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 int dmesg_restrict = IS_ENABLED(CONFIG_SECURITY_DMESG_RESTRICT);
496 static int syslog_action_restricted(int type)
501 * Unless restricted, we allow "read all" and "get buffer size"
504 return type != SYSLOG_ACTION_READ_ALL &&
505 type != SYSLOG_ACTION_SIZE_BUFFER;
508 static int check_syslog_permissions(int type, int source)
511 * If this is from /proc/kmsg and we've already opened it, then we've
512 * already done the capabilities checks at open time.
514 if (source == SYSLOG_FROM_PROC && type != SYSLOG_ACTION_OPEN)
517 if (syslog_action_restricted(type)) {
518 if (capable(CAP_SYSLOG))
521 * For historical reasons, accept CAP_SYS_ADMIN too, with
524 if (capable(CAP_SYS_ADMIN)) {
525 pr_warn_once("%s (%d): Attempt to access syslog with "
526 "CAP_SYS_ADMIN but no CAP_SYSLOG "
528 current->comm, task_pid_nr(current));
534 return security_syslog(type);
537 static void append_char(char **pp, char *e, char c)
543 static ssize_t info_print_ext_header(char *buf, size_t size,
544 struct printk_info *info)
546 u64 ts_usec = info->ts_nsec;
548 #ifdef CONFIG_PRINTK_CALLER
549 u32 id = info->caller_id;
551 snprintf(caller, sizeof(caller), ",caller=%c%u",
552 id & 0x80000000 ? 'C' : 'T', id & ~0x80000000);
557 do_div(ts_usec, 1000);
559 return scnprintf(buf, size, "%u,%llu,%llu,%c%s;",
560 (info->facility << 3) | info->level, info->seq,
561 ts_usec, info->flags & LOG_CONT ? 'c' : '-', caller);
564 static ssize_t msg_add_ext_text(char *buf, size_t size,
565 const char *text, size_t text_len,
568 char *p = buf, *e = buf + size;
571 /* escape non-printable characters */
572 for (i = 0; i < text_len; i++) {
573 unsigned char c = text[i];
575 if (c < ' ' || c >= 127 || c == '\\')
576 p += scnprintf(p, e - p, "\\x%02x", c);
578 append_char(&p, e, c);
580 append_char(&p, e, endc);
585 static ssize_t msg_add_dict_text(char *buf, size_t size,
586 const char *key, const char *val)
588 size_t val_len = strlen(val);
594 len = msg_add_ext_text(buf, size, "", 0, ' '); /* dict prefix */
595 len += msg_add_ext_text(buf + len, size - len, key, strlen(key), '=');
596 len += msg_add_ext_text(buf + len, size - len, val, val_len, '\n');
601 static ssize_t msg_print_ext_body(char *buf, size_t size,
602 char *text, size_t text_len,
603 struct dev_printk_info *dev_info)
607 len = msg_add_ext_text(buf, size, text, text_len, '\n');
612 len += msg_add_dict_text(buf + len, size - len, "SUBSYSTEM",
613 dev_info->subsystem);
614 len += msg_add_dict_text(buf + len, size - len, "DEVICE",
620 /* /dev/kmsg - userspace message inject/listen interface */
621 struct devkmsg_user {
623 struct ratelimit_state rs;
625 char buf[CONSOLE_EXT_LOG_MAX];
627 struct printk_info info;
628 char text_buf[CONSOLE_EXT_LOG_MAX];
629 struct printk_record record;
632 static __printf(3, 4) __cold
633 int devkmsg_emit(int facility, int level, const char *fmt, ...)
639 r = vprintk_emit(facility, level, NULL, fmt, args);
645 static ssize_t devkmsg_write(struct kiocb *iocb, struct iov_iter *from)
648 int level = default_message_loglevel;
649 int facility = 1; /* LOG_USER */
650 struct file *file = iocb->ki_filp;
651 struct devkmsg_user *user = file->private_data;
652 size_t len = iov_iter_count(from);
655 if (!user || len > LOG_LINE_MAX)
658 /* Ignore when user logging is disabled. */
659 if (devkmsg_log & DEVKMSG_LOG_MASK_OFF)
662 /* Ratelimit when not explicitly enabled. */
663 if (!(devkmsg_log & DEVKMSG_LOG_MASK_ON)) {
664 if (!___ratelimit(&user->rs, current->comm))
668 buf = kmalloc(len+1, GFP_KERNEL);
673 if (!copy_from_iter_full(buf, len, from)) {
679 * Extract and skip the syslog prefix <[0-9]*>. Coming from userspace
680 * the decimal value represents 32bit, the lower 3 bit are the log
681 * level, the rest are the log facility.
683 * If no prefix or no userspace facility is specified, we
684 * enforce LOG_USER, to be able to reliably distinguish
685 * kernel-generated messages from userspace-injected ones.
688 if (line[0] == '<') {
692 u = simple_strtoul(line + 1, &endp, 10);
693 if (endp && endp[0] == '>') {
694 level = LOG_LEVEL(u);
695 if (LOG_FACILITY(u) != 0)
696 facility = LOG_FACILITY(u);
702 devkmsg_emit(facility, level, "%s", line);
707 static ssize_t devkmsg_read(struct file *file, char __user *buf,
708 size_t count, loff_t *ppos)
710 struct devkmsg_user *user = file->private_data;
711 struct printk_record *r = &user->record;
718 ret = mutex_lock_interruptible(&user->lock);
723 if (!prb_read_valid(prb, user->seq, r)) {
724 if (file->f_flags & O_NONBLOCK) {
731 ret = wait_event_interruptible(log_wait,
732 prb_read_valid(prb, user->seq, r));
738 if (user->seq < prb_first_valid_seq(prb)) {
739 /* our last seen message is gone, return error and reset */
740 user->seq = prb_first_valid_seq(prb);
746 len = info_print_ext_header(user->buf, sizeof(user->buf), r->info);
747 len += msg_print_ext_body(user->buf + len, sizeof(user->buf) - len,
748 &r->text_buf[0], r->info->text_len,
751 user->seq = r->info->seq + 1;
759 if (copy_to_user(buf, user->buf, len)) {
765 mutex_unlock(&user->lock);
770 * Be careful when modifying this function!!!
772 * Only few operations are supported because the device works only with the
773 * entire variable length messages (records). Non-standard values are
774 * returned in the other cases and has been this way for quite some time.
775 * User space applications might depend on this behavior.
777 static loff_t devkmsg_llseek(struct file *file, loff_t offset, int whence)
779 struct devkmsg_user *user = file->private_data;
790 /* the first record */
791 user->seq = prb_first_valid_seq(prb);
795 * The first record after the last SYSLOG_ACTION_CLEAR,
796 * like issued by 'dmesg -c'. Reading /dev/kmsg itself
797 * changes no global state, and does not clear anything.
799 user->seq = clear_seq;
802 /* after the last record */
803 user->seq = prb_next_seq(prb);
812 static __poll_t devkmsg_poll(struct file *file, poll_table *wait)
814 struct devkmsg_user *user = file->private_data;
818 return EPOLLERR|EPOLLNVAL;
820 poll_wait(file, &log_wait, wait);
823 if (prb_read_valid(prb, user->seq, NULL)) {
824 /* return error when data has vanished underneath us */
825 if (user->seq < prb_first_valid_seq(prb))
826 ret = EPOLLIN|EPOLLRDNORM|EPOLLERR|EPOLLPRI;
828 ret = EPOLLIN|EPOLLRDNORM;
835 static int devkmsg_open(struct inode *inode, struct file *file)
837 struct devkmsg_user *user;
840 if (devkmsg_log & DEVKMSG_LOG_MASK_OFF)
843 /* write-only does not need any file context */
844 if ((file->f_flags & O_ACCMODE) != O_WRONLY) {
845 err = check_syslog_permissions(SYSLOG_ACTION_READ_ALL,
851 user = kmalloc(sizeof(struct devkmsg_user), GFP_KERNEL);
855 ratelimit_default_init(&user->rs);
856 ratelimit_set_flags(&user->rs, RATELIMIT_MSG_ON_RELEASE);
858 mutex_init(&user->lock);
860 prb_rec_init_rd(&user->record, &user->info,
861 &user->text_buf[0], sizeof(user->text_buf));
864 user->seq = prb_first_valid_seq(prb);
867 file->private_data = user;
871 static int devkmsg_release(struct inode *inode, struct file *file)
873 struct devkmsg_user *user = file->private_data;
878 ratelimit_state_exit(&user->rs);
880 mutex_destroy(&user->lock);
885 const struct file_operations kmsg_fops = {
886 .open = devkmsg_open,
887 .read = devkmsg_read,
888 .write_iter = devkmsg_write,
889 .llseek = devkmsg_llseek,
890 .poll = devkmsg_poll,
891 .release = devkmsg_release,
894 #ifdef CONFIG_CRASH_CORE
896 * This appends the listed symbols to /proc/vmcore
898 * /proc/vmcore is used by various utilities, like crash and makedumpfile to
899 * obtain access to symbols that are otherwise very difficult to locate. These
900 * symbols are specifically used so that utilities can access and extract the
901 * dmesg log from a vmcore file after a crash.
903 void log_buf_vmcoreinfo_setup(void)
905 struct dev_printk_info *dev_info = NULL;
907 VMCOREINFO_SYMBOL(prb);
908 VMCOREINFO_SYMBOL(printk_rb_static);
909 VMCOREINFO_SYMBOL(clear_seq);
912 * Export struct size and field offsets. User space tools can
913 * parse it and detect any changes to structure down the line.
916 VMCOREINFO_STRUCT_SIZE(printk_ringbuffer);
917 VMCOREINFO_OFFSET(printk_ringbuffer, desc_ring);
918 VMCOREINFO_OFFSET(printk_ringbuffer, text_data_ring);
919 VMCOREINFO_OFFSET(printk_ringbuffer, fail);
921 VMCOREINFO_STRUCT_SIZE(prb_desc_ring);
922 VMCOREINFO_OFFSET(prb_desc_ring, count_bits);
923 VMCOREINFO_OFFSET(prb_desc_ring, descs);
924 VMCOREINFO_OFFSET(prb_desc_ring, infos);
925 VMCOREINFO_OFFSET(prb_desc_ring, head_id);
926 VMCOREINFO_OFFSET(prb_desc_ring, tail_id);
928 VMCOREINFO_STRUCT_SIZE(prb_desc);
929 VMCOREINFO_OFFSET(prb_desc, state_var);
930 VMCOREINFO_OFFSET(prb_desc, text_blk_lpos);
932 VMCOREINFO_STRUCT_SIZE(prb_data_blk_lpos);
933 VMCOREINFO_OFFSET(prb_data_blk_lpos, begin);
934 VMCOREINFO_OFFSET(prb_data_blk_lpos, next);
936 VMCOREINFO_STRUCT_SIZE(printk_info);
937 VMCOREINFO_OFFSET(printk_info, seq);
938 VMCOREINFO_OFFSET(printk_info, ts_nsec);
939 VMCOREINFO_OFFSET(printk_info, text_len);
940 VMCOREINFO_OFFSET(printk_info, caller_id);
941 VMCOREINFO_OFFSET(printk_info, dev_info);
943 VMCOREINFO_STRUCT_SIZE(dev_printk_info);
944 VMCOREINFO_OFFSET(dev_printk_info, subsystem);
945 VMCOREINFO_LENGTH(printk_info_subsystem, sizeof(dev_info->subsystem));
946 VMCOREINFO_OFFSET(dev_printk_info, device);
947 VMCOREINFO_LENGTH(printk_info_device, sizeof(dev_info->device));
949 VMCOREINFO_STRUCT_SIZE(prb_data_ring);
950 VMCOREINFO_OFFSET(prb_data_ring, size_bits);
951 VMCOREINFO_OFFSET(prb_data_ring, data);
952 VMCOREINFO_OFFSET(prb_data_ring, head_lpos);
953 VMCOREINFO_OFFSET(prb_data_ring, tail_lpos);
955 VMCOREINFO_SIZE(atomic_long_t);
956 VMCOREINFO_TYPE_OFFSET(atomic_long_t, counter);
960 /* requested log_buf_len from kernel cmdline */
961 static unsigned long __initdata new_log_buf_len;
963 /* we practice scaling the ring buffer by powers of 2 */
964 static void __init log_buf_len_update(u64 size)
966 if (size > (u64)LOG_BUF_LEN_MAX) {
967 size = (u64)LOG_BUF_LEN_MAX;
968 pr_err("log_buf over 2G is not supported.\n");
972 size = roundup_pow_of_two(size);
973 if (size > log_buf_len)
974 new_log_buf_len = (unsigned long)size;
977 /* save requested log_buf_len since it's too early to process it */
978 static int __init log_buf_len_setup(char *str)
985 size = memparse(str, &str);
987 log_buf_len_update(size);
991 early_param("log_buf_len", log_buf_len_setup);
994 #define __LOG_CPU_MAX_BUF_LEN (1 << CONFIG_LOG_CPU_MAX_BUF_SHIFT)
996 static void __init log_buf_add_cpu(void)
998 unsigned int cpu_extra;
1001 * archs should set up cpu_possible_bits properly with
1002 * set_cpu_possible() after setup_arch() but just in
1003 * case lets ensure this is valid.
1005 if (num_possible_cpus() == 1)
1008 cpu_extra = (num_possible_cpus() - 1) * __LOG_CPU_MAX_BUF_LEN;
1010 /* by default this will only continue through for large > 64 CPUs */
1011 if (cpu_extra <= __LOG_BUF_LEN / 2)
1014 pr_info("log_buf_len individual max cpu contribution: %d bytes\n",
1015 __LOG_CPU_MAX_BUF_LEN);
1016 pr_info("log_buf_len total cpu_extra contributions: %d bytes\n",
1018 pr_info("log_buf_len min size: %d bytes\n", __LOG_BUF_LEN);
1020 log_buf_len_update(cpu_extra + __LOG_BUF_LEN);
1022 #else /* !CONFIG_SMP */
1023 static inline void log_buf_add_cpu(void) {}
1024 #endif /* CONFIG_SMP */
1026 static void __init set_percpu_data_ready(void)
1029 /* Make sure we set this flag only after printk_safe() init is done */
1031 __printk_percpu_data_ready = true;
1034 static unsigned int __init add_to_rb(struct printk_ringbuffer *rb,
1035 struct printk_record *r)
1037 struct prb_reserved_entry e;
1038 struct printk_record dest_r;
1040 prb_rec_init_wr(&dest_r, r->info->text_len);
1042 if (!prb_reserve(&e, rb, &dest_r))
1045 memcpy(&dest_r.text_buf[0], &r->text_buf[0], r->info->text_len);
1046 dest_r.info->text_len = r->info->text_len;
1047 dest_r.info->facility = r->info->facility;
1048 dest_r.info->level = r->info->level;
1049 dest_r.info->flags = r->info->flags;
1050 dest_r.info->ts_nsec = r->info->ts_nsec;
1051 dest_r.info->caller_id = r->info->caller_id;
1052 memcpy(&dest_r.info->dev_info, &r->info->dev_info, sizeof(dest_r.info->dev_info));
1054 prb_final_commit(&e);
1056 return prb_record_text_space(&e);
1059 static char setup_text_buf[LOG_LINE_MAX] __initdata;
1061 void __init setup_log_buf(int early)
1063 struct printk_info *new_infos;
1064 unsigned int new_descs_count;
1065 struct prb_desc *new_descs;
1066 struct printk_info info;
1067 struct printk_record r;
1068 size_t new_descs_size;
1069 size_t new_infos_size;
1070 unsigned long flags;
1076 * Some archs call setup_log_buf() multiple times - first is very
1077 * early, e.g. from setup_arch(), and second - when percpu_areas
1081 set_percpu_data_ready();
1083 if (log_buf != __log_buf)
1086 if (!early && !new_log_buf_len)
1089 if (!new_log_buf_len)
1092 new_descs_count = new_log_buf_len >> PRB_AVGBITS;
1093 if (new_descs_count == 0) {
1094 pr_err("new_log_buf_len: %lu too small\n", new_log_buf_len);
1098 new_log_buf = memblock_alloc(new_log_buf_len, LOG_ALIGN);
1099 if (unlikely(!new_log_buf)) {
1100 pr_err("log_buf_len: %lu text bytes not available\n",
1105 new_descs_size = new_descs_count * sizeof(struct prb_desc);
1106 new_descs = memblock_alloc(new_descs_size, LOG_ALIGN);
1107 if (unlikely(!new_descs)) {
1108 pr_err("log_buf_len: %zu desc bytes not available\n",
1110 goto err_free_log_buf;
1113 new_infos_size = new_descs_count * sizeof(struct printk_info);
1114 new_infos = memblock_alloc(new_infos_size, LOG_ALIGN);
1115 if (unlikely(!new_infos)) {
1116 pr_err("log_buf_len: %zu info bytes not available\n",
1118 goto err_free_descs;
1121 prb_rec_init_rd(&r, &info, &setup_text_buf[0], sizeof(setup_text_buf));
1123 prb_init(&printk_rb_dynamic,
1124 new_log_buf, ilog2(new_log_buf_len),
1125 new_descs, ilog2(new_descs_count),
1128 printk_safe_enter_irqsave(flags);
1130 log_buf_len = new_log_buf_len;
1131 log_buf = new_log_buf;
1132 new_log_buf_len = 0;
1134 free = __LOG_BUF_LEN;
1135 prb_for_each_record(0, &printk_rb_static, seq, &r)
1136 free -= add_to_rb(&printk_rb_dynamic, &r);
1139 * This is early enough that everything is still running on the
1140 * boot CPU and interrupts are disabled. So no new messages will
1141 * appear during the transition to the dynamic buffer.
1143 prb = &printk_rb_dynamic;
1145 printk_safe_exit_irqrestore(flags);
1147 if (seq != prb_next_seq(&printk_rb_static)) {
1148 pr_err("dropped %llu messages\n",
1149 prb_next_seq(&printk_rb_static) - seq);
1152 pr_info("log_buf_len: %u bytes\n", log_buf_len);
1153 pr_info("early log buf free: %u(%u%%)\n",
1154 free, (free * 100) / __LOG_BUF_LEN);
1158 memblock_free(__pa(new_descs), new_descs_size);
1160 memblock_free(__pa(new_log_buf), new_log_buf_len);
1163 static bool __read_mostly ignore_loglevel;
1165 static int __init ignore_loglevel_setup(char *str)
1167 ignore_loglevel = true;
1168 pr_info("debug: ignoring loglevel setting.\n");
1173 early_param("ignore_loglevel", ignore_loglevel_setup);
1174 module_param(ignore_loglevel, bool, S_IRUGO | S_IWUSR);
1175 MODULE_PARM_DESC(ignore_loglevel,
1176 "ignore loglevel setting (prints all kernel messages to the console)");
1178 static bool suppress_message_printing(int level)
1180 return (level >= console_loglevel && !ignore_loglevel);
1183 #ifdef CONFIG_BOOT_PRINTK_DELAY
1185 static int boot_delay; /* msecs delay after each printk during bootup */
1186 static unsigned long long loops_per_msec; /* based on boot_delay */
1188 static int __init boot_delay_setup(char *str)
1192 lpj = preset_lpj ? preset_lpj : 1000000; /* some guess */
1193 loops_per_msec = (unsigned long long)lpj / 1000 * HZ;
1195 get_option(&str, &boot_delay);
1196 if (boot_delay > 10 * 1000)
1199 pr_debug("boot_delay: %u, preset_lpj: %ld, lpj: %lu, "
1200 "HZ: %d, loops_per_msec: %llu\n",
1201 boot_delay, preset_lpj, lpj, HZ, loops_per_msec);
1204 early_param("boot_delay", boot_delay_setup);
1206 static void boot_delay_msec(int level)
1208 unsigned long long k;
1209 unsigned long timeout;
1211 if ((boot_delay == 0 || system_state >= SYSTEM_RUNNING)
1212 || suppress_message_printing(level)) {
1216 k = (unsigned long long)loops_per_msec * boot_delay;
1218 timeout = jiffies + msecs_to_jiffies(boot_delay);
1223 * use (volatile) jiffies to prevent
1224 * compiler reduction; loop termination via jiffies
1225 * is secondary and may or may not happen.
1227 if (time_after(jiffies, timeout))
1229 touch_nmi_watchdog();
1233 static inline void boot_delay_msec(int level)
1238 static bool printk_time = IS_ENABLED(CONFIG_PRINTK_TIME);
1239 module_param_named(time, printk_time, bool, S_IRUGO | S_IWUSR);
1241 static size_t print_syslog(unsigned int level, char *buf)
1243 return sprintf(buf, "<%u>", level);
1246 static size_t print_time(u64 ts, char *buf)
1248 unsigned long rem_nsec = do_div(ts, 1000000000);
1250 return sprintf(buf, "[%5lu.%06lu]",
1251 (unsigned long)ts, rem_nsec / 1000);
1254 #ifdef CONFIG_PRINTK_CALLER
1255 static size_t print_caller(u32 id, char *buf)
1259 snprintf(caller, sizeof(caller), "%c%u",
1260 id & 0x80000000 ? 'C' : 'T', id & ~0x80000000);
1261 return sprintf(buf, "[%6s]", caller);
1264 #define print_caller(id, buf) 0
1267 static size_t info_print_prefix(const struct printk_info *info, bool syslog,
1268 bool time, char *buf)
1273 len = print_syslog((info->facility << 3) | info->level, buf);
1276 len += print_time(info->ts_nsec, buf + len);
1278 len += print_caller(info->caller_id, buf + len);
1280 if (IS_ENABLED(CONFIG_PRINTK_CALLER) || time) {
1289 * Prepare the record for printing. The text is shifted within the given
1290 * buffer to avoid a need for another one. The following operations are
1293 * - Add prefix for each line.
1294 * - Drop truncated lines that no longer fit into the buffer.
1295 * - Add the trailing newline that has been removed in vprintk_store().
1296 * - Add a string terminator.
1298 * Since the produced string is always terminated, the maximum possible
1299 * return value is @r->text_buf_size - 1;
1301 * Return: The length of the updated/prepared text, including the added
1302 * prefixes and the newline. The terminator is not counted. The dropped
1303 * line(s) are not counted.
1305 static size_t record_print_text(struct printk_record *r, bool syslog,
1308 size_t text_len = r->info->text_len;
1309 size_t buf_size = r->text_buf_size;
1310 char *text = r->text_buf;
1311 char prefix[PREFIX_MAX];
1312 bool truncated = false;
1319 * If the message was truncated because the buffer was not large
1320 * enough, treat the available text as if it were the full text.
1322 if (text_len > buf_size)
1323 text_len = buf_size;
1325 prefix_len = info_print_prefix(r->info, syslog, time, prefix);
1328 * @text_len: bytes of unprocessed text
1329 * @line_len: bytes of current line _without_ newline
1330 * @text: pointer to beginning of current line
1331 * @len: number of bytes prepared in r->text_buf
1334 next = memchr(text, '\n', text_len);
1336 line_len = next - text;
1338 /* Drop truncated line(s). */
1341 line_len = text_len;
1345 * Truncate the text if there is not enough space to add the
1346 * prefix and a trailing newline and a terminator.
1348 if (len + prefix_len + text_len + 1 + 1 > buf_size) {
1349 /* Drop even the current line if no space. */
1350 if (len + prefix_len + line_len + 1 + 1 > buf_size)
1353 text_len = buf_size - len - prefix_len - 1 - 1;
1357 memmove(text + prefix_len, text, text_len);
1358 memcpy(text, prefix, prefix_len);
1361 * Increment the prepared length to include the text and
1362 * prefix that were just moved+copied. Also increment for the
1363 * newline at the end of this line. If this is the last line,
1364 * there is no newline, but it will be added immediately below.
1366 len += prefix_len + line_len + 1;
1367 if (text_len == line_len) {
1369 * This is the last line. Add the trailing newline
1370 * removed in vprintk_store().
1372 text[prefix_len + line_len] = '\n';
1377 * Advance beyond the added prefix and the related line with
1380 text += prefix_len + line_len + 1;
1383 * The remaining text has only decreased by the line with its
1386 * Note that @text_len can become zero. It happens when @text
1387 * ended with a newline (either due to truncation or the
1388 * original string ending with "\n\n"). The loop is correctly
1389 * repeated and (if not truncated) an empty line with a prefix
1392 text_len -= line_len + 1;
1396 * If a buffer was provided, it will be terminated. Space for the
1397 * string terminator is guaranteed to be available. The terminator is
1398 * not counted in the return value.
1401 r->text_buf[len] = 0;
1406 static size_t get_record_print_text_size(struct printk_info *info,
1407 unsigned int line_count,
1408 bool syslog, bool time)
1410 char prefix[PREFIX_MAX];
1413 prefix_len = info_print_prefix(info, syslog, time, prefix);
1416 * Each line will be preceded with a prefix. The intermediate
1417 * newlines are already within the text, but a final trailing
1418 * newline will be added.
1420 return ((prefix_len * line_count) + info->text_len + 1);
1423 static int syslog_print(char __user *buf, int size)
1425 struct printk_info info;
1426 struct printk_record r;
1430 text = kmalloc(LOG_LINE_MAX + PREFIX_MAX, GFP_KERNEL);
1434 prb_rec_init_rd(&r, &info, text, LOG_LINE_MAX + PREFIX_MAX);
1441 if (!prb_read_valid(prb, syslog_seq, &r)) {
1442 logbuf_unlock_irq();
1445 if (r.info->seq != syslog_seq) {
1446 /* message is gone, move to next valid one */
1447 syslog_seq = r.info->seq;
1452 * To keep reading/counting partial line consistent,
1453 * use printk_time value as of the beginning of a line.
1455 if (!syslog_partial)
1456 syslog_time = printk_time;
1458 skip = syslog_partial;
1459 n = record_print_text(&r, true, syslog_time);
1460 if (n - syslog_partial <= size) {
1461 /* message fits into buffer, move forward */
1462 syslog_seq = r.info->seq + 1;
1463 n -= syslog_partial;
1466 /* partial read(), remember position */
1468 syslog_partial += n;
1471 logbuf_unlock_irq();
1476 if (copy_to_user(buf, text + skip, n)) {
1491 static int syslog_print_all(char __user *buf, int size, bool clear)
1493 struct printk_info info;
1494 unsigned int line_count;
1495 struct printk_record r;
1501 text = kmalloc(LOG_LINE_MAX + PREFIX_MAX, GFP_KERNEL);
1508 * Find first record that fits, including all following records,
1509 * into the user-provided buffer for this dump.
1511 prb_for_each_info(clear_seq, prb, seq, &info, &line_count)
1512 len += get_record_print_text_size(&info, line_count, true, time);
1514 /* move first record forward until length fits into the buffer */
1515 prb_for_each_info(clear_seq, prb, seq, &info, &line_count) {
1518 len -= get_record_print_text_size(&info, line_count, true, time);
1521 prb_rec_init_rd(&r, &info, text, LOG_LINE_MAX + PREFIX_MAX);
1524 prb_for_each_record(seq, prb, seq, &r) {
1527 textlen = record_print_text(&r, true, time);
1529 if (len + textlen > size) {
1534 logbuf_unlock_irq();
1535 if (copy_to_user(buf + len, text, textlen))
1547 logbuf_unlock_irq();
1553 static void syslog_clear(void)
1556 clear_seq = prb_next_seq(prb);
1557 logbuf_unlock_irq();
1560 int do_syslog(int type, char __user *buf, int len, int source)
1563 static int saved_console_loglevel = LOGLEVEL_DEFAULT;
1566 error = check_syslog_permissions(type, source);
1571 case SYSLOG_ACTION_CLOSE: /* Close log */
1573 case SYSLOG_ACTION_OPEN: /* Open log */
1575 case SYSLOG_ACTION_READ: /* Read from log */
1576 if (!buf || len < 0)
1580 if (!access_ok(buf, len))
1582 error = wait_event_interruptible(log_wait,
1583 prb_read_valid(prb, syslog_seq, NULL));
1586 error = syslog_print(buf, len);
1588 /* Read/clear last kernel messages */
1589 case SYSLOG_ACTION_READ_CLEAR:
1592 /* Read last kernel messages */
1593 case SYSLOG_ACTION_READ_ALL:
1594 if (!buf || len < 0)
1598 if (!access_ok(buf, len))
1600 error = syslog_print_all(buf, len, clear);
1602 /* Clear ring buffer */
1603 case SYSLOG_ACTION_CLEAR:
1606 /* Disable logging to console */
1607 case SYSLOG_ACTION_CONSOLE_OFF:
1608 if (saved_console_loglevel == LOGLEVEL_DEFAULT)
1609 saved_console_loglevel = console_loglevel;
1610 console_loglevel = minimum_console_loglevel;
1612 /* Enable logging to console */
1613 case SYSLOG_ACTION_CONSOLE_ON:
1614 if (saved_console_loglevel != LOGLEVEL_DEFAULT) {
1615 console_loglevel = saved_console_loglevel;
1616 saved_console_loglevel = LOGLEVEL_DEFAULT;
1619 /* Set level of messages printed to console */
1620 case SYSLOG_ACTION_CONSOLE_LEVEL:
1621 if (len < 1 || len > 8)
1623 if (len < minimum_console_loglevel)
1624 len = minimum_console_loglevel;
1625 console_loglevel = len;
1626 /* Implicitly re-enable logging to console */
1627 saved_console_loglevel = LOGLEVEL_DEFAULT;
1629 /* Number of chars in the log buffer */
1630 case SYSLOG_ACTION_SIZE_UNREAD:
1632 if (syslog_seq < prb_first_valid_seq(prb)) {
1633 /* messages are gone, move to first one */
1634 syslog_seq = prb_first_valid_seq(prb);
1637 if (source == SYSLOG_FROM_PROC) {
1639 * Short-cut for poll(/"proc/kmsg") which simply checks
1640 * for pending data, not the size; return the count of
1641 * records, not the length.
1643 error = prb_next_seq(prb) - syslog_seq;
1645 bool time = syslog_partial ? syslog_time : printk_time;
1646 struct printk_info info;
1647 unsigned int line_count;
1650 prb_for_each_info(syslog_seq, prb, seq, &info,
1652 error += get_record_print_text_size(&info, line_count,
1656 error -= syslog_partial;
1658 logbuf_unlock_irq();
1660 /* Size of the log buffer */
1661 case SYSLOG_ACTION_SIZE_BUFFER:
1662 error = log_buf_len;
1672 SYSCALL_DEFINE3(syslog, int, type, char __user *, buf, int, len)
1674 return do_syslog(type, buf, len, SYSLOG_FROM_READER);
1678 * Special console_lock variants that help to reduce the risk of soft-lockups.
1679 * They allow to pass console_lock to another printk() call using a busy wait.
1682 #ifdef CONFIG_LOCKDEP
1683 static struct lockdep_map console_owner_dep_map = {
1684 .name = "console_owner"
1688 static DEFINE_RAW_SPINLOCK(console_owner_lock);
1689 static struct task_struct *console_owner;
1690 static bool console_waiter;
1693 * console_lock_spinning_enable - mark beginning of code where another
1694 * thread might safely busy wait
1696 * This basically converts console_lock into a spinlock. This marks
1697 * the section where the console_lock owner can not sleep, because
1698 * there may be a waiter spinning (like a spinlock). Also it must be
1699 * ready to hand over the lock at the end of the section.
1701 static void console_lock_spinning_enable(void)
1703 raw_spin_lock(&console_owner_lock);
1704 console_owner = current;
1705 raw_spin_unlock(&console_owner_lock);
1707 /* The waiter may spin on us after setting console_owner */
1708 spin_acquire(&console_owner_dep_map, 0, 0, _THIS_IP_);
1712 * console_lock_spinning_disable_and_check - mark end of code where another
1713 * thread was able to busy wait and check if there is a waiter
1715 * This is called at the end of the section where spinning is allowed.
1716 * It has two functions. First, it is a signal that it is no longer
1717 * safe to start busy waiting for the lock. Second, it checks if
1718 * there is a busy waiter and passes the lock rights to her.
1720 * Important: Callers lose the lock if there was a busy waiter.
1721 * They must not touch items synchronized by console_lock
1724 * Return: 1 if the lock rights were passed, 0 otherwise.
1726 static int console_lock_spinning_disable_and_check(void)
1730 raw_spin_lock(&console_owner_lock);
1731 waiter = READ_ONCE(console_waiter);
1732 console_owner = NULL;
1733 raw_spin_unlock(&console_owner_lock);
1736 spin_release(&console_owner_dep_map, _THIS_IP_);
1740 /* The waiter is now free to continue */
1741 WRITE_ONCE(console_waiter, false);
1743 spin_release(&console_owner_dep_map, _THIS_IP_);
1746 * Hand off console_lock to waiter. The waiter will perform
1747 * the up(). After this, the waiter is the console_lock owner.
1749 mutex_release(&console_lock_dep_map, _THIS_IP_);
1754 * console_trylock_spinning - try to get console_lock by busy waiting
1756 * This allows to busy wait for the console_lock when the current
1757 * owner is running in specially marked sections. It means that
1758 * the current owner is running and cannot reschedule until it
1759 * is ready to lose the lock.
1761 * Return: 1 if we got the lock, 0 othrewise
1763 static int console_trylock_spinning(void)
1765 struct task_struct *owner = NULL;
1768 unsigned long flags;
1770 if (console_trylock())
1773 printk_safe_enter_irqsave(flags);
1775 raw_spin_lock(&console_owner_lock);
1776 owner = READ_ONCE(console_owner);
1777 waiter = READ_ONCE(console_waiter);
1778 if (!waiter && owner && owner != current) {
1779 WRITE_ONCE(console_waiter, true);
1782 raw_spin_unlock(&console_owner_lock);
1785 * If there is an active printk() writing to the
1786 * consoles, instead of having it write our data too,
1787 * see if we can offload that load from the active
1788 * printer, and do some printing ourselves.
1789 * Go into a spin only if there isn't already a waiter
1790 * spinning, and there is an active printer, and
1791 * that active printer isn't us (recursive printk?).
1794 printk_safe_exit_irqrestore(flags);
1798 /* We spin waiting for the owner to release us */
1799 spin_acquire(&console_owner_dep_map, 0, 0, _THIS_IP_);
1800 /* Owner will clear console_waiter on hand off */
1801 while (READ_ONCE(console_waiter))
1803 spin_release(&console_owner_dep_map, _THIS_IP_);
1805 printk_safe_exit_irqrestore(flags);
1807 * The owner passed the console lock to us.
1808 * Since we did not spin on console lock, annotate
1809 * this as a trylock. Otherwise lockdep will
1812 mutex_acquire(&console_lock_dep_map, 0, 1, _THIS_IP_);
1818 * Call the console drivers, asking them to write out
1819 * log_buf[start] to log_buf[end - 1].
1820 * The console_lock must be held.
1822 static void call_console_drivers(const char *ext_text, size_t ext_len,
1823 const char *text, size_t len)
1825 static char dropped_text[64];
1826 size_t dropped_len = 0;
1827 struct console *con;
1829 trace_console_rcuidle(text, len);
1831 if (!console_drivers)
1834 if (console_dropped) {
1835 dropped_len = snprintf(dropped_text, sizeof(dropped_text),
1836 "** %lu printk messages dropped **\n",
1838 console_dropped = 0;
1841 for_each_console(con) {
1842 if (exclusive_console && con != exclusive_console)
1844 if (!(con->flags & CON_ENABLED))
1848 if (!cpu_online(smp_processor_id()) &&
1849 !(con->flags & CON_ANYTIME))
1851 if (con->flags & CON_EXTENDED)
1852 con->write(con, ext_text, ext_len);
1855 con->write(con, dropped_text, dropped_len);
1856 con->write(con, text, len);
1861 int printk_delay_msec __read_mostly;
1863 static inline void printk_delay(void)
1865 if (unlikely(printk_delay_msec)) {
1866 int m = printk_delay_msec;
1870 touch_nmi_watchdog();
1875 static inline u32 printk_caller_id(void)
1877 return in_task() ? task_pid_nr(current) :
1878 0x80000000 + raw_smp_processor_id();
1882 * parse_prefix - Parse level and control flags.
1884 * @text: The terminated text message.
1885 * @level: A pointer to the current level value, will be updated.
1886 * @lflags: A pointer to the current log flags, will be updated.
1888 * @level may be NULL if the caller is not interested in the parsed value.
1889 * Otherwise the variable pointed to by @level must be set to
1890 * LOGLEVEL_DEFAULT in order to be updated with the parsed value.
1892 * @lflags may be NULL if the caller is not interested in the parsed value.
1893 * Otherwise the variable pointed to by @lflags will be OR'd with the parsed
1896 * Return: The length of the parsed level and control flags.
1898 static u16 parse_prefix(char *text, int *level, enum log_flags *lflags)
1904 kern_level = printk_get_level(text);
1908 switch (kern_level) {
1910 if (level && *level == LOGLEVEL_DEFAULT)
1911 *level = kern_level - '0';
1913 case 'c': /* KERN_CONT */
1915 *lflags |= LOG_CONT;
1925 static u16 printk_sprint(char *text, u16 size, int facility, enum log_flags *lflags,
1926 const char *fmt, va_list args)
1930 text_len = vscnprintf(text, size, fmt, args);
1932 /* Mark and strip a trailing newline. */
1933 if (text_len && text[text_len - 1] == '\n') {
1935 *lflags |= LOG_NEWLINE;
1938 /* Strip log level and control flags. */
1939 if (facility == 0) {
1942 prefix_len = parse_prefix(text, NULL, NULL);
1944 text_len -= prefix_len;
1945 memmove(text, text + prefix_len, text_len);
1953 int vprintk_store(int facility, int level,
1954 const struct dev_printk_info *dev_info,
1955 const char *fmt, va_list args)
1957 const u32 caller_id = printk_caller_id();
1958 struct prb_reserved_entry e;
1959 enum log_flags lflags = 0;
1960 struct printk_record r;
1961 u16 trunc_msg_len = 0;
1969 * Since the duration of printk() can vary depending on the message
1970 * and state of the ringbuffer, grab the timestamp now so that it is
1971 * close to the call of printk(). This provides a more deterministic
1972 * timestamp with respect to the caller.
1974 ts_nsec = local_clock();
1977 * The sprintf needs to come first since the syslog prefix might be
1978 * passed in as a parameter. An extra byte must be reserved so that
1979 * later the vscnprintf() into the reserved buffer has room for the
1980 * terminating '\0', which is not counted by vsnprintf().
1982 va_copy(args2, args);
1983 reserve_size = vsnprintf(&prefix_buf[0], sizeof(prefix_buf), fmt, args2) + 1;
1986 if (reserve_size > LOG_LINE_MAX)
1987 reserve_size = LOG_LINE_MAX;
1989 /* Extract log level or control flags. */
1991 parse_prefix(&prefix_buf[0], &level, &lflags);
1993 if (level == LOGLEVEL_DEFAULT)
1994 level = default_message_loglevel;
1997 lflags |= LOG_NEWLINE;
1999 if (lflags & LOG_CONT) {
2000 prb_rec_init_wr(&r, reserve_size);
2001 if (prb_reserve_in_last(&e, prb, &r, caller_id, LOG_LINE_MAX)) {
2002 text_len = printk_sprint(&r.text_buf[r.info->text_len], reserve_size,
2003 facility, &lflags, fmt, args);
2004 r.info->text_len += text_len;
2006 if (lflags & LOG_NEWLINE) {
2007 r.info->flags |= LOG_NEWLINE;
2008 prb_final_commit(&e);
2018 * Explicitly initialize the record before every prb_reserve() call.
2019 * prb_reserve_in_last() and prb_reserve() purposely invalidate the
2020 * structure when they fail.
2022 prb_rec_init_wr(&r, reserve_size);
2023 if (!prb_reserve(&e, prb, &r)) {
2024 /* truncate the message if it is too long for empty buffer */
2025 truncate_msg(&reserve_size, &trunc_msg_len);
2027 prb_rec_init_wr(&r, reserve_size + trunc_msg_len);
2028 if (!prb_reserve(&e, prb, &r))
2033 text_len = printk_sprint(&r.text_buf[0], reserve_size, facility, &lflags, fmt, args);
2035 memcpy(&r.text_buf[text_len], trunc_msg, trunc_msg_len);
2036 r.info->text_len = text_len + trunc_msg_len;
2037 r.info->facility = facility;
2038 r.info->level = level & 7;
2039 r.info->flags = lflags & 0x1f;
2040 r.info->ts_nsec = ts_nsec;
2041 r.info->caller_id = caller_id;
2043 memcpy(&r.info->dev_info, dev_info, sizeof(r.info->dev_info));
2045 /* A message without a trailing newline can be continued. */
2046 if (!(lflags & LOG_NEWLINE))
2049 prb_final_commit(&e);
2051 return (text_len + trunc_msg_len);
2054 asmlinkage int vprintk_emit(int facility, int level,
2055 const struct dev_printk_info *dev_info,
2056 const char *fmt, va_list args)
2059 bool in_sched = false;
2060 unsigned long flags;
2062 /* Suppress unimportant messages after panic happens */
2063 if (unlikely(suppress_printk))
2066 if (level == LOGLEVEL_SCHED) {
2067 level = LOGLEVEL_DEFAULT;
2071 boot_delay_msec(level);
2074 printk_safe_enter_irqsave(flags);
2075 printed_len = vprintk_store(facility, level, dev_info, fmt, args);
2076 printk_safe_exit_irqrestore(flags);
2078 /* If called from the scheduler, we can not call up(). */
2081 * Disable preemption to avoid being preempted while holding
2082 * console_sem which would prevent anyone from printing to
2087 * Try to acquire and then immediately release the console
2088 * semaphore. The release will print out buffers and wake up
2089 * /dev/kmsg and syslog() users.
2091 if (console_trylock_spinning())
2099 EXPORT_SYMBOL(vprintk_emit);
2101 asmlinkage int vprintk(const char *fmt, va_list args)
2103 return vprintk_func(fmt, args);
2105 EXPORT_SYMBOL(vprintk);
2107 int vprintk_default(const char *fmt, va_list args)
2109 return vprintk_emit(0, LOGLEVEL_DEFAULT, NULL, fmt, args);
2111 EXPORT_SYMBOL_GPL(vprintk_default);
2114 * printk - print a kernel message
2115 * @fmt: format string
2117 * This is printk(). It can be called from any context. We want it to work.
2119 * We try to grab the console_lock. If we succeed, it's easy - we log the
2120 * output and call the console drivers. If we fail to get the semaphore, we
2121 * place the output into the log buffer and return. The current holder of
2122 * the console_sem will notice the new output in console_unlock(); and will
2123 * send it to the consoles before releasing the lock.
2125 * One effect of this deferred printing is that code which calls printk() and
2126 * then changes console_loglevel may break. This is because console_loglevel
2127 * is inspected when the actual printing occurs.
2132 * See the vsnprintf() documentation for format string extensions over C99.
2134 asmlinkage __visible int printk(const char *fmt, ...)
2139 va_start(args, fmt);
2140 r = vprintk_func(fmt, args);
2145 EXPORT_SYMBOL(printk);
2147 #else /* CONFIG_PRINTK */
2149 #define LOG_LINE_MAX 0
2150 #define PREFIX_MAX 0
2151 #define printk_time false
2153 #define prb_read_valid(rb, seq, r) false
2154 #define prb_first_valid_seq(rb) 0
2156 static u64 syslog_seq;
2157 static u64 console_seq;
2158 static u64 exclusive_console_stop_seq;
2159 static unsigned long console_dropped;
2161 static size_t record_print_text(const struct printk_record *r,
2162 bool syslog, bool time)
2166 static ssize_t info_print_ext_header(char *buf, size_t size,
2167 struct printk_info *info)
2171 static ssize_t msg_print_ext_body(char *buf, size_t size,
2172 char *text, size_t text_len,
2173 struct dev_printk_info *dev_info) { return 0; }
2174 static void console_lock_spinning_enable(void) { }
2175 static int console_lock_spinning_disable_and_check(void) { return 0; }
2176 static void call_console_drivers(const char *ext_text, size_t ext_len,
2177 const char *text, size_t len) {}
2178 static bool suppress_message_printing(int level) { return false; }
2180 #endif /* CONFIG_PRINTK */
2182 #ifdef CONFIG_EARLY_PRINTK
2183 struct console *early_console;
2185 asmlinkage __visible void early_printk(const char *fmt, ...)
2195 n = vscnprintf(buf, sizeof(buf), fmt, ap);
2198 early_console->write(early_console, buf, n);
2202 static int __add_preferred_console(char *name, int idx, char *options,
2203 char *brl_options, bool user_specified)
2205 struct console_cmdline *c;
2209 * See if this tty is not yet registered, and
2210 * if we have a slot free.
2212 for (i = 0, c = console_cmdline;
2213 i < MAX_CMDLINECONSOLES && c->name[0];
2215 if (strcmp(c->name, name) == 0 && c->index == idx) {
2217 preferred_console = i;
2219 c->user_specified = true;
2223 if (i == MAX_CMDLINECONSOLES)
2226 preferred_console = i;
2227 strlcpy(c->name, name, sizeof(c->name));
2228 c->options = options;
2229 c->user_specified = user_specified;
2230 braille_set_options(c, brl_options);
2236 static int __init console_msg_format_setup(char *str)
2238 if (!strcmp(str, "syslog"))
2239 console_msg_format = MSG_FORMAT_SYSLOG;
2240 if (!strcmp(str, "default"))
2241 console_msg_format = MSG_FORMAT_DEFAULT;
2244 __setup("console_msg_format=", console_msg_format_setup);
2247 * Set up a console. Called via do_early_param() in init/main.c
2248 * for each "console=" parameter in the boot command line.
2250 static int __init console_setup(char *str)
2252 char buf[sizeof(console_cmdline[0].name) + 4]; /* 4 for "ttyS" */
2253 char *s, *options, *brl_options = NULL;
2257 * console="" or console=null have been suggested as a way to
2258 * disable console output. Use ttynull that has been created
2259 * for exacly this purpose.
2261 if (str[0] == 0 || strcmp(str, "null") == 0) {
2262 __add_preferred_console("ttynull", 0, NULL, NULL, true);
2266 if (_braille_console_setup(&str, &brl_options))
2270 * Decode str into name, index, options.
2272 if (str[0] >= '0' && str[0] <= '9') {
2273 strcpy(buf, "ttyS");
2274 strncpy(buf + 4, str, sizeof(buf) - 5);
2276 strncpy(buf, str, sizeof(buf) - 1);
2278 buf[sizeof(buf) - 1] = 0;
2279 options = strchr(str, ',');
2283 if (!strcmp(str, "ttya"))
2284 strcpy(buf, "ttyS0");
2285 if (!strcmp(str, "ttyb"))
2286 strcpy(buf, "ttyS1");
2288 for (s = buf; *s; s++)
2289 if (isdigit(*s) || *s == ',')
2291 idx = simple_strtoul(s, NULL, 10);
2294 __add_preferred_console(buf, idx, options, brl_options, true);
2295 console_set_on_cmdline = 1;
2298 __setup("console=", console_setup);
2301 * add_preferred_console - add a device to the list of preferred consoles.
2302 * @name: device name
2303 * @idx: device index
2304 * @options: options for this console
2306 * The last preferred console added will be used for kernel messages
2307 * and stdin/out/err for init. Normally this is used by console_setup
2308 * above to handle user-supplied console arguments; however it can also
2309 * be used by arch-specific code either to override the user or more
2310 * commonly to provide a default console (ie from PROM variables) when
2311 * the user has not supplied one.
2313 int add_preferred_console(char *name, int idx, char *options)
2315 return __add_preferred_console(name, idx, options, NULL, false);
2318 bool console_suspend_enabled = true;
2319 EXPORT_SYMBOL(console_suspend_enabled);
2321 static int __init console_suspend_disable(char *str)
2323 console_suspend_enabled = false;
2326 __setup("no_console_suspend", console_suspend_disable);
2327 module_param_named(console_suspend, console_suspend_enabled,
2328 bool, S_IRUGO | S_IWUSR);
2329 MODULE_PARM_DESC(console_suspend, "suspend console during suspend"
2330 " and hibernate operations");
2333 * suspend_console - suspend the console subsystem
2335 * This disables printk() while we go into suspend states
2337 void suspend_console(void)
2339 if (!console_suspend_enabled)
2341 pr_info("Suspending console(s) (use no_console_suspend to debug)\n");
2343 console_suspended = 1;
2347 void resume_console(void)
2349 if (!console_suspend_enabled)
2352 console_suspended = 0;
2357 * console_cpu_notify - print deferred console messages after CPU hotplug
2360 * If printk() is called from a CPU that is not online yet, the messages
2361 * will be printed on the console only if there are CON_ANYTIME consoles.
2362 * This function is called when a new CPU comes online (or fails to come
2363 * up) or goes offline.
2365 static int console_cpu_notify(unsigned int cpu)
2367 if (!cpuhp_tasks_frozen) {
2368 /* If trylock fails, someone else is doing the printing */
2369 if (console_trylock())
2376 * console_lock - lock the console system for exclusive use.
2378 * Acquires a lock which guarantees that the caller has
2379 * exclusive access to the console system and the console_drivers list.
2381 * Can sleep, returns nothing.
2383 void console_lock(void)
2388 if (console_suspended)
2391 console_may_schedule = 1;
2393 EXPORT_SYMBOL(console_lock);
2396 * console_trylock - try to lock the console system for exclusive use.
2398 * Try to acquire a lock which guarantees that the caller has exclusive
2399 * access to the console system and the console_drivers list.
2401 * returns 1 on success, and 0 on failure to acquire the lock.
2403 int console_trylock(void)
2405 if (down_trylock_console_sem())
2407 if (console_suspended) {
2412 console_may_schedule = 0;
2415 EXPORT_SYMBOL(console_trylock);
2417 int is_console_locked(void)
2419 return console_locked;
2421 EXPORT_SYMBOL(is_console_locked);
2424 * Check if we have any console that is capable of printing while cpu is
2425 * booting or shutting down. Requires console_sem.
2427 static int have_callable_console(void)
2429 struct console *con;
2431 for_each_console(con)
2432 if ((con->flags & CON_ENABLED) &&
2433 (con->flags & CON_ANYTIME))
2440 * Can we actually use the console at this time on this cpu?
2442 * Console drivers may assume that per-cpu resources have been allocated. So
2443 * unless they're explicitly marked as being able to cope (CON_ANYTIME) don't
2444 * call them until this CPU is officially up.
2446 static inline int can_use_console(void)
2448 return cpu_online(raw_smp_processor_id()) || have_callable_console();
2452 * console_unlock - unlock the console system
2454 * Releases the console_lock which the caller holds on the console system
2455 * and the console driver list.
2457 * While the console_lock was held, console output may have been buffered
2458 * by printk(). If this is the case, console_unlock(); emits
2459 * the output prior to releasing the lock.
2461 * If there is output waiting, we wake /dev/kmsg and syslog() users.
2463 * console_unlock(); may be called from any context.
2465 void console_unlock(void)
2467 static char ext_text[CONSOLE_EXT_LOG_MAX];
2468 static char text[LOG_LINE_MAX + PREFIX_MAX];
2469 unsigned long flags;
2470 bool do_cond_resched, retry;
2471 struct printk_info info;
2472 struct printk_record r;
2474 if (console_suspended) {
2479 prb_rec_init_rd(&r, &info, text, sizeof(text));
2482 * Console drivers are called with interrupts disabled, so
2483 * @console_may_schedule should be cleared before; however, we may
2484 * end up dumping a lot of lines, for example, if called from
2485 * console registration path, and should invoke cond_resched()
2486 * between lines if allowable. Not doing so can cause a very long
2487 * scheduling stall on a slow console leading to RCU stall and
2488 * softlockup warnings which exacerbate the issue with more
2489 * messages practically incapacitating the system.
2491 * console_trylock() is not able to detect the preemptive
2492 * context reliably. Therefore the value must be stored before
2493 * and cleared after the "again" goto label.
2495 do_cond_resched = console_may_schedule;
2497 console_may_schedule = 0;
2500 * We released the console_sem lock, so we need to recheck if
2501 * cpu is online and (if not) is there at least one CON_ANYTIME
2504 if (!can_use_console()) {
2514 printk_safe_enter_irqsave(flags);
2515 raw_spin_lock(&logbuf_lock);
2517 if (!prb_read_valid(prb, console_seq, &r))
2520 if (console_seq != r.info->seq) {
2521 console_dropped += r.info->seq - console_seq;
2522 console_seq = r.info->seq;
2525 if (suppress_message_printing(r.info->level)) {
2527 * Skip record we have buffered and already printed
2528 * directly to the console when we received it, and
2529 * record that has level above the console loglevel.
2535 /* Output to all consoles once old messages replayed. */
2536 if (unlikely(exclusive_console &&
2537 console_seq >= exclusive_console_stop_seq)) {
2538 exclusive_console = NULL;
2542 * Handle extended console text first because later
2543 * record_print_text() will modify the record buffer in-place.
2545 if (nr_ext_console_drivers) {
2546 ext_len = info_print_ext_header(ext_text,
2549 ext_len += msg_print_ext_body(ext_text + ext_len,
2550 sizeof(ext_text) - ext_len,
2555 len = record_print_text(&r,
2556 console_msg_format & MSG_FORMAT_SYSLOG,
2559 raw_spin_unlock(&logbuf_lock);
2562 * While actively printing out messages, if another printk()
2563 * were to occur on another CPU, it may wait for this one to
2564 * finish. This task can not be preempted if there is a
2565 * waiter waiting to take over.
2567 console_lock_spinning_enable();
2569 stop_critical_timings(); /* don't trace print latency */
2570 call_console_drivers(ext_text, ext_len, text, len);
2571 start_critical_timings();
2573 if (console_lock_spinning_disable_and_check()) {
2574 printk_safe_exit_irqrestore(flags);
2578 printk_safe_exit_irqrestore(flags);
2580 if (do_cond_resched)
2586 raw_spin_unlock(&logbuf_lock);
2591 * Someone could have filled up the buffer again, so re-check if there's
2592 * something to flush. In case we cannot trylock the console_sem again,
2593 * there's a new owner and the console_unlock() from them will do the
2594 * flush, no worries.
2596 raw_spin_lock(&logbuf_lock);
2597 retry = prb_read_valid(prb, console_seq, NULL);
2598 raw_spin_unlock(&logbuf_lock);
2599 printk_safe_exit_irqrestore(flags);
2601 if (retry && console_trylock())
2604 EXPORT_SYMBOL(console_unlock);
2607 * console_conditional_schedule - yield the CPU if required
2609 * If the console code is currently allowed to sleep, and
2610 * if this CPU should yield the CPU to another task, do
2613 * Must be called within console_lock();.
2615 void __sched console_conditional_schedule(void)
2617 if (console_may_schedule)
2620 EXPORT_SYMBOL(console_conditional_schedule);
2622 void console_unblank(void)
2627 * console_unblank can no longer be called in interrupt context unless
2628 * oops_in_progress is set to 1..
2630 if (oops_in_progress) {
2631 if (down_trylock_console_sem() != 0)
2637 console_may_schedule = 0;
2639 if ((c->flags & CON_ENABLED) && c->unblank)
2645 * console_flush_on_panic - flush console content on panic
2646 * @mode: flush all messages in buffer or just the pending ones
2648 * Immediately output all pending messages no matter what.
2650 void console_flush_on_panic(enum con_flush_mode mode)
2653 * If someone else is holding the console lock, trylock will fail
2654 * and may_schedule may be set. Ignore and proceed to unlock so
2655 * that messages are flushed out. As this can be called from any
2656 * context and we don't want to get preempted while flushing,
2657 * ensure may_schedule is cleared.
2660 console_may_schedule = 0;
2662 if (mode == CONSOLE_REPLAY_ALL) {
2663 unsigned long flags;
2665 logbuf_lock_irqsave(flags);
2666 console_seq = prb_first_valid_seq(prb);
2667 logbuf_unlock_irqrestore(flags);
2673 * Return the console tty driver structure and its associated index
2675 struct tty_driver *console_device(int *index)
2678 struct tty_driver *driver = NULL;
2681 for_each_console(c) {
2684 driver = c->device(c, index);
2693 * Prevent further output on the passed console device so that (for example)
2694 * serial drivers can disable console output before suspending a port, and can
2695 * re-enable output afterwards.
2697 void console_stop(struct console *console)
2700 console->flags &= ~CON_ENABLED;
2703 EXPORT_SYMBOL(console_stop);
2705 void console_start(struct console *console)
2708 console->flags |= CON_ENABLED;
2711 EXPORT_SYMBOL(console_start);
2713 static int __read_mostly keep_bootcon;
2715 static int __init keep_bootcon_setup(char *str)
2718 pr_info("debug: skip boot console de-registration.\n");
2723 early_param("keep_bootcon", keep_bootcon_setup);
2726 * This is called by register_console() to try to match
2727 * the newly registered console with any of the ones selected
2728 * by either the command line or add_preferred_console() and
2731 * Care need to be taken with consoles that are statically
2732 * enabled such as netconsole
2734 static int try_enable_new_console(struct console *newcon, bool user_specified)
2736 struct console_cmdline *c;
2739 for (i = 0, c = console_cmdline;
2740 i < MAX_CMDLINECONSOLES && c->name[0];
2742 if (c->user_specified != user_specified)
2744 if (!newcon->match ||
2745 newcon->match(newcon, c->name, c->index, c->options) != 0) {
2746 /* default matching */
2747 BUILD_BUG_ON(sizeof(c->name) != sizeof(newcon->name));
2748 if (strcmp(c->name, newcon->name) != 0)
2750 if (newcon->index >= 0 &&
2751 newcon->index != c->index)
2753 if (newcon->index < 0)
2754 newcon->index = c->index;
2756 if (_braille_register_console(newcon, c))
2759 if (newcon->setup &&
2760 (err = newcon->setup(newcon, c->options)) != 0)
2763 newcon->flags |= CON_ENABLED;
2764 if (i == preferred_console) {
2765 newcon->flags |= CON_CONSDEV;
2766 has_preferred_console = true;
2772 * Some consoles, such as pstore and netconsole, can be enabled even
2773 * without matching. Accept the pre-enabled consoles only when match()
2774 * and setup() had a chance to be called.
2776 if (newcon->flags & CON_ENABLED && c->user_specified == user_specified)
2783 * The console driver calls this routine during kernel initialization
2784 * to register the console printing procedure with printk() and to
2785 * print any messages that were printed by the kernel before the
2786 * console driver was initialized.
2788 * This can happen pretty early during the boot process (because of
2789 * early_printk) - sometimes before setup_arch() completes - be careful
2790 * of what kernel features are used - they may not be initialised yet.
2792 * There are two types of consoles - bootconsoles (early_printk) and
2793 * "real" consoles (everything which is not a bootconsole) which are
2794 * handled differently.
2795 * - Any number of bootconsoles can be registered at any time.
2796 * - As soon as a "real" console is registered, all bootconsoles
2797 * will be unregistered automatically.
2798 * - Once a "real" console is registered, any attempt to register a
2799 * bootconsoles will be rejected
2801 void register_console(struct console *newcon)
2803 unsigned long flags;
2804 struct console *bcon = NULL;
2807 for_each_console(bcon) {
2808 if (WARN(bcon == newcon, "console '%s%d' already registered\n",
2809 bcon->name, bcon->index))
2814 * before we register a new CON_BOOT console, make sure we don't
2815 * already have a valid console
2817 if (newcon->flags & CON_BOOT) {
2818 for_each_console(bcon) {
2819 if (!(bcon->flags & CON_BOOT)) {
2820 pr_info("Too late to register bootconsole %s%d\n",
2821 newcon->name, newcon->index);
2827 if (console_drivers && console_drivers->flags & CON_BOOT)
2828 bcon = console_drivers;
2830 if (!has_preferred_console || bcon || !console_drivers)
2831 has_preferred_console = preferred_console >= 0;
2834 * See if we want to use this console driver. If we
2835 * didn't select a console we take the first one
2836 * that registers here.
2838 if (!has_preferred_console) {
2839 if (newcon->index < 0)
2841 if (newcon->setup == NULL ||
2842 newcon->setup(newcon, NULL) == 0) {
2843 newcon->flags |= CON_ENABLED;
2844 if (newcon->device) {
2845 newcon->flags |= CON_CONSDEV;
2846 has_preferred_console = true;
2851 /* See if this console matches one we selected on the command line */
2852 err = try_enable_new_console(newcon, true);
2854 /* If not, try to match against the platform default(s) */
2856 err = try_enable_new_console(newcon, false);
2858 /* printk() messages are not printed to the Braille console. */
2859 if (err || newcon->flags & CON_BRL)
2863 * If we have a bootconsole, and are switching to a real console,
2864 * don't print everything out again, since when the boot console, and
2865 * the real console are the same physical device, it's annoying to
2866 * see the beginning boot messages twice
2868 if (bcon && ((newcon->flags & (CON_CONSDEV | CON_BOOT)) == CON_CONSDEV))
2869 newcon->flags &= ~CON_PRINTBUFFER;
2872 * Put this console in the list - keep the
2873 * preferred driver at the head of the list.
2876 if ((newcon->flags & CON_CONSDEV) || console_drivers == NULL) {
2877 newcon->next = console_drivers;
2878 console_drivers = newcon;
2880 newcon->next->flags &= ~CON_CONSDEV;
2881 /* Ensure this flag is always set for the head of the list */
2882 newcon->flags |= CON_CONSDEV;
2884 newcon->next = console_drivers->next;
2885 console_drivers->next = newcon;
2888 if (newcon->flags & CON_EXTENDED)
2889 nr_ext_console_drivers++;
2891 if (newcon->flags & CON_PRINTBUFFER) {
2893 * console_unlock(); will print out the buffered messages
2896 logbuf_lock_irqsave(flags);
2898 * We're about to replay the log buffer. Only do this to the
2899 * just-registered console to avoid excessive message spam to
2900 * the already-registered consoles.
2902 * Set exclusive_console with disabled interrupts to reduce
2903 * race window with eventual console_flush_on_panic() that
2904 * ignores console_lock.
2906 exclusive_console = newcon;
2907 exclusive_console_stop_seq = console_seq;
2908 console_seq = syslog_seq;
2909 logbuf_unlock_irqrestore(flags);
2912 console_sysfs_notify();
2915 * By unregistering the bootconsoles after we enable the real console
2916 * we get the "console xxx enabled" message on all the consoles -
2917 * boot consoles, real consoles, etc - this is to ensure that end
2918 * users know there might be something in the kernel's log buffer that
2919 * went to the bootconsole (that they do not see on the real console)
2921 pr_info("%sconsole [%s%d] enabled\n",
2922 (newcon->flags & CON_BOOT) ? "boot" : "" ,
2923 newcon->name, newcon->index);
2925 ((newcon->flags & (CON_CONSDEV | CON_BOOT)) == CON_CONSDEV) &&
2927 /* We need to iterate through all boot consoles, to make
2928 * sure we print everything out, before we unregister them.
2930 for_each_console(bcon)
2931 if (bcon->flags & CON_BOOT)
2932 unregister_console(bcon);
2935 EXPORT_SYMBOL(register_console);
2937 int unregister_console(struct console *console)
2939 struct console *con;
2942 pr_info("%sconsole [%s%d] disabled\n",
2943 (console->flags & CON_BOOT) ? "boot" : "" ,
2944 console->name, console->index);
2946 res = _braille_unregister_console(console);
2954 if (console_drivers == console) {
2955 console_drivers=console->next;
2958 for_each_console(con) {
2959 if (con->next == console) {
2960 con->next = console->next;
2968 goto out_disable_unlock;
2970 if (console->flags & CON_EXTENDED)
2971 nr_ext_console_drivers--;
2974 * If this isn't the last console and it has CON_CONSDEV set, we
2975 * need to set it on the next preferred console.
2977 if (console_drivers != NULL && console->flags & CON_CONSDEV)
2978 console_drivers->flags |= CON_CONSDEV;
2980 console->flags &= ~CON_ENABLED;
2982 console_sysfs_notify();
2985 res = console->exit(console);
2990 console->flags &= ~CON_ENABLED;
2995 EXPORT_SYMBOL(unregister_console);
2998 * Initialize the console device. This is called *early*, so
2999 * we can't necessarily depend on lots of kernel help here.
3000 * Just do some early initializations, and do the complex setup
3003 void __init console_init(void)
3007 initcall_entry_t *ce;
3009 /* Setup the default TTY line discipline. */
3013 * set up the console device so that later boot sequences can
3014 * inform about problems etc..
3016 ce = __con_initcall_start;
3017 trace_initcall_level("console");
3018 while (ce < __con_initcall_end) {
3019 call = initcall_from_entry(ce);
3020 trace_initcall_start(call);
3022 trace_initcall_finish(call, ret);
3028 * Some boot consoles access data that is in the init section and which will
3029 * be discarded after the initcalls have been run. To make sure that no code
3030 * will access this data, unregister the boot consoles in a late initcall.
3032 * If for some reason, such as deferred probe or the driver being a loadable
3033 * module, the real console hasn't registered yet at this point, there will
3034 * be a brief interval in which no messages are logged to the console, which
3035 * makes it difficult to diagnose problems that occur during this time.
3037 * To mitigate this problem somewhat, only unregister consoles whose memory
3038 * intersects with the init section. Note that all other boot consoles will
3039 * get unregistred when the real preferred console is registered.
3041 static int __init printk_late_init(void)
3043 struct console *con;
3046 for_each_console(con) {
3047 if (!(con->flags & CON_BOOT))
3050 /* Check addresses that might be used for enabled consoles. */
3051 if (init_section_intersects(con, sizeof(*con)) ||
3052 init_section_contains(con->write, 0) ||
3053 init_section_contains(con->read, 0) ||
3054 init_section_contains(con->device, 0) ||
3055 init_section_contains(con->unblank, 0) ||
3056 init_section_contains(con->data, 0)) {
3058 * Please, consider moving the reported consoles out
3059 * of the init section.
3061 pr_warn("bootconsole [%s%d] uses init memory and must be disabled even before the real one is ready\n",
3062 con->name, con->index);
3063 unregister_console(con);
3066 ret = cpuhp_setup_state_nocalls(CPUHP_PRINTK_DEAD, "printk:dead", NULL,
3067 console_cpu_notify);
3069 ret = cpuhp_setup_state_nocalls(CPUHP_AP_ONLINE_DYN, "printk:online",
3070 console_cpu_notify, NULL);
3074 late_initcall(printk_late_init);
3076 #if defined CONFIG_PRINTK
3078 * Delayed printk version, for scheduler-internal messages:
3080 #define PRINTK_PENDING_WAKEUP 0x01
3081 #define PRINTK_PENDING_OUTPUT 0x02
3083 static DEFINE_PER_CPU(int, printk_pending);
3085 static void wake_up_klogd_work_func(struct irq_work *irq_work)
3087 int pending = __this_cpu_xchg(printk_pending, 0);
3089 if (pending & PRINTK_PENDING_OUTPUT) {
3090 /* If trylock fails, someone else is doing the printing */
3091 if (console_trylock())
3095 if (pending & PRINTK_PENDING_WAKEUP)
3096 wake_up_interruptible(&log_wait);
3099 static DEFINE_PER_CPU(struct irq_work, wake_up_klogd_work) =
3100 IRQ_WORK_INIT_LAZY(wake_up_klogd_work_func);
3102 void wake_up_klogd(void)
3104 if (!printk_percpu_data_ready())
3108 if (waitqueue_active(&log_wait)) {
3109 this_cpu_or(printk_pending, PRINTK_PENDING_WAKEUP);
3110 irq_work_queue(this_cpu_ptr(&wake_up_klogd_work));
3115 void defer_console_output(void)
3117 if (!printk_percpu_data_ready())
3121 __this_cpu_or(printk_pending, PRINTK_PENDING_OUTPUT);
3122 irq_work_queue(this_cpu_ptr(&wake_up_klogd_work));
3126 int vprintk_deferred(const char *fmt, va_list args)
3130 r = vprintk_emit(0, LOGLEVEL_SCHED, NULL, fmt, args);
3131 defer_console_output();
3136 int printk_deferred(const char *fmt, ...)
3141 va_start(args, fmt);
3142 r = vprintk_deferred(fmt, args);
3149 * printk rate limiting, lifted from the networking subsystem.
3151 * This enforces a rate limit: not more than 10 kernel messages
3152 * every 5s to make a denial-of-service attack impossible.
3154 DEFINE_RATELIMIT_STATE(printk_ratelimit_state, 5 * HZ, 10);
3156 int __printk_ratelimit(const char *func)
3158 return ___ratelimit(&printk_ratelimit_state, func);
3160 EXPORT_SYMBOL(__printk_ratelimit);
3163 * printk_timed_ratelimit - caller-controlled printk ratelimiting
3164 * @caller_jiffies: pointer to caller's state
3165 * @interval_msecs: minimum interval between prints
3167 * printk_timed_ratelimit() returns true if more than @interval_msecs
3168 * milliseconds have elapsed since the last time printk_timed_ratelimit()
3171 bool printk_timed_ratelimit(unsigned long *caller_jiffies,
3172 unsigned int interval_msecs)
3174 unsigned long elapsed = jiffies - *caller_jiffies;
3176 if (*caller_jiffies && elapsed <= msecs_to_jiffies(interval_msecs))
3179 *caller_jiffies = jiffies;
3182 EXPORT_SYMBOL(printk_timed_ratelimit);
3184 static DEFINE_SPINLOCK(dump_list_lock);
3185 static LIST_HEAD(dump_list);
3188 * kmsg_dump_register - register a kernel log dumper.
3189 * @dumper: pointer to the kmsg_dumper structure
3191 * Adds a kernel log dumper to the system. The dump callback in the
3192 * structure will be called when the kernel oopses or panics and must be
3193 * set. Returns zero on success and %-EINVAL or %-EBUSY otherwise.
3195 int kmsg_dump_register(struct kmsg_dumper *dumper)
3197 unsigned long flags;
3200 /* The dump callback needs to be set */
3204 spin_lock_irqsave(&dump_list_lock, flags);
3205 /* Don't allow registering multiple times */
3206 if (!dumper->registered) {
3207 dumper->registered = 1;
3208 list_add_tail_rcu(&dumper->list, &dump_list);
3211 spin_unlock_irqrestore(&dump_list_lock, flags);
3215 EXPORT_SYMBOL_GPL(kmsg_dump_register);
3218 * kmsg_dump_unregister - unregister a kmsg dumper.
3219 * @dumper: pointer to the kmsg_dumper structure
3221 * Removes a dump device from the system. Returns zero on success and
3222 * %-EINVAL otherwise.
3224 int kmsg_dump_unregister(struct kmsg_dumper *dumper)
3226 unsigned long flags;
3229 spin_lock_irqsave(&dump_list_lock, flags);
3230 if (dumper->registered) {
3231 dumper->registered = 0;
3232 list_del_rcu(&dumper->list);
3235 spin_unlock_irqrestore(&dump_list_lock, flags);
3240 EXPORT_SYMBOL_GPL(kmsg_dump_unregister);
3242 static bool always_kmsg_dump;
3243 module_param_named(always_kmsg_dump, always_kmsg_dump, bool, S_IRUGO | S_IWUSR);
3245 const char *kmsg_dump_reason_str(enum kmsg_dump_reason reason)
3248 case KMSG_DUMP_PANIC:
3250 case KMSG_DUMP_OOPS:
3252 case KMSG_DUMP_EMERG:
3254 case KMSG_DUMP_SHUTDOWN:
3260 EXPORT_SYMBOL_GPL(kmsg_dump_reason_str);
3263 * kmsg_dump - dump kernel log to kernel message dumpers.
3264 * @reason: the reason (oops, panic etc) for dumping
3266 * Call each of the registered dumper's dump() callback, which can
3267 * retrieve the kmsg records with kmsg_dump_get_line() or
3268 * kmsg_dump_get_buffer().
3270 void kmsg_dump(enum kmsg_dump_reason reason)
3272 struct kmsg_dumper *dumper;
3273 unsigned long flags;
3276 list_for_each_entry_rcu(dumper, &dump_list, list) {
3277 enum kmsg_dump_reason max_reason = dumper->max_reason;
3280 * If client has not provided a specific max_reason, default
3281 * to KMSG_DUMP_OOPS, unless always_kmsg_dump was set.
3283 if (max_reason == KMSG_DUMP_UNDEF) {
3284 max_reason = always_kmsg_dump ? KMSG_DUMP_MAX :
3287 if (reason > max_reason)
3290 /* initialize iterator with data about the stored records */
3291 dumper->active = true;
3293 logbuf_lock_irqsave(flags);
3294 dumper->cur_seq = clear_seq;
3295 dumper->next_seq = prb_next_seq(prb);
3296 logbuf_unlock_irqrestore(flags);
3298 /* invoke dumper which will iterate over records */
3299 dumper->dump(dumper, reason);
3301 /* reset iterator */
3302 dumper->active = false;
3308 * kmsg_dump_get_line_nolock - retrieve one kmsg log line (unlocked version)
3309 * @dumper: registered kmsg dumper
3310 * @syslog: include the "<4>" prefixes
3311 * @line: buffer to copy the line to
3312 * @size: maximum size of the buffer
3313 * @len: length of line placed into buffer
3315 * Start at the beginning of the kmsg buffer, with the oldest kmsg
3316 * record, and copy one record into the provided buffer.
3318 * Consecutive calls will return the next available record moving
3319 * towards the end of the buffer with the youngest messages.
3321 * A return value of FALSE indicates that there are no more records to
3324 * The function is similar to kmsg_dump_get_line(), but grabs no locks.
3326 bool kmsg_dump_get_line_nolock(struct kmsg_dumper *dumper, bool syslog,
3327 char *line, size_t size, size_t *len)
3329 struct printk_info info;
3330 unsigned int line_count;
3331 struct printk_record r;
3335 prb_rec_init_rd(&r, &info, line, size);
3337 if (!dumper->active)
3340 /* Read text or count text lines? */
3342 if (!prb_read_valid(prb, dumper->cur_seq, &r))
3344 l = record_print_text(&r, syslog, printk_time);
3346 if (!prb_read_valid_info(prb, dumper->cur_seq,
3347 &info, &line_count)) {
3350 l = get_record_print_text_size(&info, line_count, syslog,
3355 dumper->cur_seq = r.info->seq + 1;
3364 * kmsg_dump_get_line - retrieve one kmsg log line
3365 * @dumper: registered kmsg dumper
3366 * @syslog: include the "<4>" prefixes
3367 * @line: buffer to copy the line to
3368 * @size: maximum size of the buffer
3369 * @len: length of line placed into buffer
3371 * Start at the beginning of the kmsg buffer, with the oldest kmsg
3372 * record, and copy one record into the provided buffer.
3374 * Consecutive calls will return the next available record moving
3375 * towards the end of the buffer with the youngest messages.
3377 * A return value of FALSE indicates that there are no more records to
3380 bool kmsg_dump_get_line(struct kmsg_dumper *dumper, bool syslog,
3381 char *line, size_t size, size_t *len)
3383 unsigned long flags;
3386 logbuf_lock_irqsave(flags);
3387 ret = kmsg_dump_get_line_nolock(dumper, syslog, line, size, len);
3388 logbuf_unlock_irqrestore(flags);
3392 EXPORT_SYMBOL_GPL(kmsg_dump_get_line);
3395 * kmsg_dump_get_buffer - copy kmsg log lines
3396 * @dumper: registered kmsg dumper
3397 * @syslog: include the "<4>" prefixes
3398 * @buf: buffer to copy the line to
3399 * @size: maximum size of the buffer
3400 * @len: length of line placed into buffer
3402 * Start at the end of the kmsg buffer and fill the provided buffer
3403 * with as many of the *youngest* kmsg records that fit into it.
3404 * If the buffer is large enough, all available kmsg records will be
3405 * copied with a single call.
3407 * Consecutive calls will fill the buffer with the next block of
3408 * available older records, not including the earlier retrieved ones.
3410 * A return value of FALSE indicates that there are no more records to
3413 bool kmsg_dump_get_buffer(struct kmsg_dumper *dumper, bool syslog,
3414 char *buf, size_t size, size_t *len)
3416 struct printk_info info;
3417 unsigned int line_count;
3418 struct printk_record r;
3419 unsigned long flags;
3424 bool time = printk_time;
3426 prb_rec_init_rd(&r, &info, buf, size);
3428 if (!dumper->active || !buf || !size)
3431 logbuf_lock_irqsave(flags);
3432 if (dumper->cur_seq < prb_first_valid_seq(prb)) {
3433 /* messages are gone, move to first available one */
3434 dumper->cur_seq = prb_first_valid_seq(prb);
3438 if (dumper->cur_seq >= dumper->next_seq) {
3439 logbuf_unlock_irqrestore(flags);
3443 /* calculate length of entire buffer */
3444 seq = dumper->cur_seq;
3445 while (prb_read_valid_info(prb, seq, &info, &line_count)) {
3446 if (r.info->seq >= dumper->next_seq)
3448 l += get_record_print_text_size(&info, line_count, syslog, time);
3449 seq = r.info->seq + 1;
3452 /* move first record forward until length fits into the buffer */
3453 seq = dumper->cur_seq;
3454 while (l >= size && prb_read_valid_info(prb, seq,
3455 &info, &line_count)) {
3456 if (r.info->seq >= dumper->next_seq)
3458 l -= get_record_print_text_size(&info, line_count, syslog, time);
3459 seq = r.info->seq + 1;
3462 /* last message in next interation */
3465 /* actually read text into the buffer now */
3467 while (prb_read_valid(prb, seq, &r)) {
3468 if (r.info->seq >= dumper->next_seq)
3471 l += record_print_text(&r, syslog, time);
3473 /* adjust record to store to remaining buffer space */
3474 prb_rec_init_rd(&r, &info, buf + l, size - l);
3476 seq = r.info->seq + 1;
3479 dumper->next_seq = next_seq;
3481 logbuf_unlock_irqrestore(flags);
3487 EXPORT_SYMBOL_GPL(kmsg_dump_get_buffer);
3490 * kmsg_dump_rewind_nolock - reset the iterator (unlocked version)
3491 * @dumper: registered kmsg dumper
3493 * Reset the dumper's iterator so that kmsg_dump_get_line() and
3494 * kmsg_dump_get_buffer() can be called again and used multiple
3495 * times within the same dumper.dump() callback.
3497 * The function is similar to kmsg_dump_rewind(), but grabs no locks.
3499 void kmsg_dump_rewind_nolock(struct kmsg_dumper *dumper)
3501 dumper->cur_seq = clear_seq;
3502 dumper->next_seq = prb_next_seq(prb);
3506 * kmsg_dump_rewind - reset the iterator
3507 * @dumper: registered kmsg dumper
3509 * Reset the dumper's iterator so that kmsg_dump_get_line() and
3510 * kmsg_dump_get_buffer() can be called again and used multiple
3511 * times within the same dumper.dump() callback.
3513 void kmsg_dump_rewind(struct kmsg_dumper *dumper)
3515 unsigned long flags;
3517 logbuf_lock_irqsave(flags);
3518 kmsg_dump_rewind_nolock(dumper);
3519 logbuf_unlock_irqrestore(flags);
3521 EXPORT_SYMBOL_GPL(kmsg_dump_rewind);