| 1 | /* $NetBSD: kern_runq.c,v 1.45 2016/07/07 06:55:43 msaitoh Exp $ */ |
| 2 | |
| 3 | /* |
| 4 | * Copyright (c) 2007, 2008 Mindaugas Rasiukevicius <rmind at NetBSD org> |
| 5 | * All rights reserved. |
| 6 | * |
| 7 | * Redistribution and use in source and binary forms, with or without |
| 8 | * modification, are permitted provided that the following conditions |
| 9 | * are met: |
| 10 | * 1. Redistributions of source code must retain the above copyright |
| 11 | * notice, this list of conditions and the following disclaimer. |
| 12 | * 2. Redistributions in binary form must reproduce the above copyright |
| 13 | * notice, this list of conditions and the following disclaimer in the |
| 14 | * documentation and/or other materials provided with the distribution. |
| 15 | * |
| 16 | * THIS SOFTWARE IS PROVIDED BY THE AUTHOR AND CONTRIBUTORS ``AS IS'' AND |
| 17 | * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE |
| 18 | * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE |
| 19 | * ARE DISCLAIMED. IN NO EVENT SHALL THE AUTHOR OR CONTRIBUTORS BE LIABLE |
| 20 | * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL |
| 21 | * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS |
| 22 | * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) |
| 23 | * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT |
| 24 | * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY |
| 25 | * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF |
| 26 | * SUCH DAMAGE. |
| 27 | */ |
| 28 | |
| 29 | #include <sys/cdefs.h> |
| 30 | __KERNEL_RCSID(0, "$NetBSD: kern_runq.c,v 1.45 2016/07/07 06:55:43 msaitoh Exp $" ); |
| 31 | |
| 32 | #include "opt_dtrace.h" |
| 33 | |
| 34 | #include <sys/param.h> |
| 35 | #include <sys/kernel.h> |
| 36 | #include <sys/bitops.h> |
| 37 | #include <sys/cpu.h> |
| 38 | #include <sys/idle.h> |
| 39 | #include <sys/intr.h> |
| 40 | #include <sys/kmem.h> |
| 41 | #include <sys/lwp.h> |
| 42 | #include <sys/mutex.h> |
| 43 | #include <sys/proc.h> |
| 44 | #include <sys/sched.h> |
| 45 | #include <sys/syscallargs.h> |
| 46 | #include <sys/sysctl.h> |
| 47 | #include <sys/systm.h> |
| 48 | #include <sys/types.h> |
| 49 | #include <sys/evcnt.h> |
| 50 | |
| 51 | /* |
| 52 | * Priority related definitions. |
| 53 | */ |
| 54 | #define PRI_TS_COUNT (NPRI_USER) |
| 55 | #define PRI_RT_COUNT (PRI_COUNT - PRI_TS_COUNT) |
| 56 | #define PRI_HTS_RANGE (PRI_TS_COUNT / 10) |
| 57 | |
| 58 | #define PRI_HIGHEST_TS (MAXPRI_USER) |
| 59 | |
| 60 | /* |
| 61 | * Bits per map. |
| 62 | */ |
| 63 | #define BITMAP_BITS (32) |
| 64 | #define BITMAP_SHIFT (5) |
| 65 | #define BITMAP_MSB (0x80000000U) |
| 66 | #define BITMAP_MASK (BITMAP_BITS - 1) |
| 67 | |
| 68 | /* |
| 69 | * Structures, runqueue. |
| 70 | */ |
| 71 | |
| 72 | const int schedppq = 1; |
| 73 | |
| 74 | typedef struct { |
| 75 | TAILQ_HEAD(, lwp) q_head; |
| 76 | } queue_t; |
| 77 | |
| 78 | typedef struct { |
| 79 | /* Bitmap */ |
| 80 | uint32_t r_bitmap[PRI_COUNT >> BITMAP_SHIFT]; |
| 81 | /* Counters */ |
| 82 | u_int r_count; /* Count of the threads */ |
| 83 | u_int r_avgcount; /* Average count of threads */ |
| 84 | u_int r_mcount; /* Count of migratable threads */ |
| 85 | /* Runqueues */ |
| 86 | queue_t r_rt_queue[PRI_RT_COUNT]; |
| 87 | queue_t r_ts_queue[PRI_TS_COUNT]; |
| 88 | /* Event counters */ |
| 89 | struct evcnt r_ev_pull; |
| 90 | struct evcnt r_ev_push; |
| 91 | struct evcnt r_ev_stay; |
| 92 | struct evcnt r_ev_localize; |
| 93 | } runqueue_t; |
| 94 | |
| 95 | static void * sched_getrq(runqueue_t *, const pri_t); |
| 96 | #ifdef MULTIPROCESSOR |
| 97 | static lwp_t * sched_catchlwp(struct cpu_info *); |
| 98 | static void sched_balance(void *); |
| 99 | #endif |
| 100 | |
| 101 | /* |
| 102 | * Preemption control. |
| 103 | */ |
| 104 | int sched_upreempt_pri = 0; |
| 105 | #ifdef __HAVE_PREEMPTION |
| 106 | # ifdef DEBUG |
| 107 | int sched_kpreempt_pri = 0; |
| 108 | # else |
| 109 | int sched_kpreempt_pri = PRI_USER_RT; |
| 110 | # endif |
| 111 | #else |
| 112 | int sched_kpreempt_pri = 1000; |
| 113 | #endif |
| 114 | |
| 115 | /* |
| 116 | * Migration and balancing. |
| 117 | */ |
| 118 | static u_int cacheht_time; /* Cache hotness time */ |
| 119 | static u_int min_catch; /* Minimal LWP count for catching */ |
| 120 | static u_int balance_period; /* Balance period */ |
| 121 | static struct cpu_info *worker_ci; /* Victim CPU */ |
| 122 | #ifdef MULTIPROCESSOR |
| 123 | static struct callout balance_ch; /* Callout of balancer */ |
| 124 | #endif |
| 125 | |
| 126 | #ifdef KDTRACE_HOOKS |
| 127 | struct lwp *curthread; |
| 128 | #endif |
| 129 | |
| 130 | void |
| 131 | runq_init(void) |
| 132 | { |
| 133 | |
| 134 | /* Balancing */ |
| 135 | worker_ci = curcpu(); |
| 136 | cacheht_time = mstohz(3); /* ~3 ms */ |
| 137 | balance_period = mstohz(300); /* ~300 ms */ |
| 138 | |
| 139 | /* Minimal count of LWPs for catching */ |
| 140 | min_catch = 1; |
| 141 | |
| 142 | /* Initialize balancing callout and run it */ |
| 143 | #ifdef MULTIPROCESSOR |
| 144 | callout_init(&balance_ch, CALLOUT_MPSAFE); |
| 145 | callout_setfunc(&balance_ch, sched_balance, NULL); |
| 146 | callout_schedule(&balance_ch, balance_period); |
| 147 | #endif |
| 148 | } |
| 149 | |
| 150 | void |
| 151 | sched_cpuattach(struct cpu_info *ci) |
| 152 | { |
| 153 | runqueue_t *ci_rq; |
| 154 | void *rq_ptr; |
| 155 | u_int i, size; |
| 156 | |
| 157 | if (ci->ci_schedstate.spc_lwplock == NULL) { |
| 158 | ci->ci_schedstate.spc_lwplock = |
| 159 | mutex_obj_alloc(MUTEX_DEFAULT, IPL_SCHED); |
| 160 | } |
| 161 | if (ci == lwp0.l_cpu) { |
| 162 | /* Initialize the scheduler structure of the primary LWP */ |
| 163 | lwp0.l_mutex = ci->ci_schedstate.spc_lwplock; |
| 164 | } |
| 165 | if (ci->ci_schedstate.spc_mutex != NULL) { |
| 166 | /* Already initialized. */ |
| 167 | return; |
| 168 | } |
| 169 | |
| 170 | /* Allocate the run queue */ |
| 171 | size = roundup2(sizeof(runqueue_t), coherency_unit) + coherency_unit; |
| 172 | rq_ptr = kmem_zalloc(size, KM_SLEEP); |
| 173 | if (rq_ptr == NULL) { |
| 174 | panic("sched_cpuattach: could not allocate the runqueue" ); |
| 175 | } |
| 176 | ci_rq = (void *)(roundup2((uintptr_t)(rq_ptr), coherency_unit)); |
| 177 | |
| 178 | /* Initialize run queues */ |
| 179 | ci->ci_schedstate.spc_mutex = |
| 180 | mutex_obj_alloc(MUTEX_DEFAULT, IPL_SCHED); |
| 181 | for (i = 0; i < PRI_RT_COUNT; i++) |
| 182 | TAILQ_INIT(&ci_rq->r_rt_queue[i].q_head); |
| 183 | for (i = 0; i < PRI_TS_COUNT; i++) |
| 184 | TAILQ_INIT(&ci_rq->r_ts_queue[i].q_head); |
| 185 | |
| 186 | ci->ci_schedstate.spc_sched_info = ci_rq; |
| 187 | |
| 188 | evcnt_attach_dynamic(&ci_rq->r_ev_pull, EVCNT_TYPE_MISC, NULL, |
| 189 | cpu_name(ci), "runqueue pull" ); |
| 190 | evcnt_attach_dynamic(&ci_rq->r_ev_push, EVCNT_TYPE_MISC, NULL, |
| 191 | cpu_name(ci), "runqueue push" ); |
| 192 | evcnt_attach_dynamic(&ci_rq->r_ev_stay, EVCNT_TYPE_MISC, NULL, |
| 193 | cpu_name(ci), "runqueue stay" ); |
| 194 | evcnt_attach_dynamic(&ci_rq->r_ev_localize, EVCNT_TYPE_MISC, NULL, |
| 195 | cpu_name(ci), "runqueue localize" ); |
| 196 | } |
| 197 | |
| 198 | /* |
| 199 | * Control of the runqueue. |
| 200 | */ |
| 201 | |
| 202 | static inline void * |
| 203 | sched_getrq(runqueue_t *ci_rq, const pri_t prio) |
| 204 | { |
| 205 | |
| 206 | KASSERT(prio < PRI_COUNT); |
| 207 | return (prio <= PRI_HIGHEST_TS) ? |
| 208 | &ci_rq->r_ts_queue[prio].q_head : |
| 209 | &ci_rq->r_rt_queue[prio - PRI_HIGHEST_TS - 1].q_head; |
| 210 | } |
| 211 | |
| 212 | void |
| 213 | sched_enqueue(struct lwp *l, bool swtch) |
| 214 | { |
| 215 | runqueue_t *ci_rq; |
| 216 | struct schedstate_percpu *spc; |
| 217 | TAILQ_HEAD(, lwp) *q_head; |
| 218 | const pri_t eprio = lwp_eprio(l); |
| 219 | struct cpu_info *ci; |
| 220 | int type; |
| 221 | |
| 222 | ci = l->l_cpu; |
| 223 | spc = &ci->ci_schedstate; |
| 224 | ci_rq = spc->spc_sched_info; |
| 225 | KASSERT(lwp_locked(l, l->l_cpu->ci_schedstate.spc_mutex)); |
| 226 | |
| 227 | /* Update the last run time on switch */ |
| 228 | if (__predict_true(swtch == true)) |
| 229 | l->l_rticksum += (hardclock_ticks - l->l_rticks); |
| 230 | else if (l->l_rticks == 0) |
| 231 | l->l_rticks = hardclock_ticks; |
| 232 | |
| 233 | /* Enqueue the thread */ |
| 234 | q_head = sched_getrq(ci_rq, eprio); |
| 235 | if (TAILQ_EMPTY(q_head)) { |
| 236 | u_int i; |
| 237 | uint32_t q; |
| 238 | |
| 239 | /* Mark bit */ |
| 240 | i = eprio >> BITMAP_SHIFT; |
| 241 | q = BITMAP_MSB >> (eprio & BITMAP_MASK); |
| 242 | KASSERT((ci_rq->r_bitmap[i] & q) == 0); |
| 243 | ci_rq->r_bitmap[i] |= q; |
| 244 | } |
| 245 | TAILQ_INSERT_TAIL(q_head, l, l_runq); |
| 246 | ci_rq->r_count++; |
| 247 | if ((l->l_pflag & LP_BOUND) == 0) |
| 248 | ci_rq->r_mcount++; |
| 249 | |
| 250 | /* |
| 251 | * Update the value of highest priority in the runqueue, |
| 252 | * if priority of this thread is higher. |
| 253 | */ |
| 254 | if (eprio > spc->spc_maxpriority) |
| 255 | spc->spc_maxpriority = eprio; |
| 256 | |
| 257 | sched_newts(l); |
| 258 | |
| 259 | /* |
| 260 | * Wake the chosen CPU or cause a preemption if the newly |
| 261 | * enqueued thread has higher priority. Don't cause a |
| 262 | * preemption if the thread is yielding (swtch). |
| 263 | */ |
| 264 | if (!swtch && eprio > spc->spc_curpriority) { |
| 265 | if (eprio >= sched_kpreempt_pri) |
| 266 | type = RESCHED_KPREEMPT; |
| 267 | else if (eprio >= sched_upreempt_pri) |
| 268 | type = RESCHED_IMMED; |
| 269 | else |
| 270 | type = RESCHED_LAZY; |
| 271 | cpu_need_resched(ci, type); |
| 272 | } |
| 273 | } |
| 274 | |
| 275 | void |
| 276 | sched_dequeue(struct lwp *l) |
| 277 | { |
| 278 | runqueue_t *ci_rq; |
| 279 | TAILQ_HEAD(, lwp) *q_head; |
| 280 | struct schedstate_percpu *spc; |
| 281 | const pri_t eprio = lwp_eprio(l); |
| 282 | |
| 283 | spc = & l->l_cpu->ci_schedstate; |
| 284 | ci_rq = spc->spc_sched_info; |
| 285 | KASSERT(lwp_locked(l, spc->spc_mutex)); |
| 286 | |
| 287 | KASSERT(eprio <= spc->spc_maxpriority); |
| 288 | KASSERT(ci_rq->r_bitmap[eprio >> BITMAP_SHIFT] != 0); |
| 289 | KASSERT(ci_rq->r_count > 0); |
| 290 | |
| 291 | if (spc->spc_migrating == l) |
| 292 | spc->spc_migrating = NULL; |
| 293 | |
| 294 | ci_rq->r_count--; |
| 295 | if ((l->l_pflag & LP_BOUND) == 0) |
| 296 | ci_rq->r_mcount--; |
| 297 | |
| 298 | q_head = sched_getrq(ci_rq, eprio); |
| 299 | TAILQ_REMOVE(q_head, l, l_runq); |
| 300 | if (TAILQ_EMPTY(q_head)) { |
| 301 | u_int i; |
| 302 | uint32_t q; |
| 303 | |
| 304 | /* Unmark bit */ |
| 305 | i = eprio >> BITMAP_SHIFT; |
| 306 | q = BITMAP_MSB >> (eprio & BITMAP_MASK); |
| 307 | KASSERT((ci_rq->r_bitmap[i] & q) != 0); |
| 308 | ci_rq->r_bitmap[i] &= ~q; |
| 309 | |
| 310 | /* |
| 311 | * Update the value of highest priority in the runqueue, in a |
| 312 | * case it was a last thread in the queue of highest priority. |
| 313 | */ |
| 314 | if (eprio != spc->spc_maxpriority) |
| 315 | return; |
| 316 | |
| 317 | do { |
| 318 | if (ci_rq->r_bitmap[i] != 0) { |
| 319 | q = ffs(ci_rq->r_bitmap[i]); |
| 320 | spc->spc_maxpriority = |
| 321 | (i << BITMAP_SHIFT) + (BITMAP_BITS - q); |
| 322 | return; |
| 323 | } |
| 324 | } while (i--); |
| 325 | |
| 326 | /* If not found - set the lowest value */ |
| 327 | spc->spc_maxpriority = 0; |
| 328 | } |
| 329 | } |
| 330 | |
| 331 | /* |
| 332 | * Migration and balancing. |
| 333 | */ |
| 334 | |
| 335 | #ifdef MULTIPROCESSOR |
| 336 | |
| 337 | /* Estimate if LWP is cache-hot */ |
| 338 | static inline bool |
| 339 | lwp_cache_hot(const struct lwp *l) |
| 340 | { |
| 341 | |
| 342 | if (__predict_false(l->l_slptime || l->l_rticks == 0)) |
| 343 | return false; |
| 344 | |
| 345 | return (hardclock_ticks - l->l_rticks <= cacheht_time); |
| 346 | } |
| 347 | |
| 348 | /* Check if LWP can migrate to the chosen CPU */ |
| 349 | static inline bool |
| 350 | sched_migratable(const struct lwp *l, struct cpu_info *ci) |
| 351 | { |
| 352 | const struct schedstate_percpu *spc = &ci->ci_schedstate; |
| 353 | KASSERT(lwp_locked(__UNCONST(l), NULL)); |
| 354 | |
| 355 | /* Is CPU offline? */ |
| 356 | if (__predict_false(spc->spc_flags & SPCF_OFFLINE)) |
| 357 | return false; |
| 358 | |
| 359 | /* Is affinity set? */ |
| 360 | if (__predict_false(l->l_affinity)) |
| 361 | return kcpuset_isset(l->l_affinity, cpu_index(ci)); |
| 362 | |
| 363 | /* Is there a processor-set? */ |
| 364 | return (spc->spc_psid == l->l_psid); |
| 365 | } |
| 366 | |
| 367 | /* |
| 368 | * Estimate the migration of LWP to the other CPU. |
| 369 | * Take and return the CPU, if migration is needed. |
| 370 | */ |
| 371 | struct cpu_info * |
| 372 | sched_takecpu(struct lwp *l) |
| 373 | { |
| 374 | struct cpu_info *ci, *tci, *pivot, *next; |
| 375 | struct schedstate_percpu *spc; |
| 376 | runqueue_t *ci_rq, *ici_rq; |
| 377 | pri_t eprio, lpri, pri; |
| 378 | |
| 379 | KASSERT(lwp_locked(l, NULL)); |
| 380 | |
| 381 | /* If thread is strictly bound, do not estimate other CPUs */ |
| 382 | ci = l->l_cpu; |
| 383 | if (l->l_pflag & LP_BOUND) |
| 384 | return ci; |
| 385 | |
| 386 | spc = &ci->ci_schedstate; |
| 387 | ci_rq = spc->spc_sched_info; |
| 388 | |
| 389 | /* Make sure that thread is in appropriate processor-set */ |
| 390 | if (__predict_true(spc->spc_psid == l->l_psid)) { |
| 391 | /* If CPU of this thread is idling - run there */ |
| 392 | if (ci_rq->r_count == 0) { |
| 393 | ci_rq->r_ev_stay.ev_count++; |
| 394 | return ci; |
| 395 | } |
| 396 | /* Stay if thread is cache-hot */ |
| 397 | eprio = lwp_eprio(l); |
| 398 | if (__predict_true(l->l_stat != LSIDL) && |
| 399 | lwp_cache_hot(l) && eprio >= spc->spc_curpriority) { |
| 400 | ci_rq->r_ev_stay.ev_count++; |
| 401 | return ci; |
| 402 | } |
| 403 | } else { |
| 404 | eprio = lwp_eprio(l); |
| 405 | } |
| 406 | |
| 407 | /* Run on current CPU if priority of thread is higher */ |
| 408 | ci = curcpu(); |
| 409 | spc = &ci->ci_schedstate; |
| 410 | if (eprio > spc->spc_curpriority && sched_migratable(l, ci)) { |
| 411 | ci_rq = spc->spc_sched_info; |
| 412 | ci_rq->r_ev_localize.ev_count++; |
| 413 | return ci; |
| 414 | } |
| 415 | |
| 416 | /* |
| 417 | * Look for the CPU with the lowest priority thread. In case of |
| 418 | * equal priority, choose the CPU with the fewest of threads. |
| 419 | */ |
| 420 | pivot = l->l_cpu; |
| 421 | ci = pivot; |
| 422 | tci = pivot; |
| 423 | lpri = PRI_COUNT; |
| 424 | do { |
| 425 | if ((next = cpu_lookup(cpu_index(ci) + 1)) == NULL) { |
| 426 | /* Reached the end, start from the beginning. */ |
| 427 | next = cpu_lookup(0); |
| 428 | } |
| 429 | spc = &ci->ci_schedstate; |
| 430 | ici_rq = spc->spc_sched_info; |
| 431 | pri = MAX(spc->spc_curpriority, spc->spc_maxpriority); |
| 432 | if (pri > lpri) |
| 433 | continue; |
| 434 | |
| 435 | if (pri == lpri && ci_rq->r_count < ici_rq->r_count) |
| 436 | continue; |
| 437 | |
| 438 | if (!sched_migratable(l, ci)) |
| 439 | continue; |
| 440 | |
| 441 | lpri = pri; |
| 442 | tci = ci; |
| 443 | ci_rq = ici_rq; |
| 444 | } while (ci = next, ci != pivot); |
| 445 | |
| 446 | ci_rq = tci->ci_schedstate.spc_sched_info; |
| 447 | ci_rq->r_ev_push.ev_count++; |
| 448 | |
| 449 | return tci; |
| 450 | } |
| 451 | |
| 452 | /* |
| 453 | * Tries to catch an LWP from the runqueue of other CPU. |
| 454 | */ |
| 455 | static struct lwp * |
| 456 | sched_catchlwp(struct cpu_info *ci) |
| 457 | { |
| 458 | struct cpu_info *curci = curcpu(); |
| 459 | struct schedstate_percpu *spc, *curspc; |
| 460 | TAILQ_HEAD(, lwp) *q_head; |
| 461 | runqueue_t *ci_rq; |
| 462 | struct lwp *l; |
| 463 | |
| 464 | curspc = &curci->ci_schedstate; |
| 465 | spc = &ci->ci_schedstate; |
| 466 | KASSERT(curspc->spc_psid == spc->spc_psid); |
| 467 | |
| 468 | ci_rq = spc->spc_sched_info; |
| 469 | if (ci_rq->r_mcount < min_catch) { |
| 470 | spc_unlock(ci); |
| 471 | return NULL; |
| 472 | } |
| 473 | |
| 474 | /* Take the highest priority thread */ |
| 475 | q_head = sched_getrq(ci_rq, spc->spc_maxpriority); |
| 476 | l = TAILQ_FIRST(q_head); |
| 477 | |
| 478 | for (;;) { |
| 479 | /* Check the first and next result from the queue */ |
| 480 | if (l == NULL) { |
| 481 | break; |
| 482 | } |
| 483 | KASSERTMSG(l->l_stat == LSRUN, "%s l %p (%s) l_stat %d" , |
| 484 | ci->ci_data.cpu_name, |
| 485 | l, (l->l_name ? l->l_name : l->l_proc->p_comm), l->l_stat); |
| 486 | |
| 487 | /* Look for threads, whose are allowed to migrate */ |
| 488 | if ((l->l_pflag & LP_BOUND) || lwp_cache_hot(l) || |
| 489 | !sched_migratable(l, curci)) { |
| 490 | l = TAILQ_NEXT(l, l_runq); |
| 491 | continue; |
| 492 | } |
| 493 | |
| 494 | /* Grab the thread, and move to the local run queue */ |
| 495 | sched_dequeue(l); |
| 496 | |
| 497 | /* |
| 498 | * If LWP is still context switching, we may need to |
| 499 | * spin-wait before changing its CPU. |
| 500 | */ |
| 501 | if (__predict_false(l->l_ctxswtch != 0)) { |
| 502 | u_int count; |
| 503 | count = SPINLOCK_BACKOFF_MIN; |
| 504 | while (l->l_ctxswtch) |
| 505 | SPINLOCK_BACKOFF(count); |
| 506 | } |
| 507 | l->l_cpu = curci; |
| 508 | ci_rq->r_ev_pull.ev_count++; |
| 509 | lwp_unlock_to(l, curspc->spc_mutex); |
| 510 | sched_enqueue(l, false); |
| 511 | return l; |
| 512 | } |
| 513 | spc_unlock(ci); |
| 514 | |
| 515 | return l; |
| 516 | } |
| 517 | |
| 518 | /* |
| 519 | * Periodical calculations for balancing. |
| 520 | */ |
| 521 | static void |
| 522 | sched_balance(void *nocallout) |
| 523 | { |
| 524 | struct cpu_info *ci, *hci; |
| 525 | runqueue_t *ci_rq; |
| 526 | CPU_INFO_ITERATOR cii; |
| 527 | u_int highest; |
| 528 | |
| 529 | hci = curcpu(); |
| 530 | highest = 0; |
| 531 | |
| 532 | /* Make lockless countings */ |
| 533 | for (CPU_INFO_FOREACH(cii, ci)) { |
| 534 | ci_rq = ci->ci_schedstate.spc_sched_info; |
| 535 | |
| 536 | /* Average count of the threads */ |
| 537 | ci_rq->r_avgcount = (ci_rq->r_avgcount + ci_rq->r_mcount) >> 1; |
| 538 | |
| 539 | /* Look for CPU with the highest average */ |
| 540 | if (ci_rq->r_avgcount > highest) { |
| 541 | hci = ci; |
| 542 | highest = ci_rq->r_avgcount; |
| 543 | } |
| 544 | } |
| 545 | |
| 546 | /* Update the worker */ |
| 547 | worker_ci = hci; |
| 548 | |
| 549 | if (nocallout == NULL) |
| 550 | callout_schedule(&balance_ch, balance_period); |
| 551 | } |
| 552 | |
| 553 | /* |
| 554 | * Called from each CPU's idle loop. |
| 555 | */ |
| 556 | void |
| 557 | sched_idle(void) |
| 558 | { |
| 559 | struct cpu_info *ci = curcpu(), *tci = NULL; |
| 560 | struct schedstate_percpu *spc, *tspc; |
| 561 | runqueue_t *ci_rq; |
| 562 | bool dlock = false; |
| 563 | |
| 564 | /* Check if there is a migrating LWP */ |
| 565 | spc = &ci->ci_schedstate; |
| 566 | if (spc->spc_migrating == NULL) |
| 567 | goto no_migration; |
| 568 | |
| 569 | spc_lock(ci); |
| 570 | for (;;) { |
| 571 | struct lwp *l; |
| 572 | |
| 573 | l = spc->spc_migrating; |
| 574 | if (l == NULL) |
| 575 | break; |
| 576 | |
| 577 | /* |
| 578 | * If second attempt, and target CPU has changed, |
| 579 | * drop the old lock. |
| 580 | */ |
| 581 | if (dlock == true && tci != l->l_target_cpu) { |
| 582 | KASSERT(tci != NULL); |
| 583 | spc_unlock(tci); |
| 584 | dlock = false; |
| 585 | } |
| 586 | |
| 587 | /* |
| 588 | * Nothing to do if destination has changed to the |
| 589 | * local CPU, or migration was done by other CPU. |
| 590 | */ |
| 591 | tci = l->l_target_cpu; |
| 592 | if (tci == NULL || tci == ci) { |
| 593 | spc->spc_migrating = NULL; |
| 594 | l->l_target_cpu = NULL; |
| 595 | break; |
| 596 | } |
| 597 | tspc = &tci->ci_schedstate; |
| 598 | |
| 599 | /* |
| 600 | * Double-lock the runqueues. |
| 601 | * We do that only once. |
| 602 | */ |
| 603 | if (dlock == false) { |
| 604 | dlock = true; |
| 605 | if (ci < tci) { |
| 606 | spc_lock(tci); |
| 607 | } else if (!mutex_tryenter(tspc->spc_mutex)) { |
| 608 | spc_unlock(ci); |
| 609 | spc_lock(tci); |
| 610 | spc_lock(ci); |
| 611 | /* Check the situation again.. */ |
| 612 | continue; |
| 613 | } |
| 614 | } |
| 615 | |
| 616 | /* Migrate the thread */ |
| 617 | KASSERT(l->l_stat == LSRUN); |
| 618 | spc->spc_migrating = NULL; |
| 619 | l->l_target_cpu = NULL; |
| 620 | sched_dequeue(l); |
| 621 | l->l_cpu = tci; |
| 622 | lwp_setlock(l, tspc->spc_mutex); |
| 623 | sched_enqueue(l, false); |
| 624 | break; |
| 625 | } |
| 626 | if (dlock == true) { |
| 627 | KASSERT(tci != NULL); |
| 628 | spc_unlock(tci); |
| 629 | } |
| 630 | spc_unlock(ci); |
| 631 | |
| 632 | no_migration: |
| 633 | ci_rq = spc->spc_sched_info; |
| 634 | if ((spc->spc_flags & SPCF_OFFLINE) != 0 || ci_rq->r_count != 0) { |
| 635 | return; |
| 636 | } |
| 637 | |
| 638 | /* Reset the counter, and call the balancer */ |
| 639 | ci_rq->r_avgcount = 0; |
| 640 | sched_balance(ci); |
| 641 | tci = worker_ci; |
| 642 | tspc = &tci->ci_schedstate; |
| 643 | if (ci == tci || spc->spc_psid != tspc->spc_psid) |
| 644 | return; |
| 645 | spc_dlock(ci, tci); |
| 646 | (void)sched_catchlwp(tci); |
| 647 | spc_unlock(ci); |
| 648 | } |
| 649 | |
| 650 | #else |
| 651 | |
| 652 | /* |
| 653 | * stubs for !MULTIPROCESSOR |
| 654 | */ |
| 655 | |
| 656 | struct cpu_info * |
| 657 | sched_takecpu(struct lwp *l) |
| 658 | { |
| 659 | |
| 660 | return l->l_cpu; |
| 661 | } |
| 662 | |
| 663 | void |
| 664 | sched_idle(void) |
| 665 | { |
| 666 | |
| 667 | } |
| 668 | #endif /* MULTIPROCESSOR */ |
| 669 | |
| 670 | /* |
| 671 | * Scheduling statistics and balancing. |
| 672 | */ |
| 673 | void |
| 674 | sched_lwp_stats(struct lwp *l) |
| 675 | { |
| 676 | int batch; |
| 677 | |
| 678 | KASSERT(lwp_locked(l, NULL)); |
| 679 | |
| 680 | /* Update sleep time */ |
| 681 | if (l->l_stat == LSSLEEP || l->l_stat == LSSTOP || |
| 682 | l->l_stat == LSSUSPENDED) |
| 683 | l->l_slptime++; |
| 684 | |
| 685 | /* |
| 686 | * Set that thread is more CPU-bound, if sum of run time exceeds the |
| 687 | * sum of sleep time. Check if thread is CPU-bound a first time. |
| 688 | */ |
| 689 | batch = (l->l_rticksum > l->l_slpticksum); |
| 690 | if (batch != 0) { |
| 691 | if ((l->l_flag & LW_BATCH) == 0) |
| 692 | batch = 0; |
| 693 | l->l_flag |= LW_BATCH; |
| 694 | } else |
| 695 | l->l_flag &= ~LW_BATCH; |
| 696 | |
| 697 | /* |
| 698 | * If thread is CPU-bound and never sleeps, it would occupy the CPU. |
| 699 | * In such case reset the value of last sleep, and check it later, if |
| 700 | * it is still zero - perform the migration, unmark the batch flag. |
| 701 | */ |
| 702 | if (batch && (l->l_slptime + l->l_slpticksum) == 0) { |
| 703 | if (l->l_slpticks == 0) { |
| 704 | if (l->l_target_cpu == NULL && |
| 705 | (l->l_stat == LSRUN || l->l_stat == LSONPROC)) { |
| 706 | struct cpu_info *ci = sched_takecpu(l); |
| 707 | l->l_target_cpu = (ci != l->l_cpu) ? ci : NULL; |
| 708 | } |
| 709 | l->l_flag &= ~LW_BATCH; |
| 710 | } else { |
| 711 | l->l_slpticks = 0; |
| 712 | } |
| 713 | } |
| 714 | |
| 715 | /* Reset the time sums */ |
| 716 | l->l_slpticksum = 0; |
| 717 | l->l_rticksum = 0; |
| 718 | |
| 719 | /* Scheduler-specific hook */ |
| 720 | sched_pstats_hook(l, batch); |
| 721 | #ifdef KDTRACE_HOOKS |
| 722 | curthread = l; |
| 723 | #endif |
| 724 | } |
| 725 | |
| 726 | /* |
| 727 | * Scheduler mill. |
| 728 | */ |
| 729 | struct lwp * |
| 730 | sched_nextlwp(void) |
| 731 | { |
| 732 | struct cpu_info *ci = curcpu(); |
| 733 | struct schedstate_percpu *spc; |
| 734 | TAILQ_HEAD(, lwp) *q_head; |
| 735 | runqueue_t *ci_rq; |
| 736 | struct lwp *l; |
| 737 | |
| 738 | /* Return to idle LWP if there is a migrating thread */ |
| 739 | spc = &ci->ci_schedstate; |
| 740 | if (__predict_false(spc->spc_migrating != NULL)) |
| 741 | return NULL; |
| 742 | ci_rq = spc->spc_sched_info; |
| 743 | |
| 744 | #ifdef MULTIPROCESSOR |
| 745 | /* If runqueue is empty, try to catch some thread from other CPU */ |
| 746 | if (__predict_false(ci_rq->r_count == 0)) { |
| 747 | struct schedstate_percpu *cspc; |
| 748 | struct cpu_info *cci; |
| 749 | |
| 750 | /* Offline CPUs should not perform this, however */ |
| 751 | if (__predict_false(spc->spc_flags & SPCF_OFFLINE)) |
| 752 | return NULL; |
| 753 | |
| 754 | /* Reset the counter, and call the balancer */ |
| 755 | ci_rq->r_avgcount = 0; |
| 756 | sched_balance(ci); |
| 757 | cci = worker_ci; |
| 758 | cspc = &cci->ci_schedstate; |
| 759 | if (ci == cci || spc->spc_psid != cspc->spc_psid || |
| 760 | !mutex_tryenter(cci->ci_schedstate.spc_mutex)) |
| 761 | return NULL; |
| 762 | return sched_catchlwp(cci); |
| 763 | } |
| 764 | #else |
| 765 | if (__predict_false(ci_rq->r_count == 0)) |
| 766 | return NULL; |
| 767 | #endif |
| 768 | |
| 769 | /* Take the highest priority thread */ |
| 770 | KASSERT(ci_rq->r_bitmap[spc->spc_maxpriority >> BITMAP_SHIFT]); |
| 771 | q_head = sched_getrq(ci_rq, spc->spc_maxpriority); |
| 772 | l = TAILQ_FIRST(q_head); |
| 773 | KASSERT(l != NULL); |
| 774 | |
| 775 | sched_oncpu(l); |
| 776 | l->l_rticks = hardclock_ticks; |
| 777 | |
| 778 | return l; |
| 779 | } |
| 780 | |
| 781 | /* |
| 782 | * sched_curcpu_runnable_p: return if curcpu() should exit the idle loop. |
| 783 | */ |
| 784 | |
| 785 | bool |
| 786 | sched_curcpu_runnable_p(void) |
| 787 | { |
| 788 | const struct cpu_info *ci; |
| 789 | const struct schedstate_percpu *spc; |
| 790 | const runqueue_t *ci_rq; |
| 791 | bool rv; |
| 792 | |
| 793 | kpreempt_disable(); |
| 794 | ci = curcpu(); |
| 795 | spc = &ci->ci_schedstate; |
| 796 | ci_rq = spc->spc_sched_info; |
| 797 | |
| 798 | #ifndef __HAVE_FAST_SOFTINTS |
| 799 | if (ci->ci_data.cpu_softints) { |
| 800 | kpreempt_enable(); |
| 801 | return true; |
| 802 | } |
| 803 | #endif |
| 804 | |
| 805 | rv = (ci_rq->r_count != 0) ? true : false; |
| 806 | kpreempt_enable(); |
| 807 | |
| 808 | return rv; |
| 809 | } |
| 810 | |
| 811 | /* |
| 812 | * Sysctl nodes and initialization. |
| 813 | */ |
| 814 | |
| 815 | SYSCTL_SETUP(sysctl_sched_setup, "sysctl sched setup" ) |
| 816 | { |
| 817 | const struct sysctlnode *node = NULL; |
| 818 | |
| 819 | sysctl_createv(clog, 0, NULL, &node, |
| 820 | CTLFLAG_PERMANENT, |
| 821 | CTLTYPE_NODE, "sched" , |
| 822 | SYSCTL_DESCR("Scheduler options" ), |
| 823 | NULL, 0, NULL, 0, |
| 824 | CTL_KERN, CTL_CREATE, CTL_EOL); |
| 825 | |
| 826 | if (node == NULL) |
| 827 | return; |
| 828 | |
| 829 | sysctl_createv(clog, 0, &node, NULL, |
| 830 | CTLFLAG_PERMANENT | CTLFLAG_READWRITE, |
| 831 | CTLTYPE_INT, "cacheht_time" , |
| 832 | SYSCTL_DESCR("Cache hotness time (in ticks)" ), |
| 833 | NULL, 0, &cacheht_time, 0, |
| 834 | CTL_CREATE, CTL_EOL); |
| 835 | sysctl_createv(clog, 0, &node, NULL, |
| 836 | CTLFLAG_PERMANENT | CTLFLAG_READWRITE, |
| 837 | CTLTYPE_INT, "balance_period" , |
| 838 | SYSCTL_DESCR("Balance period (in ticks)" ), |
| 839 | NULL, 0, &balance_period, 0, |
| 840 | CTL_CREATE, CTL_EOL); |
| 841 | sysctl_createv(clog, 0, &node, NULL, |
| 842 | CTLFLAG_PERMANENT | CTLFLAG_READWRITE, |
| 843 | CTLTYPE_INT, "min_catch" , |
| 844 | SYSCTL_DESCR("Minimal count of threads for catching" ), |
| 845 | NULL, 0, &min_catch, 0, |
| 846 | CTL_CREATE, CTL_EOL); |
| 847 | sysctl_createv(clog, 0, &node, NULL, |
| 848 | CTLFLAG_PERMANENT | CTLFLAG_READWRITE, |
| 849 | CTLTYPE_INT, "timesoftints" , |
| 850 | SYSCTL_DESCR("Track CPU time for soft interrupts" ), |
| 851 | NULL, 0, &softint_timing, 0, |
| 852 | CTL_CREATE, CTL_EOL); |
| 853 | sysctl_createv(clog, 0, &node, NULL, |
| 854 | CTLFLAG_PERMANENT | CTLFLAG_READWRITE, |
| 855 | CTLTYPE_INT, "kpreempt_pri" , |
| 856 | SYSCTL_DESCR("Minimum priority to trigger kernel preemption" ), |
| 857 | NULL, 0, &sched_kpreempt_pri, 0, |
| 858 | CTL_CREATE, CTL_EOL); |
| 859 | sysctl_createv(clog, 0, &node, NULL, |
| 860 | CTLFLAG_PERMANENT | CTLFLAG_READWRITE, |
| 861 | CTLTYPE_INT, "upreempt_pri" , |
| 862 | SYSCTL_DESCR("Minimum priority to trigger user preemption" ), |
| 863 | NULL, 0, &sched_upreempt_pri, 0, |
| 864 | CTL_CREATE, CTL_EOL); |
| 865 | } |
| 866 | |
| 867 | /* |
| 868 | * Debugging. |
| 869 | */ |
| 870 | |
| 871 | #ifdef DDB |
| 872 | |
| 873 | void |
| 874 | sched_print_runqueue(void (*pr)(const char *, ...)) |
| 875 | { |
| 876 | runqueue_t *ci_rq; |
| 877 | struct cpu_info *ci, *tci; |
| 878 | struct schedstate_percpu *spc; |
| 879 | struct lwp *l; |
| 880 | struct proc *p; |
| 881 | CPU_INFO_ITERATOR cii; |
| 882 | |
| 883 | for (CPU_INFO_FOREACH(cii, ci)) { |
| 884 | int i; |
| 885 | |
| 886 | spc = &ci->ci_schedstate; |
| 887 | ci_rq = spc->spc_sched_info; |
| 888 | |
| 889 | (*pr)("Run-queue (CPU = %u):\n" , ci->ci_index); |
| 890 | (*pr)(" pid.lid = %d.%d, r_count = %u, r_avgcount = %u, " |
| 891 | "maxpri = %d, mlwp = %p\n" , |
| 892 | #ifdef MULTIPROCESSOR |
| 893 | ci->ci_curlwp->l_proc->p_pid, ci->ci_curlwp->l_lid, |
| 894 | #else |
| 895 | curlwp->l_proc->p_pid, curlwp->l_lid, |
| 896 | #endif |
| 897 | ci_rq->r_count, ci_rq->r_avgcount, spc->spc_maxpriority, |
| 898 | spc->spc_migrating); |
| 899 | i = (PRI_COUNT >> BITMAP_SHIFT) - 1; |
| 900 | do { |
| 901 | uint32_t q; |
| 902 | q = ci_rq->r_bitmap[i]; |
| 903 | (*pr)(" bitmap[%d] => [ %d (0x%x) ]\n" , i, ffs(q), q); |
| 904 | } while (i--); |
| 905 | } |
| 906 | |
| 907 | (*pr)(" %5s %4s %4s %10s %3s %18s %4s %4s %s\n" , |
| 908 | "LID" , "PRI" , "EPRI" , "FL" , "ST" , "LWP" , "CPU" , "TCI" , "LRTICKS" ); |
| 909 | |
| 910 | PROCLIST_FOREACH(p, &allproc) { |
| 911 | (*pr)(" /- %d (%s)\n" , (int)p->p_pid, p->p_comm); |
| 912 | LIST_FOREACH(l, &p->p_lwps, l_sibling) { |
| 913 | ci = l->l_cpu; |
| 914 | tci = l->l_target_cpu; |
| 915 | (*pr)(" | %5d %4u %4u 0x%8.8x %3s %18p %4u %4d %u\n" , |
| 916 | (int)l->l_lid, l->l_priority, lwp_eprio(l), |
| 917 | l->l_flag, l->l_stat == LSRUN ? "RQ" : |
| 918 | (l->l_stat == LSSLEEP ? "SQ" : "-" ), |
| 919 | l, ci->ci_index, (tci ? tci->ci_index : -1), |
| 920 | (u_int)(hardclock_ticks - l->l_rticks)); |
| 921 | } |
| 922 | } |
| 923 | } |
| 924 | |
| 925 | #endif |
| 926 | |