Commit | Line | Data |
---|---|---|
15637ed4 RG |
1 | /*- |
2 | * Copyright (c) 1982, 1986, 1991 The Regents of the University of California. | |
3 | * All rights reserved. | |
4 | * | |
5 | * Redistribution and use in source and binary forms, with or without | |
6 | * modification, are permitted provided that the following conditions | |
7 | * are met: | |
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. | |
13 | * 3. All advertising materials mentioning features or use of this software | |
14 | * must display the following acknowledgement: | |
15 | * This product includes software developed by the University of | |
16 | * California, Berkeley and its contributors. | |
17 | * 4. Neither the name of the University nor the names of its contributors | |
18 | * may be used to endorse or promote products derived from this software | |
19 | * without specific prior written permission. | |
20 | * | |
21 | * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND | |
22 | * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE | |
23 | * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE | |
24 | * ARE DISCLAIMED. IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE | |
25 | * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL | |
26 | * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS | |
27 | * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) | |
28 | * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT | |
29 | * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY | |
30 | * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF | |
31 | * SUCH DAMAGE. | |
32 | * | |
600f7f07 | 33 | * from: @(#)kern_clock.c 7.16 (Berkeley) 5/9/91 |
415253c8 | 34 | * $Id: kern_clock.c,v 1.11 1993/12/19 00:51:20 wollman Exp $ |
15637ed4 RG |
35 | */ |
36 | ||
37 | #include "param.h" | |
38 | #include "systm.h" | |
39 | #include "dkstat.h" | |
40 | #include "callout.h" | |
41 | #include "kernel.h" | |
42 | #include "proc.h" | |
fde1aeb2 | 43 | #include "signalvar.h" |
15637ed4 RG |
44 | #include "resourcevar.h" |
45 | ||
46 | #include "machine/cpu.h" | |
47 | ||
a702c214 NW |
48 | #include "resource.h" |
49 | #include "vm/vm.h" | |
50 | ||
15637ed4 RG |
51 | #ifdef GPROF |
52 | #include "gprof.h" | |
53 | #endif | |
54 | ||
4c45483e GW |
55 | static void gatherstats(clockframe *); |
56 | ||
bbc3f849 GW |
57 | /* From callout.h */ |
58 | struct callout *callfree, *callout, calltodo; | |
59 | int ncallout; | |
60 | ||
15637ed4 RG |
61 | /* |
62 | * Clock handling routines. | |
63 | * | |
64 | * This code is written to operate with two timers which run | |
65 | * independently of each other. The main clock, running at hz | |
66 | * times per second, is used to do scheduling and timeout calculations. | |
67 | * The second timer does resource utilization estimation statistically | |
68 | * based on the state of the machine phz times a second. Both functions | |
69 | * can be performed by a single clock (ie hz == phz), however the | |
70 | * statistics will be much more prone to errors. Ideally a machine | |
71 | * would have separate clocks measuring time spent in user state, system | |
72 | * state, interrupt state, and idle state. These clocks would allow a non- | |
73 | * approximate measure of resource utilization. | |
74 | */ | |
75 | ||
76 | /* | |
77 | * TODO: | |
78 | * time of day, system/user timing, timeouts, profiling on separate timers | |
79 | * allocate more timeout table slots when table overflows. | |
80 | */ | |
81 | ||
82 | /* | |
83 | * Bump a timeval by a small number of usec's. | |
84 | */ | |
85 | #define BUMPTIME(t, usec) { \ | |
86 | register struct timeval *tp = (t); \ | |
87 | \ | |
88 | tp->tv_usec += (usec); \ | |
89 | if (tp->tv_usec >= 1000000) { \ | |
90 | tp->tv_usec -= 1000000; \ | |
91 | tp->tv_sec++; \ | |
92 | } \ | |
93 | } | |
94 | ||
95 | /* | |
96 | * The hz hardware interval timer. | |
97 | * We update the events relating to real time. | |
98 | * If this timer is also being used to gather statistics, | |
99 | * we run through the statistics gathering routine as well. | |
100 | */ | |
4c45483e | 101 | void |
15637ed4 RG |
102 | hardclock(frame) |
103 | clockframe frame; | |
104 | { | |
105 | register struct callout *p1; | |
106 | register struct proc *p = curproc; | |
4c45483e | 107 | register struct pstats *pstats = 0; |
a702c214 NW |
108 | register struct rusage *ru; |
109 | register struct vmspace *vm; | |
15637ed4 RG |
110 | register int s; |
111 | int needsoft = 0; | |
112 | extern int tickdelta; | |
113 | extern long timedelta; | |
114 | ||
115 | /* | |
116 | * Update real-time timeout queue. | |
117 | * At front of queue are some number of events which are ``due''. | |
118 | * The time to these is <= 0 and if negative represents the | |
119 | * number of ticks which have passed since it was supposed to happen. | |
120 | * The rest of the q elements (times > 0) are events yet to happen, | |
121 | * where the time for each is given as a delta from the previous. | |
122 | * Decrementing just the first of these serves to decrement the time | |
123 | * to all events. | |
124 | */ | |
125 | p1 = calltodo.c_next; | |
126 | while (p1) { | |
127 | if (--p1->c_time > 0) | |
128 | break; | |
129 | needsoft = 1; | |
130 | if (p1->c_time == 0) | |
131 | break; | |
132 | p1 = p1->c_next; | |
133 | } | |
134 | ||
135 | /* | |
136 | * Curproc (now in p) is null if no process is running. | |
137 | * We assume that curproc is set in user mode! | |
138 | */ | |
139 | if (p) | |
140 | pstats = p->p_stats; | |
141 | /* | |
142 | * Charge the time out based on the mode the cpu is in. | |
143 | * Here again we fudge for the lack of proper interval timers | |
144 | * assuming that the current state has been around at least | |
145 | * one tick. | |
146 | */ | |
147 | if (CLKF_USERMODE(&frame)) { | |
148 | if (pstats->p_prof.pr_scale) | |
149 | needsoft = 1; | |
150 | /* | |
151 | * CPU was in user state. Increment | |
152 | * user time counter, and process process-virtual time | |
153 | * interval timer. | |
154 | */ | |
155 | BUMPTIME(&p->p_utime, tick); | |
156 | if (timerisset(&pstats->p_timer[ITIMER_VIRTUAL].it_value) && | |
157 | itimerdecr(&pstats->p_timer[ITIMER_VIRTUAL], tick) == 0) | |
158 | psignal(p, SIGVTALRM); | |
159 | } else { | |
160 | /* | |
161 | * CPU was in system state. | |
162 | */ | |
163 | if (p) | |
164 | BUMPTIME(&p->p_stime, tick); | |
165 | } | |
166 | ||
a702c214 NW |
167 | /* bump the resource usage of integral space use */ |
168 | if (p && pstats && (ru = &pstats->p_ru) && (vm = p->p_vmspace)) { | |
d2ca7c11 DG |
169 | ru->ru_ixrss += vm->vm_tsize * NBPG / 1024; |
170 | ru->ru_idrss += vm->vm_dsize * NBPG / 1024; | |
171 | ru->ru_isrss += vm->vm_ssize * NBPG / 1024; | |
172 | if ((vm->vm_pmap.pm_stats.resident_count * NBPG / 1024) > | |
173 | ru->ru_maxrss) { | |
174 | ru->ru_maxrss = | |
175 | vm->vm_pmap.pm_stats.resident_count * NBPG / 1024; | |
176 | } | |
a702c214 NW |
177 | } |
178 | ||
15637ed4 RG |
179 | /* |
180 | * If the cpu is currently scheduled to a process, then | |
181 | * charge it with resource utilization for a tick, updating | |
182 | * statistics which run in (user+system) virtual time, | |
183 | * such as the cpu time limit and profiling timers. | |
184 | * This assumes that the current process has been running | |
185 | * the entire last tick. | |
186 | */ | |
187 | if (p) { | |
188 | if ((p->p_utime.tv_sec+p->p_stime.tv_sec+1) > | |
189 | p->p_rlimit[RLIMIT_CPU].rlim_cur) { | |
190 | psignal(p, SIGXCPU); | |
191 | if (p->p_rlimit[RLIMIT_CPU].rlim_cur < | |
192 | p->p_rlimit[RLIMIT_CPU].rlim_max) | |
193 | p->p_rlimit[RLIMIT_CPU].rlim_cur += 5; | |
194 | } | |
195 | if (timerisset(&pstats->p_timer[ITIMER_PROF].it_value) && | |
196 | itimerdecr(&pstats->p_timer[ITIMER_PROF], tick) == 0) | |
197 | psignal(p, SIGPROF); | |
198 | ||
199 | /* | |
200 | * We adjust the priority of the current process. | |
201 | * The priority of a process gets worse as it accumulates | |
202 | * CPU time. The cpu usage estimator (p_cpu) is increased here | |
203 | * and the formula for computing priorities (in kern_synch.c) | |
204 | * will compute a different value each time the p_cpu increases | |
205 | * by 4. The cpu usage estimator ramps up quite quickly when | |
206 | * the process is running (linearly), and decays away | |
207 | * exponentially, * at a rate which is proportionally slower | |
208 | * when the system is busy. The basic principal is that the | |
209 | * system will 90% forget that a process used a lot of CPU | |
210 | * time in 5*loadav seconds. This causes the system to favor | |
211 | * processes which haven't run much recently, and to | |
212 | * round-robin among other processes. | |
213 | */ | |
214 | p->p_cpticks++; | |
215 | if (++p->p_cpu == 0) | |
216 | p->p_cpu--; | |
217 | if ((p->p_cpu&3) == 0) { | |
218 | setpri(p); | |
219 | if (p->p_pri >= PUSER) | |
220 | p->p_pri = p->p_usrpri; | |
221 | } | |
222 | } | |
223 | ||
224 | /* | |
225 | * If the alternate clock has not made itself known then | |
226 | * we must gather the statistics. | |
227 | */ | |
228 | if (phz == 0) | |
229 | gatherstats(&frame); | |
230 | ||
231 | /* | |
232 | * Increment the time-of-day, and schedule | |
233 | * processing of the callouts at a very low cpu priority, | |
234 | * so we don't keep the relatively high clock interrupt | |
235 | * priority any longer than necessary. | |
236 | */ | |
237 | if (timedelta == 0) | |
238 | BUMPTIME(&time, tick) | |
239 | else { | |
240 | register delta; | |
241 | ||
242 | if (timedelta < 0) { | |
243 | delta = tick - tickdelta; | |
244 | timedelta += tickdelta; | |
245 | } else { | |
246 | delta = tick + tickdelta; | |
247 | timedelta -= tickdelta; | |
248 | } | |
249 | BUMPTIME(&time, delta); | |
250 | } | |
15637ed4 RG |
251 | if (needsoft) { |
252 | #if 0 | |
253 | /* | |
254 | * XXX - hardclock runs at splhigh, so the splsoftclock is useless and | |
255 | * softclock runs at splhigh as well if we do this. It is not much of | |
256 | * an optimization, since the "software interrupt" is done with a call | |
257 | * from doreti, and the overhead of checking there is sometimes less | |
258 | * than checking here. Moreover, the whole %$$%$^ frame is passed by | |
259 | * value here. | |
260 | */ | |
261 | if (CLKF_BASEPRI(&frame)) { | |
262 | /* | |
263 | * Save the overhead of a software interrupt; | |
264 | * it will happen as soon as we return, so do it now. | |
265 | */ | |
266 | (void) splsoftclock(); | |
267 | softclock(frame); | |
268 | } else | |
269 | #endif | |
270 | setsoftclock(); | |
271 | } | |
272 | } | |
273 | ||
274 | int dk_ndrive = DK_NDRIVE; | |
275 | /* | |
276 | * Gather statistics on resource utilization. | |
277 | * | |
278 | * We make a gross assumption: that the system has been in the | |
279 | * state it is in (user state, kernel state, interrupt state, | |
280 | * or idle state) for the entire last time interval, and | |
281 | * update statistics accordingly. | |
282 | */ | |
4c45483e | 283 | void |
15637ed4 RG |
284 | gatherstats(framep) |
285 | clockframe *framep; | |
286 | { | |
287 | register int cpstate, s; | |
288 | ||
289 | /* | |
290 | * Determine what state the cpu is in. | |
291 | */ | |
292 | if (CLKF_USERMODE(framep)) { | |
293 | /* | |
294 | * CPU was in user state. | |
295 | */ | |
296 | if (curproc->p_nice > NZERO) | |
297 | cpstate = CP_NICE; | |
298 | else | |
299 | cpstate = CP_USER; | |
300 | } else { | |
301 | /* | |
302 | * CPU was in system state. If profiling kernel | |
303 | * increment a counter. If no process is running | |
304 | * then this is a system tick if we were running | |
305 | * at a non-zero IPL (in a driver). If a process is running, | |
306 | * then we charge it with system time even if we were | |
307 | * at a non-zero IPL, since the system often runs | |
308 | * this way during processing of system calls. | |
309 | * This is approximate, but the lack of true interval | |
310 | * timers makes doing anything else difficult. | |
311 | */ | |
312 | cpstate = CP_SYS; | |
313 | if (curproc == NULL && CLKF_BASEPRI(framep)) | |
314 | cpstate = CP_IDLE; | |
315 | #ifdef GPROF | |
316 | s = (u_long) CLKF_PC(framep) - (u_long) s_lowpc; | |
317 | if (profiling < 2 && s < s_textsize) | |
318 | kcount[s / (HISTFRACTION * sizeof (*kcount))]++; | |
319 | #endif | |
320 | } | |
321 | /* | |
322 | * We maintain statistics shown by user-level statistics | |
323 | * programs: the amount of time in each cpu state, and | |
324 | * the amount of time each of DK_NDRIVE ``drives'' is busy. | |
325 | */ | |
326 | cp_time[cpstate]++; | |
327 | for (s = 0; s < DK_NDRIVE; s++) | |
328 | if (dk_busy&(1<<s)) | |
329 | dk_time[s]++; | |
330 | } | |
331 | ||
332 | /* | |
333 | * Software priority level clock interrupt. | |
334 | * Run periodic events from timeout queue. | |
335 | */ | |
336 | /*ARGSUSED*/ | |
4c45483e | 337 | void |
15637ed4 RG |
338 | softclock(frame) |
339 | clockframe frame; | |
340 | { | |
341 | ||
342 | for (;;) { | |
343 | register struct callout *p1; | |
344 | register caddr_t arg; | |
4c45483e | 345 | register timeout_func_t func; |
15637ed4 RG |
346 | register int a, s; |
347 | ||
348 | s = splhigh(); | |
349 | if ((p1 = calltodo.c_next) == 0 || p1->c_time > 0) { | |
350 | splx(s); | |
351 | break; | |
352 | } | |
353 | arg = p1->c_arg; func = p1->c_func; a = p1->c_time; | |
354 | calltodo.c_next = p1->c_next; | |
355 | p1->c_next = callfree; | |
356 | callfree = p1; | |
357 | splx(s); | |
358 | (*func)(arg, a); | |
359 | } | |
360 | ||
361 | /* | |
362 | * If no process to work with, we're finished. | |
363 | */ | |
364 | if (curproc == 0) return; | |
365 | ||
366 | /* | |
367 | * If trapped user-mode and profiling, give it | |
368 | * a profiling tick. | |
369 | */ | |
370 | if (CLKF_USERMODE(&frame)) { | |
371 | register struct proc *p = curproc; | |
372 | ||
373 | if (p->p_stats->p_prof.pr_scale) | |
374 | profile_tick(p, &frame); | |
375 | /* | |
376 | * Check to see if process has accumulated | |
377 | * more than 10 minutes of user time. If so | |
378 | * reduce priority to give others a chance. | |
379 | */ | |
380 | if (p->p_ucred->cr_uid && p->p_nice == NZERO && | |
381 | p->p_utime.tv_sec > 10 * 60) { | |
382 | p->p_nice = NZERO + 4; | |
383 | setpri(p); | |
384 | p->p_pri = p->p_usrpri; | |
385 | } | |
386 | } | |
387 | } | |
388 | ||
389 | /* | |
390 | * Arrange that (*func)(arg) is called in t/hz seconds. | |
391 | */ | |
4c45483e | 392 | void |
15637ed4 | 393 | timeout(func, arg, t) |
4c45483e | 394 | timeout_func_t func; |
15637ed4 RG |
395 | caddr_t arg; |
396 | register int t; | |
397 | { | |
398 | register struct callout *p1, *p2, *pnew; | |
399 | register int s = splhigh(); | |
400 | ||
401 | if (t <= 0) | |
402 | t = 1; | |
403 | pnew = callfree; | |
404 | if (pnew == NULL) | |
405 | panic("timeout table overflow"); | |
406 | callfree = pnew->c_next; | |
407 | pnew->c_arg = arg; | |
408 | pnew->c_func = func; | |
409 | for (p1 = &calltodo; (p2 = p1->c_next) && p2->c_time < t; p1 = p2) | |
410 | if (p2->c_time > 0) | |
411 | t -= p2->c_time; | |
412 | p1->c_next = pnew; | |
413 | pnew->c_next = p2; | |
414 | pnew->c_time = t; | |
415 | if (p2) | |
416 | p2->c_time -= t; | |
417 | splx(s); | |
418 | } | |
419 | ||
420 | /* | |
421 | * untimeout is called to remove a function timeout call | |
422 | * from the callout structure. | |
423 | */ | |
4c45483e | 424 | void |
15637ed4 | 425 | untimeout(func, arg) |
4c45483e | 426 | timeout_func_t func; |
15637ed4 RG |
427 | caddr_t arg; |
428 | { | |
429 | register struct callout *p1, *p2; | |
430 | register int s; | |
431 | ||
432 | s = splhigh(); | |
433 | for (p1 = &calltodo; (p2 = p1->c_next) != 0; p1 = p2) { | |
434 | if (p2->c_func == func && p2->c_arg == arg) { | |
435 | if (p2->c_next && p2->c_time > 0) | |
436 | p2->c_next->c_time += p2->c_time; | |
437 | p1->c_next = p2->c_next; | |
438 | p2->c_next = callfree; | |
439 | callfree = p2; | |
440 | break; | |
441 | } | |
442 | } | |
443 | splx(s); | |
444 | } | |
445 | ||
446 | /* | |
447 | * Compute number of hz until specified time. | |
448 | * Used to compute third argument to timeout() from an | |
449 | * absolute time. | |
450 | */ | |
c95f3c5b AC |
451 | |
452 | /* XXX clock_t */ | |
4c45483e | 453 | u_long |
15637ed4 RG |
454 | hzto(tv) |
455 | struct timeval *tv; | |
456 | { | |
c95f3c5b | 457 | register unsigned long ticks; |
15637ed4 | 458 | register long sec; |
c95f3c5b AC |
459 | register long usec; |
460 | int s; | |
15637ed4 RG |
461 | |
462 | /* | |
c95f3c5b AC |
463 | * If the number of usecs in the whole seconds part of the time |
464 | * difference fits in a long, then the total number of usecs will | |
465 | * fit in an unsigned long. Compute the total and convert it to | |
466 | * ticks, rounding up and adding 1 to allow for the current tick | |
467 | * to expire. Rounding also depends on unsigned long arithmetic | |
468 | * to avoid overflow. | |
469 | * | |
470 | * Otherwise, if the number of ticks in the whole seconds part of | |
471 | * the time difference fits in a long, then convert the parts to | |
472 | * ticks separately and add, using similar rounding methods and | |
473 | * overflow avoidance. This method would work in the previous | |
474 | * case but it is slightly slower and assumes that hz is integral. | |
15637ed4 | 475 | * |
c95f3c5b AC |
476 | * Otherwise, round the time difference down to the maximum |
477 | * representable value. | |
478 | * | |
479 | * Maximum value for any timeout in 10ms ticks is 248 days. | |
15637ed4 | 480 | */ |
c95f3c5b | 481 | s = splhigh(); |
15637ed4 | 482 | sec = tv->tv_sec - time.tv_sec; |
c95f3c5b | 483 | usec = tv->tv_usec - time.tv_usec; |
da6d4c06 | 484 | splx(s); |
c95f3c5b AC |
485 | if (usec < 0) { |
486 | sec--; | |
487 | usec += 1000000; | |
488 | } | |
489 | if (sec < 0) { | |
90945f8b | 490 | #ifdef DIAGNOSTIC |
c95f3c5b AC |
491 | printf("hzto: negative time difference %ld sec %ld usec\n", |
492 | sec, usec); | |
493 | #endif | |
494 | ticks = 1; | |
495 | } else if (sec <= LONG_MAX / 1000000) | |
496 | ticks = (sec * 1000000 + (unsigned long)usec + (tick - 1)) | |
497 | / tick + 1; | |
498 | else if (sec <= LONG_MAX / hz) | |
499 | ticks = sec * hz | |
500 | + ((unsigned long)usec + (tick - 1)) / tick + 1; | |
501 | else | |
502 | ticks = LONG_MAX; | |
503 | #define CLOCK_T_MAX INT_MAX /* XXX should be ULONG_MAX */ | |
504 | if (ticks > CLOCK_T_MAX) | |
505 | ticks = CLOCK_T_MAX; | |
15637ed4 RG |
506 | return (ticks); |
507 | } |