1 // SPDX-License-Identifier: GPL-2.0-only
3 * linux/kernel/printk.c
5 * Copyright (C) 1991, 1992 Linus Torvalds
7 * Modified to make sys_syslog() more flexible: added commands to
8 * return the last 4k of kernel messages, regardless of whether
9 * they've been read or not. Added option to suppress kernel printk's
10 * to the console. Added hook for sending the console messages
11 * elsewhere, in preparation for a serial line console (someday).
13 * Modified for sysctl support, 1/8/97, Chris Horn.
14 * Fixed SMP synchronization, 08/08/99, Manfred Spraul
15 * manfred@colorfullife.com
16 * Rewrote bits to get rid of console_lock
17 * 01Mar01 Andrew Morton
20 #define pr_fmt(fmt) KBUILD_MODNAME ": " fmt
22 #include <linux/kernel.h>
24 #include <linux/tty.h>
25 #include <linux/tty_driver.h>
26 #include <linux/console.h>
27 #include <linux/init.h>
28 #include <linux/jiffies.h>
29 #include <linux/nmi.h>
30 #include <linux/module.h>
31 #include <linux/moduleparam.h>
32 #include <linux/delay.h>
33 #include <linux/smp.h>
34 #include <linux/security.h>
35 #include <linux/memblock.h>
36 #include <linux/syscalls.h>
37 #include <linux/crash_core.h>
38 #include <linux/ratelimit.h>
39 #include <linux/kmsg_dump.h>
40 #include <linux/syslog.h>
41 #include <linux/cpu.h>
42 #include <linux/rculist.h>
43 #include <linux/poll.h>
44 #include <linux/irq_work.h>
45 #include <linux/ctype.h>
46 #include <linux/uio.h>
47 #include <linux/sched/clock.h>
48 #include <linux/sched/debug.h>
49 #include <linux/sched/task_stack.h>
51 #include <linux/uaccess.h>
52 #include <asm/sections.h>
54 #include <trace/events/initcall.h>
55 #define CREATE_TRACE_POINTS
56 #include <trace/events/printk.h>
58 #include "printk_ringbuffer.h"
59 #include "console_cmdline.h"
63 int console_printk[4] = {
64 CONSOLE_LOGLEVEL_DEFAULT, /* console_loglevel */
65 MESSAGE_LOGLEVEL_DEFAULT, /* default_message_loglevel */
66 CONSOLE_LOGLEVEL_MIN, /* minimum_console_loglevel */
67 CONSOLE_LOGLEVEL_DEFAULT, /* default_console_loglevel */
69 EXPORT_SYMBOL_GPL(console_printk);
71 atomic_t ignore_console_lock_warning __read_mostly = ATOMIC_INIT(0);
72 EXPORT_SYMBOL(ignore_console_lock_warning);
75 * Low level drivers may need that to know if they can schedule in
76 * their unblank() callback or not. So let's export it.
79 EXPORT_SYMBOL(oops_in_progress);
82 * console_sem protects the console_drivers list, and also
83 * provides serialisation for access to the entire console
86 static DEFINE_SEMAPHORE(console_sem);
87 struct console *console_drivers;
88 EXPORT_SYMBOL_GPL(console_drivers);
91 * System may need to suppress printk message under certain
92 * circumstances, like after kernel panic happens.
94 int __read_mostly suppress_printk;
97 static struct lockdep_map console_lock_dep_map = {
98 .name = "console_lock"
102 enum devkmsg_log_bits {
103 __DEVKMSG_LOG_BIT_ON = 0,
104 __DEVKMSG_LOG_BIT_OFF,
105 __DEVKMSG_LOG_BIT_LOCK,
108 enum devkmsg_log_masks {
109 DEVKMSG_LOG_MASK_ON = BIT(__DEVKMSG_LOG_BIT_ON),
110 DEVKMSG_LOG_MASK_OFF = BIT(__DEVKMSG_LOG_BIT_OFF),
111 DEVKMSG_LOG_MASK_LOCK = BIT(__DEVKMSG_LOG_BIT_LOCK),
114 /* Keep both the 'on' and 'off' bits clear, i.e. ratelimit by default: */
115 #define DEVKMSG_LOG_MASK_DEFAULT 0
117 static unsigned int __read_mostly devkmsg_log = DEVKMSG_LOG_MASK_DEFAULT;
119 static int __control_devkmsg(char *str)
126 len = str_has_prefix(str, "on");
128 devkmsg_log = DEVKMSG_LOG_MASK_ON;
132 len = str_has_prefix(str, "off");
134 devkmsg_log = DEVKMSG_LOG_MASK_OFF;
138 len = str_has_prefix(str, "ratelimit");
140 devkmsg_log = DEVKMSG_LOG_MASK_DEFAULT;
147 static int __init control_devkmsg(char *str)
149 if (__control_devkmsg(str) < 0)
153 * Set sysctl string accordingly:
155 if (devkmsg_log == DEVKMSG_LOG_MASK_ON)
156 strcpy(devkmsg_log_str, "on");
157 else if (devkmsg_log == DEVKMSG_LOG_MASK_OFF)
158 strcpy(devkmsg_log_str, "off");
159 /* else "ratelimit" which is set by default. */
162 * Sysctl cannot change it anymore. The kernel command line setting of
163 * this parameter is to force the setting to be permanent throughout the
164 * runtime of the system. This is a precation measure against userspace
165 * trying to be a smarta** and attempting to change it up on us.
167 devkmsg_log |= DEVKMSG_LOG_MASK_LOCK;
171 __setup("printk.devkmsg=", control_devkmsg);
173 char devkmsg_log_str[DEVKMSG_STR_MAX_SIZE] = "ratelimit";
175 int devkmsg_sysctl_set_loglvl(struct ctl_table *table, int write,
176 void *buffer, size_t *lenp, loff_t *ppos)
178 char old_str[DEVKMSG_STR_MAX_SIZE];
183 if (devkmsg_log & DEVKMSG_LOG_MASK_LOCK)
187 strncpy(old_str, devkmsg_log_str, DEVKMSG_STR_MAX_SIZE);
190 err = proc_dostring(table, write, buffer, lenp, ppos);
195 err = __control_devkmsg(devkmsg_log_str);
198 * Do not accept an unknown string OR a known string with
201 if (err < 0 || (err + 1 != *lenp)) {
203 /* ... and restore old setting. */
205 strncpy(devkmsg_log_str, old_str, DEVKMSG_STR_MAX_SIZE);
214 /* Number of registered extended console drivers. */
215 static int nr_ext_console_drivers;
218 * Helper macros to handle lockdep when locking/unlocking console_sem. We use
219 * macros instead of functions so that _RET_IP_ contains useful information.
221 #define down_console_sem() do { \
223 mutex_acquire(&console_lock_dep_map, 0, 0, _RET_IP_);\
226 static int __down_trylock_console_sem(unsigned long ip)
232 * Here and in __up_console_sem() we need to be in safe mode,
233 * because spindump/WARN/etc from under console ->lock will
234 * deadlock in printk()->down_trylock_console_sem() otherwise.
236 printk_safe_enter_irqsave(flags);
237 lock_failed = down_trylock(&console_sem);
238 printk_safe_exit_irqrestore(flags);
242 mutex_acquire(&console_lock_dep_map, 0, 1, ip);
245 #define down_trylock_console_sem() __down_trylock_console_sem(_RET_IP_)
247 static void __up_console_sem(unsigned long ip)
251 mutex_release(&console_lock_dep_map, ip);
253 printk_safe_enter_irqsave(flags);
255 printk_safe_exit_irqrestore(flags);
257 #define up_console_sem() __up_console_sem(_RET_IP_)
260 * This is used for debugging the mess that is the VT code by
261 * keeping track if we have the console semaphore held. It's
262 * definitely not the perfect debug tool (we don't know if _WE_
263 * hold it and are racing, but it helps tracking those weird code
264 * paths in the console code where we end up in places I want
265 * locked without the console sempahore held).
267 static int console_locked, console_suspended;
270 * If exclusive_console is non-NULL then only this console is to be printed to.
272 static struct console *exclusive_console;
275 * Array of consoles built from command line options (console=)
278 #define MAX_CMDLINECONSOLES 8
280 static struct console_cmdline console_cmdline[MAX_CMDLINECONSOLES];
282 static int preferred_console = -1;
283 static bool has_preferred_console;
284 int console_set_on_cmdline;
285 EXPORT_SYMBOL(console_set_on_cmdline);
287 /* Flag: console code may call schedule() */
288 static int console_may_schedule;
290 enum con_msg_format_flags {
291 MSG_FORMAT_DEFAULT = 0,
292 MSG_FORMAT_SYSLOG = (1 << 0),
295 static int console_msg_format = MSG_FORMAT_DEFAULT;
298 * The printk log buffer consists of a sequenced collection of records, each
299 * containing variable length message text. Every record also contains its
300 * own meta-data (@info).
302 * Every record meta-data carries the timestamp in microseconds, as well as
303 * the standard userspace syslog level and syslog facility. The usual kernel
304 * messages use LOG_KERN; userspace-injected messages always carry a matching
305 * syslog facility, by default LOG_USER. The origin of every message can be
306 * reliably determined that way.
308 * The human readable log message of a record is available in @text, the
309 * length of the message text in @text_len. The stored message is not
312 * Optionally, a record can carry a dictionary of properties (key/value
313 * pairs), to provide userspace with a machine-readable message context.
315 * Examples for well-defined, commonly used property names are:
316 * DEVICE=b12:8 device identifier
320 * +sound:card0 subsystem:devname
321 * SUBSYSTEM=pci driver-core subsystem name
323 * Valid characters in property names are [a-zA-Z0-9.-_]. Property names
324 * and values are terminated by a '\0' character.
326 * Example of record values:
327 * record.text_buf = "it's a line" (unterminated)
328 * record.info.seq = 56
329 * record.info.ts_nsec = 36863
330 * record.info.text_len = 11
331 * record.info.facility = 0 (LOG_KERN)
332 * record.info.flags = 0
333 * record.info.level = 3 (LOG_ERR)
334 * record.info.caller_id = 299 (task 299)
335 * record.info.dev_info.subsystem = "pci" (terminated)
336 * record.info.dev_info.device = "+pci:0000:00:01.0" (terminated)
338 * The 'struct printk_info' buffer must never be directly exported to
339 * userspace, it is a kernel-private implementation detail that might
340 * need to be changed in the future, when the requirements change.
342 * /dev/kmsg exports the structured data in the following line format:
343 * "<level>,<sequnum>,<timestamp>,<contflag>[,additional_values, ... ];<message text>\n"
345 * Users of the export format should ignore possible additional values
346 * separated by ',', and find the message after the ';' character.
348 * The optional key/value pairs are attached as continuation lines starting
349 * with a space character and terminated by a newline. All possible
350 * non-prinatable characters are escaped in the "\xff" notation.
354 LOG_NEWLINE = 2, /* text ended with a newline */
355 LOG_CONT = 8, /* text is a fragment of a continuation line */
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 * Helper macros to lock/unlock logbuf_lock and switch between
367 * printk-safe/unsafe modes.
369 #define logbuf_lock_irq() \
371 printk_safe_enter_irq(); \
372 raw_spin_lock(&logbuf_lock); \
375 #define logbuf_unlock_irq() \
377 raw_spin_unlock(&logbuf_lock); \
378 printk_safe_exit_irq(); \
381 #define logbuf_lock_irqsave(flags) \
383 printk_safe_enter_irqsave(flags); \
384 raw_spin_lock(&logbuf_lock); \
387 #define logbuf_unlock_irqrestore(flags) \
389 raw_spin_unlock(&logbuf_lock); \
390 printk_safe_exit_irqrestore(flags); \
394 DECLARE_WAIT_QUEUE_HEAD(log_wait);
395 /* the next printk record to read by syslog(READ) or /proc/kmsg */
396 static u64 syslog_seq;
397 static size_t syslog_partial;
398 static bool syslog_time;
400 /* the next printk record to write to the console */
401 static u64 console_seq;
402 static u64 exclusive_console_stop_seq;
403 static unsigned long console_dropped;
405 /* the next printk record to read after the last 'clear' command */
406 static u64 clear_seq;
408 #ifdef CONFIG_PRINTK_CALLER
409 #define PREFIX_MAX 48
411 #define PREFIX_MAX 32
413 #define LOG_LINE_MAX (1024 - PREFIX_MAX)
415 #define LOG_LEVEL(v) ((v) & 0x07)
416 #define LOG_FACILITY(v) ((v) >> 3 & 0xff)
419 #define LOG_ALIGN __alignof__(unsigned long)
420 #define __LOG_BUF_LEN (1 << CONFIG_LOG_BUF_SHIFT)
421 #define LOG_BUF_LEN_MAX (u32)(1 << 31)
422 static char __log_buf[__LOG_BUF_LEN] __aligned(LOG_ALIGN);
423 static char *log_buf = __log_buf;
424 static u32 log_buf_len = __LOG_BUF_LEN;
427 * Define the average message size. This only affects the number of
428 * descriptors that will be available. Underestimating is better than
429 * overestimating (too many available descriptors is better than not enough).
431 #define PRB_AVGBITS 5 /* 32 character average length */
433 #if CONFIG_LOG_BUF_SHIFT <= PRB_AVGBITS
434 #error CONFIG_LOG_BUF_SHIFT value too small.
436 _DEFINE_PRINTKRB(printk_rb_static, CONFIG_LOG_BUF_SHIFT - PRB_AVGBITS,
437 PRB_AVGBITS, &__log_buf[0]);
439 static struct printk_ringbuffer printk_rb_dynamic;
441 static struct printk_ringbuffer *prb = &printk_rb_static;
444 * We cannot access per-CPU data (e.g. per-CPU flush irq_work) before
445 * per_cpu_areas are initialised. This variable is set to true when
446 * it's safe to access per-CPU data.
448 static bool __printk_percpu_data_ready __read_mostly;
450 bool printk_percpu_data_ready(void)
452 return __printk_percpu_data_ready;
455 /* Return log buffer address */
456 char *log_buf_addr_get(void)
461 /* Return log buffer size */
462 u32 log_buf_len_get(void)
468 * Define how much of the log buffer we could take at maximum. The value
469 * must be greater than two. Note that only half of the buffer is available
470 * when the index points to the middle.
472 #define MAX_LOG_TAKE_PART 4
473 static const char trunc_msg[] = "<truncated>";
475 static void truncate_msg(u16 *text_len, u16 *trunc_msg_len)
478 * The message should not take the whole buffer. Otherwise, it might
479 * get removed too soon.
481 u32 max_text_len = log_buf_len / MAX_LOG_TAKE_PART;
483 if (*text_len > max_text_len)
484 *text_len = max_text_len;
486 /* enable the warning message (if there is room) */
487 *trunc_msg_len = strlen(trunc_msg);
488 if (*text_len >= *trunc_msg_len)
489 *text_len -= *trunc_msg_len;
494 /* insert record into the buffer, discard old ones, update heads */
495 static int log_store(u32 caller_id, int facility, int level,
496 enum log_flags flags, u64 ts_nsec,
497 const struct dev_printk_info *dev_info,
498 const char *text, u16 text_len)
500 struct prb_reserved_entry e;
501 struct printk_record r;
502 u16 trunc_msg_len = 0;
504 prb_rec_init_wr(&r, text_len);
506 if (!prb_reserve(&e, prb, &r)) {
507 /* truncate the message if it is too long for empty buffer */
508 truncate_msg(&text_len, &trunc_msg_len);
509 prb_rec_init_wr(&r, text_len + trunc_msg_len);
510 /* survive when the log buffer is too small for trunc_msg */
511 if (!prb_reserve(&e, prb, &r))
516 memcpy(&r.text_buf[0], text, text_len);
518 memcpy(&r.text_buf[text_len], trunc_msg, trunc_msg_len);
519 r.info->text_len = text_len + trunc_msg_len;
520 r.info->facility = facility;
521 r.info->level = level & 7;
522 r.info->flags = flags & 0x1f;
524 r.info->ts_nsec = ts_nsec;
526 r.info->ts_nsec = local_clock();
527 r.info->caller_id = caller_id;
529 memcpy(&r.info->dev_info, dev_info, sizeof(r.info->dev_info));
532 if ((flags & LOG_CONT) || !(flags & LOG_NEWLINE))
535 prb_final_commit(&e);
537 return (text_len + trunc_msg_len);
540 int dmesg_restrict = IS_ENABLED(CONFIG_SECURITY_DMESG_RESTRICT);
542 static int syslog_action_restricted(int type)
547 * Unless restricted, we allow "read all" and "get buffer size"
550 return type != SYSLOG_ACTION_READ_ALL &&
551 type != SYSLOG_ACTION_SIZE_BUFFER;
554 static int check_syslog_permissions(int type, int source)
557 * If this is from /proc/kmsg and we've already opened it, then we've
558 * already done the capabilities checks at open time.
560 if (source == SYSLOG_FROM_PROC && type != SYSLOG_ACTION_OPEN)
563 if (syslog_action_restricted(type)) {
564 if (capable(CAP_SYSLOG))
567 * For historical reasons, accept CAP_SYS_ADMIN too, with
570 if (capable(CAP_SYS_ADMIN)) {
571 pr_warn_once("%s (%d): Attempt to access syslog with "
572 "CAP_SYS_ADMIN but no CAP_SYSLOG "
574 current->comm, task_pid_nr(current));
580 return security_syslog(type);
583 static void append_char(char **pp, char *e, char c)
589 static ssize_t info_print_ext_header(char *buf, size_t size,
590 struct printk_info *info)
592 u64 ts_usec = info->ts_nsec;
594 #ifdef CONFIG_PRINTK_CALLER
595 u32 id = info->caller_id;
597 snprintf(caller, sizeof(caller), ",caller=%c%u",
598 id & 0x80000000 ? 'C' : 'T', id & ~0x80000000);
603 do_div(ts_usec, 1000);
605 return scnprintf(buf, size, "%u,%llu,%llu,%c%s;",
606 (info->facility << 3) | info->level, info->seq,
607 ts_usec, info->flags & LOG_CONT ? 'c' : '-', caller);
610 static ssize_t msg_add_ext_text(char *buf, size_t size,
611 const char *text, size_t text_len,
614 char *p = buf, *e = buf + size;
617 /* escape non-printable characters */
618 for (i = 0; i < text_len; i++) {
619 unsigned char c = text[i];
621 if (c < ' ' || c >= 127 || c == '\\')
622 p += scnprintf(p, e - p, "\\x%02x", c);
624 append_char(&p, e, c);
626 append_char(&p, e, endc);
631 static ssize_t msg_add_dict_text(char *buf, size_t size,
632 const char *key, const char *val)
634 size_t val_len = strlen(val);
640 len = msg_add_ext_text(buf, size, "", 0, ' '); /* dict prefix */
641 len += msg_add_ext_text(buf + len, size - len, key, strlen(key), '=');
642 len += msg_add_ext_text(buf + len, size - len, val, val_len, '\n');
647 static ssize_t msg_print_ext_body(char *buf, size_t size,
648 char *text, size_t text_len,
649 struct dev_printk_info *dev_info)
653 len = msg_add_ext_text(buf, size, text, text_len, '\n');
658 len += msg_add_dict_text(buf + len, size - len, "SUBSYSTEM",
659 dev_info->subsystem);
660 len += msg_add_dict_text(buf + len, size - len, "DEVICE",
666 /* /dev/kmsg - userspace message inject/listen interface */
667 struct devkmsg_user {
669 struct ratelimit_state rs;
671 char buf[CONSOLE_EXT_LOG_MAX];
673 struct printk_info info;
674 char text_buf[CONSOLE_EXT_LOG_MAX];
675 struct printk_record record;
678 static __printf(3, 4) __cold
679 int devkmsg_emit(int facility, int level, const char *fmt, ...)
685 r = vprintk_emit(facility, level, NULL, fmt, args);
691 static ssize_t devkmsg_write(struct kiocb *iocb, struct iov_iter *from)
694 int level = default_message_loglevel;
695 int facility = 1; /* LOG_USER */
696 struct file *file = iocb->ki_filp;
697 struct devkmsg_user *user = file->private_data;
698 size_t len = iov_iter_count(from);
701 if (!user || len > LOG_LINE_MAX)
704 /* Ignore when user logging is disabled. */
705 if (devkmsg_log & DEVKMSG_LOG_MASK_OFF)
708 /* Ratelimit when not explicitly enabled. */
709 if (!(devkmsg_log & DEVKMSG_LOG_MASK_ON)) {
710 if (!___ratelimit(&user->rs, current->comm))
714 buf = kmalloc(len+1, GFP_KERNEL);
719 if (!copy_from_iter_full(buf, len, from)) {
725 * Extract and skip the syslog prefix <[0-9]*>. Coming from userspace
726 * the decimal value represents 32bit, the lower 3 bit are the log
727 * level, the rest are the log facility.
729 * If no prefix or no userspace facility is specified, we
730 * enforce LOG_USER, to be able to reliably distinguish
731 * kernel-generated messages from userspace-injected ones.
734 if (line[0] == '<') {
738 u = simple_strtoul(line + 1, &endp, 10);
739 if (endp && endp[0] == '>') {
740 level = LOG_LEVEL(u);
741 if (LOG_FACILITY(u) != 0)
742 facility = LOG_FACILITY(u);
749 devkmsg_emit(facility, level, "%s", line);
754 static ssize_t devkmsg_read(struct file *file, char __user *buf,
755 size_t count, loff_t *ppos)
757 struct devkmsg_user *user = file->private_data;
758 struct printk_record *r = &user->record;
765 ret = mutex_lock_interruptible(&user->lock);
770 if (!prb_read_valid(prb, user->seq, r)) {
771 if (file->f_flags & O_NONBLOCK) {
778 ret = wait_event_interruptible(log_wait,
779 prb_read_valid(prb, user->seq, r));
785 if (user->seq < prb_first_valid_seq(prb)) {
786 /* our last seen message is gone, return error and reset */
787 user->seq = prb_first_valid_seq(prb);
793 len = info_print_ext_header(user->buf, sizeof(user->buf), r->info);
794 len += msg_print_ext_body(user->buf + len, sizeof(user->buf) - len,
795 &r->text_buf[0], r->info->text_len,
798 user->seq = r->info->seq + 1;
806 if (copy_to_user(buf, user->buf, len)) {
812 mutex_unlock(&user->lock);
816 static loff_t devkmsg_llseek(struct file *file, loff_t offset, int whence)
818 struct devkmsg_user *user = file->private_data;
829 /* the first record */
830 user->seq = prb_first_valid_seq(prb);
834 * The first record after the last SYSLOG_ACTION_CLEAR,
835 * like issued by 'dmesg -c'. Reading /dev/kmsg itself
836 * changes no global state, and does not clear anything.
838 user->seq = clear_seq;
841 /* after the last record */
842 user->seq = prb_next_seq(prb);
851 static __poll_t devkmsg_poll(struct file *file, poll_table *wait)
853 struct devkmsg_user *user = file->private_data;
857 return EPOLLERR|EPOLLNVAL;
859 poll_wait(file, &log_wait, wait);
862 if (prb_read_valid(prb, user->seq, NULL)) {
863 /* return error when data has vanished underneath us */
864 if (user->seq < prb_first_valid_seq(prb))
865 ret = EPOLLIN|EPOLLRDNORM|EPOLLERR|EPOLLPRI;
867 ret = EPOLLIN|EPOLLRDNORM;
874 static int devkmsg_open(struct inode *inode, struct file *file)
876 struct devkmsg_user *user;
879 if (devkmsg_log & DEVKMSG_LOG_MASK_OFF)
882 /* write-only does not need any file context */
883 if ((file->f_flags & O_ACCMODE) != O_WRONLY) {
884 err = check_syslog_permissions(SYSLOG_ACTION_READ_ALL,
890 user = kmalloc(sizeof(struct devkmsg_user), GFP_KERNEL);
894 ratelimit_default_init(&user->rs);
895 ratelimit_set_flags(&user->rs, RATELIMIT_MSG_ON_RELEASE);
897 mutex_init(&user->lock);
899 prb_rec_init_rd(&user->record, &user->info,
900 &user->text_buf[0], sizeof(user->text_buf));
903 user->seq = prb_first_valid_seq(prb);
906 file->private_data = user;
910 static int devkmsg_release(struct inode *inode, struct file *file)
912 struct devkmsg_user *user = file->private_data;
917 ratelimit_state_exit(&user->rs);
919 mutex_destroy(&user->lock);
924 const struct file_operations kmsg_fops = {
925 .open = devkmsg_open,
926 .read = devkmsg_read,
927 .write_iter = devkmsg_write,
928 .llseek = devkmsg_llseek,
929 .poll = devkmsg_poll,
930 .release = devkmsg_release,
933 #ifdef CONFIG_CRASH_CORE
935 * This appends the listed symbols to /proc/vmcore
937 * /proc/vmcore is used by various utilities, like crash and makedumpfile to
938 * obtain access to symbols that are otherwise very difficult to locate. These
939 * symbols are specifically used so that utilities can access and extract the
940 * dmesg log from a vmcore file after a crash.
942 void log_buf_vmcoreinfo_setup(void)
944 struct dev_printk_info *dev_info = NULL;
946 VMCOREINFO_SYMBOL(prb);
947 VMCOREINFO_SYMBOL(printk_rb_static);
948 VMCOREINFO_SYMBOL(clear_seq);
951 * Export struct size and field offsets. User space tools can
952 * parse it and detect any changes to structure down the line.
955 VMCOREINFO_STRUCT_SIZE(printk_ringbuffer);
956 VMCOREINFO_OFFSET(printk_ringbuffer, desc_ring);
957 VMCOREINFO_OFFSET(printk_ringbuffer, text_data_ring);
958 VMCOREINFO_OFFSET(printk_ringbuffer, fail);
960 VMCOREINFO_STRUCT_SIZE(prb_desc_ring);
961 VMCOREINFO_OFFSET(prb_desc_ring, count_bits);
962 VMCOREINFO_OFFSET(prb_desc_ring, descs);
963 VMCOREINFO_OFFSET(prb_desc_ring, infos);
964 VMCOREINFO_OFFSET(prb_desc_ring, head_id);
965 VMCOREINFO_OFFSET(prb_desc_ring, tail_id);
967 VMCOREINFO_STRUCT_SIZE(prb_desc);
968 VMCOREINFO_OFFSET(prb_desc, state_var);
969 VMCOREINFO_OFFSET(prb_desc, text_blk_lpos);
971 VMCOREINFO_STRUCT_SIZE(prb_data_blk_lpos);
972 VMCOREINFO_OFFSET(prb_data_blk_lpos, begin);
973 VMCOREINFO_OFFSET(prb_data_blk_lpos, next);
975 VMCOREINFO_STRUCT_SIZE(printk_info);
976 VMCOREINFO_OFFSET(printk_info, seq);
977 VMCOREINFO_OFFSET(printk_info, ts_nsec);
978 VMCOREINFO_OFFSET(printk_info, text_len);
979 VMCOREINFO_OFFSET(printk_info, caller_id);
980 VMCOREINFO_OFFSET(printk_info, dev_info);
982 VMCOREINFO_STRUCT_SIZE(dev_printk_info);
983 VMCOREINFO_OFFSET(dev_printk_info, subsystem);
984 VMCOREINFO_LENGTH(printk_info_subsystem, sizeof(dev_info->subsystem));
985 VMCOREINFO_OFFSET(dev_printk_info, device);
986 VMCOREINFO_LENGTH(printk_info_device, sizeof(dev_info->device));
988 VMCOREINFO_STRUCT_SIZE(prb_data_ring);
989 VMCOREINFO_OFFSET(prb_data_ring, size_bits);
990 VMCOREINFO_OFFSET(prb_data_ring, data);
991 VMCOREINFO_OFFSET(prb_data_ring, head_lpos);
992 VMCOREINFO_OFFSET(prb_data_ring, tail_lpos);
994 VMCOREINFO_SIZE(atomic_long_t);
995 VMCOREINFO_TYPE_OFFSET(atomic_long_t, counter);
999 /* requested log_buf_len from kernel cmdline */
1000 static unsigned long __initdata new_log_buf_len;
1002 /* we practice scaling the ring buffer by powers of 2 */
1003 static void __init log_buf_len_update(u64 size)
1005 if (size > (u64)LOG_BUF_LEN_MAX) {
1006 size = (u64)LOG_BUF_LEN_MAX;
1007 pr_err("log_buf over 2G is not supported.\n");
1011 size = roundup_pow_of_two(size);
1012 if (size > log_buf_len)
1013 new_log_buf_len = (unsigned long)size;
1016 /* save requested log_buf_len since it's too early to process it */
1017 static int __init log_buf_len_setup(char *str)
1024 size = memparse(str, &str);
1026 log_buf_len_update(size);
1030 early_param("log_buf_len", log_buf_len_setup);
1033 #define __LOG_CPU_MAX_BUF_LEN (1 << CONFIG_LOG_CPU_MAX_BUF_SHIFT)
1035 static void __init log_buf_add_cpu(void)
1037 unsigned int cpu_extra;
1040 * archs should set up cpu_possible_bits properly with
1041 * set_cpu_possible() after setup_arch() but just in
1042 * case lets ensure this is valid.
1044 if (num_possible_cpus() == 1)
1047 cpu_extra = (num_possible_cpus() - 1) * __LOG_CPU_MAX_BUF_LEN;
1049 /* by default this will only continue through for large > 64 CPUs */
1050 if (cpu_extra <= __LOG_BUF_LEN / 2)
1053 pr_info("log_buf_len individual max cpu contribution: %d bytes\n",
1054 __LOG_CPU_MAX_BUF_LEN);
1055 pr_info("log_buf_len total cpu_extra contributions: %d bytes\n",
1057 pr_info("log_buf_len min size: %d bytes\n", __LOG_BUF_LEN);
1059 log_buf_len_update(cpu_extra + __LOG_BUF_LEN);
1061 #else /* !CONFIG_SMP */
1062 static inline void log_buf_add_cpu(void) {}
1063 #endif /* CONFIG_SMP */
1065 static void __init set_percpu_data_ready(void)
1068 /* Make sure we set this flag only after printk_safe() init is done */
1070 __printk_percpu_data_ready = true;
1073 static unsigned int __init add_to_rb(struct printk_ringbuffer *rb,
1074 struct printk_record *r)
1076 struct prb_reserved_entry e;
1077 struct printk_record dest_r;
1079 prb_rec_init_wr(&dest_r, r->info->text_len);
1081 if (!prb_reserve(&e, rb, &dest_r))
1084 memcpy(&dest_r.text_buf[0], &r->text_buf[0], r->info->text_len);
1085 dest_r.info->text_len = r->info->text_len;
1086 dest_r.info->facility = r->info->facility;
1087 dest_r.info->level = r->info->level;
1088 dest_r.info->flags = r->info->flags;
1089 dest_r.info->ts_nsec = r->info->ts_nsec;
1090 dest_r.info->caller_id = r->info->caller_id;
1091 memcpy(&dest_r.info->dev_info, &r->info->dev_info, sizeof(dest_r.info->dev_info));
1093 prb_final_commit(&e);
1095 return prb_record_text_space(&e);
1098 static char setup_text_buf[CONSOLE_EXT_LOG_MAX] __initdata;
1100 void __init setup_log_buf(int early)
1102 struct printk_info *new_infos;
1103 unsigned int new_descs_count;
1104 struct prb_desc *new_descs;
1105 struct printk_info info;
1106 struct printk_record r;
1107 size_t new_descs_size;
1108 size_t new_infos_size;
1109 unsigned long flags;
1115 * Some archs call setup_log_buf() multiple times - first is very
1116 * early, e.g. from setup_arch(), and second - when percpu_areas
1120 set_percpu_data_ready();
1122 if (log_buf != __log_buf)
1125 if (!early && !new_log_buf_len)
1128 if (!new_log_buf_len)
1131 new_descs_count = new_log_buf_len >> PRB_AVGBITS;
1132 if (new_descs_count == 0) {
1133 pr_err("new_log_buf_len: %lu too small\n", new_log_buf_len);
1137 new_log_buf = memblock_alloc(new_log_buf_len, LOG_ALIGN);
1138 if (unlikely(!new_log_buf)) {
1139 pr_err("log_buf_len: %lu text bytes not available\n",
1144 new_descs_size = new_descs_count * sizeof(struct prb_desc);
1145 new_descs = memblock_alloc(new_descs_size, LOG_ALIGN);
1146 if (unlikely(!new_descs)) {
1147 pr_err("log_buf_len: %zu desc bytes not available\n",
1149 goto err_free_log_buf;
1152 new_infos_size = new_descs_count * sizeof(struct printk_info);
1153 new_infos = memblock_alloc(new_infos_size, LOG_ALIGN);
1154 if (unlikely(!new_infos)) {
1155 pr_err("log_buf_len: %zu info bytes not available\n",
1157 goto err_free_descs;
1160 prb_rec_init_rd(&r, &info, &setup_text_buf[0], sizeof(setup_text_buf));
1162 prb_init(&printk_rb_dynamic,
1163 new_log_buf, ilog2(new_log_buf_len),
1164 new_descs, ilog2(new_descs_count),
1167 logbuf_lock_irqsave(flags);
1169 log_buf_len = new_log_buf_len;
1170 log_buf = new_log_buf;
1171 new_log_buf_len = 0;
1173 free = __LOG_BUF_LEN;
1174 prb_for_each_record(0, &printk_rb_static, seq, &r)
1175 free -= add_to_rb(&printk_rb_dynamic, &r);
1178 * This is early enough that everything is still running on the
1179 * boot CPU and interrupts are disabled. So no new messages will
1180 * appear during the transition to the dynamic buffer.
1182 prb = &printk_rb_dynamic;
1184 logbuf_unlock_irqrestore(flags);
1186 if (seq != prb_next_seq(&printk_rb_static)) {
1187 pr_err("dropped %llu messages\n",
1188 prb_next_seq(&printk_rb_static) - seq);
1191 pr_info("log_buf_len: %u bytes\n", log_buf_len);
1192 pr_info("early log buf free: %u(%u%%)\n",
1193 free, (free * 100) / __LOG_BUF_LEN);
1197 memblock_free(__pa(new_descs), new_descs_size);
1199 memblock_free(__pa(new_log_buf), new_log_buf_len);
1202 static bool __read_mostly ignore_loglevel;
1204 static int __init ignore_loglevel_setup(char *str)
1206 ignore_loglevel = true;
1207 pr_info("debug: ignoring loglevel setting.\n");
1212 early_param("ignore_loglevel", ignore_loglevel_setup);
1213 module_param(ignore_loglevel, bool, S_IRUGO | S_IWUSR);
1214 MODULE_PARM_DESC(ignore_loglevel,
1215 "ignore loglevel setting (prints all kernel messages to the console)");
1217 static bool suppress_message_printing(int level)
1219 return (level >= console_loglevel && !ignore_loglevel);
1222 #ifdef CONFIG_BOOT_PRINTK_DELAY
1224 static int boot_delay; /* msecs delay after each printk during bootup */
1225 static unsigned long long loops_per_msec; /* based on boot_delay */
1227 static int __init boot_delay_setup(char *str)
1231 lpj = preset_lpj ? preset_lpj : 1000000; /* some guess */
1232 loops_per_msec = (unsigned long long)lpj / 1000 * HZ;
1234 get_option(&str, &boot_delay);
1235 if (boot_delay > 10 * 1000)
1238 pr_debug("boot_delay: %u, preset_lpj: %ld, lpj: %lu, "
1239 "HZ: %d, loops_per_msec: %llu\n",
1240 boot_delay, preset_lpj, lpj, HZ, loops_per_msec);
1243 early_param("boot_delay", boot_delay_setup);
1245 static void boot_delay_msec(int level)
1247 unsigned long long k;
1248 unsigned long timeout;
1250 if ((boot_delay == 0 || system_state >= SYSTEM_RUNNING)
1251 || suppress_message_printing(level)) {
1255 k = (unsigned long long)loops_per_msec * boot_delay;
1257 timeout = jiffies + msecs_to_jiffies(boot_delay);
1262 * use (volatile) jiffies to prevent
1263 * compiler reduction; loop termination via jiffies
1264 * is secondary and may or may not happen.
1266 if (time_after(jiffies, timeout))
1268 touch_nmi_watchdog();
1272 static inline void boot_delay_msec(int level)
1277 static bool printk_time = IS_ENABLED(CONFIG_PRINTK_TIME);
1278 module_param_named(time, printk_time, bool, S_IRUGO | S_IWUSR);
1280 static size_t print_syslog(unsigned int level, char *buf)
1282 return sprintf(buf, "<%u>", level);
1285 static size_t print_time(u64 ts, char *buf)
1287 unsigned long rem_nsec = do_div(ts, 1000000000);
1289 return sprintf(buf, "[%5lu.%06lu]",
1290 (unsigned long)ts, rem_nsec / 1000);
1293 #ifdef CONFIG_PRINTK_CALLER
1294 static size_t print_caller(u32 id, char *buf)
1298 snprintf(caller, sizeof(caller), "%c%u",
1299 id & 0x80000000 ? 'C' : 'T', id & ~0x80000000);
1300 return sprintf(buf, "[%6s]", caller);
1303 #define print_caller(id, buf) 0
1306 static size_t info_print_prefix(const struct printk_info *info, bool syslog,
1307 bool time, char *buf)
1312 len = print_syslog((info->facility << 3) | info->level, buf);
1315 len += print_time(info->ts_nsec, buf + len);
1317 len += print_caller(info->caller_id, buf + len);
1319 if (IS_ENABLED(CONFIG_PRINTK_CALLER) || time) {
1328 * Prepare the record for printing. The text is shifted within the given
1329 * buffer to avoid a need for another one. The following operations are
1332 * - Add prefix for each line.
1333 * - Add the trailing newline that has been removed in vprintk_store().
1334 * - Drop truncated lines that do not longer fit into the buffer.
1336 * Return: The length of the updated/prepared text, including the added
1337 * prefixes and the newline. The dropped line(s) are not counted.
1339 static size_t record_print_text(struct printk_record *r, bool syslog,
1342 size_t text_len = r->info->text_len;
1343 size_t buf_size = r->text_buf_size;
1344 char *text = r->text_buf;
1345 char prefix[PREFIX_MAX];
1346 bool truncated = false;
1352 prefix_len = info_print_prefix(r->info, syslog, time, prefix);
1355 * @text_len: bytes of unprocessed text
1356 * @line_len: bytes of current line _without_ newline
1357 * @text: pointer to beginning of current line
1358 * @len: number of bytes prepared in r->text_buf
1361 next = memchr(text, '\n', text_len);
1363 line_len = next - text;
1365 /* Drop truncated line(s). */
1368 line_len = text_len;
1372 * Truncate the text if there is not enough space to add the
1373 * prefix and a trailing newline.
1375 if (len + prefix_len + text_len + 1 > buf_size) {
1376 /* Drop even the current line if no space. */
1377 if (len + prefix_len + line_len + 1 > buf_size)
1380 text_len = buf_size - len - prefix_len - 1;
1384 memmove(text + prefix_len, text, text_len);
1385 memcpy(text, prefix, prefix_len);
1387 len += prefix_len + line_len + 1;
1389 if (text_len == line_len) {
1391 * Add the trailing newline removed in
1394 text[prefix_len + line_len] = '\n';
1399 * Advance beyond the added prefix and the related line with
1402 text += prefix_len + line_len + 1;
1405 * The remaining text has only decreased by the line with its
1408 * Note that @text_len can become zero. It happens when @text
1409 * ended with a newline (either due to truncation or the
1410 * original string ending with "\n\n"). The loop is correctly
1411 * repeated and (if not truncated) an empty line with a prefix
1414 text_len -= line_len + 1;
1420 static size_t get_record_print_text_size(struct printk_info *info,
1421 unsigned int line_count,
1422 bool syslog, bool time)
1424 char prefix[PREFIX_MAX];
1427 prefix_len = info_print_prefix(info, syslog, time, prefix);
1430 * Each line will be preceded with a prefix. The intermediate
1431 * newlines are already within the text, but a final trailing
1432 * newline will be added.
1434 return ((prefix_len * line_count) + info->text_len + 1);
1437 static int syslog_print(char __user *buf, int size)
1439 struct printk_info info;
1440 struct printk_record r;
1444 text = kmalloc(LOG_LINE_MAX + PREFIX_MAX, GFP_KERNEL);
1448 prb_rec_init_rd(&r, &info, text, LOG_LINE_MAX + PREFIX_MAX);
1455 if (!prb_read_valid(prb, syslog_seq, &r)) {
1456 logbuf_unlock_irq();
1459 if (r.info->seq != syslog_seq) {
1460 /* message is gone, move to next valid one */
1461 syslog_seq = r.info->seq;
1466 * To keep reading/counting partial line consistent,
1467 * use printk_time value as of the beginning of a line.
1469 if (!syslog_partial)
1470 syslog_time = printk_time;
1472 skip = syslog_partial;
1473 n = record_print_text(&r, true, syslog_time);
1474 if (n - syslog_partial <= size) {
1475 /* message fits into buffer, move forward */
1476 syslog_seq = r.info->seq + 1;
1477 n -= syslog_partial;
1480 /* partial read(), remember position */
1482 syslog_partial += n;
1485 logbuf_unlock_irq();
1490 if (copy_to_user(buf, text + skip, n)) {
1505 static int syslog_print_all(char __user *buf, int size, bool clear)
1507 struct printk_info info;
1508 unsigned int line_count;
1509 struct printk_record r;
1515 text = kmalloc(LOG_LINE_MAX + PREFIX_MAX, GFP_KERNEL);
1522 * Find first record that fits, including all following records,
1523 * into the user-provided buffer for this dump.
1525 prb_for_each_info(clear_seq, prb, seq, &info, &line_count)
1526 len += get_record_print_text_size(&info, line_count, true, time);
1528 /* move first record forward until length fits into the buffer */
1529 prb_for_each_info(clear_seq, prb, seq, &info, &line_count) {
1532 len -= get_record_print_text_size(&info, line_count, true, time);
1535 prb_rec_init_rd(&r, &info, text, LOG_LINE_MAX + PREFIX_MAX);
1538 prb_for_each_record(seq, prb, seq, &r) {
1541 textlen = record_print_text(&r, true, time);
1543 if (len + textlen > size) {
1548 logbuf_unlock_irq();
1549 if (copy_to_user(buf + len, text, textlen))
1561 logbuf_unlock_irq();
1567 static void syslog_clear(void)
1570 clear_seq = prb_next_seq(prb);
1571 logbuf_unlock_irq();
1574 int do_syslog(int type, char __user *buf, int len, int source)
1577 static int saved_console_loglevel = LOGLEVEL_DEFAULT;
1580 error = check_syslog_permissions(type, source);
1585 case SYSLOG_ACTION_CLOSE: /* Close log */
1587 case SYSLOG_ACTION_OPEN: /* Open log */
1589 case SYSLOG_ACTION_READ: /* Read from log */
1590 if (!buf || len < 0)
1594 if (!access_ok(buf, len))
1596 error = wait_event_interruptible(log_wait,
1597 prb_read_valid(prb, syslog_seq, NULL));
1600 error = syslog_print(buf, len);
1602 /* Read/clear last kernel messages */
1603 case SYSLOG_ACTION_READ_CLEAR:
1606 /* Read last kernel messages */
1607 case SYSLOG_ACTION_READ_ALL:
1608 if (!buf || len < 0)
1612 if (!access_ok(buf, len))
1614 error = syslog_print_all(buf, len, clear);
1616 /* Clear ring buffer */
1617 case SYSLOG_ACTION_CLEAR:
1620 /* Disable logging to console */
1621 case SYSLOG_ACTION_CONSOLE_OFF:
1622 if (saved_console_loglevel == LOGLEVEL_DEFAULT)
1623 saved_console_loglevel = console_loglevel;
1624 console_loglevel = minimum_console_loglevel;
1626 /* Enable logging to console */
1627 case SYSLOG_ACTION_CONSOLE_ON:
1628 if (saved_console_loglevel != LOGLEVEL_DEFAULT) {
1629 console_loglevel = saved_console_loglevel;
1630 saved_console_loglevel = LOGLEVEL_DEFAULT;
1633 /* Set level of messages printed to console */
1634 case SYSLOG_ACTION_CONSOLE_LEVEL:
1635 if (len < 1 || len > 8)
1637 if (len < minimum_console_loglevel)
1638 len = minimum_console_loglevel;
1639 console_loglevel = len;
1640 /* Implicitly re-enable logging to console */
1641 saved_console_loglevel = LOGLEVEL_DEFAULT;
1643 /* Number of chars in the log buffer */
1644 case SYSLOG_ACTION_SIZE_UNREAD:
1646 if (syslog_seq < prb_first_valid_seq(prb)) {
1647 /* messages are gone, move to first one */
1648 syslog_seq = prb_first_valid_seq(prb);
1651 if (source == SYSLOG_FROM_PROC) {
1653 * Short-cut for poll(/"proc/kmsg") which simply checks
1654 * for pending data, not the size; return the count of
1655 * records, not the length.
1657 error = prb_next_seq(prb) - syslog_seq;
1659 bool time = syslog_partial ? syslog_time : printk_time;
1660 struct printk_info info;
1661 unsigned int line_count;
1664 prb_for_each_info(syslog_seq, prb, seq, &info,
1666 error += get_record_print_text_size(&info, line_count,
1670 error -= syslog_partial;
1672 logbuf_unlock_irq();
1674 /* Size of the log buffer */
1675 case SYSLOG_ACTION_SIZE_BUFFER:
1676 error = log_buf_len;
1686 SYSCALL_DEFINE3(syslog, int, type, char __user *, buf, int, len)
1688 return do_syslog(type, buf, len, SYSLOG_FROM_READER);
1692 * Special console_lock variants that help to reduce the risk of soft-lockups.
1693 * They allow to pass console_lock to another printk() call using a busy wait.
1696 #ifdef CONFIG_LOCKDEP
1697 static struct lockdep_map console_owner_dep_map = {
1698 .name = "console_owner"
1702 static DEFINE_RAW_SPINLOCK(console_owner_lock);
1703 static struct task_struct *console_owner;
1704 static bool console_waiter;
1707 * console_lock_spinning_enable - mark beginning of code where another
1708 * thread might safely busy wait
1710 * This basically converts console_lock into a spinlock. This marks
1711 * the section where the console_lock owner can not sleep, because
1712 * there may be a waiter spinning (like a spinlock). Also it must be
1713 * ready to hand over the lock at the end of the section.
1715 static void console_lock_spinning_enable(void)
1717 raw_spin_lock(&console_owner_lock);
1718 console_owner = current;
1719 raw_spin_unlock(&console_owner_lock);
1721 /* The waiter may spin on us after setting console_owner */
1722 spin_acquire(&console_owner_dep_map, 0, 0, _THIS_IP_);
1726 * console_lock_spinning_disable_and_check - mark end of code where another
1727 * thread was able to busy wait and check if there is a waiter
1729 * This is called at the end of the section where spinning is allowed.
1730 * It has two functions. First, it is a signal that it is no longer
1731 * safe to start busy waiting for the lock. Second, it checks if
1732 * there is a busy waiter and passes the lock rights to her.
1734 * Important: Callers lose the lock if there was a busy waiter.
1735 * They must not touch items synchronized by console_lock
1738 * Return: 1 if the lock rights were passed, 0 otherwise.
1740 static int console_lock_spinning_disable_and_check(void)
1744 raw_spin_lock(&console_owner_lock);
1745 waiter = READ_ONCE(console_waiter);
1746 console_owner = NULL;
1747 raw_spin_unlock(&console_owner_lock);
1750 spin_release(&console_owner_dep_map, _THIS_IP_);
1754 /* The waiter is now free to continue */
1755 WRITE_ONCE(console_waiter, false);
1757 spin_release(&console_owner_dep_map, _THIS_IP_);
1760 * Hand off console_lock to waiter. The waiter will perform
1761 * the up(). After this, the waiter is the console_lock owner.
1763 mutex_release(&console_lock_dep_map, _THIS_IP_);
1768 * console_trylock_spinning - try to get console_lock by busy waiting
1770 * This allows to busy wait for the console_lock when the current
1771 * owner is running in specially marked sections. It means that
1772 * the current owner is running and cannot reschedule until it
1773 * is ready to lose the lock.
1775 * Return: 1 if we got the lock, 0 othrewise
1777 static int console_trylock_spinning(void)
1779 struct task_struct *owner = NULL;
1782 unsigned long flags;
1784 if (console_trylock())
1787 printk_safe_enter_irqsave(flags);
1789 raw_spin_lock(&console_owner_lock);
1790 owner = READ_ONCE(console_owner);
1791 waiter = READ_ONCE(console_waiter);
1792 if (!waiter && owner && owner != current) {
1793 WRITE_ONCE(console_waiter, true);
1796 raw_spin_unlock(&console_owner_lock);
1799 * If there is an active printk() writing to the
1800 * consoles, instead of having it write our data too,
1801 * see if we can offload that load from the active
1802 * printer, and do some printing ourselves.
1803 * Go into a spin only if there isn't already a waiter
1804 * spinning, and there is an active printer, and
1805 * that active printer isn't us (recursive printk?).
1808 printk_safe_exit_irqrestore(flags);
1812 /* We spin waiting for the owner to release us */
1813 spin_acquire(&console_owner_dep_map, 0, 0, _THIS_IP_);
1814 /* Owner will clear console_waiter on hand off */
1815 while (READ_ONCE(console_waiter))
1817 spin_release(&console_owner_dep_map, _THIS_IP_);
1819 printk_safe_exit_irqrestore(flags);
1821 * The owner passed the console lock to us.
1822 * Since we did not spin on console lock, annotate
1823 * this as a trylock. Otherwise lockdep will
1826 mutex_acquire(&console_lock_dep_map, 0, 1, _THIS_IP_);
1832 * Call the console drivers, asking them to write out
1833 * log_buf[start] to log_buf[end - 1].
1834 * The console_lock must be held.
1836 static void call_console_drivers(const char *ext_text, size_t ext_len,
1837 const char *text, size_t len)
1839 static char dropped_text[64];
1840 size_t dropped_len = 0;
1841 struct console *con;
1843 trace_console_rcuidle(text, len);
1845 if (!console_drivers)
1848 if (console_dropped) {
1849 dropped_len = snprintf(dropped_text, sizeof(dropped_text),
1850 "** %lu printk messages dropped **\n",
1852 console_dropped = 0;
1855 for_each_console(con) {
1856 if (exclusive_console && con != exclusive_console)
1858 if (!(con->flags & CON_ENABLED))
1862 if (!cpu_online(smp_processor_id()) &&
1863 !(con->flags & CON_ANYTIME))
1865 if (con->flags & CON_EXTENDED)
1866 con->write(con, ext_text, ext_len);
1869 con->write(con, dropped_text, dropped_len);
1870 con->write(con, text, len);
1875 int printk_delay_msec __read_mostly;
1877 static inline void printk_delay(void)
1879 if (unlikely(printk_delay_msec)) {
1880 int m = printk_delay_msec;
1884 touch_nmi_watchdog();
1889 static inline u32 printk_caller_id(void)
1891 return in_task() ? task_pid_nr(current) :
1892 0x80000000 + raw_smp_processor_id();
1895 static size_t log_output(int facility, int level, enum log_flags lflags,
1896 const struct dev_printk_info *dev_info,
1897 char *text, size_t text_len)
1899 const u32 caller_id = printk_caller_id();
1901 if (lflags & LOG_CONT) {
1902 struct prb_reserved_entry e;
1903 struct printk_record r;
1905 prb_rec_init_wr(&r, text_len);
1906 if (prb_reserve_in_last(&e, prb, &r, caller_id)) {
1907 memcpy(&r.text_buf[r.info->text_len], text, text_len);
1908 r.info->text_len += text_len;
1909 if (lflags & LOG_NEWLINE) {
1910 r.info->flags |= LOG_NEWLINE;
1911 prb_final_commit(&e);
1919 /* Store it in the record log */
1920 return log_store(caller_id, facility, level, lflags, 0,
1921 dev_info, text, text_len);
1924 /* Must be called under logbuf_lock. */
1925 int vprintk_store(int facility, int level,
1926 const struct dev_printk_info *dev_info,
1927 const char *fmt, va_list args)
1929 static char textbuf[LOG_LINE_MAX];
1930 char *text = textbuf;
1932 enum log_flags lflags = 0;
1935 * The printf needs to come first; we need the syslog
1936 * prefix which might be passed-in as a parameter.
1938 text_len = vscnprintf(text, sizeof(textbuf), fmt, args);
1940 /* mark and strip a trailing newline */
1941 if (text_len && text[text_len-1] == '\n') {
1943 lflags |= LOG_NEWLINE;
1946 /* strip kernel syslog prefix and extract log level or control flags */
1947 if (facility == 0) {
1950 while ((kern_level = printk_get_level(text)) != 0) {
1951 switch (kern_level) {
1953 if (level == LOGLEVEL_DEFAULT)
1954 level = kern_level - '0';
1956 case 'c': /* KERN_CONT */
1965 if (level == LOGLEVEL_DEFAULT)
1966 level = default_message_loglevel;
1969 lflags |= LOG_NEWLINE;
1971 return log_output(facility, level, lflags, dev_info, text, text_len);
1974 asmlinkage int vprintk_emit(int facility, int level,
1975 const struct dev_printk_info *dev_info,
1976 const char *fmt, va_list args)
1979 bool in_sched = false;
1980 unsigned long flags;
1982 /* Suppress unimportant messages after panic happens */
1983 if (unlikely(suppress_printk))
1986 if (level == LOGLEVEL_SCHED) {
1987 level = LOGLEVEL_DEFAULT;
1991 boot_delay_msec(level);
1994 /* This stops the holder of console_sem just where we want him */
1995 logbuf_lock_irqsave(flags);
1996 printed_len = vprintk_store(facility, level, dev_info, fmt, args);
1997 logbuf_unlock_irqrestore(flags);
1999 /* If called from the scheduler, we can not call up(). */
2002 * Disable preemption to avoid being preempted while holding
2003 * console_sem which would prevent anyone from printing to
2008 * Try to acquire and then immediately release the console
2009 * semaphore. The release will print out buffers and wake up
2010 * /dev/kmsg and syslog() users.
2012 if (console_trylock_spinning())
2020 EXPORT_SYMBOL(vprintk_emit);
2022 asmlinkage int vprintk(const char *fmt, va_list args)
2024 return vprintk_func(fmt, args);
2026 EXPORT_SYMBOL(vprintk);
2028 int vprintk_default(const char *fmt, va_list args)
2030 return vprintk_emit(0, LOGLEVEL_DEFAULT, NULL, fmt, args);
2032 EXPORT_SYMBOL_GPL(vprintk_default);
2035 * printk - print a kernel message
2036 * @fmt: format string
2038 * This is printk(). It can be called from any context. We want it to work.
2040 * We try to grab the console_lock. If we succeed, it's easy - we log the
2041 * output and call the console drivers. If we fail to get the semaphore, we
2042 * place the output into the log buffer and return. The current holder of
2043 * the console_sem will notice the new output in console_unlock(); and will
2044 * send it to the consoles before releasing the lock.
2046 * One effect of this deferred printing is that code which calls printk() and
2047 * then changes console_loglevel may break. This is because console_loglevel
2048 * is inspected when the actual printing occurs.
2053 * See the vsnprintf() documentation for format string extensions over C99.
2055 asmlinkage __visible int printk(const char *fmt, ...)
2060 va_start(args, fmt);
2061 r = vprintk_func(fmt, args);
2066 EXPORT_SYMBOL(printk);
2068 #else /* CONFIG_PRINTK */
2070 #define LOG_LINE_MAX 0
2071 #define PREFIX_MAX 0
2072 #define printk_time false
2074 #define prb_read_valid(rb, seq, r) false
2075 #define prb_first_valid_seq(rb) 0
2077 static u64 syslog_seq;
2078 static u64 console_seq;
2079 static u64 exclusive_console_stop_seq;
2080 static unsigned long console_dropped;
2082 static size_t record_print_text(const struct printk_record *r,
2083 bool syslog, bool time)
2087 static ssize_t info_print_ext_header(char *buf, size_t size,
2088 struct printk_info *info)
2092 static ssize_t msg_print_ext_body(char *buf, size_t size,
2093 char *text, size_t text_len,
2094 struct dev_printk_info *dev_info) { return 0; }
2095 static void console_lock_spinning_enable(void) { }
2096 static int console_lock_spinning_disable_and_check(void) { return 0; }
2097 static void call_console_drivers(const char *ext_text, size_t ext_len,
2098 const char *text, size_t len) {}
2099 static bool suppress_message_printing(int level) { return false; }
2101 #endif /* CONFIG_PRINTK */
2103 #ifdef CONFIG_EARLY_PRINTK
2104 struct console *early_console;
2106 asmlinkage __visible void early_printk(const char *fmt, ...)
2116 n = vscnprintf(buf, sizeof(buf), fmt, ap);
2119 early_console->write(early_console, buf, n);
2123 static int __add_preferred_console(char *name, int idx, char *options,
2124 char *brl_options, bool user_specified)
2126 struct console_cmdline *c;
2130 * See if this tty is not yet registered, and
2131 * if we have a slot free.
2133 for (i = 0, c = console_cmdline;
2134 i < MAX_CMDLINECONSOLES && c->name[0];
2136 if (strcmp(c->name, name) == 0 && c->index == idx) {
2138 preferred_console = i;
2140 c->user_specified = true;
2144 if (i == MAX_CMDLINECONSOLES)
2147 preferred_console = i;
2148 strlcpy(c->name, name, sizeof(c->name));
2149 c->options = options;
2150 c->user_specified = user_specified;
2151 braille_set_options(c, brl_options);
2157 static int __init console_msg_format_setup(char *str)
2159 if (!strcmp(str, "syslog"))
2160 console_msg_format = MSG_FORMAT_SYSLOG;
2161 if (!strcmp(str, "default"))
2162 console_msg_format = MSG_FORMAT_DEFAULT;
2165 __setup("console_msg_format=", console_msg_format_setup);
2168 * Set up a console. Called via do_early_param() in init/main.c
2169 * for each "console=" parameter in the boot command line.
2171 static int __init console_setup(char *str)
2173 char buf[sizeof(console_cmdline[0].name) + 4]; /* 4 for "ttyS" */
2174 char *s, *options, *brl_options = NULL;
2180 if (_braille_console_setup(&str, &brl_options))
2184 * Decode str into name, index, options.
2186 if (str[0] >= '0' && str[0] <= '9') {
2187 strcpy(buf, "ttyS");
2188 strncpy(buf + 4, str, sizeof(buf) - 5);
2190 strncpy(buf, str, sizeof(buf) - 1);
2192 buf[sizeof(buf) - 1] = 0;
2193 options = strchr(str, ',');
2197 if (!strcmp(str, "ttya"))
2198 strcpy(buf, "ttyS0");
2199 if (!strcmp(str, "ttyb"))
2200 strcpy(buf, "ttyS1");
2202 for (s = buf; *s; s++)
2203 if (isdigit(*s) || *s == ',')
2205 idx = simple_strtoul(s, NULL, 10);
2208 __add_preferred_console(buf, idx, options, brl_options, true);
2209 console_set_on_cmdline = 1;
2212 __setup("console=", console_setup);
2215 * add_preferred_console - add a device to the list of preferred consoles.
2216 * @name: device name
2217 * @idx: device index
2218 * @options: options for this console
2220 * The last preferred console added will be used for kernel messages
2221 * and stdin/out/err for init. Normally this is used by console_setup
2222 * above to handle user-supplied console arguments; however it can also
2223 * be used by arch-specific code either to override the user or more
2224 * commonly to provide a default console (ie from PROM variables) when
2225 * the user has not supplied one.
2227 int add_preferred_console(char *name, int idx, char *options)
2229 return __add_preferred_console(name, idx, options, NULL, false);
2232 bool console_suspend_enabled = true;
2233 EXPORT_SYMBOL(console_suspend_enabled);
2235 static int __init console_suspend_disable(char *str)
2237 console_suspend_enabled = false;
2240 __setup("no_console_suspend", console_suspend_disable);
2241 module_param_named(console_suspend, console_suspend_enabled,
2242 bool, S_IRUGO | S_IWUSR);
2243 MODULE_PARM_DESC(console_suspend, "suspend console during suspend"
2244 " and hibernate operations");
2247 * suspend_console - suspend the console subsystem
2249 * This disables printk() while we go into suspend states
2251 void suspend_console(void)
2253 if (!console_suspend_enabled)
2255 pr_info("Suspending console(s) (use no_console_suspend to debug)\n");
2257 console_suspended = 1;
2261 void resume_console(void)
2263 if (!console_suspend_enabled)
2266 console_suspended = 0;
2271 * console_cpu_notify - print deferred console messages after CPU hotplug
2274 * If printk() is called from a CPU that is not online yet, the messages
2275 * will be printed on the console only if there are CON_ANYTIME consoles.
2276 * This function is called when a new CPU comes online (or fails to come
2277 * up) or goes offline.
2279 static int console_cpu_notify(unsigned int cpu)
2281 if (!cpuhp_tasks_frozen) {
2282 /* If trylock fails, someone else is doing the printing */
2283 if (console_trylock())
2290 * console_lock - lock the console system for exclusive use.
2292 * Acquires a lock which guarantees that the caller has
2293 * exclusive access to the console system and the console_drivers list.
2295 * Can sleep, returns nothing.
2297 void console_lock(void)
2302 if (console_suspended)
2305 console_may_schedule = 1;
2307 EXPORT_SYMBOL(console_lock);
2310 * console_trylock - try to lock the console system for exclusive use.
2312 * Try to acquire a lock which guarantees that the caller has exclusive
2313 * access to the console system and the console_drivers list.
2315 * returns 1 on success, and 0 on failure to acquire the lock.
2317 int console_trylock(void)
2319 if (down_trylock_console_sem())
2321 if (console_suspended) {
2326 console_may_schedule = 0;
2329 EXPORT_SYMBOL(console_trylock);
2331 int is_console_locked(void)
2333 return console_locked;
2335 EXPORT_SYMBOL(is_console_locked);
2338 * Check if we have any console that is capable of printing while cpu is
2339 * booting or shutting down. Requires console_sem.
2341 static int have_callable_console(void)
2343 struct console *con;
2345 for_each_console(con)
2346 if ((con->flags & CON_ENABLED) &&
2347 (con->flags & CON_ANYTIME))
2354 * Can we actually use the console at this time on this cpu?
2356 * Console drivers may assume that per-cpu resources have been allocated. So
2357 * unless they're explicitly marked as being able to cope (CON_ANYTIME) don't
2358 * call them until this CPU is officially up.
2360 static inline int can_use_console(void)
2362 return cpu_online(raw_smp_processor_id()) || have_callable_console();
2366 * console_unlock - unlock the console system
2368 * Releases the console_lock which the caller holds on the console system
2369 * and the console driver list.
2371 * While the console_lock was held, console output may have been buffered
2372 * by printk(). If this is the case, console_unlock(); emits
2373 * the output prior to releasing the lock.
2375 * If there is output waiting, we wake /dev/kmsg and syslog() users.
2377 * console_unlock(); may be called from any context.
2379 void console_unlock(void)
2381 static char ext_text[CONSOLE_EXT_LOG_MAX];
2382 static char text[LOG_LINE_MAX + PREFIX_MAX];
2383 unsigned long flags;
2384 bool do_cond_resched, retry;
2385 struct printk_info info;
2386 struct printk_record r;
2388 if (console_suspended) {
2393 prb_rec_init_rd(&r, &info, text, sizeof(text));
2396 * Console drivers are called with interrupts disabled, so
2397 * @console_may_schedule should be cleared before; however, we may
2398 * end up dumping a lot of lines, for example, if called from
2399 * console registration path, and should invoke cond_resched()
2400 * between lines if allowable. Not doing so can cause a very long
2401 * scheduling stall on a slow console leading to RCU stall and
2402 * softlockup warnings which exacerbate the issue with more
2403 * messages practically incapacitating the system.
2405 * console_trylock() is not able to detect the preemptive
2406 * context reliably. Therefore the value must be stored before
2407 * and cleared after the the "again" goto label.
2409 do_cond_resched = console_may_schedule;
2411 console_may_schedule = 0;
2414 * We released the console_sem lock, so we need to recheck if
2415 * cpu is online and (if not) is there at least one CON_ANYTIME
2418 if (!can_use_console()) {
2428 printk_safe_enter_irqsave(flags);
2429 raw_spin_lock(&logbuf_lock);
2431 if (!prb_read_valid(prb, console_seq, &r))
2434 if (console_seq != r.info->seq) {
2435 console_dropped += r.info->seq - console_seq;
2436 console_seq = r.info->seq;
2439 if (suppress_message_printing(r.info->level)) {
2441 * Skip record we have buffered and already printed
2442 * directly to the console when we received it, and
2443 * record that has level above the console loglevel.
2449 /* Output to all consoles once old messages replayed. */
2450 if (unlikely(exclusive_console &&
2451 console_seq >= exclusive_console_stop_seq)) {
2452 exclusive_console = NULL;
2456 * Handle extended console text first because later
2457 * record_print_text() will modify the record buffer in-place.
2459 if (nr_ext_console_drivers) {
2460 ext_len = info_print_ext_header(ext_text,
2463 ext_len += msg_print_ext_body(ext_text + ext_len,
2464 sizeof(ext_text) - ext_len,
2469 len = record_print_text(&r,
2470 console_msg_format & MSG_FORMAT_SYSLOG,
2473 raw_spin_unlock(&logbuf_lock);
2476 * While actively printing out messages, if another printk()
2477 * were to occur on another CPU, it may wait for this one to
2478 * finish. This task can not be preempted if there is a
2479 * waiter waiting to take over.
2481 console_lock_spinning_enable();
2483 stop_critical_timings(); /* don't trace print latency */
2484 call_console_drivers(ext_text, ext_len, text, len);
2485 start_critical_timings();
2487 if (console_lock_spinning_disable_and_check()) {
2488 printk_safe_exit_irqrestore(flags);
2492 printk_safe_exit_irqrestore(flags);
2494 if (do_cond_resched)
2500 raw_spin_unlock(&logbuf_lock);
2505 * Someone could have filled up the buffer again, so re-check if there's
2506 * something to flush. In case we cannot trylock the console_sem again,
2507 * there's a new owner and the console_unlock() from them will do the
2508 * flush, no worries.
2510 raw_spin_lock(&logbuf_lock);
2511 retry = prb_read_valid(prb, console_seq, NULL);
2512 raw_spin_unlock(&logbuf_lock);
2513 printk_safe_exit_irqrestore(flags);
2515 if (retry && console_trylock())
2518 EXPORT_SYMBOL(console_unlock);
2521 * console_conditional_schedule - yield the CPU if required
2523 * If the console code is currently allowed to sleep, and
2524 * if this CPU should yield the CPU to another task, do
2527 * Must be called within console_lock();.
2529 void __sched console_conditional_schedule(void)
2531 if (console_may_schedule)
2534 EXPORT_SYMBOL(console_conditional_schedule);
2536 void console_unblank(void)
2541 * console_unblank can no longer be called in interrupt context unless
2542 * oops_in_progress is set to 1..
2544 if (oops_in_progress) {
2545 if (down_trylock_console_sem() != 0)
2551 console_may_schedule = 0;
2553 if ((c->flags & CON_ENABLED) && c->unblank)
2559 * console_flush_on_panic - flush console content on panic
2560 * @mode: flush all messages in buffer or just the pending ones
2562 * Immediately output all pending messages no matter what.
2564 void console_flush_on_panic(enum con_flush_mode mode)
2567 * If someone else is holding the console lock, trylock will fail
2568 * and may_schedule may be set. Ignore and proceed to unlock so
2569 * that messages are flushed out. As this can be called from any
2570 * context and we don't want to get preempted while flushing,
2571 * ensure may_schedule is cleared.
2574 console_may_schedule = 0;
2576 if (mode == CONSOLE_REPLAY_ALL) {
2577 unsigned long flags;
2579 logbuf_lock_irqsave(flags);
2580 console_seq = prb_first_valid_seq(prb);
2581 logbuf_unlock_irqrestore(flags);
2587 * Return the console tty driver structure and its associated index
2589 struct tty_driver *console_device(int *index)
2592 struct tty_driver *driver = NULL;
2595 for_each_console(c) {
2598 driver = c->device(c, index);
2607 * Prevent further output on the passed console device so that (for example)
2608 * serial drivers can disable console output before suspending a port, and can
2609 * re-enable output afterwards.
2611 void console_stop(struct console *console)
2614 console->flags &= ~CON_ENABLED;
2617 EXPORT_SYMBOL(console_stop);
2619 void console_start(struct console *console)
2622 console->flags |= CON_ENABLED;
2625 EXPORT_SYMBOL(console_start);
2627 static int __read_mostly keep_bootcon;
2629 static int __init keep_bootcon_setup(char *str)
2632 pr_info("debug: skip boot console de-registration.\n");
2637 early_param("keep_bootcon", keep_bootcon_setup);
2640 * This is called by register_console() to try to match
2641 * the newly registered console with any of the ones selected
2642 * by either the command line or add_preferred_console() and
2645 * Care need to be taken with consoles that are statically
2646 * enabled such as netconsole
2648 static int try_enable_new_console(struct console *newcon, bool user_specified)
2650 struct console_cmdline *c;
2653 for (i = 0, c = console_cmdline;
2654 i < MAX_CMDLINECONSOLES && c->name[0];
2656 if (c->user_specified != user_specified)
2658 if (!newcon->match ||
2659 newcon->match(newcon, c->name, c->index, c->options) != 0) {
2660 /* default matching */
2661 BUILD_BUG_ON(sizeof(c->name) != sizeof(newcon->name));
2662 if (strcmp(c->name, newcon->name) != 0)
2664 if (newcon->index >= 0 &&
2665 newcon->index != c->index)
2667 if (newcon->index < 0)
2668 newcon->index = c->index;
2670 if (_braille_register_console(newcon, c))
2673 if (newcon->setup &&
2674 newcon->setup(newcon, c->options) != 0)
2677 newcon->flags |= CON_ENABLED;
2678 if (i == preferred_console) {
2679 newcon->flags |= CON_CONSDEV;
2680 has_preferred_console = true;
2686 * Some consoles, such as pstore and netconsole, can be enabled even
2687 * without matching. Accept the pre-enabled consoles only when match()
2688 * and setup() had a change to be called.
2690 if (newcon->flags & CON_ENABLED && c->user_specified == user_specified)
2697 * The console driver calls this routine during kernel initialization
2698 * to register the console printing procedure with printk() and to
2699 * print any messages that were printed by the kernel before the
2700 * console driver was initialized.
2702 * This can happen pretty early during the boot process (because of
2703 * early_printk) - sometimes before setup_arch() completes - be careful
2704 * of what kernel features are used - they may not be initialised yet.
2706 * There are two types of consoles - bootconsoles (early_printk) and
2707 * "real" consoles (everything which is not a bootconsole) which are
2708 * handled differently.
2709 * - Any number of bootconsoles can be registered at any time.
2710 * - As soon as a "real" console is registered, all bootconsoles
2711 * will be unregistered automatically.
2712 * - Once a "real" console is registered, any attempt to register a
2713 * bootconsoles will be rejected
2715 void register_console(struct console *newcon)
2717 unsigned long flags;
2718 struct console *bcon = NULL;
2721 for_each_console(bcon) {
2722 if (WARN(bcon == newcon, "console '%s%d' already registered\n",
2723 bcon->name, bcon->index))
2728 * before we register a new CON_BOOT console, make sure we don't
2729 * already have a valid console
2731 if (newcon->flags & CON_BOOT) {
2732 for_each_console(bcon) {
2733 if (!(bcon->flags & CON_BOOT)) {
2734 pr_info("Too late to register bootconsole %s%d\n",
2735 newcon->name, newcon->index);
2741 if (console_drivers && console_drivers->flags & CON_BOOT)
2742 bcon = console_drivers;
2744 if (!has_preferred_console || bcon || !console_drivers)
2745 has_preferred_console = preferred_console >= 0;
2748 * See if we want to use this console driver. If we
2749 * didn't select a console we take the first one
2750 * that registers here.
2752 if (!has_preferred_console) {
2753 if (newcon->index < 0)
2755 if (newcon->setup == NULL ||
2756 newcon->setup(newcon, NULL) == 0) {
2757 newcon->flags |= CON_ENABLED;
2758 if (newcon->device) {
2759 newcon->flags |= CON_CONSDEV;
2760 has_preferred_console = true;
2765 /* See if this console matches one we selected on the command line */
2766 err = try_enable_new_console(newcon, true);
2768 /* If not, try to match against the platform default(s) */
2770 err = try_enable_new_console(newcon, false);
2772 /* printk() messages are not printed to the Braille console. */
2773 if (err || newcon->flags & CON_BRL)
2777 * If we have a bootconsole, and are switching to a real console,
2778 * don't print everything out again, since when the boot console, and
2779 * the real console are the same physical device, it's annoying to
2780 * see the beginning boot messages twice
2782 if (bcon && ((newcon->flags & (CON_CONSDEV | CON_BOOT)) == CON_CONSDEV))
2783 newcon->flags &= ~CON_PRINTBUFFER;
2786 * Put this console in the list - keep the
2787 * preferred driver at the head of the list.
2790 if ((newcon->flags & CON_CONSDEV) || console_drivers == NULL) {
2791 newcon->next = console_drivers;
2792 console_drivers = newcon;
2794 newcon->next->flags &= ~CON_CONSDEV;
2795 /* Ensure this flag is always set for the head of the list */
2796 newcon->flags |= CON_CONSDEV;
2798 newcon->next = console_drivers->next;
2799 console_drivers->next = newcon;
2802 if (newcon->flags & CON_EXTENDED)
2803 nr_ext_console_drivers++;
2805 if (newcon->flags & CON_PRINTBUFFER) {
2807 * console_unlock(); will print out the buffered messages
2810 logbuf_lock_irqsave(flags);
2812 * We're about to replay the log buffer. Only do this to the
2813 * just-registered console to avoid excessive message spam to
2814 * the already-registered consoles.
2816 * Set exclusive_console with disabled interrupts to reduce
2817 * race window with eventual console_flush_on_panic() that
2818 * ignores console_lock.
2820 exclusive_console = newcon;
2821 exclusive_console_stop_seq = console_seq;
2822 console_seq = syslog_seq;
2823 logbuf_unlock_irqrestore(flags);
2826 console_sysfs_notify();
2829 * By unregistering the bootconsoles after we enable the real console
2830 * we get the "console xxx enabled" message on all the consoles -
2831 * boot consoles, real consoles, etc - this is to ensure that end
2832 * users know there might be something in the kernel's log buffer that
2833 * went to the bootconsole (that they do not see on the real console)
2835 pr_info("%sconsole [%s%d] enabled\n",
2836 (newcon->flags & CON_BOOT) ? "boot" : "" ,
2837 newcon->name, newcon->index);
2839 ((newcon->flags & (CON_CONSDEV | CON_BOOT)) == CON_CONSDEV) &&
2841 /* We need to iterate through all boot consoles, to make
2842 * sure we print everything out, before we unregister them.
2844 for_each_console(bcon)
2845 if (bcon->flags & CON_BOOT)
2846 unregister_console(bcon);
2849 EXPORT_SYMBOL(register_console);
2851 int unregister_console(struct console *console)
2853 struct console *con;
2856 pr_info("%sconsole [%s%d] disabled\n",
2857 (console->flags & CON_BOOT) ? "boot" : "" ,
2858 console->name, console->index);
2860 res = _braille_unregister_console(console);
2868 if (console_drivers == console) {
2869 console_drivers=console->next;
2872 for_each_console(con) {
2873 if (con->next == console) {
2874 con->next = console->next;
2882 goto out_disable_unlock;
2884 if (console->flags & CON_EXTENDED)
2885 nr_ext_console_drivers--;
2888 * If this isn't the last console and it has CON_CONSDEV set, we
2889 * need to set it on the next preferred console.
2891 if (console_drivers != NULL && console->flags & CON_CONSDEV)
2892 console_drivers->flags |= CON_CONSDEV;
2894 console->flags &= ~CON_ENABLED;
2896 console_sysfs_notify();
2899 res = console->exit(console);
2904 console->flags &= ~CON_ENABLED;
2909 EXPORT_SYMBOL(unregister_console);
2912 * Initialize the console device. This is called *early*, so
2913 * we can't necessarily depend on lots of kernel help here.
2914 * Just do some early initializations, and do the complex setup
2917 void __init console_init(void)
2921 initcall_entry_t *ce;
2923 /* Setup the default TTY line discipline. */
2927 * set up the console device so that later boot sequences can
2928 * inform about problems etc..
2930 ce = __con_initcall_start;
2931 trace_initcall_level("console");
2932 while (ce < __con_initcall_end) {
2933 call = initcall_from_entry(ce);
2934 trace_initcall_start(call);
2936 trace_initcall_finish(call, ret);
2942 * Some boot consoles access data that is in the init section and which will
2943 * be discarded after the initcalls have been run. To make sure that no code
2944 * will access this data, unregister the boot consoles in a late initcall.
2946 * If for some reason, such as deferred probe or the driver being a loadable
2947 * module, the real console hasn't registered yet at this point, there will
2948 * be a brief interval in which no messages are logged to the console, which
2949 * makes it difficult to diagnose problems that occur during this time.
2951 * To mitigate this problem somewhat, only unregister consoles whose memory
2952 * intersects with the init section. Note that all other boot consoles will
2953 * get unregistred when the real preferred console is registered.
2955 static int __init printk_late_init(void)
2957 struct console *con;
2960 for_each_console(con) {
2961 if (!(con->flags & CON_BOOT))
2964 /* Check addresses that might be used for enabled consoles. */
2965 if (init_section_intersects(con, sizeof(*con)) ||
2966 init_section_contains(con->write, 0) ||
2967 init_section_contains(con->read, 0) ||
2968 init_section_contains(con->device, 0) ||
2969 init_section_contains(con->unblank, 0) ||
2970 init_section_contains(con->data, 0)) {
2972 * Please, consider moving the reported consoles out
2973 * of the init section.
2975 pr_warn("bootconsole [%s%d] uses init memory and must be disabled even before the real one is ready\n",
2976 con->name, con->index);
2977 unregister_console(con);
2980 ret = cpuhp_setup_state_nocalls(CPUHP_PRINTK_DEAD, "printk:dead", NULL,
2981 console_cpu_notify);
2983 ret = cpuhp_setup_state_nocalls(CPUHP_AP_ONLINE_DYN, "printk:online",
2984 console_cpu_notify, NULL);
2988 late_initcall(printk_late_init);
2990 #if defined CONFIG_PRINTK
2992 * Delayed printk version, for scheduler-internal messages:
2994 #define PRINTK_PENDING_WAKEUP 0x01
2995 #define PRINTK_PENDING_OUTPUT 0x02
2997 static DEFINE_PER_CPU(int, printk_pending);
2999 static void wake_up_klogd_work_func(struct irq_work *irq_work)
3001 int pending = __this_cpu_xchg(printk_pending, 0);
3003 if (pending & PRINTK_PENDING_OUTPUT) {
3004 /* If trylock fails, someone else is doing the printing */
3005 if (console_trylock())
3009 if (pending & PRINTK_PENDING_WAKEUP)
3010 wake_up_interruptible(&log_wait);
3013 static DEFINE_PER_CPU(struct irq_work, wake_up_klogd_work) = {
3014 .func = wake_up_klogd_work_func,
3015 .flags = ATOMIC_INIT(IRQ_WORK_LAZY),
3018 void wake_up_klogd(void)
3020 if (!printk_percpu_data_ready())
3024 if (waitqueue_active(&log_wait)) {
3025 this_cpu_or(printk_pending, PRINTK_PENDING_WAKEUP);
3026 irq_work_queue(this_cpu_ptr(&wake_up_klogd_work));
3031 void defer_console_output(void)
3033 if (!printk_percpu_data_ready())
3037 __this_cpu_or(printk_pending, PRINTK_PENDING_OUTPUT);
3038 irq_work_queue(this_cpu_ptr(&wake_up_klogd_work));
3042 int vprintk_deferred(const char *fmt, va_list args)
3046 r = vprintk_emit(0, LOGLEVEL_SCHED, NULL, fmt, args);
3047 defer_console_output();
3052 int printk_deferred(const char *fmt, ...)
3057 va_start(args, fmt);
3058 r = vprintk_deferred(fmt, args);
3065 * printk rate limiting, lifted from the networking subsystem.
3067 * This enforces a rate limit: not more than 10 kernel messages
3068 * every 5s to make a denial-of-service attack impossible.
3070 DEFINE_RATELIMIT_STATE(printk_ratelimit_state, 5 * HZ, 10);
3072 int __printk_ratelimit(const char *func)
3074 return ___ratelimit(&printk_ratelimit_state, func);
3076 EXPORT_SYMBOL(__printk_ratelimit);
3079 * printk_timed_ratelimit - caller-controlled printk ratelimiting
3080 * @caller_jiffies: pointer to caller's state
3081 * @interval_msecs: minimum interval between prints
3083 * printk_timed_ratelimit() returns true if more than @interval_msecs
3084 * milliseconds have elapsed since the last time printk_timed_ratelimit()
3087 bool printk_timed_ratelimit(unsigned long *caller_jiffies,
3088 unsigned int interval_msecs)
3090 unsigned long elapsed = jiffies - *caller_jiffies;
3092 if (*caller_jiffies && elapsed <= msecs_to_jiffies(interval_msecs))
3095 *caller_jiffies = jiffies;
3098 EXPORT_SYMBOL(printk_timed_ratelimit);
3100 static DEFINE_SPINLOCK(dump_list_lock);
3101 static LIST_HEAD(dump_list);
3104 * kmsg_dump_register - register a kernel log dumper.
3105 * @dumper: pointer to the kmsg_dumper structure
3107 * Adds a kernel log dumper to the system. The dump callback in the
3108 * structure will be called when the kernel oopses or panics and must be
3109 * set. Returns zero on success and %-EINVAL or %-EBUSY otherwise.
3111 int kmsg_dump_register(struct kmsg_dumper *dumper)
3113 unsigned long flags;
3116 /* The dump callback needs to be set */
3120 spin_lock_irqsave(&dump_list_lock, flags);
3121 /* Don't allow registering multiple times */
3122 if (!dumper->registered) {
3123 dumper->registered = 1;
3124 list_add_tail_rcu(&dumper->list, &dump_list);
3127 spin_unlock_irqrestore(&dump_list_lock, flags);
3131 EXPORT_SYMBOL_GPL(kmsg_dump_register);
3134 * kmsg_dump_unregister - unregister a kmsg dumper.
3135 * @dumper: pointer to the kmsg_dumper structure
3137 * Removes a dump device from the system. Returns zero on success and
3138 * %-EINVAL otherwise.
3140 int kmsg_dump_unregister(struct kmsg_dumper *dumper)
3142 unsigned long flags;
3145 spin_lock_irqsave(&dump_list_lock, flags);
3146 if (dumper->registered) {
3147 dumper->registered = 0;
3148 list_del_rcu(&dumper->list);
3151 spin_unlock_irqrestore(&dump_list_lock, flags);
3156 EXPORT_SYMBOL_GPL(kmsg_dump_unregister);
3158 static bool always_kmsg_dump;
3159 module_param_named(always_kmsg_dump, always_kmsg_dump, bool, S_IRUGO | S_IWUSR);
3161 const char *kmsg_dump_reason_str(enum kmsg_dump_reason reason)
3164 case KMSG_DUMP_PANIC:
3166 case KMSG_DUMP_OOPS:
3168 case KMSG_DUMP_EMERG:
3170 case KMSG_DUMP_SHUTDOWN:
3176 EXPORT_SYMBOL_GPL(kmsg_dump_reason_str);
3179 * kmsg_dump - dump kernel log to kernel message dumpers.
3180 * @reason: the reason (oops, panic etc) for dumping
3182 * Call each of the registered dumper's dump() callback, which can
3183 * retrieve the kmsg records with kmsg_dump_get_line() or
3184 * kmsg_dump_get_buffer().
3186 void kmsg_dump(enum kmsg_dump_reason reason)
3188 struct kmsg_dumper *dumper;
3189 unsigned long flags;
3192 list_for_each_entry_rcu(dumper, &dump_list, list) {
3193 enum kmsg_dump_reason max_reason = dumper->max_reason;
3196 * If client has not provided a specific max_reason, default
3197 * to KMSG_DUMP_OOPS, unless always_kmsg_dump was set.
3199 if (max_reason == KMSG_DUMP_UNDEF) {
3200 max_reason = always_kmsg_dump ? KMSG_DUMP_MAX :
3203 if (reason > max_reason)
3206 /* initialize iterator with data about the stored records */
3207 dumper->active = true;
3209 logbuf_lock_irqsave(flags);
3210 dumper->cur_seq = clear_seq;
3211 dumper->next_seq = prb_next_seq(prb);
3212 logbuf_unlock_irqrestore(flags);
3214 /* invoke dumper which will iterate over records */
3215 dumper->dump(dumper, reason);
3217 /* reset iterator */
3218 dumper->active = false;
3224 * kmsg_dump_get_line_nolock - retrieve one kmsg log line (unlocked version)
3225 * @dumper: registered kmsg dumper
3226 * @syslog: include the "<4>" prefixes
3227 * @line: buffer to copy the line to
3228 * @size: maximum size of the buffer
3229 * @len: length of line placed into buffer
3231 * Start at the beginning of the kmsg buffer, with the oldest kmsg
3232 * record, and copy one record into the provided buffer.
3234 * Consecutive calls will return the next available record moving
3235 * towards the end of the buffer with the youngest messages.
3237 * A return value of FALSE indicates that there are no more records to
3240 * The function is similar to kmsg_dump_get_line(), but grabs no locks.
3242 bool kmsg_dump_get_line_nolock(struct kmsg_dumper *dumper, bool syslog,
3243 char *line, size_t size, size_t *len)
3245 struct printk_info info;
3246 unsigned int line_count;
3247 struct printk_record r;
3251 prb_rec_init_rd(&r, &info, line, size);
3253 if (!dumper->active)
3256 /* Read text or count text lines? */
3258 if (!prb_read_valid(prb, dumper->cur_seq, &r))
3260 l = record_print_text(&r, syslog, printk_time);
3262 if (!prb_read_valid_info(prb, dumper->cur_seq,
3263 &info, &line_count)) {
3266 l = get_record_print_text_size(&info, line_count, syslog,
3271 dumper->cur_seq = r.info->seq + 1;
3280 * kmsg_dump_get_line - retrieve one kmsg log line
3281 * @dumper: registered kmsg dumper
3282 * @syslog: include the "<4>" prefixes
3283 * @line: buffer to copy the line to
3284 * @size: maximum size of the buffer
3285 * @len: length of line placed into buffer
3287 * Start at the beginning of the kmsg buffer, with the oldest kmsg
3288 * record, and copy one record into the provided buffer.
3290 * Consecutive calls will return the next available record moving
3291 * towards the end of the buffer with the youngest messages.
3293 * A return value of FALSE indicates that there are no more records to
3296 bool kmsg_dump_get_line(struct kmsg_dumper *dumper, bool syslog,
3297 char *line, size_t size, size_t *len)
3299 unsigned long flags;
3302 logbuf_lock_irqsave(flags);
3303 ret = kmsg_dump_get_line_nolock(dumper, syslog, line, size, len);
3304 logbuf_unlock_irqrestore(flags);
3308 EXPORT_SYMBOL_GPL(kmsg_dump_get_line);
3311 * kmsg_dump_get_buffer - copy kmsg log lines
3312 * @dumper: registered kmsg dumper
3313 * @syslog: include the "<4>" prefixes
3314 * @buf: buffer to copy the line to
3315 * @size: maximum size of the buffer
3316 * @len: length of line placed into buffer
3318 * Start at the end of the kmsg buffer and fill the provided buffer
3319 * with as many of the the *youngest* kmsg records that fit into it.
3320 * If the buffer is large enough, all available kmsg records will be
3321 * copied with a single call.
3323 * Consecutive calls will fill the buffer with the next block of
3324 * available older records, not including the earlier retrieved ones.
3326 * A return value of FALSE indicates that there are no more records to
3329 bool kmsg_dump_get_buffer(struct kmsg_dumper *dumper, bool syslog,
3330 char *buf, size_t size, size_t *len)
3332 struct printk_info info;
3333 unsigned int line_count;
3334 struct printk_record r;
3335 unsigned long flags;
3340 bool time = printk_time;
3342 prb_rec_init_rd(&r, &info, buf, size);
3344 if (!dumper->active || !buf || !size)
3347 logbuf_lock_irqsave(flags);
3348 if (dumper->cur_seq < prb_first_valid_seq(prb)) {
3349 /* messages are gone, move to first available one */
3350 dumper->cur_seq = prb_first_valid_seq(prb);
3354 if (dumper->cur_seq >= dumper->next_seq) {
3355 logbuf_unlock_irqrestore(flags);
3359 /* calculate length of entire buffer */
3360 seq = dumper->cur_seq;
3361 while (prb_read_valid_info(prb, seq, &info, &line_count)) {
3362 if (r.info->seq >= dumper->next_seq)
3364 l += get_record_print_text_size(&info, line_count, true, time);
3365 seq = r.info->seq + 1;
3368 /* move first record forward until length fits into the buffer */
3369 seq = dumper->cur_seq;
3370 while (l >= size && prb_read_valid_info(prb, seq,
3371 &info, &line_count)) {
3372 if (r.info->seq >= dumper->next_seq)
3374 l -= get_record_print_text_size(&info, line_count, true, time);
3375 seq = r.info->seq + 1;
3378 /* last message in next interation */
3381 /* actually read text into the buffer now */
3383 while (prb_read_valid(prb, seq, &r)) {
3384 if (r.info->seq >= dumper->next_seq)
3387 l += record_print_text(&r, syslog, time);
3389 /* adjust record to store to remaining buffer space */
3390 prb_rec_init_rd(&r, &info, buf + l, size - l);
3392 seq = r.info->seq + 1;
3395 dumper->next_seq = next_seq;
3397 logbuf_unlock_irqrestore(flags);
3403 EXPORT_SYMBOL_GPL(kmsg_dump_get_buffer);
3406 * kmsg_dump_rewind_nolock - reset the iterator (unlocked version)
3407 * @dumper: registered kmsg dumper
3409 * Reset the dumper's iterator so that kmsg_dump_get_line() and
3410 * kmsg_dump_get_buffer() can be called again and used multiple
3411 * times within the same dumper.dump() callback.
3413 * The function is similar to kmsg_dump_rewind(), but grabs no locks.
3415 void kmsg_dump_rewind_nolock(struct kmsg_dumper *dumper)
3417 dumper->cur_seq = clear_seq;
3418 dumper->next_seq = prb_next_seq(prb);
3422 * kmsg_dump_rewind - reset the iterator
3423 * @dumper: registered kmsg dumper
3425 * Reset the dumper's iterator so that kmsg_dump_get_line() and
3426 * kmsg_dump_get_buffer() can be called again and used multiple
3427 * times within the same dumper.dump() callback.
3429 void kmsg_dump_rewind(struct kmsg_dumper *dumper)
3431 unsigned long flags;
3433 logbuf_lock_irqsave(flags);
3434 kmsg_dump_rewind_nolock(dumper);
3435 logbuf_unlock_irqrestore(flags);
3437 EXPORT_SYMBOL_GPL(kmsg_dump_rewind);