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 (r->info->seq != user->seq) {
739 /* our last seen message is gone, return error and reset */
740 user->seq = r->info->seq;
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;
815 struct printk_info info;
819 return EPOLLERR|EPOLLNVAL;
821 poll_wait(file, &log_wait, wait);
824 if (prb_read_valid_info(prb, user->seq, &info, NULL)) {
825 /* return error when data has vanished underneath us */
826 if (info.seq != user->seq)
827 ret = EPOLLIN|EPOLLRDNORM|EPOLLERR|EPOLLPRI;
829 ret = EPOLLIN|EPOLLRDNORM;
836 static int devkmsg_open(struct inode *inode, struct file *file)
838 struct devkmsg_user *user;
841 if (devkmsg_log & DEVKMSG_LOG_MASK_OFF)
844 /* write-only does not need any file context */
845 if ((file->f_flags & O_ACCMODE) != O_WRONLY) {
846 err = check_syslog_permissions(SYSLOG_ACTION_READ_ALL,
852 user = kmalloc(sizeof(struct devkmsg_user), GFP_KERNEL);
856 ratelimit_default_init(&user->rs);
857 ratelimit_set_flags(&user->rs, RATELIMIT_MSG_ON_RELEASE);
859 mutex_init(&user->lock);
861 prb_rec_init_rd(&user->record, &user->info,
862 &user->text_buf[0], sizeof(user->text_buf));
865 user->seq = prb_first_valid_seq(prb);
868 file->private_data = user;
872 static int devkmsg_release(struct inode *inode, struct file *file)
874 struct devkmsg_user *user = file->private_data;
879 ratelimit_state_exit(&user->rs);
881 mutex_destroy(&user->lock);
886 const struct file_operations kmsg_fops = {
887 .open = devkmsg_open,
888 .read = devkmsg_read,
889 .write_iter = devkmsg_write,
890 .llseek = devkmsg_llseek,
891 .poll = devkmsg_poll,
892 .release = devkmsg_release,
895 #ifdef CONFIG_CRASH_CORE
897 * This appends the listed symbols to /proc/vmcore
899 * /proc/vmcore is used by various utilities, like crash and makedumpfile to
900 * obtain access to symbols that are otherwise very difficult to locate. These
901 * symbols are specifically used so that utilities can access and extract the
902 * dmesg log from a vmcore file after a crash.
904 void log_buf_vmcoreinfo_setup(void)
906 struct dev_printk_info *dev_info = NULL;
908 VMCOREINFO_SYMBOL(prb);
909 VMCOREINFO_SYMBOL(printk_rb_static);
910 VMCOREINFO_SYMBOL(clear_seq);
913 * Export struct size and field offsets. User space tools can
914 * parse it and detect any changes to structure down the line.
917 VMCOREINFO_STRUCT_SIZE(printk_ringbuffer);
918 VMCOREINFO_OFFSET(printk_ringbuffer, desc_ring);
919 VMCOREINFO_OFFSET(printk_ringbuffer, text_data_ring);
920 VMCOREINFO_OFFSET(printk_ringbuffer, fail);
922 VMCOREINFO_STRUCT_SIZE(prb_desc_ring);
923 VMCOREINFO_OFFSET(prb_desc_ring, count_bits);
924 VMCOREINFO_OFFSET(prb_desc_ring, descs);
925 VMCOREINFO_OFFSET(prb_desc_ring, infos);
926 VMCOREINFO_OFFSET(prb_desc_ring, head_id);
927 VMCOREINFO_OFFSET(prb_desc_ring, tail_id);
929 VMCOREINFO_STRUCT_SIZE(prb_desc);
930 VMCOREINFO_OFFSET(prb_desc, state_var);
931 VMCOREINFO_OFFSET(prb_desc, text_blk_lpos);
933 VMCOREINFO_STRUCT_SIZE(prb_data_blk_lpos);
934 VMCOREINFO_OFFSET(prb_data_blk_lpos, begin);
935 VMCOREINFO_OFFSET(prb_data_blk_lpos, next);
937 VMCOREINFO_STRUCT_SIZE(printk_info);
938 VMCOREINFO_OFFSET(printk_info, seq);
939 VMCOREINFO_OFFSET(printk_info, ts_nsec);
940 VMCOREINFO_OFFSET(printk_info, text_len);
941 VMCOREINFO_OFFSET(printk_info, caller_id);
942 VMCOREINFO_OFFSET(printk_info, dev_info);
944 VMCOREINFO_STRUCT_SIZE(dev_printk_info);
945 VMCOREINFO_OFFSET(dev_printk_info, subsystem);
946 VMCOREINFO_LENGTH(printk_info_subsystem, sizeof(dev_info->subsystem));
947 VMCOREINFO_OFFSET(dev_printk_info, device);
948 VMCOREINFO_LENGTH(printk_info_device, sizeof(dev_info->device));
950 VMCOREINFO_STRUCT_SIZE(prb_data_ring);
951 VMCOREINFO_OFFSET(prb_data_ring, size_bits);
952 VMCOREINFO_OFFSET(prb_data_ring, data);
953 VMCOREINFO_OFFSET(prb_data_ring, head_lpos);
954 VMCOREINFO_OFFSET(prb_data_ring, tail_lpos);
956 VMCOREINFO_SIZE(atomic_long_t);
957 VMCOREINFO_TYPE_OFFSET(atomic_long_t, counter);
961 /* requested log_buf_len from kernel cmdline */
962 static unsigned long __initdata new_log_buf_len;
964 /* we practice scaling the ring buffer by powers of 2 */
965 static void __init log_buf_len_update(u64 size)
967 if (size > (u64)LOG_BUF_LEN_MAX) {
968 size = (u64)LOG_BUF_LEN_MAX;
969 pr_err("log_buf over 2G is not supported.\n");
973 size = roundup_pow_of_two(size);
974 if (size > log_buf_len)
975 new_log_buf_len = (unsigned long)size;
978 /* save requested log_buf_len since it's too early to process it */
979 static int __init log_buf_len_setup(char *str)
986 size = memparse(str, &str);
988 log_buf_len_update(size);
992 early_param("log_buf_len", log_buf_len_setup);
995 #define __LOG_CPU_MAX_BUF_LEN (1 << CONFIG_LOG_CPU_MAX_BUF_SHIFT)
997 static void __init log_buf_add_cpu(void)
999 unsigned int cpu_extra;
1002 * archs should set up cpu_possible_bits properly with
1003 * set_cpu_possible() after setup_arch() but just in
1004 * case lets ensure this is valid.
1006 if (num_possible_cpus() == 1)
1009 cpu_extra = (num_possible_cpus() - 1) * __LOG_CPU_MAX_BUF_LEN;
1011 /* by default this will only continue through for large > 64 CPUs */
1012 if (cpu_extra <= __LOG_BUF_LEN / 2)
1015 pr_info("log_buf_len individual max cpu contribution: %d bytes\n",
1016 __LOG_CPU_MAX_BUF_LEN);
1017 pr_info("log_buf_len total cpu_extra contributions: %d bytes\n",
1019 pr_info("log_buf_len min size: %d bytes\n", __LOG_BUF_LEN);
1021 log_buf_len_update(cpu_extra + __LOG_BUF_LEN);
1023 #else /* !CONFIG_SMP */
1024 static inline void log_buf_add_cpu(void) {}
1025 #endif /* CONFIG_SMP */
1027 static void __init set_percpu_data_ready(void)
1030 /* Make sure we set this flag only after printk_safe() init is done */
1032 __printk_percpu_data_ready = true;
1035 static unsigned int __init add_to_rb(struct printk_ringbuffer *rb,
1036 struct printk_record *r)
1038 struct prb_reserved_entry e;
1039 struct printk_record dest_r;
1041 prb_rec_init_wr(&dest_r, r->info->text_len);
1043 if (!prb_reserve(&e, rb, &dest_r))
1046 memcpy(&dest_r.text_buf[0], &r->text_buf[0], r->info->text_len);
1047 dest_r.info->text_len = r->info->text_len;
1048 dest_r.info->facility = r->info->facility;
1049 dest_r.info->level = r->info->level;
1050 dest_r.info->flags = r->info->flags;
1051 dest_r.info->ts_nsec = r->info->ts_nsec;
1052 dest_r.info->caller_id = r->info->caller_id;
1053 memcpy(&dest_r.info->dev_info, &r->info->dev_info, sizeof(dest_r.info->dev_info));
1055 prb_final_commit(&e);
1057 return prb_record_text_space(&e);
1060 static char setup_text_buf[LOG_LINE_MAX] __initdata;
1062 void __init setup_log_buf(int early)
1064 struct printk_info *new_infos;
1065 unsigned int new_descs_count;
1066 struct prb_desc *new_descs;
1067 struct printk_info info;
1068 struct printk_record r;
1069 size_t new_descs_size;
1070 size_t new_infos_size;
1071 unsigned long flags;
1077 * Some archs call setup_log_buf() multiple times - first is very
1078 * early, e.g. from setup_arch(), and second - when percpu_areas
1082 set_percpu_data_ready();
1084 if (log_buf != __log_buf)
1087 if (!early && !new_log_buf_len)
1090 if (!new_log_buf_len)
1093 new_descs_count = new_log_buf_len >> PRB_AVGBITS;
1094 if (new_descs_count == 0) {
1095 pr_err("new_log_buf_len: %lu too small\n", new_log_buf_len);
1099 new_log_buf = memblock_alloc(new_log_buf_len, LOG_ALIGN);
1100 if (unlikely(!new_log_buf)) {
1101 pr_err("log_buf_len: %lu text bytes not available\n",
1106 new_descs_size = new_descs_count * sizeof(struct prb_desc);
1107 new_descs = memblock_alloc(new_descs_size, LOG_ALIGN);
1108 if (unlikely(!new_descs)) {
1109 pr_err("log_buf_len: %zu desc bytes not available\n",
1111 goto err_free_log_buf;
1114 new_infos_size = new_descs_count * sizeof(struct printk_info);
1115 new_infos = memblock_alloc(new_infos_size, LOG_ALIGN);
1116 if (unlikely(!new_infos)) {
1117 pr_err("log_buf_len: %zu info bytes not available\n",
1119 goto err_free_descs;
1122 prb_rec_init_rd(&r, &info, &setup_text_buf[0], sizeof(setup_text_buf));
1124 prb_init(&printk_rb_dynamic,
1125 new_log_buf, ilog2(new_log_buf_len),
1126 new_descs, ilog2(new_descs_count),
1129 printk_safe_enter_irqsave(flags);
1131 log_buf_len = new_log_buf_len;
1132 log_buf = new_log_buf;
1133 new_log_buf_len = 0;
1135 free = __LOG_BUF_LEN;
1136 prb_for_each_record(0, &printk_rb_static, seq, &r)
1137 free -= add_to_rb(&printk_rb_dynamic, &r);
1140 * This is early enough that everything is still running on the
1141 * boot CPU and interrupts are disabled. So no new messages will
1142 * appear during the transition to the dynamic buffer.
1144 prb = &printk_rb_dynamic;
1146 printk_safe_exit_irqrestore(flags);
1148 if (seq != prb_next_seq(&printk_rb_static)) {
1149 pr_err("dropped %llu messages\n",
1150 prb_next_seq(&printk_rb_static) - seq);
1153 pr_info("log_buf_len: %u bytes\n", log_buf_len);
1154 pr_info("early log buf free: %u(%u%%)\n",
1155 free, (free * 100) / __LOG_BUF_LEN);
1159 memblock_free(__pa(new_descs), new_descs_size);
1161 memblock_free(__pa(new_log_buf), new_log_buf_len);
1164 static bool __read_mostly ignore_loglevel;
1166 static int __init ignore_loglevel_setup(char *str)
1168 ignore_loglevel = true;
1169 pr_info("debug: ignoring loglevel setting.\n");
1174 early_param("ignore_loglevel", ignore_loglevel_setup);
1175 module_param(ignore_loglevel, bool, S_IRUGO | S_IWUSR);
1176 MODULE_PARM_DESC(ignore_loglevel,
1177 "ignore loglevel setting (prints all kernel messages to the console)");
1179 static bool suppress_message_printing(int level)
1181 return (level >= console_loglevel && !ignore_loglevel);
1184 #ifdef CONFIG_BOOT_PRINTK_DELAY
1186 static int boot_delay; /* msecs delay after each printk during bootup */
1187 static unsigned long long loops_per_msec; /* based on boot_delay */
1189 static int __init boot_delay_setup(char *str)
1193 lpj = preset_lpj ? preset_lpj : 1000000; /* some guess */
1194 loops_per_msec = (unsigned long long)lpj / 1000 * HZ;
1196 get_option(&str, &boot_delay);
1197 if (boot_delay > 10 * 1000)
1200 pr_debug("boot_delay: %u, preset_lpj: %ld, lpj: %lu, "
1201 "HZ: %d, loops_per_msec: %llu\n",
1202 boot_delay, preset_lpj, lpj, HZ, loops_per_msec);
1205 early_param("boot_delay", boot_delay_setup);
1207 static void boot_delay_msec(int level)
1209 unsigned long long k;
1210 unsigned long timeout;
1212 if ((boot_delay == 0 || system_state >= SYSTEM_RUNNING)
1213 || suppress_message_printing(level)) {
1217 k = (unsigned long long)loops_per_msec * boot_delay;
1219 timeout = jiffies + msecs_to_jiffies(boot_delay);
1224 * use (volatile) jiffies to prevent
1225 * compiler reduction; loop termination via jiffies
1226 * is secondary and may or may not happen.
1228 if (time_after(jiffies, timeout))
1230 touch_nmi_watchdog();
1234 static inline void boot_delay_msec(int level)
1239 static bool printk_time = IS_ENABLED(CONFIG_PRINTK_TIME);
1240 module_param_named(time, printk_time, bool, S_IRUGO | S_IWUSR);
1242 static size_t print_syslog(unsigned int level, char *buf)
1244 return sprintf(buf, "<%u>", level);
1247 static size_t print_time(u64 ts, char *buf)
1249 unsigned long rem_nsec = do_div(ts, 1000000000);
1251 return sprintf(buf, "[%5lu.%06lu]",
1252 (unsigned long)ts, rem_nsec / 1000);
1255 #ifdef CONFIG_PRINTK_CALLER
1256 static size_t print_caller(u32 id, char *buf)
1260 snprintf(caller, sizeof(caller), "%c%u",
1261 id & 0x80000000 ? 'C' : 'T', id & ~0x80000000);
1262 return sprintf(buf, "[%6s]", caller);
1265 #define print_caller(id, buf) 0
1268 static size_t info_print_prefix(const struct printk_info *info, bool syslog,
1269 bool time, char *buf)
1274 len = print_syslog((info->facility << 3) | info->level, buf);
1277 len += print_time(info->ts_nsec, buf + len);
1279 len += print_caller(info->caller_id, buf + len);
1281 if (IS_ENABLED(CONFIG_PRINTK_CALLER) || time) {
1290 * Prepare the record for printing. The text is shifted within the given
1291 * buffer to avoid a need for another one. The following operations are
1294 * - Add prefix for each line.
1295 * - Drop truncated lines that no longer fit into the buffer.
1296 * - Add the trailing newline that has been removed in vprintk_store().
1297 * - Add a string terminator.
1299 * Since the produced string is always terminated, the maximum possible
1300 * return value is @r->text_buf_size - 1;
1302 * Return: The length of the updated/prepared text, including the added
1303 * prefixes and the newline. The terminator is not counted. The dropped
1304 * line(s) are not counted.
1306 static size_t record_print_text(struct printk_record *r, bool syslog,
1309 size_t text_len = r->info->text_len;
1310 size_t buf_size = r->text_buf_size;
1311 char *text = r->text_buf;
1312 char prefix[PREFIX_MAX];
1313 bool truncated = false;
1320 * If the message was truncated because the buffer was not large
1321 * enough, treat the available text as if it were the full text.
1323 if (text_len > buf_size)
1324 text_len = buf_size;
1326 prefix_len = info_print_prefix(r->info, syslog, time, prefix);
1329 * @text_len: bytes of unprocessed text
1330 * @line_len: bytes of current line _without_ newline
1331 * @text: pointer to beginning of current line
1332 * @len: number of bytes prepared in r->text_buf
1335 next = memchr(text, '\n', text_len);
1337 line_len = next - text;
1339 /* Drop truncated line(s). */
1342 line_len = text_len;
1346 * Truncate the text if there is not enough space to add the
1347 * prefix and a trailing newline and a terminator.
1349 if (len + prefix_len + text_len + 1 + 1 > buf_size) {
1350 /* Drop even the current line if no space. */
1351 if (len + prefix_len + line_len + 1 + 1 > buf_size)
1354 text_len = buf_size - len - prefix_len - 1 - 1;
1358 memmove(text + prefix_len, text, text_len);
1359 memcpy(text, prefix, prefix_len);
1362 * Increment the prepared length to include the text and
1363 * prefix that were just moved+copied. Also increment for the
1364 * newline at the end of this line. If this is the last line,
1365 * there is no newline, but it will be added immediately below.
1367 len += prefix_len + line_len + 1;
1368 if (text_len == line_len) {
1370 * This is the last line. Add the trailing newline
1371 * removed in vprintk_store().
1373 text[prefix_len + line_len] = '\n';
1378 * Advance beyond the added prefix and the related line with
1381 text += prefix_len + line_len + 1;
1384 * The remaining text has only decreased by the line with its
1387 * Note that @text_len can become zero. It happens when @text
1388 * ended with a newline (either due to truncation or the
1389 * original string ending with "\n\n"). The loop is correctly
1390 * repeated and (if not truncated) an empty line with a prefix
1393 text_len -= line_len + 1;
1397 * If a buffer was provided, it will be terminated. Space for the
1398 * string terminator is guaranteed to be available. The terminator is
1399 * not counted in the return value.
1402 r->text_buf[len] = 0;
1407 static size_t get_record_print_text_size(struct printk_info *info,
1408 unsigned int line_count,
1409 bool syslog, bool time)
1411 char prefix[PREFIX_MAX];
1414 prefix_len = info_print_prefix(info, syslog, time, prefix);
1417 * Each line will be preceded with a prefix. The intermediate
1418 * newlines are already within the text, but a final trailing
1419 * newline will be added.
1421 return ((prefix_len * line_count) + info->text_len + 1);
1424 static int syslog_print(char __user *buf, int size)
1426 struct printk_info info;
1427 struct printk_record r;
1431 text = kmalloc(LOG_LINE_MAX + PREFIX_MAX, GFP_KERNEL);
1435 prb_rec_init_rd(&r, &info, text, LOG_LINE_MAX + PREFIX_MAX);
1442 if (!prb_read_valid(prb, syslog_seq, &r)) {
1443 logbuf_unlock_irq();
1446 if (r.info->seq != syslog_seq) {
1447 /* message is gone, move to next valid one */
1448 syslog_seq = r.info->seq;
1453 * To keep reading/counting partial line consistent,
1454 * use printk_time value as of the beginning of a line.
1456 if (!syslog_partial)
1457 syslog_time = printk_time;
1459 skip = syslog_partial;
1460 n = record_print_text(&r, true, syslog_time);
1461 if (n - syslog_partial <= size) {
1462 /* message fits into buffer, move forward */
1463 syslog_seq = r.info->seq + 1;
1464 n -= syslog_partial;
1467 /* partial read(), remember position */
1469 syslog_partial += n;
1472 logbuf_unlock_irq();
1477 if (copy_to_user(buf, text + skip, n)) {
1492 static int syslog_print_all(char __user *buf, int size, bool clear)
1494 struct printk_info info;
1495 unsigned int line_count;
1496 struct printk_record r;
1502 text = kmalloc(LOG_LINE_MAX + PREFIX_MAX, GFP_KERNEL);
1509 * Find first record that fits, including all following records,
1510 * into the user-provided buffer for this dump.
1512 prb_for_each_info(clear_seq, prb, seq, &info, &line_count)
1513 len += get_record_print_text_size(&info, line_count, true, time);
1515 /* move first record forward until length fits into the buffer */
1516 prb_for_each_info(clear_seq, prb, seq, &info, &line_count) {
1519 len -= get_record_print_text_size(&info, line_count, true, time);
1522 prb_rec_init_rd(&r, &info, text, LOG_LINE_MAX + PREFIX_MAX);
1525 prb_for_each_record(seq, prb, seq, &r) {
1528 textlen = record_print_text(&r, true, time);
1530 if (len + textlen > size) {
1535 logbuf_unlock_irq();
1536 if (copy_to_user(buf + len, text, textlen))
1548 logbuf_unlock_irq();
1554 static void syslog_clear(void)
1557 clear_seq = prb_next_seq(prb);
1558 logbuf_unlock_irq();
1561 int do_syslog(int type, char __user *buf, int len, int source)
1563 struct printk_info info;
1565 static int saved_console_loglevel = LOGLEVEL_DEFAULT;
1568 error = check_syslog_permissions(type, source);
1573 case SYSLOG_ACTION_CLOSE: /* Close log */
1575 case SYSLOG_ACTION_OPEN: /* Open log */
1577 case SYSLOG_ACTION_READ: /* Read from log */
1578 if (!buf || len < 0)
1582 if (!access_ok(buf, len))
1584 error = wait_event_interruptible(log_wait,
1585 prb_read_valid(prb, syslog_seq, NULL));
1588 error = syslog_print(buf, len);
1590 /* Read/clear last kernel messages */
1591 case SYSLOG_ACTION_READ_CLEAR:
1594 /* Read last kernel messages */
1595 case SYSLOG_ACTION_READ_ALL:
1596 if (!buf || len < 0)
1600 if (!access_ok(buf, len))
1602 error = syslog_print_all(buf, len, clear);
1604 /* Clear ring buffer */
1605 case SYSLOG_ACTION_CLEAR:
1608 /* Disable logging to console */
1609 case SYSLOG_ACTION_CONSOLE_OFF:
1610 if (saved_console_loglevel == LOGLEVEL_DEFAULT)
1611 saved_console_loglevel = console_loglevel;
1612 console_loglevel = minimum_console_loglevel;
1614 /* Enable logging to console */
1615 case SYSLOG_ACTION_CONSOLE_ON:
1616 if (saved_console_loglevel != LOGLEVEL_DEFAULT) {
1617 console_loglevel = saved_console_loglevel;
1618 saved_console_loglevel = LOGLEVEL_DEFAULT;
1621 /* Set level of messages printed to console */
1622 case SYSLOG_ACTION_CONSOLE_LEVEL:
1623 if (len < 1 || len > 8)
1625 if (len < minimum_console_loglevel)
1626 len = minimum_console_loglevel;
1627 console_loglevel = len;
1628 /* Implicitly re-enable logging to console */
1629 saved_console_loglevel = LOGLEVEL_DEFAULT;
1631 /* Number of chars in the log buffer */
1632 case SYSLOG_ACTION_SIZE_UNREAD:
1634 if (!prb_read_valid_info(prb, syslog_seq, &info, NULL)) {
1635 /* No unread messages. */
1636 logbuf_unlock_irq();
1639 if (info.seq != syslog_seq) {
1640 /* messages are gone, move to first one */
1641 syslog_seq = info.seq;
1644 if (source == SYSLOG_FROM_PROC) {
1646 * Short-cut for poll(/"proc/kmsg") which simply checks
1647 * for pending data, not the size; return the count of
1648 * records, not the length.
1650 error = prb_next_seq(prb) - syslog_seq;
1652 bool time = syslog_partial ? syslog_time : printk_time;
1653 unsigned int line_count;
1656 prb_for_each_info(syslog_seq, prb, seq, &info,
1658 error += get_record_print_text_size(&info, line_count,
1662 error -= syslog_partial;
1664 logbuf_unlock_irq();
1666 /* Size of the log buffer */
1667 case SYSLOG_ACTION_SIZE_BUFFER:
1668 error = log_buf_len;
1678 SYSCALL_DEFINE3(syslog, int, type, char __user *, buf, int, len)
1680 return do_syslog(type, buf, len, SYSLOG_FROM_READER);
1684 * Special console_lock variants that help to reduce the risk of soft-lockups.
1685 * They allow to pass console_lock to another printk() call using a busy wait.
1688 #ifdef CONFIG_LOCKDEP
1689 static struct lockdep_map console_owner_dep_map = {
1690 .name = "console_owner"
1694 static DEFINE_RAW_SPINLOCK(console_owner_lock);
1695 static struct task_struct *console_owner;
1696 static bool console_waiter;
1699 * console_lock_spinning_enable - mark beginning of code where another
1700 * thread might safely busy wait
1702 * This basically converts console_lock into a spinlock. This marks
1703 * the section where the console_lock owner can not sleep, because
1704 * there may be a waiter spinning (like a spinlock). Also it must be
1705 * ready to hand over the lock at the end of the section.
1707 static void console_lock_spinning_enable(void)
1709 raw_spin_lock(&console_owner_lock);
1710 console_owner = current;
1711 raw_spin_unlock(&console_owner_lock);
1713 /* The waiter may spin on us after setting console_owner */
1714 spin_acquire(&console_owner_dep_map, 0, 0, _THIS_IP_);
1718 * console_lock_spinning_disable_and_check - mark end of code where another
1719 * thread was able to busy wait and check if there is a waiter
1721 * This is called at the end of the section where spinning is allowed.
1722 * It has two functions. First, it is a signal that it is no longer
1723 * safe to start busy waiting for the lock. Second, it checks if
1724 * there is a busy waiter and passes the lock rights to her.
1726 * Important: Callers lose the lock if there was a busy waiter.
1727 * They must not touch items synchronized by console_lock
1730 * Return: 1 if the lock rights were passed, 0 otherwise.
1732 static int console_lock_spinning_disable_and_check(void)
1736 raw_spin_lock(&console_owner_lock);
1737 waiter = READ_ONCE(console_waiter);
1738 console_owner = NULL;
1739 raw_spin_unlock(&console_owner_lock);
1742 spin_release(&console_owner_dep_map, _THIS_IP_);
1746 /* The waiter is now free to continue */
1747 WRITE_ONCE(console_waiter, false);
1749 spin_release(&console_owner_dep_map, _THIS_IP_);
1752 * Hand off console_lock to waiter. The waiter will perform
1753 * the up(). After this, the waiter is the console_lock owner.
1755 mutex_release(&console_lock_dep_map, _THIS_IP_);
1760 * console_trylock_spinning - try to get console_lock by busy waiting
1762 * This allows to busy wait for the console_lock when the current
1763 * owner is running in specially marked sections. It means that
1764 * the current owner is running and cannot reschedule until it
1765 * is ready to lose the lock.
1767 * Return: 1 if we got the lock, 0 othrewise
1769 static int console_trylock_spinning(void)
1771 struct task_struct *owner = NULL;
1774 unsigned long flags;
1776 if (console_trylock())
1779 printk_safe_enter_irqsave(flags);
1781 raw_spin_lock(&console_owner_lock);
1782 owner = READ_ONCE(console_owner);
1783 waiter = READ_ONCE(console_waiter);
1784 if (!waiter && owner && owner != current) {
1785 WRITE_ONCE(console_waiter, true);
1788 raw_spin_unlock(&console_owner_lock);
1791 * If there is an active printk() writing to the
1792 * consoles, instead of having it write our data too,
1793 * see if we can offload that load from the active
1794 * printer, and do some printing ourselves.
1795 * Go into a spin only if there isn't already a waiter
1796 * spinning, and there is an active printer, and
1797 * that active printer isn't us (recursive printk?).
1800 printk_safe_exit_irqrestore(flags);
1804 /* We spin waiting for the owner to release us */
1805 spin_acquire(&console_owner_dep_map, 0, 0, _THIS_IP_);
1806 /* Owner will clear console_waiter on hand off */
1807 while (READ_ONCE(console_waiter))
1809 spin_release(&console_owner_dep_map, _THIS_IP_);
1811 printk_safe_exit_irqrestore(flags);
1813 * The owner passed the console lock to us.
1814 * Since we did not spin on console lock, annotate
1815 * this as a trylock. Otherwise lockdep will
1818 mutex_acquire(&console_lock_dep_map, 0, 1, _THIS_IP_);
1824 * Call the console drivers, asking them to write out
1825 * log_buf[start] to log_buf[end - 1].
1826 * The console_lock must be held.
1828 static void call_console_drivers(const char *ext_text, size_t ext_len,
1829 const char *text, size_t len)
1831 static char dropped_text[64];
1832 size_t dropped_len = 0;
1833 struct console *con;
1835 trace_console_rcuidle(text, len);
1837 if (!console_drivers)
1840 if (console_dropped) {
1841 dropped_len = snprintf(dropped_text, sizeof(dropped_text),
1842 "** %lu printk messages dropped **\n",
1844 console_dropped = 0;
1847 for_each_console(con) {
1848 if (exclusive_console && con != exclusive_console)
1850 if (!(con->flags & CON_ENABLED))
1854 if (!cpu_online(smp_processor_id()) &&
1855 !(con->flags & CON_ANYTIME))
1857 if (con->flags & CON_EXTENDED)
1858 con->write(con, ext_text, ext_len);
1861 con->write(con, dropped_text, dropped_len);
1862 con->write(con, text, len);
1867 int printk_delay_msec __read_mostly;
1869 static inline void printk_delay(void)
1871 if (unlikely(printk_delay_msec)) {
1872 int m = printk_delay_msec;
1876 touch_nmi_watchdog();
1881 static inline u32 printk_caller_id(void)
1883 return in_task() ? task_pid_nr(current) :
1884 0x80000000 + raw_smp_processor_id();
1888 * parse_prefix - Parse level and control flags.
1890 * @text: The terminated text message.
1891 * @level: A pointer to the current level value, will be updated.
1892 * @lflags: A pointer to the current log flags, will be updated.
1894 * @level may be NULL if the caller is not interested in the parsed value.
1895 * Otherwise the variable pointed to by @level must be set to
1896 * LOGLEVEL_DEFAULT in order to be updated with the parsed value.
1898 * @lflags may be NULL if the caller is not interested in the parsed value.
1899 * Otherwise the variable pointed to by @lflags will be OR'd with the parsed
1902 * Return: The length of the parsed level and control flags.
1904 static u16 parse_prefix(char *text, int *level, enum log_flags *lflags)
1910 kern_level = printk_get_level(text);
1914 switch (kern_level) {
1916 if (level && *level == LOGLEVEL_DEFAULT)
1917 *level = kern_level - '0';
1919 case 'c': /* KERN_CONT */
1921 *lflags |= LOG_CONT;
1931 static u16 printk_sprint(char *text, u16 size, int facility, enum log_flags *lflags,
1932 const char *fmt, va_list args)
1936 text_len = vscnprintf(text, size, fmt, args);
1938 /* Mark and strip a trailing newline. */
1939 if (text_len && text[text_len - 1] == '\n') {
1941 *lflags |= LOG_NEWLINE;
1944 /* Strip log level and control flags. */
1945 if (facility == 0) {
1948 prefix_len = parse_prefix(text, NULL, NULL);
1950 text_len -= prefix_len;
1951 memmove(text, text + prefix_len, text_len);
1959 int vprintk_store(int facility, int level,
1960 const struct dev_printk_info *dev_info,
1961 const char *fmt, va_list args)
1963 const u32 caller_id = printk_caller_id();
1964 struct prb_reserved_entry e;
1965 enum log_flags lflags = 0;
1966 struct printk_record r;
1967 u16 trunc_msg_len = 0;
1975 * Since the duration of printk() can vary depending on the message
1976 * and state of the ringbuffer, grab the timestamp now so that it is
1977 * close to the call of printk(). This provides a more deterministic
1978 * timestamp with respect to the caller.
1980 ts_nsec = local_clock();
1983 * The sprintf needs to come first since the syslog prefix might be
1984 * passed in as a parameter. An extra byte must be reserved so that
1985 * later the vscnprintf() into the reserved buffer has room for the
1986 * terminating '\0', which is not counted by vsnprintf().
1988 va_copy(args2, args);
1989 reserve_size = vsnprintf(&prefix_buf[0], sizeof(prefix_buf), fmt, args2) + 1;
1992 if (reserve_size > LOG_LINE_MAX)
1993 reserve_size = LOG_LINE_MAX;
1995 /* Extract log level or control flags. */
1997 parse_prefix(&prefix_buf[0], &level, &lflags);
1999 if (level == LOGLEVEL_DEFAULT)
2000 level = default_message_loglevel;
2003 lflags |= LOG_NEWLINE;
2005 if (lflags & LOG_CONT) {
2006 prb_rec_init_wr(&r, reserve_size);
2007 if (prb_reserve_in_last(&e, prb, &r, caller_id, LOG_LINE_MAX)) {
2008 text_len = printk_sprint(&r.text_buf[r.info->text_len], reserve_size,
2009 facility, &lflags, fmt, args);
2010 r.info->text_len += text_len;
2012 if (lflags & LOG_NEWLINE) {
2013 r.info->flags |= LOG_NEWLINE;
2014 prb_final_commit(&e);
2024 * Explicitly initialize the record before every prb_reserve() call.
2025 * prb_reserve_in_last() and prb_reserve() purposely invalidate the
2026 * structure when they fail.
2028 prb_rec_init_wr(&r, reserve_size);
2029 if (!prb_reserve(&e, prb, &r)) {
2030 /* truncate the message if it is too long for empty buffer */
2031 truncate_msg(&reserve_size, &trunc_msg_len);
2033 prb_rec_init_wr(&r, reserve_size + trunc_msg_len);
2034 if (!prb_reserve(&e, prb, &r))
2039 text_len = printk_sprint(&r.text_buf[0], reserve_size, facility, &lflags, fmt, args);
2041 memcpy(&r.text_buf[text_len], trunc_msg, trunc_msg_len);
2042 r.info->text_len = text_len + trunc_msg_len;
2043 r.info->facility = facility;
2044 r.info->level = level & 7;
2045 r.info->flags = lflags & 0x1f;
2046 r.info->ts_nsec = ts_nsec;
2047 r.info->caller_id = caller_id;
2049 memcpy(&r.info->dev_info, dev_info, sizeof(r.info->dev_info));
2051 /* A message without a trailing newline can be continued. */
2052 if (!(lflags & LOG_NEWLINE))
2055 prb_final_commit(&e);
2057 return (text_len + trunc_msg_len);
2060 asmlinkage int vprintk_emit(int facility, int level,
2061 const struct dev_printk_info *dev_info,
2062 const char *fmt, va_list args)
2065 bool in_sched = false;
2066 unsigned long flags;
2068 /* Suppress unimportant messages after panic happens */
2069 if (unlikely(suppress_printk))
2072 if (level == LOGLEVEL_SCHED) {
2073 level = LOGLEVEL_DEFAULT;
2077 boot_delay_msec(level);
2080 printk_safe_enter_irqsave(flags);
2081 printed_len = vprintk_store(facility, level, dev_info, fmt, args);
2082 printk_safe_exit_irqrestore(flags);
2084 /* If called from the scheduler, we can not call up(). */
2087 * Disable preemption to avoid being preempted while holding
2088 * console_sem which would prevent anyone from printing to
2093 * Try to acquire and then immediately release the console
2094 * semaphore. The release will print out buffers and wake up
2095 * /dev/kmsg and syslog() users.
2097 if (console_trylock_spinning())
2105 EXPORT_SYMBOL(vprintk_emit);
2107 asmlinkage int vprintk(const char *fmt, va_list args)
2109 return vprintk_func(fmt, args);
2111 EXPORT_SYMBOL(vprintk);
2113 int vprintk_default(const char *fmt, va_list args)
2115 return vprintk_emit(0, LOGLEVEL_DEFAULT, NULL, fmt, args);
2117 EXPORT_SYMBOL_GPL(vprintk_default);
2120 * printk - print a kernel message
2121 * @fmt: format string
2123 * This is printk(). It can be called from any context. We want it to work.
2125 * We try to grab the console_lock. If we succeed, it's easy - we log the
2126 * output and call the console drivers. If we fail to get the semaphore, we
2127 * place the output into the log buffer and return. The current holder of
2128 * the console_sem will notice the new output in console_unlock(); and will
2129 * send it to the consoles before releasing the lock.
2131 * One effect of this deferred printing is that code which calls printk() and
2132 * then changes console_loglevel may break. This is because console_loglevel
2133 * is inspected when the actual printing occurs.
2138 * See the vsnprintf() documentation for format string extensions over C99.
2140 asmlinkage __visible int printk(const char *fmt, ...)
2145 va_start(args, fmt);
2146 r = vprintk_func(fmt, args);
2151 EXPORT_SYMBOL(printk);
2153 #else /* CONFIG_PRINTK */
2155 #define LOG_LINE_MAX 0
2156 #define PREFIX_MAX 0
2157 #define printk_time false
2159 #define prb_read_valid(rb, seq, r) false
2160 #define prb_first_valid_seq(rb) 0
2162 static u64 syslog_seq;
2163 static u64 console_seq;
2164 static u64 exclusive_console_stop_seq;
2165 static unsigned long console_dropped;
2167 static size_t record_print_text(const struct printk_record *r,
2168 bool syslog, bool time)
2172 static ssize_t info_print_ext_header(char *buf, size_t size,
2173 struct printk_info *info)
2177 static ssize_t msg_print_ext_body(char *buf, size_t size,
2178 char *text, size_t text_len,
2179 struct dev_printk_info *dev_info) { return 0; }
2180 static void console_lock_spinning_enable(void) { }
2181 static int console_lock_spinning_disable_and_check(void) { return 0; }
2182 static void call_console_drivers(const char *ext_text, size_t ext_len,
2183 const char *text, size_t len) {}
2184 static bool suppress_message_printing(int level) { return false; }
2186 #endif /* CONFIG_PRINTK */
2188 #ifdef CONFIG_EARLY_PRINTK
2189 struct console *early_console;
2191 asmlinkage __visible void early_printk(const char *fmt, ...)
2201 n = vscnprintf(buf, sizeof(buf), fmt, ap);
2204 early_console->write(early_console, buf, n);
2208 static int __add_preferred_console(char *name, int idx, char *options,
2209 char *brl_options, bool user_specified)
2211 struct console_cmdline *c;
2215 * See if this tty is not yet registered, and
2216 * if we have a slot free.
2218 for (i = 0, c = console_cmdline;
2219 i < MAX_CMDLINECONSOLES && c->name[0];
2221 if (strcmp(c->name, name) == 0 && c->index == idx) {
2223 preferred_console = i;
2225 c->user_specified = true;
2229 if (i == MAX_CMDLINECONSOLES)
2232 preferred_console = i;
2233 strlcpy(c->name, name, sizeof(c->name));
2234 c->options = options;
2235 c->user_specified = user_specified;
2236 braille_set_options(c, brl_options);
2242 static int __init console_msg_format_setup(char *str)
2244 if (!strcmp(str, "syslog"))
2245 console_msg_format = MSG_FORMAT_SYSLOG;
2246 if (!strcmp(str, "default"))
2247 console_msg_format = MSG_FORMAT_DEFAULT;
2250 __setup("console_msg_format=", console_msg_format_setup);
2253 * Set up a console. Called via do_early_param() in init/main.c
2254 * for each "console=" parameter in the boot command line.
2256 static int __init console_setup(char *str)
2258 char buf[sizeof(console_cmdline[0].name) + 4]; /* 4 for "ttyS" */
2259 char *s, *options, *brl_options = NULL;
2263 * console="" or console=null have been suggested as a way to
2264 * disable console output. Use ttynull that has been created
2265 * for exacly this purpose.
2267 if (str[0] == 0 || strcmp(str, "null") == 0) {
2268 __add_preferred_console("ttynull", 0, NULL, NULL, true);
2272 if (_braille_console_setup(&str, &brl_options))
2276 * Decode str into name, index, options.
2278 if (str[0] >= '0' && str[0] <= '9') {
2279 strcpy(buf, "ttyS");
2280 strncpy(buf + 4, str, sizeof(buf) - 5);
2282 strncpy(buf, str, sizeof(buf) - 1);
2284 buf[sizeof(buf) - 1] = 0;
2285 options = strchr(str, ',');
2289 if (!strcmp(str, "ttya"))
2290 strcpy(buf, "ttyS0");
2291 if (!strcmp(str, "ttyb"))
2292 strcpy(buf, "ttyS1");
2294 for (s = buf; *s; s++)
2295 if (isdigit(*s) || *s == ',')
2297 idx = simple_strtoul(s, NULL, 10);
2300 __add_preferred_console(buf, idx, options, brl_options, true);
2301 console_set_on_cmdline = 1;
2304 __setup("console=", console_setup);
2307 * add_preferred_console - add a device to the list of preferred consoles.
2308 * @name: device name
2309 * @idx: device index
2310 * @options: options for this console
2312 * The last preferred console added will be used for kernel messages
2313 * and stdin/out/err for init. Normally this is used by console_setup
2314 * above to handle user-supplied console arguments; however it can also
2315 * be used by arch-specific code either to override the user or more
2316 * commonly to provide a default console (ie from PROM variables) when
2317 * the user has not supplied one.
2319 int add_preferred_console(char *name, int idx, char *options)
2321 return __add_preferred_console(name, idx, options, NULL, false);
2324 bool console_suspend_enabled = true;
2325 EXPORT_SYMBOL(console_suspend_enabled);
2327 static int __init console_suspend_disable(char *str)
2329 console_suspend_enabled = false;
2332 __setup("no_console_suspend", console_suspend_disable);
2333 module_param_named(console_suspend, console_suspend_enabled,
2334 bool, S_IRUGO | S_IWUSR);
2335 MODULE_PARM_DESC(console_suspend, "suspend console during suspend"
2336 " and hibernate operations");
2339 * suspend_console - suspend the console subsystem
2341 * This disables printk() while we go into suspend states
2343 void suspend_console(void)
2345 if (!console_suspend_enabled)
2347 pr_info("Suspending console(s) (use no_console_suspend to debug)\n");
2349 console_suspended = 1;
2353 void resume_console(void)
2355 if (!console_suspend_enabled)
2358 console_suspended = 0;
2363 * console_cpu_notify - print deferred console messages after CPU hotplug
2366 * If printk() is called from a CPU that is not online yet, the messages
2367 * will be printed on the console only if there are CON_ANYTIME consoles.
2368 * This function is called when a new CPU comes online (or fails to come
2369 * up) or goes offline.
2371 static int console_cpu_notify(unsigned int cpu)
2373 if (!cpuhp_tasks_frozen) {
2374 /* If trylock fails, someone else is doing the printing */
2375 if (console_trylock())
2382 * console_lock - lock the console system for exclusive use.
2384 * Acquires a lock which guarantees that the caller has
2385 * exclusive access to the console system and the console_drivers list.
2387 * Can sleep, returns nothing.
2389 void console_lock(void)
2394 if (console_suspended)
2397 console_may_schedule = 1;
2399 EXPORT_SYMBOL(console_lock);
2402 * console_trylock - try to lock the console system for exclusive use.
2404 * Try to acquire a lock which guarantees that the caller has exclusive
2405 * access to the console system and the console_drivers list.
2407 * returns 1 on success, and 0 on failure to acquire the lock.
2409 int console_trylock(void)
2411 if (down_trylock_console_sem())
2413 if (console_suspended) {
2418 console_may_schedule = 0;
2421 EXPORT_SYMBOL(console_trylock);
2423 int is_console_locked(void)
2425 return console_locked;
2427 EXPORT_SYMBOL(is_console_locked);
2430 * Check if we have any console that is capable of printing while cpu is
2431 * booting or shutting down. Requires console_sem.
2433 static int have_callable_console(void)
2435 struct console *con;
2437 for_each_console(con)
2438 if ((con->flags & CON_ENABLED) &&
2439 (con->flags & CON_ANYTIME))
2446 * Can we actually use the console at this time on this cpu?
2448 * Console drivers may assume that per-cpu resources have been allocated. So
2449 * unless they're explicitly marked as being able to cope (CON_ANYTIME) don't
2450 * call them until this CPU is officially up.
2452 static inline int can_use_console(void)
2454 return cpu_online(raw_smp_processor_id()) || have_callable_console();
2458 * console_unlock - unlock the console system
2460 * Releases the console_lock which the caller holds on the console system
2461 * and the console driver list.
2463 * While the console_lock was held, console output may have been buffered
2464 * by printk(). If this is the case, console_unlock(); emits
2465 * the output prior to releasing the lock.
2467 * If there is output waiting, we wake /dev/kmsg and syslog() users.
2469 * console_unlock(); may be called from any context.
2471 void console_unlock(void)
2473 static char ext_text[CONSOLE_EXT_LOG_MAX];
2474 static char text[LOG_LINE_MAX + PREFIX_MAX];
2475 unsigned long flags;
2476 bool do_cond_resched, retry;
2477 struct printk_info info;
2478 struct printk_record r;
2480 if (console_suspended) {
2485 prb_rec_init_rd(&r, &info, text, sizeof(text));
2488 * Console drivers are called with interrupts disabled, so
2489 * @console_may_schedule should be cleared before; however, we may
2490 * end up dumping a lot of lines, for example, if called from
2491 * console registration path, and should invoke cond_resched()
2492 * between lines if allowable. Not doing so can cause a very long
2493 * scheduling stall on a slow console leading to RCU stall and
2494 * softlockup warnings which exacerbate the issue with more
2495 * messages practically incapacitating the system.
2497 * console_trylock() is not able to detect the preemptive
2498 * context reliably. Therefore the value must be stored before
2499 * and cleared after the "again" goto label.
2501 do_cond_resched = console_may_schedule;
2503 console_may_schedule = 0;
2506 * We released the console_sem lock, so we need to recheck if
2507 * cpu is online and (if not) is there at least one CON_ANYTIME
2510 if (!can_use_console()) {
2520 printk_safe_enter_irqsave(flags);
2521 raw_spin_lock(&logbuf_lock);
2523 if (!prb_read_valid(prb, console_seq, &r))
2526 if (console_seq != r.info->seq) {
2527 console_dropped += r.info->seq - console_seq;
2528 console_seq = r.info->seq;
2531 if (suppress_message_printing(r.info->level)) {
2533 * Skip record we have buffered and already printed
2534 * directly to the console when we received it, and
2535 * record that has level above the console loglevel.
2541 /* Output to all consoles once old messages replayed. */
2542 if (unlikely(exclusive_console &&
2543 console_seq >= exclusive_console_stop_seq)) {
2544 exclusive_console = NULL;
2548 * Handle extended console text first because later
2549 * record_print_text() will modify the record buffer in-place.
2551 if (nr_ext_console_drivers) {
2552 ext_len = info_print_ext_header(ext_text,
2555 ext_len += msg_print_ext_body(ext_text + ext_len,
2556 sizeof(ext_text) - ext_len,
2561 len = record_print_text(&r,
2562 console_msg_format & MSG_FORMAT_SYSLOG,
2565 raw_spin_unlock(&logbuf_lock);
2568 * While actively printing out messages, if another printk()
2569 * were to occur on another CPU, it may wait for this one to
2570 * finish. This task can not be preempted if there is a
2571 * waiter waiting to take over.
2573 console_lock_spinning_enable();
2575 stop_critical_timings(); /* don't trace print latency */
2576 call_console_drivers(ext_text, ext_len, text, len);
2577 start_critical_timings();
2579 if (console_lock_spinning_disable_and_check()) {
2580 printk_safe_exit_irqrestore(flags);
2584 printk_safe_exit_irqrestore(flags);
2586 if (do_cond_resched)
2592 raw_spin_unlock(&logbuf_lock);
2597 * Someone could have filled up the buffer again, so re-check if there's
2598 * something to flush. In case we cannot trylock the console_sem again,
2599 * there's a new owner and the console_unlock() from them will do the
2600 * flush, no worries.
2602 raw_spin_lock(&logbuf_lock);
2603 retry = prb_read_valid(prb, console_seq, NULL);
2604 raw_spin_unlock(&logbuf_lock);
2605 printk_safe_exit_irqrestore(flags);
2607 if (retry && console_trylock())
2610 EXPORT_SYMBOL(console_unlock);
2613 * console_conditional_schedule - yield the CPU if required
2615 * If the console code is currently allowed to sleep, and
2616 * if this CPU should yield the CPU to another task, do
2619 * Must be called within console_lock();.
2621 void __sched console_conditional_schedule(void)
2623 if (console_may_schedule)
2626 EXPORT_SYMBOL(console_conditional_schedule);
2628 void console_unblank(void)
2633 * console_unblank can no longer be called in interrupt context unless
2634 * oops_in_progress is set to 1..
2636 if (oops_in_progress) {
2637 if (down_trylock_console_sem() != 0)
2643 console_may_schedule = 0;
2645 if ((c->flags & CON_ENABLED) && c->unblank)
2651 * console_flush_on_panic - flush console content on panic
2652 * @mode: flush all messages in buffer or just the pending ones
2654 * Immediately output all pending messages no matter what.
2656 void console_flush_on_panic(enum con_flush_mode mode)
2659 * If someone else is holding the console lock, trylock will fail
2660 * and may_schedule may be set. Ignore and proceed to unlock so
2661 * that messages are flushed out. As this can be called from any
2662 * context and we don't want to get preempted while flushing,
2663 * ensure may_schedule is cleared.
2666 console_may_schedule = 0;
2668 if (mode == CONSOLE_REPLAY_ALL) {
2669 unsigned long flags;
2671 logbuf_lock_irqsave(flags);
2672 console_seq = prb_first_valid_seq(prb);
2673 logbuf_unlock_irqrestore(flags);
2679 * Return the console tty driver structure and its associated index
2681 struct tty_driver *console_device(int *index)
2684 struct tty_driver *driver = NULL;
2687 for_each_console(c) {
2690 driver = c->device(c, index);
2699 * Prevent further output on the passed console device so that (for example)
2700 * serial drivers can disable console output before suspending a port, and can
2701 * re-enable output afterwards.
2703 void console_stop(struct console *console)
2706 console->flags &= ~CON_ENABLED;
2709 EXPORT_SYMBOL(console_stop);
2711 void console_start(struct console *console)
2714 console->flags |= CON_ENABLED;
2717 EXPORT_SYMBOL(console_start);
2719 static int __read_mostly keep_bootcon;
2721 static int __init keep_bootcon_setup(char *str)
2724 pr_info("debug: skip boot console de-registration.\n");
2729 early_param("keep_bootcon", keep_bootcon_setup);
2732 * This is called by register_console() to try to match
2733 * the newly registered console with any of the ones selected
2734 * by either the command line or add_preferred_console() and
2737 * Care need to be taken with consoles that are statically
2738 * enabled such as netconsole
2740 static int try_enable_new_console(struct console *newcon, bool user_specified)
2742 struct console_cmdline *c;
2745 for (i = 0, c = console_cmdline;
2746 i < MAX_CMDLINECONSOLES && c->name[0];
2748 if (c->user_specified != user_specified)
2750 if (!newcon->match ||
2751 newcon->match(newcon, c->name, c->index, c->options) != 0) {
2752 /* default matching */
2753 BUILD_BUG_ON(sizeof(c->name) != sizeof(newcon->name));
2754 if (strcmp(c->name, newcon->name) != 0)
2756 if (newcon->index >= 0 &&
2757 newcon->index != c->index)
2759 if (newcon->index < 0)
2760 newcon->index = c->index;
2762 if (_braille_register_console(newcon, c))
2765 if (newcon->setup &&
2766 (err = newcon->setup(newcon, c->options)) != 0)
2769 newcon->flags |= CON_ENABLED;
2770 if (i == preferred_console) {
2771 newcon->flags |= CON_CONSDEV;
2772 has_preferred_console = true;
2778 * Some consoles, such as pstore and netconsole, can be enabled even
2779 * without matching. Accept the pre-enabled consoles only when match()
2780 * and setup() had a chance to be called.
2782 if (newcon->flags & CON_ENABLED && c->user_specified == user_specified)
2789 * The console driver calls this routine during kernel initialization
2790 * to register the console printing procedure with printk() and to
2791 * print any messages that were printed by the kernel before the
2792 * console driver was initialized.
2794 * This can happen pretty early during the boot process (because of
2795 * early_printk) - sometimes before setup_arch() completes - be careful
2796 * of what kernel features are used - they may not be initialised yet.
2798 * There are two types of consoles - bootconsoles (early_printk) and
2799 * "real" consoles (everything which is not a bootconsole) which are
2800 * handled differently.
2801 * - Any number of bootconsoles can be registered at any time.
2802 * - As soon as a "real" console is registered, all bootconsoles
2803 * will be unregistered automatically.
2804 * - Once a "real" console is registered, any attempt to register a
2805 * bootconsoles will be rejected
2807 void register_console(struct console *newcon)
2809 unsigned long flags;
2810 struct console *bcon = NULL;
2813 for_each_console(bcon) {
2814 if (WARN(bcon == newcon, "console '%s%d' already registered\n",
2815 bcon->name, bcon->index))
2820 * before we register a new CON_BOOT console, make sure we don't
2821 * already have a valid console
2823 if (newcon->flags & CON_BOOT) {
2824 for_each_console(bcon) {
2825 if (!(bcon->flags & CON_BOOT)) {
2826 pr_info("Too late to register bootconsole %s%d\n",
2827 newcon->name, newcon->index);
2833 if (console_drivers && console_drivers->flags & CON_BOOT)
2834 bcon = console_drivers;
2836 if (!has_preferred_console || bcon || !console_drivers)
2837 has_preferred_console = preferred_console >= 0;
2840 * See if we want to use this console driver. If we
2841 * didn't select a console we take the first one
2842 * that registers here.
2844 if (!has_preferred_console) {
2845 if (newcon->index < 0)
2847 if (newcon->setup == NULL ||
2848 newcon->setup(newcon, NULL) == 0) {
2849 newcon->flags |= CON_ENABLED;
2850 if (newcon->device) {
2851 newcon->flags |= CON_CONSDEV;
2852 has_preferred_console = true;
2857 /* See if this console matches one we selected on the command line */
2858 err = try_enable_new_console(newcon, true);
2860 /* If not, try to match against the platform default(s) */
2862 err = try_enable_new_console(newcon, false);
2864 /* printk() messages are not printed to the Braille console. */
2865 if (err || newcon->flags & CON_BRL)
2869 * If we have a bootconsole, and are switching to a real console,
2870 * don't print everything out again, since when the boot console, and
2871 * the real console are the same physical device, it's annoying to
2872 * see the beginning boot messages twice
2874 if (bcon && ((newcon->flags & (CON_CONSDEV | CON_BOOT)) == CON_CONSDEV))
2875 newcon->flags &= ~CON_PRINTBUFFER;
2878 * Put this console in the list - keep the
2879 * preferred driver at the head of the list.
2882 if ((newcon->flags & CON_CONSDEV) || console_drivers == NULL) {
2883 newcon->next = console_drivers;
2884 console_drivers = newcon;
2886 newcon->next->flags &= ~CON_CONSDEV;
2887 /* Ensure this flag is always set for the head of the list */
2888 newcon->flags |= CON_CONSDEV;
2890 newcon->next = console_drivers->next;
2891 console_drivers->next = newcon;
2894 if (newcon->flags & CON_EXTENDED)
2895 nr_ext_console_drivers++;
2897 if (newcon->flags & CON_PRINTBUFFER) {
2899 * console_unlock(); will print out the buffered messages
2902 logbuf_lock_irqsave(flags);
2904 * We're about to replay the log buffer. Only do this to the
2905 * just-registered console to avoid excessive message spam to
2906 * the already-registered consoles.
2908 * Set exclusive_console with disabled interrupts to reduce
2909 * race window with eventual console_flush_on_panic() that
2910 * ignores console_lock.
2912 exclusive_console = newcon;
2913 exclusive_console_stop_seq = console_seq;
2914 console_seq = syslog_seq;
2915 logbuf_unlock_irqrestore(flags);
2918 console_sysfs_notify();
2921 * By unregistering the bootconsoles after we enable the real console
2922 * we get the "console xxx enabled" message on all the consoles -
2923 * boot consoles, real consoles, etc - this is to ensure that end
2924 * users know there might be something in the kernel's log buffer that
2925 * went to the bootconsole (that they do not see on the real console)
2927 pr_info("%sconsole [%s%d] enabled\n",
2928 (newcon->flags & CON_BOOT) ? "boot" : "" ,
2929 newcon->name, newcon->index);
2931 ((newcon->flags & (CON_CONSDEV | CON_BOOT)) == CON_CONSDEV) &&
2933 /* We need to iterate through all boot consoles, to make
2934 * sure we print everything out, before we unregister them.
2936 for_each_console(bcon)
2937 if (bcon->flags & CON_BOOT)
2938 unregister_console(bcon);
2941 EXPORT_SYMBOL(register_console);
2943 int unregister_console(struct console *console)
2945 struct console *con;
2948 pr_info("%sconsole [%s%d] disabled\n",
2949 (console->flags & CON_BOOT) ? "boot" : "" ,
2950 console->name, console->index);
2952 res = _braille_unregister_console(console);
2960 if (console_drivers == console) {
2961 console_drivers=console->next;
2964 for_each_console(con) {
2965 if (con->next == console) {
2966 con->next = console->next;
2974 goto out_disable_unlock;
2976 if (console->flags & CON_EXTENDED)
2977 nr_ext_console_drivers--;
2980 * If this isn't the last console and it has CON_CONSDEV set, we
2981 * need to set it on the next preferred console.
2983 if (console_drivers != NULL && console->flags & CON_CONSDEV)
2984 console_drivers->flags |= CON_CONSDEV;
2986 console->flags &= ~CON_ENABLED;
2988 console_sysfs_notify();
2991 res = console->exit(console);
2996 console->flags &= ~CON_ENABLED;
3001 EXPORT_SYMBOL(unregister_console);
3004 * Initialize the console device. This is called *early*, so
3005 * we can't necessarily depend on lots of kernel help here.
3006 * Just do some early initializations, and do the complex setup
3009 void __init console_init(void)
3013 initcall_entry_t *ce;
3015 /* Setup the default TTY line discipline. */
3019 * set up the console device so that later boot sequences can
3020 * inform about problems etc..
3022 ce = __con_initcall_start;
3023 trace_initcall_level("console");
3024 while (ce < __con_initcall_end) {
3025 call = initcall_from_entry(ce);
3026 trace_initcall_start(call);
3028 trace_initcall_finish(call, ret);
3034 * Some boot consoles access data that is in the init section and which will
3035 * be discarded after the initcalls have been run. To make sure that no code
3036 * will access this data, unregister the boot consoles in a late initcall.
3038 * If for some reason, such as deferred probe or the driver being a loadable
3039 * module, the real console hasn't registered yet at this point, there will
3040 * be a brief interval in which no messages are logged to the console, which
3041 * makes it difficult to diagnose problems that occur during this time.
3043 * To mitigate this problem somewhat, only unregister consoles whose memory
3044 * intersects with the init section. Note that all other boot consoles will
3045 * get unregistred when the real preferred console is registered.
3047 static int __init printk_late_init(void)
3049 struct console *con;
3052 for_each_console(con) {
3053 if (!(con->flags & CON_BOOT))
3056 /* Check addresses that might be used for enabled consoles. */
3057 if (init_section_intersects(con, sizeof(*con)) ||
3058 init_section_contains(con->write, 0) ||
3059 init_section_contains(con->read, 0) ||
3060 init_section_contains(con->device, 0) ||
3061 init_section_contains(con->unblank, 0) ||
3062 init_section_contains(con->data, 0)) {
3064 * Please, consider moving the reported consoles out
3065 * of the init section.
3067 pr_warn("bootconsole [%s%d] uses init memory and must be disabled even before the real one is ready\n",
3068 con->name, con->index);
3069 unregister_console(con);
3072 ret = cpuhp_setup_state_nocalls(CPUHP_PRINTK_DEAD, "printk:dead", NULL,
3073 console_cpu_notify);
3075 ret = cpuhp_setup_state_nocalls(CPUHP_AP_ONLINE_DYN, "printk:online",
3076 console_cpu_notify, NULL);
3080 late_initcall(printk_late_init);
3082 #if defined CONFIG_PRINTK
3084 * Delayed printk version, for scheduler-internal messages:
3086 #define PRINTK_PENDING_WAKEUP 0x01
3087 #define PRINTK_PENDING_OUTPUT 0x02
3089 static DEFINE_PER_CPU(int, printk_pending);
3091 static void wake_up_klogd_work_func(struct irq_work *irq_work)
3093 int pending = __this_cpu_xchg(printk_pending, 0);
3095 if (pending & PRINTK_PENDING_OUTPUT) {
3096 /* If trylock fails, someone else is doing the printing */
3097 if (console_trylock())
3101 if (pending & PRINTK_PENDING_WAKEUP)
3102 wake_up_interruptible(&log_wait);
3105 static DEFINE_PER_CPU(struct irq_work, wake_up_klogd_work) =
3106 IRQ_WORK_INIT_LAZY(wake_up_klogd_work_func);
3108 void wake_up_klogd(void)
3110 if (!printk_percpu_data_ready())
3114 if (waitqueue_active(&log_wait)) {
3115 this_cpu_or(printk_pending, PRINTK_PENDING_WAKEUP);
3116 irq_work_queue(this_cpu_ptr(&wake_up_klogd_work));
3121 void defer_console_output(void)
3123 if (!printk_percpu_data_ready())
3127 __this_cpu_or(printk_pending, PRINTK_PENDING_OUTPUT);
3128 irq_work_queue(this_cpu_ptr(&wake_up_klogd_work));
3132 int vprintk_deferred(const char *fmt, va_list args)
3136 r = vprintk_emit(0, LOGLEVEL_SCHED, NULL, fmt, args);
3137 defer_console_output();
3142 int printk_deferred(const char *fmt, ...)
3147 va_start(args, fmt);
3148 r = vprintk_deferred(fmt, args);
3155 * printk rate limiting, lifted from the networking subsystem.
3157 * This enforces a rate limit: not more than 10 kernel messages
3158 * every 5s to make a denial-of-service attack impossible.
3160 DEFINE_RATELIMIT_STATE(printk_ratelimit_state, 5 * HZ, 10);
3162 int __printk_ratelimit(const char *func)
3164 return ___ratelimit(&printk_ratelimit_state, func);
3166 EXPORT_SYMBOL(__printk_ratelimit);
3169 * printk_timed_ratelimit - caller-controlled printk ratelimiting
3170 * @caller_jiffies: pointer to caller's state
3171 * @interval_msecs: minimum interval between prints
3173 * printk_timed_ratelimit() returns true if more than @interval_msecs
3174 * milliseconds have elapsed since the last time printk_timed_ratelimit()
3177 bool printk_timed_ratelimit(unsigned long *caller_jiffies,
3178 unsigned int interval_msecs)
3180 unsigned long elapsed = jiffies - *caller_jiffies;
3182 if (*caller_jiffies && elapsed <= msecs_to_jiffies(interval_msecs))
3185 *caller_jiffies = jiffies;
3188 EXPORT_SYMBOL(printk_timed_ratelimit);
3190 static DEFINE_SPINLOCK(dump_list_lock);
3191 static LIST_HEAD(dump_list);
3194 * kmsg_dump_register - register a kernel log dumper.
3195 * @dumper: pointer to the kmsg_dumper structure
3197 * Adds a kernel log dumper to the system. The dump callback in the
3198 * structure will be called when the kernel oopses or panics and must be
3199 * set. Returns zero on success and %-EINVAL or %-EBUSY otherwise.
3201 int kmsg_dump_register(struct kmsg_dumper *dumper)
3203 unsigned long flags;
3206 /* The dump callback needs to be set */
3210 spin_lock_irqsave(&dump_list_lock, flags);
3211 /* Don't allow registering multiple times */
3212 if (!dumper->registered) {
3213 dumper->registered = 1;
3214 list_add_tail_rcu(&dumper->list, &dump_list);
3217 spin_unlock_irqrestore(&dump_list_lock, flags);
3221 EXPORT_SYMBOL_GPL(kmsg_dump_register);
3224 * kmsg_dump_unregister - unregister a kmsg dumper.
3225 * @dumper: pointer to the kmsg_dumper structure
3227 * Removes a dump device from the system. Returns zero on success and
3228 * %-EINVAL otherwise.
3230 int kmsg_dump_unregister(struct kmsg_dumper *dumper)
3232 unsigned long flags;
3235 spin_lock_irqsave(&dump_list_lock, flags);
3236 if (dumper->registered) {
3237 dumper->registered = 0;
3238 list_del_rcu(&dumper->list);
3241 spin_unlock_irqrestore(&dump_list_lock, flags);
3246 EXPORT_SYMBOL_GPL(kmsg_dump_unregister);
3248 static bool always_kmsg_dump;
3249 module_param_named(always_kmsg_dump, always_kmsg_dump, bool, S_IRUGO | S_IWUSR);
3251 const char *kmsg_dump_reason_str(enum kmsg_dump_reason reason)
3254 case KMSG_DUMP_PANIC:
3256 case KMSG_DUMP_OOPS:
3258 case KMSG_DUMP_EMERG:
3260 case KMSG_DUMP_SHUTDOWN:
3266 EXPORT_SYMBOL_GPL(kmsg_dump_reason_str);
3269 * kmsg_dump - dump kernel log to kernel message dumpers.
3270 * @reason: the reason (oops, panic etc) for dumping
3272 * Call each of the registered dumper's dump() callback, which can
3273 * retrieve the kmsg records with kmsg_dump_get_line() or
3274 * kmsg_dump_get_buffer().
3276 void kmsg_dump(enum kmsg_dump_reason reason)
3278 struct kmsg_dumper *dumper;
3279 unsigned long flags;
3282 list_for_each_entry_rcu(dumper, &dump_list, list) {
3283 enum kmsg_dump_reason max_reason = dumper->max_reason;
3286 * If client has not provided a specific max_reason, default
3287 * to KMSG_DUMP_OOPS, unless always_kmsg_dump was set.
3289 if (max_reason == KMSG_DUMP_UNDEF) {
3290 max_reason = always_kmsg_dump ? KMSG_DUMP_MAX :
3293 if (reason > max_reason)
3296 /* initialize iterator with data about the stored records */
3297 dumper->active = true;
3299 logbuf_lock_irqsave(flags);
3300 dumper->cur_seq = clear_seq;
3301 dumper->next_seq = prb_next_seq(prb);
3302 logbuf_unlock_irqrestore(flags);
3304 /* invoke dumper which will iterate over records */
3305 dumper->dump(dumper, reason);
3307 /* reset iterator */
3308 dumper->active = false;
3314 * kmsg_dump_get_line_nolock - retrieve one kmsg log line (unlocked version)
3315 * @dumper: registered kmsg dumper
3316 * @syslog: include the "<4>" prefixes
3317 * @line: buffer to copy the line to
3318 * @size: maximum size of the buffer
3319 * @len: length of line placed into buffer
3321 * Start at the beginning of the kmsg buffer, with the oldest kmsg
3322 * record, and copy one record into the provided buffer.
3324 * Consecutive calls will return the next available record moving
3325 * towards the end of the buffer with the youngest messages.
3327 * A return value of FALSE indicates that there are no more records to
3330 * The function is similar to kmsg_dump_get_line(), but grabs no locks.
3332 bool kmsg_dump_get_line_nolock(struct kmsg_dumper *dumper, bool syslog,
3333 char *line, size_t size, size_t *len)
3335 struct printk_info info;
3336 unsigned int line_count;
3337 struct printk_record r;
3341 prb_rec_init_rd(&r, &info, line, size);
3343 if (!dumper->active)
3346 /* Read text or count text lines? */
3348 if (!prb_read_valid(prb, dumper->cur_seq, &r))
3350 l = record_print_text(&r, syslog, printk_time);
3352 if (!prb_read_valid_info(prb, dumper->cur_seq,
3353 &info, &line_count)) {
3356 l = get_record_print_text_size(&info, line_count, syslog,
3361 dumper->cur_seq = r.info->seq + 1;
3370 * kmsg_dump_get_line - retrieve one kmsg log line
3371 * @dumper: registered kmsg dumper
3372 * @syslog: include the "<4>" prefixes
3373 * @line: buffer to copy the line to
3374 * @size: maximum size of the buffer
3375 * @len: length of line placed into buffer
3377 * Start at the beginning of the kmsg buffer, with the oldest kmsg
3378 * record, and copy one record into the provided buffer.
3380 * Consecutive calls will return the next available record moving
3381 * towards the end of the buffer with the youngest messages.
3383 * A return value of FALSE indicates that there are no more records to
3386 bool kmsg_dump_get_line(struct kmsg_dumper *dumper, bool syslog,
3387 char *line, size_t size, size_t *len)
3389 unsigned long flags;
3392 logbuf_lock_irqsave(flags);
3393 ret = kmsg_dump_get_line_nolock(dumper, syslog, line, size, len);
3394 logbuf_unlock_irqrestore(flags);
3398 EXPORT_SYMBOL_GPL(kmsg_dump_get_line);
3401 * kmsg_dump_get_buffer - copy kmsg log lines
3402 * @dumper: registered kmsg dumper
3403 * @syslog: include the "<4>" prefixes
3404 * @buf: buffer to copy the line to
3405 * @size: maximum size of the buffer
3406 * @len: length of line placed into buffer
3408 * Start at the end of the kmsg buffer and fill the provided buffer
3409 * with as many of the *youngest* kmsg records that fit into it.
3410 * If the buffer is large enough, all available kmsg records will be
3411 * copied with a single call.
3413 * Consecutive calls will fill the buffer with the next block of
3414 * available older records, not including the earlier retrieved ones.
3416 * A return value of FALSE indicates that there are no more records to
3419 bool kmsg_dump_get_buffer(struct kmsg_dumper *dumper, bool syslog,
3420 char *buf, size_t size, size_t *len)
3422 struct printk_info info;
3423 unsigned int line_count;
3424 struct printk_record r;
3425 unsigned long flags;
3430 bool time = printk_time;
3432 prb_rec_init_rd(&r, &info, buf, size);
3434 if (!dumper->active || !buf || !size)
3437 logbuf_lock_irqsave(flags);
3438 if (prb_read_valid_info(prb, dumper->cur_seq, &info, NULL)) {
3439 if (info.seq != dumper->cur_seq) {
3440 /* messages are gone, move to first available one */
3441 dumper->cur_seq = info.seq;
3446 if (dumper->cur_seq >= dumper->next_seq) {
3447 logbuf_unlock_irqrestore(flags);
3451 /* calculate length of entire buffer */
3452 seq = dumper->cur_seq;
3453 while (prb_read_valid_info(prb, seq, &info, &line_count)) {
3454 if (r.info->seq >= dumper->next_seq)
3456 l += get_record_print_text_size(&info, line_count, syslog, time);
3457 seq = r.info->seq + 1;
3460 /* move first record forward until length fits into the buffer */
3461 seq = dumper->cur_seq;
3462 while (l >= size && prb_read_valid_info(prb, seq,
3463 &info, &line_count)) {
3464 if (r.info->seq >= dumper->next_seq)
3466 l -= get_record_print_text_size(&info, line_count, syslog, time);
3467 seq = r.info->seq + 1;
3470 /* last message in next interation */
3473 /* actually read text into the buffer now */
3475 while (prb_read_valid(prb, seq, &r)) {
3476 if (r.info->seq >= dumper->next_seq)
3479 l += record_print_text(&r, syslog, time);
3481 /* adjust record to store to remaining buffer space */
3482 prb_rec_init_rd(&r, &info, buf + l, size - l);
3484 seq = r.info->seq + 1;
3487 dumper->next_seq = next_seq;
3489 logbuf_unlock_irqrestore(flags);
3495 EXPORT_SYMBOL_GPL(kmsg_dump_get_buffer);
3498 * kmsg_dump_rewind_nolock - reset the iterator (unlocked version)
3499 * @dumper: registered kmsg dumper
3501 * Reset the dumper's iterator so that kmsg_dump_get_line() and
3502 * kmsg_dump_get_buffer() can be called again and used multiple
3503 * times within the same dumper.dump() callback.
3505 * The function is similar to kmsg_dump_rewind(), but grabs no locks.
3507 void kmsg_dump_rewind_nolock(struct kmsg_dumper *dumper)
3509 dumper->cur_seq = clear_seq;
3510 dumper->next_seq = prb_next_seq(prb);
3514 * kmsg_dump_rewind - reset the iterator
3515 * @dumper: registered kmsg dumper
3517 * Reset the dumper's iterator so that kmsg_dump_get_line() and
3518 * kmsg_dump_get_buffer() can be called again and used multiple
3519 * times within the same dumper.dump() callback.
3521 void kmsg_dump_rewind(struct kmsg_dumper *dumper)
3523 unsigned long flags;
3525 logbuf_lock_irqsave(flags);
3526 kmsg_dump_rewind_nolock(dumper);
3527 logbuf_unlock_irqrestore(flags);
3529 EXPORT_SYMBOL_GPL(kmsg_dump_rewind);