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 * Helper macros to handle lockdep when locking/unlocking console_sem. We use
228 * macros instead of functions so that _RET_IP_ contains useful information.
230 #define down_console_sem() do { \
232 mutex_acquire(&console_lock_dep_map, 0, 0, _RET_IP_);\
235 static int __down_trylock_console_sem(unsigned long ip)
241 * Here and in __up_console_sem() we need to be in safe mode,
242 * because spindump/WARN/etc from under console ->lock will
243 * deadlock in printk()->down_trylock_console_sem() otherwise.
245 printk_safe_enter_irqsave(flags);
246 lock_failed = down_trylock(&console_sem);
247 printk_safe_exit_irqrestore(flags);
251 mutex_acquire(&console_lock_dep_map, 0, 1, ip);
254 #define down_trylock_console_sem() __down_trylock_console_sem(_RET_IP_)
256 static void __up_console_sem(unsigned long ip)
260 mutex_release(&console_lock_dep_map, ip);
262 printk_safe_enter_irqsave(flags);
264 printk_safe_exit_irqrestore(flags);
266 #define up_console_sem() __up_console_sem(_RET_IP_)
268 static bool panic_in_progress(void)
270 return unlikely(atomic_read(&panic_cpu) != PANIC_CPU_INVALID);
274 * This is used for debugging the mess that is the VT code by
275 * keeping track if we have the console semaphore held. It's
276 * definitely not the perfect debug tool (we don't know if _WE_
277 * hold it and are racing, but it helps tracking those weird code
278 * paths in the console code where we end up in places I want
279 * locked without the console semaphore held).
281 static int console_locked, console_suspended;
284 * Array of consoles built from command line options (console=)
287 #define MAX_CMDLINECONSOLES 8
289 static struct console_cmdline console_cmdline[MAX_CMDLINECONSOLES];
291 static int preferred_console = -1;
292 int console_set_on_cmdline;
293 EXPORT_SYMBOL(console_set_on_cmdline);
295 /* Flag: console code may call schedule() */
296 static int console_may_schedule;
298 enum con_msg_format_flags {
299 MSG_FORMAT_DEFAULT = 0,
300 MSG_FORMAT_SYSLOG = (1 << 0),
303 static int console_msg_format = MSG_FORMAT_DEFAULT;
306 * The printk log buffer consists of a sequenced collection of records, each
307 * containing variable length message text. Every record also contains its
308 * own meta-data (@info).
310 * Every record meta-data carries the timestamp in microseconds, as well as
311 * the standard userspace syslog level and syslog facility. The usual kernel
312 * messages use LOG_KERN; userspace-injected messages always carry a matching
313 * syslog facility, by default LOG_USER. The origin of every message can be
314 * reliably determined that way.
316 * The human readable log message of a record is available in @text, the
317 * length of the message text in @text_len. The stored message is not
320 * Optionally, a record can carry a dictionary of properties (key/value
321 * pairs), to provide userspace with a machine-readable message context.
323 * Examples for well-defined, commonly used property names are:
324 * DEVICE=b12:8 device identifier
328 * +sound:card0 subsystem:devname
329 * SUBSYSTEM=pci driver-core subsystem name
331 * Valid characters in property names are [a-zA-Z0-9.-_]. Property names
332 * and values are terminated by a '\0' character.
334 * Example of record values:
335 * record.text_buf = "it's a line" (unterminated)
336 * record.info.seq = 56
337 * record.info.ts_nsec = 36863
338 * record.info.text_len = 11
339 * record.info.facility = 0 (LOG_KERN)
340 * record.info.flags = 0
341 * record.info.level = 3 (LOG_ERR)
342 * record.info.caller_id = 299 (task 299)
343 * record.info.dev_info.subsystem = "pci" (terminated)
344 * record.info.dev_info.device = "+pci:0000:00:01.0" (terminated)
346 * The 'struct printk_info' buffer must never be directly exported to
347 * userspace, it is a kernel-private implementation detail that might
348 * need to be changed in the future, when the requirements change.
350 * /dev/kmsg exports the structured data in the following line format:
351 * "<level>,<sequnum>,<timestamp>,<contflag>[,additional_values, ... ];<message text>\n"
353 * Users of the export format should ignore possible additional values
354 * separated by ',', and find the message after the ';' character.
356 * The optional key/value pairs are attached as continuation lines starting
357 * with a space character and terminated by a newline. All possible
358 * non-prinatable characters are escaped in the "\xff" notation.
361 /* syslog_lock protects syslog_* variables and write access to clear_seq. */
362 static DEFINE_MUTEX(syslog_lock);
365 static atomic_t printk_prefer_direct = ATOMIC_INIT(0);
368 * printk_prefer_direct_enter - cause printk() calls to attempt direct
369 * printing to all enabled consoles
371 * Since it is not possible to call into the console printing code from any
372 * context, there is no guarantee that direct printing will occur.
374 * This globally effects all printk() callers.
376 * Context: Any context.
378 void printk_prefer_direct_enter(void)
380 atomic_inc(&printk_prefer_direct);
384 * printk_prefer_direct_exit - restore printk() behavior
386 * Context: Any context.
388 void printk_prefer_direct_exit(void)
390 WARN_ON(atomic_dec_if_positive(&printk_prefer_direct) < 0);
393 DECLARE_WAIT_QUEUE_HEAD(log_wait);
394 /* All 3 protected by @syslog_lock. */
395 /* the next printk record to read by syslog(READ) or /proc/kmsg */
396 static u64 syslog_seq;
397 static size_t syslog_partial;
398 static bool syslog_time;
401 seqcount_latch_t latch;
406 * The next printk record to read after the last 'clear' command. There are
407 * two copies (updated with seqcount_latch) so that reads can locklessly
408 * access a valid value. Writers are synchronized by @syslog_lock.
410 static struct latched_seq clear_seq = {
411 .latch = SEQCNT_LATCH_ZERO(clear_seq.latch),
416 #ifdef CONFIG_PRINTK_CALLER
417 #define PREFIX_MAX 48
419 #define PREFIX_MAX 32
422 /* the maximum size of a formatted record (i.e. with prefix added per line) */
423 #define CONSOLE_LOG_MAX 1024
425 /* the maximum size for a dropped text message */
426 #define DROPPED_TEXT_MAX 64
428 /* the maximum size allowed to be reserved for a record */
429 #define LOG_LINE_MAX (CONSOLE_LOG_MAX - PREFIX_MAX)
431 #define LOG_LEVEL(v) ((v) & 0x07)
432 #define LOG_FACILITY(v) ((v) >> 3 & 0xff)
435 #define LOG_ALIGN __alignof__(unsigned long)
436 #define __LOG_BUF_LEN (1 << CONFIG_LOG_BUF_SHIFT)
437 #define LOG_BUF_LEN_MAX (u32)(1 << 31)
438 static char __log_buf[__LOG_BUF_LEN] __aligned(LOG_ALIGN);
439 static char *log_buf = __log_buf;
440 static u32 log_buf_len = __LOG_BUF_LEN;
443 * Define the average message size. This only affects the number of
444 * descriptors that will be available. Underestimating is better than
445 * overestimating (too many available descriptors is better than not enough).
447 #define PRB_AVGBITS 5 /* 32 character average length */
449 #if CONFIG_LOG_BUF_SHIFT <= PRB_AVGBITS
450 #error CONFIG_LOG_BUF_SHIFT value too small.
452 _DEFINE_PRINTKRB(printk_rb_static, CONFIG_LOG_BUF_SHIFT - PRB_AVGBITS,
453 PRB_AVGBITS, &__log_buf[0]);
455 static struct printk_ringbuffer printk_rb_dynamic;
457 static struct printk_ringbuffer *prb = &printk_rb_static;
460 * We cannot access per-CPU data (e.g. per-CPU flush irq_work) before
461 * per_cpu_areas are initialised. This variable is set to true when
462 * it's safe to access per-CPU data.
464 static bool __printk_percpu_data_ready __read_mostly;
466 bool printk_percpu_data_ready(void)
468 return __printk_percpu_data_ready;
471 /* Must be called under syslog_lock. */
472 static void latched_seq_write(struct latched_seq *ls, u64 val)
474 raw_write_seqcount_latch(&ls->latch);
476 raw_write_seqcount_latch(&ls->latch);
480 /* Can be called from any context. */
481 static u64 latched_seq_read_nolock(struct latched_seq *ls)
488 seq = raw_read_seqcount_latch(&ls->latch);
491 } while (read_seqcount_latch_retry(&ls->latch, seq));
496 /* Return log buffer address */
497 char *log_buf_addr_get(void)
502 /* Return log buffer size */
503 u32 log_buf_len_get(void)
509 * Define how much of the log buffer we could take at maximum. The value
510 * must be greater than two. Note that only half of the buffer is available
511 * when the index points to the middle.
513 #define MAX_LOG_TAKE_PART 4
514 static const char trunc_msg[] = "<truncated>";
516 static void truncate_msg(u16 *text_len, u16 *trunc_msg_len)
519 * The message should not take the whole buffer. Otherwise, it might
520 * get removed too soon.
522 u32 max_text_len = log_buf_len / MAX_LOG_TAKE_PART;
524 if (*text_len > max_text_len)
525 *text_len = max_text_len;
527 /* enable the warning message (if there is room) */
528 *trunc_msg_len = strlen(trunc_msg);
529 if (*text_len >= *trunc_msg_len)
530 *text_len -= *trunc_msg_len;
535 int dmesg_restrict = IS_ENABLED(CONFIG_SECURITY_DMESG_RESTRICT);
537 static int syslog_action_restricted(int type)
542 * Unless restricted, we allow "read all" and "get buffer size"
545 return type != SYSLOG_ACTION_READ_ALL &&
546 type != SYSLOG_ACTION_SIZE_BUFFER;
549 static int check_syslog_permissions(int type, int source)
552 * If this is from /proc/kmsg and we've already opened it, then we've
553 * already done the capabilities checks at open time.
555 if (source == SYSLOG_FROM_PROC && type != SYSLOG_ACTION_OPEN)
558 if (syslog_action_restricted(type)) {
559 if (capable(CAP_SYSLOG))
562 * For historical reasons, accept CAP_SYS_ADMIN too, with
565 if (capable(CAP_SYS_ADMIN)) {
566 pr_warn_once("%s (%d): Attempt to access syslog with "
567 "CAP_SYS_ADMIN but no CAP_SYSLOG "
569 current->comm, task_pid_nr(current));
575 return security_syslog(type);
578 static void append_char(char **pp, char *e, char c)
584 static ssize_t info_print_ext_header(char *buf, size_t size,
585 struct printk_info *info)
587 u64 ts_usec = info->ts_nsec;
589 #ifdef CONFIG_PRINTK_CALLER
590 u32 id = info->caller_id;
592 snprintf(caller, sizeof(caller), ",caller=%c%u",
593 id & 0x80000000 ? 'C' : 'T', id & ~0x80000000);
598 do_div(ts_usec, 1000);
600 return scnprintf(buf, size, "%u,%llu,%llu,%c%s;",
601 (info->facility << 3) | info->level, info->seq,
602 ts_usec, info->flags & LOG_CONT ? 'c' : '-', caller);
605 static ssize_t msg_add_ext_text(char *buf, size_t size,
606 const char *text, size_t text_len,
609 char *p = buf, *e = buf + size;
612 /* escape non-printable characters */
613 for (i = 0; i < text_len; i++) {
614 unsigned char c = text[i];
616 if (c < ' ' || c >= 127 || c == '\\')
617 p += scnprintf(p, e - p, "\\x%02x", c);
619 append_char(&p, e, c);
621 append_char(&p, e, endc);
626 static ssize_t msg_add_dict_text(char *buf, size_t size,
627 const char *key, const char *val)
629 size_t val_len = strlen(val);
635 len = msg_add_ext_text(buf, size, "", 0, ' '); /* dict prefix */
636 len += msg_add_ext_text(buf + len, size - len, key, strlen(key), '=');
637 len += msg_add_ext_text(buf + len, size - len, val, val_len, '\n');
642 static ssize_t msg_print_ext_body(char *buf, size_t size,
643 char *text, size_t text_len,
644 struct dev_printk_info *dev_info)
648 len = msg_add_ext_text(buf, size, text, text_len, '\n');
653 len += msg_add_dict_text(buf + len, size - len, "SUBSYSTEM",
654 dev_info->subsystem);
655 len += msg_add_dict_text(buf + len, size - len, "DEVICE",
661 /* /dev/kmsg - userspace message inject/listen interface */
662 struct devkmsg_user {
664 struct ratelimit_state rs;
666 char buf[CONSOLE_EXT_LOG_MAX];
668 struct printk_info info;
669 char text_buf[CONSOLE_EXT_LOG_MAX];
670 struct printk_record record;
673 static __printf(3, 4) __cold
674 int devkmsg_emit(int facility, int level, const char *fmt, ...)
680 r = vprintk_emit(facility, level, NULL, fmt, args);
686 static ssize_t devkmsg_write(struct kiocb *iocb, struct iov_iter *from)
689 int level = default_message_loglevel;
690 int facility = 1; /* LOG_USER */
691 struct file *file = iocb->ki_filp;
692 struct devkmsg_user *user = file->private_data;
693 size_t len = iov_iter_count(from);
696 if (!user || len > LOG_LINE_MAX)
699 /* Ignore when user logging is disabled. */
700 if (devkmsg_log & DEVKMSG_LOG_MASK_OFF)
703 /* Ratelimit when not explicitly enabled. */
704 if (!(devkmsg_log & DEVKMSG_LOG_MASK_ON)) {
705 if (!___ratelimit(&user->rs, current->comm))
709 buf = kmalloc(len+1, GFP_KERNEL);
714 if (!copy_from_iter_full(buf, len, from)) {
720 * Extract and skip the syslog prefix <[0-9]*>. Coming from userspace
721 * the decimal value represents 32bit, the lower 3 bit are the log
722 * level, the rest are the log facility.
724 * If no prefix or no userspace facility is specified, we
725 * enforce LOG_USER, to be able to reliably distinguish
726 * kernel-generated messages from userspace-injected ones.
729 if (line[0] == '<') {
733 u = simple_strtoul(line + 1, &endp, 10);
734 if (endp && endp[0] == '>') {
735 level = LOG_LEVEL(u);
736 if (LOG_FACILITY(u) != 0)
737 facility = LOG_FACILITY(u);
743 devkmsg_emit(facility, level, "%s", line);
748 static ssize_t devkmsg_read(struct file *file, char __user *buf,
749 size_t count, loff_t *ppos)
751 struct devkmsg_user *user = file->private_data;
752 struct printk_record *r = &user->record;
759 ret = mutex_lock_interruptible(&user->lock);
763 if (!prb_read_valid(prb, atomic64_read(&user->seq), r)) {
764 if (file->f_flags & O_NONBLOCK) {
770 * Guarantee this task is visible on the waitqueue before
771 * checking the wake condition.
773 * The full memory barrier within set_current_state() of
774 * prepare_to_wait_event() pairs with the full memory barrier
775 * within wq_has_sleeper().
777 * This pairs with __wake_up_klogd:A.
779 ret = wait_event_interruptible(log_wait,
781 atomic64_read(&user->seq), r)); /* LMM(devkmsg_read:A) */
786 if (r->info->seq != atomic64_read(&user->seq)) {
787 /* our last seen message is gone, return error and reset */
788 atomic64_set(&user->seq, r->info->seq);
793 len = info_print_ext_header(user->buf, sizeof(user->buf), r->info);
794 len += msg_print_ext_body(user->buf + len, sizeof(user->buf) - len,
795 &r->text_buf[0], r->info->text_len,
798 atomic64_set(&user->seq, r->info->seq + 1);
805 if (copy_to_user(buf, user->buf, len)) {
811 mutex_unlock(&user->lock);
816 * Be careful when modifying this function!!!
818 * Only few operations are supported because the device works only with the
819 * entire variable length messages (records). Non-standard values are
820 * returned in the other cases and has been this way for quite some time.
821 * User space applications might depend on this behavior.
823 static loff_t devkmsg_llseek(struct file *file, loff_t offset, int whence)
825 struct devkmsg_user *user = file->private_data;
835 /* the first record */
836 atomic64_set(&user->seq, prb_first_valid_seq(prb));
840 * The first record after the last SYSLOG_ACTION_CLEAR,
841 * like issued by 'dmesg -c'. Reading /dev/kmsg itself
842 * changes no global state, and does not clear anything.
844 atomic64_set(&user->seq, latched_seq_read_nolock(&clear_seq));
847 /* after the last record */
848 atomic64_set(&user->seq, prb_next_seq(prb));
856 static __poll_t devkmsg_poll(struct file *file, poll_table *wait)
858 struct devkmsg_user *user = file->private_data;
859 struct printk_info info;
863 return EPOLLERR|EPOLLNVAL;
865 poll_wait(file, &log_wait, wait);
867 if (prb_read_valid_info(prb, atomic64_read(&user->seq), &info, NULL)) {
868 /* return error when data has vanished underneath us */
869 if (info.seq != atomic64_read(&user->seq))
870 ret = EPOLLIN|EPOLLRDNORM|EPOLLERR|EPOLLPRI;
872 ret = EPOLLIN|EPOLLRDNORM;
878 static int devkmsg_open(struct inode *inode, struct file *file)
880 struct devkmsg_user *user;
883 if (devkmsg_log & DEVKMSG_LOG_MASK_OFF)
886 /* write-only does not need any file context */
887 if ((file->f_flags & O_ACCMODE) != O_WRONLY) {
888 err = check_syslog_permissions(SYSLOG_ACTION_READ_ALL,
894 user = kvmalloc(sizeof(struct devkmsg_user), GFP_KERNEL);
898 ratelimit_default_init(&user->rs);
899 ratelimit_set_flags(&user->rs, RATELIMIT_MSG_ON_RELEASE);
901 mutex_init(&user->lock);
903 prb_rec_init_rd(&user->record, &user->info,
904 &user->text_buf[0], sizeof(user->text_buf));
906 atomic64_set(&user->seq, prb_first_valid_seq(prb));
908 file->private_data = user;
912 static int devkmsg_release(struct inode *inode, struct file *file)
914 struct devkmsg_user *user = file->private_data;
919 ratelimit_state_exit(&user->rs);
921 mutex_destroy(&user->lock);
926 const struct file_operations kmsg_fops = {
927 .open = devkmsg_open,
928 .read = devkmsg_read,
929 .write_iter = devkmsg_write,
930 .llseek = devkmsg_llseek,
931 .poll = devkmsg_poll,
932 .release = devkmsg_release,
935 #ifdef CONFIG_CRASH_CORE
937 * This appends the listed symbols to /proc/vmcore
939 * /proc/vmcore is used by various utilities, like crash and makedumpfile to
940 * obtain access to symbols that are otherwise very difficult to locate. These
941 * symbols are specifically used so that utilities can access and extract the
942 * dmesg log from a vmcore file after a crash.
944 void log_buf_vmcoreinfo_setup(void)
946 struct dev_printk_info *dev_info = NULL;
948 VMCOREINFO_SYMBOL(prb);
949 VMCOREINFO_SYMBOL(printk_rb_static);
950 VMCOREINFO_SYMBOL(clear_seq);
953 * Export struct size and field offsets. User space tools can
954 * parse it and detect any changes to structure down the line.
957 VMCOREINFO_STRUCT_SIZE(printk_ringbuffer);
958 VMCOREINFO_OFFSET(printk_ringbuffer, desc_ring);
959 VMCOREINFO_OFFSET(printk_ringbuffer, text_data_ring);
960 VMCOREINFO_OFFSET(printk_ringbuffer, fail);
962 VMCOREINFO_STRUCT_SIZE(prb_desc_ring);
963 VMCOREINFO_OFFSET(prb_desc_ring, count_bits);
964 VMCOREINFO_OFFSET(prb_desc_ring, descs);
965 VMCOREINFO_OFFSET(prb_desc_ring, infos);
966 VMCOREINFO_OFFSET(prb_desc_ring, head_id);
967 VMCOREINFO_OFFSET(prb_desc_ring, tail_id);
969 VMCOREINFO_STRUCT_SIZE(prb_desc);
970 VMCOREINFO_OFFSET(prb_desc, state_var);
971 VMCOREINFO_OFFSET(prb_desc, text_blk_lpos);
973 VMCOREINFO_STRUCT_SIZE(prb_data_blk_lpos);
974 VMCOREINFO_OFFSET(prb_data_blk_lpos, begin);
975 VMCOREINFO_OFFSET(prb_data_blk_lpos, next);
977 VMCOREINFO_STRUCT_SIZE(printk_info);
978 VMCOREINFO_OFFSET(printk_info, seq);
979 VMCOREINFO_OFFSET(printk_info, ts_nsec);
980 VMCOREINFO_OFFSET(printk_info, text_len);
981 VMCOREINFO_OFFSET(printk_info, caller_id);
982 VMCOREINFO_OFFSET(printk_info, dev_info);
984 VMCOREINFO_STRUCT_SIZE(dev_printk_info);
985 VMCOREINFO_OFFSET(dev_printk_info, subsystem);
986 VMCOREINFO_LENGTH(printk_info_subsystem, sizeof(dev_info->subsystem));
987 VMCOREINFO_OFFSET(dev_printk_info, device);
988 VMCOREINFO_LENGTH(printk_info_device, sizeof(dev_info->device));
990 VMCOREINFO_STRUCT_SIZE(prb_data_ring);
991 VMCOREINFO_OFFSET(prb_data_ring, size_bits);
992 VMCOREINFO_OFFSET(prb_data_ring, data);
993 VMCOREINFO_OFFSET(prb_data_ring, head_lpos);
994 VMCOREINFO_OFFSET(prb_data_ring, tail_lpos);
996 VMCOREINFO_SIZE(atomic_long_t);
997 VMCOREINFO_TYPE_OFFSET(atomic_long_t, counter);
999 VMCOREINFO_STRUCT_SIZE(latched_seq);
1000 VMCOREINFO_OFFSET(latched_seq, val);
1004 /* requested log_buf_len from kernel cmdline */
1005 static unsigned long __initdata new_log_buf_len;
1007 /* we practice scaling the ring buffer by powers of 2 */
1008 static void __init log_buf_len_update(u64 size)
1010 if (size > (u64)LOG_BUF_LEN_MAX) {
1011 size = (u64)LOG_BUF_LEN_MAX;
1012 pr_err("log_buf over 2G is not supported.\n");
1016 size = roundup_pow_of_two(size);
1017 if (size > log_buf_len)
1018 new_log_buf_len = (unsigned long)size;
1021 /* save requested log_buf_len since it's too early to process it */
1022 static int __init log_buf_len_setup(char *str)
1029 size = memparse(str, &str);
1031 log_buf_len_update(size);
1035 early_param("log_buf_len", log_buf_len_setup);
1038 #define __LOG_CPU_MAX_BUF_LEN (1 << CONFIG_LOG_CPU_MAX_BUF_SHIFT)
1040 static void __init log_buf_add_cpu(void)
1042 unsigned int cpu_extra;
1045 * archs should set up cpu_possible_bits properly with
1046 * set_cpu_possible() after setup_arch() but just in
1047 * case lets ensure this is valid.
1049 if (num_possible_cpus() == 1)
1052 cpu_extra = (num_possible_cpus() - 1) * __LOG_CPU_MAX_BUF_LEN;
1054 /* by default this will only continue through for large > 64 CPUs */
1055 if (cpu_extra <= __LOG_BUF_LEN / 2)
1058 pr_info("log_buf_len individual max cpu contribution: %d bytes\n",
1059 __LOG_CPU_MAX_BUF_LEN);
1060 pr_info("log_buf_len total cpu_extra contributions: %d bytes\n",
1062 pr_info("log_buf_len min size: %d bytes\n", __LOG_BUF_LEN);
1064 log_buf_len_update(cpu_extra + __LOG_BUF_LEN);
1066 #else /* !CONFIG_SMP */
1067 static inline void log_buf_add_cpu(void) {}
1068 #endif /* CONFIG_SMP */
1070 static void __init set_percpu_data_ready(void)
1072 __printk_percpu_data_ready = true;
1075 static unsigned int __init add_to_rb(struct printk_ringbuffer *rb,
1076 struct printk_record *r)
1078 struct prb_reserved_entry e;
1079 struct printk_record dest_r;
1081 prb_rec_init_wr(&dest_r, r->info->text_len);
1083 if (!prb_reserve(&e, rb, &dest_r))
1086 memcpy(&dest_r.text_buf[0], &r->text_buf[0], r->info->text_len);
1087 dest_r.info->text_len = r->info->text_len;
1088 dest_r.info->facility = r->info->facility;
1089 dest_r.info->level = r->info->level;
1090 dest_r.info->flags = r->info->flags;
1091 dest_r.info->ts_nsec = r->info->ts_nsec;
1092 dest_r.info->caller_id = r->info->caller_id;
1093 memcpy(&dest_r.info->dev_info, &r->info->dev_info, sizeof(dest_r.info->dev_info));
1095 prb_final_commit(&e);
1097 return prb_record_text_space(&e);
1100 static char setup_text_buf[LOG_LINE_MAX] __initdata;
1102 void __init setup_log_buf(int early)
1104 struct printk_info *new_infos;
1105 unsigned int new_descs_count;
1106 struct prb_desc *new_descs;
1107 struct printk_info info;
1108 struct printk_record r;
1109 unsigned int text_size;
1110 size_t new_descs_size;
1111 size_t new_infos_size;
1112 unsigned long flags;
1118 * Some archs call setup_log_buf() multiple times - first is very
1119 * early, e.g. from setup_arch(), and second - when percpu_areas
1123 set_percpu_data_ready();
1125 if (log_buf != __log_buf)
1128 if (!early && !new_log_buf_len)
1131 if (!new_log_buf_len)
1134 new_descs_count = new_log_buf_len >> PRB_AVGBITS;
1135 if (new_descs_count == 0) {
1136 pr_err("new_log_buf_len: %lu too small\n", new_log_buf_len);
1140 new_log_buf = memblock_alloc(new_log_buf_len, LOG_ALIGN);
1141 if (unlikely(!new_log_buf)) {
1142 pr_err("log_buf_len: %lu text bytes not available\n",
1147 new_descs_size = new_descs_count * sizeof(struct prb_desc);
1148 new_descs = memblock_alloc(new_descs_size, LOG_ALIGN);
1149 if (unlikely(!new_descs)) {
1150 pr_err("log_buf_len: %zu desc bytes not available\n",
1152 goto err_free_log_buf;
1155 new_infos_size = new_descs_count * sizeof(struct printk_info);
1156 new_infos = memblock_alloc(new_infos_size, LOG_ALIGN);
1157 if (unlikely(!new_infos)) {
1158 pr_err("log_buf_len: %zu info bytes not available\n",
1160 goto err_free_descs;
1163 prb_rec_init_rd(&r, &info, &setup_text_buf[0], sizeof(setup_text_buf));
1165 prb_init(&printk_rb_dynamic,
1166 new_log_buf, ilog2(new_log_buf_len),
1167 new_descs, ilog2(new_descs_count),
1170 local_irq_save(flags);
1172 log_buf_len = new_log_buf_len;
1173 log_buf = new_log_buf;
1174 new_log_buf_len = 0;
1176 free = __LOG_BUF_LEN;
1177 prb_for_each_record(0, &printk_rb_static, seq, &r) {
1178 text_size = add_to_rb(&printk_rb_dynamic, &r);
1179 if (text_size > free)
1185 prb = &printk_rb_dynamic;
1187 local_irq_restore(flags);
1190 * Copy any remaining messages that might have appeared from
1191 * NMI context after copying but before switching to the
1194 prb_for_each_record(seq, &printk_rb_static, seq, &r) {
1195 text_size = add_to_rb(&printk_rb_dynamic, &r);
1196 if (text_size > free)
1202 if (seq != prb_next_seq(&printk_rb_static)) {
1203 pr_err("dropped %llu messages\n",
1204 prb_next_seq(&printk_rb_static) - seq);
1207 pr_info("log_buf_len: %u bytes\n", log_buf_len);
1208 pr_info("early log buf free: %u(%u%%)\n",
1209 free, (free * 100) / __LOG_BUF_LEN);
1213 memblock_free(new_descs, new_descs_size);
1215 memblock_free(new_log_buf, new_log_buf_len);
1218 static bool __read_mostly ignore_loglevel;
1220 static int __init ignore_loglevel_setup(char *str)
1222 ignore_loglevel = true;
1223 pr_info("debug: ignoring loglevel setting.\n");
1228 early_param("ignore_loglevel", ignore_loglevel_setup);
1229 module_param(ignore_loglevel, bool, S_IRUGO | S_IWUSR);
1230 MODULE_PARM_DESC(ignore_loglevel,
1231 "ignore loglevel setting (prints all kernel messages to the console)");
1233 static bool suppress_message_printing(int level)
1235 return (level >= console_loglevel && !ignore_loglevel);
1238 #ifdef CONFIG_BOOT_PRINTK_DELAY
1240 static int boot_delay; /* msecs delay after each printk during bootup */
1241 static unsigned long long loops_per_msec; /* based on boot_delay */
1243 static int __init boot_delay_setup(char *str)
1247 lpj = preset_lpj ? preset_lpj : 1000000; /* some guess */
1248 loops_per_msec = (unsigned long long)lpj / 1000 * HZ;
1250 get_option(&str, &boot_delay);
1251 if (boot_delay > 10 * 1000)
1254 pr_debug("boot_delay: %u, preset_lpj: %ld, lpj: %lu, "
1255 "HZ: %d, loops_per_msec: %llu\n",
1256 boot_delay, preset_lpj, lpj, HZ, loops_per_msec);
1259 early_param("boot_delay", boot_delay_setup);
1261 static void boot_delay_msec(int level)
1263 unsigned long long k;
1264 unsigned long timeout;
1266 if ((boot_delay == 0 || system_state >= SYSTEM_RUNNING)
1267 || suppress_message_printing(level)) {
1271 k = (unsigned long long)loops_per_msec * boot_delay;
1273 timeout = jiffies + msecs_to_jiffies(boot_delay);
1278 * use (volatile) jiffies to prevent
1279 * compiler reduction; loop termination via jiffies
1280 * is secondary and may or may not happen.
1282 if (time_after(jiffies, timeout))
1284 touch_nmi_watchdog();
1288 static inline void boot_delay_msec(int level)
1293 static bool printk_time = IS_ENABLED(CONFIG_PRINTK_TIME);
1294 module_param_named(time, printk_time, bool, S_IRUGO | S_IWUSR);
1296 static size_t print_syslog(unsigned int level, char *buf)
1298 return sprintf(buf, "<%u>", level);
1301 static size_t print_time(u64 ts, char *buf)
1303 unsigned long rem_nsec = do_div(ts, 1000000000);
1305 return sprintf(buf, "[%5lu.%06lu]",
1306 (unsigned long)ts, rem_nsec / 1000);
1309 #ifdef CONFIG_PRINTK_CALLER
1310 static size_t print_caller(u32 id, char *buf)
1314 snprintf(caller, sizeof(caller), "%c%u",
1315 id & 0x80000000 ? 'C' : 'T', id & ~0x80000000);
1316 return sprintf(buf, "[%6s]", caller);
1319 #define print_caller(id, buf) 0
1322 static size_t info_print_prefix(const struct printk_info *info, bool syslog,
1323 bool time, char *buf)
1328 len = print_syslog((info->facility << 3) | info->level, buf);
1331 len += print_time(info->ts_nsec, buf + len);
1333 len += print_caller(info->caller_id, buf + len);
1335 if (IS_ENABLED(CONFIG_PRINTK_CALLER) || time) {
1344 * Prepare the record for printing. The text is shifted within the given
1345 * buffer to avoid a need for another one. The following operations are
1348 * - Add prefix for each line.
1349 * - Drop truncated lines that no longer fit into the buffer.
1350 * - Add the trailing newline that has been removed in vprintk_store().
1351 * - Add a string terminator.
1353 * Since the produced string is always terminated, the maximum possible
1354 * return value is @r->text_buf_size - 1;
1356 * Return: The length of the updated/prepared text, including the added
1357 * prefixes and the newline. The terminator is not counted. The dropped
1358 * line(s) are not counted.
1360 static size_t record_print_text(struct printk_record *r, bool syslog,
1363 size_t text_len = r->info->text_len;
1364 size_t buf_size = r->text_buf_size;
1365 char *text = r->text_buf;
1366 char prefix[PREFIX_MAX];
1367 bool truncated = false;
1374 * If the message was truncated because the buffer was not large
1375 * enough, treat the available text as if it were the full text.
1377 if (text_len > buf_size)
1378 text_len = buf_size;
1380 prefix_len = info_print_prefix(r->info, syslog, time, prefix);
1383 * @text_len: bytes of unprocessed text
1384 * @line_len: bytes of current line _without_ newline
1385 * @text: pointer to beginning of current line
1386 * @len: number of bytes prepared in r->text_buf
1389 next = memchr(text, '\n', text_len);
1391 line_len = next - text;
1393 /* Drop truncated line(s). */
1396 line_len = text_len;
1400 * Truncate the text if there is not enough space to add the
1401 * prefix and a trailing newline and a terminator.
1403 if (len + prefix_len + text_len + 1 + 1 > buf_size) {
1404 /* Drop even the current line if no space. */
1405 if (len + prefix_len + line_len + 1 + 1 > buf_size)
1408 text_len = buf_size - len - prefix_len - 1 - 1;
1412 memmove(text + prefix_len, text, text_len);
1413 memcpy(text, prefix, prefix_len);
1416 * Increment the prepared length to include the text and
1417 * prefix that were just moved+copied. Also increment for the
1418 * newline at the end of this line. If this is the last line,
1419 * there is no newline, but it will be added immediately below.
1421 len += prefix_len + line_len + 1;
1422 if (text_len == line_len) {
1424 * This is the last line. Add the trailing newline
1425 * removed in vprintk_store().
1427 text[prefix_len + line_len] = '\n';
1432 * Advance beyond the added prefix and the related line with
1435 text += prefix_len + line_len + 1;
1438 * The remaining text has only decreased by the line with its
1441 * Note that @text_len can become zero. It happens when @text
1442 * ended with a newline (either due to truncation or the
1443 * original string ending with "\n\n"). The loop is correctly
1444 * repeated and (if not truncated) an empty line with a prefix
1447 text_len -= line_len + 1;
1451 * If a buffer was provided, it will be terminated. Space for the
1452 * string terminator is guaranteed to be available. The terminator is
1453 * not counted in the return value.
1456 r->text_buf[len] = 0;
1461 static size_t get_record_print_text_size(struct printk_info *info,
1462 unsigned int line_count,
1463 bool syslog, bool time)
1465 char prefix[PREFIX_MAX];
1468 prefix_len = info_print_prefix(info, syslog, time, prefix);
1471 * Each line will be preceded with a prefix. The intermediate
1472 * newlines are already within the text, but a final trailing
1473 * newline will be added.
1475 return ((prefix_len * line_count) + info->text_len + 1);
1479 * Beginning with @start_seq, find the first record where it and all following
1480 * records up to (but not including) @max_seq fit into @size.
1482 * @max_seq is simply an upper bound and does not need to exist. If the caller
1483 * does not require an upper bound, -1 can be used for @max_seq.
1485 static u64 find_first_fitting_seq(u64 start_seq, u64 max_seq, size_t size,
1486 bool syslog, bool time)
1488 struct printk_info info;
1489 unsigned int line_count;
1493 /* Determine the size of the records up to @max_seq. */
1494 prb_for_each_info(start_seq, prb, seq, &info, &line_count) {
1495 if (info.seq >= max_seq)
1497 len += get_record_print_text_size(&info, line_count, syslog, time);
1501 * Adjust the upper bound for the next loop to avoid subtracting
1502 * lengths that were never added.
1508 * Move first record forward until length fits into the buffer. Ignore
1509 * newest messages that were not counted in the above cycle. Messages
1510 * might appear and get lost in the meantime. This is a best effort
1511 * that prevents an infinite loop that could occur with a retry.
1513 prb_for_each_info(start_seq, prb, seq, &info, &line_count) {
1514 if (len <= size || info.seq >= max_seq)
1516 len -= get_record_print_text_size(&info, line_count, syslog, time);
1522 /* The caller is responsible for making sure @size is greater than 0. */
1523 static int syslog_print(char __user *buf, int size)
1525 struct printk_info info;
1526 struct printk_record r;
1531 text = kmalloc(CONSOLE_LOG_MAX, GFP_KERNEL);
1535 prb_rec_init_rd(&r, &info, text, CONSOLE_LOG_MAX);
1537 mutex_lock(&syslog_lock);
1540 * Wait for the @syslog_seq record to be available. @syslog_seq may
1541 * change while waiting.
1546 mutex_unlock(&syslog_lock);
1548 * Guarantee this task is visible on the waitqueue before
1549 * checking the wake condition.
1551 * The full memory barrier within set_current_state() of
1552 * prepare_to_wait_event() pairs with the full memory barrier
1553 * within wq_has_sleeper().
1555 * This pairs with __wake_up_klogd:A.
1557 len = wait_event_interruptible(log_wait,
1558 prb_read_valid(prb, seq, NULL)); /* LMM(syslog_print:A) */
1559 mutex_lock(&syslog_lock);
1563 } while (syslog_seq != seq);
1566 * Copy records that fit into the buffer. The above cycle makes sure
1567 * that the first record is always available.
1574 if (!prb_read_valid(prb, syslog_seq, &r))
1577 if (r.info->seq != syslog_seq) {
1578 /* message is gone, move to next valid one */
1579 syslog_seq = r.info->seq;
1584 * To keep reading/counting partial line consistent,
1585 * use printk_time value as of the beginning of a line.
1587 if (!syslog_partial)
1588 syslog_time = printk_time;
1590 skip = syslog_partial;
1591 n = record_print_text(&r, true, syslog_time);
1592 if (n - syslog_partial <= size) {
1593 /* message fits into buffer, move forward */
1594 syslog_seq = r.info->seq + 1;
1595 n -= syslog_partial;
1598 /* partial read(), remember position */
1600 syslog_partial += n;
1607 mutex_unlock(&syslog_lock);
1608 err = copy_to_user(buf, text + skip, n);
1609 mutex_lock(&syslog_lock);
1622 mutex_unlock(&syslog_lock);
1627 static int syslog_print_all(char __user *buf, int size, bool clear)
1629 struct printk_info info;
1630 struct printk_record r;
1636 text = kmalloc(CONSOLE_LOG_MAX, GFP_KERNEL);
1642 * Find first record that fits, including all following records,
1643 * into the user-provided buffer for this dump.
1645 seq = find_first_fitting_seq(latched_seq_read_nolock(&clear_seq), -1,
1648 prb_rec_init_rd(&r, &info, text, CONSOLE_LOG_MAX);
1651 prb_for_each_record(seq, prb, seq, &r) {
1654 textlen = record_print_text(&r, true, time);
1656 if (len + textlen > size) {
1661 if (copy_to_user(buf + len, text, textlen))
1671 mutex_lock(&syslog_lock);
1672 latched_seq_write(&clear_seq, seq);
1673 mutex_unlock(&syslog_lock);
1680 static void syslog_clear(void)
1682 mutex_lock(&syslog_lock);
1683 latched_seq_write(&clear_seq, prb_next_seq(prb));
1684 mutex_unlock(&syslog_lock);
1687 int do_syslog(int type, char __user *buf, int len, int source)
1689 struct printk_info info;
1691 static int saved_console_loglevel = LOGLEVEL_DEFAULT;
1694 error = check_syslog_permissions(type, source);
1699 case SYSLOG_ACTION_CLOSE: /* Close log */
1701 case SYSLOG_ACTION_OPEN: /* Open log */
1703 case SYSLOG_ACTION_READ: /* Read from log */
1704 if (!buf || len < 0)
1708 if (!access_ok(buf, len))
1710 error = syslog_print(buf, len);
1712 /* Read/clear last kernel messages */
1713 case SYSLOG_ACTION_READ_CLEAR:
1716 /* Read last kernel messages */
1717 case SYSLOG_ACTION_READ_ALL:
1718 if (!buf || len < 0)
1722 if (!access_ok(buf, len))
1724 error = syslog_print_all(buf, len, clear);
1726 /* Clear ring buffer */
1727 case SYSLOG_ACTION_CLEAR:
1730 /* Disable logging to console */
1731 case SYSLOG_ACTION_CONSOLE_OFF:
1732 if (saved_console_loglevel == LOGLEVEL_DEFAULT)
1733 saved_console_loglevel = console_loglevel;
1734 console_loglevel = minimum_console_loglevel;
1736 /* Enable logging to console */
1737 case SYSLOG_ACTION_CONSOLE_ON:
1738 if (saved_console_loglevel != LOGLEVEL_DEFAULT) {
1739 console_loglevel = saved_console_loglevel;
1740 saved_console_loglevel = LOGLEVEL_DEFAULT;
1743 /* Set level of messages printed to console */
1744 case SYSLOG_ACTION_CONSOLE_LEVEL:
1745 if (len < 1 || len > 8)
1747 if (len < minimum_console_loglevel)
1748 len = minimum_console_loglevel;
1749 console_loglevel = len;
1750 /* Implicitly re-enable logging to console */
1751 saved_console_loglevel = LOGLEVEL_DEFAULT;
1753 /* Number of chars in the log buffer */
1754 case SYSLOG_ACTION_SIZE_UNREAD:
1755 mutex_lock(&syslog_lock);
1756 if (!prb_read_valid_info(prb, syslog_seq, &info, NULL)) {
1757 /* No unread messages. */
1758 mutex_unlock(&syslog_lock);
1761 if (info.seq != syslog_seq) {
1762 /* messages are gone, move to first one */
1763 syslog_seq = info.seq;
1766 if (source == SYSLOG_FROM_PROC) {
1768 * Short-cut for poll(/"proc/kmsg") which simply checks
1769 * for pending data, not the size; return the count of
1770 * records, not the length.
1772 error = prb_next_seq(prb) - syslog_seq;
1774 bool time = syslog_partial ? syslog_time : printk_time;
1775 unsigned int line_count;
1778 prb_for_each_info(syslog_seq, prb, seq, &info,
1780 error += get_record_print_text_size(&info, line_count,
1784 error -= syslog_partial;
1786 mutex_unlock(&syslog_lock);
1788 /* Size of the log buffer */
1789 case SYSLOG_ACTION_SIZE_BUFFER:
1790 error = log_buf_len;
1800 SYSCALL_DEFINE3(syslog, int, type, char __user *, buf, int, len)
1802 return do_syslog(type, buf, len, SYSLOG_FROM_READER);
1806 * Special console_lock variants that help to reduce the risk of soft-lockups.
1807 * They allow to pass console_lock to another printk() call using a busy wait.
1810 #ifdef CONFIG_LOCKDEP
1811 static struct lockdep_map console_owner_dep_map = {
1812 .name = "console_owner"
1816 static DEFINE_RAW_SPINLOCK(console_owner_lock);
1817 static struct task_struct *console_owner;
1818 static bool console_waiter;
1821 * console_lock_spinning_enable - mark beginning of code where another
1822 * thread might safely busy wait
1824 * This basically converts console_lock into a spinlock. This marks
1825 * the section where the console_lock owner can not sleep, because
1826 * there may be a waiter spinning (like a spinlock). Also it must be
1827 * ready to hand over the lock at the end of the section.
1829 static void console_lock_spinning_enable(void)
1831 raw_spin_lock(&console_owner_lock);
1832 console_owner = current;
1833 raw_spin_unlock(&console_owner_lock);
1835 /* The waiter may spin on us after setting console_owner */
1836 spin_acquire(&console_owner_dep_map, 0, 0, _THIS_IP_);
1840 * console_lock_spinning_disable_and_check - mark end of code where another
1841 * thread was able to busy wait and check if there is a waiter
1843 * This is called at the end of the section where spinning is allowed.
1844 * It has two functions. First, it is a signal that it is no longer
1845 * safe to start busy waiting for the lock. Second, it checks if
1846 * there is a busy waiter and passes the lock rights to her.
1848 * Important: Callers lose the lock if there was a busy waiter.
1849 * They must not touch items synchronized by console_lock
1852 * Return: 1 if the lock rights were passed, 0 otherwise.
1854 static int console_lock_spinning_disable_and_check(void)
1858 raw_spin_lock(&console_owner_lock);
1859 waiter = READ_ONCE(console_waiter);
1860 console_owner = NULL;
1861 raw_spin_unlock(&console_owner_lock);
1864 spin_release(&console_owner_dep_map, _THIS_IP_);
1868 /* The waiter is now free to continue */
1869 WRITE_ONCE(console_waiter, false);
1871 spin_release(&console_owner_dep_map, _THIS_IP_);
1874 * Hand off console_lock to waiter. The waiter will perform
1875 * the up(). After this, the waiter is the console_lock owner.
1877 mutex_release(&console_lock_dep_map, _THIS_IP_);
1882 * console_trylock_spinning - try to get console_lock by busy waiting
1884 * This allows to busy wait for the console_lock when the current
1885 * owner is running in specially marked sections. It means that
1886 * the current owner is running and cannot reschedule until it
1887 * is ready to lose the lock.
1889 * Return: 1 if we got the lock, 0 othrewise
1891 static int console_trylock_spinning(void)
1893 struct task_struct *owner = NULL;
1896 unsigned long flags;
1898 if (console_trylock())
1902 * It's unsafe to spin once a panic has begun. If we are the
1903 * panic CPU, we may have already halted the owner of the
1904 * console_sem. If we are not the panic CPU, then we should
1905 * avoid taking console_sem, so the panic CPU has a better
1906 * chance of cleanly acquiring it later.
1908 if (panic_in_progress())
1911 printk_safe_enter_irqsave(flags);
1913 raw_spin_lock(&console_owner_lock);
1914 owner = READ_ONCE(console_owner);
1915 waiter = READ_ONCE(console_waiter);
1916 if (!waiter && owner && owner != current) {
1917 WRITE_ONCE(console_waiter, true);
1920 raw_spin_unlock(&console_owner_lock);
1923 * If there is an active printk() writing to the
1924 * consoles, instead of having it write our data too,
1925 * see if we can offload that load from the active
1926 * printer, and do some printing ourselves.
1927 * Go into a spin only if there isn't already a waiter
1928 * spinning, and there is an active printer, and
1929 * that active printer isn't us (recursive printk?).
1932 printk_safe_exit_irqrestore(flags);
1936 /* We spin waiting for the owner to release us */
1937 spin_acquire(&console_owner_dep_map, 0, 0, _THIS_IP_);
1938 /* Owner will clear console_waiter on hand off */
1939 while (READ_ONCE(console_waiter))
1941 spin_release(&console_owner_dep_map, _THIS_IP_);
1943 printk_safe_exit_irqrestore(flags);
1945 * The owner passed the console lock to us.
1946 * Since we did not spin on console lock, annotate
1947 * this as a trylock. Otherwise lockdep will
1950 mutex_acquire(&console_lock_dep_map, 0, 1, _THIS_IP_);
1956 * Call the specified console driver, asking it to write out the specified
1957 * text and length. If @dropped_text is non-NULL and any records have been
1958 * dropped, a dropped message will be written out first.
1960 static void call_console_driver(struct console *con, const char *text, size_t len,
1965 if (con->dropped && dropped_text) {
1966 dropped_len = snprintf(dropped_text, DROPPED_TEXT_MAX,
1967 "** %lu printk messages dropped **\n",
1970 con->write(con, dropped_text, dropped_len);
1973 con->write(con, text, len);
1977 * Recursion is tracked separately on each CPU. If NMIs are supported, an
1978 * additional NMI context per CPU is also separately tracked. Until per-CPU
1979 * is available, a separate "early tracking" is performed.
1981 static DEFINE_PER_CPU(u8, printk_count);
1982 static u8 printk_count_early;
1983 #ifdef CONFIG_HAVE_NMI
1984 static DEFINE_PER_CPU(u8, printk_count_nmi);
1985 static u8 printk_count_nmi_early;
1989 * Recursion is limited to keep the output sane. printk() should not require
1990 * more than 1 level of recursion (allowing, for example, printk() to trigger
1991 * a WARN), but a higher value is used in case some printk-internal errors
1992 * exist, such as the ringbuffer validation checks failing.
1994 #define PRINTK_MAX_RECURSION 3
1997 * Return a pointer to the dedicated counter for the CPU+context of the
2000 static u8 *__printk_recursion_counter(void)
2002 #ifdef CONFIG_HAVE_NMI
2004 if (printk_percpu_data_ready())
2005 return this_cpu_ptr(&printk_count_nmi);
2006 return &printk_count_nmi_early;
2009 if (printk_percpu_data_ready())
2010 return this_cpu_ptr(&printk_count);
2011 return &printk_count_early;
2015 * Enter recursion tracking. Interrupts are disabled to simplify tracking.
2016 * The caller must check the boolean return value to see if the recursion is
2017 * allowed. On failure, interrupts are not disabled.
2019 * @recursion_ptr must be a variable of type (u8 *) and is the same variable
2020 * that is passed to printk_exit_irqrestore().
2022 #define printk_enter_irqsave(recursion_ptr, flags) \
2024 bool success = true; \
2026 typecheck(u8 *, recursion_ptr); \
2027 local_irq_save(flags); \
2028 (recursion_ptr) = __printk_recursion_counter(); \
2029 if (*(recursion_ptr) > PRINTK_MAX_RECURSION) { \
2030 local_irq_restore(flags); \
2033 (*(recursion_ptr))++; \
2038 /* Exit recursion tracking, restoring interrupts. */
2039 #define printk_exit_irqrestore(recursion_ptr, flags) \
2041 typecheck(u8 *, recursion_ptr); \
2042 (*(recursion_ptr))--; \
2043 local_irq_restore(flags); \
2046 int printk_delay_msec __read_mostly;
2048 static inline void printk_delay(int level)
2050 boot_delay_msec(level);
2052 if (unlikely(printk_delay_msec)) {
2053 int m = printk_delay_msec;
2057 touch_nmi_watchdog();
2062 static inline u32 printk_caller_id(void)
2064 return in_task() ? task_pid_nr(current) :
2065 0x80000000 + smp_processor_id();
2069 * printk_parse_prefix - Parse level and control flags.
2071 * @text: The terminated text message.
2072 * @level: A pointer to the current level value, will be updated.
2073 * @flags: A pointer to the current printk_info flags, will be updated.
2075 * @level may be NULL if the caller is not interested in the parsed value.
2076 * Otherwise the variable pointed to by @level must be set to
2077 * LOGLEVEL_DEFAULT in order to be updated with the parsed value.
2079 * @flags may be NULL if the caller is not interested in the parsed value.
2080 * Otherwise the variable pointed to by @flags will be OR'd with the parsed
2083 * Return: The length of the parsed level and control flags.
2085 u16 printk_parse_prefix(const char *text, int *level,
2086 enum printk_info_flags *flags)
2092 kern_level = printk_get_level(text);
2096 switch (kern_level) {
2098 if (level && *level == LOGLEVEL_DEFAULT)
2099 *level = kern_level - '0';
2101 case 'c': /* KERN_CONT */
2114 static u16 printk_sprint(char *text, u16 size, int facility,
2115 enum printk_info_flags *flags, const char *fmt,
2120 text_len = vscnprintf(text, size, fmt, args);
2122 /* Mark and strip a trailing newline. */
2123 if (text_len && text[text_len - 1] == '\n') {
2125 *flags |= LOG_NEWLINE;
2128 /* Strip log level and control flags. */
2129 if (facility == 0) {
2132 prefix_len = printk_parse_prefix(text, NULL, NULL);
2134 text_len -= prefix_len;
2135 memmove(text, text + prefix_len, text_len);
2139 trace_console_rcuidle(text, text_len);
2145 int vprintk_store(int facility, int level,
2146 const struct dev_printk_info *dev_info,
2147 const char *fmt, va_list args)
2149 struct prb_reserved_entry e;
2150 enum printk_info_flags flags = 0;
2151 struct printk_record r;
2152 unsigned long irqflags;
2153 u16 trunc_msg_len = 0;
2163 if (!printk_enter_irqsave(recursion_ptr, irqflags))
2167 * Since the duration of printk() can vary depending on the message
2168 * and state of the ringbuffer, grab the timestamp now so that it is
2169 * close to the call of printk(). This provides a more deterministic
2170 * timestamp with respect to the caller.
2172 ts_nsec = local_clock();
2174 caller_id = printk_caller_id();
2177 * The sprintf needs to come first since the syslog prefix might be
2178 * passed in as a parameter. An extra byte must be reserved so that
2179 * later the vscnprintf() into the reserved buffer has room for the
2180 * terminating '\0', which is not counted by vsnprintf().
2182 va_copy(args2, args);
2183 reserve_size = vsnprintf(&prefix_buf[0], sizeof(prefix_buf), fmt, args2) + 1;
2186 if (reserve_size > LOG_LINE_MAX)
2187 reserve_size = LOG_LINE_MAX;
2189 /* Extract log level or control flags. */
2191 printk_parse_prefix(&prefix_buf[0], &level, &flags);
2193 if (level == LOGLEVEL_DEFAULT)
2194 level = default_message_loglevel;
2197 flags |= LOG_NEWLINE;
2199 if (flags & LOG_CONT) {
2200 prb_rec_init_wr(&r, reserve_size);
2201 if (prb_reserve_in_last(&e, prb, &r, caller_id, LOG_LINE_MAX)) {
2202 text_len = printk_sprint(&r.text_buf[r.info->text_len], reserve_size,
2203 facility, &flags, fmt, args);
2204 r.info->text_len += text_len;
2206 if (flags & LOG_NEWLINE) {
2207 r.info->flags |= LOG_NEWLINE;
2208 prb_final_commit(&e);
2219 * Explicitly initialize the record before every prb_reserve() call.
2220 * prb_reserve_in_last() and prb_reserve() purposely invalidate the
2221 * structure when they fail.
2223 prb_rec_init_wr(&r, reserve_size);
2224 if (!prb_reserve(&e, prb, &r)) {
2225 /* truncate the message if it is too long for empty buffer */
2226 truncate_msg(&reserve_size, &trunc_msg_len);
2228 prb_rec_init_wr(&r, reserve_size + trunc_msg_len);
2229 if (!prb_reserve(&e, prb, &r))
2234 text_len = printk_sprint(&r.text_buf[0], reserve_size, facility, &flags, fmt, args);
2236 memcpy(&r.text_buf[text_len], trunc_msg, trunc_msg_len);
2237 r.info->text_len = text_len + trunc_msg_len;
2238 r.info->facility = facility;
2239 r.info->level = level & 7;
2240 r.info->flags = flags & 0x1f;
2241 r.info->ts_nsec = ts_nsec;
2242 r.info->caller_id = caller_id;
2244 memcpy(&r.info->dev_info, dev_info, sizeof(r.info->dev_info));
2246 /* A message without a trailing newline can be continued. */
2247 if (!(flags & LOG_NEWLINE))
2250 prb_final_commit(&e);
2252 ret = text_len + trunc_msg_len;
2254 printk_exit_irqrestore(recursion_ptr, irqflags);
2258 asmlinkage int vprintk_emit(int facility, int level,
2259 const struct dev_printk_info *dev_info,
2260 const char *fmt, va_list args)
2263 bool in_sched = false;
2265 /* Suppress unimportant messages after panic happens */
2266 if (unlikely(suppress_printk))
2269 if (unlikely(suppress_panic_printk) &&
2270 atomic_read(&panic_cpu) != raw_smp_processor_id())
2273 if (level == LOGLEVEL_SCHED) {
2274 level = LOGLEVEL_DEFAULT;
2278 printk_delay(level);
2280 printed_len = vprintk_store(facility, level, dev_info, fmt, args);
2282 /* If called from the scheduler, we can not call up(). */
2285 * The caller may be holding system-critical or
2286 * timing-sensitive locks. Disable preemption during
2287 * printing of all remaining records to all consoles so that
2288 * this context can return as soon as possible. Hopefully
2289 * another printk() caller will take over the printing.
2293 * Try to acquire and then immediately release the console
2294 * semaphore. The release will print out buffers. With the
2295 * spinning variant, this context tries to take over the
2296 * printing from another printing context.
2298 if (console_trylock_spinning())
2306 EXPORT_SYMBOL(vprintk_emit);
2308 int vprintk_default(const char *fmt, va_list args)
2310 return vprintk_emit(0, LOGLEVEL_DEFAULT, NULL, fmt, args);
2312 EXPORT_SYMBOL_GPL(vprintk_default);
2314 asmlinkage __visible int _printk(const char *fmt, ...)
2319 va_start(args, fmt);
2320 r = vprintk(fmt, args);
2325 EXPORT_SYMBOL(_printk);
2327 static bool __pr_flush(struct console *con, int timeout_ms, bool reset_on_progress);
2329 #else /* CONFIG_PRINTK */
2331 #define CONSOLE_LOG_MAX 0
2332 #define DROPPED_TEXT_MAX 0
2333 #define printk_time false
2335 #define prb_read_valid(rb, seq, r) false
2336 #define prb_first_valid_seq(rb) 0
2337 #define prb_next_seq(rb) 0
2339 static u64 syslog_seq;
2341 static size_t record_print_text(const struct printk_record *r,
2342 bool syslog, bool time)
2346 static ssize_t info_print_ext_header(char *buf, size_t size,
2347 struct printk_info *info)
2351 static ssize_t msg_print_ext_body(char *buf, size_t size,
2352 char *text, size_t text_len,
2353 struct dev_printk_info *dev_info) { return 0; }
2354 static void console_lock_spinning_enable(void) { }
2355 static int console_lock_spinning_disable_and_check(void) { return 0; }
2356 static void call_console_driver(struct console *con, const char *text, size_t len,
2360 static bool suppress_message_printing(int level) { return false; }
2361 static bool __pr_flush(struct console *con, int timeout_ms, bool reset_on_progress) { return true; }
2363 #endif /* CONFIG_PRINTK */
2365 #ifdef CONFIG_EARLY_PRINTK
2366 struct console *early_console;
2368 asmlinkage __visible void early_printk(const char *fmt, ...)
2378 n = vscnprintf(buf, sizeof(buf), fmt, ap);
2381 early_console->write(early_console, buf, n);
2385 static void set_user_specified(struct console_cmdline *c, bool user_specified)
2387 if (!user_specified)
2391 * @c console was defined by the user on the command line.
2392 * Do not clear when added twice also by SPCR or the device tree.
2394 c->user_specified = true;
2395 /* At least one console defined by the user on the command line. */
2396 console_set_on_cmdline = 1;
2399 static int __add_preferred_console(char *name, int idx, char *options,
2400 char *brl_options, bool user_specified)
2402 struct console_cmdline *c;
2406 * See if this tty is not yet registered, and
2407 * if we have a slot free.
2409 for (i = 0, c = console_cmdline;
2410 i < MAX_CMDLINECONSOLES && c->name[0];
2412 if (strcmp(c->name, name) == 0 && c->index == idx) {
2414 preferred_console = i;
2415 set_user_specified(c, user_specified);
2419 if (i == MAX_CMDLINECONSOLES)
2422 preferred_console = i;
2423 strlcpy(c->name, name, sizeof(c->name));
2424 c->options = options;
2425 set_user_specified(c, user_specified);
2426 braille_set_options(c, brl_options);
2432 static int __init console_msg_format_setup(char *str)
2434 if (!strcmp(str, "syslog"))
2435 console_msg_format = MSG_FORMAT_SYSLOG;
2436 if (!strcmp(str, "default"))
2437 console_msg_format = MSG_FORMAT_DEFAULT;
2440 __setup("console_msg_format=", console_msg_format_setup);
2443 * Set up a console. Called via do_early_param() in init/main.c
2444 * for each "console=" parameter in the boot command line.
2446 static int __init console_setup(char *str)
2448 char buf[sizeof(console_cmdline[0].name) + 4]; /* 4 for "ttyS" */
2449 char *s, *options, *brl_options = NULL;
2453 * console="" or console=null have been suggested as a way to
2454 * disable console output. Use ttynull that has been created
2455 * for exactly this purpose.
2457 if (str[0] == 0 || strcmp(str, "null") == 0) {
2458 __add_preferred_console("ttynull", 0, NULL, NULL, true);
2462 if (_braille_console_setup(&str, &brl_options))
2466 * Decode str into name, index, options.
2468 if (str[0] >= '0' && str[0] <= '9') {
2469 strcpy(buf, "ttyS");
2470 strncpy(buf + 4, str, sizeof(buf) - 5);
2472 strncpy(buf, str, sizeof(buf) - 1);
2474 buf[sizeof(buf) - 1] = 0;
2475 options = strchr(str, ',');
2479 if (!strcmp(str, "ttya"))
2480 strcpy(buf, "ttyS0");
2481 if (!strcmp(str, "ttyb"))
2482 strcpy(buf, "ttyS1");
2484 for (s = buf; *s; s++)
2485 if (isdigit(*s) || *s == ',')
2487 idx = simple_strtoul(s, NULL, 10);
2490 __add_preferred_console(buf, idx, options, brl_options, true);
2493 __setup("console=", console_setup);
2496 * add_preferred_console - add a device to the list of preferred consoles.
2497 * @name: device name
2498 * @idx: device index
2499 * @options: options for this console
2501 * The last preferred console added will be used for kernel messages
2502 * and stdin/out/err for init. Normally this is used by console_setup
2503 * above to handle user-supplied console arguments; however it can also
2504 * be used by arch-specific code either to override the user or more
2505 * commonly to provide a default console (ie from PROM variables) when
2506 * the user has not supplied one.
2508 int add_preferred_console(char *name, int idx, char *options)
2510 return __add_preferred_console(name, idx, options, NULL, false);
2513 bool console_suspend_enabled = true;
2514 EXPORT_SYMBOL(console_suspend_enabled);
2516 static int __init console_suspend_disable(char *str)
2518 console_suspend_enabled = false;
2521 __setup("no_console_suspend", console_suspend_disable);
2522 module_param_named(console_suspend, console_suspend_enabled,
2523 bool, S_IRUGO | S_IWUSR);
2524 MODULE_PARM_DESC(console_suspend, "suspend console during suspend"
2525 " and hibernate operations");
2527 static bool printk_console_no_auto_verbose;
2529 void console_verbose(void)
2531 if (console_loglevel && !printk_console_no_auto_verbose)
2532 console_loglevel = CONSOLE_LOGLEVEL_MOTORMOUTH;
2534 EXPORT_SYMBOL_GPL(console_verbose);
2536 module_param_named(console_no_auto_verbose, printk_console_no_auto_verbose, bool, 0644);
2537 MODULE_PARM_DESC(console_no_auto_verbose, "Disable console loglevel raise to highest on oops/panic/etc");
2540 * suspend_console - suspend the console subsystem
2542 * This disables printk() while we go into suspend states
2544 void suspend_console(void)
2546 if (!console_suspend_enabled)
2548 pr_info("Suspending console(s) (use no_console_suspend to debug)\n");
2549 pr_flush(1000, true);
2551 console_suspended = 1;
2555 void resume_console(void)
2557 if (!console_suspend_enabled)
2560 console_suspended = 0;
2562 pr_flush(1000, true);
2566 * console_cpu_notify - print deferred console messages after CPU hotplug
2569 * If printk() is called from a CPU that is not online yet, the messages
2570 * will be printed on the console only if there are CON_ANYTIME consoles.
2571 * This function is called when a new CPU comes online (or fails to come
2572 * up) or goes offline.
2574 static int console_cpu_notify(unsigned int cpu)
2576 if (!cpuhp_tasks_frozen) {
2577 /* If trylock fails, someone else is doing the printing */
2578 if (console_trylock())
2585 * console_lock - lock the console system for exclusive use.
2587 * Acquires a lock which guarantees that the caller has
2588 * exclusive access to the console system and the console_drivers list.
2590 * Can sleep, returns nothing.
2592 void console_lock(void)
2597 if (console_suspended)
2600 console_may_schedule = 1;
2602 EXPORT_SYMBOL(console_lock);
2605 * console_trylock - try to lock the console system for exclusive use.
2607 * Try to acquire a lock which guarantees that the caller has exclusive
2608 * access to the console system and the console_drivers list.
2610 * returns 1 on success, and 0 on failure to acquire the lock.
2612 int console_trylock(void)
2614 if (down_trylock_console_sem())
2616 if (console_suspended) {
2621 console_may_schedule = 0;
2624 EXPORT_SYMBOL(console_trylock);
2626 int is_console_locked(void)
2628 return console_locked;
2630 EXPORT_SYMBOL(is_console_locked);
2633 * Return true when this CPU should unlock console_sem without pushing all
2634 * messages to the console. This reduces the chance that the console is
2635 * locked when the panic CPU tries to use it.
2637 static bool abandon_console_lock_in_panic(void)
2639 if (!panic_in_progress())
2643 * We can use raw_smp_processor_id() here because it is impossible for
2644 * the task to be migrated to the panic_cpu, or away from it. If
2645 * panic_cpu has already been set, and we're not currently executing on
2646 * that CPU, then we never will be.
2648 return atomic_read(&panic_cpu) != raw_smp_processor_id();
2652 * Check if the given console is currently capable and allowed to print
2655 * Requires the console_lock.
2657 static inline bool console_is_usable(struct console *con)
2659 if (!(con->flags & CON_ENABLED))
2666 * Console drivers may assume that per-cpu resources have been
2667 * allocated. So unless they're explicitly marked as being able to
2668 * cope (CON_ANYTIME) don't call them until this CPU is officially up.
2670 if (!cpu_online(raw_smp_processor_id()) &&
2671 !(con->flags & CON_ANYTIME))
2677 static void __console_unlock(void)
2684 * Print one record for the given console. The record printed is whatever
2685 * record is the next available record for the given console.
2687 * @text is a buffer of size CONSOLE_LOG_MAX.
2689 * If extended messages should be printed, @ext_text is a buffer of size
2690 * CONSOLE_EXT_LOG_MAX. Otherwise @ext_text must be NULL.
2692 * If dropped messages should be printed, @dropped_text is a buffer of size
2693 * DROPPED_TEXT_MAX. Otherwise @dropped_text must be NULL.
2695 * @handover will be set to true if a printk waiter has taken over the
2696 * console_lock, in which case the caller is no longer holding the
2697 * console_lock. Otherwise it is set to false.
2699 * Returns false if the given console has no next record to print, otherwise
2702 * Requires the console_lock.
2704 static bool console_emit_next_record(struct console *con, char *text, char *ext_text,
2705 char *dropped_text, bool *handover)
2707 static int panic_console_dropped;
2708 struct printk_info info;
2709 struct printk_record r;
2710 unsigned long flags;
2714 prb_rec_init_rd(&r, &info, text, CONSOLE_LOG_MAX);
2718 if (!prb_read_valid(prb, con->seq, &r))
2721 if (con->seq != r.info->seq) {
2722 con->dropped += r.info->seq - con->seq;
2723 con->seq = r.info->seq;
2724 if (panic_in_progress() && panic_console_dropped++ > 10) {
2725 suppress_panic_printk = 1;
2726 pr_warn_once("Too many dropped messages. Suppress messages on non-panic CPUs to prevent livelock.\n");
2730 /* Skip record that has level above the console loglevel. */
2731 if (suppress_message_printing(r.info->level)) {
2737 write_text = ext_text;
2738 len = info_print_ext_header(ext_text, CONSOLE_EXT_LOG_MAX, r.info);
2739 len += msg_print_ext_body(ext_text + len, CONSOLE_EXT_LOG_MAX - len,
2740 &r.text_buf[0], r.info->text_len, &r.info->dev_info);
2743 len = record_print_text(&r, console_msg_format & MSG_FORMAT_SYSLOG, printk_time);
2747 * While actively printing out messages, if another printk()
2748 * were to occur on another CPU, it may wait for this one to
2749 * finish. This task can not be preempted if there is a
2750 * waiter waiting to take over.
2752 * Interrupts are disabled because the hand over to a waiter
2753 * must not be interrupted until the hand over is completed
2754 * (@console_waiter is cleared).
2756 printk_safe_enter_irqsave(flags);
2757 console_lock_spinning_enable();
2759 stop_critical_timings(); /* don't trace print latency */
2760 call_console_driver(con, write_text, len, dropped_text);
2761 start_critical_timings();
2765 *handover = console_lock_spinning_disable_and_check();
2766 printk_safe_exit_irqrestore(flags);
2772 * Print out all remaining records to all consoles.
2774 * @do_cond_resched is set by the caller. It can be true only in schedulable
2777 * @next_seq is set to the sequence number after the last available record.
2778 * The value is valid only when this function returns true. It means that all
2779 * usable consoles are completely flushed.
2781 * @handover will be set to true if a printk waiter has taken over the
2782 * console_lock, in which case the caller is no longer holding the
2783 * console_lock. Otherwise it is set to false.
2785 * Returns true when there was at least one usable console and all messages
2786 * were flushed to all usable consoles. A returned false informs the caller
2787 * that everything was not flushed (either there were no usable consoles or
2788 * another context has taken over printing or it is a panic situation and this
2789 * is not the panic CPU). Regardless the reason, the caller should assume it
2790 * is not useful to immediately try again.
2792 * Requires the console_lock.
2794 static bool console_flush_all(bool do_cond_resched, u64 *next_seq, bool *handover)
2796 static char dropped_text[DROPPED_TEXT_MAX];
2797 static char ext_text[CONSOLE_EXT_LOG_MAX];
2798 static char text[CONSOLE_LOG_MAX];
2799 bool any_usable = false;
2800 struct console *con;
2807 any_progress = false;
2809 for_each_console(con) {
2812 if (!console_is_usable(con))
2816 if (con->flags & CON_EXTENDED) {
2817 /* Extended consoles do not print "dropped messages". */
2818 progress = console_emit_next_record(con, &text[0],
2822 progress = console_emit_next_record(con, &text[0],
2823 NULL, &dropped_text[0],
2829 /* Track the next of the highest seq flushed. */
2830 if (con->seq > *next_seq)
2831 *next_seq = con->seq;
2835 any_progress = true;
2837 /* Allow panic_cpu to take over the consoles safely. */
2838 if (abandon_console_lock_in_panic())
2841 if (do_cond_resched)
2844 } while (any_progress);
2850 * console_unlock - unlock the console system
2852 * Releases the console_lock which the caller holds on the console system
2853 * and the console driver list.
2855 * While the console_lock was held, console output may have been buffered
2856 * by printk(). If this is the case, console_unlock(); emits
2857 * the output prior to releasing the lock.
2859 * console_unlock(); may be called from any context.
2861 void console_unlock(void)
2863 bool do_cond_resched;
2868 if (console_suspended) {
2874 * Console drivers are called with interrupts disabled, so
2875 * @console_may_schedule should be cleared before; however, we may
2876 * end up dumping a lot of lines, for example, if called from
2877 * console registration path, and should invoke cond_resched()
2878 * between lines if allowable. Not doing so can cause a very long
2879 * scheduling stall on a slow console leading to RCU stall and
2880 * softlockup warnings which exacerbate the issue with more
2881 * messages practically incapacitating the system. Therefore, create
2882 * a local to use for the printing loop.
2884 do_cond_resched = console_may_schedule;
2887 console_may_schedule = 0;
2889 flushed = console_flush_all(do_cond_resched, &next_seq, &handover);
2894 * Abort if there was a failure to flush all messages to all
2895 * usable consoles. Either it is not possible to flush (in
2896 * which case it would be an infinite loop of retrying) or
2897 * another context has taken over printing.
2903 * Some context may have added new records after
2904 * console_flush_all() but before unlocking the console.
2905 * Re-check if there is a new record to flush. If the trylock
2906 * fails, another context is already handling the printing.
2908 } while (prb_read_valid(prb, next_seq, NULL) && console_trylock());
2910 EXPORT_SYMBOL(console_unlock);
2913 * console_conditional_schedule - yield the CPU if required
2915 * If the console code is currently allowed to sleep, and
2916 * if this CPU should yield the CPU to another task, do
2919 * Must be called within console_lock();.
2921 void __sched console_conditional_schedule(void)
2923 if (console_may_schedule)
2926 EXPORT_SYMBOL(console_conditional_schedule);
2928 void console_unblank(void)
2933 * console_unblank can no longer be called in interrupt context unless
2934 * oops_in_progress is set to 1..
2936 if (oops_in_progress) {
2937 if (down_trylock_console_sem() != 0)
2943 console_may_schedule = 0;
2945 if ((c->flags & CON_ENABLED) && c->unblank)
2949 if (!oops_in_progress)
2950 pr_flush(1000, true);
2954 * console_flush_on_panic - flush console content on panic
2955 * @mode: flush all messages in buffer or just the pending ones
2957 * Immediately output all pending messages no matter what.
2959 void console_flush_on_panic(enum con_flush_mode mode)
2962 * If someone else is holding the console lock, trylock will fail
2963 * and may_schedule may be set. Ignore and proceed to unlock so
2964 * that messages are flushed out. As this can be called from any
2965 * context and we don't want to get preempted while flushing,
2966 * ensure may_schedule is cleared.
2969 console_may_schedule = 0;
2971 if (mode == CONSOLE_REPLAY_ALL) {
2975 seq = prb_first_valid_seq(prb);
2983 * Return the console tty driver structure and its associated index
2985 struct tty_driver *console_device(int *index)
2988 struct tty_driver *driver = NULL;
2991 for_each_console(c) {
2994 driver = c->device(c, index);
3003 * Prevent further output on the passed console device so that (for example)
3004 * serial drivers can disable console output before suspending a port, and can
3005 * re-enable output afterwards.
3007 void console_stop(struct console *console)
3009 __pr_flush(console, 1000, true);
3011 console->flags &= ~CON_ENABLED;
3014 EXPORT_SYMBOL(console_stop);
3016 void console_start(struct console *console)
3019 console->flags |= CON_ENABLED;
3021 __pr_flush(console, 1000, true);
3023 EXPORT_SYMBOL(console_start);
3025 static int __read_mostly keep_bootcon;
3027 static int __init keep_bootcon_setup(char *str)
3030 pr_info("debug: skip boot console de-registration.\n");
3035 early_param("keep_bootcon", keep_bootcon_setup);
3038 * This is called by register_console() to try to match
3039 * the newly registered console with any of the ones selected
3040 * by either the command line or add_preferred_console() and
3043 * Care need to be taken with consoles that are statically
3044 * enabled such as netconsole
3046 static int try_enable_preferred_console(struct console *newcon,
3047 bool user_specified)
3049 struct console_cmdline *c;
3052 for (i = 0, c = console_cmdline;
3053 i < MAX_CMDLINECONSOLES && c->name[0];
3055 if (c->user_specified != user_specified)
3057 if (!newcon->match ||
3058 newcon->match(newcon, c->name, c->index, c->options) != 0) {
3059 /* default matching */
3060 BUILD_BUG_ON(sizeof(c->name) != sizeof(newcon->name));
3061 if (strcmp(c->name, newcon->name) != 0)
3063 if (newcon->index >= 0 &&
3064 newcon->index != c->index)
3066 if (newcon->index < 0)
3067 newcon->index = c->index;
3069 if (_braille_register_console(newcon, c))
3072 if (newcon->setup &&
3073 (err = newcon->setup(newcon, c->options)) != 0)
3076 newcon->flags |= CON_ENABLED;
3077 if (i == preferred_console)
3078 newcon->flags |= CON_CONSDEV;
3083 * Some consoles, such as pstore and netconsole, can be enabled even
3084 * without matching. Accept the pre-enabled consoles only when match()
3085 * and setup() had a chance to be called.
3087 if (newcon->flags & CON_ENABLED && c->user_specified == user_specified)
3093 /* Try to enable the console unconditionally */
3094 static void try_enable_default_console(struct console *newcon)
3096 if (newcon->index < 0)
3099 if (newcon->setup && newcon->setup(newcon, NULL) != 0)
3102 newcon->flags |= CON_ENABLED;
3105 newcon->flags |= CON_CONSDEV;
3108 #define con_printk(lvl, con, fmt, ...) \
3109 printk(lvl pr_fmt("%sconsole [%s%d] " fmt), \
3110 (con->flags & CON_BOOT) ? "boot" : "", \
3111 con->name, con->index, ##__VA_ARGS__)
3114 * The console driver calls this routine during kernel initialization
3115 * to register the console printing procedure with printk() and to
3116 * print any messages that were printed by the kernel before the
3117 * console driver was initialized.
3119 * This can happen pretty early during the boot process (because of
3120 * early_printk) - sometimes before setup_arch() completes - be careful
3121 * of what kernel features are used - they may not be initialised yet.
3123 * There are two types of consoles - bootconsoles (early_printk) and
3124 * "real" consoles (everything which is not a bootconsole) which are
3125 * handled differently.
3126 * - Any number of bootconsoles can be registered at any time.
3127 * - As soon as a "real" console is registered, all bootconsoles
3128 * will be unregistered automatically.
3129 * - Once a "real" console is registered, any attempt to register a
3130 * bootconsoles will be rejected
3132 void register_console(struct console *newcon)
3134 struct console *con;
3135 bool bootcon_enabled = false;
3136 bool realcon_enabled = false;
3139 for_each_console(con) {
3140 if (WARN(con == newcon, "console '%s%d' already registered\n",
3141 con->name, con->index))
3145 for_each_console(con) {
3146 if (con->flags & CON_BOOT)
3147 bootcon_enabled = true;
3149 realcon_enabled = true;
3152 /* Do not register boot consoles when there already is a real one. */
3153 if (newcon->flags & CON_BOOT && realcon_enabled) {
3154 pr_info("Too late to register bootconsole %s%d\n",
3155 newcon->name, newcon->index);
3160 * See if we want to enable this console driver by default.
3162 * Nope when a console is preferred by the command line, device
3165 * The first real console with tty binding (driver) wins. More
3166 * consoles might get enabled before the right one is found.
3168 * Note that a console with tty binding will have CON_CONSDEV
3169 * flag set and will be first in the list.
3171 if (preferred_console < 0) {
3172 if (!console_drivers || !console_drivers->device ||
3173 console_drivers->flags & CON_BOOT) {
3174 try_enable_default_console(newcon);
3178 /* See if this console matches one we selected on the command line */
3179 err = try_enable_preferred_console(newcon, true);
3181 /* If not, try to match against the platform default(s) */
3183 err = try_enable_preferred_console(newcon, false);
3185 /* printk() messages are not printed to the Braille console. */
3186 if (err || newcon->flags & CON_BRL)
3190 * If we have a bootconsole, and are switching to a real console,
3191 * don't print everything out again, since when the boot console, and
3192 * the real console are the same physical device, it's annoying to
3193 * see the beginning boot messages twice
3195 if (bootcon_enabled &&
3196 ((newcon->flags & (CON_CONSDEV | CON_BOOT)) == CON_CONSDEV)) {
3197 newcon->flags &= ~CON_PRINTBUFFER;
3201 * Put this console in the list - keep the
3202 * preferred driver at the head of the list.
3205 if ((newcon->flags & CON_CONSDEV) || console_drivers == NULL) {
3206 newcon->next = console_drivers;
3207 console_drivers = newcon;
3209 newcon->next->flags &= ~CON_CONSDEV;
3210 /* Ensure this flag is always set for the head of the list */
3211 newcon->flags |= CON_CONSDEV;
3213 newcon->next = console_drivers->next;
3214 console_drivers->next = newcon;
3217 if (newcon->flags & CON_EXTENDED)
3218 nr_ext_console_drivers++;
3220 newcon->dropped = 0;
3221 if (newcon->flags & CON_PRINTBUFFER) {
3222 /* Get a consistent copy of @syslog_seq. */
3223 mutex_lock(&syslog_lock);
3224 newcon->seq = syslog_seq;
3225 mutex_unlock(&syslog_lock);
3227 /* Begin with next message. */
3228 newcon->seq = prb_next_seq(prb);
3231 console_sysfs_notify();
3234 * By unregistering the bootconsoles after we enable the real console
3235 * we get the "console xxx enabled" message on all the consoles -
3236 * boot consoles, real consoles, etc - this is to ensure that end
3237 * users know there might be something in the kernel's log buffer that
3238 * went to the bootconsole (that they do not see on the real console)
3240 con_printk(KERN_INFO, newcon, "enabled\n");
3241 if (bootcon_enabled &&
3242 ((newcon->flags & (CON_CONSDEV | CON_BOOT)) == CON_CONSDEV) &&
3244 /* We need to iterate through all boot consoles, to make
3245 * sure we print everything out, before we unregister them.
3247 for_each_console(con)
3248 if (con->flags & CON_BOOT)
3249 unregister_console(con);
3252 EXPORT_SYMBOL(register_console);
3254 int unregister_console(struct console *console)
3256 struct console *con;
3259 con_printk(KERN_INFO, console, "disabled\n");
3261 res = _braille_unregister_console(console);
3269 if (console_drivers == console) {
3270 console_drivers=console->next;
3273 for_each_console(con) {
3274 if (con->next == console) {
3275 con->next = console->next;
3283 goto out_disable_unlock;
3285 if (console->flags & CON_EXTENDED)
3286 nr_ext_console_drivers--;
3289 * If this isn't the last console and it has CON_CONSDEV set, we
3290 * need to set it on the next preferred console.
3292 if (console_drivers != NULL && console->flags & CON_CONSDEV)
3293 console_drivers->flags |= CON_CONSDEV;
3295 console->flags &= ~CON_ENABLED;
3297 console_sysfs_notify();
3300 res = console->exit(console);
3305 console->flags &= ~CON_ENABLED;
3310 EXPORT_SYMBOL(unregister_console);
3313 * Initialize the console device. This is called *early*, so
3314 * we can't necessarily depend on lots of kernel help here.
3315 * Just do some early initializations, and do the complex setup
3318 void __init console_init(void)
3322 initcall_entry_t *ce;
3324 /* Setup the default TTY line discipline. */
3328 * set up the console device so that later boot sequences can
3329 * inform about problems etc..
3331 ce = __con_initcall_start;
3332 trace_initcall_level("console");
3333 while (ce < __con_initcall_end) {
3334 call = initcall_from_entry(ce);
3335 trace_initcall_start(call);
3337 trace_initcall_finish(call, ret);
3343 * Some boot consoles access data that is in the init section and which will
3344 * be discarded after the initcalls have been run. To make sure that no code
3345 * will access this data, unregister the boot consoles in a late initcall.
3347 * If for some reason, such as deferred probe or the driver being a loadable
3348 * module, the real console hasn't registered yet at this point, there will
3349 * be a brief interval in which no messages are logged to the console, which
3350 * makes it difficult to diagnose problems that occur during this time.
3352 * To mitigate this problem somewhat, only unregister consoles whose memory
3353 * intersects with the init section. Note that all other boot consoles will
3354 * get unregistered when the real preferred console is registered.
3356 static int __init printk_late_init(void)
3358 struct console *con;
3361 for_each_console(con) {
3362 if (!(con->flags & CON_BOOT))
3365 /* Check addresses that might be used for enabled consoles. */
3366 if (init_section_intersects(con, sizeof(*con)) ||
3367 init_section_contains(con->write, 0) ||
3368 init_section_contains(con->read, 0) ||
3369 init_section_contains(con->device, 0) ||
3370 init_section_contains(con->unblank, 0) ||
3371 init_section_contains(con->data, 0)) {
3373 * Please, consider moving the reported consoles out
3374 * of the init section.
3376 pr_warn("bootconsole [%s%d] uses init memory and must be disabled even before the real one is ready\n",
3377 con->name, con->index);
3378 unregister_console(con);
3381 ret = cpuhp_setup_state_nocalls(CPUHP_PRINTK_DEAD, "printk:dead", NULL,
3382 console_cpu_notify);
3384 ret = cpuhp_setup_state_nocalls(CPUHP_AP_ONLINE_DYN, "printk:online",
3385 console_cpu_notify, NULL);
3387 printk_sysctl_init();
3390 late_initcall(printk_late_init);
3392 #if defined CONFIG_PRINTK
3393 /* If @con is specified, only wait for that console. Otherwise wait for all. */
3394 static bool __pr_flush(struct console *con, int timeout_ms, bool reset_on_progress)
3396 int remaining = timeout_ms;
3405 seq = prb_next_seq(prb);
3411 for_each_console(c) {
3412 if (con && con != c)
3414 if (!console_is_usable(c))
3416 printk_seq = c->seq;
3417 if (printk_seq < seq)
3418 diff += seq - printk_seq;
3422 if (diff != last_diff && reset_on_progress)
3423 remaining = timeout_ms;
3425 if (diff == 0 || remaining == 0)
3428 if (remaining < 0) {
3429 /* no timeout limit */
3431 } else if (remaining < 100) {
3446 * pr_flush() - Wait for printing threads to catch up.
3448 * @timeout_ms: The maximum time (in ms) to wait.
3449 * @reset_on_progress: Reset the timeout if forward progress is seen.
3451 * A value of 0 for @timeout_ms means no waiting will occur. A value of -1
3452 * represents infinite waiting.
3454 * If @reset_on_progress is true, the timeout will be reset whenever any
3455 * printer has been seen to make some forward progress.
3457 * Context: Process context. May sleep while acquiring console lock.
3458 * Return: true if all enabled printers are caught up.
3460 bool pr_flush(int timeout_ms, bool reset_on_progress)
3462 return __pr_flush(NULL, timeout_ms, reset_on_progress);
3464 EXPORT_SYMBOL(pr_flush);
3467 * Delayed printk version, for scheduler-internal messages:
3469 #define PRINTK_PENDING_WAKEUP 0x01
3470 #define PRINTK_PENDING_OUTPUT 0x02
3472 static DEFINE_PER_CPU(int, printk_pending);
3474 static void wake_up_klogd_work_func(struct irq_work *irq_work)
3476 int pending = this_cpu_xchg(printk_pending, 0);
3478 if (pending & PRINTK_PENDING_OUTPUT) {
3479 /* If trylock fails, someone else is doing the printing */
3480 if (console_trylock())
3484 if (pending & PRINTK_PENDING_WAKEUP)
3485 wake_up_interruptible(&log_wait);
3488 static DEFINE_PER_CPU(struct irq_work, wake_up_klogd_work) =
3489 IRQ_WORK_INIT_LAZY(wake_up_klogd_work_func);
3491 static void __wake_up_klogd(int val)
3493 if (!printk_percpu_data_ready())
3498 * Guarantee any new records can be seen by tasks preparing to wait
3499 * before this context checks if the wait queue is empty.
3501 * The full memory barrier within wq_has_sleeper() pairs with the full
3502 * memory barrier within set_current_state() of
3503 * prepare_to_wait_event(), which is called after ___wait_event() adds
3504 * the waiter but before it has checked the wait condition.
3506 * This pairs with devkmsg_read:A and syslog_print:A.
3508 if (wq_has_sleeper(&log_wait) || /* LMM(__wake_up_klogd:A) */
3509 (val & PRINTK_PENDING_OUTPUT)) {
3510 this_cpu_or(printk_pending, val);
3511 irq_work_queue(this_cpu_ptr(&wake_up_klogd_work));
3516 void wake_up_klogd(void)
3518 __wake_up_klogd(PRINTK_PENDING_WAKEUP);
3521 void defer_console_output(void)
3524 * New messages may have been added directly to the ringbuffer
3525 * using vprintk_store(), so wake any waiters as well.
3527 __wake_up_klogd(PRINTK_PENDING_WAKEUP | PRINTK_PENDING_OUTPUT);
3530 void printk_trigger_flush(void)
3532 defer_console_output();
3535 int vprintk_deferred(const char *fmt, va_list args)
3539 r = vprintk_emit(0, LOGLEVEL_SCHED, NULL, fmt, args);
3540 defer_console_output();
3545 int _printk_deferred(const char *fmt, ...)
3550 va_start(args, fmt);
3551 r = vprintk_deferred(fmt, args);
3558 * printk rate limiting, lifted from the networking subsystem.
3560 * This enforces a rate limit: not more than 10 kernel messages
3561 * every 5s to make a denial-of-service attack impossible.
3563 DEFINE_RATELIMIT_STATE(printk_ratelimit_state, 5 * HZ, 10);
3565 int __printk_ratelimit(const char *func)
3567 return ___ratelimit(&printk_ratelimit_state, func);
3569 EXPORT_SYMBOL(__printk_ratelimit);
3572 * printk_timed_ratelimit - caller-controlled printk ratelimiting
3573 * @caller_jiffies: pointer to caller's state
3574 * @interval_msecs: minimum interval between prints
3576 * printk_timed_ratelimit() returns true if more than @interval_msecs
3577 * milliseconds have elapsed since the last time printk_timed_ratelimit()
3580 bool printk_timed_ratelimit(unsigned long *caller_jiffies,
3581 unsigned int interval_msecs)
3583 unsigned long elapsed = jiffies - *caller_jiffies;
3585 if (*caller_jiffies && elapsed <= msecs_to_jiffies(interval_msecs))
3588 *caller_jiffies = jiffies;
3591 EXPORT_SYMBOL(printk_timed_ratelimit);
3593 static DEFINE_SPINLOCK(dump_list_lock);
3594 static LIST_HEAD(dump_list);
3597 * kmsg_dump_register - register a kernel log dumper.
3598 * @dumper: pointer to the kmsg_dumper structure
3600 * Adds a kernel log dumper to the system. The dump callback in the
3601 * structure will be called when the kernel oopses or panics and must be
3602 * set. Returns zero on success and %-EINVAL or %-EBUSY otherwise.
3604 int kmsg_dump_register(struct kmsg_dumper *dumper)
3606 unsigned long flags;
3609 /* The dump callback needs to be set */
3613 spin_lock_irqsave(&dump_list_lock, flags);
3614 /* Don't allow registering multiple times */
3615 if (!dumper->registered) {
3616 dumper->registered = 1;
3617 list_add_tail_rcu(&dumper->list, &dump_list);
3620 spin_unlock_irqrestore(&dump_list_lock, flags);
3624 EXPORT_SYMBOL_GPL(kmsg_dump_register);
3627 * kmsg_dump_unregister - unregister a kmsg dumper.
3628 * @dumper: pointer to the kmsg_dumper structure
3630 * Removes a dump device from the system. Returns zero on success and
3631 * %-EINVAL otherwise.
3633 int kmsg_dump_unregister(struct kmsg_dumper *dumper)
3635 unsigned long flags;
3638 spin_lock_irqsave(&dump_list_lock, flags);
3639 if (dumper->registered) {
3640 dumper->registered = 0;
3641 list_del_rcu(&dumper->list);
3644 spin_unlock_irqrestore(&dump_list_lock, flags);
3649 EXPORT_SYMBOL_GPL(kmsg_dump_unregister);
3651 static bool always_kmsg_dump;
3652 module_param_named(always_kmsg_dump, always_kmsg_dump, bool, S_IRUGO | S_IWUSR);
3654 const char *kmsg_dump_reason_str(enum kmsg_dump_reason reason)
3657 case KMSG_DUMP_PANIC:
3659 case KMSG_DUMP_OOPS:
3661 case KMSG_DUMP_EMERG:
3663 case KMSG_DUMP_SHUTDOWN:
3669 EXPORT_SYMBOL_GPL(kmsg_dump_reason_str);
3672 * kmsg_dump - dump kernel log to kernel message dumpers.
3673 * @reason: the reason (oops, panic etc) for dumping
3675 * Call each of the registered dumper's dump() callback, which can
3676 * retrieve the kmsg records with kmsg_dump_get_line() or
3677 * kmsg_dump_get_buffer().
3679 void kmsg_dump(enum kmsg_dump_reason reason)
3681 struct kmsg_dumper *dumper;
3684 list_for_each_entry_rcu(dumper, &dump_list, list) {
3685 enum kmsg_dump_reason max_reason = dumper->max_reason;
3688 * If client has not provided a specific max_reason, default
3689 * to KMSG_DUMP_OOPS, unless always_kmsg_dump was set.
3691 if (max_reason == KMSG_DUMP_UNDEF) {
3692 max_reason = always_kmsg_dump ? KMSG_DUMP_MAX :
3695 if (reason > max_reason)
3698 /* invoke dumper which will iterate over records */
3699 dumper->dump(dumper, reason);
3705 * kmsg_dump_get_line - retrieve one kmsg log line
3706 * @iter: kmsg dump iterator
3707 * @syslog: include the "<4>" prefixes
3708 * @line: buffer to copy the line to
3709 * @size: maximum size of the buffer
3710 * @len: length of line placed into buffer
3712 * Start at the beginning of the kmsg buffer, with the oldest kmsg
3713 * record, and copy one record into the provided buffer.
3715 * Consecutive calls will return the next available record moving
3716 * towards the end of the buffer with the youngest messages.
3718 * A return value of FALSE indicates that there are no more records to
3721 bool kmsg_dump_get_line(struct kmsg_dump_iter *iter, bool syslog,
3722 char *line, size_t size, size_t *len)
3724 u64 min_seq = latched_seq_read_nolock(&clear_seq);
3725 struct printk_info info;
3726 unsigned int line_count;
3727 struct printk_record r;
3731 if (iter->cur_seq < min_seq)
3732 iter->cur_seq = min_seq;
3734 prb_rec_init_rd(&r, &info, line, size);
3736 /* Read text or count text lines? */
3738 if (!prb_read_valid(prb, iter->cur_seq, &r))
3740 l = record_print_text(&r, syslog, printk_time);
3742 if (!prb_read_valid_info(prb, iter->cur_seq,
3743 &info, &line_count)) {
3746 l = get_record_print_text_size(&info, line_count, syslog,
3751 iter->cur_seq = r.info->seq + 1;
3758 EXPORT_SYMBOL_GPL(kmsg_dump_get_line);
3761 * kmsg_dump_get_buffer - copy kmsg log lines
3762 * @iter: kmsg dump iterator
3763 * @syslog: include the "<4>" prefixes
3764 * @buf: buffer to copy the line to
3765 * @size: maximum size of the buffer
3766 * @len_out: length of line placed into buffer
3768 * Start at the end of the kmsg buffer and fill the provided buffer
3769 * with as many of the *youngest* kmsg records that fit into it.
3770 * If the buffer is large enough, all available kmsg records will be
3771 * copied with a single call.
3773 * Consecutive calls will fill the buffer with the next block of
3774 * available older records, not including the earlier retrieved ones.
3776 * A return value of FALSE indicates that there are no more records to
3779 bool kmsg_dump_get_buffer(struct kmsg_dump_iter *iter, bool syslog,
3780 char *buf, size_t size, size_t *len_out)
3782 u64 min_seq = latched_seq_read_nolock(&clear_seq);
3783 struct printk_info info;
3784 struct printk_record r;
3789 bool time = printk_time;
3794 if (iter->cur_seq < min_seq)
3795 iter->cur_seq = min_seq;
3797 if (prb_read_valid_info(prb, iter->cur_seq, &info, NULL)) {
3798 if (info.seq != iter->cur_seq) {
3799 /* messages are gone, move to first available one */
3800 iter->cur_seq = info.seq;
3805 if (iter->cur_seq >= iter->next_seq)
3809 * Find first record that fits, including all following records,
3810 * into the user-provided buffer for this dump. Pass in size-1
3811 * because this function (by way of record_print_text()) will
3812 * not write more than size-1 bytes of text into @buf.
3814 seq = find_first_fitting_seq(iter->cur_seq, iter->next_seq,
3815 size - 1, syslog, time);
3818 * Next kmsg_dump_get_buffer() invocation will dump block of
3819 * older records stored right before this one.
3823 prb_rec_init_rd(&r, &info, buf, size);
3826 prb_for_each_record(seq, prb, seq, &r) {
3827 if (r.info->seq >= iter->next_seq)
3830 len += record_print_text(&r, syslog, time);
3832 /* Adjust record to store to remaining buffer space. */
3833 prb_rec_init_rd(&r, &info, buf + len, size - len);
3836 iter->next_seq = next_seq;
3843 EXPORT_SYMBOL_GPL(kmsg_dump_get_buffer);
3846 * kmsg_dump_rewind - reset the iterator
3847 * @iter: kmsg dump iterator
3849 * Reset the dumper's iterator so that kmsg_dump_get_line() and
3850 * kmsg_dump_get_buffer() can be called again and used multiple
3851 * times within the same dumper.dump() callback.
3853 void kmsg_dump_rewind(struct kmsg_dump_iter *iter)
3855 iter->cur_seq = latched_seq_read_nolock(&clear_seq);
3856 iter->next_seq = prb_next_seq(prb);
3858 EXPORT_SYMBOL_GPL(kmsg_dump_rewind);
3863 static atomic_t printk_cpu_sync_owner = ATOMIC_INIT(-1);
3864 static atomic_t printk_cpu_sync_nested = ATOMIC_INIT(0);
3867 * __printk_cpu_sync_wait() - Busy wait until the printk cpu-reentrant
3868 * spinning lock is not owned by any CPU.
3870 * Context: Any context.
3872 void __printk_cpu_sync_wait(void)
3876 } while (atomic_read(&printk_cpu_sync_owner) != -1);
3878 EXPORT_SYMBOL(__printk_cpu_sync_wait);
3881 * __printk_cpu_sync_try_get() - Try to acquire the printk cpu-reentrant
3884 * If no processor has the lock, the calling processor takes the lock and
3885 * becomes the owner. If the calling processor is already the owner of the
3886 * lock, this function succeeds immediately.
3888 * Context: Any context. Expects interrupts to be disabled.
3889 * Return: 1 on success, otherwise 0.
3891 int __printk_cpu_sync_try_get(void)
3896 cpu = smp_processor_id();
3899 * Guarantee loads and stores from this CPU when it is the lock owner
3900 * are _not_ visible to the previous lock owner. This pairs with
3901 * __printk_cpu_sync_put:B.
3903 * Memory barrier involvement:
3905 * If __printk_cpu_sync_try_get:A reads from __printk_cpu_sync_put:B,
3906 * then __printk_cpu_sync_put:A can never read from
3907 * __printk_cpu_sync_try_get:B.
3911 * RELEASE from __printk_cpu_sync_put:A to __printk_cpu_sync_put:B
3912 * of the previous CPU
3914 * ACQUIRE from __printk_cpu_sync_try_get:A to
3915 * __printk_cpu_sync_try_get:B of this CPU
3917 old = atomic_cmpxchg_acquire(&printk_cpu_sync_owner, -1,
3918 cpu); /* LMM(__printk_cpu_sync_try_get:A) */
3921 * This CPU is now the owner and begins loading/storing
3922 * data: LMM(__printk_cpu_sync_try_get:B)
3926 } else if (old == cpu) {
3927 /* This CPU is already the owner. */
3928 atomic_inc(&printk_cpu_sync_nested);
3934 EXPORT_SYMBOL(__printk_cpu_sync_try_get);
3937 * __printk_cpu_sync_put() - Release the printk cpu-reentrant spinning lock.
3939 * The calling processor must be the owner of the lock.
3941 * Context: Any context. Expects interrupts to be disabled.
3943 void __printk_cpu_sync_put(void)
3945 if (atomic_read(&printk_cpu_sync_nested)) {
3946 atomic_dec(&printk_cpu_sync_nested);
3951 * This CPU is finished loading/storing data:
3952 * LMM(__printk_cpu_sync_put:A)
3956 * Guarantee loads and stores from this CPU when it was the
3957 * lock owner are visible to the next lock owner. This pairs
3958 * with __printk_cpu_sync_try_get:A.
3960 * Memory barrier involvement:
3962 * If __printk_cpu_sync_try_get:A reads from __printk_cpu_sync_put:B,
3963 * then __printk_cpu_sync_try_get:B reads from __printk_cpu_sync_put:A.
3967 * RELEASE from __printk_cpu_sync_put:A to __printk_cpu_sync_put:B
3970 * ACQUIRE from __printk_cpu_sync_try_get:A to
3971 * __printk_cpu_sync_try_get:B of the next CPU
3973 atomic_set_release(&printk_cpu_sync_owner,
3974 -1); /* LMM(__printk_cpu_sync_put:B) */
3976 EXPORT_SYMBOL(__printk_cpu_sync_put);
3977 #endif /* CONFIG_SMP */