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);
128 enum devkmsg_log_bits {
129 __DEVKMSG_LOG_BIT_ON = 0,
130 __DEVKMSG_LOG_BIT_OFF,
131 __DEVKMSG_LOG_BIT_LOCK,
134 enum devkmsg_log_masks {
135 DEVKMSG_LOG_MASK_ON = BIT(__DEVKMSG_LOG_BIT_ON),
136 DEVKMSG_LOG_MASK_OFF = BIT(__DEVKMSG_LOG_BIT_OFF),
137 DEVKMSG_LOG_MASK_LOCK = BIT(__DEVKMSG_LOG_BIT_LOCK),
140 /* Keep both the 'on' and 'off' bits clear, i.e. ratelimit by default: */
141 #define DEVKMSG_LOG_MASK_DEFAULT 0
143 static unsigned int __read_mostly devkmsg_log = DEVKMSG_LOG_MASK_DEFAULT;
145 static int __control_devkmsg(char *str)
152 len = str_has_prefix(str, "on");
154 devkmsg_log = DEVKMSG_LOG_MASK_ON;
158 len = str_has_prefix(str, "off");
160 devkmsg_log = DEVKMSG_LOG_MASK_OFF;
164 len = str_has_prefix(str, "ratelimit");
166 devkmsg_log = DEVKMSG_LOG_MASK_DEFAULT;
173 static int __init control_devkmsg(char *str)
175 if (__control_devkmsg(str) < 0) {
176 pr_warn("printk.devkmsg: bad option string '%s'\n", str);
181 * Set sysctl string accordingly:
183 if (devkmsg_log == DEVKMSG_LOG_MASK_ON)
184 strcpy(devkmsg_log_str, "on");
185 else if (devkmsg_log == DEVKMSG_LOG_MASK_OFF)
186 strcpy(devkmsg_log_str, "off");
187 /* else "ratelimit" which is set by default. */
190 * Sysctl cannot change it anymore. The kernel command line setting of
191 * this parameter is to force the setting to be permanent throughout the
192 * runtime of the system. This is a precation measure against userspace
193 * trying to be a smarta** and attempting to change it up on us.
195 devkmsg_log |= DEVKMSG_LOG_MASK_LOCK;
199 __setup("printk.devkmsg=", control_devkmsg);
201 char devkmsg_log_str[DEVKMSG_STR_MAX_SIZE] = "ratelimit";
202 #if defined(CONFIG_PRINTK) && defined(CONFIG_SYSCTL)
203 int devkmsg_sysctl_set_loglvl(struct ctl_table *table, int write,
204 void *buffer, size_t *lenp, loff_t *ppos)
206 char old_str[DEVKMSG_STR_MAX_SIZE];
211 if (devkmsg_log & DEVKMSG_LOG_MASK_LOCK)
215 strncpy(old_str, devkmsg_log_str, DEVKMSG_STR_MAX_SIZE);
218 err = proc_dostring(table, write, buffer, lenp, ppos);
223 err = __control_devkmsg(devkmsg_log_str);
226 * Do not accept an unknown string OR a known string with
229 if (err < 0 || (err + 1 != *lenp)) {
231 /* ... and restore old setting. */
233 strncpy(devkmsg_log_str, old_str, DEVKMSG_STR_MAX_SIZE);
241 #endif /* CONFIG_PRINTK && CONFIG_SYSCTL */
244 * console_list_lock - Lock the console list
246 * For console list or console->flags updates
248 void console_list_lock(void)
251 * In unregister_console() and console_force_preferred_locked(),
252 * synchronize_srcu() is called with the console_list_lock held.
253 * Therefore it is not allowed that the console_list_lock is taken
254 * with the srcu_lock held.
256 * Detecting if this context is really in the read-side critical
257 * section is only possible if the appropriate debug options are
260 WARN_ON_ONCE(debug_lockdep_rcu_enabled() &&
261 srcu_read_lock_held(&console_srcu));
263 mutex_lock(&console_mutex);
265 EXPORT_SYMBOL(console_list_lock);
268 * console_list_unlock - Unlock the console list
270 * Counterpart to console_list_lock()
272 void console_list_unlock(void)
274 mutex_unlock(&console_mutex);
276 EXPORT_SYMBOL(console_list_unlock);
279 * console_srcu_read_lock - Register a new reader for the
280 * SRCU-protected console list
282 * Use for_each_console_srcu() to iterate the console list
284 * Context: Any context.
285 * Return: A cookie to pass to console_srcu_read_unlock().
287 int console_srcu_read_lock(void)
289 return srcu_read_lock_nmisafe(&console_srcu);
291 EXPORT_SYMBOL(console_srcu_read_lock);
294 * console_srcu_read_unlock - Unregister an old reader from
295 * the SRCU-protected console list
296 * @cookie: cookie returned from console_srcu_read_lock()
298 * Counterpart to console_srcu_read_lock()
300 void console_srcu_read_unlock(int cookie)
302 srcu_read_unlock_nmisafe(&console_srcu, cookie);
304 EXPORT_SYMBOL(console_srcu_read_unlock);
307 * Helper macros to handle lockdep when locking/unlocking console_sem. We use
308 * macros instead of functions so that _RET_IP_ contains useful information.
310 #define down_console_sem() do { \
312 mutex_acquire(&console_lock_dep_map, 0, 0, _RET_IP_);\
315 static int __down_trylock_console_sem(unsigned long ip)
321 * Here and in __up_console_sem() we need to be in safe mode,
322 * because spindump/WARN/etc from under console ->lock will
323 * deadlock in printk()->down_trylock_console_sem() otherwise.
325 printk_safe_enter_irqsave(flags);
326 lock_failed = down_trylock(&console_sem);
327 printk_safe_exit_irqrestore(flags);
331 mutex_acquire(&console_lock_dep_map, 0, 1, ip);
334 #define down_trylock_console_sem() __down_trylock_console_sem(_RET_IP_)
336 static void __up_console_sem(unsigned long ip)
340 mutex_release(&console_lock_dep_map, ip);
342 printk_safe_enter_irqsave(flags);
344 printk_safe_exit_irqrestore(flags);
346 #define up_console_sem() __up_console_sem(_RET_IP_)
348 static bool panic_in_progress(void)
350 return unlikely(atomic_read(&panic_cpu) != PANIC_CPU_INVALID);
354 * This is used for debugging the mess that is the VT code by
355 * keeping track if we have the console semaphore held. It's
356 * definitely not the perfect debug tool (we don't know if _WE_
357 * hold it and are racing, but it helps tracking those weird code
358 * paths in the console code where we end up in places I want
359 * locked without the console semaphore held).
361 static int console_locked, console_suspended;
364 * Array of consoles built from command line options (console=)
367 #define MAX_CMDLINECONSOLES 8
369 static struct console_cmdline console_cmdline[MAX_CMDLINECONSOLES];
371 static int preferred_console = -1;
372 int console_set_on_cmdline;
373 EXPORT_SYMBOL(console_set_on_cmdline);
375 /* Flag: console code may call schedule() */
376 static int console_may_schedule;
378 enum con_msg_format_flags {
379 MSG_FORMAT_DEFAULT = 0,
380 MSG_FORMAT_SYSLOG = (1 << 0),
383 static int console_msg_format = MSG_FORMAT_DEFAULT;
386 * The printk log buffer consists of a sequenced collection of records, each
387 * containing variable length message text. Every record also contains its
388 * own meta-data (@info).
390 * Every record meta-data carries the timestamp in microseconds, as well as
391 * the standard userspace syslog level and syslog facility. The usual kernel
392 * messages use LOG_KERN; userspace-injected messages always carry a matching
393 * syslog facility, by default LOG_USER. The origin of every message can be
394 * reliably determined that way.
396 * The human readable log message of a record is available in @text, the
397 * length of the message text in @text_len. The stored message is not
400 * Optionally, a record can carry a dictionary of properties (key/value
401 * pairs), to provide userspace with a machine-readable message context.
403 * Examples for well-defined, commonly used property names are:
404 * DEVICE=b12:8 device identifier
408 * +sound:card0 subsystem:devname
409 * SUBSYSTEM=pci driver-core subsystem name
411 * Valid characters in property names are [a-zA-Z0-9.-_]. Property names
412 * and values are terminated by a '\0' character.
414 * Example of record values:
415 * record.text_buf = "it's a line" (unterminated)
416 * record.info.seq = 56
417 * record.info.ts_nsec = 36863
418 * record.info.text_len = 11
419 * record.info.facility = 0 (LOG_KERN)
420 * record.info.flags = 0
421 * record.info.level = 3 (LOG_ERR)
422 * record.info.caller_id = 299 (task 299)
423 * record.info.dev_info.subsystem = "pci" (terminated)
424 * record.info.dev_info.device = "+pci:0000:00:01.0" (terminated)
426 * The 'struct printk_info' buffer must never be directly exported to
427 * userspace, it is a kernel-private implementation detail that might
428 * need to be changed in the future, when the requirements change.
430 * /dev/kmsg exports the structured data in the following line format:
431 * "<level>,<sequnum>,<timestamp>,<contflag>[,additional_values, ... ];<message text>\n"
433 * Users of the export format should ignore possible additional values
434 * separated by ',', and find the message after the ';' character.
436 * The optional key/value pairs are attached as continuation lines starting
437 * with a space character and terminated by a newline. All possible
438 * non-prinatable characters are escaped in the "\xff" notation.
441 /* syslog_lock protects syslog_* variables and write access to clear_seq. */
442 static DEFINE_MUTEX(syslog_lock);
445 DECLARE_WAIT_QUEUE_HEAD(log_wait);
446 /* All 3 protected by @syslog_lock. */
447 /* the next printk record to read by syslog(READ) or /proc/kmsg */
448 static u64 syslog_seq;
449 static size_t syslog_partial;
450 static bool syslog_time;
453 seqcount_latch_t latch;
458 * The next printk record to read after the last 'clear' command. There are
459 * two copies (updated with seqcount_latch) so that reads can locklessly
460 * access a valid value. Writers are synchronized by @syslog_lock.
462 static struct latched_seq clear_seq = {
463 .latch = SEQCNT_LATCH_ZERO(clear_seq.latch),
468 #ifdef CONFIG_PRINTK_CALLER
469 #define PREFIX_MAX 48
471 #define PREFIX_MAX 32
474 /* the maximum size of a formatted record (i.e. with prefix added per line) */
475 #define CONSOLE_LOG_MAX 1024
477 /* the maximum size for a dropped text message */
478 #define DROPPED_TEXT_MAX 64
480 /* the maximum size allowed to be reserved for a record */
481 #define LOG_LINE_MAX (CONSOLE_LOG_MAX - PREFIX_MAX)
483 #define LOG_LEVEL(v) ((v) & 0x07)
484 #define LOG_FACILITY(v) ((v) >> 3 & 0xff)
487 #define LOG_ALIGN __alignof__(unsigned long)
488 #define __LOG_BUF_LEN (1 << CONFIG_LOG_BUF_SHIFT)
489 #define LOG_BUF_LEN_MAX (u32)(1 << 31)
490 static char __log_buf[__LOG_BUF_LEN] __aligned(LOG_ALIGN);
491 static char *log_buf = __log_buf;
492 static u32 log_buf_len = __LOG_BUF_LEN;
495 * Define the average message size. This only affects the number of
496 * descriptors that will be available. Underestimating is better than
497 * overestimating (too many available descriptors is better than not enough).
499 #define PRB_AVGBITS 5 /* 32 character average length */
501 #if CONFIG_LOG_BUF_SHIFT <= PRB_AVGBITS
502 #error CONFIG_LOG_BUF_SHIFT value too small.
504 _DEFINE_PRINTKRB(printk_rb_static, CONFIG_LOG_BUF_SHIFT - PRB_AVGBITS,
505 PRB_AVGBITS, &__log_buf[0]);
507 static struct printk_ringbuffer printk_rb_dynamic;
509 static struct printk_ringbuffer *prb = &printk_rb_static;
512 * We cannot access per-CPU data (e.g. per-CPU flush irq_work) before
513 * per_cpu_areas are initialised. This variable is set to true when
514 * it's safe to access per-CPU data.
516 static bool __printk_percpu_data_ready __ro_after_init;
518 bool printk_percpu_data_ready(void)
520 return __printk_percpu_data_ready;
523 /* Must be called under syslog_lock. */
524 static void latched_seq_write(struct latched_seq *ls, u64 val)
526 raw_write_seqcount_latch(&ls->latch);
528 raw_write_seqcount_latch(&ls->latch);
532 /* Can be called from any context. */
533 static u64 latched_seq_read_nolock(struct latched_seq *ls)
540 seq = raw_read_seqcount_latch(&ls->latch);
543 } while (read_seqcount_latch_retry(&ls->latch, seq));
548 /* Return log buffer address */
549 char *log_buf_addr_get(void)
554 /* Return log buffer size */
555 u32 log_buf_len_get(void)
561 * Define how much of the log buffer we could take at maximum. The value
562 * must be greater than two. Note that only half of the buffer is available
563 * when the index points to the middle.
565 #define MAX_LOG_TAKE_PART 4
566 static const char trunc_msg[] = "<truncated>";
568 static void truncate_msg(u16 *text_len, u16 *trunc_msg_len)
571 * The message should not take the whole buffer. Otherwise, it might
572 * get removed too soon.
574 u32 max_text_len = log_buf_len / MAX_LOG_TAKE_PART;
576 if (*text_len > max_text_len)
577 *text_len = max_text_len;
579 /* enable the warning message (if there is room) */
580 *trunc_msg_len = strlen(trunc_msg);
581 if (*text_len >= *trunc_msg_len)
582 *text_len -= *trunc_msg_len;
587 int dmesg_restrict = IS_ENABLED(CONFIG_SECURITY_DMESG_RESTRICT);
589 static int syslog_action_restricted(int type)
594 * Unless restricted, we allow "read all" and "get buffer size"
597 return type != SYSLOG_ACTION_READ_ALL &&
598 type != SYSLOG_ACTION_SIZE_BUFFER;
601 static int check_syslog_permissions(int type, int source)
604 * If this is from /proc/kmsg and we've already opened it, then we've
605 * already done the capabilities checks at open time.
607 if (source == SYSLOG_FROM_PROC && type != SYSLOG_ACTION_OPEN)
610 if (syslog_action_restricted(type)) {
611 if (capable(CAP_SYSLOG))
614 * For historical reasons, accept CAP_SYS_ADMIN too, with
617 if (capable(CAP_SYS_ADMIN)) {
618 pr_warn_once("%s (%d): Attempt to access syslog with "
619 "CAP_SYS_ADMIN but no CAP_SYSLOG "
621 current->comm, task_pid_nr(current));
627 return security_syslog(type);
630 static void append_char(char **pp, char *e, char c)
636 static ssize_t info_print_ext_header(char *buf, size_t size,
637 struct printk_info *info)
639 u64 ts_usec = info->ts_nsec;
641 #ifdef CONFIG_PRINTK_CALLER
642 u32 id = info->caller_id;
644 snprintf(caller, sizeof(caller), ",caller=%c%u",
645 id & 0x80000000 ? 'C' : 'T', id & ~0x80000000);
650 do_div(ts_usec, 1000);
652 return scnprintf(buf, size, "%u,%llu,%llu,%c%s;",
653 (info->facility << 3) | info->level, info->seq,
654 ts_usec, info->flags & LOG_CONT ? 'c' : '-', caller);
657 static ssize_t msg_add_ext_text(char *buf, size_t size,
658 const char *text, size_t text_len,
661 char *p = buf, *e = buf + size;
664 /* escape non-printable characters */
665 for (i = 0; i < text_len; i++) {
666 unsigned char c = text[i];
668 if (c < ' ' || c >= 127 || c == '\\')
669 p += scnprintf(p, e - p, "\\x%02x", c);
671 append_char(&p, e, c);
673 append_char(&p, e, endc);
678 static ssize_t msg_add_dict_text(char *buf, size_t size,
679 const char *key, const char *val)
681 size_t val_len = strlen(val);
687 len = msg_add_ext_text(buf, size, "", 0, ' '); /* dict prefix */
688 len += msg_add_ext_text(buf + len, size - len, key, strlen(key), '=');
689 len += msg_add_ext_text(buf + len, size - len, val, val_len, '\n');
694 static ssize_t msg_print_ext_body(char *buf, size_t size,
695 char *text, size_t text_len,
696 struct dev_printk_info *dev_info)
700 len = msg_add_ext_text(buf, size, text, text_len, '\n');
705 len += msg_add_dict_text(buf + len, size - len, "SUBSYSTEM",
706 dev_info->subsystem);
707 len += msg_add_dict_text(buf + len, size - len, "DEVICE",
713 /* /dev/kmsg - userspace message inject/listen interface */
714 struct devkmsg_user {
716 struct ratelimit_state rs;
718 char buf[CONSOLE_EXT_LOG_MAX];
720 struct printk_info info;
721 char text_buf[CONSOLE_EXT_LOG_MAX];
722 struct printk_record record;
725 static __printf(3, 4) __cold
726 int devkmsg_emit(int facility, int level, const char *fmt, ...)
732 r = vprintk_emit(facility, level, NULL, fmt, args);
738 static ssize_t devkmsg_write(struct kiocb *iocb, struct iov_iter *from)
741 int level = default_message_loglevel;
742 int facility = 1; /* LOG_USER */
743 struct file *file = iocb->ki_filp;
744 struct devkmsg_user *user = file->private_data;
745 size_t len = iov_iter_count(from);
748 if (!user || len > LOG_LINE_MAX)
751 /* Ignore when user logging is disabled. */
752 if (devkmsg_log & DEVKMSG_LOG_MASK_OFF)
755 /* Ratelimit when not explicitly enabled. */
756 if (!(devkmsg_log & DEVKMSG_LOG_MASK_ON)) {
757 if (!___ratelimit(&user->rs, current->comm))
761 buf = kmalloc(len+1, GFP_KERNEL);
766 if (!copy_from_iter_full(buf, len, from)) {
772 * Extract and skip the syslog prefix <[0-9]*>. Coming from userspace
773 * the decimal value represents 32bit, the lower 3 bit are the log
774 * level, the rest are the log facility.
776 * If no prefix or no userspace facility is specified, we
777 * enforce LOG_USER, to be able to reliably distinguish
778 * kernel-generated messages from userspace-injected ones.
781 if (line[0] == '<') {
785 u = simple_strtoul(line + 1, &endp, 10);
786 if (endp && endp[0] == '>') {
787 level = LOG_LEVEL(u);
788 if (LOG_FACILITY(u) != 0)
789 facility = LOG_FACILITY(u);
795 devkmsg_emit(facility, level, "%s", line);
800 static ssize_t devkmsg_read(struct file *file, char __user *buf,
801 size_t count, loff_t *ppos)
803 struct devkmsg_user *user = file->private_data;
804 struct printk_record *r = &user->record;
811 ret = mutex_lock_interruptible(&user->lock);
815 if (!prb_read_valid(prb, atomic64_read(&user->seq), r)) {
816 if (file->f_flags & O_NONBLOCK) {
822 * Guarantee this task is visible on the waitqueue before
823 * checking the wake condition.
825 * The full memory barrier within set_current_state() of
826 * prepare_to_wait_event() pairs with the full memory barrier
827 * within wq_has_sleeper().
829 * This pairs with __wake_up_klogd:A.
831 ret = wait_event_interruptible(log_wait,
833 atomic64_read(&user->seq), r)); /* LMM(devkmsg_read:A) */
838 if (r->info->seq != atomic64_read(&user->seq)) {
839 /* our last seen message is gone, return error and reset */
840 atomic64_set(&user->seq, r->info->seq);
845 len = info_print_ext_header(user->buf, sizeof(user->buf), r->info);
846 len += msg_print_ext_body(user->buf + len, sizeof(user->buf) - len,
847 &r->text_buf[0], r->info->text_len,
850 atomic64_set(&user->seq, r->info->seq + 1);
857 if (copy_to_user(buf, user->buf, len)) {
863 mutex_unlock(&user->lock);
868 * Be careful when modifying this function!!!
870 * Only few operations are supported because the device works only with the
871 * entire variable length messages (records). Non-standard values are
872 * returned in the other cases and has been this way for quite some time.
873 * User space applications might depend on this behavior.
875 static loff_t devkmsg_llseek(struct file *file, loff_t offset, int whence)
877 struct devkmsg_user *user = file->private_data;
887 /* the first record */
888 atomic64_set(&user->seq, prb_first_valid_seq(prb));
892 * The first record after the last SYSLOG_ACTION_CLEAR,
893 * like issued by 'dmesg -c'. Reading /dev/kmsg itself
894 * changes no global state, and does not clear anything.
896 atomic64_set(&user->seq, latched_seq_read_nolock(&clear_seq));
899 /* after the last record */
900 atomic64_set(&user->seq, prb_next_seq(prb));
908 static __poll_t devkmsg_poll(struct file *file, poll_table *wait)
910 struct devkmsg_user *user = file->private_data;
911 struct printk_info info;
915 return EPOLLERR|EPOLLNVAL;
917 poll_wait(file, &log_wait, wait);
919 if (prb_read_valid_info(prb, atomic64_read(&user->seq), &info, NULL)) {
920 /* return error when data has vanished underneath us */
921 if (info.seq != atomic64_read(&user->seq))
922 ret = EPOLLIN|EPOLLRDNORM|EPOLLERR|EPOLLPRI;
924 ret = EPOLLIN|EPOLLRDNORM;
930 static int devkmsg_open(struct inode *inode, struct file *file)
932 struct devkmsg_user *user;
935 if (devkmsg_log & DEVKMSG_LOG_MASK_OFF)
938 /* write-only does not need any file context */
939 if ((file->f_flags & O_ACCMODE) != O_WRONLY) {
940 err = check_syslog_permissions(SYSLOG_ACTION_READ_ALL,
946 user = kvmalloc(sizeof(struct devkmsg_user), GFP_KERNEL);
950 ratelimit_default_init(&user->rs);
951 ratelimit_set_flags(&user->rs, RATELIMIT_MSG_ON_RELEASE);
953 mutex_init(&user->lock);
955 prb_rec_init_rd(&user->record, &user->info,
956 &user->text_buf[0], sizeof(user->text_buf));
958 atomic64_set(&user->seq, prb_first_valid_seq(prb));
960 file->private_data = user;
964 static int devkmsg_release(struct inode *inode, struct file *file)
966 struct devkmsg_user *user = file->private_data;
971 ratelimit_state_exit(&user->rs);
973 mutex_destroy(&user->lock);
978 const struct file_operations kmsg_fops = {
979 .open = devkmsg_open,
980 .read = devkmsg_read,
981 .write_iter = devkmsg_write,
982 .llseek = devkmsg_llseek,
983 .poll = devkmsg_poll,
984 .release = devkmsg_release,
987 #ifdef CONFIG_CRASH_CORE
989 * This appends the listed symbols to /proc/vmcore
991 * /proc/vmcore is used by various utilities, like crash and makedumpfile to
992 * obtain access to symbols that are otherwise very difficult to locate. These
993 * symbols are specifically used so that utilities can access and extract the
994 * dmesg log from a vmcore file after a crash.
996 void log_buf_vmcoreinfo_setup(void)
998 struct dev_printk_info *dev_info = NULL;
1000 VMCOREINFO_SYMBOL(prb);
1001 VMCOREINFO_SYMBOL(printk_rb_static);
1002 VMCOREINFO_SYMBOL(clear_seq);
1005 * Export struct size and field offsets. User space tools can
1006 * parse it and detect any changes to structure down the line.
1009 VMCOREINFO_STRUCT_SIZE(printk_ringbuffer);
1010 VMCOREINFO_OFFSET(printk_ringbuffer, desc_ring);
1011 VMCOREINFO_OFFSET(printk_ringbuffer, text_data_ring);
1012 VMCOREINFO_OFFSET(printk_ringbuffer, fail);
1014 VMCOREINFO_STRUCT_SIZE(prb_desc_ring);
1015 VMCOREINFO_OFFSET(prb_desc_ring, count_bits);
1016 VMCOREINFO_OFFSET(prb_desc_ring, descs);
1017 VMCOREINFO_OFFSET(prb_desc_ring, infos);
1018 VMCOREINFO_OFFSET(prb_desc_ring, head_id);
1019 VMCOREINFO_OFFSET(prb_desc_ring, tail_id);
1021 VMCOREINFO_STRUCT_SIZE(prb_desc);
1022 VMCOREINFO_OFFSET(prb_desc, state_var);
1023 VMCOREINFO_OFFSET(prb_desc, text_blk_lpos);
1025 VMCOREINFO_STRUCT_SIZE(prb_data_blk_lpos);
1026 VMCOREINFO_OFFSET(prb_data_blk_lpos, begin);
1027 VMCOREINFO_OFFSET(prb_data_blk_lpos, next);
1029 VMCOREINFO_STRUCT_SIZE(printk_info);
1030 VMCOREINFO_OFFSET(printk_info, seq);
1031 VMCOREINFO_OFFSET(printk_info, ts_nsec);
1032 VMCOREINFO_OFFSET(printk_info, text_len);
1033 VMCOREINFO_OFFSET(printk_info, caller_id);
1034 VMCOREINFO_OFFSET(printk_info, dev_info);
1036 VMCOREINFO_STRUCT_SIZE(dev_printk_info);
1037 VMCOREINFO_OFFSET(dev_printk_info, subsystem);
1038 VMCOREINFO_LENGTH(printk_info_subsystem, sizeof(dev_info->subsystem));
1039 VMCOREINFO_OFFSET(dev_printk_info, device);
1040 VMCOREINFO_LENGTH(printk_info_device, sizeof(dev_info->device));
1042 VMCOREINFO_STRUCT_SIZE(prb_data_ring);
1043 VMCOREINFO_OFFSET(prb_data_ring, size_bits);
1044 VMCOREINFO_OFFSET(prb_data_ring, data);
1045 VMCOREINFO_OFFSET(prb_data_ring, head_lpos);
1046 VMCOREINFO_OFFSET(prb_data_ring, tail_lpos);
1048 VMCOREINFO_SIZE(atomic_long_t);
1049 VMCOREINFO_TYPE_OFFSET(atomic_long_t, counter);
1051 VMCOREINFO_STRUCT_SIZE(latched_seq);
1052 VMCOREINFO_OFFSET(latched_seq, val);
1056 /* requested log_buf_len from kernel cmdline */
1057 static unsigned long __initdata new_log_buf_len;
1059 /* we practice scaling the ring buffer by powers of 2 */
1060 static void __init log_buf_len_update(u64 size)
1062 if (size > (u64)LOG_BUF_LEN_MAX) {
1063 size = (u64)LOG_BUF_LEN_MAX;
1064 pr_err("log_buf over 2G is not supported.\n");
1068 size = roundup_pow_of_two(size);
1069 if (size > log_buf_len)
1070 new_log_buf_len = (unsigned long)size;
1073 /* save requested log_buf_len since it's too early to process it */
1074 static int __init log_buf_len_setup(char *str)
1081 size = memparse(str, &str);
1083 log_buf_len_update(size);
1087 early_param("log_buf_len", log_buf_len_setup);
1090 #define __LOG_CPU_MAX_BUF_LEN (1 << CONFIG_LOG_CPU_MAX_BUF_SHIFT)
1092 static void __init log_buf_add_cpu(void)
1094 unsigned int cpu_extra;
1097 * archs should set up cpu_possible_bits properly with
1098 * set_cpu_possible() after setup_arch() but just in
1099 * case lets ensure this is valid.
1101 if (num_possible_cpus() == 1)
1104 cpu_extra = (num_possible_cpus() - 1) * __LOG_CPU_MAX_BUF_LEN;
1106 /* by default this will only continue through for large > 64 CPUs */
1107 if (cpu_extra <= __LOG_BUF_LEN / 2)
1110 pr_info("log_buf_len individual max cpu contribution: %d bytes\n",
1111 __LOG_CPU_MAX_BUF_LEN);
1112 pr_info("log_buf_len total cpu_extra contributions: %d bytes\n",
1114 pr_info("log_buf_len min size: %d bytes\n", __LOG_BUF_LEN);
1116 log_buf_len_update(cpu_extra + __LOG_BUF_LEN);
1118 #else /* !CONFIG_SMP */
1119 static inline void log_buf_add_cpu(void) {}
1120 #endif /* CONFIG_SMP */
1122 static void __init set_percpu_data_ready(void)
1124 __printk_percpu_data_ready = true;
1127 static unsigned int __init add_to_rb(struct printk_ringbuffer *rb,
1128 struct printk_record *r)
1130 struct prb_reserved_entry e;
1131 struct printk_record dest_r;
1133 prb_rec_init_wr(&dest_r, r->info->text_len);
1135 if (!prb_reserve(&e, rb, &dest_r))
1138 memcpy(&dest_r.text_buf[0], &r->text_buf[0], r->info->text_len);
1139 dest_r.info->text_len = r->info->text_len;
1140 dest_r.info->facility = r->info->facility;
1141 dest_r.info->level = r->info->level;
1142 dest_r.info->flags = r->info->flags;
1143 dest_r.info->ts_nsec = r->info->ts_nsec;
1144 dest_r.info->caller_id = r->info->caller_id;
1145 memcpy(&dest_r.info->dev_info, &r->info->dev_info, sizeof(dest_r.info->dev_info));
1147 prb_final_commit(&e);
1149 return prb_record_text_space(&e);
1152 static char setup_text_buf[LOG_LINE_MAX] __initdata;
1154 void __init setup_log_buf(int early)
1156 struct printk_info *new_infos;
1157 unsigned int new_descs_count;
1158 struct prb_desc *new_descs;
1159 struct printk_info info;
1160 struct printk_record r;
1161 unsigned int text_size;
1162 size_t new_descs_size;
1163 size_t new_infos_size;
1164 unsigned long flags;
1170 * Some archs call setup_log_buf() multiple times - first is very
1171 * early, e.g. from setup_arch(), and second - when percpu_areas
1175 set_percpu_data_ready();
1177 if (log_buf != __log_buf)
1180 if (!early && !new_log_buf_len)
1183 if (!new_log_buf_len)
1186 new_descs_count = new_log_buf_len >> PRB_AVGBITS;
1187 if (new_descs_count == 0) {
1188 pr_err("new_log_buf_len: %lu too small\n", new_log_buf_len);
1192 new_log_buf = memblock_alloc(new_log_buf_len, LOG_ALIGN);
1193 if (unlikely(!new_log_buf)) {
1194 pr_err("log_buf_len: %lu text bytes not available\n",
1199 new_descs_size = new_descs_count * sizeof(struct prb_desc);
1200 new_descs = memblock_alloc(new_descs_size, LOG_ALIGN);
1201 if (unlikely(!new_descs)) {
1202 pr_err("log_buf_len: %zu desc bytes not available\n",
1204 goto err_free_log_buf;
1207 new_infos_size = new_descs_count * sizeof(struct printk_info);
1208 new_infos = memblock_alloc(new_infos_size, LOG_ALIGN);
1209 if (unlikely(!new_infos)) {
1210 pr_err("log_buf_len: %zu info bytes not available\n",
1212 goto err_free_descs;
1215 prb_rec_init_rd(&r, &info, &setup_text_buf[0], sizeof(setup_text_buf));
1217 prb_init(&printk_rb_dynamic,
1218 new_log_buf, ilog2(new_log_buf_len),
1219 new_descs, ilog2(new_descs_count),
1222 local_irq_save(flags);
1224 log_buf_len = new_log_buf_len;
1225 log_buf = new_log_buf;
1226 new_log_buf_len = 0;
1228 free = __LOG_BUF_LEN;
1229 prb_for_each_record(0, &printk_rb_static, seq, &r) {
1230 text_size = add_to_rb(&printk_rb_dynamic, &r);
1231 if (text_size > free)
1237 prb = &printk_rb_dynamic;
1239 local_irq_restore(flags);
1242 * Copy any remaining messages that might have appeared from
1243 * NMI context after copying but before switching to the
1246 prb_for_each_record(seq, &printk_rb_static, seq, &r) {
1247 text_size = add_to_rb(&printk_rb_dynamic, &r);
1248 if (text_size > free)
1254 if (seq != prb_next_seq(&printk_rb_static)) {
1255 pr_err("dropped %llu messages\n",
1256 prb_next_seq(&printk_rb_static) - seq);
1259 pr_info("log_buf_len: %u bytes\n", log_buf_len);
1260 pr_info("early log buf free: %u(%u%%)\n",
1261 free, (free * 100) / __LOG_BUF_LEN);
1265 memblock_free(new_descs, new_descs_size);
1267 memblock_free(new_log_buf, new_log_buf_len);
1270 static bool __read_mostly ignore_loglevel;
1272 static int __init ignore_loglevel_setup(char *str)
1274 ignore_loglevel = true;
1275 pr_info("debug: ignoring loglevel setting.\n");
1280 early_param("ignore_loglevel", ignore_loglevel_setup);
1281 module_param(ignore_loglevel, bool, S_IRUGO | S_IWUSR);
1282 MODULE_PARM_DESC(ignore_loglevel,
1283 "ignore loglevel setting (prints all kernel messages to the console)");
1285 static bool suppress_message_printing(int level)
1287 return (level >= console_loglevel && !ignore_loglevel);
1290 #ifdef CONFIG_BOOT_PRINTK_DELAY
1292 static int boot_delay; /* msecs delay after each printk during bootup */
1293 static unsigned long long loops_per_msec; /* based on boot_delay */
1295 static int __init boot_delay_setup(char *str)
1299 lpj = preset_lpj ? preset_lpj : 1000000; /* some guess */
1300 loops_per_msec = (unsigned long long)lpj / 1000 * HZ;
1302 get_option(&str, &boot_delay);
1303 if (boot_delay > 10 * 1000)
1306 pr_debug("boot_delay: %u, preset_lpj: %ld, lpj: %lu, "
1307 "HZ: %d, loops_per_msec: %llu\n",
1308 boot_delay, preset_lpj, lpj, HZ, loops_per_msec);
1311 early_param("boot_delay", boot_delay_setup);
1313 static void boot_delay_msec(int level)
1315 unsigned long long k;
1316 unsigned long timeout;
1318 if ((boot_delay == 0 || system_state >= SYSTEM_RUNNING)
1319 || suppress_message_printing(level)) {
1323 k = (unsigned long long)loops_per_msec * boot_delay;
1325 timeout = jiffies + msecs_to_jiffies(boot_delay);
1330 * use (volatile) jiffies to prevent
1331 * compiler reduction; loop termination via jiffies
1332 * is secondary and may or may not happen.
1334 if (time_after(jiffies, timeout))
1336 touch_nmi_watchdog();
1340 static inline void boot_delay_msec(int level)
1345 static bool printk_time = IS_ENABLED(CONFIG_PRINTK_TIME);
1346 module_param_named(time, printk_time, bool, S_IRUGO | S_IWUSR);
1348 static size_t print_syslog(unsigned int level, char *buf)
1350 return sprintf(buf, "<%u>", level);
1353 static size_t print_time(u64 ts, char *buf)
1355 unsigned long rem_nsec = do_div(ts, 1000000000);
1357 return sprintf(buf, "[%5lu.%06lu]",
1358 (unsigned long)ts, rem_nsec / 1000);
1361 #ifdef CONFIG_PRINTK_CALLER
1362 static size_t print_caller(u32 id, char *buf)
1366 snprintf(caller, sizeof(caller), "%c%u",
1367 id & 0x80000000 ? 'C' : 'T', id & ~0x80000000);
1368 return sprintf(buf, "[%6s]", caller);
1371 #define print_caller(id, buf) 0
1374 static size_t info_print_prefix(const struct printk_info *info, bool syslog,
1375 bool time, char *buf)
1380 len = print_syslog((info->facility << 3) | info->level, buf);
1383 len += print_time(info->ts_nsec, buf + len);
1385 len += print_caller(info->caller_id, buf + len);
1387 if (IS_ENABLED(CONFIG_PRINTK_CALLER) || time) {
1396 * Prepare the record for printing. The text is shifted within the given
1397 * buffer to avoid a need for another one. The following operations are
1400 * - Add prefix for each line.
1401 * - Drop truncated lines that no longer fit into the buffer.
1402 * - Add the trailing newline that has been removed in vprintk_store().
1403 * - Add a string terminator.
1405 * Since the produced string is always terminated, the maximum possible
1406 * return value is @r->text_buf_size - 1;
1408 * Return: The length of the updated/prepared text, including the added
1409 * prefixes and the newline. The terminator is not counted. The dropped
1410 * line(s) are not counted.
1412 static size_t record_print_text(struct printk_record *r, bool syslog,
1415 size_t text_len = r->info->text_len;
1416 size_t buf_size = r->text_buf_size;
1417 char *text = r->text_buf;
1418 char prefix[PREFIX_MAX];
1419 bool truncated = false;
1426 * If the message was truncated because the buffer was not large
1427 * enough, treat the available text as if it were the full text.
1429 if (text_len > buf_size)
1430 text_len = buf_size;
1432 prefix_len = info_print_prefix(r->info, syslog, time, prefix);
1435 * @text_len: bytes of unprocessed text
1436 * @line_len: bytes of current line _without_ newline
1437 * @text: pointer to beginning of current line
1438 * @len: number of bytes prepared in r->text_buf
1441 next = memchr(text, '\n', text_len);
1443 line_len = next - text;
1445 /* Drop truncated line(s). */
1448 line_len = text_len;
1452 * Truncate the text if there is not enough space to add the
1453 * prefix and a trailing newline and a terminator.
1455 if (len + prefix_len + text_len + 1 + 1 > buf_size) {
1456 /* Drop even the current line if no space. */
1457 if (len + prefix_len + line_len + 1 + 1 > buf_size)
1460 text_len = buf_size - len - prefix_len - 1 - 1;
1464 memmove(text + prefix_len, text, text_len);
1465 memcpy(text, prefix, prefix_len);
1468 * Increment the prepared length to include the text and
1469 * prefix that were just moved+copied. Also increment for the
1470 * newline at the end of this line. If this is the last line,
1471 * there is no newline, but it will be added immediately below.
1473 len += prefix_len + line_len + 1;
1474 if (text_len == line_len) {
1476 * This is the last line. Add the trailing newline
1477 * removed in vprintk_store().
1479 text[prefix_len + line_len] = '\n';
1484 * Advance beyond the added prefix and the related line with
1487 text += prefix_len + line_len + 1;
1490 * The remaining text has only decreased by the line with its
1493 * Note that @text_len can become zero. It happens when @text
1494 * ended with a newline (either due to truncation or the
1495 * original string ending with "\n\n"). The loop is correctly
1496 * repeated and (if not truncated) an empty line with a prefix
1499 text_len -= line_len + 1;
1503 * If a buffer was provided, it will be terminated. Space for the
1504 * string terminator is guaranteed to be available. The terminator is
1505 * not counted in the return value.
1508 r->text_buf[len] = 0;
1513 static size_t get_record_print_text_size(struct printk_info *info,
1514 unsigned int line_count,
1515 bool syslog, bool time)
1517 char prefix[PREFIX_MAX];
1520 prefix_len = info_print_prefix(info, syslog, time, prefix);
1523 * Each line will be preceded with a prefix. The intermediate
1524 * newlines are already within the text, but a final trailing
1525 * newline will be added.
1527 return ((prefix_len * line_count) + info->text_len + 1);
1531 * Beginning with @start_seq, find the first record where it and all following
1532 * records up to (but not including) @max_seq fit into @size.
1534 * @max_seq is simply an upper bound and does not need to exist. If the caller
1535 * does not require an upper bound, -1 can be used for @max_seq.
1537 static u64 find_first_fitting_seq(u64 start_seq, u64 max_seq, size_t size,
1538 bool syslog, bool time)
1540 struct printk_info info;
1541 unsigned int line_count;
1545 /* Determine the size of the records up to @max_seq. */
1546 prb_for_each_info(start_seq, prb, seq, &info, &line_count) {
1547 if (info.seq >= max_seq)
1549 len += get_record_print_text_size(&info, line_count, syslog, time);
1553 * Adjust the upper bound for the next loop to avoid subtracting
1554 * lengths that were never added.
1560 * Move first record forward until length fits into the buffer. Ignore
1561 * newest messages that were not counted in the above cycle. Messages
1562 * might appear and get lost in the meantime. This is a best effort
1563 * that prevents an infinite loop that could occur with a retry.
1565 prb_for_each_info(start_seq, prb, seq, &info, &line_count) {
1566 if (len <= size || info.seq >= max_seq)
1568 len -= get_record_print_text_size(&info, line_count, syslog, time);
1574 /* The caller is responsible for making sure @size is greater than 0. */
1575 static int syslog_print(char __user *buf, int size)
1577 struct printk_info info;
1578 struct printk_record r;
1583 text = kmalloc(CONSOLE_LOG_MAX, GFP_KERNEL);
1587 prb_rec_init_rd(&r, &info, text, CONSOLE_LOG_MAX);
1589 mutex_lock(&syslog_lock);
1592 * Wait for the @syslog_seq record to be available. @syslog_seq may
1593 * change while waiting.
1598 mutex_unlock(&syslog_lock);
1600 * Guarantee this task is visible on the waitqueue before
1601 * checking the wake condition.
1603 * The full memory barrier within set_current_state() of
1604 * prepare_to_wait_event() pairs with the full memory barrier
1605 * within wq_has_sleeper().
1607 * This pairs with __wake_up_klogd:A.
1609 len = wait_event_interruptible(log_wait,
1610 prb_read_valid(prb, seq, NULL)); /* LMM(syslog_print:A) */
1611 mutex_lock(&syslog_lock);
1615 } while (syslog_seq != seq);
1618 * Copy records that fit into the buffer. The above cycle makes sure
1619 * that the first record is always available.
1626 if (!prb_read_valid(prb, syslog_seq, &r))
1629 if (r.info->seq != syslog_seq) {
1630 /* message is gone, move to next valid one */
1631 syslog_seq = r.info->seq;
1636 * To keep reading/counting partial line consistent,
1637 * use printk_time value as of the beginning of a line.
1639 if (!syslog_partial)
1640 syslog_time = printk_time;
1642 skip = syslog_partial;
1643 n = record_print_text(&r, true, syslog_time);
1644 if (n - syslog_partial <= size) {
1645 /* message fits into buffer, move forward */
1646 syslog_seq = r.info->seq + 1;
1647 n -= syslog_partial;
1650 /* partial read(), remember position */
1652 syslog_partial += n;
1659 mutex_unlock(&syslog_lock);
1660 err = copy_to_user(buf, text + skip, n);
1661 mutex_lock(&syslog_lock);
1674 mutex_unlock(&syslog_lock);
1679 static int syslog_print_all(char __user *buf, int size, bool clear)
1681 struct printk_info info;
1682 struct printk_record r;
1688 text = kmalloc(CONSOLE_LOG_MAX, GFP_KERNEL);
1694 * Find first record that fits, including all following records,
1695 * into the user-provided buffer for this dump.
1697 seq = find_first_fitting_seq(latched_seq_read_nolock(&clear_seq), -1,
1700 prb_rec_init_rd(&r, &info, text, CONSOLE_LOG_MAX);
1703 prb_for_each_record(seq, prb, seq, &r) {
1706 textlen = record_print_text(&r, true, time);
1708 if (len + textlen > size) {
1713 if (copy_to_user(buf + len, text, textlen))
1723 mutex_lock(&syslog_lock);
1724 latched_seq_write(&clear_seq, seq);
1725 mutex_unlock(&syslog_lock);
1732 static void syslog_clear(void)
1734 mutex_lock(&syslog_lock);
1735 latched_seq_write(&clear_seq, prb_next_seq(prb));
1736 mutex_unlock(&syslog_lock);
1739 int do_syslog(int type, char __user *buf, int len, int source)
1741 struct printk_info info;
1743 static int saved_console_loglevel = LOGLEVEL_DEFAULT;
1746 error = check_syslog_permissions(type, source);
1751 case SYSLOG_ACTION_CLOSE: /* Close log */
1753 case SYSLOG_ACTION_OPEN: /* Open log */
1755 case SYSLOG_ACTION_READ: /* Read from log */
1756 if (!buf || len < 0)
1760 if (!access_ok(buf, len))
1762 error = syslog_print(buf, len);
1764 /* Read/clear last kernel messages */
1765 case SYSLOG_ACTION_READ_CLEAR:
1768 /* Read last kernel messages */
1769 case SYSLOG_ACTION_READ_ALL:
1770 if (!buf || len < 0)
1774 if (!access_ok(buf, len))
1776 error = syslog_print_all(buf, len, clear);
1778 /* Clear ring buffer */
1779 case SYSLOG_ACTION_CLEAR:
1782 /* Disable logging to console */
1783 case SYSLOG_ACTION_CONSOLE_OFF:
1784 if (saved_console_loglevel == LOGLEVEL_DEFAULT)
1785 saved_console_loglevel = console_loglevel;
1786 console_loglevel = minimum_console_loglevel;
1788 /* Enable logging to console */
1789 case SYSLOG_ACTION_CONSOLE_ON:
1790 if (saved_console_loglevel != LOGLEVEL_DEFAULT) {
1791 console_loglevel = saved_console_loglevel;
1792 saved_console_loglevel = LOGLEVEL_DEFAULT;
1795 /* Set level of messages printed to console */
1796 case SYSLOG_ACTION_CONSOLE_LEVEL:
1797 if (len < 1 || len > 8)
1799 if (len < minimum_console_loglevel)
1800 len = minimum_console_loglevel;
1801 console_loglevel = len;
1802 /* Implicitly re-enable logging to console */
1803 saved_console_loglevel = LOGLEVEL_DEFAULT;
1805 /* Number of chars in the log buffer */
1806 case SYSLOG_ACTION_SIZE_UNREAD:
1807 mutex_lock(&syslog_lock);
1808 if (!prb_read_valid_info(prb, syslog_seq, &info, NULL)) {
1809 /* No unread messages. */
1810 mutex_unlock(&syslog_lock);
1813 if (info.seq != syslog_seq) {
1814 /* messages are gone, move to first one */
1815 syslog_seq = info.seq;
1818 if (source == SYSLOG_FROM_PROC) {
1820 * Short-cut for poll(/"proc/kmsg") which simply checks
1821 * for pending data, not the size; return the count of
1822 * records, not the length.
1824 error = prb_next_seq(prb) - syslog_seq;
1826 bool time = syslog_partial ? syslog_time : printk_time;
1827 unsigned int line_count;
1830 prb_for_each_info(syslog_seq, prb, seq, &info,
1832 error += get_record_print_text_size(&info, line_count,
1836 error -= syslog_partial;
1838 mutex_unlock(&syslog_lock);
1840 /* Size of the log buffer */
1841 case SYSLOG_ACTION_SIZE_BUFFER:
1842 error = log_buf_len;
1852 SYSCALL_DEFINE3(syslog, int, type, char __user *, buf, int, len)
1854 return do_syslog(type, buf, len, SYSLOG_FROM_READER);
1858 * Special console_lock variants that help to reduce the risk of soft-lockups.
1859 * They allow to pass console_lock to another printk() call using a busy wait.
1862 #ifdef CONFIG_LOCKDEP
1863 static struct lockdep_map console_owner_dep_map = {
1864 .name = "console_owner"
1868 static DEFINE_RAW_SPINLOCK(console_owner_lock);
1869 static struct task_struct *console_owner;
1870 static bool console_waiter;
1873 * console_lock_spinning_enable - mark beginning of code where another
1874 * thread might safely busy wait
1876 * This basically converts console_lock into a spinlock. This marks
1877 * the section where the console_lock owner can not sleep, because
1878 * there may be a waiter spinning (like a spinlock). Also it must be
1879 * ready to hand over the lock at the end of the section.
1881 static void console_lock_spinning_enable(void)
1883 raw_spin_lock(&console_owner_lock);
1884 console_owner = current;
1885 raw_spin_unlock(&console_owner_lock);
1887 /* The waiter may spin on us after setting console_owner */
1888 spin_acquire(&console_owner_dep_map, 0, 0, _THIS_IP_);
1892 * console_lock_spinning_disable_and_check - mark end of code where another
1893 * thread was able to busy wait and check if there is a waiter
1895 * This is called at the end of the section where spinning is allowed.
1896 * It has two functions. First, it is a signal that it is no longer
1897 * safe to start busy waiting for the lock. Second, it checks if
1898 * there is a busy waiter and passes the lock rights to her.
1900 * Important: Callers lose both the console_lock and the SRCU read lock if
1901 * there was a busy waiter. They must not touch items synchronized by
1902 * console_lock or SRCU read lock in this case.
1904 * Return: 1 if the lock rights were passed, 0 otherwise.
1906 static int console_lock_spinning_disable_and_check(int cookie)
1910 raw_spin_lock(&console_owner_lock);
1911 waiter = READ_ONCE(console_waiter);
1912 console_owner = NULL;
1913 raw_spin_unlock(&console_owner_lock);
1916 spin_release(&console_owner_dep_map, _THIS_IP_);
1920 /* The waiter is now free to continue */
1921 WRITE_ONCE(console_waiter, false);
1923 spin_release(&console_owner_dep_map, _THIS_IP_);
1926 * Preserve lockdep lock ordering. Release the SRCU read lock before
1927 * releasing the console_lock.
1929 console_srcu_read_unlock(cookie);
1932 * Hand off console_lock to waiter. The waiter will perform
1933 * the up(). After this, the waiter is the console_lock owner.
1935 mutex_release(&console_lock_dep_map, _THIS_IP_);
1940 * console_trylock_spinning - try to get console_lock by busy waiting
1942 * This allows to busy wait for the console_lock when the current
1943 * owner is running in specially marked sections. It means that
1944 * the current owner is running and cannot reschedule until it
1945 * is ready to lose the lock.
1947 * Return: 1 if we got the lock, 0 othrewise
1949 static int console_trylock_spinning(void)
1951 struct task_struct *owner = NULL;
1954 unsigned long flags;
1956 if (console_trylock())
1960 * It's unsafe to spin once a panic has begun. If we are the
1961 * panic CPU, we may have already halted the owner of the
1962 * console_sem. If we are not the panic CPU, then we should
1963 * avoid taking console_sem, so the panic CPU has a better
1964 * chance of cleanly acquiring it later.
1966 if (panic_in_progress())
1969 printk_safe_enter_irqsave(flags);
1971 raw_spin_lock(&console_owner_lock);
1972 owner = READ_ONCE(console_owner);
1973 waiter = READ_ONCE(console_waiter);
1974 if (!waiter && owner && owner != current) {
1975 WRITE_ONCE(console_waiter, true);
1978 raw_spin_unlock(&console_owner_lock);
1981 * If there is an active printk() writing to the
1982 * consoles, instead of having it write our data too,
1983 * see if we can offload that load from the active
1984 * printer, and do some printing ourselves.
1985 * Go into a spin only if there isn't already a waiter
1986 * spinning, and there is an active printer, and
1987 * that active printer isn't us (recursive printk?).
1990 printk_safe_exit_irqrestore(flags);
1994 /* We spin waiting for the owner to release us */
1995 spin_acquire(&console_owner_dep_map, 0, 0, _THIS_IP_);
1996 /* Owner will clear console_waiter on hand off */
1997 while (READ_ONCE(console_waiter))
1999 spin_release(&console_owner_dep_map, _THIS_IP_);
2001 printk_safe_exit_irqrestore(flags);
2003 * The owner passed the console lock to us.
2004 * Since we did not spin on console lock, annotate
2005 * this as a trylock. Otherwise lockdep will
2008 mutex_acquire(&console_lock_dep_map, 0, 1, _THIS_IP_);
2014 * Call the specified console driver, asking it to write out the specified
2015 * text and length. If @dropped_text is non-NULL and any records have been
2016 * dropped, a dropped message will be written out first.
2018 static void call_console_driver(struct console *con, const char *text, size_t len,
2023 if (con->dropped && dropped_text) {
2024 dropped_len = snprintf(dropped_text, DROPPED_TEXT_MAX,
2025 "** %lu printk messages dropped **\n",
2028 con->write(con, dropped_text, dropped_len);
2031 con->write(con, text, len);
2035 * Recursion is tracked separately on each CPU. If NMIs are supported, an
2036 * additional NMI context per CPU is also separately tracked. Until per-CPU
2037 * is available, a separate "early tracking" is performed.
2039 static DEFINE_PER_CPU(u8, printk_count);
2040 static u8 printk_count_early;
2041 #ifdef CONFIG_HAVE_NMI
2042 static DEFINE_PER_CPU(u8, printk_count_nmi);
2043 static u8 printk_count_nmi_early;
2047 * Recursion is limited to keep the output sane. printk() should not require
2048 * more than 1 level of recursion (allowing, for example, printk() to trigger
2049 * a WARN), but a higher value is used in case some printk-internal errors
2050 * exist, such as the ringbuffer validation checks failing.
2052 #define PRINTK_MAX_RECURSION 3
2055 * Return a pointer to the dedicated counter for the CPU+context of the
2058 static u8 *__printk_recursion_counter(void)
2060 #ifdef CONFIG_HAVE_NMI
2062 if (printk_percpu_data_ready())
2063 return this_cpu_ptr(&printk_count_nmi);
2064 return &printk_count_nmi_early;
2067 if (printk_percpu_data_ready())
2068 return this_cpu_ptr(&printk_count);
2069 return &printk_count_early;
2073 * Enter recursion tracking. Interrupts are disabled to simplify tracking.
2074 * The caller must check the boolean return value to see if the recursion is
2075 * allowed. On failure, interrupts are not disabled.
2077 * @recursion_ptr must be a variable of type (u8 *) and is the same variable
2078 * that is passed to printk_exit_irqrestore().
2080 #define printk_enter_irqsave(recursion_ptr, flags) \
2082 bool success = true; \
2084 typecheck(u8 *, recursion_ptr); \
2085 local_irq_save(flags); \
2086 (recursion_ptr) = __printk_recursion_counter(); \
2087 if (*(recursion_ptr) > PRINTK_MAX_RECURSION) { \
2088 local_irq_restore(flags); \
2091 (*(recursion_ptr))++; \
2096 /* Exit recursion tracking, restoring interrupts. */
2097 #define printk_exit_irqrestore(recursion_ptr, flags) \
2099 typecheck(u8 *, recursion_ptr); \
2100 (*(recursion_ptr))--; \
2101 local_irq_restore(flags); \
2104 int printk_delay_msec __read_mostly;
2106 static inline void printk_delay(int level)
2108 boot_delay_msec(level);
2110 if (unlikely(printk_delay_msec)) {
2111 int m = printk_delay_msec;
2115 touch_nmi_watchdog();
2120 static inline u32 printk_caller_id(void)
2122 return in_task() ? task_pid_nr(current) :
2123 0x80000000 + smp_processor_id();
2127 * printk_parse_prefix - Parse level and control flags.
2129 * @text: The terminated text message.
2130 * @level: A pointer to the current level value, will be updated.
2131 * @flags: A pointer to the current printk_info flags, will be updated.
2133 * @level may be NULL if the caller is not interested in the parsed value.
2134 * Otherwise the variable pointed to by @level must be set to
2135 * LOGLEVEL_DEFAULT in order to be updated with the parsed value.
2137 * @flags may be NULL if the caller is not interested in the parsed value.
2138 * Otherwise the variable pointed to by @flags will be OR'd with the parsed
2141 * Return: The length of the parsed level and control flags.
2143 u16 printk_parse_prefix(const char *text, int *level,
2144 enum printk_info_flags *flags)
2150 kern_level = printk_get_level(text);
2154 switch (kern_level) {
2156 if (level && *level == LOGLEVEL_DEFAULT)
2157 *level = kern_level - '0';
2159 case 'c': /* KERN_CONT */
2172 static u16 printk_sprint(char *text, u16 size, int facility,
2173 enum printk_info_flags *flags, const char *fmt,
2178 text_len = vscnprintf(text, size, fmt, args);
2180 /* Mark and strip a trailing newline. */
2181 if (text_len && text[text_len - 1] == '\n') {
2183 *flags |= LOG_NEWLINE;
2186 /* Strip log level and control flags. */
2187 if (facility == 0) {
2190 prefix_len = printk_parse_prefix(text, NULL, NULL);
2192 text_len -= prefix_len;
2193 memmove(text, text + prefix_len, text_len);
2197 trace_console_rcuidle(text, text_len);
2203 int vprintk_store(int facility, int level,
2204 const struct dev_printk_info *dev_info,
2205 const char *fmt, va_list args)
2207 struct prb_reserved_entry e;
2208 enum printk_info_flags flags = 0;
2209 struct printk_record r;
2210 unsigned long irqflags;
2211 u16 trunc_msg_len = 0;
2221 if (!printk_enter_irqsave(recursion_ptr, irqflags))
2225 * Since the duration of printk() can vary depending on the message
2226 * and state of the ringbuffer, grab the timestamp now so that it is
2227 * close to the call of printk(). This provides a more deterministic
2228 * timestamp with respect to the caller.
2230 ts_nsec = local_clock();
2232 caller_id = printk_caller_id();
2235 * The sprintf needs to come first since the syslog prefix might be
2236 * passed in as a parameter. An extra byte must be reserved so that
2237 * later the vscnprintf() into the reserved buffer has room for the
2238 * terminating '\0', which is not counted by vsnprintf().
2240 va_copy(args2, args);
2241 reserve_size = vsnprintf(&prefix_buf[0], sizeof(prefix_buf), fmt, args2) + 1;
2244 if (reserve_size > LOG_LINE_MAX)
2245 reserve_size = LOG_LINE_MAX;
2247 /* Extract log level or control flags. */
2249 printk_parse_prefix(&prefix_buf[0], &level, &flags);
2251 if (level == LOGLEVEL_DEFAULT)
2252 level = default_message_loglevel;
2255 flags |= LOG_NEWLINE;
2257 if (flags & LOG_CONT) {
2258 prb_rec_init_wr(&r, reserve_size);
2259 if (prb_reserve_in_last(&e, prb, &r, caller_id, LOG_LINE_MAX)) {
2260 text_len = printk_sprint(&r.text_buf[r.info->text_len], reserve_size,
2261 facility, &flags, fmt, args);
2262 r.info->text_len += text_len;
2264 if (flags & LOG_NEWLINE) {
2265 r.info->flags |= LOG_NEWLINE;
2266 prb_final_commit(&e);
2277 * Explicitly initialize the record before every prb_reserve() call.
2278 * prb_reserve_in_last() and prb_reserve() purposely invalidate the
2279 * structure when they fail.
2281 prb_rec_init_wr(&r, reserve_size);
2282 if (!prb_reserve(&e, prb, &r)) {
2283 /* truncate the message if it is too long for empty buffer */
2284 truncate_msg(&reserve_size, &trunc_msg_len);
2286 prb_rec_init_wr(&r, reserve_size + trunc_msg_len);
2287 if (!prb_reserve(&e, prb, &r))
2292 text_len = printk_sprint(&r.text_buf[0], reserve_size, facility, &flags, fmt, args);
2294 memcpy(&r.text_buf[text_len], trunc_msg, trunc_msg_len);
2295 r.info->text_len = text_len + trunc_msg_len;
2296 r.info->facility = facility;
2297 r.info->level = level & 7;
2298 r.info->flags = flags & 0x1f;
2299 r.info->ts_nsec = ts_nsec;
2300 r.info->caller_id = caller_id;
2302 memcpy(&r.info->dev_info, dev_info, sizeof(r.info->dev_info));
2304 /* A message without a trailing newline can be continued. */
2305 if (!(flags & LOG_NEWLINE))
2308 prb_final_commit(&e);
2310 ret = text_len + trunc_msg_len;
2312 printk_exit_irqrestore(recursion_ptr, irqflags);
2316 asmlinkage int vprintk_emit(int facility, int level,
2317 const struct dev_printk_info *dev_info,
2318 const char *fmt, va_list args)
2321 bool in_sched = false;
2323 /* Suppress unimportant messages after panic happens */
2324 if (unlikely(suppress_printk))
2327 if (unlikely(suppress_panic_printk) &&
2328 atomic_read(&panic_cpu) != raw_smp_processor_id())
2331 if (level == LOGLEVEL_SCHED) {
2332 level = LOGLEVEL_DEFAULT;
2336 printk_delay(level);
2338 printed_len = vprintk_store(facility, level, dev_info, fmt, args);
2340 /* If called from the scheduler, we can not call up(). */
2343 * The caller may be holding system-critical or
2344 * timing-sensitive locks. Disable preemption during
2345 * printing of all remaining records to all consoles so that
2346 * this context can return as soon as possible. Hopefully
2347 * another printk() caller will take over the printing.
2351 * Try to acquire and then immediately release the console
2352 * semaphore. The release will print out buffers. With the
2353 * spinning variant, this context tries to take over the
2354 * printing from another printing context.
2356 if (console_trylock_spinning())
2364 EXPORT_SYMBOL(vprintk_emit);
2366 int vprintk_default(const char *fmt, va_list args)
2368 return vprintk_emit(0, LOGLEVEL_DEFAULT, NULL, fmt, args);
2370 EXPORT_SYMBOL_GPL(vprintk_default);
2372 asmlinkage __visible int _printk(const char *fmt, ...)
2377 va_start(args, fmt);
2378 r = vprintk(fmt, args);
2383 EXPORT_SYMBOL(_printk);
2385 static bool pr_flush(int timeout_ms, bool reset_on_progress);
2386 static bool __pr_flush(struct console *con, int timeout_ms, bool reset_on_progress);
2388 #else /* CONFIG_PRINTK */
2390 #define CONSOLE_LOG_MAX 0
2391 #define DROPPED_TEXT_MAX 0
2392 #define printk_time false
2394 #define prb_read_valid(rb, seq, r) false
2395 #define prb_first_valid_seq(rb) 0
2396 #define prb_next_seq(rb) 0
2398 static u64 syslog_seq;
2400 static size_t record_print_text(const struct printk_record *r,
2401 bool syslog, bool time)
2405 static ssize_t info_print_ext_header(char *buf, size_t size,
2406 struct printk_info *info)
2410 static ssize_t msg_print_ext_body(char *buf, size_t size,
2411 char *text, size_t text_len,
2412 struct dev_printk_info *dev_info) { return 0; }
2413 static void console_lock_spinning_enable(void) { }
2414 static int console_lock_spinning_disable_and_check(int cookie) { return 0; }
2415 static void call_console_driver(struct console *con, const char *text, size_t len,
2419 static bool suppress_message_printing(int level) { return false; }
2420 static bool pr_flush(int timeout_ms, bool reset_on_progress) { return true; }
2421 static bool __pr_flush(struct console *con, int timeout_ms, bool reset_on_progress) { return true; }
2423 #endif /* CONFIG_PRINTK */
2425 #ifdef CONFIG_EARLY_PRINTK
2426 struct console *early_console;
2428 asmlinkage __visible void early_printk(const char *fmt, ...)
2438 n = vscnprintf(buf, sizeof(buf), fmt, ap);
2441 early_console->write(early_console, buf, n);
2445 static void set_user_specified(struct console_cmdline *c, bool user_specified)
2447 if (!user_specified)
2451 * @c console was defined by the user on the command line.
2452 * Do not clear when added twice also by SPCR or the device tree.
2454 c->user_specified = true;
2455 /* At least one console defined by the user on the command line. */
2456 console_set_on_cmdline = 1;
2459 static int __add_preferred_console(char *name, int idx, char *options,
2460 char *brl_options, bool user_specified)
2462 struct console_cmdline *c;
2466 * See if this tty is not yet registered, and
2467 * if we have a slot free.
2469 for (i = 0, c = console_cmdline;
2470 i < MAX_CMDLINECONSOLES && c->name[0];
2472 if (strcmp(c->name, name) == 0 && c->index == idx) {
2474 preferred_console = i;
2475 set_user_specified(c, user_specified);
2479 if (i == MAX_CMDLINECONSOLES)
2482 preferred_console = i;
2483 strscpy(c->name, name, sizeof(c->name));
2484 c->options = options;
2485 set_user_specified(c, user_specified);
2486 braille_set_options(c, brl_options);
2492 static int __init console_msg_format_setup(char *str)
2494 if (!strcmp(str, "syslog"))
2495 console_msg_format = MSG_FORMAT_SYSLOG;
2496 if (!strcmp(str, "default"))
2497 console_msg_format = MSG_FORMAT_DEFAULT;
2500 __setup("console_msg_format=", console_msg_format_setup);
2503 * Set up a console. Called via do_early_param() in init/main.c
2504 * for each "console=" parameter in the boot command line.
2506 static int __init console_setup(char *str)
2508 char buf[sizeof(console_cmdline[0].name) + 4]; /* 4 for "ttyS" */
2509 char *s, *options, *brl_options = NULL;
2513 * console="" or console=null have been suggested as a way to
2514 * disable console output. Use ttynull that has been created
2515 * for exactly this purpose.
2517 if (str[0] == 0 || strcmp(str, "null") == 0) {
2518 __add_preferred_console("ttynull", 0, NULL, NULL, true);
2522 if (_braille_console_setup(&str, &brl_options))
2526 * Decode str into name, index, options.
2528 if (str[0] >= '0' && str[0] <= '9') {
2529 strcpy(buf, "ttyS");
2530 strncpy(buf + 4, str, sizeof(buf) - 5);
2532 strncpy(buf, str, sizeof(buf) - 1);
2534 buf[sizeof(buf) - 1] = 0;
2535 options = strchr(str, ',');
2539 if (!strcmp(str, "ttya"))
2540 strcpy(buf, "ttyS0");
2541 if (!strcmp(str, "ttyb"))
2542 strcpy(buf, "ttyS1");
2544 for (s = buf; *s; s++)
2545 if (isdigit(*s) || *s == ',')
2547 idx = simple_strtoul(s, NULL, 10);
2550 __add_preferred_console(buf, idx, options, brl_options, true);
2553 __setup("console=", console_setup);
2556 * add_preferred_console - add a device to the list of preferred consoles.
2557 * @name: device name
2558 * @idx: device index
2559 * @options: options for this console
2561 * The last preferred console added will be used for kernel messages
2562 * and stdin/out/err for init. Normally this is used by console_setup
2563 * above to handle user-supplied console arguments; however it can also
2564 * be used by arch-specific code either to override the user or more
2565 * commonly to provide a default console (ie from PROM variables) when
2566 * the user has not supplied one.
2568 int add_preferred_console(char *name, int idx, char *options)
2570 return __add_preferred_console(name, idx, options, NULL, false);
2573 bool console_suspend_enabled = true;
2574 EXPORT_SYMBOL(console_suspend_enabled);
2576 static int __init console_suspend_disable(char *str)
2578 console_suspend_enabled = false;
2581 __setup("no_console_suspend", console_suspend_disable);
2582 module_param_named(console_suspend, console_suspend_enabled,
2583 bool, S_IRUGO | S_IWUSR);
2584 MODULE_PARM_DESC(console_suspend, "suspend console during suspend"
2585 " and hibernate operations");
2587 static bool printk_console_no_auto_verbose;
2589 void console_verbose(void)
2591 if (console_loglevel && !printk_console_no_auto_verbose)
2592 console_loglevel = CONSOLE_LOGLEVEL_MOTORMOUTH;
2594 EXPORT_SYMBOL_GPL(console_verbose);
2596 module_param_named(console_no_auto_verbose, printk_console_no_auto_verbose, bool, 0644);
2597 MODULE_PARM_DESC(console_no_auto_verbose, "Disable console loglevel raise to highest on oops/panic/etc");
2600 * suspend_console - suspend the console subsystem
2602 * This disables printk() while we go into suspend states
2604 void suspend_console(void)
2606 if (!console_suspend_enabled)
2608 pr_info("Suspending console(s) (use no_console_suspend to debug)\n");
2609 pr_flush(1000, true);
2611 console_suspended = 1;
2615 void resume_console(void)
2617 if (!console_suspend_enabled)
2620 console_suspended = 0;
2622 pr_flush(1000, true);
2626 * console_cpu_notify - print deferred console messages after CPU hotplug
2629 * If printk() is called from a CPU that is not online yet, the messages
2630 * will be printed on the console only if there are CON_ANYTIME consoles.
2631 * This function is called when a new CPU comes online (or fails to come
2632 * up) or goes offline.
2634 static int console_cpu_notify(unsigned int cpu)
2636 if (!cpuhp_tasks_frozen) {
2637 /* If trylock fails, someone else is doing the printing */
2638 if (console_trylock())
2645 * console_lock - block the console subsystem from printing
2647 * Acquires a lock which guarantees that no consoles will
2648 * be in or enter their write() callback.
2650 * Can sleep, returns nothing.
2652 void console_lock(void)
2657 if (console_suspended)
2660 console_may_schedule = 1;
2662 EXPORT_SYMBOL(console_lock);
2665 * console_trylock - try to block the console subsystem from printing
2667 * Try to acquire a lock which guarantees that no consoles will
2668 * be in or enter their write() callback.
2670 * returns 1 on success, and 0 on failure to acquire the lock.
2672 int console_trylock(void)
2674 if (down_trylock_console_sem())
2676 if (console_suspended) {
2681 console_may_schedule = 0;
2684 EXPORT_SYMBOL(console_trylock);
2686 int is_console_locked(void)
2688 return console_locked;
2690 EXPORT_SYMBOL(is_console_locked);
2693 * Return true when this CPU should unlock console_sem without pushing all
2694 * messages to the console. This reduces the chance that the console is
2695 * locked when the panic CPU tries to use it.
2697 static bool abandon_console_lock_in_panic(void)
2699 if (!panic_in_progress())
2703 * We can use raw_smp_processor_id() here because it is impossible for
2704 * the task to be migrated to the panic_cpu, or away from it. If
2705 * panic_cpu has already been set, and we're not currently executing on
2706 * that CPU, then we never will be.
2708 return atomic_read(&panic_cpu) != raw_smp_processor_id();
2712 * Check if the given console is currently capable and allowed to print
2715 * Requires the console_srcu_read_lock.
2717 static inline bool console_is_usable(struct console *con)
2719 short flags = console_srcu_read_flags(con);
2721 if (!(flags & CON_ENABLED))
2728 * Console drivers may assume that per-cpu resources have been
2729 * allocated. So unless they're explicitly marked as being able to
2730 * cope (CON_ANYTIME) don't call them until this CPU is officially up.
2732 if (!cpu_online(raw_smp_processor_id()) && !(flags & CON_ANYTIME))
2738 static void __console_unlock(void)
2745 * Print one record for the given console. The record printed is whatever
2746 * record is the next available record for the given console.
2748 * @text is a buffer of size CONSOLE_LOG_MAX.
2750 * If extended messages should be printed, @ext_text is a buffer of size
2751 * CONSOLE_EXT_LOG_MAX. Otherwise @ext_text must be NULL.
2753 * If dropped messages should be printed, @dropped_text is a buffer of size
2754 * DROPPED_TEXT_MAX. Otherwise @dropped_text must be NULL.
2756 * @handover will be set to true if a printk waiter has taken over the
2757 * console_lock, in which case the caller is no longer holding both the
2758 * console_lock and the SRCU read lock. Otherwise it is set to false.
2760 * @cookie is the cookie from the SRCU read lock.
2762 * Returns false if the given console has no next record to print, otherwise
2765 * Requires the console_lock and the SRCU read lock.
2767 static bool console_emit_next_record(struct console *con, char *text, char *ext_text,
2768 char *dropped_text, bool *handover, int cookie)
2770 static int panic_console_dropped;
2771 struct printk_info info;
2772 struct printk_record r;
2773 unsigned long flags;
2777 prb_rec_init_rd(&r, &info, text, CONSOLE_LOG_MAX);
2781 if (!prb_read_valid(prb, con->seq, &r))
2784 if (con->seq != r.info->seq) {
2785 con->dropped += r.info->seq - con->seq;
2786 con->seq = r.info->seq;
2787 if (panic_in_progress() && panic_console_dropped++ > 10) {
2788 suppress_panic_printk = 1;
2789 pr_warn_once("Too many dropped messages. Suppress messages on non-panic CPUs to prevent livelock.\n");
2793 /* Skip record that has level above the console loglevel. */
2794 if (suppress_message_printing(r.info->level)) {
2800 write_text = ext_text;
2801 len = info_print_ext_header(ext_text, CONSOLE_EXT_LOG_MAX, r.info);
2802 len += msg_print_ext_body(ext_text + len, CONSOLE_EXT_LOG_MAX - len,
2803 &r.text_buf[0], r.info->text_len, &r.info->dev_info);
2806 len = record_print_text(&r, console_msg_format & MSG_FORMAT_SYSLOG, printk_time);
2810 * While actively printing out messages, if another printk()
2811 * were to occur on another CPU, it may wait for this one to
2812 * finish. This task can not be preempted if there is a
2813 * waiter waiting to take over.
2815 * Interrupts are disabled because the hand over to a waiter
2816 * must not be interrupted until the hand over is completed
2817 * (@console_waiter is cleared).
2819 printk_safe_enter_irqsave(flags);
2820 console_lock_spinning_enable();
2822 stop_critical_timings(); /* don't trace print latency */
2823 call_console_driver(con, write_text, len, dropped_text);
2824 start_critical_timings();
2828 *handover = console_lock_spinning_disable_and_check(cookie);
2829 printk_safe_exit_irqrestore(flags);
2835 * Print out all remaining records to all consoles.
2837 * @do_cond_resched is set by the caller. It can be true only in schedulable
2840 * @next_seq is set to the sequence number after the last available record.
2841 * The value is valid only when this function returns true. It means that all
2842 * usable consoles are completely flushed.
2844 * @handover will be set to true if a printk waiter has taken over the
2845 * console_lock, in which case the caller is no longer holding the
2846 * console_lock. Otherwise it is set to false.
2848 * Returns true when there was at least one usable console and all messages
2849 * were flushed to all usable consoles. A returned false informs the caller
2850 * that everything was not flushed (either there were no usable consoles or
2851 * another context has taken over printing or it is a panic situation and this
2852 * is not the panic CPU). Regardless the reason, the caller should assume it
2853 * is not useful to immediately try again.
2855 * Requires the console_lock.
2857 static bool console_flush_all(bool do_cond_resched, u64 *next_seq, bool *handover)
2859 static char dropped_text[DROPPED_TEXT_MAX];
2860 static char ext_text[CONSOLE_EXT_LOG_MAX];
2861 static char text[CONSOLE_LOG_MAX];
2862 bool any_usable = false;
2863 struct console *con;
2871 any_progress = false;
2873 cookie = console_srcu_read_lock();
2874 for_each_console_srcu(con) {
2877 if (!console_is_usable(con))
2881 if (console_srcu_read_flags(con) & CON_EXTENDED) {
2882 /* Extended consoles do not print "dropped messages". */
2883 progress = console_emit_next_record(con, &text[0],
2887 progress = console_emit_next_record(con, &text[0],
2888 NULL, &dropped_text[0],
2893 * If a handover has occurred, the SRCU read lock
2894 * is already released.
2899 /* Track the next of the highest seq flushed. */
2900 if (con->seq > *next_seq)
2901 *next_seq = con->seq;
2905 any_progress = true;
2907 /* Allow panic_cpu to take over the consoles safely. */
2908 if (abandon_console_lock_in_panic())
2911 if (do_cond_resched)
2914 console_srcu_read_unlock(cookie);
2915 } while (any_progress);
2920 console_srcu_read_unlock(cookie);
2925 * console_unlock - unblock the console subsystem from printing
2927 * Releases the console_lock which the caller holds to block printing of
2928 * the console subsystem.
2930 * While the console_lock was held, console output may have been buffered
2931 * by printk(). If this is the case, console_unlock(); emits
2932 * the output prior to releasing the lock.
2934 * console_unlock(); may be called from any context.
2936 void console_unlock(void)
2938 bool do_cond_resched;
2943 if (console_suspended) {
2949 * Console drivers are called with interrupts disabled, so
2950 * @console_may_schedule should be cleared before; however, we may
2951 * end up dumping a lot of lines, for example, if called from
2952 * console registration path, and should invoke cond_resched()
2953 * between lines if allowable. Not doing so can cause a very long
2954 * scheduling stall on a slow console leading to RCU stall and
2955 * softlockup warnings which exacerbate the issue with more
2956 * messages practically incapacitating the system. Therefore, create
2957 * a local to use for the printing loop.
2959 do_cond_resched = console_may_schedule;
2962 console_may_schedule = 0;
2964 flushed = console_flush_all(do_cond_resched, &next_seq, &handover);
2969 * Abort if there was a failure to flush all messages to all
2970 * usable consoles. Either it is not possible to flush (in
2971 * which case it would be an infinite loop of retrying) or
2972 * another context has taken over printing.
2978 * Some context may have added new records after
2979 * console_flush_all() but before unlocking the console.
2980 * Re-check if there is a new record to flush. If the trylock
2981 * fails, another context is already handling the printing.
2983 } while (prb_read_valid(prb, next_seq, NULL) && console_trylock());
2985 EXPORT_SYMBOL(console_unlock);
2988 * console_conditional_schedule - yield the CPU if required
2990 * If the console code is currently allowed to sleep, and
2991 * if this CPU should yield the CPU to another task, do
2994 * Must be called within console_lock();.
2996 void __sched console_conditional_schedule(void)
2998 if (console_may_schedule)
3001 EXPORT_SYMBOL(console_conditional_schedule);
3003 void console_unblank(void)
3009 * Stop console printing because the unblank() callback may
3010 * assume the console is not within its write() callback.
3012 * If @oops_in_progress is set, this may be an atomic context.
3013 * In that case, attempt a trylock as best-effort.
3015 if (oops_in_progress) {
3016 if (down_trylock_console_sem() != 0)
3022 console_may_schedule = 0;
3024 cookie = console_srcu_read_lock();
3025 for_each_console_srcu(c) {
3026 if ((console_srcu_read_flags(c) & CON_ENABLED) && c->unblank)
3029 console_srcu_read_unlock(cookie);
3033 if (!oops_in_progress)
3034 pr_flush(1000, true);
3038 * console_flush_on_panic - flush console content on panic
3039 * @mode: flush all messages in buffer or just the pending ones
3041 * Immediately output all pending messages no matter what.
3043 void console_flush_on_panic(enum con_flush_mode mode)
3046 * If someone else is holding the console lock, trylock will fail
3047 * and may_schedule may be set. Ignore and proceed to unlock so
3048 * that messages are flushed out. As this can be called from any
3049 * context and we don't want to get preempted while flushing,
3050 * ensure may_schedule is cleared.
3053 console_may_schedule = 0;
3055 if (mode == CONSOLE_REPLAY_ALL) {
3060 seq = prb_first_valid_seq(prb);
3062 cookie = console_srcu_read_lock();
3063 for_each_console_srcu(c) {
3065 * If the above console_trylock() failed, this is an
3066 * unsynchronized assignment. But in that case, the
3067 * kernel is in "hope and pray" mode anyway.
3071 console_srcu_read_unlock(cookie);
3077 * Return the console tty driver structure and its associated index
3079 struct tty_driver *console_device(int *index)
3082 struct tty_driver *driver = NULL;
3086 * Take console_lock to serialize device() callback with
3087 * other console operations. For example, fg_console is
3088 * modified under console_lock when switching vt.
3092 cookie = console_srcu_read_lock();
3093 for_each_console_srcu(c) {
3096 driver = c->device(c, index);
3100 console_srcu_read_unlock(cookie);
3107 * Prevent further output on the passed console device so that (for example)
3108 * serial drivers can disable console output before suspending a port, and can
3109 * re-enable output afterwards.
3111 void console_stop(struct console *console)
3113 __pr_flush(console, 1000, true);
3114 console_list_lock();
3115 console_srcu_write_flags(console, console->flags & ~CON_ENABLED);
3116 console_list_unlock();
3119 * Ensure that all SRCU list walks have completed. All contexts must
3120 * be able to see that this console is disabled so that (for example)
3121 * the caller can suspend the port without risk of another context
3124 synchronize_srcu(&console_srcu);
3126 EXPORT_SYMBOL(console_stop);
3128 void console_start(struct console *console)
3130 console_list_lock();
3131 console_srcu_write_flags(console, console->flags | CON_ENABLED);
3132 console_list_unlock();
3133 __pr_flush(console, 1000, true);
3135 EXPORT_SYMBOL(console_start);
3137 static int __read_mostly keep_bootcon;
3139 static int __init keep_bootcon_setup(char *str)
3142 pr_info("debug: skip boot console de-registration.\n");
3147 early_param("keep_bootcon", keep_bootcon_setup);
3150 * This is called by register_console() to try to match
3151 * the newly registered console with any of the ones selected
3152 * by either the command line or add_preferred_console() and
3155 * Care need to be taken with consoles that are statically
3156 * enabled such as netconsole
3158 static int try_enable_preferred_console(struct console *newcon,
3159 bool user_specified)
3161 struct console_cmdline *c;
3164 for (i = 0, c = console_cmdline;
3165 i < MAX_CMDLINECONSOLES && c->name[0];
3167 if (c->user_specified != user_specified)
3169 if (!newcon->match ||
3170 newcon->match(newcon, c->name, c->index, c->options) != 0) {
3171 /* default matching */
3172 BUILD_BUG_ON(sizeof(c->name) != sizeof(newcon->name));
3173 if (strcmp(c->name, newcon->name) != 0)
3175 if (newcon->index >= 0 &&
3176 newcon->index != c->index)
3178 if (newcon->index < 0)
3179 newcon->index = c->index;
3181 if (_braille_register_console(newcon, c))
3184 if (newcon->setup &&
3185 (err = newcon->setup(newcon, c->options)) != 0)
3188 newcon->flags |= CON_ENABLED;
3189 if (i == preferred_console)
3190 newcon->flags |= CON_CONSDEV;
3195 * Some consoles, such as pstore and netconsole, can be enabled even
3196 * without matching. Accept the pre-enabled consoles only when match()
3197 * and setup() had a chance to be called.
3199 if (newcon->flags & CON_ENABLED && c->user_specified == user_specified)
3205 /* Try to enable the console unconditionally */
3206 static void try_enable_default_console(struct console *newcon)
3208 if (newcon->index < 0)
3211 if (newcon->setup && newcon->setup(newcon, NULL) != 0)
3214 newcon->flags |= CON_ENABLED;
3217 newcon->flags |= CON_CONSDEV;
3220 #define con_printk(lvl, con, fmt, ...) \
3221 printk(lvl pr_fmt("%sconsole [%s%d] " fmt), \
3222 (con->flags & CON_BOOT) ? "boot" : "", \
3223 con->name, con->index, ##__VA_ARGS__)
3225 static void console_init_seq(struct console *newcon, bool bootcon_registered)
3227 struct console *con;
3230 if (newcon->flags & (CON_PRINTBUFFER | CON_BOOT)) {
3231 /* Get a consistent copy of @syslog_seq. */
3232 mutex_lock(&syslog_lock);
3233 newcon->seq = syslog_seq;
3234 mutex_unlock(&syslog_lock);
3236 /* Begin with next message added to ringbuffer. */
3237 newcon->seq = prb_next_seq(prb);
3240 * If any enabled boot consoles are due to be unregistered
3241 * shortly, some may not be caught up and may be the same
3242 * device as @newcon. Since it is not known which boot console
3243 * is the same device, flush all consoles and, if necessary,
3244 * start with the message of the enabled boot console that is
3245 * the furthest behind.
3247 if (bootcon_registered && !keep_bootcon) {
3249 * Hold the console_lock to stop console printing and
3250 * guarantee safe access to console->seq.
3255 * Flush all consoles and set the console to start at
3256 * the next unprinted sequence number.
3258 if (!console_flush_all(true, &newcon->seq, &handover)) {
3260 * Flushing failed. Just choose the lowest
3261 * sequence of the enabled boot consoles.
3265 * If there was a handover, this context no
3266 * longer holds the console_lock.
3271 newcon->seq = prb_next_seq(prb);
3272 for_each_console(con) {
3273 if ((con->flags & CON_BOOT) &&
3274 (con->flags & CON_ENABLED) &&
3275 con->seq < newcon->seq) {
3276 newcon->seq = con->seq;
3286 #define console_first() \
3287 hlist_entry(console_list.first, struct console, node)
3289 static int unregister_console_locked(struct console *console);
3292 * The console driver calls this routine during kernel initialization
3293 * to register the console printing procedure with printk() and to
3294 * print any messages that were printed by the kernel before the
3295 * console driver was initialized.
3297 * This can happen pretty early during the boot process (because of
3298 * early_printk) - sometimes before setup_arch() completes - be careful
3299 * of what kernel features are used - they may not be initialised yet.
3301 * There are two types of consoles - bootconsoles (early_printk) and
3302 * "real" consoles (everything which is not a bootconsole) which are
3303 * handled differently.
3304 * - Any number of bootconsoles can be registered at any time.
3305 * - As soon as a "real" console is registered, all bootconsoles
3306 * will be unregistered automatically.
3307 * - Once a "real" console is registered, any attempt to register a
3308 * bootconsoles will be rejected
3310 void register_console(struct console *newcon)
3312 struct console *con;
3313 bool bootcon_registered = false;
3314 bool realcon_registered = false;
3317 console_list_lock();
3319 for_each_console(con) {
3320 if (WARN(con == newcon, "console '%s%d' already registered\n",
3321 con->name, con->index)) {
3325 if (con->flags & CON_BOOT)
3326 bootcon_registered = true;
3328 realcon_registered = true;
3331 /* Do not register boot consoles when there already is a real one. */
3332 if ((newcon->flags & CON_BOOT) && realcon_registered) {
3333 pr_info("Too late to register bootconsole %s%d\n",
3334 newcon->name, newcon->index);
3339 * See if we want to enable this console driver by default.
3341 * Nope when a console is preferred by the command line, device
3344 * The first real console with tty binding (driver) wins. More
3345 * consoles might get enabled before the right one is found.
3347 * Note that a console with tty binding will have CON_CONSDEV
3348 * flag set and will be first in the list.
3350 if (preferred_console < 0) {
3351 if (hlist_empty(&console_list) || !console_first()->device ||
3352 console_first()->flags & CON_BOOT) {
3353 try_enable_default_console(newcon);
3357 /* See if this console matches one we selected on the command line */
3358 err = try_enable_preferred_console(newcon, true);
3360 /* If not, try to match against the platform default(s) */
3362 err = try_enable_preferred_console(newcon, false);
3364 /* printk() messages are not printed to the Braille console. */
3365 if (err || newcon->flags & CON_BRL)
3369 * If we have a bootconsole, and are switching to a real console,
3370 * don't print everything out again, since when the boot console, and
3371 * the real console are the same physical device, it's annoying to
3372 * see the beginning boot messages twice
3374 if (bootcon_registered &&
3375 ((newcon->flags & (CON_CONSDEV | CON_BOOT)) == CON_CONSDEV)) {
3376 newcon->flags &= ~CON_PRINTBUFFER;
3379 newcon->dropped = 0;
3380 console_init_seq(newcon, bootcon_registered);
3383 * Put this console in the list - keep the
3384 * preferred driver at the head of the list.
3386 if (hlist_empty(&console_list)) {
3387 /* Ensure CON_CONSDEV is always set for the head. */
3388 newcon->flags |= CON_CONSDEV;
3389 hlist_add_head_rcu(&newcon->node, &console_list);
3391 } else if (newcon->flags & CON_CONSDEV) {
3392 /* Only the new head can have CON_CONSDEV set. */
3393 console_srcu_write_flags(console_first(), console_first()->flags & ~CON_CONSDEV);
3394 hlist_add_head_rcu(&newcon->node, &console_list);
3397 hlist_add_behind_rcu(&newcon->node, console_list.first);
3401 * No need to synchronize SRCU here! The caller does not rely
3402 * on all contexts being able to see the new console before
3403 * register_console() completes.
3406 console_sysfs_notify();
3409 * By unregistering the bootconsoles after we enable the real console
3410 * we get the "console xxx enabled" message on all the consoles -
3411 * boot consoles, real consoles, etc - this is to ensure that end
3412 * users know there might be something in the kernel's log buffer that
3413 * went to the bootconsole (that they do not see on the real console)
3415 con_printk(KERN_INFO, newcon, "enabled\n");
3416 if (bootcon_registered &&
3417 ((newcon->flags & (CON_CONSDEV | CON_BOOT)) == CON_CONSDEV) &&
3419 struct hlist_node *tmp;
3421 hlist_for_each_entry_safe(con, tmp, &console_list, node) {
3422 if (con->flags & CON_BOOT)
3423 unregister_console_locked(con);
3427 console_list_unlock();
3429 EXPORT_SYMBOL(register_console);
3431 /* Must be called under console_list_lock(). */
3432 static int unregister_console_locked(struct console *console)
3436 lockdep_assert_console_list_lock_held();
3438 con_printk(KERN_INFO, console, "disabled\n");
3440 res = _braille_unregister_console(console);
3446 /* Disable it unconditionally */
3447 console_srcu_write_flags(console, console->flags & ~CON_ENABLED);
3449 if (!console_is_registered_locked(console))
3452 hlist_del_init_rcu(&console->node);
3456 * If this isn't the last console and it has CON_CONSDEV set, we
3457 * need to set it on the next preferred console.
3460 * The above makes no sense as there is no guarantee that the next
3461 * console has any device attached. Oh well....
3463 if (!hlist_empty(&console_list) && console->flags & CON_CONSDEV)
3464 console_srcu_write_flags(console_first(), console_first()->flags | CON_CONSDEV);
3467 * Ensure that all SRCU list walks have completed. All contexts
3468 * must not be able to see this console in the list so that any
3469 * exit/cleanup routines can be performed safely.
3471 synchronize_srcu(&console_srcu);
3473 console_sysfs_notify();
3476 res = console->exit(console);
3481 int unregister_console(struct console *console)
3485 console_list_lock();
3486 res = unregister_console_locked(console);
3487 console_list_unlock();
3490 EXPORT_SYMBOL(unregister_console);
3493 * console_force_preferred_locked - force a registered console preferred
3494 * @con: The registered console to force preferred.
3496 * Must be called under console_list_lock().
3498 void console_force_preferred_locked(struct console *con)
3500 struct console *cur_pref_con;
3502 if (!console_is_registered_locked(con))
3505 cur_pref_con = console_first();
3507 /* Already preferred? */
3508 if (cur_pref_con == con)
3512 * Delete, but do not re-initialize the entry. This allows the console
3513 * to continue to appear registered (via any hlist_unhashed_lockless()
3514 * checks), even though it was briefly removed from the console list.
3516 hlist_del_rcu(&con->node);
3519 * Ensure that all SRCU list walks have completed so that the console
3520 * can be added to the beginning of the console list and its forward
3521 * list pointer can be re-initialized.
3523 synchronize_srcu(&console_srcu);
3525 con->flags |= CON_CONSDEV;
3526 WARN_ON(!con->device);
3528 /* Only the new head can have CON_CONSDEV set. */
3529 console_srcu_write_flags(cur_pref_con, cur_pref_con->flags & ~CON_CONSDEV);
3530 hlist_add_head_rcu(&con->node, &console_list);
3532 EXPORT_SYMBOL(console_force_preferred_locked);
3535 * Initialize the console device. This is called *early*, so
3536 * we can't necessarily depend on lots of kernel help here.
3537 * Just do some early initializations, and do the complex setup
3540 void __init console_init(void)
3544 initcall_entry_t *ce;
3546 /* Setup the default TTY line discipline. */
3550 * set up the console device so that later boot sequences can
3551 * inform about problems etc..
3553 ce = __con_initcall_start;
3554 trace_initcall_level("console");
3555 while (ce < __con_initcall_end) {
3556 call = initcall_from_entry(ce);
3557 trace_initcall_start(call);
3559 trace_initcall_finish(call, ret);
3565 * Some boot consoles access data that is in the init section and which will
3566 * be discarded after the initcalls have been run. To make sure that no code
3567 * will access this data, unregister the boot consoles in a late initcall.
3569 * If for some reason, such as deferred probe or the driver being a loadable
3570 * module, the real console hasn't registered yet at this point, there will
3571 * be a brief interval in which no messages are logged to the console, which
3572 * makes it difficult to diagnose problems that occur during this time.
3574 * To mitigate this problem somewhat, only unregister consoles whose memory
3575 * intersects with the init section. Note that all other boot consoles will
3576 * get unregistered when the real preferred console is registered.
3578 static int __init printk_late_init(void)
3580 struct hlist_node *tmp;
3581 struct console *con;
3584 console_list_lock();
3585 hlist_for_each_entry_safe(con, tmp, &console_list, node) {
3586 if (!(con->flags & CON_BOOT))
3589 /* Check addresses that might be used for enabled consoles. */
3590 if (init_section_intersects(con, sizeof(*con)) ||
3591 init_section_contains(con->write, 0) ||
3592 init_section_contains(con->read, 0) ||
3593 init_section_contains(con->device, 0) ||
3594 init_section_contains(con->unblank, 0) ||
3595 init_section_contains(con->data, 0)) {
3597 * Please, consider moving the reported consoles out
3598 * of the init section.
3600 pr_warn("bootconsole [%s%d] uses init memory and must be disabled even before the real one is ready\n",
3601 con->name, con->index);
3602 unregister_console_locked(con);
3605 console_list_unlock();
3607 ret = cpuhp_setup_state_nocalls(CPUHP_PRINTK_DEAD, "printk:dead", NULL,
3608 console_cpu_notify);
3610 ret = cpuhp_setup_state_nocalls(CPUHP_AP_ONLINE_DYN, "printk:online",
3611 console_cpu_notify, NULL);
3613 printk_sysctl_init();
3616 late_initcall(printk_late_init);
3618 #if defined CONFIG_PRINTK
3619 /* If @con is specified, only wait for that console. Otherwise wait for all. */
3620 static bool __pr_flush(struct console *con, int timeout_ms, bool reset_on_progress)
3622 int remaining = timeout_ms;
3632 seq = prb_next_seq(prb);
3638 * Hold the console_lock to guarantee safe access to
3639 * console->seq and to prevent changes to @console_suspended
3640 * until all consoles have been processed.
3644 cookie = console_srcu_read_lock();
3645 for_each_console_srcu(c) {
3646 if (con && con != c)
3648 if (!console_is_usable(c))
3650 printk_seq = c->seq;
3651 if (printk_seq < seq)
3652 diff += seq - printk_seq;
3654 console_srcu_read_unlock(cookie);
3657 * If consoles are suspended, it cannot be expected that they
3658 * make forward progress, so timeout immediately. @diff is
3659 * still used to return a valid flush status.
3661 if (console_suspended)
3663 else if (diff != last_diff && reset_on_progress)
3664 remaining = timeout_ms;
3668 if (diff == 0 || remaining == 0)
3671 if (remaining < 0) {
3672 /* no timeout limit */
3674 } else if (remaining < 100) {
3689 * pr_flush() - Wait for printing threads to catch up.
3691 * @timeout_ms: The maximum time (in ms) to wait.
3692 * @reset_on_progress: Reset the timeout if forward progress is seen.
3694 * A value of 0 for @timeout_ms means no waiting will occur. A value of -1
3695 * represents infinite waiting.
3697 * If @reset_on_progress is true, the timeout will be reset whenever any
3698 * printer has been seen to make some forward progress.
3700 * Context: Process context. May sleep while acquiring console lock.
3701 * Return: true if all enabled printers are caught up.
3703 static bool pr_flush(int timeout_ms, bool reset_on_progress)
3705 return __pr_flush(NULL, timeout_ms, reset_on_progress);
3709 * Delayed printk version, for scheduler-internal messages:
3711 #define PRINTK_PENDING_WAKEUP 0x01
3712 #define PRINTK_PENDING_OUTPUT 0x02
3714 static DEFINE_PER_CPU(int, printk_pending);
3716 static void wake_up_klogd_work_func(struct irq_work *irq_work)
3718 int pending = this_cpu_xchg(printk_pending, 0);
3720 if (pending & PRINTK_PENDING_OUTPUT) {
3721 /* If trylock fails, someone else is doing the printing */
3722 if (console_trylock())
3726 if (pending & PRINTK_PENDING_WAKEUP)
3727 wake_up_interruptible(&log_wait);
3730 static DEFINE_PER_CPU(struct irq_work, wake_up_klogd_work) =
3731 IRQ_WORK_INIT_LAZY(wake_up_klogd_work_func);
3733 static void __wake_up_klogd(int val)
3735 if (!printk_percpu_data_ready())
3740 * Guarantee any new records can be seen by tasks preparing to wait
3741 * before this context checks if the wait queue is empty.
3743 * The full memory barrier within wq_has_sleeper() pairs with the full
3744 * memory barrier within set_current_state() of
3745 * prepare_to_wait_event(), which is called after ___wait_event() adds
3746 * the waiter but before it has checked the wait condition.
3748 * This pairs with devkmsg_read:A and syslog_print:A.
3750 if (wq_has_sleeper(&log_wait) || /* LMM(__wake_up_klogd:A) */
3751 (val & PRINTK_PENDING_OUTPUT)) {
3752 this_cpu_or(printk_pending, val);
3753 irq_work_queue(this_cpu_ptr(&wake_up_klogd_work));
3758 void wake_up_klogd(void)
3760 __wake_up_klogd(PRINTK_PENDING_WAKEUP);
3763 void defer_console_output(void)
3766 * New messages may have been added directly to the ringbuffer
3767 * using vprintk_store(), so wake any waiters as well.
3769 __wake_up_klogd(PRINTK_PENDING_WAKEUP | PRINTK_PENDING_OUTPUT);
3772 void printk_trigger_flush(void)
3774 defer_console_output();
3777 int vprintk_deferred(const char *fmt, va_list args)
3781 r = vprintk_emit(0, LOGLEVEL_SCHED, NULL, fmt, args);
3782 defer_console_output();
3787 int _printk_deferred(const char *fmt, ...)
3792 va_start(args, fmt);
3793 r = vprintk_deferred(fmt, args);
3800 * printk rate limiting, lifted from the networking subsystem.
3802 * This enforces a rate limit: not more than 10 kernel messages
3803 * every 5s to make a denial-of-service attack impossible.
3805 DEFINE_RATELIMIT_STATE(printk_ratelimit_state, 5 * HZ, 10);
3807 int __printk_ratelimit(const char *func)
3809 return ___ratelimit(&printk_ratelimit_state, func);
3811 EXPORT_SYMBOL(__printk_ratelimit);
3814 * printk_timed_ratelimit - caller-controlled printk ratelimiting
3815 * @caller_jiffies: pointer to caller's state
3816 * @interval_msecs: minimum interval between prints
3818 * printk_timed_ratelimit() returns true if more than @interval_msecs
3819 * milliseconds have elapsed since the last time printk_timed_ratelimit()
3822 bool printk_timed_ratelimit(unsigned long *caller_jiffies,
3823 unsigned int interval_msecs)
3825 unsigned long elapsed = jiffies - *caller_jiffies;
3827 if (*caller_jiffies && elapsed <= msecs_to_jiffies(interval_msecs))
3830 *caller_jiffies = jiffies;
3833 EXPORT_SYMBOL(printk_timed_ratelimit);
3835 static DEFINE_SPINLOCK(dump_list_lock);
3836 static LIST_HEAD(dump_list);
3839 * kmsg_dump_register - register a kernel log dumper.
3840 * @dumper: pointer to the kmsg_dumper structure
3842 * Adds a kernel log dumper to the system. The dump callback in the
3843 * structure will be called when the kernel oopses or panics and must be
3844 * set. Returns zero on success and %-EINVAL or %-EBUSY otherwise.
3846 int kmsg_dump_register(struct kmsg_dumper *dumper)
3848 unsigned long flags;
3851 /* The dump callback needs to be set */
3855 spin_lock_irqsave(&dump_list_lock, flags);
3856 /* Don't allow registering multiple times */
3857 if (!dumper->registered) {
3858 dumper->registered = 1;
3859 list_add_tail_rcu(&dumper->list, &dump_list);
3862 spin_unlock_irqrestore(&dump_list_lock, flags);
3866 EXPORT_SYMBOL_GPL(kmsg_dump_register);
3869 * kmsg_dump_unregister - unregister a kmsg dumper.
3870 * @dumper: pointer to the kmsg_dumper structure
3872 * Removes a dump device from the system. Returns zero on success and
3873 * %-EINVAL otherwise.
3875 int kmsg_dump_unregister(struct kmsg_dumper *dumper)
3877 unsigned long flags;
3880 spin_lock_irqsave(&dump_list_lock, flags);
3881 if (dumper->registered) {
3882 dumper->registered = 0;
3883 list_del_rcu(&dumper->list);
3886 spin_unlock_irqrestore(&dump_list_lock, flags);
3891 EXPORT_SYMBOL_GPL(kmsg_dump_unregister);
3893 static bool always_kmsg_dump;
3894 module_param_named(always_kmsg_dump, always_kmsg_dump, bool, S_IRUGO | S_IWUSR);
3896 const char *kmsg_dump_reason_str(enum kmsg_dump_reason reason)
3899 case KMSG_DUMP_PANIC:
3901 case KMSG_DUMP_OOPS:
3903 case KMSG_DUMP_EMERG:
3905 case KMSG_DUMP_SHUTDOWN:
3911 EXPORT_SYMBOL_GPL(kmsg_dump_reason_str);
3914 * kmsg_dump - dump kernel log to kernel message dumpers.
3915 * @reason: the reason (oops, panic etc) for dumping
3917 * Call each of the registered dumper's dump() callback, which can
3918 * retrieve the kmsg records with kmsg_dump_get_line() or
3919 * kmsg_dump_get_buffer().
3921 void kmsg_dump(enum kmsg_dump_reason reason)
3923 struct kmsg_dumper *dumper;
3926 list_for_each_entry_rcu(dumper, &dump_list, list) {
3927 enum kmsg_dump_reason max_reason = dumper->max_reason;
3930 * If client has not provided a specific max_reason, default
3931 * to KMSG_DUMP_OOPS, unless always_kmsg_dump was set.
3933 if (max_reason == KMSG_DUMP_UNDEF) {
3934 max_reason = always_kmsg_dump ? KMSG_DUMP_MAX :
3937 if (reason > max_reason)
3940 /* invoke dumper which will iterate over records */
3941 dumper->dump(dumper, reason);
3947 * kmsg_dump_get_line - retrieve one kmsg log line
3948 * @iter: kmsg dump iterator
3949 * @syslog: include the "<4>" prefixes
3950 * @line: buffer to copy the line to
3951 * @size: maximum size of the buffer
3952 * @len: length of line placed into buffer
3954 * Start at the beginning of the kmsg buffer, with the oldest kmsg
3955 * record, and copy one record into the provided buffer.
3957 * Consecutive calls will return the next available record moving
3958 * towards the end of the buffer with the youngest messages.
3960 * A return value of FALSE indicates that there are no more records to
3963 bool kmsg_dump_get_line(struct kmsg_dump_iter *iter, bool syslog,
3964 char *line, size_t size, size_t *len)
3966 u64 min_seq = latched_seq_read_nolock(&clear_seq);
3967 struct printk_info info;
3968 unsigned int line_count;
3969 struct printk_record r;
3973 if (iter->cur_seq < min_seq)
3974 iter->cur_seq = min_seq;
3976 prb_rec_init_rd(&r, &info, line, size);
3978 /* Read text or count text lines? */
3980 if (!prb_read_valid(prb, iter->cur_seq, &r))
3982 l = record_print_text(&r, syslog, printk_time);
3984 if (!prb_read_valid_info(prb, iter->cur_seq,
3985 &info, &line_count)) {
3988 l = get_record_print_text_size(&info, line_count, syslog,
3993 iter->cur_seq = r.info->seq + 1;
4000 EXPORT_SYMBOL_GPL(kmsg_dump_get_line);
4003 * kmsg_dump_get_buffer - copy kmsg log lines
4004 * @iter: kmsg dump iterator
4005 * @syslog: include the "<4>" prefixes
4006 * @buf: buffer to copy the line to
4007 * @size: maximum size of the buffer
4008 * @len_out: length of line placed into buffer
4010 * Start at the end of the kmsg buffer and fill the provided buffer
4011 * with as many of the *youngest* kmsg records that fit into it.
4012 * If the buffer is large enough, all available kmsg records will be
4013 * copied with a single call.
4015 * Consecutive calls will fill the buffer with the next block of
4016 * available older records, not including the earlier retrieved ones.
4018 * A return value of FALSE indicates that there are no more records to
4021 bool kmsg_dump_get_buffer(struct kmsg_dump_iter *iter, bool syslog,
4022 char *buf, size_t size, size_t *len_out)
4024 u64 min_seq = latched_seq_read_nolock(&clear_seq);
4025 struct printk_info info;
4026 struct printk_record r;
4031 bool time = printk_time;
4036 if (iter->cur_seq < min_seq)
4037 iter->cur_seq = min_seq;
4039 if (prb_read_valid_info(prb, iter->cur_seq, &info, NULL)) {
4040 if (info.seq != iter->cur_seq) {
4041 /* messages are gone, move to first available one */
4042 iter->cur_seq = info.seq;
4047 if (iter->cur_seq >= iter->next_seq)
4051 * Find first record that fits, including all following records,
4052 * into the user-provided buffer for this dump. Pass in size-1
4053 * because this function (by way of record_print_text()) will
4054 * not write more than size-1 bytes of text into @buf.
4056 seq = find_first_fitting_seq(iter->cur_seq, iter->next_seq,
4057 size - 1, syslog, time);
4060 * Next kmsg_dump_get_buffer() invocation will dump block of
4061 * older records stored right before this one.
4065 prb_rec_init_rd(&r, &info, buf, size);
4068 prb_for_each_record(seq, prb, seq, &r) {
4069 if (r.info->seq >= iter->next_seq)
4072 len += record_print_text(&r, syslog, time);
4074 /* Adjust record to store to remaining buffer space. */
4075 prb_rec_init_rd(&r, &info, buf + len, size - len);
4078 iter->next_seq = next_seq;
4085 EXPORT_SYMBOL_GPL(kmsg_dump_get_buffer);
4088 * kmsg_dump_rewind - reset the iterator
4089 * @iter: kmsg dump iterator
4091 * Reset the dumper's iterator so that kmsg_dump_get_line() and
4092 * kmsg_dump_get_buffer() can be called again and used multiple
4093 * times within the same dumper.dump() callback.
4095 void kmsg_dump_rewind(struct kmsg_dump_iter *iter)
4097 iter->cur_seq = latched_seq_read_nolock(&clear_seq);
4098 iter->next_seq = prb_next_seq(prb);
4100 EXPORT_SYMBOL_GPL(kmsg_dump_rewind);
4105 static atomic_t printk_cpu_sync_owner = ATOMIC_INIT(-1);
4106 static atomic_t printk_cpu_sync_nested = ATOMIC_INIT(0);
4109 * __printk_cpu_sync_wait() - Busy wait until the printk cpu-reentrant
4110 * spinning lock is not owned by any CPU.
4112 * Context: Any context.
4114 void __printk_cpu_sync_wait(void)
4118 } while (atomic_read(&printk_cpu_sync_owner) != -1);
4120 EXPORT_SYMBOL(__printk_cpu_sync_wait);
4123 * __printk_cpu_sync_try_get() - Try to acquire the printk cpu-reentrant
4126 * If no processor has the lock, the calling processor takes the lock and
4127 * becomes the owner. If the calling processor is already the owner of the
4128 * lock, this function succeeds immediately.
4130 * Context: Any context. Expects interrupts to be disabled.
4131 * Return: 1 on success, otherwise 0.
4133 int __printk_cpu_sync_try_get(void)
4138 cpu = smp_processor_id();
4141 * Guarantee loads and stores from this CPU when it is the lock owner
4142 * are _not_ visible to the previous lock owner. This pairs with
4143 * __printk_cpu_sync_put:B.
4145 * Memory barrier involvement:
4147 * If __printk_cpu_sync_try_get:A reads from __printk_cpu_sync_put:B,
4148 * then __printk_cpu_sync_put:A can never read from
4149 * __printk_cpu_sync_try_get:B.
4153 * RELEASE from __printk_cpu_sync_put:A to __printk_cpu_sync_put:B
4154 * of the previous CPU
4156 * ACQUIRE from __printk_cpu_sync_try_get:A to
4157 * __printk_cpu_sync_try_get:B of this CPU
4159 old = atomic_cmpxchg_acquire(&printk_cpu_sync_owner, -1,
4160 cpu); /* LMM(__printk_cpu_sync_try_get:A) */
4163 * This CPU is now the owner and begins loading/storing
4164 * data: LMM(__printk_cpu_sync_try_get:B)
4168 } else if (old == cpu) {
4169 /* This CPU is already the owner. */
4170 atomic_inc(&printk_cpu_sync_nested);
4176 EXPORT_SYMBOL(__printk_cpu_sync_try_get);
4179 * __printk_cpu_sync_put() - Release the printk cpu-reentrant spinning lock.
4181 * The calling processor must be the owner of the lock.
4183 * Context: Any context. Expects interrupts to be disabled.
4185 void __printk_cpu_sync_put(void)
4187 if (atomic_read(&printk_cpu_sync_nested)) {
4188 atomic_dec(&printk_cpu_sync_nested);
4193 * This CPU is finished loading/storing data:
4194 * LMM(__printk_cpu_sync_put:A)
4198 * Guarantee loads and stores from this CPU when it was the
4199 * lock owner are visible to the next lock owner. This pairs
4200 * with __printk_cpu_sync_try_get:A.
4202 * Memory barrier involvement:
4204 * If __printk_cpu_sync_try_get:A reads from __printk_cpu_sync_put:B,
4205 * then __printk_cpu_sync_try_get:B reads from __printk_cpu_sync_put:A.
4209 * RELEASE from __printk_cpu_sync_put:A to __printk_cpu_sync_put:B
4212 * ACQUIRE from __printk_cpu_sync_try_get:A to
4213 * __printk_cpu_sync_try_get:B of the next CPU
4215 atomic_set_release(&printk_cpu_sync_owner,
4216 -1); /* LMM(__printk_cpu_sync_put:B) */
4218 EXPORT_SYMBOL(__printk_cpu_sync_put);
4219 #endif /* CONFIG_SMP */