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 * 28 */ 29 30 #include <sys/param.h> 31 #include <sys/systm.h> 32 #include <sys/malloc.h> 33 #include <sys/kernel.h> 34 #include <sys/sysctl.h> 35 #include <sys/thread.h> 36 #include <sys/proc.h> 37 #include <sys/random.h> 38 #include <sys/serialize.h> 39 #include <sys/interrupt.h> 40 #include <sys/bus.h> 41 #include <sys/machintr.h> 42 43 #include <machine/frame.h> 44 45 #include <sys/thread2.h> 46 #include <sys/mplock2.h> 47 48 struct intr_info; 49 50 typedef struct intrec { 51 struct intrec *next; 52 struct intr_info *info; 53 inthand2_t *handler; 54 void *argument; 55 char *name; 56 int intr; 57 int intr_flags; 58 struct lwkt_serialize *serializer; 59 } *intrec_t; 60 61 struct intr_info { 62 intrec_t i_reclist; 63 struct thread *i_thread; /* don't embed struct thread */ 64 struct random_softc i_random; 65 long i_count; /* interrupts dispatched */ 66 int i_running; 67 short i_mplock_required; 68 short i_flags; 69 int i_fast; 70 int i_slow; 71 int i_state; 72 int i_errorticks; 73 unsigned long i_straycount; 74 int i_cpuid; 75 int i_intr; 76 }; 77 78 struct intr_info_block { 79 struct intr_info ary[MAXCPU][MAX_INTS]; 80 }; 81 82 static struct intr_info_block *intr_block; 83 static struct intr_info *swi_info_ary[MAX_SOFTINTS]; 84 85 static int max_installed_hard_intr[MAXCPU]; 86 87 MALLOC_DEFINE(M_INTRMNG, "intrmng", "interrupt management"); 88 89 90 #define EMERGENCY_INTR_POLLING_FREQ_MAX 20000 91 92 /* 93 * Assert that callers into interrupt handlers don't return with 94 * dangling tokens, spinlocks, or mp locks. 95 */ 96 #ifdef INVARIANTS 97 98 #define TD_INVARIANTS_DECLARE \ 99 int spincount; \ 100 lwkt_tokref_t curstop 101 102 #define TD_INVARIANTS_GET(td) \ 103 do { \ 104 spincount = (td)->td_gd->gd_spinlocks; \ 105 curstop = (td)->td_toks_stop; \ 106 } while(0) 107 108 #define TD_INVARIANTS_TEST(td, name) \ 109 do { \ 110 KASSERT(spincount == (td)->td_gd->gd_spinlocks, \ 111 ("spincount mismatch after interrupt handler %s", \ 112 name)); \ 113 KASSERT(curstop == (td)->td_toks_stop, \ 114 ("token count mismatch after interrupt handler %s", \ 115 name)); \ 116 } while(0) 117 118 #else 119 120 /* !INVARIANTS */ 121 122 #define TD_INVARIANTS_DECLARE 123 #define TD_INVARIANTS_GET(td) 124 #define TD_INVARIANTS_TEST(td, name) 125 126 #endif /* ndef INVARIANTS */ 127 128 static int sysctl_emergency_freq(SYSCTL_HANDLER_ARGS); 129 static int sysctl_emergency_enable(SYSCTL_HANDLER_ARGS); 130 static void emergency_intr_timer_callback(systimer_t, int, struct intrframe *); 131 static void ithread_handler(void *arg); 132 static void ithread_emergency(void *arg); 133 static void report_stray_interrupt(struct intr_info *info, const char *func); 134 static void int_moveto_destcpu(int *, int); 135 static void int_moveto_origcpu(int, int); 136 static void sched_ithd_intern(struct intr_info *info); 137 138 static struct systimer emergency_intr_timer[MAXCPU]; 139 static struct thread *emergency_intr_thread[MAXCPU]; 140 141 #define ISTATE_NOTHREAD 0 142 #define ISTATE_NORMAL 1 143 #define ISTATE_LIVELOCKED 2 144 145 static int livelock_limit = 40000; 146 static int livelock_limit_hi = 120000; 147 static int livelock_lowater = 20000; 148 static int livelock_debug = -1; 149 SYSCTL_INT(_kern, OID_AUTO, livelock_limit, 150 CTLFLAG_RW, &livelock_limit, 0, "Livelock interrupt rate limit"); 151 SYSCTL_INT(_kern, OID_AUTO, livelock_limit_hi, 152 CTLFLAG_RW, &livelock_limit_hi, 0, 153 "Livelock interrupt rate limit (high frequency)"); 154 SYSCTL_INT(_kern, OID_AUTO, livelock_lowater, 155 CTLFLAG_RW, &livelock_lowater, 0, "Livelock low-water mark restore"); 156 SYSCTL_INT(_kern, OID_AUTO, livelock_debug, 157 CTLFLAG_RW, &livelock_debug, 0, "Livelock debug intr#"); 158 159 static int emergency_intr_enable = 0; /* emergency interrupt polling */ 160 TUNABLE_INT("kern.emergency_intr_enable", &emergency_intr_enable); 161 SYSCTL_PROC(_kern, OID_AUTO, emergency_intr_enable, CTLTYPE_INT | CTLFLAG_RW, 162 0, 0, sysctl_emergency_enable, "I", "Emergency Interrupt Poll Enable"); 163 164 static int emergency_intr_freq = 10; /* emergency polling frequency */ 165 TUNABLE_INT("kern.emergency_intr_freq", &emergency_intr_freq); 166 SYSCTL_PROC(_kern, OID_AUTO, emergency_intr_freq, CTLTYPE_INT | CTLFLAG_RW, 167 0, 0, sysctl_emergency_freq, "I", "Emergency Interrupt Poll Frequency"); 168 169 /* 170 * Sysctl support routines 171 */ 172 static int 173 sysctl_emergency_enable(SYSCTL_HANDLER_ARGS) 174 { 175 int error, enabled, cpuid, freq; 176 177 enabled = emergency_intr_enable; 178 error = sysctl_handle_int(oidp, &enabled, 0, req); 179 if (error || req->newptr == NULL) 180 return error; 181 emergency_intr_enable = enabled; 182 if (emergency_intr_enable) 183 freq = emergency_intr_freq; 184 else 185 freq = 1; 186 187 for (cpuid = 0; cpuid < ncpus; ++cpuid) 188 systimer_adjust_periodic(&emergency_intr_timer[cpuid], freq); 189 return 0; 190 } 191 192 static int 193 sysctl_emergency_freq(SYSCTL_HANDLER_ARGS) 194 { 195 int error, phz, cpuid, freq; 196 197 phz = emergency_intr_freq; 198 error = sysctl_handle_int(oidp, &phz, 0, req); 199 if (error || req->newptr == NULL) 200 return error; 201 if (phz <= 0) 202 return EINVAL; 203 else if (phz > EMERGENCY_INTR_POLLING_FREQ_MAX) 204 phz = EMERGENCY_INTR_POLLING_FREQ_MAX; 205 206 emergency_intr_freq = phz; 207 if (emergency_intr_enable) 208 freq = emergency_intr_freq; 209 else 210 freq = 1; 211 212 for (cpuid = 0; cpuid < ncpus; ++cpuid) 213 systimer_adjust_periodic(&emergency_intr_timer[cpuid], freq); 214 return 0; 215 } 216 217 /* 218 * Register an SWI or INTerrupt handler. 219 */ 220 void * 221 register_swi(int intr, inthand2_t *handler, void *arg, const char *name, 222 struct lwkt_serialize *serializer, int cpuid) 223 { 224 if (intr < FIRST_SOFTINT || intr >= MAX_INTS) 225 panic("register_swi: bad intr %d", intr); 226 227 if (cpuid < 0) 228 cpuid = intr % ncpus; 229 return(register_int(intr, handler, arg, name, serializer, 0, cpuid)); 230 } 231 232 void * 233 register_swi_mp(int intr, inthand2_t *handler, void *arg, const char *name, 234 struct lwkt_serialize *serializer, int cpuid) 235 { 236 if (intr < FIRST_SOFTINT || intr >= MAX_INTS) 237 panic("register_swi: bad intr %d", intr); 238 239 if (cpuid < 0) 240 cpuid = intr % ncpus; 241 return(register_int(intr, handler, arg, name, serializer, 242 INTR_MPSAFE, cpuid)); 243 } 244 245 void * 246 register_int(int intr, inthand2_t *handler, void *arg, const char *name, 247 struct lwkt_serialize *serializer, int intr_flags, int cpuid) 248 { 249 struct intr_info *info; 250 struct intrec **list; 251 intrec_t rec = NULL; 252 int orig_cpuid; 253 254 KKASSERT(cpuid >= 0 && cpuid < ncpus); 255 256 if (intr < 0 || intr >= MAX_INTS) 257 panic("register_int: bad intr %d", intr); 258 if (name == NULL) 259 name = "???"; 260 info = &intr_block->ary[cpuid][intr]; 261 262 int_moveto_destcpu(&orig_cpuid, cpuid); 263 264 /* 265 * This intr has been registered as exclusive one, so 266 * it can't shared. 267 */ 268 if (info->i_flags & INTR_EXCL) 269 goto done; 270 271 /* 272 * This intr has been registered as shared one, so it 273 * can't be used for exclusive handler. 274 */ 275 list = &info->i_reclist; 276 if ((intr_flags & INTR_EXCL) && *list != NULL) 277 goto done; 278 279 /* 280 * Construct an interrupt handler record 281 */ 282 rec = kmalloc(sizeof(struct intrec), M_DEVBUF, M_INTWAIT); 283 rec->name = kmalloc(strlen(name) + 1, M_DEVBUF, M_INTWAIT); 284 strcpy(rec->name, name); 285 286 rec->info = info; 287 rec->handler = handler; 288 rec->argument = arg; 289 rec->intr = intr; 290 rec->intr_flags = intr_flags; 291 rec->next = NULL; 292 rec->serializer = serializer; 293 294 /* 295 * Create an emergency polling thread and set up a systimer to wake 296 * it up. objcache isn't operational yet so use kmalloc. 297 * 298 * objcache may not be operational yet, use kmalloc(). 299 */ 300 if (emergency_intr_thread[cpuid] == NULL) { 301 emergency_intr_thread[cpuid] = kmalloc(sizeof(struct thread), M_DEVBUF, 302 M_INTWAIT | M_ZERO); 303 lwkt_create(ithread_emergency, NULL, NULL, 304 emergency_intr_thread[cpuid], 305 TDF_NOSTART | TDF_INTTHREAD, cpuid, "ithreadE %d", 306 cpuid); 307 systimer_init_periodic_nq(&emergency_intr_timer[cpuid], 308 emergency_intr_timer_callback, 309 emergency_intr_thread[cpuid], 310 (emergency_intr_enable ? emergency_intr_freq : 1)); 311 } 312 313 /* 314 * Create an interrupt thread if necessary, leave it in an unscheduled 315 * state. 316 */ 317 if (info->i_state == ISTATE_NOTHREAD) { 318 info->i_state = ISTATE_NORMAL; 319 info->i_thread = kmalloc(sizeof(struct thread), M_DEVBUF, 320 M_INTWAIT | M_ZERO); 321 lwkt_create(ithread_handler, (void *)(intptr_t)intr, NULL, 322 info->i_thread, TDF_NOSTART | TDF_INTTHREAD, cpuid, 323 "ithread%d %d", intr, cpuid); 324 if (intr >= FIRST_SOFTINT) 325 lwkt_setpri(info->i_thread, TDPRI_SOFT_NORM); 326 else 327 lwkt_setpri(info->i_thread, TDPRI_INT_MED); 328 info->i_thread->td_preemptable = lwkt_preempt; 329 } 330 331 /* 332 * Keep track of how many fast and slow interrupts we have. 333 * Set i_mplock_required if any handler in the chain requires 334 * the MP lock to operate. 335 */ 336 if ((intr_flags & INTR_MPSAFE) == 0) { 337 info->i_mplock_required = 1; 338 kprintf("interrupt uses mplock: %s\n", name); 339 } 340 if (intr_flags & INTR_CLOCK) 341 ++info->i_fast; 342 else 343 ++info->i_slow; 344 345 info->i_flags |= (intr_flags & INTR_EXCL); 346 if (info->i_slow + info->i_fast == 1 && (intr_flags & INTR_HIFREQ)) { 347 /* 348 * Allow high frequency interrupt, if this intr is not 349 * shared yet. 350 */ 351 info->i_flags |= INTR_HIFREQ; 352 } else { 353 info->i_flags &= ~INTR_HIFREQ; 354 } 355 356 /* 357 * Enable random number generation keying off of this interrupt. 358 */ 359 if ((intr_flags & INTR_NOENTROPY) == 0 && info->i_random.sc_enabled == 0) { 360 info->i_random.sc_enabled = 1; 361 info->i_random.sc_intr = intr; 362 } 363 364 /* 365 * Add the record to the interrupt list. 366 */ 367 crit_enter(); 368 while (*list != NULL) 369 list = &(*list)->next; 370 *list = rec; 371 crit_exit(); 372 373 /* 374 * Update max_installed_hard_intr to make the emergency intr poll 375 * a bit more efficient. 376 */ 377 if (intr < FIRST_SOFTINT) { 378 if (max_installed_hard_intr[cpuid] <= intr) 379 max_installed_hard_intr[cpuid] = intr + 1; 380 } 381 382 if (intr >= FIRST_SOFTINT) 383 swi_info_ary[intr - FIRST_SOFTINT] = info; 384 385 /* 386 * Setup the machine level interrupt vector 387 */ 388 if (intr < FIRST_SOFTINT && info->i_slow + info->i_fast == 1) 389 machintr_intr_setup(intr, intr_flags); 390 391 done: 392 int_moveto_origcpu(orig_cpuid, cpuid); 393 return(rec); 394 } 395 396 void 397 unregister_swi(void *id, int intr, int cpuid) 398 { 399 if (cpuid < 0) 400 cpuid = intr % ncpus; 401 402 unregister_int(id, cpuid); 403 } 404 405 void 406 unregister_int(void *id, int cpuid) 407 { 408 struct intr_info *info; 409 struct intrec **list; 410 intrec_t rec; 411 int intr, orig_cpuid; 412 413 KKASSERT(cpuid >= 0 && cpuid < ncpus); 414 415 intr = ((intrec_t)id)->intr; 416 417 if (intr < 0 || intr >= MAX_INTS) 418 panic("register_int: bad intr %d", intr); 419 420 info = &intr_block->ary[cpuid][intr]; 421 422 int_moveto_destcpu(&orig_cpuid, cpuid); 423 424 /* 425 * Remove the interrupt descriptor, adjust the descriptor count, 426 * and teardown the machine level vector if this was the last interrupt. 427 */ 428 crit_enter(); 429 list = &info->i_reclist; 430 while ((rec = *list) != NULL) { 431 if (rec == id) 432 break; 433 list = &rec->next; 434 } 435 if (rec) { 436 intrec_t rec0; 437 438 *list = rec->next; 439 if (rec->intr_flags & INTR_CLOCK) 440 --info->i_fast; 441 else 442 --info->i_slow; 443 if (intr < FIRST_SOFTINT && info->i_fast + info->i_slow == 0) 444 machintr_intr_teardown(intr); 445 446 /* 447 * Clear i_mplock_required if no handlers in the chain require the 448 * MP lock. 449 */ 450 for (rec0 = info->i_reclist; rec0; rec0 = rec0->next) { 451 if ((rec0->intr_flags & INTR_MPSAFE) == 0) 452 break; 453 } 454 if (rec0 == NULL) 455 info->i_mplock_required = 0; 456 } 457 458 if (info->i_reclist == NULL) { 459 info->i_flags = 0; 460 if (intr >= FIRST_SOFTINT) 461 swi_info_ary[intr - FIRST_SOFTINT] = NULL; 462 } else if (info->i_fast + info->i_slow == 1 && 463 (info->i_reclist->intr_flags & INTR_HIFREQ)) { 464 /* Unshared high frequency interrupt. */ 465 info->i_flags |= INTR_HIFREQ; 466 } 467 468 crit_exit(); 469 470 int_moveto_origcpu(orig_cpuid, cpuid); 471 472 /* 473 * Free the record. 474 */ 475 if (rec != NULL) { 476 kfree(rec->name, M_DEVBUF); 477 kfree(rec, M_DEVBUF); 478 } else { 479 kprintf("warning: unregister_int: int %d handler for %s not found\n", 480 intr, ((intrec_t)id)->name); 481 } 482 } 483 484 long 485 get_interrupt_counter(int intr, int cpuid) 486 { 487 struct intr_info *info; 488 489 KKASSERT(cpuid >= 0 && cpuid < ncpus); 490 491 if (intr < 0 || intr >= MAX_INTS) 492 panic("register_int: bad intr %d", intr); 493 info = &intr_block->ary[cpuid][intr]; 494 return(info->i_count); 495 } 496 497 void 498 register_randintr(int intr) 499 { 500 struct intr_info *info; 501 int cpuid; 502 503 if (intr < 0 || intr >= MAX_INTS) 504 panic("register_randintr: bad intr %d", intr); 505 506 for (cpuid = 0; cpuid < ncpus; ++cpuid) { 507 info = &intr_block->ary[cpuid][intr]; 508 info->i_random.sc_intr = intr; 509 info->i_random.sc_enabled = 1; 510 } 511 } 512 513 void 514 unregister_randintr(int intr) 515 { 516 struct intr_info *info; 517 int cpuid; 518 519 if (intr < 0 || intr >= MAX_INTS) 520 panic("register_swi: bad intr %d", intr); 521 522 for (cpuid = 0; cpuid < ncpus; ++cpuid) { 523 info = &intr_block->ary[cpuid][intr]; 524 info->i_random.sc_enabled = -1; 525 } 526 } 527 528 int 529 next_registered_randintr(int intr) 530 { 531 struct intr_info *info; 532 533 if (intr < 0 || intr >= MAX_INTS) 534 panic("register_swi: bad intr %d", intr); 535 536 while (intr < MAX_INTS) { 537 int cpuid; 538 539 for (cpuid = 0; cpuid < ncpus; ++cpuid) { 540 info = &intr_block->ary[cpuid][intr]; 541 if (info->i_random.sc_enabled > 0) 542 return intr; 543 } 544 ++intr; 545 } 546 return intr; 547 } 548 549 /* 550 * Dispatch an interrupt. If there's nothing to do we have a stray 551 * interrupt and can just return, leaving the interrupt masked. 552 * 553 * We need to schedule the interrupt and set its i_running bit. If 554 * we are not on the interrupt thread's cpu we have to send a message 555 * to the correct cpu that will issue the desired action (interlocking 556 * with the interrupt thread's critical section). We do NOT attempt to 557 * reschedule interrupts whos i_running bit is already set because 558 * this would prematurely wakeup a livelock-limited interrupt thread. 559 * 560 * i_running is only tested/set on the same cpu as the interrupt thread. 561 * 562 * We are NOT in a critical section, which will allow the scheduled 563 * interrupt to preempt us. The MP lock might *NOT* be held here. 564 */ 565 static void 566 sched_ithd_remote(void *arg) 567 { 568 sched_ithd_intern(arg); 569 } 570 571 static void 572 sched_ithd_intern(struct intr_info *info) 573 { 574 ++info->i_count; 575 if (info->i_state != ISTATE_NOTHREAD) { 576 if (info->i_reclist == NULL) { 577 report_stray_interrupt(info, "sched_ithd"); 578 } else { 579 if (info->i_thread->td_gd == mycpu) { 580 if (info->i_running == 0) { 581 info->i_running = 1; 582 if (info->i_state != ISTATE_LIVELOCKED) 583 lwkt_schedule(info->i_thread); /* MIGHT PREEMPT */ 584 } 585 } else { 586 lwkt_send_ipiq(info->i_thread->td_gd, sched_ithd_remote, info); 587 } 588 } 589 } else { 590 report_stray_interrupt(info, "sched_ithd"); 591 } 592 } 593 594 void 595 sched_ithd_soft(int intr) 596 { 597 struct intr_info *info; 598 599 KKASSERT(intr >= FIRST_SOFTINT && intr < MAX_INTS); 600 601 info = swi_info_ary[intr - FIRST_SOFTINT]; 602 if (info != NULL) { 603 sched_ithd_intern(info); 604 } else { 605 kprintf("unregistered softint %d got scheduled on cpu%d\n", 606 intr, mycpuid); 607 } 608 } 609 610 void 611 sched_ithd_hard(int intr) 612 { 613 KKASSERT(intr >= 0 && intr < MAX_HARDINTS); 614 sched_ithd_intern(&intr_block->ary[mycpuid][intr]); 615 } 616 617 #ifdef _KERNEL_VIRTUAL 618 619 void 620 sched_ithd_hard_virtual(int intr) 621 { 622 KKASSERT(intr >= 0 && intr < MAX_HARDINTS); 623 sched_ithd_intern(&intr_block->ary[0][intr]); 624 } 625 626 void * 627 register_int_virtual(int intr, inthand2_t *handler, void *arg, const char *name, 628 struct lwkt_serialize *serializer, int intr_flags) 629 { 630 return register_int(intr, handler, arg, name, serializer, intr_flags, 0); 631 } 632 633 void 634 unregister_int_virtual(void *id) 635 { 636 unregister_int(id, 0); 637 } 638 639 #endif /* _KERN_VIRTUAL */ 640 641 static void 642 report_stray_interrupt(struct intr_info *info, const char *func) 643 { 644 ++info->i_straycount; 645 if (info->i_straycount < 10) { 646 if (info->i_errorticks == ticks) 647 return; 648 info->i_errorticks = ticks; 649 kprintf("%s: stray interrupt %d on cpu%d\n", 650 func, info->i_intr, mycpuid); 651 } else if (info->i_straycount == 10) { 652 kprintf("%s: %ld stray interrupts %d on cpu%d - " 653 "there will be no further reports\n", func, 654 info->i_straycount, info->i_intr, mycpuid); 655 } 656 } 657 658 /* 659 * This is run from a periodic SYSTIMER (and thus must be MP safe, the BGL 660 * might not be held). 661 */ 662 static void 663 ithread_livelock_wakeup(systimer_t st, int in_ipi __unused, 664 struct intrframe *frame __unused) 665 { 666 struct intr_info *info; 667 668 info = &intr_block->ary[mycpuid][(int)(intptr_t)st->data]; 669 if (info->i_state != ISTATE_NOTHREAD) 670 lwkt_schedule(info->i_thread); 671 } 672 673 /* 674 * Schedule ithread within fast intr handler 675 * 676 * Temporarily bump the current thread's td_nest_count to prevent deep 677 * preemptions and splz/doreti stacks. 678 */ 679 static __inline void 680 ithread_fast_sched(int intr, thread_t td) 681 { 682 ++td->td_nest_count; 683 crit_exit_quick(td); 684 sched_ithd_hard(intr); 685 crit_enter_quick(td); 686 --td->td_nest_count; 687 } 688 689 /* 690 * This function is called directly from the ICU or APIC vector code assembly 691 * to process an interrupt. The critical section and interrupt deferral 692 * checks have already been done but the function is entered WITHOUT 693 * a critical section held. The BGL may or may not be held. 694 * 695 * Must return non-zero if we do not want the vector code to re-enable 696 * the interrupt (which we don't if we have to schedule the interrupt) 697 */ 698 int ithread_fast_handler(struct intrframe *frame); 699 700 int 701 ithread_fast_handler(struct intrframe *frame) 702 { 703 int intr; 704 struct intr_info *info; 705 struct intrec **list; 706 int must_schedule; 707 int got_mplock; 708 TD_INVARIANTS_DECLARE; 709 intrec_t rec, nrec; 710 globaldata_t gd; 711 thread_t td; 712 713 intr = frame->if_vec; 714 gd = mycpu; 715 td = curthread; 716 717 /* We must be in critical section. */ 718 KKASSERT(td->td_critcount); 719 720 info = &intr_block->ary[mycpuid][intr]; 721 722 /* 723 * If we are not processing any FAST interrupts, just schedule the thing. 724 */ 725 if (info->i_fast == 0) { 726 ++gd->gd_cnt.v_intr; 727 ithread_fast_sched(intr, td); 728 return(1); 729 } 730 731 /* 732 * This should not normally occur since interrupts ought to be 733 * masked if the ithread has been scheduled or is running. 734 */ 735 if (info->i_running) 736 return(1); 737 738 /* 739 * Bump the interrupt nesting level to process any FAST interrupts. 740 * Obtain the MP lock as necessary. If the MP lock cannot be obtained, 741 * schedule the interrupt thread to deal with the issue instead. 742 * 743 * To reduce overhead, just leave the MP lock held once it has been 744 * obtained. 745 */ 746 ++gd->gd_intr_nesting_level; 747 ++gd->gd_cnt.v_intr; 748 must_schedule = info->i_slow; 749 got_mplock = 0; 750 751 TD_INVARIANTS_GET(td); 752 list = &info->i_reclist; 753 754 for (rec = *list; rec; rec = nrec) { 755 /* rec may be invalid after call */ 756 nrec = rec->next; 757 758 if (rec->intr_flags & INTR_CLOCK) { 759 if ((rec->intr_flags & INTR_MPSAFE) == 0 && got_mplock == 0) { 760 if (try_mplock() == 0) { 761 /* Couldn't get the MP lock; just schedule it. */ 762 must_schedule = 1; 763 break; 764 } 765 got_mplock = 1; 766 } 767 if (rec->serializer) { 768 must_schedule += lwkt_serialize_handler_try( 769 rec->serializer, rec->handler, 770 rec->argument, frame); 771 } else { 772 rec->handler(rec->argument, frame); 773 } 774 TD_INVARIANTS_TEST(td, rec->name); 775 } 776 } 777 778 /* 779 * Cleanup 780 */ 781 --gd->gd_intr_nesting_level; 782 if (got_mplock) 783 rel_mplock(); 784 785 /* 786 * If we had a problem, or mixed fast and slow interrupt handlers are 787 * registered, schedule the ithread to catch the missed records (it 788 * will just re-run all of them). A return value of 0 indicates that 789 * all handlers have been run and the interrupt can be re-enabled, and 790 * a non-zero return indicates that the interrupt thread controls 791 * re-enablement. 792 */ 793 if (must_schedule > 0) 794 ithread_fast_sched(intr, td); 795 else if (must_schedule == 0) 796 ++info->i_count; 797 return(must_schedule); 798 } 799 800 /* 801 * Interrupt threads run this as their main loop. 802 * 803 * The handler begins execution outside a critical section and no MP lock. 804 * 805 * The i_running state starts at 0. When an interrupt occurs, the hardware 806 * interrupt is disabled and sched_ithd_hard(). The HW interrupt remains 807 * disabled until all routines have run. We then call machintr_intr_enable() 808 * to reenable the HW interrupt and deschedule us until the next interrupt. 809 * 810 * We are responsible for atomically checking i_running. i_running for our 811 * irq is only set in the context of our cpu, so a critical section is a 812 * sufficient interlock. 813 */ 814 #define LIVELOCK_TIMEFRAME(freq) ((freq) >> 2) /* 1/4 second */ 815 816 static void 817 ithread_handler(void *arg) 818 { 819 struct intr_info *info; 820 int use_limit; 821 uint32_t lseconds; 822 int intr, cpuid = mycpuid; 823 int mpheld; 824 struct intrec **list; 825 intrec_t rec, nrec; 826 globaldata_t gd; 827 struct systimer ill_timer; /* enforced freq. timer */ 828 u_int ill_count; /* interrupt livelock counter */ 829 int upper_limit; /* interrupt livelock upper limit */ 830 TD_INVARIANTS_DECLARE; 831 832 ill_count = 0; 833 intr = (int)(intptr_t)arg; 834 info = &intr_block->ary[cpuid][intr]; 835 list = &info->i_reclist; 836 837 /* 838 * The loop must be entered with one critical section held. The thread 839 * does not hold the mplock on startup. 840 */ 841 gd = mycpu; 842 lseconds = gd->gd_time_seconds; 843 crit_enter_gd(gd); 844 mpheld = 0; 845 846 for (;;) { 847 /* 848 * The chain is only considered MPSAFE if all its interrupt handlers 849 * are MPSAFE. However, if intr_mpsafe has been turned off we 850 * always operate with the BGL. 851 */ 852 if (info->i_mplock_required != mpheld) { 853 if (info->i_mplock_required) { 854 KKASSERT(mpheld == 0); 855 get_mplock(); 856 mpheld = 1; 857 } else { 858 KKASSERT(mpheld != 0); 859 rel_mplock(); 860 mpheld = 0; 861 } 862 } 863 864 TD_INVARIANTS_GET(gd->gd_curthread); 865 866 /* 867 * If an interrupt is pending, clear i_running and execute the 868 * handlers. Note that certain types of interrupts can re-trigger 869 * and set i_running again. 870 * 871 * Each handler is run in a critical section. Note that we run both 872 * FAST and SLOW designated service routines. 873 */ 874 if (info->i_running) { 875 ++ill_count; 876 info->i_running = 0; 877 878 if (*list == NULL) 879 report_stray_interrupt(info, "ithread_handler"); 880 881 for (rec = *list; rec; rec = nrec) { 882 /* rec may be invalid after call */ 883 nrec = rec->next; 884 if (rec->handler == NULL) { 885 kprintf("NULL HANDLER %s\n", rec->name); 886 } else 887 if (rec->serializer) { 888 lwkt_serialize_handler_call(rec->serializer, rec->handler, 889 rec->argument, NULL); 890 } else { 891 rec->handler(rec->argument, NULL); 892 } 893 TD_INVARIANTS_TEST(gd->gd_curthread, rec->name); 894 } 895 } 896 897 /* 898 * This is our interrupt hook to add rate randomness to the random 899 * number generator. 900 */ 901 if (info->i_random.sc_enabled > 0) 902 add_interrupt_randomness(intr); 903 904 /* 905 * Unmask the interrupt to allow it to trigger again. This only 906 * applies to certain types of interrupts (typ level interrupts). 907 * This can result in the interrupt retriggering, but the retrigger 908 * will not be processed until we cycle our critical section. 909 * 910 * Only unmask interrupts while handlers are installed. It is 911 * possible to hit a situation where no handlers are installed 912 * due to a device driver livelocking and then tearing down its 913 * interrupt on close (the parallel bus being a good example). 914 */ 915 if (intr < FIRST_SOFTINT && *list) 916 machintr_intr_enable(intr); 917 918 /* 919 * Do a quick exit/enter to catch any higher-priority interrupt 920 * sources, such as the statclock, so thread time accounting 921 * will still work. This may also cause an interrupt to re-trigger. 922 */ 923 crit_exit_gd(gd); 924 crit_enter_gd(gd); 925 926 /* 927 * LIVELOCK STATE MACHINE 928 */ 929 switch(info->i_state) { 930 case ISTATE_NORMAL: 931 /* 932 * Reset the count each second. 933 */ 934 if (lseconds != gd->gd_time_seconds) { 935 lseconds = gd->gd_time_seconds; 936 ill_count = 0; 937 } 938 939 /* 940 * If we did not exceed the frequency limit, we are done. 941 * If the interrupt has not retriggered we deschedule ourselves. 942 */ 943 if (info->i_flags & INTR_HIFREQ) 944 upper_limit = livelock_limit_hi; 945 else 946 upper_limit = livelock_limit; 947 if (ill_count <= upper_limit) { 948 if (info->i_running == 0) { 949 lwkt_deschedule_self(gd->gd_curthread); 950 lwkt_switch(); 951 } 952 break; 953 } 954 955 /* 956 * Otherwise we are livelocked. Set up a periodic systimer 957 * to wake the thread up at the limit frequency. 958 */ 959 kprintf("intr %d on cpu%d at %d/%d hz, livelocked limit engaged!\n", 960 intr, cpuid, ill_count, upper_limit); 961 info->i_state = ISTATE_LIVELOCKED; 962 if ((use_limit = upper_limit) < 100) 963 use_limit = 100; 964 else if (use_limit > 500000) 965 use_limit = 500000; 966 systimer_init_periodic_nq(&ill_timer, ithread_livelock_wakeup, 967 (void *)(intptr_t)intr, use_limit); 968 /* fall through */ 969 case ISTATE_LIVELOCKED: 970 /* 971 * Wait for our periodic timer to go off. Since the interrupt 972 * has re-armed it can still set i_running, but it will not 973 * reschedule us while we are in a livelocked state. 974 */ 975 lwkt_deschedule_self(gd->gd_curthread); 976 lwkt_switch(); 977 978 /* 979 * Check once a second to see if the livelock condition no 980 * longer applies. 981 */ 982 if (lseconds != gd->gd_time_seconds) { 983 lseconds = gd->gd_time_seconds; 984 if (ill_count < livelock_lowater) { 985 info->i_state = ISTATE_NORMAL; 986 systimer_del(&ill_timer); 987 kprintf("intr %d on cpu%d at %d/%d hz, livelock removed\n", 988 intr, cpuid, ill_count, livelock_lowater); 989 } else if (livelock_debug == intr || 990 (bootverbose && cold)) { 991 kprintf("intr %d on cpu%d at %d/%d hz, in livelock\n", 992 intr, cpuid, ill_count, livelock_lowater); 993 } 994 ill_count = 0; 995 } 996 break; 997 } 998 } 999 /* NOT REACHED */ 1000 } 1001 1002 /* 1003 * Emergency interrupt polling thread. The thread begins execution 1004 * outside a critical section with the BGL held. 1005 * 1006 * If emergency interrupt polling is enabled, this thread will 1007 * execute all system interrupts not marked INTR_NOPOLL at the 1008 * specified polling frequency. 1009 * 1010 * WARNING! This thread runs *ALL* interrupt service routines that 1011 * are not marked INTR_NOPOLL, which basically means everything except 1012 * the 8254 clock interrupt and the ATA interrupt. It has very high 1013 * overhead and should only be used in situations where the machine 1014 * cannot otherwise be made to work. Due to the severe performance 1015 * degredation, it should not be enabled on production machines. 1016 */ 1017 static void 1018 ithread_emergency(void *arg __unused) 1019 { 1020 globaldata_t gd = mycpu; 1021 struct intr_info *info; 1022 intrec_t rec, nrec; 1023 int intr, cpuid = mycpuid; 1024 TD_INVARIANTS_DECLARE; 1025 1026 get_mplock(); 1027 crit_enter_gd(gd); 1028 TD_INVARIANTS_GET(gd->gd_curthread); 1029 1030 for (;;) { 1031 for (intr = 0; intr < max_installed_hard_intr[cpuid]; ++intr) { 1032 info = &intr_block->ary[cpuid][intr]; 1033 for (rec = info->i_reclist; rec; rec = nrec) { 1034 /* rec may be invalid after call */ 1035 nrec = rec->next; 1036 if ((rec->intr_flags & INTR_NOPOLL) == 0) { 1037 if (rec->serializer) { 1038 lwkt_serialize_handler_try(rec->serializer, 1039 rec->handler, rec->argument, NULL); 1040 } else { 1041 rec->handler(rec->argument, NULL); 1042 } 1043 TD_INVARIANTS_TEST(gd->gd_curthread, rec->name); 1044 } 1045 } 1046 } 1047 lwkt_deschedule_self(gd->gd_curthread); 1048 lwkt_switch(); 1049 } 1050 /* NOT REACHED */ 1051 } 1052 1053 /* 1054 * Systimer callback - schedule the emergency interrupt poll thread 1055 * if emergency polling is enabled. 1056 */ 1057 static 1058 void 1059 emergency_intr_timer_callback(systimer_t info, int in_ipi __unused, 1060 struct intrframe *frame __unused) 1061 { 1062 if (emergency_intr_enable) 1063 lwkt_schedule(info->data); 1064 } 1065 1066 /* 1067 * Sysctls used by systat and others: hw.intrnames and hw.intrcnt. 1068 * The data for this machine dependent, and the declarations are in machine 1069 * dependent code. The layout of intrnames and intrcnt however is machine 1070 * independent. 1071 * 1072 * We do not know the length of intrcnt and intrnames at compile time, so 1073 * calculate things at run time. 1074 */ 1075 1076 static int 1077 sysctl_intrnames(SYSCTL_HANDLER_ARGS) 1078 { 1079 struct intr_info *info; 1080 intrec_t rec; 1081 int error = 0; 1082 int len; 1083 int intr, cpuid; 1084 char buf[64]; 1085 1086 for (cpuid = 0; cpuid < ncpus; ++cpuid) { 1087 for (intr = 0; error == 0 && intr < MAX_INTS; ++intr) { 1088 info = &intr_block->ary[cpuid][intr]; 1089 1090 len = 0; 1091 buf[0] = 0; 1092 for (rec = info->i_reclist; rec; rec = rec->next) { 1093 ksnprintf(buf + len, sizeof(buf) - len, "%s%s", 1094 (len ? "/" : ""), rec->name); 1095 len += strlen(buf + len); 1096 } 1097 if (len == 0) { 1098 ksnprintf(buf, sizeof(buf), "irq%d", intr); 1099 len = strlen(buf); 1100 } 1101 error = SYSCTL_OUT(req, buf, len + 1); 1102 } 1103 } 1104 return (error); 1105 } 1106 1107 SYSCTL_PROC(_hw, OID_AUTO, intrnames, CTLTYPE_OPAQUE | CTLFLAG_RD, 1108 NULL, 0, sysctl_intrnames, "", "Interrupt Names"); 1109 1110 static int 1111 sysctl_intrcnt_all(SYSCTL_HANDLER_ARGS) 1112 { 1113 struct intr_info *info; 1114 int error = 0; 1115 int intr, cpuid; 1116 1117 for (cpuid = 0; cpuid < ncpus; ++cpuid) { 1118 for (intr = 0; intr < MAX_INTS; ++intr) { 1119 info = &intr_block->ary[cpuid][intr]; 1120 1121 error = SYSCTL_OUT(req, &info->i_count, sizeof(info->i_count)); 1122 if (error) 1123 goto failed; 1124 } 1125 } 1126 failed: 1127 return(error); 1128 } 1129 1130 SYSCTL_PROC(_hw, OID_AUTO, intrcnt_all, CTLTYPE_OPAQUE | CTLFLAG_RD, 1131 NULL, 0, sysctl_intrcnt_all, "", "Interrupt Counts"); 1132 1133 SYSCTL_PROC(_hw, OID_AUTO, intrcnt, CTLTYPE_OPAQUE | CTLFLAG_RD, 1134 NULL, 0, sysctl_intrcnt_all, "", "Interrupt Counts"); 1135 1136 static void 1137 int_moveto_destcpu(int *orig_cpuid0, int cpuid) 1138 { 1139 int orig_cpuid = mycpuid; 1140 1141 if (cpuid != orig_cpuid) 1142 lwkt_migratecpu(cpuid); 1143 1144 *orig_cpuid0 = orig_cpuid; 1145 } 1146 1147 static void 1148 int_moveto_origcpu(int orig_cpuid, int cpuid) 1149 { 1150 if (cpuid != orig_cpuid) 1151 lwkt_migratecpu(orig_cpuid); 1152 } 1153 1154 static void 1155 intr_init(void *dummy __unused) 1156 { 1157 int cpuid; 1158 1159 kprintf("Initialize MI interrupts\n"); 1160 1161 intr_block = kmalloc(sizeof(*intr_block), M_INTRMNG, 1162 M_INTWAIT | M_ZERO); 1163 1164 for (cpuid = 0; cpuid < ncpus; ++cpuid) { 1165 int intr; 1166 1167 for (intr = 0; intr < MAX_INTS; ++intr) { 1168 struct intr_info *info = &intr_block->ary[cpuid][intr]; 1169 1170 info->i_cpuid = cpuid; 1171 info->i_intr = intr; 1172 } 1173 } 1174 } 1175 SYSINIT(intr_init, SI_BOOT2_FINISH_PIC, SI_ORDER_ANY, intr_init, NULL); 1176