kern_synch.c revision 1.41 1 1.41 christos /* $NetBSD: kern_synch.c,v 1.41 1996/10/13 02:32:35 christos Exp $ */
2 1.26 cgd
3 1.26 cgd /*-
4 1.26 cgd * Copyright (c) 1982, 1986, 1990, 1991, 1993
5 1.26 cgd * The Regents of the University of California. All rights reserved.
6 1.26 cgd * (c) UNIX System Laboratories, Inc.
7 1.26 cgd * All or some portions of this file are derived from material licensed
8 1.26 cgd * to the University of California by American Telephone and Telegraph
9 1.26 cgd * Co. or Unix System Laboratories, Inc. and are reproduced herein with
10 1.26 cgd * the permission of UNIX System Laboratories, Inc.
11 1.26 cgd *
12 1.26 cgd * Redistribution and use in source and binary forms, with or without
13 1.26 cgd * modification, are permitted provided that the following conditions
14 1.26 cgd * are met:
15 1.26 cgd * 1. Redistributions of source code must retain the above copyright
16 1.26 cgd * notice, this list of conditions and the following disclaimer.
17 1.26 cgd * 2. Redistributions in binary form must reproduce the above copyright
18 1.26 cgd * notice, this list of conditions and the following disclaimer in the
19 1.26 cgd * documentation and/or other materials provided with the distribution.
20 1.26 cgd * 3. All advertising materials mentioning features or use of this software
21 1.26 cgd * must display the following acknowledgement:
22 1.26 cgd * This product includes software developed by the University of
23 1.26 cgd * California, Berkeley and its contributors.
24 1.26 cgd * 4. Neither the name of the University nor the names of its contributors
25 1.26 cgd * may be used to endorse or promote products derived from this software
26 1.26 cgd * without specific prior written permission.
27 1.26 cgd *
28 1.26 cgd * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
29 1.26 cgd * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
30 1.26 cgd * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
31 1.26 cgd * ARE DISCLAIMED. IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
32 1.26 cgd * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
33 1.26 cgd * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
34 1.26 cgd * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
35 1.26 cgd * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
36 1.26 cgd * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
37 1.26 cgd * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
38 1.26 cgd * SUCH DAMAGE.
39 1.26 cgd *
40 1.26 cgd * @(#)kern_synch.c 8.6 (Berkeley) 1/21/94
41 1.26 cgd */
42 1.26 cgd
43 1.26 cgd #include <sys/param.h>
44 1.26 cgd #include <sys/systm.h>
45 1.26 cgd #include <sys/proc.h>
46 1.26 cgd #include <sys/kernel.h>
47 1.26 cgd #include <sys/buf.h>
48 1.26 cgd #include <sys/signalvar.h>
49 1.26 cgd #include <sys/resourcevar.h>
50 1.34 christos #include <vm/vm.h>
51 1.26 cgd #ifdef KTRACE
52 1.26 cgd #include <sys/ktrace.h>
53 1.26 cgd #endif
54 1.26 cgd
55 1.26 cgd #include <machine/cpu.h>
56 1.34 christos
57 1.26 cgd u_char curpriority; /* usrpri of curproc */
58 1.26 cgd int lbolt; /* once a second sleep address */
59 1.26 cgd
60 1.34 christos void roundrobin __P((void *));
61 1.34 christos void schedcpu __P((void *));
62 1.34 christos void updatepri __P((struct proc *));
63 1.34 christos void endtsleep __P((void *));
64 1.34 christos
65 1.26 cgd /*
66 1.26 cgd * Force switch among equal priority processes every 100ms.
67 1.26 cgd */
68 1.26 cgd /* ARGSUSED */
69 1.26 cgd void
70 1.26 cgd roundrobin(arg)
71 1.26 cgd void *arg;
72 1.26 cgd {
73 1.26 cgd
74 1.26 cgd need_resched();
75 1.26 cgd timeout(roundrobin, NULL, hz / 10);
76 1.26 cgd }
77 1.26 cgd
78 1.26 cgd /*
79 1.26 cgd * Constants for digital decay and forget:
80 1.26 cgd * 90% of (p_estcpu) usage in 5 * loadav time
81 1.26 cgd * 95% of (p_pctcpu) usage in 60 seconds (load insensitive)
82 1.26 cgd * Note that, as ps(1) mentions, this can let percentages
83 1.26 cgd * total over 100% (I've seen 137.9% for 3 processes).
84 1.26 cgd *
85 1.26 cgd * Note that hardclock updates p_estcpu and p_cpticks independently.
86 1.26 cgd *
87 1.26 cgd * We wish to decay away 90% of p_estcpu in (5 * loadavg) seconds.
88 1.26 cgd * That is, the system wants to compute a value of decay such
89 1.26 cgd * that the following for loop:
90 1.26 cgd * for (i = 0; i < (5 * loadavg); i++)
91 1.26 cgd * p_estcpu *= decay;
92 1.26 cgd * will compute
93 1.26 cgd * p_estcpu *= 0.1;
94 1.26 cgd * for all values of loadavg:
95 1.26 cgd *
96 1.26 cgd * Mathematically this loop can be expressed by saying:
97 1.26 cgd * decay ** (5 * loadavg) ~= .1
98 1.26 cgd *
99 1.26 cgd * The system computes decay as:
100 1.26 cgd * decay = (2 * loadavg) / (2 * loadavg + 1)
101 1.26 cgd *
102 1.26 cgd * We wish to prove that the system's computation of decay
103 1.26 cgd * will always fulfill the equation:
104 1.26 cgd * decay ** (5 * loadavg) ~= .1
105 1.26 cgd *
106 1.26 cgd * If we compute b as:
107 1.26 cgd * b = 2 * loadavg
108 1.26 cgd * then
109 1.26 cgd * decay = b / (b + 1)
110 1.26 cgd *
111 1.26 cgd * We now need to prove two things:
112 1.26 cgd * 1) Given factor ** (5 * loadavg) ~= .1, prove factor == b/(b+1)
113 1.26 cgd * 2) Given b/(b+1) ** power ~= .1, prove power == (5 * loadavg)
114 1.26 cgd *
115 1.26 cgd * Facts:
116 1.26 cgd * For x close to zero, exp(x) =~ 1 + x, since
117 1.26 cgd * exp(x) = 0! + x**1/1! + x**2/2! + ... .
118 1.26 cgd * therefore exp(-1/b) =~ 1 - (1/b) = (b-1)/b.
119 1.26 cgd * For x close to zero, ln(1+x) =~ x, since
120 1.26 cgd * ln(1+x) = x - x**2/2 + x**3/3 - ... -1 < x < 1
121 1.26 cgd * therefore ln(b/(b+1)) = ln(1 - 1/(b+1)) =~ -1/(b+1).
122 1.26 cgd * ln(.1) =~ -2.30
123 1.26 cgd *
124 1.26 cgd * Proof of (1):
125 1.26 cgd * Solve (factor)**(power) =~ .1 given power (5*loadav):
126 1.26 cgd * solving for factor,
127 1.26 cgd * ln(factor) =~ (-2.30/5*loadav), or
128 1.26 cgd * factor =~ exp(-1/((5/2.30)*loadav)) =~ exp(-1/(2*loadav)) =
129 1.26 cgd * exp(-1/b) =~ (b-1)/b =~ b/(b+1). QED
130 1.26 cgd *
131 1.26 cgd * Proof of (2):
132 1.26 cgd * Solve (factor)**(power) =~ .1 given factor == (b/(b+1)):
133 1.26 cgd * solving for power,
134 1.26 cgd * power*ln(b/(b+1)) =~ -2.30, or
135 1.26 cgd * power =~ 2.3 * (b + 1) = 4.6*loadav + 2.3 =~ 5*loadav. QED
136 1.26 cgd *
137 1.26 cgd * Actual power values for the implemented algorithm are as follows:
138 1.26 cgd * loadav: 1 2 3 4
139 1.26 cgd * power: 5.68 10.32 14.94 19.55
140 1.26 cgd */
141 1.26 cgd
142 1.26 cgd /* calculations for digital decay to forget 90% of usage in 5*loadav sec */
143 1.26 cgd #define loadfactor(loadav) (2 * (loadav))
144 1.26 cgd #define decay_cpu(loadfac, cpu) (((loadfac) * (cpu)) / ((loadfac) + FSCALE))
145 1.26 cgd
146 1.26 cgd /* decay 95% of `p_pctcpu' in 60 seconds; see CCPU_SHIFT before changing */
147 1.26 cgd fixpt_t ccpu = 0.95122942450071400909 * FSCALE; /* exp(-1/20) */
148 1.26 cgd
149 1.26 cgd /*
150 1.26 cgd * If `ccpu' is not equal to `exp(-1/20)' and you still want to use the
151 1.26 cgd * faster/more-accurate formula, you'll have to estimate CCPU_SHIFT below
152 1.26 cgd * and possibly adjust FSHIFT in "param.h" so that (FSHIFT >= CCPU_SHIFT).
153 1.26 cgd *
154 1.26 cgd * To estimate CCPU_SHIFT for exp(-1/20), the following formula was used:
155 1.26 cgd * 1 - exp(-1/20) ~= 0.0487 ~= 0.0488 == 1 (fixed pt, *11* bits).
156 1.26 cgd *
157 1.26 cgd * If you dont want to bother with the faster/more-accurate formula, you
158 1.26 cgd * can set CCPU_SHIFT to (FSHIFT + 1) which will use a slower/less-accurate
159 1.26 cgd * (more general) method of calculating the %age of CPU used by a process.
160 1.26 cgd */
161 1.26 cgd #define CCPU_SHIFT 11
162 1.26 cgd
163 1.26 cgd /*
164 1.26 cgd * Recompute process priorities, every hz ticks.
165 1.26 cgd */
166 1.26 cgd /* ARGSUSED */
167 1.26 cgd void
168 1.26 cgd schedcpu(arg)
169 1.26 cgd void *arg;
170 1.26 cgd {
171 1.26 cgd register fixpt_t loadfac = loadfactor(averunnable.ldavg[0]);
172 1.26 cgd register struct proc *p;
173 1.26 cgd register int s;
174 1.26 cgd register unsigned int newcpu;
175 1.26 cgd
176 1.26 cgd wakeup((caddr_t)&lbolt);
177 1.27 mycroft for (p = allproc.lh_first; p != 0; p = p->p_list.le_next) {
178 1.26 cgd /*
179 1.26 cgd * Increment time in/out of memory and sleep time
180 1.26 cgd * (if sleeping). We ignore overflow; with 16-bit int's
181 1.26 cgd * (remember them?) overflow takes 45 days.
182 1.26 cgd */
183 1.26 cgd p->p_swtime++;
184 1.26 cgd if (p->p_stat == SSLEEP || p->p_stat == SSTOP)
185 1.26 cgd p->p_slptime++;
186 1.26 cgd p->p_pctcpu = (p->p_pctcpu * ccpu) >> FSHIFT;
187 1.26 cgd /*
188 1.26 cgd * If the process has slept the entire second,
189 1.26 cgd * stop recalculating its priority until it wakes up.
190 1.26 cgd */
191 1.26 cgd if (p->p_slptime > 1)
192 1.26 cgd continue;
193 1.26 cgd s = splstatclock(); /* prevent state changes */
194 1.26 cgd /*
195 1.26 cgd * p_pctcpu is only for ps.
196 1.26 cgd */
197 1.26 cgd #if (FSHIFT >= CCPU_SHIFT)
198 1.26 cgd p->p_pctcpu += (hz == 100)?
199 1.26 cgd ((fixpt_t) p->p_cpticks) << (FSHIFT - CCPU_SHIFT):
200 1.26 cgd 100 * (((fixpt_t) p->p_cpticks)
201 1.26 cgd << (FSHIFT - CCPU_SHIFT)) / hz;
202 1.26 cgd #else
203 1.26 cgd p->p_pctcpu += ((FSCALE - ccpu) *
204 1.26 cgd (p->p_cpticks * FSCALE / hz)) >> FSHIFT;
205 1.26 cgd #endif
206 1.26 cgd p->p_cpticks = 0;
207 1.39 ws newcpu = (u_int)decay_cpu(loadfac, p->p_estcpu)
208 1.39 ws + p->p_nice - NZERO;
209 1.26 cgd p->p_estcpu = min(newcpu, UCHAR_MAX);
210 1.26 cgd resetpriority(p);
211 1.26 cgd if (p->p_priority >= PUSER) {
212 1.26 cgd #define PPQ (128 / NQS) /* priorities per queue */
213 1.26 cgd if ((p != curproc) &&
214 1.26 cgd p->p_stat == SRUN &&
215 1.26 cgd (p->p_flag & P_INMEM) &&
216 1.26 cgd (p->p_priority / PPQ) != (p->p_usrpri / PPQ)) {
217 1.26 cgd remrq(p);
218 1.26 cgd p->p_priority = p->p_usrpri;
219 1.26 cgd setrunqueue(p);
220 1.26 cgd } else
221 1.26 cgd p->p_priority = p->p_usrpri;
222 1.26 cgd }
223 1.26 cgd splx(s);
224 1.26 cgd }
225 1.26 cgd vmmeter();
226 1.26 cgd if (bclnlist != NULL)
227 1.26 cgd wakeup((caddr_t)pageproc);
228 1.26 cgd timeout(schedcpu, (void *)0, hz);
229 1.26 cgd }
230 1.26 cgd
231 1.26 cgd /*
232 1.26 cgd * Recalculate the priority of a process after it has slept for a while.
233 1.26 cgd * For all load averages >= 1 and max p_estcpu of 255, sleeping for at
234 1.26 cgd * least six times the loadfactor will decay p_estcpu to zero.
235 1.26 cgd */
236 1.26 cgd void
237 1.26 cgd updatepri(p)
238 1.26 cgd register struct proc *p;
239 1.26 cgd {
240 1.26 cgd register unsigned int newcpu = p->p_estcpu;
241 1.26 cgd register fixpt_t loadfac = loadfactor(averunnable.ldavg[0]);
242 1.26 cgd
243 1.26 cgd if (p->p_slptime > 5 * loadfac)
244 1.26 cgd p->p_estcpu = 0;
245 1.26 cgd else {
246 1.26 cgd p->p_slptime--; /* the first time was done in schedcpu */
247 1.26 cgd while (newcpu && --p->p_slptime)
248 1.26 cgd newcpu = (int) decay_cpu(loadfac, newcpu);
249 1.26 cgd p->p_estcpu = min(newcpu, UCHAR_MAX);
250 1.26 cgd }
251 1.26 cgd resetpriority(p);
252 1.26 cgd }
253 1.26 cgd
254 1.26 cgd /*
255 1.26 cgd * We're only looking at 7 bits of the address; everything is
256 1.26 cgd * aligned to 4, lots of things are aligned to greater powers
257 1.26 cgd * of 2. Shift right by 8, i.e. drop the bottom 256 worth.
258 1.26 cgd */
259 1.26 cgd #define TABLESIZE 128
260 1.30 cgd #define LOOKUP(x) (((long)(x) >> 8) & (TABLESIZE - 1))
261 1.26 cgd struct slpque {
262 1.26 cgd struct proc *sq_head;
263 1.26 cgd struct proc **sq_tailp;
264 1.26 cgd } slpque[TABLESIZE];
265 1.26 cgd
266 1.26 cgd /*
267 1.26 cgd * During autoconfiguration or after a panic, a sleep will simply
268 1.26 cgd * lower the priority briefly to allow interrupts, then return.
269 1.26 cgd * The priority to be used (safepri) is machine-dependent, thus this
270 1.26 cgd * value is initialized and maintained in the machine-dependent layers.
271 1.26 cgd * This priority will typically be 0, or the lowest priority
272 1.26 cgd * that is safe for use on the interrupt stack; it can be made
273 1.26 cgd * higher to block network software interrupts after panics.
274 1.26 cgd */
275 1.26 cgd int safepri;
276 1.26 cgd
277 1.26 cgd /*
278 1.26 cgd * General sleep call. Suspends the current process until a wakeup is
279 1.26 cgd * performed on the specified identifier. The process will then be made
280 1.26 cgd * runnable with the specified priority. Sleeps at most timo/hz seconds
281 1.26 cgd * (0 means no timeout). If pri includes PCATCH flag, signals are checked
282 1.26 cgd * before and after sleeping, else signals are not checked. Returns 0 if
283 1.26 cgd * awakened, EWOULDBLOCK if the timeout expires. If PCATCH is set and a
284 1.26 cgd * signal needs to be delivered, ERESTART is returned if the current system
285 1.26 cgd * call should be restarted if possible, and EINTR is returned if the system
286 1.26 cgd * call should be interrupted by the signal (return EINTR).
287 1.26 cgd */
288 1.26 cgd int
289 1.26 cgd tsleep(ident, priority, wmesg, timo)
290 1.26 cgd void *ident;
291 1.26 cgd int priority, timo;
292 1.26 cgd char *wmesg;
293 1.26 cgd {
294 1.26 cgd register struct proc *p = curproc;
295 1.26 cgd register struct slpque *qp;
296 1.26 cgd register s;
297 1.26 cgd int sig, catch = priority & PCATCH;
298 1.26 cgd extern int cold;
299 1.26 cgd void endtsleep __P((void *));
300 1.26 cgd
301 1.26 cgd #ifdef KTRACE
302 1.26 cgd if (KTRPOINT(p, KTR_CSW))
303 1.26 cgd ktrcsw(p->p_tracep, 1, 0);
304 1.26 cgd #endif
305 1.26 cgd s = splhigh();
306 1.26 cgd if (cold || panicstr) {
307 1.26 cgd /*
308 1.26 cgd * After a panic, or during autoconfiguration,
309 1.26 cgd * just give interrupts a chance, then just return;
310 1.26 cgd * don't run any other procs or panic below,
311 1.26 cgd * in case this is the idle process and already asleep.
312 1.26 cgd */
313 1.26 cgd splx(safepri);
314 1.26 cgd splx(s);
315 1.26 cgd return (0);
316 1.26 cgd }
317 1.26 cgd #ifdef DIAGNOSTIC
318 1.26 cgd if (ident == NULL || p->p_stat != SRUN || p->p_back)
319 1.26 cgd panic("tsleep");
320 1.26 cgd #endif
321 1.26 cgd p->p_wchan = ident;
322 1.26 cgd p->p_wmesg = wmesg;
323 1.26 cgd p->p_slptime = 0;
324 1.26 cgd p->p_priority = priority & PRIMASK;
325 1.26 cgd qp = &slpque[LOOKUP(ident)];
326 1.26 cgd if (qp->sq_head == 0)
327 1.26 cgd qp->sq_head = p;
328 1.26 cgd else
329 1.26 cgd *qp->sq_tailp = p;
330 1.26 cgd *(qp->sq_tailp = &p->p_forw) = 0;
331 1.26 cgd if (timo)
332 1.26 cgd timeout(endtsleep, (void *)p, timo);
333 1.26 cgd /*
334 1.26 cgd * We put ourselves on the sleep queue and start our timeout
335 1.26 cgd * before calling CURSIG, as we could stop there, and a wakeup
336 1.26 cgd * or a SIGCONT (or both) could occur while we were stopped.
337 1.26 cgd * A SIGCONT would cause us to be marked as SSLEEP
338 1.26 cgd * without resuming us, thus we must be ready for sleep
339 1.26 cgd * when CURSIG is called. If the wakeup happens while we're
340 1.26 cgd * stopped, p->p_wchan will be 0 upon return from CURSIG.
341 1.26 cgd */
342 1.26 cgd if (catch) {
343 1.26 cgd p->p_flag |= P_SINTR;
344 1.34 christos if ((sig = CURSIG(p)) != 0) {
345 1.26 cgd if (p->p_wchan)
346 1.26 cgd unsleep(p);
347 1.26 cgd p->p_stat = SRUN;
348 1.26 cgd goto resume;
349 1.26 cgd }
350 1.26 cgd if (p->p_wchan == 0) {
351 1.26 cgd catch = 0;
352 1.26 cgd goto resume;
353 1.26 cgd }
354 1.26 cgd } else
355 1.26 cgd sig = 0;
356 1.26 cgd p->p_stat = SSLEEP;
357 1.26 cgd p->p_stats->p_ru.ru_nvcsw++;
358 1.26 cgd mi_switch();
359 1.26 cgd #ifdef DDB
360 1.26 cgd /* handy breakpoint location after process "wakes" */
361 1.26 cgd asm(".globl bpendtsleep ; bpendtsleep:");
362 1.26 cgd #endif
363 1.26 cgd resume:
364 1.26 cgd curpriority = p->p_usrpri;
365 1.26 cgd splx(s);
366 1.26 cgd p->p_flag &= ~P_SINTR;
367 1.26 cgd if (p->p_flag & P_TIMEOUT) {
368 1.26 cgd p->p_flag &= ~P_TIMEOUT;
369 1.26 cgd if (sig == 0) {
370 1.26 cgd #ifdef KTRACE
371 1.26 cgd if (KTRPOINT(p, KTR_CSW))
372 1.26 cgd ktrcsw(p->p_tracep, 0, 0);
373 1.26 cgd #endif
374 1.26 cgd return (EWOULDBLOCK);
375 1.26 cgd }
376 1.26 cgd } else if (timo)
377 1.26 cgd untimeout(endtsleep, (void *)p);
378 1.34 christos if (catch && (sig != 0 || (sig = CURSIG(p)) != 0)) {
379 1.26 cgd #ifdef KTRACE
380 1.26 cgd if (KTRPOINT(p, KTR_CSW))
381 1.26 cgd ktrcsw(p->p_tracep, 0, 0);
382 1.26 cgd #endif
383 1.26 cgd if (p->p_sigacts->ps_sigintr & sigmask(sig))
384 1.26 cgd return (EINTR);
385 1.26 cgd return (ERESTART);
386 1.26 cgd }
387 1.26 cgd #ifdef KTRACE
388 1.26 cgd if (KTRPOINT(p, KTR_CSW))
389 1.26 cgd ktrcsw(p->p_tracep, 0, 0);
390 1.26 cgd #endif
391 1.26 cgd return (0);
392 1.26 cgd }
393 1.26 cgd
394 1.26 cgd /*
395 1.26 cgd * Implement timeout for tsleep.
396 1.26 cgd * If process hasn't been awakened (wchan non-zero),
397 1.26 cgd * set timeout flag and undo the sleep. If proc
398 1.26 cgd * is stopped, just unsleep so it will remain stopped.
399 1.26 cgd */
400 1.26 cgd void
401 1.26 cgd endtsleep(arg)
402 1.26 cgd void *arg;
403 1.26 cgd {
404 1.26 cgd register struct proc *p;
405 1.26 cgd int s;
406 1.26 cgd
407 1.26 cgd p = (struct proc *)arg;
408 1.26 cgd s = splhigh();
409 1.26 cgd if (p->p_wchan) {
410 1.26 cgd if (p->p_stat == SSLEEP)
411 1.26 cgd setrunnable(p);
412 1.26 cgd else
413 1.26 cgd unsleep(p);
414 1.26 cgd p->p_flag |= P_TIMEOUT;
415 1.26 cgd }
416 1.26 cgd splx(s);
417 1.26 cgd }
418 1.26 cgd
419 1.26 cgd /*
420 1.26 cgd * Short-term, non-interruptable sleep.
421 1.26 cgd */
422 1.26 cgd void
423 1.26 cgd sleep(ident, priority)
424 1.26 cgd void *ident;
425 1.26 cgd int priority;
426 1.26 cgd {
427 1.26 cgd register struct proc *p = curproc;
428 1.26 cgd register struct slpque *qp;
429 1.26 cgd register s;
430 1.26 cgd extern int cold;
431 1.26 cgd
432 1.26 cgd #ifdef DIAGNOSTIC
433 1.26 cgd if (priority > PZERO) {
434 1.41 christos printf("sleep called with priority %d > PZERO, wchan: %p\n",
435 1.26 cgd priority, ident);
436 1.26 cgd panic("old sleep");
437 1.26 cgd }
438 1.26 cgd #endif
439 1.26 cgd s = splhigh();
440 1.26 cgd if (cold || panicstr) {
441 1.26 cgd /*
442 1.26 cgd * After a panic, or during autoconfiguration,
443 1.26 cgd * just give interrupts a chance, then just return;
444 1.26 cgd * don't run any other procs or panic below,
445 1.26 cgd * in case this is the idle process and already asleep.
446 1.26 cgd */
447 1.26 cgd splx(safepri);
448 1.26 cgd splx(s);
449 1.26 cgd return;
450 1.26 cgd }
451 1.26 cgd #ifdef DIAGNOSTIC
452 1.26 cgd if (ident == NULL || p->p_stat != SRUN || p->p_back)
453 1.26 cgd panic("sleep");
454 1.26 cgd #endif
455 1.26 cgd p->p_wchan = ident;
456 1.26 cgd p->p_wmesg = NULL;
457 1.26 cgd p->p_slptime = 0;
458 1.26 cgd p->p_priority = priority;
459 1.26 cgd qp = &slpque[LOOKUP(ident)];
460 1.26 cgd if (qp->sq_head == 0)
461 1.26 cgd qp->sq_head = p;
462 1.26 cgd else
463 1.26 cgd *qp->sq_tailp = p;
464 1.26 cgd *(qp->sq_tailp = &p->p_forw) = 0;
465 1.26 cgd p->p_stat = SSLEEP;
466 1.26 cgd p->p_stats->p_ru.ru_nvcsw++;
467 1.26 cgd #ifdef KTRACE
468 1.26 cgd if (KTRPOINT(p, KTR_CSW))
469 1.26 cgd ktrcsw(p->p_tracep, 1, 0);
470 1.26 cgd #endif
471 1.26 cgd mi_switch();
472 1.26 cgd #ifdef DDB
473 1.26 cgd /* handy breakpoint location after process "wakes" */
474 1.26 cgd asm(".globl bpendsleep ; bpendsleep:");
475 1.26 cgd #endif
476 1.26 cgd #ifdef KTRACE
477 1.26 cgd if (KTRPOINT(p, KTR_CSW))
478 1.26 cgd ktrcsw(p->p_tracep, 0, 0);
479 1.26 cgd #endif
480 1.26 cgd curpriority = p->p_usrpri;
481 1.26 cgd splx(s);
482 1.26 cgd }
483 1.26 cgd
484 1.26 cgd /*
485 1.26 cgd * Remove a process from its wait queue
486 1.26 cgd */
487 1.26 cgd void
488 1.26 cgd unsleep(p)
489 1.26 cgd register struct proc *p;
490 1.26 cgd {
491 1.26 cgd register struct slpque *qp;
492 1.26 cgd register struct proc **hp;
493 1.26 cgd int s;
494 1.26 cgd
495 1.26 cgd s = splhigh();
496 1.26 cgd if (p->p_wchan) {
497 1.26 cgd hp = &(qp = &slpque[LOOKUP(p->p_wchan)])->sq_head;
498 1.26 cgd while (*hp != p)
499 1.26 cgd hp = &(*hp)->p_forw;
500 1.26 cgd *hp = p->p_forw;
501 1.26 cgd if (qp->sq_tailp == &p->p_forw)
502 1.26 cgd qp->sq_tailp = hp;
503 1.26 cgd p->p_wchan = 0;
504 1.26 cgd }
505 1.26 cgd splx(s);
506 1.26 cgd }
507 1.26 cgd
508 1.26 cgd /*
509 1.26 cgd * Make all processes sleeping on the specified identifier runnable.
510 1.26 cgd */
511 1.26 cgd void
512 1.26 cgd wakeup(ident)
513 1.26 cgd register void *ident;
514 1.26 cgd {
515 1.26 cgd register struct slpque *qp;
516 1.26 cgd register struct proc *p, **q;
517 1.26 cgd int s;
518 1.26 cgd
519 1.26 cgd s = splhigh();
520 1.26 cgd qp = &slpque[LOOKUP(ident)];
521 1.26 cgd restart:
522 1.34 christos for (q = &qp->sq_head; (p = *q) != NULL; ) {
523 1.26 cgd #ifdef DIAGNOSTIC
524 1.34 christos if (p->p_back || (p->p_stat != SSLEEP && p->p_stat != SSTOP))
525 1.26 cgd panic("wakeup");
526 1.26 cgd #endif
527 1.26 cgd if (p->p_wchan == ident) {
528 1.26 cgd p->p_wchan = 0;
529 1.26 cgd *q = p->p_forw;
530 1.26 cgd if (qp->sq_tailp == &p->p_forw)
531 1.26 cgd qp->sq_tailp = q;
532 1.26 cgd if (p->p_stat == SSLEEP) {
533 1.26 cgd /* OPTIMIZED EXPANSION OF setrunnable(p); */
534 1.26 cgd if (p->p_slptime > 1)
535 1.26 cgd updatepri(p);
536 1.26 cgd p->p_slptime = 0;
537 1.26 cgd p->p_stat = SRUN;
538 1.26 cgd if (p->p_flag & P_INMEM)
539 1.26 cgd setrunqueue(p);
540 1.26 cgd /*
541 1.26 cgd * Since curpriority is a user priority,
542 1.26 cgd * p->p_priority is always better than
543 1.26 cgd * curpriority.
544 1.26 cgd */
545 1.26 cgd if ((p->p_flag & P_INMEM) == 0)
546 1.26 cgd wakeup((caddr_t)&proc0);
547 1.26 cgd else
548 1.26 cgd need_resched();
549 1.26 cgd /* END INLINE EXPANSION */
550 1.26 cgd goto restart;
551 1.26 cgd }
552 1.26 cgd } else
553 1.26 cgd q = &p->p_forw;
554 1.26 cgd }
555 1.26 cgd splx(s);
556 1.26 cgd }
557 1.26 cgd
558 1.26 cgd /*
559 1.26 cgd * The machine independent parts of mi_switch().
560 1.26 cgd * Must be called at splstatclock() or higher.
561 1.26 cgd */
562 1.26 cgd void
563 1.26 cgd mi_switch()
564 1.26 cgd {
565 1.26 cgd register struct proc *p = curproc; /* XXX */
566 1.26 cgd register struct rlimit *rlim;
567 1.26 cgd register long s, u;
568 1.26 cgd struct timeval tv;
569 1.26 cgd
570 1.26 cgd /*
571 1.26 cgd * Compute the amount of time during which the current
572 1.26 cgd * process was running, and add that to its total so far.
573 1.26 cgd */
574 1.26 cgd microtime(&tv);
575 1.26 cgd u = p->p_rtime.tv_usec + (tv.tv_usec - runtime.tv_usec);
576 1.26 cgd s = p->p_rtime.tv_sec + (tv.tv_sec - runtime.tv_sec);
577 1.26 cgd if (u < 0) {
578 1.26 cgd u += 1000000;
579 1.26 cgd s--;
580 1.26 cgd } else if (u >= 1000000) {
581 1.26 cgd u -= 1000000;
582 1.26 cgd s++;
583 1.26 cgd }
584 1.26 cgd p->p_rtime.tv_usec = u;
585 1.26 cgd p->p_rtime.tv_sec = s;
586 1.26 cgd
587 1.26 cgd /*
588 1.26 cgd * Check if the process exceeds its cpu resource allocation.
589 1.26 cgd * If over max, kill it. In any case, if it has run for more
590 1.26 cgd * than 10 minutes, reduce priority to give others a chance.
591 1.26 cgd */
592 1.26 cgd rlim = &p->p_rlimit[RLIMIT_CPU];
593 1.26 cgd if (s >= rlim->rlim_cur) {
594 1.26 cgd if (s >= rlim->rlim_max)
595 1.26 cgd psignal(p, SIGKILL);
596 1.26 cgd else {
597 1.26 cgd psignal(p, SIGXCPU);
598 1.26 cgd if (rlim->rlim_cur < rlim->rlim_max)
599 1.26 cgd rlim->rlim_cur += 5;
600 1.26 cgd }
601 1.26 cgd }
602 1.38 explorer if (autonicetime && s > autonicetime && p->p_ucred->cr_uid && p->p_nice == NZERO) {
603 1.39 ws p->p_nice = autoniceval + NZERO;
604 1.26 cgd resetpriority(p);
605 1.26 cgd }
606 1.26 cgd
607 1.26 cgd /*
608 1.26 cgd * Pick a new current process and record its start time.
609 1.26 cgd */
610 1.26 cgd cnt.v_swtch++;
611 1.26 cgd cpu_switch(p);
612 1.26 cgd microtime(&runtime);
613 1.26 cgd }
614 1.26 cgd
615 1.26 cgd /*
616 1.26 cgd * Initialize the (doubly-linked) run queues
617 1.26 cgd * to be empty.
618 1.26 cgd */
619 1.26 cgd void
620 1.26 cgd rqinit()
621 1.26 cgd {
622 1.26 cgd register int i;
623 1.26 cgd
624 1.26 cgd for (i = 0; i < NQS; i++)
625 1.26 cgd qs[i].ph_link = qs[i].ph_rlink = (struct proc *)&qs[i];
626 1.26 cgd }
627 1.26 cgd
628 1.26 cgd /*
629 1.26 cgd * Change process state to be runnable,
630 1.26 cgd * placing it on the run queue if it is in memory,
631 1.26 cgd * and awakening the swapper if it isn't in memory.
632 1.26 cgd */
633 1.26 cgd void
634 1.26 cgd setrunnable(p)
635 1.26 cgd register struct proc *p;
636 1.26 cgd {
637 1.26 cgd register int s;
638 1.26 cgd
639 1.26 cgd s = splhigh();
640 1.26 cgd switch (p->p_stat) {
641 1.26 cgd case 0:
642 1.26 cgd case SRUN:
643 1.26 cgd case SZOMB:
644 1.26 cgd default:
645 1.26 cgd panic("setrunnable");
646 1.26 cgd case SSTOP:
647 1.33 mycroft /*
648 1.33 mycroft * If we're being traced (possibly because someone attached us
649 1.33 mycroft * while we were stopped), check for a signal from the debugger.
650 1.33 mycroft */
651 1.33 mycroft if ((p->p_flag & P_TRACED) != 0 && p->p_xstat != 0)
652 1.33 mycroft p->p_siglist |= sigmask(p->p_xstat);
653 1.26 cgd case SSLEEP:
654 1.26 cgd unsleep(p); /* e.g. when sending signals */
655 1.26 cgd break;
656 1.26 cgd
657 1.26 cgd case SIDL:
658 1.26 cgd break;
659 1.26 cgd }
660 1.26 cgd p->p_stat = SRUN;
661 1.26 cgd if (p->p_flag & P_INMEM)
662 1.26 cgd setrunqueue(p);
663 1.26 cgd splx(s);
664 1.26 cgd if (p->p_slptime > 1)
665 1.26 cgd updatepri(p);
666 1.26 cgd p->p_slptime = 0;
667 1.26 cgd if ((p->p_flag & P_INMEM) == 0)
668 1.26 cgd wakeup((caddr_t)&proc0);
669 1.26 cgd else if (p->p_priority < curpriority)
670 1.26 cgd need_resched();
671 1.26 cgd }
672 1.26 cgd
673 1.26 cgd /*
674 1.26 cgd * Compute the priority of a process when running in user mode.
675 1.26 cgd * Arrange to reschedule if the resulting priority is better
676 1.26 cgd * than that of the current process.
677 1.26 cgd */
678 1.26 cgd void
679 1.26 cgd resetpriority(p)
680 1.26 cgd register struct proc *p;
681 1.26 cgd {
682 1.26 cgd register unsigned int newpriority;
683 1.26 cgd
684 1.39 ws newpriority = PUSER + p->p_estcpu / 4 + 2 * (p->p_nice - NZERO);
685 1.26 cgd newpriority = min(newpriority, MAXPRI);
686 1.26 cgd p->p_usrpri = newpriority;
687 1.26 cgd if (newpriority < curpriority)
688 1.26 cgd need_resched();
689 1.26 cgd }
690 1.26 cgd
691 1.26 cgd #ifdef DDB
692 1.34 christos #include <machine/db_machdep.h>
693 1.34 christos
694 1.34 christos #include <ddb/db_interface.h>
695 1.34 christos #include <ddb/db_output.h>
696 1.34 christos
697 1.28 mycroft void
698 1.28 mycroft db_show_all_procs(addr, haddr, count, modif)
699 1.34 christos db_expr_t addr;
700 1.28 mycroft int haddr;
701 1.34 christos db_expr_t count;
702 1.28 mycroft char *modif;
703 1.28 mycroft {
704 1.28 mycroft int map = modif[0] == 'm';
705 1.28 mycroft int doingzomb = 0;
706 1.28 mycroft struct proc *p, *pp;
707 1.26 cgd
708 1.28 mycroft p = allproc.lh_first;
709 1.28 mycroft db_printf(" pid proc addr %s comm wchan\n",
710 1.29 mycroft map ? "map " : "uid ppid pgrp flag stat em ");
711 1.28 mycroft while (p != 0) {
712 1.28 mycroft pp = p->p_pptr;
713 1.28 mycroft if (p->p_stat) {
714 1.36 christos db_printf("%5d %p %p ",
715 1.28 mycroft p->p_pid, p, p->p_addr);
716 1.28 mycroft if (map)
717 1.36 christos db_printf("%p %s ",
718 1.28 mycroft p->p_vmspace, p->p_comm);
719 1.28 mycroft else
720 1.32 christos db_printf("%3d %5d %5d %06x %d %s %s ",
721 1.28 mycroft p->p_cred->p_ruid, pp ? pp->p_pid : -1,
722 1.28 mycroft p->p_pgrp->pg_id, p->p_flag, p->p_stat,
723 1.32 christos p->p_emul->e_name, p->p_comm);
724 1.28 mycroft if (p->p_wchan) {
725 1.28 mycroft if (p->p_wmesg)
726 1.28 mycroft db_printf("%s ", p->p_wmesg);
727 1.36 christos db_printf("%p", p->p_wchan);
728 1.28 mycroft }
729 1.28 mycroft db_printf("\n");
730 1.28 mycroft }
731 1.28 mycroft p = p->p_list.le_next;
732 1.28 mycroft if (p == 0 && doingzomb == 0) {
733 1.28 mycroft doingzomb = 1;
734 1.28 mycroft p = zombproc.lh_first;
735 1.28 mycroft }
736 1.28 mycroft }
737 1.26 cgd }
738 1.26 cgd #endif
739