1 /*-
2 * Copyright (c) 2003 John Baldwin <jhb@FreeBSD.org>
3 * Copyright (c) 1996, by Steve Passe
4 * All rights reserved.
5 *
6 * Redistribution and use in source and binary forms, with or without
7 * modification, are permitted provided that the following conditions
8 * are met:
9 * 1. Redistributions of source code must retain the above copyright
10 * notice, this list of conditions and the following disclaimer.
11 * 2. The name of the developer may NOT be used to endorse or promote products
12 * derived from this software without specific prior written permission.
13 * 3. Neither the name of the author nor the names of any co-contributors
14 * may be used to endorse or promote products derived from this software
15 * without specific prior written permission.
16 *
17 * THIS SOFTWARE IS PROVIDED BY THE AUTHOR AND CONTRIBUTORS ``AS IS'' AND
18 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
19 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
20 * ARE DISCLAIMED. IN NO EVENT SHALL THE AUTHOR OR CONTRIBUTORS BE LIABLE
21 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
22 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
23 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
24 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
25 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
26 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
27 * SUCH DAMAGE.
28 */
29
30 /*
31 * Local APIC support on Pentium and later processors.
32 */
33
34 #include <sys/cdefs.h>
35 __FBSDID("$FreeBSD$");
36
37 #include "opt_hwpmc_hooks.h"
38 #include "opt_kdtrace.h"
39
40 #include "opt_ddb.h"
41
42 #include <sys/param.h>
43 #include <sys/systm.h>
44 #include <sys/bus.h>
45 #include <sys/kernel.h>
46 #include <sys/lock.h>
47 #include <sys/mutex.h>
48 #include <sys/pcpu.h>
49 #include <sys/smp.h>
50
51 #include <vm/vm.h>
52 #include <vm/pmap.h>
53
54 #include <machine/apicreg.h>
55 #include <machine/cpu.h>
56 #include <machine/cputypes.h>
57 #include <machine/frame.h>
58 #include <machine/intr_machdep.h>
59 #include <machine/apicvar.h>
60 #include <machine/md_var.h>
61 #include <machine/smp.h>
62 #include <machine/specialreg.h>
63
64 #ifdef DDB
65 #include <sys/interrupt.h>
66 #include <ddb/ddb.h>
67 #endif
68
69 #ifdef KDTRACE_HOOKS
70 #include <sys/dtrace_bsd.h>
71 cyclic_clock_func_t lapic_cyclic_clock_func[MAXCPU];
72 #endif
73
74 /* Sanity checks on IDT vectors. */
75 CTASSERT(APIC_IO_INTS + APIC_NUM_IOINTS == APIC_TIMER_INT);
76 CTASSERT(APIC_TIMER_INT < APIC_LOCAL_INTS);
77 CTASSERT(APIC_LOCAL_INTS == 240);
78 CTASSERT(IPI_STOP < APIC_SPURIOUS_INT);
79
80 /* Magic IRQ values for the timer and syscalls. */
81 #define IRQ_TIMER (NUM_IO_INTS + 1)
82 #define IRQ_SYSCALL (NUM_IO_INTS + 2)
83
84 /*
85 * Support for local APICs. Local APICs manage interrupts on each
86 * individual processor as opposed to I/O APICs which receive interrupts
87 * from I/O devices and then forward them on to the local APICs.
88 *
89 * Local APICs can also send interrupts to each other thus providing the
90 * mechanism for IPIs.
91 */
92
93 struct lvt {
94 u_int lvt_edgetrigger:1;
95 u_int lvt_activehi:1;
96 u_int lvt_masked:1;
97 u_int lvt_active:1;
98 u_int lvt_mode:16;
99 u_int lvt_vector:8;
100 };
101
102 struct lapic {
103 struct lvt la_lvts[LVT_MAX + 1];
104 u_int la_id:8;
105 u_int la_cluster:4;
106 u_int la_cluster_id:2;
107 u_int la_present:1;
108 u_long *la_timer_count;
109 u_long la_hard_ticks;
110 u_long la_stat_ticks;
111 u_long la_prof_ticks;
112 } static lapics[MAX_APIC_ID + 1];
113
114 /* XXX: should thermal be an NMI? */
115
116 /* Global defaults for local APIC LVT entries. */
117 static struct lvt lvts[LVT_MAX + 1] = {
118 { 1, 1, 1, 1, APIC_LVT_DM_EXTINT, 0 }, /* LINT0: masked ExtINT */
119 { 1, 1, 0, 1, APIC_LVT_DM_NMI, 0 }, /* LINT1: NMI */
120 { 1, 1, 1, 1, APIC_LVT_DM_FIXED, APIC_TIMER_INT }, /* Timer */
121 { 1, 1, 1, 1, APIC_LVT_DM_FIXED, APIC_ERROR_INT }, /* Error */
122 { 1, 1, 0, 1, APIC_LVT_DM_NMI, 0 }, /* PMC */
123 { 1, 1, 1, 1, APIC_LVT_DM_FIXED, APIC_THERMAL_INT }, /* Thermal */
124 };
125
126 static inthand_t *ioint_handlers[] = {
127 NULL, /* 0 - 31 */
128 IDTVEC(apic_isr1), /* 32 - 63 */
129 IDTVEC(apic_isr2), /* 64 - 95 */
130 IDTVEC(apic_isr3), /* 96 - 127 */
131 IDTVEC(apic_isr4), /* 128 - 159 */
132 IDTVEC(apic_isr5), /* 160 - 191 */
133 IDTVEC(apic_isr6), /* 192 - 223 */
134 IDTVEC(apic_isr7), /* 224 - 255 */
135 };
136
137 /* Include IDT_SYSCALL to make indexing easier. */
138 static u_int ioint_irqs[APIC_NUM_IOINTS + 1];
139
140 static u_int32_t lapic_timer_divisors[] = {
141 APIC_TDCR_1, APIC_TDCR_2, APIC_TDCR_4, APIC_TDCR_8, APIC_TDCR_16,
142 APIC_TDCR_32, APIC_TDCR_64, APIC_TDCR_128
143 };
144
145 extern inthand_t IDTVEC(rsvd);
146
147 volatile lapic_t *lapic;
148 vm_paddr_t lapic_paddr;
149 static u_long lapic_timer_divisor, lapic_timer_period, lapic_timer_hz;
150
151 static void lapic_enable(void);
152 static void lapic_resume(struct pic *pic);
153 static void lapic_timer_enable_intr(void);
154 static void lapic_timer_oneshot(u_int count);
155 static void lapic_timer_periodic(u_int count);
156 static void lapic_timer_set_divisor(u_int divisor);
157 static uint32_t lvt_mode(struct lapic *la, u_int pin, uint32_t value);
158
159 struct pic lapic_pic = { .pic_resume = lapic_resume };
160
161 static uint32_t
162 lvt_mode(struct lapic *la, u_int pin, uint32_t value)
163 {
164 struct lvt *lvt;
165
166 KASSERT(pin <= LVT_MAX, ("%s: pin %u out of range", __func__, pin));
167 if (la->la_lvts[pin].lvt_active)
168 lvt = &la->la_lvts[pin];
169 else
170 lvt = &lvts[pin];
171
172 value &= ~(APIC_LVT_M | APIC_LVT_TM | APIC_LVT_IIPP | APIC_LVT_DM |
173 APIC_LVT_VECTOR);
174 if (lvt->lvt_edgetrigger == 0)
175 value |= APIC_LVT_TM;
176 if (lvt->lvt_activehi == 0)
177 value |= APIC_LVT_IIPP_INTALO;
178 if (lvt->lvt_masked)
179 value |= APIC_LVT_M;
180 value |= lvt->lvt_mode;
181 switch (lvt->lvt_mode) {
182 case APIC_LVT_DM_NMI:
183 case APIC_LVT_DM_SMI:
184 case APIC_LVT_DM_INIT:
185 case APIC_LVT_DM_EXTINT:
186 if (!lvt->lvt_edgetrigger) {
187 printf("lapic%u: Forcing LINT%u to edge trigger\n",
188 la->la_id, pin);
189 value |= APIC_LVT_TM;
190 }
191 /* Use a vector of 0. */
192 break;
193 case APIC_LVT_DM_FIXED:
194 value |= lvt->lvt_vector;
195 break;
196 default:
197 panic("bad APIC LVT delivery mode: %#x\n", value);
198 }
199 return (value);
200 }
201
202 /*
203 * Map the local APIC and setup necessary interrupt vectors.
204 */
205 void
206 lapic_init(vm_paddr_t addr)
207 {
208
209 /* Map the local APIC and setup the spurious interrupt handler. */
210 KASSERT(trunc_page(addr) == addr,
211 ("local APIC not aligned on a page boundary"));
212 lapic = pmap_mapdev(addr, sizeof(lapic_t));
213 lapic_paddr = addr;
214 setidt(APIC_SPURIOUS_INT, IDTVEC(spuriousint), SDT_SYS386IGT, SEL_KPL,
215 GSEL(GCODE_SEL, SEL_KPL));
216
217 /* Perform basic initialization of the BSP's local APIC. */
218 lapic_enable();
219 ioint_irqs[IDT_SYSCALL - APIC_IO_INTS] = IRQ_SYSCALL;
220
221 /* Set BSP's per-CPU local APIC ID. */
222 PCPU_SET(apic_id, lapic_id());
223
224 /* Local APIC timer interrupt. */
225 setidt(APIC_TIMER_INT, IDTVEC(timerint), SDT_SYS386IGT, SEL_KPL,
226 GSEL(GCODE_SEL, SEL_KPL));
227 ioint_irqs[APIC_TIMER_INT - APIC_IO_INTS] = IRQ_TIMER;
228
229 /* XXX: error/thermal interrupts */
230 }
231
232 /*
233 * Create a local APIC instance.
234 */
235 void
236 lapic_create(u_int apic_id, int boot_cpu)
237 {
238 int i;
239
240 if (apic_id > MAX_APIC_ID) {
241 printf("APIC: Ignoring local APIC with ID %d\n", apic_id);
242 if (boot_cpu)
243 panic("Can't ignore BSP");
244 return;
245 }
246 KASSERT(!lapics[apic_id].la_present, ("duplicate local APIC %u",
247 apic_id));
248
249 /*
250 * Assume no local LVT overrides and a cluster of 0 and
251 * intra-cluster ID of 0.
252 */
253 lapics[apic_id].la_present = 1;
254 lapics[apic_id].la_id = apic_id;
255 for (i = 0; i < LVT_MAX; i++) {
256 lapics[apic_id].la_lvts[i] = lvts[i];
257 lapics[apic_id].la_lvts[i].lvt_active = 0;
258 }
259
260 #ifdef SMP
261 cpu_add(apic_id, boot_cpu);
262 #endif
263 }
264
265 /*
266 * Dump contents of local APIC registers
267 */
268 void
269 lapic_dump(const char* str)
270 {
271
272 printf("cpu%d %s:\n", PCPU_GET(cpuid), str);
273 printf(" ID: 0x%08x VER: 0x%08x LDR: 0x%08x DFR: 0x%08x\n",
274 lapic->id, lapic->version, lapic->ldr, lapic->dfr);
275 printf(" lint0: 0x%08x lint1: 0x%08x TPR: 0x%08x SVR: 0x%08x\n",
276 lapic->lvt_lint0, lapic->lvt_lint1, lapic->tpr, lapic->svr);
277 printf(" timer: 0x%08x therm: 0x%08x err: 0x%08x pcm: 0x%08x\n",
278 lapic->lvt_timer, lapic->lvt_thermal, lapic->lvt_error,
279 lapic->lvt_pcint);
280 }
281
282 void
283 lapic_setup(int boot)
284 {
285 struct lapic *la;
286 u_int32_t maxlvt;
287 register_t eflags;
288 char buf[MAXCOMLEN + 1];
289
290 la = &lapics[lapic_id()];
291 KASSERT(la->la_present, ("missing APIC structure"));
292 eflags = intr_disable();
293 maxlvt = (lapic->version & APIC_VER_MAXLVT) >> MAXLVTSHIFT;
294
295 /* Initialize the TPR to allow all interrupts. */
296 lapic_set_tpr(0);
297
298 /* Setup spurious vector and enable the local APIC. */
299 lapic_enable();
300
301 /* Program LINT[01] LVT entries. */
302 lapic->lvt_lint0 = lvt_mode(la, LVT_LINT0, lapic->lvt_lint0);
303 lapic->lvt_lint1 = lvt_mode(la, LVT_LINT1, lapic->lvt_lint1);
304 #ifdef HWPMC_HOOKS
305 /* Program the PMC LVT entry if present. */
306 if (maxlvt >= LVT_PMC)
307 lapic->lvt_pcint = lvt_mode(la, LVT_PMC, lapic->lvt_pcint);
308 #endif
309
310 /* Program timer LVT and setup handler. */
311 lapic->lvt_timer = lvt_mode(la, LVT_TIMER, lapic->lvt_timer);
312 if (boot) {
313 snprintf(buf, sizeof(buf), "cpu%d: timer", PCPU_GET(cpuid));
314 intrcnt_add(buf, &la->la_timer_count);
315 }
316
317 /* We don't setup the timer during boot on the BSP until later. */
318 if (!(boot && PCPU_GET(cpuid) == 0)) {
319 KASSERT(lapic_timer_period != 0, ("lapic%u: zero divisor",
320 lapic_id()));
321 lapic_timer_set_divisor(lapic_timer_divisor);
322 lapic_timer_periodic(lapic_timer_period);
323 lapic_timer_enable_intr();
324 }
325
326 /* XXX: Error and thermal LVTs */
327
328 if (strcmp(cpu_vendor, "AuthenticAMD") == 0) {
329 /*
330 * Detect the presence of C1E capability mostly on latest
331 * dual-cores (or future) k8 family. This feature renders
332 * the local APIC timer dead, so we disable it by reading
333 * the Interrupt Pending Message register and clearing both
334 * C1eOnCmpHalt (bit 28) and SmiOnCmpHalt (bit 27).
335 *
336 * Reference:
337 * "BIOS and Kernel Developer's Guide for AMD NPT
338 * Family 0Fh Processors"
339 * #32559 revision 3.00
340 */
341 if ((cpu_id & 0x00000f00) == 0x00000f00 &&
342 (cpu_id & 0x0fff0000) >= 0x00040000) {
343 uint64_t msr;
344
345 msr = rdmsr(0xc0010055);
346 if (msr & 0x18000000)
347 wrmsr(0xc0010055, msr & ~0x18000000ULL);
348 }
349 }
350
351 intr_restore(eflags);
352 }
353
354 /*
355 * Called by cpu_initclocks() on the BSP to setup the local APIC timer so
356 * that it can drive hardclock, statclock, and profclock. This function
357 * returns true if it is able to use the local APIC timer to drive the
358 * clocks and false if it is not able.
359 */
360 int
361 lapic_setup_clock(void)
362 {
363 u_long value;
364
365 /* Can't drive the timer without a local APIC. */
366 if (lapic == NULL)
367 return (0);
368
369 /* Start off with a divisor of 2 (power on reset default). */
370 lapic_timer_divisor = 2;
371
372 /* Try to calibrate the local APIC timer. */
373 do {
374 lapic_timer_set_divisor(lapic_timer_divisor);
375 lapic_timer_oneshot(APIC_TIMER_MAX_COUNT);
376 DELAY(2000000);
377 value = APIC_TIMER_MAX_COUNT - lapic->ccr_timer;
378 if (value != APIC_TIMER_MAX_COUNT)
379 break;
380 lapic_timer_divisor <<= 1;
381 } while (lapic_timer_divisor <= 128);
382 if (lapic_timer_divisor > 128)
383 panic("lapic: Divisor too big");
384 value /= 2;
385 if (bootverbose)
386 printf("lapic: Divisor %lu, Frequency %lu hz\n",
387 lapic_timer_divisor, value);
388
389 /*
390 * We want to run stathz in the neighborhood of 128hz. We would
391 * like profhz to run as often as possible, so we let it run on
392 * each clock tick. We try to honor the requested 'hz' value as
393 * much as possible.
394 *
395 * If 'hz' is above 1500, then we just let the lapic timer
396 * (and profhz) run at hz. If 'hz' is below 1500 but above
397 * 750, then we let the lapic timer run at 2 * 'hz'. If 'hz'
398 * is below 750 then we let the lapic timer run at 4 * 'hz'.
399 */
400 if (hz >= 1500)
401 lapic_timer_hz = hz;
402 else if (hz >= 750)
403 lapic_timer_hz = hz * 2;
404 else
405 lapic_timer_hz = hz * 4;
406 if (lapic_timer_hz < 128)
407 stathz = lapic_timer_hz;
408 else
409 stathz = lapic_timer_hz / (lapic_timer_hz / 128);
410 profhz = lapic_timer_hz;
411 lapic_timer_period = value / lapic_timer_hz;
412
413 /*
414 * Start up the timer on the BSP. The APs will kick off their
415 * timer during lapic_setup().
416 */
417 lapic_timer_periodic(lapic_timer_period);
418 lapic_timer_enable_intr();
419 return (1);
420 }
421
422 void
423 lapic_disable(void)
424 {
425 uint32_t value;
426
427 /* Software disable the local APIC. */
428 value = lapic->svr;
429 value &= ~APIC_SVR_SWEN;
430 lapic->svr = value;
431 }
432
433 static void
434 lapic_enable(void)
435 {
436 u_int32_t value;
437
438 /* Program the spurious vector to enable the local APIC. */
439 value = lapic->svr;
440 value &= ~(APIC_SVR_VECTOR | APIC_SVR_FOCUS);
441 value |= (APIC_SVR_FEN | APIC_SVR_SWEN | APIC_SPURIOUS_INT);
442 lapic->svr = value;
443 }
444
445 /* Reset the local APIC on the BSP during resume. */
446 static void
447 lapic_resume(struct pic *pic)
448 {
449
450 lapic_setup(0);
451 }
452
453 int
454 lapic_id(void)
455 {
456
457 KASSERT(lapic != NULL, ("local APIC is not mapped"));
458 return (lapic->id >> APIC_ID_SHIFT);
459 }
460
461 int
462 lapic_intr_pending(u_int vector)
463 {
464 volatile u_int32_t *irr;
465
466 /*
467 * The IRR registers are an array of 128-bit registers each of
468 * which only describes 32 interrupts in the low 32 bits.. Thus,
469 * we divide the vector by 32 to get the 128-bit index. We then
470 * multiply that index by 4 to get the equivalent index from
471 * treating the IRR as an array of 32-bit registers. Finally, we
472 * modulus the vector by 32 to determine the individual bit to
473 * test.
474 */
475 irr = &lapic->irr0;
476 return (irr[(vector / 32) * 4] & 1 << (vector % 32));
477 }
478
479 void
480 lapic_set_logical_id(u_int apic_id, u_int cluster, u_int cluster_id)
481 {
482 struct lapic *la;
483
484 KASSERT(lapics[apic_id].la_present, ("%s: APIC %u doesn't exist",
485 __func__, apic_id));
486 KASSERT(cluster <= APIC_MAX_CLUSTER, ("%s: cluster %u too big",
487 __func__, cluster));
488 KASSERT(cluster_id <= APIC_MAX_INTRACLUSTER_ID,
489 ("%s: intra cluster id %u too big", __func__, cluster_id));
490 la = &lapics[apic_id];
491 la->la_cluster = cluster;
492 la->la_cluster_id = cluster_id;
493 }
494
495 int
496 lapic_set_lvt_mask(u_int apic_id, u_int pin, u_char masked)
497 {
498
499 if (pin > LVT_MAX)
500 return (EINVAL);
501 if (apic_id == APIC_ID_ALL) {
502 lvts[pin].lvt_masked = masked;
503 if (bootverbose)
504 printf("lapic:");
505 } else {
506 KASSERT(lapics[apic_id].la_present,
507 ("%s: missing APIC %u", __func__, apic_id));
508 lapics[apic_id].la_lvts[pin].lvt_masked = masked;
509 lapics[apic_id].la_lvts[pin].lvt_active = 1;
510 if (bootverbose)
511 printf("lapic%u:", apic_id);
512 }
513 if (bootverbose)
514 printf(" LINT%u %s\n", pin, masked ? "masked" : "unmasked");
515 return (0);
516 }
517
518 int
519 lapic_set_lvt_mode(u_int apic_id, u_int pin, u_int32_t mode)
520 {
521 struct lvt *lvt;
522
523 if (pin > LVT_MAX)
524 return (EINVAL);
525 if (apic_id == APIC_ID_ALL) {
526 lvt = &lvts[pin];
527 if (bootverbose)
528 printf("lapic:");
529 } else {
530 KASSERT(lapics[apic_id].la_present,
531 ("%s: missing APIC %u", __func__, apic_id));
532 lvt = &lapics[apic_id].la_lvts[pin];
533 lvt->lvt_active = 1;
534 if (bootverbose)
535 printf("lapic%u:", apic_id);
536 }
537 lvt->lvt_mode = mode;
538 switch (mode) {
539 case APIC_LVT_DM_NMI:
540 case APIC_LVT_DM_SMI:
541 case APIC_LVT_DM_INIT:
542 case APIC_LVT_DM_EXTINT:
543 lvt->lvt_edgetrigger = 1;
544 lvt->lvt_activehi = 1;
545 if (mode == APIC_LVT_DM_EXTINT)
546 lvt->lvt_masked = 1;
547 else
548 lvt->lvt_masked = 0;
549 break;
550 default:
551 panic("Unsupported delivery mode: 0x%x\n", mode);
552 }
553 if (bootverbose) {
554 printf(" Routing ");
555 switch (mode) {
556 case APIC_LVT_DM_NMI:
557 printf("NMI");
558 break;
559 case APIC_LVT_DM_SMI:
560 printf("SMI");
561 break;
562 case APIC_LVT_DM_INIT:
563 printf("INIT");
564 break;
565 case APIC_LVT_DM_EXTINT:
566 printf("ExtINT");
567 break;
568 }
569 printf(" -> LINT%u\n", pin);
570 }
571 return (0);
572 }
573
574 int
575 lapic_set_lvt_polarity(u_int apic_id, u_int pin, enum intr_polarity pol)
576 {
577
578 if (pin > LVT_MAX || pol == INTR_POLARITY_CONFORM)
579 return (EINVAL);
580 if (apic_id == APIC_ID_ALL) {
581 lvts[pin].lvt_activehi = (pol == INTR_POLARITY_HIGH);
582 if (bootverbose)
583 printf("lapic:");
584 } else {
585 KASSERT(lapics[apic_id].la_present,
586 ("%s: missing APIC %u", __func__, apic_id));
587 lapics[apic_id].la_lvts[pin].lvt_active = 1;
588 lapics[apic_id].la_lvts[pin].lvt_activehi =
589 (pol == INTR_POLARITY_HIGH);
590 if (bootverbose)
591 printf("lapic%u:", apic_id);
592 }
593 if (bootverbose)
594 printf(" LINT%u polarity: %s\n", pin,
595 pol == INTR_POLARITY_HIGH ? "high" : "low");
596 return (0);
597 }
598
599 int
600 lapic_set_lvt_triggermode(u_int apic_id, u_int pin, enum intr_trigger trigger)
601 {
602
603 if (pin > LVT_MAX || trigger == INTR_TRIGGER_CONFORM)
604 return (EINVAL);
605 if (apic_id == APIC_ID_ALL) {
606 lvts[pin].lvt_edgetrigger = (trigger == INTR_TRIGGER_EDGE);
607 if (bootverbose)
608 printf("lapic:");
609 } else {
610 KASSERT(lapics[apic_id].la_present,
611 ("%s: missing APIC %u", __func__, apic_id));
612 lapics[apic_id].la_lvts[pin].lvt_edgetrigger =
613 (trigger == INTR_TRIGGER_EDGE);
614 lapics[apic_id].la_lvts[pin].lvt_active = 1;
615 if (bootverbose)
616 printf("lapic%u:", apic_id);
617 }
618 if (bootverbose)
619 printf(" LINT%u trigger: %s\n", pin,
620 trigger == INTR_TRIGGER_EDGE ? "edge" : "level");
621 return (0);
622 }
623
624 /*
625 * Adjust the TPR of the current CPU so that it blocks all interrupts below
626 * the passed in vector.
627 */
628 void
629 lapic_set_tpr(u_int vector)
630 {
631 #ifdef CHEAP_TPR
632 lapic->tpr = vector;
633 #else
634 u_int32_t tpr;
635
636 tpr = lapic->tpr & ~APIC_TPR_PRIO;
637 tpr |= vector;
638 lapic->tpr = tpr;
639 #endif
640 }
641
642 void
643 lapic_eoi(void)
644 {
645
646 lapic->eoi = 0;
647 }
648
649 void
650 lapic_handle_intr(int vector, struct trapframe *frame)
651 {
652 struct intsrc *isrc;
653
654 if (vector == -1)
655 panic("Couldn't get vector from ISR!");
656 isrc = intr_lookup_source(apic_idt_to_irq(vector));
657 intr_execute_handlers(isrc, frame);
658 }
659
660 void
661 lapic_handle_timer(struct trapframe *frame)
662 {
663 struct lapic *la;
664
665 /* Send EOI first thing. */
666 lapic_eoi();
667
668 #if defined(SMP) && !defined(SCHED_ULE)
669 /*
670 * Don't do any accounting for the disabled HTT cores, since it
671 * will provide misleading numbers for the userland.
672 *
673 * No locking is necessary here, since even if we loose the race
674 * when hlt_cpus_mask changes it is not a big deal, really.
675 *
676 * Don't do that for ULE, since ULE doesn't consider hlt_cpus_mask
677 * and unlike other schedulers it actually schedules threads to
678 * those CPUs.
679 */
680 if ((hlt_cpus_mask & (1 << PCPU_GET(cpuid))) != 0)
681 return;
682 #endif
683
684 /* Look up our local APIC structure for the tick counters. */
685 la = &lapics[PCPU_GET(apic_id)];
686 (*la->la_timer_count)++;
687 critical_enter();
688
689 #ifdef KDTRACE_HOOKS
690 /*
691 * If the DTrace hooks are configured and a callback function
692 * has been registered, then call it to process the high speed
693 * timers.
694 */
695 int cpu = PCPU_GET(cpuid);
696 if (lapic_cyclic_clock_func[cpu] != NULL)
697 (*lapic_cyclic_clock_func[cpu])(frame);
698 #endif
699
700 /* Fire hardclock at hz. */
701 la->la_hard_ticks += hz;
702 if (la->la_hard_ticks >= lapic_timer_hz) {
703 la->la_hard_ticks -= lapic_timer_hz;
704 if (PCPU_GET(cpuid) == 0)
705 hardclock(TRAPF_USERMODE(frame), TRAPF_PC(frame));
706 else
707 hardclock_cpu(TRAPF_USERMODE(frame));
708 }
709
710 /* Fire statclock at stathz. */
711 la->la_stat_ticks += stathz;
712 if (la->la_stat_ticks >= lapic_timer_hz) {
713 la->la_stat_ticks -= lapic_timer_hz;
714 statclock(TRAPF_USERMODE(frame));
715 }
716
717 /* Fire profclock at profhz, but only when needed. */
718 la->la_prof_ticks += profhz;
719 if (la->la_prof_ticks >= lapic_timer_hz) {
720 la->la_prof_ticks -= lapic_timer_hz;
721 if (profprocs != 0)
722 profclock(TRAPF_USERMODE(frame), TRAPF_PC(frame));
723 }
724 critical_exit();
725 }
726
727 static void
728 lapic_timer_set_divisor(u_int divisor)
729 {
730
731 KASSERT(powerof2(divisor), ("lapic: invalid divisor %u", divisor));
732 KASSERT(ffs(divisor) <= sizeof(lapic_timer_divisors) /
733 sizeof(u_int32_t), ("lapic: invalid divisor %u", divisor));
734 lapic->dcr_timer = lapic_timer_divisors[ffs(divisor) - 1];
735 }
736
737 static void
738 lapic_timer_oneshot(u_int count)
739 {
740 u_int32_t value;
741
742 value = lapic->lvt_timer;
743 value &= ~APIC_LVTT_TM;
744 value |= APIC_LVTT_TM_ONE_SHOT;
745 lapic->lvt_timer = value;
746 lapic->icr_timer = count;
747 }
748
749 static void
750 lapic_timer_periodic(u_int count)
751 {
752 u_int32_t value;
753
754 value = lapic->lvt_timer;
755 value &= ~APIC_LVTT_TM;
756 value |= APIC_LVTT_TM_PERIODIC;
757 lapic->lvt_timer = value;
758 lapic->icr_timer = count;
759 }
760
761 static void
762 lapic_timer_enable_intr(void)
763 {
764 u_int32_t value;
765
766 value = lapic->lvt_timer;
767 value &= ~APIC_LVT_M;
768 lapic->lvt_timer = value;
769 }
770
771 /* Request a free IDT vector to be used by the specified IRQ. */
772 u_int
773 apic_alloc_vector(u_int irq)
774 {
775 u_int vector;
776
777 KASSERT(irq < NUM_IO_INTS, ("Invalid IRQ %u", irq));
778
779 /*
780 * Search for a free vector. Currently we just use a very simple
781 * algorithm to find the first free vector.
782 */
783 mtx_lock_spin(&icu_lock);
784 for (vector = 0; vector < APIC_NUM_IOINTS; vector++) {
785 if (ioint_irqs[vector] != 0)
786 continue;
787 ioint_irqs[vector] = irq;
788 mtx_unlock_spin(&icu_lock);
789 return (vector + APIC_IO_INTS);
790 }
791 mtx_unlock_spin(&icu_lock);
792 panic("Couldn't find an APIC vector for IRQ %u", irq);
793 }
794
795 /*
796 * Request 'count' free contiguous IDT vectors to be used by 'count'
797 * IRQs. 'count' must be a power of two and the vectors will be
798 * aligned on a boundary of 'align'. If the request cannot be
799 * satisfied, 0 is returned.
800 */
801 u_int
802 apic_alloc_vectors(u_int *irqs, u_int count, u_int align)
803 {
804 u_int first, run, vector;
805
806 KASSERT(powerof2(count), ("bad count"));
807 KASSERT(powerof2(align), ("bad align"));
808 KASSERT(align >= count, ("align < count"));
809 #ifdef INVARIANTS
810 for (run = 0; run < count; run++)
811 KASSERT(irqs[run] < NUM_IO_INTS, ("Invalid IRQ %u at index %u",
812 irqs[run], run));
813 #endif
814
815 /*
816 * Search for 'count' free vectors. As with apic_alloc_vector(),
817 * this just uses a simple first fit algorithm.
818 */
819 run = 0;
820 first = 0;
821 mtx_lock_spin(&icu_lock);
822 for (vector = 0; vector < APIC_NUM_IOINTS; vector++) {
823
824 /* Vector is in use, end run. */
825 if (ioint_irqs[vector] != 0) {
826 run = 0;
827 first = 0;
828 continue;
829 }
830
831 /* Start a new run if run == 0 and vector is aligned. */
832 if (run == 0) {
833 if ((vector & (align - 1)) != 0)
834 continue;
835 first = vector;
836 }
837 run++;
838
839 /* Keep looping if the run isn't long enough yet. */
840 if (run < count)
841 continue;
842
843 /* Found a run, assign IRQs and return the first vector. */
844 for (vector = 0; vector < count; vector++)
845 ioint_irqs[first + vector] = irqs[vector];
846 mtx_unlock_spin(&icu_lock);
847 return (first + APIC_IO_INTS);
848 }
849 mtx_unlock_spin(&icu_lock);
850 printf("APIC: Couldn't find APIC vectors for %u IRQs\n", count);
851 return (0);
852 }
853
854 void
855 apic_enable_vector(u_int vector)
856 {
857
858 KASSERT(vector != IDT_SYSCALL, ("Attempt to overwrite syscall entry"));
859 KASSERT(ioint_handlers[vector / 32] != NULL,
860 ("No ISR handler for vector %u", vector));
861 setidt(vector, ioint_handlers[vector / 32], SDT_SYS386IGT, SEL_KPL,
862 GSEL(GCODE_SEL, SEL_KPL));
863 }
864
865 void
866 apic_disable_vector(u_int vector)
867 {
868
869 KASSERT(vector != IDT_SYSCALL, ("Attempt to overwrite syscall entry"));
870 KASSERT(ioint_handlers[vector / 32] != NULL,
871 ("No ISR handler for vector %u", vector));
872 setidt(vector, &IDTVEC(rsvd), SDT_SYS386TGT, SEL_KPL,
873 GSEL(GCODE_SEL, SEL_KPL));
874 }
875
876 /* Release an APIC vector when it's no longer in use. */
877 void
878 apic_free_vector(u_int vector, u_int irq)
879 {
880 KASSERT(vector >= APIC_IO_INTS && vector != IDT_SYSCALL &&
881 vector <= APIC_IO_INTS + APIC_NUM_IOINTS,
882 ("Vector %u does not map to an IRQ line", vector));
883 KASSERT(irq < NUM_IO_INTS, ("Invalid IRQ %u", irq));
884 KASSERT(ioint_irqs[vector - APIC_IO_INTS] == irq, ("IRQ mismatch"));
885 mtx_lock_spin(&icu_lock);
886 ioint_irqs[vector - APIC_IO_INTS] = 0;
887 mtx_unlock_spin(&icu_lock);
888 }
889
890 /* Map an IDT vector (APIC) to an IRQ (interrupt source). */
891 u_int
892 apic_idt_to_irq(u_int vector)
893 {
894
895 KASSERT(vector >= APIC_IO_INTS && vector != IDT_SYSCALL &&
896 vector <= APIC_IO_INTS + APIC_NUM_IOINTS,
897 ("Vector %u does not map to an IRQ line", vector));
898 return (ioint_irqs[vector - APIC_IO_INTS]);
899 }
900
901 #ifdef DDB
902 /*
903 * Dump data about APIC IDT vector mappings.
904 */
905 DB_SHOW_COMMAND(apic, db_show_apic)
906 {
907 struct intsrc *isrc;
908 int i, verbose;
909 u_int irq;
910
911 if (strcmp(modif, "vv") == 0)
912 verbose = 2;
913 else if (strcmp(modif, "v") == 0)
914 verbose = 1;
915 else
916 verbose = 0;
917 for (i = 0; i < APIC_NUM_IOINTS + 1 && !db_pager_quit; i++) {
918 irq = ioint_irqs[i];
919 if (irq != 0 && irq != IRQ_SYSCALL) {
920 db_printf("vec 0x%2x -> ", i + APIC_IO_INTS);
921 if (irq == IRQ_TIMER)
922 db_printf("lapic timer\n");
923 else if (irq < NUM_IO_INTS) {
924 isrc = intr_lookup_source(irq);
925 if (isrc == NULL || verbose == 0)
926 db_printf("IRQ %u\n", irq);
927 else
928 db_dump_intr_event(isrc->is_event,
929 verbose == 2);
930 } else
931 db_printf("IRQ %u ???\n", irq);
932 }
933 }
934 }
935
936 static void
937 dump_mask(const char *prefix, uint32_t v, int base)
938 {
939 int i, first;
940
941 first = 1;
942 for (i = 0; i < 32; i++)
943 if (v & (1 << i)) {
944 if (first) {
945 db_printf("%s:", prefix);
946 first = 0;
947 }
948 db_printf(" %02x", base + i);
949 }
950 if (!first)
951 db_printf("\n");
952 }
953
954 /* Show info from the lapic regs for this CPU. */
955 DB_SHOW_COMMAND(lapic, db_show_lapic)
956 {
957 uint32_t v;
958
959 db_printf("lapic ID = %d\n", lapic_id());
960 v = lapic->version;
961 db_printf("version = %d.%d\n", (v & APIC_VER_VERSION) >> 4,
962 v & 0xf);
963 db_printf("max LVT = %d\n", (v & APIC_VER_MAXLVT) >> MAXLVTSHIFT);
964 v = lapic->svr;
965 db_printf("SVR = %02x (%s)\n", v & APIC_SVR_VECTOR,
966 v & APIC_SVR_ENABLE ? "enabled" : "disabled");
967 db_printf("TPR = %02x\n", lapic->tpr);
968
969 #define dump_field(prefix, index) \
970 dump_mask(__XSTRING(prefix ## index), lapic->prefix ## index, \
971 index * 32)
972
973 db_printf("In-service Interrupts:\n");
974 dump_field(isr, 0);
975 dump_field(isr, 1);
976 dump_field(isr, 2);
977 dump_field(isr, 3);
978 dump_field(isr, 4);
979 dump_field(isr, 5);
980 dump_field(isr, 6);
981 dump_field(isr, 7);
982
983 db_printf("TMR Interrupts:\n");
984 dump_field(tmr, 0);
985 dump_field(tmr, 1);
986 dump_field(tmr, 2);
987 dump_field(tmr, 3);
988 dump_field(tmr, 4);
989 dump_field(tmr, 5);
990 dump_field(tmr, 6);
991 dump_field(tmr, 7);
992
993 db_printf("IRR Interrupts:\n");
994 dump_field(irr, 0);
995 dump_field(irr, 1);
996 dump_field(irr, 2);
997 dump_field(irr, 3);
998 dump_field(irr, 4);
999 dump_field(irr, 5);
1000 dump_field(irr, 6);
1001 dump_field(irr, 7);
1002
1003 #undef dump_field
1004 }
1005 #endif
1006
1007 /*
1008 * APIC probing support code. This includes code to manage enumerators.
1009 */
1010
1011 static SLIST_HEAD(, apic_enumerator) enumerators =
1012 SLIST_HEAD_INITIALIZER(enumerators);
1013 static struct apic_enumerator *best_enum;
1014
1015 void
1016 apic_register_enumerator(struct apic_enumerator *enumerator)
1017 {
1018 #ifdef INVARIANTS
1019 struct apic_enumerator *apic_enum;
1020
1021 SLIST_FOREACH(apic_enum, &enumerators, apic_next) {
1022 if (apic_enum == enumerator)
1023 panic("%s: Duplicate register of %s", __func__,
1024 enumerator->apic_name);
1025 }
1026 #endif
1027 SLIST_INSERT_HEAD(&enumerators, enumerator, apic_next);
1028 }
1029
1030 /*
1031 * Probe the APIC enumerators, enumerate CPUs, and initialize the
1032 * local APIC.
1033 */
1034 static void
1035 apic_init(void *dummy __unused)
1036 {
1037 struct apic_enumerator *enumerator;
1038 uint64_t apic_base;
1039 int retval, best;
1040
1041 /* We only support built in local APICs. */
1042 if (!(cpu_feature & CPUID_APIC))
1043 return;
1044
1045 /* Don't probe if APIC mode is disabled. */
1046 if (resource_disabled("apic", 0))
1047 return;
1048
1049 /* First, probe all the enumerators to find the best match. */
1050 best_enum = NULL;
1051 best = 0;
1052 SLIST_FOREACH(enumerator, &enumerators, apic_next) {
1053 retval = enumerator->apic_probe();
1054 if (retval > 0)
1055 continue;
1056 if (best_enum == NULL || best < retval) {
1057 best_enum = enumerator;
1058 best = retval;
1059 }
1060 }
1061 if (best_enum == NULL) {
1062 if (bootverbose)
1063 printf("APIC: Could not find any APICs.\n");
1064 return;
1065 }
1066
1067 if (bootverbose)
1068 printf("APIC: Using the %s enumerator.\n",
1069 best_enum->apic_name);
1070
1071 /*
1072 * To work around an errata, we disable the local APIC on some
1073 * CPUs during early startup. We need to turn the local APIC back
1074 * on on such CPUs now.
1075 */
1076 if (cpu == CPU_686 && strcmp(cpu_vendor, "GenuineIntel") == 0 &&
1077 (cpu_id & 0xff0) == 0x610) {
1078 apic_base = rdmsr(MSR_APICBASE);
1079 apic_base |= APICBASE_ENABLED;
1080 wrmsr(MSR_APICBASE, apic_base);
1081 }
1082
1083 /* Second, probe the CPU's in the system. */
1084 retval = best_enum->apic_probe_cpus();
1085 if (retval != 0)
1086 printf("%s: Failed to probe CPUs: returned %d\n",
1087 best_enum->apic_name, retval);
1088
1089 /* Third, initialize the local APIC. */
1090 retval = best_enum->apic_setup_local();
1091 if (retval != 0)
1092 printf("%s: Failed to setup the local APIC: returned %d\n",
1093 best_enum->apic_name, retval);
1094 }
1095 SYSINIT(apic_init, SI_SUB_CPU, SI_ORDER_SECOND, apic_init, NULL);
1096
1097 /*
1098 * Setup the I/O APICs.
1099 */
1100 static void
1101 apic_setup_io(void *dummy __unused)
1102 {
1103 int retval;
1104
1105 if (best_enum == NULL)
1106 return;
1107 retval = best_enum->apic_setup_io();
1108 if (retval != 0)
1109 printf("%s: Failed to setup I/O APICs: returned %d\n",
1110 best_enum->apic_name, retval);
1111
1112 /*
1113 * Finish setting up the local APIC on the BSP once we know how to
1114 * properly program the LINT pins.
1115 */
1116 lapic_setup(1);
1117 intr_register_pic(&lapic_pic);
1118 if (bootverbose)
1119 lapic_dump("BSP");
1120
1121 /* Enable the MSI "pic". */
1122 msi_init();
1123 }
1124 SYSINIT(apic_setup_io, SI_SUB_INTR, SI_ORDER_SECOND, apic_setup_io, NULL);
1125
1126 #ifdef SMP
1127 /*
1128 * Inter Processor Interrupt functions. The lapic_ipi_*() functions are
1129 * private to the sys/i386 code. The public interface for the rest of the
1130 * kernel is defined in mp_machdep.c.
1131 */
1132 int
1133 lapic_ipi_wait(int delay)
1134 {
1135 int x, incr;
1136
1137 /*
1138 * Wait delay loops for IPI to be sent. This is highly bogus
1139 * since this is sensitive to CPU clock speed. If delay is
1140 * -1, we wait forever.
1141 */
1142 if (delay == -1) {
1143 incr = 0;
1144 delay = 1;
1145 } else
1146 incr = 1;
1147 for (x = 0; x < delay; x += incr) {
1148 if ((lapic->icr_lo & APIC_DELSTAT_MASK) == APIC_DELSTAT_IDLE)
1149 return (1);
1150 ia32_pause();
1151 }
1152 return (0);
1153 }
1154
1155 void
1156 lapic_ipi_raw(register_t icrlo, u_int dest)
1157 {
1158 register_t value, eflags;
1159
1160 /* XXX: Need more sanity checking of icrlo? */
1161 KASSERT(lapic != NULL, ("%s called too early", __func__));
1162 KASSERT((dest & ~(APIC_ID_MASK >> APIC_ID_SHIFT)) == 0,
1163 ("%s: invalid dest field", __func__));
1164 KASSERT((icrlo & APIC_ICRLO_RESV_MASK) == 0,
1165 ("%s: reserved bits set in ICR LO register", __func__));
1166
1167 /* Set destination in ICR HI register if it is being used. */
1168 eflags = intr_disable();
1169 if ((icrlo & APIC_DEST_MASK) == APIC_DEST_DESTFLD) {
1170 value = lapic->icr_hi;
1171 value &= ~APIC_ID_MASK;
1172 value |= dest << APIC_ID_SHIFT;
1173 lapic->icr_hi = value;
1174 }
1175
1176 /* Program the contents of the IPI and dispatch it. */
1177 value = lapic->icr_lo;
1178 value &= APIC_ICRLO_RESV_MASK;
1179 value |= icrlo;
1180 lapic->icr_lo = value;
1181 intr_restore(eflags);
1182 }
1183
1184 #define BEFORE_SPIN 1000000
1185 #ifdef DETECT_DEADLOCK
1186 #define AFTER_SPIN 1000
1187 #endif
1188
1189 void
1190 lapic_ipi_vectored(u_int vector, int dest)
1191 {
1192 register_t icrlo, destfield;
1193
1194 KASSERT((vector & ~APIC_VECTOR_MASK) == 0,
1195 ("%s: invalid vector %d", __func__, vector));
1196
1197 icrlo = vector | APIC_DELMODE_FIXED | APIC_DESTMODE_PHY |
1198 APIC_LEVEL_DEASSERT | APIC_TRIGMOD_EDGE;
1199 destfield = 0;
1200 switch (dest) {
1201 case APIC_IPI_DEST_SELF:
1202 icrlo |= APIC_DEST_SELF;
1203 break;
1204 case APIC_IPI_DEST_ALL:
1205 icrlo |= APIC_DEST_ALLISELF;
1206 break;
1207 case APIC_IPI_DEST_OTHERS:
1208 icrlo |= APIC_DEST_ALLESELF;
1209 break;
1210 default:
1211 KASSERT((dest & ~(APIC_ID_MASK >> APIC_ID_SHIFT)) == 0,
1212 ("%s: invalid destination 0x%x", __func__, dest));
1213 destfield = dest;
1214 }
1215
1216 /* Wait for an earlier IPI to finish. */
1217 if (!lapic_ipi_wait(BEFORE_SPIN)) {
1218 if (panicstr != NULL)
1219 return;
1220 else
1221 panic("APIC: Previous IPI is stuck");
1222 }
1223
1224 lapic_ipi_raw(icrlo, destfield);
1225
1226 #ifdef DETECT_DEADLOCK
1227 /* Wait for IPI to be delivered. */
1228 if (!lapic_ipi_wait(AFTER_SPIN)) {
1229 #ifdef needsattention
1230 /*
1231 * XXX FIXME:
1232 *
1233 * The above function waits for the message to actually be
1234 * delivered. It breaks out after an arbitrary timeout
1235 * since the message should eventually be delivered (at
1236 * least in theory) and that if it wasn't we would catch
1237 * the failure with the check above when the next IPI is
1238 * sent.
1239 *
1240 * We could skip this wait entirely, EXCEPT it probably
1241 * protects us from other routines that assume that the
1242 * message was delivered and acted upon when this function
1243 * returns.
1244 */
1245 printf("APIC: IPI might be stuck\n");
1246 #else /* !needsattention */
1247 /* Wait until mesage is sent without a timeout. */
1248 while (lapic->icr_lo & APIC_DELSTAT_PEND)
1249 ia32_pause();
1250 #endif /* needsattention */
1251 }
1252 #endif /* DETECT_DEADLOCK */
1253 }
1254 #endif /* SMP */
Cache object: f822eb12aae8aaa74711eedf58156efe
|