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 and dictionary text. Every record
300 * also contains its 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. The
314 * length of the dictionary is available in @dict_len. The dictionary is not
317 * Examples for well-defined, commonly used property names are:
318 * DEVICE=b12:8 device identifier
322 * +sound:card0 subsystem:devname
323 * SUBSYSTEM=pci driver-core subsystem name
325 * Valid characters in property names are [a-zA-Z0-9.-_]. The plain text value
326 * follows directly after a '=' character. Every property is terminated by
327 * a '\0' character. The last property is not terminated.
329 * Example of record values:
330 * record.text_buf = "it's a line" (unterminated)
331 * record.dict_buf = "DEVICE=b8:2\0DRIVER=bug" (unterminated)
332 * record.info.seq = 56
333 * record.info.ts_nsec = 36863
334 * record.info.text_len = 11
335 * record.info.dict_len = 22
336 * record.info.facility = 0 (LOG_KERN)
337 * record.info.flags = 0
338 * record.info.level = 3 (LOG_ERR)
339 * record.info.caller_id = 299 (task 299)
341 * The 'struct printk_info' buffer must never be directly exported to
342 * userspace, it is a kernel-private implementation detail that might
343 * need to be changed in the future, when the requirements change.
345 * /dev/kmsg exports the structured data in the following line format:
346 * "<level>,<sequnum>,<timestamp>,<contflag>[,additional_values, ... ];<message text>\n"
348 * Users of the export format should ignore possible additional values
349 * separated by ',', and find the message after the ';' character.
351 * The optional key/value pairs are attached as continuation lines starting
352 * with a space character and terminated by a newline. All possible
353 * non-prinatable characters are escaped in the "\xff" notation.
357 LOG_NEWLINE = 2, /* text ended with a newline */
358 LOG_CONT = 8, /* text is a fragment of a continuation line */
362 * The logbuf_lock protects kmsg buffer, indices, counters. This can be taken
363 * within the scheduler's rq lock. It must be released before calling
364 * console_unlock() or anything else that might wake up a process.
366 DEFINE_RAW_SPINLOCK(logbuf_lock);
369 * Helper macros to lock/unlock logbuf_lock and switch between
370 * printk-safe/unsafe modes.
372 #define logbuf_lock_irq() \
374 printk_safe_enter_irq(); \
375 raw_spin_lock(&logbuf_lock); \
378 #define logbuf_unlock_irq() \
380 raw_spin_unlock(&logbuf_lock); \
381 printk_safe_exit_irq(); \
384 #define logbuf_lock_irqsave(flags) \
386 printk_safe_enter_irqsave(flags); \
387 raw_spin_lock(&logbuf_lock); \
390 #define logbuf_unlock_irqrestore(flags) \
392 raw_spin_unlock(&logbuf_lock); \
393 printk_safe_exit_irqrestore(flags); \
397 DECLARE_WAIT_QUEUE_HEAD(log_wait);
398 /* the next printk record to read by syslog(READ) or /proc/kmsg */
399 static u64 syslog_seq;
400 static size_t syslog_partial;
401 static bool syslog_time;
403 /* the next printk record to write to the console */
404 static u64 console_seq;
405 static u64 exclusive_console_stop_seq;
406 static unsigned long console_dropped;
408 /* the next printk record to read after the last 'clear' command */
409 static u64 clear_seq;
411 #ifdef CONFIG_PRINTK_CALLER
412 #define PREFIX_MAX 48
414 #define PREFIX_MAX 32
416 #define LOG_LINE_MAX (1024 - PREFIX_MAX)
418 #define LOG_LEVEL(v) ((v) & 0x07)
419 #define LOG_FACILITY(v) ((v) >> 3 & 0xff)
422 #define LOG_ALIGN __alignof__(unsigned long)
423 #define __LOG_BUF_LEN (1 << CONFIG_LOG_BUF_SHIFT)
424 #define LOG_BUF_LEN_MAX (u32)(1 << 31)
425 static char __log_buf[__LOG_BUF_LEN] __aligned(LOG_ALIGN);
426 static char *log_buf = __log_buf;
427 static u32 log_buf_len = __LOG_BUF_LEN;
430 * Define the average message size. This only affects the number of
431 * descriptors that will be available. Underestimating is better than
432 * overestimating (too many available descriptors is better than not enough).
433 * The dictionary buffer will be the same size as the text buffer.
435 #define PRB_AVGBITS 5 /* 32 character average length */
437 #if CONFIG_LOG_BUF_SHIFT <= PRB_AVGBITS
438 #error CONFIG_LOG_BUF_SHIFT value too small.
440 _DEFINE_PRINTKRB(printk_rb_static, CONFIG_LOG_BUF_SHIFT - PRB_AVGBITS,
441 PRB_AVGBITS, PRB_AVGBITS, &__log_buf[0]);
443 static struct printk_ringbuffer printk_rb_dynamic;
445 static struct printk_ringbuffer *prb = &printk_rb_static;
448 * We cannot access per-CPU data (e.g. per-CPU flush irq_work) before
449 * per_cpu_areas are initialised. This variable is set to true when
450 * it's safe to access per-CPU data.
452 static bool __printk_percpu_data_ready __read_mostly;
454 bool printk_percpu_data_ready(void)
456 return __printk_percpu_data_ready;
459 /* Return log buffer address */
460 char *log_buf_addr_get(void)
465 /* Return log buffer size */
466 u32 log_buf_len_get(void)
472 * Define how much of the log buffer we could take at maximum. The value
473 * must be greater than two. Note that only half of the buffer is available
474 * when the index points to the middle.
476 #define MAX_LOG_TAKE_PART 4
477 static const char trunc_msg[] = "<truncated>";
479 static void truncate_msg(u16 *text_len, u16 *trunc_msg_len)
482 * The message should not take the whole buffer. Otherwise, it might
483 * get removed too soon.
485 u32 max_text_len = log_buf_len / MAX_LOG_TAKE_PART;
487 if (*text_len > max_text_len)
488 *text_len = max_text_len;
490 /* enable the warning message (if there is room) */
491 *trunc_msg_len = strlen(trunc_msg);
492 if (*text_len >= *trunc_msg_len)
493 *text_len -= *trunc_msg_len;
498 /* insert record into the buffer, discard old ones, update heads */
499 static int log_store(u32 caller_id, int facility, int level,
500 enum log_flags flags, u64 ts_nsec,
501 const char *dict, u16 dict_len,
502 const char *text, u16 text_len)
504 struct prb_reserved_entry e;
505 struct printk_record r;
506 u16 trunc_msg_len = 0;
508 prb_rec_init_wr(&r, text_len, dict_len);
510 if (!prb_reserve(&e, prb, &r)) {
511 /* truncate the message if it is too long for empty buffer */
512 truncate_msg(&text_len, &trunc_msg_len);
513 prb_rec_init_wr(&r, text_len + trunc_msg_len, dict_len);
514 /* survive when the log buffer is too small for trunc_msg */
515 if (!prb_reserve(&e, prb, &r))
520 memcpy(&r.text_buf[0], text, text_len);
522 memcpy(&r.text_buf[text_len], trunc_msg, trunc_msg_len);
523 r.info->text_len = text_len + trunc_msg_len;
525 memcpy(&r.dict_buf[0], dict, dict_len);
526 r.info->dict_len = dict_len;
528 r.info->facility = facility;
529 r.info->level = level & 7;
530 r.info->flags = flags & 0x1f;
532 r.info->ts_nsec = ts_nsec;
534 r.info->ts_nsec = local_clock();
535 r.info->caller_id = caller_id;
538 if ((flags & LOG_CONT) || !(flags & LOG_NEWLINE))
541 prb_final_commit(&e);
543 return (text_len + trunc_msg_len);
546 int dmesg_restrict = IS_ENABLED(CONFIG_SECURITY_DMESG_RESTRICT);
548 static int syslog_action_restricted(int type)
553 * Unless restricted, we allow "read all" and "get buffer size"
556 return type != SYSLOG_ACTION_READ_ALL &&
557 type != SYSLOG_ACTION_SIZE_BUFFER;
560 static int check_syslog_permissions(int type, int source)
563 * If this is from /proc/kmsg and we've already opened it, then we've
564 * already done the capabilities checks at open time.
566 if (source == SYSLOG_FROM_PROC && type != SYSLOG_ACTION_OPEN)
569 if (syslog_action_restricted(type)) {
570 if (capable(CAP_SYSLOG))
573 * For historical reasons, accept CAP_SYS_ADMIN too, with
576 if (capable(CAP_SYS_ADMIN)) {
577 pr_warn_once("%s (%d): Attempt to access syslog with "
578 "CAP_SYS_ADMIN but no CAP_SYSLOG "
580 current->comm, task_pid_nr(current));
586 return security_syslog(type);
589 static void append_char(char **pp, char *e, char c)
595 static ssize_t info_print_ext_header(char *buf, size_t size,
596 struct printk_info *info)
598 u64 ts_usec = info->ts_nsec;
600 #ifdef CONFIG_PRINTK_CALLER
601 u32 id = info->caller_id;
603 snprintf(caller, sizeof(caller), ",caller=%c%u",
604 id & 0x80000000 ? 'C' : 'T', id & ~0x80000000);
609 do_div(ts_usec, 1000);
611 return scnprintf(buf, size, "%u,%llu,%llu,%c%s;",
612 (info->facility << 3) | info->level, info->seq,
613 ts_usec, info->flags & LOG_CONT ? 'c' : '-', caller);
616 static ssize_t msg_print_ext_body(char *buf, size_t size,
617 char *dict, size_t dict_len,
618 char *text, size_t text_len)
620 char *p = buf, *e = buf + size;
623 /* escape non-printable characters */
624 for (i = 0; i < text_len; i++) {
625 unsigned char c = text[i];
627 if (c < ' ' || c >= 127 || c == '\\')
628 p += scnprintf(p, e - p, "\\x%02x", c);
630 append_char(&p, e, c);
632 append_char(&p, e, '\n');
637 for (i = 0; i < dict_len; i++) {
638 unsigned char c = dict[i];
641 append_char(&p, e, ' ');
646 append_char(&p, e, '\n');
651 if (c < ' ' || c >= 127 || c == '\\') {
652 p += scnprintf(p, e - p, "\\x%02x", c);
656 append_char(&p, e, c);
658 append_char(&p, e, '\n');
664 /* /dev/kmsg - userspace message inject/listen interface */
665 struct devkmsg_user {
667 struct ratelimit_state rs;
669 char buf[CONSOLE_EXT_LOG_MAX];
671 struct printk_info info;
672 char text_buf[CONSOLE_EXT_LOG_MAX];
673 char dict_buf[CONSOLE_EXT_LOG_MAX];
674 struct printk_record record;
677 static __printf(3, 4) __cold
678 int devkmsg_emit(int facility, int level, const char *fmt, ...)
684 r = vprintk_emit(facility, level, NULL, 0, fmt, args);
690 static ssize_t devkmsg_write(struct kiocb *iocb, struct iov_iter *from)
693 int level = default_message_loglevel;
694 int facility = 1; /* LOG_USER */
695 struct file *file = iocb->ki_filp;
696 struct devkmsg_user *user = file->private_data;
697 size_t len = iov_iter_count(from);
700 if (!user || len > LOG_LINE_MAX)
703 /* Ignore when user logging is disabled. */
704 if (devkmsg_log & DEVKMSG_LOG_MASK_OFF)
707 /* Ratelimit when not explicitly enabled. */
708 if (!(devkmsg_log & DEVKMSG_LOG_MASK_ON)) {
709 if (!___ratelimit(&user->rs, current->comm))
713 buf = kmalloc(len+1, GFP_KERNEL);
718 if (!copy_from_iter_full(buf, len, from)) {
724 * Extract and skip the syslog prefix <[0-9]*>. Coming from userspace
725 * the decimal value represents 32bit, the lower 3 bit are the log
726 * level, the rest are the log facility.
728 * If no prefix or no userspace facility is specified, we
729 * enforce LOG_USER, to be able to reliably distinguish
730 * kernel-generated messages from userspace-injected ones.
733 if (line[0] == '<') {
737 u = simple_strtoul(line + 1, &endp, 10);
738 if (endp && endp[0] == '>') {
739 level = LOG_LEVEL(u);
740 if (LOG_FACILITY(u) != 0)
741 facility = LOG_FACILITY(u);
748 devkmsg_emit(facility, level, "%s", line);
753 static ssize_t devkmsg_read(struct file *file, char __user *buf,
754 size_t count, loff_t *ppos)
756 struct devkmsg_user *user = file->private_data;
757 struct printk_record *r = &user->record;
764 ret = mutex_lock_interruptible(&user->lock);
769 if (!prb_read_valid(prb, user->seq, r)) {
770 if (file->f_flags & O_NONBLOCK) {
777 ret = wait_event_interruptible(log_wait,
778 prb_read_valid(prb, user->seq, r));
784 if (user->seq < prb_first_valid_seq(prb)) {
785 /* our last seen message is gone, return error and reset */
786 user->seq = prb_first_valid_seq(prb);
792 len = info_print_ext_header(user->buf, sizeof(user->buf), r->info);
793 len += msg_print_ext_body(user->buf + len, sizeof(user->buf) - len,
794 &r->dict_buf[0], r->info->dict_len,
795 &r->text_buf[0], r->info->text_len);
797 user->seq = r->info->seq + 1;
805 if (copy_to_user(buf, user->buf, len)) {
811 mutex_unlock(&user->lock);
815 static loff_t devkmsg_llseek(struct file *file, loff_t offset, int whence)
817 struct devkmsg_user *user = file->private_data;
828 /* the first record */
829 user->seq = prb_first_valid_seq(prb);
833 * The first record after the last SYSLOG_ACTION_CLEAR,
834 * like issued by 'dmesg -c'. Reading /dev/kmsg itself
835 * changes no global state, and does not clear anything.
837 user->seq = clear_seq;
840 /* after the last record */
841 user->seq = prb_next_seq(prb);
850 static __poll_t devkmsg_poll(struct file *file, poll_table *wait)
852 struct devkmsg_user *user = file->private_data;
856 return EPOLLERR|EPOLLNVAL;
858 poll_wait(file, &log_wait, wait);
861 if (prb_read_valid(prb, user->seq, NULL)) {
862 /* return error when data has vanished underneath us */
863 if (user->seq < prb_first_valid_seq(prb))
864 ret = EPOLLIN|EPOLLRDNORM|EPOLLERR|EPOLLPRI;
866 ret = EPOLLIN|EPOLLRDNORM;
873 static int devkmsg_open(struct inode *inode, struct file *file)
875 struct devkmsg_user *user;
878 if (devkmsg_log & DEVKMSG_LOG_MASK_OFF)
881 /* write-only does not need any file context */
882 if ((file->f_flags & O_ACCMODE) != O_WRONLY) {
883 err = check_syslog_permissions(SYSLOG_ACTION_READ_ALL,
889 user = kmalloc(sizeof(struct devkmsg_user), GFP_KERNEL);
893 ratelimit_default_init(&user->rs);
894 ratelimit_set_flags(&user->rs, RATELIMIT_MSG_ON_RELEASE);
896 mutex_init(&user->lock);
898 prb_rec_init_rd(&user->record, &user->info,
899 &user->text_buf[0], sizeof(user->text_buf),
900 &user->dict_buf[0], sizeof(user->dict_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 VMCOREINFO_SYMBOL(prb);
945 VMCOREINFO_SYMBOL(printk_rb_static);
946 VMCOREINFO_SYMBOL(clear_seq);
949 * Export struct size and field offsets. User space tools can
950 * parse it and detect any changes to structure down the line.
953 VMCOREINFO_STRUCT_SIZE(printk_ringbuffer);
954 VMCOREINFO_OFFSET(printk_ringbuffer, desc_ring);
955 VMCOREINFO_OFFSET(printk_ringbuffer, text_data_ring);
956 VMCOREINFO_OFFSET(printk_ringbuffer, dict_data_ring);
957 VMCOREINFO_OFFSET(printk_ringbuffer, fail);
959 VMCOREINFO_STRUCT_SIZE(prb_desc_ring);
960 VMCOREINFO_OFFSET(prb_desc_ring, count_bits);
961 VMCOREINFO_OFFSET(prb_desc_ring, descs);
962 VMCOREINFO_OFFSET(prb_desc_ring, head_id);
963 VMCOREINFO_OFFSET(prb_desc_ring, tail_id);
965 VMCOREINFO_STRUCT_SIZE(prb_desc);
966 VMCOREINFO_OFFSET(prb_desc, info);
967 VMCOREINFO_OFFSET(prb_desc, state_var);
968 VMCOREINFO_OFFSET(prb_desc, text_blk_lpos);
969 VMCOREINFO_OFFSET(prb_desc, dict_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, dict_len);
980 VMCOREINFO_OFFSET(printk_info, caller_id);
982 VMCOREINFO_STRUCT_SIZE(prb_data_ring);
983 VMCOREINFO_OFFSET(prb_data_ring, size_bits);
984 VMCOREINFO_OFFSET(prb_data_ring, data);
985 VMCOREINFO_OFFSET(prb_data_ring, head_lpos);
986 VMCOREINFO_OFFSET(prb_data_ring, tail_lpos);
988 VMCOREINFO_SIZE(atomic_long_t);
989 VMCOREINFO_TYPE_OFFSET(atomic_long_t, counter);
993 /* requested log_buf_len from kernel cmdline */
994 static unsigned long __initdata new_log_buf_len;
996 /* we practice scaling the ring buffer by powers of 2 */
997 static void __init log_buf_len_update(u64 size)
999 if (size > (u64)LOG_BUF_LEN_MAX) {
1000 size = (u64)LOG_BUF_LEN_MAX;
1001 pr_err("log_buf over 2G is not supported.\n");
1005 size = roundup_pow_of_two(size);
1006 if (size > log_buf_len)
1007 new_log_buf_len = (unsigned long)size;
1010 /* save requested log_buf_len since it's too early to process it */
1011 static int __init log_buf_len_setup(char *str)
1018 size = memparse(str, &str);
1020 log_buf_len_update(size);
1024 early_param("log_buf_len", log_buf_len_setup);
1027 #define __LOG_CPU_MAX_BUF_LEN (1 << CONFIG_LOG_CPU_MAX_BUF_SHIFT)
1029 static void __init log_buf_add_cpu(void)
1031 unsigned int cpu_extra;
1034 * archs should set up cpu_possible_bits properly with
1035 * set_cpu_possible() after setup_arch() but just in
1036 * case lets ensure this is valid.
1038 if (num_possible_cpus() == 1)
1041 cpu_extra = (num_possible_cpus() - 1) * __LOG_CPU_MAX_BUF_LEN;
1043 /* by default this will only continue through for large > 64 CPUs */
1044 if (cpu_extra <= __LOG_BUF_LEN / 2)
1047 pr_info("log_buf_len individual max cpu contribution: %d bytes\n",
1048 __LOG_CPU_MAX_BUF_LEN);
1049 pr_info("log_buf_len total cpu_extra contributions: %d bytes\n",
1051 pr_info("log_buf_len min size: %d bytes\n", __LOG_BUF_LEN);
1053 log_buf_len_update(cpu_extra + __LOG_BUF_LEN);
1055 #else /* !CONFIG_SMP */
1056 static inline void log_buf_add_cpu(void) {}
1057 #endif /* CONFIG_SMP */
1059 static void __init set_percpu_data_ready(void)
1062 /* Make sure we set this flag only after printk_safe() init is done */
1064 __printk_percpu_data_ready = true;
1067 static unsigned int __init add_to_rb(struct printk_ringbuffer *rb,
1068 struct printk_record *r)
1070 struct prb_reserved_entry e;
1071 struct printk_record dest_r;
1073 prb_rec_init_wr(&dest_r, r->info->text_len, r->info->dict_len);
1075 if (!prb_reserve(&e, rb, &dest_r))
1078 memcpy(&dest_r.text_buf[0], &r->text_buf[0], r->info->text_len);
1079 dest_r.info->text_len = r->info->text_len;
1080 if (dest_r.dict_buf) {
1081 memcpy(&dest_r.dict_buf[0], &r->dict_buf[0], r->info->dict_len);
1082 dest_r.info->dict_len = r->info->dict_len;
1084 dest_r.info->facility = r->info->facility;
1085 dest_r.info->level = r->info->level;
1086 dest_r.info->flags = r->info->flags;
1087 dest_r.info->ts_nsec = r->info->ts_nsec;
1088 dest_r.info->caller_id = r->info->caller_id;
1090 prb_final_commit(&e);
1092 return prb_record_text_space(&e);
1095 static char setup_text_buf[CONSOLE_EXT_LOG_MAX] __initdata;
1096 static char setup_dict_buf[CONSOLE_EXT_LOG_MAX] __initdata;
1098 void __init setup_log_buf(int early)
1100 unsigned int new_descs_count;
1101 struct prb_desc *new_descs;
1102 struct printk_info info;
1103 struct printk_record r;
1104 size_t new_descs_size;
1105 unsigned long flags;
1112 * Some archs call setup_log_buf() multiple times - first is very
1113 * early, e.g. from setup_arch(), and second - when percpu_areas
1117 set_percpu_data_ready();
1119 if (log_buf != __log_buf)
1122 if (!early && !new_log_buf_len)
1125 if (!new_log_buf_len)
1128 new_descs_count = new_log_buf_len >> PRB_AVGBITS;
1129 if (new_descs_count == 0) {
1130 pr_err("new_log_buf_len: %lu too small\n", new_log_buf_len);
1134 new_log_buf = memblock_alloc(new_log_buf_len, LOG_ALIGN);
1135 if (unlikely(!new_log_buf)) {
1136 pr_err("log_buf_len: %lu text bytes not available\n",
1141 new_dict_buf = memblock_alloc(new_log_buf_len, LOG_ALIGN);
1142 if (unlikely(!new_dict_buf)) {
1143 pr_err("log_buf_len: %lu dict bytes not available\n",
1145 memblock_free(__pa(new_log_buf), new_log_buf_len);
1149 new_descs_size = new_descs_count * sizeof(struct prb_desc);
1150 new_descs = memblock_alloc(new_descs_size, LOG_ALIGN);
1151 if (unlikely(!new_descs)) {
1152 pr_err("log_buf_len: %zu desc bytes not available\n",
1154 memblock_free(__pa(new_dict_buf), new_log_buf_len);
1155 memblock_free(__pa(new_log_buf), new_log_buf_len);
1159 prb_rec_init_rd(&r, &info,
1160 &setup_text_buf[0], sizeof(setup_text_buf),
1161 &setup_dict_buf[0], sizeof(setup_dict_buf));
1163 prb_init(&printk_rb_dynamic,
1164 new_log_buf, ilog2(new_log_buf_len),
1165 new_dict_buf, ilog2(new_log_buf_len),
1166 new_descs, ilog2(new_descs_count));
1168 logbuf_lock_irqsave(flags);
1170 log_buf_len = new_log_buf_len;
1171 log_buf = new_log_buf;
1172 new_log_buf_len = 0;
1174 free = __LOG_BUF_LEN;
1175 prb_for_each_record(0, &printk_rb_static, seq, &r)
1176 free -= add_to_rb(&printk_rb_dynamic, &r);
1179 * This is early enough that everything is still running on the
1180 * boot CPU and interrupts are disabled. So no new messages will
1181 * appear during the transition to the dynamic buffer.
1183 prb = &printk_rb_dynamic;
1185 logbuf_unlock_irqrestore(flags);
1187 if (seq != prb_next_seq(&printk_rb_static)) {
1188 pr_err("dropped %llu messages\n",
1189 prb_next_seq(&printk_rb_static) - seq);
1192 pr_info("log_buf_len: %u bytes\n", log_buf_len);
1193 pr_info("early log buf free: %u(%u%%)\n",
1194 free, (free * 100) / __LOG_BUF_LEN);
1197 static bool __read_mostly ignore_loglevel;
1199 static int __init ignore_loglevel_setup(char *str)
1201 ignore_loglevel = true;
1202 pr_info("debug: ignoring loglevel setting.\n");
1207 early_param("ignore_loglevel", ignore_loglevel_setup);
1208 module_param(ignore_loglevel, bool, S_IRUGO | S_IWUSR);
1209 MODULE_PARM_DESC(ignore_loglevel,
1210 "ignore loglevel setting (prints all kernel messages to the console)");
1212 static bool suppress_message_printing(int level)
1214 return (level >= console_loglevel && !ignore_loglevel);
1217 #ifdef CONFIG_BOOT_PRINTK_DELAY
1219 static int boot_delay; /* msecs delay after each printk during bootup */
1220 static unsigned long long loops_per_msec; /* based on boot_delay */
1222 static int __init boot_delay_setup(char *str)
1226 lpj = preset_lpj ? preset_lpj : 1000000; /* some guess */
1227 loops_per_msec = (unsigned long long)lpj / 1000 * HZ;
1229 get_option(&str, &boot_delay);
1230 if (boot_delay > 10 * 1000)
1233 pr_debug("boot_delay: %u, preset_lpj: %ld, lpj: %lu, "
1234 "HZ: %d, loops_per_msec: %llu\n",
1235 boot_delay, preset_lpj, lpj, HZ, loops_per_msec);
1238 early_param("boot_delay", boot_delay_setup);
1240 static void boot_delay_msec(int level)
1242 unsigned long long k;
1243 unsigned long timeout;
1245 if ((boot_delay == 0 || system_state >= SYSTEM_RUNNING)
1246 || suppress_message_printing(level)) {
1250 k = (unsigned long long)loops_per_msec * boot_delay;
1252 timeout = jiffies + msecs_to_jiffies(boot_delay);
1257 * use (volatile) jiffies to prevent
1258 * compiler reduction; loop termination via jiffies
1259 * is secondary and may or may not happen.
1261 if (time_after(jiffies, timeout))
1263 touch_nmi_watchdog();
1267 static inline void boot_delay_msec(int level)
1272 static bool printk_time = IS_ENABLED(CONFIG_PRINTK_TIME);
1273 module_param_named(time, printk_time, bool, S_IRUGO | S_IWUSR);
1275 static size_t print_syslog(unsigned int level, char *buf)
1277 return sprintf(buf, "<%u>", level);
1280 static size_t print_time(u64 ts, char *buf)
1282 unsigned long rem_nsec = do_div(ts, 1000000000);
1284 return sprintf(buf, "[%5lu.%06lu]",
1285 (unsigned long)ts, rem_nsec / 1000);
1288 #ifdef CONFIG_PRINTK_CALLER
1289 static size_t print_caller(u32 id, char *buf)
1293 snprintf(caller, sizeof(caller), "%c%u",
1294 id & 0x80000000 ? 'C' : 'T', id & ~0x80000000);
1295 return sprintf(buf, "[%6s]", caller);
1298 #define print_caller(id, buf) 0
1301 static size_t info_print_prefix(const struct printk_info *info, bool syslog,
1302 bool time, char *buf)
1307 len = print_syslog((info->facility << 3) | info->level, buf);
1310 len += print_time(info->ts_nsec, buf + len);
1312 len += print_caller(info->caller_id, buf + len);
1314 if (IS_ENABLED(CONFIG_PRINTK_CALLER) || time) {
1323 * Prepare the record for printing. The text is shifted within the given
1324 * buffer to avoid a need for another one. The following operations are
1327 * - Add prefix for each line.
1328 * - Add the trailing newline that has been removed in vprintk_store().
1329 * - Drop truncated lines that do not longer fit into the buffer.
1331 * Return: The length of the updated/prepared text, including the added
1332 * prefixes and the newline. The dropped line(s) are not counted.
1334 static size_t record_print_text(struct printk_record *r, bool syslog,
1337 size_t text_len = r->info->text_len;
1338 size_t buf_size = r->text_buf_size;
1339 char *text = r->text_buf;
1340 char prefix[PREFIX_MAX];
1341 bool truncated = false;
1347 prefix_len = info_print_prefix(r->info, syslog, time, prefix);
1350 * @text_len: bytes of unprocessed text
1351 * @line_len: bytes of current line _without_ newline
1352 * @text: pointer to beginning of current line
1353 * @len: number of bytes prepared in r->text_buf
1356 next = memchr(text, '\n', text_len);
1358 line_len = next - text;
1360 /* Drop truncated line(s). */
1363 line_len = text_len;
1367 * Truncate the text if there is not enough space to add the
1368 * prefix and a trailing newline.
1370 if (len + prefix_len + text_len + 1 > buf_size) {
1371 /* Drop even the current line if no space. */
1372 if (len + prefix_len + line_len + 1 > buf_size)
1375 text_len = buf_size - len - prefix_len - 1;
1379 memmove(text + prefix_len, text, text_len);
1380 memcpy(text, prefix, prefix_len);
1382 len += prefix_len + line_len + 1;
1384 if (text_len == line_len) {
1386 * Add the trailing newline removed in
1389 text[prefix_len + line_len] = '\n';
1394 * Advance beyond the added prefix and the related line with
1397 text += prefix_len + line_len + 1;
1400 * The remaining text has only decreased by the line with its
1403 * Note that @text_len can become zero. It happens when @text
1404 * ended with a newline (either due to truncation or the
1405 * original string ending with "\n\n"). The loop is correctly
1406 * repeated and (if not truncated) an empty line with a prefix
1409 text_len -= line_len + 1;
1415 static size_t get_record_print_text_size(struct printk_info *info,
1416 unsigned int line_count,
1417 bool syslog, bool time)
1419 char prefix[PREFIX_MAX];
1422 prefix_len = info_print_prefix(info, syslog, time, prefix);
1425 * Each line will be preceded with a prefix. The intermediate
1426 * newlines are already within the text, but a final trailing
1427 * newline will be added.
1429 return ((prefix_len * line_count) + info->text_len + 1);
1432 static int syslog_print(char __user *buf, int size)
1434 struct printk_info info;
1435 struct printk_record r;
1439 text = kmalloc(LOG_LINE_MAX + PREFIX_MAX, GFP_KERNEL);
1443 prb_rec_init_rd(&r, &info, text, LOG_LINE_MAX + PREFIX_MAX, NULL, 0);
1450 if (!prb_read_valid(prb, syslog_seq, &r)) {
1451 logbuf_unlock_irq();
1454 if (r.info->seq != syslog_seq) {
1455 /* message is gone, move to next valid one */
1456 syslog_seq = r.info->seq;
1461 * To keep reading/counting partial line consistent,
1462 * use printk_time value as of the beginning of a line.
1464 if (!syslog_partial)
1465 syslog_time = printk_time;
1467 skip = syslog_partial;
1468 n = record_print_text(&r, true, syslog_time);
1469 if (n - syslog_partial <= size) {
1470 /* message fits into buffer, move forward */
1471 syslog_seq = r.info->seq + 1;
1472 n -= syslog_partial;
1475 /* partial read(), remember position */
1477 syslog_partial += n;
1480 logbuf_unlock_irq();
1485 if (copy_to_user(buf, text + skip, n)) {
1500 static int syslog_print_all(char __user *buf, int size, bool clear)
1502 struct printk_info info;
1503 unsigned int line_count;
1504 struct printk_record r;
1510 text = kmalloc(LOG_LINE_MAX + PREFIX_MAX, GFP_KERNEL);
1517 * Find first record that fits, including all following records,
1518 * into the user-provided buffer for this dump.
1520 prb_for_each_info(clear_seq, prb, seq, &info, &line_count)
1521 len += get_record_print_text_size(&info, line_count, true, time);
1523 /* move first record forward until length fits into the buffer */
1524 prb_for_each_info(clear_seq, prb, seq, &info, &line_count) {
1527 len -= get_record_print_text_size(&info, line_count, true, time);
1530 prb_rec_init_rd(&r, &info, text, LOG_LINE_MAX + PREFIX_MAX, NULL, 0);
1533 prb_for_each_record(seq, prb, seq, &r) {
1536 textlen = record_print_text(&r, true, time);
1538 if (len + textlen > size) {
1543 logbuf_unlock_irq();
1544 if (copy_to_user(buf + len, text, textlen))
1556 logbuf_unlock_irq();
1562 static void syslog_clear(void)
1565 clear_seq = prb_next_seq(prb);
1566 logbuf_unlock_irq();
1569 int do_syslog(int type, char __user *buf, int len, int source)
1572 static int saved_console_loglevel = LOGLEVEL_DEFAULT;
1575 error = check_syslog_permissions(type, source);
1580 case SYSLOG_ACTION_CLOSE: /* Close log */
1582 case SYSLOG_ACTION_OPEN: /* Open log */
1584 case SYSLOG_ACTION_READ: /* Read from log */
1585 if (!buf || len < 0)
1589 if (!access_ok(buf, len))
1591 error = wait_event_interruptible(log_wait,
1592 prb_read_valid(prb, syslog_seq, NULL));
1595 error = syslog_print(buf, len);
1597 /* Read/clear last kernel messages */
1598 case SYSLOG_ACTION_READ_CLEAR:
1601 /* Read last kernel messages */
1602 case SYSLOG_ACTION_READ_ALL:
1603 if (!buf || len < 0)
1607 if (!access_ok(buf, len))
1609 error = syslog_print_all(buf, len, clear);
1611 /* Clear ring buffer */
1612 case SYSLOG_ACTION_CLEAR:
1615 /* Disable logging to console */
1616 case SYSLOG_ACTION_CONSOLE_OFF:
1617 if (saved_console_loglevel == LOGLEVEL_DEFAULT)
1618 saved_console_loglevel = console_loglevel;
1619 console_loglevel = minimum_console_loglevel;
1621 /* Enable logging to console */
1622 case SYSLOG_ACTION_CONSOLE_ON:
1623 if (saved_console_loglevel != LOGLEVEL_DEFAULT) {
1624 console_loglevel = saved_console_loglevel;
1625 saved_console_loglevel = LOGLEVEL_DEFAULT;
1628 /* Set level of messages printed to console */
1629 case SYSLOG_ACTION_CONSOLE_LEVEL:
1630 if (len < 1 || len > 8)
1632 if (len < minimum_console_loglevel)
1633 len = minimum_console_loglevel;
1634 console_loglevel = len;
1635 /* Implicitly re-enable logging to console */
1636 saved_console_loglevel = LOGLEVEL_DEFAULT;
1638 /* Number of chars in the log buffer */
1639 case SYSLOG_ACTION_SIZE_UNREAD:
1641 if (syslog_seq < prb_first_valid_seq(prb)) {
1642 /* messages are gone, move to first one */
1643 syslog_seq = prb_first_valid_seq(prb);
1646 if (source == SYSLOG_FROM_PROC) {
1648 * Short-cut for poll(/"proc/kmsg") which simply checks
1649 * for pending data, not the size; return the count of
1650 * records, not the length.
1652 error = prb_next_seq(prb) - syslog_seq;
1654 bool time = syslog_partial ? syslog_time : printk_time;
1655 struct printk_info info;
1656 unsigned int line_count;
1659 prb_for_each_info(syslog_seq, prb, seq, &info,
1661 error += get_record_print_text_size(&info, line_count,
1665 error -= syslog_partial;
1667 logbuf_unlock_irq();
1669 /* Size of the log buffer */
1670 case SYSLOG_ACTION_SIZE_BUFFER:
1671 error = log_buf_len;
1681 SYSCALL_DEFINE3(syslog, int, type, char __user *, buf, int, len)
1683 return do_syslog(type, buf, len, SYSLOG_FROM_READER);
1687 * Special console_lock variants that help to reduce the risk of soft-lockups.
1688 * They allow to pass console_lock to another printk() call using a busy wait.
1691 #ifdef CONFIG_LOCKDEP
1692 static struct lockdep_map console_owner_dep_map = {
1693 .name = "console_owner"
1697 static DEFINE_RAW_SPINLOCK(console_owner_lock);
1698 static struct task_struct *console_owner;
1699 static bool console_waiter;
1702 * console_lock_spinning_enable - mark beginning of code where another
1703 * thread might safely busy wait
1705 * This basically converts console_lock into a spinlock. This marks
1706 * the section where the console_lock owner can not sleep, because
1707 * there may be a waiter spinning (like a spinlock). Also it must be
1708 * ready to hand over the lock at the end of the section.
1710 static void console_lock_spinning_enable(void)
1712 raw_spin_lock(&console_owner_lock);
1713 console_owner = current;
1714 raw_spin_unlock(&console_owner_lock);
1716 /* The waiter may spin on us after setting console_owner */
1717 spin_acquire(&console_owner_dep_map, 0, 0, _THIS_IP_);
1721 * console_lock_spinning_disable_and_check - mark end of code where another
1722 * thread was able to busy wait and check if there is a waiter
1724 * This is called at the end of the section where spinning is allowed.
1725 * It has two functions. First, it is a signal that it is no longer
1726 * safe to start busy waiting for the lock. Second, it checks if
1727 * there is a busy waiter and passes the lock rights to her.
1729 * Important: Callers lose the lock if there was a busy waiter.
1730 * They must not touch items synchronized by console_lock
1733 * Return: 1 if the lock rights were passed, 0 otherwise.
1735 static int console_lock_spinning_disable_and_check(void)
1739 raw_spin_lock(&console_owner_lock);
1740 waiter = READ_ONCE(console_waiter);
1741 console_owner = NULL;
1742 raw_spin_unlock(&console_owner_lock);
1745 spin_release(&console_owner_dep_map, _THIS_IP_);
1749 /* The waiter is now free to continue */
1750 WRITE_ONCE(console_waiter, false);
1752 spin_release(&console_owner_dep_map, _THIS_IP_);
1755 * Hand off console_lock to waiter. The waiter will perform
1756 * the up(). After this, the waiter is the console_lock owner.
1758 mutex_release(&console_lock_dep_map, _THIS_IP_);
1763 * console_trylock_spinning - try to get console_lock by busy waiting
1765 * This allows to busy wait for the console_lock when the current
1766 * owner is running in specially marked sections. It means that
1767 * the current owner is running and cannot reschedule until it
1768 * is ready to lose the lock.
1770 * Return: 1 if we got the lock, 0 othrewise
1772 static int console_trylock_spinning(void)
1774 struct task_struct *owner = NULL;
1777 unsigned long flags;
1779 if (console_trylock())
1782 printk_safe_enter_irqsave(flags);
1784 raw_spin_lock(&console_owner_lock);
1785 owner = READ_ONCE(console_owner);
1786 waiter = READ_ONCE(console_waiter);
1787 if (!waiter && owner && owner != current) {
1788 WRITE_ONCE(console_waiter, true);
1791 raw_spin_unlock(&console_owner_lock);
1794 * If there is an active printk() writing to the
1795 * consoles, instead of having it write our data too,
1796 * see if we can offload that load from the active
1797 * printer, and do some printing ourselves.
1798 * Go into a spin only if there isn't already a waiter
1799 * spinning, and there is an active printer, and
1800 * that active printer isn't us (recursive printk?).
1803 printk_safe_exit_irqrestore(flags);
1807 /* We spin waiting for the owner to release us */
1808 spin_acquire(&console_owner_dep_map, 0, 0, _THIS_IP_);
1809 /* Owner will clear console_waiter on hand off */
1810 while (READ_ONCE(console_waiter))
1812 spin_release(&console_owner_dep_map, _THIS_IP_);
1814 printk_safe_exit_irqrestore(flags);
1816 * The owner passed the console lock to us.
1817 * Since we did not spin on console lock, annotate
1818 * this as a trylock. Otherwise lockdep will
1821 mutex_acquire(&console_lock_dep_map, 0, 1, _THIS_IP_);
1827 * Call the console drivers, asking them to write out
1828 * log_buf[start] to log_buf[end - 1].
1829 * The console_lock must be held.
1831 static void call_console_drivers(const char *ext_text, size_t ext_len,
1832 const char *text, size_t len)
1834 static char dropped_text[64];
1835 size_t dropped_len = 0;
1836 struct console *con;
1838 trace_console_rcuidle(text, len);
1840 if (!console_drivers)
1843 if (console_dropped) {
1844 dropped_len = snprintf(dropped_text, sizeof(dropped_text),
1845 "** %lu printk messages dropped **\n",
1847 console_dropped = 0;
1850 for_each_console(con) {
1851 if (exclusive_console && con != exclusive_console)
1853 if (!(con->flags & CON_ENABLED))
1857 if (!cpu_online(smp_processor_id()) &&
1858 !(con->flags & CON_ANYTIME))
1860 if (con->flags & CON_EXTENDED)
1861 con->write(con, ext_text, ext_len);
1864 con->write(con, dropped_text, dropped_len);
1865 con->write(con, text, len);
1870 int printk_delay_msec __read_mostly;
1872 static inline void printk_delay(void)
1874 if (unlikely(printk_delay_msec)) {
1875 int m = printk_delay_msec;
1879 touch_nmi_watchdog();
1884 static inline u32 printk_caller_id(void)
1886 return in_task() ? task_pid_nr(current) :
1887 0x80000000 + raw_smp_processor_id();
1890 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)
1892 const u32 caller_id = printk_caller_id();
1894 if (lflags & LOG_CONT) {
1895 struct prb_reserved_entry e;
1896 struct printk_record r;
1898 prb_rec_init_wr(&r, text_len, 0);
1899 if (prb_reserve_in_last(&e, prb, &r, caller_id)) {
1900 memcpy(&r.text_buf[r.info->text_len], text, text_len);
1901 r.info->text_len += text_len;
1902 if (lflags & LOG_NEWLINE) {
1903 r.info->flags |= LOG_NEWLINE;
1904 prb_final_commit(&e);
1912 /* Store it in the record log */
1913 return log_store(caller_id, facility, level, lflags, 0,
1914 dict, dictlen, text, text_len);
1917 /* Must be called under logbuf_lock. */
1918 int vprintk_store(int facility, int level,
1919 const char *dict, size_t dictlen,
1920 const char *fmt, va_list args)
1922 static char textbuf[LOG_LINE_MAX];
1923 char *text = textbuf;
1925 enum log_flags lflags = 0;
1928 * The printf needs to come first; we need the syslog
1929 * prefix which might be passed-in as a parameter.
1931 text_len = vscnprintf(text, sizeof(textbuf), fmt, args);
1933 /* mark and strip a trailing newline */
1934 if (text_len && text[text_len-1] == '\n') {
1936 lflags |= LOG_NEWLINE;
1939 /* strip kernel syslog prefix and extract log level or control flags */
1940 if (facility == 0) {
1943 while ((kern_level = printk_get_level(text)) != 0) {
1944 switch (kern_level) {
1946 if (level == LOGLEVEL_DEFAULT)
1947 level = kern_level - '0';
1949 case 'c': /* KERN_CONT */
1958 if (level == LOGLEVEL_DEFAULT)
1959 level = default_message_loglevel;
1962 lflags |= LOG_NEWLINE;
1964 return log_output(facility, level, lflags,
1965 dict, dictlen, text, text_len);
1968 asmlinkage int vprintk_emit(int facility, int level,
1969 const char *dict, size_t dictlen,
1970 const char *fmt, va_list args)
1973 bool in_sched = false;
1974 unsigned long flags;
1976 /* Suppress unimportant messages after panic happens */
1977 if (unlikely(suppress_printk))
1980 if (level == LOGLEVEL_SCHED) {
1981 level = LOGLEVEL_DEFAULT;
1985 boot_delay_msec(level);
1988 /* This stops the holder of console_sem just where we want him */
1989 logbuf_lock_irqsave(flags);
1990 printed_len = vprintk_store(facility, level, dict, dictlen, fmt, args);
1991 logbuf_unlock_irqrestore(flags);
1993 /* If called from the scheduler, we can not call up(). */
1996 * Disable preemption to avoid being preempted while holding
1997 * console_sem which would prevent anyone from printing to
2002 * Try to acquire and then immediately release the console
2003 * semaphore. The release will print out buffers and wake up
2004 * /dev/kmsg and syslog() users.
2006 if (console_trylock_spinning())
2014 EXPORT_SYMBOL(vprintk_emit);
2016 asmlinkage int vprintk(const char *fmt, va_list args)
2018 return vprintk_func(fmt, args);
2020 EXPORT_SYMBOL(vprintk);
2022 int vprintk_default(const char *fmt, va_list args)
2024 return vprintk_emit(0, LOGLEVEL_DEFAULT, NULL, 0, fmt, args);
2026 EXPORT_SYMBOL_GPL(vprintk_default);
2029 * printk - print a kernel message
2030 * @fmt: format string
2032 * This is printk(). It can be called from any context. We want it to work.
2034 * We try to grab the console_lock. If we succeed, it's easy - we log the
2035 * output and call the console drivers. If we fail to get the semaphore, we
2036 * place the output into the log buffer and return. The current holder of
2037 * the console_sem will notice the new output in console_unlock(); and will
2038 * send it to the consoles before releasing the lock.
2040 * One effect of this deferred printing is that code which calls printk() and
2041 * then changes console_loglevel may break. This is because console_loglevel
2042 * is inspected when the actual printing occurs.
2047 * See the vsnprintf() documentation for format string extensions over C99.
2049 asmlinkage __visible int printk(const char *fmt, ...)
2054 va_start(args, fmt);
2055 r = vprintk_func(fmt, args);
2060 EXPORT_SYMBOL(printk);
2062 #else /* CONFIG_PRINTK */
2064 #define LOG_LINE_MAX 0
2065 #define PREFIX_MAX 0
2066 #define printk_time false
2068 #define prb_read_valid(rb, seq, r) false
2069 #define prb_first_valid_seq(rb) 0
2071 static u64 syslog_seq;
2072 static u64 console_seq;
2073 static u64 exclusive_console_stop_seq;
2074 static unsigned long console_dropped;
2076 static size_t record_print_text(const struct printk_record *r,
2077 bool syslog, bool time)
2081 static ssize_t info_print_ext_header(char *buf, size_t size,
2082 struct printk_info *info)
2086 static ssize_t msg_print_ext_body(char *buf, size_t size,
2087 char *dict, size_t dict_len,
2088 char *text, size_t text_len) { return 0; }
2089 static void console_lock_spinning_enable(void) { }
2090 static int console_lock_spinning_disable_and_check(void) { return 0; }
2091 static void call_console_drivers(const char *ext_text, size_t ext_len,
2092 const char *text, size_t len) {}
2093 static bool suppress_message_printing(int level) { return false; }
2095 #endif /* CONFIG_PRINTK */
2097 #ifdef CONFIG_EARLY_PRINTK
2098 struct console *early_console;
2100 asmlinkage __visible void early_printk(const char *fmt, ...)
2110 n = vscnprintf(buf, sizeof(buf), fmt, ap);
2113 early_console->write(early_console, buf, n);
2117 static int __add_preferred_console(char *name, int idx, char *options,
2118 char *brl_options, bool user_specified)
2120 struct console_cmdline *c;
2124 * See if this tty is not yet registered, and
2125 * if we have a slot free.
2127 for (i = 0, c = console_cmdline;
2128 i < MAX_CMDLINECONSOLES && c->name[0];
2130 if (strcmp(c->name, name) == 0 && c->index == idx) {
2132 preferred_console = i;
2134 c->user_specified = true;
2138 if (i == MAX_CMDLINECONSOLES)
2141 preferred_console = i;
2142 strlcpy(c->name, name, sizeof(c->name));
2143 c->options = options;
2144 c->user_specified = user_specified;
2145 braille_set_options(c, brl_options);
2151 static int __init console_msg_format_setup(char *str)
2153 if (!strcmp(str, "syslog"))
2154 console_msg_format = MSG_FORMAT_SYSLOG;
2155 if (!strcmp(str, "default"))
2156 console_msg_format = MSG_FORMAT_DEFAULT;
2159 __setup("console_msg_format=", console_msg_format_setup);
2162 * Set up a console. Called via do_early_param() in init/main.c
2163 * for each "console=" parameter in the boot command line.
2165 static int __init console_setup(char *str)
2167 char buf[sizeof(console_cmdline[0].name) + 4]; /* 4 for "ttyS" */
2168 char *s, *options, *brl_options = NULL;
2174 if (_braille_console_setup(&str, &brl_options))
2178 * Decode str into name, index, options.
2180 if (str[0] >= '0' && str[0] <= '9') {
2181 strcpy(buf, "ttyS");
2182 strncpy(buf + 4, str, sizeof(buf) - 5);
2184 strncpy(buf, str, sizeof(buf) - 1);
2186 buf[sizeof(buf) - 1] = 0;
2187 options = strchr(str, ',');
2191 if (!strcmp(str, "ttya"))
2192 strcpy(buf, "ttyS0");
2193 if (!strcmp(str, "ttyb"))
2194 strcpy(buf, "ttyS1");
2196 for (s = buf; *s; s++)
2197 if (isdigit(*s) || *s == ',')
2199 idx = simple_strtoul(s, NULL, 10);
2202 __add_preferred_console(buf, idx, options, brl_options, true);
2203 console_set_on_cmdline = 1;
2206 __setup("console=", console_setup);
2209 * add_preferred_console - add a device to the list of preferred consoles.
2210 * @name: device name
2211 * @idx: device index
2212 * @options: options for this console
2214 * The last preferred console added will be used for kernel messages
2215 * and stdin/out/err for init. Normally this is used by console_setup
2216 * above to handle user-supplied console arguments; however it can also
2217 * be used by arch-specific code either to override the user or more
2218 * commonly to provide a default console (ie from PROM variables) when
2219 * the user has not supplied one.
2221 int add_preferred_console(char *name, int idx, char *options)
2223 return __add_preferred_console(name, idx, options, NULL, false);
2226 bool console_suspend_enabled = true;
2227 EXPORT_SYMBOL(console_suspend_enabled);
2229 static int __init console_suspend_disable(char *str)
2231 console_suspend_enabled = false;
2234 __setup("no_console_suspend", console_suspend_disable);
2235 module_param_named(console_suspend, console_suspend_enabled,
2236 bool, S_IRUGO | S_IWUSR);
2237 MODULE_PARM_DESC(console_suspend, "suspend console during suspend"
2238 " and hibernate operations");
2241 * suspend_console - suspend the console subsystem
2243 * This disables printk() while we go into suspend states
2245 void suspend_console(void)
2247 if (!console_suspend_enabled)
2249 pr_info("Suspending console(s) (use no_console_suspend to debug)\n");
2251 console_suspended = 1;
2255 void resume_console(void)
2257 if (!console_suspend_enabled)
2260 console_suspended = 0;
2265 * console_cpu_notify - print deferred console messages after CPU hotplug
2268 * If printk() is called from a CPU that is not online yet, the messages
2269 * will be printed on the console only if there are CON_ANYTIME consoles.
2270 * This function is called when a new CPU comes online (or fails to come
2271 * up) or goes offline.
2273 static int console_cpu_notify(unsigned int cpu)
2275 if (!cpuhp_tasks_frozen) {
2276 /* If trylock fails, someone else is doing the printing */
2277 if (console_trylock())
2284 * console_lock - lock the console system for exclusive use.
2286 * Acquires a lock which guarantees that the caller has
2287 * exclusive access to the console system and the console_drivers list.
2289 * Can sleep, returns nothing.
2291 void console_lock(void)
2296 if (console_suspended)
2299 console_may_schedule = 1;
2301 EXPORT_SYMBOL(console_lock);
2304 * console_trylock - try to lock the console system for exclusive use.
2306 * Try to acquire a lock which guarantees that the caller has exclusive
2307 * access to the console system and the console_drivers list.
2309 * returns 1 on success, and 0 on failure to acquire the lock.
2311 int console_trylock(void)
2313 if (down_trylock_console_sem())
2315 if (console_suspended) {
2320 console_may_schedule = 0;
2323 EXPORT_SYMBOL(console_trylock);
2325 int is_console_locked(void)
2327 return console_locked;
2329 EXPORT_SYMBOL(is_console_locked);
2332 * Check if we have any console that is capable of printing while cpu is
2333 * booting or shutting down. Requires console_sem.
2335 static int have_callable_console(void)
2337 struct console *con;
2339 for_each_console(con)
2340 if ((con->flags & CON_ENABLED) &&
2341 (con->flags & CON_ANYTIME))
2348 * Can we actually use the console at this time on this cpu?
2350 * Console drivers may assume that per-cpu resources have been allocated. So
2351 * unless they're explicitly marked as being able to cope (CON_ANYTIME) don't
2352 * call them until this CPU is officially up.
2354 static inline int can_use_console(void)
2356 return cpu_online(raw_smp_processor_id()) || have_callable_console();
2360 * console_unlock - unlock the console system
2362 * Releases the console_lock which the caller holds on the console system
2363 * and the console driver list.
2365 * While the console_lock was held, console output may have been buffered
2366 * by printk(). If this is the case, console_unlock(); emits
2367 * the output prior to releasing the lock.
2369 * If there is output waiting, we wake /dev/kmsg and syslog() users.
2371 * console_unlock(); may be called from any context.
2373 void console_unlock(void)
2375 static char ext_text[CONSOLE_EXT_LOG_MAX];
2376 static char text[LOG_LINE_MAX + PREFIX_MAX];
2377 static char dict[LOG_LINE_MAX];
2378 unsigned long flags;
2379 bool do_cond_resched, retry;
2380 struct printk_info info;
2381 struct printk_record r;
2383 if (console_suspended) {
2388 prb_rec_init_rd(&r, &info, text, sizeof(text), dict, sizeof(dict));
2391 * Console drivers are called with interrupts disabled, so
2392 * @console_may_schedule should be cleared before; however, we may
2393 * end up dumping a lot of lines, for example, if called from
2394 * console registration path, and should invoke cond_resched()
2395 * between lines if allowable. Not doing so can cause a very long
2396 * scheduling stall on a slow console leading to RCU stall and
2397 * softlockup warnings which exacerbate the issue with more
2398 * messages practically incapacitating the system.
2400 * console_trylock() is not able to detect the preemptive
2401 * context reliably. Therefore the value must be stored before
2402 * and cleared after the the "again" goto label.
2404 do_cond_resched = console_may_schedule;
2406 console_may_schedule = 0;
2409 * We released the console_sem lock, so we need to recheck if
2410 * cpu is online and (if not) is there at least one CON_ANYTIME
2413 if (!can_use_console()) {
2423 printk_safe_enter_irqsave(flags);
2424 raw_spin_lock(&logbuf_lock);
2426 if (!prb_read_valid(prb, console_seq, &r))
2429 if (console_seq != r.info->seq) {
2430 console_dropped += r.info->seq - console_seq;
2431 console_seq = r.info->seq;
2434 if (suppress_message_printing(r.info->level)) {
2436 * Skip record we have buffered and already printed
2437 * directly to the console when we received it, and
2438 * record that has level above the console loglevel.
2444 /* Output to all consoles once old messages replayed. */
2445 if (unlikely(exclusive_console &&
2446 console_seq >= exclusive_console_stop_seq)) {
2447 exclusive_console = NULL;
2451 * Handle extended console text first because later
2452 * record_print_text() will modify the record buffer in-place.
2454 if (nr_ext_console_drivers) {
2455 ext_len = info_print_ext_header(ext_text,
2458 ext_len += msg_print_ext_body(ext_text + ext_len,
2459 sizeof(ext_text) - ext_len,
2465 len = record_print_text(&r,
2466 console_msg_format & MSG_FORMAT_SYSLOG,
2469 raw_spin_unlock(&logbuf_lock);
2472 * While actively printing out messages, if another printk()
2473 * were to occur on another CPU, it may wait for this one to
2474 * finish. This task can not be preempted if there is a
2475 * waiter waiting to take over.
2477 console_lock_spinning_enable();
2479 stop_critical_timings(); /* don't trace print latency */
2480 call_console_drivers(ext_text, ext_len, text, len);
2481 start_critical_timings();
2483 if (console_lock_spinning_disable_and_check()) {
2484 printk_safe_exit_irqrestore(flags);
2488 printk_safe_exit_irqrestore(flags);
2490 if (do_cond_resched)
2496 raw_spin_unlock(&logbuf_lock);
2501 * Someone could have filled up the buffer again, so re-check if there's
2502 * something to flush. In case we cannot trylock the console_sem again,
2503 * there's a new owner and the console_unlock() from them will do the
2504 * flush, no worries.
2506 raw_spin_lock(&logbuf_lock);
2507 retry = prb_read_valid(prb, console_seq, NULL);
2508 raw_spin_unlock(&logbuf_lock);
2509 printk_safe_exit_irqrestore(flags);
2511 if (retry && console_trylock())
2514 EXPORT_SYMBOL(console_unlock);
2517 * console_conditional_schedule - yield the CPU if required
2519 * If the console code is currently allowed to sleep, and
2520 * if this CPU should yield the CPU to another task, do
2523 * Must be called within console_lock();.
2525 void __sched console_conditional_schedule(void)
2527 if (console_may_schedule)
2530 EXPORT_SYMBOL(console_conditional_schedule);
2532 void console_unblank(void)
2537 * console_unblank can no longer be called in interrupt context unless
2538 * oops_in_progress is set to 1..
2540 if (oops_in_progress) {
2541 if (down_trylock_console_sem() != 0)
2547 console_may_schedule = 0;
2549 if ((c->flags & CON_ENABLED) && c->unblank)
2555 * console_flush_on_panic - flush console content on panic
2556 * @mode: flush all messages in buffer or just the pending ones
2558 * Immediately output all pending messages no matter what.
2560 void console_flush_on_panic(enum con_flush_mode mode)
2563 * If someone else is holding the console lock, trylock will fail
2564 * and may_schedule may be set. Ignore and proceed to unlock so
2565 * that messages are flushed out. As this can be called from any
2566 * context and we don't want to get preempted while flushing,
2567 * ensure may_schedule is cleared.
2570 console_may_schedule = 0;
2572 if (mode == CONSOLE_REPLAY_ALL) {
2573 unsigned long flags;
2575 logbuf_lock_irqsave(flags);
2576 console_seq = prb_first_valid_seq(prb);
2577 logbuf_unlock_irqrestore(flags);
2583 * Return the console tty driver structure and its associated index
2585 struct tty_driver *console_device(int *index)
2588 struct tty_driver *driver = NULL;
2591 for_each_console(c) {
2594 driver = c->device(c, index);
2603 * Prevent further output on the passed console device so that (for example)
2604 * serial drivers can disable console output before suspending a port, and can
2605 * re-enable output afterwards.
2607 void console_stop(struct console *console)
2610 console->flags &= ~CON_ENABLED;
2613 EXPORT_SYMBOL(console_stop);
2615 void console_start(struct console *console)
2618 console->flags |= CON_ENABLED;
2621 EXPORT_SYMBOL(console_start);
2623 static int __read_mostly keep_bootcon;
2625 static int __init keep_bootcon_setup(char *str)
2628 pr_info("debug: skip boot console de-registration.\n");
2633 early_param("keep_bootcon", keep_bootcon_setup);
2636 * This is called by register_console() to try to match
2637 * the newly registered console with any of the ones selected
2638 * by either the command line or add_preferred_console() and
2641 * Care need to be taken with consoles that are statically
2642 * enabled such as netconsole
2644 static int try_enable_new_console(struct console *newcon, bool user_specified)
2646 struct console_cmdline *c;
2649 for (i = 0, c = console_cmdline;
2650 i < MAX_CMDLINECONSOLES && c->name[0];
2652 if (c->user_specified != user_specified)
2654 if (!newcon->match ||
2655 newcon->match(newcon, c->name, c->index, c->options) != 0) {
2656 /* default matching */
2657 BUILD_BUG_ON(sizeof(c->name) != sizeof(newcon->name));
2658 if (strcmp(c->name, newcon->name) != 0)
2660 if (newcon->index >= 0 &&
2661 newcon->index != c->index)
2663 if (newcon->index < 0)
2664 newcon->index = c->index;
2666 if (_braille_register_console(newcon, c))
2669 if (newcon->setup &&
2670 newcon->setup(newcon, c->options) != 0)
2673 newcon->flags |= CON_ENABLED;
2674 if (i == preferred_console) {
2675 newcon->flags |= CON_CONSDEV;
2676 has_preferred_console = true;
2682 * Some consoles, such as pstore and netconsole, can be enabled even
2683 * without matching. Accept the pre-enabled consoles only when match()
2684 * and setup() had a change to be called.
2686 if (newcon->flags & CON_ENABLED && c->user_specified == user_specified)
2693 * The console driver calls this routine during kernel initialization
2694 * to register the console printing procedure with printk() and to
2695 * print any messages that were printed by the kernel before the
2696 * console driver was initialized.
2698 * This can happen pretty early during the boot process (because of
2699 * early_printk) - sometimes before setup_arch() completes - be careful
2700 * of what kernel features are used - they may not be initialised yet.
2702 * There are two types of consoles - bootconsoles (early_printk) and
2703 * "real" consoles (everything which is not a bootconsole) which are
2704 * handled differently.
2705 * - Any number of bootconsoles can be registered at any time.
2706 * - As soon as a "real" console is registered, all bootconsoles
2707 * will be unregistered automatically.
2708 * - Once a "real" console is registered, any attempt to register a
2709 * bootconsoles will be rejected
2711 void register_console(struct console *newcon)
2713 unsigned long flags;
2714 struct console *bcon = NULL;
2717 for_each_console(bcon) {
2718 if (WARN(bcon == newcon, "console '%s%d' already registered\n",
2719 bcon->name, bcon->index))
2724 * before we register a new CON_BOOT console, make sure we don't
2725 * already have a valid console
2727 if (newcon->flags & CON_BOOT) {
2728 for_each_console(bcon) {
2729 if (!(bcon->flags & CON_BOOT)) {
2730 pr_info("Too late to register bootconsole %s%d\n",
2731 newcon->name, newcon->index);
2737 if (console_drivers && console_drivers->flags & CON_BOOT)
2738 bcon = console_drivers;
2740 if (!has_preferred_console || bcon || !console_drivers)
2741 has_preferred_console = preferred_console >= 0;
2744 * See if we want to use this console driver. If we
2745 * didn't select a console we take the first one
2746 * that registers here.
2748 if (!has_preferred_console) {
2749 if (newcon->index < 0)
2751 if (newcon->setup == NULL ||
2752 newcon->setup(newcon, NULL) == 0) {
2753 newcon->flags |= CON_ENABLED;
2754 if (newcon->device) {
2755 newcon->flags |= CON_CONSDEV;
2756 has_preferred_console = true;
2761 /* See if this console matches one we selected on the command line */
2762 err = try_enable_new_console(newcon, true);
2764 /* If not, try to match against the platform default(s) */
2766 err = try_enable_new_console(newcon, false);
2768 /* printk() messages are not printed to the Braille console. */
2769 if (err || newcon->flags & CON_BRL)
2773 * If we have a bootconsole, and are switching to a real console,
2774 * don't print everything out again, since when the boot console, and
2775 * the real console are the same physical device, it's annoying to
2776 * see the beginning boot messages twice
2778 if (bcon && ((newcon->flags & (CON_CONSDEV | CON_BOOT)) == CON_CONSDEV))
2779 newcon->flags &= ~CON_PRINTBUFFER;
2782 * Put this console in the list - keep the
2783 * preferred driver at the head of the list.
2786 if ((newcon->flags & CON_CONSDEV) || console_drivers == NULL) {
2787 newcon->next = console_drivers;
2788 console_drivers = newcon;
2790 newcon->next->flags &= ~CON_CONSDEV;
2791 /* Ensure this flag is always set for the head of the list */
2792 newcon->flags |= CON_CONSDEV;
2794 newcon->next = console_drivers->next;
2795 console_drivers->next = newcon;
2798 if (newcon->flags & CON_EXTENDED)
2799 nr_ext_console_drivers++;
2801 if (newcon->flags & CON_PRINTBUFFER) {
2803 * console_unlock(); will print out the buffered messages
2806 logbuf_lock_irqsave(flags);
2808 * We're about to replay the log buffer. Only do this to the
2809 * just-registered console to avoid excessive message spam to
2810 * the already-registered consoles.
2812 * Set exclusive_console with disabled interrupts to reduce
2813 * race window with eventual console_flush_on_panic() that
2814 * ignores console_lock.
2816 exclusive_console = newcon;
2817 exclusive_console_stop_seq = console_seq;
2818 console_seq = syslog_seq;
2819 logbuf_unlock_irqrestore(flags);
2822 console_sysfs_notify();
2825 * By unregistering the bootconsoles after we enable the real console
2826 * we get the "console xxx enabled" message on all the consoles -
2827 * boot consoles, real consoles, etc - this is to ensure that end
2828 * users know there might be something in the kernel's log buffer that
2829 * went to the bootconsole (that they do not see on the real console)
2831 pr_info("%sconsole [%s%d] enabled\n",
2832 (newcon->flags & CON_BOOT) ? "boot" : "" ,
2833 newcon->name, newcon->index);
2835 ((newcon->flags & (CON_CONSDEV | CON_BOOT)) == CON_CONSDEV) &&
2837 /* We need to iterate through all boot consoles, to make
2838 * sure we print everything out, before we unregister them.
2840 for_each_console(bcon)
2841 if (bcon->flags & CON_BOOT)
2842 unregister_console(bcon);
2845 EXPORT_SYMBOL(register_console);
2847 int unregister_console(struct console *console)
2849 struct console *con;
2852 pr_info("%sconsole [%s%d] disabled\n",
2853 (console->flags & CON_BOOT) ? "boot" : "" ,
2854 console->name, console->index);
2856 res = _braille_unregister_console(console);
2864 if (console_drivers == console) {
2865 console_drivers=console->next;
2868 for_each_console(con) {
2869 if (con->next == console) {
2870 con->next = console->next;
2878 goto out_disable_unlock;
2880 if (console->flags & CON_EXTENDED)
2881 nr_ext_console_drivers--;
2884 * If this isn't the last console and it has CON_CONSDEV set, we
2885 * need to set it on the next preferred console.
2887 if (console_drivers != NULL && console->flags & CON_CONSDEV)
2888 console_drivers->flags |= CON_CONSDEV;
2890 console->flags &= ~CON_ENABLED;
2892 console_sysfs_notify();
2895 res = console->exit(console);
2900 console->flags &= ~CON_ENABLED;
2905 EXPORT_SYMBOL(unregister_console);
2908 * Initialize the console device. This is called *early*, so
2909 * we can't necessarily depend on lots of kernel help here.
2910 * Just do some early initializations, and do the complex setup
2913 void __init console_init(void)
2917 initcall_entry_t *ce;
2919 /* Setup the default TTY line discipline. */
2923 * set up the console device so that later boot sequences can
2924 * inform about problems etc..
2926 ce = __con_initcall_start;
2927 trace_initcall_level("console");
2928 while (ce < __con_initcall_end) {
2929 call = initcall_from_entry(ce);
2930 trace_initcall_start(call);
2932 trace_initcall_finish(call, ret);
2938 * Some boot consoles access data that is in the init section and which will
2939 * be discarded after the initcalls have been run. To make sure that no code
2940 * will access this data, unregister the boot consoles in a late initcall.
2942 * If for some reason, such as deferred probe or the driver being a loadable
2943 * module, the real console hasn't registered yet at this point, there will
2944 * be a brief interval in which no messages are logged to the console, which
2945 * makes it difficult to diagnose problems that occur during this time.
2947 * To mitigate this problem somewhat, only unregister consoles whose memory
2948 * intersects with the init section. Note that all other boot consoles will
2949 * get unregistred when the real preferred console is registered.
2951 static int __init printk_late_init(void)
2953 struct console *con;
2956 for_each_console(con) {
2957 if (!(con->flags & CON_BOOT))
2960 /* Check addresses that might be used for enabled consoles. */
2961 if (init_section_intersects(con, sizeof(*con)) ||
2962 init_section_contains(con->write, 0) ||
2963 init_section_contains(con->read, 0) ||
2964 init_section_contains(con->device, 0) ||
2965 init_section_contains(con->unblank, 0) ||
2966 init_section_contains(con->data, 0)) {
2968 * Please, consider moving the reported consoles out
2969 * of the init section.
2971 pr_warn("bootconsole [%s%d] uses init memory and must be disabled even before the real one is ready\n",
2972 con->name, con->index);
2973 unregister_console(con);
2976 ret = cpuhp_setup_state_nocalls(CPUHP_PRINTK_DEAD, "printk:dead", NULL,
2977 console_cpu_notify);
2979 ret = cpuhp_setup_state_nocalls(CPUHP_AP_ONLINE_DYN, "printk:online",
2980 console_cpu_notify, NULL);
2984 late_initcall(printk_late_init);
2986 #if defined CONFIG_PRINTK
2988 * Delayed printk version, for scheduler-internal messages:
2990 #define PRINTK_PENDING_WAKEUP 0x01
2991 #define PRINTK_PENDING_OUTPUT 0x02
2993 static DEFINE_PER_CPU(int, printk_pending);
2995 static void wake_up_klogd_work_func(struct irq_work *irq_work)
2997 int pending = __this_cpu_xchg(printk_pending, 0);
2999 if (pending & PRINTK_PENDING_OUTPUT) {
3000 /* If trylock fails, someone else is doing the printing */
3001 if (console_trylock())
3005 if (pending & PRINTK_PENDING_WAKEUP)
3006 wake_up_interruptible(&log_wait);
3009 static DEFINE_PER_CPU(struct irq_work, wake_up_klogd_work) = {
3010 .func = wake_up_klogd_work_func,
3011 .flags = ATOMIC_INIT(IRQ_WORK_LAZY),
3014 void wake_up_klogd(void)
3016 if (!printk_percpu_data_ready())
3020 if (waitqueue_active(&log_wait)) {
3021 this_cpu_or(printk_pending, PRINTK_PENDING_WAKEUP);
3022 irq_work_queue(this_cpu_ptr(&wake_up_klogd_work));
3027 void defer_console_output(void)
3029 if (!printk_percpu_data_ready())
3033 __this_cpu_or(printk_pending, PRINTK_PENDING_OUTPUT);
3034 irq_work_queue(this_cpu_ptr(&wake_up_klogd_work));
3038 int vprintk_deferred(const char *fmt, va_list args)
3042 r = vprintk_emit(0, LOGLEVEL_SCHED, NULL, 0, fmt, args);
3043 defer_console_output();
3048 int printk_deferred(const char *fmt, ...)
3053 va_start(args, fmt);
3054 r = vprintk_deferred(fmt, args);
3061 * printk rate limiting, lifted from the networking subsystem.
3063 * This enforces a rate limit: not more than 10 kernel messages
3064 * every 5s to make a denial-of-service attack impossible.
3066 DEFINE_RATELIMIT_STATE(printk_ratelimit_state, 5 * HZ, 10);
3068 int __printk_ratelimit(const char *func)
3070 return ___ratelimit(&printk_ratelimit_state, func);
3072 EXPORT_SYMBOL(__printk_ratelimit);
3075 * printk_timed_ratelimit - caller-controlled printk ratelimiting
3076 * @caller_jiffies: pointer to caller's state
3077 * @interval_msecs: minimum interval between prints
3079 * printk_timed_ratelimit() returns true if more than @interval_msecs
3080 * milliseconds have elapsed since the last time printk_timed_ratelimit()
3083 bool printk_timed_ratelimit(unsigned long *caller_jiffies,
3084 unsigned int interval_msecs)
3086 unsigned long elapsed = jiffies - *caller_jiffies;
3088 if (*caller_jiffies && elapsed <= msecs_to_jiffies(interval_msecs))
3091 *caller_jiffies = jiffies;
3094 EXPORT_SYMBOL(printk_timed_ratelimit);
3096 static DEFINE_SPINLOCK(dump_list_lock);
3097 static LIST_HEAD(dump_list);
3100 * kmsg_dump_register - register a kernel log dumper.
3101 * @dumper: pointer to the kmsg_dumper structure
3103 * Adds a kernel log dumper to the system. The dump callback in the
3104 * structure will be called when the kernel oopses or panics and must be
3105 * set. Returns zero on success and %-EINVAL or %-EBUSY otherwise.
3107 int kmsg_dump_register(struct kmsg_dumper *dumper)
3109 unsigned long flags;
3112 /* The dump callback needs to be set */
3116 spin_lock_irqsave(&dump_list_lock, flags);
3117 /* Don't allow registering multiple times */
3118 if (!dumper->registered) {
3119 dumper->registered = 1;
3120 list_add_tail_rcu(&dumper->list, &dump_list);
3123 spin_unlock_irqrestore(&dump_list_lock, flags);
3127 EXPORT_SYMBOL_GPL(kmsg_dump_register);
3130 * kmsg_dump_unregister - unregister a kmsg dumper.
3131 * @dumper: pointer to the kmsg_dumper structure
3133 * Removes a dump device from the system. Returns zero on success and
3134 * %-EINVAL otherwise.
3136 int kmsg_dump_unregister(struct kmsg_dumper *dumper)
3138 unsigned long flags;
3141 spin_lock_irqsave(&dump_list_lock, flags);
3142 if (dumper->registered) {
3143 dumper->registered = 0;
3144 list_del_rcu(&dumper->list);
3147 spin_unlock_irqrestore(&dump_list_lock, flags);
3152 EXPORT_SYMBOL_GPL(kmsg_dump_unregister);
3154 static bool always_kmsg_dump;
3155 module_param_named(always_kmsg_dump, always_kmsg_dump, bool, S_IRUGO | S_IWUSR);
3157 const char *kmsg_dump_reason_str(enum kmsg_dump_reason reason)
3160 case KMSG_DUMP_PANIC:
3162 case KMSG_DUMP_OOPS:
3164 case KMSG_DUMP_EMERG:
3166 case KMSG_DUMP_SHUTDOWN:
3172 EXPORT_SYMBOL_GPL(kmsg_dump_reason_str);
3175 * kmsg_dump - dump kernel log to kernel message dumpers.
3176 * @reason: the reason (oops, panic etc) for dumping
3178 * Call each of the registered dumper's dump() callback, which can
3179 * retrieve the kmsg records with kmsg_dump_get_line() or
3180 * kmsg_dump_get_buffer().
3182 void kmsg_dump(enum kmsg_dump_reason reason)
3184 struct kmsg_dumper *dumper;
3185 unsigned long flags;
3188 list_for_each_entry_rcu(dumper, &dump_list, list) {
3189 enum kmsg_dump_reason max_reason = dumper->max_reason;
3192 * If client has not provided a specific max_reason, default
3193 * to KMSG_DUMP_OOPS, unless always_kmsg_dump was set.
3195 if (max_reason == KMSG_DUMP_UNDEF) {
3196 max_reason = always_kmsg_dump ? KMSG_DUMP_MAX :
3199 if (reason > max_reason)
3202 /* initialize iterator with data about the stored records */
3203 dumper->active = true;
3205 logbuf_lock_irqsave(flags);
3206 dumper->cur_seq = clear_seq;
3207 dumper->next_seq = prb_next_seq(prb);
3208 logbuf_unlock_irqrestore(flags);
3210 /* invoke dumper which will iterate over records */
3211 dumper->dump(dumper, reason);
3213 /* reset iterator */
3214 dumper->active = false;
3220 * kmsg_dump_get_line_nolock - retrieve one kmsg log line (unlocked version)
3221 * @dumper: registered kmsg dumper
3222 * @syslog: include the "<4>" prefixes
3223 * @line: buffer to copy the line to
3224 * @size: maximum size of the buffer
3225 * @len: length of line placed into buffer
3227 * Start at the beginning of the kmsg buffer, with the oldest kmsg
3228 * record, and copy one record into the provided buffer.
3230 * Consecutive calls will return the next available record moving
3231 * towards the end of the buffer with the youngest messages.
3233 * A return value of FALSE indicates that there are no more records to
3236 * The function is similar to kmsg_dump_get_line(), but grabs no locks.
3238 bool kmsg_dump_get_line_nolock(struct kmsg_dumper *dumper, bool syslog,
3239 char *line, size_t size, size_t *len)
3241 struct printk_info info;
3242 unsigned int line_count;
3243 struct printk_record r;
3247 prb_rec_init_rd(&r, &info, line, size, NULL, 0);
3249 if (!dumper->active)
3252 /* Read text or count text lines? */
3254 if (!prb_read_valid(prb, dumper->cur_seq, &r))
3256 l = record_print_text(&r, syslog, printk_time);
3258 if (!prb_read_valid_info(prb, dumper->cur_seq,
3259 &info, &line_count)) {
3262 l = get_record_print_text_size(&info, line_count, syslog,
3267 dumper->cur_seq = r.info->seq + 1;
3276 * kmsg_dump_get_line - retrieve one kmsg log line
3277 * @dumper: registered kmsg dumper
3278 * @syslog: include the "<4>" prefixes
3279 * @line: buffer to copy the line to
3280 * @size: maximum size of the buffer
3281 * @len: length of line placed into buffer
3283 * Start at the beginning of the kmsg buffer, with the oldest kmsg
3284 * record, and copy one record into the provided buffer.
3286 * Consecutive calls will return the next available record moving
3287 * towards the end of the buffer with the youngest messages.
3289 * A return value of FALSE indicates that there are no more records to
3292 bool kmsg_dump_get_line(struct kmsg_dumper *dumper, bool syslog,
3293 char *line, size_t size, size_t *len)
3295 unsigned long flags;
3298 logbuf_lock_irqsave(flags);
3299 ret = kmsg_dump_get_line_nolock(dumper, syslog, line, size, len);
3300 logbuf_unlock_irqrestore(flags);
3304 EXPORT_SYMBOL_GPL(kmsg_dump_get_line);
3307 * kmsg_dump_get_buffer - copy kmsg log lines
3308 * @dumper: registered kmsg dumper
3309 * @syslog: include the "<4>" prefixes
3310 * @buf: buffer to copy the line to
3311 * @size: maximum size of the buffer
3312 * @len: length of line placed into buffer
3314 * Start at the end of the kmsg buffer and fill the provided buffer
3315 * with as many of the the *youngest* kmsg records that fit into it.
3316 * If the buffer is large enough, all available kmsg records will be
3317 * copied with a single call.
3319 * Consecutive calls will fill the buffer with the next block of
3320 * available older records, not including the earlier retrieved ones.
3322 * A return value of FALSE indicates that there are no more records to
3325 bool kmsg_dump_get_buffer(struct kmsg_dumper *dumper, bool syslog,
3326 char *buf, size_t size, size_t *len)
3328 struct printk_info info;
3329 unsigned int line_count;
3330 struct printk_record r;
3331 unsigned long flags;
3336 bool time = printk_time;
3338 prb_rec_init_rd(&r, &info, buf, size, NULL, 0);
3340 if (!dumper->active || !buf || !size)
3343 logbuf_lock_irqsave(flags);
3344 if (dumper->cur_seq < prb_first_valid_seq(prb)) {
3345 /* messages are gone, move to first available one */
3346 dumper->cur_seq = prb_first_valid_seq(prb);
3350 if (dumper->cur_seq >= dumper->next_seq) {
3351 logbuf_unlock_irqrestore(flags);
3355 /* calculate length of entire buffer */
3356 seq = dumper->cur_seq;
3357 while (prb_read_valid_info(prb, seq, &info, &line_count)) {
3358 if (r.info->seq >= dumper->next_seq)
3360 l += get_record_print_text_size(&info, line_count, true, time);
3361 seq = r.info->seq + 1;
3364 /* move first record forward until length fits into the buffer */
3365 seq = dumper->cur_seq;
3366 while (l >= size && prb_read_valid_info(prb, seq,
3367 &info, &line_count)) {
3368 if (r.info->seq >= dumper->next_seq)
3370 l -= get_record_print_text_size(&info, line_count, true, time);
3371 seq = r.info->seq + 1;
3374 /* last message in next interation */
3377 /* actually read text into the buffer now */
3379 while (prb_read_valid(prb, seq, &r)) {
3380 if (r.info->seq >= dumper->next_seq)
3383 l += record_print_text(&r, syslog, time);
3385 /* adjust record to store to remaining buffer space */
3386 prb_rec_init_rd(&r, &info, buf + l, size - l, NULL, 0);
3388 seq = r.info->seq + 1;
3391 dumper->next_seq = next_seq;
3393 logbuf_unlock_irqrestore(flags);
3399 EXPORT_SYMBOL_GPL(kmsg_dump_get_buffer);
3402 * kmsg_dump_rewind_nolock - reset the iterator (unlocked version)
3403 * @dumper: registered kmsg dumper
3405 * Reset the dumper's iterator so that kmsg_dump_get_line() and
3406 * kmsg_dump_get_buffer() can be called again and used multiple
3407 * times within the same dumper.dump() callback.
3409 * The function is similar to kmsg_dump_rewind(), but grabs no locks.
3411 void kmsg_dump_rewind_nolock(struct kmsg_dumper *dumper)
3413 dumper->cur_seq = clear_seq;
3414 dumper->next_seq = prb_next_seq(prb);
3418 * kmsg_dump_rewind - reset the iterator
3419 * @dumper: registered kmsg dumper
3421 * Reset the dumper's iterator so that kmsg_dump_get_line() and
3422 * kmsg_dump_get_buffer() can be called again and used multiple
3423 * times within the same dumper.dump() callback.
3425 void kmsg_dump_rewind(struct kmsg_dumper *dumper)
3427 unsigned long flags;
3429 logbuf_lock_irqsave(flags);
3430 kmsg_dump_rewind_nolock(dumper);
3431 logbuf_unlock_irqrestore(flags);
3433 EXPORT_SYMBOL_GPL(kmsg_dump_rewind);