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_mutex protects console_list updates and console->flags updates.
83 * The flags are synchronized only for consoles that are registered, i.e.
84 * accessible via the console list.
86 static DEFINE_MUTEX(console_mutex);
89 * console_sem protects updates to console->seq and console_suspended,
90 * and also provides serialization for console printing.
92 static DEFINE_SEMAPHORE(console_sem);
93 HLIST_HEAD(console_list);
94 EXPORT_SYMBOL_GPL(console_list);
95 DEFINE_STATIC_SRCU(console_srcu);
98 * System may need to suppress printk message under certain
99 * circumstances, like after kernel panic happens.
101 int __read_mostly suppress_printk;
104 * During panic, heavy printk by other CPUs can delay the
105 * panic and risk deadlock on console resources.
107 static int __read_mostly suppress_panic_printk;
109 #ifdef CONFIG_LOCKDEP
110 static struct lockdep_map console_lock_dep_map = {
111 .name = "console_lock"
114 void lockdep_assert_console_list_lock_held(void)
116 lockdep_assert_held(&console_mutex);
118 EXPORT_SYMBOL(lockdep_assert_console_list_lock_held);
121 #ifdef CONFIG_DEBUG_LOCK_ALLOC
122 bool console_srcu_read_lock_is_held(void)
124 return srcu_read_lock_held(&console_srcu);
126 EXPORT_SYMBOL(console_srcu_read_lock_is_held);
129 enum devkmsg_log_bits {
130 __DEVKMSG_LOG_BIT_ON = 0,
131 __DEVKMSG_LOG_BIT_OFF,
132 __DEVKMSG_LOG_BIT_LOCK,
135 enum devkmsg_log_masks {
136 DEVKMSG_LOG_MASK_ON = BIT(__DEVKMSG_LOG_BIT_ON),
137 DEVKMSG_LOG_MASK_OFF = BIT(__DEVKMSG_LOG_BIT_OFF),
138 DEVKMSG_LOG_MASK_LOCK = BIT(__DEVKMSG_LOG_BIT_LOCK),
141 /* Keep both the 'on' and 'off' bits clear, i.e. ratelimit by default: */
142 #define DEVKMSG_LOG_MASK_DEFAULT 0
144 static unsigned int __read_mostly devkmsg_log = DEVKMSG_LOG_MASK_DEFAULT;
146 static int __control_devkmsg(char *str)
153 len = str_has_prefix(str, "on");
155 devkmsg_log = DEVKMSG_LOG_MASK_ON;
159 len = str_has_prefix(str, "off");
161 devkmsg_log = DEVKMSG_LOG_MASK_OFF;
165 len = str_has_prefix(str, "ratelimit");
167 devkmsg_log = DEVKMSG_LOG_MASK_DEFAULT;
174 static int __init control_devkmsg(char *str)
176 if (__control_devkmsg(str) < 0) {
177 pr_warn("printk.devkmsg: bad option string '%s'\n", str);
182 * Set sysctl string accordingly:
184 if (devkmsg_log == DEVKMSG_LOG_MASK_ON)
185 strcpy(devkmsg_log_str, "on");
186 else if (devkmsg_log == DEVKMSG_LOG_MASK_OFF)
187 strcpy(devkmsg_log_str, "off");
188 /* else "ratelimit" which is set by default. */
191 * Sysctl cannot change it anymore. The kernel command line setting of
192 * this parameter is to force the setting to be permanent throughout the
193 * runtime of the system. This is a precation measure against userspace
194 * trying to be a smarta** and attempting to change it up on us.
196 devkmsg_log |= DEVKMSG_LOG_MASK_LOCK;
200 __setup("printk.devkmsg=", control_devkmsg);
202 char devkmsg_log_str[DEVKMSG_STR_MAX_SIZE] = "ratelimit";
203 #if defined(CONFIG_PRINTK) && defined(CONFIG_SYSCTL)
204 int devkmsg_sysctl_set_loglvl(struct ctl_table *table, int write,
205 void *buffer, size_t *lenp, loff_t *ppos)
207 char old_str[DEVKMSG_STR_MAX_SIZE];
212 if (devkmsg_log & DEVKMSG_LOG_MASK_LOCK)
216 strncpy(old_str, devkmsg_log_str, DEVKMSG_STR_MAX_SIZE);
219 err = proc_dostring(table, write, buffer, lenp, ppos);
224 err = __control_devkmsg(devkmsg_log_str);
227 * Do not accept an unknown string OR a known string with
230 if (err < 0 || (err + 1 != *lenp)) {
232 /* ... and restore old setting. */
234 strncpy(devkmsg_log_str, old_str, DEVKMSG_STR_MAX_SIZE);
242 #endif /* CONFIG_PRINTK && CONFIG_SYSCTL */
245 * console_list_lock - Lock the console list
247 * For console list or console->flags updates
249 void console_list_lock(void)
252 * In unregister_console() and console_force_preferred_locked(),
253 * synchronize_srcu() is called with the console_list_lock held.
254 * Therefore it is not allowed that the console_list_lock is taken
255 * with the srcu_lock held.
257 * Detecting if this context is really in the read-side critical
258 * section is only possible if the appropriate debug options are
261 WARN_ON_ONCE(debug_lockdep_rcu_enabled() &&
262 srcu_read_lock_held(&console_srcu));
264 mutex_lock(&console_mutex);
266 EXPORT_SYMBOL(console_list_lock);
269 * console_list_unlock - Unlock the console list
271 * Counterpart to console_list_lock()
273 void console_list_unlock(void)
275 mutex_unlock(&console_mutex);
277 EXPORT_SYMBOL(console_list_unlock);
280 * console_srcu_read_lock - Register a new reader for the
281 * SRCU-protected console list
283 * Use for_each_console_srcu() to iterate the console list
285 * Context: Any context.
286 * Return: A cookie to pass to console_srcu_read_unlock().
288 int console_srcu_read_lock(void)
290 return srcu_read_lock_nmisafe(&console_srcu);
292 EXPORT_SYMBOL(console_srcu_read_lock);
295 * console_srcu_read_unlock - Unregister an old reader from
296 * the SRCU-protected console list
297 * @cookie: cookie returned from console_srcu_read_lock()
299 * Counterpart to console_srcu_read_lock()
301 void console_srcu_read_unlock(int cookie)
303 srcu_read_unlock_nmisafe(&console_srcu, cookie);
305 EXPORT_SYMBOL(console_srcu_read_unlock);
308 * Helper macros to handle lockdep when locking/unlocking console_sem. We use
309 * macros instead of functions so that _RET_IP_ contains useful information.
311 #define down_console_sem() do { \
313 mutex_acquire(&console_lock_dep_map, 0, 0, _RET_IP_);\
316 static int __down_trylock_console_sem(unsigned long ip)
322 * Here and in __up_console_sem() we need to be in safe mode,
323 * because spindump/WARN/etc from under console ->lock will
324 * deadlock in printk()->down_trylock_console_sem() otherwise.
326 printk_safe_enter_irqsave(flags);
327 lock_failed = down_trylock(&console_sem);
328 printk_safe_exit_irqrestore(flags);
332 mutex_acquire(&console_lock_dep_map, 0, 1, ip);
335 #define down_trylock_console_sem() __down_trylock_console_sem(_RET_IP_)
337 static void __up_console_sem(unsigned long ip)
341 mutex_release(&console_lock_dep_map, ip);
343 printk_safe_enter_irqsave(flags);
345 printk_safe_exit_irqrestore(flags);
347 #define up_console_sem() __up_console_sem(_RET_IP_)
349 static bool panic_in_progress(void)
351 return unlikely(atomic_read(&panic_cpu) != PANIC_CPU_INVALID);
355 * This is used for debugging the mess that is the VT code by
356 * keeping track if we have the console semaphore held. It's
357 * definitely not the perfect debug tool (we don't know if _WE_
358 * hold it and are racing, but it helps tracking those weird code
359 * paths in the console code where we end up in places I want
360 * locked without the console semaphore held).
362 static int console_locked, console_suspended;
365 * Array of consoles built from command line options (console=)
368 #define MAX_CMDLINECONSOLES 8
370 static struct console_cmdline console_cmdline[MAX_CMDLINECONSOLES];
372 static int preferred_console = -1;
373 int console_set_on_cmdline;
374 EXPORT_SYMBOL(console_set_on_cmdline);
376 /* Flag: console code may call schedule() */
377 static int console_may_schedule;
379 enum con_msg_format_flags {
380 MSG_FORMAT_DEFAULT = 0,
381 MSG_FORMAT_SYSLOG = (1 << 0),
384 static int console_msg_format = MSG_FORMAT_DEFAULT;
387 * The printk log buffer consists of a sequenced collection of records, each
388 * containing variable length message text. Every record also contains its
389 * own meta-data (@info).
391 * Every record meta-data carries the timestamp in microseconds, as well as
392 * the standard userspace syslog level and syslog facility. The usual kernel
393 * messages use LOG_KERN; userspace-injected messages always carry a matching
394 * syslog facility, by default LOG_USER. The origin of every message can be
395 * reliably determined that way.
397 * The human readable log message of a record is available in @text, the
398 * length of the message text in @text_len. The stored message is not
401 * Optionally, a record can carry a dictionary of properties (key/value
402 * pairs), to provide userspace with a machine-readable message context.
404 * Examples for well-defined, commonly used property names are:
405 * DEVICE=b12:8 device identifier
409 * +sound:card0 subsystem:devname
410 * SUBSYSTEM=pci driver-core subsystem name
412 * Valid characters in property names are [a-zA-Z0-9.-_]. Property names
413 * and values are terminated by a '\0' character.
415 * Example of record values:
416 * record.text_buf = "it's a line" (unterminated)
417 * record.info.seq = 56
418 * record.info.ts_nsec = 36863
419 * record.info.text_len = 11
420 * record.info.facility = 0 (LOG_KERN)
421 * record.info.flags = 0
422 * record.info.level = 3 (LOG_ERR)
423 * record.info.caller_id = 299 (task 299)
424 * record.info.dev_info.subsystem = "pci" (terminated)
425 * record.info.dev_info.device = "+pci:0000:00:01.0" (terminated)
427 * The 'struct printk_info' buffer must never be directly exported to
428 * userspace, it is a kernel-private implementation detail that might
429 * need to be changed in the future, when the requirements change.
431 * /dev/kmsg exports the structured data in the following line format:
432 * "<level>,<sequnum>,<timestamp>,<contflag>[,additional_values, ... ];<message text>\n"
434 * Users of the export format should ignore possible additional values
435 * separated by ',', and find the message after the ';' character.
437 * The optional key/value pairs are attached as continuation lines starting
438 * with a space character and terminated by a newline. All possible
439 * non-prinatable characters are escaped in the "\xff" notation.
442 /* syslog_lock protects syslog_* variables and write access to clear_seq. */
443 static DEFINE_MUTEX(syslog_lock);
446 DECLARE_WAIT_QUEUE_HEAD(log_wait);
447 /* All 3 protected by @syslog_lock. */
448 /* the next printk record to read by syslog(READ) or /proc/kmsg */
449 static u64 syslog_seq;
450 static size_t syslog_partial;
451 static bool syslog_time;
454 seqcount_latch_t latch;
459 * The next printk record to read after the last 'clear' command. There are
460 * two copies (updated with seqcount_latch) so that reads can locklessly
461 * access a valid value. Writers are synchronized by @syslog_lock.
463 static struct latched_seq clear_seq = {
464 .latch = SEQCNT_LATCH_ZERO(clear_seq.latch),
469 #define LOG_LEVEL(v) ((v) & 0x07)
470 #define LOG_FACILITY(v) ((v) >> 3 & 0xff)
473 #define LOG_ALIGN __alignof__(unsigned long)
474 #define __LOG_BUF_LEN (1 << CONFIG_LOG_BUF_SHIFT)
475 #define LOG_BUF_LEN_MAX (u32)(1 << 31)
476 static char __log_buf[__LOG_BUF_LEN] __aligned(LOG_ALIGN);
477 static char *log_buf = __log_buf;
478 static u32 log_buf_len = __LOG_BUF_LEN;
481 * Define the average message size. This only affects the number of
482 * descriptors that will be available. Underestimating is better than
483 * overestimating (too many available descriptors is better than not enough).
485 #define PRB_AVGBITS 5 /* 32 character average length */
487 #if CONFIG_LOG_BUF_SHIFT <= PRB_AVGBITS
488 #error CONFIG_LOG_BUF_SHIFT value too small.
490 _DEFINE_PRINTKRB(printk_rb_static, CONFIG_LOG_BUF_SHIFT - PRB_AVGBITS,
491 PRB_AVGBITS, &__log_buf[0]);
493 static struct printk_ringbuffer printk_rb_dynamic;
495 static struct printk_ringbuffer *prb = &printk_rb_static;
498 * We cannot access per-CPU data (e.g. per-CPU flush irq_work) before
499 * per_cpu_areas are initialised. This variable is set to true when
500 * it's safe to access per-CPU data.
502 static bool __printk_percpu_data_ready __ro_after_init;
504 bool printk_percpu_data_ready(void)
506 return __printk_percpu_data_ready;
509 /* Must be called under syslog_lock. */
510 static void latched_seq_write(struct latched_seq *ls, u64 val)
512 raw_write_seqcount_latch(&ls->latch);
514 raw_write_seqcount_latch(&ls->latch);
518 /* Can be called from any context. */
519 static u64 latched_seq_read_nolock(struct latched_seq *ls)
526 seq = raw_read_seqcount_latch(&ls->latch);
529 } while (read_seqcount_latch_retry(&ls->latch, seq));
534 /* Return log buffer address */
535 char *log_buf_addr_get(void)
540 /* Return log buffer size */
541 u32 log_buf_len_get(void)
547 * Define how much of the log buffer we could take at maximum. The value
548 * must be greater than two. Note that only half of the buffer is available
549 * when the index points to the middle.
551 #define MAX_LOG_TAKE_PART 4
552 static const char trunc_msg[] = "<truncated>";
554 static void truncate_msg(u16 *text_len, u16 *trunc_msg_len)
557 * The message should not take the whole buffer. Otherwise, it might
558 * get removed too soon.
560 u32 max_text_len = log_buf_len / MAX_LOG_TAKE_PART;
562 if (*text_len > max_text_len)
563 *text_len = max_text_len;
565 /* enable the warning message (if there is room) */
566 *trunc_msg_len = strlen(trunc_msg);
567 if (*text_len >= *trunc_msg_len)
568 *text_len -= *trunc_msg_len;
573 int dmesg_restrict = IS_ENABLED(CONFIG_SECURITY_DMESG_RESTRICT);
575 static int syslog_action_restricted(int type)
580 * Unless restricted, we allow "read all" and "get buffer size"
583 return type != SYSLOG_ACTION_READ_ALL &&
584 type != SYSLOG_ACTION_SIZE_BUFFER;
587 static int check_syslog_permissions(int type, int source)
590 * If this is from /proc/kmsg and we've already opened it, then we've
591 * already done the capabilities checks at open time.
593 if (source == SYSLOG_FROM_PROC && type != SYSLOG_ACTION_OPEN)
596 if (syslog_action_restricted(type)) {
597 if (capable(CAP_SYSLOG))
600 * For historical reasons, accept CAP_SYS_ADMIN too, with
603 if (capable(CAP_SYS_ADMIN)) {
604 pr_warn_once("%s (%d): Attempt to access syslog with "
605 "CAP_SYS_ADMIN but no CAP_SYSLOG "
607 current->comm, task_pid_nr(current));
613 return security_syslog(type);
616 static void append_char(char **pp, char *e, char c)
622 static ssize_t info_print_ext_header(char *buf, size_t size,
623 struct printk_info *info)
625 u64 ts_usec = info->ts_nsec;
627 #ifdef CONFIG_PRINTK_CALLER
628 u32 id = info->caller_id;
630 snprintf(caller, sizeof(caller), ",caller=%c%u",
631 id & 0x80000000 ? 'C' : 'T', id & ~0x80000000);
636 do_div(ts_usec, 1000);
638 return scnprintf(buf, size, "%u,%llu,%llu,%c%s;",
639 (info->facility << 3) | info->level, info->seq,
640 ts_usec, info->flags & LOG_CONT ? 'c' : '-', caller);
643 static ssize_t msg_add_ext_text(char *buf, size_t size,
644 const char *text, size_t text_len,
647 char *p = buf, *e = buf + size;
650 /* escape non-printable characters */
651 for (i = 0; i < text_len; i++) {
652 unsigned char c = text[i];
654 if (c < ' ' || c >= 127 || c == '\\')
655 p += scnprintf(p, e - p, "\\x%02x", c);
657 append_char(&p, e, c);
659 append_char(&p, e, endc);
664 static ssize_t msg_add_dict_text(char *buf, size_t size,
665 const char *key, const char *val)
667 size_t val_len = strlen(val);
673 len = msg_add_ext_text(buf, size, "", 0, ' '); /* dict prefix */
674 len += msg_add_ext_text(buf + len, size - len, key, strlen(key), '=');
675 len += msg_add_ext_text(buf + len, size - len, val, val_len, '\n');
680 static ssize_t msg_print_ext_body(char *buf, size_t size,
681 char *text, size_t text_len,
682 struct dev_printk_info *dev_info)
686 len = msg_add_ext_text(buf, size, text, text_len, '\n');
691 len += msg_add_dict_text(buf + len, size - len, "SUBSYSTEM",
692 dev_info->subsystem);
693 len += msg_add_dict_text(buf + len, size - len, "DEVICE",
699 static bool printk_get_next_message(struct printk_message *pmsg, u64 seq,
700 bool is_extended, bool may_supress);
702 /* /dev/kmsg - userspace message inject/listen interface */
703 struct devkmsg_user {
705 struct ratelimit_state rs;
707 struct printk_buffers pbufs;
710 static __printf(3, 4) __cold
711 int devkmsg_emit(int facility, int level, const char *fmt, ...)
717 r = vprintk_emit(facility, level, NULL, fmt, args);
723 static ssize_t devkmsg_write(struct kiocb *iocb, struct iov_iter *from)
726 int level = default_message_loglevel;
727 int facility = 1; /* LOG_USER */
728 struct file *file = iocb->ki_filp;
729 struct devkmsg_user *user = file->private_data;
730 size_t len = iov_iter_count(from);
733 if (!user || len > PRINTKRB_RECORD_MAX)
736 /* Ignore when user logging is disabled. */
737 if (devkmsg_log & DEVKMSG_LOG_MASK_OFF)
740 /* Ratelimit when not explicitly enabled. */
741 if (!(devkmsg_log & DEVKMSG_LOG_MASK_ON)) {
742 if (!___ratelimit(&user->rs, current->comm))
746 buf = kmalloc(len+1, GFP_KERNEL);
751 if (!copy_from_iter_full(buf, len, from)) {
757 * Extract and skip the syslog prefix <[0-9]*>. Coming from userspace
758 * the decimal value represents 32bit, the lower 3 bit are the log
759 * level, the rest are the log facility.
761 * If no prefix or no userspace facility is specified, we
762 * enforce LOG_USER, to be able to reliably distinguish
763 * kernel-generated messages from userspace-injected ones.
766 if (line[0] == '<') {
770 u = simple_strtoul(line + 1, &endp, 10);
771 if (endp && endp[0] == '>') {
772 level = LOG_LEVEL(u);
773 if (LOG_FACILITY(u) != 0)
774 facility = LOG_FACILITY(u);
780 devkmsg_emit(facility, level, "%s", line);
785 static ssize_t devkmsg_read(struct file *file, char __user *buf,
786 size_t count, loff_t *ppos)
788 struct devkmsg_user *user = file->private_data;
789 char *outbuf = &user->pbufs.outbuf[0];
790 struct printk_message pmsg = {
791 .pbufs = &user->pbufs,
798 ret = mutex_lock_interruptible(&user->lock);
802 if (!printk_get_next_message(&pmsg, atomic64_read(&user->seq), true, false)) {
803 if (file->f_flags & O_NONBLOCK) {
809 * Guarantee this task is visible on the waitqueue before
810 * checking the wake condition.
812 * The full memory barrier within set_current_state() of
813 * prepare_to_wait_event() pairs with the full memory barrier
814 * within wq_has_sleeper().
816 * This pairs with __wake_up_klogd:A.
818 ret = wait_event_interruptible(log_wait,
819 printk_get_next_message(&pmsg, atomic64_read(&user->seq), true,
820 false)); /* LMM(devkmsg_read:A) */
826 /* our last seen message is gone, return error and reset */
827 atomic64_set(&user->seq, pmsg.seq);
832 atomic64_set(&user->seq, pmsg.seq + 1);
834 if (pmsg.outbuf_len > count) {
839 if (copy_to_user(buf, outbuf, pmsg.outbuf_len)) {
843 ret = pmsg.outbuf_len;
845 mutex_unlock(&user->lock);
850 * Be careful when modifying this function!!!
852 * Only few operations are supported because the device works only with the
853 * entire variable length messages (records). Non-standard values are
854 * returned in the other cases and has been this way for quite some time.
855 * User space applications might depend on this behavior.
857 static loff_t devkmsg_llseek(struct file *file, loff_t offset, int whence)
859 struct devkmsg_user *user = file->private_data;
869 /* the first record */
870 atomic64_set(&user->seq, prb_first_valid_seq(prb));
874 * The first record after the last SYSLOG_ACTION_CLEAR,
875 * like issued by 'dmesg -c'. Reading /dev/kmsg itself
876 * changes no global state, and does not clear anything.
878 atomic64_set(&user->seq, latched_seq_read_nolock(&clear_seq));
881 /* after the last record */
882 atomic64_set(&user->seq, prb_next_seq(prb));
890 static __poll_t devkmsg_poll(struct file *file, poll_table *wait)
892 struct devkmsg_user *user = file->private_data;
893 struct printk_info info;
897 return EPOLLERR|EPOLLNVAL;
899 poll_wait(file, &log_wait, wait);
901 if (prb_read_valid_info(prb, atomic64_read(&user->seq), &info, NULL)) {
902 /* return error when data has vanished underneath us */
903 if (info.seq != atomic64_read(&user->seq))
904 ret = EPOLLIN|EPOLLRDNORM|EPOLLERR|EPOLLPRI;
906 ret = EPOLLIN|EPOLLRDNORM;
912 static int devkmsg_open(struct inode *inode, struct file *file)
914 struct devkmsg_user *user;
917 if (devkmsg_log & DEVKMSG_LOG_MASK_OFF)
920 /* write-only does not need any file context */
921 if ((file->f_flags & O_ACCMODE) != O_WRONLY) {
922 err = check_syslog_permissions(SYSLOG_ACTION_READ_ALL,
928 user = kvmalloc(sizeof(struct devkmsg_user), GFP_KERNEL);
932 ratelimit_default_init(&user->rs);
933 ratelimit_set_flags(&user->rs, RATELIMIT_MSG_ON_RELEASE);
935 mutex_init(&user->lock);
937 atomic64_set(&user->seq, prb_first_valid_seq(prb));
939 file->private_data = user;
943 static int devkmsg_release(struct inode *inode, struct file *file)
945 struct devkmsg_user *user = file->private_data;
950 ratelimit_state_exit(&user->rs);
952 mutex_destroy(&user->lock);
957 const struct file_operations kmsg_fops = {
958 .open = devkmsg_open,
959 .read = devkmsg_read,
960 .write_iter = devkmsg_write,
961 .llseek = devkmsg_llseek,
962 .poll = devkmsg_poll,
963 .release = devkmsg_release,
966 #ifdef CONFIG_CRASH_CORE
968 * This appends the listed symbols to /proc/vmcore
970 * /proc/vmcore is used by various utilities, like crash and makedumpfile to
971 * obtain access to symbols that are otherwise very difficult to locate. These
972 * symbols are specifically used so that utilities can access and extract the
973 * dmesg log from a vmcore file after a crash.
975 void log_buf_vmcoreinfo_setup(void)
977 struct dev_printk_info *dev_info = NULL;
979 VMCOREINFO_SYMBOL(prb);
980 VMCOREINFO_SYMBOL(printk_rb_static);
981 VMCOREINFO_SYMBOL(clear_seq);
984 * Export struct size and field offsets. User space tools can
985 * parse it and detect any changes to structure down the line.
988 VMCOREINFO_STRUCT_SIZE(printk_ringbuffer);
989 VMCOREINFO_OFFSET(printk_ringbuffer, desc_ring);
990 VMCOREINFO_OFFSET(printk_ringbuffer, text_data_ring);
991 VMCOREINFO_OFFSET(printk_ringbuffer, fail);
993 VMCOREINFO_STRUCT_SIZE(prb_desc_ring);
994 VMCOREINFO_OFFSET(prb_desc_ring, count_bits);
995 VMCOREINFO_OFFSET(prb_desc_ring, descs);
996 VMCOREINFO_OFFSET(prb_desc_ring, infos);
997 VMCOREINFO_OFFSET(prb_desc_ring, head_id);
998 VMCOREINFO_OFFSET(prb_desc_ring, tail_id);
1000 VMCOREINFO_STRUCT_SIZE(prb_desc);
1001 VMCOREINFO_OFFSET(prb_desc, state_var);
1002 VMCOREINFO_OFFSET(prb_desc, text_blk_lpos);
1004 VMCOREINFO_STRUCT_SIZE(prb_data_blk_lpos);
1005 VMCOREINFO_OFFSET(prb_data_blk_lpos, begin);
1006 VMCOREINFO_OFFSET(prb_data_blk_lpos, next);
1008 VMCOREINFO_STRUCT_SIZE(printk_info);
1009 VMCOREINFO_OFFSET(printk_info, seq);
1010 VMCOREINFO_OFFSET(printk_info, ts_nsec);
1011 VMCOREINFO_OFFSET(printk_info, text_len);
1012 VMCOREINFO_OFFSET(printk_info, caller_id);
1013 VMCOREINFO_OFFSET(printk_info, dev_info);
1015 VMCOREINFO_STRUCT_SIZE(dev_printk_info);
1016 VMCOREINFO_OFFSET(dev_printk_info, subsystem);
1017 VMCOREINFO_LENGTH(printk_info_subsystem, sizeof(dev_info->subsystem));
1018 VMCOREINFO_OFFSET(dev_printk_info, device);
1019 VMCOREINFO_LENGTH(printk_info_device, sizeof(dev_info->device));
1021 VMCOREINFO_STRUCT_SIZE(prb_data_ring);
1022 VMCOREINFO_OFFSET(prb_data_ring, size_bits);
1023 VMCOREINFO_OFFSET(prb_data_ring, data);
1024 VMCOREINFO_OFFSET(prb_data_ring, head_lpos);
1025 VMCOREINFO_OFFSET(prb_data_ring, tail_lpos);
1027 VMCOREINFO_SIZE(atomic_long_t);
1028 VMCOREINFO_TYPE_OFFSET(atomic_long_t, counter);
1030 VMCOREINFO_STRUCT_SIZE(latched_seq);
1031 VMCOREINFO_OFFSET(latched_seq, val);
1035 /* requested log_buf_len from kernel cmdline */
1036 static unsigned long __initdata new_log_buf_len;
1038 /* we practice scaling the ring buffer by powers of 2 */
1039 static void __init log_buf_len_update(u64 size)
1041 if (size > (u64)LOG_BUF_LEN_MAX) {
1042 size = (u64)LOG_BUF_LEN_MAX;
1043 pr_err("log_buf over 2G is not supported.\n");
1047 size = roundup_pow_of_two(size);
1048 if (size > log_buf_len)
1049 new_log_buf_len = (unsigned long)size;
1052 /* save requested log_buf_len since it's too early to process it */
1053 static int __init log_buf_len_setup(char *str)
1060 size = memparse(str, &str);
1062 log_buf_len_update(size);
1066 early_param("log_buf_len", log_buf_len_setup);
1069 #define __LOG_CPU_MAX_BUF_LEN (1 << CONFIG_LOG_CPU_MAX_BUF_SHIFT)
1071 static void __init log_buf_add_cpu(void)
1073 unsigned int cpu_extra;
1076 * archs should set up cpu_possible_bits properly with
1077 * set_cpu_possible() after setup_arch() but just in
1078 * case lets ensure this is valid.
1080 if (num_possible_cpus() == 1)
1083 cpu_extra = (num_possible_cpus() - 1) * __LOG_CPU_MAX_BUF_LEN;
1085 /* by default this will only continue through for large > 64 CPUs */
1086 if (cpu_extra <= __LOG_BUF_LEN / 2)
1089 pr_info("log_buf_len individual max cpu contribution: %d bytes\n",
1090 __LOG_CPU_MAX_BUF_LEN);
1091 pr_info("log_buf_len total cpu_extra contributions: %d bytes\n",
1093 pr_info("log_buf_len min size: %d bytes\n", __LOG_BUF_LEN);
1095 log_buf_len_update(cpu_extra + __LOG_BUF_LEN);
1097 #else /* !CONFIG_SMP */
1098 static inline void log_buf_add_cpu(void) {}
1099 #endif /* CONFIG_SMP */
1101 static void __init set_percpu_data_ready(void)
1103 __printk_percpu_data_ready = true;
1106 static unsigned int __init add_to_rb(struct printk_ringbuffer *rb,
1107 struct printk_record *r)
1109 struct prb_reserved_entry e;
1110 struct printk_record dest_r;
1112 prb_rec_init_wr(&dest_r, r->info->text_len);
1114 if (!prb_reserve(&e, rb, &dest_r))
1117 memcpy(&dest_r.text_buf[0], &r->text_buf[0], r->info->text_len);
1118 dest_r.info->text_len = r->info->text_len;
1119 dest_r.info->facility = r->info->facility;
1120 dest_r.info->level = r->info->level;
1121 dest_r.info->flags = r->info->flags;
1122 dest_r.info->ts_nsec = r->info->ts_nsec;
1123 dest_r.info->caller_id = r->info->caller_id;
1124 memcpy(&dest_r.info->dev_info, &r->info->dev_info, sizeof(dest_r.info->dev_info));
1126 prb_final_commit(&e);
1128 return prb_record_text_space(&e);
1131 static char setup_text_buf[PRINTKRB_RECORD_MAX] __initdata;
1133 void __init setup_log_buf(int early)
1135 struct printk_info *new_infos;
1136 unsigned int new_descs_count;
1137 struct prb_desc *new_descs;
1138 struct printk_info info;
1139 struct printk_record r;
1140 unsigned int text_size;
1141 size_t new_descs_size;
1142 size_t new_infos_size;
1143 unsigned long flags;
1149 * Some archs call setup_log_buf() multiple times - first is very
1150 * early, e.g. from setup_arch(), and second - when percpu_areas
1154 set_percpu_data_ready();
1156 if (log_buf != __log_buf)
1159 if (!early && !new_log_buf_len)
1162 if (!new_log_buf_len)
1165 new_descs_count = new_log_buf_len >> PRB_AVGBITS;
1166 if (new_descs_count == 0) {
1167 pr_err("new_log_buf_len: %lu too small\n", new_log_buf_len);
1171 new_log_buf = memblock_alloc(new_log_buf_len, LOG_ALIGN);
1172 if (unlikely(!new_log_buf)) {
1173 pr_err("log_buf_len: %lu text bytes not available\n",
1178 new_descs_size = new_descs_count * sizeof(struct prb_desc);
1179 new_descs = memblock_alloc(new_descs_size, LOG_ALIGN);
1180 if (unlikely(!new_descs)) {
1181 pr_err("log_buf_len: %zu desc bytes not available\n",
1183 goto err_free_log_buf;
1186 new_infos_size = new_descs_count * sizeof(struct printk_info);
1187 new_infos = memblock_alloc(new_infos_size, LOG_ALIGN);
1188 if (unlikely(!new_infos)) {
1189 pr_err("log_buf_len: %zu info bytes not available\n",
1191 goto err_free_descs;
1194 prb_rec_init_rd(&r, &info, &setup_text_buf[0], sizeof(setup_text_buf));
1196 prb_init(&printk_rb_dynamic,
1197 new_log_buf, ilog2(new_log_buf_len),
1198 new_descs, ilog2(new_descs_count),
1201 local_irq_save(flags);
1203 log_buf_len = new_log_buf_len;
1204 log_buf = new_log_buf;
1205 new_log_buf_len = 0;
1207 free = __LOG_BUF_LEN;
1208 prb_for_each_record(0, &printk_rb_static, seq, &r) {
1209 text_size = add_to_rb(&printk_rb_dynamic, &r);
1210 if (text_size > free)
1216 prb = &printk_rb_dynamic;
1218 local_irq_restore(flags);
1221 * Copy any remaining messages that might have appeared from
1222 * NMI context after copying but before switching to the
1225 prb_for_each_record(seq, &printk_rb_static, seq, &r) {
1226 text_size = add_to_rb(&printk_rb_dynamic, &r);
1227 if (text_size > free)
1233 if (seq != prb_next_seq(&printk_rb_static)) {
1234 pr_err("dropped %llu messages\n",
1235 prb_next_seq(&printk_rb_static) - seq);
1238 pr_info("log_buf_len: %u bytes\n", log_buf_len);
1239 pr_info("early log buf free: %u(%u%%)\n",
1240 free, (free * 100) / __LOG_BUF_LEN);
1244 memblock_free(new_descs, new_descs_size);
1246 memblock_free(new_log_buf, new_log_buf_len);
1249 static bool __read_mostly ignore_loglevel;
1251 static int __init ignore_loglevel_setup(char *str)
1253 ignore_loglevel = true;
1254 pr_info("debug: ignoring loglevel setting.\n");
1259 early_param("ignore_loglevel", ignore_loglevel_setup);
1260 module_param(ignore_loglevel, bool, S_IRUGO | S_IWUSR);
1261 MODULE_PARM_DESC(ignore_loglevel,
1262 "ignore loglevel setting (prints all kernel messages to the console)");
1264 static bool suppress_message_printing(int level)
1266 return (level >= console_loglevel && !ignore_loglevel);
1269 #ifdef CONFIG_BOOT_PRINTK_DELAY
1271 static int boot_delay; /* msecs delay after each printk during bootup */
1272 static unsigned long long loops_per_msec; /* based on boot_delay */
1274 static int __init boot_delay_setup(char *str)
1278 lpj = preset_lpj ? preset_lpj : 1000000; /* some guess */
1279 loops_per_msec = (unsigned long long)lpj / 1000 * HZ;
1281 get_option(&str, &boot_delay);
1282 if (boot_delay > 10 * 1000)
1285 pr_debug("boot_delay: %u, preset_lpj: %ld, lpj: %lu, "
1286 "HZ: %d, loops_per_msec: %llu\n",
1287 boot_delay, preset_lpj, lpj, HZ, loops_per_msec);
1290 early_param("boot_delay", boot_delay_setup);
1292 static void boot_delay_msec(int level)
1294 unsigned long long k;
1295 unsigned long timeout;
1297 if ((boot_delay == 0 || system_state >= SYSTEM_RUNNING)
1298 || suppress_message_printing(level)) {
1302 k = (unsigned long long)loops_per_msec * boot_delay;
1304 timeout = jiffies + msecs_to_jiffies(boot_delay);
1309 * use (volatile) jiffies to prevent
1310 * compiler reduction; loop termination via jiffies
1311 * is secondary and may or may not happen.
1313 if (time_after(jiffies, timeout))
1315 touch_nmi_watchdog();
1319 static inline void boot_delay_msec(int level)
1324 static bool printk_time = IS_ENABLED(CONFIG_PRINTK_TIME);
1325 module_param_named(time, printk_time, bool, S_IRUGO | S_IWUSR);
1327 static size_t print_syslog(unsigned int level, char *buf)
1329 return sprintf(buf, "<%u>", level);
1332 static size_t print_time(u64 ts, char *buf)
1334 unsigned long rem_nsec = do_div(ts, 1000000000);
1336 return sprintf(buf, "[%5lu.%06lu]",
1337 (unsigned long)ts, rem_nsec / 1000);
1340 #ifdef CONFIG_PRINTK_CALLER
1341 static size_t print_caller(u32 id, char *buf)
1345 snprintf(caller, sizeof(caller), "%c%u",
1346 id & 0x80000000 ? 'C' : 'T', id & ~0x80000000);
1347 return sprintf(buf, "[%6s]", caller);
1350 #define print_caller(id, buf) 0
1353 static size_t info_print_prefix(const struct printk_info *info, bool syslog,
1354 bool time, char *buf)
1359 len = print_syslog((info->facility << 3) | info->level, buf);
1362 len += print_time(info->ts_nsec, buf + len);
1364 len += print_caller(info->caller_id, buf + len);
1366 if (IS_ENABLED(CONFIG_PRINTK_CALLER) || time) {
1375 * Prepare the record for printing. The text is shifted within the given
1376 * buffer to avoid a need for another one. The following operations are
1379 * - Add prefix for each line.
1380 * - Drop truncated lines that no longer fit into the buffer.
1381 * - Add the trailing newline that has been removed in vprintk_store().
1382 * - Add a string terminator.
1384 * Since the produced string is always terminated, the maximum possible
1385 * return value is @r->text_buf_size - 1;
1387 * Return: The length of the updated/prepared text, including the added
1388 * prefixes and the newline. The terminator is not counted. The dropped
1389 * line(s) are not counted.
1391 static size_t record_print_text(struct printk_record *r, bool syslog,
1394 size_t text_len = r->info->text_len;
1395 size_t buf_size = r->text_buf_size;
1396 char *text = r->text_buf;
1397 char prefix[PRINTK_PREFIX_MAX];
1398 bool truncated = false;
1405 * If the message was truncated because the buffer was not large
1406 * enough, treat the available text as if it were the full text.
1408 if (text_len > buf_size)
1409 text_len = buf_size;
1411 prefix_len = info_print_prefix(r->info, syslog, time, prefix);
1414 * @text_len: bytes of unprocessed text
1415 * @line_len: bytes of current line _without_ newline
1416 * @text: pointer to beginning of current line
1417 * @len: number of bytes prepared in r->text_buf
1420 next = memchr(text, '\n', text_len);
1422 line_len = next - text;
1424 /* Drop truncated line(s). */
1427 line_len = text_len;
1431 * Truncate the text if there is not enough space to add the
1432 * prefix and a trailing newline and a terminator.
1434 if (len + prefix_len + text_len + 1 + 1 > buf_size) {
1435 /* Drop even the current line if no space. */
1436 if (len + prefix_len + line_len + 1 + 1 > buf_size)
1439 text_len = buf_size - len - prefix_len - 1 - 1;
1443 memmove(text + prefix_len, text, text_len);
1444 memcpy(text, prefix, prefix_len);
1447 * Increment the prepared length to include the text and
1448 * prefix that were just moved+copied. Also increment for the
1449 * newline at the end of this line. If this is the last line,
1450 * there is no newline, but it will be added immediately below.
1452 len += prefix_len + line_len + 1;
1453 if (text_len == line_len) {
1455 * This is the last line. Add the trailing newline
1456 * removed in vprintk_store().
1458 text[prefix_len + line_len] = '\n';
1463 * Advance beyond the added prefix and the related line with
1466 text += prefix_len + line_len + 1;
1469 * The remaining text has only decreased by the line with its
1472 * Note that @text_len can become zero. It happens when @text
1473 * ended with a newline (either due to truncation or the
1474 * original string ending with "\n\n"). The loop is correctly
1475 * repeated and (if not truncated) an empty line with a prefix
1478 text_len -= line_len + 1;
1482 * If a buffer was provided, it will be terminated. Space for the
1483 * string terminator is guaranteed to be available. The terminator is
1484 * not counted in the return value.
1487 r->text_buf[len] = 0;
1492 static size_t get_record_print_text_size(struct printk_info *info,
1493 unsigned int line_count,
1494 bool syslog, bool time)
1496 char prefix[PRINTK_PREFIX_MAX];
1499 prefix_len = info_print_prefix(info, syslog, time, prefix);
1502 * Each line will be preceded with a prefix. The intermediate
1503 * newlines are already within the text, but a final trailing
1504 * newline will be added.
1506 return ((prefix_len * line_count) + info->text_len + 1);
1510 * Beginning with @start_seq, find the first record where it and all following
1511 * records up to (but not including) @max_seq fit into @size.
1513 * @max_seq is simply an upper bound and does not need to exist. If the caller
1514 * does not require an upper bound, -1 can be used for @max_seq.
1516 static u64 find_first_fitting_seq(u64 start_seq, u64 max_seq, size_t size,
1517 bool syslog, bool time)
1519 struct printk_info info;
1520 unsigned int line_count;
1524 /* Determine the size of the records up to @max_seq. */
1525 prb_for_each_info(start_seq, prb, seq, &info, &line_count) {
1526 if (info.seq >= max_seq)
1528 len += get_record_print_text_size(&info, line_count, syslog, time);
1532 * Adjust the upper bound for the next loop to avoid subtracting
1533 * lengths that were never added.
1539 * Move first record forward until length fits into the buffer. Ignore
1540 * newest messages that were not counted in the above cycle. Messages
1541 * might appear and get lost in the meantime. This is a best effort
1542 * that prevents an infinite loop that could occur with a retry.
1544 prb_for_each_info(start_seq, prb, seq, &info, &line_count) {
1545 if (len <= size || info.seq >= max_seq)
1547 len -= get_record_print_text_size(&info, line_count, syslog, time);
1553 /* The caller is responsible for making sure @size is greater than 0. */
1554 static int syslog_print(char __user *buf, int size)
1556 struct printk_info info;
1557 struct printk_record r;
1562 text = kmalloc(PRINTK_MESSAGE_MAX, GFP_KERNEL);
1566 prb_rec_init_rd(&r, &info, text, PRINTK_MESSAGE_MAX);
1568 mutex_lock(&syslog_lock);
1571 * Wait for the @syslog_seq record to be available. @syslog_seq may
1572 * change while waiting.
1577 mutex_unlock(&syslog_lock);
1579 * Guarantee this task is visible on the waitqueue before
1580 * checking the wake condition.
1582 * The full memory barrier within set_current_state() of
1583 * prepare_to_wait_event() pairs with the full memory barrier
1584 * within wq_has_sleeper().
1586 * This pairs with __wake_up_klogd:A.
1588 len = wait_event_interruptible(log_wait,
1589 prb_read_valid(prb, seq, NULL)); /* LMM(syslog_print:A) */
1590 mutex_lock(&syslog_lock);
1594 } while (syslog_seq != seq);
1597 * Copy records that fit into the buffer. The above cycle makes sure
1598 * that the first record is always available.
1605 if (!prb_read_valid(prb, syslog_seq, &r))
1608 if (r.info->seq != syslog_seq) {
1609 /* message is gone, move to next valid one */
1610 syslog_seq = r.info->seq;
1615 * To keep reading/counting partial line consistent,
1616 * use printk_time value as of the beginning of a line.
1618 if (!syslog_partial)
1619 syslog_time = printk_time;
1621 skip = syslog_partial;
1622 n = record_print_text(&r, true, syslog_time);
1623 if (n - syslog_partial <= size) {
1624 /* message fits into buffer, move forward */
1625 syslog_seq = r.info->seq + 1;
1626 n -= syslog_partial;
1629 /* partial read(), remember position */
1631 syslog_partial += n;
1638 mutex_unlock(&syslog_lock);
1639 err = copy_to_user(buf, text + skip, n);
1640 mutex_lock(&syslog_lock);
1653 mutex_unlock(&syslog_lock);
1658 static int syslog_print_all(char __user *buf, int size, bool clear)
1660 struct printk_info info;
1661 struct printk_record r;
1667 text = kmalloc(PRINTK_MESSAGE_MAX, GFP_KERNEL);
1673 * Find first record that fits, including all following records,
1674 * into the user-provided buffer for this dump.
1676 seq = find_first_fitting_seq(latched_seq_read_nolock(&clear_seq), -1,
1679 prb_rec_init_rd(&r, &info, text, PRINTK_MESSAGE_MAX);
1682 prb_for_each_record(seq, prb, seq, &r) {
1685 textlen = record_print_text(&r, true, time);
1687 if (len + textlen > size) {
1692 if (copy_to_user(buf + len, text, textlen))
1702 mutex_lock(&syslog_lock);
1703 latched_seq_write(&clear_seq, seq);
1704 mutex_unlock(&syslog_lock);
1711 static void syslog_clear(void)
1713 mutex_lock(&syslog_lock);
1714 latched_seq_write(&clear_seq, prb_next_seq(prb));
1715 mutex_unlock(&syslog_lock);
1718 int do_syslog(int type, char __user *buf, int len, int source)
1720 struct printk_info info;
1722 static int saved_console_loglevel = LOGLEVEL_DEFAULT;
1725 error = check_syslog_permissions(type, source);
1730 case SYSLOG_ACTION_CLOSE: /* Close log */
1732 case SYSLOG_ACTION_OPEN: /* Open log */
1734 case SYSLOG_ACTION_READ: /* Read from log */
1735 if (!buf || len < 0)
1739 if (!access_ok(buf, len))
1741 error = syslog_print(buf, len);
1743 /* Read/clear last kernel messages */
1744 case SYSLOG_ACTION_READ_CLEAR:
1747 /* Read last kernel messages */
1748 case SYSLOG_ACTION_READ_ALL:
1749 if (!buf || len < 0)
1753 if (!access_ok(buf, len))
1755 error = syslog_print_all(buf, len, clear);
1757 /* Clear ring buffer */
1758 case SYSLOG_ACTION_CLEAR:
1761 /* Disable logging to console */
1762 case SYSLOG_ACTION_CONSOLE_OFF:
1763 if (saved_console_loglevel == LOGLEVEL_DEFAULT)
1764 saved_console_loglevel = console_loglevel;
1765 console_loglevel = minimum_console_loglevel;
1767 /* Enable logging to console */
1768 case SYSLOG_ACTION_CONSOLE_ON:
1769 if (saved_console_loglevel != LOGLEVEL_DEFAULT) {
1770 console_loglevel = saved_console_loglevel;
1771 saved_console_loglevel = LOGLEVEL_DEFAULT;
1774 /* Set level of messages printed to console */
1775 case SYSLOG_ACTION_CONSOLE_LEVEL:
1776 if (len < 1 || len > 8)
1778 if (len < minimum_console_loglevel)
1779 len = minimum_console_loglevel;
1780 console_loglevel = len;
1781 /* Implicitly re-enable logging to console */
1782 saved_console_loglevel = LOGLEVEL_DEFAULT;
1784 /* Number of chars in the log buffer */
1785 case SYSLOG_ACTION_SIZE_UNREAD:
1786 mutex_lock(&syslog_lock);
1787 if (!prb_read_valid_info(prb, syslog_seq, &info, NULL)) {
1788 /* No unread messages. */
1789 mutex_unlock(&syslog_lock);
1792 if (info.seq != syslog_seq) {
1793 /* messages are gone, move to first one */
1794 syslog_seq = info.seq;
1797 if (source == SYSLOG_FROM_PROC) {
1799 * Short-cut for poll(/"proc/kmsg") which simply checks
1800 * for pending data, not the size; return the count of
1801 * records, not the length.
1803 error = prb_next_seq(prb) - syslog_seq;
1805 bool time = syslog_partial ? syslog_time : printk_time;
1806 unsigned int line_count;
1809 prb_for_each_info(syslog_seq, prb, seq, &info,
1811 error += get_record_print_text_size(&info, line_count,
1815 error -= syslog_partial;
1817 mutex_unlock(&syslog_lock);
1819 /* Size of the log buffer */
1820 case SYSLOG_ACTION_SIZE_BUFFER:
1821 error = log_buf_len;
1831 SYSCALL_DEFINE3(syslog, int, type, char __user *, buf, int, len)
1833 return do_syslog(type, buf, len, SYSLOG_FROM_READER);
1837 * Special console_lock variants that help to reduce the risk of soft-lockups.
1838 * They allow to pass console_lock to another printk() call using a busy wait.
1841 #ifdef CONFIG_LOCKDEP
1842 static struct lockdep_map console_owner_dep_map = {
1843 .name = "console_owner"
1847 static DEFINE_RAW_SPINLOCK(console_owner_lock);
1848 static struct task_struct *console_owner;
1849 static bool console_waiter;
1852 * console_lock_spinning_enable - mark beginning of code where another
1853 * thread might safely busy wait
1855 * This basically converts console_lock into a spinlock. This marks
1856 * the section where the console_lock owner can not sleep, because
1857 * there may be a waiter spinning (like a spinlock). Also it must be
1858 * ready to hand over the lock at the end of the section.
1860 static void console_lock_spinning_enable(void)
1862 raw_spin_lock(&console_owner_lock);
1863 console_owner = current;
1864 raw_spin_unlock(&console_owner_lock);
1866 /* The waiter may spin on us after setting console_owner */
1867 spin_acquire(&console_owner_dep_map, 0, 0, _THIS_IP_);
1871 * console_lock_spinning_disable_and_check - mark end of code where another
1872 * thread was able to busy wait and check if there is a waiter
1873 * @cookie: cookie returned from console_srcu_read_lock()
1875 * This is called at the end of the section where spinning is allowed.
1876 * It has two functions. First, it is a signal that it is no longer
1877 * safe to start busy waiting for the lock. Second, it checks if
1878 * there is a busy waiter and passes the lock rights to her.
1880 * Important: Callers lose both the console_lock and the SRCU read lock if
1881 * there was a busy waiter. They must not touch items synchronized by
1882 * console_lock or SRCU read lock in this case.
1884 * Return: 1 if the lock rights were passed, 0 otherwise.
1886 static int console_lock_spinning_disable_and_check(int cookie)
1890 raw_spin_lock(&console_owner_lock);
1891 waiter = READ_ONCE(console_waiter);
1892 console_owner = NULL;
1893 raw_spin_unlock(&console_owner_lock);
1896 spin_release(&console_owner_dep_map, _THIS_IP_);
1900 /* The waiter is now free to continue */
1901 WRITE_ONCE(console_waiter, false);
1903 spin_release(&console_owner_dep_map, _THIS_IP_);
1906 * Preserve lockdep lock ordering. Release the SRCU read lock before
1907 * releasing the console_lock.
1909 console_srcu_read_unlock(cookie);
1912 * Hand off console_lock to waiter. The waiter will perform
1913 * the up(). After this, the waiter is the console_lock owner.
1915 mutex_release(&console_lock_dep_map, _THIS_IP_);
1920 * console_trylock_spinning - try to get console_lock by busy waiting
1922 * This allows to busy wait for the console_lock when the current
1923 * owner is running in specially marked sections. It means that
1924 * the current owner is running and cannot reschedule until it
1925 * is ready to lose the lock.
1927 * Return: 1 if we got the lock, 0 othrewise
1929 static int console_trylock_spinning(void)
1931 struct task_struct *owner = NULL;
1934 unsigned long flags;
1936 if (console_trylock())
1940 * It's unsafe to spin once a panic has begun. If we are the
1941 * panic CPU, we may have already halted the owner of the
1942 * console_sem. If we are not the panic CPU, then we should
1943 * avoid taking console_sem, so the panic CPU has a better
1944 * chance of cleanly acquiring it later.
1946 if (panic_in_progress())
1949 printk_safe_enter_irqsave(flags);
1951 raw_spin_lock(&console_owner_lock);
1952 owner = READ_ONCE(console_owner);
1953 waiter = READ_ONCE(console_waiter);
1954 if (!waiter && owner && owner != current) {
1955 WRITE_ONCE(console_waiter, true);
1958 raw_spin_unlock(&console_owner_lock);
1961 * If there is an active printk() writing to the
1962 * consoles, instead of having it write our data too,
1963 * see if we can offload that load from the active
1964 * printer, and do some printing ourselves.
1965 * Go into a spin only if there isn't already a waiter
1966 * spinning, and there is an active printer, and
1967 * that active printer isn't us (recursive printk?).
1970 printk_safe_exit_irqrestore(flags);
1974 /* We spin waiting for the owner to release us */
1975 spin_acquire(&console_owner_dep_map, 0, 0, _THIS_IP_);
1976 /* Owner will clear console_waiter on hand off */
1977 while (READ_ONCE(console_waiter))
1979 spin_release(&console_owner_dep_map, _THIS_IP_);
1981 printk_safe_exit_irqrestore(flags);
1983 * The owner passed the console lock to us.
1984 * Since we did not spin on console lock, annotate
1985 * this as a trylock. Otherwise lockdep will
1988 mutex_acquire(&console_lock_dep_map, 0, 1, _THIS_IP_);
1994 * Recursion is tracked separately on each CPU. If NMIs are supported, an
1995 * additional NMI context per CPU is also separately tracked. Until per-CPU
1996 * is available, a separate "early tracking" is performed.
1998 static DEFINE_PER_CPU(u8, printk_count);
1999 static u8 printk_count_early;
2000 #ifdef CONFIG_HAVE_NMI
2001 static DEFINE_PER_CPU(u8, printk_count_nmi);
2002 static u8 printk_count_nmi_early;
2006 * Recursion is limited to keep the output sane. printk() should not require
2007 * more than 1 level of recursion (allowing, for example, printk() to trigger
2008 * a WARN), but a higher value is used in case some printk-internal errors
2009 * exist, such as the ringbuffer validation checks failing.
2011 #define PRINTK_MAX_RECURSION 3
2014 * Return a pointer to the dedicated counter for the CPU+context of the
2017 static u8 *__printk_recursion_counter(void)
2019 #ifdef CONFIG_HAVE_NMI
2021 if (printk_percpu_data_ready())
2022 return this_cpu_ptr(&printk_count_nmi);
2023 return &printk_count_nmi_early;
2026 if (printk_percpu_data_ready())
2027 return this_cpu_ptr(&printk_count);
2028 return &printk_count_early;
2032 * Enter recursion tracking. Interrupts are disabled to simplify tracking.
2033 * The caller must check the boolean return value to see if the recursion is
2034 * allowed. On failure, interrupts are not disabled.
2036 * @recursion_ptr must be a variable of type (u8 *) and is the same variable
2037 * that is passed to printk_exit_irqrestore().
2039 #define printk_enter_irqsave(recursion_ptr, flags) \
2041 bool success = true; \
2043 typecheck(u8 *, recursion_ptr); \
2044 local_irq_save(flags); \
2045 (recursion_ptr) = __printk_recursion_counter(); \
2046 if (*(recursion_ptr) > PRINTK_MAX_RECURSION) { \
2047 local_irq_restore(flags); \
2050 (*(recursion_ptr))++; \
2055 /* Exit recursion tracking, restoring interrupts. */
2056 #define printk_exit_irqrestore(recursion_ptr, flags) \
2058 typecheck(u8 *, recursion_ptr); \
2059 (*(recursion_ptr))--; \
2060 local_irq_restore(flags); \
2063 int printk_delay_msec __read_mostly;
2065 static inline void printk_delay(int level)
2067 boot_delay_msec(level);
2069 if (unlikely(printk_delay_msec)) {
2070 int m = printk_delay_msec;
2074 touch_nmi_watchdog();
2079 static inline u32 printk_caller_id(void)
2081 return in_task() ? task_pid_nr(current) :
2082 0x80000000 + smp_processor_id();
2086 * printk_parse_prefix - Parse level and control flags.
2088 * @text: The terminated text message.
2089 * @level: A pointer to the current level value, will be updated.
2090 * @flags: A pointer to the current printk_info flags, will be updated.
2092 * @level may be NULL if the caller is not interested in the parsed value.
2093 * Otherwise the variable pointed to by @level must be set to
2094 * LOGLEVEL_DEFAULT in order to be updated with the parsed value.
2096 * @flags may be NULL if the caller is not interested in the parsed value.
2097 * Otherwise the variable pointed to by @flags will be OR'd with the parsed
2100 * Return: The length of the parsed level and control flags.
2102 u16 printk_parse_prefix(const char *text, int *level,
2103 enum printk_info_flags *flags)
2109 kern_level = printk_get_level(text);
2113 switch (kern_level) {
2115 if (level && *level == LOGLEVEL_DEFAULT)
2116 *level = kern_level - '0';
2118 case 'c': /* KERN_CONT */
2131 static u16 printk_sprint(char *text, u16 size, int facility,
2132 enum printk_info_flags *flags, const char *fmt,
2137 text_len = vscnprintf(text, size, fmt, args);
2139 /* Mark and strip a trailing newline. */
2140 if (text_len && text[text_len - 1] == '\n') {
2142 *flags |= LOG_NEWLINE;
2145 /* Strip log level and control flags. */
2146 if (facility == 0) {
2149 prefix_len = printk_parse_prefix(text, NULL, NULL);
2151 text_len -= prefix_len;
2152 memmove(text, text + prefix_len, text_len);
2156 trace_console(text, text_len);
2162 int vprintk_store(int facility, int level,
2163 const struct dev_printk_info *dev_info,
2164 const char *fmt, va_list args)
2166 struct prb_reserved_entry e;
2167 enum printk_info_flags flags = 0;
2168 struct printk_record r;
2169 unsigned long irqflags;
2170 u16 trunc_msg_len = 0;
2180 if (!printk_enter_irqsave(recursion_ptr, irqflags))
2184 * Since the duration of printk() can vary depending on the message
2185 * and state of the ringbuffer, grab the timestamp now so that it is
2186 * close to the call of printk(). This provides a more deterministic
2187 * timestamp with respect to the caller.
2189 ts_nsec = local_clock();
2191 caller_id = printk_caller_id();
2194 * The sprintf needs to come first since the syslog prefix might be
2195 * passed in as a parameter. An extra byte must be reserved so that
2196 * later the vscnprintf() into the reserved buffer has room for the
2197 * terminating '\0', which is not counted by vsnprintf().
2199 va_copy(args2, args);
2200 reserve_size = vsnprintf(&prefix_buf[0], sizeof(prefix_buf), fmt, args2) + 1;
2203 if (reserve_size > PRINTKRB_RECORD_MAX)
2204 reserve_size = PRINTKRB_RECORD_MAX;
2206 /* Extract log level or control flags. */
2208 printk_parse_prefix(&prefix_buf[0], &level, &flags);
2210 if (level == LOGLEVEL_DEFAULT)
2211 level = default_message_loglevel;
2214 flags |= LOG_NEWLINE;
2216 if (flags & LOG_CONT) {
2217 prb_rec_init_wr(&r, reserve_size);
2218 if (prb_reserve_in_last(&e, prb, &r, caller_id, PRINTKRB_RECORD_MAX)) {
2219 text_len = printk_sprint(&r.text_buf[r.info->text_len], reserve_size,
2220 facility, &flags, fmt, args);
2221 r.info->text_len += text_len;
2223 if (flags & LOG_NEWLINE) {
2224 r.info->flags |= LOG_NEWLINE;
2225 prb_final_commit(&e);
2236 * Explicitly initialize the record before every prb_reserve() call.
2237 * prb_reserve_in_last() and prb_reserve() purposely invalidate the
2238 * structure when they fail.
2240 prb_rec_init_wr(&r, reserve_size);
2241 if (!prb_reserve(&e, prb, &r)) {
2242 /* truncate the message if it is too long for empty buffer */
2243 truncate_msg(&reserve_size, &trunc_msg_len);
2245 prb_rec_init_wr(&r, reserve_size + trunc_msg_len);
2246 if (!prb_reserve(&e, prb, &r))
2251 text_len = printk_sprint(&r.text_buf[0], reserve_size, facility, &flags, fmt, args);
2253 memcpy(&r.text_buf[text_len], trunc_msg, trunc_msg_len);
2254 r.info->text_len = text_len + trunc_msg_len;
2255 r.info->facility = facility;
2256 r.info->level = level & 7;
2257 r.info->flags = flags & 0x1f;
2258 r.info->ts_nsec = ts_nsec;
2259 r.info->caller_id = caller_id;
2261 memcpy(&r.info->dev_info, dev_info, sizeof(r.info->dev_info));
2263 /* A message without a trailing newline can be continued. */
2264 if (!(flags & LOG_NEWLINE))
2267 prb_final_commit(&e);
2269 ret = text_len + trunc_msg_len;
2271 printk_exit_irqrestore(recursion_ptr, irqflags);
2275 asmlinkage int vprintk_emit(int facility, int level,
2276 const struct dev_printk_info *dev_info,
2277 const char *fmt, va_list args)
2280 bool in_sched = false;
2282 /* Suppress unimportant messages after panic happens */
2283 if (unlikely(suppress_printk))
2286 if (unlikely(suppress_panic_printk) &&
2287 atomic_read(&panic_cpu) != raw_smp_processor_id())
2290 if (level == LOGLEVEL_SCHED) {
2291 level = LOGLEVEL_DEFAULT;
2295 printk_delay(level);
2297 printed_len = vprintk_store(facility, level, dev_info, fmt, args);
2299 /* If called from the scheduler, we can not call up(). */
2302 * The caller may be holding system-critical or
2303 * timing-sensitive locks. Disable preemption during
2304 * printing of all remaining records to all consoles so that
2305 * this context can return as soon as possible. Hopefully
2306 * another printk() caller will take over the printing.
2310 * Try to acquire and then immediately release the console
2311 * semaphore. The release will print out buffers. With the
2312 * spinning variant, this context tries to take over the
2313 * printing from another printing context.
2315 if (console_trylock_spinning())
2323 EXPORT_SYMBOL(vprintk_emit);
2325 int vprintk_default(const char *fmt, va_list args)
2327 return vprintk_emit(0, LOGLEVEL_DEFAULT, NULL, fmt, args);
2329 EXPORT_SYMBOL_GPL(vprintk_default);
2331 asmlinkage __visible int _printk(const char *fmt, ...)
2336 va_start(args, fmt);
2337 r = vprintk(fmt, args);
2342 EXPORT_SYMBOL(_printk);
2344 static bool pr_flush(int timeout_ms, bool reset_on_progress);
2345 static bool __pr_flush(struct console *con, int timeout_ms, bool reset_on_progress);
2347 #else /* CONFIG_PRINTK */
2349 #define printk_time false
2351 #define prb_read_valid(rb, seq, r) false
2352 #define prb_first_valid_seq(rb) 0
2353 #define prb_next_seq(rb) 0
2355 static u64 syslog_seq;
2357 static size_t record_print_text(const struct printk_record *r,
2358 bool syslog, bool time)
2362 static ssize_t info_print_ext_header(char *buf, size_t size,
2363 struct printk_info *info)
2367 static ssize_t msg_print_ext_body(char *buf, size_t size,
2368 char *text, size_t text_len,
2369 struct dev_printk_info *dev_info) { return 0; }
2370 static void console_lock_spinning_enable(void) { }
2371 static int console_lock_spinning_disable_and_check(int cookie) { return 0; }
2372 static bool suppress_message_printing(int level) { return false; }
2373 static bool pr_flush(int timeout_ms, bool reset_on_progress) { return true; }
2374 static bool __pr_flush(struct console *con, int timeout_ms, bool reset_on_progress) { return true; }
2376 #endif /* CONFIG_PRINTK */
2378 #ifdef CONFIG_EARLY_PRINTK
2379 struct console *early_console;
2381 asmlinkage __visible void early_printk(const char *fmt, ...)
2391 n = vscnprintf(buf, sizeof(buf), fmt, ap);
2394 early_console->write(early_console, buf, n);
2398 static void set_user_specified(struct console_cmdline *c, bool user_specified)
2400 if (!user_specified)
2404 * @c console was defined by the user on the command line.
2405 * Do not clear when added twice also by SPCR or the device tree.
2407 c->user_specified = true;
2408 /* At least one console defined by the user on the command line. */
2409 console_set_on_cmdline = 1;
2412 static int __add_preferred_console(char *name, int idx, char *options,
2413 char *brl_options, bool user_specified)
2415 struct console_cmdline *c;
2419 * See if this tty is not yet registered, and
2420 * if we have a slot free.
2422 for (i = 0, c = console_cmdline;
2423 i < MAX_CMDLINECONSOLES && c->name[0];
2425 if (strcmp(c->name, name) == 0 && c->index == idx) {
2427 preferred_console = i;
2428 set_user_specified(c, user_specified);
2432 if (i == MAX_CMDLINECONSOLES)
2435 preferred_console = i;
2436 strscpy(c->name, name, sizeof(c->name));
2437 c->options = options;
2438 set_user_specified(c, user_specified);
2439 braille_set_options(c, brl_options);
2445 static int __init console_msg_format_setup(char *str)
2447 if (!strcmp(str, "syslog"))
2448 console_msg_format = MSG_FORMAT_SYSLOG;
2449 if (!strcmp(str, "default"))
2450 console_msg_format = MSG_FORMAT_DEFAULT;
2453 __setup("console_msg_format=", console_msg_format_setup);
2456 * Set up a console. Called via do_early_param() in init/main.c
2457 * for each "console=" parameter in the boot command line.
2459 static int __init console_setup(char *str)
2461 char buf[sizeof(console_cmdline[0].name) + 4]; /* 4 for "ttyS" */
2462 char *s, *options, *brl_options = NULL;
2466 * console="" or console=null have been suggested as a way to
2467 * disable console output. Use ttynull that has been created
2468 * for exactly this purpose.
2470 if (str[0] == 0 || strcmp(str, "null") == 0) {
2471 __add_preferred_console("ttynull", 0, NULL, NULL, true);
2475 if (_braille_console_setup(&str, &brl_options))
2479 * Decode str into name, index, options.
2481 if (str[0] >= '0' && str[0] <= '9') {
2482 strcpy(buf, "ttyS");
2483 strncpy(buf + 4, str, sizeof(buf) - 5);
2485 strncpy(buf, str, sizeof(buf) - 1);
2487 buf[sizeof(buf) - 1] = 0;
2488 options = strchr(str, ',');
2492 if (!strcmp(str, "ttya"))
2493 strcpy(buf, "ttyS0");
2494 if (!strcmp(str, "ttyb"))
2495 strcpy(buf, "ttyS1");
2497 for (s = buf; *s; s++)
2498 if (isdigit(*s) || *s == ',')
2500 idx = simple_strtoul(s, NULL, 10);
2503 __add_preferred_console(buf, idx, options, brl_options, true);
2506 __setup("console=", console_setup);
2509 * add_preferred_console - add a device to the list of preferred consoles.
2510 * @name: device name
2511 * @idx: device index
2512 * @options: options for this console
2514 * The last preferred console added will be used for kernel messages
2515 * and stdin/out/err for init. Normally this is used by console_setup
2516 * above to handle user-supplied console arguments; however it can also
2517 * be used by arch-specific code either to override the user or more
2518 * commonly to provide a default console (ie from PROM variables) when
2519 * the user has not supplied one.
2521 int add_preferred_console(char *name, int idx, char *options)
2523 return __add_preferred_console(name, idx, options, NULL, false);
2526 bool console_suspend_enabled = true;
2527 EXPORT_SYMBOL(console_suspend_enabled);
2529 static int __init console_suspend_disable(char *str)
2531 console_suspend_enabled = false;
2534 __setup("no_console_suspend", console_suspend_disable);
2535 module_param_named(console_suspend, console_suspend_enabled,
2536 bool, S_IRUGO | S_IWUSR);
2537 MODULE_PARM_DESC(console_suspend, "suspend console during suspend"
2538 " and hibernate operations");
2540 static bool printk_console_no_auto_verbose;
2542 void console_verbose(void)
2544 if (console_loglevel && !printk_console_no_auto_verbose)
2545 console_loglevel = CONSOLE_LOGLEVEL_MOTORMOUTH;
2547 EXPORT_SYMBOL_GPL(console_verbose);
2549 module_param_named(console_no_auto_verbose, printk_console_no_auto_verbose, bool, 0644);
2550 MODULE_PARM_DESC(console_no_auto_verbose, "Disable console loglevel raise to highest on oops/panic/etc");
2553 * suspend_console - suspend the console subsystem
2555 * This disables printk() while we go into suspend states
2557 void suspend_console(void)
2559 if (!console_suspend_enabled)
2561 pr_info("Suspending console(s) (use no_console_suspend to debug)\n");
2562 pr_flush(1000, true);
2564 console_suspended = 1;
2568 void resume_console(void)
2570 if (!console_suspend_enabled)
2573 console_suspended = 0;
2575 pr_flush(1000, true);
2579 * console_cpu_notify - print deferred console messages after CPU hotplug
2582 * If printk() is called from a CPU that is not online yet, the messages
2583 * will be printed on the console only if there are CON_ANYTIME consoles.
2584 * This function is called when a new CPU comes online (or fails to come
2585 * up) or goes offline.
2587 static int console_cpu_notify(unsigned int cpu)
2589 if (!cpuhp_tasks_frozen) {
2590 /* If trylock fails, someone else is doing the printing */
2591 if (console_trylock())
2598 * console_lock - block the console subsystem from printing
2600 * Acquires a lock which guarantees that no consoles will
2601 * be in or enter their write() callback.
2603 * Can sleep, returns nothing.
2605 void console_lock(void)
2610 if (console_suspended)
2613 console_may_schedule = 1;
2615 EXPORT_SYMBOL(console_lock);
2618 * console_trylock - try to block the console subsystem from printing
2620 * Try to acquire a lock which guarantees that no consoles will
2621 * be in or enter their write() callback.
2623 * returns 1 on success, and 0 on failure to acquire the lock.
2625 int console_trylock(void)
2627 if (down_trylock_console_sem())
2629 if (console_suspended) {
2634 console_may_schedule = 0;
2637 EXPORT_SYMBOL(console_trylock);
2639 int is_console_locked(void)
2641 return console_locked;
2643 EXPORT_SYMBOL(is_console_locked);
2646 * Return true when this CPU should unlock console_sem without pushing all
2647 * messages to the console. This reduces the chance that the console is
2648 * locked when the panic CPU tries to use it.
2650 static bool abandon_console_lock_in_panic(void)
2652 if (!panic_in_progress())
2656 * We can use raw_smp_processor_id() here because it is impossible for
2657 * the task to be migrated to the panic_cpu, or away from it. If
2658 * panic_cpu has already been set, and we're not currently executing on
2659 * that CPU, then we never will be.
2661 return atomic_read(&panic_cpu) != raw_smp_processor_id();
2665 * Check if the given console is currently capable and allowed to print
2668 * Requires the console_srcu_read_lock.
2670 static inline bool console_is_usable(struct console *con)
2672 short flags = console_srcu_read_flags(con);
2674 if (!(flags & CON_ENABLED))
2681 * Console drivers may assume that per-cpu resources have been
2682 * allocated. So unless they're explicitly marked as being able to
2683 * cope (CON_ANYTIME) don't call them until this CPU is officially up.
2685 if (!cpu_online(raw_smp_processor_id()) && !(flags & CON_ANYTIME))
2691 static void __console_unlock(void)
2698 * Prepend the message in @pmsg->pbufs->outbuf with a "dropped message". This
2699 * is achieved by shifting the existing message over and inserting the dropped
2702 * @pmsg is the printk message to prepend.
2704 * @dropped is the dropped count to report in the dropped message.
2706 * If the message text in @pmsg->pbufs->outbuf does not have enough space for
2707 * the dropped message, the message text will be sufficiently truncated.
2709 * If @pmsg->pbufs->outbuf is modified, @pmsg->outbuf_len is updated.
2711 #ifdef CONFIG_PRINTK
2712 static void console_prepend_dropped(struct printk_message *pmsg, unsigned long dropped)
2714 struct printk_buffers *pbufs = pmsg->pbufs;
2715 const size_t scratchbuf_sz = sizeof(pbufs->scratchbuf);
2716 const size_t outbuf_sz = sizeof(pbufs->outbuf);
2717 char *scratchbuf = &pbufs->scratchbuf[0];
2718 char *outbuf = &pbufs->outbuf[0];
2721 len = scnprintf(scratchbuf, scratchbuf_sz,
2722 "** %lu printk messages dropped **\n", dropped);
2725 * Make sure outbuf is sufficiently large before prepending.
2726 * Keep at least the prefix when the message must be truncated.
2727 * It is a rather theoretical problem when someone tries to
2728 * use a minimalist buffer.
2730 if (WARN_ON_ONCE(len + PRINTK_PREFIX_MAX >= outbuf_sz))
2733 if (pmsg->outbuf_len + len >= outbuf_sz) {
2734 /* Truncate the message, but keep it terminated. */
2735 pmsg->outbuf_len = outbuf_sz - (len + 1);
2736 outbuf[pmsg->outbuf_len] = 0;
2739 memmove(outbuf + len, outbuf, pmsg->outbuf_len + 1);
2740 memcpy(outbuf, scratchbuf, len);
2741 pmsg->outbuf_len += len;
2744 #define console_prepend_dropped(pmsg, dropped)
2745 #endif /* CONFIG_PRINTK */
2748 * Read and format the specified record (or a later record if the specified
2749 * record is not available).
2751 * @pmsg will contain the formatted result. @pmsg->pbufs must point to a
2752 * struct printk_buffers.
2754 * @seq is the record to read and format. If it is not available, the next
2755 * valid record is read.
2757 * @is_extended specifies if the message should be formatted for extended
2760 * @may_supress specifies if records may be skipped based on loglevel.
2762 * Returns false if no record is available. Otherwise true and all fields
2763 * of @pmsg are valid. (See the documentation of struct printk_message
2764 * for information about the @pmsg fields.)
2766 static bool printk_get_next_message(struct printk_message *pmsg, u64 seq,
2767 bool is_extended, bool may_suppress)
2769 static int panic_console_dropped;
2771 struct printk_buffers *pbufs = pmsg->pbufs;
2772 const size_t scratchbuf_sz = sizeof(pbufs->scratchbuf);
2773 const size_t outbuf_sz = sizeof(pbufs->outbuf);
2774 char *scratchbuf = &pbufs->scratchbuf[0];
2775 char *outbuf = &pbufs->outbuf[0];
2776 struct printk_info info;
2777 struct printk_record r;
2781 * Formatting extended messages requires a separate buffer, so use the
2782 * scratch buffer to read in the ringbuffer text.
2784 * Formatting normal messages is done in-place, so read the ringbuffer
2785 * text directly into the output buffer.
2788 prb_rec_init_rd(&r, &info, scratchbuf, scratchbuf_sz);
2790 prb_rec_init_rd(&r, &info, outbuf, outbuf_sz);
2792 if (!prb_read_valid(prb, seq, &r))
2795 pmsg->seq = r.info->seq;
2796 pmsg->dropped = r.info->seq - seq;
2799 * Check for dropped messages in panic here so that printk
2800 * suppression can occur as early as possible if necessary.
2802 if (pmsg->dropped &&
2803 panic_in_progress() &&
2804 panic_console_dropped++ > 10) {
2805 suppress_panic_printk = 1;
2806 pr_warn_once("Too many dropped messages. Suppress messages on non-panic CPUs to prevent livelock.\n");
2809 /* Skip record that has level above the console loglevel. */
2810 if (may_suppress && suppress_message_printing(r.info->level))
2814 len = info_print_ext_header(outbuf, outbuf_sz, r.info);
2815 len += msg_print_ext_body(outbuf + len, outbuf_sz - len,
2816 &r.text_buf[0], r.info->text_len, &r.info->dev_info);
2818 len = record_print_text(&r, console_msg_format & MSG_FORMAT_SYSLOG, printk_time);
2821 pmsg->outbuf_len = len;
2826 * Print one record for the given console. The record printed is whatever
2827 * record is the next available record for the given console.
2829 * @handover will be set to true if a printk waiter has taken over the
2830 * console_lock, in which case the caller is no longer holding both the
2831 * console_lock and the SRCU read lock. Otherwise it is set to false.
2833 * @cookie is the cookie from the SRCU read lock.
2835 * Returns false if the given console has no next record to print, otherwise
2838 * Requires the console_lock and the SRCU read lock.
2840 static bool console_emit_next_record(struct console *con, bool *handover, int cookie)
2842 static struct printk_buffers pbufs;
2844 bool is_extended = console_srcu_read_flags(con) & CON_EXTENDED;
2845 char *outbuf = &pbufs.outbuf[0];
2846 struct printk_message pmsg = {
2849 unsigned long flags;
2853 if (!printk_get_next_message(&pmsg, con->seq, is_extended, true))
2856 con->dropped += pmsg.dropped;
2858 /* Skip messages of formatted length 0. */
2859 if (pmsg.outbuf_len == 0) {
2860 con->seq = pmsg.seq + 1;
2864 if (con->dropped && !is_extended) {
2865 console_prepend_dropped(&pmsg, con->dropped);
2870 * While actively printing out messages, if another printk()
2871 * were to occur on another CPU, it may wait for this one to
2872 * finish. This task can not be preempted if there is a
2873 * waiter waiting to take over.
2875 * Interrupts are disabled because the hand over to a waiter
2876 * must not be interrupted until the hand over is completed
2877 * (@console_waiter is cleared).
2879 printk_safe_enter_irqsave(flags);
2880 console_lock_spinning_enable();
2882 /* Do not trace print latency. */
2883 stop_critical_timings();
2885 /* Write everything out to the hardware. */
2886 con->write(con, outbuf, pmsg.outbuf_len);
2888 start_critical_timings();
2890 con->seq = pmsg.seq + 1;
2892 *handover = console_lock_spinning_disable_and_check(cookie);
2893 printk_safe_exit_irqrestore(flags);
2899 * Print out all remaining records to all consoles.
2901 * @do_cond_resched is set by the caller. It can be true only in schedulable
2904 * @next_seq is set to the sequence number after the last available record.
2905 * The value is valid only when this function returns true. It means that all
2906 * usable consoles are completely flushed.
2908 * @handover will be set to true if a printk waiter has taken over the
2909 * console_lock, in which case the caller is no longer holding the
2910 * console_lock. Otherwise it is set to false.
2912 * Returns true when there was at least one usable console and all messages
2913 * were flushed to all usable consoles. A returned false informs the caller
2914 * that everything was not flushed (either there were no usable consoles or
2915 * another context has taken over printing or it is a panic situation and this
2916 * is not the panic CPU). Regardless the reason, the caller should assume it
2917 * is not useful to immediately try again.
2919 * Requires the console_lock.
2921 static bool console_flush_all(bool do_cond_resched, u64 *next_seq, bool *handover)
2923 bool any_usable = false;
2924 struct console *con;
2932 any_progress = false;
2934 cookie = console_srcu_read_lock();
2935 for_each_console_srcu(con) {
2938 if (!console_is_usable(con))
2942 progress = console_emit_next_record(con, handover, cookie);
2945 * If a handover has occurred, the SRCU read lock
2946 * is already released.
2951 /* Track the next of the highest seq flushed. */
2952 if (con->seq > *next_seq)
2953 *next_seq = con->seq;
2957 any_progress = true;
2959 /* Allow panic_cpu to take over the consoles safely. */
2960 if (abandon_console_lock_in_panic())
2963 if (do_cond_resched)
2966 console_srcu_read_unlock(cookie);
2967 } while (any_progress);
2972 console_srcu_read_unlock(cookie);
2977 * console_unlock - unblock the console subsystem from printing
2979 * Releases the console_lock which the caller holds to block printing of
2980 * the console subsystem.
2982 * While the console_lock was held, console output may have been buffered
2983 * by printk(). If this is the case, console_unlock(); emits
2984 * the output prior to releasing the lock.
2986 * console_unlock(); may be called from any context.
2988 void console_unlock(void)
2990 bool do_cond_resched;
2995 if (console_suspended) {
3001 * Console drivers are called with interrupts disabled, so
3002 * @console_may_schedule should be cleared before; however, we may
3003 * end up dumping a lot of lines, for example, if called from
3004 * console registration path, and should invoke cond_resched()
3005 * between lines if allowable. Not doing so can cause a very long
3006 * scheduling stall on a slow console leading to RCU stall and
3007 * softlockup warnings which exacerbate the issue with more
3008 * messages practically incapacitating the system. Therefore, create
3009 * a local to use for the printing loop.
3011 do_cond_resched = console_may_schedule;
3014 console_may_schedule = 0;
3016 flushed = console_flush_all(do_cond_resched, &next_seq, &handover);
3021 * Abort if there was a failure to flush all messages to all
3022 * usable consoles. Either it is not possible to flush (in
3023 * which case it would be an infinite loop of retrying) or
3024 * another context has taken over printing.
3030 * Some context may have added new records after
3031 * console_flush_all() but before unlocking the console.
3032 * Re-check if there is a new record to flush. If the trylock
3033 * fails, another context is already handling the printing.
3035 } while (prb_read_valid(prb, next_seq, NULL) && console_trylock());
3037 EXPORT_SYMBOL(console_unlock);
3040 * console_conditional_schedule - yield the CPU if required
3042 * If the console code is currently allowed to sleep, and
3043 * if this CPU should yield the CPU to another task, do
3046 * Must be called within console_lock();.
3048 void __sched console_conditional_schedule(void)
3050 if (console_may_schedule)
3053 EXPORT_SYMBOL(console_conditional_schedule);
3055 void console_unblank(void)
3061 * Stop console printing because the unblank() callback may
3062 * assume the console is not within its write() callback.
3064 * If @oops_in_progress is set, this may be an atomic context.
3065 * In that case, attempt a trylock as best-effort.
3067 if (oops_in_progress) {
3068 if (down_trylock_console_sem() != 0)
3074 console_may_schedule = 0;
3076 cookie = console_srcu_read_lock();
3077 for_each_console_srcu(c) {
3078 if ((console_srcu_read_flags(c) & CON_ENABLED) && c->unblank)
3081 console_srcu_read_unlock(cookie);
3085 if (!oops_in_progress)
3086 pr_flush(1000, true);
3090 * console_flush_on_panic - flush console content on panic
3091 * @mode: flush all messages in buffer or just the pending ones
3093 * Immediately output all pending messages no matter what.
3095 void console_flush_on_panic(enum con_flush_mode mode)
3098 * If someone else is holding the console lock, trylock will fail
3099 * and may_schedule may be set. Ignore and proceed to unlock so
3100 * that messages are flushed out. As this can be called from any
3101 * context and we don't want to get preempted while flushing,
3102 * ensure may_schedule is cleared.
3105 console_may_schedule = 0;
3107 if (mode == CONSOLE_REPLAY_ALL) {
3112 seq = prb_first_valid_seq(prb);
3114 cookie = console_srcu_read_lock();
3115 for_each_console_srcu(c) {
3117 * If the above console_trylock() failed, this is an
3118 * unsynchronized assignment. But in that case, the
3119 * kernel is in "hope and pray" mode anyway.
3123 console_srcu_read_unlock(cookie);
3129 * Return the console tty driver structure and its associated index
3131 struct tty_driver *console_device(int *index)
3134 struct tty_driver *driver = NULL;
3138 * Take console_lock to serialize device() callback with
3139 * other console operations. For example, fg_console is
3140 * modified under console_lock when switching vt.
3144 cookie = console_srcu_read_lock();
3145 for_each_console_srcu(c) {
3148 driver = c->device(c, index);
3152 console_srcu_read_unlock(cookie);
3159 * Prevent further output on the passed console device so that (for example)
3160 * serial drivers can disable console output before suspending a port, and can
3161 * re-enable output afterwards.
3163 void console_stop(struct console *console)
3165 __pr_flush(console, 1000, true);
3166 console_list_lock();
3167 console_srcu_write_flags(console, console->flags & ~CON_ENABLED);
3168 console_list_unlock();
3171 * Ensure that all SRCU list walks have completed. All contexts must
3172 * be able to see that this console is disabled so that (for example)
3173 * the caller can suspend the port without risk of another context
3176 synchronize_srcu(&console_srcu);
3178 EXPORT_SYMBOL(console_stop);
3180 void console_start(struct console *console)
3182 console_list_lock();
3183 console_srcu_write_flags(console, console->flags | CON_ENABLED);
3184 console_list_unlock();
3185 __pr_flush(console, 1000, true);
3187 EXPORT_SYMBOL(console_start);
3189 static int __read_mostly keep_bootcon;
3191 static int __init keep_bootcon_setup(char *str)
3194 pr_info("debug: skip boot console de-registration.\n");
3199 early_param("keep_bootcon", keep_bootcon_setup);
3202 * This is called by register_console() to try to match
3203 * the newly registered console with any of the ones selected
3204 * by either the command line or add_preferred_console() and
3207 * Care need to be taken with consoles that are statically
3208 * enabled such as netconsole
3210 static int try_enable_preferred_console(struct console *newcon,
3211 bool user_specified)
3213 struct console_cmdline *c;
3216 for (i = 0, c = console_cmdline;
3217 i < MAX_CMDLINECONSOLES && c->name[0];
3219 if (c->user_specified != user_specified)
3221 if (!newcon->match ||
3222 newcon->match(newcon, c->name, c->index, c->options) != 0) {
3223 /* default matching */
3224 BUILD_BUG_ON(sizeof(c->name) != sizeof(newcon->name));
3225 if (strcmp(c->name, newcon->name) != 0)
3227 if (newcon->index >= 0 &&
3228 newcon->index != c->index)
3230 if (newcon->index < 0)
3231 newcon->index = c->index;
3233 if (_braille_register_console(newcon, c))
3236 if (newcon->setup &&
3237 (err = newcon->setup(newcon, c->options)) != 0)
3240 newcon->flags |= CON_ENABLED;
3241 if (i == preferred_console)
3242 newcon->flags |= CON_CONSDEV;
3247 * Some consoles, such as pstore and netconsole, can be enabled even
3248 * without matching. Accept the pre-enabled consoles only when match()
3249 * and setup() had a chance to be called.
3251 if (newcon->flags & CON_ENABLED && c->user_specified == user_specified)
3257 /* Try to enable the console unconditionally */
3258 static void try_enable_default_console(struct console *newcon)
3260 if (newcon->index < 0)
3263 if (newcon->setup && newcon->setup(newcon, NULL) != 0)
3266 newcon->flags |= CON_ENABLED;
3269 newcon->flags |= CON_CONSDEV;
3272 #define con_printk(lvl, con, fmt, ...) \
3273 printk(lvl pr_fmt("%sconsole [%s%d] " fmt), \
3274 (con->flags & CON_BOOT) ? "boot" : "", \
3275 con->name, con->index, ##__VA_ARGS__)
3277 static void console_init_seq(struct console *newcon, bool bootcon_registered)
3279 struct console *con;
3282 if (newcon->flags & (CON_PRINTBUFFER | CON_BOOT)) {
3283 /* Get a consistent copy of @syslog_seq. */
3284 mutex_lock(&syslog_lock);
3285 newcon->seq = syslog_seq;
3286 mutex_unlock(&syslog_lock);
3288 /* Begin with next message added to ringbuffer. */
3289 newcon->seq = prb_next_seq(prb);
3292 * If any enabled boot consoles are due to be unregistered
3293 * shortly, some may not be caught up and may be the same
3294 * device as @newcon. Since it is not known which boot console
3295 * is the same device, flush all consoles and, if necessary,
3296 * start with the message of the enabled boot console that is
3297 * the furthest behind.
3299 if (bootcon_registered && !keep_bootcon) {
3301 * Hold the console_lock to stop console printing and
3302 * guarantee safe access to console->seq.
3307 * Flush all consoles and set the console to start at
3308 * the next unprinted sequence number.
3310 if (!console_flush_all(true, &newcon->seq, &handover)) {
3312 * Flushing failed. Just choose the lowest
3313 * sequence of the enabled boot consoles.
3317 * If there was a handover, this context no
3318 * longer holds the console_lock.
3323 newcon->seq = prb_next_seq(prb);
3324 for_each_console(con) {
3325 if ((con->flags & CON_BOOT) &&
3326 (con->flags & CON_ENABLED) &&
3327 con->seq < newcon->seq) {
3328 newcon->seq = con->seq;
3338 #define console_first() \
3339 hlist_entry(console_list.first, struct console, node)
3341 static int unregister_console_locked(struct console *console);
3344 * The console driver calls this routine during kernel initialization
3345 * to register the console printing procedure with printk() and to
3346 * print any messages that were printed by the kernel before the
3347 * console driver was initialized.
3349 * This can happen pretty early during the boot process (because of
3350 * early_printk) - sometimes before setup_arch() completes - be careful
3351 * of what kernel features are used - they may not be initialised yet.
3353 * There are two types of consoles - bootconsoles (early_printk) and
3354 * "real" consoles (everything which is not a bootconsole) which are
3355 * handled differently.
3356 * - Any number of bootconsoles can be registered at any time.
3357 * - As soon as a "real" console is registered, all bootconsoles
3358 * will be unregistered automatically.
3359 * - Once a "real" console is registered, any attempt to register a
3360 * bootconsoles will be rejected
3362 void register_console(struct console *newcon)
3364 struct console *con;
3365 bool bootcon_registered = false;
3366 bool realcon_registered = false;
3369 console_list_lock();
3371 for_each_console(con) {
3372 if (WARN(con == newcon, "console '%s%d' already registered\n",
3373 con->name, con->index)) {
3377 if (con->flags & CON_BOOT)
3378 bootcon_registered = true;
3380 realcon_registered = true;
3383 /* Do not register boot consoles when there already is a real one. */
3384 if ((newcon->flags & CON_BOOT) && realcon_registered) {
3385 pr_info("Too late to register bootconsole %s%d\n",
3386 newcon->name, newcon->index);
3391 * See if we want to enable this console driver by default.
3393 * Nope when a console is preferred by the command line, device
3396 * The first real console with tty binding (driver) wins. More
3397 * consoles might get enabled before the right one is found.
3399 * Note that a console with tty binding will have CON_CONSDEV
3400 * flag set and will be first in the list.
3402 if (preferred_console < 0) {
3403 if (hlist_empty(&console_list) || !console_first()->device ||
3404 console_first()->flags & CON_BOOT) {
3405 try_enable_default_console(newcon);
3409 /* See if this console matches one we selected on the command line */
3410 err = try_enable_preferred_console(newcon, true);
3412 /* If not, try to match against the platform default(s) */
3414 err = try_enable_preferred_console(newcon, false);
3416 /* printk() messages are not printed to the Braille console. */
3417 if (err || newcon->flags & CON_BRL)
3421 * If we have a bootconsole, and are switching to a real console,
3422 * don't print everything out again, since when the boot console, and
3423 * the real console are the same physical device, it's annoying to
3424 * see the beginning boot messages twice
3426 if (bootcon_registered &&
3427 ((newcon->flags & (CON_CONSDEV | CON_BOOT)) == CON_CONSDEV)) {
3428 newcon->flags &= ~CON_PRINTBUFFER;
3431 newcon->dropped = 0;
3432 console_init_seq(newcon, bootcon_registered);
3435 * Put this console in the list - keep the
3436 * preferred driver at the head of the list.
3438 if (hlist_empty(&console_list)) {
3439 /* Ensure CON_CONSDEV is always set for the head. */
3440 newcon->flags |= CON_CONSDEV;
3441 hlist_add_head_rcu(&newcon->node, &console_list);
3443 } else if (newcon->flags & CON_CONSDEV) {
3444 /* Only the new head can have CON_CONSDEV set. */
3445 console_srcu_write_flags(console_first(), console_first()->flags & ~CON_CONSDEV);
3446 hlist_add_head_rcu(&newcon->node, &console_list);
3449 hlist_add_behind_rcu(&newcon->node, console_list.first);
3453 * No need to synchronize SRCU here! The caller does not rely
3454 * on all contexts being able to see the new console before
3455 * register_console() completes.
3458 console_sysfs_notify();
3461 * By unregistering the bootconsoles after we enable the real console
3462 * we get the "console xxx enabled" message on all the consoles -
3463 * boot consoles, real consoles, etc - this is to ensure that end
3464 * users know there might be something in the kernel's log buffer that
3465 * went to the bootconsole (that they do not see on the real console)
3467 con_printk(KERN_INFO, newcon, "enabled\n");
3468 if (bootcon_registered &&
3469 ((newcon->flags & (CON_CONSDEV | CON_BOOT)) == CON_CONSDEV) &&
3471 struct hlist_node *tmp;
3473 hlist_for_each_entry_safe(con, tmp, &console_list, node) {
3474 if (con->flags & CON_BOOT)
3475 unregister_console_locked(con);
3479 console_list_unlock();
3481 EXPORT_SYMBOL(register_console);
3483 /* Must be called under console_list_lock(). */
3484 static int unregister_console_locked(struct console *console)
3488 lockdep_assert_console_list_lock_held();
3490 con_printk(KERN_INFO, console, "disabled\n");
3492 res = _braille_unregister_console(console);
3498 /* Disable it unconditionally */
3499 console_srcu_write_flags(console, console->flags & ~CON_ENABLED);
3501 if (!console_is_registered_locked(console))
3504 hlist_del_init_rcu(&console->node);
3508 * If this isn't the last console and it has CON_CONSDEV set, we
3509 * need to set it on the next preferred console.
3512 * The above makes no sense as there is no guarantee that the next
3513 * console has any device attached. Oh well....
3515 if (!hlist_empty(&console_list) && console->flags & CON_CONSDEV)
3516 console_srcu_write_flags(console_first(), console_first()->flags | CON_CONSDEV);
3519 * Ensure that all SRCU list walks have completed. All contexts
3520 * must not be able to see this console in the list so that any
3521 * exit/cleanup routines can be performed safely.
3523 synchronize_srcu(&console_srcu);
3525 console_sysfs_notify();
3528 res = console->exit(console);
3533 int unregister_console(struct console *console)
3537 console_list_lock();
3538 res = unregister_console_locked(console);
3539 console_list_unlock();
3542 EXPORT_SYMBOL(unregister_console);
3545 * console_force_preferred_locked - force a registered console preferred
3546 * @con: The registered console to force preferred.
3548 * Must be called under console_list_lock().
3550 void console_force_preferred_locked(struct console *con)
3552 struct console *cur_pref_con;
3554 if (!console_is_registered_locked(con))
3557 cur_pref_con = console_first();
3559 /* Already preferred? */
3560 if (cur_pref_con == con)
3564 * Delete, but do not re-initialize the entry. This allows the console
3565 * to continue to appear registered (via any hlist_unhashed_lockless()
3566 * checks), even though it was briefly removed from the console list.
3568 hlist_del_rcu(&con->node);
3571 * Ensure that all SRCU list walks have completed so that the console
3572 * can be added to the beginning of the console list and its forward
3573 * list pointer can be re-initialized.
3575 synchronize_srcu(&console_srcu);
3577 con->flags |= CON_CONSDEV;
3578 WARN_ON(!con->device);
3580 /* Only the new head can have CON_CONSDEV set. */
3581 console_srcu_write_flags(cur_pref_con, cur_pref_con->flags & ~CON_CONSDEV);
3582 hlist_add_head_rcu(&con->node, &console_list);
3584 EXPORT_SYMBOL(console_force_preferred_locked);
3587 * Initialize the console device. This is called *early*, so
3588 * we can't necessarily depend on lots of kernel help here.
3589 * Just do some early initializations, and do the complex setup
3592 void __init console_init(void)
3596 initcall_entry_t *ce;
3598 /* Setup the default TTY line discipline. */
3602 * set up the console device so that later boot sequences can
3603 * inform about problems etc..
3605 ce = __con_initcall_start;
3606 trace_initcall_level("console");
3607 while (ce < __con_initcall_end) {
3608 call = initcall_from_entry(ce);
3609 trace_initcall_start(call);
3611 trace_initcall_finish(call, ret);
3617 * Some boot consoles access data that is in the init section and which will
3618 * be discarded after the initcalls have been run. To make sure that no code
3619 * will access this data, unregister the boot consoles in a late initcall.
3621 * If for some reason, such as deferred probe or the driver being a loadable
3622 * module, the real console hasn't registered yet at this point, there will
3623 * be a brief interval in which no messages are logged to the console, which
3624 * makes it difficult to diagnose problems that occur during this time.
3626 * To mitigate this problem somewhat, only unregister consoles whose memory
3627 * intersects with the init section. Note that all other boot consoles will
3628 * get unregistered when the real preferred console is registered.
3630 static int __init printk_late_init(void)
3632 struct hlist_node *tmp;
3633 struct console *con;
3636 console_list_lock();
3637 hlist_for_each_entry_safe(con, tmp, &console_list, node) {
3638 if (!(con->flags & CON_BOOT))
3641 /* Check addresses that might be used for enabled consoles. */
3642 if (init_section_intersects(con, sizeof(*con)) ||
3643 init_section_contains(con->write, 0) ||
3644 init_section_contains(con->read, 0) ||
3645 init_section_contains(con->device, 0) ||
3646 init_section_contains(con->unblank, 0) ||
3647 init_section_contains(con->data, 0)) {
3649 * Please, consider moving the reported consoles out
3650 * of the init section.
3652 pr_warn("bootconsole [%s%d] uses init memory and must be disabled even before the real one is ready\n",
3653 con->name, con->index);
3654 unregister_console_locked(con);
3657 console_list_unlock();
3659 ret = cpuhp_setup_state_nocalls(CPUHP_PRINTK_DEAD, "printk:dead", NULL,
3660 console_cpu_notify);
3662 ret = cpuhp_setup_state_nocalls(CPUHP_AP_ONLINE_DYN, "printk:online",
3663 console_cpu_notify, NULL);
3665 printk_sysctl_init();
3668 late_initcall(printk_late_init);
3670 #if defined CONFIG_PRINTK
3671 /* If @con is specified, only wait for that console. Otherwise wait for all. */
3672 static bool __pr_flush(struct console *con, int timeout_ms, bool reset_on_progress)
3674 int remaining = timeout_ms;
3684 seq = prb_next_seq(prb);
3690 * Hold the console_lock to guarantee safe access to
3691 * console->seq and to prevent changes to @console_suspended
3692 * until all consoles have been processed.
3696 cookie = console_srcu_read_lock();
3697 for_each_console_srcu(c) {
3698 if (con && con != c)
3700 if (!console_is_usable(c))
3702 printk_seq = c->seq;
3703 if (printk_seq < seq)
3704 diff += seq - printk_seq;
3706 console_srcu_read_unlock(cookie);
3709 * If consoles are suspended, it cannot be expected that they
3710 * make forward progress, so timeout immediately. @diff is
3711 * still used to return a valid flush status.
3713 if (console_suspended)
3715 else if (diff != last_diff && reset_on_progress)
3716 remaining = timeout_ms;
3720 if (diff == 0 || remaining == 0)
3723 if (remaining < 0) {
3724 /* no timeout limit */
3726 } else if (remaining < 100) {
3741 * pr_flush() - Wait for printing threads to catch up.
3743 * @timeout_ms: The maximum time (in ms) to wait.
3744 * @reset_on_progress: Reset the timeout if forward progress is seen.
3746 * A value of 0 for @timeout_ms means no waiting will occur. A value of -1
3747 * represents infinite waiting.
3749 * If @reset_on_progress is true, the timeout will be reset whenever any
3750 * printer has been seen to make some forward progress.
3752 * Context: Process context. May sleep while acquiring console lock.
3753 * Return: true if all enabled printers are caught up.
3755 static bool pr_flush(int timeout_ms, bool reset_on_progress)
3757 return __pr_flush(NULL, timeout_ms, reset_on_progress);
3761 * Delayed printk version, for scheduler-internal messages:
3763 #define PRINTK_PENDING_WAKEUP 0x01
3764 #define PRINTK_PENDING_OUTPUT 0x02
3766 static DEFINE_PER_CPU(int, printk_pending);
3768 static void wake_up_klogd_work_func(struct irq_work *irq_work)
3770 int pending = this_cpu_xchg(printk_pending, 0);
3772 if (pending & PRINTK_PENDING_OUTPUT) {
3773 /* If trylock fails, someone else is doing the printing */
3774 if (console_trylock())
3778 if (pending & PRINTK_PENDING_WAKEUP)
3779 wake_up_interruptible(&log_wait);
3782 static DEFINE_PER_CPU(struct irq_work, wake_up_klogd_work) =
3783 IRQ_WORK_INIT_LAZY(wake_up_klogd_work_func);
3785 static void __wake_up_klogd(int val)
3787 if (!printk_percpu_data_ready())
3792 * Guarantee any new records can be seen by tasks preparing to wait
3793 * before this context checks if the wait queue is empty.
3795 * The full memory barrier within wq_has_sleeper() pairs with the full
3796 * memory barrier within set_current_state() of
3797 * prepare_to_wait_event(), which is called after ___wait_event() adds
3798 * the waiter but before it has checked the wait condition.
3800 * This pairs with devkmsg_read:A and syslog_print:A.
3802 if (wq_has_sleeper(&log_wait) || /* LMM(__wake_up_klogd:A) */
3803 (val & PRINTK_PENDING_OUTPUT)) {
3804 this_cpu_or(printk_pending, val);
3805 irq_work_queue(this_cpu_ptr(&wake_up_klogd_work));
3810 void wake_up_klogd(void)
3812 __wake_up_klogd(PRINTK_PENDING_WAKEUP);
3815 void defer_console_output(void)
3818 * New messages may have been added directly to the ringbuffer
3819 * using vprintk_store(), so wake any waiters as well.
3821 __wake_up_klogd(PRINTK_PENDING_WAKEUP | PRINTK_PENDING_OUTPUT);
3824 void printk_trigger_flush(void)
3826 defer_console_output();
3829 int vprintk_deferred(const char *fmt, va_list args)
3833 r = vprintk_emit(0, LOGLEVEL_SCHED, NULL, fmt, args);
3834 defer_console_output();
3839 int _printk_deferred(const char *fmt, ...)
3844 va_start(args, fmt);
3845 r = vprintk_deferred(fmt, args);
3852 * printk rate limiting, lifted from the networking subsystem.
3854 * This enforces a rate limit: not more than 10 kernel messages
3855 * every 5s to make a denial-of-service attack impossible.
3857 DEFINE_RATELIMIT_STATE(printk_ratelimit_state, 5 * HZ, 10);
3859 int __printk_ratelimit(const char *func)
3861 return ___ratelimit(&printk_ratelimit_state, func);
3863 EXPORT_SYMBOL(__printk_ratelimit);
3866 * printk_timed_ratelimit - caller-controlled printk ratelimiting
3867 * @caller_jiffies: pointer to caller's state
3868 * @interval_msecs: minimum interval between prints
3870 * printk_timed_ratelimit() returns true if more than @interval_msecs
3871 * milliseconds have elapsed since the last time printk_timed_ratelimit()
3874 bool printk_timed_ratelimit(unsigned long *caller_jiffies,
3875 unsigned int interval_msecs)
3877 unsigned long elapsed = jiffies - *caller_jiffies;
3879 if (*caller_jiffies && elapsed <= msecs_to_jiffies(interval_msecs))
3882 *caller_jiffies = jiffies;
3885 EXPORT_SYMBOL(printk_timed_ratelimit);
3887 static DEFINE_SPINLOCK(dump_list_lock);
3888 static LIST_HEAD(dump_list);
3891 * kmsg_dump_register - register a kernel log dumper.
3892 * @dumper: pointer to the kmsg_dumper structure
3894 * Adds a kernel log dumper to the system. The dump callback in the
3895 * structure will be called when the kernel oopses or panics and must be
3896 * set. Returns zero on success and %-EINVAL or %-EBUSY otherwise.
3898 int kmsg_dump_register(struct kmsg_dumper *dumper)
3900 unsigned long flags;
3903 /* The dump callback needs to be set */
3907 spin_lock_irqsave(&dump_list_lock, flags);
3908 /* Don't allow registering multiple times */
3909 if (!dumper->registered) {
3910 dumper->registered = 1;
3911 list_add_tail_rcu(&dumper->list, &dump_list);
3914 spin_unlock_irqrestore(&dump_list_lock, flags);
3918 EXPORT_SYMBOL_GPL(kmsg_dump_register);
3921 * kmsg_dump_unregister - unregister a kmsg dumper.
3922 * @dumper: pointer to the kmsg_dumper structure
3924 * Removes a dump device from the system. Returns zero on success and
3925 * %-EINVAL otherwise.
3927 int kmsg_dump_unregister(struct kmsg_dumper *dumper)
3929 unsigned long flags;
3932 spin_lock_irqsave(&dump_list_lock, flags);
3933 if (dumper->registered) {
3934 dumper->registered = 0;
3935 list_del_rcu(&dumper->list);
3938 spin_unlock_irqrestore(&dump_list_lock, flags);
3943 EXPORT_SYMBOL_GPL(kmsg_dump_unregister);
3945 static bool always_kmsg_dump;
3946 module_param_named(always_kmsg_dump, always_kmsg_dump, bool, S_IRUGO | S_IWUSR);
3948 const char *kmsg_dump_reason_str(enum kmsg_dump_reason reason)
3951 case KMSG_DUMP_PANIC:
3953 case KMSG_DUMP_OOPS:
3955 case KMSG_DUMP_EMERG:
3957 case KMSG_DUMP_SHUTDOWN:
3963 EXPORT_SYMBOL_GPL(kmsg_dump_reason_str);
3966 * kmsg_dump - dump kernel log to kernel message dumpers.
3967 * @reason: the reason (oops, panic etc) for dumping
3969 * Call each of the registered dumper's dump() callback, which can
3970 * retrieve the kmsg records with kmsg_dump_get_line() or
3971 * kmsg_dump_get_buffer().
3973 void kmsg_dump(enum kmsg_dump_reason reason)
3975 struct kmsg_dumper *dumper;
3978 list_for_each_entry_rcu(dumper, &dump_list, list) {
3979 enum kmsg_dump_reason max_reason = dumper->max_reason;
3982 * If client has not provided a specific max_reason, default
3983 * to KMSG_DUMP_OOPS, unless always_kmsg_dump was set.
3985 if (max_reason == KMSG_DUMP_UNDEF) {
3986 max_reason = always_kmsg_dump ? KMSG_DUMP_MAX :
3989 if (reason > max_reason)
3992 /* invoke dumper which will iterate over records */
3993 dumper->dump(dumper, reason);
3999 * kmsg_dump_get_line - retrieve one kmsg log line
4000 * @iter: kmsg dump iterator
4001 * @syslog: include the "<4>" prefixes
4002 * @line: buffer to copy the line to
4003 * @size: maximum size of the buffer
4004 * @len: length of line placed into buffer
4006 * Start at the beginning of the kmsg buffer, with the oldest kmsg
4007 * record, and copy one record into the provided buffer.
4009 * Consecutive calls will return the next available record moving
4010 * towards the end of the buffer with the youngest messages.
4012 * A return value of FALSE indicates that there are no more records to
4015 bool kmsg_dump_get_line(struct kmsg_dump_iter *iter, bool syslog,
4016 char *line, size_t size, size_t *len)
4018 u64 min_seq = latched_seq_read_nolock(&clear_seq);
4019 struct printk_info info;
4020 unsigned int line_count;
4021 struct printk_record r;
4025 if (iter->cur_seq < min_seq)
4026 iter->cur_seq = min_seq;
4028 prb_rec_init_rd(&r, &info, line, size);
4030 /* Read text or count text lines? */
4032 if (!prb_read_valid(prb, iter->cur_seq, &r))
4034 l = record_print_text(&r, syslog, printk_time);
4036 if (!prb_read_valid_info(prb, iter->cur_seq,
4037 &info, &line_count)) {
4040 l = get_record_print_text_size(&info, line_count, syslog,
4045 iter->cur_seq = r.info->seq + 1;
4052 EXPORT_SYMBOL_GPL(kmsg_dump_get_line);
4055 * kmsg_dump_get_buffer - copy kmsg log lines
4056 * @iter: kmsg dump iterator
4057 * @syslog: include the "<4>" prefixes
4058 * @buf: buffer to copy the line to
4059 * @size: maximum size of the buffer
4060 * @len_out: length of line placed into buffer
4062 * Start at the end of the kmsg buffer and fill the provided buffer
4063 * with as many of the *youngest* kmsg records that fit into it.
4064 * If the buffer is large enough, all available kmsg records will be
4065 * copied with a single call.
4067 * Consecutive calls will fill the buffer with the next block of
4068 * available older records, not including the earlier retrieved ones.
4070 * A return value of FALSE indicates that there are no more records to
4073 bool kmsg_dump_get_buffer(struct kmsg_dump_iter *iter, bool syslog,
4074 char *buf, size_t size, size_t *len_out)
4076 u64 min_seq = latched_seq_read_nolock(&clear_seq);
4077 struct printk_info info;
4078 struct printk_record r;
4083 bool time = printk_time;
4088 if (iter->cur_seq < min_seq)
4089 iter->cur_seq = min_seq;
4091 if (prb_read_valid_info(prb, iter->cur_seq, &info, NULL)) {
4092 if (info.seq != iter->cur_seq) {
4093 /* messages are gone, move to first available one */
4094 iter->cur_seq = info.seq;
4099 if (iter->cur_seq >= iter->next_seq)
4103 * Find first record that fits, including all following records,
4104 * into the user-provided buffer for this dump. Pass in size-1
4105 * because this function (by way of record_print_text()) will
4106 * not write more than size-1 bytes of text into @buf.
4108 seq = find_first_fitting_seq(iter->cur_seq, iter->next_seq,
4109 size - 1, syslog, time);
4112 * Next kmsg_dump_get_buffer() invocation will dump block of
4113 * older records stored right before this one.
4117 prb_rec_init_rd(&r, &info, buf, size);
4120 prb_for_each_record(seq, prb, seq, &r) {
4121 if (r.info->seq >= iter->next_seq)
4124 len += record_print_text(&r, syslog, time);
4126 /* Adjust record to store to remaining buffer space. */
4127 prb_rec_init_rd(&r, &info, buf + len, size - len);
4130 iter->next_seq = next_seq;
4137 EXPORT_SYMBOL_GPL(kmsg_dump_get_buffer);
4140 * kmsg_dump_rewind - reset the iterator
4141 * @iter: kmsg dump iterator
4143 * Reset the dumper's iterator so that kmsg_dump_get_line() and
4144 * kmsg_dump_get_buffer() can be called again and used multiple
4145 * times within the same dumper.dump() callback.
4147 void kmsg_dump_rewind(struct kmsg_dump_iter *iter)
4149 iter->cur_seq = latched_seq_read_nolock(&clear_seq);
4150 iter->next_seq = prb_next_seq(prb);
4152 EXPORT_SYMBOL_GPL(kmsg_dump_rewind);
4157 static atomic_t printk_cpu_sync_owner = ATOMIC_INIT(-1);
4158 static atomic_t printk_cpu_sync_nested = ATOMIC_INIT(0);
4161 * __printk_cpu_sync_wait() - Busy wait until the printk cpu-reentrant
4162 * spinning lock is not owned by any CPU.
4164 * Context: Any context.
4166 void __printk_cpu_sync_wait(void)
4170 } while (atomic_read(&printk_cpu_sync_owner) != -1);
4172 EXPORT_SYMBOL(__printk_cpu_sync_wait);
4175 * __printk_cpu_sync_try_get() - Try to acquire the printk cpu-reentrant
4178 * If no processor has the lock, the calling processor takes the lock and
4179 * becomes the owner. If the calling processor is already the owner of the
4180 * lock, this function succeeds immediately.
4182 * Context: Any context. Expects interrupts to be disabled.
4183 * Return: 1 on success, otherwise 0.
4185 int __printk_cpu_sync_try_get(void)
4190 cpu = smp_processor_id();
4193 * Guarantee loads and stores from this CPU when it is the lock owner
4194 * are _not_ visible to the previous lock owner. This pairs with
4195 * __printk_cpu_sync_put:B.
4197 * Memory barrier involvement:
4199 * If __printk_cpu_sync_try_get:A reads from __printk_cpu_sync_put:B,
4200 * then __printk_cpu_sync_put:A can never read from
4201 * __printk_cpu_sync_try_get:B.
4205 * RELEASE from __printk_cpu_sync_put:A to __printk_cpu_sync_put:B
4206 * of the previous CPU
4208 * ACQUIRE from __printk_cpu_sync_try_get:A to
4209 * __printk_cpu_sync_try_get:B of this CPU
4211 old = atomic_cmpxchg_acquire(&printk_cpu_sync_owner, -1,
4212 cpu); /* LMM(__printk_cpu_sync_try_get:A) */
4215 * This CPU is now the owner and begins loading/storing
4216 * data: LMM(__printk_cpu_sync_try_get:B)
4220 } else if (old == cpu) {
4221 /* This CPU is already the owner. */
4222 atomic_inc(&printk_cpu_sync_nested);
4228 EXPORT_SYMBOL(__printk_cpu_sync_try_get);
4231 * __printk_cpu_sync_put() - Release the printk cpu-reentrant spinning lock.
4233 * The calling processor must be the owner of the lock.
4235 * Context: Any context. Expects interrupts to be disabled.
4237 void __printk_cpu_sync_put(void)
4239 if (atomic_read(&printk_cpu_sync_nested)) {
4240 atomic_dec(&printk_cpu_sync_nested);
4245 * This CPU is finished loading/storing data:
4246 * LMM(__printk_cpu_sync_put:A)
4250 * Guarantee loads and stores from this CPU when it was the
4251 * lock owner are visible to the next lock owner. This pairs
4252 * with __printk_cpu_sync_try_get:A.
4254 * Memory barrier involvement:
4256 * If __printk_cpu_sync_try_get:A reads from __printk_cpu_sync_put:B,
4257 * then __printk_cpu_sync_try_get:B reads from __printk_cpu_sync_put:A.
4261 * RELEASE from __printk_cpu_sync_put:A to __printk_cpu_sync_put:B
4264 * ACQUIRE from __printk_cpu_sync_try_get:A to
4265 * __printk_cpu_sync_try_get:B of the next CPU
4267 atomic_set_release(&printk_cpu_sync_owner,
4268 -1); /* LMM(__printk_cpu_sync_put:B) */
4270 EXPORT_SYMBOL(__printk_cpu_sync_put);
4271 #endif /* CONFIG_SMP */