2 * linux/kernel/printk.c
4 * Copyright (C) 1991, 1992 Linus Torvalds
6 * Modified to make sys_syslog() more flexible: added commands to
7 * return the last 4k of kernel messages, regardless of whether
8 * they've been read or not. Added option to suppress kernel printk's
9 * to the console. Added hook for sending the console messages
10 * elsewhere, in preparation for a serial line console (someday).
12 * Modified for sysctl support, 1/8/97, Chris Horn.
13 * Fixed SMP synchronization, 08/08/99, Manfred Spraul
14 * manfred@colorfullife.com
15 * Rewrote bits to get rid of console_lock
16 * 01Mar01 Andrew Morton
19 #define pr_fmt(fmt) KBUILD_MODNAME ": " fmt
21 #include <linux/kernel.h>
23 #include <linux/tty.h>
24 #include <linux/tty_driver.h>
25 #include <linux/console.h>
26 #include <linux/init.h>
27 #include <linux/jiffies.h>
28 #include <linux/nmi.h>
29 #include <linux/module.h>
30 #include <linux/moduleparam.h>
31 #include <linux/delay.h>
32 #include <linux/smp.h>
33 #include <linux/security.h>
34 #include <linux/bootmem.h>
35 #include <linux/memblock.h>
36 #include <linux/syscalls.h>
37 #include <linux/crash_core.h>
38 #include <linux/kdb.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>
52 #include <linux/uaccess.h>
53 #include <asm/sections.h>
55 #include <trace/events/initcall.h>
56 #define CREATE_TRACE_POINTS
57 #include <trace/events/printk.h>
59 #include "console_cmdline.h"
63 int console_printk[4] = {
64 CONSOLE_LOGLEVEL_DEFAULT, /* console_loglevel */
65 MESSAGE_LOGLEVEL_DEFAULT, /* default_message_loglevel */
66 CONSOLE_LOGLEVEL_MIN, /* minimum_console_loglevel */
67 CONSOLE_LOGLEVEL_DEFAULT, /* default_console_loglevel */
70 atomic_t ignore_console_lock_warning __read_mostly = ATOMIC_INIT(0);
71 EXPORT_SYMBOL(ignore_console_lock_warning);
74 * Low level drivers may need that to know if they can schedule in
75 * their unblank() callback or not. So let's export it.
78 EXPORT_SYMBOL(oops_in_progress);
81 * console_sem protects the console_drivers list, and also
82 * provides serialisation for access to the entire console
85 static DEFINE_SEMAPHORE(console_sem);
86 struct console *console_drivers;
87 EXPORT_SYMBOL_GPL(console_drivers);
90 static struct lockdep_map console_lock_dep_map = {
91 .name = "console_lock"
95 enum devkmsg_log_bits {
96 __DEVKMSG_LOG_BIT_ON = 0,
97 __DEVKMSG_LOG_BIT_OFF,
98 __DEVKMSG_LOG_BIT_LOCK,
101 enum devkmsg_log_masks {
102 DEVKMSG_LOG_MASK_ON = BIT(__DEVKMSG_LOG_BIT_ON),
103 DEVKMSG_LOG_MASK_OFF = BIT(__DEVKMSG_LOG_BIT_OFF),
104 DEVKMSG_LOG_MASK_LOCK = BIT(__DEVKMSG_LOG_BIT_LOCK),
107 /* Keep both the 'on' and 'off' bits clear, i.e. ratelimit by default: */
108 #define DEVKMSG_LOG_MASK_DEFAULT 0
110 static unsigned int __read_mostly devkmsg_log = DEVKMSG_LOG_MASK_DEFAULT;
112 static int __control_devkmsg(char *str)
117 if (!strncmp(str, "on", 2)) {
118 devkmsg_log = DEVKMSG_LOG_MASK_ON;
120 } else if (!strncmp(str, "off", 3)) {
121 devkmsg_log = DEVKMSG_LOG_MASK_OFF;
123 } else if (!strncmp(str, "ratelimit", 9)) {
124 devkmsg_log = DEVKMSG_LOG_MASK_DEFAULT;
130 static int __init control_devkmsg(char *str)
132 if (__control_devkmsg(str) < 0)
136 * Set sysctl string accordingly:
138 if (devkmsg_log == DEVKMSG_LOG_MASK_ON)
139 strcpy(devkmsg_log_str, "on");
140 else if (devkmsg_log == DEVKMSG_LOG_MASK_OFF)
141 strcpy(devkmsg_log_str, "off");
142 /* else "ratelimit" which is set by default. */
145 * Sysctl cannot change it anymore. The kernel command line setting of
146 * this parameter is to force the setting to be permanent throughout the
147 * runtime of the system. This is a precation measure against userspace
148 * trying to be a smarta** and attempting to change it up on us.
150 devkmsg_log |= DEVKMSG_LOG_MASK_LOCK;
154 __setup("printk.devkmsg=", control_devkmsg);
156 char devkmsg_log_str[DEVKMSG_STR_MAX_SIZE] = "ratelimit";
158 int devkmsg_sysctl_set_loglvl(struct ctl_table *table, int write,
159 void __user *buffer, size_t *lenp, loff_t *ppos)
161 char old_str[DEVKMSG_STR_MAX_SIZE];
166 if (devkmsg_log & DEVKMSG_LOG_MASK_LOCK)
170 strncpy(old_str, devkmsg_log_str, DEVKMSG_STR_MAX_SIZE);
173 err = proc_dostring(table, write, buffer, lenp, ppos);
178 err = __control_devkmsg(devkmsg_log_str);
181 * Do not accept an unknown string OR a known string with
184 if (err < 0 || (err + 1 != *lenp)) {
186 /* ... and restore old setting. */
188 strncpy(devkmsg_log_str, old_str, DEVKMSG_STR_MAX_SIZE);
197 /* Number of registered extended console drivers. */
198 static int nr_ext_console_drivers;
201 * Helper macros to handle lockdep when locking/unlocking console_sem. We use
202 * macros instead of functions so that _RET_IP_ contains useful information.
204 #define down_console_sem() do { \
206 mutex_acquire(&console_lock_dep_map, 0, 0, _RET_IP_);\
209 static int __down_trylock_console_sem(unsigned long ip)
215 * Here and in __up_console_sem() we need to be in safe mode,
216 * because spindump/WARN/etc from under console ->lock will
217 * deadlock in printk()->down_trylock_console_sem() otherwise.
219 printk_safe_enter_irqsave(flags);
220 lock_failed = down_trylock(&console_sem);
221 printk_safe_exit_irqrestore(flags);
225 mutex_acquire(&console_lock_dep_map, 0, 1, ip);
228 #define down_trylock_console_sem() __down_trylock_console_sem(_RET_IP_)
230 static void __up_console_sem(unsigned long ip)
234 mutex_release(&console_lock_dep_map, 1, ip);
236 printk_safe_enter_irqsave(flags);
238 printk_safe_exit_irqrestore(flags);
240 #define up_console_sem() __up_console_sem(_RET_IP_)
243 * This is used for debugging the mess that is the VT code by
244 * keeping track if we have the console semaphore held. It's
245 * definitely not the perfect debug tool (we don't know if _WE_
246 * hold it and are racing, but it helps tracking those weird code
247 * paths in the console code where we end up in places I want
248 * locked without the console sempahore held).
250 static int console_locked, console_suspended;
253 * If exclusive_console is non-NULL then only this console is to be printed to.
255 static struct console *exclusive_console;
258 * Array of consoles built from command line options (console=)
261 #define MAX_CMDLINECONSOLES 8
263 static struct console_cmdline console_cmdline[MAX_CMDLINECONSOLES];
265 static int preferred_console = -1;
266 int console_set_on_cmdline;
267 EXPORT_SYMBOL(console_set_on_cmdline);
269 /* Flag: console code may call schedule() */
270 static int console_may_schedule;
272 enum con_msg_format_flags {
273 MSG_FORMAT_DEFAULT = 0,
274 MSG_FORMAT_SYSLOG = (1 << 0),
277 static int console_msg_format = MSG_FORMAT_DEFAULT;
280 * The printk log buffer consists of a chain of concatenated variable
281 * length records. Every record starts with a record header, containing
282 * the overall length of the record.
284 * The heads to the first and last entry in the buffer, as well as the
285 * sequence numbers of these entries are maintained when messages are
288 * If the heads indicate available messages, the length in the header
289 * tells the start next message. A length == 0 for the next message
290 * indicates a wrap-around to the beginning of the buffer.
292 * Every record carries the monotonic timestamp in microseconds, as well as
293 * the standard userspace syslog level and syslog facility. The usual
294 * kernel messages use LOG_KERN; userspace-injected messages always carry
295 * a matching syslog facility, by default LOG_USER. The origin of every
296 * message can be reliably determined that way.
298 * The human readable log message directly follows the message header. The
299 * length of the message text is stored in the header, the stored message
302 * Optionally, a message can carry a dictionary of properties (key/value pairs),
303 * to provide userspace with a machine-readable message context.
305 * Examples for well-defined, commonly used property names are:
306 * DEVICE=b12:8 device identifier
310 * +sound:card0 subsystem:devname
311 * SUBSYSTEM=pci driver-core subsystem name
313 * Valid characters in property names are [a-zA-Z0-9.-_]. The plain text value
314 * follows directly after a '=' character. Every property is terminated by
315 * a '\0' character. The last property is not terminated.
317 * Example of a message structure:
318 * 0000 ff 8f 00 00 00 00 00 00 monotonic time in nsec
319 * 0008 34 00 record is 52 bytes long
320 * 000a 0b 00 text is 11 bytes long
321 * 000c 1f 00 dictionary is 23 bytes long
322 * 000e 03 00 LOG_KERN (facility) LOG_ERR (level)
323 * 0010 69 74 27 73 20 61 20 6c "it's a l"
325 * 001b 44 45 56 49 43 "DEVIC"
326 * 45 3d 62 38 3a 32 00 44 "E=b8:2\0D"
327 * 52 49 56 45 52 3d 62 75 "RIVER=bu"
329 * 0032 00 00 00 padding to next message header
331 * The 'struct printk_log' buffer header must never be directly exported to
332 * userspace, it is a kernel-private implementation detail that might
333 * need to be changed in the future, when the requirements change.
335 * /dev/kmsg exports the structured data in the following line format:
336 * "<level>,<sequnum>,<timestamp>,<contflag>[,additional_values, ... ];<message text>\n"
338 * Users of the export format should ignore possible additional values
339 * separated by ',', and find the message after the ';' character.
341 * The optional key/value pairs are attached as continuation lines starting
342 * with a space character and terminated by a newline. All possible
343 * non-prinatable characters are escaped in the "\xff" notation.
347 LOG_NEWLINE = 2, /* text ended with a newline */
348 LOG_PREFIX = 4, /* text started with a prefix */
349 LOG_CONT = 8, /* text is a fragment of a continuation line */
353 u64 ts_nsec; /* timestamp in nanoseconds */
354 u16 len; /* length of entire record */
355 u16 text_len; /* length of text buffer */
356 u16 dict_len; /* length of dictionary buffer */
357 u8 facility; /* syslog facility */
358 u8 flags:5; /* internal record flags */
359 u8 level:3; /* syslog level */
360 #ifdef CONFIG_PRINTK_CALLER
361 u32 caller_id; /* thread id or processor id */
364 #ifdef CONFIG_HAVE_EFFICIENT_UNALIGNED_ACCESS
365 __packed __aligned(4)
370 * The logbuf_lock protects kmsg buffer, indices, counters. This can be taken
371 * within the scheduler's rq lock. It must be released before calling
372 * console_unlock() or anything else that might wake up a process.
374 DEFINE_RAW_SPINLOCK(logbuf_lock);
377 * Helper macros to lock/unlock logbuf_lock and switch between
378 * printk-safe/unsafe modes.
380 #define logbuf_lock_irq() \
382 printk_safe_enter_irq(); \
383 raw_spin_lock(&logbuf_lock); \
386 #define logbuf_unlock_irq() \
388 raw_spin_unlock(&logbuf_lock); \
389 printk_safe_exit_irq(); \
392 #define logbuf_lock_irqsave(flags) \
394 printk_safe_enter_irqsave(flags); \
395 raw_spin_lock(&logbuf_lock); \
398 #define logbuf_unlock_irqrestore(flags) \
400 raw_spin_unlock(&logbuf_lock); \
401 printk_safe_exit_irqrestore(flags); \
405 DECLARE_WAIT_QUEUE_HEAD(log_wait);
406 /* the next printk record to read by syslog(READ) or /proc/kmsg */
407 static u64 syslog_seq;
408 static u32 syslog_idx;
409 static size_t syslog_partial;
410 static bool syslog_time;
412 /* index and sequence number of the first record stored in the buffer */
413 static u64 log_first_seq;
414 static u32 log_first_idx;
416 /* index and sequence number of the next record to store in the buffer */
417 static u64 log_next_seq;
418 static u32 log_next_idx;
420 /* the next printk record to write to the console */
421 static u64 console_seq;
422 static u32 console_idx;
423 static u64 exclusive_console_stop_seq;
425 /* the next printk record to read after the last 'clear' command */
426 static u64 clear_seq;
427 static u32 clear_idx;
429 #ifdef CONFIG_PRINTK_CALLER
430 #define PREFIX_MAX 48
432 #define PREFIX_MAX 32
434 #define LOG_LINE_MAX (1024 - PREFIX_MAX)
436 #define LOG_LEVEL(v) ((v) & 0x07)
437 #define LOG_FACILITY(v) ((v) >> 3 & 0xff)
440 #define LOG_ALIGN __alignof__(struct printk_log)
441 #define __LOG_BUF_LEN (1 << CONFIG_LOG_BUF_SHIFT)
442 #define LOG_BUF_LEN_MAX (u32)(1 << 31)
443 static char __log_buf[__LOG_BUF_LEN] __aligned(LOG_ALIGN);
444 static char *log_buf = __log_buf;
445 static u32 log_buf_len = __LOG_BUF_LEN;
447 /* Return log buffer address */
448 char *log_buf_addr_get(void)
453 /* Return log buffer size */
454 u32 log_buf_len_get(void)
459 /* human readable text of the record */
460 static char *log_text(const struct printk_log *msg)
462 return (char *)msg + sizeof(struct printk_log);
465 /* optional key/value pair dictionary attached to the record */
466 static char *log_dict(const struct printk_log *msg)
468 return (char *)msg + sizeof(struct printk_log) + msg->text_len;
471 /* get record by index; idx must point to valid msg */
472 static struct printk_log *log_from_idx(u32 idx)
474 struct printk_log *msg = (struct printk_log *)(log_buf + idx);
477 * A length == 0 record is the end of buffer marker. Wrap around and
478 * read the message at the start of the buffer.
481 return (struct printk_log *)log_buf;
485 /* get next record; idx must point to valid msg */
486 static u32 log_next(u32 idx)
488 struct printk_log *msg = (struct printk_log *)(log_buf + idx);
490 /* length == 0 indicates the end of the buffer; wrap */
492 * A length == 0 record is the end of buffer marker. Wrap around and
493 * read the message at the start of the buffer as *this* one, and
494 * return the one after that.
497 msg = (struct printk_log *)log_buf;
500 return idx + msg->len;
504 * Check whether there is enough free space for the given message.
506 * The same values of first_idx and next_idx mean that the buffer
507 * is either empty or full.
509 * If the buffer is empty, we must respect the position of the indexes.
510 * They cannot be reset to the beginning of the buffer.
512 static int logbuf_has_space(u32 msg_size, bool empty)
516 if (log_next_idx > log_first_idx || empty)
517 free = max(log_buf_len - log_next_idx, log_first_idx);
519 free = log_first_idx - log_next_idx;
522 * We need space also for an empty header that signalizes wrapping
525 return free >= msg_size + sizeof(struct printk_log);
528 static int log_make_free_space(u32 msg_size)
530 while (log_first_seq < log_next_seq &&
531 !logbuf_has_space(msg_size, false)) {
532 /* drop old messages until we have enough contiguous space */
533 log_first_idx = log_next(log_first_idx);
537 if (clear_seq < log_first_seq) {
538 clear_seq = log_first_seq;
539 clear_idx = log_first_idx;
542 /* sequence numbers are equal, so the log buffer is empty */
543 if (logbuf_has_space(msg_size, log_first_seq == log_next_seq))
549 /* compute the message size including the padding bytes */
550 static u32 msg_used_size(u16 text_len, u16 dict_len, u32 *pad_len)
554 size = sizeof(struct printk_log) + text_len + dict_len;
555 *pad_len = (-size) & (LOG_ALIGN - 1);
562 * Define how much of the log buffer we could take at maximum. The value
563 * must be greater than two. Note that only half of the buffer is available
564 * when the index points to the middle.
566 #define MAX_LOG_TAKE_PART 4
567 static const char trunc_msg[] = "<truncated>";
569 static u32 truncate_msg(u16 *text_len, u16 *trunc_msg_len,
570 u16 *dict_len, u32 *pad_len)
573 * The message should not take the whole buffer. Otherwise, it might
574 * get removed too soon.
576 u32 max_text_len = log_buf_len / MAX_LOG_TAKE_PART;
577 if (*text_len > max_text_len)
578 *text_len = max_text_len;
579 /* enable the warning message */
580 *trunc_msg_len = strlen(trunc_msg);
581 /* disable the "dict" completely */
583 /* compute the size again, count also the warning message */
584 return msg_used_size(*text_len + *trunc_msg_len, 0, pad_len);
587 /* insert record into the buffer, discard old ones, update heads */
588 static int log_store(u32 caller_id, int facility, int level,
589 enum log_flags flags, u64 ts_nsec,
590 const char *dict, u16 dict_len,
591 const char *text, u16 text_len)
593 struct printk_log *msg;
595 u16 trunc_msg_len = 0;
597 /* number of '\0' padding bytes to next message */
598 size = msg_used_size(text_len, dict_len, &pad_len);
600 if (log_make_free_space(size)) {
601 /* truncate the message if it is too long for empty buffer */
602 size = truncate_msg(&text_len, &trunc_msg_len,
603 &dict_len, &pad_len);
604 /* survive when the log buffer is too small for trunc_msg */
605 if (log_make_free_space(size))
609 if (log_next_idx + size + sizeof(struct printk_log) > log_buf_len) {
611 * This message + an additional empty header does not fit
612 * at the end of the buffer. Add an empty header with len == 0
613 * to signify a wrap around.
615 memset(log_buf + log_next_idx, 0, sizeof(struct printk_log));
620 msg = (struct printk_log *)(log_buf + log_next_idx);
621 memcpy(log_text(msg), text, text_len);
622 msg->text_len = text_len;
624 memcpy(log_text(msg) + text_len, trunc_msg, trunc_msg_len);
625 msg->text_len += trunc_msg_len;
627 memcpy(log_dict(msg), dict, dict_len);
628 msg->dict_len = dict_len;
629 msg->facility = facility;
630 msg->level = level & 7;
631 msg->flags = flags & 0x1f;
633 msg->ts_nsec = ts_nsec;
635 msg->ts_nsec = local_clock();
636 #ifdef CONFIG_PRINTK_CALLER
637 msg->caller_id = caller_id;
639 memset(log_dict(msg) + dict_len, 0, pad_len);
643 log_next_idx += msg->len;
646 return msg->text_len;
649 int dmesg_restrict = IS_ENABLED(CONFIG_SECURITY_DMESG_RESTRICT);
651 static int syslog_action_restricted(int type)
656 * Unless restricted, we allow "read all" and "get buffer size"
659 return type != SYSLOG_ACTION_READ_ALL &&
660 type != SYSLOG_ACTION_SIZE_BUFFER;
663 static int check_syslog_permissions(int type, int source)
666 * If this is from /proc/kmsg and we've already opened it, then we've
667 * already done the capabilities checks at open time.
669 if (source == SYSLOG_FROM_PROC && type != SYSLOG_ACTION_OPEN)
672 if (syslog_action_restricted(type)) {
673 if (capable(CAP_SYSLOG))
676 * For historical reasons, accept CAP_SYS_ADMIN too, with
679 if (capable(CAP_SYS_ADMIN)) {
680 pr_warn_once("%s (%d): Attempt to access syslog with "
681 "CAP_SYS_ADMIN but no CAP_SYSLOG "
683 current->comm, task_pid_nr(current));
689 return security_syslog(type);
692 static void append_char(char **pp, char *e, char c)
698 static ssize_t msg_print_ext_header(char *buf, size_t size,
699 struct printk_log *msg, u64 seq)
701 u64 ts_usec = msg->ts_nsec;
703 #ifdef CONFIG_PRINTK_CALLER
704 u32 id = msg->caller_id;
706 snprintf(caller, sizeof(caller), ",caller=%c%u",
707 id & 0x80000000 ? 'C' : 'T', id & ~0x80000000);
712 do_div(ts_usec, 1000);
714 return scnprintf(buf, size, "%u,%llu,%llu,%c%s;",
715 (msg->facility << 3) | msg->level, seq, ts_usec,
716 msg->flags & LOG_CONT ? 'c' : '-', caller);
719 static ssize_t msg_print_ext_body(char *buf, size_t size,
720 char *dict, size_t dict_len,
721 char *text, size_t text_len)
723 char *p = buf, *e = buf + size;
726 /* escape non-printable characters */
727 for (i = 0; i < text_len; i++) {
728 unsigned char c = text[i];
730 if (c < ' ' || c >= 127 || c == '\\')
731 p += scnprintf(p, e - p, "\\x%02x", c);
733 append_char(&p, e, c);
735 append_char(&p, e, '\n');
740 for (i = 0; i < dict_len; i++) {
741 unsigned char c = dict[i];
744 append_char(&p, e, ' ');
749 append_char(&p, e, '\n');
754 if (c < ' ' || c >= 127 || c == '\\') {
755 p += scnprintf(p, e - p, "\\x%02x", c);
759 append_char(&p, e, c);
761 append_char(&p, e, '\n');
767 /* /dev/kmsg - userspace message inject/listen interface */
768 struct devkmsg_user {
771 struct ratelimit_state rs;
773 char buf[CONSOLE_EXT_LOG_MAX];
776 static __printf(3, 4) __cold
777 int devkmsg_emit(int facility, int level, const char *fmt, ...)
783 r = vprintk_emit(facility, level, NULL, 0, fmt, args);
789 static ssize_t devkmsg_write(struct kiocb *iocb, struct iov_iter *from)
792 int level = default_message_loglevel;
793 int facility = 1; /* LOG_USER */
794 struct file *file = iocb->ki_filp;
795 struct devkmsg_user *user = file->private_data;
796 size_t len = iov_iter_count(from);
799 if (!user || len > LOG_LINE_MAX)
802 /* Ignore when user logging is disabled. */
803 if (devkmsg_log & DEVKMSG_LOG_MASK_OFF)
806 /* Ratelimit when not explicitly enabled. */
807 if (!(devkmsg_log & DEVKMSG_LOG_MASK_ON)) {
808 if (!___ratelimit(&user->rs, current->comm))
812 buf = kmalloc(len+1, GFP_KERNEL);
817 if (!copy_from_iter_full(buf, len, from)) {
823 * Extract and skip the syslog prefix <[0-9]*>. Coming from userspace
824 * the decimal value represents 32bit, the lower 3 bit are the log
825 * level, the rest are the log facility.
827 * If no prefix or no userspace facility is specified, we
828 * enforce LOG_USER, to be able to reliably distinguish
829 * kernel-generated messages from userspace-injected ones.
832 if (line[0] == '<') {
836 u = simple_strtoul(line + 1, &endp, 10);
837 if (endp && endp[0] == '>') {
838 level = LOG_LEVEL(u);
839 if (LOG_FACILITY(u) != 0)
840 facility = LOG_FACILITY(u);
847 devkmsg_emit(facility, level, "%s", line);
852 static ssize_t devkmsg_read(struct file *file, char __user *buf,
853 size_t count, loff_t *ppos)
855 struct devkmsg_user *user = file->private_data;
856 struct printk_log *msg;
863 ret = mutex_lock_interruptible(&user->lock);
868 while (user->seq == log_next_seq) {
869 if (file->f_flags & O_NONBLOCK) {
876 ret = wait_event_interruptible(log_wait,
877 user->seq != log_next_seq);
883 if (user->seq < log_first_seq) {
884 /* our last seen message is gone, return error and reset */
885 user->idx = log_first_idx;
886 user->seq = log_first_seq;
892 msg = log_from_idx(user->idx);
893 len = msg_print_ext_header(user->buf, sizeof(user->buf),
895 len += msg_print_ext_body(user->buf + len, sizeof(user->buf) - len,
896 log_dict(msg), msg->dict_len,
897 log_text(msg), msg->text_len);
899 user->idx = log_next(user->idx);
908 if (copy_to_user(buf, user->buf, len)) {
914 mutex_unlock(&user->lock);
918 static loff_t devkmsg_llseek(struct file *file, loff_t offset, int whence)
920 struct devkmsg_user *user = file->private_data;
931 /* the first record */
932 user->idx = log_first_idx;
933 user->seq = log_first_seq;
937 * The first record after the last SYSLOG_ACTION_CLEAR,
938 * like issued by 'dmesg -c'. Reading /dev/kmsg itself
939 * changes no global state, and does not clear anything.
941 user->idx = clear_idx;
942 user->seq = clear_seq;
945 /* after the last record */
946 user->idx = log_next_idx;
947 user->seq = log_next_seq;
956 static __poll_t devkmsg_poll(struct file *file, poll_table *wait)
958 struct devkmsg_user *user = file->private_data;
962 return EPOLLERR|EPOLLNVAL;
964 poll_wait(file, &log_wait, wait);
967 if (user->seq < log_next_seq) {
968 /* return error when data has vanished underneath us */
969 if (user->seq < log_first_seq)
970 ret = EPOLLIN|EPOLLRDNORM|EPOLLERR|EPOLLPRI;
972 ret = EPOLLIN|EPOLLRDNORM;
979 static int devkmsg_open(struct inode *inode, struct file *file)
981 struct devkmsg_user *user;
984 if (devkmsg_log & DEVKMSG_LOG_MASK_OFF)
987 /* write-only does not need any file context */
988 if ((file->f_flags & O_ACCMODE) != O_WRONLY) {
989 err = check_syslog_permissions(SYSLOG_ACTION_READ_ALL,
995 user = kmalloc(sizeof(struct devkmsg_user), GFP_KERNEL);
999 ratelimit_default_init(&user->rs);
1000 ratelimit_set_flags(&user->rs, RATELIMIT_MSG_ON_RELEASE);
1002 mutex_init(&user->lock);
1005 user->idx = log_first_idx;
1006 user->seq = log_first_seq;
1007 logbuf_unlock_irq();
1009 file->private_data = user;
1013 static int devkmsg_release(struct inode *inode, struct file *file)
1015 struct devkmsg_user *user = file->private_data;
1020 ratelimit_state_exit(&user->rs);
1022 mutex_destroy(&user->lock);
1027 const struct file_operations kmsg_fops = {
1028 .open = devkmsg_open,
1029 .read = devkmsg_read,
1030 .write_iter = devkmsg_write,
1031 .llseek = devkmsg_llseek,
1032 .poll = devkmsg_poll,
1033 .release = devkmsg_release,
1036 #ifdef CONFIG_CRASH_CORE
1038 * This appends the listed symbols to /proc/vmcore
1040 * /proc/vmcore is used by various utilities, like crash and makedumpfile to
1041 * obtain access to symbols that are otherwise very difficult to locate. These
1042 * symbols are specifically used so that utilities can access and extract the
1043 * dmesg log from a vmcore file after a crash.
1045 void log_buf_vmcoreinfo_setup(void)
1047 VMCOREINFO_SYMBOL(log_buf);
1048 VMCOREINFO_SYMBOL(log_buf_len);
1049 VMCOREINFO_SYMBOL(log_first_idx);
1050 VMCOREINFO_SYMBOL(clear_idx);
1051 VMCOREINFO_SYMBOL(log_next_idx);
1053 * Export struct printk_log size and field offsets. User space tools can
1054 * parse it and detect any changes to structure down the line.
1056 VMCOREINFO_STRUCT_SIZE(printk_log);
1057 VMCOREINFO_OFFSET(printk_log, ts_nsec);
1058 VMCOREINFO_OFFSET(printk_log, len);
1059 VMCOREINFO_OFFSET(printk_log, text_len);
1060 VMCOREINFO_OFFSET(printk_log, dict_len);
1061 #ifdef CONFIG_PRINTK_CALLER
1062 VMCOREINFO_OFFSET(printk_log, caller_id);
1067 /* requested log_buf_len from kernel cmdline */
1068 static unsigned long __initdata new_log_buf_len;
1070 /* we practice scaling the ring buffer by powers of 2 */
1071 static void __init log_buf_len_update(u64 size)
1073 if (size > (u64)LOG_BUF_LEN_MAX) {
1074 size = (u64)LOG_BUF_LEN_MAX;
1075 pr_err("log_buf over 2G is not supported.\n");
1079 size = roundup_pow_of_two(size);
1080 if (size > log_buf_len)
1081 new_log_buf_len = (unsigned long)size;
1084 /* save requested log_buf_len since it's too early to process it */
1085 static int __init log_buf_len_setup(char *str)
1092 size = memparse(str, &str);
1094 log_buf_len_update(size);
1098 early_param("log_buf_len", log_buf_len_setup);
1101 #define __LOG_CPU_MAX_BUF_LEN (1 << CONFIG_LOG_CPU_MAX_BUF_SHIFT)
1103 static void __init log_buf_add_cpu(void)
1105 unsigned int cpu_extra;
1108 * archs should set up cpu_possible_bits properly with
1109 * set_cpu_possible() after setup_arch() but just in
1110 * case lets ensure this is valid.
1112 if (num_possible_cpus() == 1)
1115 cpu_extra = (num_possible_cpus() - 1) * __LOG_CPU_MAX_BUF_LEN;
1117 /* by default this will only continue through for large > 64 CPUs */
1118 if (cpu_extra <= __LOG_BUF_LEN / 2)
1121 pr_info("log_buf_len individual max cpu contribution: %d bytes\n",
1122 __LOG_CPU_MAX_BUF_LEN);
1123 pr_info("log_buf_len total cpu_extra contributions: %d bytes\n",
1125 pr_info("log_buf_len min size: %d bytes\n", __LOG_BUF_LEN);
1127 log_buf_len_update(cpu_extra + __LOG_BUF_LEN);
1129 #else /* !CONFIG_SMP */
1130 static inline void log_buf_add_cpu(void) {}
1131 #endif /* CONFIG_SMP */
1133 void __init setup_log_buf(int early)
1135 unsigned long flags;
1139 if (log_buf != __log_buf)
1142 if (!early && !new_log_buf_len)
1145 if (!new_log_buf_len)
1150 memblock_virt_alloc(new_log_buf_len, LOG_ALIGN);
1152 new_log_buf = memblock_virt_alloc_nopanic(new_log_buf_len,
1156 if (unlikely(!new_log_buf)) {
1157 pr_err("log_buf_len: %lu bytes not available\n",
1162 logbuf_lock_irqsave(flags);
1163 log_buf_len = new_log_buf_len;
1164 log_buf = new_log_buf;
1165 new_log_buf_len = 0;
1166 free = __LOG_BUF_LEN - log_next_idx;
1167 memcpy(log_buf, __log_buf, __LOG_BUF_LEN);
1168 logbuf_unlock_irqrestore(flags);
1170 pr_info("log_buf_len: %u bytes\n", log_buf_len);
1171 pr_info("early log buf free: %u(%u%%)\n",
1172 free, (free * 100) / __LOG_BUF_LEN);
1175 static bool __read_mostly ignore_loglevel;
1177 static int __init ignore_loglevel_setup(char *str)
1179 ignore_loglevel = true;
1180 pr_info("debug: ignoring loglevel setting.\n");
1185 early_param("ignore_loglevel", ignore_loglevel_setup);
1186 module_param(ignore_loglevel, bool, S_IRUGO | S_IWUSR);
1187 MODULE_PARM_DESC(ignore_loglevel,
1188 "ignore loglevel setting (prints all kernel messages to the console)");
1190 static bool suppress_message_printing(int level)
1192 return (level >= console_loglevel && !ignore_loglevel);
1195 #ifdef CONFIG_BOOT_PRINTK_DELAY
1197 static int boot_delay; /* msecs delay after each printk during bootup */
1198 static unsigned long long loops_per_msec; /* based on boot_delay */
1200 static int __init boot_delay_setup(char *str)
1204 lpj = preset_lpj ? preset_lpj : 1000000; /* some guess */
1205 loops_per_msec = (unsigned long long)lpj / 1000 * HZ;
1207 get_option(&str, &boot_delay);
1208 if (boot_delay > 10 * 1000)
1211 pr_debug("boot_delay: %u, preset_lpj: %ld, lpj: %lu, "
1212 "HZ: %d, loops_per_msec: %llu\n",
1213 boot_delay, preset_lpj, lpj, HZ, loops_per_msec);
1216 early_param("boot_delay", boot_delay_setup);
1218 static void boot_delay_msec(int level)
1220 unsigned long long k;
1221 unsigned long timeout;
1223 if ((boot_delay == 0 || system_state >= SYSTEM_RUNNING)
1224 || suppress_message_printing(level)) {
1228 k = (unsigned long long)loops_per_msec * boot_delay;
1230 timeout = jiffies + msecs_to_jiffies(boot_delay);
1235 * use (volatile) jiffies to prevent
1236 * compiler reduction; loop termination via jiffies
1237 * is secondary and may or may not happen.
1239 if (time_after(jiffies, timeout))
1241 touch_nmi_watchdog();
1245 static inline void boot_delay_msec(int level)
1250 static bool printk_time = IS_ENABLED(CONFIG_PRINTK_TIME);
1251 module_param_named(time, printk_time, bool, S_IRUGO | S_IWUSR);
1253 static size_t print_syslog(unsigned int level, char *buf)
1255 return sprintf(buf, "<%u>", level);
1258 static size_t print_time(u64 ts, char *buf)
1260 unsigned long rem_nsec = do_div(ts, 1000000000);
1262 return sprintf(buf, "[%5lu.%06lu]",
1263 (unsigned long)ts, rem_nsec / 1000);
1266 #ifdef CONFIG_PRINTK_CALLER
1267 static size_t print_caller(u32 id, char *buf)
1271 snprintf(caller, sizeof(caller), "%c%u",
1272 id & 0x80000000 ? 'C' : 'T', id & ~0x80000000);
1273 return sprintf(buf, "[%6s]", caller);
1276 #define print_caller(id, buf) 0
1279 static size_t print_prefix(const struct printk_log *msg, bool syslog,
1280 bool time, char *buf)
1285 len = print_syslog((msg->facility << 3) | msg->level, buf);
1288 len += print_time(msg->ts_nsec, buf + len);
1290 len += print_caller(msg->caller_id, buf + len);
1292 if (IS_ENABLED(CONFIG_PRINTK_CALLER) || time) {
1300 static size_t msg_print_text(const struct printk_log *msg, bool syslog,
1301 bool time, char *buf, size_t size)
1303 const char *text = log_text(msg);
1304 size_t text_size = msg->text_len;
1306 char prefix[PREFIX_MAX];
1307 const size_t prefix_len = print_prefix(msg, syslog, time, prefix);
1310 const char *next = memchr(text, '\n', text_size);
1314 text_len = next - text;
1316 text_size -= next - text;
1318 text_len = text_size;
1322 if (prefix_len + text_len + 1 >= size - len)
1325 memcpy(buf + len, prefix, prefix_len);
1327 memcpy(buf + len, text, text_len);
1331 /* SYSLOG_ACTION_* buffer size only calculation */
1332 len += prefix_len + text_len + 1;
1341 static int syslog_print(char __user *buf, int size)
1344 struct printk_log *msg;
1347 text = kmalloc(LOG_LINE_MAX + PREFIX_MAX, GFP_KERNEL);
1356 if (syslog_seq < log_first_seq) {
1357 /* messages are gone, move to first one */
1358 syslog_seq = log_first_seq;
1359 syslog_idx = log_first_idx;
1362 if (syslog_seq == log_next_seq) {
1363 logbuf_unlock_irq();
1368 * To keep reading/counting partial line consistent,
1369 * use printk_time value as of the beginning of a line.
1371 if (!syslog_partial)
1372 syslog_time = printk_time;
1374 skip = syslog_partial;
1375 msg = log_from_idx(syslog_idx);
1376 n = msg_print_text(msg, true, syslog_time, text,
1377 LOG_LINE_MAX + PREFIX_MAX);
1378 if (n - syslog_partial <= size) {
1379 /* message fits into buffer, move forward */
1380 syslog_idx = log_next(syslog_idx);
1382 n -= syslog_partial;
1385 /* partial read(), remember position */
1387 syslog_partial += n;
1390 logbuf_unlock_irq();
1395 if (copy_to_user(buf, text + skip, n)) {
1410 static int syslog_print_all(char __user *buf, int size, bool clear)
1419 text = kmalloc(LOG_LINE_MAX + PREFIX_MAX, GFP_KERNEL);
1426 * Find first record that fits, including all following records,
1427 * into the user-provided buffer for this dump.
1431 while (seq < log_next_seq) {
1432 struct printk_log *msg = log_from_idx(idx);
1434 len += msg_print_text(msg, true, time, NULL, 0);
1435 idx = log_next(idx);
1439 /* move first record forward until length fits into the buffer */
1442 while (len > size && seq < log_next_seq) {
1443 struct printk_log *msg = log_from_idx(idx);
1445 len -= msg_print_text(msg, true, time, NULL, 0);
1446 idx = log_next(idx);
1450 /* last message fitting into this dump */
1451 next_seq = log_next_seq;
1454 while (len >= 0 && seq < next_seq) {
1455 struct printk_log *msg = log_from_idx(idx);
1456 int textlen = msg_print_text(msg, true, time, text,
1457 LOG_LINE_MAX + PREFIX_MAX);
1459 idx = log_next(idx);
1462 logbuf_unlock_irq();
1463 if (copy_to_user(buf + len, text, textlen))
1469 if (seq < log_first_seq) {
1470 /* messages are gone, move to next one */
1471 seq = log_first_seq;
1472 idx = log_first_idx;
1477 clear_seq = log_next_seq;
1478 clear_idx = log_next_idx;
1480 logbuf_unlock_irq();
1486 static void syslog_clear(void)
1489 clear_seq = log_next_seq;
1490 clear_idx = log_next_idx;
1491 logbuf_unlock_irq();
1494 int do_syslog(int type, char __user *buf, int len, int source)
1497 static int saved_console_loglevel = LOGLEVEL_DEFAULT;
1500 error = check_syslog_permissions(type, source);
1505 case SYSLOG_ACTION_CLOSE: /* Close log */
1507 case SYSLOG_ACTION_OPEN: /* Open log */
1509 case SYSLOG_ACTION_READ: /* Read from log */
1510 if (!buf || len < 0)
1514 if (!access_ok(VERIFY_WRITE, buf, len))
1516 error = wait_event_interruptible(log_wait,
1517 syslog_seq != log_next_seq);
1520 error = syslog_print(buf, len);
1522 /* Read/clear last kernel messages */
1523 case SYSLOG_ACTION_READ_CLEAR:
1526 /* Read last kernel messages */
1527 case SYSLOG_ACTION_READ_ALL:
1528 if (!buf || len < 0)
1532 if (!access_ok(VERIFY_WRITE, buf, len))
1534 error = syslog_print_all(buf, len, clear);
1536 /* Clear ring buffer */
1537 case SYSLOG_ACTION_CLEAR:
1540 /* Disable logging to console */
1541 case SYSLOG_ACTION_CONSOLE_OFF:
1542 if (saved_console_loglevel == LOGLEVEL_DEFAULT)
1543 saved_console_loglevel = console_loglevel;
1544 console_loglevel = minimum_console_loglevel;
1546 /* Enable logging to console */
1547 case SYSLOG_ACTION_CONSOLE_ON:
1548 if (saved_console_loglevel != LOGLEVEL_DEFAULT) {
1549 console_loglevel = saved_console_loglevel;
1550 saved_console_loglevel = LOGLEVEL_DEFAULT;
1553 /* Set level of messages printed to console */
1554 case SYSLOG_ACTION_CONSOLE_LEVEL:
1555 if (len < 1 || len > 8)
1557 if (len < minimum_console_loglevel)
1558 len = minimum_console_loglevel;
1559 console_loglevel = len;
1560 /* Implicitly re-enable logging to console */
1561 saved_console_loglevel = LOGLEVEL_DEFAULT;
1563 /* Number of chars in the log buffer */
1564 case SYSLOG_ACTION_SIZE_UNREAD:
1566 if (syslog_seq < log_first_seq) {
1567 /* messages are gone, move to first one */
1568 syslog_seq = log_first_seq;
1569 syslog_idx = log_first_idx;
1572 if (source == SYSLOG_FROM_PROC) {
1574 * Short-cut for poll(/"proc/kmsg") which simply checks
1575 * for pending data, not the size; return the count of
1576 * records, not the length.
1578 error = log_next_seq - syslog_seq;
1580 u64 seq = syslog_seq;
1581 u32 idx = syslog_idx;
1582 bool time = syslog_partial ? syslog_time : printk_time;
1584 while (seq < log_next_seq) {
1585 struct printk_log *msg = log_from_idx(idx);
1587 error += msg_print_text(msg, true, time, NULL,
1590 idx = log_next(idx);
1593 error -= syslog_partial;
1595 logbuf_unlock_irq();
1597 /* Size of the log buffer */
1598 case SYSLOG_ACTION_SIZE_BUFFER:
1599 error = log_buf_len;
1609 SYSCALL_DEFINE3(syslog, int, type, char __user *, buf, int, len)
1611 return do_syslog(type, buf, len, SYSLOG_FROM_READER);
1615 * Special console_lock variants that help to reduce the risk of soft-lockups.
1616 * They allow to pass console_lock to another printk() call using a busy wait.
1619 #ifdef CONFIG_LOCKDEP
1620 static struct lockdep_map console_owner_dep_map = {
1621 .name = "console_owner"
1625 static DEFINE_RAW_SPINLOCK(console_owner_lock);
1626 static struct task_struct *console_owner;
1627 static bool console_waiter;
1630 * console_lock_spinning_enable - mark beginning of code where another
1631 * thread might safely busy wait
1633 * This basically converts console_lock into a spinlock. This marks
1634 * the section where the console_lock owner can not sleep, because
1635 * there may be a waiter spinning (like a spinlock). Also it must be
1636 * ready to hand over the lock at the end of the section.
1638 static void console_lock_spinning_enable(void)
1640 raw_spin_lock(&console_owner_lock);
1641 console_owner = current;
1642 raw_spin_unlock(&console_owner_lock);
1644 /* The waiter may spin on us after setting console_owner */
1645 spin_acquire(&console_owner_dep_map, 0, 0, _THIS_IP_);
1649 * console_lock_spinning_disable_and_check - mark end of code where another
1650 * thread was able to busy wait and check if there is a waiter
1652 * This is called at the end of the section where spinning is allowed.
1653 * It has two functions. First, it is a signal that it is no longer
1654 * safe to start busy waiting for the lock. Second, it checks if
1655 * there is a busy waiter and passes the lock rights to her.
1657 * Important: Callers lose the lock if there was a busy waiter.
1658 * They must not touch items synchronized by console_lock
1661 * Return: 1 if the lock rights were passed, 0 otherwise.
1663 static int console_lock_spinning_disable_and_check(void)
1667 raw_spin_lock(&console_owner_lock);
1668 waiter = READ_ONCE(console_waiter);
1669 console_owner = NULL;
1670 raw_spin_unlock(&console_owner_lock);
1673 spin_release(&console_owner_dep_map, 1, _THIS_IP_);
1677 /* The waiter is now free to continue */
1678 WRITE_ONCE(console_waiter, false);
1680 spin_release(&console_owner_dep_map, 1, _THIS_IP_);
1683 * Hand off console_lock to waiter. The waiter will perform
1684 * the up(). After this, the waiter is the console_lock owner.
1686 mutex_release(&console_lock_dep_map, 1, _THIS_IP_);
1691 * console_trylock_spinning - try to get console_lock by busy waiting
1693 * This allows to busy wait for the console_lock when the current
1694 * owner is running in specially marked sections. It means that
1695 * the current owner is running and cannot reschedule until it
1696 * is ready to lose the lock.
1698 * Return: 1 if we got the lock, 0 othrewise
1700 static int console_trylock_spinning(void)
1702 struct task_struct *owner = NULL;
1705 unsigned long flags;
1707 if (console_trylock())
1710 printk_safe_enter_irqsave(flags);
1712 raw_spin_lock(&console_owner_lock);
1713 owner = READ_ONCE(console_owner);
1714 waiter = READ_ONCE(console_waiter);
1715 if (!waiter && owner && owner != current) {
1716 WRITE_ONCE(console_waiter, true);
1719 raw_spin_unlock(&console_owner_lock);
1722 * If there is an active printk() writing to the
1723 * consoles, instead of having it write our data too,
1724 * see if we can offload that load from the active
1725 * printer, and do some printing ourselves.
1726 * Go into a spin only if there isn't already a waiter
1727 * spinning, and there is an active printer, and
1728 * that active printer isn't us (recursive printk?).
1731 printk_safe_exit_irqrestore(flags);
1735 /* We spin waiting for the owner to release us */
1736 spin_acquire(&console_owner_dep_map, 0, 0, _THIS_IP_);
1737 /* Owner will clear console_waiter on hand off */
1738 while (READ_ONCE(console_waiter))
1740 spin_release(&console_owner_dep_map, 1, _THIS_IP_);
1742 printk_safe_exit_irqrestore(flags);
1744 * The owner passed the console lock to us.
1745 * Since we did not spin on console lock, annotate
1746 * this as a trylock. Otherwise lockdep will
1749 mutex_acquire(&console_lock_dep_map, 0, 1, _THIS_IP_);
1755 * Call the console drivers, asking them to write out
1756 * log_buf[start] to log_buf[end - 1].
1757 * The console_lock must be held.
1759 static void call_console_drivers(const char *ext_text, size_t ext_len,
1760 const char *text, size_t len)
1762 struct console *con;
1764 trace_console_rcuidle(text, len);
1766 if (!console_drivers)
1769 for_each_console(con) {
1770 if (exclusive_console && con != exclusive_console)
1772 if (!(con->flags & CON_ENABLED))
1776 if (!cpu_online(smp_processor_id()) &&
1777 !(con->flags & CON_ANYTIME))
1779 if (con->flags & CON_EXTENDED)
1780 con->write(con, ext_text, ext_len);
1782 con->write(con, text, len);
1786 int printk_delay_msec __read_mostly;
1788 static inline void printk_delay(void)
1790 if (unlikely(printk_delay_msec)) {
1791 int m = printk_delay_msec;
1795 touch_nmi_watchdog();
1800 static inline u32 printk_caller_id(void)
1802 return in_task() ? task_pid_nr(current) :
1803 0x80000000 + raw_smp_processor_id();
1807 * Continuation lines are buffered, and not committed to the record buffer
1808 * until the line is complete, or a race forces it. The line fragments
1809 * though, are printed immediately to the consoles to ensure everything has
1810 * reached the console in case of a kernel crash.
1812 static struct cont {
1813 char buf[LOG_LINE_MAX];
1814 size_t len; /* length == 0 means unused buffer */
1815 u32 caller_id; /* printk_caller_id() of first print */
1816 u64 ts_nsec; /* time of first print */
1817 u8 level; /* log level of first message */
1818 u8 facility; /* log facility of first message */
1819 enum log_flags flags; /* prefix, newline flags */
1822 static void cont_flush(void)
1827 log_store(cont.caller_id, cont.facility, cont.level, cont.flags,
1828 cont.ts_nsec, NULL, 0, cont.buf, cont.len);
1832 static bool cont_add(u32 caller_id, int facility, int level,
1833 enum log_flags flags, const char *text, size_t len)
1835 /* If the line gets too long, split it up in separate records. */
1836 if (cont.len + len > sizeof(cont.buf)) {
1842 cont.facility = facility;
1844 cont.caller_id = caller_id;
1845 cont.ts_nsec = local_clock();
1849 memcpy(cont.buf + cont.len, text, len);
1852 // The original flags come from the first line,
1853 // but later continuations can add a newline.
1854 if (flags & LOG_NEWLINE) {
1855 cont.flags |= LOG_NEWLINE;
1862 static size_t log_output(int facility, int level, enum log_flags lflags, const char *dict, size_t dictlen, char *text, size_t text_len)
1864 const u32 caller_id = printk_caller_id();
1867 * If an earlier line was buffered, and we're a continuation
1868 * write from the same context, try to add it to the buffer.
1871 if (cont.caller_id == caller_id && (lflags & LOG_CONT)) {
1872 if (cont_add(caller_id, facility, level, lflags, text, text_len))
1875 /* Otherwise, make sure it's flushed */
1879 /* Skip empty continuation lines that couldn't be added - they just flush */
1880 if (!text_len && (lflags & LOG_CONT))
1883 /* If it doesn't end in a newline, try to buffer the current line */
1884 if (!(lflags & LOG_NEWLINE)) {
1885 if (cont_add(caller_id, facility, level, lflags, text, text_len))
1889 /* Store it in the record log */
1890 return log_store(caller_id, facility, level, lflags, 0,
1891 dict, dictlen, text, text_len);
1894 /* Must be called under logbuf_lock. */
1895 int vprintk_store(int facility, int level,
1896 const char *dict, size_t dictlen,
1897 const char *fmt, va_list args)
1899 static char textbuf[LOG_LINE_MAX];
1900 char *text = textbuf;
1902 enum log_flags lflags = 0;
1905 * The printf needs to come first; we need the syslog
1906 * prefix which might be passed-in as a parameter.
1908 text_len = vscnprintf(text, sizeof(textbuf), fmt, args);
1910 /* mark and strip a trailing newline */
1911 if (text_len && text[text_len-1] == '\n') {
1913 lflags |= LOG_NEWLINE;
1916 /* strip kernel syslog prefix and extract log level or control flags */
1917 if (facility == 0) {
1920 while ((kern_level = printk_get_level(text)) != 0) {
1921 switch (kern_level) {
1923 if (level == LOGLEVEL_DEFAULT)
1924 level = kern_level - '0';
1926 case 'd': /* KERN_DEFAULT */
1927 lflags |= LOG_PREFIX;
1929 case 'c': /* KERN_CONT */
1938 if (level == LOGLEVEL_DEFAULT)
1939 level = default_message_loglevel;
1942 lflags |= LOG_PREFIX|LOG_NEWLINE;
1944 return log_output(facility, level, lflags,
1945 dict, dictlen, text, text_len);
1948 asmlinkage int vprintk_emit(int facility, int level,
1949 const char *dict, size_t dictlen,
1950 const char *fmt, va_list args)
1953 bool in_sched = false, pending_output;
1954 unsigned long flags;
1957 if (level == LOGLEVEL_SCHED) {
1958 level = LOGLEVEL_DEFAULT;
1962 boot_delay_msec(level);
1965 /* This stops the holder of console_sem just where we want him */
1966 logbuf_lock_irqsave(flags);
1967 curr_log_seq = log_next_seq;
1968 printed_len = vprintk_store(facility, level, dict, dictlen, fmt, args);
1969 pending_output = (curr_log_seq != log_next_seq);
1970 logbuf_unlock_irqrestore(flags);
1972 /* If called from the scheduler, we can not call up(). */
1973 if (!in_sched && pending_output) {
1975 * Disable preemption to avoid being preempted while holding
1976 * console_sem which would prevent anyone from printing to
1981 * Try to acquire and then immediately release the console
1982 * semaphore. The release will print out buffers and wake up
1983 * /dev/kmsg and syslog() users.
1985 if (console_trylock_spinning())
1994 EXPORT_SYMBOL(vprintk_emit);
1996 asmlinkage int vprintk(const char *fmt, va_list args)
1998 return vprintk_func(fmt, args);
2000 EXPORT_SYMBOL(vprintk);
2002 int vprintk_default(const char *fmt, va_list args)
2006 #ifdef CONFIG_KGDB_KDB
2007 /* Allow to pass printk() to kdb but avoid a recursion. */
2008 if (unlikely(kdb_trap_printk && kdb_printf_cpu < 0)) {
2009 r = vkdb_printf(KDB_MSGSRC_PRINTK, fmt, args);
2013 r = vprintk_emit(0, LOGLEVEL_DEFAULT, NULL, 0, fmt, args);
2017 EXPORT_SYMBOL_GPL(vprintk_default);
2020 * printk - print a kernel message
2021 * @fmt: format string
2023 * This is printk(). It can be called from any context. We want it to work.
2025 * We try to grab the console_lock. If we succeed, it's easy - we log the
2026 * output and call the console drivers. If we fail to get the semaphore, we
2027 * place the output into the log buffer and return. The current holder of
2028 * the console_sem will notice the new output in console_unlock(); and will
2029 * send it to the consoles before releasing the lock.
2031 * One effect of this deferred printing is that code which calls printk() and
2032 * then changes console_loglevel may break. This is because console_loglevel
2033 * is inspected when the actual printing occurs.
2038 * See the vsnprintf() documentation for format string extensions over C99.
2040 asmlinkage __visible int printk(const char *fmt, ...)
2045 va_start(args, fmt);
2046 r = vprintk_func(fmt, args);
2051 EXPORT_SYMBOL(printk);
2053 #else /* CONFIG_PRINTK */
2055 #define LOG_LINE_MAX 0
2056 #define PREFIX_MAX 0
2057 #define printk_time false
2059 static u64 syslog_seq;
2060 static u32 syslog_idx;
2061 static u64 console_seq;
2062 static u32 console_idx;
2063 static u64 exclusive_console_stop_seq;
2064 static u64 log_first_seq;
2065 static u32 log_first_idx;
2066 static u64 log_next_seq;
2067 static char *log_text(const struct printk_log *msg) { return NULL; }
2068 static char *log_dict(const struct printk_log *msg) { return NULL; }
2069 static struct printk_log *log_from_idx(u32 idx) { return NULL; }
2070 static u32 log_next(u32 idx) { return 0; }
2071 static ssize_t msg_print_ext_header(char *buf, size_t size,
2072 struct printk_log *msg,
2073 u64 seq) { return 0; }
2074 static ssize_t msg_print_ext_body(char *buf, size_t size,
2075 char *dict, size_t dict_len,
2076 char *text, size_t text_len) { return 0; }
2077 static void console_lock_spinning_enable(void) { }
2078 static int console_lock_spinning_disable_and_check(void) { return 0; }
2079 static void call_console_drivers(const char *ext_text, size_t ext_len,
2080 const char *text, size_t len) {}
2081 static size_t msg_print_text(const struct printk_log *msg, bool syslog,
2082 bool time, char *buf, size_t size) { return 0; }
2083 static bool suppress_message_printing(int level) { return false; }
2085 #endif /* CONFIG_PRINTK */
2087 #ifdef CONFIG_EARLY_PRINTK
2088 struct console *early_console;
2090 asmlinkage __visible void early_printk(const char *fmt, ...)
2100 n = vscnprintf(buf, sizeof(buf), fmt, ap);
2103 early_console->write(early_console, buf, n);
2107 static int __add_preferred_console(char *name, int idx, char *options,
2110 struct console_cmdline *c;
2114 * See if this tty is not yet registered, and
2115 * if we have a slot free.
2117 for (i = 0, c = console_cmdline;
2118 i < MAX_CMDLINECONSOLES && c->name[0];
2120 if (strcmp(c->name, name) == 0 && c->index == idx) {
2122 preferred_console = i;
2126 if (i == MAX_CMDLINECONSOLES)
2129 preferred_console = i;
2130 strlcpy(c->name, name, sizeof(c->name));
2131 c->options = options;
2132 braille_set_options(c, brl_options);
2138 static int __init console_msg_format_setup(char *str)
2140 if (!strcmp(str, "syslog"))
2141 console_msg_format = MSG_FORMAT_SYSLOG;
2142 if (!strcmp(str, "default"))
2143 console_msg_format = MSG_FORMAT_DEFAULT;
2146 __setup("console_msg_format=", console_msg_format_setup);
2149 * Set up a console. Called via do_early_param() in init/main.c
2150 * for each "console=" parameter in the boot command line.
2152 static int __init console_setup(char *str)
2154 char buf[sizeof(console_cmdline[0].name) + 4]; /* 4 for "ttyS" */
2155 char *s, *options, *brl_options = NULL;
2158 if (_braille_console_setup(&str, &brl_options))
2162 * Decode str into name, index, options.
2164 if (str[0] >= '0' && str[0] <= '9') {
2165 strcpy(buf, "ttyS");
2166 strncpy(buf + 4, str, sizeof(buf) - 5);
2168 strncpy(buf, str, sizeof(buf) - 1);
2170 buf[sizeof(buf) - 1] = 0;
2171 options = strchr(str, ',');
2175 if (!strcmp(str, "ttya"))
2176 strcpy(buf, "ttyS0");
2177 if (!strcmp(str, "ttyb"))
2178 strcpy(buf, "ttyS1");
2180 for (s = buf; *s; s++)
2181 if (isdigit(*s) || *s == ',')
2183 idx = simple_strtoul(s, NULL, 10);
2186 __add_preferred_console(buf, idx, options, brl_options);
2187 console_set_on_cmdline = 1;
2190 __setup("console=", console_setup);
2193 * add_preferred_console - add a device to the list of preferred consoles.
2194 * @name: device name
2195 * @idx: device index
2196 * @options: options for this console
2198 * The last preferred console added will be used for kernel messages
2199 * and stdin/out/err for init. Normally this is used by console_setup
2200 * above to handle user-supplied console arguments; however it can also
2201 * be used by arch-specific code either to override the user or more
2202 * commonly to provide a default console (ie from PROM variables) when
2203 * the user has not supplied one.
2205 int add_preferred_console(char *name, int idx, char *options)
2207 return __add_preferred_console(name, idx, options, NULL);
2210 bool console_suspend_enabled = true;
2211 EXPORT_SYMBOL(console_suspend_enabled);
2213 static int __init console_suspend_disable(char *str)
2215 console_suspend_enabled = false;
2218 __setup("no_console_suspend", console_suspend_disable);
2219 module_param_named(console_suspend, console_suspend_enabled,
2220 bool, S_IRUGO | S_IWUSR);
2221 MODULE_PARM_DESC(console_suspend, "suspend console during suspend"
2222 " and hibernate operations");
2225 * suspend_console - suspend the console subsystem
2227 * This disables printk() while we go into suspend states
2229 void suspend_console(void)
2231 if (!console_suspend_enabled)
2233 pr_info("Suspending console(s) (use no_console_suspend to debug)\n");
2235 console_suspended = 1;
2239 void resume_console(void)
2241 if (!console_suspend_enabled)
2244 console_suspended = 0;
2249 * console_cpu_notify - print deferred console messages after CPU hotplug
2252 * If printk() is called from a CPU that is not online yet, the messages
2253 * will be printed on the console only if there are CON_ANYTIME consoles.
2254 * This function is called when a new CPU comes online (or fails to come
2255 * up) or goes offline.
2257 static int console_cpu_notify(unsigned int cpu)
2259 if (!cpuhp_tasks_frozen) {
2260 /* If trylock fails, someone else is doing the printing */
2261 if (console_trylock())
2268 * console_lock - lock the console system for exclusive use.
2270 * Acquires a lock which guarantees that the caller has
2271 * exclusive access to the console system and the console_drivers list.
2273 * Can sleep, returns nothing.
2275 void console_lock(void)
2280 if (console_suspended)
2283 console_may_schedule = 1;
2285 EXPORT_SYMBOL(console_lock);
2288 * console_trylock - try to lock the console system for exclusive use.
2290 * Try to acquire a lock which guarantees that the caller has exclusive
2291 * access to the console system and the console_drivers list.
2293 * returns 1 on success, and 0 on failure to acquire the lock.
2295 int console_trylock(void)
2297 if (down_trylock_console_sem())
2299 if (console_suspended) {
2304 console_may_schedule = 0;
2307 EXPORT_SYMBOL(console_trylock);
2309 int is_console_locked(void)
2311 return console_locked;
2313 EXPORT_SYMBOL(is_console_locked);
2316 * Check if we have any console that is capable of printing while cpu is
2317 * booting or shutting down. Requires console_sem.
2319 static int have_callable_console(void)
2321 struct console *con;
2323 for_each_console(con)
2324 if ((con->flags & CON_ENABLED) &&
2325 (con->flags & CON_ANYTIME))
2332 * Can we actually use the console at this time on this cpu?
2334 * Console drivers may assume that per-cpu resources have been allocated. So
2335 * unless they're explicitly marked as being able to cope (CON_ANYTIME) don't
2336 * call them until this CPU is officially up.
2338 static inline int can_use_console(void)
2340 return cpu_online(raw_smp_processor_id()) || have_callable_console();
2344 * console_unlock - unlock the console system
2346 * Releases the console_lock which the caller holds on the console system
2347 * and the console driver list.
2349 * While the console_lock was held, console output may have been buffered
2350 * by printk(). If this is the case, console_unlock(); emits
2351 * the output prior to releasing the lock.
2353 * If there is output waiting, we wake /dev/kmsg and syslog() users.
2355 * console_unlock(); may be called from any context.
2357 void console_unlock(void)
2359 static char ext_text[CONSOLE_EXT_LOG_MAX];
2360 static char text[LOG_LINE_MAX + PREFIX_MAX];
2361 unsigned long flags;
2362 bool do_cond_resched, retry;
2364 if (console_suspended) {
2370 * Console drivers are called with interrupts disabled, so
2371 * @console_may_schedule should be cleared before; however, we may
2372 * end up dumping a lot of lines, for example, if called from
2373 * console registration path, and should invoke cond_resched()
2374 * between lines if allowable. Not doing so can cause a very long
2375 * scheduling stall on a slow console leading to RCU stall and
2376 * softlockup warnings which exacerbate the issue with more
2377 * messages practically incapacitating the system.
2379 * console_trylock() is not able to detect the preemptive
2380 * context reliably. Therefore the value must be stored before
2381 * and cleared after the the "again" goto label.
2383 do_cond_resched = console_may_schedule;
2385 console_may_schedule = 0;
2388 * We released the console_sem lock, so we need to recheck if
2389 * cpu is online and (if not) is there at least one CON_ANYTIME
2392 if (!can_use_console()) {
2399 struct printk_log *msg;
2403 printk_safe_enter_irqsave(flags);
2404 raw_spin_lock(&logbuf_lock);
2405 if (console_seq < log_first_seq) {
2407 "** %llu printk messages dropped **\n",
2408 log_first_seq - console_seq);
2410 /* messages are gone, move to first one */
2411 console_seq = log_first_seq;
2412 console_idx = log_first_idx;
2417 if (console_seq == log_next_seq)
2420 msg = log_from_idx(console_idx);
2421 if (suppress_message_printing(msg->level)) {
2423 * Skip record we have buffered and already printed
2424 * directly to the console when we received it, and
2425 * record that has level above the console loglevel.
2427 console_idx = log_next(console_idx);
2432 /* Output to all consoles once old messages replayed. */
2433 if (unlikely(exclusive_console &&
2434 console_seq >= exclusive_console_stop_seq)) {
2435 exclusive_console = NULL;
2438 len += msg_print_text(msg,
2439 console_msg_format & MSG_FORMAT_SYSLOG,
2440 printk_time, text + len, sizeof(text) - len);
2441 if (nr_ext_console_drivers) {
2442 ext_len = msg_print_ext_header(ext_text,
2445 ext_len += msg_print_ext_body(ext_text + ext_len,
2446 sizeof(ext_text) - ext_len,
2447 log_dict(msg), msg->dict_len,
2448 log_text(msg), msg->text_len);
2450 console_idx = log_next(console_idx);
2452 raw_spin_unlock(&logbuf_lock);
2455 * While actively printing out messages, if another printk()
2456 * were to occur on another CPU, it may wait for this one to
2457 * finish. This task can not be preempted if there is a
2458 * waiter waiting to take over.
2460 console_lock_spinning_enable();
2462 stop_critical_timings(); /* don't trace print latency */
2463 call_console_drivers(ext_text, ext_len, text, len);
2464 start_critical_timings();
2466 if (console_lock_spinning_disable_and_check()) {
2467 printk_safe_exit_irqrestore(flags);
2471 printk_safe_exit_irqrestore(flags);
2473 if (do_cond_resched)
2479 raw_spin_unlock(&logbuf_lock);
2484 * Someone could have filled up the buffer again, so re-check if there's
2485 * something to flush. In case we cannot trylock the console_sem again,
2486 * there's a new owner and the console_unlock() from them will do the
2487 * flush, no worries.
2489 raw_spin_lock(&logbuf_lock);
2490 retry = console_seq != log_next_seq;
2491 raw_spin_unlock(&logbuf_lock);
2492 printk_safe_exit_irqrestore(flags);
2494 if (retry && console_trylock())
2497 EXPORT_SYMBOL(console_unlock);
2500 * console_conditional_schedule - yield the CPU if required
2502 * If the console code is currently allowed to sleep, and
2503 * if this CPU should yield the CPU to another task, do
2506 * Must be called within console_lock();.
2508 void __sched console_conditional_schedule(void)
2510 if (console_may_schedule)
2513 EXPORT_SYMBOL(console_conditional_schedule);
2515 void console_unblank(void)
2520 * console_unblank can no longer be called in interrupt context unless
2521 * oops_in_progress is set to 1..
2523 if (oops_in_progress) {
2524 if (down_trylock_console_sem() != 0)
2530 console_may_schedule = 0;
2532 if ((c->flags & CON_ENABLED) && c->unblank)
2538 * console_flush_on_panic - flush console content on panic
2540 * Immediately output all pending messages no matter what.
2542 void console_flush_on_panic(void)
2545 * If someone else is holding the console lock, trylock will fail
2546 * and may_schedule may be set. Ignore and proceed to unlock so
2547 * that messages are flushed out. As this can be called from any
2548 * context and we don't want to get preempted while flushing,
2549 * ensure may_schedule is cleared.
2552 console_may_schedule = 0;
2557 * Return the console tty driver structure and its associated index
2559 struct tty_driver *console_device(int *index)
2562 struct tty_driver *driver = NULL;
2565 for_each_console(c) {
2568 driver = c->device(c, index);
2577 * Prevent further output on the passed console device so that (for example)
2578 * serial drivers can disable console output before suspending a port, and can
2579 * re-enable output afterwards.
2581 void console_stop(struct console *console)
2584 console->flags &= ~CON_ENABLED;
2587 EXPORT_SYMBOL(console_stop);
2589 void console_start(struct console *console)
2592 console->flags |= CON_ENABLED;
2595 EXPORT_SYMBOL(console_start);
2597 static int __read_mostly keep_bootcon;
2599 static int __init keep_bootcon_setup(char *str)
2602 pr_info("debug: skip boot console de-registration.\n");
2607 early_param("keep_bootcon", keep_bootcon_setup);
2610 * The console driver calls this routine during kernel initialization
2611 * to register the console printing procedure with printk() and to
2612 * print any messages that were printed by the kernel before the
2613 * console driver was initialized.
2615 * This can happen pretty early during the boot process (because of
2616 * early_printk) - sometimes before setup_arch() completes - be careful
2617 * of what kernel features are used - they may not be initialised yet.
2619 * There are two types of consoles - bootconsoles (early_printk) and
2620 * "real" consoles (everything which is not a bootconsole) which are
2621 * handled differently.
2622 * - Any number of bootconsoles can be registered at any time.
2623 * - As soon as a "real" console is registered, all bootconsoles
2624 * will be unregistered automatically.
2625 * - Once a "real" console is registered, any attempt to register a
2626 * bootconsoles will be rejected
2628 void register_console(struct console *newcon)
2631 unsigned long flags;
2632 struct console *bcon = NULL;
2633 struct console_cmdline *c;
2634 static bool has_preferred;
2636 if (console_drivers)
2637 for_each_console(bcon)
2638 if (WARN(bcon == newcon,
2639 "console '%s%d' already registered\n",
2640 bcon->name, bcon->index))
2644 * before we register a new CON_BOOT console, make sure we don't
2645 * already have a valid console
2647 if (console_drivers && newcon->flags & CON_BOOT) {
2648 /* find the last or real console */
2649 for_each_console(bcon) {
2650 if (!(bcon->flags & CON_BOOT)) {
2651 pr_info("Too late to register bootconsole %s%d\n",
2652 newcon->name, newcon->index);
2658 if (console_drivers && console_drivers->flags & CON_BOOT)
2659 bcon = console_drivers;
2661 if (!has_preferred || bcon || !console_drivers)
2662 has_preferred = preferred_console >= 0;
2665 * See if we want to use this console driver. If we
2666 * didn't select a console we take the first one
2667 * that registers here.
2669 if (!has_preferred) {
2670 if (newcon->index < 0)
2672 if (newcon->setup == NULL ||
2673 newcon->setup(newcon, NULL) == 0) {
2674 newcon->flags |= CON_ENABLED;
2675 if (newcon->device) {
2676 newcon->flags |= CON_CONSDEV;
2677 has_preferred = true;
2683 * See if this console matches one we selected on
2686 for (i = 0, c = console_cmdline;
2687 i < MAX_CMDLINECONSOLES && c->name[0];
2689 if (!newcon->match ||
2690 newcon->match(newcon, c->name, c->index, c->options) != 0) {
2691 /* default matching */
2692 BUILD_BUG_ON(sizeof(c->name) != sizeof(newcon->name));
2693 if (strcmp(c->name, newcon->name) != 0)
2695 if (newcon->index >= 0 &&
2696 newcon->index != c->index)
2698 if (newcon->index < 0)
2699 newcon->index = c->index;
2701 if (_braille_register_console(newcon, c))
2704 if (newcon->setup &&
2705 newcon->setup(newcon, c->options) != 0)
2709 newcon->flags |= CON_ENABLED;
2710 if (i == preferred_console) {
2711 newcon->flags |= CON_CONSDEV;
2712 has_preferred = true;
2717 if (!(newcon->flags & CON_ENABLED))
2721 * If we have a bootconsole, and are switching to a real console,
2722 * don't print everything out again, since when the boot console, and
2723 * the real console are the same physical device, it's annoying to
2724 * see the beginning boot messages twice
2726 if (bcon && ((newcon->flags & (CON_CONSDEV | CON_BOOT)) == CON_CONSDEV))
2727 newcon->flags &= ~CON_PRINTBUFFER;
2730 * Put this console in the list - keep the
2731 * preferred driver at the head of the list.
2734 if ((newcon->flags & CON_CONSDEV) || console_drivers == NULL) {
2735 newcon->next = console_drivers;
2736 console_drivers = newcon;
2738 newcon->next->flags &= ~CON_CONSDEV;
2740 newcon->next = console_drivers->next;
2741 console_drivers->next = newcon;
2744 if (newcon->flags & CON_EXTENDED)
2745 nr_ext_console_drivers++;
2747 if (newcon->flags & CON_PRINTBUFFER) {
2749 * console_unlock(); will print out the buffered messages
2752 logbuf_lock_irqsave(flags);
2753 console_seq = syslog_seq;
2754 console_idx = syslog_idx;
2756 * We're about to replay the log buffer. Only do this to the
2757 * just-registered console to avoid excessive message spam to
2758 * the already-registered consoles.
2760 * Set exclusive_console with disabled interrupts to reduce
2761 * race window with eventual console_flush_on_panic() that
2762 * ignores console_lock.
2764 exclusive_console = newcon;
2765 exclusive_console_stop_seq = console_seq;
2766 logbuf_unlock_irqrestore(flags);
2769 console_sysfs_notify();
2772 * By unregistering the bootconsoles after we enable the real console
2773 * we get the "console xxx enabled" message on all the consoles -
2774 * boot consoles, real consoles, etc - this is to ensure that end
2775 * users know there might be something in the kernel's log buffer that
2776 * went to the bootconsole (that they do not see on the real console)
2778 pr_info("%sconsole [%s%d] enabled\n",
2779 (newcon->flags & CON_BOOT) ? "boot" : "" ,
2780 newcon->name, newcon->index);
2782 ((newcon->flags & (CON_CONSDEV | CON_BOOT)) == CON_CONSDEV) &&
2784 /* We need to iterate through all boot consoles, to make
2785 * sure we print everything out, before we unregister them.
2787 for_each_console(bcon)
2788 if (bcon->flags & CON_BOOT)
2789 unregister_console(bcon);
2792 EXPORT_SYMBOL(register_console);
2794 int unregister_console(struct console *console)
2796 struct console *a, *b;
2799 pr_info("%sconsole [%s%d] disabled\n",
2800 (console->flags & CON_BOOT) ? "boot" : "" ,
2801 console->name, console->index);
2803 res = _braille_unregister_console(console);
2809 if (console_drivers == console) {
2810 console_drivers=console->next;
2812 } else if (console_drivers) {
2813 for (a=console_drivers->next, b=console_drivers ;
2814 a; b=a, a=b->next) {
2823 if (!res && (console->flags & CON_EXTENDED))
2824 nr_ext_console_drivers--;
2827 * If this isn't the last console and it has CON_CONSDEV set, we
2828 * need to set it on the next preferred console.
2830 if (console_drivers != NULL && console->flags & CON_CONSDEV)
2831 console_drivers->flags |= CON_CONSDEV;
2833 console->flags &= ~CON_ENABLED;
2835 console_sysfs_notify();
2838 EXPORT_SYMBOL(unregister_console);
2841 * Initialize the console device. This is called *early*, so
2842 * we can't necessarily depend on lots of kernel help here.
2843 * Just do some early initializations, and do the complex setup
2846 void __init console_init(void)
2850 initcall_entry_t *ce;
2852 /* Setup the default TTY line discipline. */
2856 * set up the console device so that later boot sequences can
2857 * inform about problems etc..
2859 ce = __con_initcall_start;
2860 trace_initcall_level("console");
2861 while (ce < __con_initcall_end) {
2862 call = initcall_from_entry(ce);
2863 trace_initcall_start(call);
2865 trace_initcall_finish(call, ret);
2871 * Some boot consoles access data that is in the init section and which will
2872 * be discarded after the initcalls have been run. To make sure that no code
2873 * will access this data, unregister the boot consoles in a late initcall.
2875 * If for some reason, such as deferred probe or the driver being a loadable
2876 * module, the real console hasn't registered yet at this point, there will
2877 * be a brief interval in which no messages are logged to the console, which
2878 * makes it difficult to diagnose problems that occur during this time.
2880 * To mitigate this problem somewhat, only unregister consoles whose memory
2881 * intersects with the init section. Note that all other boot consoles will
2882 * get unregistred when the real preferred console is registered.
2884 static int __init printk_late_init(void)
2886 struct console *con;
2889 for_each_console(con) {
2890 if (!(con->flags & CON_BOOT))
2893 /* Check addresses that might be used for enabled consoles. */
2894 if (init_section_intersects(con, sizeof(*con)) ||
2895 init_section_contains(con->write, 0) ||
2896 init_section_contains(con->read, 0) ||
2897 init_section_contains(con->device, 0) ||
2898 init_section_contains(con->unblank, 0) ||
2899 init_section_contains(con->data, 0)) {
2901 * Please, consider moving the reported consoles out
2902 * of the init section.
2904 pr_warn("bootconsole [%s%d] uses init memory and must be disabled even before the real one is ready\n",
2905 con->name, con->index);
2906 unregister_console(con);
2909 ret = cpuhp_setup_state_nocalls(CPUHP_PRINTK_DEAD, "printk:dead", NULL,
2910 console_cpu_notify);
2912 ret = cpuhp_setup_state_nocalls(CPUHP_AP_ONLINE_DYN, "printk:online",
2913 console_cpu_notify, NULL);
2917 late_initcall(printk_late_init);
2919 #if defined CONFIG_PRINTK
2921 * Delayed printk version, for scheduler-internal messages:
2923 #define PRINTK_PENDING_WAKEUP 0x01
2924 #define PRINTK_PENDING_OUTPUT 0x02
2926 static DEFINE_PER_CPU(int, printk_pending);
2928 static void wake_up_klogd_work_func(struct irq_work *irq_work)
2930 int pending = __this_cpu_xchg(printk_pending, 0);
2932 if (pending & PRINTK_PENDING_OUTPUT) {
2933 /* If trylock fails, someone else is doing the printing */
2934 if (console_trylock())
2938 if (pending & PRINTK_PENDING_WAKEUP)
2939 wake_up_interruptible(&log_wait);
2942 static DEFINE_PER_CPU(struct irq_work, wake_up_klogd_work) = {
2943 .func = wake_up_klogd_work_func,
2944 .flags = IRQ_WORK_LAZY,
2947 void wake_up_klogd(void)
2950 if (waitqueue_active(&log_wait)) {
2951 this_cpu_or(printk_pending, PRINTK_PENDING_WAKEUP);
2952 irq_work_queue(this_cpu_ptr(&wake_up_klogd_work));
2957 void defer_console_output(void)
2960 __this_cpu_or(printk_pending, PRINTK_PENDING_OUTPUT);
2961 irq_work_queue(this_cpu_ptr(&wake_up_klogd_work));
2965 int vprintk_deferred(const char *fmt, va_list args)
2969 r = vprintk_emit(0, LOGLEVEL_SCHED, NULL, 0, fmt, args);
2970 defer_console_output();
2975 int printk_deferred(const char *fmt, ...)
2980 va_start(args, fmt);
2981 r = vprintk_deferred(fmt, args);
2988 * printk rate limiting, lifted from the networking subsystem.
2990 * This enforces a rate limit: not more than 10 kernel messages
2991 * every 5s to make a denial-of-service attack impossible.
2993 DEFINE_RATELIMIT_STATE(printk_ratelimit_state, 5 * HZ, 10);
2995 int __printk_ratelimit(const char *func)
2997 return ___ratelimit(&printk_ratelimit_state, func);
2999 EXPORT_SYMBOL(__printk_ratelimit);
3002 * printk_timed_ratelimit - caller-controlled printk ratelimiting
3003 * @caller_jiffies: pointer to caller's state
3004 * @interval_msecs: minimum interval between prints
3006 * printk_timed_ratelimit() returns true if more than @interval_msecs
3007 * milliseconds have elapsed since the last time printk_timed_ratelimit()
3010 bool printk_timed_ratelimit(unsigned long *caller_jiffies,
3011 unsigned int interval_msecs)
3013 unsigned long elapsed = jiffies - *caller_jiffies;
3015 if (*caller_jiffies && elapsed <= msecs_to_jiffies(interval_msecs))
3018 *caller_jiffies = jiffies;
3021 EXPORT_SYMBOL(printk_timed_ratelimit);
3023 static DEFINE_SPINLOCK(dump_list_lock);
3024 static LIST_HEAD(dump_list);
3027 * kmsg_dump_register - register a kernel log dumper.
3028 * @dumper: pointer to the kmsg_dumper structure
3030 * Adds a kernel log dumper to the system. The dump callback in the
3031 * structure will be called when the kernel oopses or panics and must be
3032 * set. Returns zero on success and %-EINVAL or %-EBUSY otherwise.
3034 int kmsg_dump_register(struct kmsg_dumper *dumper)
3036 unsigned long flags;
3039 /* The dump callback needs to be set */
3043 spin_lock_irqsave(&dump_list_lock, flags);
3044 /* Don't allow registering multiple times */
3045 if (!dumper->registered) {
3046 dumper->registered = 1;
3047 list_add_tail_rcu(&dumper->list, &dump_list);
3050 spin_unlock_irqrestore(&dump_list_lock, flags);
3054 EXPORT_SYMBOL_GPL(kmsg_dump_register);
3057 * kmsg_dump_unregister - unregister a kmsg dumper.
3058 * @dumper: pointer to the kmsg_dumper structure
3060 * Removes a dump device from the system. Returns zero on success and
3061 * %-EINVAL otherwise.
3063 int kmsg_dump_unregister(struct kmsg_dumper *dumper)
3065 unsigned long flags;
3068 spin_lock_irqsave(&dump_list_lock, flags);
3069 if (dumper->registered) {
3070 dumper->registered = 0;
3071 list_del_rcu(&dumper->list);
3074 spin_unlock_irqrestore(&dump_list_lock, flags);
3079 EXPORT_SYMBOL_GPL(kmsg_dump_unregister);
3081 static bool always_kmsg_dump;
3082 module_param_named(always_kmsg_dump, always_kmsg_dump, bool, S_IRUGO | S_IWUSR);
3085 * kmsg_dump - dump kernel log to kernel message dumpers.
3086 * @reason: the reason (oops, panic etc) for dumping
3088 * Call each of the registered dumper's dump() callback, which can
3089 * retrieve the kmsg records with kmsg_dump_get_line() or
3090 * kmsg_dump_get_buffer().
3092 void kmsg_dump(enum kmsg_dump_reason reason)
3094 struct kmsg_dumper *dumper;
3095 unsigned long flags;
3097 if ((reason > KMSG_DUMP_OOPS) && !always_kmsg_dump)
3101 list_for_each_entry_rcu(dumper, &dump_list, list) {
3102 if (dumper->max_reason && reason > dumper->max_reason)
3105 /* initialize iterator with data about the stored records */
3106 dumper->active = true;
3108 logbuf_lock_irqsave(flags);
3109 dumper->cur_seq = clear_seq;
3110 dumper->cur_idx = clear_idx;
3111 dumper->next_seq = log_next_seq;
3112 dumper->next_idx = log_next_idx;
3113 logbuf_unlock_irqrestore(flags);
3115 /* invoke dumper which will iterate over records */
3116 dumper->dump(dumper, reason);
3118 /* reset iterator */
3119 dumper->active = false;
3125 * kmsg_dump_get_line_nolock - retrieve one kmsg log line (unlocked version)
3126 * @dumper: registered kmsg dumper
3127 * @syslog: include the "<4>" prefixes
3128 * @line: buffer to copy the line to
3129 * @size: maximum size of the buffer
3130 * @len: length of line placed into buffer
3132 * Start at the beginning of the kmsg buffer, with the oldest kmsg
3133 * record, and copy one record into the provided buffer.
3135 * Consecutive calls will return the next available record moving
3136 * towards the end of the buffer with the youngest messages.
3138 * A return value of FALSE indicates that there are no more records to
3141 * The function is similar to kmsg_dump_get_line(), but grabs no locks.
3143 bool kmsg_dump_get_line_nolock(struct kmsg_dumper *dumper, bool syslog,
3144 char *line, size_t size, size_t *len)
3146 struct printk_log *msg;
3150 if (!dumper->active)
3153 if (dumper->cur_seq < log_first_seq) {
3154 /* messages are gone, move to first available one */
3155 dumper->cur_seq = log_first_seq;
3156 dumper->cur_idx = log_first_idx;
3160 if (dumper->cur_seq >= log_next_seq)
3163 msg = log_from_idx(dumper->cur_idx);
3164 l = msg_print_text(msg, syslog, printk_time, line, size);
3166 dumper->cur_idx = log_next(dumper->cur_idx);
3176 * kmsg_dump_get_line - retrieve one kmsg log line
3177 * @dumper: registered kmsg dumper
3178 * @syslog: include the "<4>" prefixes
3179 * @line: buffer to copy the line to
3180 * @size: maximum size of the buffer
3181 * @len: length of line placed into buffer
3183 * Start at the beginning of the kmsg buffer, with the oldest kmsg
3184 * record, and copy one record into the provided buffer.
3186 * Consecutive calls will return the next available record moving
3187 * towards the end of the buffer with the youngest messages.
3189 * A return value of FALSE indicates that there are no more records to
3192 bool kmsg_dump_get_line(struct kmsg_dumper *dumper, bool syslog,
3193 char *line, size_t size, size_t *len)
3195 unsigned long flags;
3198 logbuf_lock_irqsave(flags);
3199 ret = kmsg_dump_get_line_nolock(dumper, syslog, line, size, len);
3200 logbuf_unlock_irqrestore(flags);
3204 EXPORT_SYMBOL_GPL(kmsg_dump_get_line);
3207 * kmsg_dump_get_buffer - copy kmsg log lines
3208 * @dumper: registered kmsg dumper
3209 * @syslog: include the "<4>" prefixes
3210 * @buf: buffer to copy the line to
3211 * @size: maximum size of the buffer
3212 * @len: length of line placed into buffer
3214 * Start at the end of the kmsg buffer and fill the provided buffer
3215 * with as many of the the *youngest* kmsg records that fit into it.
3216 * If the buffer is large enough, all available kmsg records will be
3217 * copied with a single call.
3219 * Consecutive calls will fill the buffer with the next block of
3220 * available older records, not including the earlier retrieved ones.
3222 * A return value of FALSE indicates that there are no more records to
3225 bool kmsg_dump_get_buffer(struct kmsg_dumper *dumper, bool syslog,
3226 char *buf, size_t size, size_t *len)
3228 unsigned long flags;
3235 bool time = printk_time;
3237 if (!dumper->active)
3240 logbuf_lock_irqsave(flags);
3241 if (dumper->cur_seq < log_first_seq) {
3242 /* messages are gone, move to first available one */
3243 dumper->cur_seq = log_first_seq;
3244 dumper->cur_idx = log_first_idx;
3248 if (dumper->cur_seq >= dumper->next_seq) {
3249 logbuf_unlock_irqrestore(flags);
3253 /* calculate length of entire buffer */
3254 seq = dumper->cur_seq;
3255 idx = dumper->cur_idx;
3256 while (seq < dumper->next_seq) {
3257 struct printk_log *msg = log_from_idx(idx);
3259 l += msg_print_text(msg, true, time, NULL, 0);
3260 idx = log_next(idx);
3264 /* move first record forward until length fits into the buffer */
3265 seq = dumper->cur_seq;
3266 idx = dumper->cur_idx;
3267 while (l > size && seq < dumper->next_seq) {
3268 struct printk_log *msg = log_from_idx(idx);
3270 l -= msg_print_text(msg, true, time, NULL, 0);
3271 idx = log_next(idx);
3275 /* last message in next interation */
3280 while (seq < dumper->next_seq) {
3281 struct printk_log *msg = log_from_idx(idx);
3283 l += msg_print_text(msg, syslog, time, buf + l, size - l);
3284 idx = log_next(idx);
3288 dumper->next_seq = next_seq;
3289 dumper->next_idx = next_idx;
3291 logbuf_unlock_irqrestore(flags);
3297 EXPORT_SYMBOL_GPL(kmsg_dump_get_buffer);
3300 * kmsg_dump_rewind_nolock - reset the interator (unlocked version)
3301 * @dumper: registered kmsg dumper
3303 * Reset the dumper's iterator so that kmsg_dump_get_line() and
3304 * kmsg_dump_get_buffer() can be called again and used multiple
3305 * times within the same dumper.dump() callback.
3307 * The function is similar to kmsg_dump_rewind(), but grabs no locks.
3309 void kmsg_dump_rewind_nolock(struct kmsg_dumper *dumper)
3311 dumper->cur_seq = clear_seq;
3312 dumper->cur_idx = clear_idx;
3313 dumper->next_seq = log_next_seq;
3314 dumper->next_idx = log_next_idx;
3318 * kmsg_dump_rewind - reset the interator
3319 * @dumper: registered kmsg dumper
3321 * Reset the dumper's iterator so that kmsg_dump_get_line() and
3322 * kmsg_dump_get_buffer() can be called again and used multiple
3323 * times within the same dumper.dump() callback.
3325 void kmsg_dump_rewind(struct kmsg_dumper *dumper)
3327 unsigned long flags;
3329 logbuf_lock_irqsave(flags);
3330 kmsg_dump_rewind_nolock(dumper);
3331 logbuf_unlock_irqrestore(flags);
3333 EXPORT_SYMBOL_GPL(kmsg_dump_rewind);