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: releng/6.4/sys/amd64/amd64/local_apic.c 184456 2008-10-29 21:31:01Z sobomax $");
36
37 #include "opt_hwpmc_hooks.h"
38
39 #include "opt_ddb.h"
40
41 #include <sys/param.h>
42 #include <sys/systm.h>
43 #include <sys/bus.h>
44 #include <sys/kernel.h>
45 #include <sys/lock.h>
46 #include <sys/mutex.h>
47 #include <sys/pcpu.h>
48 #include <sys/smp.h>
49 #include <sys/proc.h>
50
51 #include <vm/vm.h>
52 #include <vm/pmap.h>
53
54 #include <machine/apicreg.h>
55 #include <machine/cputypes.h>
56 #include <machine/frame.h>
57 #include <machine/intr_machdep.h>
58 #include <machine/apicvar.h>
59 #include <machine/md_var.h>
60 #include <machine/smp.h>
61 #include <machine/specialreg.h>
62
63 #ifdef DDB
64 #include <sys/interrupt.h>
65 #include <ddb/ddb.h>
66 #endif
67
68 /* Sanity checks on IDT vectors. */
69 CTASSERT(APIC_IO_INTS + APIC_NUM_IOINTS == APIC_TIMER_INT);
70 CTASSERT(APIC_TIMER_INT < APIC_LOCAL_INTS);
71 CTASSERT(APIC_LOCAL_INTS == 240);
72 CTASSERT(IPI_STOP < APIC_SPURIOUS_INT);
73
74 /* Magic IRQ values for the timer and syscalls. */
75 #define IRQ_TIMER (NUM_IO_INTS + 1)
76 #define IRQ_SYSCALL (NUM_IO_INTS + 2)
77
78 /*
79 * Support for local APICs. Local APICs manage interrupts on each
80 * individual processor as opposed to I/O APICs which receive interrupts
81 * from I/O devices and then forward them on to the local APICs.
82 *
83 * Local APICs can also send interrupts to each other thus providing the
84 * mechanism for IPIs.
85 */
86
87 struct lvt {
88 u_int lvt_edgetrigger:1;
89 u_int lvt_activehi:1;
90 u_int lvt_masked:1;
91 u_int lvt_active:1;
92 u_int lvt_mode:16;
93 u_int lvt_vector:8;
94 };
95
96 struct lapic {
97 struct lvt la_lvts[LVT_MAX + 1];
98 u_int la_id:8;
99 u_int la_cluster:4;
100 u_int la_cluster_id:2;
101 u_int la_present:1;
102 u_long *la_timer_count;
103 u_long la_hard_ticks;
104 u_long la_stat_ticks;
105 u_long la_prof_ticks;
106 } static lapics[MAX_APIC_ID + 1];
107
108 /* XXX: should thermal be an NMI? */
109
110 /* Global defaults for local APIC LVT entries. */
111 static struct lvt lvts[LVT_MAX + 1] = {
112 { 1, 1, 1, 1, APIC_LVT_DM_EXTINT, 0 }, /* LINT0: masked ExtINT */
113 { 1, 1, 0, 1, APIC_LVT_DM_NMI, 0 }, /* LINT1: NMI */
114 { 1, 1, 1, 1, APIC_LVT_DM_FIXED, APIC_TIMER_INT }, /* Timer */
115 { 1, 1, 1, 1, APIC_LVT_DM_FIXED, APIC_ERROR_INT }, /* Error */
116 { 1, 1, 0, 1, APIC_LVT_DM_NMI, 0 }, /* PMC */
117 { 1, 1, 1, 1, APIC_LVT_DM_FIXED, APIC_THERMAL_INT }, /* Thermal */
118 };
119
120 static inthand_t *ioint_handlers[] = {
121 NULL, /* 0 - 31 */
122 IDTVEC(apic_isr1), /* 32 - 63 */
123 IDTVEC(apic_isr2), /* 64 - 95 */
124 IDTVEC(apic_isr3), /* 96 - 127 */
125 IDTVEC(apic_isr4), /* 128 - 159 */
126 IDTVEC(apic_isr5), /* 160 - 191 */
127 IDTVEC(apic_isr6), /* 192 - 223 */
128 IDTVEC(apic_isr7), /* 224 - 255 */
129 };
130
131 /* Include IDT_SYSCALL to make indexing easier. */
132 static u_int ioint_irqs[APIC_NUM_IOINTS + 1];
133
134 static u_int32_t lapic_timer_divisors[] = {
135 APIC_TDCR_1, APIC_TDCR_2, APIC_TDCR_4, APIC_TDCR_8, APIC_TDCR_16,
136 APIC_TDCR_32, APIC_TDCR_64, APIC_TDCR_128
137 };
138
139 volatile lapic_t *lapic;
140 static u_long lapic_timer_divisor, lapic_timer_period, lapic_timer_hz;
141
142 static void lapic_enable(void);
143 static void lapic_resume(struct pic *pic);
144 static void lapic_timer_enable_intr(void);
145 static void lapic_timer_oneshot(u_int count);
146 static void lapic_timer_periodic(u_int count);
147 static void lapic_timer_set_divisor(u_int divisor);
148 static uint32_t lvt_mode(struct lapic *la, u_int pin, uint32_t value);
149
150 struct pic lapic_pic = { .pic_resume = lapic_resume };
151
152 static uint32_t
153 lvt_mode(struct lapic *la, u_int pin, uint32_t value)
154 {
155 struct lvt *lvt;
156
157 KASSERT(pin <= LVT_MAX, ("%s: pin %u out of range", __func__, pin));
158 if (la->la_lvts[pin].lvt_active)
159 lvt = &la->la_lvts[pin];
160 else
161 lvt = &lvts[pin];
162
163 value &= ~(APIC_LVT_M | APIC_LVT_TM | APIC_LVT_IIPP | APIC_LVT_DM |
164 APIC_LVT_VECTOR);
165 if (lvt->lvt_edgetrigger == 0)
166 value |= APIC_LVT_TM;
167 if (lvt->lvt_activehi == 0)
168 value |= APIC_LVT_IIPP_INTALO;
169 if (lvt->lvt_masked)
170 value |= APIC_LVT_M;
171 value |= lvt->lvt_mode;
172 switch (lvt->lvt_mode) {
173 case APIC_LVT_DM_NMI:
174 case APIC_LVT_DM_SMI:
175 case APIC_LVT_DM_INIT:
176 case APIC_LVT_DM_EXTINT:
177 if (!lvt->lvt_edgetrigger) {
178 printf("lapic%u: Forcing LINT%u to edge trigger\n",
179 la->la_id, pin);
180 value |= APIC_LVT_TM;
181 }
182 /* Use a vector of 0. */
183 break;
184 case APIC_LVT_DM_FIXED:
185 value |= lvt->lvt_vector;
186 break;
187 default:
188 panic("bad APIC LVT delivery mode: %#x\n", value);
189 }
190 return (value);
191 }
192
193 /*
194 * Map the local APIC and setup necessary interrupt vectors.
195 */
196 void
197 lapic_init(uintptr_t addr)
198 {
199
200 /* Map the local APIC and setup the spurious interrupt handler. */
201 KASSERT(trunc_page(addr) == addr,
202 ("local APIC not aligned on a page boundary"));
203 lapic = (lapic_t *)pmap_mapdev(addr, sizeof(lapic_t));
204 setidt(APIC_SPURIOUS_INT, IDTVEC(spuriousint), SDT_SYSIGT, SEL_KPL, 0);
205
206 /* Perform basic initialization of the BSP's local APIC. */
207 lapic_enable();
208 ioint_irqs[IDT_SYSCALL - APIC_IO_INTS] = IRQ_SYSCALL;
209
210 /* Set BSP's per-CPU local APIC ID. */
211 PCPU_SET(apic_id, lapic_id());
212 intr_add_cpu(PCPU_GET(apic_id));
213
214 /* Local APIC timer interrupt. */
215 setidt(APIC_TIMER_INT, IDTVEC(timerint), SDT_SYSIGT, SEL_KPL, 0);
216 ioint_irqs[APIC_TIMER_INT - APIC_IO_INTS] = IRQ_TIMER;
217
218 /* XXX: error/thermal interrupts */
219 }
220
221 /*
222 * Create a local APIC instance.
223 */
224 void
225 lapic_create(u_int apic_id, int boot_cpu)
226 {
227 int i;
228
229 if (apic_id > MAX_APIC_ID) {
230 printf("APIC: Ignoring local APIC with ID %d\n", apic_id);
231 if (boot_cpu)
232 panic("Can't ignore BSP");
233 return;
234 }
235 KASSERT(!lapics[apic_id].la_present, ("duplicate local APIC %u",
236 apic_id));
237
238 /*
239 * Assume no local LVT overrides and a cluster of 0 and
240 * intra-cluster ID of 0.
241 */
242 lapics[apic_id].la_present = 1;
243 lapics[apic_id].la_id = apic_id;
244 for (i = 0; i < LVT_MAX; i++) {
245 lapics[apic_id].la_lvts[i] = lvts[i];
246 lapics[apic_id].la_lvts[i].lvt_active = 0;
247 }
248
249 #ifdef SMP
250 cpu_add(apic_id, boot_cpu);
251 #endif
252 }
253
254 /*
255 * Dump contents of local APIC registers
256 */
257 void
258 lapic_dump(const char* str)
259 {
260
261 printf("cpu%d %s:\n", PCPU_GET(cpuid), str);
262 printf(" ID: 0x%08x VER: 0x%08x LDR: 0x%08x DFR: 0x%08x\n",
263 lapic->id, lapic->version, lapic->ldr, lapic->dfr);
264 printf(" lint0: 0x%08x lint1: 0x%08x TPR: 0x%08x SVR: 0x%08x\n",
265 lapic->lvt_lint0, lapic->lvt_lint1, lapic->tpr, lapic->svr);
266 printf(" timer: 0x%08x therm: 0x%08x err: 0x%08x pcm: 0x%08x\n",
267 lapic->lvt_timer, lapic->lvt_thermal, lapic->lvt_error,
268 lapic->lvt_pcint);
269 }
270
271 void
272 lapic_setup(int boot)
273 {
274 struct lapic *la;
275 u_int32_t maxlvt;
276 register_t eflags;
277 char buf[MAXCOMLEN + 1];
278
279 la = &lapics[lapic_id()];
280 KASSERT(la->la_present, ("missing APIC structure"));
281 eflags = intr_disable();
282 maxlvt = (lapic->version & APIC_VER_MAXLVT) >> MAXLVTSHIFT;
283
284 /* Initialize the TPR to allow all interrupts. */
285 lapic_set_tpr(0);
286
287 /* Setup spurious vector and enable the local APIC. */
288 lapic_enable();
289
290 /* Program LINT[01] LVT entries. */
291 lapic->lvt_lint0 = lvt_mode(la, LVT_LINT0, lapic->lvt_lint0);
292 lapic->lvt_lint1 = lvt_mode(la, LVT_LINT1, lapic->lvt_lint1);
293 #ifdef HWPMC_HOOKS
294 /* Program the PMC LVT entry if present. */
295 if (maxlvt >= LVT_PMC)
296 lapic->lvt_pcint = lvt_mode(la, LVT_PMC, lapic->lvt_pcint);
297 #endif
298
299 /* Program timer LVT and setup handler. */
300 lapic->lvt_timer = lvt_mode(la, LVT_TIMER, lapic->lvt_timer);
301 if (boot) {
302 snprintf(buf, sizeof(buf), "cpu%d: timer", PCPU_GET(cpuid));
303 intrcnt_add(buf, &la->la_timer_count);
304 }
305
306 /* We don't setup the timer during boot on the BSP until later. */
307 if (!(boot && PCPU_GET(cpuid) == 0)) {
308 KASSERT(lapic_timer_period != 0, ("lapic%u: zero divisor",
309 lapic_id()));
310 lapic_timer_set_divisor(lapic_timer_divisor);
311 lapic_timer_periodic(lapic_timer_period);
312 lapic_timer_enable_intr();
313 }
314
315 /* XXX: Error and thermal LVTs */
316
317 if (strcmp(cpu_vendor, "AuthenticAMD") == 0) {
318 /*
319 * Detect the presence of C1E capability mostly on latest
320 * dual-cores (or future) k8 family. This feature renders
321 * the local APIC timer dead, so we disable it by reading
322 * the Interrupt Pending Message register and clearing both
323 * C1eOnCmpHalt (bit 28) and SmiOnCmpHalt (bit 27).
324 *
325 * Reference:
326 * "BIOS and Kernel Developer's Guide for AMD NPT
327 * Family 0Fh Processors"
328 * #32559 revision 3.00
329 */
330 if ((cpu_id & 0x00000f00) == 0x00000f00 &&
331 (cpu_id & 0x0fff0000) >= 0x00040000) {
332 uint64_t msr;
333
334 msr = rdmsr(0xc0010055);
335 if (msr & 0x18000000)
336 wrmsr(0xc0010055, msr & ~0x18000000ULL);
337 }
338 }
339
340 intr_restore(eflags);
341 }
342
343 /*
344 * Called by cpu_initclocks() on the BSP to setup the local APIC timer so
345 * that it can drive hardclock, statclock, and profclock. This function
346 * returns true if it is able to use the local APIC timer to drive the
347 * clocks and false if it is not able.
348 */
349 int
350 lapic_setup_clock(void)
351 {
352 u_long value;
353
354 /* Can't drive the timer without a local APIC. */
355 if (lapic == NULL)
356 return (0);
357
358 /* Start off with a divisor of 2 (power on reset default). */
359 lapic_timer_divisor = 2;
360
361 /* Try to calibrate the local APIC timer. */
362 do {
363 lapic_timer_set_divisor(lapic_timer_divisor);
364 lapic_timer_oneshot(APIC_TIMER_MAX_COUNT);
365 DELAY(2000000);
366 value = APIC_TIMER_MAX_COUNT - lapic->ccr_timer;
367 if (value != APIC_TIMER_MAX_COUNT)
368 break;
369 lapic_timer_divisor <<= 1;
370 } while (lapic_timer_divisor <= 128);
371 if (lapic_timer_divisor > 128)
372 panic("lapic: Divisor too big");
373 value /= 2;
374 if (bootverbose)
375 printf("lapic: Divisor %lu, Frequency %lu hz\n",
376 lapic_timer_divisor, value);
377
378 /*
379 * We want to run stathz in the neighborhood of 128hz. We would
380 * like profhz to run as often as possible, so we let it run on
381 * each clock tick. We try to honor the requested 'hz' value as
382 * much as possible.
383 *
384 * If 'hz' is above 1500, then we just let the lapic timer
385 * (and profhz) run at hz. If 'hz' is below 1500 but above
386 * 750, then we let the lapic timer run at 2 * 'hz'. If 'hz'
387 * is below 750 then we let the lapic timer run at 4 * 'hz'.
388 */
389 if (hz >= 1500)
390 lapic_timer_hz = hz;
391 else if (hz >= 750)
392 lapic_timer_hz = hz * 2;
393 else
394 lapic_timer_hz = hz * 4;
395 if (lapic_timer_hz < 128)
396 stathz = lapic_timer_hz;
397 else
398 stathz = lapic_timer_hz / (lapic_timer_hz / 128);
399 profhz = lapic_timer_hz;
400 lapic_timer_period = value / lapic_timer_hz;
401
402 /*
403 * Start up the timer on the BSP. The APs will kick off their
404 * timer during lapic_setup().
405 */
406 lapic_timer_periodic(lapic_timer_period);
407 lapic_timer_enable_intr();
408 return (1);
409 }
410
411 void
412 lapic_disable(void)
413 {
414 uint32_t value;
415
416 /* Software disable the local APIC. */
417 value = lapic->svr;
418 value &= ~APIC_SVR_SWEN;
419 lapic->svr = value;
420 }
421
422 static void
423 lapic_enable(void)
424 {
425 u_int32_t value;
426
427 /* Program the spurious vector to enable the local APIC. */
428 value = lapic->svr;
429 value &= ~(APIC_SVR_VECTOR | APIC_SVR_FOCUS);
430 value |= (APIC_SVR_FEN | APIC_SVR_SWEN | APIC_SPURIOUS_INT);
431 lapic->svr = value;
432 }
433
434 /* Reset the local APIC on the BSP during resume. */
435 static void
436 lapic_resume(struct pic *pic)
437 {
438
439 lapic_setup(0);
440 }
441
442 int
443 lapic_id(void)
444 {
445
446 KASSERT(lapic != NULL, ("local APIC is not mapped"));
447 return (lapic->id >> APIC_ID_SHIFT);
448 }
449
450 int
451 lapic_intr_pending(u_int vector)
452 {
453 volatile u_int32_t *irr;
454
455 /*
456 * The IRR registers are an array of 128-bit registers each of
457 * which only describes 32 interrupts in the low 32 bits.. Thus,
458 * we divide the vector by 32 to get the 128-bit index. We then
459 * multiply that index by 4 to get the equivalent index from
460 * treating the IRR as an array of 32-bit registers. Finally, we
461 * modulus the vector by 32 to determine the individual bit to
462 * test.
463 */
464 irr = &lapic->irr0;
465 return (irr[(vector / 32) * 4] & 1 << (vector % 32));
466 }
467
468 void
469 lapic_set_logical_id(u_int apic_id, u_int cluster, u_int cluster_id)
470 {
471 struct lapic *la;
472
473 KASSERT(lapics[apic_id].la_present, ("%s: APIC %u doesn't exist",
474 __func__, apic_id));
475 KASSERT(cluster <= APIC_MAX_CLUSTER, ("%s: cluster %u too big",
476 __func__, cluster));
477 KASSERT(cluster_id <= APIC_MAX_INTRACLUSTER_ID,
478 ("%s: intra cluster id %u too big", __func__, cluster_id));
479 la = &lapics[apic_id];
480 la->la_cluster = cluster;
481 la->la_cluster_id = cluster_id;
482 }
483
484 int
485 lapic_set_lvt_mask(u_int apic_id, u_int pin, u_char masked)
486 {
487
488 if (pin > LVT_MAX)
489 return (EINVAL);
490 if (apic_id == APIC_ID_ALL) {
491 lvts[pin].lvt_masked = masked;
492 if (bootverbose)
493 printf("lapic:");
494 } else {
495 KASSERT(lapics[apic_id].la_present,
496 ("%s: missing APIC %u", __func__, apic_id));
497 lapics[apic_id].la_lvts[pin].lvt_masked = masked;
498 lapics[apic_id].la_lvts[pin].lvt_active = 1;
499 if (bootverbose)
500 printf("lapic%u:", apic_id);
501 }
502 if (bootverbose)
503 printf(" LINT%u %s\n", pin, masked ? "masked" : "unmasked");
504 return (0);
505 }
506
507 int
508 lapic_set_lvt_mode(u_int apic_id, u_int pin, u_int32_t mode)
509 {
510 struct lvt *lvt;
511
512 if (pin > LVT_MAX)
513 return (EINVAL);
514 if (apic_id == APIC_ID_ALL) {
515 lvt = &lvts[pin];
516 if (bootverbose)
517 printf("lapic:");
518 } else {
519 KASSERT(lapics[apic_id].la_present,
520 ("%s: missing APIC %u", __func__, apic_id));
521 lvt = &lapics[apic_id].la_lvts[pin];
522 lvt->lvt_active = 1;
523 if (bootverbose)
524 printf("lapic%u:", apic_id);
525 }
526 lvt->lvt_mode = mode;
527 switch (mode) {
528 case APIC_LVT_DM_NMI:
529 case APIC_LVT_DM_SMI:
530 case APIC_LVT_DM_INIT:
531 case APIC_LVT_DM_EXTINT:
532 lvt->lvt_edgetrigger = 1;
533 lvt->lvt_activehi = 1;
534 if (mode == APIC_LVT_DM_EXTINT)
535 lvt->lvt_masked = 1;
536 else
537 lvt->lvt_masked = 0;
538 break;
539 default:
540 panic("Unsupported delivery mode: 0x%x\n", mode);
541 }
542 if (bootverbose) {
543 printf(" Routing ");
544 switch (mode) {
545 case APIC_LVT_DM_NMI:
546 printf("NMI");
547 break;
548 case APIC_LVT_DM_SMI:
549 printf("SMI");
550 break;
551 case APIC_LVT_DM_INIT:
552 printf("INIT");
553 break;
554 case APIC_LVT_DM_EXTINT:
555 printf("ExtINT");
556 break;
557 }
558 printf(" -> LINT%u\n", pin);
559 }
560 return (0);
561 }
562
563 int
564 lapic_set_lvt_polarity(u_int apic_id, u_int pin, enum intr_polarity pol)
565 {
566
567 if (pin > LVT_MAX || pol == INTR_POLARITY_CONFORM)
568 return (EINVAL);
569 if (apic_id == APIC_ID_ALL) {
570 lvts[pin].lvt_activehi = (pol == INTR_POLARITY_HIGH);
571 if (bootverbose)
572 printf("lapic:");
573 } else {
574 KASSERT(lapics[apic_id].la_present,
575 ("%s: missing APIC %u", __func__, apic_id));
576 lapics[apic_id].la_lvts[pin].lvt_active = 1;
577 lapics[apic_id].la_lvts[pin].lvt_activehi =
578 (pol == INTR_POLARITY_HIGH);
579 if (bootverbose)
580 printf("lapic%u:", apic_id);
581 }
582 if (bootverbose)
583 printf(" LINT%u polarity: %s\n", pin,
584 pol == INTR_POLARITY_HIGH ? "high" : "low");
585 return (0);
586 }
587
588 int
589 lapic_set_lvt_triggermode(u_int apic_id, u_int pin, enum intr_trigger trigger)
590 {
591
592 if (pin > LVT_MAX || trigger == INTR_TRIGGER_CONFORM)
593 return (EINVAL);
594 if (apic_id == APIC_ID_ALL) {
595 lvts[pin].lvt_edgetrigger = (trigger == INTR_TRIGGER_EDGE);
596 if (bootverbose)
597 printf("lapic:");
598 } else {
599 KASSERT(lapics[apic_id].la_present,
600 ("%s: missing APIC %u", __func__, apic_id));
601 lapics[apic_id].la_lvts[pin].lvt_edgetrigger =
602 (trigger == INTR_TRIGGER_EDGE);
603 lapics[apic_id].la_lvts[pin].lvt_active = 1;
604 if (bootverbose)
605 printf("lapic%u:", apic_id);
606 }
607 if (bootverbose)
608 printf(" LINT%u trigger: %s\n", pin,
609 trigger == INTR_TRIGGER_EDGE ? "edge" : "level");
610 return (0);
611 }
612
613 /*
614 * Adjust the TPR of the current CPU so that it blocks all interrupts below
615 * the passed in vector.
616 */
617 void
618 lapic_set_tpr(u_int vector)
619 {
620 #ifdef CHEAP_TPR
621 lapic->tpr = vector;
622 #else
623 u_int32_t tpr;
624
625 tpr = lapic->tpr & ~APIC_TPR_PRIO;
626 tpr |= vector;
627 lapic->tpr = tpr;
628 #endif
629 }
630
631 void
632 lapic_eoi(void)
633 {
634
635 lapic->eoi = 0;
636 }
637
638 void
639 lapic_handle_intr(void *cookie, struct intrframe frame)
640 {
641 struct intsrc *isrc;
642 int vec = (uintptr_t)cookie;
643
644 if (vec == -1)
645 panic("Couldn't get vector from ISR!");
646 isrc = intr_lookup_source(apic_idt_to_irq(vec));
647 intr_execute_handlers(isrc, &frame);
648 }
649
650 void
651 lapic_handle_timer(struct clockframe frame)
652 {
653 struct lapic *la;
654
655 #if defined(SMP) && !defined(SCHED_ULE)
656 /*
657 * Don't do any accounting for the disabled HTT cores, since it
658 * will provide misleading numbers for the userland.
659 *
660 * No locking is necessary here, since even if we loose the race
661 * when hlt_cpus_mask changes it is not a big deal, really.
662 *
663 * Don't do that for ULE, since ULE doesn't consider hlt_cpus_mask
664 * and unlike other schedulers it actually schedules threads to
665 * those CPUs.
666 */
667 if ((hlt_cpus_mask & (1 << PCPU_GET(cpuid))) != 0)
668 return;
669 #endif
670
671 la = &lapics[PCPU_GET(apic_id)];
672 (*la->la_timer_count)++;
673 critical_enter();
674
675 /* Fire hardclock at hz. */
676 la->la_hard_ticks += hz;
677 if (la->la_hard_ticks >= lapic_timer_hz) {
678 la->la_hard_ticks -= lapic_timer_hz;
679 if (PCPU_GET(cpuid) == 0)
680 hardclock(&frame);
681 else
682 hardclock_process(&frame);
683 }
684
685 /* Fire statclock at stathz. */
686 la->la_stat_ticks += stathz;
687 if (la->la_stat_ticks >= lapic_timer_hz) {
688 la->la_stat_ticks -= lapic_timer_hz;
689 statclock(&frame);
690 }
691
692 /* Fire profclock at profhz, but only when needed. */
693 la->la_prof_ticks += profhz;
694 if (la->la_prof_ticks >= lapic_timer_hz) {
695 la->la_prof_ticks -= lapic_timer_hz;
696 if (profprocs != 0)
697 profclock(&frame);
698 }
699 critical_exit();
700 }
701
702 static void
703 lapic_timer_set_divisor(u_int divisor)
704 {
705
706 KASSERT(powerof2(divisor), ("lapic: invalid divisor %u", divisor));
707 KASSERT(ffs(divisor) <= sizeof(lapic_timer_divisors) /
708 sizeof(u_int32_t), ("lapic: invalid divisor %u", divisor));
709 lapic->dcr_timer = lapic_timer_divisors[ffs(divisor) - 1];
710 }
711
712 static void
713 lapic_timer_oneshot(u_int count)
714 {
715 u_int32_t value;
716
717 value = lapic->lvt_timer;
718 value &= ~APIC_LVTT_TM;
719 value |= APIC_LVTT_TM_ONE_SHOT;
720 lapic->lvt_timer = value;
721 lapic->icr_timer = count;
722 }
723
724 static void
725 lapic_timer_periodic(u_int count)
726 {
727 u_int32_t value;
728
729 value = lapic->lvt_timer;
730 value &= ~APIC_LVTT_TM;
731 value |= APIC_LVTT_TM_PERIODIC;
732 lapic->lvt_timer = value;
733 lapic->icr_timer = count;
734 }
735
736 static void
737 lapic_timer_enable_intr(void)
738 {
739 u_int32_t value;
740
741 value = lapic->lvt_timer;
742 value &= ~APIC_LVT_M;
743 lapic->lvt_timer = value;
744 }
745
746 /* Request a free IDT vector to be used by the specified IRQ. */
747 u_int
748 apic_alloc_vector(u_int irq)
749 {
750 u_int vector;
751
752 KASSERT(irq < NUM_IO_INTS, ("Invalid IRQ %u", irq));
753
754 /*
755 * Search for a free vector. Currently we just use a very simple
756 * algorithm to find the first free vector.
757 */
758 mtx_lock_spin(&icu_lock);
759 for (vector = 0; vector < APIC_NUM_IOINTS; vector++) {
760 if (ioint_irqs[vector] != 0)
761 continue;
762 ioint_irqs[vector] = irq;
763 mtx_unlock_spin(&icu_lock);
764 return (vector + APIC_IO_INTS);
765 }
766 mtx_unlock_spin(&icu_lock);
767 panic("Couldn't find an APIC vector for IRQ %u", irq);
768 }
769
770 /*
771 * Request 'count' free contiguous IDT vectors to be used by 'count'
772 * IRQs. 'count' must be a power of two and the vectors will be
773 * aligned on a boundary of 'align'. If the request cannot be
774 * satisfied, 0 is returned.
775 */
776 u_int
777 apic_alloc_vectors(u_int *irqs, u_int count, u_int align)
778 {
779 u_int first, run, vector;
780
781 KASSERT(powerof2(count), ("bad count"));
782 KASSERT(powerof2(align), ("bad align"));
783 KASSERT(align >= count, ("align < count"));
784 #ifdef INVARIANTS
785 for (run = 0; run < count; run++)
786 KASSERT(irqs[run] < NUM_IO_INTS, ("Invalid IRQ %u at index %u",
787 irqs[run], run));
788 #endif
789
790 /*
791 * Search for 'count' free vectors. As with apic_alloc_vector(),
792 * this just uses a simple first fit algorithm.
793 */
794 run = 0;
795 first = 0;
796 mtx_lock_spin(&icu_lock);
797 for (vector = 0; vector < APIC_NUM_IOINTS; vector++) {
798
799 /* Vector is in use, end run. */
800 if (ioint_irqs[vector] != 0) {
801 run = 0;
802 first = 0;
803 continue;
804 }
805
806 /* Start a new run if run == 0 and vector is aligned. */
807 if (run == 0) {
808 if ((vector & (align - 1)) != 0)
809 continue;
810 first = vector;
811 }
812 run++;
813
814 /* Keep looping if the run isn't long enough yet. */
815 if (run < count)
816 continue;
817
818 /* Found a run, assign IRQs and return the first vector. */
819 for (vector = 0; vector < count; vector++)
820 ioint_irqs[first + vector] = irqs[vector];
821 mtx_unlock_spin(&icu_lock);
822 return (first + APIC_IO_INTS);
823 }
824 mtx_unlock_spin(&icu_lock);
825 printf("APIC: Couldn't find APIC vectors for %u IRQs\n", count);
826 return (0);
827 }
828
829 void
830 apic_enable_vector(u_int vector)
831 {
832
833 KASSERT(vector != IDT_SYSCALL, ("Attempt to overwrite syscall entry"));
834 KASSERT(ioint_handlers[vector / 32] != NULL,
835 ("No ISR handler for vector %u", vector));
836 setidt(vector, ioint_handlers[vector / 32], SDT_SYSIGT, SEL_KPL, 0);
837 }
838
839 /* Release an APIC vector when it's no longer in use. */
840 void
841 apic_free_vector(u_int vector, u_int irq)
842 {
843 KASSERT(vector >= APIC_IO_INTS && vector != IDT_SYSCALL &&
844 vector <= APIC_IO_INTS + APIC_NUM_IOINTS,
845 ("Vector %u does not map to an IRQ line", vector));
846 KASSERT(irq < NUM_IO_INTS, ("Invalid IRQ %u", irq));
847 KASSERT(ioint_irqs[vector - APIC_IO_INTS] == irq, ("IRQ mismatch"));
848 mtx_lock_spin(&icu_lock);
849 ioint_irqs[vector - APIC_IO_INTS] = 0;
850 mtx_unlock_spin(&icu_lock);
851 }
852
853 /* Map an IDT vector (APIC) to an IRQ (interrupt source). */
854 u_int
855 apic_idt_to_irq(u_int vector)
856 {
857
858 KASSERT(vector >= APIC_IO_INTS && vector != IDT_SYSCALL &&
859 vector <= APIC_IO_INTS + APIC_NUM_IOINTS,
860 ("Vector %u does not map to an IRQ line", vector));
861 return (ioint_irqs[vector - APIC_IO_INTS]);
862 }
863
864 #ifdef DDB
865 /*
866 * Dump data about APIC IDT vector mappings.
867 */
868 DB_SHOW_COMMAND(apic, db_show_apic)
869 {
870 struct intsrc *isrc;
871 int quit, i, verbose;
872 u_int irq;
873
874 quit = 0;
875 if (strcmp(modif, "vv") == 0)
876 verbose = 2;
877 else if (strcmp(modif, "v") == 0)
878 verbose = 1;
879 else
880 verbose = 0;
881 db_setup_paging(db_simple_pager, &quit, db_lines_per_page);
882 for (i = 0; i < APIC_NUM_IOINTS + 1 && !quit; i++) {
883 irq = ioint_irqs[i];
884 if (irq != 0 && irq != IRQ_SYSCALL) {
885 db_printf("vec 0x%2x -> ", i + APIC_IO_INTS);
886 if (irq == IRQ_TIMER)
887 db_printf("lapic timer\n");
888 else if (irq < NUM_IO_INTS) {
889 isrc = intr_lookup_source(irq);
890 if (isrc == NULL || verbose == 0)
891 db_printf("IRQ %u\n", irq);
892 else
893 db_dump_intr_event(isrc->is_event,
894 verbose == 2);
895 } else
896 db_printf("IRQ %u ???\n", irq);
897 }
898 }
899 }
900
901 static void
902 dump_mask(const char *prefix, uint32_t v, int base)
903 {
904 int i, first;
905
906 first = 1;
907 for (i = 0; i < 32; i++)
908 if (v & (1 << i)) {
909 if (first) {
910 db_printf("%s:", prefix);
911 first = 0;
912 }
913 db_printf(" %02x", base + i);
914 }
915 if (!first)
916 db_printf("\n");
917 }
918
919 /* Show info from the lapic regs for this CPU. */
920 DB_SHOW_COMMAND(lapic, db_show_lapic)
921 {
922 uint32_t v;
923
924 db_printf("lapic ID = %d\n", lapic_id());
925 v = lapic->version;
926 db_printf("version = %d.%d\n", (v & APIC_VER_VERSION) >> 4,
927 v & 0xf);
928 db_printf("max LVT = %d\n", (v & APIC_VER_MAXLVT) >> MAXLVTSHIFT);
929 v = lapic->svr;
930 db_printf("SVR = %02x (%s)\n", v & APIC_SVR_VECTOR,
931 v & APIC_SVR_ENABLE ? "enabled" : "disabled");
932 db_printf("TPR = %02x\n", lapic->tpr);
933
934 #define dump_field(prefix, index) \
935 dump_mask(__XSTRING(prefix ## index), lapic->prefix ## index, \
936 index * 32)
937
938 db_printf("In-service Interrupts:\n");
939 dump_field(isr, 0);
940 dump_field(isr, 1);
941 dump_field(isr, 2);
942 dump_field(isr, 3);
943 dump_field(isr, 4);
944 dump_field(isr, 5);
945 dump_field(isr, 6);
946 dump_field(isr, 7);
947
948 db_printf("TMR Interrupts:\n");
949 dump_field(tmr, 0);
950 dump_field(tmr, 1);
951 dump_field(tmr, 2);
952 dump_field(tmr, 3);
953 dump_field(tmr, 4);
954 dump_field(tmr, 5);
955 dump_field(tmr, 6);
956 dump_field(tmr, 7);
957
958 db_printf("IRR Interrupts:\n");
959 dump_field(irr, 0);
960 dump_field(irr, 1);
961 dump_field(irr, 2);
962 dump_field(irr, 3);
963 dump_field(irr, 4);
964 dump_field(irr, 5);
965 dump_field(irr, 6);
966 dump_field(irr, 7);
967
968 #undef dump_field
969 }
970 #endif
971
972 /*
973 * APIC probing support code. This includes code to manage enumerators.
974 */
975
976 static SLIST_HEAD(, apic_enumerator) enumerators =
977 SLIST_HEAD_INITIALIZER(enumerators);
978 static struct apic_enumerator *best_enum;
979
980 void
981 apic_register_enumerator(struct apic_enumerator *enumerator)
982 {
983 #ifdef INVARIANTS
984 struct apic_enumerator *apic_enum;
985
986 SLIST_FOREACH(apic_enum, &enumerators, apic_next) {
987 if (apic_enum == enumerator)
988 panic("%s: Duplicate register of %s", __func__,
989 enumerator->apic_name);
990 }
991 #endif
992 SLIST_INSERT_HEAD(&enumerators, enumerator, apic_next);
993 }
994
995 /*
996 * We have to look for CPU's very, very early because certain subsystems
997 * want to know how many CPU's we have extremely early on in the boot
998 * process.
999 */
1000 static void
1001 apic_init(void *dummy __unused)
1002 {
1003 struct apic_enumerator *enumerator;
1004 int retval, best;
1005
1006 /* Don't probe if APIC mode is disabled. */
1007 if (resource_disabled("apic", 0))
1008 return;
1009
1010 /* First, probe all the enumerators to find the best match. */
1011 best_enum = NULL;
1012 best = 0;
1013 SLIST_FOREACH(enumerator, &enumerators, apic_next) {
1014 retval = enumerator->apic_probe();
1015 if (retval > 0)
1016 continue;
1017 if (best_enum == NULL || best < retval) {
1018 best_enum = enumerator;
1019 best = retval;
1020 }
1021 }
1022 if (best_enum == NULL) {
1023 if (bootverbose)
1024 printf("APIC: Could not find any APICs.\n");
1025 return;
1026 }
1027
1028 if (bootverbose)
1029 printf("APIC: Using the %s enumerator.\n",
1030 best_enum->apic_name);
1031
1032 /* Second, probe the CPU's in the system. */
1033 retval = best_enum->apic_probe_cpus();
1034 if (retval != 0)
1035 printf("%s: Failed to probe CPUs: returned %d\n",
1036 best_enum->apic_name, retval);
1037 }
1038 SYSINIT(apic_init, SI_SUB_TUNABLES - 1, SI_ORDER_SECOND, apic_init, NULL)
1039
1040 /*
1041 * Setup the local APIC. We have to do this prior to starting up the APs
1042 * in the SMP case.
1043 */
1044 static void
1045 apic_setup_local(void *dummy __unused)
1046 {
1047 int retval;
1048
1049 if (best_enum == NULL)
1050 return;
1051 retval = best_enum->apic_setup_local();
1052 if (retval != 0)
1053 printf("%s: Failed to setup the local APIC: returned %d\n",
1054 best_enum->apic_name, retval);
1055 #ifdef SMP
1056 /* Last, setup the cpu topology now that we have probed CPUs */
1057 mp_topology();
1058 #endif
1059 }
1060 SYSINIT(apic_setup_local, SI_SUB_CPU, SI_ORDER_FIRST, apic_setup_local, NULL)
1061
1062 /*
1063 * Setup the I/O APICs.
1064 */
1065 static void
1066 apic_setup_io(void *dummy __unused)
1067 {
1068 int retval;
1069
1070 if (best_enum == NULL)
1071 return;
1072 retval = best_enum->apic_setup_io();
1073 if (retval != 0)
1074 printf("%s: Failed to setup I/O APICs: returned %d\n",
1075 best_enum->apic_name, retval);
1076
1077 /*
1078 * Finish setting up the local APIC on the BSP once we know how to
1079 * properly program the LINT pins.
1080 */
1081 lapic_setup(1);
1082 intr_register_pic(&lapic_pic);
1083 if (bootverbose)
1084 lapic_dump("BSP");
1085
1086 /* Enable the MSI "pic". */
1087 msi_init();
1088 }
1089 SYSINIT(apic_setup_io, SI_SUB_INTR, SI_ORDER_SECOND, apic_setup_io, NULL)
1090
1091 #ifdef SMP
1092 /*
1093 * Inter Processor Interrupt functions. The lapic_ipi_*() functions are
1094 * private to the sys/amd64 code. The public interface for the rest of the
1095 * kernel is defined in mp_machdep.c.
1096 */
1097 int
1098 lapic_ipi_wait(int delay)
1099 {
1100 int x, incr;
1101
1102 /*
1103 * Wait delay loops for IPI to be sent. This is highly bogus
1104 * since this is sensitive to CPU clock speed. If delay is
1105 * -1, we wait forever.
1106 */
1107 if (delay == -1) {
1108 incr = 0;
1109 delay = 1;
1110 } else
1111 incr = 1;
1112 for (x = 0; x < delay; x += incr) {
1113 if ((lapic->icr_lo & APIC_DELSTAT_MASK) == APIC_DELSTAT_IDLE)
1114 return (1);
1115 ia32_pause();
1116 }
1117 return (0);
1118 }
1119
1120 void
1121 lapic_ipi_raw(register_t icrlo, u_int dest)
1122 {
1123 register_t value, eflags;
1124
1125 /* XXX: Need more sanity checking of icrlo? */
1126 KASSERT(lapic != NULL, ("%s called too early", __func__));
1127 KASSERT((dest & ~(APIC_ID_MASK >> APIC_ID_SHIFT)) == 0,
1128 ("%s: invalid dest field", __func__));
1129 KASSERT((icrlo & APIC_ICRLO_RESV_MASK) == 0,
1130 ("%s: reserved bits set in ICR LO register", __func__));
1131
1132 /* Set destination in ICR HI register if it is being used. */
1133 eflags = intr_disable();
1134 if ((icrlo & APIC_DEST_MASK) == APIC_DEST_DESTFLD) {
1135 value = lapic->icr_hi;
1136 value &= ~APIC_ID_MASK;
1137 value |= dest << APIC_ID_SHIFT;
1138 lapic->icr_hi = value;
1139 }
1140
1141 /* Program the contents of the IPI and dispatch it. */
1142 value = lapic->icr_lo;
1143 value &= APIC_ICRLO_RESV_MASK;
1144 value |= icrlo;
1145 lapic->icr_lo = value;
1146 intr_restore(eflags);
1147 }
1148
1149 #define BEFORE_SPIN 1000000
1150 #ifdef DETECT_DEADLOCK
1151 #define AFTER_SPIN 1000
1152 #endif
1153
1154 void
1155 lapic_ipi_vectored(u_int vector, int dest)
1156 {
1157 register_t icrlo, destfield;
1158
1159 KASSERT((vector & ~APIC_VECTOR_MASK) == 0,
1160 ("%s: invalid vector %d", __func__, vector));
1161
1162 icrlo = vector | APIC_DELMODE_FIXED | APIC_DESTMODE_PHY |
1163 APIC_LEVEL_DEASSERT | APIC_TRIGMOD_EDGE;
1164 destfield = 0;
1165 switch (dest) {
1166 case APIC_IPI_DEST_SELF:
1167 icrlo |= APIC_DEST_SELF;
1168 break;
1169 case APIC_IPI_DEST_ALL:
1170 icrlo |= APIC_DEST_ALLISELF;
1171 break;
1172 case APIC_IPI_DEST_OTHERS:
1173 icrlo |= APIC_DEST_ALLESELF;
1174 break;
1175 default:
1176 KASSERT((dest & ~(APIC_ID_MASK >> APIC_ID_SHIFT)) == 0,
1177 ("%s: invalid destination 0x%x", __func__, dest));
1178 destfield = dest;
1179 }
1180
1181 /* Wait for an earlier IPI to finish. */
1182 if (!lapic_ipi_wait(BEFORE_SPIN)) {
1183 if (panicstr != NULL)
1184 return;
1185 else
1186 panic("APIC: Previous IPI is stuck");
1187 }
1188
1189 lapic_ipi_raw(icrlo, destfield);
1190
1191 #ifdef DETECT_DEADLOCK
1192 /* Wait for IPI to be delivered. */
1193 if (!lapic_ipi_wait(AFTER_SPIN)) {
1194 #ifdef needsattention
1195 /*
1196 * XXX FIXME:
1197 *
1198 * The above function waits for the message to actually be
1199 * delivered. It breaks out after an arbitrary timeout
1200 * since the message should eventually be delivered (at
1201 * least in theory) and that if it wasn't we would catch
1202 * the failure with the check above when the next IPI is
1203 * sent.
1204 *
1205 * We could skip this wait entirely, EXCEPT it probably
1206 * protects us from other routines that assume that the
1207 * message was delivered and acted upon when this function
1208 * returns.
1209 */
1210 printf("APIC: IPI might be stuck\n");
1211 #else /* !needsattention */
1212 /* Wait until mesage is sent without a timeout. */
1213 while (lapic->icr_lo & APIC_DELSTAT_PEND)
1214 ia32_pause();
1215 #endif /* needsattention */
1216 }
1217 #endif /* DETECT_DEADLOCK */
1218 }
1219 #endif /* SMP */
Cache object: eb9965beaf50909b8d61cda5d7d6375e
|