xref: /linux/kernel/printk/printk.c (revision 8a7f26aa072a4e63bc3b1adfe649546709db722e)
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 #include <linux/panic.h>
52 
53 #include <linux/uaccess.h>
54 #include <asm/sections.h>
55 
56 #include <trace/events/initcall.h>
57 #define CREATE_TRACE_POINTS
58 #include <trace/events/printk.h>
59 
60 #include "printk_ringbuffer.h"
61 #include "console_cmdline.h"
62 #include "braille.h"
63 #include "internal.h"
64 
65 int console_printk[4] = {
66 	CONSOLE_LOGLEVEL_DEFAULT,	/* console_loglevel */
67 	MESSAGE_LOGLEVEL_DEFAULT,	/* default_message_loglevel */
68 	CONSOLE_LOGLEVEL_MIN,		/* minimum_console_loglevel */
69 	CONSOLE_LOGLEVEL_DEFAULT,	/* default_console_loglevel */
70 };
71 EXPORT_SYMBOL_GPL(console_printk);
72 
73 atomic_t ignore_console_lock_warning __read_mostly = ATOMIC_INIT(0);
74 EXPORT_SYMBOL(ignore_console_lock_warning);
75 
76 EXPORT_TRACEPOINT_SYMBOL_GPL(console);
77 
78 /*
79  * Low level drivers may need that to know if they can schedule in
80  * their unblank() callback or not. So let's export it.
81  */
82 int oops_in_progress;
83 EXPORT_SYMBOL(oops_in_progress);
84 
85 /*
86  * console_mutex protects console_list updates and console->flags updates.
87  * The flags are synchronized only for consoles that are registered, i.e.
88  * accessible via the console list.
89  */
90 static DEFINE_MUTEX(console_mutex);
91 
92 /*
93  * console_sem protects updates to console->seq
94  * and also provides serialization for console printing.
95  */
96 static DEFINE_SEMAPHORE(console_sem, 1);
97 HLIST_HEAD(console_list);
98 EXPORT_SYMBOL_GPL(console_list);
99 DEFINE_STATIC_SRCU(console_srcu);
100 
101 /*
102  * System may need to suppress printk message under certain
103  * circumstances, like after kernel panic happens.
104  */
105 int __read_mostly suppress_printk;
106 
107 #ifdef CONFIG_LOCKDEP
108 static struct lockdep_map console_lock_dep_map = {
109 	.name = "console_lock"
110 };
111 
112 void lockdep_assert_console_list_lock_held(void)
113 {
114 	lockdep_assert_held(&console_mutex);
115 }
116 EXPORT_SYMBOL(lockdep_assert_console_list_lock_held);
117 #endif
118 
119 #ifdef CONFIG_DEBUG_LOCK_ALLOC
120 bool console_srcu_read_lock_is_held(void)
121 {
122 	return srcu_read_lock_held(&console_srcu);
123 }
124 EXPORT_SYMBOL(console_srcu_read_lock_is_held);
125 #endif
126 
127 enum devkmsg_log_bits {
128 	__DEVKMSG_LOG_BIT_ON = 0,
129 	__DEVKMSG_LOG_BIT_OFF,
130 	__DEVKMSG_LOG_BIT_LOCK,
131 };
132 
133 enum devkmsg_log_masks {
134 	DEVKMSG_LOG_MASK_ON             = BIT(__DEVKMSG_LOG_BIT_ON),
135 	DEVKMSG_LOG_MASK_OFF            = BIT(__DEVKMSG_LOG_BIT_OFF),
136 	DEVKMSG_LOG_MASK_LOCK           = BIT(__DEVKMSG_LOG_BIT_LOCK),
137 };
138 
139 /* Keep both the 'on' and 'off' bits clear, i.e. ratelimit by default: */
140 #define DEVKMSG_LOG_MASK_DEFAULT	0
141 
142 static unsigned int __read_mostly devkmsg_log = DEVKMSG_LOG_MASK_DEFAULT;
143 
144 static int __control_devkmsg(char *str)
145 {
146 	size_t len;
147 
148 	if (!str)
149 		return -EINVAL;
150 
151 	len = str_has_prefix(str, "on");
152 	if (len) {
153 		devkmsg_log = DEVKMSG_LOG_MASK_ON;
154 		return len;
155 	}
156 
157 	len = str_has_prefix(str, "off");
158 	if (len) {
159 		devkmsg_log = DEVKMSG_LOG_MASK_OFF;
160 		return len;
161 	}
162 
163 	len = str_has_prefix(str, "ratelimit");
164 	if (len) {
165 		devkmsg_log = DEVKMSG_LOG_MASK_DEFAULT;
166 		return len;
167 	}
168 
169 	return -EINVAL;
170 }
171 
172 static int __init control_devkmsg(char *str)
173 {
174 	if (__control_devkmsg(str) < 0) {
175 		pr_warn("printk.devkmsg: bad option string '%s'\n", str);
176 		return 1;
177 	}
178 
179 	/*
180 	 * Set sysctl string accordingly:
181 	 */
182 	if (devkmsg_log == DEVKMSG_LOG_MASK_ON)
183 		strscpy(devkmsg_log_str, "on");
184 	else if (devkmsg_log == DEVKMSG_LOG_MASK_OFF)
185 		strscpy(devkmsg_log_str, "off");
186 	/* else "ratelimit" which is set by default. */
187 
188 	/*
189 	 * Sysctl cannot change it anymore. The kernel command line setting of
190 	 * this parameter is to force the setting to be permanent throughout the
191 	 * runtime of the system. This is a precautionary measure against userspace
192 	 * trying to be a smarta** and attempting to change it up on us.
193 	 */
194 	devkmsg_log |= DEVKMSG_LOG_MASK_LOCK;
195 
196 	return 1;
197 }
198 __setup("printk.devkmsg=", control_devkmsg);
199 
200 char devkmsg_log_str[DEVKMSG_STR_MAX_SIZE] = "ratelimit";
201 #if defined(CONFIG_PRINTK) && defined(CONFIG_SYSCTL)
202 int devkmsg_sysctl_set_loglvl(const struct ctl_table *table, int write,
203 			      void *buffer, size_t *lenp, loff_t *ppos)
204 {
205 	char old_str[DEVKMSG_STR_MAX_SIZE];
206 	unsigned int old;
207 	int err;
208 
209 	if (write) {
210 		if (devkmsg_log & DEVKMSG_LOG_MASK_LOCK)
211 			return -EINVAL;
212 
213 		old = devkmsg_log;
214 		strscpy(old_str, devkmsg_log_str);
215 	}
216 
217 	err = proc_dostring(table, write, buffer, lenp, ppos);
218 	if (err)
219 		return err;
220 
221 	if (write) {
222 		err = __control_devkmsg(devkmsg_log_str);
223 
224 		/*
225 		 * Do not accept an unknown string OR a known string with
226 		 * trailing crap...
227 		 */
228 		if (err < 0 || (err + 1 != *lenp)) {
229 
230 			/* ... and restore old setting. */
231 			devkmsg_log = old;
232 			strscpy(devkmsg_log_str, old_str);
233 
234 			return -EINVAL;
235 		}
236 	}
237 
238 	return 0;
239 }
240 #endif /* CONFIG_PRINTK && CONFIG_SYSCTL */
241 
242 /**
243  * console_list_lock - Lock the console list
244  *
245  * For console list or console->flags updates
246  */
247 void console_list_lock(void)
248 	__acquires(&console_mutex)
249 {
250 	/*
251 	 * In unregister_console() and console_force_preferred_locked(),
252 	 * synchronize_srcu() is called with the console_list_lock held.
253 	 * Therefore it is not allowed that the console_list_lock is taken
254 	 * with the srcu_lock held.
255 	 *
256 	 * Detecting if this context is really in the read-side critical
257 	 * section is only possible if the appropriate debug options are
258 	 * enabled.
259 	 */
260 	WARN_ON_ONCE(debug_lockdep_rcu_enabled() &&
261 		     srcu_read_lock_held(&console_srcu));
262 
263 	mutex_lock(&console_mutex);
264 }
265 EXPORT_SYMBOL(console_list_lock);
266 
267 /**
268  * console_list_unlock - Unlock the console list
269  *
270  * Counterpart to console_list_lock()
271  */
272 void console_list_unlock(void)
273 	__releases(&console_mutex)
274 {
275 	mutex_unlock(&console_mutex);
276 }
277 EXPORT_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  */
288 int console_srcu_read_lock(void)
289 	__acquires(&console_srcu)
290 {
291 	return srcu_read_lock_nmisafe(&console_srcu);
292 }
293 EXPORT_SYMBOL(console_srcu_read_lock);
294 
295 /**
296  * console_srcu_read_unlock - Unregister an old reader from
297  *	the SRCU-protected console list
298  * @cookie: cookie returned from console_srcu_read_lock()
299  *
300  * Counterpart to console_srcu_read_lock()
301  */
302 void console_srcu_read_unlock(int cookie)
303 	__releases(&console_srcu)
304 {
305 	srcu_read_unlock_nmisafe(&console_srcu, cookie);
306 }
307 EXPORT_SYMBOL(console_srcu_read_unlock);
308 
309 /*
310  * Helper macros to handle lockdep when locking/unlocking console_sem. We use
311  * macros instead of functions so that _RET_IP_ contains useful information.
312  */
313 #define down_console_sem() do { \
314 	down(&console_sem);\
315 	mutex_acquire(&console_lock_dep_map, 0, 0, _RET_IP_);\
316 } while (0)
317 
318 static int __down_trylock_console_sem(unsigned long ip)
319 {
320 	int lock_failed;
321 	unsigned long flags;
322 
323 	/*
324 	 * Here and in __up_console_sem() we need to be in safe mode,
325 	 * because spindump/WARN/etc from under console ->lock will
326 	 * deadlock in printk()->down_trylock_console_sem() otherwise.
327 	 */
328 	printk_safe_enter_irqsave(flags);
329 	lock_failed = down_trylock(&console_sem);
330 	printk_safe_exit_irqrestore(flags);
331 
332 	if (lock_failed)
333 		return 1;
334 	mutex_acquire(&console_lock_dep_map, 0, 1, ip);
335 	return 0;
336 }
337 #define down_trylock_console_sem() __down_trylock_console_sem(_RET_IP_)
338 
339 static void __up_console_sem(unsigned long ip)
340 {
341 	unsigned long flags;
342 
343 	mutex_release(&console_lock_dep_map, ip);
344 
345 	printk_safe_enter_irqsave(flags);
346 	up(&console_sem);
347 	printk_safe_exit_irqrestore(flags);
348 }
349 #define up_console_sem() __up_console_sem(_RET_IP_)
350 
351 /*
352  * This is used for debugging the mess that is the VT code by
353  * keeping track if we have the console semaphore held. It's
354  * definitely not the perfect debug tool (we don't know if _WE_
355  * hold it and are racing, but it helps tracking those weird code
356  * paths in the console code where we end up in places I want
357  * locked without the console semaphore held).
358  */
359 static int console_locked;
360 
361 /*
362  *	Array of consoles built from command line options (console=)
363  */
364 
365 #define MAX_CMDLINECONSOLES 8
366 
367 static struct console_cmdline console_cmdline[MAX_CMDLINECONSOLES];
368 
369 static int preferred_console = -1;
370 int console_set_on_cmdline;
371 EXPORT_SYMBOL(console_set_on_cmdline);
372 
373 /* Flag: console code may call schedule() */
374 static int console_may_schedule;
375 
376 enum con_msg_format_flags {
377 	MSG_FORMAT_DEFAULT	= 0,
378 	MSG_FORMAT_SYSLOG	= (1 << 0),
379 };
380 
381 static int console_msg_format = MSG_FORMAT_DEFAULT;
382 
383 /*
384  * The printk log buffer consists of a sequenced collection of records, each
385  * containing variable length message text. Every record also contains its
386  * own meta-data (@info).
387  *
388  * Every record meta-data carries the timestamp in microseconds, as well as
389  * the standard userspace syslog level and syslog facility. The usual kernel
390  * messages use LOG_KERN; userspace-injected messages always carry a matching
391  * syslog facility, by default LOG_USER. The origin of every message can be
392  * reliably determined that way.
393  *
394  * The human readable log message of a record is available in @text, the
395  * length of the message text in @text_len. The stored message is not
396  * terminated.
397  *
398  * Optionally, a record can carry a dictionary of properties (key/value
399  * pairs), to provide userspace with a machine-readable message context.
400  *
401  * Examples for well-defined, commonly used property names are:
402  *   DEVICE=b12:8               device identifier
403  *                                b12:8         block dev_t
404  *                                c127:3        char dev_t
405  *                                n8            netdev ifindex
406  *                                +sound:card0  subsystem:devname
407  *   SUBSYSTEM=pci              driver-core subsystem name
408  *
409  * Valid characters in property names are [a-zA-Z0-9.-_]. Property names
410  * and values are terminated by a '\0' character.
411  *
412  * Example of record values:
413  *   record.text_buf                = "it's a line" (unterminated)
414  *   record.info.seq                = 56
415  *   record.info.ts_nsec            = 36863
416  *   record.info.text_len           = 11
417  *   record.info.facility           = 0 (LOG_KERN)
418  *   record.info.flags              = 0
419  *   record.info.level              = 3 (LOG_ERR)
420  *   record.info.caller_id          = 299 (task 299)
421  *   record.info.dev_info.subsystem = "pci" (terminated)
422  *   record.info.dev_info.device    = "+pci:0000:00:01.0" (terminated)
423  *
424  * The 'struct printk_info' buffer must never be directly exported to
425  * userspace, it is a kernel-private implementation detail that might
426  * need to be changed in the future, when the requirements change.
427  *
428  * /dev/kmsg exports the structured data in the following line format:
429  *   "<level>,<sequnum>,<timestamp>,<contflag>[,additional_values, ... ];<message text>\n"
430  *
431  * Users of the export format should ignore possible additional values
432  * separated by ',', and find the message after the ';' character.
433  *
434  * The optional key/value pairs are attached as continuation lines starting
435  * with a space character and terminated by a newline. All possible
436  * non-prinatable characters are escaped in the "\xff" notation.
437  */
438 
439 /* syslog_lock protects syslog_* variables and write access to clear_seq. */
440 static DEFINE_MUTEX(syslog_lock);
441 
442 /*
443  * Specifies if a legacy console is registered. If legacy consoles are
444  * present, it is necessary to perform the console lock/unlock dance
445  * whenever console flushing should occur.
446  */
447 bool have_legacy_console;
448 
449 /*
450  * Specifies if an nbcon console is registered. If nbcon consoles are present,
451  * synchronous printing of legacy consoles will not occur during panic until
452  * the backtrace has been stored to the ringbuffer.
453  */
454 bool have_nbcon_console;
455 
456 /*
457  * Specifies if a boot console is registered. If boot consoles are present,
458  * nbcon consoles cannot print simultaneously and must be synchronized by
459  * the console lock. This is because boot consoles and nbcon consoles may
460  * have mapped the same hardware.
461  */
462 bool have_boot_console;
463 
464 /* See printk_legacy_allow_panic_sync() for details. */
465 bool legacy_allow_panic_sync;
466 
467 /* Avoid using irq_work when suspending. */
468 bool console_irqwork_blocked;
469 
470 #ifdef CONFIG_PRINTK
471 DECLARE_WAIT_QUEUE_HEAD(log_wait);
472 static DECLARE_WAIT_QUEUE_HEAD(legacy_wait);
473 /* All 3 protected by @syslog_lock. */
474 /* the next printk record to read by syslog(READ) or /proc/kmsg */
475 static u64 syslog_seq;
476 static size_t syslog_partial;
477 static bool syslog_time;
478 
479 /* True when _all_ printer threads are available for printing. */
480 bool printk_kthreads_running;
481 
482 struct latched_seq {
483 	seqcount_latch_t	latch;
484 	u64			val[2];
485 };
486 
487 /*
488  * The next printk record to read after the last 'clear' command. There are
489  * two copies (updated with seqcount_latch) so that reads can locklessly
490  * access a valid value. Writers are synchronized by @syslog_lock.
491  */
492 static struct latched_seq clear_seq = {
493 	.latch		= SEQCNT_LATCH_ZERO(clear_seq.latch),
494 	.val[0]		= 0,
495 	.val[1]		= 0,
496 };
497 
498 #define LOG_LEVEL(v)		((v) & 0x07)
499 #define LOG_FACILITY(v)		((v) >> 3 & 0xff)
500 
501 /* record buffer */
502 #define LOG_ALIGN __alignof__(unsigned long)
503 #define __LOG_BUF_LEN (1 << CONFIG_LOG_BUF_SHIFT)
504 #define LOG_BUF_LEN_MAX ((u32)1 << 31)
505 static char __log_buf[__LOG_BUF_LEN] __aligned(LOG_ALIGN);
506 static char *log_buf = __log_buf;
507 static u32 log_buf_len = __LOG_BUF_LEN;
508 
509 /*
510  * Define the average message size. This only affects the number of
511  * descriptors that will be available. Underestimating is better than
512  * overestimating (too many available descriptors is better than not enough).
513  */
514 #define PRB_AVGBITS 5	/* 32 character average length */
515 
516 #if CONFIG_LOG_BUF_SHIFT <= PRB_AVGBITS
517 #error CONFIG_LOG_BUF_SHIFT value too small.
518 #endif
519 _DEFINE_PRINTKRB(printk_rb_static, CONFIG_LOG_BUF_SHIFT - PRB_AVGBITS,
520 		 PRB_AVGBITS, &__log_buf[0]);
521 
522 static struct printk_ringbuffer printk_rb_dynamic;
523 
524 struct printk_ringbuffer *prb = &printk_rb_static;
525 
526 /*
527  * We cannot access per-CPU data (e.g. per-CPU flush irq_work) before
528  * per_cpu_areas are initialised. This variable is set to true when
529  * it's safe to access per-CPU data.
530  */
531 static bool __printk_percpu_data_ready __ro_after_init;
532 
533 bool printk_percpu_data_ready(void)
534 {
535 	return __printk_percpu_data_ready;
536 }
537 
538 /* Must be called under syslog_lock. */
539 static void latched_seq_write(struct latched_seq *ls, u64 val)
540 {
541 	write_seqcount_latch_begin(&ls->latch);
542 	ls->val[0] = val;
543 	write_seqcount_latch(&ls->latch);
544 	ls->val[1] = val;
545 	write_seqcount_latch_end(&ls->latch);
546 }
547 
548 /* Can be called from any context. */
549 static u64 latched_seq_read_nolock(struct latched_seq *ls)
550 {
551 	unsigned int seq;
552 	unsigned int idx;
553 	u64 val;
554 
555 	do {
556 		seq = read_seqcount_latch(&ls->latch);
557 		idx = seq & 0x1;
558 		val = ls->val[idx];
559 	} while (read_seqcount_latch_retry(&ls->latch, seq));
560 
561 	return val;
562 }
563 
564 /* Return log buffer address */
565 char *log_buf_addr_get(void)
566 {
567 	return log_buf;
568 }
569 
570 /* Return log buffer size */
571 u32 log_buf_len_get(void)
572 {
573 	return log_buf_len;
574 }
575 
576 /*
577  * Define how much of the log buffer we could take at maximum. The value
578  * must be greater than two. Note that only half of the buffer is available
579  * when the index points to the middle.
580  */
581 #define MAX_LOG_TAKE_PART 4
582 static const char trunc_msg[] = "<truncated>";
583 
584 static void truncate_msg(u16 *text_len, u16 *trunc_msg_len)
585 {
586 	/*
587 	 * The message should not take the whole buffer. Otherwise, it might
588 	 * get removed too soon.
589 	 */
590 	u32 max_text_len = log_buf_len / MAX_LOG_TAKE_PART;
591 
592 	if (*text_len > max_text_len)
593 		*text_len = max_text_len;
594 
595 	/* enable the warning message (if there is room) */
596 	*trunc_msg_len = strlen(trunc_msg);
597 	if (*text_len >= *trunc_msg_len)
598 		*text_len -= *trunc_msg_len;
599 	else
600 		*trunc_msg_len = 0;
601 }
602 
603 int dmesg_restrict = IS_ENABLED(CONFIG_SECURITY_DMESG_RESTRICT);
604 
605 static int syslog_action_restricted(int type)
606 {
607 	if (dmesg_restrict)
608 		return 1;
609 	/*
610 	 * Unless restricted, we allow "read all" and "get buffer size"
611 	 * for everybody.
612 	 */
613 	return type != SYSLOG_ACTION_READ_ALL &&
614 	       type != SYSLOG_ACTION_SIZE_BUFFER;
615 }
616 
617 static int check_syslog_permissions(int type, int source)
618 {
619 	/*
620 	 * If this is from /proc/kmsg and we've already opened it, then we've
621 	 * already done the capabilities checks at open time.
622 	 */
623 	if (source == SYSLOG_FROM_PROC && type != SYSLOG_ACTION_OPEN)
624 		goto ok;
625 
626 	if (syslog_action_restricted(type)) {
627 		if (capable(CAP_SYSLOG))
628 			goto ok;
629 		return -EPERM;
630 	}
631 ok:
632 	return security_syslog(type);
633 }
634 
635 static void append_char(char **pp, char *e, char c)
636 {
637 	if (*pp < e)
638 		*(*pp)++ = c;
639 }
640 
641 static ssize_t info_print_ext_header(char *buf, size_t size,
642 				     struct printk_info *info)
643 {
644 	u64 ts_usec = info->ts_nsec;
645 	char caller[20];
646 #ifdef CONFIG_PRINTK_CALLER
647 	u32 id = info->caller_id;
648 
649 	snprintf(caller, sizeof(caller), ",caller=%c%u",
650 		 id & 0x80000000 ? 'C' : 'T', id & ~0x80000000);
651 #else
652 	caller[0] = '\0';
653 #endif
654 
655 	do_div(ts_usec, 1000);
656 
657 	return scnprintf(buf, size, "%u,%llu,%llu,%c%s;",
658 			 (info->facility << 3) | info->level, info->seq,
659 			 ts_usec, info->flags & LOG_CONT ? 'c' : '-', caller);
660 }
661 
662 static ssize_t msg_add_ext_text(char *buf, size_t size,
663 				const char *text, size_t text_len,
664 				unsigned char endc)
665 {
666 	char *p = buf, *e = buf + size;
667 	size_t i;
668 
669 	/* escape non-printable characters */
670 	for (i = 0; i < text_len; i++) {
671 		unsigned char c = text[i];
672 
673 		if (c < ' ' || c >= 127 || c == '\\')
674 			p += scnprintf(p, e - p, "\\x%02x", c);
675 		else
676 			append_char(&p, e, c);
677 	}
678 	append_char(&p, e, endc);
679 
680 	return p - buf;
681 }
682 
683 static ssize_t msg_add_dict_text(char *buf, size_t size,
684 				 const char *key, const char *val)
685 {
686 	size_t val_len = strlen(val);
687 	ssize_t len;
688 
689 	if (!val_len)
690 		return 0;
691 
692 	len = msg_add_ext_text(buf, size, "", 0, ' ');	/* dict prefix */
693 	len += msg_add_ext_text(buf + len, size - len, key, strlen(key), '=');
694 	len += msg_add_ext_text(buf + len, size - len, val, val_len, '\n');
695 
696 	return len;
697 }
698 
699 static ssize_t msg_print_ext_body(char *buf, size_t size,
700 				  char *text, size_t text_len,
701 				  struct dev_printk_info *dev_info)
702 {
703 	ssize_t len;
704 
705 	len = msg_add_ext_text(buf, size, text, text_len, '\n');
706 
707 	if (!dev_info)
708 		goto out;
709 
710 	len += msg_add_dict_text(buf + len, size - len, "SUBSYSTEM",
711 				 dev_info->subsystem);
712 	len += msg_add_dict_text(buf + len, size - len, "DEVICE",
713 				 dev_info->device);
714 out:
715 	return len;
716 }
717 
718 /* /dev/kmsg - userspace message inject/listen interface */
719 struct devkmsg_user {
720 	atomic64_t seq;
721 	struct ratelimit_state rs;
722 	struct mutex lock;
723 	struct printk_buffers pbufs;
724 };
725 
726 static __printf(3, 4) __cold
727 int 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 
739 static 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 (len > PRINTKRB_RECORD_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 
801 static 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 	char *outbuf = &user->pbufs.outbuf[0];
806 	struct printk_message pmsg = {
807 		.pbufs = &user->pbufs,
808 	};
809 	ssize_t ret;
810 
811 	ret = mutex_lock_interruptible(&user->lock);
812 	if (ret)
813 		return ret;
814 
815 	if (!printk_get_next_message(&pmsg, atomic64_read(&user->seq), true, false)) {
816 		if (file->f_flags & O_NONBLOCK) {
817 			ret = -EAGAIN;
818 			goto out;
819 		}
820 
821 		/*
822 		 * Guarantee this task is visible on the waitqueue before
823 		 * checking the wake condition.
824 		 *
825 		 * The full memory barrier within set_current_state() of
826 		 * prepare_to_wait_event() pairs with the full memory barrier
827 		 * within wq_has_sleeper().
828 		 *
829 		 * This pairs with __wake_up_klogd:A.
830 		 */
831 		ret = wait_event_interruptible(log_wait,
832 				printk_get_next_message(&pmsg, atomic64_read(&user->seq), true,
833 							false)); /* LMM(devkmsg_read:A) */
834 		if (ret)
835 			goto out;
836 	}
837 
838 	if (pmsg.dropped) {
839 		/* our last seen message is gone, return error and reset */
840 		atomic64_set(&user->seq, pmsg.seq);
841 		ret = -EPIPE;
842 		goto out;
843 	}
844 
845 	atomic64_set(&user->seq, pmsg.seq + 1);
846 
847 	if (pmsg.outbuf_len > count) {
848 		ret = -EINVAL;
849 		goto out;
850 	}
851 
852 	if (copy_to_user(buf, outbuf, pmsg.outbuf_len)) {
853 		ret = -EFAULT;
854 		goto out;
855 	}
856 	ret = pmsg.outbuf_len;
857 out:
858 	mutex_unlock(&user->lock);
859 	return ret;
860 }
861 
862 /*
863  * Be careful when modifying this function!!!
864  *
865  * Only few operations are supported because the device works only with the
866  * entire variable length messages (records). Non-standard values are
867  * returned in the other cases and has been this way for quite some time.
868  * User space applications might depend on this behavior.
869  */
870 static loff_t devkmsg_llseek(struct file *file, loff_t offset, int whence)
871 {
872 	struct devkmsg_user *user = file->private_data;
873 	loff_t ret = 0;
874 
875 	if (offset)
876 		return -ESPIPE;
877 
878 	switch (whence) {
879 	case SEEK_SET:
880 		/* the first record */
881 		atomic64_set(&user->seq, prb_first_valid_seq(prb));
882 		break;
883 	case SEEK_DATA:
884 		/*
885 		 * The first record after the last SYSLOG_ACTION_CLEAR,
886 		 * like issued by 'dmesg -c'. Reading /dev/kmsg itself
887 		 * changes no global state, and does not clear anything.
888 		 */
889 		atomic64_set(&user->seq, latched_seq_read_nolock(&clear_seq));
890 		break;
891 	case SEEK_END:
892 		/* after the last record */
893 		atomic64_set(&user->seq, prb_next_seq(prb));
894 		break;
895 	default:
896 		ret = -EINVAL;
897 	}
898 	return ret;
899 }
900 
901 static __poll_t devkmsg_poll(struct file *file, poll_table *wait)
902 {
903 	struct devkmsg_user *user = file->private_data;
904 	struct printk_info info;
905 	__poll_t ret = 0;
906 
907 	poll_wait(file, &log_wait, wait);
908 
909 	if (prb_read_valid_info(prb, atomic64_read(&user->seq), &info, NULL)) {
910 		/* return error when data has vanished underneath us */
911 		if (info.seq != atomic64_read(&user->seq))
912 			ret = EPOLLIN|EPOLLRDNORM|EPOLLERR|EPOLLPRI;
913 		else
914 			ret = EPOLLIN|EPOLLRDNORM;
915 	}
916 
917 	return ret;
918 }
919 
920 static int devkmsg_open(struct inode *inode, struct file *file)
921 {
922 	struct devkmsg_user *user;
923 	int err;
924 
925 	if (devkmsg_log & DEVKMSG_LOG_MASK_OFF)
926 		return -EPERM;
927 
928 	/* write-only does not need any file context */
929 	if ((file->f_flags & O_ACCMODE) != O_WRONLY) {
930 		err = check_syslog_permissions(SYSLOG_ACTION_READ_ALL,
931 					       SYSLOG_FROM_READER);
932 		if (err)
933 			return err;
934 	}
935 
936 	user = kvmalloc_obj(struct devkmsg_user);
937 	if (!user)
938 		return -ENOMEM;
939 
940 	ratelimit_default_init(&user->rs);
941 	ratelimit_set_flags(&user->rs, RATELIMIT_MSG_ON_RELEASE);
942 
943 	mutex_init(&user->lock);
944 
945 	atomic64_set(&user->seq, prb_first_valid_seq(prb));
946 
947 	file->private_data = user;
948 	return 0;
949 }
950 
951 static int devkmsg_release(struct inode *inode, struct file *file)
952 {
953 	struct devkmsg_user *user = file->private_data;
954 
955 	ratelimit_state_exit(&user->rs);
956 
957 	mutex_destroy(&user->lock);
958 	kvfree(user);
959 	return 0;
960 }
961 
962 const struct file_operations kmsg_fops = {
963 	.open = devkmsg_open,
964 	.read = devkmsg_read,
965 	.write_iter = devkmsg_write,
966 	.llseek = devkmsg_llseek,
967 	.poll = devkmsg_poll,
968 	.release = devkmsg_release,
969 };
970 
971 #ifdef CONFIG_VMCORE_INFO
972 /*
973  * This appends the listed symbols to /proc/vmcore
974  *
975  * /proc/vmcore is used by various utilities, like crash and makedumpfile to
976  * obtain access to symbols that are otherwise very difficult to locate.  These
977  * symbols are specifically used so that utilities can access and extract the
978  * dmesg log from a vmcore file after a crash.
979  */
980 void log_buf_vmcoreinfo_setup(void)
981 {
982 	struct dev_printk_info *dev_info = NULL;
983 
984 	VMCOREINFO_SYMBOL(prb);
985 	VMCOREINFO_SYMBOL(printk_rb_static);
986 	VMCOREINFO_SYMBOL(clear_seq);
987 
988 	/*
989 	 * Export struct size and field offsets. User space tools can
990 	 * parse it and detect any changes to structure down the line.
991 	 */
992 
993 	VMCOREINFO_STRUCT_SIZE(printk_ringbuffer);
994 	VMCOREINFO_OFFSET(printk_ringbuffer, desc_ring);
995 	VMCOREINFO_OFFSET(printk_ringbuffer, text_data_ring);
996 	VMCOREINFO_OFFSET(printk_ringbuffer, fail);
997 
998 	VMCOREINFO_STRUCT_SIZE(prb_desc_ring);
999 	VMCOREINFO_OFFSET(prb_desc_ring, count_bits);
1000 	VMCOREINFO_OFFSET(prb_desc_ring, descs);
1001 	VMCOREINFO_OFFSET(prb_desc_ring, infos);
1002 	VMCOREINFO_OFFSET(prb_desc_ring, head_id);
1003 	VMCOREINFO_OFFSET(prb_desc_ring, tail_id);
1004 
1005 	VMCOREINFO_STRUCT_SIZE(prb_desc);
1006 	VMCOREINFO_OFFSET(prb_desc, state_var);
1007 	VMCOREINFO_OFFSET(prb_desc, text_blk_lpos);
1008 
1009 	VMCOREINFO_STRUCT_SIZE(prb_data_blk_lpos);
1010 	VMCOREINFO_OFFSET(prb_data_blk_lpos, begin);
1011 	VMCOREINFO_OFFSET(prb_data_blk_lpos, next);
1012 
1013 	VMCOREINFO_STRUCT_SIZE(printk_info);
1014 	VMCOREINFO_OFFSET(printk_info, seq);
1015 	VMCOREINFO_OFFSET(printk_info, ts_nsec);
1016 	VMCOREINFO_OFFSET(printk_info, text_len);
1017 	VMCOREINFO_OFFSET(printk_info, caller_id);
1018 	VMCOREINFO_OFFSET(printk_info, dev_info);
1019 
1020 	VMCOREINFO_STRUCT_SIZE(dev_printk_info);
1021 	VMCOREINFO_OFFSET(dev_printk_info, subsystem);
1022 	VMCOREINFO_LENGTH(printk_info_subsystem, sizeof(dev_info->subsystem));
1023 	VMCOREINFO_OFFSET(dev_printk_info, device);
1024 	VMCOREINFO_LENGTH(printk_info_device, sizeof(dev_info->device));
1025 
1026 	VMCOREINFO_STRUCT_SIZE(prb_data_ring);
1027 	VMCOREINFO_OFFSET(prb_data_ring, size_bits);
1028 	VMCOREINFO_OFFSET(prb_data_ring, data);
1029 	VMCOREINFO_OFFSET(prb_data_ring, head_lpos);
1030 	VMCOREINFO_OFFSET(prb_data_ring, tail_lpos);
1031 
1032 	VMCOREINFO_SIZE(atomic_long_t);
1033 	VMCOREINFO_TYPE_OFFSET(atomic_long_t, counter);
1034 
1035 	VMCOREINFO_STRUCT_SIZE(latched_seq);
1036 	VMCOREINFO_OFFSET(latched_seq, val);
1037 }
1038 #endif
1039 
1040 /* requested log_buf_len from kernel cmdline */
1041 static unsigned long __initdata new_log_buf_len;
1042 
1043 /* we practice scaling the ring buffer by powers of 2 */
1044 static void __init log_buf_len_update(u64 size)
1045 {
1046 	if (size > (u64)LOG_BUF_LEN_MAX) {
1047 		size = (u64)LOG_BUF_LEN_MAX;
1048 		pr_err("log_buf over 2G is not supported.\n");
1049 	}
1050 
1051 	if (size)
1052 		size = roundup_pow_of_two(size);
1053 	if (size > log_buf_len)
1054 		new_log_buf_len = (unsigned long)size;
1055 }
1056 
1057 /* save requested log_buf_len since it's too early to process it */
1058 static int __init log_buf_len_setup(char *str)
1059 {
1060 	u64 size;
1061 
1062 	if (!str)
1063 		return -EINVAL;
1064 
1065 	size = memparse(str, &str);
1066 
1067 	log_buf_len_update(size);
1068 
1069 	return 0;
1070 }
1071 early_param("log_buf_len", log_buf_len_setup);
1072 
1073 #ifdef CONFIG_SMP
1074 #define __LOG_CPU_MAX_BUF_LEN (1 << CONFIG_LOG_CPU_MAX_BUF_SHIFT)
1075 
1076 static void __init log_buf_add_cpu(void)
1077 {
1078 	unsigned int cpu_extra;
1079 
1080 	/*
1081 	 * archs should set up cpu_possible_bits properly with
1082 	 * set_cpu_possible() after setup_arch() but just in
1083 	 * case lets ensure this is valid.
1084 	 */
1085 	if (num_possible_cpus() == 1)
1086 		return;
1087 
1088 	cpu_extra = num_possible_cpus() * __LOG_CPU_MAX_BUF_LEN;
1089 
1090 	/* by default this will only continue through for more than 16 CPUs */
1091 	if (cpu_extra <= __LOG_BUF_LEN / 2)
1092 		return;
1093 
1094 	pr_info("log_buf_len individual max cpu contribution: %d bytes\n",
1095 		__LOG_CPU_MAX_BUF_LEN);
1096 	pr_info("log_buf_len total cpu_extra contributions: %d bytes\n",
1097 		cpu_extra);
1098 	pr_info("log_buf_len min size: %d bytes\n", __LOG_BUF_LEN);
1099 
1100 	log_buf_len_update(cpu_extra + __LOG_BUF_LEN);
1101 }
1102 #else /* !CONFIG_SMP */
1103 static inline void log_buf_add_cpu(void) {}
1104 #endif /* CONFIG_SMP */
1105 
1106 static void __init set_percpu_data_ready(void)
1107 {
1108 	__printk_percpu_data_ready = true;
1109 }
1110 
1111 static unsigned int __init add_to_rb(struct printk_ringbuffer *rb,
1112 				     struct printk_record *r)
1113 {
1114 	struct prb_reserved_entry e;
1115 	struct printk_record dest_r;
1116 
1117 	prb_rec_init_wr(&dest_r, r->info->text_len);
1118 
1119 	if (!prb_reserve(&e, rb, &dest_r))
1120 		return 0;
1121 
1122 	memcpy(&dest_r.text_buf[0], &r->text_buf[0], r->info->text_len);
1123 	dest_r.info->text_len = r->info->text_len;
1124 	dest_r.info->facility = r->info->facility;
1125 	dest_r.info->level = r->info->level;
1126 	dest_r.info->flags = r->info->flags;
1127 	dest_r.info->ts_nsec = r->info->ts_nsec;
1128 	dest_r.info->caller_id = r->info->caller_id;
1129 	memcpy(&dest_r.info->dev_info, &r->info->dev_info, sizeof(dest_r.info->dev_info));
1130 
1131 	prb_final_commit(&e);
1132 
1133 	return prb_record_text_space(&e);
1134 }
1135 
1136 static char setup_text_buf[PRINTKRB_RECORD_MAX] __initdata;
1137 
1138 static void print_log_buf_usage_stats(void)
1139 {
1140 	unsigned int descs_count = log_buf_len >> PRB_AVGBITS;
1141 	size_t meta_data_size;
1142 
1143 	meta_data_size = descs_count * (sizeof(struct prb_desc) + sizeof(struct printk_info));
1144 
1145 	pr_info("log buffer data + meta data: %u + %zu = %zu bytes\n",
1146 		log_buf_len, meta_data_size, log_buf_len + meta_data_size);
1147 }
1148 
1149 void __init setup_log_buf(int early)
1150 {
1151 	struct printk_info *new_infos;
1152 	unsigned int new_descs_count;
1153 	struct prb_desc *new_descs;
1154 	struct printk_info info;
1155 	struct printk_record r;
1156 	unsigned int text_size;
1157 	size_t new_descs_size;
1158 	size_t new_infos_size;
1159 	unsigned long flags;
1160 	char *new_log_buf;
1161 	unsigned int free;
1162 	u64 seq;
1163 
1164 	/*
1165 	 * Some archs call setup_log_buf() multiple times - first is very
1166 	 * early, e.g. from setup_arch(), and second - when percpu_areas
1167 	 * are initialised.
1168 	 */
1169 	if (!early)
1170 		set_percpu_data_ready();
1171 
1172 	if (log_buf != __log_buf)
1173 		return;
1174 
1175 	if (!early && !new_log_buf_len)
1176 		log_buf_add_cpu();
1177 
1178 	if (!new_log_buf_len) {
1179 		/* Show the memory stats only once. */
1180 		if (!early)
1181 			goto out;
1182 
1183 		return;
1184 	}
1185 
1186 	new_descs_count = new_log_buf_len >> PRB_AVGBITS;
1187 	if (new_descs_count == 0) {
1188 		pr_err("new_log_buf_len: %lu too small\n", new_log_buf_len);
1189 		goto out;
1190 	}
1191 
1192 	new_log_buf = memblock_alloc(new_log_buf_len, LOG_ALIGN);
1193 	if (unlikely(!new_log_buf)) {
1194 		pr_err("log_buf_len: %lu text bytes not available\n",
1195 		       new_log_buf_len);
1196 		goto out;
1197 	}
1198 
1199 	new_descs_size = new_descs_count * sizeof(struct prb_desc);
1200 	new_descs = memblock_alloc(new_descs_size, LOG_ALIGN);
1201 	if (unlikely(!new_descs)) {
1202 		pr_err("log_buf_len: %zu desc bytes not available\n",
1203 		       new_descs_size);
1204 		goto err_free_log_buf;
1205 	}
1206 
1207 	new_infos_size = new_descs_count * sizeof(struct printk_info);
1208 	new_infos = memblock_alloc(new_infos_size, LOG_ALIGN);
1209 	if (unlikely(!new_infos)) {
1210 		pr_err("log_buf_len: %zu info bytes not available\n",
1211 		       new_infos_size);
1212 		goto err_free_descs;
1213 	}
1214 
1215 	prb_rec_init_rd(&r, &info, &setup_text_buf[0], sizeof(setup_text_buf));
1216 
1217 	prb_init(&printk_rb_dynamic,
1218 		 new_log_buf, ilog2(new_log_buf_len),
1219 		 new_descs, ilog2(new_descs_count),
1220 		 new_infos);
1221 
1222 	local_irq_save(flags);
1223 
1224 	log_buf_len = new_log_buf_len;
1225 	log_buf = new_log_buf;
1226 	new_log_buf_len = 0;
1227 
1228 	free = __LOG_BUF_LEN;
1229 	prb_for_each_record(0, &printk_rb_static, seq, &r) {
1230 		text_size = add_to_rb(&printk_rb_dynamic, &r);
1231 		if (text_size > free)
1232 			free = 0;
1233 		else
1234 			free -= text_size;
1235 	}
1236 
1237 	prb = &printk_rb_dynamic;
1238 
1239 	local_irq_restore(flags);
1240 
1241 	/*
1242 	 * Copy any remaining messages that might have appeared from
1243 	 * NMI context after copying but before switching to the
1244 	 * dynamic buffer.
1245 	 */
1246 	prb_for_each_record(seq, &printk_rb_static, seq, &r) {
1247 		text_size = add_to_rb(&printk_rb_dynamic, &r);
1248 		if (text_size > free)
1249 			free = 0;
1250 		else
1251 			free -= text_size;
1252 	}
1253 
1254 	if (seq != prb_next_seq(&printk_rb_static)) {
1255 		pr_err("dropped %llu messages\n",
1256 		       prb_next_seq(&printk_rb_static) - seq);
1257 	}
1258 
1259 	print_log_buf_usage_stats();
1260 	pr_info("early log buf free: %u(%u%%)\n",
1261 		free, (free * 100) / __LOG_BUF_LEN);
1262 	return;
1263 
1264 err_free_descs:
1265 	memblock_free(new_descs, new_descs_size);
1266 err_free_log_buf:
1267 	memblock_free(new_log_buf, new_log_buf_len);
1268 out:
1269 	print_log_buf_usage_stats();
1270 }
1271 
1272 static bool __read_mostly ignore_loglevel;
1273 
1274 static int __init ignore_loglevel_setup(char *str)
1275 {
1276 	ignore_loglevel = true;
1277 	pr_info("debug: ignoring loglevel setting.\n");
1278 
1279 	return 0;
1280 }
1281 
1282 early_param("ignore_loglevel", ignore_loglevel_setup);
1283 module_param(ignore_loglevel, bool, S_IRUGO | S_IWUSR);
1284 MODULE_PARM_DESC(ignore_loglevel,
1285 		 "ignore loglevel setting (prints all kernel messages to the console)");
1286 
1287 static bool suppress_message_printing(int level)
1288 {
1289 	return (level >= console_loglevel && !ignore_loglevel);
1290 }
1291 
1292 #ifdef CONFIG_BOOT_PRINTK_DELAY
1293 
1294 static int boot_delay; /* msecs delay after each printk during bootup */
1295 static unsigned long long loops_per_msec;	/* based on boot_delay */
1296 
1297 static int __init boot_delay_setup(char *str)
1298 {
1299 	unsigned long lpj;
1300 
1301 	lpj = preset_lpj ? preset_lpj : 1000000;	/* some guess */
1302 	loops_per_msec = (unsigned long long)lpj / 1000 * HZ;
1303 
1304 	get_option(&str, &boot_delay);
1305 	if (boot_delay > 10 * 1000)
1306 		boot_delay = 0;
1307 
1308 	pr_debug("boot_delay: %u, preset_lpj: %ld, lpj: %lu, "
1309 		"HZ: %d, loops_per_msec: %llu\n",
1310 		boot_delay, preset_lpj, lpj, HZ, loops_per_msec);
1311 	return 0;
1312 }
1313 early_param("boot_delay", boot_delay_setup);
1314 
1315 static void boot_delay_msec(int level)
1316 {
1317 	unsigned long long k;
1318 	unsigned long timeout;
1319 	bool suppress = !is_printk_force_console() &&
1320 			suppress_message_printing(level);
1321 
1322 	if ((boot_delay == 0 || system_state >= SYSTEM_RUNNING) || suppress)
1323 		return;
1324 
1325 	k = (unsigned long long)loops_per_msec * boot_delay;
1326 
1327 	timeout = jiffies + msecs_to_jiffies(boot_delay);
1328 	while (k) {
1329 		k--;
1330 		cpu_relax();
1331 		/*
1332 		 * use (volatile) jiffies to prevent
1333 		 * compiler reduction; loop termination via jiffies
1334 		 * is secondary and may or may not happen.
1335 		 */
1336 		if (time_after(jiffies, timeout))
1337 			break;
1338 		touch_nmi_watchdog();
1339 	}
1340 }
1341 #else
1342 static inline void boot_delay_msec(int level)
1343 {
1344 }
1345 #endif
1346 
1347 static bool printk_time = IS_ENABLED(CONFIG_PRINTK_TIME);
1348 module_param_named(time, printk_time, bool, S_IRUGO | S_IWUSR);
1349 
1350 static size_t print_syslog(unsigned int level, char *buf)
1351 {
1352 	return sprintf(buf, "<%u>", level);
1353 }
1354 
1355 static size_t print_time(u64 ts, char *buf)
1356 {
1357 	unsigned long rem_nsec = do_div(ts, 1000000000);
1358 
1359 	return sprintf(buf, "[%5lu.%06lu]",
1360 		       (unsigned long)ts, rem_nsec / 1000);
1361 }
1362 
1363 #ifdef CONFIG_PRINTK_CALLER
1364 static size_t print_caller(u32 id, char *buf)
1365 {
1366 	char caller[12];
1367 
1368 	snprintf(caller, sizeof(caller), "%c%u",
1369 		 id & 0x80000000 ? 'C' : 'T', id & ~0x80000000);
1370 	return sprintf(buf, "[%6s]", caller);
1371 }
1372 #else
1373 #define print_caller(id, buf) 0
1374 #endif
1375 
1376 static size_t info_print_prefix(const struct printk_info  *info, bool syslog,
1377 				bool time, char *buf)
1378 {
1379 	size_t len = 0;
1380 
1381 	if (syslog)
1382 		len = print_syslog((info->facility << 3) | info->level, buf);
1383 
1384 	if (time)
1385 		len += print_time(info->ts_nsec, buf + len);
1386 
1387 	len += print_caller(info->caller_id, buf + len);
1388 
1389 	if (IS_ENABLED(CONFIG_PRINTK_CALLER) || time) {
1390 		buf[len++] = ' ';
1391 		buf[len] = '\0';
1392 	}
1393 
1394 	return len;
1395 }
1396 
1397 /*
1398  * Prepare the record for printing. The text is shifted within the given
1399  * buffer to avoid a need for another one. The following operations are
1400  * done:
1401  *
1402  *   - Add prefix for each line.
1403  *   - Drop truncated lines that no longer fit into the buffer.
1404  *   - Add the trailing newline that has been removed in vprintk_store().
1405  *   - Add a string terminator.
1406  *
1407  * Since the produced string is always terminated, the maximum possible
1408  * return value is @r->text_buf_size - 1;
1409  *
1410  * Return: The length of the updated/prepared text, including the added
1411  * prefixes and the newline. The terminator is not counted. The dropped
1412  * line(s) are not counted.
1413  */
1414 static size_t record_print_text(struct printk_record *r, bool syslog,
1415 				bool time)
1416 {
1417 	size_t text_len = r->info->text_len;
1418 	size_t buf_size = r->text_buf_size;
1419 	char *text = r->text_buf;
1420 	char prefix[PRINTK_PREFIX_MAX];
1421 	bool truncated = false;
1422 	size_t prefix_len;
1423 	size_t line_len;
1424 	size_t len = 0;
1425 	char *next;
1426 
1427 	/*
1428 	 * If the message was truncated because the buffer was not large
1429 	 * enough, treat the available text as if it were the full text.
1430 	 */
1431 	if (text_len > buf_size)
1432 		text_len = buf_size;
1433 
1434 	prefix_len = info_print_prefix(r->info, syslog, time, prefix);
1435 
1436 	/*
1437 	 * @text_len: bytes of unprocessed text
1438 	 * @line_len: bytes of current line _without_ newline
1439 	 * @text:     pointer to beginning of current line
1440 	 * @len:      number of bytes prepared in r->text_buf
1441 	 */
1442 	for (;;) {
1443 		next = memchr(text, '\n', text_len);
1444 		if (next) {
1445 			line_len = next - text;
1446 		} else {
1447 			/* Drop truncated line(s). */
1448 			if (truncated)
1449 				break;
1450 			line_len = text_len;
1451 		}
1452 
1453 		/*
1454 		 * Truncate the text if there is not enough space to add the
1455 		 * prefix and a trailing newline and a terminator.
1456 		 */
1457 		if (len + prefix_len + text_len + 1 + 1 > buf_size) {
1458 			/* Drop even the current line if no space. */
1459 			if (len + prefix_len + line_len + 1 + 1 > buf_size)
1460 				break;
1461 
1462 			text_len = buf_size - len - prefix_len - 1 - 1;
1463 			truncated = true;
1464 		}
1465 
1466 		memmove(text + prefix_len, text, text_len);
1467 		memcpy(text, prefix, prefix_len);
1468 
1469 		/*
1470 		 * Increment the prepared length to include the text and
1471 		 * prefix that were just moved+copied. Also increment for the
1472 		 * newline at the end of this line. If this is the last line,
1473 		 * there is no newline, but it will be added immediately below.
1474 		 */
1475 		len += prefix_len + line_len + 1;
1476 		if (text_len == line_len) {
1477 			/*
1478 			 * This is the last line. Add the trailing newline
1479 			 * removed in vprintk_store().
1480 			 */
1481 			text[prefix_len + line_len] = '\n';
1482 			break;
1483 		}
1484 
1485 		/*
1486 		 * Advance beyond the added prefix and the related line with
1487 		 * its newline.
1488 		 */
1489 		text += prefix_len + line_len + 1;
1490 
1491 		/*
1492 		 * The remaining text has only decreased by the line with its
1493 		 * newline.
1494 		 *
1495 		 * Note that @text_len can become zero. It happens when @text
1496 		 * ended with a newline (either due to truncation or the
1497 		 * original string ending with "\n\n"). The loop is correctly
1498 		 * repeated and (if not truncated) an empty line with a prefix
1499 		 * will be prepared.
1500 		 */
1501 		text_len -= line_len + 1;
1502 	}
1503 
1504 	/*
1505 	 * If a buffer was provided, it will be terminated. Space for the
1506 	 * string terminator is guaranteed to be available. The terminator is
1507 	 * not counted in the return value.
1508 	 */
1509 	if (buf_size > 0)
1510 		r->text_buf[len] = 0;
1511 
1512 	return len;
1513 }
1514 
1515 static size_t get_record_print_text_size(struct printk_info *info,
1516 					 unsigned int line_count,
1517 					 bool syslog, bool time)
1518 {
1519 	char prefix[PRINTK_PREFIX_MAX];
1520 	size_t prefix_len;
1521 
1522 	prefix_len = info_print_prefix(info, syslog, time, prefix);
1523 
1524 	/*
1525 	 * Each line will be preceded with a prefix. The intermediate
1526 	 * newlines are already within the text, but a final trailing
1527 	 * newline will be added.
1528 	 */
1529 	return ((prefix_len * line_count) + info->text_len + 1);
1530 }
1531 
1532 /*
1533  * Beginning with @start_seq, find the first record where it and all following
1534  * records up to (but not including) @max_seq fit into @size.
1535  *
1536  * @max_seq is simply an upper bound and does not need to exist. If the caller
1537  * does not require an upper bound, -1 can be used for @max_seq.
1538  */
1539 static u64 find_first_fitting_seq(u64 start_seq, u64 max_seq, size_t size,
1540 				  bool syslog, bool time)
1541 {
1542 	struct printk_info info;
1543 	unsigned int line_count;
1544 	size_t len = 0;
1545 	u64 seq;
1546 
1547 	/* Determine the size of the records up to @max_seq. */
1548 	prb_for_each_info(start_seq, prb, seq, &info, &line_count) {
1549 		if (info.seq >= max_seq)
1550 			break;
1551 		len += get_record_print_text_size(&info, line_count, syslog, time);
1552 	}
1553 
1554 	/*
1555 	 * Adjust the upper bound for the next loop to avoid subtracting
1556 	 * lengths that were never added.
1557 	 */
1558 	if (seq < max_seq)
1559 		max_seq = seq;
1560 
1561 	/*
1562 	 * Move first record forward until length fits into the buffer. Ignore
1563 	 * newest messages that were not counted in the above cycle. Messages
1564 	 * might appear and get lost in the meantime. This is a best effort
1565 	 * that prevents an infinite loop that could occur with a retry.
1566 	 */
1567 	prb_for_each_info(start_seq, prb, seq, &info, &line_count) {
1568 		if (len <= size || info.seq >= max_seq)
1569 			break;
1570 		len -= get_record_print_text_size(&info, line_count, syslog, time);
1571 	}
1572 
1573 	return seq;
1574 }
1575 
1576 /* The caller is responsible for making sure @size is greater than 0. */
1577 static int syslog_print(char __user *buf, int size)
1578 {
1579 	struct printk_info info;
1580 	struct printk_record r;
1581 	char *text;
1582 	int len = 0;
1583 	u64 seq;
1584 
1585 	text = kmalloc(PRINTK_MESSAGE_MAX, GFP_KERNEL);
1586 	if (!text)
1587 		return -ENOMEM;
1588 
1589 	prb_rec_init_rd(&r, &info, text, PRINTK_MESSAGE_MAX);
1590 
1591 	mutex_lock(&syslog_lock);
1592 
1593 	/*
1594 	 * Wait for the @syslog_seq record to be available. @syslog_seq may
1595 	 * change while waiting.
1596 	 */
1597 	do {
1598 		seq = syslog_seq;
1599 
1600 		mutex_unlock(&syslog_lock);
1601 		/*
1602 		 * Guarantee this task is visible on the waitqueue before
1603 		 * checking the wake condition.
1604 		 *
1605 		 * The full memory barrier within set_current_state() of
1606 		 * prepare_to_wait_event() pairs with the full memory barrier
1607 		 * within wq_has_sleeper().
1608 		 *
1609 		 * This pairs with __wake_up_klogd:A.
1610 		 */
1611 		len = wait_event_interruptible(log_wait,
1612 				prb_read_valid(prb, seq, NULL)); /* LMM(syslog_print:A) */
1613 		mutex_lock(&syslog_lock);
1614 
1615 		if (len)
1616 			goto out;
1617 	} while (syslog_seq != seq);
1618 
1619 	/*
1620 	 * Copy records that fit into the buffer. The above cycle makes sure
1621 	 * that the first record is always available.
1622 	 */
1623 	do {
1624 		size_t n;
1625 		size_t skip;
1626 		int err;
1627 
1628 		if (!prb_read_valid(prb, syslog_seq, &r))
1629 			break;
1630 
1631 		if (r.info->seq != syslog_seq) {
1632 			/* message is gone, move to next valid one */
1633 			syslog_seq = r.info->seq;
1634 			syslog_partial = 0;
1635 		}
1636 
1637 		/*
1638 		 * To keep reading/counting partial line consistent,
1639 		 * use printk_time value as of the beginning of a line.
1640 		 */
1641 		if (!syslog_partial)
1642 			syslog_time = printk_time;
1643 
1644 		skip = syslog_partial;
1645 		n = record_print_text(&r, true, syslog_time);
1646 		if (n - syslog_partial <= size) {
1647 			/* message fits into buffer, move forward */
1648 			syslog_seq = r.info->seq + 1;
1649 			n -= syslog_partial;
1650 			syslog_partial = 0;
1651 		} else if (!len){
1652 			/* partial read(), remember position */
1653 			n = size;
1654 			syslog_partial += n;
1655 		} else
1656 			n = 0;
1657 
1658 		if (!n)
1659 			break;
1660 
1661 		mutex_unlock(&syslog_lock);
1662 		err = copy_to_user(buf, text + skip, n);
1663 		mutex_lock(&syslog_lock);
1664 
1665 		if (err) {
1666 			if (!len)
1667 				len = -EFAULT;
1668 			break;
1669 		}
1670 
1671 		len += n;
1672 		size -= n;
1673 		buf += n;
1674 	} while (size);
1675 out:
1676 	mutex_unlock(&syslog_lock);
1677 	kfree(text);
1678 	return len;
1679 }
1680 
1681 static int syslog_print_all(char __user *buf, int size, bool clear)
1682 {
1683 	struct printk_info info;
1684 	struct printk_record r;
1685 	char *text;
1686 	int len = 0;
1687 	u64 seq;
1688 	bool time;
1689 
1690 	text = kmalloc(PRINTK_MESSAGE_MAX, GFP_KERNEL);
1691 	if (!text)
1692 		return -ENOMEM;
1693 
1694 	time = printk_time;
1695 	/*
1696 	 * Find first record that fits, including all following records,
1697 	 * into the user-provided buffer for this dump.
1698 	 */
1699 	seq = find_first_fitting_seq(latched_seq_read_nolock(&clear_seq), -1,
1700 				     size, true, time);
1701 
1702 	prb_rec_init_rd(&r, &info, text, PRINTK_MESSAGE_MAX);
1703 
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 
1733 static 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 
1740 int 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 
1853 SYSCALL_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
1864 static struct lockdep_map console_owner_dep_map = {
1865 	.name = "console_owner"
1866 };
1867 #endif
1868 
1869 static DEFINE_RAW_SPINLOCK(console_owner_lock);
1870 static struct task_struct *console_owner;
1871 static 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  */
1882 void console_lock_spinning_enable(void)
1883 {
1884 	/*
1885 	 * Do not use spinning in panic(). The panic CPU wants to keep the lock.
1886 	 * Non-panic CPUs abandon the flush anyway.
1887 	 *
1888 	 * Just keep the lockdep annotation. The panic-CPU should avoid
1889 	 * taking console_owner_lock because it might cause a deadlock.
1890 	 * This looks like the easiest way how to prevent false lockdep
1891 	 * reports without handling races a lockless way.
1892 	 */
1893 	if (panic_in_progress())
1894 		goto lockdep;
1895 
1896 	raw_spin_lock(&console_owner_lock);
1897 	console_owner = current;
1898 	raw_spin_unlock(&console_owner_lock);
1899 
1900 lockdep:
1901 	/* The waiter may spin on us after setting console_owner */
1902 	spin_acquire(&console_owner_dep_map, 0, 0, _THIS_IP_);
1903 }
1904 
1905 /**
1906  * console_lock_spinning_disable_and_check - mark end of code where another
1907  *	thread was able to busy wait and check if there is a waiter
1908  * @cookie: cookie returned from console_srcu_read_lock()
1909  *
1910  * This is called at the end of the section where spinning is allowed.
1911  * It has two functions. First, it is a signal that it is no longer
1912  * safe to start busy waiting for the lock. Second, it checks if
1913  * there is a busy waiter and passes the lock rights to her.
1914  *
1915  * Important: Callers lose both the console_lock and the SRCU read lock if
1916  *	there was a busy waiter. They must not touch items synchronized by
1917  *	console_lock or SRCU read lock in this case.
1918  *
1919  * Return: 1 if the lock rights were passed, 0 otherwise.
1920  */
1921 int console_lock_spinning_disable_and_check(int cookie)
1922 {
1923 	int waiter;
1924 
1925 	/*
1926 	 * Ignore spinning waiters during panic() because they might get stopped
1927 	 * or blocked at any time,
1928 	 *
1929 	 * It is safe because nobody is allowed to start spinning during panic
1930 	 * in the first place. If there has been a waiter then non panic CPUs
1931 	 * might stay spinning. They would get stopped anyway. The panic context
1932 	 * will never start spinning and an interrupted spin on panic CPU will
1933 	 * never continue.
1934 	 */
1935 	if (panic_in_progress()) {
1936 		/* Keep lockdep happy. */
1937 		spin_release(&console_owner_dep_map, _THIS_IP_);
1938 		return 0;
1939 	}
1940 
1941 	raw_spin_lock(&console_owner_lock);
1942 	waiter = READ_ONCE(console_waiter);
1943 	console_owner = NULL;
1944 	raw_spin_unlock(&console_owner_lock);
1945 
1946 	if (!waiter) {
1947 		spin_release(&console_owner_dep_map, _THIS_IP_);
1948 		return 0;
1949 	}
1950 
1951 	/* The waiter is now free to continue */
1952 	WRITE_ONCE(console_waiter, false);
1953 
1954 	spin_release(&console_owner_dep_map, _THIS_IP_);
1955 
1956 	/*
1957 	 * Preserve lockdep lock ordering. Release the SRCU read lock before
1958 	 * releasing the console_lock.
1959 	 */
1960 	console_srcu_read_unlock(cookie);
1961 
1962 	/*
1963 	 * Hand off console_lock to waiter. The waiter will perform
1964 	 * the up(). After this, the waiter is the console_lock owner.
1965 	 */
1966 	mutex_release(&console_lock_dep_map, _THIS_IP_);
1967 	return 1;
1968 }
1969 
1970 /**
1971  * console_trylock_spinning - try to get console_lock by busy waiting
1972  *
1973  * This allows to busy wait for the console_lock when the current
1974  * owner is running in specially marked sections. It means that
1975  * the current owner is running and cannot reschedule until it
1976  * is ready to lose the lock.
1977  *
1978  * Return: 1 if we got the lock, 0 otherwise
1979  */
1980 static int console_trylock_spinning(void)
1981 {
1982 	struct task_struct *owner = NULL;
1983 	bool waiter;
1984 	bool spin = false;
1985 	unsigned long flags;
1986 
1987 	if (console_trylock())
1988 		return 1;
1989 
1990 	/*
1991 	 * It's unsafe to spin once a panic has begun. If we are the
1992 	 * panic CPU, we may have already halted the owner of the
1993 	 * console_sem. If we are not the panic CPU, then we should
1994 	 * avoid taking console_sem, so the panic CPU has a better
1995 	 * chance of cleanly acquiring it later.
1996 	 */
1997 	if (panic_in_progress())
1998 		return 0;
1999 
2000 	printk_safe_enter_irqsave(flags);
2001 
2002 	raw_spin_lock(&console_owner_lock);
2003 	owner = READ_ONCE(console_owner);
2004 	waiter = READ_ONCE(console_waiter);
2005 	if (!waiter && owner && owner != current) {
2006 		WRITE_ONCE(console_waiter, true);
2007 		spin = true;
2008 	}
2009 	raw_spin_unlock(&console_owner_lock);
2010 
2011 	/*
2012 	 * If there is an active printk() writing to the
2013 	 * consoles, instead of having it write our data too,
2014 	 * see if we can offload that load from the active
2015 	 * printer, and do some printing ourselves.
2016 	 * Go into a spin only if there isn't already a waiter
2017 	 * spinning, and there is an active printer, and
2018 	 * that active printer isn't us (recursive printk?).
2019 	 */
2020 	if (!spin) {
2021 		printk_safe_exit_irqrestore(flags);
2022 		return 0;
2023 	}
2024 
2025 	/* We spin waiting for the owner to release us */
2026 	spin_acquire(&console_owner_dep_map, 0, 0, _THIS_IP_);
2027 	/* Owner will clear console_waiter on hand off */
2028 	while (READ_ONCE(console_waiter))
2029 		cpu_relax();
2030 	spin_release(&console_owner_dep_map, _THIS_IP_);
2031 
2032 	printk_safe_exit_irqrestore(flags);
2033 	/*
2034 	 * The owner passed the console lock to us.
2035 	 * Since we did not spin on console lock, annotate
2036 	 * this as a trylock. Otherwise lockdep will
2037 	 * complain.
2038 	 */
2039 	mutex_acquire(&console_lock_dep_map, 0, 1, _THIS_IP_);
2040 
2041 	/*
2042 	 * Update @console_may_schedule for trylock because the previous
2043 	 * owner may have been schedulable.
2044 	 */
2045 	console_may_schedule = 0;
2046 
2047 	return 1;
2048 }
2049 
2050 /*
2051  * Recursion is tracked separately on each CPU. If NMIs are supported, an
2052  * additional NMI context per CPU is also separately tracked. Until per-CPU
2053  * is available, a separate "early tracking" is performed.
2054  */
2055 static DEFINE_PER_CPU(u8, printk_count);
2056 static u8 printk_count_early;
2057 #ifdef CONFIG_HAVE_NMI
2058 static DEFINE_PER_CPU(u8, printk_count_nmi);
2059 static u8 printk_count_nmi_early;
2060 #endif
2061 
2062 /*
2063  * Recursion is limited to keep the output sane. printk() should not require
2064  * more than 1 level of recursion (allowing, for example, printk() to trigger
2065  * a WARN), but a higher value is used in case some printk-internal errors
2066  * exist, such as the ringbuffer validation checks failing.
2067  */
2068 #define PRINTK_MAX_RECURSION 3
2069 
2070 /*
2071  * Return a pointer to the dedicated counter for the CPU+context of the
2072  * caller.
2073  */
2074 static u8 *__printk_recursion_counter(void)
2075 {
2076 #ifdef CONFIG_HAVE_NMI
2077 	if (in_nmi()) {
2078 		if (printk_percpu_data_ready())
2079 			return this_cpu_ptr(&printk_count_nmi);
2080 		return &printk_count_nmi_early;
2081 	}
2082 #endif
2083 	if (printk_percpu_data_ready())
2084 		return this_cpu_ptr(&printk_count);
2085 	return &printk_count_early;
2086 }
2087 
2088 /*
2089  * Enter recursion tracking. Interrupts are disabled to simplify tracking.
2090  * The caller must check the boolean return value to see if the recursion is
2091  * allowed. On failure, interrupts are not disabled.
2092  *
2093  * @recursion_ptr must be a variable of type (u8 *) and is the same variable
2094  * that is passed to printk_exit_irqrestore().
2095  */
2096 #define printk_enter_irqsave(recursion_ptr, flags)	\
2097 ({							\
2098 	bool success = true;				\
2099 							\
2100 	typecheck(u8 *, recursion_ptr);			\
2101 	local_irq_save(flags);				\
2102 	(recursion_ptr) = __printk_recursion_counter();	\
2103 	if (*(recursion_ptr) > PRINTK_MAX_RECURSION) {	\
2104 		local_irq_restore(flags);		\
2105 		success = false;			\
2106 	} else {					\
2107 		(*(recursion_ptr))++;			\
2108 	}						\
2109 	success;					\
2110 })
2111 
2112 /* Exit recursion tracking, restoring interrupts. */
2113 #define printk_exit_irqrestore(recursion_ptr, flags)	\
2114 	do {						\
2115 		typecheck(u8 *, recursion_ptr);		\
2116 		(*(recursion_ptr))--;			\
2117 		local_irq_restore(flags);		\
2118 	} while (0)
2119 
2120 int printk_delay_msec __read_mostly;
2121 
2122 static inline void printk_delay(int level)
2123 {
2124 	boot_delay_msec(level);
2125 
2126 	if (unlikely(printk_delay_msec)) {
2127 		int m = printk_delay_msec;
2128 
2129 		while (m--) {
2130 			mdelay(1);
2131 			touch_nmi_watchdog();
2132 		}
2133 	}
2134 }
2135 
2136 #define CALLER_ID_MASK 0x80000000
2137 
2138 static inline u32 printk_caller_id(void)
2139 {
2140 	return in_task() ? task_pid_nr(current) :
2141 		CALLER_ID_MASK + smp_processor_id();
2142 }
2143 
2144 #ifdef CONFIG_PRINTK_EXECUTION_CTX
2145 /* Store the opposite info than caller_id. */
2146 static u32 printk_caller_id2(void)
2147 {
2148 	return !in_task() ? task_pid_nr(current) :
2149 		CALLER_ID_MASK + smp_processor_id();
2150 }
2151 
2152 static pid_t printk_info_get_pid(const struct printk_info *info)
2153 {
2154 	u32 caller_id = info->caller_id;
2155 	u32 caller_id2 = info->caller_id2;
2156 
2157 	return caller_id & CALLER_ID_MASK ? caller_id2 : caller_id;
2158 }
2159 
2160 static int printk_info_get_cpu(const struct printk_info *info)
2161 {
2162 	u32 caller_id = info->caller_id;
2163 	u32 caller_id2 = info->caller_id2;
2164 
2165 	return ((caller_id & CALLER_ID_MASK ?
2166 		 caller_id : caller_id2) & ~CALLER_ID_MASK);
2167 }
2168 #endif
2169 
2170 /**
2171  * printk_parse_prefix - Parse level and control flags.
2172  *
2173  * @text:     The terminated text message.
2174  * @level:    A pointer to the current level value, will be updated.
2175  * @flags:    A pointer to the current printk_info flags, will be updated.
2176  *
2177  * @level may be NULL if the caller is not interested in the parsed value.
2178  * Otherwise the variable pointed to by @level must be set to
2179  * LOGLEVEL_DEFAULT in order to be updated with the parsed value.
2180  *
2181  * @flags may be NULL if the caller is not interested in the parsed value.
2182  * Otherwise the variable pointed to by @flags will be OR'd with the parsed
2183  * value.
2184  *
2185  * Return: The length of the parsed level and control flags.
2186  */
2187 u16 printk_parse_prefix(const char *text, int *level,
2188 			enum printk_info_flags *flags)
2189 {
2190 	u16 prefix_len = 0;
2191 	int kern_level;
2192 
2193 	while (*text) {
2194 		kern_level = printk_get_level(text);
2195 		if (!kern_level)
2196 			break;
2197 
2198 		switch (kern_level) {
2199 		case '0' ... '7':
2200 			if (level && *level == LOGLEVEL_DEFAULT)
2201 				*level = kern_level - '0';
2202 			break;
2203 		case 'c':	/* KERN_CONT */
2204 			if (flags)
2205 				*flags |= LOG_CONT;
2206 		}
2207 
2208 		prefix_len += 2;
2209 		text += 2;
2210 	}
2211 
2212 	return prefix_len;
2213 }
2214 
2215 __printf(5, 0)
2216 static u16 printk_sprint(char *text, u16 size, int facility,
2217 			 enum printk_info_flags *flags, const char *fmt,
2218 			 va_list args)
2219 {
2220 	u16 text_len;
2221 
2222 	text_len = vscnprintf(text, size, fmt, args);
2223 
2224 	/* Mark and strip a trailing newline. */
2225 	if (text_len && text[text_len - 1] == '\n') {
2226 		text_len--;
2227 		*flags |= LOG_NEWLINE;
2228 	}
2229 
2230 	/* Strip log level and control flags. */
2231 	if (facility == 0) {
2232 		u16 prefix_len;
2233 
2234 		prefix_len = printk_parse_prefix(text, NULL, NULL);
2235 		if (prefix_len) {
2236 			text_len -= prefix_len;
2237 			memmove(text, text + prefix_len, text_len);
2238 		}
2239 	}
2240 
2241 	trace_console(text, text_len);
2242 
2243 	return text_len;
2244 }
2245 
2246 #ifdef CONFIG_PRINTK_EXECUTION_CTX
2247 static void printk_store_execution_ctx(struct printk_info *info)
2248 {
2249 	info->caller_id2 = printk_caller_id2();
2250 	get_task_comm(info->comm, current);
2251 }
2252 
2253 static void pmsg_load_execution_ctx(struct printk_message *pmsg,
2254 				    const struct printk_info *info)
2255 {
2256 	pmsg->cpu = printk_info_get_cpu(info);
2257 	pmsg->pid = printk_info_get_pid(info);
2258 	memcpy(pmsg->comm, info->comm, sizeof(pmsg->comm));
2259 	static_assert(sizeof(pmsg->comm) == sizeof(info->comm));
2260 }
2261 #else
2262 static void printk_store_execution_ctx(struct printk_info *info) {}
2263 
2264 static void pmsg_load_execution_ctx(struct printk_message *pmsg,
2265 				    const struct printk_info *info) {}
2266 #endif
2267 
2268 __printf(4, 0)
2269 int vprintk_store(int facility, int level,
2270 		  const struct dev_printk_info *dev_info,
2271 		  const char *fmt, va_list args)
2272 {
2273 	struct prb_reserved_entry e;
2274 	enum printk_info_flags flags = 0;
2275 	struct printk_record r;
2276 	unsigned long irqflags;
2277 	u16 trunc_msg_len = 0;
2278 	char prefix_buf[8];
2279 	u8 *recursion_ptr;
2280 	u16 reserve_size;
2281 	va_list args2;
2282 	u32 caller_id;
2283 	u16 text_len;
2284 	int ret = 0;
2285 	u64 ts_nsec;
2286 
2287 	if (!printk_enter_irqsave(recursion_ptr, irqflags))
2288 		return 0;
2289 
2290 	/*
2291 	 * Since the duration of printk() can vary depending on the message
2292 	 * and state of the ringbuffer, grab the timestamp now so that it is
2293 	 * close to the call of printk(). This provides a more deterministic
2294 	 * timestamp with respect to the caller.
2295 	 */
2296 	ts_nsec = local_clock();
2297 
2298 	caller_id = printk_caller_id();
2299 
2300 	/*
2301 	 * The sprintf needs to come first since the syslog prefix might be
2302 	 * passed in as a parameter. An extra byte must be reserved so that
2303 	 * later the vscnprintf() into the reserved buffer has room for the
2304 	 * terminating '\0', which is not counted by vsnprintf().
2305 	 */
2306 	va_copy(args2, args);
2307 	reserve_size = vsnprintf(&prefix_buf[0], sizeof(prefix_buf), fmt, args2) + 1;
2308 	va_end(args2);
2309 
2310 	if (reserve_size > PRINTKRB_RECORD_MAX)
2311 		reserve_size = PRINTKRB_RECORD_MAX;
2312 
2313 	/* Extract log level or control flags. */
2314 	if (facility == 0)
2315 		printk_parse_prefix(&prefix_buf[0], &level, &flags);
2316 
2317 	if (level == LOGLEVEL_DEFAULT)
2318 		level = default_message_loglevel;
2319 
2320 	if (dev_info)
2321 		flags |= LOG_NEWLINE;
2322 
2323 	if (is_printk_force_console())
2324 		flags |= LOG_FORCE_CON;
2325 
2326 	if (flags & LOG_CONT) {
2327 		prb_rec_init_wr(&r, reserve_size);
2328 		if (prb_reserve_in_last(&e, prb, &r, caller_id, PRINTKRB_RECORD_MAX)) {
2329 			text_len = printk_sprint(&r.text_buf[r.info->text_len], reserve_size,
2330 						 facility, &flags, fmt, args);
2331 			r.info->text_len += text_len;
2332 
2333 			if (flags & LOG_FORCE_CON)
2334 				r.info->flags |= LOG_FORCE_CON;
2335 
2336 			if (flags & LOG_NEWLINE) {
2337 				r.info->flags |= LOG_NEWLINE;
2338 				prb_final_commit(&e);
2339 			} else {
2340 				prb_commit(&e);
2341 			}
2342 
2343 			ret = text_len;
2344 			goto out;
2345 		}
2346 	}
2347 
2348 	/*
2349 	 * Explicitly initialize the record before every prb_reserve() call.
2350 	 * prb_reserve_in_last() and prb_reserve() purposely invalidate the
2351 	 * structure when they fail.
2352 	 */
2353 	prb_rec_init_wr(&r, reserve_size);
2354 	if (!prb_reserve(&e, prb, &r)) {
2355 		/* truncate the message if it is too long for empty buffer */
2356 		truncate_msg(&reserve_size, &trunc_msg_len);
2357 
2358 		prb_rec_init_wr(&r, reserve_size + trunc_msg_len);
2359 		if (!prb_reserve(&e, prb, &r))
2360 			goto out;
2361 	}
2362 
2363 	/* fill message */
2364 	text_len = printk_sprint(&r.text_buf[0], reserve_size, facility, &flags, fmt, args);
2365 	if (trunc_msg_len)
2366 		memcpy(&r.text_buf[text_len], trunc_msg, trunc_msg_len);
2367 	r.info->text_len = text_len + trunc_msg_len;
2368 	r.info->facility = facility;
2369 	r.info->level = level & 7;
2370 	r.info->flags = flags & 0x1f;
2371 	r.info->ts_nsec = ts_nsec;
2372 	r.info->caller_id = caller_id;
2373 	if (dev_info)
2374 		memcpy(&r.info->dev_info, dev_info, sizeof(r.info->dev_info));
2375 	printk_store_execution_ctx(r.info);
2376 
2377 	/* A message without a trailing newline can be continued. */
2378 	if (!(flags & LOG_NEWLINE))
2379 		prb_commit(&e);
2380 	else
2381 		prb_final_commit(&e);
2382 
2383 	ret = text_len + trunc_msg_len;
2384 out:
2385 	printk_exit_irqrestore(recursion_ptr, irqflags);
2386 	return ret;
2387 }
2388 
2389 /*
2390  * This acts as a one-way switch to allow legacy consoles to print from
2391  * the printk() caller context on a panic CPU. It also attempts to flush
2392  * the legacy consoles in this context.
2393  */
2394 void printk_legacy_allow_panic_sync(void)
2395 {
2396 	struct console_flush_type ft;
2397 
2398 	legacy_allow_panic_sync = true;
2399 
2400 	printk_get_console_flush_type(&ft);
2401 	if (ft.legacy_direct) {
2402 		if (console_trylock())
2403 			console_unlock();
2404 	}
2405 }
2406 
2407 bool __read_mostly debug_non_panic_cpus;
2408 
2409 #ifdef CONFIG_PRINTK_CALLER
2410 static int __init debug_non_panic_cpus_setup(char *str)
2411 {
2412 	debug_non_panic_cpus = true;
2413 	pr_info("allow messages from non-panic CPUs in panic()\n");
2414 
2415 	return 0;
2416 }
2417 early_param("debug_non_panic_cpus", debug_non_panic_cpus_setup);
2418 module_param(debug_non_panic_cpus, bool, 0644);
2419 MODULE_PARM_DESC(debug_non_panic_cpus,
2420 		 "allow messages from non-panic CPUs in panic()");
2421 #endif
2422 
2423 asmlinkage int vprintk_emit(int facility, int level,
2424 			    const struct dev_printk_info *dev_info,
2425 			    const char *fmt, va_list args)
2426 {
2427 	struct console_flush_type ft;
2428 	int printed_len;
2429 
2430 	/* Suppress unimportant messages after panic happens */
2431 	if (unlikely(suppress_printk))
2432 		return 0;
2433 
2434 	/*
2435 	 * The messages on the panic CPU are the most important. If
2436 	 * non-panic CPUs are generating any messages, they will be
2437 	 * silently dropped.
2438 	 */
2439 	if (panic_on_other_cpu() &&
2440 	    !debug_non_panic_cpus &&
2441 	    !panic_triggering_all_cpu_backtrace)
2442 		return 0;
2443 
2444 	printk_get_console_flush_type(&ft);
2445 
2446 	/* If called from the scheduler, we can not call up(). */
2447 	if (level == LOGLEVEL_SCHED) {
2448 		level = LOGLEVEL_DEFAULT;
2449 		ft.legacy_offload |= ft.legacy_direct && !console_irqwork_blocked;
2450 		ft.legacy_direct = false;
2451 	}
2452 
2453 	printk_delay(level);
2454 
2455 	printed_len = vprintk_store(facility, level, dev_info, fmt, args);
2456 
2457 	if (ft.nbcon_atomic)
2458 		nbcon_atomic_flush_pending();
2459 
2460 	if (ft.nbcon_offload)
2461 		nbcon_kthreads_wake();
2462 
2463 	if (ft.legacy_direct) {
2464 		/*
2465 		 * The caller may be holding system-critical or
2466 		 * timing-sensitive locks. Disable preemption during
2467 		 * printing of all remaining records to all consoles so that
2468 		 * this context can return as soon as possible. Hopefully
2469 		 * another printk() caller will take over the printing.
2470 		 */
2471 		preempt_disable();
2472 		/*
2473 		 * Try to acquire and then immediately release the console
2474 		 * semaphore. The release will print out buffers. With the
2475 		 * spinning variant, this context tries to take over the
2476 		 * printing from another printing context.
2477 		 */
2478 		if (console_trylock_spinning())
2479 			console_unlock();
2480 		preempt_enable();
2481 	}
2482 
2483 	if (ft.legacy_offload)
2484 		defer_console_output();
2485 	else if (!console_irqwork_blocked)
2486 		wake_up_klogd();
2487 
2488 	return printed_len;
2489 }
2490 EXPORT_SYMBOL(vprintk_emit);
2491 
2492 int vprintk_default(const char *fmt, va_list args)
2493 {
2494 	return vprintk_emit(0, LOGLEVEL_DEFAULT, NULL, fmt, args);
2495 }
2496 EXPORT_SYMBOL_GPL(vprintk_default);
2497 
2498 asmlinkage __visible int _printk(const char *fmt, ...)
2499 {
2500 	va_list args;
2501 	int r;
2502 
2503 	va_start(args, fmt);
2504 	r = vprintk(fmt, args);
2505 	va_end(args);
2506 
2507 	return r;
2508 }
2509 EXPORT_SYMBOL(_printk);
2510 
2511 static bool __pr_flush(struct console *con, int timeout_ms, bool reset_on_progress);
2512 
2513 #else /* CONFIG_PRINTK */
2514 
2515 #define printk_time		false
2516 
2517 #define prb_read_valid(rb, seq, r)	false
2518 #define prb_first_valid_seq(rb)		0
2519 #define prb_next_seq(rb)		0
2520 
2521 static u64 syslog_seq;
2522 
2523 static bool __pr_flush(struct console *con, int timeout_ms, bool reset_on_progress) { return true; }
2524 
2525 #endif /* CONFIG_PRINTK */
2526 
2527 #ifdef CONFIG_EARLY_PRINTK
2528 struct console *early_console;
2529 
2530 asmlinkage __visible void early_printk(const char *fmt, ...)
2531 {
2532 	va_list ap;
2533 	char buf[512];
2534 	int n;
2535 
2536 	if (!early_console)
2537 		return;
2538 
2539 	va_start(ap, fmt);
2540 	n = vscnprintf(buf, sizeof(buf), fmt, ap);
2541 	va_end(ap);
2542 
2543 	early_console->write(early_console, buf, n);
2544 }
2545 #endif
2546 
2547 static void set_user_specified(struct console_cmdline *c, bool user_specified)
2548 {
2549 	if (!user_specified)
2550 		return;
2551 
2552 	/*
2553 	 * @c console was defined by the user on the command line.
2554 	 * Do not clear when added twice also by SPCR or the device tree.
2555 	 */
2556 	c->user_specified = true;
2557 	/* At least one console defined by the user on the command line. */
2558 	console_set_on_cmdline = 1;
2559 }
2560 
2561 static int __add_preferred_console(const char *name, const short idx,
2562 				   const char *devname, char *options,
2563 				   char *brl_options, bool user_specified)
2564 {
2565 	struct console_cmdline *c;
2566 	int i;
2567 
2568 	if (!name && !devname)
2569 		return -EINVAL;
2570 
2571 	/*
2572 	 * We use a signed short index for struct console for device drivers to
2573 	 * indicate a not yet assigned index or port. However, a negative index
2574 	 * value is not valid when the console name and index are defined on
2575 	 * the command line.
2576 	 */
2577 	if (name && idx < 0)
2578 		return -EINVAL;
2579 
2580 	/*
2581 	 *	See if this tty is not yet registered, and
2582 	 *	if we have a slot free.
2583 	 */
2584 	for (i = 0, c = console_cmdline;
2585 	     i < MAX_CMDLINECONSOLES && (c->name[0] || c->devname[0]);
2586 	     i++, c++) {
2587 		if ((name && strcmp(c->name, name) == 0 && c->index == idx) ||
2588 		    (devname && strcmp(c->devname, devname) == 0)) {
2589 			if (!brl_options)
2590 				preferred_console = i;
2591 			set_user_specified(c, user_specified);
2592 			return 0;
2593 		}
2594 	}
2595 	if (i == MAX_CMDLINECONSOLES)
2596 		return -E2BIG;
2597 	if (!brl_options)
2598 		preferred_console = i;
2599 	if (name)
2600 		strscpy(c->name, name);
2601 	if (devname)
2602 		strscpy(c->devname, devname);
2603 	c->options = options;
2604 	set_user_specified(c, user_specified);
2605 	braille_set_options(c, brl_options);
2606 
2607 	c->index = idx;
2608 	return 0;
2609 }
2610 
2611 static int __init console_msg_format_setup(char *str)
2612 {
2613 	if (!strcmp(str, "syslog"))
2614 		console_msg_format = MSG_FORMAT_SYSLOG;
2615 	if (!strcmp(str, "default"))
2616 		console_msg_format = MSG_FORMAT_DEFAULT;
2617 	return 1;
2618 }
2619 __setup("console_msg_format=", console_msg_format_setup);
2620 
2621 /*
2622  * Set up a console.  Called via do_early_param() in init/main.c
2623  * for each "console=" parameter in the boot command line.
2624  */
2625 static int __init console_setup(char *str)
2626 {
2627 	static_assert(sizeof(console_cmdline[0].devname) >= sizeof(console_cmdline[0].name) + 4);
2628 	char buf[sizeof(console_cmdline[0].devname)];
2629 	char *brl_options = NULL;
2630 	char *ttyname = NULL;
2631 	char *devname = NULL;
2632 	char *options;
2633 	char *s;
2634 	int idx;
2635 
2636 	/*
2637 	 * console="" or console=null have been suggested as a way to
2638 	 * disable console output. Use ttynull that has been created
2639 	 * for exactly this purpose.
2640 	 */
2641 	if (str[0] == 0 || strcmp(str, "null") == 0) {
2642 		__add_preferred_console("ttynull", 0, NULL, NULL, NULL, true);
2643 		return 1;
2644 	}
2645 
2646 	if (_braille_console_setup(&str, &brl_options))
2647 		return 1;
2648 
2649 	/*
2650 	 * Decode str into name, index and options. Start with options, since
2651 	 * it might also contain a ':' used for DEVNAME.
2652 	 */
2653 	options = strchr(str, ',');
2654 	if (options)
2655 		*(options++) = 0;
2656 
2657 	/* For a DEVNAME:0.0 style console the character device is unknown early */
2658 	if (strchr(str, ':'))
2659 		devname = buf;
2660 	else
2661 		ttyname = buf;
2662 
2663 	if (ttyname && isdigit(str[0]))
2664 		scnprintf(buf, sizeof(buf), "ttyS%s", str);
2665 	else
2666 		strscpy(buf, str);
2667 
2668 #ifdef __sparc__
2669 	if (!strcmp(str, "ttya"))
2670 		strscpy(buf, "ttyS0");
2671 	if (!strcmp(str, "ttyb"))
2672 		strscpy(buf, "ttyS1");
2673 #endif
2674 
2675 	if (ttyname) {
2676 		/* Detect @idx in ttyname and remove it. */
2677 		for (s = ttyname; *s; s++) {
2678 			if (isdigit(*s))
2679 				break;
2680 		}
2681 		idx = simple_strtoul(s, NULL, 10);
2682 		*s = 0;
2683 	} else {
2684 		/* @idx will get defined when devname matches. */
2685 		idx = -1;
2686 	}
2687 
2688 	__add_preferred_console(ttyname, idx, devname, options, brl_options, true);
2689 	return 1;
2690 }
2691 __setup("console=", console_setup);
2692 
2693 /**
2694  * add_preferred_console - add a device to the list of preferred consoles.
2695  * @name: device name
2696  * @idx: device index
2697  * @options: options for this console
2698  *
2699  * The last preferred console added will be used for kernel messages
2700  * and stdin/out/err for init.  Normally this is used by console_setup
2701  * above to handle user-supplied console arguments; however it can also
2702  * be used by arch-specific code either to override the user or more
2703  * commonly to provide a default console (ie from PROM variables) when
2704  * the user has not supplied one.
2705  */
2706 int add_preferred_console(const char *name, const short idx, char *options)
2707 {
2708 	return __add_preferred_console(name, idx, NULL, options, NULL, false);
2709 }
2710 
2711 /**
2712  * match_devname_and_update_preferred_console - Update a preferred console
2713  *	when matching devname is found.
2714  * @devname: DEVNAME:0.0 style device name
2715  * @name: Name of the corresponding console driver, e.g. "ttyS"
2716  * @idx: Console index, e.g. port number.
2717  *
2718  * The function checks whether a device with the given @devname is
2719  * preferred via the console=DEVNAME:0.0 command line option.
2720  * It fills the missing console driver name and console index
2721  * so that a later register_console() call could find (match)
2722  * and enable this device.
2723  *
2724  * It might be used when a driver subsystem initializes particular
2725  * devices with already known DEVNAME:0.0 style names. And it
2726  * could predict which console driver name and index this device
2727  * would later get associated with.
2728  *
2729  * Return: 0 on success, negative error code on failure.
2730  */
2731 int match_devname_and_update_preferred_console(const char *devname,
2732 					       const char *name,
2733 					       const short idx)
2734 {
2735 	struct console_cmdline *c = console_cmdline;
2736 	int i;
2737 
2738 	if (!devname || !strlen(devname) || !name || !strlen(name) || idx < 0)
2739 		return -EINVAL;
2740 
2741 	for (i = 0; i < MAX_CMDLINECONSOLES && (c->name[0] || c->devname[0]);
2742 	     i++, c++) {
2743 		if (!strcmp(devname, c->devname)) {
2744 			pr_info("associate the preferred console \"%s\" with \"%s%d\"\n",
2745 				devname, name, idx);
2746 			strscpy(c->name, name);
2747 			c->index = idx;
2748 			return 0;
2749 		}
2750 	}
2751 
2752 	return -ENOENT;
2753 }
2754 EXPORT_SYMBOL_GPL(match_devname_and_update_preferred_console);
2755 
2756 bool console_suspend_enabled = true;
2757 EXPORT_SYMBOL(console_suspend_enabled);
2758 
2759 static int __init console_suspend_disable(char *str)
2760 {
2761 	console_suspend_enabled = false;
2762 	return 1;
2763 }
2764 __setup("no_console_suspend", console_suspend_disable);
2765 module_param_named(console_suspend, console_suspend_enabled,
2766 		bool, S_IRUGO | S_IWUSR);
2767 MODULE_PARM_DESC(console_suspend, "suspend console during suspend"
2768 	" and hibernate operations");
2769 
2770 static bool printk_console_no_auto_verbose;
2771 
2772 void console_verbose(void)
2773 {
2774 	if (console_loglevel && !printk_console_no_auto_verbose)
2775 		console_loglevel = CONSOLE_LOGLEVEL_MOTORMOUTH;
2776 }
2777 EXPORT_SYMBOL_GPL(console_verbose);
2778 
2779 module_param_named(console_no_auto_verbose, printk_console_no_auto_verbose, bool, 0644);
2780 MODULE_PARM_DESC(console_no_auto_verbose, "Disable console loglevel raise to highest on oops/panic/etc");
2781 
2782 /**
2783  * console_suspend_all - suspend the console subsystem
2784  *
2785  * This disables printk() while we go into suspend states
2786  */
2787 void console_suspend_all(void)
2788 {
2789 	struct console *con;
2790 
2791 	if (console_suspend_enabled)
2792 		pr_info("Suspending console(s) (use no_console_suspend to debug)\n");
2793 
2794 	/*
2795 	 * Flush any console backlog and then avoid queueing irq_work until
2796 	 * console_resume_all(). Until then deferred printing is no longer
2797 	 * triggered, NBCON consoles transition to atomic flushing, and
2798 	 * any klogd waiters are not triggered.
2799 	 */
2800 	pr_flush(1000, true);
2801 	console_irqwork_blocked = true;
2802 
2803 	if (!console_suspend_enabled)
2804 		return;
2805 
2806 	console_list_lock();
2807 	for_each_console(con)
2808 		console_srcu_write_flags(con, con->flags | CON_SUSPENDED);
2809 	console_list_unlock();
2810 
2811 	/*
2812 	 * Ensure that all SRCU list walks have completed. All printing
2813 	 * contexts must be able to see that they are suspended so that it
2814 	 * is guaranteed that all printing has stopped when this function
2815 	 * completes.
2816 	 */
2817 	synchronize_srcu(&console_srcu);
2818 }
2819 
2820 void console_resume_all(void)
2821 {
2822 	struct console_flush_type ft;
2823 	struct console *con;
2824 
2825 	/*
2826 	 * Allow queueing irq_work. After restoring console state, deferred
2827 	 * printing and any klogd waiters need to be triggered in case there
2828 	 * is now a console backlog.
2829 	 */
2830 	console_irqwork_blocked = false;
2831 
2832 	if (console_suspend_enabled) {
2833 		console_list_lock();
2834 		for_each_console(con)
2835 			console_srcu_write_flags(con, con->flags & ~CON_SUSPENDED);
2836 		console_list_unlock();
2837 
2838 		/*
2839 		 * Ensure that all SRCU list walks have completed. All printing
2840 		 * contexts must be able to see they are no longer suspended so
2841 		 * that they are guaranteed to wake up and resume printing.
2842 		 */
2843 		synchronize_srcu(&console_srcu);
2844 	}
2845 
2846 	printk_get_console_flush_type(&ft);
2847 	if (ft.nbcon_offload)
2848 		nbcon_kthreads_wake();
2849 	if (ft.legacy_offload)
2850 		defer_console_output();
2851 	else
2852 		wake_up_klogd();
2853 
2854 	pr_flush(1000, true);
2855 }
2856 
2857 /**
2858  * console_cpu_notify - print deferred console messages after CPU hotplug
2859  * @cpu: unused
2860  *
2861  * If printk() is called from a CPU that is not online yet, the messages
2862  * will be printed on the console only if there are CON_ANYTIME consoles.
2863  * This function is called when a new CPU comes online (or fails to come
2864  * up) or goes offline.
2865  */
2866 static int console_cpu_notify(unsigned int cpu)
2867 {
2868 	struct console_flush_type ft;
2869 
2870 	if (!cpuhp_tasks_frozen) {
2871 		printk_get_console_flush_type(&ft);
2872 		if (ft.nbcon_atomic)
2873 			nbcon_atomic_flush_pending();
2874 		if (ft.legacy_direct) {
2875 			if (console_trylock())
2876 				console_unlock();
2877 		}
2878 	}
2879 	return 0;
2880 }
2881 
2882 /**
2883  * console_lock - block the console subsystem from printing
2884  *
2885  * Acquires a lock which guarantees that no consoles will
2886  * be in or enter their write() callback.
2887  *
2888  * Can sleep, returns nothing.
2889  */
2890 void console_lock(void)
2891 {
2892 	might_sleep();
2893 
2894 	/* On panic, the console_lock must be left to the panic cpu. */
2895 	while (panic_on_other_cpu())
2896 		msleep(1000);
2897 
2898 	down_console_sem();
2899 	console_locked = 1;
2900 	console_may_schedule = 1;
2901 }
2902 EXPORT_SYMBOL(console_lock);
2903 
2904 /**
2905  * console_trylock - try to block the console subsystem from printing
2906  *
2907  * Try to acquire a lock which guarantees that no consoles will
2908  * be in or enter their write() callback.
2909  *
2910  * returns 1 on success, and 0 on failure to acquire the lock.
2911  */
2912 int console_trylock(void)
2913 {
2914 	/* On panic, the console_lock must be left to the panic cpu. */
2915 	if (panic_on_other_cpu())
2916 		return 0;
2917 	if (down_trylock_console_sem())
2918 		return 0;
2919 	console_locked = 1;
2920 	console_may_schedule = 0;
2921 	return 1;
2922 }
2923 EXPORT_SYMBOL(console_trylock);
2924 
2925 int is_console_locked(void)
2926 {
2927 	return console_locked;
2928 }
2929 EXPORT_SYMBOL(is_console_locked);
2930 
2931 static void __console_unlock(void)
2932 {
2933 	console_locked = 0;
2934 	up_console_sem();
2935 }
2936 
2937 #ifdef CONFIG_PRINTK
2938 
2939 /*
2940  * Prepend the message in @pmsg->pbufs->outbuf. This is achieved by shifting
2941  * the existing message over and inserting the scratchbuf message.
2942  *
2943  * @pmsg is the original printk message.
2944  * @fmt is the printf format of the message which will prepend the existing one.
2945  *
2946  * If there is not enough space in @pmsg->pbufs->outbuf, the existing
2947  * message text will be sufficiently truncated.
2948  *
2949  * If @pmsg->pbufs->outbuf is modified, @pmsg->outbuf_len is updated.
2950  */
2951 __printf(2, 3)
2952 static void console_prepend_message(struct printk_message *pmsg, const char *fmt, ...)
2953 {
2954 	struct printk_buffers *pbufs = pmsg->pbufs;
2955 	const size_t scratchbuf_sz = sizeof(pbufs->scratchbuf);
2956 	const size_t outbuf_sz = sizeof(pbufs->outbuf);
2957 	char *scratchbuf = &pbufs->scratchbuf[0];
2958 	char *outbuf = &pbufs->outbuf[0];
2959 	va_list args;
2960 	size_t len;
2961 
2962 	va_start(args, fmt);
2963 	len = vscnprintf(scratchbuf, scratchbuf_sz, fmt, args);
2964 	va_end(args);
2965 
2966 	/*
2967 	 * Make sure outbuf is sufficiently large before prepending.
2968 	 * Keep at least the prefix when the message must be truncated.
2969 	 * It is a rather theoretical problem when someone tries to
2970 	 * use a minimalist buffer.
2971 	 */
2972 	if (WARN_ON_ONCE(len + PRINTK_PREFIX_MAX >= outbuf_sz))
2973 		return;
2974 
2975 	if (pmsg->outbuf_len + len >= outbuf_sz) {
2976 		/* Truncate the message, but keep it terminated. */
2977 		pmsg->outbuf_len = outbuf_sz - (len + 1);
2978 		outbuf[pmsg->outbuf_len] = 0;
2979 	}
2980 
2981 	memmove(outbuf + len, outbuf, pmsg->outbuf_len + 1);
2982 	memcpy(outbuf, scratchbuf, len);
2983 	pmsg->outbuf_len += len;
2984 }
2985 
2986 /*
2987  * Prepend the message in @pmsg->pbufs->outbuf with a "dropped message".
2988  * @pmsg->outbuf_len is updated appropriately.
2989  *
2990  * @pmsg is the printk message to prepend.
2991  *
2992  * @dropped is the dropped count to report in the dropped message.
2993  */
2994 void console_prepend_dropped(struct printk_message *pmsg, unsigned long dropped)
2995 {
2996 	console_prepend_message(pmsg, "** %lu printk messages dropped **\n", dropped);
2997 }
2998 
2999 /*
3000  * Prepend the message in @pmsg->pbufs->outbuf with a "replay message".
3001  * @pmsg->outbuf_len is updated appropriately.
3002  *
3003  * @pmsg is the printk message to prepend.
3004  */
3005 void console_prepend_replay(struct printk_message *pmsg)
3006 {
3007 	console_prepend_message(pmsg, "** replaying previous printk message **\n");
3008 }
3009 
3010 /*
3011  * Read and format the specified record (or a later record if the specified
3012  * record is not available).
3013  *
3014  * @pmsg will contain the formatted result. @pmsg->pbufs must point to a
3015  * struct printk_buffers.
3016  *
3017  * @seq is the record to read and format. If it is not available, the next
3018  * valid record is read.
3019  *
3020  * @is_extended specifies if the message should be formatted for extended
3021  * console output.
3022  *
3023  * @may_supress specifies if records may be skipped based on loglevel.
3024  *
3025  * Returns false if no record is available. Otherwise true and all fields
3026  * of @pmsg are valid. (See the documentation of struct printk_message
3027  * for information about the @pmsg fields.)
3028  */
3029 bool printk_get_next_message(struct printk_message *pmsg, u64 seq,
3030 			     bool is_extended, bool may_suppress)
3031 {
3032 	struct printk_buffers *pbufs = pmsg->pbufs;
3033 	const size_t scratchbuf_sz = sizeof(pbufs->scratchbuf);
3034 	const size_t outbuf_sz = sizeof(pbufs->outbuf);
3035 	char *scratchbuf = &pbufs->scratchbuf[0];
3036 	char *outbuf = &pbufs->outbuf[0];
3037 	struct printk_info info;
3038 	struct printk_record r;
3039 	size_t len = 0;
3040 	bool force_con;
3041 
3042 	/*
3043 	 * Formatting extended messages requires a separate buffer, so use the
3044 	 * scratch buffer to read in the ringbuffer text.
3045 	 *
3046 	 * Formatting normal messages is done in-place, so read the ringbuffer
3047 	 * text directly into the output buffer.
3048 	 */
3049 	if (is_extended)
3050 		prb_rec_init_rd(&r, &info, scratchbuf, scratchbuf_sz);
3051 	else
3052 		prb_rec_init_rd(&r, &info, outbuf, outbuf_sz);
3053 
3054 	if (!prb_read_valid(prb, seq, &r))
3055 		return false;
3056 
3057 	pmsg->seq = r.info->seq;
3058 	pmsg->dropped = r.info->seq - seq;
3059 	force_con = r.info->flags & LOG_FORCE_CON;
3060 	pmsg_load_execution_ctx(pmsg, r.info);
3061 
3062 	/*
3063 	 * Skip records that are not forced to be printed on consoles and that
3064 	 * has level above the console loglevel.
3065 	 */
3066 	if (!force_con && may_suppress && suppress_message_printing(r.info->level))
3067 		goto out;
3068 
3069 	if (is_extended) {
3070 		len = info_print_ext_header(outbuf, outbuf_sz, r.info);
3071 		len += msg_print_ext_body(outbuf + len, outbuf_sz - len,
3072 					  &r.text_buf[0], r.info->text_len, &r.info->dev_info);
3073 	} else {
3074 		len = record_print_text(&r, console_msg_format & MSG_FORMAT_SYSLOG, printk_time);
3075 	}
3076 out:
3077 	pmsg->outbuf_len = len;
3078 	return true;
3079 }
3080 
3081 /*
3082  * The legacy console always acquires a spinlock_t from its printing
3083  * callback. This violates lock nesting if the caller acquired an always
3084  * spinning lock (raw_spinlock_t) while invoking printk(). This is not a
3085  * problem on PREEMPT_RT because legacy consoles print always from a
3086  * dedicated thread and never from within printk(). Therefore we tell
3087  * lockdep that a sleeping spin lock (spinlock_t) is valid here.
3088  */
3089 #ifdef CONFIG_PREEMPT_RT
3090 static inline void printk_legacy_allow_spinlock_enter(void) { }
3091 static inline void printk_legacy_allow_spinlock_exit(void) { }
3092 #else
3093 static DEFINE_WAIT_OVERRIDE_MAP(printk_legacy_map, LD_WAIT_CONFIG);
3094 
3095 static inline void printk_legacy_allow_spinlock_enter(void)
3096 {
3097 	lock_map_acquire_try(&printk_legacy_map);
3098 }
3099 
3100 static inline void printk_legacy_allow_spinlock_exit(void)
3101 {
3102 	lock_map_release(&printk_legacy_map);
3103 }
3104 #endif /* CONFIG_PREEMPT_RT */
3105 
3106 /*
3107  * Used as the printk buffers for non-panic, serialized console printing.
3108  * This is for legacy (!CON_NBCON) as well as all boot (CON_BOOT) consoles.
3109  * Its usage requires the console_lock held.
3110  */
3111 struct printk_buffers printk_shared_pbufs;
3112 
3113 /*
3114  * Print one record for the given console. The record printed is whatever
3115  * record is the next available record for the given console.
3116  *
3117  * @handover will be set to true if a printk waiter has taken over the
3118  * console_lock, in which case the caller is no longer holding both the
3119  * console_lock and the SRCU read lock. Otherwise it is set to false.
3120  *
3121  * @cookie is the cookie from the SRCU read lock.
3122  *
3123  * Returns false if the given console has no next record to print, otherwise
3124  * true.
3125  *
3126  * Requires the console_lock and the SRCU read lock.
3127  */
3128 static bool console_emit_next_record(struct console *con, bool *handover, int cookie)
3129 {
3130 	bool is_extended = console_srcu_read_flags(con) & CON_EXTENDED;
3131 	char *outbuf = &printk_shared_pbufs.outbuf[0];
3132 	struct printk_message pmsg = {
3133 		.pbufs = &printk_shared_pbufs,
3134 	};
3135 	unsigned long flags;
3136 
3137 	*handover = false;
3138 
3139 	if (!printk_get_next_message(&pmsg, con->seq, is_extended, true))
3140 		return false;
3141 
3142 	con->dropped += pmsg.dropped;
3143 
3144 	/* Skip messages of formatted length 0. */
3145 	if (pmsg.outbuf_len == 0) {
3146 		con->seq = pmsg.seq + 1;
3147 		goto skip;
3148 	}
3149 
3150 	if (con->dropped && !is_extended) {
3151 		console_prepend_dropped(&pmsg, con->dropped);
3152 		con->dropped = 0;
3153 	}
3154 
3155 	/* Write everything out to the hardware. */
3156 
3157 	if (force_legacy_kthread() && !panic_in_progress()) {
3158 		/*
3159 		 * With forced threading this function is in a task context
3160 		 * (either legacy kthread or get_init_console_seq()). There
3161 		 * is no need for concern about printk reentrance, handovers,
3162 		 * or lockdep complaints.
3163 		 */
3164 
3165 		con->write(con, outbuf, pmsg.outbuf_len);
3166 		con->seq = pmsg.seq + 1;
3167 	} else {
3168 		/*
3169 		 * While actively printing out messages, if another printk()
3170 		 * were to occur on another CPU, it may wait for this one to
3171 		 * finish. This task can not be preempted if there is a
3172 		 * waiter waiting to take over.
3173 		 *
3174 		 * Interrupts are disabled because the hand over to a waiter
3175 		 * must not be interrupted until the hand over is completed
3176 		 * (@console_waiter is cleared).
3177 		 */
3178 		printk_safe_enter_irqsave(flags);
3179 		console_lock_spinning_enable();
3180 
3181 		/* Do not trace print latency. */
3182 		stop_critical_timings();
3183 
3184 		printk_legacy_allow_spinlock_enter();
3185 		con->write(con, outbuf, pmsg.outbuf_len);
3186 		printk_legacy_allow_spinlock_exit();
3187 
3188 		start_critical_timings();
3189 
3190 		con->seq = pmsg.seq + 1;
3191 
3192 		*handover = console_lock_spinning_disable_and_check(cookie);
3193 		printk_safe_exit_irqrestore(flags);
3194 	}
3195 skip:
3196 	return true;
3197 }
3198 
3199 #else
3200 
3201 static bool console_emit_next_record(struct console *con, bool *handover, int cookie)
3202 {
3203 	*handover = false;
3204 	return false;
3205 }
3206 
3207 static inline void printk_kthreads_check_locked(void) { }
3208 
3209 #endif /* CONFIG_PRINTK */
3210 
3211 
3212 /*
3213  * Print out one record for each console.
3214  *
3215  * @do_cond_resched is set by the caller. It can be true only in schedulable
3216  * context.
3217  *
3218  * @next_seq is set to the sequence number after the last available record.
3219  * The value is valid only when all usable consoles were flushed. It is
3220  * when the function returns true (can do the job) and @try_again parameter
3221  * is set to false, see below.
3222  *
3223  * @handover will be set to true if a printk waiter has taken over the
3224  * console_lock, in which case the caller is no longer holding the
3225  * console_lock. Otherwise it is set to false.
3226  *
3227  * @try_again will be set to true when it still makes sense to call this
3228  * function again. The function could do the job, see the return value.
3229  * And some consoles still make progress.
3230  *
3231  * Returns true when the function could do the job. Some consoles are usable,
3232  * and there was no takeover and no panic_on_other_cpu().
3233  *
3234  * Requires the console_lock.
3235  */
3236 static bool console_flush_one_record(bool do_cond_resched, u64 *next_seq, bool *handover,
3237 				     bool *try_again)
3238 {
3239 	struct console_flush_type ft;
3240 	bool any_usable = false;
3241 	struct console *con;
3242 	int cookie;
3243 
3244 	*try_again = false;
3245 
3246 	printk_get_console_flush_type(&ft);
3247 
3248 	cookie = console_srcu_read_lock();
3249 	for_each_console_srcu(con) {
3250 		short flags = console_srcu_read_flags(con);
3251 		u64 printk_seq;
3252 		bool progress;
3253 
3254 		/*
3255 		 * console_flush_one_record() is only responsible for
3256 		 * nbcon consoles when the nbcon consoles cannot print via
3257 		 * their atomic or threaded flushing.
3258 		 */
3259 		if ((flags & CON_NBCON) && (ft.nbcon_atomic || ft.nbcon_offload))
3260 			continue;
3261 
3262 		if (!console_is_usable(con, flags, !do_cond_resched))
3263 			continue;
3264 		any_usable = true;
3265 
3266 		if (flags & CON_NBCON) {
3267 			progress = nbcon_legacy_emit_next_record(con, handover, cookie,
3268 								 !do_cond_resched);
3269 		} else {
3270 			progress = console_emit_next_record(con, handover, cookie);
3271 		}
3272 
3273 		/*
3274 		 * If a handover has occurred, the SRCU read lock
3275 		 * is already released.
3276 		 */
3277 		if (*handover)
3278 			goto fail;
3279 
3280 		/*
3281 		 * @con can be used here now that it is certain that this
3282 		 * context is still holding the SRCU read lock.
3283 		 */
3284 		if (flags & CON_NBCON)
3285 			printk_seq = nbcon_seq_read(con);
3286 		else
3287 			printk_seq = con->seq;
3288 
3289 		/* Track the next of the highest seq flushed. */
3290 		if (printk_seq > *next_seq)
3291 			*next_seq = printk_seq;
3292 
3293 		if (!progress)
3294 			continue;
3295 
3296 		/*
3297 		 * A usable console made progress. There might still be
3298 		 * pending messages.
3299 		 */
3300 		*try_again = true;
3301 
3302 		/* Allow panic_cpu to take over the consoles safely. */
3303 		if (panic_on_other_cpu())
3304 			goto fail_srcu;
3305 
3306 		if (do_cond_resched)
3307 			cond_resched();
3308 	}
3309 	console_srcu_read_unlock(cookie);
3310 
3311 	return any_usable;
3312 
3313 fail_srcu:
3314 	console_srcu_read_unlock(cookie);
3315 fail:
3316 	*try_again = false;
3317 	return false;
3318 }
3319 
3320 /*
3321  * Print out all remaining records to all consoles.
3322  *
3323  * @do_cond_resched is set by the caller. It can be true only in schedulable
3324  * context.
3325  *
3326  * @next_seq is set to the sequence number after the last available record.
3327  * The value is valid only when this function returns true. It means that all
3328  * usable consoles are completely flushed.
3329  *
3330  * @handover will be set to true if a printk waiter has taken over the
3331  * console_lock, in which case the caller is no longer holding the
3332  * console_lock. Otherwise it is set to false.
3333  *
3334  * Returns true when there was at least one usable console and all messages
3335  * were flushed to all usable consoles. A returned false informs the caller
3336  * that everything was not flushed (either there were no usable consoles or
3337  * another context has taken over printing or it is a panic situation and this
3338  * is not the panic CPU). Regardless the reason, the caller should assume it
3339  * is not useful to immediately try again.
3340  *
3341  * Requires the console_lock.
3342  */
3343 static bool console_flush_all(bool do_cond_resched, u64 *next_seq, bool *handover)
3344 {
3345 	bool try_again;
3346 	bool ret;
3347 
3348 	*next_seq = 0;
3349 	*handover = false;
3350 
3351 	do {
3352 		ret = console_flush_one_record(do_cond_resched, next_seq,
3353 					       handover, &try_again);
3354 	} while (try_again);
3355 
3356 	return ret;
3357 }
3358 
3359 static void __console_flush_and_unlock(void)
3360 {
3361 	bool do_cond_resched;
3362 	bool handover;
3363 	bool flushed;
3364 	u64 next_seq;
3365 
3366 	/*
3367 	 * Console drivers are called with interrupts disabled, so
3368 	 * @console_may_schedule should be cleared before; however, we may
3369 	 * end up dumping a lot of lines, for example, if called from
3370 	 * console registration path, and should invoke cond_resched()
3371 	 * between lines if allowable.  Not doing so can cause a very long
3372 	 * scheduling stall on a slow console leading to RCU stall and
3373 	 * softlockup warnings which exacerbate the issue with more
3374 	 * messages practically incapacitating the system. Therefore, create
3375 	 * a local to use for the printing loop.
3376 	 */
3377 	do_cond_resched = console_may_schedule;
3378 
3379 	do {
3380 		console_may_schedule = 0;
3381 
3382 		flushed = console_flush_all(do_cond_resched, &next_seq, &handover);
3383 		if (!handover)
3384 			__console_unlock();
3385 
3386 		/*
3387 		 * Abort if there was a failure to flush all messages to all
3388 		 * usable consoles. Either it is not possible to flush (in
3389 		 * which case it would be an infinite loop of retrying) or
3390 		 * another context has taken over printing.
3391 		 */
3392 		if (!flushed)
3393 			break;
3394 
3395 		/*
3396 		 * Some context may have added new records after
3397 		 * console_flush_all() but before unlocking the console.
3398 		 * Re-check if there is a new record to flush. If the trylock
3399 		 * fails, another context is already handling the printing.
3400 		 */
3401 	} while (prb_read_valid(prb, next_seq, NULL) && console_trylock());
3402 }
3403 
3404 /**
3405  * console_unlock - unblock the legacy console subsystem from printing
3406  *
3407  * Releases the console_lock which the caller holds to block printing of
3408  * the legacy console subsystem.
3409  *
3410  * While the console_lock was held, console output may have been buffered
3411  * by printk(). If this is the case, console_unlock() emits the output on
3412  * legacy consoles prior to releasing the lock.
3413  *
3414  * console_unlock(); may be called from any context.
3415  */
3416 void console_unlock(void)
3417 {
3418 	struct console_flush_type ft;
3419 
3420 	printk_get_console_flush_type(&ft);
3421 	if (ft.legacy_direct)
3422 		__console_flush_and_unlock();
3423 	else
3424 		__console_unlock();
3425 }
3426 EXPORT_SYMBOL(console_unlock);
3427 
3428 void console_unblank(void)
3429 {
3430 	bool found_unblank = false;
3431 	struct console *c;
3432 	int cookie;
3433 
3434 	/*
3435 	 * First check if there are any consoles implementing the unblank()
3436 	 * callback. If not, there is no reason to continue and take the
3437 	 * console lock, which in particular can be dangerous if
3438 	 * @oops_in_progress is set.
3439 	 */
3440 	cookie = console_srcu_read_lock();
3441 	for_each_console_srcu(c) {
3442 		if (!console_is_usable(c, console_srcu_read_flags(c), true))
3443 			continue;
3444 
3445 		if (c->unblank) {
3446 			found_unblank = true;
3447 			break;
3448 		}
3449 	}
3450 	console_srcu_read_unlock(cookie);
3451 	if (!found_unblank)
3452 		return;
3453 
3454 	/*
3455 	 * Stop console printing because the unblank() callback may
3456 	 * assume the console is not within its write() callback.
3457 	 *
3458 	 * If @oops_in_progress is set, this may be an atomic context.
3459 	 * In that case, attempt a trylock as best-effort.
3460 	 */
3461 	if (oops_in_progress) {
3462 		/* Semaphores are not NMI-safe. */
3463 		if (in_nmi())
3464 			return;
3465 
3466 		/*
3467 		 * Attempting to trylock the console lock can deadlock
3468 		 * if another CPU was stopped while modifying the
3469 		 * semaphore. "Hope and pray" that this is not the
3470 		 * current situation.
3471 		 */
3472 		if (down_trylock_console_sem() != 0)
3473 			return;
3474 	} else
3475 		console_lock();
3476 
3477 	console_locked = 1;
3478 	console_may_schedule = 0;
3479 
3480 	cookie = console_srcu_read_lock();
3481 	for_each_console_srcu(c) {
3482 		if (!console_is_usable(c, console_srcu_read_flags(c), true))
3483 			continue;
3484 
3485 		if (c->unblank)
3486 			c->unblank();
3487 	}
3488 	console_srcu_read_unlock(cookie);
3489 
3490 	console_unlock();
3491 
3492 	if (!oops_in_progress)
3493 		pr_flush(1000, true);
3494 }
3495 
3496 /*
3497  * Rewind all consoles to the oldest available record.
3498  *
3499  * IMPORTANT: The function is safe only when called under
3500  *            console_lock(). It is not enforced because
3501  *            it is used as a best effort in panic().
3502  */
3503 static void __console_rewind_all(void)
3504 {
3505 	struct console *c;
3506 	short flags;
3507 	int cookie;
3508 	u64 seq;
3509 
3510 	seq = prb_first_valid_seq(prb);
3511 
3512 	cookie = console_srcu_read_lock();
3513 	for_each_console_srcu(c) {
3514 		flags = console_srcu_read_flags(c);
3515 
3516 		if (flags & CON_NBCON) {
3517 			nbcon_seq_force(c, seq);
3518 		} else {
3519 			/*
3520 			 * This assignment is safe only when called under
3521 			 * console_lock(). On panic, legacy consoles are
3522 			 * only best effort.
3523 			 */
3524 			c->seq = seq;
3525 		}
3526 	}
3527 	console_srcu_read_unlock(cookie);
3528 }
3529 
3530 /**
3531  * console_flush_on_panic - flush console content on panic
3532  * @mode: flush all messages in buffer or just the pending ones
3533  *
3534  * Immediately output all pending messages no matter what.
3535  */
3536 void console_flush_on_panic(enum con_flush_mode mode)
3537 {
3538 	struct console_flush_type ft;
3539 	bool handover;
3540 	u64 next_seq;
3541 
3542 	/*
3543 	 * Ignore the console lock and flush out the messages. Attempting a
3544 	 * trylock would not be useful because:
3545 	 *
3546 	 *   - if it is contended, it must be ignored anyway
3547 	 *   - console_lock() and console_trylock() block and fail
3548 	 *     respectively in panic for non-panic CPUs
3549 	 *   - semaphores are not NMI-safe
3550 	 */
3551 
3552 	/*
3553 	 * If another context is holding the console lock,
3554 	 * @console_may_schedule might be set. Clear it so that
3555 	 * this context does not call cond_resched() while flushing.
3556 	 */
3557 	console_may_schedule = 0;
3558 
3559 	if (mode == CONSOLE_REPLAY_ALL)
3560 		__console_rewind_all();
3561 
3562 	printk_get_console_flush_type(&ft);
3563 	if (ft.nbcon_atomic)
3564 		nbcon_atomic_flush_pending();
3565 
3566 	/* Flush legacy consoles once allowed, even when dangerous. */
3567 	if (legacy_allow_panic_sync)
3568 		console_flush_all(false, &next_seq, &handover);
3569 }
3570 
3571 /*
3572  * Return the console tty driver structure and its associated index
3573  */
3574 struct tty_driver *console_device(int *index)
3575 {
3576 	struct console *c;
3577 	struct tty_driver *driver = NULL;
3578 	int cookie;
3579 
3580 	/*
3581 	 * Take console_lock to serialize device() callback with
3582 	 * other console operations. For example, fg_console is
3583 	 * modified under console_lock when switching vt.
3584 	 */
3585 	console_lock();
3586 
3587 	cookie = console_srcu_read_lock();
3588 	for_each_console_srcu(c) {
3589 		if (!c->device)
3590 			continue;
3591 		driver = c->device(c, index);
3592 		if (driver)
3593 			break;
3594 	}
3595 	console_srcu_read_unlock(cookie);
3596 
3597 	console_unlock();
3598 	return driver;
3599 }
3600 
3601 /*
3602  * Prevent further output on the passed console device so that (for example)
3603  * serial drivers can suspend console output before suspending a port, and can
3604  * re-enable output afterwards.
3605  */
3606 void console_suspend(struct console *console)
3607 {
3608 	__pr_flush(console, 1000, true);
3609 	console_list_lock();
3610 	console_srcu_write_flags(console, console->flags & ~CON_ENABLED);
3611 	console_list_unlock();
3612 
3613 	/*
3614 	 * Ensure that all SRCU list walks have completed. All contexts must
3615 	 * be able to see that this console is disabled so that (for example)
3616 	 * the caller can suspend the port without risk of another context
3617 	 * using the port.
3618 	 */
3619 	synchronize_srcu(&console_srcu);
3620 }
3621 EXPORT_SYMBOL(console_suspend);
3622 
3623 void console_resume(struct console *console)
3624 {
3625 	struct console_flush_type ft;
3626 	bool is_nbcon;
3627 
3628 	console_list_lock();
3629 	console_srcu_write_flags(console, console->flags | CON_ENABLED);
3630 	is_nbcon = console->flags & CON_NBCON;
3631 	console_list_unlock();
3632 
3633 	/*
3634 	 * Ensure that all SRCU list walks have completed. The related
3635 	 * printing context must be able to see it is enabled so that
3636 	 * it is guaranteed to wake up and resume printing.
3637 	 */
3638 	synchronize_srcu(&console_srcu);
3639 
3640 	printk_get_console_flush_type(&ft);
3641 	if (is_nbcon && ft.nbcon_offload)
3642 		nbcon_kthread_wake(console);
3643 	else if (ft.legacy_offload)
3644 		defer_console_output();
3645 
3646 	__pr_flush(console, 1000, true);
3647 }
3648 EXPORT_SYMBOL(console_resume);
3649 
3650 #ifdef CONFIG_PRINTK
3651 static int unregister_console_locked(struct console *console);
3652 
3653 /* True when system boot is far enough to create printer threads. */
3654 bool printk_kthreads_ready __ro_after_init;
3655 
3656 static struct task_struct *printk_legacy_kthread;
3657 
3658 static bool legacy_kthread_should_wakeup(void)
3659 {
3660 	struct console_flush_type ft;
3661 	struct console *con;
3662 	bool ret = false;
3663 	int cookie;
3664 
3665 	if (kthread_should_stop())
3666 		return true;
3667 
3668 	printk_get_console_flush_type(&ft);
3669 
3670 	cookie = console_srcu_read_lock();
3671 	for_each_console_srcu(con) {
3672 		short flags = console_srcu_read_flags(con);
3673 		u64 printk_seq;
3674 
3675 		/*
3676 		 * The legacy printer thread is only responsible for nbcon
3677 		 * consoles when the nbcon consoles cannot print via their
3678 		 * atomic or threaded flushing.
3679 		 */
3680 		if ((flags & CON_NBCON) && (ft.nbcon_atomic || ft.nbcon_offload))
3681 			continue;
3682 
3683 		if (!console_is_usable(con, flags, false))
3684 			continue;
3685 
3686 		if (flags & CON_NBCON) {
3687 			printk_seq = nbcon_seq_read(con);
3688 		} else {
3689 			/*
3690 			 * It is safe to read @seq because only this
3691 			 * thread context updates @seq.
3692 			 */
3693 			printk_seq = con->seq;
3694 		}
3695 
3696 		if (prb_read_valid(prb, printk_seq, NULL)) {
3697 			ret = true;
3698 			break;
3699 		}
3700 	}
3701 	console_srcu_read_unlock(cookie);
3702 
3703 	return ret;
3704 }
3705 
3706 static int legacy_kthread_func(void *unused)
3707 {
3708 	bool try_again;
3709 
3710 wait_for_event:
3711 	wait_event_interruptible(legacy_wait, legacy_kthread_should_wakeup());
3712 
3713 	do {
3714 		bool handover = false;
3715 		u64 next_seq = 0;
3716 
3717 		if (kthread_should_stop())
3718 			return 0;
3719 
3720 		console_lock();
3721 		console_flush_one_record(true, &next_seq, &handover, &try_again);
3722 		if (!handover)
3723 			__console_unlock();
3724 
3725 	} while (try_again);
3726 
3727 	goto wait_for_event;
3728 }
3729 
3730 static bool legacy_kthread_create(void)
3731 {
3732 	struct task_struct *kt;
3733 
3734 	lockdep_assert_console_list_lock_held();
3735 
3736 	kt = kthread_run(legacy_kthread_func, NULL, "pr/legacy");
3737 	if (IS_ERR(kt)) {
3738 		pr_err("failed to start legacy printing thread\n");
3739 		return false;
3740 	}
3741 
3742 	printk_legacy_kthread = kt;
3743 
3744 	/*
3745 	 * It is important that console printing threads are scheduled
3746 	 * shortly after a printk call and with generous runtime budgets.
3747 	 */
3748 	sched_set_normal(printk_legacy_kthread, -20);
3749 
3750 	return true;
3751 }
3752 
3753 /**
3754  * printk_kthreads_shutdown - shutdown all threaded printers
3755  * @data: syscore context
3756  *
3757  * On system shutdown all threaded printers are stopped. This allows printk
3758  * to transition back to atomic printing, thus providing a robust mechanism
3759  * for the final shutdown/reboot messages to be output.
3760  */
3761 static void printk_kthreads_shutdown(void *data)
3762 {
3763 	struct console *con;
3764 
3765 	console_list_lock();
3766 	if (printk_kthreads_running) {
3767 		printk_kthreads_running = false;
3768 
3769 		for_each_console(con) {
3770 			if (con->flags & CON_NBCON)
3771 				nbcon_kthread_stop(con);
3772 		}
3773 
3774 		/*
3775 		 * The threads may have been stopped while printing a
3776 		 * backlog. Flush any records left over.
3777 		 */
3778 		nbcon_atomic_flush_pending();
3779 	}
3780 	console_list_unlock();
3781 }
3782 
3783 static const struct syscore_ops printk_syscore_ops = {
3784 	.shutdown = printk_kthreads_shutdown,
3785 };
3786 
3787 static struct syscore printk_syscore = {
3788 	.ops = &printk_syscore_ops,
3789 };
3790 
3791 /*
3792  * If appropriate, start nbcon kthreads and set @printk_kthreads_running.
3793  * If any kthreads fail to start, those consoles are unregistered.
3794  *
3795  * Called after consoles are registered or unregistered.
3796  *
3797  * This MUST NOT be called within an hlist_for_each_entry_safe(console_list)
3798  * loop because it can unregister multiple consoles if any kthreads fail to
3799  * start. That would lead to use-after-free scenarios within the
3800  * hlist_for_each_entry_safe(console_list) loop of the caller.
3801  *
3802  * Must be called under console_list_lock().
3803  */
3804 static void printk_kthreads_check_locked(void)
3805 {
3806 	struct hlist_node *tmp;
3807 	struct console *con;
3808 
3809 	lockdep_assert_console_list_lock_held();
3810 
3811 	if (!printk_kthreads_ready)
3812 		return;
3813 
3814 	/* Start or stop the legacy kthread when needed. */
3815 	if (have_legacy_console || have_boot_console) {
3816 		if (!printk_legacy_kthread &&
3817 		    force_legacy_kthread() &&
3818 		    !legacy_kthread_create()) {
3819 			/*
3820 			 * All legacy consoles must be unregistered. If there
3821 			 * are any nbcon consoles, they will set up their own
3822 			 * kthread.
3823 			 */
3824 			hlist_for_each_entry_safe(con, tmp, &console_list, node) {
3825 				if (con->flags & CON_NBCON)
3826 					continue;
3827 
3828 				unregister_console_locked(con);
3829 			}
3830 		}
3831 	} else if (printk_legacy_kthread) {
3832 		kthread_stop(printk_legacy_kthread);
3833 		printk_legacy_kthread = NULL;
3834 	}
3835 
3836 	/*
3837 	 * Printer threads cannot be started as long as any boot console is
3838 	 * registered because there is no way to synchronize the hardware
3839 	 * registers between boot console code and regular console code.
3840 	 * It can only be known that there will be no new boot consoles when
3841 	 * an nbcon console is registered.
3842 	 */
3843 	if (have_boot_console || !have_nbcon_console) {
3844 		/* Clear flag in case all nbcon consoles unregistered. */
3845 		printk_kthreads_running = false;
3846 		return;
3847 	}
3848 
3849 	if (printk_kthreads_running)
3850 		return;
3851 
3852 	hlist_for_each_entry_safe(con, tmp, &console_list, node) {
3853 		if (!(con->flags & CON_NBCON))
3854 			continue;
3855 
3856 		if (!nbcon_kthread_create(con))
3857 			unregister_console_locked(con);
3858 	}
3859 
3860 	printk_kthreads_running = true;
3861 }
3862 
3863 static int __init printk_set_kthreads_ready(void)
3864 {
3865 	register_syscore(&printk_syscore);
3866 
3867 	console_list_lock();
3868 	printk_kthreads_ready = true;
3869 	printk_kthreads_check_locked();
3870 	console_list_unlock();
3871 
3872 	return 0;
3873 }
3874 subsys_initcall(printk_set_kthreads_ready);
3875 #endif /* CONFIG_PRINTK */
3876 
3877 static int __read_mostly keep_bootcon;
3878 
3879 static int __init keep_bootcon_setup(char *str)
3880 {
3881 	keep_bootcon = 1;
3882 	pr_info("debug: skip boot console de-registration.\n");
3883 
3884 	return 0;
3885 }
3886 
3887 early_param("keep_bootcon", keep_bootcon_setup);
3888 
3889 static int console_call_setup(struct console *newcon, char *options)
3890 {
3891 	int err;
3892 
3893 	if (!newcon->setup)
3894 		return 0;
3895 
3896 	/* Synchronize with possible boot console. */
3897 	console_lock();
3898 	err = newcon->setup(newcon, options);
3899 	console_unlock();
3900 
3901 	return err;
3902 }
3903 
3904 /*
3905  * This is called by register_console() to try to match
3906  * the newly registered console with any of the ones selected
3907  * by either the command line or add_preferred_console() and
3908  * setup/enable it.
3909  */
3910 static int try_enable_preferred_console(struct console *newcon,
3911 					bool user_specified)
3912 {
3913 	struct console_cmdline *c;
3914 	int i, err;
3915 
3916 	for (i = 0, c = console_cmdline;
3917 	     i < MAX_CMDLINECONSOLES && (c->name[0] || c->devname[0]);
3918 	     i++, c++) {
3919 		/* Console not yet initialized? */
3920 		if (!c->name[0])
3921 			continue;
3922 		if (c->user_specified != user_specified)
3923 			continue;
3924 		if (!newcon->match ||
3925 		    newcon->match(newcon, c->name, c->index, c->options) != 0) {
3926 			/* default matching */
3927 			BUILD_BUG_ON(sizeof(c->name) != sizeof(newcon->name));
3928 			if (strcmp(c->name, newcon->name) != 0)
3929 				continue;
3930 			if (newcon->index >= 0 &&
3931 			    newcon->index != c->index)
3932 				continue;
3933 			if (newcon->index < 0)
3934 				newcon->index = c->index;
3935 
3936 			if (_braille_register_console(newcon, c))
3937 				return 0;
3938 
3939 			err = console_call_setup(newcon, c->options);
3940 			if (err)
3941 				return err;
3942 		}
3943 		newcon->flags |= CON_ENABLED;
3944 		if (i == preferred_console)
3945 			newcon->flags |= CON_CONSDEV;
3946 		return 0;
3947 	}
3948 
3949 	return -ENOENT;
3950 }
3951 
3952 /* Try to enable the console unconditionally */
3953 static void try_enable_default_console(struct console *newcon)
3954 {
3955 	if (newcon->index < 0)
3956 		newcon->index = 0;
3957 
3958 	if (console_call_setup(newcon, NULL) != 0)
3959 		return;
3960 
3961 	newcon->flags |= CON_ENABLED;
3962 
3963 	if (newcon->device)
3964 		newcon->flags |= CON_CONSDEV;
3965 }
3966 
3967 /* Return the starting sequence number for a newly registered console. */
3968 static u64 get_init_console_seq(struct console *newcon, bool bootcon_registered)
3969 {
3970 	struct console *con;
3971 	bool handover;
3972 	u64 init_seq;
3973 
3974 	if (newcon->flags & (CON_PRINTBUFFER | CON_BOOT)) {
3975 		/* Get a consistent copy of @syslog_seq. */
3976 		mutex_lock(&syslog_lock);
3977 		init_seq = syslog_seq;
3978 		mutex_unlock(&syslog_lock);
3979 	} else {
3980 		/* Begin with next message added to ringbuffer. */
3981 		init_seq = prb_next_seq(prb);
3982 
3983 		/*
3984 		 * If any enabled boot consoles are due to be unregistered
3985 		 * shortly, some may not be caught up and may be the same
3986 		 * device as @newcon. Since it is not known which boot console
3987 		 * is the same device, flush all consoles and, if necessary,
3988 		 * start with the message of the enabled boot console that is
3989 		 * the furthest behind.
3990 		 */
3991 		if (bootcon_registered && !keep_bootcon) {
3992 			/*
3993 			 * Hold the console_lock to stop console printing and
3994 			 * guarantee safe access to console->seq.
3995 			 */
3996 			console_lock();
3997 
3998 			/*
3999 			 * Flush all consoles and set the console to start at
4000 			 * the next unprinted sequence number.
4001 			 */
4002 			if (!console_flush_all(true, &init_seq, &handover)) {
4003 				/*
4004 				 * Flushing failed. Just choose the lowest
4005 				 * sequence of the enabled boot consoles.
4006 				 */
4007 
4008 				/*
4009 				 * If there was a handover, this context no
4010 				 * longer holds the console_lock.
4011 				 */
4012 				if (handover)
4013 					console_lock();
4014 
4015 				init_seq = prb_next_seq(prb);
4016 				for_each_console(con) {
4017 					u64 seq;
4018 
4019 					if (!(con->flags & CON_BOOT) ||
4020 					    !(con->flags & CON_ENABLED)) {
4021 						continue;
4022 					}
4023 
4024 					if (con->flags & CON_NBCON)
4025 						seq = nbcon_seq_read(con);
4026 					else
4027 						seq = con->seq;
4028 
4029 					if (seq < init_seq)
4030 						init_seq = seq;
4031 				}
4032 			}
4033 
4034 			console_unlock();
4035 		}
4036 	}
4037 
4038 	return init_seq;
4039 }
4040 
4041 #define console_first()				\
4042 	hlist_entry(console_list.first, struct console, node)
4043 
4044 static int unregister_console_locked(struct console *console);
4045 
4046 /*
4047  * The console driver calls this routine during kernel initialization
4048  * to register the console printing procedure with printk() and to
4049  * print any messages that were printed by the kernel before the
4050  * console driver was initialized.
4051  *
4052  * This can happen pretty early during the boot process (because of
4053  * early_printk) - sometimes before setup_arch() completes - be careful
4054  * of what kernel features are used - they may not be initialised yet.
4055  *
4056  * There are two types of consoles - bootconsoles (early_printk) and
4057  * "real" consoles (everything which is not a bootconsole) which are
4058  * handled differently.
4059  *  - Any number of bootconsoles can be registered at any time.
4060  *  - As soon as a "real" console is registered, all bootconsoles
4061  *    will be unregistered automatically.
4062  *  - Once a "real" console is registered, any attempt to register a
4063  *    bootconsoles will be rejected
4064  */
4065 void register_console(struct console *newcon)
4066 {
4067 	bool use_device_lock = (newcon->flags & CON_NBCON) && newcon->write_atomic;
4068 	bool bootcon_registered = false;
4069 	bool realcon_registered = false;
4070 	struct console *con;
4071 	unsigned long flags;
4072 	u64 init_seq;
4073 	int err;
4074 
4075 	console_list_lock();
4076 
4077 	for_each_console(con) {
4078 		if (WARN(con == newcon, "console '%s%d' already registered\n",
4079 					 con->name, con->index)) {
4080 			goto unlock;
4081 		}
4082 
4083 		if (con->flags & CON_BOOT)
4084 			bootcon_registered = true;
4085 		else
4086 			realcon_registered = true;
4087 	}
4088 
4089 	/* Do not register boot consoles when there already is a real one. */
4090 	if ((newcon->flags & CON_BOOT) && realcon_registered) {
4091 		pr_info("Too late to register bootconsole %s%d\n",
4092 			newcon->name, newcon->index);
4093 		goto unlock;
4094 	}
4095 
4096 	if (newcon->flags & CON_NBCON) {
4097 		/*
4098 		 * Ensure the nbcon console buffers can be allocated
4099 		 * before modifying any global data.
4100 		 */
4101 		if (!nbcon_alloc(newcon))
4102 			goto unlock;
4103 	}
4104 
4105 	/*
4106 	 * See if we want to enable this console driver by default.
4107 	 *
4108 	 * Nope when a console is preferred by the command line, device
4109 	 * tree, or SPCR.
4110 	 *
4111 	 * The first real console with tty binding (driver) wins. More
4112 	 * consoles might get enabled before the right one is found.
4113 	 *
4114 	 * Note that a console with tty binding will have CON_CONSDEV
4115 	 * flag set and will be first in the list.
4116 	 */
4117 	if (preferred_console < 0) {
4118 		if (hlist_empty(&console_list) || !console_first()->device ||
4119 		    console_first()->flags & CON_BOOT) {
4120 			try_enable_default_console(newcon);
4121 		}
4122 	}
4123 
4124 	/* See if this console matches one we selected on the command line */
4125 	err = try_enable_preferred_console(newcon, true);
4126 
4127 	/* If not, try to match against the platform default(s) */
4128 	if (err == -ENOENT)
4129 		err = try_enable_preferred_console(newcon, false);
4130 
4131 	/*
4132 	 * Some consoles, such as pstore and netconsole, can be enabled even
4133 	 * without matching. Accept them at this stage when they had a chance
4134 	 * to match() and call setup().
4135 	 */
4136 	if (err == -ENOENT && (newcon->flags & CON_ENABLED))
4137 		err = 0;
4138 
4139 	/* printk() messages are not printed to the Braille console. */
4140 	if (err || newcon->flags & CON_BRL) {
4141 		if (newcon->flags & CON_NBCON)
4142 			nbcon_free(newcon);
4143 		goto unlock;
4144 	}
4145 
4146 	/*
4147 	 * If we have a bootconsole, and are switching to a real console,
4148 	 * don't print everything out again, since when the boot console, and
4149 	 * the real console are the same physical device, it's annoying to
4150 	 * see the beginning boot messages twice
4151 	 */
4152 	if (bootcon_registered &&
4153 	    ((newcon->flags & (CON_CONSDEV | CON_BOOT)) == CON_CONSDEV)) {
4154 		newcon->flags &= ~CON_PRINTBUFFER;
4155 	}
4156 
4157 	newcon->dropped = 0;
4158 	init_seq = get_init_console_seq(newcon, bootcon_registered);
4159 
4160 	if (newcon->flags & CON_NBCON) {
4161 		have_nbcon_console = true;
4162 		nbcon_seq_force(newcon, init_seq);
4163 	} else {
4164 		have_legacy_console = true;
4165 		newcon->seq = init_seq;
4166 	}
4167 
4168 	if (newcon->flags & CON_BOOT)
4169 		have_boot_console = true;
4170 
4171 	/*
4172 	 * If another context is actively using the hardware of this new
4173 	 * console, it will not be aware of the nbcon synchronization. This
4174 	 * is a risk that two contexts could access the hardware
4175 	 * simultaneously if this new console is used for atomic printing
4176 	 * and the other context is still using the hardware.
4177 	 *
4178 	 * Use the driver synchronization to ensure that the hardware is not
4179 	 * in use while this new console transitions to being registered.
4180 	 */
4181 	if (use_device_lock)
4182 		newcon->device_lock(newcon, &flags);
4183 
4184 	/*
4185 	 * Put this console in the list - keep the
4186 	 * preferred driver at the head of the list.
4187 	 */
4188 	if (hlist_empty(&console_list)) {
4189 		/* Ensure CON_CONSDEV is always set for the head. */
4190 		newcon->flags |= CON_CONSDEV;
4191 		hlist_add_head_rcu(&newcon->node, &console_list);
4192 
4193 	} else if (newcon->flags & CON_CONSDEV) {
4194 		/* Only the new head can have CON_CONSDEV set. */
4195 		console_srcu_write_flags(console_first(), console_first()->flags & ~CON_CONSDEV);
4196 		hlist_add_head_rcu(&newcon->node, &console_list);
4197 
4198 	} else {
4199 		hlist_add_behind_rcu(&newcon->node, console_list.first);
4200 	}
4201 
4202 	/*
4203 	 * No need to synchronize SRCU here! The caller does not rely
4204 	 * on all contexts being able to see the new console before
4205 	 * register_console() completes.
4206 	 */
4207 
4208 	/* This new console is now registered. */
4209 	if (use_device_lock)
4210 		newcon->device_unlock(newcon, flags);
4211 
4212 	console_sysfs_notify();
4213 
4214 	/*
4215 	 * By unregistering the bootconsoles after we enable the real console
4216 	 * we get the "console xxx enabled" message on all the consoles -
4217 	 * boot consoles, real consoles, etc - this is to ensure that end
4218 	 * users know there might be something in the kernel's log buffer that
4219 	 * went to the bootconsole (that they do not see on the real console)
4220 	 */
4221 	con_printk(KERN_INFO, newcon, "enabled\n");
4222 	if (bootcon_registered &&
4223 	    ((newcon->flags & (CON_CONSDEV | CON_BOOT)) == CON_CONSDEV) &&
4224 	    !keep_bootcon) {
4225 		struct hlist_node *tmp;
4226 
4227 		hlist_for_each_entry_safe(con, tmp, &console_list, node) {
4228 			if (con->flags & CON_BOOT)
4229 				unregister_console_locked(con);
4230 		}
4231 	}
4232 
4233 	/* Changed console list, may require printer threads to start/stop. */
4234 	printk_kthreads_check_locked();
4235 unlock:
4236 	console_list_unlock();
4237 }
4238 EXPORT_SYMBOL(register_console);
4239 
4240 /*
4241  * Unregister the specified console.
4242  *
4243  * This can be called safely from within an
4244  * hlist_for_each_entry_safe(console_list) loop because it unregisters
4245  * _only_ the specified console.
4246  *
4247  * Afterwards the caller should call printk_kthreads_check_locked() since
4248  * unregistering a console may require printer threads to start/stop. Note
4249  * that printk_kthreads_check_locked() cannot be called safely from within
4250  * an hlist_for_each_entry_safe(console_list) loop.
4251  *
4252  * Must be called under console_list_lock().
4253  */
4254 static int unregister_console_locked(struct console *console)
4255 {
4256 	bool use_device_lock = (console->flags & CON_NBCON) && console->write_atomic;
4257 	bool found_legacy_con = false;
4258 	bool found_nbcon_con = false;
4259 	bool found_boot_con = false;
4260 	unsigned long flags;
4261 	struct console *c;
4262 	int res;
4263 
4264 	lockdep_assert_console_list_lock_held();
4265 
4266 	con_printk(KERN_INFO, console, "disabled\n");
4267 
4268 	res = _braille_unregister_console(console);
4269 	if (res < 0)
4270 		return res;
4271 	if (res > 0)
4272 		return 0;
4273 
4274 	if (!console_is_registered_locked(console))
4275 		res = -ENODEV;
4276 	else if (console_is_usable(console, console->flags, true))
4277 		__pr_flush(console, 1000, true);
4278 
4279 	/* Disable it unconditionally */
4280 	console_srcu_write_flags(console, console->flags & ~CON_ENABLED);
4281 
4282 	if (res < 0)
4283 		return res;
4284 
4285 	/*
4286 	 * Use the driver synchronization to ensure that the hardware is not
4287 	 * in use while this console transitions to being unregistered.
4288 	 */
4289 	if (use_device_lock)
4290 		console->device_lock(console, &flags);
4291 
4292 	hlist_del_init_rcu(&console->node);
4293 
4294 	if (use_device_lock)
4295 		console->device_unlock(console, flags);
4296 
4297 	/*
4298 	 * <HISTORICAL>
4299 	 * If this isn't the last console and it has CON_CONSDEV set, we
4300 	 * need to set it on the next preferred console.
4301 	 * </HISTORICAL>
4302 	 *
4303 	 * The above makes no sense as there is no guarantee that the next
4304 	 * console has any device attached. Oh well....
4305 	 */
4306 	if (!hlist_empty(&console_list) && console->flags & CON_CONSDEV)
4307 		console_srcu_write_flags(console_first(), console_first()->flags | CON_CONSDEV);
4308 
4309 	/*
4310 	 * Ensure that all SRCU list walks have completed. All contexts
4311 	 * must not be able to see this console in the list so that any
4312 	 * exit/cleanup routines can be performed safely.
4313 	 */
4314 	synchronize_srcu(&console_srcu);
4315 
4316 	/*
4317 	 * With this console gone, the global flags tracking registered
4318 	 * console types may have changed. Update them.
4319 	 */
4320 	for_each_console(c) {
4321 		if (c->flags & CON_BOOT)
4322 			found_boot_con = true;
4323 
4324 		if (c->flags & CON_NBCON)
4325 			found_nbcon_con = true;
4326 		else
4327 			found_legacy_con = true;
4328 	}
4329 	if (!found_boot_con)
4330 		have_boot_console = found_boot_con;
4331 	if (!found_legacy_con)
4332 		have_legacy_console = found_legacy_con;
4333 	if (!found_nbcon_con)
4334 		have_nbcon_console = found_nbcon_con;
4335 
4336 	/* @have_nbcon_console must be updated before calling nbcon_free(). */
4337 	if (console->flags & CON_NBCON)
4338 		nbcon_free(console);
4339 
4340 	console_sysfs_notify();
4341 
4342 	if (console->exit)
4343 		res = console->exit(console);
4344 
4345 	return res;
4346 }
4347 
4348 int unregister_console(struct console *console)
4349 {
4350 	int res;
4351 
4352 	console_list_lock();
4353 
4354 	res = unregister_console_locked(console);
4355 
4356 	/* Changed console list, may require printer threads to start/stop. */
4357 	printk_kthreads_check_locked();
4358 
4359 	console_list_unlock();
4360 	return res;
4361 }
4362 EXPORT_SYMBOL(unregister_console);
4363 
4364 /**
4365  * console_force_preferred_locked - force a registered console preferred
4366  * @con: The registered console to force preferred.
4367  *
4368  * Must be called under console_list_lock().
4369  */
4370 void console_force_preferred_locked(struct console *con)
4371 {
4372 	struct console *cur_pref_con;
4373 
4374 	if (!console_is_registered_locked(con))
4375 		return;
4376 
4377 	cur_pref_con = console_first();
4378 
4379 	/* Already preferred? */
4380 	if (cur_pref_con == con)
4381 		return;
4382 
4383 	/*
4384 	 * Delete, but do not re-initialize the entry. This allows the console
4385 	 * to continue to appear registered (via any hlist_unhashed_lockless()
4386 	 * checks), even though it was briefly removed from the console list.
4387 	 */
4388 	hlist_del_rcu(&con->node);
4389 
4390 	/*
4391 	 * Ensure that all SRCU list walks have completed so that the console
4392 	 * can be added to the beginning of the console list and its forward
4393 	 * list pointer can be re-initialized.
4394 	 */
4395 	synchronize_srcu(&console_srcu);
4396 
4397 	con->flags |= CON_CONSDEV;
4398 	WARN_ON(!con->device);
4399 
4400 	/* Only the new head can have CON_CONSDEV set. */
4401 	console_srcu_write_flags(cur_pref_con, cur_pref_con->flags & ~CON_CONSDEV);
4402 	hlist_add_head_rcu(&con->node, &console_list);
4403 }
4404 EXPORT_SYMBOL(console_force_preferred_locked);
4405 
4406 /*
4407  * Initialize the console device. This is called *early*, so
4408  * we can't necessarily depend on lots of kernel help here.
4409  * Just do some early initializations, and do the complex setup
4410  * later.
4411  */
4412 void __init console_init(void)
4413 {
4414 	int ret;
4415 	initcall_t call;
4416 	initcall_entry_t *ce;
4417 
4418 #ifdef CONFIG_NULL_TTY_DEFAULT_CONSOLE
4419 	if (!console_set_on_cmdline)
4420 		add_preferred_console("ttynull", 0, NULL);
4421 #endif
4422 
4423 	/* Setup the default TTY line discipline. */
4424 	n_tty_init();
4425 
4426 	/*
4427 	 * set up the console device so that later boot sequences can
4428 	 * inform about problems etc..
4429 	 */
4430 	ce = __con_initcall_start;
4431 	trace_initcall_level("console");
4432 	while (ce < __con_initcall_end) {
4433 		call = initcall_from_entry(ce);
4434 		trace_initcall_start(call);
4435 		ret = call();
4436 		trace_initcall_finish(call, ret);
4437 		ce++;
4438 	}
4439 }
4440 
4441 /*
4442  * Some boot consoles access data that is in the init section and which will
4443  * be discarded after the initcalls have been run. To make sure that no code
4444  * will access this data, unregister the boot consoles in a late initcall.
4445  *
4446  * If for some reason, such as deferred probe or the driver being a loadable
4447  * module, the real console hasn't registered yet at this point, there will
4448  * be a brief interval in which no messages are logged to the console, which
4449  * makes it difficult to diagnose problems that occur during this time.
4450  *
4451  * If a real console has already registered, remove all remaining boot consoles.
4452  * Otherwise, mitigate the no-console interval by removing only boot consoles
4453  * whose memory intersects with the init section.
4454  */
4455 static int __init printk_late_init(void)
4456 {
4457 	bool have_real_console = false;
4458 	struct hlist_node *tmp;
4459 	struct console *con;
4460 	int ret;
4461 
4462 	console_list_lock();
4463 	for_each_console(con) {
4464 		if (!(con->flags & CON_BOOT)) {
4465 			have_real_console = true;
4466 			break;
4467 		}
4468 	}
4469 
4470 	hlist_for_each_entry_safe(con, tmp, &console_list, node) {
4471 		if (!(con->flags & CON_BOOT))
4472 			continue;
4473 
4474 		/*
4475 		 * Keep the boot console when requested or as a fallback
4476 		 * unless it is using an init section.
4477 		 */
4478 		if (keep_bootcon || !have_real_console) {
4479 			if (!init_section_intersects(con, sizeof(*con)) &&
4480 			    !init_section_contains(con->write, 0) &&
4481 			    !init_section_contains(con->read, 0) &&
4482 			    !init_section_contains(con->device, 0) &&
4483 			    !init_section_contains(con->unblank, 0) &&
4484 			    !init_section_contains(con->data, 0))
4485 				continue;
4486 
4487 			/*
4488 			 * Please, consider moving the reported consoles out
4489 			 * of the init section.
4490 			 */
4491 			pr_warn("bootconsole [%s%d] uses init memory and must be disabled even before the real one is ready\n",
4492 				con->name, con->index);
4493 		}
4494 
4495 		unregister_console_locked(con);
4496 	}
4497 
4498 	/* Changed console list, may require printer threads to start/stop. */
4499 	printk_kthreads_check_locked();
4500 
4501 	console_list_unlock();
4502 
4503 	ret = cpuhp_setup_state_nocalls(CPUHP_PRINTK_DEAD, "printk:dead", NULL,
4504 					console_cpu_notify);
4505 	WARN_ON(ret < 0);
4506 	ret = cpuhp_setup_state_nocalls(CPUHP_AP_ONLINE_DYN, "printk:online",
4507 					console_cpu_notify, NULL);
4508 	WARN_ON(ret < 0);
4509 	printk_sysctl_init();
4510 	return 0;
4511 }
4512 late_initcall(printk_late_init);
4513 
4514 #if defined CONFIG_PRINTK
4515 /* If @con is specified, only wait for that console. Otherwise wait for all. */
4516 static bool __pr_flush(struct console *con, int timeout_ms, bool reset_on_progress)
4517 {
4518 	unsigned long timeout_jiffies = msecs_to_jiffies(timeout_ms);
4519 	unsigned long remaining_jiffies = timeout_jiffies;
4520 	struct console_flush_type ft;
4521 	struct console *c;
4522 	u64 last_diff = 0;
4523 	u64 printk_seq;
4524 	short flags;
4525 	int cookie;
4526 	u64 diff;
4527 	u64 seq;
4528 
4529 	/* Sorry, pr_flush() will not work this early. */
4530 	if (system_state < SYSTEM_SCHEDULING)
4531 		return false;
4532 
4533 	might_sleep();
4534 
4535 	seq = prb_next_reserve_seq(prb);
4536 
4537 	/* Flush the consoles so that records up to @seq are printed. */
4538 	printk_get_console_flush_type(&ft);
4539 	if (ft.nbcon_atomic)
4540 		nbcon_atomic_flush_pending();
4541 	if (ft.legacy_direct) {
4542 		console_lock();
4543 		console_unlock();
4544 	}
4545 
4546 	for (;;) {
4547 		unsigned long begin_jiffies;
4548 		unsigned long slept_jiffies;
4549 
4550 		diff = 0;
4551 
4552 		/*
4553 		 * Hold the console_lock to guarantee safe access to
4554 		 * console->seq. Releasing console_lock flushes more
4555 		 * records in case @seq is still not printed on all
4556 		 * usable consoles.
4557 		 *
4558 		 * Holding the console_lock is not necessary if there
4559 		 * are no legacy or boot consoles. However, such a
4560 		 * console could register at any time. Always hold the
4561 		 * console_lock as a precaution rather than
4562 		 * synchronizing against register_console().
4563 		 */
4564 		console_lock();
4565 
4566 		cookie = console_srcu_read_lock();
4567 		for_each_console_srcu(c) {
4568 			if (con && con != c)
4569 				continue;
4570 
4571 			flags = console_srcu_read_flags(c);
4572 
4573 			/*
4574 			 * If consoles are not usable, it cannot be expected
4575 			 * that they make forward progress, so only increment
4576 			 * @diff for usable consoles.
4577 			 */
4578 			if (!console_is_usable(c, flags, true) &&
4579 			    !console_is_usable(c, flags, false)) {
4580 				continue;
4581 			}
4582 
4583 			if (flags & CON_NBCON) {
4584 				printk_seq = nbcon_seq_read(c);
4585 			} else {
4586 				printk_seq = c->seq;
4587 			}
4588 
4589 			if (printk_seq < seq)
4590 				diff += seq - printk_seq;
4591 		}
4592 		console_srcu_read_unlock(cookie);
4593 
4594 		if (diff != last_diff && reset_on_progress)
4595 			remaining_jiffies = timeout_jiffies;
4596 
4597 		console_unlock();
4598 
4599 		/* Note: @diff is 0 if there are no usable consoles. */
4600 		if (diff == 0 || remaining_jiffies == 0)
4601 			break;
4602 
4603 		/* msleep(1) might sleep much longer. Check time by jiffies. */
4604 		begin_jiffies = jiffies;
4605 		msleep(1);
4606 		slept_jiffies = jiffies - begin_jiffies;
4607 
4608 		remaining_jiffies -= min(slept_jiffies, remaining_jiffies);
4609 
4610 		last_diff = diff;
4611 	}
4612 
4613 	return (diff == 0);
4614 }
4615 
4616 /**
4617  * pr_flush() - Wait for printing threads to catch up.
4618  *
4619  * @timeout_ms:        The maximum time (in ms) to wait.
4620  * @reset_on_progress: Reset the timeout if forward progress is seen.
4621  *
4622  * A value of 0 for @timeout_ms means no waiting will occur. A value of -1
4623  * represents infinite waiting.
4624  *
4625  * If @reset_on_progress is true, the timeout will be reset whenever any
4626  * printer has been seen to make some forward progress.
4627  *
4628  * Context: Process context. May sleep while acquiring console lock.
4629  * Return: true if all usable printers are caught up.
4630  */
4631 bool pr_flush(int timeout_ms, bool reset_on_progress)
4632 {
4633 	return __pr_flush(NULL, timeout_ms, reset_on_progress);
4634 }
4635 
4636 /*
4637  * Delayed printk version, for scheduler-internal messages:
4638  */
4639 #define PRINTK_PENDING_OUTPUT	0x01
4640 
4641 static void pending_wake_fn(struct irq_work *irq_work)
4642 {
4643 	wake_up_interruptible(&log_wait);
4644 }
4645 
4646 static void pending_output_fn(struct irq_work *irq_work)
4647 {
4648 	if (force_legacy_kthread()) {
4649 		if (printk_legacy_kthread)
4650 			wake_up_interruptible(&legacy_wait);
4651 	} else {
4652 		if (console_trylock())
4653 			console_unlock();
4654 	}
4655 }
4656 
4657 static struct irq_work pending_wakeup_work = IRQ_WORK_INIT_LAZY(pending_wake_fn);
4658 static struct irq_work pending_output_work = IRQ_WORK_INIT_LAZY(pending_output_fn);
4659 
4660 static void __wake_up_klogd(int val)
4661 {
4662 	if (!printk_percpu_data_ready())
4663 		return;
4664 
4665 	/*
4666 	 * It is not allowed to call this function when console irq_work
4667 	 * is blocked.
4668 	 */
4669 	if (WARN_ON_ONCE(console_irqwork_blocked))
4670 		return;
4671 
4672 	/*
4673 	 * Guarantee any new records can be seen by tasks preparing to wait
4674 	 * before this context checks if the wait queue is empty.
4675 	 *
4676 	 * The full memory barrier within wq_has_sleeper() pairs with the full
4677 	 * memory barrier within set_current_state() of
4678 	 * prepare_to_wait_event(), which is called after ___wait_event() adds
4679 	 * the waiter but before it has checked the wait condition.
4680 	 *
4681 	 * This pairs with devkmsg_read:A and syslog_print:A.
4682 	 */
4683 	if (wq_has_sleeper(&log_wait)) /* LMM(__wake_up_klogd:A) */
4684 		irq_work_queue(&pending_wakeup_work);
4685 
4686 	if (val & PRINTK_PENDING_OUTPUT)
4687 		irq_work_queue(&pending_output_work);
4688 }
4689 
4690 /**
4691  * wake_up_klogd - Wake kernel logging daemon
4692  *
4693  * Use this function when new records have been added to the ringbuffer
4694  * and the console printing of those records has already occurred or is
4695  * known to be handled by some other context. This function will only
4696  * wake the logging daemon.
4697  *
4698  * Context: Any context.
4699  */
4700 void wake_up_klogd(void)
4701 {
4702 	__wake_up_klogd(0);
4703 }
4704 
4705 /**
4706  * defer_console_output - Wake kernel logging daemon and trigger
4707  *	console printing in a deferred context
4708  *
4709  * Use this function when new records have been added to the ringbuffer,
4710  * this context is responsible for console printing those records, but
4711  * the current context is not allowed to perform the console printing.
4712  * Trigger an irq_work context to perform the console printing. This
4713  * function also wakes the logging daemon.
4714  *
4715  * Context: Any context.
4716  */
4717 void defer_console_output(void)
4718 {
4719 	/*
4720 	 * New messages may have been added directly to the ringbuffer
4721 	 * using vprintk_store(), so wake any waiters as well.
4722 	 */
4723 	__wake_up_klogd(PRINTK_PENDING_OUTPUT);
4724 }
4725 
4726 /**
4727  * printk_trigger_flush - Attempt to flush printk buffer to consoles.
4728  *
4729  * If possible, flush the printk buffer to all consoles in the caller's
4730  * context. If offloading is available, trigger deferred printing.
4731  *
4732  * This is best effort. Depending on the system state, console states,
4733  * and caller context, no actual flushing may result from this call.
4734  */
4735 void printk_trigger_flush(void)
4736 {
4737 	struct console_flush_type ft;
4738 
4739 	printk_get_console_flush_type(&ft);
4740 	if (ft.nbcon_atomic)
4741 		nbcon_atomic_flush_pending();
4742 	if (ft.nbcon_offload)
4743 		nbcon_kthreads_wake();
4744 	if (ft.legacy_direct) {
4745 		if (console_trylock())
4746 			console_unlock();
4747 	}
4748 	if (ft.legacy_offload)
4749 		defer_console_output();
4750 }
4751 
4752 int vprintk_deferred(const char *fmt, va_list args)
4753 {
4754 	return vprintk_emit(0, LOGLEVEL_SCHED, NULL, fmt, args);
4755 }
4756 
4757 int _printk_deferred(const char *fmt, ...)
4758 {
4759 	va_list args;
4760 	int r;
4761 
4762 	va_start(args, fmt);
4763 	r = vprintk_deferred(fmt, args);
4764 	va_end(args);
4765 
4766 	return r;
4767 }
4768 
4769 /*
4770  * printk rate limiting, lifted from the networking subsystem.
4771  *
4772  * This enforces a rate limit: not more than 10 kernel messages
4773  * every 5s to make a denial-of-service attack impossible.
4774  */
4775 DEFINE_RATELIMIT_STATE(printk_ratelimit_state, 5 * HZ, 10);
4776 
4777 int __printk_ratelimit(const char *func)
4778 {
4779 	return ___ratelimit(&printk_ratelimit_state, func);
4780 }
4781 EXPORT_SYMBOL(__printk_ratelimit);
4782 
4783 /**
4784  * printk_timed_ratelimit - caller-controlled printk ratelimiting
4785  * @caller_jiffies: pointer to caller's state
4786  * @interval_msecs: minimum interval between prints
4787  *
4788  * printk_timed_ratelimit() returns true if more than @interval_msecs
4789  * milliseconds have elapsed since the last time printk_timed_ratelimit()
4790  * returned true.
4791  */
4792 bool printk_timed_ratelimit(unsigned long *caller_jiffies,
4793 			unsigned int interval_msecs)
4794 {
4795 	unsigned long elapsed = jiffies - *caller_jiffies;
4796 
4797 	if (*caller_jiffies && elapsed <= msecs_to_jiffies(interval_msecs))
4798 		return false;
4799 
4800 	*caller_jiffies = jiffies;
4801 	return true;
4802 }
4803 EXPORT_SYMBOL(printk_timed_ratelimit);
4804 
4805 static DEFINE_SPINLOCK(dump_list_lock);
4806 static LIST_HEAD(dump_list);
4807 
4808 /**
4809  * kmsg_dump_register - register a kernel log dumper.
4810  * @dumper: pointer to the kmsg_dumper structure
4811  *
4812  * Adds a kernel log dumper to the system. The dump callback in the
4813  * structure will be called when the kernel oopses or panics and must be
4814  * set. Returns zero on success and %-EINVAL or %-EBUSY otherwise.
4815  */
4816 int kmsg_dump_register(struct kmsg_dumper *dumper)
4817 {
4818 	unsigned long flags;
4819 	int err = -EBUSY;
4820 
4821 	/* The dump callback needs to be set */
4822 	if (!dumper->dump)
4823 		return -EINVAL;
4824 
4825 	spin_lock_irqsave(&dump_list_lock, flags);
4826 	/* Don't allow registering multiple times */
4827 	if (!dumper->registered) {
4828 		dumper->registered = 1;
4829 		list_add_tail_rcu(&dumper->list, &dump_list);
4830 		err = 0;
4831 	}
4832 	spin_unlock_irqrestore(&dump_list_lock, flags);
4833 
4834 	return err;
4835 }
4836 EXPORT_SYMBOL_GPL(kmsg_dump_register);
4837 
4838 /**
4839  * kmsg_dump_unregister - unregister a kmsg dumper.
4840  * @dumper: pointer to the kmsg_dumper structure
4841  *
4842  * Removes a dump device from the system. Returns zero on success and
4843  * %-EINVAL otherwise.
4844  */
4845 int kmsg_dump_unregister(struct kmsg_dumper *dumper)
4846 {
4847 	unsigned long flags;
4848 	int err = -EINVAL;
4849 
4850 	spin_lock_irqsave(&dump_list_lock, flags);
4851 	if (dumper->registered) {
4852 		dumper->registered = 0;
4853 		list_del_rcu(&dumper->list);
4854 		err = 0;
4855 	}
4856 	spin_unlock_irqrestore(&dump_list_lock, flags);
4857 	synchronize_rcu();
4858 
4859 	return err;
4860 }
4861 EXPORT_SYMBOL_GPL(kmsg_dump_unregister);
4862 
4863 static bool always_kmsg_dump;
4864 module_param_named(always_kmsg_dump, always_kmsg_dump, bool, S_IRUGO | S_IWUSR);
4865 
4866 const char *kmsg_dump_reason_str(enum kmsg_dump_reason reason)
4867 {
4868 	switch (reason) {
4869 	case KMSG_DUMP_PANIC:
4870 		return "Panic";
4871 	case KMSG_DUMP_OOPS:
4872 		return "Oops";
4873 	case KMSG_DUMP_EMERG:
4874 		return "Emergency";
4875 	case KMSG_DUMP_SHUTDOWN:
4876 		return "Shutdown";
4877 	default:
4878 		return "Unknown";
4879 	}
4880 }
4881 EXPORT_SYMBOL_GPL(kmsg_dump_reason_str);
4882 
4883 /**
4884  * kmsg_dump_desc - dump kernel log to kernel message dumpers.
4885  * @reason: the reason (oops, panic etc) for dumping
4886  * @desc: a short string to describe what caused the panic or oops. Can be NULL
4887  * if no additional description is available.
4888  *
4889  * Call each of the registered dumper's dump() callback, which can
4890  * retrieve the kmsg records with kmsg_dump_get_line() or
4891  * kmsg_dump_get_buffer().
4892  */
4893 void kmsg_dump_desc(enum kmsg_dump_reason reason, const char *desc)
4894 {
4895 	struct kmsg_dumper *dumper;
4896 	struct kmsg_dump_detail detail = {
4897 		.reason = reason,
4898 		.description = desc};
4899 
4900 	rcu_read_lock();
4901 	list_for_each_entry_rcu(dumper, &dump_list, list) {
4902 		enum kmsg_dump_reason max_reason = dumper->max_reason;
4903 
4904 		/*
4905 		 * If client has not provided a specific max_reason, default
4906 		 * to KMSG_DUMP_OOPS, unless always_kmsg_dump was set.
4907 		 */
4908 		if (max_reason == KMSG_DUMP_UNDEF) {
4909 			max_reason = always_kmsg_dump ? KMSG_DUMP_MAX :
4910 							KMSG_DUMP_OOPS;
4911 		}
4912 		if (reason > max_reason)
4913 			continue;
4914 
4915 		/* invoke dumper which will iterate over records */
4916 		dumper->dump(dumper, &detail);
4917 	}
4918 	rcu_read_unlock();
4919 }
4920 
4921 /**
4922  * kmsg_dump_get_line - retrieve one kmsg log line
4923  * @iter: kmsg dump iterator
4924  * @syslog: include the "<4>" prefixes
4925  * @line: buffer to copy the line to
4926  * @size: maximum size of the buffer
4927  * @len: length of line placed into buffer
4928  *
4929  * Start at the beginning of the kmsg buffer, with the oldest kmsg
4930  * record, and copy one record into the provided buffer.
4931  *
4932  * Consecutive calls will return the next available record moving
4933  * towards the end of the buffer with the youngest messages.
4934  *
4935  * A return value of FALSE indicates that there are no more records to
4936  * read.
4937  */
4938 bool kmsg_dump_get_line(struct kmsg_dump_iter *iter, bool syslog,
4939 			char *line, size_t size, size_t *len)
4940 {
4941 	u64 min_seq = latched_seq_read_nolock(&clear_seq);
4942 	struct printk_info info;
4943 	unsigned int line_count;
4944 	struct printk_record r;
4945 	size_t l = 0;
4946 	bool ret = false;
4947 
4948 	if (iter->cur_seq < min_seq)
4949 		iter->cur_seq = min_seq;
4950 
4951 	prb_rec_init_rd(&r, &info, line, size);
4952 
4953 	/* Read text or count text lines? */
4954 	if (line) {
4955 		if (!prb_read_valid(prb, iter->cur_seq, &r))
4956 			goto out;
4957 		l = record_print_text(&r, syslog, printk_time);
4958 	} else {
4959 		if (!prb_read_valid_info(prb, iter->cur_seq,
4960 					 &info, &line_count)) {
4961 			goto out;
4962 		}
4963 		l = get_record_print_text_size(&info, line_count, syslog,
4964 					       printk_time);
4965 
4966 	}
4967 
4968 	iter->cur_seq = r.info->seq + 1;
4969 	ret = true;
4970 out:
4971 	if (len)
4972 		*len = l;
4973 	return ret;
4974 }
4975 EXPORT_SYMBOL_GPL(kmsg_dump_get_line);
4976 
4977 /**
4978  * kmsg_dump_get_buffer - copy kmsg log lines
4979  * @iter: kmsg dump iterator
4980  * @syslog: include the "<4>" prefixes
4981  * @buf: buffer to copy the line to
4982  * @size: maximum size of the buffer
4983  * @len_out: length of line placed into buffer
4984  *
4985  * Start at the end of the kmsg buffer and fill the provided buffer
4986  * with as many of the *youngest* kmsg records that fit into it.
4987  * If the buffer is large enough, all available kmsg records will be
4988  * copied with a single call.
4989  *
4990  * Consecutive calls will fill the buffer with the next block of
4991  * available older records, not including the earlier retrieved ones.
4992  *
4993  * A return value of FALSE indicates that there are no more records to
4994  * read.
4995  */
4996 bool kmsg_dump_get_buffer(struct kmsg_dump_iter *iter, bool syslog,
4997 			  char *buf, size_t size, size_t *len_out)
4998 {
4999 	u64 min_seq = latched_seq_read_nolock(&clear_seq);
5000 	struct printk_info info;
5001 	struct printk_record r;
5002 	u64 seq;
5003 	u64 next_seq;
5004 	size_t len = 0;
5005 	bool ret = false;
5006 	bool time = printk_time;
5007 
5008 	if (!buf || !size)
5009 		goto out;
5010 
5011 	if (iter->cur_seq < min_seq)
5012 		iter->cur_seq = min_seq;
5013 
5014 	if (prb_read_valid_info(prb, iter->cur_seq, &info, NULL)) {
5015 		if (info.seq != iter->cur_seq) {
5016 			/* messages are gone, move to first available one */
5017 			iter->cur_seq = info.seq;
5018 		}
5019 	}
5020 
5021 	/* last entry */
5022 	if (iter->cur_seq >= iter->next_seq)
5023 		goto out;
5024 
5025 	/*
5026 	 * Find first record that fits, including all following records,
5027 	 * into the user-provided buffer for this dump. Pass in size-1
5028 	 * because this function (by way of record_print_text()) will
5029 	 * not write more than size-1 bytes of text into @buf.
5030 	 */
5031 	seq = find_first_fitting_seq(iter->cur_seq, iter->next_seq,
5032 				     size - 1, syslog, time);
5033 
5034 	/*
5035 	 * Next kmsg_dump_get_buffer() invocation will dump block of
5036 	 * older records stored right before this one.
5037 	 */
5038 	next_seq = seq;
5039 
5040 	prb_rec_init_rd(&r, &info, buf, size);
5041 
5042 	prb_for_each_record(seq, prb, seq, &r) {
5043 		if (r.info->seq >= iter->next_seq)
5044 			break;
5045 
5046 		len += record_print_text(&r, syslog, time);
5047 
5048 		/* Adjust record to store to remaining buffer space. */
5049 		prb_rec_init_rd(&r, &info, buf + len, size - len);
5050 	}
5051 
5052 	iter->next_seq = next_seq;
5053 	ret = true;
5054 out:
5055 	if (len_out)
5056 		*len_out = len;
5057 	return ret;
5058 }
5059 EXPORT_SYMBOL_GPL(kmsg_dump_get_buffer);
5060 
5061 /**
5062  * kmsg_dump_rewind - reset the iterator
5063  * @iter: kmsg dump iterator
5064  *
5065  * Reset the dumper's iterator so that kmsg_dump_get_line() and
5066  * kmsg_dump_get_buffer() can be called again and used multiple
5067  * times within the same dumper.dump() callback.
5068  */
5069 void kmsg_dump_rewind(struct kmsg_dump_iter *iter)
5070 {
5071 	iter->cur_seq = latched_seq_read_nolock(&clear_seq);
5072 	iter->next_seq = prb_next_seq(prb);
5073 }
5074 EXPORT_SYMBOL_GPL(kmsg_dump_rewind);
5075 
5076 /**
5077  * console_try_replay_all - try to replay kernel log on consoles
5078  *
5079  * Try to obtain lock on console subsystem and replay all
5080  * available records in printk buffer on the consoles.
5081  * Does nothing if lock is not obtained.
5082  *
5083  * Context: Any, except for NMI.
5084  */
5085 void console_try_replay_all(void)
5086 {
5087 	struct console_flush_type ft;
5088 
5089 	printk_get_console_flush_type(&ft);
5090 	if (console_trylock()) {
5091 		__console_rewind_all();
5092 		if (ft.nbcon_atomic)
5093 			nbcon_atomic_flush_pending();
5094 		if (ft.nbcon_offload)
5095 			nbcon_kthreads_wake();
5096 		if (ft.legacy_offload)
5097 			defer_console_output();
5098 		/* Consoles are flushed as part of console_unlock(). */
5099 		console_unlock();
5100 	}
5101 }
5102 #endif
5103 
5104 #ifdef CONFIG_SMP
5105 static atomic_t printk_cpu_sync_owner = ATOMIC_INIT(-1);
5106 static atomic_t printk_cpu_sync_nested = ATOMIC_INIT(0);
5107 
5108 bool is_printk_cpu_sync_owner(void)
5109 {
5110 	return (atomic_read(&printk_cpu_sync_owner) == raw_smp_processor_id());
5111 }
5112 
5113 /**
5114  * __printk_cpu_sync_wait() - Busy wait until the printk cpu-reentrant
5115  *                            spinning lock is not owned by any CPU.
5116  *
5117  * Context: Any context.
5118  */
5119 void __printk_cpu_sync_wait(void)
5120 {
5121 	do {
5122 		cpu_relax();
5123 	} while (atomic_read(&printk_cpu_sync_owner) != -1);
5124 }
5125 EXPORT_SYMBOL(__printk_cpu_sync_wait);
5126 
5127 /**
5128  * __printk_cpu_sync_try_get() - Try to acquire the printk cpu-reentrant
5129  *                               spinning lock.
5130  *
5131  * If no processor has the lock, the calling processor takes the lock and
5132  * becomes the owner. If the calling processor is already the owner of the
5133  * lock, this function succeeds immediately.
5134  *
5135  * Context: Any context. Expects interrupts to be disabled.
5136  * Return: 1 on success, otherwise 0.
5137  */
5138 int __printk_cpu_sync_try_get(void)
5139 {
5140 	int cpu;
5141 	int old;
5142 
5143 	cpu = smp_processor_id();
5144 
5145 	/*
5146 	 * Guarantee loads and stores from this CPU when it is the lock owner
5147 	 * are _not_ visible to the previous lock owner. This pairs with
5148 	 * __printk_cpu_sync_put:B.
5149 	 *
5150 	 * Memory barrier involvement:
5151 	 *
5152 	 * If __printk_cpu_sync_try_get:A reads from __printk_cpu_sync_put:B,
5153 	 * then __printk_cpu_sync_put:A can never read from
5154 	 * __printk_cpu_sync_try_get:B.
5155 	 *
5156 	 * Relies on:
5157 	 *
5158 	 * RELEASE from __printk_cpu_sync_put:A to __printk_cpu_sync_put:B
5159 	 * of the previous CPU
5160 	 *    matching
5161 	 * ACQUIRE from __printk_cpu_sync_try_get:A to
5162 	 * __printk_cpu_sync_try_get:B of this CPU
5163 	 */
5164 	old = atomic_cmpxchg_acquire(&printk_cpu_sync_owner, -1,
5165 				     cpu); /* LMM(__printk_cpu_sync_try_get:A) */
5166 	if (old == -1) {
5167 		/*
5168 		 * This CPU is now the owner and begins loading/storing
5169 		 * data: LMM(__printk_cpu_sync_try_get:B)
5170 		 */
5171 		return 1;
5172 
5173 	} else if (old == cpu) {
5174 		/* This CPU is already the owner. */
5175 		atomic_inc(&printk_cpu_sync_nested);
5176 		return 1;
5177 	}
5178 
5179 	return 0;
5180 }
5181 EXPORT_SYMBOL(__printk_cpu_sync_try_get);
5182 
5183 /**
5184  * __printk_cpu_sync_put() - Release the printk cpu-reentrant spinning lock.
5185  *
5186  * The calling processor must be the owner of the lock.
5187  *
5188  * Context: Any context. Expects interrupts to be disabled.
5189  */
5190 void __printk_cpu_sync_put(void)
5191 {
5192 	if (atomic_read(&printk_cpu_sync_nested)) {
5193 		atomic_dec(&printk_cpu_sync_nested);
5194 		return;
5195 	}
5196 
5197 	/*
5198 	 * This CPU is finished loading/storing data:
5199 	 * LMM(__printk_cpu_sync_put:A)
5200 	 */
5201 
5202 	/*
5203 	 * Guarantee loads and stores from this CPU when it was the
5204 	 * lock owner are visible to the next lock owner. This pairs
5205 	 * with __printk_cpu_sync_try_get:A.
5206 	 *
5207 	 * Memory barrier involvement:
5208 	 *
5209 	 * If __printk_cpu_sync_try_get:A reads from __printk_cpu_sync_put:B,
5210 	 * then __printk_cpu_sync_try_get:B reads from __printk_cpu_sync_put:A.
5211 	 *
5212 	 * Relies on:
5213 	 *
5214 	 * RELEASE from __printk_cpu_sync_put:A to __printk_cpu_sync_put:B
5215 	 * of this CPU
5216 	 *    matching
5217 	 * ACQUIRE from __printk_cpu_sync_try_get:A to
5218 	 * __printk_cpu_sync_try_get:B of the next CPU
5219 	 */
5220 	atomic_set_release(&printk_cpu_sync_owner,
5221 			   -1); /* LMM(__printk_cpu_sync_put:B) */
5222 }
5223 EXPORT_SYMBOL(__printk_cpu_sync_put);
5224 #endif /* CONFIG_SMP */
5225