/*- * Copyright (c) 1982, 1986, 1991, 1993 * The Regents of the University of California. All rights reserved. * (c) UNIX System Laboratories, Inc. * All or some portions of this file are derived from material licensed * to the University of California by American Telephone and Telegraph * Co. or Unix System Laboratories, Inc. and are reproduced herein with * the permission of UNIX System Laboratories, Inc. * * 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. * 3. All advertising materials mentioning features or use of this software * must display the following acknowledgement: * This product includes software developed by the University of * California, Berkeley and its contributors. * 4. Neither the name of the University nor the names of its contributors * may be used to endorse or promote products derived from this software * without specific prior written permission. * * THIS SOFTWARE IS PROVIDED BY THE REGENTS 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 REGENTS 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. * * @(#)kern_clock.c 8.5 (Berkeley) 1/21/94 * $Id: kern_clock.c,v 1.55 1998/02/06 12:13:22 eivind Exp $ */ #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #define CLOCK_HAIR /* XXX */ #include #include #ifdef GPROF #include #endif #if defined(SMP) && defined(BETTER_CLOCK) #include #endif static void initclocks __P((void *dummy)); SYSINIT(clocks, SI_SUB_CLOCKS, SI_ORDER_FIRST, initclocks, NULL) /* Some of these don't belong here, but it's easiest to concentrate them. */ #if defined(SMP) && defined(BETTER_CLOCK) long cp_time[CPUSTATES]; #else static long cp_time[CPUSTATES]; #endif long dk_seek[DK_NDRIVE]; static long dk_time[DK_NDRIVE]; /* time busy (in statclock ticks) */ long dk_wds[DK_NDRIVE]; long dk_wpms[DK_NDRIVE]; long dk_xfer[DK_NDRIVE]; int dk_busy; int dk_ndrive = 0; char dk_names[DK_NDRIVE][DK_NAMELEN]; long tk_cancc; long tk_nin; long tk_nout; long tk_rawcc; /* * Clock handling routines. * * This code is written to operate with two timers that run independently of * each other. The main clock, running hz times per second, is used to keep * track of real time. The second timer handles kernel and user profiling, * and does resource use estimation. If the second timer is programmable, * it is randomized to avoid aliasing between the two clocks. For example, * the randomization prevents an adversary from always giving up the cpu * just before its quantum expires. Otherwise, it would never accumulate * cpu ticks. The mean frequency of the second timer is stathz. * * If no second timer exists, stathz will be zero; in this case we drive * profiling and statistics off the main clock. This WILL NOT be accurate; * do not do it unless absolutely necessary. * * The statistics clock may (or may not) be run at a higher rate while * profiling. This profile clock runs at profhz. We require that profhz * be an integral multiple of stathz. * * If the statistics clock is running fast, it must be divided by the ratio * profhz/stathz for statistics. (For profiling, every tick counts.) */ /* * TODO: * allocate more timeout table slots when table overflows. */ /* * Bump a timeval by a small number of usec's. */ #define BUMPTIME(t, usec) { \ register volatile struct timeval *tp = (t); \ register long us; \ \ tp->tv_usec = us = tp->tv_usec + (usec); \ if (us >= 1000000) { \ tp->tv_usec = us - 1000000; \ tp->tv_sec++; \ } \ } int stathz; int profhz; static int profprocs; int ticks; static int psdiv, pscnt; /* prof => stat divider */ int psratio; /* ratio: prof / stat */ volatile struct timeval time; volatile struct timeval mono_time; /* * Initialize clock frequencies and start both clocks running. */ /* ARGSUSED*/ static void initclocks(dummy) void *dummy; { register int i; /* * Set divisors to 1 (normal case) and let the machine-specific * code do its bit. */ psdiv = pscnt = 1; cpu_initclocks(); /* * Compute profhz/stathz, and fix profhz if needed. */ i = stathz ? stathz : hz; if (profhz == 0) profhz = i; psratio = profhz / i; } /* * The real-time timer, interrupting hz times per second. */ void hardclock(frame) register struct clockframe *frame; { register struct proc *p; int time_update; struct timeval newtime = time; long ltemp; p = curproc; if (p) { register struct pstats *pstats; /* * Run current process's virtual and profile time, as needed. */ pstats = p->p_stats; if (CLKF_USERMODE(frame) && timerisset(&pstats->p_timer[ITIMER_VIRTUAL].it_value) && itimerdecr(&pstats->p_timer[ITIMER_VIRTUAL], tick) == 0) psignal(p, SIGVTALRM); if (timerisset(&pstats->p_timer[ITIMER_PROF].it_value) && itimerdecr(&pstats->p_timer[ITIMER_PROF], tick) == 0) psignal(p, SIGPROF); } #if defined(SMP) && defined(BETTER_CLOCK) forward_hardclock(pscnt); #endif /* * If no separate statistics clock is available, run it from here. */ if (stathz == 0) statclock(frame); /* * Increment the time-of-day. */ ticks++; if (timedelta == 0) { time_update = CPU_THISTICKLEN(tick); } else { time_update = CPU_THISTICKLEN(tick) + tickdelta; timedelta -= tickdelta; } BUMPTIME(&mono_time, time_update); /* * Compute the phase adjustment. If the low-order bits * (time_phase) of the update overflow, bump the high-order bits * (time_update). */ time_phase += time_adj; if (time_phase <= -FINEUSEC) { ltemp = -time_phase >> SHIFT_SCALE; time_phase += ltemp << SHIFT_SCALE; time_update -= ltemp; } else if (time_phase >= FINEUSEC) { ltemp = time_phase >> SHIFT_SCALE; time_phase -= ltemp << SHIFT_SCALE; time_update += ltemp; } newtime.tv_usec += time_update; /* * On rollover of the second the phase adjustment to be used for * the next second is calculated. Also, the maximum error is * increased by the tolerance. If the PPS frequency discipline * code is present, the phase is increased to compensate for the * CPU clock oscillator frequency error. * * On a 32-bit machine and given parameters in the timex.h * header file, the maximum phase adjustment is +-512 ms and * maximum frequency offset is a tad less than) +-512 ppm. On a * 64-bit machine, you shouldn't need to ask. */ if (newtime.tv_usec >= 1000000) { newtime.tv_usec -= 1000000; newtime.tv_sec++; ntp_update_second(&newtime.tv_sec); } CPU_CLOCKUPDATE(&time, &newtime); if (TAILQ_FIRST(&callwheel[ticks & callwheelmask]) != NULL) setsoftclock(); } void gettime(struct timeval *tvp) { int s; s = splclock(); /* XXX should use microtime() iff tv_usec is used. */ *tvp = time; splx(s); } /* * Compute number of hz until specified time. Used to * compute third argument to timeout() from an absolute time. */ int hzto(tv) struct timeval *tv; { register unsigned long ticks; register long sec, usec; int s; /* * If the number of usecs in the whole seconds part of the time * difference fits in a long, then the total number of usecs will * fit in an unsigned long. Compute the total and convert it to * ticks, rounding up and adding 1 to allow for the current tick * to expire. Rounding also depends on unsigned long arithmetic * to avoid overflow. * * Otherwise, if the number of ticks in the whole seconds part of * the time difference fits in a long, then convert the parts to * ticks separately and add, using similar rounding methods and * overflow avoidance. This method would work in the previous * case but it is slightly slower and assumes that hz is integral. * * Otherwise, round the time difference down to the maximum * representable value. * * If ints have 32 bits, then the maximum value for any timeout in * 10ms ticks is 248 days. */ s = splclock(); sec = tv->tv_sec - time.tv_sec; usec = tv->tv_usec - time.tv_usec; splx(s); if (usec < 0) { sec--; usec += 1000000; } if (sec < 0) { #ifdef DIAGNOSTIC printf("hzto: negative time difference %ld sec %ld usec\n", sec, usec); #endif ticks = 1; } else if (sec <= LONG_MAX / 1000000) ticks = (sec * 1000000 + (unsigned long)usec + (tick - 1)) / tick + 1; else if (sec <= LONG_MAX / hz) ticks = sec * hz + ((unsigned long)usec + (tick - 1)) / tick + 1; else ticks = LONG_MAX; if (ticks > INT_MAX) ticks = INT_MAX; return (ticks); } /* * Start profiling on a process. * * Kernel profiling passes proc0 which never exits and hence * keeps the profile clock running constantly. */ void startprofclock(p) register struct proc *p; { int s; if ((p->p_flag & P_PROFIL) == 0) { p->p_flag |= P_PROFIL; if (++profprocs == 1 && stathz != 0) { s = splstatclock(); psdiv = pscnt = psratio; setstatclockrate(profhz); splx(s); } } } /* * Stop profiling on a process. */ void stopprofclock(p) register struct proc *p; { int s; if (p->p_flag & P_PROFIL) { p->p_flag &= ~P_PROFIL; if (--profprocs == 0 && stathz != 0) { s = splstatclock(); psdiv = pscnt = 1; setstatclockrate(stathz); splx(s); } } } /* * Statistics clock. Grab profile sample, and if divider reaches 0, * do process and kernel statistics. */ void statclock(frame) register struct clockframe *frame; { #ifdef GPROF register struct gmonparam *g; #endif register struct proc *p; register int i; struct pstats *pstats; long rss; struct rusage *ru; struct vmspace *vm; if (CLKF_USERMODE(frame)) { p = curproc; if (p->p_flag & P_PROFIL) addupc_intr(p, CLKF_PC(frame), 1); #if defined(SMP) && defined(BETTER_CLOCK) if (stathz != 0) forward_statclock(pscnt); #endif if (--pscnt > 0) return; /* * Came from user mode; CPU was in user state. * If this process is being profiled record the tick. */ p->p_uticks++; if (p->p_nice > NZERO) cp_time[CP_NICE]++; else cp_time[CP_USER]++; } else { #ifdef GPROF /* * Kernel statistics are just like addupc_intr, only easier. */ g = &_gmonparam; if (g->state == GMON_PROF_ON) { i = CLKF_PC(frame) - g->lowpc; if (i < g->textsize) { i /= HISTFRACTION * sizeof(*g->kcount); g->kcount[i]++; } } #endif #if defined(SMP) && defined(BETTER_CLOCK) if (stathz != 0) forward_statclock(pscnt); #endif if (--pscnt > 0) return; /* * Came from kernel mode, so we were: * - handling an interrupt, * - doing syscall or trap work on behalf of the current * user process, or * - spinning in the idle loop. * Whichever it is, charge the time as appropriate. * Note that we charge interrupts to the current process, * regardless of whether they are ``for'' that process, * so that we know how much of its real time was spent * in ``non-process'' (i.e., interrupt) work. */ p = curproc; if (CLKF_INTR(frame)) { if (p != NULL) p->p_iticks++; cp_time[CP_INTR]++; } else if (p != NULL) { p->p_sticks++; cp_time[CP_SYS]++; } else cp_time[CP_IDLE]++; } pscnt = psdiv; /* * We maintain statistics shown by user-level statistics * programs: the amount of time in each cpu state, and * the amount of time each of DK_NDRIVE ``drives'' is busy. * * XXX should either run linked list of drives, or (better) * grab timestamps in the start & done code. */ for (i = 0; i < DK_NDRIVE; i++) if (dk_busy & (1 << i)) dk_time[i]++; /* * We adjust the priority of the current process. The priority of * a process gets worse as it accumulates CPU time. The cpu usage * estimator (p_estcpu) is increased here. The formula for computing * priorities (in kern_synch.c) will compute a different value each * time p_estcpu increases by 4. The cpu usage estimator ramps up * quite quickly when the process is running (linearly), and decays * away exponentially, at a rate which is proportionally slower when * the system is busy. The basic principal is that the system will * 90% forget that the process used a lot of CPU time in 5 * loadav * seconds. This causes the system to favor processes which haven't * run much recently, and to round-robin among other processes. */ if (p != NULL) { p->p_cpticks++; if (++p->p_estcpu == 0) p->p_estcpu--; if ((p->p_estcpu & 3) == 0) { resetpriority(p); if (p->p_priority >= PUSER) p->p_priority = p->p_usrpri; } /* Update resource usage integrals and maximums. */ if ((pstats = p->p_stats) != NULL && (ru = &pstats->p_ru) != NULL && (vm = p->p_vmspace) != NULL) { ru->ru_ixrss += vm->vm_tsize * PAGE_SIZE / 1024; ru->ru_idrss += vm->vm_dsize * PAGE_SIZE / 1024; ru->ru_isrss += vm->vm_ssize * PAGE_SIZE / 1024; rss = vm->vm_pmap.pm_stats.resident_count * PAGE_SIZE / 1024; if (ru->ru_maxrss < rss) ru->ru_maxrss = rss; } } } /* * Return information about system clocks. */ static int sysctl_kern_clockrate SYSCTL_HANDLER_ARGS { struct clockinfo clkinfo; /* * Construct clockinfo structure. */ clkinfo.hz = hz; clkinfo.tick = tick; clkinfo.tickadj = tickadj; clkinfo.profhz = profhz; clkinfo.stathz = stathz ? stathz : hz; return (sysctl_handle_opaque(oidp, &clkinfo, sizeof clkinfo, req)); } SYSCTL_PROC(_kern, KERN_CLOCKRATE, clockrate, CTLTYPE_STRUCT|CTLFLAG_RD, 0, 0, sysctl_kern_clockrate, "S,clockinfo",""); void nanotime(ts) struct timespec *ts; { struct timeval tv; microtime(&tv); ts->tv_sec = tv.tv_sec; ts->tv_nsec = tv.tv_usec * 1000; }