sched_m2.c revision 1.40 1 1.40 christos /* $NetBSD: sched_m2.c,v 1.40 2024/01/24 16:11:48 christos Exp $ */
2 1.1 rmind
3 1.1 rmind /*
4 1.15 rmind * Copyright (c) 2007, 2008 Mindaugas Rasiukevicius <rmind at NetBSD org>
5 1.12 rmind * All rights reserved.
6 1.1 rmind *
7 1.1 rmind * Redistribution and use in source and binary forms, with or without
8 1.1 rmind * modification, are permitted provided that the following conditions
9 1.1 rmind * are met:
10 1.1 rmind * 1. Redistributions of source code must retain the above copyright
11 1.1 rmind * notice, this list of conditions and the following disclaimer.
12 1.1 rmind * 2. Redistributions in binary form must reproduce the above copyright
13 1.1 rmind * notice, this list of conditions and the following disclaimer in the
14 1.1 rmind * documentation and/or other materials provided with the distribution.
15 1.1 rmind *
16 1.19 rmind * THIS SOFTWARE IS PROVIDED BY THE AUTHOR AND CONTRIBUTORS ``AS IS'' AND
17 1.19 rmind * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
18 1.19 rmind * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
19 1.19 rmind * ARE DISCLAIMED. IN NO EVENT SHALL THE AUTHOR OR CONTRIBUTORS BE LIABLE
20 1.19 rmind * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
21 1.19 rmind * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
22 1.19 rmind * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
23 1.19 rmind * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
24 1.19 rmind * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
25 1.19 rmind * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
26 1.19 rmind * SUCH DAMAGE.
27 1.1 rmind */
28 1.1 rmind
29 1.1 rmind /*
30 1.1 rmind * TODO:
31 1.1 rmind * - Implementation of fair share queue;
32 1.1 rmind * - Support for NUMA;
33 1.1 rmind */
34 1.1 rmind
35 1.1 rmind #include <sys/cdefs.h>
36 1.40 christos __KERNEL_RCSID(0, "$NetBSD: sched_m2.c,v 1.40 2024/01/24 16:11:48 christos Exp $");
37 1.1 rmind
38 1.1 rmind #include <sys/param.h>
39 1.1 rmind
40 1.1 rmind #include <sys/cpu.h>
41 1.1 rmind #include <sys/callout.h>
42 1.1 rmind #include <sys/errno.h>
43 1.1 rmind #include <sys/kernel.h>
44 1.1 rmind #include <sys/kmem.h>
45 1.1 rmind #include <sys/lwp.h>
46 1.1 rmind #include <sys/mutex.h>
47 1.1 rmind #include <sys/pool.h>
48 1.1 rmind #include <sys/proc.h>
49 1.15 rmind #include <sys/pset.h>
50 1.1 rmind #include <sys/resource.h>
51 1.1 rmind #include <sys/resourcevar.h>
52 1.1 rmind #include <sys/sched.h>
53 1.1 rmind #include <sys/syscallargs.h>
54 1.1 rmind #include <sys/sysctl.h>
55 1.1 rmind #include <sys/types.h>
56 1.1 rmind
57 1.1 rmind /*
58 1.32 maxv * Priority related definitions.
59 1.1 rmind */
60 1.10 ad #define PRI_TS_COUNT (NPRI_USER)
61 1.10 ad #define PRI_RT_COUNT (PRI_COUNT - PRI_TS_COUNT)
62 1.10 ad #define PRI_HTS_RANGE (PRI_TS_COUNT / 10)
63 1.10 ad
64 1.11 rmind #define PRI_HIGHEST_TS (MAXPRI_USER)
65 1.10 ad
66 1.1 rmind /*
67 1.1 rmind * Time-slices and priorities.
68 1.1 rmind */
69 1.1 rmind static u_int min_ts; /* Minimal time-slice */
70 1.1 rmind static u_int max_ts; /* Maximal time-slice */
71 1.1 rmind static u_int ts_map[PRI_COUNT]; /* Map of time-slices */
72 1.1 rmind static pri_t high_pri[PRI_COUNT]; /* Map for priority increase */
73 1.39 ad u_int sched_rrticks; /* Real-time time-slice */
74 1.1 rmind
75 1.25 rmind static void sched_precalcts(void);
76 1.25 rmind
77 1.1 rmind /*
78 1.1 rmind * Initialization and setup.
79 1.1 rmind */
80 1.1 rmind
81 1.1 rmind void
82 1.1 rmind sched_rqinit(void)
83 1.1 rmind {
84 1.1 rmind if (hz < 100) {
85 1.1 rmind panic("sched_rqinit: value of HZ is too low\n");
86 1.1 rmind }
87 1.1 rmind
88 1.1 rmind /* Default timing ranges */
89 1.26 rmind min_ts = mstohz(20); /* ~20 ms */
90 1.26 rmind max_ts = mstohz(150); /* ~150 ms */
91 1.39 ad sched_rrticks = mstohz(100); /* ~100 ms */
92 1.1 rmind sched_precalcts();
93 1.1 rmind
94 1.40 christos #ifdef notyet
95 1.30 christos /* Need to set the name etc. This does not belong here */
96 1.1 rmind /* Attach the primary CPU here */
97 1.30 christos sched_cpuattach(curcpu());
98 1.30 christos #endif
99 1.1 rmind
100 1.10 ad sched_lwp_fork(NULL, &lwp0);
101 1.40 christos #ifdef notyet
102 1.40 christos /* without attaching the primary CPU l_mutex does not get initialized */
103 1.40 christos lwp_lock(&lwp0);
104 1.1 rmind sched_newts(&lwp0);
105 1.40 christos lwp_unlock(&lwp0);
106 1.40 christos #else
107 1.40 christos /* gross */
108 1.40 christos lwp0.l_sched.timeslice = ts_map[lwp0.l_auxprio];
109 1.40 christos #endif
110 1.1 rmind }
111 1.1 rmind
112 1.1 rmind /* Pre-calculate the time-slices for the priorities */
113 1.1 rmind static void
114 1.1 rmind sched_precalcts(void)
115 1.1 rmind {
116 1.1 rmind pri_t p;
117 1.1 rmind
118 1.10 ad /* Time-sharing range */
119 1.10 ad for (p = 0; p <= PRI_HIGHEST_TS; p++) {
120 1.10 ad ts_map[p] = max_ts -
121 1.10 ad (p * 100 / (PRI_TS_COUNT - 1) * (max_ts - min_ts) / 100);
122 1.10 ad high_pri[p] = (PRI_HIGHEST_TS - PRI_HTS_RANGE) +
123 1.10 ad ((p * PRI_HTS_RANGE) / (PRI_TS_COUNT - 1));
124 1.10 ad }
125 1.10 ad
126 1.10 ad /* Real-time range */
127 1.10 ad for (p = (PRI_HIGHEST_TS + 1); p < PRI_COUNT; p++) {
128 1.39 ad ts_map[p] = sched_rrticks;
129 1.1 rmind high_pri[p] = p;
130 1.1 rmind }
131 1.1 rmind }
132 1.1 rmind
133 1.1 rmind /*
134 1.1 rmind * Hooks.
135 1.1 rmind */
136 1.1 rmind
137 1.1 rmind void
138 1.1 rmind sched_proc_fork(struct proc *parent, struct proc *child)
139 1.1 rmind {
140 1.1 rmind struct lwp *l;
141 1.1 rmind
142 1.1 rmind LIST_FOREACH(l, &child->p_lwps, l_sibling) {
143 1.1 rmind lwp_lock(l);
144 1.1 rmind sched_newts(l);
145 1.1 rmind lwp_unlock(l);
146 1.1 rmind }
147 1.1 rmind }
148 1.1 rmind
149 1.1 rmind void
150 1.1 rmind sched_proc_exit(struct proc *child, struct proc *parent)
151 1.1 rmind {
152 1.1 rmind
153 1.1 rmind }
154 1.1 rmind
155 1.1 rmind void
156 1.10 ad sched_lwp_fork(struct lwp *l1, struct lwp *l2)
157 1.1 rmind {
158 1.1 rmind
159 1.1 rmind }
160 1.1 rmind
161 1.1 rmind void
162 1.10 ad sched_lwp_collect(struct lwp *l)
163 1.10 ad {
164 1.10 ad
165 1.10 ad }
166 1.10 ad
167 1.10 ad void
168 1.1 rmind sched_setrunnable(struct lwp *l)
169 1.1 rmind {
170 1.1 rmind
171 1.1 rmind }
172 1.1 rmind
173 1.1 rmind void
174 1.1 rmind sched_schedclock(struct lwp *l)
175 1.1 rmind {
176 1.1 rmind
177 1.1 rmind }
178 1.1 rmind
179 1.1 rmind /*
180 1.1 rmind * Priorities and time-slice.
181 1.1 rmind */
182 1.1 rmind
183 1.1 rmind void
184 1.1 rmind sched_nice(struct proc *p, int prio)
185 1.1 rmind {
186 1.27 rmind struct lwp *l;
187 1.27 rmind int n;
188 1.27 rmind
189 1.27 rmind KASSERT(mutex_owned(p->p_lock));
190 1.1 rmind
191 1.27 rmind p->p_nice = prio;
192 1.27 rmind n = (prio - NZERO) >> 2;
193 1.27 rmind if (n == 0)
194 1.27 rmind return;
195 1.27 rmind
196 1.27 rmind LIST_FOREACH(l, &p->p_lwps, l_sibling) {
197 1.27 rmind lwp_lock(l);
198 1.27 rmind if (l->l_class == SCHED_OTHER) {
199 1.27 rmind pri_t pri = l->l_priority - n;
200 1.33 riastrad pri = (n < 0) ? uimin(pri, PRI_HIGHEST_TS) : imax(pri, 0);
201 1.27 rmind lwp_changepri(l, pri);
202 1.27 rmind }
203 1.27 rmind lwp_unlock(l);
204 1.27 rmind }
205 1.1 rmind }
206 1.1 rmind
207 1.1 rmind /* Recalculate the time-slice */
208 1.24 ad void
209 1.1 rmind sched_newts(struct lwp *l)
210 1.1 rmind {
211 1.1 rmind
212 1.26 rmind l->l_sched.timeslice = ts_map[lwp_eprio(l)];
213 1.1 rmind }
214 1.1 rmind
215 1.1 rmind void
216 1.1 rmind sched_slept(struct lwp *l)
217 1.1 rmind {
218 1.1 rmind
219 1.1 rmind /*
220 1.10 ad * If thread is in time-sharing queue and batch flag is not marked,
221 1.29 mbalmer * increase the priority, and run with the lower time-quantum.
222 1.1 rmind */
223 1.25 rmind if (l->l_priority < PRI_HIGHEST_TS && (l->l_flag & LW_BATCH) == 0) {
224 1.27 rmind struct proc *p = l->l_proc;
225 1.27 rmind
226 1.10 ad KASSERT(l->l_class == SCHED_OTHER);
227 1.27 rmind if (__predict_false(p->p_nice < NZERO)) {
228 1.33 riastrad const int n = uimax((NZERO - p->p_nice) >> 2, 1);
229 1.33 riastrad l->l_priority = uimin(l->l_priority + n, PRI_HIGHEST_TS);
230 1.27 rmind } else {
231 1.27 rmind l->l_priority++;
232 1.27 rmind }
233 1.10 ad }
234 1.1 rmind }
235 1.1 rmind
236 1.1 rmind void
237 1.1 rmind sched_wakeup(struct lwp *l)
238 1.1 rmind {
239 1.1 rmind
240 1.1 rmind /* If thread was sleeping a second or more - set a high priority */
241 1.25 rmind if (l->l_slptime >= 1)
242 1.10 ad l->l_priority = high_pri[l->l_priority];
243 1.1 rmind }
244 1.1 rmind
245 1.1 rmind void
246 1.25 rmind sched_pstats_hook(struct lwp *l, int batch)
247 1.1 rmind {
248 1.11 rmind pri_t prio;
249 1.1 rmind
250 1.22 rmind /*
251 1.22 rmind * Estimate threads on time-sharing queue only, however,
252 1.22 rmind * exclude the highest priority for performance purposes.
253 1.22 rmind */
254 1.26 rmind KASSERT(lwp_locked(l, NULL));
255 1.10 ad if (l->l_priority >= PRI_HIGHEST_TS)
256 1.1 rmind return;
257 1.16 rmind KASSERT(l->l_class == SCHED_OTHER);
258 1.1 rmind
259 1.10 ad /* If it is CPU-bound not a first time - decrease the priority */
260 1.11 rmind prio = l->l_priority;
261 1.11 rmind if (batch && prio != 0)
262 1.11 rmind prio--;
263 1.10 ad
264 1.1 rmind /* If thread was not ran a second or more - set a high priority */
265 1.11 rmind if (l->l_stat == LSRUN) {
266 1.38 maxv if (l->l_rticks && (getticks() - l->l_rticks >= hz))
267 1.11 rmind prio = high_pri[prio];
268 1.11 rmind /* Re-enqueue the thread if priority has changed */
269 1.11 rmind if (prio != l->l_priority)
270 1.11 rmind lwp_changepri(l, prio);
271 1.11 rmind } else {
272 1.11 rmind /* In other states, change the priority directly */
273 1.11 rmind l->l_priority = prio;
274 1.11 rmind }
275 1.1 rmind }
276 1.1 rmind
277 1.24 ad void
278 1.24 ad sched_oncpu(lwp_t *l)
279 1.16 rmind {
280 1.26 rmind struct schedstate_percpu *spc = &l->l_cpu->ci_schedstate;
281 1.1 rmind
282 1.1 rmind /* Update the counters */
283 1.26 rmind KASSERT(l->l_sched.timeslice >= min_ts);
284 1.26 rmind KASSERT(l->l_sched.timeslice <= max_ts);
285 1.26 rmind spc->spc_ticks = l->l_sched.timeslice;
286 1.1 rmind }
287 1.1 rmind
288 1.1 rmind /*
289 1.1 rmind * Time-driven events.
290 1.1 rmind */
291 1.1 rmind
292 1.1 rmind /*
293 1.34 ad * Called once per time-quantum, with the running LWP lock held (spc_lwplock).
294 1.1 rmind */
295 1.1 rmind void
296 1.1 rmind sched_tick(struct cpu_info *ci)
297 1.1 rmind {
298 1.1 rmind struct schedstate_percpu *spc = &ci->ci_schedstate;
299 1.36 ad struct lwp *l = ci->ci_onproc;
300 1.27 rmind struct proc *p;
301 1.1 rmind
302 1.26 rmind if (__predict_false(CURCPU_IDLE_P()))
303 1.2 rmind return;
304 1.1 rmind
305 1.34 ad lwp_lock(l);
306 1.34 ad KASSERT(l->l_mutex != spc->spc_mutex);
307 1.10 ad switch (l->l_class) {
308 1.2 rmind case SCHED_FIFO:
309 1.2 rmind /*
310 1.2 rmind * Update the time-quantum, and continue running,
311 1.2 rmind * if thread runs on FIFO real-time policy.
312 1.2 rmind */
313 1.16 rmind KASSERT(l->l_priority > PRI_HIGHEST_TS);
314 1.26 rmind spc->spc_ticks = l->l_sched.timeslice;
315 1.34 ad lwp_unlock(l);
316 1.1 rmind return;
317 1.2 rmind case SCHED_OTHER:
318 1.10 ad /*
319 1.10 ad * If thread is in time-sharing queue, decrease the priority,
320 1.10 ad * and run with a higher time-quantum.
321 1.10 ad */
322 1.16 rmind KASSERT(l->l_priority <= PRI_HIGHEST_TS);
323 1.27 rmind if (l->l_priority == 0)
324 1.27 rmind break;
325 1.27 rmind
326 1.27 rmind p = l->l_proc;
327 1.27 rmind if (__predict_false(p->p_nice > NZERO)) {
328 1.33 riastrad const int n = uimax((p->p_nice - NZERO) >> 2, 1);
329 1.27 rmind l->l_priority = imax(l->l_priority - n, 0);
330 1.27 rmind } else
331 1.10 ad l->l_priority--;
332 1.2 rmind break;
333 1.1 rmind }
334 1.1 rmind
335 1.1 rmind /*
336 1.2 rmind * If there are higher priority threads or threads in the same queue,
337 1.2 rmind * mark that thread should yield, otherwise, continue running.
338 1.1 rmind */
339 1.24 ad if (lwp_eprio(l) <= spc->spc_maxpriority || l->l_target_cpu) {
340 1.1 rmind spc->spc_flags |= SPCF_SHOULDYIELD;
341 1.37 ad spc_lock(ci);
342 1.37 ad sched_resched_cpu(ci, MAXPRI_KTHREAD, true);
343 1.37 ad /* spc now unlocked */
344 1.1 rmind } else
345 1.34 ad spc->spc_ticks = l->l_sched.timeslice;
346 1.34 ad lwp_unlock(l);
347 1.1 rmind }
348 1.1 rmind
349 1.1 rmind /*
350 1.1 rmind * Sysctl nodes and initialization.
351 1.1 rmind */
352 1.1 rmind
353 1.1 rmind static int
354 1.15 rmind sysctl_sched_rtts(SYSCTLFN_ARGS)
355 1.15 rmind {
356 1.15 rmind struct sysctlnode node;
357 1.39 ad int rttsms = hztoms(sched_rrticks);
358 1.15 rmind
359 1.15 rmind node = *rnode;
360 1.15 rmind node.sysctl_data = &rttsms;
361 1.15 rmind return sysctl_lookup(SYSCTLFN_CALL(&node));
362 1.15 rmind }
363 1.15 rmind
364 1.15 rmind static int
365 1.1 rmind sysctl_sched_mints(SYSCTLFN_ARGS)
366 1.1 rmind {
367 1.1 rmind struct sysctlnode node;
368 1.1 rmind struct cpu_info *ci;
369 1.1 rmind int error, newsize;
370 1.1 rmind CPU_INFO_ITERATOR cii;
371 1.1 rmind
372 1.1 rmind node = *rnode;
373 1.1 rmind node.sysctl_data = &newsize;
374 1.1 rmind
375 1.1 rmind newsize = hztoms(min_ts);
376 1.1 rmind error = sysctl_lookup(SYSCTLFN_CALL(&node));
377 1.1 rmind if (error || newp == NULL)
378 1.1 rmind return error;
379 1.1 rmind
380 1.8 rmind newsize = mstohz(newsize);
381 1.1 rmind if (newsize < 1 || newsize > hz || newsize >= max_ts)
382 1.1 rmind return EINVAL;
383 1.1 rmind
384 1.1 rmind /* It is safe to do this in such order */
385 1.1 rmind for (CPU_INFO_FOREACH(cii, ci))
386 1.1 rmind spc_lock(ci);
387 1.1 rmind
388 1.8 rmind min_ts = newsize;
389 1.1 rmind sched_precalcts();
390 1.1 rmind
391 1.1 rmind for (CPU_INFO_FOREACH(cii, ci))
392 1.1 rmind spc_unlock(ci);
393 1.1 rmind
394 1.1 rmind return 0;
395 1.1 rmind }
396 1.1 rmind
397 1.1 rmind static int
398 1.1 rmind sysctl_sched_maxts(SYSCTLFN_ARGS)
399 1.1 rmind {
400 1.1 rmind struct sysctlnode node;
401 1.1 rmind struct cpu_info *ci;
402 1.1 rmind int error, newsize;
403 1.1 rmind CPU_INFO_ITERATOR cii;
404 1.1 rmind
405 1.1 rmind node = *rnode;
406 1.1 rmind node.sysctl_data = &newsize;
407 1.1 rmind
408 1.1 rmind newsize = hztoms(max_ts);
409 1.1 rmind error = sysctl_lookup(SYSCTLFN_CALL(&node));
410 1.1 rmind if (error || newp == NULL)
411 1.1 rmind return error;
412 1.1 rmind
413 1.8 rmind newsize = mstohz(newsize);
414 1.1 rmind if (newsize < 10 || newsize > hz || newsize <= min_ts)
415 1.1 rmind return EINVAL;
416 1.1 rmind
417 1.1 rmind /* It is safe to do this in such order */
418 1.1 rmind for (CPU_INFO_FOREACH(cii, ci))
419 1.1 rmind spc_lock(ci);
420 1.1 rmind
421 1.8 rmind max_ts = newsize;
422 1.1 rmind sched_precalcts();
423 1.1 rmind
424 1.1 rmind for (CPU_INFO_FOREACH(cii, ci))
425 1.1 rmind spc_unlock(ci);
426 1.1 rmind
427 1.1 rmind return 0;
428 1.1 rmind }
429 1.1 rmind
430 1.24 ad SYSCTL_SETUP(sysctl_sched_m2_setup, "sysctl sched setup")
431 1.1 rmind {
432 1.1 rmind const struct sysctlnode *node = NULL;
433 1.1 rmind
434 1.1 rmind sysctl_createv(clog, 0, NULL, &node,
435 1.1 rmind CTLFLAG_PERMANENT,
436 1.1 rmind CTLTYPE_NODE, "sched",
437 1.1 rmind SYSCTL_DESCR("Scheduler options"),
438 1.1 rmind NULL, 0, NULL, 0,
439 1.1 rmind CTL_KERN, CTL_CREATE, CTL_EOL);
440 1.1 rmind
441 1.1 rmind if (node == NULL)
442 1.1 rmind return;
443 1.1 rmind
444 1.24 ad sysctl_createv(NULL, 0, &node, NULL,
445 1.1 rmind CTLFLAG_PERMANENT,
446 1.1 rmind CTLTYPE_STRING, "name", NULL,
447 1.1 rmind NULL, 0, __UNCONST("M2"), 0,
448 1.1 rmind CTL_CREATE, CTL_EOL);
449 1.24 ad sysctl_createv(NULL, 0, &node, NULL,
450 1.15 rmind CTLFLAG_PERMANENT,
451 1.15 rmind CTLTYPE_INT, "rtts",
452 1.32 maxv SYSCTL_DESCR("Round-robin time quantum (in milliseconds)"),
453 1.15 rmind sysctl_sched_rtts, 0, NULL, 0,
454 1.15 rmind CTL_CREATE, CTL_EOL);
455 1.24 ad sysctl_createv(NULL, 0, &node, NULL,
456 1.1 rmind CTLFLAG_PERMANENT | CTLFLAG_READWRITE,
457 1.1 rmind CTLTYPE_INT, "maxts",
458 1.32 maxv SYSCTL_DESCR("Maximal time quantum (in milliseconds)"),
459 1.1 rmind sysctl_sched_maxts, 0, &max_ts, 0,
460 1.1 rmind CTL_CREATE, CTL_EOL);
461 1.24 ad sysctl_createv(NULL, 0, &node, NULL,
462 1.1 rmind CTLFLAG_PERMANENT | CTLFLAG_READWRITE,
463 1.1 rmind CTLTYPE_INT, "mints",
464 1.32 maxv SYSCTL_DESCR("Minimal time quantum (in milliseconds)"),
465 1.1 rmind sysctl_sched_mints, 0, &min_ts, 0,
466 1.1 rmind CTL_CREATE, CTL_EOL);
467 1.1 rmind }
468