Rev 1708 | 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 | |||
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> |
1571 | jermar | 47 | #include <time/delay.h> |
378 | jermar | 48 | #include <arch/asm.h> |
49 | #include <arch/faddr.h> |
||
1104 | jermar | 50 | #include <atomic.h> |
378 | jermar | 51 | #include <synch/spinlock.h> |
1 | jermar | 52 | #include <config.h> |
53 | #include <context.h> |
||
54 | #include <func.h> |
||
55 | #include <arch.h> |
||
788 | jermar | 56 | #include <adt/list.h> |
68 | decky | 57 | #include <panic.h> |
1 | jermar | 58 | #include <typedefs.h> |
378 | jermar | 59 | #include <cpu.h> |
195 | vana | 60 | #include <print.h> |
227 | jermar | 61 | #include <debug.h> |
1 | jermar | 62 | |
1187 | jermar | 63 | static void before_task_runs(void); |
64 | static void before_thread_runs(void); |
||
65 | static void after_thread_ran(void); |
||
898 | jermar | 66 | static void scheduler_separated_stack(void); |
195 | vana | 67 | |
898 | jermar | 68 | atomic_t nrdy; /**< Number of ready threads in the system. */ |
69 | |||
1187 | jermar | 70 | /** Carry out actions before new task runs. */ |
71 | void before_task_runs(void) |
||
72 | { |
||
73 | before_task_runs_arch(); |
||
74 | } |
||
75 | |||
897 | jermar | 76 | /** Take actions before new thread runs. |
107 | decky | 77 | * |
118 | jermar | 78 | * Perform actions that need to be |
79 | * taken before the newly selected |
||
80 | * tread is passed control. |
||
107 | decky | 81 | * |
827 | palkovsky | 82 | * THREAD->lock is locked on entry |
83 | * |
||
107 | decky | 84 | */ |
52 | vana | 85 | void before_thread_runs(void) |
86 | { |
||
309 | palkovsky | 87 | before_thread_runs_arch(); |
906 | palkovsky | 88 | #ifdef CONFIG_FPU_LAZY |
309 | palkovsky | 89 | if(THREAD==CPU->fpu_owner) |
90 | fpu_enable(); |
||
91 | else |
||
92 | fpu_disable(); |
||
906 | palkovsky | 93 | #else |
309 | palkovsky | 94 | fpu_enable(); |
95 | if (THREAD->fpu_context_exists) |
||
906 | palkovsky | 96 | fpu_context_restore(THREAD->saved_fpu_context); |
309 | palkovsky | 97 | else { |
906 | palkovsky | 98 | fpu_init(); |
309 | palkovsky | 99 | THREAD->fpu_context_exists=1; |
100 | } |
||
906 | palkovsky | 101 | #endif |
52 | vana | 102 | } |
103 | |||
898 | jermar | 104 | /** Take actions after THREAD had run. |
897 | jermar | 105 | * |
106 | * Perform actions that need to be |
||
107 | * taken after the running thread |
||
898 | jermar | 108 | * had been preempted by the scheduler. |
897 | jermar | 109 | * |
110 | * THREAD->lock is locked on entry |
||
111 | * |
||
112 | */ |
||
113 | void after_thread_ran(void) |
||
114 | { |
||
115 | after_thread_ran_arch(); |
||
116 | } |
||
117 | |||
458 | decky | 118 | #ifdef CONFIG_FPU_LAZY |
309 | palkovsky | 119 | void scheduler_fpu_lazy_request(void) |
120 | { |
||
907 | palkovsky | 121 | restart: |
309 | palkovsky | 122 | fpu_enable(); |
827 | palkovsky | 123 | spinlock_lock(&CPU->lock); |
124 | |||
125 | /* Save old context */ |
||
309 | palkovsky | 126 | if (CPU->fpu_owner != NULL) { |
827 | palkovsky | 127 | spinlock_lock(&CPU->fpu_owner->lock); |
906 | palkovsky | 128 | fpu_context_save(CPU->fpu_owner->saved_fpu_context); |
309 | palkovsky | 129 | /* don't prevent migration */ |
130 | CPU->fpu_owner->fpu_context_engaged=0; |
||
827 | palkovsky | 131 | spinlock_unlock(&CPU->fpu_owner->lock); |
907 | palkovsky | 132 | CPU->fpu_owner = NULL; |
309 | palkovsky | 133 | } |
827 | palkovsky | 134 | |
135 | spinlock_lock(&THREAD->lock); |
||
898 | jermar | 136 | if (THREAD->fpu_context_exists) { |
906 | palkovsky | 137 | fpu_context_restore(THREAD->saved_fpu_context); |
898 | jermar | 138 | } else { |
906 | palkovsky | 139 | /* Allocate FPU context */ |
140 | if (!THREAD->saved_fpu_context) { |
||
141 | /* Might sleep */ |
||
142 | spinlock_unlock(&THREAD->lock); |
||
907 | palkovsky | 143 | spinlock_unlock(&CPU->lock); |
906 | palkovsky | 144 | THREAD->saved_fpu_context = slab_alloc(fpu_context_slab, |
145 | 0); |
||
907 | palkovsky | 146 | /* We may have switched CPUs during slab_alloc */ |
147 | goto restart; |
||
906 | palkovsky | 148 | } |
149 | fpu_init(); |
||
309 | palkovsky | 150 | THREAD->fpu_context_exists=1; |
151 | } |
||
152 | CPU->fpu_owner=THREAD; |
||
153 | THREAD->fpu_context_engaged = 1; |
||
898 | jermar | 154 | spinlock_unlock(&THREAD->lock); |
827 | palkovsky | 155 | |
156 | spinlock_unlock(&CPU->lock); |
||
309 | palkovsky | 157 | } |
158 | #endif |
||
52 | vana | 159 | |
107 | decky | 160 | /** Initialize scheduler |
161 | * |
||
162 | * Initialize kernel scheduler. |
||
163 | * |
||
164 | */ |
||
1 | jermar | 165 | void scheduler_init(void) |
166 | { |
||
167 | } |
||
168 | |||
107 | decky | 169 | /** Get thread to be scheduled |
170 | * |
||
171 | * Get the optimal thread to be scheduled |
||
109 | jermar | 172 | * according to thread accounting and scheduler |
107 | decky | 173 | * policy. |
174 | * |
||
175 | * @return Thread to be scheduled. |
||
176 | * |
||
177 | */ |
||
483 | jermar | 178 | static thread_t *find_best_thread(void) |
1 | jermar | 179 | { |
180 | thread_t *t; |
||
181 | runq_t *r; |
||
783 | palkovsky | 182 | int i; |
1 | jermar | 183 | |
227 | jermar | 184 | ASSERT(CPU != NULL); |
185 | |||
1 | jermar | 186 | loop: |
413 | jermar | 187 | interrupts_enable(); |
1 | jermar | 188 | |
783 | palkovsky | 189 | if (atomic_get(&CPU->nrdy) == 0) { |
1 | jermar | 190 | /* |
191 | * For there was nothing to run, the CPU goes to sleep |
||
192 | * until a hardware interrupt or an IPI comes. |
||
193 | * This improves energy saving and hyperthreading. |
||
194 | */ |
||
785 | jermar | 195 | |
196 | /* |
||
197 | * An interrupt might occur right now and wake up a thread. |
||
198 | * In such case, the CPU will continue to go to sleep |
||
199 | * even though there is a runnable thread. |
||
200 | */ |
||
201 | |||
1 | jermar | 202 | cpu_sleep(); |
203 | goto loop; |
||
204 | } |
||
205 | |||
413 | jermar | 206 | interrupts_disable(); |
114 | jermar | 207 | |
898 | jermar | 208 | for (i = 0; i<RQ_COUNT; i++) { |
15 | jermar | 209 | r = &CPU->rq[i]; |
1 | jermar | 210 | spinlock_lock(&r->lock); |
211 | if (r->n == 0) { |
||
212 | /* |
||
213 | * If this queue is empty, try a lower-priority queue. |
||
214 | */ |
||
215 | spinlock_unlock(&r->lock); |
||
216 | continue; |
||
217 | } |
||
213 | jermar | 218 | |
783 | palkovsky | 219 | atomic_dec(&CPU->nrdy); |
475 | jermar | 220 | atomic_dec(&nrdy); |
1 | jermar | 221 | r->n--; |
222 | |||
223 | /* |
||
224 | * Take the first thread from the queue. |
||
225 | */ |
||
226 | t = list_get_instance(r->rq_head.next, thread_t, rq_link); |
||
227 | list_remove(&t->rq_link); |
||
228 | |||
229 | spinlock_unlock(&r->lock); |
||
230 | |||
231 | spinlock_lock(&t->lock); |
||
15 | jermar | 232 | t->cpu = CPU; |
1 | jermar | 233 | |
234 | t->ticks = us2ticks((i+1)*10000); |
||
898 | jermar | 235 | t->priority = i; /* correct rq index */ |
1 | jermar | 236 | |
237 | /* |
||
238 | * Clear the X_STOLEN flag so that t can be migrated when load balancing needs emerge. |
||
239 | */ |
||
240 | t->flags &= ~X_STOLEN; |
||
241 | spinlock_unlock(&t->lock); |
||
242 | |||
243 | return t; |
||
244 | } |
||
245 | goto loop; |
||
246 | |||
247 | } |
||
248 | |||
107 | decky | 249 | /** Prevent rq starvation |
250 | * |
||
251 | * Prevent low priority threads from starving in rq's. |
||
252 | * |
||
253 | * When the function decides to relink rq's, it reconnects |
||
254 | * respective pointers so that in result threads with 'pri' |
||
1708 | jermar | 255 | * greater or equal start are moved to a higher-priority queue. |
107 | decky | 256 | * |
257 | * @param start Threshold priority. |
||
258 | * |
||
1 | jermar | 259 | */ |
452 | decky | 260 | static void relink_rq(int start) |
1 | jermar | 261 | { |
262 | link_t head; |
||
263 | runq_t *r; |
||
264 | int i, n; |
||
265 | |||
266 | list_initialize(&head); |
||
15 | jermar | 267 | spinlock_lock(&CPU->lock); |
268 | if (CPU->needs_relink > NEEDS_RELINK_MAX) { |
||
1 | jermar | 269 | for (i = start; i<RQ_COUNT-1; i++) { |
270 | /* remember and empty rq[i + 1] */ |
||
15 | jermar | 271 | r = &CPU->rq[i + 1]; |
1 | jermar | 272 | spinlock_lock(&r->lock); |
273 | list_concat(&head, &r->rq_head); |
||
274 | n = r->n; |
||
275 | r->n = 0; |
||
276 | spinlock_unlock(&r->lock); |
||
277 | |||
278 | /* append rq[i + 1] to rq[i] */ |
||
15 | jermar | 279 | r = &CPU->rq[i]; |
1 | jermar | 280 | spinlock_lock(&r->lock); |
281 | list_concat(&r->rq_head, &head); |
||
282 | r->n += n; |
||
283 | spinlock_unlock(&r->lock); |
||
284 | } |
||
15 | jermar | 285 | CPU->needs_relink = 0; |
1 | jermar | 286 | } |
784 | palkovsky | 287 | spinlock_unlock(&CPU->lock); |
1 | jermar | 288 | |
289 | } |
||
290 | |||
898 | jermar | 291 | /** The scheduler |
292 | * |
||
293 | * The thread scheduling procedure. |
||
294 | * Passes control directly to |
||
295 | * scheduler_separated_stack(). |
||
296 | * |
||
297 | */ |
||
298 | void scheduler(void) |
||
299 | { |
||
300 | volatile ipl_t ipl; |
||
107 | decky | 301 | |
898 | jermar | 302 | ASSERT(CPU != NULL); |
303 | |||
304 | ipl = interrupts_disable(); |
||
305 | |||
306 | if (atomic_get(&haltstate)) |
||
307 | halt(); |
||
1007 | decky | 308 | |
898 | jermar | 309 | if (THREAD) { |
310 | spinlock_lock(&THREAD->lock); |
||
906 | palkovsky | 311 | #ifndef CONFIG_FPU_LAZY |
312 | fpu_context_save(THREAD->saved_fpu_context); |
||
313 | #endif |
||
898 | jermar | 314 | if (!context_save(&THREAD->saved_context)) { |
315 | /* |
||
316 | * This is the place where threads leave scheduler(); |
||
317 | */ |
||
318 | spinlock_unlock(&THREAD->lock); |
||
319 | interrupts_restore(THREAD->saved_context.ipl); |
||
1007 | decky | 320 | |
898 | jermar | 321 | return; |
322 | } |
||
323 | |||
324 | /* |
||
325 | * Interrupt priority level of preempted thread is recorded here |
||
326 | * to facilitate scheduler() invocations from interrupts_disable()'d |
||
327 | * code (e.g. waitq_sleep_timeout()). |
||
328 | */ |
||
329 | THREAD->saved_context.ipl = ipl; |
||
330 | } |
||
331 | |||
332 | /* |
||
333 | * Through the 'THE' structure, we keep track of THREAD, TASK, CPU, VM |
||
334 | * and preemption counter. At this point THE could be coming either |
||
335 | * from THREAD's or CPU's stack. |
||
336 | */ |
||
337 | the_copy(THE, (the_t *) CPU->stack); |
||
338 | |||
339 | /* |
||
340 | * We may not keep the old stack. |
||
341 | * Reason: If we kept the old stack and got blocked, for instance, in |
||
342 | * find_best_thread(), the old thread could get rescheduled by another |
||
343 | * CPU and overwrite the part of its own stack that was also used by |
||
344 | * the scheduler on this CPU. |
||
345 | * |
||
346 | * Moreover, we have to bypass the compiler-generated POP sequence |
||
347 | * which is fooled by SP being set to the very top of the stack. |
||
348 | * Therefore the scheduler() function continues in |
||
349 | * scheduler_separated_stack(). |
||
350 | */ |
||
351 | context_save(&CPU->saved_context); |
||
352 | context_set(&CPU->saved_context, FADDR(scheduler_separated_stack), (__address) CPU->stack, CPU_STACK_SIZE); |
||
353 | context_restore(&CPU->saved_context); |
||
354 | /* not reached */ |
||
355 | } |
||
356 | |||
107 | decky | 357 | /** Scheduler stack switch wrapper |
358 | * |
||
359 | * Second part of the scheduler() function |
||
360 | * using new stack. Handling the actual context |
||
361 | * switch to a new thread. |
||
362 | * |
||
787 | palkovsky | 363 | * Assume THREAD->lock is held. |
107 | decky | 364 | */ |
898 | jermar | 365 | void scheduler_separated_stack(void) |
1 | jermar | 366 | { |
367 | int priority; |
||
1007 | decky | 368 | |
227 | jermar | 369 | ASSERT(CPU != NULL); |
1007 | decky | 370 | |
15 | jermar | 371 | if (THREAD) { |
898 | jermar | 372 | /* must be run after the switch to scheduler stack */ |
897 | jermar | 373 | after_thread_ran(); |
374 | |||
15 | jermar | 375 | switch (THREAD->state) { |
1 | jermar | 376 | case Running: |
125 | jermar | 377 | spinlock_unlock(&THREAD->lock); |
378 | thread_ready(THREAD); |
||
379 | break; |
||
1 | jermar | 380 | |
381 | case Exiting: |
||
1571 | jermar | 382 | repeat: |
383 | if (THREAD->detached) { |
||
384 | thread_destroy(THREAD); |
||
385 | } else { |
||
386 | /* |
||
387 | * The thread structure is kept allocated until somebody |
||
388 | * calls thread_detach() on it. |
||
389 | */ |
||
390 | if (!spinlock_trylock(&THREAD->join_wq.lock)) { |
||
391 | /* |
||
392 | * Avoid deadlock. |
||
393 | */ |
||
394 | spinlock_unlock(&THREAD->lock); |
||
395 | delay(10); |
||
396 | spinlock_lock(&THREAD->lock); |
||
397 | goto repeat; |
||
398 | } |
||
399 | _waitq_wakeup_unsafe(&THREAD->join_wq, false); |
||
400 | spinlock_unlock(&THREAD->join_wq.lock); |
||
401 | |||
402 | THREAD->state = Undead; |
||
403 | spinlock_unlock(&THREAD->lock); |
||
404 | } |
||
125 | jermar | 405 | break; |
787 | palkovsky | 406 | |
1 | jermar | 407 | case Sleeping: |
125 | jermar | 408 | /* |
409 | * Prefer the thread after it's woken up. |
||
410 | */ |
||
413 | jermar | 411 | THREAD->priority = -1; |
1 | jermar | 412 | |
125 | jermar | 413 | /* |
414 | * We need to release wq->lock which we locked in waitq_sleep(). |
||
415 | * Address of wq->lock is kept in THREAD->sleep_queue. |
||
416 | */ |
||
417 | spinlock_unlock(&THREAD->sleep_queue->lock); |
||
1 | jermar | 418 | |
125 | jermar | 419 | /* |
420 | * Check for possible requests for out-of-context invocation. |
||
421 | */ |
||
422 | if (THREAD->call_me) { |
||
423 | THREAD->call_me(THREAD->call_me_with); |
||
424 | THREAD->call_me = NULL; |
||
425 | THREAD->call_me_with = NULL; |
||
426 | } |
||
1 | jermar | 427 | |
125 | jermar | 428 | spinlock_unlock(&THREAD->lock); |
1 | jermar | 429 | |
125 | jermar | 430 | break; |
431 | |||
1 | jermar | 432 | default: |
125 | jermar | 433 | /* |
434 | * Entering state is unexpected. |
||
435 | */ |
||
436 | panic("tid%d: unexpected state %s\n", THREAD->tid, thread_states[THREAD->state]); |
||
437 | break; |
||
1 | jermar | 438 | } |
897 | jermar | 439 | |
15 | jermar | 440 | THREAD = NULL; |
1 | jermar | 441 | } |
198 | jermar | 442 | |
15 | jermar | 443 | THREAD = find_best_thread(); |
1 | jermar | 444 | |
15 | jermar | 445 | spinlock_lock(&THREAD->lock); |
413 | jermar | 446 | priority = THREAD->priority; |
15 | jermar | 447 | spinlock_unlock(&THREAD->lock); |
192 | jermar | 448 | |
1 | jermar | 449 | relink_rq(priority); |
450 | |||
451 | /* |
||
452 | * If both the old and the new task are the same, lots of work is avoided. |
||
453 | */ |
||
15 | jermar | 454 | if (TASK != THREAD->task) { |
703 | jermar | 455 | as_t *as1 = NULL; |
456 | as_t *as2; |
||
1 | jermar | 457 | |
15 | jermar | 458 | if (TASK) { |
459 | spinlock_lock(&TASK->lock); |
||
703 | jermar | 460 | as1 = TASK->as; |
15 | jermar | 461 | spinlock_unlock(&TASK->lock); |
1 | jermar | 462 | } |
463 | |||
15 | jermar | 464 | spinlock_lock(&THREAD->task->lock); |
703 | jermar | 465 | as2 = THREAD->task->as; |
15 | jermar | 466 | spinlock_unlock(&THREAD->task->lock); |
1 | jermar | 467 | |
468 | /* |
||
703 | jermar | 469 | * Note that it is possible for two tasks to share one address space. |
1 | jermar | 470 | */ |
703 | jermar | 471 | if (as1 != as2) { |
1 | jermar | 472 | /* |
703 | jermar | 473 | * Both tasks and address spaces are different. |
1 | jermar | 474 | * Replace the old one with the new one. |
475 | */ |
||
823 | jermar | 476 | as_switch(as1, as2); |
1 | jermar | 477 | } |
906 | palkovsky | 478 | TASK = THREAD->task; |
1187 | jermar | 479 | before_task_runs(); |
1 | jermar | 480 | } |
481 | |||
1380 | jermar | 482 | spinlock_lock(&THREAD->lock); |
15 | jermar | 483 | THREAD->state = Running; |
1 | jermar | 484 | |
906 | palkovsky | 485 | #ifdef SCHEDULER_VERBOSE |
1196 | cejka | 486 | printf("cpu%d: tid %d (priority=%d,ticks=%lld,nrdy=%ld)\n", CPU->id, THREAD->tid, THREAD->priority, THREAD->ticks, atomic_get(&CPU->nrdy)); |
906 | palkovsky | 487 | #endif |
1 | jermar | 488 | |
213 | jermar | 489 | /* |
897 | jermar | 490 | * Some architectures provide late kernel PA2KA(identity) |
491 | * mapping in a page fault handler. However, the page fault |
||
492 | * handler uses the kernel stack of the running thread and |
||
493 | * therefore cannot be used to map it. The kernel stack, if |
||
494 | * necessary, is to be mapped in before_thread_runs(). This |
||
495 | * function must be executed before the switch to the new stack. |
||
496 | */ |
||
497 | before_thread_runs(); |
||
498 | |||
499 | /* |
||
213 | jermar | 500 | * Copy the knowledge of CPU, TASK, THREAD and preemption counter to thread's stack. |
501 | */ |
||
184 | jermar | 502 | the_copy(THE, (the_t *) THREAD->kstack); |
503 | |||
15 | jermar | 504 | context_restore(&THREAD->saved_context); |
1 | jermar | 505 | /* not reached */ |
506 | } |
||
507 | |||
458 | decky | 508 | #ifdef CONFIG_SMP |
107 | decky | 509 | /** Load balancing thread |
510 | * |
||
511 | * SMP load balancing thread, supervising thread supplies |
||
512 | * for the CPU it's wired to. |
||
513 | * |
||
514 | * @param arg Generic thread argument (unused). |
||
515 | * |
||
1 | jermar | 516 | */ |
517 | void kcpulb(void *arg) |
||
518 | { |
||
519 | thread_t *t; |
||
783 | palkovsky | 520 | int count, average, i, j, k = 0; |
413 | jermar | 521 | ipl_t ipl; |
1 | jermar | 522 | |
1576 | jermar | 523 | /* |
524 | * Detach kcpulb as nobody will call thread_join_timeout() on it. |
||
525 | */ |
||
526 | thread_detach(THREAD); |
||
527 | |||
1 | jermar | 528 | loop: |
529 | /* |
||
779 | jermar | 530 | * Work in 1s intervals. |
1 | jermar | 531 | */ |
779 | jermar | 532 | thread_sleep(1); |
1 | jermar | 533 | |
534 | not_satisfied: |
||
535 | /* |
||
536 | * Calculate the number of threads that will be migrated/stolen from |
||
537 | * other CPU's. Note that situation can have changed between two |
||
538 | * passes. Each time get the most up to date counts. |
||
539 | */ |
||
784 | palkovsky | 540 | average = atomic_get(&nrdy) / config.cpu_active + 1; |
783 | palkovsky | 541 | count = average - atomic_get(&CPU->nrdy); |
1 | jermar | 542 | |
784 | palkovsky | 543 | if (count <= 0) |
1 | jermar | 544 | goto satisfied; |
545 | |||
546 | /* |
||
547 | * Searching least priority queues on all CPU's first and most priority queues on all CPU's last. |
||
548 | */ |
||
549 | for (j=RQ_COUNT-1; j >= 0; j--) { |
||
550 | for (i=0; i < config.cpu_active; i++) { |
||
551 | link_t *l; |
||
552 | runq_t *r; |
||
553 | cpu_t *cpu; |
||
554 | |||
555 | cpu = &cpus[(i + k) % config.cpu_active]; |
||
556 | |||
557 | /* |
||
558 | * Not interested in ourselves. |
||
559 | * Doesn't require interrupt disabling for kcpulb is X_WIRED. |
||
560 | */ |
||
15 | jermar | 561 | if (CPU == cpu) |
783 | palkovsky | 562 | continue; |
563 | if (atomic_get(&cpu->nrdy) <= average) |
||
564 | continue; |
||
1 | jermar | 565 | |
784 | palkovsky | 566 | ipl = interrupts_disable(); |
115 | jermar | 567 | r = &cpu->rq[j]; |
1 | jermar | 568 | spinlock_lock(&r->lock); |
569 | if (r->n == 0) { |
||
570 | spinlock_unlock(&r->lock); |
||
413 | jermar | 571 | interrupts_restore(ipl); |
1 | jermar | 572 | continue; |
573 | } |
||
574 | |||
575 | t = NULL; |
||
576 | l = r->rq_head.prev; /* search rq from the back */ |
||
577 | while (l != &r->rq_head) { |
||
578 | t = list_get_instance(l, thread_t, rq_link); |
||
579 | /* |
||
125 | jermar | 580 | * We don't want to steal CPU-wired threads neither threads already stolen. |
1 | jermar | 581 | * The latter prevents threads from migrating between CPU's without ever being run. |
125 | jermar | 582 | * We don't want to steal threads whose FPU context is still in CPU. |
73 | vana | 583 | */ |
1 | jermar | 584 | spinlock_lock(&t->lock); |
73 | vana | 585 | if ( (!(t->flags & (X_WIRED | X_STOLEN))) && (!(t->fpu_context_engaged)) ) { |
1 | jermar | 586 | /* |
587 | * Remove t from r. |
||
588 | */ |
||
589 | spinlock_unlock(&t->lock); |
||
590 | |||
783 | palkovsky | 591 | atomic_dec(&cpu->nrdy); |
475 | jermar | 592 | atomic_dec(&nrdy); |
1 | jermar | 593 | |
125 | jermar | 594 | r->n--; |
1 | jermar | 595 | list_remove(&t->rq_link); |
596 | |||
597 | break; |
||
598 | } |
||
599 | spinlock_unlock(&t->lock); |
||
600 | l = l->prev; |
||
601 | t = NULL; |
||
602 | } |
||
603 | spinlock_unlock(&r->lock); |
||
604 | |||
605 | if (t) { |
||
606 | /* |
||
607 | * Ready t on local CPU |
||
608 | */ |
||
609 | spinlock_lock(&t->lock); |
||
906 | palkovsky | 610 | #ifdef KCPULB_VERBOSE |
1196 | cejka | 611 | printf("kcpulb%d: TID %d -> cpu%d, nrdy=%ld, avg=%nd\n", CPU->id, t->tid, CPU->id, atomic_get(&CPU->nrdy), atomic_get(&nrdy) / config.cpu_active); |
906 | palkovsky | 612 | #endif |
1 | jermar | 613 | t->flags |= X_STOLEN; |
1115 | jermar | 614 | t->state = Entering; |
1 | jermar | 615 | spinlock_unlock(&t->lock); |
616 | |||
617 | thread_ready(t); |
||
618 | |||
413 | jermar | 619 | interrupts_restore(ipl); |
1 | jermar | 620 | |
621 | if (--count == 0) |
||
622 | goto satisfied; |
||
623 | |||
624 | /* |
||
125 | jermar | 625 | * We are not satisfied yet, focus on another CPU next time. |
1 | jermar | 626 | */ |
627 | k++; |
||
628 | |||
629 | continue; |
||
630 | } |
||
413 | jermar | 631 | interrupts_restore(ipl); |
1 | jermar | 632 | } |
633 | } |
||
634 | |||
783 | palkovsky | 635 | if (atomic_get(&CPU->nrdy)) { |
1 | jermar | 636 | /* |
637 | * Be a little bit light-weight and let migrated threads run. |
||
638 | */ |
||
639 | scheduler(); |
||
779 | jermar | 640 | } else { |
1 | jermar | 641 | /* |
642 | * We failed to migrate a single thread. |
||
779 | jermar | 643 | * Give up this turn. |
1 | jermar | 644 | */ |
779 | jermar | 645 | goto loop; |
1 | jermar | 646 | } |
647 | |||
648 | goto not_satisfied; |
||
125 | jermar | 649 | |
1 | jermar | 650 | satisfied: |
651 | goto loop; |
||
652 | } |
||
653 | |||
458 | decky | 654 | #endif /* CONFIG_SMP */ |
775 | palkovsky | 655 | |
656 | |||
657 | /** Print information about threads & scheduler queues */ |
||
658 | void sched_print_list(void) |
||
659 | { |
||
660 | ipl_t ipl; |
||
661 | int cpu,i; |
||
662 | runq_t *r; |
||
663 | thread_t *t; |
||
664 | link_t *cur; |
||
665 | |||
666 | /* We are going to mess with scheduler structures, |
||
667 | * let's not be interrupted */ |
||
668 | ipl = interrupts_disable(); |
||
669 | for (cpu=0;cpu < config.cpu_count; cpu++) { |
||
898 | jermar | 670 | |
775 | palkovsky | 671 | if (!cpus[cpu].active) |
672 | continue; |
||
898 | jermar | 673 | |
775 | palkovsky | 674 | spinlock_lock(&cpus[cpu].lock); |
1221 | decky | 675 | printf("cpu%d: address=%p, nrdy=%ld, needs_relink=%ld\n", |
1062 | jermar | 676 | cpus[cpu].id, &cpus[cpu], atomic_get(&cpus[cpu].nrdy), cpus[cpu].needs_relink); |
775 | palkovsky | 677 | |
678 | for (i=0; i<RQ_COUNT; i++) { |
||
679 | r = &cpus[cpu].rq[i]; |
||
680 | spinlock_lock(&r->lock); |
||
681 | if (!r->n) { |
||
682 | spinlock_unlock(&r->lock); |
||
683 | continue; |
||
684 | } |
||
898 | jermar | 685 | printf("\trq[%d]: ", i); |
775 | palkovsky | 686 | for (cur=r->rq_head.next; cur!=&r->rq_head; cur=cur->next) { |
687 | t = list_get_instance(cur, thread_t, rq_link); |
||
688 | printf("%d(%s) ", t->tid, |
||
689 | thread_states[t->state]); |
||
690 | } |
||
691 | printf("\n"); |
||
692 | spinlock_unlock(&r->lock); |
||
693 | } |
||
694 | spinlock_unlock(&cpus[cpu].lock); |
||
695 | } |
||
696 | |||
697 | interrupts_restore(ipl); |
||
698 | } |
||
1702 | cejka | 699 | |
1757 | jermar | 700 | /** @} |
1702 | cejka | 701 | */ |