Rev 1248 | Rev 1380 | 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 | |||
1248 | jermar | 29 | /** |
30 | * @file scheduler.c |
||
31 | * @brief Scheduler and load balancing. |
||
32 | * |
||
1264 | jermar | 33 | * This file contains the scheduler and kcpulb kernel thread which |
1248 | jermar | 34 | * performs load-balancing of per-CPU run queues. |
35 | */ |
||
36 | |||
1 | jermar | 37 | #include <proc/scheduler.h> |
38 | #include <proc/thread.h> |
||
39 | #include <proc/task.h> |
||
378 | jermar | 40 | #include <mm/frame.h> |
41 | #include <mm/page.h> |
||
703 | jermar | 42 | #include <mm/as.h> |
378 | jermar | 43 | #include <arch/asm.h> |
44 | #include <arch/faddr.h> |
||
1104 | jermar | 45 | #include <atomic.h> |
378 | jermar | 46 | #include <synch/spinlock.h> |
1 | jermar | 47 | #include <config.h> |
48 | #include <context.h> |
||
49 | #include <func.h> |
||
50 | #include <arch.h> |
||
788 | jermar | 51 | #include <adt/list.h> |
68 | decky | 52 | #include <panic.h> |
1 | jermar | 53 | #include <typedefs.h> |
378 | jermar | 54 | #include <cpu.h> |
195 | vana | 55 | #include <print.h> |
227 | jermar | 56 | #include <debug.h> |
1 | jermar | 57 | |
1187 | jermar | 58 | static void before_task_runs(void); |
59 | static void before_thread_runs(void); |
||
60 | static void after_thread_ran(void); |
||
898 | jermar | 61 | static void scheduler_separated_stack(void); |
195 | vana | 62 | |
898 | jermar | 63 | atomic_t nrdy; /**< Number of ready threads in the system. */ |
64 | |||
1187 | jermar | 65 | /** Carry out actions before new task runs. */ |
66 | void before_task_runs(void) |
||
67 | { |
||
68 | before_task_runs_arch(); |
||
69 | } |
||
70 | |||
897 | jermar | 71 | /** Take actions before new thread runs. |
107 | decky | 72 | * |
118 | jermar | 73 | * Perform actions that need to be |
74 | * taken before the newly selected |
||
75 | * tread is passed control. |
||
107 | decky | 76 | * |
827 | palkovsky | 77 | * THREAD->lock is locked on entry |
78 | * |
||
107 | decky | 79 | */ |
52 | vana | 80 | void before_thread_runs(void) |
81 | { |
||
309 | palkovsky | 82 | before_thread_runs_arch(); |
906 | palkovsky | 83 | #ifdef CONFIG_FPU_LAZY |
309 | palkovsky | 84 | if(THREAD==CPU->fpu_owner) |
85 | fpu_enable(); |
||
86 | else |
||
87 | fpu_disable(); |
||
906 | palkovsky | 88 | #else |
309 | palkovsky | 89 | fpu_enable(); |
90 | if (THREAD->fpu_context_exists) |
||
906 | palkovsky | 91 | fpu_context_restore(THREAD->saved_fpu_context); |
309 | palkovsky | 92 | else { |
906 | palkovsky | 93 | fpu_init(); |
309 | palkovsky | 94 | THREAD->fpu_context_exists=1; |
95 | } |
||
906 | palkovsky | 96 | #endif |
52 | vana | 97 | } |
98 | |||
898 | jermar | 99 | /** Take actions after THREAD had run. |
897 | jermar | 100 | * |
101 | * Perform actions that need to be |
||
102 | * taken after the running thread |
||
898 | jermar | 103 | * had been preempted by the scheduler. |
897 | jermar | 104 | * |
105 | * THREAD->lock is locked on entry |
||
106 | * |
||
107 | */ |
||
108 | void after_thread_ran(void) |
||
109 | { |
||
110 | after_thread_ran_arch(); |
||
111 | } |
||
112 | |||
458 | decky | 113 | #ifdef CONFIG_FPU_LAZY |
309 | palkovsky | 114 | void scheduler_fpu_lazy_request(void) |
115 | { |
||
907 | palkovsky | 116 | restart: |
309 | palkovsky | 117 | fpu_enable(); |
827 | palkovsky | 118 | spinlock_lock(&CPU->lock); |
119 | |||
120 | /* Save old context */ |
||
309 | palkovsky | 121 | if (CPU->fpu_owner != NULL) { |
827 | palkovsky | 122 | spinlock_lock(&CPU->fpu_owner->lock); |
906 | palkovsky | 123 | fpu_context_save(CPU->fpu_owner->saved_fpu_context); |
309 | palkovsky | 124 | /* don't prevent migration */ |
125 | CPU->fpu_owner->fpu_context_engaged=0; |
||
827 | palkovsky | 126 | spinlock_unlock(&CPU->fpu_owner->lock); |
907 | palkovsky | 127 | CPU->fpu_owner = NULL; |
309 | palkovsky | 128 | } |
827 | palkovsky | 129 | |
130 | spinlock_lock(&THREAD->lock); |
||
898 | jermar | 131 | if (THREAD->fpu_context_exists) { |
906 | palkovsky | 132 | fpu_context_restore(THREAD->saved_fpu_context); |
898 | jermar | 133 | } else { |
906 | palkovsky | 134 | /* Allocate FPU context */ |
135 | if (!THREAD->saved_fpu_context) { |
||
136 | /* Might sleep */ |
||
137 | spinlock_unlock(&THREAD->lock); |
||
907 | palkovsky | 138 | spinlock_unlock(&CPU->lock); |
906 | palkovsky | 139 | THREAD->saved_fpu_context = slab_alloc(fpu_context_slab, |
140 | 0); |
||
907 | palkovsky | 141 | /* We may have switched CPUs during slab_alloc */ |
142 | goto restart; |
||
906 | palkovsky | 143 | } |
144 | fpu_init(); |
||
309 | palkovsky | 145 | THREAD->fpu_context_exists=1; |
146 | } |
||
147 | CPU->fpu_owner=THREAD; |
||
148 | THREAD->fpu_context_engaged = 1; |
||
898 | jermar | 149 | spinlock_unlock(&THREAD->lock); |
827 | palkovsky | 150 | |
151 | spinlock_unlock(&CPU->lock); |
||
309 | palkovsky | 152 | } |
153 | #endif |
||
52 | vana | 154 | |
107 | decky | 155 | /** Initialize scheduler |
156 | * |
||
157 | * Initialize kernel scheduler. |
||
158 | * |
||
159 | */ |
||
1 | jermar | 160 | void scheduler_init(void) |
161 | { |
||
162 | } |
||
163 | |||
107 | decky | 164 | /** Get thread to be scheduled |
165 | * |
||
166 | * Get the optimal thread to be scheduled |
||
109 | jermar | 167 | * according to thread accounting and scheduler |
107 | decky | 168 | * policy. |
169 | * |
||
170 | * @return Thread to be scheduled. |
||
171 | * |
||
172 | */ |
||
483 | jermar | 173 | static thread_t *find_best_thread(void) |
1 | jermar | 174 | { |
175 | thread_t *t; |
||
176 | runq_t *r; |
||
783 | palkovsky | 177 | int i; |
1 | jermar | 178 | |
227 | jermar | 179 | ASSERT(CPU != NULL); |
180 | |||
1 | jermar | 181 | loop: |
413 | jermar | 182 | interrupts_enable(); |
1 | jermar | 183 | |
783 | palkovsky | 184 | if (atomic_get(&CPU->nrdy) == 0) { |
1 | jermar | 185 | /* |
186 | * For there was nothing to run, the CPU goes to sleep |
||
187 | * until a hardware interrupt or an IPI comes. |
||
188 | * This improves energy saving and hyperthreading. |
||
189 | */ |
||
785 | jermar | 190 | |
191 | /* |
||
192 | * An interrupt might occur right now and wake up a thread. |
||
193 | * In such case, the CPU will continue to go to sleep |
||
194 | * even though there is a runnable thread. |
||
195 | */ |
||
196 | |||
1 | jermar | 197 | cpu_sleep(); |
198 | goto loop; |
||
199 | } |
||
200 | |||
413 | jermar | 201 | interrupts_disable(); |
114 | jermar | 202 | |
898 | jermar | 203 | for (i = 0; i<RQ_COUNT; i++) { |
15 | jermar | 204 | r = &CPU->rq[i]; |
1 | jermar | 205 | spinlock_lock(&r->lock); |
206 | if (r->n == 0) { |
||
207 | /* |
||
208 | * If this queue is empty, try a lower-priority queue. |
||
209 | */ |
||
210 | spinlock_unlock(&r->lock); |
||
211 | continue; |
||
212 | } |
||
213 | jermar | 213 | |
783 | palkovsky | 214 | atomic_dec(&CPU->nrdy); |
475 | jermar | 215 | atomic_dec(&nrdy); |
1 | jermar | 216 | r->n--; |
217 | |||
218 | /* |
||
219 | * Take the first thread from the queue. |
||
220 | */ |
||
221 | t = list_get_instance(r->rq_head.next, thread_t, rq_link); |
||
222 | list_remove(&t->rq_link); |
||
223 | |||
224 | spinlock_unlock(&r->lock); |
||
225 | |||
226 | spinlock_lock(&t->lock); |
||
15 | jermar | 227 | t->cpu = CPU; |
1 | jermar | 228 | |
229 | t->ticks = us2ticks((i+1)*10000); |
||
898 | jermar | 230 | t->priority = i; /* correct rq index */ |
1 | jermar | 231 | |
232 | /* |
||
233 | * Clear the X_STOLEN flag so that t can be migrated when load balancing needs emerge. |
||
234 | */ |
||
235 | t->flags &= ~X_STOLEN; |
||
236 | spinlock_unlock(&t->lock); |
||
237 | |||
238 | return t; |
||
239 | } |
||
240 | goto loop; |
||
241 | |||
242 | } |
||
243 | |||
107 | decky | 244 | /** Prevent rq starvation |
245 | * |
||
246 | * Prevent low priority threads from starving in rq's. |
||
247 | * |
||
248 | * When the function decides to relink rq's, it reconnects |
||
249 | * respective pointers so that in result threads with 'pri' |
||
1229 | jermar | 250 | * greater or equal @start are moved to a higher-priority queue. |
107 | decky | 251 | * |
252 | * @param start Threshold priority. |
||
253 | * |
||
1 | jermar | 254 | */ |
452 | decky | 255 | static void relink_rq(int start) |
1 | jermar | 256 | { |
257 | link_t head; |
||
258 | runq_t *r; |
||
259 | int i, n; |
||
260 | |||
261 | list_initialize(&head); |
||
15 | jermar | 262 | spinlock_lock(&CPU->lock); |
263 | if (CPU->needs_relink > NEEDS_RELINK_MAX) { |
||
1 | jermar | 264 | for (i = start; i<RQ_COUNT-1; i++) { |
265 | /* remember and empty rq[i + 1] */ |
||
15 | jermar | 266 | r = &CPU->rq[i + 1]; |
1 | jermar | 267 | spinlock_lock(&r->lock); |
268 | list_concat(&head, &r->rq_head); |
||
269 | n = r->n; |
||
270 | r->n = 0; |
||
271 | spinlock_unlock(&r->lock); |
||
272 | |||
273 | /* append rq[i + 1] to rq[i] */ |
||
15 | jermar | 274 | r = &CPU->rq[i]; |
1 | jermar | 275 | spinlock_lock(&r->lock); |
276 | list_concat(&r->rq_head, &head); |
||
277 | r->n += n; |
||
278 | spinlock_unlock(&r->lock); |
||
279 | } |
||
15 | jermar | 280 | CPU->needs_relink = 0; |
1 | jermar | 281 | } |
784 | palkovsky | 282 | spinlock_unlock(&CPU->lock); |
1 | jermar | 283 | |
284 | } |
||
285 | |||
898 | jermar | 286 | /** The scheduler |
287 | * |
||
288 | * The thread scheduling procedure. |
||
289 | * Passes control directly to |
||
290 | * scheduler_separated_stack(). |
||
291 | * |
||
292 | */ |
||
293 | void scheduler(void) |
||
294 | { |
||
295 | volatile ipl_t ipl; |
||
107 | decky | 296 | |
898 | jermar | 297 | ASSERT(CPU != NULL); |
298 | |||
299 | ipl = interrupts_disable(); |
||
300 | |||
301 | if (atomic_get(&haltstate)) |
||
302 | halt(); |
||
1007 | decky | 303 | |
898 | jermar | 304 | if (THREAD) { |
305 | spinlock_lock(&THREAD->lock); |
||
906 | palkovsky | 306 | #ifndef CONFIG_FPU_LAZY |
307 | fpu_context_save(THREAD->saved_fpu_context); |
||
308 | #endif |
||
898 | jermar | 309 | if (!context_save(&THREAD->saved_context)) { |
310 | /* |
||
311 | * This is the place where threads leave scheduler(); |
||
312 | */ |
||
313 | spinlock_unlock(&THREAD->lock); |
||
314 | interrupts_restore(THREAD->saved_context.ipl); |
||
1007 | decky | 315 | |
898 | jermar | 316 | return; |
317 | } |
||
318 | |||
319 | /* |
||
320 | * Interrupt priority level of preempted thread is recorded here |
||
321 | * to facilitate scheduler() invocations from interrupts_disable()'d |
||
322 | * code (e.g. waitq_sleep_timeout()). |
||
323 | */ |
||
324 | THREAD->saved_context.ipl = ipl; |
||
325 | } |
||
326 | |||
327 | /* |
||
328 | * Through the 'THE' structure, we keep track of THREAD, TASK, CPU, VM |
||
329 | * and preemption counter. At this point THE could be coming either |
||
330 | * from THREAD's or CPU's stack. |
||
331 | */ |
||
332 | the_copy(THE, (the_t *) CPU->stack); |
||
333 | |||
334 | /* |
||
335 | * We may not keep the old stack. |
||
336 | * Reason: If we kept the old stack and got blocked, for instance, in |
||
337 | * find_best_thread(), the old thread could get rescheduled by another |
||
338 | * CPU and overwrite the part of its own stack that was also used by |
||
339 | * the scheduler on this CPU. |
||
340 | * |
||
341 | * Moreover, we have to bypass the compiler-generated POP sequence |
||
342 | * which is fooled by SP being set to the very top of the stack. |
||
343 | * Therefore the scheduler() function continues in |
||
344 | * scheduler_separated_stack(). |
||
345 | */ |
||
346 | context_save(&CPU->saved_context); |
||
347 | context_set(&CPU->saved_context, FADDR(scheduler_separated_stack), (__address) CPU->stack, CPU_STACK_SIZE); |
||
348 | context_restore(&CPU->saved_context); |
||
349 | /* not reached */ |
||
350 | } |
||
351 | |||
107 | decky | 352 | /** Scheduler stack switch wrapper |
353 | * |
||
354 | * Second part of the scheduler() function |
||
355 | * using new stack. Handling the actual context |
||
356 | * switch to a new thread. |
||
357 | * |
||
787 | palkovsky | 358 | * Assume THREAD->lock is held. |
107 | decky | 359 | */ |
898 | jermar | 360 | void scheduler_separated_stack(void) |
1 | jermar | 361 | { |
362 | int priority; |
||
1007 | decky | 363 | |
227 | jermar | 364 | ASSERT(CPU != NULL); |
1007 | decky | 365 | |
15 | jermar | 366 | if (THREAD) { |
898 | jermar | 367 | /* must be run after the switch to scheduler stack */ |
897 | jermar | 368 | after_thread_ran(); |
369 | |||
15 | jermar | 370 | switch (THREAD->state) { |
1 | jermar | 371 | case Running: |
125 | jermar | 372 | spinlock_unlock(&THREAD->lock); |
373 | thread_ready(THREAD); |
||
374 | break; |
||
1 | jermar | 375 | |
376 | case Exiting: |
||
787 | palkovsky | 377 | thread_destroy(THREAD); |
125 | jermar | 378 | break; |
787 | palkovsky | 379 | |
1 | jermar | 380 | case Sleeping: |
125 | jermar | 381 | /* |
382 | * Prefer the thread after it's woken up. |
||
383 | */ |
||
413 | jermar | 384 | THREAD->priority = -1; |
1 | jermar | 385 | |
125 | jermar | 386 | /* |
387 | * We need to release wq->lock which we locked in waitq_sleep(). |
||
388 | * Address of wq->lock is kept in THREAD->sleep_queue. |
||
389 | */ |
||
390 | spinlock_unlock(&THREAD->sleep_queue->lock); |
||
1 | jermar | 391 | |
125 | jermar | 392 | /* |
393 | * Check for possible requests for out-of-context invocation. |
||
394 | */ |
||
395 | if (THREAD->call_me) { |
||
396 | THREAD->call_me(THREAD->call_me_with); |
||
397 | THREAD->call_me = NULL; |
||
398 | THREAD->call_me_with = NULL; |
||
399 | } |
||
1 | jermar | 400 | |
125 | jermar | 401 | spinlock_unlock(&THREAD->lock); |
1 | jermar | 402 | |
125 | jermar | 403 | break; |
404 | |||
1 | jermar | 405 | default: |
125 | jermar | 406 | /* |
407 | * Entering state is unexpected. |
||
408 | */ |
||
409 | panic("tid%d: unexpected state %s\n", THREAD->tid, thread_states[THREAD->state]); |
||
410 | break; |
||
1 | jermar | 411 | } |
897 | jermar | 412 | |
15 | jermar | 413 | THREAD = NULL; |
1 | jermar | 414 | } |
198 | jermar | 415 | |
15 | jermar | 416 | THREAD = find_best_thread(); |
1 | jermar | 417 | |
15 | jermar | 418 | spinlock_lock(&THREAD->lock); |
413 | jermar | 419 | priority = THREAD->priority; |
15 | jermar | 420 | spinlock_unlock(&THREAD->lock); |
192 | jermar | 421 | |
1 | jermar | 422 | relink_rq(priority); |
423 | |||
15 | jermar | 424 | spinlock_lock(&THREAD->lock); |
1 | jermar | 425 | |
426 | /* |
||
427 | * If both the old and the new task are the same, lots of work is avoided. |
||
428 | */ |
||
15 | jermar | 429 | if (TASK != THREAD->task) { |
703 | jermar | 430 | as_t *as1 = NULL; |
431 | as_t *as2; |
||
1 | jermar | 432 | |
15 | jermar | 433 | if (TASK) { |
434 | spinlock_lock(&TASK->lock); |
||
703 | jermar | 435 | as1 = TASK->as; |
15 | jermar | 436 | spinlock_unlock(&TASK->lock); |
1 | jermar | 437 | } |
438 | |||
15 | jermar | 439 | spinlock_lock(&THREAD->task->lock); |
703 | jermar | 440 | as2 = THREAD->task->as; |
15 | jermar | 441 | spinlock_unlock(&THREAD->task->lock); |
1 | jermar | 442 | |
443 | /* |
||
703 | jermar | 444 | * Note that it is possible for two tasks to share one address space. |
1 | jermar | 445 | */ |
703 | jermar | 446 | if (as1 != as2) { |
1 | jermar | 447 | /* |
703 | jermar | 448 | * Both tasks and address spaces are different. |
1 | jermar | 449 | * Replace the old one with the new one. |
450 | */ |
||
823 | jermar | 451 | as_switch(as1, as2); |
1 | jermar | 452 | } |
906 | palkovsky | 453 | TASK = THREAD->task; |
1187 | jermar | 454 | before_task_runs(); |
1 | jermar | 455 | } |
456 | |||
15 | jermar | 457 | THREAD->state = Running; |
1 | jermar | 458 | |
906 | palkovsky | 459 | #ifdef SCHEDULER_VERBOSE |
1196 | cejka | 460 | 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 | 461 | #endif |
1 | jermar | 462 | |
213 | jermar | 463 | /* |
897 | jermar | 464 | * Some architectures provide late kernel PA2KA(identity) |
465 | * mapping in a page fault handler. However, the page fault |
||
466 | * handler uses the kernel stack of the running thread and |
||
467 | * therefore cannot be used to map it. The kernel stack, if |
||
468 | * necessary, is to be mapped in before_thread_runs(). This |
||
469 | * function must be executed before the switch to the new stack. |
||
470 | */ |
||
471 | before_thread_runs(); |
||
472 | |||
473 | /* |
||
213 | jermar | 474 | * Copy the knowledge of CPU, TASK, THREAD and preemption counter to thread's stack. |
475 | */ |
||
184 | jermar | 476 | the_copy(THE, (the_t *) THREAD->kstack); |
477 | |||
15 | jermar | 478 | context_restore(&THREAD->saved_context); |
1 | jermar | 479 | /* not reached */ |
480 | } |
||
481 | |||
458 | decky | 482 | #ifdef CONFIG_SMP |
107 | decky | 483 | /** Load balancing thread |
484 | * |
||
485 | * SMP load balancing thread, supervising thread supplies |
||
486 | * for the CPU it's wired to. |
||
487 | * |
||
488 | * @param arg Generic thread argument (unused). |
||
489 | * |
||
1 | jermar | 490 | */ |
491 | void kcpulb(void *arg) |
||
492 | { |
||
493 | thread_t *t; |
||
783 | palkovsky | 494 | int count, average, i, j, k = 0; |
413 | jermar | 495 | ipl_t ipl; |
1 | jermar | 496 | |
497 | loop: |
||
498 | /* |
||
779 | jermar | 499 | * Work in 1s intervals. |
1 | jermar | 500 | */ |
779 | jermar | 501 | thread_sleep(1); |
1 | jermar | 502 | |
503 | not_satisfied: |
||
504 | /* |
||
505 | * Calculate the number of threads that will be migrated/stolen from |
||
506 | * other CPU's. Note that situation can have changed between two |
||
507 | * passes. Each time get the most up to date counts. |
||
508 | */ |
||
784 | palkovsky | 509 | average = atomic_get(&nrdy) / config.cpu_active + 1; |
783 | palkovsky | 510 | count = average - atomic_get(&CPU->nrdy); |
1 | jermar | 511 | |
784 | palkovsky | 512 | if (count <= 0) |
1 | jermar | 513 | goto satisfied; |
514 | |||
515 | /* |
||
516 | * Searching least priority queues on all CPU's first and most priority queues on all CPU's last. |
||
517 | */ |
||
518 | for (j=RQ_COUNT-1; j >= 0; j--) { |
||
519 | for (i=0; i < config.cpu_active; i++) { |
||
520 | link_t *l; |
||
521 | runq_t *r; |
||
522 | cpu_t *cpu; |
||
523 | |||
524 | cpu = &cpus[(i + k) % config.cpu_active]; |
||
525 | |||
526 | /* |
||
527 | * Not interested in ourselves. |
||
528 | * Doesn't require interrupt disabling for kcpulb is X_WIRED. |
||
529 | */ |
||
15 | jermar | 530 | if (CPU == cpu) |
783 | palkovsky | 531 | continue; |
532 | if (atomic_get(&cpu->nrdy) <= average) |
||
533 | continue; |
||
1 | jermar | 534 | |
784 | palkovsky | 535 | ipl = interrupts_disable(); |
115 | jermar | 536 | r = &cpu->rq[j]; |
1 | jermar | 537 | spinlock_lock(&r->lock); |
538 | if (r->n == 0) { |
||
539 | spinlock_unlock(&r->lock); |
||
413 | jermar | 540 | interrupts_restore(ipl); |
1 | jermar | 541 | continue; |
542 | } |
||
543 | |||
544 | t = NULL; |
||
545 | l = r->rq_head.prev; /* search rq from the back */ |
||
546 | while (l != &r->rq_head) { |
||
547 | t = list_get_instance(l, thread_t, rq_link); |
||
548 | /* |
||
125 | jermar | 549 | * We don't want to steal CPU-wired threads neither threads already stolen. |
1 | jermar | 550 | * The latter prevents threads from migrating between CPU's without ever being run. |
125 | jermar | 551 | * We don't want to steal threads whose FPU context is still in CPU. |
73 | vana | 552 | */ |
1 | jermar | 553 | spinlock_lock(&t->lock); |
73 | vana | 554 | if ( (!(t->flags & (X_WIRED | X_STOLEN))) && (!(t->fpu_context_engaged)) ) { |
1 | jermar | 555 | /* |
556 | * Remove t from r. |
||
557 | */ |
||
558 | spinlock_unlock(&t->lock); |
||
559 | |||
783 | palkovsky | 560 | atomic_dec(&cpu->nrdy); |
475 | jermar | 561 | atomic_dec(&nrdy); |
1 | jermar | 562 | |
125 | jermar | 563 | r->n--; |
1 | jermar | 564 | list_remove(&t->rq_link); |
565 | |||
566 | break; |
||
567 | } |
||
568 | spinlock_unlock(&t->lock); |
||
569 | l = l->prev; |
||
570 | t = NULL; |
||
571 | } |
||
572 | spinlock_unlock(&r->lock); |
||
573 | |||
574 | if (t) { |
||
575 | /* |
||
576 | * Ready t on local CPU |
||
577 | */ |
||
578 | spinlock_lock(&t->lock); |
||
906 | palkovsky | 579 | #ifdef KCPULB_VERBOSE |
1196 | cejka | 580 | 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 | 581 | #endif |
1 | jermar | 582 | t->flags |= X_STOLEN; |
1115 | jermar | 583 | t->state = Entering; |
1 | jermar | 584 | spinlock_unlock(&t->lock); |
585 | |||
586 | thread_ready(t); |
||
587 | |||
413 | jermar | 588 | interrupts_restore(ipl); |
1 | jermar | 589 | |
590 | if (--count == 0) |
||
591 | goto satisfied; |
||
592 | |||
593 | /* |
||
125 | jermar | 594 | * We are not satisfied yet, focus on another CPU next time. |
1 | jermar | 595 | */ |
596 | k++; |
||
597 | |||
598 | continue; |
||
599 | } |
||
413 | jermar | 600 | interrupts_restore(ipl); |
1 | jermar | 601 | } |
602 | } |
||
603 | |||
783 | palkovsky | 604 | if (atomic_get(&CPU->nrdy)) { |
1 | jermar | 605 | /* |
606 | * Be a little bit light-weight and let migrated threads run. |
||
607 | */ |
||
608 | scheduler(); |
||
779 | jermar | 609 | } else { |
1 | jermar | 610 | /* |
611 | * We failed to migrate a single thread. |
||
779 | jermar | 612 | * Give up this turn. |
1 | jermar | 613 | */ |
779 | jermar | 614 | goto loop; |
1 | jermar | 615 | } |
616 | |||
617 | goto not_satisfied; |
||
125 | jermar | 618 | |
1 | jermar | 619 | satisfied: |
620 | goto loop; |
||
621 | } |
||
622 | |||
458 | decky | 623 | #endif /* CONFIG_SMP */ |
775 | palkovsky | 624 | |
625 | |||
626 | /** Print information about threads & scheduler queues */ |
||
627 | void sched_print_list(void) |
||
628 | { |
||
629 | ipl_t ipl; |
||
630 | int cpu,i; |
||
631 | runq_t *r; |
||
632 | thread_t *t; |
||
633 | link_t *cur; |
||
634 | |||
635 | /* We are going to mess with scheduler structures, |
||
636 | * let's not be interrupted */ |
||
637 | ipl = interrupts_disable(); |
||
638 | for (cpu=0;cpu < config.cpu_count; cpu++) { |
||
898 | jermar | 639 | |
775 | palkovsky | 640 | if (!cpus[cpu].active) |
641 | continue; |
||
898 | jermar | 642 | |
775 | palkovsky | 643 | spinlock_lock(&cpus[cpu].lock); |
1221 | decky | 644 | printf("cpu%d: address=%p, nrdy=%ld, needs_relink=%ld\n", |
1062 | jermar | 645 | cpus[cpu].id, &cpus[cpu], atomic_get(&cpus[cpu].nrdy), cpus[cpu].needs_relink); |
775 | palkovsky | 646 | |
647 | for (i=0; i<RQ_COUNT; i++) { |
||
648 | r = &cpus[cpu].rq[i]; |
||
649 | spinlock_lock(&r->lock); |
||
650 | if (!r->n) { |
||
651 | spinlock_unlock(&r->lock); |
||
652 | continue; |
||
653 | } |
||
898 | jermar | 654 | printf("\trq[%d]: ", i); |
775 | palkovsky | 655 | for (cur=r->rq_head.next; cur!=&r->rq_head; cur=cur->next) { |
656 | t = list_get_instance(cur, thread_t, rq_link); |
||
657 | printf("%d(%s) ", t->tid, |
||
658 | thread_states[t->state]); |
||
659 | } |
||
660 | printf("\n"); |
||
661 | spinlock_unlock(&r->lock); |
||
662 | } |
||
663 | spinlock_unlock(&cpus[cpu].lock); |
||
664 | } |
||
665 | |||
666 | interrupts_restore(ipl); |
||
667 | } |