Rev 2089 | Rev 2183 | Go to most recent revision | Details | Compare with Previous | Last modification | View Log | RSS feed
Rev | Author | Line No. | Line |
---|---|---|---|
1 | jermar | 1 | /* |
2071 | jermar | 2 | * Copyright (c) 2001-2007 Jakub Jermar |
1 | 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 | |||
1757 | jermar | 29 | /** @addtogroup genericproc |
1702 | cejka | 30 | * @{ |
31 | */ |
||
32 | |||
1248 | jermar | 33 | /** |
1702 | cejka | 34 | * @file |
1248 | jermar | 35 | * @brief Scheduler and load balancing. |
36 | * |
||
1264 | jermar | 37 | * This file contains the scheduler and kcpulb kernel thread which |
1248 | jermar | 38 | * performs load-balancing of per-CPU run queues. |
39 | */ |
||
40 | |||
1 | jermar | 41 | #include <proc/scheduler.h> |
42 | #include <proc/thread.h> |
||
43 | #include <proc/task.h> |
||
378 | jermar | 44 | #include <mm/frame.h> |
45 | #include <mm/page.h> |
||
703 | jermar | 46 | #include <mm/as.h> |
2089 | decky | 47 | #include <time/timeout.h> |
1571 | jermar | 48 | #include <time/delay.h> |
378 | jermar | 49 | #include <arch/asm.h> |
50 | #include <arch/faddr.h> |
||
2030 | decky | 51 | #include <arch/cycle.h> |
1104 | jermar | 52 | #include <atomic.h> |
378 | jermar | 53 | #include <synch/spinlock.h> |
1 | jermar | 54 | #include <config.h> |
55 | #include <context.h> |
||
2089 | decky | 56 | #include <fpu_context.h> |
1 | jermar | 57 | #include <func.h> |
58 | #include <arch.h> |
||
788 | jermar | 59 | #include <adt/list.h> |
68 | decky | 60 | #include <panic.h> |
378 | jermar | 61 | #include <cpu.h> |
195 | vana | 62 | #include <print.h> |
227 | jermar | 63 | #include <debug.h> |
1 | jermar | 64 | |
1187 | jermar | 65 | static void before_task_runs(void); |
66 | static void before_thread_runs(void); |
||
67 | static void after_thread_ran(void); |
||
898 | jermar | 68 | static void scheduler_separated_stack(void); |
195 | vana | 69 | |
898 | jermar | 70 | atomic_t nrdy; /**< Number of ready threads in the system. */ |
71 | |||
1187 | jermar | 72 | /** Carry out actions before new task runs. */ |
73 | void before_task_runs(void) |
||
74 | { |
||
75 | before_task_runs_arch(); |
||
76 | } |
||
77 | |||
897 | jermar | 78 | /** Take actions before new thread runs. |
107 | decky | 79 | * |
118 | jermar | 80 | * Perform actions that need to be |
81 | * taken before the newly selected |
||
82 | * tread is passed control. |
||
107 | decky | 83 | * |
827 | palkovsky | 84 | * THREAD->lock is locked on entry |
85 | * |
||
107 | decky | 86 | */ |
52 | vana | 87 | void before_thread_runs(void) |
88 | { |
||
309 | palkovsky | 89 | before_thread_runs_arch(); |
906 | palkovsky | 90 | #ifdef CONFIG_FPU_LAZY |
1882 | jermar | 91 | if(THREAD == CPU->fpu_owner) |
309 | palkovsky | 92 | fpu_enable(); |
93 | else |
||
94 | fpu_disable(); |
||
906 | palkovsky | 95 | #else |
309 | palkovsky | 96 | fpu_enable(); |
97 | if (THREAD->fpu_context_exists) |
||
906 | palkovsky | 98 | fpu_context_restore(THREAD->saved_fpu_context); |
309 | palkovsky | 99 | else { |
906 | palkovsky | 100 | fpu_init(); |
1882 | jermar | 101 | THREAD->fpu_context_exists = 1; |
309 | palkovsky | 102 | } |
906 | palkovsky | 103 | #endif |
52 | vana | 104 | } |
105 | |||
898 | jermar | 106 | /** Take actions after THREAD had run. |
897 | jermar | 107 | * |
108 | * Perform actions that need to be |
||
109 | * taken after the running thread |
||
898 | jermar | 110 | * had been preempted by the scheduler. |
897 | jermar | 111 | * |
112 | * THREAD->lock is locked on entry |
||
113 | * |
||
114 | */ |
||
115 | void after_thread_ran(void) |
||
116 | { |
||
117 | after_thread_ran_arch(); |
||
118 | } |
||
119 | |||
458 | decky | 120 | #ifdef CONFIG_FPU_LAZY |
309 | palkovsky | 121 | void scheduler_fpu_lazy_request(void) |
122 | { |
||
907 | palkovsky | 123 | restart: |
309 | palkovsky | 124 | fpu_enable(); |
827 | palkovsky | 125 | spinlock_lock(&CPU->lock); |
126 | |||
127 | /* Save old context */ |
||
309 | palkovsky | 128 | if (CPU->fpu_owner != NULL) { |
827 | palkovsky | 129 | spinlock_lock(&CPU->fpu_owner->lock); |
906 | palkovsky | 130 | fpu_context_save(CPU->fpu_owner->saved_fpu_context); |
309 | palkovsky | 131 | /* don't prevent migration */ |
1882 | jermar | 132 | CPU->fpu_owner->fpu_context_engaged = 0; |
827 | palkovsky | 133 | spinlock_unlock(&CPU->fpu_owner->lock); |
907 | palkovsky | 134 | CPU->fpu_owner = NULL; |
309 | palkovsky | 135 | } |
827 | palkovsky | 136 | |
137 | spinlock_lock(&THREAD->lock); |
||
898 | jermar | 138 | if (THREAD->fpu_context_exists) { |
906 | palkovsky | 139 | fpu_context_restore(THREAD->saved_fpu_context); |
898 | jermar | 140 | } else { |
906 | palkovsky | 141 | /* Allocate FPU context */ |
142 | if (!THREAD->saved_fpu_context) { |
||
143 | /* Might sleep */ |
||
144 | spinlock_unlock(&THREAD->lock); |
||
907 | palkovsky | 145 | spinlock_unlock(&CPU->lock); |
2067 | jermar | 146 | THREAD->saved_fpu_context = |
2118 | decky | 147 | (fpu_context_t *) slab_alloc(fpu_context_slab, 0); |
907 | palkovsky | 148 | /* We may have switched CPUs during slab_alloc */ |
149 | goto restart; |
||
906 | palkovsky | 150 | } |
151 | fpu_init(); |
||
1882 | jermar | 152 | THREAD->fpu_context_exists = 1; |
309 | palkovsky | 153 | } |
1882 | jermar | 154 | CPU->fpu_owner = THREAD; |
309 | palkovsky | 155 | THREAD->fpu_context_engaged = 1; |
898 | jermar | 156 | spinlock_unlock(&THREAD->lock); |
827 | palkovsky | 157 | |
158 | spinlock_unlock(&CPU->lock); |
||
309 | palkovsky | 159 | } |
160 | #endif |
||
52 | vana | 161 | |
107 | decky | 162 | /** Initialize scheduler |
163 | * |
||
164 | * Initialize kernel scheduler. |
||
165 | * |
||
166 | */ |
||
1 | jermar | 167 | void scheduler_init(void) |
168 | { |
||
169 | } |
||
170 | |||
107 | decky | 171 | /** Get thread to be scheduled |
172 | * |
||
173 | * Get the optimal thread to be scheduled |
||
109 | jermar | 174 | * according to thread accounting and scheduler |
107 | decky | 175 | * policy. |
176 | * |
||
177 | * @return Thread to be scheduled. |
||
178 | * |
||
179 | */ |
||
483 | jermar | 180 | static thread_t *find_best_thread(void) |
1 | jermar | 181 | { |
182 | thread_t *t; |
||
183 | runq_t *r; |
||
783 | palkovsky | 184 | int i; |
1 | jermar | 185 | |
227 | jermar | 186 | ASSERT(CPU != NULL); |
187 | |||
1 | jermar | 188 | loop: |
413 | jermar | 189 | interrupts_enable(); |
1 | jermar | 190 | |
783 | palkovsky | 191 | if (atomic_get(&CPU->nrdy) == 0) { |
1 | jermar | 192 | /* |
193 | * For there was nothing to run, the CPU goes to sleep |
||
194 | * until a hardware interrupt or an IPI comes. |
||
195 | * This improves energy saving and hyperthreading. |
||
196 | */ |
||
785 | jermar | 197 | |
198 | /* |
||
199 | * An interrupt might occur right now and wake up a thread. |
||
200 | * In such case, the CPU will continue to go to sleep |
||
201 | * even though there is a runnable thread. |
||
202 | */ |
||
203 | |||
1 | jermar | 204 | cpu_sleep(); |
205 | goto loop; |
||
206 | } |
||
207 | |||
413 | jermar | 208 | interrupts_disable(); |
114 | jermar | 209 | |
898 | jermar | 210 | for (i = 0; i<RQ_COUNT; i++) { |
15 | jermar | 211 | r = &CPU->rq[i]; |
1 | jermar | 212 | spinlock_lock(&r->lock); |
213 | if (r->n == 0) { |
||
214 | /* |
||
215 | * If this queue is empty, try a lower-priority queue. |
||
216 | */ |
||
217 | spinlock_unlock(&r->lock); |
||
218 | continue; |
||
219 | } |
||
213 | jermar | 220 | |
783 | palkovsky | 221 | atomic_dec(&CPU->nrdy); |
475 | jermar | 222 | atomic_dec(&nrdy); |
1 | jermar | 223 | r->n--; |
224 | |||
225 | /* |
||
226 | * Take the first thread from the queue. |
||
227 | */ |
||
228 | t = list_get_instance(r->rq_head.next, thread_t, rq_link); |
||
229 | list_remove(&t->rq_link); |
||
230 | |||
231 | spinlock_unlock(&r->lock); |
||
232 | |||
233 | spinlock_lock(&t->lock); |
||
15 | jermar | 234 | t->cpu = CPU; |
1 | jermar | 235 | |
2067 | jermar | 236 | t->ticks = us2ticks((i + 1) * 10000); |
898 | jermar | 237 | t->priority = i; /* correct rq index */ |
1 | jermar | 238 | |
239 | /* |
||
1854 | jermar | 240 | * Clear the THREAD_FLAG_STOLEN flag so that t can be migrated |
241 | * when load balancing needs emerge. |
||
1 | jermar | 242 | */ |
1854 | jermar | 243 | t->flags &= ~THREAD_FLAG_STOLEN; |
1 | jermar | 244 | spinlock_unlock(&t->lock); |
245 | |||
246 | return t; |
||
247 | } |
||
248 | goto loop; |
||
249 | |||
250 | } |
||
251 | |||
107 | decky | 252 | /** Prevent rq starvation |
253 | * |
||
254 | * Prevent low priority threads from starving in rq's. |
||
255 | * |
||
256 | * When the function decides to relink rq's, it reconnects |
||
257 | * respective pointers so that in result threads with 'pri' |
||
1708 | jermar | 258 | * greater or equal start are moved to a higher-priority queue. |
107 | decky | 259 | * |
260 | * @param start Threshold priority. |
||
261 | * |
||
1 | jermar | 262 | */ |
452 | decky | 263 | static void relink_rq(int start) |
1 | jermar | 264 | { |
265 | link_t head; |
||
266 | runq_t *r; |
||
267 | int i, n; |
||
268 | |||
269 | list_initialize(&head); |
||
15 | jermar | 270 | spinlock_lock(&CPU->lock); |
271 | if (CPU->needs_relink > NEEDS_RELINK_MAX) { |
||
2067 | jermar | 272 | for (i = start; i < RQ_COUNT - 1; i++) { |
1 | jermar | 273 | /* remember and empty rq[i + 1] */ |
15 | jermar | 274 | r = &CPU->rq[i + 1]; |
1 | jermar | 275 | spinlock_lock(&r->lock); |
276 | list_concat(&head, &r->rq_head); |
||
277 | n = r->n; |
||
278 | r->n = 0; |
||
279 | spinlock_unlock(&r->lock); |
||
280 | |||
281 | /* append rq[i + 1] to rq[i] */ |
||
15 | jermar | 282 | r = &CPU->rq[i]; |
1 | jermar | 283 | spinlock_lock(&r->lock); |
284 | list_concat(&r->rq_head, &head); |
||
285 | r->n += n; |
||
286 | spinlock_unlock(&r->lock); |
||
287 | } |
||
15 | jermar | 288 | CPU->needs_relink = 0; |
1 | jermar | 289 | } |
784 | palkovsky | 290 | spinlock_unlock(&CPU->lock); |
1 | jermar | 291 | |
292 | } |
||
293 | |||
898 | jermar | 294 | /** The scheduler |
295 | * |
||
296 | * The thread scheduling procedure. |
||
297 | * Passes control directly to |
||
298 | * scheduler_separated_stack(). |
||
299 | * |
||
300 | */ |
||
301 | void scheduler(void) |
||
302 | { |
||
303 | volatile ipl_t ipl; |
||
107 | decky | 304 | |
898 | jermar | 305 | ASSERT(CPU != NULL); |
306 | |||
307 | ipl = interrupts_disable(); |
||
308 | |||
309 | if (atomic_get(&haltstate)) |
||
310 | halt(); |
||
1007 | decky | 311 | |
898 | jermar | 312 | if (THREAD) { |
313 | spinlock_lock(&THREAD->lock); |
||
2030 | decky | 314 | |
315 | /* Update thread accounting */ |
||
316 | THREAD->cycles += get_cycle() - THREAD->last_cycle; |
||
317 | |||
906 | palkovsky | 318 | #ifndef CONFIG_FPU_LAZY |
319 | fpu_context_save(THREAD->saved_fpu_context); |
||
320 | #endif |
||
898 | jermar | 321 | if (!context_save(&THREAD->saved_context)) { |
322 | /* |
||
323 | * This is the place where threads leave scheduler(); |
||
324 | */ |
||
2030 | decky | 325 | |
326 | /* Save current CPU cycle */ |
||
327 | THREAD->last_cycle = get_cycle(); |
||
328 | |||
898 | jermar | 329 | spinlock_unlock(&THREAD->lock); |
330 | interrupts_restore(THREAD->saved_context.ipl); |
||
1007 | decky | 331 | |
898 | jermar | 332 | return; |
333 | } |
||
334 | |||
335 | /* |
||
2067 | jermar | 336 | * Interrupt priority level of preempted thread is recorded |
337 | * here to facilitate scheduler() invocations from |
||
338 | * interrupts_disable()'d code (e.g. waitq_sleep_timeout()). |
||
898 | jermar | 339 | */ |
340 | THREAD->saved_context.ipl = ipl; |
||
341 | } |
||
342 | |||
343 | /* |
||
344 | * Through the 'THE' structure, we keep track of THREAD, TASK, CPU, VM |
||
345 | * and preemption counter. At this point THE could be coming either |
||
346 | * from THREAD's or CPU's stack. |
||
347 | */ |
||
348 | the_copy(THE, (the_t *) CPU->stack); |
||
349 | |||
350 | /* |
||
351 | * We may not keep the old stack. |
||
352 | * Reason: If we kept the old stack and got blocked, for instance, in |
||
353 | * find_best_thread(), the old thread could get rescheduled by another |
||
354 | * CPU and overwrite the part of its own stack that was also used by |
||
355 | * the scheduler on this CPU. |
||
356 | * |
||
357 | * Moreover, we have to bypass the compiler-generated POP sequence |
||
358 | * which is fooled by SP being set to the very top of the stack. |
||
359 | * Therefore the scheduler() function continues in |
||
360 | * scheduler_separated_stack(). |
||
361 | */ |
||
362 | context_save(&CPU->saved_context); |
||
1854 | jermar | 363 | context_set(&CPU->saved_context, FADDR(scheduler_separated_stack), |
2087 | jermar | 364 | (uintptr_t) CPU->stack, CPU_STACK_SIZE); |
898 | jermar | 365 | context_restore(&CPU->saved_context); |
366 | /* not reached */ |
||
367 | } |
||
368 | |||
107 | decky | 369 | /** Scheduler stack switch wrapper |
370 | * |
||
371 | * Second part of the scheduler() function |
||
372 | * using new stack. Handling the actual context |
||
373 | * switch to a new thread. |
||
374 | * |
||
787 | palkovsky | 375 | * Assume THREAD->lock is held. |
107 | decky | 376 | */ |
898 | jermar | 377 | void scheduler_separated_stack(void) |
1 | jermar | 378 | { |
379 | int priority; |
||
1007 | decky | 380 | |
227 | jermar | 381 | ASSERT(CPU != NULL); |
1007 | decky | 382 | |
15 | jermar | 383 | if (THREAD) { |
898 | jermar | 384 | /* must be run after the switch to scheduler stack */ |
897 | jermar | 385 | after_thread_ran(); |
386 | |||
15 | jermar | 387 | switch (THREAD->state) { |
1888 | jermar | 388 | case Running: |
125 | jermar | 389 | spinlock_unlock(&THREAD->lock); |
390 | thread_ready(THREAD); |
||
391 | break; |
||
1 | jermar | 392 | |
1888 | jermar | 393 | case Exiting: |
1571 | jermar | 394 | repeat: |
2040 | decky | 395 | if (THREAD->detached) { |
1571 | jermar | 396 | thread_destroy(THREAD); |
397 | } else { |
||
398 | /* |
||
2067 | jermar | 399 | * The thread structure is kept allocated until |
400 | * somebody calls thread_detach() on it. |
||
1571 | jermar | 401 | */ |
402 | if (!spinlock_trylock(&THREAD->join_wq.lock)) { |
||
403 | /* |
||
404 | * Avoid deadlock. |
||
405 | */ |
||
406 | spinlock_unlock(&THREAD->lock); |
||
407 | delay(10); |
||
408 | spinlock_lock(&THREAD->lock); |
||
409 | goto repeat; |
||
410 | } |
||
411 | _waitq_wakeup_unsafe(&THREAD->join_wq, false); |
||
412 | spinlock_unlock(&THREAD->join_wq.lock); |
||
413 | |||
414 | THREAD->state = Undead; |
||
415 | spinlock_unlock(&THREAD->lock); |
||
416 | } |
||
125 | jermar | 417 | break; |
787 | palkovsky | 418 | |
1888 | jermar | 419 | case Sleeping: |
125 | jermar | 420 | /* |
421 | * Prefer the thread after it's woken up. |
||
422 | */ |
||
413 | jermar | 423 | THREAD->priority = -1; |
1 | jermar | 424 | |
125 | jermar | 425 | /* |
2067 | jermar | 426 | * We need to release wq->lock which we locked in |
427 | * waitq_sleep(). Address of wq->lock is kept in |
||
428 | * THREAD->sleep_queue. |
||
125 | jermar | 429 | */ |
430 | spinlock_unlock(&THREAD->sleep_queue->lock); |
||
1 | jermar | 431 | |
125 | jermar | 432 | /* |
2067 | jermar | 433 | * Check for possible requests for out-of-context |
434 | * invocation. |
||
125 | jermar | 435 | */ |
436 | if (THREAD->call_me) { |
||
437 | THREAD->call_me(THREAD->call_me_with); |
||
438 | THREAD->call_me = NULL; |
||
439 | THREAD->call_me_with = NULL; |
||
440 | } |
||
1 | jermar | 441 | |
125 | jermar | 442 | spinlock_unlock(&THREAD->lock); |
1 | jermar | 443 | |
125 | jermar | 444 | break; |
445 | |||
1888 | jermar | 446 | default: |
125 | jermar | 447 | /* |
448 | * Entering state is unexpected. |
||
449 | */ |
||
2067 | jermar | 450 | panic("tid%d: unexpected state %s\n", THREAD->tid, |
451 | thread_states[THREAD->state]); |
||
125 | jermar | 452 | break; |
1 | jermar | 453 | } |
897 | jermar | 454 | |
15 | jermar | 455 | THREAD = NULL; |
1 | jermar | 456 | } |
198 | jermar | 457 | |
15 | jermar | 458 | THREAD = find_best_thread(); |
1 | jermar | 459 | |
15 | jermar | 460 | spinlock_lock(&THREAD->lock); |
413 | jermar | 461 | priority = THREAD->priority; |
15 | jermar | 462 | spinlock_unlock(&THREAD->lock); |
192 | jermar | 463 | |
1 | jermar | 464 | relink_rq(priority); |
465 | |||
466 | /* |
||
2067 | jermar | 467 | * If both the old and the new task are the same, lots of work is |
468 | * avoided. |
||
1 | jermar | 469 | */ |
15 | jermar | 470 | if (TASK != THREAD->task) { |
703 | jermar | 471 | as_t *as1 = NULL; |
472 | as_t *as2; |
||
1 | jermar | 473 | |
15 | jermar | 474 | if (TASK) { |
475 | spinlock_lock(&TASK->lock); |
||
703 | jermar | 476 | as1 = TASK->as; |
15 | jermar | 477 | spinlock_unlock(&TASK->lock); |
1 | jermar | 478 | } |
479 | |||
15 | jermar | 480 | spinlock_lock(&THREAD->task->lock); |
703 | jermar | 481 | as2 = THREAD->task->as; |
15 | jermar | 482 | spinlock_unlock(&THREAD->task->lock); |
1 | jermar | 483 | |
484 | /* |
||
2067 | jermar | 485 | * Note that it is possible for two tasks to share one address |
486 | * space. |
||
1 | jermar | 487 | */ |
703 | jermar | 488 | if (as1 != as2) { |
1 | jermar | 489 | /* |
703 | jermar | 490 | * Both tasks and address spaces are different. |
1 | jermar | 491 | * Replace the old one with the new one. |
492 | */ |
||
823 | jermar | 493 | as_switch(as1, as2); |
1 | jermar | 494 | } |
906 | palkovsky | 495 | TASK = THREAD->task; |
1187 | jermar | 496 | before_task_runs(); |
1 | jermar | 497 | } |
498 | |||
1380 | jermar | 499 | spinlock_lock(&THREAD->lock); |
15 | jermar | 500 | THREAD->state = Running; |
1 | jermar | 501 | |
906 | palkovsky | 502 | #ifdef SCHEDULER_VERBOSE |
2067 | jermar | 503 | printf("cpu%d: tid %d (priority=%d, ticks=%lld, nrdy=%ld)\n", |
2087 | jermar | 504 | CPU->id, THREAD->tid, THREAD->priority, THREAD->ticks, |
505 | atomic_get(&CPU->nrdy)); |
||
906 | palkovsky | 506 | #endif |
1 | jermar | 507 | |
213 | jermar | 508 | /* |
897 | jermar | 509 | * Some architectures provide late kernel PA2KA(identity) |
510 | * mapping in a page fault handler. However, the page fault |
||
511 | * handler uses the kernel stack of the running thread and |
||
512 | * therefore cannot be used to map it. The kernel stack, if |
||
513 | * necessary, is to be mapped in before_thread_runs(). This |
||
514 | * function must be executed before the switch to the new stack. |
||
515 | */ |
||
516 | before_thread_runs(); |
||
517 | |||
518 | /* |
||
2067 | jermar | 519 | * Copy the knowledge of CPU, TASK, THREAD and preemption counter to |
520 | * thread's stack. |
||
213 | jermar | 521 | */ |
184 | jermar | 522 | the_copy(THE, (the_t *) THREAD->kstack); |
523 | |||
15 | jermar | 524 | context_restore(&THREAD->saved_context); |
1 | jermar | 525 | /* not reached */ |
526 | } |
||
527 | |||
458 | decky | 528 | #ifdef CONFIG_SMP |
107 | decky | 529 | /** Load balancing thread |
530 | * |
||
531 | * SMP load balancing thread, supervising thread supplies |
||
532 | * for the CPU it's wired to. |
||
533 | * |
||
534 | * @param arg Generic thread argument (unused). |
||
535 | * |
||
1 | jermar | 536 | */ |
537 | void kcpulb(void *arg) |
||
538 | { |
||
539 | thread_t *t; |
||
2118 | decky | 540 | int count, average, j, k = 0; |
541 | unsigned int i; |
||
413 | jermar | 542 | ipl_t ipl; |
1 | jermar | 543 | |
1576 | jermar | 544 | /* |
545 | * Detach kcpulb as nobody will call thread_join_timeout() on it. |
||
546 | */ |
||
547 | thread_detach(THREAD); |
||
548 | |||
1 | jermar | 549 | loop: |
550 | /* |
||
779 | jermar | 551 | * Work in 1s intervals. |
1 | jermar | 552 | */ |
779 | jermar | 553 | thread_sleep(1); |
1 | jermar | 554 | |
555 | not_satisfied: |
||
556 | /* |
||
557 | * Calculate the number of threads that will be migrated/stolen from |
||
558 | * other CPU's. Note that situation can have changed between two |
||
559 | * passes. Each time get the most up to date counts. |
||
560 | */ |
||
784 | palkovsky | 561 | average = atomic_get(&nrdy) / config.cpu_active + 1; |
783 | palkovsky | 562 | count = average - atomic_get(&CPU->nrdy); |
1 | jermar | 563 | |
784 | palkovsky | 564 | if (count <= 0) |
1 | jermar | 565 | goto satisfied; |
566 | |||
567 | /* |
||
2067 | jermar | 568 | * Searching least priority queues on all CPU's first and most priority |
569 | * queues on all CPU's last. |
||
1 | jermar | 570 | */ |
2067 | jermar | 571 | for (j= RQ_COUNT - 1; j >= 0; j--) { |
572 | for (i = 0; i < config.cpu_active; i++) { |
||
1 | jermar | 573 | link_t *l; |
574 | runq_t *r; |
||
575 | cpu_t *cpu; |
||
576 | |||
577 | cpu = &cpus[(i + k) % config.cpu_active]; |
||
578 | |||
579 | /* |
||
580 | * Not interested in ourselves. |
||
2067 | jermar | 581 | * Doesn't require interrupt disabling for kcpulb has |
582 | * THREAD_FLAG_WIRED. |
||
1 | jermar | 583 | */ |
15 | jermar | 584 | if (CPU == cpu) |
783 | palkovsky | 585 | continue; |
586 | if (atomic_get(&cpu->nrdy) <= average) |
||
587 | continue; |
||
1 | jermar | 588 | |
784 | palkovsky | 589 | ipl = interrupts_disable(); |
115 | jermar | 590 | r = &cpu->rq[j]; |
1 | jermar | 591 | spinlock_lock(&r->lock); |
592 | if (r->n == 0) { |
||
593 | spinlock_unlock(&r->lock); |
||
413 | jermar | 594 | interrupts_restore(ipl); |
1 | jermar | 595 | continue; |
596 | } |
||
597 | |||
598 | t = NULL; |
||
599 | l = r->rq_head.prev; /* search rq from the back */ |
||
600 | while (l != &r->rq_head) { |
||
601 | t = list_get_instance(l, thread_t, rq_link); |
||
602 | /* |
||
2067 | jermar | 603 | * We don't want to steal CPU-wired threads |
604 | * neither threads already stolen. The latter |
||
605 | * prevents threads from migrating between CPU's |
||
606 | * without ever being run. We don't want to |
||
607 | * steal threads whose FPU context is still in |
||
608 | * CPU. |
||
73 | vana | 609 | */ |
1 | jermar | 610 | spinlock_lock(&t->lock); |
2067 | jermar | 611 | if ((!(t->flags & (THREAD_FLAG_WIRED | |
612 | THREAD_FLAG_STOLEN))) && |
||
1854 | jermar | 613 | (!(t->fpu_context_engaged)) ) { |
1 | jermar | 614 | /* |
615 | * Remove t from r. |
||
616 | */ |
||
617 | spinlock_unlock(&t->lock); |
||
618 | |||
783 | palkovsky | 619 | atomic_dec(&cpu->nrdy); |
475 | jermar | 620 | atomic_dec(&nrdy); |
1 | jermar | 621 | |
125 | jermar | 622 | r->n--; |
1 | jermar | 623 | list_remove(&t->rq_link); |
624 | |||
625 | break; |
||
626 | } |
||
627 | spinlock_unlock(&t->lock); |
||
628 | l = l->prev; |
||
629 | t = NULL; |
||
630 | } |
||
631 | spinlock_unlock(&r->lock); |
||
632 | |||
633 | if (t) { |
||
634 | /* |
||
635 | * Ready t on local CPU |
||
636 | */ |
||
637 | spinlock_lock(&t->lock); |
||
906 | palkovsky | 638 | #ifdef KCPULB_VERBOSE |
2067 | jermar | 639 | printf("kcpulb%d: TID %d -> cpu%d, nrdy=%ld, " |
2087 | jermar | 640 | "avg=%nd\n", CPU->id, t->tid, CPU->id, |
641 | atomic_get(&CPU->nrdy), |
||
642 | atomic_get(&nrdy) / config.cpu_active); |
||
906 | palkovsky | 643 | #endif |
1854 | jermar | 644 | t->flags |= THREAD_FLAG_STOLEN; |
1115 | jermar | 645 | t->state = Entering; |
1 | jermar | 646 | spinlock_unlock(&t->lock); |
647 | |||
648 | thread_ready(t); |
||
649 | |||
413 | jermar | 650 | interrupts_restore(ipl); |
1 | jermar | 651 | |
652 | if (--count == 0) |
||
653 | goto satisfied; |
||
654 | |||
655 | /* |
||
2067 | jermar | 656 | * We are not satisfied yet, focus on another |
657 | * CPU next time. |
||
1 | jermar | 658 | */ |
659 | k++; |
||
660 | |||
661 | continue; |
||
662 | } |
||
413 | jermar | 663 | interrupts_restore(ipl); |
1 | jermar | 664 | } |
665 | } |
||
666 | |||
783 | palkovsky | 667 | if (atomic_get(&CPU->nrdy)) { |
1 | jermar | 668 | /* |
669 | * Be a little bit light-weight and let migrated threads run. |
||
670 | */ |
||
671 | scheduler(); |
||
779 | jermar | 672 | } else { |
1 | jermar | 673 | /* |
674 | * We failed to migrate a single thread. |
||
779 | jermar | 675 | * Give up this turn. |
1 | jermar | 676 | */ |
779 | jermar | 677 | goto loop; |
1 | jermar | 678 | } |
679 | |||
680 | goto not_satisfied; |
||
125 | jermar | 681 | |
1 | jermar | 682 | satisfied: |
683 | goto loop; |
||
684 | } |
||
685 | |||
458 | decky | 686 | #endif /* CONFIG_SMP */ |
775 | palkovsky | 687 | |
688 | |||
689 | /** Print information about threads & scheduler queues */ |
||
690 | void sched_print_list(void) |
||
691 | { |
||
692 | ipl_t ipl; |
||
2118 | decky | 693 | unsigned int cpu, i; |
775 | palkovsky | 694 | runq_t *r; |
695 | thread_t *t; |
||
696 | link_t *cur; |
||
697 | |||
698 | /* We are going to mess with scheduler structures, |
||
699 | * let's not be interrupted */ |
||
700 | ipl = interrupts_disable(); |
||
2118 | decky | 701 | for (cpu = 0; cpu < config.cpu_count; cpu++) { |
898 | jermar | 702 | |
775 | palkovsky | 703 | if (!cpus[cpu].active) |
704 | continue; |
||
898 | jermar | 705 | |
775 | palkovsky | 706 | spinlock_lock(&cpus[cpu].lock); |
1221 | decky | 707 | printf("cpu%d: address=%p, nrdy=%ld, needs_relink=%ld\n", |
2087 | jermar | 708 | cpus[cpu].id, &cpus[cpu], atomic_get(&cpus[cpu].nrdy), |
709 | cpus[cpu].needs_relink); |
||
775 | palkovsky | 710 | |
2067 | jermar | 711 | for (i = 0; i < RQ_COUNT; i++) { |
775 | palkovsky | 712 | r = &cpus[cpu].rq[i]; |
713 | spinlock_lock(&r->lock); |
||
714 | if (!r->n) { |
||
715 | spinlock_unlock(&r->lock); |
||
716 | continue; |
||
717 | } |
||
898 | jermar | 718 | printf("\trq[%d]: ", i); |
2067 | jermar | 719 | for (cur = r->rq_head.next; cur != &r->rq_head; |
720 | cur = cur->next) { |
||
775 | palkovsky | 721 | t = list_get_instance(cur, thread_t, rq_link); |
722 | printf("%d(%s) ", t->tid, |
||
2087 | jermar | 723 | thread_states[t->state]); |
775 | palkovsky | 724 | } |
725 | printf("\n"); |
||
726 | spinlock_unlock(&r->lock); |
||
727 | } |
||
728 | spinlock_unlock(&cpus[cpu].lock); |
||
729 | } |
||
730 | |||
731 | interrupts_restore(ipl); |
||
732 | } |
||
1702 | cejka | 733 | |
1757 | jermar | 734 | /** @} |
1702 | cejka | 735 | */ |