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 88 int intr_info_size = sizeof(intr_info_ary) / sizeof(intr_info_ary[0]); 89 90 static struct systimer emergency_intr_timer; 91 static struct thread emergency_intr_thread; 92 93 #define ISTATE_NOTHREAD 0 94 #define ISTATE_NORMAL 1 95 #define ISTATE_LIVELOCKED 2 96 97 #ifdef SMP 98 static int intr_mpsafe = 1; 99 TUNABLE_INT("kern.intr_mpsafe", &intr_mpsafe); 100 SYSCTL_INT(_kern, OID_AUTO, intr_mpsafe, 101 CTLFLAG_RW, &intr_mpsafe, 0, "Run INTR_MPSAFE handlers without the BGL"); 102 #endif 103 static int livelock_limit = 40000; 104 static int livelock_lowater = 20000; 105 static int livelock_debug = -1; 106 SYSCTL_INT(_kern, OID_AUTO, livelock_limit, 107 CTLFLAG_RW, &livelock_limit, 0, "Livelock interrupt rate limit"); 108 SYSCTL_INT(_kern, OID_AUTO, livelock_lowater, 109 CTLFLAG_RW, &livelock_lowater, 0, "Livelock low-water mark restore"); 110 SYSCTL_INT(_kern, OID_AUTO, livelock_debug, 111 CTLFLAG_RW, &livelock_debug, 0, "Livelock debug intr#"); 112 113 static int emergency_intr_enable = 0; /* emergency interrupt polling */ 114 TUNABLE_INT("kern.emergency_intr_enable", &emergency_intr_enable); 115 SYSCTL_PROC(_kern, OID_AUTO, emergency_intr_enable, CTLTYPE_INT | CTLFLAG_RW, 116 0, 0, sysctl_emergency_enable, "I", "Emergency Interrupt Poll Enable"); 117 118 static int emergency_intr_freq = 10; /* emergency polling frequency */ 119 TUNABLE_INT("kern.emergency_intr_freq", &emergency_intr_freq); 120 SYSCTL_PROC(_kern, OID_AUTO, emergency_intr_freq, CTLTYPE_INT | CTLFLAG_RW, 121 0, 0, sysctl_emergency_freq, "I", "Emergency Interrupt Poll Frequency"); 122 123 /* 124 * Sysctl support routines 125 */ 126 static int 127 sysctl_emergency_enable(SYSCTL_HANDLER_ARGS) 128 { 129 int error, enabled; 130 131 enabled = emergency_intr_enable; 132 error = sysctl_handle_int(oidp, &enabled, 0, req); 133 if (error || req->newptr == NULL) 134 return error; 135 emergency_intr_enable = enabled; 136 if (emergency_intr_enable) { 137 systimer_adjust_periodic(&emergency_intr_timer, 138 emergency_intr_freq); 139 } else { 140 systimer_adjust_periodic(&emergency_intr_timer, 1); 141 } 142 return 0; 143 } 144 145 static int 146 sysctl_emergency_freq(SYSCTL_HANDLER_ARGS) 147 { 148 int error, phz; 149 150 phz = emergency_intr_freq; 151 error = sysctl_handle_int(oidp, &phz, 0, req); 152 if (error || req->newptr == NULL) 153 return error; 154 if (phz <= 0) 155 return EINVAL; 156 else if (phz > EMERGENCY_INTR_POLLING_FREQ_MAX) 157 phz = EMERGENCY_INTR_POLLING_FREQ_MAX; 158 159 emergency_intr_freq = phz; 160 if (emergency_intr_enable) { 161 systimer_adjust_periodic(&emergency_intr_timer, 162 emergency_intr_freq); 163 } else { 164 systimer_adjust_periodic(&emergency_intr_timer, 1); 165 } 166 return 0; 167 } 168 169 /* 170 * Register an SWI or INTerrupt handler. 171 */ 172 void * 173 register_swi(int intr, inthand2_t *handler, void *arg, const char *name, 174 struct lwkt_serialize *serializer) 175 { 176 if (intr < FIRST_SOFTINT || intr >= MAX_INTS) 177 panic("register_swi: bad intr %d", intr); 178 return(register_int(intr, handler, arg, name, serializer, 0)); 179 } 180 181 void * 182 register_int(int intr, inthand2_t *handler, void *arg, const char *name, 183 struct lwkt_serialize *serializer, int intr_flags) 184 { 185 struct intr_info *info; 186 struct intrec **list; 187 intrec_t rec; 188 189 if (intr < 0 || intr >= MAX_INTS) 190 panic("register_int: bad intr %d", intr); 191 if (name == NULL) 192 name = "???"; 193 info = &intr_info_ary[intr]; 194 195 /* 196 * Construct an interrupt handler record 197 */ 198 rec = kmalloc(sizeof(struct intrec), M_DEVBUF, M_INTWAIT); 199 rec->name = kmalloc(strlen(name) + 1, M_DEVBUF, M_INTWAIT); 200 strcpy(rec->name, name); 201 202 rec->info = info; 203 rec->handler = handler; 204 rec->argument = arg; 205 rec->intr = intr; 206 rec->intr_flags = intr_flags; 207 rec->next = NULL; 208 rec->serializer = serializer; 209 210 /* 211 * Create an emergency polling thread and set up a systimer to wake 212 * it up. 213 */ 214 if (emergency_intr_thread.td_kstack == NULL) { 215 lwkt_create(ithread_emergency, NULL, NULL, 216 &emergency_intr_thread, TDF_STOPREQ|TDF_INTTHREAD, -1, 217 "ithread emerg"); 218 systimer_init_periodic_nq(&emergency_intr_timer, 219 emergency_intr_timer_callback, &emergency_intr_thread, 220 (emergency_intr_enable ? emergency_intr_freq : 1)); 221 } 222 223 /* 224 * Create an interrupt thread if necessary, leave it in an unscheduled 225 * state. 226 */ 227 if (info->i_state == ISTATE_NOTHREAD) { 228 info->i_state = ISTATE_NORMAL; 229 lwkt_create((void *)ithread_handler, (void *)intr, NULL, 230 &info->i_thread, TDF_STOPREQ|TDF_INTTHREAD|TDF_MPSAFE, -1, 231 "ithread %d", intr); 232 if (intr >= FIRST_SOFTINT) 233 lwkt_setpri(&info->i_thread, TDPRI_SOFT_NORM); 234 else 235 lwkt_setpri(&info->i_thread, TDPRI_INT_MED); 236 info->i_thread.td_preemptable = lwkt_preempt; 237 } 238 239 list = &info->i_reclist; 240 241 /* 242 * Keep track of how many fast and slow interrupts we have. 243 * Set i_mplock_required if any handler in the chain requires 244 * the MP lock to operate. 245 */ 246 if ((intr_flags & INTR_MPSAFE) == 0) 247 info->i_mplock_required = 1; 248 if (intr_flags & INTR_FAST) 249 ++info->i_fast; 250 else 251 ++info->i_slow; 252 253 /* 254 * Enable random number generation keying off of this interrupt. 255 */ 256 if ((intr_flags & INTR_NOENTROPY) == 0 && info->i_random.sc_enabled == 0) { 257 info->i_random.sc_enabled = 1; 258 info->i_random.sc_intr = intr; 259 } 260 261 /* 262 * Add the record to the interrupt list. 263 */ 264 crit_enter(); 265 while (*list != NULL) 266 list = &(*list)->next; 267 *list = rec; 268 crit_exit(); 269 270 /* 271 * Update max_installed_hard_intr to make the emergency intr poll 272 * a bit more efficient. 273 */ 274 if (intr < FIRST_SOFTINT) { 275 if (max_installed_hard_intr <= intr) 276 max_installed_hard_intr = intr + 1; 277 } else { 278 if (max_installed_soft_intr <= intr) 279 max_installed_soft_intr = intr + 1; 280 } 281 282 /* 283 * Setup the machine level interrupt vector 284 * 285 * XXX temporary workaround for some ACPI brokedness. ACPI installs 286 * its interrupt too early, before the IOAPICs have been configured, 287 * which means the IOAPIC is not enabled by the registration of the 288 * ACPI interrupt. Anything else sharing that IRQ will wind up not 289 * being enabled. Temporarily work around the problem by always 290 * installing and enabling on every new interrupt handler, even 291 * if one has already been setup on that irq. 292 */ 293 if (intr < FIRST_SOFTINT /* && info->i_slow + info->i_fast == 1*/) { 294 if (machintr_vector_setup(intr, intr_flags)) 295 kprintf("machintr_vector_setup: failed on irq %d\n", intr); 296 } 297 298 return(rec); 299 } 300 301 void 302 unregister_swi(void *id) 303 { 304 unregister_int(id); 305 } 306 307 void 308 unregister_int(void *id) 309 { 310 struct intr_info *info; 311 struct intrec **list; 312 intrec_t rec; 313 int intr; 314 315 intr = ((intrec_t)id)->intr; 316 317 if (intr < 0 || intr >= MAX_INTS) 318 panic("register_int: bad intr %d", intr); 319 320 info = &intr_info_ary[intr]; 321 322 /* 323 * Remove the interrupt descriptor, adjust the descriptor count, 324 * and teardown the machine level vector if this was the last interrupt. 325 */ 326 crit_enter(); 327 list = &info->i_reclist; 328 while ((rec = *list) != NULL) { 329 if (rec == id) 330 break; 331 list = &rec->next; 332 } 333 if (rec) { 334 intrec_t rec0; 335 336 *list = rec->next; 337 if (rec->intr_flags & INTR_FAST) 338 --info->i_fast; 339 else 340 --info->i_slow; 341 if (intr < FIRST_SOFTINT && info->i_fast + info->i_slow == 0) 342 machintr_vector_teardown(intr); 343 344 /* 345 * Clear i_mplock_required if no handlers in the chain require the 346 * MP lock. 347 */ 348 for (rec0 = info->i_reclist; rec0; rec0 = rec0->next) { 349 if ((rec0->intr_flags & INTR_MPSAFE) == 0) 350 break; 351 } 352 if (rec0 == NULL) 353 info->i_mplock_required = 0; 354 } 355 356 crit_exit(); 357 358 /* 359 * Free the record. 360 */ 361 if (rec != NULL) { 362 kfree(rec->name, M_DEVBUF); 363 kfree(rec, M_DEVBUF); 364 } else { 365 kprintf("warning: unregister_int: int %d handler for %s not found\n", 366 intr, ((intrec_t)id)->name); 367 } 368 } 369 370 const char * 371 get_registered_name(int intr) 372 { 373 intrec_t rec; 374 375 if (intr < 0 || intr >= MAX_INTS) 376 panic("register_int: bad intr %d", intr); 377 378 if ((rec = intr_info_ary[intr].i_reclist) == NULL) 379 return(NULL); 380 else if (rec->next) 381 return("mux"); 382 else 383 return(rec->name); 384 } 385 386 int 387 count_registered_ints(int intr) 388 { 389 struct intr_info *info; 390 391 if (intr < 0 || intr >= MAX_INTS) 392 panic("register_int: bad intr %d", intr); 393 info = &intr_info_ary[intr]; 394 return(info->i_fast + info->i_slow); 395 } 396 397 long 398 get_interrupt_counter(int intr) 399 { 400 struct intr_info *info; 401 402 if (intr < 0 || intr >= MAX_INTS) 403 panic("register_int: bad intr %d", intr); 404 info = &intr_info_ary[intr]; 405 return(info->i_count); 406 } 407 408 409 void 410 swi_setpriority(int intr, int pri) 411 { 412 struct intr_info *info; 413 414 if (intr < FIRST_SOFTINT || intr >= MAX_INTS) 415 panic("register_swi: bad intr %d", intr); 416 info = &intr_info_ary[intr]; 417 if (info->i_state != ISTATE_NOTHREAD) 418 lwkt_setpri(&info->i_thread, pri); 419 } 420 421 void 422 register_randintr(int intr) 423 { 424 struct intr_info *info; 425 426 if (intr < 0 || intr >= MAX_INTS) 427 panic("register_randintr: bad intr %d", intr); 428 info = &intr_info_ary[intr]; 429 info->i_random.sc_intr = intr; 430 info->i_random.sc_enabled = 1; 431 } 432 433 void 434 unregister_randintr(int intr) 435 { 436 struct intr_info *info; 437 438 if (intr < 0 || intr >= MAX_INTS) 439 panic("register_swi: bad intr %d", intr); 440 info = &intr_info_ary[intr]; 441 info->i_random.sc_enabled = -1; 442 } 443 444 int 445 next_registered_randintr(int intr) 446 { 447 struct intr_info *info; 448 449 if (intr < 0 || intr >= MAX_INTS) 450 panic("register_swi: bad intr %d", intr); 451 while (intr < MAX_INTS) { 452 info = &intr_info_ary[intr]; 453 if (info->i_random.sc_enabled > 0) 454 break; 455 ++intr; 456 } 457 return(intr); 458 } 459 460 /* 461 * Dispatch an interrupt. If there's nothing to do we have a stray 462 * interrupt and can just return, leaving the interrupt masked. 463 * 464 * We need to schedule the interrupt and set its i_running bit. If 465 * we are not on the interrupt thread's cpu we have to send a message 466 * to the correct cpu that will issue the desired action (interlocking 467 * with the interrupt thread's critical section). We do NOT attempt to 468 * reschedule interrupts whos i_running bit is already set because 469 * this would prematurely wakeup a livelock-limited interrupt thread. 470 * 471 * i_running is only tested/set on the same cpu as the interrupt thread. 472 * 473 * We are NOT in a critical section, which will allow the scheduled 474 * interrupt to preempt us. The MP lock might *NOT* be held here. 475 */ 476 #ifdef SMP 477 478 static void 479 sched_ithd_remote(void *arg) 480 { 481 sched_ithd((int)arg); 482 } 483 484 #endif 485 486 void 487 sched_ithd(int intr) 488 { 489 struct intr_info *info; 490 491 info = &intr_info_ary[intr]; 492 493 ++info->i_count; 494 if (info->i_state != ISTATE_NOTHREAD) { 495 if (info->i_reclist == NULL) { 496 report_stray_interrupt(intr, info); 497 } else { 498 #ifdef SMP 499 if (info->i_thread.td_gd == mycpu) { 500 if (info->i_running == 0) { 501 info->i_running = 1; 502 if (info->i_state != ISTATE_LIVELOCKED) 503 lwkt_schedule(&info->i_thread); /* MIGHT PREEMPT */ 504 } 505 } else { 506 lwkt_send_ipiq(info->i_thread.td_gd, 507 sched_ithd_remote, (void *)intr); 508 } 509 #else 510 if (info->i_running == 0) { 511 info->i_running = 1; 512 if (info->i_state != ISTATE_LIVELOCKED) 513 lwkt_schedule(&info->i_thread); /* MIGHT PREEMPT */ 514 } 515 #endif 516 } 517 } else { 518 report_stray_interrupt(intr, info); 519 } 520 } 521 522 static void 523 report_stray_interrupt(int intr, struct intr_info *info) 524 { 525 ++info->i_straycount; 526 if (info->i_straycount < 10) { 527 if (info->i_errorticks == ticks) 528 return; 529 info->i_errorticks = ticks; 530 kprintf("sched_ithd: stray interrupt %d on cpu %d\n", 531 intr, mycpuid); 532 } else if (info->i_straycount == 10) { 533 kprintf("sched_ithd: %ld stray interrupts %d on cpu %d - " 534 "there will be no further reports\n", 535 info->i_straycount, intr, mycpuid); 536 } 537 } 538 539 /* 540 * This is run from a periodic SYSTIMER (and thus must be MP safe, the BGL 541 * might not be held). 542 */ 543 static void 544 ithread_livelock_wakeup(systimer_t st) 545 { 546 struct intr_info *info; 547 548 info = &intr_info_ary[(int)st->data]; 549 if (info->i_state != ISTATE_NOTHREAD) 550 lwkt_schedule(&info->i_thread); 551 } 552 553 /* 554 * This function is called directly from the ICU or APIC vector code assembly 555 * to process an interrupt. The critical section and interrupt deferral 556 * checks have already been done but the function is entered WITHOUT 557 * a critical section held. The BGL may or may not be held. 558 * 559 * Must return non-zero if we do not want the vector code to re-enable 560 * the interrupt (which we don't if we have to schedule the interrupt) 561 */ 562 int ithread_fast_handler(struct intrframe *frame); 563 564 int 565 ithread_fast_handler(struct intrframe *frame) 566 { 567 int intr; 568 struct intr_info *info; 569 struct intrec **list; 570 int must_schedule; 571 #ifdef SMP 572 int got_mplock; 573 #endif 574 intrec_t rec, next_rec; 575 globaldata_t gd; 576 577 intr = frame->if_vec; 578 gd = mycpu; 579 580 info = &intr_info_ary[intr]; 581 582 /* 583 * If we are not processing any FAST interrupts, just schedule the thing. 584 * (since we aren't in a critical section, this can result in a 585 * preemption) 586 * 587 * XXX Protect sched_ithd() call with gd_intr_nesting_level? Interrupts 588 * aren't enabled, but still... 589 */ 590 if (info->i_fast == 0) { 591 ++gd->gd_cnt.v_intr; 592 sched_ithd(intr); 593 return(1); 594 } 595 596 /* 597 * This should not normally occur since interrupts ought to be 598 * masked if the ithread has been scheduled or is running. 599 */ 600 if (info->i_running) 601 return(1); 602 603 /* 604 * Bump the interrupt nesting level to process any FAST interrupts. 605 * Obtain the MP lock as necessary. If the MP lock cannot be obtained, 606 * schedule the interrupt thread to deal with the issue instead. 607 * 608 * To reduce overhead, just leave the MP lock held once it has been 609 * obtained. 610 */ 611 crit_enter_gd(gd); 612 ++gd->gd_intr_nesting_level; 613 ++gd->gd_cnt.v_intr; 614 must_schedule = info->i_slow; 615 #ifdef SMP 616 got_mplock = 0; 617 #endif 618 619 list = &info->i_reclist; 620 for (rec = *list; rec; rec = next_rec) { 621 next_rec = rec->next; /* rec may be invalid after call */ 622 623 if (rec->intr_flags & INTR_FAST) { 624 #ifdef SMP 625 if ((rec->intr_flags & INTR_MPSAFE) == 0 && got_mplock == 0) { 626 if (try_mplock() == 0) { 627 int owner; 628 629 /* 630 * If we couldn't get the MP lock try to forward it 631 * to the cpu holding the MP lock, setting must_schedule 632 * to -1 so we do not schedule and also do not unmask 633 * the interrupt. Otherwise just schedule it. 634 */ 635 owner = owner_mplock(); 636 if (owner >= 0 && owner != gd->gd_cpuid) { 637 lwkt_send_ipiq_bycpu(owner, forward_fastint_remote, 638 (void *)intr); 639 must_schedule = -1; 640 ++gd->gd_cnt.v_forwarded_ints; 641 } else { 642 must_schedule = 1; 643 } 644 break; 645 } 646 got_mplock = 1; 647 } 648 #endif 649 if (rec->serializer) { 650 must_schedule += lwkt_serialize_handler_try( 651 rec->serializer, rec->handler, 652 rec->argument, frame); 653 } else { 654 rec->handler(rec->argument, frame); 655 } 656 } 657 } 658 659 /* 660 * Cleanup 661 */ 662 --gd->gd_intr_nesting_level; 663 #ifdef SMP 664 if (got_mplock) 665 rel_mplock(); 666 #endif 667 crit_exit_gd(gd); 668 669 /* 670 * If we had a problem, schedule the thread to catch the missed 671 * records (it will just re-run all of them). A return value of 0 672 * indicates that all handlers have been run and the interrupt can 673 * be re-enabled, and a non-zero return indicates that the interrupt 674 * thread controls re-enablement. 675 */ 676 if (must_schedule > 0) 677 sched_ithd(intr); 678 else if (must_schedule == 0) 679 ++info->i_count; 680 return(must_schedule); 681 } 682 683 #if 0 684 685 6: ; \ 686 /* could not get the MP lock, forward the interrupt */ \ 687 movl mp_lock, %eax ; /* check race */ \ 688 cmpl $MP_FREE_LOCK,%eax ; \ 689 je 2b ; \ 690 incl PCPU(cnt)+V_FORWARDED_INTS ; \ 691 subl $12,%esp ; \ 692 movl $irq_num,8(%esp) ; \ 693 movl $forward_fastint_remote,4(%esp) ; \ 694 movl %eax,(%esp) ; \ 695 call lwkt_send_ipiq_bycpu ; \ 696 addl $12,%esp ; \ 697 jmp 5f ; 698 699 #endif 700 701 702 /* 703 * Interrupt threads run this as their main loop. 704 * 705 * The handler begins execution outside a critical section and with the BGL 706 * held. 707 * 708 * The i_running state starts at 0. When an interrupt occurs, the hardware 709 * interrupt is disabled and sched_ithd() The HW interrupt remains disabled 710 * until all routines have run. We then call ithread_done() to reenable 711 * the HW interrupt and deschedule us until the next interrupt. 712 * 713 * We are responsible for atomically checking i_running and ithread_done() 714 * is responsible for atomically checking for platform-specific delayed 715 * interrupts. i_running for our irq is only set in the context of our cpu, 716 * so a critical section is a sufficient interlock. 717 */ 718 #define LIVELOCK_TIMEFRAME(freq) ((freq) >> 2) /* 1/4 second */ 719 720 static void 721 ithread_handler(void *arg) 722 { 723 struct intr_info *info; 724 int use_limit; 725 __uint32_t lseconds; 726 int intr; 727 int mpheld; 728 struct intrec **list; 729 intrec_t rec, nrec; 730 globaldata_t gd; 731 struct systimer ill_timer; /* enforced freq. timer */ 732 u_int ill_count; /* interrupt livelock counter */ 733 734 ill_count = 0; 735 intr = (int)arg; 736 info = &intr_info_ary[intr]; 737 list = &info->i_reclist; 738 gd = mycpu; 739 lseconds = gd->gd_time_seconds; 740 741 /* 742 * The loop must be entered with one critical section held. The thread 743 * is created with TDF_MPSAFE so the MP lock is not held on start. 744 */ 745 crit_enter_gd(gd); 746 mpheld = 0; 747 748 for (;;) { 749 /* 750 * The chain is only considered MPSAFE if all its interrupt handlers 751 * are MPSAFE. However, if intr_mpsafe has been turned off we 752 * always operate with the BGL. 753 */ 754 #ifdef SMP 755 if (intr_mpsafe == 0) { 756 if (mpheld == 0) { 757 get_mplock(); 758 mpheld = 1; 759 } 760 } else if (info->i_mplock_required != mpheld) { 761 if (info->i_mplock_required) { 762 KKASSERT(mpheld == 0); 763 get_mplock(); 764 mpheld = 1; 765 } else { 766 KKASSERT(mpheld != 0); 767 rel_mplock(); 768 mpheld = 0; 769 } 770 } 771 #endif 772 773 /* 774 * If an interrupt is pending, clear i_running and execute the 775 * handlers. Note that certain types of interrupts can re-trigger 776 * and set i_running again. 777 * 778 * Each handler is run in a critical section. Note that we run both 779 * FAST and SLOW designated service routines. 780 */ 781 if (info->i_running) { 782 ++ill_count; 783 info->i_running = 0; 784 785 if (*list == NULL) 786 report_stray_interrupt(intr, info); 787 788 for (rec = *list; rec; rec = nrec) { 789 nrec = rec->next; 790 if (rec->serializer) { 791 lwkt_serialize_handler_call(rec->serializer, rec->handler, 792 rec->argument, NULL); 793 } else { 794 rec->handler(rec->argument, NULL); 795 } 796 } 797 } 798 799 /* 800 * This is our interrupt hook to add rate randomness to the random 801 * number generator. 802 */ 803 if (info->i_random.sc_enabled > 0) 804 add_interrupt_randomness(intr); 805 806 /* 807 * Unmask the interrupt to allow it to trigger again. This only 808 * applies to certain types of interrupts (typ level interrupts). 809 * This can result in the interrupt retriggering, but the retrigger 810 * will not be processed until we cycle our critical section. 811 * 812 * Only unmask interrupts while handlers are installed. It is 813 * possible to hit a situation where no handlers are installed 814 * due to a device driver livelocking and then tearing down its 815 * interrupt on close (the parallel bus being a good example). 816 */ 817 if (*list) 818 machintr_intren(intr); 819 820 /* 821 * Do a quick exit/enter to catch any higher-priority interrupt 822 * sources, such as the statclock, so thread time accounting 823 * will still work. This may also cause an interrupt to re-trigger. 824 */ 825 crit_exit_gd(gd); 826 crit_enter_gd(gd); 827 828 /* 829 * LIVELOCK STATE MACHINE 830 */ 831 switch(info->i_state) { 832 case ISTATE_NORMAL: 833 /* 834 * Reset the count each second. 835 */ 836 if (lseconds != gd->gd_time_seconds) { 837 lseconds = gd->gd_time_seconds; 838 ill_count = 0; 839 } 840 841 /* 842 * If we did not exceed the frequency limit, we are done. 843 * If the interrupt has not retriggered we deschedule ourselves. 844 */ 845 if (ill_count <= livelock_limit) { 846 if (info->i_running == 0) { 847 lwkt_deschedule_self(gd->gd_curthread); 848 lwkt_switch(); 849 } 850 break; 851 } 852 853 /* 854 * Otherwise we are livelocked. Set up a periodic systimer 855 * to wake the thread up at the limit frequency. 856 */ 857 kprintf("intr %d at %d/%d hz, livelocked limit engaged!\n", 858 intr, ill_count, livelock_limit); 859 info->i_state = ISTATE_LIVELOCKED; 860 if ((use_limit = livelock_limit) < 100) 861 use_limit = 100; 862 else if (use_limit > 500000) 863 use_limit = 500000; 864 systimer_init_periodic(&ill_timer, ithread_livelock_wakeup, 865 (void *)intr, use_limit); 866 /* fall through */ 867 case ISTATE_LIVELOCKED: 868 /* 869 * Wait for our periodic timer to go off. Since the interrupt 870 * has re-armed it can still set i_running, but it will not 871 * reschedule us while we are in a livelocked state. 872 */ 873 lwkt_deschedule_self(gd->gd_curthread); 874 lwkt_switch(); 875 876 /* 877 * Check once a second to see if the livelock condition no 878 * longer applies. 879 */ 880 if (lseconds != gd->gd_time_seconds) { 881 lseconds = gd->gd_time_seconds; 882 if (ill_count < livelock_lowater) { 883 info->i_state = ISTATE_NORMAL; 884 systimer_del(&ill_timer); 885 kprintf("intr %d at %d/%d hz, livelock removed\n", 886 intr, ill_count, livelock_lowater); 887 } else if (livelock_debug == intr || 888 (bootverbose && cold)) { 889 kprintf("intr %d at %d/%d hz, in livelock\n", 890 intr, ill_count, livelock_lowater); 891 } 892 ill_count = 0; 893 } 894 break; 895 } 896 } 897 /* not reached */ 898 } 899 900 /* 901 * Emergency interrupt polling thread. The thread begins execution 902 * outside a critical section with the BGL held. 903 * 904 * If emergency interrupt polling is enabled, this thread will 905 * execute all system interrupts not marked INTR_NOPOLL at the 906 * specified polling frequency. 907 * 908 * WARNING! This thread runs *ALL* interrupt service routines that 909 * are not marked INTR_NOPOLL, which basically means everything except 910 * the 8254 clock interrupt and the ATA interrupt. It has very high 911 * overhead and should only be used in situations where the machine 912 * cannot otherwise be made to work. Due to the severe performance 913 * degredation, it should not be enabled on production machines. 914 */ 915 static void 916 ithread_emergency(void *arg __unused) 917 { 918 struct intr_info *info; 919 intrec_t rec, nrec; 920 int intr; 921 922 for (;;) { 923 for (intr = 0; intr < max_installed_hard_intr; ++intr) { 924 info = &intr_info_ary[intr]; 925 for (rec = info->i_reclist; rec; rec = nrec) { 926 if ((rec->intr_flags & INTR_NOPOLL) == 0) { 927 if (rec->serializer) { 928 lwkt_serialize_handler_call(rec->serializer, 929 rec->handler, rec->argument, NULL); 930 } else { 931 rec->handler(rec->argument, NULL); 932 } 933 } 934 nrec = rec->next; 935 } 936 } 937 lwkt_deschedule_self(curthread); 938 lwkt_switch(); 939 } 940 } 941 942 /* 943 * Systimer callback - schedule the emergency interrupt poll thread 944 * if emergency polling is enabled. 945 */ 946 static 947 void 948 emergency_intr_timer_callback(systimer_t info, struct intrframe *frame __unused) 949 { 950 if (emergency_intr_enable) 951 lwkt_schedule(info->data); 952 } 953 954 int 955 ithread_cpuid(int intr) 956 { 957 const struct intr_info *info; 958 959 KKASSERT(intr >= 0 && intr < MAX_INTS); 960 info = &intr_info_ary[intr]; 961 962 if (info->i_state == ISTATE_NOTHREAD) 963 return -1; 964 return info->i_thread.td_gd->gd_cpuid; 965 } 966 967 /* 968 * Sysctls used by systat and others: hw.intrnames and hw.intrcnt. 969 * The data for this machine dependent, and the declarations are in machine 970 * dependent code. The layout of intrnames and intrcnt however is machine 971 * independent. 972 * 973 * We do not know the length of intrcnt and intrnames at compile time, so 974 * calculate things at run time. 975 */ 976 977 static int 978 sysctl_intrnames(SYSCTL_HANDLER_ARGS) 979 { 980 struct intr_info *info; 981 intrec_t rec; 982 int error = 0; 983 int len; 984 int intr; 985 char buf[64]; 986 987 for (intr = 0; error == 0 && intr < MAX_INTS; ++intr) { 988 info = &intr_info_ary[intr]; 989 990 len = 0; 991 buf[0] = 0; 992 for (rec = info->i_reclist; rec; rec = rec->next) { 993 ksnprintf(buf + len, sizeof(buf) - len, "%s%s", 994 (len ? "/" : ""), rec->name); 995 len += strlen(buf + len); 996 } 997 if (len == 0) { 998 ksnprintf(buf, sizeof(buf), "irq%d", intr); 999 len = strlen(buf); 1000 } 1001 error = SYSCTL_OUT(req, buf, len + 1); 1002 } 1003 return (error); 1004 } 1005 1006 1007 SYSCTL_PROC(_hw, OID_AUTO, intrnames, CTLTYPE_OPAQUE | CTLFLAG_RD, 1008 NULL, 0, sysctl_intrnames, "", "Interrupt Names"); 1009 1010 static int 1011 sysctl_intrcnt(SYSCTL_HANDLER_ARGS) 1012 { 1013 struct intr_info *info; 1014 int error = 0; 1015 int intr; 1016 1017 for (intr = 0; intr < max_installed_hard_intr; ++intr) { 1018 info = &intr_info_ary[intr]; 1019 1020 error = SYSCTL_OUT(req, &info->i_count, sizeof(info->i_count)); 1021 if (error) 1022 goto failed; 1023 } 1024 for (intr = FIRST_SOFTINT; intr < max_installed_soft_intr; ++intr) { 1025 info = &intr_info_ary[intr]; 1026 1027 error = SYSCTL_OUT(req, &info->i_count, sizeof(info->i_count)); 1028 if (error) 1029 goto failed; 1030 } 1031 failed: 1032 return(error); 1033 } 1034 1035 SYSCTL_PROC(_hw, OID_AUTO, intrcnt, CTLTYPE_OPAQUE | CTLFLAG_RD, 1036 NULL, 0, sysctl_intrcnt, "", "Interrupt Counts"); 1037 1038