1/*-
2 * SPDX-License-Identifier: BSD-2-Clause
3 *
4 * Copyright (c) 2000 John Baldwin <jhb@FreeBSD.org>
5 *
6 * Redistribution and use in source and binary forms, with or without
7 * modification, are permitted provided that the following conditions
8 * are met:
9 * 1. Redistributions of source code must retain the above copyright
10 *    notice, this list of conditions and the following disclaimer.
11 * 2. Redistributions in binary form must reproduce the above copyright
12 *    notice, this list of conditions and the following disclaimer in the
13 *    documentation and/or other materials provided with the distribution.
14 *
15 * THIS SOFTWARE IS PROVIDED BY THE AUTHOR AND CONTRIBUTORS ``AS IS'' AND
16 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
17 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
18 * ARE DISCLAIMED.  IN NO EVENT SHALL THE AUTHOR OR CONTRIBUTORS BE LIABLE
19 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
20 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
21 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
22 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
23 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
24 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
25 * SUCH DAMAGE.
26 */
27
28/*
29 * This module holds the global variables used by KTR and the ktr_tracepoint()
30 * function that does the actual tracing.
31 */
32
33#include <sys/cdefs.h>
34#include "opt_ddb.h"
35#include "opt_ktr.h"
36#include "opt_alq.h"
37
38#include <sys/param.h>
39#include <sys/queue.h>
40#include <sys/alq.h>
41#include <sys/cons.h>
42#include <sys/cpuset.h>
43#include <sys/kdb.h>
44#include <sys/kernel.h>
45#include <sys/ktr.h>
46#include <sys/libkern.h>
47#include <sys/lock.h>
48#include <sys/malloc.h>
49#include <sys/mutex.h>
50#include <sys/proc.h>
51#include <sys/smp.h>
52#include <sys/sysctl.h>
53#include <sys/systm.h>
54#include <sys/time.h>
55
56#include <machine/cpu.h>
57
58#ifdef DDB
59#include <ddb/ddb.h>
60#include <ddb/db_output.h>
61#endif
62
63#ifndef KTR_BOOT_ENTRIES
64#define	KTR_BOOT_ENTRIES	1024
65#endif
66
67#ifndef KTR_ENTRIES
68#define	KTR_ENTRIES	1024
69#endif
70
71/* Limit the allocations to something manageable. */
72#define	KTR_ENTRIES_MAX	(8 * 1024 * 1024)
73
74#ifndef KTR_MASK
75#define	KTR_MASK	(0)
76#endif
77
78#ifndef KTR_CPUMASK
79#define	KTR_CPUMASK	CPUSET_FSET
80#endif
81
82#ifndef KTR_TIME
83#define	KTR_TIME	get_cyclecount()
84#endif
85
86#ifndef KTR_CPU
87#define	KTR_CPU		PCPU_GET(cpuid)
88#endif
89
90static MALLOC_DEFINE(M_KTR, "KTR", "KTR");
91
92FEATURE(ktr, "Kernel support for KTR kernel tracing facility");
93
94volatile int	ktr_idx = 0;
95uint64_t ktr_mask = KTR_MASK;
96uint64_t ktr_compile = KTR_COMPILE;
97int	ktr_entries = KTR_BOOT_ENTRIES;
98int	ktr_version = KTR_VERSION;
99struct	ktr_entry ktr_buf_init[KTR_BOOT_ENTRIES];
100struct	ktr_entry *ktr_buf = ktr_buf_init;
101cpuset_t ktr_cpumask = CPUSET_T_INITIALIZER(KTR_CPUMASK);
102
103static SYSCTL_NODE(_debug, OID_AUTO, ktr, CTLFLAG_RD | CTLFLAG_MPSAFE, 0,
104    "KTR options");
105
106SYSCTL_INT(_debug_ktr, OID_AUTO, version, CTLFLAG_RD,
107    &ktr_version, 0, "Version of the KTR interface");
108
109SYSCTL_UQUAD(_debug_ktr, OID_AUTO, compile, CTLFLAG_RD,
110    &ktr_compile, 0, "Bitmask of KTR event classes compiled into the kernel");
111
112static int
113sysctl_debug_ktr_cpumask(SYSCTL_HANDLER_ARGS)
114{
115	char lktr_cpumask_str[CPUSETBUFSIZ];
116	cpuset_t imask;
117	int error;
118
119	cpusetobj_strprint(lktr_cpumask_str, &ktr_cpumask);
120	error = sysctl_handle_string(oidp, lktr_cpumask_str,
121	    sizeof(lktr_cpumask_str), req);
122	if (error != 0 || req->newptr == NULL)
123		return (error);
124	if (cpusetobj_strscan(&imask, lktr_cpumask_str) == -1)
125		return (EINVAL);
126	CPU_COPY(&imask, &ktr_cpumask);
127
128	return (error);
129}
130SYSCTL_PROC(_debug_ktr, OID_AUTO, cpumask,
131    CTLFLAG_RWTUN | CTLFLAG_MPSAFE | CTLTYPE_STRING, NULL, 0,
132    sysctl_debug_ktr_cpumask, "S",
133    "Bitmask of CPUs on which KTR logging is enabled");
134
135static int
136sysctl_debug_ktr_clear(SYSCTL_HANDLER_ARGS)
137{
138	int clear, error;
139
140	clear = 0;
141	error = sysctl_handle_int(oidp, &clear, 0, req);
142	if (error || !req->newptr)
143		return (error);
144
145	if (clear) {
146		bzero(ktr_buf, sizeof(*ktr_buf) * ktr_entries);
147		ktr_idx = 0;
148	}
149
150	return (error);
151}
152SYSCTL_PROC(_debug_ktr, OID_AUTO, clear,
153    CTLTYPE_INT | CTLFLAG_RW | CTLFLAG_NEEDGIANT, 0, 0,
154    sysctl_debug_ktr_clear, "I",
155    "Clear KTR Buffer");
156
157/*
158 * This is a sysctl proc so that it is serialized as !MPSAFE along with
159 * the other ktr sysctl procs.
160 */
161static int
162sysctl_debug_ktr_mask(SYSCTL_HANDLER_ARGS)
163{
164	uint64_t mask;
165	int error;
166
167	mask = ktr_mask;
168	error = sysctl_handle_64(oidp, &mask, 0, req);
169	if (error || !req->newptr)
170		return (error);
171	ktr_mask = mask;
172	return (error);
173}
174
175SYSCTL_PROC(_debug_ktr, OID_AUTO, mask,
176    CTLTYPE_U64 | CTLFLAG_RWTUN | CTLFLAG_NEEDGIANT,
177    0, 0, sysctl_debug_ktr_mask, "QU",
178    "Bitmask of KTR event classes for which logging is enabled");
179
180#if KTR_ENTRIES > KTR_BOOT_ENTRIES
181/*
182 * A simplified version of sysctl_debug_ktr_entries.
183 * No need to care about SMP, scheduling, etc.
184 */
185static void
186ktr_entries_initializer(void *dummy __unused)
187{
188	uint64_t mask;
189
190	/* Temporarily disable ktr in case malloc() is being traced. */
191	mask = ktr_mask;
192	ktr_mask = 0;
193	ktr_buf = malloc(sizeof(*ktr_buf) * KTR_ENTRIES, M_KTR,
194	    M_WAITOK | M_ZERO);
195	memcpy(ktr_buf, ktr_buf_init + ktr_idx,
196	    (KTR_BOOT_ENTRIES - ktr_idx) * sizeof(*ktr_buf));
197	if (ktr_idx != 0) {
198		memcpy(ktr_buf + KTR_BOOT_ENTRIES - ktr_idx, ktr_buf_init,
199		    ktr_idx * sizeof(*ktr_buf));
200		ktr_idx = KTR_BOOT_ENTRIES;
201	}
202	ktr_entries = KTR_ENTRIES;
203	ktr_mask = mask;
204}
205SYSINIT(ktr_entries_initializer, SI_SUB_KMEM, SI_ORDER_ANY,
206    ktr_entries_initializer, NULL);
207#endif
208
209static int
210sysctl_debug_ktr_entries(SYSCTL_HANDLER_ARGS)
211{
212	uint64_t mask;
213	int entries, error;
214	struct ktr_entry *buf, *oldbuf;
215
216	entries = ktr_entries;
217	error = sysctl_handle_int(oidp, &entries, 0, req);
218	if (error || !req->newptr)
219		return (error);
220	if (entries > KTR_ENTRIES_MAX)
221		return (ERANGE);
222	/* Disable ktr temporarily. */
223	mask = ktr_mask;
224	ktr_mask = 0;
225	/* Wait for threads to go idle. */
226	if ((error = quiesce_all_cpus("ktrent", PCATCH)) != 0) {
227		ktr_mask = mask;
228		return (error);
229	}
230	if (ktr_buf != ktr_buf_init)
231		oldbuf = ktr_buf;
232	else
233		oldbuf = NULL;
234	/* Allocate a new buffer. */
235	buf = malloc(sizeof(*buf) * entries, M_KTR, M_WAITOK | M_ZERO);
236	/* Install the new buffer and restart ktr. */
237	ktr_buf = buf;
238	ktr_entries = entries;
239	ktr_idx = 0;
240	ktr_mask = mask;
241	if (oldbuf != NULL)
242		free(oldbuf, M_KTR);
243
244	return (error);
245}
246
247SYSCTL_PROC(_debug_ktr, OID_AUTO, entries,
248    CTLTYPE_INT | CTLFLAG_RW | CTLFLAG_NEEDGIANT,
249    0, 0, sysctl_debug_ktr_entries, "I",
250    "Number of entries in the KTR buffer");
251
252#ifdef KTR_VERBOSE
253int	ktr_verbose = KTR_VERBOSE;
254TUNABLE_INT("debug.ktr.verbose", &ktr_verbose);
255SYSCTL_INT(_debug_ktr, OID_AUTO, verbose, CTLFLAG_RW, &ktr_verbose, 0, "");
256#endif
257
258#ifdef KTR_ALQ
259struct alq *ktr_alq;
260char	ktr_alq_file[MAXPATHLEN] = "/tmp/ktr.out";
261int	ktr_alq_cnt = 0;
262int	ktr_alq_depth = KTR_ENTRIES;
263int	ktr_alq_enabled = 0;
264int	ktr_alq_failed = 0;
265int	ktr_alq_max = 0;
266
267SYSCTL_INT(_debug_ktr, OID_AUTO, alq_max, CTLFLAG_RW, &ktr_alq_max, 0,
268    "Maximum number of entries to write");
269SYSCTL_INT(_debug_ktr, OID_AUTO, alq_cnt, CTLFLAG_RD, &ktr_alq_cnt, 0,
270    "Current number of written entries");
271SYSCTL_INT(_debug_ktr, OID_AUTO, alq_failed, CTLFLAG_RD, &ktr_alq_failed, 0,
272    "Number of times we overran the buffer");
273SYSCTL_INT(_debug_ktr, OID_AUTO, alq_depth, CTLFLAG_RW, &ktr_alq_depth, 0,
274    "Number of items in the write buffer");
275SYSCTL_STRING(_debug_ktr, OID_AUTO, alq_file, CTLFLAG_RW, ktr_alq_file,
276    sizeof(ktr_alq_file), "KTR logging file");
277
278static int
279sysctl_debug_ktr_alq_enable(SYSCTL_HANDLER_ARGS)
280{
281	int error;
282	int enable;
283
284	enable = ktr_alq_enabled;
285
286	error = sysctl_handle_int(oidp, &enable, 0, req);
287	if (error || !req->newptr)
288		return (error);
289
290	if (enable) {
291		if (ktr_alq_enabled)
292			return (0);
293		error = alq_open(&ktr_alq, (const char *)ktr_alq_file,
294		    req->td->td_ucred, ALQ_DEFAULT_CMODE,
295		    sizeof(struct ktr_entry), ktr_alq_depth);
296		if (error == 0) {
297			ktr_alq_cnt = 0;
298			ktr_alq_failed = 0;
299			ktr_alq_enabled = 1;
300		}
301	} else {
302		if (ktr_alq_enabled == 0)
303			return (0);
304		ktr_alq_enabled = 0;
305		alq_close(ktr_alq);
306		ktr_alq = NULL;
307	}
308
309	return (error);
310}
311SYSCTL_PROC(_debug_ktr, OID_AUTO, alq_enable,
312    CTLTYPE_INT | CTLFLAG_RW | CTLFLAG_NEEDGIANT, 0, 0,
313    sysctl_debug_ktr_alq_enable, "I",
314    "Enable KTR logging");
315#endif
316
317void
318ktr_tracepoint(uint64_t mask, const char *file, int line, const char *format,
319    u_long arg1, u_long arg2, u_long arg3, u_long arg4, u_long arg5,
320    u_long arg6)
321{
322	struct ktr_entry *entry;
323#ifdef KTR_ALQ
324	struct ale *ale = NULL;
325#endif
326	int newindex, saveindex;
327#if defined(KTR_VERBOSE) || defined(KTR_ALQ)
328	struct thread *td;
329#endif
330	int cpu;
331
332	if (KERNEL_PANICKED() || kdb_active)
333		return;
334	if ((ktr_mask & mask) == 0 || ktr_buf == NULL)
335		return;
336	cpu = KTR_CPU;
337	if (!CPU_ISSET(cpu, &ktr_cpumask))
338		return;
339#if defined(KTR_VERBOSE) || defined(KTR_ALQ)
340	td = curthread;
341	if (td->td_pflags & TDP_INKTR)
342		return;
343	td->td_pflags |= TDP_INKTR;
344#endif
345#ifdef KTR_ALQ
346	if (ktr_alq_enabled) {
347		if (td->td_critnest == 0 &&
348		    (TD_IS_IDLETHREAD(td)) == 0 &&
349		    td != ald_thread) {
350			if (ktr_alq_max && ktr_alq_cnt > ktr_alq_max)
351				goto done;
352			if ((ale = alq_get(ktr_alq, ALQ_NOWAIT)) == NULL) {
353				ktr_alq_failed++;
354				goto done;
355			}
356			ktr_alq_cnt++;
357			entry = (struct ktr_entry *)ale->ae_data;
358		} else {
359			goto done;
360		}
361	} else
362#endif
363	{
364		do {
365			saveindex = ktr_idx;
366			newindex = (saveindex + 1) % ktr_entries;
367		} while (atomic_cmpset_rel_int(&ktr_idx, saveindex, newindex) == 0);
368		entry = &ktr_buf[saveindex];
369	}
370	entry->ktr_timestamp = KTR_TIME;
371	entry->ktr_cpu = cpu;
372	entry->ktr_thread = curthread;
373	if (file != NULL)
374		while (strncmp(file, "../", 3) == 0)
375			file += 3;
376	entry->ktr_file = file;
377	entry->ktr_line = line;
378#ifdef KTR_VERBOSE
379	if (ktr_verbose) {
380#ifdef SMP
381		printf("cpu%d ", cpu);
382#endif
383		if (ktr_verbose > 1) {
384			printf("%s.%d\t", entry->ktr_file,
385			    entry->ktr_line);
386		}
387		printf(format, arg1, arg2, arg3, arg4, arg5, arg6);
388		printf("\n");
389	}
390#endif
391	entry->ktr_desc = format;
392	entry->ktr_parms[0] = arg1;
393	entry->ktr_parms[1] = arg2;
394	entry->ktr_parms[2] = arg3;
395	entry->ktr_parms[3] = arg4;
396	entry->ktr_parms[4] = arg5;
397	entry->ktr_parms[5] = arg6;
398#ifdef KTR_ALQ
399	if (ktr_alq_enabled && ale)
400		alq_post(ktr_alq, ale);
401done:
402#endif
403#if defined(KTR_VERBOSE) || defined(KTR_ALQ)
404	td->td_pflags &= ~TDP_INKTR;
405#endif
406}
407
408#ifdef DDB
409
410struct tstate {
411	int	cur;
412	int	first;
413};
414static	struct tstate tstate;
415static	int db_ktr_verbose;
416static	int db_mach_vtrace(void);
417
418DB_SHOW_COMMAND_FLAGS(ktr, db_ktr_all, DB_CMD_MEMSAFE)
419{
420
421	tstate.cur = (ktr_idx - 1) % ktr_entries;
422	tstate.first = -1;
423	db_ktr_verbose = 0;
424	db_ktr_verbose |= (strchr(modif, 'v') != NULL) ? 2 : 0;
425	db_ktr_verbose |= (strchr(modif, 'V') != NULL) ? 1 : 0; /* just timestamp please */
426	if (strchr(modif, 'a') != NULL) {
427		db_disable_pager();
428		while (cncheckc() == -1)
429			if (db_mach_vtrace() == 0)
430				break;
431	} else {
432		while (!db_pager_quit)
433			if (db_mach_vtrace() == 0)
434				break;
435	}
436}
437
438static int
439db_mach_vtrace(void)
440{
441	struct ktr_entry	*kp;
442
443	if (tstate.cur == tstate.first || ktr_buf == NULL) {
444		db_printf("--- End of trace buffer ---\n");
445		return (0);
446	}
447	kp = &ktr_buf[tstate.cur];
448
449	/* Skip over unused entries. */
450	if (kp->ktr_desc == NULL) {
451		db_printf("--- End of trace buffer ---\n");
452		return (0);
453	}
454	db_printf("%d (%p", tstate.cur, kp->ktr_thread);
455#ifdef SMP
456	db_printf(":cpu%d", kp->ktr_cpu);
457#endif
458	db_printf(")");
459	if (db_ktr_verbose >= 1) {
460		db_printf(" %10.10lld", (long long)kp->ktr_timestamp);
461	}
462	if (db_ktr_verbose >= 2) {
463		db_printf(" %s.%d", kp->ktr_file, kp->ktr_line);
464	}
465	db_printf(": ");
466	db_printf(kp->ktr_desc, kp->ktr_parms[0], kp->ktr_parms[1],
467	    kp->ktr_parms[2], kp->ktr_parms[3], kp->ktr_parms[4],
468	    kp->ktr_parms[5]);
469	db_printf("\n");
470
471	if (tstate.first == -1)
472		tstate.first = tstate.cur;
473
474	if (--tstate.cur < 0)
475		tstate.cur = ktr_entries - 1;
476
477	return (1);
478}
479
480#endif	/* DDB */
481