1 // SPDX-License-Identifier: GPL-2.0-only
3 * linux/kernel/printk.c
5 * Copyright (C) 1991, 1992 Linus Torvalds
7 * Modified to make sys_syslog() more flexible: added commands to
8 * return the last 4k of kernel messages, regardless of whether
9 * they've been read or not. Added option to suppress kernel printk's
10 * to the console. Added hook for sending the console messages
11 * elsewhere, in preparation for a serial line console (someday).
13 * Modified for sysctl support, 1/8/97, Chris Horn.
14 * Fixed SMP synchronization, 08/08/99, Manfred Spraul
15 * manfred@colorfullife.com
16 * Rewrote bits to get rid of console_lock
17 * 01Mar01 Andrew Morton
20 #define pr_fmt(fmt) KBUILD_MODNAME ": " fmt
22 #include <linux/kernel.h>
24 #include <linux/tty.h>
25 #include <linux/tty_driver.h>
26 #include <linux/console.h>
27 #include <linux/init.h>
28 #include <linux/jiffies.h>
29 #include <linux/nmi.h>
30 #include <linux/module.h>
31 #include <linux/moduleparam.h>
32 #include <linux/delay.h>
33 #include <linux/smp.h>
34 #include <linux/security.h>
35 #include <linux/memblock.h>
36 #include <linux/syscalls.h>
37 #include <linux/crash_core.h>
38 #include <linux/ratelimit.h>
39 #include <linux/kmsg_dump.h>
40 #include <linux/syslog.h>
41 #include <linux/cpu.h>
42 #include <linux/rculist.h>
43 #include <linux/poll.h>
44 #include <linux/irq_work.h>
45 #include <linux/ctype.h>
46 #include <linux/uio.h>
47 #include <linux/sched/clock.h>
48 #include <linux/sched/debug.h>
49 #include <linux/sched/task_stack.h>
51 #include <linux/uaccess.h>
52 #include <asm/sections.h>
54 #include <trace/events/initcall.h>
55 #define CREATE_TRACE_POINTS
56 #include <trace/events/printk.h>
58 #include "printk_ringbuffer.h"
59 #include "console_cmdline.h"
63 int console_printk[4] = {
64 CONSOLE_LOGLEVEL_DEFAULT, /* console_loglevel */
65 MESSAGE_LOGLEVEL_DEFAULT, /* default_message_loglevel */
66 CONSOLE_LOGLEVEL_MIN, /* minimum_console_loglevel */
67 CONSOLE_LOGLEVEL_DEFAULT, /* default_console_loglevel */
69 EXPORT_SYMBOL_GPL(console_printk);
71 atomic_t ignore_console_lock_warning __read_mostly = ATOMIC_INIT(0);
72 EXPORT_SYMBOL(ignore_console_lock_warning);
75 * Low level drivers may need that to know if they can schedule in
76 * their unblank() callback or not. So let's export it.
79 EXPORT_SYMBOL(oops_in_progress);
82 * console_mutex protects console_list updates and console->flags updates.
83 * The flags are synchronized only for consoles that are registered, i.e.
84 * accessible via the console list.
86 static DEFINE_MUTEX(console_mutex);
89 * console_sem protects updates to console->seq and console_suspended,
90 * and also provides serialization for console printing.
92 static DEFINE_SEMAPHORE(console_sem);
93 HLIST_HEAD(console_list);
94 EXPORT_SYMBOL_GPL(console_list);
95 DEFINE_STATIC_SRCU(console_srcu);
98 * System may need to suppress printk message under certain
99 * circumstances, like after kernel panic happens.
101 int __read_mostly suppress_printk;
104 * During panic, heavy printk by other CPUs can delay the
105 * panic and risk deadlock on console resources.
107 static int __read_mostly suppress_panic_printk;
109 #ifdef CONFIG_LOCKDEP
110 static struct lockdep_map console_lock_dep_map = {
111 .name = "console_lock"
114 void lockdep_assert_console_list_lock_held(void)
116 lockdep_assert_held(&console_mutex);
118 EXPORT_SYMBOL(lockdep_assert_console_list_lock_held);
121 #ifdef CONFIG_DEBUG_LOCK_ALLOC
122 bool console_srcu_read_lock_is_held(void)
124 return srcu_read_lock_held(&console_srcu);
126 EXPORT_SYMBOL(console_srcu_read_lock_is_held);
129 enum devkmsg_log_bits {
130 __DEVKMSG_LOG_BIT_ON = 0,
131 __DEVKMSG_LOG_BIT_OFF,
132 __DEVKMSG_LOG_BIT_LOCK,
135 enum devkmsg_log_masks {
136 DEVKMSG_LOG_MASK_ON = BIT(__DEVKMSG_LOG_BIT_ON),
137 DEVKMSG_LOG_MASK_OFF = BIT(__DEVKMSG_LOG_BIT_OFF),
138 DEVKMSG_LOG_MASK_LOCK = BIT(__DEVKMSG_LOG_BIT_LOCK),
141 /* Keep both the 'on' and 'off' bits clear, i.e. ratelimit by default: */
142 #define DEVKMSG_LOG_MASK_DEFAULT 0
144 static unsigned int __read_mostly devkmsg_log = DEVKMSG_LOG_MASK_DEFAULT;
146 static int __control_devkmsg(char *str)
153 len = str_has_prefix(str, "on");
155 devkmsg_log = DEVKMSG_LOG_MASK_ON;
159 len = str_has_prefix(str, "off");
161 devkmsg_log = DEVKMSG_LOG_MASK_OFF;
165 len = str_has_prefix(str, "ratelimit");
167 devkmsg_log = DEVKMSG_LOG_MASK_DEFAULT;
174 static int __init control_devkmsg(char *str)
176 if (__control_devkmsg(str) < 0) {
177 pr_warn("printk.devkmsg: bad option string '%s'\n", str);
182 * Set sysctl string accordingly:
184 if (devkmsg_log == DEVKMSG_LOG_MASK_ON)
185 strcpy(devkmsg_log_str, "on");
186 else if (devkmsg_log == DEVKMSG_LOG_MASK_OFF)
187 strcpy(devkmsg_log_str, "off");
188 /* else "ratelimit" which is set by default. */
191 * Sysctl cannot change it anymore. The kernel command line setting of
192 * this parameter is to force the setting to be permanent throughout the
193 * runtime of the system. This is a precation measure against userspace
194 * trying to be a smarta** and attempting to change it up on us.
196 devkmsg_log |= DEVKMSG_LOG_MASK_LOCK;
200 __setup("printk.devkmsg=", control_devkmsg);
202 char devkmsg_log_str[DEVKMSG_STR_MAX_SIZE] = "ratelimit";
203 #if defined(CONFIG_PRINTK) && defined(CONFIG_SYSCTL)
204 int devkmsg_sysctl_set_loglvl(struct ctl_table *table, int write,
205 void *buffer, size_t *lenp, loff_t *ppos)
207 char old_str[DEVKMSG_STR_MAX_SIZE];
212 if (devkmsg_log & DEVKMSG_LOG_MASK_LOCK)
216 strncpy(old_str, devkmsg_log_str, DEVKMSG_STR_MAX_SIZE);
219 err = proc_dostring(table, write, buffer, lenp, ppos);
224 err = __control_devkmsg(devkmsg_log_str);
227 * Do not accept an unknown string OR a known string with
230 if (err < 0 || (err + 1 != *lenp)) {
232 /* ... and restore old setting. */
234 strncpy(devkmsg_log_str, old_str, DEVKMSG_STR_MAX_SIZE);
242 #endif /* CONFIG_PRINTK && CONFIG_SYSCTL */
245 * console_list_lock - Lock the console list
247 * For console list or console->flags updates
249 void console_list_lock(void)
252 * In unregister_console() and console_force_preferred_locked(),
253 * synchronize_srcu() is called with the console_list_lock held.
254 * Therefore it is not allowed that the console_list_lock is taken
255 * with the srcu_lock held.
257 * Detecting if this context is really in the read-side critical
258 * section is only possible if the appropriate debug options are
261 WARN_ON_ONCE(debug_lockdep_rcu_enabled() &&
262 srcu_read_lock_held(&console_srcu));
264 mutex_lock(&console_mutex);
266 EXPORT_SYMBOL(console_list_lock);
269 * console_list_unlock - Unlock the console list
271 * Counterpart to console_list_lock()
273 void console_list_unlock(void)
275 mutex_unlock(&console_mutex);
277 EXPORT_SYMBOL(console_list_unlock);
280 * console_srcu_read_lock - Register a new reader for the
281 * SRCU-protected console list
283 * Use for_each_console_srcu() to iterate the console list
285 * Context: Any context.
286 * Return: A cookie to pass to console_srcu_read_unlock().
288 int console_srcu_read_lock(void)
290 return srcu_read_lock_nmisafe(&console_srcu);
292 EXPORT_SYMBOL(console_srcu_read_lock);
295 * console_srcu_read_unlock - Unregister an old reader from
296 * the SRCU-protected console list
297 * @cookie: cookie returned from console_srcu_read_lock()
299 * Counterpart to console_srcu_read_lock()
301 void console_srcu_read_unlock(int cookie)
303 srcu_read_unlock_nmisafe(&console_srcu, cookie);
305 EXPORT_SYMBOL(console_srcu_read_unlock);
308 * Helper macros to handle lockdep when locking/unlocking console_sem. We use
309 * macros instead of functions so that _RET_IP_ contains useful information.
311 #define down_console_sem() do { \
313 mutex_acquire(&console_lock_dep_map, 0, 0, _RET_IP_);\
316 static int __down_trylock_console_sem(unsigned long ip)
322 * Here and in __up_console_sem() we need to be in safe mode,
323 * because spindump/WARN/etc from under console ->lock will
324 * deadlock in printk()->down_trylock_console_sem() otherwise.
326 printk_safe_enter_irqsave(flags);
327 lock_failed = down_trylock(&console_sem);
328 printk_safe_exit_irqrestore(flags);
332 mutex_acquire(&console_lock_dep_map, 0, 1, ip);
335 #define down_trylock_console_sem() __down_trylock_console_sem(_RET_IP_)
337 static void __up_console_sem(unsigned long ip)
341 mutex_release(&console_lock_dep_map, ip);
343 printk_safe_enter_irqsave(flags);
345 printk_safe_exit_irqrestore(flags);
347 #define up_console_sem() __up_console_sem(_RET_IP_)
349 static bool panic_in_progress(void)
351 return unlikely(atomic_read(&panic_cpu) != PANIC_CPU_INVALID);
355 * This is used for debugging the mess that is the VT code by
356 * keeping track if we have the console semaphore held. It's
357 * definitely not the perfect debug tool (we don't know if _WE_
358 * hold it and are racing, but it helps tracking those weird code
359 * paths in the console code where we end up in places I want
360 * locked without the console semaphore held).
362 static int console_locked, console_suspended;
365 * Array of consoles built from command line options (console=)
368 #define MAX_CMDLINECONSOLES 8
370 static struct console_cmdline console_cmdline[MAX_CMDLINECONSOLES];
372 static int preferred_console = -1;
373 int console_set_on_cmdline;
374 EXPORT_SYMBOL(console_set_on_cmdline);
376 /* Flag: console code may call schedule() */
377 static int console_may_schedule;
379 enum con_msg_format_flags {
380 MSG_FORMAT_DEFAULT = 0,
381 MSG_FORMAT_SYSLOG = (1 << 0),
384 static int console_msg_format = MSG_FORMAT_DEFAULT;
387 * The printk log buffer consists of a sequenced collection of records, each
388 * containing variable length message text. Every record also contains its
389 * own meta-data (@info).
391 * Every record meta-data carries the timestamp in microseconds, as well as
392 * the standard userspace syslog level and syslog facility. The usual kernel
393 * messages use LOG_KERN; userspace-injected messages always carry a matching
394 * syslog facility, by default LOG_USER. The origin of every message can be
395 * reliably determined that way.
397 * The human readable log message of a record is available in @text, the
398 * length of the message text in @text_len. The stored message is not
401 * Optionally, a record can carry a dictionary of properties (key/value
402 * pairs), to provide userspace with a machine-readable message context.
404 * Examples for well-defined, commonly used property names are:
405 * DEVICE=b12:8 device identifier
409 * +sound:card0 subsystem:devname
410 * SUBSYSTEM=pci driver-core subsystem name
412 * Valid characters in property names are [a-zA-Z0-9.-_]. Property names
413 * and values are terminated by a '\0' character.
415 * Example of record values:
416 * record.text_buf = "it's a line" (unterminated)
417 * record.info.seq = 56
418 * record.info.ts_nsec = 36863
419 * record.info.text_len = 11
420 * record.info.facility = 0 (LOG_KERN)
421 * record.info.flags = 0
422 * record.info.level = 3 (LOG_ERR)
423 * record.info.caller_id = 299 (task 299)
424 * record.info.dev_info.subsystem = "pci" (terminated)
425 * record.info.dev_info.device = "+pci:0000:00:01.0" (terminated)
427 * The 'struct printk_info' buffer must never be directly exported to
428 * userspace, it is a kernel-private implementation detail that might
429 * need to be changed in the future, when the requirements change.
431 * /dev/kmsg exports the structured data in the following line format:
432 * "<level>,<sequnum>,<timestamp>,<contflag>[,additional_values, ... ];<message text>\n"
434 * Users of the export format should ignore possible additional values
435 * separated by ',', and find the message after the ';' character.
437 * The optional key/value pairs are attached as continuation lines starting
438 * with a space character and terminated by a newline. All possible
439 * non-prinatable characters are escaped in the "\xff" notation.
442 /* syslog_lock protects syslog_* variables and write access to clear_seq. */
443 static DEFINE_MUTEX(syslog_lock);
446 DECLARE_WAIT_QUEUE_HEAD(log_wait);
447 /* All 3 protected by @syslog_lock. */
448 /* the next printk record to read by syslog(READ) or /proc/kmsg */
449 static u64 syslog_seq;
450 static size_t syslog_partial;
451 static bool syslog_time;
454 seqcount_latch_t latch;
459 * The next printk record to read after the last 'clear' command. There are
460 * two copies (updated with seqcount_latch) so that reads can locklessly
461 * access a valid value. Writers are synchronized by @syslog_lock.
463 static struct latched_seq clear_seq = {
464 .latch = SEQCNT_LATCH_ZERO(clear_seq.latch),
469 #ifdef CONFIG_PRINTK_CALLER
470 #define PREFIX_MAX 48
472 #define PREFIX_MAX 32
475 /* the maximum size of a formatted record (i.e. with prefix added per line) */
476 #define CONSOLE_LOG_MAX 1024
478 /* the maximum size for a dropped text message */
479 #define DROPPED_TEXT_MAX 64
481 /* the maximum size allowed to be reserved for a record */
482 #define LOG_LINE_MAX (CONSOLE_LOG_MAX - PREFIX_MAX)
484 #define LOG_LEVEL(v) ((v) & 0x07)
485 #define LOG_FACILITY(v) ((v) >> 3 & 0xff)
488 #define LOG_ALIGN __alignof__(unsigned long)
489 #define __LOG_BUF_LEN (1 << CONFIG_LOG_BUF_SHIFT)
490 #define LOG_BUF_LEN_MAX (u32)(1 << 31)
491 static char __log_buf[__LOG_BUF_LEN] __aligned(LOG_ALIGN);
492 static char *log_buf = __log_buf;
493 static u32 log_buf_len = __LOG_BUF_LEN;
496 * Define the average message size. This only affects the number of
497 * descriptors that will be available. Underestimating is better than
498 * overestimating (too many available descriptors is better than not enough).
500 #define PRB_AVGBITS 5 /* 32 character average length */
502 #if CONFIG_LOG_BUF_SHIFT <= PRB_AVGBITS
503 #error CONFIG_LOG_BUF_SHIFT value too small.
505 _DEFINE_PRINTKRB(printk_rb_static, CONFIG_LOG_BUF_SHIFT - PRB_AVGBITS,
506 PRB_AVGBITS, &__log_buf[0]);
508 static struct printk_ringbuffer printk_rb_dynamic;
510 static struct printk_ringbuffer *prb = &printk_rb_static;
513 * We cannot access per-CPU data (e.g. per-CPU flush irq_work) before
514 * per_cpu_areas are initialised. This variable is set to true when
515 * it's safe to access per-CPU data.
517 static bool __printk_percpu_data_ready __ro_after_init;
519 bool printk_percpu_data_ready(void)
521 return __printk_percpu_data_ready;
524 /* Must be called under syslog_lock. */
525 static void latched_seq_write(struct latched_seq *ls, u64 val)
527 raw_write_seqcount_latch(&ls->latch);
529 raw_write_seqcount_latch(&ls->latch);
533 /* Can be called from any context. */
534 static u64 latched_seq_read_nolock(struct latched_seq *ls)
541 seq = raw_read_seqcount_latch(&ls->latch);
544 } while (read_seqcount_latch_retry(&ls->latch, seq));
549 /* Return log buffer address */
550 char *log_buf_addr_get(void)
555 /* Return log buffer size */
556 u32 log_buf_len_get(void)
562 * Define how much of the log buffer we could take at maximum. The value
563 * must be greater than two. Note that only half of the buffer is available
564 * when the index points to the middle.
566 #define MAX_LOG_TAKE_PART 4
567 static const char trunc_msg[] = "<truncated>";
569 static void truncate_msg(u16 *text_len, u16 *trunc_msg_len)
572 * The message should not take the whole buffer. Otherwise, it might
573 * get removed too soon.
575 u32 max_text_len = log_buf_len / MAX_LOG_TAKE_PART;
577 if (*text_len > max_text_len)
578 *text_len = max_text_len;
580 /* enable the warning message (if there is room) */
581 *trunc_msg_len = strlen(trunc_msg);
582 if (*text_len >= *trunc_msg_len)
583 *text_len -= *trunc_msg_len;
588 int dmesg_restrict = IS_ENABLED(CONFIG_SECURITY_DMESG_RESTRICT);
590 static int syslog_action_restricted(int type)
595 * Unless restricted, we allow "read all" and "get buffer size"
598 return type != SYSLOG_ACTION_READ_ALL &&
599 type != SYSLOG_ACTION_SIZE_BUFFER;
602 static int check_syslog_permissions(int type, int source)
605 * If this is from /proc/kmsg and we've already opened it, then we've
606 * already done the capabilities checks at open time.
608 if (source == SYSLOG_FROM_PROC && type != SYSLOG_ACTION_OPEN)
611 if (syslog_action_restricted(type)) {
612 if (capable(CAP_SYSLOG))
615 * For historical reasons, accept CAP_SYS_ADMIN too, with
618 if (capable(CAP_SYS_ADMIN)) {
619 pr_warn_once("%s (%d): Attempt to access syslog with "
620 "CAP_SYS_ADMIN but no CAP_SYSLOG "
622 current->comm, task_pid_nr(current));
628 return security_syslog(type);
631 static void append_char(char **pp, char *e, char c)
637 static ssize_t info_print_ext_header(char *buf, size_t size,
638 struct printk_info *info)
640 u64 ts_usec = info->ts_nsec;
642 #ifdef CONFIG_PRINTK_CALLER
643 u32 id = info->caller_id;
645 snprintf(caller, sizeof(caller), ",caller=%c%u",
646 id & 0x80000000 ? 'C' : 'T', id & ~0x80000000);
651 do_div(ts_usec, 1000);
653 return scnprintf(buf, size, "%u,%llu,%llu,%c%s;",
654 (info->facility << 3) | info->level, info->seq,
655 ts_usec, info->flags & LOG_CONT ? 'c' : '-', caller);
658 static ssize_t msg_add_ext_text(char *buf, size_t size,
659 const char *text, size_t text_len,
662 char *p = buf, *e = buf + size;
665 /* escape non-printable characters */
666 for (i = 0; i < text_len; i++) {
667 unsigned char c = text[i];
669 if (c < ' ' || c >= 127 || c == '\\')
670 p += scnprintf(p, e - p, "\\x%02x", c);
672 append_char(&p, e, c);
674 append_char(&p, e, endc);
679 static ssize_t msg_add_dict_text(char *buf, size_t size,
680 const char *key, const char *val)
682 size_t val_len = strlen(val);
688 len = msg_add_ext_text(buf, size, "", 0, ' '); /* dict prefix */
689 len += msg_add_ext_text(buf + len, size - len, key, strlen(key), '=');
690 len += msg_add_ext_text(buf + len, size - len, val, val_len, '\n');
695 static ssize_t msg_print_ext_body(char *buf, size_t size,
696 char *text, size_t text_len,
697 struct dev_printk_info *dev_info)
701 len = msg_add_ext_text(buf, size, text, text_len, '\n');
706 len += msg_add_dict_text(buf + len, size - len, "SUBSYSTEM",
707 dev_info->subsystem);
708 len += msg_add_dict_text(buf + len, size - len, "DEVICE",
714 /* /dev/kmsg - userspace message inject/listen interface */
715 struct devkmsg_user {
717 struct ratelimit_state rs;
719 char buf[CONSOLE_EXT_LOG_MAX];
721 struct printk_info info;
722 char text_buf[CONSOLE_EXT_LOG_MAX];
723 struct printk_record record;
726 static __printf(3, 4) __cold
727 int devkmsg_emit(int facility, int level, const char *fmt, ...)
733 r = vprintk_emit(facility, level, NULL, fmt, args);
739 static ssize_t devkmsg_write(struct kiocb *iocb, struct iov_iter *from)
742 int level = default_message_loglevel;
743 int facility = 1; /* LOG_USER */
744 struct file *file = iocb->ki_filp;
745 struct devkmsg_user *user = file->private_data;
746 size_t len = iov_iter_count(from);
749 if (!user || len > LOG_LINE_MAX)
752 /* Ignore when user logging is disabled. */
753 if (devkmsg_log & DEVKMSG_LOG_MASK_OFF)
756 /* Ratelimit when not explicitly enabled. */
757 if (!(devkmsg_log & DEVKMSG_LOG_MASK_ON)) {
758 if (!___ratelimit(&user->rs, current->comm))
762 buf = kmalloc(len+1, GFP_KERNEL);
767 if (!copy_from_iter_full(buf, len, from)) {
773 * Extract and skip the syslog prefix <[0-9]*>. Coming from userspace
774 * the decimal value represents 32bit, the lower 3 bit are the log
775 * level, the rest are the log facility.
777 * If no prefix or no userspace facility is specified, we
778 * enforce LOG_USER, to be able to reliably distinguish
779 * kernel-generated messages from userspace-injected ones.
782 if (line[0] == '<') {
786 u = simple_strtoul(line + 1, &endp, 10);
787 if (endp && endp[0] == '>') {
788 level = LOG_LEVEL(u);
789 if (LOG_FACILITY(u) != 0)
790 facility = LOG_FACILITY(u);
796 devkmsg_emit(facility, level, "%s", line);
801 static ssize_t devkmsg_read(struct file *file, char __user *buf,
802 size_t count, loff_t *ppos)
804 struct devkmsg_user *user = file->private_data;
805 struct printk_record *r = &user->record;
812 ret = mutex_lock_interruptible(&user->lock);
816 if (!prb_read_valid(prb, atomic64_read(&user->seq), r)) {
817 if (file->f_flags & O_NONBLOCK) {
823 * Guarantee this task is visible on the waitqueue before
824 * checking the wake condition.
826 * The full memory barrier within set_current_state() of
827 * prepare_to_wait_event() pairs with the full memory barrier
828 * within wq_has_sleeper().
830 * This pairs with __wake_up_klogd:A.
832 ret = wait_event_interruptible(log_wait,
834 atomic64_read(&user->seq), r)); /* LMM(devkmsg_read:A) */
839 if (r->info->seq != atomic64_read(&user->seq)) {
840 /* our last seen message is gone, return error and reset */
841 atomic64_set(&user->seq, r->info->seq);
846 len = info_print_ext_header(user->buf, sizeof(user->buf), r->info);
847 len += msg_print_ext_body(user->buf + len, sizeof(user->buf) - len,
848 &r->text_buf[0], r->info->text_len,
851 atomic64_set(&user->seq, r->info->seq + 1);
858 if (copy_to_user(buf, user->buf, len)) {
864 mutex_unlock(&user->lock);
869 * Be careful when modifying this function!!!
871 * Only few operations are supported because the device works only with the
872 * entire variable length messages (records). Non-standard values are
873 * returned in the other cases and has been this way for quite some time.
874 * User space applications might depend on this behavior.
876 static loff_t devkmsg_llseek(struct file *file, loff_t offset, int whence)
878 struct devkmsg_user *user = file->private_data;
888 /* the first record */
889 atomic64_set(&user->seq, prb_first_valid_seq(prb));
893 * The first record after the last SYSLOG_ACTION_CLEAR,
894 * like issued by 'dmesg -c'. Reading /dev/kmsg itself
895 * changes no global state, and does not clear anything.
897 atomic64_set(&user->seq, latched_seq_read_nolock(&clear_seq));
900 /* after the last record */
901 atomic64_set(&user->seq, prb_next_seq(prb));
909 static __poll_t devkmsg_poll(struct file *file, poll_table *wait)
911 struct devkmsg_user *user = file->private_data;
912 struct printk_info info;
916 return EPOLLERR|EPOLLNVAL;
918 poll_wait(file, &log_wait, wait);
920 if (prb_read_valid_info(prb, atomic64_read(&user->seq), &info, NULL)) {
921 /* return error when data has vanished underneath us */
922 if (info.seq != atomic64_read(&user->seq))
923 ret = EPOLLIN|EPOLLRDNORM|EPOLLERR|EPOLLPRI;
925 ret = EPOLLIN|EPOLLRDNORM;
931 static int devkmsg_open(struct inode *inode, struct file *file)
933 struct devkmsg_user *user;
936 if (devkmsg_log & DEVKMSG_LOG_MASK_OFF)
939 /* write-only does not need any file context */
940 if ((file->f_flags & O_ACCMODE) != O_WRONLY) {
941 err = check_syslog_permissions(SYSLOG_ACTION_READ_ALL,
947 user = kvmalloc(sizeof(struct devkmsg_user), GFP_KERNEL);
951 ratelimit_default_init(&user->rs);
952 ratelimit_set_flags(&user->rs, RATELIMIT_MSG_ON_RELEASE);
954 mutex_init(&user->lock);
956 prb_rec_init_rd(&user->record, &user->info,
957 &user->text_buf[0], sizeof(user->text_buf));
959 atomic64_set(&user->seq, prb_first_valid_seq(prb));
961 file->private_data = user;
965 static int devkmsg_release(struct inode *inode, struct file *file)
967 struct devkmsg_user *user = file->private_data;
972 ratelimit_state_exit(&user->rs);
974 mutex_destroy(&user->lock);
979 const struct file_operations kmsg_fops = {
980 .open = devkmsg_open,
981 .read = devkmsg_read,
982 .write_iter = devkmsg_write,
983 .llseek = devkmsg_llseek,
984 .poll = devkmsg_poll,
985 .release = devkmsg_release,
988 #ifdef CONFIG_CRASH_CORE
990 * This appends the listed symbols to /proc/vmcore
992 * /proc/vmcore is used by various utilities, like crash and makedumpfile to
993 * obtain access to symbols that are otherwise very difficult to locate. These
994 * symbols are specifically used so that utilities can access and extract the
995 * dmesg log from a vmcore file after a crash.
997 void log_buf_vmcoreinfo_setup(void)
999 struct dev_printk_info *dev_info = NULL;
1001 VMCOREINFO_SYMBOL(prb);
1002 VMCOREINFO_SYMBOL(printk_rb_static);
1003 VMCOREINFO_SYMBOL(clear_seq);
1006 * Export struct size and field offsets. User space tools can
1007 * parse it and detect any changes to structure down the line.
1010 VMCOREINFO_STRUCT_SIZE(printk_ringbuffer);
1011 VMCOREINFO_OFFSET(printk_ringbuffer, desc_ring);
1012 VMCOREINFO_OFFSET(printk_ringbuffer, text_data_ring);
1013 VMCOREINFO_OFFSET(printk_ringbuffer, fail);
1015 VMCOREINFO_STRUCT_SIZE(prb_desc_ring);
1016 VMCOREINFO_OFFSET(prb_desc_ring, count_bits);
1017 VMCOREINFO_OFFSET(prb_desc_ring, descs);
1018 VMCOREINFO_OFFSET(prb_desc_ring, infos);
1019 VMCOREINFO_OFFSET(prb_desc_ring, head_id);
1020 VMCOREINFO_OFFSET(prb_desc_ring, tail_id);
1022 VMCOREINFO_STRUCT_SIZE(prb_desc);
1023 VMCOREINFO_OFFSET(prb_desc, state_var);
1024 VMCOREINFO_OFFSET(prb_desc, text_blk_lpos);
1026 VMCOREINFO_STRUCT_SIZE(prb_data_blk_lpos);
1027 VMCOREINFO_OFFSET(prb_data_blk_lpos, begin);
1028 VMCOREINFO_OFFSET(prb_data_blk_lpos, next);
1030 VMCOREINFO_STRUCT_SIZE(printk_info);
1031 VMCOREINFO_OFFSET(printk_info, seq);
1032 VMCOREINFO_OFFSET(printk_info, ts_nsec);
1033 VMCOREINFO_OFFSET(printk_info, text_len);
1034 VMCOREINFO_OFFSET(printk_info, caller_id);
1035 VMCOREINFO_OFFSET(printk_info, dev_info);
1037 VMCOREINFO_STRUCT_SIZE(dev_printk_info);
1038 VMCOREINFO_OFFSET(dev_printk_info, subsystem);
1039 VMCOREINFO_LENGTH(printk_info_subsystem, sizeof(dev_info->subsystem));
1040 VMCOREINFO_OFFSET(dev_printk_info, device);
1041 VMCOREINFO_LENGTH(printk_info_device, sizeof(dev_info->device));
1043 VMCOREINFO_STRUCT_SIZE(prb_data_ring);
1044 VMCOREINFO_OFFSET(prb_data_ring, size_bits);
1045 VMCOREINFO_OFFSET(prb_data_ring, data);
1046 VMCOREINFO_OFFSET(prb_data_ring, head_lpos);
1047 VMCOREINFO_OFFSET(prb_data_ring, tail_lpos);
1049 VMCOREINFO_SIZE(atomic_long_t);
1050 VMCOREINFO_TYPE_OFFSET(atomic_long_t, counter);
1052 VMCOREINFO_STRUCT_SIZE(latched_seq);
1053 VMCOREINFO_OFFSET(latched_seq, val);
1057 /* requested log_buf_len from kernel cmdline */
1058 static unsigned long __initdata new_log_buf_len;
1060 /* we practice scaling the ring buffer by powers of 2 */
1061 static void __init log_buf_len_update(u64 size)
1063 if (size > (u64)LOG_BUF_LEN_MAX) {
1064 size = (u64)LOG_BUF_LEN_MAX;
1065 pr_err("log_buf over 2G is not supported.\n");
1069 size = roundup_pow_of_two(size);
1070 if (size > log_buf_len)
1071 new_log_buf_len = (unsigned long)size;
1074 /* save requested log_buf_len since it's too early to process it */
1075 static int __init log_buf_len_setup(char *str)
1082 size = memparse(str, &str);
1084 log_buf_len_update(size);
1088 early_param("log_buf_len", log_buf_len_setup);
1091 #define __LOG_CPU_MAX_BUF_LEN (1 << CONFIG_LOG_CPU_MAX_BUF_SHIFT)
1093 static void __init log_buf_add_cpu(void)
1095 unsigned int cpu_extra;
1098 * archs should set up cpu_possible_bits properly with
1099 * set_cpu_possible() after setup_arch() but just in
1100 * case lets ensure this is valid.
1102 if (num_possible_cpus() == 1)
1105 cpu_extra = (num_possible_cpus() - 1) * __LOG_CPU_MAX_BUF_LEN;
1107 /* by default this will only continue through for large > 64 CPUs */
1108 if (cpu_extra <= __LOG_BUF_LEN / 2)
1111 pr_info("log_buf_len individual max cpu contribution: %d bytes\n",
1112 __LOG_CPU_MAX_BUF_LEN);
1113 pr_info("log_buf_len total cpu_extra contributions: %d bytes\n",
1115 pr_info("log_buf_len min size: %d bytes\n", __LOG_BUF_LEN);
1117 log_buf_len_update(cpu_extra + __LOG_BUF_LEN);
1119 #else /* !CONFIG_SMP */
1120 static inline void log_buf_add_cpu(void) {}
1121 #endif /* CONFIG_SMP */
1123 static void __init set_percpu_data_ready(void)
1125 __printk_percpu_data_ready = true;
1128 static unsigned int __init add_to_rb(struct printk_ringbuffer *rb,
1129 struct printk_record *r)
1131 struct prb_reserved_entry e;
1132 struct printk_record dest_r;
1134 prb_rec_init_wr(&dest_r, r->info->text_len);
1136 if (!prb_reserve(&e, rb, &dest_r))
1139 memcpy(&dest_r.text_buf[0], &r->text_buf[0], r->info->text_len);
1140 dest_r.info->text_len = r->info->text_len;
1141 dest_r.info->facility = r->info->facility;
1142 dest_r.info->level = r->info->level;
1143 dest_r.info->flags = r->info->flags;
1144 dest_r.info->ts_nsec = r->info->ts_nsec;
1145 dest_r.info->caller_id = r->info->caller_id;
1146 memcpy(&dest_r.info->dev_info, &r->info->dev_info, sizeof(dest_r.info->dev_info));
1148 prb_final_commit(&e);
1150 return prb_record_text_space(&e);
1153 static char setup_text_buf[LOG_LINE_MAX] __initdata;
1155 void __init setup_log_buf(int early)
1157 struct printk_info *new_infos;
1158 unsigned int new_descs_count;
1159 struct prb_desc *new_descs;
1160 struct printk_info info;
1161 struct printk_record r;
1162 unsigned int text_size;
1163 size_t new_descs_size;
1164 size_t new_infos_size;
1165 unsigned long flags;
1171 * Some archs call setup_log_buf() multiple times - first is very
1172 * early, e.g. from setup_arch(), and second - when percpu_areas
1176 set_percpu_data_ready();
1178 if (log_buf != __log_buf)
1181 if (!early && !new_log_buf_len)
1184 if (!new_log_buf_len)
1187 new_descs_count = new_log_buf_len >> PRB_AVGBITS;
1188 if (new_descs_count == 0) {
1189 pr_err("new_log_buf_len: %lu too small\n", new_log_buf_len);
1193 new_log_buf = memblock_alloc(new_log_buf_len, LOG_ALIGN);
1194 if (unlikely(!new_log_buf)) {
1195 pr_err("log_buf_len: %lu text bytes not available\n",
1200 new_descs_size = new_descs_count * sizeof(struct prb_desc);
1201 new_descs = memblock_alloc(new_descs_size, LOG_ALIGN);
1202 if (unlikely(!new_descs)) {
1203 pr_err("log_buf_len: %zu desc bytes not available\n",
1205 goto err_free_log_buf;
1208 new_infos_size = new_descs_count * sizeof(struct printk_info);
1209 new_infos = memblock_alloc(new_infos_size, LOG_ALIGN);
1210 if (unlikely(!new_infos)) {
1211 pr_err("log_buf_len: %zu info bytes not available\n",
1213 goto err_free_descs;
1216 prb_rec_init_rd(&r, &info, &setup_text_buf[0], sizeof(setup_text_buf));
1218 prb_init(&printk_rb_dynamic,
1219 new_log_buf, ilog2(new_log_buf_len),
1220 new_descs, ilog2(new_descs_count),
1223 local_irq_save(flags);
1225 log_buf_len = new_log_buf_len;
1226 log_buf = new_log_buf;
1227 new_log_buf_len = 0;
1229 free = __LOG_BUF_LEN;
1230 prb_for_each_record(0, &printk_rb_static, seq, &r) {
1231 text_size = add_to_rb(&printk_rb_dynamic, &r);
1232 if (text_size > free)
1238 prb = &printk_rb_dynamic;
1240 local_irq_restore(flags);
1243 * Copy any remaining messages that might have appeared from
1244 * NMI context after copying but before switching to the
1247 prb_for_each_record(seq, &printk_rb_static, seq, &r) {
1248 text_size = add_to_rb(&printk_rb_dynamic, &r);
1249 if (text_size > free)
1255 if (seq != prb_next_seq(&printk_rb_static)) {
1256 pr_err("dropped %llu messages\n",
1257 prb_next_seq(&printk_rb_static) - seq);
1260 pr_info("log_buf_len: %u bytes\n", log_buf_len);
1261 pr_info("early log buf free: %u(%u%%)\n",
1262 free, (free * 100) / __LOG_BUF_LEN);
1266 memblock_free(new_descs, new_descs_size);
1268 memblock_free(new_log_buf, new_log_buf_len);
1271 static bool __read_mostly ignore_loglevel;
1273 static int __init ignore_loglevel_setup(char *str)
1275 ignore_loglevel = true;
1276 pr_info("debug: ignoring loglevel setting.\n");
1281 early_param("ignore_loglevel", ignore_loglevel_setup);
1282 module_param(ignore_loglevel, bool, S_IRUGO | S_IWUSR);
1283 MODULE_PARM_DESC(ignore_loglevel,
1284 "ignore loglevel setting (prints all kernel messages to the console)");
1286 static bool suppress_message_printing(int level)
1288 return (level >= console_loglevel && !ignore_loglevel);
1291 #ifdef CONFIG_BOOT_PRINTK_DELAY
1293 static int boot_delay; /* msecs delay after each printk during bootup */
1294 static unsigned long long loops_per_msec; /* based on boot_delay */
1296 static int __init boot_delay_setup(char *str)
1300 lpj = preset_lpj ? preset_lpj : 1000000; /* some guess */
1301 loops_per_msec = (unsigned long long)lpj / 1000 * HZ;
1303 get_option(&str, &boot_delay);
1304 if (boot_delay > 10 * 1000)
1307 pr_debug("boot_delay: %u, preset_lpj: %ld, lpj: %lu, "
1308 "HZ: %d, loops_per_msec: %llu\n",
1309 boot_delay, preset_lpj, lpj, HZ, loops_per_msec);
1312 early_param("boot_delay", boot_delay_setup);
1314 static void boot_delay_msec(int level)
1316 unsigned long long k;
1317 unsigned long timeout;
1319 if ((boot_delay == 0 || system_state >= SYSTEM_RUNNING)
1320 || suppress_message_printing(level)) {
1324 k = (unsigned long long)loops_per_msec * boot_delay;
1326 timeout = jiffies + msecs_to_jiffies(boot_delay);
1331 * use (volatile) jiffies to prevent
1332 * compiler reduction; loop termination via jiffies
1333 * is secondary and may or may not happen.
1335 if (time_after(jiffies, timeout))
1337 touch_nmi_watchdog();
1341 static inline void boot_delay_msec(int level)
1346 static bool printk_time = IS_ENABLED(CONFIG_PRINTK_TIME);
1347 module_param_named(time, printk_time, bool, S_IRUGO | S_IWUSR);
1349 static size_t print_syslog(unsigned int level, char *buf)
1351 return sprintf(buf, "<%u>", level);
1354 static size_t print_time(u64 ts, char *buf)
1356 unsigned long rem_nsec = do_div(ts, 1000000000);
1358 return sprintf(buf, "[%5lu.%06lu]",
1359 (unsigned long)ts, rem_nsec / 1000);
1362 #ifdef CONFIG_PRINTK_CALLER
1363 static size_t print_caller(u32 id, char *buf)
1367 snprintf(caller, sizeof(caller), "%c%u",
1368 id & 0x80000000 ? 'C' : 'T', id & ~0x80000000);
1369 return sprintf(buf, "[%6s]", caller);
1372 #define print_caller(id, buf) 0
1375 static size_t info_print_prefix(const struct printk_info *info, bool syslog,
1376 bool time, char *buf)
1381 len = print_syslog((info->facility << 3) | info->level, buf);
1384 len += print_time(info->ts_nsec, buf + len);
1386 len += print_caller(info->caller_id, buf + len);
1388 if (IS_ENABLED(CONFIG_PRINTK_CALLER) || time) {
1397 * Prepare the record for printing. The text is shifted within the given
1398 * buffer to avoid a need for another one. The following operations are
1401 * - Add prefix for each line.
1402 * - Drop truncated lines that no longer fit into the buffer.
1403 * - Add the trailing newline that has been removed in vprintk_store().
1404 * - Add a string terminator.
1406 * Since the produced string is always terminated, the maximum possible
1407 * return value is @r->text_buf_size - 1;
1409 * Return: The length of the updated/prepared text, including the added
1410 * prefixes and the newline. The terminator is not counted. The dropped
1411 * line(s) are not counted.
1413 static size_t record_print_text(struct printk_record *r, bool syslog,
1416 size_t text_len = r->info->text_len;
1417 size_t buf_size = r->text_buf_size;
1418 char *text = r->text_buf;
1419 char prefix[PREFIX_MAX];
1420 bool truncated = false;
1427 * If the message was truncated because the buffer was not large
1428 * enough, treat the available text as if it were the full text.
1430 if (text_len > buf_size)
1431 text_len = buf_size;
1433 prefix_len = info_print_prefix(r->info, syslog, time, prefix);
1436 * @text_len: bytes of unprocessed text
1437 * @line_len: bytes of current line _without_ newline
1438 * @text: pointer to beginning of current line
1439 * @len: number of bytes prepared in r->text_buf
1442 next = memchr(text, '\n', text_len);
1444 line_len = next - text;
1446 /* Drop truncated line(s). */
1449 line_len = text_len;
1453 * Truncate the text if there is not enough space to add the
1454 * prefix and a trailing newline and a terminator.
1456 if (len + prefix_len + text_len + 1 + 1 > buf_size) {
1457 /* Drop even the current line if no space. */
1458 if (len + prefix_len + line_len + 1 + 1 > buf_size)
1461 text_len = buf_size - len - prefix_len - 1 - 1;
1465 memmove(text + prefix_len, text, text_len);
1466 memcpy(text, prefix, prefix_len);
1469 * Increment the prepared length to include the text and
1470 * prefix that were just moved+copied. Also increment for the
1471 * newline at the end of this line. If this is the last line,
1472 * there is no newline, but it will be added immediately below.
1474 len += prefix_len + line_len + 1;
1475 if (text_len == line_len) {
1477 * This is the last line. Add the trailing newline
1478 * removed in vprintk_store().
1480 text[prefix_len + line_len] = '\n';
1485 * Advance beyond the added prefix and the related line with
1488 text += prefix_len + line_len + 1;
1491 * The remaining text has only decreased by the line with its
1494 * Note that @text_len can become zero. It happens when @text
1495 * ended with a newline (either due to truncation or the
1496 * original string ending with "\n\n"). The loop is correctly
1497 * repeated and (if not truncated) an empty line with a prefix
1500 text_len -= line_len + 1;
1504 * If a buffer was provided, it will be terminated. Space for the
1505 * string terminator is guaranteed to be available. The terminator is
1506 * not counted in the return value.
1509 r->text_buf[len] = 0;
1514 static size_t get_record_print_text_size(struct printk_info *info,
1515 unsigned int line_count,
1516 bool syslog, bool time)
1518 char prefix[PREFIX_MAX];
1521 prefix_len = info_print_prefix(info, syslog, time, prefix);
1524 * Each line will be preceded with a prefix. The intermediate
1525 * newlines are already within the text, but a final trailing
1526 * newline will be added.
1528 return ((prefix_len * line_count) + info->text_len + 1);
1532 * Beginning with @start_seq, find the first record where it and all following
1533 * records up to (but not including) @max_seq fit into @size.
1535 * @max_seq is simply an upper bound and does not need to exist. If the caller
1536 * does not require an upper bound, -1 can be used for @max_seq.
1538 static u64 find_first_fitting_seq(u64 start_seq, u64 max_seq, size_t size,
1539 bool syslog, bool time)
1541 struct printk_info info;
1542 unsigned int line_count;
1546 /* Determine the size of the records up to @max_seq. */
1547 prb_for_each_info(start_seq, prb, seq, &info, &line_count) {
1548 if (info.seq >= max_seq)
1550 len += get_record_print_text_size(&info, line_count, syslog, time);
1554 * Adjust the upper bound for the next loop to avoid subtracting
1555 * lengths that were never added.
1561 * Move first record forward until length fits into the buffer. Ignore
1562 * newest messages that were not counted in the above cycle. Messages
1563 * might appear and get lost in the meantime. This is a best effort
1564 * that prevents an infinite loop that could occur with a retry.
1566 prb_for_each_info(start_seq, prb, seq, &info, &line_count) {
1567 if (len <= size || info.seq >= max_seq)
1569 len -= get_record_print_text_size(&info, line_count, syslog, time);
1575 /* The caller is responsible for making sure @size is greater than 0. */
1576 static int syslog_print(char __user *buf, int size)
1578 struct printk_info info;
1579 struct printk_record r;
1584 text = kmalloc(CONSOLE_LOG_MAX, GFP_KERNEL);
1588 prb_rec_init_rd(&r, &info, text, CONSOLE_LOG_MAX);
1590 mutex_lock(&syslog_lock);
1593 * Wait for the @syslog_seq record to be available. @syslog_seq may
1594 * change while waiting.
1599 mutex_unlock(&syslog_lock);
1601 * Guarantee this task is visible on the waitqueue before
1602 * checking the wake condition.
1604 * The full memory barrier within set_current_state() of
1605 * prepare_to_wait_event() pairs with the full memory barrier
1606 * within wq_has_sleeper().
1608 * This pairs with __wake_up_klogd:A.
1610 len = wait_event_interruptible(log_wait,
1611 prb_read_valid(prb, seq, NULL)); /* LMM(syslog_print:A) */
1612 mutex_lock(&syslog_lock);
1616 } while (syslog_seq != seq);
1619 * Copy records that fit into the buffer. The above cycle makes sure
1620 * that the first record is always available.
1627 if (!prb_read_valid(prb, syslog_seq, &r))
1630 if (r.info->seq != syslog_seq) {
1631 /* message is gone, move to next valid one */
1632 syslog_seq = r.info->seq;
1637 * To keep reading/counting partial line consistent,
1638 * use printk_time value as of the beginning of a line.
1640 if (!syslog_partial)
1641 syslog_time = printk_time;
1643 skip = syslog_partial;
1644 n = record_print_text(&r, true, syslog_time);
1645 if (n - syslog_partial <= size) {
1646 /* message fits into buffer, move forward */
1647 syslog_seq = r.info->seq + 1;
1648 n -= syslog_partial;
1651 /* partial read(), remember position */
1653 syslog_partial += n;
1660 mutex_unlock(&syslog_lock);
1661 err = copy_to_user(buf, text + skip, n);
1662 mutex_lock(&syslog_lock);
1675 mutex_unlock(&syslog_lock);
1680 static int syslog_print_all(char __user *buf, int size, bool clear)
1682 struct printk_info info;
1683 struct printk_record r;
1689 text = kmalloc(CONSOLE_LOG_MAX, GFP_KERNEL);
1695 * Find first record that fits, including all following records,
1696 * into the user-provided buffer for this dump.
1698 seq = find_first_fitting_seq(latched_seq_read_nolock(&clear_seq), -1,
1701 prb_rec_init_rd(&r, &info, text, CONSOLE_LOG_MAX);
1704 prb_for_each_record(seq, prb, seq, &r) {
1707 textlen = record_print_text(&r, true, time);
1709 if (len + textlen > size) {
1714 if (copy_to_user(buf + len, text, textlen))
1724 mutex_lock(&syslog_lock);
1725 latched_seq_write(&clear_seq, seq);
1726 mutex_unlock(&syslog_lock);
1733 static void syslog_clear(void)
1735 mutex_lock(&syslog_lock);
1736 latched_seq_write(&clear_seq, prb_next_seq(prb));
1737 mutex_unlock(&syslog_lock);
1740 int do_syslog(int type, char __user *buf, int len, int source)
1742 struct printk_info info;
1744 static int saved_console_loglevel = LOGLEVEL_DEFAULT;
1747 error = check_syslog_permissions(type, source);
1752 case SYSLOG_ACTION_CLOSE: /* Close log */
1754 case SYSLOG_ACTION_OPEN: /* Open log */
1756 case SYSLOG_ACTION_READ: /* Read from log */
1757 if (!buf || len < 0)
1761 if (!access_ok(buf, len))
1763 error = syslog_print(buf, len);
1765 /* Read/clear last kernel messages */
1766 case SYSLOG_ACTION_READ_CLEAR:
1769 /* Read last kernel messages */
1770 case SYSLOG_ACTION_READ_ALL:
1771 if (!buf || len < 0)
1775 if (!access_ok(buf, len))
1777 error = syslog_print_all(buf, len, clear);
1779 /* Clear ring buffer */
1780 case SYSLOG_ACTION_CLEAR:
1783 /* Disable logging to console */
1784 case SYSLOG_ACTION_CONSOLE_OFF:
1785 if (saved_console_loglevel == LOGLEVEL_DEFAULT)
1786 saved_console_loglevel = console_loglevel;
1787 console_loglevel = minimum_console_loglevel;
1789 /* Enable logging to console */
1790 case SYSLOG_ACTION_CONSOLE_ON:
1791 if (saved_console_loglevel != LOGLEVEL_DEFAULT) {
1792 console_loglevel = saved_console_loglevel;
1793 saved_console_loglevel = LOGLEVEL_DEFAULT;
1796 /* Set level of messages printed to console */
1797 case SYSLOG_ACTION_CONSOLE_LEVEL:
1798 if (len < 1 || len > 8)
1800 if (len < minimum_console_loglevel)
1801 len = minimum_console_loglevel;
1802 console_loglevel = len;
1803 /* Implicitly re-enable logging to console */
1804 saved_console_loglevel = LOGLEVEL_DEFAULT;
1806 /* Number of chars in the log buffer */
1807 case SYSLOG_ACTION_SIZE_UNREAD:
1808 mutex_lock(&syslog_lock);
1809 if (!prb_read_valid_info(prb, syslog_seq, &info, NULL)) {
1810 /* No unread messages. */
1811 mutex_unlock(&syslog_lock);
1814 if (info.seq != syslog_seq) {
1815 /* messages are gone, move to first one */
1816 syslog_seq = info.seq;
1819 if (source == SYSLOG_FROM_PROC) {
1821 * Short-cut for poll(/"proc/kmsg") which simply checks
1822 * for pending data, not the size; return the count of
1823 * records, not the length.
1825 error = prb_next_seq(prb) - syslog_seq;
1827 bool time = syslog_partial ? syslog_time : printk_time;
1828 unsigned int line_count;
1831 prb_for_each_info(syslog_seq, prb, seq, &info,
1833 error += get_record_print_text_size(&info, line_count,
1837 error -= syslog_partial;
1839 mutex_unlock(&syslog_lock);
1841 /* Size of the log buffer */
1842 case SYSLOG_ACTION_SIZE_BUFFER:
1843 error = log_buf_len;
1853 SYSCALL_DEFINE3(syslog, int, type, char __user *, buf, int, len)
1855 return do_syslog(type, buf, len, SYSLOG_FROM_READER);
1859 * Special console_lock variants that help to reduce the risk of soft-lockups.
1860 * They allow to pass console_lock to another printk() call using a busy wait.
1863 #ifdef CONFIG_LOCKDEP
1864 static struct lockdep_map console_owner_dep_map = {
1865 .name = "console_owner"
1869 static DEFINE_RAW_SPINLOCK(console_owner_lock);
1870 static struct task_struct *console_owner;
1871 static bool console_waiter;
1874 * console_lock_spinning_enable - mark beginning of code where another
1875 * thread might safely busy wait
1877 * This basically converts console_lock into a spinlock. This marks
1878 * the section where the console_lock owner can not sleep, because
1879 * there may be a waiter spinning (like a spinlock). Also it must be
1880 * ready to hand over the lock at the end of the section.
1882 static void console_lock_spinning_enable(void)
1884 raw_spin_lock(&console_owner_lock);
1885 console_owner = current;
1886 raw_spin_unlock(&console_owner_lock);
1888 /* The waiter may spin on us after setting console_owner */
1889 spin_acquire(&console_owner_dep_map, 0, 0, _THIS_IP_);
1893 * console_lock_spinning_disable_and_check - mark end of code where another
1894 * thread was able to busy wait and check if there is a waiter
1895 * @cookie: cookie returned from console_srcu_read_lock()
1897 * This is called at the end of the section where spinning is allowed.
1898 * It has two functions. First, it is a signal that it is no longer
1899 * safe to start busy waiting for the lock. Second, it checks if
1900 * there is a busy waiter and passes the lock rights to her.
1902 * Important: Callers lose both the console_lock and the SRCU read lock if
1903 * there was a busy waiter. They must not touch items synchronized by
1904 * console_lock or SRCU read lock in this case.
1906 * Return: 1 if the lock rights were passed, 0 otherwise.
1908 static int console_lock_spinning_disable_and_check(int cookie)
1912 raw_spin_lock(&console_owner_lock);
1913 waiter = READ_ONCE(console_waiter);
1914 console_owner = NULL;
1915 raw_spin_unlock(&console_owner_lock);
1918 spin_release(&console_owner_dep_map, _THIS_IP_);
1922 /* The waiter is now free to continue */
1923 WRITE_ONCE(console_waiter, false);
1925 spin_release(&console_owner_dep_map, _THIS_IP_);
1928 * Preserve lockdep lock ordering. Release the SRCU read lock before
1929 * releasing the console_lock.
1931 console_srcu_read_unlock(cookie);
1934 * Hand off console_lock to waiter. The waiter will perform
1935 * the up(). After this, the waiter is the console_lock owner.
1937 mutex_release(&console_lock_dep_map, _THIS_IP_);
1942 * console_trylock_spinning - try to get console_lock by busy waiting
1944 * This allows to busy wait for the console_lock when the current
1945 * owner is running in specially marked sections. It means that
1946 * the current owner is running and cannot reschedule until it
1947 * is ready to lose the lock.
1949 * Return: 1 if we got the lock, 0 othrewise
1951 static int console_trylock_spinning(void)
1953 struct task_struct *owner = NULL;
1956 unsigned long flags;
1958 if (console_trylock())
1962 * It's unsafe to spin once a panic has begun. If we are the
1963 * panic CPU, we may have already halted the owner of the
1964 * console_sem. If we are not the panic CPU, then we should
1965 * avoid taking console_sem, so the panic CPU has a better
1966 * chance of cleanly acquiring it later.
1968 if (panic_in_progress())
1971 printk_safe_enter_irqsave(flags);
1973 raw_spin_lock(&console_owner_lock);
1974 owner = READ_ONCE(console_owner);
1975 waiter = READ_ONCE(console_waiter);
1976 if (!waiter && owner && owner != current) {
1977 WRITE_ONCE(console_waiter, true);
1980 raw_spin_unlock(&console_owner_lock);
1983 * If there is an active printk() writing to the
1984 * consoles, instead of having it write our data too,
1985 * see if we can offload that load from the active
1986 * printer, and do some printing ourselves.
1987 * Go into a spin only if there isn't already a waiter
1988 * spinning, and there is an active printer, and
1989 * that active printer isn't us (recursive printk?).
1992 printk_safe_exit_irqrestore(flags);
1996 /* We spin waiting for the owner to release us */
1997 spin_acquire(&console_owner_dep_map, 0, 0, _THIS_IP_);
1998 /* Owner will clear console_waiter on hand off */
1999 while (READ_ONCE(console_waiter))
2001 spin_release(&console_owner_dep_map, _THIS_IP_);
2003 printk_safe_exit_irqrestore(flags);
2005 * The owner passed the console lock to us.
2006 * Since we did not spin on console lock, annotate
2007 * this as a trylock. Otherwise lockdep will
2010 mutex_acquire(&console_lock_dep_map, 0, 1, _THIS_IP_);
2016 * Call the specified console driver, asking it to write out the specified
2017 * text and length. If @dropped_text is non-NULL and any records have been
2018 * dropped, a dropped message will be written out first.
2020 static void call_console_driver(struct console *con, const char *text, size_t len,
2025 if (con->dropped && dropped_text) {
2026 dropped_len = snprintf(dropped_text, DROPPED_TEXT_MAX,
2027 "** %lu printk messages dropped **\n",
2030 con->write(con, dropped_text, dropped_len);
2033 con->write(con, text, len);
2037 * Recursion is tracked separately on each CPU. If NMIs are supported, an
2038 * additional NMI context per CPU is also separately tracked. Until per-CPU
2039 * is available, a separate "early tracking" is performed.
2041 static DEFINE_PER_CPU(u8, printk_count);
2042 static u8 printk_count_early;
2043 #ifdef CONFIG_HAVE_NMI
2044 static DEFINE_PER_CPU(u8, printk_count_nmi);
2045 static u8 printk_count_nmi_early;
2049 * Recursion is limited to keep the output sane. printk() should not require
2050 * more than 1 level of recursion (allowing, for example, printk() to trigger
2051 * a WARN), but a higher value is used in case some printk-internal errors
2052 * exist, such as the ringbuffer validation checks failing.
2054 #define PRINTK_MAX_RECURSION 3
2057 * Return a pointer to the dedicated counter for the CPU+context of the
2060 static u8 *__printk_recursion_counter(void)
2062 #ifdef CONFIG_HAVE_NMI
2064 if (printk_percpu_data_ready())
2065 return this_cpu_ptr(&printk_count_nmi);
2066 return &printk_count_nmi_early;
2069 if (printk_percpu_data_ready())
2070 return this_cpu_ptr(&printk_count);
2071 return &printk_count_early;
2075 * Enter recursion tracking. Interrupts are disabled to simplify tracking.
2076 * The caller must check the boolean return value to see if the recursion is
2077 * allowed. On failure, interrupts are not disabled.
2079 * @recursion_ptr must be a variable of type (u8 *) and is the same variable
2080 * that is passed to printk_exit_irqrestore().
2082 #define printk_enter_irqsave(recursion_ptr, flags) \
2084 bool success = true; \
2086 typecheck(u8 *, recursion_ptr); \
2087 local_irq_save(flags); \
2088 (recursion_ptr) = __printk_recursion_counter(); \
2089 if (*(recursion_ptr) > PRINTK_MAX_RECURSION) { \
2090 local_irq_restore(flags); \
2093 (*(recursion_ptr))++; \
2098 /* Exit recursion tracking, restoring interrupts. */
2099 #define printk_exit_irqrestore(recursion_ptr, flags) \
2101 typecheck(u8 *, recursion_ptr); \
2102 (*(recursion_ptr))--; \
2103 local_irq_restore(flags); \
2106 int printk_delay_msec __read_mostly;
2108 static inline void printk_delay(int level)
2110 boot_delay_msec(level);
2112 if (unlikely(printk_delay_msec)) {
2113 int m = printk_delay_msec;
2117 touch_nmi_watchdog();
2122 static inline u32 printk_caller_id(void)
2124 return in_task() ? task_pid_nr(current) :
2125 0x80000000 + smp_processor_id();
2129 * printk_parse_prefix - Parse level and control flags.
2131 * @text: The terminated text message.
2132 * @level: A pointer to the current level value, will be updated.
2133 * @flags: A pointer to the current printk_info flags, will be updated.
2135 * @level may be NULL if the caller is not interested in the parsed value.
2136 * Otherwise the variable pointed to by @level must be set to
2137 * LOGLEVEL_DEFAULT in order to be updated with the parsed value.
2139 * @flags may be NULL if the caller is not interested in the parsed value.
2140 * Otherwise the variable pointed to by @flags will be OR'd with the parsed
2143 * Return: The length of the parsed level and control flags.
2145 u16 printk_parse_prefix(const char *text, int *level,
2146 enum printk_info_flags *flags)
2152 kern_level = printk_get_level(text);
2156 switch (kern_level) {
2158 if (level && *level == LOGLEVEL_DEFAULT)
2159 *level = kern_level - '0';
2161 case 'c': /* KERN_CONT */
2174 static u16 printk_sprint(char *text, u16 size, int facility,
2175 enum printk_info_flags *flags, const char *fmt,
2180 text_len = vscnprintf(text, size, fmt, args);
2182 /* Mark and strip a trailing newline. */
2183 if (text_len && text[text_len - 1] == '\n') {
2185 *flags |= LOG_NEWLINE;
2188 /* Strip log level and control flags. */
2189 if (facility == 0) {
2192 prefix_len = printk_parse_prefix(text, NULL, NULL);
2194 text_len -= prefix_len;
2195 memmove(text, text + prefix_len, text_len);
2199 trace_console_rcuidle(text, text_len);
2205 int vprintk_store(int facility, int level,
2206 const struct dev_printk_info *dev_info,
2207 const char *fmt, va_list args)
2209 struct prb_reserved_entry e;
2210 enum printk_info_flags flags = 0;
2211 struct printk_record r;
2212 unsigned long irqflags;
2213 u16 trunc_msg_len = 0;
2223 if (!printk_enter_irqsave(recursion_ptr, irqflags))
2227 * Since the duration of printk() can vary depending on the message
2228 * and state of the ringbuffer, grab the timestamp now so that it is
2229 * close to the call of printk(). This provides a more deterministic
2230 * timestamp with respect to the caller.
2232 ts_nsec = local_clock();
2234 caller_id = printk_caller_id();
2237 * The sprintf needs to come first since the syslog prefix might be
2238 * passed in as a parameter. An extra byte must be reserved so that
2239 * later the vscnprintf() into the reserved buffer has room for the
2240 * terminating '\0', which is not counted by vsnprintf().
2242 va_copy(args2, args);
2243 reserve_size = vsnprintf(&prefix_buf[0], sizeof(prefix_buf), fmt, args2) + 1;
2246 if (reserve_size > LOG_LINE_MAX)
2247 reserve_size = LOG_LINE_MAX;
2249 /* Extract log level or control flags. */
2251 printk_parse_prefix(&prefix_buf[0], &level, &flags);
2253 if (level == LOGLEVEL_DEFAULT)
2254 level = default_message_loglevel;
2257 flags |= LOG_NEWLINE;
2259 if (flags & LOG_CONT) {
2260 prb_rec_init_wr(&r, reserve_size);
2261 if (prb_reserve_in_last(&e, prb, &r, caller_id, LOG_LINE_MAX)) {
2262 text_len = printk_sprint(&r.text_buf[r.info->text_len], reserve_size,
2263 facility, &flags, fmt, args);
2264 r.info->text_len += text_len;
2266 if (flags & LOG_NEWLINE) {
2267 r.info->flags |= LOG_NEWLINE;
2268 prb_final_commit(&e);
2279 * Explicitly initialize the record before every prb_reserve() call.
2280 * prb_reserve_in_last() and prb_reserve() purposely invalidate the
2281 * structure when they fail.
2283 prb_rec_init_wr(&r, reserve_size);
2284 if (!prb_reserve(&e, prb, &r)) {
2285 /* truncate the message if it is too long for empty buffer */
2286 truncate_msg(&reserve_size, &trunc_msg_len);
2288 prb_rec_init_wr(&r, reserve_size + trunc_msg_len);
2289 if (!prb_reserve(&e, prb, &r))
2294 text_len = printk_sprint(&r.text_buf[0], reserve_size, facility, &flags, fmt, args);
2296 memcpy(&r.text_buf[text_len], trunc_msg, trunc_msg_len);
2297 r.info->text_len = text_len + trunc_msg_len;
2298 r.info->facility = facility;
2299 r.info->level = level & 7;
2300 r.info->flags = flags & 0x1f;
2301 r.info->ts_nsec = ts_nsec;
2302 r.info->caller_id = caller_id;
2304 memcpy(&r.info->dev_info, dev_info, sizeof(r.info->dev_info));
2306 /* A message without a trailing newline can be continued. */
2307 if (!(flags & LOG_NEWLINE))
2310 prb_final_commit(&e);
2312 ret = text_len + trunc_msg_len;
2314 printk_exit_irqrestore(recursion_ptr, irqflags);
2318 asmlinkage int vprintk_emit(int facility, int level,
2319 const struct dev_printk_info *dev_info,
2320 const char *fmt, va_list args)
2323 bool in_sched = false;
2325 /* Suppress unimportant messages after panic happens */
2326 if (unlikely(suppress_printk))
2329 if (unlikely(suppress_panic_printk) &&
2330 atomic_read(&panic_cpu) != raw_smp_processor_id())
2333 if (level == LOGLEVEL_SCHED) {
2334 level = LOGLEVEL_DEFAULT;
2338 printk_delay(level);
2340 printed_len = vprintk_store(facility, level, dev_info, fmt, args);
2342 /* If called from the scheduler, we can not call up(). */
2345 * The caller may be holding system-critical or
2346 * timing-sensitive locks. Disable preemption during
2347 * printing of all remaining records to all consoles so that
2348 * this context can return as soon as possible. Hopefully
2349 * another printk() caller will take over the printing.
2353 * Try to acquire and then immediately release the console
2354 * semaphore. The release will print out buffers. With the
2355 * spinning variant, this context tries to take over the
2356 * printing from another printing context.
2358 if (console_trylock_spinning())
2366 EXPORT_SYMBOL(vprintk_emit);
2368 int vprintk_default(const char *fmt, va_list args)
2370 return vprintk_emit(0, LOGLEVEL_DEFAULT, NULL, fmt, args);
2372 EXPORT_SYMBOL_GPL(vprintk_default);
2374 asmlinkage __visible int _printk(const char *fmt, ...)
2379 va_start(args, fmt);
2380 r = vprintk(fmt, args);
2385 EXPORT_SYMBOL(_printk);
2387 static bool pr_flush(int timeout_ms, bool reset_on_progress);
2388 static bool __pr_flush(struct console *con, int timeout_ms, bool reset_on_progress);
2390 #else /* CONFIG_PRINTK */
2392 #define CONSOLE_LOG_MAX 0
2393 #define DROPPED_TEXT_MAX 0
2394 #define printk_time false
2396 #define prb_read_valid(rb, seq, r) false
2397 #define prb_first_valid_seq(rb) 0
2398 #define prb_next_seq(rb) 0
2400 static u64 syslog_seq;
2402 static size_t record_print_text(const struct printk_record *r,
2403 bool syslog, bool time)
2407 static ssize_t info_print_ext_header(char *buf, size_t size,
2408 struct printk_info *info)
2412 static ssize_t msg_print_ext_body(char *buf, size_t size,
2413 char *text, size_t text_len,
2414 struct dev_printk_info *dev_info) { return 0; }
2415 static void console_lock_spinning_enable(void) { }
2416 static int console_lock_spinning_disable_and_check(int cookie) { return 0; }
2417 static void call_console_driver(struct console *con, const char *text, size_t len,
2421 static bool suppress_message_printing(int level) { return false; }
2422 static bool pr_flush(int timeout_ms, bool reset_on_progress) { return true; }
2423 static bool __pr_flush(struct console *con, int timeout_ms, bool reset_on_progress) { return true; }
2425 #endif /* CONFIG_PRINTK */
2427 #ifdef CONFIG_EARLY_PRINTK
2428 struct console *early_console;
2430 asmlinkage __visible void early_printk(const char *fmt, ...)
2440 n = vscnprintf(buf, sizeof(buf), fmt, ap);
2443 early_console->write(early_console, buf, n);
2447 static void set_user_specified(struct console_cmdline *c, bool user_specified)
2449 if (!user_specified)
2453 * @c console was defined by the user on the command line.
2454 * Do not clear when added twice also by SPCR or the device tree.
2456 c->user_specified = true;
2457 /* At least one console defined by the user on the command line. */
2458 console_set_on_cmdline = 1;
2461 static int __add_preferred_console(char *name, int idx, char *options,
2462 char *brl_options, bool user_specified)
2464 struct console_cmdline *c;
2468 * See if this tty is not yet registered, and
2469 * if we have a slot free.
2471 for (i = 0, c = console_cmdline;
2472 i < MAX_CMDLINECONSOLES && c->name[0];
2474 if (strcmp(c->name, name) == 0 && c->index == idx) {
2476 preferred_console = i;
2477 set_user_specified(c, user_specified);
2481 if (i == MAX_CMDLINECONSOLES)
2484 preferred_console = i;
2485 strscpy(c->name, name, sizeof(c->name));
2486 c->options = options;
2487 set_user_specified(c, user_specified);
2488 braille_set_options(c, brl_options);
2494 static int __init console_msg_format_setup(char *str)
2496 if (!strcmp(str, "syslog"))
2497 console_msg_format = MSG_FORMAT_SYSLOG;
2498 if (!strcmp(str, "default"))
2499 console_msg_format = MSG_FORMAT_DEFAULT;
2502 __setup("console_msg_format=", console_msg_format_setup);
2505 * Set up a console. Called via do_early_param() in init/main.c
2506 * for each "console=" parameter in the boot command line.
2508 static int __init console_setup(char *str)
2510 char buf[sizeof(console_cmdline[0].name) + 4]; /* 4 for "ttyS" */
2511 char *s, *options, *brl_options = NULL;
2515 * console="" or console=null have been suggested as a way to
2516 * disable console output. Use ttynull that has been created
2517 * for exactly this purpose.
2519 if (str[0] == 0 || strcmp(str, "null") == 0) {
2520 __add_preferred_console("ttynull", 0, NULL, NULL, true);
2524 if (_braille_console_setup(&str, &brl_options))
2528 * Decode str into name, index, options.
2530 if (str[0] >= '0' && str[0] <= '9') {
2531 strcpy(buf, "ttyS");
2532 strncpy(buf + 4, str, sizeof(buf) - 5);
2534 strncpy(buf, str, sizeof(buf) - 1);
2536 buf[sizeof(buf) - 1] = 0;
2537 options = strchr(str, ',');
2541 if (!strcmp(str, "ttya"))
2542 strcpy(buf, "ttyS0");
2543 if (!strcmp(str, "ttyb"))
2544 strcpy(buf, "ttyS1");
2546 for (s = buf; *s; s++)
2547 if (isdigit(*s) || *s == ',')
2549 idx = simple_strtoul(s, NULL, 10);
2552 __add_preferred_console(buf, idx, options, brl_options, true);
2555 __setup("console=", console_setup);
2558 * add_preferred_console - add a device to the list of preferred consoles.
2559 * @name: device name
2560 * @idx: device index
2561 * @options: options for this console
2563 * The last preferred console added will be used for kernel messages
2564 * and stdin/out/err for init. Normally this is used by console_setup
2565 * above to handle user-supplied console arguments; however it can also
2566 * be used by arch-specific code either to override the user or more
2567 * commonly to provide a default console (ie from PROM variables) when
2568 * the user has not supplied one.
2570 int add_preferred_console(char *name, int idx, char *options)
2572 return __add_preferred_console(name, idx, options, NULL, false);
2575 bool console_suspend_enabled = true;
2576 EXPORT_SYMBOL(console_suspend_enabled);
2578 static int __init console_suspend_disable(char *str)
2580 console_suspend_enabled = false;
2583 __setup("no_console_suspend", console_suspend_disable);
2584 module_param_named(console_suspend, console_suspend_enabled,
2585 bool, S_IRUGO | S_IWUSR);
2586 MODULE_PARM_DESC(console_suspend, "suspend console during suspend"
2587 " and hibernate operations");
2589 static bool printk_console_no_auto_verbose;
2591 void console_verbose(void)
2593 if (console_loglevel && !printk_console_no_auto_verbose)
2594 console_loglevel = CONSOLE_LOGLEVEL_MOTORMOUTH;
2596 EXPORT_SYMBOL_GPL(console_verbose);
2598 module_param_named(console_no_auto_verbose, printk_console_no_auto_verbose, bool, 0644);
2599 MODULE_PARM_DESC(console_no_auto_verbose, "Disable console loglevel raise to highest on oops/panic/etc");
2602 * suspend_console - suspend the console subsystem
2604 * This disables printk() while we go into suspend states
2606 void suspend_console(void)
2608 if (!console_suspend_enabled)
2610 pr_info("Suspending console(s) (use no_console_suspend to debug)\n");
2611 pr_flush(1000, true);
2613 console_suspended = 1;
2617 void resume_console(void)
2619 if (!console_suspend_enabled)
2622 console_suspended = 0;
2624 pr_flush(1000, true);
2628 * console_cpu_notify - print deferred console messages after CPU hotplug
2631 * If printk() is called from a CPU that is not online yet, the messages
2632 * will be printed on the console only if there are CON_ANYTIME consoles.
2633 * This function is called when a new CPU comes online (or fails to come
2634 * up) or goes offline.
2636 static int console_cpu_notify(unsigned int cpu)
2638 if (!cpuhp_tasks_frozen) {
2639 /* If trylock fails, someone else is doing the printing */
2640 if (console_trylock())
2647 * console_lock - block the console subsystem from printing
2649 * Acquires a lock which guarantees that no consoles will
2650 * be in or enter their write() callback.
2652 * Can sleep, returns nothing.
2654 void console_lock(void)
2659 if (console_suspended)
2662 console_may_schedule = 1;
2664 EXPORT_SYMBOL(console_lock);
2667 * console_trylock - try to block the console subsystem from printing
2669 * Try to acquire a lock which guarantees that no consoles will
2670 * be in or enter their write() callback.
2672 * returns 1 on success, and 0 on failure to acquire the lock.
2674 int console_trylock(void)
2676 if (down_trylock_console_sem())
2678 if (console_suspended) {
2683 console_may_schedule = 0;
2686 EXPORT_SYMBOL(console_trylock);
2688 int is_console_locked(void)
2690 return console_locked;
2692 EXPORT_SYMBOL(is_console_locked);
2695 * Return true when this CPU should unlock console_sem without pushing all
2696 * messages to the console. This reduces the chance that the console is
2697 * locked when the panic CPU tries to use it.
2699 static bool abandon_console_lock_in_panic(void)
2701 if (!panic_in_progress())
2705 * We can use raw_smp_processor_id() here because it is impossible for
2706 * the task to be migrated to the panic_cpu, or away from it. If
2707 * panic_cpu has already been set, and we're not currently executing on
2708 * that CPU, then we never will be.
2710 return atomic_read(&panic_cpu) != raw_smp_processor_id();
2714 * Check if the given console is currently capable and allowed to print
2717 * Requires the console_srcu_read_lock.
2719 static inline bool console_is_usable(struct console *con)
2721 short flags = console_srcu_read_flags(con);
2723 if (!(flags & CON_ENABLED))
2730 * Console drivers may assume that per-cpu resources have been
2731 * allocated. So unless they're explicitly marked as being able to
2732 * cope (CON_ANYTIME) don't call them until this CPU is officially up.
2734 if (!cpu_online(raw_smp_processor_id()) && !(flags & CON_ANYTIME))
2740 static void __console_unlock(void)
2747 * Print one record for the given console. The record printed is whatever
2748 * record is the next available record for the given console.
2750 * @text is a buffer of size CONSOLE_LOG_MAX.
2752 * If extended messages should be printed, @ext_text is a buffer of size
2753 * CONSOLE_EXT_LOG_MAX. Otherwise @ext_text must be NULL.
2755 * If dropped messages should be printed, @dropped_text is a buffer of size
2756 * DROPPED_TEXT_MAX. Otherwise @dropped_text must be NULL.
2758 * @handover will be set to true if a printk waiter has taken over the
2759 * console_lock, in which case the caller is no longer holding both the
2760 * console_lock and the SRCU read lock. Otherwise it is set to false.
2762 * @cookie is the cookie from the SRCU read lock.
2764 * Returns false if the given console has no next record to print, otherwise
2767 * Requires the console_lock and the SRCU read lock.
2769 static bool console_emit_next_record(struct console *con, char *text, char *ext_text,
2770 char *dropped_text, bool *handover, int cookie)
2772 static int panic_console_dropped;
2773 struct printk_info info;
2774 struct printk_record r;
2775 unsigned long flags;
2779 prb_rec_init_rd(&r, &info, text, CONSOLE_LOG_MAX);
2783 if (!prb_read_valid(prb, con->seq, &r))
2786 if (con->seq != r.info->seq) {
2787 con->dropped += r.info->seq - con->seq;
2788 con->seq = r.info->seq;
2789 if (panic_in_progress() && panic_console_dropped++ > 10) {
2790 suppress_panic_printk = 1;
2791 pr_warn_once("Too many dropped messages. Suppress messages on non-panic CPUs to prevent livelock.\n");
2795 /* Skip record that has level above the console loglevel. */
2796 if (suppress_message_printing(r.info->level)) {
2802 write_text = ext_text;
2803 len = info_print_ext_header(ext_text, CONSOLE_EXT_LOG_MAX, r.info);
2804 len += msg_print_ext_body(ext_text + len, CONSOLE_EXT_LOG_MAX - len,
2805 &r.text_buf[0], r.info->text_len, &r.info->dev_info);
2808 len = record_print_text(&r, console_msg_format & MSG_FORMAT_SYSLOG, printk_time);
2812 * While actively printing out messages, if another printk()
2813 * were to occur on another CPU, it may wait for this one to
2814 * finish. This task can not be preempted if there is a
2815 * waiter waiting to take over.
2817 * Interrupts are disabled because the hand over to a waiter
2818 * must not be interrupted until the hand over is completed
2819 * (@console_waiter is cleared).
2821 printk_safe_enter_irqsave(flags);
2822 console_lock_spinning_enable();
2824 stop_critical_timings(); /* don't trace print latency */
2825 call_console_driver(con, write_text, len, dropped_text);
2826 start_critical_timings();
2830 *handover = console_lock_spinning_disable_and_check(cookie);
2831 printk_safe_exit_irqrestore(flags);
2837 * Print out all remaining records to all consoles.
2839 * @do_cond_resched is set by the caller. It can be true only in schedulable
2842 * @next_seq is set to the sequence number after the last available record.
2843 * The value is valid only when this function returns true. It means that all
2844 * usable consoles are completely flushed.
2846 * @handover will be set to true if a printk waiter has taken over the
2847 * console_lock, in which case the caller is no longer holding the
2848 * console_lock. Otherwise it is set to false.
2850 * Returns true when there was at least one usable console and all messages
2851 * were flushed to all usable consoles. A returned false informs the caller
2852 * that everything was not flushed (either there were no usable consoles or
2853 * another context has taken over printing or it is a panic situation and this
2854 * is not the panic CPU). Regardless the reason, the caller should assume it
2855 * is not useful to immediately try again.
2857 * Requires the console_lock.
2859 static bool console_flush_all(bool do_cond_resched, u64 *next_seq, bool *handover)
2861 static char dropped_text[DROPPED_TEXT_MAX];
2862 static char ext_text[CONSOLE_EXT_LOG_MAX];
2863 static char text[CONSOLE_LOG_MAX];
2864 bool any_usable = false;
2865 struct console *con;
2873 any_progress = false;
2875 cookie = console_srcu_read_lock();
2876 for_each_console_srcu(con) {
2879 if (!console_is_usable(con))
2883 if (console_srcu_read_flags(con) & CON_EXTENDED) {
2884 /* Extended consoles do not print "dropped messages". */
2885 progress = console_emit_next_record(con, &text[0],
2889 progress = console_emit_next_record(con, &text[0],
2890 NULL, &dropped_text[0],
2895 * If a handover has occurred, the SRCU read lock
2896 * is already released.
2901 /* Track the next of the highest seq flushed. */
2902 if (con->seq > *next_seq)
2903 *next_seq = con->seq;
2907 any_progress = true;
2909 /* Allow panic_cpu to take over the consoles safely. */
2910 if (abandon_console_lock_in_panic())
2913 if (do_cond_resched)
2916 console_srcu_read_unlock(cookie);
2917 } while (any_progress);
2922 console_srcu_read_unlock(cookie);
2927 * console_unlock - unblock the console subsystem from printing
2929 * Releases the console_lock which the caller holds to block printing of
2930 * the console subsystem.
2932 * While the console_lock was held, console output may have been buffered
2933 * by printk(). If this is the case, console_unlock(); emits
2934 * the output prior to releasing the lock.
2936 * console_unlock(); may be called from any context.
2938 void console_unlock(void)
2940 bool do_cond_resched;
2945 if (console_suspended) {
2951 * Console drivers are called with interrupts disabled, so
2952 * @console_may_schedule should be cleared before; however, we may
2953 * end up dumping a lot of lines, for example, if called from
2954 * console registration path, and should invoke cond_resched()
2955 * between lines if allowable. Not doing so can cause a very long
2956 * scheduling stall on a slow console leading to RCU stall and
2957 * softlockup warnings which exacerbate the issue with more
2958 * messages practically incapacitating the system. Therefore, create
2959 * a local to use for the printing loop.
2961 do_cond_resched = console_may_schedule;
2964 console_may_schedule = 0;
2966 flushed = console_flush_all(do_cond_resched, &next_seq, &handover);
2971 * Abort if there was a failure to flush all messages to all
2972 * usable consoles. Either it is not possible to flush (in
2973 * which case it would be an infinite loop of retrying) or
2974 * another context has taken over printing.
2980 * Some context may have added new records after
2981 * console_flush_all() but before unlocking the console.
2982 * Re-check if there is a new record to flush. If the trylock
2983 * fails, another context is already handling the printing.
2985 } while (prb_read_valid(prb, next_seq, NULL) && console_trylock());
2987 EXPORT_SYMBOL(console_unlock);
2990 * console_conditional_schedule - yield the CPU if required
2992 * If the console code is currently allowed to sleep, and
2993 * if this CPU should yield the CPU to another task, do
2996 * Must be called within console_lock();.
2998 void __sched console_conditional_schedule(void)
3000 if (console_may_schedule)
3003 EXPORT_SYMBOL(console_conditional_schedule);
3005 void console_unblank(void)
3011 * Stop console printing because the unblank() callback may
3012 * assume the console is not within its write() callback.
3014 * If @oops_in_progress is set, this may be an atomic context.
3015 * In that case, attempt a trylock as best-effort.
3017 if (oops_in_progress) {
3018 if (down_trylock_console_sem() != 0)
3024 console_may_schedule = 0;
3026 cookie = console_srcu_read_lock();
3027 for_each_console_srcu(c) {
3028 if ((console_srcu_read_flags(c) & CON_ENABLED) && c->unblank)
3031 console_srcu_read_unlock(cookie);
3035 if (!oops_in_progress)
3036 pr_flush(1000, true);
3040 * console_flush_on_panic - flush console content on panic
3041 * @mode: flush all messages in buffer or just the pending ones
3043 * Immediately output all pending messages no matter what.
3045 void console_flush_on_panic(enum con_flush_mode mode)
3048 * If someone else is holding the console lock, trylock will fail
3049 * and may_schedule may be set. Ignore and proceed to unlock so
3050 * that messages are flushed out. As this can be called from any
3051 * context and we don't want to get preempted while flushing,
3052 * ensure may_schedule is cleared.
3055 console_may_schedule = 0;
3057 if (mode == CONSOLE_REPLAY_ALL) {
3062 seq = prb_first_valid_seq(prb);
3064 cookie = console_srcu_read_lock();
3065 for_each_console_srcu(c) {
3067 * If the above console_trylock() failed, this is an
3068 * unsynchronized assignment. But in that case, the
3069 * kernel is in "hope and pray" mode anyway.
3073 console_srcu_read_unlock(cookie);
3079 * Return the console tty driver structure and its associated index
3081 struct tty_driver *console_device(int *index)
3084 struct tty_driver *driver = NULL;
3088 * Take console_lock to serialize device() callback with
3089 * other console operations. For example, fg_console is
3090 * modified under console_lock when switching vt.
3094 cookie = console_srcu_read_lock();
3095 for_each_console_srcu(c) {
3098 driver = c->device(c, index);
3102 console_srcu_read_unlock(cookie);
3109 * Prevent further output on the passed console device so that (for example)
3110 * serial drivers can disable console output before suspending a port, and can
3111 * re-enable output afterwards.
3113 void console_stop(struct console *console)
3115 __pr_flush(console, 1000, true);
3116 console_list_lock();
3117 console_srcu_write_flags(console, console->flags & ~CON_ENABLED);
3118 console_list_unlock();
3121 * Ensure that all SRCU list walks have completed. All contexts must
3122 * be able to see that this console is disabled so that (for example)
3123 * the caller can suspend the port without risk of another context
3126 synchronize_srcu(&console_srcu);
3128 EXPORT_SYMBOL(console_stop);
3130 void console_start(struct console *console)
3132 console_list_lock();
3133 console_srcu_write_flags(console, console->flags | CON_ENABLED);
3134 console_list_unlock();
3135 __pr_flush(console, 1000, true);
3137 EXPORT_SYMBOL(console_start);
3139 static int __read_mostly keep_bootcon;
3141 static int __init keep_bootcon_setup(char *str)
3144 pr_info("debug: skip boot console de-registration.\n");
3149 early_param("keep_bootcon", keep_bootcon_setup);
3152 * This is called by register_console() to try to match
3153 * the newly registered console with any of the ones selected
3154 * by either the command line or add_preferred_console() and
3157 * Care need to be taken with consoles that are statically
3158 * enabled such as netconsole
3160 static int try_enable_preferred_console(struct console *newcon,
3161 bool user_specified)
3163 struct console_cmdline *c;
3166 for (i = 0, c = console_cmdline;
3167 i < MAX_CMDLINECONSOLES && c->name[0];
3169 if (c->user_specified != user_specified)
3171 if (!newcon->match ||
3172 newcon->match(newcon, c->name, c->index, c->options) != 0) {
3173 /* default matching */
3174 BUILD_BUG_ON(sizeof(c->name) != sizeof(newcon->name));
3175 if (strcmp(c->name, newcon->name) != 0)
3177 if (newcon->index >= 0 &&
3178 newcon->index != c->index)
3180 if (newcon->index < 0)
3181 newcon->index = c->index;
3183 if (_braille_register_console(newcon, c))
3186 if (newcon->setup &&
3187 (err = newcon->setup(newcon, c->options)) != 0)
3190 newcon->flags |= CON_ENABLED;
3191 if (i == preferred_console)
3192 newcon->flags |= CON_CONSDEV;
3197 * Some consoles, such as pstore and netconsole, can be enabled even
3198 * without matching. Accept the pre-enabled consoles only when match()
3199 * and setup() had a chance to be called.
3201 if (newcon->flags & CON_ENABLED && c->user_specified == user_specified)
3207 /* Try to enable the console unconditionally */
3208 static void try_enable_default_console(struct console *newcon)
3210 if (newcon->index < 0)
3213 if (newcon->setup && newcon->setup(newcon, NULL) != 0)
3216 newcon->flags |= CON_ENABLED;
3219 newcon->flags |= CON_CONSDEV;
3222 #define con_printk(lvl, con, fmt, ...) \
3223 printk(lvl pr_fmt("%sconsole [%s%d] " fmt), \
3224 (con->flags & CON_BOOT) ? "boot" : "", \
3225 con->name, con->index, ##__VA_ARGS__)
3227 static void console_init_seq(struct console *newcon, bool bootcon_registered)
3229 struct console *con;
3232 if (newcon->flags & (CON_PRINTBUFFER | CON_BOOT)) {
3233 /* Get a consistent copy of @syslog_seq. */
3234 mutex_lock(&syslog_lock);
3235 newcon->seq = syslog_seq;
3236 mutex_unlock(&syslog_lock);
3238 /* Begin with next message added to ringbuffer. */
3239 newcon->seq = prb_next_seq(prb);
3242 * If any enabled boot consoles are due to be unregistered
3243 * shortly, some may not be caught up and may be the same
3244 * device as @newcon. Since it is not known which boot console
3245 * is the same device, flush all consoles and, if necessary,
3246 * start with the message of the enabled boot console that is
3247 * the furthest behind.
3249 if (bootcon_registered && !keep_bootcon) {
3251 * Hold the console_lock to stop console printing and
3252 * guarantee safe access to console->seq.
3257 * Flush all consoles and set the console to start at
3258 * the next unprinted sequence number.
3260 if (!console_flush_all(true, &newcon->seq, &handover)) {
3262 * Flushing failed. Just choose the lowest
3263 * sequence of the enabled boot consoles.
3267 * If there was a handover, this context no
3268 * longer holds the console_lock.
3273 newcon->seq = prb_next_seq(prb);
3274 for_each_console(con) {
3275 if ((con->flags & CON_BOOT) &&
3276 (con->flags & CON_ENABLED) &&
3277 con->seq < newcon->seq) {
3278 newcon->seq = con->seq;
3288 #define console_first() \
3289 hlist_entry(console_list.first, struct console, node)
3291 static int unregister_console_locked(struct console *console);
3294 * The console driver calls this routine during kernel initialization
3295 * to register the console printing procedure with printk() and to
3296 * print any messages that were printed by the kernel before the
3297 * console driver was initialized.
3299 * This can happen pretty early during the boot process (because of
3300 * early_printk) - sometimes before setup_arch() completes - be careful
3301 * of what kernel features are used - they may not be initialised yet.
3303 * There are two types of consoles - bootconsoles (early_printk) and
3304 * "real" consoles (everything which is not a bootconsole) which are
3305 * handled differently.
3306 * - Any number of bootconsoles can be registered at any time.
3307 * - As soon as a "real" console is registered, all bootconsoles
3308 * will be unregistered automatically.
3309 * - Once a "real" console is registered, any attempt to register a
3310 * bootconsoles will be rejected
3312 void register_console(struct console *newcon)
3314 struct console *con;
3315 bool bootcon_registered = false;
3316 bool realcon_registered = false;
3319 console_list_lock();
3321 for_each_console(con) {
3322 if (WARN(con == newcon, "console '%s%d' already registered\n",
3323 con->name, con->index)) {
3327 if (con->flags & CON_BOOT)
3328 bootcon_registered = true;
3330 realcon_registered = true;
3333 /* Do not register boot consoles when there already is a real one. */
3334 if ((newcon->flags & CON_BOOT) && realcon_registered) {
3335 pr_info("Too late to register bootconsole %s%d\n",
3336 newcon->name, newcon->index);
3341 * See if we want to enable this console driver by default.
3343 * Nope when a console is preferred by the command line, device
3346 * The first real console with tty binding (driver) wins. More
3347 * consoles might get enabled before the right one is found.
3349 * Note that a console with tty binding will have CON_CONSDEV
3350 * flag set and will be first in the list.
3352 if (preferred_console < 0) {
3353 if (hlist_empty(&console_list) || !console_first()->device ||
3354 console_first()->flags & CON_BOOT) {
3355 try_enable_default_console(newcon);
3359 /* See if this console matches one we selected on the command line */
3360 err = try_enable_preferred_console(newcon, true);
3362 /* If not, try to match against the platform default(s) */
3364 err = try_enable_preferred_console(newcon, false);
3366 /* printk() messages are not printed to the Braille console. */
3367 if (err || newcon->flags & CON_BRL)
3371 * If we have a bootconsole, and are switching to a real console,
3372 * don't print everything out again, since when the boot console, and
3373 * the real console are the same physical device, it's annoying to
3374 * see the beginning boot messages twice
3376 if (bootcon_registered &&
3377 ((newcon->flags & (CON_CONSDEV | CON_BOOT)) == CON_CONSDEV)) {
3378 newcon->flags &= ~CON_PRINTBUFFER;
3381 newcon->dropped = 0;
3382 console_init_seq(newcon, bootcon_registered);
3385 * Put this console in the list - keep the
3386 * preferred driver at the head of the list.
3388 if (hlist_empty(&console_list)) {
3389 /* Ensure CON_CONSDEV is always set for the head. */
3390 newcon->flags |= CON_CONSDEV;
3391 hlist_add_head_rcu(&newcon->node, &console_list);
3393 } else if (newcon->flags & CON_CONSDEV) {
3394 /* Only the new head can have CON_CONSDEV set. */
3395 console_srcu_write_flags(console_first(), console_first()->flags & ~CON_CONSDEV);
3396 hlist_add_head_rcu(&newcon->node, &console_list);
3399 hlist_add_behind_rcu(&newcon->node, console_list.first);
3403 * No need to synchronize SRCU here! The caller does not rely
3404 * on all contexts being able to see the new console before
3405 * register_console() completes.
3408 console_sysfs_notify();
3411 * By unregistering the bootconsoles after we enable the real console
3412 * we get the "console xxx enabled" message on all the consoles -
3413 * boot consoles, real consoles, etc - this is to ensure that end
3414 * users know there might be something in the kernel's log buffer that
3415 * went to the bootconsole (that they do not see on the real console)
3417 con_printk(KERN_INFO, newcon, "enabled\n");
3418 if (bootcon_registered &&
3419 ((newcon->flags & (CON_CONSDEV | CON_BOOT)) == CON_CONSDEV) &&
3421 struct hlist_node *tmp;
3423 hlist_for_each_entry_safe(con, tmp, &console_list, node) {
3424 if (con->flags & CON_BOOT)
3425 unregister_console_locked(con);
3429 console_list_unlock();
3431 EXPORT_SYMBOL(register_console);
3433 /* Must be called under console_list_lock(). */
3434 static int unregister_console_locked(struct console *console)
3438 lockdep_assert_console_list_lock_held();
3440 con_printk(KERN_INFO, console, "disabled\n");
3442 res = _braille_unregister_console(console);
3448 /* Disable it unconditionally */
3449 console_srcu_write_flags(console, console->flags & ~CON_ENABLED);
3451 if (!console_is_registered_locked(console))
3454 hlist_del_init_rcu(&console->node);
3458 * If this isn't the last console and it has CON_CONSDEV set, we
3459 * need to set it on the next preferred console.
3462 * The above makes no sense as there is no guarantee that the next
3463 * console has any device attached. Oh well....
3465 if (!hlist_empty(&console_list) && console->flags & CON_CONSDEV)
3466 console_srcu_write_flags(console_first(), console_first()->flags | CON_CONSDEV);
3469 * Ensure that all SRCU list walks have completed. All contexts
3470 * must not be able to see this console in the list so that any
3471 * exit/cleanup routines can be performed safely.
3473 synchronize_srcu(&console_srcu);
3475 console_sysfs_notify();
3478 res = console->exit(console);
3483 int unregister_console(struct console *console)
3487 console_list_lock();
3488 res = unregister_console_locked(console);
3489 console_list_unlock();
3492 EXPORT_SYMBOL(unregister_console);
3495 * console_force_preferred_locked - force a registered console preferred
3496 * @con: The registered console to force preferred.
3498 * Must be called under console_list_lock().
3500 void console_force_preferred_locked(struct console *con)
3502 struct console *cur_pref_con;
3504 if (!console_is_registered_locked(con))
3507 cur_pref_con = console_first();
3509 /* Already preferred? */
3510 if (cur_pref_con == con)
3514 * Delete, but do not re-initialize the entry. This allows the console
3515 * to continue to appear registered (via any hlist_unhashed_lockless()
3516 * checks), even though it was briefly removed from the console list.
3518 hlist_del_rcu(&con->node);
3521 * Ensure that all SRCU list walks have completed so that the console
3522 * can be added to the beginning of the console list and its forward
3523 * list pointer can be re-initialized.
3525 synchronize_srcu(&console_srcu);
3527 con->flags |= CON_CONSDEV;
3528 WARN_ON(!con->device);
3530 /* Only the new head can have CON_CONSDEV set. */
3531 console_srcu_write_flags(cur_pref_con, cur_pref_con->flags & ~CON_CONSDEV);
3532 hlist_add_head_rcu(&con->node, &console_list);
3534 EXPORT_SYMBOL(console_force_preferred_locked);
3537 * Initialize the console device. This is called *early*, so
3538 * we can't necessarily depend on lots of kernel help here.
3539 * Just do some early initializations, and do the complex setup
3542 void __init console_init(void)
3546 initcall_entry_t *ce;
3548 /* Setup the default TTY line discipline. */
3552 * set up the console device so that later boot sequences can
3553 * inform about problems etc..
3555 ce = __con_initcall_start;
3556 trace_initcall_level("console");
3557 while (ce < __con_initcall_end) {
3558 call = initcall_from_entry(ce);
3559 trace_initcall_start(call);
3561 trace_initcall_finish(call, ret);
3567 * Some boot consoles access data that is in the init section and which will
3568 * be discarded after the initcalls have been run. To make sure that no code
3569 * will access this data, unregister the boot consoles in a late initcall.
3571 * If for some reason, such as deferred probe or the driver being a loadable
3572 * module, the real console hasn't registered yet at this point, there will
3573 * be a brief interval in which no messages are logged to the console, which
3574 * makes it difficult to diagnose problems that occur during this time.
3576 * To mitigate this problem somewhat, only unregister consoles whose memory
3577 * intersects with the init section. Note that all other boot consoles will
3578 * get unregistered when the real preferred console is registered.
3580 static int __init printk_late_init(void)
3582 struct hlist_node *tmp;
3583 struct console *con;
3586 console_list_lock();
3587 hlist_for_each_entry_safe(con, tmp, &console_list, node) {
3588 if (!(con->flags & CON_BOOT))
3591 /* Check addresses that might be used for enabled consoles. */
3592 if (init_section_intersects(con, sizeof(*con)) ||
3593 init_section_contains(con->write, 0) ||
3594 init_section_contains(con->read, 0) ||
3595 init_section_contains(con->device, 0) ||
3596 init_section_contains(con->unblank, 0) ||
3597 init_section_contains(con->data, 0)) {
3599 * Please, consider moving the reported consoles out
3600 * of the init section.
3602 pr_warn("bootconsole [%s%d] uses init memory and must be disabled even before the real one is ready\n",
3603 con->name, con->index);
3604 unregister_console_locked(con);
3607 console_list_unlock();
3609 ret = cpuhp_setup_state_nocalls(CPUHP_PRINTK_DEAD, "printk:dead", NULL,
3610 console_cpu_notify);
3612 ret = cpuhp_setup_state_nocalls(CPUHP_AP_ONLINE_DYN, "printk:online",
3613 console_cpu_notify, NULL);
3615 printk_sysctl_init();
3618 late_initcall(printk_late_init);
3620 #if defined CONFIG_PRINTK
3621 /* If @con is specified, only wait for that console. Otherwise wait for all. */
3622 static bool __pr_flush(struct console *con, int timeout_ms, bool reset_on_progress)
3624 int remaining = timeout_ms;
3634 seq = prb_next_seq(prb);
3640 * Hold the console_lock to guarantee safe access to
3641 * console->seq and to prevent changes to @console_suspended
3642 * until all consoles have been processed.
3646 cookie = console_srcu_read_lock();
3647 for_each_console_srcu(c) {
3648 if (con && con != c)
3650 if (!console_is_usable(c))
3652 printk_seq = c->seq;
3653 if (printk_seq < seq)
3654 diff += seq - printk_seq;
3656 console_srcu_read_unlock(cookie);
3659 * If consoles are suspended, it cannot be expected that they
3660 * make forward progress, so timeout immediately. @diff is
3661 * still used to return a valid flush status.
3663 if (console_suspended)
3665 else if (diff != last_diff && reset_on_progress)
3666 remaining = timeout_ms;
3670 if (diff == 0 || remaining == 0)
3673 if (remaining < 0) {
3674 /* no timeout limit */
3676 } else if (remaining < 100) {
3691 * pr_flush() - Wait for printing threads to catch up.
3693 * @timeout_ms: The maximum time (in ms) to wait.
3694 * @reset_on_progress: Reset the timeout if forward progress is seen.
3696 * A value of 0 for @timeout_ms means no waiting will occur. A value of -1
3697 * represents infinite waiting.
3699 * If @reset_on_progress is true, the timeout will be reset whenever any
3700 * printer has been seen to make some forward progress.
3702 * Context: Process context. May sleep while acquiring console lock.
3703 * Return: true if all enabled printers are caught up.
3705 static bool pr_flush(int timeout_ms, bool reset_on_progress)
3707 return __pr_flush(NULL, timeout_ms, reset_on_progress);
3711 * Delayed printk version, for scheduler-internal messages:
3713 #define PRINTK_PENDING_WAKEUP 0x01
3714 #define PRINTK_PENDING_OUTPUT 0x02
3716 static DEFINE_PER_CPU(int, printk_pending);
3718 static void wake_up_klogd_work_func(struct irq_work *irq_work)
3720 int pending = this_cpu_xchg(printk_pending, 0);
3722 if (pending & PRINTK_PENDING_OUTPUT) {
3723 /* If trylock fails, someone else is doing the printing */
3724 if (console_trylock())
3728 if (pending & PRINTK_PENDING_WAKEUP)
3729 wake_up_interruptible(&log_wait);
3732 static DEFINE_PER_CPU(struct irq_work, wake_up_klogd_work) =
3733 IRQ_WORK_INIT_LAZY(wake_up_klogd_work_func);
3735 static void __wake_up_klogd(int val)
3737 if (!printk_percpu_data_ready())
3742 * Guarantee any new records can be seen by tasks preparing to wait
3743 * before this context checks if the wait queue is empty.
3745 * The full memory barrier within wq_has_sleeper() pairs with the full
3746 * memory barrier within set_current_state() of
3747 * prepare_to_wait_event(), which is called after ___wait_event() adds
3748 * the waiter but before it has checked the wait condition.
3750 * This pairs with devkmsg_read:A and syslog_print:A.
3752 if (wq_has_sleeper(&log_wait) || /* LMM(__wake_up_klogd:A) */
3753 (val & PRINTK_PENDING_OUTPUT)) {
3754 this_cpu_or(printk_pending, val);
3755 irq_work_queue(this_cpu_ptr(&wake_up_klogd_work));
3760 void wake_up_klogd(void)
3762 __wake_up_klogd(PRINTK_PENDING_WAKEUP);
3765 void defer_console_output(void)
3768 * New messages may have been added directly to the ringbuffer
3769 * using vprintk_store(), so wake any waiters as well.
3771 __wake_up_klogd(PRINTK_PENDING_WAKEUP | PRINTK_PENDING_OUTPUT);
3774 void printk_trigger_flush(void)
3776 defer_console_output();
3779 int vprintk_deferred(const char *fmt, va_list args)
3783 r = vprintk_emit(0, LOGLEVEL_SCHED, NULL, fmt, args);
3784 defer_console_output();
3789 int _printk_deferred(const char *fmt, ...)
3794 va_start(args, fmt);
3795 r = vprintk_deferred(fmt, args);
3802 * printk rate limiting, lifted from the networking subsystem.
3804 * This enforces a rate limit: not more than 10 kernel messages
3805 * every 5s to make a denial-of-service attack impossible.
3807 DEFINE_RATELIMIT_STATE(printk_ratelimit_state, 5 * HZ, 10);
3809 int __printk_ratelimit(const char *func)
3811 return ___ratelimit(&printk_ratelimit_state, func);
3813 EXPORT_SYMBOL(__printk_ratelimit);
3816 * printk_timed_ratelimit - caller-controlled printk ratelimiting
3817 * @caller_jiffies: pointer to caller's state
3818 * @interval_msecs: minimum interval between prints
3820 * printk_timed_ratelimit() returns true if more than @interval_msecs
3821 * milliseconds have elapsed since the last time printk_timed_ratelimit()
3824 bool printk_timed_ratelimit(unsigned long *caller_jiffies,
3825 unsigned int interval_msecs)
3827 unsigned long elapsed = jiffies - *caller_jiffies;
3829 if (*caller_jiffies && elapsed <= msecs_to_jiffies(interval_msecs))
3832 *caller_jiffies = jiffies;
3835 EXPORT_SYMBOL(printk_timed_ratelimit);
3837 static DEFINE_SPINLOCK(dump_list_lock);
3838 static LIST_HEAD(dump_list);
3841 * kmsg_dump_register - register a kernel log dumper.
3842 * @dumper: pointer to the kmsg_dumper structure
3844 * Adds a kernel log dumper to the system. The dump callback in the
3845 * structure will be called when the kernel oopses or panics and must be
3846 * set. Returns zero on success and %-EINVAL or %-EBUSY otherwise.
3848 int kmsg_dump_register(struct kmsg_dumper *dumper)
3850 unsigned long flags;
3853 /* The dump callback needs to be set */
3857 spin_lock_irqsave(&dump_list_lock, flags);
3858 /* Don't allow registering multiple times */
3859 if (!dumper->registered) {
3860 dumper->registered = 1;
3861 list_add_tail_rcu(&dumper->list, &dump_list);
3864 spin_unlock_irqrestore(&dump_list_lock, flags);
3868 EXPORT_SYMBOL_GPL(kmsg_dump_register);
3871 * kmsg_dump_unregister - unregister a kmsg dumper.
3872 * @dumper: pointer to the kmsg_dumper structure
3874 * Removes a dump device from the system. Returns zero on success and
3875 * %-EINVAL otherwise.
3877 int kmsg_dump_unregister(struct kmsg_dumper *dumper)
3879 unsigned long flags;
3882 spin_lock_irqsave(&dump_list_lock, flags);
3883 if (dumper->registered) {
3884 dumper->registered = 0;
3885 list_del_rcu(&dumper->list);
3888 spin_unlock_irqrestore(&dump_list_lock, flags);
3893 EXPORT_SYMBOL_GPL(kmsg_dump_unregister);
3895 static bool always_kmsg_dump;
3896 module_param_named(always_kmsg_dump, always_kmsg_dump, bool, S_IRUGO | S_IWUSR);
3898 const char *kmsg_dump_reason_str(enum kmsg_dump_reason reason)
3901 case KMSG_DUMP_PANIC:
3903 case KMSG_DUMP_OOPS:
3905 case KMSG_DUMP_EMERG:
3907 case KMSG_DUMP_SHUTDOWN:
3913 EXPORT_SYMBOL_GPL(kmsg_dump_reason_str);
3916 * kmsg_dump - dump kernel log to kernel message dumpers.
3917 * @reason: the reason (oops, panic etc) for dumping
3919 * Call each of the registered dumper's dump() callback, which can
3920 * retrieve the kmsg records with kmsg_dump_get_line() or
3921 * kmsg_dump_get_buffer().
3923 void kmsg_dump(enum kmsg_dump_reason reason)
3925 struct kmsg_dumper *dumper;
3928 list_for_each_entry_rcu(dumper, &dump_list, list) {
3929 enum kmsg_dump_reason max_reason = dumper->max_reason;
3932 * If client has not provided a specific max_reason, default
3933 * to KMSG_DUMP_OOPS, unless always_kmsg_dump was set.
3935 if (max_reason == KMSG_DUMP_UNDEF) {
3936 max_reason = always_kmsg_dump ? KMSG_DUMP_MAX :
3939 if (reason > max_reason)
3942 /* invoke dumper which will iterate over records */
3943 dumper->dump(dumper, reason);
3949 * kmsg_dump_get_line - retrieve one kmsg log line
3950 * @iter: kmsg dump iterator
3951 * @syslog: include the "<4>" prefixes
3952 * @line: buffer to copy the line to
3953 * @size: maximum size of the buffer
3954 * @len: length of line placed into buffer
3956 * Start at the beginning of the kmsg buffer, with the oldest kmsg
3957 * record, and copy one record into the provided buffer.
3959 * Consecutive calls will return the next available record moving
3960 * towards the end of the buffer with the youngest messages.
3962 * A return value of FALSE indicates that there are no more records to
3965 bool kmsg_dump_get_line(struct kmsg_dump_iter *iter, bool syslog,
3966 char *line, size_t size, size_t *len)
3968 u64 min_seq = latched_seq_read_nolock(&clear_seq);
3969 struct printk_info info;
3970 unsigned int line_count;
3971 struct printk_record r;
3975 if (iter->cur_seq < min_seq)
3976 iter->cur_seq = min_seq;
3978 prb_rec_init_rd(&r, &info, line, size);
3980 /* Read text or count text lines? */
3982 if (!prb_read_valid(prb, iter->cur_seq, &r))
3984 l = record_print_text(&r, syslog, printk_time);
3986 if (!prb_read_valid_info(prb, iter->cur_seq,
3987 &info, &line_count)) {
3990 l = get_record_print_text_size(&info, line_count, syslog,
3995 iter->cur_seq = r.info->seq + 1;
4002 EXPORT_SYMBOL_GPL(kmsg_dump_get_line);
4005 * kmsg_dump_get_buffer - copy kmsg log lines
4006 * @iter: kmsg dump iterator
4007 * @syslog: include the "<4>" prefixes
4008 * @buf: buffer to copy the line to
4009 * @size: maximum size of the buffer
4010 * @len_out: length of line placed into buffer
4012 * Start at the end of the kmsg buffer and fill the provided buffer
4013 * with as many of the *youngest* kmsg records that fit into it.
4014 * If the buffer is large enough, all available kmsg records will be
4015 * copied with a single call.
4017 * Consecutive calls will fill the buffer with the next block of
4018 * available older records, not including the earlier retrieved ones.
4020 * A return value of FALSE indicates that there are no more records to
4023 bool kmsg_dump_get_buffer(struct kmsg_dump_iter *iter, bool syslog,
4024 char *buf, size_t size, size_t *len_out)
4026 u64 min_seq = latched_seq_read_nolock(&clear_seq);
4027 struct printk_info info;
4028 struct printk_record r;
4033 bool time = printk_time;
4038 if (iter->cur_seq < min_seq)
4039 iter->cur_seq = min_seq;
4041 if (prb_read_valid_info(prb, iter->cur_seq, &info, NULL)) {
4042 if (info.seq != iter->cur_seq) {
4043 /* messages are gone, move to first available one */
4044 iter->cur_seq = info.seq;
4049 if (iter->cur_seq >= iter->next_seq)
4053 * Find first record that fits, including all following records,
4054 * into the user-provided buffer for this dump. Pass in size-1
4055 * because this function (by way of record_print_text()) will
4056 * not write more than size-1 bytes of text into @buf.
4058 seq = find_first_fitting_seq(iter->cur_seq, iter->next_seq,
4059 size - 1, syslog, time);
4062 * Next kmsg_dump_get_buffer() invocation will dump block of
4063 * older records stored right before this one.
4067 prb_rec_init_rd(&r, &info, buf, size);
4070 prb_for_each_record(seq, prb, seq, &r) {
4071 if (r.info->seq >= iter->next_seq)
4074 len += record_print_text(&r, syslog, time);
4076 /* Adjust record to store to remaining buffer space. */
4077 prb_rec_init_rd(&r, &info, buf + len, size - len);
4080 iter->next_seq = next_seq;
4087 EXPORT_SYMBOL_GPL(kmsg_dump_get_buffer);
4090 * kmsg_dump_rewind - reset the iterator
4091 * @iter: kmsg dump iterator
4093 * Reset the dumper's iterator so that kmsg_dump_get_line() and
4094 * kmsg_dump_get_buffer() can be called again and used multiple
4095 * times within the same dumper.dump() callback.
4097 void kmsg_dump_rewind(struct kmsg_dump_iter *iter)
4099 iter->cur_seq = latched_seq_read_nolock(&clear_seq);
4100 iter->next_seq = prb_next_seq(prb);
4102 EXPORT_SYMBOL_GPL(kmsg_dump_rewind);
4107 static atomic_t printk_cpu_sync_owner = ATOMIC_INIT(-1);
4108 static atomic_t printk_cpu_sync_nested = ATOMIC_INIT(0);
4111 * __printk_cpu_sync_wait() - Busy wait until the printk cpu-reentrant
4112 * spinning lock is not owned by any CPU.
4114 * Context: Any context.
4116 void __printk_cpu_sync_wait(void)
4120 } while (atomic_read(&printk_cpu_sync_owner) != -1);
4122 EXPORT_SYMBOL(__printk_cpu_sync_wait);
4125 * __printk_cpu_sync_try_get() - Try to acquire the printk cpu-reentrant
4128 * If no processor has the lock, the calling processor takes the lock and
4129 * becomes the owner. If the calling processor is already the owner of the
4130 * lock, this function succeeds immediately.
4132 * Context: Any context. Expects interrupts to be disabled.
4133 * Return: 1 on success, otherwise 0.
4135 int __printk_cpu_sync_try_get(void)
4140 cpu = smp_processor_id();
4143 * Guarantee loads and stores from this CPU when it is the lock owner
4144 * are _not_ visible to the previous lock owner. This pairs with
4145 * __printk_cpu_sync_put:B.
4147 * Memory barrier involvement:
4149 * If __printk_cpu_sync_try_get:A reads from __printk_cpu_sync_put:B,
4150 * then __printk_cpu_sync_put:A can never read from
4151 * __printk_cpu_sync_try_get:B.
4155 * RELEASE from __printk_cpu_sync_put:A to __printk_cpu_sync_put:B
4156 * of the previous CPU
4158 * ACQUIRE from __printk_cpu_sync_try_get:A to
4159 * __printk_cpu_sync_try_get:B of this CPU
4161 old = atomic_cmpxchg_acquire(&printk_cpu_sync_owner, -1,
4162 cpu); /* LMM(__printk_cpu_sync_try_get:A) */
4165 * This CPU is now the owner and begins loading/storing
4166 * data: LMM(__printk_cpu_sync_try_get:B)
4170 } else if (old == cpu) {
4171 /* This CPU is already the owner. */
4172 atomic_inc(&printk_cpu_sync_nested);
4178 EXPORT_SYMBOL(__printk_cpu_sync_try_get);
4181 * __printk_cpu_sync_put() - Release the printk cpu-reentrant spinning lock.
4183 * The calling processor must be the owner of the lock.
4185 * Context: Any context. Expects interrupts to be disabled.
4187 void __printk_cpu_sync_put(void)
4189 if (atomic_read(&printk_cpu_sync_nested)) {
4190 atomic_dec(&printk_cpu_sync_nested);
4195 * This CPU is finished loading/storing data:
4196 * LMM(__printk_cpu_sync_put:A)
4200 * Guarantee loads and stores from this CPU when it was the
4201 * lock owner are visible to the next lock owner. This pairs
4202 * with __printk_cpu_sync_try_get:A.
4204 * Memory barrier involvement:
4206 * If __printk_cpu_sync_try_get:A reads from __printk_cpu_sync_put:B,
4207 * then __printk_cpu_sync_try_get:B reads from __printk_cpu_sync_put:A.
4211 * RELEASE from __printk_cpu_sync_put:A to __printk_cpu_sync_put:B
4214 * ACQUIRE from __printk_cpu_sync_try_get:A to
4215 * __printk_cpu_sync_try_get:B of the next CPU
4217 atomic_set_release(&printk_cpu_sync_owner,
4218 -1); /* LMM(__printk_cpu_sync_put:B) */
4220 EXPORT_SYMBOL(__printk_cpu_sync_put);
4221 #endif /* CONFIG_SMP */