source: mainline/src/proc/scheduler.c@ cd95d784

lfn serial ticket/834-toolchain-update topic/msim-upgrade topic/simplify-dev-export
Last change on this file since cd95d784 was cd95d784, checked in by Jakub Vana <jakub.vana@…>, 20 years ago
  • Property mode set to 100644
File size: 12.9 KB
Line 
1/*
2 * Copyright (C) 2001-2004 Jakub Jermar
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 *
9 * - Redistributions of source code must retain the above copyright
10 * notice, this list of conditions and the following disclaimer.
11 * - Redistributions in binary form must reproduce the above copyright
12 * notice, this list of conditions and the following disclaimer in the
13 * documentation and/or other materials provided with the distribution.
14 * - The name of the author may not be used to endorse or promote products
15 * derived from this software without specific prior written permission.
16 *
17 * THIS SOFTWARE IS PROVIDED BY THE AUTHOR ``AS IS'' AND ANY EXPRESS OR
18 * IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE IMPLIED WARRANTIES
19 * OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE ARE DISCLAIMED.
20 * IN NO EVENT SHALL THE AUTHOR BE LIABLE FOR ANY DIRECT, INDIRECT,
21 * INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT
22 * NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE,
23 * DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY
24 * THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT
25 * (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE OF
26 * THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
27 */
28
29#include <proc/scheduler.h>
30#include <proc/thread.h>
31#include <proc/task.h>
32#include <cpu.h>
33#include <mm/vm.h>
34#include <config.h>
35#include <context.h>
36#include <func.h>
37#include <arch.h>
38#include <arch/asm.h>
39#include <list.h>
40#include <panic.h>
41#include <typedefs.h>
42#include <mm/page.h>
43#include <synch/spinlock.h>
44#include <arch/faddr.h>
45#include <arch/atomic.h>
46#include <print.h>
47#include <mm/frame.h>
48#include <mm/heap.h>
49
50
51volatile int nrdy;
52
53
54/** Take actions before new thread runs
55 *
56 * Perform actions that need to be
57 * taken before the newly selected
58 * tread is passed control.
59 *
60 */
61void before_thread_runs(void)
62{
63 before_thread_runs_arch();
64 fpu_context_restore(&(THREAD->saved_fpu_context));
65}
66
67
68/** Initialize scheduler
69 *
70 * Initialize kernel scheduler.
71 *
72 */
73void scheduler_init(void)
74{
75}
76
77
78/** Get thread to be scheduled
79 *
80 * Get the optimal thread to be scheduled
81 * according to thread accounting and scheduler
82 * policy.
83 *
84 * @return Thread to be scheduled.
85 *
86 */
87struct thread *find_best_thread(void)
88{
89 thread_t *t;
90 runq_t *r;
91 int i, n;
92
93loop:
94 cpu_priority_high();
95
96 spinlock_lock(&CPU->lock);
97 n = CPU->nrdy;
98 spinlock_unlock(&CPU->lock);
99
100 cpu_priority_low();
101
102 if (n == 0) {
103 #ifdef __SMP__
104 /*
105 * If the load balancing thread is not running, wake it up and
106 * set CPU-private flag that the kcpulb has been started.
107 */
108 if (test_and_set(&CPU->kcpulbstarted) == 0) {
109 waitq_wakeup(&CPU->kcpulb_wq, 0);
110 goto loop;
111 }
112 #endif /* __SMP__ */
113
114 /*
115 * For there was nothing to run, the CPU goes to sleep
116 * until a hardware interrupt or an IPI comes.
117 * This improves energy saving and hyperthreading.
118 * On the other hand, several hardware interrupts can be ignored.
119 */
120 cpu_sleep();
121 goto loop;
122 }
123
124 cpu_priority_high();
125
126 i = 0;
127retry:
128 for (; i<RQ_COUNT; i++) {
129 r = &CPU->rq[i];
130 spinlock_lock(&r->lock);
131 if (r->n == 0) {
132 /*
133 * If this queue is empty, try a lower-priority queue.
134 */
135 spinlock_unlock(&r->lock);
136 continue;
137 }
138
139 /* avoid deadlock with relink_rq() */
140 if (!spinlock_trylock(&CPU->lock)) {
141 /*
142 * Unlock r and try again.
143 */
144 spinlock_unlock(&r->lock);
145 goto retry;
146 }
147 CPU->nrdy--;
148 spinlock_unlock(&CPU->lock);
149
150 atomic_dec(&nrdy);
151 r->n--;
152
153 /*
154 * Take the first thread from the queue.
155 */
156 t = list_get_instance(r->rq_head.next, thread_t, rq_link);
157 list_remove(&t->rq_link);
158
159 spinlock_unlock(&r->lock);
160
161 spinlock_lock(&t->lock);
162 t->cpu = CPU;
163
164 t->ticks = us2ticks((i+1)*10000);
165 t->pri = i; /* eventually correct rq index */
166
167 /*
168 * Clear the X_STOLEN flag so that t can be migrated when load balancing needs emerge.
169 */
170 t->flags &= ~X_STOLEN;
171 spinlock_unlock(&t->lock);
172
173 return t;
174 }
175 goto loop;
176
177}
178
179
180/** Prevent rq starvation
181 *
182 * Prevent low priority threads from starving in rq's.
183 *
184 * When the function decides to relink rq's, it reconnects
185 * respective pointers so that in result threads with 'pri'
186 * greater or equal 'start' are moved to a higher-priority queue.
187 *
188 * @param start Threshold priority.
189 *
190 */
191void relink_rq(int start)
192{
193 link_t head;
194 runq_t *r;
195 int i, n;
196
197 list_initialize(&head);
198 spinlock_lock(&CPU->lock);
199 if (CPU->needs_relink > NEEDS_RELINK_MAX) {
200 for (i = start; i<RQ_COUNT-1; i++) {
201 /* remember and empty rq[i + 1] */
202 r = &CPU->rq[i + 1];
203 spinlock_lock(&r->lock);
204 list_concat(&head, &r->rq_head);
205 n = r->n;
206 r->n = 0;
207 spinlock_unlock(&r->lock);
208
209 /* append rq[i + 1] to rq[i] */
210 r = &CPU->rq[i];
211 spinlock_lock(&r->lock);
212 list_concat(&r->rq_head, &head);
213 r->n += n;
214 spinlock_unlock(&r->lock);
215 }
216 CPU->needs_relink = 0;
217 }
218 spinlock_unlock(&CPU->lock);
219
220}
221
222
223/** The scheduler
224 *
225 * The thread scheduling procedure.
226 *
227 */
228void scheduler(void)
229{
230 volatile pri_t pri;
231
232 pri = cpu_priority_high();
233
234 if (haltstate)
235 halt();
236
237 if (THREAD) {
238 spinlock_lock(&THREAD->lock);
239 fpu_context_save(&(THREAD->saved_fpu_context));
240 if (!context_save(&THREAD->saved_context)) {
241 /*
242 * This is the place where threads leave scheduler();
243 */
244 before_thread_runs();
245 spinlock_unlock(&THREAD->lock);
246 cpu_priority_restore(THREAD->saved_context.pri);
247 return;
248 }
249
250 /*
251 * CPU priority of preempted thread is recorded here
252 * to facilitate scheduler() invocations from
253 * cpu_priority_high()'ed code (e.g. waitq_sleep_timeout()).
254 */
255 THREAD->saved_context.pri = pri;
256 }
257
258 /*
259 * Through the 'THE' structure, we keep track of THREAD, TASK, CPU
260 * and preemption counter. At this point THE could be coming either
261 * from THREAD's or CPU's stack.
262 */
263 the_copy(THE, (the_t *) CPU->stack);
264
265 /*
266 * We may not keep the old stack.
267 * Reason: If we kept the old stack and got blocked, for instance, in
268 * find_best_thread(), the old thread could get rescheduled by another
269 * CPU and overwrite the part of its own stack that was also used by
270 * the scheduler on this CPU.
271 *
272 * Moreover, we have to bypass the compiler-generated POP sequence
273 * which is fooled by SP being set to the very top of the stack.
274 * Therefore the scheduler() function continues in
275 * scheduler_separated_stack().
276 */
277 context_save(&CPU->saved_context);
278 context_set(&CPU->saved_context, FADDR(scheduler_separated_stack), CPU->stack, CPU_STACK_SIZE);
279 context_restore(&CPU->saved_context);
280 /* not reached */
281}
282
283
284/** Scheduler stack switch wrapper
285 *
286 * Second part of the scheduler() function
287 * using new stack. Handling the actual context
288 * switch to a new thread.
289 *
290 */
291void scheduler_separated_stack(void)
292{
293 int priority;
294
295 if (THREAD) {
296 switch (THREAD->state) {
297 case Running:
298 THREAD->state = Ready;
299 spinlock_unlock(&THREAD->lock);
300 thread_ready(THREAD);
301 break;
302
303 case Exiting:
304 frame_free((__address) THREAD->kstack);
305 if (THREAD->ustack) {
306 frame_free((__address) THREAD->ustack);
307 }
308
309 /*
310 * Detach from the containing task.
311 */
312 spinlock_lock(&TASK->lock);
313 list_remove(&THREAD->th_link);
314 spinlock_unlock(&TASK->lock);
315
316 spinlock_unlock(&THREAD->lock);
317
318 spinlock_lock(&threads_lock);
319 list_remove(&THREAD->threads_link);
320 spinlock_unlock(&threads_lock);
321
322 spinlock_lock(&CPU->lock);
323 if(CPU->fpu_owner==THREAD) CPU->fpu_owner=NULL;
324 spinlock_unlock(&CPU->lock);
325
326 free(THREAD);
327
328 break;
329
330 case Sleeping:
331 /*
332 * Prefer the thread after it's woken up.
333 */
334 THREAD->pri = -1;
335
336 /*
337 * We need to release wq->lock which we locked in waitq_sleep().
338 * Address of wq->lock is kept in THREAD->sleep_queue.
339 */
340 spinlock_unlock(&THREAD->sleep_queue->lock);
341
342 /*
343 * Check for possible requests for out-of-context invocation.
344 */
345 if (THREAD->call_me) {
346 THREAD->call_me(THREAD->call_me_with);
347 THREAD->call_me = NULL;
348 THREAD->call_me_with = NULL;
349 }
350
351 spinlock_unlock(&THREAD->lock);
352
353 break;
354
355 default:
356 /*
357 * Entering state is unexpected.
358 */
359 panic("tid%d: unexpected state %s\n", THREAD->tid, thread_states[THREAD->state]);
360 break;
361 }
362 THREAD = NULL;
363 }
364
365
366 THREAD = find_best_thread();
367
368 spinlock_lock(&THREAD->lock);
369 priority = THREAD->pri;
370 spinlock_unlock(&THREAD->lock);
371
372 relink_rq(priority);
373
374 spinlock_lock(&THREAD->lock);
375
376 /*
377 * If both the old and the new task are the same, lots of work is avoided.
378 */
379 if (TASK != THREAD->task) {
380 vm_t *m1 = NULL;
381 vm_t *m2;
382
383 if (TASK) {
384 spinlock_lock(&TASK->lock);
385 m1 = TASK->vm;
386 spinlock_unlock(&TASK->lock);
387 }
388
389 spinlock_lock(&THREAD->task->lock);
390 m2 = THREAD->task->vm;
391 spinlock_unlock(&THREAD->task->lock);
392
393 /*
394 * Note that it is possible for two tasks to share one vm mapping.
395 */
396 if (m1 != m2) {
397 /*
398 * Both tasks and vm mappings are different.
399 * Replace the old one with the new one.
400 */
401 vm_install(m2);
402 }
403 TASK = THREAD->task;
404 }
405
406 THREAD->state = Running;
407
408 #ifdef SCHEDULER_VERBOSE
409 printf("cpu%d: tid %d (pri=%d,ticks=%d,nrdy=%d)\n", CPU->id, THREAD->tid, THREAD->pri, THREAD->ticks, CPU->nrdy);
410 #endif
411
412 /*
413 * Copy the knowledge of CPU, TASK, THREAD and preemption counter to thread's stack.
414 */
415 the_copy(THE, (the_t *) THREAD->kstack);
416
417 context_restore(&THREAD->saved_context);
418 /* not reached */
419}
420
421
422#ifdef __SMP__
423/** Load balancing thread
424 *
425 * SMP load balancing thread, supervising thread supplies
426 * for the CPU it's wired to.
427 *
428 * @param arg Generic thread argument (unused).
429 *
430 */
431void kcpulb(void *arg)
432{
433 thread_t *t;
434 int count, i, j, k = 0;
435 pri_t pri;
436
437loop:
438 /*
439 * Sleep until there's some work to do.
440 */
441 waitq_sleep(&CPU->kcpulb_wq);
442
443not_satisfied:
444 /*
445 * Calculate the number of threads that will be migrated/stolen from
446 * other CPU's. Note that situation can have changed between two
447 * passes. Each time get the most up to date counts.
448 */
449 pri = cpu_priority_high();
450 spinlock_lock(&CPU->lock);
451 count = nrdy / config.cpu_active;
452 count -= CPU->nrdy;
453 spinlock_unlock(&CPU->lock);
454 cpu_priority_restore(pri);
455
456 if (count <= 0)
457 goto satisfied;
458
459 /*
460 * Searching least priority queues on all CPU's first and most priority queues on all CPU's last.
461 */
462 for (j=RQ_COUNT-1; j >= 0; j--) {
463 for (i=0; i < config.cpu_active; i++) {
464 link_t *l;
465 runq_t *r;
466 cpu_t *cpu;
467
468 cpu = &cpus[(i + k) % config.cpu_active];
469
470 /*
471 * Not interested in ourselves.
472 * Doesn't require interrupt disabling for kcpulb is X_WIRED.
473 */
474 if (CPU == cpu)
475 continue;
476
477restart: pri = cpu_priority_high();
478 r = &cpu->rq[j];
479 spinlock_lock(&r->lock);
480 if (r->n == 0) {
481 spinlock_unlock(&r->lock);
482 cpu_priority_restore(pri);
483 continue;
484 }
485
486 t = NULL;
487 l = r->rq_head.prev; /* search rq from the back */
488 while (l != &r->rq_head) {
489 t = list_get_instance(l, thread_t, rq_link);
490 /*
491 * We don't want to steal CPU-wired threads neither threads already stolen.
492 * The latter prevents threads from migrating between CPU's without ever being run.
493 * We don't want to steal threads whose FPU context is still in CPU.
494 */
495 spinlock_lock(&t->lock);
496 if ( (!(t->flags & (X_WIRED | X_STOLEN))) && (!(t->fpu_context_engaged)) ) {
497
498 /*
499 * Remove t from r.
500 */
501
502 spinlock_unlock(&t->lock);
503
504 /*
505 * Here we have to avoid deadlock with relink_rq(),
506 * because it locks cpu and r in a different order than we do.
507 */
508 if (!spinlock_trylock(&cpu->lock)) {
509 /* Release all locks and try again. */
510 spinlock_unlock(&r->lock);
511 cpu_priority_restore(pri);
512 goto restart;
513 }
514 cpu->nrdy--;
515 spinlock_unlock(&cpu->lock);
516
517 atomic_dec(&nrdy);
518
519 r->n--;
520 list_remove(&t->rq_link);
521
522 break;
523 }
524 spinlock_unlock(&t->lock);
525 l = l->prev;
526 t = NULL;
527 }
528 spinlock_unlock(&r->lock);
529
530 if (t) {
531 /*
532 * Ready t on local CPU
533 */
534 spinlock_lock(&t->lock);
535 #ifdef KCPULB_VERBOSE
536 printf("kcpulb%d: TID %d -> cpu%d, nrdy=%d, avg=%d\n", CPU->id, t->tid, CPU->id, CPU->nrdy, nrdy / config.cpu_active);
537 #endif
538 t->flags |= X_STOLEN;
539 spinlock_unlock(&t->lock);
540
541 thread_ready(t);
542
543 cpu_priority_restore(pri);
544
545 if (--count == 0)
546 goto satisfied;
547
548 /*
549 * We are not satisfied yet, focus on another CPU next time.
550 */
551 k++;
552
553 continue;
554 }
555 cpu_priority_restore(pri);
556 }
557 }
558
559 if (CPU->nrdy) {
560 /*
561 * Be a little bit light-weight and let migrated threads run.
562 */
563 scheduler();
564 }
565 else {
566 /*
567 * We failed to migrate a single thread.
568 * Something more sophisticated should be done.
569 */
570 scheduler();
571 }
572
573 goto not_satisfied;
574
575satisfied:
576 /*
577 * Tell find_best_thread() to wake us up later again.
578 */
579 CPU->kcpulbstarted = 0;
580 goto loop;
581}
582
583#endif /* __SMP__ */
Note: See TracBrowser for help on using the repository browser.