1 // SPDX-License-Identifier: GPL-2.0-only
3 * linux/kernel/printk.c
5 * Copyright (C) 1991, 1992 Linus Torvalds
7 * Modified to make sys_syslog() more flexible: added commands to
8 * return the last 4k of kernel messages, regardless of whether
9 * they've been read or not. Added option to suppress kernel printk's
10 * to the console. Added hook for sending the console messages
11 * elsewhere, in preparation for a serial line console (someday).
13 * Modified for sysctl support, 1/8/97, Chris Horn.
14 * Fixed SMP synchronization, 08/08/99, Manfred Spraul
15 * manfred@colorfullife.com
16 * Rewrote bits to get rid of console_lock
17 * 01Mar01 Andrew Morton
20 #define pr_fmt(fmt) KBUILD_MODNAME ": " fmt
22 #include <linux/kernel.h>
24 #include <linux/tty.h>
25 #include <linux/tty_driver.h>
26 #include <linux/console.h>
27 #include <linux/init.h>
28 #include <linux/jiffies.h>
29 #include <linux/nmi.h>
30 #include <linux/module.h>
31 #include <linux/moduleparam.h>
32 #include <linux/delay.h>
33 #include <linux/smp.h>
34 #include <linux/security.h>
35 #include <linux/memblock.h>
36 #include <linux/syscalls.h>
37 #include <linux/crash_core.h>
38 #include <linux/ratelimit.h>
39 #include <linux/kmsg_dump.h>
40 #include <linux/syslog.h>
41 #include <linux/cpu.h>
42 #include <linux/rculist.h>
43 #include <linux/poll.h>
44 #include <linux/irq_work.h>
45 #include <linux/ctype.h>
46 #include <linux/uio.h>
47 #include <linux/sched/clock.h>
48 #include <linux/sched/debug.h>
49 #include <linux/sched/task_stack.h>
51 #include <linux/uaccess.h>
52 #include <asm/sections.h>
54 #include <trace/events/initcall.h>
55 #define CREATE_TRACE_POINTS
56 #include <trace/events/printk.h>
58 #include "printk_ringbuffer.h"
59 #include "console_cmdline.h"
63 int console_printk[4] = {
64 CONSOLE_LOGLEVEL_DEFAULT, /* console_loglevel */
65 MESSAGE_LOGLEVEL_DEFAULT, /* default_message_loglevel */
66 CONSOLE_LOGLEVEL_MIN, /* minimum_console_loglevel */
67 CONSOLE_LOGLEVEL_DEFAULT, /* default_console_loglevel */
69 EXPORT_SYMBOL_GPL(console_printk);
71 atomic_t ignore_console_lock_warning __read_mostly = ATOMIC_INIT(0);
72 EXPORT_SYMBOL(ignore_console_lock_warning);
74 EXPORT_TRACEPOINT_SYMBOL_GPL(console);
77 * Low level drivers may need that to know if they can schedule in
78 * their unblank() callback or not. So let's export it.
81 EXPORT_SYMBOL(oops_in_progress);
84 * console_mutex protects console_list updates and console->flags updates.
85 * The flags are synchronized only for consoles that are registered, i.e.
86 * accessible via the console list.
88 static DEFINE_MUTEX(console_mutex);
91 * console_sem protects updates to console->seq
92 * and also provides serialization for console printing.
94 static DEFINE_SEMAPHORE(console_sem, 1);
95 HLIST_HEAD(console_list);
96 EXPORT_SYMBOL_GPL(console_list);
97 DEFINE_STATIC_SRCU(console_srcu);
100 * System may need to suppress printk message under certain
101 * circumstances, like after kernel panic happens.
103 int __read_mostly suppress_printk;
105 #ifdef CONFIG_LOCKDEP
106 static struct lockdep_map console_lock_dep_map = {
107 .name = "console_lock"
110 void lockdep_assert_console_list_lock_held(void)
112 lockdep_assert_held(&console_mutex);
114 EXPORT_SYMBOL(lockdep_assert_console_list_lock_held);
117 #ifdef CONFIG_DEBUG_LOCK_ALLOC
118 bool console_srcu_read_lock_is_held(void)
120 return srcu_read_lock_held(&console_srcu);
122 EXPORT_SYMBOL(console_srcu_read_lock_is_held);
125 enum devkmsg_log_bits {
126 __DEVKMSG_LOG_BIT_ON = 0,
127 __DEVKMSG_LOG_BIT_OFF,
128 __DEVKMSG_LOG_BIT_LOCK,
131 enum devkmsg_log_masks {
132 DEVKMSG_LOG_MASK_ON = BIT(__DEVKMSG_LOG_BIT_ON),
133 DEVKMSG_LOG_MASK_OFF = BIT(__DEVKMSG_LOG_BIT_OFF),
134 DEVKMSG_LOG_MASK_LOCK = BIT(__DEVKMSG_LOG_BIT_LOCK),
137 /* Keep both the 'on' and 'off' bits clear, i.e. ratelimit by default: */
138 #define DEVKMSG_LOG_MASK_DEFAULT 0
140 static unsigned int __read_mostly devkmsg_log = DEVKMSG_LOG_MASK_DEFAULT;
142 static int __control_devkmsg(char *str)
149 len = str_has_prefix(str, "on");
151 devkmsg_log = DEVKMSG_LOG_MASK_ON;
155 len = str_has_prefix(str, "off");
157 devkmsg_log = DEVKMSG_LOG_MASK_OFF;
161 len = str_has_prefix(str, "ratelimit");
163 devkmsg_log = DEVKMSG_LOG_MASK_DEFAULT;
170 static int __init control_devkmsg(char *str)
172 if (__control_devkmsg(str) < 0) {
173 pr_warn("printk.devkmsg: bad option string '%s'\n", str);
178 * Set sysctl string accordingly:
180 if (devkmsg_log == DEVKMSG_LOG_MASK_ON)
181 strcpy(devkmsg_log_str, "on");
182 else if (devkmsg_log == DEVKMSG_LOG_MASK_OFF)
183 strcpy(devkmsg_log_str, "off");
184 /* else "ratelimit" which is set by default. */
187 * Sysctl cannot change it anymore. The kernel command line setting of
188 * this parameter is to force the setting to be permanent throughout the
189 * runtime of the system. This is a precation measure against userspace
190 * trying to be a smarta** and attempting to change it up on us.
192 devkmsg_log |= DEVKMSG_LOG_MASK_LOCK;
196 __setup("printk.devkmsg=", control_devkmsg);
198 char devkmsg_log_str[DEVKMSG_STR_MAX_SIZE] = "ratelimit";
199 #if defined(CONFIG_PRINTK) && defined(CONFIG_SYSCTL)
200 int devkmsg_sysctl_set_loglvl(struct ctl_table *table, int write,
201 void *buffer, size_t *lenp, loff_t *ppos)
203 char old_str[DEVKMSG_STR_MAX_SIZE];
208 if (devkmsg_log & DEVKMSG_LOG_MASK_LOCK)
212 strncpy(old_str, devkmsg_log_str, DEVKMSG_STR_MAX_SIZE);
215 err = proc_dostring(table, write, buffer, lenp, ppos);
220 err = __control_devkmsg(devkmsg_log_str);
223 * Do not accept an unknown string OR a known string with
226 if (err < 0 || (err + 1 != *lenp)) {
228 /* ... and restore old setting. */
230 strncpy(devkmsg_log_str, old_str, DEVKMSG_STR_MAX_SIZE);
238 #endif /* CONFIG_PRINTK && CONFIG_SYSCTL */
241 * console_list_lock - Lock the console list
243 * For console list or console->flags updates
245 void console_list_lock(void)
248 * In unregister_console() and console_force_preferred_locked(),
249 * synchronize_srcu() is called with the console_list_lock held.
250 * Therefore it is not allowed that the console_list_lock is taken
251 * with the srcu_lock held.
253 * Detecting if this context is really in the read-side critical
254 * section is only possible if the appropriate debug options are
257 WARN_ON_ONCE(debug_lockdep_rcu_enabled() &&
258 srcu_read_lock_held(&console_srcu));
260 mutex_lock(&console_mutex);
262 EXPORT_SYMBOL(console_list_lock);
265 * console_list_unlock - Unlock the console list
267 * Counterpart to console_list_lock()
269 void console_list_unlock(void)
271 mutex_unlock(&console_mutex);
273 EXPORT_SYMBOL(console_list_unlock);
276 * console_srcu_read_lock - Register a new reader for the
277 * SRCU-protected console list
279 * Use for_each_console_srcu() to iterate the console list
281 * Context: Any context.
282 * Return: A cookie to pass to console_srcu_read_unlock().
284 int console_srcu_read_lock(void)
286 return srcu_read_lock_nmisafe(&console_srcu);
288 EXPORT_SYMBOL(console_srcu_read_lock);
291 * console_srcu_read_unlock - Unregister an old reader from
292 * the SRCU-protected console list
293 * @cookie: cookie returned from console_srcu_read_lock()
295 * Counterpart to console_srcu_read_lock()
297 void console_srcu_read_unlock(int cookie)
299 srcu_read_unlock_nmisafe(&console_srcu, cookie);
301 EXPORT_SYMBOL(console_srcu_read_unlock);
304 * Helper macros to handle lockdep when locking/unlocking console_sem. We use
305 * macros instead of functions so that _RET_IP_ contains useful information.
307 #define down_console_sem() do { \
309 mutex_acquire(&console_lock_dep_map, 0, 0, _RET_IP_);\
312 static int __down_trylock_console_sem(unsigned long ip)
318 * Here and in __up_console_sem() we need to be in safe mode,
319 * because spindump/WARN/etc from under console ->lock will
320 * deadlock in printk()->down_trylock_console_sem() otherwise.
322 printk_safe_enter_irqsave(flags);
323 lock_failed = down_trylock(&console_sem);
324 printk_safe_exit_irqrestore(flags);
328 mutex_acquire(&console_lock_dep_map, 0, 1, ip);
331 #define down_trylock_console_sem() __down_trylock_console_sem(_RET_IP_)
333 static void __up_console_sem(unsigned long ip)
337 mutex_release(&console_lock_dep_map, ip);
339 printk_safe_enter_irqsave(flags);
341 printk_safe_exit_irqrestore(flags);
343 #define up_console_sem() __up_console_sem(_RET_IP_)
345 static bool panic_in_progress(void)
347 return unlikely(atomic_read(&panic_cpu) != PANIC_CPU_INVALID);
350 /* Return true if a panic is in progress on the current CPU. */
351 bool this_cpu_in_panic(void)
354 * We can use raw_smp_processor_id() here because it is impossible for
355 * the task to be migrated to the panic_cpu, or away from it. If
356 * panic_cpu has already been set, and we're not currently executing on
357 * that CPU, then we never will be.
359 return unlikely(atomic_read(&panic_cpu) == raw_smp_processor_id());
363 * Return true if a panic is in progress on a remote CPU.
365 * On true, the local CPU should immediately release any printing resources
366 * that may be needed by the panic CPU.
368 bool other_cpu_in_panic(void)
370 return (panic_in_progress() && !this_cpu_in_panic());
374 * This is used for debugging the mess that is the VT code by
375 * keeping track if we have the console semaphore held. It's
376 * definitely not the perfect debug tool (we don't know if _WE_
377 * hold it and are racing, but it helps tracking those weird code
378 * paths in the console code where we end up in places I want
379 * locked without the console semaphore held).
381 static int console_locked;
384 * Array of consoles built from command line options (console=)
387 #define MAX_CMDLINECONSOLES 8
389 static struct console_cmdline console_cmdline[MAX_CMDLINECONSOLES];
391 static int preferred_console = -1;
392 int console_set_on_cmdline;
393 EXPORT_SYMBOL(console_set_on_cmdline);
395 /* Flag: console code may call schedule() */
396 static int console_may_schedule;
398 enum con_msg_format_flags {
399 MSG_FORMAT_DEFAULT = 0,
400 MSG_FORMAT_SYSLOG = (1 << 0),
403 static int console_msg_format = MSG_FORMAT_DEFAULT;
406 * The printk log buffer consists of a sequenced collection of records, each
407 * containing variable length message text. Every record also contains its
408 * own meta-data (@info).
410 * Every record meta-data carries the timestamp in microseconds, as well as
411 * the standard userspace syslog level and syslog facility. The usual kernel
412 * messages use LOG_KERN; userspace-injected messages always carry a matching
413 * syslog facility, by default LOG_USER. The origin of every message can be
414 * reliably determined that way.
416 * The human readable log message of a record is available in @text, the
417 * length of the message text in @text_len. The stored message is not
420 * Optionally, a record can carry a dictionary of properties (key/value
421 * pairs), to provide userspace with a machine-readable message context.
423 * Examples for well-defined, commonly used property names are:
424 * DEVICE=b12:8 device identifier
428 * +sound:card0 subsystem:devname
429 * SUBSYSTEM=pci driver-core subsystem name
431 * Valid characters in property names are [a-zA-Z0-9.-_]. Property names
432 * and values are terminated by a '\0' character.
434 * Example of record values:
435 * record.text_buf = "it's a line" (unterminated)
436 * record.info.seq = 56
437 * record.info.ts_nsec = 36863
438 * record.info.text_len = 11
439 * record.info.facility = 0 (LOG_KERN)
440 * record.info.flags = 0
441 * record.info.level = 3 (LOG_ERR)
442 * record.info.caller_id = 299 (task 299)
443 * record.info.dev_info.subsystem = "pci" (terminated)
444 * record.info.dev_info.device = "+pci:0000:00:01.0" (terminated)
446 * The 'struct printk_info' buffer must never be directly exported to
447 * userspace, it is a kernel-private implementation detail that might
448 * need to be changed in the future, when the requirements change.
450 * /dev/kmsg exports the structured data in the following line format:
451 * "<level>,<sequnum>,<timestamp>,<contflag>[,additional_values, ... ];<message text>\n"
453 * Users of the export format should ignore possible additional values
454 * separated by ',', and find the message after the ';' character.
456 * The optional key/value pairs are attached as continuation lines starting
457 * with a space character and terminated by a newline. All possible
458 * non-prinatable characters are escaped in the "\xff" notation.
461 /* syslog_lock protects syslog_* variables and write access to clear_seq. */
462 static DEFINE_MUTEX(syslog_lock);
466 * During panic, heavy printk by other CPUs can delay the
467 * panic and risk deadlock on console resources.
469 static int __read_mostly suppress_panic_printk;
471 DECLARE_WAIT_QUEUE_HEAD(log_wait);
472 /* All 3 protected by @syslog_lock. */
473 /* the next printk record to read by syslog(READ) or /proc/kmsg */
474 static u64 syslog_seq;
475 static size_t syslog_partial;
476 static bool syslog_time;
479 seqcount_latch_t latch;
484 * The next printk record to read after the last 'clear' command. There are
485 * two copies (updated with seqcount_latch) so that reads can locklessly
486 * access a valid value. Writers are synchronized by @syslog_lock.
488 static struct latched_seq clear_seq = {
489 .latch = SEQCNT_LATCH_ZERO(clear_seq.latch),
494 #define LOG_LEVEL(v) ((v) & 0x07)
495 #define LOG_FACILITY(v) ((v) >> 3 & 0xff)
498 #define LOG_ALIGN __alignof__(unsigned long)
499 #define __LOG_BUF_LEN (1 << CONFIG_LOG_BUF_SHIFT)
500 #define LOG_BUF_LEN_MAX (u32)(1 << 31)
501 static char __log_buf[__LOG_BUF_LEN] __aligned(LOG_ALIGN);
502 static char *log_buf = __log_buf;
503 static u32 log_buf_len = __LOG_BUF_LEN;
506 * Define the average message size. This only affects the number of
507 * descriptors that will be available. Underestimating is better than
508 * overestimating (too many available descriptors is better than not enough).
510 #define PRB_AVGBITS 5 /* 32 character average length */
512 #if CONFIG_LOG_BUF_SHIFT <= PRB_AVGBITS
513 #error CONFIG_LOG_BUF_SHIFT value too small.
515 _DEFINE_PRINTKRB(printk_rb_static, CONFIG_LOG_BUF_SHIFT - PRB_AVGBITS,
516 PRB_AVGBITS, &__log_buf[0]);
518 static struct printk_ringbuffer printk_rb_dynamic;
520 struct printk_ringbuffer *prb = &printk_rb_static;
523 * We cannot access per-CPU data (e.g. per-CPU flush irq_work) before
524 * per_cpu_areas are initialised. This variable is set to true when
525 * it's safe to access per-CPU data.
527 static bool __printk_percpu_data_ready __ro_after_init;
529 bool printk_percpu_data_ready(void)
531 return __printk_percpu_data_ready;
534 /* Must be called under syslog_lock. */
535 static void latched_seq_write(struct latched_seq *ls, u64 val)
537 raw_write_seqcount_latch(&ls->latch);
539 raw_write_seqcount_latch(&ls->latch);
543 /* Can be called from any context. */
544 static u64 latched_seq_read_nolock(struct latched_seq *ls)
551 seq = raw_read_seqcount_latch(&ls->latch);
554 } while (raw_read_seqcount_latch_retry(&ls->latch, seq));
559 /* Return log buffer address */
560 char *log_buf_addr_get(void)
565 /* Return log buffer size */
566 u32 log_buf_len_get(void)
572 * Define how much of the log buffer we could take at maximum. The value
573 * must be greater than two. Note that only half of the buffer is available
574 * when the index points to the middle.
576 #define MAX_LOG_TAKE_PART 4
577 static const char trunc_msg[] = "<truncated>";
579 static void truncate_msg(u16 *text_len, u16 *trunc_msg_len)
582 * The message should not take the whole buffer. Otherwise, it might
583 * get removed too soon.
585 u32 max_text_len = log_buf_len / MAX_LOG_TAKE_PART;
587 if (*text_len > max_text_len)
588 *text_len = max_text_len;
590 /* enable the warning message (if there is room) */
591 *trunc_msg_len = strlen(trunc_msg);
592 if (*text_len >= *trunc_msg_len)
593 *text_len -= *trunc_msg_len;
598 int dmesg_restrict = IS_ENABLED(CONFIG_SECURITY_DMESG_RESTRICT);
600 static int syslog_action_restricted(int type)
605 * Unless restricted, we allow "read all" and "get buffer size"
608 return type != SYSLOG_ACTION_READ_ALL &&
609 type != SYSLOG_ACTION_SIZE_BUFFER;
612 static int check_syslog_permissions(int type, int source)
615 * If this is from /proc/kmsg and we've already opened it, then we've
616 * already done the capabilities checks at open time.
618 if (source == SYSLOG_FROM_PROC && type != SYSLOG_ACTION_OPEN)
621 if (syslog_action_restricted(type)) {
622 if (capable(CAP_SYSLOG))
625 * For historical reasons, accept CAP_SYS_ADMIN too, with
628 if (capable(CAP_SYS_ADMIN)) {
629 pr_warn_once("%s (%d): Attempt to access syslog with "
630 "CAP_SYS_ADMIN but no CAP_SYSLOG "
632 current->comm, task_pid_nr(current));
638 return security_syslog(type);
641 static void append_char(char **pp, char *e, char c)
647 static ssize_t info_print_ext_header(char *buf, size_t size,
648 struct printk_info *info)
650 u64 ts_usec = info->ts_nsec;
652 #ifdef CONFIG_PRINTK_CALLER
653 u32 id = info->caller_id;
655 snprintf(caller, sizeof(caller), ",caller=%c%u",
656 id & 0x80000000 ? 'C' : 'T', id & ~0x80000000);
661 do_div(ts_usec, 1000);
663 return scnprintf(buf, size, "%u,%llu,%llu,%c%s;",
664 (info->facility << 3) | info->level, info->seq,
665 ts_usec, info->flags & LOG_CONT ? 'c' : '-', caller);
668 static ssize_t msg_add_ext_text(char *buf, size_t size,
669 const char *text, size_t text_len,
672 char *p = buf, *e = buf + size;
675 /* escape non-printable characters */
676 for (i = 0; i < text_len; i++) {
677 unsigned char c = text[i];
679 if (c < ' ' || c >= 127 || c == '\\')
680 p += scnprintf(p, e - p, "\\x%02x", c);
682 append_char(&p, e, c);
684 append_char(&p, e, endc);
689 static ssize_t msg_add_dict_text(char *buf, size_t size,
690 const char *key, const char *val)
692 size_t val_len = strlen(val);
698 len = msg_add_ext_text(buf, size, "", 0, ' '); /* dict prefix */
699 len += msg_add_ext_text(buf + len, size - len, key, strlen(key), '=');
700 len += msg_add_ext_text(buf + len, size - len, val, val_len, '\n');
705 static ssize_t msg_print_ext_body(char *buf, size_t size,
706 char *text, size_t text_len,
707 struct dev_printk_info *dev_info)
711 len = msg_add_ext_text(buf, size, text, text_len, '\n');
716 len += msg_add_dict_text(buf + len, size - len, "SUBSYSTEM",
717 dev_info->subsystem);
718 len += msg_add_dict_text(buf + len, size - len, "DEVICE",
724 /* /dev/kmsg - userspace message inject/listen interface */
725 struct devkmsg_user {
727 struct ratelimit_state rs;
729 struct printk_buffers pbufs;
732 static __printf(3, 4) __cold
733 int devkmsg_emit(int facility, int level, const char *fmt, ...)
739 r = vprintk_emit(facility, level, NULL, fmt, args);
745 static ssize_t devkmsg_write(struct kiocb *iocb, struct iov_iter *from)
748 int level = default_message_loglevel;
749 int facility = 1; /* LOG_USER */
750 struct file *file = iocb->ki_filp;
751 struct devkmsg_user *user = file->private_data;
752 size_t len = iov_iter_count(from);
755 if (len > PRINTKRB_RECORD_MAX)
758 /* Ignore when user logging is disabled. */
759 if (devkmsg_log & DEVKMSG_LOG_MASK_OFF)
762 /* Ratelimit when not explicitly enabled. */
763 if (!(devkmsg_log & DEVKMSG_LOG_MASK_ON)) {
764 if (!___ratelimit(&user->rs, current->comm))
768 buf = kmalloc(len+1, GFP_KERNEL);
773 if (!copy_from_iter_full(buf, len, from)) {
779 * Extract and skip the syslog prefix <[0-9]*>. Coming from userspace
780 * the decimal value represents 32bit, the lower 3 bit are the log
781 * level, the rest are the log facility.
783 * If no prefix or no userspace facility is specified, we
784 * enforce LOG_USER, to be able to reliably distinguish
785 * kernel-generated messages from userspace-injected ones.
788 if (line[0] == '<') {
792 u = simple_strtoul(line + 1, &endp, 10);
793 if (endp && endp[0] == '>') {
794 level = LOG_LEVEL(u);
795 if (LOG_FACILITY(u) != 0)
796 facility = LOG_FACILITY(u);
802 devkmsg_emit(facility, level, "%s", line);
807 static ssize_t devkmsg_read(struct file *file, char __user *buf,
808 size_t count, loff_t *ppos)
810 struct devkmsg_user *user = file->private_data;
811 char *outbuf = &user->pbufs.outbuf[0];
812 struct printk_message pmsg = {
813 .pbufs = &user->pbufs,
817 ret = mutex_lock_interruptible(&user->lock);
821 if (!printk_get_next_message(&pmsg, atomic64_read(&user->seq), true, false)) {
822 if (file->f_flags & O_NONBLOCK) {
828 * Guarantee this task is visible on the waitqueue before
829 * checking the wake condition.
831 * The full memory barrier within set_current_state() of
832 * prepare_to_wait_event() pairs with the full memory barrier
833 * within wq_has_sleeper().
835 * This pairs with __wake_up_klogd:A.
837 ret = wait_event_interruptible(log_wait,
838 printk_get_next_message(&pmsg, atomic64_read(&user->seq), true,
839 false)); /* LMM(devkmsg_read:A) */
845 /* our last seen message is gone, return error and reset */
846 atomic64_set(&user->seq, pmsg.seq);
851 atomic64_set(&user->seq, pmsg.seq + 1);
853 if (pmsg.outbuf_len > count) {
858 if (copy_to_user(buf, outbuf, pmsg.outbuf_len)) {
862 ret = pmsg.outbuf_len;
864 mutex_unlock(&user->lock);
869 * Be careful when modifying this function!!!
871 * Only few operations are supported because the device works only with the
872 * entire variable length messages (records). Non-standard values are
873 * returned in the other cases and has been this way for quite some time.
874 * User space applications might depend on this behavior.
876 static loff_t devkmsg_llseek(struct file *file, loff_t offset, int whence)
878 struct devkmsg_user *user = file->private_data;
886 /* the first record */
887 atomic64_set(&user->seq, prb_first_valid_seq(prb));
891 * The first record after the last SYSLOG_ACTION_CLEAR,
892 * like issued by 'dmesg -c'. Reading /dev/kmsg itself
893 * changes no global state, and does not clear anything.
895 atomic64_set(&user->seq, latched_seq_read_nolock(&clear_seq));
898 /* after the last record */
899 atomic64_set(&user->seq, prb_next_seq(prb));
907 static __poll_t devkmsg_poll(struct file *file, poll_table *wait)
909 struct devkmsg_user *user = file->private_data;
910 struct printk_info info;
913 poll_wait(file, &log_wait, wait);
915 if (prb_read_valid_info(prb, atomic64_read(&user->seq), &info, NULL)) {
916 /* return error when data has vanished underneath us */
917 if (info.seq != atomic64_read(&user->seq))
918 ret = EPOLLIN|EPOLLRDNORM|EPOLLERR|EPOLLPRI;
920 ret = EPOLLIN|EPOLLRDNORM;
926 static int devkmsg_open(struct inode *inode, struct file *file)
928 struct devkmsg_user *user;
931 if (devkmsg_log & DEVKMSG_LOG_MASK_OFF)
934 /* write-only does not need any file context */
935 if ((file->f_flags & O_ACCMODE) != O_WRONLY) {
936 err = check_syslog_permissions(SYSLOG_ACTION_READ_ALL,
942 user = kvmalloc(sizeof(struct devkmsg_user), GFP_KERNEL);
946 ratelimit_default_init(&user->rs);
947 ratelimit_set_flags(&user->rs, RATELIMIT_MSG_ON_RELEASE);
949 mutex_init(&user->lock);
951 atomic64_set(&user->seq, prb_first_valid_seq(prb));
953 file->private_data = user;
957 static int devkmsg_release(struct inode *inode, struct file *file)
959 struct devkmsg_user *user = file->private_data;
961 ratelimit_state_exit(&user->rs);
963 mutex_destroy(&user->lock);
968 const struct file_operations kmsg_fops = {
969 .open = devkmsg_open,
970 .read = devkmsg_read,
971 .write_iter = devkmsg_write,
972 .llseek = devkmsg_llseek,
973 .poll = devkmsg_poll,
974 .release = devkmsg_release,
977 #ifdef CONFIG_CRASH_CORE
979 * This appends the listed symbols to /proc/vmcore
981 * /proc/vmcore is used by various utilities, like crash and makedumpfile to
982 * obtain access to symbols that are otherwise very difficult to locate. These
983 * symbols are specifically used so that utilities can access and extract the
984 * dmesg log from a vmcore file after a crash.
986 void log_buf_vmcoreinfo_setup(void)
988 struct dev_printk_info *dev_info = NULL;
990 VMCOREINFO_SYMBOL(prb);
991 VMCOREINFO_SYMBOL(printk_rb_static);
992 VMCOREINFO_SYMBOL(clear_seq);
995 * Export struct size and field offsets. User space tools can
996 * parse it and detect any changes to structure down the line.
999 VMCOREINFO_STRUCT_SIZE(printk_ringbuffer);
1000 VMCOREINFO_OFFSET(printk_ringbuffer, desc_ring);
1001 VMCOREINFO_OFFSET(printk_ringbuffer, text_data_ring);
1002 VMCOREINFO_OFFSET(printk_ringbuffer, fail);
1004 VMCOREINFO_STRUCT_SIZE(prb_desc_ring);
1005 VMCOREINFO_OFFSET(prb_desc_ring, count_bits);
1006 VMCOREINFO_OFFSET(prb_desc_ring, descs);
1007 VMCOREINFO_OFFSET(prb_desc_ring, infos);
1008 VMCOREINFO_OFFSET(prb_desc_ring, head_id);
1009 VMCOREINFO_OFFSET(prb_desc_ring, tail_id);
1011 VMCOREINFO_STRUCT_SIZE(prb_desc);
1012 VMCOREINFO_OFFSET(prb_desc, state_var);
1013 VMCOREINFO_OFFSET(prb_desc, text_blk_lpos);
1015 VMCOREINFO_STRUCT_SIZE(prb_data_blk_lpos);
1016 VMCOREINFO_OFFSET(prb_data_blk_lpos, begin);
1017 VMCOREINFO_OFFSET(prb_data_blk_lpos, next);
1019 VMCOREINFO_STRUCT_SIZE(printk_info);
1020 VMCOREINFO_OFFSET(printk_info, seq);
1021 VMCOREINFO_OFFSET(printk_info, ts_nsec);
1022 VMCOREINFO_OFFSET(printk_info, text_len);
1023 VMCOREINFO_OFFSET(printk_info, caller_id);
1024 VMCOREINFO_OFFSET(printk_info, dev_info);
1026 VMCOREINFO_STRUCT_SIZE(dev_printk_info);
1027 VMCOREINFO_OFFSET(dev_printk_info, subsystem);
1028 VMCOREINFO_LENGTH(printk_info_subsystem, sizeof(dev_info->subsystem));
1029 VMCOREINFO_OFFSET(dev_printk_info, device);
1030 VMCOREINFO_LENGTH(printk_info_device, sizeof(dev_info->device));
1032 VMCOREINFO_STRUCT_SIZE(prb_data_ring);
1033 VMCOREINFO_OFFSET(prb_data_ring, size_bits);
1034 VMCOREINFO_OFFSET(prb_data_ring, data);
1035 VMCOREINFO_OFFSET(prb_data_ring, head_lpos);
1036 VMCOREINFO_OFFSET(prb_data_ring, tail_lpos);
1038 VMCOREINFO_SIZE(atomic_long_t);
1039 VMCOREINFO_TYPE_OFFSET(atomic_long_t, counter);
1041 VMCOREINFO_STRUCT_SIZE(latched_seq);
1042 VMCOREINFO_OFFSET(latched_seq, val);
1046 /* requested log_buf_len from kernel cmdline */
1047 static unsigned long __initdata new_log_buf_len;
1049 /* we practice scaling the ring buffer by powers of 2 */
1050 static void __init log_buf_len_update(u64 size)
1052 if (size > (u64)LOG_BUF_LEN_MAX) {
1053 size = (u64)LOG_BUF_LEN_MAX;
1054 pr_err("log_buf over 2G is not supported.\n");
1058 size = roundup_pow_of_two(size);
1059 if (size > log_buf_len)
1060 new_log_buf_len = (unsigned long)size;
1063 /* save requested log_buf_len since it's too early to process it */
1064 static int __init log_buf_len_setup(char *str)
1071 size = memparse(str, &str);
1073 log_buf_len_update(size);
1077 early_param("log_buf_len", log_buf_len_setup);
1080 #define __LOG_CPU_MAX_BUF_LEN (1 << CONFIG_LOG_CPU_MAX_BUF_SHIFT)
1082 static void __init log_buf_add_cpu(void)
1084 unsigned int cpu_extra;
1087 * archs should set up cpu_possible_bits properly with
1088 * set_cpu_possible() after setup_arch() but just in
1089 * case lets ensure this is valid.
1091 if (num_possible_cpus() == 1)
1094 cpu_extra = (num_possible_cpus() - 1) * __LOG_CPU_MAX_BUF_LEN;
1096 /* by default this will only continue through for large > 64 CPUs */
1097 if (cpu_extra <= __LOG_BUF_LEN / 2)
1100 pr_info("log_buf_len individual max cpu contribution: %d bytes\n",
1101 __LOG_CPU_MAX_BUF_LEN);
1102 pr_info("log_buf_len total cpu_extra contributions: %d bytes\n",
1104 pr_info("log_buf_len min size: %d bytes\n", __LOG_BUF_LEN);
1106 log_buf_len_update(cpu_extra + __LOG_BUF_LEN);
1108 #else /* !CONFIG_SMP */
1109 static inline void log_buf_add_cpu(void) {}
1110 #endif /* CONFIG_SMP */
1112 static void __init set_percpu_data_ready(void)
1114 __printk_percpu_data_ready = true;
1117 static unsigned int __init add_to_rb(struct printk_ringbuffer *rb,
1118 struct printk_record *r)
1120 struct prb_reserved_entry e;
1121 struct printk_record dest_r;
1123 prb_rec_init_wr(&dest_r, r->info->text_len);
1125 if (!prb_reserve(&e, rb, &dest_r))
1128 memcpy(&dest_r.text_buf[0], &r->text_buf[0], r->info->text_len);
1129 dest_r.info->text_len = r->info->text_len;
1130 dest_r.info->facility = r->info->facility;
1131 dest_r.info->level = r->info->level;
1132 dest_r.info->flags = r->info->flags;
1133 dest_r.info->ts_nsec = r->info->ts_nsec;
1134 dest_r.info->caller_id = r->info->caller_id;
1135 memcpy(&dest_r.info->dev_info, &r->info->dev_info, sizeof(dest_r.info->dev_info));
1137 prb_final_commit(&e);
1139 return prb_record_text_space(&e);
1142 static char setup_text_buf[PRINTKRB_RECORD_MAX] __initdata;
1144 void __init setup_log_buf(int early)
1146 struct printk_info *new_infos;
1147 unsigned int new_descs_count;
1148 struct prb_desc *new_descs;
1149 struct printk_info info;
1150 struct printk_record r;
1151 unsigned int text_size;
1152 size_t new_descs_size;
1153 size_t new_infos_size;
1154 unsigned long flags;
1160 * Some archs call setup_log_buf() multiple times - first is very
1161 * early, e.g. from setup_arch(), and second - when percpu_areas
1165 set_percpu_data_ready();
1167 if (log_buf != __log_buf)
1170 if (!early && !new_log_buf_len)
1173 if (!new_log_buf_len)
1176 new_descs_count = new_log_buf_len >> PRB_AVGBITS;
1177 if (new_descs_count == 0) {
1178 pr_err("new_log_buf_len: %lu too small\n", new_log_buf_len);
1182 new_log_buf = memblock_alloc(new_log_buf_len, LOG_ALIGN);
1183 if (unlikely(!new_log_buf)) {
1184 pr_err("log_buf_len: %lu text bytes not available\n",
1189 new_descs_size = new_descs_count * sizeof(struct prb_desc);
1190 new_descs = memblock_alloc(new_descs_size, LOG_ALIGN);
1191 if (unlikely(!new_descs)) {
1192 pr_err("log_buf_len: %zu desc bytes not available\n",
1194 goto err_free_log_buf;
1197 new_infos_size = new_descs_count * sizeof(struct printk_info);
1198 new_infos = memblock_alloc(new_infos_size, LOG_ALIGN);
1199 if (unlikely(!new_infos)) {
1200 pr_err("log_buf_len: %zu info bytes not available\n",
1202 goto err_free_descs;
1205 prb_rec_init_rd(&r, &info, &setup_text_buf[0], sizeof(setup_text_buf));
1207 prb_init(&printk_rb_dynamic,
1208 new_log_buf, ilog2(new_log_buf_len),
1209 new_descs, ilog2(new_descs_count),
1212 local_irq_save(flags);
1214 log_buf_len = new_log_buf_len;
1215 log_buf = new_log_buf;
1216 new_log_buf_len = 0;
1218 free = __LOG_BUF_LEN;
1219 prb_for_each_record(0, &printk_rb_static, seq, &r) {
1220 text_size = add_to_rb(&printk_rb_dynamic, &r);
1221 if (text_size > free)
1227 prb = &printk_rb_dynamic;
1229 local_irq_restore(flags);
1232 * Copy any remaining messages that might have appeared from
1233 * NMI context after copying but before switching to the
1236 prb_for_each_record(seq, &printk_rb_static, seq, &r) {
1237 text_size = add_to_rb(&printk_rb_dynamic, &r);
1238 if (text_size > free)
1244 if (seq != prb_next_seq(&printk_rb_static)) {
1245 pr_err("dropped %llu messages\n",
1246 prb_next_seq(&printk_rb_static) - seq);
1249 pr_info("log_buf_len: %u bytes\n", log_buf_len);
1250 pr_info("early log buf free: %u(%u%%)\n",
1251 free, (free * 100) / __LOG_BUF_LEN);
1255 memblock_free(new_descs, new_descs_size);
1257 memblock_free(new_log_buf, new_log_buf_len);
1260 static bool __read_mostly ignore_loglevel;
1262 static int __init ignore_loglevel_setup(char *str)
1264 ignore_loglevel = true;
1265 pr_info("debug: ignoring loglevel setting.\n");
1270 early_param("ignore_loglevel", ignore_loglevel_setup);
1271 module_param(ignore_loglevel, bool, S_IRUGO | S_IWUSR);
1272 MODULE_PARM_DESC(ignore_loglevel,
1273 "ignore loglevel setting (prints all kernel messages to the console)");
1275 static bool suppress_message_printing(int level)
1277 return (level >= console_loglevel && !ignore_loglevel);
1280 #ifdef CONFIG_BOOT_PRINTK_DELAY
1282 static int boot_delay; /* msecs delay after each printk during bootup */
1283 static unsigned long long loops_per_msec; /* based on boot_delay */
1285 static int __init boot_delay_setup(char *str)
1289 lpj = preset_lpj ? preset_lpj : 1000000; /* some guess */
1290 loops_per_msec = (unsigned long long)lpj / 1000 * HZ;
1292 get_option(&str, &boot_delay);
1293 if (boot_delay > 10 * 1000)
1296 pr_debug("boot_delay: %u, preset_lpj: %ld, lpj: %lu, "
1297 "HZ: %d, loops_per_msec: %llu\n",
1298 boot_delay, preset_lpj, lpj, HZ, loops_per_msec);
1301 early_param("boot_delay", boot_delay_setup);
1303 static void boot_delay_msec(int level)
1305 unsigned long long k;
1306 unsigned long timeout;
1308 if ((boot_delay == 0 || system_state >= SYSTEM_RUNNING)
1309 || suppress_message_printing(level)) {
1313 k = (unsigned long long)loops_per_msec * boot_delay;
1315 timeout = jiffies + msecs_to_jiffies(boot_delay);
1320 * use (volatile) jiffies to prevent
1321 * compiler reduction; loop termination via jiffies
1322 * is secondary and may or may not happen.
1324 if (time_after(jiffies, timeout))
1326 touch_nmi_watchdog();
1330 static inline void boot_delay_msec(int level)
1335 static bool printk_time = IS_ENABLED(CONFIG_PRINTK_TIME);
1336 module_param_named(time, printk_time, bool, S_IRUGO | S_IWUSR);
1338 static size_t print_syslog(unsigned int level, char *buf)
1340 return sprintf(buf, "<%u>", level);
1343 static size_t print_time(u64 ts, char *buf)
1345 unsigned long rem_nsec = do_div(ts, 1000000000);
1347 return sprintf(buf, "[%5lu.%06lu]",
1348 (unsigned long)ts, rem_nsec / 1000);
1351 #ifdef CONFIG_PRINTK_CALLER
1352 static size_t print_caller(u32 id, char *buf)
1356 snprintf(caller, sizeof(caller), "%c%u",
1357 id & 0x80000000 ? 'C' : 'T', id & ~0x80000000);
1358 return sprintf(buf, "[%6s]", caller);
1361 #define print_caller(id, buf) 0
1364 static size_t info_print_prefix(const struct printk_info *info, bool syslog,
1365 bool time, char *buf)
1370 len = print_syslog((info->facility << 3) | info->level, buf);
1373 len += print_time(info->ts_nsec, buf + len);
1375 len += print_caller(info->caller_id, buf + len);
1377 if (IS_ENABLED(CONFIG_PRINTK_CALLER) || time) {
1386 * Prepare the record for printing. The text is shifted within the given
1387 * buffer to avoid a need for another one. The following operations are
1390 * - Add prefix for each line.
1391 * - Drop truncated lines that no longer fit into the buffer.
1392 * - Add the trailing newline that has been removed in vprintk_store().
1393 * - Add a string terminator.
1395 * Since the produced string is always terminated, the maximum possible
1396 * return value is @r->text_buf_size - 1;
1398 * Return: The length of the updated/prepared text, including the added
1399 * prefixes and the newline. The terminator is not counted. The dropped
1400 * line(s) are not counted.
1402 static size_t record_print_text(struct printk_record *r, bool syslog,
1405 size_t text_len = r->info->text_len;
1406 size_t buf_size = r->text_buf_size;
1407 char *text = r->text_buf;
1408 char prefix[PRINTK_PREFIX_MAX];
1409 bool truncated = false;
1416 * If the message was truncated because the buffer was not large
1417 * enough, treat the available text as if it were the full text.
1419 if (text_len > buf_size)
1420 text_len = buf_size;
1422 prefix_len = info_print_prefix(r->info, syslog, time, prefix);
1425 * @text_len: bytes of unprocessed text
1426 * @line_len: bytes of current line _without_ newline
1427 * @text: pointer to beginning of current line
1428 * @len: number of bytes prepared in r->text_buf
1431 next = memchr(text, '\n', text_len);
1433 line_len = next - text;
1435 /* Drop truncated line(s). */
1438 line_len = text_len;
1442 * Truncate the text if there is not enough space to add the
1443 * prefix and a trailing newline and a terminator.
1445 if (len + prefix_len + text_len + 1 + 1 > buf_size) {
1446 /* Drop even the current line if no space. */
1447 if (len + prefix_len + line_len + 1 + 1 > buf_size)
1450 text_len = buf_size - len - prefix_len - 1 - 1;
1454 memmove(text + prefix_len, text, text_len);
1455 memcpy(text, prefix, prefix_len);
1458 * Increment the prepared length to include the text and
1459 * prefix that were just moved+copied. Also increment for the
1460 * newline at the end of this line. If this is the last line,
1461 * there is no newline, but it will be added immediately below.
1463 len += prefix_len + line_len + 1;
1464 if (text_len == line_len) {
1466 * This is the last line. Add the trailing newline
1467 * removed in vprintk_store().
1469 text[prefix_len + line_len] = '\n';
1474 * Advance beyond the added prefix and the related line with
1477 text += prefix_len + line_len + 1;
1480 * The remaining text has only decreased by the line with its
1483 * Note that @text_len can become zero. It happens when @text
1484 * ended with a newline (either due to truncation or the
1485 * original string ending with "\n\n"). The loop is correctly
1486 * repeated and (if not truncated) an empty line with a prefix
1489 text_len -= line_len + 1;
1493 * If a buffer was provided, it will be terminated. Space for the
1494 * string terminator is guaranteed to be available. The terminator is
1495 * not counted in the return value.
1498 r->text_buf[len] = 0;
1503 static size_t get_record_print_text_size(struct printk_info *info,
1504 unsigned int line_count,
1505 bool syslog, bool time)
1507 char prefix[PRINTK_PREFIX_MAX];
1510 prefix_len = info_print_prefix(info, syslog, time, prefix);
1513 * Each line will be preceded with a prefix. The intermediate
1514 * newlines are already within the text, but a final trailing
1515 * newline will be added.
1517 return ((prefix_len * line_count) + info->text_len + 1);
1521 * Beginning with @start_seq, find the first record where it and all following
1522 * records up to (but not including) @max_seq fit into @size.
1524 * @max_seq is simply an upper bound and does not need to exist. If the caller
1525 * does not require an upper bound, -1 can be used for @max_seq.
1527 static u64 find_first_fitting_seq(u64 start_seq, u64 max_seq, size_t size,
1528 bool syslog, bool time)
1530 struct printk_info info;
1531 unsigned int line_count;
1535 /* Determine the size of the records up to @max_seq. */
1536 prb_for_each_info(start_seq, prb, seq, &info, &line_count) {
1537 if (info.seq >= max_seq)
1539 len += get_record_print_text_size(&info, line_count, syslog, time);
1543 * Adjust the upper bound for the next loop to avoid subtracting
1544 * lengths that were never added.
1550 * Move first record forward until length fits into the buffer. Ignore
1551 * newest messages that were not counted in the above cycle. Messages
1552 * might appear and get lost in the meantime. This is a best effort
1553 * that prevents an infinite loop that could occur with a retry.
1555 prb_for_each_info(start_seq, prb, seq, &info, &line_count) {
1556 if (len <= size || info.seq >= max_seq)
1558 len -= get_record_print_text_size(&info, line_count, syslog, time);
1564 /* The caller is responsible for making sure @size is greater than 0. */
1565 static int syslog_print(char __user *buf, int size)
1567 struct printk_info info;
1568 struct printk_record r;
1573 text = kmalloc(PRINTK_MESSAGE_MAX, GFP_KERNEL);
1577 prb_rec_init_rd(&r, &info, text, PRINTK_MESSAGE_MAX);
1579 mutex_lock(&syslog_lock);
1582 * Wait for the @syslog_seq record to be available. @syslog_seq may
1583 * change while waiting.
1588 mutex_unlock(&syslog_lock);
1590 * Guarantee this task is visible on the waitqueue before
1591 * checking the wake condition.
1593 * The full memory barrier within set_current_state() of
1594 * prepare_to_wait_event() pairs with the full memory barrier
1595 * within wq_has_sleeper().
1597 * This pairs with __wake_up_klogd:A.
1599 len = wait_event_interruptible(log_wait,
1600 prb_read_valid(prb, seq, NULL)); /* LMM(syslog_print:A) */
1601 mutex_lock(&syslog_lock);
1605 } while (syslog_seq != seq);
1608 * Copy records that fit into the buffer. The above cycle makes sure
1609 * that the first record is always available.
1616 if (!prb_read_valid(prb, syslog_seq, &r))
1619 if (r.info->seq != syslog_seq) {
1620 /* message is gone, move to next valid one */
1621 syslog_seq = r.info->seq;
1626 * To keep reading/counting partial line consistent,
1627 * use printk_time value as of the beginning of a line.
1629 if (!syslog_partial)
1630 syslog_time = printk_time;
1632 skip = syslog_partial;
1633 n = record_print_text(&r, true, syslog_time);
1634 if (n - syslog_partial <= size) {
1635 /* message fits into buffer, move forward */
1636 syslog_seq = r.info->seq + 1;
1637 n -= syslog_partial;
1640 /* partial read(), remember position */
1642 syslog_partial += n;
1649 mutex_unlock(&syslog_lock);
1650 err = copy_to_user(buf, text + skip, n);
1651 mutex_lock(&syslog_lock);
1664 mutex_unlock(&syslog_lock);
1669 static int syslog_print_all(char __user *buf, int size, bool clear)
1671 struct printk_info info;
1672 struct printk_record r;
1678 text = kmalloc(PRINTK_MESSAGE_MAX, GFP_KERNEL);
1684 * Find first record that fits, including all following records,
1685 * into the user-provided buffer for this dump.
1687 seq = find_first_fitting_seq(latched_seq_read_nolock(&clear_seq), -1,
1690 prb_rec_init_rd(&r, &info, text, PRINTK_MESSAGE_MAX);
1692 prb_for_each_record(seq, prb, seq, &r) {
1695 textlen = record_print_text(&r, true, time);
1697 if (len + textlen > size) {
1702 if (copy_to_user(buf + len, text, textlen))
1712 mutex_lock(&syslog_lock);
1713 latched_seq_write(&clear_seq, seq);
1714 mutex_unlock(&syslog_lock);
1721 static void syslog_clear(void)
1723 mutex_lock(&syslog_lock);
1724 latched_seq_write(&clear_seq, prb_next_seq(prb));
1725 mutex_unlock(&syslog_lock);
1728 int do_syslog(int type, char __user *buf, int len, int source)
1730 struct printk_info info;
1732 static int saved_console_loglevel = LOGLEVEL_DEFAULT;
1735 error = check_syslog_permissions(type, source);
1740 case SYSLOG_ACTION_CLOSE: /* Close log */
1742 case SYSLOG_ACTION_OPEN: /* Open log */
1744 case SYSLOG_ACTION_READ: /* Read from log */
1745 if (!buf || len < 0)
1749 if (!access_ok(buf, len))
1751 error = syslog_print(buf, len);
1753 /* Read/clear last kernel messages */
1754 case SYSLOG_ACTION_READ_CLEAR:
1757 /* Read last kernel messages */
1758 case SYSLOG_ACTION_READ_ALL:
1759 if (!buf || len < 0)
1763 if (!access_ok(buf, len))
1765 error = syslog_print_all(buf, len, clear);
1767 /* Clear ring buffer */
1768 case SYSLOG_ACTION_CLEAR:
1771 /* Disable logging to console */
1772 case SYSLOG_ACTION_CONSOLE_OFF:
1773 if (saved_console_loglevel == LOGLEVEL_DEFAULT)
1774 saved_console_loglevel = console_loglevel;
1775 console_loglevel = minimum_console_loglevel;
1777 /* Enable logging to console */
1778 case SYSLOG_ACTION_CONSOLE_ON:
1779 if (saved_console_loglevel != LOGLEVEL_DEFAULT) {
1780 console_loglevel = saved_console_loglevel;
1781 saved_console_loglevel = LOGLEVEL_DEFAULT;
1784 /* Set level of messages printed to console */
1785 case SYSLOG_ACTION_CONSOLE_LEVEL:
1786 if (len < 1 || len > 8)
1788 if (len < minimum_console_loglevel)
1789 len = minimum_console_loglevel;
1790 console_loglevel = len;
1791 /* Implicitly re-enable logging to console */
1792 saved_console_loglevel = LOGLEVEL_DEFAULT;
1794 /* Number of chars in the log buffer */
1795 case SYSLOG_ACTION_SIZE_UNREAD:
1796 mutex_lock(&syslog_lock);
1797 if (!prb_read_valid_info(prb, syslog_seq, &info, NULL)) {
1798 /* No unread messages. */
1799 mutex_unlock(&syslog_lock);
1802 if (info.seq != syslog_seq) {
1803 /* messages are gone, move to first one */
1804 syslog_seq = info.seq;
1807 if (source == SYSLOG_FROM_PROC) {
1809 * Short-cut for poll(/"proc/kmsg") which simply checks
1810 * for pending data, not the size; return the count of
1811 * records, not the length.
1813 error = prb_next_seq(prb) - syslog_seq;
1815 bool time = syslog_partial ? syslog_time : printk_time;
1816 unsigned int line_count;
1819 prb_for_each_info(syslog_seq, prb, seq, &info,
1821 error += get_record_print_text_size(&info, line_count,
1825 error -= syslog_partial;
1827 mutex_unlock(&syslog_lock);
1829 /* Size of the log buffer */
1830 case SYSLOG_ACTION_SIZE_BUFFER:
1831 error = log_buf_len;
1841 SYSCALL_DEFINE3(syslog, int, type, char __user *, buf, int, len)
1843 return do_syslog(type, buf, len, SYSLOG_FROM_READER);
1847 * Special console_lock variants that help to reduce the risk of soft-lockups.
1848 * They allow to pass console_lock to another printk() call using a busy wait.
1851 #ifdef CONFIG_LOCKDEP
1852 static struct lockdep_map console_owner_dep_map = {
1853 .name = "console_owner"
1857 static DEFINE_RAW_SPINLOCK(console_owner_lock);
1858 static struct task_struct *console_owner;
1859 static bool console_waiter;
1862 * console_lock_spinning_enable - mark beginning of code where another
1863 * thread might safely busy wait
1865 * This basically converts console_lock into a spinlock. This marks
1866 * the section where the console_lock owner can not sleep, because
1867 * there may be a waiter spinning (like a spinlock). Also it must be
1868 * ready to hand over the lock at the end of the section.
1870 static void console_lock_spinning_enable(void)
1872 raw_spin_lock(&console_owner_lock);
1873 console_owner = current;
1874 raw_spin_unlock(&console_owner_lock);
1876 /* The waiter may spin on us after setting console_owner */
1877 spin_acquire(&console_owner_dep_map, 0, 0, _THIS_IP_);
1881 * console_lock_spinning_disable_and_check - mark end of code where another
1882 * thread was able to busy wait and check if there is a waiter
1883 * @cookie: cookie returned from console_srcu_read_lock()
1885 * This is called at the end of the section where spinning is allowed.
1886 * It has two functions. First, it is a signal that it is no longer
1887 * safe to start busy waiting for the lock. Second, it checks if
1888 * there is a busy waiter and passes the lock rights to her.
1890 * Important: Callers lose both the console_lock and the SRCU read lock if
1891 * there was a busy waiter. They must not touch items synchronized by
1892 * console_lock or SRCU read lock in this case.
1894 * Return: 1 if the lock rights were passed, 0 otherwise.
1896 static int console_lock_spinning_disable_and_check(int cookie)
1900 raw_spin_lock(&console_owner_lock);
1901 waiter = READ_ONCE(console_waiter);
1902 console_owner = NULL;
1903 raw_spin_unlock(&console_owner_lock);
1906 spin_release(&console_owner_dep_map, _THIS_IP_);
1910 /* The waiter is now free to continue */
1911 WRITE_ONCE(console_waiter, false);
1913 spin_release(&console_owner_dep_map, _THIS_IP_);
1916 * Preserve lockdep lock ordering. Release the SRCU read lock before
1917 * releasing the console_lock.
1919 console_srcu_read_unlock(cookie);
1922 * Hand off console_lock to waiter. The waiter will perform
1923 * the up(). After this, the waiter is the console_lock owner.
1925 mutex_release(&console_lock_dep_map, _THIS_IP_);
1930 * console_trylock_spinning - try to get console_lock by busy waiting
1932 * This allows to busy wait for the console_lock when the current
1933 * owner is running in specially marked sections. It means that
1934 * the current owner is running and cannot reschedule until it
1935 * is ready to lose the lock.
1937 * Return: 1 if we got the lock, 0 othrewise
1939 static int console_trylock_spinning(void)
1941 struct task_struct *owner = NULL;
1944 unsigned long flags;
1946 if (console_trylock())
1950 * It's unsafe to spin once a panic has begun. If we are the
1951 * panic CPU, we may have already halted the owner of the
1952 * console_sem. If we are not the panic CPU, then we should
1953 * avoid taking console_sem, so the panic CPU has a better
1954 * chance of cleanly acquiring it later.
1956 if (panic_in_progress())
1959 printk_safe_enter_irqsave(flags);
1961 raw_spin_lock(&console_owner_lock);
1962 owner = READ_ONCE(console_owner);
1963 waiter = READ_ONCE(console_waiter);
1964 if (!waiter && owner && owner != current) {
1965 WRITE_ONCE(console_waiter, true);
1968 raw_spin_unlock(&console_owner_lock);
1971 * If there is an active printk() writing to the
1972 * consoles, instead of having it write our data too,
1973 * see if we can offload that load from the active
1974 * printer, and do some printing ourselves.
1975 * Go into a spin only if there isn't already a waiter
1976 * spinning, and there is an active printer, and
1977 * that active printer isn't us (recursive printk?).
1980 printk_safe_exit_irqrestore(flags);
1984 /* We spin waiting for the owner to release us */
1985 spin_acquire(&console_owner_dep_map, 0, 0, _THIS_IP_);
1986 /* Owner will clear console_waiter on hand off */
1987 while (READ_ONCE(console_waiter))
1989 spin_release(&console_owner_dep_map, _THIS_IP_);
1991 printk_safe_exit_irqrestore(flags);
1993 * The owner passed the console lock to us.
1994 * Since we did not spin on console lock, annotate
1995 * this as a trylock. Otherwise lockdep will
1998 mutex_acquire(&console_lock_dep_map, 0, 1, _THIS_IP_);
2004 * Recursion is tracked separately on each CPU. If NMIs are supported, an
2005 * additional NMI context per CPU is also separately tracked. Until per-CPU
2006 * is available, a separate "early tracking" is performed.
2008 static DEFINE_PER_CPU(u8, printk_count);
2009 static u8 printk_count_early;
2010 #ifdef CONFIG_HAVE_NMI
2011 static DEFINE_PER_CPU(u8, printk_count_nmi);
2012 static u8 printk_count_nmi_early;
2016 * Recursion is limited to keep the output sane. printk() should not require
2017 * more than 1 level of recursion (allowing, for example, printk() to trigger
2018 * a WARN), but a higher value is used in case some printk-internal errors
2019 * exist, such as the ringbuffer validation checks failing.
2021 #define PRINTK_MAX_RECURSION 3
2024 * Return a pointer to the dedicated counter for the CPU+context of the
2027 static u8 *__printk_recursion_counter(void)
2029 #ifdef CONFIG_HAVE_NMI
2031 if (printk_percpu_data_ready())
2032 return this_cpu_ptr(&printk_count_nmi);
2033 return &printk_count_nmi_early;
2036 if (printk_percpu_data_ready())
2037 return this_cpu_ptr(&printk_count);
2038 return &printk_count_early;
2042 * Enter recursion tracking. Interrupts are disabled to simplify tracking.
2043 * The caller must check the boolean return value to see if the recursion is
2044 * allowed. On failure, interrupts are not disabled.
2046 * @recursion_ptr must be a variable of type (u8 *) and is the same variable
2047 * that is passed to printk_exit_irqrestore().
2049 #define printk_enter_irqsave(recursion_ptr, flags) \
2051 bool success = true; \
2053 typecheck(u8 *, recursion_ptr); \
2054 local_irq_save(flags); \
2055 (recursion_ptr) = __printk_recursion_counter(); \
2056 if (*(recursion_ptr) > PRINTK_MAX_RECURSION) { \
2057 local_irq_restore(flags); \
2060 (*(recursion_ptr))++; \
2065 /* Exit recursion tracking, restoring interrupts. */
2066 #define printk_exit_irqrestore(recursion_ptr, flags) \
2068 typecheck(u8 *, recursion_ptr); \
2069 (*(recursion_ptr))--; \
2070 local_irq_restore(flags); \
2073 int printk_delay_msec __read_mostly;
2075 static inline void printk_delay(int level)
2077 boot_delay_msec(level);
2079 if (unlikely(printk_delay_msec)) {
2080 int m = printk_delay_msec;
2084 touch_nmi_watchdog();
2089 static inline u32 printk_caller_id(void)
2091 return in_task() ? task_pid_nr(current) :
2092 0x80000000 + smp_processor_id();
2096 * printk_parse_prefix - Parse level and control flags.
2098 * @text: The terminated text message.
2099 * @level: A pointer to the current level value, will be updated.
2100 * @flags: A pointer to the current printk_info flags, will be updated.
2102 * @level may be NULL if the caller is not interested in the parsed value.
2103 * Otherwise the variable pointed to by @level must be set to
2104 * LOGLEVEL_DEFAULT in order to be updated with the parsed value.
2106 * @flags may be NULL if the caller is not interested in the parsed value.
2107 * Otherwise the variable pointed to by @flags will be OR'd with the parsed
2110 * Return: The length of the parsed level and control flags.
2112 u16 printk_parse_prefix(const char *text, int *level,
2113 enum printk_info_flags *flags)
2119 kern_level = printk_get_level(text);
2123 switch (kern_level) {
2125 if (level && *level == LOGLEVEL_DEFAULT)
2126 *level = kern_level - '0';
2128 case 'c': /* KERN_CONT */
2141 static u16 printk_sprint(char *text, u16 size, int facility,
2142 enum printk_info_flags *flags, const char *fmt,
2147 text_len = vscnprintf(text, size, fmt, args);
2149 /* Mark and strip a trailing newline. */
2150 if (text_len && text[text_len - 1] == '\n') {
2152 *flags |= LOG_NEWLINE;
2155 /* Strip log level and control flags. */
2156 if (facility == 0) {
2159 prefix_len = printk_parse_prefix(text, NULL, NULL);
2161 text_len -= prefix_len;
2162 memmove(text, text + prefix_len, text_len);
2166 trace_console(text, text_len);
2172 int vprintk_store(int facility, int level,
2173 const struct dev_printk_info *dev_info,
2174 const char *fmt, va_list args)
2176 struct prb_reserved_entry e;
2177 enum printk_info_flags flags = 0;
2178 struct printk_record r;
2179 unsigned long irqflags;
2180 u16 trunc_msg_len = 0;
2190 if (!printk_enter_irqsave(recursion_ptr, irqflags))
2194 * Since the duration of printk() can vary depending on the message
2195 * and state of the ringbuffer, grab the timestamp now so that it is
2196 * close to the call of printk(). This provides a more deterministic
2197 * timestamp with respect to the caller.
2199 ts_nsec = local_clock();
2201 caller_id = printk_caller_id();
2204 * The sprintf needs to come first since the syslog prefix might be
2205 * passed in as a parameter. An extra byte must be reserved so that
2206 * later the vscnprintf() into the reserved buffer has room for the
2207 * terminating '\0', which is not counted by vsnprintf().
2209 va_copy(args2, args);
2210 reserve_size = vsnprintf(&prefix_buf[0], sizeof(prefix_buf), fmt, args2) + 1;
2213 if (reserve_size > PRINTKRB_RECORD_MAX)
2214 reserve_size = PRINTKRB_RECORD_MAX;
2216 /* Extract log level or control flags. */
2218 printk_parse_prefix(&prefix_buf[0], &level, &flags);
2220 if (level == LOGLEVEL_DEFAULT)
2221 level = default_message_loglevel;
2224 flags |= LOG_NEWLINE;
2226 if (flags & LOG_CONT) {
2227 prb_rec_init_wr(&r, reserve_size);
2228 if (prb_reserve_in_last(&e, prb, &r, caller_id, PRINTKRB_RECORD_MAX)) {
2229 text_len = printk_sprint(&r.text_buf[r.info->text_len], reserve_size,
2230 facility, &flags, fmt, args);
2231 r.info->text_len += text_len;
2233 if (flags & LOG_NEWLINE) {
2234 r.info->flags |= LOG_NEWLINE;
2235 prb_final_commit(&e);
2246 * Explicitly initialize the record before every prb_reserve() call.
2247 * prb_reserve_in_last() and prb_reserve() purposely invalidate the
2248 * structure when they fail.
2250 prb_rec_init_wr(&r, reserve_size);
2251 if (!prb_reserve(&e, prb, &r)) {
2252 /* truncate the message if it is too long for empty buffer */
2253 truncate_msg(&reserve_size, &trunc_msg_len);
2255 prb_rec_init_wr(&r, reserve_size + trunc_msg_len);
2256 if (!prb_reserve(&e, prb, &r))
2261 text_len = printk_sprint(&r.text_buf[0], reserve_size, facility, &flags, fmt, args);
2263 memcpy(&r.text_buf[text_len], trunc_msg, trunc_msg_len);
2264 r.info->text_len = text_len + trunc_msg_len;
2265 r.info->facility = facility;
2266 r.info->level = level & 7;
2267 r.info->flags = flags & 0x1f;
2268 r.info->ts_nsec = ts_nsec;
2269 r.info->caller_id = caller_id;
2271 memcpy(&r.info->dev_info, dev_info, sizeof(r.info->dev_info));
2273 /* A message without a trailing newline can be continued. */
2274 if (!(flags & LOG_NEWLINE))
2277 prb_final_commit(&e);
2279 ret = text_len + trunc_msg_len;
2281 printk_exit_irqrestore(recursion_ptr, irqflags);
2285 asmlinkage int vprintk_emit(int facility, int level,
2286 const struct dev_printk_info *dev_info,
2287 const char *fmt, va_list args)
2290 bool in_sched = false;
2292 /* Suppress unimportant messages after panic happens */
2293 if (unlikely(suppress_printk))
2296 if (unlikely(suppress_panic_printk) && other_cpu_in_panic())
2299 if (level == LOGLEVEL_SCHED) {
2300 level = LOGLEVEL_DEFAULT;
2304 printk_delay(level);
2306 printed_len = vprintk_store(facility, level, dev_info, fmt, args);
2308 /* If called from the scheduler, we can not call up(). */
2311 * The caller may be holding system-critical or
2312 * timing-sensitive locks. Disable preemption during
2313 * printing of all remaining records to all consoles so that
2314 * this context can return as soon as possible. Hopefully
2315 * another printk() caller will take over the printing.
2319 * Try to acquire and then immediately release the console
2320 * semaphore. The release will print out buffers. With the
2321 * spinning variant, this context tries to take over the
2322 * printing from another printing context.
2324 if (console_trylock_spinning())
2330 defer_console_output();
2336 EXPORT_SYMBOL(vprintk_emit);
2338 int vprintk_default(const char *fmt, va_list args)
2340 return vprintk_emit(0, LOGLEVEL_DEFAULT, NULL, fmt, args);
2342 EXPORT_SYMBOL_GPL(vprintk_default);
2344 asmlinkage __visible int _printk(const char *fmt, ...)
2349 va_start(args, fmt);
2350 r = vprintk(fmt, args);
2355 EXPORT_SYMBOL(_printk);
2357 static bool pr_flush(int timeout_ms, bool reset_on_progress);
2358 static bool __pr_flush(struct console *con, int timeout_ms, bool reset_on_progress);
2360 #else /* CONFIG_PRINTK */
2362 #define printk_time false
2364 #define prb_read_valid(rb, seq, r) false
2365 #define prb_first_valid_seq(rb) 0
2366 #define prb_next_seq(rb) 0
2368 static u64 syslog_seq;
2370 static bool pr_flush(int timeout_ms, bool reset_on_progress) { return true; }
2371 static bool __pr_flush(struct console *con, int timeout_ms, bool reset_on_progress) { return true; }
2373 #endif /* CONFIG_PRINTK */
2375 #ifdef CONFIG_EARLY_PRINTK
2376 struct console *early_console;
2378 asmlinkage __visible void early_printk(const char *fmt, ...)
2388 n = vscnprintf(buf, sizeof(buf), fmt, ap);
2391 early_console->write(early_console, buf, n);
2395 static void set_user_specified(struct console_cmdline *c, bool user_specified)
2397 if (!user_specified)
2401 * @c console was defined by the user on the command line.
2402 * Do not clear when added twice also by SPCR or the device tree.
2404 c->user_specified = true;
2405 /* At least one console defined by the user on the command line. */
2406 console_set_on_cmdline = 1;
2409 static int __add_preferred_console(char *name, int idx, char *options,
2410 char *brl_options, bool user_specified)
2412 struct console_cmdline *c;
2416 * See if this tty is not yet registered, and
2417 * if we have a slot free.
2419 for (i = 0, c = console_cmdline;
2420 i < MAX_CMDLINECONSOLES && c->name[0];
2422 if (strcmp(c->name, name) == 0 && c->index == idx) {
2424 preferred_console = i;
2425 set_user_specified(c, user_specified);
2429 if (i == MAX_CMDLINECONSOLES)
2432 preferred_console = i;
2433 strscpy(c->name, name, sizeof(c->name));
2434 c->options = options;
2435 set_user_specified(c, user_specified);
2436 braille_set_options(c, brl_options);
2442 static int __init console_msg_format_setup(char *str)
2444 if (!strcmp(str, "syslog"))
2445 console_msg_format = MSG_FORMAT_SYSLOG;
2446 if (!strcmp(str, "default"))
2447 console_msg_format = MSG_FORMAT_DEFAULT;
2450 __setup("console_msg_format=", console_msg_format_setup);
2453 * Set up a console. Called via do_early_param() in init/main.c
2454 * for each "console=" parameter in the boot command line.
2456 static int __init console_setup(char *str)
2458 char buf[sizeof(console_cmdline[0].name) + 4]; /* 4 for "ttyS" */
2459 char *s, *options, *brl_options = NULL;
2463 * console="" or console=null have been suggested as a way to
2464 * disable console output. Use ttynull that has been created
2465 * for exactly this purpose.
2467 if (str[0] == 0 || strcmp(str, "null") == 0) {
2468 __add_preferred_console("ttynull", 0, NULL, NULL, true);
2472 if (_braille_console_setup(&str, &brl_options))
2476 * Decode str into name, index, options.
2478 if (str[0] >= '0' && str[0] <= '9') {
2479 strcpy(buf, "ttyS");
2480 strncpy(buf + 4, str, sizeof(buf) - 5);
2482 strncpy(buf, str, sizeof(buf) - 1);
2484 buf[sizeof(buf) - 1] = 0;
2485 options = strchr(str, ',');
2489 if (!strcmp(str, "ttya"))
2490 strcpy(buf, "ttyS0");
2491 if (!strcmp(str, "ttyb"))
2492 strcpy(buf, "ttyS1");
2494 for (s = buf; *s; s++)
2495 if (isdigit(*s) || *s == ',')
2497 idx = simple_strtoul(s, NULL, 10);
2500 __add_preferred_console(buf, idx, options, brl_options, true);
2503 __setup("console=", console_setup);
2506 * add_preferred_console - add a device to the list of preferred consoles.
2507 * @name: device name
2508 * @idx: device index
2509 * @options: options for this console
2511 * The last preferred console added will be used for kernel messages
2512 * and stdin/out/err for init. Normally this is used by console_setup
2513 * above to handle user-supplied console arguments; however it can also
2514 * be used by arch-specific code either to override the user or more
2515 * commonly to provide a default console (ie from PROM variables) when
2516 * the user has not supplied one.
2518 int add_preferred_console(char *name, int idx, char *options)
2520 return __add_preferred_console(name, idx, options, NULL, false);
2523 bool console_suspend_enabled = true;
2524 EXPORT_SYMBOL(console_suspend_enabled);
2526 static int __init console_suspend_disable(char *str)
2528 console_suspend_enabled = false;
2531 __setup("no_console_suspend", console_suspend_disable);
2532 module_param_named(console_suspend, console_suspend_enabled,
2533 bool, S_IRUGO | S_IWUSR);
2534 MODULE_PARM_DESC(console_suspend, "suspend console during suspend"
2535 " and hibernate operations");
2537 static bool printk_console_no_auto_verbose;
2539 void console_verbose(void)
2541 if (console_loglevel && !printk_console_no_auto_verbose)
2542 console_loglevel = CONSOLE_LOGLEVEL_MOTORMOUTH;
2544 EXPORT_SYMBOL_GPL(console_verbose);
2546 module_param_named(console_no_auto_verbose, printk_console_no_auto_verbose, bool, 0644);
2547 MODULE_PARM_DESC(console_no_auto_verbose, "Disable console loglevel raise to highest on oops/panic/etc");
2550 * suspend_console - suspend the console subsystem
2552 * This disables printk() while we go into suspend states
2554 void suspend_console(void)
2556 struct console *con;
2558 if (!console_suspend_enabled)
2560 pr_info("Suspending console(s) (use no_console_suspend to debug)\n");
2561 pr_flush(1000, true);
2563 console_list_lock();
2564 for_each_console(con)
2565 console_srcu_write_flags(con, con->flags | CON_SUSPENDED);
2566 console_list_unlock();
2569 * Ensure that all SRCU list walks have completed. All printing
2570 * contexts must be able to see that they are suspended so that it
2571 * is guaranteed that all printing has stopped when this function
2574 synchronize_srcu(&console_srcu);
2577 void resume_console(void)
2579 struct console *con;
2581 if (!console_suspend_enabled)
2584 console_list_lock();
2585 for_each_console(con)
2586 console_srcu_write_flags(con, con->flags & ~CON_SUSPENDED);
2587 console_list_unlock();
2590 * Ensure that all SRCU list walks have completed. All printing
2591 * contexts must be able to see they are no longer suspended so
2592 * that they are guaranteed to wake up and resume printing.
2594 synchronize_srcu(&console_srcu);
2596 pr_flush(1000, true);
2600 * console_cpu_notify - print deferred console messages after CPU hotplug
2603 * If printk() is called from a CPU that is not online yet, the messages
2604 * will be printed on the console only if there are CON_ANYTIME consoles.
2605 * This function is called when a new CPU comes online (or fails to come
2606 * up) or goes offline.
2608 static int console_cpu_notify(unsigned int cpu)
2610 if (!cpuhp_tasks_frozen) {
2611 /* If trylock fails, someone else is doing the printing */
2612 if (console_trylock())
2619 * console_lock - block the console subsystem from printing
2621 * Acquires a lock which guarantees that no consoles will
2622 * be in or enter their write() callback.
2624 * Can sleep, returns nothing.
2626 void console_lock(void)
2630 /* On panic, the console_lock must be left to the panic cpu. */
2631 while (other_cpu_in_panic())
2636 console_may_schedule = 1;
2638 EXPORT_SYMBOL(console_lock);
2641 * console_trylock - try to block the console subsystem from printing
2643 * Try to acquire a lock which guarantees that no consoles will
2644 * be in or enter their write() callback.
2646 * returns 1 on success, and 0 on failure to acquire the lock.
2648 int console_trylock(void)
2650 /* On panic, the console_lock must be left to the panic cpu. */
2651 if (other_cpu_in_panic())
2653 if (down_trylock_console_sem())
2656 console_may_schedule = 0;
2659 EXPORT_SYMBOL(console_trylock);
2661 int is_console_locked(void)
2663 return console_locked;
2665 EXPORT_SYMBOL(is_console_locked);
2668 * Check if the given console is currently capable and allowed to print
2671 * Requires the console_srcu_read_lock.
2673 static inline bool console_is_usable(struct console *con)
2675 short flags = console_srcu_read_flags(con);
2677 if (!(flags & CON_ENABLED))
2680 if ((flags & CON_SUSPENDED))
2687 * Console drivers may assume that per-cpu resources have been
2688 * allocated. So unless they're explicitly marked as being able to
2689 * cope (CON_ANYTIME) don't call them until this CPU is officially up.
2691 if (!cpu_online(raw_smp_processor_id()) && !(flags & CON_ANYTIME))
2697 static void __console_unlock(void)
2703 #ifdef CONFIG_PRINTK
2706 * Prepend the message in @pmsg->pbufs->outbuf with a "dropped message". This
2707 * is achieved by shifting the existing message over and inserting the dropped
2710 * @pmsg is the printk message to prepend.
2712 * @dropped is the dropped count to report in the dropped message.
2714 * If the message text in @pmsg->pbufs->outbuf does not have enough space for
2715 * the dropped message, the message text will be sufficiently truncated.
2717 * If @pmsg->pbufs->outbuf is modified, @pmsg->outbuf_len is updated.
2719 void console_prepend_dropped(struct printk_message *pmsg, unsigned long dropped)
2721 struct printk_buffers *pbufs = pmsg->pbufs;
2722 const size_t scratchbuf_sz = sizeof(pbufs->scratchbuf);
2723 const size_t outbuf_sz = sizeof(pbufs->outbuf);
2724 char *scratchbuf = &pbufs->scratchbuf[0];
2725 char *outbuf = &pbufs->outbuf[0];
2728 len = scnprintf(scratchbuf, scratchbuf_sz,
2729 "** %lu printk messages dropped **\n", dropped);
2732 * Make sure outbuf is sufficiently large before prepending.
2733 * Keep at least the prefix when the message must be truncated.
2734 * It is a rather theoretical problem when someone tries to
2735 * use a minimalist buffer.
2737 if (WARN_ON_ONCE(len + PRINTK_PREFIX_MAX >= outbuf_sz))
2740 if (pmsg->outbuf_len + len >= outbuf_sz) {
2741 /* Truncate the message, but keep it terminated. */
2742 pmsg->outbuf_len = outbuf_sz - (len + 1);
2743 outbuf[pmsg->outbuf_len] = 0;
2746 memmove(outbuf + len, outbuf, pmsg->outbuf_len + 1);
2747 memcpy(outbuf, scratchbuf, len);
2748 pmsg->outbuf_len += len;
2752 * Read and format the specified record (or a later record if the specified
2753 * record is not available).
2755 * @pmsg will contain the formatted result. @pmsg->pbufs must point to a
2756 * struct printk_buffers.
2758 * @seq is the record to read and format. If it is not available, the next
2759 * valid record is read.
2761 * @is_extended specifies if the message should be formatted for extended
2764 * @may_supress specifies if records may be skipped based on loglevel.
2766 * Returns false if no record is available. Otherwise true and all fields
2767 * of @pmsg are valid. (See the documentation of struct printk_message
2768 * for information about the @pmsg fields.)
2770 bool printk_get_next_message(struct printk_message *pmsg, u64 seq,
2771 bool is_extended, bool may_suppress)
2773 static int panic_console_dropped;
2775 struct printk_buffers *pbufs = pmsg->pbufs;
2776 const size_t scratchbuf_sz = sizeof(pbufs->scratchbuf);
2777 const size_t outbuf_sz = sizeof(pbufs->outbuf);
2778 char *scratchbuf = &pbufs->scratchbuf[0];
2779 char *outbuf = &pbufs->outbuf[0];
2780 struct printk_info info;
2781 struct printk_record r;
2785 * Formatting extended messages requires a separate buffer, so use the
2786 * scratch buffer to read in the ringbuffer text.
2788 * Formatting normal messages is done in-place, so read the ringbuffer
2789 * text directly into the output buffer.
2792 prb_rec_init_rd(&r, &info, scratchbuf, scratchbuf_sz);
2794 prb_rec_init_rd(&r, &info, outbuf, outbuf_sz);
2796 if (!prb_read_valid(prb, seq, &r))
2799 pmsg->seq = r.info->seq;
2800 pmsg->dropped = r.info->seq - seq;
2803 * Check for dropped messages in panic here so that printk
2804 * suppression can occur as early as possible if necessary.
2806 if (pmsg->dropped &&
2807 panic_in_progress() &&
2808 panic_console_dropped++ > 10) {
2809 suppress_panic_printk = 1;
2810 pr_warn_once("Too many dropped messages. Suppress messages on non-panic CPUs to prevent livelock.\n");
2813 /* Skip record that has level above the console loglevel. */
2814 if (may_suppress && suppress_message_printing(r.info->level))
2818 len = info_print_ext_header(outbuf, outbuf_sz, r.info);
2819 len += msg_print_ext_body(outbuf + len, outbuf_sz - len,
2820 &r.text_buf[0], r.info->text_len, &r.info->dev_info);
2822 len = record_print_text(&r, console_msg_format & MSG_FORMAT_SYSLOG, printk_time);
2825 pmsg->outbuf_len = len;
2830 * Used as the printk buffers for non-panic, serialized console printing.
2831 * This is for legacy (!CON_NBCON) as well as all boot (CON_BOOT) consoles.
2832 * Its usage requires the console_lock held.
2834 struct printk_buffers printk_shared_pbufs;
2837 * Print one record for the given console. The record printed is whatever
2838 * record is the next available record for the given console.
2840 * @handover will be set to true if a printk waiter has taken over the
2841 * console_lock, in which case the caller is no longer holding both the
2842 * console_lock and the SRCU read lock. Otherwise it is set to false.
2844 * @cookie is the cookie from the SRCU read lock.
2846 * Returns false if the given console has no next record to print, otherwise
2849 * Requires the console_lock and the SRCU read lock.
2851 static bool console_emit_next_record(struct console *con, bool *handover, int cookie)
2853 bool is_extended = console_srcu_read_flags(con) & CON_EXTENDED;
2854 char *outbuf = &printk_shared_pbufs.outbuf[0];
2855 struct printk_message pmsg = {
2856 .pbufs = &printk_shared_pbufs,
2858 unsigned long flags;
2862 if (!printk_get_next_message(&pmsg, con->seq, is_extended, true))
2865 con->dropped += pmsg.dropped;
2867 /* Skip messages of formatted length 0. */
2868 if (pmsg.outbuf_len == 0) {
2869 con->seq = pmsg.seq + 1;
2873 if (con->dropped && !is_extended) {
2874 console_prepend_dropped(&pmsg, con->dropped);
2879 * While actively printing out messages, if another printk()
2880 * were to occur on another CPU, it may wait for this one to
2881 * finish. This task can not be preempted if there is a
2882 * waiter waiting to take over.
2884 * Interrupts are disabled because the hand over to a waiter
2885 * must not be interrupted until the hand over is completed
2886 * (@console_waiter is cleared).
2888 printk_safe_enter_irqsave(flags);
2889 console_lock_spinning_enable();
2891 /* Do not trace print latency. */
2892 stop_critical_timings();
2894 /* Write everything out to the hardware. */
2895 con->write(con, outbuf, pmsg.outbuf_len);
2897 start_critical_timings();
2899 con->seq = pmsg.seq + 1;
2901 *handover = console_lock_spinning_disable_and_check(cookie);
2902 printk_safe_exit_irqrestore(flags);
2909 static bool console_emit_next_record(struct console *con, bool *handover, int cookie)
2915 #endif /* CONFIG_PRINTK */
2918 * Print out all remaining records to all consoles.
2920 * @do_cond_resched is set by the caller. It can be true only in schedulable
2923 * @next_seq is set to the sequence number after the last available record.
2924 * The value is valid only when this function returns true. It means that all
2925 * usable consoles are completely flushed.
2927 * @handover will be set to true if a printk waiter has taken over the
2928 * console_lock, in which case the caller is no longer holding the
2929 * console_lock. Otherwise it is set to false.
2931 * Returns true when there was at least one usable console and all messages
2932 * were flushed to all usable consoles. A returned false informs the caller
2933 * that everything was not flushed (either there were no usable consoles or
2934 * another context has taken over printing or it is a panic situation and this
2935 * is not the panic CPU). Regardless the reason, the caller should assume it
2936 * is not useful to immediately try again.
2938 * Requires the console_lock.
2940 static bool console_flush_all(bool do_cond_resched, u64 *next_seq, bool *handover)
2942 bool any_usable = false;
2943 struct console *con;
2951 any_progress = false;
2953 cookie = console_srcu_read_lock();
2954 for_each_console_srcu(con) {
2957 if (!console_is_usable(con))
2961 progress = console_emit_next_record(con, handover, cookie);
2964 * If a handover has occurred, the SRCU read lock
2965 * is already released.
2970 /* Track the next of the highest seq flushed. */
2971 if (con->seq > *next_seq)
2972 *next_seq = con->seq;
2976 any_progress = true;
2978 /* Allow panic_cpu to take over the consoles safely. */
2979 if (other_cpu_in_panic())
2982 if (do_cond_resched)
2985 console_srcu_read_unlock(cookie);
2986 } while (any_progress);
2991 console_srcu_read_unlock(cookie);
2996 * console_unlock - unblock the console subsystem from printing
2998 * Releases the console_lock which the caller holds to block printing of
2999 * the console subsystem.
3001 * While the console_lock was held, console output may have been buffered
3002 * by printk(). If this is the case, console_unlock(); emits
3003 * the output prior to releasing the lock.
3005 * console_unlock(); may be called from any context.
3007 void console_unlock(void)
3009 bool do_cond_resched;
3015 * Console drivers are called with interrupts disabled, so
3016 * @console_may_schedule should be cleared before; however, we may
3017 * end up dumping a lot of lines, for example, if called from
3018 * console registration path, and should invoke cond_resched()
3019 * between lines if allowable. Not doing so can cause a very long
3020 * scheduling stall on a slow console leading to RCU stall and
3021 * softlockup warnings which exacerbate the issue with more
3022 * messages practically incapacitating the system. Therefore, create
3023 * a local to use for the printing loop.
3025 do_cond_resched = console_may_schedule;
3028 console_may_schedule = 0;
3030 flushed = console_flush_all(do_cond_resched, &next_seq, &handover);
3035 * Abort if there was a failure to flush all messages to all
3036 * usable consoles. Either it is not possible to flush (in
3037 * which case it would be an infinite loop of retrying) or
3038 * another context has taken over printing.
3044 * Some context may have added new records after
3045 * console_flush_all() but before unlocking the console.
3046 * Re-check if there is a new record to flush. If the trylock
3047 * fails, another context is already handling the printing.
3049 } while (prb_read_valid(prb, next_seq, NULL) && console_trylock());
3051 EXPORT_SYMBOL(console_unlock);
3054 * console_conditional_schedule - yield the CPU if required
3056 * If the console code is currently allowed to sleep, and
3057 * if this CPU should yield the CPU to another task, do
3060 * Must be called within console_lock();.
3062 void __sched console_conditional_schedule(void)
3064 if (console_may_schedule)
3067 EXPORT_SYMBOL(console_conditional_schedule);
3069 void console_unblank(void)
3071 bool found_unblank = false;
3076 * First check if there are any consoles implementing the unblank()
3077 * callback. If not, there is no reason to continue and take the
3078 * console lock, which in particular can be dangerous if
3079 * @oops_in_progress is set.
3081 cookie = console_srcu_read_lock();
3082 for_each_console_srcu(c) {
3083 if ((console_srcu_read_flags(c) & CON_ENABLED) && c->unblank) {
3084 found_unblank = true;
3088 console_srcu_read_unlock(cookie);
3093 * Stop console printing because the unblank() callback may
3094 * assume the console is not within its write() callback.
3096 * If @oops_in_progress is set, this may be an atomic context.
3097 * In that case, attempt a trylock as best-effort.
3099 if (oops_in_progress) {
3100 /* Semaphores are not NMI-safe. */
3105 * Attempting to trylock the console lock can deadlock
3106 * if another CPU was stopped while modifying the
3107 * semaphore. "Hope and pray" that this is not the
3108 * current situation.
3110 if (down_trylock_console_sem() != 0)
3116 console_may_schedule = 0;
3118 cookie = console_srcu_read_lock();
3119 for_each_console_srcu(c) {
3120 if ((console_srcu_read_flags(c) & CON_ENABLED) && c->unblank)
3123 console_srcu_read_unlock(cookie);
3127 if (!oops_in_progress)
3128 pr_flush(1000, true);
3132 * console_flush_on_panic - flush console content on panic
3133 * @mode: flush all messages in buffer or just the pending ones
3135 * Immediately output all pending messages no matter what.
3137 void console_flush_on_panic(enum con_flush_mode mode)
3143 * Ignore the console lock and flush out the messages. Attempting a
3144 * trylock would not be useful because:
3146 * - if it is contended, it must be ignored anyway
3147 * - console_lock() and console_trylock() block and fail
3148 * respectively in panic for non-panic CPUs
3149 * - semaphores are not NMI-safe
3153 * If another context is holding the console lock,
3154 * @console_may_schedule might be set. Clear it so that
3155 * this context does not call cond_resched() while flushing.
3157 console_may_schedule = 0;
3159 if (mode == CONSOLE_REPLAY_ALL) {
3165 seq = prb_first_valid_seq(prb);
3167 cookie = console_srcu_read_lock();
3168 for_each_console_srcu(c) {
3169 flags = console_srcu_read_flags(c);
3171 if (flags & CON_NBCON) {
3172 nbcon_seq_force(c, seq);
3175 * This is an unsynchronized assignment. On
3176 * panic legacy consoles are only best effort.
3181 console_srcu_read_unlock(cookie);
3184 console_flush_all(false, &next_seq, &handover);
3188 * Return the console tty driver structure and its associated index
3190 struct tty_driver *console_device(int *index)
3193 struct tty_driver *driver = NULL;
3197 * Take console_lock to serialize device() callback with
3198 * other console operations. For example, fg_console is
3199 * modified under console_lock when switching vt.
3203 cookie = console_srcu_read_lock();
3204 for_each_console_srcu(c) {
3207 driver = c->device(c, index);
3211 console_srcu_read_unlock(cookie);
3218 * Prevent further output on the passed console device so that (for example)
3219 * serial drivers can disable console output before suspending a port, and can
3220 * re-enable output afterwards.
3222 void console_stop(struct console *console)
3224 __pr_flush(console, 1000, true);
3225 console_list_lock();
3226 console_srcu_write_flags(console, console->flags & ~CON_ENABLED);
3227 console_list_unlock();
3230 * Ensure that all SRCU list walks have completed. All contexts must
3231 * be able to see that this console is disabled so that (for example)
3232 * the caller can suspend the port without risk of another context
3235 synchronize_srcu(&console_srcu);
3237 EXPORT_SYMBOL(console_stop);
3239 void console_start(struct console *console)
3241 console_list_lock();
3242 console_srcu_write_flags(console, console->flags | CON_ENABLED);
3243 console_list_unlock();
3244 __pr_flush(console, 1000, true);
3246 EXPORT_SYMBOL(console_start);
3248 static int __read_mostly keep_bootcon;
3250 static int __init keep_bootcon_setup(char *str)
3253 pr_info("debug: skip boot console de-registration.\n");
3258 early_param("keep_bootcon", keep_bootcon_setup);
3261 * This is called by register_console() to try to match
3262 * the newly registered console with any of the ones selected
3263 * by either the command line or add_preferred_console() and
3266 * Care need to be taken with consoles that are statically
3267 * enabled such as netconsole
3269 static int try_enable_preferred_console(struct console *newcon,
3270 bool user_specified)
3272 struct console_cmdline *c;
3275 for (i = 0, c = console_cmdline;
3276 i < MAX_CMDLINECONSOLES && c->name[0];
3278 if (c->user_specified != user_specified)
3280 if (!newcon->match ||
3281 newcon->match(newcon, c->name, c->index, c->options) != 0) {
3282 /* default matching */
3283 BUILD_BUG_ON(sizeof(c->name) != sizeof(newcon->name));
3284 if (strcmp(c->name, newcon->name) != 0)
3286 if (newcon->index >= 0 &&
3287 newcon->index != c->index)
3289 if (newcon->index < 0)
3290 newcon->index = c->index;
3292 if (_braille_register_console(newcon, c))
3295 if (newcon->setup &&
3296 (err = newcon->setup(newcon, c->options)) != 0)
3299 newcon->flags |= CON_ENABLED;
3300 if (i == preferred_console)
3301 newcon->flags |= CON_CONSDEV;
3306 * Some consoles, such as pstore and netconsole, can be enabled even
3307 * without matching. Accept the pre-enabled consoles only when match()
3308 * and setup() had a chance to be called.
3310 if (newcon->flags & CON_ENABLED && c->user_specified == user_specified)
3316 /* Try to enable the console unconditionally */
3317 static void try_enable_default_console(struct console *newcon)
3319 if (newcon->index < 0)
3322 if (newcon->setup && newcon->setup(newcon, NULL) != 0)
3325 newcon->flags |= CON_ENABLED;
3328 newcon->flags |= CON_CONSDEV;
3331 static void console_init_seq(struct console *newcon, bool bootcon_registered)
3333 struct console *con;
3336 if (newcon->flags & (CON_PRINTBUFFER | CON_BOOT)) {
3337 /* Get a consistent copy of @syslog_seq. */
3338 mutex_lock(&syslog_lock);
3339 newcon->seq = syslog_seq;
3340 mutex_unlock(&syslog_lock);
3342 /* Begin with next message added to ringbuffer. */
3343 newcon->seq = prb_next_seq(prb);
3346 * If any enabled boot consoles are due to be unregistered
3347 * shortly, some may not be caught up and may be the same
3348 * device as @newcon. Since it is not known which boot console
3349 * is the same device, flush all consoles and, if necessary,
3350 * start with the message of the enabled boot console that is
3351 * the furthest behind.
3353 if (bootcon_registered && !keep_bootcon) {
3355 * Hold the console_lock to stop console printing and
3356 * guarantee safe access to console->seq.
3361 * Flush all consoles and set the console to start at
3362 * the next unprinted sequence number.
3364 if (!console_flush_all(true, &newcon->seq, &handover)) {
3366 * Flushing failed. Just choose the lowest
3367 * sequence of the enabled boot consoles.
3371 * If there was a handover, this context no
3372 * longer holds the console_lock.
3377 newcon->seq = prb_next_seq(prb);
3378 for_each_console(con) {
3379 if ((con->flags & CON_BOOT) &&
3380 (con->flags & CON_ENABLED) &&
3381 con->seq < newcon->seq) {
3382 newcon->seq = con->seq;
3392 #define console_first() \
3393 hlist_entry(console_list.first, struct console, node)
3395 static int unregister_console_locked(struct console *console);
3398 * The console driver calls this routine during kernel initialization
3399 * to register the console printing procedure with printk() and to
3400 * print any messages that were printed by the kernel before the
3401 * console driver was initialized.
3403 * This can happen pretty early during the boot process (because of
3404 * early_printk) - sometimes before setup_arch() completes - be careful
3405 * of what kernel features are used - they may not be initialised yet.
3407 * There are two types of consoles - bootconsoles (early_printk) and
3408 * "real" consoles (everything which is not a bootconsole) which are
3409 * handled differently.
3410 * - Any number of bootconsoles can be registered at any time.
3411 * - As soon as a "real" console is registered, all bootconsoles
3412 * will be unregistered automatically.
3413 * - Once a "real" console is registered, any attempt to register a
3414 * bootconsoles will be rejected
3416 void register_console(struct console *newcon)
3418 struct console *con;
3419 bool bootcon_registered = false;
3420 bool realcon_registered = false;
3423 console_list_lock();
3425 for_each_console(con) {
3426 if (WARN(con == newcon, "console '%s%d' already registered\n",
3427 con->name, con->index)) {
3431 if (con->flags & CON_BOOT)
3432 bootcon_registered = true;
3434 realcon_registered = true;
3437 /* Do not register boot consoles when there already is a real one. */
3438 if ((newcon->flags & CON_BOOT) && realcon_registered) {
3439 pr_info("Too late to register bootconsole %s%d\n",
3440 newcon->name, newcon->index);
3444 if (newcon->flags & CON_NBCON) {
3446 * Ensure the nbcon console buffers can be allocated
3447 * before modifying any global data.
3449 if (!nbcon_alloc(newcon))
3454 * See if we want to enable this console driver by default.
3456 * Nope when a console is preferred by the command line, device
3459 * The first real console with tty binding (driver) wins. More
3460 * consoles might get enabled before the right one is found.
3462 * Note that a console with tty binding will have CON_CONSDEV
3463 * flag set and will be first in the list.
3465 if (preferred_console < 0) {
3466 if (hlist_empty(&console_list) || !console_first()->device ||
3467 console_first()->flags & CON_BOOT) {
3468 try_enable_default_console(newcon);
3472 /* See if this console matches one we selected on the command line */
3473 err = try_enable_preferred_console(newcon, true);
3475 /* If not, try to match against the platform default(s) */
3477 err = try_enable_preferred_console(newcon, false);
3479 /* printk() messages are not printed to the Braille console. */
3480 if (err || newcon->flags & CON_BRL) {
3481 if (newcon->flags & CON_NBCON)
3487 * If we have a bootconsole, and are switching to a real console,
3488 * don't print everything out again, since when the boot console, and
3489 * the real console are the same physical device, it's annoying to
3490 * see the beginning boot messages twice
3492 if (bootcon_registered &&
3493 ((newcon->flags & (CON_CONSDEV | CON_BOOT)) == CON_CONSDEV)) {
3494 newcon->flags &= ~CON_PRINTBUFFER;
3497 newcon->dropped = 0;
3498 console_init_seq(newcon, bootcon_registered);
3500 if (newcon->flags & CON_NBCON)
3504 * Put this console in the list - keep the
3505 * preferred driver at the head of the list.
3507 if (hlist_empty(&console_list)) {
3508 /* Ensure CON_CONSDEV is always set for the head. */
3509 newcon->flags |= CON_CONSDEV;
3510 hlist_add_head_rcu(&newcon->node, &console_list);
3512 } else if (newcon->flags & CON_CONSDEV) {
3513 /* Only the new head can have CON_CONSDEV set. */
3514 console_srcu_write_flags(console_first(), console_first()->flags & ~CON_CONSDEV);
3515 hlist_add_head_rcu(&newcon->node, &console_list);
3518 hlist_add_behind_rcu(&newcon->node, console_list.first);
3522 * No need to synchronize SRCU here! The caller does not rely
3523 * on all contexts being able to see the new console before
3524 * register_console() completes.
3527 console_sysfs_notify();
3530 * By unregistering the bootconsoles after we enable the real console
3531 * we get the "console xxx enabled" message on all the consoles -
3532 * boot consoles, real consoles, etc - this is to ensure that end
3533 * users know there might be something in the kernel's log buffer that
3534 * went to the bootconsole (that they do not see on the real console)
3536 con_printk(KERN_INFO, newcon, "enabled\n");
3537 if (bootcon_registered &&
3538 ((newcon->flags & (CON_CONSDEV | CON_BOOT)) == CON_CONSDEV) &&
3540 struct hlist_node *tmp;
3542 hlist_for_each_entry_safe(con, tmp, &console_list, node) {
3543 if (con->flags & CON_BOOT)
3544 unregister_console_locked(con);
3548 console_list_unlock();
3550 EXPORT_SYMBOL(register_console);
3552 /* Must be called under console_list_lock(). */
3553 static int unregister_console_locked(struct console *console)
3557 lockdep_assert_console_list_lock_held();
3559 con_printk(KERN_INFO, console, "disabled\n");
3561 res = _braille_unregister_console(console);
3567 /* Disable it unconditionally */
3568 console_srcu_write_flags(console, console->flags & ~CON_ENABLED);
3570 if (!console_is_registered_locked(console))
3573 hlist_del_init_rcu(&console->node);
3577 * If this isn't the last console and it has CON_CONSDEV set, we
3578 * need to set it on the next preferred console.
3581 * The above makes no sense as there is no guarantee that the next
3582 * console has any device attached. Oh well....
3584 if (!hlist_empty(&console_list) && console->flags & CON_CONSDEV)
3585 console_srcu_write_flags(console_first(), console_first()->flags | CON_CONSDEV);
3588 * Ensure that all SRCU list walks have completed. All contexts
3589 * must not be able to see this console in the list so that any
3590 * exit/cleanup routines can be performed safely.
3592 synchronize_srcu(&console_srcu);
3594 if (console->flags & CON_NBCON)
3595 nbcon_free(console);
3597 console_sysfs_notify();
3600 res = console->exit(console);
3605 int unregister_console(struct console *console)
3609 console_list_lock();
3610 res = unregister_console_locked(console);
3611 console_list_unlock();
3614 EXPORT_SYMBOL(unregister_console);
3617 * console_force_preferred_locked - force a registered console preferred
3618 * @con: The registered console to force preferred.
3620 * Must be called under console_list_lock().
3622 void console_force_preferred_locked(struct console *con)
3624 struct console *cur_pref_con;
3626 if (!console_is_registered_locked(con))
3629 cur_pref_con = console_first();
3631 /* Already preferred? */
3632 if (cur_pref_con == con)
3636 * Delete, but do not re-initialize the entry. This allows the console
3637 * to continue to appear registered (via any hlist_unhashed_lockless()
3638 * checks), even though it was briefly removed from the console list.
3640 hlist_del_rcu(&con->node);
3643 * Ensure that all SRCU list walks have completed so that the console
3644 * can be added to the beginning of the console list and its forward
3645 * list pointer can be re-initialized.
3647 synchronize_srcu(&console_srcu);
3649 con->flags |= CON_CONSDEV;
3650 WARN_ON(!con->device);
3652 /* Only the new head can have CON_CONSDEV set. */
3653 console_srcu_write_flags(cur_pref_con, cur_pref_con->flags & ~CON_CONSDEV);
3654 hlist_add_head_rcu(&con->node, &console_list);
3656 EXPORT_SYMBOL(console_force_preferred_locked);
3659 * Initialize the console device. This is called *early*, so
3660 * we can't necessarily depend on lots of kernel help here.
3661 * Just do some early initializations, and do the complex setup
3664 void __init console_init(void)
3668 initcall_entry_t *ce;
3670 /* Setup the default TTY line discipline. */
3674 * set up the console device so that later boot sequences can
3675 * inform about problems etc..
3677 ce = __con_initcall_start;
3678 trace_initcall_level("console");
3679 while (ce < __con_initcall_end) {
3680 call = initcall_from_entry(ce);
3681 trace_initcall_start(call);
3683 trace_initcall_finish(call, ret);
3689 * Some boot consoles access data that is in the init section and which will
3690 * be discarded after the initcalls have been run. To make sure that no code
3691 * will access this data, unregister the boot consoles in a late initcall.
3693 * If for some reason, such as deferred probe or the driver being a loadable
3694 * module, the real console hasn't registered yet at this point, there will
3695 * be a brief interval in which no messages are logged to the console, which
3696 * makes it difficult to diagnose problems that occur during this time.
3698 * To mitigate this problem somewhat, only unregister consoles whose memory
3699 * intersects with the init section. Note that all other boot consoles will
3700 * get unregistered when the real preferred console is registered.
3702 static int __init printk_late_init(void)
3704 struct hlist_node *tmp;
3705 struct console *con;
3708 console_list_lock();
3709 hlist_for_each_entry_safe(con, tmp, &console_list, node) {
3710 if (!(con->flags & CON_BOOT))
3713 /* Check addresses that might be used for enabled consoles. */
3714 if (init_section_intersects(con, sizeof(*con)) ||
3715 init_section_contains(con->write, 0) ||
3716 init_section_contains(con->read, 0) ||
3717 init_section_contains(con->device, 0) ||
3718 init_section_contains(con->unblank, 0) ||
3719 init_section_contains(con->data, 0)) {
3721 * Please, consider moving the reported consoles out
3722 * of the init section.
3724 pr_warn("bootconsole [%s%d] uses init memory and must be disabled even before the real one is ready\n",
3725 con->name, con->index);
3726 unregister_console_locked(con);
3729 console_list_unlock();
3731 ret = cpuhp_setup_state_nocalls(CPUHP_PRINTK_DEAD, "printk:dead", NULL,
3732 console_cpu_notify);
3734 ret = cpuhp_setup_state_nocalls(CPUHP_AP_ONLINE_DYN, "printk:online",
3735 console_cpu_notify, NULL);
3737 printk_sysctl_init();
3740 late_initcall(printk_late_init);
3742 #if defined CONFIG_PRINTK
3743 /* If @con is specified, only wait for that console. Otherwise wait for all. */
3744 static bool __pr_flush(struct console *con, int timeout_ms, bool reset_on_progress)
3746 unsigned long timeout_jiffies = msecs_to_jiffies(timeout_ms);
3747 unsigned long remaining_jiffies = timeout_jiffies;
3758 seq = prb_next_reserve_seq(prb);
3760 /* Flush the consoles so that records up to @seq are printed. */
3765 unsigned long begin_jiffies;
3766 unsigned long slept_jiffies;
3771 * Hold the console_lock to guarantee safe access to
3772 * console->seq. Releasing console_lock flushes more
3773 * records in case @seq is still not printed on all
3778 cookie = console_srcu_read_lock();
3779 for_each_console_srcu(c) {
3780 if (con && con != c)
3783 flags = console_srcu_read_flags(c);
3786 * If consoles are not usable, it cannot be expected
3787 * that they make forward progress, so only increment
3788 * @diff for usable consoles.
3790 if (!console_is_usable(c))
3793 if (flags & CON_NBCON) {
3794 printk_seq = nbcon_seq_read(c);
3796 printk_seq = c->seq;
3799 if (printk_seq < seq)
3800 diff += seq - printk_seq;
3802 console_srcu_read_unlock(cookie);
3804 if (diff != last_diff && reset_on_progress)
3805 remaining_jiffies = timeout_jiffies;
3809 /* Note: @diff is 0 if there are no usable consoles. */
3810 if (diff == 0 || remaining_jiffies == 0)
3813 /* msleep(1) might sleep much longer. Check time by jiffies. */
3814 begin_jiffies = jiffies;
3816 slept_jiffies = jiffies - begin_jiffies;
3818 remaining_jiffies -= min(slept_jiffies, remaining_jiffies);
3827 * pr_flush() - Wait for printing threads to catch up.
3829 * @timeout_ms: The maximum time (in ms) to wait.
3830 * @reset_on_progress: Reset the timeout if forward progress is seen.
3832 * A value of 0 for @timeout_ms means no waiting will occur. A value of -1
3833 * represents infinite waiting.
3835 * If @reset_on_progress is true, the timeout will be reset whenever any
3836 * printer has been seen to make some forward progress.
3838 * Context: Process context. May sleep while acquiring console lock.
3839 * Return: true if all usable printers are caught up.
3841 static bool pr_flush(int timeout_ms, bool reset_on_progress)
3843 return __pr_flush(NULL, timeout_ms, reset_on_progress);
3847 * Delayed printk version, for scheduler-internal messages:
3849 #define PRINTK_PENDING_WAKEUP 0x01
3850 #define PRINTK_PENDING_OUTPUT 0x02
3852 static DEFINE_PER_CPU(int, printk_pending);
3854 static void wake_up_klogd_work_func(struct irq_work *irq_work)
3856 int pending = this_cpu_xchg(printk_pending, 0);
3858 if (pending & PRINTK_PENDING_OUTPUT) {
3859 /* If trylock fails, someone else is doing the printing */
3860 if (console_trylock())
3864 if (pending & PRINTK_PENDING_WAKEUP)
3865 wake_up_interruptible(&log_wait);
3868 static DEFINE_PER_CPU(struct irq_work, wake_up_klogd_work) =
3869 IRQ_WORK_INIT_LAZY(wake_up_klogd_work_func);
3871 static void __wake_up_klogd(int val)
3873 if (!printk_percpu_data_ready())
3878 * Guarantee any new records can be seen by tasks preparing to wait
3879 * before this context checks if the wait queue is empty.
3881 * The full memory barrier within wq_has_sleeper() pairs with the full
3882 * memory barrier within set_current_state() of
3883 * prepare_to_wait_event(), which is called after ___wait_event() adds
3884 * the waiter but before it has checked the wait condition.
3886 * This pairs with devkmsg_read:A and syslog_print:A.
3888 if (wq_has_sleeper(&log_wait) || /* LMM(__wake_up_klogd:A) */
3889 (val & PRINTK_PENDING_OUTPUT)) {
3890 this_cpu_or(printk_pending, val);
3891 irq_work_queue(this_cpu_ptr(&wake_up_klogd_work));
3897 * wake_up_klogd - Wake kernel logging daemon
3899 * Use this function when new records have been added to the ringbuffer
3900 * and the console printing of those records has already occurred or is
3901 * known to be handled by some other context. This function will only
3902 * wake the logging daemon.
3904 * Context: Any context.
3906 void wake_up_klogd(void)
3908 __wake_up_klogd(PRINTK_PENDING_WAKEUP);
3912 * defer_console_output - Wake kernel logging daemon and trigger
3913 * console printing in a deferred context
3915 * Use this function when new records have been added to the ringbuffer,
3916 * this context is responsible for console printing those records, but
3917 * the current context is not allowed to perform the console printing.
3918 * Trigger an irq_work context to perform the console printing. This
3919 * function also wakes the logging daemon.
3921 * Context: Any context.
3923 void defer_console_output(void)
3926 * New messages may have been added directly to the ringbuffer
3927 * using vprintk_store(), so wake any waiters as well.
3929 __wake_up_klogd(PRINTK_PENDING_WAKEUP | PRINTK_PENDING_OUTPUT);
3932 void printk_trigger_flush(void)
3934 defer_console_output();
3937 int vprintk_deferred(const char *fmt, va_list args)
3939 return vprintk_emit(0, LOGLEVEL_SCHED, NULL, fmt, args);
3942 int _printk_deferred(const char *fmt, ...)
3947 va_start(args, fmt);
3948 r = vprintk_deferred(fmt, args);
3955 * printk rate limiting, lifted from the networking subsystem.
3957 * This enforces a rate limit: not more than 10 kernel messages
3958 * every 5s to make a denial-of-service attack impossible.
3960 DEFINE_RATELIMIT_STATE(printk_ratelimit_state, 5 * HZ, 10);
3962 int __printk_ratelimit(const char *func)
3964 return ___ratelimit(&printk_ratelimit_state, func);
3966 EXPORT_SYMBOL(__printk_ratelimit);
3969 * printk_timed_ratelimit - caller-controlled printk ratelimiting
3970 * @caller_jiffies: pointer to caller's state
3971 * @interval_msecs: minimum interval between prints
3973 * printk_timed_ratelimit() returns true if more than @interval_msecs
3974 * milliseconds have elapsed since the last time printk_timed_ratelimit()
3977 bool printk_timed_ratelimit(unsigned long *caller_jiffies,
3978 unsigned int interval_msecs)
3980 unsigned long elapsed = jiffies - *caller_jiffies;
3982 if (*caller_jiffies && elapsed <= msecs_to_jiffies(interval_msecs))
3985 *caller_jiffies = jiffies;
3988 EXPORT_SYMBOL(printk_timed_ratelimit);
3990 static DEFINE_SPINLOCK(dump_list_lock);
3991 static LIST_HEAD(dump_list);
3994 * kmsg_dump_register - register a kernel log dumper.
3995 * @dumper: pointer to the kmsg_dumper structure
3997 * Adds a kernel log dumper to the system. The dump callback in the
3998 * structure will be called when the kernel oopses or panics and must be
3999 * set. Returns zero on success and %-EINVAL or %-EBUSY otherwise.
4001 int kmsg_dump_register(struct kmsg_dumper *dumper)
4003 unsigned long flags;
4006 /* The dump callback needs to be set */
4010 spin_lock_irqsave(&dump_list_lock, flags);
4011 /* Don't allow registering multiple times */
4012 if (!dumper->registered) {
4013 dumper->registered = 1;
4014 list_add_tail_rcu(&dumper->list, &dump_list);
4017 spin_unlock_irqrestore(&dump_list_lock, flags);
4021 EXPORT_SYMBOL_GPL(kmsg_dump_register);
4024 * kmsg_dump_unregister - unregister a kmsg dumper.
4025 * @dumper: pointer to the kmsg_dumper structure
4027 * Removes a dump device from the system. Returns zero on success and
4028 * %-EINVAL otherwise.
4030 int kmsg_dump_unregister(struct kmsg_dumper *dumper)
4032 unsigned long flags;
4035 spin_lock_irqsave(&dump_list_lock, flags);
4036 if (dumper->registered) {
4037 dumper->registered = 0;
4038 list_del_rcu(&dumper->list);
4041 spin_unlock_irqrestore(&dump_list_lock, flags);
4046 EXPORT_SYMBOL_GPL(kmsg_dump_unregister);
4048 static bool always_kmsg_dump;
4049 module_param_named(always_kmsg_dump, always_kmsg_dump, bool, S_IRUGO | S_IWUSR);
4051 const char *kmsg_dump_reason_str(enum kmsg_dump_reason reason)
4054 case KMSG_DUMP_PANIC:
4056 case KMSG_DUMP_OOPS:
4058 case KMSG_DUMP_EMERG:
4060 case KMSG_DUMP_SHUTDOWN:
4066 EXPORT_SYMBOL_GPL(kmsg_dump_reason_str);
4069 * kmsg_dump - dump kernel log to kernel message dumpers.
4070 * @reason: the reason (oops, panic etc) for dumping
4072 * Call each of the registered dumper's dump() callback, which can
4073 * retrieve the kmsg records with kmsg_dump_get_line() or
4074 * kmsg_dump_get_buffer().
4076 void kmsg_dump(enum kmsg_dump_reason reason)
4078 struct kmsg_dumper *dumper;
4081 list_for_each_entry_rcu(dumper, &dump_list, list) {
4082 enum kmsg_dump_reason max_reason = dumper->max_reason;
4085 * If client has not provided a specific max_reason, default
4086 * to KMSG_DUMP_OOPS, unless always_kmsg_dump was set.
4088 if (max_reason == KMSG_DUMP_UNDEF) {
4089 max_reason = always_kmsg_dump ? KMSG_DUMP_MAX :
4092 if (reason > max_reason)
4095 /* invoke dumper which will iterate over records */
4096 dumper->dump(dumper, reason);
4102 * kmsg_dump_get_line - retrieve one kmsg log line
4103 * @iter: kmsg dump iterator
4104 * @syslog: include the "<4>" prefixes
4105 * @line: buffer to copy the line to
4106 * @size: maximum size of the buffer
4107 * @len: length of line placed into buffer
4109 * Start at the beginning of the kmsg buffer, with the oldest kmsg
4110 * record, and copy one record into the provided buffer.
4112 * Consecutive calls will return the next available record moving
4113 * towards the end of the buffer with the youngest messages.
4115 * A return value of FALSE indicates that there are no more records to
4118 bool kmsg_dump_get_line(struct kmsg_dump_iter *iter, bool syslog,
4119 char *line, size_t size, size_t *len)
4121 u64 min_seq = latched_seq_read_nolock(&clear_seq);
4122 struct printk_info info;
4123 unsigned int line_count;
4124 struct printk_record r;
4128 if (iter->cur_seq < min_seq)
4129 iter->cur_seq = min_seq;
4131 prb_rec_init_rd(&r, &info, line, size);
4133 /* Read text or count text lines? */
4135 if (!prb_read_valid(prb, iter->cur_seq, &r))
4137 l = record_print_text(&r, syslog, printk_time);
4139 if (!prb_read_valid_info(prb, iter->cur_seq,
4140 &info, &line_count)) {
4143 l = get_record_print_text_size(&info, line_count, syslog,
4148 iter->cur_seq = r.info->seq + 1;
4155 EXPORT_SYMBOL_GPL(kmsg_dump_get_line);
4158 * kmsg_dump_get_buffer - copy kmsg log lines
4159 * @iter: kmsg dump iterator
4160 * @syslog: include the "<4>" prefixes
4161 * @buf: buffer to copy the line to
4162 * @size: maximum size of the buffer
4163 * @len_out: length of line placed into buffer
4165 * Start at the end of the kmsg buffer and fill the provided buffer
4166 * with as many of the *youngest* kmsg records that fit into it.
4167 * If the buffer is large enough, all available kmsg records will be
4168 * copied with a single call.
4170 * Consecutive calls will fill the buffer with the next block of
4171 * available older records, not including the earlier retrieved ones.
4173 * A return value of FALSE indicates that there are no more records to
4176 bool kmsg_dump_get_buffer(struct kmsg_dump_iter *iter, bool syslog,
4177 char *buf, size_t size, size_t *len_out)
4179 u64 min_seq = latched_seq_read_nolock(&clear_seq);
4180 struct printk_info info;
4181 struct printk_record r;
4186 bool time = printk_time;
4191 if (iter->cur_seq < min_seq)
4192 iter->cur_seq = min_seq;
4194 if (prb_read_valid_info(prb, iter->cur_seq, &info, NULL)) {
4195 if (info.seq != iter->cur_seq) {
4196 /* messages are gone, move to first available one */
4197 iter->cur_seq = info.seq;
4202 if (iter->cur_seq >= iter->next_seq)
4206 * Find first record that fits, including all following records,
4207 * into the user-provided buffer for this dump. Pass in size-1
4208 * because this function (by way of record_print_text()) will
4209 * not write more than size-1 bytes of text into @buf.
4211 seq = find_first_fitting_seq(iter->cur_seq, iter->next_seq,
4212 size - 1, syslog, time);
4215 * Next kmsg_dump_get_buffer() invocation will dump block of
4216 * older records stored right before this one.
4220 prb_rec_init_rd(&r, &info, buf, size);
4222 prb_for_each_record(seq, prb, seq, &r) {
4223 if (r.info->seq >= iter->next_seq)
4226 len += record_print_text(&r, syslog, time);
4228 /* Adjust record to store to remaining buffer space. */
4229 prb_rec_init_rd(&r, &info, buf + len, size - len);
4232 iter->next_seq = next_seq;
4239 EXPORT_SYMBOL_GPL(kmsg_dump_get_buffer);
4242 * kmsg_dump_rewind - reset the iterator
4243 * @iter: kmsg dump iterator
4245 * Reset the dumper's iterator so that kmsg_dump_get_line() and
4246 * kmsg_dump_get_buffer() can be called again and used multiple
4247 * times within the same dumper.dump() callback.
4249 void kmsg_dump_rewind(struct kmsg_dump_iter *iter)
4251 iter->cur_seq = latched_seq_read_nolock(&clear_seq);
4252 iter->next_seq = prb_next_seq(prb);
4254 EXPORT_SYMBOL_GPL(kmsg_dump_rewind);
4259 static atomic_t printk_cpu_sync_owner = ATOMIC_INIT(-1);
4260 static atomic_t printk_cpu_sync_nested = ATOMIC_INIT(0);
4263 * __printk_cpu_sync_wait() - Busy wait until the printk cpu-reentrant
4264 * spinning lock is not owned by any CPU.
4266 * Context: Any context.
4268 void __printk_cpu_sync_wait(void)
4272 } while (atomic_read(&printk_cpu_sync_owner) != -1);
4274 EXPORT_SYMBOL(__printk_cpu_sync_wait);
4277 * __printk_cpu_sync_try_get() - Try to acquire the printk cpu-reentrant
4280 * If no processor has the lock, the calling processor takes the lock and
4281 * becomes the owner. If the calling processor is already the owner of the
4282 * lock, this function succeeds immediately.
4284 * Context: Any context. Expects interrupts to be disabled.
4285 * Return: 1 on success, otherwise 0.
4287 int __printk_cpu_sync_try_get(void)
4292 cpu = smp_processor_id();
4295 * Guarantee loads and stores from this CPU when it is the lock owner
4296 * are _not_ visible to the previous lock owner. This pairs with
4297 * __printk_cpu_sync_put:B.
4299 * Memory barrier involvement:
4301 * If __printk_cpu_sync_try_get:A reads from __printk_cpu_sync_put:B,
4302 * then __printk_cpu_sync_put:A can never read from
4303 * __printk_cpu_sync_try_get:B.
4307 * RELEASE from __printk_cpu_sync_put:A to __printk_cpu_sync_put:B
4308 * of the previous CPU
4310 * ACQUIRE from __printk_cpu_sync_try_get:A to
4311 * __printk_cpu_sync_try_get:B of this CPU
4313 old = atomic_cmpxchg_acquire(&printk_cpu_sync_owner, -1,
4314 cpu); /* LMM(__printk_cpu_sync_try_get:A) */
4317 * This CPU is now the owner and begins loading/storing
4318 * data: LMM(__printk_cpu_sync_try_get:B)
4322 } else if (old == cpu) {
4323 /* This CPU is already the owner. */
4324 atomic_inc(&printk_cpu_sync_nested);
4330 EXPORT_SYMBOL(__printk_cpu_sync_try_get);
4333 * __printk_cpu_sync_put() - Release the printk cpu-reentrant spinning lock.
4335 * The calling processor must be the owner of the lock.
4337 * Context: Any context. Expects interrupts to be disabled.
4339 void __printk_cpu_sync_put(void)
4341 if (atomic_read(&printk_cpu_sync_nested)) {
4342 atomic_dec(&printk_cpu_sync_nested);
4347 * This CPU is finished loading/storing data:
4348 * LMM(__printk_cpu_sync_put:A)
4352 * Guarantee loads and stores from this CPU when it was the
4353 * lock owner are visible to the next lock owner. This pairs
4354 * with __printk_cpu_sync_try_get:A.
4356 * Memory barrier involvement:
4358 * If __printk_cpu_sync_try_get:A reads from __printk_cpu_sync_put:B,
4359 * then __printk_cpu_sync_try_get:B reads from __printk_cpu_sync_put:A.
4363 * RELEASE from __printk_cpu_sync_put:A to __printk_cpu_sync_put:B
4366 * ACQUIRE from __printk_cpu_sync_try_get:A to
4367 * __printk_cpu_sync_try_get:B of the next CPU
4369 atomic_set_release(&printk_cpu_sync_owner,
4370 -1); /* LMM(__printk_cpu_sync_put:B) */
4372 EXPORT_SYMBOL(__printk_cpu_sync_put);
4373 #endif /* CONFIG_SMP */