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 DECLARE_WAIT_QUEUE_HEAD(log_wait);
366 /* All 3 protected by @syslog_lock. */
367 /* the next printk record to read by syslog(READ) or /proc/kmsg */
368 static u64 syslog_seq;
369 static size_t syslog_partial;
370 static bool syslog_time;
373 seqcount_latch_t latch;
378 * The next printk record to read after the last 'clear' command. There are
379 * two copies (updated with seqcount_latch) so that reads can locklessly
380 * access a valid value. Writers are synchronized by @syslog_lock.
382 static struct latched_seq clear_seq = {
383 .latch = SEQCNT_LATCH_ZERO(clear_seq.latch),
388 #ifdef CONFIG_PRINTK_CALLER
389 #define PREFIX_MAX 48
391 #define PREFIX_MAX 32
394 /* the maximum size of a formatted record (i.e. with prefix added per line) */
395 #define CONSOLE_LOG_MAX 1024
397 /* the maximum size for a dropped text message */
398 #define DROPPED_TEXT_MAX 64
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 if (!prb_read_valid(prb, atomic64_read(&user->seq), r)) {
736 if (file->f_flags & O_NONBLOCK) {
742 * Guarantee this task is visible on the waitqueue before
743 * checking the wake condition.
745 * The full memory barrier within set_current_state() of
746 * prepare_to_wait_event() pairs with the full memory barrier
747 * within wq_has_sleeper().
749 * This pairs with __wake_up_klogd:A.
751 ret = wait_event_interruptible(log_wait,
753 atomic64_read(&user->seq), r)); /* LMM(devkmsg_read:A) */
758 if (r->info->seq != atomic64_read(&user->seq)) {
759 /* our last seen message is gone, return error and reset */
760 atomic64_set(&user->seq, r->info->seq);
765 len = info_print_ext_header(user->buf, sizeof(user->buf), r->info);
766 len += msg_print_ext_body(user->buf + len, sizeof(user->buf) - len,
767 &r->text_buf[0], r->info->text_len,
770 atomic64_set(&user->seq, r->info->seq + 1);
777 if (copy_to_user(buf, user->buf, len)) {
783 mutex_unlock(&user->lock);
788 * Be careful when modifying this function!!!
790 * Only few operations are supported because the device works only with the
791 * entire variable length messages (records). Non-standard values are
792 * returned in the other cases and has been this way for quite some time.
793 * User space applications might depend on this behavior.
795 static loff_t devkmsg_llseek(struct file *file, loff_t offset, int whence)
797 struct devkmsg_user *user = file->private_data;
807 /* the first record */
808 atomic64_set(&user->seq, prb_first_valid_seq(prb));
812 * The first record after the last SYSLOG_ACTION_CLEAR,
813 * like issued by 'dmesg -c'. Reading /dev/kmsg itself
814 * changes no global state, and does not clear anything.
816 atomic64_set(&user->seq, latched_seq_read_nolock(&clear_seq));
819 /* after the last record */
820 atomic64_set(&user->seq, prb_next_seq(prb));
828 static __poll_t devkmsg_poll(struct file *file, poll_table *wait)
830 struct devkmsg_user *user = file->private_data;
831 struct printk_info info;
835 return EPOLLERR|EPOLLNVAL;
837 poll_wait(file, &log_wait, wait);
839 if (prb_read_valid_info(prb, atomic64_read(&user->seq), &info, NULL)) {
840 /* return error when data has vanished underneath us */
841 if (info.seq != atomic64_read(&user->seq))
842 ret = EPOLLIN|EPOLLRDNORM|EPOLLERR|EPOLLPRI;
844 ret = EPOLLIN|EPOLLRDNORM;
850 static int devkmsg_open(struct inode *inode, struct file *file)
852 struct devkmsg_user *user;
855 if (devkmsg_log & DEVKMSG_LOG_MASK_OFF)
858 /* write-only does not need any file context */
859 if ((file->f_flags & O_ACCMODE) != O_WRONLY) {
860 err = check_syslog_permissions(SYSLOG_ACTION_READ_ALL,
866 user = kvmalloc(sizeof(struct devkmsg_user), GFP_KERNEL);
870 ratelimit_default_init(&user->rs);
871 ratelimit_set_flags(&user->rs, RATELIMIT_MSG_ON_RELEASE);
873 mutex_init(&user->lock);
875 prb_rec_init_rd(&user->record, &user->info,
876 &user->text_buf[0], sizeof(user->text_buf));
878 atomic64_set(&user->seq, prb_first_valid_seq(prb));
880 file->private_data = user;
884 static int devkmsg_release(struct inode *inode, struct file *file)
886 struct devkmsg_user *user = file->private_data;
891 ratelimit_state_exit(&user->rs);
893 mutex_destroy(&user->lock);
898 const struct file_operations kmsg_fops = {
899 .open = devkmsg_open,
900 .read = devkmsg_read,
901 .write_iter = devkmsg_write,
902 .llseek = devkmsg_llseek,
903 .poll = devkmsg_poll,
904 .release = devkmsg_release,
907 #ifdef CONFIG_CRASH_CORE
909 * This appends the listed symbols to /proc/vmcore
911 * /proc/vmcore is used by various utilities, like crash and makedumpfile to
912 * obtain access to symbols that are otherwise very difficult to locate. These
913 * symbols are specifically used so that utilities can access and extract the
914 * dmesg log from a vmcore file after a crash.
916 void log_buf_vmcoreinfo_setup(void)
918 struct dev_printk_info *dev_info = NULL;
920 VMCOREINFO_SYMBOL(prb);
921 VMCOREINFO_SYMBOL(printk_rb_static);
922 VMCOREINFO_SYMBOL(clear_seq);
925 * Export struct size and field offsets. User space tools can
926 * parse it and detect any changes to structure down the line.
929 VMCOREINFO_STRUCT_SIZE(printk_ringbuffer);
930 VMCOREINFO_OFFSET(printk_ringbuffer, desc_ring);
931 VMCOREINFO_OFFSET(printk_ringbuffer, text_data_ring);
932 VMCOREINFO_OFFSET(printk_ringbuffer, fail);
934 VMCOREINFO_STRUCT_SIZE(prb_desc_ring);
935 VMCOREINFO_OFFSET(prb_desc_ring, count_bits);
936 VMCOREINFO_OFFSET(prb_desc_ring, descs);
937 VMCOREINFO_OFFSET(prb_desc_ring, infos);
938 VMCOREINFO_OFFSET(prb_desc_ring, head_id);
939 VMCOREINFO_OFFSET(prb_desc_ring, tail_id);
941 VMCOREINFO_STRUCT_SIZE(prb_desc);
942 VMCOREINFO_OFFSET(prb_desc, state_var);
943 VMCOREINFO_OFFSET(prb_desc, text_blk_lpos);
945 VMCOREINFO_STRUCT_SIZE(prb_data_blk_lpos);
946 VMCOREINFO_OFFSET(prb_data_blk_lpos, begin);
947 VMCOREINFO_OFFSET(prb_data_blk_lpos, next);
949 VMCOREINFO_STRUCT_SIZE(printk_info);
950 VMCOREINFO_OFFSET(printk_info, seq);
951 VMCOREINFO_OFFSET(printk_info, ts_nsec);
952 VMCOREINFO_OFFSET(printk_info, text_len);
953 VMCOREINFO_OFFSET(printk_info, caller_id);
954 VMCOREINFO_OFFSET(printk_info, dev_info);
956 VMCOREINFO_STRUCT_SIZE(dev_printk_info);
957 VMCOREINFO_OFFSET(dev_printk_info, subsystem);
958 VMCOREINFO_LENGTH(printk_info_subsystem, sizeof(dev_info->subsystem));
959 VMCOREINFO_OFFSET(dev_printk_info, device);
960 VMCOREINFO_LENGTH(printk_info_device, sizeof(dev_info->device));
962 VMCOREINFO_STRUCT_SIZE(prb_data_ring);
963 VMCOREINFO_OFFSET(prb_data_ring, size_bits);
964 VMCOREINFO_OFFSET(prb_data_ring, data);
965 VMCOREINFO_OFFSET(prb_data_ring, head_lpos);
966 VMCOREINFO_OFFSET(prb_data_ring, tail_lpos);
968 VMCOREINFO_SIZE(atomic_long_t);
969 VMCOREINFO_TYPE_OFFSET(atomic_long_t, counter);
971 VMCOREINFO_STRUCT_SIZE(latched_seq);
972 VMCOREINFO_OFFSET(latched_seq, val);
976 /* requested log_buf_len from kernel cmdline */
977 static unsigned long __initdata new_log_buf_len;
979 /* we practice scaling the ring buffer by powers of 2 */
980 static void __init log_buf_len_update(u64 size)
982 if (size > (u64)LOG_BUF_LEN_MAX) {
983 size = (u64)LOG_BUF_LEN_MAX;
984 pr_err("log_buf over 2G is not supported.\n");
988 size = roundup_pow_of_two(size);
989 if (size > log_buf_len)
990 new_log_buf_len = (unsigned long)size;
993 /* save requested log_buf_len since it's too early to process it */
994 static int __init log_buf_len_setup(char *str)
1001 size = memparse(str, &str);
1003 log_buf_len_update(size);
1007 early_param("log_buf_len", log_buf_len_setup);
1010 #define __LOG_CPU_MAX_BUF_LEN (1 << CONFIG_LOG_CPU_MAX_BUF_SHIFT)
1012 static void __init log_buf_add_cpu(void)
1014 unsigned int cpu_extra;
1017 * archs should set up cpu_possible_bits properly with
1018 * set_cpu_possible() after setup_arch() but just in
1019 * case lets ensure this is valid.
1021 if (num_possible_cpus() == 1)
1024 cpu_extra = (num_possible_cpus() - 1) * __LOG_CPU_MAX_BUF_LEN;
1026 /* by default this will only continue through for large > 64 CPUs */
1027 if (cpu_extra <= __LOG_BUF_LEN / 2)
1030 pr_info("log_buf_len individual max cpu contribution: %d bytes\n",
1031 __LOG_CPU_MAX_BUF_LEN);
1032 pr_info("log_buf_len total cpu_extra contributions: %d bytes\n",
1034 pr_info("log_buf_len min size: %d bytes\n", __LOG_BUF_LEN);
1036 log_buf_len_update(cpu_extra + __LOG_BUF_LEN);
1038 #else /* !CONFIG_SMP */
1039 static inline void log_buf_add_cpu(void) {}
1040 #endif /* CONFIG_SMP */
1042 static void __init set_percpu_data_ready(void)
1044 __printk_percpu_data_ready = true;
1047 static unsigned int __init add_to_rb(struct printk_ringbuffer *rb,
1048 struct printk_record *r)
1050 struct prb_reserved_entry e;
1051 struct printk_record dest_r;
1053 prb_rec_init_wr(&dest_r, r->info->text_len);
1055 if (!prb_reserve(&e, rb, &dest_r))
1058 memcpy(&dest_r.text_buf[0], &r->text_buf[0], r->info->text_len);
1059 dest_r.info->text_len = r->info->text_len;
1060 dest_r.info->facility = r->info->facility;
1061 dest_r.info->level = r->info->level;
1062 dest_r.info->flags = r->info->flags;
1063 dest_r.info->ts_nsec = r->info->ts_nsec;
1064 dest_r.info->caller_id = r->info->caller_id;
1065 memcpy(&dest_r.info->dev_info, &r->info->dev_info, sizeof(dest_r.info->dev_info));
1067 prb_final_commit(&e);
1069 return prb_record_text_space(&e);
1072 static char setup_text_buf[LOG_LINE_MAX] __initdata;
1074 void __init setup_log_buf(int early)
1076 struct printk_info *new_infos;
1077 unsigned int new_descs_count;
1078 struct prb_desc *new_descs;
1079 struct printk_info info;
1080 struct printk_record r;
1081 unsigned int text_size;
1082 size_t new_descs_size;
1083 size_t new_infos_size;
1084 unsigned long flags;
1090 * Some archs call setup_log_buf() multiple times - first is very
1091 * early, e.g. from setup_arch(), and second - when percpu_areas
1095 set_percpu_data_ready();
1097 if (log_buf != __log_buf)
1100 if (!early && !new_log_buf_len)
1103 if (!new_log_buf_len)
1106 new_descs_count = new_log_buf_len >> PRB_AVGBITS;
1107 if (new_descs_count == 0) {
1108 pr_err("new_log_buf_len: %lu too small\n", new_log_buf_len);
1112 new_log_buf = memblock_alloc(new_log_buf_len, LOG_ALIGN);
1113 if (unlikely(!new_log_buf)) {
1114 pr_err("log_buf_len: %lu text bytes not available\n",
1119 new_descs_size = new_descs_count * sizeof(struct prb_desc);
1120 new_descs = memblock_alloc(new_descs_size, LOG_ALIGN);
1121 if (unlikely(!new_descs)) {
1122 pr_err("log_buf_len: %zu desc bytes not available\n",
1124 goto err_free_log_buf;
1127 new_infos_size = new_descs_count * sizeof(struct printk_info);
1128 new_infos = memblock_alloc(new_infos_size, LOG_ALIGN);
1129 if (unlikely(!new_infos)) {
1130 pr_err("log_buf_len: %zu info bytes not available\n",
1132 goto err_free_descs;
1135 prb_rec_init_rd(&r, &info, &setup_text_buf[0], sizeof(setup_text_buf));
1137 prb_init(&printk_rb_dynamic,
1138 new_log_buf, ilog2(new_log_buf_len),
1139 new_descs, ilog2(new_descs_count),
1142 local_irq_save(flags);
1144 log_buf_len = new_log_buf_len;
1145 log_buf = new_log_buf;
1146 new_log_buf_len = 0;
1148 free = __LOG_BUF_LEN;
1149 prb_for_each_record(0, &printk_rb_static, seq, &r) {
1150 text_size = add_to_rb(&printk_rb_dynamic, &r);
1151 if (text_size > free)
1157 prb = &printk_rb_dynamic;
1159 local_irq_restore(flags);
1162 * Copy any remaining messages that might have appeared from
1163 * NMI context after copying but before switching to the
1166 prb_for_each_record(seq, &printk_rb_static, seq, &r) {
1167 text_size = add_to_rb(&printk_rb_dynamic, &r);
1168 if (text_size > free)
1174 if (seq != prb_next_seq(&printk_rb_static)) {
1175 pr_err("dropped %llu messages\n",
1176 prb_next_seq(&printk_rb_static) - seq);
1179 pr_info("log_buf_len: %u bytes\n", log_buf_len);
1180 pr_info("early log buf free: %u(%u%%)\n",
1181 free, (free * 100) / __LOG_BUF_LEN);
1185 memblock_free(new_descs, new_descs_size);
1187 memblock_free(new_log_buf, new_log_buf_len);
1190 static bool __read_mostly ignore_loglevel;
1192 static int __init ignore_loglevel_setup(char *str)
1194 ignore_loglevel = true;
1195 pr_info("debug: ignoring loglevel setting.\n");
1200 early_param("ignore_loglevel", ignore_loglevel_setup);
1201 module_param(ignore_loglevel, bool, S_IRUGO | S_IWUSR);
1202 MODULE_PARM_DESC(ignore_loglevel,
1203 "ignore loglevel setting (prints all kernel messages to the console)");
1205 static bool suppress_message_printing(int level)
1207 return (level >= console_loglevel && !ignore_loglevel);
1210 #ifdef CONFIG_BOOT_PRINTK_DELAY
1212 static int boot_delay; /* msecs delay after each printk during bootup */
1213 static unsigned long long loops_per_msec; /* based on boot_delay */
1215 static int __init boot_delay_setup(char *str)
1219 lpj = preset_lpj ? preset_lpj : 1000000; /* some guess */
1220 loops_per_msec = (unsigned long long)lpj / 1000 * HZ;
1222 get_option(&str, &boot_delay);
1223 if (boot_delay > 10 * 1000)
1226 pr_debug("boot_delay: %u, preset_lpj: %ld, lpj: %lu, "
1227 "HZ: %d, loops_per_msec: %llu\n",
1228 boot_delay, preset_lpj, lpj, HZ, loops_per_msec);
1231 early_param("boot_delay", boot_delay_setup);
1233 static void boot_delay_msec(int level)
1235 unsigned long long k;
1236 unsigned long timeout;
1238 if ((boot_delay == 0 || system_state >= SYSTEM_RUNNING)
1239 || suppress_message_printing(level)) {
1243 k = (unsigned long long)loops_per_msec * boot_delay;
1245 timeout = jiffies + msecs_to_jiffies(boot_delay);
1250 * use (volatile) jiffies to prevent
1251 * compiler reduction; loop termination via jiffies
1252 * is secondary and may or may not happen.
1254 if (time_after(jiffies, timeout))
1256 touch_nmi_watchdog();
1260 static inline void boot_delay_msec(int level)
1265 static bool printk_time = IS_ENABLED(CONFIG_PRINTK_TIME);
1266 module_param_named(time, printk_time, bool, S_IRUGO | S_IWUSR);
1268 static size_t print_syslog(unsigned int level, char *buf)
1270 return sprintf(buf, "<%u>", level);
1273 static size_t print_time(u64 ts, char *buf)
1275 unsigned long rem_nsec = do_div(ts, 1000000000);
1277 return sprintf(buf, "[%5lu.%06lu]",
1278 (unsigned long)ts, rem_nsec / 1000);
1281 #ifdef CONFIG_PRINTK_CALLER
1282 static size_t print_caller(u32 id, char *buf)
1286 snprintf(caller, sizeof(caller), "%c%u",
1287 id & 0x80000000 ? 'C' : 'T', id & ~0x80000000);
1288 return sprintf(buf, "[%6s]", caller);
1291 #define print_caller(id, buf) 0
1294 static size_t info_print_prefix(const struct printk_info *info, bool syslog,
1295 bool time, char *buf)
1300 len = print_syslog((info->facility << 3) | info->level, buf);
1303 len += print_time(info->ts_nsec, buf + len);
1305 len += print_caller(info->caller_id, buf + len);
1307 if (IS_ENABLED(CONFIG_PRINTK_CALLER) || time) {
1316 * Prepare the record for printing. The text is shifted within the given
1317 * buffer to avoid a need for another one. The following operations are
1320 * - Add prefix for each line.
1321 * - Drop truncated lines that no longer fit into the buffer.
1322 * - Add the trailing newline that has been removed in vprintk_store().
1323 * - Add a string terminator.
1325 * Since the produced string is always terminated, the maximum possible
1326 * return value is @r->text_buf_size - 1;
1328 * Return: The length of the updated/prepared text, including the added
1329 * prefixes and the newline. The terminator is not counted. The dropped
1330 * line(s) are not counted.
1332 static size_t record_print_text(struct printk_record *r, bool syslog,
1335 size_t text_len = r->info->text_len;
1336 size_t buf_size = r->text_buf_size;
1337 char *text = r->text_buf;
1338 char prefix[PREFIX_MAX];
1339 bool truncated = false;
1346 * If the message was truncated because the buffer was not large
1347 * enough, treat the available text as if it were the full text.
1349 if (text_len > buf_size)
1350 text_len = buf_size;
1352 prefix_len = info_print_prefix(r->info, syslog, time, prefix);
1355 * @text_len: bytes of unprocessed text
1356 * @line_len: bytes of current line _without_ newline
1357 * @text: pointer to beginning of current line
1358 * @len: number of bytes prepared in r->text_buf
1361 next = memchr(text, '\n', text_len);
1363 line_len = next - text;
1365 /* Drop truncated line(s). */
1368 line_len = text_len;
1372 * Truncate the text if there is not enough space to add the
1373 * prefix and a trailing newline and a terminator.
1375 if (len + prefix_len + text_len + 1 + 1 > buf_size) {
1376 /* Drop even the current line if no space. */
1377 if (len + prefix_len + line_len + 1 + 1 > buf_size)
1380 text_len = buf_size - len - prefix_len - 1 - 1;
1384 memmove(text + prefix_len, text, text_len);
1385 memcpy(text, prefix, prefix_len);
1388 * Increment the prepared length to include the text and
1389 * prefix that were just moved+copied. Also increment for the
1390 * newline at the end of this line. If this is the last line,
1391 * there is no newline, but it will be added immediately below.
1393 len += prefix_len + line_len + 1;
1394 if (text_len == line_len) {
1396 * This is the last line. Add the trailing newline
1397 * removed in vprintk_store().
1399 text[prefix_len + line_len] = '\n';
1404 * Advance beyond the added prefix and the related line with
1407 text += prefix_len + line_len + 1;
1410 * The remaining text has only decreased by the line with its
1413 * Note that @text_len can become zero. It happens when @text
1414 * ended with a newline (either due to truncation or the
1415 * original string ending with "\n\n"). The loop is correctly
1416 * repeated and (if not truncated) an empty line with a prefix
1419 text_len -= line_len + 1;
1423 * If a buffer was provided, it will be terminated. Space for the
1424 * string terminator is guaranteed to be available. The terminator is
1425 * not counted in the return value.
1428 r->text_buf[len] = 0;
1433 static size_t get_record_print_text_size(struct printk_info *info,
1434 unsigned int line_count,
1435 bool syslog, bool time)
1437 char prefix[PREFIX_MAX];
1440 prefix_len = info_print_prefix(info, syslog, time, prefix);
1443 * Each line will be preceded with a prefix. The intermediate
1444 * newlines are already within the text, but a final trailing
1445 * newline will be added.
1447 return ((prefix_len * line_count) + info->text_len + 1);
1451 * Beginning with @start_seq, find the first record where it and all following
1452 * records up to (but not including) @max_seq fit into @size.
1454 * @max_seq is simply an upper bound and does not need to exist. If the caller
1455 * does not require an upper bound, -1 can be used for @max_seq.
1457 static u64 find_first_fitting_seq(u64 start_seq, u64 max_seq, size_t size,
1458 bool syslog, bool time)
1460 struct printk_info info;
1461 unsigned int line_count;
1465 /* Determine the size of the records up to @max_seq. */
1466 prb_for_each_info(start_seq, prb, seq, &info, &line_count) {
1467 if (info.seq >= max_seq)
1469 len += get_record_print_text_size(&info, line_count, syslog, time);
1473 * Adjust the upper bound for the next loop to avoid subtracting
1474 * lengths that were never added.
1480 * Move first record forward until length fits into the buffer. Ignore
1481 * newest messages that were not counted in the above cycle. Messages
1482 * might appear and get lost in the meantime. This is a best effort
1483 * that prevents an infinite loop that could occur with a retry.
1485 prb_for_each_info(start_seq, prb, seq, &info, &line_count) {
1486 if (len <= size || info.seq >= max_seq)
1488 len -= get_record_print_text_size(&info, line_count, syslog, time);
1494 /* The caller is responsible for making sure @size is greater than 0. */
1495 static int syslog_print(char __user *buf, int size)
1497 struct printk_info info;
1498 struct printk_record r;
1503 text = kmalloc(CONSOLE_LOG_MAX, GFP_KERNEL);
1507 prb_rec_init_rd(&r, &info, text, CONSOLE_LOG_MAX);
1509 mutex_lock(&syslog_lock);
1512 * Wait for the @syslog_seq record to be available. @syslog_seq may
1513 * change while waiting.
1518 mutex_unlock(&syslog_lock);
1520 * Guarantee this task is visible on the waitqueue before
1521 * checking the wake condition.
1523 * The full memory barrier within set_current_state() of
1524 * prepare_to_wait_event() pairs with the full memory barrier
1525 * within wq_has_sleeper().
1527 * This pairs with __wake_up_klogd:A.
1529 len = wait_event_interruptible(log_wait,
1530 prb_read_valid(prb, seq, NULL)); /* LMM(syslog_print:A) */
1531 mutex_lock(&syslog_lock);
1535 } while (syslog_seq != seq);
1538 * Copy records that fit into the buffer. The above cycle makes sure
1539 * that the first record is always available.
1546 if (!prb_read_valid(prb, syslog_seq, &r))
1549 if (r.info->seq != syslog_seq) {
1550 /* message is gone, move to next valid one */
1551 syslog_seq = r.info->seq;
1556 * To keep reading/counting partial line consistent,
1557 * use printk_time value as of the beginning of a line.
1559 if (!syslog_partial)
1560 syslog_time = printk_time;
1562 skip = syslog_partial;
1563 n = record_print_text(&r, true, syslog_time);
1564 if (n - syslog_partial <= size) {
1565 /* message fits into buffer, move forward */
1566 syslog_seq = r.info->seq + 1;
1567 n -= syslog_partial;
1570 /* partial read(), remember position */
1572 syslog_partial += n;
1579 mutex_unlock(&syslog_lock);
1580 err = copy_to_user(buf, text + skip, n);
1581 mutex_lock(&syslog_lock);
1594 mutex_unlock(&syslog_lock);
1599 static int syslog_print_all(char __user *buf, int size, bool clear)
1601 struct printk_info info;
1602 struct printk_record r;
1608 text = kmalloc(CONSOLE_LOG_MAX, GFP_KERNEL);
1614 * Find first record that fits, including all following records,
1615 * into the user-provided buffer for this dump.
1617 seq = find_first_fitting_seq(latched_seq_read_nolock(&clear_seq), -1,
1620 prb_rec_init_rd(&r, &info, text, CONSOLE_LOG_MAX);
1623 prb_for_each_record(seq, prb, seq, &r) {
1626 textlen = record_print_text(&r, true, time);
1628 if (len + textlen > size) {
1633 if (copy_to_user(buf + len, text, textlen))
1643 mutex_lock(&syslog_lock);
1644 latched_seq_write(&clear_seq, seq);
1645 mutex_unlock(&syslog_lock);
1652 static void syslog_clear(void)
1654 mutex_lock(&syslog_lock);
1655 latched_seq_write(&clear_seq, prb_next_seq(prb));
1656 mutex_unlock(&syslog_lock);
1659 int do_syslog(int type, char __user *buf, int len, int source)
1661 struct printk_info info;
1663 static int saved_console_loglevel = LOGLEVEL_DEFAULT;
1666 error = check_syslog_permissions(type, source);
1671 case SYSLOG_ACTION_CLOSE: /* Close log */
1673 case SYSLOG_ACTION_OPEN: /* Open log */
1675 case SYSLOG_ACTION_READ: /* Read from log */
1676 if (!buf || len < 0)
1680 if (!access_ok(buf, len))
1682 error = syslog_print(buf, len);
1684 /* Read/clear last kernel messages */
1685 case SYSLOG_ACTION_READ_CLEAR:
1688 /* Read last kernel messages */
1689 case SYSLOG_ACTION_READ_ALL:
1690 if (!buf || len < 0)
1694 if (!access_ok(buf, len))
1696 error = syslog_print_all(buf, len, clear);
1698 /* Clear ring buffer */
1699 case SYSLOG_ACTION_CLEAR:
1702 /* Disable logging to console */
1703 case SYSLOG_ACTION_CONSOLE_OFF:
1704 if (saved_console_loglevel == LOGLEVEL_DEFAULT)
1705 saved_console_loglevel = console_loglevel;
1706 console_loglevel = minimum_console_loglevel;
1708 /* Enable logging to console */
1709 case SYSLOG_ACTION_CONSOLE_ON:
1710 if (saved_console_loglevel != LOGLEVEL_DEFAULT) {
1711 console_loglevel = saved_console_loglevel;
1712 saved_console_loglevel = LOGLEVEL_DEFAULT;
1715 /* Set level of messages printed to console */
1716 case SYSLOG_ACTION_CONSOLE_LEVEL:
1717 if (len < 1 || len > 8)
1719 if (len < minimum_console_loglevel)
1720 len = minimum_console_loglevel;
1721 console_loglevel = len;
1722 /* Implicitly re-enable logging to console */
1723 saved_console_loglevel = LOGLEVEL_DEFAULT;
1725 /* Number of chars in the log buffer */
1726 case SYSLOG_ACTION_SIZE_UNREAD:
1727 mutex_lock(&syslog_lock);
1728 if (!prb_read_valid_info(prb, syslog_seq, &info, NULL)) {
1729 /* No unread messages. */
1730 mutex_unlock(&syslog_lock);
1733 if (info.seq != syslog_seq) {
1734 /* messages are gone, move to first one */
1735 syslog_seq = info.seq;
1738 if (source == SYSLOG_FROM_PROC) {
1740 * Short-cut for poll(/"proc/kmsg") which simply checks
1741 * for pending data, not the size; return the count of
1742 * records, not the length.
1744 error = prb_next_seq(prb) - syslog_seq;
1746 bool time = syslog_partial ? syslog_time : printk_time;
1747 unsigned int line_count;
1750 prb_for_each_info(syslog_seq, prb, seq, &info,
1752 error += get_record_print_text_size(&info, line_count,
1756 error -= syslog_partial;
1758 mutex_unlock(&syslog_lock);
1760 /* Size of the log buffer */
1761 case SYSLOG_ACTION_SIZE_BUFFER:
1762 error = log_buf_len;
1772 SYSCALL_DEFINE3(syslog, int, type, char __user *, buf, int, len)
1774 return do_syslog(type, buf, len, SYSLOG_FROM_READER);
1778 * Special console_lock variants that help to reduce the risk of soft-lockups.
1779 * They allow to pass console_lock to another printk() call using a busy wait.
1782 #ifdef CONFIG_LOCKDEP
1783 static struct lockdep_map console_owner_dep_map = {
1784 .name = "console_owner"
1788 static DEFINE_RAW_SPINLOCK(console_owner_lock);
1789 static struct task_struct *console_owner;
1790 static bool console_waiter;
1793 * console_lock_spinning_enable - mark beginning of code where another
1794 * thread might safely busy wait
1796 * This basically converts console_lock into a spinlock. This marks
1797 * the section where the console_lock owner can not sleep, because
1798 * there may be a waiter spinning (like a spinlock). Also it must be
1799 * ready to hand over the lock at the end of the section.
1801 static void console_lock_spinning_enable(void)
1803 raw_spin_lock(&console_owner_lock);
1804 console_owner = current;
1805 raw_spin_unlock(&console_owner_lock);
1807 /* The waiter may spin on us after setting console_owner */
1808 spin_acquire(&console_owner_dep_map, 0, 0, _THIS_IP_);
1812 * console_lock_spinning_disable_and_check - mark end of code where another
1813 * thread was able to busy wait and check if there is a waiter
1815 * This is called at the end of the section where spinning is allowed.
1816 * It has two functions. First, it is a signal that it is no longer
1817 * safe to start busy waiting for the lock. Second, it checks if
1818 * there is a busy waiter and passes the lock rights to her.
1820 * Important: Callers lose the lock if there was a busy waiter.
1821 * They must not touch items synchronized by console_lock
1824 * Return: 1 if the lock rights were passed, 0 otherwise.
1826 static int console_lock_spinning_disable_and_check(void)
1830 raw_spin_lock(&console_owner_lock);
1831 waiter = READ_ONCE(console_waiter);
1832 console_owner = NULL;
1833 raw_spin_unlock(&console_owner_lock);
1836 spin_release(&console_owner_dep_map, _THIS_IP_);
1840 /* The waiter is now free to continue */
1841 WRITE_ONCE(console_waiter, false);
1843 spin_release(&console_owner_dep_map, _THIS_IP_);
1846 * Hand off console_lock to waiter. The waiter will perform
1847 * the up(). After this, the waiter is the console_lock owner.
1849 mutex_release(&console_lock_dep_map, _THIS_IP_);
1854 * console_trylock_spinning - try to get console_lock by busy waiting
1856 * This allows to busy wait for the console_lock when the current
1857 * owner is running in specially marked sections. It means that
1858 * the current owner is running and cannot reschedule until it
1859 * is ready to lose the lock.
1861 * Return: 1 if we got the lock, 0 othrewise
1863 static int console_trylock_spinning(void)
1865 struct task_struct *owner = NULL;
1868 unsigned long flags;
1870 if (console_trylock())
1874 * It's unsafe to spin once a panic has begun. If we are the
1875 * panic CPU, we may have already halted the owner of the
1876 * console_sem. If we are not the panic CPU, then we should
1877 * avoid taking console_sem, so the panic CPU has a better
1878 * chance of cleanly acquiring it later.
1880 if (panic_in_progress())
1883 printk_safe_enter_irqsave(flags);
1885 raw_spin_lock(&console_owner_lock);
1886 owner = READ_ONCE(console_owner);
1887 waiter = READ_ONCE(console_waiter);
1888 if (!waiter && owner && owner != current) {
1889 WRITE_ONCE(console_waiter, true);
1892 raw_spin_unlock(&console_owner_lock);
1895 * If there is an active printk() writing to the
1896 * consoles, instead of having it write our data too,
1897 * see if we can offload that load from the active
1898 * printer, and do some printing ourselves.
1899 * Go into a spin only if there isn't already a waiter
1900 * spinning, and there is an active printer, and
1901 * that active printer isn't us (recursive printk?).
1904 printk_safe_exit_irqrestore(flags);
1908 /* We spin waiting for the owner to release us */
1909 spin_acquire(&console_owner_dep_map, 0, 0, _THIS_IP_);
1910 /* Owner will clear console_waiter on hand off */
1911 while (READ_ONCE(console_waiter))
1913 spin_release(&console_owner_dep_map, _THIS_IP_);
1915 printk_safe_exit_irqrestore(flags);
1917 * The owner passed the console lock to us.
1918 * Since we did not spin on console lock, annotate
1919 * this as a trylock. Otherwise lockdep will
1922 mutex_acquire(&console_lock_dep_map, 0, 1, _THIS_IP_);
1928 * Call the specified console driver, asking it to write out the specified
1929 * text and length. If @dropped_text is non-NULL and any records have been
1930 * dropped, a dropped message will be written out first.
1932 static void call_console_driver(struct console *con, const char *text, size_t len,
1937 if (con->dropped && dropped_text) {
1938 dropped_len = snprintf(dropped_text, DROPPED_TEXT_MAX,
1939 "** %lu printk messages dropped **\n",
1942 con->write(con, dropped_text, dropped_len);
1945 con->write(con, text, len);
1949 * Recursion is tracked separately on each CPU. If NMIs are supported, an
1950 * additional NMI context per CPU is also separately tracked. Until per-CPU
1951 * is available, a separate "early tracking" is performed.
1953 static DEFINE_PER_CPU(u8, printk_count);
1954 static u8 printk_count_early;
1955 #ifdef CONFIG_HAVE_NMI
1956 static DEFINE_PER_CPU(u8, printk_count_nmi);
1957 static u8 printk_count_nmi_early;
1961 * Recursion is limited to keep the output sane. printk() should not require
1962 * more than 1 level of recursion (allowing, for example, printk() to trigger
1963 * a WARN), but a higher value is used in case some printk-internal errors
1964 * exist, such as the ringbuffer validation checks failing.
1966 #define PRINTK_MAX_RECURSION 3
1969 * Return a pointer to the dedicated counter for the CPU+context of the
1972 static u8 *__printk_recursion_counter(void)
1974 #ifdef CONFIG_HAVE_NMI
1976 if (printk_percpu_data_ready())
1977 return this_cpu_ptr(&printk_count_nmi);
1978 return &printk_count_nmi_early;
1981 if (printk_percpu_data_ready())
1982 return this_cpu_ptr(&printk_count);
1983 return &printk_count_early;
1987 * Enter recursion tracking. Interrupts are disabled to simplify tracking.
1988 * The caller must check the boolean return value to see if the recursion is
1989 * allowed. On failure, interrupts are not disabled.
1991 * @recursion_ptr must be a variable of type (u8 *) and is the same variable
1992 * that is passed to printk_exit_irqrestore().
1994 #define printk_enter_irqsave(recursion_ptr, flags) \
1996 bool success = true; \
1998 typecheck(u8 *, recursion_ptr); \
1999 local_irq_save(flags); \
2000 (recursion_ptr) = __printk_recursion_counter(); \
2001 if (*(recursion_ptr) > PRINTK_MAX_RECURSION) { \
2002 local_irq_restore(flags); \
2005 (*(recursion_ptr))++; \
2010 /* Exit recursion tracking, restoring interrupts. */
2011 #define printk_exit_irqrestore(recursion_ptr, flags) \
2013 typecheck(u8 *, recursion_ptr); \
2014 (*(recursion_ptr))--; \
2015 local_irq_restore(flags); \
2018 int printk_delay_msec __read_mostly;
2020 static inline void printk_delay(int level)
2022 boot_delay_msec(level);
2024 if (unlikely(printk_delay_msec)) {
2025 int m = printk_delay_msec;
2029 touch_nmi_watchdog();
2034 static inline u32 printk_caller_id(void)
2036 return in_task() ? task_pid_nr(current) :
2037 0x80000000 + smp_processor_id();
2041 * printk_parse_prefix - Parse level and control flags.
2043 * @text: The terminated text message.
2044 * @level: A pointer to the current level value, will be updated.
2045 * @flags: A pointer to the current printk_info flags, will be updated.
2047 * @level may be NULL if the caller is not interested in the parsed value.
2048 * Otherwise the variable pointed to by @level must be set to
2049 * LOGLEVEL_DEFAULT in order to be updated with the parsed value.
2051 * @flags may be NULL if the caller is not interested in the parsed value.
2052 * Otherwise the variable pointed to by @flags will be OR'd with the parsed
2055 * Return: The length of the parsed level and control flags.
2057 u16 printk_parse_prefix(const char *text, int *level,
2058 enum printk_info_flags *flags)
2064 kern_level = printk_get_level(text);
2068 switch (kern_level) {
2070 if (level && *level == LOGLEVEL_DEFAULT)
2071 *level = kern_level - '0';
2073 case 'c': /* KERN_CONT */
2086 static u16 printk_sprint(char *text, u16 size, int facility,
2087 enum printk_info_flags *flags, const char *fmt,
2092 text_len = vscnprintf(text, size, fmt, args);
2094 /* Mark and strip a trailing newline. */
2095 if (text_len && text[text_len - 1] == '\n') {
2097 *flags |= LOG_NEWLINE;
2100 /* Strip log level and control flags. */
2101 if (facility == 0) {
2104 prefix_len = printk_parse_prefix(text, NULL, NULL);
2106 text_len -= prefix_len;
2107 memmove(text, text + prefix_len, text_len);
2111 trace_console_rcuidle(text, text_len);
2117 int vprintk_store(int facility, int level,
2118 const struct dev_printk_info *dev_info,
2119 const char *fmt, va_list args)
2121 struct prb_reserved_entry e;
2122 enum printk_info_flags flags = 0;
2123 struct printk_record r;
2124 unsigned long irqflags;
2125 u16 trunc_msg_len = 0;
2135 if (!printk_enter_irqsave(recursion_ptr, irqflags))
2139 * Since the duration of printk() can vary depending on the message
2140 * and state of the ringbuffer, grab the timestamp now so that it is
2141 * close to the call of printk(). This provides a more deterministic
2142 * timestamp with respect to the caller.
2144 ts_nsec = local_clock();
2146 caller_id = printk_caller_id();
2149 * The sprintf needs to come first since the syslog prefix might be
2150 * passed in as a parameter. An extra byte must be reserved so that
2151 * later the vscnprintf() into the reserved buffer has room for the
2152 * terminating '\0', which is not counted by vsnprintf().
2154 va_copy(args2, args);
2155 reserve_size = vsnprintf(&prefix_buf[0], sizeof(prefix_buf), fmt, args2) + 1;
2158 if (reserve_size > LOG_LINE_MAX)
2159 reserve_size = LOG_LINE_MAX;
2161 /* Extract log level or control flags. */
2163 printk_parse_prefix(&prefix_buf[0], &level, &flags);
2165 if (level == LOGLEVEL_DEFAULT)
2166 level = default_message_loglevel;
2169 flags |= LOG_NEWLINE;
2171 if (flags & LOG_CONT) {
2172 prb_rec_init_wr(&r, reserve_size);
2173 if (prb_reserve_in_last(&e, prb, &r, caller_id, LOG_LINE_MAX)) {
2174 text_len = printk_sprint(&r.text_buf[r.info->text_len], reserve_size,
2175 facility, &flags, fmt, args);
2176 r.info->text_len += text_len;
2178 if (flags & LOG_NEWLINE) {
2179 r.info->flags |= LOG_NEWLINE;
2180 prb_final_commit(&e);
2191 * Explicitly initialize the record before every prb_reserve() call.
2192 * prb_reserve_in_last() and prb_reserve() purposely invalidate the
2193 * structure when they fail.
2195 prb_rec_init_wr(&r, reserve_size);
2196 if (!prb_reserve(&e, prb, &r)) {
2197 /* truncate the message if it is too long for empty buffer */
2198 truncate_msg(&reserve_size, &trunc_msg_len);
2200 prb_rec_init_wr(&r, reserve_size + trunc_msg_len);
2201 if (!prb_reserve(&e, prb, &r))
2206 text_len = printk_sprint(&r.text_buf[0], reserve_size, facility, &flags, fmt, args);
2208 memcpy(&r.text_buf[text_len], trunc_msg, trunc_msg_len);
2209 r.info->text_len = text_len + trunc_msg_len;
2210 r.info->facility = facility;
2211 r.info->level = level & 7;
2212 r.info->flags = flags & 0x1f;
2213 r.info->ts_nsec = ts_nsec;
2214 r.info->caller_id = caller_id;
2216 memcpy(&r.info->dev_info, dev_info, sizeof(r.info->dev_info));
2218 /* A message without a trailing newline can be continued. */
2219 if (!(flags & LOG_NEWLINE))
2222 prb_final_commit(&e);
2224 ret = text_len + trunc_msg_len;
2226 printk_exit_irqrestore(recursion_ptr, irqflags);
2230 asmlinkage int vprintk_emit(int facility, int level,
2231 const struct dev_printk_info *dev_info,
2232 const char *fmt, va_list args)
2235 bool in_sched = false;
2237 /* Suppress unimportant messages after panic happens */
2238 if (unlikely(suppress_printk))
2241 if (unlikely(suppress_panic_printk) &&
2242 atomic_read(&panic_cpu) != raw_smp_processor_id())
2245 if (level == LOGLEVEL_SCHED) {
2246 level = LOGLEVEL_DEFAULT;
2250 printk_delay(level);
2252 printed_len = vprintk_store(facility, level, dev_info, fmt, args);
2254 /* If called from the scheduler, we can not call up(). */
2257 * The caller may be holding system-critical or
2258 * timing-sensitive locks. Disable preemption during
2259 * printing of all remaining records to all consoles so that
2260 * this context can return as soon as possible. Hopefully
2261 * another printk() caller will take over the printing.
2265 * Try to acquire and then immediately release the console
2266 * semaphore. The release will print out buffers. With the
2267 * spinning variant, this context tries to take over the
2268 * printing from another printing context.
2270 if (console_trylock_spinning())
2278 EXPORT_SYMBOL(vprintk_emit);
2280 int vprintk_default(const char *fmt, va_list args)
2282 return vprintk_emit(0, LOGLEVEL_DEFAULT, NULL, fmt, args);
2284 EXPORT_SYMBOL_GPL(vprintk_default);
2286 asmlinkage __visible int _printk(const char *fmt, ...)
2291 va_start(args, fmt);
2292 r = vprintk(fmt, args);
2297 EXPORT_SYMBOL(_printk);
2299 static bool __pr_flush(struct console *con, int timeout_ms, bool reset_on_progress);
2301 #else /* CONFIG_PRINTK */
2303 #define CONSOLE_LOG_MAX 0
2304 #define DROPPED_TEXT_MAX 0
2305 #define printk_time false
2307 #define prb_read_valid(rb, seq, r) false
2308 #define prb_first_valid_seq(rb) 0
2309 #define prb_next_seq(rb) 0
2311 static u64 syslog_seq;
2313 static size_t record_print_text(const struct printk_record *r,
2314 bool syslog, bool time)
2318 static ssize_t info_print_ext_header(char *buf, size_t size,
2319 struct printk_info *info)
2323 static ssize_t msg_print_ext_body(char *buf, size_t size,
2324 char *text, size_t text_len,
2325 struct dev_printk_info *dev_info) { return 0; }
2326 static void console_lock_spinning_enable(void) { }
2327 static int console_lock_spinning_disable_and_check(void) { return 0; }
2328 static void call_console_driver(struct console *con, const char *text, size_t len,
2332 static bool suppress_message_printing(int level) { return false; }
2333 static bool __pr_flush(struct console *con, int timeout_ms, bool reset_on_progress) { return true; }
2335 #endif /* CONFIG_PRINTK */
2337 #ifdef CONFIG_EARLY_PRINTK
2338 struct console *early_console;
2340 asmlinkage __visible void early_printk(const char *fmt, ...)
2350 n = vscnprintf(buf, sizeof(buf), fmt, ap);
2353 early_console->write(early_console, buf, n);
2357 static void set_user_specified(struct console_cmdline *c, bool user_specified)
2359 if (!user_specified)
2363 * @c console was defined by the user on the command line.
2364 * Do not clear when added twice also by SPCR or the device tree.
2366 c->user_specified = true;
2367 /* At least one console defined by the user on the command line. */
2368 console_set_on_cmdline = 1;
2371 static int __add_preferred_console(char *name, int idx, char *options,
2372 char *brl_options, bool user_specified)
2374 struct console_cmdline *c;
2378 * See if this tty is not yet registered, and
2379 * if we have a slot free.
2381 for (i = 0, c = console_cmdline;
2382 i < MAX_CMDLINECONSOLES && c->name[0];
2384 if (strcmp(c->name, name) == 0 && c->index == idx) {
2386 preferred_console = i;
2387 set_user_specified(c, user_specified);
2391 if (i == MAX_CMDLINECONSOLES)
2394 preferred_console = i;
2395 strlcpy(c->name, name, sizeof(c->name));
2396 c->options = options;
2397 set_user_specified(c, user_specified);
2398 braille_set_options(c, brl_options);
2404 static int __init console_msg_format_setup(char *str)
2406 if (!strcmp(str, "syslog"))
2407 console_msg_format = MSG_FORMAT_SYSLOG;
2408 if (!strcmp(str, "default"))
2409 console_msg_format = MSG_FORMAT_DEFAULT;
2412 __setup("console_msg_format=", console_msg_format_setup);
2415 * Set up a console. Called via do_early_param() in init/main.c
2416 * for each "console=" parameter in the boot command line.
2418 static int __init console_setup(char *str)
2420 char buf[sizeof(console_cmdline[0].name) + 4]; /* 4 for "ttyS" */
2421 char *s, *options, *brl_options = NULL;
2425 * console="" or console=null have been suggested as a way to
2426 * disable console output. Use ttynull that has been created
2427 * for exactly this purpose.
2429 if (str[0] == 0 || strcmp(str, "null") == 0) {
2430 __add_preferred_console("ttynull", 0, NULL, NULL, true);
2434 if (_braille_console_setup(&str, &brl_options))
2438 * Decode str into name, index, options.
2440 if (str[0] >= '0' && str[0] <= '9') {
2441 strcpy(buf, "ttyS");
2442 strncpy(buf + 4, str, sizeof(buf) - 5);
2444 strncpy(buf, str, sizeof(buf) - 1);
2446 buf[sizeof(buf) - 1] = 0;
2447 options = strchr(str, ',');
2451 if (!strcmp(str, "ttya"))
2452 strcpy(buf, "ttyS0");
2453 if (!strcmp(str, "ttyb"))
2454 strcpy(buf, "ttyS1");
2456 for (s = buf; *s; s++)
2457 if (isdigit(*s) || *s == ',')
2459 idx = simple_strtoul(s, NULL, 10);
2462 __add_preferred_console(buf, idx, options, brl_options, true);
2465 __setup("console=", console_setup);
2468 * add_preferred_console - add a device to the list of preferred consoles.
2469 * @name: device name
2470 * @idx: device index
2471 * @options: options for this console
2473 * The last preferred console added will be used for kernel messages
2474 * and stdin/out/err for init. Normally this is used by console_setup
2475 * above to handle user-supplied console arguments; however it can also
2476 * be used by arch-specific code either to override the user or more
2477 * commonly to provide a default console (ie from PROM variables) when
2478 * the user has not supplied one.
2480 int add_preferred_console(char *name, int idx, char *options)
2482 return __add_preferred_console(name, idx, options, NULL, false);
2485 bool console_suspend_enabled = true;
2486 EXPORT_SYMBOL(console_suspend_enabled);
2488 static int __init console_suspend_disable(char *str)
2490 console_suspend_enabled = false;
2493 __setup("no_console_suspend", console_suspend_disable);
2494 module_param_named(console_suspend, console_suspend_enabled,
2495 bool, S_IRUGO | S_IWUSR);
2496 MODULE_PARM_DESC(console_suspend, "suspend console during suspend"
2497 " and hibernate operations");
2499 static bool printk_console_no_auto_verbose;
2501 void console_verbose(void)
2503 if (console_loglevel && !printk_console_no_auto_verbose)
2504 console_loglevel = CONSOLE_LOGLEVEL_MOTORMOUTH;
2506 EXPORT_SYMBOL_GPL(console_verbose);
2508 module_param_named(console_no_auto_verbose, printk_console_no_auto_verbose, bool, 0644);
2509 MODULE_PARM_DESC(console_no_auto_verbose, "Disable console loglevel raise to highest on oops/panic/etc");
2512 * suspend_console - suspend the console subsystem
2514 * This disables printk() while we go into suspend states
2516 void suspend_console(void)
2518 if (!console_suspend_enabled)
2520 pr_info("Suspending console(s) (use no_console_suspend to debug)\n");
2521 pr_flush(1000, true);
2523 console_suspended = 1;
2527 void resume_console(void)
2529 if (!console_suspend_enabled)
2532 console_suspended = 0;
2534 pr_flush(1000, true);
2538 * console_cpu_notify - print deferred console messages after CPU hotplug
2541 * If printk() is called from a CPU that is not online yet, the messages
2542 * will be printed on the console only if there are CON_ANYTIME consoles.
2543 * This function is called when a new CPU comes online (or fails to come
2544 * up) or goes offline.
2546 static int console_cpu_notify(unsigned int cpu)
2548 if (!cpuhp_tasks_frozen) {
2549 /* If trylock fails, someone else is doing the printing */
2550 if (console_trylock())
2557 * console_lock - lock the console system for exclusive use.
2559 * Acquires a lock which guarantees that the caller has
2560 * exclusive access to the console system and the console_drivers list.
2562 * Can sleep, returns nothing.
2564 void console_lock(void)
2569 if (console_suspended)
2572 console_may_schedule = 1;
2574 EXPORT_SYMBOL(console_lock);
2577 * console_trylock - try to lock the console system for exclusive use.
2579 * Try to acquire a lock which guarantees that the caller has exclusive
2580 * access to the console system and the console_drivers list.
2582 * returns 1 on success, and 0 on failure to acquire the lock.
2584 int console_trylock(void)
2586 if (down_trylock_console_sem())
2588 if (console_suspended) {
2593 console_may_schedule = 0;
2596 EXPORT_SYMBOL(console_trylock);
2598 int is_console_locked(void)
2600 return console_locked;
2602 EXPORT_SYMBOL(is_console_locked);
2605 * Return true when this CPU should unlock console_sem without pushing all
2606 * messages to the console. This reduces the chance that the console is
2607 * locked when the panic CPU tries to use it.
2609 static bool abandon_console_lock_in_panic(void)
2611 if (!panic_in_progress())
2615 * We can use raw_smp_processor_id() here because it is impossible for
2616 * the task to be migrated to the panic_cpu, or away from it. If
2617 * panic_cpu has already been set, and we're not currently executing on
2618 * that CPU, then we never will be.
2620 return atomic_read(&panic_cpu) != raw_smp_processor_id();
2624 * Check if the given console is currently capable and allowed to print
2627 * Requires the console_lock.
2629 static inline bool console_is_usable(struct console *con)
2631 if (!(con->flags & CON_ENABLED))
2638 * Console drivers may assume that per-cpu resources have been
2639 * allocated. So unless they're explicitly marked as being able to
2640 * cope (CON_ANYTIME) don't call them until this CPU is officially up.
2642 if (!cpu_online(raw_smp_processor_id()) &&
2643 !(con->flags & CON_ANYTIME))
2649 static void __console_unlock(void)
2656 * Print one record for the given console. The record printed is whatever
2657 * record is the next available record for the given console.
2659 * @text is a buffer of size CONSOLE_LOG_MAX.
2661 * If extended messages should be printed, @ext_text is a buffer of size
2662 * CONSOLE_EXT_LOG_MAX. Otherwise @ext_text must be NULL.
2664 * If dropped messages should be printed, @dropped_text is a buffer of size
2665 * DROPPED_TEXT_MAX. Otherwise @dropped_text must be NULL.
2667 * @handover will be set to true if a printk waiter has taken over the
2668 * console_lock, in which case the caller is no longer holding the
2669 * console_lock. Otherwise it is set to false.
2671 * Returns false if the given console has no next record to print, otherwise
2674 * Requires the console_lock.
2676 static bool console_emit_next_record(struct console *con, char *text, char *ext_text,
2677 char *dropped_text, bool *handover)
2679 static int panic_console_dropped;
2680 struct printk_info info;
2681 struct printk_record r;
2682 unsigned long flags;
2686 prb_rec_init_rd(&r, &info, text, CONSOLE_LOG_MAX);
2690 if (!prb_read_valid(prb, con->seq, &r))
2693 if (con->seq != r.info->seq) {
2694 con->dropped += r.info->seq - con->seq;
2695 con->seq = r.info->seq;
2696 if (panic_in_progress() && panic_console_dropped++ > 10) {
2697 suppress_panic_printk = 1;
2698 pr_warn_once("Too many dropped messages. Suppress messages on non-panic CPUs to prevent livelock.\n");
2702 /* Skip record that has level above the console loglevel. */
2703 if (suppress_message_printing(r.info->level)) {
2709 write_text = ext_text;
2710 len = info_print_ext_header(ext_text, CONSOLE_EXT_LOG_MAX, r.info);
2711 len += msg_print_ext_body(ext_text + len, CONSOLE_EXT_LOG_MAX - len,
2712 &r.text_buf[0], r.info->text_len, &r.info->dev_info);
2715 len = record_print_text(&r, console_msg_format & MSG_FORMAT_SYSLOG, printk_time);
2719 * While actively printing out messages, if another printk()
2720 * were to occur on another CPU, it may wait for this one to
2721 * finish. This task can not be preempted if there is a
2722 * waiter waiting to take over.
2724 * Interrupts are disabled because the hand over to a waiter
2725 * must not be interrupted until the hand over is completed
2726 * (@console_waiter is cleared).
2728 printk_safe_enter_irqsave(flags);
2729 console_lock_spinning_enable();
2731 stop_critical_timings(); /* don't trace print latency */
2732 call_console_driver(con, write_text, len, dropped_text);
2733 start_critical_timings();
2737 *handover = console_lock_spinning_disable_and_check();
2738 printk_safe_exit_irqrestore(flags);
2744 * Print out all remaining records to all consoles.
2746 * @do_cond_resched is set by the caller. It can be true only in schedulable
2749 * @next_seq is set to the sequence number after the last available record.
2750 * The value is valid only when this function returns true. It means that all
2751 * usable consoles are completely flushed.
2753 * @handover will be set to true if a printk waiter has taken over the
2754 * console_lock, in which case the caller is no longer holding the
2755 * console_lock. Otherwise it is set to false.
2757 * Returns true when there was at least one usable console and all messages
2758 * were flushed to all usable consoles. A returned false informs the caller
2759 * that everything was not flushed (either there were no usable consoles or
2760 * another context has taken over printing or it is a panic situation and this
2761 * is not the panic CPU). Regardless the reason, the caller should assume it
2762 * is not useful to immediately try again.
2764 * Requires the console_lock.
2766 static bool console_flush_all(bool do_cond_resched, u64 *next_seq, bool *handover)
2768 static char dropped_text[DROPPED_TEXT_MAX];
2769 static char ext_text[CONSOLE_EXT_LOG_MAX];
2770 static char text[CONSOLE_LOG_MAX];
2771 bool any_usable = false;
2772 struct console *con;
2779 any_progress = false;
2781 for_each_console(con) {
2784 if (!console_is_usable(con))
2788 if (con->flags & CON_EXTENDED) {
2789 /* Extended consoles do not print "dropped messages". */
2790 progress = console_emit_next_record(con, &text[0],
2794 progress = console_emit_next_record(con, &text[0],
2795 NULL, &dropped_text[0],
2801 /* Track the next of the highest seq flushed. */
2802 if (con->seq > *next_seq)
2803 *next_seq = con->seq;
2807 any_progress = true;
2809 /* Allow panic_cpu to take over the consoles safely. */
2810 if (abandon_console_lock_in_panic())
2813 if (do_cond_resched)
2816 } while (any_progress);
2822 * console_unlock - unlock the console system
2824 * Releases the console_lock which the caller holds on the console system
2825 * and the console driver list.
2827 * While the console_lock was held, console output may have been buffered
2828 * by printk(). If this is the case, console_unlock(); emits
2829 * the output prior to releasing the lock.
2831 * console_unlock(); may be called from any context.
2833 void console_unlock(void)
2835 bool do_cond_resched;
2840 if (console_suspended) {
2846 * Console drivers are called with interrupts disabled, so
2847 * @console_may_schedule should be cleared before; however, we may
2848 * end up dumping a lot of lines, for example, if called from
2849 * console registration path, and should invoke cond_resched()
2850 * between lines if allowable. Not doing so can cause a very long
2851 * scheduling stall on a slow console leading to RCU stall and
2852 * softlockup warnings which exacerbate the issue with more
2853 * messages practically incapacitating the system. Therefore, create
2854 * a local to use for the printing loop.
2856 do_cond_resched = console_may_schedule;
2859 console_may_schedule = 0;
2861 flushed = console_flush_all(do_cond_resched, &next_seq, &handover);
2866 * Abort if there was a failure to flush all messages to all
2867 * usable consoles. Either it is not possible to flush (in
2868 * which case it would be an infinite loop of retrying) or
2869 * another context has taken over printing.
2875 * Some context may have added new records after
2876 * console_flush_all() but before unlocking the console.
2877 * Re-check if there is a new record to flush. If the trylock
2878 * fails, another context is already handling the printing.
2880 } while (prb_read_valid(prb, next_seq, NULL) && console_trylock());
2882 EXPORT_SYMBOL(console_unlock);
2885 * console_conditional_schedule - yield the CPU if required
2887 * If the console code is currently allowed to sleep, and
2888 * if this CPU should yield the CPU to another task, do
2891 * Must be called within console_lock();.
2893 void __sched console_conditional_schedule(void)
2895 if (console_may_schedule)
2898 EXPORT_SYMBOL(console_conditional_schedule);
2900 void console_unblank(void)
2905 * console_unblank can no longer be called in interrupt context unless
2906 * oops_in_progress is set to 1..
2908 if (oops_in_progress) {
2909 if (down_trylock_console_sem() != 0)
2915 console_may_schedule = 0;
2917 if ((c->flags & CON_ENABLED) && c->unblank)
2921 if (!oops_in_progress)
2922 pr_flush(1000, true);
2926 * console_flush_on_panic - flush console content on panic
2927 * @mode: flush all messages in buffer or just the pending ones
2929 * Immediately output all pending messages no matter what.
2931 void console_flush_on_panic(enum con_flush_mode mode)
2934 * If someone else is holding the console lock, trylock will fail
2935 * and may_schedule may be set. Ignore and proceed to unlock so
2936 * that messages are flushed out. As this can be called from any
2937 * context and we don't want to get preempted while flushing,
2938 * ensure may_schedule is cleared.
2941 console_may_schedule = 0;
2943 if (mode == CONSOLE_REPLAY_ALL) {
2947 seq = prb_first_valid_seq(prb);
2955 * Return the console tty driver structure and its associated index
2957 struct tty_driver *console_device(int *index)
2960 struct tty_driver *driver = NULL;
2963 for_each_console(c) {
2966 driver = c->device(c, index);
2975 * Prevent further output on the passed console device so that (for example)
2976 * serial drivers can disable console output before suspending a port, and can
2977 * re-enable output afterwards.
2979 void console_stop(struct console *console)
2981 __pr_flush(console, 1000, true);
2983 console->flags &= ~CON_ENABLED;
2986 EXPORT_SYMBOL(console_stop);
2988 void console_start(struct console *console)
2991 console->flags |= CON_ENABLED;
2993 __pr_flush(console, 1000, true);
2995 EXPORT_SYMBOL(console_start);
2997 static int __read_mostly keep_bootcon;
2999 static int __init keep_bootcon_setup(char *str)
3002 pr_info("debug: skip boot console de-registration.\n");
3007 early_param("keep_bootcon", keep_bootcon_setup);
3010 * This is called by register_console() to try to match
3011 * the newly registered console with any of the ones selected
3012 * by either the command line or add_preferred_console() and
3015 * Care need to be taken with consoles that are statically
3016 * enabled such as netconsole
3018 static int try_enable_preferred_console(struct console *newcon,
3019 bool user_specified)
3021 struct console_cmdline *c;
3024 for (i = 0, c = console_cmdline;
3025 i < MAX_CMDLINECONSOLES && c->name[0];
3027 if (c->user_specified != user_specified)
3029 if (!newcon->match ||
3030 newcon->match(newcon, c->name, c->index, c->options) != 0) {
3031 /* default matching */
3032 BUILD_BUG_ON(sizeof(c->name) != sizeof(newcon->name));
3033 if (strcmp(c->name, newcon->name) != 0)
3035 if (newcon->index >= 0 &&
3036 newcon->index != c->index)
3038 if (newcon->index < 0)
3039 newcon->index = c->index;
3041 if (_braille_register_console(newcon, c))
3044 if (newcon->setup &&
3045 (err = newcon->setup(newcon, c->options)) != 0)
3048 newcon->flags |= CON_ENABLED;
3049 if (i == preferred_console)
3050 newcon->flags |= CON_CONSDEV;
3055 * Some consoles, such as pstore and netconsole, can be enabled even
3056 * without matching. Accept the pre-enabled consoles only when match()
3057 * and setup() had a chance to be called.
3059 if (newcon->flags & CON_ENABLED && c->user_specified == user_specified)
3065 /* Try to enable the console unconditionally */
3066 static void try_enable_default_console(struct console *newcon)
3068 if (newcon->index < 0)
3071 if (newcon->setup && newcon->setup(newcon, NULL) != 0)
3074 newcon->flags |= CON_ENABLED;
3077 newcon->flags |= CON_CONSDEV;
3080 #define con_printk(lvl, con, fmt, ...) \
3081 printk(lvl pr_fmt("%sconsole [%s%d] " fmt), \
3082 (con->flags & CON_BOOT) ? "boot" : "", \
3083 con->name, con->index, ##__VA_ARGS__)
3086 * The console driver calls this routine during kernel initialization
3087 * to register the console printing procedure with printk() and to
3088 * print any messages that were printed by the kernel before the
3089 * console driver was initialized.
3091 * This can happen pretty early during the boot process (because of
3092 * early_printk) - sometimes before setup_arch() completes - be careful
3093 * of what kernel features are used - they may not be initialised yet.
3095 * There are two types of consoles - bootconsoles (early_printk) and
3096 * "real" consoles (everything which is not a bootconsole) which are
3097 * handled differently.
3098 * - Any number of bootconsoles can be registered at any time.
3099 * - As soon as a "real" console is registered, all bootconsoles
3100 * will be unregistered automatically.
3101 * - Once a "real" console is registered, any attempt to register a
3102 * bootconsoles will be rejected
3104 void register_console(struct console *newcon)
3106 struct console *con;
3107 bool bootcon_enabled = false;
3108 bool realcon_enabled = false;
3111 for_each_console(con) {
3112 if (WARN(con == newcon, "console '%s%d' already registered\n",
3113 con->name, con->index))
3117 for_each_console(con) {
3118 if (con->flags & CON_BOOT)
3119 bootcon_enabled = true;
3121 realcon_enabled = true;
3124 /* Do not register boot consoles when there already is a real one. */
3125 if (newcon->flags & CON_BOOT && realcon_enabled) {
3126 pr_info("Too late to register bootconsole %s%d\n",
3127 newcon->name, newcon->index);
3132 * See if we want to enable this console driver by default.
3134 * Nope when a console is preferred by the command line, device
3137 * The first real console with tty binding (driver) wins. More
3138 * consoles might get enabled before the right one is found.
3140 * Note that a console with tty binding will have CON_CONSDEV
3141 * flag set and will be first in the list.
3143 if (preferred_console < 0) {
3144 if (!console_drivers || !console_drivers->device ||
3145 console_drivers->flags & CON_BOOT) {
3146 try_enable_default_console(newcon);
3150 /* See if this console matches one we selected on the command line */
3151 err = try_enable_preferred_console(newcon, true);
3153 /* If not, try to match against the platform default(s) */
3155 err = try_enable_preferred_console(newcon, false);
3157 /* printk() messages are not printed to the Braille console. */
3158 if (err || newcon->flags & CON_BRL)
3162 * If we have a bootconsole, and are switching to a real console,
3163 * don't print everything out again, since when the boot console, and
3164 * the real console are the same physical device, it's annoying to
3165 * see the beginning boot messages twice
3167 if (bootcon_enabled &&
3168 ((newcon->flags & (CON_CONSDEV | CON_BOOT)) == CON_CONSDEV)) {
3169 newcon->flags &= ~CON_PRINTBUFFER;
3173 * Put this console in the list - keep the
3174 * preferred driver at the head of the list.
3177 if ((newcon->flags & CON_CONSDEV) || console_drivers == NULL) {
3178 newcon->next = console_drivers;
3179 console_drivers = newcon;
3181 newcon->next->flags &= ~CON_CONSDEV;
3182 /* Ensure this flag is always set for the head of the list */
3183 newcon->flags |= CON_CONSDEV;
3185 newcon->next = console_drivers->next;
3186 console_drivers->next = newcon;
3189 if (newcon->flags & CON_EXTENDED)
3190 nr_ext_console_drivers++;
3192 newcon->dropped = 0;
3193 if (newcon->flags & CON_PRINTBUFFER) {
3194 /* Get a consistent copy of @syslog_seq. */
3195 mutex_lock(&syslog_lock);
3196 newcon->seq = syslog_seq;
3197 mutex_unlock(&syslog_lock);
3199 /* Begin with next message. */
3200 newcon->seq = prb_next_seq(prb);
3203 console_sysfs_notify();
3206 * By unregistering the bootconsoles after we enable the real console
3207 * we get the "console xxx enabled" message on all the consoles -
3208 * boot consoles, real consoles, etc - this is to ensure that end
3209 * users know there might be something in the kernel's log buffer that
3210 * went to the bootconsole (that they do not see on the real console)
3212 con_printk(KERN_INFO, newcon, "enabled\n");
3213 if (bootcon_enabled &&
3214 ((newcon->flags & (CON_CONSDEV | CON_BOOT)) == CON_CONSDEV) &&
3216 /* We need to iterate through all boot consoles, to make
3217 * sure we print everything out, before we unregister them.
3219 for_each_console(con)
3220 if (con->flags & CON_BOOT)
3221 unregister_console(con);
3224 EXPORT_SYMBOL(register_console);
3226 int unregister_console(struct console *console)
3228 struct console *con;
3231 con_printk(KERN_INFO, console, "disabled\n");
3233 res = _braille_unregister_console(console);
3241 if (console_drivers == console) {
3242 console_drivers=console->next;
3245 for_each_console(con) {
3246 if (con->next == console) {
3247 con->next = console->next;
3255 goto out_disable_unlock;
3257 if (console->flags & CON_EXTENDED)
3258 nr_ext_console_drivers--;
3261 * If this isn't the last console and it has CON_CONSDEV set, we
3262 * need to set it on the next preferred console.
3264 if (console_drivers != NULL && console->flags & CON_CONSDEV)
3265 console_drivers->flags |= CON_CONSDEV;
3267 console->flags &= ~CON_ENABLED;
3269 console_sysfs_notify();
3272 res = console->exit(console);
3277 console->flags &= ~CON_ENABLED;
3282 EXPORT_SYMBOL(unregister_console);
3285 * Initialize the console device. This is called *early*, so
3286 * we can't necessarily depend on lots of kernel help here.
3287 * Just do some early initializations, and do the complex setup
3290 void __init console_init(void)
3294 initcall_entry_t *ce;
3296 /* Setup the default TTY line discipline. */
3300 * set up the console device so that later boot sequences can
3301 * inform about problems etc..
3303 ce = __con_initcall_start;
3304 trace_initcall_level("console");
3305 while (ce < __con_initcall_end) {
3306 call = initcall_from_entry(ce);
3307 trace_initcall_start(call);
3309 trace_initcall_finish(call, ret);
3315 * Some boot consoles access data that is in the init section and which will
3316 * be discarded after the initcalls have been run. To make sure that no code
3317 * will access this data, unregister the boot consoles in a late initcall.
3319 * If for some reason, such as deferred probe or the driver being a loadable
3320 * module, the real console hasn't registered yet at this point, there will
3321 * be a brief interval in which no messages are logged to the console, which
3322 * makes it difficult to diagnose problems that occur during this time.
3324 * To mitigate this problem somewhat, only unregister consoles whose memory
3325 * intersects with the init section. Note that all other boot consoles will
3326 * get unregistered when the real preferred console is registered.
3328 static int __init printk_late_init(void)
3330 struct console *con;
3333 for_each_console(con) {
3334 if (!(con->flags & CON_BOOT))
3337 /* Check addresses that might be used for enabled consoles. */
3338 if (init_section_intersects(con, sizeof(*con)) ||
3339 init_section_contains(con->write, 0) ||
3340 init_section_contains(con->read, 0) ||
3341 init_section_contains(con->device, 0) ||
3342 init_section_contains(con->unblank, 0) ||
3343 init_section_contains(con->data, 0)) {
3345 * Please, consider moving the reported consoles out
3346 * of the init section.
3348 pr_warn("bootconsole [%s%d] uses init memory and must be disabled even before the real one is ready\n",
3349 con->name, con->index);
3350 unregister_console(con);
3353 ret = cpuhp_setup_state_nocalls(CPUHP_PRINTK_DEAD, "printk:dead", NULL,
3354 console_cpu_notify);
3356 ret = cpuhp_setup_state_nocalls(CPUHP_AP_ONLINE_DYN, "printk:online",
3357 console_cpu_notify, NULL);
3359 printk_sysctl_init();
3362 late_initcall(printk_late_init);
3364 #if defined CONFIG_PRINTK
3365 /* If @con is specified, only wait for that console. Otherwise wait for all. */
3366 static bool __pr_flush(struct console *con, int timeout_ms, bool reset_on_progress)
3368 int remaining = timeout_ms;
3377 seq = prb_next_seq(prb);
3383 for_each_console(c) {
3384 if (con && con != c)
3386 if (!console_is_usable(c))
3388 printk_seq = c->seq;
3389 if (printk_seq < seq)
3390 diff += seq - printk_seq;
3394 if (diff != last_diff && reset_on_progress)
3395 remaining = timeout_ms;
3397 if (diff == 0 || remaining == 0)
3400 if (remaining < 0) {
3401 /* no timeout limit */
3403 } else if (remaining < 100) {
3418 * pr_flush() - Wait for printing threads to catch up.
3420 * @timeout_ms: The maximum time (in ms) to wait.
3421 * @reset_on_progress: Reset the timeout if forward progress is seen.
3423 * A value of 0 for @timeout_ms means no waiting will occur. A value of -1
3424 * represents infinite waiting.
3426 * If @reset_on_progress is true, the timeout will be reset whenever any
3427 * printer has been seen to make some forward progress.
3429 * Context: Process context. May sleep while acquiring console lock.
3430 * Return: true if all enabled printers are caught up.
3432 bool pr_flush(int timeout_ms, bool reset_on_progress)
3434 return __pr_flush(NULL, timeout_ms, reset_on_progress);
3436 EXPORT_SYMBOL(pr_flush);
3439 * Delayed printk version, for scheduler-internal messages:
3441 #define PRINTK_PENDING_WAKEUP 0x01
3442 #define PRINTK_PENDING_OUTPUT 0x02
3444 static DEFINE_PER_CPU(int, printk_pending);
3446 static void wake_up_klogd_work_func(struct irq_work *irq_work)
3448 int pending = this_cpu_xchg(printk_pending, 0);
3450 if (pending & PRINTK_PENDING_OUTPUT) {
3451 /* If trylock fails, someone else is doing the printing */
3452 if (console_trylock())
3456 if (pending & PRINTK_PENDING_WAKEUP)
3457 wake_up_interruptible(&log_wait);
3460 static DEFINE_PER_CPU(struct irq_work, wake_up_klogd_work) =
3461 IRQ_WORK_INIT_LAZY(wake_up_klogd_work_func);
3463 static void __wake_up_klogd(int val)
3465 if (!printk_percpu_data_ready())
3470 * Guarantee any new records can be seen by tasks preparing to wait
3471 * before this context checks if the wait queue is empty.
3473 * The full memory barrier within wq_has_sleeper() pairs with the full
3474 * memory barrier within set_current_state() of
3475 * prepare_to_wait_event(), which is called after ___wait_event() adds
3476 * the waiter but before it has checked the wait condition.
3478 * This pairs with devkmsg_read:A and syslog_print:A.
3480 if (wq_has_sleeper(&log_wait) || /* LMM(__wake_up_klogd:A) */
3481 (val & PRINTK_PENDING_OUTPUT)) {
3482 this_cpu_or(printk_pending, val);
3483 irq_work_queue(this_cpu_ptr(&wake_up_klogd_work));
3488 void wake_up_klogd(void)
3490 __wake_up_klogd(PRINTK_PENDING_WAKEUP);
3493 void defer_console_output(void)
3496 * New messages may have been added directly to the ringbuffer
3497 * using vprintk_store(), so wake any waiters as well.
3499 __wake_up_klogd(PRINTK_PENDING_WAKEUP | PRINTK_PENDING_OUTPUT);
3502 void printk_trigger_flush(void)
3504 defer_console_output();
3507 int vprintk_deferred(const char *fmt, va_list args)
3511 r = vprintk_emit(0, LOGLEVEL_SCHED, NULL, fmt, args);
3512 defer_console_output();
3517 int _printk_deferred(const char *fmt, ...)
3522 va_start(args, fmt);
3523 r = vprintk_deferred(fmt, args);
3530 * printk rate limiting, lifted from the networking subsystem.
3532 * This enforces a rate limit: not more than 10 kernel messages
3533 * every 5s to make a denial-of-service attack impossible.
3535 DEFINE_RATELIMIT_STATE(printk_ratelimit_state, 5 * HZ, 10);
3537 int __printk_ratelimit(const char *func)
3539 return ___ratelimit(&printk_ratelimit_state, func);
3541 EXPORT_SYMBOL(__printk_ratelimit);
3544 * printk_timed_ratelimit - caller-controlled printk ratelimiting
3545 * @caller_jiffies: pointer to caller's state
3546 * @interval_msecs: minimum interval between prints
3548 * printk_timed_ratelimit() returns true if more than @interval_msecs
3549 * milliseconds have elapsed since the last time printk_timed_ratelimit()
3552 bool printk_timed_ratelimit(unsigned long *caller_jiffies,
3553 unsigned int interval_msecs)
3555 unsigned long elapsed = jiffies - *caller_jiffies;
3557 if (*caller_jiffies && elapsed <= msecs_to_jiffies(interval_msecs))
3560 *caller_jiffies = jiffies;
3563 EXPORT_SYMBOL(printk_timed_ratelimit);
3565 static DEFINE_SPINLOCK(dump_list_lock);
3566 static LIST_HEAD(dump_list);
3569 * kmsg_dump_register - register a kernel log dumper.
3570 * @dumper: pointer to the kmsg_dumper structure
3572 * Adds a kernel log dumper to the system. The dump callback in the
3573 * structure will be called when the kernel oopses or panics and must be
3574 * set. Returns zero on success and %-EINVAL or %-EBUSY otherwise.
3576 int kmsg_dump_register(struct kmsg_dumper *dumper)
3578 unsigned long flags;
3581 /* The dump callback needs to be set */
3585 spin_lock_irqsave(&dump_list_lock, flags);
3586 /* Don't allow registering multiple times */
3587 if (!dumper->registered) {
3588 dumper->registered = 1;
3589 list_add_tail_rcu(&dumper->list, &dump_list);
3592 spin_unlock_irqrestore(&dump_list_lock, flags);
3596 EXPORT_SYMBOL_GPL(kmsg_dump_register);
3599 * kmsg_dump_unregister - unregister a kmsg dumper.
3600 * @dumper: pointer to the kmsg_dumper structure
3602 * Removes a dump device from the system. Returns zero on success and
3603 * %-EINVAL otherwise.
3605 int kmsg_dump_unregister(struct kmsg_dumper *dumper)
3607 unsigned long flags;
3610 spin_lock_irqsave(&dump_list_lock, flags);
3611 if (dumper->registered) {
3612 dumper->registered = 0;
3613 list_del_rcu(&dumper->list);
3616 spin_unlock_irqrestore(&dump_list_lock, flags);
3621 EXPORT_SYMBOL_GPL(kmsg_dump_unregister);
3623 static bool always_kmsg_dump;
3624 module_param_named(always_kmsg_dump, always_kmsg_dump, bool, S_IRUGO | S_IWUSR);
3626 const char *kmsg_dump_reason_str(enum kmsg_dump_reason reason)
3629 case KMSG_DUMP_PANIC:
3631 case KMSG_DUMP_OOPS:
3633 case KMSG_DUMP_EMERG:
3635 case KMSG_DUMP_SHUTDOWN:
3641 EXPORT_SYMBOL_GPL(kmsg_dump_reason_str);
3644 * kmsg_dump - dump kernel log to kernel message dumpers.
3645 * @reason: the reason (oops, panic etc) for dumping
3647 * Call each of the registered dumper's dump() callback, which can
3648 * retrieve the kmsg records with kmsg_dump_get_line() or
3649 * kmsg_dump_get_buffer().
3651 void kmsg_dump(enum kmsg_dump_reason reason)
3653 struct kmsg_dumper *dumper;
3656 list_for_each_entry_rcu(dumper, &dump_list, list) {
3657 enum kmsg_dump_reason max_reason = dumper->max_reason;
3660 * If client has not provided a specific max_reason, default
3661 * to KMSG_DUMP_OOPS, unless always_kmsg_dump was set.
3663 if (max_reason == KMSG_DUMP_UNDEF) {
3664 max_reason = always_kmsg_dump ? KMSG_DUMP_MAX :
3667 if (reason > max_reason)
3670 /* invoke dumper which will iterate over records */
3671 dumper->dump(dumper, reason);
3677 * kmsg_dump_get_line - retrieve one kmsg log line
3678 * @iter: kmsg dump iterator
3679 * @syslog: include the "<4>" prefixes
3680 * @line: buffer to copy the line to
3681 * @size: maximum size of the buffer
3682 * @len: length of line placed into buffer
3684 * Start at the beginning of the kmsg buffer, with the oldest kmsg
3685 * record, and copy one record into the provided buffer.
3687 * Consecutive calls will return the next available record moving
3688 * towards the end of the buffer with the youngest messages.
3690 * A return value of FALSE indicates that there are no more records to
3693 bool kmsg_dump_get_line(struct kmsg_dump_iter *iter, bool syslog,
3694 char *line, size_t size, size_t *len)
3696 u64 min_seq = latched_seq_read_nolock(&clear_seq);
3697 struct printk_info info;
3698 unsigned int line_count;
3699 struct printk_record r;
3703 if (iter->cur_seq < min_seq)
3704 iter->cur_seq = min_seq;
3706 prb_rec_init_rd(&r, &info, line, size);
3708 /* Read text or count text lines? */
3710 if (!prb_read_valid(prb, iter->cur_seq, &r))
3712 l = record_print_text(&r, syslog, printk_time);
3714 if (!prb_read_valid_info(prb, iter->cur_seq,
3715 &info, &line_count)) {
3718 l = get_record_print_text_size(&info, line_count, syslog,
3723 iter->cur_seq = r.info->seq + 1;
3730 EXPORT_SYMBOL_GPL(kmsg_dump_get_line);
3733 * kmsg_dump_get_buffer - copy kmsg log lines
3734 * @iter: kmsg dump iterator
3735 * @syslog: include the "<4>" prefixes
3736 * @buf: buffer to copy the line to
3737 * @size: maximum size of the buffer
3738 * @len_out: length of line placed into buffer
3740 * Start at the end of the kmsg buffer and fill the provided buffer
3741 * with as many of the *youngest* kmsg records that fit into it.
3742 * If the buffer is large enough, all available kmsg records will be
3743 * copied with a single call.
3745 * Consecutive calls will fill the buffer with the next block of
3746 * available older records, not including the earlier retrieved ones.
3748 * A return value of FALSE indicates that there are no more records to
3751 bool kmsg_dump_get_buffer(struct kmsg_dump_iter *iter, bool syslog,
3752 char *buf, size_t size, size_t *len_out)
3754 u64 min_seq = latched_seq_read_nolock(&clear_seq);
3755 struct printk_info info;
3756 struct printk_record r;
3761 bool time = printk_time;
3766 if (iter->cur_seq < min_seq)
3767 iter->cur_seq = min_seq;
3769 if (prb_read_valid_info(prb, iter->cur_seq, &info, NULL)) {
3770 if (info.seq != iter->cur_seq) {
3771 /* messages are gone, move to first available one */
3772 iter->cur_seq = info.seq;
3777 if (iter->cur_seq >= iter->next_seq)
3781 * Find first record that fits, including all following records,
3782 * into the user-provided buffer for this dump. Pass in size-1
3783 * because this function (by way of record_print_text()) will
3784 * not write more than size-1 bytes of text into @buf.
3786 seq = find_first_fitting_seq(iter->cur_seq, iter->next_seq,
3787 size - 1, syslog, time);
3790 * Next kmsg_dump_get_buffer() invocation will dump block of
3791 * older records stored right before this one.
3795 prb_rec_init_rd(&r, &info, buf, size);
3798 prb_for_each_record(seq, prb, seq, &r) {
3799 if (r.info->seq >= iter->next_seq)
3802 len += record_print_text(&r, syslog, time);
3804 /* Adjust record to store to remaining buffer space. */
3805 prb_rec_init_rd(&r, &info, buf + len, size - len);
3808 iter->next_seq = next_seq;
3815 EXPORT_SYMBOL_GPL(kmsg_dump_get_buffer);
3818 * kmsg_dump_rewind - reset the iterator
3819 * @iter: kmsg dump iterator
3821 * Reset the dumper's iterator so that kmsg_dump_get_line() and
3822 * kmsg_dump_get_buffer() can be called again and used multiple
3823 * times within the same dumper.dump() callback.
3825 void kmsg_dump_rewind(struct kmsg_dump_iter *iter)
3827 iter->cur_seq = latched_seq_read_nolock(&clear_seq);
3828 iter->next_seq = prb_next_seq(prb);
3830 EXPORT_SYMBOL_GPL(kmsg_dump_rewind);
3835 static atomic_t printk_cpu_sync_owner = ATOMIC_INIT(-1);
3836 static atomic_t printk_cpu_sync_nested = ATOMIC_INIT(0);
3839 * __printk_cpu_sync_wait() - Busy wait until the printk cpu-reentrant
3840 * spinning lock is not owned by any CPU.
3842 * Context: Any context.
3844 void __printk_cpu_sync_wait(void)
3848 } while (atomic_read(&printk_cpu_sync_owner) != -1);
3850 EXPORT_SYMBOL(__printk_cpu_sync_wait);
3853 * __printk_cpu_sync_try_get() - Try to acquire the printk cpu-reentrant
3856 * If no processor has the lock, the calling processor takes the lock and
3857 * becomes the owner. If the calling processor is already the owner of the
3858 * lock, this function succeeds immediately.
3860 * Context: Any context. Expects interrupts to be disabled.
3861 * Return: 1 on success, otherwise 0.
3863 int __printk_cpu_sync_try_get(void)
3868 cpu = smp_processor_id();
3871 * Guarantee loads and stores from this CPU when it is the lock owner
3872 * are _not_ visible to the previous lock owner. This pairs with
3873 * __printk_cpu_sync_put:B.
3875 * Memory barrier involvement:
3877 * If __printk_cpu_sync_try_get:A reads from __printk_cpu_sync_put:B,
3878 * then __printk_cpu_sync_put:A can never read from
3879 * __printk_cpu_sync_try_get:B.
3883 * RELEASE from __printk_cpu_sync_put:A to __printk_cpu_sync_put:B
3884 * of the previous CPU
3886 * ACQUIRE from __printk_cpu_sync_try_get:A to
3887 * __printk_cpu_sync_try_get:B of this CPU
3889 old = atomic_cmpxchg_acquire(&printk_cpu_sync_owner, -1,
3890 cpu); /* LMM(__printk_cpu_sync_try_get:A) */
3893 * This CPU is now the owner and begins loading/storing
3894 * data: LMM(__printk_cpu_sync_try_get:B)
3898 } else if (old == cpu) {
3899 /* This CPU is already the owner. */
3900 atomic_inc(&printk_cpu_sync_nested);
3906 EXPORT_SYMBOL(__printk_cpu_sync_try_get);
3909 * __printk_cpu_sync_put() - Release the printk cpu-reentrant spinning lock.
3911 * The calling processor must be the owner of the lock.
3913 * Context: Any context. Expects interrupts to be disabled.
3915 void __printk_cpu_sync_put(void)
3917 if (atomic_read(&printk_cpu_sync_nested)) {
3918 atomic_dec(&printk_cpu_sync_nested);
3923 * This CPU is finished loading/storing data:
3924 * LMM(__printk_cpu_sync_put:A)
3928 * Guarantee loads and stores from this CPU when it was the
3929 * lock owner are visible to the next lock owner. This pairs
3930 * with __printk_cpu_sync_try_get:A.
3932 * Memory barrier involvement:
3934 * If __printk_cpu_sync_try_get:A reads from __printk_cpu_sync_put:B,
3935 * then __printk_cpu_sync_try_get:B reads from __printk_cpu_sync_put:A.
3939 * RELEASE from __printk_cpu_sync_put:A to __printk_cpu_sync_put:B
3942 * ACQUIRE from __printk_cpu_sync_try_get:A to
3943 * __printk_cpu_sync_try_get:B of the next CPU
3945 atomic_set_release(&printk_cpu_sync_owner,
3946 -1); /* LMM(__printk_cpu_sync_put:B) */
3948 EXPORT_SYMBOL(__printk_cpu_sync_put);
3949 #endif /* CONFIG_SMP */