| 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 info_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 | } intr_info_ary[MAX_INTS]; |
| 74 | |
| 75 | int max_installed_hard_intr; |
| 76 | int max_installed_soft_intr; |
| 77 | |
| 78 | #define EMERGENCY_INTR_POLLING_FREQ_MAX 20000 |
| 79 | |
| 80 | /* |
| 81 | * Assert that callers into interrupt handlers don't return with |
| 82 | * dangling tokens, spinlocks, or mp locks. |
| 83 | */ |
| 84 | #ifdef INVARIANTS |
| 85 | |
| 86 | #define TD_INVARIANTS_DECLARE \ |
| 87 | int spincount; \ |
| 88 | lwkt_tokref_t curstop |
| 89 | |
| 90 | #define TD_INVARIANTS_GET(td) \ |
| 91 | do { \ |
| 92 | spincount = (td)->td_gd->gd_spinlocks_wr; \ |
| 93 | curstop = (td)->td_toks_stop; \ |
| 94 | } while(0) |
| 95 | |
| 96 | #define TD_INVARIANTS_TEST(td, name) \ |
| 97 | do { \ |
| 98 | KASSERT(spincount == (td)->td_gd->gd_spinlocks_wr, \ |
| 99 | ("spincount mismatch after interrupt handler %s", \ |
| 100 | name)); \ |
| 101 | KASSERT(curstop == (td)->td_toks_stop, \ |
| 102 | ("token count mismatch after interrupt handler %s", \ |
| 103 | name)); \ |
| 104 | } while(0) |
| 105 | |
| 106 | #else |
| 107 | |
| 108 | /* !INVARIANTS */ |
| 109 | |
| 110 | #define TD_INVARIANTS_DECLARE |
| 111 | #define TD_INVARIANTS_GET(td) |
| 112 | #define TD_INVARIANTS_TEST(td, name) |
| 113 | |
| 114 | #endif /* ndef INVARIANTS */ |
| 115 | |
| 116 | static int sysctl_emergency_freq(SYSCTL_HANDLER_ARGS); |
| 117 | static int sysctl_emergency_enable(SYSCTL_HANDLER_ARGS); |
| 118 | static void emergency_intr_timer_callback(systimer_t, int, struct intrframe *); |
| 119 | static void ithread_handler(void *arg); |
| 120 | static void ithread_emergency(void *arg); |
| 121 | static void report_stray_interrupt(int intr, struct intr_info *info); |
| 122 | static void int_moveto_destcpu(int *, int); |
| 123 | static void int_moveto_origcpu(int, int); |
| 124 | |
| 125 | int intr_info_size = NELEM(intr_info_ary); |
| 126 | |
| 127 | static struct systimer emergency_intr_timer; |
| 128 | static struct thread emergency_intr_thread; |
| 129 | |
| 130 | #define ISTATE_NOTHREAD 0 |
| 131 | #define ISTATE_NORMAL 1 |
| 132 | #define ISTATE_LIVELOCKED 2 |
| 133 | |
| 134 | static int livelock_limit = 40000; |
| 135 | static int livelock_lowater = 20000; |
| 136 | static int livelock_debug = -1; |
| 137 | SYSCTL_INT(_kern, OID_AUTO, livelock_limit, |
| 138 | CTLFLAG_RW, &livelock_limit, 0, "Livelock interrupt rate limit"); |
| 139 | SYSCTL_INT(_kern, OID_AUTO, livelock_lowater, |
| 140 | CTLFLAG_RW, &livelock_lowater, 0, "Livelock low-water mark restore"); |
| 141 | SYSCTL_INT(_kern, OID_AUTO, livelock_debug, |
| 142 | CTLFLAG_RW, &livelock_debug, 0, "Livelock debug intr#"); |
| 143 | |
| 144 | static int emergency_intr_enable = 0; /* emergency interrupt polling */ |
| 145 | TUNABLE_INT("kern.emergency_intr_enable", &emergency_intr_enable); |
| 146 | SYSCTL_PROC(_kern, OID_AUTO, emergency_intr_enable, CTLTYPE_INT | CTLFLAG_RW, |
| 147 | 0, 0, sysctl_emergency_enable, "I", "Emergency Interrupt Poll Enable"); |
| 148 | |
| 149 | static int emergency_intr_freq = 10; /* emergency polling frequency */ |
| 150 | TUNABLE_INT("kern.emergency_intr_freq", &emergency_intr_freq); |
| 151 | SYSCTL_PROC(_kern, OID_AUTO, emergency_intr_freq, CTLTYPE_INT | CTLFLAG_RW, |
| 152 | 0, 0, sysctl_emergency_freq, "I", "Emergency Interrupt Poll Frequency"); |
| 153 | |
| 154 | /* |
| 155 | * Sysctl support routines |
| 156 | */ |
| 157 | static int |
| 158 | sysctl_emergency_enable(SYSCTL_HANDLER_ARGS) |
| 159 | { |
| 160 | int error, enabled; |
| 161 | |
| 162 | enabled = emergency_intr_enable; |
| 163 | error = sysctl_handle_int(oidp, &enabled, 0, req); |
| 164 | if (error || req->newptr == NULL) |
| 165 | return error; |
| 166 | emergency_intr_enable = enabled; |
| 167 | if (emergency_intr_enable) { |
| 168 | systimer_adjust_periodic(&emergency_intr_timer, |
| 169 | emergency_intr_freq); |
| 170 | } else { |
| 171 | systimer_adjust_periodic(&emergency_intr_timer, 1); |
| 172 | } |
| 173 | return 0; |
| 174 | } |
| 175 | |
| 176 | static int |
| 177 | sysctl_emergency_freq(SYSCTL_HANDLER_ARGS) |
| 178 | { |
| 179 | int error, phz; |
| 180 | |
| 181 | phz = emergency_intr_freq; |
| 182 | error = sysctl_handle_int(oidp, &phz, 0, req); |
| 183 | if (error || req->newptr == NULL) |
| 184 | return error; |
| 185 | if (phz <= 0) |
| 186 | return EINVAL; |
| 187 | else if (phz > EMERGENCY_INTR_POLLING_FREQ_MAX) |
| 188 | phz = EMERGENCY_INTR_POLLING_FREQ_MAX; |
| 189 | |
| 190 | emergency_intr_freq = phz; |
| 191 | if (emergency_intr_enable) { |
| 192 | systimer_adjust_periodic(&emergency_intr_timer, |
| 193 | emergency_intr_freq); |
| 194 | } else { |
| 195 | systimer_adjust_periodic(&emergency_intr_timer, 1); |
| 196 | } |
| 197 | return 0; |
| 198 | } |
| 199 | |
| 200 | /* |
| 201 | * Register an SWI or INTerrupt handler. |
| 202 | */ |
| 203 | void * |
| 204 | register_swi(int intr, inthand2_t *handler, void *arg, const char *name, |
| 205 | struct lwkt_serialize *serializer) |
| 206 | { |
| 207 | if (intr < FIRST_SOFTINT || intr >= MAX_INTS) |
| 208 | panic("register_swi: bad intr %d", intr); |
| 209 | return(register_int(intr, handler, arg, name, serializer, 0, 0)); |
| 210 | } |
| 211 | |
| 212 | void * |
| 213 | register_swi_mp(int intr, inthand2_t *handler, void *arg, const char *name, |
| 214 | struct lwkt_serialize *serializer) |
| 215 | { |
| 216 | if (intr < FIRST_SOFTINT || intr >= MAX_INTS) |
| 217 | panic("register_swi: bad intr %d", intr); |
| 218 | return(register_int(intr, handler, arg, name, serializer, INTR_MPSAFE, 0)); |
| 219 | } |
| 220 | |
| 221 | void * |
| 222 | register_int(int intr, inthand2_t *handler, void *arg, const char *name, |
| 223 | struct lwkt_serialize *serializer, int intr_flags, int cpuid) |
| 224 | { |
| 225 | struct intr_info *info; |
| 226 | struct intrec **list; |
| 227 | intrec_t rec; |
| 228 | int orig_cpuid; |
| 229 | |
| 230 | KKASSERT(cpuid >= 0 && cpuid < ncpus); |
| 231 | |
| 232 | if (intr < 0 || intr >= MAX_INTS) |
| 233 | panic("register_int: bad intr %d", intr); |
| 234 | if (name == NULL) |
| 235 | name = "???"; |
| 236 | info = &intr_info_ary[intr]; |
| 237 | |
| 238 | /* |
| 239 | * Construct an interrupt handler record |
| 240 | */ |
| 241 | rec = kmalloc(sizeof(struct intrec), M_DEVBUF, M_INTWAIT); |
| 242 | rec->name = kmalloc(strlen(name) + 1, M_DEVBUF, M_INTWAIT); |
| 243 | strcpy(rec->name, name); |
| 244 | |
| 245 | rec->info = info; |
| 246 | rec->handler = handler; |
| 247 | rec->argument = arg; |
| 248 | rec->intr = intr; |
| 249 | rec->intr_flags = intr_flags; |
| 250 | rec->next = NULL; |
| 251 | rec->serializer = serializer; |
| 252 | |
| 253 | /* |
| 254 | * Create an emergency polling thread and set up a systimer to wake |
| 255 | * it up. |
| 256 | */ |
| 257 | if (emergency_intr_thread.td_kstack == NULL) { |
| 258 | lwkt_create(ithread_emergency, NULL, NULL, &emergency_intr_thread, |
| 259 | TDF_STOPREQ | TDF_INTTHREAD, ncpus - 1, "ithread emerg"); |
| 260 | systimer_init_periodic_nq(&emergency_intr_timer, |
| 261 | emergency_intr_timer_callback, &emergency_intr_thread, |
| 262 | (emergency_intr_enable ? emergency_intr_freq : 1)); |
| 263 | } |
| 264 | |
| 265 | int_moveto_destcpu(&orig_cpuid, cpuid); |
| 266 | |
| 267 | /* |
| 268 | * Create an interrupt thread if necessary, leave it in an unscheduled |
| 269 | * state. |
| 270 | */ |
| 271 | if (info->i_state == ISTATE_NOTHREAD) { |
| 272 | info->i_state = ISTATE_NORMAL; |
| 273 | lwkt_create(ithread_handler, (void *)(intptr_t)intr, NULL, |
| 274 | &info->i_thread, TDF_STOPREQ | TDF_INTTHREAD, cpuid, |
| 275 | "ithread %d", intr); |
| 276 | if (intr >= FIRST_SOFTINT) |
| 277 | lwkt_setpri(&info->i_thread, TDPRI_SOFT_NORM); |
| 278 | else |
| 279 | lwkt_setpri(&info->i_thread, TDPRI_INT_MED); |
| 280 | info->i_thread.td_preemptable = lwkt_preempt; |
| 281 | } |
| 282 | |
| 283 | list = &info->i_reclist; |
| 284 | |
| 285 | /* |
| 286 | * Keep track of how many fast and slow interrupts we have. |
| 287 | * Set i_mplock_required if any handler in the chain requires |
| 288 | * the MP lock to operate. |
| 289 | */ |
| 290 | if ((intr_flags & INTR_MPSAFE) == 0) |
| 291 | info->i_mplock_required = 1; |
| 292 | if (intr_flags & INTR_CLOCK) |
| 293 | ++info->i_fast; |
| 294 | else |
| 295 | ++info->i_slow; |
| 296 | |
| 297 | /* |
| 298 | * Enable random number generation keying off of this interrupt. |
| 299 | */ |
| 300 | if ((intr_flags & INTR_NOENTROPY) == 0 && info->i_random.sc_enabled == 0) { |
| 301 | info->i_random.sc_enabled = 1; |
| 302 | info->i_random.sc_intr = intr; |
| 303 | } |
| 304 | |
| 305 | /* |
| 306 | * Add the record to the interrupt list. |
| 307 | */ |
| 308 | crit_enter(); |
| 309 | while (*list != NULL) |
| 310 | list = &(*list)->next; |
| 311 | *list = rec; |
| 312 | crit_exit(); |
| 313 | |
| 314 | /* |
| 315 | * Update max_installed_hard_intr to make the emergency intr poll |
| 316 | * a bit more efficient. |
| 317 | */ |
| 318 | if (intr < FIRST_SOFTINT) { |
| 319 | if (max_installed_hard_intr <= intr) |
| 320 | max_installed_hard_intr = intr + 1; |
| 321 | } else { |
| 322 | if (max_installed_soft_intr <= intr) |
| 323 | max_installed_soft_intr = intr + 1; |
| 324 | } |
| 325 | |
| 326 | /* |
| 327 | * Setup the machine level interrupt vector |
| 328 | */ |
| 329 | if (intr < FIRST_SOFTINT && info->i_slow + info->i_fast == 1) |
| 330 | machintr_intr_setup(intr, intr_flags); |
| 331 | |
| 332 | int_moveto_origcpu(orig_cpuid, cpuid); |
| 333 | |
| 334 | return(rec); |
| 335 | } |
| 336 | |
| 337 | void |
| 338 | unregister_swi(void *id) |
| 339 | { |
| 340 | unregister_int(id, 0); |
| 341 | } |
| 342 | |
| 343 | void |
| 344 | unregister_int(void *id, int cpuid) |
| 345 | { |
| 346 | struct intr_info *info; |
| 347 | struct intrec **list; |
| 348 | intrec_t rec; |
| 349 | int intr, orig_cpuid; |
| 350 | |
| 351 | KKASSERT(cpuid >= 0 && cpuid < ncpus); |
| 352 | |
| 353 | intr = ((intrec_t)id)->intr; |
| 354 | |
| 355 | if (intr < 0 || intr >= MAX_INTS) |
| 356 | panic("register_int: bad intr %d", intr); |
| 357 | |
| 358 | info = &intr_info_ary[intr]; |
| 359 | |
| 360 | int_moveto_destcpu(&orig_cpuid, cpuid); |
| 361 | |
| 362 | /* |
| 363 | * Remove the interrupt descriptor, adjust the descriptor count, |
| 364 | * and teardown the machine level vector if this was the last interrupt. |
| 365 | */ |
| 366 | crit_enter(); |
| 367 | list = &info->i_reclist; |
| 368 | while ((rec = *list) != NULL) { |
| 369 | if (rec == id) |
| 370 | break; |
| 371 | list = &rec->next; |
| 372 | } |
| 373 | if (rec) { |
| 374 | intrec_t rec0; |
| 375 | |
| 376 | *list = rec->next; |
| 377 | if (rec->intr_flags & INTR_CLOCK) |
| 378 | --info->i_fast; |
| 379 | else |
| 380 | --info->i_slow; |
| 381 | if (intr < FIRST_SOFTINT && info->i_fast + info->i_slow == 0) |
| 382 | machintr_intr_teardown(intr); |
| 383 | |
| 384 | /* |
| 385 | * Clear i_mplock_required if no handlers in the chain require the |
| 386 | * MP lock. |
| 387 | */ |
| 388 | for (rec0 = info->i_reclist; rec0; rec0 = rec0->next) { |
| 389 | if ((rec0->intr_flags & INTR_MPSAFE) == 0) |
| 390 | break; |
| 391 | } |
| 392 | if (rec0 == NULL) |
| 393 | info->i_mplock_required = 0; |
| 394 | } |
| 395 | |
| 396 | crit_exit(); |
| 397 | |
| 398 | int_moveto_origcpu(orig_cpuid, cpuid); |
| 399 | |
| 400 | /* |
| 401 | * Free the record. |
| 402 | */ |
| 403 | if (rec != NULL) { |
| 404 | kfree(rec->name, M_DEVBUF); |
| 405 | kfree(rec, M_DEVBUF); |
| 406 | } else { |
| 407 | kprintf("warning: unregister_int: int %d handler for %s not found\n", |
| 408 | intr, ((intrec_t)id)->name); |
| 409 | } |
| 410 | } |
| 411 | |
| 412 | const char * |
| 413 | get_registered_name(int intr) |
| 414 | { |
| 415 | intrec_t rec; |
| 416 | |
| 417 | if (intr < 0 || intr >= MAX_INTS) |
| 418 | panic("register_int: bad intr %d", intr); |
| 419 | |
| 420 | if ((rec = intr_info_ary[intr].i_reclist) == NULL) |
| 421 | return(NULL); |
| 422 | else if (rec->next) |
| 423 | return("mux"); |
| 424 | else |
| 425 | return(rec->name); |
| 426 | } |
| 427 | |
| 428 | int |
| 429 | count_registered_ints(int intr) |
| 430 | { |
| 431 | struct intr_info *info; |
| 432 | |
| 433 | if (intr < 0 || intr >= MAX_INTS) |
| 434 | panic("register_int: bad intr %d", intr); |
| 435 | info = &intr_info_ary[intr]; |
| 436 | return(info->i_fast + info->i_slow); |
| 437 | } |
| 438 | |
| 439 | long |
| 440 | get_interrupt_counter(int intr) |
| 441 | { |
| 442 | struct intr_info *info; |
| 443 | |
| 444 | if (intr < 0 || intr >= MAX_INTS) |
| 445 | panic("register_int: bad intr %d", intr); |
| 446 | info = &intr_info_ary[intr]; |
| 447 | return(info->i_count); |
| 448 | } |
| 449 | |
| 450 | |
| 451 | void |
| 452 | swi_setpriority(int intr, int pri) |
| 453 | { |
| 454 | struct intr_info *info; |
| 455 | |
| 456 | if (intr < FIRST_SOFTINT || intr >= MAX_INTS) |
| 457 | panic("register_swi: bad intr %d", intr); |
| 458 | info = &intr_info_ary[intr]; |
| 459 | if (info->i_state != ISTATE_NOTHREAD) |
| 460 | lwkt_setpri(&info->i_thread, pri); |
| 461 | } |
| 462 | |
| 463 | void |
| 464 | register_randintr(int intr) |
| 465 | { |
| 466 | struct intr_info *info; |
| 467 | |
| 468 | if (intr < 0 || intr >= MAX_INTS) |
| 469 | panic("register_randintr: bad intr %d", intr); |
| 470 | info = &intr_info_ary[intr]; |
| 471 | info->i_random.sc_intr = intr; |
| 472 | info->i_random.sc_enabled = 1; |
| 473 | } |
| 474 | |
| 475 | void |
| 476 | unregister_randintr(int intr) |
| 477 | { |
| 478 | struct intr_info *info; |
| 479 | |
| 480 | if (intr < 0 || intr >= MAX_INTS) |
| 481 | panic("register_swi: bad intr %d", intr); |
| 482 | info = &intr_info_ary[intr]; |
| 483 | info->i_random.sc_enabled = -1; |
| 484 | } |
| 485 | |
| 486 | int |
| 487 | next_registered_randintr(int intr) |
| 488 | { |
| 489 | struct intr_info *info; |
| 490 | |
| 491 | if (intr < 0 || intr >= MAX_INTS) |
| 492 | panic("register_swi: bad intr %d", intr); |
| 493 | while (intr < MAX_INTS) { |
| 494 | info = &intr_info_ary[intr]; |
| 495 | if (info->i_random.sc_enabled > 0) |
| 496 | break; |
| 497 | ++intr; |
| 498 | } |
| 499 | return(intr); |
| 500 | } |
| 501 | |
| 502 | /* |
| 503 | * Dispatch an interrupt. If there's nothing to do we have a stray |
| 504 | * interrupt and can just return, leaving the interrupt masked. |
| 505 | * |
| 506 | * We need to schedule the interrupt and set its i_running bit. If |
| 507 | * we are not on the interrupt thread's cpu we have to send a message |
| 508 | * to the correct cpu that will issue the desired action (interlocking |
| 509 | * with the interrupt thread's critical section). We do NOT attempt to |
| 510 | * reschedule interrupts whos i_running bit is already set because |
| 511 | * this would prematurely wakeup a livelock-limited interrupt thread. |
| 512 | * |
| 513 | * i_running is only tested/set on the same cpu as the interrupt thread. |
| 514 | * |
| 515 | * We are NOT in a critical section, which will allow the scheduled |
| 516 | * interrupt to preempt us. The MP lock might *NOT* be held here. |
| 517 | */ |
| 518 | #ifdef SMP |
| 519 | |
| 520 | static void |
| 521 | sched_ithd_remote(void *arg) |
| 522 | { |
| 523 | sched_ithd((int)(intptr_t)arg); |
| 524 | } |
| 525 | |
| 526 | #endif |
| 527 | |
| 528 | void |
| 529 | sched_ithd(int intr) |
| 530 | { |
| 531 | struct intr_info *info; |
| 532 | |
| 533 | info = &intr_info_ary[intr]; |
| 534 | |
| 535 | ++info->i_count; |
| 536 | if (info->i_state != ISTATE_NOTHREAD) { |
| 537 | if (info->i_reclist == NULL) { |
| 538 | report_stray_interrupt(intr, info); |
| 539 | } else { |
| 540 | #ifdef SMP |
| 541 | if (info->i_thread.td_gd == mycpu) { |
| 542 | if (info->i_running == 0) { |
| 543 | info->i_running = 1; |
| 544 | if (info->i_state != ISTATE_LIVELOCKED) |
| 545 | lwkt_schedule(&info->i_thread); /* MIGHT PREEMPT */ |
| 546 | } |
| 547 | } else { |
| 548 | lwkt_send_ipiq(info->i_thread.td_gd, |
| 549 | sched_ithd_remote, (void *)(intptr_t)intr); |
| 550 | } |
| 551 | #else |
| 552 | if (info->i_running == 0) { |
| 553 | info->i_running = 1; |
| 554 | if (info->i_state != ISTATE_LIVELOCKED) |
| 555 | lwkt_schedule(&info->i_thread); /* MIGHT PREEMPT */ |
| 556 | } |
| 557 | #endif |
| 558 | } |
| 559 | } else { |
| 560 | report_stray_interrupt(intr, info); |
| 561 | } |
| 562 | } |
| 563 | |
| 564 | static void |
| 565 | report_stray_interrupt(int intr, struct intr_info *info) |
| 566 | { |
| 567 | ++info->i_straycount; |
| 568 | if (info->i_straycount < 10) { |
| 569 | if (info->i_errorticks == ticks) |
| 570 | return; |
| 571 | info->i_errorticks = ticks; |
| 572 | kprintf("sched_ithd: stray interrupt %d on cpu %d\n", |
| 573 | intr, mycpuid); |
| 574 | } else if (info->i_straycount == 10) { |
| 575 | kprintf("sched_ithd: %ld stray interrupts %d on cpu %d - " |
| 576 | "there will be no further reports\n", |
| 577 | info->i_straycount, intr, mycpuid); |
| 578 | } |
| 579 | } |
| 580 | |
| 581 | /* |
| 582 | * This is run from a periodic SYSTIMER (and thus must be MP safe, the BGL |
| 583 | * might not be held). |
| 584 | */ |
| 585 | static void |
| 586 | ithread_livelock_wakeup(systimer_t st, int in_ipi __unused, |
| 587 | struct intrframe *frame __unused) |
| 588 | { |
| 589 | struct intr_info *info; |
| 590 | |
| 591 | info = &intr_info_ary[(int)(intptr_t)st->data]; |
| 592 | if (info->i_state != ISTATE_NOTHREAD) |
| 593 | lwkt_schedule(&info->i_thread); |
| 594 | } |
| 595 | |
| 596 | /* |
| 597 | * Schedule ithread within fast intr handler |
| 598 | * |
| 599 | * XXX Protect sched_ithd() call with gd_intr_nesting_level? |
| 600 | * Interrupts aren't enabled, but still... |
| 601 | */ |
| 602 | static __inline void |
| 603 | ithread_fast_sched(int intr, thread_t td) |
| 604 | { |
| 605 | ++td->td_nest_count; |
| 606 | |
| 607 | /* |
| 608 | * We are already in critical section, exit it now to |
| 609 | * allow preemption. |
| 610 | */ |
| 611 | crit_exit_quick(td); |
| 612 | sched_ithd(intr); |
| 613 | crit_enter_quick(td); |
| 614 | |
| 615 | --td->td_nest_count; |
| 616 | } |
| 617 | |
| 618 | /* |
| 619 | * This function is called directly from the ICU or APIC vector code assembly |
| 620 | * to process an interrupt. The critical section and interrupt deferral |
| 621 | * checks have already been done but the function is entered WITHOUT |
| 622 | * a critical section held. The BGL may or may not be held. |
| 623 | * |
| 624 | * Must return non-zero if we do not want the vector code to re-enable |
| 625 | * the interrupt (which we don't if we have to schedule the interrupt) |
| 626 | */ |
| 627 | int ithread_fast_handler(struct intrframe *frame); |
| 628 | |
| 629 | int |
| 630 | ithread_fast_handler(struct intrframe *frame) |
| 631 | { |
| 632 | int intr; |
| 633 | struct intr_info *info; |
| 634 | struct intrec **list; |
| 635 | int must_schedule; |
| 636 | #ifdef SMP |
| 637 | int got_mplock; |
| 638 | #endif |
| 639 | TD_INVARIANTS_DECLARE; |
| 640 | intrec_t rec, nrec; |
| 641 | globaldata_t gd; |
| 642 | thread_t td; |
| 643 | |
| 644 | intr = frame->if_vec; |
| 645 | gd = mycpu; |
| 646 | td = curthread; |
| 647 | |
| 648 | /* We must be in critical section. */ |
| 649 | KKASSERT(td->td_critcount); |
| 650 | |
| 651 | info = &intr_info_ary[intr]; |
| 652 | |
| 653 | /* |
| 654 | * If we are not processing any FAST interrupts, just schedule the thing. |
| 655 | */ |
| 656 | if (info->i_fast == 0) { |
| 657 | ++gd->gd_cnt.v_intr; |
| 658 | ithread_fast_sched(intr, td); |
| 659 | return(1); |
| 660 | } |
| 661 | |
| 662 | /* |
| 663 | * This should not normally occur since interrupts ought to be |
| 664 | * masked if the ithread has been scheduled or is running. |
| 665 | */ |
| 666 | if (info->i_running) |
| 667 | return(1); |
| 668 | |
| 669 | /* |
| 670 | * Bump the interrupt nesting level to process any FAST interrupts. |
| 671 | * Obtain the MP lock as necessary. If the MP lock cannot be obtained, |
| 672 | * schedule the interrupt thread to deal with the issue instead. |
| 673 | * |
| 674 | * To reduce overhead, just leave the MP lock held once it has been |
| 675 | * obtained. |
| 676 | */ |
| 677 | ++gd->gd_intr_nesting_level; |
| 678 | ++gd->gd_cnt.v_intr; |
| 679 | must_schedule = info->i_slow; |
| 680 | #ifdef SMP |
| 681 | got_mplock = 0; |
| 682 | #endif |
| 683 | |
| 684 | TD_INVARIANTS_GET(td); |
| 685 | list = &info->i_reclist; |
| 686 | |
| 687 | for (rec = *list; rec; rec = nrec) { |
| 688 | /* rec may be invalid after call */ |
| 689 | nrec = rec->next; |
| 690 | |
| 691 | if (rec->intr_flags & INTR_CLOCK) { |
| 692 | #ifdef SMP |
| 693 | if ((rec->intr_flags & INTR_MPSAFE) == 0 && got_mplock == 0) { |
| 694 | if (try_mplock() == 0) { |
| 695 | /* Couldn't get the MP lock; just schedule it. */ |
| 696 | must_schedule = 1; |
| 697 | break; |
| 698 | } |
| 699 | got_mplock = 1; |
| 700 | } |
| 701 | #endif |
| 702 | if (rec->serializer) { |
| 703 | must_schedule += lwkt_serialize_handler_try( |
| 704 | rec->serializer, rec->handler, |
| 705 | rec->argument, frame); |
| 706 | } else { |
| 707 | rec->handler(rec->argument, frame); |
| 708 | } |
| 709 | TD_INVARIANTS_TEST(td, rec->name); |
| 710 | } |
| 711 | } |
| 712 | |
| 713 | /* |
| 714 | * Cleanup |
| 715 | */ |
| 716 | --gd->gd_intr_nesting_level; |
| 717 | #ifdef SMP |
| 718 | if (got_mplock) |
| 719 | rel_mplock(); |
| 720 | #endif |
| 721 | |
| 722 | /* |
| 723 | * If we had a problem, or mixed fast and slow interrupt handlers are |
| 724 | * registered, schedule the ithread to catch the missed records (it |
| 725 | * will just re-run all of them). A return value of 0 indicates that |
| 726 | * all handlers have been run and the interrupt can be re-enabled, and |
| 727 | * a non-zero return indicates that the interrupt thread controls |
| 728 | * re-enablement. |
| 729 | */ |
| 730 | if (must_schedule > 0) |
| 731 | ithread_fast_sched(intr, td); |
| 732 | else if (must_schedule == 0) |
| 733 | ++info->i_count; |
| 734 | return(must_schedule); |
| 735 | } |
| 736 | |
| 737 | /* |
| 738 | * Interrupt threads run this as their main loop. |
| 739 | * |
| 740 | * The handler begins execution outside a critical section and no MP lock. |
| 741 | * |
| 742 | * The i_running state starts at 0. When an interrupt occurs, the hardware |
| 743 | * interrupt is disabled and sched_ithd() The HW interrupt remains disabled |
| 744 | * until all routines have run. We then call ithread_done() to reenable |
| 745 | * the HW interrupt and deschedule us until the next interrupt. |
| 746 | * |
| 747 | * We are responsible for atomically checking i_running and ithread_done() |
| 748 | * is responsible for atomically checking for platform-specific delayed |
| 749 | * interrupts. i_running for our irq is only set in the context of our cpu, |
| 750 | * so a critical section is a sufficient interlock. |
| 751 | */ |
| 752 | #define LIVELOCK_TIMEFRAME(freq) ((freq) >> 2) /* 1/4 second */ |
| 753 | |
| 754 | static void |
| 755 | ithread_handler(void *arg) |
| 756 | { |
| 757 | struct intr_info *info; |
| 758 | int use_limit; |
| 759 | __uint32_t lseconds; |
| 760 | int intr; |
| 761 | int mpheld; |
| 762 | struct intrec **list; |
| 763 | intrec_t rec, nrec; |
| 764 | globaldata_t gd; |
| 765 | struct systimer ill_timer; /* enforced freq. timer */ |
| 766 | u_int ill_count; /* interrupt livelock counter */ |
| 767 | TD_INVARIANTS_DECLARE; |
| 768 | |
| 769 | ill_count = 0; |
| 770 | intr = (int)(intptr_t)arg; |
| 771 | info = &intr_info_ary[intr]; |
| 772 | list = &info->i_reclist; |
| 773 | |
| 774 | /* |
| 775 | * The loop must be entered with one critical section held. The thread |
| 776 | * does not hold the mplock on startup. |
| 777 | */ |
| 778 | gd = mycpu; |
| 779 | lseconds = gd->gd_time_seconds; |
| 780 | crit_enter_gd(gd); |
| 781 | mpheld = 0; |
| 782 | |
| 783 | for (;;) { |
| 784 | /* |
| 785 | * The chain is only considered MPSAFE if all its interrupt handlers |
| 786 | * are MPSAFE. However, if intr_mpsafe has been turned off we |
| 787 | * always operate with the BGL. |
| 788 | */ |
| 789 | #ifdef SMP |
| 790 | if (info->i_mplock_required != mpheld) { |
| 791 | if (info->i_mplock_required) { |
| 792 | KKASSERT(mpheld == 0); |
| 793 | get_mplock(); |
| 794 | mpheld = 1; |
| 795 | } else { |
| 796 | KKASSERT(mpheld != 0); |
| 797 | rel_mplock(); |
| 798 | mpheld = 0; |
| 799 | } |
| 800 | } |
| 801 | #endif |
| 802 | |
| 803 | TD_INVARIANTS_GET(gd->gd_curthread); |
| 804 | |
| 805 | /* |
| 806 | * If an interrupt is pending, clear i_running and execute the |
| 807 | * handlers. Note that certain types of interrupts can re-trigger |
| 808 | * and set i_running again. |
| 809 | * |
| 810 | * Each handler is run in a critical section. Note that we run both |
| 811 | * FAST and SLOW designated service routines. |
| 812 | */ |
| 813 | if (info->i_running) { |
| 814 | ++ill_count; |
| 815 | info->i_running = 0; |
| 816 | |
| 817 | if (*list == NULL) |
| 818 | report_stray_interrupt(intr, info); |
| 819 | |
| 820 | for (rec = *list; rec; rec = nrec) { |
| 821 | /* rec may be invalid after call */ |
| 822 | nrec = rec->next; |
| 823 | if (rec->serializer) { |
| 824 | lwkt_serialize_handler_call(rec->serializer, rec->handler, |
| 825 | rec->argument, NULL); |
| 826 | } else { |
| 827 | rec->handler(rec->argument, NULL); |
| 828 | } |
| 829 | TD_INVARIANTS_TEST(gd->gd_curthread, rec->name); |
| 830 | } |
| 831 | } |
| 832 | |
| 833 | /* |
| 834 | * This is our interrupt hook to add rate randomness to the random |
| 835 | * number generator. |
| 836 | */ |
| 837 | if (info->i_random.sc_enabled > 0) |
| 838 | add_interrupt_randomness(intr); |
| 839 | |
| 840 | /* |
| 841 | * Unmask the interrupt to allow it to trigger again. This only |
| 842 | * applies to certain types of interrupts (typ level interrupts). |
| 843 | * This can result in the interrupt retriggering, but the retrigger |
| 844 | * will not be processed until we cycle our critical section. |
| 845 | * |
| 846 | * Only unmask interrupts while handlers are installed. It is |
| 847 | * possible to hit a situation where no handlers are installed |
| 848 | * due to a device driver livelocking and then tearing down its |
| 849 | * interrupt on close (the parallel bus being a good example). |
| 850 | */ |
| 851 | if (intr < FIRST_SOFTINT && *list) |
| 852 | machintr_intr_enable(intr); |
| 853 | |
| 854 | /* |
| 855 | * Do a quick exit/enter to catch any higher-priority interrupt |
| 856 | * sources, such as the statclock, so thread time accounting |
| 857 | * will still work. This may also cause an interrupt to re-trigger. |
| 858 | */ |
| 859 | crit_exit_gd(gd); |
| 860 | crit_enter_gd(gd); |
| 861 | |
| 862 | /* |
| 863 | * LIVELOCK STATE MACHINE |
| 864 | */ |
| 865 | switch(info->i_state) { |
| 866 | case ISTATE_NORMAL: |
| 867 | /* |
| 868 | * Reset the count each second. |
| 869 | */ |
| 870 | if (lseconds != gd->gd_time_seconds) { |
| 871 | lseconds = gd->gd_time_seconds; |
| 872 | ill_count = 0; |
| 873 | } |
| 874 | |
| 875 | /* |
| 876 | * If we did not exceed the frequency limit, we are done. |
| 877 | * If the interrupt has not retriggered we deschedule ourselves. |
| 878 | */ |
| 879 | if (ill_count <= livelock_limit) { |
| 880 | if (info->i_running == 0) { |
| 881 | lwkt_deschedule_self(gd->gd_curthread); |
| 882 | lwkt_switch(); |
| 883 | } |
| 884 | break; |
| 885 | } |
| 886 | |
| 887 | /* |
| 888 | * Otherwise we are livelocked. Set up a periodic systimer |
| 889 | * to wake the thread up at the limit frequency. |
| 890 | */ |
| 891 | kprintf("intr %d at %d/%d hz, livelocked limit engaged!\n", |
| 892 | intr, ill_count, livelock_limit); |
| 893 | info->i_state = ISTATE_LIVELOCKED; |
| 894 | if ((use_limit = livelock_limit) < 100) |
| 895 | use_limit = 100; |
| 896 | else if (use_limit > 500000) |
| 897 | use_limit = 500000; |
| 898 | systimer_init_periodic_nq(&ill_timer, ithread_livelock_wakeup, |
| 899 | (void *)(intptr_t)intr, use_limit); |
| 900 | /* fall through */ |
| 901 | case ISTATE_LIVELOCKED: |
| 902 | /* |
| 903 | * Wait for our periodic timer to go off. Since the interrupt |
| 904 | * has re-armed it can still set i_running, but it will not |
| 905 | * reschedule us while we are in a livelocked state. |
| 906 | */ |
| 907 | lwkt_deschedule_self(gd->gd_curthread); |
| 908 | lwkt_switch(); |
| 909 | |
| 910 | /* |
| 911 | * Check once a second to see if the livelock condition no |
| 912 | * longer applies. |
| 913 | */ |
| 914 | if (lseconds != gd->gd_time_seconds) { |
| 915 | lseconds = gd->gd_time_seconds; |
| 916 | if (ill_count < livelock_lowater) { |
| 917 | info->i_state = ISTATE_NORMAL; |
| 918 | systimer_del(&ill_timer); |
| 919 | kprintf("intr %d at %d/%d hz, livelock removed\n", |
| 920 | intr, ill_count, livelock_lowater); |
| 921 | } else if (livelock_debug == intr || |
| 922 | (bootverbose && cold)) { |
| 923 | kprintf("intr %d at %d/%d hz, in livelock\n", |
| 924 | intr, ill_count, livelock_lowater); |
| 925 | } |
| 926 | ill_count = 0; |
| 927 | } |
| 928 | break; |
| 929 | } |
| 930 | } |
| 931 | /* NOT REACHED */ |
| 932 | } |
| 933 | |
| 934 | /* |
| 935 | * Emergency interrupt polling thread. The thread begins execution |
| 936 | * outside a critical section with the BGL held. |
| 937 | * |
| 938 | * If emergency interrupt polling is enabled, this thread will |
| 939 | * execute all system interrupts not marked INTR_NOPOLL at the |
| 940 | * specified polling frequency. |
| 941 | * |
| 942 | * WARNING! This thread runs *ALL* interrupt service routines that |
| 943 | * are not marked INTR_NOPOLL, which basically means everything except |
| 944 | * the 8254 clock interrupt and the ATA interrupt. It has very high |
| 945 | * overhead and should only be used in situations where the machine |
| 946 | * cannot otherwise be made to work. Due to the severe performance |
| 947 | * degredation, it should not be enabled on production machines. |
| 948 | */ |
| 949 | static void |
| 950 | ithread_emergency(void *arg __unused) |
| 951 | { |
| 952 | globaldata_t gd = mycpu; |
| 953 | struct intr_info *info; |
| 954 | intrec_t rec, nrec; |
| 955 | int intr; |
| 956 | TD_INVARIANTS_DECLARE; |
| 957 | |
| 958 | get_mplock(); |
| 959 | crit_enter_gd(gd); |
| 960 | TD_INVARIANTS_GET(gd->gd_curthread); |
| 961 | |
| 962 | for (;;) { |
| 963 | for (intr = 0; intr < max_installed_hard_intr; ++intr) { |
| 964 | info = &intr_info_ary[intr]; |
| 965 | for (rec = info->i_reclist; rec; rec = nrec) { |
| 966 | /* rec may be invalid after call */ |
| 967 | nrec = rec->next; |
| 968 | if ((rec->intr_flags & INTR_NOPOLL) == 0) { |
| 969 | if (rec->serializer) { |
| 970 | lwkt_serialize_handler_try(rec->serializer, |
| 971 | rec->handler, rec->argument, NULL); |
| 972 | } else { |
| 973 | rec->handler(rec->argument, NULL); |
| 974 | } |
| 975 | TD_INVARIANTS_TEST(gd->gd_curthread, rec->name); |
| 976 | } |
| 977 | } |
| 978 | } |
| 979 | lwkt_deschedule_self(gd->gd_curthread); |
| 980 | lwkt_switch(); |
| 981 | } |
| 982 | /* NOT REACHED */ |
| 983 | } |
| 984 | |
| 985 | /* |
| 986 | * Systimer callback - schedule the emergency interrupt poll thread |
| 987 | * if emergency polling is enabled. |
| 988 | */ |
| 989 | static |
| 990 | void |
| 991 | emergency_intr_timer_callback(systimer_t info, int in_ipi __unused, |
| 992 | struct intrframe *frame __unused) |
| 993 | { |
| 994 | if (emergency_intr_enable) |
| 995 | lwkt_schedule(info->data); |
| 996 | } |
| 997 | |
| 998 | int |
| 999 | ithread_cpuid(int intr) |
| 1000 | { |
| 1001 | const struct intr_info *info; |
| 1002 | |
| 1003 | KKASSERT(intr >= 0 && intr < MAX_INTS); |
| 1004 | info = &intr_info_ary[intr]; |
| 1005 | |
| 1006 | if (info->i_state == ISTATE_NOTHREAD) |
| 1007 | return -1; |
| 1008 | return info->i_thread.td_gd->gd_cpuid; |
| 1009 | } |
| 1010 | |
| 1011 | /* |
| 1012 | * Sysctls used by systat and others: hw.intrnames and hw.intrcnt. |
| 1013 | * The data for this machine dependent, and the declarations are in machine |
| 1014 | * dependent code. The layout of intrnames and intrcnt however is machine |
| 1015 | * independent. |
| 1016 | * |
| 1017 | * We do not know the length of intrcnt and intrnames at compile time, so |
| 1018 | * calculate things at run time. |
| 1019 | */ |
| 1020 | |
| 1021 | static int |
| 1022 | sysctl_intrnames(SYSCTL_HANDLER_ARGS) |
| 1023 | { |
| 1024 | struct intr_info *info; |
| 1025 | intrec_t rec; |
| 1026 | int error = 0; |
| 1027 | int len; |
| 1028 | int intr; |
| 1029 | char buf[64]; |
| 1030 | |
| 1031 | for (intr = 0; error == 0 && intr < MAX_INTS; ++intr) { |
| 1032 | info = &intr_info_ary[intr]; |
| 1033 | |
| 1034 | len = 0; |
| 1035 | buf[0] = 0; |
| 1036 | for (rec = info->i_reclist; rec; rec = rec->next) { |
| 1037 | ksnprintf(buf + len, sizeof(buf) - len, "%s%s", |
| 1038 | (len ? "/" : ""), rec->name); |
| 1039 | len += strlen(buf + len); |
| 1040 | } |
| 1041 | if (len == 0) { |
| 1042 | ksnprintf(buf, sizeof(buf), "irq%d", intr); |
| 1043 | len = strlen(buf); |
| 1044 | } |
| 1045 | error = SYSCTL_OUT(req, buf, len + 1); |
| 1046 | } |
| 1047 | return (error); |
| 1048 | } |
| 1049 | |
| 1050 | |
| 1051 | SYSCTL_PROC(_hw, OID_AUTO, intrnames, CTLTYPE_OPAQUE | CTLFLAG_RD, |
| 1052 | NULL, 0, sysctl_intrnames, "", "Interrupt Names"); |
| 1053 | |
| 1054 | static int |
| 1055 | sysctl_intrcnt(SYSCTL_HANDLER_ARGS) |
| 1056 | { |
| 1057 | struct intr_info *info; |
| 1058 | int error = 0; |
| 1059 | int intr; |
| 1060 | |
| 1061 | for (intr = 0; intr < max_installed_hard_intr; ++intr) { |
| 1062 | info = &intr_info_ary[intr]; |
| 1063 | |
| 1064 | error = SYSCTL_OUT(req, &info->i_count, sizeof(info->i_count)); |
| 1065 | if (error) |
| 1066 | goto failed; |
| 1067 | } |
| 1068 | for (intr = FIRST_SOFTINT; intr < max_installed_soft_intr; ++intr) { |
| 1069 | info = &intr_info_ary[intr]; |
| 1070 | |
| 1071 | error = SYSCTL_OUT(req, &info->i_count, sizeof(info->i_count)); |
| 1072 | if (error) |
| 1073 | goto failed; |
| 1074 | } |
| 1075 | failed: |
| 1076 | return(error); |
| 1077 | } |
| 1078 | |
| 1079 | SYSCTL_PROC(_hw, OID_AUTO, intrcnt, CTLTYPE_OPAQUE | CTLFLAG_RD, |
| 1080 | NULL, 0, sysctl_intrcnt, "", "Interrupt Counts"); |
| 1081 | |
| 1082 | static int |
| 1083 | sysctl_intrcnt_all(SYSCTL_HANDLER_ARGS) |
| 1084 | { |
| 1085 | struct intr_info *info; |
| 1086 | int error = 0; |
| 1087 | int intr; |
| 1088 | |
| 1089 | for (intr = 0; intr < MAX_INTS; ++intr) { |
| 1090 | info = &intr_info_ary[intr]; |
| 1091 | |
| 1092 | error = SYSCTL_OUT(req, &info->i_count, sizeof(info->i_count)); |
| 1093 | if (error) |
| 1094 | goto failed; |
| 1095 | } |
| 1096 | failed: |
| 1097 | return(error); |
| 1098 | } |
| 1099 | |
| 1100 | SYSCTL_PROC(_hw, OID_AUTO, intrcnt_all, CTLTYPE_OPAQUE | CTLFLAG_RD, |
| 1101 | NULL, 0, sysctl_intrcnt_all, "", "Interrupt Counts"); |
| 1102 | |
| 1103 | static void |
| 1104 | int_moveto_destcpu(int *orig_cpuid0, int cpuid) |
| 1105 | { |
| 1106 | int orig_cpuid = mycpuid; |
| 1107 | |
| 1108 | if (cpuid != orig_cpuid) |
| 1109 | lwkt_migratecpu(cpuid); |
| 1110 | |
| 1111 | *orig_cpuid0 = orig_cpuid; |
| 1112 | } |
| 1113 | |
| 1114 | static void |
| 1115 | int_moveto_origcpu(int orig_cpuid, int cpuid) |
| 1116 | { |
| 1117 | if (cpuid != orig_cpuid) |
| 1118 | lwkt_migratecpu(orig_cpuid); |
| 1119 | } |