Commit | Line | Data |
---|---|---|
da7c5cc6 | 1 | /* |
0880b18e | 2 | * Copyright (c) 1982, 1986 Regents of the University of California. |
da7c5cc6 KM |
3 | * All rights reserved. The Berkeley software License Agreement |
4 | * specifies the terms and conditions for redistribution. | |
5 | * | |
9d4095a1 | 6 | * @(#)kern_clock.c 7.11 (Berkeley) %G% |
da7c5cc6 | 7 | */ |
961945a8 | 8 | |
94368568 JB |
9 | #include "param.h" |
10 | #include "systm.h" | |
fb1db32c | 11 | #include "dkstat.h" |
94368568 | 12 | #include "callout.h" |
d9c2f47f | 13 | #include "user.h" |
94368568 JB |
14 | #include "kernel.h" |
15 | #include "proc.h" | |
83be5fac | 16 | |
40ed2c45 KM |
17 | #include "machine/reg.h" |
18 | #include "machine/psl.h" | |
8e8cbcca | 19 | |
fb1db32c | 20 | #if defined(vax) || defined(tahoe) |
40ed2c45 KM |
21 | #include "machine/mtpr.h" |
22 | #include "machine/clock.h" | |
961945a8 | 23 | #endif |
933220e9 KM |
24 | #if defined(hp300) |
25 | #include "machine/mtpr.h" | |
26 | #endif | |
d293217c WN |
27 | #ifdef i386 |
28 | #include "machine/frame.h" | |
29 | #include "machine/segments.h" | |
30 | #endif | |
961945a8 | 31 | |
8487304f | 32 | #ifdef GPROF |
94368568 | 33 | #include "gprof.h" |
8487304f KM |
34 | #endif |
35 | ||
45e9acec MK |
36 | #define ADJTIME /* For now... */ |
37 | #define ADJ_TICK 1000 | |
38 | int adjtimedelta; | |
39 | ||
76b2a182 BJ |
40 | /* |
41 | * Clock handling routines. | |
42 | * | |
53a32545 SL |
43 | * This code is written to operate with two timers which run |
44 | * independently of each other. The main clock, running at hz | |
45 | * times per second, is used to do scheduling and timeout calculations. | |
46 | * The second timer does resource utilization estimation statistically | |
47 | * based on the state of the machine phz times a second. Both functions | |
48 | * can be performed by a single clock (ie hz == phz), however the | |
49 | * statistics will be much more prone to errors. Ideally a machine | |
50 | * would have separate clocks measuring time spent in user state, system | |
51 | * state, interrupt state, and idle state. These clocks would allow a non- | |
52 | * approximate measure of resource utilization. | |
76b2a182 | 53 | */ |
6602c75b | 54 | |
76b2a182 BJ |
55 | /* |
56 | * TODO: | |
88a7a62a SL |
57 | * time of day, system/user timing, timeouts, profiling on separate timers |
58 | * allocate more timeout table slots when table overflows. | |
76b2a182 | 59 | */ |
9c5cfb8b | 60 | |
ad8023d1 KM |
61 | /* |
62 | * Bump a timeval by a small number of usec's. | |
63 | */ | |
ad8023d1 KM |
64 | #define BUMPTIME(t, usec) { \ |
65 | register struct timeval *tp = (t); \ | |
66 | \ | |
67 | tp->tv_usec += (usec); \ | |
68 | if (tp->tv_usec >= 1000000) { \ | |
69 | tp->tv_usec -= 1000000; \ | |
70 | tp->tv_sec++; \ | |
71 | } \ | |
72 | } | |
83be5fac | 73 | |
76b2a182 | 74 | /* |
53a32545 SL |
75 | * The hz hardware interval timer. |
76 | * We update the events relating to real time. | |
77 | * If this timer is also being used to gather statistics, | |
78 | * we run through the statistics gathering routine as well. | |
76b2a182 | 79 | */ |
260ea681 | 80 | /*ARGSUSED*/ |
d293217c | 81 | #ifndef i386 |
f403d99f | 82 | hardclock(pc, ps) |
4512b9a4 | 83 | caddr_t pc; |
460ab27f | 84 | int ps; |
d293217c WN |
85 | #else |
86 | hardclock(frame) | |
87 | struct intrframe frame; | |
88 | #define pc frame.if_eip | |
89 | #endif | |
83be5fac | 90 | { |
0a34b6fd | 91 | register struct callout *p1; |
53fbb3b3 | 92 | register struct proc *p = u.u_procp; |
0b355a6e | 93 | register int s; |
83be5fac | 94 | |
76b2a182 BJ |
95 | /* |
96 | * Update real-time timeout queue. | |
97 | * At front of queue are some number of events which are ``due''. | |
98 | * The time to these is <= 0 and if negative represents the | |
99 | * number of ticks which have passed since it was supposed to happen. | |
100 | * The rest of the q elements (times > 0) are events yet to happen, | |
101 | * where the time for each is given as a delta from the previous. | |
102 | * Decrementing just the first of these serves to decrement the time | |
103 | * to all events. | |
104 | */ | |
88a7a62a SL |
105 | p1 = calltodo.c_next; |
106 | while (p1) { | |
107 | if (--p1->c_time > 0) | |
108 | break; | |
88a7a62a SL |
109 | if (p1->c_time == 0) |
110 | break; | |
111 | p1 = p1->c_next; | |
112 | } | |
5da67d35 | 113 | |
76b2a182 BJ |
114 | /* |
115 | * Charge the time out based on the mode the cpu is in. | |
116 | * Here again we fudge for the lack of proper interval timers | |
117 | * assuming that the current state has been around at least | |
118 | * one tick. | |
119 | */ | |
d293217c WN |
120 | #ifdef i386 |
121 | if (ISPL(frame.if_cs) == SEL_UPL) { | |
122 | #else | |
83be5fac | 123 | if (USERMODE(ps)) { |
d293217c | 124 | #endif |
76b2a182 BJ |
125 | /* |
126 | * CPU was in user state. Increment | |
127 | * user time counter, and process process-virtual time | |
877ef342 | 128 | * interval timer. |
76b2a182 | 129 | */ |
53fbb3b3 | 130 | BUMPTIME(&p->p_utime, tick); |
27b91f59 BJ |
131 | if (timerisset(&u.u_timer[ITIMER_VIRTUAL].it_value) && |
132 | itimerdecr(&u.u_timer[ITIMER_VIRTUAL], tick) == 0) | |
53fbb3b3 | 133 | psignal(p, SIGVTALRM); |
83be5fac | 134 | } else { |
76b2a182 | 135 | /* |
0b355a6e | 136 | * CPU was in system state. |
76b2a182 | 137 | */ |
9c5cfb8b | 138 | if (!noproc) |
53fbb3b3 | 139 | BUMPTIME(&p->p_stime, tick); |
83be5fac | 140 | } |
27b91f59 | 141 | |
9fb1a8d0 SL |
142 | /* |
143 | * If the cpu is currently scheduled to a process, then | |
144 | * charge it with resource utilization for a tick, updating | |
145 | * statistics which run in (user+system) virtual time, | |
146 | * such as the cpu time limit and profiling timers. | |
147 | * This assumes that the current process has been running | |
148 | * the entire last tick. | |
149 | */ | |
405de916 | 150 | if (noproc == 0) { |
53fbb3b3 | 151 | if ((p->p_utime.tv_sec+p->p_stime.tv_sec+1) > |
9fb1a8d0 | 152 | u.u_rlimit[RLIMIT_CPU].rlim_cur) { |
53fbb3b3 | 153 | psignal(p, SIGXCPU); |
9fb1a8d0 SL |
154 | if (u.u_rlimit[RLIMIT_CPU].rlim_cur < |
155 | u.u_rlimit[RLIMIT_CPU].rlim_max) | |
156 | u.u_rlimit[RLIMIT_CPU].rlim_cur += 5; | |
157 | } | |
158 | if (timerisset(&u.u_timer[ITIMER_PROF].it_value) && | |
159 | itimerdecr(&u.u_timer[ITIMER_PROF], tick) == 0) | |
53fbb3b3 | 160 | psignal(p, SIGPROF); |
9fb1a8d0 SL |
161 | } |
162 | ||
76b2a182 BJ |
163 | /* |
164 | * We adjust the priority of the current process. | |
165 | * The priority of a process gets worse as it accumulates | |
166 | * CPU time. The cpu usage estimator (p_cpu) is increased here | |
167 | * and the formula for computing priorities (in kern_synch.c) | |
168 | * will compute a different value each time the p_cpu increases | |
169 | * by 4. The cpu usage estimator ramps up quite quickly when | |
170 | * the process is running (linearly), and decays away exponentially, | |
171 | * at a rate which is proportionally slower when the system is | |
172 | * busy. The basic principal is that the system will 90% forget | |
173 | * that a process used a lot of CPU time in 5*loadav seconds. | |
174 | * This causes the system to favor processes which haven't run | |
175 | * much recently, and to round-robin among other processes. | |
176 | */ | |
83be5fac | 177 | if (!noproc) { |
27b91f59 BJ |
178 | p->p_cpticks++; |
179 | if (++p->p_cpu == 0) | |
180 | p->p_cpu--; | |
76b2a182 | 181 | if ((p->p_cpu&3) == 0) { |
27b91f59 BJ |
182 | (void) setpri(p); |
183 | if (p->p_pri >= PUSER) | |
184 | p->p_pri = p->p_usrpri; | |
83be5fac BJ |
185 | } |
186 | } | |
76b2a182 | 187 | |
53a32545 SL |
188 | /* |
189 | * If the alternate clock has not made itself known then | |
190 | * we must gather the statistics. | |
191 | */ | |
192 | if (phz == 0) | |
d293217c WN |
193 | #ifdef i386 |
194 | gatherstats(pc, ISPL(frame.if_cs), frame.if_ppl); | |
195 | #else | |
53a32545 | 196 | gatherstats(pc, ps); |
d293217c | 197 | #endif |
53a32545 | 198 | |
76b2a182 BJ |
199 | /* |
200 | * Increment the time-of-day, and schedule | |
201 | * processing of the callouts at a very low cpu priority, | |
202 | * so we don't keep the relatively high clock interrupt | |
203 | * priority any longer than necessary. | |
204 | */ | |
45e9acec MK |
205 | #ifdef ADJTIME |
206 | if (adjtimedelta == 0) | |
207 | bumptime(&time, tick); | |
208 | else { | |
209 | if (adjtimedelta < 0) { | |
210 | bumptime(&time, tick-ADJ_TICK); | |
211 | adjtimedelta++; | |
212 | } else { | |
213 | bumptime(&time, tick+ADJ_TICK); | |
214 | adjtimedelta--; | |
215 | } | |
216 | } | |
217 | #else | |
4ca0d0d6 | 218 | if (timedelta == 0) |
99e47f6b MK |
219 | BUMPTIME(&time, tick) |
220 | else { | |
221 | register delta; | |
222 | ||
4ca0d0d6 MK |
223 | if (timedelta < 0) { |
224 | delta = tick - tickdelta; | |
225 | timedelta += tickdelta; | |
99e47f6b | 226 | } else { |
4ca0d0d6 MK |
227 | delta = tick + tickdelta; |
228 | timedelta -= tickdelta; | |
99e47f6b MK |
229 | } |
230 | BUMPTIME(&time, delta); | |
231 | } | |
45e9acec | 232 | #endif |
ca6b57a4 | 233 | setsoftclock(); |
f403d99f BJ |
234 | } |
235 | ||
d976d466 | 236 | int dk_ndrive = DK_NDRIVE; |
53a32545 SL |
237 | /* |
238 | * Gather statistics on resource utilization. | |
239 | * | |
240 | * We make a gross assumption: that the system has been in the | |
241 | * state it is in (user state, kernel state, interrupt state, | |
242 | * or idle state) for the entire last time interval, and | |
243 | * update statistics accordingly. | |
244 | */ | |
88a7a62a | 245 | /*ARGSUSED*/ |
d293217c WN |
246 | #ifdef i386 |
247 | #undef pc | |
248 | gatherstats(pc, ps, ppl) | |
249 | #else | |
53a32545 | 250 | gatherstats(pc, ps) |
d293217c | 251 | #endif |
53a32545 SL |
252 | caddr_t pc; |
253 | int ps; | |
254 | { | |
9c5cfb8b | 255 | register int cpstate, s; |
53a32545 SL |
256 | |
257 | /* | |
258 | * Determine what state the cpu is in. | |
259 | */ | |
d293217c WN |
260 | #ifdef i386 |
261 | if (ps == SEL_UPL) { | |
262 | #else | |
53a32545 | 263 | if (USERMODE(ps)) { |
d293217c | 264 | #endif |
53a32545 SL |
265 | /* |
266 | * CPU was in user state. | |
267 | */ | |
268 | if (u.u_procp->p_nice > NZERO) | |
269 | cpstate = CP_NICE; | |
270 | else | |
271 | cpstate = CP_USER; | |
272 | } else { | |
273 | /* | |
274 | * CPU was in system state. If profiling kernel | |
0b355a6e JB |
275 | * increment a counter. If no process is running |
276 | * then this is a system tick if we were running | |
277 | * at a non-zero IPL (in a driver). If a process is running, | |
278 | * then we charge it with system time even if we were | |
279 | * at a non-zero IPL, since the system often runs | |
280 | * this way during processing of system calls. | |
281 | * This is approximate, but the lack of true interval | |
282 | * timers makes doing anything else difficult. | |
53a32545 SL |
283 | */ |
284 | cpstate = CP_SYS; | |
d293217c WN |
285 | #if defined(i386) |
286 | if (noproc && ps == 0) | |
287 | #else | |
53a32545 | 288 | if (noproc && BASEPRI(ps)) |
d293217c | 289 | #endif |
53a32545 SL |
290 | cpstate = CP_IDLE; |
291 | #ifdef GPROF | |
292 | s = pc - s_lowpc; | |
293 | if (profiling < 2 && s < s_textsize) | |
294 | kcount[s / (HISTFRACTION * sizeof (*kcount))]++; | |
295 | #endif | |
296 | } | |
297 | /* | |
298 | * We maintain statistics shown by user-level statistics | |
299 | * programs: the amount of time in each cpu state, and | |
300 | * the amount of time each of DK_NDRIVE ``drives'' is busy. | |
301 | */ | |
302 | cp_time[cpstate]++; | |
303 | for (s = 0; s < DK_NDRIVE; s++) | |
fb1db32c | 304 | if (dk_busy&(1<<s)) |
53a32545 SL |
305 | dk_time[s]++; |
306 | } | |
307 | ||
76b2a182 BJ |
308 | /* |
309 | * Software priority level clock interrupt. | |
310 | * Run periodic events from timeout queue. | |
311 | */ | |
260ea681 | 312 | /*ARGSUSED*/ |
d293217c WN |
313 | #ifdef i386 |
314 | softclock(frame) | |
315 | struct intrframe frame; | |
316 | #define pc frame.if_eip | |
317 | #else | |
f403d99f | 318 | softclock(pc, ps) |
4512b9a4 | 319 | caddr_t pc; |
460ab27f | 320 | int ps; |
d293217c | 321 | #endif |
f403d99f | 322 | { |
f403d99f | 323 | |
27b91f59 | 324 | for (;;) { |
76b2a182 BJ |
325 | register struct callout *p1; |
326 | register caddr_t arg; | |
327 | register int (*func)(); | |
328 | register int a, s; | |
329 | ||
9c5cfb8b | 330 | s = splhigh(); |
27b91f59 BJ |
331 | if ((p1 = calltodo.c_next) == 0 || p1->c_time > 0) { |
332 | splx(s); | |
333 | break; | |
f403d99f | 334 | } |
76b2a182 | 335 | arg = p1->c_arg; func = p1->c_func; a = p1->c_time; |
27b91f59 | 336 | calltodo.c_next = p1->c_next; |
27b91f59 BJ |
337 | p1->c_next = callfree; |
338 | callfree = p1; | |
4f083fd7 | 339 | splx(s); |
d01b68d6 | 340 | (*func)(arg, a); |
f403d99f | 341 | } |
877ef342 | 342 | /* |
db1f1262 SL |
343 | * If trapped user-mode and profiling, give it |
344 | * a profiling tick. | |
877ef342 | 345 | */ |
d293217c WN |
346 | #ifdef i386 |
347 | if (ISPL(frame.if_cs) == SEL_UPL) { | |
348 | #else | |
db1f1262 | 349 | if (USERMODE(ps)) { |
d293217c | 350 | #endif |
db1f1262 SL |
351 | register struct proc *p = u.u_procp; |
352 | ||
353 | if (u.u_prof.pr_scale) { | |
354 | p->p_flag |= SOWEUPC; | |
355 | aston(); | |
356 | } | |
db1f1262 SL |
357 | /* |
358 | * Check to see if process has accumulated | |
359 | * more than 10 minutes of user time. If so | |
360 | * reduce priority to give others a chance. | |
361 | */ | |
362 | if (p->p_uid && p->p_nice == NZERO && | |
53fbb3b3 | 363 | p->p_utime.tv_sec > 10 * 60) { |
db1f1262 SL |
364 | p->p_nice = NZERO+4; |
365 | (void) setpri(p); | |
366 | p->p_pri = p->p_usrpri; | |
367 | } | |
877ef342 | 368 | } |
83be5fac BJ |
369 | } |
370 | ||
88a7a62a SL |
371 | /* |
372 | * Arrange that (*fun)(arg) is called in t/hz seconds. | |
83be5fac | 373 | */ |
88a7a62a | 374 | timeout(fun, arg, t) |
4512b9a4 BJ |
375 | int (*fun)(); |
376 | caddr_t arg; | |
88a7a62a | 377 | register int t; |
83be5fac | 378 | { |
c4710996 | 379 | register struct callout *p1, *p2, *pnew; |
9c5cfb8b | 380 | register int s = splhigh(); |
83be5fac | 381 | |
ba96129b | 382 | if (t <= 0) |
88a7a62a | 383 | t = 1; |
c4710996 BJ |
384 | pnew = callfree; |
385 | if (pnew == NULL) | |
386 | panic("timeout table overflow"); | |
387 | callfree = pnew->c_next; | |
388 | pnew->c_arg = arg; | |
389 | pnew->c_func = fun; | |
390 | for (p1 = &calltodo; (p2 = p1->c_next) && p2->c_time < t; p1 = p2) | |
d45b61eb SL |
391 | if (p2->c_time > 0) |
392 | t -= p2->c_time; | |
c4710996 BJ |
393 | p1->c_next = pnew; |
394 | pnew->c_next = p2; | |
395 | pnew->c_time = t; | |
396 | if (p2) | |
397 | p2->c_time -= t; | |
83be5fac BJ |
398 | splx(s); |
399 | } | |
1fa9ff62 SL |
400 | |
401 | /* | |
402 | * untimeout is called to remove a function timeout call | |
403 | * from the callout structure. | |
404 | */ | |
27b91f59 | 405 | untimeout(fun, arg) |
1fa9ff62 SL |
406 | int (*fun)(); |
407 | caddr_t arg; | |
408 | { | |
1fa9ff62 SL |
409 | register struct callout *p1, *p2; |
410 | register int s; | |
411 | ||
9c5cfb8b | 412 | s = splhigh(); |
1fa9ff62 SL |
413 | for (p1 = &calltodo; (p2 = p1->c_next) != 0; p1 = p2) { |
414 | if (p2->c_func == fun && p2->c_arg == arg) { | |
d01b68d6 | 415 | if (p2->c_next && p2->c_time > 0) |
1fa9ff62 SL |
416 | p2->c_next->c_time += p2->c_time; |
417 | p1->c_next = p2->c_next; | |
418 | p2->c_next = callfree; | |
419 | callfree = p2; | |
420 | break; | |
421 | } | |
422 | } | |
423 | splx(s); | |
424 | } | |
d01b68d6 | 425 | |
76b2a182 BJ |
426 | /* |
427 | * Compute number of hz until specified time. | |
428 | * Used to compute third argument to timeout() from an | |
429 | * absolute time. | |
430 | */ | |
d01b68d6 BJ |
431 | hzto(tv) |
432 | struct timeval *tv; | |
433 | { | |
76b2a182 BJ |
434 | register long ticks; |
435 | register long sec; | |
9c5cfb8b | 436 | int s = splhigh(); |
d01b68d6 | 437 | |
76b2a182 BJ |
438 | /* |
439 | * If number of milliseconds will fit in 32 bit arithmetic, | |
440 | * then compute number of milliseconds to time and scale to | |
441 | * ticks. Otherwise just compute number of hz in time, rounding | |
442 | * times greater than representible to maximum value. | |
443 | * | |
444 | * Delta times less than 25 days can be computed ``exactly''. | |
445 | * Maximum value for any timeout in 10ms ticks is 250 days. | |
446 | */ | |
447 | sec = tv->tv_sec - time.tv_sec; | |
448 | if (sec <= 0x7fffffff / 1000 - 1000) | |
449 | ticks = ((tv->tv_sec - time.tv_sec) * 1000 + | |
450 | (tv->tv_usec - time.tv_usec) / 1000) / (tick / 1000); | |
451 | else if (sec <= 0x7fffffff / hz) | |
452 | ticks = sec * hz; | |
453 | else | |
454 | ticks = 0x7fffffff; | |
d01b68d6 BJ |
455 | splx(s); |
456 | return (ticks); | |
457 | } | |
88a7a62a | 458 | |
330a6ff9 KM |
459 | /* ARGSUSED */ |
460 | profil(p, uap, retval) | |
461 | struct proc *p; | |
462 | register struct args { | |
88a7a62a SL |
463 | short *bufbase; |
464 | unsigned bufsize; | |
465 | unsigned pcoffset; | |
466 | unsigned pcscale; | |
330a6ff9 KM |
467 | } *uap; |
468 | int *retval; | |
469 | { | |
88a7a62a SL |
470 | register struct uprof *upp = &u.u_prof; |
471 | ||
472 | upp->pr_base = uap->bufbase; | |
473 | upp->pr_size = uap->bufsize; | |
474 | upp->pr_off = uap->pcoffset; | |
475 | upp->pr_scale = uap->pcscale; | |
d9c2f47f | 476 | return (0); |
88a7a62a | 477 | } |