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; 64 struct random_softc i_random; 65 int i_running; 66 long i_count; /* interrupts dispatched */ 67 int i_mplock_required; 68 int i_fast; 69 int i_slow; 70 int i_state; 71 int i_errorticks; 72 unsigned long i_straycount; 73 int i_cpuid; 74 int i_intr; 75 }; 76 77 static struct intr_info intr_info_ary[MAXCPU][MAX_INTS]; 78 static struct intr_info *swi_info_ary[MAX_SOFTINTS]; 79 80 static int max_installed_hard_intr[MAXCPU]; 81 82 #define EMERGENCY_INTR_POLLING_FREQ_MAX 20000 83 84 /* 85 * Assert that callers into interrupt handlers don't return with 86 * dangling tokens, spinlocks, or mp locks. 87 */ 88 #ifdef INVARIANTS 89 90 #define TD_INVARIANTS_DECLARE \ 91 int spincount; \ 92 lwkt_tokref_t curstop 93 94 #define TD_INVARIANTS_GET(td) \ 95 do { \ 96 spincount = (td)->td_gd->gd_spinlocks; \ 97 curstop = (td)->td_toks_stop; \ 98 } while(0) 99 100 #define TD_INVARIANTS_TEST(td, name) \ 101 do { \ 102 KASSERT(spincount == (td)->td_gd->gd_spinlocks, \ 103 ("spincount mismatch after interrupt handler %s", \ 104 name)); \ 105 KASSERT(curstop == (td)->td_toks_stop, \ 106 ("token count mismatch after interrupt handler %s", \ 107 name)); \ 108 } while(0) 109 110 #else 111 112 /* !INVARIANTS */ 113 114 #define TD_INVARIANTS_DECLARE 115 #define TD_INVARIANTS_GET(td) 116 #define TD_INVARIANTS_TEST(td, name) 117 118 #endif /* ndef INVARIANTS */ 119 120 static int sysctl_emergency_freq(SYSCTL_HANDLER_ARGS); 121 static int sysctl_emergency_enable(SYSCTL_HANDLER_ARGS); 122 static void emergency_intr_timer_callback(systimer_t, int, struct intrframe *); 123 static void ithread_handler(void *arg); 124 static void ithread_emergency(void *arg); 125 static void report_stray_interrupt(struct intr_info *info, const char *func); 126 static void int_moveto_destcpu(int *, int); 127 static void int_moveto_origcpu(int, int); 128 static void sched_ithd_intern(struct intr_info *info); 129 130 static struct systimer emergency_intr_timer[MAXCPU]; 131 static struct thread emergency_intr_thread[MAXCPU]; 132 133 #define ISTATE_NOTHREAD 0 134 #define ISTATE_NORMAL 1 135 #define ISTATE_LIVELOCKED 2 136 137 static int livelock_limit = 40000; 138 static int livelock_lowater = 20000; 139 static int livelock_debug = -1; 140 SYSCTL_INT(_kern, OID_AUTO, livelock_limit, 141 CTLFLAG_RW, &livelock_limit, 0, "Livelock interrupt rate limit"); 142 SYSCTL_INT(_kern, OID_AUTO, livelock_lowater, 143 CTLFLAG_RW, &livelock_lowater, 0, "Livelock low-water mark restore"); 144 SYSCTL_INT(_kern, OID_AUTO, livelock_debug, 145 CTLFLAG_RW, &livelock_debug, 0, "Livelock debug intr#"); 146 147 static int emergency_intr_enable = 0; /* emergency interrupt polling */ 148 TUNABLE_INT("kern.emergency_intr_enable", &emergency_intr_enable); 149 SYSCTL_PROC(_kern, OID_AUTO, emergency_intr_enable, CTLTYPE_INT | CTLFLAG_RW, 150 0, 0, sysctl_emergency_enable, "I", "Emergency Interrupt Poll Enable"); 151 152 static int emergency_intr_freq = 10; /* emergency polling frequency */ 153 TUNABLE_INT("kern.emergency_intr_freq", &emergency_intr_freq); 154 SYSCTL_PROC(_kern, OID_AUTO, emergency_intr_freq, CTLTYPE_INT | CTLFLAG_RW, 155 0, 0, sysctl_emergency_freq, "I", "Emergency Interrupt Poll Frequency"); 156 157 /* 158 * Sysctl support routines 159 */ 160 static int 161 sysctl_emergency_enable(SYSCTL_HANDLER_ARGS) 162 { 163 int error, enabled, cpuid, freq; 164 165 enabled = emergency_intr_enable; 166 error = sysctl_handle_int(oidp, &enabled, 0, req); 167 if (error || req->newptr == NULL) 168 return error; 169 emergency_intr_enable = enabled; 170 if (emergency_intr_enable) 171 freq = emergency_intr_freq; 172 else 173 freq = 1; 174 175 for (cpuid = 0; cpuid < ncpus; ++cpuid) 176 systimer_adjust_periodic(&emergency_intr_timer[cpuid], freq); 177 return 0; 178 } 179 180 static int 181 sysctl_emergency_freq(SYSCTL_HANDLER_ARGS) 182 { 183 int error, phz, cpuid, freq; 184 185 phz = emergency_intr_freq; 186 error = sysctl_handle_int(oidp, &phz, 0, req); 187 if (error || req->newptr == NULL) 188 return error; 189 if (phz <= 0) 190 return EINVAL; 191 else if (phz > EMERGENCY_INTR_POLLING_FREQ_MAX) 192 phz = EMERGENCY_INTR_POLLING_FREQ_MAX; 193 194 emergency_intr_freq = phz; 195 if (emergency_intr_enable) 196 freq = emergency_intr_freq; 197 else 198 freq = 1; 199 200 for (cpuid = 0; cpuid < ncpus; ++cpuid) 201 systimer_adjust_periodic(&emergency_intr_timer[cpuid], freq); 202 return 0; 203 } 204 205 /* 206 * Register an SWI or INTerrupt handler. 207 */ 208 void * 209 register_swi(int intr, inthand2_t *handler, void *arg, const char *name, 210 struct lwkt_serialize *serializer, int cpuid) 211 { 212 if (intr < FIRST_SOFTINT || intr >= MAX_INTS) 213 panic("register_swi: bad intr %d", intr); 214 215 if (cpuid < 0) 216 cpuid = intr % ncpus; 217 return(register_int(intr, handler, arg, name, serializer, 0, cpuid)); 218 } 219 220 void * 221 register_swi_mp(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, 230 INTR_MPSAFE, cpuid)); 231 } 232 233 void * 234 register_int(int intr, inthand2_t *handler, void *arg, const char *name, 235 struct lwkt_serialize *serializer, int intr_flags, int cpuid) 236 { 237 struct intr_info *info; 238 struct intrec **list; 239 intrec_t rec; 240 int orig_cpuid; 241 242 KKASSERT(cpuid >= 0 && cpuid < ncpus); 243 244 if (intr < 0 || intr >= MAX_INTS) 245 panic("register_int: bad intr %d", intr); 246 if (name == NULL) 247 name = "???"; 248 info = &intr_info_ary[cpuid][intr]; 249 250 /* 251 * Construct an interrupt handler record 252 */ 253 rec = kmalloc(sizeof(struct intrec), M_DEVBUF, M_INTWAIT); 254 rec->name = kmalloc(strlen(name) + 1, M_DEVBUF, M_INTWAIT); 255 strcpy(rec->name, name); 256 257 rec->info = info; 258 rec->handler = handler; 259 rec->argument = arg; 260 rec->intr = intr; 261 rec->intr_flags = intr_flags; 262 rec->next = NULL; 263 rec->serializer = serializer; 264 265 int_moveto_destcpu(&orig_cpuid, cpuid); 266 267 /* 268 * Create an emergency polling thread and set up a systimer to wake 269 * it up. 270 */ 271 if (emergency_intr_thread[cpuid].td_kstack == NULL) { 272 lwkt_create(ithread_emergency, NULL, NULL, 273 &emergency_intr_thread[cpuid], 274 TDF_NOSTART | TDF_INTTHREAD, cpuid, "ithreadE %d", 275 cpuid); 276 systimer_init_periodic_nq(&emergency_intr_timer[cpuid], 277 emergency_intr_timer_callback, 278 &emergency_intr_thread[cpuid], 279 (emergency_intr_enable ? emergency_intr_freq : 1)); 280 } 281 282 /* 283 * Create an interrupt thread if necessary, leave it in an unscheduled 284 * state. 285 */ 286 if (info->i_state == ISTATE_NOTHREAD) { 287 info->i_state = ISTATE_NORMAL; 288 lwkt_create(ithread_handler, (void *)(intptr_t)intr, NULL, 289 &info->i_thread, TDF_NOSTART | TDF_INTTHREAD, cpuid, 290 "ithread%d %d", intr, cpuid); 291 if (intr >= FIRST_SOFTINT) 292 lwkt_setpri(&info->i_thread, TDPRI_SOFT_NORM); 293 else 294 lwkt_setpri(&info->i_thread, TDPRI_INT_MED); 295 info->i_thread.td_preemptable = lwkt_preempt; 296 } 297 298 list = &info->i_reclist; 299 300 /* 301 * Keep track of how many fast and slow interrupts we have. 302 * Set i_mplock_required if any handler in the chain requires 303 * the MP lock to operate. 304 */ 305 if ((intr_flags & INTR_MPSAFE) == 0) 306 info->i_mplock_required = 1; 307 if (intr_flags & INTR_CLOCK) 308 ++info->i_fast; 309 else 310 ++info->i_slow; 311 312 /* 313 * Enable random number generation keying off of this interrupt. 314 */ 315 if ((intr_flags & INTR_NOENTROPY) == 0 && info->i_random.sc_enabled == 0) { 316 info->i_random.sc_enabled = 1; 317 info->i_random.sc_intr = intr; 318 } 319 320 /* 321 * Add the record to the interrupt list. 322 */ 323 crit_enter(); 324 while (*list != NULL) 325 list = &(*list)->next; 326 *list = rec; 327 crit_exit(); 328 329 /* 330 * Update max_installed_hard_intr to make the emergency intr poll 331 * a bit more efficient. 332 */ 333 if (intr < FIRST_SOFTINT) { 334 if (max_installed_hard_intr[cpuid] <= intr) 335 max_installed_hard_intr[cpuid] = intr + 1; 336 } 337 338 if (intr >= FIRST_SOFTINT) 339 swi_info_ary[intr - FIRST_SOFTINT] = info; 340 341 /* 342 * Setup the machine level interrupt vector 343 */ 344 if (intr < FIRST_SOFTINT && info->i_slow + info->i_fast == 1) 345 machintr_intr_setup(intr, intr_flags); 346 347 int_moveto_origcpu(orig_cpuid, cpuid); 348 349 return(rec); 350 } 351 352 void 353 unregister_swi(void *id, int intr, int cpuid) 354 { 355 if (cpuid < 0) 356 cpuid = intr % ncpus; 357 358 unregister_int(id, cpuid); 359 } 360 361 void 362 unregister_int(void *id, int cpuid) 363 { 364 struct intr_info *info; 365 struct intrec **list; 366 intrec_t rec; 367 int intr, orig_cpuid; 368 369 KKASSERT(cpuid >= 0 && cpuid < ncpus); 370 371 intr = ((intrec_t)id)->intr; 372 373 if (intr < 0 || intr >= MAX_INTS) 374 panic("register_int: bad intr %d", intr); 375 376 info = &intr_info_ary[cpuid][intr]; 377 378 int_moveto_destcpu(&orig_cpuid, cpuid); 379 380 /* 381 * Remove the interrupt descriptor, adjust the descriptor count, 382 * and teardown the machine level vector if this was the last interrupt. 383 */ 384 crit_enter(); 385 list = &info->i_reclist; 386 while ((rec = *list) != NULL) { 387 if (rec == id) 388 break; 389 list = &rec->next; 390 } 391 if (rec) { 392 intrec_t rec0; 393 394 *list = rec->next; 395 if (rec->intr_flags & INTR_CLOCK) 396 --info->i_fast; 397 else 398 --info->i_slow; 399 if (intr < FIRST_SOFTINT && info->i_fast + info->i_slow == 0) 400 machintr_intr_teardown(intr); 401 402 /* 403 * Clear i_mplock_required if no handlers in the chain require the 404 * MP lock. 405 */ 406 for (rec0 = info->i_reclist; rec0; rec0 = rec0->next) { 407 if ((rec0->intr_flags & INTR_MPSAFE) == 0) 408 break; 409 } 410 if (rec0 == NULL) 411 info->i_mplock_required = 0; 412 } 413 414 if (intr >= FIRST_SOFTINT && info->i_reclist == NULL) 415 swi_info_ary[intr - FIRST_SOFTINT] = NULL; 416 417 crit_exit(); 418 419 int_moveto_origcpu(orig_cpuid, cpuid); 420 421 /* 422 * Free the record. 423 */ 424 if (rec != NULL) { 425 kfree(rec->name, M_DEVBUF); 426 kfree(rec, M_DEVBUF); 427 } else { 428 kprintf("warning: unregister_int: int %d handler for %s not found\n", 429 intr, ((intrec_t)id)->name); 430 } 431 } 432 433 long 434 get_interrupt_counter(int intr, int cpuid) 435 { 436 struct intr_info *info; 437 438 KKASSERT(cpuid >= 0 && cpuid < ncpus); 439 440 if (intr < 0 || intr >= MAX_INTS) 441 panic("register_int: bad intr %d", intr); 442 info = &intr_info_ary[cpuid][intr]; 443 return(info->i_count); 444 } 445 446 void 447 register_randintr(int intr) 448 { 449 struct intr_info *info; 450 int cpuid; 451 452 if (intr < 0 || intr >= MAX_INTS) 453 panic("register_randintr: bad intr %d", intr); 454 455 for (cpuid = 0; cpuid < ncpus; ++cpuid) { 456 info = &intr_info_ary[cpuid][intr]; 457 info->i_random.sc_intr = intr; 458 info->i_random.sc_enabled = 1; 459 } 460 } 461 462 void 463 unregister_randintr(int intr) 464 { 465 struct intr_info *info; 466 int cpuid; 467 468 if (intr < 0 || intr >= MAX_INTS) 469 panic("register_swi: bad intr %d", intr); 470 471 for (cpuid = 0; cpuid < ncpus; ++cpuid) { 472 info = &intr_info_ary[cpuid][intr]; 473 info->i_random.sc_enabled = -1; 474 } 475 } 476 477 int 478 next_registered_randintr(int intr) 479 { 480 struct intr_info *info; 481 482 if (intr < 0 || intr >= MAX_INTS) 483 panic("register_swi: bad intr %d", intr); 484 485 while (intr < MAX_INTS) { 486 int cpuid; 487 488 for (cpuid = 0; cpuid < ncpus; ++cpuid) { 489 info = &intr_info_ary[cpuid][intr]; 490 if (info->i_random.sc_enabled > 0) 491 return intr; 492 } 493 ++intr; 494 } 495 return intr; 496 } 497 498 /* 499 * Dispatch an interrupt. If there's nothing to do we have a stray 500 * interrupt and can just return, leaving the interrupt masked. 501 * 502 * We need to schedule the interrupt and set its i_running bit. If 503 * we are not on the interrupt thread's cpu we have to send a message 504 * to the correct cpu that will issue the desired action (interlocking 505 * with the interrupt thread's critical section). We do NOT attempt to 506 * reschedule interrupts whos i_running bit is already set because 507 * this would prematurely wakeup a livelock-limited interrupt thread. 508 * 509 * i_running is only tested/set on the same cpu as the interrupt thread. 510 * 511 * We are NOT in a critical section, which will allow the scheduled 512 * interrupt to preempt us. The MP lock might *NOT* be held here. 513 */ 514 static void 515 sched_ithd_remote(void *arg) 516 { 517 sched_ithd_intern(arg); 518 } 519 520 static void 521 sched_ithd_intern(struct intr_info *info) 522 { 523 ++info->i_count; 524 if (info->i_state != ISTATE_NOTHREAD) { 525 if (info->i_reclist == NULL) { 526 report_stray_interrupt(info, "sched_ithd"); 527 } else { 528 if (info->i_thread.td_gd == mycpu) { 529 if (info->i_running == 0) { 530 info->i_running = 1; 531 if (info->i_state != ISTATE_LIVELOCKED) 532 lwkt_schedule(&info->i_thread); /* MIGHT PREEMPT */ 533 } 534 } else { 535 lwkt_send_ipiq(info->i_thread.td_gd, sched_ithd_remote, info); 536 } 537 } 538 } else { 539 report_stray_interrupt(info, "sched_ithd"); 540 } 541 } 542 543 void 544 sched_ithd_soft(int intr) 545 { 546 struct intr_info *info; 547 548 KKASSERT(intr >= FIRST_SOFTINT && intr < MAX_INTS); 549 550 info = swi_info_ary[intr - FIRST_SOFTINT]; 551 if (info != NULL) { 552 sched_ithd_intern(info); 553 } else { 554 kprintf("unregistered softint %d got scheduled on cpu%d\n", 555 intr, mycpuid); 556 } 557 } 558 559 void 560 sched_ithd_hard(int intr) 561 { 562 KKASSERT(intr >= 0 && intr < MAX_HARDINTS); 563 sched_ithd_intern(&intr_info_ary[mycpuid][intr]); 564 } 565 566 #ifdef _KERNEL_VIRTUAL 567 568 void 569 sched_ithd_hard_virtual(int intr) 570 { 571 KKASSERT(intr >= 0 && intr < MAX_HARDINTS); 572 sched_ithd_intern(&intr_info_ary[0][intr]); 573 } 574 575 void * 576 register_int_virtual(int intr, inthand2_t *handler, void *arg, const char *name, 577 struct lwkt_serialize *serializer, int intr_flags) 578 { 579 return register_int(intr, handler, arg, name, serializer, intr_flags, 0); 580 } 581 582 void 583 unregister_int_virtual(void *id) 584 { 585 unregister_int(id, 0); 586 } 587 588 #endif /* _KERN_VIRTUAL */ 589 590 static void 591 report_stray_interrupt(struct intr_info *info, const char *func) 592 { 593 ++info->i_straycount; 594 if (info->i_straycount < 10) { 595 if (info->i_errorticks == ticks) 596 return; 597 info->i_errorticks = ticks; 598 kprintf("%s: stray interrupt %d on cpu%d\n", 599 func, info->i_intr, mycpuid); 600 } else if (info->i_straycount == 10) { 601 kprintf("%s: %ld stray interrupts %d on cpu%d - " 602 "there will be no further reports\n", func, 603 info->i_straycount, info->i_intr, mycpuid); 604 } 605 } 606 607 /* 608 * This is run from a periodic SYSTIMER (and thus must be MP safe, the BGL 609 * might not be held). 610 */ 611 static void 612 ithread_livelock_wakeup(systimer_t st, int in_ipi __unused, 613 struct intrframe *frame __unused) 614 { 615 struct intr_info *info; 616 617 info = &intr_info_ary[mycpuid][(int)(intptr_t)st->data]; 618 if (info->i_state != ISTATE_NOTHREAD) 619 lwkt_schedule(&info->i_thread); 620 } 621 622 /* 623 * Schedule ithread within fast intr handler 624 * 625 * XXX Protect sched_ithd_hard() call with gd_intr_nesting_level? 626 * Interrupts aren't enabled, but still... 627 */ 628 static __inline void 629 ithread_fast_sched(int intr, thread_t td) 630 { 631 ++td->td_nest_count; 632 633 /* 634 * We are already in critical section, exit it now to 635 * allow preemption. 636 */ 637 crit_exit_quick(td); 638 sched_ithd_hard(intr); 639 crit_enter_quick(td); 640 641 --td->td_nest_count; 642 } 643 644 /* 645 * This function is called directly from the ICU or APIC vector code assembly 646 * to process an interrupt. The critical section and interrupt deferral 647 * checks have already been done but the function is entered WITHOUT 648 * a critical section held. The BGL may or may not be held. 649 * 650 * Must return non-zero if we do not want the vector code to re-enable 651 * the interrupt (which we don't if we have to schedule the interrupt) 652 */ 653 int ithread_fast_handler(struct intrframe *frame); 654 655 int 656 ithread_fast_handler(struct intrframe *frame) 657 { 658 int intr; 659 struct intr_info *info; 660 struct intrec **list; 661 int must_schedule; 662 int got_mplock; 663 TD_INVARIANTS_DECLARE; 664 intrec_t rec, nrec; 665 globaldata_t gd; 666 thread_t td; 667 668 intr = frame->if_vec; 669 gd = mycpu; 670 td = curthread; 671 672 /* We must be in critical section. */ 673 KKASSERT(td->td_critcount); 674 675 info = &intr_info_ary[mycpuid][intr]; 676 677 /* 678 * If we are not processing any FAST interrupts, just schedule the thing. 679 */ 680 if (info->i_fast == 0) { 681 ++gd->gd_cnt.v_intr; 682 ithread_fast_sched(intr, td); 683 return(1); 684 } 685 686 /* 687 * This should not normally occur since interrupts ought to be 688 * masked if the ithread has been scheduled or is running. 689 */ 690 if (info->i_running) 691 return(1); 692 693 /* 694 * Bump the interrupt nesting level to process any FAST interrupts. 695 * Obtain the MP lock as necessary. If the MP lock cannot be obtained, 696 * schedule the interrupt thread to deal with the issue instead. 697 * 698 * To reduce overhead, just leave the MP lock held once it has been 699 * obtained. 700 */ 701 ++gd->gd_intr_nesting_level; 702 ++gd->gd_cnt.v_intr; 703 must_schedule = info->i_slow; 704 got_mplock = 0; 705 706 TD_INVARIANTS_GET(td); 707 list = &info->i_reclist; 708 709 for (rec = *list; rec; rec = nrec) { 710 /* rec may be invalid after call */ 711 nrec = rec->next; 712 713 if (rec->intr_flags & INTR_CLOCK) { 714 if ((rec->intr_flags & INTR_MPSAFE) == 0 && got_mplock == 0) { 715 if (try_mplock() == 0) { 716 /* Couldn't get the MP lock; just schedule it. */ 717 must_schedule = 1; 718 break; 719 } 720 got_mplock = 1; 721 } 722 if (rec->serializer) { 723 must_schedule += lwkt_serialize_handler_try( 724 rec->serializer, rec->handler, 725 rec->argument, frame); 726 } else { 727 rec->handler(rec->argument, frame); 728 } 729 TD_INVARIANTS_TEST(td, rec->name); 730 } 731 } 732 733 /* 734 * Cleanup 735 */ 736 --gd->gd_intr_nesting_level; 737 if (got_mplock) 738 rel_mplock(); 739 740 /* 741 * If we had a problem, or mixed fast and slow interrupt handlers are 742 * registered, schedule the ithread to catch the missed records (it 743 * will just re-run all of them). A return value of 0 indicates that 744 * all handlers have been run and the interrupt can be re-enabled, and 745 * a non-zero return indicates that the interrupt thread controls 746 * re-enablement. 747 */ 748 if (must_schedule > 0) 749 ithread_fast_sched(intr, td); 750 else if (must_schedule == 0) 751 ++info->i_count; 752 return(must_schedule); 753 } 754 755 /* 756 * Interrupt threads run this as their main loop. 757 * 758 * The handler begins execution outside a critical section and no MP lock. 759 * 760 * The i_running state starts at 0. When an interrupt occurs, the hardware 761 * interrupt is disabled and sched_ithd_hard() The HW interrupt remains 762 * disabled until all routines have run. We then call ithread_done() to 763 * reenable the HW interrupt and deschedule us until the next interrupt. 764 * 765 * We are responsible for atomically checking i_running and ithread_done() 766 * is responsible for atomically checking for platform-specific delayed 767 * interrupts. i_running for our irq is only set in the context of our cpu, 768 * so a critical section is a sufficient interlock. 769 */ 770 #define LIVELOCK_TIMEFRAME(freq) ((freq) >> 2) /* 1/4 second */ 771 772 static void 773 ithread_handler(void *arg) 774 { 775 struct intr_info *info; 776 int use_limit; 777 __uint32_t lseconds; 778 int intr, cpuid = mycpuid; 779 int mpheld; 780 struct intrec **list; 781 intrec_t rec, nrec; 782 globaldata_t gd; 783 struct systimer ill_timer; /* enforced freq. timer */ 784 u_int ill_count; /* interrupt livelock counter */ 785 TD_INVARIANTS_DECLARE; 786 787 ill_count = 0; 788 intr = (int)(intptr_t)arg; 789 info = &intr_info_ary[cpuid][intr]; 790 list = &info->i_reclist; 791 792 /* 793 * The loop must be entered with one critical section held. The thread 794 * does not hold the mplock on startup. 795 */ 796 gd = mycpu; 797 lseconds = gd->gd_time_seconds; 798 crit_enter_gd(gd); 799 mpheld = 0; 800 801 for (;;) { 802 /* 803 * The chain is only considered MPSAFE if all its interrupt handlers 804 * are MPSAFE. However, if intr_mpsafe has been turned off we 805 * always operate with the BGL. 806 */ 807 if (info->i_mplock_required != mpheld) { 808 if (info->i_mplock_required) { 809 KKASSERT(mpheld == 0); 810 get_mplock(); 811 mpheld = 1; 812 } else { 813 KKASSERT(mpheld != 0); 814 rel_mplock(); 815 mpheld = 0; 816 } 817 } 818 819 TD_INVARIANTS_GET(gd->gd_curthread); 820 821 /* 822 * If an interrupt is pending, clear i_running and execute the 823 * handlers. Note that certain types of interrupts can re-trigger 824 * and set i_running again. 825 * 826 * Each handler is run in a critical section. Note that we run both 827 * FAST and SLOW designated service routines. 828 */ 829 if (info->i_running) { 830 ++ill_count; 831 info->i_running = 0; 832 833 if (*list == NULL) 834 report_stray_interrupt(info, "ithread_handler"); 835 836 for (rec = *list; rec; rec = nrec) { 837 /* rec may be invalid after call */ 838 nrec = rec->next; 839 if (rec->serializer) { 840 lwkt_serialize_handler_call(rec->serializer, rec->handler, 841 rec->argument, NULL); 842 } else { 843 rec->handler(rec->argument, NULL); 844 } 845 TD_INVARIANTS_TEST(gd->gd_curthread, rec->name); 846 } 847 } 848 849 /* 850 * This is our interrupt hook to add rate randomness to the random 851 * number generator. 852 */ 853 if (info->i_random.sc_enabled > 0) 854 add_interrupt_randomness(intr); 855 856 /* 857 * Unmask the interrupt to allow it to trigger again. This only 858 * applies to certain types of interrupts (typ level interrupts). 859 * This can result in the interrupt retriggering, but the retrigger 860 * will not be processed until we cycle our critical section. 861 * 862 * Only unmask interrupts while handlers are installed. It is 863 * possible to hit a situation where no handlers are installed 864 * due to a device driver livelocking and then tearing down its 865 * interrupt on close (the parallel bus being a good example). 866 */ 867 if (intr < FIRST_SOFTINT && *list) 868 machintr_intr_enable(intr); 869 870 /* 871 * Do a quick exit/enter to catch any higher-priority interrupt 872 * sources, such as the statclock, so thread time accounting 873 * will still work. This may also cause an interrupt to re-trigger. 874 */ 875 crit_exit_gd(gd); 876 crit_enter_gd(gd); 877 878 /* 879 * LIVELOCK STATE MACHINE 880 */ 881 switch(info->i_state) { 882 case ISTATE_NORMAL: 883 /* 884 * Reset the count each second. 885 */ 886 if (lseconds != gd->gd_time_seconds) { 887 lseconds = gd->gd_time_seconds; 888 ill_count = 0; 889 } 890 891 /* 892 * If we did not exceed the frequency limit, we are done. 893 * If the interrupt has not retriggered we deschedule ourselves. 894 */ 895 if (ill_count <= livelock_limit) { 896 if (info->i_running == 0) { 897 lwkt_deschedule_self(gd->gd_curthread); 898 lwkt_switch(); 899 } 900 break; 901 } 902 903 /* 904 * Otherwise we are livelocked. Set up a periodic systimer 905 * to wake the thread up at the limit frequency. 906 */ 907 kprintf("intr %d on cpu%d at %d/%d hz, livelocked limit engaged!\n", 908 intr, cpuid, ill_count, livelock_limit); 909 info->i_state = ISTATE_LIVELOCKED; 910 if ((use_limit = livelock_limit) < 100) 911 use_limit = 100; 912 else if (use_limit > 500000) 913 use_limit = 500000; 914 systimer_init_periodic_nq(&ill_timer, ithread_livelock_wakeup, 915 (void *)(intptr_t)intr, use_limit); 916 /* fall through */ 917 case ISTATE_LIVELOCKED: 918 /* 919 * Wait for our periodic timer to go off. Since the interrupt 920 * has re-armed it can still set i_running, but it will not 921 * reschedule us while we are in a livelocked state. 922 */ 923 lwkt_deschedule_self(gd->gd_curthread); 924 lwkt_switch(); 925 926 /* 927 * Check once a second to see if the livelock condition no 928 * longer applies. 929 */ 930 if (lseconds != gd->gd_time_seconds) { 931 lseconds = gd->gd_time_seconds; 932 if (ill_count < livelock_lowater) { 933 info->i_state = ISTATE_NORMAL; 934 systimer_del(&ill_timer); 935 kprintf("intr %d on cpu%d at %d/%d hz, livelock removed\n", 936 intr, cpuid, ill_count, livelock_lowater); 937 } else if (livelock_debug == intr || 938 (bootverbose && cold)) { 939 kprintf("intr %d on cpu%d at %d/%d hz, in livelock\n", 940 intr, cpuid, ill_count, livelock_lowater); 941 } 942 ill_count = 0; 943 } 944 break; 945 } 946 } 947 /* NOT REACHED */ 948 } 949 950 /* 951 * Emergency interrupt polling thread. The thread begins execution 952 * outside a critical section with the BGL held. 953 * 954 * If emergency interrupt polling is enabled, this thread will 955 * execute all system interrupts not marked INTR_NOPOLL at the 956 * specified polling frequency. 957 * 958 * WARNING! This thread runs *ALL* interrupt service routines that 959 * are not marked INTR_NOPOLL, which basically means everything except 960 * the 8254 clock interrupt and the ATA interrupt. It has very high 961 * overhead and should only be used in situations where the machine 962 * cannot otherwise be made to work. Due to the severe performance 963 * degredation, it should not be enabled on production machines. 964 */ 965 static void 966 ithread_emergency(void *arg __unused) 967 { 968 globaldata_t gd = mycpu; 969 struct intr_info *info; 970 intrec_t rec, nrec; 971 int intr, cpuid = mycpuid; 972 TD_INVARIANTS_DECLARE; 973 974 get_mplock(); 975 crit_enter_gd(gd); 976 TD_INVARIANTS_GET(gd->gd_curthread); 977 978 for (;;) { 979 for (intr = 0; intr < max_installed_hard_intr[cpuid]; ++intr) { 980 info = &intr_info_ary[cpuid][intr]; 981 for (rec = info->i_reclist; rec; rec = nrec) { 982 /* rec may be invalid after call */ 983 nrec = rec->next; 984 if ((rec->intr_flags & INTR_NOPOLL) == 0) { 985 if (rec->serializer) { 986 lwkt_serialize_handler_try(rec->serializer, 987 rec->handler, rec->argument, NULL); 988 } else { 989 rec->handler(rec->argument, NULL); 990 } 991 TD_INVARIANTS_TEST(gd->gd_curthread, rec->name); 992 } 993 } 994 } 995 lwkt_deschedule_self(gd->gd_curthread); 996 lwkt_switch(); 997 } 998 /* NOT REACHED */ 999 } 1000 1001 /* 1002 * Systimer callback - schedule the emergency interrupt poll thread 1003 * if emergency polling is enabled. 1004 */ 1005 static 1006 void 1007 emergency_intr_timer_callback(systimer_t info, int in_ipi __unused, 1008 struct intrframe *frame __unused) 1009 { 1010 if (emergency_intr_enable) 1011 lwkt_schedule(info->data); 1012 } 1013 1014 /* 1015 * Sysctls used by systat and others: hw.intrnames and hw.intrcnt. 1016 * The data for this machine dependent, and the declarations are in machine 1017 * dependent code. The layout of intrnames and intrcnt however is machine 1018 * independent. 1019 * 1020 * We do not know the length of intrcnt and intrnames at compile time, so 1021 * calculate things at run time. 1022 */ 1023 1024 static int 1025 sysctl_intrnames(SYSCTL_HANDLER_ARGS) 1026 { 1027 struct intr_info *info; 1028 intrec_t rec; 1029 int error = 0; 1030 int len; 1031 int intr, cpuid; 1032 char buf[64]; 1033 1034 for (cpuid = 0; cpuid < ncpus; ++cpuid) { 1035 for (intr = 0; error == 0 && intr < MAX_INTS; ++intr) { 1036 info = &intr_info_ary[cpuid][intr]; 1037 1038 len = 0; 1039 buf[0] = 0; 1040 for (rec = info->i_reclist; rec; rec = rec->next) { 1041 ksnprintf(buf + len, sizeof(buf) - len, "%s%s", 1042 (len ? "/" : ""), rec->name); 1043 len += strlen(buf + len); 1044 } 1045 if (len == 0) { 1046 ksnprintf(buf, sizeof(buf), "irq%d", intr); 1047 len = strlen(buf); 1048 } 1049 error = SYSCTL_OUT(req, buf, len + 1); 1050 } 1051 } 1052 return (error); 1053 } 1054 1055 SYSCTL_PROC(_hw, OID_AUTO, intrnames, CTLTYPE_OPAQUE | CTLFLAG_RD, 1056 NULL, 0, sysctl_intrnames, "", "Interrupt Names"); 1057 1058 static int 1059 sysctl_intrcnt_all(SYSCTL_HANDLER_ARGS) 1060 { 1061 struct intr_info *info; 1062 int error = 0; 1063 int intr, cpuid; 1064 1065 for (cpuid = 0; cpuid < ncpus; ++cpuid) { 1066 for (intr = 0; intr < MAX_INTS; ++intr) { 1067 info = &intr_info_ary[cpuid][intr]; 1068 1069 error = SYSCTL_OUT(req, &info->i_count, sizeof(info->i_count)); 1070 if (error) 1071 goto failed; 1072 } 1073 } 1074 failed: 1075 return(error); 1076 } 1077 1078 SYSCTL_PROC(_hw, OID_AUTO, intrcnt_all, CTLTYPE_OPAQUE | CTLFLAG_RD, 1079 NULL, 0, sysctl_intrcnt_all, "", "Interrupt Counts"); 1080 1081 SYSCTL_PROC(_hw, OID_AUTO, intrcnt, CTLTYPE_OPAQUE | CTLFLAG_RD, 1082 NULL, 0, sysctl_intrcnt_all, "", "Interrupt Counts"); 1083 1084 static void 1085 int_moveto_destcpu(int *orig_cpuid0, int cpuid) 1086 { 1087 int orig_cpuid = mycpuid; 1088 1089 if (cpuid != orig_cpuid) 1090 lwkt_migratecpu(cpuid); 1091 1092 *orig_cpuid0 = orig_cpuid; 1093 } 1094 1095 static void 1096 int_moveto_origcpu(int orig_cpuid, int cpuid) 1097 { 1098 if (cpuid != orig_cpuid) 1099 lwkt_migratecpu(orig_cpuid); 1100 } 1101 1102 static void 1103 intr_init(void *dummy __unused) 1104 { 1105 int cpuid; 1106 1107 kprintf("Initialize MI interrupts\n"); 1108 1109 for (cpuid = 0; cpuid < ncpus; ++cpuid) { 1110 int intr; 1111 1112 for (intr = 0; intr < MAX_INTS; ++intr) { 1113 struct intr_info *info = &intr_info_ary[cpuid][intr]; 1114 1115 info->i_cpuid = cpuid; 1116 info->i_intr = intr; 1117 } 1118 } 1119 } 1120 SYSINIT(intr_init, SI_BOOT2_FINISH_PIC, SI_ORDER_ANY, intr_init, NULL); 1121