/* Author: Ben Gras 17 march 2006 */ /* Modified for ProcFS by Alen Stojanov and David van Moolenbroek */ #define _MINIX 1 #ifndef __NBSD_LIBC #define _POSIX_SOURCE 1 #endif #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #define TIMECYCLEKEY 't' u32_t system_hz; /* name of cpu cycle types, in the order they appear in /psinfo. */ char *cputimenames[] = { "user", "ipc", "kernelcall" }; #define CPUTIMENAMES (sizeof(cputimenames)/sizeof(cputimenames[0])) #define CPUTIME(m, i) (m & (1L << (i))) unsigned int nr_procs, nr_tasks; int nr_total; #define SLOT_NR(e) (_ENDPOINT_P(e) + nr_tasks) #define TC_BUFFER 1024 /* Size of termcap(3) buffer */ #define TC_STRINGS 200 /* Enough room for cm,cl,so,se */ char *Tclr_all; int blockedverbose = 0; #define USED 0x1 #define IS_TASK 0x2 #define IS_SYSTEM 0x4 #define BLOCKED 0x8 struct proc { int p_flags; endpoint_t p_endpoint; pid_t p_pid; u64_t p_cpucycles[CPUTIMENAMES]; int p_priority; endpoint_t p_blocked; time_t p_user_time; vir_bytes p_memory; uid_t p_effuid; int p_nice; char p_name[PROC_NAME_LEN+1]; }; struct proc *proc = NULL, *prev_proc = NULL; void parse_file(pid_t pid) { char path[PATH_MAX], name[256], type, state; int version, endpt, effuid; unsigned long cycles_hi, cycles_lo; FILE *fp; struct proc *p; int slot; int i; sprintf(path, "%d/psinfo", pid); if ((fp = fopen(path, "r")) == NULL) return; if (fscanf(fp, "%d", &version) != 1) { fclose(fp); return; } if (version != PSINFO_VERSION) { fputs("procfs version mismatch!\n", stderr); exit(1); } if (fscanf(fp, " %c %d", &type, &endpt) != 2) { fclose(fp); return; } slot = SLOT_NR(endpt); if(slot < 0 || slot >= nr_total) { fprintf(stderr, "top: unreasonable endpoint number %d\n", endpt); fclose(fp); return; } p = &proc[slot]; if (type == TYPE_TASK) p->p_flags |= IS_TASK; else if (type == TYPE_SYSTEM) p->p_flags |= IS_SYSTEM; p->p_endpoint = endpt; p->p_pid = pid; if (fscanf(fp, " %255s %c %d %d %lu %*u %lu %lu", name, &state, &p->p_blocked, &p->p_priority, &p->p_user_time, &cycles_hi, &cycles_lo) != 7) { fclose(fp); return; } strncpy(p->p_name, name, sizeof(p->p_name)-1); p->p_name[sizeof(p->p_name)-1] = 0; if (state != STATE_RUN) p->p_flags |= BLOCKED; p->p_cpucycles[0] = make64(cycles_lo, cycles_hi); p->p_memory = 0L; if (!(p->p_flags & IS_TASK)) { int i; if ((i=fscanf(fp, " %lu %*u %*u %*c %*d %*u %u %*u %d %*c %*d %*u", &p->p_memory, &effuid, &p->p_nice)) != 3) { fclose(fp); return; } p->p_effuid = effuid; } for(i = 1; i < CPUTIMENAMES; i++) { if(fscanf(fp, " %lu %lu", &cycles_hi, &cycles_lo) == 2) { p->p_cpucycles[i] = make64(cycles_lo, cycles_hi); } else { p->p_cpucycles[i] = make64(0, 0); } } p->p_flags |= USED; fclose(fp); } void parse_dir(void) { DIR *p_dir; struct dirent *p_ent; pid_t pid; char *end; if ((p_dir = opendir(".")) == NULL) { perror("opendir on " _PATH_PROC); exit(1); } for (p_ent = readdir(p_dir); p_ent != NULL; p_ent = readdir(p_dir)) { pid = strtol(p_ent->d_name, &end, 10); if (!end[0] && pid != 0) parse_file(pid); } closedir(p_dir); } void get_procs(void) { struct proc *p; int i; p = prev_proc; prev_proc = proc; proc = p; if (proc == NULL) { proc = malloc(nr_total * sizeof(proc[0])); if (proc == NULL) { fprintf(stderr, "Out of memory!\n"); exit(1); } } for (i = 0; i < nr_total; i++) proc[i].p_flags = 0; parse_dir(); } int print_memory(void) { FILE *fp; unsigned int pagesize; unsigned long total, free, largest, cached; if ((fp = fopen("meminfo", "r")) == NULL) return 0; if (fscanf(fp, "%u %lu %lu %lu %lu", &pagesize, &total, &free, &largest, &cached) != 5) { fclose(fp); return 0; } fclose(fp); printf("main memory: %ldK total, %ldK free, %ldK contig free, " "%ldK cached\n", (pagesize * total)/1024, (pagesize * free)/1024, (pagesize * largest)/1024, (pagesize * cached)/1024); return 1; } int print_load(double *loads, int nloads) { int i; printf("load averages: "); for(i = 0; i < nloads; i++) printf("%s %.2f", (i > 0) ? "," : "", loads[i]); printf("\n"); return 1; } int print_proc_summary(struct proc *proc) { int p, alive, running, sleeping; alive = running = sleeping = 0; for(p = 0; p < nr_total; p++) { if (proc[p].p_endpoint == IDLE) continue; if(!(proc[p].p_flags & USED)) continue; alive++; if(proc[p].p_flags & BLOCKED) sleeping++; else running++; } printf("%d processes: %d running, %d sleeping\n", alive, running, sleeping); return 1; } struct tp { struct proc *p; u64_t ticks; }; int cmp_ticks(const void *v1, const void *v2) { int c; struct tp *p1 = (struct tp *) v1, *p2 = (struct tp *) v2; int p1blocked, p2blocked; p1blocked = !!(p1->p->p_flags & BLOCKED); p2blocked = !!(p2->p->p_flags & BLOCKED); /* Primarily order by used number of cpu cycles. * * Exception: if in blockedverbose mode, a blocked * process is always printed after an unblocked * process, and used cpu cycles don't matter. * * In both cases, process slot number is a tie breaker. */ if(blockedverbose && (p1blocked || p2blocked)) { if(!p1blocked && p2blocked) return -1; if( p2blocked && !p1blocked) return 1; } else if((c=cmp64(p1->ticks, p2->ticks)) != 0) return -c; /* Process slot number is a tie breaker. */ return (int) (p1->p - p2->p); } struct tp *lookup(endpoint_t who, struct tp *tptab, int np) { int t; for(t = 0; t < np; t++) if(who == tptab[t].p->p_endpoint) return &tptab[t]; fprintf(stderr, "lookup: tp %d (0x%x) not found.\n", who, who); abort(); return NULL; } /* * since we don't have true div64(u64_t, u64_t) we scale the 64 bit counters to * 32. Since the samplig happens every ~1s and the counters count CPU cycles * during this period, unless we have extremely fast CPU, shifting the counters * by 12 make them 32bit counters which we can use for normal integer * arithmetics */ #define SCALE (1 << 12) double ktotal = 0; void print_proc(struct tp *tp, u32_t tcyc) { int euid = 0; static struct passwd *who = NULL; static int last_who = -1; char *name = ""; unsigned long pcyc; int ticks; struct proc *pr = tp->p; printf("%5d ", pr->p_pid); euid = pr->p_effuid; name = pr->p_name; if(last_who != euid || !who) { who = getpwuid(euid); last_who = euid; } if(who && who->pw_name) printf("%-8s ", who->pw_name); else if(!(pr->p_flags & IS_TASK)) printf("%8d ", pr->p_effuid); else printf(" "); printf(" %2d ", pr->p_priority); if(!(pr->p_flags & IS_TASK)) { printf(" %3d ", pr->p_nice); } else printf(" "); printf("%6ldK", (pr->p_memory + 512) / 1024); printf("%6s", (pr->p_flags & BLOCKED) ? "" : "RUN"); ticks = pr->p_user_time; printf(" %3u:%02u ", (ticks/system_hz/60), (ticks/system_hz)%60); pcyc = div64u(tp->ticks, SCALE); printf("%6.2f%% %s", 100.0*pcyc/tcyc, name); } char *cputimemodename(int cputimemode) { static char name[100]; int i; name[0] = '\0'; for(i = 0; i < CPUTIMENAMES; i++) { if(CPUTIME(cputimemode, i)) { assert(strlen(name) + strlen(cputimenames[i]) < sizeof(name)); strcat(name, cputimenames[i]); strcat(name, " "); } } return name; } u64_t cputicks(struct proc *p1, struct proc *p2, int timemode) { int i; u64_t t = make64(0, 0); for(i = 0; i < CPUTIMENAMES; i++) { if(!CPUTIME(timemode, i)) continue; if(p1->p_endpoint == p2->p_endpoint) { t = add64(t, sub64(p2->p_cpucycles[i], p1->p_cpucycles[i])); } else { t = add64(t, p2->p_cpucycles[i]); } } return t; } void print_procs(int maxlines, struct proc *proc1, struct proc *proc2, int cputimemode) { int p, nprocs; u64_t idleticks = cvu64(0); u64_t kernelticks = cvu64(0); u64_t systemticks = cvu64(0); u64_t userticks = cvu64(0); u64_t total_ticks = cvu64(0); unsigned long tcyc; unsigned long tmp; int blockedseen = 0; static struct tp *tick_procs = NULL; if (tick_procs == NULL) { tick_procs = malloc(nr_total * sizeof(tick_procs[0])); if (tick_procs == NULL) { fprintf(stderr, "Out of memory!\n"); exit(1); } } for(p = nprocs = 0; p < nr_total; p++) { u64_t uticks; if(!(proc2[p].p_flags & USED)) continue; tick_procs[nprocs].p = proc2 + p; tick_procs[nprocs].ticks = cputicks(&proc1[p], &proc2[p], cputimemode); uticks = cputicks(&proc1[p], &proc2[p], 1); total_ticks = add64(total_ticks, uticks); if(p-NR_TASKS == IDLE) { idleticks = uticks; continue; } if(p-NR_TASKS == KERNEL) { kernelticks = uticks; continue; } if(!(proc2[p].p_flags & IS_TASK)) { if(proc2[p].p_flags & IS_SYSTEM) systemticks = add64(systemticks, tick_procs[nprocs].ticks); else userticks = add64(userticks, tick_procs[nprocs].ticks); } nprocs++; } if (!cmp64u(total_ticks, 0)) return; qsort(tick_procs, nprocs, sizeof(tick_procs[0]), cmp_ticks); tcyc = div64u(total_ticks, SCALE); tmp = div64u(userticks, SCALE); printf("CPU states: %6.2f%% user, ", 100.0*(tmp)/tcyc); tmp = div64u(systemticks, SCALE); printf("%6.2f%% system, ", 100.0*tmp/tcyc); tmp = div64u(kernelticks, SCALE); printf("%6.2f%% kernel, ", 100.0*tmp/tcyc); tmp = div64u(idleticks, SCALE); printf("%6.2f%% idle", 100.0*tmp/tcyc); #define NEWLINE do { printf("\n"); if(--maxlines <= 0) { return; } } while(0) NEWLINE; printf("CPU time displayed (press '%c' to cycle): %s", TIMECYCLEKEY, cputimemodename(cputimemode)); NEWLINE; NEWLINE; printf(" PID USERNAME PRI NICE SIZE STATE TIME CPU COMMAND"); NEWLINE; for(p = 0; p < nprocs; p++) { struct proc *pr; int level = 0; pr = tick_procs[p].p; if(pr->p_flags & IS_TASK) { /* skip old kernel tasks as they don't run anymore */ continue; } /* If we're in blocked verbose mode, indicate start of * blocked processes. */ if(blockedverbose && (pr->p_flags & BLOCKED) && !blockedseen) { NEWLINE; printf("Blocked processes:"); NEWLINE; blockedseen = 1; } print_proc(&tick_procs[p], tcyc); NEWLINE; if(!blockedverbose) continue; /* Traverse dependency chain if blocked. */ while(pr->p_flags & BLOCKED) { endpoint_t dep = NONE; struct tp *tpdep; level += 5; if((dep = pr->p_blocked) == NONE) { printf("not blocked on a process"); NEWLINE; break; } if(dep == ANY) break; tpdep = lookup(dep, tick_procs, nprocs); pr = tpdep->p; printf("%*s> ", level, ""); print_proc(tpdep, tcyc); NEWLINE; } } } void showtop(int cputimemode, int r) { #define NLOADS 3 double loads[NLOADS]; int nloads, lines = 0; struct winsize winsize; if(ioctl(STDIN_FILENO, TIOCGWINSZ, &winsize) != 0) { perror("TIOCGWINSZ"); fprintf(stderr, "TIOCGWINSZ failed\n"); exit(1); } get_procs(); if (prev_proc == NULL) get_procs(); if((nloads = getloadavg(loads, NLOADS)) != NLOADS) { fprintf(stderr, "getloadavg() failed - %d loads\n", nloads); exit(1); } printf("%s", Tclr_all); lines += print_load(loads, NLOADS); lines += print_proc_summary(proc); lines += print_memory(); if(winsize.ws_row > 0) r = winsize.ws_row; print_procs(r - lines - 2, prev_proc, proc, cputimemode); fflush(NULL); } void init(int *rows) { char *term; static char buffer[TC_BUFFER], strings[TC_STRINGS]; char *s = strings, *v; *rows = 0; if(!(term = getenv("TERM"))) { fprintf(stderr, "No TERM set\n"); exit(1); } if ( tgetent( buffer, term ) != 1 ) { fprintf(stderr, "tgetent failed for term %s\n", term); exit(1); } initscr(); cbreak(); if ( (Tclr_all = tgetstr( "cl", &s )) == NULL ) Tclr_all = "\f"; if((v = tgetstr ("li", &s)) != NULL) sscanf(v, "%d", rows); if(*rows < 1) *rows = 24; } void sigwinch(int sig) { } void getkinfo(void) { FILE *fp; if ((fp = fopen("kinfo", "r")) == NULL) { fprintf(stderr, "opening " _PATH_PROC "kinfo failed\n"); exit(1); } if (fscanf(fp, "%u %u", &nr_procs, &nr_tasks) != 2) { fprintf(stderr, "reading from " _PATH_PROC "kinfo failed\n"); exit(1); } fclose(fp); nr_total = (int) (nr_procs + nr_tasks); } int main(int argc, char *argv[]) { int r, c, s = 0; int cputimemode = 1; /* bitmap. */ if (chdir(_PATH_PROC) != 0) { perror("chdir to " _PATH_PROC); return 1; } system_hz = (u32_t) sysconf(_SC_CLK_TCK); getkinfo(); init(&r); while((c=getopt(argc, argv, "s:B")) != EOF) { switch(c) { case 's': s = atoi(optarg); break; case 'B': blockedverbose = 1; break; default: fprintf(stderr, "Usage: %s [-s] [-B]\n", argv[0]); return 1; } } if(s < 1) s = 2; /* Catch window size changes so display is updated properly * right away. */ signal(SIGWINCH, sigwinch); while(1) { fd_set fds; int ns; struct timeval tv; showtop(cputimemode, r); tv.tv_sec = s; tv.tv_usec = 0; FD_ZERO(&fds); FD_SET(STDIN_FILENO, &fds); if((ns=select(STDIN_FILENO+1, &fds, NULL, NULL, &tv)) < 0 && errno != EINTR) { perror("select"); sleep(1); } if(ns > 0 && FD_ISSET(STDIN_FILENO, &fds)) { char c; if(read(STDIN_FILENO, &c, 1) == 1) { switch(c) { case 'q': putchar('\r'); return 0; break; case TIMECYCLEKEY: cputimemode++; if(cputimemode >= (1L << CPUTIMENAMES)) cputimemode = 1; break; } } } } return 0; }