1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139
140
141
142
143
144
145
146
147
148
149
150
151
152
153
154
155
156
157
158
159
160
161
162
163
164
165
166
167
168
169
170
171
172
173
174
175
176
177
178
179
180
181
182
183
184
185
186
187
188
189
190
191
192
193
194
195
196
197
198
199
200
201
202
203
204
205
206
207
208
209
210
211
212
213
214
215
216
217
218
219
220
221
222
223
224
225
226
227
228
229
230
231
232
233
234
235
236
237
238
239
240
241
242
243
244
245
246
247
248
249
250
251
252
253
254
255
256
257
258
259
260
261
262
263
264
265
266
267
268
269
270
271
272
273
274
275
276
277
278
279
280
281
282
283
284
285
286
287
288
289
290
291
292
293
294
295
296
297
298
299
300
301
302
303
304
305
306
307
308
309
310
311
312
313
314
315
316
317
318
319
320
321
322
323
324
325
326
327
328
329
330
331
332
333
334
335
336
337
338
339
340
341
342
343
344
345
346
347
348
349
350
351
352
353
354
355
356
357
358
359
360
361
362
363
364
365
366
367
368
369
370
371
372
373
374
375
376
377
378
379
380
381
382
383
384
385
386
387
388
389
390
391
392
393
394
395
396
397
398
399
400
401
402
403
404
405
406
407
408
409
410
411
412
413
414
415
416
417
418
419
420
421
422
423
424
425
426
427
428
429
430
431
432
433
434
435
436
437
438
439
440
441
442
443
444
445
446
447
448
449
450
451
452
453
454
455
456
457
458
459
460
461
462
463
464
465
466
467
468
469
470
471
472
473
474
475
476
477
478
479
480
481
482
483
484
485
486
487
488
489
490
491
492
493
494
495
496
497
498
499
500
501
502
503
504
505
506
507
508
509
510
511
512
513
514
515
516
517
518
519
520
521
522
523
524
525
526
527
528
529
530
531
532
533
534
535
536
537
538
539
540
541
542
543
544
545
546
547
548
549
550
551
552
553
554
555
556
557
558
559
560
561
562
563
564
565
566
567
568
569
570
571
572
573
|
/* $OpenBSD: sched_bsd.c,v 1.25 2011/03/07 07:07:13 guenther Exp $ */
/* $NetBSD: kern_synch.c,v 1.37 1996/04/22 01:38:37 christos Exp $ */
/*-
* Copyright (c) 1982, 1986, 1990, 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. 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_synch.c 8.6 (Berkeley) 1/21/94
*/
#include <sys/param.h>
#include <sys/systm.h>
#include <sys/proc.h>
#include <sys/kernel.h>
#include <sys/buf.h>
#include <sys/signalvar.h>
#include <sys/resourcevar.h>
#include <uvm/uvm_extern.h>
#include <sys/sched.h>
#include <sys/timeout.h>
#ifdef KTRACE
#include <sys/ktrace.h>
#endif
#include <machine/cpu.h>
int lbolt; /* once a second sleep address */
int rrticks_init; /* # of hardclock ticks per roundrobin() */
#ifdef MULTIPROCESSOR
struct __mp_lock sched_lock;
#endif
void scheduler_start(void);
void roundrobin(struct cpu_info *);
void schedcpu(void *);
void updatepri(struct proc *);
void endtsleep(void *);
void
scheduler_start(void)
{
static struct timeout schedcpu_to;
/*
* We avoid polluting the global namespace by keeping the scheduler
* timeouts static in this function.
* We setup the timeouts here and kick schedcpu and roundrobin once to
* make them do their job.
*/
timeout_set(&schedcpu_to, schedcpu, &schedcpu_to);
rrticks_init = hz / 10;
schedcpu(&schedcpu_to);
}
/*
* Force switch among equal priority processes every 100ms.
*/
void
roundrobin(struct cpu_info *ci)
{
struct schedstate_percpu *spc = &ci->ci_schedstate;
spc->spc_rrticks = rrticks_init;
if (ci->ci_curproc != NULL) {
if (spc->spc_schedflags & SPCF_SEENRR) {
/*
* The process has already been through a roundrobin
* without switching and may be hogging the CPU.
* Indicate that the process should yield.
*/
atomic_setbits_int(&spc->spc_schedflags,
SPCF_SHOULDYIELD);
} else {
atomic_setbits_int(&spc->spc_schedflags,
SPCF_SEENRR);
}
}
if (spc->spc_nrun)
need_resched(ci);
}
/*
* Constants for digital decay and forget:
* 90% of (p_estcpu) usage in 5 * loadav time
* 95% of (p_pctcpu) usage in 60 seconds (load insensitive)
* Note that, as ps(1) mentions, this can let percentages
* total over 100% (I've seen 137.9% for 3 processes).
*
* Note that hardclock updates p_estcpu and p_cpticks independently.
*
* We wish to decay away 90% of p_estcpu in (5 * loadavg) seconds.
* That is, the system wants to compute a value of decay such
* that the following for loop:
* for (i = 0; i < (5 * loadavg); i++)
* p_estcpu *= decay;
* will compute
* p_estcpu *= 0.1;
* for all values of loadavg:
*
* Mathematically this loop can be expressed by saying:
* decay ** (5 * loadavg) ~= .1
*
* The system computes decay as:
* decay = (2 * loadavg) / (2 * loadavg + 1)
*
* We wish to prove that the system's computation of decay
* will always fulfill the equation:
* decay ** (5 * loadavg) ~= .1
*
* If we compute b as:
* b = 2 * loadavg
* then
* decay = b / (b + 1)
*
* We now need to prove two things:
* 1) Given factor ** (5 * loadavg) ~= .1, prove factor == b/(b+1)
* 2) Given b/(b+1) ** power ~= .1, prove power == (5 * loadavg)
*
* Facts:
* For x close to zero, exp(x) =~ 1 + x, since
* exp(x) = 0! + x**1/1! + x**2/2! + ... .
* therefore exp(-1/b) =~ 1 - (1/b) = (b-1)/b.
* For x close to zero, ln(1+x) =~ x, since
* ln(1+x) = x - x**2/2 + x**3/3 - ... -1 < x < 1
* therefore ln(b/(b+1)) = ln(1 - 1/(b+1)) =~ -1/(b+1).
* ln(.1) =~ -2.30
*
* Proof of (1):
* Solve (factor)**(power) =~ .1 given power (5*loadav):
* solving for factor,
* ln(factor) =~ (-2.30/5*loadav), or
* factor =~ exp(-1/((5/2.30)*loadav)) =~ exp(-1/(2*loadav)) =
* exp(-1/b) =~ (b-1)/b =~ b/(b+1). QED
*
* Proof of (2):
* Solve (factor)**(power) =~ .1 given factor == (b/(b+1)):
* solving for power,
* power*ln(b/(b+1)) =~ -2.30, or
* power =~ 2.3 * (b + 1) = 4.6*loadav + 2.3 =~ 5*loadav. QED
*
* Actual power values for the implemented algorithm are as follows:
* loadav: 1 2 3 4
* power: 5.68 10.32 14.94 19.55
*/
/* calculations for digital decay to forget 90% of usage in 5*loadav sec */
#define loadfactor(loadav) (2 * (loadav))
#define decay_cpu(loadfac, cpu) (((loadfac) * (cpu)) / ((loadfac) + FSCALE))
/* decay 95% of `p_pctcpu' in 60 seconds; see CCPU_SHIFT before changing */
fixpt_t ccpu = 0.95122942450071400909 * FSCALE; /* exp(-1/20) */
/*
* If `ccpu' is not equal to `exp(-1/20)' and you still want to use the
* faster/more-accurate formula, you'll have to estimate CCPU_SHIFT below
* and possibly adjust FSHIFT in "param.h" so that (FSHIFT >= CCPU_SHIFT).
*
* To estimate CCPU_SHIFT for exp(-1/20), the following formula was used:
* 1 - exp(-1/20) ~= 0.0487 ~= 0.0488 == 1 (fixed pt, *11* bits).
*
* If you don't want to bother with the faster/more-accurate formula, you
* can set CCPU_SHIFT to (FSHIFT + 1) which will use a slower/less-accurate
* (more general) method of calculating the %age of CPU used by a process.
*/
#define CCPU_SHIFT 11
/*
* Recompute process priorities, every second.
*/
void
schedcpu(void *arg)
{
struct timeout *to = (struct timeout *)arg;
fixpt_t loadfac = loadfactor(averunnable.ldavg[0]);
struct proc *p;
int s;
unsigned int newcpu;
int phz;
/*
* If we have a statistics clock, use that to calculate CPU
* time, otherwise revert to using the profiling clock (which,
* in turn, defaults to hz if there is no separate profiling
* clock available)
*/
phz = stathz ? stathz : profhz;
KASSERT(phz);
LIST_FOREACH(p, &allproc, p_list) {
/*
* Increment time in/out of memory and sleep time
* (if sleeping). We ignore overflow; with 16-bit int's
* (remember them?) overflow takes 45 days.
*/
p->p_swtime++;
if (p->p_stat == SSLEEP || p->p_stat == SSTOP)
p->p_slptime++;
p->p_pctcpu = (p->p_pctcpu * ccpu) >> FSHIFT;
/*
* If the process has slept the entire second,
* stop recalculating its priority until it wakes up.
*/
if (p->p_slptime > 1)
continue;
SCHED_LOCK(s);
/*
* p_pctcpu is only for ps.
*/
#if (FSHIFT >= CCPU_SHIFT)
p->p_pctcpu += (phz == 100)?
((fixpt_t) p->p_cpticks) << (FSHIFT - CCPU_SHIFT):
100 * (((fixpt_t) p->p_cpticks)
<< (FSHIFT - CCPU_SHIFT)) / phz;
#else
p->p_pctcpu += ((FSCALE - ccpu) *
(p->p_cpticks * FSCALE / phz)) >> FSHIFT;
#endif
p->p_cpticks = 0;
newcpu = (u_int) decay_cpu(loadfac, p->p_estcpu);
p->p_estcpu = newcpu;
resetpriority(p);
if (p->p_priority >= PUSER) {
if (p->p_stat == SRUN &&
(p->p_priority / SCHED_PPQ) !=
(p->p_usrpri / SCHED_PPQ)) {
remrunqueue(p);
p->p_priority = p->p_usrpri;
setrunqueue(p);
} else
p->p_priority = p->p_usrpri;
}
SCHED_UNLOCK(s);
}
uvm_meter();
wakeup(&lbolt);
timeout_add_sec(to, 1);
}
/*
* Recalculate the priority of a process after it has slept for a while.
* For all load averages >= 1 and max p_estcpu of 255, sleeping for at
* least six times the loadfactor will decay p_estcpu to zero.
*/
void
updatepri(struct proc *p)
{
unsigned int newcpu = p->p_estcpu;
fixpt_t loadfac = loadfactor(averunnable.ldavg[0]);
SCHED_ASSERT_LOCKED();
if (p->p_slptime > 5 * loadfac)
p->p_estcpu = 0;
else {
p->p_slptime--; /* the first time was done in schedcpu */
while (newcpu && --p->p_slptime)
newcpu = (int) decay_cpu(loadfac, newcpu);
p->p_estcpu = newcpu;
}
resetpriority(p);
}
/*
* General yield call. Puts the current process back on its run queue and
* performs a voluntary context switch.
*/
void
yield(void)
{
struct proc *p = curproc;
int s;
SCHED_LOCK(s);
p->p_priority = p->p_usrpri;
p->p_stat = SRUN;
setrunqueue(p);
p->p_stats->p_ru.ru_nvcsw++;
mi_switch();
SCHED_UNLOCK(s);
}
/*
* General preemption call. Puts the current process back on its run queue
* and performs an involuntary context switch. If a process is supplied,
* we switch to that process. Otherwise, we use the normal process selection
* criteria.
*/
void
preempt(struct proc *newp)
{
struct proc *p = curproc;
int s;
/*
* XXX Switching to a specific process is not supported yet.
*/
if (newp != NULL)
panic("preempt: cpu_preempt not yet implemented");
SCHED_LOCK(s);
p->p_priority = p->p_usrpri;
p->p_stat = SRUN;
p->p_cpu = sched_choosecpu(p);
setrunqueue(p);
p->p_stats->p_ru.ru_nivcsw++;
mi_switch();
SCHED_UNLOCK(s);
}
void
mi_switch(void)
{
struct schedstate_percpu *spc = &curcpu()->ci_schedstate;
struct proc *p = curproc;
struct proc *nextproc;
struct rlimit *rlim;
struct timeval tv;
#ifdef MULTIPROCESSOR
int hold_count;
int sched_count;
#endif
assertwaitok();
KASSERT(p->p_stat != SONPROC);
SCHED_ASSERT_LOCKED();
#ifdef MULTIPROCESSOR
/*
* Release the kernel_lock, as we are about to yield the CPU.
*/
sched_count = __mp_release_all_but_one(&sched_lock);
if (p->p_flag & P_BIGLOCK)
hold_count = __mp_release_all(&kernel_lock);
#endif
/*
* Compute the amount of time during which the current
* process was running, and add that to its total so far.
*/
microuptime(&tv);
if (timercmp(&tv, &spc->spc_runtime, <)) {
#if 0
printf("uptime is not monotonic! "
"tv=%lu.%06lu, runtime=%lu.%06lu\n",
tv.tv_sec, tv.tv_usec, spc->spc_runtime.tv_sec,
spc->spc_runtime.tv_usec);
#endif
} else {
timersub(&tv, &spc->spc_runtime, &tv);
timeradd(&p->p_rtime, &tv, &p->p_rtime);
}
/*
* Check if the process exceeds its cpu resource allocation.
* If over max, kill it.
*/
rlim = &p->p_rlimit[RLIMIT_CPU];
if ((rlim_t)p->p_rtime.tv_sec >= rlim->rlim_cur) {
if ((rlim_t)p->p_rtime.tv_sec >= rlim->rlim_max) {
psignal(p, SIGKILL);
} else {
psignal(p, SIGXCPU);
if (rlim->rlim_cur < rlim->rlim_max)
rlim->rlim_cur += 5;
}
}
/*
* Process is about to yield the CPU; clear the appropriate
* scheduling flags.
*/
atomic_clearbits_int(&spc->spc_schedflags, SPCF_SWITCHCLEAR);
nextproc = sched_chooseproc();
if (p != nextproc) {
uvmexp.swtch++;
cpu_switchto(p, nextproc);
} else {
p->p_stat = SONPROC;
}
clear_resched(curcpu());
SCHED_ASSERT_LOCKED();
/*
* To preserve lock ordering, we need to release the sched lock
* and grab it after we grab the big lock.
* In the future, when the sched lock isn't recursive, we'll
* just release it here.
*/
#ifdef MULTIPROCESSOR
__mp_unlock(&sched_lock);
#endif
SCHED_ASSERT_UNLOCKED();
/*
* We're running again; record our new start time. We might
* be running on a new CPU now, so don't use the cache'd
* schedstate_percpu pointer.
*/
KASSERT(p->p_cpu == curcpu());
microuptime(&p->p_cpu->ci_schedstate.spc_runtime);
#ifdef MULTIPROCESSOR
/*
* Reacquire the kernel_lock now. We do this after we've
* released the scheduler lock to avoid deadlock, and before
* we reacquire the interlock and the scheduler lock.
*/
if (p->p_flag & P_BIGLOCK)
__mp_acquire_count(&kernel_lock, hold_count);
__mp_acquire_count(&sched_lock, sched_count + 1);
#endif
}
static __inline void
resched_proc(struct proc *p, u_char pri)
{
struct cpu_info *ci;
/*
* XXXSMP
* Since p->p_cpu persists across a context switch,
* this gives us *very weak* processor affinity, in
* that we notify the CPU on which the process last
* ran that it should try to switch.
*
* This does not guarantee that the process will run on
* that processor next, because another processor might
* grab it the next time it performs a context switch.
*
* This also does not handle the case where its last
* CPU is running a higher-priority process, but every
* other CPU is running a lower-priority process. There
* are ways to handle this situation, but they're not
* currently very pretty, and we also need to weigh the
* cost of moving a process from one CPU to another.
*
* XXXSMP
* There is also the issue of locking the other CPU's
* sched state, which we currently do not do.
*/
ci = (p->p_cpu != NULL) ? p->p_cpu : curcpu();
if (pri < ci->ci_schedstate.spc_curpriority)
need_resched(ci);
}
/*
* Change process state to be runnable,
* placing it on the run queue if it is in memory,
* and awakening the swapper if it isn't in memory.
*/
void
setrunnable(struct proc *p)
{
SCHED_ASSERT_LOCKED();
switch (p->p_stat) {
case 0:
case SRUN:
case SONPROC:
case SZOMB:
case SDEAD:
case SIDL:
default:
panic("setrunnable");
case SSTOP:
/*
* If we're being traced (possibly because someone attached us
* while we were stopped), check for a signal from the debugger.
*/
if ((p->p_flag & P_TRACED) != 0 && p->p_xstat != 0)
atomic_setbits_int(&p->p_siglist, sigmask(p->p_xstat));
case SSLEEP:
unsleep(p); /* e.g. when sending signals */
break;
}
p->p_stat = SRUN;
p->p_cpu = sched_choosecpu(p);
setrunqueue(p);
if (p->p_slptime > 1)
updatepri(p);
p->p_slptime = 0;
resched_proc(p, p->p_priority);
}
/*
* Compute the priority of a process when running in user mode.
* Arrange to reschedule if the resulting priority is better
* than that of the current process.
*/
void
resetpriority(struct proc *p)
{
unsigned int newpriority;
SCHED_ASSERT_LOCKED();
newpriority = PUSER + p->p_estcpu +
NICE_WEIGHT * (p->p_p->ps_nice - NZERO);
newpriority = min(newpriority, MAXPRI);
p->p_usrpri = newpriority;
resched_proc(p, p->p_usrpri);
}
/*
* 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. This can
* cause a switch, but unless the priority crosses a PPQ boundary the actual
* queue will not change. 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
* principle 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.
*/
void
schedclock(struct proc *p)
{
int s;
SCHED_LOCK(s);
p->p_estcpu = ESTCPULIM(p->p_estcpu + 1);
resetpriority(p);
if (p->p_priority >= PUSER)
p->p_priority = p->p_usrpri;
SCHED_UNLOCK(s);
}
|