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 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 info->i_thread = kmalloc(sizeof(struct thread), M_DEVBUF, 289 M_INTWAIT | M_ZERO); 290 lwkt_create(ithread_handler, (void *)(intptr_t)intr, NULL, 291 info->i_thread, TDF_NOSTART | TDF_INTTHREAD, cpuid, 292 "ithread%d %d", intr, cpuid); 293 if (intr >= FIRST_SOFTINT) 294 lwkt_setpri(info->i_thread, TDPRI_SOFT_NORM); 295 else 296 lwkt_setpri(info->i_thread, TDPRI_INT_MED); 297 info->i_thread->td_preemptable = lwkt_preempt; 298 } 299 300 list = &info->i_reclist; 301 302 /* 303 * Keep track of how many fast and slow interrupts we have. 304 * Set i_mplock_required if any handler in the chain requires 305 * the MP lock to operate. 306 */ 307 if ((intr_flags & INTR_MPSAFE) == 0) 308 info->i_mplock_required = 1; 309 if (intr_flags & INTR_CLOCK) 310 ++info->i_fast; 311 else 312 ++info->i_slow; 313 314 /* 315 * Enable random number generation keying off of this interrupt. 316 */ 317 if ((intr_flags & INTR_NOENTROPY) == 0 && info->i_random.sc_enabled == 0) { 318 info->i_random.sc_enabled = 1; 319 info->i_random.sc_intr = intr; 320 } 321 322 /* 323 * Add the record to the interrupt list. 324 */ 325 crit_enter(); 326 while (*list != NULL) 327 list = &(*list)->next; 328 *list = rec; 329 crit_exit(); 330 331 /* 332 * Update max_installed_hard_intr to make the emergency intr poll 333 * a bit more efficient. 334 */ 335 if (intr < FIRST_SOFTINT) { 336 if (max_installed_hard_intr[cpuid] <= intr) 337 max_installed_hard_intr[cpuid] = intr + 1; 338 } 339 340 if (intr >= FIRST_SOFTINT) 341 swi_info_ary[intr - FIRST_SOFTINT] = info; 342 343 /* 344 * Setup the machine level interrupt vector 345 */ 346 if (intr < FIRST_SOFTINT && info->i_slow + info->i_fast == 1) 347 machintr_intr_setup(intr, intr_flags); 348 349 int_moveto_origcpu(orig_cpuid, cpuid); 350 351 return(rec); 352 } 353 354 void 355 unregister_swi(void *id, int intr, int cpuid) 356 { 357 if (cpuid < 0) 358 cpuid = intr % ncpus; 359 360 unregister_int(id, cpuid); 361 } 362 363 void 364 unregister_int(void *id, int cpuid) 365 { 366 struct intr_info *info; 367 struct intrec **list; 368 intrec_t rec; 369 int intr, orig_cpuid; 370 371 KKASSERT(cpuid >= 0 && cpuid < ncpus); 372 373 intr = ((intrec_t)id)->intr; 374 375 if (intr < 0 || intr >= MAX_INTS) 376 panic("register_int: bad intr %d", intr); 377 378 info = &intr_info_ary[cpuid][intr]; 379 380 int_moveto_destcpu(&orig_cpuid, cpuid); 381 382 /* 383 * Remove the interrupt descriptor, adjust the descriptor count, 384 * and teardown the machine level vector if this was the last interrupt. 385 */ 386 crit_enter(); 387 list = &info->i_reclist; 388 while ((rec = *list) != NULL) { 389 if (rec == id) 390 break; 391 list = &rec->next; 392 } 393 if (rec) { 394 intrec_t rec0; 395 396 *list = rec->next; 397 if (rec->intr_flags & INTR_CLOCK) 398 --info->i_fast; 399 else 400 --info->i_slow; 401 if (intr < FIRST_SOFTINT && info->i_fast + info->i_slow == 0) 402 machintr_intr_teardown(intr); 403 404 /* 405 * Clear i_mplock_required if no handlers in the chain require the 406 * MP lock. 407 */ 408 for (rec0 = info->i_reclist; rec0; rec0 = rec0->next) { 409 if ((rec0->intr_flags & INTR_MPSAFE) == 0) 410 break; 411 } 412 if (rec0 == NULL) 413 info->i_mplock_required = 0; 414 } 415 416 if (intr >= FIRST_SOFTINT && info->i_reclist == NULL) 417 swi_info_ary[intr - FIRST_SOFTINT] = NULL; 418 419 crit_exit(); 420 421 int_moveto_origcpu(orig_cpuid, cpuid); 422 423 /* 424 * Free the record. 425 */ 426 if (rec != NULL) { 427 kfree(rec->name, M_DEVBUF); 428 kfree(rec, M_DEVBUF); 429 } else { 430 kprintf("warning: unregister_int: int %d handler for %s not found\n", 431 intr, ((intrec_t)id)->name); 432 } 433 } 434 435 long 436 get_interrupt_counter(int intr, int cpuid) 437 { 438 struct intr_info *info; 439 440 KKASSERT(cpuid >= 0 && cpuid < ncpus); 441 442 if (intr < 0 || intr >= MAX_INTS) 443 panic("register_int: bad intr %d", intr); 444 info = &intr_info_ary[cpuid][intr]; 445 return(info->i_count); 446 } 447 448 void 449 register_randintr(int intr) 450 { 451 struct intr_info *info; 452 int cpuid; 453 454 if (intr < 0 || intr >= MAX_INTS) 455 panic("register_randintr: bad intr %d", intr); 456 457 for (cpuid = 0; cpuid < ncpus; ++cpuid) { 458 info = &intr_info_ary[cpuid][intr]; 459 info->i_random.sc_intr = intr; 460 info->i_random.sc_enabled = 1; 461 } 462 } 463 464 void 465 unregister_randintr(int intr) 466 { 467 struct intr_info *info; 468 int cpuid; 469 470 if (intr < 0 || intr >= MAX_INTS) 471 panic("register_swi: bad intr %d", intr); 472 473 for (cpuid = 0; cpuid < ncpus; ++cpuid) { 474 info = &intr_info_ary[cpuid][intr]; 475 info->i_random.sc_enabled = -1; 476 } 477 } 478 479 int 480 next_registered_randintr(int intr) 481 { 482 struct intr_info *info; 483 484 if (intr < 0 || intr >= MAX_INTS) 485 panic("register_swi: bad intr %d", intr); 486 487 while (intr < MAX_INTS) { 488 int cpuid; 489 490 for (cpuid = 0; cpuid < ncpus; ++cpuid) { 491 info = &intr_info_ary[cpuid][intr]; 492 if (info->i_random.sc_enabled > 0) 493 return intr; 494 } 495 ++intr; 496 } 497 return intr; 498 } 499 500 /* 501 * Dispatch an interrupt. If there's nothing to do we have a stray 502 * interrupt and can just return, leaving the interrupt masked. 503 * 504 * We need to schedule the interrupt and set its i_running bit. If 505 * we are not on the interrupt thread's cpu we have to send a message 506 * to the correct cpu that will issue the desired action (interlocking 507 * with the interrupt thread's critical section). We do NOT attempt to 508 * reschedule interrupts whos i_running bit is already set because 509 * this would prematurely wakeup a livelock-limited interrupt thread. 510 * 511 * i_running is only tested/set on the same cpu as the interrupt thread. 512 * 513 * We are NOT in a critical section, which will allow the scheduled 514 * interrupt to preempt us. The MP lock might *NOT* be held here. 515 */ 516 static void 517 sched_ithd_remote(void *arg) 518 { 519 sched_ithd_intern(arg); 520 } 521 522 static void 523 sched_ithd_intern(struct intr_info *info) 524 { 525 ++info->i_count; 526 if (info->i_state != ISTATE_NOTHREAD) { 527 if (info->i_reclist == NULL) { 528 report_stray_interrupt(info, "sched_ithd"); 529 } else { 530 if (info->i_thread->td_gd == mycpu) { 531 if (info->i_running == 0) { 532 info->i_running = 1; 533 if (info->i_state != ISTATE_LIVELOCKED) 534 lwkt_schedule(info->i_thread); /* MIGHT PREEMPT */ 535 } 536 } else { 537 lwkt_send_ipiq(info->i_thread->td_gd, sched_ithd_remote, info); 538 } 539 } 540 } else { 541 report_stray_interrupt(info, "sched_ithd"); 542 } 543 } 544 545 void 546 sched_ithd_soft(int intr) 547 { 548 struct intr_info *info; 549 550 KKASSERT(intr >= FIRST_SOFTINT && intr < MAX_INTS); 551 552 info = swi_info_ary[intr - FIRST_SOFTINT]; 553 if (info != NULL) { 554 sched_ithd_intern(info); 555 } else { 556 kprintf("unregistered softint %d got scheduled on cpu%d\n", 557 intr, mycpuid); 558 } 559 } 560 561 void 562 sched_ithd_hard(int intr) 563 { 564 KKASSERT(intr >= 0 && intr < MAX_HARDINTS); 565 sched_ithd_intern(&intr_info_ary[mycpuid][intr]); 566 } 567 568 #ifdef _KERNEL_VIRTUAL 569 570 void 571 sched_ithd_hard_virtual(int intr) 572 { 573 KKASSERT(intr >= 0 && intr < MAX_HARDINTS); 574 sched_ithd_intern(&intr_info_ary[0][intr]); 575 } 576 577 void * 578 register_int_virtual(int intr, inthand2_t *handler, void *arg, const char *name, 579 struct lwkt_serialize *serializer, int intr_flags) 580 { 581 return register_int(intr, handler, arg, name, serializer, intr_flags, 0); 582 } 583 584 void 585 unregister_int_virtual(void *id) 586 { 587 unregister_int(id, 0); 588 } 589 590 #endif /* _KERN_VIRTUAL */ 591 592 static void 593 report_stray_interrupt(struct intr_info *info, const char *func) 594 { 595 ++info->i_straycount; 596 if (info->i_straycount < 10) { 597 if (info->i_errorticks == ticks) 598 return; 599 info->i_errorticks = ticks; 600 kprintf("%s: stray interrupt %d on cpu%d\n", 601 func, info->i_intr, mycpuid); 602 } else if (info->i_straycount == 10) { 603 kprintf("%s: %ld stray interrupts %d on cpu%d - " 604 "there will be no further reports\n", func, 605 info->i_straycount, info->i_intr, mycpuid); 606 } 607 } 608 609 /* 610 * This is run from a periodic SYSTIMER (and thus must be MP safe, the BGL 611 * might not be held). 612 */ 613 static void 614 ithread_livelock_wakeup(systimer_t st, int in_ipi __unused, 615 struct intrframe *frame __unused) 616 { 617 struct intr_info *info; 618 619 info = &intr_info_ary[mycpuid][(int)(intptr_t)st->data]; 620 if (info->i_state != ISTATE_NOTHREAD) 621 lwkt_schedule(info->i_thread); 622 } 623 624 /* 625 * Schedule ithread within fast intr handler 626 * 627 * XXX Protect sched_ithd_hard() call with gd_intr_nesting_level? 628 * Interrupts aren't enabled, but still... 629 */ 630 static __inline void 631 ithread_fast_sched(int intr, thread_t td) 632 { 633 ++td->td_nest_count; 634 635 /* 636 * We are already in critical section, exit it now to 637 * allow preemption. 638 */ 639 crit_exit_quick(td); 640 sched_ithd_hard(intr); 641 crit_enter_quick(td); 642 643 --td->td_nest_count; 644 } 645 646 /* 647 * This function is called directly from the ICU or APIC vector code assembly 648 * to process an interrupt. The critical section and interrupt deferral 649 * checks have already been done but the function is entered WITHOUT 650 * a critical section held. The BGL may or may not be held. 651 * 652 * Must return non-zero if we do not want the vector code to re-enable 653 * the interrupt (which we don't if we have to schedule the interrupt) 654 */ 655 int ithread_fast_handler(struct intrframe *frame); 656 657 int 658 ithread_fast_handler(struct intrframe *frame) 659 { 660 int intr; 661 struct intr_info *info; 662 struct intrec **list; 663 int must_schedule; 664 int got_mplock; 665 TD_INVARIANTS_DECLARE; 666 intrec_t rec, nrec; 667 globaldata_t gd; 668 thread_t td; 669 670 intr = frame->if_vec; 671 gd = mycpu; 672 td = curthread; 673 674 /* We must be in critical section. */ 675 KKASSERT(td->td_critcount); 676 677 info = &intr_info_ary[mycpuid][intr]; 678 679 /* 680 * If we are not processing any FAST interrupts, just schedule the thing. 681 */ 682 if (info->i_fast == 0) { 683 ++gd->gd_cnt.v_intr; 684 ithread_fast_sched(intr, td); 685 return(1); 686 } 687 688 /* 689 * This should not normally occur since interrupts ought to be 690 * masked if the ithread has been scheduled or is running. 691 */ 692 if (info->i_running) 693 return(1); 694 695 /* 696 * Bump the interrupt nesting level to process any FAST interrupts. 697 * Obtain the MP lock as necessary. If the MP lock cannot be obtained, 698 * schedule the interrupt thread to deal with the issue instead. 699 * 700 * To reduce overhead, just leave the MP lock held once it has been 701 * obtained. 702 */ 703 ++gd->gd_intr_nesting_level; 704 ++gd->gd_cnt.v_intr; 705 must_schedule = info->i_slow; 706 got_mplock = 0; 707 708 TD_INVARIANTS_GET(td); 709 list = &info->i_reclist; 710 711 for (rec = *list; rec; rec = nrec) { 712 /* rec may be invalid after call */ 713 nrec = rec->next; 714 715 if (rec->intr_flags & INTR_CLOCK) { 716 if ((rec->intr_flags & INTR_MPSAFE) == 0 && got_mplock == 0) { 717 if (try_mplock() == 0) { 718 /* Couldn't get the MP lock; just schedule it. */ 719 must_schedule = 1; 720 break; 721 } 722 got_mplock = 1; 723 } 724 if (rec->serializer) { 725 must_schedule += lwkt_serialize_handler_try( 726 rec->serializer, rec->handler, 727 rec->argument, frame); 728 } else { 729 rec->handler(rec->argument, frame); 730 } 731 TD_INVARIANTS_TEST(td, rec->name); 732 } 733 } 734 735 /* 736 * Cleanup 737 */ 738 --gd->gd_intr_nesting_level; 739 if (got_mplock) 740 rel_mplock(); 741 742 /* 743 * If we had a problem, or mixed fast and slow interrupt handlers are 744 * registered, schedule the ithread to catch the missed records (it 745 * will just re-run all of them). A return value of 0 indicates that 746 * all handlers have been run and the interrupt can be re-enabled, and 747 * a non-zero return indicates that the interrupt thread controls 748 * re-enablement. 749 */ 750 if (must_schedule > 0) 751 ithread_fast_sched(intr, td); 752 else if (must_schedule == 0) 753 ++info->i_count; 754 return(must_schedule); 755 } 756 757 /* 758 * Interrupt threads run this as their main loop. 759 * 760 * The handler begins execution outside a critical section and no MP lock. 761 * 762 * The i_running state starts at 0. When an interrupt occurs, the hardware 763 * interrupt is disabled and sched_ithd_hard(). The HW interrupt remains 764 * disabled until all routines have run. We then call machintr_intr_enable() 765 * to reenable the HW interrupt and deschedule us until the next interrupt. 766 * 767 * We are responsible for atomically checking i_running. i_running for our 768 * irq is only set in the context of our cpu, so a critical section is a 769 * sufficient interlock. 770 */ 771 #define LIVELOCK_TIMEFRAME(freq) ((freq) >> 2) /* 1/4 second */ 772 773 static void 774 ithread_handler(void *arg) 775 { 776 struct intr_info *info; 777 int use_limit; 778 uint32_t lseconds; 779 int intr, cpuid = mycpuid; 780 int mpheld; 781 struct intrec **list; 782 intrec_t rec, nrec; 783 globaldata_t gd; 784 struct systimer ill_timer; /* enforced freq. timer */ 785 u_int ill_count; /* interrupt livelock counter */ 786 TD_INVARIANTS_DECLARE; 787 788 ill_count = 0; 789 intr = (int)(intptr_t)arg; 790 info = &intr_info_ary[cpuid][intr]; 791 list = &info->i_reclist; 792 793 /* 794 * The loop must be entered with one critical section held. The thread 795 * does not hold the mplock on startup. 796 */ 797 gd = mycpu; 798 lseconds = gd->gd_time_seconds; 799 crit_enter_gd(gd); 800 mpheld = 0; 801 802 for (;;) { 803 /* 804 * The chain is only considered MPSAFE if all its interrupt handlers 805 * are MPSAFE. However, if intr_mpsafe has been turned off we 806 * always operate with the BGL. 807 */ 808 if (info->i_mplock_required != mpheld) { 809 if (info->i_mplock_required) { 810 KKASSERT(mpheld == 0); 811 get_mplock(); 812 mpheld = 1; 813 } else { 814 KKASSERT(mpheld != 0); 815 rel_mplock(); 816 mpheld = 0; 817 } 818 } 819 820 TD_INVARIANTS_GET(gd->gd_curthread); 821 822 /* 823 * If an interrupt is pending, clear i_running and execute the 824 * handlers. Note that certain types of interrupts can re-trigger 825 * and set i_running again. 826 * 827 * Each handler is run in a critical section. Note that we run both 828 * FAST and SLOW designated service routines. 829 */ 830 if (info->i_running) { 831 ++ill_count; 832 info->i_running = 0; 833 834 if (*list == NULL) 835 report_stray_interrupt(info, "ithread_handler"); 836 837 for (rec = *list; rec; rec = nrec) { 838 /* rec may be invalid after call */ 839 nrec = rec->next; 840 if (rec->serializer) { 841 lwkt_serialize_handler_call(rec->serializer, rec->handler, 842 rec->argument, NULL); 843 } else { 844 rec->handler(rec->argument, NULL); 845 } 846 TD_INVARIANTS_TEST(gd->gd_curthread, rec->name); 847 } 848 } 849 850 /* 851 * This is our interrupt hook to add rate randomness to the random 852 * number generator. 853 */ 854 if (info->i_random.sc_enabled > 0) 855 add_interrupt_randomness(intr); 856 857 /* 858 * Unmask the interrupt to allow it to trigger again. This only 859 * applies to certain types of interrupts (typ level interrupts). 860 * This can result in the interrupt retriggering, but the retrigger 861 * will not be processed until we cycle our critical section. 862 * 863 * Only unmask interrupts while handlers are installed. It is 864 * possible to hit a situation where no handlers are installed 865 * due to a device driver livelocking and then tearing down its 866 * interrupt on close (the parallel bus being a good example). 867 */ 868 if (intr < FIRST_SOFTINT && *list) 869 machintr_intr_enable(intr); 870 871 /* 872 * Do a quick exit/enter to catch any higher-priority interrupt 873 * sources, such as the statclock, so thread time accounting 874 * will still work. This may also cause an interrupt to re-trigger. 875 */ 876 crit_exit_gd(gd); 877 crit_enter_gd(gd); 878 879 /* 880 * LIVELOCK STATE MACHINE 881 */ 882 switch(info->i_state) { 883 case ISTATE_NORMAL: 884 /* 885 * Reset the count each second. 886 */ 887 if (lseconds != gd->gd_time_seconds) { 888 lseconds = gd->gd_time_seconds; 889 ill_count = 0; 890 } 891 892 /* 893 * If we did not exceed the frequency limit, we are done. 894 * If the interrupt has not retriggered we deschedule ourselves. 895 */ 896 if (ill_count <= livelock_limit) { 897 if (info->i_running == 0) { 898 lwkt_deschedule_self(gd->gd_curthread); 899 lwkt_switch(); 900 } 901 break; 902 } 903 904 /* 905 * Otherwise we are livelocked. Set up a periodic systimer 906 * to wake the thread up at the limit frequency. 907 */ 908 kprintf("intr %d on cpu%d at %d/%d hz, livelocked limit engaged!\n", 909 intr, cpuid, ill_count, livelock_limit); 910 info->i_state = ISTATE_LIVELOCKED; 911 if ((use_limit = livelock_limit) < 100) 912 use_limit = 100; 913 else if (use_limit > 500000) 914 use_limit = 500000; 915 systimer_init_periodic_nq(&ill_timer, ithread_livelock_wakeup, 916 (void *)(intptr_t)intr, use_limit); 917 /* fall through */ 918 case ISTATE_LIVELOCKED: 919 /* 920 * Wait for our periodic timer to go off. Since the interrupt 921 * has re-armed it can still set i_running, but it will not 922 * reschedule us while we are in a livelocked state. 923 */ 924 lwkt_deschedule_self(gd->gd_curthread); 925 lwkt_switch(); 926 927 /* 928 * Check once a second to see if the livelock condition no 929 * longer applies. 930 */ 931 if (lseconds != gd->gd_time_seconds) { 932 lseconds = gd->gd_time_seconds; 933 if (ill_count < livelock_lowater) { 934 info->i_state = ISTATE_NORMAL; 935 systimer_del(&ill_timer); 936 kprintf("intr %d on cpu%d at %d/%d hz, livelock removed\n", 937 intr, cpuid, ill_count, livelock_lowater); 938 } else if (livelock_debug == intr || 939 (bootverbose && cold)) { 940 kprintf("intr %d on cpu%d at %d/%d hz, in livelock\n", 941 intr, cpuid, ill_count, livelock_lowater); 942 } 943 ill_count = 0; 944 } 945 break; 946 } 947 } 948 /* NOT REACHED */ 949 } 950 951 /* 952 * Emergency interrupt polling thread. The thread begins execution 953 * outside a critical section with the BGL held. 954 * 955 * If emergency interrupt polling is enabled, this thread will 956 * execute all system interrupts not marked INTR_NOPOLL at the 957 * specified polling frequency. 958 * 959 * WARNING! This thread runs *ALL* interrupt service routines that 960 * are not marked INTR_NOPOLL, which basically means everything except 961 * the 8254 clock interrupt and the ATA interrupt. It has very high 962 * overhead and should only be used in situations where the machine 963 * cannot otherwise be made to work. Due to the severe performance 964 * degredation, it should not be enabled on production machines. 965 */ 966 static void 967 ithread_emergency(void *arg __unused) 968 { 969 globaldata_t gd = mycpu; 970 struct intr_info *info; 971 intrec_t rec, nrec; 972 int intr, cpuid = mycpuid; 973 TD_INVARIANTS_DECLARE; 974 975 get_mplock(); 976 crit_enter_gd(gd); 977 TD_INVARIANTS_GET(gd->gd_curthread); 978 979 for (;;) { 980 for (intr = 0; intr < max_installed_hard_intr[cpuid]; ++intr) { 981 info = &intr_info_ary[cpuid][intr]; 982 for (rec = info->i_reclist; rec; rec = nrec) { 983 /* rec may be invalid after call */ 984 nrec = rec->next; 985 if ((rec->intr_flags & INTR_NOPOLL) == 0) { 986 if (rec->serializer) { 987 lwkt_serialize_handler_try(rec->serializer, 988 rec->handler, rec->argument, NULL); 989 } else { 990 rec->handler(rec->argument, NULL); 991 } 992 TD_INVARIANTS_TEST(gd->gd_curthread, rec->name); 993 } 994 } 995 } 996 lwkt_deschedule_self(gd->gd_curthread); 997 lwkt_switch(); 998 } 999 /* NOT REACHED */ 1000 } 1001 1002 /* 1003 * Systimer callback - schedule the emergency interrupt poll thread 1004 * if emergency polling is enabled. 1005 */ 1006 static 1007 void 1008 emergency_intr_timer_callback(systimer_t info, int in_ipi __unused, 1009 struct intrframe *frame __unused) 1010 { 1011 if (emergency_intr_enable) 1012 lwkt_schedule(info->data); 1013 } 1014 1015 /* 1016 * Sysctls used by systat and others: hw.intrnames and hw.intrcnt. 1017 * The data for this machine dependent, and the declarations are in machine 1018 * dependent code. The layout of intrnames and intrcnt however is machine 1019 * independent. 1020 * 1021 * We do not know the length of intrcnt and intrnames at compile time, so 1022 * calculate things at run time. 1023 */ 1024 1025 static int 1026 sysctl_intrnames(SYSCTL_HANDLER_ARGS) 1027 { 1028 struct intr_info *info; 1029 intrec_t rec; 1030 int error = 0; 1031 int len; 1032 int intr, cpuid; 1033 char buf[64]; 1034 1035 for (cpuid = 0; cpuid < ncpus; ++cpuid) { 1036 for (intr = 0; error == 0 && intr < MAX_INTS; ++intr) { 1037 info = &intr_info_ary[cpuid][intr]; 1038 1039 len = 0; 1040 buf[0] = 0; 1041 for (rec = info->i_reclist; rec; rec = rec->next) { 1042 ksnprintf(buf + len, sizeof(buf) - len, "%s%s", 1043 (len ? "/" : ""), rec->name); 1044 len += strlen(buf + len); 1045 } 1046 if (len == 0) { 1047 ksnprintf(buf, sizeof(buf), "irq%d", intr); 1048 len = strlen(buf); 1049 } 1050 error = SYSCTL_OUT(req, buf, len + 1); 1051 } 1052 } 1053 return (error); 1054 } 1055 1056 SYSCTL_PROC(_hw, OID_AUTO, intrnames, CTLTYPE_OPAQUE | CTLFLAG_RD, 1057 NULL, 0, sysctl_intrnames, "", "Interrupt Names"); 1058 1059 static int 1060 sysctl_intrcnt_all(SYSCTL_HANDLER_ARGS) 1061 { 1062 struct intr_info *info; 1063 int error = 0; 1064 int intr, cpuid; 1065 1066 for (cpuid = 0; cpuid < ncpus; ++cpuid) { 1067 for (intr = 0; intr < MAX_INTS; ++intr) { 1068 info = &intr_info_ary[cpuid][intr]; 1069 1070 error = SYSCTL_OUT(req, &info->i_count, sizeof(info->i_count)); 1071 if (error) 1072 goto failed; 1073 } 1074 } 1075 failed: 1076 return(error); 1077 } 1078 1079 SYSCTL_PROC(_hw, OID_AUTO, intrcnt_all, CTLTYPE_OPAQUE | CTLFLAG_RD, 1080 NULL, 0, sysctl_intrcnt_all, "", "Interrupt Counts"); 1081 1082 SYSCTL_PROC(_hw, OID_AUTO, intrcnt, CTLTYPE_OPAQUE | CTLFLAG_RD, 1083 NULL, 0, sysctl_intrcnt_all, "", "Interrupt Counts"); 1084 1085 static void 1086 int_moveto_destcpu(int *orig_cpuid0, int cpuid) 1087 { 1088 int orig_cpuid = mycpuid; 1089 1090 if (cpuid != orig_cpuid) 1091 lwkt_migratecpu(cpuid); 1092 1093 *orig_cpuid0 = orig_cpuid; 1094 } 1095 1096 static void 1097 int_moveto_origcpu(int orig_cpuid, int cpuid) 1098 { 1099 if (cpuid != orig_cpuid) 1100 lwkt_migratecpu(orig_cpuid); 1101 } 1102 1103 static void 1104 intr_init(void *dummy __unused) 1105 { 1106 int cpuid; 1107 1108 kprintf("Initialize MI interrupts\n"); 1109 1110 for (cpuid = 0; cpuid < ncpus; ++cpuid) { 1111 int intr; 1112 1113 for (intr = 0; intr < MAX_INTS; ++intr) { 1114 struct intr_info *info = &intr_info_ary[cpuid][intr]; 1115 1116 info->i_cpuid = cpuid; 1117 info->i_intr = intr; 1118 } 1119 } 1120 } 1121 SYSINIT(intr_init, SI_BOOT2_FINISH_PIC, SI_ORDER_ANY, intr_init, NULL); 1122