#else /* CONFIG_PRINTK */
2468
 
 
2469#define printk_time		false
2470
2471#define prb_read_valid(rb, seq, r)	false
2472#define prb_first_valid_seq(rb)		0
2473#define prb_next_seq(rb)		0
2474
2475static u64 syslog_seq;
2476
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
2477static bool pr_flush(int timeout_ms, bool reset_on_progress) { return true; }
2478static bool __pr_flush(struct console *con, int timeout_ms, bool reset_on_progress) { return true; }
2479
2480#endif /* CONFIG_PRINTK */
2481
2482#ifdef CONFIG_EARLY_PRINTK
2483struct console *early_console;
2484
2485asmlinkage __visible void early_printk(const char *fmt, ...)
2486{
2487	va_list ap;
2488	char buf[512];
2489	int n;
2490
2491	if (!early_console)
2492		return;
2493
2494	va_start(ap, fmt);
2495	n = vscnprintf(buf, sizeof(buf), fmt, ap);
2496	va_end(ap);
2497
2498	early_console->write(early_console, buf, n);
2499}
2500#endif
2501
2502static void set_user_specified(struct console_cmdline *c, bool user_specified)
2503{
2504	if (!user_specified)
2505		return;
2506
2507	/*
2508	 * @c console was defined by the user on the command line.
2509	 * Do not clear when added twice also by SPCR or the device tree.
2510	 */
2511	c->user_specified = true;
2512	/* At least one console defined by the user on the command line. */
2513	console_set_on_cmdline = 1;
2514}
2515
2516static int __add_preferred_console(const char *name, const short idx,
2517				   const char *devname, char *options,
2518				   char *brl_options, bool user_specified)
2519{
2520	struct console_cmdline *c;
2521	int i;
2522
2523	if (!name && !devname)
2524		return -EINVAL;
2525
2526	/*
2527	 * We use a signed short index for struct console for device drivers to
2528	 * indicate a not yet assigned index or port. However, a negative index
2529	 * value is not valid when the console name and index are defined on
2530	 * the command line.
2531	 */
2532	if (name && idx < 0)
2533		return -EINVAL;
2534
2535	/*
2536	 *	See if this tty is not yet registered, and
2537	 *	if we have a slot free.
2538	 */
2539	for (i = 0, c = console_cmdline;
2540	     i < MAX_CMDLINECONSOLES && (c->name[0] || c->devname[0]);
2541	     i++, c++) {
2542		if ((name && strcmp(c->name, name) == 0 && c->index == idx) ||
2543		    (devname && strcmp(c->devname, devname) == 0)) {
2544			if (!brl_options)
2545				preferred_console = i;
2546			set_user_specified(c, user_specified);
2547			return 0;
2548		}
2549	}
2550	if (i == MAX_CMDLINECONSOLES)
2551		return -E2BIG;
2552	if (!brl_options)
2553		preferred_console = i;
2554	if (name)
2555		strscpy(c->name, name);
2556	if (devname)
2557		strscpy(c->devname, devname);
2558	c->options = options;
2559	set_user_specified(c, user_specified);
2560	braille_set_options(c, brl_options);
2561
2562	c->index = idx;
2563	return 0;
2564}
2565
2566static int __init console_msg_format_setup(char *str)
2567{
2568	if (!strcmp(str, "syslog"))
2569		console_msg_format = MSG_FORMAT_SYSLOG;
2570	if (!strcmp(str, "default"))
2571		console_msg_format = MSG_FORMAT_DEFAULT;
2572	return 1;
2573}
2574__setup("console_msg_format=", console_msg_format_setup);
2575
2576/*
2577 * Set up a console.  Called via do_early_param() in init/main.c
2578 * for each "console=" parameter in the boot command line.
2579 */
2580static int __init console_setup(char *str)
2581{
2582	static_assert(sizeof(console_cmdline[0].devname) >= sizeof(console_cmdline[0].name) + 4);
2583	char buf[sizeof(console_cmdline[0].devname)];
2584	char *brl_options = NULL;
2585	char *ttyname = NULL;
2586	char *devname = NULL;
2587	char *options;
2588	char *s;
2589	int idx;
2590
2591	/*
2592	 * console="" or console=null have been suggested as a way to
2593	 * disable console output. Use ttynull that has been created
2594	 * for exactly this purpose.
2595	 */
2596	if (str[0] == 0 || strcmp(str, "null") == 0) {
2597		__add_preferred_console("ttynull", 0, NULL, NULL, NULL, true);
2598		return 1;
2599	}
2600
2601	if (_braille_console_setup(&str, &brl_options))
2602		return 1;
2603
2604	/* For a DEVNAME:0.0 style console the character device is unknown early */
2605	if (strchr(str, ':'))
2606		devname = buf;
2607	else
2608		ttyname = buf;
2609
2610	/*
2611	 * Decode str into name, index, options.
2612	 */
2613	if (ttyname && isdigit(str[0]))
2614		scnprintf(buf, sizeof(buf), "ttyS%s", str);
2615	else
2616		strscpy(buf, str);
2617
 
 
2618	options = strchr(str, ',');
2619	if (options)
2620		*(options++) = 0;
2621
2622#ifdef __sparc__
2623	if (!strcmp(str, "ttya"))
2624		strscpy(buf, "ttyS0");
2625	if (!strcmp(str, "ttyb"))
2626		strscpy(buf, "ttyS1");
2627#endif
2628
2629	for (s = buf; *s; s++)
2630		if ((ttyname && isdigit(*s)) || *s == ',')
2631			break;
2632
2633	/* @idx will get defined when devname matches. */
2634	if (devname)
2635		idx = -1;
2636	else
2637		idx = simple_strtoul(s, NULL, 10);
2638
2639	*s = 0;
2640
2641	__add_preferred_console(ttyname, idx, devname, options, brl_options, true);
2642	return 1;
2643}
2644__setup("console=", console_setup);
2645
2646/**
2647 * add_preferred_console - add a device to the list of preferred consoles.
2648 * @name: device name
2649 * @idx: device index
2650 * @options: options for this console
2651 *
2652 * The last preferred console added will be used for kernel messages
2653 * and stdin/out/err for init.  Normally this is used by console_setup
2654 * above to handle user-supplied console arguments; however it can also
2655 * be used by arch-specific code either to override the user or more
2656 * commonly to provide a default console (ie from PROM variables) when
2657 * the user has not supplied one.
2658 */
2659int add_preferred_console(const char *name, const short idx, char *options)
2660{
2661	return __add_preferred_console(name, idx, NULL, options, NULL, false);
2662}
2663
2664/**
2665 * match_devname_and_update_preferred_console - Update a preferred console
2666 *	when matching devname is found.
2667 * @devname: DEVNAME:0.0 style device name
2668 * @name: Name of the corresponding console driver, e.g. "ttyS"
2669 * @idx: Console index, e.g. port number.
2670 *
2671 * The function checks whether a device with the given @devname is
2672 * preferred via the console=DEVNAME:0.0 command line option.
2673 * It fills the missing console driver name and console index
2674 * so that a later register_console() call could find (match)
2675 * and enable this device.
2676 *
2677 * It might be used when a driver subsystem initializes particular
2678 * devices with already known DEVNAME:0.0 style names. And it
2679 * could predict which console driver name and index this device
2680 * would later get associated with.
2681 *
2682 * Return: 0 on success, negative error code on failure.
2683 */
2684int match_devname_and_update_preferred_console(const char *devname,
2685					       const char *name,
2686					       const short idx)
2687{
2688	struct console_cmdline *c = console_cmdline;
2689	int i;
2690
2691	if (!devname || !strlen(devname) || !name || !strlen(name) || idx < 0)
2692		return -EINVAL;
2693
2694	for (i = 0; i < MAX_CMDLINECONSOLES && (c->name[0] || c->devname[0]);
2695	     i++, c++) {
2696		if (!strcmp(devname, c->devname)) {
2697			pr_info("associate the preferred console \"%s\" with \"%s%d\"\n",
2698				devname, name, idx);
2699			strscpy(c->name, name);
2700			c->index = idx;
2701			return 0;
2702		}
2703	}
2704
2705	return -ENOENT;
2706}
2707EXPORT_SYMBOL_GPL(match_devname_and_update_preferred_console);
2708
2709bool console_suspend_enabled = true;
2710EXPORT_SYMBOL(console_suspend_enabled);
2711
2712static int __init console_suspend_disable(char *str)
2713{
2714	console_suspend_enabled = false;
2715	return 1;
2716}
2717__setup("no_console_suspend", console_suspend_disable);
2718module_param_named(console_suspend, console_suspend_enabled,
2719		bool, S_IRUGO | S_IWUSR);
2720MODULE_PARM_DESC(console_suspend, "suspend console during suspend"
2721	" and hibernate operations");
2722
2723static bool printk_console_no_auto_verbose;
2724
2725void console_verbose(void)
2726{
2727	if (console_loglevel && !printk_console_no_auto_verbose)
2728		console_loglevel = CONSOLE_LOGLEVEL_MOTORMOUTH;
2729}
2730EXPORT_SYMBOL_GPL(console_verbose);
2731
2732module_param_named(console_no_auto_verbose, printk_console_no_auto_verbose, bool, 0644);
2733MODULE_PARM_DESC(console_no_auto_verbose, "Disable console loglevel raise to highest on oops/panic/etc");
2734
2735/**
2736 * suspend_console - suspend the console subsystem
2737 *
2738 * This disables printk() while we go into suspend states
2739 */
2740void suspend_console(void)
2741{
2742	struct console *con;
2743
2744	if (!console_suspend_enabled)
2745		return;
2746	pr_info("Suspending console(s) (use no_console_suspend to debug)\n");
2747	pr_flush(1000, true);
2748
2749	console_list_lock();
2750	for_each_console(con)
2751		console_srcu_write_flags(con, con->flags | CON_SUSPENDED);
2752	console_list_unlock();
2753
2754	/*
2755	 * Ensure that all SRCU list walks have completed. All printing
2756	 * contexts must be able to see that they are suspended so that it
2757	 * is guaranteed that all printing has stopped when this function
2758	 * completes.
2759	 */
2760	synchronize_srcu(&console_srcu);
2761}
2762
2763void resume_console(void)
2764{
2765	struct console_flush_type ft;
2766	struct console *con;
2767
2768	if (!console_suspend_enabled)
2769		return;
2770
2771	console_list_lock();
2772	for_each_console(con)
2773		console_srcu_write_flags(con, con->flags & ~CON_SUSPENDED);
2774	console_list_unlock();
2775
2776	/*
2777	 * Ensure that all SRCU list walks have completed. All printing
2778	 * contexts must be able to see they are no longer suspended so
2779	 * that they are guaranteed to wake up and resume printing.
2780	 */
2781	synchronize_srcu(&console_srcu);
2782
2783	printk_get_console_flush_type(&ft);
2784	if (ft.nbcon_offload)
2785		nbcon_kthreads_wake();
2786	if (ft.legacy_offload)
2787		defer_console_output();
2788
2789	pr_flush(1000, true);
2790}
2791
2792/**
2793 * console_cpu_notify - print deferred console messages after CPU hotplug
2794 * @cpu: unused
2795 *
2796 * If printk() is called from a CPU that is not online yet, the messages
2797 * will be printed on the console only if there are CON_ANYTIME consoles.
2798 * This function is called when a new CPU comes online (or fails to come
2799 * up) or goes offline.
2800 */
2801static int console_cpu_notify(unsigned int cpu)
2802{
2803	struct console_flush_type ft;
2804
2805	if (!cpuhp_tasks_frozen) {
2806		printk_get_console_flush_type(&ft);
2807		if (ft.nbcon_atomic)
2808			nbcon_atomic_flush_pending();
2809		if (ft.legacy_direct) {
2810			if (console_trylock())
2811				console_unlock();
2812		}
2813	}
2814	return 0;
2815}
2816
2817/**
2818 * console_lock - block the console subsystem from printing
2819 *
2820 * Acquires a lock which guarantees that no consoles will
2821 * be in or enter their write() callback.
2822 *
2823 * Can sleep, returns nothing.
2824 */
2825void console_lock(void)
2826{
2827	might_sleep();
2828
2829	/* On panic, the console_lock must be left to the panic cpu. */
2830	while (other_cpu_in_panic())
2831		msleep(1000);
2832
2833	down_console_sem();
 
 
2834	console_locked = 1;
2835	console_may_schedule = 1;
2836}
2837EXPORT_SYMBOL(console_lock);
2838
2839/**
2840 * console_trylock - try to block the console subsystem from printing
2841 *
2842 * Try to acquire a lock which guarantees that no consoles will
2843 * be in or enter their write() callback.
2844 *
2845 * returns 1 on success, and 0 on failure to acquire the lock.
2846 */
2847int console_trylock(void)
2848{
2849	/* On panic, the console_lock must be left to the panic cpu. */
2850	if (other_cpu_in_panic())
2851		return 0;
2852	if (down_trylock_console_sem())
2853		return 0;
 
 
 
 
2854	console_locked = 1;
2855	console_may_schedule = 0;
2856	return 1;
2857}
2858EXPORT_SYMBOL(console_trylock);
2859
2860int is_console_locked(void)
2861{
2862	return console_locked;
2863}
2864EXPORT_SYMBOL(is_console_locked);
2865
2866static void __console_unlock(void)
2867{
2868	console_locked = 0;
2869	up_console_sem();
2870}
2871
2872#ifdef CONFIG_PRINTK
2873
2874/*
2875 * Prepend the message in @pmsg->pbufs->outbuf. This is achieved by shifting
2876 * the existing message over and inserting the scratchbuf message.
2877 *
2878 * @pmsg is the original printk message.
2879 * @fmt is the printf format of the message which will prepend the existing one.
2880 *
2881 * If there is not enough space in @pmsg->pbufs->outbuf, the existing
2882 * message text will be sufficiently truncated.
2883 *
2884 * If @pmsg->pbufs->outbuf is modified, @pmsg->outbuf_len is updated.
2885 */
2886__printf(2, 3)
2887static void console_prepend_message(struct printk_message *pmsg, const char *fmt, ...)
2888{
2889	struct printk_buffers *pbufs = pmsg->pbufs;
2890	const size_t scratchbuf_sz = sizeof(pbufs->scratchbuf);
2891	const size_t outbuf_sz = sizeof(pbufs->outbuf);
2892	char *scratchbuf = &pbufs->scratchbuf[0];
2893	char *outbuf = &pbufs->outbuf[0];
2894	va_list args;
2895	size_t len;
2896
2897	va_start(args, fmt);
2898	len = vscnprintf(scratchbuf, scratchbuf_sz, fmt, args);
2899	va_end(args);
2900
2901	/*
2902	 * Make sure outbuf is sufficiently large before prepending.
2903	 * Keep at least the prefix when the message must be truncated.
2904	 * It is a rather theoretical problem when someone tries to
2905	 * use a minimalist buffer.
2906	 */
2907	if (WARN_ON_ONCE(len + PRINTK_PREFIX_MAX >= outbuf_sz))
2908		return;
2909
2910	if (pmsg->outbuf_len + len >= outbuf_sz) {
2911		/* Truncate the message, but keep it terminated. */
2912		pmsg->outbuf_len = outbuf_sz - (len + 1);
2913		outbuf[pmsg->outbuf_len] = 0;
2914	}
2915
2916	memmove(outbuf + len, outbuf, pmsg->outbuf_len + 1);
2917	memcpy(outbuf, scratchbuf, len);
2918	pmsg->outbuf_len += len;
2919}
2920
2921/*
2922 * Prepend the message in @pmsg->pbufs->outbuf with a "dropped message".
2923 * @pmsg->outbuf_len is updated appropriately.
2924 *
2925 * @pmsg is the printk message to prepend.
2926 *
2927 * @dropped is the dropped count to report in the dropped message.
2928 */
2929void console_prepend_dropped(struct printk_message *pmsg, unsigned long dropped)
2930{
2931	console_prepend_message(pmsg, "** %lu printk messages dropped **\n", dropped);
2932}
2933
2934/*
2935 * Prepend the message in @pmsg->pbufs->outbuf with a "replay message".
2936 * @pmsg->outbuf_len is updated appropriately.
2937 *
2938 * @pmsg is the printk message to prepend.
2939 */
2940void console_prepend_replay(struct printk_message *pmsg)
2941{
2942	console_prepend_message(pmsg, "** replaying previous printk message **\n");
2943}
2944
2945/*
2946 * Read and format the specified record (or a later record if the specified
2947 * record is not available).
2948 *
2949 * @pmsg will contain the formatted result. @pmsg->pbufs must point to a
2950 * struct printk_buffers.
2951 *
2952 * @seq is the record to read and format. If it is not available, the next
2953 * valid record is read.
2954 *
2955 * @is_extended specifies if the message should be formatted for extended
2956 * console output.
2957 *
2958 * @may_supress specifies if records may be skipped based on loglevel.
2959 *
2960 * Returns false if no record is available. Otherwise true and all fields
2961 * of @pmsg are valid. (See the documentation of struct printk_message
2962 * for information about the @pmsg fields.)
2963 */
2964bool printk_get_next_message(struct printk_message *pmsg, u64 seq,
2965			     bool is_extended, bool may_suppress)
2966{
2967	struct printk_buffers *pbufs = pmsg->pbufs;
2968	const size_t scratchbuf_sz = sizeof(pbufs->scratchbuf);
2969	const size_t outbuf_sz = sizeof(pbufs->outbuf);
2970	char *scratchbuf = &pbufs->scratchbuf[0];
2971	char *outbuf = &pbufs->outbuf[0];
2972	struct printk_info info;
2973	struct printk_record r;
2974	size_t len = 0;
2975	bool force_con;
2976
2977	/*
2978	 * Formatting extended messages requires a separate buffer, so use the
2979	 * scratch buffer to read in the ringbuffer text.
2980	 *
2981	 * Formatting normal messages is done in-place, so read the ringbuffer
2982	 * text directly into the output buffer.
2983	 */
2984	if (is_extended)
2985		prb_rec_init_rd(&r, &info, scratchbuf, scratchbuf_sz);
2986	else
2987		prb_rec_init_rd(&r, &info, outbuf, outbuf_sz);
2988
2989	if (!prb_read_valid(prb, seq, &r))
2990		return false;
2991
2992	pmsg->seq = r.info->seq;
2993	pmsg->dropped = r.info->seq - seq;
2994	force_con = r.info->flags & LOG_FORCE_CON;
2995
2996	/*
2997	 * Skip records that are not forced to be printed on consoles and that
2998	 * has level above the console loglevel.
 
2999	 */
3000	if (!force_con && may_suppress && suppress_message_printing(r.info->level))
3001		goto out;
3002
3003	if (is_extended) {
3004		len = info_print_ext_header(outbuf, outbuf_sz, r.info);
3005		len += msg_print_ext_body(outbuf + len, outbuf_sz - len,
3006					  &r.text_buf[0], r.info->text_len, &r.info->dev_info);
3007	} else {
3008		len = record_print_text(&r, console_msg_format & MSG_FORMAT_SYSLOG, printk_time);
3009	}
3010out:
3011	pmsg->outbuf_len = len;
3012	return true;
3013}
3014
3015/*
3016 * Legacy console printing from printk() caller context does not respect
3017 * raw_spinlock/spinlock nesting. For !PREEMPT_RT the lockdep warning is a
3018 * false positive. For PREEMPT_RT the false positive condition does not
3019 * occur.
3020 *
3021 * This map is used to temporarily establish LD_WAIT_SLEEP context for the
3022 * console write() callback when legacy printing to avoid false positive
3023 * lockdep complaints, thus allowing lockdep to continue to function for
3024 * real issues.
3025 */
3026#ifdef CONFIG_PREEMPT_RT
3027static inline void printk_legacy_allow_spinlock_enter(void) { }
3028static inline void printk_legacy_allow_spinlock_exit(void) { }
3029#else
3030static DEFINE_WAIT_OVERRIDE_MAP(printk_legacy_map, LD_WAIT_SLEEP);
3031
3032static inline void printk_legacy_allow_spinlock_enter(void)
3033{
3034	lock_map_acquire_try(&printk_legacy_map);
3035}
3036
3037static inline void printk_legacy_allow_spinlock_exit(void)
3038{
3039	lock_map_release(&printk_legacy_map);
 
3040}
3041#endif /* CONFIG_PREEMPT_RT */
3042
3043/*
3044 * Used as the printk buffers for non-panic, serialized console printing.
3045 * This is for legacy (!CON_NBCON) as well as all boot (CON_BOOT) consoles.
3046 * Its usage requires the console_lock held.
3047 */
3048struct printk_buffers printk_shared_pbufs;
3049
3050/*
3051 * Print one record for the given console. The record printed is whatever
3052 * record is the next available record for the given console.
3053 *
 
 
 
 
 
 
 
 
3054 * @handover will be set to true if a printk waiter has taken over the
3055 * console_lock, in which case the caller is no longer holding both the
3056 * console_lock and the SRCU read lock. Otherwise it is set to false.
3057 *
3058 * @cookie is the cookie from the SRCU read lock.
3059 *
3060 * Returns false if the given console has no next record to print, otherwise
3061 * true.
3062 *
3063 * Requires the console_lock and the SRCU read lock.
3064 */
3065static bool console_emit_next_record(struct console *con, bool *handover, int cookie)
 
3066{
3067	bool is_extended = console_srcu_read_flags(con) & CON_EXTENDED;
3068	char *outbuf = &printk_shared_pbufs.outbuf[0];
3069	struct printk_message pmsg = {
3070		.pbufs = &printk_shared_pbufs,
3071	};
3072	unsigned long flags;
 
 
 
 
3073
3074	*handover = false;
3075
3076	if (!printk_get_next_message(&pmsg, con->seq, is_extended, true))
3077		return false;
3078
3079	con->dropped += pmsg.dropped;
3080
3081	/* Skip messages of formatted length 0. */
3082	if (pmsg.outbuf_len == 0) {
3083		con->seq = pmsg.seq + 1;
3084		goto skip;
 
3085	}
3086
3087	if (con->dropped && !is_extended) {
3088		console_prepend_dropped(&pmsg, con->dropped);
3089		con->dropped = 0;
 
3090	}
3091
3092	/* Write everything out to the hardware. */
3093
3094	if (force_legacy_kthread() && !panic_in_progress()) {
3095		/*
3096		 * With forced threading this function is in a task context
3097		 * (either legacy kthread or get_init_console_seq()). There
3098		 * is no need for concern about printk reentrance, handovers,
3099		 * or lockdep complaints.
3100		 */
3101
3102		con->write(con, outbuf, pmsg.outbuf_len);
3103		con->seq = pmsg.seq + 1;
3104	} else {
3105		/*
3106		 * While actively printing out messages, if another printk()
3107		 * were to occur on another CPU, it may wait for this one to
3108		 * finish. This task can not be preempted if there is a
3109		 * waiter waiting to take over.
3110		 *
3111		 * Interrupts are disabled because the hand over to a waiter
3112		 * must not be interrupted until the hand over is completed
3113		 * (@console_waiter is cleared).
3114		 */
3115		printk_safe_enter_irqsave(flags);
3116		console_lock_spinning_enable();
3117
3118		/* Do not trace print latency. */
3119		stop_critical_timings();
3120
3121		printk_legacy_allow_spinlock_enter();
3122		con->write(con, outbuf, pmsg.outbuf_len);
3123		printk_legacy_allow_spinlock_exit();
 
 
 
 
 
 
 
 
 
3124
3125		start_critical_timings();
 
 
3126
3127		con->seq = pmsg.seq + 1;
3128
3129		*handover = console_lock_spinning_disable_and_check(cookie);
3130		printk_safe_exit_irqrestore(flags);
3131	}
3132skip:
3133	return true;
3134}
3135
3136#else
3137
3138static bool console_emit_next_record(struct console *con, bool *handover, int cookie)
3139{
3140	*handover = false;
3141	return false;
3142}
3143
3144static inline void printk_kthreads_check_locked(void) { }
3145
3146#endif /* CONFIG_PRINTK */
3147
3148/*
3149 * Print out all remaining records to all consoles.
3150 *
3151 * @do_cond_resched is set by the caller. It can be true only in schedulable
3152 * context.
3153 *
3154 * @next_seq is set to the sequence number after the last available record.
3155 * The value is valid only when this function returns true. It means that all
3156 * usable consoles are completely flushed.
3157 *
3158 * @handover will be set to true if a printk waiter has taken over the
3159 * console_lock, in which case the caller is no longer holding the
3160 * console_lock. Otherwise it is set to false.
3161 *
3162 * Returns true when there was at least one usable console and all messages
3163 * were flushed to all usable consoles. A returned false informs the caller
3164 * that everything was not flushed (either there were no usable consoles or
3165 * another context has taken over printing or it is a panic situation and this
3166 * is not the panic CPU). Regardless the reason, the caller should assume it
3167 * is not useful to immediately try again.
3168 *
3169 * Requires the console_lock.
3170 */
3171static bool console_flush_all(bool do_cond_resched, u64 *next_seq, bool *handover)
3172{
3173	struct console_flush_type ft;
 
 
3174	bool any_usable = false;
3175	struct console *con;
3176	bool any_progress;
3177	int cookie;
3178
3179	*next_seq = 0;
3180	*handover = false;
3181
3182	do {
3183		any_progress = false;
3184
3185		printk_get_console_flush_type(&ft);
3186
3187		cookie = console_srcu_read_lock();
3188		for_each_console_srcu(con) {
3189			short flags = console_srcu_read_flags(con);
3190			u64 printk_seq;
3191			bool progress;
3192
3193			/*
3194			 * console_flush_all() is only responsible for nbcon
3195			 * consoles when the nbcon consoles cannot print via
3196			 * their atomic or threaded flushing.
3197			 */
3198			if ((flags & CON_NBCON) && (ft.nbcon_atomic || ft.nbcon_offload))
3199				continue;
3200
3201			if (!console_is_usable(con, flags, !do_cond_resched))
3202				continue;
3203			any_usable = true;
3204
3205			if (flags & CON_NBCON) {
3206				progress = nbcon_legacy_emit_next_record(con, handover, cookie,
3207									 !do_cond_resched);
3208				printk_seq = nbcon_seq_read(con);
 
3209			} else {
3210				progress = console_emit_next_record(con, handover, cookie);
3211				printk_seq = con->seq;
 
3212			}
3213
3214			/*
3215			 * If a handover has occurred, the SRCU read lock
3216			 * is already released.
3217			 */
3218			if (*handover)
3219				return false;
3220
3221			/* Track the next of the highest seq flushed. */
3222			if (printk_seq > *next_seq)
3223				*next_seq = printk_seq;
3224
3225			if (!progress)
3226				continue;
3227			any_progress = true;
3228
3229			/* Allow panic_cpu to take over the consoles safely. */
3230			if (other_cpu_in_panic())
3231				goto abandon;
3232
3233			if (do_cond_resched)
3234				cond_resched();
3235		}
3236		console_srcu_read_unlock(cookie);
3237	} while (any_progress);
3238
3239	return any_usable;
3240
3241abandon:
3242	console_srcu_read_unlock(cookie);
3243	return false;
3244}
3245
3246static void __console_flush_and_unlock(void)
 
 
 
 
 
 
 
 
 
 
 
 
3247{
3248	bool do_cond_resched;
3249	bool handover;
3250	bool flushed;
3251	u64 next_seq;
3252
 
 
 
 
 
3253	/*
3254	 * Console drivers are called with interrupts disabled, so
3255	 * @console_may_schedule should be cleared before; however, we may
3256	 * end up dumping a lot of lines, for example, if called from
3257	 * console registration path, and should invoke cond_resched()
3258	 * between lines if allowable.  Not doing so can cause a very long
3259	 * scheduling stall on a slow console leading to RCU stall and
3260	 * softlockup warnings which exacerbate the issue with more
3261	 * messages practically incapacitating the system. Therefore, create
3262	 * a local to use for the printing loop.
3263	 */
3264	do_cond_resched = console_may_schedule;
3265
3266	do {
3267		console_may_schedule = 0;
3268
3269		flushed = console_flush_all(do_cond_resched, &next_seq, &handover);
3270		if (!handover)
3271			__console_unlock();
3272
3273		/*
3274		 * Abort if there was a failure to flush all messages to all
3275		 * usable consoles. Either it is not possible to flush (in
3276		 * which case it would be an infinite loop of retrying) or
3277		 * another context has taken over printing.
3278		 */
3279		if (!flushed)
3280			break;
3281
3282		/*
3283		 * Some context may have added new records after
3284		 * console_flush_all() but before unlocking the console.
3285		 * Re-check if there is a new record to flush. If the trylock
3286		 * fails, another context is already handling the printing.
3287		 */
3288	} while (prb_read_valid(prb, next_seq, NULL) && console_trylock());
3289}
3290
3291/**
3292 * console_unlock - unblock the legacy console subsystem from printing
3293 *
3294 * Releases the console_lock which the caller holds to block printing of
3295 * the legacy console subsystem.
3296 *
3297 * While the console_lock was held, console output may have been buffered
3298 * by printk(). If this is the case, console_unlock() emits the output on
3299 * legacy consoles prior to releasing the lock.
3300 *
3301 * console_unlock(); may be called from any context.
3302 */
3303void console_unlock(void)
3304{
3305	struct console_flush_type ft;
3306
3307	printk_get_console_flush_type(&ft);
3308	if (ft.legacy_direct)
3309		__console_flush_and_unlock();
3310	else
3311		__console_unlock();
3312}
3313EXPORT_SYMBOL(console_unlock);
3314
3315/**
3316 * console_conditional_schedule - yield the CPU if required
3317 *
3318 * If the console code is currently allowed to sleep, and
3319 * if this CPU should yield the CPU to another task, do
3320 * so here.
3321 *
3322 * Must be called within console_lock();.
3323 */
3324void __sched console_conditional_schedule(void)
3325{
3326	if (console_may_schedule)
3327		cond_resched();
3328}
3329EXPORT_SYMBOL(console_conditional_schedule);
3330
3331void console_unblank(void)
3332{
3333	bool found_unblank = false;
3334	struct console *c;
3335	int cookie;
3336
3337	/*
3338	 * First check if there are any consoles implementing the unblank()
3339	 * callback. If not, there is no reason to continue and take the
3340	 * console lock, which in particular can be dangerous if
3341	 * @oops_in_progress is set.
3342	 */
3343	cookie = console_srcu_read_lock();
3344	for_each_console_srcu(c) {
3345		if ((console_srcu_read_flags(c) & CON_ENABLED) && c->unblank) {
3346			found_unblank = true;
3347			break;
3348		}
3349	}
3350	console_srcu_read_unlock(cookie);
3351	if (!found_unblank)
3352		return;
3353
3354	/*
3355	 * Stop console printing because the unblank() callback may
3356	 * assume the console is not within its write() callback.
3357	 *
3358	 * If @oops_in_progress is set, this may be an atomic context.
3359	 * In that case, attempt a trylock as best-effort.
3360	 */
3361	if (oops_in_progress) {
3362		/* Semaphores are not NMI-safe. */
3363		if (in_nmi())
3364			return;
3365
3366		/*
3367		 * Attempting to trylock the console lock can deadlock
3368		 * if another CPU was stopped while modifying the
3369		 * semaphore. "Hope and pray" that this is not the
3370		 * current situation.
3371		 */
3372		if (down_trylock_console_sem() != 0)
3373			return;
3374	} else
3375		console_lock();
3376
3377	console_locked = 1;
3378	console_may_schedule = 0;
3379
3380	cookie = console_srcu_read_lock();
3381	for_each_console_srcu(c) {
3382		if ((console_srcu_read_flags(c) & CON_ENABLED) && c->unblank)
3383			c->unblank();
3384	}
3385	console_srcu_read_unlock(cookie);
3386
3387	console_unlock();
3388
3389	if (!oops_in_progress)
3390		pr_flush(1000, true);
3391}
3392
3393/*
3394 * Rewind all consoles to the oldest available record.
3395 *
3396 * IMPORTANT: The function is safe only when called under
3397 *            console_lock(). It is not enforced because
3398 *            it is used as a best effort in panic().
3399 */
3400static void __console_rewind_all(void)
3401{
3402	struct console *c;
3403	short flags;
3404	int cookie;
3405	u64 seq;
3406
3407	seq = prb_first_valid_seq(prb);
3408
3409	cookie = console_srcu_read_lock();
3410	for_each_console_srcu(c) {
3411		flags = console_srcu_read_flags(c);
3412
3413		if (flags & CON_NBCON) {
3414			nbcon_seq_force(c, seq);
3415		} else {
3416			/*
3417			 * This assignment is safe only when called under
3418			 * console_lock(). On panic, legacy consoles are
3419			 * only best effort.
3420			 */
3421			c->seq = seq;
3422		}
3423	}
3424	console_srcu_read_unlock(cookie);
3425}
3426
3427/**
3428 * console_flush_on_panic - flush console content on panic
3429 * @mode: flush all messages in buffer or just the pending ones
3430 *
3431 * Immediately output all pending messages no matter what.
3432 */
3433void console_flush_on_panic(enum con_flush_mode mode)
3434{
3435	struct console_flush_type ft;
3436	bool handover;
3437	u64 next_seq;
3438
3439	/*
3440	 * Ignore the console lock and flush out the messages. Attempting a
3441	 * trylock would not be useful because:
3442	 *
3443	 *   - if it is contended, it must be ignored anyway
3444	 *   - console_lock() and console_trylock() block and fail
3445	 *     respectively in panic for non-panic CPUs
3446	 *   - semaphores are not NMI-safe
3447	 */
3448
3449	/*
3450	 * If another context is holding the console lock,
3451	 * @console_may_schedule might be set. Clear it so that
3452	 * this context does not call cond_resched() while flushing.
 
 
3453	 */
 
3454	console_may_schedule = 0;
3455
3456	if (mode == CONSOLE_REPLAY_ALL)
3457		__console_rewind_all();
 
 
 
 
3458
3459	printk_get_console_flush_type(&ft);
3460	if (ft.nbcon_atomic)
3461		nbcon_atomic_flush_pending();
3462
3463	/* Flush legacy consoles once allowed, even when dangerous. */
3464	if (legacy_allow_panic_sync)
3465		console_flush_all(false, &next_seq, &handover);
 
 
 
 
 
3466}
3467
3468/*
3469 * Return the console tty driver structure and its associated index
3470 */
3471struct tty_driver *console_device(int *index)
3472{
3473	struct console *c;
3474	struct tty_driver *driver = NULL;
3475	int cookie;
3476
3477	/*
3478	 * Take console_lock to serialize device() callback with
3479	 * other console operations. For example, fg_console is
3480	 * modified under console_lock when switching vt.
3481	 */
3482	console_lock();
3483
3484	cookie = console_srcu_read_lock();
3485	for_each_console_srcu(c) {
3486		if (!c->device)
3487			continue;
3488		driver = c->device(c, index);
3489		if (driver)
3490			break;
3491	}
3492	console_srcu_read_unlock(cookie);
3493
3494	console_unlock();
3495	return driver;
3496}
3497
3498/*
3499 * Prevent further output on the passed console device so that (for example)
3500 * serial drivers can disable console output before suspending a port, and can
3501 * re-enable output afterwards.
3502 */
3503void console_stop(struct console *console)
3504{
3505	__pr_flush(console, 1000, true);
3506	console_list_lock();
3507	console_srcu_write_flags(console, console->flags & ~CON_ENABLED);
3508	console_list_unlock();
3509
3510	/*
3511	 * Ensure that all SRCU list walks have completed. All contexts must
3512	 * be able to see that this console is disabled so that (for example)
3513	 * the caller can suspend the port without risk of another context
3514	 * using the port.
3515	 */
3516	synchronize_srcu(&console_srcu);
3517}
3518EXPORT_SYMBOL(console_stop);
3519
3520void console_start(struct console *console)
3521{
3522	struct console_flush_type ft;
3523	bool is_nbcon;
3524
3525	console_list_lock();
3526	console_srcu_write_flags(console, console->flags | CON_ENABLED);
3527	is_nbcon = console->flags & CON_NBCON;
3528	console_list_unlock();
3529
3530	/*
3531	 * Ensure that all SRCU list walks have completed. The related
3532	 * printing context must be able to see it is enabled so that
3533	 * it is guaranteed to wake up and resume printing.
3534	 */
3535	synchronize_srcu(&console_srcu);
3536
3537	printk_get_console_flush_type(&ft);
3538	if (is_nbcon && ft.nbcon_offload)
3539		nbcon_kthread_wake(console);
3540	else if (ft.legacy_offload)
3541		defer_console_output();
3542
3543	__pr_flush(console, 1000, true);
3544}
3545EXPORT_SYMBOL(console_start);
3546
3547#ifdef CONFIG_PRINTK
3548static int unregister_console_locked(struct console *console);
3549
3550/* True when system boot is far enough to create printer threads. */
3551static bool printk_kthreads_ready __ro_after_init;
3552
3553static struct task_struct *printk_legacy_kthread;
3554
3555static bool legacy_kthread_should_wakeup(void)
3556{
3557	struct console_flush_type ft;
3558	struct console *con;
3559	bool ret = false;
3560	int cookie;
3561
3562	if (kthread_should_stop())
3563		return true;
3564
3565	printk_get_console_flush_type(&ft);
3566
3567	cookie = console_srcu_read_lock();
3568	for_each_console_srcu(con) {
3569		short flags = console_srcu_read_flags(con);
3570		u64 printk_seq;
3571
3572		/*
3573		 * The legacy printer thread is only responsible for nbcon
3574		 * consoles when the nbcon consoles cannot print via their
3575		 * atomic or threaded flushing.
3576		 */
3577		if ((flags & CON_NBCON) && (ft.nbcon_atomic || ft.nbcon_offload))
3578			continue;
3579
3580		if (!console_is_usable(con, flags, false))
3581			continue;
3582
3583		if (flags & CON_NBCON) {
3584			printk_seq = nbcon_seq_read(con);
3585		} else {
3586			/*
3587			 * It is safe to read @seq because only this
3588			 * thread context updates @seq.
3589			 */
3590			printk_seq = con->seq;
3591		}
3592
3593		if (prb_read_valid(prb, printk_seq, NULL)) {
3594			ret = true;
3595			break;
3596		}
3597	}
3598	console_srcu_read_unlock(cookie);
3599
3600	return ret;
3601}
3602
3603static int legacy_kthread_func(void *unused)
3604{
3605	for (;;) {
3606		wait_event_interruptible(legacy_wait, legacy_kthread_should_wakeup());
3607
3608		if (kthread_should_stop())
3609			break;
3610
3611		console_lock();
3612		__console_flush_and_unlock();
3613	}
3614
3615	return 0;
3616}
3617
3618static bool legacy_kthread_create(void)
3619{
3620	struct task_struct *kt;
3621
3622	lockdep_assert_console_list_lock_held();
3623
3624	kt = kthread_run(legacy_kthread_func, NULL, "pr/legacy");
3625	if (WARN_ON(IS_ERR(kt))) {
3626		pr_err("failed to start legacy printing thread\n");
3627		return false;
3628	}
3629
3630	printk_legacy_kthread = kt;
3631
3632	/*
3633	 * It is important that console printing threads are scheduled
3634	 * shortly after a printk call and with generous runtime budgets.
3635	 */
3636	sched_set_normal(printk_legacy_kthread, -20);
3637
3638	return true;
3639}
3640
3641/**
3642 * printk_kthreads_shutdown - shutdown all threaded printers
3643 *
3644 * On system shutdown all threaded printers are stopped. This allows printk
3645 * to transition back to atomic printing, thus providing a robust mechanism
3646 * for the final shutdown/reboot messages to be output.
3647 */
3648static void printk_kthreads_shutdown(void)
3649{
3650	struct console *con;
3651
3652	console_list_lock();
3653	if (printk_kthreads_running) {
3654		printk_kthreads_running = false;
3655
3656		for_each_console(con) {
3657			if (con->flags & CON_NBCON)
3658				nbcon_kthread_stop(con);
3659		}
3660
3661		/*
3662		 * The threads may have been stopped while printing a
3663		 * backlog. Flush any records left over.
3664		 */
3665		nbcon_atomic_flush_pending();
3666	}
3667	console_list_unlock();
3668}
3669
3670static struct syscore_ops printk_syscore_ops = {
3671	.shutdown = printk_kthreads_shutdown,
3672};
3673
3674/*
3675 * If appropriate, start nbcon kthreads and set @printk_kthreads_running.
3676 * If any kthreads fail to start, those consoles are unregistered.
3677 *
3678 * Must be called under console_list_lock().
3679 */
3680static void printk_kthreads_check_locked(void)
3681{
3682	struct hlist_node *tmp;
3683	struct console *con;
3684
3685	lockdep_assert_console_list_lock_held();
3686
3687	if (!printk_kthreads_ready)
3688		return;
3689
3690	if (have_legacy_console || have_boot_console) {
3691		if (!printk_legacy_kthread &&
3692		    force_legacy_kthread() &&
3693		    !legacy_kthread_create()) {
3694			/*
3695			 * All legacy consoles must be unregistered. If there
3696			 * are any nbcon consoles, they will set up their own
3697			 * kthread.
3698			 */
3699			hlist_for_each_entry_safe(con, tmp, &console_list, node) {
3700				if (con->flags & CON_NBCON)
3701					continue;
3702
3703				unregister_console_locked(con);
3704			}
3705		}
3706	} else if (printk_legacy_kthread) {
3707		kthread_stop(printk_legacy_kthread);
3708		printk_legacy_kthread = NULL;
3709	}
3710
3711	/*
3712	 * Printer threads cannot be started as long as any boot console is
3713	 * registered because there is no way to synchronize the hardware
3714	 * registers between boot console code and regular console code.
3715	 * It can only be known that there will be no new boot consoles when
3716	 * an nbcon console is registered.
3717	 */
3718	if (have_boot_console || !have_nbcon_console) {
3719		/* Clear flag in case all nbcon consoles unregistered. */
3720		printk_kthreads_running = false;
3721		return;
3722	}
3723
3724	if (printk_kthreads_running)
3725		return;
3726
3727	hlist_for_each_entry_safe(con, tmp, &console_list, node) {
3728		if (!(con->flags & CON_NBCON))
3729			continue;
3730
3731		if (!nbcon_kthread_create(con))
3732			unregister_console_locked(con);
3733	}
3734
3735	printk_kthreads_running = true;
3736}
3737
3738static int __init printk_set_kthreads_ready(void)
3739{
3740	register_syscore_ops(&printk_syscore_ops);
3741
3742	console_list_lock();
3743	printk_kthreads_ready = true;
3744	printk_kthreads_check_locked();
3745	console_list_unlock();
3746
3747	return 0;
3748}
3749early_initcall(printk_set_kthreads_ready);
3750#endif /* CONFIG_PRINTK */
3751
3752static int __read_mostly keep_bootcon;
3753
3754static int __init keep_bootcon_setup(char *str)
3755{
3756	keep_bootcon = 1;
3757	pr_info("debug: skip boot console de-registration.\n");
3758
3759	return 0;
3760}
3761
3762early_param("keep_bootcon", keep_bootcon_setup);
3763
3764static int console_call_setup(struct console *newcon, char *options)
3765{
3766	int err;
3767
3768	if (!newcon->setup)
3769		return 0;
3770
3771	/* Synchronize with possible boot console. */
3772	console_lock();
3773	err = newcon->setup(newcon, options);
3774	console_unlock();
3775
3776	return err;
3777}
3778
3779/*
3780 * This is called by register_console() to try to match
3781 * the newly registered console with any of the ones selected
3782 * by either the command line or add_preferred_console() and
3783 * setup/enable it.
3784 *
3785 * Care need to be taken with consoles that are statically
3786 * enabled such as netconsole
3787 */
3788static int try_enable_preferred_console(struct console *newcon,
3789					bool user_specified)
3790{
3791	struct console_cmdline *c;
3792	int i, err;
3793
3794	for (i = 0, c = console_cmdline;
3795	     i < MAX_CMDLINECONSOLES && (c->name[0] || c->devname[0]);
3796	     i++, c++) {
3797		/* Console not yet initialized? */
3798		if (!c->name[0])
3799			continue;
3800		if (c->user_specified != user_specified)
3801			continue;
3802		if (!newcon->match ||
3803		    newcon->match(newcon, c->name, c->index, c->options) != 0) {
3804			/* default matching */
3805			BUILD_BUG_ON(sizeof(c->name) != sizeof(newcon->name));
3806			if (strcmp(c->name, newcon->name) != 0)
3807				continue;
3808			if (newcon->index >= 0 &&
3809			    newcon->index != c->index)
3810				continue;
3811			if (newcon->index < 0)
3812				newcon->index = c->index;
3813
3814			if (_braille_register_console(newcon, c))
3815				return 0;
3816
3817			err = console_call_setup(newcon, c->options);
3818			if (err)
3819				return err;
3820		}
3821		newcon->flags |= CON_ENABLED;
3822		if (i == preferred_console)
3823			newcon->flags |= CON_CONSDEV;
3824		return 0;
3825	}
3826
3827	/*
3828	 * Some consoles, such as pstore and netconsole, can be enabled even
3829	 * without matching. Accept the pre-enabled consoles only when match()
3830	 * and setup() had a chance to be called.
3831	 */
3832	if (newcon->flags & CON_ENABLED && c->user_specified ==	user_specified)
3833		return 0;
3834
3835	return -ENOENT;
3836}
3837
3838/* Try to enable the console unconditionally */
3839static void try_enable_default_console(struct console *newcon)
3840{
3841	if (newcon->index < 0)
3842		newcon->index = 0;
3843
3844	if (console_call_setup(newcon, NULL) != 0)
3845		return;
3846
3847	newcon->flags |= CON_ENABLED;
3848
3849	if (newcon->device)
3850		newcon->flags |= CON_CONSDEV;
3851}
3852
3853/* Return the starting sequence number for a newly registered console. */
3854static u64 get_init_console_seq(struct console *newcon, bool bootcon_registered)
 
 
 
 
3855{
3856	struct console *con;
3857	bool handover;
3858	u64 init_seq;
3859
3860	if (newcon->flags & (CON_PRINTBUFFER | CON_BOOT)) {
3861		/* Get a consistent copy of @syslog_seq. */
3862		mutex_lock(&syslog_lock);
3863		init_seq = syslog_seq;
3864		mutex_unlock(&syslog_lock);
3865	} else {
3866		/* Begin with next message added to ringbuffer. */
3867		init_seq = prb_next_seq(prb);
3868
3869		/*
3870		 * If any enabled boot consoles are due to be unregistered
3871		 * shortly, some may not be caught up and may be the same
3872		 * device as @newcon. Since it is not known which boot console
3873		 * is the same device, flush all consoles and, if necessary,
3874		 * start with the message of the enabled boot console that is
3875		 * the furthest behind.
3876		 */
3877		if (bootcon_registered && !keep_bootcon) {
3878			/*
3879			 * Hold the console_lock to stop console printing and
3880			 * guarantee safe access to console->seq.
3881			 */
3882			console_lock();
3883
3884			/*
3885			 * Flush all consoles and set the console to start at
3886			 * the next unprinted sequence number.
3887			 */
3888			if (!console_flush_all(true, &init_seq, &handover)) {
3889				/*
3890				 * Flushing failed. Just choose the lowest
3891				 * sequence of the enabled boot consoles.
3892				 */
3893
3894				/*
3895				 * If there was a handover, this context no
3896				 * longer holds the console_lock.
3897				 */
3898				if (handover)
3899					console_lock();
3900
3901				init_seq = prb_next_seq(prb);
3902				for_each_console(con) {
3903					u64 seq;
3904
3905					if (!(con->flags & CON_BOOT) ||
3906					    !(con->flags & CON_ENABLED)) {
3907						continue;
3908					}
3909
3910					if (con->flags & CON_NBCON)
3911						seq = nbcon_seq_read(con);
3912					else
3913						seq = con->seq;
3914
3915					if (seq < init_seq)
3916						init_seq = seq;
3917				}
3918			}
3919
3920			console_unlock();
3921		}
3922	}
3923
3924	return init_seq;
3925}
3926
3927#define console_first()				\
3928	hlist_entry(console_list.first, struct console, node)
3929
3930static int unregister_console_locked(struct console *console);
3931
3932/*
3933 * The console driver calls this routine during kernel initialization
3934 * to register the console printing procedure with printk() and to
3935 * print any messages that were printed by the kernel before the
3936 * console driver was initialized.
3937 *
3938 * This can happen pretty early during the boot process (because of
3939 * early_printk) - sometimes before setup_arch() completes - be careful
3940 * of what kernel features are used - they may not be initialised yet.
3941 *
3942 * There are two types of consoles - bootconsoles (early_printk) and
3943 * "real" consoles (everything which is not a bootconsole) which are
3944 * handled differently.
3945 *  - Any number of bootconsoles can be registered at any time.
3946 *  - As soon as a "real" console is registered, all bootconsoles
3947 *    will be unregistered automatically.
3948 *  - Once a "real" console is registered, any attempt to register a
3949 *    bootconsoles will be rejected
3950 */
3951void register_console(struct console *newcon)
3952{
3953	bool use_device_lock = (newcon->flags & CON_NBCON) && newcon->write_atomic;
3954	bool bootcon_registered = false;
3955	bool realcon_registered = false;
3956	struct console *con;
3957	unsigned long flags;
3958	u64 init_seq;
3959	int err;
3960
3961	console_list_lock();
3962
3963	for_each_console(con) {
3964		if (WARN(con == newcon, "console '%s%d' already registered\n",
3965					 con->name, con->index)) {
3966			goto unlock;
3967		}
3968
3969		if (con->flags & CON_BOOT)
3970			bootcon_registered = true;
3971		else
3972			realcon_registered = true;
3973	}
3974
3975	/* Do not register boot consoles when there already is a real one. */
3976	if ((newcon->flags & CON_BOOT) && realcon_registered) {
3977		pr_info("Too late to register bootconsole %s%d\n",
3978			newcon->name, newcon->index);
3979		goto unlock;
3980	}
3981
3982	if (newcon->flags & CON_NBCON) {
3983		/*
3984		 * Ensure the nbcon console buffers can be allocated
3985		 * before modifying any global data.
3986		 */
3987		if (!nbcon_alloc(newcon))
3988			goto unlock;
3989	}
3990
3991	/*
3992	 * See if we want to enable this console driver by default.
3993	 *
3994	 * Nope when a console is preferred by the command line, device
3995	 * tree, or SPCR.
3996	 *
3997	 * The first real console with tty binding (driver) wins. More
3998	 * consoles might get enabled before the right one is found.
3999	 *
4000	 * Note that a console with tty binding will have CON_CONSDEV
4001	 * flag set and will be first in the list.
4002	 */
4003	if (preferred_console < 0) {
4004		if (hlist_empty(&console_list) || !console_first()->device ||
4005		    console_first()->flags & CON_BOOT) {
4006			try_enable_default_console(newcon);
4007		}
4008	}
4009
4010	/* See if this console matches one we selected on the command line */
4011	err = try_enable_preferred_console(newcon, true);
4012
4013	/* If not, try to match against the platform default(s) */
4014	if (err == -ENOENT)
4015		err = try_enable_preferred_console(newcon, false);
4016
4017	/* printk() messages are not printed to the Braille console. */
4018	if (err || newcon->flags & CON_BRL) {
4019		if (newcon->flags & CON_NBCON)
4020			nbcon_free(newcon);
4021		goto unlock;
4022	}
4023
4024	/*
4025	 * If we have a bootconsole, and are switching to a real console,
4026	 * don't print everything out again, since when the boot console, and
4027	 * the real console are the same physical device, it's annoying to
4028	 * see the beginning boot messages twice
4029	 */
4030	if (bootcon_registered &&
4031	    ((newcon->flags & (CON_CONSDEV | CON_BOOT)) == CON_CONSDEV)) {
4032		newcon->flags &= ~CON_PRINTBUFFER;
4033	}
4034
4035	newcon->dropped = 0;
4036	init_seq = get_init_console_seq(newcon, bootcon_registered);
4037
4038	if (newcon->flags & CON_NBCON) {
4039		have_nbcon_console = true;
4040		nbcon_seq_force(newcon, init_seq);
4041	} else {
4042		have_legacy_console = true;
4043		newcon->seq = init_seq;
4044	}
4045
4046	if (newcon->flags & CON_BOOT)
4047		have_boot_console = true;
4048
4049	/*
4050	 * If another context is actively using the hardware of this new
4051	 * console, it will not be aware of the nbcon synchronization. This
4052	 * is a risk that two contexts could access the hardware
4053	 * simultaneously if this new console is used for atomic printing
4054	 * and the other context is still using the hardware.
4055	 *
4056	 * Use the driver synchronization to ensure that the hardware is not
4057	 * in use while this new console transitions to being registered.
4058	 */
4059	if (use_device_lock)
4060		newcon->device_lock(newcon, &flags);
4061
4062	/*
4063	 * Put this console in the list - keep the
4064	 * preferred driver at the head of the list.
4065	 */
4066	if (hlist_empty(&console_list)) {
4067		/* Ensure CON_CONSDEV is always set for the head. */
4068		newcon->flags |= CON_CONSDEV;
4069		hlist_add_head_rcu(&newcon->node, &console_list);
4070
4071	} else if (newcon->flags & CON_CONSDEV) {
4072		/* Only the new head can have CON_CONSDEV set. */
4073		console_srcu_write_flags(console_first(), console_first()->flags & ~CON_CONSDEV);
4074		hlist_add_head_rcu(&newcon->node, &console_list);
4075
4076	} else {
4077		hlist_add_behind_rcu(&newcon->node, console_list.first);
4078	}
4079
4080	/*
4081	 * No need to synchronize SRCU here! The caller does not rely
4082	 * on all contexts being able to see the new console before
4083	 * register_console() completes.
4084	 */
4085
4086	/* This new console is now registered. */
4087	if (use_device_lock)
4088		newcon->device_unlock(newcon, flags);
4089
4090	console_sysfs_notify();
4091
4092	/*
4093	 * By unregistering the bootconsoles after we enable the real console
4094	 * we get the "console xxx enabled" message on all the consoles -
4095	 * boot consoles, real consoles, etc - this is to ensure that end
4096	 * users know there might be something in the kernel's log buffer that
4097	 * went to the bootconsole (that they do not see on the real console)
4098	 */
4099	con_printk(KERN_INFO, newcon, "enabled\n");
4100	if (bootcon_registered &&
4101	    ((newcon->flags & (CON_CONSDEV | CON_BOOT)) == CON_CONSDEV) &&
4102	    !keep_bootcon) {
4103		struct hlist_node *tmp;
4104
4105		hlist_for_each_entry_safe(con, tmp, &console_list, node) {
4106			if (con->flags & CON_BOOT)
4107				unregister_console_locked(con);
4108		}
4109	}
4110
4111	/* Changed console list, may require printer threads to start/stop. */
4112	printk_kthreads_check_locked();
4113unlock:
4114	console_list_unlock();
4115}
4116EXPORT_SYMBOL(register_console);
4117
4118/* Must be called under console_list_lock(). */
4119static int unregister_console_locked(struct console *console)
4120{
4121	bool use_device_lock = (console->flags & CON_NBCON) && console->write_atomic;
4122	bool found_legacy_con = false;
4123	bool found_nbcon_con = false;
4124	bool found_boot_con = false;
4125	unsigned long flags;
4126	struct console *c;
4127	int res;
4128
4129	lockdep_assert_console_list_lock_held();
4130
4131	con_printk(KERN_INFO, console, "disabled\n");
4132
4133	res = _braille_unregister_console(console);
4134	if (res < 0)
4135		return res;
4136	if (res > 0)
4137		return 0;
4138
4139	if (!console_is_registered_locked(console))
4140		res = -ENODEV;
4141	else if (console_is_usable(console, console->flags, true))
4142		__pr_flush(console, 1000, true);
4143
4144	/* Disable it unconditionally */
4145	console_srcu_write_flags(console, console->flags & ~CON_ENABLED);
4146
4147	if (res < 0)
4148		return res;
4149
4150	/*
4151	 * Use the driver synchronization to ensure that the hardware is not
4152	 * in use while this console transitions to being unregistered.
4153	 */
4154	if (use_device_lock)
4155		console->device_lock(console, &flags);
4156
4157	hlist_del_init_rcu(&console->node);
4158
4159	if (use_device_lock)
4160		console->device_unlock(console, flags);
4161
4162	/*
4163	 * <HISTORICAL>
4164	 * If this isn't the last console and it has CON_CONSDEV set, we
4165	 * need to set it on the next preferred console.
4166	 * </HISTORICAL>
4167	 *
4168	 * The above makes no sense as there is no guarantee that the next
4169	 * console has any device attached. Oh well....
4170	 */
4171	if (!hlist_empty(&console_list) && console->flags & CON_CONSDEV)
4172		console_srcu_write_flags(console_first(), console_first()->flags | CON_CONSDEV);
4173
4174	/*
4175	 * Ensure that all SRCU list walks have completed. All contexts
4176	 * must not be able to see this console in the list so that any
4177	 * exit/cleanup routines can be performed safely.
4178	 */
4179	synchronize_srcu(&console_srcu);
4180
4181	if (console->flags & CON_NBCON)
4182		nbcon_free(console);
4183
4184	console_sysfs_notify();
4185
4186	if (console->exit)
4187		res = console->exit(console);
4188
4189	/*
4190	 * With this console gone, the global flags tracking registered
4191	 * console types may have changed. Update them.
4192	 */
4193	for_each_console(c) {
4194		if (c->flags & CON_BOOT)
4195			found_boot_con = true;
4196
4197		if (c->flags & CON_NBCON)
4198			found_nbcon_con = true;
4199		else
4200			found_legacy_con = true;
4201	}
4202	if (!found_boot_con)
4203		have_boot_console = found_boot_con;
4204	if (!found_legacy_con)
4205		have_legacy_console = found_legacy_con;
4206	if (!found_nbcon_con)
4207		have_nbcon_console = found_nbcon_con;
4208
4209	/* Changed console list, may require printer threads to start/stop. */
4210	printk_kthreads_check_locked();
4211
4212	return res;
4213}
4214
4215int unregister_console(struct console *console)
4216{
4217	int res;
4218
4219	console_list_lock();
4220	res = unregister_console_locked(console);
4221	console_list_unlock();
4222	return res;
4223}
4224EXPORT_SYMBOL(unregister_console);
4225
4226/**
4227 * console_force_preferred_locked - force a registered console preferred
4228 * @con: The registered console to force preferred.
4229 *
4230 * Must be called under console_list_lock().
4231 */
4232void console_force_preferred_locked(struct console *con)
4233{
4234	struct console *cur_pref_con;
4235
4236	if (!console_is_registered_locked(con))
4237		return;
4238
4239	cur_pref_con = console_first();
4240
4241	/* Already preferred? */
4242	if (cur_pref_con == con)
4243		return;
4244
4245	/*
4246	 * Delete, but do not re-initialize the entry. This allows the console
4247	 * to continue to appear registered (via any hlist_unhashed_lockless()
4248	 * checks), even though it was briefly removed from the console list.
4249	 */
4250	hlist_del_rcu(&con->node);
4251
4252	/*
4253	 * Ensure that all SRCU list walks have completed so that the console
4254	 * can be added to the beginning of the console list and its forward
4255	 * list pointer can be re-initialized.
4256	 */
4257	synchronize_srcu(&console_srcu);
4258
4259	con->flags |= CON_CONSDEV;
4260	WARN_ON(!con->device);
4261
4262	/* Only the new head can have CON_CONSDEV set. */
4263	console_srcu_write_flags(cur_pref_con, cur_pref_con->flags & ~CON_CONSDEV);
4264	hlist_add_head_rcu(&con->node, &console_list);
4265}
4266EXPORT_SYMBOL(console_force_preferred_locked);
4267
4268/*
4269 * Initialize the console device. This is called *early*, so
4270 * we can't necessarily depend on lots of kernel help here.
4271 * Just do some early initializations, and do the complex setup
4272 * later.
4273 */
4274void __init console_init(void)
4275{
4276	int ret;
4277	initcall_t call;
4278	initcall_entry_t *ce;
4279
4280	/* Setup the default TTY line discipline. */
4281	n_tty_init();
4282
4283	/*
4284	 * set up the console device so that later boot sequences can
4285	 * inform about problems etc..
4286	 */
4287	ce = __con_initcall_start;
4288	trace_initcall_level("console");
4289	while (ce < __con_initcall_end) {
4290		call = initcall_from_entry(ce);
4291		trace_initcall_start(call);
4292		ret = call();
4293		trace_initcall_finish(call, ret);
4294		ce++;
4295	}
4296}
4297
4298/*
4299 * Some boot consoles access data that is in the init section and which will
4300 * be discarded after the initcalls have been run. To make sure that no code
4301 * will access this data, unregister the boot consoles in a late initcall.
4302 *
4303 * If for some reason, such as deferred probe or the driver being a loadable
4304 * module, the real console hasn't registered yet at this point, there will
4305 * be a brief interval in which no messages are logged to the console, which
4306 * makes it difficult to diagnose problems that occur during this time.
4307 *
4308 * To mitigate this problem somewhat, only unregister consoles whose memory
4309 * intersects with the init section. Note that all other boot consoles will
4310 * get unregistered when the real preferred console is registered.
4311 */
4312static int __init printk_late_init(void)
4313{
4314	struct hlist_node *tmp;
4315	struct console *con;
4316	int ret;
4317
4318	console_list_lock();
4319	hlist_for_each_entry_safe(con, tmp, &console_list, node) {
4320		if (!(con->flags & CON_BOOT))
4321			continue;
4322
4323		/* Check addresses that might be used for enabled consoles. */
4324		if (init_section_intersects(con, sizeof(*con)) ||
4325		    init_section_contains(con->write, 0) ||
4326		    init_section_contains(con->read, 0) ||
4327		    init_section_contains(con->device, 0) ||
4328		    init_section_contains(con->unblank, 0) ||
4329		    init_section_contains(con->data, 0)) {
4330			/*
4331			 * Please, consider moving the reported consoles out
4332			 * of the init section.
4333			 */
4334			pr_warn("bootconsole [%s%d] uses init memory and must be disabled even before the real one is ready\n",
4335				con->name, con->index);
4336			unregister_console_locked(con);
4337		}
4338	}
4339	console_list_unlock();
4340
4341	ret = cpuhp_setup_state_nocalls(CPUHP_PRINTK_DEAD, "printk:dead", NULL,
4342					console_cpu_notify);
4343	WARN_ON(ret < 0);
4344	ret = cpuhp_setup_state_nocalls(CPUHP_AP_ONLINE_DYN, "printk:online",
4345					console_cpu_notify, NULL);
4346	WARN_ON(ret < 0);
4347	printk_sysctl_init();
4348	return 0;
4349}
4350late_initcall(printk_late_init);
4351
4352#if defined CONFIG_PRINTK
4353/* If @con is specified, only wait for that console. Otherwise wait for all. */
4354static bool __pr_flush(struct console *con, int timeout_ms, bool reset_on_progress)
4355{
4356	unsigned long timeout_jiffies = msecs_to_jiffies(timeout_ms);
4357	unsigned long remaining_jiffies = timeout_jiffies;
4358	struct console_flush_type ft;
4359	struct console *c;
4360	u64 last_diff = 0;
4361	u64 printk_seq;
4362	short flags;
4363	int cookie;
4364	u64 diff;
4365	u64 seq;
4366
4367	/* Sorry, pr_flush() will not work this early. */
4368	if (system_state < SYSTEM_SCHEDULING)
4369		return false;
4370
4371	might_sleep();
4372
4373	seq = prb_next_reserve_seq(prb);
4374
4375	/* Flush the consoles so that records up to @seq are printed. */
4376	printk_get_console_flush_type(&ft);
4377	if (ft.nbcon_atomic)
4378		nbcon_atomic_flush_pending();
4379	if (ft.legacy_direct) {
4380		console_lock();
4381		console_unlock();
4382	}
4383
4384	for (;;) {
4385		unsigned long begin_jiffies;
4386		unsigned long slept_jiffies;
4387
4388		diff = 0;
4389
4390		/*
4391		 * Hold the console_lock to guarantee safe access to
4392		 * console->seq. Releasing console_lock flushes more
4393		 * records in case @seq is still not printed on all
4394		 * usable consoles.
4395		 *
4396		 * Holding the console_lock is not necessary if there
4397		 * are no legacy or boot consoles. However, such a
4398		 * console could register at any time. Always hold the
4399		 * console_lock as a precaution rather than
4400		 * synchronizing against register_console().
4401		 */
4402		console_lock();
4403
4404		cookie = console_srcu_read_lock();
4405		for_each_console_srcu(c) {
4406			if (con && con != c)
4407				continue;
4408
4409			flags = console_srcu_read_flags(c);
4410
4411			/*
4412			 * If consoles are not usable, it cannot be expected
4413			 * that they make forward progress, so only increment
4414			 * @diff for usable consoles.
4415			 */
4416			if (!console_is_usable(c, flags, true) &&
4417			    !console_is_usable(c, flags, false)) {
4418				continue;
4419			}
4420
4421			if (flags & CON_NBCON) {
4422				printk_seq = nbcon_seq_read(c);
4423			} else {
4424				printk_seq = c->seq;
4425			}
4426
4427			if (printk_seq < seq)
4428				diff += seq - printk_seq;
4429		}
4430		console_srcu_read_unlock(cookie);
4431
4432		if (diff != last_diff && reset_on_progress)
4433			remaining_jiffies = timeout_jiffies;
 
 
 
 
 
 
 
4434
4435		console_unlock();
4436
4437		/* Note: @diff is 0 if there are no usable consoles. */
4438		if (diff == 0 || remaining_jiffies == 0)
4439			break;
4440
4441		/* msleep(1) might sleep much longer. Check time by jiffies. */
4442		begin_jiffies = jiffies;
4443		msleep(1);
4444		slept_jiffies = jiffies - begin_jiffies;
4445
4446		remaining_jiffies -= min(slept_jiffies, remaining_jiffies);
 
 
 
 
4447
4448		last_diff = diff;
4449	}
4450
4451	return (diff == 0);
4452}
4453
4454/**
4455 * pr_flush() - Wait for printing threads to catch up.
4456 *
4457 * @timeout_ms:        The maximum time (in ms) to wait.
4458 * @reset_on_progress: Reset the timeout if forward progress is seen.
4459 *
4460 * A value of 0 for @timeout_ms means no waiting will occur. A value of -1
4461 * represents infinite waiting.
4462 *
4463 * If @reset_on_progress is true, the timeout will be reset whenever any
4464 * printer has been seen to make some forward progress.
4465 *
4466 * Context: Process context. May sleep while acquiring console lock.
4467 * Return: true if all usable printers are caught up.
4468 */
4469static bool pr_flush(int timeout_ms, bool reset_on_progress)
4470{
4471	return __pr_flush(NULL, timeout_ms, reset_on_progress);
4472}
4473
4474/*
4475 * Delayed printk version, for scheduler-internal messages:
4476 */
4477#define PRINTK_PENDING_WAKEUP	0x01
4478#define PRINTK_PENDING_OUTPUT	0x02
4479
4480static DEFINE_PER_CPU(int, printk_pending);
4481
4482static void wake_up_klogd_work_func(struct irq_work *irq_work)
4483{
4484	int pending = this_cpu_xchg(printk_pending, 0);
4485
4486	if (pending & PRINTK_PENDING_OUTPUT) {
4487		if (force_legacy_kthread()) {
4488			if (printk_legacy_kthread)
4489				wake_up_interruptible(&legacy_wait);
4490		} else {
4491			if (console_trylock())
4492				console_unlock();
4493		}
4494	}
4495
4496	if (pending & PRINTK_PENDING_WAKEUP)
4497		wake_up_interruptible(&log_wait);
4498}
4499
4500static DEFINE_PER_CPU(struct irq_work, wake_up_klogd_work) =
4501	IRQ_WORK_INIT_LAZY(wake_up_klogd_work_func);
4502
4503static void __wake_up_klogd(int val)
4504{
4505	if (!printk_percpu_data_ready())
4506		return;
4507
4508	preempt_disable();
4509	/*
4510	 * Guarantee any new records can be seen by tasks preparing to wait
4511	 * before this context checks if the wait queue is empty.
4512	 *
4513	 * The full memory barrier within wq_has_sleeper() pairs with the full
4514	 * memory barrier within set_current_state() of
4515	 * prepare_to_wait_event(), which is called after ___wait_event() adds
4516	 * the waiter but before it has checked the wait condition.
4517	 *
4518	 * This pairs with devkmsg_read:A and syslog_print:A.
4519	 */
4520	if (wq_has_sleeper(&log_wait) || /* LMM(__wake_up_klogd:A) */
4521	    (val & PRINTK_PENDING_OUTPUT)) {
4522		this_cpu_or(printk_pending, val);
4523		irq_work_queue(this_cpu_ptr(&wake_up_klogd_work));
4524	}
4525	preempt_enable();
4526}
4527
4528/**
4529 * wake_up_klogd - Wake kernel logging daemon
4530 *
4531 * Use this function when new records have been added to the ringbuffer
4532 * and the console printing of those records has already occurred or is
4533 * known to be handled by some other context. This function will only
4534 * wake the logging daemon.
4535 *
4536 * Context: Any context.
4537 */
4538void wake_up_klogd(void)
4539{
4540	__wake_up_klogd(PRINTK_PENDING_WAKEUP);
4541}
4542
4543/**
4544 * defer_console_output - Wake kernel logging daemon and trigger
4545 *	console printing in a deferred context
4546 *
4547 * Use this function when new records have been added to the ringbuffer,
4548 * this context is responsible for console printing those records, but
4549 * the current context is not allowed to perform the console printing.
4550 * Trigger an irq_work context to perform the console printing. This
4551 * function also wakes the logging daemon.
4552 *
4553 * Context: Any context.
4554 */
4555void defer_console_output(void)
4556{
4557	/*
4558	 * New messages may have been added directly to the ringbuffer
4559	 * using vprintk_store(), so wake any waiters as well.
4560	 */
4561	__wake_up_klogd(PRINTK_PENDING_WAKEUP | PRINTK_PENDING_OUTPUT);
4562}
4563
4564void printk_trigger_flush(void)
4565{
4566	defer_console_output();
4567}
4568
4569int vprintk_deferred(const char *fmt, va_list args)
4570{
4571	return vprintk_emit(0, LOGLEVEL_SCHED, NULL, fmt, args);
 
 
 
 
 
4572}
4573
4574int _printk_deferred(const char *fmt, ...)
4575{
4576	va_list args;
4577	int r;
4578
4579	va_start(args, fmt);
4580	r = vprintk_deferred(fmt, args);
4581	va_end(args);
4582
4583	return r;
4584}
4585
4586/*
4587 * printk rate limiting, lifted from the networking subsystem.
4588 *
4589 * This enforces a rate limit: not more than 10 kernel messages
4590 * every 5s to make a denial-of-service attack impossible.
4591 */
4592DEFINE_RATELIMIT_STATE(printk_ratelimit_state, 5 * HZ, 10);
4593
4594int __printk_ratelimit(const char *func)
4595{
4596	return ___ratelimit(&printk_ratelimit_state, func);
4597}
4598EXPORT_SYMBOL(__printk_ratelimit);
4599
4600/**
4601 * printk_timed_ratelimit - caller-controlled printk ratelimiting
4602 * @caller_jiffies: pointer to caller's state
4603 * @interval_msecs: minimum interval between prints
4604 *
4605 * printk_timed_ratelimit() returns true if more than @interval_msecs
4606 * milliseconds have elapsed since the last time printk_timed_ratelimit()
4607 * returned true.
4608 */
4609bool printk_timed_ratelimit(unsigned long *caller_jiffies,
4610			unsigned int interval_msecs)
4611{
4612	unsigned long elapsed = jiffies - *caller_jiffies;
4613
4614	if (*caller_jiffies && elapsed <= msecs_to_jiffies(interval_msecs))
4615		return false;
4616
4617	*caller_jiffies = jiffies;
4618	return true;
4619}
4620EXPORT_SYMBOL(printk_timed_ratelimit);
4621
4622static DEFINE_SPINLOCK(dump_list_lock);
4623static LIST_HEAD(dump_list);
4624
4625/**
4626 * kmsg_dump_register - register a kernel log dumper.
4627 * @dumper: pointer to the kmsg_dumper structure
4628 *
4629 * Adds a kernel log dumper to the system. The dump callback in the
4630 * structure will be called when the kernel oopses or panics and must be
4631 * set. Returns zero on success and %-EINVAL or %-EBUSY otherwise.
4632 */
4633int kmsg_dump_register(struct kmsg_dumper *dumper)
4634{
4635	unsigned long flags;
4636	int err = -EBUSY;
4637
4638	/* The dump callback needs to be set */
4639	if (!dumper->dump)
4640		return -EINVAL;
4641
4642	spin_lock_irqsave(&dump_list_lock, flags);
4643	/* Don't allow registering multiple times */
4644	if (!dumper->registered) {
4645		dumper->registered = 1;
4646		list_add_tail_rcu(&dumper->list, &dump_list);
4647		err = 0;
4648	}
4649	spin_unlock_irqrestore(&dump_list_lock, flags);
4650
4651	return err;
4652}
4653EXPORT_SYMBOL_GPL(kmsg_dump_register);
4654
4655/**
4656 * kmsg_dump_unregister - unregister a kmsg dumper.
4657 * @dumper: pointer to the kmsg_dumper structure
4658 *
4659 * Removes a dump device from the system. Returns zero on success and
4660 * %-EINVAL otherwise.
4661 */
4662int kmsg_dump_unregister(struct kmsg_dumper *dumper)
4663{
4664	unsigned long flags;
4665	int err = -EINVAL;
4666
4667	spin_lock_irqsave(&dump_list_lock, flags);
4668	if (dumper->registered) {
4669		dumper->registered = 0;
4670		list_del_rcu(&dumper->list);
4671		err = 0;
4672	}
4673	spin_unlock_irqrestore(&dump_list_lock, flags);
4674	synchronize_rcu();
4675
4676	return err;
4677}
4678EXPORT_SYMBOL_GPL(kmsg_dump_unregister);
4679
4680static bool always_kmsg_dump;
4681module_param_named(always_kmsg_dump, always_kmsg_dump, bool, S_IRUGO | S_IWUSR);
4682
4683const char *kmsg_dump_reason_str(enum kmsg_dump_reason reason)
4684{
4685	switch (reason) {
4686	case KMSG_DUMP_PANIC:
4687		return "Panic";
4688	case KMSG_DUMP_OOPS:
4689		return "Oops";
4690	case KMSG_DUMP_EMERG:
4691		return "Emergency";
4692	case KMSG_DUMP_SHUTDOWN:
4693		return "Shutdown";
4694	default:
4695		return "Unknown";
4696	}
4697}
4698EXPORT_SYMBOL_GPL(kmsg_dump_reason_str);
4699
4700/**
4701 * kmsg_dump_desc - dump kernel log to kernel message dumpers.
4702 * @reason: the reason (oops, panic etc) for dumping
4703 * @desc: a short string to describe what caused the panic or oops. Can be NULL
4704 * if no additional description is available.
4705 *
4706 * Call each of the registered dumper's dump() callback, which can
4707 * retrieve the kmsg records with kmsg_dump_get_line() or
4708 * kmsg_dump_get_buffer().
4709 */
4710void kmsg_dump_desc(enum kmsg_dump_reason reason, const char *desc)
4711{
4712	struct kmsg_dumper *dumper;
4713	struct kmsg_dump_detail detail = {
4714		.reason = reason,
4715		.description = desc};
4716
4717	rcu_read_lock();
4718	list_for_each_entry_rcu(dumper, &dump_list, list) {
4719		enum kmsg_dump_reason max_reason = dumper->max_reason;
4720
4721		/*
4722		 * If client has not provided a specific max_reason, default
4723		 * to KMSG_DUMP_OOPS, unless always_kmsg_dump was set.
4724		 */
4725		if (max_reason == KMSG_DUMP_UNDEF) {
4726			max_reason = always_kmsg_dump ? KMSG_DUMP_MAX :
4727							KMSG_DUMP_OOPS;
4728		}
4729		if (reason > max_reason)
4730			continue;
4731
4732		/* invoke dumper which will iterate over records */
4733		dumper->dump(dumper, &detail);
4734	}
4735	rcu_read_unlock();
4736}
4737
4738/**
4739 * kmsg_dump_get_line - retrieve one kmsg log line
4740 * @iter: kmsg dump iterator
4741 * @syslog: include the "<4>" prefixes
4742 * @line: buffer to copy the line to
4743 * @size: maximum size of the buffer
4744 * @len: length of line placed into buffer
4745 *
4746 * Start at the beginning of the kmsg buffer, with the oldest kmsg
4747 * record, and copy one record into the provided buffer.
4748 *
4749 * Consecutive calls will return the next available record moving
4750 * towards the end of the buffer with the youngest messages.
4751 *
4752 * A return value of FALSE indicates that there are no more records to
4753 * read.
4754 */
4755bool kmsg_dump_get_line(struct kmsg_dump_iter *iter, bool syslog,
4756			char *line, size_t size, size_t *len)
4757{
4758	u64 min_seq = latched_seq_read_nolock(&clear_seq);
4759	struct printk_info info;
4760	unsigned int line_count;
4761	struct printk_record r;
4762	size_t l = 0;
4763	bool ret = false;
4764
4765	if (iter->cur_seq < min_seq)
4766		iter->cur_seq = min_seq;
4767
4768	prb_rec_init_rd(&r, &info, line, size);
4769
4770	/* Read text or count text lines? */
4771	if (line) {
4772		if (!prb_read_valid(prb, iter->cur_seq, &r))
4773			goto out;
4774		l = record_print_text(&r, syslog, printk_time);
4775	} else {
4776		if (!prb_read_valid_info(prb, iter->cur_seq,
4777					 &info, &line_count)) {
4778			goto out;
4779		}
4780		l = get_record_print_text_size(&info, line_count, syslog,
4781					       printk_time);
4782
4783	}
4784
4785	iter->cur_seq = r.info->seq + 1;
4786	ret = true;
4787out:
4788	if (len)
4789		*len = l;
4790	return ret;
4791}
4792EXPORT_SYMBOL_GPL(kmsg_dump_get_line);
4793
4794/**
4795 * kmsg_dump_get_buffer - copy kmsg log lines
4796 * @iter: kmsg dump iterator
4797 * @syslog: include the "<4>" prefixes
4798 * @buf: buffer to copy the line to
4799 * @size: maximum size of the buffer
4800 * @len_out: length of line placed into buffer
4801 *
4802 * Start at the end of the kmsg buffer and fill the provided buffer
4803 * with as many of the *youngest* kmsg records that fit into it.
4804 * If the buffer is large enough, all available kmsg records will be
4805 * copied with a single call.
4806 *
4807 * Consecutive calls will fill the buffer with the next block of
4808 * available older records, not including the earlier retrieved ones.
4809 *
4810 * A return value of FALSE indicates that there are no more records to
4811 * read.
4812 */
4813bool kmsg_dump_get_buffer(struct kmsg_dump_iter *iter, bool syslog,
4814			  char *buf, size_t size, size_t *len_out)
4815{
4816	u64 min_seq = latched_seq_read_nolock(&clear_seq);
4817	struct printk_info info;
4818	struct printk_record r;
4819	u64 seq;
4820	u64 next_seq;
4821	size_t len = 0;
4822	bool ret = false;
4823	bool time = printk_time;
4824
4825	if (!buf || !size)
4826		goto out;
4827
4828	if (iter->cur_seq < min_seq)
4829		iter->cur_seq = min_seq;
4830
4831	if (prb_read_valid_info(prb, iter->cur_seq, &info, NULL)) {
4832		if (info.seq != iter->cur_seq) {
4833			/* messages are gone, move to first available one */
4834			iter->cur_seq = info.seq;
4835		}
4836	}
4837
4838	/* last entry */
4839	if (iter->cur_seq >= iter->next_seq)
4840		goto out;
4841
4842	/*
4843	 * Find first record that fits, including all following records,
4844	 * into the user-provided buffer for this dump. Pass in size-1
4845	 * because this function (by way of record_print_text()) will
4846	 * not write more than size-1 bytes of text into @buf.
4847	 */
4848	seq = find_first_fitting_seq(iter->cur_seq, iter->next_seq,
4849				     size - 1, syslog, time);
4850
4851	/*
4852	 * Next kmsg_dump_get_buffer() invocation will dump block of
4853	 * older records stored right before this one.
4854	 */
4855	next_seq = seq;
4856
4857	prb_rec_init_rd(&r, &info, buf, size);
4858
 
4859	prb_for_each_record(seq, prb, seq, &r) {
4860		if (r.info->seq >= iter->next_seq)
4861			break;
4862
4863		len += record_print_text(&r, syslog, time);
4864
4865		/* Adjust record to store to remaining buffer space. */
4866		prb_rec_init_rd(&r, &info, buf + len, size - len);
4867	}
4868
4869	iter->next_seq = next_seq;
4870	ret = true;
4871out:
4872	if (len_out)
4873		*len_out = len;
4874	return ret;
4875}
4876EXPORT_SYMBOL_GPL(kmsg_dump_get_buffer);
4877
4878/**
4879 * kmsg_dump_rewind - reset the iterator
4880 * @iter: kmsg dump iterator
4881 *
4882 * Reset the dumper's iterator so that kmsg_dump_get_line() and
4883 * kmsg_dump_get_buffer() can be called again and used multiple
4884 * times within the same dumper.dump() callback.
4885 */
4886void kmsg_dump_rewind(struct kmsg_dump_iter *iter)
4887{
4888	iter->cur_seq = latched_seq_read_nolock(&clear_seq);
4889	iter->next_seq = prb_next_seq(prb);
4890}
4891EXPORT_SYMBOL_GPL(kmsg_dump_rewind);
4892
4893/**
4894 * console_try_replay_all - try to replay kernel log on consoles
4895 *
4896 * Try to obtain lock on console subsystem and replay all
4897 * available records in printk buffer on the consoles.
4898 * Does nothing if lock is not obtained.
4899 *
4900 * Context: Any, except for NMI.
4901 */
4902void console_try_replay_all(void)
4903{
4904	struct console_flush_type ft;
4905
4906	printk_get_console_flush_type(&ft);
4907	if (console_trylock()) {
4908		__console_rewind_all();
4909		if (ft.nbcon_atomic)
4910			nbcon_atomic_flush_pending();
4911		if (ft.nbcon_offload)
4912			nbcon_kthreads_wake();
4913		if (ft.legacy_offload)
4914			defer_console_output();
4915		/* Consoles are flushed as part of console_unlock(). */
4916		console_unlock();
4917	}
4918}
4919#endif
4920
4921#ifdef CONFIG_SMP
4922static atomic_t printk_cpu_sync_owner = ATOMIC_INIT(-1);
4923static atomic_t printk_cpu_sync_nested = ATOMIC_INIT(0);
4924
4925bool is_printk_cpu_sync_owner(void)
4926{
4927	return (atomic_read(&printk_cpu_sync_owner) == raw_smp_processor_id());
4928}
4929
4930/**
4931 * __printk_cpu_sync_wait() - Busy wait until the printk cpu-reentrant
4932 *                            spinning lock is not owned by any CPU.
4933 *
4934 * Context: Any context.
4935 */
4936void __printk_cpu_sync_wait(void)
4937{
4938	do {
4939		cpu_relax();
4940	} while (atomic_read(&printk_cpu_sync_owner) != -1);
4941}
4942EXPORT_SYMBOL(__printk_cpu_sync_wait);
4943
4944/**
4945 * __printk_cpu_sync_try_get() - Try to acquire the printk cpu-reentrant
4946 *                               spinning lock.
4947 *
4948 * If no processor has the lock, the calling processor takes the lock and
4949 * becomes the owner. If the calling processor is already the owner of the
4950 * lock, this function succeeds immediately.
4951 *
4952 * Context: Any context. Expects interrupts to be disabled.
4953 * Return: 1 on success, otherwise 0.
4954 */
4955int __printk_cpu_sync_try_get(void)
4956{
4957	int cpu;
4958	int old;
4959
4960	cpu = smp_processor_id();
4961
4962	/*
4963	 * Guarantee loads and stores from this CPU when it is the lock owner
4964	 * are _not_ visible to the previous lock owner. This pairs with
4965	 * __printk_cpu_sync_put:B.
4966	 *
4967	 * Memory barrier involvement:
4968	 *
4969	 * If __printk_cpu_sync_try_get:A reads from __printk_cpu_sync_put:B,
4970	 * then __printk_cpu_sync_put:A can never read from
4971	 * __printk_cpu_sync_try_get:B.
4972	 *
4973	 * Relies on:
4974	 *
4975	 * RELEASE from __printk_cpu_sync_put:A to __printk_cpu_sync_put:B
4976	 * of the previous CPU
4977	 *    matching
4978	 * ACQUIRE from __printk_cpu_sync_try_get:A to
4979	 * __printk_cpu_sync_try_get:B of this CPU
4980	 */
4981	old = atomic_cmpxchg_acquire(&printk_cpu_sync_owner, -1,
4982				     cpu); /* LMM(__printk_cpu_sync_try_get:A) */
4983	if (old == -1) {
4984		/*
4985		 * This CPU is now the owner and begins loading/storing
4986		 * data: LMM(__printk_cpu_sync_try_get:B)
4987		 */
4988		return 1;
4989
4990	} else if (old == cpu) {
4991		/* This CPU is already the owner. */
4992		atomic_inc(&printk_cpu_sync_nested);
4993		return 1;
4994	}
4995
4996	return 0;
4997}
4998EXPORT_SYMBOL(__printk_cpu_sync_try_get);
4999
5000/**
5001 * __printk_cpu_sync_put() - Release the printk cpu-reentrant spinning lock.
5002 *
5003 * The calling processor must be the owner of the lock.
5004 *
5005 * Context: Any context. Expects interrupts to be disabled.
5006 */
5007void __printk_cpu_sync_put(void)
5008{
5009	if (atomic_read(&printk_cpu_sync_nested)) {
5010		atomic_dec(&printk_cpu_sync_nested);
5011		return;
5012	}
5013
5014	/*
5015	 * This CPU is finished loading/storing data:
5016	 * LMM(__printk_cpu_sync_put:A)
5017	 */
5018
5019	/*
5020	 * Guarantee loads and stores from this CPU when it was the
5021	 * lock owner are visible to the next lock owner. This pairs
5022	 * with __printk_cpu_sync_try_get:A.
5023	 *
5024	 * Memory barrier involvement:
5025	 *
5026	 * If __printk_cpu_sync_try_get:A reads from __printk_cpu_sync_put:B,
5027	 * then __printk_cpu_sync_try_get:B reads from __printk_cpu_sync_put:A.
5028	 *
5029	 * Relies on:
5030	 *
5031	 * RELEASE from __printk_cpu_sync_put:A to __printk_cpu_sync_put:B
5032	 * of this CPU
5033	 *    matching
5034	 * ACQUIRE from __printk_cpu_sync_try_get:A to
5035	 * __printk_cpu_sync_try_get:B of the next CPU
5036	 */
5037	atomic_set_release(&printk_cpu_sync_owner,
5038			   -1); /* LMM(__printk_cpu_sync_put:B) */
5039}
5040EXPORT_SYMBOL(__printk_cpu_sync_put);
5041#endif /* CONFIG_SMP */
v6.2
   1// SPDX-License-Identifier: GPL-2.0-only
   2/*
   3 *  linux/kernel/printk.c
   4 *
   5 *  Copyright (C) 1991, 1992  Linus Torvalds
   6 *
   7 * Modified to make sys_syslog() more flexible: added commands to
   8 * return the last 4k of kernel messages, regardless of whether
   9 * they've been read or not.  Added option to suppress kernel printk's
  10 * to the console.  Added hook for sending the console messages
  11 * elsewhere, in preparation for a serial line console (someday).
  12 * Ted Ts'o, 2/11/93.
  13 * Modified for sysctl support, 1/8/97, Chris Horn.
  14 * Fixed SMP synchronization, 08/08/99, Manfred Spraul
  15 *     manfred@colorfullife.com
  16 * Rewrote bits to get rid of console_lock
  17 *	01Mar01 Andrew Morton
  18 */
  19
  20#define pr_fmt(fmt) KBUILD_MODNAME ": " fmt
  21
  22#include <linux/kernel.h>
  23#include <linux/mm.h>
  24#include <linux/tty.h>
  25#include <linux/tty_driver.h>
  26#include <linux/console.h>
  27#include <linux/init.h>
  28#include <linux/jiffies.h>
  29#include <linux/nmi.h>
  30#include <linux/module.h>
  31#include <linux/moduleparam.h>
  32#include <linux/delay.h>
  33#include <linux/smp.h>
  34#include <linux/security.h>
  35#include <linux/memblock.h>
  36#include <linux/syscalls.h>
  37#include <linux/crash_core.h>
 
  38#include <linux/ratelimit.h>
  39#include <linux/kmsg_dump.h>
  40#include <linux/syslog.h>
  41#include <linux/cpu.h>
  42#include <linux/rculist.h>
  43#include <linux/poll.h>
  44#include <linux/irq_work.h>
  45#include <linux/ctype.h>
  46#include <linux/uio.h>
  47#include <linux/sched/clock.h>
  48#include <linux/sched/debug.h>
  49#include <linux/sched/task_stack.h>
  50
  51#include <linux/uaccess.h>
  52#include <asm/sections.h>
  53
  54#include <trace/events/initcall.h>
  55#define CREATE_TRACE_POINTS
  56#include <trace/events/printk.h>
  57
  58#include "printk_ringbuffer.h"
  59#include "console_cmdline.h"
  60#include "braille.h"
  61#include "internal.h"
  62
  63int console_printk[4] = {
  64	CONSOLE_LOGLEVEL_DEFAULT,	/* console_loglevel */
  65	MESSAGE_LOGLEVEL_DEFAULT,	/* default_message_loglevel */
  66	CONSOLE_LOGLEVEL_MIN,		/* minimum_console_loglevel */
  67	CONSOLE_LOGLEVEL_DEFAULT,	/* default_console_loglevel */
  68};
  69EXPORT_SYMBOL_GPL(console_printk);
  70
  71atomic_t ignore_console_lock_warning __read_mostly = ATOMIC_INIT(0);
  72EXPORT_SYMBOL(ignore_console_lock_warning);
  73
 
 
  74/*
  75 * Low level drivers may need that to know if they can schedule in
  76 * their unblank() callback or not. So let's export it.
  77 */
  78int oops_in_progress;
  79EXPORT_SYMBOL(oops_in_progress);
  80
  81/*
  82 * console_mutex protects console_list updates and console->flags updates.
  83 * The flags are synchronized only for consoles that are registered, i.e.
  84 * accessible via the console list.
  85 */
  86static DEFINE_MUTEX(console_mutex);
  87
  88/*
  89 * console_sem protects updates to console->seq and console_suspended,
  90 * and also provides serialization for console printing.
  91 */
  92static DEFINE_SEMAPHORE(console_sem);
  93HLIST_HEAD(console_list);
  94EXPORT_SYMBOL_GPL(console_list);
  95DEFINE_STATIC_SRCU(console_srcu);
  96
  97/*
  98 * System may need to suppress printk message under certain
  99 * circumstances, like after kernel panic happens.
 100 */
 101int __read_mostly suppress_printk;
 102
 103/*
 104 * During panic, heavy printk by other CPUs can delay the
 105 * panic and risk deadlock on console resources.
 106 */
 107static int __read_mostly suppress_panic_printk;
 108
 109#ifdef CONFIG_LOCKDEP
 110static struct lockdep_map console_lock_dep_map = {
 111	.name = "console_lock"
 112};
 113
 114void lockdep_assert_console_list_lock_held(void)
 115{
 116	lockdep_assert_held(&console_mutex);
 117}
 118EXPORT_SYMBOL(lockdep_assert_console_list_lock_held);
 119#endif
 120
 121#ifdef CONFIG_DEBUG_LOCK_ALLOC
 122bool console_srcu_read_lock_is_held(void)
 123{
 124	return srcu_read_lock_held(&console_srcu);
 125}
 126EXPORT_SYMBOL(console_srcu_read_lock_is_held);
 127#endif
 128
 129enum devkmsg_log_bits {
 130	__DEVKMSG_LOG_BIT_ON = 0,
 131	__DEVKMSG_LOG_BIT_OFF,
 132	__DEVKMSG_LOG_BIT_LOCK,
 133};
 134
 135enum devkmsg_log_masks {
 136	DEVKMSG_LOG_MASK_ON             = BIT(__DEVKMSG_LOG_BIT_ON),
 137	DEVKMSG_LOG_MASK_OFF            = BIT(__DEVKMSG_LOG_BIT_OFF),
 138	DEVKMSG_LOG_MASK_LOCK           = BIT(__DEVKMSG_LOG_BIT_LOCK),
 139};
 140
 141/* Keep both the 'on' and 'off' bits clear, i.e. ratelimit by default: */
 142#define DEVKMSG_LOG_MASK_DEFAULT	0
 143
 144static unsigned int __read_mostly devkmsg_log = DEVKMSG_LOG_MASK_DEFAULT;
 145
 146static int __control_devkmsg(char *str)
 147{
 148	size_t len;
 149
 150	if (!str)
 151		return -EINVAL;
 152
 153	len = str_has_prefix(str, "on");
 154	if (len) {
 155		devkmsg_log = DEVKMSG_LOG_MASK_ON;
 156		return len;
 157	}
 158
 159	len = str_has_prefix(str, "off");
 160	if (len) {
 161		devkmsg_log = DEVKMSG_LOG_MASK_OFF;
 162		return len;
 163	}
 164
 165	len = str_has_prefix(str, "ratelimit");
 166	if (len) {
 167		devkmsg_log = DEVKMSG_LOG_MASK_DEFAULT;
 168		return len;
 169	}
 170
 171	return -EINVAL;
 172}
 173
 174static int __init control_devkmsg(char *str)
 175{
 176	if (__control_devkmsg(str) < 0) {
 177		pr_warn("printk.devkmsg: bad option string '%s'\n", str);
 178		return 1;
 179	}
 180
 181	/*
 182	 * Set sysctl string accordingly:
 183	 */
 184	if (devkmsg_log == DEVKMSG_LOG_MASK_ON)
 185		strcpy(devkmsg_log_str, "on");
 186	else if (devkmsg_log == DEVKMSG_LOG_MASK_OFF)
 187		strcpy(devkmsg_log_str, "off");
 188	/* else "ratelimit" which is set by default. */
 189
 190	/*
 191	 * Sysctl cannot change it anymore. The kernel command line setting of
 192	 * this parameter is to force the setting to be permanent throughout the
 193	 * runtime of the system. This is a precation measure against userspace
 194	 * trying to be a smarta** and attempting to change it up on us.
 195	 */
 196	devkmsg_log |= DEVKMSG_LOG_MASK_LOCK;
 197
 198	return 1;
 199}
 200__setup("printk.devkmsg=", control_devkmsg);
 201
 202char devkmsg_log_str[DEVKMSG_STR_MAX_SIZE] = "ratelimit";
 203#if defined(CONFIG_PRINTK) && defined(CONFIG_SYSCTL)
 204int devkmsg_sysctl_set_loglvl(struct ctl_table *table, int write,
 205			      void *buffer, size_t *lenp, loff_t *ppos)
 206{
 207	char old_str[DEVKMSG_STR_MAX_SIZE];
 208	unsigned int old;
 209	int err;
 210
 211	if (write) {
 212		if (devkmsg_log & DEVKMSG_LOG_MASK_LOCK)
 213			return -EINVAL;
 214
 215		old = devkmsg_log;
 216		strncpy(old_str, devkmsg_log_str, DEVKMSG_STR_MAX_SIZE);
 217	}
 218
 219	err = proc_dostring(table, write, buffer, lenp, ppos);
 220	if (err)
 221		return err;
 222
 223	if (write) {
 224		err = __control_devkmsg(devkmsg_log_str);
 225
 226		/*
 227		 * Do not accept an unknown string OR a known string with
 228		 * trailing crap...
 229		 */
 230		if (err < 0 || (err + 1 != *lenp)) {
 231
 232			/* ... and restore old setting. */
 233			devkmsg_log = old;
 234			strncpy(devkmsg_log_str, old_str, DEVKMSG_STR_MAX_SIZE);
 235
 236			return -EINVAL;
 237		}
 238	}
 239
 240	return 0;
 241}
 242#endif /* CONFIG_PRINTK && CONFIG_SYSCTL */
 243
 244/**
 245 * console_list_lock - Lock the console list
 246 *
 247 * For console list or console->flags updates
 248 */
 249void console_list_lock(void)
 250{
 251	/*
 252	 * In unregister_console() and console_force_preferred_locked(),
 253	 * synchronize_srcu() is called with the console_list_lock held.
 254	 * Therefore it is not allowed that the console_list_lock is taken
 255	 * with the srcu_lock held.
 256	 *
 257	 * Detecting if this context is really in the read-side critical
 258	 * section is only possible if the appropriate debug options are
 259	 * enabled.
 260	 */
 261	WARN_ON_ONCE(debug_lockdep_rcu_enabled() &&
 262		     srcu_read_lock_held(&console_srcu));
 263
 264	mutex_lock(&console_mutex);
 265}
 266EXPORT_SYMBOL(console_list_lock);
 267
 268/**
 269 * console_list_unlock - Unlock the console list
 270 *
 271 * Counterpart to console_list_lock()
 272 */
 273void console_list_unlock(void)
 274{
 275	mutex_unlock(&console_mutex);
 276}
 277EXPORT_SYMBOL(console_list_unlock);
 278
 279/**
 280 * console_srcu_read_lock - Register a new reader for the
 281 *	SRCU-protected console list
 282 *
 283 * Use for_each_console_srcu() to iterate the console list
 284 *
 285 * Context: Any context.
 286 * Return: A cookie to pass to console_srcu_read_unlock().
 287 */
 288int console_srcu_read_lock(void)
 
 289{
 290	return srcu_read_lock_nmisafe(&console_srcu);
 291}
 292EXPORT_SYMBOL(console_srcu_read_lock);
 293
 294/**
 295 * console_srcu_read_unlock - Unregister an old reader from
 296 *	the SRCU-protected console list
 297 * @cookie: cookie returned from console_srcu_read_lock()
 298 *
 299 * Counterpart to console_srcu_read_lock()
 300 */
 301void console_srcu_read_unlock(int cookie)
 
 302{
 303	srcu_read_unlock_nmisafe(&console_srcu, cookie);
 304}
 305EXPORT_SYMBOL(console_srcu_read_unlock);
 306
 307/*
 308 * Helper macros to handle lockdep when locking/unlocking console_sem. We use
 309 * macros instead of functions so that _RET_IP_ contains useful information.
 310 */
 311#define down_console_sem() do { \
 312	down(&console_sem);\
 313	mutex_acquire(&console_lock_dep_map, 0, 0, _RET_IP_);\
 314} while (0)
 315
 316static int __down_trylock_console_sem(unsigned long ip)
 317{
 318	int lock_failed;
 319	unsigned long flags;
 320
 321	/*
 322	 * Here and in __up_console_sem() we need to be in safe mode,
 323	 * because spindump/WARN/etc from under console ->lock will
 324	 * deadlock in printk()->down_trylock_console_sem() otherwise.
 325	 */
 326	printk_safe_enter_irqsave(flags);
 327	lock_failed = down_trylock(&console_sem);
 328	printk_safe_exit_irqrestore(flags);
 329
 330	if (lock_failed)
 331		return 1;
 332	mutex_acquire(&console_lock_dep_map, 0, 1, ip);
 333	return 0;
 334}
 335#define down_trylock_console_sem() __down_trylock_console_sem(_RET_IP_)
 336
 337static void __up_console_sem(unsigned long ip)
 338{
 339	unsigned long flags;
 340
 341	mutex_release(&console_lock_dep_map, ip);
 342
 343	printk_safe_enter_irqsave(flags);
 344	up(&console_sem);
 345	printk_safe_exit_irqrestore(flags);
 346}
 347#define up_console_sem() __up_console_sem(_RET_IP_)
 348
 349static bool panic_in_progress(void)
 350{
 351	return unlikely(atomic_read(&panic_cpu) != PANIC_CPU_INVALID);
 352}
 353
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 354/*
 355 * This is used for debugging the mess that is the VT code by
 356 * keeping track if we have the console semaphore held. It's
 357 * definitely not the perfect debug tool (we don't know if _WE_
 358 * hold it and are racing, but it helps tracking those weird code
 359 * paths in the console code where we end up in places I want
 360 * locked without the console semaphore held).
 361 */
 362static int console_locked, console_suspended;
 363
 364/*
 365 *	Array of consoles built from command line options (console=)
 366 */
 367
 368#define MAX_CMDLINECONSOLES 8
 369
 370static struct console_cmdline console_cmdline[MAX_CMDLINECONSOLES];
 371
 372static int preferred_console = -1;
 373int console_set_on_cmdline;
 374EXPORT_SYMBOL(console_set_on_cmdline);
 375
 376/* Flag: console code may call schedule() */
 377static int console_may_schedule;
 378
 379enum con_msg_format_flags {
 380	MSG_FORMAT_DEFAULT	= 0,
 381	MSG_FORMAT_SYSLOG	= (1 << 0),
 382};
 383
 384static int console_msg_format = MSG_FORMAT_DEFAULT;
 385
 386/*
 387 * The printk log buffer consists of a sequenced collection of records, each
 388 * containing variable length message text. Every record also contains its
 389 * own meta-data (@info).
 390 *
 391 * Every record meta-data carries the timestamp in microseconds, as well as
 392 * the standard userspace syslog level and syslog facility. The usual kernel
 393 * messages use LOG_KERN; userspace-injected messages always carry a matching
 394 * syslog facility, by default LOG_USER. The origin of every message can be
 395 * reliably determined that way.
 396 *
 397 * The human readable log message of a record is available in @text, the
 398 * length of the message text in @text_len. The stored message is not
 399 * terminated.
 400 *
 401 * Optionally, a record can carry a dictionary of properties (key/value
 402 * pairs), to provide userspace with a machine-readable message context.
 403 *
 404 * Examples for well-defined, commonly used property names are:
 405 *   DEVICE=b12:8               device identifier
 406 *                                b12:8         block dev_t
 407 *                                c127:3        char dev_t
 408 *                                n8            netdev ifindex
 409 *                                +sound:card0  subsystem:devname
 410 *   SUBSYSTEM=pci              driver-core subsystem name
 411 *
 412 * Valid characters in property names are [a-zA-Z0-9.-_]. Property names
 413 * and values are terminated by a '\0' character.
 414 *
 415 * Example of record values:
 416 *   record.text_buf                = "it's a line" (unterminated)
 417 *   record.info.seq                = 56
 418 *   record.info.ts_nsec            = 36863
 419 *   record.info.text_len           = 11
 420 *   record.info.facility           = 0 (LOG_KERN)
 421 *   record.info.flags              = 0
 422 *   record.info.level              = 3 (LOG_ERR)
 423 *   record.info.caller_id          = 299 (task 299)
 424 *   record.info.dev_info.subsystem = "pci" (terminated)
 425 *   record.info.dev_info.device    = "+pci:0000:00:01.0" (terminated)
 426 *
 427 * The 'struct printk_info' buffer must never be directly exported to
 428 * userspace, it is a kernel-private implementation detail that might
 429 * need to be changed in the future, when the requirements change.
 430 *
 431 * /dev/kmsg exports the structured data in the following line format:
 432 *   "<level>,<sequnum>,<timestamp>,<contflag>[,additional_values, ... ];<message text>\n"
 433 *
 434 * Users of the export format should ignore possible additional values
 435 * separated by ',', and find the message after the ';' character.
 436 *
 437 * The optional key/value pairs are attached as continuation lines starting
 438 * with a space character and terminated by a newline. All possible
 439 * non-prinatable characters are escaped in the "\xff" notation.
 440 */
 441
 442/* syslog_lock protects syslog_* variables and write access to clear_seq. */
 443static DEFINE_MUTEX(syslog_lock);
 444
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 445#ifdef CONFIG_PRINTK
 446DECLARE_WAIT_QUEUE_HEAD(log_wait);
 
 447/* All 3 protected by @syslog_lock. */
 448/* the next printk record to read by syslog(READ) or /proc/kmsg */
 449static u64 syslog_seq;
 450static size_t syslog_partial;
 451static bool syslog_time;
 452
 
 
 
 453struct latched_seq {
 454	seqcount_latch_t	latch;
 455	u64			val[2];
 456};
 457
 458/*
 459 * The next printk record to read after the last 'clear' command. There are
 460 * two copies (updated with seqcount_latch) so that reads can locklessly
 461 * access a valid value. Writers are synchronized by @syslog_lock.
 462 */
 463static struct latched_seq clear_seq = {
 464	.latch		= SEQCNT_LATCH_ZERO(clear_seq.latch),
 465	.val[0]		= 0,
 466	.val[1]		= 0,
 467};
 468
 469#ifdef CONFIG_PRINTK_CALLER
 470#define PREFIX_MAX		48
 471#else
 472#define PREFIX_MAX		32
 473#endif
 474
 475/* the maximum size of a formatted record (i.e. with prefix added per line) */
 476#define CONSOLE_LOG_MAX		1024
 477
 478/* the maximum size for a dropped text message */
 479#define DROPPED_TEXT_MAX	64
 480
 481/* the maximum size allowed to be reserved for a record */
 482#define LOG_LINE_MAX		(CONSOLE_LOG_MAX - PREFIX_MAX)
 483
 484#define LOG_LEVEL(v)		((v) & 0x07)
 485#define LOG_FACILITY(v)		((v) >> 3 & 0xff)
 486
 487/* record buffer */
 488#define LOG_ALIGN __alignof__(unsigned long)
 489#define __LOG_BUF_LEN (1 << CONFIG_LOG_BUF_SHIFT)
 490#define LOG_BUF_LEN_MAX (u32)(1 << 31)
 491static char __log_buf[__LOG_BUF_LEN] __aligned(LOG_ALIGN);
 492static char *log_buf = __log_buf;
 493static u32 log_buf_len = __LOG_BUF_LEN;
 494
 495/*
 496 * Define the average message size. This only affects the number of
 497 * descriptors that will be available. Underestimating is better than
 498 * overestimating (too many available descriptors is better than not enough).
 499 */
 500#define PRB_AVGBITS 5	/* 32 character average length */
 501
 502#if CONFIG_LOG_BUF_SHIFT <= PRB_AVGBITS
 503#error CONFIG_LOG_BUF_SHIFT value too small.
 504#endif
 505_DEFINE_PRINTKRB(printk_rb_static, CONFIG_LOG_BUF_SHIFT - PRB_AVGBITS,
 506		 PRB_AVGBITS, &__log_buf[0]);
 507
 508static struct printk_ringbuffer printk_rb_dynamic;
 509
 510static struct printk_ringbuffer *prb = &printk_rb_static;
 511
 512/*
 513 * We cannot access per-CPU data (e.g. per-CPU flush irq_work) before
 514 * per_cpu_areas are initialised. This variable is set to true when
 515 * it's safe to access per-CPU data.
 516 */
 517static bool __printk_percpu_data_ready __ro_after_init;
 518
 519bool printk_percpu_data_ready(void)
 520{
 521	return __printk_percpu_data_ready;
 522}
 523
 524/* Must be called under syslog_lock. */
 525static void latched_seq_write(struct latched_seq *ls, u64 val)
 526{
 527	raw_write_seqcount_latch(&ls->latch);
 528	ls->val[0] = val;
 529	raw_write_seqcount_latch(&ls->latch);
 530	ls->val[1] = val;
 
 531}
 532
 533/* Can be called from any context. */
 534static u64 latched_seq_read_nolock(struct latched_seq *ls)
 535{
 536	unsigned int seq;
 537	unsigned int idx;
 538	u64 val;
 539
 540	do {
 541		seq = raw_read_seqcount_latch(&ls->latch);
 542		idx = seq & 0x1;
 543		val = ls->val[idx];
 544	} while (read_seqcount_latch_retry(&ls->latch, seq));
 545
 546	return val;
 547}
 548
 549/* Return log buffer address */
 550char *log_buf_addr_get(void)
 551{
 552	return log_buf;
 553}
 554
 555/* Return log buffer size */
 556u32 log_buf_len_get(void)
 557{
 558	return log_buf_len;
 559}
 560
 561/*
 562 * Define how much of the log buffer we could take at maximum. The value
 563 * must be greater than two. Note that only half of the buffer is available
 564 * when the index points to the middle.
 565 */
 566#define MAX_LOG_TAKE_PART 4
 567static const char trunc_msg[] = "<truncated>";
 568
 569static void truncate_msg(u16 *text_len, u16 *trunc_msg_len)
 570{
 571	/*
 572	 * The message should not take the whole buffer. Otherwise, it might
 573	 * get removed too soon.
 574	 */
 575	u32 max_text_len = log_buf_len / MAX_LOG_TAKE_PART;
 576
 577	if (*text_len > max_text_len)
 578		*text_len = max_text_len;
 579
 580	/* enable the warning message (if there is room) */
 581	*trunc_msg_len = strlen(trunc_msg);
 582	if (*text_len >= *trunc_msg_len)
 583		*text_len -= *trunc_msg_len;
 584	else
 585		*trunc_msg_len = 0;
 586}
 587
 588int dmesg_restrict = IS_ENABLED(CONFIG_SECURITY_DMESG_RESTRICT);
 589
 590static int syslog_action_restricted(int type)
 591{
 592	if (dmesg_restrict)
 593		return 1;
 594	/*
 595	 * Unless restricted, we allow "read all" and "get buffer size"
 596	 * for everybody.
 597	 */
 598	return type != SYSLOG_ACTION_READ_ALL &&
 599	       type != SYSLOG_ACTION_SIZE_BUFFER;
 600}
 601
 602static int check_syslog_permissions(int type, int source)
 603{
 604	/*
 605	 * If this is from /proc/kmsg and we've already opened it, then we've
 606	 * already done the capabilities checks at open time.
 607	 */
 608	if (source == SYSLOG_FROM_PROC && type != SYSLOG_ACTION_OPEN)
 609		goto ok;
 610
 611	if (syslog_action_restricted(type)) {
 612		if (capable(CAP_SYSLOG))
 613			goto ok;
 614		/*
 615		 * For historical reasons, accept CAP_SYS_ADMIN too, with
 616		 * a warning.
 617		 */
 618		if (capable(CAP_SYS_ADMIN)) {
 619			pr_warn_once("%s (%d): Attempt to access syslog with "
 620				     "CAP_SYS_ADMIN but no CAP_SYSLOG "
 621				     "(deprecated).\n",
 622				 current->comm, task_pid_nr(current));
 623			goto ok;
 624		}
 625		return -EPERM;
 626	}
 627ok:
 628	return security_syslog(type);
 629}
 630
 631static void append_char(char **pp, char *e, char c)
 632{
 633	if (*pp < e)
 634		*(*pp)++ = c;
 635}
 636
 637static ssize_t info_print_ext_header(char *buf, size_t size,
 638				     struct printk_info *info)
 639{
 640	u64 ts_usec = info->ts_nsec;
 641	char caller[20];
 642#ifdef CONFIG_PRINTK_CALLER
 643	u32 id = info->caller_id;
 644
 645	snprintf(caller, sizeof(caller), ",caller=%c%u",
 646		 id & 0x80000000 ? 'C' : 'T', id & ~0x80000000);
 647#else
 648	caller[0] = '\0';
 649#endif
 650
 651	do_div(ts_usec, 1000);
 652
 653	return scnprintf(buf, size, "%u,%llu,%llu,%c%s;",
 654			 (info->facility << 3) | info->level, info->seq,
 655			 ts_usec, info->flags & LOG_CONT ? 'c' : '-', caller);
 656}
 657
 658static ssize_t msg_add_ext_text(char *buf, size_t size,
 659				const char *text, size_t text_len,
 660				unsigned char endc)
 661{
 662	char *p = buf, *e = buf + size;
 663	size_t i;
 664
 665	/* escape non-printable characters */
 666	for (i = 0; i < text_len; i++) {
 667		unsigned char c = text[i];
 668
 669		if (c < ' ' || c >= 127 || c == '\\')
 670			p += scnprintf(p, e - p, "\\x%02x", c);
 671		else
 672			append_char(&p, e, c);
 673	}
 674	append_char(&p, e, endc);
 675
 676	return p - buf;
 677}
 678
 679static ssize_t msg_add_dict_text(char *buf, size_t size,
 680				 const char *key, const char *val)
 681{
 682	size_t val_len = strlen(val);
 683	ssize_t len;
 684
 685	if (!val_len)
 686		return 0;
 687
 688	len = msg_add_ext_text(buf, size, "", 0, ' ');	/* dict prefix */
 689	len += msg_add_ext_text(buf + len, size - len, key, strlen(key), '=');
 690	len += msg_add_ext_text(buf + len, size - len, val, val_len, '\n');
 691
 692	return len;
 693}
 694
 695static ssize_t msg_print_ext_body(char *buf, size_t size,
 696				  char *text, size_t text_len,
 697				  struct dev_printk_info *dev_info)
 698{
 699	ssize_t len;
 700
 701	len = msg_add_ext_text(buf, size, text, text_len, '\n');
 702
 703	if (!dev_info)
 704		goto out;
 705
 706	len += msg_add_dict_text(buf + len, size - len, "SUBSYSTEM",
 707				 dev_info->subsystem);
 708	len += msg_add_dict_text(buf + len, size - len, "DEVICE",
 709				 dev_info->device);
 710out:
 711	return len;
 712}
 713
 714/* /dev/kmsg - userspace message inject/listen interface */
 715struct devkmsg_user {
 716	atomic64_t seq;
 717	struct ratelimit_state rs;
 718	struct mutex lock;
 719	char buf[CONSOLE_EXT_LOG_MAX];
 720
 721	struct printk_info info;
 722	char text_buf[CONSOLE_EXT_LOG_MAX];
 723	struct printk_record record;
 724};
 725
 726static __printf(3, 4) __cold
 727int devkmsg_emit(int facility, int level, const char *fmt, ...)
 728{
 729	va_list args;
 730	int r;
 731
 732	va_start(args, fmt);
 733	r = vprintk_emit(facility, level, NULL, fmt, args);
 734	va_end(args);
 735
 736	return r;
 737}
 738
 739static ssize_t devkmsg_write(struct kiocb *iocb, struct iov_iter *from)
 740{
 741	char *buf, *line;
 742	int level = default_message_loglevel;
 743	int facility = 1;	/* LOG_USER */
 744	struct file *file = iocb->ki_filp;
 745	struct devkmsg_user *user = file->private_data;
 746	size_t len = iov_iter_count(from);
 747	ssize_t ret = len;
 748
 749	if (!user || len > LOG_LINE_MAX)
 750		return -EINVAL;
 751
 752	/* Ignore when user logging is disabled. */
 753	if (devkmsg_log & DEVKMSG_LOG_MASK_OFF)
 754		return len;
 755
 756	/* Ratelimit when not explicitly enabled. */
 757	if (!(devkmsg_log & DEVKMSG_LOG_MASK_ON)) {
 758		if (!___ratelimit(&user->rs, current->comm))
 759			return ret;
 760	}
 761
 762	buf = kmalloc(len+1, GFP_KERNEL);
 763	if (buf == NULL)
 764		return -ENOMEM;
 765
 766	buf[len] = '\0';
 767	if (!copy_from_iter_full(buf, len, from)) {
 768		kfree(buf);
 769		return -EFAULT;
 770	}
 771
 772	/*
 773	 * Extract and skip the syslog prefix <[0-9]*>. Coming from userspace
 774	 * the decimal value represents 32bit, the lower 3 bit are the log
 775	 * level, the rest are the log facility.
 776	 *
 777	 * If no prefix or no userspace facility is specified, we
 778	 * enforce LOG_USER, to be able to reliably distinguish
 779	 * kernel-generated messages from userspace-injected ones.
 780	 */
 781	line = buf;
 782	if (line[0] == '<') {
 783		char *endp = NULL;
 784		unsigned int u;
 785
 786		u = simple_strtoul(line + 1, &endp, 10);
 787		if (endp && endp[0] == '>') {
 788			level = LOG_LEVEL(u);
 789			if (LOG_FACILITY(u) != 0)
 790				facility = LOG_FACILITY(u);
 791			endp++;
 792			line = endp;
 793		}
 794	}
 795
 796	devkmsg_emit(facility, level, "%s", line);
 797	kfree(buf);
 798	return ret;
 799}
 800
 801static ssize_t devkmsg_read(struct file *file, char __user *buf,
 802			    size_t count, loff_t *ppos)
 803{
 804	struct devkmsg_user *user = file->private_data;
 805	struct printk_record *r = &user->record;
 806	size_t len;
 
 
 807	ssize_t ret;
 808
 809	if (!user)
 810		return -EBADF;
 811
 812	ret = mutex_lock_interruptible(&user->lock);
 813	if (ret)
 814		return ret;
 815
 816	if (!prb_read_valid(prb, atomic64_read(&user->seq), r)) {
 817		if (file->f_flags & O_NONBLOCK) {
 818			ret = -EAGAIN;
 819			goto out;
 820		}
 821
 822		/*
 823		 * Guarantee this task is visible on the waitqueue before
 824		 * checking the wake condition.
 825		 *
 826		 * The full memory barrier within set_current_state() of
 827		 * prepare_to_wait_event() pairs with the full memory barrier
 828		 * within wq_has_sleeper().
 829		 *
 830		 * This pairs with __wake_up_klogd:A.
 831		 */
 832		ret = wait_event_interruptible(log_wait,
 833				prb_read_valid(prb,
 834					atomic64_read(&user->seq), r)); /* LMM(devkmsg_read:A) */
 835		if (ret)
 836			goto out;
 837	}
 838
 839	if (r->info->seq != atomic64_read(&user->seq)) {
 840		/* our last seen message is gone, return error and reset */
 841		atomic64_set(&user->seq, r->info->seq);
 842		ret = -EPIPE;
 843		goto out;
 844	}
 845
 846	len = info_print_ext_header(user->buf, sizeof(user->buf), r->info);
 847	len += msg_print_ext_body(user->buf + len, sizeof(user->buf) - len,
 848				  &r->text_buf[0], r->info->text_len,
 849				  &r->info->dev_info);
 850
 851	atomic64_set(&user->seq, r->info->seq + 1);
 852
 853	if (len > count) {
 854		ret = -EINVAL;
 855		goto out;
 856	}
 857
 858	if (copy_to_user(buf, user->buf, len)) {
 859		ret = -EFAULT;
 860		goto out;
 861	}
 862	ret = len;
 863out:
 864	mutex_unlock(&user->lock);
 865	return ret;
 866}
 867
 868/*
 869 * Be careful when modifying this function!!!
 870 *
 871 * Only few operations are supported because the device works only with the
 872 * entire variable length messages (records). Non-standard values are
 873 * returned in the other cases and has been this way for quite some time.
 874 * User space applications might depend on this behavior.
 875 */
 876static loff_t devkmsg_llseek(struct file *file, loff_t offset, int whence)
 877{
 878	struct devkmsg_user *user = file->private_data;
 879	loff_t ret = 0;
 880
 881	if (!user)
 882		return -EBADF;
 883	if (offset)
 884		return -ESPIPE;
 885
 886	switch (whence) {
 887	case SEEK_SET:
 888		/* the first record */
 889		atomic64_set(&user->seq, prb_first_valid_seq(prb));
 890		break;
 891	case SEEK_DATA:
 892		/*
 893		 * The first record after the last SYSLOG_ACTION_CLEAR,
 894		 * like issued by 'dmesg -c'. Reading /dev/kmsg itself
 895		 * changes no global state, and does not clear anything.
 896		 */
 897		atomic64_set(&user->seq, latched_seq_read_nolock(&clear_seq));
 898		break;
 899	case SEEK_END:
 900		/* after the last record */
 901		atomic64_set(&user->seq, prb_next_seq(prb));
 902		break;
 903	default:
 904		ret = -EINVAL;
 905	}
 906	return ret;
 907}
 908
 909static __poll_t devkmsg_poll(struct file *file, poll_table *wait)
 910{
 911	struct devkmsg_user *user = file->private_data;
 912	struct printk_info info;
 913	__poll_t ret = 0;
 914
 915	if (!user)
 916		return EPOLLERR|EPOLLNVAL;
 917
 918	poll_wait(file, &log_wait, wait);
 919
 920	if (prb_read_valid_info(prb, atomic64_read(&user->seq), &info, NULL)) {
 921		/* return error when data has vanished underneath us */
 922		if (info.seq != atomic64_read(&user->seq))
 923			ret = EPOLLIN|EPOLLRDNORM|EPOLLERR|EPOLLPRI;
 924		else
 925			ret = EPOLLIN|EPOLLRDNORM;
 926	}
 927
 928	return ret;
 929}
 930
 931static int devkmsg_open(struct inode *inode, struct file *file)
 932{
 933	struct devkmsg_user *user;
 934	int err;
 935
 936	if (devkmsg_log & DEVKMSG_LOG_MASK_OFF)
 937		return -EPERM;
 938
 939	/* write-only does not need any file context */
 940	if ((file->f_flags & O_ACCMODE) != O_WRONLY) {
 941		err = check_syslog_permissions(SYSLOG_ACTION_READ_ALL,
 942					       SYSLOG_FROM_READER);
 943		if (err)
 944			return err;
 945	}
 946
 947	user = kvmalloc(sizeof(struct devkmsg_user), GFP_KERNEL);
 948	if (!user)
 949		return -ENOMEM;
 950
 951	ratelimit_default_init(&user->rs);
 952	ratelimit_set_flags(&user->rs, RATELIMIT_MSG_ON_RELEASE);
 953
 954	mutex_init(&user->lock);
 955
 956	prb_rec_init_rd(&user->record, &user->info,
 957			&user->text_buf[0], sizeof(user->text_buf));
 958
 959	atomic64_set(&user->seq, prb_first_valid_seq(prb));
 960
 961	file->private_data = user;
 962	return 0;
 963}
 964
 965static int devkmsg_release(struct inode *inode, struct file *file)
 966{
 967	struct devkmsg_user *user = file->private_data;
 968
 969	if (!user)
 970		return 0;
 971
 972	ratelimit_state_exit(&user->rs);
 973
 974	mutex_destroy(&user->lock);
 975	kvfree(user);
 976	return 0;
 977}
 978
 979const struct file_operations kmsg_fops = {
 980	.open = devkmsg_open,
 981	.read = devkmsg_read,
 982	.write_iter = devkmsg_write,
 983	.llseek = devkmsg_llseek,
 984	.poll = devkmsg_poll,
 985	.release = devkmsg_release,
 986};
 987
 988#ifdef CONFIG_CRASH_CORE
 989/*
 990 * This appends the listed symbols to /proc/vmcore
 991 *
 992 * /proc/vmcore is used by various utilities, like crash and makedumpfile to
 993 * obtain access to symbols that are otherwise very difficult to locate.  These
 994 * symbols are specifically used so that utilities can access and extract the
 995 * dmesg log from a vmcore file after a crash.
 996 */
 997void log_buf_vmcoreinfo_setup(void)
 998{
 999	struct dev_printk_info *dev_info = NULL;
1000
1001	VMCOREINFO_SYMBOL(prb);
1002	VMCOREINFO_SYMBOL(printk_rb_static);
1003	VMCOREINFO_SYMBOL(clear_seq);
1004
1005	/*
1006	 * Export struct size and field offsets. User space tools can
1007	 * parse it and detect any changes to structure down the line.
1008	 */
1009
1010	VMCOREINFO_STRUCT_SIZE(printk_ringbuffer);
1011	VMCOREINFO_OFFSET(printk_ringbuffer, desc_ring);
1012	VMCOREINFO_OFFSET(printk_ringbuffer, text_data_ring);
1013	VMCOREINFO_OFFSET(printk_ringbuffer, fail);
1014
1015	VMCOREINFO_STRUCT_SIZE(prb_desc_ring);
1016	VMCOREINFO_OFFSET(prb_desc_ring, count_bits);
1017	VMCOREINFO_OFFSET(prb_desc_ring, descs);
1018	VMCOREINFO_OFFSET(prb_desc_ring, infos);
1019	VMCOREINFO_OFFSET(prb_desc_ring, head_id);
1020	VMCOREINFO_OFFSET(prb_desc_ring, tail_id);
1021
1022	VMCOREINFO_STRUCT_SIZE(prb_desc);
1023	VMCOREINFO_OFFSET(prb_desc, state_var);
1024	VMCOREINFO_OFFSET(prb_desc, text_blk_lpos);
1025
1026	VMCOREINFO_STRUCT_SIZE(prb_data_blk_lpos);
1027	VMCOREINFO_OFFSET(prb_data_blk_lpos, begin);
1028	VMCOREINFO_OFFSET(prb_data_blk_lpos, next);
1029
1030	VMCOREINFO_STRUCT_SIZE(printk_info);
1031	VMCOREINFO_OFFSET(printk_info, seq);
1032	VMCOREINFO_OFFSET(printk_info, ts_nsec);
1033	VMCOREINFO_OFFSET(printk_info, text_len);
1034	VMCOREINFO_OFFSET(printk_info, caller_id);
1035	VMCOREINFO_OFFSET(printk_info, dev_info);
1036
1037	VMCOREINFO_STRUCT_SIZE(dev_printk_info);
1038	VMCOREINFO_OFFSET(dev_printk_info, subsystem);
1039	VMCOREINFO_LENGTH(printk_info_subsystem, sizeof(dev_info->subsystem));
1040	VMCOREINFO_OFFSET(dev_printk_info, device);
1041	VMCOREINFO_LENGTH(printk_info_device, sizeof(dev_info->device));
1042
1043	VMCOREINFO_STRUCT_SIZE(prb_data_ring);
1044	VMCOREINFO_OFFSET(prb_data_ring, size_bits);
1045	VMCOREINFO_OFFSET(prb_data_ring, data);
1046	VMCOREINFO_OFFSET(prb_data_ring, head_lpos);
1047	VMCOREINFO_OFFSET(prb_data_ring, tail_lpos);
1048
1049	VMCOREINFO_SIZE(atomic_long_t);
1050	VMCOREINFO_TYPE_OFFSET(atomic_long_t, counter);
1051
1052	VMCOREINFO_STRUCT_SIZE(latched_seq);
1053	VMCOREINFO_OFFSET(latched_seq, val);
1054}
1055#endif
1056
1057/* requested log_buf_len from kernel cmdline */
1058static unsigned long __initdata new_log_buf_len;
1059
1060/* we practice scaling the ring buffer by powers of 2 */
1061static void __init log_buf_len_update(u64 size)
1062{
1063	if (size > (u64)LOG_BUF_LEN_MAX) {
1064		size = (u64)LOG_BUF_LEN_MAX;
1065		pr_err("log_buf over 2G is not supported.\n");
1066	}
1067
1068	if (size)
1069		size = roundup_pow_of_two(size);
1070	if (size > log_buf_len)
1071		new_log_buf_len = (unsigned long)size;
1072}
1073
1074/* save requested log_buf_len since it's too early to process it */
1075static int __init log_buf_len_setup(char *str)
1076{
1077	u64 size;
1078
1079	if (!str)
1080		return -EINVAL;
1081
1082	size = memparse(str, &str);
1083
1084	log_buf_len_update(size);
1085
1086	return 0;
1087}
1088early_param("log_buf_len", log_buf_len_setup);
1089
1090#ifdef CONFIG_SMP
1091#define __LOG_CPU_MAX_BUF_LEN (1 << CONFIG_LOG_CPU_MAX_BUF_SHIFT)
1092
1093static void __init log_buf_add_cpu(void)
1094{
1095	unsigned int cpu_extra;
1096
1097	/*
1098	 * archs should set up cpu_possible_bits properly with
1099	 * set_cpu_possible() after setup_arch() but just in
1100	 * case lets ensure this is valid.
1101	 */
1102	if (num_possible_cpus() == 1)
1103		return;
1104
1105	cpu_extra = (num_possible_cpus() - 1) * __LOG_CPU_MAX_BUF_LEN;
1106
1107	/* by default this will only continue through for large > 64 CPUs */
1108	if (cpu_extra <= __LOG_BUF_LEN / 2)
1109		return;
1110
1111	pr_info("log_buf_len individual max cpu contribution: %d bytes\n",
1112		__LOG_CPU_MAX_BUF_LEN);
1113	pr_info("log_buf_len total cpu_extra contributions: %d bytes\n",
1114		cpu_extra);
1115	pr_info("log_buf_len min size: %d bytes\n", __LOG_BUF_LEN);
1116
1117	log_buf_len_update(cpu_extra + __LOG_BUF_LEN);
1118}
1119#else /* !CONFIG_SMP */
1120static inline void log_buf_add_cpu(void) {}
1121#endif /* CONFIG_SMP */
1122
1123static void __init set_percpu_data_ready(void)
1124{
1125	__printk_percpu_data_ready = true;
1126}
1127
1128static unsigned int __init add_to_rb(struct printk_ringbuffer *rb,
1129				     struct printk_record *r)
1130{
1131	struct prb_reserved_entry e;
1132	struct printk_record dest_r;
1133
1134	prb_rec_init_wr(&dest_r, r->info->text_len);
1135
1136	if (!prb_reserve(&e, rb, &dest_r))
1137		return 0;
1138
1139	memcpy(&dest_r.text_buf[0], &r->text_buf[0], r->info->text_len);
1140	dest_r.info->text_len = r->info->text_len;
1141	dest_r.info->facility = r->info->facility;
1142	dest_r.info->level = r->info->level;
1143	dest_r.info->flags = r->info->flags;
1144	dest_r.info->ts_nsec = r->info->ts_nsec;
1145	dest_r.info->caller_id = r->info->caller_id;
1146	memcpy(&dest_r.info->dev_info, &r->info->dev_info, sizeof(dest_r.info->dev_info));
1147
1148	prb_final_commit(&e);
1149
1150	return prb_record_text_space(&e);
1151}
1152
1153static char setup_text_buf[LOG_LINE_MAX] __initdata;
 
 
 
 
 
 
 
 
 
 
 
1154
1155void __init setup_log_buf(int early)
1156{
1157	struct printk_info *new_infos;
1158	unsigned int new_descs_count;
1159	struct prb_desc *new_descs;
1160	struct printk_info info;
1161	struct printk_record r;
1162	unsigned int text_size;
1163	size_t new_descs_size;
1164	size_t new_infos_size;
1165	unsigned long flags;
1166	char *new_log_buf;
1167	unsigned int free;
1168	u64 seq;
1169
1170	/*
1171	 * Some archs call setup_log_buf() multiple times - first is very
1172	 * early, e.g. from setup_arch(), and second - when percpu_areas
1173	 * are initialised.
1174	 */
1175	if (!early)
1176		set_percpu_data_ready();
1177
1178	if (log_buf != __log_buf)
1179		return;
1180
1181	if (!early && !new_log_buf_len)
1182		log_buf_add_cpu();
1183
1184	if (!new_log_buf_len)
 
 
 
 
1185		return;
 
1186
1187	new_descs_count = new_log_buf_len >> PRB_AVGBITS;
1188	if (new_descs_count == 0) {
1189		pr_err("new_log_buf_len: %lu too small\n", new_log_buf_len);
1190		return;
1191	}
1192
1193	new_log_buf = memblock_alloc(new_log_buf_len, LOG_ALIGN);
1194	if (unlikely(!new_log_buf)) {
1195		pr_err("log_buf_len: %lu text bytes not available\n",
1196		       new_log_buf_len);
1197		return;
1198	}
1199
1200	new_descs_size = new_descs_count * sizeof(struct prb_desc);
1201	new_descs = memblock_alloc(new_descs_size, LOG_ALIGN);
1202	if (unlikely(!new_descs)) {
1203		pr_err("log_buf_len: %zu desc bytes not available\n",
1204		       new_descs_size);
1205		goto err_free_log_buf;
1206	}
1207
1208	new_infos_size = new_descs_count * sizeof(struct printk_info);
1209	new_infos = memblock_alloc(new_infos_size, LOG_ALIGN);
1210	if (unlikely(!new_infos)) {
1211		pr_err("log_buf_len: %zu info bytes not available\n",
1212		       new_infos_size);
1213		goto err_free_descs;
1214	}
1215
1216	prb_rec_init_rd(&r, &info, &setup_text_buf[0], sizeof(setup_text_buf));
1217
1218	prb_init(&printk_rb_dynamic,
1219		 new_log_buf, ilog2(new_log_buf_len),
1220		 new_descs, ilog2(new_descs_count),
1221		 new_infos);
1222
1223	local_irq_save(flags);
1224
1225	log_buf_len = new_log_buf_len;
1226	log_buf = new_log_buf;
1227	new_log_buf_len = 0;
1228
1229	free = __LOG_BUF_LEN;
1230	prb_for_each_record(0, &printk_rb_static, seq, &r) {
1231		text_size = add_to_rb(&printk_rb_dynamic, &r);
1232		if (text_size > free)
1233			free = 0;
1234		else
1235			free -= text_size;
1236	}
1237
1238	prb = &printk_rb_dynamic;
1239
1240	local_irq_restore(flags);
1241
1242	/*
1243	 * Copy any remaining messages that might have appeared from
1244	 * NMI context after copying but before switching to the
1245	 * dynamic buffer.
1246	 */
1247	prb_for_each_record(seq, &printk_rb_static, seq, &r) {
1248		text_size = add_to_rb(&printk_rb_dynamic, &r);
1249		if (text_size > free)
1250			free = 0;
1251		else
1252			free -= text_size;
1253	}
1254
1255	if (seq != prb_next_seq(&printk_rb_static)) {
1256		pr_err("dropped %llu messages\n",
1257		       prb_next_seq(&printk_rb_static) - seq);
1258	}
1259
1260	pr_info("log_buf_len: %u bytes\n", log_buf_len);
1261	pr_info("early log buf free: %u(%u%%)\n",
1262		free, (free * 100) / __LOG_BUF_LEN);
1263	return;
1264
1265err_free_descs:
1266	memblock_free(new_descs, new_descs_size);
1267err_free_log_buf:
1268	memblock_free(new_log_buf, new_log_buf_len);
 
 
1269}
1270
1271static bool __read_mostly ignore_loglevel;
1272
1273static int __init ignore_loglevel_setup(char *str)
1274{
1275	ignore_loglevel = true;
1276	pr_info("debug: ignoring loglevel setting.\n");
1277
1278	return 0;
1279}
1280
1281early_param("ignore_loglevel", ignore_loglevel_setup);
1282module_param(ignore_loglevel, bool, S_IRUGO | S_IWUSR);
1283MODULE_PARM_DESC(ignore_loglevel,
1284		 "ignore loglevel setting (prints all kernel messages to the console)");
1285
1286static bool suppress_message_printing(int level)
1287{
1288	return (level >= console_loglevel && !ignore_loglevel);
1289}
1290
1291#ifdef CONFIG_BOOT_PRINTK_DELAY
1292
1293static int boot_delay; /* msecs delay after each printk during bootup */
1294static unsigned long long loops_per_msec;	/* based on boot_delay */
1295
1296static int __init boot_delay_setup(char *str)
1297{
1298	unsigned long lpj;
1299
1300	lpj = preset_lpj ? preset_lpj : 1000000;	/* some guess */
1301	loops_per_msec = (unsigned long long)lpj / 1000 * HZ;
1302
1303	get_option(&str, &boot_delay);
1304	if (boot_delay > 10 * 1000)
1305		boot_delay = 0;
1306
1307	pr_debug("boot_delay: %u, preset_lpj: %ld, lpj: %lu, "
1308		"HZ: %d, loops_per_msec: %llu\n",
1309		boot_delay, preset_lpj, lpj, HZ, loops_per_msec);
1310	return 0;
1311}
1312early_param("boot_delay", boot_delay_setup);
1313
1314static void boot_delay_msec(int level)
1315{
1316	unsigned long long k;
1317	unsigned long timeout;
 
 
1318
1319	if ((boot_delay == 0 || system_state >= SYSTEM_RUNNING)
1320		|| suppress_message_printing(level)) {
1321		return;
1322	}
1323
1324	k = (unsigned long long)loops_per_msec * boot_delay;
1325
1326	timeout = jiffies + msecs_to_jiffies(boot_delay);
1327	while (k) {
1328		k--;
1329		cpu_relax();
1330		/*
1331		 * use (volatile) jiffies to prevent
1332		 * compiler reduction; loop termination via jiffies
1333		 * is secondary and may or may not happen.
1334		 */
1335		if (time_after(jiffies, timeout))
1336			break;
1337		touch_nmi_watchdog();
1338	}
1339}
1340#else
1341static inline void boot_delay_msec(int level)
1342{
1343}
1344#endif
1345
1346static bool printk_time = IS_ENABLED(CONFIG_PRINTK_TIME);
1347module_param_named(time, printk_time, bool, S_IRUGO | S_IWUSR);
1348
1349static size_t print_syslog(unsigned int level, char *buf)
1350{
1351	return sprintf(buf, "<%u>", level);
1352}
1353
1354static size_t print_time(u64 ts, char *buf)
1355{
1356	unsigned long rem_nsec = do_div(ts, 1000000000);
1357
1358	return sprintf(buf, "[%5lu.%06lu]",
1359		       (unsigned long)ts, rem_nsec / 1000);
1360}
1361
1362#ifdef CONFIG_PRINTK_CALLER
1363static size_t print_caller(u32 id, char *buf)
1364{
1365	char caller[12];
1366
1367	snprintf(caller, sizeof(caller), "%c%u",
1368		 id & 0x80000000 ? 'C' : 'T', id & ~0x80000000);
1369	return sprintf(buf, "[%6s]", caller);
1370}
1371#else
1372#define print_caller(id, buf) 0
1373#endif
1374
1375static size_t info_print_prefix(const struct printk_info  *info, bool syslog,
1376				bool time, char *buf)
1377{
1378	size_t len = 0;
1379
1380	if (syslog)
1381		len = print_syslog((info->facility << 3) | info->level, buf);
1382
1383	if (time)
1384		len += print_time(info->ts_nsec, buf + len);
1385
1386	len += print_caller(info->caller_id, buf + len);
1387
1388	if (IS_ENABLED(CONFIG_PRINTK_CALLER) || time) {
1389		buf[len++] = ' ';
1390		buf[len] = '\0';
1391	}
1392
1393	return len;
1394}
1395
1396/*
1397 * Prepare the record for printing. The text is shifted within the given
1398 * buffer to avoid a need for another one. The following operations are
1399 * done:
1400 *
1401 *   - Add prefix for each line.
1402 *   - Drop truncated lines that no longer fit into the buffer.
1403 *   - Add the trailing newline that has been removed in vprintk_store().
1404 *   - Add a string terminator.
1405 *
1406 * Since the produced string is always terminated, the maximum possible
1407 * return value is @r->text_buf_size - 1;
1408 *
1409 * Return: The length of the updated/prepared text, including the added
1410 * prefixes and the newline. The terminator is not counted. The dropped
1411 * line(s) are not counted.
1412 */
1413static size_t record_print_text(struct printk_record *r, bool syslog,
1414				bool time)
1415{
1416	size_t text_len = r->info->text_len;
1417	size_t buf_size = r->text_buf_size;
1418	char *text = r->text_buf;
1419	char prefix[PREFIX_MAX];
1420	bool truncated = false;
1421	size_t prefix_len;
1422	size_t line_len;
1423	size_t len = 0;
1424	char *next;
1425
1426	/*
1427	 * If the message was truncated because the buffer was not large
1428	 * enough, treat the available text as if it were the full text.
1429	 */
1430	if (text_len > buf_size)
1431		text_len = buf_size;
1432
1433	prefix_len = info_print_prefix(r->info, syslog, time, prefix);
1434
1435	/*
1436	 * @text_len: bytes of unprocessed text
1437	 * @line_len: bytes of current line _without_ newline
1438	 * @text:     pointer to beginning of current line
1439	 * @len:      number of bytes prepared in r->text_buf
1440	 */
1441	for (;;) {
1442		next = memchr(text, '\n', text_len);
1443		if (next) {
1444			line_len = next - text;
1445		} else {
1446			/* Drop truncated line(s). */
1447			if (truncated)
1448				break;
1449			line_len = text_len;
1450		}
1451
1452		/*
1453		 * Truncate the text if there is not enough space to add the
1454		 * prefix and a trailing newline and a terminator.
1455		 */
1456		if (len + prefix_len + text_len + 1 + 1 > buf_size) {
1457			/* Drop even the current line if no space. */
1458			if (len + prefix_len + line_len + 1 + 1 > buf_size)
1459				break;
1460
1461			text_len = buf_size - len - prefix_len - 1 - 1;
1462			truncated = true;
1463		}
1464
1465		memmove(text + prefix_len, text, text_len);
1466		memcpy(text, prefix, prefix_len);
1467
1468		/*
1469		 * Increment the prepared length to include the text and
1470		 * prefix that were just moved+copied. Also increment for the
1471		 * newline at the end of this line. If this is the last line,
1472		 * there is no newline, but it will be added immediately below.
1473		 */
1474		len += prefix_len + line_len + 1;
1475		if (text_len == line_len) {
1476			/*
1477			 * This is the last line. Add the trailing newline
1478			 * removed in vprintk_store().
1479			 */
1480			text[prefix_len + line_len] = '\n';
1481			break;
1482		}
1483
1484		/*
1485		 * Advance beyond the added prefix and the related line with
1486		 * its newline.
1487		 */
1488		text += prefix_len + line_len + 1;
1489
1490		/*
1491		 * The remaining text has only decreased by the line with its
1492		 * newline.
1493		 *
1494		 * Note that @text_len can become zero. It happens when @text
1495		 * ended with a newline (either due to truncation or the
1496		 * original string ending with "\n\n"). The loop is correctly
1497		 * repeated and (if not truncated) an empty line with a prefix
1498		 * will be prepared.
1499		 */
1500		text_len -= line_len + 1;
1501	}
1502
1503	/*
1504	 * If a buffer was provided, it will be terminated. Space for the
1505	 * string terminator is guaranteed to be available. The terminator is
1506	 * not counted in the return value.
1507	 */
1508	if (buf_size > 0)
1509		r->text_buf[len] = 0;
1510
1511	return len;
1512}
1513
1514static size_t get_record_print_text_size(struct printk_info *info,
1515					 unsigned int line_count,
1516					 bool syslog, bool time)
1517{
1518	char prefix[PREFIX_MAX];
1519	size_t prefix_len;
1520
1521	prefix_len = info_print_prefix(info, syslog, time, prefix);
1522
1523	/*
1524	 * Each line will be preceded with a prefix. The intermediate
1525	 * newlines are already within the text, but a final trailing
1526	 * newline will be added.
1527	 */
1528	return ((prefix_len * line_count) + info->text_len + 1);
1529}
1530
1531/*
1532 * Beginning with @start_seq, find the first record where it and all following
1533 * records up to (but not including) @max_seq fit into @size.
1534 *
1535 * @max_seq is simply an upper bound and does not need to exist. If the caller
1536 * does not require an upper bound, -1 can be used for @max_seq.
1537 */
1538static u64 find_first_fitting_seq(u64 start_seq, u64 max_seq, size_t size,
1539				  bool syslog, bool time)
1540{
1541	struct printk_info info;
1542	unsigned int line_count;
1543	size_t len = 0;
1544	u64 seq;
1545
1546	/* Determine the size of the records up to @max_seq. */
1547	prb_for_each_info(start_seq, prb, seq, &info, &line_count) {
1548		if (info.seq >= max_seq)
1549			break;
1550		len += get_record_print_text_size(&info, line_count, syslog, time);
1551	}
1552
1553	/*
1554	 * Adjust the upper bound for the next loop to avoid subtracting
1555	 * lengths that were never added.
1556	 */
1557	if (seq < max_seq)
1558		max_seq = seq;
1559
1560	/*
1561	 * Move first record forward until length fits into the buffer. Ignore
1562	 * newest messages that were not counted in the above cycle. Messages
1563	 * might appear and get lost in the meantime. This is a best effort
1564	 * that prevents an infinite loop that could occur with a retry.
1565	 */
1566	prb_for_each_info(start_seq, prb, seq, &info, &line_count) {
1567		if (len <= size || info.seq >= max_seq)
1568			break;
1569		len -= get_record_print_text_size(&info, line_count, syslog, time);
1570	}
1571
1572	return seq;
1573}
1574
1575/* The caller is responsible for making sure @size is greater than 0. */
1576static int syslog_print(char __user *buf, int size)
1577{
1578	struct printk_info info;
1579	struct printk_record r;
1580	char *text;
1581	int len = 0;
1582	u64 seq;
1583
1584	text = kmalloc(CONSOLE_LOG_MAX, GFP_KERNEL);
1585	if (!text)
1586		return -ENOMEM;
1587
1588	prb_rec_init_rd(&r, &info, text, CONSOLE_LOG_MAX);
1589
1590	mutex_lock(&syslog_lock);
1591
1592	/*
1593	 * Wait for the @syslog_seq record to be available. @syslog_seq may
1594	 * change while waiting.
1595	 */
1596	do {
1597		seq = syslog_seq;
1598
1599		mutex_unlock(&syslog_lock);
1600		/*
1601		 * Guarantee this task is visible on the waitqueue before
1602		 * checking the wake condition.
1603		 *
1604		 * The full memory barrier within set_current_state() of
1605		 * prepare_to_wait_event() pairs with the full memory barrier
1606		 * within wq_has_sleeper().
1607		 *
1608		 * This pairs with __wake_up_klogd:A.
1609		 */
1610		len = wait_event_interruptible(log_wait,
1611				prb_read_valid(prb, seq, NULL)); /* LMM(syslog_print:A) */
1612		mutex_lock(&syslog_lock);
1613
1614		if (len)
1615			goto out;
1616	} while (syslog_seq != seq);
1617
1618	/*
1619	 * Copy records that fit into the buffer. The above cycle makes sure
1620	 * that the first record is always available.
1621	 */
1622	do {
1623		size_t n;
1624		size_t skip;
1625		int err;
1626
1627		if (!prb_read_valid(prb, syslog_seq, &r))
1628			break;
1629
1630		if (r.info->seq != syslog_seq) {
1631			/* message is gone, move to next valid one */
1632			syslog_seq = r.info->seq;
1633			syslog_partial = 0;
1634		}
1635
1636		/*
1637		 * To keep reading/counting partial line consistent,
1638		 * use printk_time value as of the beginning of a line.
1639		 */
1640		if (!syslog_partial)
1641			syslog_time = printk_time;
1642
1643		skip = syslog_partial;
1644		n = record_print_text(&r, true, syslog_time);
1645		if (n - syslog_partial <= size) {
1646			/* message fits into buffer, move forward */
1647			syslog_seq = r.info->seq + 1;
1648			n -= syslog_partial;
1649			syslog_partial = 0;
1650		} else if (!len){
1651			/* partial read(), remember position */
1652			n = size;
1653			syslog_partial += n;
1654		} else
1655			n = 0;
1656
1657		if (!n)
1658			break;
1659
1660		mutex_unlock(&syslog_lock);
1661		err = copy_to_user(buf, text + skip, n);
1662		mutex_lock(&syslog_lock);
1663
1664		if (err) {
1665			if (!len)
1666				len = -EFAULT;
1667			break;
1668		}
1669
1670		len += n;
1671		size -= n;
1672		buf += n;
1673	} while (size);
1674out:
1675	mutex_unlock(&syslog_lock);
1676	kfree(text);
1677	return len;
1678}
1679
1680static int syslog_print_all(char __user *buf, int size, bool clear)
1681{
1682	struct printk_info info;
1683	struct printk_record r;
1684	char *text;
1685	int len = 0;
1686	u64 seq;
1687	bool time;
1688
1689	text = kmalloc(CONSOLE_LOG_MAX, GFP_KERNEL);
1690	if (!text)
1691		return -ENOMEM;
1692
1693	time = printk_time;
1694	/*
1695	 * Find first record that fits, including all following records,
1696	 * into the user-provided buffer for this dump.
1697	 */
1698	seq = find_first_fitting_seq(latched_seq_read_nolock(&clear_seq), -1,
1699				     size, true, time);
1700
1701	prb_rec_init_rd(&r, &info, text, CONSOLE_LOG_MAX);
1702
1703	len = 0;
1704	prb_for_each_record(seq, prb, seq, &r) {
1705		int textlen;
1706
1707		textlen = record_print_text(&r, true, time);
1708
1709		if (len + textlen > size) {
1710			seq--;
1711			break;
1712		}
1713
1714		if (copy_to_user(buf + len, text, textlen))
1715			len = -EFAULT;
1716		else
1717			len += textlen;
1718
1719		if (len < 0)
1720			break;
1721	}
1722
1723	if (clear) {
1724		mutex_lock(&syslog_lock);
1725		latched_seq_write(&clear_seq, seq);
1726		mutex_unlock(&syslog_lock);
1727	}
1728
1729	kfree(text);
1730	return len;
1731}
1732
1733static void syslog_clear(void)
1734{
1735	mutex_lock(&syslog_lock);
1736	latched_seq_write(&clear_seq, prb_next_seq(prb));
1737	mutex_unlock(&syslog_lock);
1738}
1739
1740int do_syslog(int type, char __user *buf, int len, int source)
1741{
1742	struct printk_info info;
1743	bool clear = false;
1744	static int saved_console_loglevel = LOGLEVEL_DEFAULT;
1745	int error;
1746
1747	error = check_syslog_permissions(type, source);
1748	if (error)
1749		return error;
1750
1751	switch (type) {
1752	case SYSLOG_ACTION_CLOSE:	/* Close log */
1753		break;
1754	case SYSLOG_ACTION_OPEN:	/* Open log */
1755		break;
1756	case SYSLOG_ACTION_READ:	/* Read from log */
1757		if (!buf || len < 0)
1758			return -EINVAL;
1759		if (!len)
1760			return 0;
1761		if (!access_ok(buf, len))
1762			return -EFAULT;
1763		error = syslog_print(buf, len);
1764		break;
1765	/* Read/clear last kernel messages */
1766	case SYSLOG_ACTION_READ_CLEAR:
1767		clear = true;
1768		fallthrough;
1769	/* Read last kernel messages */
1770	case SYSLOG_ACTION_READ_ALL:
1771		if (!buf || len < 0)
1772			return -EINVAL;
1773		if (!len)
1774			return 0;
1775		if (!access_ok(buf, len))
1776			return -EFAULT;
1777		error = syslog_print_all(buf, len, clear);
1778		break;
1779	/* Clear ring buffer */
1780	case SYSLOG_ACTION_CLEAR:
1781		syslog_clear();
1782		break;
1783	/* Disable logging to console */
1784	case SYSLOG_ACTION_CONSOLE_OFF:
1785		if (saved_console_loglevel == LOGLEVEL_DEFAULT)
1786			saved_console_loglevel = console_loglevel;
1787		console_loglevel = minimum_console_loglevel;
1788		break;
1789	/* Enable logging to console */
1790	case SYSLOG_ACTION_CONSOLE_ON:
1791		if (saved_console_loglevel != LOGLEVEL_DEFAULT) {
1792			console_loglevel = saved_console_loglevel;
1793			saved_console_loglevel = LOGLEVEL_DEFAULT;
1794		}
1795		break;
1796	/* Set level of messages printed to console */
1797	case SYSLOG_ACTION_CONSOLE_LEVEL:
1798		if (len < 1 || len > 8)
1799			return -EINVAL;
1800		if (len < minimum_console_loglevel)
1801			len = minimum_console_loglevel;
1802		console_loglevel = len;
1803		/* Implicitly re-enable logging to console */
1804		saved_console_loglevel = LOGLEVEL_DEFAULT;
1805		break;
1806	/* Number of chars in the log buffer */
1807	case SYSLOG_ACTION_SIZE_UNREAD:
1808		mutex_lock(&syslog_lock);
1809		if (!prb_read_valid_info(prb, syslog_seq, &info, NULL)) {
1810			/* No unread messages. */
1811			mutex_unlock(&syslog_lock);
1812			return 0;
1813		}
1814		if (info.seq != syslog_seq) {
1815			/* messages are gone, move to first one */
1816			syslog_seq = info.seq;
1817			syslog_partial = 0;
1818		}
1819		if (source == SYSLOG_FROM_PROC) {
1820			/*
1821			 * Short-cut for poll(/"proc/kmsg") which simply checks
1822			 * for pending data, not the size; return the count of
1823			 * records, not the length.
1824			 */
1825			error = prb_next_seq(prb) - syslog_seq;
1826		} else {
1827			bool time = syslog_partial ? syslog_time : printk_time;
1828			unsigned int line_count;
1829			u64 seq;
1830
1831			prb_for_each_info(syslog_seq, prb, seq, &info,
1832					  &line_count) {
1833				error += get_record_print_text_size(&info, line_count,
1834								    true, time);
1835				time = printk_time;
1836			}
1837			error -= syslog_partial;
1838		}
1839		mutex_unlock(&syslog_lock);
1840		break;
1841	/* Size of the log buffer */
1842	case SYSLOG_ACTION_SIZE_BUFFER:
1843		error = log_buf_len;
1844		break;
1845	default:
1846		error = -EINVAL;
1847		break;
1848	}
1849
1850	return error;
1851}
1852
1853SYSCALL_DEFINE3(syslog, int, type, char __user *, buf, int, len)
1854{
1855	return do_syslog(type, buf, len, SYSLOG_FROM_READER);
1856}
1857
1858/*
1859 * Special console_lock variants that help to reduce the risk of soft-lockups.
1860 * They allow to pass console_lock to another printk() call using a busy wait.
1861 */
1862
1863#ifdef CONFIG_LOCKDEP
1864static struct lockdep_map console_owner_dep_map = {
1865	.name = "console_owner"
1866};
1867#endif
1868
1869static DEFINE_RAW_SPINLOCK(console_owner_lock);
1870static struct task_struct *console_owner;
1871static bool console_waiter;
1872
1873/**
1874 * console_lock_spinning_enable - mark beginning of code where another
1875 *	thread might safely busy wait
1876 *
1877 * This basically converts console_lock into a spinlock. This marks
1878 * the section where the console_lock owner can not sleep, because
1879 * there may be a waiter spinning (like a spinlock). Also it must be
1880 * ready to hand over the lock at the end of the section.
1881 */
1882static void console_lock_spinning_enable(void)
1883{
 
 
 
 
 
 
 
 
 
 
 
 
1884	raw_spin_lock(&console_owner_lock);
1885	console_owner = current;
1886	raw_spin_unlock(&console_owner_lock);
1887
 
1888	/* The waiter may spin on us after setting console_owner */
1889	spin_acquire(&console_owner_dep_map, 0, 0, _THIS_IP_);
1890}
1891
1892/**
1893 * console_lock_spinning_disable_and_check - mark end of code where another
1894 *	thread was able to busy wait and check if there is a waiter
1895 * @cookie: cookie returned from console_srcu_read_lock()
1896 *
1897 * This is called at the end of the section where spinning is allowed.
1898 * It has two functions. First, it is a signal that it is no longer
1899 * safe to start busy waiting for the lock. Second, it checks if
1900 * there is a busy waiter and passes the lock rights to her.
1901 *
1902 * Important: Callers lose both the console_lock and the SRCU read lock if
1903 *	there was a busy waiter. They must not touch items synchronized by
1904 *	console_lock or SRCU read lock in this case.
1905 *
1906 * Return: 1 if the lock rights were passed, 0 otherwise.
1907 */
1908static int console_lock_spinning_disable_and_check(int cookie)
1909{
1910	int waiter;
1911
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1912	raw_spin_lock(&console_owner_lock);
1913	waiter = READ_ONCE(console_waiter);
1914	console_owner = NULL;
1915	raw_spin_unlock(&console_owner_lock);
1916
1917	if (!waiter) {
1918		spin_release(&console_owner_dep_map, _THIS_IP_);
1919		return 0;
1920	}
1921
1922	/* The waiter is now free to continue */
1923	WRITE_ONCE(console_waiter, false);
1924
1925	spin_release(&console_owner_dep_map, _THIS_IP_);
1926
1927	/*
1928	 * Preserve lockdep lock ordering. Release the SRCU read lock before
1929	 * releasing the console_lock.
1930	 */
1931	console_srcu_read_unlock(cookie);
1932
1933	/*
1934	 * Hand off console_lock to waiter. The waiter will perform
1935	 * the up(). After this, the waiter is the console_lock owner.
1936	 */
1937	mutex_release(&console_lock_dep_map, _THIS_IP_);
1938	return 1;
1939}
1940
1941/**
1942 * console_trylock_spinning - try to get console_lock by busy waiting
1943 *
1944 * This allows to busy wait for the console_lock when the current
1945 * owner is running in specially marked sections. It means that
1946 * the current owner is running and cannot reschedule until it
1947 * is ready to lose the lock.
1948 *
1949 * Return: 1 if we got the lock, 0 othrewise
1950 */
1951static int console_trylock_spinning(void)
1952{
1953	struct task_struct *owner = NULL;
1954	bool waiter;
1955	bool spin = false;
1956	unsigned long flags;
1957
1958	if (console_trylock())
1959		return 1;
1960
1961	/*
1962	 * It's unsafe to spin once a panic has begun. If we are the
1963	 * panic CPU, we may have already halted the owner of the
1964	 * console_sem. If we are not the panic CPU, then we should
1965	 * avoid taking console_sem, so the panic CPU has a better
1966	 * chance of cleanly acquiring it later.
1967	 */
1968	if (panic_in_progress())
1969		return 0;
1970
1971	printk_safe_enter_irqsave(flags);
1972
1973	raw_spin_lock(&console_owner_lock);
1974	owner = READ_ONCE(console_owner);
1975	waiter = READ_ONCE(console_waiter);
1976	if (!waiter && owner && owner != current) {
1977		WRITE_ONCE(console_waiter, true);
1978		spin = true;
1979	}
1980	raw_spin_unlock(&console_owner_lock);
1981
1982	/*
1983	 * If there is an active printk() writing to the
1984	 * consoles, instead of having it write our data too,
1985	 * see if we can offload that load from the active
1986	 * printer, and do some printing ourselves.
1987	 * Go into a spin only if there isn't already a waiter
1988	 * spinning, and there is an active printer, and
1989	 * that active printer isn't us (recursive printk?).
1990	 */
1991	if (!spin) {
1992		printk_safe_exit_irqrestore(flags);
1993		return 0;
1994	}
1995
1996	/* We spin waiting for the owner to release us */
1997	spin_acquire(&console_owner_dep_map, 0, 0, _THIS_IP_);
1998	/* Owner will clear console_waiter on hand off */
1999	while (READ_ONCE(console_waiter))
2000		cpu_relax();
2001	spin_release(&console_owner_dep_map, _THIS_IP_);
2002
2003	printk_safe_exit_irqrestore(flags);
2004	/*
2005	 * The owner passed the console lock to us.
2006	 * Since we did not spin on console lock, annotate
2007	 * this as a trylock. Otherwise lockdep will
2008	 * complain.
2009	 */
2010	mutex_acquire(&console_lock_dep_map, 0, 1, _THIS_IP_);
2011
 
 
 
 
 
 
2012	return 1;
2013}
2014
2015/*
2016 * Call the specified console driver, asking it to write out the specified
2017 * text and length. If @dropped_text is non-NULL and any records have been
2018 * dropped, a dropped message will be written out first.
2019 */
2020static void call_console_driver(struct console *con, const char *text, size_t len,
2021				char *dropped_text)
2022{
2023	size_t dropped_len;
2024
2025	if (con->dropped && dropped_text) {
2026		dropped_len = snprintf(dropped_text, DROPPED_TEXT_MAX,
2027				       "** %lu printk messages dropped **\n",
2028				       con->dropped);
2029		con->dropped = 0;
2030		con->write(con, dropped_text, dropped_len);
2031	}
2032
2033	con->write(con, text, len);
2034}
2035
2036/*
2037 * Recursion is tracked separately on each CPU. If NMIs are supported, an
2038 * additional NMI context per CPU is also separately tracked. Until per-CPU
2039 * is available, a separate "early tracking" is performed.
2040 */
2041static DEFINE_PER_CPU(u8, printk_count);
2042static u8 printk_count_early;
2043#ifdef CONFIG_HAVE_NMI
2044static DEFINE_PER_CPU(u8, printk_count_nmi);
2045static u8 printk_count_nmi_early;
2046#endif
2047
2048/*
2049 * Recursion is limited to keep the output sane. printk() should not require
2050 * more than 1 level of recursion (allowing, for example, printk() to trigger
2051 * a WARN), but a higher value is used in case some printk-internal errors
2052 * exist, such as the ringbuffer validation checks failing.
2053 */
2054#define PRINTK_MAX_RECURSION 3
2055
2056/*
2057 * Return a pointer to the dedicated counter for the CPU+context of the
2058 * caller.
2059 */
2060static u8 *__printk_recursion_counter(void)
2061{
2062#ifdef CONFIG_HAVE_NMI
2063	if (in_nmi()) {
2064		if (printk_percpu_data_ready())
2065			return this_cpu_ptr(&printk_count_nmi);
2066		return &printk_count_nmi_early;
2067	}
2068#endif
2069	if (printk_percpu_data_ready())
2070		return this_cpu_ptr(&printk_count);
2071	return &printk_count_early;
2072}
2073
2074/*
2075 * Enter recursion tracking. Interrupts are disabled to simplify tracking.
2076 * The caller must check the boolean return value to see if the recursion is
2077 * allowed. On failure, interrupts are not disabled.
2078 *
2079 * @recursion_ptr must be a variable of type (u8 *) and is the same variable
2080 * that is passed to printk_exit_irqrestore().
2081 */
2082#define printk_enter_irqsave(recursion_ptr, flags)	\
2083({							\
2084	bool success = true;				\
2085							\
2086	typecheck(u8 *, recursion_ptr);			\
2087	local_irq_save(flags);				\
2088	(recursion_ptr) = __printk_recursion_counter();	\
2089	if (*(recursion_ptr) > PRINTK_MAX_RECURSION) {	\
2090		local_irq_restore(flags);		\
2091		success = false;			\
2092	} else {					\
2093		(*(recursion_ptr))++;			\
2094	}						\
2095	success;					\
2096})
2097
2098/* Exit recursion tracking, restoring interrupts. */
2099#define printk_exit_irqrestore(recursion_ptr, flags)	\
2100	do {						\
2101		typecheck(u8 *, recursion_ptr);		\
2102		(*(recursion_ptr))--;			\
2103		local_irq_restore(flags);		\
2104	} while (0)
2105
2106int printk_delay_msec __read_mostly;
2107
2108static inline void printk_delay(int level)
2109{
2110	boot_delay_msec(level);
2111
2112	if (unlikely(printk_delay_msec)) {
2113		int m = printk_delay_msec;
2114
2115		while (m--) {
2116			mdelay(1);
2117			touch_nmi_watchdog();
2118		}
2119	}
2120}
2121
2122static inline u32 printk_caller_id(void)
2123{
2124	return in_task() ? task_pid_nr(current) :
2125		0x80000000 + smp_processor_id();
2126}
2127
2128/**
2129 * printk_parse_prefix - Parse level and control flags.
2130 *
2131 * @text:     The terminated text message.
2132 * @level:    A pointer to the current level value, will be updated.
2133 * @flags:    A pointer to the current printk_info flags, will be updated.
2134 *
2135 * @level may be NULL if the caller is not interested in the parsed value.
2136 * Otherwise the variable pointed to by @level must be set to
2137 * LOGLEVEL_DEFAULT in order to be updated with the parsed value.
2138 *
2139 * @flags may be NULL if the caller is not interested in the parsed value.
2140 * Otherwise the variable pointed to by @flags will be OR'd with the parsed
2141 * value.
2142 *
2143 * Return: The length of the parsed level and control flags.
2144 */
2145u16 printk_parse_prefix(const char *text, int *level,
2146			enum printk_info_flags *flags)
2147{
2148	u16 prefix_len = 0;
2149	int kern_level;
2150
2151	while (*text) {
2152		kern_level = printk_get_level(text);
2153		if (!kern_level)
2154			break;
2155
2156		switch (kern_level) {
2157		case '0' ... '7':
2158			if (level && *level == LOGLEVEL_DEFAULT)
2159				*level = kern_level - '0';
2160			break;
2161		case 'c':	/* KERN_CONT */
2162			if (flags)
2163				*flags |= LOG_CONT;
2164		}
2165
2166		prefix_len += 2;
2167		text += 2;
2168	}
2169
2170	return prefix_len;
2171}
2172
2173__printf(5, 0)
2174static u16 printk_sprint(char *text, u16 size, int facility,
2175			 enum printk_info_flags *flags, const char *fmt,
2176			 va_list args)
2177{
2178	u16 text_len;
2179
2180	text_len = vscnprintf(text, size, fmt, args);
2181
2182	/* Mark and strip a trailing newline. */
2183	if (text_len && text[text_len - 1] == '\n') {
2184		text_len--;
2185		*flags |= LOG_NEWLINE;
2186	}
2187
2188	/* Strip log level and control flags. */
2189	if (facility == 0) {
2190		u16 prefix_len;
2191
2192		prefix_len = printk_parse_prefix(text, NULL, NULL);
2193		if (prefix_len) {
2194			text_len -= prefix_len;
2195			memmove(text, text + prefix_len, text_len);
2196		}
2197	}
2198
2199	trace_console_rcuidle(text, text_len);
2200
2201	return text_len;
2202}
2203
2204__printf(4, 0)
2205int vprintk_store(int facility, int level,
2206		  const struct dev_printk_info *dev_info,
2207		  const char *fmt, va_list args)
2208{
2209	struct prb_reserved_entry e;
2210	enum printk_info_flags flags = 0;
2211	struct printk_record r;
2212	unsigned long irqflags;
2213	u16 trunc_msg_len = 0;
2214	char prefix_buf[8];
2215	u8 *recursion_ptr;
2216	u16 reserve_size;
2217	va_list args2;
2218	u32 caller_id;
2219	u16 text_len;
2220	int ret = 0;
2221	u64 ts_nsec;
2222
2223	if (!printk_enter_irqsave(recursion_ptr, irqflags))
2224		return 0;
2225
2226	/*
2227	 * Since the duration of printk() can vary depending on the message
2228	 * and state of the ringbuffer, grab the timestamp now so that it is
2229	 * close to the call of printk(). This provides a more deterministic
2230	 * timestamp with respect to the caller.
2231	 */
2232	ts_nsec = local_clock();
2233
2234	caller_id = printk_caller_id();
2235
2236	/*
2237	 * The sprintf needs to come first since the syslog prefix might be
2238	 * passed in as a parameter. An extra byte must be reserved so that
2239	 * later the vscnprintf() into the reserved buffer has room for the
2240	 * terminating '\0', which is not counted by vsnprintf().
2241	 */
2242	va_copy(args2, args);
2243	reserve_size = vsnprintf(&prefix_buf[0], sizeof(prefix_buf), fmt, args2) + 1;
2244	va_end(args2);
2245
2246	if (reserve_size > LOG_LINE_MAX)
2247		reserve_size = LOG_LINE_MAX;
2248
2249	/* Extract log level or control flags. */
2250	if (facility == 0)
2251		printk_parse_prefix(&prefix_buf[0], &level, &flags);
2252
2253	if (level == LOGLEVEL_DEFAULT)
2254		level = default_message_loglevel;
2255
2256	if (dev_info)
2257		flags |= LOG_NEWLINE;
2258
 
 
 
2259	if (flags & LOG_CONT) {
2260		prb_rec_init_wr(&r, reserve_size);
2261		if (prb_reserve_in_last(&e, prb, &r, caller_id, LOG_LINE_MAX)) {
2262			text_len = printk_sprint(&r.text_buf[r.info->text_len], reserve_size,
2263						 facility, &flags, fmt, args);
2264			r.info->text_len += text_len;
2265
 
 
 
2266			if (flags & LOG_NEWLINE) {
2267				r.info->flags |= LOG_NEWLINE;
2268				prb_final_commit(&e);
2269			} else {
2270				prb_commit(&e);
2271			}
2272
2273			ret = text_len;
2274			goto out;
2275		}
2276	}
2277
2278	/*
2279	 * Explicitly initialize the record before every prb_reserve() call.
2280	 * prb_reserve_in_last() and prb_reserve() purposely invalidate the
2281	 * structure when they fail.
2282	 */
2283	prb_rec_init_wr(&r, reserve_size);
2284	if (!prb_reserve(&e, prb, &r)) {
2285		/* truncate the message if it is too long for empty buffer */
2286		truncate_msg(&reserve_size, &trunc_msg_len);
2287
2288		prb_rec_init_wr(&r, reserve_size + trunc_msg_len);
2289		if (!prb_reserve(&e, prb, &r))
2290			goto out;
2291	}
2292
2293	/* fill message */
2294	text_len = printk_sprint(&r.text_buf[0], reserve_size, facility, &flags, fmt, args);
2295	if (trunc_msg_len)
2296		memcpy(&r.text_buf[text_len], trunc_msg, trunc_msg_len);
2297	r.info->text_len = text_len + trunc_msg_len;
2298	r.info->facility = facility;
2299	r.info->level = level & 7;
2300	r.info->flags = flags & 0x1f;
2301	r.info->ts_nsec = ts_nsec;
2302	r.info->caller_id = caller_id;
2303	if (dev_info)
2304		memcpy(&r.info->dev_info, dev_info, sizeof(r.info->dev_info));
2305
2306	/* A message without a trailing newline can be continued. */
2307	if (!(flags & LOG_NEWLINE))
2308		prb_commit(&e);
2309	else
2310		prb_final_commit(&e);
2311
2312	ret = text_len + trunc_msg_len;
2313out:
2314	printk_exit_irqrestore(recursion_ptr, irqflags);
2315	return ret;
2316}
2317
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
2318asmlinkage int vprintk_emit(int facility, int level,
2319			    const struct dev_printk_info *dev_info,
2320			    const char *fmt, va_list args)
2321{
 
2322	int printed_len;
2323	bool in_sched = false;
2324
2325	/* Suppress unimportant messages after panic happens */
2326	if (unlikely(suppress_printk))
2327		return 0;
2328
2329	if (unlikely(suppress_panic_printk) &&
2330	    atomic_read(&panic_cpu) != raw_smp_processor_id())
 
 
 
 
2331		return 0;
2332
 
 
 
2333	if (level == LOGLEVEL_SCHED) {
2334		level = LOGLEVEL_DEFAULT;
2335		in_sched = true;
 
2336	}
2337
2338	printk_delay(level);
2339
2340	printed_len = vprintk_store(facility, level, dev_info, fmt, args);
2341
2342	/* If called from the scheduler, we can not call up(). */
2343	if (!in_sched) {
 
 
 
 
 
2344		/*
2345		 * The caller may be holding system-critical or
2346		 * timing-sensitive locks. Disable preemption during
2347		 * printing of all remaining records to all consoles so that
2348		 * this context can return as soon as possible. Hopefully
2349		 * another printk() caller will take over the printing.
2350		 */
2351		preempt_disable();
2352		/*
2353		 * Try to acquire and then immediately release the console
2354		 * semaphore. The release will print out buffers. With the
2355		 * spinning variant, this context tries to take over the
2356		 * printing from another printing context.
2357		 */
2358		if (console_trylock_spinning())
2359			console_unlock();
2360		preempt_enable();
2361	}
2362
2363	wake_up_klogd();
 
 
 
 
2364	return printed_len;
2365}
2366EXPORT_SYMBOL(vprintk_emit);
2367
2368int vprintk_default(const char *fmt, va_list args)
2369{
2370	return vprintk_emit(0, LOGLEVEL_DEFAULT, NULL, fmt, args);
2371}
2372EXPORT_SYMBOL_GPL(vprintk_default);
2373
2374asmlinkage __visible int _printk(const char *fmt, ...)
2375{
2376	va_list args;
2377	int r;
2378
2379	va_start(args, fmt);
2380	r = vprintk(fmt, args);
2381	va_end(args);
2382
2383	return r;
2384}
2385EXPORT_SYMBOL(_printk);
2386
2387static bool pr_flush(int timeout_ms, bool reset_on_progress);
2388static bool __pr_flush(struct console *con, int timeout_ms, bool reset_on_progress);
2389
2390#else /* CONFIG_PRINTK */
2391
2392#define CONSOLE_LOG_MAX		0
2393#define DROPPED_TEXT_MAX	0
2394#define printk_time		false
2395
2396#define prb_read_valid(rb, seq, r)	false
2397#define prb_first_valid_seq(rb)		0
2398#define prb_next_seq(rb)		0
2399
2400static u64 syslog_seq;
2401
2402static size_t record_print_text(const struct printk_record *r,
2403				bool syslog, bool time)
2404{
2405	return 0;
2406}
2407static ssize_t info_print_ext_header(char *buf, size_t size,
2408				     struct printk_info *info)
2409{
2410	return 0;
2411}
2412static ssize_t msg_print_ext_body(char *buf, size_t size,
2413				  char *text, size_t text_len,
2414				  struct dev_printk_info *dev_info) { return 0; }
2415static void console_lock_spinning_enable(void) { }
2416static int console_lock_spinning_disable_and_check(int cookie) { return 0; }
2417static void call_console_driver(struct console *con, const char *text, size_t len,
2418				char *dropped_text)
2419{
2420}
2421static bool suppress_message_printing(int level) { return false; }
2422static bool pr_flush(int timeout_ms, bool reset_on_progress) { return true; }
2423static bool __pr_flush(struct console *con, int timeout_ms, bool reset_on_progress) { return true; }
2424
2425#endif /* CONFIG_PRINTK */
2426
2427#ifdef CONFIG_EARLY_PRINTK
2428struct console *early_console;
2429
2430asmlinkage __visible void early_printk(const char *fmt, ...)
2431{
2432	va_list ap;
2433	char buf[512];
2434	int n;
2435
2436	if (!early_console)
2437		return;
2438
2439	va_start(ap, fmt);
2440	n = vscnprintf(buf, sizeof(buf), fmt, ap);
2441	va_end(ap);
2442
2443	early_console->write(early_console, buf, n);
2444}
2445#endif
2446
2447static void set_user_specified(struct console_cmdline *c, bool user_specified)
2448{
2449	if (!user_specified)
2450		return;
2451
2452	/*
2453	 * @c console was defined by the user on the command line.
2454	 * Do not clear when added twice also by SPCR or the device tree.
2455	 */
2456	c->user_specified = true;
2457	/* At least one console defined by the user on the command line. */
2458	console_set_on_cmdline = 1;
2459}
2460
2461static int __add_preferred_console(char *name, int idx, char *options,
 
2462				   char *brl_options, bool user_specified)
2463{
2464	struct console_cmdline *c;
2465	int i;
2466
 
 
 
 
 
 
 
 
 
 
 
 
2467	/*
2468	 *	See if this tty is not yet registered, and
2469	 *	if we have a slot free.
2470	 */
2471	for (i = 0, c = console_cmdline;
2472	     i < MAX_CMDLINECONSOLES && c->name[0];
2473	     i++, c++) {
2474		if (strcmp(c->name, name) == 0 && c->index == idx) {
 
2475			if (!brl_options)
2476				preferred_console = i;
2477			set_user_specified(c, user_specified);
2478			return 0;
2479		}
2480	}
2481	if (i == MAX_CMDLINECONSOLES)
2482		return -E2BIG;
2483	if (!brl_options)
2484		preferred_console = i;
2485	strscpy(c->name, name, sizeof(c->name));
 
 
 
2486	c->options = options;
2487	set_user_specified(c, user_specified);
2488	braille_set_options(c, brl_options);
2489
2490	c->index = idx;
2491	return 0;
2492}
2493
2494static int __init console_msg_format_setup(char *str)
2495{
2496	if (!strcmp(str, "syslog"))
2497		console_msg_format = MSG_FORMAT_SYSLOG;
2498	if (!strcmp(str, "default"))
2499		console_msg_format = MSG_FORMAT_DEFAULT;
2500	return 1;
2501}
2502__setup("console_msg_format=", console_msg_format_setup);
2503
2504/*
2505 * Set up a console.  Called via do_early_param() in init/main.c
2506 * for each "console=" parameter in the boot command line.
2507 */
2508static int __init console_setup(char *str)
2509{
2510	char buf[sizeof(console_cmdline[0].name) + 4]; /* 4 for "ttyS" */
2511	char *s, *options, *brl_options = NULL;
 
 
 
 
 
2512	int idx;
2513
2514	/*
2515	 * console="" or console=null have been suggested as a way to
2516	 * disable console output. Use ttynull that has been created
2517	 * for exactly this purpose.
2518	 */
2519	if (str[0] == 0 || strcmp(str, "null") == 0) {
2520		__add_preferred_console("ttynull", 0, NULL, NULL, true);
2521		return 1;
2522	}
2523
2524	if (_braille_console_setup(&str, &brl_options))
2525		return 1;
2526
 
 
 
 
 
 
2527	/*
2528	 * Decode str into name, index, options.
2529	 */
2530	if (str[0] >= '0' && str[0] <= '9') {
2531		strcpy(buf, "ttyS");
2532		strncpy(buf + 4, str, sizeof(buf) - 5);
2533	} else {
2534		strncpy(buf, str, sizeof(buf) - 1);
2535	}
2536	buf[sizeof(buf) - 1] = 0;
2537	options = strchr(str, ',');
2538	if (options)
2539		*(options++) = 0;
 
2540#ifdef __sparc__
2541	if (!strcmp(str, "ttya"))
2542		strcpy(buf, "ttyS0");
2543	if (!strcmp(str, "ttyb"))
2544		strcpy(buf, "ttyS1");
2545#endif
 
2546	for (s = buf; *s; s++)
2547		if (isdigit(*s) || *s == ',')
2548			break;
2549	idx = simple_strtoul(s, NULL, 10);
 
 
 
 
 
 
2550	*s = 0;
2551
2552	__add_preferred_console(buf, idx, options, brl_options, true);
2553	return 1;
2554}
2555__setup("console=", console_setup);
2556
2557/**
2558 * add_preferred_console - add a device to the list of preferred consoles.
2559 * @name: device name
2560 * @idx: device index
2561 * @options: options for this console
2562 *
2563 * The last preferred console added will be used for kernel messages
2564 * and stdin/out/err for init.  Normally this is used by console_setup
2565 * above to handle user-supplied console arguments; however it can also
2566 * be used by arch-specific code either to override the user or more
2567 * commonly to provide a default console (ie from PROM variables) when
2568 * the user has not supplied one.
2569 */
2570int add_preferred_console(char *name, int idx, char *options)
2571{
2572	return __add_preferred_console(name, idx, options, NULL, false);
2573}
2574
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
2575bool console_suspend_enabled = true;
2576EXPORT_SYMBOL(console_suspend_enabled);
2577
2578static int __init console_suspend_disable(char *str)
2579{
2580	console_suspend_enabled = false;
2581	return 1;
2582}
2583__setup("no_console_suspend", console_suspend_disable);
2584module_param_named(console_suspend, console_suspend_enabled,
2585		bool, S_IRUGO | S_IWUSR);
2586MODULE_PARM_DESC(console_suspend, "suspend console during suspend"
2587	" and hibernate operations");
2588
2589static bool printk_console_no_auto_verbose;
2590
2591void console_verbose(void)
2592{
2593	if (console_loglevel && !printk_console_no_auto_verbose)
2594		console_loglevel = CONSOLE_LOGLEVEL_MOTORMOUTH;
2595}
2596EXPORT_SYMBOL_GPL(console_verbose);
2597
2598module_param_named(console_no_auto_verbose, printk_console_no_auto_verbose, bool, 0644);
2599MODULE_PARM_DESC(console_no_auto_verbose, "Disable console loglevel raise to highest on oops/panic/etc");
2600
2601/**
2602 * suspend_console - suspend the console subsystem
2603 *
2604 * This disables printk() while we go into suspend states
2605 */
2606void suspend_console(void)
2607{
 
 
2608	if (!console_suspend_enabled)
2609		return;
2610	pr_info("Suspending console(s) (use no_console_suspend to debug)\n");
2611	pr_flush(1000, true);
2612	console_lock();
2613	console_suspended = 1;
2614	up_console_sem();
 
 
 
 
 
 
 
 
 
 
2615}
2616
2617void resume_console(void)
2618{
 
 
 
2619	if (!console_suspend_enabled)
2620		return;
2621	down_console_sem();
2622	console_suspended = 0;
2623	console_unlock();
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
2624	pr_flush(1000, true);
2625}
2626
2627/**
2628 * console_cpu_notify - print deferred console messages after CPU hotplug
2629 * @cpu: unused
2630 *
2631 * If printk() is called from a CPU that is not online yet, the messages
2632 * will be printed on the console only if there are CON_ANYTIME consoles.
2633 * This function is called when a new CPU comes online (or fails to come
2634 * up) or goes offline.
2635 */
2636static int console_cpu_notify(unsigned int cpu)
2637{
 
 
2638	if (!cpuhp_tasks_frozen) {
2639		/* If trylock fails, someone else is doing the printing */
2640		if (console_trylock())
2641			console_unlock();
 
 
 
 
2642	}
2643	return 0;
2644}
2645
2646/**
2647 * console_lock - block the console subsystem from printing
2648 *
2649 * Acquires a lock which guarantees that no consoles will
2650 * be in or enter their write() callback.
2651 *
2652 * Can sleep, returns nothing.
2653 */
2654void console_lock(void)
2655{
2656	might_sleep();
2657
 
 
 
 
2658	down_console_sem();
2659	if (console_suspended)
2660		return;
2661	console_locked = 1;
2662	console_may_schedule = 1;
2663}
2664EXPORT_SYMBOL(console_lock);
2665
2666/**
2667 * console_trylock - try to block the console subsystem from printing
2668 *
2669 * Try to acquire a lock which guarantees that no consoles will
2670 * be in or enter their write() callback.
2671 *
2672 * returns 1 on success, and 0 on failure to acquire the lock.
2673 */
2674int console_trylock(void)
2675{
 
 
 
2676	if (down_trylock_console_sem())
2677		return 0;
2678	if (console_suspended) {
2679		up_console_sem();
2680		return 0;
2681	}
2682	console_locked = 1;
2683	console_may_schedule = 0;
2684	return 1;
2685}
2686EXPORT_SYMBOL(console_trylock);
2687
2688int is_console_locked(void)
2689{
2690	return console_locked;
2691}
2692EXPORT_SYMBOL(is_console_locked);
2693
 
 
 
 
 
 
 
 
2694/*
2695 * Return true when this CPU should unlock console_sem without pushing all
2696 * messages to the console. This reduces the chance that the console is
2697 * locked when the panic CPU tries to use it.
 
 
 
 
 
 
 
2698 */
2699static bool abandon_console_lock_in_panic(void)
 
2700{
2701	if (!panic_in_progress())
2702		return false;
 
 
 
 
 
 
 
 
 
2703
2704	/*
2705	 * We can use raw_smp_processor_id() here because it is impossible for
2706	 * the task to be migrated to the panic_cpu, or away from it. If
2707	 * panic_cpu has already been set, and we're not currently executing on
2708	 * that CPU, then we never will be.
2709	 */
2710	return atomic_read(&panic_cpu) != raw_smp_processor_id();
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
2711}
2712
2713/*
2714 * Check if the given console is currently capable and allowed to print
2715 * records.
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
2716 *
2717 * Requires the console_srcu_read_lock.
 
 
2718 */
2719static inline bool console_is_usable(struct console *con)
 
2720{
2721	short flags = console_srcu_read_flags(con);
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
2722
2723	if (!(flags & CON_ENABLED))
2724		return false;
2725
2726	if (!con->write)
2727		return false;
 
2728
2729	/*
2730	 * Console drivers may assume that per-cpu resources have been
2731	 * allocated. So unless they're explicitly marked as being able to
2732	 * cope (CON_ANYTIME) don't call them until this CPU is officially up.
2733	 */
2734	if (!cpu_online(raw_smp_processor_id()) && !(flags & CON_ANYTIME))
2735		return false;
2736
 
 
 
 
 
 
 
 
 
2737	return true;
2738}
2739
2740static void __console_unlock(void)
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
2741{
2742	console_locked = 0;
2743	up_console_sem();
2744}
 
 
 
 
 
 
 
 
2745
2746/*
2747 * Print one record for the given console. The record printed is whatever
2748 * record is the next available record for the given console.
2749 *
2750 * @text is a buffer of size CONSOLE_LOG_MAX.
2751 *
2752 * If extended messages should be printed, @ext_text is a buffer of size
2753 * CONSOLE_EXT_LOG_MAX. Otherwise @ext_text must be NULL.
2754 *
2755 * If dropped messages should be printed, @dropped_text is a buffer of size
2756 * DROPPED_TEXT_MAX. Otherwise @dropped_text must be NULL.
2757 *
2758 * @handover will be set to true if a printk waiter has taken over the
2759 * console_lock, in which case the caller is no longer holding both the
2760 * console_lock and the SRCU read lock. Otherwise it is set to false.
2761 *
2762 * @cookie is the cookie from the SRCU read lock.
2763 *
2764 * Returns false if the given console has no next record to print, otherwise
2765 * true.
2766 *
2767 * Requires the console_lock and the SRCU read lock.
2768 */
2769static bool console_emit_next_record(struct console *con, char *text, char *ext_text,
2770				     char *dropped_text, bool *handover, int cookie)
2771{
2772	static int panic_console_dropped;
2773	struct printk_info info;
2774	struct printk_record r;
 
 
2775	unsigned long flags;
2776	char *write_text;
2777	size_t len;
2778
2779	prb_rec_init_rd(&r, &info, text, CONSOLE_LOG_MAX);
2780
2781	*handover = false;
2782
2783	if (!prb_read_valid(prb, con->seq, &r))
2784		return false;
2785
2786	if (con->seq != r.info->seq) {
2787		con->dropped += r.info->seq - con->seq;
2788		con->seq = r.info->seq;
2789		if (panic_in_progress() && panic_console_dropped++ > 10) {
2790			suppress_panic_printk = 1;
2791			pr_warn_once("Too many dropped messages. Suppress messages on non-panic CPUs to prevent livelock.\n");
2792		}
2793	}
2794
2795	/* Skip record that has level above the console loglevel. */
2796	if (suppress_message_printing(r.info->level)) {
2797		con->seq++;
2798		goto skip;
2799	}
2800
2801	if (ext_text) {
2802		write_text = ext_text;
2803		len = info_print_ext_header(ext_text, CONSOLE_EXT_LOG_MAX, r.info);
2804		len += msg_print_ext_body(ext_text + len, CONSOLE_EXT_LOG_MAX - len,
2805					  &r.text_buf[0], r.info->text_len, &r.info->dev_info);
 
 
 
 
 
 
 
2806	} else {
2807		write_text = text;
2808		len = record_print_text(&r, console_msg_format & MSG_FORMAT_SYSLOG, printk_time);
2809	}
 
 
 
 
 
 
 
 
 
 
 
 
2810
2811	/*
2812	 * While actively printing out messages, if another printk()
2813	 * were to occur on another CPU, it may wait for this one to
2814	 * finish. This task can not be preempted if there is a
2815	 * waiter waiting to take over.
2816	 *
2817	 * Interrupts are disabled because the hand over to a waiter
2818	 * must not be interrupted until the hand over is completed
2819	 * (@console_waiter is cleared).
2820	 */
2821	printk_safe_enter_irqsave(flags);
2822	console_lock_spinning_enable();
2823
2824	stop_critical_timings();	/* don't trace print latency */
2825	call_console_driver(con, write_text, len, dropped_text);
2826	start_critical_timings();
2827
2828	con->seq++;
2829
2830	*handover = console_lock_spinning_disable_and_check(cookie);
2831	printk_safe_exit_irqrestore(flags);
 
2832skip:
2833	return true;
2834}
2835
 
 
 
 
 
 
 
 
 
 
 
 
2836/*
2837 * Print out all remaining records to all consoles.
2838 *
2839 * @do_cond_resched is set by the caller. It can be true only in schedulable
2840 * context.
2841 *
2842 * @next_seq is set to the sequence number after the last available record.
2843 * The value is valid only when this function returns true. It means that all
2844 * usable consoles are completely flushed.
2845 *
2846 * @handover will be set to true if a printk waiter has taken over the
2847 * console_lock, in which case the caller is no longer holding the
2848 * console_lock. Otherwise it is set to false.
2849 *
2850 * Returns true when there was at least one usable console and all messages
2851 * were flushed to all usable consoles. A returned false informs the caller
2852 * that everything was not flushed (either there were no usable consoles or
2853 * another context has taken over printing or it is a panic situation and this
2854 * is not the panic CPU). Regardless the reason, the caller should assume it
2855 * is not useful to immediately try again.
2856 *
2857 * Requires the console_lock.
2858 */
2859static bool console_flush_all(bool do_cond_resched, u64 *next_seq, bool *handover)
2860{
2861	static char dropped_text[DROPPED_TEXT_MAX];
2862	static char ext_text[CONSOLE_EXT_LOG_MAX];
2863	static char text[CONSOLE_LOG_MAX];
2864	bool any_usable = false;
2865	struct console *con;
2866	bool any_progress;
2867	int cookie;
2868
2869	*next_seq = 0;
2870	*handover = false;
2871
2872	do {
2873		any_progress = false;
2874
 
 
2875		cookie = console_srcu_read_lock();
2876		for_each_console_srcu(con) {
 
 
2877			bool progress;
2878
2879			if (!console_is_usable(con))
 
 
 
 
 
 
 
 
2880				continue;
2881			any_usable = true;
2882
2883			if (console_srcu_read_flags(con) & CON_EXTENDED) {
2884				/* Extended consoles do not print "dropped messages". */
2885				progress = console_emit_next_record(con, &text[0],
2886								    &ext_text[0], NULL,
2887								    handover, cookie);
2888			} else {
2889				progress = console_emit_next_record(con, &text[0],
2890								    NULL, &dropped_text[0],
2891								    handover, cookie);
2892			}
2893
2894			/*
2895			 * If a handover has occurred, the SRCU read lock
2896			 * is already released.
2897			 */
2898			if (*handover)
2899				return false;
2900
2901			/* Track the next of the highest seq flushed. */
2902			if (con->seq > *next_seq)
2903				*next_seq = con->seq;
2904
2905			if (!progress)
2906				continue;
2907			any_progress = true;
2908
2909			/* Allow panic_cpu to take over the consoles safely. */
2910			if (abandon_console_lock_in_panic())
2911				goto abandon;
2912
2913			if (do_cond_resched)
2914				cond_resched();
2915		}
2916		console_srcu_read_unlock(cookie);
2917	} while (any_progress);
2918
2919	return any_usable;
2920
2921abandon:
2922	console_srcu_read_unlock(cookie);
2923	return false;
2924}
2925
2926/**
2927 * console_unlock - unblock the console subsystem from printing
2928 *
2929 * Releases the console_lock which the caller holds to block printing of
2930 * the console subsystem.
2931 *
2932 * While the console_lock was held, console output may have been buffered
2933 * by printk().  If this is the case, console_unlock(); emits
2934 * the output prior to releasing the lock.
2935 *
2936 * console_unlock(); may be called from any context.
2937 */
2938void console_unlock(void)
2939{
2940	bool do_cond_resched;
2941	bool handover;
2942	bool flushed;
2943	u64 next_seq;
2944
2945	if (console_suspended) {
2946		up_console_sem();
2947		return;
2948	}
2949
2950	/*
2951	 * Console drivers are called with interrupts disabled, so
2952	 * @console_may_schedule should be cleared before; however, we may
2953	 * end up dumping a lot of lines, for example, if called from
2954	 * console registration path, and should invoke cond_resched()
2955	 * between lines if allowable.  Not doing so can cause a very long
2956	 * scheduling stall on a slow console leading to RCU stall and
2957	 * softlockup warnings which exacerbate the issue with more
2958	 * messages practically incapacitating the system. Therefore, create
2959	 * a local to use for the printing loop.
2960	 */
2961	do_cond_resched = console_may_schedule;
2962
2963	do {
2964		console_may_schedule = 0;
2965
2966		flushed = console_flush_all(do_cond_resched, &next_seq, &handover);
2967		if (!handover)
2968			__console_unlock();
2969
2970		/*
2971		 * Abort if there was a failure to flush all messages to all
2972		 * usable consoles. Either it is not possible to flush (in
2973		 * which case it would be an infinite loop of retrying) or
2974		 * another context has taken over printing.
2975		 */
2976		if (!flushed)
2977			break;
2978
2979		/*
2980		 * Some context may have added new records after
2981		 * console_flush_all() but before unlocking the console.
2982		 * Re-check if there is a new record to flush. If the trylock
2983		 * fails, another context is already handling the printing.
2984		 */
2985	} while (prb_read_valid(prb, next_seq, NULL) && console_trylock());
2986}
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
2987EXPORT_SYMBOL(console_unlock);
2988
2989/**
2990 * console_conditional_schedule - yield the CPU if required
2991 *
2992 * If the console code is currently allowed to sleep, and
2993 * if this CPU should yield the CPU to another task, do
2994 * so here.
2995 *
2996 * Must be called within console_lock();.
2997 */
2998void __sched console_conditional_schedule(void)
2999{
3000	if (console_may_schedule)
3001		cond_resched();
3002}
3003EXPORT_SYMBOL(console_conditional_schedule);
3004
3005void console_unblank(void)
3006{
 
3007	struct console *c;
3008	int cookie;
3009
3010	/*
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
3011	 * Stop console printing because the unblank() callback may
3012	 * assume the console is not within its write() callback.
3013	 *
3014	 * If @oops_in_progress is set, this may be an atomic context.
3015	 * In that case, attempt a trylock as best-effort.
3016	 */
3017	if (oops_in_progress) {
 
 
 
 
 
 
 
 
 
 
3018		if (down_trylock_console_sem() != 0)
3019			return;
3020	} else
3021		console_lock();
3022
3023	console_locked = 1;
3024	console_may_schedule = 0;
3025
3026	cookie = console_srcu_read_lock();
3027	for_each_console_srcu(c) {
3028		if ((console_srcu_read_flags(c) & CON_ENABLED) && c->unblank)
3029			c->unblank();
3030	}
3031	console_srcu_read_unlock(cookie);
3032
3033	console_unlock();
3034
3035	if (!oops_in_progress)
3036		pr_flush(1000, true);
3037}
3038
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
3039/**
3040 * console_flush_on_panic - flush console content on panic
3041 * @mode: flush all messages in buffer or just the pending ones
3042 *
3043 * Immediately output all pending messages no matter what.
3044 */
3045void console_flush_on_panic(enum con_flush_mode mode)
3046{
 
 
 
 
 
 
 
 
 
 
 
 
 
 
3047	/*
3048	 * If someone else is holding the console lock, trylock will fail
3049	 * and may_schedule may be set.  Ignore and proceed to unlock so
3050	 * that messages are flushed out.  As this can be called from any
3051	 * context and we don't want to get preempted while flushing,
3052	 * ensure may_schedule is cleared.
3053	 */
3054	console_trylock();
3055	console_may_schedule = 0;
3056
3057	if (mode == CONSOLE_REPLAY_ALL) {
3058		struct console *c;
3059		int cookie;
3060		u64 seq;
3061
3062		seq = prb_first_valid_seq(prb);
3063
3064		cookie = console_srcu_read_lock();
3065		for_each_console_srcu(c) {
3066			/*
3067			 * If the above console_trylock() failed, this is an
3068			 * unsynchronized assignment. But in that case, the
3069			 * kernel is in "hope and pray" mode anyway.
3070			 */
3071			c->seq = seq;
3072		}
3073		console_srcu_read_unlock(cookie);
3074	}
3075	console_unlock();
3076}
3077
3078/*
3079 * Return the console tty driver structure and its associated index
3080 */
3081struct tty_driver *console_device(int *index)
3082{
3083	struct console *c;
3084	struct tty_driver *driver = NULL;
3085	int cookie;
3086
3087	/*
3088	 * Take console_lock to serialize device() callback with
3089	 * other console operations. For example, fg_console is
3090	 * modified under console_lock when switching vt.
3091	 */
3092	console_lock();
3093
3094	cookie = console_srcu_read_lock();
3095	for_each_console_srcu(c) {
3096		if (!c->device)
3097			continue;
3098		driver = c->device(c, index);
3099		if (driver)
3100			break;
3101	}
3102	console_srcu_read_unlock(cookie);
3103
3104	console_unlock();
3105	return driver;
3106}
3107
3108/*
3109 * Prevent further output on the passed console device so that (for example)
3110 * serial drivers can disable console output before suspending a port, and can
3111 * re-enable output afterwards.
3112 */
3113void console_stop(struct console *console)
3114{
3115	__pr_flush(console, 1000, true);
3116	console_list_lock();
3117	console_srcu_write_flags(console, console->flags & ~CON_ENABLED);
3118	console_list_unlock();
3119
3120	/*
3121	 * Ensure that all SRCU list walks have completed. All contexts must
3122	 * be able to see that this console is disabled so that (for example)
3123	 * the caller can suspend the port without risk of another context
3124	 * using the port.
3125	 */
3126	synchronize_srcu(&console_srcu);
3127}
3128EXPORT_SYMBOL(console_stop);
3129
3130void console_start(struct console *console)
3131{
 
 
 
3132	console_list_lock();
3133	console_srcu_write_flags(console, console->flags | CON_ENABLED);
 
3134	console_list_unlock();
 
 
 
 
 
 
 
 
 
 
 
 
 
 
3135	__pr_flush(console, 1000, true);
3136}
3137EXPORT_SYMBOL(console_start);
3138
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
3139static int __read_mostly keep_bootcon;
3140
3141static int __init keep_bootcon_setup(char *str)
3142{
3143	keep_bootcon = 1;
3144	pr_info("debug: skip boot console de-registration.\n");
3145
3146	return 0;
3147}
3148
3149early_param("keep_bootcon", keep_bootcon_setup);
3150
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
3151/*
3152 * This is called by register_console() to try to match
3153 * the newly registered console with any of the ones selected
3154 * by either the command line or add_preferred_console() and
3155 * setup/enable it.
3156 *
3157 * Care need to be taken with consoles that are statically
3158 * enabled such as netconsole
3159 */
3160static int try_enable_preferred_console(struct console *newcon,
3161					bool user_specified)
3162{
3163	struct console_cmdline *c;
3164	int i, err;
3165
3166	for (i = 0, c = console_cmdline;
3167	     i < MAX_CMDLINECONSOLES && c->name[0];
3168	     i++, c++) {
 
 
 
3169		if (c->user_specified != user_specified)
3170			continue;
3171		if (!newcon->match ||
3172		    newcon->match(newcon, c->name, c->index, c->options) != 0) {
3173			/* default matching */
3174			BUILD_BUG_ON(sizeof(c->name) != sizeof(newcon->name));
3175			if (strcmp(c->name, newcon->name) != 0)
3176				continue;
3177			if (newcon->index >= 0 &&
3178			    newcon->index != c->index)
3179				continue;
3180			if (newcon->index < 0)
3181				newcon->index = c->index;
3182
3183			if (_braille_register_console(newcon, c))
3184				return 0;
3185
3186			if (newcon->setup &&
3187			    (err = newcon->setup(newcon, c->options)) != 0)
3188				return err;
3189		}
3190		newcon->flags |= CON_ENABLED;
3191		if (i == preferred_console)
3192			newcon->flags |= CON_CONSDEV;
3193		return 0;
3194	}
3195
3196	/*
3197	 * Some consoles, such as pstore and netconsole, can be enabled even
3198	 * without matching. Accept the pre-enabled consoles only when match()
3199	 * and setup() had a chance to be called.
3200	 */
3201	if (newcon->flags & CON_ENABLED && c->user_specified ==	user_specified)
3202		return 0;
3203
3204	return -ENOENT;
3205}
3206
3207/* Try to enable the console unconditionally */
3208static void try_enable_default_console(struct console *newcon)
3209{
3210	if (newcon->index < 0)
3211		newcon->index = 0;
3212
3213	if (newcon->setup && newcon->setup(newcon, NULL) != 0)
3214		return;
3215
3216	newcon->flags |= CON_ENABLED;
3217
3218	if (newcon->device)
3219		newcon->flags |= CON_CONSDEV;
3220}
3221
3222#define con_printk(lvl, con, fmt, ...)			\
3223	printk(lvl pr_fmt("%sconsole [%s%d] " fmt),	\
3224	       (con->flags & CON_BOOT) ? "boot" : "",	\
3225	       con->name, con->index, ##__VA_ARGS__)
3226
3227static void console_init_seq(struct console *newcon, bool bootcon_registered)
3228{
3229	struct console *con;
3230	bool handover;
 
3231
3232	if (newcon->flags & (CON_PRINTBUFFER | CON_BOOT)) {
3233		/* Get a consistent copy of @syslog_seq. */
3234		mutex_lock(&syslog_lock);
3235		newcon->seq = syslog_seq;
3236		mutex_unlock(&syslog_lock);
3237	} else {
3238		/* Begin with next message added to ringbuffer. */
3239		newcon->seq = prb_next_seq(prb);
3240
3241		/*
3242		 * If any enabled boot consoles are due to be unregistered
3243		 * shortly, some may not be caught up and may be the same
3244		 * device as @newcon. Since it is not known which boot console
3245		 * is the same device, flush all consoles and, if necessary,
3246		 * start with the message of the enabled boot console that is
3247		 * the furthest behind.
3248		 */
3249		if (bootcon_registered && !keep_bootcon) {
3250			/*
3251			 * Hold the console_lock to stop console printing and
3252			 * guarantee safe access to console->seq.
3253			 */
3254			console_lock();
3255
3256			/*
3257			 * Flush all consoles and set the console to start at
3258			 * the next unprinted sequence number.
3259			 */
3260			if (!console_flush_all(true, &newcon->seq, &handover)) {
3261				/*
3262				 * Flushing failed. Just choose the lowest
3263				 * sequence of the enabled boot consoles.
3264				 */
3265
3266				/*
3267				 * If there was a handover, this context no
3268				 * longer holds the console_lock.
3269				 */
3270				if (handover)
3271					console_lock();
3272
3273				newcon->seq = prb_next_seq(prb);
3274				for_each_console(con) {
3275					if ((con->flags & CON_BOOT) &&
3276					    (con->flags & CON_ENABLED) &&
3277					    con->seq < newcon->seq) {
3278						newcon->seq = con->seq;
 
3279					}
 
 
 
 
 
 
 
 
3280				}
3281			}
3282
3283			console_unlock();
3284		}
3285	}
 
 
3286}
3287
3288#define console_first()				\
3289	hlist_entry(console_list.first, struct console, node)
3290
3291static int unregister_console_locked(struct console *console);
3292
3293/*
3294 * The console driver calls this routine during kernel initialization
3295 * to register the console printing procedure with printk() and to
3296 * print any messages that were printed by the kernel before the
3297 * console driver was initialized.
3298 *
3299 * This can happen pretty early during the boot process (because of
3300 * early_printk) - sometimes before setup_arch() completes - be careful
3301 * of what kernel features are used - they may not be initialised yet.
3302 *
3303 * There are two types of consoles - bootconsoles (early_printk) and
3304 * "real" consoles (everything which is not a bootconsole) which are
3305 * handled differently.
3306 *  - Any number of bootconsoles can be registered at any time.
3307 *  - As soon as a "real" console is registered, all bootconsoles
3308 *    will be unregistered automatically.
3309 *  - Once a "real" console is registered, any attempt to register a
3310 *    bootconsoles will be rejected
3311 */
3312void register_console(struct console *newcon)
3313{
3314	struct console *con;
3315	bool bootcon_registered = false;
3316	bool realcon_registered = false;
 
 
 
3317	int err;
3318
3319	console_list_lock();
3320
3321	for_each_console(con) {
3322		if (WARN(con == newcon, "console '%s%d' already registered\n",
3323					 con->name, con->index)) {
3324			goto unlock;
3325		}
3326
3327		if (con->flags & CON_BOOT)
3328			bootcon_registered = true;
3329		else
3330			realcon_registered = true;
3331	}
3332
3333	/* Do not register boot consoles when there already is a real one. */
3334	if ((newcon->flags & CON_BOOT) && realcon_registered) {
3335		pr_info("Too late to register bootconsole %s%d\n",
3336			newcon->name, newcon->index);
3337		goto unlock;
3338	}
3339
 
 
 
 
 
 
 
 
 
3340	/*
3341	 * See if we want to enable this console driver by default.
3342	 *
3343	 * Nope when a console is preferred by the command line, device
3344	 * tree, or SPCR.
3345	 *
3346	 * The first real console with tty binding (driver) wins. More
3347	 * consoles might get enabled before the right one is found.
3348	 *
3349	 * Note that a console with tty binding will have CON_CONSDEV
3350	 * flag set and will be first in the list.
3351	 */
3352	if (preferred_console < 0) {
3353		if (hlist_empty(&console_list) || !console_first()->device ||
3354		    console_first()->flags & CON_BOOT) {
3355			try_enable_default_console(newcon);
3356		}
3357	}
3358
3359	/* See if this console matches one we selected on the command line */
3360	err = try_enable_preferred_console(newcon, true);
3361
3362	/* If not, try to match against the platform default(s) */
3363	if (err == -ENOENT)
3364		err = try_enable_preferred_console(newcon, false);
3365
3366	/* printk() messages are not printed to the Braille console. */
3367	if (err || newcon->flags & CON_BRL)
 
 
3368		goto unlock;
 
3369
3370	/*
3371	 * If we have a bootconsole, and are switching to a real console,
3372	 * don't print everything out again, since when the boot console, and
3373	 * the real console are the same physical device, it's annoying to
3374	 * see the beginning boot messages twice
3375	 */
3376	if (bootcon_registered &&
3377	    ((newcon->flags & (CON_CONSDEV | CON_BOOT)) == CON_CONSDEV)) {
3378		newcon->flags &= ~CON_PRINTBUFFER;
3379	}
3380
3381	newcon->dropped = 0;
3382	console_init_seq(newcon, bootcon_registered);
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
3383
3384	/*
3385	 * Put this console in the list - keep the
3386	 * preferred driver at the head of the list.
3387	 */
3388	if (hlist_empty(&console_list)) {
3389		/* Ensure CON_CONSDEV is always set for the head. */
3390		newcon->flags |= CON_CONSDEV;
3391		hlist_add_head_rcu(&newcon->node, &console_list);
3392
3393	} else if (newcon->flags & CON_CONSDEV) {
3394		/* Only the new head can have CON_CONSDEV set. */
3395		console_srcu_write_flags(console_first(), console_first()->flags & ~CON_CONSDEV);
3396		hlist_add_head_rcu(&newcon->node, &console_list);
3397
3398	} else {
3399		hlist_add_behind_rcu(&newcon->node, console_list.first);
3400	}
3401
3402	/*
3403	 * No need to synchronize SRCU here! The caller does not rely
3404	 * on all contexts being able to see the new console before
3405	 * register_console() completes.
3406	 */
3407
 
 
 
 
3408	console_sysfs_notify();
3409
3410	/*
3411	 * By unregistering the bootconsoles after we enable the real console
3412	 * we get the "console xxx enabled" message on all the consoles -
3413	 * boot consoles, real consoles, etc - this is to ensure that end
3414	 * users know there might be something in the kernel's log buffer that
3415	 * went to the bootconsole (that they do not see on the real console)
3416	 */
3417	con_printk(KERN_INFO, newcon, "enabled\n");
3418	if (bootcon_registered &&
3419	    ((newcon->flags & (CON_CONSDEV | CON_BOOT)) == CON_CONSDEV) &&
3420	    !keep_bootcon) {
3421		struct hlist_node *tmp;
3422
3423		hlist_for_each_entry_safe(con, tmp, &console_list, node) {
3424			if (con->flags & CON_BOOT)
3425				unregister_console_locked(con);
3426		}
3427	}
 
 
 
3428unlock:
3429	console_list_unlock();
3430}
3431EXPORT_SYMBOL(register_console);
3432
3433/* Must be called under console_list_lock(). */
3434static int unregister_console_locked(struct console *console)
3435{
 
 
 
 
 
 
3436	int res;
3437
3438	lockdep_assert_console_list_lock_held();
3439
3440	con_printk(KERN_INFO, console, "disabled\n");
3441
3442	res = _braille_unregister_console(console);
3443	if (res < 0)
3444		return res;
3445	if (res > 0)
3446		return 0;
3447
 
 
 
 
 
3448	/* Disable it unconditionally */
3449	console_srcu_write_flags(console, console->flags & ~CON_ENABLED);
3450
3451	if (!console_is_registered_locked(console))
3452		return -ENODEV;
 
 
 
 
 
 
 
3453
3454	hlist_del_init_rcu(&console->node);
3455
 
 
 
3456	/*
3457	 * <HISTORICAL>
3458	 * If this isn't the last console and it has CON_CONSDEV set, we
3459	 * need to set it on the next preferred console.
3460	 * </HISTORICAL>
3461	 *
3462	 * The above makes no sense as there is no guarantee that the next
3463	 * console has any device attached. Oh well....
3464	 */
3465	if (!hlist_empty(&console_list) && console->flags & CON_CONSDEV)
3466		console_srcu_write_flags(console_first(), console_first()->flags | CON_CONSDEV);
3467
3468	/*
3469	 * Ensure that all SRCU list walks have completed. All contexts
3470	 * must not be able to see this console in the list so that any
3471	 * exit/cleanup routines can be performed safely.
3472	 */
3473	synchronize_srcu(&console_srcu);
3474
 
 
 
3475	console_sysfs_notify();
3476
3477	if (console->exit)
3478		res = console->exit(console);
3479
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
3480	return res;
3481}
3482
3483int unregister_console(struct console *console)
3484{
3485	int res;
3486
3487	console_list_lock();
3488	res = unregister_console_locked(console);
3489	console_list_unlock();
3490	return res;
3491}
3492EXPORT_SYMBOL(unregister_console);
3493
3494/**
3495 * console_force_preferred_locked - force a registered console preferred
3496 * @con: The registered console to force preferred.
3497 *
3498 * Must be called under console_list_lock().
3499 */
3500void console_force_preferred_locked(struct console *con)
3501{
3502	struct console *cur_pref_con;
3503
3504	if (!console_is_registered_locked(con))
3505		return;
3506
3507	cur_pref_con = console_first();
3508
3509	/* Already preferred? */
3510	if (cur_pref_con == con)
3511		return;
3512
3513	/*
3514	 * Delete, but do not re-initialize the entry. This allows the console
3515	 * to continue to appear registered (via any hlist_unhashed_lockless()
3516	 * checks), even though it was briefly removed from the console list.
3517	 */
3518	hlist_del_rcu(&con->node);
3519
3520	/*
3521	 * Ensure that all SRCU list walks have completed so that the console
3522	 * can be added to the beginning of the console list and its forward
3523	 * list pointer can be re-initialized.
3524	 */
3525	synchronize_srcu(&console_srcu);
3526
3527	con->flags |= CON_CONSDEV;
3528	WARN_ON(!con->device);
3529
3530	/* Only the new head can have CON_CONSDEV set. */
3531	console_srcu_write_flags(cur_pref_con, cur_pref_con->flags & ~CON_CONSDEV);
3532	hlist_add_head_rcu(&con->node, &console_list);
3533}
3534EXPORT_SYMBOL(console_force_preferred_locked);
3535
3536/*
3537 * Initialize the console device. This is called *early*, so
3538 * we can't necessarily depend on lots of kernel help here.
3539 * Just do some early initializations, and do the complex setup
3540 * later.
3541 */
3542void __init console_init(void)
3543{
3544	int ret;
3545	initcall_t call;
3546	initcall_entry_t *ce;
3547
3548	/* Setup the default TTY line discipline. */
3549	n_tty_init();
3550
3551	/*
3552	 * set up the console device so that later boot sequences can
3553	 * inform about problems etc..
3554	 */
3555	ce = __con_initcall_start;
3556	trace_initcall_level("console");
3557	while (ce < __con_initcall_end) {
3558		call = initcall_from_entry(ce);
3559		trace_initcall_start(call);
3560		ret = call();
3561		trace_initcall_finish(call, ret);
3562		ce++;
3563	}
3564}
3565
3566/*
3567 * Some boot consoles access data that is in the init section and which will
3568 * be discarded after the initcalls have been run. To make sure that no code
3569 * will access this data, unregister the boot consoles in a late initcall.
3570 *
3571 * If for some reason, such as deferred probe or the driver being a loadable
3572 * module, the real console hasn't registered yet at this point, there will
3573 * be a brief interval in which no messages are logged to the console, which
3574 * makes it difficult to diagnose problems that occur during this time.
3575 *
3576 * To mitigate this problem somewhat, only unregister consoles whose memory
3577 * intersects with the init section. Note that all other boot consoles will
3578 * get unregistered when the real preferred console is registered.
3579 */
3580static int __init printk_late_init(void)
3581{
3582	struct hlist_node *tmp;
3583	struct console *con;
3584	int ret;
3585
3586	console_list_lock();
3587	hlist_for_each_entry_safe(con, tmp, &console_list, node) {
3588		if (!(con->flags & CON_BOOT))
3589			continue;
3590
3591		/* Check addresses that might be used for enabled consoles. */
3592		if (init_section_intersects(con, sizeof(*con)) ||
3593		    init_section_contains(con->write, 0) ||
3594		    init_section_contains(con->read, 0) ||
3595		    init_section_contains(con->device, 0) ||
3596		    init_section_contains(con->unblank, 0) ||
3597		    init_section_contains(con->data, 0)) {
3598			/*
3599			 * Please, consider moving the reported consoles out
3600			 * of the init section.
3601			 */
3602			pr_warn("bootconsole [%s%d] uses init memory and must be disabled even before the real one is ready\n",
3603				con->name, con->index);
3604			unregister_console_locked(con);
3605		}
3606	}
3607	console_list_unlock();
3608
3609	ret = cpuhp_setup_state_nocalls(CPUHP_PRINTK_DEAD, "printk:dead", NULL,
3610					console_cpu_notify);
3611	WARN_ON(ret < 0);
3612	ret = cpuhp_setup_state_nocalls(CPUHP_AP_ONLINE_DYN, "printk:online",
3613					console_cpu_notify, NULL);
3614	WARN_ON(ret < 0);
3615	printk_sysctl_init();
3616	return 0;
3617}
3618late_initcall(printk_late_init);
3619
3620#if defined CONFIG_PRINTK
3621/* If @con is specified, only wait for that console. Otherwise wait for all. */
3622static bool __pr_flush(struct console *con, int timeout_ms, bool reset_on_progress)
3623{
3624	int remaining = timeout_ms;
 
 
3625	struct console *c;
3626	u64 last_diff = 0;
3627	u64 printk_seq;
 
3628	int cookie;
3629	u64 diff;
3630	u64 seq;
3631
 
 
 
 
3632	might_sleep();
3633
3634	seq = prb_next_seq(prb);
 
 
 
 
 
 
 
 
 
3635
3636	for (;;) {
 
 
 
3637		diff = 0;
3638
3639		/*
3640		 * Hold the console_lock to guarantee safe access to
3641		 * console->seq and to prevent changes to @console_suspended
3642		 * until all consoles have been processed.
 
 
 
 
 
 
 
3643		 */
3644		console_lock();
3645
3646		cookie = console_srcu_read_lock();
3647		for_each_console_srcu(c) {
3648			if (con && con != c)
3649				continue;
3650			if (!console_is_usable(c))
 
 
 
 
 
 
 
 
 
3651				continue;
3652			printk_seq = c->seq;
 
 
 
 
 
 
 
3653			if (printk_seq < seq)
3654				diff += seq - printk_seq;
3655		}
3656		console_srcu_read_unlock(cookie);
3657
3658		/*
3659		 * If consoles are suspended, it cannot be expected that they
3660		 * make forward progress, so timeout immediately. @diff is
3661		 * still used to return a valid flush status.
3662		 */
3663		if (console_suspended)
3664			remaining = 0;
3665		else if (diff != last_diff && reset_on_progress)
3666			remaining = timeout_ms;
3667
3668		console_unlock();
3669
3670		if (diff == 0 || remaining == 0)
 
3671			break;
3672
3673		if (remaining < 0) {
3674			/* no timeout limit */
3675			msleep(100);
3676		} else if (remaining < 100) {
3677			msleep(remaining);
3678			remaining = 0;
3679		} else {
3680			msleep(100);
3681			remaining -= 100;
3682		}
3683
3684		last_diff = diff;
3685	}
3686
3687	return (diff == 0);
3688}
3689
3690/**
3691 * pr_flush() - Wait for printing threads to catch up.
3692 *
3693 * @timeout_ms:        The maximum time (in ms) to wait.
3694 * @reset_on_progress: Reset the timeout if forward progress is seen.
3695 *
3696 * A value of 0 for @timeout_ms means no waiting will occur. A value of -1
3697 * represents infinite waiting.
3698 *
3699 * If @reset_on_progress is true, the timeout will be reset whenever any
3700 * printer has been seen to make some forward progress.
3701 *
3702 * Context: Process context. May sleep while acquiring console lock.
3703 * Return: true if all enabled printers are caught up.
3704 */
3705static bool pr_flush(int timeout_ms, bool reset_on_progress)
3706{
3707	return __pr_flush(NULL, timeout_ms, reset_on_progress);
3708}
3709
3710/*
3711 * Delayed printk version, for scheduler-internal messages:
3712 */
3713#define PRINTK_PENDING_WAKEUP	0x01
3714#define PRINTK_PENDING_OUTPUT	0x02
3715
3716static DEFINE_PER_CPU(int, printk_pending);
3717
3718static void wake_up_klogd_work_func(struct irq_work *irq_work)
3719{
3720	int pending = this_cpu_xchg(printk_pending, 0);
3721
3722	if (pending & PRINTK_PENDING_OUTPUT) {
3723		/* If trylock fails, someone else is doing the printing */
3724		if (console_trylock())
3725			console_unlock();
 
 
 
 
3726	}
3727
3728	if (pending & PRINTK_PENDING_WAKEUP)
3729		wake_up_interruptible(&log_wait);
3730}
3731
3732static DEFINE_PER_CPU(struct irq_work, wake_up_klogd_work) =
3733	IRQ_WORK_INIT_LAZY(wake_up_klogd_work_func);
3734
3735static void __wake_up_klogd(int val)
3736{
3737	if (!printk_percpu_data_ready())
3738		return;
3739
3740	preempt_disable();
3741	/*
3742	 * Guarantee any new records can be seen by tasks preparing to wait
3743	 * before this context checks if the wait queue is empty.
3744	 *
3745	 * The full memory barrier within wq_has_sleeper() pairs with the full
3746	 * memory barrier within set_current_state() of
3747	 * prepare_to_wait_event(), which is called after ___wait_event() adds
3748	 * the waiter but before it has checked the wait condition.
3749	 *
3750	 * This pairs with devkmsg_read:A and syslog_print:A.
3751	 */
3752	if (wq_has_sleeper(&log_wait) || /* LMM(__wake_up_klogd:A) */
3753	    (val & PRINTK_PENDING_OUTPUT)) {
3754		this_cpu_or(printk_pending, val);
3755		irq_work_queue(this_cpu_ptr(&wake_up_klogd_work));
3756	}
3757	preempt_enable();
3758}
3759
 
 
 
 
 
 
 
 
 
 
3760void wake_up_klogd(void)
3761{
3762	__wake_up_klogd(PRINTK_PENDING_WAKEUP);
3763}
3764
 
 
 
 
 
 
 
 
 
 
 
 
3765void defer_console_output(void)
3766{
3767	/*
3768	 * New messages may have been added directly to the ringbuffer
3769	 * using vprintk_store(), so wake any waiters as well.
3770	 */
3771	__wake_up_klogd(PRINTK_PENDING_WAKEUP | PRINTK_PENDING_OUTPUT);
3772}
3773
3774void printk_trigger_flush(void)
3775{
3776	defer_console_output();
3777}
3778
3779int vprintk_deferred(const char *fmt, va_list args)
3780{
3781	int r;
3782
3783	r = vprintk_emit(0, LOGLEVEL_SCHED, NULL, fmt, args);
3784	defer_console_output();
3785
3786	return r;
3787}
3788
3789int _printk_deferred(const char *fmt, ...)
3790{
3791	va_list args;
3792	int r;
3793
3794	va_start(args, fmt);
3795	r = vprintk_deferred(fmt, args);
3796	va_end(args);
3797
3798	return r;
3799}
3800
3801/*
3802 * printk rate limiting, lifted from the networking subsystem.
3803 *
3804 * This enforces a rate limit: not more than 10 kernel messages
3805 * every 5s to make a denial-of-service attack impossible.
3806 */
3807DEFINE_RATELIMIT_STATE(printk_ratelimit_state, 5 * HZ, 10);
3808
3809int __printk_ratelimit(const char *func)
3810{
3811	return ___ratelimit(&printk_ratelimit_state, func);
3812}
3813EXPORT_SYMBOL(__printk_ratelimit);
3814
3815/**
3816 * printk_timed_ratelimit - caller-controlled printk ratelimiting
3817 * @caller_jiffies: pointer to caller's state
3818 * @interval_msecs: minimum interval between prints
3819 *
3820 * printk_timed_ratelimit() returns true if more than @interval_msecs
3821 * milliseconds have elapsed since the last time printk_timed_ratelimit()
3822 * returned true.
3823 */
3824bool printk_timed_ratelimit(unsigned long *caller_jiffies,
3825			unsigned int interval_msecs)
3826{
3827	unsigned long elapsed = jiffies - *caller_jiffies;
3828
3829	if (*caller_jiffies && elapsed <= msecs_to_jiffies(interval_msecs))
3830		return false;
3831
3832	*caller_jiffies = jiffies;
3833	return true;
3834}
3835EXPORT_SYMBOL(printk_timed_ratelimit);
3836
3837static DEFINE_SPINLOCK(dump_list_lock);
3838static LIST_HEAD(dump_list);
3839
3840/**
3841 * kmsg_dump_register - register a kernel log dumper.
3842 * @dumper: pointer to the kmsg_dumper structure
3843 *
3844 * Adds a kernel log dumper to the system. The dump callback in the
3845 * structure will be called when the kernel oopses or panics and must be
3846 * set. Returns zero on success and %-EINVAL or %-EBUSY otherwise.
3847 */
3848int kmsg_dump_register(struct kmsg_dumper *dumper)
3849{
3850	unsigned long flags;
3851	int err = -EBUSY;
3852
3853	/* The dump callback needs to be set */
3854	if (!dumper->dump)
3855		return -EINVAL;
3856
3857	spin_lock_irqsave(&dump_list_lock, flags);
3858	/* Don't allow registering multiple times */
3859	if (!dumper->registered) {
3860		dumper->registered = 1;
3861		list_add_tail_rcu(&dumper->list, &dump_list);
3862		err = 0;
3863	}
3864	spin_unlock_irqrestore(&dump_list_lock, flags);
3865
3866	return err;
3867}
3868EXPORT_SYMBOL_GPL(kmsg_dump_register);
3869
3870/**
3871 * kmsg_dump_unregister - unregister a kmsg dumper.
3872 * @dumper: pointer to the kmsg_dumper structure
3873 *
3874 * Removes a dump device from the system. Returns zero on success and
3875 * %-EINVAL otherwise.
3876 */
3877int kmsg_dump_unregister(struct kmsg_dumper *dumper)
3878{
3879	unsigned long flags;
3880	int err = -EINVAL;
3881
3882	spin_lock_irqsave(&dump_list_lock, flags);
3883	if (dumper->registered) {
3884		dumper->registered = 0;
3885		list_del_rcu(&dumper->list);
3886		err = 0;
3887	}
3888	spin_unlock_irqrestore(&dump_list_lock, flags);
3889	synchronize_rcu();
3890
3891	return err;
3892}
3893EXPORT_SYMBOL_GPL(kmsg_dump_unregister);
3894
3895static bool always_kmsg_dump;
3896module_param_named(always_kmsg_dump, always_kmsg_dump, bool, S_IRUGO | S_IWUSR);
3897
3898const char *kmsg_dump_reason_str(enum kmsg_dump_reason reason)
3899{
3900	switch (reason) {
3901	case KMSG_DUMP_PANIC:
3902		return "Panic";
3903	case KMSG_DUMP_OOPS:
3904		return "Oops";
3905	case KMSG_DUMP_EMERG:
3906		return "Emergency";
3907	case KMSG_DUMP_SHUTDOWN:
3908		return "Shutdown";
3909	default:
3910		return "Unknown";
3911	}
3912}
3913EXPORT_SYMBOL_GPL(kmsg_dump_reason_str);
3914
3915/**
3916 * kmsg_dump - dump kernel log to kernel message dumpers.
3917 * @reason: the reason (oops, panic etc) for dumping
 
 
3918 *
3919 * Call each of the registered dumper's dump() callback, which can
3920 * retrieve the kmsg records with kmsg_dump_get_line() or
3921 * kmsg_dump_get_buffer().
3922 */
3923void kmsg_dump(enum kmsg_dump_reason reason)
3924{
3925	struct kmsg_dumper *dumper;
 
 
 
3926
3927	rcu_read_lock();
3928	list_for_each_entry_rcu(dumper, &dump_list, list) {
3929		enum kmsg_dump_reason max_reason = dumper->max_reason;
3930
3931		/*
3932		 * If client has not provided a specific max_reason, default
3933		 * to KMSG_DUMP_OOPS, unless always_kmsg_dump was set.
3934		 */
3935		if (max_reason == KMSG_DUMP_UNDEF) {
3936			max_reason = always_kmsg_dump ? KMSG_DUMP_MAX :
3937							KMSG_DUMP_OOPS;
3938		}
3939		if (reason > max_reason)
3940			continue;
3941
3942		/* invoke dumper which will iterate over records */
3943		dumper->dump(dumper, reason);
3944	}
3945	rcu_read_unlock();
3946}
3947
3948/**
3949 * kmsg_dump_get_line - retrieve one kmsg log line
3950 * @iter: kmsg dump iterator
3951 * @syslog: include the "<4>" prefixes
3952 * @line: buffer to copy the line to
3953 * @size: maximum size of the buffer
3954 * @len: length of line placed into buffer
3955 *
3956 * Start at the beginning of the kmsg buffer, with the oldest kmsg
3957 * record, and copy one record into the provided buffer.
3958 *
3959 * Consecutive calls will return the next available record moving
3960 * towards the end of the buffer with the youngest messages.
3961 *
3962 * A return value of FALSE indicates that there are no more records to
3963 * read.
3964 */
3965bool kmsg_dump_get_line(struct kmsg_dump_iter *iter, bool syslog,
3966			char *line, size_t size, size_t *len)
3967{
3968	u64 min_seq = latched_seq_read_nolock(&clear_seq);
3969	struct printk_info info;
3970	unsigned int line_count;
3971	struct printk_record r;
3972	size_t l = 0;
3973	bool ret = false;
3974
3975	if (iter->cur_seq < min_seq)
3976		iter->cur_seq = min_seq;
3977
3978	prb_rec_init_rd(&r, &info, line, size);
3979
3980	/* Read text or count text lines? */
3981	if (line) {
3982		if (!prb_read_valid(prb, iter->cur_seq, &r))
3983			goto out;
3984		l = record_print_text(&r, syslog, printk_time);
3985	} else {
3986		if (!prb_read_valid_info(prb, iter->cur_seq,
3987					 &info, &line_count)) {
3988			goto out;
3989		}
3990		l = get_record_print_text_size(&info, line_count, syslog,
3991					       printk_time);
3992
3993	}
3994
3995	iter->cur_seq = r.info->seq + 1;
3996	ret = true;
3997out:
3998	if (len)
3999		*len = l;
4000	return ret;
4001}
4002EXPORT_SYMBOL_GPL(kmsg_dump_get_line);
4003
4004/**
4005 * kmsg_dump_get_buffer - copy kmsg log lines
4006 * @iter: kmsg dump iterator
4007 * @syslog: include the "<4>" prefixes
4008 * @buf: buffer to copy the line to
4009 * @size: maximum size of the buffer
4010 * @len_out: length of line placed into buffer
4011 *
4012 * Start at the end of the kmsg buffer and fill the provided buffer
4013 * with as many of the *youngest* kmsg records that fit into it.
4014 * If the buffer is large enough, all available kmsg records will be
4015 * copied with a single call.
4016 *
4017 * Consecutive calls will fill the buffer with the next block of
4018 * available older records, not including the earlier retrieved ones.
4019 *
4020 * A return value of FALSE indicates that there are no more records to
4021 * read.
4022 */
4023bool kmsg_dump_get_buffer(struct kmsg_dump_iter *iter, bool syslog,
4024			  char *buf, size_t size, size_t *len_out)
4025{
4026	u64 min_seq = latched_seq_read_nolock(&clear_seq);
4027	struct printk_info info;
4028	struct printk_record r;
4029	u64 seq;
4030	u64 next_seq;
4031	size_t len = 0;
4032	bool ret = false;
4033	bool time = printk_time;
4034
4035	if (!buf || !size)
4036		goto out;
4037
4038	if (iter->cur_seq < min_seq)
4039		iter->cur_seq = min_seq;
4040
4041	if (prb_read_valid_info(prb, iter->cur_seq, &info, NULL)) {
4042		if (info.seq != iter->cur_seq) {
4043			/* messages are gone, move to first available one */
4044			iter->cur_seq = info.seq;
4045		}
4046	}
4047
4048	/* last entry */
4049	if (iter->cur_seq >= iter->next_seq)
4050		goto out;
4051
4052	/*
4053	 * Find first record that fits, including all following records,
4054	 * into the user-provided buffer for this dump. Pass in size-1
4055	 * because this function (by way of record_print_text()) will
4056	 * not write more than size-1 bytes of text into @buf.
4057	 */
4058	seq = find_first_fitting_seq(iter->cur_seq, iter->next_seq,
4059				     size - 1, syslog, time);
4060
4061	/*
4062	 * Next kmsg_dump_get_buffer() invocation will dump block of
4063	 * older records stored right before this one.
4064	 */
4065	next_seq = seq;
4066
4067	prb_rec_init_rd(&r, &info, buf, size);
4068
4069	len = 0;
4070	prb_for_each_record(seq, prb, seq, &r) {
4071		if (r.info->seq >= iter->next_seq)
4072			break;
4073
4074		len += record_print_text(&r, syslog, time);
4075
4076		/* Adjust record to store to remaining buffer space. */
4077		prb_rec_init_rd(&r, &info, buf + len, size - len);
4078	}
4079
4080	iter->next_seq = next_seq;
4081	ret = true;
4082out:
4083	if (len_out)
4084		*len_out = len;
4085	return ret;
4086}
4087EXPORT_SYMBOL_GPL(kmsg_dump_get_buffer);
4088
4089/**
4090 * kmsg_dump_rewind - reset the iterator
4091 * @iter: kmsg dump iterator
4092 *
4093 * Reset the dumper's iterator so that kmsg_dump_get_line() and
4094 * kmsg_dump_get_buffer() can be called again and used multiple
4095 * times within the same dumper.dump() callback.
4096 */
4097void kmsg_dump_rewind(struct kmsg_dump_iter *iter)
4098{
4099	iter->cur_seq = latched_seq_read_nolock(&clear_seq);
4100	iter->next_seq = prb_next_seq(prb);
4101}
4102EXPORT_SYMBOL_GPL(kmsg_dump_rewind);
4103
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
4104#endif
4105
4106#ifdef CONFIG_SMP
4107static atomic_t printk_cpu_sync_owner = ATOMIC_INIT(-1);
4108static atomic_t printk_cpu_sync_nested = ATOMIC_INIT(0);
 
 
 
 
 
4109
4110/**
4111 * __printk_cpu_sync_wait() - Busy wait until the printk cpu-reentrant
4112 *                            spinning lock is not owned by any CPU.
4113 *
4114 * Context: Any context.
4115 */
4116void __printk_cpu_sync_wait(void)
4117{
4118	do {
4119		cpu_relax();
4120	} while (atomic_read(&printk_cpu_sync_owner) != -1);
4121}
4122EXPORT_SYMBOL(__printk_cpu_sync_wait);
4123
4124/**
4125 * __printk_cpu_sync_try_get() - Try to acquire the printk cpu-reentrant
4126 *                               spinning lock.
4127 *
4128 * If no processor has the lock, the calling processor takes the lock and
4129 * becomes the owner. If the calling processor is already the owner of the
4130 * lock, this function succeeds immediately.
4131 *
4132 * Context: Any context. Expects interrupts to be disabled.
4133 * Return: 1 on success, otherwise 0.
4134 */
4135int __printk_cpu_sync_try_get(void)
4136{
4137	int cpu;
4138	int old;
4139
4140	cpu = smp_processor_id();
4141
4142	/*
4143	 * Guarantee loads and stores from this CPU when it is the lock owner
4144	 * are _not_ visible to the previous lock owner. This pairs with
4145	 * __printk_cpu_sync_put:B.
4146	 *
4147	 * Memory barrier involvement:
4148	 *
4149	 * If __printk_cpu_sync_try_get:A reads from __printk_cpu_sync_put:B,
4150	 * then __printk_cpu_sync_put:A can never read from
4151	 * __printk_cpu_sync_try_get:B.
4152	 *
4153	 * Relies on:
4154	 *
4155	 * RELEASE from __printk_cpu_sync_put:A to __printk_cpu_sync_put:B
4156	 * of the previous CPU
4157	 *    matching
4158	 * ACQUIRE from __printk_cpu_sync_try_get:A to
4159	 * __printk_cpu_sync_try_get:B of this CPU
4160	 */
4161	old = atomic_cmpxchg_acquire(&printk_cpu_sync_owner, -1,
4162				     cpu); /* LMM(__printk_cpu_sync_try_get:A) */
4163	if (old == -1) {
4164		/*
4165		 * This CPU is now the owner and begins loading/storing
4166		 * data: LMM(__printk_cpu_sync_try_get:B)
4167		 */
4168		return 1;
4169
4170	} else if (old == cpu) {
4171		/* This CPU is already the owner. */
4172		atomic_inc(&printk_cpu_sync_nested);
4173		return 1;
4174	}
4175
4176	return 0;
4177}
4178EXPORT_SYMBOL(__printk_cpu_sync_try_get);
4179
4180/**
4181 * __printk_cpu_sync_put() - Release the printk cpu-reentrant spinning lock.
4182 *
4183 * The calling processor must be the owner of the lock.
4184 *
4185 * Context: Any context. Expects interrupts to be disabled.
4186 */
4187void __printk_cpu_sync_put(void)
4188{
4189	if (atomic_read(&printk_cpu_sync_nested)) {
4190		atomic_dec(&printk_cpu_sync_nested);
4191		return;
4192	}
4193
4194	/*
4195	 * This CPU is finished loading/storing data:
4196	 * LMM(__printk_cpu_sync_put:A)
4197	 */
4198
4199	/*
4200	 * Guarantee loads and stores from this CPU when it was the
4201	 * lock owner are visible to the next lock owner. This pairs
4202	 * with __printk_cpu_sync_try_get:A.
4203	 *
4204	 * Memory barrier involvement:
4205	 *
4206	 * If __printk_cpu_sync_try_get:A reads from __printk_cpu_sync_put:B,
4207	 * then __printk_cpu_sync_try_get:B reads from __printk_cpu_sync_put:A.
4208	 *
4209	 * Relies on:
4210	 *
4211	 * RELEASE from __printk_cpu_sync_put:A to __printk_cpu_sync_put:B
4212	 * of this CPU
4213	 *    matching
4214	 * ACQUIRE from __printk_cpu_sync_try_get:A to
4215	 * __printk_cpu_sync_try_get:B of the next CPU
4216	 */
4217	atomic_set_release(&printk_cpu_sync_owner,
4218			   -1); /* LMM(__printk_cpu_sync_put:B) */
4219}
4220EXPORT_SYMBOL(__printk_cpu_sync_put);
4221#endif /* CONFIG_SMP */