1 /* 2 * Copyright (c) 2003 Matthew Dillon <dillon@backplane.com> All rights reserved. 3 * Copyright (c) 1997, Stefan Esser <se@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 unmodified, this list of conditions, and the following 10 * disclaimer. 11 * 2. Redistributions in binary form must reproduce the above copyright 12 * notice, this list of conditions and the following disclaimer in the 13 * documentation and/or other materials provided with the distribution. 14 * 15 * THIS SOFTWARE IS PROVIDED BY THE AUTHOR ``AS IS'' AND ANY EXPRESS OR 16 * IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE IMPLIED WARRANTIES 17 * OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE ARE DISCLAIMED. 18 * IN NO EVENT SHALL THE AUTHOR BE LIABLE FOR ANY DIRECT, INDIRECT, 19 * INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT 20 * NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, 21 * DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY 22 * THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT 23 * (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE OF 24 * THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE. 25 * 26 * $FreeBSD: src/sys/kern/kern_intr.c,v 1.24.2.1 2001/10/14 20:05:50 luigi Exp $ 27 * $DragonFly: src/sys/kern/kern_intr.c,v 1.55 2008/09/01 12:49:00 sephe Exp $ 28 * 29 */ 30 31 #include <sys/param.h> 32 #include <sys/systm.h> 33 #include <sys/malloc.h> 34 #include <sys/kernel.h> 35 #include <sys/sysctl.h> 36 #include <sys/thread.h> 37 #include <sys/proc.h> 38 #include <sys/random.h> 39 #include <sys/serialize.h> 40 #include <sys/interrupt.h> 41 #include <sys/bus.h> 42 #include <sys/machintr.h> 43 44 #include <machine/frame.h> 45 46 #include <sys/interrupt.h> 47 48 #include <sys/thread2.h> 49 #include <sys/mplock2.h> 50 51 struct info_info; 52 53 typedef struct intrec { 54 struct intrec *next; 55 struct intr_info *info; 56 inthand2_t *handler; 57 void *argument; 58 char *name; 59 int intr; 60 int intr_flags; 61 struct lwkt_serialize *serializer; 62 } *intrec_t; 63 64 struct intr_info { 65 intrec_t i_reclist; 66 struct thread i_thread; 67 struct random_softc i_random; 68 int i_running; 69 long i_count; /* interrupts dispatched */ 70 int i_mplock_required; 71 int i_fast; 72 int i_slow; 73 int i_state; 74 int i_errorticks; 75 unsigned long i_straycount; 76 } intr_info_ary[MAX_INTS]; 77 78 int max_installed_hard_intr; 79 int max_installed_soft_intr; 80 81 #define EMERGENCY_INTR_POLLING_FREQ_MAX 20000 82 83 static int sysctl_emergency_freq(SYSCTL_HANDLER_ARGS); 84 static int sysctl_emergency_enable(SYSCTL_HANDLER_ARGS); 85 static void emergency_intr_timer_callback(systimer_t, struct intrframe *); 86 static void ithread_handler(void *arg); 87 static void ithread_emergency(void *arg); 88 static void report_stray_interrupt(int intr, struct intr_info *info); 89 static void int_moveto_destcpu(int *, int *, int); 90 static void int_moveto_origcpu(int, int); 91 #ifdef SMP 92 static void intr_get_mplock(void); 93 #endif 94 95 int intr_info_size = sizeof(intr_info_ary) / sizeof(intr_info_ary[0]); 96 97 static struct systimer emergency_intr_timer; 98 static struct thread emergency_intr_thread; 99 100 #define ISTATE_NOTHREAD 0 101 #define ISTATE_NORMAL 1 102 #define ISTATE_LIVELOCKED 2 103 104 #ifdef SMP 105 static int intr_mpsafe = 1; 106 static int intr_migrate = 0; 107 static int intr_migrate_count; 108 TUNABLE_INT("kern.intr_mpsafe", &intr_mpsafe); 109 SYSCTL_INT(_kern, OID_AUTO, intr_mpsafe, 110 CTLFLAG_RW, &intr_mpsafe, 0, "Run INTR_MPSAFE handlers without the BGL"); 111 SYSCTL_INT(_kern, OID_AUTO, intr_migrate, 112 CTLFLAG_RW, &intr_migrate, 0, "Migrate to cpu holding BGL"); 113 SYSCTL_INT(_kern, OID_AUTO, intr_migrate_count, 114 CTLFLAG_RW, &intr_migrate_count, 0, ""); 115 #endif 116 static int livelock_limit = 40000; 117 static int livelock_lowater = 20000; 118 static int livelock_debug = -1; 119 SYSCTL_INT(_kern, OID_AUTO, livelock_limit, 120 CTLFLAG_RW, &livelock_limit, 0, "Livelock interrupt rate limit"); 121 SYSCTL_INT(_kern, OID_AUTO, livelock_lowater, 122 CTLFLAG_RW, &livelock_lowater, 0, "Livelock low-water mark restore"); 123 SYSCTL_INT(_kern, OID_AUTO, livelock_debug, 124 CTLFLAG_RW, &livelock_debug, 0, "Livelock debug intr#"); 125 126 static int emergency_intr_enable = 0; /* emergency interrupt polling */ 127 TUNABLE_INT("kern.emergency_intr_enable", &emergency_intr_enable); 128 SYSCTL_PROC(_kern, OID_AUTO, emergency_intr_enable, CTLTYPE_INT | CTLFLAG_RW, 129 0, 0, sysctl_emergency_enable, "I", "Emergency Interrupt Poll Enable"); 130 131 static int emergency_intr_freq = 10; /* emergency polling frequency */ 132 TUNABLE_INT("kern.emergency_intr_freq", &emergency_intr_freq); 133 SYSCTL_PROC(_kern, OID_AUTO, emergency_intr_freq, CTLTYPE_INT | CTLFLAG_RW, 134 0, 0, sysctl_emergency_freq, "I", "Emergency Interrupt Poll Frequency"); 135 136 /* 137 * Sysctl support routines 138 */ 139 static int 140 sysctl_emergency_enable(SYSCTL_HANDLER_ARGS) 141 { 142 int error, enabled; 143 144 enabled = emergency_intr_enable; 145 error = sysctl_handle_int(oidp, &enabled, 0, req); 146 if (error || req->newptr == NULL) 147 return error; 148 emergency_intr_enable = enabled; 149 if (emergency_intr_enable) { 150 systimer_adjust_periodic(&emergency_intr_timer, 151 emergency_intr_freq); 152 } else { 153 systimer_adjust_periodic(&emergency_intr_timer, 1); 154 } 155 return 0; 156 } 157 158 static int 159 sysctl_emergency_freq(SYSCTL_HANDLER_ARGS) 160 { 161 int error, phz; 162 163 phz = emergency_intr_freq; 164 error = sysctl_handle_int(oidp, &phz, 0, req); 165 if (error || req->newptr == NULL) 166 return error; 167 if (phz <= 0) 168 return EINVAL; 169 else if (phz > EMERGENCY_INTR_POLLING_FREQ_MAX) 170 phz = EMERGENCY_INTR_POLLING_FREQ_MAX; 171 172 emergency_intr_freq = phz; 173 if (emergency_intr_enable) { 174 systimer_adjust_periodic(&emergency_intr_timer, 175 emergency_intr_freq); 176 } else { 177 systimer_adjust_periodic(&emergency_intr_timer, 1); 178 } 179 return 0; 180 } 181 182 /* 183 * Register an SWI or INTerrupt handler. 184 */ 185 void * 186 register_swi(int intr, inthand2_t *handler, void *arg, const char *name, 187 struct lwkt_serialize *serializer) 188 { 189 if (intr < FIRST_SOFTINT || intr >= MAX_INTS) 190 panic("register_swi: bad intr %d", intr); 191 return(register_int(intr, handler, arg, name, serializer, 0)); 192 } 193 194 void * 195 register_swi_mp(int intr, inthand2_t *handler, void *arg, const char *name, 196 struct lwkt_serialize *serializer) 197 { 198 if (intr < FIRST_SOFTINT || intr >= MAX_INTS) 199 panic("register_swi: bad intr %d", intr); 200 return(register_int(intr, handler, arg, name, serializer, INTR_MPSAFE)); 201 } 202 203 void * 204 register_int(int intr, inthand2_t *handler, void *arg, const char *name, 205 struct lwkt_serialize *serializer, int intr_flags) 206 { 207 struct intr_info *info; 208 struct intrec **list; 209 intrec_t rec; 210 int orig_cpuid, cpuid; 211 212 if (intr < 0 || intr >= MAX_INTS) 213 panic("register_int: bad intr %d", intr); 214 if (name == NULL) 215 name = "???"; 216 info = &intr_info_ary[intr]; 217 218 /* 219 * Construct an interrupt handler record 220 */ 221 rec = kmalloc(sizeof(struct intrec), M_DEVBUF, M_INTWAIT); 222 rec->name = kmalloc(strlen(name) + 1, M_DEVBUF, M_INTWAIT); 223 strcpy(rec->name, name); 224 225 rec->info = info; 226 rec->handler = handler; 227 rec->argument = arg; 228 rec->intr = intr; 229 rec->intr_flags = intr_flags; 230 rec->next = NULL; 231 rec->serializer = serializer; 232 233 /* 234 * Create an emergency polling thread and set up a systimer to wake 235 * it up. 236 */ 237 if (emergency_intr_thread.td_kstack == NULL) { 238 lwkt_create(ithread_emergency, NULL, NULL, 239 &emergency_intr_thread, TDF_STOPREQ|TDF_INTTHREAD, -1, 240 "ithread emerg"); 241 systimer_init_periodic_nq(&emergency_intr_timer, 242 emergency_intr_timer_callback, &emergency_intr_thread, 243 (emergency_intr_enable ? emergency_intr_freq : 1)); 244 } 245 246 int_moveto_destcpu(&orig_cpuid, &cpuid, intr); 247 248 /* 249 * Create an interrupt thread if necessary, leave it in an unscheduled 250 * state. 251 */ 252 if (info->i_state == ISTATE_NOTHREAD) { 253 info->i_state = ISTATE_NORMAL; 254 lwkt_create((void *)ithread_handler, (void *)(intptr_t)intr, NULL, 255 &info->i_thread, TDF_STOPREQ|TDF_INTTHREAD|TDF_MPSAFE, -1, 256 "ithread %d", intr); 257 if (intr >= FIRST_SOFTINT) 258 lwkt_setpri(&info->i_thread, TDPRI_SOFT_NORM); 259 else 260 lwkt_setpri(&info->i_thread, TDPRI_INT_MED); 261 info->i_thread.td_preemptable = lwkt_preempt; 262 } 263 264 list = &info->i_reclist; 265 266 /* 267 * Keep track of how many fast and slow interrupts we have. 268 * Set i_mplock_required if any handler in the chain requires 269 * the MP lock to operate. 270 */ 271 if ((intr_flags & INTR_MPSAFE) == 0) 272 info->i_mplock_required = 1; 273 if (intr_flags & INTR_CLOCK) 274 ++info->i_fast; 275 else 276 ++info->i_slow; 277 278 /* 279 * Enable random number generation keying off of this interrupt. 280 */ 281 if ((intr_flags & INTR_NOENTROPY) == 0 && info->i_random.sc_enabled == 0) { 282 info->i_random.sc_enabled = 1; 283 info->i_random.sc_intr = intr; 284 } 285 286 /* 287 * Add the record to the interrupt list. 288 */ 289 crit_enter(); 290 while (*list != NULL) 291 list = &(*list)->next; 292 *list = rec; 293 crit_exit(); 294 295 /* 296 * Update max_installed_hard_intr to make the emergency intr poll 297 * a bit more efficient. 298 */ 299 if (intr < FIRST_SOFTINT) { 300 if (max_installed_hard_intr <= intr) 301 max_installed_hard_intr = intr + 1; 302 } else { 303 if (max_installed_soft_intr <= intr) 304 max_installed_soft_intr = intr + 1; 305 } 306 307 /* 308 * Setup the machine level interrupt vector 309 * 310 * XXX temporary workaround for some ACPI brokedness. ACPI installs 311 * its interrupt too early, before the IOAPICs have been configured, 312 * which means the IOAPIC is not enabled by the registration of the 313 * ACPI interrupt. Anything else sharing that IRQ will wind up not 314 * being enabled. Temporarily work around the problem by always 315 * installing and enabling on every new interrupt handler, even 316 * if one has already been setup on that irq. 317 */ 318 if (intr < FIRST_SOFTINT /* && info->i_slow + info->i_fast == 1*/) { 319 if (machintr_vector_setup(intr, intr_flags)) 320 kprintf("machintr_vector_setup: failed on irq %d\n", intr); 321 } 322 323 int_moveto_origcpu(orig_cpuid, cpuid); 324 325 return(rec); 326 } 327 328 void 329 unregister_swi(void *id) 330 { 331 unregister_int(id); 332 } 333 334 void 335 unregister_int(void *id) 336 { 337 struct intr_info *info; 338 struct intrec **list; 339 intrec_t rec; 340 int intr, orig_cpuid, cpuid; 341 342 intr = ((intrec_t)id)->intr; 343 344 if (intr < 0 || intr >= MAX_INTS) 345 panic("register_int: bad intr %d", intr); 346 347 info = &intr_info_ary[intr]; 348 349 int_moveto_destcpu(&orig_cpuid, &cpuid, intr); 350 351 /* 352 * Remove the interrupt descriptor, adjust the descriptor count, 353 * and teardown the machine level vector if this was the last interrupt. 354 */ 355 crit_enter(); 356 list = &info->i_reclist; 357 while ((rec = *list) != NULL) { 358 if (rec == id) 359 break; 360 list = &rec->next; 361 } 362 if (rec) { 363 intrec_t rec0; 364 365 *list = rec->next; 366 if (rec->intr_flags & INTR_CLOCK) 367 --info->i_fast; 368 else 369 --info->i_slow; 370 if (intr < FIRST_SOFTINT && info->i_fast + info->i_slow == 0) 371 machintr_vector_teardown(intr); 372 373 /* 374 * Clear i_mplock_required if no handlers in the chain require the 375 * MP lock. 376 */ 377 for (rec0 = info->i_reclist; rec0; rec0 = rec0->next) { 378 if ((rec0->intr_flags & INTR_MPSAFE) == 0) 379 break; 380 } 381 if (rec0 == NULL) 382 info->i_mplock_required = 0; 383 } 384 385 crit_exit(); 386 387 int_moveto_origcpu(orig_cpuid, cpuid); 388 389 /* 390 * Free the record. 391 */ 392 if (rec != NULL) { 393 kfree(rec->name, M_DEVBUF); 394 kfree(rec, M_DEVBUF); 395 } else { 396 kprintf("warning: unregister_int: int %d handler for %s not found\n", 397 intr, ((intrec_t)id)->name); 398 } 399 } 400 401 const char * 402 get_registered_name(int intr) 403 { 404 intrec_t rec; 405 406 if (intr < 0 || intr >= MAX_INTS) 407 panic("register_int: bad intr %d", intr); 408 409 if ((rec = intr_info_ary[intr].i_reclist) == NULL) 410 return(NULL); 411 else if (rec->next) 412 return("mux"); 413 else 414 return(rec->name); 415 } 416 417 int 418 count_registered_ints(int intr) 419 { 420 struct intr_info *info; 421 422 if (intr < 0 || intr >= MAX_INTS) 423 panic("register_int: bad intr %d", intr); 424 info = &intr_info_ary[intr]; 425 return(info->i_fast + info->i_slow); 426 } 427 428 long 429 get_interrupt_counter(int intr) 430 { 431 struct intr_info *info; 432 433 if (intr < 0 || intr >= MAX_INTS) 434 panic("register_int: bad intr %d", intr); 435 info = &intr_info_ary[intr]; 436 return(info->i_count); 437 } 438 439 440 void 441 swi_setpriority(int intr, int pri) 442 { 443 struct intr_info *info; 444 445 if (intr < FIRST_SOFTINT || intr >= MAX_INTS) 446 panic("register_swi: bad intr %d", intr); 447 info = &intr_info_ary[intr]; 448 if (info->i_state != ISTATE_NOTHREAD) 449 lwkt_setpri(&info->i_thread, pri); 450 } 451 452 void 453 register_randintr(int intr) 454 { 455 struct intr_info *info; 456 457 if (intr < 0 || intr >= MAX_INTS) 458 panic("register_randintr: bad intr %d", intr); 459 info = &intr_info_ary[intr]; 460 info->i_random.sc_intr = intr; 461 info->i_random.sc_enabled = 1; 462 } 463 464 void 465 unregister_randintr(int intr) 466 { 467 struct intr_info *info; 468 469 if (intr < 0 || intr >= MAX_INTS) 470 panic("register_swi: bad intr %d", intr); 471 info = &intr_info_ary[intr]; 472 info->i_random.sc_enabled = -1; 473 } 474 475 int 476 next_registered_randintr(int intr) 477 { 478 struct intr_info *info; 479 480 if (intr < 0 || intr >= MAX_INTS) 481 panic("register_swi: bad intr %d", intr); 482 while (intr < MAX_INTS) { 483 info = &intr_info_ary[intr]; 484 if (info->i_random.sc_enabled > 0) 485 break; 486 ++intr; 487 } 488 return(intr); 489 } 490 491 /* 492 * Dispatch an interrupt. If there's nothing to do we have a stray 493 * interrupt and can just return, leaving the interrupt masked. 494 * 495 * We need to schedule the interrupt and set its i_running bit. If 496 * we are not on the interrupt thread's cpu we have to send a message 497 * to the correct cpu that will issue the desired action (interlocking 498 * with the interrupt thread's critical section). We do NOT attempt to 499 * reschedule interrupts whos i_running bit is already set because 500 * this would prematurely wakeup a livelock-limited interrupt thread. 501 * 502 * i_running is only tested/set on the same cpu as the interrupt thread. 503 * 504 * We are NOT in a critical section, which will allow the scheduled 505 * interrupt to preempt us. The MP lock might *NOT* be held here. 506 */ 507 #ifdef SMP 508 509 static void 510 sched_ithd_remote(void *arg) 511 { 512 sched_ithd((int)(intptr_t)arg); 513 } 514 515 #endif 516 517 void 518 sched_ithd(int intr) 519 { 520 struct intr_info *info; 521 522 info = &intr_info_ary[intr]; 523 524 ++info->i_count; 525 if (info->i_state != ISTATE_NOTHREAD) { 526 if (info->i_reclist == NULL) { 527 report_stray_interrupt(intr, info); 528 } else { 529 #ifdef SMP 530 if (info->i_thread.td_gd == mycpu) { 531 if (info->i_running == 0) { 532 info->i_running = 1; 533 if (info->i_state != ISTATE_LIVELOCKED) 534 lwkt_schedule(&info->i_thread); /* MIGHT PREEMPT */ 535 } 536 } else { 537 lwkt_send_ipiq(info->i_thread.td_gd, 538 sched_ithd_remote, (void *)(intptr_t)intr); 539 } 540 #else 541 if (info->i_running == 0) { 542 info->i_running = 1; 543 if (info->i_state != ISTATE_LIVELOCKED) 544 lwkt_schedule(&info->i_thread); /* MIGHT PREEMPT */ 545 } 546 #endif 547 } 548 } else { 549 report_stray_interrupt(intr, info); 550 } 551 } 552 553 static void 554 report_stray_interrupt(int intr, struct intr_info *info) 555 { 556 ++info->i_straycount; 557 if (info->i_straycount < 10) { 558 if (info->i_errorticks == ticks) 559 return; 560 info->i_errorticks = ticks; 561 kprintf("sched_ithd: stray interrupt %d on cpu %d\n", 562 intr, mycpuid); 563 } else if (info->i_straycount == 10) { 564 kprintf("sched_ithd: %ld stray interrupts %d on cpu %d - " 565 "there will be no further reports\n", 566 info->i_straycount, intr, mycpuid); 567 } 568 } 569 570 /* 571 * This is run from a periodic SYSTIMER (and thus must be MP safe, the BGL 572 * might not be held). 573 */ 574 static void 575 ithread_livelock_wakeup(systimer_t st) 576 { 577 struct intr_info *info; 578 579 info = &intr_info_ary[(int)(intptr_t)st->data]; 580 if (info->i_state != ISTATE_NOTHREAD) 581 lwkt_schedule(&info->i_thread); 582 } 583 584 /* 585 * Schedule ithread within fast intr handler 586 * 587 * XXX Protect sched_ithd() call with gd_intr_nesting_level? 588 * Interrupts aren't enabled, but still... 589 */ 590 static __inline void 591 ithread_fast_sched(int intr, thread_t td) 592 { 593 ++td->td_nest_count; 594 595 /* 596 * We are already in critical section, exit it now to 597 * allow preemption. 598 */ 599 crit_exit_quick(td); 600 sched_ithd(intr); 601 crit_enter_quick(td); 602 603 --td->td_nest_count; 604 } 605 606 /* 607 * This function is called directly from the ICU or APIC vector code assembly 608 * to process an interrupt. The critical section and interrupt deferral 609 * checks have already been done but the function is entered WITHOUT 610 * a critical section held. The BGL may or may not be held. 611 * 612 * Must return non-zero if we do not want the vector code to re-enable 613 * the interrupt (which we don't if we have to schedule the interrupt) 614 */ 615 int ithread_fast_handler(struct intrframe *frame); 616 617 int 618 ithread_fast_handler(struct intrframe *frame) 619 { 620 int intr; 621 struct intr_info *info; 622 struct intrec **list; 623 int must_schedule; 624 #ifdef SMP 625 int got_mplock; 626 #endif 627 intrec_t rec, next_rec; 628 globaldata_t gd; 629 thread_t td; 630 631 intr = frame->if_vec; 632 gd = mycpu; 633 td = curthread; 634 635 /* We must be in critical section. */ 636 KKASSERT(td->td_critcount); 637 638 info = &intr_info_ary[intr]; 639 640 /* 641 * If we are not processing any FAST interrupts, just schedule the thing. 642 */ 643 if (info->i_fast == 0) { 644 ++gd->gd_cnt.v_intr; 645 ithread_fast_sched(intr, td); 646 return(1); 647 } 648 649 /* 650 * This should not normally occur since interrupts ought to be 651 * masked if the ithread has been scheduled or is running. 652 */ 653 if (info->i_running) 654 return(1); 655 656 /* 657 * Bump the interrupt nesting level to process any FAST interrupts. 658 * Obtain the MP lock as necessary. If the MP lock cannot be obtained, 659 * schedule the interrupt thread to deal with the issue instead. 660 * 661 * To reduce overhead, just leave the MP lock held once it has been 662 * obtained. 663 */ 664 ++gd->gd_intr_nesting_level; 665 ++gd->gd_cnt.v_intr; 666 must_schedule = info->i_slow; 667 #ifdef SMP 668 got_mplock = 0; 669 #endif 670 671 list = &info->i_reclist; 672 for (rec = *list; rec; rec = next_rec) { 673 next_rec = rec->next; /* rec may be invalid after call */ 674 675 if (rec->intr_flags & INTR_CLOCK) { 676 #ifdef SMP 677 if ((rec->intr_flags & INTR_MPSAFE) == 0 && got_mplock == 0) { 678 if (try_mplock() == 0) { 679 /* Couldn't get the MP lock; just schedule it. */ 680 must_schedule = 1; 681 break; 682 } 683 got_mplock = 1; 684 } 685 #endif 686 if (rec->serializer) { 687 must_schedule += lwkt_serialize_handler_try( 688 rec->serializer, rec->handler, 689 rec->argument, frame); 690 } else { 691 rec->handler(rec->argument, frame); 692 } 693 } 694 } 695 696 /* 697 * Cleanup 698 */ 699 --gd->gd_intr_nesting_level; 700 #ifdef SMP 701 if (got_mplock) 702 rel_mplock(); 703 #endif 704 705 /* 706 * If we had a problem, or mixed fast and slow interrupt handlers are 707 * registered, schedule the ithread to catch the missed records (it 708 * will just re-run all of them). A return value of 0 indicates that 709 * all handlers have been run and the interrupt can be re-enabled, and 710 * a non-zero return indicates that the interrupt thread controls 711 * re-enablement. 712 */ 713 if (must_schedule > 0) 714 ithread_fast_sched(intr, td); 715 else if (must_schedule == 0) 716 ++info->i_count; 717 return(must_schedule); 718 } 719 720 /* 721 * Interrupt threads run this as their main loop. 722 * 723 * The handler begins execution outside a critical section and with the BGL 724 * held. 725 * 726 * The i_running state starts at 0. When an interrupt occurs, the hardware 727 * interrupt is disabled and sched_ithd() The HW interrupt remains disabled 728 * until all routines have run. We then call ithread_done() to reenable 729 * the HW interrupt and deschedule us until the next interrupt. 730 * 731 * We are responsible for atomically checking i_running and ithread_done() 732 * is responsible for atomically checking for platform-specific delayed 733 * interrupts. i_running for our irq is only set in the context of our cpu, 734 * so a critical section is a sufficient interlock. 735 */ 736 #define LIVELOCK_TIMEFRAME(freq) ((freq) >> 2) /* 1/4 second */ 737 738 static void 739 ithread_handler(void *arg) 740 { 741 struct intr_info *info; 742 int use_limit; 743 __uint32_t lseconds; 744 int intr; 745 int mpheld; 746 struct intrec **list; 747 intrec_t rec, nrec; 748 globaldata_t gd; 749 struct systimer ill_timer; /* enforced freq. timer */ 750 u_int ill_count; /* interrupt livelock counter */ 751 752 ill_count = 0; 753 intr = (int)(intptr_t)arg; 754 info = &intr_info_ary[intr]; 755 list = &info->i_reclist; 756 757 /* 758 * The loop must be entered with one critical section held. The thread 759 * is created with TDF_MPSAFE so the MP lock is not held on start. 760 */ 761 gd = mycpu; 762 lseconds = gd->gd_time_seconds; 763 crit_enter_gd(gd); 764 mpheld = 0; 765 766 for (;;) { 767 /* 768 * The chain is only considered MPSAFE if all its interrupt handlers 769 * are MPSAFE. However, if intr_mpsafe has been turned off we 770 * always operate with the BGL. 771 */ 772 #ifdef SMP 773 if (intr_mpsafe == 0) { 774 if (mpheld == 0) { 775 intr_get_mplock(); 776 mpheld = 1; 777 } 778 } else if (info->i_mplock_required != mpheld) { 779 if (info->i_mplock_required) { 780 KKASSERT(mpheld == 0); 781 intr_get_mplock(); 782 mpheld = 1; 783 } else { 784 KKASSERT(mpheld != 0); 785 rel_mplock(); 786 mpheld = 0; 787 } 788 } 789 790 /* 791 * scheduled cpu may have changed, see intr_get_mplock() 792 */ 793 gd = mycpu; 794 #endif 795 796 /* 797 * If an interrupt is pending, clear i_running and execute the 798 * handlers. Note that certain types of interrupts can re-trigger 799 * and set i_running again. 800 * 801 * Each handler is run in a critical section. Note that we run both 802 * FAST and SLOW designated service routines. 803 */ 804 if (info->i_running) { 805 ++ill_count; 806 info->i_running = 0; 807 808 if (*list == NULL) 809 report_stray_interrupt(intr, info); 810 811 for (rec = *list; rec; rec = nrec) { 812 nrec = rec->next; 813 if (rec->serializer) { 814 lwkt_serialize_handler_call(rec->serializer, rec->handler, 815 rec->argument, NULL); 816 } else { 817 rec->handler(rec->argument, NULL); 818 } 819 } 820 } 821 822 /* 823 * This is our interrupt hook to add rate randomness to the random 824 * number generator. 825 */ 826 if (info->i_random.sc_enabled > 0) 827 add_interrupt_randomness(intr); 828 829 /* 830 * Unmask the interrupt to allow it to trigger again. This only 831 * applies to certain types of interrupts (typ level interrupts). 832 * This can result in the interrupt retriggering, but the retrigger 833 * will not be processed until we cycle our critical section. 834 * 835 * Only unmask interrupts while handlers are installed. It is 836 * possible to hit a situation where no handlers are installed 837 * due to a device driver livelocking and then tearing down its 838 * interrupt on close (the parallel bus being a good example). 839 */ 840 if (*list) 841 machintr_intren(intr); 842 843 /* 844 * Do a quick exit/enter to catch any higher-priority interrupt 845 * sources, such as the statclock, so thread time accounting 846 * will still work. This may also cause an interrupt to re-trigger. 847 */ 848 crit_exit_gd(gd); 849 crit_enter_gd(gd); 850 851 /* 852 * LIVELOCK STATE MACHINE 853 */ 854 switch(info->i_state) { 855 case ISTATE_NORMAL: 856 /* 857 * Reset the count each second. 858 */ 859 if (lseconds != gd->gd_time_seconds) { 860 lseconds = gd->gd_time_seconds; 861 ill_count = 0; 862 } 863 864 /* 865 * If we did not exceed the frequency limit, we are done. 866 * If the interrupt has not retriggered we deschedule ourselves. 867 */ 868 if (ill_count <= livelock_limit) { 869 if (info->i_running == 0) { 870 #ifdef SMP 871 if (mpheld && intr_migrate) { 872 rel_mplock(); 873 mpheld = 0; 874 } 875 #endif 876 lwkt_deschedule_self(gd->gd_curthread); 877 lwkt_switch(); 878 } 879 break; 880 } 881 882 /* 883 * Otherwise we are livelocked. Set up a periodic systimer 884 * to wake the thread up at the limit frequency. 885 */ 886 kprintf("intr %d at %d/%d hz, livelocked limit engaged!\n", 887 intr, ill_count, livelock_limit); 888 info->i_state = ISTATE_LIVELOCKED; 889 if ((use_limit = livelock_limit) < 100) 890 use_limit = 100; 891 else if (use_limit > 500000) 892 use_limit = 500000; 893 systimer_init_periodic_nq(&ill_timer, ithread_livelock_wakeup, 894 (void *)(intptr_t)intr, use_limit); 895 /* fall through */ 896 case ISTATE_LIVELOCKED: 897 /* 898 * Wait for our periodic timer to go off. Since the interrupt 899 * has re-armed it can still set i_running, but it will not 900 * reschedule us while we are in a livelocked state. 901 */ 902 lwkt_deschedule_self(gd->gd_curthread); 903 lwkt_switch(); 904 905 /* 906 * Check once a second to see if the livelock condition no 907 * longer applies. 908 */ 909 if (lseconds != gd->gd_time_seconds) { 910 lseconds = gd->gd_time_seconds; 911 if (ill_count < livelock_lowater) { 912 info->i_state = ISTATE_NORMAL; 913 systimer_del(&ill_timer); 914 kprintf("intr %d at %d/%d hz, livelock removed\n", 915 intr, ill_count, livelock_lowater); 916 } else if (livelock_debug == intr || 917 (bootverbose && cold)) { 918 kprintf("intr %d at %d/%d hz, in livelock\n", 919 intr, ill_count, livelock_lowater); 920 } 921 ill_count = 0; 922 } 923 break; 924 } 925 } 926 /* not reached */ 927 } 928 929 #ifdef SMP 930 931 /* 932 * An interrupt thread is trying to get the MP lock. To avoid cpu-bound 933 * code in the kernel on cpu X from interfering we chase the MP lock. 934 */ 935 static void 936 intr_get_mplock(void) 937 { 938 int owner; 939 940 if (intr_migrate == 0) { 941 get_mplock(); 942 return; 943 } 944 while (try_mplock() == 0) { 945 owner = owner_mplock(); 946 if (owner >= 0 && owner != mycpu->gd_cpuid) { 947 lwkt_migratecpu(owner); 948 ++intr_migrate_count; 949 } else { 950 lwkt_switch(); 951 } 952 } 953 } 954 955 #endif 956 957 /* 958 * Emergency interrupt polling thread. The thread begins execution 959 * outside a critical section with the BGL held. 960 * 961 * If emergency interrupt polling is enabled, this thread will 962 * execute all system interrupts not marked INTR_NOPOLL at the 963 * specified polling frequency. 964 * 965 * WARNING! This thread runs *ALL* interrupt service routines that 966 * are not marked INTR_NOPOLL, which basically means everything except 967 * the 8254 clock interrupt and the ATA interrupt. It has very high 968 * overhead and should only be used in situations where the machine 969 * cannot otherwise be made to work. Due to the severe performance 970 * degredation, it should not be enabled on production machines. 971 */ 972 static void 973 ithread_emergency(void *arg __unused) 974 { 975 struct intr_info *info; 976 intrec_t rec, nrec; 977 int intr; 978 979 for (;;) { 980 for (intr = 0; intr < max_installed_hard_intr; ++intr) { 981 info = &intr_info_ary[intr]; 982 for (rec = info->i_reclist; rec; rec = nrec) { 983 if ((rec->intr_flags & INTR_NOPOLL) == 0) { 984 if (rec->serializer) { 985 lwkt_serialize_handler_call(rec->serializer, 986 rec->handler, rec->argument, NULL); 987 } else { 988 rec->handler(rec->argument, NULL); 989 } 990 } 991 nrec = rec->next; 992 } 993 } 994 lwkt_deschedule_self(curthread); 995 lwkt_switch(); 996 } 997 } 998 999 /* 1000 * Systimer callback - schedule the emergency interrupt poll thread 1001 * if emergency polling is enabled. 1002 */ 1003 static 1004 void 1005 emergency_intr_timer_callback(systimer_t info, struct intrframe *frame __unused) 1006 { 1007 if (emergency_intr_enable) 1008 lwkt_schedule(info->data); 1009 } 1010 1011 int 1012 ithread_cpuid(int intr) 1013 { 1014 const struct intr_info *info; 1015 1016 KKASSERT(intr >= 0 && intr < MAX_INTS); 1017 info = &intr_info_ary[intr]; 1018 1019 if (info->i_state == ISTATE_NOTHREAD) 1020 return -1; 1021 return info->i_thread.td_gd->gd_cpuid; 1022 } 1023 1024 /* 1025 * Sysctls used by systat and others: hw.intrnames and hw.intrcnt. 1026 * The data for this machine dependent, and the declarations are in machine 1027 * dependent code. The layout of intrnames and intrcnt however is machine 1028 * independent. 1029 * 1030 * We do not know the length of intrcnt and intrnames at compile time, so 1031 * calculate things at run time. 1032 */ 1033 1034 static int 1035 sysctl_intrnames(SYSCTL_HANDLER_ARGS) 1036 { 1037 struct intr_info *info; 1038 intrec_t rec; 1039 int error = 0; 1040 int len; 1041 int intr; 1042 char buf[64]; 1043 1044 for (intr = 0; error == 0 && intr < MAX_INTS; ++intr) { 1045 info = &intr_info_ary[intr]; 1046 1047 len = 0; 1048 buf[0] = 0; 1049 for (rec = info->i_reclist; rec; rec = rec->next) { 1050 ksnprintf(buf + len, sizeof(buf) - len, "%s%s", 1051 (len ? "/" : ""), rec->name); 1052 len += strlen(buf + len); 1053 } 1054 if (len == 0) { 1055 ksnprintf(buf, sizeof(buf), "irq%d", intr); 1056 len = strlen(buf); 1057 } 1058 error = SYSCTL_OUT(req, buf, len + 1); 1059 } 1060 return (error); 1061 } 1062 1063 1064 SYSCTL_PROC(_hw, OID_AUTO, intrnames, CTLTYPE_OPAQUE | CTLFLAG_RD, 1065 NULL, 0, sysctl_intrnames, "", "Interrupt Names"); 1066 1067 static int 1068 sysctl_intrcnt(SYSCTL_HANDLER_ARGS) 1069 { 1070 struct intr_info *info; 1071 int error = 0; 1072 int intr; 1073 1074 for (intr = 0; intr < max_installed_hard_intr; ++intr) { 1075 info = &intr_info_ary[intr]; 1076 1077 error = SYSCTL_OUT(req, &info->i_count, sizeof(info->i_count)); 1078 if (error) 1079 goto failed; 1080 } 1081 for (intr = FIRST_SOFTINT; intr < max_installed_soft_intr; ++intr) { 1082 info = &intr_info_ary[intr]; 1083 1084 error = SYSCTL_OUT(req, &info->i_count, sizeof(info->i_count)); 1085 if (error) 1086 goto failed; 1087 } 1088 failed: 1089 return(error); 1090 } 1091 1092 SYSCTL_PROC(_hw, OID_AUTO, intrcnt, CTLTYPE_OPAQUE | CTLFLAG_RD, 1093 NULL, 0, sysctl_intrcnt, "", "Interrupt Counts"); 1094 1095 static void 1096 int_moveto_destcpu(int *orig_cpuid0, int *cpuid0, int intr) 1097 { 1098 int orig_cpuid = mycpuid, cpuid; 1099 char envpath[32]; 1100 1101 cpuid = orig_cpuid; 1102 ksnprintf(envpath, sizeof(envpath), "hw.irq.%d.dest", intr); 1103 kgetenv_int(envpath, &cpuid); 1104 if (cpuid >= ncpus) 1105 cpuid = orig_cpuid; 1106 1107 if (cpuid != orig_cpuid) 1108 lwkt_migratecpu(cpuid); 1109 1110 *orig_cpuid0 = orig_cpuid; 1111 *cpuid0 = cpuid; 1112 } 1113 1114 static void 1115 int_moveto_origcpu(int orig_cpuid, int cpuid) 1116 { 1117 if (cpuid != orig_cpuid) 1118 lwkt_migratecpu(orig_cpuid); 1119 } 1120