2 * Copyright (c) 2003-2008 Joseph Koshy
5 * Redistribution and use in source and binary forms, with or without
6 * modification, are permitted provided that the following conditions
8 * 1. Redistributions of source code must retain the above copyright
9 * notice, this list of conditions and the following disclaimer.
10 * 2. Redistributions in binary form must reproduce the above copyright
11 * notice, this list of conditions and the following disclaimer in the
12 * documentation and/or other materials provided with the distribution.
14 * THIS SOFTWARE IS PROVIDED BY THE AUTHOR AND CONTRIBUTORS ``AS IS'' AND
15 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
16 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
17 * ARE DISCLAIMED. IN NO EVENT SHALL THE AUTHOR OR CONTRIBUTORS BE LIABLE
18 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
19 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
20 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
21 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
22 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
23 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
27 #include <sys/cdefs.h>
28 __FBSDID("$FreeBSD$");
30 #include <sys/types.h>
31 #include <sys/cpuset.h>
32 #include <sys/param.h>
33 #include <sys/socket.h>
53 #include "libpmcstat.h"
56 * Get PMC record by id, apply merge policy.
59 static struct pmcstat_pmcrecord *
60 pmcstat_lookup_pmcid(pmc_id_t pmcid, int pmcstat_mergepmc)
62 struct pmcstat_pmcrecord *pr;
64 LIST_FOREACH(pr, &pmcstat_pmcs, pr_next) {
65 if (pr->pr_pmcid == pmcid) {
76 * Add a {pmcid,name} mapping.
80 pmcstat_pmcid_add(pmc_id_t pmcid, pmcstat_interned_string ps,
81 struct pmcstat_args *args, struct pmc_plugins *plugins,
84 struct pmcstat_pmcrecord *pr, *prm;
86 /* Replace an existing name for the PMC. */
88 LIST_FOREACH(pr, &pmcstat_pmcs, pr_next)
89 if (pr->pr_pmcid == pmcid) {
92 } else if (pr->pr_pmcname == ps)
96 * Otherwise, allocate a new descriptor and call the
99 if ((pr = malloc(sizeof(*pr))) == NULL)
100 err(EX_OSERR, "ERROR: Cannot allocate pmc record");
102 pr->pr_pmcid = pmcid;
104 pr->pr_pmcin = (*pmcstat_npmcs)++;
106 pr->pr_dubious_frames = 0;
107 pr->pr_merge = prm == NULL ? pr : prm;
109 LIST_INSERT_HEAD(&pmcstat_pmcs, pr, pr_next);
111 if (plugins[args->pa_pplugin].pl_newpmc != NULL)
112 plugins[args->pa_pplugin].pl_newpmc(ps, pr);
113 if (plugins[args->pa_plugin].pl_newpmc != NULL)
114 plugins[args->pa_plugin].pl_newpmc(ps, pr);
118 * Unmap images in the range [start..end) associated with process
123 pmcstat_image_unmap(struct pmcstat_process *pp, uintfptr_t start,
126 struct pmcstat_pcmap *pcm, *pcmtmp, *pcmnew;
133 * - we could have the range completely in the middle of an
134 * existing pcmap; in this case we have to split the pcmap
135 * structure into two (i.e., generate a 'hole').
136 * - we could have the range covering multiple pcmaps; these
137 * will have to be removed.
138 * - we could have either 'start' or 'end' falling in the
139 * middle of a pcmap; in this case shorten the entry.
141 TAILQ_FOREACH_SAFE(pcm, &pp->pp_map, ppm_next, pcmtmp) {
142 assert(pcm->ppm_lowpc < pcm->ppm_highpc);
143 if (pcm->ppm_highpc <= start)
145 if (pcm->ppm_lowpc >= end)
147 if (pcm->ppm_lowpc >= start && pcm->ppm_highpc <= end) {
149 * The current pcmap is completely inside the
150 * unmapped range: remove it entirely.
152 TAILQ_REMOVE(&pp->pp_map, pcm, ppm_next);
154 } else if (pcm->ppm_lowpc < start && pcm->ppm_highpc > end) {
156 * Split this pcmap into two; curtail the
157 * current map to end at [start-1], and start
158 * the new one at [end].
160 if ((pcmnew = malloc(sizeof(*pcmnew))) == NULL)
162 "ERROR: Cannot split a map entry");
164 pcmnew->ppm_image = pcm->ppm_image;
166 pcmnew->ppm_lowpc = end;
167 pcmnew->ppm_highpc = pcm->ppm_highpc;
169 pcm->ppm_highpc = start;
171 TAILQ_INSERT_AFTER(&pp->pp_map, pcm, pcmnew, ppm_next);
174 } else if (pcm->ppm_lowpc < start && pcm->ppm_highpc <= end)
175 pcm->ppm_highpc = start;
176 else if (pcm->ppm_lowpc >= start && pcm->ppm_highpc > end)
177 pcm->ppm_lowpc = end;
184 * Convert a hwpmc(4) log to profile information. A system-wide
185 * callgraph is generated if FLAG_DO_CALLGRAPHS is set. gmon.out
186 * files usable by gprof(1) are created if FLAG_DO_GPROF is set.
189 pmcstat_analyze_log(struct pmcstat_args *args,
190 struct pmc_plugins *plugins,
191 struct pmcstat_stats *pmcstat_stats,
192 struct pmcstat_process *pmcstat_kernproc,
193 int pmcstat_mergepmc,
195 int *ps_samples_period)
197 uint32_t cpu, cpuflags;
199 struct pmcstat_image *image;
200 struct pmcstat_process *pp, *ppnew;
201 struct pmcstat_pcmap *ppm, *ppmtmp;
203 struct pmcstat_pmcrecord *pmcr;
204 pmcstat_interned_string image_path;
206 assert(args->pa_flags & FLAG_DO_ANALYSIS);
208 if (elf_version(EV_CURRENT) == EV_NONE)
209 err(EX_UNAVAILABLE, "Elf library initialization failed");
211 while (pmclog_read(args->pa_logparser, &ev) == 0) {
212 assert(ev.pl_state == PMCLOG_OK);
214 switch (ev.pl_type) {
215 case PMCLOG_TYPE_INITIALIZE:
216 if ((ev.pl_u.pl_i.pl_version & 0xFF000000) !=
217 PMC_VERSION_MAJOR << 24 && args->pa_verbosity > 0)
219 "WARNING: Log version 0x%x does not match compiled version 0x%x.",
220 ev.pl_u.pl_i.pl_version, PMC_VERSION_MAJOR);
223 case PMCLOG_TYPE_MAP_IN:
225 * Introduce an address range mapping for a
226 * userland process or the kernel (pid == -1).
228 * We always allocate a process descriptor so
229 * that subsequent samples seen for this
230 * address range are mapped to the current
231 * object being mapped in.
233 pid = ev.pl_u.pl_mi.pl_pid;
235 pp = pmcstat_kernproc;
237 pp = pmcstat_process_lookup(pid,
242 image_path = pmcstat_string_intern(ev.pl_u.pl_mi.
244 image = pmcstat_image_from_path(image_path, pid == -1,
246 if (image->pi_type == PMCSTAT_IMAGE_UNKNOWN)
247 pmcstat_image_determine_type(image, args);
248 if (image->pi_type != PMCSTAT_IMAGE_INDETERMINABLE)
249 pmcstat_image_link(pp, image,
250 ev.pl_u.pl_mi.pl_start);
253 case PMCLOG_TYPE_MAP_OUT:
255 * Remove an address map.
257 pid = ev.pl_u.pl_mo.pl_pid;
259 pp = pmcstat_kernproc;
261 pp = pmcstat_process_lookup(pid, 0);
263 if (pp == NULL) /* unknown process */
266 pmcstat_image_unmap(pp, ev.pl_u.pl_mo.pl_start,
267 ev.pl_u.pl_mo.pl_end);
270 case PMCLOG_TYPE_CALLCHAIN:
271 pmcstat_stats->ps_samples_total++;
272 *ps_samples_period += 1;
274 cpuflags = ev.pl_u.pl_cc.pl_cpuflags;
275 cpu = PMC_CALLCHAIN_CPUFLAGS_TO_CPU(cpuflags);
277 if ((args->pa_flags & FLAG_FILTER_THREAD_ID) &&
278 args->pa_tid != ev.pl_u.pl_cc.pl_tid) {
279 pmcstat_stats->ps_samples_skipped++;
282 /* Filter on the CPU id. */
283 if (!CPU_ISSET(cpu, &(args->pa_cpumask))) {
284 pmcstat_stats->ps_samples_skipped++;
288 pp = pmcstat_process_lookup(ev.pl_u.pl_cc.pl_pid,
291 /* Get PMC record. */
292 pmcr = pmcstat_lookup_pmcid(ev.pl_u.pl_cc.pl_pmcid, pmcstat_mergepmc);
293 assert(pmcr != NULL);
297 * Call the plugins processing
300 if (plugins[args->pa_pplugin].pl_process != NULL)
301 plugins[args->pa_pplugin].pl_process(
303 ev.pl_u.pl_cc.pl_npc,
305 PMC_CALLCHAIN_CPUFLAGS_TO_USERMODE(cpuflags),
307 plugins[args->pa_plugin].pl_process(
309 ev.pl_u.pl_cc.pl_npc,
311 PMC_CALLCHAIN_CPUFLAGS_TO_USERMODE(cpuflags),
315 case PMCLOG_TYPE_PMCALLOCATE:
317 * Record the association pmc id between this
320 pmcstat_pmcid_add(ev.pl_u.pl_a.pl_pmcid,
321 pmcstat_string_intern(ev.pl_u.pl_a.pl_evname),
322 args, plugins, pmcstat_npmcs);
325 case PMCLOG_TYPE_PMCALLOCATEDYN:
327 * Record the association pmc id between this
330 pmcstat_pmcid_add(ev.pl_u.pl_ad.pl_pmcid,
331 pmcstat_string_intern(ev.pl_u.pl_ad.pl_evname),
332 args, plugins, pmcstat_npmcs);
335 case PMCLOG_TYPE_PROCEXEC:
337 * Change the executable image associated with
340 pp = pmcstat_process_lookup(ev.pl_u.pl_x.pl_pid,
343 /* delete the current process map */
344 TAILQ_FOREACH_SAFE(ppm, &pp->pp_map, ppm_next, ppmtmp) {
345 TAILQ_REMOVE(&pp->pp_map, ppm, ppm_next);
350 * Associate this process image.
352 image_path = pmcstat_string_intern(
353 ev.pl_u.pl_x.pl_pathname);
354 assert(image_path != NULL);
355 pmcstat_process_exec(pp, image_path,
356 ev.pl_u.pl_x.pl_entryaddr, args,
357 plugins, pmcstat_stats);
360 case PMCLOG_TYPE_PROCEXIT:
363 * Due to the way the log is generated, the
364 * last few samples corresponding to a process
365 * may appear in the log after the process
366 * exit event is recorded. Thus we keep the
367 * process' descriptor and associated data
368 * structures around, but mark the process as
371 pp = pmcstat_process_lookup(ev.pl_u.pl_e.pl_pid, 0);
374 pp->pp_isactive = 0; /* mark as a zombie */
377 case PMCLOG_TYPE_SYSEXIT:
378 pp = pmcstat_process_lookup(ev.pl_u.pl_se.pl_pid, 0);
381 pp->pp_isactive = 0; /* make a zombie */
384 case PMCLOG_TYPE_PROCFORK:
387 * Allocate a process descriptor for the new
391 pmcstat_process_lookup(ev.pl_u.pl_f.pl_newpid,
395 * If we had been tracking the parent, clone
398 pp = pmcstat_process_lookup(ev.pl_u.pl_f.pl_oldpid, 0);
401 TAILQ_FOREACH(ppm, &pp->pp_map, ppm_next)
402 pmcstat_image_link(ppnew, ppm->ppm_image,
406 default: /* other types of entries are not relevant */
411 if (ev.pl_state == PMCLOG_EOF)
412 return (PMCSTAT_FINISHED);
413 else if (ev.pl_state == PMCLOG_REQUIRE_DATA)
414 return (PMCSTAT_RUNNING);
417 "ERROR: event parsing failed state: %d type: %d (record %jd, offset 0x%jx)",
418 ev.pl_state, ev.pl_type, (uintmax_t) ev.pl_count + 1, ev.pl_offset);
422 * Open a log file, for reading or writing.
424 * The function returns the fd of a successfully opened log or -1 in
429 pmcstat_open_log(const char *path, int mode)
433 const char *p, *errstr;
434 struct addrinfo hints, *res, *res0;
435 char hostname[MAXHOSTNAMELEN];
441 * If 'path' is "-" then open one of stdin or stdout depending
442 * on the value of 'mode'.
444 * If 'path' contains a ':' and does not start with a '/' or '.',
445 * and is being opened for writing, treat it as a "host:port"
446 * specification and open a network socket.
448 * Otherwise, treat 'path' as a file name and open that.
450 if (path[0] == '-' && path[1] == '\0')
451 fd = (mode == PMCSTAT_OPEN_FOR_READ) ? 0 : 1;
452 else if (path[0] != '/' &&
453 path[0] != '.' && strchr(path, ':') != NULL) {
455 p = strrchr(path, ':');
457 if (p == path || hlen >= sizeof(hostname)) {
458 errstr = strerror(EINVAL);
462 assert(hlen < sizeof(hostname));
463 (void) strncpy(hostname, path, hlen);
464 hostname[hlen] = '\0';
466 (void) memset(&hints, 0, sizeof(hints));
467 hints.ai_family = AF_UNSPEC;
468 hints.ai_socktype = SOCK_STREAM;
469 if ((error = getaddrinfo(hostname, p+1, &hints, &res0)) != 0) {
470 errstr = gai_strerror(error);
475 for (res = res0; res; res = res->ai_next) {
476 if ((fd = socket(res->ai_family, res->ai_socktype,
477 res->ai_protocol)) < 0) {
478 errstr = strerror(errno);
481 if (mode == PMCSTAT_OPEN_FOR_READ) {
482 if (bind(fd, res->ai_addr, res->ai_addrlen) < 0) {
483 errstr = strerror(errno);
489 cfd = accept(fd, NULL, NULL);
492 errstr = strerror(errno);
498 if (connect(fd, res->ai_addr, res->ai_addrlen) < 0) {
499 errstr = strerror(errno);
510 } else if ((fd = open(path, mode == PMCSTAT_OPEN_FOR_READ ?
511 O_RDONLY : (O_WRONLY|O_CREAT|O_TRUNC),
512 S_IRUSR|S_IWUSR|S_IRGRP|S_IROTH)) < 0)
513 errstr = strerror(errno);
517 errx(EX_OSERR, "ERROR: Cannot open \"%s\" for %s: %s.", path,
518 (mode == PMCSTAT_OPEN_FOR_READ ? "reading" : "writing"),
525 * Close a logfile, after first flushing all in-module queued data.
529 pmcstat_close_log(struct pmcstat_args *args)
531 /* If a local logfile is configured ask the kernel to stop
532 * and flush data. Kernel will close the file when data is flushed
533 * so keep the status to EXITING.
535 if (args->pa_logfd != -1) {
536 if (pmc_close_logfile() < 0)
537 err(EX_OSERR, "ERROR: logging failed");
540 return (args->pa_flags & FLAG_HAS_PIPE ? PMCSTAT_EXITING :
549 pmcstat_initialize_logging(struct pmcstat_process **pmcstat_kernproc,
550 struct pmcstat_args *args, struct pmc_plugins *plugins,
551 int *pmcstat_npmcs, int *pmcstat_mergepmc)
553 struct pmcstat_process *pmcstat_kp;
556 /* use a convenient format for 'ldd' output */
557 if (setenv("LD_TRACE_LOADED_OBJECTS_FMT1","%o \"%p\" %x\n",1) != 0)
558 err(EX_OSERR, "ERROR: Cannot setenv");
560 /* Initialize hash tables */
561 pmcstat_string_initialize();
562 for (i = 0; i < PMCSTAT_NHASH; i++) {
563 LIST_INIT(&pmcstat_image_hash[i]);
564 LIST_INIT(&pmcstat_process_hash[i]);
568 * Create a fake 'process' entry for the kernel with pid -1.
569 * hwpmc(4) will subsequently inform us about where the kernel
570 * and any loaded kernel modules are mapped.
572 if ((pmcstat_kp = pmcstat_process_lookup((pid_t) -1,
573 PMCSTAT_ALLOCATE)) == NULL)
574 err(EX_OSERR, "ERROR: Cannot initialize logging");
576 *pmcstat_kernproc = pmcstat_kp;
581 /* Merge PMC with same name. */
582 *pmcstat_mergepmc = args->pa_mergepmc;
588 if (plugins[args->pa_pplugin].pl_init != NULL)
589 plugins[args->pa_pplugin].pl_init();
590 if (plugins[args->pa_plugin].pl_init != NULL)
591 plugins[args->pa_plugin].pl_init();
599 pmcstat_shutdown_logging(struct pmcstat_args *args,
600 struct pmc_plugins *plugins,
601 struct pmcstat_stats *pmcstat_stats)
603 struct pmcstat_image *pi, *pitmp;
604 struct pmcstat_process *pp, *pptmp;
605 struct pmcstat_pcmap *ppm, *ppmtmp;
609 /* determine where to send the map file */
611 if (args->pa_mapfilename != NULL)
612 mf = (strcmp(args->pa_mapfilename, "-") == 0) ?
613 args->pa_printfile : fopen(args->pa_mapfilename, "w");
615 if (mf == NULL && args->pa_flags & FLAG_DO_GPROF &&
616 args->pa_verbosity >= 2)
617 mf = args->pa_printfile;
620 (void) fprintf(mf, "MAP:\n");
623 * Shutdown the plugins
626 if (plugins[args->pa_plugin].pl_shutdown != NULL)
627 plugins[args->pa_plugin].pl_shutdown(mf);
628 if (plugins[args->pa_pplugin].pl_shutdown != NULL)
629 plugins[args->pa_pplugin].pl_shutdown(mf);
631 for (i = 0; i < PMCSTAT_NHASH; i++) {
632 LIST_FOREACH_SAFE(pi, &pmcstat_image_hash[i], pi_next,
634 if (plugins[args->pa_plugin].pl_shutdownimage != NULL)
635 plugins[args->pa_plugin].pl_shutdownimage(pi);
636 if (plugins[args->pa_pplugin].pl_shutdownimage != NULL)
637 plugins[args->pa_pplugin].pl_shutdownimage(pi);
639 free(pi->pi_symbols);
640 if (pi->pi_addr2line != NULL)
641 pclose(pi->pi_addr2line);
642 LIST_REMOVE(pi, pi_next);
646 LIST_FOREACH_SAFE(pp, &pmcstat_process_hash[i], pp_next,
648 TAILQ_FOREACH_SAFE(ppm, &pp->pp_map, ppm_next, ppmtmp) {
649 TAILQ_REMOVE(&pp->pp_map, ppm, ppm_next);
652 LIST_REMOVE(pp, pp_next);
657 pmcstat_string_shutdown();
660 * Print errors unless -q was specified. Print all statistics
663 #define PRINT(N,V) do { \
664 if (pmcstat_stats->ps_##V || args->pa_verbosity >= 2) \
665 (void) fprintf(args->pa_printfile, " %-40s %d\n",\
666 N, pmcstat_stats->ps_##V); \
669 if (args->pa_verbosity >= 1 && (args->pa_flags & FLAG_DO_ANALYSIS)) {
670 (void) fprintf(args->pa_printfile, "CONVERSION STATISTICS:\n");
671 PRINT("#exec/a.out", exec_aout);
672 PRINT("#exec/elf", exec_elf);
673 PRINT("#exec/unknown", exec_indeterminable);
674 PRINT("#exec handling errors", exec_errors);
675 PRINT("#samples/total", samples_total);
676 PRINT("#samples/unclaimed", samples_unknown_offset);
677 PRINT("#samples/unknown-object", samples_indeterminable);
678 PRINT("#samples/unknown-function", samples_unknown_function);
679 PRINT("#callchain/dubious-frames", callchain_dubious_frames);