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