Subversion Repositories HelenOS

Rev

Rev 99 | Rev 109 | 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>
1 jermar 45
 
46
#ifdef __SMP__
11 jermar 47
#include <arch/smp/atomic.h>
1 jermar 48
#endif /* __SMP__ */
49
 
50
/*
51
 * NOTE ON ATOMIC READS:
52
 * Some architectures cannot read __u32 atomically.
53
 * For that reason, all accesses to nrdy and the likes must be protected by spinlock.
54
 */
55
 
56
spinlock_t nrdylock;
57
volatile int nrdy;
58
 
107 decky 59
 
60
/** Initialize context switching
61
 *
62
 * Initialize context switching and lazy FPU
63
 * context switching.
64
 *
65
 */
52 vana 66
void before_thread_runs(void)
67
{
68
    before_thread_runs_arch();
57 vana 69
    fpu_context_restore(&(THREAD->saved_fpu_context));
52 vana 70
}
71
 
72
 
107 decky 73
/** Initialize scheduler
74
 *
75
 * Initialize kernel scheduler.
76
 *
77
 */
1 jermar 78
void scheduler_init(void)
79
{
80
    spinlock_initialize(&nrdylock);
81
}
82
 
107 decky 83
 
84
/** Get thread to be scheduled
85
 *
86
 * Get the optimal thread to be scheduled
87
 * according thread accounting and scheduler
88
 * policy.
89
 *
90
 * @return Thread to be scheduled.
91
 *
92
 */
1 jermar 93
struct thread *find_best_thread(void)
94
{
95
    thread_t *t;
96
    runq_t *r;
97
    int i, n;
98
 
99
loop:
100
    cpu_priority_high();
101
 
15 jermar 102
    spinlock_lock(&CPU->lock);
103
    n = CPU->nrdy;
104
    spinlock_unlock(&CPU->lock);
1 jermar 105
 
106
    cpu_priority_low();
107
 
108
    if (n == 0) {
109
        #ifdef __SMP__
110
        /*
111
         * If the load balancing thread is not running, wake it up and
112
         * set CPU-private flag that the kcpulb has been started.
113
         */
15 jermar 114
        if (test_and_set(&CPU->kcpulbstarted) == 0) {
115
                waitq_wakeup(&CPU->kcpulb_wq, 0);
1 jermar 116
            goto loop;
117
        }
118
        #endif /* __SMP__ */
119
 
120
        /*
121
         * For there was nothing to run, the CPU goes to sleep
122
         * until a hardware interrupt or an IPI comes.
123
         * This improves energy saving and hyperthreading.
124
         * On the other hand, several hardware interrupts can be ignored.
125
         */
126
         cpu_sleep();
127
         goto loop;
128
    }
129
 
130
    cpu_priority_high();
131
 
132
    for (i = 0; i<RQ_COUNT; i++) {
15 jermar 133
        r = &CPU->rq[i];
1 jermar 134
        spinlock_lock(&r->lock);
135
        if (r->n == 0) {
136
            /*
137
             * If this queue is empty, try a lower-priority queue.
138
             */
139
            spinlock_unlock(&r->lock);
140
            continue;
141
        }
142
 
143
        spinlock_lock(&nrdylock);
144
        nrdy--;
145
        spinlock_unlock(&nrdylock);    
146
 
15 jermar 147
        spinlock_lock(&CPU->lock);
148
        CPU->nrdy--;
149
        spinlock_unlock(&CPU->lock);
1 jermar 150
 
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();
15 jermar 245
                spinlock_unlock(&THREAD->lock);
246
            cpu_priority_restore(THREAD->saved_context.pri);
1 jermar 247
            return;
248
        }
15 jermar 249
        THREAD->saved_context.pri = pri;
1 jermar 250
    }
251
 
252
    /*
253
     * We may not keep the old stack.
254
     * Reason: If we kept the old stack and got blocked, for instance, in
255
     * find_best_thread(), the old thread could get rescheduled by another
256
     * CPU and overwrite the part of its own stack that was also used by
257
     * the scheduler on this CPU.
258
     *
259
     * Moreover, we have to bypass the compiler-generated POP sequence
260
     * which is fooled by SP being set to the very top of the stack.
261
     * Therefore the scheduler() function continues in
262
     * scheduler_separated_stack().
263
     */
15 jermar 264
    context_save(&CPU->saved_context);
97 jermar 265
    context_set(&CPU->saved_context, FADDR(scheduler_separated_stack), CPU->stack, CPU_STACK_SIZE);
15 jermar 266
    context_restore(&CPU->saved_context);
1 jermar 267
    /* not reached */
268
}
269
 
107 decky 270
 
271
/** Scheduler stack switch wrapper
272
 *
273
 * Second part of the scheduler() function
274
 * using new stack. Handling the actual context
275
 * switch to a new thread.
276
 *
277
 */
1 jermar 278
void scheduler_separated_stack(void)
279
{
280
    int priority;
281
 
15 jermar 282
    if (THREAD) {
283
        switch (THREAD->state) {
1 jermar 284
            case Running:
15 jermar 285
                THREAD->state = Ready;
286
                spinlock_unlock(&THREAD->lock);
287
                thread_ready(THREAD);
1 jermar 288
                break;
289
 
290
            case Exiting:
15 jermar 291
                frame_free((__address) THREAD->kstack);
292
                if (THREAD->ustack) {
293
                    frame_free((__address) THREAD->ustack);
1 jermar 294
                }
295
 
296
                /*
297
                 * Detach from the containing task.
298
                 */
15 jermar 299
                spinlock_lock(&TASK->lock);
300
                list_remove(&THREAD->th_link);
301
                spinlock_unlock(&TASK->lock);
1 jermar 302
 
15 jermar 303
                spinlock_unlock(&THREAD->lock);
1 jermar 304
 
305
                spinlock_lock(&threads_lock);
15 jermar 306
                list_remove(&THREAD->threads_link);
1 jermar 307
                spinlock_unlock(&threads_lock);
73 vana 308
 
99 jermar 309
                spinlock_lock(&CPU->lock);
310
                if(CPU->fpu_owner==THREAD) CPU->fpu_owner=NULL;
311
                spinlock_unlock(&CPU->lock);
73 vana 312
 
1 jermar 313
 
15 jermar 314
                free(THREAD);
1 jermar 315
 
316
                break;
317
 
318
            case Sleeping:
319
                /*
320
                 * Prefer the thread after it's woken up.
321
                 */
15 jermar 322
                THREAD->pri = -1;
1 jermar 323
 
324
                /*
325
                 * We need to release wq->lock which we locked in waitq_sleep().
15 jermar 326
                 * Address of wq->lock is kept in THREAD->sleep_queue.
1 jermar 327
                 */
15 jermar 328
                spinlock_unlock(&THREAD->sleep_queue->lock);
1 jermar 329
 
330
                /*
331
                 * Check for possible requests for out-of-context invocation.
332
                 */
15 jermar 333
                if (THREAD->call_me) {
334
                    THREAD->call_me(THREAD->call_me_with);
335
                    THREAD->call_me = NULL;
336
                    THREAD->call_me_with = NULL;
1 jermar 337
                }
338
 
15 jermar 339
                spinlock_unlock(&THREAD->lock);
1 jermar 340
 
341
                break;
342
 
343
            default:
344
                /*
345
                 * Entering state is unexpected.
346
                 */
15 jermar 347
                panic("tid%d: unexpected state %s\n", THREAD->tid, thread_states[THREAD->state]);
1 jermar 348
                break;
349
        }
15 jermar 350
        THREAD = NULL;
1 jermar 351
    }
352
 
15 jermar 353
    THREAD = find_best_thread();
1 jermar 354
 
15 jermar 355
    spinlock_lock(&THREAD->lock);
356
    priority = THREAD->pri;
357
    spinlock_unlock(&THREAD->lock);
1 jermar 358
 
359
    relink_rq(priority);       
360
 
15 jermar 361
    spinlock_lock(&THREAD->lock);  
1 jermar 362
 
363
    /*
364
     * If both the old and the new task are the same, lots of work is avoided.
365
     */
15 jermar 366
    if (TASK != THREAD->task) {
1 jermar 367
        vm_t *m1 = NULL;
368
        vm_t *m2;
369
 
15 jermar 370
        if (TASK) {
371
            spinlock_lock(&TASK->lock);
372
            m1 = TASK->vm;
373
            spinlock_unlock(&TASK->lock);
1 jermar 374
        }
375
 
15 jermar 376
        spinlock_lock(&THREAD->task->lock);
377
        m2 = THREAD->task->vm;
378
        spinlock_unlock(&THREAD->task->lock);
1 jermar 379
 
380
        /*
381
         * Note that it is possible for two tasks to share one vm mapping.
382
         */
383
        if (m1 != m2) {
384
            /*
385
             * Both tasks and vm mappings are different.
386
             * Replace the old one with the new one.
387
             */
388
            if (m1) {
389
                vm_uninstall(m1);
390
            }
391
            vm_install(m2);
392
        }
15 jermar 393
        TASK = THREAD->task;   
1 jermar 394
    }
395
 
15 jermar 396
    THREAD->state = Running;
1 jermar 397
 
398
    #ifdef SCHEDULER_VERBOSE
15 jermar 399
    printf("cpu%d: tid %d (pri=%d,ticks=%d,nrdy=%d)\n", CPU->id, THREAD->tid, THREAD->pri, THREAD->ticks, CPU->nrdy);
1 jermar 400
    #endif  
401
 
15 jermar 402
    context_restore(&THREAD->saved_context);
1 jermar 403
    /* not reached */
404
}
405
 
107 decky 406
 
1 jermar 407
#ifdef __SMP__
107 decky 408
/** Load balancing thread
409
 *
410
 * SMP load balancing thread, supervising thread supplies
411
 * for the CPU it's wired to.
412
 *
413
 * @param arg Generic thread argument (unused).
414
 *
1 jermar 415
 */
416
void kcpulb(void *arg)
417
{
418
    thread_t *t;
419
    int count, i, j, k = 0;
420
    pri_t pri;
421
 
422
loop:
423
    /*
424
     * Sleep until there's some work to do.
425
     */
15 jermar 426
    waitq_sleep(&CPU->kcpulb_wq);
1 jermar 427
 
428
not_satisfied:
429
    /*
430
     * Calculate the number of threads that will be migrated/stolen from
431
     * other CPU's. Note that situation can have changed between two
432
     * passes. Each time get the most up to date counts.
433
     */
434
    pri = cpu_priority_high();
15 jermar 435
    spinlock_lock(&CPU->lock);
1 jermar 436
    count = nrdy / config.cpu_active;
15 jermar 437
    count -= CPU->nrdy;
438
    spinlock_unlock(&CPU->lock);
1 jermar 439
    cpu_priority_restore(pri);
440
 
441
    if (count <= 0)
442
        goto satisfied;
443
 
444
    /*
445
     * Searching least priority queues on all CPU's first and most priority queues on all CPU's last.
446
     */
447
    for (j=RQ_COUNT-1; j >= 0; j--) {
448
        for (i=0; i < config.cpu_active; i++) {
449
            link_t *l;
450
            runq_t *r;
451
            cpu_t *cpu;
452
 
453
            cpu = &cpus[(i + k) % config.cpu_active];
454
            r = &cpu->rq[j];
455
 
456
            /*
457
             * Not interested in ourselves.
458
             * Doesn't require interrupt disabling for kcpulb is X_WIRED.
459
             */
15 jermar 460
            if (CPU == cpu)
1 jermar 461
                continue;
462
 
463
restart:        pri = cpu_priority_high();
464
            spinlock_lock(&r->lock);
465
            if (r->n == 0) {
466
                spinlock_unlock(&r->lock);
467
                cpu_priority_restore(pri);
468
                continue;
469
            }
470
 
471
            t = NULL;
472
            l = r->rq_head.prev;    /* search rq from the back */
473
            while (l != &r->rq_head) {
474
                t = list_get_instance(l, thread_t, rq_link);
475
                /*
476
                     * We don't want to steal CPU-wired threads neither threads already stolen.
477
                 * The latter prevents threads from migrating between CPU's without ever being run.
73 vana 478
                     * We don't want to steal threads whose FPU context is still in CPU
479
                 */
1 jermar 480
                spinlock_lock(&t->lock);
73 vana 481
                if ( (!(t->flags & (X_WIRED | X_STOLEN))) && (!(t->fpu_context_engaged)) ) {
1 jermar 482
                    /*
483
                     * Remove t from r.
484
                     */
485
 
486
                    spinlock_unlock(&t->lock);
487
 
488
                    /*
489
                     * Here we have to avoid deadlock with relink_rq(),
490
                     * because it locks cpu and r in a different order than we do.
491
                     */
492
                    if (!spinlock_trylock(&cpu->lock)) {
493
                        /* Release all locks and try again. */
494
                        spinlock_unlock(&r->lock);
495
                        cpu_priority_restore(pri);
496
                        goto restart;
497
                    }
498
                    cpu->nrdy--;
499
                    spinlock_unlock(&cpu->lock);
500
 
501
                    spinlock_lock(&nrdylock);
502
                    nrdy--;
503
                    spinlock_unlock(&nrdylock);                
504
 
505
                        r->n--;
506
                    list_remove(&t->rq_link);
507
 
508
                    break;
509
                }
510
                spinlock_unlock(&t->lock);
511
                l = l->prev;
512
                t = NULL;
513
            }
514
            spinlock_unlock(&r->lock);
515
 
516
            if (t) {
517
                /*
518
                 * Ready t on local CPU
519
                 */
520
                spinlock_lock(&t->lock);
521
                #ifdef KCPULB_VERBOSE
15 jermar 522
                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 523
                #endif
524
                t->flags |= X_STOLEN;
525
                spinlock_unlock(&t->lock);
526
 
527
                thread_ready(t);
528
 
529
                cpu_priority_restore(pri);
530
 
531
                if (--count == 0)
532
                    goto satisfied;
533
 
534
                /*
535
                             * We are not satisfied yet, focus on another CPU next time.
536
                 */
537
                k++;
538
 
539
                continue;
540
            }
541
            cpu_priority_restore(pri);
542
        }
543
    }
544
 
15 jermar 545
    if (CPU->nrdy) {
1 jermar 546
        /*
547
         * Be a little bit light-weight and let migrated threads run.
548
         */
549
        scheduler();
550
    }
551
    else {
552
        /*
553
         * We failed to migrate a single thread.
554
         * Something more sophisticated should be done.
555
         */
556
        scheduler();
557
    }
558
 
559
    goto not_satisfied;
560
 
561
satisfied:
562
    /*
563
     * Tell find_best_thread() to wake us up later again.
564
     */
15 jermar 565
    CPU->kcpulbstarted = 0;
1 jermar 566
    goto loop;
567
}
568
 
569
#endif /* __SMP__ */