sh_eth: fix EESIPR values for SH77{34|63}
[linux/fpc-iii.git] / kernel / printk / printk.c
blob8b2696420abb5de7347429f52d9a3d46d4dbcd24
1 /*
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).
11 * Ted Ts'o, 2/11/93.
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>
20 #include <linux/mm.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 <linux/uaccess.h>
50 #include <asm/sections.h>
52 #define CREATE_TRACE_POINTS
53 #include <trace/events/printk.h>
55 #include "console_cmdline.h"
56 #include "braille.h"
57 #include "internal.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.
70 int oops_in_progress;
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
76 * driver system.
78 static DEFINE_SEMAPHORE(console_sem);
79 struct console *console_drivers;
80 EXPORT_SYMBOL_GPL(console_drivers);
82 #ifdef CONFIG_LOCKDEP
83 static struct lockdep_map console_lock_dep_map = {
84 .name = "console_lock"
86 #endif
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)
107 if (!str)
108 return -EINVAL;
110 if (!strncmp(str, "on", 2)) {
111 devkmsg_log = DEVKMSG_LOG_MASK_ON;
112 return 2;
113 } else if (!strncmp(str, "off", 3)) {
114 devkmsg_log = DEVKMSG_LOG_MASK_OFF;
115 return 3;
116 } else if (!strncmp(str, "ratelimit", 9)) {
117 devkmsg_log = DEVKMSG_LOG_MASK_DEFAULT;
118 return 9;
120 return -EINVAL;
123 static int __init control_devkmsg(char *str)
125 if (__control_devkmsg(str) < 0)
126 return 1;
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;
148 return 0;
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];
158 unsigned int old;
159 int err;
161 if (write) {
162 if (devkmsg_log & DEVKMSG_LOG_MASK_LOCK)
163 return -EINVAL;
165 old = devkmsg_log;
166 strncpy(old_str, devkmsg_log_str, DEVKMSG_STR_MAX_SIZE);
169 err = proc_dostring(table, write, buffer, lenp, ppos);
170 if (err)
171 return err;
173 if (write) {
174 err = __control_devkmsg(devkmsg_log_str);
177 * Do not accept an unknown string OR a known string with
178 * trailing crap...
180 if (err < 0 || (err + 1 != *lenp)) {
182 /* ... and restore old setting. */
183 devkmsg_log = old;
184 strncpy(devkmsg_log_str, old_str, DEVKMSG_STR_MAX_SIZE);
186 return -EINVAL;
190 return 0;
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 { \
210 down(&console_sem);\
211 mutex_acquire(&console_lock_dep_map, 0, 0, _RET_IP_);\
212 } while (0)
214 static int __down_trylock_console_sem(unsigned long ip)
216 if (down_trylock(&console_sem))
217 return 1;
218 mutex_acquire(&console_lock_dep_map, 0, 1, ip);
219 return 0;
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_);\
225 up(&console_sem);\
226 } while (0)
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);
256 /* Flag: console code may call schedule() */
257 static int console_may_schedule;
260 * The printk log buffer consists of a chain of concatenated variable
261 * length records. Every record starts with a record header, containing
262 * the overall length of the record.
264 * The heads to the first and last entry in the buffer, as well as the
265 * sequence numbers of these entries are maintained when messages are
266 * stored.
268 * If the heads indicate available messages, the length in the header
269 * tells the start next message. A length == 0 for the next message
270 * indicates a wrap-around to the beginning of the buffer.
272 * Every record carries the monotonic timestamp in microseconds, as well as
273 * the standard userspace syslog level and syslog facility. The usual
274 * kernel messages use LOG_KERN; userspace-injected messages always carry
275 * a matching syslog facility, by default LOG_USER. The origin of every
276 * message can be reliably determined that way.
278 * The human readable log message directly follows the message header. The
279 * length of the message text is stored in the header, the stored message
280 * is not terminated.
282 * Optionally, a message can carry a dictionary of properties (key/value pairs),
283 * to provide userspace with a machine-readable message context.
285 * Examples for well-defined, commonly used property names are:
286 * DEVICE=b12:8 device identifier
287 * b12:8 block dev_t
288 * c127:3 char dev_t
289 * n8 netdev ifindex
290 * +sound:card0 subsystem:devname
291 * SUBSYSTEM=pci driver-core subsystem name
293 * Valid characters in property names are [a-zA-Z0-9.-_]. The plain text value
294 * follows directly after a '=' character. Every property is terminated by
295 * a '\0' character. The last property is not terminated.
297 * Example of a message structure:
298 * 0000 ff 8f 00 00 00 00 00 00 monotonic time in nsec
299 * 0008 34 00 record is 52 bytes long
300 * 000a 0b 00 text is 11 bytes long
301 * 000c 1f 00 dictionary is 23 bytes long
302 * 000e 03 00 LOG_KERN (facility) LOG_ERR (level)
303 * 0010 69 74 27 73 20 61 20 6c "it's a l"
304 * 69 6e 65 "ine"
305 * 001b 44 45 56 49 43 "DEVIC"
306 * 45 3d 62 38 3a 32 00 44 "E=b8:2\0D"
307 * 52 49 56 45 52 3d 62 75 "RIVER=bu"
308 * 67 "g"
309 * 0032 00 00 00 padding to next message header
311 * The 'struct printk_log' buffer header must never be directly exported to
312 * userspace, it is a kernel-private implementation detail that might
313 * need to be changed in the future, when the requirements change.
315 * /dev/kmsg exports the structured data in the following line format:
316 * "<level>,<sequnum>,<timestamp>,<contflag>[,additional_values, ... ];<message text>\n"
318 * Users of the export format should ignore possible additional values
319 * separated by ',', and find the message after the ';' character.
321 * The optional key/value pairs are attached as continuation lines starting
322 * with a space character and terminated by a newline. All possible
323 * non-prinatable characters are escaped in the "\xff" notation.
326 enum log_flags {
327 LOG_NOCONS = 1, /* already flushed, do not print to console */
328 LOG_NEWLINE = 2, /* text ended with a newline */
329 LOG_PREFIX = 4, /* text started with a prefix */
330 LOG_CONT = 8, /* text is a fragment of a continuation line */
333 struct printk_log {
334 u64 ts_nsec; /* timestamp in nanoseconds */
335 u16 len; /* length of entire record */
336 u16 text_len; /* length of text buffer */
337 u16 dict_len; /* length of dictionary buffer */
338 u8 facility; /* syslog facility */
339 u8 flags:5; /* internal record flags */
340 u8 level:3; /* syslog level */
342 #ifdef CONFIG_HAVE_EFFICIENT_UNALIGNED_ACCESS
343 __packed __aligned(4)
344 #endif
348 * The logbuf_lock protects kmsg buffer, indices, counters. This can be taken
349 * within the scheduler's rq lock. It must be released before calling
350 * console_unlock() or anything else that might wake up a process.
352 DEFINE_RAW_SPINLOCK(logbuf_lock);
354 #ifdef CONFIG_PRINTK
355 DECLARE_WAIT_QUEUE_HEAD(log_wait);
356 /* the next printk record to read by syslog(READ) or /proc/kmsg */
357 static u64 syslog_seq;
358 static u32 syslog_idx;
359 static size_t syslog_partial;
361 /* index and sequence number of the first record stored in the buffer */
362 static u64 log_first_seq;
363 static u32 log_first_idx;
365 /* index and sequence number of the next record to store in the buffer */
366 static u64 log_next_seq;
367 static u32 log_next_idx;
369 /* the next printk record to write to the console */
370 static u64 console_seq;
371 static u32 console_idx;
373 /* the next printk record to read after the last 'clear' command */
374 static u64 clear_seq;
375 static u32 clear_idx;
377 #define PREFIX_MAX 32
378 #define LOG_LINE_MAX (1024 - PREFIX_MAX)
380 #define LOG_LEVEL(v) ((v) & 0x07)
381 #define LOG_FACILITY(v) ((v) >> 3 & 0xff)
383 /* record buffer */
384 #define LOG_ALIGN __alignof__(struct printk_log)
385 #define __LOG_BUF_LEN (1 << CONFIG_LOG_BUF_SHIFT)
386 static char __log_buf[__LOG_BUF_LEN] __aligned(LOG_ALIGN);
387 static char *log_buf = __log_buf;
388 static u32 log_buf_len = __LOG_BUF_LEN;
390 /* Return log buffer address */
391 char *log_buf_addr_get(void)
393 return log_buf;
396 /* Return log buffer size */
397 u32 log_buf_len_get(void)
399 return log_buf_len;
402 /* human readable text of the record */
403 static char *log_text(const struct printk_log *msg)
405 return (char *)msg + sizeof(struct printk_log);
408 /* optional key/value pair dictionary attached to the record */
409 static char *log_dict(const struct printk_log *msg)
411 return (char *)msg + sizeof(struct printk_log) + msg->text_len;
414 /* get record by index; idx must point to valid msg */
415 static struct printk_log *log_from_idx(u32 idx)
417 struct printk_log *msg = (struct printk_log *)(log_buf + idx);
420 * A length == 0 record is the end of buffer marker. Wrap around and
421 * read the message at the start of the buffer.
423 if (!msg->len)
424 return (struct printk_log *)log_buf;
425 return msg;
428 /* get next record; idx must point to valid msg */
429 static u32 log_next(u32 idx)
431 struct printk_log *msg = (struct printk_log *)(log_buf + idx);
433 /* length == 0 indicates the end of the buffer; wrap */
435 * A length == 0 record is the end of buffer marker. Wrap around and
436 * read the message at the start of the buffer as *this* one, and
437 * return the one after that.
439 if (!msg->len) {
440 msg = (struct printk_log *)log_buf;
441 return msg->len;
443 return idx + msg->len;
447 * Check whether there is enough free space for the given message.
449 * The same values of first_idx and next_idx mean that the buffer
450 * is either empty or full.
452 * If the buffer is empty, we must respect the position of the indexes.
453 * They cannot be reset to the beginning of the buffer.
455 static int logbuf_has_space(u32 msg_size, bool empty)
457 u32 free;
459 if (log_next_idx > log_first_idx || empty)
460 free = max(log_buf_len - log_next_idx, log_first_idx);
461 else
462 free = log_first_idx - log_next_idx;
465 * We need space also for an empty header that signalizes wrapping
466 * of the buffer.
468 return free >= msg_size + sizeof(struct printk_log);
471 static int log_make_free_space(u32 msg_size)
473 while (log_first_seq < log_next_seq &&
474 !logbuf_has_space(msg_size, false)) {
475 /* drop old messages until we have enough contiguous space */
476 log_first_idx = log_next(log_first_idx);
477 log_first_seq++;
480 if (clear_seq < log_first_seq) {
481 clear_seq = log_first_seq;
482 clear_idx = log_first_idx;
485 /* sequence numbers are equal, so the log buffer is empty */
486 if (logbuf_has_space(msg_size, log_first_seq == log_next_seq))
487 return 0;
489 return -ENOMEM;
492 /* compute the message size including the padding bytes */
493 static u32 msg_used_size(u16 text_len, u16 dict_len, u32 *pad_len)
495 u32 size;
497 size = sizeof(struct printk_log) + text_len + dict_len;
498 *pad_len = (-size) & (LOG_ALIGN - 1);
499 size += *pad_len;
501 return size;
505 * Define how much of the log buffer we could take at maximum. The value
506 * must be greater than two. Note that only half of the buffer is available
507 * when the index points to the middle.
509 #define MAX_LOG_TAKE_PART 4
510 static const char trunc_msg[] = "<truncated>";
512 static u32 truncate_msg(u16 *text_len, u16 *trunc_msg_len,
513 u16 *dict_len, u32 *pad_len)
516 * The message should not take the whole buffer. Otherwise, it might
517 * get removed too soon.
519 u32 max_text_len = log_buf_len / MAX_LOG_TAKE_PART;
520 if (*text_len > max_text_len)
521 *text_len = max_text_len;
522 /* enable the warning message */
523 *trunc_msg_len = strlen(trunc_msg);
524 /* disable the "dict" completely */
525 *dict_len = 0;
526 /* compute the size again, count also the warning message */
527 return msg_used_size(*text_len + *trunc_msg_len, 0, pad_len);
530 /* insert record into the buffer, discard old ones, update heads */
531 static int log_store(int facility, int level,
532 enum log_flags flags, u64 ts_nsec,
533 const char *dict, u16 dict_len,
534 const char *text, u16 text_len)
536 struct printk_log *msg;
537 u32 size, pad_len;
538 u16 trunc_msg_len = 0;
540 /* number of '\0' padding bytes to next message */
541 size = msg_used_size(text_len, dict_len, &pad_len);
543 if (log_make_free_space(size)) {
544 /* truncate the message if it is too long for empty buffer */
545 size = truncate_msg(&text_len, &trunc_msg_len,
546 &dict_len, &pad_len);
547 /* survive when the log buffer is too small for trunc_msg */
548 if (log_make_free_space(size))
549 return 0;
552 if (log_next_idx + size + sizeof(struct printk_log) > log_buf_len) {
554 * This message + an additional empty header does not fit
555 * at the end of the buffer. Add an empty header with len == 0
556 * to signify a wrap around.
558 memset(log_buf + log_next_idx, 0, sizeof(struct printk_log));
559 log_next_idx = 0;
562 /* fill message */
563 msg = (struct printk_log *)(log_buf + log_next_idx);
564 memcpy(log_text(msg), text, text_len);
565 msg->text_len = text_len;
566 if (trunc_msg_len) {
567 memcpy(log_text(msg) + text_len, trunc_msg, trunc_msg_len);
568 msg->text_len += trunc_msg_len;
570 memcpy(log_dict(msg), dict, dict_len);
571 msg->dict_len = dict_len;
572 msg->facility = facility;
573 msg->level = level & 7;
574 msg->flags = flags & 0x1f;
575 if (ts_nsec > 0)
576 msg->ts_nsec = ts_nsec;
577 else
578 msg->ts_nsec = local_clock();
579 memset(log_dict(msg) + dict_len, 0, pad_len);
580 msg->len = size;
582 /* insert message */
583 log_next_idx += msg->len;
584 log_next_seq++;
586 return msg->text_len;
589 int dmesg_restrict = IS_ENABLED(CONFIG_SECURITY_DMESG_RESTRICT);
591 static int syslog_action_restricted(int type)
593 if (dmesg_restrict)
594 return 1;
596 * Unless restricted, we allow "read all" and "get buffer size"
597 * for everybody.
599 return type != SYSLOG_ACTION_READ_ALL &&
600 type != SYSLOG_ACTION_SIZE_BUFFER;
603 int check_syslog_permissions(int type, int source)
606 * If this is from /proc/kmsg and we've already opened it, then we've
607 * already done the capabilities checks at open time.
609 if (source == SYSLOG_FROM_PROC && type != SYSLOG_ACTION_OPEN)
610 goto ok;
612 if (syslog_action_restricted(type)) {
613 if (capable(CAP_SYSLOG))
614 goto ok;
616 * For historical reasons, accept CAP_SYS_ADMIN too, with
617 * a warning.
619 if (capable(CAP_SYS_ADMIN)) {
620 pr_warn_once("%s (%d): Attempt to access syslog with "
621 "CAP_SYS_ADMIN but no CAP_SYSLOG "
622 "(deprecated).\n",
623 current->comm, task_pid_nr(current));
624 goto ok;
626 return -EPERM;
629 return security_syslog(type);
631 EXPORT_SYMBOL_GPL(check_syslog_permissions);
633 static void append_char(char **pp, char *e, char c)
635 if (*pp < e)
636 *(*pp)++ = c;
639 static ssize_t msg_print_ext_header(char *buf, size_t size,
640 struct printk_log *msg, u64 seq)
642 u64 ts_usec = msg->ts_nsec;
644 do_div(ts_usec, 1000);
646 return scnprintf(buf, size, "%u,%llu,%llu,%c;",
647 (msg->facility << 3) | msg->level, seq, ts_usec,
648 msg->flags & LOG_CONT ? 'c' : '-');
651 static ssize_t msg_print_ext_body(char *buf, size_t size,
652 char *dict, size_t dict_len,
653 char *text, size_t text_len)
655 char *p = buf, *e = buf + size;
656 size_t i;
658 /* escape non-printable characters */
659 for (i = 0; i < text_len; i++) {
660 unsigned char c = text[i];
662 if (c < ' ' || c >= 127 || c == '\\')
663 p += scnprintf(p, e - p, "\\x%02x", c);
664 else
665 append_char(&p, e, c);
667 append_char(&p, e, '\n');
669 if (dict_len) {
670 bool line = true;
672 for (i = 0; i < dict_len; i++) {
673 unsigned char c = dict[i];
675 if (line) {
676 append_char(&p, e, ' ');
677 line = false;
680 if (c == '\0') {
681 append_char(&p, e, '\n');
682 line = true;
683 continue;
686 if (c < ' ' || c >= 127 || c == '\\') {
687 p += scnprintf(p, e - p, "\\x%02x", c);
688 continue;
691 append_char(&p, e, c);
693 append_char(&p, e, '\n');
696 return p - buf;
699 /* /dev/kmsg - userspace message inject/listen interface */
700 struct devkmsg_user {
701 u64 seq;
702 u32 idx;
703 struct ratelimit_state rs;
704 struct mutex lock;
705 char buf[CONSOLE_EXT_LOG_MAX];
708 static ssize_t devkmsg_write(struct kiocb *iocb, struct iov_iter *from)
710 char *buf, *line;
711 int level = default_message_loglevel;
712 int facility = 1; /* LOG_USER */
713 struct file *file = iocb->ki_filp;
714 struct devkmsg_user *user = file->private_data;
715 size_t len = iov_iter_count(from);
716 ssize_t ret = len;
718 if (!user || len > LOG_LINE_MAX)
719 return -EINVAL;
721 /* Ignore when user logging is disabled. */
722 if (devkmsg_log & DEVKMSG_LOG_MASK_OFF)
723 return len;
725 /* Ratelimit when not explicitly enabled. */
726 if (!(devkmsg_log & DEVKMSG_LOG_MASK_ON)) {
727 if (!___ratelimit(&user->rs, current->comm))
728 return ret;
731 buf = kmalloc(len+1, GFP_KERNEL);
732 if (buf == NULL)
733 return -ENOMEM;
735 buf[len] = '\0';
736 if (!copy_from_iter_full(buf, len, from)) {
737 kfree(buf);
738 return -EFAULT;
742 * Extract and skip the syslog prefix <[0-9]*>. Coming from userspace
743 * the decimal value represents 32bit, the lower 3 bit are the log
744 * level, the rest are the log facility.
746 * If no prefix or no userspace facility is specified, we
747 * enforce LOG_USER, to be able to reliably distinguish
748 * kernel-generated messages from userspace-injected ones.
750 line = buf;
751 if (line[0] == '<') {
752 char *endp = NULL;
753 unsigned int u;
755 u = simple_strtoul(line + 1, &endp, 10);
756 if (endp && endp[0] == '>') {
757 level = LOG_LEVEL(u);
758 if (LOG_FACILITY(u) != 0)
759 facility = LOG_FACILITY(u);
760 endp++;
761 len -= endp - line;
762 line = endp;
766 printk_emit(facility, level, NULL, 0, "%s", line);
767 kfree(buf);
768 return ret;
771 static ssize_t devkmsg_read(struct file *file, char __user *buf,
772 size_t count, loff_t *ppos)
774 struct devkmsg_user *user = file->private_data;
775 struct printk_log *msg;
776 size_t len;
777 ssize_t ret;
779 if (!user)
780 return -EBADF;
782 ret = mutex_lock_interruptible(&user->lock);
783 if (ret)
784 return ret;
785 raw_spin_lock_irq(&logbuf_lock);
786 while (user->seq == log_next_seq) {
787 if (file->f_flags & O_NONBLOCK) {
788 ret = -EAGAIN;
789 raw_spin_unlock_irq(&logbuf_lock);
790 goto out;
793 raw_spin_unlock_irq(&logbuf_lock);
794 ret = wait_event_interruptible(log_wait,
795 user->seq != log_next_seq);
796 if (ret)
797 goto out;
798 raw_spin_lock_irq(&logbuf_lock);
801 if (user->seq < log_first_seq) {
802 /* our last seen message is gone, return error and reset */
803 user->idx = log_first_idx;
804 user->seq = log_first_seq;
805 ret = -EPIPE;
806 raw_spin_unlock_irq(&logbuf_lock);
807 goto out;
810 msg = log_from_idx(user->idx);
811 len = msg_print_ext_header(user->buf, sizeof(user->buf),
812 msg, user->seq);
813 len += msg_print_ext_body(user->buf + len, sizeof(user->buf) - len,
814 log_dict(msg), msg->dict_len,
815 log_text(msg), msg->text_len);
817 user->idx = log_next(user->idx);
818 user->seq++;
819 raw_spin_unlock_irq(&logbuf_lock);
821 if (len > count) {
822 ret = -EINVAL;
823 goto out;
826 if (copy_to_user(buf, user->buf, len)) {
827 ret = -EFAULT;
828 goto out;
830 ret = len;
831 out:
832 mutex_unlock(&user->lock);
833 return ret;
836 static loff_t devkmsg_llseek(struct file *file, loff_t offset, int whence)
838 struct devkmsg_user *user = file->private_data;
839 loff_t ret = 0;
841 if (!user)
842 return -EBADF;
843 if (offset)
844 return -ESPIPE;
846 raw_spin_lock_irq(&logbuf_lock);
847 switch (whence) {
848 case SEEK_SET:
849 /* the first record */
850 user->idx = log_first_idx;
851 user->seq = log_first_seq;
852 break;
853 case SEEK_DATA:
855 * The first record after the last SYSLOG_ACTION_CLEAR,
856 * like issued by 'dmesg -c'. Reading /dev/kmsg itself
857 * changes no global state, and does not clear anything.
859 user->idx = clear_idx;
860 user->seq = clear_seq;
861 break;
862 case SEEK_END:
863 /* after the last record */
864 user->idx = log_next_idx;
865 user->seq = log_next_seq;
866 break;
867 default:
868 ret = -EINVAL;
870 raw_spin_unlock_irq(&logbuf_lock);
871 return ret;
874 static unsigned int devkmsg_poll(struct file *file, poll_table *wait)
876 struct devkmsg_user *user = file->private_data;
877 int ret = 0;
879 if (!user)
880 return POLLERR|POLLNVAL;
882 poll_wait(file, &log_wait, wait);
884 raw_spin_lock_irq(&logbuf_lock);
885 if (user->seq < log_next_seq) {
886 /* return error when data has vanished underneath us */
887 if (user->seq < log_first_seq)
888 ret = POLLIN|POLLRDNORM|POLLERR|POLLPRI;
889 else
890 ret = POLLIN|POLLRDNORM;
892 raw_spin_unlock_irq(&logbuf_lock);
894 return ret;
897 static int devkmsg_open(struct inode *inode, struct file *file)
899 struct devkmsg_user *user;
900 int err;
902 if (devkmsg_log & DEVKMSG_LOG_MASK_OFF)
903 return -EPERM;
905 /* write-only does not need any file context */
906 if ((file->f_flags & O_ACCMODE) != O_WRONLY) {
907 err = check_syslog_permissions(SYSLOG_ACTION_READ_ALL,
908 SYSLOG_FROM_READER);
909 if (err)
910 return err;
913 user = kmalloc(sizeof(struct devkmsg_user), GFP_KERNEL);
914 if (!user)
915 return -ENOMEM;
917 ratelimit_default_init(&user->rs);
918 ratelimit_set_flags(&user->rs, RATELIMIT_MSG_ON_RELEASE);
920 mutex_init(&user->lock);
922 raw_spin_lock_irq(&logbuf_lock);
923 user->idx = log_first_idx;
924 user->seq = log_first_seq;
925 raw_spin_unlock_irq(&logbuf_lock);
927 file->private_data = user;
928 return 0;
931 static int devkmsg_release(struct inode *inode, struct file *file)
933 struct devkmsg_user *user = file->private_data;
935 if (!user)
936 return 0;
938 ratelimit_state_exit(&user->rs);
940 mutex_destroy(&user->lock);
941 kfree(user);
942 return 0;
945 const struct file_operations kmsg_fops = {
946 .open = devkmsg_open,
947 .read = devkmsg_read,
948 .write_iter = devkmsg_write,
949 .llseek = devkmsg_llseek,
950 .poll = devkmsg_poll,
951 .release = devkmsg_release,
954 #ifdef CONFIG_KEXEC_CORE
956 * This appends the listed symbols to /proc/vmcore
958 * /proc/vmcore is used by various utilities, like crash and makedumpfile to
959 * obtain access to symbols that are otherwise very difficult to locate. These
960 * symbols are specifically used so that utilities can access and extract the
961 * dmesg log from a vmcore file after a crash.
963 void log_buf_kexec_setup(void)
965 VMCOREINFO_SYMBOL(log_buf);
966 VMCOREINFO_SYMBOL(log_buf_len);
967 VMCOREINFO_SYMBOL(log_first_idx);
968 VMCOREINFO_SYMBOL(clear_idx);
969 VMCOREINFO_SYMBOL(log_next_idx);
971 * Export struct printk_log size and field offsets. User space tools can
972 * parse it and detect any changes to structure down the line.
974 VMCOREINFO_STRUCT_SIZE(printk_log);
975 VMCOREINFO_OFFSET(printk_log, ts_nsec);
976 VMCOREINFO_OFFSET(printk_log, len);
977 VMCOREINFO_OFFSET(printk_log, text_len);
978 VMCOREINFO_OFFSET(printk_log, dict_len);
980 #endif
982 /* requested log_buf_len from kernel cmdline */
983 static unsigned long __initdata new_log_buf_len;
985 /* we practice scaling the ring buffer by powers of 2 */
986 static void __init log_buf_len_update(unsigned size)
988 if (size)
989 size = roundup_pow_of_two(size);
990 if (size > log_buf_len)
991 new_log_buf_len = size;
994 /* save requested log_buf_len since it's too early to process it */
995 static int __init log_buf_len_setup(char *str)
997 unsigned size = memparse(str, &str);
999 log_buf_len_update(size);
1001 return 0;
1003 early_param("log_buf_len", log_buf_len_setup);
1005 #ifdef CONFIG_SMP
1006 #define __LOG_CPU_MAX_BUF_LEN (1 << CONFIG_LOG_CPU_MAX_BUF_SHIFT)
1008 static void __init log_buf_add_cpu(void)
1010 unsigned int cpu_extra;
1013 * archs should set up cpu_possible_bits properly with
1014 * set_cpu_possible() after setup_arch() but just in
1015 * case lets ensure this is valid.
1017 if (num_possible_cpus() == 1)
1018 return;
1020 cpu_extra = (num_possible_cpus() - 1) * __LOG_CPU_MAX_BUF_LEN;
1022 /* by default this will only continue through for large > 64 CPUs */
1023 if (cpu_extra <= __LOG_BUF_LEN / 2)
1024 return;
1026 pr_info("log_buf_len individual max cpu contribution: %d bytes\n",
1027 __LOG_CPU_MAX_BUF_LEN);
1028 pr_info("log_buf_len total cpu_extra contributions: %d bytes\n",
1029 cpu_extra);
1030 pr_info("log_buf_len min size: %d bytes\n", __LOG_BUF_LEN);
1032 log_buf_len_update(cpu_extra + __LOG_BUF_LEN);
1034 #else /* !CONFIG_SMP */
1035 static inline void log_buf_add_cpu(void) {}
1036 #endif /* CONFIG_SMP */
1038 void __init setup_log_buf(int early)
1040 unsigned long flags;
1041 char *new_log_buf;
1042 int free;
1044 if (log_buf != __log_buf)
1045 return;
1047 if (!early && !new_log_buf_len)
1048 log_buf_add_cpu();
1050 if (!new_log_buf_len)
1051 return;
1053 if (early) {
1054 new_log_buf =
1055 memblock_virt_alloc(new_log_buf_len, LOG_ALIGN);
1056 } else {
1057 new_log_buf = memblock_virt_alloc_nopanic(new_log_buf_len,
1058 LOG_ALIGN);
1061 if (unlikely(!new_log_buf)) {
1062 pr_err("log_buf_len: %ld bytes not available\n",
1063 new_log_buf_len);
1064 return;
1067 raw_spin_lock_irqsave(&logbuf_lock, flags);
1068 log_buf_len = new_log_buf_len;
1069 log_buf = new_log_buf;
1070 new_log_buf_len = 0;
1071 free = __LOG_BUF_LEN - log_next_idx;
1072 memcpy(log_buf, __log_buf, __LOG_BUF_LEN);
1073 raw_spin_unlock_irqrestore(&logbuf_lock, flags);
1075 pr_info("log_buf_len: %d bytes\n", log_buf_len);
1076 pr_info("early log buf free: %d(%d%%)\n",
1077 free, (free * 100) / __LOG_BUF_LEN);
1080 static bool __read_mostly ignore_loglevel;
1082 static int __init ignore_loglevel_setup(char *str)
1084 ignore_loglevel = true;
1085 pr_info("debug: ignoring loglevel setting.\n");
1087 return 0;
1090 early_param("ignore_loglevel", ignore_loglevel_setup);
1091 module_param(ignore_loglevel, bool, S_IRUGO | S_IWUSR);
1092 MODULE_PARM_DESC(ignore_loglevel,
1093 "ignore loglevel setting (prints all kernel messages to the console)");
1095 static bool suppress_message_printing(int level)
1097 return (level >= console_loglevel && !ignore_loglevel);
1100 #ifdef CONFIG_BOOT_PRINTK_DELAY
1102 static int boot_delay; /* msecs delay after each printk during bootup */
1103 static unsigned long long loops_per_msec; /* based on boot_delay */
1105 static int __init boot_delay_setup(char *str)
1107 unsigned long lpj;
1109 lpj = preset_lpj ? preset_lpj : 1000000; /* some guess */
1110 loops_per_msec = (unsigned long long)lpj / 1000 * HZ;
1112 get_option(&str, &boot_delay);
1113 if (boot_delay > 10 * 1000)
1114 boot_delay = 0;
1116 pr_debug("boot_delay: %u, preset_lpj: %ld, lpj: %lu, "
1117 "HZ: %d, loops_per_msec: %llu\n",
1118 boot_delay, preset_lpj, lpj, HZ, loops_per_msec);
1119 return 0;
1121 early_param("boot_delay", boot_delay_setup);
1123 static void boot_delay_msec(int level)
1125 unsigned long long k;
1126 unsigned long timeout;
1128 if ((boot_delay == 0 || system_state != SYSTEM_BOOTING)
1129 || suppress_message_printing(level)) {
1130 return;
1133 k = (unsigned long long)loops_per_msec * boot_delay;
1135 timeout = jiffies + msecs_to_jiffies(boot_delay);
1136 while (k) {
1137 k--;
1138 cpu_relax();
1140 * use (volatile) jiffies to prevent
1141 * compiler reduction; loop termination via jiffies
1142 * is secondary and may or may not happen.
1144 if (time_after(jiffies, timeout))
1145 break;
1146 touch_nmi_watchdog();
1149 #else
1150 static inline void boot_delay_msec(int level)
1153 #endif
1155 static bool printk_time = IS_ENABLED(CONFIG_PRINTK_TIME);
1156 module_param_named(time, printk_time, bool, S_IRUGO | S_IWUSR);
1158 static size_t print_time(u64 ts, char *buf)
1160 unsigned long rem_nsec;
1162 if (!printk_time)
1163 return 0;
1165 rem_nsec = do_div(ts, 1000000000);
1167 if (!buf)
1168 return snprintf(NULL, 0, "[%5lu.000000] ", (unsigned long)ts);
1170 return sprintf(buf, "[%5lu.%06lu] ",
1171 (unsigned long)ts, rem_nsec / 1000);
1174 static size_t print_prefix(const struct printk_log *msg, bool syslog, char *buf)
1176 size_t len = 0;
1177 unsigned int prefix = (msg->facility << 3) | msg->level;
1179 if (syslog) {
1180 if (buf) {
1181 len += sprintf(buf, "<%u>", prefix);
1182 } else {
1183 len += 3;
1184 if (prefix > 999)
1185 len += 3;
1186 else if (prefix > 99)
1187 len += 2;
1188 else if (prefix > 9)
1189 len++;
1193 len += print_time(msg->ts_nsec, buf ? buf + len : NULL);
1194 return len;
1197 static size_t msg_print_text(const struct printk_log *msg, bool syslog, char *buf, size_t size)
1199 const char *text = log_text(msg);
1200 size_t text_size = msg->text_len;
1201 size_t len = 0;
1203 do {
1204 const char *next = memchr(text, '\n', text_size);
1205 size_t text_len;
1207 if (next) {
1208 text_len = next - text;
1209 next++;
1210 text_size -= next - text;
1211 } else {
1212 text_len = text_size;
1215 if (buf) {
1216 if (print_prefix(msg, syslog, NULL) +
1217 text_len + 1 >= size - len)
1218 break;
1220 len += print_prefix(msg, syslog, buf + len);
1221 memcpy(buf + len, text, text_len);
1222 len += text_len;
1223 buf[len++] = '\n';
1224 } else {
1225 /* SYSLOG_ACTION_* buffer size only calculation */
1226 len += print_prefix(msg, syslog, NULL);
1227 len += text_len;
1228 len++;
1231 text = next;
1232 } while (text);
1234 return len;
1237 static int syslog_print(char __user *buf, int size)
1239 char *text;
1240 struct printk_log *msg;
1241 int len = 0;
1243 text = kmalloc(LOG_LINE_MAX + PREFIX_MAX, GFP_KERNEL);
1244 if (!text)
1245 return -ENOMEM;
1247 while (size > 0) {
1248 size_t n;
1249 size_t skip;
1251 raw_spin_lock_irq(&logbuf_lock);
1252 if (syslog_seq < log_first_seq) {
1253 /* messages are gone, move to first one */
1254 syslog_seq = log_first_seq;
1255 syslog_idx = log_first_idx;
1256 syslog_partial = 0;
1258 if (syslog_seq == log_next_seq) {
1259 raw_spin_unlock_irq(&logbuf_lock);
1260 break;
1263 skip = syslog_partial;
1264 msg = log_from_idx(syslog_idx);
1265 n = msg_print_text(msg, true, text, LOG_LINE_MAX + PREFIX_MAX);
1266 if (n - syslog_partial <= size) {
1267 /* message fits into buffer, move forward */
1268 syslog_idx = log_next(syslog_idx);
1269 syslog_seq++;
1270 n -= syslog_partial;
1271 syslog_partial = 0;
1272 } else if (!len){
1273 /* partial read(), remember position */
1274 n = size;
1275 syslog_partial += n;
1276 } else
1277 n = 0;
1278 raw_spin_unlock_irq(&logbuf_lock);
1280 if (!n)
1281 break;
1283 if (copy_to_user(buf, text + skip, n)) {
1284 if (!len)
1285 len = -EFAULT;
1286 break;
1289 len += n;
1290 size -= n;
1291 buf += n;
1294 kfree(text);
1295 return len;
1298 static int syslog_print_all(char __user *buf, int size, bool clear)
1300 char *text;
1301 int len = 0;
1303 text = kmalloc(LOG_LINE_MAX + PREFIX_MAX, GFP_KERNEL);
1304 if (!text)
1305 return -ENOMEM;
1307 raw_spin_lock_irq(&logbuf_lock);
1308 if (buf) {
1309 u64 next_seq;
1310 u64 seq;
1311 u32 idx;
1314 * Find first record that fits, including all following records,
1315 * into the user-provided buffer for this dump.
1317 seq = clear_seq;
1318 idx = clear_idx;
1319 while (seq < log_next_seq) {
1320 struct printk_log *msg = log_from_idx(idx);
1322 len += msg_print_text(msg, true, NULL, 0);
1323 idx = log_next(idx);
1324 seq++;
1327 /* move first record forward until length fits into the buffer */
1328 seq = clear_seq;
1329 idx = clear_idx;
1330 while (len > size && seq < log_next_seq) {
1331 struct printk_log *msg = log_from_idx(idx);
1333 len -= msg_print_text(msg, true, NULL, 0);
1334 idx = log_next(idx);
1335 seq++;
1338 /* last message fitting into this dump */
1339 next_seq = log_next_seq;
1341 len = 0;
1342 while (len >= 0 && seq < next_seq) {
1343 struct printk_log *msg = log_from_idx(idx);
1344 int textlen;
1346 textlen = msg_print_text(msg, true, text,
1347 LOG_LINE_MAX + PREFIX_MAX);
1348 if (textlen < 0) {
1349 len = textlen;
1350 break;
1352 idx = log_next(idx);
1353 seq++;
1355 raw_spin_unlock_irq(&logbuf_lock);
1356 if (copy_to_user(buf + len, text, textlen))
1357 len = -EFAULT;
1358 else
1359 len += textlen;
1360 raw_spin_lock_irq(&logbuf_lock);
1362 if (seq < log_first_seq) {
1363 /* messages are gone, move to next one */
1364 seq = log_first_seq;
1365 idx = log_first_idx;
1370 if (clear) {
1371 clear_seq = log_next_seq;
1372 clear_idx = log_next_idx;
1374 raw_spin_unlock_irq(&logbuf_lock);
1376 kfree(text);
1377 return len;
1380 int do_syslog(int type, char __user *buf, int len, int source)
1382 bool clear = false;
1383 static int saved_console_loglevel = LOGLEVEL_DEFAULT;
1384 int error;
1386 error = check_syslog_permissions(type, source);
1387 if (error)
1388 goto out;
1390 switch (type) {
1391 case SYSLOG_ACTION_CLOSE: /* Close log */
1392 break;
1393 case SYSLOG_ACTION_OPEN: /* Open log */
1394 break;
1395 case SYSLOG_ACTION_READ: /* Read from log */
1396 error = -EINVAL;
1397 if (!buf || len < 0)
1398 goto out;
1399 error = 0;
1400 if (!len)
1401 goto out;
1402 if (!access_ok(VERIFY_WRITE, buf, len)) {
1403 error = -EFAULT;
1404 goto out;
1406 error = wait_event_interruptible(log_wait,
1407 syslog_seq != log_next_seq);
1408 if (error)
1409 goto out;
1410 error = syslog_print(buf, len);
1411 break;
1412 /* Read/clear last kernel messages */
1413 case SYSLOG_ACTION_READ_CLEAR:
1414 clear = true;
1415 /* FALL THRU */
1416 /* Read last kernel messages */
1417 case SYSLOG_ACTION_READ_ALL:
1418 error = -EINVAL;
1419 if (!buf || len < 0)
1420 goto out;
1421 error = 0;
1422 if (!len)
1423 goto out;
1424 if (!access_ok(VERIFY_WRITE, buf, len)) {
1425 error = -EFAULT;
1426 goto out;
1428 error = syslog_print_all(buf, len, clear);
1429 break;
1430 /* Clear ring buffer */
1431 case SYSLOG_ACTION_CLEAR:
1432 syslog_print_all(NULL, 0, true);
1433 break;
1434 /* Disable logging to console */
1435 case SYSLOG_ACTION_CONSOLE_OFF:
1436 if (saved_console_loglevel == LOGLEVEL_DEFAULT)
1437 saved_console_loglevel = console_loglevel;
1438 console_loglevel = minimum_console_loglevel;
1439 break;
1440 /* Enable logging to console */
1441 case SYSLOG_ACTION_CONSOLE_ON:
1442 if (saved_console_loglevel != LOGLEVEL_DEFAULT) {
1443 console_loglevel = saved_console_loglevel;
1444 saved_console_loglevel = LOGLEVEL_DEFAULT;
1446 break;
1447 /* Set level of messages printed to console */
1448 case SYSLOG_ACTION_CONSOLE_LEVEL:
1449 error = -EINVAL;
1450 if (len < 1 || len > 8)
1451 goto out;
1452 if (len < minimum_console_loglevel)
1453 len = minimum_console_loglevel;
1454 console_loglevel = len;
1455 /* Implicitly re-enable logging to console */
1456 saved_console_loglevel = LOGLEVEL_DEFAULT;
1457 error = 0;
1458 break;
1459 /* Number of chars in the log buffer */
1460 case SYSLOG_ACTION_SIZE_UNREAD:
1461 raw_spin_lock_irq(&logbuf_lock);
1462 if (syslog_seq < log_first_seq) {
1463 /* messages are gone, move to first one */
1464 syslog_seq = log_first_seq;
1465 syslog_idx = log_first_idx;
1466 syslog_partial = 0;
1468 if (source == SYSLOG_FROM_PROC) {
1470 * Short-cut for poll(/"proc/kmsg") which simply checks
1471 * for pending data, not the size; return the count of
1472 * records, not the length.
1474 error = log_next_seq - syslog_seq;
1475 } else {
1476 u64 seq = syslog_seq;
1477 u32 idx = syslog_idx;
1479 error = 0;
1480 while (seq < log_next_seq) {
1481 struct printk_log *msg = log_from_idx(idx);
1483 error += msg_print_text(msg, true, NULL, 0);
1484 idx = log_next(idx);
1485 seq++;
1487 error -= syslog_partial;
1489 raw_spin_unlock_irq(&logbuf_lock);
1490 break;
1491 /* Size of the log buffer */
1492 case SYSLOG_ACTION_SIZE_BUFFER:
1493 error = log_buf_len;
1494 break;
1495 default:
1496 error = -EINVAL;
1497 break;
1499 out:
1500 return error;
1503 SYSCALL_DEFINE3(syslog, int, type, char __user *, buf, int, len)
1505 return do_syslog(type, buf, len, SYSLOG_FROM_READER);
1509 * Call the console drivers, asking them to write out
1510 * log_buf[start] to log_buf[end - 1].
1511 * The console_lock must be held.
1513 static void call_console_drivers(int level,
1514 const char *ext_text, size_t ext_len,
1515 const char *text, size_t len)
1517 struct console *con;
1519 trace_console(text, len);
1521 if (!console_drivers)
1522 return;
1524 for_each_console(con) {
1525 if (exclusive_console && con != exclusive_console)
1526 continue;
1527 if (!(con->flags & CON_ENABLED))
1528 continue;
1529 if (!con->write)
1530 continue;
1531 if (!cpu_online(smp_processor_id()) &&
1532 !(con->flags & CON_ANYTIME))
1533 continue;
1534 if (con->flags & CON_EXTENDED)
1535 con->write(con, ext_text, ext_len);
1536 else
1537 con->write(con, text, len);
1542 * Zap console related locks when oopsing.
1543 * To leave time for slow consoles to print a full oops,
1544 * only zap at most once every 30 seconds.
1546 static void zap_locks(void)
1548 static unsigned long oops_timestamp;
1550 if (time_after_eq(jiffies, oops_timestamp) &&
1551 !time_after(jiffies, oops_timestamp + 30 * HZ))
1552 return;
1554 oops_timestamp = jiffies;
1556 debug_locks_off();
1557 /* If a crash is occurring, make sure we can't deadlock */
1558 raw_spin_lock_init(&logbuf_lock);
1559 /* And make sure that we print immediately */
1560 sema_init(&console_sem, 1);
1563 int printk_delay_msec __read_mostly;
1565 static inline void printk_delay(void)
1567 if (unlikely(printk_delay_msec)) {
1568 int m = printk_delay_msec;
1570 while (m--) {
1571 mdelay(1);
1572 touch_nmi_watchdog();
1578 * Continuation lines are buffered, and not committed to the record buffer
1579 * until the line is complete, or a race forces it. The line fragments
1580 * though, are printed immediately to the consoles to ensure everything has
1581 * reached the console in case of a kernel crash.
1583 static struct cont {
1584 char buf[LOG_LINE_MAX];
1585 size_t len; /* length == 0 means unused buffer */
1586 struct task_struct *owner; /* task of first print*/
1587 u64 ts_nsec; /* time of first print */
1588 u8 level; /* log level of first message */
1589 u8 facility; /* log facility of first message */
1590 enum log_flags flags; /* prefix, newline flags */
1591 } cont;
1593 static void cont_flush(void)
1595 if (cont.len == 0)
1596 return;
1598 log_store(cont.facility, cont.level, cont.flags, cont.ts_nsec,
1599 NULL, 0, cont.buf, cont.len);
1600 cont.len = 0;
1603 static bool cont_add(int facility, int level, enum log_flags flags, const char *text, size_t len)
1606 * If ext consoles are present, flush and skip in-kernel
1607 * continuation. See nr_ext_console_drivers definition. Also, if
1608 * the line gets too long, split it up in separate records.
1610 if (nr_ext_console_drivers || cont.len + len > sizeof(cont.buf)) {
1611 cont_flush();
1612 return false;
1615 if (!cont.len) {
1616 cont.facility = facility;
1617 cont.level = level;
1618 cont.owner = current;
1619 cont.ts_nsec = local_clock();
1620 cont.flags = flags;
1623 memcpy(cont.buf + cont.len, text, len);
1624 cont.len += len;
1626 // The original flags come from the first line,
1627 // but later continuations can add a newline.
1628 if (flags & LOG_NEWLINE) {
1629 cont.flags |= LOG_NEWLINE;
1630 cont_flush();
1633 if (cont.len > (sizeof(cont.buf) * 80) / 100)
1634 cont_flush();
1636 return true;
1639 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)
1642 * If an earlier line was buffered, and we're a continuation
1643 * write from the same process, try to add it to the buffer.
1645 if (cont.len) {
1646 if (cont.owner == current && (lflags & LOG_CONT)) {
1647 if (cont_add(facility, level, lflags, text, text_len))
1648 return text_len;
1650 /* Otherwise, make sure it's flushed */
1651 cont_flush();
1654 /* Skip empty continuation lines that couldn't be added - they just flush */
1655 if (!text_len && (lflags & LOG_CONT))
1656 return 0;
1658 /* If it doesn't end in a newline, try to buffer the current line */
1659 if (!(lflags & LOG_NEWLINE)) {
1660 if (cont_add(facility, level, lflags, text, text_len))
1661 return text_len;
1664 /* Store it in the record log */
1665 return log_store(facility, level, lflags, 0, dict, dictlen, text, text_len);
1668 asmlinkage int vprintk_emit(int facility, int level,
1669 const char *dict, size_t dictlen,
1670 const char *fmt, va_list args)
1672 static bool recursion_bug;
1673 static char textbuf[LOG_LINE_MAX];
1674 char *text = textbuf;
1675 size_t text_len = 0;
1676 enum log_flags lflags = 0;
1677 unsigned long flags;
1678 int this_cpu;
1679 int printed_len = 0;
1680 int nmi_message_lost;
1681 bool in_sched = false;
1682 /* cpu currently holding logbuf_lock in this function */
1683 static unsigned int logbuf_cpu = UINT_MAX;
1685 if (level == LOGLEVEL_SCHED) {
1686 level = LOGLEVEL_DEFAULT;
1687 in_sched = true;
1690 boot_delay_msec(level);
1691 printk_delay();
1693 local_irq_save(flags);
1694 this_cpu = smp_processor_id();
1697 * Ouch, printk recursed into itself!
1699 if (unlikely(logbuf_cpu == this_cpu)) {
1701 * If a crash is occurring during printk() on this CPU,
1702 * then try to get the crash message out but make sure
1703 * we can't deadlock. Otherwise just return to avoid the
1704 * recursion and return - but flag the recursion so that
1705 * it can be printed at the next appropriate moment:
1707 if (!oops_in_progress && !lockdep_recursing(current)) {
1708 recursion_bug = true;
1709 local_irq_restore(flags);
1710 return 0;
1712 zap_locks();
1715 lockdep_off();
1716 /* This stops the holder of console_sem just where we want him */
1717 raw_spin_lock(&logbuf_lock);
1718 logbuf_cpu = this_cpu;
1720 if (unlikely(recursion_bug)) {
1721 static const char recursion_msg[] =
1722 "BUG: recent printk recursion!";
1724 recursion_bug = false;
1725 /* emit KERN_CRIT message */
1726 printed_len += log_store(0, 2, LOG_PREFIX|LOG_NEWLINE, 0,
1727 NULL, 0, recursion_msg,
1728 strlen(recursion_msg));
1731 nmi_message_lost = get_nmi_message_lost();
1732 if (unlikely(nmi_message_lost)) {
1733 text_len = scnprintf(textbuf, sizeof(textbuf),
1734 "BAD LUCK: lost %d message(s) from NMI context!",
1735 nmi_message_lost);
1736 printed_len += log_store(0, 2, LOG_PREFIX|LOG_NEWLINE, 0,
1737 NULL, 0, textbuf, text_len);
1741 * The printf needs to come first; we need the syslog
1742 * prefix which might be passed-in as a parameter.
1744 text_len = vscnprintf(text, sizeof(textbuf), fmt, args);
1746 /* mark and strip a trailing newline */
1747 if (text_len && text[text_len-1] == '\n') {
1748 text_len--;
1749 lflags |= LOG_NEWLINE;
1752 /* strip kernel syslog prefix and extract log level or control flags */
1753 if (facility == 0) {
1754 int kern_level;
1756 while ((kern_level = printk_get_level(text)) != 0) {
1757 switch (kern_level) {
1758 case '0' ... '7':
1759 if (level == LOGLEVEL_DEFAULT)
1760 level = kern_level - '0';
1761 /* fallthrough */
1762 case 'd': /* KERN_DEFAULT */
1763 lflags |= LOG_PREFIX;
1764 break;
1765 case 'c': /* KERN_CONT */
1766 lflags |= LOG_CONT;
1769 text_len -= 2;
1770 text += 2;
1774 if (level == LOGLEVEL_DEFAULT)
1775 level = default_message_loglevel;
1777 if (dict)
1778 lflags |= LOG_PREFIX|LOG_NEWLINE;
1780 printed_len += log_output(facility, level, lflags, dict, dictlen, text, text_len);
1782 logbuf_cpu = UINT_MAX;
1783 raw_spin_unlock(&logbuf_lock);
1784 lockdep_on();
1785 local_irq_restore(flags);
1787 /* If called from the scheduler, we can not call up(). */
1788 if (!in_sched) {
1789 lockdep_off();
1791 * Try to acquire and then immediately release the console
1792 * semaphore. The release will print out buffers and wake up
1793 * /dev/kmsg and syslog() users.
1795 if (console_trylock())
1796 console_unlock();
1797 lockdep_on();
1800 return printed_len;
1802 EXPORT_SYMBOL(vprintk_emit);
1804 asmlinkage int vprintk(const char *fmt, va_list args)
1806 return vprintk_emit(0, LOGLEVEL_DEFAULT, NULL, 0, fmt, args);
1808 EXPORT_SYMBOL(vprintk);
1810 asmlinkage int printk_emit(int facility, int level,
1811 const char *dict, size_t dictlen,
1812 const char *fmt, ...)
1814 va_list args;
1815 int r;
1817 va_start(args, fmt);
1818 r = vprintk_emit(facility, level, dict, dictlen, fmt, args);
1819 va_end(args);
1821 return r;
1823 EXPORT_SYMBOL(printk_emit);
1825 int vprintk_default(const char *fmt, va_list args)
1827 int r;
1829 #ifdef CONFIG_KGDB_KDB
1830 /* Allow to pass printk() to kdb but avoid a recursion. */
1831 if (unlikely(kdb_trap_printk && kdb_printf_cpu < 0)) {
1832 r = vkdb_printf(KDB_MSGSRC_PRINTK, fmt, args);
1833 return r;
1835 #endif
1836 r = vprintk_emit(0, LOGLEVEL_DEFAULT, NULL, 0, fmt, args);
1838 return r;
1840 EXPORT_SYMBOL_GPL(vprintk_default);
1843 * printk - print a kernel message
1844 * @fmt: format string
1846 * This is printk(). It can be called from any context. We want it to work.
1848 * We try to grab the console_lock. If we succeed, it's easy - we log the
1849 * output and call the console drivers. If we fail to get the semaphore, we
1850 * place the output into the log buffer and return. The current holder of
1851 * the console_sem will notice the new output in console_unlock(); and will
1852 * send it to the consoles before releasing the lock.
1854 * One effect of this deferred printing is that code which calls printk() and
1855 * then changes console_loglevel may break. This is because console_loglevel
1856 * is inspected when the actual printing occurs.
1858 * See also:
1859 * printf(3)
1861 * See the vsnprintf() documentation for format string extensions over C99.
1863 asmlinkage __visible int printk(const char *fmt, ...)
1865 va_list args;
1866 int r;
1868 va_start(args, fmt);
1869 r = vprintk_func(fmt, args);
1870 va_end(args);
1872 return r;
1874 EXPORT_SYMBOL(printk);
1876 #else /* CONFIG_PRINTK */
1878 #define LOG_LINE_MAX 0
1879 #define PREFIX_MAX 0
1881 static u64 syslog_seq;
1882 static u32 syslog_idx;
1883 static u64 console_seq;
1884 static u32 console_idx;
1885 static u64 log_first_seq;
1886 static u32 log_first_idx;
1887 static u64 log_next_seq;
1888 static char *log_text(const struct printk_log *msg) { return NULL; }
1889 static char *log_dict(const struct printk_log *msg) { return NULL; }
1890 static struct printk_log *log_from_idx(u32 idx) { return NULL; }
1891 static u32 log_next(u32 idx) { return 0; }
1892 static ssize_t msg_print_ext_header(char *buf, size_t size,
1893 struct printk_log *msg,
1894 u64 seq) { return 0; }
1895 static ssize_t msg_print_ext_body(char *buf, size_t size,
1896 char *dict, size_t dict_len,
1897 char *text, size_t text_len) { return 0; }
1898 static void call_console_drivers(int level,
1899 const char *ext_text, size_t ext_len,
1900 const char *text, size_t len) {}
1901 static size_t msg_print_text(const struct printk_log *msg,
1902 bool syslog, char *buf, size_t size) { return 0; }
1903 static bool suppress_message_printing(int level) { return false; }
1905 /* Still needs to be defined for users */
1906 DEFINE_PER_CPU(printk_func_t, printk_func);
1908 #endif /* CONFIG_PRINTK */
1910 #ifdef CONFIG_EARLY_PRINTK
1911 struct console *early_console;
1913 asmlinkage __visible void early_printk(const char *fmt, ...)
1915 va_list ap;
1916 char buf[512];
1917 int n;
1919 if (!early_console)
1920 return;
1922 va_start(ap, fmt);
1923 n = vscnprintf(buf, sizeof(buf), fmt, ap);
1924 va_end(ap);
1926 early_console->write(early_console, buf, n);
1928 #endif
1930 static int __add_preferred_console(char *name, int idx, char *options,
1931 char *brl_options)
1933 struct console_cmdline *c;
1934 int i;
1937 * See if this tty is not yet registered, and
1938 * if we have a slot free.
1940 for (i = 0, c = console_cmdline;
1941 i < MAX_CMDLINECONSOLES && c->name[0];
1942 i++, c++) {
1943 if (strcmp(c->name, name) == 0 && c->index == idx) {
1944 if (!brl_options)
1945 selected_console = i;
1946 return 0;
1949 if (i == MAX_CMDLINECONSOLES)
1950 return -E2BIG;
1951 if (!brl_options)
1952 selected_console = i;
1953 strlcpy(c->name, name, sizeof(c->name));
1954 c->options = options;
1955 braille_set_options(c, brl_options);
1957 c->index = idx;
1958 return 0;
1961 * Set up a console. Called via do_early_param() in init/main.c
1962 * for each "console=" parameter in the boot command line.
1964 static int __init console_setup(char *str)
1966 char buf[sizeof(console_cmdline[0].name) + 4]; /* 4 for "ttyS" */
1967 char *s, *options, *brl_options = NULL;
1968 int idx;
1970 if (_braille_console_setup(&str, &brl_options))
1971 return 1;
1974 * Decode str into name, index, options.
1976 if (str[0] >= '0' && str[0] <= '9') {
1977 strcpy(buf, "ttyS");
1978 strncpy(buf + 4, str, sizeof(buf) - 5);
1979 } else {
1980 strncpy(buf, str, sizeof(buf) - 1);
1982 buf[sizeof(buf) - 1] = 0;
1983 options = strchr(str, ',');
1984 if (options)
1985 *(options++) = 0;
1986 #ifdef __sparc__
1987 if (!strcmp(str, "ttya"))
1988 strcpy(buf, "ttyS0");
1989 if (!strcmp(str, "ttyb"))
1990 strcpy(buf, "ttyS1");
1991 #endif
1992 for (s = buf; *s; s++)
1993 if (isdigit(*s) || *s == ',')
1994 break;
1995 idx = simple_strtoul(s, NULL, 10);
1996 *s = 0;
1998 __add_preferred_console(buf, idx, options, brl_options);
1999 console_set_on_cmdline = 1;
2000 return 1;
2002 __setup("console=", console_setup);
2005 * add_preferred_console - add a device to the list of preferred consoles.
2006 * @name: device name
2007 * @idx: device index
2008 * @options: options for this console
2010 * The last preferred console added will be used for kernel messages
2011 * and stdin/out/err for init. Normally this is used by console_setup
2012 * above to handle user-supplied console arguments; however it can also
2013 * be used by arch-specific code either to override the user or more
2014 * commonly to provide a default console (ie from PROM variables) when
2015 * the user has not supplied one.
2017 int add_preferred_console(char *name, int idx, char *options)
2019 return __add_preferred_console(name, idx, options, NULL);
2022 bool console_suspend_enabled = true;
2023 EXPORT_SYMBOL(console_suspend_enabled);
2025 static int __init console_suspend_disable(char *str)
2027 console_suspend_enabled = false;
2028 return 1;
2030 __setup("no_console_suspend", console_suspend_disable);
2031 module_param_named(console_suspend, console_suspend_enabled,
2032 bool, S_IRUGO | S_IWUSR);
2033 MODULE_PARM_DESC(console_suspend, "suspend console during suspend"
2034 " and hibernate operations");
2037 * suspend_console - suspend the console subsystem
2039 * This disables printk() while we go into suspend states
2041 void suspend_console(void)
2043 if (!console_suspend_enabled)
2044 return;
2045 printk("Suspending console(s) (use no_console_suspend to debug)\n");
2046 console_lock();
2047 console_suspended = 1;
2048 up_console_sem();
2051 void resume_console(void)
2053 if (!console_suspend_enabled)
2054 return;
2055 down_console_sem();
2056 console_suspended = 0;
2057 console_unlock();
2061 * console_cpu_notify - print deferred console messages after CPU hotplug
2062 * @cpu: unused
2064 * If printk() is called from a CPU that is not online yet, the messages
2065 * will be spooled but will not show up on the console. This function is
2066 * called when a new CPU comes online (or fails to come up), and ensures
2067 * that any such output gets printed.
2069 static int console_cpu_notify(unsigned int cpu)
2071 if (!cpuhp_tasks_frozen) {
2072 console_lock();
2073 console_unlock();
2075 return 0;
2079 * console_lock - lock the console system for exclusive use.
2081 * Acquires a lock which guarantees that the caller has
2082 * exclusive access to the console system and the console_drivers list.
2084 * Can sleep, returns nothing.
2086 void console_lock(void)
2088 might_sleep();
2090 down_console_sem();
2091 if (console_suspended)
2092 return;
2093 console_locked = 1;
2094 console_may_schedule = 1;
2096 EXPORT_SYMBOL(console_lock);
2099 * console_trylock - try to lock the console system for exclusive use.
2101 * Try to acquire a lock which guarantees that the caller has exclusive
2102 * access to the console system and the console_drivers list.
2104 * returns 1 on success, and 0 on failure to acquire the lock.
2106 int console_trylock(void)
2108 if (down_trylock_console_sem())
2109 return 0;
2110 if (console_suspended) {
2111 up_console_sem();
2112 return 0;
2114 console_locked = 1;
2116 * When PREEMPT_COUNT disabled we can't reliably detect if it's
2117 * safe to schedule (e.g. calling printk while holding a spin_lock),
2118 * because preempt_disable()/preempt_enable() are just barriers there
2119 * and preempt_count() is always 0.
2121 * RCU read sections have a separate preemption counter when
2122 * PREEMPT_RCU enabled thus we must take extra care and check
2123 * rcu_preempt_depth(), otherwise RCU read sections modify
2124 * preempt_count().
2126 console_may_schedule = !oops_in_progress &&
2127 preemptible() &&
2128 !rcu_preempt_depth();
2129 return 1;
2131 EXPORT_SYMBOL(console_trylock);
2133 int is_console_locked(void)
2135 return console_locked;
2139 * Check if we have any console that is capable of printing while cpu is
2140 * booting or shutting down. Requires console_sem.
2142 static int have_callable_console(void)
2144 struct console *con;
2146 for_each_console(con)
2147 if ((con->flags & CON_ENABLED) &&
2148 (con->flags & CON_ANYTIME))
2149 return 1;
2151 return 0;
2155 * Can we actually use the console at this time on this cpu?
2157 * Console drivers may assume that per-cpu resources have been allocated. So
2158 * unless they're explicitly marked as being able to cope (CON_ANYTIME) don't
2159 * call them until this CPU is officially up.
2161 static inline int can_use_console(void)
2163 return cpu_online(raw_smp_processor_id()) || have_callable_console();
2167 * console_unlock - unlock the console system
2169 * Releases the console_lock which the caller holds on the console system
2170 * and the console driver list.
2172 * While the console_lock was held, console output may have been buffered
2173 * by printk(). If this is the case, console_unlock(); emits
2174 * the output prior to releasing the lock.
2176 * If there is output waiting, we wake /dev/kmsg and syslog() users.
2178 * console_unlock(); may be called from any context.
2180 void console_unlock(void)
2182 static char ext_text[CONSOLE_EXT_LOG_MAX];
2183 static char text[LOG_LINE_MAX + PREFIX_MAX];
2184 static u64 seen_seq;
2185 unsigned long flags;
2186 bool wake_klogd = false;
2187 bool do_cond_resched, retry;
2189 if (console_suspended) {
2190 up_console_sem();
2191 return;
2195 * Console drivers are called under logbuf_lock, so
2196 * @console_may_schedule should be cleared before; however, we may
2197 * end up dumping a lot of lines, for example, if called from
2198 * console registration path, and should invoke cond_resched()
2199 * between lines if allowable. Not doing so can cause a very long
2200 * scheduling stall on a slow console leading to RCU stall and
2201 * softlockup warnings which exacerbate the issue with more
2202 * messages practically incapacitating the system.
2204 do_cond_resched = console_may_schedule;
2205 console_may_schedule = 0;
2207 again:
2209 * We released the console_sem lock, so we need to recheck if
2210 * cpu is online and (if not) is there at least one CON_ANYTIME
2211 * console.
2213 if (!can_use_console()) {
2214 console_locked = 0;
2215 up_console_sem();
2216 return;
2219 for (;;) {
2220 struct printk_log *msg;
2221 size_t ext_len = 0;
2222 size_t len;
2223 int level;
2225 raw_spin_lock_irqsave(&logbuf_lock, flags);
2226 if (seen_seq != log_next_seq) {
2227 wake_klogd = true;
2228 seen_seq = log_next_seq;
2231 if (console_seq < log_first_seq) {
2232 len = sprintf(text, "** %u printk messages dropped ** ",
2233 (unsigned)(log_first_seq - console_seq));
2235 /* messages are gone, move to first one */
2236 console_seq = log_first_seq;
2237 console_idx = log_first_idx;
2238 } else {
2239 len = 0;
2241 skip:
2242 if (console_seq == log_next_seq)
2243 break;
2245 msg = log_from_idx(console_idx);
2246 level = msg->level;
2247 if (suppress_message_printing(level)) {
2249 * Skip record we have buffered and already printed
2250 * directly to the console when we received it, and
2251 * record that has level above the console loglevel.
2253 console_idx = log_next(console_idx);
2254 console_seq++;
2255 goto skip;
2258 len += msg_print_text(msg, false, text + len, sizeof(text) - len);
2259 if (nr_ext_console_drivers) {
2260 ext_len = msg_print_ext_header(ext_text,
2261 sizeof(ext_text),
2262 msg, console_seq);
2263 ext_len += msg_print_ext_body(ext_text + ext_len,
2264 sizeof(ext_text) - ext_len,
2265 log_dict(msg), msg->dict_len,
2266 log_text(msg), msg->text_len);
2268 console_idx = log_next(console_idx);
2269 console_seq++;
2270 raw_spin_unlock(&logbuf_lock);
2272 stop_critical_timings(); /* don't trace print latency */
2273 call_console_drivers(level, ext_text, ext_len, text, len);
2274 start_critical_timings();
2275 local_irq_restore(flags);
2277 if (do_cond_resched)
2278 cond_resched();
2280 console_locked = 0;
2282 /* Release the exclusive_console once it is used */
2283 if (unlikely(exclusive_console))
2284 exclusive_console = NULL;
2286 raw_spin_unlock(&logbuf_lock);
2288 up_console_sem();
2291 * Someone could have filled up the buffer again, so re-check if there's
2292 * something to flush. In case we cannot trylock the console_sem again,
2293 * there's a new owner and the console_unlock() from them will do the
2294 * flush, no worries.
2296 raw_spin_lock(&logbuf_lock);
2297 retry = console_seq != log_next_seq;
2298 raw_spin_unlock_irqrestore(&logbuf_lock, flags);
2300 if (retry && console_trylock())
2301 goto again;
2303 if (wake_klogd)
2304 wake_up_klogd();
2306 EXPORT_SYMBOL(console_unlock);
2309 * console_conditional_schedule - yield the CPU if required
2311 * If the console code is currently allowed to sleep, and
2312 * if this CPU should yield the CPU to another task, do
2313 * so here.
2315 * Must be called within console_lock();.
2317 void __sched console_conditional_schedule(void)
2319 if (console_may_schedule)
2320 cond_resched();
2322 EXPORT_SYMBOL(console_conditional_schedule);
2324 void console_unblank(void)
2326 struct console *c;
2329 * console_unblank can no longer be called in interrupt context unless
2330 * oops_in_progress is set to 1..
2332 if (oops_in_progress) {
2333 if (down_trylock_console_sem() != 0)
2334 return;
2335 } else
2336 console_lock();
2338 console_locked = 1;
2339 console_may_schedule = 0;
2340 for_each_console(c)
2341 if ((c->flags & CON_ENABLED) && c->unblank)
2342 c->unblank();
2343 console_unlock();
2347 * console_flush_on_panic - flush console content on panic
2349 * Immediately output all pending messages no matter what.
2351 void console_flush_on_panic(void)
2354 * If someone else is holding the console lock, trylock will fail
2355 * and may_schedule may be set. Ignore and proceed to unlock so
2356 * that messages are flushed out. As this can be called from any
2357 * context and we don't want to get preempted while flushing,
2358 * ensure may_schedule is cleared.
2360 console_trylock();
2361 console_may_schedule = 0;
2362 console_unlock();
2366 * Return the console tty driver structure and its associated index
2368 struct tty_driver *console_device(int *index)
2370 struct console *c;
2371 struct tty_driver *driver = NULL;
2373 console_lock();
2374 for_each_console(c) {
2375 if (!c->device)
2376 continue;
2377 driver = c->device(c, index);
2378 if (driver)
2379 break;
2381 console_unlock();
2382 return driver;
2386 * Prevent further output on the passed console device so that (for example)
2387 * serial drivers can disable console output before suspending a port, and can
2388 * re-enable output afterwards.
2390 void console_stop(struct console *console)
2392 console_lock();
2393 console->flags &= ~CON_ENABLED;
2394 console_unlock();
2396 EXPORT_SYMBOL(console_stop);
2398 void console_start(struct console *console)
2400 console_lock();
2401 console->flags |= CON_ENABLED;
2402 console_unlock();
2404 EXPORT_SYMBOL(console_start);
2406 static int __read_mostly keep_bootcon;
2408 static int __init keep_bootcon_setup(char *str)
2410 keep_bootcon = 1;
2411 pr_info("debug: skip boot console de-registration.\n");
2413 return 0;
2416 early_param("keep_bootcon", keep_bootcon_setup);
2419 * The console driver calls this routine during kernel initialization
2420 * to register the console printing procedure with printk() and to
2421 * print any messages that were printed by the kernel before the
2422 * console driver was initialized.
2424 * This can happen pretty early during the boot process (because of
2425 * early_printk) - sometimes before setup_arch() completes - be careful
2426 * of what kernel features are used - they may not be initialised yet.
2428 * There are two types of consoles - bootconsoles (early_printk) and
2429 * "real" consoles (everything which is not a bootconsole) which are
2430 * handled differently.
2431 * - Any number of bootconsoles can be registered at any time.
2432 * - As soon as a "real" console is registered, all bootconsoles
2433 * will be unregistered automatically.
2434 * - Once a "real" console is registered, any attempt to register a
2435 * bootconsoles will be rejected
2437 void register_console(struct console *newcon)
2439 int i;
2440 unsigned long flags;
2441 struct console *bcon = NULL;
2442 struct console_cmdline *c;
2444 if (console_drivers)
2445 for_each_console(bcon)
2446 if (WARN(bcon == newcon,
2447 "console '%s%d' already registered\n",
2448 bcon->name, bcon->index))
2449 return;
2452 * before we register a new CON_BOOT console, make sure we don't
2453 * already have a valid console
2455 if (console_drivers && newcon->flags & CON_BOOT) {
2456 /* find the last or real console */
2457 for_each_console(bcon) {
2458 if (!(bcon->flags & CON_BOOT)) {
2459 pr_info("Too late to register bootconsole %s%d\n",
2460 newcon->name, newcon->index);
2461 return;
2466 if (console_drivers && console_drivers->flags & CON_BOOT)
2467 bcon = console_drivers;
2469 if (preferred_console < 0 || bcon || !console_drivers)
2470 preferred_console = selected_console;
2473 * See if we want to use this console driver. If we
2474 * didn't select a console we take the first one
2475 * that registers here.
2477 if (preferred_console < 0) {
2478 if (newcon->index < 0)
2479 newcon->index = 0;
2480 if (newcon->setup == NULL ||
2481 newcon->setup(newcon, NULL) == 0) {
2482 newcon->flags |= CON_ENABLED;
2483 if (newcon->device) {
2484 newcon->flags |= CON_CONSDEV;
2485 preferred_console = 0;
2491 * See if this console matches one we selected on
2492 * the command line.
2494 for (i = 0, c = console_cmdline;
2495 i < MAX_CMDLINECONSOLES && c->name[0];
2496 i++, c++) {
2497 if (!newcon->match ||
2498 newcon->match(newcon, c->name, c->index, c->options) != 0) {
2499 /* default matching */
2500 BUILD_BUG_ON(sizeof(c->name) != sizeof(newcon->name));
2501 if (strcmp(c->name, newcon->name) != 0)
2502 continue;
2503 if (newcon->index >= 0 &&
2504 newcon->index != c->index)
2505 continue;
2506 if (newcon->index < 0)
2507 newcon->index = c->index;
2509 if (_braille_register_console(newcon, c))
2510 return;
2512 if (newcon->setup &&
2513 newcon->setup(newcon, c->options) != 0)
2514 break;
2517 newcon->flags |= CON_ENABLED;
2518 if (i == selected_console) {
2519 newcon->flags |= CON_CONSDEV;
2520 preferred_console = selected_console;
2522 break;
2525 if (!(newcon->flags & CON_ENABLED))
2526 return;
2529 * If we have a bootconsole, and are switching to a real console,
2530 * don't print everything out again, since when the boot console, and
2531 * the real console are the same physical device, it's annoying to
2532 * see the beginning boot messages twice
2534 if (bcon && ((newcon->flags & (CON_CONSDEV | CON_BOOT)) == CON_CONSDEV))
2535 newcon->flags &= ~CON_PRINTBUFFER;
2538 * Put this console in the list - keep the
2539 * preferred driver at the head of the list.
2541 console_lock();
2542 if ((newcon->flags & CON_CONSDEV) || console_drivers == NULL) {
2543 newcon->next = console_drivers;
2544 console_drivers = newcon;
2545 if (newcon->next)
2546 newcon->next->flags &= ~CON_CONSDEV;
2547 } else {
2548 newcon->next = console_drivers->next;
2549 console_drivers->next = newcon;
2552 if (newcon->flags & CON_EXTENDED)
2553 if (!nr_ext_console_drivers++)
2554 pr_info("printk: continuation disabled due to ext consoles, expect more fragments in /dev/kmsg\n");
2556 if (newcon->flags & CON_PRINTBUFFER) {
2558 * console_unlock(); will print out the buffered messages
2559 * for us.
2561 raw_spin_lock_irqsave(&logbuf_lock, flags);
2562 console_seq = syslog_seq;
2563 console_idx = syslog_idx;
2564 raw_spin_unlock_irqrestore(&logbuf_lock, flags);
2566 * We're about to replay the log buffer. Only do this to the
2567 * just-registered console to avoid excessive message spam to
2568 * the already-registered consoles.
2570 exclusive_console = newcon;
2572 console_unlock();
2573 console_sysfs_notify();
2576 * By unregistering the bootconsoles after we enable the real console
2577 * we get the "console xxx enabled" message on all the consoles -
2578 * boot consoles, real consoles, etc - this is to ensure that end
2579 * users know there might be something in the kernel's log buffer that
2580 * went to the bootconsole (that they do not see on the real console)
2582 pr_info("%sconsole [%s%d] enabled\n",
2583 (newcon->flags & CON_BOOT) ? "boot" : "" ,
2584 newcon->name, newcon->index);
2585 if (bcon &&
2586 ((newcon->flags & (CON_CONSDEV | CON_BOOT)) == CON_CONSDEV) &&
2587 !keep_bootcon) {
2588 /* We need to iterate through all boot consoles, to make
2589 * sure we print everything out, before we unregister them.
2591 for_each_console(bcon)
2592 if (bcon->flags & CON_BOOT)
2593 unregister_console(bcon);
2596 EXPORT_SYMBOL(register_console);
2598 int unregister_console(struct console *console)
2600 struct console *a, *b;
2601 int res;
2603 pr_info("%sconsole [%s%d] disabled\n",
2604 (console->flags & CON_BOOT) ? "boot" : "" ,
2605 console->name, console->index);
2607 res = _braille_unregister_console(console);
2608 if (res)
2609 return res;
2611 res = 1;
2612 console_lock();
2613 if (console_drivers == console) {
2614 console_drivers=console->next;
2615 res = 0;
2616 } else if (console_drivers) {
2617 for (a=console_drivers->next, b=console_drivers ;
2618 a; b=a, a=b->next) {
2619 if (a == console) {
2620 b->next = a->next;
2621 res = 0;
2622 break;
2627 if (!res && (console->flags & CON_EXTENDED))
2628 nr_ext_console_drivers--;
2631 * If this isn't the last console and it has CON_CONSDEV set, we
2632 * need to set it on the next preferred console.
2634 if (console_drivers != NULL && console->flags & CON_CONSDEV)
2635 console_drivers->flags |= CON_CONSDEV;
2637 console->flags &= ~CON_ENABLED;
2638 console_unlock();
2639 console_sysfs_notify();
2640 return res;
2642 EXPORT_SYMBOL(unregister_console);
2645 * Some boot consoles access data that is in the init section and which will
2646 * be discarded after the initcalls have been run. To make sure that no code
2647 * will access this data, unregister the boot consoles in a late initcall.
2649 * If for some reason, such as deferred probe or the driver being a loadable
2650 * module, the real console hasn't registered yet at this point, there will
2651 * be a brief interval in which no messages are logged to the console, which
2652 * makes it difficult to diagnose problems that occur during this time.
2654 * To mitigate this problem somewhat, only unregister consoles whose memory
2655 * intersects with the init section. Note that code exists elsewhere to get
2656 * rid of the boot console as soon as the proper console shows up, so there
2657 * won't be side-effects from postponing the removal.
2659 static int __init printk_late_init(void)
2661 struct console *con;
2662 int ret;
2664 for_each_console(con) {
2665 if (!keep_bootcon && con->flags & CON_BOOT) {
2667 * Make sure to unregister boot consoles whose data
2668 * resides in the init section before the init section
2669 * is discarded. Boot consoles whose data will stick
2670 * around will automatically be unregistered when the
2671 * proper console replaces them.
2673 if (init_section_intersects(con, sizeof(*con)))
2674 unregister_console(con);
2677 ret = cpuhp_setup_state_nocalls(CPUHP_PRINTK_DEAD, "printk:dead", NULL,
2678 console_cpu_notify);
2679 WARN_ON(ret < 0);
2680 ret = cpuhp_setup_state_nocalls(CPUHP_AP_ONLINE_DYN, "printk:online",
2681 console_cpu_notify, NULL);
2682 WARN_ON(ret < 0);
2683 return 0;
2685 late_initcall(printk_late_init);
2687 #if defined CONFIG_PRINTK
2689 * Delayed printk version, for scheduler-internal messages:
2691 #define PRINTK_PENDING_WAKEUP 0x01
2692 #define PRINTK_PENDING_OUTPUT 0x02
2694 static DEFINE_PER_CPU(int, printk_pending);
2696 static void wake_up_klogd_work_func(struct irq_work *irq_work)
2698 int pending = __this_cpu_xchg(printk_pending, 0);
2700 if (pending & PRINTK_PENDING_OUTPUT) {
2701 /* If trylock fails, someone else is doing the printing */
2702 if (console_trylock())
2703 console_unlock();
2706 if (pending & PRINTK_PENDING_WAKEUP)
2707 wake_up_interruptible(&log_wait);
2710 static DEFINE_PER_CPU(struct irq_work, wake_up_klogd_work) = {
2711 .func = wake_up_klogd_work_func,
2712 .flags = IRQ_WORK_LAZY,
2715 void wake_up_klogd(void)
2717 preempt_disable();
2718 if (waitqueue_active(&log_wait)) {
2719 this_cpu_or(printk_pending, PRINTK_PENDING_WAKEUP);
2720 irq_work_queue(this_cpu_ptr(&wake_up_klogd_work));
2722 preempt_enable();
2725 int printk_deferred(const char *fmt, ...)
2727 va_list args;
2728 int r;
2730 preempt_disable();
2731 va_start(args, fmt);
2732 r = vprintk_emit(0, LOGLEVEL_SCHED, NULL, 0, fmt, args);
2733 va_end(args);
2735 __this_cpu_or(printk_pending, PRINTK_PENDING_OUTPUT);
2736 irq_work_queue(this_cpu_ptr(&wake_up_klogd_work));
2737 preempt_enable();
2739 return r;
2743 * printk rate limiting, lifted from the networking subsystem.
2745 * This enforces a rate limit: not more than 10 kernel messages
2746 * every 5s to make a denial-of-service attack impossible.
2748 DEFINE_RATELIMIT_STATE(printk_ratelimit_state, 5 * HZ, 10);
2750 int __printk_ratelimit(const char *func)
2752 return ___ratelimit(&printk_ratelimit_state, func);
2754 EXPORT_SYMBOL(__printk_ratelimit);
2757 * printk_timed_ratelimit - caller-controlled printk ratelimiting
2758 * @caller_jiffies: pointer to caller's state
2759 * @interval_msecs: minimum interval between prints
2761 * printk_timed_ratelimit() returns true if more than @interval_msecs
2762 * milliseconds have elapsed since the last time printk_timed_ratelimit()
2763 * returned true.
2765 bool printk_timed_ratelimit(unsigned long *caller_jiffies,
2766 unsigned int interval_msecs)
2768 unsigned long elapsed = jiffies - *caller_jiffies;
2770 if (*caller_jiffies && elapsed <= msecs_to_jiffies(interval_msecs))
2771 return false;
2773 *caller_jiffies = jiffies;
2774 return true;
2776 EXPORT_SYMBOL(printk_timed_ratelimit);
2778 static DEFINE_SPINLOCK(dump_list_lock);
2779 static LIST_HEAD(dump_list);
2782 * kmsg_dump_register - register a kernel log dumper.
2783 * @dumper: pointer to the kmsg_dumper structure
2785 * Adds a kernel log dumper to the system. The dump callback in the
2786 * structure will be called when the kernel oopses or panics and must be
2787 * set. Returns zero on success and %-EINVAL or %-EBUSY otherwise.
2789 int kmsg_dump_register(struct kmsg_dumper *dumper)
2791 unsigned long flags;
2792 int err = -EBUSY;
2794 /* The dump callback needs to be set */
2795 if (!dumper->dump)
2796 return -EINVAL;
2798 spin_lock_irqsave(&dump_list_lock, flags);
2799 /* Don't allow registering multiple times */
2800 if (!dumper->registered) {
2801 dumper->registered = 1;
2802 list_add_tail_rcu(&dumper->list, &dump_list);
2803 err = 0;
2805 spin_unlock_irqrestore(&dump_list_lock, flags);
2807 return err;
2809 EXPORT_SYMBOL_GPL(kmsg_dump_register);
2812 * kmsg_dump_unregister - unregister a kmsg dumper.
2813 * @dumper: pointer to the kmsg_dumper structure
2815 * Removes a dump device from the system. Returns zero on success and
2816 * %-EINVAL otherwise.
2818 int kmsg_dump_unregister(struct kmsg_dumper *dumper)
2820 unsigned long flags;
2821 int err = -EINVAL;
2823 spin_lock_irqsave(&dump_list_lock, flags);
2824 if (dumper->registered) {
2825 dumper->registered = 0;
2826 list_del_rcu(&dumper->list);
2827 err = 0;
2829 spin_unlock_irqrestore(&dump_list_lock, flags);
2830 synchronize_rcu();
2832 return err;
2834 EXPORT_SYMBOL_GPL(kmsg_dump_unregister);
2836 static bool always_kmsg_dump;
2837 module_param_named(always_kmsg_dump, always_kmsg_dump, bool, S_IRUGO | S_IWUSR);
2840 * kmsg_dump - dump kernel log to kernel message dumpers.
2841 * @reason: the reason (oops, panic etc) for dumping
2843 * Call each of the registered dumper's dump() callback, which can
2844 * retrieve the kmsg records with kmsg_dump_get_line() or
2845 * kmsg_dump_get_buffer().
2847 void kmsg_dump(enum kmsg_dump_reason reason)
2849 struct kmsg_dumper *dumper;
2850 unsigned long flags;
2852 if ((reason > KMSG_DUMP_OOPS) && !always_kmsg_dump)
2853 return;
2855 rcu_read_lock();
2856 list_for_each_entry_rcu(dumper, &dump_list, list) {
2857 if (dumper->max_reason && reason > dumper->max_reason)
2858 continue;
2860 /* initialize iterator with data about the stored records */
2861 dumper->active = true;
2863 raw_spin_lock_irqsave(&logbuf_lock, flags);
2864 dumper->cur_seq = clear_seq;
2865 dumper->cur_idx = clear_idx;
2866 dumper->next_seq = log_next_seq;
2867 dumper->next_idx = log_next_idx;
2868 raw_spin_unlock_irqrestore(&logbuf_lock, flags);
2870 /* invoke dumper which will iterate over records */
2871 dumper->dump(dumper, reason);
2873 /* reset iterator */
2874 dumper->active = false;
2876 rcu_read_unlock();
2880 * kmsg_dump_get_line_nolock - retrieve one kmsg log line (unlocked version)
2881 * @dumper: registered kmsg dumper
2882 * @syslog: include the "<4>" prefixes
2883 * @line: buffer to copy the line to
2884 * @size: maximum size of the buffer
2885 * @len: length of line placed into buffer
2887 * Start at the beginning of the kmsg buffer, with the oldest kmsg
2888 * record, and copy one record into the provided buffer.
2890 * Consecutive calls will return the next available record moving
2891 * towards the end of the buffer with the youngest messages.
2893 * A return value of FALSE indicates that there are no more records to
2894 * read.
2896 * The function is similar to kmsg_dump_get_line(), but grabs no locks.
2898 bool kmsg_dump_get_line_nolock(struct kmsg_dumper *dumper, bool syslog,
2899 char *line, size_t size, size_t *len)
2901 struct printk_log *msg;
2902 size_t l = 0;
2903 bool ret = false;
2905 if (!dumper->active)
2906 goto out;
2908 if (dumper->cur_seq < log_first_seq) {
2909 /* messages are gone, move to first available one */
2910 dumper->cur_seq = log_first_seq;
2911 dumper->cur_idx = log_first_idx;
2914 /* last entry */
2915 if (dumper->cur_seq >= log_next_seq)
2916 goto out;
2918 msg = log_from_idx(dumper->cur_idx);
2919 l = msg_print_text(msg, syslog, line, size);
2921 dumper->cur_idx = log_next(dumper->cur_idx);
2922 dumper->cur_seq++;
2923 ret = true;
2924 out:
2925 if (len)
2926 *len = l;
2927 return ret;
2931 * kmsg_dump_get_line - retrieve one kmsg log line
2932 * @dumper: registered kmsg dumper
2933 * @syslog: include the "<4>" prefixes
2934 * @line: buffer to copy the line to
2935 * @size: maximum size of the buffer
2936 * @len: length of line placed into buffer
2938 * Start at the beginning of the kmsg buffer, with the oldest kmsg
2939 * record, and copy one record into the provided buffer.
2941 * Consecutive calls will return the next available record moving
2942 * towards the end of the buffer with the youngest messages.
2944 * A return value of FALSE indicates that there are no more records to
2945 * read.
2947 bool kmsg_dump_get_line(struct kmsg_dumper *dumper, bool syslog,
2948 char *line, size_t size, size_t *len)
2950 unsigned long flags;
2951 bool ret;
2953 raw_spin_lock_irqsave(&logbuf_lock, flags);
2954 ret = kmsg_dump_get_line_nolock(dumper, syslog, line, size, len);
2955 raw_spin_unlock_irqrestore(&logbuf_lock, flags);
2957 return ret;
2959 EXPORT_SYMBOL_GPL(kmsg_dump_get_line);
2962 * kmsg_dump_get_buffer - copy kmsg log lines
2963 * @dumper: registered kmsg dumper
2964 * @syslog: include the "<4>" prefixes
2965 * @buf: buffer to copy the line to
2966 * @size: maximum size of the buffer
2967 * @len: length of line placed into buffer
2969 * Start at the end of the kmsg buffer and fill the provided buffer
2970 * with as many of the the *youngest* kmsg records that fit into it.
2971 * If the buffer is large enough, all available kmsg records will be
2972 * copied with a single call.
2974 * Consecutive calls will fill the buffer with the next block of
2975 * available older records, not including the earlier retrieved ones.
2977 * A return value of FALSE indicates that there are no more records to
2978 * read.
2980 bool kmsg_dump_get_buffer(struct kmsg_dumper *dumper, bool syslog,
2981 char *buf, size_t size, size_t *len)
2983 unsigned long flags;
2984 u64 seq;
2985 u32 idx;
2986 u64 next_seq;
2987 u32 next_idx;
2988 size_t l = 0;
2989 bool ret = false;
2991 if (!dumper->active)
2992 goto out;
2994 raw_spin_lock_irqsave(&logbuf_lock, flags);
2995 if (dumper->cur_seq < log_first_seq) {
2996 /* messages are gone, move to first available one */
2997 dumper->cur_seq = log_first_seq;
2998 dumper->cur_idx = log_first_idx;
3001 /* last entry */
3002 if (dumper->cur_seq >= dumper->next_seq) {
3003 raw_spin_unlock_irqrestore(&logbuf_lock, flags);
3004 goto out;
3007 /* calculate length of entire buffer */
3008 seq = dumper->cur_seq;
3009 idx = dumper->cur_idx;
3010 while (seq < dumper->next_seq) {
3011 struct printk_log *msg = log_from_idx(idx);
3013 l += msg_print_text(msg, true, NULL, 0);
3014 idx = log_next(idx);
3015 seq++;
3018 /* move first record forward until length fits into the buffer */
3019 seq = dumper->cur_seq;
3020 idx = dumper->cur_idx;
3021 while (l > size && seq < dumper->next_seq) {
3022 struct printk_log *msg = log_from_idx(idx);
3024 l -= msg_print_text(msg, true, NULL, 0);
3025 idx = log_next(idx);
3026 seq++;
3029 /* last message in next interation */
3030 next_seq = seq;
3031 next_idx = idx;
3033 l = 0;
3034 while (seq < dumper->next_seq) {
3035 struct printk_log *msg = log_from_idx(idx);
3037 l += msg_print_text(msg, syslog, buf + l, size - l);
3038 idx = log_next(idx);
3039 seq++;
3042 dumper->next_seq = next_seq;
3043 dumper->next_idx = next_idx;
3044 ret = true;
3045 raw_spin_unlock_irqrestore(&logbuf_lock, flags);
3046 out:
3047 if (len)
3048 *len = l;
3049 return ret;
3051 EXPORT_SYMBOL_GPL(kmsg_dump_get_buffer);
3054 * kmsg_dump_rewind_nolock - reset the interator (unlocked version)
3055 * @dumper: registered kmsg dumper
3057 * Reset the dumper's iterator so that kmsg_dump_get_line() and
3058 * kmsg_dump_get_buffer() can be called again and used multiple
3059 * times within the same dumper.dump() callback.
3061 * The function is similar to kmsg_dump_rewind(), but grabs no locks.
3063 void kmsg_dump_rewind_nolock(struct kmsg_dumper *dumper)
3065 dumper->cur_seq = clear_seq;
3066 dumper->cur_idx = clear_idx;
3067 dumper->next_seq = log_next_seq;
3068 dumper->next_idx = log_next_idx;
3072 * kmsg_dump_rewind - reset the interator
3073 * @dumper: registered kmsg dumper
3075 * Reset the dumper's iterator so that kmsg_dump_get_line() and
3076 * kmsg_dump_get_buffer() can be called again and used multiple
3077 * times within the same dumper.dump() callback.
3079 void kmsg_dump_rewind(struct kmsg_dumper *dumper)
3081 unsigned long flags;
3083 raw_spin_lock_irqsave(&logbuf_lock, flags);
3084 kmsg_dump_rewind_nolock(dumper);
3085 raw_spin_unlock_irqrestore(&logbuf_lock, flags);
3087 EXPORT_SYMBOL_GPL(kmsg_dump_rewind);
3089 static char dump_stack_arch_desc_str[128];
3092 * dump_stack_set_arch_desc - set arch-specific str to show with task dumps
3093 * @fmt: printf-style format string
3094 * @...: arguments for the format string
3096 * The configured string will be printed right after utsname during task
3097 * dumps. Usually used to add arch-specific system identifiers. If an
3098 * arch wants to make use of such an ID string, it should initialize this
3099 * as soon as possible during boot.
3101 void __init dump_stack_set_arch_desc(const char *fmt, ...)
3103 va_list args;
3105 va_start(args, fmt);
3106 vsnprintf(dump_stack_arch_desc_str, sizeof(dump_stack_arch_desc_str),
3107 fmt, args);
3108 va_end(args);
3112 * dump_stack_print_info - print generic debug info for dump_stack()
3113 * @log_lvl: log level
3115 * Arch-specific dump_stack() implementations can use this function to
3116 * print out the same debug information as the generic dump_stack().
3118 void dump_stack_print_info(const char *log_lvl)
3120 printk("%sCPU: %d PID: %d Comm: %.20s %s %s %.*s\n",
3121 log_lvl, raw_smp_processor_id(), current->pid, current->comm,
3122 print_tainted(), init_utsname()->release,
3123 (int)strcspn(init_utsname()->version, " "),
3124 init_utsname()->version);
3126 if (dump_stack_arch_desc_str[0] != '\0')
3127 printk("%sHardware name: %s\n",
3128 log_lvl, dump_stack_arch_desc_str);
3130 print_worker_info(log_lvl, current);
3134 * show_regs_print_info - print generic debug info for show_regs()
3135 * @log_lvl: log level
3137 * show_regs() implementations can use this function to print out generic
3138 * debug information.
3140 void show_regs_print_info(const char *log_lvl)
3142 dump_stack_print_info(log_lvl);
3144 printk("%stask: %p task.stack: %p\n",
3145 log_lvl, current, task_stack_page(current));
3148 #endif