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);
74 EXPORT_TRACEPOINT_SYMBOL_GPL(console);
77 * Low level drivers may need that to know if they can schedule in
78 * their unblank() callback or not. So let's export it.
81 EXPORT_SYMBOL(oops_in_progress);
84 * console_mutex protects console_list updates and console->flags updates.
85 * The flags are synchronized only for consoles that are registered, i.e.
86 * accessible via the console list.
88 static DEFINE_MUTEX(console_mutex);
91 * console_sem protects updates to console->seq and console_suspended,
92 * and also provides serialization for console printing.
94 static DEFINE_SEMAPHORE(console_sem);
95 HLIST_HEAD(console_list);
96 EXPORT_SYMBOL_GPL(console_list);
97 DEFINE_STATIC_SRCU(console_srcu);
100 * System may need to suppress printk message under certain
101 * circumstances, like after kernel panic happens.
103 int __read_mostly suppress_printk;
106 * During panic, heavy printk by other CPUs can delay the
107 * panic and risk deadlock on console resources.
109 static int __read_mostly suppress_panic_printk;
111 #ifdef CONFIG_LOCKDEP
112 static struct lockdep_map console_lock_dep_map = {
113 .name = "console_lock"
116 void lockdep_assert_console_list_lock_held(void)
118 lockdep_assert_held(&console_mutex);
120 EXPORT_SYMBOL(lockdep_assert_console_list_lock_held);
123 #ifdef CONFIG_DEBUG_LOCK_ALLOC
124 bool console_srcu_read_lock_is_held(void)
126 return srcu_read_lock_held(&console_srcu);
128 EXPORT_SYMBOL(console_srcu_read_lock_is_held);
131 enum devkmsg_log_bits {
132 __DEVKMSG_LOG_BIT_ON = 0,
133 __DEVKMSG_LOG_BIT_OFF,
134 __DEVKMSG_LOG_BIT_LOCK,
137 enum devkmsg_log_masks {
138 DEVKMSG_LOG_MASK_ON = BIT(__DEVKMSG_LOG_BIT_ON),
139 DEVKMSG_LOG_MASK_OFF = BIT(__DEVKMSG_LOG_BIT_OFF),
140 DEVKMSG_LOG_MASK_LOCK = BIT(__DEVKMSG_LOG_BIT_LOCK),
143 /* Keep both the 'on' and 'off' bits clear, i.e. ratelimit by default: */
144 #define DEVKMSG_LOG_MASK_DEFAULT 0
146 static unsigned int __read_mostly devkmsg_log = DEVKMSG_LOG_MASK_DEFAULT;
148 static int __control_devkmsg(char *str)
155 len = str_has_prefix(str, "on");
157 devkmsg_log = DEVKMSG_LOG_MASK_ON;
161 len = str_has_prefix(str, "off");
163 devkmsg_log = DEVKMSG_LOG_MASK_OFF;
167 len = str_has_prefix(str, "ratelimit");
169 devkmsg_log = DEVKMSG_LOG_MASK_DEFAULT;
176 static int __init control_devkmsg(char *str)
178 if (__control_devkmsg(str) < 0) {
179 pr_warn("printk.devkmsg: bad option string '%s'\n", str);
184 * Set sysctl string accordingly:
186 if (devkmsg_log == DEVKMSG_LOG_MASK_ON)
187 strcpy(devkmsg_log_str, "on");
188 else if (devkmsg_log == DEVKMSG_LOG_MASK_OFF)
189 strcpy(devkmsg_log_str, "off");
190 /* else "ratelimit" which is set by default. */
193 * Sysctl cannot change it anymore. The kernel command line setting of
194 * this parameter is to force the setting to be permanent throughout the
195 * runtime of the system. This is a precation measure against userspace
196 * trying to be a smarta** and attempting to change it up on us.
198 devkmsg_log |= DEVKMSG_LOG_MASK_LOCK;
202 __setup("printk.devkmsg=", control_devkmsg);
204 char devkmsg_log_str[DEVKMSG_STR_MAX_SIZE] = "ratelimit";
205 #if defined(CONFIG_PRINTK) && defined(CONFIG_SYSCTL)
206 int devkmsg_sysctl_set_loglvl(struct ctl_table *table, int write,
207 void *buffer, size_t *lenp, loff_t *ppos)
209 char old_str[DEVKMSG_STR_MAX_SIZE];
214 if (devkmsg_log & DEVKMSG_LOG_MASK_LOCK)
218 strncpy(old_str, devkmsg_log_str, DEVKMSG_STR_MAX_SIZE);
221 err = proc_dostring(table, write, buffer, lenp, ppos);
226 err = __control_devkmsg(devkmsg_log_str);
229 * Do not accept an unknown string OR a known string with
232 if (err < 0 || (err + 1 != *lenp)) {
234 /* ... and restore old setting. */
236 strncpy(devkmsg_log_str, old_str, DEVKMSG_STR_MAX_SIZE);
244 #endif /* CONFIG_PRINTK && CONFIG_SYSCTL */
247 * console_list_lock - Lock the console list
249 * For console list or console->flags updates
251 void console_list_lock(void)
254 * In unregister_console() and console_force_preferred_locked(),
255 * synchronize_srcu() is called with the console_list_lock held.
256 * Therefore it is not allowed that the console_list_lock is taken
257 * with the srcu_lock held.
259 * Detecting if this context is really in the read-side critical
260 * section is only possible if the appropriate debug options are
263 WARN_ON_ONCE(debug_lockdep_rcu_enabled() &&
264 srcu_read_lock_held(&console_srcu));
266 mutex_lock(&console_mutex);
268 EXPORT_SYMBOL(console_list_lock);
271 * console_list_unlock - Unlock the console list
273 * Counterpart to console_list_lock()
275 void console_list_unlock(void)
277 mutex_unlock(&console_mutex);
279 EXPORT_SYMBOL(console_list_unlock);
282 * console_srcu_read_lock - Register a new reader for the
283 * SRCU-protected console list
285 * Use for_each_console_srcu() to iterate the console list
287 * Context: Any context.
288 * Return: A cookie to pass to console_srcu_read_unlock().
290 int console_srcu_read_lock(void)
292 return srcu_read_lock_nmisafe(&console_srcu);
294 EXPORT_SYMBOL(console_srcu_read_lock);
297 * console_srcu_read_unlock - Unregister an old reader from
298 * the SRCU-protected console list
299 * @cookie: cookie returned from console_srcu_read_lock()
301 * Counterpart to console_srcu_read_lock()
303 void console_srcu_read_unlock(int cookie)
305 srcu_read_unlock_nmisafe(&console_srcu, cookie);
307 EXPORT_SYMBOL(console_srcu_read_unlock);
310 * Helper macros to handle lockdep when locking/unlocking console_sem. We use
311 * macros instead of functions so that _RET_IP_ contains useful information.
313 #define down_console_sem() do { \
315 mutex_acquire(&console_lock_dep_map, 0, 0, _RET_IP_);\
318 static int __down_trylock_console_sem(unsigned long ip)
324 * Here and in __up_console_sem() we need to be in safe mode,
325 * because spindump/WARN/etc from under console ->lock will
326 * deadlock in printk()->down_trylock_console_sem() otherwise.
328 printk_safe_enter_irqsave(flags);
329 lock_failed = down_trylock(&console_sem);
330 printk_safe_exit_irqrestore(flags);
334 mutex_acquire(&console_lock_dep_map, 0, 1, ip);
337 #define down_trylock_console_sem() __down_trylock_console_sem(_RET_IP_)
339 static void __up_console_sem(unsigned long ip)
343 mutex_release(&console_lock_dep_map, ip);
345 printk_safe_enter_irqsave(flags);
347 printk_safe_exit_irqrestore(flags);
349 #define up_console_sem() __up_console_sem(_RET_IP_)
351 static bool panic_in_progress(void)
353 return unlikely(atomic_read(&panic_cpu) != PANIC_CPU_INVALID);
357 * This is used for debugging the mess that is the VT code by
358 * keeping track if we have the console semaphore held. It's
359 * definitely not the perfect debug tool (we don't know if _WE_
360 * hold it and are racing, but it helps tracking those weird code
361 * paths in the console code where we end up in places I want
362 * locked without the console semaphore held).
364 static int console_locked, console_suspended;
367 * Array of consoles built from command line options (console=)
370 #define MAX_CMDLINECONSOLES 8
372 static struct console_cmdline console_cmdline[MAX_CMDLINECONSOLES];
374 static int preferred_console = -1;
375 int console_set_on_cmdline;
376 EXPORT_SYMBOL(console_set_on_cmdline);
378 /* Flag: console code may call schedule() */
379 static int console_may_schedule;
381 enum con_msg_format_flags {
382 MSG_FORMAT_DEFAULT = 0,
383 MSG_FORMAT_SYSLOG = (1 << 0),
386 static int console_msg_format = MSG_FORMAT_DEFAULT;
389 * The printk log buffer consists of a sequenced collection of records, each
390 * containing variable length message text. Every record also contains its
391 * own meta-data (@info).
393 * Every record meta-data carries the timestamp in microseconds, as well as
394 * the standard userspace syslog level and syslog facility. The usual kernel
395 * messages use LOG_KERN; userspace-injected messages always carry a matching
396 * syslog facility, by default LOG_USER. The origin of every message can be
397 * reliably determined that way.
399 * The human readable log message of a record is available in @text, the
400 * length of the message text in @text_len. The stored message is not
403 * Optionally, a record can carry a dictionary of properties (key/value
404 * pairs), to provide userspace with a machine-readable message context.
406 * Examples for well-defined, commonly used property names are:
407 * DEVICE=b12:8 device identifier
411 * +sound:card0 subsystem:devname
412 * SUBSYSTEM=pci driver-core subsystem name
414 * Valid characters in property names are [a-zA-Z0-9.-_]. Property names
415 * and values are terminated by a '\0' character.
417 * Example of record values:
418 * record.text_buf = "it's a line" (unterminated)
419 * record.info.seq = 56
420 * record.info.ts_nsec = 36863
421 * record.info.text_len = 11
422 * record.info.facility = 0 (LOG_KERN)
423 * record.info.flags = 0
424 * record.info.level = 3 (LOG_ERR)
425 * record.info.caller_id = 299 (task 299)
426 * record.info.dev_info.subsystem = "pci" (terminated)
427 * record.info.dev_info.device = "+pci:0000:00:01.0" (terminated)
429 * The 'struct printk_info' buffer must never be directly exported to
430 * userspace, it is a kernel-private implementation detail that might
431 * need to be changed in the future, when the requirements change.
433 * /dev/kmsg exports the structured data in the following line format:
434 * "<level>,<sequnum>,<timestamp>,<contflag>[,additional_values, ... ];<message text>\n"
436 * Users of the export format should ignore possible additional values
437 * separated by ',', and find the message after the ';' character.
439 * The optional key/value pairs are attached as continuation lines starting
440 * with a space character and terminated by a newline. All possible
441 * non-prinatable characters are escaped in the "\xff" notation.
444 /* syslog_lock protects syslog_* variables and write access to clear_seq. */
445 static DEFINE_MUTEX(syslog_lock);
448 DECLARE_WAIT_QUEUE_HEAD(log_wait);
449 /* All 3 protected by @syslog_lock. */
450 /* the next printk record to read by syslog(READ) or /proc/kmsg */
451 static u64 syslog_seq;
452 static size_t syslog_partial;
453 static bool syslog_time;
456 seqcount_latch_t latch;
461 * The next printk record to read after the last 'clear' command. There are
462 * two copies (updated with seqcount_latch) so that reads can locklessly
463 * access a valid value. Writers are synchronized by @syslog_lock.
465 static struct latched_seq clear_seq = {
466 .latch = SEQCNT_LATCH_ZERO(clear_seq.latch),
471 #define LOG_LEVEL(v) ((v) & 0x07)
472 #define LOG_FACILITY(v) ((v) >> 3 & 0xff)
475 #define LOG_ALIGN __alignof__(unsigned long)
476 #define __LOG_BUF_LEN (1 << CONFIG_LOG_BUF_SHIFT)
477 #define LOG_BUF_LEN_MAX (u32)(1 << 31)
478 static char __log_buf[__LOG_BUF_LEN] __aligned(LOG_ALIGN);
479 static char *log_buf = __log_buf;
480 static u32 log_buf_len = __LOG_BUF_LEN;
483 * Define the average message size. This only affects the number of
484 * descriptors that will be available. Underestimating is better than
485 * overestimating (too many available descriptors is better than not enough).
487 #define PRB_AVGBITS 5 /* 32 character average length */
489 #if CONFIG_LOG_BUF_SHIFT <= PRB_AVGBITS
490 #error CONFIG_LOG_BUF_SHIFT value too small.
492 _DEFINE_PRINTKRB(printk_rb_static, CONFIG_LOG_BUF_SHIFT - PRB_AVGBITS,
493 PRB_AVGBITS, &__log_buf[0]);
495 static struct printk_ringbuffer printk_rb_dynamic;
497 static struct printk_ringbuffer *prb = &printk_rb_static;
500 * We cannot access per-CPU data (e.g. per-CPU flush irq_work) before
501 * per_cpu_areas are initialised. This variable is set to true when
502 * it's safe to access per-CPU data.
504 static bool __printk_percpu_data_ready __ro_after_init;
506 bool printk_percpu_data_ready(void)
508 return __printk_percpu_data_ready;
511 /* Must be called under syslog_lock. */
512 static void latched_seq_write(struct latched_seq *ls, u64 val)
514 raw_write_seqcount_latch(&ls->latch);
516 raw_write_seqcount_latch(&ls->latch);
520 /* Can be called from any context. */
521 static u64 latched_seq_read_nolock(struct latched_seq *ls)
528 seq = raw_read_seqcount_latch(&ls->latch);
531 } while (read_seqcount_latch_retry(&ls->latch, seq));
536 /* Return log buffer address */
537 char *log_buf_addr_get(void)
542 /* Return log buffer size */
543 u32 log_buf_len_get(void)
549 * Define how much of the log buffer we could take at maximum. The value
550 * must be greater than two. Note that only half of the buffer is available
551 * when the index points to the middle.
553 #define MAX_LOG_TAKE_PART 4
554 static const char trunc_msg[] = "<truncated>";
556 static void truncate_msg(u16 *text_len, u16 *trunc_msg_len)
559 * The message should not take the whole buffer. Otherwise, it might
560 * get removed too soon.
562 u32 max_text_len = log_buf_len / MAX_LOG_TAKE_PART;
564 if (*text_len > max_text_len)
565 *text_len = max_text_len;
567 /* enable the warning message (if there is room) */
568 *trunc_msg_len = strlen(trunc_msg);
569 if (*text_len >= *trunc_msg_len)
570 *text_len -= *trunc_msg_len;
575 int dmesg_restrict = IS_ENABLED(CONFIG_SECURITY_DMESG_RESTRICT);
577 static int syslog_action_restricted(int type)
582 * Unless restricted, we allow "read all" and "get buffer size"
585 return type != SYSLOG_ACTION_READ_ALL &&
586 type != SYSLOG_ACTION_SIZE_BUFFER;
589 static int check_syslog_permissions(int type, int source)
592 * If this is from /proc/kmsg and we've already opened it, then we've
593 * already done the capabilities checks at open time.
595 if (source == SYSLOG_FROM_PROC && type != SYSLOG_ACTION_OPEN)
598 if (syslog_action_restricted(type)) {
599 if (capable(CAP_SYSLOG))
602 * For historical reasons, accept CAP_SYS_ADMIN too, with
605 if (capable(CAP_SYS_ADMIN)) {
606 pr_warn_once("%s (%d): Attempt to access syslog with "
607 "CAP_SYS_ADMIN but no CAP_SYSLOG "
609 current->comm, task_pid_nr(current));
615 return security_syslog(type);
618 static void append_char(char **pp, char *e, char c)
624 static ssize_t info_print_ext_header(char *buf, size_t size,
625 struct printk_info *info)
627 u64 ts_usec = info->ts_nsec;
629 #ifdef CONFIG_PRINTK_CALLER
630 u32 id = info->caller_id;
632 snprintf(caller, sizeof(caller), ",caller=%c%u",
633 id & 0x80000000 ? 'C' : 'T', id & ~0x80000000);
638 do_div(ts_usec, 1000);
640 return scnprintf(buf, size, "%u,%llu,%llu,%c%s;",
641 (info->facility << 3) | info->level, info->seq,
642 ts_usec, info->flags & LOG_CONT ? 'c' : '-', caller);
645 static ssize_t msg_add_ext_text(char *buf, size_t size,
646 const char *text, size_t text_len,
649 char *p = buf, *e = buf + size;
652 /* escape non-printable characters */
653 for (i = 0; i < text_len; i++) {
654 unsigned char c = text[i];
656 if (c < ' ' || c >= 127 || c == '\\')
657 p += scnprintf(p, e - p, "\\x%02x", c);
659 append_char(&p, e, c);
661 append_char(&p, e, endc);
666 static ssize_t msg_add_dict_text(char *buf, size_t size,
667 const char *key, const char *val)
669 size_t val_len = strlen(val);
675 len = msg_add_ext_text(buf, size, "", 0, ' '); /* dict prefix */
676 len += msg_add_ext_text(buf + len, size - len, key, strlen(key), '=');
677 len += msg_add_ext_text(buf + len, size - len, val, val_len, '\n');
682 static ssize_t msg_print_ext_body(char *buf, size_t size,
683 char *text, size_t text_len,
684 struct dev_printk_info *dev_info)
688 len = msg_add_ext_text(buf, size, text, text_len, '\n');
693 len += msg_add_dict_text(buf + len, size - len, "SUBSYSTEM",
694 dev_info->subsystem);
695 len += msg_add_dict_text(buf + len, size - len, "DEVICE",
701 static bool printk_get_next_message(struct printk_message *pmsg, u64 seq,
702 bool is_extended, bool may_supress);
704 /* /dev/kmsg - userspace message inject/listen interface */
705 struct devkmsg_user {
707 struct ratelimit_state rs;
709 struct printk_buffers pbufs;
712 static __printf(3, 4) __cold
713 int devkmsg_emit(int facility, int level, const char *fmt, ...)
719 r = vprintk_emit(facility, level, NULL, fmt, args);
725 static ssize_t devkmsg_write(struct kiocb *iocb, struct iov_iter *from)
728 int level = default_message_loglevel;
729 int facility = 1; /* LOG_USER */
730 struct file *file = iocb->ki_filp;
731 struct devkmsg_user *user = file->private_data;
732 size_t len = iov_iter_count(from);
735 if (!user || len > PRINTKRB_RECORD_MAX)
738 /* Ignore when user logging is disabled. */
739 if (devkmsg_log & DEVKMSG_LOG_MASK_OFF)
742 /* Ratelimit when not explicitly enabled. */
743 if (!(devkmsg_log & DEVKMSG_LOG_MASK_ON)) {
744 if (!___ratelimit(&user->rs, current->comm))
748 buf = kmalloc(len+1, GFP_KERNEL);
753 if (!copy_from_iter_full(buf, len, from)) {
759 * Extract and skip the syslog prefix <[0-9]*>. Coming from userspace
760 * the decimal value represents 32bit, the lower 3 bit are the log
761 * level, the rest are the log facility.
763 * If no prefix or no userspace facility is specified, we
764 * enforce LOG_USER, to be able to reliably distinguish
765 * kernel-generated messages from userspace-injected ones.
768 if (line[0] == '<') {
772 u = simple_strtoul(line + 1, &endp, 10);
773 if (endp && endp[0] == '>') {
774 level = LOG_LEVEL(u);
775 if (LOG_FACILITY(u) != 0)
776 facility = LOG_FACILITY(u);
782 devkmsg_emit(facility, level, "%s", line);
787 static ssize_t devkmsg_read(struct file *file, char __user *buf,
788 size_t count, loff_t *ppos)
790 struct devkmsg_user *user = file->private_data;
791 char *outbuf = &user->pbufs.outbuf[0];
792 struct printk_message pmsg = {
793 .pbufs = &user->pbufs,
800 ret = mutex_lock_interruptible(&user->lock);
804 if (!printk_get_next_message(&pmsg, atomic64_read(&user->seq), true, false)) {
805 if (file->f_flags & O_NONBLOCK) {
811 * Guarantee this task is visible on the waitqueue before
812 * checking the wake condition.
814 * The full memory barrier within set_current_state() of
815 * prepare_to_wait_event() pairs with the full memory barrier
816 * within wq_has_sleeper().
818 * This pairs with __wake_up_klogd:A.
820 ret = wait_event_interruptible(log_wait,
821 printk_get_next_message(&pmsg, atomic64_read(&user->seq), true,
822 false)); /* LMM(devkmsg_read:A) */
828 /* our last seen message is gone, return error and reset */
829 atomic64_set(&user->seq, pmsg.seq);
834 atomic64_set(&user->seq, pmsg.seq + 1);
836 if (pmsg.outbuf_len > count) {
841 if (copy_to_user(buf, outbuf, pmsg.outbuf_len)) {
845 ret = pmsg.outbuf_len;
847 mutex_unlock(&user->lock);
852 * Be careful when modifying this function!!!
854 * Only few operations are supported because the device works only with the
855 * entire variable length messages (records). Non-standard values are
856 * returned in the other cases and has been this way for quite some time.
857 * User space applications might depend on this behavior.
859 static loff_t devkmsg_llseek(struct file *file, loff_t offset, int whence)
861 struct devkmsg_user *user = file->private_data;
871 /* the first record */
872 atomic64_set(&user->seq, prb_first_valid_seq(prb));
876 * The first record after the last SYSLOG_ACTION_CLEAR,
877 * like issued by 'dmesg -c'. Reading /dev/kmsg itself
878 * changes no global state, and does not clear anything.
880 atomic64_set(&user->seq, latched_seq_read_nolock(&clear_seq));
883 /* after the last record */
884 atomic64_set(&user->seq, prb_next_seq(prb));
892 static __poll_t devkmsg_poll(struct file *file, poll_table *wait)
894 struct devkmsg_user *user = file->private_data;
895 struct printk_info info;
899 return EPOLLERR|EPOLLNVAL;
901 poll_wait(file, &log_wait, wait);
903 if (prb_read_valid_info(prb, atomic64_read(&user->seq), &info, NULL)) {
904 /* return error when data has vanished underneath us */
905 if (info.seq != atomic64_read(&user->seq))
906 ret = EPOLLIN|EPOLLRDNORM|EPOLLERR|EPOLLPRI;
908 ret = EPOLLIN|EPOLLRDNORM;
914 static int devkmsg_open(struct inode *inode, struct file *file)
916 struct devkmsg_user *user;
919 if (devkmsg_log & DEVKMSG_LOG_MASK_OFF)
922 /* write-only does not need any file context */
923 if ((file->f_flags & O_ACCMODE) != O_WRONLY) {
924 err = check_syslog_permissions(SYSLOG_ACTION_READ_ALL,
930 user = kvmalloc(sizeof(struct devkmsg_user), GFP_KERNEL);
934 ratelimit_default_init(&user->rs);
935 ratelimit_set_flags(&user->rs, RATELIMIT_MSG_ON_RELEASE);
937 mutex_init(&user->lock);
939 atomic64_set(&user->seq, prb_first_valid_seq(prb));
941 file->private_data = user;
945 static int devkmsg_release(struct inode *inode, struct file *file)
947 struct devkmsg_user *user = file->private_data;
952 ratelimit_state_exit(&user->rs);
954 mutex_destroy(&user->lock);
959 const struct file_operations kmsg_fops = {
960 .open = devkmsg_open,
961 .read = devkmsg_read,
962 .write_iter = devkmsg_write,
963 .llseek = devkmsg_llseek,
964 .poll = devkmsg_poll,
965 .release = devkmsg_release,
968 #ifdef CONFIG_CRASH_CORE
970 * This appends the listed symbols to /proc/vmcore
972 * /proc/vmcore is used by various utilities, like crash and makedumpfile to
973 * obtain access to symbols that are otherwise very difficult to locate. These
974 * symbols are specifically used so that utilities can access and extract the
975 * dmesg log from a vmcore file after a crash.
977 void log_buf_vmcoreinfo_setup(void)
979 struct dev_printk_info *dev_info = NULL;
981 VMCOREINFO_SYMBOL(prb);
982 VMCOREINFO_SYMBOL(printk_rb_static);
983 VMCOREINFO_SYMBOL(clear_seq);
986 * Export struct size and field offsets. User space tools can
987 * parse it and detect any changes to structure down the line.
990 VMCOREINFO_STRUCT_SIZE(printk_ringbuffer);
991 VMCOREINFO_OFFSET(printk_ringbuffer, desc_ring);
992 VMCOREINFO_OFFSET(printk_ringbuffer, text_data_ring);
993 VMCOREINFO_OFFSET(printk_ringbuffer, fail);
995 VMCOREINFO_STRUCT_SIZE(prb_desc_ring);
996 VMCOREINFO_OFFSET(prb_desc_ring, count_bits);
997 VMCOREINFO_OFFSET(prb_desc_ring, descs);
998 VMCOREINFO_OFFSET(prb_desc_ring, infos);
999 VMCOREINFO_OFFSET(prb_desc_ring, head_id);
1000 VMCOREINFO_OFFSET(prb_desc_ring, tail_id);
1002 VMCOREINFO_STRUCT_SIZE(prb_desc);
1003 VMCOREINFO_OFFSET(prb_desc, state_var);
1004 VMCOREINFO_OFFSET(prb_desc, text_blk_lpos);
1006 VMCOREINFO_STRUCT_SIZE(prb_data_blk_lpos);
1007 VMCOREINFO_OFFSET(prb_data_blk_lpos, begin);
1008 VMCOREINFO_OFFSET(prb_data_blk_lpos, next);
1010 VMCOREINFO_STRUCT_SIZE(printk_info);
1011 VMCOREINFO_OFFSET(printk_info, seq);
1012 VMCOREINFO_OFFSET(printk_info, ts_nsec);
1013 VMCOREINFO_OFFSET(printk_info, text_len);
1014 VMCOREINFO_OFFSET(printk_info, caller_id);
1015 VMCOREINFO_OFFSET(printk_info, dev_info);
1017 VMCOREINFO_STRUCT_SIZE(dev_printk_info);
1018 VMCOREINFO_OFFSET(dev_printk_info, subsystem);
1019 VMCOREINFO_LENGTH(printk_info_subsystem, sizeof(dev_info->subsystem));
1020 VMCOREINFO_OFFSET(dev_printk_info, device);
1021 VMCOREINFO_LENGTH(printk_info_device, sizeof(dev_info->device));
1023 VMCOREINFO_STRUCT_SIZE(prb_data_ring);
1024 VMCOREINFO_OFFSET(prb_data_ring, size_bits);
1025 VMCOREINFO_OFFSET(prb_data_ring, data);
1026 VMCOREINFO_OFFSET(prb_data_ring, head_lpos);
1027 VMCOREINFO_OFFSET(prb_data_ring, tail_lpos);
1029 VMCOREINFO_SIZE(atomic_long_t);
1030 VMCOREINFO_TYPE_OFFSET(atomic_long_t, counter);
1032 VMCOREINFO_STRUCT_SIZE(latched_seq);
1033 VMCOREINFO_OFFSET(latched_seq, val);
1037 /* requested log_buf_len from kernel cmdline */
1038 static unsigned long __initdata new_log_buf_len;
1040 /* we practice scaling the ring buffer by powers of 2 */
1041 static void __init log_buf_len_update(u64 size)
1043 if (size > (u64)LOG_BUF_LEN_MAX) {
1044 size = (u64)LOG_BUF_LEN_MAX;
1045 pr_err("log_buf over 2G is not supported.\n");
1049 size = roundup_pow_of_two(size);
1050 if (size > log_buf_len)
1051 new_log_buf_len = (unsigned long)size;
1054 /* save requested log_buf_len since it's too early to process it */
1055 static int __init log_buf_len_setup(char *str)
1062 size = memparse(str, &str);
1064 log_buf_len_update(size);
1068 early_param("log_buf_len", log_buf_len_setup);
1071 #define __LOG_CPU_MAX_BUF_LEN (1 << CONFIG_LOG_CPU_MAX_BUF_SHIFT)
1073 static void __init log_buf_add_cpu(void)
1075 unsigned int cpu_extra;
1078 * archs should set up cpu_possible_bits properly with
1079 * set_cpu_possible() after setup_arch() but just in
1080 * case lets ensure this is valid.
1082 if (num_possible_cpus() == 1)
1085 cpu_extra = (num_possible_cpus() - 1) * __LOG_CPU_MAX_BUF_LEN;
1087 /* by default this will only continue through for large > 64 CPUs */
1088 if (cpu_extra <= __LOG_BUF_LEN / 2)
1091 pr_info("log_buf_len individual max cpu contribution: %d bytes\n",
1092 __LOG_CPU_MAX_BUF_LEN);
1093 pr_info("log_buf_len total cpu_extra contributions: %d bytes\n",
1095 pr_info("log_buf_len min size: %d bytes\n", __LOG_BUF_LEN);
1097 log_buf_len_update(cpu_extra + __LOG_BUF_LEN);
1099 #else /* !CONFIG_SMP */
1100 static inline void log_buf_add_cpu(void) {}
1101 #endif /* CONFIG_SMP */
1103 static void __init set_percpu_data_ready(void)
1105 __printk_percpu_data_ready = true;
1108 static unsigned int __init add_to_rb(struct printk_ringbuffer *rb,
1109 struct printk_record *r)
1111 struct prb_reserved_entry e;
1112 struct printk_record dest_r;
1114 prb_rec_init_wr(&dest_r, r->info->text_len);
1116 if (!prb_reserve(&e, rb, &dest_r))
1119 memcpy(&dest_r.text_buf[0], &r->text_buf[0], r->info->text_len);
1120 dest_r.info->text_len = r->info->text_len;
1121 dest_r.info->facility = r->info->facility;
1122 dest_r.info->level = r->info->level;
1123 dest_r.info->flags = r->info->flags;
1124 dest_r.info->ts_nsec = r->info->ts_nsec;
1125 dest_r.info->caller_id = r->info->caller_id;
1126 memcpy(&dest_r.info->dev_info, &r->info->dev_info, sizeof(dest_r.info->dev_info));
1128 prb_final_commit(&e);
1130 return prb_record_text_space(&e);
1133 static char setup_text_buf[PRINTKRB_RECORD_MAX] __initdata;
1135 void __init setup_log_buf(int early)
1137 struct printk_info *new_infos;
1138 unsigned int new_descs_count;
1139 struct prb_desc *new_descs;
1140 struct printk_info info;
1141 struct printk_record r;
1142 unsigned int text_size;
1143 size_t new_descs_size;
1144 size_t new_infos_size;
1145 unsigned long flags;
1151 * Some archs call setup_log_buf() multiple times - first is very
1152 * early, e.g. from setup_arch(), and second - when percpu_areas
1156 set_percpu_data_ready();
1158 if (log_buf != __log_buf)
1161 if (!early && !new_log_buf_len)
1164 if (!new_log_buf_len)
1167 new_descs_count = new_log_buf_len >> PRB_AVGBITS;
1168 if (new_descs_count == 0) {
1169 pr_err("new_log_buf_len: %lu too small\n", new_log_buf_len);
1173 new_log_buf = memblock_alloc(new_log_buf_len, LOG_ALIGN);
1174 if (unlikely(!new_log_buf)) {
1175 pr_err("log_buf_len: %lu text bytes not available\n",
1180 new_descs_size = new_descs_count * sizeof(struct prb_desc);
1181 new_descs = memblock_alloc(new_descs_size, LOG_ALIGN);
1182 if (unlikely(!new_descs)) {
1183 pr_err("log_buf_len: %zu desc bytes not available\n",
1185 goto err_free_log_buf;
1188 new_infos_size = new_descs_count * sizeof(struct printk_info);
1189 new_infos = memblock_alloc(new_infos_size, LOG_ALIGN);
1190 if (unlikely(!new_infos)) {
1191 pr_err("log_buf_len: %zu info bytes not available\n",
1193 goto err_free_descs;
1196 prb_rec_init_rd(&r, &info, &setup_text_buf[0], sizeof(setup_text_buf));
1198 prb_init(&printk_rb_dynamic,
1199 new_log_buf, ilog2(new_log_buf_len),
1200 new_descs, ilog2(new_descs_count),
1203 local_irq_save(flags);
1205 log_buf_len = new_log_buf_len;
1206 log_buf = new_log_buf;
1207 new_log_buf_len = 0;
1209 free = __LOG_BUF_LEN;
1210 prb_for_each_record(0, &printk_rb_static, seq, &r) {
1211 text_size = add_to_rb(&printk_rb_dynamic, &r);
1212 if (text_size > free)
1218 prb = &printk_rb_dynamic;
1220 local_irq_restore(flags);
1223 * Copy any remaining messages that might have appeared from
1224 * NMI context after copying but before switching to the
1227 prb_for_each_record(seq, &printk_rb_static, seq, &r) {
1228 text_size = add_to_rb(&printk_rb_dynamic, &r);
1229 if (text_size > free)
1235 if (seq != prb_next_seq(&printk_rb_static)) {
1236 pr_err("dropped %llu messages\n",
1237 prb_next_seq(&printk_rb_static) - seq);
1240 pr_info("log_buf_len: %u bytes\n", log_buf_len);
1241 pr_info("early log buf free: %u(%u%%)\n",
1242 free, (free * 100) / __LOG_BUF_LEN);
1246 memblock_free(new_descs, new_descs_size);
1248 memblock_free(new_log_buf, new_log_buf_len);
1251 static bool __read_mostly ignore_loglevel;
1253 static int __init ignore_loglevel_setup(char *str)
1255 ignore_loglevel = true;
1256 pr_info("debug: ignoring loglevel setting.\n");
1261 early_param("ignore_loglevel", ignore_loglevel_setup);
1262 module_param(ignore_loglevel, bool, S_IRUGO | S_IWUSR);
1263 MODULE_PARM_DESC(ignore_loglevel,
1264 "ignore loglevel setting (prints all kernel messages to the console)");
1266 static bool suppress_message_printing(int level)
1268 return (level >= console_loglevel && !ignore_loglevel);
1271 #ifdef CONFIG_BOOT_PRINTK_DELAY
1273 static int boot_delay; /* msecs delay after each printk during bootup */
1274 static unsigned long long loops_per_msec; /* based on boot_delay */
1276 static int __init boot_delay_setup(char *str)
1280 lpj = preset_lpj ? preset_lpj : 1000000; /* some guess */
1281 loops_per_msec = (unsigned long long)lpj / 1000 * HZ;
1283 get_option(&str, &boot_delay);
1284 if (boot_delay > 10 * 1000)
1287 pr_debug("boot_delay: %u, preset_lpj: %ld, lpj: %lu, "
1288 "HZ: %d, loops_per_msec: %llu\n",
1289 boot_delay, preset_lpj, lpj, HZ, loops_per_msec);
1292 early_param("boot_delay", boot_delay_setup);
1294 static void boot_delay_msec(int level)
1296 unsigned long long k;
1297 unsigned long timeout;
1299 if ((boot_delay == 0 || system_state >= SYSTEM_RUNNING)
1300 || suppress_message_printing(level)) {
1304 k = (unsigned long long)loops_per_msec * boot_delay;
1306 timeout = jiffies + msecs_to_jiffies(boot_delay);
1311 * use (volatile) jiffies to prevent
1312 * compiler reduction; loop termination via jiffies
1313 * is secondary and may or may not happen.
1315 if (time_after(jiffies, timeout))
1317 touch_nmi_watchdog();
1321 static inline void boot_delay_msec(int level)
1326 static bool printk_time = IS_ENABLED(CONFIG_PRINTK_TIME);
1327 module_param_named(time, printk_time, bool, S_IRUGO | S_IWUSR);
1329 static size_t print_syslog(unsigned int level, char *buf)
1331 return sprintf(buf, "<%u>", level);
1334 static size_t print_time(u64 ts, char *buf)
1336 unsigned long rem_nsec = do_div(ts, 1000000000);
1338 return sprintf(buf, "[%5lu.%06lu]",
1339 (unsigned long)ts, rem_nsec / 1000);
1342 #ifdef CONFIG_PRINTK_CALLER
1343 static size_t print_caller(u32 id, char *buf)
1347 snprintf(caller, sizeof(caller), "%c%u",
1348 id & 0x80000000 ? 'C' : 'T', id & ~0x80000000);
1349 return sprintf(buf, "[%6s]", caller);
1352 #define print_caller(id, buf) 0
1355 static size_t info_print_prefix(const struct printk_info *info, bool syslog,
1356 bool time, char *buf)
1361 len = print_syslog((info->facility << 3) | info->level, buf);
1364 len += print_time(info->ts_nsec, buf + len);
1366 len += print_caller(info->caller_id, buf + len);
1368 if (IS_ENABLED(CONFIG_PRINTK_CALLER) || time) {
1377 * Prepare the record for printing. The text is shifted within the given
1378 * buffer to avoid a need for another one. The following operations are
1381 * - Add prefix for each line.
1382 * - Drop truncated lines that no longer fit into the buffer.
1383 * - Add the trailing newline that has been removed in vprintk_store().
1384 * - Add a string terminator.
1386 * Since the produced string is always terminated, the maximum possible
1387 * return value is @r->text_buf_size - 1;
1389 * Return: The length of the updated/prepared text, including the added
1390 * prefixes and the newline. The terminator is not counted. The dropped
1391 * line(s) are not counted.
1393 static size_t record_print_text(struct printk_record *r, bool syslog,
1396 size_t text_len = r->info->text_len;
1397 size_t buf_size = r->text_buf_size;
1398 char *text = r->text_buf;
1399 char prefix[PRINTK_PREFIX_MAX];
1400 bool truncated = false;
1407 * If the message was truncated because the buffer was not large
1408 * enough, treat the available text as if it were the full text.
1410 if (text_len > buf_size)
1411 text_len = buf_size;
1413 prefix_len = info_print_prefix(r->info, syslog, time, prefix);
1416 * @text_len: bytes of unprocessed text
1417 * @line_len: bytes of current line _without_ newline
1418 * @text: pointer to beginning of current line
1419 * @len: number of bytes prepared in r->text_buf
1422 next = memchr(text, '\n', text_len);
1424 line_len = next - text;
1426 /* Drop truncated line(s). */
1429 line_len = text_len;
1433 * Truncate the text if there is not enough space to add the
1434 * prefix and a trailing newline and a terminator.
1436 if (len + prefix_len + text_len + 1 + 1 > buf_size) {
1437 /* Drop even the current line if no space. */
1438 if (len + prefix_len + line_len + 1 + 1 > buf_size)
1441 text_len = buf_size - len - prefix_len - 1 - 1;
1445 memmove(text + prefix_len, text, text_len);
1446 memcpy(text, prefix, prefix_len);
1449 * Increment the prepared length to include the text and
1450 * prefix that were just moved+copied. Also increment for the
1451 * newline at the end of this line. If this is the last line,
1452 * there is no newline, but it will be added immediately below.
1454 len += prefix_len + line_len + 1;
1455 if (text_len == line_len) {
1457 * This is the last line. Add the trailing newline
1458 * removed in vprintk_store().
1460 text[prefix_len + line_len] = '\n';
1465 * Advance beyond the added prefix and the related line with
1468 text += prefix_len + line_len + 1;
1471 * The remaining text has only decreased by the line with its
1474 * Note that @text_len can become zero. It happens when @text
1475 * ended with a newline (either due to truncation or the
1476 * original string ending with "\n\n"). The loop is correctly
1477 * repeated and (if not truncated) an empty line with a prefix
1480 text_len -= line_len + 1;
1484 * If a buffer was provided, it will be terminated. Space for the
1485 * string terminator is guaranteed to be available. The terminator is
1486 * not counted in the return value.
1489 r->text_buf[len] = 0;
1494 static size_t get_record_print_text_size(struct printk_info *info,
1495 unsigned int line_count,
1496 bool syslog, bool time)
1498 char prefix[PRINTK_PREFIX_MAX];
1501 prefix_len = info_print_prefix(info, syslog, time, prefix);
1504 * Each line will be preceded with a prefix. The intermediate
1505 * newlines are already within the text, but a final trailing
1506 * newline will be added.
1508 return ((prefix_len * line_count) + info->text_len + 1);
1512 * Beginning with @start_seq, find the first record where it and all following
1513 * records up to (but not including) @max_seq fit into @size.
1515 * @max_seq is simply an upper bound and does not need to exist. If the caller
1516 * does not require an upper bound, -1 can be used for @max_seq.
1518 static u64 find_first_fitting_seq(u64 start_seq, u64 max_seq, size_t size,
1519 bool syslog, bool time)
1521 struct printk_info info;
1522 unsigned int line_count;
1526 /* Determine the size of the records up to @max_seq. */
1527 prb_for_each_info(start_seq, prb, seq, &info, &line_count) {
1528 if (info.seq >= max_seq)
1530 len += get_record_print_text_size(&info, line_count, syslog, time);
1534 * Adjust the upper bound for the next loop to avoid subtracting
1535 * lengths that were never added.
1541 * Move first record forward until length fits into the buffer. Ignore
1542 * newest messages that were not counted in the above cycle. Messages
1543 * might appear and get lost in the meantime. This is a best effort
1544 * that prevents an infinite loop that could occur with a retry.
1546 prb_for_each_info(start_seq, prb, seq, &info, &line_count) {
1547 if (len <= size || info.seq >= max_seq)
1549 len -= get_record_print_text_size(&info, line_count, syslog, time);
1555 /* The caller is responsible for making sure @size is greater than 0. */
1556 static int syslog_print(char __user *buf, int size)
1558 struct printk_info info;
1559 struct printk_record r;
1564 text = kmalloc(PRINTK_MESSAGE_MAX, GFP_KERNEL);
1568 prb_rec_init_rd(&r, &info, text, PRINTK_MESSAGE_MAX);
1570 mutex_lock(&syslog_lock);
1573 * Wait for the @syslog_seq record to be available. @syslog_seq may
1574 * change while waiting.
1579 mutex_unlock(&syslog_lock);
1581 * Guarantee this task is visible on the waitqueue before
1582 * checking the wake condition.
1584 * The full memory barrier within set_current_state() of
1585 * prepare_to_wait_event() pairs with the full memory barrier
1586 * within wq_has_sleeper().
1588 * This pairs with __wake_up_klogd:A.
1590 len = wait_event_interruptible(log_wait,
1591 prb_read_valid(prb, seq, NULL)); /* LMM(syslog_print:A) */
1592 mutex_lock(&syslog_lock);
1596 } while (syslog_seq != seq);
1599 * Copy records that fit into the buffer. The above cycle makes sure
1600 * that the first record is always available.
1607 if (!prb_read_valid(prb, syslog_seq, &r))
1610 if (r.info->seq != syslog_seq) {
1611 /* message is gone, move to next valid one */
1612 syslog_seq = r.info->seq;
1617 * To keep reading/counting partial line consistent,
1618 * use printk_time value as of the beginning of a line.
1620 if (!syslog_partial)
1621 syslog_time = printk_time;
1623 skip = syslog_partial;
1624 n = record_print_text(&r, true, syslog_time);
1625 if (n - syslog_partial <= size) {
1626 /* message fits into buffer, move forward */
1627 syslog_seq = r.info->seq + 1;
1628 n -= syslog_partial;
1631 /* partial read(), remember position */
1633 syslog_partial += n;
1640 mutex_unlock(&syslog_lock);
1641 err = copy_to_user(buf, text + skip, n);
1642 mutex_lock(&syslog_lock);
1655 mutex_unlock(&syslog_lock);
1660 static int syslog_print_all(char __user *buf, int size, bool clear)
1662 struct printk_info info;
1663 struct printk_record r;
1669 text = kmalloc(PRINTK_MESSAGE_MAX, GFP_KERNEL);
1675 * Find first record that fits, including all following records,
1676 * into the user-provided buffer for this dump.
1678 seq = find_first_fitting_seq(latched_seq_read_nolock(&clear_seq), -1,
1681 prb_rec_init_rd(&r, &info, text, PRINTK_MESSAGE_MAX);
1684 prb_for_each_record(seq, prb, seq, &r) {
1687 textlen = record_print_text(&r, true, time);
1689 if (len + textlen > size) {
1694 if (copy_to_user(buf + len, text, textlen))
1704 mutex_lock(&syslog_lock);
1705 latched_seq_write(&clear_seq, seq);
1706 mutex_unlock(&syslog_lock);
1713 static void syslog_clear(void)
1715 mutex_lock(&syslog_lock);
1716 latched_seq_write(&clear_seq, prb_next_seq(prb));
1717 mutex_unlock(&syslog_lock);
1720 int do_syslog(int type, char __user *buf, int len, int source)
1722 struct printk_info info;
1724 static int saved_console_loglevel = LOGLEVEL_DEFAULT;
1727 error = check_syslog_permissions(type, source);
1732 case SYSLOG_ACTION_CLOSE: /* Close log */
1734 case SYSLOG_ACTION_OPEN: /* Open log */
1736 case SYSLOG_ACTION_READ: /* Read from log */
1737 if (!buf || len < 0)
1741 if (!access_ok(buf, len))
1743 error = syslog_print(buf, len);
1745 /* Read/clear last kernel messages */
1746 case SYSLOG_ACTION_READ_CLEAR:
1749 /* Read last kernel messages */
1750 case SYSLOG_ACTION_READ_ALL:
1751 if (!buf || len < 0)
1755 if (!access_ok(buf, len))
1757 error = syslog_print_all(buf, len, clear);
1759 /* Clear ring buffer */
1760 case SYSLOG_ACTION_CLEAR:
1763 /* Disable logging to console */
1764 case SYSLOG_ACTION_CONSOLE_OFF:
1765 if (saved_console_loglevel == LOGLEVEL_DEFAULT)
1766 saved_console_loglevel = console_loglevel;
1767 console_loglevel = minimum_console_loglevel;
1769 /* Enable logging to console */
1770 case SYSLOG_ACTION_CONSOLE_ON:
1771 if (saved_console_loglevel != LOGLEVEL_DEFAULT) {
1772 console_loglevel = saved_console_loglevel;
1773 saved_console_loglevel = LOGLEVEL_DEFAULT;
1776 /* Set level of messages printed to console */
1777 case SYSLOG_ACTION_CONSOLE_LEVEL:
1778 if (len < 1 || len > 8)
1780 if (len < minimum_console_loglevel)
1781 len = minimum_console_loglevel;
1782 console_loglevel = len;
1783 /* Implicitly re-enable logging to console */
1784 saved_console_loglevel = LOGLEVEL_DEFAULT;
1786 /* Number of chars in the log buffer */
1787 case SYSLOG_ACTION_SIZE_UNREAD:
1788 mutex_lock(&syslog_lock);
1789 if (!prb_read_valid_info(prb, syslog_seq, &info, NULL)) {
1790 /* No unread messages. */
1791 mutex_unlock(&syslog_lock);
1794 if (info.seq != syslog_seq) {
1795 /* messages are gone, move to first one */
1796 syslog_seq = info.seq;
1799 if (source == SYSLOG_FROM_PROC) {
1801 * Short-cut for poll(/"proc/kmsg") which simply checks
1802 * for pending data, not the size; return the count of
1803 * records, not the length.
1805 error = prb_next_seq(prb) - syslog_seq;
1807 bool time = syslog_partial ? syslog_time : printk_time;
1808 unsigned int line_count;
1811 prb_for_each_info(syslog_seq, prb, seq, &info,
1813 error += get_record_print_text_size(&info, line_count,
1817 error -= syslog_partial;
1819 mutex_unlock(&syslog_lock);
1821 /* Size of the log buffer */
1822 case SYSLOG_ACTION_SIZE_BUFFER:
1823 error = log_buf_len;
1833 SYSCALL_DEFINE3(syslog, int, type, char __user *, buf, int, len)
1835 return do_syslog(type, buf, len, SYSLOG_FROM_READER);
1839 * Special console_lock variants that help to reduce the risk of soft-lockups.
1840 * They allow to pass console_lock to another printk() call using a busy wait.
1843 #ifdef CONFIG_LOCKDEP
1844 static struct lockdep_map console_owner_dep_map = {
1845 .name = "console_owner"
1849 static DEFINE_RAW_SPINLOCK(console_owner_lock);
1850 static struct task_struct *console_owner;
1851 static bool console_waiter;
1854 * console_lock_spinning_enable - mark beginning of code where another
1855 * thread might safely busy wait
1857 * This basically converts console_lock into a spinlock. This marks
1858 * the section where the console_lock owner can not sleep, because
1859 * there may be a waiter spinning (like a spinlock). Also it must be
1860 * ready to hand over the lock at the end of the section.
1862 static void console_lock_spinning_enable(void)
1864 raw_spin_lock(&console_owner_lock);
1865 console_owner = current;
1866 raw_spin_unlock(&console_owner_lock);
1868 /* The waiter may spin on us after setting console_owner */
1869 spin_acquire(&console_owner_dep_map, 0, 0, _THIS_IP_);
1873 * console_lock_spinning_disable_and_check - mark end of code where another
1874 * thread was able to busy wait and check if there is a waiter
1875 * @cookie: cookie returned from console_srcu_read_lock()
1877 * This is called at the end of the section where spinning is allowed.
1878 * It has two functions. First, it is a signal that it is no longer
1879 * safe to start busy waiting for the lock. Second, it checks if
1880 * there is a busy waiter and passes the lock rights to her.
1882 * Important: Callers lose both the console_lock and the SRCU read lock if
1883 * there was a busy waiter. They must not touch items synchronized by
1884 * console_lock or SRCU read lock in this case.
1886 * Return: 1 if the lock rights were passed, 0 otherwise.
1888 static int console_lock_spinning_disable_and_check(int cookie)
1892 raw_spin_lock(&console_owner_lock);
1893 waiter = READ_ONCE(console_waiter);
1894 console_owner = NULL;
1895 raw_spin_unlock(&console_owner_lock);
1898 spin_release(&console_owner_dep_map, _THIS_IP_);
1902 /* The waiter is now free to continue */
1903 WRITE_ONCE(console_waiter, false);
1905 spin_release(&console_owner_dep_map, _THIS_IP_);
1908 * Preserve lockdep lock ordering. Release the SRCU read lock before
1909 * releasing the console_lock.
1911 console_srcu_read_unlock(cookie);
1914 * Hand off console_lock to waiter. The waiter will perform
1915 * the up(). After this, the waiter is the console_lock owner.
1917 mutex_release(&console_lock_dep_map, _THIS_IP_);
1922 * console_trylock_spinning - try to get console_lock by busy waiting
1924 * This allows to busy wait for the console_lock when the current
1925 * owner is running in specially marked sections. It means that
1926 * the current owner is running and cannot reschedule until it
1927 * is ready to lose the lock.
1929 * Return: 1 if we got the lock, 0 othrewise
1931 static int console_trylock_spinning(void)
1933 struct task_struct *owner = NULL;
1936 unsigned long flags;
1938 if (console_trylock())
1942 * It's unsafe to spin once a panic has begun. If we are the
1943 * panic CPU, we may have already halted the owner of the
1944 * console_sem. If we are not the panic CPU, then we should
1945 * avoid taking console_sem, so the panic CPU has a better
1946 * chance of cleanly acquiring it later.
1948 if (panic_in_progress())
1951 printk_safe_enter_irqsave(flags);
1953 raw_spin_lock(&console_owner_lock);
1954 owner = READ_ONCE(console_owner);
1955 waiter = READ_ONCE(console_waiter);
1956 if (!waiter && owner && owner != current) {
1957 WRITE_ONCE(console_waiter, true);
1960 raw_spin_unlock(&console_owner_lock);
1963 * If there is an active printk() writing to the
1964 * consoles, instead of having it write our data too,
1965 * see if we can offload that load from the active
1966 * printer, and do some printing ourselves.
1967 * Go into a spin only if there isn't already a waiter
1968 * spinning, and there is an active printer, and
1969 * that active printer isn't us (recursive printk?).
1972 printk_safe_exit_irqrestore(flags);
1976 /* We spin waiting for the owner to release us */
1977 spin_acquire(&console_owner_dep_map, 0, 0, _THIS_IP_);
1978 /* Owner will clear console_waiter on hand off */
1979 while (READ_ONCE(console_waiter))
1981 spin_release(&console_owner_dep_map, _THIS_IP_);
1983 printk_safe_exit_irqrestore(flags);
1985 * The owner passed the console lock to us.
1986 * Since we did not spin on console lock, annotate
1987 * this as a trylock. Otherwise lockdep will
1990 mutex_acquire(&console_lock_dep_map, 0, 1, _THIS_IP_);
1996 * Recursion is tracked separately on each CPU. If NMIs are supported, an
1997 * additional NMI context per CPU is also separately tracked. Until per-CPU
1998 * is available, a separate "early tracking" is performed.
2000 static DEFINE_PER_CPU(u8, printk_count);
2001 static u8 printk_count_early;
2002 #ifdef CONFIG_HAVE_NMI
2003 static DEFINE_PER_CPU(u8, printk_count_nmi);
2004 static u8 printk_count_nmi_early;
2008 * Recursion is limited to keep the output sane. printk() should not require
2009 * more than 1 level of recursion (allowing, for example, printk() to trigger
2010 * a WARN), but a higher value is used in case some printk-internal errors
2011 * exist, such as the ringbuffer validation checks failing.
2013 #define PRINTK_MAX_RECURSION 3
2016 * Return a pointer to the dedicated counter for the CPU+context of the
2019 static u8 *__printk_recursion_counter(void)
2021 #ifdef CONFIG_HAVE_NMI
2023 if (printk_percpu_data_ready())
2024 return this_cpu_ptr(&printk_count_nmi);
2025 return &printk_count_nmi_early;
2028 if (printk_percpu_data_ready())
2029 return this_cpu_ptr(&printk_count);
2030 return &printk_count_early;
2034 * Enter recursion tracking. Interrupts are disabled to simplify tracking.
2035 * The caller must check the boolean return value to see if the recursion is
2036 * allowed. On failure, interrupts are not disabled.
2038 * @recursion_ptr must be a variable of type (u8 *) and is the same variable
2039 * that is passed to printk_exit_irqrestore().
2041 #define printk_enter_irqsave(recursion_ptr, flags) \
2043 bool success = true; \
2045 typecheck(u8 *, recursion_ptr); \
2046 local_irq_save(flags); \
2047 (recursion_ptr) = __printk_recursion_counter(); \
2048 if (*(recursion_ptr) > PRINTK_MAX_RECURSION) { \
2049 local_irq_restore(flags); \
2052 (*(recursion_ptr))++; \
2057 /* Exit recursion tracking, restoring interrupts. */
2058 #define printk_exit_irqrestore(recursion_ptr, flags) \
2060 typecheck(u8 *, recursion_ptr); \
2061 (*(recursion_ptr))--; \
2062 local_irq_restore(flags); \
2065 int printk_delay_msec __read_mostly;
2067 static inline void printk_delay(int level)
2069 boot_delay_msec(level);
2071 if (unlikely(printk_delay_msec)) {
2072 int m = printk_delay_msec;
2076 touch_nmi_watchdog();
2081 static inline u32 printk_caller_id(void)
2083 return in_task() ? task_pid_nr(current) :
2084 0x80000000 + smp_processor_id();
2088 * printk_parse_prefix - Parse level and control flags.
2090 * @text: The terminated text message.
2091 * @level: A pointer to the current level value, will be updated.
2092 * @flags: A pointer to the current printk_info flags, will be updated.
2094 * @level may be NULL if the caller is not interested in the parsed value.
2095 * Otherwise the variable pointed to by @level must be set to
2096 * LOGLEVEL_DEFAULT in order to be updated with the parsed value.
2098 * @flags may be NULL if the caller is not interested in the parsed value.
2099 * Otherwise the variable pointed to by @flags will be OR'd with the parsed
2102 * Return: The length of the parsed level and control flags.
2104 u16 printk_parse_prefix(const char *text, int *level,
2105 enum printk_info_flags *flags)
2111 kern_level = printk_get_level(text);
2115 switch (kern_level) {
2117 if (level && *level == LOGLEVEL_DEFAULT)
2118 *level = kern_level - '0';
2120 case 'c': /* KERN_CONT */
2133 static u16 printk_sprint(char *text, u16 size, int facility,
2134 enum printk_info_flags *flags, const char *fmt,
2139 text_len = vscnprintf(text, size, fmt, args);
2141 /* Mark and strip a trailing newline. */
2142 if (text_len && text[text_len - 1] == '\n') {
2144 *flags |= LOG_NEWLINE;
2147 /* Strip log level and control flags. */
2148 if (facility == 0) {
2151 prefix_len = printk_parse_prefix(text, NULL, NULL);
2153 text_len -= prefix_len;
2154 memmove(text, text + prefix_len, text_len);
2158 trace_console(text, text_len);
2164 int vprintk_store(int facility, int level,
2165 const struct dev_printk_info *dev_info,
2166 const char *fmt, va_list args)
2168 struct prb_reserved_entry e;
2169 enum printk_info_flags flags = 0;
2170 struct printk_record r;
2171 unsigned long irqflags;
2172 u16 trunc_msg_len = 0;
2182 if (!printk_enter_irqsave(recursion_ptr, irqflags))
2186 * Since the duration of printk() can vary depending on the message
2187 * and state of the ringbuffer, grab the timestamp now so that it is
2188 * close to the call of printk(). This provides a more deterministic
2189 * timestamp with respect to the caller.
2191 ts_nsec = local_clock();
2193 caller_id = printk_caller_id();
2196 * The sprintf needs to come first since the syslog prefix might be
2197 * passed in as a parameter. An extra byte must be reserved so that
2198 * later the vscnprintf() into the reserved buffer has room for the
2199 * terminating '\0', which is not counted by vsnprintf().
2201 va_copy(args2, args);
2202 reserve_size = vsnprintf(&prefix_buf[0], sizeof(prefix_buf), fmt, args2) + 1;
2205 if (reserve_size > PRINTKRB_RECORD_MAX)
2206 reserve_size = PRINTKRB_RECORD_MAX;
2208 /* Extract log level or control flags. */
2210 printk_parse_prefix(&prefix_buf[0], &level, &flags);
2212 if (level == LOGLEVEL_DEFAULT)
2213 level = default_message_loglevel;
2216 flags |= LOG_NEWLINE;
2218 if (flags & LOG_CONT) {
2219 prb_rec_init_wr(&r, reserve_size);
2220 if (prb_reserve_in_last(&e, prb, &r, caller_id, PRINTKRB_RECORD_MAX)) {
2221 text_len = printk_sprint(&r.text_buf[r.info->text_len], reserve_size,
2222 facility, &flags, fmt, args);
2223 r.info->text_len += text_len;
2225 if (flags & LOG_NEWLINE) {
2226 r.info->flags |= LOG_NEWLINE;
2227 prb_final_commit(&e);
2238 * Explicitly initialize the record before every prb_reserve() call.
2239 * prb_reserve_in_last() and prb_reserve() purposely invalidate the
2240 * structure when they fail.
2242 prb_rec_init_wr(&r, reserve_size);
2243 if (!prb_reserve(&e, prb, &r)) {
2244 /* truncate the message if it is too long for empty buffer */
2245 truncate_msg(&reserve_size, &trunc_msg_len);
2247 prb_rec_init_wr(&r, reserve_size + trunc_msg_len);
2248 if (!prb_reserve(&e, prb, &r))
2253 text_len = printk_sprint(&r.text_buf[0], reserve_size, facility, &flags, fmt, args);
2255 memcpy(&r.text_buf[text_len], trunc_msg, trunc_msg_len);
2256 r.info->text_len = text_len + trunc_msg_len;
2257 r.info->facility = facility;
2258 r.info->level = level & 7;
2259 r.info->flags = flags & 0x1f;
2260 r.info->ts_nsec = ts_nsec;
2261 r.info->caller_id = caller_id;
2263 memcpy(&r.info->dev_info, dev_info, sizeof(r.info->dev_info));
2265 /* A message without a trailing newline can be continued. */
2266 if (!(flags & LOG_NEWLINE))
2269 prb_final_commit(&e);
2271 ret = text_len + trunc_msg_len;
2273 printk_exit_irqrestore(recursion_ptr, irqflags);
2277 asmlinkage int vprintk_emit(int facility, int level,
2278 const struct dev_printk_info *dev_info,
2279 const char *fmt, va_list args)
2282 bool in_sched = false;
2284 /* Suppress unimportant messages after panic happens */
2285 if (unlikely(suppress_printk))
2288 if (unlikely(suppress_panic_printk) &&
2289 atomic_read(&panic_cpu) != raw_smp_processor_id())
2292 if (level == LOGLEVEL_SCHED) {
2293 level = LOGLEVEL_DEFAULT;
2297 printk_delay(level);
2299 printed_len = vprintk_store(facility, level, dev_info, fmt, args);
2301 /* If called from the scheduler, we can not call up(). */
2304 * The caller may be holding system-critical or
2305 * timing-sensitive locks. Disable preemption during
2306 * printing of all remaining records to all consoles so that
2307 * this context can return as soon as possible. Hopefully
2308 * another printk() caller will take over the printing.
2312 * Try to acquire and then immediately release the console
2313 * semaphore. The release will print out buffers. With the
2314 * spinning variant, this context tries to take over the
2315 * printing from another printing context.
2317 if (console_trylock_spinning())
2325 EXPORT_SYMBOL(vprintk_emit);
2327 int vprintk_default(const char *fmt, va_list args)
2329 return vprintk_emit(0, LOGLEVEL_DEFAULT, NULL, fmt, args);
2331 EXPORT_SYMBOL_GPL(vprintk_default);
2333 asmlinkage __visible int _printk(const char *fmt, ...)
2338 va_start(args, fmt);
2339 r = vprintk(fmt, args);
2344 EXPORT_SYMBOL(_printk);
2346 static bool pr_flush(int timeout_ms, bool reset_on_progress);
2347 static bool __pr_flush(struct console *con, int timeout_ms, bool reset_on_progress);
2349 #else /* CONFIG_PRINTK */
2351 #define printk_time false
2353 #define prb_read_valid(rb, seq, r) false
2354 #define prb_first_valid_seq(rb) 0
2355 #define prb_next_seq(rb) 0
2357 static u64 syslog_seq;
2359 static size_t record_print_text(const struct printk_record *r,
2360 bool syslog, bool time)
2364 static ssize_t info_print_ext_header(char *buf, size_t size,
2365 struct printk_info *info)
2369 static ssize_t msg_print_ext_body(char *buf, size_t size,
2370 char *text, size_t text_len,
2371 struct dev_printk_info *dev_info) { return 0; }
2372 static void console_lock_spinning_enable(void) { }
2373 static int console_lock_spinning_disable_and_check(int cookie) { return 0; }
2374 static bool suppress_message_printing(int level) { return false; }
2375 static bool pr_flush(int timeout_ms, bool reset_on_progress) { return true; }
2376 static bool __pr_flush(struct console *con, int timeout_ms, bool reset_on_progress) { return true; }
2378 #endif /* CONFIG_PRINTK */
2380 #ifdef CONFIG_EARLY_PRINTK
2381 struct console *early_console;
2383 asmlinkage __visible void early_printk(const char *fmt, ...)
2393 n = vscnprintf(buf, sizeof(buf), fmt, ap);
2396 early_console->write(early_console, buf, n);
2400 static void set_user_specified(struct console_cmdline *c, bool user_specified)
2402 if (!user_specified)
2406 * @c console was defined by the user on the command line.
2407 * Do not clear when added twice also by SPCR or the device tree.
2409 c->user_specified = true;
2410 /* At least one console defined by the user on the command line. */
2411 console_set_on_cmdline = 1;
2414 static int __add_preferred_console(char *name, int idx, char *options,
2415 char *brl_options, bool user_specified)
2417 struct console_cmdline *c;
2421 * See if this tty is not yet registered, and
2422 * if we have a slot free.
2424 for (i = 0, c = console_cmdline;
2425 i < MAX_CMDLINECONSOLES && c->name[0];
2427 if (strcmp(c->name, name) == 0 && c->index == idx) {
2429 preferred_console = i;
2430 set_user_specified(c, user_specified);
2434 if (i == MAX_CMDLINECONSOLES)
2437 preferred_console = i;
2438 strscpy(c->name, name, sizeof(c->name));
2439 c->options = options;
2440 set_user_specified(c, user_specified);
2441 braille_set_options(c, brl_options);
2447 static int __init console_msg_format_setup(char *str)
2449 if (!strcmp(str, "syslog"))
2450 console_msg_format = MSG_FORMAT_SYSLOG;
2451 if (!strcmp(str, "default"))
2452 console_msg_format = MSG_FORMAT_DEFAULT;
2455 __setup("console_msg_format=", console_msg_format_setup);
2458 * Set up a console. Called via do_early_param() in init/main.c
2459 * for each "console=" parameter in the boot command line.
2461 static int __init console_setup(char *str)
2463 char buf[sizeof(console_cmdline[0].name) + 4]; /* 4 for "ttyS" */
2464 char *s, *options, *brl_options = NULL;
2468 * console="" or console=null have been suggested as a way to
2469 * disable console output. Use ttynull that has been created
2470 * for exactly this purpose.
2472 if (str[0] == 0 || strcmp(str, "null") == 0) {
2473 __add_preferred_console("ttynull", 0, NULL, NULL, true);
2477 if (_braille_console_setup(&str, &brl_options))
2481 * Decode str into name, index, options.
2483 if (str[0] >= '0' && str[0] <= '9') {
2484 strcpy(buf, "ttyS");
2485 strncpy(buf + 4, str, sizeof(buf) - 5);
2487 strncpy(buf, str, sizeof(buf) - 1);
2489 buf[sizeof(buf) - 1] = 0;
2490 options = strchr(str, ',');
2494 if (!strcmp(str, "ttya"))
2495 strcpy(buf, "ttyS0");
2496 if (!strcmp(str, "ttyb"))
2497 strcpy(buf, "ttyS1");
2499 for (s = buf; *s; s++)
2500 if (isdigit(*s) || *s == ',')
2502 idx = simple_strtoul(s, NULL, 10);
2505 __add_preferred_console(buf, idx, options, brl_options, true);
2508 __setup("console=", console_setup);
2511 * add_preferred_console - add a device to the list of preferred consoles.
2512 * @name: device name
2513 * @idx: device index
2514 * @options: options for this console
2516 * The last preferred console added will be used for kernel messages
2517 * and stdin/out/err for init. Normally this is used by console_setup
2518 * above to handle user-supplied console arguments; however it can also
2519 * be used by arch-specific code either to override the user or more
2520 * commonly to provide a default console (ie from PROM variables) when
2521 * the user has not supplied one.
2523 int add_preferred_console(char *name, int idx, char *options)
2525 return __add_preferred_console(name, idx, options, NULL, false);
2528 bool console_suspend_enabled = true;
2529 EXPORT_SYMBOL(console_suspend_enabled);
2531 static int __init console_suspend_disable(char *str)
2533 console_suspend_enabled = false;
2536 __setup("no_console_suspend", console_suspend_disable);
2537 module_param_named(console_suspend, console_suspend_enabled,
2538 bool, S_IRUGO | S_IWUSR);
2539 MODULE_PARM_DESC(console_suspend, "suspend console during suspend"
2540 " and hibernate operations");
2542 static bool printk_console_no_auto_verbose;
2544 void console_verbose(void)
2546 if (console_loglevel && !printk_console_no_auto_verbose)
2547 console_loglevel = CONSOLE_LOGLEVEL_MOTORMOUTH;
2549 EXPORT_SYMBOL_GPL(console_verbose);
2551 module_param_named(console_no_auto_verbose, printk_console_no_auto_verbose, bool, 0644);
2552 MODULE_PARM_DESC(console_no_auto_verbose, "Disable console loglevel raise to highest on oops/panic/etc");
2555 * suspend_console - suspend the console subsystem
2557 * This disables printk() while we go into suspend states
2559 void suspend_console(void)
2561 if (!console_suspend_enabled)
2563 pr_info("Suspending console(s) (use no_console_suspend to debug)\n");
2564 pr_flush(1000, true);
2566 console_suspended = 1;
2570 void resume_console(void)
2572 if (!console_suspend_enabled)
2575 console_suspended = 0;
2577 pr_flush(1000, true);
2581 * console_cpu_notify - print deferred console messages after CPU hotplug
2584 * If printk() is called from a CPU that is not online yet, the messages
2585 * will be printed on the console only if there are CON_ANYTIME consoles.
2586 * This function is called when a new CPU comes online (or fails to come
2587 * up) or goes offline.
2589 static int console_cpu_notify(unsigned int cpu)
2591 if (!cpuhp_tasks_frozen) {
2592 /* If trylock fails, someone else is doing the printing */
2593 if (console_trylock())
2600 * console_lock - block the console subsystem from printing
2602 * Acquires a lock which guarantees that no consoles will
2603 * be in or enter their write() callback.
2605 * Can sleep, returns nothing.
2607 void console_lock(void)
2612 if (console_suspended)
2615 console_may_schedule = 1;
2617 EXPORT_SYMBOL(console_lock);
2620 * console_trylock - try to block the console subsystem from printing
2622 * Try to acquire a lock which guarantees that no consoles will
2623 * be in or enter their write() callback.
2625 * returns 1 on success, and 0 on failure to acquire the lock.
2627 int console_trylock(void)
2629 if (down_trylock_console_sem())
2631 if (console_suspended) {
2636 console_may_schedule = 0;
2639 EXPORT_SYMBOL(console_trylock);
2641 int is_console_locked(void)
2643 return console_locked;
2645 EXPORT_SYMBOL(is_console_locked);
2648 * Return true when this CPU should unlock console_sem without pushing all
2649 * messages to the console. This reduces the chance that the console is
2650 * locked when the panic CPU tries to use it.
2652 static bool abandon_console_lock_in_panic(void)
2654 if (!panic_in_progress())
2658 * We can use raw_smp_processor_id() here because it is impossible for
2659 * the task to be migrated to the panic_cpu, or away from it. If
2660 * panic_cpu has already been set, and we're not currently executing on
2661 * that CPU, then we never will be.
2663 return atomic_read(&panic_cpu) != raw_smp_processor_id();
2667 * Check if the given console is currently capable and allowed to print
2670 * Requires the console_srcu_read_lock.
2672 static inline bool console_is_usable(struct console *con)
2674 short flags = console_srcu_read_flags(con);
2676 if (!(flags & CON_ENABLED))
2683 * Console drivers may assume that per-cpu resources have been
2684 * allocated. So unless they're explicitly marked as being able to
2685 * cope (CON_ANYTIME) don't call them until this CPU is officially up.
2687 if (!cpu_online(raw_smp_processor_id()) && !(flags & CON_ANYTIME))
2693 static void __console_unlock(void)
2700 * Prepend the message in @pmsg->pbufs->outbuf with a "dropped message". This
2701 * is achieved by shifting the existing message over and inserting the dropped
2704 * @pmsg is the printk message to prepend.
2706 * @dropped is the dropped count to report in the dropped message.
2708 * If the message text in @pmsg->pbufs->outbuf does not have enough space for
2709 * the dropped message, the message text will be sufficiently truncated.
2711 * If @pmsg->pbufs->outbuf is modified, @pmsg->outbuf_len is updated.
2713 #ifdef CONFIG_PRINTK
2714 static void console_prepend_dropped(struct printk_message *pmsg, unsigned long dropped)
2716 struct printk_buffers *pbufs = pmsg->pbufs;
2717 const size_t scratchbuf_sz = sizeof(pbufs->scratchbuf);
2718 const size_t outbuf_sz = sizeof(pbufs->outbuf);
2719 char *scratchbuf = &pbufs->scratchbuf[0];
2720 char *outbuf = &pbufs->outbuf[0];
2723 len = scnprintf(scratchbuf, scratchbuf_sz,
2724 "** %lu printk messages dropped **\n", dropped);
2727 * Make sure outbuf is sufficiently large before prepending.
2728 * Keep at least the prefix when the message must be truncated.
2729 * It is a rather theoretical problem when someone tries to
2730 * use a minimalist buffer.
2732 if (WARN_ON_ONCE(len + PRINTK_PREFIX_MAX >= outbuf_sz))
2735 if (pmsg->outbuf_len + len >= outbuf_sz) {
2736 /* Truncate the message, but keep it terminated. */
2737 pmsg->outbuf_len = outbuf_sz - (len + 1);
2738 outbuf[pmsg->outbuf_len] = 0;
2741 memmove(outbuf + len, outbuf, pmsg->outbuf_len + 1);
2742 memcpy(outbuf, scratchbuf, len);
2743 pmsg->outbuf_len += len;
2746 #define console_prepend_dropped(pmsg, dropped)
2747 #endif /* CONFIG_PRINTK */
2750 * Read and format the specified record (or a later record if the specified
2751 * record is not available).
2753 * @pmsg will contain the formatted result. @pmsg->pbufs must point to a
2754 * struct printk_buffers.
2756 * @seq is the record to read and format. If it is not available, the next
2757 * valid record is read.
2759 * @is_extended specifies if the message should be formatted for extended
2762 * @may_supress specifies if records may be skipped based on loglevel.
2764 * Returns false if no record is available. Otherwise true and all fields
2765 * of @pmsg are valid. (See the documentation of struct printk_message
2766 * for information about the @pmsg fields.)
2768 static bool printk_get_next_message(struct printk_message *pmsg, u64 seq,
2769 bool is_extended, bool may_suppress)
2771 static int panic_console_dropped;
2773 struct printk_buffers *pbufs = pmsg->pbufs;
2774 const size_t scratchbuf_sz = sizeof(pbufs->scratchbuf);
2775 const size_t outbuf_sz = sizeof(pbufs->outbuf);
2776 char *scratchbuf = &pbufs->scratchbuf[0];
2777 char *outbuf = &pbufs->outbuf[0];
2778 struct printk_info info;
2779 struct printk_record r;
2783 * Formatting extended messages requires a separate buffer, so use the
2784 * scratch buffer to read in the ringbuffer text.
2786 * Formatting normal messages is done in-place, so read the ringbuffer
2787 * text directly into the output buffer.
2790 prb_rec_init_rd(&r, &info, scratchbuf, scratchbuf_sz);
2792 prb_rec_init_rd(&r, &info, outbuf, outbuf_sz);
2794 if (!prb_read_valid(prb, seq, &r))
2797 pmsg->seq = r.info->seq;
2798 pmsg->dropped = r.info->seq - seq;
2801 * Check for dropped messages in panic here so that printk
2802 * suppression can occur as early as possible if necessary.
2804 if (pmsg->dropped &&
2805 panic_in_progress() &&
2806 panic_console_dropped++ > 10) {
2807 suppress_panic_printk = 1;
2808 pr_warn_once("Too many dropped messages. Suppress messages on non-panic CPUs to prevent livelock.\n");
2811 /* Skip record that has level above the console loglevel. */
2812 if (may_suppress && suppress_message_printing(r.info->level))
2816 len = info_print_ext_header(outbuf, outbuf_sz, r.info);
2817 len += msg_print_ext_body(outbuf + len, outbuf_sz - len,
2818 &r.text_buf[0], r.info->text_len, &r.info->dev_info);
2820 len = record_print_text(&r, console_msg_format & MSG_FORMAT_SYSLOG, printk_time);
2823 pmsg->outbuf_len = len;
2828 * Print one record for the given console. The record printed is whatever
2829 * record is the next available record for the given console.
2831 * @handover will be set to true if a printk waiter has taken over the
2832 * console_lock, in which case the caller is no longer holding both the
2833 * console_lock and the SRCU read lock. Otherwise it is set to false.
2835 * @cookie is the cookie from the SRCU read lock.
2837 * Returns false if the given console has no next record to print, otherwise
2840 * Requires the console_lock and the SRCU read lock.
2842 static bool console_emit_next_record(struct console *con, bool *handover, int cookie)
2844 static struct printk_buffers pbufs;
2846 bool is_extended = console_srcu_read_flags(con) & CON_EXTENDED;
2847 char *outbuf = &pbufs.outbuf[0];
2848 struct printk_message pmsg = {
2851 unsigned long flags;
2855 if (!printk_get_next_message(&pmsg, con->seq, is_extended, true))
2858 con->dropped += pmsg.dropped;
2860 /* Skip messages of formatted length 0. */
2861 if (pmsg.outbuf_len == 0) {
2862 con->seq = pmsg.seq + 1;
2866 if (con->dropped && !is_extended) {
2867 console_prepend_dropped(&pmsg, con->dropped);
2872 * While actively printing out messages, if another printk()
2873 * were to occur on another CPU, it may wait for this one to
2874 * finish. This task can not be preempted if there is a
2875 * waiter waiting to take over.
2877 * Interrupts are disabled because the hand over to a waiter
2878 * must not be interrupted until the hand over is completed
2879 * (@console_waiter is cleared).
2881 printk_safe_enter_irqsave(flags);
2882 console_lock_spinning_enable();
2884 /* Do not trace print latency. */
2885 stop_critical_timings();
2887 /* Write everything out to the hardware. */
2888 con->write(con, outbuf, pmsg.outbuf_len);
2890 start_critical_timings();
2892 con->seq = pmsg.seq + 1;
2894 *handover = console_lock_spinning_disable_and_check(cookie);
2895 printk_safe_exit_irqrestore(flags);
2901 * Print out all remaining records to all consoles.
2903 * @do_cond_resched is set by the caller. It can be true only in schedulable
2906 * @next_seq is set to the sequence number after the last available record.
2907 * The value is valid only when this function returns true. It means that all
2908 * usable consoles are completely flushed.
2910 * @handover will be set to true if a printk waiter has taken over the
2911 * console_lock, in which case the caller is no longer holding the
2912 * console_lock. Otherwise it is set to false.
2914 * Returns true when there was at least one usable console and all messages
2915 * were flushed to all usable consoles. A returned false informs the caller
2916 * that everything was not flushed (either there were no usable consoles or
2917 * another context has taken over printing or it is a panic situation and this
2918 * is not the panic CPU). Regardless the reason, the caller should assume it
2919 * is not useful to immediately try again.
2921 * Requires the console_lock.
2923 static bool console_flush_all(bool do_cond_resched, u64 *next_seq, bool *handover)
2925 bool any_usable = false;
2926 struct console *con;
2934 any_progress = false;
2936 cookie = console_srcu_read_lock();
2937 for_each_console_srcu(con) {
2940 if (!console_is_usable(con))
2944 progress = console_emit_next_record(con, handover, cookie);
2947 * If a handover has occurred, the SRCU read lock
2948 * is already released.
2953 /* Track the next of the highest seq flushed. */
2954 if (con->seq > *next_seq)
2955 *next_seq = con->seq;
2959 any_progress = true;
2961 /* Allow panic_cpu to take over the consoles safely. */
2962 if (abandon_console_lock_in_panic())
2965 if (do_cond_resched)
2968 console_srcu_read_unlock(cookie);
2969 } while (any_progress);
2974 console_srcu_read_unlock(cookie);
2979 * console_unlock - unblock the console subsystem from printing
2981 * Releases the console_lock which the caller holds to block printing of
2982 * the console subsystem.
2984 * While the console_lock was held, console output may have been buffered
2985 * by printk(). If this is the case, console_unlock(); emits
2986 * the output prior to releasing the lock.
2988 * console_unlock(); may be called from any context.
2990 void console_unlock(void)
2992 bool do_cond_resched;
2997 if (console_suspended) {
3003 * Console drivers are called with interrupts disabled, so
3004 * @console_may_schedule should be cleared before; however, we may
3005 * end up dumping a lot of lines, for example, if called from
3006 * console registration path, and should invoke cond_resched()
3007 * between lines if allowable. Not doing so can cause a very long
3008 * scheduling stall on a slow console leading to RCU stall and
3009 * softlockup warnings which exacerbate the issue with more
3010 * messages practically incapacitating the system. Therefore, create
3011 * a local to use for the printing loop.
3013 do_cond_resched = console_may_schedule;
3016 console_may_schedule = 0;
3018 flushed = console_flush_all(do_cond_resched, &next_seq, &handover);
3023 * Abort if there was a failure to flush all messages to all
3024 * usable consoles. Either it is not possible to flush (in
3025 * which case it would be an infinite loop of retrying) or
3026 * another context has taken over printing.
3032 * Some context may have added new records after
3033 * console_flush_all() but before unlocking the console.
3034 * Re-check if there is a new record to flush. If the trylock
3035 * fails, another context is already handling the printing.
3037 } while (prb_read_valid(prb, next_seq, NULL) && console_trylock());
3039 EXPORT_SYMBOL(console_unlock);
3042 * console_conditional_schedule - yield the CPU if required
3044 * If the console code is currently allowed to sleep, and
3045 * if this CPU should yield the CPU to another task, do
3048 * Must be called within console_lock();.
3050 void __sched console_conditional_schedule(void)
3052 if (console_may_schedule)
3055 EXPORT_SYMBOL(console_conditional_schedule);
3057 void console_unblank(void)
3063 * Stop console printing because the unblank() callback may
3064 * assume the console is not within its write() callback.
3066 * If @oops_in_progress is set, this may be an atomic context.
3067 * In that case, attempt a trylock as best-effort.
3069 if (oops_in_progress) {
3070 if (down_trylock_console_sem() != 0)
3076 console_may_schedule = 0;
3078 cookie = console_srcu_read_lock();
3079 for_each_console_srcu(c) {
3080 if ((console_srcu_read_flags(c) & CON_ENABLED) && c->unblank)
3083 console_srcu_read_unlock(cookie);
3087 if (!oops_in_progress)
3088 pr_flush(1000, true);
3092 * console_flush_on_panic - flush console content on panic
3093 * @mode: flush all messages in buffer or just the pending ones
3095 * Immediately output all pending messages no matter what.
3097 void console_flush_on_panic(enum con_flush_mode mode)
3100 * If someone else is holding the console lock, trylock will fail
3101 * and may_schedule may be set. Ignore and proceed to unlock so
3102 * that messages are flushed out. As this can be called from any
3103 * context and we don't want to get preempted while flushing,
3104 * ensure may_schedule is cleared.
3107 console_may_schedule = 0;
3109 if (mode == CONSOLE_REPLAY_ALL) {
3114 seq = prb_first_valid_seq(prb);
3116 cookie = console_srcu_read_lock();
3117 for_each_console_srcu(c) {
3119 * If the above console_trylock() failed, this is an
3120 * unsynchronized assignment. But in that case, the
3121 * kernel is in "hope and pray" mode anyway.
3125 console_srcu_read_unlock(cookie);
3131 * Return the console tty driver structure and its associated index
3133 struct tty_driver *console_device(int *index)
3136 struct tty_driver *driver = NULL;
3140 * Take console_lock to serialize device() callback with
3141 * other console operations. For example, fg_console is
3142 * modified under console_lock when switching vt.
3146 cookie = console_srcu_read_lock();
3147 for_each_console_srcu(c) {
3150 driver = c->device(c, index);
3154 console_srcu_read_unlock(cookie);
3161 * Prevent further output on the passed console device so that (for example)
3162 * serial drivers can disable console output before suspending a port, and can
3163 * re-enable output afterwards.
3165 void console_stop(struct console *console)
3167 __pr_flush(console, 1000, true);
3168 console_list_lock();
3169 console_srcu_write_flags(console, console->flags & ~CON_ENABLED);
3170 console_list_unlock();
3173 * Ensure that all SRCU list walks have completed. All contexts must
3174 * be able to see that this console is disabled so that (for example)
3175 * the caller can suspend the port without risk of another context
3178 synchronize_srcu(&console_srcu);
3180 EXPORT_SYMBOL(console_stop);
3182 void console_start(struct console *console)
3184 console_list_lock();
3185 console_srcu_write_flags(console, console->flags | CON_ENABLED);
3186 console_list_unlock();
3187 __pr_flush(console, 1000, true);
3189 EXPORT_SYMBOL(console_start);
3191 static int __read_mostly keep_bootcon;
3193 static int __init keep_bootcon_setup(char *str)
3196 pr_info("debug: skip boot console de-registration.\n");
3201 early_param("keep_bootcon", keep_bootcon_setup);
3204 * This is called by register_console() to try to match
3205 * the newly registered console with any of the ones selected
3206 * by either the command line or add_preferred_console() and
3209 * Care need to be taken with consoles that are statically
3210 * enabled such as netconsole
3212 static int try_enable_preferred_console(struct console *newcon,
3213 bool user_specified)
3215 struct console_cmdline *c;
3218 for (i = 0, c = console_cmdline;
3219 i < MAX_CMDLINECONSOLES && c->name[0];
3221 if (c->user_specified != user_specified)
3223 if (!newcon->match ||
3224 newcon->match(newcon, c->name, c->index, c->options) != 0) {
3225 /* default matching */
3226 BUILD_BUG_ON(sizeof(c->name) != sizeof(newcon->name));
3227 if (strcmp(c->name, newcon->name) != 0)
3229 if (newcon->index >= 0 &&
3230 newcon->index != c->index)
3232 if (newcon->index < 0)
3233 newcon->index = c->index;
3235 if (_braille_register_console(newcon, c))
3238 if (newcon->setup &&
3239 (err = newcon->setup(newcon, c->options)) != 0)
3242 newcon->flags |= CON_ENABLED;
3243 if (i == preferred_console)
3244 newcon->flags |= CON_CONSDEV;
3249 * Some consoles, such as pstore and netconsole, can be enabled even
3250 * without matching. Accept the pre-enabled consoles only when match()
3251 * and setup() had a chance to be called.
3253 if (newcon->flags & CON_ENABLED && c->user_specified == user_specified)
3259 /* Try to enable the console unconditionally */
3260 static void try_enable_default_console(struct console *newcon)
3262 if (newcon->index < 0)
3265 if (newcon->setup && newcon->setup(newcon, NULL) != 0)
3268 newcon->flags |= CON_ENABLED;
3271 newcon->flags |= CON_CONSDEV;
3274 #define con_printk(lvl, con, fmt, ...) \
3275 printk(lvl pr_fmt("%sconsole [%s%d] " fmt), \
3276 (con->flags & CON_BOOT) ? "boot" : "", \
3277 con->name, con->index, ##__VA_ARGS__)
3279 static void console_init_seq(struct console *newcon, bool bootcon_registered)
3281 struct console *con;
3284 if (newcon->flags & (CON_PRINTBUFFER | CON_BOOT)) {
3285 /* Get a consistent copy of @syslog_seq. */
3286 mutex_lock(&syslog_lock);
3287 newcon->seq = syslog_seq;
3288 mutex_unlock(&syslog_lock);
3290 /* Begin with next message added to ringbuffer. */
3291 newcon->seq = prb_next_seq(prb);
3294 * If any enabled boot consoles are due to be unregistered
3295 * shortly, some may not be caught up and may be the same
3296 * device as @newcon. Since it is not known which boot console
3297 * is the same device, flush all consoles and, if necessary,
3298 * start with the message of the enabled boot console that is
3299 * the furthest behind.
3301 if (bootcon_registered && !keep_bootcon) {
3303 * Hold the console_lock to stop console printing and
3304 * guarantee safe access to console->seq.
3309 * Flush all consoles and set the console to start at
3310 * the next unprinted sequence number.
3312 if (!console_flush_all(true, &newcon->seq, &handover)) {
3314 * Flushing failed. Just choose the lowest
3315 * sequence of the enabled boot consoles.
3319 * If there was a handover, this context no
3320 * longer holds the console_lock.
3325 newcon->seq = prb_next_seq(prb);
3326 for_each_console(con) {
3327 if ((con->flags & CON_BOOT) &&
3328 (con->flags & CON_ENABLED) &&
3329 con->seq < newcon->seq) {
3330 newcon->seq = con->seq;
3340 #define console_first() \
3341 hlist_entry(console_list.first, struct console, node)
3343 static int unregister_console_locked(struct console *console);
3346 * The console driver calls this routine during kernel initialization
3347 * to register the console printing procedure with printk() and to
3348 * print any messages that were printed by the kernel before the
3349 * console driver was initialized.
3351 * This can happen pretty early during the boot process (because of
3352 * early_printk) - sometimes before setup_arch() completes - be careful
3353 * of what kernel features are used - they may not be initialised yet.
3355 * There are two types of consoles - bootconsoles (early_printk) and
3356 * "real" consoles (everything which is not a bootconsole) which are
3357 * handled differently.
3358 * - Any number of bootconsoles can be registered at any time.
3359 * - As soon as a "real" console is registered, all bootconsoles
3360 * will be unregistered automatically.
3361 * - Once a "real" console is registered, any attempt to register a
3362 * bootconsoles will be rejected
3364 void register_console(struct console *newcon)
3366 struct console *con;
3367 bool bootcon_registered = false;
3368 bool realcon_registered = false;
3371 console_list_lock();
3373 for_each_console(con) {
3374 if (WARN(con == newcon, "console '%s%d' already registered\n",
3375 con->name, con->index)) {
3379 if (con->flags & CON_BOOT)
3380 bootcon_registered = true;
3382 realcon_registered = true;
3385 /* Do not register boot consoles when there already is a real one. */
3386 if ((newcon->flags & CON_BOOT) && realcon_registered) {
3387 pr_info("Too late to register bootconsole %s%d\n",
3388 newcon->name, newcon->index);
3393 * See if we want to enable this console driver by default.
3395 * Nope when a console is preferred by the command line, device
3398 * The first real console with tty binding (driver) wins. More
3399 * consoles might get enabled before the right one is found.
3401 * Note that a console with tty binding will have CON_CONSDEV
3402 * flag set and will be first in the list.
3404 if (preferred_console < 0) {
3405 if (hlist_empty(&console_list) || !console_first()->device ||
3406 console_first()->flags & CON_BOOT) {
3407 try_enable_default_console(newcon);
3411 /* See if this console matches one we selected on the command line */
3412 err = try_enable_preferred_console(newcon, true);
3414 /* If not, try to match against the platform default(s) */
3416 err = try_enable_preferred_console(newcon, false);
3418 /* printk() messages are not printed to the Braille console. */
3419 if (err || newcon->flags & CON_BRL)
3423 * If we have a bootconsole, and are switching to a real console,
3424 * don't print everything out again, since when the boot console, and
3425 * the real console are the same physical device, it's annoying to
3426 * see the beginning boot messages twice
3428 if (bootcon_registered &&
3429 ((newcon->flags & (CON_CONSDEV | CON_BOOT)) == CON_CONSDEV)) {
3430 newcon->flags &= ~CON_PRINTBUFFER;
3433 newcon->dropped = 0;
3434 console_init_seq(newcon, bootcon_registered);
3437 * Put this console in the list - keep the
3438 * preferred driver at the head of the list.
3440 if (hlist_empty(&console_list)) {
3441 /* Ensure CON_CONSDEV is always set for the head. */
3442 newcon->flags |= CON_CONSDEV;
3443 hlist_add_head_rcu(&newcon->node, &console_list);
3445 } else if (newcon->flags & CON_CONSDEV) {
3446 /* Only the new head can have CON_CONSDEV set. */
3447 console_srcu_write_flags(console_first(), console_first()->flags & ~CON_CONSDEV);
3448 hlist_add_head_rcu(&newcon->node, &console_list);
3451 hlist_add_behind_rcu(&newcon->node, console_list.first);
3455 * No need to synchronize SRCU here! The caller does not rely
3456 * on all contexts being able to see the new console before
3457 * register_console() completes.
3460 console_sysfs_notify();
3463 * By unregistering the bootconsoles after we enable the real console
3464 * we get the "console xxx enabled" message on all the consoles -
3465 * boot consoles, real consoles, etc - this is to ensure that end
3466 * users know there might be something in the kernel's log buffer that
3467 * went to the bootconsole (that they do not see on the real console)
3469 con_printk(KERN_INFO, newcon, "enabled\n");
3470 if (bootcon_registered &&
3471 ((newcon->flags & (CON_CONSDEV | CON_BOOT)) == CON_CONSDEV) &&
3473 struct hlist_node *tmp;
3475 hlist_for_each_entry_safe(con, tmp, &console_list, node) {
3476 if (con->flags & CON_BOOT)
3477 unregister_console_locked(con);
3481 console_list_unlock();
3483 EXPORT_SYMBOL(register_console);
3485 /* Must be called under console_list_lock(). */
3486 static int unregister_console_locked(struct console *console)
3490 lockdep_assert_console_list_lock_held();
3492 con_printk(KERN_INFO, console, "disabled\n");
3494 res = _braille_unregister_console(console);
3500 /* Disable it unconditionally */
3501 console_srcu_write_flags(console, console->flags & ~CON_ENABLED);
3503 if (!console_is_registered_locked(console))
3506 hlist_del_init_rcu(&console->node);
3510 * If this isn't the last console and it has CON_CONSDEV set, we
3511 * need to set it on the next preferred console.
3514 * The above makes no sense as there is no guarantee that the next
3515 * console has any device attached. Oh well....
3517 if (!hlist_empty(&console_list) && console->flags & CON_CONSDEV)
3518 console_srcu_write_flags(console_first(), console_first()->flags | CON_CONSDEV);
3521 * Ensure that all SRCU list walks have completed. All contexts
3522 * must not be able to see this console in the list so that any
3523 * exit/cleanup routines can be performed safely.
3525 synchronize_srcu(&console_srcu);
3527 console_sysfs_notify();
3530 res = console->exit(console);
3535 int unregister_console(struct console *console)
3539 console_list_lock();
3540 res = unregister_console_locked(console);
3541 console_list_unlock();
3544 EXPORT_SYMBOL(unregister_console);
3547 * console_force_preferred_locked - force a registered console preferred
3548 * @con: The registered console to force preferred.
3550 * Must be called under console_list_lock().
3552 void console_force_preferred_locked(struct console *con)
3554 struct console *cur_pref_con;
3556 if (!console_is_registered_locked(con))
3559 cur_pref_con = console_first();
3561 /* Already preferred? */
3562 if (cur_pref_con == con)
3566 * Delete, but do not re-initialize the entry. This allows the console
3567 * to continue to appear registered (via any hlist_unhashed_lockless()
3568 * checks), even though it was briefly removed from the console list.
3570 hlist_del_rcu(&con->node);
3573 * Ensure that all SRCU list walks have completed so that the console
3574 * can be added to the beginning of the console list and its forward
3575 * list pointer can be re-initialized.
3577 synchronize_srcu(&console_srcu);
3579 con->flags |= CON_CONSDEV;
3580 WARN_ON(!con->device);
3582 /* Only the new head can have CON_CONSDEV set. */
3583 console_srcu_write_flags(cur_pref_con, cur_pref_con->flags & ~CON_CONSDEV);
3584 hlist_add_head_rcu(&con->node, &console_list);
3586 EXPORT_SYMBOL(console_force_preferred_locked);
3589 * Initialize the console device. This is called *early*, so
3590 * we can't necessarily depend on lots of kernel help here.
3591 * Just do some early initializations, and do the complex setup
3594 void __init console_init(void)
3598 initcall_entry_t *ce;
3600 /* Setup the default TTY line discipline. */
3604 * set up the console device so that later boot sequences can
3605 * inform about problems etc..
3607 ce = __con_initcall_start;
3608 trace_initcall_level("console");
3609 while (ce < __con_initcall_end) {
3610 call = initcall_from_entry(ce);
3611 trace_initcall_start(call);
3613 trace_initcall_finish(call, ret);
3619 * Some boot consoles access data that is in the init section and which will
3620 * be discarded after the initcalls have been run. To make sure that no code
3621 * will access this data, unregister the boot consoles in a late initcall.
3623 * If for some reason, such as deferred probe or the driver being a loadable
3624 * module, the real console hasn't registered yet at this point, there will
3625 * be a brief interval in which no messages are logged to the console, which
3626 * makes it difficult to diagnose problems that occur during this time.
3628 * To mitigate this problem somewhat, only unregister consoles whose memory
3629 * intersects with the init section. Note that all other boot consoles will
3630 * get unregistered when the real preferred console is registered.
3632 static int __init printk_late_init(void)
3634 struct hlist_node *tmp;
3635 struct console *con;
3638 console_list_lock();
3639 hlist_for_each_entry_safe(con, tmp, &console_list, node) {
3640 if (!(con->flags & CON_BOOT))
3643 /* Check addresses that might be used for enabled consoles. */
3644 if (init_section_intersects(con, sizeof(*con)) ||
3645 init_section_contains(con->write, 0) ||
3646 init_section_contains(con->read, 0) ||
3647 init_section_contains(con->device, 0) ||
3648 init_section_contains(con->unblank, 0) ||
3649 init_section_contains(con->data, 0)) {
3651 * Please, consider moving the reported consoles out
3652 * of the init section.
3654 pr_warn("bootconsole [%s%d] uses init memory and must be disabled even before the real one is ready\n",
3655 con->name, con->index);
3656 unregister_console_locked(con);
3659 console_list_unlock();
3661 ret = cpuhp_setup_state_nocalls(CPUHP_PRINTK_DEAD, "printk:dead", NULL,
3662 console_cpu_notify);
3664 ret = cpuhp_setup_state_nocalls(CPUHP_AP_ONLINE_DYN, "printk:online",
3665 console_cpu_notify, NULL);
3667 printk_sysctl_init();
3670 late_initcall(printk_late_init);
3672 #if defined CONFIG_PRINTK
3673 /* If @con is specified, only wait for that console. Otherwise wait for all. */
3674 static bool __pr_flush(struct console *con, int timeout_ms, bool reset_on_progress)
3676 int remaining = timeout_ms;
3686 seq = prb_next_seq(prb);
3692 * Hold the console_lock to guarantee safe access to
3693 * console->seq and to prevent changes to @console_suspended
3694 * until all consoles have been processed.
3698 cookie = console_srcu_read_lock();
3699 for_each_console_srcu(c) {
3700 if (con && con != c)
3702 if (!console_is_usable(c))
3704 printk_seq = c->seq;
3705 if (printk_seq < seq)
3706 diff += seq - printk_seq;
3708 console_srcu_read_unlock(cookie);
3711 * If consoles are suspended, it cannot be expected that they
3712 * make forward progress, so timeout immediately. @diff is
3713 * still used to return a valid flush status.
3715 if (console_suspended)
3717 else if (diff != last_diff && reset_on_progress)
3718 remaining = timeout_ms;
3722 if (diff == 0 || remaining == 0)
3725 if (remaining < 0) {
3726 /* no timeout limit */
3728 } else if (remaining < 100) {
3743 * pr_flush() - Wait for printing threads to catch up.
3745 * @timeout_ms: The maximum time (in ms) to wait.
3746 * @reset_on_progress: Reset the timeout if forward progress is seen.
3748 * A value of 0 for @timeout_ms means no waiting will occur. A value of -1
3749 * represents infinite waiting.
3751 * If @reset_on_progress is true, the timeout will be reset whenever any
3752 * printer has been seen to make some forward progress.
3754 * Context: Process context. May sleep while acquiring console lock.
3755 * Return: true if all enabled printers are caught up.
3757 static bool pr_flush(int timeout_ms, bool reset_on_progress)
3759 return __pr_flush(NULL, timeout_ms, reset_on_progress);
3763 * Delayed printk version, for scheduler-internal messages:
3765 #define PRINTK_PENDING_WAKEUP 0x01
3766 #define PRINTK_PENDING_OUTPUT 0x02
3768 static DEFINE_PER_CPU(int, printk_pending);
3770 static void wake_up_klogd_work_func(struct irq_work *irq_work)
3772 int pending = this_cpu_xchg(printk_pending, 0);
3774 if (pending & PRINTK_PENDING_OUTPUT) {
3775 /* If trylock fails, someone else is doing the printing */
3776 if (console_trylock())
3780 if (pending & PRINTK_PENDING_WAKEUP)
3781 wake_up_interruptible(&log_wait);
3784 static DEFINE_PER_CPU(struct irq_work, wake_up_klogd_work) =
3785 IRQ_WORK_INIT_LAZY(wake_up_klogd_work_func);
3787 static void __wake_up_klogd(int val)
3789 if (!printk_percpu_data_ready())
3794 * Guarantee any new records can be seen by tasks preparing to wait
3795 * before this context checks if the wait queue is empty.
3797 * The full memory barrier within wq_has_sleeper() pairs with the full
3798 * memory barrier within set_current_state() of
3799 * prepare_to_wait_event(), which is called after ___wait_event() adds
3800 * the waiter but before it has checked the wait condition.
3802 * This pairs with devkmsg_read:A and syslog_print:A.
3804 if (wq_has_sleeper(&log_wait) || /* LMM(__wake_up_klogd:A) */
3805 (val & PRINTK_PENDING_OUTPUT)) {
3806 this_cpu_or(printk_pending, val);
3807 irq_work_queue(this_cpu_ptr(&wake_up_klogd_work));
3812 void wake_up_klogd(void)
3814 __wake_up_klogd(PRINTK_PENDING_WAKEUP);
3817 void defer_console_output(void)
3820 * New messages may have been added directly to the ringbuffer
3821 * using vprintk_store(), so wake any waiters as well.
3823 __wake_up_klogd(PRINTK_PENDING_WAKEUP | PRINTK_PENDING_OUTPUT);
3826 void printk_trigger_flush(void)
3828 defer_console_output();
3831 int vprintk_deferred(const char *fmt, va_list args)
3835 r = vprintk_emit(0, LOGLEVEL_SCHED, NULL, fmt, args);
3836 defer_console_output();
3841 int _printk_deferred(const char *fmt, ...)
3846 va_start(args, fmt);
3847 r = vprintk_deferred(fmt, args);
3854 * printk rate limiting, lifted from the networking subsystem.
3856 * This enforces a rate limit: not more than 10 kernel messages
3857 * every 5s to make a denial-of-service attack impossible.
3859 DEFINE_RATELIMIT_STATE(printk_ratelimit_state, 5 * HZ, 10);
3861 int __printk_ratelimit(const char *func)
3863 return ___ratelimit(&printk_ratelimit_state, func);
3865 EXPORT_SYMBOL(__printk_ratelimit);
3868 * printk_timed_ratelimit - caller-controlled printk ratelimiting
3869 * @caller_jiffies: pointer to caller's state
3870 * @interval_msecs: minimum interval between prints
3872 * printk_timed_ratelimit() returns true if more than @interval_msecs
3873 * milliseconds have elapsed since the last time printk_timed_ratelimit()
3876 bool printk_timed_ratelimit(unsigned long *caller_jiffies,
3877 unsigned int interval_msecs)
3879 unsigned long elapsed = jiffies - *caller_jiffies;
3881 if (*caller_jiffies && elapsed <= msecs_to_jiffies(interval_msecs))
3884 *caller_jiffies = jiffies;
3887 EXPORT_SYMBOL(printk_timed_ratelimit);
3889 static DEFINE_SPINLOCK(dump_list_lock);
3890 static LIST_HEAD(dump_list);
3893 * kmsg_dump_register - register a kernel log dumper.
3894 * @dumper: pointer to the kmsg_dumper structure
3896 * Adds a kernel log dumper to the system. The dump callback in the
3897 * structure will be called when the kernel oopses or panics and must be
3898 * set. Returns zero on success and %-EINVAL or %-EBUSY otherwise.
3900 int kmsg_dump_register(struct kmsg_dumper *dumper)
3902 unsigned long flags;
3905 /* The dump callback needs to be set */
3909 spin_lock_irqsave(&dump_list_lock, flags);
3910 /* Don't allow registering multiple times */
3911 if (!dumper->registered) {
3912 dumper->registered = 1;
3913 list_add_tail_rcu(&dumper->list, &dump_list);
3916 spin_unlock_irqrestore(&dump_list_lock, flags);
3920 EXPORT_SYMBOL_GPL(kmsg_dump_register);
3923 * kmsg_dump_unregister - unregister a kmsg dumper.
3924 * @dumper: pointer to the kmsg_dumper structure
3926 * Removes a dump device from the system. Returns zero on success and
3927 * %-EINVAL otherwise.
3929 int kmsg_dump_unregister(struct kmsg_dumper *dumper)
3931 unsigned long flags;
3934 spin_lock_irqsave(&dump_list_lock, flags);
3935 if (dumper->registered) {
3936 dumper->registered = 0;
3937 list_del_rcu(&dumper->list);
3940 spin_unlock_irqrestore(&dump_list_lock, flags);
3945 EXPORT_SYMBOL_GPL(kmsg_dump_unregister);
3947 static bool always_kmsg_dump;
3948 module_param_named(always_kmsg_dump, always_kmsg_dump, bool, S_IRUGO | S_IWUSR);
3950 const char *kmsg_dump_reason_str(enum kmsg_dump_reason reason)
3953 case KMSG_DUMP_PANIC:
3955 case KMSG_DUMP_OOPS:
3957 case KMSG_DUMP_EMERG:
3959 case KMSG_DUMP_SHUTDOWN:
3965 EXPORT_SYMBOL_GPL(kmsg_dump_reason_str);
3968 * kmsg_dump - dump kernel log to kernel message dumpers.
3969 * @reason: the reason (oops, panic etc) for dumping
3971 * Call each of the registered dumper's dump() callback, which can
3972 * retrieve the kmsg records with kmsg_dump_get_line() or
3973 * kmsg_dump_get_buffer().
3975 void kmsg_dump(enum kmsg_dump_reason reason)
3977 struct kmsg_dumper *dumper;
3980 list_for_each_entry_rcu(dumper, &dump_list, list) {
3981 enum kmsg_dump_reason max_reason = dumper->max_reason;
3984 * If client has not provided a specific max_reason, default
3985 * to KMSG_DUMP_OOPS, unless always_kmsg_dump was set.
3987 if (max_reason == KMSG_DUMP_UNDEF) {
3988 max_reason = always_kmsg_dump ? KMSG_DUMP_MAX :
3991 if (reason > max_reason)
3994 /* invoke dumper which will iterate over records */
3995 dumper->dump(dumper, reason);
4001 * kmsg_dump_get_line - retrieve one kmsg log line
4002 * @iter: kmsg dump iterator
4003 * @syslog: include the "<4>" prefixes
4004 * @line: buffer to copy the line to
4005 * @size: maximum size of the buffer
4006 * @len: length of line placed into buffer
4008 * Start at the beginning of the kmsg buffer, with the oldest kmsg
4009 * record, and copy one record into the provided buffer.
4011 * Consecutive calls will return the next available record moving
4012 * towards the end of the buffer with the youngest messages.
4014 * A return value of FALSE indicates that there are no more records to
4017 bool kmsg_dump_get_line(struct kmsg_dump_iter *iter, bool syslog,
4018 char *line, size_t size, size_t *len)
4020 u64 min_seq = latched_seq_read_nolock(&clear_seq);
4021 struct printk_info info;
4022 unsigned int line_count;
4023 struct printk_record r;
4027 if (iter->cur_seq < min_seq)
4028 iter->cur_seq = min_seq;
4030 prb_rec_init_rd(&r, &info, line, size);
4032 /* Read text or count text lines? */
4034 if (!prb_read_valid(prb, iter->cur_seq, &r))
4036 l = record_print_text(&r, syslog, printk_time);
4038 if (!prb_read_valid_info(prb, iter->cur_seq,
4039 &info, &line_count)) {
4042 l = get_record_print_text_size(&info, line_count, syslog,
4047 iter->cur_seq = r.info->seq + 1;
4054 EXPORT_SYMBOL_GPL(kmsg_dump_get_line);
4057 * kmsg_dump_get_buffer - copy kmsg log lines
4058 * @iter: kmsg dump iterator
4059 * @syslog: include the "<4>" prefixes
4060 * @buf: buffer to copy the line to
4061 * @size: maximum size of the buffer
4062 * @len_out: length of line placed into buffer
4064 * Start at the end of the kmsg buffer and fill the provided buffer
4065 * with as many of the *youngest* kmsg records that fit into it.
4066 * If the buffer is large enough, all available kmsg records will be
4067 * copied with a single call.
4069 * Consecutive calls will fill the buffer with the next block of
4070 * available older records, not including the earlier retrieved ones.
4072 * A return value of FALSE indicates that there are no more records to
4075 bool kmsg_dump_get_buffer(struct kmsg_dump_iter *iter, bool syslog,
4076 char *buf, size_t size, size_t *len_out)
4078 u64 min_seq = latched_seq_read_nolock(&clear_seq);
4079 struct printk_info info;
4080 struct printk_record r;
4085 bool time = printk_time;
4090 if (iter->cur_seq < min_seq)
4091 iter->cur_seq = min_seq;
4093 if (prb_read_valid_info(prb, iter->cur_seq, &info, NULL)) {
4094 if (info.seq != iter->cur_seq) {
4095 /* messages are gone, move to first available one */
4096 iter->cur_seq = info.seq;
4101 if (iter->cur_seq >= iter->next_seq)
4105 * Find first record that fits, including all following records,
4106 * into the user-provided buffer for this dump. Pass in size-1
4107 * because this function (by way of record_print_text()) will
4108 * not write more than size-1 bytes of text into @buf.
4110 seq = find_first_fitting_seq(iter->cur_seq, iter->next_seq,
4111 size - 1, syslog, time);
4114 * Next kmsg_dump_get_buffer() invocation will dump block of
4115 * older records stored right before this one.
4119 prb_rec_init_rd(&r, &info, buf, size);
4122 prb_for_each_record(seq, prb, seq, &r) {
4123 if (r.info->seq >= iter->next_seq)
4126 len += record_print_text(&r, syslog, time);
4128 /* Adjust record to store to remaining buffer space. */
4129 prb_rec_init_rd(&r, &info, buf + len, size - len);
4132 iter->next_seq = next_seq;
4139 EXPORT_SYMBOL_GPL(kmsg_dump_get_buffer);
4142 * kmsg_dump_rewind - reset the iterator
4143 * @iter: kmsg dump iterator
4145 * Reset the dumper's iterator so that kmsg_dump_get_line() and
4146 * kmsg_dump_get_buffer() can be called again and used multiple
4147 * times within the same dumper.dump() callback.
4149 void kmsg_dump_rewind(struct kmsg_dump_iter *iter)
4151 iter->cur_seq = latched_seq_read_nolock(&clear_seq);
4152 iter->next_seq = prb_next_seq(prb);
4154 EXPORT_SYMBOL_GPL(kmsg_dump_rewind);
4159 static atomic_t printk_cpu_sync_owner = ATOMIC_INIT(-1);
4160 static atomic_t printk_cpu_sync_nested = ATOMIC_INIT(0);
4163 * __printk_cpu_sync_wait() - Busy wait until the printk cpu-reentrant
4164 * spinning lock is not owned by any CPU.
4166 * Context: Any context.
4168 void __printk_cpu_sync_wait(void)
4172 } while (atomic_read(&printk_cpu_sync_owner) != -1);
4174 EXPORT_SYMBOL(__printk_cpu_sync_wait);
4177 * __printk_cpu_sync_try_get() - Try to acquire the printk cpu-reentrant
4180 * If no processor has the lock, the calling processor takes the lock and
4181 * becomes the owner. If the calling processor is already the owner of the
4182 * lock, this function succeeds immediately.
4184 * Context: Any context. Expects interrupts to be disabled.
4185 * Return: 1 on success, otherwise 0.
4187 int __printk_cpu_sync_try_get(void)
4192 cpu = smp_processor_id();
4195 * Guarantee loads and stores from this CPU when it is the lock owner
4196 * are _not_ visible to the previous lock owner. This pairs with
4197 * __printk_cpu_sync_put:B.
4199 * Memory barrier involvement:
4201 * If __printk_cpu_sync_try_get:A reads from __printk_cpu_sync_put:B,
4202 * then __printk_cpu_sync_put:A can never read from
4203 * __printk_cpu_sync_try_get:B.
4207 * RELEASE from __printk_cpu_sync_put:A to __printk_cpu_sync_put:B
4208 * of the previous CPU
4210 * ACQUIRE from __printk_cpu_sync_try_get:A to
4211 * __printk_cpu_sync_try_get:B of this CPU
4213 old = atomic_cmpxchg_acquire(&printk_cpu_sync_owner, -1,
4214 cpu); /* LMM(__printk_cpu_sync_try_get:A) */
4217 * This CPU is now the owner and begins loading/storing
4218 * data: LMM(__printk_cpu_sync_try_get:B)
4222 } else if (old == cpu) {
4223 /* This CPU is already the owner. */
4224 atomic_inc(&printk_cpu_sync_nested);
4230 EXPORT_SYMBOL(__printk_cpu_sync_try_get);
4233 * __printk_cpu_sync_put() - Release the printk cpu-reentrant spinning lock.
4235 * The calling processor must be the owner of the lock.
4237 * Context: Any context. Expects interrupts to be disabled.
4239 void __printk_cpu_sync_put(void)
4241 if (atomic_read(&printk_cpu_sync_nested)) {
4242 atomic_dec(&printk_cpu_sync_nested);
4247 * This CPU is finished loading/storing data:
4248 * LMM(__printk_cpu_sync_put:A)
4252 * Guarantee loads and stores from this CPU when it was the
4253 * lock owner are visible to the next lock owner. This pairs
4254 * with __printk_cpu_sync_try_get:A.
4256 * Memory barrier involvement:
4258 * If __printk_cpu_sync_try_get:A reads from __printk_cpu_sync_put:B,
4259 * then __printk_cpu_sync_try_get:B reads from __printk_cpu_sync_put:A.
4263 * RELEASE from __printk_cpu_sync_put:A to __printk_cpu_sync_put:B
4266 * ACQUIRE from __printk_cpu_sync_try_get:A to
4267 * __printk_cpu_sync_try_get:B of the next CPU
4269 atomic_set_release(&printk_cpu_sync_owner,
4270 -1); /* LMM(__printk_cpu_sync_put:B) */
4272 EXPORT_SYMBOL(__printk_cpu_sync_put);
4273 #endif /* CONFIG_SMP */