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 #include <linux/kernel.h>
21 #include <linux/tty.h>
22 #include <linux/tty_driver.h>
23 #include <linux/console.h>
24 #include <linux/init.h>
25 #include <linux/jiffies.h>
26 #include <linux/nmi.h>
27 #include <linux/module.h>
28 #include <linux/moduleparam.h>
29 #include <linux/delay.h>
30 #include <linux/smp.h>
31 #include <linux/security.h>
32 #include <linux/bootmem.h>
33 #include <linux/memblock.h>
34 #include <linux/syscalls.h>
35 #include <linux/kexec.h>
36 #include <linux/kdb.h>
37 #include <linux/ratelimit.h>
38 #include <linux/kmsg_dump.h>
39 #include <linux/syslog.h>
40 #include <linux/cpu.h>
41 #include <linux/notifier.h>
42 #include <linux/rculist.h>
43 #include <linux/poll.h>
44 #include <linux/irq_work.h>
45 #include <linux/utsname.h>
46 #include <linux/ctype.h>
47 #include <linux/uio.h>
49 #include <asm/uaccess.h>
50 #include <asm/sections.h>
52 #define CREATE_TRACE_POINTS
53 #include <trace/events/printk.h>
55 #include "console_cmdline.h"
59 int console_printk[4] = {
60 CONSOLE_LOGLEVEL_DEFAULT, /* console_loglevel */
61 MESSAGE_LOGLEVEL_DEFAULT, /* default_message_loglevel */
62 CONSOLE_LOGLEVEL_MIN, /* minimum_console_loglevel */
63 CONSOLE_LOGLEVEL_DEFAULT, /* default_console_loglevel */
67 * Low level drivers may need that to know if they can schedule in
68 * their unblank() callback or not. So let's export it.
71 EXPORT_SYMBOL(oops_in_progress);
74 * console_sem protects the console_drivers list, and also
75 * provides serialisation for access to the entire console
78 static DEFINE_SEMAPHORE(console_sem);
79 struct console *console_drivers;
80 EXPORT_SYMBOL_GPL(console_drivers);
83 static struct lockdep_map console_lock_dep_map = {
84 .name = "console_lock"
88 enum devkmsg_log_bits {
89 __DEVKMSG_LOG_BIT_ON = 0,
90 __DEVKMSG_LOG_BIT_OFF,
91 __DEVKMSG_LOG_BIT_LOCK,
94 enum devkmsg_log_masks {
95 DEVKMSG_LOG_MASK_ON = BIT(__DEVKMSG_LOG_BIT_ON),
96 DEVKMSG_LOG_MASK_OFF = BIT(__DEVKMSG_LOG_BIT_OFF),
97 DEVKMSG_LOG_MASK_LOCK = BIT(__DEVKMSG_LOG_BIT_LOCK),
100 /* Keep both the 'on' and 'off' bits clear, i.e. ratelimit by default: */
101 #define DEVKMSG_LOG_MASK_DEFAULT 0
103 static unsigned int __read_mostly devkmsg_log = DEVKMSG_LOG_MASK_DEFAULT;
105 static int __control_devkmsg(char *str)
110 if (!strncmp(str, "on", 2)) {
111 devkmsg_log = DEVKMSG_LOG_MASK_ON;
113 } else if (!strncmp(str, "off", 3)) {
114 devkmsg_log = DEVKMSG_LOG_MASK_OFF;
116 } else if (!strncmp(str, "ratelimit", 9)) {
117 devkmsg_log = DEVKMSG_LOG_MASK_DEFAULT;
123 static int __init control_devkmsg(char *str)
125 if (__control_devkmsg(str) < 0)
129 * Set sysctl string accordingly:
131 if (devkmsg_log == DEVKMSG_LOG_MASK_ON) {
132 memset(devkmsg_log_str, 0, DEVKMSG_STR_MAX_SIZE);
133 strncpy(devkmsg_log_str, "on", 2);
134 } else if (devkmsg_log == DEVKMSG_LOG_MASK_OFF) {
135 memset(devkmsg_log_str, 0, DEVKMSG_STR_MAX_SIZE);
136 strncpy(devkmsg_log_str, "off", 3);
138 /* else "ratelimit" which is set by default. */
141 * Sysctl cannot change it anymore. The kernel command line setting of
142 * this parameter is to force the setting to be permanent throughout the
143 * runtime of the system. This is a precation measure against userspace
144 * trying to be a smarta** and attempting to change it up on us.
146 devkmsg_log |= DEVKMSG_LOG_MASK_LOCK;
150 __setup("printk.devkmsg=", control_devkmsg);
152 char devkmsg_log_str[DEVKMSG_STR_MAX_SIZE] = "ratelimit";
154 int devkmsg_sysctl_set_loglvl(struct ctl_table *table, int write,
155 void __user *buffer, size_t *lenp, loff_t *ppos)
157 char old_str[DEVKMSG_STR_MAX_SIZE];
162 if (devkmsg_log & DEVKMSG_LOG_MASK_LOCK)
166 strncpy(old_str, devkmsg_log_str, DEVKMSG_STR_MAX_SIZE);
169 err = proc_dostring(table, write, buffer, lenp, ppos);
174 err = __control_devkmsg(devkmsg_log_str);
177 * Do not accept an unknown string OR a known string with
180 if (err < 0 || (err + 1 != *lenp)) {
182 /* ... and restore old setting. */
184 strncpy(devkmsg_log_str, old_str, DEVKMSG_STR_MAX_SIZE);
194 * Number of registered extended console drivers.
196 * If extended consoles are present, in-kernel cont reassembly is disabled
197 * and each fragment is stored as a separate log entry with proper
198 * continuation flag so that every emitted message has full metadata. This
199 * doesn't change the result for regular consoles or /proc/kmsg. For
200 * /dev/kmsg, as long as the reader concatenates messages according to
201 * consecutive continuation flags, the end result should be the same too.
203 static int nr_ext_console_drivers;
206 * Helper macros to handle lockdep when locking/unlocking console_sem. We use
207 * macros instead of functions so that _RET_IP_ contains useful information.
209 #define down_console_sem() do { \
211 mutex_acquire(&console_lock_dep_map, 0, 0, _RET_IP_);\
214 static int __down_trylock_console_sem(unsigned long ip)
216 if (down_trylock(&console_sem))
218 mutex_acquire(&console_lock_dep_map, 0, 1, ip);
221 #define down_trylock_console_sem() __down_trylock_console_sem(_RET_IP_)
223 #define up_console_sem() do { \
224 mutex_release(&console_lock_dep_map, 1, _RET_IP_);\
229 * This is used for debugging the mess that is the VT code by
230 * keeping track if we have the console semaphore held. It's
231 * definitely not the perfect debug tool (we don't know if _WE_
232 * hold it and are racing, but it helps tracking those weird code
233 * paths in the console code where we end up in places I want
234 * locked without the console sempahore held).
236 static int console_locked, console_suspended;
239 * If exclusive_console is non-NULL then only this console is to be printed to.
241 static struct console *exclusive_console;
244 * Array of consoles built from command line options (console=)
247 #define MAX_CMDLINECONSOLES 8
249 static struct console_cmdline console_cmdline[MAX_CMDLINECONSOLES];
251 static int selected_console = -1;
252 static int preferred_console = -1;
253 int console_set_on_cmdline;
254 EXPORT_SYMBOL(console_set_on_cmdline);
257 static bool of_specified_console;
259 void console_set_by_of(void)
261 of_specified_console = true;
264 # define of_specified_console false
267 /* Flag: console code may call schedule() */
268 static int console_may_schedule;
271 * The printk log buffer consists of a chain of concatenated variable
272 * length records. Every record starts with a record header, containing
273 * the overall length of the record.
275 * The heads to the first and last entry in the buffer, as well as the
276 * sequence numbers of these entries are maintained when messages are
279 * If the heads indicate available messages, the length in the header
280 * tells the start next message. A length == 0 for the next message
281 * indicates a wrap-around to the beginning of the buffer.
283 * Every record carries the monotonic timestamp in microseconds, as well as
284 * the standard userspace syslog level and syslog facility. The usual
285 * kernel messages use LOG_KERN; userspace-injected messages always carry
286 * a matching syslog facility, by default LOG_USER. The origin of every
287 * message can be reliably determined that way.
289 * The human readable log message directly follows the message header. The
290 * length of the message text is stored in the header, the stored message
293 * Optionally, a message can carry a dictionary of properties (key/value pairs),
294 * to provide userspace with a machine-readable message context.
296 * Examples for well-defined, commonly used property names are:
297 * DEVICE=b12:8 device identifier
301 * +sound:card0 subsystem:devname
302 * SUBSYSTEM=pci driver-core subsystem name
304 * Valid characters in property names are [a-zA-Z0-9.-_]. The plain text value
305 * follows directly after a '=' character. Every property is terminated by
306 * a '\0' character. The last property is not terminated.
308 * Example of a message structure:
309 * 0000 ff 8f 00 00 00 00 00 00 monotonic time in nsec
310 * 0008 34 00 record is 52 bytes long
311 * 000a 0b 00 text is 11 bytes long
312 * 000c 1f 00 dictionary is 23 bytes long
313 * 000e 03 00 LOG_KERN (facility) LOG_ERR (level)
314 * 0010 69 74 27 73 20 61 20 6c "it's a l"
316 * 001b 44 45 56 49 43 "DEVIC"
317 * 45 3d 62 38 3a 32 00 44 "E=b8:2\0D"
318 * 52 49 56 45 52 3d 62 75 "RIVER=bu"
320 * 0032 00 00 00 padding to next message header
322 * The 'struct printk_log' buffer header must never be directly exported to
323 * userspace, it is a kernel-private implementation detail that might
324 * need to be changed in the future, when the requirements change.
326 * /dev/kmsg exports the structured data in the following line format:
327 * "<level>,<sequnum>,<timestamp>,<contflag>[,additional_values, ... ];<message text>\n"
329 * Users of the export format should ignore possible additional values
330 * separated by ',', and find the message after the ';' character.
332 * The optional key/value pairs are attached as continuation lines starting
333 * with a space character and terminated by a newline. All possible
334 * non-prinatable characters are escaped in the "\xff" notation.
338 LOG_NOCONS = 1, /* already flushed, do not print to console */
339 LOG_NEWLINE = 2, /* text ended with a newline */
340 LOG_PREFIX = 4, /* text started with a prefix */
341 LOG_CONT = 8, /* text is a fragment of a continuation line */
345 u64 ts_nsec; /* timestamp in nanoseconds */
346 u16 len; /* length of entire record */
347 u16 text_len; /* length of text buffer */
348 u16 dict_len; /* length of dictionary buffer */
349 u8 facility; /* syslog facility */
350 u8 flags:5; /* internal record flags */
351 u8 level:3; /* syslog level */
353 #ifdef CONFIG_HAVE_EFFICIENT_UNALIGNED_ACCESS
354 __packed __aligned(4)
359 * The logbuf_lock protects kmsg buffer, indices, counters. This can be taken
360 * within the scheduler's rq lock. It must be released before calling
361 * console_unlock() or anything else that might wake up a process.
363 DEFINE_RAW_SPINLOCK(logbuf_lock);
366 DECLARE_WAIT_QUEUE_HEAD(log_wait);
367 /* the next printk record to read by syslog(READ) or /proc/kmsg */
368 static u64 syslog_seq;
369 static u32 syslog_idx;
370 static enum log_flags syslog_prev;
371 static size_t syslog_partial;
373 /* index and sequence number of the first record stored in the buffer */
374 static u64 log_first_seq;
375 static u32 log_first_idx;
377 /* index and sequence number of the next record to store in the buffer */
378 static u64 log_next_seq;
379 static u32 log_next_idx;
381 /* the next printk record to write to the console */
382 static u64 console_seq;
383 static u32 console_idx;
384 static enum log_flags console_prev;
386 /* the next printk record to read after the last 'clear' command */
387 static u64 clear_seq;
388 static u32 clear_idx;
390 #define PREFIX_MAX 32
391 #define LOG_LINE_MAX (1024 - PREFIX_MAX)
393 #define LOG_LEVEL(v) ((v) & 0x07)
394 #define LOG_FACILITY(v) ((v) >> 3 & 0xff)
397 #define LOG_ALIGN __alignof__(struct printk_log)
398 #define __LOG_BUF_LEN (1 << CONFIG_LOG_BUF_SHIFT)
399 static char __log_buf[__LOG_BUF_LEN] __aligned(LOG_ALIGN);
400 static char *log_buf = __log_buf;
401 static u32 log_buf_len = __LOG_BUF_LEN;
403 /* Return log buffer address */
404 char *log_buf_addr_get(void)
409 /* Return log buffer size */
410 u32 log_buf_len_get(void)
415 /* human readable text of the record */
416 static char *log_text(const struct printk_log *msg)
418 return (char *)msg + sizeof(struct printk_log);
421 /* optional key/value pair dictionary attached to the record */
422 static char *log_dict(const struct printk_log *msg)
424 return (char *)msg + sizeof(struct printk_log) + msg->text_len;
427 /* get record by index; idx must point to valid msg */
428 static struct printk_log *log_from_idx(u32 idx)
430 struct printk_log *msg = (struct printk_log *)(log_buf + idx);
433 * A length == 0 record is the end of buffer marker. Wrap around and
434 * read the message at the start of the buffer.
437 return (struct printk_log *)log_buf;
441 /* get next record; idx must point to valid msg */
442 static u32 log_next(u32 idx)
444 struct printk_log *msg = (struct printk_log *)(log_buf + idx);
446 /* length == 0 indicates the end of the buffer; wrap */
448 * A length == 0 record is the end of buffer marker. Wrap around and
449 * read the message at the start of the buffer as *this* one, and
450 * return the one after that.
453 msg = (struct printk_log *)log_buf;
456 return idx + msg->len;
460 * Check whether there is enough free space for the given message.
462 * The same values of first_idx and next_idx mean that the buffer
463 * is either empty or full.
465 * If the buffer is empty, we must respect the position of the indexes.
466 * They cannot be reset to the beginning of the buffer.
468 static int logbuf_has_space(u32 msg_size, bool empty)
472 if (log_next_idx > log_first_idx || empty)
473 free = max(log_buf_len - log_next_idx, log_first_idx);
475 free = log_first_idx - log_next_idx;
478 * We need space also for an empty header that signalizes wrapping
481 return free >= msg_size + sizeof(struct printk_log);
484 static int log_make_free_space(u32 msg_size)
486 while (log_first_seq < log_next_seq &&
487 !logbuf_has_space(msg_size, false)) {
488 /* drop old messages until we have enough contiguous space */
489 log_first_idx = log_next(log_first_idx);
493 if (clear_seq < log_first_seq) {
494 clear_seq = log_first_seq;
495 clear_idx = log_first_idx;
498 /* sequence numbers are equal, so the log buffer is empty */
499 if (logbuf_has_space(msg_size, log_first_seq == log_next_seq))
505 /* compute the message size including the padding bytes */
506 static u32 msg_used_size(u16 text_len, u16 dict_len, u32 *pad_len)
510 size = sizeof(struct printk_log) + text_len + dict_len;
511 *pad_len = (-size) & (LOG_ALIGN - 1);
518 * Define how much of the log buffer we could take at maximum. The value
519 * must be greater than two. Note that only half of the buffer is available
520 * when the index points to the middle.
522 #define MAX_LOG_TAKE_PART 4
523 static const char trunc_msg[] = "<truncated>";
525 static u32 truncate_msg(u16 *text_len, u16 *trunc_msg_len,
526 u16 *dict_len, u32 *pad_len)
529 * The message should not take the whole buffer. Otherwise, it might
530 * get removed too soon.
532 u32 max_text_len = log_buf_len / MAX_LOG_TAKE_PART;
533 if (*text_len > max_text_len)
534 *text_len = max_text_len;
535 /* enable the warning message */
536 *trunc_msg_len = strlen(trunc_msg);
537 /* disable the "dict" completely */
539 /* compute the size again, count also the warning message */
540 return msg_used_size(*text_len + *trunc_msg_len, 0, pad_len);
543 /* insert record into the buffer, discard old ones, update heads */
544 static int log_store(int facility, int level,
545 enum log_flags flags, u64 ts_nsec,
546 const char *dict, u16 dict_len,
547 const char *text, u16 text_len)
549 struct printk_log *msg;
551 u16 trunc_msg_len = 0;
553 /* number of '\0' padding bytes to next message */
554 size = msg_used_size(text_len, dict_len, &pad_len);
556 if (log_make_free_space(size)) {
557 /* truncate the message if it is too long for empty buffer */
558 size = truncate_msg(&text_len, &trunc_msg_len,
559 &dict_len, &pad_len);
560 /* survive when the log buffer is too small for trunc_msg */
561 if (log_make_free_space(size))
565 if (log_next_idx + size + sizeof(struct printk_log) > log_buf_len) {
567 * This message + an additional empty header does not fit
568 * at the end of the buffer. Add an empty header with len == 0
569 * to signify a wrap around.
571 memset(log_buf + log_next_idx, 0, sizeof(struct printk_log));
576 msg = (struct printk_log *)(log_buf + log_next_idx);
577 memcpy(log_text(msg), text, text_len);
578 msg->text_len = text_len;
580 memcpy(log_text(msg) + text_len, trunc_msg, trunc_msg_len);
581 msg->text_len += trunc_msg_len;
583 memcpy(log_dict(msg), dict, dict_len);
584 msg->dict_len = dict_len;
585 msg->facility = facility;
586 msg->level = level & 7;
587 msg->flags = flags & 0x1f;
589 msg->ts_nsec = ts_nsec;
591 msg->ts_nsec = local_clock();
592 memset(log_dict(msg) + dict_len, 0, pad_len);
596 log_next_idx += msg->len;
599 return msg->text_len;
602 int dmesg_restrict = IS_ENABLED(CONFIG_SECURITY_DMESG_RESTRICT);
604 static int syslog_action_restricted(int type)
609 * Unless restricted, we allow "read all" and "get buffer size"
612 return type != SYSLOG_ACTION_READ_ALL &&
613 type != SYSLOG_ACTION_SIZE_BUFFER;
616 int check_syslog_permissions(int type, int source)
619 * If this is from /proc/kmsg and we've already opened it, then we've
620 * already done the capabilities checks at open time.
622 if (source == SYSLOG_FROM_PROC && type != SYSLOG_ACTION_OPEN)
625 if (syslog_action_restricted(type)) {
626 if (capable(CAP_SYSLOG))
629 * For historical reasons, accept CAP_SYS_ADMIN too, with
632 if (capable(CAP_SYS_ADMIN)) {
633 pr_warn_once("%s (%d): Attempt to access syslog with "
634 "CAP_SYS_ADMIN but no CAP_SYSLOG "
636 current->comm, task_pid_nr(current));
642 return security_syslog(type);
644 EXPORT_SYMBOL_GPL(check_syslog_permissions);
646 static void append_char(char **pp, char *e, char c)
652 static ssize_t msg_print_ext_header(char *buf, size_t size,
653 struct printk_log *msg, u64 seq,
654 enum log_flags prev_flags)
656 u64 ts_usec = msg->ts_nsec;
659 do_div(ts_usec, 1000);
662 * If we couldn't merge continuation line fragments during the print,
663 * export the stored flags to allow an optional external merge of the
664 * records. Merging the records isn't always neccessarily correct, like
665 * when we hit a race during printing. In most cases though, it produces
666 * better readable output. 'c' in the record flags mark the first
667 * fragment of a line, '+' the following.
669 if (msg->flags & LOG_CONT && !(prev_flags & LOG_CONT))
671 else if ((msg->flags & LOG_CONT) ||
672 ((prev_flags & LOG_CONT) && !(msg->flags & LOG_PREFIX)))
675 return scnprintf(buf, size, "%u,%llu,%llu,%c;",
676 (msg->facility << 3) | msg->level, seq, ts_usec, cont);
679 static ssize_t msg_print_ext_body(char *buf, size_t size,
680 char *dict, size_t dict_len,
681 char *text, size_t text_len)
683 char *p = buf, *e = buf + size;
686 /* escape non-printable characters */
687 for (i = 0; i < text_len; i++) {
688 unsigned char c = text[i];
690 if (c < ' ' || c >= 127 || c == '\\')
691 p += scnprintf(p, e - p, "\\x%02x", c);
693 append_char(&p, e, c);
695 append_char(&p, e, '\n');
700 for (i = 0; i < dict_len; i++) {
701 unsigned char c = dict[i];
704 append_char(&p, e, ' ');
709 append_char(&p, e, '\n');
714 if (c < ' ' || c >= 127 || c == '\\') {
715 p += scnprintf(p, e - p, "\\x%02x", c);
719 append_char(&p, e, c);
721 append_char(&p, e, '\n');
727 /* /dev/kmsg - userspace message inject/listen interface */
728 struct devkmsg_user {
732 struct ratelimit_state rs;
734 char buf[CONSOLE_EXT_LOG_MAX];
737 static ssize_t devkmsg_write(struct kiocb *iocb, struct iov_iter *from)
740 int level = default_message_loglevel;
741 int facility = 1; /* LOG_USER */
742 struct file *file = iocb->ki_filp;
743 struct devkmsg_user *user = file->private_data;
744 size_t len = iov_iter_count(from);
747 if (!user || len > LOG_LINE_MAX)
750 /* Ignore when user logging is disabled. */
751 if (devkmsg_log & DEVKMSG_LOG_MASK_OFF)
754 /* Ratelimit when not explicitly enabled. */
755 if (!(devkmsg_log & DEVKMSG_LOG_MASK_ON)) {
756 if (!___ratelimit(&user->rs, current->comm))
760 buf = kmalloc(len+1, GFP_KERNEL);
765 if (copy_from_iter(buf, len, from) != len) {
771 * Extract and skip the syslog prefix <[0-9]*>. Coming from userspace
772 * the decimal value represents 32bit, the lower 3 bit are the log
773 * level, the rest are the log facility.
775 * If no prefix or no userspace facility is specified, we
776 * enforce LOG_USER, to be able to reliably distinguish
777 * kernel-generated messages from userspace-injected ones.
780 if (line[0] == '<') {
784 u = simple_strtoul(line + 1, &endp, 10);
785 if (endp && endp[0] == '>') {
786 level = LOG_LEVEL(u);
787 if (LOG_FACILITY(u) != 0)
788 facility = LOG_FACILITY(u);
795 printk_emit(facility, level, NULL, 0, "%s", line);
800 static ssize_t devkmsg_read(struct file *file, char __user *buf,
801 size_t count, loff_t *ppos)
803 struct devkmsg_user *user = file->private_data;
804 struct printk_log *msg;
811 ret = mutex_lock_interruptible(&user->lock);
814 raw_spin_lock_irq(&logbuf_lock);
815 while (user->seq == log_next_seq) {
816 if (file->f_flags & O_NONBLOCK) {
818 raw_spin_unlock_irq(&logbuf_lock);
822 raw_spin_unlock_irq(&logbuf_lock);
823 ret = wait_event_interruptible(log_wait,
824 user->seq != log_next_seq);
827 raw_spin_lock_irq(&logbuf_lock);
830 if (user->seq < log_first_seq) {
831 /* our last seen message is gone, return error and reset */
832 user->idx = log_first_idx;
833 user->seq = log_first_seq;
835 raw_spin_unlock_irq(&logbuf_lock);
839 msg = log_from_idx(user->idx);
840 len = msg_print_ext_header(user->buf, sizeof(user->buf),
841 msg, user->seq, user->prev);
842 len += msg_print_ext_body(user->buf + len, sizeof(user->buf) - len,
843 log_dict(msg), msg->dict_len,
844 log_text(msg), msg->text_len);
846 user->prev = msg->flags;
847 user->idx = log_next(user->idx);
849 raw_spin_unlock_irq(&logbuf_lock);
856 if (copy_to_user(buf, user->buf, len)) {
862 mutex_unlock(&user->lock);
866 static loff_t devkmsg_llseek(struct file *file, loff_t offset, int whence)
868 struct devkmsg_user *user = file->private_data;
876 raw_spin_lock_irq(&logbuf_lock);
879 /* the first record */
880 user->idx = log_first_idx;
881 user->seq = log_first_seq;
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.
889 user->idx = clear_idx;
890 user->seq = clear_seq;
893 /* after the last record */
894 user->idx = log_next_idx;
895 user->seq = log_next_seq;
900 raw_spin_unlock_irq(&logbuf_lock);
904 static unsigned int devkmsg_poll(struct file *file, poll_table *wait)
906 struct devkmsg_user *user = file->private_data;
910 return POLLERR|POLLNVAL;
912 poll_wait(file, &log_wait, wait);
914 raw_spin_lock_irq(&logbuf_lock);
915 if (user->seq < log_next_seq) {
916 /* return error when data has vanished underneath us */
917 if (user->seq < log_first_seq)
918 ret = POLLIN|POLLRDNORM|POLLERR|POLLPRI;
920 ret = POLLIN|POLLRDNORM;
922 raw_spin_unlock_irq(&logbuf_lock);
927 static int devkmsg_open(struct inode *inode, struct file *file)
929 struct devkmsg_user *user;
932 if (devkmsg_log & DEVKMSG_LOG_MASK_OFF)
935 /* write-only does not need any file context */
936 if ((file->f_flags & O_ACCMODE) != O_WRONLY) {
937 err = check_syslog_permissions(SYSLOG_ACTION_READ_ALL,
943 user = kmalloc(sizeof(struct devkmsg_user), GFP_KERNEL);
947 ratelimit_default_init(&user->rs);
948 ratelimit_set_flags(&user->rs, RATELIMIT_MSG_ON_RELEASE);
950 mutex_init(&user->lock);
952 raw_spin_lock_irq(&logbuf_lock);
953 user->idx = log_first_idx;
954 user->seq = log_first_seq;
955 raw_spin_unlock_irq(&logbuf_lock);
957 file->private_data = user;
961 static int devkmsg_release(struct inode *inode, struct file *file)
963 struct devkmsg_user *user = file->private_data;
968 ratelimit_state_exit(&user->rs);
970 mutex_destroy(&user->lock);
975 const struct file_operations kmsg_fops = {
976 .open = devkmsg_open,
977 .read = devkmsg_read,
978 .write_iter = devkmsg_write,
979 .llseek = devkmsg_llseek,
980 .poll = devkmsg_poll,
981 .release = devkmsg_release,
984 #ifdef CONFIG_KEXEC_CORE
986 * This appends the listed symbols to /proc/vmcore
988 * /proc/vmcore is used by various utilities, like crash and makedumpfile to
989 * obtain access to symbols that are otherwise very difficult to locate. These
990 * symbols are specifically used so that utilities can access and extract the
991 * dmesg log from a vmcore file after a crash.
993 void log_buf_kexec_setup(void)
995 VMCOREINFO_SYMBOL(log_buf);
996 VMCOREINFO_SYMBOL(log_buf_len);
997 VMCOREINFO_SYMBOL(log_first_idx);
998 VMCOREINFO_SYMBOL(clear_idx);
999 VMCOREINFO_SYMBOL(log_next_idx);
1001 * Export struct printk_log size and field offsets. User space tools can
1002 * parse it and detect any changes to structure down the line.
1004 VMCOREINFO_STRUCT_SIZE(printk_log);
1005 VMCOREINFO_OFFSET(printk_log, ts_nsec);
1006 VMCOREINFO_OFFSET(printk_log, len);
1007 VMCOREINFO_OFFSET(printk_log, text_len);
1008 VMCOREINFO_OFFSET(printk_log, dict_len);
1012 /* requested log_buf_len from kernel cmdline */
1013 static unsigned long __initdata new_log_buf_len;
1015 /* we practice scaling the ring buffer by powers of 2 */
1016 static void __init log_buf_len_update(unsigned size)
1019 size = roundup_pow_of_two(size);
1020 if (size > log_buf_len)
1021 new_log_buf_len = size;
1024 /* save requested log_buf_len since it's too early to process it */
1025 static int __init log_buf_len_setup(char *str)
1027 unsigned size = memparse(str, &str);
1029 log_buf_len_update(size);
1033 early_param("log_buf_len", log_buf_len_setup);
1036 #define __LOG_CPU_MAX_BUF_LEN (1 << CONFIG_LOG_CPU_MAX_BUF_SHIFT)
1038 static void __init log_buf_add_cpu(void)
1040 unsigned int cpu_extra;
1043 * archs should set up cpu_possible_bits properly with
1044 * set_cpu_possible() after setup_arch() but just in
1045 * case lets ensure this is valid.
1047 if (num_possible_cpus() == 1)
1050 cpu_extra = (num_possible_cpus() - 1) * __LOG_CPU_MAX_BUF_LEN;
1052 /* by default this will only continue through for large > 64 CPUs */
1053 if (cpu_extra <= __LOG_BUF_LEN / 2)
1056 pr_info("log_buf_len individual max cpu contribution: %d bytes\n",
1057 __LOG_CPU_MAX_BUF_LEN);
1058 pr_info("log_buf_len total cpu_extra contributions: %d bytes\n",
1060 pr_info("log_buf_len min size: %d bytes\n", __LOG_BUF_LEN);
1062 log_buf_len_update(cpu_extra + __LOG_BUF_LEN);
1064 #else /* !CONFIG_SMP */
1065 static inline void log_buf_add_cpu(void) {}
1066 #endif /* CONFIG_SMP */
1068 void __init setup_log_buf(int early)
1070 unsigned long flags;
1074 if (log_buf != __log_buf)
1077 if (!early && !new_log_buf_len)
1080 if (!new_log_buf_len)
1085 memblock_virt_alloc(new_log_buf_len, LOG_ALIGN);
1087 new_log_buf = memblock_virt_alloc_nopanic(new_log_buf_len,
1091 if (unlikely(!new_log_buf)) {
1092 pr_err("log_buf_len: %ld bytes not available\n",
1097 raw_spin_lock_irqsave(&logbuf_lock, flags);
1098 log_buf_len = new_log_buf_len;
1099 log_buf = new_log_buf;
1100 new_log_buf_len = 0;
1101 free = __LOG_BUF_LEN - log_next_idx;
1102 memcpy(log_buf, __log_buf, __LOG_BUF_LEN);
1103 raw_spin_unlock_irqrestore(&logbuf_lock, flags);
1105 pr_info("log_buf_len: %d bytes\n", log_buf_len);
1106 pr_info("early log buf free: %d(%d%%)\n",
1107 free, (free * 100) / __LOG_BUF_LEN);
1110 static bool __read_mostly ignore_loglevel;
1112 static int __init ignore_loglevel_setup(char *str)
1114 ignore_loglevel = true;
1115 pr_info("debug: ignoring loglevel setting.\n");
1120 early_param("ignore_loglevel", ignore_loglevel_setup);
1121 module_param(ignore_loglevel, bool, S_IRUGO | S_IWUSR);
1122 MODULE_PARM_DESC(ignore_loglevel,
1123 "ignore loglevel setting (prints all kernel messages to the console)");
1125 static bool suppress_message_printing(int level)
1127 return (level >= console_loglevel && !ignore_loglevel);
1130 #ifdef CONFIG_BOOT_PRINTK_DELAY
1132 static int boot_delay; /* msecs delay after each printk during bootup */
1133 static unsigned long long loops_per_msec; /* based on boot_delay */
1135 static int __init boot_delay_setup(char *str)
1139 lpj = preset_lpj ? preset_lpj : 1000000; /* some guess */
1140 loops_per_msec = (unsigned long long)lpj / 1000 * HZ;
1142 get_option(&str, &boot_delay);
1143 if (boot_delay > 10 * 1000)
1146 pr_debug("boot_delay: %u, preset_lpj: %ld, lpj: %lu, "
1147 "HZ: %d, loops_per_msec: %llu\n",
1148 boot_delay, preset_lpj, lpj, HZ, loops_per_msec);
1151 early_param("boot_delay", boot_delay_setup);
1153 static void boot_delay_msec(int level)
1155 unsigned long long k;
1156 unsigned long timeout;
1158 if ((boot_delay == 0 || system_state != SYSTEM_BOOTING)
1159 || suppress_message_printing(level)) {
1163 k = (unsigned long long)loops_per_msec * boot_delay;
1165 timeout = jiffies + msecs_to_jiffies(boot_delay);
1170 * use (volatile) jiffies to prevent
1171 * compiler reduction; loop termination via jiffies
1172 * is secondary and may or may not happen.
1174 if (time_after(jiffies, timeout))
1176 touch_nmi_watchdog();
1180 static inline void boot_delay_msec(int level)
1185 static bool printk_time = IS_ENABLED(CONFIG_PRINTK_TIME);
1186 module_param_named(time, printk_time, bool, S_IRUGO | S_IWUSR);
1188 static size_t print_time(u64 ts, char *buf)
1190 unsigned long rem_nsec;
1195 rem_nsec = do_div(ts, 1000000000);
1198 return snprintf(NULL, 0, "[%5lu.000000] ", (unsigned long)ts);
1200 return sprintf(buf, "[%5lu.%06lu] ",
1201 (unsigned long)ts, rem_nsec / 1000);
1204 static size_t print_prefix(const struct printk_log *msg, bool syslog, char *buf)
1207 unsigned int prefix = (msg->facility << 3) | msg->level;
1211 len += sprintf(buf, "<%u>", prefix);
1216 else if (prefix > 99)
1218 else if (prefix > 9)
1223 len += print_time(msg->ts_nsec, buf ? buf + len : NULL);
1227 static size_t msg_print_text(const struct printk_log *msg, enum log_flags prev,
1228 bool syslog, char *buf, size_t size)
1230 const char *text = log_text(msg);
1231 size_t text_size = msg->text_len;
1233 bool newline = true;
1236 if ((prev & LOG_CONT) && !(msg->flags & LOG_PREFIX))
1239 if (msg->flags & LOG_CONT) {
1240 if ((prev & LOG_CONT) && !(prev & LOG_NEWLINE))
1243 if (!(msg->flags & LOG_NEWLINE))
1248 const char *next = memchr(text, '\n', text_size);
1252 text_len = next - text;
1254 text_size -= next - text;
1256 text_len = text_size;
1260 if (print_prefix(msg, syslog, NULL) +
1261 text_len + 1 >= size - len)
1265 len += print_prefix(msg, syslog, buf + len);
1266 memcpy(buf + len, text, text_len);
1268 if (next || newline)
1271 /* SYSLOG_ACTION_* buffer size only calculation */
1273 len += print_prefix(msg, syslog, NULL);
1275 if (next || newline)
1286 static int syslog_print(char __user *buf, int size)
1289 struct printk_log *msg;
1292 text = kmalloc(LOG_LINE_MAX + PREFIX_MAX, GFP_KERNEL);
1300 raw_spin_lock_irq(&logbuf_lock);
1301 if (syslog_seq < log_first_seq) {
1302 /* messages are gone, move to first one */
1303 syslog_seq = log_first_seq;
1304 syslog_idx = log_first_idx;
1308 if (syslog_seq == log_next_seq) {
1309 raw_spin_unlock_irq(&logbuf_lock);
1313 skip = syslog_partial;
1314 msg = log_from_idx(syslog_idx);
1315 n = msg_print_text(msg, syslog_prev, true, text,
1316 LOG_LINE_MAX + PREFIX_MAX);
1317 if (n - syslog_partial <= size) {
1318 /* message fits into buffer, move forward */
1319 syslog_idx = log_next(syslog_idx);
1321 syslog_prev = msg->flags;
1322 n -= syslog_partial;
1325 /* partial read(), remember position */
1327 syslog_partial += n;
1330 raw_spin_unlock_irq(&logbuf_lock);
1335 if (copy_to_user(buf, text + skip, n)) {
1350 static int syslog_print_all(char __user *buf, int size, bool clear)
1355 text = kmalloc(LOG_LINE_MAX + PREFIX_MAX, GFP_KERNEL);
1359 raw_spin_lock_irq(&logbuf_lock);
1364 enum log_flags prev;
1367 * Find first record that fits, including all following records,
1368 * into the user-provided buffer for this dump.
1373 while (seq < log_next_seq) {
1374 struct printk_log *msg = log_from_idx(idx);
1376 len += msg_print_text(msg, prev, true, NULL, 0);
1378 idx = log_next(idx);
1382 /* move first record forward until length fits into the buffer */
1386 while (len > size && seq < log_next_seq) {
1387 struct printk_log *msg = log_from_idx(idx);
1389 len -= msg_print_text(msg, prev, true, NULL, 0);
1391 idx = log_next(idx);
1395 /* last message fitting into this dump */
1396 next_seq = log_next_seq;
1399 while (len >= 0 && seq < next_seq) {
1400 struct printk_log *msg = log_from_idx(idx);
1403 textlen = msg_print_text(msg, prev, true, text,
1404 LOG_LINE_MAX + PREFIX_MAX);
1409 idx = log_next(idx);
1413 raw_spin_unlock_irq(&logbuf_lock);
1414 if (copy_to_user(buf + len, text, textlen))
1418 raw_spin_lock_irq(&logbuf_lock);
1420 if (seq < log_first_seq) {
1421 /* messages are gone, move to next one */
1422 seq = log_first_seq;
1423 idx = log_first_idx;
1430 clear_seq = log_next_seq;
1431 clear_idx = log_next_idx;
1433 raw_spin_unlock_irq(&logbuf_lock);
1439 int do_syslog(int type, char __user *buf, int len, int source)
1442 static int saved_console_loglevel = LOGLEVEL_DEFAULT;
1445 error = check_syslog_permissions(type, source);
1450 case SYSLOG_ACTION_CLOSE: /* Close log */
1452 case SYSLOG_ACTION_OPEN: /* Open log */
1454 case SYSLOG_ACTION_READ: /* Read from log */
1456 if (!buf || len < 0)
1461 if (!access_ok(VERIFY_WRITE, buf, len)) {
1465 error = wait_event_interruptible(log_wait,
1466 syslog_seq != log_next_seq);
1469 error = syslog_print(buf, len);
1471 /* Read/clear last kernel messages */
1472 case SYSLOG_ACTION_READ_CLEAR:
1475 /* Read last kernel messages */
1476 case SYSLOG_ACTION_READ_ALL:
1478 if (!buf || len < 0)
1483 if (!access_ok(VERIFY_WRITE, buf, len)) {
1487 error = syslog_print_all(buf, len, clear);
1489 /* Clear ring buffer */
1490 case SYSLOG_ACTION_CLEAR:
1491 syslog_print_all(NULL, 0, true);
1493 /* Disable logging to console */
1494 case SYSLOG_ACTION_CONSOLE_OFF:
1495 if (saved_console_loglevel == LOGLEVEL_DEFAULT)
1496 saved_console_loglevel = console_loglevel;
1497 console_loglevel = minimum_console_loglevel;
1499 /* Enable logging to console */
1500 case SYSLOG_ACTION_CONSOLE_ON:
1501 if (saved_console_loglevel != LOGLEVEL_DEFAULT) {
1502 console_loglevel = saved_console_loglevel;
1503 saved_console_loglevel = LOGLEVEL_DEFAULT;
1506 /* Set level of messages printed to console */
1507 case SYSLOG_ACTION_CONSOLE_LEVEL:
1509 if (len < 1 || len > 8)
1511 if (len < minimum_console_loglevel)
1512 len = minimum_console_loglevel;
1513 console_loglevel = len;
1514 /* Implicitly re-enable logging to console */
1515 saved_console_loglevel = LOGLEVEL_DEFAULT;
1518 /* Number of chars in the log buffer */
1519 case SYSLOG_ACTION_SIZE_UNREAD:
1520 raw_spin_lock_irq(&logbuf_lock);
1521 if (syslog_seq < log_first_seq) {
1522 /* messages are gone, move to first one */
1523 syslog_seq = log_first_seq;
1524 syslog_idx = log_first_idx;
1528 if (source == SYSLOG_FROM_PROC) {
1530 * Short-cut for poll(/"proc/kmsg") which simply checks
1531 * for pending data, not the size; return the count of
1532 * records, not the length.
1534 error = log_next_seq - syslog_seq;
1536 u64 seq = syslog_seq;
1537 u32 idx = syslog_idx;
1538 enum log_flags prev = syslog_prev;
1541 while (seq < log_next_seq) {
1542 struct printk_log *msg = log_from_idx(idx);
1544 error += msg_print_text(msg, prev, true, NULL, 0);
1545 idx = log_next(idx);
1549 error -= syslog_partial;
1551 raw_spin_unlock_irq(&logbuf_lock);
1553 /* Size of the log buffer */
1554 case SYSLOG_ACTION_SIZE_BUFFER:
1555 error = log_buf_len;
1565 SYSCALL_DEFINE3(syslog, int, type, char __user *, buf, int, len)
1567 return do_syslog(type, buf, len, SYSLOG_FROM_READER);
1571 * Call the console drivers, asking them to write out
1572 * log_buf[start] to log_buf[end - 1].
1573 * The console_lock must be held.
1575 static void call_console_drivers(int level,
1576 const char *ext_text, size_t ext_len,
1577 const char *text, size_t len)
1579 struct console *con;
1581 trace_console(text, len);
1583 if (!console_drivers)
1586 for_each_console(con) {
1587 if (exclusive_console && con != exclusive_console)
1589 if (!(con->flags & CON_ENABLED))
1593 if (!cpu_online(smp_processor_id()) &&
1594 !(con->flags & CON_ANYTIME))
1596 if (con->flags & CON_EXTENDED)
1597 con->write(con, ext_text, ext_len);
1599 con->write(con, text, len);
1604 * Zap console related locks when oopsing.
1605 * To leave time for slow consoles to print a full oops,
1606 * only zap at most once every 30 seconds.
1608 static void zap_locks(void)
1610 static unsigned long oops_timestamp;
1612 if (time_after_eq(jiffies, oops_timestamp) &&
1613 !time_after(jiffies, oops_timestamp + 30 * HZ))
1616 oops_timestamp = jiffies;
1619 /* If a crash is occurring, make sure we can't deadlock */
1620 raw_spin_lock_init(&logbuf_lock);
1621 /* And make sure that we print immediately */
1622 sema_init(&console_sem, 1);
1625 int printk_delay_msec __read_mostly;
1627 static inline void printk_delay(void)
1629 if (unlikely(printk_delay_msec)) {
1630 int m = printk_delay_msec;
1634 touch_nmi_watchdog();
1640 * Continuation lines are buffered, and not committed to the record buffer
1641 * until the line is complete, or a race forces it. The line fragments
1642 * though, are printed immediately to the consoles to ensure everything has
1643 * reached the console in case of a kernel crash.
1645 static struct cont {
1646 char buf[LOG_LINE_MAX];
1647 size_t len; /* length == 0 means unused buffer */
1648 size_t cons; /* bytes written to console */
1649 struct task_struct *owner; /* task of first print*/
1650 u64 ts_nsec; /* time of first print */
1651 u8 level; /* log level of first message */
1652 u8 facility; /* log facility of first message */
1653 enum log_flags flags; /* prefix, newline flags */
1654 bool flushed:1; /* buffer sealed and committed */
1657 static void cont_flush(enum log_flags flags)
1666 * If a fragment of this line was directly flushed to the
1667 * console; wait for the console to pick up the rest of the
1668 * line. LOG_NOCONS suppresses a duplicated output.
1670 log_store(cont.facility, cont.level, flags | LOG_NOCONS,
1671 cont.ts_nsec, NULL, 0, cont.buf, cont.len);
1673 cont.flushed = true;
1676 * If no fragment of this line ever reached the console,
1677 * just submit it to the store and free the buffer.
1679 log_store(cont.facility, cont.level, flags, 0,
1680 NULL, 0, cont.buf, cont.len);
1685 static bool cont_add(int facility, int level, const char *text, size_t len)
1687 if (cont.len && cont.flushed)
1691 * If ext consoles are present, flush and skip in-kernel
1692 * continuation. See nr_ext_console_drivers definition. Also, if
1693 * the line gets too long, split it up in separate records.
1695 if (nr_ext_console_drivers || cont.len + len > sizeof(cont.buf)) {
1696 cont_flush(LOG_CONT);
1701 cont.facility = facility;
1703 cont.owner = current;
1704 cont.ts_nsec = local_clock();
1707 cont.flushed = false;
1710 memcpy(cont.buf + cont.len, text, len);
1713 if (cont.len > (sizeof(cont.buf) * 80) / 100)
1714 cont_flush(LOG_CONT);
1719 static size_t cont_print_text(char *text, size_t size)
1724 if (cont.cons == 0 && (console_prev & LOG_NEWLINE)) {
1725 textlen += print_time(cont.ts_nsec, text);
1729 len = cont.len - cont.cons;
1733 memcpy(text + textlen, cont.buf + cont.cons, len);
1735 cont.cons = cont.len;
1739 if (cont.flags & LOG_NEWLINE)
1740 text[textlen++] = '\n';
1741 /* got everything, release buffer */
1747 asmlinkage int vprintk_emit(int facility, int level,
1748 const char *dict, size_t dictlen,
1749 const char *fmt, va_list args)
1751 static bool recursion_bug;
1752 static char textbuf[LOG_LINE_MAX];
1753 char *text = textbuf;
1754 size_t text_len = 0;
1755 enum log_flags lflags = 0;
1756 unsigned long flags;
1758 int printed_len = 0;
1759 int nmi_message_lost;
1760 bool in_sched = false;
1761 /* cpu currently holding logbuf_lock in this function */
1762 static unsigned int logbuf_cpu = UINT_MAX;
1764 if (level == LOGLEVEL_SCHED) {
1765 level = LOGLEVEL_DEFAULT;
1769 boot_delay_msec(level);
1772 local_irq_save(flags);
1773 this_cpu = smp_processor_id();
1776 * Ouch, printk recursed into itself!
1778 if (unlikely(logbuf_cpu == this_cpu)) {
1780 * If a crash is occurring during printk() on this CPU,
1781 * then try to get the crash message out but make sure
1782 * we can't deadlock. Otherwise just return to avoid the
1783 * recursion and return - but flag the recursion so that
1784 * it can be printed at the next appropriate moment:
1786 if (!oops_in_progress && !lockdep_recursing(current)) {
1787 recursion_bug = true;
1788 local_irq_restore(flags);
1795 /* This stops the holder of console_sem just where we want him */
1796 raw_spin_lock(&logbuf_lock);
1797 logbuf_cpu = this_cpu;
1799 if (unlikely(recursion_bug)) {
1800 static const char recursion_msg[] =
1801 "BUG: recent printk recursion!";
1803 recursion_bug = false;
1804 /* emit KERN_CRIT message */
1805 printed_len += log_store(0, 2, LOG_PREFIX|LOG_NEWLINE, 0,
1806 NULL, 0, recursion_msg,
1807 strlen(recursion_msg));
1810 nmi_message_lost = get_nmi_message_lost();
1811 if (unlikely(nmi_message_lost)) {
1812 text_len = scnprintf(textbuf, sizeof(textbuf),
1813 "BAD LUCK: lost %d message(s) from NMI context!",
1815 printed_len += log_store(0, 2, LOG_PREFIX|LOG_NEWLINE, 0,
1816 NULL, 0, textbuf, text_len);
1820 * The printf needs to come first; we need the syslog
1821 * prefix which might be passed-in as a parameter.
1823 text_len = vscnprintf(text, sizeof(textbuf), fmt, args);
1825 /* mark and strip a trailing newline */
1826 if (text_len && text[text_len-1] == '\n') {
1828 lflags |= LOG_NEWLINE;
1831 /* strip kernel syslog prefix and extract log level or control flags */
1832 if (facility == 0) {
1833 int kern_level = printk_get_level(text);
1836 const char *end_of_header = printk_skip_level(text);
1837 switch (kern_level) {
1839 if (level == LOGLEVEL_DEFAULT)
1840 level = kern_level - '0';
1842 case 'd': /* KERN_DEFAULT */
1843 lflags |= LOG_PREFIX;
1846 * No need to check length here because vscnprintf
1847 * put '\0' at the end of the string. Only valid and
1848 * newly printed level is detected.
1850 text_len -= end_of_header - text;
1851 text = (char *)end_of_header;
1855 if (level == LOGLEVEL_DEFAULT)
1856 level = default_message_loglevel;
1859 lflags |= LOG_PREFIX|LOG_NEWLINE;
1861 if (!(lflags & LOG_NEWLINE)) {
1863 * Flush the conflicting buffer. An earlier newline was missing,
1864 * or another task also prints continuation lines.
1866 if (cont.len && (lflags & LOG_PREFIX || cont.owner != current))
1867 cont_flush(LOG_NEWLINE);
1869 /* buffer line if possible, otherwise store it right away */
1870 if (cont_add(facility, level, text, text_len))
1871 printed_len += text_len;
1873 printed_len += log_store(facility, level,
1874 lflags | LOG_CONT, 0,
1875 dict, dictlen, text, text_len);
1877 bool stored = false;
1880 * If an earlier newline was missing and it was the same task,
1881 * either merge it with the current buffer and flush, or if
1882 * there was a race with interrupts (prefix == true) then just
1883 * flush it out and store this line separately.
1884 * If the preceding printk was from a different task and missed
1885 * a newline, flush and append the newline.
1888 if (cont.owner == current && !(lflags & LOG_PREFIX))
1889 stored = cont_add(facility, level, text,
1891 cont_flush(LOG_NEWLINE);
1895 printed_len += text_len;
1897 printed_len += log_store(facility, level, lflags, 0,
1898 dict, dictlen, text, text_len);
1901 logbuf_cpu = UINT_MAX;
1902 raw_spin_unlock(&logbuf_lock);
1904 local_irq_restore(flags);
1906 /* If called from the scheduler, we can not call up(). */
1910 * Try to acquire and then immediately release the console
1911 * semaphore. The release will print out buffers and wake up
1912 * /dev/kmsg and syslog() users.
1914 if (console_trylock())
1921 EXPORT_SYMBOL(vprintk_emit);
1923 asmlinkage int vprintk(const char *fmt, va_list args)
1925 return vprintk_emit(0, LOGLEVEL_DEFAULT, NULL, 0, fmt, args);
1927 EXPORT_SYMBOL(vprintk);
1929 asmlinkage int printk_emit(int facility, int level,
1930 const char *dict, size_t dictlen,
1931 const char *fmt, ...)
1936 va_start(args, fmt);
1937 r = vprintk_emit(facility, level, dict, dictlen, fmt, args);
1942 EXPORT_SYMBOL(printk_emit);
1944 int vprintk_default(const char *fmt, va_list args)
1948 #ifdef CONFIG_KGDB_KDB
1949 if (unlikely(kdb_trap_printk)) {
1950 r = vkdb_printf(KDB_MSGSRC_PRINTK, fmt, args);
1954 r = vprintk_emit(0, LOGLEVEL_DEFAULT, NULL, 0, fmt, args);
1958 EXPORT_SYMBOL_GPL(vprintk_default);
1961 * printk - print a kernel message
1962 * @fmt: format string
1964 * This is printk(). It can be called from any context. We want it to work.
1966 * We try to grab the console_lock. If we succeed, it's easy - we log the
1967 * output and call the console drivers. If we fail to get the semaphore, we
1968 * place the output into the log buffer and return. The current holder of
1969 * the console_sem will notice the new output in console_unlock(); and will
1970 * send it to the consoles before releasing the lock.
1972 * One effect of this deferred printing is that code which calls printk() and
1973 * then changes console_loglevel may break. This is because console_loglevel
1974 * is inspected when the actual printing occurs.
1979 * See the vsnprintf() documentation for format string extensions over C99.
1981 asmlinkage __visible int printk(const char *fmt, ...)
1986 va_start(args, fmt);
1987 r = vprintk_func(fmt, args);
1992 EXPORT_SYMBOL(printk);
1994 #else /* CONFIG_PRINTK */
1996 #define LOG_LINE_MAX 0
1997 #define PREFIX_MAX 0
1999 static u64 syslog_seq;
2000 static u32 syslog_idx;
2001 static u64 console_seq;
2002 static u32 console_idx;
2003 static enum log_flags syslog_prev;
2004 static u64 log_first_seq;
2005 static u32 log_first_idx;
2006 static u64 log_next_seq;
2007 static enum log_flags console_prev;
2008 static struct cont {
2014 static char *log_text(const struct printk_log *msg) { return NULL; }
2015 static char *log_dict(const struct printk_log *msg) { return NULL; }
2016 static struct printk_log *log_from_idx(u32 idx) { return NULL; }
2017 static u32 log_next(u32 idx) { return 0; }
2018 static ssize_t msg_print_ext_header(char *buf, size_t size,
2019 struct printk_log *msg, u64 seq,
2020 enum log_flags prev_flags) { return 0; }
2021 static ssize_t msg_print_ext_body(char *buf, size_t size,
2022 char *dict, size_t dict_len,
2023 char *text, size_t text_len) { return 0; }
2024 static void call_console_drivers(int level,
2025 const char *ext_text, size_t ext_len,
2026 const char *text, size_t len) {}
2027 static size_t msg_print_text(const struct printk_log *msg, enum log_flags prev,
2028 bool syslog, char *buf, size_t size) { return 0; }
2029 static size_t cont_print_text(char *text, size_t size) { return 0; }
2030 static bool suppress_message_printing(int level) { return false; }
2032 /* Still needs to be defined for users */
2033 DEFINE_PER_CPU(printk_func_t, printk_func);
2035 #endif /* CONFIG_PRINTK */
2037 #ifdef CONFIG_EARLY_PRINTK
2038 struct console *early_console;
2040 asmlinkage __visible void early_printk(const char *fmt, ...)
2050 n = vscnprintf(buf, sizeof(buf), fmt, ap);
2053 early_console->write(early_console, buf, n);
2057 static int __add_preferred_console(char *name, int idx, char *options,
2060 struct console_cmdline *c;
2064 * See if this tty is not yet registered, and
2065 * if we have a slot free.
2067 for (i = 0, c = console_cmdline;
2068 i < MAX_CMDLINECONSOLES && c->name[0];
2070 if (strcmp(c->name, name) == 0 && c->index == idx) {
2072 selected_console = i;
2076 if (i == MAX_CMDLINECONSOLES)
2079 selected_console = i;
2080 strlcpy(c->name, name, sizeof(c->name));
2081 c->options = options;
2082 braille_set_options(c, brl_options);
2088 * Set up a console. Called via do_early_param() in init/main.c
2089 * for each "console=" parameter in the boot command line.
2091 static int __init console_setup(char *str)
2093 char buf[sizeof(console_cmdline[0].name) + 4]; /* 4 for "ttyS" */
2094 char *s, *options, *brl_options = NULL;
2097 if (_braille_console_setup(&str, &brl_options))
2101 * Decode str into name, index, options.
2103 if (str[0] >= '0' && str[0] <= '9') {
2104 strcpy(buf, "ttyS");
2105 strncpy(buf + 4, str, sizeof(buf) - 5);
2107 strncpy(buf, str, sizeof(buf) - 1);
2109 buf[sizeof(buf) - 1] = 0;
2110 options = strchr(str, ',');
2114 if (!strcmp(str, "ttya"))
2115 strcpy(buf, "ttyS0");
2116 if (!strcmp(str, "ttyb"))
2117 strcpy(buf, "ttyS1");
2119 for (s = buf; *s; s++)
2120 if (isdigit(*s) || *s == ',')
2122 idx = simple_strtoul(s, NULL, 10);
2125 __add_preferred_console(buf, idx, options, brl_options);
2126 console_set_on_cmdline = 1;
2129 __setup("console=", console_setup);
2132 * add_preferred_console - add a device to the list of preferred consoles.
2133 * @name: device name
2134 * @idx: device index
2135 * @options: options for this console
2137 * The last preferred console added will be used for kernel messages
2138 * and stdin/out/err for init. Normally this is used by console_setup
2139 * above to handle user-supplied console arguments; however it can also
2140 * be used by arch-specific code either to override the user or more
2141 * commonly to provide a default console (ie from PROM variables) when
2142 * the user has not supplied one.
2144 int add_preferred_console(char *name, int idx, char *options)
2146 return __add_preferred_console(name, idx, options, NULL);
2149 bool console_suspend_enabled = true;
2150 EXPORT_SYMBOL(console_suspend_enabled);
2152 static int __init console_suspend_disable(char *str)
2154 console_suspend_enabled = false;
2157 __setup("no_console_suspend", console_suspend_disable);
2158 module_param_named(console_suspend, console_suspend_enabled,
2159 bool, S_IRUGO | S_IWUSR);
2160 MODULE_PARM_DESC(console_suspend, "suspend console during suspend"
2161 " and hibernate operations");
2164 * suspend_console - suspend the console subsystem
2166 * This disables printk() while we go into suspend states
2168 void suspend_console(void)
2170 if (!console_suspend_enabled)
2172 printk("Suspending console(s) (use no_console_suspend to debug)\n");
2174 console_suspended = 1;
2178 void resume_console(void)
2180 if (!console_suspend_enabled)
2183 console_suspended = 0;
2188 * console_cpu_notify - print deferred console messages after CPU hotplug
2189 * @self: notifier struct
2190 * @action: CPU hotplug event
2193 * If printk() is called from a CPU that is not online yet, the messages
2194 * will be spooled but will not show up on the console. This function is
2195 * called when a new CPU comes online (or fails to come up), and ensures
2196 * that any such output gets printed.
2198 static int console_cpu_notify(struct notifier_block *self,
2199 unsigned long action, void *hcpu)
2204 case CPU_DOWN_FAILED:
2205 case CPU_UP_CANCELED:
2213 * console_lock - lock the console system for exclusive use.
2215 * Acquires a lock which guarantees that the caller has
2216 * exclusive access to the console system and the console_drivers list.
2218 * Can sleep, returns nothing.
2220 void console_lock(void)
2225 if (console_suspended)
2228 console_may_schedule = 1;
2230 EXPORT_SYMBOL(console_lock);
2233 * console_trylock - try to lock the console system for exclusive use.
2235 * Try to acquire a lock which guarantees that the caller has exclusive
2236 * access to the console system and the console_drivers list.
2238 * returns 1 on success, and 0 on failure to acquire the lock.
2240 int console_trylock(void)
2242 if (down_trylock_console_sem())
2244 if (console_suspended) {
2250 * When PREEMPT_COUNT disabled we can't reliably detect if it's
2251 * safe to schedule (e.g. calling printk while holding a spin_lock),
2252 * because preempt_disable()/preempt_enable() are just barriers there
2253 * and preempt_count() is always 0.
2255 * RCU read sections have a separate preemption counter when
2256 * PREEMPT_RCU enabled thus we must take extra care and check
2257 * rcu_preempt_depth(), otherwise RCU read sections modify
2260 console_may_schedule = !oops_in_progress &&
2262 !rcu_preempt_depth();
2265 EXPORT_SYMBOL(console_trylock);
2267 int is_console_locked(void)
2269 return console_locked;
2273 * Check if we have any console that is capable of printing while cpu is
2274 * booting or shutting down. Requires console_sem.
2276 static int have_callable_console(void)
2278 struct console *con;
2280 for_each_console(con)
2281 if ((con->flags & CON_ENABLED) &&
2282 (con->flags & CON_ANYTIME))
2289 * Can we actually use the console at this time on this cpu?
2291 * Console drivers may assume that per-cpu resources have been allocated. So
2292 * unless they're explicitly marked as being able to cope (CON_ANYTIME) don't
2293 * call them until this CPU is officially up.
2295 static inline int can_use_console(void)
2297 return cpu_online(raw_smp_processor_id()) || have_callable_console();
2300 static void console_cont_flush(char *text, size_t size)
2302 unsigned long flags;
2305 raw_spin_lock_irqsave(&logbuf_lock, flags);
2310 if (suppress_message_printing(cont.level)) {
2311 cont.cons = cont.len;
2318 * We still queue earlier records, likely because the console was
2319 * busy. The earlier ones need to be printed before this one, we
2320 * did not flush any fragment so far, so just let it queue up.
2322 if (console_seq < log_next_seq && !cont.cons)
2325 len = cont_print_text(text, size);
2326 raw_spin_unlock(&logbuf_lock);
2327 stop_critical_timings();
2328 call_console_drivers(cont.level, NULL, 0, text, len);
2329 start_critical_timings();
2330 local_irq_restore(flags);
2333 raw_spin_unlock_irqrestore(&logbuf_lock, flags);
2337 * console_unlock - unlock the console system
2339 * Releases the console_lock which the caller holds on the console system
2340 * and the console driver list.
2342 * While the console_lock was held, console output may have been buffered
2343 * by printk(). If this is the case, console_unlock(); emits
2344 * the output prior to releasing the lock.
2346 * If there is output waiting, we wake /dev/kmsg and syslog() users.
2348 * console_unlock(); may be called from any context.
2350 void console_unlock(void)
2352 static char ext_text[CONSOLE_EXT_LOG_MAX];
2353 static char text[LOG_LINE_MAX + PREFIX_MAX];
2354 static u64 seen_seq;
2355 unsigned long flags;
2356 bool wake_klogd = false;
2357 bool do_cond_resched, retry;
2359 if (console_suspended) {
2365 * Console drivers are called under logbuf_lock, so
2366 * @console_may_schedule should be cleared before; however, we may
2367 * end up dumping a lot of lines, for example, if called from
2368 * console registration path, and should invoke cond_resched()
2369 * between lines if allowable. Not doing so can cause a very long
2370 * scheduling stall on a slow console leading to RCU stall and
2371 * softlockup warnings which exacerbate the issue with more
2372 * messages practically incapacitating the system.
2374 do_cond_resched = console_may_schedule;
2375 console_may_schedule = 0;
2379 * We released the console_sem lock, so we need to recheck if
2380 * cpu is online and (if not) is there at least one CON_ANYTIME
2383 if (!can_use_console()) {
2389 /* flush buffered message fragment immediately to console */
2390 console_cont_flush(text, sizeof(text));
2393 struct printk_log *msg;
2398 raw_spin_lock_irqsave(&logbuf_lock, flags);
2399 if (seen_seq != log_next_seq) {
2401 seen_seq = log_next_seq;
2404 if (console_seq < log_first_seq) {
2405 len = sprintf(text, "** %u printk messages dropped ** ",
2406 (unsigned)(log_first_seq - console_seq));
2408 /* messages are gone, move to first one */
2409 console_seq = log_first_seq;
2410 console_idx = log_first_idx;
2416 if (console_seq == log_next_seq)
2419 msg = log_from_idx(console_idx);
2421 if ((msg->flags & LOG_NOCONS) ||
2422 suppress_message_printing(level)) {
2424 * Skip record we have buffered and already printed
2425 * directly to the console when we received it, and
2426 * record that has level above the console loglevel.
2428 console_idx = log_next(console_idx);
2431 * We will get here again when we register a new
2432 * CON_PRINTBUFFER console. Clear the flag so we
2433 * will properly dump everything later.
2435 msg->flags &= ~LOG_NOCONS;
2436 console_prev = msg->flags;
2440 len += msg_print_text(msg, console_prev, false,
2441 text + len, sizeof(text) - len);
2442 if (nr_ext_console_drivers) {
2443 ext_len = msg_print_ext_header(ext_text,
2445 msg, console_seq, console_prev);
2446 ext_len += msg_print_ext_body(ext_text + ext_len,
2447 sizeof(ext_text) - ext_len,
2448 log_dict(msg), msg->dict_len,
2449 log_text(msg), msg->text_len);
2451 console_idx = log_next(console_idx);
2453 console_prev = msg->flags;
2454 raw_spin_unlock(&logbuf_lock);
2456 stop_critical_timings(); /* don't trace print latency */
2457 call_console_drivers(level, ext_text, ext_len, text, len);
2458 start_critical_timings();
2459 local_irq_restore(flags);
2461 if (do_cond_resched)
2466 /* Release the exclusive_console once it is used */
2467 if (unlikely(exclusive_console))
2468 exclusive_console = NULL;
2470 raw_spin_unlock(&logbuf_lock);
2475 * Someone could have filled up the buffer again, so re-check if there's
2476 * something to flush. In case we cannot trylock the console_sem again,
2477 * there's a new owner and the console_unlock() from them will do the
2478 * flush, no worries.
2480 raw_spin_lock(&logbuf_lock);
2481 retry = console_seq != log_next_seq;
2482 raw_spin_unlock_irqrestore(&logbuf_lock, flags);
2484 if (retry && console_trylock())
2490 EXPORT_SYMBOL(console_unlock);
2493 * console_conditional_schedule - yield the CPU if required
2495 * If the console code is currently allowed to sleep, and
2496 * if this CPU should yield the CPU to another task, do
2499 * Must be called within console_lock();.
2501 void __sched console_conditional_schedule(void)
2503 if (console_may_schedule)
2506 EXPORT_SYMBOL(console_conditional_schedule);
2508 void console_unblank(void)
2513 * console_unblank can no longer be called in interrupt context unless
2514 * oops_in_progress is set to 1..
2516 if (oops_in_progress) {
2517 if (down_trylock_console_sem() != 0)
2523 console_may_schedule = 0;
2525 if ((c->flags & CON_ENABLED) && c->unblank)
2531 * console_flush_on_panic - flush console content on panic
2533 * Immediately output all pending messages no matter what.
2535 void console_flush_on_panic(void)
2538 * If someone else is holding the console lock, trylock will fail
2539 * and may_schedule may be set. Ignore and proceed to unlock so
2540 * that messages are flushed out. As this can be called from any
2541 * context and we don't want to get preempted while flushing,
2542 * ensure may_schedule is cleared.
2545 console_may_schedule = 0;
2550 * Return the console tty driver structure and its associated index
2552 struct tty_driver *console_device(int *index)
2555 struct tty_driver *driver = NULL;
2558 for_each_console(c) {
2561 driver = c->device(c, index);
2570 * Prevent further output on the passed console device so that (for example)
2571 * serial drivers can disable console output before suspending a port, and can
2572 * re-enable output afterwards.
2574 void console_stop(struct console *console)
2577 console->flags &= ~CON_ENABLED;
2580 EXPORT_SYMBOL(console_stop);
2582 void console_start(struct console *console)
2585 console->flags |= CON_ENABLED;
2588 EXPORT_SYMBOL(console_start);
2590 static int __read_mostly keep_bootcon;
2592 static int __init keep_bootcon_setup(char *str)
2595 pr_info("debug: skip boot console de-registration.\n");
2600 early_param("keep_bootcon", keep_bootcon_setup);
2603 * The console driver calls this routine during kernel initialization
2604 * to register the console printing procedure with printk() and to
2605 * print any messages that were printed by the kernel before the
2606 * console driver was initialized.
2608 * This can happen pretty early during the boot process (because of
2609 * early_printk) - sometimes before setup_arch() completes - be careful
2610 * of what kernel features are used - they may not be initialised yet.
2612 * There are two types of consoles - bootconsoles (early_printk) and
2613 * "real" consoles (everything which is not a bootconsole) which are
2614 * handled differently.
2615 * - Any number of bootconsoles can be registered at any time.
2616 * - As soon as a "real" console is registered, all bootconsoles
2617 * will be unregistered automatically.
2618 * - Once a "real" console is registered, any attempt to register a
2619 * bootconsoles will be rejected
2621 void register_console(struct console *newcon)
2624 unsigned long flags;
2625 struct console *bcon = NULL;
2626 struct console_cmdline *c;
2628 if (console_drivers)
2629 for_each_console(bcon)
2630 if (WARN(bcon == newcon,
2631 "console '%s%d' already registered\n",
2632 bcon->name, bcon->index))
2636 * before we register a new CON_BOOT console, make sure we don't
2637 * already have a valid console
2639 if (console_drivers && newcon->flags & CON_BOOT) {
2640 /* find the last or real console */
2641 for_each_console(bcon) {
2642 if (!(bcon->flags & CON_BOOT)) {
2643 pr_info("Too late to register bootconsole %s%d\n",
2644 newcon->name, newcon->index);
2650 if (console_drivers && console_drivers->flags & CON_BOOT)
2651 bcon = console_drivers;
2653 if (preferred_console < 0 || bcon || !console_drivers)
2654 preferred_console = selected_console;
2657 * See if we want to use this console driver. If we
2658 * didn't select a console we take the first one
2659 * that registers here.
2661 if (preferred_console < 0 && !of_specified_console) {
2662 if (newcon->index < 0)
2664 if (newcon->setup == NULL ||
2665 newcon->setup(newcon, NULL) == 0) {
2666 newcon->flags |= CON_ENABLED;
2667 if (newcon->device) {
2668 newcon->flags |= CON_CONSDEV;
2669 preferred_console = 0;
2675 * See if this console matches one we selected on
2678 for (i = 0, c = console_cmdline;
2679 i < MAX_CMDLINECONSOLES && c->name[0];
2681 if (!newcon->match ||
2682 newcon->match(newcon, c->name, c->index, c->options) != 0) {
2683 /* default matching */
2684 BUILD_BUG_ON(sizeof(c->name) != sizeof(newcon->name));
2685 if (strcmp(c->name, newcon->name) != 0)
2687 if (newcon->index >= 0 &&
2688 newcon->index != c->index)
2690 if (newcon->index < 0)
2691 newcon->index = c->index;
2693 if (_braille_register_console(newcon, c))
2696 if (newcon->setup &&
2697 newcon->setup(newcon, c->options) != 0)
2701 newcon->flags |= CON_ENABLED;
2702 if (i == selected_console) {
2703 newcon->flags |= CON_CONSDEV;
2704 preferred_console = selected_console;
2709 if (!(newcon->flags & CON_ENABLED))
2713 * If we have a bootconsole, and are switching to a real console,
2714 * don't print everything out again, since when the boot console, and
2715 * the real console are the same physical device, it's annoying to
2716 * see the beginning boot messages twice
2718 if (bcon && ((newcon->flags & (CON_CONSDEV | CON_BOOT)) == CON_CONSDEV))
2719 newcon->flags &= ~CON_PRINTBUFFER;
2722 * Put this console in the list - keep the
2723 * preferred driver at the head of the list.
2726 if ((newcon->flags & CON_CONSDEV) || console_drivers == NULL) {
2727 newcon->next = console_drivers;
2728 console_drivers = newcon;
2730 newcon->next->flags &= ~CON_CONSDEV;
2732 newcon->next = console_drivers->next;
2733 console_drivers->next = newcon;
2736 if (newcon->flags & CON_EXTENDED)
2737 if (!nr_ext_console_drivers++)
2738 pr_info("printk: continuation disabled due to ext consoles, expect more fragments in /dev/kmsg\n");
2740 if (newcon->flags & CON_PRINTBUFFER) {
2742 * console_unlock(); will print out the buffered messages
2745 raw_spin_lock_irqsave(&logbuf_lock, flags);
2746 console_seq = syslog_seq;
2747 console_idx = syslog_idx;
2748 console_prev = syslog_prev;
2749 raw_spin_unlock_irqrestore(&logbuf_lock, flags);
2751 * We're about to replay the log buffer. Only do this to the
2752 * just-registered console to avoid excessive message spam to
2753 * the already-registered consoles.
2755 exclusive_console = newcon;
2758 console_sysfs_notify();
2761 * By unregistering the bootconsoles after we enable the real console
2762 * we get the "console xxx enabled" message on all the consoles -
2763 * boot consoles, real consoles, etc - this is to ensure that end
2764 * users know there might be something in the kernel's log buffer that
2765 * went to the bootconsole (that they do not see on the real console)
2767 pr_info("%sconsole [%s%d] enabled\n",
2768 (newcon->flags & CON_BOOT) ? "boot" : "" ,
2769 newcon->name, newcon->index);
2771 ((newcon->flags & (CON_CONSDEV | CON_BOOT)) == CON_CONSDEV) &&
2773 /* We need to iterate through all boot consoles, to make
2774 * sure we print everything out, before we unregister them.
2776 for_each_console(bcon)
2777 if (bcon->flags & CON_BOOT)
2778 unregister_console(bcon);
2781 EXPORT_SYMBOL(register_console);
2783 int unregister_console(struct console *console)
2785 struct console *a, *b;
2788 pr_info("%sconsole [%s%d] disabled\n",
2789 (console->flags & CON_BOOT) ? "boot" : "" ,
2790 console->name, console->index);
2792 res = _braille_unregister_console(console);
2798 if (console_drivers == console) {
2799 console_drivers=console->next;
2801 } else if (console_drivers) {
2802 for (a=console_drivers->next, b=console_drivers ;
2803 a; b=a, a=b->next) {
2812 if (!res && (console->flags & CON_EXTENDED))
2813 nr_ext_console_drivers--;
2816 * If this isn't the last console and it has CON_CONSDEV set, we
2817 * need to set it on the next preferred console.
2819 if (console_drivers != NULL && console->flags & CON_CONSDEV)
2820 console_drivers->flags |= CON_CONSDEV;
2822 console->flags &= ~CON_ENABLED;
2824 console_sysfs_notify();
2827 EXPORT_SYMBOL(unregister_console);
2830 * Some boot consoles access data that is in the init section and which will
2831 * be discarded after the initcalls have been run. To make sure that no code
2832 * will access this data, unregister the boot consoles in a late initcall.
2834 * If for some reason, such as deferred probe or the driver being a loadable
2835 * module, the real console hasn't registered yet at this point, there will
2836 * be a brief interval in which no messages are logged to the console, which
2837 * makes it difficult to diagnose problems that occur during this time.
2839 * To mitigate this problem somewhat, only unregister consoles whose memory
2840 * intersects with the init section. Note that code exists elsewhere to get
2841 * rid of the boot console as soon as the proper console shows up, so there
2842 * won't be side-effects from postponing the removal.
2844 static int __init printk_late_init(void)
2846 struct console *con;
2848 for_each_console(con) {
2849 if (!keep_bootcon && con->flags & CON_BOOT) {
2851 * Make sure to unregister boot consoles whose data
2852 * resides in the init section before the init section
2853 * is discarded. Boot consoles whose data will stick
2854 * around will automatically be unregistered when the
2855 * proper console replaces them.
2857 if (init_section_intersects(con, sizeof(*con)))
2858 unregister_console(con);
2861 hotcpu_notifier(console_cpu_notify, 0);
2864 late_initcall(printk_late_init);
2866 #if defined CONFIG_PRINTK
2868 * Delayed printk version, for scheduler-internal messages:
2870 #define PRINTK_PENDING_WAKEUP 0x01
2871 #define PRINTK_PENDING_OUTPUT 0x02
2873 static DEFINE_PER_CPU(int, printk_pending);
2875 static void wake_up_klogd_work_func(struct irq_work *irq_work)
2877 int pending = __this_cpu_xchg(printk_pending, 0);
2879 if (pending & PRINTK_PENDING_OUTPUT) {
2880 /* If trylock fails, someone else is doing the printing */
2881 if (console_trylock())
2885 if (pending & PRINTK_PENDING_WAKEUP)
2886 wake_up_interruptible(&log_wait);
2889 static DEFINE_PER_CPU(struct irq_work, wake_up_klogd_work) = {
2890 .func = wake_up_klogd_work_func,
2891 .flags = IRQ_WORK_LAZY,
2894 void wake_up_klogd(void)
2897 if (waitqueue_active(&log_wait)) {
2898 this_cpu_or(printk_pending, PRINTK_PENDING_WAKEUP);
2899 irq_work_queue(this_cpu_ptr(&wake_up_klogd_work));
2904 int printk_deferred(const char *fmt, ...)
2910 va_start(args, fmt);
2911 r = vprintk_emit(0, LOGLEVEL_SCHED, NULL, 0, fmt, args);
2914 __this_cpu_or(printk_pending, PRINTK_PENDING_OUTPUT);
2915 irq_work_queue(this_cpu_ptr(&wake_up_klogd_work));
2922 * printk rate limiting, lifted from the networking subsystem.
2924 * This enforces a rate limit: not more than 10 kernel messages
2925 * every 5s to make a denial-of-service attack impossible.
2927 DEFINE_RATELIMIT_STATE(printk_ratelimit_state, 5 * HZ, 10);
2929 int __printk_ratelimit(const char *func)
2931 return ___ratelimit(&printk_ratelimit_state, func);
2933 EXPORT_SYMBOL(__printk_ratelimit);
2936 * printk_timed_ratelimit - caller-controlled printk ratelimiting
2937 * @caller_jiffies: pointer to caller's state
2938 * @interval_msecs: minimum interval between prints
2940 * printk_timed_ratelimit() returns true if more than @interval_msecs
2941 * milliseconds have elapsed since the last time printk_timed_ratelimit()
2944 bool printk_timed_ratelimit(unsigned long *caller_jiffies,
2945 unsigned int interval_msecs)
2947 unsigned long elapsed = jiffies - *caller_jiffies;
2949 if (*caller_jiffies && elapsed <= msecs_to_jiffies(interval_msecs))
2952 *caller_jiffies = jiffies;
2955 EXPORT_SYMBOL(printk_timed_ratelimit);
2957 static DEFINE_SPINLOCK(dump_list_lock);
2958 static LIST_HEAD(dump_list);
2961 * kmsg_dump_register - register a kernel log dumper.
2962 * @dumper: pointer to the kmsg_dumper structure
2964 * Adds a kernel log dumper to the system. The dump callback in the
2965 * structure will be called when the kernel oopses or panics and must be
2966 * set. Returns zero on success and %-EINVAL or %-EBUSY otherwise.
2968 int kmsg_dump_register(struct kmsg_dumper *dumper)
2970 unsigned long flags;
2973 /* The dump callback needs to be set */
2977 spin_lock_irqsave(&dump_list_lock, flags);
2978 /* Don't allow registering multiple times */
2979 if (!dumper->registered) {
2980 dumper->registered = 1;
2981 list_add_tail_rcu(&dumper->list, &dump_list);
2984 spin_unlock_irqrestore(&dump_list_lock, flags);
2988 EXPORT_SYMBOL_GPL(kmsg_dump_register);
2991 * kmsg_dump_unregister - unregister a kmsg dumper.
2992 * @dumper: pointer to the kmsg_dumper structure
2994 * Removes a dump device from the system. Returns zero on success and
2995 * %-EINVAL otherwise.
2997 int kmsg_dump_unregister(struct kmsg_dumper *dumper)
2999 unsigned long flags;
3002 spin_lock_irqsave(&dump_list_lock, flags);
3003 if (dumper->registered) {
3004 dumper->registered = 0;
3005 list_del_rcu(&dumper->list);
3008 spin_unlock_irqrestore(&dump_list_lock, flags);
3013 EXPORT_SYMBOL_GPL(kmsg_dump_unregister);
3015 static bool always_kmsg_dump;
3016 module_param_named(always_kmsg_dump, always_kmsg_dump, bool, S_IRUGO | S_IWUSR);
3019 * kmsg_dump - dump kernel log to kernel message dumpers.
3020 * @reason: the reason (oops, panic etc) for dumping
3022 * Call each of the registered dumper's dump() callback, which can
3023 * retrieve the kmsg records with kmsg_dump_get_line() or
3024 * kmsg_dump_get_buffer().
3026 void kmsg_dump(enum kmsg_dump_reason reason)
3028 struct kmsg_dumper *dumper;
3029 unsigned long flags;
3031 if ((reason > KMSG_DUMP_OOPS) && !always_kmsg_dump)
3035 list_for_each_entry_rcu(dumper, &dump_list, list) {
3036 if (dumper->max_reason && reason > dumper->max_reason)
3039 /* initialize iterator with data about the stored records */
3040 dumper->active = true;
3042 raw_spin_lock_irqsave(&logbuf_lock, flags);
3043 dumper->cur_seq = clear_seq;
3044 dumper->cur_idx = clear_idx;
3045 dumper->next_seq = log_next_seq;
3046 dumper->next_idx = log_next_idx;
3047 raw_spin_unlock_irqrestore(&logbuf_lock, flags);
3049 /* invoke dumper which will iterate over records */
3050 dumper->dump(dumper, reason);
3052 /* reset iterator */
3053 dumper->active = false;
3059 * kmsg_dump_get_line_nolock - retrieve one kmsg log line (unlocked version)
3060 * @dumper: registered kmsg dumper
3061 * @syslog: include the "<4>" prefixes
3062 * @line: buffer to copy the line to
3063 * @size: maximum size of the buffer
3064 * @len: length of line placed into buffer
3066 * Start at the beginning of the kmsg buffer, with the oldest kmsg
3067 * record, and copy one record into the provided buffer.
3069 * Consecutive calls will return the next available record moving
3070 * towards the end of the buffer with the youngest messages.
3072 * A return value of FALSE indicates that there are no more records to
3075 * The function is similar to kmsg_dump_get_line(), but grabs no locks.
3077 bool kmsg_dump_get_line_nolock(struct kmsg_dumper *dumper, bool syslog,
3078 char *line, size_t size, size_t *len)
3080 struct printk_log *msg;
3084 if (!dumper->active)
3087 if (dumper->cur_seq < log_first_seq) {
3088 /* messages are gone, move to first available one */
3089 dumper->cur_seq = log_first_seq;
3090 dumper->cur_idx = log_first_idx;
3094 if (dumper->cur_seq >= log_next_seq)
3097 msg = log_from_idx(dumper->cur_idx);
3098 l = msg_print_text(msg, 0, syslog, line, size);
3100 dumper->cur_idx = log_next(dumper->cur_idx);
3110 * kmsg_dump_get_line - retrieve one kmsg log line
3111 * @dumper: registered kmsg dumper
3112 * @syslog: include the "<4>" prefixes
3113 * @line: buffer to copy the line to
3114 * @size: maximum size of the buffer
3115 * @len: length of line placed into buffer
3117 * Start at the beginning of the kmsg buffer, with the oldest kmsg
3118 * record, and copy one record into the provided buffer.
3120 * Consecutive calls will return the next available record moving
3121 * towards the end of the buffer with the youngest messages.
3123 * A return value of FALSE indicates that there are no more records to
3126 bool kmsg_dump_get_line(struct kmsg_dumper *dumper, bool syslog,
3127 char *line, size_t size, size_t *len)
3129 unsigned long flags;
3132 raw_spin_lock_irqsave(&logbuf_lock, flags);
3133 ret = kmsg_dump_get_line_nolock(dumper, syslog, line, size, len);
3134 raw_spin_unlock_irqrestore(&logbuf_lock, flags);
3138 EXPORT_SYMBOL_GPL(kmsg_dump_get_line);
3141 * kmsg_dump_get_buffer - copy kmsg log lines
3142 * @dumper: registered kmsg dumper
3143 * @syslog: include the "<4>" prefixes
3144 * @buf: buffer to copy the line to
3145 * @size: maximum size of the buffer
3146 * @len: length of line placed into buffer
3148 * Start at the end of the kmsg buffer and fill the provided buffer
3149 * with as many of the the *youngest* kmsg records that fit into it.
3150 * If the buffer is large enough, all available kmsg records will be
3151 * copied with a single call.
3153 * Consecutive calls will fill the buffer with the next block of
3154 * available older records, not including the earlier retrieved ones.
3156 * A return value of FALSE indicates that there are no more records to
3159 bool kmsg_dump_get_buffer(struct kmsg_dumper *dumper, bool syslog,
3160 char *buf, size_t size, size_t *len)
3162 unsigned long flags;
3167 enum log_flags prev;
3171 if (!dumper->active)
3174 raw_spin_lock_irqsave(&logbuf_lock, flags);
3175 if (dumper->cur_seq < log_first_seq) {
3176 /* messages are gone, move to first available one */
3177 dumper->cur_seq = log_first_seq;
3178 dumper->cur_idx = log_first_idx;
3182 if (dumper->cur_seq >= dumper->next_seq) {
3183 raw_spin_unlock_irqrestore(&logbuf_lock, flags);
3187 /* calculate length of entire buffer */
3188 seq = dumper->cur_seq;
3189 idx = dumper->cur_idx;
3191 while (seq < dumper->next_seq) {
3192 struct printk_log *msg = log_from_idx(idx);
3194 l += msg_print_text(msg, prev, true, NULL, 0);
3195 idx = log_next(idx);
3200 /* move first record forward until length fits into the buffer */
3201 seq = dumper->cur_seq;
3202 idx = dumper->cur_idx;
3204 while (l > size && seq < dumper->next_seq) {
3205 struct printk_log *msg = log_from_idx(idx);
3207 l -= msg_print_text(msg, prev, true, NULL, 0);
3208 idx = log_next(idx);
3213 /* last message in next interation */
3218 while (seq < dumper->next_seq) {
3219 struct printk_log *msg = log_from_idx(idx);
3221 l += msg_print_text(msg, prev, syslog, buf + l, size - l);
3222 idx = log_next(idx);
3227 dumper->next_seq = next_seq;
3228 dumper->next_idx = next_idx;
3230 raw_spin_unlock_irqrestore(&logbuf_lock, flags);
3236 EXPORT_SYMBOL_GPL(kmsg_dump_get_buffer);
3239 * kmsg_dump_rewind_nolock - reset the interator (unlocked version)
3240 * @dumper: registered kmsg dumper
3242 * Reset the dumper's iterator so that kmsg_dump_get_line() and
3243 * kmsg_dump_get_buffer() can be called again and used multiple
3244 * times within the same dumper.dump() callback.
3246 * The function is similar to kmsg_dump_rewind(), but grabs no locks.
3248 void kmsg_dump_rewind_nolock(struct kmsg_dumper *dumper)
3250 dumper->cur_seq = clear_seq;
3251 dumper->cur_idx = clear_idx;
3252 dumper->next_seq = log_next_seq;
3253 dumper->next_idx = log_next_idx;
3257 * kmsg_dump_rewind - reset the interator
3258 * @dumper: registered kmsg dumper
3260 * Reset the dumper's iterator so that kmsg_dump_get_line() and
3261 * kmsg_dump_get_buffer() can be called again and used multiple
3262 * times within the same dumper.dump() callback.
3264 void kmsg_dump_rewind(struct kmsg_dumper *dumper)
3266 unsigned long flags;
3268 raw_spin_lock_irqsave(&logbuf_lock, flags);
3269 kmsg_dump_rewind_nolock(dumper);
3270 raw_spin_unlock_irqrestore(&logbuf_lock, flags);
3272 EXPORT_SYMBOL_GPL(kmsg_dump_rewind);
3274 static char dump_stack_arch_desc_str[128];
3277 * dump_stack_set_arch_desc - set arch-specific str to show with task dumps
3278 * @fmt: printf-style format string
3279 * @...: arguments for the format string
3281 * The configured string will be printed right after utsname during task
3282 * dumps. Usually used to add arch-specific system identifiers. If an
3283 * arch wants to make use of such an ID string, it should initialize this
3284 * as soon as possible during boot.
3286 void __init dump_stack_set_arch_desc(const char *fmt, ...)
3290 va_start(args, fmt);
3291 vsnprintf(dump_stack_arch_desc_str, sizeof(dump_stack_arch_desc_str),
3297 * dump_stack_print_info - print generic debug info for dump_stack()
3298 * @log_lvl: log level
3300 * Arch-specific dump_stack() implementations can use this function to
3301 * print out the same debug information as the generic dump_stack().
3303 void dump_stack_print_info(const char *log_lvl)
3305 printk("%sCPU: %d PID: %d Comm: %.20s %s %s %.*s\n",
3306 log_lvl, raw_smp_processor_id(), current->pid, current->comm,
3307 print_tainted(), init_utsname()->release,
3308 (int)strcspn(init_utsname()->version, " "),
3309 init_utsname()->version);
3311 if (dump_stack_arch_desc_str[0] != '\0')
3312 printk("%sHardware name: %s\n",
3313 log_lvl, dump_stack_arch_desc_str);
3315 print_worker_info(log_lvl, current);
3319 * show_regs_print_info - print generic debug info for show_regs()
3320 * @log_lvl: log level
3322 * show_regs() implementations can use this function to print out generic
3323 * debug information.
3325 void show_regs_print_info(const char *log_lvl)
3327 dump_stack_print_info(log_lvl);
3329 printk("%stask: %p task.stack: %p\n",
3330 log_lvl, current, task_stack_page(current));