1 /*- 2 * Copyright (c) 1982, 1986, 1990, 1991, 1993 3 * The Regents of the University of California. All rights reserved. 4 * (c) UNIX System Laboratories, Inc. 5 * All or some portions of this file are derived from material licensed 6 * to the University of California by American Telephone and Telegraph 7 * Co. or Unix System Laboratories, Inc. and are reproduced herein with 8 * the permission of UNIX System Laboratories, Inc. 9 * 10 * Redistribution and use in source and binary forms, with or without 11 * modification, are permitted provided that the following conditions 12 * are met: 13 * 1. Redistributions of source code must retain the above copyright 14 * notice, this list of conditions and the following disclaimer. 15 * 2. Redistributions in binary form must reproduce the above copyright 16 * notice, this list of conditions and the following disclaimer in the 17 * documentation and/or other materials provided with the distribution. 18 * 3. Neither the name of the University nor the names of its contributors 19 * may be used to endorse or promote products derived from this software 20 * without specific prior written permission. 21 * 22 * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND 23 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE 24 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE 25 * ARE DISCLAIMED. IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE 26 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL 27 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS 28 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) 29 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT 30 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY 31 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF 32 * SUCH DAMAGE. 33 */ 34 35 #include <sys/cdefs.h> 36 __FBSDID("$FreeBSD$"); 37 38 #include "opt_hwpmc_hooks.h" 39 #include "opt_sched.h" 40 41 #include <sys/param.h> 42 #include <sys/systm.h> 43 #include <sys/cpuset.h> 44 #include <sys/kernel.h> 45 #include <sys/ktr.h> 46 #include <sys/lock.h> 47 #include <sys/kthread.h> 48 #include <sys/mutex.h> 49 #include <sys/proc.h> 50 #include <sys/resourcevar.h> 51 #include <sys/sched.h> 52 #include <sys/sdt.h> 53 #include <sys/smp.h> 54 #include <sys/sysctl.h> 55 #include <sys/sx.h> 56 #include <sys/turnstile.h> 57 #include <sys/umtx.h> 58 #include <machine/pcb.h> 59 #include <machine/smp.h> 60 61 #ifdef HWPMC_HOOKS 62 #include <sys/pmckern.h> 63 #endif 64 65 #ifdef KDTRACE_HOOKS 66 #include <sys/dtrace_bsd.h> 67 int dtrace_vtime_active; 68 dtrace_vtime_switch_func_t dtrace_vtime_switch_func; 69 #endif 70 71 /* 72 * INVERSE_ESTCPU_WEIGHT is only suitable for statclock() frequencies in 73 * the range 100-256 Hz (approximately). 74 */ 75 #define ESTCPULIM(e) \ 76 min((e), INVERSE_ESTCPU_WEIGHT * (NICE_WEIGHT * (PRIO_MAX - PRIO_MIN) - \ 77 RQ_PPQ) + INVERSE_ESTCPU_WEIGHT - 1) 78 #ifdef SMP 79 #define INVERSE_ESTCPU_WEIGHT (8 * smp_cpus) 80 #else 81 #define INVERSE_ESTCPU_WEIGHT 8 /* 1 / (priorities per estcpu level). */ 82 #endif 83 #define NICE_WEIGHT 1 /* Priorities per nice level. */ 84 85 #define TS_NAME_LEN (MAXCOMLEN + sizeof(" td ") + sizeof(__XSTRING(UINT_MAX))) 86 87 /* 88 * The schedulable entity that runs a context. 89 * This is an extension to the thread structure and is tailored to 90 * the requirements of this scheduler. 91 * All fields are protected by the scheduler lock. 92 */ 93 struct td_sched { 94 fixpt_t ts_pctcpu; /* %cpu during p_swtime. */ 95 u_int ts_estcpu; /* Estimated cpu utilization. */ 96 int ts_cpticks; /* Ticks of cpu time. */ 97 int ts_slptime; /* Seconds !RUNNING. */ 98 int ts_slice; /* Remaining part of time slice. */ 99 int ts_flags; 100 struct runq *ts_runq; /* runq the thread is currently on */ 101 #ifdef KTR 102 char ts_name[TS_NAME_LEN]; 103 #endif 104 }; 105 106 /* flags kept in td_flags */ 107 #define TDF_DIDRUN TDF_SCHED0 /* thread actually ran. */ 108 #define TDF_BOUND TDF_SCHED1 /* Bound to one CPU. */ 109 #define TDF_SLICEEND TDF_SCHED2 /* Thread time slice is over. */ 110 111 /* flags kept in ts_flags */ 112 #define TSF_AFFINITY 0x0001 /* Has a non-"full" CPU set. */ 113 114 #define SKE_RUNQ_PCPU(ts) \ 115 ((ts)->ts_runq != 0 && (ts)->ts_runq != &runq) 116 117 #define THREAD_CAN_SCHED(td, cpu) \ 118 CPU_ISSET((cpu), &(td)->td_cpuset->cs_mask) 119 120 _Static_assert(sizeof(struct thread) + sizeof(struct td_sched) <= 121 sizeof(struct thread0_storage), 122 "increase struct thread0_storage.t0st_sched size"); 123 124 static struct mtx sched_lock; 125 126 static int realstathz = 127; /* stathz is sometimes 0 and run off of hz. */ 127 static int sched_tdcnt; /* Total runnable threads in the system. */ 128 static int sched_slice = 12; /* Thread run time before rescheduling. */ 129 130 static void setup_runqs(void); 131 static void schedcpu(void); 132 static void schedcpu_thread(void); 133 static void sched_priority(struct thread *td, u_char prio); 134 static void sched_setup(void *dummy); 135 static void maybe_resched(struct thread *td); 136 static void updatepri(struct thread *td); 137 static void resetpriority(struct thread *td); 138 static void resetpriority_thread(struct thread *td); 139 #ifdef SMP 140 static int sched_pickcpu(struct thread *td); 141 static int forward_wakeup(int cpunum); 142 static void kick_other_cpu(int pri, int cpuid); 143 #endif 144 145 static struct kproc_desc sched_kp = { 146 "schedcpu", 147 schedcpu_thread, 148 NULL 149 }; 150 SYSINIT(schedcpu, SI_SUB_LAST, SI_ORDER_FIRST, kproc_start, 151 &sched_kp); 152 SYSINIT(sched_setup, SI_SUB_RUN_QUEUE, SI_ORDER_FIRST, sched_setup, NULL); 153 154 static void sched_initticks(void *dummy); 155 SYSINIT(sched_initticks, SI_SUB_CLOCKS, SI_ORDER_THIRD, sched_initticks, 156 NULL); 157 158 /* 159 * Global run queue. 160 */ 161 static struct runq runq; 162 163 #ifdef SMP 164 /* 165 * Per-CPU run queues 166 */ 167 static struct runq runq_pcpu[MAXCPU]; 168 long runq_length[MAXCPU]; 169 170 static cpuset_t idle_cpus_mask; 171 #endif 172 173 struct pcpuidlestat { 174 u_int idlecalls; 175 u_int oldidlecalls; 176 }; 177 static DPCPU_DEFINE(struct pcpuidlestat, idlestat); 178 179 static void 180 setup_runqs(void) 181 { 182 #ifdef SMP 183 int i; 184 185 for (i = 0; i < MAXCPU; ++i) 186 runq_init(&runq_pcpu[i]); 187 #endif 188 189 runq_init(&runq); 190 } 191 192 static int 193 sysctl_kern_quantum(SYSCTL_HANDLER_ARGS) 194 { 195 int error, new_val, period; 196 197 period = 1000000 / realstathz; 198 new_val = period * sched_slice; 199 error = sysctl_handle_int(oidp, &new_val, 0, req); 200 if (error != 0 || req->newptr == NULL) 201 return (error); 202 if (new_val <= 0) 203 return (EINVAL); 204 sched_slice = imax(1, (new_val + period / 2) / period); 205 hogticks = imax(1, (2 * hz * sched_slice + realstathz / 2) / 206 realstathz); 207 return (0); 208 } 209 210 SYSCTL_NODE(_kern, OID_AUTO, sched, CTLFLAG_RD, 0, "Scheduler"); 211 212 SYSCTL_STRING(_kern_sched, OID_AUTO, name, CTLFLAG_RD, "4BSD", 0, 213 "Scheduler name"); 214 SYSCTL_PROC(_kern_sched, OID_AUTO, quantum, CTLTYPE_INT | CTLFLAG_RW, 215 NULL, 0, sysctl_kern_quantum, "I", 216 "Quantum for timeshare threads in microseconds"); 217 SYSCTL_INT(_kern_sched, OID_AUTO, slice, CTLFLAG_RW, &sched_slice, 0, 218 "Quantum for timeshare threads in stathz ticks"); 219 #ifdef SMP 220 /* Enable forwarding of wakeups to all other cpus */ 221 static SYSCTL_NODE(_kern_sched, OID_AUTO, ipiwakeup, CTLFLAG_RD, NULL, 222 "Kernel SMP"); 223 224 static int runq_fuzz = 1; 225 SYSCTL_INT(_kern_sched, OID_AUTO, runq_fuzz, CTLFLAG_RW, &runq_fuzz, 0, ""); 226 227 static int forward_wakeup_enabled = 1; 228 SYSCTL_INT(_kern_sched_ipiwakeup, OID_AUTO, enabled, CTLFLAG_RW, 229 &forward_wakeup_enabled, 0, 230 "Forwarding of wakeup to idle CPUs"); 231 232 static int forward_wakeups_requested = 0; 233 SYSCTL_INT(_kern_sched_ipiwakeup, OID_AUTO, requested, CTLFLAG_RD, 234 &forward_wakeups_requested, 0, 235 "Requests for Forwarding of wakeup to idle CPUs"); 236 237 static int forward_wakeups_delivered = 0; 238 SYSCTL_INT(_kern_sched_ipiwakeup, OID_AUTO, delivered, CTLFLAG_RD, 239 &forward_wakeups_delivered, 0, 240 "Completed Forwarding of wakeup to idle CPUs"); 241 242 static int forward_wakeup_use_mask = 1; 243 SYSCTL_INT(_kern_sched_ipiwakeup, OID_AUTO, usemask, CTLFLAG_RW, 244 &forward_wakeup_use_mask, 0, 245 "Use the mask of idle cpus"); 246 247 static int forward_wakeup_use_loop = 0; 248 SYSCTL_INT(_kern_sched_ipiwakeup, OID_AUTO, useloop, CTLFLAG_RW, 249 &forward_wakeup_use_loop, 0, 250 "Use a loop to find idle cpus"); 251 252 #endif 253 #if 0 254 static int sched_followon = 0; 255 SYSCTL_INT(_kern_sched, OID_AUTO, followon, CTLFLAG_RW, 256 &sched_followon, 0, 257 "allow threads to share a quantum"); 258 #endif 259 260 SDT_PROVIDER_DEFINE(sched); 261 262 SDT_PROBE_DEFINE3(sched, , , change__pri, "struct thread *", 263 "struct proc *", "uint8_t"); 264 SDT_PROBE_DEFINE3(sched, , , dequeue, "struct thread *", 265 "struct proc *", "void *"); 266 SDT_PROBE_DEFINE4(sched, , , enqueue, "struct thread *", 267 "struct proc *", "void *", "int"); 268 SDT_PROBE_DEFINE4(sched, , , lend__pri, "struct thread *", 269 "struct proc *", "uint8_t", "struct thread *"); 270 SDT_PROBE_DEFINE2(sched, , , load__change, "int", "int"); 271 SDT_PROBE_DEFINE2(sched, , , off__cpu, "struct thread *", 272 "struct proc *"); 273 SDT_PROBE_DEFINE(sched, , , on__cpu); 274 SDT_PROBE_DEFINE(sched, , , remain__cpu); 275 SDT_PROBE_DEFINE2(sched, , , surrender, "struct thread *", 276 "struct proc *"); 277 278 static __inline void 279 sched_load_add(void) 280 { 281 282 sched_tdcnt++; 283 KTR_COUNTER0(KTR_SCHED, "load", "global load", sched_tdcnt); 284 SDT_PROBE2(sched, , , load__change, NOCPU, sched_tdcnt); 285 } 286 287 static __inline void 288 sched_load_rem(void) 289 { 290 291 sched_tdcnt--; 292 KTR_COUNTER0(KTR_SCHED, "load", "global load", sched_tdcnt); 293 SDT_PROBE2(sched, , , load__change, NOCPU, sched_tdcnt); 294 } 295 /* 296 * Arrange to reschedule if necessary, taking the priorities and 297 * schedulers into account. 298 */ 299 static void 300 maybe_resched(struct thread *td) 301 { 302 303 THREAD_LOCK_ASSERT(td, MA_OWNED); 304 if (td->td_priority < curthread->td_priority) 305 curthread->td_flags |= TDF_NEEDRESCHED; 306 } 307 308 /* 309 * This function is called when a thread is about to be put on run queue 310 * because it has been made runnable or its priority has been adjusted. It 311 * determines if the new thread should be immediately preempted to. If so, 312 * it switches to it and eventually returns true. If not, it returns false 313 * so that the caller may place the thread on an appropriate run queue. 314 */ 315 int 316 maybe_preempt(struct thread *td) 317 { 318 #ifdef PREEMPTION 319 struct thread *ctd; 320 int cpri, pri; 321 322 /* 323 * The new thread should not preempt the current thread if any of the 324 * following conditions are true: 325 * 326 * - The kernel is in the throes of crashing (panicstr). 327 * - The current thread has a higher (numerically lower) or 328 * equivalent priority. Note that this prevents curthread from 329 * trying to preempt to itself. 330 * - It is too early in the boot for context switches (cold is set). 331 * - The current thread has an inhibitor set or is in the process of 332 * exiting. In this case, the current thread is about to switch 333 * out anyways, so there's no point in preempting. If we did, 334 * the current thread would not be properly resumed as well, so 335 * just avoid that whole landmine. 336 * - If the new thread's priority is not a realtime priority and 337 * the current thread's priority is not an idle priority and 338 * FULL_PREEMPTION is disabled. 339 * 340 * If all of these conditions are false, but the current thread is in 341 * a nested critical section, then we have to defer the preemption 342 * until we exit the critical section. Otherwise, switch immediately 343 * to the new thread. 344 */ 345 ctd = curthread; 346 THREAD_LOCK_ASSERT(td, MA_OWNED); 347 KASSERT((td->td_inhibitors == 0), 348 ("maybe_preempt: trying to run inhibited thread")); 349 pri = td->td_priority; 350 cpri = ctd->td_priority; 351 if (panicstr != NULL || pri >= cpri || cold /* || dumping */ || 352 TD_IS_INHIBITED(ctd)) 353 return (0); 354 #ifndef FULL_PREEMPTION 355 if (pri > PRI_MAX_ITHD && cpri < PRI_MIN_IDLE) 356 return (0); 357 #endif 358 359 if (ctd->td_critnest > 1) { 360 CTR1(KTR_PROC, "maybe_preempt: in critical section %d", 361 ctd->td_critnest); 362 ctd->td_owepreempt = 1; 363 return (0); 364 } 365 /* 366 * Thread is runnable but not yet put on system run queue. 367 */ 368 MPASS(ctd->td_lock == td->td_lock); 369 MPASS(TD_ON_RUNQ(td)); 370 TD_SET_RUNNING(td); 371 CTR3(KTR_PROC, "preempting to thread %p (pid %d, %s)\n", td, 372 td->td_proc->p_pid, td->td_name); 373 mi_switch(SW_INVOL | SW_PREEMPT | SWT_PREEMPT, td); 374 /* 375 * td's lock pointer may have changed. We have to return with it 376 * locked. 377 */ 378 spinlock_enter(); 379 thread_unlock(ctd); 380 thread_lock(td); 381 spinlock_exit(); 382 return (1); 383 #else 384 return (0); 385 #endif 386 } 387 388 /* 389 * Constants for digital decay and forget: 390 * 90% of (ts_estcpu) usage in 5 * loadav time 391 * 95% of (ts_pctcpu) usage in 60 seconds (load insensitive) 392 * Note that, as ps(1) mentions, this can let percentages 393 * total over 100% (I've seen 137.9% for 3 processes). 394 * 395 * Note that schedclock() updates ts_estcpu and p_cpticks asynchronously. 396 * 397 * We wish to decay away 90% of ts_estcpu in (5 * loadavg) seconds. 398 * That is, the system wants to compute a value of decay such 399 * that the following for loop: 400 * for (i = 0; i < (5 * loadavg); i++) 401 * ts_estcpu *= decay; 402 * will compute 403 * ts_estcpu *= 0.1; 404 * for all values of loadavg: 405 * 406 * Mathematically this loop can be expressed by saying: 407 * decay ** (5 * loadavg) ~= .1 408 * 409 * The system computes decay as: 410 * decay = (2 * loadavg) / (2 * loadavg + 1) 411 * 412 * We wish to prove that the system's computation of decay 413 * will always fulfill the equation: 414 * decay ** (5 * loadavg) ~= .1 415 * 416 * If we compute b as: 417 * b = 2 * loadavg 418 * then 419 * decay = b / (b + 1) 420 * 421 * We now need to prove two things: 422 * 1) Given factor ** (5 * loadavg) ~= .1, prove factor == b/(b+1) 423 * 2) Given b/(b+1) ** power ~= .1, prove power == (5 * loadavg) 424 * 425 * Facts: 426 * For x close to zero, exp(x) =~ 1 + x, since 427 * exp(x) = 0! + x**1/1! + x**2/2! + ... . 428 * therefore exp(-1/b) =~ 1 - (1/b) = (b-1)/b. 429 * For x close to zero, ln(1+x) =~ x, since 430 * ln(1+x) = x - x**2/2 + x**3/3 - ... -1 < x < 1 431 * therefore ln(b/(b+1)) = ln(1 - 1/(b+1)) =~ -1/(b+1). 432 * ln(.1) =~ -2.30 433 * 434 * Proof of (1): 435 * Solve (factor)**(power) =~ .1 given power (5*loadav): 436 * solving for factor, 437 * ln(factor) =~ (-2.30/5*loadav), or 438 * factor =~ exp(-1/((5/2.30)*loadav)) =~ exp(-1/(2*loadav)) = 439 * exp(-1/b) =~ (b-1)/b =~ b/(b+1). QED 440 * 441 * Proof of (2): 442 * Solve (factor)**(power) =~ .1 given factor == (b/(b+1)): 443 * solving for power, 444 * power*ln(b/(b+1)) =~ -2.30, or 445 * power =~ 2.3 * (b + 1) = 4.6*loadav + 2.3 =~ 5*loadav. QED 446 * 447 * Actual power values for the implemented algorithm are as follows: 448 * loadav: 1 2 3 4 449 * power: 5.68 10.32 14.94 19.55 450 */ 451 452 /* calculations for digital decay to forget 90% of usage in 5*loadav sec */ 453 #define loadfactor(loadav) (2 * (loadav)) 454 #define decay_cpu(loadfac, cpu) (((loadfac) * (cpu)) / ((loadfac) + FSCALE)) 455 456 /* decay 95% of `ts_pctcpu' in 60 seconds; see CCPU_SHIFT before changing */ 457 static fixpt_t ccpu = 0.95122942450071400909 * FSCALE; /* exp(-1/20) */ 458 SYSCTL_UINT(_kern, OID_AUTO, ccpu, CTLFLAG_RD, &ccpu, 0, ""); 459 460 /* 461 * If `ccpu' is not equal to `exp(-1/20)' and you still want to use the 462 * faster/more-accurate formula, you'll have to estimate CCPU_SHIFT below 463 * and possibly adjust FSHIFT in "param.h" so that (FSHIFT >= CCPU_SHIFT). 464 * 465 * To estimate CCPU_SHIFT for exp(-1/20), the following formula was used: 466 * 1 - exp(-1/20) ~= 0.0487 ~= 0.0488 == 1 (fixed pt, *11* bits). 467 * 468 * If you don't want to bother with the faster/more-accurate formula, you 469 * can set CCPU_SHIFT to (FSHIFT + 1) which will use a slower/less-accurate 470 * (more general) method of calculating the %age of CPU used by a process. 471 */ 472 #define CCPU_SHIFT 11 473 474 /* 475 * Recompute process priorities, every hz ticks. 476 * MP-safe, called without the Giant mutex. 477 */ 478 /* ARGSUSED */ 479 static void 480 schedcpu(void) 481 { 482 register fixpt_t loadfac = loadfactor(averunnable.ldavg[0]); 483 struct thread *td; 484 struct proc *p; 485 struct td_sched *ts; 486 int awake; 487 488 sx_slock(&allproc_lock); 489 FOREACH_PROC_IN_SYSTEM(p) { 490 PROC_LOCK(p); 491 if (p->p_state == PRS_NEW) { 492 PROC_UNLOCK(p); 493 continue; 494 } 495 FOREACH_THREAD_IN_PROC(p, td) { 496 awake = 0; 497 ts = td_get_sched(td); 498 thread_lock(td); 499 /* 500 * Increment sleep time (if sleeping). We 501 * ignore overflow, as above. 502 */ 503 /* 504 * The td_sched slptimes are not touched in wakeup 505 * because the thread may not HAVE everything in 506 * memory? XXX I think this is out of date. 507 */ 508 if (TD_ON_RUNQ(td)) { 509 awake = 1; 510 td->td_flags &= ~TDF_DIDRUN; 511 } else if (TD_IS_RUNNING(td)) { 512 awake = 1; 513 /* Do not clear TDF_DIDRUN */ 514 } else if (td->td_flags & TDF_DIDRUN) { 515 awake = 1; 516 td->td_flags &= ~TDF_DIDRUN; 517 } 518 519 /* 520 * ts_pctcpu is only for ps and ttyinfo(). 521 */ 522 ts->ts_pctcpu = (ts->ts_pctcpu * ccpu) >> FSHIFT; 523 /* 524 * If the td_sched has been idle the entire second, 525 * stop recalculating its priority until 526 * it wakes up. 527 */ 528 if (ts->ts_cpticks != 0) { 529 #if (FSHIFT >= CCPU_SHIFT) 530 ts->ts_pctcpu += (realstathz == 100) 531 ? ((fixpt_t) ts->ts_cpticks) << 532 (FSHIFT - CCPU_SHIFT) : 533 100 * (((fixpt_t) ts->ts_cpticks) 534 << (FSHIFT - CCPU_SHIFT)) / realstathz; 535 #else 536 ts->ts_pctcpu += ((FSCALE - ccpu) * 537 (ts->ts_cpticks * 538 FSCALE / realstathz)) >> FSHIFT; 539 #endif 540 ts->ts_cpticks = 0; 541 } 542 /* 543 * If there are ANY running threads in this process, 544 * then don't count it as sleeping. 545 * XXX: this is broken. 546 */ 547 if (awake) { 548 if (ts->ts_slptime > 1) { 549 /* 550 * In an ideal world, this should not 551 * happen, because whoever woke us 552 * up from the long sleep should have 553 * unwound the slptime and reset our 554 * priority before we run at the stale 555 * priority. Should KASSERT at some 556 * point when all the cases are fixed. 557 */ 558 updatepri(td); 559 } 560 ts->ts_slptime = 0; 561 } else 562 ts->ts_slptime++; 563 if (ts->ts_slptime > 1) { 564 thread_unlock(td); 565 continue; 566 } 567 ts->ts_estcpu = decay_cpu(loadfac, ts->ts_estcpu); 568 resetpriority(td); 569 resetpriority_thread(td); 570 thread_unlock(td); 571 } 572 PROC_UNLOCK(p); 573 } 574 sx_sunlock(&allproc_lock); 575 } 576 577 /* 578 * Main loop for a kthread that executes schedcpu once a second. 579 */ 580 static void 581 schedcpu_thread(void) 582 { 583 584 for (;;) { 585 schedcpu(); 586 pause("-", hz); 587 } 588 } 589 590 /* 591 * Recalculate the priority of a process after it has slept for a while. 592 * For all load averages >= 1 and max ts_estcpu of 255, sleeping for at 593 * least six times the loadfactor will decay ts_estcpu to zero. 594 */ 595 static void 596 updatepri(struct thread *td) 597 { 598 struct td_sched *ts; 599 fixpt_t loadfac; 600 unsigned int newcpu; 601 602 ts = td_get_sched(td); 603 loadfac = loadfactor(averunnable.ldavg[0]); 604 if (ts->ts_slptime > 5 * loadfac) 605 ts->ts_estcpu = 0; 606 else { 607 newcpu = ts->ts_estcpu; 608 ts->ts_slptime--; /* was incremented in schedcpu() */ 609 while (newcpu && --ts->ts_slptime) 610 newcpu = decay_cpu(loadfac, newcpu); 611 ts->ts_estcpu = newcpu; 612 } 613 } 614 615 /* 616 * Compute the priority of a process when running in user mode. 617 * Arrange to reschedule if the resulting priority is better 618 * than that of the current process. 619 */ 620 static void 621 resetpriority(struct thread *td) 622 { 623 u_int newpriority; 624 625 if (td->td_pri_class != PRI_TIMESHARE) 626 return; 627 newpriority = PUSER + 628 td_get_sched(td)->ts_estcpu / INVERSE_ESTCPU_WEIGHT + 629 NICE_WEIGHT * (td->td_proc->p_nice - PRIO_MIN); 630 newpriority = min(max(newpriority, PRI_MIN_TIMESHARE), 631 PRI_MAX_TIMESHARE); 632 sched_user_prio(td, newpriority); 633 } 634 635 /* 636 * Update the thread's priority when the associated process's user 637 * priority changes. 638 */ 639 static void 640 resetpriority_thread(struct thread *td) 641 { 642 643 /* Only change threads with a time sharing user priority. */ 644 if (td->td_priority < PRI_MIN_TIMESHARE || 645 td->td_priority > PRI_MAX_TIMESHARE) 646 return; 647 648 /* XXX the whole needresched thing is broken, but not silly. */ 649 maybe_resched(td); 650 651 sched_prio(td, td->td_user_pri); 652 } 653 654 /* ARGSUSED */ 655 static void 656 sched_setup(void *dummy) 657 { 658 659 setup_runqs(); 660 661 /* Account for thread0. */ 662 sched_load_add(); 663 } 664 665 /* 666 * This routine determines time constants after stathz and hz are setup. 667 */ 668 static void 669 sched_initticks(void *dummy) 670 { 671 672 realstathz = stathz ? stathz : hz; 673 sched_slice = realstathz / 10; /* ~100ms */ 674 hogticks = imax(1, (2 * hz * sched_slice + realstathz / 2) / 675 realstathz); 676 } 677 678 /* External interfaces start here */ 679 680 /* 681 * Very early in the boot some setup of scheduler-specific 682 * parts of proc0 and of some scheduler resources needs to be done. 683 * Called from: 684 * proc0_init() 685 */ 686 void 687 schedinit(void) 688 { 689 690 /* 691 * Set up the scheduler specific parts of thread0. 692 */ 693 thread0.td_lock = &sched_lock; 694 td_get_sched(&thread0)->ts_slice = sched_slice; 695 mtx_init(&sched_lock, "sched lock", NULL, MTX_SPIN | MTX_RECURSE); 696 } 697 698 int 699 sched_runnable(void) 700 { 701 #ifdef SMP 702 return runq_check(&runq) + runq_check(&runq_pcpu[PCPU_GET(cpuid)]); 703 #else 704 return runq_check(&runq); 705 #endif 706 } 707 708 int 709 sched_rr_interval(void) 710 { 711 712 /* Convert sched_slice from stathz to hz. */ 713 return (imax(1, (sched_slice * hz + realstathz / 2) / realstathz)); 714 } 715 716 /* 717 * We adjust the priority of the current process. The priority of a 718 * process gets worse as it accumulates CPU time. The cpu usage 719 * estimator (ts_estcpu) is increased here. resetpriority() will 720 * compute a different priority each time ts_estcpu increases by 721 * INVERSE_ESTCPU_WEIGHT (until PRI_MAX_TIMESHARE is reached). The 722 * cpu usage estimator ramps up quite quickly when the process is 723 * running (linearly), and decays away exponentially, at a rate which 724 * is proportionally slower when the system is busy. The basic 725 * principle is that the system will 90% forget that the process used 726 * a lot of CPU time in 5 * loadav seconds. This causes the system to 727 * favor processes which haven't run much recently, and to round-robin 728 * among other processes. 729 */ 730 void 731 sched_clock(struct thread *td) 732 { 733 struct pcpuidlestat *stat; 734 struct td_sched *ts; 735 736 THREAD_LOCK_ASSERT(td, MA_OWNED); 737 ts = td_get_sched(td); 738 739 ts->ts_cpticks++; 740 ts->ts_estcpu = ESTCPULIM(ts->ts_estcpu + 1); 741 if ((ts->ts_estcpu % INVERSE_ESTCPU_WEIGHT) == 0) { 742 resetpriority(td); 743 resetpriority_thread(td); 744 } 745 746 /* 747 * Force a context switch if the current thread has used up a full 748 * time slice (default is 100ms). 749 */ 750 if (!TD_IS_IDLETHREAD(td) && --ts->ts_slice <= 0) { 751 ts->ts_slice = sched_slice; 752 td->td_flags |= TDF_NEEDRESCHED | TDF_SLICEEND; 753 } 754 755 stat = DPCPU_PTR(idlestat); 756 stat->oldidlecalls = stat->idlecalls; 757 stat->idlecalls = 0; 758 } 759 760 /* 761 * Charge child's scheduling CPU usage to parent. 762 */ 763 void 764 sched_exit(struct proc *p, struct thread *td) 765 { 766 767 KTR_STATE1(KTR_SCHED, "thread", sched_tdname(td), "proc exit", 768 "prio:%d", td->td_priority); 769 770 PROC_LOCK_ASSERT(p, MA_OWNED); 771 sched_exit_thread(FIRST_THREAD_IN_PROC(p), td); 772 } 773 774 void 775 sched_exit_thread(struct thread *td, struct thread *child) 776 { 777 778 KTR_STATE1(KTR_SCHED, "thread", sched_tdname(child), "exit", 779 "prio:%d", child->td_priority); 780 thread_lock(td); 781 td_get_sched(td)->ts_estcpu = ESTCPULIM(td_get_sched(td)->ts_estcpu + 782 td_get_sched(child)->ts_estcpu); 783 thread_unlock(td); 784 thread_lock(child); 785 if ((child->td_flags & TDF_NOLOAD) == 0) 786 sched_load_rem(); 787 thread_unlock(child); 788 } 789 790 void 791 sched_fork(struct thread *td, struct thread *childtd) 792 { 793 sched_fork_thread(td, childtd); 794 } 795 796 void 797 sched_fork_thread(struct thread *td, struct thread *childtd) 798 { 799 struct td_sched *ts, *tsc; 800 801 childtd->td_oncpu = NOCPU; 802 childtd->td_lastcpu = NOCPU; 803 childtd->td_lock = &sched_lock; 804 childtd->td_cpuset = cpuset_ref(td->td_cpuset); 805 childtd->td_priority = childtd->td_base_pri; 806 ts = td_get_sched(childtd); 807 bzero(ts, sizeof(*ts)); 808 tsc = td_get_sched(td); 809 ts->ts_estcpu = tsc->ts_estcpu; 810 ts->ts_flags |= (tsc->ts_flags & TSF_AFFINITY); 811 ts->ts_slice = 1; 812 } 813 814 void 815 sched_nice(struct proc *p, int nice) 816 { 817 struct thread *td; 818 819 PROC_LOCK_ASSERT(p, MA_OWNED); 820 p->p_nice = nice; 821 FOREACH_THREAD_IN_PROC(p, td) { 822 thread_lock(td); 823 resetpriority(td); 824 resetpriority_thread(td); 825 thread_unlock(td); 826 } 827 } 828 829 void 830 sched_class(struct thread *td, int class) 831 { 832 THREAD_LOCK_ASSERT(td, MA_OWNED); 833 td->td_pri_class = class; 834 } 835 836 /* 837 * Adjust the priority of a thread. 838 */ 839 static void 840 sched_priority(struct thread *td, u_char prio) 841 { 842 843 844 KTR_POINT3(KTR_SCHED, "thread", sched_tdname(td), "priority change", 845 "prio:%d", td->td_priority, "new prio:%d", prio, KTR_ATTR_LINKED, 846 sched_tdname(curthread)); 847 SDT_PROBE3(sched, , , change__pri, td, td->td_proc, prio); 848 if (td != curthread && prio > td->td_priority) { 849 KTR_POINT3(KTR_SCHED, "thread", sched_tdname(curthread), 850 "lend prio", "prio:%d", td->td_priority, "new prio:%d", 851 prio, KTR_ATTR_LINKED, sched_tdname(td)); 852 SDT_PROBE4(sched, , , lend__pri, td, td->td_proc, prio, 853 curthread); 854 } 855 THREAD_LOCK_ASSERT(td, MA_OWNED); 856 if (td->td_priority == prio) 857 return; 858 td->td_priority = prio; 859 if (TD_ON_RUNQ(td) && td->td_rqindex != (prio / RQ_PPQ)) { 860 sched_rem(td); 861 sched_add(td, SRQ_BORING); 862 } 863 } 864 865 /* 866 * Update a thread's priority when it is lent another thread's 867 * priority. 868 */ 869 void 870 sched_lend_prio(struct thread *td, u_char prio) 871 { 872 873 td->td_flags |= TDF_BORROWING; 874 sched_priority(td, prio); 875 } 876 877 /* 878 * Restore a thread's priority when priority propagation is 879 * over. The prio argument is the minimum priority the thread 880 * needs to have to satisfy other possible priority lending 881 * requests. If the thread's regulary priority is less 882 * important than prio the thread will keep a priority boost 883 * of prio. 884 */ 885 void 886 sched_unlend_prio(struct thread *td, u_char prio) 887 { 888 u_char base_pri; 889 890 if (td->td_base_pri >= PRI_MIN_TIMESHARE && 891 td->td_base_pri <= PRI_MAX_TIMESHARE) 892 base_pri = td->td_user_pri; 893 else 894 base_pri = td->td_base_pri; 895 if (prio >= base_pri) { 896 td->td_flags &= ~TDF_BORROWING; 897 sched_prio(td, base_pri); 898 } else 899 sched_lend_prio(td, prio); 900 } 901 902 void 903 sched_prio(struct thread *td, u_char prio) 904 { 905 u_char oldprio; 906 907 /* First, update the base priority. */ 908 td->td_base_pri = prio; 909 910 /* 911 * If the thread is borrowing another thread's priority, don't ever 912 * lower the priority. 913 */ 914 if (td->td_flags & TDF_BORROWING && td->td_priority < prio) 915 return; 916 917 /* Change the real priority. */ 918 oldprio = td->td_priority; 919 sched_priority(td, prio); 920 921 /* 922 * If the thread is on a turnstile, then let the turnstile update 923 * its state. 924 */ 925 if (TD_ON_LOCK(td) && oldprio != prio) 926 turnstile_adjust(td, oldprio); 927 } 928 929 void 930 sched_user_prio(struct thread *td, u_char prio) 931 { 932 933 THREAD_LOCK_ASSERT(td, MA_OWNED); 934 td->td_base_user_pri = prio; 935 if (td->td_lend_user_pri <= prio) 936 return; 937 td->td_user_pri = prio; 938 } 939 940 void 941 sched_lend_user_prio(struct thread *td, u_char prio) 942 { 943 944 THREAD_LOCK_ASSERT(td, MA_OWNED); 945 td->td_lend_user_pri = prio; 946 td->td_user_pri = min(prio, td->td_base_user_pri); 947 if (td->td_priority > td->td_user_pri) 948 sched_prio(td, td->td_user_pri); 949 else if (td->td_priority != td->td_user_pri) 950 td->td_flags |= TDF_NEEDRESCHED; 951 } 952 953 void 954 sched_sleep(struct thread *td, int pri) 955 { 956 957 THREAD_LOCK_ASSERT(td, MA_OWNED); 958 td->td_slptick = ticks; 959 td_get_sched(td)->ts_slptime = 0; 960 if (pri != 0 && PRI_BASE(td->td_pri_class) == PRI_TIMESHARE) 961 sched_prio(td, pri); 962 if (TD_IS_SUSPENDED(td) || pri >= PSOCK) 963 td->td_flags |= TDF_CANSWAP; 964 } 965 966 void 967 sched_switch(struct thread *td, struct thread *newtd, int flags) 968 { 969 struct mtx *tmtx; 970 struct td_sched *ts; 971 struct proc *p; 972 int preempted; 973 974 tmtx = NULL; 975 ts = td_get_sched(td); 976 p = td->td_proc; 977 978 THREAD_LOCK_ASSERT(td, MA_OWNED); 979 980 /* 981 * Switch to the sched lock to fix things up and pick 982 * a new thread. 983 * Block the td_lock in order to avoid breaking the critical path. 984 */ 985 if (td->td_lock != &sched_lock) { 986 mtx_lock_spin(&sched_lock); 987 tmtx = thread_lock_block(td); 988 } 989 990 if ((td->td_flags & TDF_NOLOAD) == 0) 991 sched_load_rem(); 992 993 td->td_lastcpu = td->td_oncpu; 994 preempted = !((td->td_flags & TDF_SLICEEND) || 995 (flags & SWT_RELINQUISH)); 996 td->td_flags &= ~(TDF_NEEDRESCHED | TDF_SLICEEND); 997 td->td_owepreempt = 0; 998 td->td_oncpu = NOCPU; 999 1000 /* 1001 * At the last moment, if this thread is still marked RUNNING, 1002 * then put it back on the run queue as it has not been suspended 1003 * or stopped or any thing else similar. We never put the idle 1004 * threads on the run queue, however. 1005 */ 1006 if (td->td_flags & TDF_IDLETD) { 1007 TD_SET_CAN_RUN(td); 1008 #ifdef SMP 1009 CPU_CLR(PCPU_GET(cpuid), &idle_cpus_mask); 1010 #endif 1011 } else { 1012 if (TD_IS_RUNNING(td)) { 1013 /* Put us back on the run queue. */ 1014 sched_add(td, preempted ? 1015 SRQ_OURSELF|SRQ_YIELDING|SRQ_PREEMPTED : 1016 SRQ_OURSELF|SRQ_YIELDING); 1017 } 1018 } 1019 if (newtd) { 1020 /* 1021 * The thread we are about to run needs to be counted 1022 * as if it had been added to the run queue and selected. 1023 * It came from: 1024 * * A preemption 1025 * * An upcall 1026 * * A followon 1027 */ 1028 KASSERT((newtd->td_inhibitors == 0), 1029 ("trying to run inhibited thread")); 1030 newtd->td_flags |= TDF_DIDRUN; 1031 TD_SET_RUNNING(newtd); 1032 if ((newtd->td_flags & TDF_NOLOAD) == 0) 1033 sched_load_add(); 1034 } else { 1035 newtd = choosethread(); 1036 MPASS(newtd->td_lock == &sched_lock); 1037 } 1038 1039 if (td != newtd) { 1040 #ifdef HWPMC_HOOKS 1041 if (PMC_PROC_IS_USING_PMCS(td->td_proc)) 1042 PMC_SWITCH_CONTEXT(td, PMC_FN_CSW_OUT); 1043 #endif 1044 1045 SDT_PROBE2(sched, , , off__cpu, newtd, newtd->td_proc); 1046 1047 /* I feel sleepy */ 1048 lock_profile_release_lock(&sched_lock.lock_object); 1049 #ifdef KDTRACE_HOOKS 1050 /* 1051 * If DTrace has set the active vtime enum to anything 1052 * other than INACTIVE (0), then it should have set the 1053 * function to call. 1054 */ 1055 if (dtrace_vtime_active) 1056 (*dtrace_vtime_switch_func)(newtd); 1057 #endif 1058 1059 cpu_switch(td, newtd, tmtx != NULL ? tmtx : td->td_lock); 1060 lock_profile_obtain_lock_success(&sched_lock.lock_object, 1061 0, 0, __FILE__, __LINE__); 1062 /* 1063 * Where am I? What year is it? 1064 * We are in the same thread that went to sleep above, 1065 * but any amount of time may have passed. All our context 1066 * will still be available as will local variables. 1067 * PCPU values however may have changed as we may have 1068 * changed CPU so don't trust cached values of them. 1069 * New threads will go to fork_exit() instead of here 1070 * so if you change things here you may need to change 1071 * things there too. 1072 * 1073 * If the thread above was exiting it will never wake 1074 * up again here, so either it has saved everything it 1075 * needed to, or the thread_wait() or wait() will 1076 * need to reap it. 1077 */ 1078 1079 SDT_PROBE0(sched, , , on__cpu); 1080 #ifdef HWPMC_HOOKS 1081 if (PMC_PROC_IS_USING_PMCS(td->td_proc)) 1082 PMC_SWITCH_CONTEXT(td, PMC_FN_CSW_IN); 1083 #endif 1084 } else 1085 SDT_PROBE0(sched, , , remain__cpu); 1086 1087 #ifdef SMP 1088 if (td->td_flags & TDF_IDLETD) 1089 CPU_SET(PCPU_GET(cpuid), &idle_cpus_mask); 1090 #endif 1091 sched_lock.mtx_lock = (uintptr_t)td; 1092 td->td_oncpu = PCPU_GET(cpuid); 1093 MPASS(td->td_lock == &sched_lock); 1094 } 1095 1096 void 1097 sched_wakeup(struct thread *td) 1098 { 1099 struct td_sched *ts; 1100 1101 THREAD_LOCK_ASSERT(td, MA_OWNED); 1102 ts = td_get_sched(td); 1103 td->td_flags &= ~TDF_CANSWAP; 1104 if (ts->ts_slptime > 1) { 1105 updatepri(td); 1106 resetpriority(td); 1107 } 1108 td->td_slptick = 0; 1109 ts->ts_slptime = 0; 1110 ts->ts_slice = sched_slice; 1111 sched_add(td, SRQ_BORING); 1112 } 1113 1114 #ifdef SMP 1115 static int 1116 forward_wakeup(int cpunum) 1117 { 1118 struct pcpu *pc; 1119 cpuset_t dontuse, map, map2; 1120 u_int id, me; 1121 int iscpuset; 1122 1123 mtx_assert(&sched_lock, MA_OWNED); 1124 1125 CTR0(KTR_RUNQ, "forward_wakeup()"); 1126 1127 if ((!forward_wakeup_enabled) || 1128 (forward_wakeup_use_mask == 0 && forward_wakeup_use_loop == 0)) 1129 return (0); 1130 if (!smp_started || cold || panicstr) 1131 return (0); 1132 1133 forward_wakeups_requested++; 1134 1135 /* 1136 * Check the idle mask we received against what we calculated 1137 * before in the old version. 1138 */ 1139 me = PCPU_GET(cpuid); 1140 1141 /* Don't bother if we should be doing it ourself. */ 1142 if (CPU_ISSET(me, &idle_cpus_mask) && 1143 (cpunum == NOCPU || me == cpunum)) 1144 return (0); 1145 1146 CPU_SETOF(me, &dontuse); 1147 CPU_OR(&dontuse, &stopped_cpus); 1148 CPU_OR(&dontuse, &hlt_cpus_mask); 1149 CPU_ZERO(&map2); 1150 if (forward_wakeup_use_loop) { 1151 STAILQ_FOREACH(pc, &cpuhead, pc_allcpu) { 1152 id = pc->pc_cpuid; 1153 if (!CPU_ISSET(id, &dontuse) && 1154 pc->pc_curthread == pc->pc_idlethread) { 1155 CPU_SET(id, &map2); 1156 } 1157 } 1158 } 1159 1160 if (forward_wakeup_use_mask) { 1161 map = idle_cpus_mask; 1162 CPU_NAND(&map, &dontuse); 1163 1164 /* If they are both on, compare and use loop if different. */ 1165 if (forward_wakeup_use_loop) { 1166 if (CPU_CMP(&map, &map2)) { 1167 printf("map != map2, loop method preferred\n"); 1168 map = map2; 1169 } 1170 } 1171 } else { 1172 map = map2; 1173 } 1174 1175 /* If we only allow a specific CPU, then mask off all the others. */ 1176 if (cpunum != NOCPU) { 1177 KASSERT((cpunum <= mp_maxcpus),("forward_wakeup: bad cpunum.")); 1178 iscpuset = CPU_ISSET(cpunum, &map); 1179 if (iscpuset == 0) 1180 CPU_ZERO(&map); 1181 else 1182 CPU_SETOF(cpunum, &map); 1183 } 1184 if (!CPU_EMPTY(&map)) { 1185 forward_wakeups_delivered++; 1186 STAILQ_FOREACH(pc, &cpuhead, pc_allcpu) { 1187 id = pc->pc_cpuid; 1188 if (!CPU_ISSET(id, &map)) 1189 continue; 1190 if (cpu_idle_wakeup(pc->pc_cpuid)) 1191 CPU_CLR(id, &map); 1192 } 1193 if (!CPU_EMPTY(&map)) 1194 ipi_selected(map, IPI_AST); 1195 return (1); 1196 } 1197 if (cpunum == NOCPU) 1198 printf("forward_wakeup: Idle processor not found\n"); 1199 return (0); 1200 } 1201 1202 static void 1203 kick_other_cpu(int pri, int cpuid) 1204 { 1205 struct pcpu *pcpu; 1206 int cpri; 1207 1208 pcpu = pcpu_find(cpuid); 1209 if (CPU_ISSET(cpuid, &idle_cpus_mask)) { 1210 forward_wakeups_delivered++; 1211 if (!cpu_idle_wakeup(cpuid)) 1212 ipi_cpu(cpuid, IPI_AST); 1213 return; 1214 } 1215 1216 cpri = pcpu->pc_curthread->td_priority; 1217 if (pri >= cpri) 1218 return; 1219 1220 #if defined(IPI_PREEMPTION) && defined(PREEMPTION) 1221 #if !defined(FULL_PREEMPTION) 1222 if (pri <= PRI_MAX_ITHD) 1223 #endif /* ! FULL_PREEMPTION */ 1224 { 1225 ipi_cpu(cpuid, IPI_PREEMPT); 1226 return; 1227 } 1228 #endif /* defined(IPI_PREEMPTION) && defined(PREEMPTION) */ 1229 1230 pcpu->pc_curthread->td_flags |= TDF_NEEDRESCHED; 1231 ipi_cpu(cpuid, IPI_AST); 1232 return; 1233 } 1234 #endif /* SMP */ 1235 1236 #ifdef SMP 1237 static int 1238 sched_pickcpu(struct thread *td) 1239 { 1240 int best, cpu; 1241 1242 mtx_assert(&sched_lock, MA_OWNED); 1243 1244 if (td->td_lastcpu != NOCPU && THREAD_CAN_SCHED(td, td->td_lastcpu)) 1245 best = td->td_lastcpu; 1246 else 1247 best = NOCPU; 1248 CPU_FOREACH(cpu) { 1249 if (!THREAD_CAN_SCHED(td, cpu)) 1250 continue; 1251 1252 if (best == NOCPU) 1253 best = cpu; 1254 else if (runq_length[cpu] < runq_length[best]) 1255 best = cpu; 1256 } 1257 KASSERT(best != NOCPU, ("no valid CPUs")); 1258 1259 return (best); 1260 } 1261 #endif 1262 1263 void 1264 sched_add(struct thread *td, int flags) 1265 #ifdef SMP 1266 { 1267 cpuset_t tidlemsk; 1268 struct td_sched *ts; 1269 u_int cpu, cpuid; 1270 int forwarded = 0; 1271 int single_cpu = 0; 1272 1273 ts = td_get_sched(td); 1274 THREAD_LOCK_ASSERT(td, MA_OWNED); 1275 KASSERT((td->td_inhibitors == 0), 1276 ("sched_add: trying to run inhibited thread")); 1277 KASSERT((TD_CAN_RUN(td) || TD_IS_RUNNING(td)), 1278 ("sched_add: bad thread state")); 1279 KASSERT(td->td_flags & TDF_INMEM, 1280 ("sched_add: thread swapped out")); 1281 1282 KTR_STATE2(KTR_SCHED, "thread", sched_tdname(td), "runq add", 1283 "prio:%d", td->td_priority, KTR_ATTR_LINKED, 1284 sched_tdname(curthread)); 1285 KTR_POINT1(KTR_SCHED, "thread", sched_tdname(curthread), "wokeup", 1286 KTR_ATTR_LINKED, sched_tdname(td)); 1287 SDT_PROBE4(sched, , , enqueue, td, td->td_proc, NULL, 1288 flags & SRQ_PREEMPTED); 1289 1290 1291 /* 1292 * Now that the thread is moving to the run-queue, set the lock 1293 * to the scheduler's lock. 1294 */ 1295 if (td->td_lock != &sched_lock) { 1296 mtx_lock_spin(&sched_lock); 1297 thread_lock_set(td, &sched_lock); 1298 } 1299 TD_SET_RUNQ(td); 1300 1301 /* 1302 * If SMP is started and the thread is pinned or otherwise limited to 1303 * a specific set of CPUs, queue the thread to a per-CPU run queue. 1304 * Otherwise, queue the thread to the global run queue. 1305 * 1306 * If SMP has not yet been started we must use the global run queue 1307 * as per-CPU state may not be initialized yet and we may crash if we 1308 * try to access the per-CPU run queues. 1309 */ 1310 if (smp_started && (td->td_pinned != 0 || td->td_flags & TDF_BOUND || 1311 ts->ts_flags & TSF_AFFINITY)) { 1312 if (td->td_pinned != 0) 1313 cpu = td->td_lastcpu; 1314 else if (td->td_flags & TDF_BOUND) { 1315 /* Find CPU from bound runq. */ 1316 KASSERT(SKE_RUNQ_PCPU(ts), 1317 ("sched_add: bound td_sched not on cpu runq")); 1318 cpu = ts->ts_runq - &runq_pcpu[0]; 1319 } else 1320 /* Find a valid CPU for our cpuset */ 1321 cpu = sched_pickcpu(td); 1322 ts->ts_runq = &runq_pcpu[cpu]; 1323 single_cpu = 1; 1324 CTR3(KTR_RUNQ, 1325 "sched_add: Put td_sched:%p(td:%p) on cpu%d runq", ts, td, 1326 cpu); 1327 } else { 1328 CTR2(KTR_RUNQ, 1329 "sched_add: adding td_sched:%p (td:%p) to gbl runq", ts, 1330 td); 1331 cpu = NOCPU; 1332 ts->ts_runq = &runq; 1333 } 1334 1335 cpuid = PCPU_GET(cpuid); 1336 if (single_cpu && cpu != cpuid) { 1337 kick_other_cpu(td->td_priority, cpu); 1338 } else { 1339 if (!single_cpu) { 1340 tidlemsk = idle_cpus_mask; 1341 CPU_NAND(&tidlemsk, &hlt_cpus_mask); 1342 CPU_CLR(cpuid, &tidlemsk); 1343 1344 if (!CPU_ISSET(cpuid, &idle_cpus_mask) && 1345 ((flags & SRQ_INTR) == 0) && 1346 !CPU_EMPTY(&tidlemsk)) 1347 forwarded = forward_wakeup(cpu); 1348 } 1349 1350 if (!forwarded) { 1351 if ((flags & SRQ_YIELDING) == 0 && maybe_preempt(td)) 1352 return; 1353 else 1354 maybe_resched(td); 1355 } 1356 } 1357 1358 if ((td->td_flags & TDF_NOLOAD) == 0) 1359 sched_load_add(); 1360 runq_add(ts->ts_runq, td, flags); 1361 if (cpu != NOCPU) 1362 runq_length[cpu]++; 1363 } 1364 #else /* SMP */ 1365 { 1366 struct td_sched *ts; 1367 1368 ts = td_get_sched(td); 1369 THREAD_LOCK_ASSERT(td, MA_OWNED); 1370 KASSERT((td->td_inhibitors == 0), 1371 ("sched_add: trying to run inhibited thread")); 1372 KASSERT((TD_CAN_RUN(td) || TD_IS_RUNNING(td)), 1373 ("sched_add: bad thread state")); 1374 KASSERT(td->td_flags & TDF_INMEM, 1375 ("sched_add: thread swapped out")); 1376 KTR_STATE2(KTR_SCHED, "thread", sched_tdname(td), "runq add", 1377 "prio:%d", td->td_priority, KTR_ATTR_LINKED, 1378 sched_tdname(curthread)); 1379 KTR_POINT1(KTR_SCHED, "thread", sched_tdname(curthread), "wokeup", 1380 KTR_ATTR_LINKED, sched_tdname(td)); 1381 SDT_PROBE4(sched, , , enqueue, td, td->td_proc, NULL, 1382 flags & SRQ_PREEMPTED); 1383 1384 /* 1385 * Now that the thread is moving to the run-queue, set the lock 1386 * to the scheduler's lock. 1387 */ 1388 if (td->td_lock != &sched_lock) { 1389 mtx_lock_spin(&sched_lock); 1390 thread_lock_set(td, &sched_lock); 1391 } 1392 TD_SET_RUNQ(td); 1393 CTR2(KTR_RUNQ, "sched_add: adding td_sched:%p (td:%p) to runq", ts, td); 1394 ts->ts_runq = &runq; 1395 1396 /* 1397 * If we are yielding (on the way out anyhow) or the thread 1398 * being saved is US, then don't try be smart about preemption 1399 * or kicking off another CPU as it won't help and may hinder. 1400 * In the YIEDLING case, we are about to run whoever is being 1401 * put in the queue anyhow, and in the OURSELF case, we are 1402 * putting ourself on the run queue which also only happens 1403 * when we are about to yield. 1404 */ 1405 if ((flags & SRQ_YIELDING) == 0) { 1406 if (maybe_preempt(td)) 1407 return; 1408 } 1409 if ((td->td_flags & TDF_NOLOAD) == 0) 1410 sched_load_add(); 1411 runq_add(ts->ts_runq, td, flags); 1412 maybe_resched(td); 1413 } 1414 #endif /* SMP */ 1415 1416 void 1417 sched_rem(struct thread *td) 1418 { 1419 struct td_sched *ts; 1420 1421 ts = td_get_sched(td); 1422 KASSERT(td->td_flags & TDF_INMEM, 1423 ("sched_rem: thread swapped out")); 1424 KASSERT(TD_ON_RUNQ(td), 1425 ("sched_rem: thread not on run queue")); 1426 mtx_assert(&sched_lock, MA_OWNED); 1427 KTR_STATE2(KTR_SCHED, "thread", sched_tdname(td), "runq rem", 1428 "prio:%d", td->td_priority, KTR_ATTR_LINKED, 1429 sched_tdname(curthread)); 1430 SDT_PROBE3(sched, , , dequeue, td, td->td_proc, NULL); 1431 1432 if ((td->td_flags & TDF_NOLOAD) == 0) 1433 sched_load_rem(); 1434 #ifdef SMP 1435 if (ts->ts_runq != &runq) 1436 runq_length[ts->ts_runq - runq_pcpu]--; 1437 #endif 1438 runq_remove(ts->ts_runq, td); 1439 TD_SET_CAN_RUN(td); 1440 } 1441 1442 /* 1443 * Select threads to run. Note that running threads still consume a 1444 * slot. 1445 */ 1446 struct thread * 1447 sched_choose(void) 1448 { 1449 struct thread *td; 1450 struct runq *rq; 1451 1452 mtx_assert(&sched_lock, MA_OWNED); 1453 #ifdef SMP 1454 struct thread *tdcpu; 1455 1456 rq = &runq; 1457 td = runq_choose_fuzz(&runq, runq_fuzz); 1458 tdcpu = runq_choose(&runq_pcpu[PCPU_GET(cpuid)]); 1459 1460 if (td == NULL || 1461 (tdcpu != NULL && 1462 tdcpu->td_priority < td->td_priority)) { 1463 CTR2(KTR_RUNQ, "choosing td %p from pcpu runq %d", tdcpu, 1464 PCPU_GET(cpuid)); 1465 td = tdcpu; 1466 rq = &runq_pcpu[PCPU_GET(cpuid)]; 1467 } else { 1468 CTR1(KTR_RUNQ, "choosing td_sched %p from main runq", td); 1469 } 1470 1471 #else 1472 rq = &runq; 1473 td = runq_choose(&runq); 1474 #endif 1475 1476 if (td) { 1477 #ifdef SMP 1478 if (td == tdcpu) 1479 runq_length[PCPU_GET(cpuid)]--; 1480 #endif 1481 runq_remove(rq, td); 1482 td->td_flags |= TDF_DIDRUN; 1483 1484 KASSERT(td->td_flags & TDF_INMEM, 1485 ("sched_choose: thread swapped out")); 1486 return (td); 1487 } 1488 return (PCPU_GET(idlethread)); 1489 } 1490 1491 void 1492 sched_preempt(struct thread *td) 1493 { 1494 1495 SDT_PROBE2(sched, , , surrender, td, td->td_proc); 1496 thread_lock(td); 1497 if (td->td_critnest > 1) 1498 td->td_owepreempt = 1; 1499 else 1500 mi_switch(SW_INVOL | SW_PREEMPT | SWT_PREEMPT, NULL); 1501 thread_unlock(td); 1502 } 1503 1504 void 1505 sched_userret(struct thread *td) 1506 { 1507 /* 1508 * XXX we cheat slightly on the locking here to avoid locking in 1509 * the usual case. Setting td_priority here is essentially an 1510 * incomplete workaround for not setting it properly elsewhere. 1511 * Now that some interrupt handlers are threads, not setting it 1512 * properly elsewhere can clobber it in the window between setting 1513 * it here and returning to user mode, so don't waste time setting 1514 * it perfectly here. 1515 */ 1516 KASSERT((td->td_flags & TDF_BORROWING) == 0, 1517 ("thread with borrowed priority returning to userland")); 1518 if (td->td_priority != td->td_user_pri) { 1519 thread_lock(td); 1520 td->td_priority = td->td_user_pri; 1521 td->td_base_pri = td->td_user_pri; 1522 thread_unlock(td); 1523 } 1524 } 1525 1526 void 1527 sched_bind(struct thread *td, int cpu) 1528 { 1529 struct td_sched *ts; 1530 1531 THREAD_LOCK_ASSERT(td, MA_OWNED|MA_NOTRECURSED); 1532 KASSERT(td == curthread, ("sched_bind: can only bind curthread")); 1533 1534 ts = td_get_sched(td); 1535 1536 td->td_flags |= TDF_BOUND; 1537 #ifdef SMP 1538 ts->ts_runq = &runq_pcpu[cpu]; 1539 if (PCPU_GET(cpuid) == cpu) 1540 return; 1541 1542 mi_switch(SW_VOL, NULL); 1543 #endif 1544 } 1545 1546 void 1547 sched_unbind(struct thread* td) 1548 { 1549 THREAD_LOCK_ASSERT(td, MA_OWNED); 1550 KASSERT(td == curthread, ("sched_unbind: can only bind curthread")); 1551 td->td_flags &= ~TDF_BOUND; 1552 } 1553 1554 int 1555 sched_is_bound(struct thread *td) 1556 { 1557 THREAD_LOCK_ASSERT(td, MA_OWNED); 1558 return (td->td_flags & TDF_BOUND); 1559 } 1560 1561 void 1562 sched_relinquish(struct thread *td) 1563 { 1564 thread_lock(td); 1565 mi_switch(SW_VOL | SWT_RELINQUISH, NULL); 1566 thread_unlock(td); 1567 } 1568 1569 int 1570 sched_load(void) 1571 { 1572 return (sched_tdcnt); 1573 } 1574 1575 int 1576 sched_sizeof_proc(void) 1577 { 1578 return (sizeof(struct proc)); 1579 } 1580 1581 int 1582 sched_sizeof_thread(void) 1583 { 1584 return (sizeof(struct thread) + sizeof(struct td_sched)); 1585 } 1586 1587 fixpt_t 1588 sched_pctcpu(struct thread *td) 1589 { 1590 struct td_sched *ts; 1591 1592 THREAD_LOCK_ASSERT(td, MA_OWNED); 1593 ts = td_get_sched(td); 1594 return (ts->ts_pctcpu); 1595 } 1596 1597 #ifdef RACCT 1598 /* 1599 * Calculates the contribution to the thread cpu usage for the latest 1600 * (unfinished) second. 1601 */ 1602 fixpt_t 1603 sched_pctcpu_delta(struct thread *td) 1604 { 1605 struct td_sched *ts; 1606 fixpt_t delta; 1607 int realstathz; 1608 1609 THREAD_LOCK_ASSERT(td, MA_OWNED); 1610 ts = td_get_sched(td); 1611 delta = 0; 1612 realstathz = stathz ? stathz : hz; 1613 if (ts->ts_cpticks != 0) { 1614 #if (FSHIFT >= CCPU_SHIFT) 1615 delta = (realstathz == 100) 1616 ? ((fixpt_t) ts->ts_cpticks) << 1617 (FSHIFT - CCPU_SHIFT) : 1618 100 * (((fixpt_t) ts->ts_cpticks) 1619 << (FSHIFT - CCPU_SHIFT)) / realstathz; 1620 #else 1621 delta = ((FSCALE - ccpu) * 1622 (ts->ts_cpticks * 1623 FSCALE / realstathz)) >> FSHIFT; 1624 #endif 1625 } 1626 1627 return (delta); 1628 } 1629 #endif 1630 1631 u_int 1632 sched_estcpu(struct thread *td) 1633 { 1634 1635 return (td_get_sched(td)->ts_estcpu); 1636 } 1637 1638 /* 1639 * The actual idle process. 1640 */ 1641 void 1642 sched_idletd(void *dummy) 1643 { 1644 struct pcpuidlestat *stat; 1645 1646 THREAD_NO_SLEEPING(); 1647 stat = DPCPU_PTR(idlestat); 1648 for (;;) { 1649 mtx_assert(&Giant, MA_NOTOWNED); 1650 1651 while (sched_runnable() == 0) { 1652 cpu_idle(stat->idlecalls + stat->oldidlecalls > 64); 1653 stat->idlecalls++; 1654 } 1655 1656 mtx_lock_spin(&sched_lock); 1657 mi_switch(SW_VOL | SWT_IDLE, NULL); 1658 mtx_unlock_spin(&sched_lock); 1659 } 1660 } 1661 1662 /* 1663 * A CPU is entering for the first time or a thread is exiting. 1664 */ 1665 void 1666 sched_throw(struct thread *td) 1667 { 1668 /* 1669 * Correct spinlock nesting. The idle thread context that we are 1670 * borrowing was created so that it would start out with a single 1671 * spin lock (sched_lock) held in fork_trampoline(). Since we've 1672 * explicitly acquired locks in this function, the nesting count 1673 * is now 2 rather than 1. Since we are nested, calling 1674 * spinlock_exit() will simply adjust the counts without allowing 1675 * spin lock using code to interrupt us. 1676 */ 1677 if (td == NULL) { 1678 mtx_lock_spin(&sched_lock); 1679 spinlock_exit(); 1680 PCPU_SET(switchtime, cpu_ticks()); 1681 PCPU_SET(switchticks, ticks); 1682 } else { 1683 lock_profile_release_lock(&sched_lock.lock_object); 1684 MPASS(td->td_lock == &sched_lock); 1685 td->td_lastcpu = td->td_oncpu; 1686 td->td_oncpu = NOCPU; 1687 } 1688 mtx_assert(&sched_lock, MA_OWNED); 1689 KASSERT(curthread->td_md.md_spinlock_count == 1, ("invalid count")); 1690 cpu_throw(td, choosethread()); /* doesn't return */ 1691 } 1692 1693 void 1694 sched_fork_exit(struct thread *td) 1695 { 1696 1697 /* 1698 * Finish setting up thread glue so that it begins execution in a 1699 * non-nested critical section with sched_lock held but not recursed. 1700 */ 1701 td->td_oncpu = PCPU_GET(cpuid); 1702 sched_lock.mtx_lock = (uintptr_t)td; 1703 lock_profile_obtain_lock_success(&sched_lock.lock_object, 1704 0, 0, __FILE__, __LINE__); 1705 THREAD_LOCK_ASSERT(td, MA_OWNED | MA_NOTRECURSED); 1706 } 1707 1708 char * 1709 sched_tdname(struct thread *td) 1710 { 1711 #ifdef KTR 1712 struct td_sched *ts; 1713 1714 ts = td_get_sched(td); 1715 if (ts->ts_name[0] == '\0') 1716 snprintf(ts->ts_name, sizeof(ts->ts_name), 1717 "%s tid %d", td->td_name, td->td_tid); 1718 return (ts->ts_name); 1719 #else 1720 return (td->td_name); 1721 #endif 1722 } 1723 1724 #ifdef KTR 1725 void 1726 sched_clear_tdname(struct thread *td) 1727 { 1728 struct td_sched *ts; 1729 1730 ts = td_get_sched(td); 1731 ts->ts_name[0] = '\0'; 1732 } 1733 #endif 1734 1735 void 1736 sched_affinity(struct thread *td) 1737 { 1738 #ifdef SMP 1739 struct td_sched *ts; 1740 int cpu; 1741 1742 THREAD_LOCK_ASSERT(td, MA_OWNED); 1743 1744 /* 1745 * Set the TSF_AFFINITY flag if there is at least one CPU this 1746 * thread can't run on. 1747 */ 1748 ts = td_get_sched(td); 1749 ts->ts_flags &= ~TSF_AFFINITY; 1750 CPU_FOREACH(cpu) { 1751 if (!THREAD_CAN_SCHED(td, cpu)) { 1752 ts->ts_flags |= TSF_AFFINITY; 1753 break; 1754 } 1755 } 1756 1757 /* 1758 * If this thread can run on all CPUs, nothing else to do. 1759 */ 1760 if (!(ts->ts_flags & TSF_AFFINITY)) 1761 return; 1762 1763 /* Pinned threads and bound threads should be left alone. */ 1764 if (td->td_pinned != 0 || td->td_flags & TDF_BOUND) 1765 return; 1766 1767 switch (td->td_state) { 1768 case TDS_RUNQ: 1769 /* 1770 * If we are on a per-CPU runqueue that is in the set, 1771 * then nothing needs to be done. 1772 */ 1773 if (ts->ts_runq != &runq && 1774 THREAD_CAN_SCHED(td, ts->ts_runq - runq_pcpu)) 1775 return; 1776 1777 /* Put this thread on a valid per-CPU runqueue. */ 1778 sched_rem(td); 1779 sched_add(td, SRQ_BORING); 1780 break; 1781 case TDS_RUNNING: 1782 /* 1783 * See if our current CPU is in the set. If not, force a 1784 * context switch. 1785 */ 1786 if (THREAD_CAN_SCHED(td, td->td_oncpu)) 1787 return; 1788 1789 td->td_flags |= TDF_NEEDRESCHED; 1790 if (td != curthread) 1791 ipi_cpu(cpu, IPI_AST); 1792 break; 1793 default: 1794 break; 1795 } 1796 #endif 1797 } 1798