/* $NetBSD: tprof.c,v 1.18.2.2 2023/06/21 22:34:51 martin Exp $ */ /* * Copyright (c) 2018 The NetBSD Foundation, Inc. * All rights reserved. * * This code is derived from software contributed to The NetBSD Foundation * by Maxime Villard. * * Redistribution and use in source and binary forms, with or without * modification, are permitted provided that the following conditions * are met: * 1. Redistributions of source code must retain the above copyright * notice, this list of conditions and the following disclaimer. * 2. Redistributions in binary form must reproduce the above copyright * notice, this list of conditions and the following disclaimer in the * documentation and/or other materials provided with the distribution. * * THIS SOFTWARE IS PROVIDED BY THE NETBSD FOUNDATION, INC. AND CONTRIBUTORS * ``AS IS'' AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED * TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR * PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE FOUNDATION OR CONTRIBUTORS * BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR * CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF * SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS * INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN * CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) * ARISING IN ANY WAY OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE * POSSIBILITY OF SUCH DAMAGE. */ /* * Copyright (c)2008 YAMAMOTO Takashi, * All rights reserved. * * Redistribution and use in source and binary forms, with or without * modification, are permitted provided that the following conditions * are met: * 1. Redistributions of source code must retain the above copyright * notice, this list of conditions and the following disclaimer. * 2. Redistributions in binary form must reproduce the above copyright * notice, this list of conditions and the following disclaimer in the * documentation and/or other materials provided with the distribution. * * THIS SOFTWARE IS PROVIDED BY THE AUTHOR AND CONTRIBUTORS ``AS IS'' AND * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE * ARE DISCLAIMED. IN NO EVENT SHALL THE AUTHOR OR CONTRIBUTORS BE LIABLE * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF * SUCH DAMAGE. */ #include #ifndef lint __RCSID("$NetBSD: tprof.c,v 1.18.2.2 2023/06/21 22:34:51 martin Exp $"); #endif /* not lint */ #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include "tprof.h" #define _PATH_TPROF "/dev/tprof" struct tprof_info tprof_info; u_int ncounters; int devfd; int outfd; int ncpu; u_int nevent; double interval = 0xffffffff; /* XXX */ const char *eventname[TPROF_MAXCOUNTERS]; u_int eventnamewidth[TPROF_MAXCOUNTERS]; #define COUNTER_COLUMNS_WIDTH 11 static void tprof_list(int, char **); static void tprof_monitor_common(bool, int, char **) __dead; static void tprof_monitor(int, char **) __dead; static void tprof_count(int, char **) __dead; static struct cmdtab { const char *label; bool takesargs; bool argsoptional; void (*func)(int, char **); } const tprof_cmdtab[] = { { "list", false, false, tprof_list }, { "monitor", true, false, tprof_monitor }, { "count", true, false, tprof_count }, { "analyze", true, true, tprof_analyze }, { "top", true, true, tprof_top }, { NULL, false, false, NULL }, }; __dead static void usage(void) { fprintf(stderr, "%s op [arguments]\n", getprogname()); fprintf(stderr, "\n"); fprintf(stderr, "\tlist\n"); fprintf(stderr, "\t\tList the available events.\n"); fprintf(stderr, "\tmonitor -e name[:option] [-e ...] [-o outfile]" " command\n"); fprintf(stderr, "\t\tMonitor the event 'name' with option 'option'\n" "\t\tcounted during the execution of 'command'.\n"); fprintf(stderr, "\tcount -e name[:option] [-e ...] [-i interval]" " command\n"); fprintf(stderr, "\t\tSame as monitor, but does not profile," " only outputs a counter.\n"); fprintf(stderr, "\tanalyze [-CkLPs] [-p pid] file\n"); fprintf(stderr, "\t\tAnalyze the samples of the file 'file'.\n"); fprintf(stderr, "\ttop [-e name [-e ...]] [-i interval] [-acu]\n"); fprintf(stderr, "\t\tDisplay profiling results in real-time.\n"); exit(EXIT_FAILURE); } static int getncpu(void) { size_t size; int mib[2]; mib[0] = CTL_HW; mib[1] = HW_NCPU; size = sizeof(ncpu); if (sysctl(mib, 2, &ncpu, &size, NULL, 0) == -1) ncpu = 1; return ncpu; } static void * process_samples(void *dummy) { for (;;) { char buf[4096]; const char *cp; ssize_t ssz; ssz = read(devfd, buf, sizeof(buf)); if (ssz == -1) { err(EXIT_FAILURE, "read"); } if (ssz == 0) { break; } cp = buf; while (ssz) { ssize_t wsz; wsz = write(outfd, cp, ssz); if (wsz == -1) { err(EXIT_FAILURE, "write"); } ssz -= wsz; cp += wsz; } } return NULL; } static void show_counters(void) { unsigned int i; int n, ret; fprintf(stderr, " "); for (i = 0; i < nevent; i++) fprintf(stderr, " %*s", eventnamewidth[i], eventname[i]); fprintf(stderr, "\n"); for (n = 0; n < ncpu; n++) { tprof_counts_t counts; memset(&counts, 0, sizeof(counts)); counts.c_cpu = n; ret = ioctl(devfd, TPROF_IOC_GETCOUNTS, &counts); if (ret == -1) err(EXIT_FAILURE, "TPROF_IOC_GETCOUNTS"); fprintf(stderr, "CPU%-3d", n); for (i = 0; i < nevent; i++) { fprintf(stderr, " %*"PRIu64, eventnamewidth[i], counts.c_count[i]); } fprintf(stderr, "\n"); } } /* XXX: avoid mixing with the output of the child process SIGINFO handler... */ static void output_delay(void) { struct timespec delay_ts; delay_ts.tv_sec = 0; delay_ts.tv_nsec = 100000000; nanosleep(&delay_ts, NULL); } static void siginfo_nothing(int signo) { __nothing; } static void siginfo_showcount(int signo) { output_delay(); show_counters(); } static void * process_stat(void *arg) { unsigned int *done = arg; double ival, fval; struct timespec ts; ival = floor(interval); fval = (1000000000 * (interval - ival)); ts.tv_sec = ival; ts.tv_nsec = fval; while (atomic_add_int_nv(done, 0) == 0) { show_counters(); nanosleep(&ts, NULL); if (errno == EINTR) /* interrupted by SIGINFO? */ output_delay(); } return NULL; } static void tprof_list(int argc, char **argv) { const char *defaultevent = tprof_cycle_event_name(); printf("%u events can be counted at the same time.\n", ncounters); if (defaultevent != NULL) printf("The default counter for monitor and top command is " "\"%s\".\n", defaultevent); tprof_event_list(); } int tprof_parse_event(tprof_param_t *param, const char *str, uint32_t flags, const char **eventnamep, char **errmsgp) { double d; uint64_t n; int error = 0; char *p, *event = NULL, *opt = NULL, *scale = NULL; bool allow_option, allow_scale; static char errmsgbuf[128]; allow_option = flags & TPROF_PARSE_EVENT_F_ALLOWOPTION; allow_scale = flags & TPROF_PARSE_EVENT_F_ALLOWSCALE; p = estrdup(str); event = p; if (allow_option) { opt = strchr(p, ':'); if (opt != NULL) { *opt++ = '\0'; p = opt; } } if (allow_scale) { scale = strchr(p, ','); if (scale != NULL) *scale++ = '\0'; } tprof_event_lookup(event, param); if (opt != NULL) { while (*opt != '\0') { switch (*opt) { case 'u': param->p_flags |= TPROF_PARAM_USER; break; case 'k': param->p_flags |= TPROF_PARAM_KERN; break; default: error = -1; snprintf(errmsgbuf, sizeof(errmsgbuf), "invalid option: '%c'", *opt); goto done; } opt++; } } else if (allow_option) { param->p_flags |= TPROF_PARAM_USER; param->p_flags |= TPROF_PARAM_KERN; } if (scale != NULL) { if (*scale == '=') { scale++; n = strtoull(scale, &p, 0); if (*p != '\0') { error = -1; } else { param->p_value2 = n; param->p_flags |= TPROF_PARAM_VALUE2_TRIGGERCOUNT; } } else { if (strncasecmp("0x", scale, 2) == 0) d = strtol(scale, &p, 0); else d = strtod(scale, &p); if (*p != '\0' || d <= 0) { error = -1; } else { param->p_value2 = 0x100000000ULL / d; param->p_flags |= TPROF_PARAM_VALUE2_SCALE; } } if (error != 0) { snprintf(errmsgbuf, sizeof(errmsgbuf), "invalid scale: %s", scale); goto done; } } done: if (eventnamep != NULL) *eventnamep = event; if (error != 0 && errmsgp != NULL) *errmsgp = errmsgbuf; return error; } const char * tprof_cycle_event_name(void) { const char *cycleevent; switch (tprof_info.ti_ident) { case TPROF_IDENT_INTEL_GENERIC: cycleevent = "unhalted-core-cycles"; break; case TPROF_IDENT_AMD_GENERIC: cycleevent = "LsNotHaltedCyc"; break; case TPROF_IDENT_ARMV8_GENERIC: case TPROF_IDENT_ARMV7_GENERIC: cycleevent = "CPU_CYCLES"; break; default: cycleevent = NULL; break; } return cycleevent; } static void tprof_monitor_common(bool do_profile, int argc, char **argv) { const char *outfile = "tprof.out"; struct tprof_stat ts; tprof_param_t params[TPROF_MAXCOUNTERS]; pid_t pid; pthread_t pt; int ret, ch, i; char *p, *errmsg; tprof_countermask_t mask = TPROF_COUNTERMASK_ALL; memset(params, 0, sizeof(params)); while ((ch = getopt(argc, argv, do_profile ? "o:e:" : "e:i:")) != -1) { switch (ch) { case 'o': outfile = optarg; break; case 'i': interval = strtod(optarg, &p); if (*p != '\0' || interval <= 0) errx(EXIT_FAILURE, "Bad/invalid interval: %s", optarg); break; case 'e': if (tprof_parse_event(¶ms[nevent], optarg, TPROF_PARSE_EVENT_F_ALLOWOPTION | (do_profile ? TPROF_PARSE_EVENT_F_ALLOWSCALE : 0), &eventname[nevent], &errmsg) != 0) { errx(EXIT_FAILURE, "%s", errmsg); } eventnamewidth[nevent] = strlen(eventname[nevent]); if (eventnamewidth[nevent] < COUNTER_COLUMNS_WIDTH) eventnamewidth[nevent] = COUNTER_COLUMNS_WIDTH; nevent++; if (nevent > __arraycount(params) || nevent > ncounters) errx(EXIT_FAILURE, "Too many events. Only a" " maximum of %d counters can be used.", ncounters); break; default: usage(); } } argc -= optind; argv += optind; if (argc == 0) usage(); if (nevent == 0) { const char *defaultevent = tprof_cycle_event_name(); if (defaultevent == NULL) errx(EXIT_FAILURE, "cpu not supported"); tprof_event_lookup(defaultevent, ¶ms[nevent]); eventname[nevent] = defaultevent; params[nevent].p_flags |= TPROF_PARAM_KERN; nevent++; } if (do_profile) { outfd = open(outfile, O_WRONLY | O_CREAT | O_TRUNC, 0666); if (outfd == -1) { err(EXIT_FAILURE, "%s", outfile); } } for (i = 0; i < (int)nevent; i++) { params[i].p_counter = i; if (do_profile) params[i].p_flags |= TPROF_PARAM_PROFILE; ret = ioctl(devfd, TPROF_IOC_CONFIGURE_EVENT, ¶ms[i]); if (ret == -1) { err(EXIT_FAILURE, "TPROF_IOC_CONFIGURE_EVENT: %s", eventname[i]); } } ret = ioctl(devfd, TPROF_IOC_START, &mask); if (ret == -1) { err(EXIT_FAILURE, "TPROF_IOC_START"); } pid = fork(); switch (pid) { case -1: err(EXIT_FAILURE, "fork"); case 0: close(devfd); execvp(argv[0], argv); _Exit(EXIT_FAILURE); } signal(SIGINT, SIG_IGN); if (do_profile) signal(SIGINFO, siginfo_showcount); else signal(SIGINFO, siginfo_nothing); unsigned int done = 0; if (do_profile) ret = pthread_create(&pt, NULL, process_samples, NULL); else ret = pthread_create(&pt, NULL, process_stat, &done); if (ret != 0) errx(1, "pthread_create: %s", strerror(ret)); for (;;) { int status; pid = wait4(-1, &status, 0, NULL); if (pid == -1) { if (errno == ECHILD) { break; } err(EXIT_FAILURE, "wait4"); } if (pid != 0 && WIFEXITED(status)) { break; } } ret = ioctl(devfd, TPROF_IOC_STOP, &mask); if (ret == -1) { err(EXIT_FAILURE, "TPROF_IOC_STOP"); } if (!do_profile) { atomic_add_int(&done, 1); /* terminate thread */ kill(0, SIGINFO); } pthread_join(pt, NULL); if (do_profile) { ret = ioctl(devfd, TPROF_IOC_GETSTAT, &ts); if (ret == -1) err(EXIT_FAILURE, "TPROF_IOC_GETSTAT"); fprintf(stderr, "\n%s statistics:\n", getprogname()); fprintf(stderr, "\tsample %" PRIu64 "\n", ts.ts_sample); fprintf(stderr, "\toverflow %" PRIu64 "\n", ts.ts_overflow); fprintf(stderr, "\tbuf %" PRIu64 "\n", ts.ts_buf); fprintf(stderr, "\temptybuf %" PRIu64 "\n", ts.ts_emptybuf); fprintf(stderr, "\tdropbuf %" PRIu64 "\n", ts.ts_dropbuf); fprintf(stderr, "\tdropbuf_sample %" PRIu64 "\n", ts.ts_dropbuf_sample); fprintf(stderr, "\n"); } show_counters(); exit(EXIT_SUCCESS); } static void tprof_monitor(int argc, char **argv) { tprof_monitor_common(true, argc, argv); } static void tprof_count(int argc, char **argv) { tprof_monitor_common(false, argc, argv); } int main(int argc, char *argv[]) { const struct cmdtab *ct; int ret; getncpu(); setprogname(argv[0]); argv += 1, argc -= 1; devfd = open(_PATH_TPROF, O_RDWR); if (devfd == -1) { err(EXIT_FAILURE, "%s", _PATH_TPROF); } ret = ioctl(devfd, TPROF_IOC_GETINFO, &tprof_info); if (ret == -1) { err(EXIT_FAILURE, "TPROF_IOC_GETINFO"); } if (tprof_info.ti_version != TPROF_VERSION) { errx(EXIT_FAILURE, "version mismatch: version=%d, expected=%d", tprof_info.ti_version, TPROF_VERSION); } if (tprof_event_init(tprof_info.ti_ident) == -1) { errx(EXIT_FAILURE, "cpu not supported"); } ret = ioctl(devfd, TPROF_IOC_GETNCOUNTERS, &ncounters); if (ret == -1) { err(EXIT_FAILURE, "TPROF_IOC_GETNCOUNTERS"); } if (ncounters == 0) { errx(EXIT_FAILURE, "no available counters"); } if (argc == 0) usage(); for (ct = tprof_cmdtab; ct->label != NULL; ct++) { if (strcmp(argv[0], ct->label) == 0) { if (!ct->argsoptional && ((ct->takesargs == 0) ^ (argv[1] == NULL))) { usage(); } (*ct->func)(argc, argv); break; } } if (ct->label == NULL) { usage(); } }