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