| 1 | /* |
| 2 | * Copyright (c) 2003 Matthew Dillon <dillon@backplane.com> All rights reserved. |
| 3 | * Copyright (c) 1997, Stefan Esser <se@freebsd.org> All rights reserved. |
| 4 | * |
| 5 | * Redistribution and use in source and binary forms, with or without |
| 6 | * modification, are permitted provided that the following conditions |
| 7 | * are met: |
| 8 | * 1. Redistributions of source code must retain the above copyright |
| 9 | * notice unmodified, this list of conditions, and the following |
| 10 | * disclaimer. |
| 11 | * 2. Redistributions in binary form must reproduce the above copyright |
| 12 | * notice, this list of conditions and the following disclaimer in the |
| 13 | * documentation and/or other materials provided with the distribution. |
| 14 | * |
| 15 | * THIS SOFTWARE IS PROVIDED BY THE AUTHOR ``AS IS'' AND ANY EXPRESS OR |
| 16 | * IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE IMPLIED WARRANTIES |
| 17 | * OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE ARE DISCLAIMED. |
| 18 | * IN NO EVENT SHALL THE AUTHOR BE LIABLE FOR ANY DIRECT, INDIRECT, |
| 19 | * INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT |
| 20 | * NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, |
| 21 | * DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY |
| 22 | * THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT |
| 23 | * (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE OF |
| 24 | * THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE. |
| 25 | * |
| 26 | * $FreeBSD: src/sys/kern/kern_intr.c,v 1.24.2.1 2001/10/14 20:05:50 luigi Exp $ |
| 27 | * $DragonFly: src/sys/kern/kern_intr.c,v 1.45 2006/12/23 00:35:04 swildner Exp $ |
| 28 | * |
| 29 | */ |
| 30 | |
| 31 | #include <sys/param.h> |
| 32 | #include <sys/systm.h> |
| 33 | #include <sys/malloc.h> |
| 34 | #include <sys/kernel.h> |
| 35 | #include <sys/sysctl.h> |
| 36 | #include <sys/thread.h> |
| 37 | #include <sys/proc.h> |
| 38 | #include <sys/thread2.h> |
| 39 | #include <sys/random.h> |
| 40 | #include <sys/serialize.h> |
| 41 | #include <sys/interrupt.h> |
| 42 | #include <sys/bus.h> |
| 43 | #include <sys/machintr.h> |
| 44 | |
| 45 | #include <machine/frame.h> |
| 46 | |
| 47 | #include <sys/interrupt.h> |
| 48 | |
| 49 | struct info_info; |
| 50 | |
| 51 | typedef struct intrec { |
| 52 | struct intrec *next; |
| 53 | struct intr_info *info; |
| 54 | inthand2_t *handler; |
| 55 | void *argument; |
| 56 | char *name; |
| 57 | int intr; |
| 58 | int intr_flags; |
| 59 | struct lwkt_serialize *serializer; |
| 60 | } *intrec_t; |
| 61 | |
| 62 | struct intr_info { |
| 63 | intrec_t i_reclist; |
| 64 | struct thread i_thread; |
| 65 | struct random_softc i_random; |
| 66 | int i_running; |
| 67 | long i_count; /* interrupts dispatched */ |
| 68 | int i_mplock_required; |
| 69 | int i_fast; |
| 70 | int i_slow; |
| 71 | int i_state; |
| 72 | } intr_info_ary[MAX_INTS]; |
| 73 | |
| 74 | int max_installed_hard_intr; |
| 75 | int max_installed_soft_intr; |
| 76 | |
| 77 | #define EMERGENCY_INTR_POLLING_FREQ_MAX 20000 |
| 78 | |
| 79 | static int sysctl_emergency_freq(SYSCTL_HANDLER_ARGS); |
| 80 | static int sysctl_emergency_enable(SYSCTL_HANDLER_ARGS); |
| 81 | static void emergency_intr_timer_callback(systimer_t, struct intrframe *); |
| 82 | static void ithread_handler(void *arg); |
| 83 | static void ithread_emergency(void *arg); |
| 84 | |
| 85 | int intr_info_size = sizeof(intr_info_ary) / sizeof(intr_info_ary[0]); |
| 86 | |
| 87 | static struct systimer emergency_intr_timer; |
| 88 | static struct thread emergency_intr_thread; |
| 89 | |
| 90 | #define ISTATE_NOTHREAD 0 |
| 91 | #define ISTATE_NORMAL 1 |
| 92 | #define ISTATE_LIVELOCKED 2 |
| 93 | |
| 94 | #ifdef SMP |
| 95 | static int intr_mpsafe = 0; |
| 96 | TUNABLE_INT("kern.intr_mpsafe", &intr_mpsafe); |
| 97 | SYSCTL_INT(_kern, OID_AUTO, intr_mpsafe, |
| 98 | CTLFLAG_RW, &intr_mpsafe, 0, "Run INTR_MPSAFE handlers without the BGL"); |
| 99 | #endif |
| 100 | static int livelock_limit = 50000; |
| 101 | static int livelock_lowater = 20000; |
| 102 | SYSCTL_INT(_kern, OID_AUTO, livelock_limit, |
| 103 | CTLFLAG_RW, &livelock_limit, 0, "Livelock interrupt rate limit"); |
| 104 | SYSCTL_INT(_kern, OID_AUTO, livelock_lowater, |
| 105 | CTLFLAG_RW, &livelock_lowater, 0, "Livelock low-water mark restore"); |
| 106 | |
| 107 | static int emergency_intr_enable = 0; /* emergency interrupt polling */ |
| 108 | TUNABLE_INT("kern.emergency_intr_enable", &emergency_intr_enable); |
| 109 | SYSCTL_PROC(_kern, OID_AUTO, emergency_intr_enable, CTLTYPE_INT | CTLFLAG_RW, |
| 110 | 0, 0, sysctl_emergency_enable, "I", "Emergency Interrupt Poll Enable"); |
| 111 | |
| 112 | static int emergency_intr_freq = 10; /* emergency polling frequency */ |
| 113 | TUNABLE_INT("kern.emergency_intr_freq", &emergency_intr_freq); |
| 114 | SYSCTL_PROC(_kern, OID_AUTO, emergency_intr_freq, CTLTYPE_INT | CTLFLAG_RW, |
| 115 | 0, 0, sysctl_emergency_freq, "I", "Emergency Interrupt Poll Frequency"); |
| 116 | |
| 117 | /* |
| 118 | * Sysctl support routines |
| 119 | */ |
| 120 | static int |
| 121 | sysctl_emergency_enable(SYSCTL_HANDLER_ARGS) |
| 122 | { |
| 123 | int error, enabled; |
| 124 | |
| 125 | enabled = emergency_intr_enable; |
| 126 | error = sysctl_handle_int(oidp, &enabled, 0, req); |
| 127 | if (error || req->newptr == NULL) |
| 128 | return error; |
| 129 | emergency_intr_enable = enabled; |
| 130 | if (emergency_intr_enable) { |
| 131 | emergency_intr_timer.periodic = |
| 132 | sys_cputimer->fromhz(emergency_intr_freq); |
| 133 | } else { |
| 134 | emergency_intr_timer.periodic = sys_cputimer->fromhz(1); |
| 135 | } |
| 136 | return 0; |
| 137 | } |
| 138 | |
| 139 | static int |
| 140 | sysctl_emergency_freq(SYSCTL_HANDLER_ARGS) |
| 141 | { |
| 142 | int error, phz; |
| 143 | |
| 144 | phz = emergency_intr_freq; |
| 145 | error = sysctl_handle_int(oidp, &phz, 0, req); |
| 146 | if (error || req->newptr == NULL) |
| 147 | return error; |
| 148 | if (phz <= 0) |
| 149 | return EINVAL; |
| 150 | else if (phz > EMERGENCY_INTR_POLLING_FREQ_MAX) |
| 151 | phz = EMERGENCY_INTR_POLLING_FREQ_MAX; |
| 152 | |
| 153 | emergency_intr_freq = phz; |
| 154 | if (emergency_intr_enable) { |
| 155 | emergency_intr_timer.periodic = |
| 156 | sys_cputimer->fromhz(emergency_intr_freq); |
| 157 | } else { |
| 158 | emergency_intr_timer.periodic = sys_cputimer->fromhz(1); |
| 159 | } |
| 160 | return 0; |
| 161 | } |
| 162 | |
| 163 | /* |
| 164 | * Register an SWI or INTerrupt handler. |
| 165 | */ |
| 166 | void * |
| 167 | register_swi(int intr, inthand2_t *handler, void *arg, const char *name, |
| 168 | struct lwkt_serialize *serializer) |
| 169 | { |
| 170 | if (intr < FIRST_SOFTINT || intr >= MAX_INTS) |
| 171 | panic("register_swi: bad intr %d", intr); |
| 172 | return(register_int(intr, handler, arg, name, serializer, 0)); |
| 173 | } |
| 174 | |
| 175 | void * |
| 176 | register_int(int intr, inthand2_t *handler, void *arg, const char *name, |
| 177 | struct lwkt_serialize *serializer, int intr_flags) |
| 178 | { |
| 179 | struct intr_info *info; |
| 180 | struct intrec **list; |
| 181 | intrec_t rec; |
| 182 | |
| 183 | if (intr < 0 || intr >= MAX_INTS) |
| 184 | panic("register_int: bad intr %d", intr); |
| 185 | if (name == NULL) |
| 186 | name = "???"; |
| 187 | info = &intr_info_ary[intr]; |
| 188 | |
| 189 | /* |
| 190 | * Construct an interrupt handler record |
| 191 | */ |
| 192 | rec = kmalloc(sizeof(struct intrec), M_DEVBUF, M_INTWAIT); |
| 193 | rec->name = kmalloc(strlen(name) + 1, M_DEVBUF, M_INTWAIT); |
| 194 | strcpy(rec->name, name); |
| 195 | |
| 196 | rec->info = info; |
| 197 | rec->handler = handler; |
| 198 | rec->argument = arg; |
| 199 | rec->intr = intr; |
| 200 | rec->intr_flags = intr_flags; |
| 201 | rec->next = NULL; |
| 202 | rec->serializer = serializer; |
| 203 | |
| 204 | /* |
| 205 | * Create an emergency polling thread and set up a systimer to wake |
| 206 | * it up. |
| 207 | */ |
| 208 | if (emergency_intr_thread.td_kstack == NULL) { |
| 209 | lwkt_create(ithread_emergency, NULL, NULL, |
| 210 | &emergency_intr_thread, TDF_STOPREQ|TDF_INTTHREAD, -1, |
| 211 | "ithread emerg"); |
| 212 | systimer_init_periodic_nq(&emergency_intr_timer, |
| 213 | emergency_intr_timer_callback, &emergency_intr_thread, |
| 214 | (emergency_intr_enable ? emergency_intr_freq : 1)); |
| 215 | } |
| 216 | |
| 217 | /* |
| 218 | * Create an interrupt thread if necessary, leave it in an unscheduled |
| 219 | * state. |
| 220 | */ |
| 221 | if (info->i_state == ISTATE_NOTHREAD) { |
| 222 | info->i_state = ISTATE_NORMAL; |
| 223 | lwkt_create((void *)ithread_handler, (void *)intr, NULL, |
| 224 | &info->i_thread, TDF_STOPREQ|TDF_INTTHREAD|TDF_MPSAFE, -1, |
| 225 | "ithread %d", intr); |
| 226 | if (intr >= FIRST_SOFTINT) |
| 227 | lwkt_setpri(&info->i_thread, TDPRI_SOFT_NORM); |
| 228 | else |
| 229 | lwkt_setpri(&info->i_thread, TDPRI_INT_MED); |
| 230 | info->i_thread.td_preemptable = lwkt_preempt; |
| 231 | } |
| 232 | |
| 233 | list = &info->i_reclist; |
| 234 | |
| 235 | /* |
| 236 | * Keep track of how many fast and slow interrupts we have. |
| 237 | * Set i_mplock_required if any handler in the chain requires |
| 238 | * the MP lock to operate. |
| 239 | */ |
| 240 | if ((intr_flags & INTR_MPSAFE) == 0) |
| 241 | info->i_mplock_required = 1; |
| 242 | if (intr_flags & INTR_FAST) |
| 243 | ++info->i_fast; |
| 244 | else |
| 245 | ++info->i_slow; |
| 246 | |
| 247 | /* |
| 248 | * Enable random number generation keying off of this interrupt. |
| 249 | */ |
| 250 | if ((intr_flags & INTR_NOENTROPY) == 0 && info->i_random.sc_enabled == 0) { |
| 251 | info->i_random.sc_enabled = 1; |
| 252 | info->i_random.sc_intr = intr; |
| 253 | } |
| 254 | |
| 255 | /* |
| 256 | * Add the record to the interrupt list. |
| 257 | */ |
| 258 | crit_enter(); |
| 259 | while (*list != NULL) |
| 260 | list = &(*list)->next; |
| 261 | *list = rec; |
| 262 | crit_exit(); |
| 263 | |
| 264 | /* |
| 265 | * Update max_installed_hard_intr to make the emergency intr poll |
| 266 | * a bit more efficient. |
| 267 | */ |
| 268 | if (intr < FIRST_SOFTINT) { |
| 269 | if (max_installed_hard_intr <= intr) |
| 270 | max_installed_hard_intr = intr + 1; |
| 271 | } else { |
| 272 | if (max_installed_soft_intr <= intr) |
| 273 | max_installed_soft_intr = intr + 1; |
| 274 | } |
| 275 | |
| 276 | /* |
| 277 | * Setup the machine level interrupt vector |
| 278 | * |
| 279 | * XXX temporary workaround for some ACPI brokedness. ACPI installs |
| 280 | * its interrupt too early, before the IOAPICs have been configured, |
| 281 | * which means the IOAPIC is not enabled by the registration of the |
| 282 | * ACPI interrupt. Anything else sharing that IRQ will wind up not |
| 283 | * being enabled. Temporarily work around the problem by always |
| 284 | * installing and enabling on every new interrupt handler, even |
| 285 | * if one has already been setup on that irq. |
| 286 | */ |
| 287 | if (intr < FIRST_SOFTINT /* && info->i_slow + info->i_fast == 1*/) { |
| 288 | if (machintr_vector_setup(intr, intr_flags)) |
| 289 | kprintf("machintr_vector_setup: failed on irq %d\n", intr); |
| 290 | } |
| 291 | |
| 292 | return(rec); |
| 293 | } |
| 294 | |
| 295 | void |
| 296 | unregister_swi(void *id) |
| 297 | { |
| 298 | unregister_int(id); |
| 299 | } |
| 300 | |
| 301 | void |
| 302 | unregister_int(void *id) |
| 303 | { |
| 304 | struct intr_info *info; |
| 305 | struct intrec **list; |
| 306 | intrec_t rec; |
| 307 | int intr; |
| 308 | |
| 309 | intr = ((intrec_t)id)->intr; |
| 310 | |
| 311 | if (intr < 0 || intr >= MAX_INTS) |
| 312 | panic("register_int: bad intr %d", intr); |
| 313 | |
| 314 | info = &intr_info_ary[intr]; |
| 315 | |
| 316 | /* |
| 317 | * Remove the interrupt descriptor, adjust the descriptor count, |
| 318 | * and teardown the machine level vector if this was the last interrupt. |
| 319 | */ |
| 320 | crit_enter(); |
| 321 | list = &info->i_reclist; |
| 322 | while ((rec = *list) != NULL) { |
| 323 | if (rec == id) |
| 324 | break; |
| 325 | list = &rec->next; |
| 326 | } |
| 327 | if (rec) { |
| 328 | intrec_t rec0; |
| 329 | |
| 330 | *list = rec->next; |
| 331 | if (rec->intr_flags & INTR_FAST) |
| 332 | --info->i_fast; |
| 333 | else |
| 334 | --info->i_slow; |
| 335 | if (intr < FIRST_SOFTINT && info->i_fast + info->i_slow == 0) |
| 336 | machintr_vector_teardown(intr); |
| 337 | |
| 338 | /* |
| 339 | * Clear i_mplock_required if no handlers in the chain require the |
| 340 | * MP lock. |
| 341 | */ |
| 342 | for (rec0 = info->i_reclist; rec0; rec0 = rec0->next) { |
| 343 | if ((rec0->intr_flags & INTR_MPSAFE) == 0) |
| 344 | break; |
| 345 | } |
| 346 | if (rec0 == NULL) |
| 347 | info->i_mplock_required = 0; |
| 348 | } |
| 349 | |
| 350 | crit_exit(); |
| 351 | |
| 352 | /* |
| 353 | * Free the record. |
| 354 | */ |
| 355 | if (rec != NULL) { |
| 356 | kfree(rec->name, M_DEVBUF); |
| 357 | kfree(rec, M_DEVBUF); |
| 358 | } else { |
| 359 | kprintf("warning: unregister_int: int %d handler for %s not found\n", |
| 360 | intr, ((intrec_t)id)->name); |
| 361 | } |
| 362 | } |
| 363 | |
| 364 | const char * |
| 365 | get_registered_name(int intr) |
| 366 | { |
| 367 | intrec_t rec; |
| 368 | |
| 369 | if (intr < 0 || intr >= MAX_INTS) |
| 370 | panic("register_int: bad intr %d", intr); |
| 371 | |
| 372 | if ((rec = intr_info_ary[intr].i_reclist) == NULL) |
| 373 | return(NULL); |
| 374 | else if (rec->next) |
| 375 | return("mux"); |
| 376 | else |
| 377 | return(rec->name); |
| 378 | } |
| 379 | |
| 380 | int |
| 381 | count_registered_ints(int intr) |
| 382 | { |
| 383 | struct intr_info *info; |
| 384 | |
| 385 | if (intr < 0 || intr >= MAX_INTS) |
| 386 | panic("register_int: bad intr %d", intr); |
| 387 | info = &intr_info_ary[intr]; |
| 388 | return(info->i_fast + info->i_slow); |
| 389 | } |
| 390 | |
| 391 | long |
| 392 | get_interrupt_counter(int intr) |
| 393 | { |
| 394 | struct intr_info *info; |
| 395 | |
| 396 | if (intr < 0 || intr >= MAX_INTS) |
| 397 | panic("register_int: bad intr %d", intr); |
| 398 | info = &intr_info_ary[intr]; |
| 399 | return(info->i_count); |
| 400 | } |
| 401 | |
| 402 | |
| 403 | void |
| 404 | swi_setpriority(int intr, int pri) |
| 405 | { |
| 406 | struct intr_info *info; |
| 407 | |
| 408 | if (intr < FIRST_SOFTINT || intr >= MAX_INTS) |
| 409 | panic("register_swi: bad intr %d", intr); |
| 410 | info = &intr_info_ary[intr]; |
| 411 | if (info->i_state != ISTATE_NOTHREAD) |
| 412 | lwkt_setpri(&info->i_thread, pri); |
| 413 | } |
| 414 | |
| 415 | void |
| 416 | register_randintr(int intr) |
| 417 | { |
| 418 | struct intr_info *info; |
| 419 | |
| 420 | if (intr < 0 || intr >= MAX_INTS) |
| 421 | panic("register_randintr: bad intr %d", intr); |
| 422 | info = &intr_info_ary[intr]; |
| 423 | info->i_random.sc_intr = intr; |
| 424 | info->i_random.sc_enabled = 1; |
| 425 | } |
| 426 | |
| 427 | void |
| 428 | unregister_randintr(int intr) |
| 429 | { |
| 430 | struct intr_info *info; |
| 431 | |
| 432 | if (intr < 0 || intr >= MAX_INTS) |
| 433 | panic("register_swi: bad intr %d", intr); |
| 434 | info = &intr_info_ary[intr]; |
| 435 | info->i_random.sc_enabled = -1; |
| 436 | } |
| 437 | |
| 438 | int |
| 439 | next_registered_randintr(int intr) |
| 440 | { |
| 441 | struct intr_info *info; |
| 442 | |
| 443 | if (intr < 0 || intr >= MAX_INTS) |
| 444 | panic("register_swi: bad intr %d", intr); |
| 445 | while (intr < MAX_INTS) { |
| 446 | info = &intr_info_ary[intr]; |
| 447 | if (info->i_random.sc_enabled > 0) |
| 448 | break; |
| 449 | ++intr; |
| 450 | } |
| 451 | return(intr); |
| 452 | } |
| 453 | |
| 454 | /* |
| 455 | * Dispatch an interrupt. If there's nothing to do we have a stray |
| 456 | * interrupt and can just return, leaving the interrupt masked. |
| 457 | * |
| 458 | * We need to schedule the interrupt and set its i_running bit. If |
| 459 | * we are not on the interrupt thread's cpu we have to send a message |
| 460 | * to the correct cpu that will issue the desired action (interlocking |
| 461 | * with the interrupt thread's critical section). We do NOT attempt to |
| 462 | * reschedule interrupts whos i_running bit is already set because |
| 463 | * this would prematurely wakeup a livelock-limited interrupt thread. |
| 464 | * |
| 465 | * i_running is only tested/set on the same cpu as the interrupt thread. |
| 466 | * |
| 467 | * We are NOT in a critical section, which will allow the scheduled |
| 468 | * interrupt to preempt us. The MP lock might *NOT* be held here. |
| 469 | */ |
| 470 | #ifdef SMP |
| 471 | |
| 472 | static void |
| 473 | sched_ithd_remote(void *arg) |
| 474 | { |
| 475 | sched_ithd((int)arg); |
| 476 | } |
| 477 | |
| 478 | #endif |
| 479 | |
| 480 | void |
| 481 | sched_ithd(int intr) |
| 482 | { |
| 483 | struct intr_info *info; |
| 484 | |
| 485 | info = &intr_info_ary[intr]; |
| 486 | |
| 487 | ++info->i_count; |
| 488 | if (info->i_state != ISTATE_NOTHREAD) { |
| 489 | if (info->i_reclist == NULL) { |
| 490 | kprintf("sched_ithd: stray interrupt %d\n", intr); |
| 491 | } else { |
| 492 | #ifdef SMP |
| 493 | if (info->i_thread.td_gd == mycpu) { |
| 494 | if (info->i_running == 0) { |
| 495 | info->i_running = 1; |
| 496 | if (info->i_state != ISTATE_LIVELOCKED) |
| 497 | lwkt_schedule(&info->i_thread); /* MIGHT PREEMPT */ |
| 498 | } |
| 499 | } else { |
| 500 | lwkt_send_ipiq(info->i_thread.td_gd, |
| 501 | sched_ithd_remote, (void *)intr); |
| 502 | } |
| 503 | #else |
| 504 | if (info->i_running == 0) { |
| 505 | info->i_running = 1; |
| 506 | if (info->i_state != ISTATE_LIVELOCKED) |
| 507 | lwkt_schedule(&info->i_thread); /* MIGHT PREEMPT */ |
| 508 | } |
| 509 | #endif |
| 510 | } |
| 511 | } else { |
| 512 | kprintf("sched_ithd: stray interrupt %d\n", intr); |
| 513 | } |
| 514 | } |
| 515 | |
| 516 | /* |
| 517 | * This is run from a periodic SYSTIMER (and thus must be MP safe, the BGL |
| 518 | * might not be held). |
| 519 | */ |
| 520 | static void |
| 521 | ithread_livelock_wakeup(systimer_t st) |
| 522 | { |
| 523 | struct intr_info *info; |
| 524 | |
| 525 | info = &intr_info_ary[(int)st->data]; |
| 526 | if (info->i_state != ISTATE_NOTHREAD) |
| 527 | lwkt_schedule(&info->i_thread); |
| 528 | } |
| 529 | |
| 530 | /* |
| 531 | * This function is called drectly from the ICU or APIC vector code assembly |
| 532 | * to process an interrupt. The critical section and interrupt deferral |
| 533 | * checks have already been done but the function is entered WITHOUT |
| 534 | * a critical section held. The BGL may or may not be held. |
| 535 | * |
| 536 | * Must return non-zero if we do not want the vector code to re-enable |
| 537 | * the interrupt (which we don't if we have to schedule the interrupt) |
| 538 | */ |
| 539 | int ithread_fast_handler(struct intrframe frame); |
| 540 | |
| 541 | int |
| 542 | ithread_fast_handler(struct intrframe frame) |
| 543 | { |
| 544 | int intr; |
| 545 | struct intr_info *info; |
| 546 | struct intrec **list; |
| 547 | int must_schedule; |
| 548 | #ifdef SMP |
| 549 | int got_mplock; |
| 550 | #endif |
| 551 | intrec_t rec, next_rec; |
| 552 | globaldata_t gd; |
| 553 | |
| 554 | intr = frame.if_vec; |
| 555 | gd = mycpu; |
| 556 | |
| 557 | info = &intr_info_ary[intr]; |
| 558 | |
| 559 | /* |
| 560 | * If we are not processing any FAST interrupts, just schedule the thing. |
| 561 | * (since we aren't in a critical section, this can result in a |
| 562 | * preemption) |
| 563 | */ |
| 564 | if (info->i_fast == 0) { |
| 565 | sched_ithd(intr); |
| 566 | return(1); |
| 567 | } |
| 568 | |
| 569 | /* |
| 570 | * This should not normally occur since interrupts ought to be |
| 571 | * masked if the ithread has been scheduled or is running. |
| 572 | */ |
| 573 | if (info->i_running) |
| 574 | return(1); |
| 575 | |
| 576 | /* |
| 577 | * Bump the interrupt nesting level to process any FAST interrupts. |
| 578 | * Obtain the MP lock as necessary. If the MP lock cannot be obtained, |
| 579 | * schedule the interrupt thread to deal with the issue instead. |
| 580 | * |
| 581 | * To reduce overhead, just leave the MP lock held once it has been |
| 582 | * obtained. |
| 583 | */ |
| 584 | crit_enter_gd(gd); |
| 585 | ++gd->gd_intr_nesting_level; |
| 586 | ++gd->gd_cnt.v_intr; |
| 587 | must_schedule = info->i_slow; |
| 588 | #ifdef SMP |
| 589 | got_mplock = 0; |
| 590 | #endif |
| 591 | |
| 592 | list = &info->i_reclist; |
| 593 | for (rec = *list; rec; rec = next_rec) { |
| 594 | next_rec = rec->next; /* rec may be invalid after call */ |
| 595 | |
| 596 | if (rec->intr_flags & INTR_FAST) { |
| 597 | #ifdef SMP |
| 598 | if ((rec->intr_flags & INTR_MPSAFE) == 0 && got_mplock == 0) { |
| 599 | if (try_mplock() == 0) { |
| 600 | int owner; |
| 601 | |
| 602 | /* |
| 603 | * If we couldn't get the MP lock try to forward it |
| 604 | * to the cpu holding the MP lock, setting must_schedule |
| 605 | * to -1 so we do not schedule and also do not unmask |
| 606 | * the interrupt. Otherwise just schedule it. |
| 607 | */ |
| 608 | owner = owner_mplock(); |
| 609 | if (owner >= 0 && owner != gd->gd_cpuid) { |
| 610 | lwkt_send_ipiq_bycpu(owner, forward_fastint_remote, |
| 611 | (void *)intr); |
| 612 | must_schedule = -1; |
| 613 | ++gd->gd_cnt.v_forwarded_ints; |
| 614 | } else { |
| 615 | must_schedule = 1; |
| 616 | } |
| 617 | break; |
| 618 | } |
| 619 | got_mplock = 1; |
| 620 | } |
| 621 | #endif |
| 622 | if (rec->serializer) { |
| 623 | must_schedule += lwkt_serialize_handler_try( |
| 624 | rec->serializer, rec->handler, |
| 625 | rec->argument, &frame); |
| 626 | } else { |
| 627 | rec->handler(rec->argument, &frame); |
| 628 | } |
| 629 | } |
| 630 | } |
| 631 | |
| 632 | /* |
| 633 | * Cleanup |
| 634 | */ |
| 635 | --gd->gd_intr_nesting_level; |
| 636 | #ifdef SMP |
| 637 | if (got_mplock) |
| 638 | rel_mplock(); |
| 639 | #endif |
| 640 | crit_exit_gd(gd); |
| 641 | |
| 642 | /* |
| 643 | * If we had a problem, schedule the thread to catch the missed |
| 644 | * records (it will just re-run all of them). A return value of 0 |
| 645 | * indicates that all handlers have been run and the interrupt can |
| 646 | * be re-enabled, and a non-zero return indicates that the interrupt |
| 647 | * thread controls re-enablement. |
| 648 | */ |
| 649 | if (must_schedule > 0) |
| 650 | sched_ithd(intr); |
| 651 | else if (must_schedule == 0) |
| 652 | ++info->i_count; |
| 653 | return(must_schedule); |
| 654 | } |
| 655 | |
| 656 | #if 0 |
| 657 | |
| 658 | 6: ; \ |
| 659 | /* could not get the MP lock, forward the interrupt */ \ |
| 660 | movl mp_lock, %eax ; /* check race */ \ |
| 661 | cmpl $MP_FREE_LOCK,%eax ; \ |
| 662 | je 2b ; \ |
| 663 | incl PCPU(cnt)+V_FORWARDED_INTS ; \ |
| 664 | subl $12,%esp ; \ |
| 665 | movl $irq_num,8(%esp) ; \ |
| 666 | movl $forward_fastint_remote,4(%esp) ; \ |
| 667 | movl %eax,(%esp) ; \ |
| 668 | call lwkt_send_ipiq_bycpu ; \ |
| 669 | addl $12,%esp ; \ |
| 670 | jmp 5f ; |
| 671 | |
| 672 | #endif |
| 673 | |
| 674 | |
| 675 | /* |
| 676 | * Interrupt threads run this as their main loop. |
| 677 | * |
| 678 | * The handler begins execution outside a critical section and with the BGL |
| 679 | * held. |
| 680 | * |
| 681 | * The i_running state starts at 0. When an interrupt occurs, the hardware |
| 682 | * interrupt is disabled and sched_ithd() The HW interrupt remains disabled |
| 683 | * until all routines have run. We then call ithread_done() to reenable |
| 684 | * the HW interrupt and deschedule us until the next interrupt. |
| 685 | * |
| 686 | * We are responsible for atomically checking i_running and ithread_done() |
| 687 | * is responsible for atomically checking for platform-specific delayed |
| 688 | * interrupts. i_running for our irq is only set in the context of our cpu, |
| 689 | * so a critical section is a sufficient interlock. |
| 690 | */ |
| 691 | #define LIVELOCK_TIMEFRAME(freq) ((freq) >> 2) /* 1/4 second */ |
| 692 | |
| 693 | static void |
| 694 | ithread_handler(void *arg) |
| 695 | { |
| 696 | struct intr_info *info; |
| 697 | int use_limit; |
| 698 | int lticks; |
| 699 | int lcount; |
| 700 | int intr; |
| 701 | int mpheld; |
| 702 | struct intrec **list; |
| 703 | intrec_t rec, nrec; |
| 704 | globaldata_t gd; |
| 705 | struct systimer ill_timer; /* enforced freq. timer */ |
| 706 | u_int ill_count; /* interrupt livelock counter */ |
| 707 | |
| 708 | ill_count = 0; |
| 709 | lticks = ticks; |
| 710 | lcount = 0; |
| 711 | intr = (int)arg; |
| 712 | info = &intr_info_ary[intr]; |
| 713 | list = &info->i_reclist; |
| 714 | gd = mycpu; |
| 715 | |
| 716 | /* |
| 717 | * The loop must be entered with one critical section held. The thread |
| 718 | * is created with TDF_MPSAFE so the MP lock is not held on start. |
| 719 | */ |
| 720 | crit_enter_gd(gd); |
| 721 | mpheld = 0; |
| 722 | |
| 723 | for (;;) { |
| 724 | /* |
| 725 | * The chain is only considered MPSAFE if all its interrupt handlers |
| 726 | * are MPSAFE. However, if intr_mpsafe has been turned off we |
| 727 | * always operate with the BGL. |
| 728 | */ |
| 729 | #ifdef SMP |
| 730 | if (intr_mpsafe == 0) { |
| 731 | if (mpheld == 0) { |
| 732 | get_mplock(); |
| 733 | mpheld = 1; |
| 734 | } |
| 735 | } else if (info->i_mplock_required != mpheld) { |
| 736 | if (info->i_mplock_required) { |
| 737 | KKASSERT(mpheld == 0); |
| 738 | get_mplock(); |
| 739 | mpheld = 1; |
| 740 | } else { |
| 741 | KKASSERT(mpheld != 0); |
| 742 | rel_mplock(); |
| 743 | mpheld = 0; |
| 744 | } |
| 745 | } |
| 746 | #endif |
| 747 | |
| 748 | /* |
| 749 | * If an interrupt is pending, clear i_running and execute the |
| 750 | * handlers. Note that certain types of interrupts can re-trigger |
| 751 | * and set i_running again. |
| 752 | * |
| 753 | * Each handler is run in a critical section. Note that we run both |
| 754 | * FAST and SLOW designated service routines. |
| 755 | */ |
| 756 | if (info->i_running) { |
| 757 | ++ill_count; |
| 758 | info->i_running = 0; |
| 759 | |
| 760 | for (rec = *list; rec; rec = nrec) { |
| 761 | nrec = rec->next; |
| 762 | if (rec->serializer) { |
| 763 | lwkt_serialize_handler_call(rec->serializer, rec->handler, |
| 764 | rec->argument, NULL); |
| 765 | } else { |
| 766 | rec->handler(rec->argument, NULL); |
| 767 | } |
| 768 | } |
| 769 | } |
| 770 | |
| 771 | /* |
| 772 | * This is our interrupt hook to add rate randomness to the random |
| 773 | * number generator. |
| 774 | */ |
| 775 | if (info->i_random.sc_enabled > 0) |
| 776 | add_interrupt_randomness(intr); |
| 777 | |
| 778 | /* |
| 779 | * Unmask the interrupt to allow it to trigger again. This only |
| 780 | * applies to certain types of interrupts (typ level interrupts). |
| 781 | * This can result in the interrupt retriggering, but the retrigger |
| 782 | * will not be processed until we cycle our critical section. |
| 783 | * |
| 784 | * Only unmask interrupts while handlers are installed. It is |
| 785 | * possible to hit a situation where no handlers are installed |
| 786 | * due to a device driver livelocking and then tearing down its |
| 787 | * interrupt on close (the parallel bus being a good example). |
| 788 | */ |
| 789 | if (*list) |
| 790 | machintr_intren(intr); |
| 791 | |
| 792 | /* |
| 793 | * Do a quick exit/enter to catch any higher-priority interrupt |
| 794 | * sources, such as the statclock, so thread time accounting |
| 795 | * will still work. This may also cause an interrupt to re-trigger. |
| 796 | */ |
| 797 | crit_exit_gd(gd); |
| 798 | crit_enter_gd(gd); |
| 799 | |
| 800 | /* |
| 801 | * LIVELOCK STATE MACHINE |
| 802 | */ |
| 803 | switch(info->i_state) { |
| 804 | case ISTATE_NORMAL: |
| 805 | /* |
| 806 | * Calculate a running average every tick. |
| 807 | */ |
| 808 | if (lticks != ticks) { |
| 809 | lticks = ticks; |
| 810 | ill_count -= ill_count / hz; |
| 811 | } |
| 812 | |
| 813 | /* |
| 814 | * If we did not exceed the frequency limit, we are done. |
| 815 | * If the interrupt has not retriggered we deschedule ourselves. |
| 816 | */ |
| 817 | if (ill_count <= livelock_limit) { |
| 818 | if (info->i_running == 0) { |
| 819 | lwkt_deschedule_self(gd->gd_curthread); |
| 820 | lwkt_switch(); |
| 821 | } |
| 822 | break; |
| 823 | } |
| 824 | |
| 825 | /* |
| 826 | * Otherwise we are livelocked. Set up a periodic systimer |
| 827 | * to wake the thread up at the limit frequency. |
| 828 | */ |
| 829 | kprintf("intr %d at %d > %d hz, livelocked limit engaged!\n", |
| 830 | intr, ill_count, livelock_limit); |
| 831 | info->i_state = ISTATE_LIVELOCKED; |
| 832 | if ((use_limit = livelock_limit) < 100) |
| 833 | use_limit = 100; |
| 834 | else if (use_limit > 500000) |
| 835 | use_limit = 500000; |
| 836 | systimer_init_periodic(&ill_timer, ithread_livelock_wakeup, |
| 837 | (void *)intr, use_limit); |
| 838 | lcount = 0; |
| 839 | /* fall through */ |
| 840 | case ISTATE_LIVELOCKED: |
| 841 | /* |
| 842 | * Wait for our periodic timer to go off. Since the interrupt |
| 843 | * has re-armed it can still set i_running, but it will not |
| 844 | * reschedule us while we are in a livelocked state. |
| 845 | */ |
| 846 | lwkt_deschedule_self(gd->gd_curthread); |
| 847 | lwkt_switch(); |
| 848 | |
| 849 | /* |
| 850 | * Check to see if the livelock condition no longer applies. |
| 851 | * The interrupt must be able to operate normally for one |
| 852 | * full second before we restore normal operation. |
| 853 | */ |
| 854 | if (lticks != ticks) { |
| 855 | lticks = ticks; |
| 856 | if (ill_count < livelock_lowater) { |
| 857 | if (++lcount >= hz) { |
| 858 | info->i_state = ISTATE_NORMAL; |
| 859 | systimer_del(&ill_timer); |
| 860 | kprintf("intr %d at %d < %d hz, livelock removed\n", |
| 861 | intr, ill_count, livelock_lowater); |
| 862 | } |
| 863 | } else { |
| 864 | lcount = 0; |
| 865 | } |
| 866 | ill_count -= ill_count / hz; |
| 867 | } |
| 868 | break; |
| 869 | } |
| 870 | } |
| 871 | /* not reached */ |
| 872 | } |
| 873 | |
| 874 | /* |
| 875 | * Emergency interrupt polling thread. The thread begins execution |
| 876 | * outside a critical section with the BGL held. |
| 877 | * |
| 878 | * If emergency interrupt polling is enabled, this thread will |
| 879 | * execute all system interrupts not marked INTR_NOPOLL at the |
| 880 | * specified polling frequency. |
| 881 | * |
| 882 | * WARNING! This thread runs *ALL* interrupt service routines that |
| 883 | * are not marked INTR_NOPOLL, which basically means everything except |
| 884 | * the 8254 clock interrupt and the ATA interrupt. It has very high |
| 885 | * overhead and should only be used in situations where the machine |
| 886 | * cannot otherwise be made to work. Due to the severe performance |
| 887 | * degredation, it should not be enabled on production machines. |
| 888 | */ |
| 889 | static void |
| 890 | ithread_emergency(void *arg __unused) |
| 891 | { |
| 892 | struct intr_info *info; |
| 893 | intrec_t rec, nrec; |
| 894 | int intr; |
| 895 | |
| 896 | for (;;) { |
| 897 | for (intr = 0; intr < max_installed_hard_intr; ++intr) { |
| 898 | info = &intr_info_ary[intr]; |
| 899 | for (rec = info->i_reclist; rec; rec = nrec) { |
| 900 | if ((rec->intr_flags & INTR_NOPOLL) == 0) { |
| 901 | if (rec->serializer) { |
| 902 | lwkt_serialize_handler_call(rec->serializer, |
| 903 | rec->handler, rec->argument, NULL); |
| 904 | } else { |
| 905 | rec->handler(rec->argument, NULL); |
| 906 | } |
| 907 | } |
| 908 | nrec = rec->next; |
| 909 | } |
| 910 | } |
| 911 | lwkt_deschedule_self(curthread); |
| 912 | lwkt_switch(); |
| 913 | } |
| 914 | } |
| 915 | |
| 916 | /* |
| 917 | * Systimer callback - schedule the emergency interrupt poll thread |
| 918 | * if emergency polling is enabled. |
| 919 | */ |
| 920 | static |
| 921 | void |
| 922 | emergency_intr_timer_callback(systimer_t info, struct intrframe *frame __unused) |
| 923 | { |
| 924 | if (emergency_intr_enable) |
| 925 | lwkt_schedule(info->data); |
| 926 | } |
| 927 | |
| 928 | /* |
| 929 | * Sysctls used by systat and others: hw.intrnames and hw.intrcnt. |
| 930 | * The data for this machine dependent, and the declarations are in machine |
| 931 | * dependent code. The layout of intrnames and intrcnt however is machine |
| 932 | * independent. |
| 933 | * |
| 934 | * We do not know the length of intrcnt and intrnames at compile time, so |
| 935 | * calculate things at run time. |
| 936 | */ |
| 937 | |
| 938 | static int |
| 939 | sysctl_intrnames(SYSCTL_HANDLER_ARGS) |
| 940 | { |
| 941 | struct intr_info *info; |
| 942 | intrec_t rec; |
| 943 | int error = 0; |
| 944 | int len; |
| 945 | int intr; |
| 946 | char buf[64]; |
| 947 | |
| 948 | for (intr = 0; error == 0 && intr < MAX_INTS; ++intr) { |
| 949 | info = &intr_info_ary[intr]; |
| 950 | |
| 951 | len = 0; |
| 952 | buf[0] = 0; |
| 953 | for (rec = info->i_reclist; rec; rec = rec->next) { |
| 954 | ksnprintf(buf + len, sizeof(buf) - len, "%s%s", |
| 955 | (len ? "/" : ""), rec->name); |
| 956 | len += strlen(buf + len); |
| 957 | } |
| 958 | if (len == 0) { |
| 959 | ksnprintf(buf, sizeof(buf), "irq%d", intr); |
| 960 | len = strlen(buf); |
| 961 | } |
| 962 | error = SYSCTL_OUT(req, buf, len + 1); |
| 963 | } |
| 964 | return (error); |
| 965 | } |
| 966 | |
| 967 | |
| 968 | SYSCTL_PROC(_hw, OID_AUTO, intrnames, CTLTYPE_OPAQUE | CTLFLAG_RD, |
| 969 | NULL, 0, sysctl_intrnames, "", "Interrupt Names"); |
| 970 | |
| 971 | static int |
| 972 | sysctl_intrcnt(SYSCTL_HANDLER_ARGS) |
| 973 | { |
| 974 | struct intr_info *info; |
| 975 | int error = 0; |
| 976 | int intr; |
| 977 | |
| 978 | for (intr = 0; intr < max_installed_hard_intr; ++intr) { |
| 979 | info = &intr_info_ary[intr]; |
| 980 | |
| 981 | error = SYSCTL_OUT(req, &info->i_count, sizeof(info->i_count)); |
| 982 | if (error) |
| 983 | goto failed; |
| 984 | } |
| 985 | for (intr = FIRST_SOFTINT; intr < max_installed_soft_intr; ++intr) { |
| 986 | info = &intr_info_ary[intr]; |
| 987 | |
| 988 | error = SYSCTL_OUT(req, &info->i_count, sizeof(info->i_count)); |
| 989 | if (error) |
| 990 | goto failed; |
| 991 | } |
| 992 | failed: |
| 993 | return(error); |
| 994 | } |
| 995 | |
| 996 | SYSCTL_PROC(_hw, OID_AUTO, intrcnt, CTLTYPE_OPAQUE | CTLFLAG_RD, |
| 997 | NULL, 0, sysctl_intrcnt, "", "Interrupt Counts"); |
| 998 | |