Rev 2030 | Rev 2068 | 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> |
||
2030 | decky | 50 | #include <arch/cycle.h> |
1104 | jermar | 51 | #include <atomic.h> |
378 | jermar | 52 | #include <synch/spinlock.h> |
1 | jermar | 53 | #include <config.h> |
54 | #include <context.h> |
||
55 | #include <func.h> |
||
56 | #include <arch.h> |
||
788 | jermar | 57 | #include <adt/list.h> |
68 | decky | 58 | #include <panic.h> |
1 | jermar | 59 | #include <typedefs.h> |
378 | jermar | 60 | #include <cpu.h> |
195 | vana | 61 | #include <print.h> |
227 | jermar | 62 | #include <debug.h> |
1 | jermar | 63 | |
1187 | jermar | 64 | static void before_task_runs(void); |
65 | static void before_thread_runs(void); |
||
66 | static void after_thread_ran(void); |
||
898 | jermar | 67 | static void scheduler_separated_stack(void); |
195 | vana | 68 | |
898 | jermar | 69 | atomic_t nrdy; /**< Number of ready threads in the system. */ |
70 | |||
1187 | jermar | 71 | /** Carry out actions before new task runs. */ |
72 | void before_task_runs(void) |
||
73 | { |
||
74 | before_task_runs_arch(); |
||
75 | } |
||
76 | |||
897 | jermar | 77 | /** Take actions before new thread runs. |
107 | decky | 78 | * |
118 | jermar | 79 | * Perform actions that need to be |
80 | * taken before the newly selected |
||
81 | * tread is passed control. |
||
107 | decky | 82 | * |
827 | palkovsky | 83 | * THREAD->lock is locked on entry |
84 | * |
||
107 | decky | 85 | */ |
52 | vana | 86 | void before_thread_runs(void) |
87 | { |
||
309 | palkovsky | 88 | before_thread_runs_arch(); |
906 | palkovsky | 89 | #ifdef CONFIG_FPU_LAZY |
1882 | jermar | 90 | if(THREAD == CPU->fpu_owner) |
309 | palkovsky | 91 | fpu_enable(); |
92 | else |
||
93 | fpu_disable(); |
||
906 | palkovsky | 94 | #else |
309 | palkovsky | 95 | fpu_enable(); |
96 | if (THREAD->fpu_context_exists) |
||
906 | palkovsky | 97 | fpu_context_restore(THREAD->saved_fpu_context); |
309 | palkovsky | 98 | else { |
906 | palkovsky | 99 | fpu_init(); |
1882 | jermar | 100 | THREAD->fpu_context_exists = 1; |
309 | palkovsky | 101 | } |
906 | palkovsky | 102 | #endif |
52 | vana | 103 | } |
104 | |||
898 | jermar | 105 | /** Take actions after THREAD had run. |
897 | jermar | 106 | * |
107 | * Perform actions that need to be |
||
108 | * taken after the running thread |
||
898 | jermar | 109 | * had been preempted by the scheduler. |
897 | jermar | 110 | * |
111 | * THREAD->lock is locked on entry |
||
112 | * |
||
113 | */ |
||
114 | void after_thread_ran(void) |
||
115 | { |
||
116 | after_thread_ran_arch(); |
||
117 | } |
||
118 | |||
458 | decky | 119 | #ifdef CONFIG_FPU_LAZY |
309 | palkovsky | 120 | void scheduler_fpu_lazy_request(void) |
121 | { |
||
907 | palkovsky | 122 | restart: |
309 | palkovsky | 123 | fpu_enable(); |
827 | palkovsky | 124 | spinlock_lock(&CPU->lock); |
125 | |||
126 | /* Save old context */ |
||
309 | palkovsky | 127 | if (CPU->fpu_owner != NULL) { |
827 | palkovsky | 128 | spinlock_lock(&CPU->fpu_owner->lock); |
906 | palkovsky | 129 | fpu_context_save(CPU->fpu_owner->saved_fpu_context); |
309 | palkovsky | 130 | /* don't prevent migration */ |
1882 | jermar | 131 | CPU->fpu_owner->fpu_context_engaged = 0; |
827 | palkovsky | 132 | spinlock_unlock(&CPU->fpu_owner->lock); |
907 | palkovsky | 133 | CPU->fpu_owner = NULL; |
309 | palkovsky | 134 | } |
827 | palkovsky | 135 | |
136 | spinlock_lock(&THREAD->lock); |
||
898 | jermar | 137 | if (THREAD->fpu_context_exists) { |
906 | palkovsky | 138 | fpu_context_restore(THREAD->saved_fpu_context); |
898 | jermar | 139 | } else { |
906 | palkovsky | 140 | /* Allocate FPU context */ |
141 | if (!THREAD->saved_fpu_context) { |
||
142 | /* Might sleep */ |
||
143 | spinlock_unlock(&THREAD->lock); |
||
907 | palkovsky | 144 | spinlock_unlock(&CPU->lock); |
1854 | jermar | 145 | THREAD->saved_fpu_context = slab_alloc(fpu_context_slab, 0); |
907 | palkovsky | 146 | /* We may have switched CPUs during slab_alloc */ |
147 | goto restart; |
||
906 | palkovsky | 148 | } |
149 | fpu_init(); |
||
1882 | jermar | 150 | THREAD->fpu_context_exists = 1; |
309 | palkovsky | 151 | } |
1882 | jermar | 152 | CPU->fpu_owner = THREAD; |
309 | palkovsky | 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 | /* |
||
1854 | jermar | 238 | * Clear the THREAD_FLAG_STOLEN flag so that t can be migrated |
239 | * when load balancing needs emerge. |
||
1 | jermar | 240 | */ |
1854 | jermar | 241 | t->flags &= ~THREAD_FLAG_STOLEN; |
1 | jermar | 242 | spinlock_unlock(&t->lock); |
243 | |||
244 | return t; |
||
245 | } |
||
246 | goto loop; |
||
247 | |||
248 | } |
||
249 | |||
107 | decky | 250 | /** Prevent rq starvation |
251 | * |
||
252 | * Prevent low priority threads from starving in rq's. |
||
253 | * |
||
254 | * When the function decides to relink rq's, it reconnects |
||
255 | * respective pointers so that in result threads with 'pri' |
||
1708 | jermar | 256 | * greater or equal start are moved to a higher-priority queue. |
107 | decky | 257 | * |
258 | * @param start Threshold priority. |
||
259 | * |
||
1 | jermar | 260 | */ |
452 | decky | 261 | static void relink_rq(int start) |
1 | jermar | 262 | { |
263 | link_t head; |
||
264 | runq_t *r; |
||
265 | int i, n; |
||
266 | |||
267 | list_initialize(&head); |
||
15 | jermar | 268 | spinlock_lock(&CPU->lock); |
269 | if (CPU->needs_relink > NEEDS_RELINK_MAX) { |
||
1 | jermar | 270 | for (i = start; i<RQ_COUNT-1; i++) { |
271 | /* remember and empty rq[i + 1] */ |
||
15 | jermar | 272 | r = &CPU->rq[i + 1]; |
1 | jermar | 273 | spinlock_lock(&r->lock); |
274 | list_concat(&head, &r->rq_head); |
||
275 | n = r->n; |
||
276 | r->n = 0; |
||
277 | spinlock_unlock(&r->lock); |
||
278 | |||
279 | /* append rq[i + 1] to rq[i] */ |
||
15 | jermar | 280 | r = &CPU->rq[i]; |
1 | jermar | 281 | spinlock_lock(&r->lock); |
282 | list_concat(&r->rq_head, &head); |
||
283 | r->n += n; |
||
284 | spinlock_unlock(&r->lock); |
||
285 | } |
||
15 | jermar | 286 | CPU->needs_relink = 0; |
1 | jermar | 287 | } |
784 | palkovsky | 288 | spinlock_unlock(&CPU->lock); |
1 | jermar | 289 | |
290 | } |
||
291 | |||
898 | jermar | 292 | /** The scheduler |
293 | * |
||
294 | * The thread scheduling procedure. |
||
295 | * Passes control directly to |
||
296 | * scheduler_separated_stack(). |
||
297 | * |
||
298 | */ |
||
299 | void scheduler(void) |
||
300 | { |
||
301 | volatile ipl_t ipl; |
||
107 | decky | 302 | |
898 | jermar | 303 | ASSERT(CPU != NULL); |
304 | |||
305 | ipl = interrupts_disable(); |
||
306 | |||
307 | if (atomic_get(&haltstate)) |
||
308 | halt(); |
||
1007 | decky | 309 | |
898 | jermar | 310 | if (THREAD) { |
311 | spinlock_lock(&THREAD->lock); |
||
2030 | decky | 312 | |
313 | /* Update thread accounting */ |
||
314 | THREAD->cycles += get_cycle() - THREAD->last_cycle; |
||
315 | |||
906 | palkovsky | 316 | #ifndef CONFIG_FPU_LAZY |
317 | fpu_context_save(THREAD->saved_fpu_context); |
||
318 | #endif |
||
898 | jermar | 319 | if (!context_save(&THREAD->saved_context)) { |
320 | /* |
||
321 | * This is the place where threads leave scheduler(); |
||
322 | */ |
||
2030 | decky | 323 | |
324 | /* Save current CPU cycle */ |
||
325 | THREAD->last_cycle = get_cycle(); |
||
326 | |||
898 | jermar | 327 | spinlock_unlock(&THREAD->lock); |
328 | interrupts_restore(THREAD->saved_context.ipl); |
||
1007 | decky | 329 | |
898 | jermar | 330 | return; |
331 | } |
||
332 | |||
333 | /* |
||
334 | * Interrupt priority level of preempted thread is recorded here |
||
335 | * to facilitate scheduler() invocations from interrupts_disable()'d |
||
336 | * code (e.g. waitq_sleep_timeout()). |
||
337 | */ |
||
338 | THREAD->saved_context.ipl = ipl; |
||
339 | } |
||
340 | |||
341 | /* |
||
342 | * Through the 'THE' structure, we keep track of THREAD, TASK, CPU, VM |
||
343 | * and preemption counter. At this point THE could be coming either |
||
344 | * from THREAD's or CPU's stack. |
||
345 | */ |
||
346 | the_copy(THE, (the_t *) CPU->stack); |
||
347 | |||
348 | /* |
||
349 | * We may not keep the old stack. |
||
350 | * Reason: If we kept the old stack and got blocked, for instance, in |
||
351 | * find_best_thread(), the old thread could get rescheduled by another |
||
352 | * CPU and overwrite the part of its own stack that was also used by |
||
353 | * the scheduler on this CPU. |
||
354 | * |
||
355 | * Moreover, we have to bypass the compiler-generated POP sequence |
||
356 | * which is fooled by SP being set to the very top of the stack. |
||
357 | * Therefore the scheduler() function continues in |
||
358 | * scheduler_separated_stack(). |
||
359 | */ |
||
360 | context_save(&CPU->saved_context); |
||
1854 | jermar | 361 | context_set(&CPU->saved_context, FADDR(scheduler_separated_stack), |
362 | (uintptr_t) CPU->stack, CPU_STACK_SIZE); |
||
898 | jermar | 363 | context_restore(&CPU->saved_context); |
364 | /* not reached */ |
||
365 | } |
||
366 | |||
107 | decky | 367 | /** Scheduler stack switch wrapper |
368 | * |
||
369 | * Second part of the scheduler() function |
||
370 | * using new stack. Handling the actual context |
||
371 | * switch to a new thread. |
||
372 | * |
||
787 | palkovsky | 373 | * Assume THREAD->lock is held. |
107 | decky | 374 | */ |
898 | jermar | 375 | void scheduler_separated_stack(void) |
1 | jermar | 376 | { |
377 | int priority; |
||
1007 | decky | 378 | |
227 | jermar | 379 | ASSERT(CPU != NULL); |
1007 | decky | 380 | |
15 | jermar | 381 | if (THREAD) { |
898 | jermar | 382 | /* must be run after the switch to scheduler stack */ |
897 | jermar | 383 | after_thread_ran(); |
384 | |||
15 | jermar | 385 | switch (THREAD->state) { |
1888 | jermar | 386 | case Running: |
125 | jermar | 387 | spinlock_unlock(&THREAD->lock); |
388 | thread_ready(THREAD); |
||
389 | break; |
||
1 | jermar | 390 | |
1888 | jermar | 391 | case Exiting: |
1571 | jermar | 392 | repeat: |
2040 | decky | 393 | if (THREAD->detached) { |
1571 | jermar | 394 | thread_destroy(THREAD); |
395 | } else { |
||
396 | /* |
||
397 | * The thread structure is kept allocated until somebody |
||
398 | * calls thread_detach() on it. |
||
399 | */ |
||
400 | if (!spinlock_trylock(&THREAD->join_wq.lock)) { |
||
401 | /* |
||
402 | * Avoid deadlock. |
||
403 | */ |
||
404 | spinlock_unlock(&THREAD->lock); |
||
405 | delay(10); |
||
406 | spinlock_lock(&THREAD->lock); |
||
407 | goto repeat; |
||
408 | } |
||
409 | _waitq_wakeup_unsafe(&THREAD->join_wq, false); |
||
410 | spinlock_unlock(&THREAD->join_wq.lock); |
||
411 | |||
412 | THREAD->state = Undead; |
||
413 | spinlock_unlock(&THREAD->lock); |
||
414 | } |
||
125 | jermar | 415 | break; |
787 | palkovsky | 416 | |
1888 | jermar | 417 | case Sleeping: |
125 | jermar | 418 | /* |
419 | * Prefer the thread after it's woken up. |
||
420 | */ |
||
413 | jermar | 421 | THREAD->priority = -1; |
1 | jermar | 422 | |
125 | jermar | 423 | /* |
424 | * We need to release wq->lock which we locked in waitq_sleep(). |
||
425 | * Address of wq->lock is kept in THREAD->sleep_queue. |
||
426 | */ |
||
427 | spinlock_unlock(&THREAD->sleep_queue->lock); |
||
1 | jermar | 428 | |
125 | jermar | 429 | /* |
430 | * Check for possible requests for out-of-context invocation. |
||
431 | */ |
||
432 | if (THREAD->call_me) { |
||
433 | THREAD->call_me(THREAD->call_me_with); |
||
434 | THREAD->call_me = NULL; |
||
435 | THREAD->call_me_with = NULL; |
||
436 | } |
||
1 | jermar | 437 | |
125 | jermar | 438 | spinlock_unlock(&THREAD->lock); |
1 | jermar | 439 | |
125 | jermar | 440 | break; |
441 | |||
1888 | jermar | 442 | default: |
125 | jermar | 443 | /* |
444 | * Entering state is unexpected. |
||
445 | */ |
||
446 | panic("tid%d: unexpected state %s\n", THREAD->tid, thread_states[THREAD->state]); |
||
447 | break; |
||
1 | jermar | 448 | } |
897 | jermar | 449 | |
15 | jermar | 450 | THREAD = NULL; |
1 | jermar | 451 | } |
198 | jermar | 452 | |
15 | jermar | 453 | THREAD = find_best_thread(); |
1 | jermar | 454 | |
15 | jermar | 455 | spinlock_lock(&THREAD->lock); |
413 | jermar | 456 | priority = THREAD->priority; |
15 | jermar | 457 | spinlock_unlock(&THREAD->lock); |
192 | jermar | 458 | |
1 | jermar | 459 | relink_rq(priority); |
460 | |||
461 | /* |
||
462 | * If both the old and the new task are the same, lots of work is avoided. |
||
463 | */ |
||
15 | jermar | 464 | if (TASK != THREAD->task) { |
703 | jermar | 465 | as_t *as1 = NULL; |
466 | as_t *as2; |
||
1 | jermar | 467 | |
15 | jermar | 468 | if (TASK) { |
469 | spinlock_lock(&TASK->lock); |
||
703 | jermar | 470 | as1 = TASK->as; |
15 | jermar | 471 | spinlock_unlock(&TASK->lock); |
1 | jermar | 472 | } |
473 | |||
15 | jermar | 474 | spinlock_lock(&THREAD->task->lock); |
703 | jermar | 475 | as2 = THREAD->task->as; |
15 | jermar | 476 | spinlock_unlock(&THREAD->task->lock); |
1 | jermar | 477 | |
478 | /* |
||
703 | jermar | 479 | * Note that it is possible for two tasks to share one address space. |
1 | jermar | 480 | */ |
703 | jermar | 481 | if (as1 != as2) { |
1 | jermar | 482 | /* |
703 | jermar | 483 | * Both tasks and address spaces are different. |
1 | jermar | 484 | * Replace the old one with the new one. |
485 | */ |
||
823 | jermar | 486 | as_switch(as1, as2); |
1 | jermar | 487 | } |
906 | palkovsky | 488 | TASK = THREAD->task; |
1187 | jermar | 489 | before_task_runs(); |
1 | jermar | 490 | } |
491 | |||
1380 | jermar | 492 | spinlock_lock(&THREAD->lock); |
15 | jermar | 493 | THREAD->state = Running; |
1 | jermar | 494 | |
906 | palkovsky | 495 | #ifdef SCHEDULER_VERBOSE |
1854 | jermar | 496 | printf("cpu%d: tid %d (priority=%d,ticks=%lld,nrdy=%ld)\n", |
497 | CPU->id, THREAD->tid, THREAD->priority, THREAD->ticks, atomic_get(&CPU->nrdy)); |
||
906 | palkovsky | 498 | #endif |
1 | jermar | 499 | |
213 | jermar | 500 | /* |
897 | jermar | 501 | * Some architectures provide late kernel PA2KA(identity) |
502 | * mapping in a page fault handler. However, the page fault |
||
503 | * handler uses the kernel stack of the running thread and |
||
504 | * therefore cannot be used to map it. The kernel stack, if |
||
505 | * necessary, is to be mapped in before_thread_runs(). This |
||
506 | * function must be executed before the switch to the new stack. |
||
507 | */ |
||
508 | before_thread_runs(); |
||
509 | |||
510 | /* |
||
213 | jermar | 511 | * Copy the knowledge of CPU, TASK, THREAD and preemption counter to thread's stack. |
512 | */ |
||
184 | jermar | 513 | the_copy(THE, (the_t *) THREAD->kstack); |
514 | |||
15 | jermar | 515 | context_restore(&THREAD->saved_context); |
1 | jermar | 516 | /* not reached */ |
517 | } |
||
518 | |||
458 | decky | 519 | #ifdef CONFIG_SMP |
107 | decky | 520 | /** Load balancing thread |
521 | * |
||
522 | * SMP load balancing thread, supervising thread supplies |
||
523 | * for the CPU it's wired to. |
||
524 | * |
||
525 | * @param arg Generic thread argument (unused). |
||
526 | * |
||
1 | jermar | 527 | */ |
528 | void kcpulb(void *arg) |
||
529 | { |
||
530 | thread_t *t; |
||
783 | palkovsky | 531 | int count, average, i, j, k = 0; |
413 | jermar | 532 | ipl_t ipl; |
1 | jermar | 533 | |
1576 | jermar | 534 | /* |
535 | * Detach kcpulb as nobody will call thread_join_timeout() on it. |
||
536 | */ |
||
537 | thread_detach(THREAD); |
||
538 | |||
1 | jermar | 539 | loop: |
540 | /* |
||
779 | jermar | 541 | * Work in 1s intervals. |
1 | jermar | 542 | */ |
779 | jermar | 543 | thread_sleep(1); |
1 | jermar | 544 | |
545 | not_satisfied: |
||
546 | /* |
||
547 | * Calculate the number of threads that will be migrated/stolen from |
||
548 | * other CPU's. Note that situation can have changed between two |
||
549 | * passes. Each time get the most up to date counts. |
||
550 | */ |
||
784 | palkovsky | 551 | average = atomic_get(&nrdy) / config.cpu_active + 1; |
783 | palkovsky | 552 | count = average - atomic_get(&CPU->nrdy); |
1 | jermar | 553 | |
784 | palkovsky | 554 | if (count <= 0) |
1 | jermar | 555 | goto satisfied; |
556 | |||
557 | /* |
||
558 | * Searching least priority queues on all CPU's first and most priority queues on all CPU's last. |
||
559 | */ |
||
560 | for (j=RQ_COUNT-1; j >= 0; j--) { |
||
561 | for (i=0; i < config.cpu_active; i++) { |
||
562 | link_t *l; |
||
563 | runq_t *r; |
||
564 | cpu_t *cpu; |
||
565 | |||
566 | cpu = &cpus[(i + k) % config.cpu_active]; |
||
567 | |||
568 | /* |
||
569 | * Not interested in ourselves. |
||
1854 | jermar | 570 | * Doesn't require interrupt disabling for kcpulb has THREAD_FLAG_WIRED. |
1 | jermar | 571 | */ |
15 | jermar | 572 | if (CPU == cpu) |
783 | palkovsky | 573 | continue; |
574 | if (atomic_get(&cpu->nrdy) <= average) |
||
575 | continue; |
||
1 | jermar | 576 | |
784 | palkovsky | 577 | ipl = interrupts_disable(); |
115 | jermar | 578 | r = &cpu->rq[j]; |
1 | jermar | 579 | spinlock_lock(&r->lock); |
580 | if (r->n == 0) { |
||
581 | spinlock_unlock(&r->lock); |
||
413 | jermar | 582 | interrupts_restore(ipl); |
1 | jermar | 583 | continue; |
584 | } |
||
585 | |||
586 | t = NULL; |
||
587 | l = r->rq_head.prev; /* search rq from the back */ |
||
588 | while (l != &r->rq_head) { |
||
589 | t = list_get_instance(l, thread_t, rq_link); |
||
590 | /* |
||
1854 | jermar | 591 | * We don't want to steal CPU-wired threads neither threads already |
592 | * stolen. The latter prevents threads from migrating between CPU's |
||
593 | * without ever being run. We don't want to steal threads whose FPU |
||
594 | * context is still in CPU. |
||
73 | vana | 595 | */ |
1 | jermar | 596 | spinlock_lock(&t->lock); |
1854 | jermar | 597 | if ((!(t->flags & (THREAD_FLAG_WIRED | THREAD_FLAG_STOLEN))) && |
598 | (!(t->fpu_context_engaged)) ) { |
||
1 | jermar | 599 | /* |
600 | * Remove t from r. |
||
601 | */ |
||
602 | spinlock_unlock(&t->lock); |
||
603 | |||
783 | palkovsky | 604 | atomic_dec(&cpu->nrdy); |
475 | jermar | 605 | atomic_dec(&nrdy); |
1 | jermar | 606 | |
125 | jermar | 607 | r->n--; |
1 | jermar | 608 | list_remove(&t->rq_link); |
609 | |||
610 | break; |
||
611 | } |
||
612 | spinlock_unlock(&t->lock); |
||
613 | l = l->prev; |
||
614 | t = NULL; |
||
615 | } |
||
616 | spinlock_unlock(&r->lock); |
||
617 | |||
618 | if (t) { |
||
619 | /* |
||
620 | * Ready t on local CPU |
||
621 | */ |
||
622 | spinlock_lock(&t->lock); |
||
906 | palkovsky | 623 | #ifdef KCPULB_VERBOSE |
1854 | jermar | 624 | printf("kcpulb%d: TID %d -> cpu%d, nrdy=%ld, avg=%nd\n", |
625 | CPU->id, t->tid, CPU->id, atomic_get(&CPU->nrdy), |
||
626 | atomic_get(&nrdy) / config.cpu_active); |
||
906 | palkovsky | 627 | #endif |
1854 | jermar | 628 | t->flags |= THREAD_FLAG_STOLEN; |
1115 | jermar | 629 | t->state = Entering; |
1 | jermar | 630 | spinlock_unlock(&t->lock); |
631 | |||
632 | thread_ready(t); |
||
633 | |||
413 | jermar | 634 | interrupts_restore(ipl); |
1 | jermar | 635 | |
636 | if (--count == 0) |
||
637 | goto satisfied; |
||
638 | |||
639 | /* |
||
125 | jermar | 640 | * We are not satisfied yet, focus on another CPU next time. |
1 | jermar | 641 | */ |
642 | k++; |
||
643 | |||
644 | continue; |
||
645 | } |
||
413 | jermar | 646 | interrupts_restore(ipl); |
1 | jermar | 647 | } |
648 | } |
||
649 | |||
783 | palkovsky | 650 | if (atomic_get(&CPU->nrdy)) { |
1 | jermar | 651 | /* |
652 | * Be a little bit light-weight and let migrated threads run. |
||
653 | */ |
||
654 | scheduler(); |
||
779 | jermar | 655 | } else { |
1 | jermar | 656 | /* |
657 | * We failed to migrate a single thread. |
||
779 | jermar | 658 | * Give up this turn. |
1 | jermar | 659 | */ |
779 | jermar | 660 | goto loop; |
1 | jermar | 661 | } |
662 | |||
663 | goto not_satisfied; |
||
125 | jermar | 664 | |
1 | jermar | 665 | satisfied: |
666 | goto loop; |
||
667 | } |
||
668 | |||
458 | decky | 669 | #endif /* CONFIG_SMP */ |
775 | palkovsky | 670 | |
671 | |||
672 | /** Print information about threads & scheduler queues */ |
||
673 | void sched_print_list(void) |
||
674 | { |
||
675 | ipl_t ipl; |
||
676 | int cpu,i; |
||
677 | runq_t *r; |
||
678 | thread_t *t; |
||
679 | link_t *cur; |
||
680 | |||
681 | /* We are going to mess with scheduler structures, |
||
682 | * let's not be interrupted */ |
||
683 | ipl = interrupts_disable(); |
||
684 | for (cpu=0;cpu < config.cpu_count; cpu++) { |
||
898 | jermar | 685 | |
775 | palkovsky | 686 | if (!cpus[cpu].active) |
687 | continue; |
||
898 | jermar | 688 | |
775 | palkovsky | 689 | spinlock_lock(&cpus[cpu].lock); |
1221 | decky | 690 | printf("cpu%d: address=%p, nrdy=%ld, needs_relink=%ld\n", |
1062 | jermar | 691 | cpus[cpu].id, &cpus[cpu], atomic_get(&cpus[cpu].nrdy), cpus[cpu].needs_relink); |
775 | palkovsky | 692 | |
693 | for (i=0; i<RQ_COUNT; i++) { |
||
694 | r = &cpus[cpu].rq[i]; |
||
695 | spinlock_lock(&r->lock); |
||
696 | if (!r->n) { |
||
697 | spinlock_unlock(&r->lock); |
||
698 | continue; |
||
699 | } |
||
898 | jermar | 700 | printf("\trq[%d]: ", i); |
775 | palkovsky | 701 | for (cur=r->rq_head.next; cur!=&r->rq_head; cur=cur->next) { |
702 | t = list_get_instance(cur, thread_t, rq_link); |
||
703 | printf("%d(%s) ", t->tid, |
||
704 | thread_states[t->state]); |
||
705 | } |
||
706 | printf("\n"); |
||
707 | spinlock_unlock(&r->lock); |
||
708 | } |
||
709 | spinlock_unlock(&cpus[cpu].lock); |
||
710 | } |
||
711 | |||
712 | interrupts_restore(ipl); |
||
713 | } |
||
1702 | cejka | 714 | |
1757 | jermar | 715 | /** @} |
1702 | cejka | 716 | */ |