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 #include <sys/types.h>
29 #include <sys/cpuset.h>
30 #include <sys/param.h>
31 #include <sys/socket.h>
51 #include "libpmcstat.h"
54 * Get PMC record by id, apply merge policy.
57 static struct pmcstat_pmcrecord *
58 pmcstat_lookup_pmcid(pmc_id_t pmcid, int pmcstat_mergepmc)
60 struct pmcstat_pmcrecord *pr;
62 LIST_FOREACH(pr, &pmcstat_pmcs, pr_next) {
63 if (pr->pr_pmcid == pmcid) {
74 * Add a {pmcid,name} mapping.
78 pmcstat_pmcid_add(pmc_id_t pmcid, pmcstat_interned_string ps,
79 struct pmcstat_args *args, struct pmc_plugins *plugins,
82 struct pmcstat_pmcrecord *pr, *prm;
84 /* Replace an existing name for the PMC. */
86 LIST_FOREACH(pr, &pmcstat_pmcs, pr_next)
87 if (pr->pr_pmcid == pmcid) {
90 } else if (pr->pr_pmcname == ps)
94 * Otherwise, allocate a new descriptor and call the
97 if ((pr = malloc(sizeof(*pr))) == NULL)
98 err(EX_OSERR, "ERROR: Cannot allocate pmc record");
100 pr->pr_pmcid = pmcid;
102 pr->pr_pmcin = (*pmcstat_npmcs)++;
104 pr->pr_dubious_frames = 0;
105 pr->pr_merge = prm == NULL ? pr : prm;
107 LIST_INSERT_HEAD(&pmcstat_pmcs, pr, pr_next);
109 if (plugins[args->pa_pplugin].pl_newpmc != NULL)
110 plugins[args->pa_pplugin].pl_newpmc(ps, pr);
111 if (plugins[args->pa_plugin].pl_newpmc != NULL)
112 plugins[args->pa_plugin].pl_newpmc(ps, pr);
116 * Unmap images in the range [start..end) associated with process
121 pmcstat_image_unmap(struct pmcstat_process *pp, uintfptr_t start,
124 struct pmcstat_pcmap *pcm, *pcmtmp, *pcmnew;
131 * - we could have the range completely in the middle of an
132 * existing pcmap; in this case we have to split the pcmap
133 * structure into two (i.e., generate a 'hole').
134 * - we could have the range covering multiple pcmaps; these
135 * will have to be removed.
136 * - we could have either 'start' or 'end' falling in the
137 * middle of a pcmap; in this case shorten the entry.
139 TAILQ_FOREACH_SAFE(pcm, &pp->pp_map, ppm_next, pcmtmp) {
140 assert(pcm->ppm_lowpc < pcm->ppm_highpc);
141 if (pcm->ppm_highpc <= start)
143 if (pcm->ppm_lowpc >= end)
145 if (pcm->ppm_lowpc >= start && pcm->ppm_highpc <= end) {
147 * The current pcmap is completely inside the
148 * unmapped range: remove it entirely.
150 TAILQ_REMOVE(&pp->pp_map, pcm, ppm_next);
152 } else if (pcm->ppm_lowpc < start && pcm->ppm_highpc > end) {
154 * Split this pcmap into two; curtail the
155 * current map to end at [start-1], and start
156 * the new one at [end].
158 if ((pcmnew = malloc(sizeof(*pcmnew))) == NULL)
160 "ERROR: Cannot split a map entry");
162 pcmnew->ppm_image = pcm->ppm_image;
164 pcmnew->ppm_lowpc = end;
165 pcmnew->ppm_highpc = pcm->ppm_highpc;
167 pcm->ppm_highpc = start;
169 TAILQ_INSERT_AFTER(&pp->pp_map, pcm, pcmnew, ppm_next);
172 } else if (pcm->ppm_lowpc < start && pcm->ppm_highpc <= end)
173 pcm->ppm_highpc = start;
174 else if (pcm->ppm_lowpc >= start && pcm->ppm_highpc > end)
175 pcm->ppm_lowpc = end;
182 * Convert a hwpmc(4) log to profile information. A system-wide
183 * callgraph is generated if FLAG_DO_CALLGRAPHS is set. gmon.out
184 * files usable by gprof(1) are created if FLAG_DO_GPROF is set.
187 pmcstat_analyze_log(struct pmcstat_args *args,
188 struct pmc_plugins *plugins,
189 struct pmcstat_stats *pmcstat_stats,
190 struct pmcstat_process *pmcstat_kernproc,
191 int pmcstat_mergepmc,
193 int *ps_samples_period)
195 uint32_t cpu, cpuflags;
197 struct pmcstat_image *image;
198 struct pmcstat_process *pp, *ppnew;
199 struct pmcstat_pcmap *ppm, *ppmtmp;
201 struct pmcstat_pmcrecord *pmcr;
202 pmcstat_interned_string image_path;
204 assert(args->pa_flags & FLAG_DO_ANALYSIS);
206 if (elf_version(EV_CURRENT) == EV_NONE)
207 err(EX_UNAVAILABLE, "Elf library initialization failed");
209 while (pmclog_read(args->pa_logparser, &ev) == 0) {
210 assert(ev.pl_state == PMCLOG_OK);
212 switch (ev.pl_type) {
213 case PMCLOG_TYPE_INITIALIZE:
214 if ((ev.pl_u.pl_i.pl_version & 0xFF000000) !=
215 PMC_VERSION_MAJOR << 24 && args->pa_verbosity > 0)
217 "WARNING: Log version 0x%x does not match compiled version 0x%x.",
218 ev.pl_u.pl_i.pl_version, PMC_VERSION_MAJOR);
221 case PMCLOG_TYPE_MAP_IN:
223 * Introduce an address range mapping for a
224 * userland process or the kernel (pid == -1).
226 * We always allocate a process descriptor so
227 * that subsequent samples seen for this
228 * address range are mapped to the current
229 * object being mapped in.
231 pid = ev.pl_u.pl_mi.pl_pid;
233 pp = pmcstat_kernproc;
235 pp = pmcstat_process_lookup(pid,
240 image_path = pmcstat_string_intern(ev.pl_u.pl_mi.
242 image = pmcstat_image_from_path(image_path, pid == -1,
244 if (image->pi_type == PMCSTAT_IMAGE_UNKNOWN)
245 pmcstat_image_determine_type(image, args);
246 if (image->pi_type != PMCSTAT_IMAGE_INDETERMINABLE)
247 pmcstat_image_link(pp, image,
248 ev.pl_u.pl_mi.pl_start);
251 case PMCLOG_TYPE_MAP_OUT:
253 * Remove an address map.
255 pid = ev.pl_u.pl_mo.pl_pid;
257 pp = pmcstat_kernproc;
259 pp = pmcstat_process_lookup(pid, 0);
261 if (pp == NULL) /* unknown process */
264 pmcstat_image_unmap(pp, ev.pl_u.pl_mo.pl_start,
265 ev.pl_u.pl_mo.pl_end);
268 case PMCLOG_TYPE_CALLCHAIN:
269 pmcstat_stats->ps_samples_total++;
270 *ps_samples_period += 1;
272 cpuflags = ev.pl_u.pl_cc.pl_cpuflags;
273 cpu = PMC_CALLCHAIN_CPUFLAGS_TO_CPU(cpuflags);
275 if ((args->pa_flags & FLAG_FILTER_THREAD_ID) &&
276 args->pa_tid != ev.pl_u.pl_cc.pl_tid) {
277 pmcstat_stats->ps_samples_skipped++;
280 /* Filter on the CPU id. */
281 if (!CPU_ISSET(cpu, &(args->pa_cpumask))) {
282 pmcstat_stats->ps_samples_skipped++;
286 pp = pmcstat_process_lookup(ev.pl_u.pl_cc.pl_pid,
289 /* Get PMC record. */
290 pmcr = pmcstat_lookup_pmcid(ev.pl_u.pl_cc.pl_pmcid, pmcstat_mergepmc);
291 assert(pmcr != NULL);
295 * Call the plugins processing
298 if (plugins[args->pa_pplugin].pl_process != NULL)
299 plugins[args->pa_pplugin].pl_process(
301 ev.pl_u.pl_cc.pl_npc,
303 PMC_CALLCHAIN_CPUFLAGS_TO_USERMODE(cpuflags),
305 plugins[args->pa_plugin].pl_process(
307 ev.pl_u.pl_cc.pl_npc,
309 PMC_CALLCHAIN_CPUFLAGS_TO_USERMODE(cpuflags),
313 case PMCLOG_TYPE_PMCALLOCATE:
315 * Record the association pmc id between this
318 pmcstat_pmcid_add(ev.pl_u.pl_a.pl_pmcid,
319 pmcstat_string_intern(ev.pl_u.pl_a.pl_evname),
320 args, plugins, pmcstat_npmcs);
323 case PMCLOG_TYPE_PMCALLOCATEDYN:
325 * Record the association pmc id between this
328 pmcstat_pmcid_add(ev.pl_u.pl_ad.pl_pmcid,
329 pmcstat_string_intern(ev.pl_u.pl_ad.pl_evname),
330 args, plugins, pmcstat_npmcs);
333 case PMCLOG_TYPE_PROCEXEC:
335 * Change the executable image associated with
338 pp = pmcstat_process_lookup(ev.pl_u.pl_x.pl_pid,
341 /* delete the current process map */
342 TAILQ_FOREACH_SAFE(ppm, &pp->pp_map, ppm_next, ppmtmp) {
343 TAILQ_REMOVE(&pp->pp_map, ppm, ppm_next);
348 * Associate this process image.
350 image_path = pmcstat_string_intern(
351 ev.pl_u.pl_x.pl_pathname);
352 assert(image_path != NULL);
353 pmcstat_process_exec(pp, image_path,
354 ev.pl_u.pl_x.pl_baseaddr, ev.pl_u.pl_x.pl_dynaddr,
355 args, plugins, pmcstat_stats);
358 case PMCLOG_TYPE_PROCEXIT:
361 * Due to the way the log is generated, the
362 * last few samples corresponding to a process
363 * may appear in the log after the process
364 * exit event is recorded. Thus we keep the
365 * process' descriptor and associated data
366 * structures around, but mark the process as
369 pp = pmcstat_process_lookup(ev.pl_u.pl_e.pl_pid, 0);
372 pp->pp_isactive = 0; /* mark as a zombie */
375 case PMCLOG_TYPE_SYSEXIT:
376 pp = pmcstat_process_lookup(ev.pl_u.pl_se.pl_pid, 0);
379 pp->pp_isactive = 0; /* make a zombie */
382 case PMCLOG_TYPE_PROCFORK:
385 * Allocate a process descriptor for the new
389 pmcstat_process_lookup(ev.pl_u.pl_f.pl_newpid,
393 * If we had been tracking the parent, clone
396 pp = pmcstat_process_lookup(ev.pl_u.pl_f.pl_oldpid, 0);
399 TAILQ_FOREACH(ppm, &pp->pp_map, ppm_next)
400 pmcstat_image_link(ppnew, ppm->ppm_image,
404 default: /* other types of entries are not relevant */
409 if (ev.pl_state == PMCLOG_EOF)
410 return (PMCSTAT_FINISHED);
411 else if (ev.pl_state == PMCLOG_REQUIRE_DATA)
412 return (PMCSTAT_RUNNING);
415 "ERROR: event parsing failed state: %d type: %d (record %jd, offset 0x%jx)",
416 ev.pl_state, ev.pl_type, (uintmax_t) ev.pl_count + 1, ev.pl_offset);
420 * Open a log file, for reading or writing.
422 * The function returns the fd of a successfully opened log or -1 in
427 pmcstat_open_log(const char *path, int mode)
431 const char *p, *errstr;
432 struct addrinfo hints, *res, *res0;
433 char hostname[MAXHOSTNAMELEN];
439 * If 'path' is "-" then open one of stdin or stdout depending
440 * on the value of 'mode'.
442 * If 'path' contains a ':' and does not start with a '/' or '.',
443 * and is being opened for writing, treat it as a "host:port"
444 * specification and open a network socket.
446 * Otherwise, treat 'path' as a file name and open that.
448 if (path[0] == '-' && path[1] == '\0')
449 fd = (mode == PMCSTAT_OPEN_FOR_READ) ? 0 : 1;
450 else if (path[0] != '/' &&
451 path[0] != '.' && strchr(path, ':') != NULL) {
453 p = strrchr(path, ':');
455 if (p == path || hlen >= sizeof(hostname)) {
456 errstr = strerror(EINVAL);
460 assert(hlen < sizeof(hostname));
461 (void) strncpy(hostname, path, hlen);
462 hostname[hlen] = '\0';
464 (void) memset(&hints, 0, sizeof(hints));
465 hints.ai_family = AF_UNSPEC;
466 hints.ai_socktype = SOCK_STREAM;
467 if ((error = getaddrinfo(hostname, p+1, &hints, &res0)) != 0) {
468 errstr = gai_strerror(error);
473 for (res = res0; res; res = res->ai_next) {
474 if ((fd = socket(res->ai_family, res->ai_socktype,
475 res->ai_protocol)) < 0) {
476 errstr = strerror(errno);
479 if (mode == PMCSTAT_OPEN_FOR_READ) {
480 if (bind(fd, res->ai_addr, res->ai_addrlen) < 0) {
481 errstr = strerror(errno);
487 cfd = accept(fd, NULL, NULL);
490 errstr = strerror(errno);
496 if (connect(fd, res->ai_addr, res->ai_addrlen) < 0) {
497 errstr = strerror(errno);
508 } else if ((fd = open(path, mode == PMCSTAT_OPEN_FOR_READ ?
509 O_RDONLY : (O_WRONLY|O_CREAT|O_TRUNC),
510 S_IRUSR|S_IWUSR|S_IRGRP|S_IROTH)) < 0)
511 errstr = strerror(errno);
515 errx(EX_OSERR, "ERROR: Cannot open \"%s\" for %s: %s.", path,
516 (mode == PMCSTAT_OPEN_FOR_READ ? "reading" : "writing"),
523 * Close a logfile, after first flushing all in-module queued data.
527 pmcstat_close_log(struct pmcstat_args *args)
529 /* If a local logfile is configured ask the kernel to stop
530 * and flush data. Kernel will close the file when data is flushed
531 * so keep the status to EXITING.
533 if (args->pa_logfd != -1) {
534 if (pmc_close_logfile() < 0)
535 err(EX_OSERR, "ERROR: logging failed");
538 return (args->pa_flags & FLAG_HAS_PIPE ? PMCSTAT_EXITING :
547 pmcstat_initialize_logging(struct pmcstat_process **pmcstat_kernproc,
548 struct pmcstat_args *args, struct pmc_plugins *plugins,
549 int *pmcstat_npmcs, int *pmcstat_mergepmc)
551 struct pmcstat_process *pmcstat_kp;
554 /* use a convenient format for 'ldd' output */
555 if (setenv("LD_TRACE_LOADED_OBJECTS_FMT1","%o \"%p\" %x\n",1) != 0)
556 err(EX_OSERR, "ERROR: Cannot setenv");
558 /* Initialize hash tables */
559 pmcstat_string_initialize();
560 for (i = 0; i < PMCSTAT_NHASH; i++) {
561 LIST_INIT(&pmcstat_image_hash[i]);
562 LIST_INIT(&pmcstat_process_hash[i]);
566 * Create a fake 'process' entry for the kernel with pid -1.
567 * hwpmc(4) will subsequently inform us about where the kernel
568 * and any loaded kernel modules are mapped.
570 if ((pmcstat_kp = pmcstat_process_lookup((pid_t) -1,
571 PMCSTAT_ALLOCATE)) == NULL)
572 err(EX_OSERR, "ERROR: Cannot initialize logging");
574 *pmcstat_kernproc = pmcstat_kp;
579 /* Merge PMC with same name. */
580 *pmcstat_mergepmc = args->pa_mergepmc;
586 if (plugins[args->pa_pplugin].pl_init != NULL)
587 plugins[args->pa_pplugin].pl_init();
588 if (plugins[args->pa_plugin].pl_init != NULL)
589 plugins[args->pa_plugin].pl_init();
597 pmcstat_shutdown_logging(struct pmcstat_args *args,
598 struct pmc_plugins *plugins,
599 struct pmcstat_stats *pmcstat_stats)
601 struct pmcstat_image *pi, *pitmp;
602 struct pmcstat_process *pp, *pptmp;
603 struct pmcstat_pcmap *ppm, *ppmtmp;
607 /* determine where to send the map file */
609 if (args->pa_mapfilename != NULL)
610 mf = (strcmp(args->pa_mapfilename, "-") == 0) ?
611 args->pa_printfile : fopen(args->pa_mapfilename, "w");
613 if (mf == NULL && args->pa_flags & FLAG_DO_GPROF &&
614 args->pa_verbosity >= 2)
615 mf = args->pa_printfile;
618 (void) fprintf(mf, "MAP:\n");
621 * Shutdown the plugins
624 if (plugins[args->pa_plugin].pl_shutdown != NULL)
625 plugins[args->pa_plugin].pl_shutdown(mf);
626 if (plugins[args->pa_pplugin].pl_shutdown != NULL)
627 plugins[args->pa_pplugin].pl_shutdown(mf);
629 for (i = 0; i < PMCSTAT_NHASH; i++) {
630 LIST_FOREACH_SAFE(pi, &pmcstat_image_hash[i], pi_next,
632 if (plugins[args->pa_plugin].pl_shutdownimage != NULL)
633 plugins[args->pa_plugin].pl_shutdownimage(pi);
634 if (plugins[args->pa_pplugin].pl_shutdownimage != NULL)
635 plugins[args->pa_pplugin].pl_shutdownimage(pi);
637 free(pi->pi_symbols);
638 if (pi->pi_addr2line != NULL)
639 pclose(pi->pi_addr2line);
640 LIST_REMOVE(pi, pi_next);
644 LIST_FOREACH_SAFE(pp, &pmcstat_process_hash[i], pp_next,
646 TAILQ_FOREACH_SAFE(ppm, &pp->pp_map, ppm_next, ppmtmp) {
647 TAILQ_REMOVE(&pp->pp_map, ppm, ppm_next);
650 LIST_REMOVE(pp, pp_next);
655 pmcstat_string_shutdown();
658 * Print errors unless -q was specified. Print all statistics
661 #define PRINT(N,V) do { \
662 if (pmcstat_stats->ps_##V || args->pa_verbosity >= 2) \
663 (void) fprintf(args->pa_printfile, " %-40s %d\n",\
664 N, pmcstat_stats->ps_##V); \
667 if (args->pa_verbosity >= 1 && (args->pa_flags & FLAG_DO_ANALYSIS)) {
668 (void) fprintf(args->pa_printfile, "CONVERSION STATISTICS:\n");
669 PRINT("#exec/a.out", exec_aout);
670 PRINT("#exec/elf", exec_elf);
671 PRINT("#exec/unknown", exec_indeterminable);
672 PRINT("#exec handling errors", exec_errors);
673 PRINT("#samples/total", samples_total);
674 PRINT("#samples/unclaimed", samples_unknown_offset);
675 PRINT("#samples/unknown-object", samples_indeterminable);
676 PRINT("#samples/unknown-function", samples_unknown_function);
677 PRINT("#callchain/dubious-frames", callchain_dubious_frames);