xref: /dragonfly/sys/kern/kern_ktr.c (revision cecb9aae)
1 /*
2  * Copyright (c) 2005 The DragonFly Project.  All rights reserved.
3  *
4  * This code is derived from software contributed to The DragonFly Project
5  * by Matthew Dillon <dillon@backplane.com>
6  *
7  * Redistribution and use in source and binary forms, with or without
8  * modification, are permitted provided that the following conditions
9  * are met:
10  *
11  * 1. Redistributions of source code must retain the above copyright
12  *    notice, this list of conditions and the following disclaimer.
13  * 2. Redistributions in binary form must reproduce the above copyright
14  *    notice, this list of conditions and the following disclaimer in
15  *    the documentation and/or other materials provided with the
16  *    distribution.
17  * 3. Neither the name of The DragonFly Project nor the names of its
18  *    contributors may be used to endorse or promote products derived
19  *    from this software without specific, prior written permission.
20  *
21  * THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS
22  * ``AS IS'' AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT
23  * LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS
24  * FOR A PARTICULAR PURPOSE ARE DISCLAIMED.  IN NO EVENT SHALL THE
25  * COPYRIGHT HOLDERS OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT,
26  * INCIDENTAL, SPECIAL, EXEMPLARY OR CONSEQUENTIAL DAMAGES (INCLUDING,
27  * BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES;
28  * LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED
29  * AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY,
30  * OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT
31  * OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
32  * SUCH DAMAGE.
33  */
34 /*
35  * The following copyright applies to the DDB command code:
36  *
37  * Copyright (c) 2000 John Baldwin <jhb@FreeBSD.org>
38  * All rights reserved.
39  *
40  * Redistribution and use in source and binary forms, with or without
41  * modification, are permitted provided that the following conditions
42  * are met:
43  * 1. Redistributions of source code must retain the above copyright
44  *    notice, this list of conditions and the following disclaimer.
45  * 2. Redistributions in binary form must reproduce the above copyright
46  *    notice, this list of conditions and the following disclaimer in the
47  *    documentation and/or other materials provided with the distribution.
48  * 3. Neither the name of the author nor the names of any co-contributors
49  *    may be used to endorse or promote products derived from this software
50  *    without specific prior written permission.
51  *
52  * THIS SOFTWARE IS PROVIDED BY THE AUTHOR AND CONTRIBUTORS ``AS IS'' AND
53  * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
54  * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
55  * ARE DISCLAIMED.  IN NO EVENT SHALL THE AUTHOR OR CONTRIBUTORS BE LIABLE
56  * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
57  * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
58  * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
59  * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
60  * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
61  * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
62  * SUCH DAMAGE.
63  */
64 
65 /*
66  * Kernel tracepoint facility.
67  */
68 
69 #include "opt_ddb.h"
70 #include "opt_ktr.h"
71 
72 #include <sys/param.h>
73 #include <sys/cons.h>
74 #include <sys/kernel.h>
75 #include <sys/libkern.h>
76 #include <sys/proc.h>
77 #include <sys/sysctl.h>
78 #include <sys/ktr.h>
79 #include <sys/systm.h>
80 #include <sys/time.h>
81 #include <sys/malloc.h>
82 #include <sys/spinlock.h>
83 #include <sys/thread2.h>
84 #include <sys/spinlock2.h>
85 #include <sys/ctype.h>
86 
87 #include <machine/cpu.h>
88 #include <machine/cpufunc.h>
89 #include <machine/specialreg.h>
90 #include <machine/md_var.h>
91 
92 #include <ddb/ddb.h>
93 
94 #ifndef KTR_ENTRIES
95 #define	KTR_ENTRIES		2048
96 #elif (KTR_ENTRIES & KTR_ENTRIES - 1)
97 #error KTR_ENTRIES must be a power of two
98 #endif
99 #define KTR_ENTRIES_MASK	(KTR_ENTRIES - 1)
100 
101 /*
102  * test logging support.  When ktr_testlogcnt is non-zero each synchronization
103  * interrupt will issue six back-to-back ktr logging messages on cpu 0
104  * so the user can determine KTR logging overheads.
105  */
106 #if !defined(KTR_TESTLOG)
107 #define KTR_TESTLOG	KTR_ALL
108 #endif
109 KTR_INFO_MASTER(testlog);
110 #if KTR_TESTLOG
111 KTR_INFO(KTR_TESTLOG, testlog, test1, 0, "test1 %d %d %d %d", int dummy1, int dummy2, int dummy3, int dummy4);
112 KTR_INFO(KTR_TESTLOG, testlog, test2, 1, "test2 %d %d %d %d", int dummy1, int dummy2, int dummy3, int dummy4);
113 KTR_INFO(KTR_TESTLOG, testlog, test3, 2, "test3 %d %d %d %d", int dummy1, int dummy2, int dummy3, int dummy4);
114 KTR_INFO(KTR_TESTLOG, testlog, test4, 3, "test4");
115 KTR_INFO(KTR_TESTLOG, testlog, test5, 4, "test5");
116 KTR_INFO(KTR_TESTLOG, testlog, test6, 5, "test6");
117 KTR_INFO(KTR_TESTLOG, testlog, pingpong, 6, "pingpong");
118 KTR_INFO(KTR_TESTLOG, testlog, pipeline, 7, "pipeline");
119 KTR_INFO(KTR_TESTLOG, testlog, crit_beg, 8, "crit_beg");
120 KTR_INFO(KTR_TESTLOG, testlog, crit_end, 9, "crit_end");
121 KTR_INFO(KTR_TESTLOG, testlog, spin_beg, 10, "spin_beg");
122 KTR_INFO(KTR_TESTLOG, testlog, spin_end, 11, "spin_end");
123 #define logtest(name)	KTR_LOG(testlog_ ## name, 0, 0, 0, 0)
124 #define logtest_noargs(name)	KTR_LOG(testlog_ ## name)
125 #endif
126 
127 MALLOC_DEFINE(M_KTR, "ktr", "ktr buffers");
128 
129 SYSCTL_NODE(_debug, OID_AUTO, ktr, CTLFLAG_RW, 0, "ktr");
130 
131 int		ktr_entries = KTR_ENTRIES;
132 SYSCTL_INT(_debug_ktr, OID_AUTO, entries, CTLFLAG_RD, &ktr_entries, 0,
133     "Size of the event buffer");
134 
135 int		ktr_version = KTR_VERSION;
136 SYSCTL_INT(_debug_ktr, OID_AUTO, version, CTLFLAG_RD, &ktr_version, 0, "");
137 
138 static int	ktr_stacktrace = 1;
139 SYSCTL_INT(_debug_ktr, OID_AUTO, stacktrace, CTLFLAG_RD, &ktr_stacktrace, 0, "");
140 
141 static int	ktr_resynchronize = 0;
142 SYSCTL_INT(_debug_ktr, OID_AUTO, resynchronize, CTLFLAG_RW,
143     &ktr_resynchronize, 0, "Resynchronize TSC 10 times a second");
144 
145 #if KTR_TESTLOG
146 static int	ktr_testlogcnt = 0;
147 SYSCTL_INT(_debug_ktr, OID_AUTO, testlogcnt, CTLFLAG_RW, &ktr_testlogcnt, 0, "");
148 static int	ktr_testipicnt = 0;
149 static int	ktr_testipicnt_remainder;
150 SYSCTL_INT(_debug_ktr, OID_AUTO, testipicnt, CTLFLAG_RW, &ktr_testipicnt, 0, "");
151 static int	ktr_testcritcnt = 0;
152 SYSCTL_INT(_debug_ktr, OID_AUTO, testcritcnt, CTLFLAG_RW, &ktr_testcritcnt, 0, "");
153 static int	ktr_testspincnt = 0;
154 SYSCTL_INT(_debug_ktr, OID_AUTO, testspincnt, CTLFLAG_RW, &ktr_testspincnt, 0, "");
155 #endif
156 
157 /*
158  * Give cpu0 a static buffer so the tracepoint facility can be used during
159  * early boot (note however that we still use a critical section, XXX).
160  */
161 static struct	ktr_entry ktr_buf0[KTR_ENTRIES];
162 
163 struct ktr_cpu ktr_cpu[MAXCPU] = {
164 	{ .core.ktr_buf = &ktr_buf0[0] }
165 };
166 
167 static int64_t	ktr_sync_tsc;
168 struct callout	ktr_resync_callout;
169 
170 #ifdef KTR_VERBOSE
171 int	ktr_verbose = KTR_VERBOSE;
172 TUNABLE_INT("debug.ktr.verbose", &ktr_verbose);
173 SYSCTL_INT(_debug_ktr, OID_AUTO, verbose, CTLFLAG_RW, &ktr_verbose, 0,
174     "Log events to the console as well");
175 #endif
176 
177 static void ktr_resync_callback(void *dummy __unused);
178 
179 extern int64_t tsc_offsets[];
180 
181 static void
182 ktr_sysinit(void *dummy)
183 {
184 	struct ktr_cpu_core *kcpu;
185 	int i;
186 
187 	for(i = 1; i < ncpus; ++i) {
188 		kcpu = &ktr_cpu[i].core;
189 		kcpu->ktr_buf = kmalloc(KTR_ENTRIES * sizeof(struct ktr_entry),
190 					M_KTR, M_WAITOK | M_ZERO);
191 	}
192 	callout_init_mp(&ktr_resync_callout);
193 	callout_reset(&ktr_resync_callout, hz / 10, ktr_resync_callback, NULL);
194 }
195 SYSINIT(ktr_sysinit, SI_BOOT2_KLD, SI_ORDER_ANY, ktr_sysinit, NULL);
196 
197 /*
198  * Try to resynchronize the TSC's for all cpus.  This is really, really nasty.
199  * We have to send an IPIQ message to all remote cpus, wait until they
200  * get into their IPIQ processing code loop, then do an even stricter hard
201  * loop to get the cpus as close to synchronized as we can to get the most
202  * accurate reading.
203  *
204  * This callback occurs on cpu0.
205  */
206 #if KTR_TESTLOG
207 static void ktr_pingpong_remote(void *dummy);
208 static void ktr_pipeline_remote(void *dummy);
209 #endif
210 
211 #ifdef _RDTSC_SUPPORTED_
212 
213 static void ktr_resync_remote(void *dummy);
214 
215 /*
216  * We use a callout callback instead of a systimer because we cannot afford
217  * to preempt anyone to do this, or we might deadlock a spin-lock or
218  * serializer between two cpus.
219  */
220 static
221 void
222 ktr_resync_callback(void *dummy __unused)
223 {
224 	struct lwkt_cpusync cs;
225 #if KTR_TESTLOG
226 	int count;
227 #endif
228 
229 	KKASSERT(mycpu->gd_cpuid == 0);
230 
231 #if KTR_TESTLOG
232 	/*
233 	 * Test logging
234 	 */
235 	if (ktr_testlogcnt) {
236 		--ktr_testlogcnt;
237 		cpu_disable_intr();
238 		logtest(test1);
239 		logtest(test2);
240 		logtest(test3);
241 		logtest_noargs(test4);
242 		logtest_noargs(test5);
243 		logtest_noargs(test6);
244 		cpu_enable_intr();
245 	}
246 
247 	/*
248 	 * Test IPI messaging
249 	 */
250 	if (ktr_testipicnt && ktr_testipicnt_remainder == 0 && ncpus > 1) {
251 		ktr_testipicnt_remainder = ktr_testipicnt;
252 		ktr_testipicnt = 0;
253 		lwkt_send_ipiq_bycpu(1, ktr_pingpong_remote, NULL);
254 	}
255 
256 	/*
257 	 * Test critical sections
258 	 */
259 	if (ktr_testcritcnt) {
260 		crit_enter();
261 		crit_exit();
262 		logtest_noargs(crit_beg);
263 		for (count = ktr_testcritcnt; count; --count) {
264 			crit_enter();
265 			crit_exit();
266 		}
267 		logtest_noargs(crit_end);
268 		ktr_testcritcnt = 0;
269 	}
270 
271 	/*
272 	 * Test spinlock sections
273 	 */
274 	if (ktr_testspincnt) {
275 		struct spinlock spin;
276 
277 		spin_init(&spin);
278 		spin_lock(&spin);
279 		spin_unlock(&spin);
280 		logtest_noargs(spin_beg);
281 		for (count = ktr_testspincnt; count; --count) {
282 			spin_lock(&spin);
283 			spin_unlock(&spin);
284 		}
285 		logtest_noargs(spin_end);
286 		ktr_testspincnt = 0;
287 	}
288 #endif
289 
290 	/*
291 	 * Resynchronize the TSC
292 	 */
293 	if (ktr_resynchronize == 0)
294 		goto done;
295 	if ((cpu_feature & CPUID_TSC) == 0)
296 		return;
297 
298 	crit_enter();
299 	lwkt_cpusync_init(&cs, smp_active_mask, ktr_resync_remote,
300 			  (void *)(intptr_t)mycpu->gd_cpuid);
301 	lwkt_cpusync_interlock(&cs);
302 	ktr_sync_tsc = rdtsc();
303 	lwkt_cpusync_deinterlock(&cs);
304 	crit_exit();
305 done:
306 	callout_reset(&ktr_resync_callout, hz / 10, ktr_resync_callback, NULL);
307 }
308 
309 /*
310  * The remote-end of the KTR synchronization protocol runs on all cpus.
311  * The one we run on the controlling cpu updates its tsc continuously
312  * until the others have finished syncing (theoretically), but we don't
313  * loop forever.
314  *
315  * This is a bit ad-hoc but we need to avoid livelocking inside an IPI
316  * callback.  rdtsc() is a synchronizing instruction (I think).
317  */
318 static void
319 ktr_resync_remote(void *arg)
320 {
321 	globaldata_t gd = mycpu;
322 	int64_t delta;
323 	int i;
324 
325 	if (gd->gd_cpuid == (int)(intptr_t)arg) {
326 		for (i = 0; i < 2000; ++i)
327 			ktr_sync_tsc = rdtsc();
328 	} else {
329 		delta = rdtsc() - ktr_sync_tsc;
330 		if (tsc_offsets[gd->gd_cpuid] == 0)
331 			tsc_offsets[gd->gd_cpuid] = delta;
332 		tsc_offsets[gd->gd_cpuid] =
333 			(tsc_offsets[gd->gd_cpuid] * 7 + delta) / 8;
334 	}
335 }
336 
337 #if KTR_TESTLOG
338 
339 static
340 void
341 ktr_pingpong_remote(void *dummy __unused)
342 {
343 	int other_cpu;
344 
345 	logtest_noargs(pingpong);
346 	other_cpu = 1 - mycpu->gd_cpuid;
347 	if (ktr_testipicnt_remainder) {
348 		--ktr_testipicnt_remainder;
349 		lwkt_send_ipiq_bycpu(other_cpu, ktr_pingpong_remote, NULL);
350 	} else {
351 		lwkt_send_ipiq_bycpu(other_cpu, ktr_pipeline_remote, NULL);
352 		lwkt_send_ipiq_bycpu(other_cpu, ktr_pipeline_remote, NULL);
353 		lwkt_send_ipiq_bycpu(other_cpu, ktr_pipeline_remote, NULL);
354 		lwkt_send_ipiq_bycpu(other_cpu, ktr_pipeline_remote, NULL);
355 		lwkt_send_ipiq_bycpu(other_cpu, ktr_pipeline_remote, NULL);
356 	}
357 }
358 
359 static
360 void
361 ktr_pipeline_remote(void *dummy __unused)
362 {
363 	logtest_noargs(pipeline);
364 }
365 
366 #endif
367 
368 #else	/* !_RDTSC_SUPPORTED_ */
369 
370 /*
371  * The resync callback for UP doesn't do anything other then run the test
372  * log messages.  If test logging is not enabled, don't bother resetting
373  * the callout.
374  */
375 static
376 void
377 ktr_resync_callback(void *dummy __unused)
378 {
379 #if KTR_TESTLOG
380 	/*
381 	 * Test logging
382 	 */
383 	if (ktr_testlogcnt) {
384 		--ktr_testlogcnt;
385 		cpu_disable_intr();
386 		logtest(test1);
387 		logtest(test2);
388 		logtest(test3);
389 		logtest_noargs(test4);
390 		logtest_noargs(test5);
391 		logtest_noargs(test6);
392 		cpu_enable_intr();
393 	}
394 	callout_reset(&ktr_resync_callout, hz / 10, ktr_resync_callback, NULL);
395 #endif
396 }
397 
398 #endif
399 
400 /*
401  * Setup the next empty slot and return it to the caller to store the data
402  * directly.
403  */
404 struct ktr_entry *
405 ktr_begin_write_entry(struct ktr_info *info, const char *file, int line)
406 {
407 	struct ktr_cpu_core *kcpu;
408 	struct ktr_entry *entry;
409 	int cpu;
410 
411 	cpu = mycpu->gd_cpuid;
412 	kcpu = &ktr_cpu[cpu].core;
413 	if (panicstr)			/* stop logging during panic */
414 		return NULL;
415 	if (kcpu->ktr_buf == NULL)	/* too early in boot */
416 		return NULL;
417 
418 	crit_enter();
419 	entry = kcpu->ktr_buf + (kcpu->ktr_idx & KTR_ENTRIES_MASK);
420 	++kcpu->ktr_idx;
421 #ifdef _RDTSC_SUPPORTED_
422 	if (cpu_feature & CPUID_TSC) {
423 		entry->ktr_timestamp = rdtsc() - tsc_offsets[cpu];
424 	} else
425 #endif
426 	{
427 		entry->ktr_timestamp = get_approximate_time_t();
428 	}
429 	entry->ktr_info = info;
430 	entry->ktr_file = file;
431 	entry->ktr_line = line;
432 	crit_exit();
433 	return entry;
434 }
435 
436 int
437 ktr_finish_write_entry(struct ktr_info *info, struct ktr_entry *entry)
438 {
439 	if (ktr_stacktrace)
440 		cpu_ktr_caller(entry);
441 #ifdef KTR_VERBOSE
442 	if (ktr_verbose && info->kf_format) {
443 		kprintf("cpu%d ", mycpu->gd_cpuid);
444 		if (ktr_verbose > 1) {
445 			kprintf("%s.%d\t", entry->ktr_file, entry->ktr_line);
446 		}
447 		return !0;
448 	}
449 #endif
450 	return 0;
451 }
452 
453 #ifdef DDB
454 
455 #define	NUM_LINES_PER_PAGE	19
456 
457 struct tstate {
458 	int	cur;
459 	int	first;
460 };
461 
462 static	int db_ktr_verbose;
463 static	int db_mach_vtrace(int cpu, struct ktr_entry *kp, int idx);
464 
465 DB_SHOW_COMMAND(ktr, db_ktr_all)
466 {
467 	struct ktr_cpu_core *kcpu;
468 	int a_flag = 0;
469 	int c;
470 	int nl = 0;
471 	int i;
472 	struct tstate tstate[MAXCPU];
473 	int printcpu = -1;
474 
475 	for(i = 0; i < ncpus; i++) {
476 		kcpu = &ktr_cpu[i].core;
477 		tstate[i].first = -1;
478 		tstate[i].cur = (kcpu->ktr_idx - 1) & KTR_ENTRIES_MASK;
479 	}
480 	db_ktr_verbose = 0;
481 	while ((c = *(modif++)) != '\0') {
482 		if (c == 'v') {
483 			db_ktr_verbose = 1;
484 		}
485 		else if (c == 'a') {
486 			a_flag = 1;
487 		}
488 		else if (c == 'c') {
489 			printcpu = 0;
490 			while ((c = *(modif++)) != '\0') {
491 				if (isdigit(c)) {
492 					printcpu *= 10;
493 					printcpu += c - '0';
494 				}
495 				else {
496 					modif++;
497 					break;
498 				}
499 			}
500 			modif--;
501 		}
502 	}
503 	if (printcpu > ncpus - 1) {
504 		db_printf("Invalid cpu number\n");
505 		return;
506 	}
507 	/*
508 	 * Lopp throug all the buffers and print the content of them, sorted
509 	 * by the timestamp.
510 	 */
511 	while (1) {
512 		int counter;
513 		u_int64_t highest_ts;
514 		int highest_cpu;
515 		struct ktr_entry *kp;
516 
517 		if (a_flag == 1 && cncheckc() != -1)
518 			return;
519 		highest_ts = 0;
520 		highest_cpu = -1;
521 		/*
522 		 * Find the lowest timestamp
523 		 */
524 		for (i = 0, counter = 0; i < ncpus; i++) {
525 			kcpu = &ktr_cpu[i].core;
526 			if (kcpu->ktr_buf == NULL)
527 				continue;
528 			if (printcpu != -1 && printcpu != i)
529 				continue;
530 			if (tstate[i].cur == -1) {
531 				counter++;
532 				if (counter == ncpus) {
533 					db_printf("--- End of trace buffer ---\n");
534 					return;
535 				}
536 				continue;
537 			}
538 			if (kcpu->ktr_buf[tstate[i].cur].ktr_timestamp > highest_ts) {
539 				highest_ts = kcpu->ktr_buf[tstate[i].cur].ktr_timestamp;
540 				highest_cpu = i;
541 			}
542 		}
543 		if (highest_cpu < 0) {
544 			db_printf("no KTR data available\n");
545 			break;
546 		}
547 		i = highest_cpu;
548 		kcpu = &ktr_cpu[i].core;
549 		kp = &kcpu->ktr_buf[tstate[i].cur];
550 		if (tstate[i].first == -1)
551 			tstate[i].first = tstate[i].cur;
552 		if (--tstate[i].cur < 0)
553 			tstate[i].cur = KTR_ENTRIES - 1;
554 		if (tstate[i].first == tstate[i].cur) {
555 			db_mach_vtrace(i, kp, tstate[i].cur + 1);
556 			tstate[i].cur = -1;
557 			continue;
558 		}
559 		if (kcpu->ktr_buf[tstate[i].cur].ktr_info == NULL)
560 			tstate[i].cur = -1;
561 		if (db_more(&nl) == -1)
562 			break;
563 		if (db_mach_vtrace(i, kp, tstate[i].cur + 1) == 0)
564 			tstate[i].cur = -1;
565 	}
566 }
567 
568 static int
569 db_mach_vtrace(int cpu, struct ktr_entry *kp, int idx)
570 {
571 	if (kp->ktr_info == NULL)
572 		return(0);
573 	db_printf("cpu%d ", cpu);
574 	db_printf("%d: ", idx);
575 	if (db_ktr_verbose) {
576 		db_printf("%10.10lld %s.%d\t", (long long)kp->ktr_timestamp,
577 		    kp->ktr_file, kp->ktr_line);
578 	}
579 	db_printf("%s\t", kp->ktr_info->kf_name);
580 	db_printf("from(%p,%p) ", kp->ktr_caller1, kp->ktr_caller2);
581 #ifdef __i386__
582 	if (kp->ktr_info->kf_format)
583 		db_vprintf(kp->ktr_info->kf_format, (__va_list)kp->ktr_data);
584 #endif
585 	db_printf("\n");
586 
587 	return(1);
588 }
589 
590 #endif	/* DDB */
591