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 static struct lockdep_map console_lock_dep_map = {
98 .name = "console_lock"
102 enum devkmsg_log_bits {
103 __DEVKMSG_LOG_BIT_ON = 0,
104 __DEVKMSG_LOG_BIT_OFF,
105 __DEVKMSG_LOG_BIT_LOCK,
108 enum devkmsg_log_masks {
109 DEVKMSG_LOG_MASK_ON = BIT(__DEVKMSG_LOG_BIT_ON),
110 DEVKMSG_LOG_MASK_OFF = BIT(__DEVKMSG_LOG_BIT_OFF),
111 DEVKMSG_LOG_MASK_LOCK = BIT(__DEVKMSG_LOG_BIT_LOCK),
114 /* Keep both the 'on' and 'off' bits clear, i.e. ratelimit by default: */
115 #define DEVKMSG_LOG_MASK_DEFAULT 0
117 static unsigned int __read_mostly devkmsg_log = DEVKMSG_LOG_MASK_DEFAULT;
119 static int __control_devkmsg(char *str)
126 len = str_has_prefix(str, "on");
128 devkmsg_log = DEVKMSG_LOG_MASK_ON;
132 len = str_has_prefix(str, "off");
134 devkmsg_log = DEVKMSG_LOG_MASK_OFF;
138 len = str_has_prefix(str, "ratelimit");
140 devkmsg_log = DEVKMSG_LOG_MASK_DEFAULT;
147 static int __init control_devkmsg(char *str)
149 if (__control_devkmsg(str) < 0)
153 * Set sysctl string accordingly:
155 if (devkmsg_log == DEVKMSG_LOG_MASK_ON)
156 strcpy(devkmsg_log_str, "on");
157 else if (devkmsg_log == DEVKMSG_LOG_MASK_OFF)
158 strcpy(devkmsg_log_str, "off");
159 /* else "ratelimit" which is set by default. */
162 * Sysctl cannot change it anymore. The kernel command line setting of
163 * this parameter is to force the setting to be permanent throughout the
164 * runtime of the system. This is a precation measure against userspace
165 * trying to be a smarta** and attempting to change it up on us.
167 devkmsg_log |= DEVKMSG_LOG_MASK_LOCK;
171 __setup("printk.devkmsg=", control_devkmsg);
173 char devkmsg_log_str[DEVKMSG_STR_MAX_SIZE] = "ratelimit";
175 int devkmsg_sysctl_set_loglvl(struct ctl_table *table, int write,
176 void *buffer, size_t *lenp, loff_t *ppos)
178 char old_str[DEVKMSG_STR_MAX_SIZE];
183 if (devkmsg_log & DEVKMSG_LOG_MASK_LOCK)
187 strncpy(old_str, devkmsg_log_str, DEVKMSG_STR_MAX_SIZE);
190 err = proc_dostring(table, write, buffer, lenp, ppos);
195 err = __control_devkmsg(devkmsg_log_str);
198 * Do not accept an unknown string OR a known string with
201 if (err < 0 || (err + 1 != *lenp)) {
203 /* ... and restore old setting. */
205 strncpy(devkmsg_log_str, old_str, DEVKMSG_STR_MAX_SIZE);
214 /* Number of registered extended console drivers. */
215 static int nr_ext_console_drivers;
218 * Helper macros to handle lockdep when locking/unlocking console_sem. We use
219 * macros instead of functions so that _RET_IP_ contains useful information.
221 #define down_console_sem() do { \
223 mutex_acquire(&console_lock_dep_map, 0, 0, _RET_IP_);\
226 static int __down_trylock_console_sem(unsigned long ip)
232 * Here and in __up_console_sem() we need to be in safe mode,
233 * because spindump/WARN/etc from under console ->lock will
234 * deadlock in printk()->down_trylock_console_sem() otherwise.
236 printk_safe_enter_irqsave(flags);
237 lock_failed = down_trylock(&console_sem);
238 printk_safe_exit_irqrestore(flags);
242 mutex_acquire(&console_lock_dep_map, 0, 1, ip);
245 #define down_trylock_console_sem() __down_trylock_console_sem(_RET_IP_)
247 static void __up_console_sem(unsigned long ip)
251 mutex_release(&console_lock_dep_map, ip);
253 printk_safe_enter_irqsave(flags);
255 printk_safe_exit_irqrestore(flags);
257 #define up_console_sem() __up_console_sem(_RET_IP_)
260 * This is used for debugging the mess that is the VT code by
261 * keeping track if we have the console semaphore held. It's
262 * definitely not the perfect debug tool (we don't know if _WE_
263 * hold it and are racing, but it helps tracking those weird code
264 * paths in the console code where we end up in places I want
265 * locked without the console semaphore held).
267 static int console_locked, console_suspended;
270 * If exclusive_console is non-NULL then only this console is to be printed to.
272 static struct console *exclusive_console;
275 * Array of consoles built from command line options (console=)
278 #define MAX_CMDLINECONSOLES 8
280 static struct console_cmdline console_cmdline[MAX_CMDLINECONSOLES];
282 static int preferred_console = -1;
283 static bool has_preferred_console;
284 int console_set_on_cmdline;
285 EXPORT_SYMBOL(console_set_on_cmdline);
287 /* Flag: console code may call schedule() */
288 static int console_may_schedule;
290 enum con_msg_format_flags {
291 MSG_FORMAT_DEFAULT = 0,
292 MSG_FORMAT_SYSLOG = (1 << 0),
295 static int console_msg_format = MSG_FORMAT_DEFAULT;
298 * The printk log buffer consists of a sequenced collection of records, each
299 * containing variable length message text. Every record also contains its
300 * own meta-data (@info).
302 * Every record meta-data carries the timestamp in microseconds, as well as
303 * the standard userspace syslog level and syslog facility. The usual kernel
304 * messages use LOG_KERN; userspace-injected messages always carry a matching
305 * syslog facility, by default LOG_USER. The origin of every message can be
306 * reliably determined that way.
308 * The human readable log message of a record is available in @text, the
309 * length of the message text in @text_len. The stored message is not
312 * Optionally, a record can carry a dictionary of properties (key/value
313 * pairs), to provide userspace with a machine-readable message context.
315 * Examples for well-defined, commonly used property names are:
316 * DEVICE=b12:8 device identifier
320 * +sound:card0 subsystem:devname
321 * SUBSYSTEM=pci driver-core subsystem name
323 * Valid characters in property names are [a-zA-Z0-9.-_]. Property names
324 * and values are terminated by a '\0' character.
326 * Example of record values:
327 * record.text_buf = "it's a line" (unterminated)
328 * record.info.seq = 56
329 * record.info.ts_nsec = 36863
330 * record.info.text_len = 11
331 * record.info.facility = 0 (LOG_KERN)
332 * record.info.flags = 0
333 * record.info.level = 3 (LOG_ERR)
334 * record.info.caller_id = 299 (task 299)
335 * record.info.dev_info.subsystem = "pci" (terminated)
336 * record.info.dev_info.device = "+pci:0000:00:01.0" (terminated)
338 * The 'struct printk_info' buffer must never be directly exported to
339 * userspace, it is a kernel-private implementation detail that might
340 * need to be changed in the future, when the requirements change.
342 * /dev/kmsg exports the structured data in the following line format:
343 * "<level>,<sequnum>,<timestamp>,<contflag>[,additional_values, ... ];<message text>\n"
345 * Users of the export format should ignore possible additional values
346 * separated by ',', and find the message after the ';' character.
348 * The optional key/value pairs are attached as continuation lines starting
349 * with a space character and terminated by a newline. All possible
350 * non-prinatable characters are escaped in the "\xff" notation.
354 LOG_NEWLINE = 2, /* text ended with a newline */
355 LOG_CONT = 8, /* text is a fragment of a continuation line */
358 /* syslog_lock protects syslog_* variables and write access to clear_seq. */
359 static DEFINE_RAW_SPINLOCK(syslog_lock);
362 DECLARE_WAIT_QUEUE_HEAD(log_wait);
363 /* All 3 protected by @syslog_lock. */
364 /* the next printk record to read by syslog(READ) or /proc/kmsg */
365 static u64 syslog_seq;
366 static size_t syslog_partial;
367 static bool syslog_time;
369 /* All 3 protected by @console_sem. */
370 /* the next printk record to write to the console */
371 static u64 console_seq;
372 static u64 exclusive_console_stop_seq;
373 static unsigned long console_dropped;
376 seqcount_latch_t latch;
381 * The next printk record to read after the last 'clear' command. There are
382 * two copies (updated with seqcount_latch) so that reads can locklessly
383 * access a valid value. Writers are synchronized by @syslog_lock.
385 static struct latched_seq clear_seq = {
386 .latch = SEQCNT_LATCH_ZERO(clear_seq.latch),
391 #ifdef CONFIG_PRINTK_CALLER
392 #define PREFIX_MAX 48
394 #define PREFIX_MAX 32
397 /* the maximum size of a formatted record (i.e. with prefix added per line) */
398 #define CONSOLE_LOG_MAX 1024
400 /* the maximum size allowed to be reserved for a record */
401 #define LOG_LINE_MAX (CONSOLE_LOG_MAX - PREFIX_MAX)
403 #define LOG_LEVEL(v) ((v) & 0x07)
404 #define LOG_FACILITY(v) ((v) >> 3 & 0xff)
407 #define LOG_ALIGN __alignof__(unsigned long)
408 #define __LOG_BUF_LEN (1 << CONFIG_LOG_BUF_SHIFT)
409 #define LOG_BUF_LEN_MAX (u32)(1 << 31)
410 static char __log_buf[__LOG_BUF_LEN] __aligned(LOG_ALIGN);
411 static char *log_buf = __log_buf;
412 static u32 log_buf_len = __LOG_BUF_LEN;
415 * Define the average message size. This only affects the number of
416 * descriptors that will be available. Underestimating is better than
417 * overestimating (too many available descriptors is better than not enough).
419 #define PRB_AVGBITS 5 /* 32 character average length */
421 #if CONFIG_LOG_BUF_SHIFT <= PRB_AVGBITS
422 #error CONFIG_LOG_BUF_SHIFT value too small.
424 _DEFINE_PRINTKRB(printk_rb_static, CONFIG_LOG_BUF_SHIFT - PRB_AVGBITS,
425 PRB_AVGBITS, &__log_buf[0]);
427 static struct printk_ringbuffer printk_rb_dynamic;
429 static struct printk_ringbuffer *prb = &printk_rb_static;
432 * We cannot access per-CPU data (e.g. per-CPU flush irq_work) before
433 * per_cpu_areas are initialised. This variable is set to true when
434 * it's safe to access per-CPU data.
436 static bool __printk_percpu_data_ready __read_mostly;
438 bool printk_percpu_data_ready(void)
440 return __printk_percpu_data_ready;
443 /* Must be called under syslog_lock. */
444 static void latched_seq_write(struct latched_seq *ls, u64 val)
446 raw_write_seqcount_latch(&ls->latch);
448 raw_write_seqcount_latch(&ls->latch);
452 /* Can be called from any context. */
453 static u64 latched_seq_read_nolock(struct latched_seq *ls)
460 seq = raw_read_seqcount_latch(&ls->latch);
463 } while (read_seqcount_latch_retry(&ls->latch, seq));
468 /* Return log buffer address */
469 char *log_buf_addr_get(void)
474 /* Return log buffer size */
475 u32 log_buf_len_get(void)
481 * Define how much of the log buffer we could take at maximum. The value
482 * must be greater than two. Note that only half of the buffer is available
483 * when the index points to the middle.
485 #define MAX_LOG_TAKE_PART 4
486 static const char trunc_msg[] = "<truncated>";
488 static void truncate_msg(u16 *text_len, u16 *trunc_msg_len)
491 * The message should not take the whole buffer. Otherwise, it might
492 * get removed too soon.
494 u32 max_text_len = log_buf_len / MAX_LOG_TAKE_PART;
496 if (*text_len > max_text_len)
497 *text_len = max_text_len;
499 /* enable the warning message (if there is room) */
500 *trunc_msg_len = strlen(trunc_msg);
501 if (*text_len >= *trunc_msg_len)
502 *text_len -= *trunc_msg_len;
507 int dmesg_restrict = IS_ENABLED(CONFIG_SECURITY_DMESG_RESTRICT);
509 static int syslog_action_restricted(int type)
514 * Unless restricted, we allow "read all" and "get buffer size"
517 return type != SYSLOG_ACTION_READ_ALL &&
518 type != SYSLOG_ACTION_SIZE_BUFFER;
521 static int check_syslog_permissions(int type, int source)
524 * If this is from /proc/kmsg and we've already opened it, then we've
525 * already done the capabilities checks at open time.
527 if (source == SYSLOG_FROM_PROC && type != SYSLOG_ACTION_OPEN)
530 if (syslog_action_restricted(type)) {
531 if (capable(CAP_SYSLOG))
534 * For historical reasons, accept CAP_SYS_ADMIN too, with
537 if (capable(CAP_SYS_ADMIN)) {
538 pr_warn_once("%s (%d): Attempt to access syslog with "
539 "CAP_SYS_ADMIN but no CAP_SYSLOG "
541 current->comm, task_pid_nr(current));
547 return security_syslog(type);
550 static void append_char(char **pp, char *e, char c)
556 static ssize_t info_print_ext_header(char *buf, size_t size,
557 struct printk_info *info)
559 u64 ts_usec = info->ts_nsec;
561 #ifdef CONFIG_PRINTK_CALLER
562 u32 id = info->caller_id;
564 snprintf(caller, sizeof(caller), ",caller=%c%u",
565 id & 0x80000000 ? 'C' : 'T', id & ~0x80000000);
570 do_div(ts_usec, 1000);
572 return scnprintf(buf, size, "%u,%llu,%llu,%c%s;",
573 (info->facility << 3) | info->level, info->seq,
574 ts_usec, info->flags & LOG_CONT ? 'c' : '-', caller);
577 static ssize_t msg_add_ext_text(char *buf, size_t size,
578 const char *text, size_t text_len,
581 char *p = buf, *e = buf + size;
584 /* escape non-printable characters */
585 for (i = 0; i < text_len; i++) {
586 unsigned char c = text[i];
588 if (c < ' ' || c >= 127 || c == '\\')
589 p += scnprintf(p, e - p, "\\x%02x", c);
591 append_char(&p, e, c);
593 append_char(&p, e, endc);
598 static ssize_t msg_add_dict_text(char *buf, size_t size,
599 const char *key, const char *val)
601 size_t val_len = strlen(val);
607 len = msg_add_ext_text(buf, size, "", 0, ' '); /* dict prefix */
608 len += msg_add_ext_text(buf + len, size - len, key, strlen(key), '=');
609 len += msg_add_ext_text(buf + len, size - len, val, val_len, '\n');
614 static ssize_t msg_print_ext_body(char *buf, size_t size,
615 char *text, size_t text_len,
616 struct dev_printk_info *dev_info)
620 len = msg_add_ext_text(buf, size, text, text_len, '\n');
625 len += msg_add_dict_text(buf + len, size - len, "SUBSYSTEM",
626 dev_info->subsystem);
627 len += msg_add_dict_text(buf + len, size - len, "DEVICE",
633 /* /dev/kmsg - userspace message inject/listen interface */
634 struct devkmsg_user {
636 struct ratelimit_state rs;
638 char buf[CONSOLE_EXT_LOG_MAX];
640 struct printk_info info;
641 char text_buf[CONSOLE_EXT_LOG_MAX];
642 struct printk_record record;
645 static __printf(3, 4) __cold
646 int devkmsg_emit(int facility, int level, const char *fmt, ...)
652 r = vprintk_emit(facility, level, NULL, fmt, args);
658 static ssize_t devkmsg_write(struct kiocb *iocb, struct iov_iter *from)
661 int level = default_message_loglevel;
662 int facility = 1; /* LOG_USER */
663 struct file *file = iocb->ki_filp;
664 struct devkmsg_user *user = file->private_data;
665 size_t len = iov_iter_count(from);
668 if (!user || len > LOG_LINE_MAX)
671 /* Ignore when user logging is disabled. */
672 if (devkmsg_log & DEVKMSG_LOG_MASK_OFF)
675 /* Ratelimit when not explicitly enabled. */
676 if (!(devkmsg_log & DEVKMSG_LOG_MASK_ON)) {
677 if (!___ratelimit(&user->rs, current->comm))
681 buf = kmalloc(len+1, GFP_KERNEL);
686 if (!copy_from_iter_full(buf, len, from)) {
692 * Extract and skip the syslog prefix <[0-9]*>. Coming from userspace
693 * the decimal value represents 32bit, the lower 3 bit are the log
694 * level, the rest are the log facility.
696 * If no prefix or no userspace facility is specified, we
697 * enforce LOG_USER, to be able to reliably distinguish
698 * kernel-generated messages from userspace-injected ones.
701 if (line[0] == '<') {
705 u = simple_strtoul(line + 1, &endp, 10);
706 if (endp && endp[0] == '>') {
707 level = LOG_LEVEL(u);
708 if (LOG_FACILITY(u) != 0)
709 facility = LOG_FACILITY(u);
715 devkmsg_emit(facility, level, "%s", line);
720 static ssize_t devkmsg_read(struct file *file, char __user *buf,
721 size_t count, loff_t *ppos)
723 struct devkmsg_user *user = file->private_data;
724 struct printk_record *r = &user->record;
731 ret = mutex_lock_interruptible(&user->lock);
735 printk_safe_enter_irq();
736 if (!prb_read_valid(prb, atomic64_read(&user->seq), r)) {
737 if (file->f_flags & O_NONBLOCK) {
739 printk_safe_exit_irq();
743 printk_safe_exit_irq();
744 ret = wait_event_interruptible(log_wait,
745 prb_read_valid(prb, atomic64_read(&user->seq), r));
748 printk_safe_enter_irq();
751 if (r->info->seq != atomic64_read(&user->seq)) {
752 /* our last seen message is gone, return error and reset */
753 atomic64_set(&user->seq, r->info->seq);
755 printk_safe_exit_irq();
759 len = info_print_ext_header(user->buf, sizeof(user->buf), r->info);
760 len += msg_print_ext_body(user->buf + len, sizeof(user->buf) - len,
761 &r->text_buf[0], r->info->text_len,
764 atomic64_set(&user->seq, r->info->seq + 1);
765 printk_safe_exit_irq();
772 if (copy_to_user(buf, user->buf, len)) {
778 mutex_unlock(&user->lock);
783 * Be careful when modifying this function!!!
785 * Only few operations are supported because the device works only with the
786 * entire variable length messages (records). Non-standard values are
787 * returned in the other cases and has been this way for quite some time.
788 * User space applications might depend on this behavior.
790 static loff_t devkmsg_llseek(struct file *file, loff_t offset, int whence)
792 struct devkmsg_user *user = file->private_data;
800 printk_safe_enter_irq();
803 /* the first record */
804 atomic64_set(&user->seq, prb_first_valid_seq(prb));
808 * The first record after the last SYSLOG_ACTION_CLEAR,
809 * like issued by 'dmesg -c'. Reading /dev/kmsg itself
810 * changes no global state, and does not clear anything.
812 atomic64_set(&user->seq, latched_seq_read_nolock(&clear_seq));
815 /* after the last record */
816 atomic64_set(&user->seq, prb_next_seq(prb));
821 printk_safe_exit_irq();
825 static __poll_t devkmsg_poll(struct file *file, poll_table *wait)
827 struct devkmsg_user *user = file->private_data;
828 struct printk_info info;
832 return EPOLLERR|EPOLLNVAL;
834 poll_wait(file, &log_wait, wait);
836 printk_safe_enter_irq();
837 if (prb_read_valid_info(prb, atomic64_read(&user->seq), &info, NULL)) {
838 /* return error when data has vanished underneath us */
839 if (info.seq != atomic64_read(&user->seq))
840 ret = EPOLLIN|EPOLLRDNORM|EPOLLERR|EPOLLPRI;
842 ret = EPOLLIN|EPOLLRDNORM;
844 printk_safe_exit_irq();
849 static int devkmsg_open(struct inode *inode, struct file *file)
851 struct devkmsg_user *user;
854 if (devkmsg_log & DEVKMSG_LOG_MASK_OFF)
857 /* write-only does not need any file context */
858 if ((file->f_flags & O_ACCMODE) != O_WRONLY) {
859 err = check_syslog_permissions(SYSLOG_ACTION_READ_ALL,
865 user = kmalloc(sizeof(struct devkmsg_user), GFP_KERNEL);
869 ratelimit_default_init(&user->rs);
870 ratelimit_set_flags(&user->rs, RATELIMIT_MSG_ON_RELEASE);
872 mutex_init(&user->lock);
874 prb_rec_init_rd(&user->record, &user->info,
875 &user->text_buf[0], sizeof(user->text_buf));
877 printk_safe_enter_irq();
878 atomic64_set(&user->seq, prb_first_valid_seq(prb));
879 printk_safe_exit_irq();
881 file->private_data = user;
885 static int devkmsg_release(struct inode *inode, struct file *file)
887 struct devkmsg_user *user = file->private_data;
892 ratelimit_state_exit(&user->rs);
894 mutex_destroy(&user->lock);
899 const struct file_operations kmsg_fops = {
900 .open = devkmsg_open,
901 .read = devkmsg_read,
902 .write_iter = devkmsg_write,
903 .llseek = devkmsg_llseek,
904 .poll = devkmsg_poll,
905 .release = devkmsg_release,
908 #ifdef CONFIG_CRASH_CORE
910 * This appends the listed symbols to /proc/vmcore
912 * /proc/vmcore is used by various utilities, like crash and makedumpfile to
913 * obtain access to symbols that are otherwise very difficult to locate. These
914 * symbols are specifically used so that utilities can access and extract the
915 * dmesg log from a vmcore file after a crash.
917 void log_buf_vmcoreinfo_setup(void)
919 struct dev_printk_info *dev_info = NULL;
921 VMCOREINFO_SYMBOL(prb);
922 VMCOREINFO_SYMBOL(printk_rb_static);
923 VMCOREINFO_SYMBOL(clear_seq);
926 * Export struct size and field offsets. User space tools can
927 * parse it and detect any changes to structure down the line.
930 VMCOREINFO_STRUCT_SIZE(printk_ringbuffer);
931 VMCOREINFO_OFFSET(printk_ringbuffer, desc_ring);
932 VMCOREINFO_OFFSET(printk_ringbuffer, text_data_ring);
933 VMCOREINFO_OFFSET(printk_ringbuffer, fail);
935 VMCOREINFO_STRUCT_SIZE(prb_desc_ring);
936 VMCOREINFO_OFFSET(prb_desc_ring, count_bits);
937 VMCOREINFO_OFFSET(prb_desc_ring, descs);
938 VMCOREINFO_OFFSET(prb_desc_ring, infos);
939 VMCOREINFO_OFFSET(prb_desc_ring, head_id);
940 VMCOREINFO_OFFSET(prb_desc_ring, tail_id);
942 VMCOREINFO_STRUCT_SIZE(prb_desc);
943 VMCOREINFO_OFFSET(prb_desc, state_var);
944 VMCOREINFO_OFFSET(prb_desc, text_blk_lpos);
946 VMCOREINFO_STRUCT_SIZE(prb_data_blk_lpos);
947 VMCOREINFO_OFFSET(prb_data_blk_lpos, begin);
948 VMCOREINFO_OFFSET(prb_data_blk_lpos, next);
950 VMCOREINFO_STRUCT_SIZE(printk_info);
951 VMCOREINFO_OFFSET(printk_info, seq);
952 VMCOREINFO_OFFSET(printk_info, ts_nsec);
953 VMCOREINFO_OFFSET(printk_info, text_len);
954 VMCOREINFO_OFFSET(printk_info, caller_id);
955 VMCOREINFO_OFFSET(printk_info, dev_info);
957 VMCOREINFO_STRUCT_SIZE(dev_printk_info);
958 VMCOREINFO_OFFSET(dev_printk_info, subsystem);
959 VMCOREINFO_LENGTH(printk_info_subsystem, sizeof(dev_info->subsystem));
960 VMCOREINFO_OFFSET(dev_printk_info, device);
961 VMCOREINFO_LENGTH(printk_info_device, sizeof(dev_info->device));
963 VMCOREINFO_STRUCT_SIZE(prb_data_ring);
964 VMCOREINFO_OFFSET(prb_data_ring, size_bits);
965 VMCOREINFO_OFFSET(prb_data_ring, data);
966 VMCOREINFO_OFFSET(prb_data_ring, head_lpos);
967 VMCOREINFO_OFFSET(prb_data_ring, tail_lpos);
969 VMCOREINFO_SIZE(atomic_long_t);
970 VMCOREINFO_TYPE_OFFSET(atomic_long_t, counter);
972 VMCOREINFO_STRUCT_SIZE(latched_seq);
973 VMCOREINFO_OFFSET(latched_seq, val);
977 /* requested log_buf_len from kernel cmdline */
978 static unsigned long __initdata new_log_buf_len;
980 /* we practice scaling the ring buffer by powers of 2 */
981 static void __init log_buf_len_update(u64 size)
983 if (size > (u64)LOG_BUF_LEN_MAX) {
984 size = (u64)LOG_BUF_LEN_MAX;
985 pr_err("log_buf over 2G is not supported.\n");
989 size = roundup_pow_of_two(size);
990 if (size > log_buf_len)
991 new_log_buf_len = (unsigned long)size;
994 /* save requested log_buf_len since it's too early to process it */
995 static int __init log_buf_len_setup(char *str)
1002 size = memparse(str, &str);
1004 log_buf_len_update(size);
1008 early_param("log_buf_len", log_buf_len_setup);
1011 #define __LOG_CPU_MAX_BUF_LEN (1 << CONFIG_LOG_CPU_MAX_BUF_SHIFT)
1013 static void __init log_buf_add_cpu(void)
1015 unsigned int cpu_extra;
1018 * archs should set up cpu_possible_bits properly with
1019 * set_cpu_possible() after setup_arch() but just in
1020 * case lets ensure this is valid.
1022 if (num_possible_cpus() == 1)
1025 cpu_extra = (num_possible_cpus() - 1) * __LOG_CPU_MAX_BUF_LEN;
1027 /* by default this will only continue through for large > 64 CPUs */
1028 if (cpu_extra <= __LOG_BUF_LEN / 2)
1031 pr_info("log_buf_len individual max cpu contribution: %d bytes\n",
1032 __LOG_CPU_MAX_BUF_LEN);
1033 pr_info("log_buf_len total cpu_extra contributions: %d bytes\n",
1035 pr_info("log_buf_len min size: %d bytes\n", __LOG_BUF_LEN);
1037 log_buf_len_update(cpu_extra + __LOG_BUF_LEN);
1039 #else /* !CONFIG_SMP */
1040 static inline void log_buf_add_cpu(void) {}
1041 #endif /* CONFIG_SMP */
1043 static void __init set_percpu_data_ready(void)
1046 /* Make sure we set this flag only after printk_safe() init is done */
1048 __printk_percpu_data_ready = true;
1051 static unsigned int __init add_to_rb(struct printk_ringbuffer *rb,
1052 struct printk_record *r)
1054 struct prb_reserved_entry e;
1055 struct printk_record dest_r;
1057 prb_rec_init_wr(&dest_r, r->info->text_len);
1059 if (!prb_reserve(&e, rb, &dest_r))
1062 memcpy(&dest_r.text_buf[0], &r->text_buf[0], r->info->text_len);
1063 dest_r.info->text_len = r->info->text_len;
1064 dest_r.info->facility = r->info->facility;
1065 dest_r.info->level = r->info->level;
1066 dest_r.info->flags = r->info->flags;
1067 dest_r.info->ts_nsec = r->info->ts_nsec;
1068 dest_r.info->caller_id = r->info->caller_id;
1069 memcpy(&dest_r.info->dev_info, &r->info->dev_info, sizeof(dest_r.info->dev_info));
1071 prb_final_commit(&e);
1073 return prb_record_text_space(&e);
1076 static char setup_text_buf[LOG_LINE_MAX] __initdata;
1078 void __init setup_log_buf(int early)
1080 struct printk_info *new_infos;
1081 unsigned int new_descs_count;
1082 struct prb_desc *new_descs;
1083 struct printk_info info;
1084 struct printk_record r;
1085 size_t new_descs_size;
1086 size_t new_infos_size;
1087 unsigned long flags;
1093 * Some archs call setup_log_buf() multiple times - first is very
1094 * early, e.g. from setup_arch(), and second - when percpu_areas
1098 set_percpu_data_ready();
1100 if (log_buf != __log_buf)
1103 if (!early && !new_log_buf_len)
1106 if (!new_log_buf_len)
1109 new_descs_count = new_log_buf_len >> PRB_AVGBITS;
1110 if (new_descs_count == 0) {
1111 pr_err("new_log_buf_len: %lu too small\n", new_log_buf_len);
1115 new_log_buf = memblock_alloc(new_log_buf_len, LOG_ALIGN);
1116 if (unlikely(!new_log_buf)) {
1117 pr_err("log_buf_len: %lu text bytes not available\n",
1122 new_descs_size = new_descs_count * sizeof(struct prb_desc);
1123 new_descs = memblock_alloc(new_descs_size, LOG_ALIGN);
1124 if (unlikely(!new_descs)) {
1125 pr_err("log_buf_len: %zu desc bytes not available\n",
1127 goto err_free_log_buf;
1130 new_infos_size = new_descs_count * sizeof(struct printk_info);
1131 new_infos = memblock_alloc(new_infos_size, LOG_ALIGN);
1132 if (unlikely(!new_infos)) {
1133 pr_err("log_buf_len: %zu info bytes not available\n",
1135 goto err_free_descs;
1138 prb_rec_init_rd(&r, &info, &setup_text_buf[0], sizeof(setup_text_buf));
1140 prb_init(&printk_rb_dynamic,
1141 new_log_buf, ilog2(new_log_buf_len),
1142 new_descs, ilog2(new_descs_count),
1145 printk_safe_enter_irqsave(flags);
1147 log_buf_len = new_log_buf_len;
1148 log_buf = new_log_buf;
1149 new_log_buf_len = 0;
1151 free = __LOG_BUF_LEN;
1152 prb_for_each_record(0, &printk_rb_static, seq, &r)
1153 free -= add_to_rb(&printk_rb_dynamic, &r);
1156 * This is early enough that everything is still running on the
1157 * boot CPU and interrupts are disabled. So no new messages will
1158 * appear during the transition to the dynamic buffer.
1160 prb = &printk_rb_dynamic;
1162 printk_safe_exit_irqrestore(flags);
1164 if (seq != prb_next_seq(&printk_rb_static)) {
1165 pr_err("dropped %llu messages\n",
1166 prb_next_seq(&printk_rb_static) - seq);
1169 pr_info("log_buf_len: %u bytes\n", log_buf_len);
1170 pr_info("early log buf free: %u(%u%%)\n",
1171 free, (free * 100) / __LOG_BUF_LEN);
1175 memblock_free(__pa(new_descs), new_descs_size);
1177 memblock_free(__pa(new_log_buf), new_log_buf_len);
1180 static bool __read_mostly ignore_loglevel;
1182 static int __init ignore_loglevel_setup(char *str)
1184 ignore_loglevel = true;
1185 pr_info("debug: ignoring loglevel setting.\n");
1190 early_param("ignore_loglevel", ignore_loglevel_setup);
1191 module_param(ignore_loglevel, bool, S_IRUGO | S_IWUSR);
1192 MODULE_PARM_DESC(ignore_loglevel,
1193 "ignore loglevel setting (prints all kernel messages to the console)");
1195 static bool suppress_message_printing(int level)
1197 return (level >= console_loglevel && !ignore_loglevel);
1200 #ifdef CONFIG_BOOT_PRINTK_DELAY
1202 static int boot_delay; /* msecs delay after each printk during bootup */
1203 static unsigned long long loops_per_msec; /* based on boot_delay */
1205 static int __init boot_delay_setup(char *str)
1209 lpj = preset_lpj ? preset_lpj : 1000000; /* some guess */
1210 loops_per_msec = (unsigned long long)lpj / 1000 * HZ;
1212 get_option(&str, &boot_delay);
1213 if (boot_delay > 10 * 1000)
1216 pr_debug("boot_delay: %u, preset_lpj: %ld, lpj: %lu, "
1217 "HZ: %d, loops_per_msec: %llu\n",
1218 boot_delay, preset_lpj, lpj, HZ, loops_per_msec);
1221 early_param("boot_delay", boot_delay_setup);
1223 static void boot_delay_msec(int level)
1225 unsigned long long k;
1226 unsigned long timeout;
1228 if ((boot_delay == 0 || system_state >= SYSTEM_RUNNING)
1229 || suppress_message_printing(level)) {
1233 k = (unsigned long long)loops_per_msec * boot_delay;
1235 timeout = jiffies + msecs_to_jiffies(boot_delay);
1240 * use (volatile) jiffies to prevent
1241 * compiler reduction; loop termination via jiffies
1242 * is secondary and may or may not happen.
1244 if (time_after(jiffies, timeout))
1246 touch_nmi_watchdog();
1250 static inline void boot_delay_msec(int level)
1255 static bool printk_time = IS_ENABLED(CONFIG_PRINTK_TIME);
1256 module_param_named(time, printk_time, bool, S_IRUGO | S_IWUSR);
1258 static size_t print_syslog(unsigned int level, char *buf)
1260 return sprintf(buf, "<%u>", level);
1263 static size_t print_time(u64 ts, char *buf)
1265 unsigned long rem_nsec = do_div(ts, 1000000000);
1267 return sprintf(buf, "[%5lu.%06lu]",
1268 (unsigned long)ts, rem_nsec / 1000);
1271 #ifdef CONFIG_PRINTK_CALLER
1272 static size_t print_caller(u32 id, char *buf)
1276 snprintf(caller, sizeof(caller), "%c%u",
1277 id & 0x80000000 ? 'C' : 'T', id & ~0x80000000);
1278 return sprintf(buf, "[%6s]", caller);
1281 #define print_caller(id, buf) 0
1284 static size_t info_print_prefix(const struct printk_info *info, bool syslog,
1285 bool time, char *buf)
1290 len = print_syslog((info->facility << 3) | info->level, buf);
1293 len += print_time(info->ts_nsec, buf + len);
1295 len += print_caller(info->caller_id, buf + len);
1297 if (IS_ENABLED(CONFIG_PRINTK_CALLER) || time) {
1306 * Prepare the record for printing. The text is shifted within the given
1307 * buffer to avoid a need for another one. The following operations are
1310 * - Add prefix for each line.
1311 * - Drop truncated lines that no longer fit into the buffer.
1312 * - Add the trailing newline that has been removed in vprintk_store().
1313 * - Add a string terminator.
1315 * Since the produced string is always terminated, the maximum possible
1316 * return value is @r->text_buf_size - 1;
1318 * Return: The length of the updated/prepared text, including the added
1319 * prefixes and the newline. The terminator is not counted. The dropped
1320 * line(s) are not counted.
1322 static size_t record_print_text(struct printk_record *r, bool syslog,
1325 size_t text_len = r->info->text_len;
1326 size_t buf_size = r->text_buf_size;
1327 char *text = r->text_buf;
1328 char prefix[PREFIX_MAX];
1329 bool truncated = false;
1336 * If the message was truncated because the buffer was not large
1337 * enough, treat the available text as if it were the full text.
1339 if (text_len > buf_size)
1340 text_len = buf_size;
1342 prefix_len = info_print_prefix(r->info, syslog, time, prefix);
1345 * @text_len: bytes of unprocessed text
1346 * @line_len: bytes of current line _without_ newline
1347 * @text: pointer to beginning of current line
1348 * @len: number of bytes prepared in r->text_buf
1351 next = memchr(text, '\n', text_len);
1353 line_len = next - text;
1355 /* Drop truncated line(s). */
1358 line_len = text_len;
1362 * Truncate the text if there is not enough space to add the
1363 * prefix and a trailing newline and a terminator.
1365 if (len + prefix_len + text_len + 1 + 1 > buf_size) {
1366 /* Drop even the current line if no space. */
1367 if (len + prefix_len + line_len + 1 + 1 > buf_size)
1370 text_len = buf_size - len - prefix_len - 1 - 1;
1374 memmove(text + prefix_len, text, text_len);
1375 memcpy(text, prefix, prefix_len);
1378 * Increment the prepared length to include the text and
1379 * prefix that were just moved+copied. Also increment for the
1380 * newline at the end of this line. If this is the last line,
1381 * there is no newline, but it will be added immediately below.
1383 len += prefix_len + line_len + 1;
1384 if (text_len == line_len) {
1386 * This is the last line. Add the trailing newline
1387 * removed in vprintk_store().
1389 text[prefix_len + line_len] = '\n';
1394 * Advance beyond the added prefix and the related line with
1397 text += prefix_len + line_len + 1;
1400 * The remaining text has only decreased by the line with its
1403 * Note that @text_len can become zero. It happens when @text
1404 * ended with a newline (either due to truncation or the
1405 * original string ending with "\n\n"). The loop is correctly
1406 * repeated and (if not truncated) an empty line with a prefix
1409 text_len -= line_len + 1;
1413 * If a buffer was provided, it will be terminated. Space for the
1414 * string terminator is guaranteed to be available. The terminator is
1415 * not counted in the return value.
1418 r->text_buf[len] = 0;
1423 static size_t get_record_print_text_size(struct printk_info *info,
1424 unsigned int line_count,
1425 bool syslog, bool time)
1427 char prefix[PREFIX_MAX];
1430 prefix_len = info_print_prefix(info, syslog, time, prefix);
1433 * Each line will be preceded with a prefix. The intermediate
1434 * newlines are already within the text, but a final trailing
1435 * newline will be added.
1437 return ((prefix_len * line_count) + info->text_len + 1);
1441 * Beginning with @start_seq, find the first record where it and all following
1442 * records up to (but not including) @max_seq fit into @size.
1444 * @max_seq is simply an upper bound and does not need to exist. If the caller
1445 * does not require an upper bound, -1 can be used for @max_seq.
1447 static u64 find_first_fitting_seq(u64 start_seq, u64 max_seq, size_t size,
1448 bool syslog, bool time)
1450 struct printk_info info;
1451 unsigned int line_count;
1455 /* Determine the size of the records up to @max_seq. */
1456 prb_for_each_info(start_seq, prb, seq, &info, &line_count) {
1457 if (info.seq >= max_seq)
1459 len += get_record_print_text_size(&info, line_count, syslog, time);
1463 * Adjust the upper bound for the next loop to avoid subtracting
1464 * lengths that were never added.
1470 * Move first record forward until length fits into the buffer. Ignore
1471 * newest messages that were not counted in the above cycle. Messages
1472 * might appear and get lost in the meantime. This is a best effort
1473 * that prevents an infinite loop that could occur with a retry.
1475 prb_for_each_info(start_seq, prb, seq, &info, &line_count) {
1476 if (len <= size || info.seq >= max_seq)
1478 len -= get_record_print_text_size(&info, line_count, syslog, time);
1484 static int syslog_print(char __user *buf, int size)
1486 struct printk_info info;
1487 struct printk_record r;
1491 text = kmalloc(CONSOLE_LOG_MAX, GFP_KERNEL);
1495 prb_rec_init_rd(&r, &info, text, CONSOLE_LOG_MAX);
1501 printk_safe_enter_irq();
1502 raw_spin_lock(&syslog_lock);
1503 if (!prb_read_valid(prb, syslog_seq, &r)) {
1504 raw_spin_unlock(&syslog_lock);
1505 printk_safe_exit_irq();
1508 if (r.info->seq != syslog_seq) {
1509 /* message is gone, move to next valid one */
1510 syslog_seq = r.info->seq;
1515 * To keep reading/counting partial line consistent,
1516 * use printk_time value as of the beginning of a line.
1518 if (!syslog_partial)
1519 syslog_time = printk_time;
1521 skip = syslog_partial;
1522 n = record_print_text(&r, true, syslog_time);
1523 if (n - syslog_partial <= size) {
1524 /* message fits into buffer, move forward */
1525 syslog_seq = r.info->seq + 1;
1526 n -= syslog_partial;
1529 /* partial read(), remember position */
1531 syslog_partial += n;
1534 raw_spin_unlock(&syslog_lock);
1535 printk_safe_exit_irq();
1540 if (copy_to_user(buf, text + skip, n)) {
1555 static int syslog_print_all(char __user *buf, int size, bool clear)
1557 struct printk_info info;
1558 struct printk_record r;
1564 text = kmalloc(CONSOLE_LOG_MAX, GFP_KERNEL);
1569 printk_safe_enter_irq();
1571 * Find first record that fits, including all following records,
1572 * into the user-provided buffer for this dump.
1574 seq = find_first_fitting_seq(latched_seq_read_nolock(&clear_seq), -1,
1577 prb_rec_init_rd(&r, &info, text, CONSOLE_LOG_MAX);
1580 prb_for_each_record(seq, prb, seq, &r) {
1583 textlen = record_print_text(&r, true, time);
1585 if (len + textlen > size) {
1590 printk_safe_exit_irq();
1591 if (copy_to_user(buf + len, text, textlen))
1595 printk_safe_enter_irq();
1602 raw_spin_lock(&syslog_lock);
1603 latched_seq_write(&clear_seq, seq);
1604 raw_spin_unlock(&syslog_lock);
1606 printk_safe_exit_irq();
1612 static void syslog_clear(void)
1614 printk_safe_enter_irq();
1615 raw_spin_lock(&syslog_lock);
1616 latched_seq_write(&clear_seq, prb_next_seq(prb));
1617 raw_spin_unlock(&syslog_lock);
1618 printk_safe_exit_irq();
1621 /* Return a consistent copy of @syslog_seq. */
1622 static u64 read_syslog_seq_irq(void)
1626 raw_spin_lock_irq(&syslog_lock);
1628 raw_spin_unlock_irq(&syslog_lock);
1633 int do_syslog(int type, char __user *buf, int len, int source)
1635 struct printk_info info;
1637 static int saved_console_loglevel = LOGLEVEL_DEFAULT;
1640 error = check_syslog_permissions(type, source);
1645 case SYSLOG_ACTION_CLOSE: /* Close log */
1647 case SYSLOG_ACTION_OPEN: /* Open log */
1649 case SYSLOG_ACTION_READ: /* Read from log */
1650 if (!buf || len < 0)
1654 if (!access_ok(buf, len))
1657 error = wait_event_interruptible(log_wait,
1658 prb_read_valid(prb, read_syslog_seq_irq(), NULL));
1661 error = syslog_print(buf, len);
1663 /* Read/clear last kernel messages */
1664 case SYSLOG_ACTION_READ_CLEAR:
1667 /* Read last kernel messages */
1668 case SYSLOG_ACTION_READ_ALL:
1669 if (!buf || len < 0)
1673 if (!access_ok(buf, len))
1675 error = syslog_print_all(buf, len, clear);
1677 /* Clear ring buffer */
1678 case SYSLOG_ACTION_CLEAR:
1681 /* Disable logging to console */
1682 case SYSLOG_ACTION_CONSOLE_OFF:
1683 if (saved_console_loglevel == LOGLEVEL_DEFAULT)
1684 saved_console_loglevel = console_loglevel;
1685 console_loglevel = minimum_console_loglevel;
1687 /* Enable logging to console */
1688 case SYSLOG_ACTION_CONSOLE_ON:
1689 if (saved_console_loglevel != LOGLEVEL_DEFAULT) {
1690 console_loglevel = saved_console_loglevel;
1691 saved_console_loglevel = LOGLEVEL_DEFAULT;
1694 /* Set level of messages printed to console */
1695 case SYSLOG_ACTION_CONSOLE_LEVEL:
1696 if (len < 1 || len > 8)
1698 if (len < minimum_console_loglevel)
1699 len = minimum_console_loglevel;
1700 console_loglevel = len;
1701 /* Implicitly re-enable logging to console */
1702 saved_console_loglevel = LOGLEVEL_DEFAULT;
1704 /* Number of chars in the log buffer */
1705 case SYSLOG_ACTION_SIZE_UNREAD:
1706 printk_safe_enter_irq();
1707 raw_spin_lock(&syslog_lock);
1708 if (!prb_read_valid_info(prb, syslog_seq, &info, NULL)) {
1709 /* No unread messages. */
1710 raw_spin_unlock(&syslog_lock);
1711 printk_safe_exit_irq();
1714 if (info.seq != syslog_seq) {
1715 /* messages are gone, move to first one */
1716 syslog_seq = info.seq;
1719 if (source == SYSLOG_FROM_PROC) {
1721 * Short-cut for poll(/"proc/kmsg") which simply checks
1722 * for pending data, not the size; return the count of
1723 * records, not the length.
1725 error = prb_next_seq(prb) - syslog_seq;
1727 bool time = syslog_partial ? syslog_time : printk_time;
1728 unsigned int line_count;
1731 prb_for_each_info(syslog_seq, prb, seq, &info,
1733 error += get_record_print_text_size(&info, line_count,
1737 error -= syslog_partial;
1739 raw_spin_unlock(&syslog_lock);
1740 printk_safe_exit_irq();
1742 /* Size of the log buffer */
1743 case SYSLOG_ACTION_SIZE_BUFFER:
1744 error = log_buf_len;
1754 SYSCALL_DEFINE3(syslog, int, type, char __user *, buf, int, len)
1756 return do_syslog(type, buf, len, SYSLOG_FROM_READER);
1760 * Special console_lock variants that help to reduce the risk of soft-lockups.
1761 * They allow to pass console_lock to another printk() call using a busy wait.
1764 #ifdef CONFIG_LOCKDEP
1765 static struct lockdep_map console_owner_dep_map = {
1766 .name = "console_owner"
1770 static DEFINE_RAW_SPINLOCK(console_owner_lock);
1771 static struct task_struct *console_owner;
1772 static bool console_waiter;
1775 * console_lock_spinning_enable - mark beginning of code where another
1776 * thread might safely busy wait
1778 * This basically converts console_lock into a spinlock. This marks
1779 * the section where the console_lock owner can not sleep, because
1780 * there may be a waiter spinning (like a spinlock). Also it must be
1781 * ready to hand over the lock at the end of the section.
1783 static void console_lock_spinning_enable(void)
1785 raw_spin_lock(&console_owner_lock);
1786 console_owner = current;
1787 raw_spin_unlock(&console_owner_lock);
1789 /* The waiter may spin on us after setting console_owner */
1790 spin_acquire(&console_owner_dep_map, 0, 0, _THIS_IP_);
1794 * console_lock_spinning_disable_and_check - mark end of code where another
1795 * thread was able to busy wait and check if there is a waiter
1797 * This is called at the end of the section where spinning is allowed.
1798 * It has two functions. First, it is a signal that it is no longer
1799 * safe to start busy waiting for the lock. Second, it checks if
1800 * there is a busy waiter and passes the lock rights to her.
1802 * Important: Callers lose the lock if there was a busy waiter.
1803 * They must not touch items synchronized by console_lock
1806 * Return: 1 if the lock rights were passed, 0 otherwise.
1808 static int console_lock_spinning_disable_and_check(void)
1812 raw_spin_lock(&console_owner_lock);
1813 waiter = READ_ONCE(console_waiter);
1814 console_owner = NULL;
1815 raw_spin_unlock(&console_owner_lock);
1818 spin_release(&console_owner_dep_map, _THIS_IP_);
1822 /* The waiter is now free to continue */
1823 WRITE_ONCE(console_waiter, false);
1825 spin_release(&console_owner_dep_map, _THIS_IP_);
1828 * Hand off console_lock to waiter. The waiter will perform
1829 * the up(). After this, the waiter is the console_lock owner.
1831 mutex_release(&console_lock_dep_map, _THIS_IP_);
1836 * console_trylock_spinning - try to get console_lock by busy waiting
1838 * This allows to busy wait for the console_lock when the current
1839 * owner is running in specially marked sections. It means that
1840 * the current owner is running and cannot reschedule until it
1841 * is ready to lose the lock.
1843 * Return: 1 if we got the lock, 0 othrewise
1845 static int console_trylock_spinning(void)
1847 struct task_struct *owner = NULL;
1850 unsigned long flags;
1852 if (console_trylock())
1855 printk_safe_enter_irqsave(flags);
1857 raw_spin_lock(&console_owner_lock);
1858 owner = READ_ONCE(console_owner);
1859 waiter = READ_ONCE(console_waiter);
1860 if (!waiter && owner && owner != current) {
1861 WRITE_ONCE(console_waiter, true);
1864 raw_spin_unlock(&console_owner_lock);
1867 * If there is an active printk() writing to the
1868 * consoles, instead of having it write our data too,
1869 * see if we can offload that load from the active
1870 * printer, and do some printing ourselves.
1871 * Go into a spin only if there isn't already a waiter
1872 * spinning, and there is an active printer, and
1873 * that active printer isn't us (recursive printk?).
1876 printk_safe_exit_irqrestore(flags);
1880 /* We spin waiting for the owner to release us */
1881 spin_acquire(&console_owner_dep_map, 0, 0, _THIS_IP_);
1882 /* Owner will clear console_waiter on hand off */
1883 while (READ_ONCE(console_waiter))
1885 spin_release(&console_owner_dep_map, _THIS_IP_);
1887 printk_safe_exit_irqrestore(flags);
1889 * The owner passed the console lock to us.
1890 * Since we did not spin on console lock, annotate
1891 * this as a trylock. Otherwise lockdep will
1894 mutex_acquire(&console_lock_dep_map, 0, 1, _THIS_IP_);
1900 * Call the console drivers, asking them to write out
1901 * log_buf[start] to log_buf[end - 1].
1902 * The console_lock must be held.
1904 static void call_console_drivers(const char *ext_text, size_t ext_len,
1905 const char *text, size_t len)
1907 static char dropped_text[64];
1908 size_t dropped_len = 0;
1909 struct console *con;
1911 trace_console_rcuidle(text, len);
1913 if (!console_drivers)
1916 if (console_dropped) {
1917 dropped_len = snprintf(dropped_text, sizeof(dropped_text),
1918 "** %lu printk messages dropped **\n",
1920 console_dropped = 0;
1923 for_each_console(con) {
1924 if (exclusive_console && con != exclusive_console)
1926 if (!(con->flags & CON_ENABLED))
1930 if (!cpu_online(smp_processor_id()) &&
1931 !(con->flags & CON_ANYTIME))
1933 if (con->flags & CON_EXTENDED)
1934 con->write(con, ext_text, ext_len);
1937 con->write(con, dropped_text, dropped_len);
1938 con->write(con, text, len);
1943 int printk_delay_msec __read_mostly;
1945 static inline void printk_delay(void)
1947 if (unlikely(printk_delay_msec)) {
1948 int m = printk_delay_msec;
1952 touch_nmi_watchdog();
1957 static inline u32 printk_caller_id(void)
1959 return in_task() ? task_pid_nr(current) :
1960 0x80000000 + raw_smp_processor_id();
1964 * parse_prefix - Parse level and control flags.
1966 * @text: The terminated text message.
1967 * @level: A pointer to the current level value, will be updated.
1968 * @lflags: A pointer to the current log flags, will be updated.
1970 * @level may be NULL if the caller is not interested in the parsed value.
1971 * Otherwise the variable pointed to by @level must be set to
1972 * LOGLEVEL_DEFAULT in order to be updated with the parsed value.
1974 * @lflags may be NULL if the caller is not interested in the parsed value.
1975 * Otherwise the variable pointed to by @lflags will be OR'd with the parsed
1978 * Return: The length of the parsed level and control flags.
1980 static u16 parse_prefix(char *text, int *level, enum log_flags *lflags)
1986 kern_level = printk_get_level(text);
1990 switch (kern_level) {
1992 if (level && *level == LOGLEVEL_DEFAULT)
1993 *level = kern_level - '0';
1995 case 'c': /* KERN_CONT */
1997 *lflags |= LOG_CONT;
2007 static u16 printk_sprint(char *text, u16 size, int facility, enum log_flags *lflags,
2008 const char *fmt, va_list args)
2012 text_len = vscnprintf(text, size, fmt, args);
2014 /* Mark and strip a trailing newline. */
2015 if (text_len && text[text_len - 1] == '\n') {
2017 *lflags |= LOG_NEWLINE;
2020 /* Strip log level and control flags. */
2021 if (facility == 0) {
2024 prefix_len = parse_prefix(text, NULL, NULL);
2026 text_len -= prefix_len;
2027 memmove(text, text + prefix_len, text_len);
2035 int vprintk_store(int facility, int level,
2036 const struct dev_printk_info *dev_info,
2037 const char *fmt, va_list args)
2039 const u32 caller_id = printk_caller_id();
2040 struct prb_reserved_entry e;
2041 enum log_flags lflags = 0;
2042 struct printk_record r;
2043 u16 trunc_msg_len = 0;
2051 * Since the duration of printk() can vary depending on the message
2052 * and state of the ringbuffer, grab the timestamp now so that it is
2053 * close to the call of printk(). This provides a more deterministic
2054 * timestamp with respect to the caller.
2056 ts_nsec = local_clock();
2059 * The sprintf needs to come first since the syslog prefix might be
2060 * passed in as a parameter. An extra byte must be reserved so that
2061 * later the vscnprintf() into the reserved buffer has room for the
2062 * terminating '\0', which is not counted by vsnprintf().
2064 va_copy(args2, args);
2065 reserve_size = vsnprintf(&prefix_buf[0], sizeof(prefix_buf), fmt, args2) + 1;
2068 if (reserve_size > LOG_LINE_MAX)
2069 reserve_size = LOG_LINE_MAX;
2071 /* Extract log level or control flags. */
2073 parse_prefix(&prefix_buf[0], &level, &lflags);
2075 if (level == LOGLEVEL_DEFAULT)
2076 level = default_message_loglevel;
2079 lflags |= LOG_NEWLINE;
2081 if (lflags & LOG_CONT) {
2082 prb_rec_init_wr(&r, reserve_size);
2083 if (prb_reserve_in_last(&e, prb, &r, caller_id, LOG_LINE_MAX)) {
2084 text_len = printk_sprint(&r.text_buf[r.info->text_len], reserve_size,
2085 facility, &lflags, fmt, args);
2086 r.info->text_len += text_len;
2088 if (lflags & LOG_NEWLINE) {
2089 r.info->flags |= LOG_NEWLINE;
2090 prb_final_commit(&e);
2100 * Explicitly initialize the record before every prb_reserve() call.
2101 * prb_reserve_in_last() and prb_reserve() purposely invalidate the
2102 * structure when they fail.
2104 prb_rec_init_wr(&r, reserve_size);
2105 if (!prb_reserve(&e, prb, &r)) {
2106 /* truncate the message if it is too long for empty buffer */
2107 truncate_msg(&reserve_size, &trunc_msg_len);
2109 prb_rec_init_wr(&r, reserve_size + trunc_msg_len);
2110 if (!prb_reserve(&e, prb, &r))
2115 text_len = printk_sprint(&r.text_buf[0], reserve_size, facility, &lflags, fmt, args);
2117 memcpy(&r.text_buf[text_len], trunc_msg, trunc_msg_len);
2118 r.info->text_len = text_len + trunc_msg_len;
2119 r.info->facility = facility;
2120 r.info->level = level & 7;
2121 r.info->flags = lflags & 0x1f;
2122 r.info->ts_nsec = ts_nsec;
2123 r.info->caller_id = caller_id;
2125 memcpy(&r.info->dev_info, dev_info, sizeof(r.info->dev_info));
2127 /* A message without a trailing newline can be continued. */
2128 if (!(lflags & LOG_NEWLINE))
2131 prb_final_commit(&e);
2133 return (text_len + trunc_msg_len);
2136 asmlinkage int vprintk_emit(int facility, int level,
2137 const struct dev_printk_info *dev_info,
2138 const char *fmt, va_list args)
2141 bool in_sched = false;
2142 unsigned long flags;
2144 /* Suppress unimportant messages after panic happens */
2145 if (unlikely(suppress_printk))
2148 if (level == LOGLEVEL_SCHED) {
2149 level = LOGLEVEL_DEFAULT;
2153 boot_delay_msec(level);
2156 printk_safe_enter_irqsave(flags);
2157 printed_len = vprintk_store(facility, level, dev_info, fmt, args);
2158 printk_safe_exit_irqrestore(flags);
2160 /* If called from the scheduler, we can not call up(). */
2163 * Disable preemption to avoid being preempted while holding
2164 * console_sem which would prevent anyone from printing to
2169 * Try to acquire and then immediately release the console
2170 * semaphore. The release will print out buffers and wake up
2171 * /dev/kmsg and syslog() users.
2173 if (console_trylock_spinning())
2181 EXPORT_SYMBOL(vprintk_emit);
2183 int vprintk_default(const char *fmt, va_list args)
2185 return vprintk_emit(0, LOGLEVEL_DEFAULT, NULL, fmt, args);
2187 EXPORT_SYMBOL_GPL(vprintk_default);
2190 * printk - print a kernel message
2191 * @fmt: format string
2193 * This is printk(). It can be called from any context. We want it to work.
2195 * We try to grab the console_lock. If we succeed, it's easy - we log the
2196 * output and call the console drivers. If we fail to get the semaphore, we
2197 * place the output into the log buffer and return. The current holder of
2198 * the console_sem will notice the new output in console_unlock(); and will
2199 * send it to the consoles before releasing the lock.
2201 * One effect of this deferred printing is that code which calls printk() and
2202 * then changes console_loglevel may break. This is because console_loglevel
2203 * is inspected when the actual printing occurs.
2208 * See the vsnprintf() documentation for format string extensions over C99.
2210 asmlinkage __visible int printk(const char *fmt, ...)
2215 va_start(args, fmt);
2216 r = vprintk(fmt, args);
2221 EXPORT_SYMBOL(printk);
2223 #else /* CONFIG_PRINTK */
2225 #define CONSOLE_LOG_MAX 0
2226 #define printk_time false
2228 #define prb_read_valid(rb, seq, r) false
2229 #define prb_first_valid_seq(rb) 0
2231 static u64 syslog_seq;
2232 static u64 console_seq;
2233 static u64 exclusive_console_stop_seq;
2234 static unsigned long console_dropped;
2236 static size_t record_print_text(const struct printk_record *r,
2237 bool syslog, bool time)
2241 static ssize_t info_print_ext_header(char *buf, size_t size,
2242 struct printk_info *info)
2246 static ssize_t msg_print_ext_body(char *buf, size_t size,
2247 char *text, size_t text_len,
2248 struct dev_printk_info *dev_info) { return 0; }
2249 static void console_lock_spinning_enable(void) { }
2250 static int console_lock_spinning_disable_and_check(void) { return 0; }
2251 static void call_console_drivers(const char *ext_text, size_t ext_len,
2252 const char *text, size_t len) {}
2253 static bool suppress_message_printing(int level) { return false; }
2255 #endif /* CONFIG_PRINTK */
2257 #ifdef CONFIG_EARLY_PRINTK
2258 struct console *early_console;
2260 asmlinkage __visible void early_printk(const char *fmt, ...)
2270 n = vscnprintf(buf, sizeof(buf), fmt, ap);
2273 early_console->write(early_console, buf, n);
2277 static int __add_preferred_console(char *name, int idx, char *options,
2278 char *brl_options, bool user_specified)
2280 struct console_cmdline *c;
2284 * See if this tty is not yet registered, and
2285 * if we have a slot free.
2287 for (i = 0, c = console_cmdline;
2288 i < MAX_CMDLINECONSOLES && c->name[0];
2290 if (strcmp(c->name, name) == 0 && c->index == idx) {
2292 preferred_console = i;
2294 c->user_specified = true;
2298 if (i == MAX_CMDLINECONSOLES)
2301 preferred_console = i;
2302 strlcpy(c->name, name, sizeof(c->name));
2303 c->options = options;
2304 c->user_specified = user_specified;
2305 braille_set_options(c, brl_options);
2311 static int __init console_msg_format_setup(char *str)
2313 if (!strcmp(str, "syslog"))
2314 console_msg_format = MSG_FORMAT_SYSLOG;
2315 if (!strcmp(str, "default"))
2316 console_msg_format = MSG_FORMAT_DEFAULT;
2319 __setup("console_msg_format=", console_msg_format_setup);
2322 * Set up a console. Called via do_early_param() in init/main.c
2323 * for each "console=" parameter in the boot command line.
2325 static int __init console_setup(char *str)
2327 char buf[sizeof(console_cmdline[0].name) + 4]; /* 4 for "ttyS" */
2328 char *s, *options, *brl_options = NULL;
2332 * console="" or console=null have been suggested as a way to
2333 * disable console output. Use ttynull that has been created
2334 * for exactly this purpose.
2336 if (str[0] == 0 || strcmp(str, "null") == 0) {
2337 __add_preferred_console("ttynull", 0, NULL, NULL, true);
2341 if (_braille_console_setup(&str, &brl_options))
2345 * Decode str into name, index, options.
2347 if (str[0] >= '0' && str[0] <= '9') {
2348 strcpy(buf, "ttyS");
2349 strncpy(buf + 4, str, sizeof(buf) - 5);
2351 strncpy(buf, str, sizeof(buf) - 1);
2353 buf[sizeof(buf) - 1] = 0;
2354 options = strchr(str, ',');
2358 if (!strcmp(str, "ttya"))
2359 strcpy(buf, "ttyS0");
2360 if (!strcmp(str, "ttyb"))
2361 strcpy(buf, "ttyS1");
2363 for (s = buf; *s; s++)
2364 if (isdigit(*s) || *s == ',')
2366 idx = simple_strtoul(s, NULL, 10);
2369 __add_preferred_console(buf, idx, options, brl_options, true);
2370 console_set_on_cmdline = 1;
2373 __setup("console=", console_setup);
2376 * add_preferred_console - add a device to the list of preferred consoles.
2377 * @name: device name
2378 * @idx: device index
2379 * @options: options for this console
2381 * The last preferred console added will be used for kernel messages
2382 * and stdin/out/err for init. Normally this is used by console_setup
2383 * above to handle user-supplied console arguments; however it can also
2384 * be used by arch-specific code either to override the user or more
2385 * commonly to provide a default console (ie from PROM variables) when
2386 * the user has not supplied one.
2388 int add_preferred_console(char *name, int idx, char *options)
2390 return __add_preferred_console(name, idx, options, NULL, false);
2393 bool console_suspend_enabled = true;
2394 EXPORT_SYMBOL(console_suspend_enabled);
2396 static int __init console_suspend_disable(char *str)
2398 console_suspend_enabled = false;
2401 __setup("no_console_suspend", console_suspend_disable);
2402 module_param_named(console_suspend, console_suspend_enabled,
2403 bool, S_IRUGO | S_IWUSR);
2404 MODULE_PARM_DESC(console_suspend, "suspend console during suspend"
2405 " and hibernate operations");
2408 * suspend_console - suspend the console subsystem
2410 * This disables printk() while we go into suspend states
2412 void suspend_console(void)
2414 if (!console_suspend_enabled)
2416 pr_info("Suspending console(s) (use no_console_suspend to debug)\n");
2418 console_suspended = 1;
2422 void resume_console(void)
2424 if (!console_suspend_enabled)
2427 console_suspended = 0;
2432 * console_cpu_notify - print deferred console messages after CPU hotplug
2435 * If printk() is called from a CPU that is not online yet, the messages
2436 * will be printed on the console only if there are CON_ANYTIME consoles.
2437 * This function is called when a new CPU comes online (or fails to come
2438 * up) or goes offline.
2440 static int console_cpu_notify(unsigned int cpu)
2442 if (!cpuhp_tasks_frozen) {
2443 /* If trylock fails, someone else is doing the printing */
2444 if (console_trylock())
2451 * console_lock - lock the console system for exclusive use.
2453 * Acquires a lock which guarantees that the caller has
2454 * exclusive access to the console system and the console_drivers list.
2456 * Can sleep, returns nothing.
2458 void console_lock(void)
2463 if (console_suspended)
2466 console_may_schedule = 1;
2468 EXPORT_SYMBOL(console_lock);
2471 * console_trylock - try to lock the console system for exclusive use.
2473 * Try to acquire a lock which guarantees that the caller has exclusive
2474 * access to the console system and the console_drivers list.
2476 * returns 1 on success, and 0 on failure to acquire the lock.
2478 int console_trylock(void)
2480 if (down_trylock_console_sem())
2482 if (console_suspended) {
2487 console_may_schedule = 0;
2490 EXPORT_SYMBOL(console_trylock);
2492 int is_console_locked(void)
2494 return console_locked;
2496 EXPORT_SYMBOL(is_console_locked);
2499 * Check if we have any console that is capable of printing while cpu is
2500 * booting or shutting down. Requires console_sem.
2502 static int have_callable_console(void)
2504 struct console *con;
2506 for_each_console(con)
2507 if ((con->flags & CON_ENABLED) &&
2508 (con->flags & CON_ANYTIME))
2515 * Can we actually use the console at this time on this cpu?
2517 * Console drivers may assume that per-cpu resources have been allocated. So
2518 * unless they're explicitly marked as being able to cope (CON_ANYTIME) don't
2519 * call them until this CPU is officially up.
2521 static inline int can_use_console(void)
2523 return cpu_online(raw_smp_processor_id()) || have_callable_console();
2527 * console_unlock - unlock the console system
2529 * Releases the console_lock which the caller holds on the console system
2530 * and the console driver list.
2532 * While the console_lock was held, console output may have been buffered
2533 * by printk(). If this is the case, console_unlock(); emits
2534 * the output prior to releasing the lock.
2536 * If there is output waiting, we wake /dev/kmsg and syslog() users.
2538 * console_unlock(); may be called from any context.
2540 void console_unlock(void)
2542 static char ext_text[CONSOLE_EXT_LOG_MAX];
2543 static char text[CONSOLE_LOG_MAX];
2544 unsigned long flags;
2545 bool do_cond_resched, retry;
2546 struct printk_info info;
2547 struct printk_record r;
2549 if (console_suspended) {
2554 prb_rec_init_rd(&r, &info, text, sizeof(text));
2557 * Console drivers are called with interrupts disabled, so
2558 * @console_may_schedule should be cleared before; however, we may
2559 * end up dumping a lot of lines, for example, if called from
2560 * console registration path, and should invoke cond_resched()
2561 * between lines if allowable. Not doing so can cause a very long
2562 * scheduling stall on a slow console leading to RCU stall and
2563 * softlockup warnings which exacerbate the issue with more
2564 * messages practically incapacitating the system.
2566 * console_trylock() is not able to detect the preemptive
2567 * context reliably. Therefore the value must be stored before
2568 * and cleared after the "again" goto label.
2570 do_cond_resched = console_may_schedule;
2572 console_may_schedule = 0;
2575 * We released the console_sem lock, so we need to recheck if
2576 * cpu is online and (if not) is there at least one CON_ANYTIME
2579 if (!can_use_console()) {
2589 printk_safe_enter_irqsave(flags);
2591 if (!prb_read_valid(prb, console_seq, &r))
2594 if (console_seq != r.info->seq) {
2595 console_dropped += r.info->seq - console_seq;
2596 console_seq = r.info->seq;
2599 if (suppress_message_printing(r.info->level)) {
2601 * Skip record we have buffered and already printed
2602 * directly to the console when we received it, and
2603 * record that has level above the console loglevel.
2609 /* Output to all consoles once old messages replayed. */
2610 if (unlikely(exclusive_console &&
2611 console_seq >= exclusive_console_stop_seq)) {
2612 exclusive_console = NULL;
2616 * Handle extended console text first because later
2617 * record_print_text() will modify the record buffer in-place.
2619 if (nr_ext_console_drivers) {
2620 ext_len = info_print_ext_header(ext_text,
2623 ext_len += msg_print_ext_body(ext_text + ext_len,
2624 sizeof(ext_text) - ext_len,
2629 len = record_print_text(&r,
2630 console_msg_format & MSG_FORMAT_SYSLOG,
2635 * While actively printing out messages, if another printk()
2636 * were to occur on another CPU, it may wait for this one to
2637 * finish. This task can not be preempted if there is a
2638 * waiter waiting to take over.
2640 console_lock_spinning_enable();
2642 stop_critical_timings(); /* don't trace print latency */
2643 call_console_drivers(ext_text, ext_len, text, len);
2644 start_critical_timings();
2646 if (console_lock_spinning_disable_and_check()) {
2647 printk_safe_exit_irqrestore(flags);
2651 printk_safe_exit_irqrestore(flags);
2653 if (do_cond_resched)
2662 * Someone could have filled up the buffer again, so re-check if there's
2663 * something to flush. In case we cannot trylock the console_sem again,
2664 * there's a new owner and the console_unlock() from them will do the
2665 * flush, no worries.
2667 retry = prb_read_valid(prb, console_seq, NULL);
2668 printk_safe_exit_irqrestore(flags);
2670 if (retry && console_trylock())
2673 EXPORT_SYMBOL(console_unlock);
2676 * console_conditional_schedule - yield the CPU if required
2678 * If the console code is currently allowed to sleep, and
2679 * if this CPU should yield the CPU to another task, do
2682 * Must be called within console_lock();.
2684 void __sched console_conditional_schedule(void)
2686 if (console_may_schedule)
2689 EXPORT_SYMBOL(console_conditional_schedule);
2691 void console_unblank(void)
2696 * console_unblank can no longer be called in interrupt context unless
2697 * oops_in_progress is set to 1..
2699 if (oops_in_progress) {
2700 if (down_trylock_console_sem() != 0)
2706 console_may_schedule = 0;
2708 if ((c->flags & CON_ENABLED) && c->unblank)
2714 * console_flush_on_panic - flush console content on panic
2715 * @mode: flush all messages in buffer or just the pending ones
2717 * Immediately output all pending messages no matter what.
2719 void console_flush_on_panic(enum con_flush_mode mode)
2722 * If someone else is holding the console lock, trylock will fail
2723 * and may_schedule may be set. Ignore and proceed to unlock so
2724 * that messages are flushed out. As this can be called from any
2725 * context and we don't want to get preempted while flushing,
2726 * ensure may_schedule is cleared.
2729 console_may_schedule = 0;
2731 if (mode == CONSOLE_REPLAY_ALL) {
2732 unsigned long flags;
2734 printk_safe_enter_irqsave(flags);
2735 console_seq = prb_first_valid_seq(prb);
2736 printk_safe_exit_irqrestore(flags);
2742 * Return the console tty driver structure and its associated index
2744 struct tty_driver *console_device(int *index)
2747 struct tty_driver *driver = NULL;
2750 for_each_console(c) {
2753 driver = c->device(c, index);
2762 * Prevent further output on the passed console device so that (for example)
2763 * serial drivers can disable console output before suspending a port, and can
2764 * re-enable output afterwards.
2766 void console_stop(struct console *console)
2769 console->flags &= ~CON_ENABLED;
2772 EXPORT_SYMBOL(console_stop);
2774 void console_start(struct console *console)
2777 console->flags |= CON_ENABLED;
2780 EXPORT_SYMBOL(console_start);
2782 static int __read_mostly keep_bootcon;
2784 static int __init keep_bootcon_setup(char *str)
2787 pr_info("debug: skip boot console de-registration.\n");
2792 early_param("keep_bootcon", keep_bootcon_setup);
2795 * This is called by register_console() to try to match
2796 * the newly registered console with any of the ones selected
2797 * by either the command line or add_preferred_console() and
2800 * Care need to be taken with consoles that are statically
2801 * enabled such as netconsole
2803 static int try_enable_new_console(struct console *newcon, bool user_specified)
2805 struct console_cmdline *c;
2808 for (i = 0, c = console_cmdline;
2809 i < MAX_CMDLINECONSOLES && c->name[0];
2811 if (c->user_specified != user_specified)
2813 if (!newcon->match ||
2814 newcon->match(newcon, c->name, c->index, c->options) != 0) {
2815 /* default matching */
2816 BUILD_BUG_ON(sizeof(c->name) != sizeof(newcon->name));
2817 if (strcmp(c->name, newcon->name) != 0)
2819 if (newcon->index >= 0 &&
2820 newcon->index != c->index)
2822 if (newcon->index < 0)
2823 newcon->index = c->index;
2825 if (_braille_register_console(newcon, c))
2828 if (newcon->setup &&
2829 (err = newcon->setup(newcon, c->options)) != 0)
2832 newcon->flags |= CON_ENABLED;
2833 if (i == preferred_console) {
2834 newcon->flags |= CON_CONSDEV;
2835 has_preferred_console = true;
2841 * Some consoles, such as pstore and netconsole, can be enabled even
2842 * without matching. Accept the pre-enabled consoles only when match()
2843 * and setup() had a chance to be called.
2845 if (newcon->flags & CON_ENABLED && c->user_specified == user_specified)
2852 * The console driver calls this routine during kernel initialization
2853 * to register the console printing procedure with printk() and to
2854 * print any messages that were printed by the kernel before the
2855 * console driver was initialized.
2857 * This can happen pretty early during the boot process (because of
2858 * early_printk) - sometimes before setup_arch() completes - be careful
2859 * of what kernel features are used - they may not be initialised yet.
2861 * There are two types of consoles - bootconsoles (early_printk) and
2862 * "real" consoles (everything which is not a bootconsole) which are
2863 * handled differently.
2864 * - Any number of bootconsoles can be registered at any time.
2865 * - As soon as a "real" console is registered, all bootconsoles
2866 * will be unregistered automatically.
2867 * - Once a "real" console is registered, any attempt to register a
2868 * bootconsoles will be rejected
2870 void register_console(struct console *newcon)
2872 unsigned long flags;
2873 struct console *bcon = NULL;
2876 for_each_console(bcon) {
2877 if (WARN(bcon == newcon, "console '%s%d' already registered\n",
2878 bcon->name, bcon->index))
2883 * before we register a new CON_BOOT console, make sure we don't
2884 * already have a valid console
2886 if (newcon->flags & CON_BOOT) {
2887 for_each_console(bcon) {
2888 if (!(bcon->flags & CON_BOOT)) {
2889 pr_info("Too late to register bootconsole %s%d\n",
2890 newcon->name, newcon->index);
2896 if (console_drivers && console_drivers->flags & CON_BOOT)
2897 bcon = console_drivers;
2899 if (!has_preferred_console || bcon || !console_drivers)
2900 has_preferred_console = preferred_console >= 0;
2903 * See if we want to use this console driver. If we
2904 * didn't select a console we take the first one
2905 * that registers here.
2907 if (!has_preferred_console) {
2908 if (newcon->index < 0)
2910 if (newcon->setup == NULL ||
2911 newcon->setup(newcon, NULL) == 0) {
2912 newcon->flags |= CON_ENABLED;
2913 if (newcon->device) {
2914 newcon->flags |= CON_CONSDEV;
2915 has_preferred_console = true;
2920 /* See if this console matches one we selected on the command line */
2921 err = try_enable_new_console(newcon, true);
2923 /* If not, try to match against the platform default(s) */
2925 err = try_enable_new_console(newcon, false);
2927 /* printk() messages are not printed to the Braille console. */
2928 if (err || newcon->flags & CON_BRL)
2932 * If we have a bootconsole, and are switching to a real console,
2933 * don't print everything out again, since when the boot console, and
2934 * the real console are the same physical device, it's annoying to
2935 * see the beginning boot messages twice
2937 if (bcon && ((newcon->flags & (CON_CONSDEV | CON_BOOT)) == CON_CONSDEV))
2938 newcon->flags &= ~CON_PRINTBUFFER;
2941 * Put this console in the list - keep the
2942 * preferred driver at the head of the list.
2945 if ((newcon->flags & CON_CONSDEV) || console_drivers == NULL) {
2946 newcon->next = console_drivers;
2947 console_drivers = newcon;
2949 newcon->next->flags &= ~CON_CONSDEV;
2950 /* Ensure this flag is always set for the head of the list */
2951 newcon->flags |= CON_CONSDEV;
2953 newcon->next = console_drivers->next;
2954 console_drivers->next = newcon;
2957 if (newcon->flags & CON_EXTENDED)
2958 nr_ext_console_drivers++;
2960 if (newcon->flags & CON_PRINTBUFFER) {
2962 * console_unlock(); will print out the buffered messages
2965 * We're about to replay the log buffer. Only do this to the
2966 * just-registered console to avoid excessive message spam to
2967 * the already-registered consoles.
2969 * Set exclusive_console with disabled interrupts to reduce
2970 * race window with eventual console_flush_on_panic() that
2971 * ignores console_lock.
2973 exclusive_console = newcon;
2974 exclusive_console_stop_seq = console_seq;
2976 /* Get a consistent copy of @syslog_seq. */
2977 raw_spin_lock_irqsave(&syslog_lock, flags);
2978 console_seq = syslog_seq;
2979 raw_spin_unlock_irqrestore(&syslog_lock, flags);
2982 console_sysfs_notify();
2985 * By unregistering the bootconsoles after we enable the real console
2986 * we get the "console xxx enabled" message on all the consoles -
2987 * boot consoles, real consoles, etc - this is to ensure that end
2988 * users know there might be something in the kernel's log buffer that
2989 * went to the bootconsole (that they do not see on the real console)
2991 pr_info("%sconsole [%s%d] enabled\n",
2992 (newcon->flags & CON_BOOT) ? "boot" : "" ,
2993 newcon->name, newcon->index);
2995 ((newcon->flags & (CON_CONSDEV | CON_BOOT)) == CON_CONSDEV) &&
2997 /* We need to iterate through all boot consoles, to make
2998 * sure we print everything out, before we unregister them.
3000 for_each_console(bcon)
3001 if (bcon->flags & CON_BOOT)
3002 unregister_console(bcon);
3005 EXPORT_SYMBOL(register_console);
3007 int unregister_console(struct console *console)
3009 struct console *con;
3012 pr_info("%sconsole [%s%d] disabled\n",
3013 (console->flags & CON_BOOT) ? "boot" : "" ,
3014 console->name, console->index);
3016 res = _braille_unregister_console(console);
3024 if (console_drivers == console) {
3025 console_drivers=console->next;
3028 for_each_console(con) {
3029 if (con->next == console) {
3030 con->next = console->next;
3038 goto out_disable_unlock;
3040 if (console->flags & CON_EXTENDED)
3041 nr_ext_console_drivers--;
3044 * If this isn't the last console and it has CON_CONSDEV set, we
3045 * need to set it on the next preferred console.
3047 if (console_drivers != NULL && console->flags & CON_CONSDEV)
3048 console_drivers->flags |= CON_CONSDEV;
3050 console->flags &= ~CON_ENABLED;
3052 console_sysfs_notify();
3055 res = console->exit(console);
3060 console->flags &= ~CON_ENABLED;
3065 EXPORT_SYMBOL(unregister_console);
3068 * Initialize the console device. This is called *early*, so
3069 * we can't necessarily depend on lots of kernel help here.
3070 * Just do some early initializations, and do the complex setup
3073 void __init console_init(void)
3077 initcall_entry_t *ce;
3079 /* Setup the default TTY line discipline. */
3083 * set up the console device so that later boot sequences can
3084 * inform about problems etc..
3086 ce = __con_initcall_start;
3087 trace_initcall_level("console");
3088 while (ce < __con_initcall_end) {
3089 call = initcall_from_entry(ce);
3090 trace_initcall_start(call);
3092 trace_initcall_finish(call, ret);
3098 * Some boot consoles access data that is in the init section and which will
3099 * be discarded after the initcalls have been run. To make sure that no code
3100 * will access this data, unregister the boot consoles in a late initcall.
3102 * If for some reason, such as deferred probe or the driver being a loadable
3103 * module, the real console hasn't registered yet at this point, there will
3104 * be a brief interval in which no messages are logged to the console, which
3105 * makes it difficult to diagnose problems that occur during this time.
3107 * To mitigate this problem somewhat, only unregister consoles whose memory
3108 * intersects with the init section. Note that all other boot consoles will
3109 * get unregistered when the real preferred console is registered.
3111 static int __init printk_late_init(void)
3113 struct console *con;
3116 for_each_console(con) {
3117 if (!(con->flags & CON_BOOT))
3120 /* Check addresses that might be used for enabled consoles. */
3121 if (init_section_intersects(con, sizeof(*con)) ||
3122 init_section_contains(con->write, 0) ||
3123 init_section_contains(con->read, 0) ||
3124 init_section_contains(con->device, 0) ||
3125 init_section_contains(con->unblank, 0) ||
3126 init_section_contains(con->data, 0)) {
3128 * Please, consider moving the reported consoles out
3129 * of the init section.
3131 pr_warn("bootconsole [%s%d] uses init memory and must be disabled even before the real one is ready\n",
3132 con->name, con->index);
3133 unregister_console(con);
3136 ret = cpuhp_setup_state_nocalls(CPUHP_PRINTK_DEAD, "printk:dead", NULL,
3137 console_cpu_notify);
3139 ret = cpuhp_setup_state_nocalls(CPUHP_AP_ONLINE_DYN, "printk:online",
3140 console_cpu_notify, NULL);
3144 late_initcall(printk_late_init);
3146 #if defined CONFIG_PRINTK
3148 * Delayed printk version, for scheduler-internal messages:
3150 #define PRINTK_PENDING_WAKEUP 0x01
3151 #define PRINTK_PENDING_OUTPUT 0x02
3153 static DEFINE_PER_CPU(int, printk_pending);
3155 static void wake_up_klogd_work_func(struct irq_work *irq_work)
3157 int pending = __this_cpu_xchg(printk_pending, 0);
3159 if (pending & PRINTK_PENDING_OUTPUT) {
3160 /* If trylock fails, someone else is doing the printing */
3161 if (console_trylock())
3165 if (pending & PRINTK_PENDING_WAKEUP)
3166 wake_up_interruptible(&log_wait);
3169 static DEFINE_PER_CPU(struct irq_work, wake_up_klogd_work) =
3170 IRQ_WORK_INIT_LAZY(wake_up_klogd_work_func);
3172 void wake_up_klogd(void)
3174 if (!printk_percpu_data_ready())
3178 if (waitqueue_active(&log_wait)) {
3179 this_cpu_or(printk_pending, PRINTK_PENDING_WAKEUP);
3180 irq_work_queue(this_cpu_ptr(&wake_up_klogd_work));
3185 void defer_console_output(void)
3187 if (!printk_percpu_data_ready())
3191 __this_cpu_or(printk_pending, PRINTK_PENDING_OUTPUT);
3192 irq_work_queue(this_cpu_ptr(&wake_up_klogd_work));
3196 int vprintk_deferred(const char *fmt, va_list args)
3200 r = vprintk_emit(0, LOGLEVEL_SCHED, NULL, fmt, args);
3201 defer_console_output();
3206 int printk_deferred(const char *fmt, ...)
3211 va_start(args, fmt);
3212 r = vprintk_deferred(fmt, args);
3219 * printk rate limiting, lifted from the networking subsystem.
3221 * This enforces a rate limit: not more than 10 kernel messages
3222 * every 5s to make a denial-of-service attack impossible.
3224 DEFINE_RATELIMIT_STATE(printk_ratelimit_state, 5 * HZ, 10);
3226 int __printk_ratelimit(const char *func)
3228 return ___ratelimit(&printk_ratelimit_state, func);
3230 EXPORT_SYMBOL(__printk_ratelimit);
3233 * printk_timed_ratelimit - caller-controlled printk ratelimiting
3234 * @caller_jiffies: pointer to caller's state
3235 * @interval_msecs: minimum interval between prints
3237 * printk_timed_ratelimit() returns true if more than @interval_msecs
3238 * milliseconds have elapsed since the last time printk_timed_ratelimit()
3241 bool printk_timed_ratelimit(unsigned long *caller_jiffies,
3242 unsigned int interval_msecs)
3244 unsigned long elapsed = jiffies - *caller_jiffies;
3246 if (*caller_jiffies && elapsed <= msecs_to_jiffies(interval_msecs))
3249 *caller_jiffies = jiffies;
3252 EXPORT_SYMBOL(printk_timed_ratelimit);
3254 static DEFINE_SPINLOCK(dump_list_lock);
3255 static LIST_HEAD(dump_list);
3258 * kmsg_dump_register - register a kernel log dumper.
3259 * @dumper: pointer to the kmsg_dumper structure
3261 * Adds a kernel log dumper to the system. The dump callback in the
3262 * structure will be called when the kernel oopses or panics and must be
3263 * set. Returns zero on success and %-EINVAL or %-EBUSY otherwise.
3265 int kmsg_dump_register(struct kmsg_dumper *dumper)
3267 unsigned long flags;
3270 /* The dump callback needs to be set */
3274 spin_lock_irqsave(&dump_list_lock, flags);
3275 /* Don't allow registering multiple times */
3276 if (!dumper->registered) {
3277 dumper->registered = 1;
3278 list_add_tail_rcu(&dumper->list, &dump_list);
3281 spin_unlock_irqrestore(&dump_list_lock, flags);
3285 EXPORT_SYMBOL_GPL(kmsg_dump_register);
3288 * kmsg_dump_unregister - unregister a kmsg dumper.
3289 * @dumper: pointer to the kmsg_dumper structure
3291 * Removes a dump device from the system. Returns zero on success and
3292 * %-EINVAL otherwise.
3294 int kmsg_dump_unregister(struct kmsg_dumper *dumper)
3296 unsigned long flags;
3299 spin_lock_irqsave(&dump_list_lock, flags);
3300 if (dumper->registered) {
3301 dumper->registered = 0;
3302 list_del_rcu(&dumper->list);
3305 spin_unlock_irqrestore(&dump_list_lock, flags);
3310 EXPORT_SYMBOL_GPL(kmsg_dump_unregister);
3312 static bool always_kmsg_dump;
3313 module_param_named(always_kmsg_dump, always_kmsg_dump, bool, S_IRUGO | S_IWUSR);
3315 const char *kmsg_dump_reason_str(enum kmsg_dump_reason reason)
3318 case KMSG_DUMP_PANIC:
3320 case KMSG_DUMP_OOPS:
3322 case KMSG_DUMP_EMERG:
3324 case KMSG_DUMP_SHUTDOWN:
3330 EXPORT_SYMBOL_GPL(kmsg_dump_reason_str);
3333 * kmsg_dump - dump kernel log to kernel message dumpers.
3334 * @reason: the reason (oops, panic etc) for dumping
3336 * Call each of the registered dumper's dump() callback, which can
3337 * retrieve the kmsg records with kmsg_dump_get_line() or
3338 * kmsg_dump_get_buffer().
3340 void kmsg_dump(enum kmsg_dump_reason reason)
3342 struct kmsg_dumper *dumper;
3345 list_for_each_entry_rcu(dumper, &dump_list, list) {
3346 enum kmsg_dump_reason max_reason = dumper->max_reason;
3349 * If client has not provided a specific max_reason, default
3350 * to KMSG_DUMP_OOPS, unless always_kmsg_dump was set.
3352 if (max_reason == KMSG_DUMP_UNDEF) {
3353 max_reason = always_kmsg_dump ? KMSG_DUMP_MAX :
3356 if (reason > max_reason)
3359 /* invoke dumper which will iterate over records */
3360 dumper->dump(dumper, reason);
3366 * kmsg_dump_get_line - retrieve one kmsg log line
3367 * @iter: kmsg dump iterator
3368 * @syslog: include the "<4>" prefixes
3369 * @line: buffer to copy the line to
3370 * @size: maximum size of the buffer
3371 * @len: length of line placed into buffer
3373 * Start at the beginning of the kmsg buffer, with the oldest kmsg
3374 * record, and copy one record into the provided buffer.
3376 * Consecutive calls will return the next available record moving
3377 * towards the end of the buffer with the youngest messages.
3379 * A return value of FALSE indicates that there are no more records to
3382 bool kmsg_dump_get_line(struct kmsg_dump_iter *iter, bool syslog,
3383 char *line, size_t size, size_t *len)
3385 u64 min_seq = latched_seq_read_nolock(&clear_seq);
3386 struct printk_info info;
3387 unsigned int line_count;
3388 struct printk_record r;
3389 unsigned long flags;
3393 if (iter->cur_seq < min_seq)
3394 iter->cur_seq = min_seq;
3396 printk_safe_enter_irqsave(flags);
3397 prb_rec_init_rd(&r, &info, line, size);
3399 /* Read text or count text lines? */
3401 if (!prb_read_valid(prb, iter->cur_seq, &r))
3403 l = record_print_text(&r, syslog, printk_time);
3405 if (!prb_read_valid_info(prb, iter->cur_seq,
3406 &info, &line_count)) {
3409 l = get_record_print_text_size(&info, line_count, syslog,
3414 iter->cur_seq = r.info->seq + 1;
3417 printk_safe_exit_irqrestore(flags);
3422 EXPORT_SYMBOL_GPL(kmsg_dump_get_line);
3425 * kmsg_dump_get_buffer - copy kmsg log lines
3426 * @iter: kmsg dump iterator
3427 * @syslog: include the "<4>" prefixes
3428 * @buf: buffer to copy the line to
3429 * @size: maximum size of the buffer
3430 * @len_out: length of line placed into buffer
3432 * Start at the end of the kmsg buffer and fill the provided buffer
3433 * with as many of the *youngest* kmsg records that fit into it.
3434 * If the buffer is large enough, all available kmsg records will be
3435 * copied with a single call.
3437 * Consecutive calls will fill the buffer with the next block of
3438 * available older records, not including the earlier retrieved ones.
3440 * A return value of FALSE indicates that there are no more records to
3443 bool kmsg_dump_get_buffer(struct kmsg_dump_iter *iter, bool syslog,
3444 char *buf, size_t size, size_t *len_out)
3446 u64 min_seq = latched_seq_read_nolock(&clear_seq);
3447 struct printk_info info;
3448 struct printk_record r;
3449 unsigned long flags;
3454 bool time = printk_time;
3459 if (iter->cur_seq < min_seq)
3460 iter->cur_seq = min_seq;
3462 printk_safe_enter_irqsave(flags);
3463 if (prb_read_valid_info(prb, iter->cur_seq, &info, NULL)) {
3464 if (info.seq != iter->cur_seq) {
3465 /* messages are gone, move to first available one */
3466 iter->cur_seq = info.seq;
3471 if (iter->cur_seq >= iter->next_seq) {
3472 printk_safe_exit_irqrestore(flags);
3477 * Find first record that fits, including all following records,
3478 * into the user-provided buffer for this dump. Pass in size-1
3479 * because this function (by way of record_print_text()) will
3480 * not write more than size-1 bytes of text into @buf.
3482 seq = find_first_fitting_seq(iter->cur_seq, iter->next_seq,
3483 size - 1, syslog, time);
3486 * Next kmsg_dump_get_buffer() invocation will dump block of
3487 * older records stored right before this one.
3491 prb_rec_init_rd(&r, &info, buf, size);
3494 prb_for_each_record(seq, prb, seq, &r) {
3495 if (r.info->seq >= iter->next_seq)
3498 len += record_print_text(&r, syslog, time);
3500 /* Adjust record to store to remaining buffer space. */
3501 prb_rec_init_rd(&r, &info, buf + len, size - len);
3504 iter->next_seq = next_seq;
3506 printk_safe_exit_irqrestore(flags);
3512 EXPORT_SYMBOL_GPL(kmsg_dump_get_buffer);
3515 * kmsg_dump_rewind - reset the iterator
3516 * @iter: kmsg dump iterator
3518 * Reset the dumper's iterator so that kmsg_dump_get_line() and
3519 * kmsg_dump_get_buffer() can be called again and used multiple
3520 * times within the same dumper.dump() callback.
3522 void kmsg_dump_rewind(struct kmsg_dump_iter *iter)
3524 unsigned long flags;
3526 printk_safe_enter_irqsave(flags);
3527 iter->cur_seq = latched_seq_read_nolock(&clear_seq);
3528 iter->next_seq = prb_next_seq(prb);
3529 printk_safe_exit_irqrestore(flags);
3531 EXPORT_SYMBOL_GPL(kmsg_dump_rewind);