1 /*- 2 * Copyright (c) 2001 3 * John Baldwin <jhb@FreeBSD.org>. 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 * 1. Redistributions of source code must retain the above copyright 9 * notice, this list of conditions and the following disclaimer. 10 * 2. Redistributions in binary form must reproduce the above copyright 11 * notice, this list of conditions and the following disclaimer in the 12 * documentation and/or other materials provided with the distribution. 13 * 4. Neither the name of the author nor the names of any co-contributors 14 * may be used to endorse or promote products derived from this software 15 * without specific prior written permission. 16 * 17 * THIS SOFTWARE IS PROVIDED BY JOHN BALDWIN AND CONTRIBUTORS ``AS IS'' AND 18 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE 19 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE 20 * ARE DISCLAIMED. IN NO EVENT SHALL JOHN BALDWIN OR THE VOICES IN HIS HEAD 21 * BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR 22 * CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF 23 * SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS 24 * INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN 25 * CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) 26 * ARISING IN ANY WAY OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF 27 * THE POSSIBILITY OF SUCH DAMAGE. 28 */ 29 30 /* 31 * This module holds the global variables and machine independent functions 32 * used for the kernel SMP support. 33 */ 34 35 #include <sys/cdefs.h> 36 __FBSDID("$FreeBSD$"); 37 38 #include <sys/param.h> 39 #include <sys/systm.h> 40 #include <sys/kernel.h> 41 #include <sys/ktr.h> 42 #include <sys/proc.h> 43 #include <sys/bus.h> 44 #include <sys/lock.h> 45 #include <sys/mutex.h> 46 #include <sys/pcpu.h> 47 #include <sys/smp.h> 48 #include <sys/sysctl.h> 49 50 #include <machine/cpu.h> 51 #include <machine/smp.h> 52 53 #include "opt_sched.h" 54 55 #ifdef SMP 56 volatile cpumask_t stopped_cpus; 57 volatile cpumask_t started_cpus; 58 cpumask_t idle_cpus_mask; 59 cpumask_t hlt_cpus_mask; 60 cpumask_t logical_cpus_mask; 61 62 void (*cpustop_restartfunc)(void); 63 #endif 64 /* This is used in modules that need to work in both SMP and UP. */ 65 cpumask_t all_cpus; 66 67 int mp_ncpus; 68 /* export this for libkvm consumers. */ 69 int mp_maxcpus = MAXCPU; 70 71 volatile int smp_started; 72 u_int mp_maxid; 73 74 SYSCTL_NODE(_kern, OID_AUTO, smp, CTLFLAG_RD, NULL, "Kernel SMP"); 75 76 SYSCTL_INT(_kern_smp, OID_AUTO, maxid, CTLFLAG_RD, &mp_maxid, 0, 77 "Max CPU ID."); 78 79 SYSCTL_INT(_kern_smp, OID_AUTO, maxcpus, CTLFLAG_RD, &mp_maxcpus, 0, 80 "Max number of CPUs that the system was compiled for."); 81 82 int smp_active = 0; /* are the APs allowed to run? */ 83 SYSCTL_INT(_kern_smp, OID_AUTO, active, CTLFLAG_RW, &smp_active, 0, 84 "Number of Auxillary Processors (APs) that were successfully started"); 85 86 int smp_disabled = 0; /* has smp been disabled? */ 87 SYSCTL_INT(_kern_smp, OID_AUTO, disabled, CTLFLAG_RDTUN, &smp_disabled, 0, 88 "SMP has been disabled from the loader"); 89 TUNABLE_INT("kern.smp.disabled", &smp_disabled); 90 91 int smp_cpus = 1; /* how many cpu's running */ 92 SYSCTL_INT(_kern_smp, OID_AUTO, cpus, CTLFLAG_RD, &smp_cpus, 0, 93 "Number of CPUs online"); 94 95 int smp_topology = 0; /* Which topology we're using. */ 96 SYSCTL_INT(_kern_smp, OID_AUTO, topology, CTLFLAG_RD, &smp_topology, 0, 97 "Topology override setting; 0 is default provided by hardware."); 98 TUNABLE_INT("kern.smp.topology", &smp_topology); 99 100 #ifdef SMP 101 /* Enable forwarding of a signal to a process running on a different CPU */ 102 static int forward_signal_enabled = 1; 103 SYSCTL_INT(_kern_smp, OID_AUTO, forward_signal_enabled, CTLFLAG_RW, 104 &forward_signal_enabled, 0, 105 "Forwarding of a signal to a process on a different CPU"); 106 107 /* Enable forwarding of roundrobin to all other cpus */ 108 static int forward_roundrobin_enabled = 1; 109 SYSCTL_INT(_kern_smp, OID_AUTO, forward_roundrobin_enabled, CTLFLAG_RW, 110 &forward_roundrobin_enabled, 0, 111 "Forwarding of roundrobin to all other CPUs"); 112 113 /* Variables needed for SMP rendezvous. */ 114 static volatile cpumask_t smp_rv_cpumask; 115 static void (*volatile smp_rv_setup_func)(void *arg); 116 static void (*volatile smp_rv_action_func)(void *arg); 117 static void (*volatile smp_rv_teardown_func)(void *arg); 118 static void * volatile smp_rv_func_arg; 119 static volatile int smp_rv_waiters[3]; 120 121 /* 122 * Shared mutex to restrict busywaits between smp_rendezvous() and 123 * smp(_targeted)_tlb_shootdown(). A deadlock occurs if both of these 124 * functions trigger at once and cause multiple CPUs to busywait with 125 * interrupts disabled. 126 */ 127 struct mtx smp_ipi_mtx; 128 129 /* 130 * Let the MD SMP code initialize mp_maxid very early if it can. 131 */ 132 static void 133 mp_setmaxid(void *dummy) 134 { 135 cpu_mp_setmaxid(); 136 } 137 SYSINIT(cpu_mp_setmaxid, SI_SUB_TUNABLES, SI_ORDER_FIRST, mp_setmaxid, NULL); 138 139 /* 140 * Call the MD SMP initialization code. 141 */ 142 static void 143 mp_start(void *dummy) 144 { 145 146 /* Probe for MP hardware. */ 147 if (smp_disabled != 0 || cpu_mp_probe() == 0) { 148 mp_ncpus = 1; 149 all_cpus = PCPU_GET(cpumask); 150 return; 151 } 152 153 mtx_init(&smp_ipi_mtx, "smp rendezvous", NULL, MTX_SPIN); 154 cpu_mp_start(); 155 printf("FreeBSD/SMP: Multiprocessor System Detected: %d CPUs\n", 156 mp_ncpus); 157 cpu_mp_announce(); 158 } 159 SYSINIT(cpu_mp, SI_SUB_CPU, SI_ORDER_THIRD, mp_start, NULL); 160 161 void 162 forward_signal(struct thread *td) 163 { 164 int id; 165 166 /* 167 * signotify() has already set TDF_ASTPENDING and TDF_NEEDSIGCHECK on 168 * this thread, so all we need to do is poke it if it is currently 169 * executing so that it executes ast(). 170 */ 171 THREAD_LOCK_ASSERT(td, MA_OWNED); 172 KASSERT(TD_IS_RUNNING(td), 173 ("forward_signal: thread is not TDS_RUNNING")); 174 175 CTR1(KTR_SMP, "forward_signal(%p)", td->td_proc); 176 177 if (!smp_started || cold || panicstr) 178 return; 179 if (!forward_signal_enabled) 180 return; 181 182 /* No need to IPI ourself. */ 183 if (td == curthread) 184 return; 185 186 id = td->td_oncpu; 187 if (id == NOCPU) 188 return; 189 ipi_selected(1 << id, IPI_AST); 190 } 191 192 void 193 forward_roundrobin(void) 194 { 195 struct pcpu *pc; 196 struct thread *td; 197 cpumask_t id, map, me; 198 199 CTR0(KTR_SMP, "forward_roundrobin()"); 200 201 if (!smp_started || cold || panicstr) 202 return; 203 if (!forward_roundrobin_enabled) 204 return; 205 map = 0; 206 me = PCPU_GET(cpumask); 207 SLIST_FOREACH(pc, &cpuhead, pc_allcpu) { 208 td = pc->pc_curthread; 209 id = pc->pc_cpumask; 210 if (id != me && (id & stopped_cpus) == 0 && 211 !TD_IS_IDLETHREAD(td)) { 212 td->td_flags |= TDF_NEEDRESCHED; 213 map |= id; 214 } 215 } 216 ipi_selected(map, IPI_AST); 217 } 218 219 /* 220 * When called the executing CPU will send an IPI to all other CPUs 221 * requesting that they halt execution. 222 * 223 * Usually (but not necessarily) called with 'other_cpus' as its arg. 224 * 225 * - Signals all CPUs in map to stop. 226 * - Waits for each to stop. 227 * 228 * Returns: 229 * -1: error 230 * 0: NA 231 * 1: ok 232 * 233 * XXX FIXME: this is not MP-safe, needs a lock to prevent multiple CPUs 234 * from executing at same time. 235 */ 236 int 237 stop_cpus(cpumask_t map) 238 { 239 int i; 240 241 if (!smp_started) 242 return 0; 243 244 CTR1(KTR_SMP, "stop_cpus(%x)", map); 245 246 /* send the stop IPI to all CPUs in map */ 247 ipi_selected(map, IPI_STOP); 248 249 i = 0; 250 while ((stopped_cpus & map) != map) { 251 /* spin */ 252 cpu_spinwait(); 253 i++; 254 #ifdef DIAGNOSTIC 255 if (i == 100000) { 256 printf("timeout stopping cpus\n"); 257 break; 258 } 259 #endif 260 } 261 262 return 1; 263 } 264 265 /* 266 * Called by a CPU to restart stopped CPUs. 267 * 268 * Usually (but not necessarily) called with 'stopped_cpus' as its arg. 269 * 270 * - Signals all CPUs in map to restart. 271 * - Waits for each to restart. 272 * 273 * Returns: 274 * -1: error 275 * 0: NA 276 * 1: ok 277 */ 278 int 279 restart_cpus(cpumask_t map) 280 { 281 282 if (!smp_started) 283 return 0; 284 285 CTR1(KTR_SMP, "restart_cpus(%x)", map); 286 287 /* signal other cpus to restart */ 288 atomic_store_rel_int(&started_cpus, map); 289 290 /* wait for each to clear its bit */ 291 while ((stopped_cpus & map) != 0) 292 cpu_spinwait(); 293 294 return 1; 295 } 296 297 /* 298 * All-CPU rendezvous. CPUs are signalled, all execute the setup function 299 * (if specified), rendezvous, execute the action function (if specified), 300 * rendezvous again, execute the teardown function (if specified), and then 301 * resume. 302 * 303 * Note that the supplied external functions _must_ be reentrant and aware 304 * that they are running in parallel and in an unknown lock context. 305 */ 306 void 307 smp_rendezvous_action(void) 308 { 309 cpumask_t map = smp_rv_cpumask; 310 int i, ncpus = 0; 311 void* local_func_arg = smp_rv_func_arg; 312 void (*local_setup_func)(void*) = smp_rv_setup_func; 313 void (*local_action_func)(void*) = smp_rv_action_func; 314 void (*local_teardown_func)(void*) = smp_rv_teardown_func; 315 316 for (i = 0; i < MAXCPU; i++) 317 if (((1 << i) & map) != 0 && pcpu_find(i) != NULL) 318 ncpus++; 319 320 /* Ensure we have up-to-date values. */ 321 atomic_add_acq_int(&smp_rv_waiters[0], 1); 322 while (smp_rv_waiters[0] < ncpus) 323 cpu_spinwait(); 324 325 /* setup function */ 326 if (local_setup_func != smp_no_rendevous_barrier) { 327 if (smp_rv_setup_func != NULL) 328 smp_rv_setup_func(smp_rv_func_arg); 329 330 /* spin on entry rendezvous */ 331 atomic_add_int(&smp_rv_waiters[1], 1); 332 while (smp_rv_waiters[1] < ncpus) 333 cpu_spinwait(); 334 } 335 336 /* action function */ 337 if (local_action_func != NULL) 338 local_action_func(local_func_arg); 339 340 /* spin on exit rendezvous */ 341 atomic_add_int(&smp_rv_waiters[2], 1); 342 if (local_teardown_func == smp_no_rendevous_barrier) 343 return; 344 while (smp_rv_waiters[2] < ncpus) 345 cpu_spinwait(); 346 347 /* teardown function */ 348 if (local_teardown_func != NULL) 349 local_teardown_func(local_func_arg); 350 } 351 352 void 353 smp_rendezvous_cpus(cpumask_t map, 354 void (* setup_func)(void *), 355 void (* action_func)(void *), 356 void (* teardown_func)(void *), 357 void *arg) 358 { 359 int i, ncpus = 0; 360 361 if (!smp_started) { 362 if (setup_func != NULL) 363 setup_func(arg); 364 if (action_func != NULL) 365 action_func(arg); 366 if (teardown_func != NULL) 367 teardown_func(arg); 368 return; 369 } 370 371 for (i = 0; i < MAXCPU; i++) 372 if (((1 << i) & map) != 0 && pcpu_find(i) != NULL) 373 ncpus++; 374 375 /* obtain rendezvous lock */ 376 mtx_lock_spin(&smp_ipi_mtx); 377 378 /* set static function pointers */ 379 smp_rv_cpumask = map; 380 smp_rv_setup_func = setup_func; 381 smp_rv_action_func = action_func; 382 smp_rv_teardown_func = teardown_func; 383 smp_rv_func_arg = arg; 384 smp_rv_waiters[1] = 0; 385 smp_rv_waiters[2] = 0; 386 atomic_store_rel_int(&smp_rv_waiters[0], 0); 387 388 /* signal other processors, which will enter the IPI with interrupts off */ 389 ipi_selected(map & ~(1 << curcpu), IPI_RENDEZVOUS); 390 391 /* Check if the current CPU is in the map */ 392 if ((map & (1 << curcpu)) != 0) 393 smp_rendezvous_action(); 394 395 if (teardown_func == smp_no_rendevous_barrier) 396 while (atomic_load_acq_int(&smp_rv_waiters[2]) < ncpus) 397 cpu_spinwait(); 398 399 /* release lock */ 400 mtx_unlock_spin(&smp_ipi_mtx); 401 } 402 403 void 404 smp_rendezvous(void (* setup_func)(void *), 405 void (* action_func)(void *), 406 void (* teardown_func)(void *), 407 void *arg) 408 { 409 smp_rendezvous_cpus(all_cpus, setup_func, action_func, teardown_func, arg); 410 } 411 412 static struct cpu_group group[MAXCPU]; 413 414 struct cpu_group * 415 smp_topo(void) 416 { 417 struct cpu_group *top; 418 419 /* 420 * Check for a fake topology request for debugging purposes. 421 */ 422 switch (smp_topology) { 423 case 1: 424 /* Dual core with no sharing. */ 425 top = smp_topo_1level(CG_SHARE_NONE, 2, 0); 426 break; 427 case 2: 428 /* No topology, all cpus are equal. */ 429 top = smp_topo_none(); 430 break; 431 case 3: 432 /* Dual core with shared L2. */ 433 top = smp_topo_1level(CG_SHARE_L2, 2, 0); 434 break; 435 case 4: 436 /* quad core, shared l3 among each package, private l2. */ 437 top = smp_topo_1level(CG_SHARE_L3, 4, 0); 438 break; 439 case 5: 440 /* quad core, 2 dualcore parts on each package share l2. */ 441 top = smp_topo_2level(CG_SHARE_NONE, 2, CG_SHARE_L2, 2, 0); 442 break; 443 case 6: 444 /* Single-core 2xHTT */ 445 top = smp_topo_1level(CG_SHARE_L1, 2, CG_FLAG_HTT); 446 break; 447 case 7: 448 /* quad core with a shared l3, 8 threads sharing L2. */ 449 top = smp_topo_2level(CG_SHARE_L3, 4, CG_SHARE_L2, 8, 450 CG_FLAG_THREAD); 451 break; 452 default: 453 /* Default, ask the system what it wants. */ 454 top = cpu_topo(); 455 break; 456 } 457 /* 458 * Verify the returned topology. 459 */ 460 if (top->cg_count != mp_ncpus) 461 panic("Built bad topology at %p. CPU count %d != %d", 462 top, top->cg_count, mp_ncpus); 463 if (top->cg_mask != all_cpus) 464 panic("Built bad topology at %p. CPU mask 0x%X != 0x%X", 465 top, top->cg_mask, all_cpus); 466 return (top); 467 } 468 469 struct cpu_group * 470 smp_topo_none(void) 471 { 472 struct cpu_group *top; 473 474 top = &group[0]; 475 top->cg_parent = NULL; 476 top->cg_child = NULL; 477 top->cg_mask = (1 << mp_ncpus) - 1; 478 top->cg_count = mp_ncpus; 479 top->cg_children = 0; 480 top->cg_level = CG_SHARE_NONE; 481 top->cg_flags = 0; 482 483 return (top); 484 } 485 486 static int 487 smp_topo_addleaf(struct cpu_group *parent, struct cpu_group *child, int share, 488 int count, int flags, int start) 489 { 490 cpumask_t mask; 491 int i; 492 493 for (mask = 0, i = 0; i < count; i++, start++) 494 mask |= (1 << start); 495 child->cg_parent = parent; 496 child->cg_child = NULL; 497 child->cg_children = 0; 498 child->cg_level = share; 499 child->cg_count = count; 500 child->cg_flags = flags; 501 child->cg_mask = mask; 502 parent->cg_children++; 503 for (; parent != NULL; parent = parent->cg_parent) { 504 if ((parent->cg_mask & child->cg_mask) != 0) 505 panic("Duplicate children in %p. mask 0x%X child 0x%X", 506 parent, parent->cg_mask, child->cg_mask); 507 parent->cg_mask |= child->cg_mask; 508 parent->cg_count += child->cg_count; 509 } 510 511 return (start); 512 } 513 514 struct cpu_group * 515 smp_topo_1level(int share, int count, int flags) 516 { 517 struct cpu_group *child; 518 struct cpu_group *top; 519 int packages; 520 int cpu; 521 int i; 522 523 cpu = 0; 524 top = &group[0]; 525 packages = mp_ncpus / count; 526 top->cg_child = child = &group[1]; 527 top->cg_level = CG_SHARE_NONE; 528 for (i = 0; i < packages; i++, child++) 529 cpu = smp_topo_addleaf(top, child, share, count, flags, cpu); 530 return (top); 531 } 532 533 struct cpu_group * 534 smp_topo_2level(int l2share, int l2count, int l1share, int l1count, 535 int l1flags) 536 { 537 struct cpu_group *top; 538 struct cpu_group *l1g; 539 struct cpu_group *l2g; 540 int cpu; 541 int i; 542 int j; 543 544 cpu = 0; 545 top = &group[0]; 546 l2g = &group[1]; 547 top->cg_child = l2g; 548 top->cg_level = CG_SHARE_NONE; 549 top->cg_children = mp_ncpus / (l2count * l1count); 550 l1g = l2g + top->cg_children; 551 for (i = 0; i < top->cg_children; i++, l2g++) { 552 l2g->cg_parent = top; 553 l2g->cg_child = l1g; 554 l2g->cg_level = l2share; 555 for (j = 0; j < l2count; j++, l1g++) 556 cpu = smp_topo_addleaf(l2g, l1g, l1share, l1count, 557 l1flags, cpu); 558 } 559 return (top); 560 } 561 562 563 struct cpu_group * 564 smp_topo_find(struct cpu_group *top, int cpu) 565 { 566 struct cpu_group *cg; 567 cpumask_t mask; 568 int children; 569 int i; 570 571 mask = (1 << cpu); 572 cg = top; 573 for (;;) { 574 if ((cg->cg_mask & mask) == 0) 575 return (NULL); 576 if (cg->cg_children == 0) 577 return (cg); 578 children = cg->cg_children; 579 for (i = 0, cg = cg->cg_child; i < children; cg++, i++) 580 if ((cg->cg_mask & mask) != 0) 581 break; 582 } 583 return (NULL); 584 } 585 #else /* !SMP */ 586 587 void 588 smp_rendezvous_cpus(cpumask_t map, 589 void (*setup_func)(void *), 590 void (*action_func)(void *), 591 void (*teardown_func)(void *), 592 void *arg) 593 { 594 if (setup_func != NULL) 595 setup_func(arg); 596 if (action_func != NULL) 597 action_func(arg); 598 if (teardown_func != NULL) 599 teardown_func(arg); 600 } 601 602 void 603 smp_rendezvous(void (*setup_func)(void *), 604 void (*action_func)(void *), 605 void (*teardown_func)(void *), 606 void *arg) 607 { 608 609 if (setup_func != NULL) 610 setup_func(arg); 611 if (action_func != NULL) 612 action_func(arg); 613 if (teardown_func != NULL) 614 teardown_func(arg); 615 } 616 617 /* 618 * Provide dummy SMP support for UP kernels. Modules that need to use SMP 619 * APIs will still work using this dummy support. 620 */ 621 static void 622 mp_setvariables_for_up(void *dummy) 623 { 624 mp_ncpus = 1; 625 mp_maxid = PCPU_GET(cpuid); 626 all_cpus = PCPU_GET(cpumask); 627 KASSERT(PCPU_GET(cpuid) == 0, ("UP must have a CPU ID of zero")); 628 } 629 SYSINIT(cpu_mp_setvariables, SI_SUB_TUNABLES, SI_ORDER_FIRST, 630 mp_setvariables_for_up, NULL); 631 #endif /* SMP */ 632 633 void 634 smp_no_rendevous_barrier(void *dummy) 635 { 636 #ifdef SMP 637 KASSERT((!smp_started),("smp_no_rendevous called and smp is started")); 638 #endif 639 } 640