Subversion Repositories HelenOS

Rev

Rev 195 | Rev 212 | Go to most recent revision | Details | Compare with Previous | Last modification | View Log | RSS feed

Rev Author Line No. Line
1 jermar 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>
68 decky 40
#include <panic.h>
1 jermar 41
#include <typedefs.h>
42
#include <mm/page.h>
43
#include <synch/spinlock.h>
76 jermar 44
#include <arch/faddr.h>
111 palkovsky 45
#include <arch/atomic.h>
195 vana 46
#include <print.h>
47
#include <mm/frame.h>
48
#include <mm/heap.h>
1 jermar 49
 
195 vana 50
 
1 jermar 51
volatile int nrdy;
52
 
107 decky 53
 
118 jermar 54
/** Take actions before new thread runs
107 decky 55
 *
118 jermar 56
 * Perform actions that need to be
57
 * taken before the newly selected
58
 * tread is passed control.
107 decky 59
 *
60
 */
52 vana 61
void before_thread_runs(void)
62
{
63
    before_thread_runs_arch();
57 vana 64
    fpu_context_restore(&(THREAD->saved_fpu_context));
52 vana 65
}
66
 
67
 
107 decky 68
/** Initialize scheduler
69
 *
70
 * Initialize kernel scheduler.
71
 *
72
 */
1 jermar 73
void scheduler_init(void)
74
{
75
}
76
 
107 decky 77
 
78
/** Get thread to be scheduled
79
 *
80
 * Get the optimal thread to be scheduled
109 jermar 81
 * according to thread accounting and scheduler
107 decky 82
 * policy.
83
 *
84
 * @return Thread to be scheduled.
85
 *
86
 */
1 jermar 87
struct thread *find_best_thread(void)
88
{
89
    thread_t *t;
90
    runq_t *r;
91
    int i, n;
92
 
93
loop:
94
    cpu_priority_high();
95
 
15 jermar 96
    spinlock_lock(&CPU->lock);
97
    n = CPU->nrdy;
98
    spinlock_unlock(&CPU->lock);
1 jermar 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
         */
15 jermar 108
        if (test_and_set(&CPU->kcpulbstarted) == 0) {
125 jermar 109
            waitq_wakeup(&CPU->kcpulb_wq, 0);
1 jermar 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();
114 jermar 125
 
126
    i = 0;
127
retry:
128
    for (; i<RQ_COUNT; i++) {
15 jermar 129
        r = &CPU->rq[i];
1 jermar 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
 
115 jermar 139
        /* avoid deadlock with relink_rq() */
114 jermar 140
        if (!spinlock_trylock(&CPU->lock)) {
141
            /*
142
             * Unlock r and try again.
143
             */
144
            spinlock_unlock(&r->lock);
145
            goto retry;
146
        }
15 jermar 147
        CPU->nrdy--;
148
        spinlock_unlock(&CPU->lock);
1 jermar 149
 
114 jermar 150
        atomic_dec(&nrdy);
1 jermar 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);
15 jermar 162
        t->cpu = CPU;
1 jermar 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
 
107 decky 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
 *
1 jermar 190
 */
191
void relink_rq(int start)
192
{
193
    link_t head;
194
    runq_t *r;
195
    int i, n;
196
 
197
    list_initialize(&head);
15 jermar 198
    spinlock_lock(&CPU->lock);
199
    if (CPU->needs_relink > NEEDS_RELINK_MAX) {
1 jermar 200
        for (i = start; i<RQ_COUNT-1; i++) {
201
            /* remember and empty rq[i + 1] */
15 jermar 202
            r = &CPU->rq[i + 1];
1 jermar 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] */
15 jermar 210
            r = &CPU->rq[i];
1 jermar 211
            spinlock_lock(&r->lock);
212
            list_concat(&r->rq_head, &head);
213
            r->n += n;
214
            spinlock_unlock(&r->lock);
215
        }
15 jermar 216
        CPU->needs_relink = 0;
1 jermar 217
    }
15 jermar 218
    spinlock_unlock(&CPU->lock);               
1 jermar 219
 
220
}
221
 
107 decky 222
 
223
/** The scheduler
224
 *
225
 * The thread scheduling procedure.
226
 *
1 jermar 227
 */
228
void scheduler(void)
229
{
230
    volatile pri_t pri;
231
 
232
    pri = cpu_priority_high();
233
 
234
    if (haltstate)
235
        halt();
236
 
15 jermar 237
    if (THREAD) {
238
        spinlock_lock(&THREAD->lock);
57 vana 239
        fpu_context_save(&(THREAD->saved_fpu_context));
15 jermar 240
        if (!context_save(&THREAD->saved_context)) {
1 jermar 241
            /*
242
             * This is the place where threads leave scheduler();
243
             */
22 jermar 244
            before_thread_runs();
125 jermar 245
            spinlock_unlock(&THREAD->lock);
15 jermar 246
            cpu_priority_restore(THREAD->saved_context.pri);
1 jermar 247
            return;
248
        }
170 jermar 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
         */
15 jermar 255
        THREAD->saved_context.pri = pri;
1 jermar 256
    }
257
 
258
    /*
184 jermar 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
    /*
1 jermar 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
     */
15 jermar 277
    context_save(&CPU->saved_context);
97 jermar 278
    context_set(&CPU->saved_context, FADDR(scheduler_separated_stack), CPU->stack, CPU_STACK_SIZE);
15 jermar 279
    context_restore(&CPU->saved_context);
1 jermar 280
    /* not reached */
281
}
282
 
107 decky 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
 */
1 jermar 291
void scheduler_separated_stack(void)
292
{
293
    int priority;
294
 
15 jermar 295
    if (THREAD) {
296
        switch (THREAD->state) {
1 jermar 297
            case Running:
125 jermar 298
            THREAD->state = Ready;
299
            spinlock_unlock(&THREAD->lock);
300
            thread_ready(THREAD);
301
            break;
1 jermar 302
 
303
            case Exiting:
125 jermar 304
            frame_free((__address) THREAD->kstack);
305
            if (THREAD->ustack) {
306
                frame_free((__address) THREAD->ustack);
307
            }
1 jermar 308
 
125 jermar 309
            /*
310
             * Detach from the containing task.
311
             */
312
            spinlock_lock(&TASK->lock);
313
            list_remove(&THREAD->th_link);
314
            spinlock_unlock(&TASK->lock);
73 vana 315
 
125 jermar 316
            spinlock_unlock(&THREAD->lock);
317
 
318
            spinlock_lock(&threads_lock);
319
            list_remove(&THREAD->threads_link);
320
            spinlock_unlock(&threads_lock);
73 vana 321
 
125 jermar 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
 
1 jermar 330
            case Sleeping:
125 jermar 331
            /*
332
             * Prefer the thread after it's woken up.
333
             */
334
            THREAD->pri = -1;
1 jermar 335
 
125 jermar 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);
1 jermar 341
 
125 jermar 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
            }
1 jermar 350
 
125 jermar 351
            spinlock_unlock(&THREAD->lock);
1 jermar 352
 
125 jermar 353
            break;
354
 
1 jermar 355
            default:
125 jermar 356
            /*
357
             * Entering state is unexpected.
358
             */
359
            panic("tid%d: unexpected state %s\n", THREAD->tid, thread_states[THREAD->state]);
360
            break;
1 jermar 361
        }
15 jermar 362
        THREAD = NULL;
1 jermar 363
    }
198 jermar 364
 
15 jermar 365
    THREAD = find_best_thread();
1 jermar 366
 
15 jermar 367
    spinlock_lock(&THREAD->lock);
368
    priority = THREAD->pri;
369
    spinlock_unlock(&THREAD->lock);
192 jermar 370
 
1 jermar 371
    relink_rq(priority);       
372
 
15 jermar 373
    spinlock_lock(&THREAD->lock);  
1 jermar 374
 
375
    /*
376
     * If both the old and the new task are the same, lots of work is avoided.
377
     */
15 jermar 378
    if (TASK != THREAD->task) {
1 jermar 379
        vm_t *m1 = NULL;
380
        vm_t *m2;
381
 
15 jermar 382
        if (TASK) {
383
            spinlock_lock(&TASK->lock);
384
            m1 = TASK->vm;
385
            spinlock_unlock(&TASK->lock);
1 jermar 386
        }
387
 
15 jermar 388
        spinlock_lock(&THREAD->task->lock);
389
        m2 = THREAD->task->vm;
390
        spinlock_unlock(&THREAD->task->lock);
1 jermar 391
 
392
        /*
393
         * Note that it is possible for two tasks to share one vm mapping.
394
         */
395
        if (m1 != m2) {
396
            /*
397
             * Both tasks and vm mappings are different.
398
             * Replace the old one with the new one.
399
             */
400
            vm_install(m2);
401
        }
15 jermar 402
        TASK = THREAD->task;   
1 jermar 403
    }
404
 
15 jermar 405
    THREAD->state = Running;
1 jermar 406
 
407
    #ifdef SCHEDULER_VERBOSE
15 jermar 408
    printf("cpu%d: tid %d (pri=%d,ticks=%d,nrdy=%d)\n", CPU->id, THREAD->tid, THREAD->pri, THREAD->ticks, CPU->nrdy);
1 jermar 409
    #endif  
410
 
184 jermar 411
    the_copy(THE, (the_t *) THREAD->kstack);
412
 
15 jermar 413
    context_restore(&THREAD->saved_context);
1 jermar 414
    /* not reached */
415
}
416
 
107 decky 417
 
1 jermar 418
#ifdef __SMP__
107 decky 419
/** Load balancing thread
420
 *
421
 * SMP load balancing thread, supervising thread supplies
422
 * for the CPU it's wired to.
423
 *
424
 * @param arg Generic thread argument (unused).
425
 *
1 jermar 426
 */
427
void kcpulb(void *arg)
428
{
429
    thread_t *t;
430
    int count, i, j, k = 0;
431
    pri_t pri;
432
 
433
loop:
434
    /*
435
     * Sleep until there's some work to do.
436
     */
15 jermar 437
    waitq_sleep(&CPU->kcpulb_wq);
1 jermar 438
 
439
not_satisfied:
440
    /*
441
     * Calculate the number of threads that will be migrated/stolen from
442
     * other CPU's. Note that situation can have changed between two
443
     * passes. Each time get the most up to date counts.
444
     */
445
    pri = cpu_priority_high();
15 jermar 446
    spinlock_lock(&CPU->lock);
1 jermar 447
    count = nrdy / config.cpu_active;
15 jermar 448
    count -= CPU->nrdy;
449
    spinlock_unlock(&CPU->lock);
1 jermar 450
    cpu_priority_restore(pri);
451
 
452
    if (count <= 0)
453
        goto satisfied;
454
 
455
    /*
456
     * Searching least priority queues on all CPU's first and most priority queues on all CPU's last.
457
     */
458
    for (j=RQ_COUNT-1; j >= 0; j--) {
459
        for (i=0; i < config.cpu_active; i++) {
460
            link_t *l;
461
            runq_t *r;
462
            cpu_t *cpu;
463
 
464
            cpu = &cpus[(i + k) % config.cpu_active];
465
 
466
            /*
467
             * Not interested in ourselves.
468
             * Doesn't require interrupt disabling for kcpulb is X_WIRED.
469
             */
15 jermar 470
            if (CPU == cpu)
115 jermar 471
                continue;              
1 jermar 472
 
473
restart:        pri = cpu_priority_high();
115 jermar 474
            r = &cpu->rq[j];
1 jermar 475
            spinlock_lock(&r->lock);
476
            if (r->n == 0) {
477
                spinlock_unlock(&r->lock);
478
                cpu_priority_restore(pri);
479
                continue;
480
            }
481
 
482
            t = NULL;
483
            l = r->rq_head.prev;    /* search rq from the back */
484
            while (l != &r->rq_head) {
485
                t = list_get_instance(l, thread_t, rq_link);
486
                /*
125 jermar 487
                 * We don't want to steal CPU-wired threads neither threads already stolen.
1 jermar 488
                 * The latter prevents threads from migrating between CPU's without ever being run.
125 jermar 489
                 * We don't want to steal threads whose FPU context is still in CPU.
73 vana 490
                 */
1 jermar 491
                spinlock_lock(&t->lock);
73 vana 492
                if ( (!(t->flags & (X_WIRED | X_STOLEN))) && (!(t->fpu_context_engaged)) ) {
115 jermar 493
 
1 jermar 494
                    /*
495
                     * Remove t from r.
496
                     */
497
 
498
                    spinlock_unlock(&t->lock);
499
 
500
                    /*
501
                     * Here we have to avoid deadlock with relink_rq(),
502
                     * because it locks cpu and r in a different order than we do.
503
                     */
504
                    if (!spinlock_trylock(&cpu->lock)) {
505
                        /* Release all locks and try again. */
506
                        spinlock_unlock(&r->lock);
507
                        cpu_priority_restore(pri);
508
                        goto restart;
509
                    }
510
                    cpu->nrdy--;
511
                    spinlock_unlock(&cpu->lock);
512
 
111 palkovsky 513
                    atomic_dec(&nrdy);
1 jermar 514
 
125 jermar 515
                    r->n--;
1 jermar 516
                    list_remove(&t->rq_link);
517
 
518
                    break;
519
                }
520
                spinlock_unlock(&t->lock);
521
                l = l->prev;
522
                t = NULL;
523
            }
524
            spinlock_unlock(&r->lock);
525
 
526
            if (t) {
527
                /*
528
                 * Ready t on local CPU
529
                 */
530
                spinlock_lock(&t->lock);
531
                #ifdef KCPULB_VERBOSE
15 jermar 532
                printf("kcpulb%d: TID %d -> cpu%d, nrdy=%d, avg=%d\n", CPU->id, t->tid, CPU->id, CPU->nrdy, nrdy / config.cpu_active);
1 jermar 533
                #endif
534
                t->flags |= X_STOLEN;
535
                spinlock_unlock(&t->lock);
536
 
537
                thread_ready(t);
538
 
539
                cpu_priority_restore(pri);
540
 
541
                if (--count == 0)
542
                    goto satisfied;
543
 
544
                /*
125 jermar 545
                 * We are not satisfied yet, focus on another CPU next time.
1 jermar 546
                 */
547
                k++;
548
 
549
                continue;
550
            }
551
            cpu_priority_restore(pri);
552
        }
553
    }
554
 
15 jermar 555
    if (CPU->nrdy) {
1 jermar 556
        /*
557
         * Be a little bit light-weight and let migrated threads run.
558
         */
559
        scheduler();
560
    }
561
    else {
562
        /*
563
         * We failed to migrate a single thread.
564
         * Something more sophisticated should be done.
565
         */
566
        scheduler();
567
    }
568
 
569
    goto not_satisfied;
125 jermar 570
 
1 jermar 571
satisfied:
572
    /*
573
     * Tell find_best_thread() to wake us up later again.
574
     */
15 jermar 575
    CPU->kcpulbstarted = 0;
1 jermar 576
    goto loop;
577
}
578
 
579
#endif /* __SMP__ */