1 // SPDX-License-Identifier: GPL-2.0-only
3 * linux/kernel/printk.c
5 * Copyright (C) 1991, 1992 Linus Torvalds
7 * Modified to make sys_syslog() more flexible: added commands to
8 * return the last 4k of kernel messages, regardless of whether
9 * they've been read or not. Added option to suppress kernel printk's
10 * to the console. Added hook for sending the console messages
11 * elsewhere, in preparation for a serial line console (someday).
13 * Modified for sysctl support, 1/8/97, Chris Horn.
14 * Fixed SMP synchronization, 08/08/99, Manfred Spraul
15 * manfred@colorfullife.com
16 * Rewrote bits to get rid of console_lock
17 * 01Mar01 Andrew Morton
20 #define pr_fmt(fmt) KBUILD_MODNAME ": " fmt
22 #include <linux/kernel.h>
24 #include <linux/tty.h>
25 #include <linux/tty_driver.h>
26 #include <linux/console.h>
27 #include <linux/init.h>
28 #include <linux/jiffies.h>
29 #include <linux/nmi.h>
30 #include <linux/module.h>
31 #include <linux/moduleparam.h>
32 #include <linux/delay.h>
33 #include <linux/smp.h>
34 #include <linux/security.h>
35 #include <linux/memblock.h>
36 #include <linux/syscalls.h>
37 #include <linux/crash_core.h>
38 #include <linux/ratelimit.h>
39 #include <linux/kmsg_dump.h>
40 #include <linux/syslog.h>
41 #include <linux/cpu.h>
42 #include <linux/rculist.h>
43 #include <linux/poll.h>
44 #include <linux/irq_work.h>
45 #include <linux/ctype.h>
46 #include <linux/uio.h>
47 #include <linux/sched/clock.h>
48 #include <linux/sched/debug.h>
49 #include <linux/sched/task_stack.h>
51 #include <linux/uaccess.h>
52 #include <asm/sections.h>
54 #include <trace/events/initcall.h>
55 #define CREATE_TRACE_POINTS
56 #include <trace/events/printk.h>
58 #include "printk_ringbuffer.h"
59 #include "console_cmdline.h"
63 int console_printk[4] = {
64 CONSOLE_LOGLEVEL_DEFAULT, /* console_loglevel */
65 MESSAGE_LOGLEVEL_DEFAULT, /* default_message_loglevel */
66 CONSOLE_LOGLEVEL_MIN, /* minimum_console_loglevel */
67 CONSOLE_LOGLEVEL_DEFAULT, /* default_console_loglevel */
69 EXPORT_SYMBOL_GPL(console_printk);
71 atomic_t ignore_console_lock_warning __read_mostly = ATOMIC_INIT(0);
72 EXPORT_SYMBOL(ignore_console_lock_warning);
75 * Low level drivers may need that to know if they can schedule in
76 * their unblank() callback or not. So let's export it.
79 EXPORT_SYMBOL(oops_in_progress);
82 * console_sem protects the console_drivers list, and also
83 * provides serialisation for access to the entire console
86 static DEFINE_SEMAPHORE(console_sem);
87 struct console *console_drivers;
88 EXPORT_SYMBOL_GPL(console_drivers);
91 * System may need to suppress printk message under certain
92 * circumstances, like after kernel panic happens.
94 int __read_mostly suppress_printk;
97 * During panic, heavy printk by other CPUs can delay the
98 * panic and risk deadlock on console resources.
100 static int __read_mostly suppress_panic_printk;
102 #ifdef CONFIG_LOCKDEP
103 static struct lockdep_map console_lock_dep_map = {
104 .name = "console_lock"
108 enum devkmsg_log_bits {
109 __DEVKMSG_LOG_BIT_ON = 0,
110 __DEVKMSG_LOG_BIT_OFF,
111 __DEVKMSG_LOG_BIT_LOCK,
114 enum devkmsg_log_masks {
115 DEVKMSG_LOG_MASK_ON = BIT(__DEVKMSG_LOG_BIT_ON),
116 DEVKMSG_LOG_MASK_OFF = BIT(__DEVKMSG_LOG_BIT_OFF),
117 DEVKMSG_LOG_MASK_LOCK = BIT(__DEVKMSG_LOG_BIT_LOCK),
120 /* Keep both the 'on' and 'off' bits clear, i.e. ratelimit by default: */
121 #define DEVKMSG_LOG_MASK_DEFAULT 0
123 static unsigned int __read_mostly devkmsg_log = DEVKMSG_LOG_MASK_DEFAULT;
125 static int __control_devkmsg(char *str)
132 len = str_has_prefix(str, "on");
134 devkmsg_log = DEVKMSG_LOG_MASK_ON;
138 len = str_has_prefix(str, "off");
140 devkmsg_log = DEVKMSG_LOG_MASK_OFF;
144 len = str_has_prefix(str, "ratelimit");
146 devkmsg_log = DEVKMSG_LOG_MASK_DEFAULT;
153 static int __init control_devkmsg(char *str)
155 if (__control_devkmsg(str) < 0) {
156 pr_warn("printk.devkmsg: bad option string '%s'\n", str);
161 * Set sysctl string accordingly:
163 if (devkmsg_log == DEVKMSG_LOG_MASK_ON)
164 strcpy(devkmsg_log_str, "on");
165 else if (devkmsg_log == DEVKMSG_LOG_MASK_OFF)
166 strcpy(devkmsg_log_str, "off");
167 /* else "ratelimit" which is set by default. */
170 * Sysctl cannot change it anymore. The kernel command line setting of
171 * this parameter is to force the setting to be permanent throughout the
172 * runtime of the system. This is a precation measure against userspace
173 * trying to be a smarta** and attempting to change it up on us.
175 devkmsg_log |= DEVKMSG_LOG_MASK_LOCK;
179 __setup("printk.devkmsg=", control_devkmsg);
181 char devkmsg_log_str[DEVKMSG_STR_MAX_SIZE] = "ratelimit";
182 #if defined(CONFIG_PRINTK) && defined(CONFIG_SYSCTL)
183 int devkmsg_sysctl_set_loglvl(struct ctl_table *table, int write,
184 void *buffer, size_t *lenp, loff_t *ppos)
186 char old_str[DEVKMSG_STR_MAX_SIZE];
191 if (devkmsg_log & DEVKMSG_LOG_MASK_LOCK)
195 strncpy(old_str, devkmsg_log_str, DEVKMSG_STR_MAX_SIZE);
198 err = proc_dostring(table, write, buffer, lenp, ppos);
203 err = __control_devkmsg(devkmsg_log_str);
206 * Do not accept an unknown string OR a known string with
209 if (err < 0 || (err + 1 != *lenp)) {
211 /* ... and restore old setting. */
213 strncpy(devkmsg_log_str, old_str, DEVKMSG_STR_MAX_SIZE);
221 #endif /* CONFIG_PRINTK && CONFIG_SYSCTL */
223 /* Number of registered extended console drivers. */
224 static int nr_ext_console_drivers;
227 * Used to synchronize printing kthreads against direct printing via
228 * console_trylock/console_unlock.
231 * -1 = console kthreads atomically blocked (via global trylock)
232 * 0 = no kthread printing, console not locked (via trylock)
233 * >0 = kthread(s) actively printing
235 * Note: For synchronizing against direct printing via
236 * console_lock/console_unlock, see the @lock variable in
239 static atomic_t console_kthreads_active = ATOMIC_INIT(0);
241 #define console_kthreads_atomic_tryblock() \
242 (atomic_cmpxchg(&console_kthreads_active, 0, -1) == 0)
243 #define console_kthreads_atomic_unblock() \
244 atomic_cmpxchg(&console_kthreads_active, -1, 0)
245 #define console_kthreads_atomically_blocked() \
246 (atomic_read(&console_kthreads_active) == -1)
248 #define console_kthread_printing_tryenter() \
249 atomic_inc_unless_negative(&console_kthreads_active)
250 #define console_kthread_printing_exit() \
251 atomic_dec(&console_kthreads_active)
254 * Helper macros to handle lockdep when locking/unlocking console_sem. We use
255 * macros instead of functions so that _RET_IP_ contains useful information.
257 #define down_console_sem() do { \
259 mutex_acquire(&console_lock_dep_map, 0, 0, _RET_IP_);\
262 static int __down_trylock_console_sem(unsigned long ip)
268 * Here and in __up_console_sem() we need to be in safe mode,
269 * because spindump/WARN/etc from under console ->lock will
270 * deadlock in printk()->down_trylock_console_sem() otherwise.
272 printk_safe_enter_irqsave(flags);
273 lock_failed = down_trylock(&console_sem);
274 printk_safe_exit_irqrestore(flags);
278 mutex_acquire(&console_lock_dep_map, 0, 1, ip);
281 #define down_trylock_console_sem() __down_trylock_console_sem(_RET_IP_)
283 static void __up_console_sem(unsigned long ip)
287 mutex_release(&console_lock_dep_map, ip);
289 printk_safe_enter_irqsave(flags);
291 printk_safe_exit_irqrestore(flags);
293 #define up_console_sem() __up_console_sem(_RET_IP_)
295 static bool panic_in_progress(void)
297 return unlikely(atomic_read(&panic_cpu) != PANIC_CPU_INVALID);
301 * Tracks whether kthread printers are all blocked. A value of true implies
302 * that the console is locked via console_lock() or the console is suspended.
303 * Writing to this variable requires holding @console_sem.
305 static bool console_kthreads_blocked;
308 * Block all kthread printers from a schedulable context.
310 * Requires holding @console_sem.
312 static void console_kthreads_block(void)
316 for_each_console(con) {
317 mutex_lock(&con->lock);
319 mutex_unlock(&con->lock);
322 console_kthreads_blocked = true;
326 * Unblock all kthread printers from a schedulable context.
328 * Requires holding @console_sem.
330 static void console_kthreads_unblock(void)
334 for_each_console(con) {
335 mutex_lock(&con->lock);
336 con->blocked = false;
337 mutex_unlock(&con->lock);
340 console_kthreads_blocked = false;
343 static int console_suspended;
346 * Array of consoles built from command line options (console=)
349 #define MAX_CMDLINECONSOLES 8
351 static struct console_cmdline console_cmdline[MAX_CMDLINECONSOLES];
353 static int preferred_console = -1;
354 int console_set_on_cmdline;
355 EXPORT_SYMBOL(console_set_on_cmdline);
357 /* Flag: console code may call schedule() */
358 static int console_may_schedule;
360 enum con_msg_format_flags {
361 MSG_FORMAT_DEFAULT = 0,
362 MSG_FORMAT_SYSLOG = (1 << 0),
365 static int console_msg_format = MSG_FORMAT_DEFAULT;
368 * The printk log buffer consists of a sequenced collection of records, each
369 * containing variable length message text. Every record also contains its
370 * own meta-data (@info).
372 * Every record meta-data carries the timestamp in microseconds, as well as
373 * the standard userspace syslog level and syslog facility. The usual kernel
374 * messages use LOG_KERN; userspace-injected messages always carry a matching
375 * syslog facility, by default LOG_USER. The origin of every message can be
376 * reliably determined that way.
378 * The human readable log message of a record is available in @text, the
379 * length of the message text in @text_len. The stored message is not
382 * Optionally, a record can carry a dictionary of properties (key/value
383 * pairs), to provide userspace with a machine-readable message context.
385 * Examples for well-defined, commonly used property names are:
386 * DEVICE=b12:8 device identifier
390 * +sound:card0 subsystem:devname
391 * SUBSYSTEM=pci driver-core subsystem name
393 * Valid characters in property names are [a-zA-Z0-9.-_]. Property names
394 * and values are terminated by a '\0' character.
396 * Example of record values:
397 * record.text_buf = "it's a line" (unterminated)
398 * record.info.seq = 56
399 * record.info.ts_nsec = 36863
400 * record.info.text_len = 11
401 * record.info.facility = 0 (LOG_KERN)
402 * record.info.flags = 0
403 * record.info.level = 3 (LOG_ERR)
404 * record.info.caller_id = 299 (task 299)
405 * record.info.dev_info.subsystem = "pci" (terminated)
406 * record.info.dev_info.device = "+pci:0000:00:01.0" (terminated)
408 * The 'struct printk_info' buffer must never be directly exported to
409 * userspace, it is a kernel-private implementation detail that might
410 * need to be changed in the future, when the requirements change.
412 * /dev/kmsg exports the structured data in the following line format:
413 * "<level>,<sequnum>,<timestamp>,<contflag>[,additional_values, ... ];<message text>\n"
415 * Users of the export format should ignore possible additional values
416 * separated by ',', and find the message after the ';' character.
418 * The optional key/value pairs are attached as continuation lines starting
419 * with a space character and terminated by a newline. All possible
420 * non-prinatable characters are escaped in the "\xff" notation.
423 /* syslog_lock protects syslog_* variables and write access to clear_seq. */
424 static DEFINE_MUTEX(syslog_lock);
427 * A flag to signify if printk_activate_kthreads() has already started the
428 * kthread printers. If true, any later registered consoles must start their
429 * own kthread directly. The flag is write protected by the console_lock.
431 static bool printk_kthreads_available;
434 static atomic_t printk_prefer_direct = ATOMIC_INIT(0);
437 * printk_prefer_direct_enter - cause printk() calls to attempt direct
438 * printing to all enabled consoles
440 * Since it is not possible to call into the console printing code from any
441 * context, there is no guarantee that direct printing will occur.
443 * This globally effects all printk() callers.
445 * Context: Any context.
447 void printk_prefer_direct_enter(void)
449 atomic_inc(&printk_prefer_direct);
453 * printk_prefer_direct_exit - restore printk() behavior
455 * Context: Any context.
457 void printk_prefer_direct_exit(void)
459 WARN_ON(atomic_dec_if_positive(&printk_prefer_direct) < 0);
463 * Calling printk() always wakes kthread printers so that they can
464 * flush the new message to their respective consoles. Also, if direct
465 * printing is allowed, printk() tries to flush the messages directly.
467 * Direct printing is allowed in situations when the kthreads
468 * are not available or the system is in a problematic state.
470 * See the implementation about possible races.
472 static inline bool allow_direct_printing(void)
475 * Checking kthread availability is a possible race because the
476 * kthread printers can become permanently disabled during runtime.
477 * However, doing that requires holding the console_lock, so any
478 * pending messages will be direct printed by console_unlock().
480 if (!printk_kthreads_available)
484 * Prefer direct printing when the system is in a problematic state.
485 * The context that sets this state will always see the updated value.
486 * The other contexts do not care. Anyway, direct printing is just a
487 * best effort. The direct output is only possible when console_lock
488 * is not already taken and no kthread printers are actively printing.
490 return (system_state > SYSTEM_RUNNING ||
492 atomic_read(&printk_prefer_direct));
495 DECLARE_WAIT_QUEUE_HEAD(log_wait);
496 /* All 3 protected by @syslog_lock. */
497 /* the next printk record to read by syslog(READ) or /proc/kmsg */
498 static u64 syslog_seq;
499 static size_t syslog_partial;
500 static bool syslog_time;
503 seqcount_latch_t latch;
508 * The next printk record to read after the last 'clear' command. There are
509 * two copies (updated with seqcount_latch) so that reads can locklessly
510 * access a valid value. Writers are synchronized by @syslog_lock.
512 static struct latched_seq clear_seq = {
513 .latch = SEQCNT_LATCH_ZERO(clear_seq.latch),
518 #ifdef CONFIG_PRINTK_CALLER
519 #define PREFIX_MAX 48
521 #define PREFIX_MAX 32
524 /* the maximum size of a formatted record (i.e. with prefix added per line) */
525 #define CONSOLE_LOG_MAX 1024
527 /* the maximum size for a dropped text message */
528 #define DROPPED_TEXT_MAX 64
530 /* the maximum size allowed to be reserved for a record */
531 #define LOG_LINE_MAX (CONSOLE_LOG_MAX - PREFIX_MAX)
533 #define LOG_LEVEL(v) ((v) & 0x07)
534 #define LOG_FACILITY(v) ((v) >> 3 & 0xff)
537 #define LOG_ALIGN __alignof__(unsigned long)
538 #define __LOG_BUF_LEN (1 << CONFIG_LOG_BUF_SHIFT)
539 #define LOG_BUF_LEN_MAX (u32)(1 << 31)
540 static char __log_buf[__LOG_BUF_LEN] __aligned(LOG_ALIGN);
541 static char *log_buf = __log_buf;
542 static u32 log_buf_len = __LOG_BUF_LEN;
545 * Define the average message size. This only affects the number of
546 * descriptors that will be available. Underestimating is better than
547 * overestimating (too many available descriptors is better than not enough).
549 #define PRB_AVGBITS 5 /* 32 character average length */
551 #if CONFIG_LOG_BUF_SHIFT <= PRB_AVGBITS
552 #error CONFIG_LOG_BUF_SHIFT value too small.
554 _DEFINE_PRINTKRB(printk_rb_static, CONFIG_LOG_BUF_SHIFT - PRB_AVGBITS,
555 PRB_AVGBITS, &__log_buf[0]);
557 static struct printk_ringbuffer printk_rb_dynamic;
559 static struct printk_ringbuffer *prb = &printk_rb_static;
562 * We cannot access per-CPU data (e.g. per-CPU flush irq_work) before
563 * per_cpu_areas are initialised. This variable is set to true when
564 * it's safe to access per-CPU data.
566 static bool __printk_percpu_data_ready __read_mostly;
568 bool printk_percpu_data_ready(void)
570 return __printk_percpu_data_ready;
573 /* Must be called under syslog_lock. */
574 static void latched_seq_write(struct latched_seq *ls, u64 val)
576 raw_write_seqcount_latch(&ls->latch);
578 raw_write_seqcount_latch(&ls->latch);
582 /* Can be called from any context. */
583 static u64 latched_seq_read_nolock(struct latched_seq *ls)
590 seq = raw_read_seqcount_latch(&ls->latch);
593 } while (read_seqcount_latch_retry(&ls->latch, seq));
598 /* Return log buffer address */
599 char *log_buf_addr_get(void)
604 /* Return log buffer size */
605 u32 log_buf_len_get(void)
611 * Define how much of the log buffer we could take at maximum. The value
612 * must be greater than two. Note that only half of the buffer is available
613 * when the index points to the middle.
615 #define MAX_LOG_TAKE_PART 4
616 static const char trunc_msg[] = "<truncated>";
618 static void truncate_msg(u16 *text_len, u16 *trunc_msg_len)
621 * The message should not take the whole buffer. Otherwise, it might
622 * get removed too soon.
624 u32 max_text_len = log_buf_len / MAX_LOG_TAKE_PART;
626 if (*text_len > max_text_len)
627 *text_len = max_text_len;
629 /* enable the warning message (if there is room) */
630 *trunc_msg_len = strlen(trunc_msg);
631 if (*text_len >= *trunc_msg_len)
632 *text_len -= *trunc_msg_len;
637 int dmesg_restrict = IS_ENABLED(CONFIG_SECURITY_DMESG_RESTRICT);
639 static int syslog_action_restricted(int type)
644 * Unless restricted, we allow "read all" and "get buffer size"
647 return type != SYSLOG_ACTION_READ_ALL &&
648 type != SYSLOG_ACTION_SIZE_BUFFER;
651 static int check_syslog_permissions(int type, int source)
654 * If this is from /proc/kmsg and we've already opened it, then we've
655 * already done the capabilities checks at open time.
657 if (source == SYSLOG_FROM_PROC && type != SYSLOG_ACTION_OPEN)
660 if (syslog_action_restricted(type)) {
661 if (capable(CAP_SYSLOG))
664 * For historical reasons, accept CAP_SYS_ADMIN too, with
667 if (capable(CAP_SYS_ADMIN)) {
668 pr_warn_once("%s (%d): Attempt to access syslog with "
669 "CAP_SYS_ADMIN but no CAP_SYSLOG "
671 current->comm, task_pid_nr(current));
677 return security_syslog(type);
680 static void append_char(char **pp, char *e, char c)
686 static ssize_t info_print_ext_header(char *buf, size_t size,
687 struct printk_info *info)
689 u64 ts_usec = info->ts_nsec;
691 #ifdef CONFIG_PRINTK_CALLER
692 u32 id = info->caller_id;
694 snprintf(caller, sizeof(caller), ",caller=%c%u",
695 id & 0x80000000 ? 'C' : 'T', id & ~0x80000000);
700 do_div(ts_usec, 1000);
702 return scnprintf(buf, size, "%u,%llu,%llu,%c%s;",
703 (info->facility << 3) | info->level, info->seq,
704 ts_usec, info->flags & LOG_CONT ? 'c' : '-', caller);
707 static ssize_t msg_add_ext_text(char *buf, size_t size,
708 const char *text, size_t text_len,
711 char *p = buf, *e = buf + size;
714 /* escape non-printable characters */
715 for (i = 0; i < text_len; i++) {
716 unsigned char c = text[i];
718 if (c < ' ' || c >= 127 || c == '\\')
719 p += scnprintf(p, e - p, "\\x%02x", c);
721 append_char(&p, e, c);
723 append_char(&p, e, endc);
728 static ssize_t msg_add_dict_text(char *buf, size_t size,
729 const char *key, const char *val)
731 size_t val_len = strlen(val);
737 len = msg_add_ext_text(buf, size, "", 0, ' '); /* dict prefix */
738 len += msg_add_ext_text(buf + len, size - len, key, strlen(key), '=');
739 len += msg_add_ext_text(buf + len, size - len, val, val_len, '\n');
744 static ssize_t msg_print_ext_body(char *buf, size_t size,
745 char *text, size_t text_len,
746 struct dev_printk_info *dev_info)
750 len = msg_add_ext_text(buf, size, text, text_len, '\n');
755 len += msg_add_dict_text(buf + len, size - len, "SUBSYSTEM",
756 dev_info->subsystem);
757 len += msg_add_dict_text(buf + len, size - len, "DEVICE",
763 /* /dev/kmsg - userspace message inject/listen interface */
764 struct devkmsg_user {
766 struct ratelimit_state rs;
768 char buf[CONSOLE_EXT_LOG_MAX];
770 struct printk_info info;
771 char text_buf[CONSOLE_EXT_LOG_MAX];
772 struct printk_record record;
775 static __printf(3, 4) __cold
776 int devkmsg_emit(int facility, int level, const char *fmt, ...)
782 r = vprintk_emit(facility, level, NULL, fmt, args);
788 static ssize_t devkmsg_write(struct kiocb *iocb, struct iov_iter *from)
791 int level = default_message_loglevel;
792 int facility = 1; /* LOG_USER */
793 struct file *file = iocb->ki_filp;
794 struct devkmsg_user *user = file->private_data;
795 size_t len = iov_iter_count(from);
798 if (!user || len > LOG_LINE_MAX)
801 /* Ignore when user logging is disabled. */
802 if (devkmsg_log & DEVKMSG_LOG_MASK_OFF)
805 /* Ratelimit when not explicitly enabled. */
806 if (!(devkmsg_log & DEVKMSG_LOG_MASK_ON)) {
807 if (!___ratelimit(&user->rs, current->comm))
811 buf = kmalloc(len+1, GFP_KERNEL);
816 if (!copy_from_iter_full(buf, len, from)) {
822 * Extract and skip the syslog prefix <[0-9]*>. Coming from userspace
823 * the decimal value represents 32bit, the lower 3 bit are the log
824 * level, the rest are the log facility.
826 * If no prefix or no userspace facility is specified, we
827 * enforce LOG_USER, to be able to reliably distinguish
828 * kernel-generated messages from userspace-injected ones.
831 if (line[0] == '<') {
835 u = simple_strtoul(line + 1, &endp, 10);
836 if (endp && endp[0] == '>') {
837 level = LOG_LEVEL(u);
838 if (LOG_FACILITY(u) != 0)
839 facility = LOG_FACILITY(u);
845 devkmsg_emit(facility, level, "%s", line);
850 static ssize_t devkmsg_read(struct file *file, char __user *buf,
851 size_t count, loff_t *ppos)
853 struct devkmsg_user *user = file->private_data;
854 struct printk_record *r = &user->record;
861 ret = mutex_lock_interruptible(&user->lock);
865 if (!prb_read_valid(prb, atomic64_read(&user->seq), r)) {
866 if (file->f_flags & O_NONBLOCK) {
872 * Guarantee this task is visible on the waitqueue before
873 * checking the wake condition.
875 * The full memory barrier within set_current_state() of
876 * prepare_to_wait_event() pairs with the full memory barrier
877 * within wq_has_sleeper().
879 * This pairs with __wake_up_klogd:A.
881 ret = wait_event_interruptible(log_wait,
883 atomic64_read(&user->seq), r)); /* LMM(devkmsg_read:A) */
888 if (r->info->seq != atomic64_read(&user->seq)) {
889 /* our last seen message is gone, return error and reset */
890 atomic64_set(&user->seq, r->info->seq);
895 len = info_print_ext_header(user->buf, sizeof(user->buf), r->info);
896 len += msg_print_ext_body(user->buf + len, sizeof(user->buf) - len,
897 &r->text_buf[0], r->info->text_len,
900 atomic64_set(&user->seq, r->info->seq + 1);
907 if (copy_to_user(buf, user->buf, len)) {
913 mutex_unlock(&user->lock);
918 * Be careful when modifying this function!!!
920 * Only few operations are supported because the device works only with the
921 * entire variable length messages (records). Non-standard values are
922 * returned in the other cases and has been this way for quite some time.
923 * User space applications might depend on this behavior.
925 static loff_t devkmsg_llseek(struct file *file, loff_t offset, int whence)
927 struct devkmsg_user *user = file->private_data;
937 /* the first record */
938 atomic64_set(&user->seq, prb_first_valid_seq(prb));
942 * The first record after the last SYSLOG_ACTION_CLEAR,
943 * like issued by 'dmesg -c'. Reading /dev/kmsg itself
944 * changes no global state, and does not clear anything.
946 atomic64_set(&user->seq, latched_seq_read_nolock(&clear_seq));
949 /* after the last record */
950 atomic64_set(&user->seq, prb_next_seq(prb));
958 static __poll_t devkmsg_poll(struct file *file, poll_table *wait)
960 struct devkmsg_user *user = file->private_data;
961 struct printk_info info;
965 return EPOLLERR|EPOLLNVAL;
967 poll_wait(file, &log_wait, wait);
969 if (prb_read_valid_info(prb, atomic64_read(&user->seq), &info, NULL)) {
970 /* return error when data has vanished underneath us */
971 if (info.seq != atomic64_read(&user->seq))
972 ret = EPOLLIN|EPOLLRDNORM|EPOLLERR|EPOLLPRI;
974 ret = EPOLLIN|EPOLLRDNORM;
980 static int devkmsg_open(struct inode *inode, struct file *file)
982 struct devkmsg_user *user;
985 if (devkmsg_log & DEVKMSG_LOG_MASK_OFF)
988 /* write-only does not need any file context */
989 if ((file->f_flags & O_ACCMODE) != O_WRONLY) {
990 err = check_syslog_permissions(SYSLOG_ACTION_READ_ALL,
996 user = kvmalloc(sizeof(struct devkmsg_user), GFP_KERNEL);
1000 ratelimit_default_init(&user->rs);
1001 ratelimit_set_flags(&user->rs, RATELIMIT_MSG_ON_RELEASE);
1003 mutex_init(&user->lock);
1005 prb_rec_init_rd(&user->record, &user->info,
1006 &user->text_buf[0], sizeof(user->text_buf));
1008 atomic64_set(&user->seq, prb_first_valid_seq(prb));
1010 file->private_data = user;
1014 static int devkmsg_release(struct inode *inode, struct file *file)
1016 struct devkmsg_user *user = file->private_data;
1021 ratelimit_state_exit(&user->rs);
1023 mutex_destroy(&user->lock);
1028 const struct file_operations kmsg_fops = {
1029 .open = devkmsg_open,
1030 .read = devkmsg_read,
1031 .write_iter = devkmsg_write,
1032 .llseek = devkmsg_llseek,
1033 .poll = devkmsg_poll,
1034 .release = devkmsg_release,
1037 #ifdef CONFIG_CRASH_CORE
1039 * This appends the listed symbols to /proc/vmcore
1041 * /proc/vmcore is used by various utilities, like crash and makedumpfile to
1042 * obtain access to symbols that are otherwise very difficult to locate. These
1043 * symbols are specifically used so that utilities can access and extract the
1044 * dmesg log from a vmcore file after a crash.
1046 void log_buf_vmcoreinfo_setup(void)
1048 struct dev_printk_info *dev_info = NULL;
1050 VMCOREINFO_SYMBOL(prb);
1051 VMCOREINFO_SYMBOL(printk_rb_static);
1052 VMCOREINFO_SYMBOL(clear_seq);
1055 * Export struct size and field offsets. User space tools can
1056 * parse it and detect any changes to structure down the line.
1059 VMCOREINFO_STRUCT_SIZE(printk_ringbuffer);
1060 VMCOREINFO_OFFSET(printk_ringbuffer, desc_ring);
1061 VMCOREINFO_OFFSET(printk_ringbuffer, text_data_ring);
1062 VMCOREINFO_OFFSET(printk_ringbuffer, fail);
1064 VMCOREINFO_STRUCT_SIZE(prb_desc_ring);
1065 VMCOREINFO_OFFSET(prb_desc_ring, count_bits);
1066 VMCOREINFO_OFFSET(prb_desc_ring, descs);
1067 VMCOREINFO_OFFSET(prb_desc_ring, infos);
1068 VMCOREINFO_OFFSET(prb_desc_ring, head_id);
1069 VMCOREINFO_OFFSET(prb_desc_ring, tail_id);
1071 VMCOREINFO_STRUCT_SIZE(prb_desc);
1072 VMCOREINFO_OFFSET(prb_desc, state_var);
1073 VMCOREINFO_OFFSET(prb_desc, text_blk_lpos);
1075 VMCOREINFO_STRUCT_SIZE(prb_data_blk_lpos);
1076 VMCOREINFO_OFFSET(prb_data_blk_lpos, begin);
1077 VMCOREINFO_OFFSET(prb_data_blk_lpos, next);
1079 VMCOREINFO_STRUCT_SIZE(printk_info);
1080 VMCOREINFO_OFFSET(printk_info, seq);
1081 VMCOREINFO_OFFSET(printk_info, ts_nsec);
1082 VMCOREINFO_OFFSET(printk_info, text_len);
1083 VMCOREINFO_OFFSET(printk_info, caller_id);
1084 VMCOREINFO_OFFSET(printk_info, dev_info);
1086 VMCOREINFO_STRUCT_SIZE(dev_printk_info);
1087 VMCOREINFO_OFFSET(dev_printk_info, subsystem);
1088 VMCOREINFO_LENGTH(printk_info_subsystem, sizeof(dev_info->subsystem));
1089 VMCOREINFO_OFFSET(dev_printk_info, device);
1090 VMCOREINFO_LENGTH(printk_info_device, sizeof(dev_info->device));
1092 VMCOREINFO_STRUCT_SIZE(prb_data_ring);
1093 VMCOREINFO_OFFSET(prb_data_ring, size_bits);
1094 VMCOREINFO_OFFSET(prb_data_ring, data);
1095 VMCOREINFO_OFFSET(prb_data_ring, head_lpos);
1096 VMCOREINFO_OFFSET(prb_data_ring, tail_lpos);
1098 VMCOREINFO_SIZE(atomic_long_t);
1099 VMCOREINFO_TYPE_OFFSET(atomic_long_t, counter);
1101 VMCOREINFO_STRUCT_SIZE(latched_seq);
1102 VMCOREINFO_OFFSET(latched_seq, val);
1106 /* requested log_buf_len from kernel cmdline */
1107 static unsigned long __initdata new_log_buf_len;
1109 /* we practice scaling the ring buffer by powers of 2 */
1110 static void __init log_buf_len_update(u64 size)
1112 if (size > (u64)LOG_BUF_LEN_MAX) {
1113 size = (u64)LOG_BUF_LEN_MAX;
1114 pr_err("log_buf over 2G is not supported.\n");
1118 size = roundup_pow_of_two(size);
1119 if (size > log_buf_len)
1120 new_log_buf_len = (unsigned long)size;
1123 /* save requested log_buf_len since it's too early to process it */
1124 static int __init log_buf_len_setup(char *str)
1131 size = memparse(str, &str);
1133 log_buf_len_update(size);
1137 early_param("log_buf_len", log_buf_len_setup);
1140 #define __LOG_CPU_MAX_BUF_LEN (1 << CONFIG_LOG_CPU_MAX_BUF_SHIFT)
1142 static void __init log_buf_add_cpu(void)
1144 unsigned int cpu_extra;
1147 * archs should set up cpu_possible_bits properly with
1148 * set_cpu_possible() after setup_arch() but just in
1149 * case lets ensure this is valid.
1151 if (num_possible_cpus() == 1)
1154 cpu_extra = (num_possible_cpus() - 1) * __LOG_CPU_MAX_BUF_LEN;
1156 /* by default this will only continue through for large > 64 CPUs */
1157 if (cpu_extra <= __LOG_BUF_LEN / 2)
1160 pr_info("log_buf_len individual max cpu contribution: %d bytes\n",
1161 __LOG_CPU_MAX_BUF_LEN);
1162 pr_info("log_buf_len total cpu_extra contributions: %d bytes\n",
1164 pr_info("log_buf_len min size: %d bytes\n", __LOG_BUF_LEN);
1166 log_buf_len_update(cpu_extra + __LOG_BUF_LEN);
1168 #else /* !CONFIG_SMP */
1169 static inline void log_buf_add_cpu(void) {}
1170 #endif /* CONFIG_SMP */
1172 static void __init set_percpu_data_ready(void)
1174 __printk_percpu_data_ready = true;
1177 static unsigned int __init add_to_rb(struct printk_ringbuffer *rb,
1178 struct printk_record *r)
1180 struct prb_reserved_entry e;
1181 struct printk_record dest_r;
1183 prb_rec_init_wr(&dest_r, r->info->text_len);
1185 if (!prb_reserve(&e, rb, &dest_r))
1188 memcpy(&dest_r.text_buf[0], &r->text_buf[0], r->info->text_len);
1189 dest_r.info->text_len = r->info->text_len;
1190 dest_r.info->facility = r->info->facility;
1191 dest_r.info->level = r->info->level;
1192 dest_r.info->flags = r->info->flags;
1193 dest_r.info->ts_nsec = r->info->ts_nsec;
1194 dest_r.info->caller_id = r->info->caller_id;
1195 memcpy(&dest_r.info->dev_info, &r->info->dev_info, sizeof(dest_r.info->dev_info));
1197 prb_final_commit(&e);
1199 return prb_record_text_space(&e);
1202 static char setup_text_buf[LOG_LINE_MAX] __initdata;
1204 void __init setup_log_buf(int early)
1206 struct printk_info *new_infos;
1207 unsigned int new_descs_count;
1208 struct prb_desc *new_descs;
1209 struct printk_info info;
1210 struct printk_record r;
1211 unsigned int text_size;
1212 size_t new_descs_size;
1213 size_t new_infos_size;
1214 unsigned long flags;
1220 * Some archs call setup_log_buf() multiple times - first is very
1221 * early, e.g. from setup_arch(), and second - when percpu_areas
1225 set_percpu_data_ready();
1227 if (log_buf != __log_buf)
1230 if (!early && !new_log_buf_len)
1233 if (!new_log_buf_len)
1236 new_descs_count = new_log_buf_len >> PRB_AVGBITS;
1237 if (new_descs_count == 0) {
1238 pr_err("new_log_buf_len: %lu too small\n", new_log_buf_len);
1242 new_log_buf = memblock_alloc(new_log_buf_len, LOG_ALIGN);
1243 if (unlikely(!new_log_buf)) {
1244 pr_err("log_buf_len: %lu text bytes not available\n",
1249 new_descs_size = new_descs_count * sizeof(struct prb_desc);
1250 new_descs = memblock_alloc(new_descs_size, LOG_ALIGN);
1251 if (unlikely(!new_descs)) {
1252 pr_err("log_buf_len: %zu desc bytes not available\n",
1254 goto err_free_log_buf;
1257 new_infos_size = new_descs_count * sizeof(struct printk_info);
1258 new_infos = memblock_alloc(new_infos_size, LOG_ALIGN);
1259 if (unlikely(!new_infos)) {
1260 pr_err("log_buf_len: %zu info bytes not available\n",
1262 goto err_free_descs;
1265 prb_rec_init_rd(&r, &info, &setup_text_buf[0], sizeof(setup_text_buf));
1267 prb_init(&printk_rb_dynamic,
1268 new_log_buf, ilog2(new_log_buf_len),
1269 new_descs, ilog2(new_descs_count),
1272 local_irq_save(flags);
1274 log_buf_len = new_log_buf_len;
1275 log_buf = new_log_buf;
1276 new_log_buf_len = 0;
1278 free = __LOG_BUF_LEN;
1279 prb_for_each_record(0, &printk_rb_static, seq, &r) {
1280 text_size = add_to_rb(&printk_rb_dynamic, &r);
1281 if (text_size > free)
1287 prb = &printk_rb_dynamic;
1289 local_irq_restore(flags);
1292 * Copy any remaining messages that might have appeared from
1293 * NMI context after copying but before switching to the
1296 prb_for_each_record(seq, &printk_rb_static, seq, &r) {
1297 text_size = add_to_rb(&printk_rb_dynamic, &r);
1298 if (text_size > free)
1304 if (seq != prb_next_seq(&printk_rb_static)) {
1305 pr_err("dropped %llu messages\n",
1306 prb_next_seq(&printk_rb_static) - seq);
1309 pr_info("log_buf_len: %u bytes\n", log_buf_len);
1310 pr_info("early log buf free: %u(%u%%)\n",
1311 free, (free * 100) / __LOG_BUF_LEN);
1315 memblock_free(new_descs, new_descs_size);
1317 memblock_free(new_log_buf, new_log_buf_len);
1320 static bool __read_mostly ignore_loglevel;
1322 static int __init ignore_loglevel_setup(char *str)
1324 ignore_loglevel = true;
1325 pr_info("debug: ignoring loglevel setting.\n");
1330 early_param("ignore_loglevel", ignore_loglevel_setup);
1331 module_param(ignore_loglevel, bool, S_IRUGO | S_IWUSR);
1332 MODULE_PARM_DESC(ignore_loglevel,
1333 "ignore loglevel setting (prints all kernel messages to the console)");
1335 static bool suppress_message_printing(int level)
1337 return (level >= console_loglevel && !ignore_loglevel);
1340 #ifdef CONFIG_BOOT_PRINTK_DELAY
1342 static int boot_delay; /* msecs delay after each printk during bootup */
1343 static unsigned long long loops_per_msec; /* based on boot_delay */
1345 static int __init boot_delay_setup(char *str)
1349 lpj = preset_lpj ? preset_lpj : 1000000; /* some guess */
1350 loops_per_msec = (unsigned long long)lpj / 1000 * HZ;
1352 get_option(&str, &boot_delay);
1353 if (boot_delay > 10 * 1000)
1356 pr_debug("boot_delay: %u, preset_lpj: %ld, lpj: %lu, "
1357 "HZ: %d, loops_per_msec: %llu\n",
1358 boot_delay, preset_lpj, lpj, HZ, loops_per_msec);
1361 early_param("boot_delay", boot_delay_setup);
1363 static void boot_delay_msec(int level)
1365 unsigned long long k;
1366 unsigned long timeout;
1368 if ((boot_delay == 0 || system_state >= SYSTEM_RUNNING)
1369 || suppress_message_printing(level)) {
1373 k = (unsigned long long)loops_per_msec * boot_delay;
1375 timeout = jiffies + msecs_to_jiffies(boot_delay);
1380 * use (volatile) jiffies to prevent
1381 * compiler reduction; loop termination via jiffies
1382 * is secondary and may or may not happen.
1384 if (time_after(jiffies, timeout))
1386 touch_nmi_watchdog();
1390 static inline void boot_delay_msec(int level)
1395 static bool printk_time = IS_ENABLED(CONFIG_PRINTK_TIME);
1396 module_param_named(time, printk_time, bool, S_IRUGO | S_IWUSR);
1398 static size_t print_syslog(unsigned int level, char *buf)
1400 return sprintf(buf, "<%u>", level);
1403 static size_t print_time(u64 ts, char *buf)
1405 unsigned long rem_nsec = do_div(ts, 1000000000);
1407 return sprintf(buf, "[%5lu.%06lu]",
1408 (unsigned long)ts, rem_nsec / 1000);
1411 #ifdef CONFIG_PRINTK_CALLER
1412 static size_t print_caller(u32 id, char *buf)
1416 snprintf(caller, sizeof(caller), "%c%u",
1417 id & 0x80000000 ? 'C' : 'T', id & ~0x80000000);
1418 return sprintf(buf, "[%6s]", caller);
1421 #define print_caller(id, buf) 0
1424 static size_t info_print_prefix(const struct printk_info *info, bool syslog,
1425 bool time, char *buf)
1430 len = print_syslog((info->facility << 3) | info->level, buf);
1433 len += print_time(info->ts_nsec, buf + len);
1435 len += print_caller(info->caller_id, buf + len);
1437 if (IS_ENABLED(CONFIG_PRINTK_CALLER) || time) {
1446 * Prepare the record for printing. The text is shifted within the given
1447 * buffer to avoid a need for another one. The following operations are
1450 * - Add prefix for each line.
1451 * - Drop truncated lines that no longer fit into the buffer.
1452 * - Add the trailing newline that has been removed in vprintk_store().
1453 * - Add a string terminator.
1455 * Since the produced string is always terminated, the maximum possible
1456 * return value is @r->text_buf_size - 1;
1458 * Return: The length of the updated/prepared text, including the added
1459 * prefixes and the newline. The terminator is not counted. The dropped
1460 * line(s) are not counted.
1462 static size_t record_print_text(struct printk_record *r, bool syslog,
1465 size_t text_len = r->info->text_len;
1466 size_t buf_size = r->text_buf_size;
1467 char *text = r->text_buf;
1468 char prefix[PREFIX_MAX];
1469 bool truncated = false;
1476 * If the message was truncated because the buffer was not large
1477 * enough, treat the available text as if it were the full text.
1479 if (text_len > buf_size)
1480 text_len = buf_size;
1482 prefix_len = info_print_prefix(r->info, syslog, time, prefix);
1485 * @text_len: bytes of unprocessed text
1486 * @line_len: bytes of current line _without_ newline
1487 * @text: pointer to beginning of current line
1488 * @len: number of bytes prepared in r->text_buf
1491 next = memchr(text, '\n', text_len);
1493 line_len = next - text;
1495 /* Drop truncated line(s). */
1498 line_len = text_len;
1502 * Truncate the text if there is not enough space to add the
1503 * prefix and a trailing newline and a terminator.
1505 if (len + prefix_len + text_len + 1 + 1 > buf_size) {
1506 /* Drop even the current line if no space. */
1507 if (len + prefix_len + line_len + 1 + 1 > buf_size)
1510 text_len = buf_size - len - prefix_len - 1 - 1;
1514 memmove(text + prefix_len, text, text_len);
1515 memcpy(text, prefix, prefix_len);
1518 * Increment the prepared length to include the text and
1519 * prefix that were just moved+copied. Also increment for the
1520 * newline at the end of this line. If this is the last line,
1521 * there is no newline, but it will be added immediately below.
1523 len += prefix_len + line_len + 1;
1524 if (text_len == line_len) {
1526 * This is the last line. Add the trailing newline
1527 * removed in vprintk_store().
1529 text[prefix_len + line_len] = '\n';
1534 * Advance beyond the added prefix and the related line with
1537 text += prefix_len + line_len + 1;
1540 * The remaining text has only decreased by the line with its
1543 * Note that @text_len can become zero. It happens when @text
1544 * ended with a newline (either due to truncation or the
1545 * original string ending with "\n\n"). The loop is correctly
1546 * repeated and (if not truncated) an empty line with a prefix
1549 text_len -= line_len + 1;
1553 * If a buffer was provided, it will be terminated. Space for the
1554 * string terminator is guaranteed to be available. The terminator is
1555 * not counted in the return value.
1558 r->text_buf[len] = 0;
1563 static size_t get_record_print_text_size(struct printk_info *info,
1564 unsigned int line_count,
1565 bool syslog, bool time)
1567 char prefix[PREFIX_MAX];
1570 prefix_len = info_print_prefix(info, syslog, time, prefix);
1573 * Each line will be preceded with a prefix. The intermediate
1574 * newlines are already within the text, but a final trailing
1575 * newline will be added.
1577 return ((prefix_len * line_count) + info->text_len + 1);
1581 * Beginning with @start_seq, find the first record where it and all following
1582 * records up to (but not including) @max_seq fit into @size.
1584 * @max_seq is simply an upper bound and does not need to exist. If the caller
1585 * does not require an upper bound, -1 can be used for @max_seq.
1587 static u64 find_first_fitting_seq(u64 start_seq, u64 max_seq, size_t size,
1588 bool syslog, bool time)
1590 struct printk_info info;
1591 unsigned int line_count;
1595 /* Determine the size of the records up to @max_seq. */
1596 prb_for_each_info(start_seq, prb, seq, &info, &line_count) {
1597 if (info.seq >= max_seq)
1599 len += get_record_print_text_size(&info, line_count, syslog, time);
1603 * Adjust the upper bound for the next loop to avoid subtracting
1604 * lengths that were never added.
1610 * Move first record forward until length fits into the buffer. Ignore
1611 * newest messages that were not counted in the above cycle. Messages
1612 * might appear and get lost in the meantime. This is a best effort
1613 * that prevents an infinite loop that could occur with a retry.
1615 prb_for_each_info(start_seq, prb, seq, &info, &line_count) {
1616 if (len <= size || info.seq >= max_seq)
1618 len -= get_record_print_text_size(&info, line_count, syslog, time);
1624 /* The caller is responsible for making sure @size is greater than 0. */
1625 static int syslog_print(char __user *buf, int size)
1627 struct printk_info info;
1628 struct printk_record r;
1633 text = kmalloc(CONSOLE_LOG_MAX, GFP_KERNEL);
1637 prb_rec_init_rd(&r, &info, text, CONSOLE_LOG_MAX);
1639 mutex_lock(&syslog_lock);
1642 * Wait for the @syslog_seq record to be available. @syslog_seq may
1643 * change while waiting.
1648 mutex_unlock(&syslog_lock);
1650 * Guarantee this task is visible on the waitqueue before
1651 * checking the wake condition.
1653 * The full memory barrier within set_current_state() of
1654 * prepare_to_wait_event() pairs with the full memory barrier
1655 * within wq_has_sleeper().
1657 * This pairs with __wake_up_klogd:A.
1659 len = wait_event_interruptible(log_wait,
1660 prb_read_valid(prb, seq, NULL)); /* LMM(syslog_print:A) */
1661 mutex_lock(&syslog_lock);
1665 } while (syslog_seq != seq);
1668 * Copy records that fit into the buffer. The above cycle makes sure
1669 * that the first record is always available.
1676 if (!prb_read_valid(prb, syslog_seq, &r))
1679 if (r.info->seq != syslog_seq) {
1680 /* message is gone, move to next valid one */
1681 syslog_seq = r.info->seq;
1686 * To keep reading/counting partial line consistent,
1687 * use printk_time value as of the beginning of a line.
1689 if (!syslog_partial)
1690 syslog_time = printk_time;
1692 skip = syslog_partial;
1693 n = record_print_text(&r, true, syslog_time);
1694 if (n - syslog_partial <= size) {
1695 /* message fits into buffer, move forward */
1696 syslog_seq = r.info->seq + 1;
1697 n -= syslog_partial;
1700 /* partial read(), remember position */
1702 syslog_partial += n;
1709 mutex_unlock(&syslog_lock);
1710 err = copy_to_user(buf, text + skip, n);
1711 mutex_lock(&syslog_lock);
1724 mutex_unlock(&syslog_lock);
1729 static int syslog_print_all(char __user *buf, int size, bool clear)
1731 struct printk_info info;
1732 struct printk_record r;
1738 text = kmalloc(CONSOLE_LOG_MAX, GFP_KERNEL);
1744 * Find first record that fits, including all following records,
1745 * into the user-provided buffer for this dump.
1747 seq = find_first_fitting_seq(latched_seq_read_nolock(&clear_seq), -1,
1750 prb_rec_init_rd(&r, &info, text, CONSOLE_LOG_MAX);
1753 prb_for_each_record(seq, prb, seq, &r) {
1756 textlen = record_print_text(&r, true, time);
1758 if (len + textlen > size) {
1763 if (copy_to_user(buf + len, text, textlen))
1773 mutex_lock(&syslog_lock);
1774 latched_seq_write(&clear_seq, seq);
1775 mutex_unlock(&syslog_lock);
1782 static void syslog_clear(void)
1784 mutex_lock(&syslog_lock);
1785 latched_seq_write(&clear_seq, prb_next_seq(prb));
1786 mutex_unlock(&syslog_lock);
1789 int do_syslog(int type, char __user *buf, int len, int source)
1791 struct printk_info info;
1793 static int saved_console_loglevel = LOGLEVEL_DEFAULT;
1796 error = check_syslog_permissions(type, source);
1801 case SYSLOG_ACTION_CLOSE: /* Close log */
1803 case SYSLOG_ACTION_OPEN: /* Open log */
1805 case SYSLOG_ACTION_READ: /* Read from log */
1806 if (!buf || len < 0)
1810 if (!access_ok(buf, len))
1812 error = syslog_print(buf, len);
1814 /* Read/clear last kernel messages */
1815 case SYSLOG_ACTION_READ_CLEAR:
1818 /* Read last kernel messages */
1819 case SYSLOG_ACTION_READ_ALL:
1820 if (!buf || len < 0)
1824 if (!access_ok(buf, len))
1826 error = syslog_print_all(buf, len, clear);
1828 /* Clear ring buffer */
1829 case SYSLOG_ACTION_CLEAR:
1832 /* Disable logging to console */
1833 case SYSLOG_ACTION_CONSOLE_OFF:
1834 if (saved_console_loglevel == LOGLEVEL_DEFAULT)
1835 saved_console_loglevel = console_loglevel;
1836 console_loglevel = minimum_console_loglevel;
1838 /* Enable logging to console */
1839 case SYSLOG_ACTION_CONSOLE_ON:
1840 if (saved_console_loglevel != LOGLEVEL_DEFAULT) {
1841 console_loglevel = saved_console_loglevel;
1842 saved_console_loglevel = LOGLEVEL_DEFAULT;
1845 /* Set level of messages printed to console */
1846 case SYSLOG_ACTION_CONSOLE_LEVEL:
1847 if (len < 1 || len > 8)
1849 if (len < minimum_console_loglevel)
1850 len = minimum_console_loglevel;
1851 console_loglevel = len;
1852 /* Implicitly re-enable logging to console */
1853 saved_console_loglevel = LOGLEVEL_DEFAULT;
1855 /* Number of chars in the log buffer */
1856 case SYSLOG_ACTION_SIZE_UNREAD:
1857 mutex_lock(&syslog_lock);
1858 if (!prb_read_valid_info(prb, syslog_seq, &info, NULL)) {
1859 /* No unread messages. */
1860 mutex_unlock(&syslog_lock);
1863 if (info.seq != syslog_seq) {
1864 /* messages are gone, move to first one */
1865 syslog_seq = info.seq;
1868 if (source == SYSLOG_FROM_PROC) {
1870 * Short-cut for poll(/"proc/kmsg") which simply checks
1871 * for pending data, not the size; return the count of
1872 * records, not the length.
1874 error = prb_next_seq(prb) - syslog_seq;
1876 bool time = syslog_partial ? syslog_time : printk_time;
1877 unsigned int line_count;
1880 prb_for_each_info(syslog_seq, prb, seq, &info,
1882 error += get_record_print_text_size(&info, line_count,
1886 error -= syslog_partial;
1888 mutex_unlock(&syslog_lock);
1890 /* Size of the log buffer */
1891 case SYSLOG_ACTION_SIZE_BUFFER:
1892 error = log_buf_len;
1902 SYSCALL_DEFINE3(syslog, int, type, char __user *, buf, int, len)
1904 return do_syslog(type, buf, len, SYSLOG_FROM_READER);
1908 * Special console_lock variants that help to reduce the risk of soft-lockups.
1909 * They allow to pass console_lock to another printk() call using a busy wait.
1912 #ifdef CONFIG_LOCKDEP
1913 static struct lockdep_map console_owner_dep_map = {
1914 .name = "console_owner"
1918 static DEFINE_RAW_SPINLOCK(console_owner_lock);
1919 static struct task_struct *console_owner;
1920 static bool console_waiter;
1923 * console_lock_spinning_enable - mark beginning of code where another
1924 * thread might safely busy wait
1926 * This basically converts console_lock into a spinlock. This marks
1927 * the section where the console_lock owner can not sleep, because
1928 * there may be a waiter spinning (like a spinlock). Also it must be
1929 * ready to hand over the lock at the end of the section.
1931 static void console_lock_spinning_enable(void)
1933 raw_spin_lock(&console_owner_lock);
1934 console_owner = current;
1935 raw_spin_unlock(&console_owner_lock);
1937 /* The waiter may spin on us after setting console_owner */
1938 spin_acquire(&console_owner_dep_map, 0, 0, _THIS_IP_);
1942 * console_lock_spinning_disable_and_check - mark end of code where another
1943 * thread was able to busy wait and check if there is a waiter
1945 * This is called at the end of the section where spinning is allowed.
1946 * It has two functions. First, it is a signal that it is no longer
1947 * safe to start busy waiting for the lock. Second, it checks if
1948 * there is a busy waiter and passes the lock rights to her.
1950 * Important: Callers lose the lock if there was a busy waiter.
1951 * They must not touch items synchronized by console_lock
1954 * Return: 1 if the lock rights were passed, 0 otherwise.
1956 static int console_lock_spinning_disable_and_check(void)
1960 raw_spin_lock(&console_owner_lock);
1961 waiter = READ_ONCE(console_waiter);
1962 console_owner = NULL;
1963 raw_spin_unlock(&console_owner_lock);
1966 spin_release(&console_owner_dep_map, _THIS_IP_);
1970 /* The waiter is now free to continue */
1971 WRITE_ONCE(console_waiter, false);
1973 spin_release(&console_owner_dep_map, _THIS_IP_);
1976 * Hand off console_lock to waiter. The waiter will perform
1977 * the up(). After this, the waiter is the console_lock owner.
1979 mutex_release(&console_lock_dep_map, _THIS_IP_);
1984 * console_trylock_spinning - try to get console_lock by busy waiting
1986 * This allows to busy wait for the console_lock when the current
1987 * owner is running in specially marked sections. It means that
1988 * the current owner is running and cannot reschedule until it
1989 * is ready to lose the lock.
1991 * Return: 1 if we got the lock, 0 othrewise
1993 static int console_trylock_spinning(void)
1995 struct task_struct *owner = NULL;
1998 unsigned long flags;
2000 if (console_trylock())
2004 * It's unsafe to spin once a panic has begun. If we are the
2005 * panic CPU, we may have already halted the owner of the
2006 * console_sem. If we are not the panic CPU, then we should
2007 * avoid taking console_sem, so the panic CPU has a better
2008 * chance of cleanly acquiring it later.
2010 if (panic_in_progress())
2013 printk_safe_enter_irqsave(flags);
2015 raw_spin_lock(&console_owner_lock);
2016 owner = READ_ONCE(console_owner);
2017 waiter = READ_ONCE(console_waiter);
2018 if (!waiter && owner && owner != current) {
2019 WRITE_ONCE(console_waiter, true);
2022 raw_spin_unlock(&console_owner_lock);
2025 * If there is an active printk() writing to the
2026 * consoles, instead of having it write our data too,
2027 * see if we can offload that load from the active
2028 * printer, and do some printing ourselves.
2029 * Go into a spin only if there isn't already a waiter
2030 * spinning, and there is an active printer, and
2031 * that active printer isn't us (recursive printk?).
2034 printk_safe_exit_irqrestore(flags);
2038 /* We spin waiting for the owner to release us */
2039 spin_acquire(&console_owner_dep_map, 0, 0, _THIS_IP_);
2040 /* Owner will clear console_waiter on hand off */
2041 while (READ_ONCE(console_waiter))
2043 spin_release(&console_owner_dep_map, _THIS_IP_);
2045 printk_safe_exit_irqrestore(flags);
2047 * The owner passed the console lock to us.
2048 * Since we did not spin on console lock, annotate
2049 * this as a trylock. Otherwise lockdep will
2052 mutex_acquire(&console_lock_dep_map, 0, 1, _THIS_IP_);
2058 * Call the specified console driver, asking it to write out the specified
2059 * text and length. If @dropped_text is non-NULL and any records have been
2060 * dropped, a dropped message will be written out first.
2062 static void call_console_driver(struct console *con, const char *text, size_t len,
2067 if (con->dropped && dropped_text) {
2068 dropped_len = snprintf(dropped_text, DROPPED_TEXT_MAX,
2069 "** %lu printk messages dropped **\n",
2072 con->write(con, dropped_text, dropped_len);
2075 con->write(con, text, len);
2079 * Recursion is tracked separately on each CPU. If NMIs are supported, an
2080 * additional NMI context per CPU is also separately tracked. Until per-CPU
2081 * is available, a separate "early tracking" is performed.
2083 static DEFINE_PER_CPU(u8, printk_count);
2084 static u8 printk_count_early;
2085 #ifdef CONFIG_HAVE_NMI
2086 static DEFINE_PER_CPU(u8, printk_count_nmi);
2087 static u8 printk_count_nmi_early;
2091 * Recursion is limited to keep the output sane. printk() should not require
2092 * more than 1 level of recursion (allowing, for example, printk() to trigger
2093 * a WARN), but a higher value is used in case some printk-internal errors
2094 * exist, such as the ringbuffer validation checks failing.
2096 #define PRINTK_MAX_RECURSION 3
2099 * Return a pointer to the dedicated counter for the CPU+context of the
2102 static u8 *__printk_recursion_counter(void)
2104 #ifdef CONFIG_HAVE_NMI
2106 if (printk_percpu_data_ready())
2107 return this_cpu_ptr(&printk_count_nmi);
2108 return &printk_count_nmi_early;
2111 if (printk_percpu_data_ready())
2112 return this_cpu_ptr(&printk_count);
2113 return &printk_count_early;
2117 * Enter recursion tracking. Interrupts are disabled to simplify tracking.
2118 * The caller must check the boolean return value to see if the recursion is
2119 * allowed. On failure, interrupts are not disabled.
2121 * @recursion_ptr must be a variable of type (u8 *) and is the same variable
2122 * that is passed to printk_exit_irqrestore().
2124 #define printk_enter_irqsave(recursion_ptr, flags) \
2126 bool success = true; \
2128 typecheck(u8 *, recursion_ptr); \
2129 local_irq_save(flags); \
2130 (recursion_ptr) = __printk_recursion_counter(); \
2131 if (*(recursion_ptr) > PRINTK_MAX_RECURSION) { \
2132 local_irq_restore(flags); \
2135 (*(recursion_ptr))++; \
2140 /* Exit recursion tracking, restoring interrupts. */
2141 #define printk_exit_irqrestore(recursion_ptr, flags) \
2143 typecheck(u8 *, recursion_ptr); \
2144 (*(recursion_ptr))--; \
2145 local_irq_restore(flags); \
2148 int printk_delay_msec __read_mostly;
2150 static inline void printk_delay(int level)
2152 boot_delay_msec(level);
2154 if (unlikely(printk_delay_msec)) {
2155 int m = printk_delay_msec;
2159 touch_nmi_watchdog();
2164 static inline u32 printk_caller_id(void)
2166 return in_task() ? task_pid_nr(current) :
2167 0x80000000 + smp_processor_id();
2171 * printk_parse_prefix - Parse level and control flags.
2173 * @text: The terminated text message.
2174 * @level: A pointer to the current level value, will be updated.
2175 * @flags: A pointer to the current printk_info flags, will be updated.
2177 * @level may be NULL if the caller is not interested in the parsed value.
2178 * Otherwise the variable pointed to by @level must be set to
2179 * LOGLEVEL_DEFAULT in order to be updated with the parsed value.
2181 * @flags may be NULL if the caller is not interested in the parsed value.
2182 * Otherwise the variable pointed to by @flags will be OR'd with the parsed
2185 * Return: The length of the parsed level and control flags.
2187 u16 printk_parse_prefix(const char *text, int *level,
2188 enum printk_info_flags *flags)
2194 kern_level = printk_get_level(text);
2198 switch (kern_level) {
2200 if (level && *level == LOGLEVEL_DEFAULT)
2201 *level = kern_level - '0';
2203 case 'c': /* KERN_CONT */
2216 static u16 printk_sprint(char *text, u16 size, int facility,
2217 enum printk_info_flags *flags, const char *fmt,
2222 text_len = vscnprintf(text, size, fmt, args);
2224 /* Mark and strip a trailing newline. */
2225 if (text_len && text[text_len - 1] == '\n') {
2227 *flags |= LOG_NEWLINE;
2230 /* Strip log level and control flags. */
2231 if (facility == 0) {
2234 prefix_len = printk_parse_prefix(text, NULL, NULL);
2236 text_len -= prefix_len;
2237 memmove(text, text + prefix_len, text_len);
2241 trace_console_rcuidle(text, text_len);
2247 int vprintk_store(int facility, int level,
2248 const struct dev_printk_info *dev_info,
2249 const char *fmt, va_list args)
2251 struct prb_reserved_entry e;
2252 enum printk_info_flags flags = 0;
2253 struct printk_record r;
2254 unsigned long irqflags;
2255 u16 trunc_msg_len = 0;
2265 if (!printk_enter_irqsave(recursion_ptr, irqflags))
2269 * Since the duration of printk() can vary depending on the message
2270 * and state of the ringbuffer, grab the timestamp now so that it is
2271 * close to the call of printk(). This provides a more deterministic
2272 * timestamp with respect to the caller.
2274 ts_nsec = local_clock();
2276 caller_id = printk_caller_id();
2279 * The sprintf needs to come first since the syslog prefix might be
2280 * passed in as a parameter. An extra byte must be reserved so that
2281 * later the vscnprintf() into the reserved buffer has room for the
2282 * terminating '\0', which is not counted by vsnprintf().
2284 va_copy(args2, args);
2285 reserve_size = vsnprintf(&prefix_buf[0], sizeof(prefix_buf), fmt, args2) + 1;
2288 if (reserve_size > LOG_LINE_MAX)
2289 reserve_size = LOG_LINE_MAX;
2291 /* Extract log level or control flags. */
2293 printk_parse_prefix(&prefix_buf[0], &level, &flags);
2295 if (level == LOGLEVEL_DEFAULT)
2296 level = default_message_loglevel;
2299 flags |= LOG_NEWLINE;
2301 if (flags & LOG_CONT) {
2302 prb_rec_init_wr(&r, reserve_size);
2303 if (prb_reserve_in_last(&e, prb, &r, caller_id, LOG_LINE_MAX)) {
2304 text_len = printk_sprint(&r.text_buf[r.info->text_len], reserve_size,
2305 facility, &flags, fmt, args);
2306 r.info->text_len += text_len;
2308 if (flags & LOG_NEWLINE) {
2309 r.info->flags |= LOG_NEWLINE;
2310 prb_final_commit(&e);
2321 * Explicitly initialize the record before every prb_reserve() call.
2322 * prb_reserve_in_last() and prb_reserve() purposely invalidate the
2323 * structure when they fail.
2325 prb_rec_init_wr(&r, reserve_size);
2326 if (!prb_reserve(&e, prb, &r)) {
2327 /* truncate the message if it is too long for empty buffer */
2328 truncate_msg(&reserve_size, &trunc_msg_len);
2330 prb_rec_init_wr(&r, reserve_size + trunc_msg_len);
2331 if (!prb_reserve(&e, prb, &r))
2336 text_len = printk_sprint(&r.text_buf[0], reserve_size, facility, &flags, fmt, args);
2338 memcpy(&r.text_buf[text_len], trunc_msg, trunc_msg_len);
2339 r.info->text_len = text_len + trunc_msg_len;
2340 r.info->facility = facility;
2341 r.info->level = level & 7;
2342 r.info->flags = flags & 0x1f;
2343 r.info->ts_nsec = ts_nsec;
2344 r.info->caller_id = caller_id;
2346 memcpy(&r.info->dev_info, dev_info, sizeof(r.info->dev_info));
2348 /* A message without a trailing newline can be continued. */
2349 if (!(flags & LOG_NEWLINE))
2352 prb_final_commit(&e);
2354 ret = text_len + trunc_msg_len;
2356 printk_exit_irqrestore(recursion_ptr, irqflags);
2360 asmlinkage int vprintk_emit(int facility, int level,
2361 const struct dev_printk_info *dev_info,
2362 const char *fmt, va_list args)
2365 bool in_sched = false;
2367 /* Suppress unimportant messages after panic happens */
2368 if (unlikely(suppress_printk))
2371 if (unlikely(suppress_panic_printk) &&
2372 atomic_read(&panic_cpu) != raw_smp_processor_id())
2375 if (level == LOGLEVEL_SCHED) {
2376 level = LOGLEVEL_DEFAULT;
2380 printk_delay(level);
2382 printed_len = vprintk_store(facility, level, dev_info, fmt, args);
2384 /* If called from the scheduler, we can not call up(). */
2385 if (!in_sched && allow_direct_printing()) {
2387 * The caller may be holding system-critical or
2388 * timing-sensitive locks. Disable preemption during direct
2389 * printing of all remaining records to all consoles so that
2390 * this context can return as soon as possible. Hopefully
2391 * another printk() caller will take over the printing.
2395 * Try to acquire and then immediately release the console
2396 * semaphore. The release will print out buffers. With the
2397 * spinning variant, this context tries to take over the
2398 * printing from another printing context.
2400 if (console_trylock_spinning())
2408 EXPORT_SYMBOL(vprintk_emit);
2410 int vprintk_default(const char *fmt, va_list args)
2412 return vprintk_emit(0, LOGLEVEL_DEFAULT, NULL, fmt, args);
2414 EXPORT_SYMBOL_GPL(vprintk_default);
2416 asmlinkage __visible int _printk(const char *fmt, ...)
2421 va_start(args, fmt);
2422 r = vprintk(fmt, args);
2427 EXPORT_SYMBOL(_printk);
2429 static bool __pr_flush(struct console *con, int timeout_ms, bool reset_on_progress);
2431 static void printk_start_kthread(struct console *con);
2433 #else /* CONFIG_PRINTK */
2435 #define CONSOLE_LOG_MAX 0
2436 #define DROPPED_TEXT_MAX 0
2437 #define printk_time false
2439 #define prb_read_valid(rb, seq, r) false
2440 #define prb_first_valid_seq(rb) 0
2441 #define prb_next_seq(rb) 0
2443 static u64 syslog_seq;
2445 static size_t record_print_text(const struct printk_record *r,
2446 bool syslog, bool time)
2450 static ssize_t info_print_ext_header(char *buf, size_t size,
2451 struct printk_info *info)
2455 static ssize_t msg_print_ext_body(char *buf, size_t size,
2456 char *text, size_t text_len,
2457 struct dev_printk_info *dev_info) { return 0; }
2458 static void console_lock_spinning_enable(void) { }
2459 static int console_lock_spinning_disable_and_check(void) { return 0; }
2460 static void call_console_driver(struct console *con, const char *text, size_t len,
2464 static bool suppress_message_printing(int level) { return false; }
2465 static bool __pr_flush(struct console *con, int timeout_ms, bool reset_on_progress) { return true; }
2466 static void printk_start_kthread(struct console *con) { }
2467 static bool allow_direct_printing(void) { return true; }
2469 #endif /* CONFIG_PRINTK */
2471 #ifdef CONFIG_EARLY_PRINTK
2472 struct console *early_console;
2474 asmlinkage __visible void early_printk(const char *fmt, ...)
2484 n = vscnprintf(buf, sizeof(buf), fmt, ap);
2487 early_console->write(early_console, buf, n);
2491 static void set_user_specified(struct console_cmdline *c, bool user_specified)
2493 if (!user_specified)
2497 * @c console was defined by the user on the command line.
2498 * Do not clear when added twice also by SPCR or the device tree.
2500 c->user_specified = true;
2501 /* At least one console defined by the user on the command line. */
2502 console_set_on_cmdline = 1;
2505 static int __add_preferred_console(char *name, int idx, char *options,
2506 char *brl_options, bool user_specified)
2508 struct console_cmdline *c;
2512 * See if this tty is not yet registered, and
2513 * if we have a slot free.
2515 for (i = 0, c = console_cmdline;
2516 i < MAX_CMDLINECONSOLES && c->name[0];
2518 if (strcmp(c->name, name) == 0 && c->index == idx) {
2520 preferred_console = i;
2521 set_user_specified(c, user_specified);
2525 if (i == MAX_CMDLINECONSOLES)
2528 preferred_console = i;
2529 strlcpy(c->name, name, sizeof(c->name));
2530 c->options = options;
2531 set_user_specified(c, user_specified);
2532 braille_set_options(c, brl_options);
2538 static int __init console_msg_format_setup(char *str)
2540 if (!strcmp(str, "syslog"))
2541 console_msg_format = MSG_FORMAT_SYSLOG;
2542 if (!strcmp(str, "default"))
2543 console_msg_format = MSG_FORMAT_DEFAULT;
2546 __setup("console_msg_format=", console_msg_format_setup);
2549 * Set up a console. Called via do_early_param() in init/main.c
2550 * for each "console=" parameter in the boot command line.
2552 static int __init console_setup(char *str)
2554 char buf[sizeof(console_cmdline[0].name) + 4]; /* 4 for "ttyS" */
2555 char *s, *options, *brl_options = NULL;
2559 * console="" or console=null have been suggested as a way to
2560 * disable console output. Use ttynull that has been created
2561 * for exactly this purpose.
2563 if (str[0] == 0 || strcmp(str, "null") == 0) {
2564 __add_preferred_console("ttynull", 0, NULL, NULL, true);
2568 if (_braille_console_setup(&str, &brl_options))
2572 * Decode str into name, index, options.
2574 if (str[0] >= '0' && str[0] <= '9') {
2575 strcpy(buf, "ttyS");
2576 strncpy(buf + 4, str, sizeof(buf) - 5);
2578 strncpy(buf, str, sizeof(buf) - 1);
2580 buf[sizeof(buf) - 1] = 0;
2581 options = strchr(str, ',');
2585 if (!strcmp(str, "ttya"))
2586 strcpy(buf, "ttyS0");
2587 if (!strcmp(str, "ttyb"))
2588 strcpy(buf, "ttyS1");
2590 for (s = buf; *s; s++)
2591 if (isdigit(*s) || *s == ',')
2593 idx = simple_strtoul(s, NULL, 10);
2596 __add_preferred_console(buf, idx, options, brl_options, true);
2599 __setup("console=", console_setup);
2602 * add_preferred_console - add a device to the list of preferred consoles.
2603 * @name: device name
2604 * @idx: device index
2605 * @options: options for this console
2607 * The last preferred console added will be used for kernel messages
2608 * and stdin/out/err for init. Normally this is used by console_setup
2609 * above to handle user-supplied console arguments; however it can also
2610 * be used by arch-specific code either to override the user or more
2611 * commonly to provide a default console (ie from PROM variables) when
2612 * the user has not supplied one.
2614 int add_preferred_console(char *name, int idx, char *options)
2616 return __add_preferred_console(name, idx, options, NULL, false);
2619 bool console_suspend_enabled = true;
2620 EXPORT_SYMBOL(console_suspend_enabled);
2622 static int __init console_suspend_disable(char *str)
2624 console_suspend_enabled = false;
2627 __setup("no_console_suspend", console_suspend_disable);
2628 module_param_named(console_suspend, console_suspend_enabled,
2629 bool, S_IRUGO | S_IWUSR);
2630 MODULE_PARM_DESC(console_suspend, "suspend console during suspend"
2631 " and hibernate operations");
2633 static bool printk_console_no_auto_verbose;
2635 void console_verbose(void)
2637 if (console_loglevel && !printk_console_no_auto_verbose)
2638 console_loglevel = CONSOLE_LOGLEVEL_MOTORMOUTH;
2640 EXPORT_SYMBOL_GPL(console_verbose);
2642 module_param_named(console_no_auto_verbose, printk_console_no_auto_verbose, bool, 0644);
2643 MODULE_PARM_DESC(console_no_auto_verbose, "Disable console loglevel raise to highest on oops/panic/etc");
2646 * suspend_console - suspend the console subsystem
2648 * This disables printk() while we go into suspend states
2650 void suspend_console(void)
2652 if (!console_suspend_enabled)
2654 pr_info("Suspending console(s) (use no_console_suspend to debug)\n");
2655 pr_flush(1000, true);
2657 console_suspended = 1;
2661 void resume_console(void)
2663 if (!console_suspend_enabled)
2666 console_suspended = 0;
2668 pr_flush(1000, true);
2672 * console_cpu_notify - print deferred console messages after CPU hotplug
2675 * If printk() is called from a CPU that is not online yet, the messages
2676 * will be printed on the console only if there are CON_ANYTIME consoles.
2677 * This function is called when a new CPU comes online (or fails to come
2678 * up) or goes offline.
2680 static int console_cpu_notify(unsigned int cpu)
2682 if (!cpuhp_tasks_frozen) {
2683 /* If trylock fails, someone else is doing the printing */
2684 if (console_trylock())
2688 * If a new CPU comes online, the conditions for
2689 * printer_should_wake() may have changed for some
2690 * kthread printer with !CON_ANYTIME.
2699 * console_lock - lock the console system for exclusive use.
2701 * Acquires a lock which guarantees that the caller has
2702 * exclusive access to the console system and the console_drivers list.
2704 * Can sleep, returns nothing.
2706 void console_lock(void)
2711 if (console_suspended)
2713 console_kthreads_block();
2714 console_may_schedule = 1;
2716 EXPORT_SYMBOL(console_lock);
2719 * console_trylock - try to lock the console system for exclusive use.
2721 * Try to acquire a lock which guarantees that the caller has exclusive
2722 * access to the console system and the console_drivers list.
2724 * returns 1 on success, and 0 on failure to acquire the lock.
2726 int console_trylock(void)
2728 if (down_trylock_console_sem())
2730 if (console_suspended) {
2734 if (!console_kthreads_atomic_tryblock()) {
2738 console_may_schedule = 0;
2741 EXPORT_SYMBOL(console_trylock);
2744 * This is used to help to make sure that certain paths within the VT code are
2745 * running with the console lock held. It is definitely not the perfect debug
2746 * tool (it is not known if the VT code is the task holding the console lock),
2747 * but it helps tracking those weird code paths in the console code such as
2748 * when the console is suspended: where the console is not locked but no
2749 * console printing may occur.
2751 * Note: This returns true when the console is suspended but is not locked.
2752 * This is intentional because the VT code must consider that situation
2753 * the same as if the console was locked.
2755 int is_console_locked(void)
2757 return (console_kthreads_blocked || atomic_read(&console_kthreads_active));
2759 EXPORT_SYMBOL(is_console_locked);
2762 * Return true when this CPU should unlock console_sem without pushing all
2763 * messages to the console. This reduces the chance that the console is
2764 * locked when the panic CPU tries to use it.
2766 static bool abandon_console_lock_in_panic(void)
2768 if (!panic_in_progress())
2772 * We can use raw_smp_processor_id() here because it is impossible for
2773 * the task to be migrated to the panic_cpu, or away from it. If
2774 * panic_cpu has already been set, and we're not currently executing on
2775 * that CPU, then we never will be.
2777 return atomic_read(&panic_cpu) != raw_smp_processor_id();
2780 static inline bool __console_is_usable(short flags)
2782 if (!(flags & CON_ENABLED))
2786 * Console drivers may assume that per-cpu resources have been
2787 * allocated. So unless they're explicitly marked as being able to
2788 * cope (CON_ANYTIME) don't call them until this CPU is officially up.
2790 if (!cpu_online(raw_smp_processor_id()) &&
2791 !(flags & CON_ANYTIME))
2798 * Check if the given console is currently capable and allowed to print
2801 * Requires holding the console_lock.
2803 static inline bool console_is_usable(struct console *con)
2808 return __console_is_usable(con->flags);
2811 static void __console_unlock(void)
2814 * Depending on whether console_lock() or console_trylock() was used,
2815 * appropriately allow the kthread printers to continue.
2817 if (console_kthreads_blocked)
2818 console_kthreads_unblock();
2820 console_kthreads_atomic_unblock();
2823 * New records may have arrived while the console was locked.
2824 * Wake the kthread printers to print them.
2832 * Print one record for the given console. The record printed is whatever
2833 * record is the next available record for the given console.
2835 * @text is a buffer of size CONSOLE_LOG_MAX.
2837 * If extended messages should be printed, @ext_text is a buffer of size
2838 * CONSOLE_EXT_LOG_MAX. Otherwise @ext_text must be NULL.
2840 * If dropped messages should be printed, @dropped_text is a buffer of size
2841 * DROPPED_TEXT_MAX. Otherwise @dropped_text must be NULL.
2843 * @handover will be set to true if a printk waiter has taken over the
2844 * console_lock, in which case the caller is no longer holding the
2845 * console_lock. Otherwise it is set to false. A NULL pointer may be provided
2846 * to disable allowing the console_lock to be taken over by a printk waiter.
2848 * Returns false if the given console has no next record to print, otherwise
2851 * Requires the console_lock if @handover is non-NULL.
2852 * Requires con->lock otherwise.
2854 static bool __console_emit_next_record(struct console *con, char *text, char *ext_text,
2855 char *dropped_text, bool *handover)
2857 static atomic_t panic_console_dropped = ATOMIC_INIT(0);
2858 struct printk_info info;
2859 struct printk_record r;
2860 unsigned long flags;
2864 prb_rec_init_rd(&r, &info, text, CONSOLE_LOG_MAX);
2869 if (!prb_read_valid(prb, con->seq, &r))
2872 if (con->seq != r.info->seq) {
2873 con->dropped += r.info->seq - con->seq;
2874 con->seq = r.info->seq;
2875 if (panic_in_progress() &&
2876 atomic_fetch_inc_relaxed(&panic_console_dropped) > 10) {
2877 suppress_panic_printk = 1;
2878 pr_warn_once("Too many dropped messages. Suppress messages on non-panic CPUs to prevent livelock.\n");
2882 /* Skip record that has level above the console loglevel. */
2883 if (suppress_message_printing(r.info->level)) {
2889 write_text = ext_text;
2890 len = info_print_ext_header(ext_text, CONSOLE_EXT_LOG_MAX, r.info);
2891 len += msg_print_ext_body(ext_text + len, CONSOLE_EXT_LOG_MAX - len,
2892 &r.text_buf[0], r.info->text_len, &r.info->dev_info);
2895 len = record_print_text(&r, console_msg_format & MSG_FORMAT_SYSLOG, printk_time);
2900 * While actively printing out messages, if another printk()
2901 * were to occur on another CPU, it may wait for this one to
2902 * finish. This task can not be preempted if there is a
2903 * waiter waiting to take over.
2905 * Interrupts are disabled because the hand over to a waiter
2906 * must not be interrupted until the hand over is completed
2907 * (@console_waiter is cleared).
2909 printk_safe_enter_irqsave(flags);
2910 console_lock_spinning_enable();
2912 /* don't trace irqsoff print latency */
2913 stop_critical_timings();
2916 call_console_driver(con, write_text, len, dropped_text);
2921 start_critical_timings();
2922 *handover = console_lock_spinning_disable_and_check();
2923 printk_safe_exit_irqrestore(flags);
2930 * Print a record for a given console, but allow another printk() caller to
2931 * take over the console_lock and continue printing.
2933 * Requires the console_lock, but depending on @handover after the call, the
2934 * caller may no longer have the console_lock.
2936 * See __console_emit_next_record() for argument and return details.
2938 static bool console_emit_next_record_transferable(struct console *con, char *text, char *ext_text,
2939 char *dropped_text, bool *handover)
2942 * Handovers are only supported if threaded printers are atomically
2943 * blocked. The context taking over the console_lock may be atomic.
2945 if (!console_kthreads_atomically_blocked()) {
2950 return __console_emit_next_record(con, text, ext_text, dropped_text, handover);
2954 * Print out all remaining records to all consoles.
2956 * @do_cond_resched is set by the caller. It can be true only in schedulable
2959 * @next_seq is set to the sequence number after the last available record.
2960 * The value is valid only when this function returns true. It means that all
2961 * usable consoles are completely flushed.
2963 * @handover will be set to true if a printk waiter has taken over the
2964 * console_lock, in which case the caller is no longer holding the
2965 * console_lock. Otherwise it is set to false.
2967 * Returns true when there was at least one usable console and all messages
2968 * were flushed to all usable consoles. A returned false informs the caller
2969 * that everything was not flushed (either there were no usable consoles or
2970 * another context has taken over printing or it is a panic situation and this
2971 * is not the panic CPU or direct printing is not preferred). Regardless the
2972 * reason, the caller should assume it is not useful to immediately try again.
2974 * Requires the console_lock.
2976 static bool console_flush_all(bool do_cond_resched, u64 *next_seq, bool *handover)
2978 static char dropped_text[DROPPED_TEXT_MAX];
2979 static char ext_text[CONSOLE_EXT_LOG_MAX];
2980 static char text[CONSOLE_LOG_MAX];
2981 bool any_usable = false;
2982 struct console *con;
2989 /* Let the kthread printers do the work if they can. */
2990 if (!allow_direct_printing())
2993 any_progress = false;
2995 for_each_console(con) {
2998 if (!console_is_usable(con))
3002 if (con->flags & CON_EXTENDED) {
3003 /* Extended consoles do not print "dropped messages". */
3004 progress = console_emit_next_record_transferable(con, &text[0],
3005 &ext_text[0], NULL, handover);
3007 progress = console_emit_next_record_transferable(con, &text[0],
3008 NULL, &dropped_text[0], handover);
3013 /* Track the next of the highest seq flushed. */
3014 if (con->seq > *next_seq)
3015 *next_seq = con->seq;
3019 any_progress = true;
3021 /* Allow panic_cpu to take over the consoles safely. */
3022 if (abandon_console_lock_in_panic())
3025 if (do_cond_resched)
3028 } while (any_progress);
3034 * console_unlock - unlock the console system
3036 * Releases the console_lock which the caller holds on the console system
3037 * and the console driver list.
3039 * While the console_lock was held, console output may have been buffered
3040 * by printk(). If this is the case, console_unlock(); emits
3041 * the output prior to releasing the lock.
3043 * console_unlock(); may be called from any context.
3045 void console_unlock(void)
3047 bool do_cond_resched;
3052 if (console_suspended) {
3058 * Console drivers are called with interrupts disabled, so
3059 * @console_may_schedule should be cleared before; however, we may
3060 * end up dumping a lot of lines, for example, if called from
3061 * console registration path, and should invoke cond_resched()
3062 * between lines if allowable. Not doing so can cause a very long
3063 * scheduling stall on a slow console leading to RCU stall and
3064 * softlockup warnings which exacerbate the issue with more
3065 * messages practically incapacitating the system. Therefore, create
3066 * a local to use for the printing loop.
3068 do_cond_resched = console_may_schedule;
3071 console_may_schedule = 0;
3073 flushed = console_flush_all(do_cond_resched, &next_seq, &handover);
3078 * Abort if there was a failure to flush all messages to all
3079 * usable consoles. Either it is not possible to flush (in
3080 * which case it would be an infinite loop of retrying) or
3081 * another context has taken over printing.
3087 * Some context may have added new records after
3088 * console_flush_all() but before unlocking the console.
3089 * Re-check if there is a new record to flush. If the trylock
3090 * fails, another context is already handling the printing.
3092 } while (prb_read_valid(prb, next_seq, NULL) && console_trylock());
3094 EXPORT_SYMBOL(console_unlock);
3097 * console_conditional_schedule - yield the CPU if required
3099 * If the console code is currently allowed to sleep, and
3100 * if this CPU should yield the CPU to another task, do
3103 * Must be called within console_lock();.
3105 void __sched console_conditional_schedule(void)
3107 if (console_may_schedule)
3110 EXPORT_SYMBOL(console_conditional_schedule);
3112 void console_unblank(void)
3117 * console_unblank can no longer be called in interrupt context unless
3118 * oops_in_progress is set to 1..
3120 if (oops_in_progress) {
3121 if (down_trylock_console_sem() != 0)
3123 if (!console_kthreads_atomic_tryblock()) {
3130 console_may_schedule = 0;
3132 if ((c->flags & CON_ENABLED) && c->unblank)
3136 if (!oops_in_progress)
3137 pr_flush(1000, true);
3141 * console_flush_on_panic - flush console content on panic
3142 * @mode: flush all messages in buffer or just the pending ones
3144 * Immediately output all pending messages no matter what.
3146 void console_flush_on_panic(enum con_flush_mode mode)
3149 * If someone else is holding the console lock, trylock will fail
3150 * and may_schedule may be set. Ignore and proceed to unlock so
3151 * that messages are flushed out. As this can be called from any
3152 * context and we don't want to get preempted while flushing,
3153 * ensure may_schedule is cleared.
3156 console_may_schedule = 0;
3158 if (mode == CONSOLE_REPLAY_ALL) {
3162 seq = prb_first_valid_seq(prb);
3170 * Return the console tty driver structure and its associated index
3172 struct tty_driver *console_device(int *index)
3175 struct tty_driver *driver = NULL;
3178 for_each_console(c) {
3181 driver = c->device(c, index);
3190 * Prevent further output on the passed console device so that (for example)
3191 * serial drivers can disable console output before suspending a port, and can
3192 * re-enable output afterwards.
3194 void console_stop(struct console *console)
3196 __pr_flush(console, 1000, true);
3198 console->flags &= ~CON_ENABLED;
3201 EXPORT_SYMBOL(console_stop);
3203 void console_start(struct console *console)
3206 console->flags |= CON_ENABLED;
3208 __pr_flush(console, 1000, true);
3210 EXPORT_SYMBOL(console_start);
3212 static int __read_mostly keep_bootcon;
3214 static int __init keep_bootcon_setup(char *str)
3217 pr_info("debug: skip boot console de-registration.\n");
3222 early_param("keep_bootcon", keep_bootcon_setup);
3225 * This is called by register_console() to try to match
3226 * the newly registered console with any of the ones selected
3227 * by either the command line or add_preferred_console() and
3230 * Care need to be taken with consoles that are statically
3231 * enabled such as netconsole
3233 static int try_enable_preferred_console(struct console *newcon,
3234 bool user_specified)
3236 struct console_cmdline *c;
3239 for (i = 0, c = console_cmdline;
3240 i < MAX_CMDLINECONSOLES && c->name[0];
3242 if (c->user_specified != user_specified)
3244 if (!newcon->match ||
3245 newcon->match(newcon, c->name, c->index, c->options) != 0) {
3246 /* default matching */
3247 BUILD_BUG_ON(sizeof(c->name) != sizeof(newcon->name));
3248 if (strcmp(c->name, newcon->name) != 0)
3250 if (newcon->index >= 0 &&
3251 newcon->index != c->index)
3253 if (newcon->index < 0)
3254 newcon->index = c->index;
3256 if (_braille_register_console(newcon, c))
3259 if (newcon->setup &&
3260 (err = newcon->setup(newcon, c->options)) != 0)
3263 newcon->flags |= CON_ENABLED;
3264 if (i == preferred_console)
3265 newcon->flags |= CON_CONSDEV;
3270 * Some consoles, such as pstore and netconsole, can be enabled even
3271 * without matching. Accept the pre-enabled consoles only when match()
3272 * and setup() had a chance to be called.
3274 if (newcon->flags & CON_ENABLED && c->user_specified == user_specified)
3280 /* Try to enable the console unconditionally */
3281 static void try_enable_default_console(struct console *newcon)
3283 if (newcon->index < 0)
3286 if (newcon->setup && newcon->setup(newcon, NULL) != 0)
3289 newcon->flags |= CON_ENABLED;
3292 newcon->flags |= CON_CONSDEV;
3295 #define con_printk(lvl, con, fmt, ...) \
3296 printk(lvl pr_fmt("%sconsole [%s%d] " fmt), \
3297 (con->flags & CON_BOOT) ? "boot" : "", \
3298 con->name, con->index, ##__VA_ARGS__)
3301 * The console driver calls this routine during kernel initialization
3302 * to register the console printing procedure with printk() and to
3303 * print any messages that were printed by the kernel before the
3304 * console driver was initialized.
3306 * This can happen pretty early during the boot process (because of
3307 * early_printk) - sometimes before setup_arch() completes - be careful
3308 * of what kernel features are used - they may not be initialised yet.
3310 * There are two types of consoles - bootconsoles (early_printk) and
3311 * "real" consoles (everything which is not a bootconsole) which are
3312 * handled differently.
3313 * - Any number of bootconsoles can be registered at any time.
3314 * - As soon as a "real" console is registered, all bootconsoles
3315 * will be unregistered automatically.
3316 * - Once a "real" console is registered, any attempt to register a
3317 * bootconsoles will be rejected
3319 void register_console(struct console *newcon)
3321 struct console *con;
3322 bool bootcon_enabled = false;
3323 bool realcon_enabled = false;
3326 for_each_console(con) {
3327 if (WARN(con == newcon, "console '%s%d' already registered\n",
3328 con->name, con->index))
3332 for_each_console(con) {
3333 if (con->flags & CON_BOOT)
3334 bootcon_enabled = true;
3336 realcon_enabled = true;
3339 /* Do not register boot consoles when there already is a real one. */
3340 if (newcon->flags & CON_BOOT && realcon_enabled) {
3341 pr_info("Too late to register bootconsole %s%d\n",
3342 newcon->name, newcon->index);
3347 * See if we want to enable this console driver by default.
3349 * Nope when a console is preferred by the command line, device
3352 * The first real console with tty binding (driver) wins. More
3353 * consoles might get enabled before the right one is found.
3355 * Note that a console with tty binding will have CON_CONSDEV
3356 * flag set and will be first in the list.
3358 if (preferred_console < 0) {
3359 if (!console_drivers || !console_drivers->device ||
3360 console_drivers->flags & CON_BOOT) {
3361 try_enable_default_console(newcon);
3365 /* See if this console matches one we selected on the command line */
3366 err = try_enable_preferred_console(newcon, true);
3368 /* If not, try to match against the platform default(s) */
3370 err = try_enable_preferred_console(newcon, false);
3372 /* printk() messages are not printed to the Braille console. */
3373 if (err || newcon->flags & CON_BRL)
3377 * If we have a bootconsole, and are switching to a real console,
3378 * don't print everything out again, since when the boot console, and
3379 * the real console are the same physical device, it's annoying to
3380 * see the beginning boot messages twice
3382 if (bootcon_enabled &&
3383 ((newcon->flags & (CON_CONSDEV | CON_BOOT)) == CON_CONSDEV)) {
3384 newcon->flags &= ~CON_PRINTBUFFER;
3388 * Put this console in the list - keep the
3389 * preferred driver at the head of the list.
3392 if ((newcon->flags & CON_CONSDEV) || console_drivers == NULL) {
3393 newcon->next = console_drivers;
3394 console_drivers = newcon;
3396 newcon->next->flags &= ~CON_CONSDEV;
3397 /* Ensure this flag is always set for the head of the list */
3398 newcon->flags |= CON_CONSDEV;
3400 newcon->next = console_drivers->next;
3401 console_drivers->next = newcon;
3404 if (newcon->flags & CON_EXTENDED)
3405 nr_ext_console_drivers++;
3407 newcon->dropped = 0;
3408 newcon->thread = NULL;
3409 newcon->blocked = true;
3410 mutex_init(&newcon->lock);
3412 if (newcon->flags & CON_PRINTBUFFER) {
3413 /* Get a consistent copy of @syslog_seq. */
3414 mutex_lock(&syslog_lock);
3415 newcon->seq = syslog_seq;
3416 mutex_unlock(&syslog_lock);
3418 /* Begin with next message. */
3419 newcon->seq = prb_next_seq(prb);
3422 if (printk_kthreads_available)
3423 printk_start_kthread(newcon);
3426 console_sysfs_notify();
3429 * By unregistering the bootconsoles after we enable the real console
3430 * we get the "console xxx enabled" message on all the consoles -
3431 * boot consoles, real consoles, etc - this is to ensure that end
3432 * users know there might be something in the kernel's log buffer that
3433 * went to the bootconsole (that they do not see on the real console)
3435 con_printk(KERN_INFO, newcon, "enabled\n");
3436 if (bootcon_enabled &&
3437 ((newcon->flags & (CON_CONSDEV | CON_BOOT)) == CON_CONSDEV) &&
3439 /* We need to iterate through all boot consoles, to make
3440 * sure we print everything out, before we unregister them.
3442 for_each_console(con)
3443 if (con->flags & CON_BOOT)
3444 unregister_console(con);
3447 EXPORT_SYMBOL(register_console);
3449 int unregister_console(struct console *console)
3451 struct task_struct *thd;
3452 struct console *con;
3455 con_printk(KERN_INFO, console, "disabled\n");
3457 res = _braille_unregister_console(console);
3465 if (console_drivers == console) {
3466 console_drivers=console->next;
3469 for_each_console(con) {
3470 if (con->next == console) {
3471 con->next = console->next;
3479 goto out_disable_unlock;
3481 if (console->flags & CON_EXTENDED)
3482 nr_ext_console_drivers--;
3485 * If this isn't the last console and it has CON_CONSDEV set, we
3486 * need to set it on the next preferred console.
3488 if (console_drivers != NULL && console->flags & CON_CONSDEV)
3489 console_drivers->flags |= CON_CONSDEV;
3491 console->flags &= ~CON_ENABLED;
3494 * console->thread can only be cleared under the console lock. But
3495 * stopping the thread must be done without the console lock. The
3496 * task that clears @thread is the task that stops the kthread.
3498 thd = console->thread;
3499 console->thread = NULL;
3506 console_sysfs_notify();
3509 res = console->exit(console);
3514 console->flags &= ~CON_ENABLED;
3519 EXPORT_SYMBOL(unregister_console);
3522 * Initialize the console device. This is called *early*, so
3523 * we can't necessarily depend on lots of kernel help here.
3524 * Just do some early initializations, and do the complex setup
3527 void __init console_init(void)
3531 initcall_entry_t *ce;
3533 /* Setup the default TTY line discipline. */
3537 * set up the console device so that later boot sequences can
3538 * inform about problems etc..
3540 ce = __con_initcall_start;
3541 trace_initcall_level("console");
3542 while (ce < __con_initcall_end) {
3543 call = initcall_from_entry(ce);
3544 trace_initcall_start(call);
3546 trace_initcall_finish(call, ret);
3552 * Some boot consoles access data that is in the init section and which will
3553 * be discarded after the initcalls have been run. To make sure that no code
3554 * will access this data, unregister the boot consoles in a late initcall.
3556 * If for some reason, such as deferred probe or the driver being a loadable
3557 * module, the real console hasn't registered yet at this point, there will
3558 * be a brief interval in which no messages are logged to the console, which
3559 * makes it difficult to diagnose problems that occur during this time.
3561 * To mitigate this problem somewhat, only unregister consoles whose memory
3562 * intersects with the init section. Note that all other boot consoles will
3563 * get unregistered when the real preferred console is registered.
3565 static int __init printk_late_init(void)
3567 struct console *con;
3570 for_each_console(con) {
3571 if (!(con->flags & CON_BOOT))
3574 /* Check addresses that might be used for enabled consoles. */
3575 if (init_section_intersects(con, sizeof(*con)) ||
3576 init_section_contains(con->write, 0) ||
3577 init_section_contains(con->read, 0) ||
3578 init_section_contains(con->device, 0) ||
3579 init_section_contains(con->unblank, 0) ||
3580 init_section_contains(con->data, 0)) {
3582 * Please, consider moving the reported consoles out
3583 * of the init section.
3585 pr_warn("bootconsole [%s%d] uses init memory and must be disabled even before the real one is ready\n",
3586 con->name, con->index);
3587 unregister_console(con);
3590 ret = cpuhp_setup_state_nocalls(CPUHP_PRINTK_DEAD, "printk:dead", NULL,
3591 console_cpu_notify);
3593 ret = cpuhp_setup_state_nocalls(CPUHP_AP_ONLINE_DYN, "printk:online",
3594 console_cpu_notify, NULL);
3596 printk_sysctl_init();
3599 late_initcall(printk_late_init);
3601 static int __init printk_activate_kthreads(void)
3603 struct console *con;
3606 printk_kthreads_available = true;
3607 for_each_console(con)
3608 printk_start_kthread(con);
3613 early_initcall(printk_activate_kthreads);
3615 #if defined CONFIG_PRINTK
3616 /* If @con is specified, only wait for that console. Otherwise wait for all. */
3617 static bool __pr_flush(struct console *con, int timeout_ms, bool reset_on_progress)
3619 int remaining = timeout_ms;
3628 seq = prb_next_seq(prb);
3634 for_each_console(c) {
3635 if (con && con != c)
3637 if (!console_is_usable(c))
3639 printk_seq = c->seq;
3640 if (printk_seq < seq)
3641 diff += seq - printk_seq;
3645 if (diff != last_diff && reset_on_progress)
3646 remaining = timeout_ms;
3648 if (diff == 0 || remaining == 0)
3651 if (remaining < 0) {
3652 /* no timeout limit */
3654 } else if (remaining < 100) {
3669 * pr_flush() - Wait for printing threads to catch up.
3671 * @timeout_ms: The maximum time (in ms) to wait.
3672 * @reset_on_progress: Reset the timeout if forward progress is seen.
3674 * A value of 0 for @timeout_ms means no waiting will occur. A value of -1
3675 * represents infinite waiting.
3677 * If @reset_on_progress is true, the timeout will be reset whenever any
3678 * printer has been seen to make some forward progress.
3680 * Context: Process context. May sleep while acquiring console lock.
3681 * Return: true if all enabled printers are caught up.
3683 bool pr_flush(int timeout_ms, bool reset_on_progress)
3685 return __pr_flush(NULL, timeout_ms, reset_on_progress);
3687 EXPORT_SYMBOL(pr_flush);
3689 static void __printk_fallback_preferred_direct(void)
3691 printk_prefer_direct_enter();
3692 pr_err("falling back to preferred direct printing\n");
3693 printk_kthreads_available = false;
3697 * Enter preferred direct printing, but never exit. Mark console threads as
3698 * unavailable. The system is then forever in preferred direct printing and
3699 * any printing threads will exit.
3701 * Must *not* be called under console_lock. Use
3702 * __printk_fallback_preferred_direct() if already holding console_lock.
3704 static void printk_fallback_preferred_direct(void)
3707 __printk_fallback_preferred_direct();
3712 * Print a record for a given console, not allowing another printk() caller
3713 * to take over. This is appropriate for contexts that do not have the
3716 * See __console_emit_next_record() for argument and return details.
3718 static bool console_emit_next_record(struct console *con, char *text, char *ext_text,
3721 return __console_emit_next_record(con, text, ext_text, dropped_text, NULL);
3724 static bool printer_should_wake(struct console *con, u64 seq)
3728 if (kthread_should_stop() || !printk_kthreads_available)
3732 console_kthreads_atomically_blocked()) {
3737 * This is an unsafe read from con->flags, but a false positive is
3738 * not a problem. Worst case it would allow the printer to wake up
3739 * although it is disabled. But the printer will notice that when
3740 * attempting to print and instead go back to sleep.
3742 flags = data_race(READ_ONCE(con->flags));
3744 if (!__console_is_usable(flags))
3747 return prb_read_valid(prb, seq, NULL);
3750 static int printk_kthread_func(void *data)
3752 struct console *con = data;
3753 char *dropped_text = NULL;
3754 char *ext_text = NULL;
3759 text = kmalloc(CONSOLE_LOG_MAX, GFP_KERNEL);
3761 con_printk(KERN_ERR, con, "failed to allocate text buffer\n");
3762 printk_fallback_preferred_direct();
3766 if (con->flags & CON_EXTENDED) {
3767 ext_text = kmalloc(CONSOLE_EXT_LOG_MAX, GFP_KERNEL);
3769 con_printk(KERN_ERR, con, "failed to allocate ext_text buffer\n");
3770 printk_fallback_preferred_direct();
3774 dropped_text = kmalloc(DROPPED_TEXT_MAX, GFP_KERNEL);
3775 if (!dropped_text) {
3776 con_printk(KERN_ERR, con, "failed to allocate dropped_text buffer\n");
3777 printk_fallback_preferred_direct();
3782 con_printk(KERN_INFO, con, "printing thread started\n");
3786 * Guarantee this task is visible on the waitqueue before
3787 * checking the wake condition.
3789 * The full memory barrier within set_current_state() of
3790 * prepare_to_wait_event() pairs with the full memory barrier
3791 * within wq_has_sleeper().
3793 * This pairs with __wake_up_klogd:A.
3795 error = wait_event_interruptible(log_wait,
3796 printer_should_wake(con, seq)); /* LMM(printk_kthread_func:A) */
3798 if (kthread_should_stop() || !printk_kthreads_available)
3804 error = mutex_lock_interruptible(&con->lock);
3809 !console_kthread_printing_tryenter()) {
3810 /* Another context has locked the console_lock. */
3811 mutex_unlock(&con->lock);
3816 * Although this context has not locked the console_lock, it
3817 * is known that the console_lock is not locked and it is not
3818 * possible for any other context to lock the console_lock.
3819 * Therefore it is safe to read con->flags.
3822 if (!__console_is_usable(con->flags)) {
3823 console_kthread_printing_exit();
3824 mutex_unlock(&con->lock);
3829 * Even though the printk kthread is always preemptible, it is
3830 * still not allowed to call cond_resched() from within
3831 * console drivers. The task may become non-preemptible in the
3832 * console driver call chain. For example, vt_console_print()
3833 * takes a spinlock and then can call into fbcon_redraw(),
3834 * which can conditionally invoke cond_resched().
3836 console_may_schedule = 0;
3837 console_emit_next_record(con, text, ext_text, dropped_text);
3841 console_kthread_printing_exit();
3843 mutex_unlock(&con->lock);
3846 con_printk(KERN_INFO, con, "printing thread stopped\n");
3848 kfree(dropped_text);
3854 * If this kthread is being stopped by another task, con->thread will
3855 * already be NULL. That is fine. The important thing is that it is
3856 * NULL after the kthread exits.
3864 /* Must be called under console_lock. */
3865 static void printk_start_kthread(struct console *con)
3868 * Do not start a kthread if there is no write() callback. The
3869 * kthreads assume the write() callback exists.
3874 con->thread = kthread_run(printk_kthread_func, con,
3875 "pr/%s%d", con->name, con->index);
3876 if (IS_ERR(con->thread)) {
3878 con_printk(KERN_ERR, con, "unable to start printing thread\n");
3879 __printk_fallback_preferred_direct();
3885 * Delayed printk version, for scheduler-internal messages:
3887 #define PRINTK_PENDING_WAKEUP 0x01
3888 #define PRINTK_PENDING_DIRECT_OUTPUT 0x02
3890 static DEFINE_PER_CPU(int, printk_pending);
3892 static void wake_up_klogd_work_func(struct irq_work *irq_work)
3894 int pending = this_cpu_xchg(printk_pending, 0);
3896 if (pending & PRINTK_PENDING_DIRECT_OUTPUT) {
3897 printk_prefer_direct_enter();
3899 /* If trylock fails, someone else is doing the printing */
3900 if (console_trylock())
3903 printk_prefer_direct_exit();
3906 if (pending & PRINTK_PENDING_WAKEUP)
3907 wake_up_interruptible_all(&log_wait);
3910 static DEFINE_PER_CPU(struct irq_work, wake_up_klogd_work) =
3911 IRQ_WORK_INIT_LAZY(wake_up_klogd_work_func);
3913 static void __wake_up_klogd(int val)
3915 if (!printk_percpu_data_ready())
3920 * Guarantee any new records can be seen by tasks preparing to wait
3921 * before this context checks if the wait queue is empty.
3923 * The full memory barrier within wq_has_sleeper() pairs with the full
3924 * memory barrier within set_current_state() of
3925 * prepare_to_wait_event(), which is called after ___wait_event() adds
3926 * the waiter but before it has checked the wait condition.
3928 * This pairs with devkmsg_read:A, syslog_print:A, and
3929 * printk_kthread_func:A.
3931 if (wq_has_sleeper(&log_wait) || /* LMM(__wake_up_klogd:A) */
3932 (val & PRINTK_PENDING_DIRECT_OUTPUT)) {
3933 this_cpu_or(printk_pending, val);
3934 irq_work_queue(this_cpu_ptr(&wake_up_klogd_work));
3939 void wake_up_klogd(void)
3941 __wake_up_klogd(PRINTK_PENDING_WAKEUP);
3944 void defer_console_output(void)
3947 * New messages may have been added directly to the ringbuffer
3948 * using vprintk_store(), so wake any waiters as well.
3950 int val = PRINTK_PENDING_WAKEUP;
3953 * Make sure that some context will print the messages when direct
3954 * printing is allowed. This happens in situations when the kthreads
3955 * may not be as reliable or perhaps unusable.
3957 if (allow_direct_printing())
3958 val |= PRINTK_PENDING_DIRECT_OUTPUT;
3960 __wake_up_klogd(val);
3963 void printk_trigger_flush(void)
3965 defer_console_output();
3968 int vprintk_deferred(const char *fmt, va_list args)
3972 r = vprintk_emit(0, LOGLEVEL_SCHED, NULL, fmt, args);
3973 defer_console_output();
3978 int _printk_deferred(const char *fmt, ...)
3983 va_start(args, fmt);
3984 r = vprintk_deferred(fmt, args);
3991 * printk rate limiting, lifted from the networking subsystem.
3993 * This enforces a rate limit: not more than 10 kernel messages
3994 * every 5s to make a denial-of-service attack impossible.
3996 DEFINE_RATELIMIT_STATE(printk_ratelimit_state, 5 * HZ, 10);
3998 int __printk_ratelimit(const char *func)
4000 return ___ratelimit(&printk_ratelimit_state, func);
4002 EXPORT_SYMBOL(__printk_ratelimit);
4005 * printk_timed_ratelimit - caller-controlled printk ratelimiting
4006 * @caller_jiffies: pointer to caller's state
4007 * @interval_msecs: minimum interval between prints
4009 * printk_timed_ratelimit() returns true if more than @interval_msecs
4010 * milliseconds have elapsed since the last time printk_timed_ratelimit()
4013 bool printk_timed_ratelimit(unsigned long *caller_jiffies,
4014 unsigned int interval_msecs)
4016 unsigned long elapsed = jiffies - *caller_jiffies;
4018 if (*caller_jiffies && elapsed <= msecs_to_jiffies(interval_msecs))
4021 *caller_jiffies = jiffies;
4024 EXPORT_SYMBOL(printk_timed_ratelimit);
4026 static DEFINE_SPINLOCK(dump_list_lock);
4027 static LIST_HEAD(dump_list);
4030 * kmsg_dump_register - register a kernel log dumper.
4031 * @dumper: pointer to the kmsg_dumper structure
4033 * Adds a kernel log dumper to the system. The dump callback in the
4034 * structure will be called when the kernel oopses or panics and must be
4035 * set. Returns zero on success and %-EINVAL or %-EBUSY otherwise.
4037 int kmsg_dump_register(struct kmsg_dumper *dumper)
4039 unsigned long flags;
4042 /* The dump callback needs to be set */
4046 spin_lock_irqsave(&dump_list_lock, flags);
4047 /* Don't allow registering multiple times */
4048 if (!dumper->registered) {
4049 dumper->registered = 1;
4050 list_add_tail_rcu(&dumper->list, &dump_list);
4053 spin_unlock_irqrestore(&dump_list_lock, flags);
4057 EXPORT_SYMBOL_GPL(kmsg_dump_register);
4060 * kmsg_dump_unregister - unregister a kmsg dumper.
4061 * @dumper: pointer to the kmsg_dumper structure
4063 * Removes a dump device from the system. Returns zero on success and
4064 * %-EINVAL otherwise.
4066 int kmsg_dump_unregister(struct kmsg_dumper *dumper)
4068 unsigned long flags;
4071 spin_lock_irqsave(&dump_list_lock, flags);
4072 if (dumper->registered) {
4073 dumper->registered = 0;
4074 list_del_rcu(&dumper->list);
4077 spin_unlock_irqrestore(&dump_list_lock, flags);
4082 EXPORT_SYMBOL_GPL(kmsg_dump_unregister);
4084 static bool always_kmsg_dump;
4085 module_param_named(always_kmsg_dump, always_kmsg_dump, bool, S_IRUGO | S_IWUSR);
4087 const char *kmsg_dump_reason_str(enum kmsg_dump_reason reason)
4090 case KMSG_DUMP_PANIC:
4092 case KMSG_DUMP_OOPS:
4094 case KMSG_DUMP_EMERG:
4096 case KMSG_DUMP_SHUTDOWN:
4102 EXPORT_SYMBOL_GPL(kmsg_dump_reason_str);
4105 * kmsg_dump - dump kernel log to kernel message dumpers.
4106 * @reason: the reason (oops, panic etc) for dumping
4108 * Call each of the registered dumper's dump() callback, which can
4109 * retrieve the kmsg records with kmsg_dump_get_line() or
4110 * kmsg_dump_get_buffer().
4112 void kmsg_dump(enum kmsg_dump_reason reason)
4114 struct kmsg_dumper *dumper;
4117 list_for_each_entry_rcu(dumper, &dump_list, list) {
4118 enum kmsg_dump_reason max_reason = dumper->max_reason;
4121 * If client has not provided a specific max_reason, default
4122 * to KMSG_DUMP_OOPS, unless always_kmsg_dump was set.
4124 if (max_reason == KMSG_DUMP_UNDEF) {
4125 max_reason = always_kmsg_dump ? KMSG_DUMP_MAX :
4128 if (reason > max_reason)
4131 /* invoke dumper which will iterate over records */
4132 dumper->dump(dumper, reason);
4138 * kmsg_dump_get_line - retrieve one kmsg log line
4139 * @iter: kmsg dump iterator
4140 * @syslog: include the "<4>" prefixes
4141 * @line: buffer to copy the line to
4142 * @size: maximum size of the buffer
4143 * @len: length of line placed into buffer
4145 * Start at the beginning of the kmsg buffer, with the oldest kmsg
4146 * record, and copy one record into the provided buffer.
4148 * Consecutive calls will return the next available record moving
4149 * towards the end of the buffer with the youngest messages.
4151 * A return value of FALSE indicates that there are no more records to
4154 bool kmsg_dump_get_line(struct kmsg_dump_iter *iter, bool syslog,
4155 char *line, size_t size, size_t *len)
4157 u64 min_seq = latched_seq_read_nolock(&clear_seq);
4158 struct printk_info info;
4159 unsigned int line_count;
4160 struct printk_record r;
4164 if (iter->cur_seq < min_seq)
4165 iter->cur_seq = min_seq;
4167 prb_rec_init_rd(&r, &info, line, size);
4169 /* Read text or count text lines? */
4171 if (!prb_read_valid(prb, iter->cur_seq, &r))
4173 l = record_print_text(&r, syslog, printk_time);
4175 if (!prb_read_valid_info(prb, iter->cur_seq,
4176 &info, &line_count)) {
4179 l = get_record_print_text_size(&info, line_count, syslog,
4184 iter->cur_seq = r.info->seq + 1;
4191 EXPORT_SYMBOL_GPL(kmsg_dump_get_line);
4194 * kmsg_dump_get_buffer - copy kmsg log lines
4195 * @iter: kmsg dump iterator
4196 * @syslog: include the "<4>" prefixes
4197 * @buf: buffer to copy the line to
4198 * @size: maximum size of the buffer
4199 * @len_out: length of line placed into buffer
4201 * Start at the end of the kmsg buffer and fill the provided buffer
4202 * with as many of the *youngest* kmsg records that fit into it.
4203 * If the buffer is large enough, all available kmsg records will be
4204 * copied with a single call.
4206 * Consecutive calls will fill the buffer with the next block of
4207 * available older records, not including the earlier retrieved ones.
4209 * A return value of FALSE indicates that there are no more records to
4212 bool kmsg_dump_get_buffer(struct kmsg_dump_iter *iter, bool syslog,
4213 char *buf, size_t size, size_t *len_out)
4215 u64 min_seq = latched_seq_read_nolock(&clear_seq);
4216 struct printk_info info;
4217 struct printk_record r;
4222 bool time = printk_time;
4227 if (iter->cur_seq < min_seq)
4228 iter->cur_seq = min_seq;
4230 if (prb_read_valid_info(prb, iter->cur_seq, &info, NULL)) {
4231 if (info.seq != iter->cur_seq) {
4232 /* messages are gone, move to first available one */
4233 iter->cur_seq = info.seq;
4238 if (iter->cur_seq >= iter->next_seq)
4242 * Find first record that fits, including all following records,
4243 * into the user-provided buffer for this dump. Pass in size-1
4244 * because this function (by way of record_print_text()) will
4245 * not write more than size-1 bytes of text into @buf.
4247 seq = find_first_fitting_seq(iter->cur_seq, iter->next_seq,
4248 size - 1, syslog, time);
4251 * Next kmsg_dump_get_buffer() invocation will dump block of
4252 * older records stored right before this one.
4256 prb_rec_init_rd(&r, &info, buf, size);
4259 prb_for_each_record(seq, prb, seq, &r) {
4260 if (r.info->seq >= iter->next_seq)
4263 len += record_print_text(&r, syslog, time);
4265 /* Adjust record to store to remaining buffer space. */
4266 prb_rec_init_rd(&r, &info, buf + len, size - len);
4269 iter->next_seq = next_seq;
4276 EXPORT_SYMBOL_GPL(kmsg_dump_get_buffer);
4279 * kmsg_dump_rewind - reset the iterator
4280 * @iter: kmsg dump iterator
4282 * Reset the dumper's iterator so that kmsg_dump_get_line() and
4283 * kmsg_dump_get_buffer() can be called again and used multiple
4284 * times within the same dumper.dump() callback.
4286 void kmsg_dump_rewind(struct kmsg_dump_iter *iter)
4288 iter->cur_seq = latched_seq_read_nolock(&clear_seq);
4289 iter->next_seq = prb_next_seq(prb);
4291 EXPORT_SYMBOL_GPL(kmsg_dump_rewind);
4296 static atomic_t printk_cpu_sync_owner = ATOMIC_INIT(-1);
4297 static atomic_t printk_cpu_sync_nested = ATOMIC_INIT(0);
4300 * __printk_cpu_sync_wait() - Busy wait until the printk cpu-reentrant
4301 * spinning lock is not owned by any CPU.
4303 * Context: Any context.
4305 void __printk_cpu_sync_wait(void)
4309 } while (atomic_read(&printk_cpu_sync_owner) != -1);
4311 EXPORT_SYMBOL(__printk_cpu_sync_wait);
4314 * __printk_cpu_sync_try_get() - Try to acquire the printk cpu-reentrant
4317 * If no processor has the lock, the calling processor takes the lock and
4318 * becomes the owner. If the calling processor is already the owner of the
4319 * lock, this function succeeds immediately.
4321 * Context: Any context. Expects interrupts to be disabled.
4322 * Return: 1 on success, otherwise 0.
4324 int __printk_cpu_sync_try_get(void)
4329 cpu = smp_processor_id();
4332 * Guarantee loads and stores from this CPU when it is the lock owner
4333 * are _not_ visible to the previous lock owner. This pairs with
4334 * __printk_cpu_sync_put:B.
4336 * Memory barrier involvement:
4338 * If __printk_cpu_sync_try_get:A reads from __printk_cpu_sync_put:B,
4339 * then __printk_cpu_sync_put:A can never read from
4340 * __printk_cpu_sync_try_get:B.
4344 * RELEASE from __printk_cpu_sync_put:A to __printk_cpu_sync_put:B
4345 * of the previous CPU
4347 * ACQUIRE from __printk_cpu_sync_try_get:A to
4348 * __printk_cpu_sync_try_get:B of this CPU
4350 old = atomic_cmpxchg_acquire(&printk_cpu_sync_owner, -1,
4351 cpu); /* LMM(__printk_cpu_sync_try_get:A) */
4354 * This CPU is now the owner and begins loading/storing
4355 * data: LMM(__printk_cpu_sync_try_get:B)
4359 } else if (old == cpu) {
4360 /* This CPU is already the owner. */
4361 atomic_inc(&printk_cpu_sync_nested);
4367 EXPORT_SYMBOL(__printk_cpu_sync_try_get);
4370 * __printk_cpu_sync_put() - Release the printk cpu-reentrant spinning lock.
4372 * The calling processor must be the owner of the lock.
4374 * Context: Any context. Expects interrupts to be disabled.
4376 void __printk_cpu_sync_put(void)
4378 if (atomic_read(&printk_cpu_sync_nested)) {
4379 atomic_dec(&printk_cpu_sync_nested);
4384 * This CPU is finished loading/storing data:
4385 * LMM(__printk_cpu_sync_put:A)
4389 * Guarantee loads and stores from this CPU when it was the
4390 * lock owner are visible to the next lock owner. This pairs
4391 * with __printk_cpu_sync_try_get:A.
4393 * Memory barrier involvement:
4395 * If __printk_cpu_sync_try_get:A reads from __printk_cpu_sync_put:B,
4396 * then __printk_cpu_sync_try_get:B reads from __printk_cpu_sync_put:A.
4400 * RELEASE from __printk_cpu_sync_put:A to __printk_cpu_sync_put:B
4403 * ACQUIRE from __printk_cpu_sync_try_get:A to
4404 * __printk_cpu_sync_try_get:B of the next CPU
4406 atomic_set_release(&printk_cpu_sync_owner,
4407 -1); /* LMM(__printk_cpu_sync_put:B) */
4409 EXPORT_SYMBOL(__printk_cpu_sync_put);
4410 #endif /* CONFIG_SMP */