xref: /freebsd/sys/x86/x86/io_apic.c (revision 076ad2f8)
1 /*-
2  * Copyright (c) 2003 John Baldwin <jhb@FreeBSD.org>
3  * 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, this list of conditions and the following disclaimer.
10  * 2. Redistributions in binary form must reproduce the above copyright
11  *    notice, this list of conditions and the following disclaimer in the
12  *    documentation and/or other materials provided with the distribution.
13  *
14  * THIS SOFTWARE IS PROVIDED BY THE AUTHOR AND CONTRIBUTORS ``AS IS'' AND
15  * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
16  * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
17  * ARE DISCLAIMED.  IN NO EVENT SHALL THE AUTHOR OR CONTRIBUTORS BE LIABLE
18  * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
19  * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
20  * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
21  * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
22  * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
23  * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
24  * SUCH DAMAGE.
25  */
26 
27 #include <sys/cdefs.h>
28 __FBSDID("$FreeBSD$");
29 
30 #include "opt_acpi.h"
31 #include "opt_isa.h"
32 
33 #include <sys/param.h>
34 #include <sys/systm.h>
35 #include <sys/bus.h>
36 #include <sys/kernel.h>
37 #include <sys/lock.h>
38 #include <sys/malloc.h>
39 #include <sys/module.h>
40 #include <sys/mutex.h>
41 #include <sys/sysctl.h>
42 
43 #include <dev/pci/pcireg.h>
44 #include <dev/pci/pcivar.h>
45 
46 #include <vm/vm.h>
47 #include <vm/pmap.h>
48 
49 #include <x86/apicreg.h>
50 #include <machine/frame.h>
51 #include <machine/intr_machdep.h>
52 #include <x86/apicvar.h>
53 #include <machine/resource.h>
54 #include <machine/segments.h>
55 #include <x86/iommu/iommu_intrmap.h>
56 
57 #define IOAPIC_ISA_INTS		16
58 #define	IOAPIC_MEM_REGION	32
59 #define	IOAPIC_REDTBL_LO(i)	(IOAPIC_REDTBL + (i) * 2)
60 #define	IOAPIC_REDTBL_HI(i)	(IOAPIC_REDTBL_LO(i) + 1)
61 
62 static MALLOC_DEFINE(M_IOAPIC, "io_apic", "I/O APIC structures");
63 
64 /*
65  * I/O APIC interrupt source driver.  Each pin is assigned an IRQ cookie
66  * as laid out in the ACPI System Interrupt number model where each I/O
67  * APIC has a contiguous chunk of the System Interrupt address space.
68  * We assume that IRQs 1 - 15 behave like ISA IRQs and that all other
69  * IRQs behave as PCI IRQs by default.  We also assume that the pin for
70  * IRQ 0 is actually an ExtINT pin.  The apic enumerators override the
71  * configuration of individual pins as indicated by their tables.
72  *
73  * Documentation for the I/O APIC: "82093AA I/O Advanced Programmable
74  * Interrupt Controller (IOAPIC)", May 1996, Intel Corp.
75  * ftp://download.intel.com/design/chipsets/datashts/29056601.pdf
76  */
77 
78 struct ioapic_intsrc {
79 	struct intsrc io_intsrc;
80 	u_int io_irq;
81 	u_int io_intpin:8;
82 	u_int io_vector:8;
83 	u_int io_cpu;
84 	u_int io_activehi:1;
85 	u_int io_edgetrigger:1;
86 	u_int io_masked:1;
87 	int io_bus:4;
88 	uint32_t io_lowreg;
89 	u_int io_remap_cookie;
90 };
91 
92 struct ioapic {
93 	struct pic io_pic;
94 	u_int io_id:8;			/* logical ID */
95 	u_int io_apic_id:4;
96 	u_int io_intbase:8;		/* System Interrupt base */
97 	u_int io_numintr:8;
98 	u_int io_haseoi:1;
99 	volatile ioapic_t *io_addr;	/* XXX: should use bus_space */
100 	vm_paddr_t io_paddr;
101 	STAILQ_ENTRY(ioapic) io_next;
102 	struct ioapic_intsrc io_pins[0];
103 };
104 
105 static u_int	ioapic_read(volatile ioapic_t *apic, int reg);
106 static void	ioapic_write(volatile ioapic_t *apic, int reg, u_int val);
107 static const char *ioapic_bus_string(int bus_type);
108 static void	ioapic_print_irq(struct ioapic_intsrc *intpin);
109 static void	ioapic_enable_source(struct intsrc *isrc);
110 static void	ioapic_disable_source(struct intsrc *isrc, int eoi);
111 static void	ioapic_eoi_source(struct intsrc *isrc);
112 static void	ioapic_enable_intr(struct intsrc *isrc);
113 static void	ioapic_disable_intr(struct intsrc *isrc);
114 static int	ioapic_vector(struct intsrc *isrc);
115 static int	ioapic_source_pending(struct intsrc *isrc);
116 static int	ioapic_config_intr(struct intsrc *isrc, enum intr_trigger trig,
117 		    enum intr_polarity pol);
118 static void	ioapic_resume(struct pic *pic, bool suspend_cancelled);
119 static int	ioapic_assign_cpu(struct intsrc *isrc, u_int apic_id);
120 static void	ioapic_program_intpin(struct ioapic_intsrc *intpin);
121 static void	ioapic_reprogram_intpin(struct intsrc *isrc);
122 
123 static STAILQ_HEAD(,ioapic) ioapic_list = STAILQ_HEAD_INITIALIZER(ioapic_list);
124 struct pic ioapic_template = {
125 	.pic_enable_source = ioapic_enable_source,
126 	.pic_disable_source = ioapic_disable_source,
127 	.pic_eoi_source = ioapic_eoi_source,
128 	.pic_enable_intr = ioapic_enable_intr,
129 	.pic_disable_intr = ioapic_disable_intr,
130 	.pic_vector = ioapic_vector,
131 	.pic_source_pending = ioapic_source_pending,
132 	.pic_suspend = NULL,
133 	.pic_resume = ioapic_resume,
134 	.pic_config_intr = ioapic_config_intr,
135 	.pic_assign_cpu = ioapic_assign_cpu,
136 	.pic_reprogram_pin = ioapic_reprogram_intpin,
137 };
138 
139 static int next_ioapic_base;
140 static u_int next_id;
141 
142 static int enable_extint;
143 SYSCTL_INT(_hw_apic, OID_AUTO, enable_extint, CTLFLAG_RDTUN, &enable_extint, 0,
144     "Enable the ExtINT pin in the first I/O APIC");
145 
146 static void
147 _ioapic_eoi_source(struct intsrc *isrc, int locked)
148 {
149 	struct ioapic_intsrc *src;
150 	struct ioapic *io;
151 	volatile uint32_t *apic_eoi;
152 	uint32_t low1;
153 
154 	lapic_eoi();
155 	if (!lapic_eoi_suppression)
156 		return;
157 	src = (struct ioapic_intsrc *)isrc;
158 	if (src->io_edgetrigger)
159 		return;
160 	io = (struct ioapic *)isrc->is_pic;
161 
162 	/*
163 	 * Handle targeted EOI for level-triggered pins, if broadcast
164 	 * EOI suppression is supported by LAPICs.
165 	 */
166 	if (io->io_haseoi) {
167 		/*
168 		 * If IOAPIC has EOI Register, simply write vector
169 		 * number into the reg.
170 		 */
171 		apic_eoi = (volatile uint32_t *)((volatile char *)
172 		    io->io_addr + IOAPIC_EOIR);
173 		*apic_eoi = src->io_vector;
174 	} else {
175 		/*
176 		 * Otherwise, if IO-APIC is too old to provide EOIR,
177 		 * do what Intel did for the Linux kernel. Temporary
178 		 * switch the pin to edge-trigger and back, masking
179 		 * the pin during the trick.
180 		 */
181 		if (!locked)
182 			mtx_lock_spin(&icu_lock);
183 		low1 = src->io_lowreg;
184 		low1 &= ~IOART_TRGRLVL;
185 		low1 |= IOART_TRGREDG | IOART_INTMSET;
186 		ioapic_write(io->io_addr, IOAPIC_REDTBL_LO(src->io_intpin),
187 		    low1);
188 		ioapic_write(io->io_addr, IOAPIC_REDTBL_LO(src->io_intpin),
189 		    src->io_lowreg);
190 		if (!locked)
191 			mtx_unlock_spin(&icu_lock);
192 	}
193 }
194 
195 static u_int
196 ioapic_read(volatile ioapic_t *apic, int reg)
197 {
198 
199 	mtx_assert(&icu_lock, MA_OWNED);
200 	apic->ioregsel = reg;
201 	return (apic->iowin);
202 }
203 
204 static void
205 ioapic_write(volatile ioapic_t *apic, int reg, u_int val)
206 {
207 
208 	mtx_assert(&icu_lock, MA_OWNED);
209 	apic->ioregsel = reg;
210 	apic->iowin = val;
211 }
212 
213 static const char *
214 ioapic_bus_string(int bus_type)
215 {
216 
217 	switch (bus_type) {
218 	case APIC_BUS_ISA:
219 		return ("ISA");
220 	case APIC_BUS_EISA:
221 		return ("EISA");
222 	case APIC_BUS_PCI:
223 		return ("PCI");
224 	default:
225 		return ("unknown");
226 	}
227 }
228 
229 static void
230 ioapic_print_irq(struct ioapic_intsrc *intpin)
231 {
232 
233 	switch (intpin->io_irq) {
234 	case IRQ_DISABLED:
235 		printf("disabled");
236 		break;
237 	case IRQ_EXTINT:
238 		printf("ExtINT");
239 		break;
240 	case IRQ_NMI:
241 		printf("NMI");
242 		break;
243 	case IRQ_SMI:
244 		printf("SMI");
245 		break;
246 	default:
247 		printf("%s IRQ %u", ioapic_bus_string(intpin->io_bus),
248 		    intpin->io_irq);
249 	}
250 }
251 
252 static void
253 ioapic_enable_source(struct intsrc *isrc)
254 {
255 	struct ioapic_intsrc *intpin = (struct ioapic_intsrc *)isrc;
256 	struct ioapic *io = (struct ioapic *)isrc->is_pic;
257 	uint32_t flags;
258 
259 	mtx_lock_spin(&icu_lock);
260 	if (intpin->io_masked) {
261 		flags = intpin->io_lowreg & ~IOART_INTMASK;
262 		ioapic_write(io->io_addr, IOAPIC_REDTBL_LO(intpin->io_intpin),
263 		    flags);
264 		intpin->io_masked = 0;
265 	}
266 	mtx_unlock_spin(&icu_lock);
267 }
268 
269 static void
270 ioapic_disable_source(struct intsrc *isrc, int eoi)
271 {
272 	struct ioapic_intsrc *intpin = (struct ioapic_intsrc *)isrc;
273 	struct ioapic *io = (struct ioapic *)isrc->is_pic;
274 	uint32_t flags;
275 
276 	mtx_lock_spin(&icu_lock);
277 	if (!intpin->io_masked && !intpin->io_edgetrigger) {
278 		flags = intpin->io_lowreg | IOART_INTMSET;
279 		ioapic_write(io->io_addr, IOAPIC_REDTBL_LO(intpin->io_intpin),
280 		    flags);
281 		intpin->io_masked = 1;
282 	}
283 
284 	if (eoi == PIC_EOI)
285 		_ioapic_eoi_source(isrc, 1);
286 
287 	mtx_unlock_spin(&icu_lock);
288 }
289 
290 static void
291 ioapic_eoi_source(struct intsrc *isrc)
292 {
293 
294 	_ioapic_eoi_source(isrc, 0);
295 }
296 
297 /*
298  * Completely program an intpin based on the data in its interrupt source
299  * structure.
300  */
301 static void
302 ioapic_program_intpin(struct ioapic_intsrc *intpin)
303 {
304 	struct ioapic *io = (struct ioapic *)intpin->io_intsrc.is_pic;
305 	uint32_t low, high, value;
306 #ifdef ACPI_DMAR
307 	int error;
308 #endif
309 
310 	/*
311 	 * If a pin is completely invalid or if it is valid but hasn't
312 	 * been enabled yet, just ensure that the pin is masked.
313 	 */
314 	mtx_assert(&icu_lock, MA_OWNED);
315 	if (intpin->io_irq == IRQ_DISABLED || (intpin->io_irq < NUM_IO_INTS &&
316 	    intpin->io_vector == 0)) {
317 		low = ioapic_read(io->io_addr,
318 		    IOAPIC_REDTBL_LO(intpin->io_intpin));
319 		if ((low & IOART_INTMASK) == IOART_INTMCLR)
320 			ioapic_write(io->io_addr,
321 			    IOAPIC_REDTBL_LO(intpin->io_intpin),
322 			    low | IOART_INTMSET);
323 #ifdef ACPI_DMAR
324 		mtx_unlock_spin(&icu_lock);
325 		iommu_unmap_ioapic_intr(io->io_apic_id,
326 		    &intpin->io_remap_cookie);
327 		mtx_lock_spin(&icu_lock);
328 #endif
329 		return;
330 	}
331 
332 #ifdef ACPI_DMAR
333 	mtx_unlock_spin(&icu_lock);
334 	error = iommu_map_ioapic_intr(io->io_apic_id,
335 	    intpin->io_cpu, intpin->io_vector, intpin->io_edgetrigger,
336 	    intpin->io_activehi, intpin->io_irq, &intpin->io_remap_cookie,
337 	    &high, &low);
338 	mtx_lock_spin(&icu_lock);
339 	if (error == 0) {
340 		ioapic_write(io->io_addr, IOAPIC_REDTBL_HI(intpin->io_intpin),
341 		    high);
342 		intpin->io_lowreg = low;
343 		ioapic_write(io->io_addr, IOAPIC_REDTBL_LO(intpin->io_intpin),
344 		    low);
345 		return;
346 	} else if (error != EOPNOTSUPP) {
347 		return;
348 	}
349 #endif
350 
351 	/* Set the destination. */
352 	low = IOART_DESTPHY;
353 	high = intpin->io_cpu << APIC_ID_SHIFT;
354 
355 	/* Program the rest of the low word. */
356 	if (intpin->io_edgetrigger)
357 		low |= IOART_TRGREDG;
358 	else
359 		low |= IOART_TRGRLVL;
360 	if (intpin->io_activehi)
361 		low |= IOART_INTAHI;
362 	else
363 		low |= IOART_INTALO;
364 	if (intpin->io_masked)
365 		low |= IOART_INTMSET;
366 	switch (intpin->io_irq) {
367 	case IRQ_EXTINT:
368 		KASSERT(intpin->io_edgetrigger,
369 		    ("ExtINT not edge triggered"));
370 		low |= IOART_DELEXINT;
371 		break;
372 	case IRQ_NMI:
373 		KASSERT(intpin->io_edgetrigger,
374 		    ("NMI not edge triggered"));
375 		low |= IOART_DELNMI;
376 		break;
377 	case IRQ_SMI:
378 		KASSERT(intpin->io_edgetrigger,
379 		    ("SMI not edge triggered"));
380 		low |= IOART_DELSMI;
381 		break;
382 	default:
383 		KASSERT(intpin->io_vector != 0, ("No vector for IRQ %u",
384 		    intpin->io_irq));
385 		low |= IOART_DELFIXED | intpin->io_vector;
386 	}
387 
388 	/* Write the values to the APIC. */
389 	value = ioapic_read(io->io_addr, IOAPIC_REDTBL_HI(intpin->io_intpin));
390 	value &= ~IOART_DEST;
391 	value |= high;
392 	ioapic_write(io->io_addr, IOAPIC_REDTBL_HI(intpin->io_intpin), value);
393 	intpin->io_lowreg = low;
394 	ioapic_write(io->io_addr, IOAPIC_REDTBL_LO(intpin->io_intpin), low);
395 }
396 
397 static void
398 ioapic_reprogram_intpin(struct intsrc *isrc)
399 {
400 
401 	mtx_lock_spin(&icu_lock);
402 	ioapic_program_intpin((struct ioapic_intsrc *)isrc);
403 	mtx_unlock_spin(&icu_lock);
404 }
405 
406 static int
407 ioapic_assign_cpu(struct intsrc *isrc, u_int apic_id)
408 {
409 	struct ioapic_intsrc *intpin = (struct ioapic_intsrc *)isrc;
410 	struct ioapic *io = (struct ioapic *)isrc->is_pic;
411 	u_int old_vector, new_vector;
412 	u_int old_id;
413 
414 	/*
415 	 * On Hyper-V:
416 	 * - Stick to the first cpu for all I/O APIC pins.
417 	 * - And don't allow destination cpu changes.
418 	 */
419 	if (vm_guest == VM_GUEST_HV) {
420 		if (intpin->io_vector)
421 			return (EINVAL);
422 		else
423 			apic_id = 0;
424 	}
425 
426 	/*
427 	 * keep 1st core as the destination for NMI
428 	 */
429 	if (intpin->io_irq == IRQ_NMI)
430 		apic_id = 0;
431 
432 	/*
433 	 * Set us up to free the old irq.
434 	 */
435 	old_vector = intpin->io_vector;
436 	old_id = intpin->io_cpu;
437 	if (old_vector && apic_id == old_id)
438 		return (0);
439 
440 	/*
441 	 * Allocate an APIC vector for this interrupt pin.  Once
442 	 * we have a vector we program the interrupt pin.
443 	 */
444 	new_vector = apic_alloc_vector(apic_id, intpin->io_irq);
445 	if (new_vector == 0)
446 		return (ENOSPC);
447 
448 	/*
449 	 * Mask the old intpin if it is enabled while it is migrated.
450 	 *
451 	 * At least some level-triggered interrupts seem to need the
452 	 * extra DELAY() to avoid being stuck in a non-EOI'd state.
453 	 */
454 	mtx_lock_spin(&icu_lock);
455 	if (!intpin->io_masked && !intpin->io_edgetrigger) {
456 		ioapic_write(io->io_addr, IOAPIC_REDTBL_LO(intpin->io_intpin),
457 		    intpin->io_lowreg | IOART_INTMSET);
458 		mtx_unlock_spin(&icu_lock);
459 		DELAY(100);
460 		mtx_lock_spin(&icu_lock);
461 	}
462 
463 	intpin->io_cpu = apic_id;
464 	intpin->io_vector = new_vector;
465 	if (isrc->is_handlers > 0)
466 		apic_enable_vector(intpin->io_cpu, intpin->io_vector);
467 	if (bootverbose) {
468 		printf("ioapic%u: routing intpin %u (", io->io_id,
469 		    intpin->io_intpin);
470 		ioapic_print_irq(intpin);
471 		printf(") to lapic %u vector %u\n", intpin->io_cpu,
472 		    intpin->io_vector);
473 	}
474 	ioapic_program_intpin(intpin);
475 	mtx_unlock_spin(&icu_lock);
476 
477 	/*
478 	 * Free the old vector after the new one is established.  This is done
479 	 * to prevent races where we could miss an interrupt.
480 	 */
481 	if (old_vector) {
482 		if (isrc->is_handlers > 0)
483 			apic_disable_vector(old_id, old_vector);
484 		apic_free_vector(old_id, old_vector, intpin->io_irq);
485 	}
486 	return (0);
487 }
488 
489 static void
490 ioapic_enable_intr(struct intsrc *isrc)
491 {
492 	struct ioapic_intsrc *intpin = (struct ioapic_intsrc *)isrc;
493 
494 	if (intpin->io_vector == 0)
495 		if (ioapic_assign_cpu(isrc, intr_next_cpu()) != 0)
496 			panic("Couldn't find an APIC vector for IRQ %d",
497 			    intpin->io_irq);
498 	apic_enable_vector(intpin->io_cpu, intpin->io_vector);
499 }
500 
501 
502 static void
503 ioapic_disable_intr(struct intsrc *isrc)
504 {
505 	struct ioapic_intsrc *intpin = (struct ioapic_intsrc *)isrc;
506 	u_int vector;
507 
508 	if (intpin->io_vector != 0) {
509 		/* Mask this interrupt pin and free its APIC vector. */
510 		vector = intpin->io_vector;
511 		apic_disable_vector(intpin->io_cpu, vector);
512 		mtx_lock_spin(&icu_lock);
513 		intpin->io_masked = 1;
514 		intpin->io_vector = 0;
515 		ioapic_program_intpin(intpin);
516 		mtx_unlock_spin(&icu_lock);
517 		apic_free_vector(intpin->io_cpu, vector, intpin->io_irq);
518 	}
519 }
520 
521 static int
522 ioapic_vector(struct intsrc *isrc)
523 {
524 	struct ioapic_intsrc *pin;
525 
526 	pin = (struct ioapic_intsrc *)isrc;
527 	return (pin->io_irq);
528 }
529 
530 static int
531 ioapic_source_pending(struct intsrc *isrc)
532 {
533 	struct ioapic_intsrc *intpin = (struct ioapic_intsrc *)isrc;
534 
535 	if (intpin->io_vector == 0)
536 		return 0;
537 	return (lapic_intr_pending(intpin->io_vector));
538 }
539 
540 static int
541 ioapic_config_intr(struct intsrc *isrc, enum intr_trigger trig,
542     enum intr_polarity pol)
543 {
544 	struct ioapic_intsrc *intpin = (struct ioapic_intsrc *)isrc;
545 	struct ioapic *io = (struct ioapic *)isrc->is_pic;
546 	int changed;
547 
548 	KASSERT(!(trig == INTR_TRIGGER_CONFORM || pol == INTR_POLARITY_CONFORM),
549 	    ("%s: Conforming trigger or polarity\n", __func__));
550 
551 	/*
552 	 * EISA interrupts always use active high polarity, so don't allow
553 	 * them to be set to active low.
554 	 *
555 	 * XXX: Should we write to the ELCR if the trigger mode changes for
556 	 * an EISA IRQ or an ISA IRQ with the ELCR present?
557 	 */
558 	mtx_lock_spin(&icu_lock);
559 	if (intpin->io_bus == APIC_BUS_EISA)
560 		pol = INTR_POLARITY_HIGH;
561 	changed = 0;
562 	if (intpin->io_edgetrigger != (trig == INTR_TRIGGER_EDGE)) {
563 		if (bootverbose)
564 			printf("ioapic%u: Changing trigger for pin %u to %s\n",
565 			    io->io_id, intpin->io_intpin,
566 			    trig == INTR_TRIGGER_EDGE ? "edge" : "level");
567 		intpin->io_edgetrigger = (trig == INTR_TRIGGER_EDGE);
568 		changed++;
569 	}
570 	if (intpin->io_activehi != (pol == INTR_POLARITY_HIGH)) {
571 		if (bootverbose)
572 			printf("ioapic%u: Changing polarity for pin %u to %s\n",
573 			    io->io_id, intpin->io_intpin,
574 			    pol == INTR_POLARITY_HIGH ? "high" : "low");
575 		intpin->io_activehi = (pol == INTR_POLARITY_HIGH);
576 		changed++;
577 	}
578 	if (changed)
579 		ioapic_program_intpin(intpin);
580 	mtx_unlock_spin(&icu_lock);
581 	return (0);
582 }
583 
584 static void
585 ioapic_resume(struct pic *pic, bool suspend_cancelled)
586 {
587 	struct ioapic *io = (struct ioapic *)pic;
588 	int i;
589 
590 	mtx_lock_spin(&icu_lock);
591 	for (i = 0; i < io->io_numintr; i++)
592 		ioapic_program_intpin(&io->io_pins[i]);
593 	mtx_unlock_spin(&icu_lock);
594 }
595 
596 /*
597  * Create a plain I/O APIC object.
598  */
599 void *
600 ioapic_create(vm_paddr_t addr, int32_t apic_id, int intbase)
601 {
602 	struct ioapic *io;
603 	struct ioapic_intsrc *intpin;
604 	volatile ioapic_t *apic;
605 	u_int numintr, i;
606 	uint32_t value;
607 
608 	/* Map the register window so we can access the device. */
609 	apic = pmap_mapdev(addr, IOAPIC_MEM_REGION);
610 	mtx_lock_spin(&icu_lock);
611 	value = ioapic_read(apic, IOAPIC_VER);
612 	mtx_unlock_spin(&icu_lock);
613 
614 	/* If it's version register doesn't seem to work, punt. */
615 	if (value == 0xffffffff) {
616 		pmap_unmapdev((vm_offset_t)apic, IOAPIC_MEM_REGION);
617 		return (NULL);
618 	}
619 
620 	/* Determine the number of vectors and set the APIC ID. */
621 	numintr = ((value & IOART_VER_MAXREDIR) >> MAXREDIRSHIFT) + 1;
622 	io = malloc(sizeof(struct ioapic) +
623 	    numintr * sizeof(struct ioapic_intsrc), M_IOAPIC, M_WAITOK);
624 	io->io_pic = ioapic_template;
625 	mtx_lock_spin(&icu_lock);
626 	io->io_id = next_id++;
627 	io->io_apic_id = ioapic_read(apic, IOAPIC_ID) >> APIC_ID_SHIFT;
628 	if (apic_id != -1 && io->io_apic_id != apic_id) {
629 		ioapic_write(apic, IOAPIC_ID, apic_id << APIC_ID_SHIFT);
630 		mtx_unlock_spin(&icu_lock);
631 		io->io_apic_id = apic_id;
632 		printf("ioapic%u: Changing APIC ID to %d\n", io->io_id,
633 		    apic_id);
634 	} else
635 		mtx_unlock_spin(&icu_lock);
636 	if (intbase == -1) {
637 		intbase = next_ioapic_base;
638 		printf("ioapic%u: Assuming intbase of %d\n", io->io_id,
639 		    intbase);
640 	} else if (intbase != next_ioapic_base && bootverbose)
641 		printf("ioapic%u: WARNING: intbase %d != expected base %d\n",
642 		    io->io_id, intbase, next_ioapic_base);
643 	io->io_intbase = intbase;
644 	next_ioapic_base = intbase + numintr;
645 	io->io_numintr = numintr;
646 	io->io_addr = apic;
647 	io->io_paddr = addr;
648 
649 	if (bootverbose) {
650 		printf("ioapic%u: ver 0x%02x maxredir 0x%02x\n", io->io_id,
651 		    (value & IOART_VER_VERSION), (value & IOART_VER_MAXREDIR)
652 		    >> MAXREDIRSHIFT);
653 	}
654 	/*
655 	 * The  summary information about IO-APIC versions is taken from
656 	 * the Linux kernel source:
657 	 *     0Xh     82489DX
658 	 *     1Xh     I/OAPIC or I/O(x)APIC which are not PCI 2.2 Compliant
659 	 *     2Xh     I/O(x)APIC which is PCI 2.2 Compliant
660 	 *     30h-FFh Reserved
661 	 * IO-APICs with version >= 0x20 have working EOIR register.
662 	 */
663 	io->io_haseoi = (value & IOART_VER_VERSION) >= 0x20;
664 
665 	/*
666 	 * Initialize pins.  Start off with interrupts disabled.  Default
667 	 * to active-hi and edge-triggered for ISA interrupts and active-lo
668 	 * and level-triggered for all others.
669 	 */
670 	bzero(io->io_pins, sizeof(struct ioapic_intsrc) * numintr);
671 	mtx_lock_spin(&icu_lock);
672 	for (i = 0, intpin = io->io_pins; i < numintr; i++, intpin++) {
673 		intpin->io_intsrc.is_pic = (struct pic *)io;
674 		intpin->io_intpin = i;
675 		intpin->io_irq = intbase + i;
676 
677 		/*
678 		 * Assume that pin 0 on the first I/O APIC is an ExtINT pin.
679 		 * Assume that pins 1-15 are ISA interrupts and that all
680 		 * other pins are PCI interrupts.
681 		 */
682 		if (intpin->io_irq == 0)
683 			ioapic_set_extint(io, i);
684 		else if (intpin->io_irq < IOAPIC_ISA_INTS) {
685 			intpin->io_bus = APIC_BUS_ISA;
686 			intpin->io_activehi = 1;
687 			intpin->io_edgetrigger = 1;
688 			intpin->io_masked = 1;
689 		} else {
690 			intpin->io_bus = APIC_BUS_PCI;
691 			intpin->io_activehi = 0;
692 			intpin->io_edgetrigger = 0;
693 			intpin->io_masked = 1;
694 		}
695 
696 		/*
697 		 * Route interrupts to the BSP by default.  Interrupts may
698 		 * be routed to other CPUs later after they are enabled.
699 		 */
700 		intpin->io_cpu = PCPU_GET(apic_id);
701 		value = ioapic_read(apic, IOAPIC_REDTBL_LO(i));
702 		ioapic_write(apic, IOAPIC_REDTBL_LO(i), value | IOART_INTMSET);
703 #ifdef ACPI_DMAR
704 		/* dummy, but sets cookie */
705 		mtx_unlock_spin(&icu_lock);
706 		iommu_map_ioapic_intr(io->io_apic_id,
707 		    intpin->io_cpu, intpin->io_vector, intpin->io_edgetrigger,
708 		    intpin->io_activehi, intpin->io_irq,
709 		    &intpin->io_remap_cookie, NULL, NULL);
710 		mtx_lock_spin(&icu_lock);
711 #endif
712 	}
713 	mtx_unlock_spin(&icu_lock);
714 
715 	return (io);
716 }
717 
718 int
719 ioapic_get_vector(void *cookie, u_int pin)
720 {
721 	struct ioapic *io;
722 
723 	io = (struct ioapic *)cookie;
724 	if (pin >= io->io_numintr)
725 		return (-1);
726 	return (io->io_pins[pin].io_irq);
727 }
728 
729 int
730 ioapic_disable_pin(void *cookie, u_int pin)
731 {
732 	struct ioapic *io;
733 
734 	io = (struct ioapic *)cookie;
735 	if (pin >= io->io_numintr)
736 		return (EINVAL);
737 	if (io->io_pins[pin].io_irq == IRQ_DISABLED)
738 		return (EINVAL);
739 	io->io_pins[pin].io_irq = IRQ_DISABLED;
740 	if (bootverbose)
741 		printf("ioapic%u: intpin %d disabled\n", io->io_id, pin);
742 	return (0);
743 }
744 
745 int
746 ioapic_remap_vector(void *cookie, u_int pin, int vector)
747 {
748 	struct ioapic *io;
749 
750 	io = (struct ioapic *)cookie;
751 	if (pin >= io->io_numintr || vector < 0)
752 		return (EINVAL);
753 	if (io->io_pins[pin].io_irq >= NUM_IO_INTS)
754 		return (EINVAL);
755 	io->io_pins[pin].io_irq = vector;
756 	if (bootverbose)
757 		printf("ioapic%u: Routing IRQ %d -> intpin %d\n", io->io_id,
758 		    vector, pin);
759 	return (0);
760 }
761 
762 int
763 ioapic_set_bus(void *cookie, u_int pin, int bus_type)
764 {
765 	struct ioapic *io;
766 
767 	if (bus_type < 0 || bus_type > APIC_BUS_MAX)
768 		return (EINVAL);
769 	io = (struct ioapic *)cookie;
770 	if (pin >= io->io_numintr)
771 		return (EINVAL);
772 	if (io->io_pins[pin].io_irq >= NUM_IO_INTS)
773 		return (EINVAL);
774 	if (io->io_pins[pin].io_bus == bus_type)
775 		return (0);
776 	io->io_pins[pin].io_bus = bus_type;
777 	if (bootverbose)
778 		printf("ioapic%u: intpin %d bus %s\n", io->io_id, pin,
779 		    ioapic_bus_string(bus_type));
780 	return (0);
781 }
782 
783 int
784 ioapic_set_nmi(void *cookie, u_int pin)
785 {
786 	struct ioapic *io;
787 
788 	io = (struct ioapic *)cookie;
789 	if (pin >= io->io_numintr)
790 		return (EINVAL);
791 	if (io->io_pins[pin].io_irq == IRQ_NMI)
792 		return (0);
793 	if (io->io_pins[pin].io_irq >= NUM_IO_INTS)
794 		return (EINVAL);
795 	io->io_pins[pin].io_bus = APIC_BUS_UNKNOWN;
796 	io->io_pins[pin].io_irq = IRQ_NMI;
797 	io->io_pins[pin].io_masked = 0;
798 	io->io_pins[pin].io_edgetrigger = 1;
799 	io->io_pins[pin].io_activehi = 1;
800 	if (bootverbose)
801 		printf("ioapic%u: Routing NMI -> intpin %d\n",
802 		    io->io_id, pin);
803 	return (0);
804 }
805 
806 int
807 ioapic_set_smi(void *cookie, u_int pin)
808 {
809 	struct ioapic *io;
810 
811 	io = (struct ioapic *)cookie;
812 	if (pin >= io->io_numintr)
813 		return (EINVAL);
814 	if (io->io_pins[pin].io_irq == IRQ_SMI)
815 		return (0);
816 	if (io->io_pins[pin].io_irq >= NUM_IO_INTS)
817 		return (EINVAL);
818 	io->io_pins[pin].io_bus = APIC_BUS_UNKNOWN;
819 	io->io_pins[pin].io_irq = IRQ_SMI;
820 	io->io_pins[pin].io_masked = 0;
821 	io->io_pins[pin].io_edgetrigger = 1;
822 	io->io_pins[pin].io_activehi = 1;
823 	if (bootverbose)
824 		printf("ioapic%u: Routing SMI -> intpin %d\n",
825 		    io->io_id, pin);
826 	return (0);
827 }
828 
829 int
830 ioapic_set_extint(void *cookie, u_int pin)
831 {
832 	struct ioapic *io;
833 
834 	io = (struct ioapic *)cookie;
835 	if (pin >= io->io_numintr)
836 		return (EINVAL);
837 	if (io->io_pins[pin].io_irq == IRQ_EXTINT)
838 		return (0);
839 	if (io->io_pins[pin].io_irq >= NUM_IO_INTS)
840 		return (EINVAL);
841 	io->io_pins[pin].io_bus = APIC_BUS_UNKNOWN;
842 	io->io_pins[pin].io_irq = IRQ_EXTINT;
843 	if (enable_extint)
844 		io->io_pins[pin].io_masked = 0;
845 	else
846 		io->io_pins[pin].io_masked = 1;
847 	io->io_pins[pin].io_edgetrigger = 1;
848 	io->io_pins[pin].io_activehi = 1;
849 	if (bootverbose)
850 		printf("ioapic%u: Routing external 8259A's -> intpin %d\n",
851 		    io->io_id, pin);
852 	return (0);
853 }
854 
855 int
856 ioapic_set_polarity(void *cookie, u_int pin, enum intr_polarity pol)
857 {
858 	struct ioapic *io;
859 	int activehi;
860 
861 	io = (struct ioapic *)cookie;
862 	if (pin >= io->io_numintr || pol == INTR_POLARITY_CONFORM)
863 		return (EINVAL);
864 	if (io->io_pins[pin].io_irq >= NUM_IO_INTS)
865 		return (EINVAL);
866 	activehi = (pol == INTR_POLARITY_HIGH);
867 	if (io->io_pins[pin].io_activehi == activehi)
868 		return (0);
869 	io->io_pins[pin].io_activehi = activehi;
870 	if (bootverbose)
871 		printf("ioapic%u: intpin %d polarity: %s\n", io->io_id, pin,
872 		    pol == INTR_POLARITY_HIGH ? "high" : "low");
873 	return (0);
874 }
875 
876 int
877 ioapic_set_triggermode(void *cookie, u_int pin, enum intr_trigger trigger)
878 {
879 	struct ioapic *io;
880 	int edgetrigger;
881 
882 	io = (struct ioapic *)cookie;
883 	if (pin >= io->io_numintr || trigger == INTR_TRIGGER_CONFORM)
884 		return (EINVAL);
885 	if (io->io_pins[pin].io_irq >= NUM_IO_INTS)
886 		return (EINVAL);
887 	edgetrigger = (trigger == INTR_TRIGGER_EDGE);
888 	if (io->io_pins[pin].io_edgetrigger == edgetrigger)
889 		return (0);
890 	io->io_pins[pin].io_edgetrigger = edgetrigger;
891 	if (bootverbose)
892 		printf("ioapic%u: intpin %d trigger: %s\n", io->io_id, pin,
893 		    trigger == INTR_TRIGGER_EDGE ? "edge" : "level");
894 	return (0);
895 }
896 
897 /*
898  * Register a complete I/O APIC object with the interrupt subsystem.
899  */
900 void
901 ioapic_register(void *cookie)
902 {
903 	struct ioapic_intsrc *pin;
904 	struct ioapic *io;
905 	volatile ioapic_t *apic;
906 	uint32_t flags;
907 	int i;
908 
909 	io = (struct ioapic *)cookie;
910 	apic = io->io_addr;
911 	mtx_lock_spin(&icu_lock);
912 	flags = ioapic_read(apic, IOAPIC_VER) & IOART_VER_VERSION;
913 	STAILQ_INSERT_TAIL(&ioapic_list, io, io_next);
914 	mtx_unlock_spin(&icu_lock);
915 	printf("ioapic%u <Version %u.%u> irqs %u-%u on motherboard\n",
916 	    io->io_id, flags >> 4, flags & 0xf, io->io_intbase,
917 	    io->io_intbase + io->io_numintr - 1);
918 
919 	/*
920 	 * Reprogram pins to handle special case pins (such as NMI and
921 	 * SMI) and register valid pins as interrupt sources.
922 	 */
923 	intr_register_pic(&io->io_pic);
924 	for (i = 0, pin = io->io_pins; i < io->io_numintr; i++, pin++) {
925 		ioapic_reprogram_intpin(&pin->io_intsrc);
926 		if (pin->io_irq < NUM_IO_INTS)
927 			intr_register_source(&pin->io_intsrc);
928 	}
929 }
930 
931 /* A simple new-bus driver to consume PCI I/O APIC devices. */
932 static int
933 ioapic_pci_probe(device_t dev)
934 {
935 
936 	if (pci_get_class(dev) == PCIC_BASEPERIPH &&
937 	    pci_get_subclass(dev) == PCIS_BASEPERIPH_PIC) {
938 		switch (pci_get_progif(dev)) {
939 		case PCIP_BASEPERIPH_PIC_IO_APIC:
940 			device_set_desc(dev, "IO APIC");
941 			break;
942 		case PCIP_BASEPERIPH_PIC_IOX_APIC:
943 			device_set_desc(dev, "IO(x) APIC");
944 			break;
945 		default:
946 			return (ENXIO);
947 		}
948 		device_quiet(dev);
949 		return (-10000);
950 	}
951 	return (ENXIO);
952 }
953 
954 static int
955 ioapic_pci_attach(device_t dev)
956 {
957 
958 	return (0);
959 }
960 
961 static device_method_t ioapic_pci_methods[] = {
962 	/* Device interface */
963 	DEVMETHOD(device_probe,		ioapic_pci_probe),
964 	DEVMETHOD(device_attach,	ioapic_pci_attach),
965 
966 	{ 0, 0 }
967 };
968 
969 DEFINE_CLASS_0(ioapic, ioapic_pci_driver, ioapic_pci_methods, 0);
970 
971 static devclass_t ioapic_devclass;
972 DRIVER_MODULE(ioapic, pci, ioapic_pci_driver, ioapic_devclass, 0, 0);
973 
974 /*
975  * A new-bus driver to consume the memory resources associated with
976  * the APICs in the system.  On some systems ACPI or PnPBIOS system
977  * resource devices may already claim these resources.  To keep from
978  * breaking those devices, we attach ourself to the nexus device after
979  * legacy0 and acpi0 and ignore any allocation failures.
980  */
981 static void
982 apic_identify(driver_t *driver, device_t parent)
983 {
984 
985 	/*
986 	 * Add at order 12.  acpi0 is probed at order 10 and legacy0
987 	 * is probed at order 11.
988 	 */
989 	if (lapic_paddr != 0)
990 		BUS_ADD_CHILD(parent, 12, "apic", 0);
991 }
992 
993 static int
994 apic_probe(device_t dev)
995 {
996 
997 	device_set_desc(dev, "APIC resources");
998 	device_quiet(dev);
999 	return (0);
1000 }
1001 
1002 static void
1003 apic_add_resource(device_t dev, int rid, vm_paddr_t base, size_t length)
1004 {
1005 	int error;
1006 
1007 	error = bus_set_resource(dev, SYS_RES_MEMORY, rid, base, length);
1008 	if (error)
1009 		panic("apic_add_resource: resource %d failed set with %d", rid,
1010 		    error);
1011 	bus_alloc_resource_any(dev, SYS_RES_MEMORY, &rid, 0);
1012 }
1013 
1014 static int
1015 apic_attach(device_t dev)
1016 {
1017 	struct ioapic *io;
1018 	int i;
1019 
1020 	/* Reserve the local APIC. */
1021 	apic_add_resource(dev, 0, lapic_paddr, LAPIC_MEM_REGION);
1022 	i = 1;
1023 	STAILQ_FOREACH(io, &ioapic_list, io_next) {
1024 		apic_add_resource(dev, i, io->io_paddr, IOAPIC_MEM_REGION);
1025 		i++;
1026 	}
1027 	return (0);
1028 }
1029 
1030 static device_method_t apic_methods[] = {
1031 	/* Device interface */
1032 	DEVMETHOD(device_identify,	apic_identify),
1033 	DEVMETHOD(device_probe,		apic_probe),
1034 	DEVMETHOD(device_attach,	apic_attach),
1035 
1036 	{ 0, 0 }
1037 };
1038 
1039 DEFINE_CLASS_0(apic, apic_driver, apic_methods, 0);
1040 
1041 static devclass_t apic_devclass;
1042 DRIVER_MODULE(apic, nexus, apic_driver, apic_devclass, 0, 0);
1043 
1044 #include "opt_ddb.h"
1045 
1046 #ifdef DDB
1047 #include <ddb/ddb.h>
1048 
1049 static const char *
1050 ioapic_delivery_mode(uint32_t mode)
1051 {
1052 
1053 	switch (mode) {
1054 	case IOART_DELFIXED:
1055 		return ("fixed");
1056 	case IOART_DELLOPRI:
1057 		return ("lowestpri");
1058 	case IOART_DELSMI:
1059 		return ("SMI");
1060 	case IOART_DELRSV1:
1061 		return ("rsrvd1");
1062 	case IOART_DELNMI:
1063 		return ("NMI");
1064 	case IOART_DELINIT:
1065 		return ("INIT");
1066 	case IOART_DELRSV2:
1067 		return ("rsrvd2");
1068 	case IOART_DELEXINT:
1069 		return ("ExtINT");
1070 	default:
1071 		return ("");
1072 	}
1073 }
1074 
1075 static u_int
1076 db_ioapic_read(volatile ioapic_t *apic, int reg)
1077 {
1078 
1079 	apic->ioregsel = reg;
1080 	return (apic->iowin);
1081 }
1082 
1083 static void
1084 db_show_ioapic_one(volatile ioapic_t *io_addr)
1085 {
1086 	uint32_t r, lo, hi;
1087 	int mre, i;
1088 
1089 	r = db_ioapic_read(io_addr, IOAPIC_VER);
1090 	mre = (r & IOART_VER_MAXREDIR) >> MAXREDIRSHIFT;
1091 	db_printf("Id 0x%08x Ver 0x%02x MRE %d\n",
1092 	    db_ioapic_read(io_addr, IOAPIC_ID), r & IOART_VER_VERSION, mre);
1093 	for (i = 0; i < mre; i++) {
1094 		lo = db_ioapic_read(io_addr, IOAPIC_REDTBL_LO(i));
1095 		hi = db_ioapic_read(io_addr, IOAPIC_REDTBL_HI(i));
1096 		db_printf("  pin %d Dest %s/%x %smasked Trig %s RemoteIRR %d "
1097 		    "Polarity %s Status %s DeliveryMode %s Vec %d\n", i,
1098 		    (lo & IOART_DESTMOD) == IOART_DESTLOG ? "log" : "phy",
1099 		    (hi & IOART_DEST) >> 24,
1100 		    (lo & IOART_INTMASK) == IOART_INTMSET ? "" : "not",
1101 		    (lo & IOART_TRGRMOD) == IOART_TRGRLVL ? "lvl" : "edge",
1102 		    (lo & IOART_REM_IRR) == IOART_REM_IRR ? 1 : 0,
1103 		    (lo & IOART_INTPOL) == IOART_INTALO ? "low" : "high",
1104 		    (lo & IOART_DELIVS) == IOART_DELIVS ? "pend" : "idle",
1105 		    ioapic_delivery_mode(lo & IOART_DELMOD),
1106 		    (lo & IOART_INTVEC));
1107 	  }
1108 }
1109 
1110 DB_SHOW_COMMAND(ioapic, db_show_ioapic)
1111 {
1112 	struct ioapic *ioapic;
1113 	int idx, i;
1114 
1115 	if (!have_addr) {
1116 		db_printf("usage: show ioapic index\n");
1117 		return;
1118 	}
1119 
1120 	idx = (int)addr;
1121 	i = 0;
1122 	STAILQ_FOREACH(ioapic, &ioapic_list, io_next) {
1123 		if (idx == i) {
1124 			db_show_ioapic_one(ioapic->io_addr);
1125 			break;
1126 		}
1127 		i++;
1128 	}
1129 }
1130 
1131 DB_SHOW_ALL_COMMAND(ioapics, db_show_all_ioapics)
1132 {
1133 	struct ioapic *ioapic;
1134 
1135 	STAILQ_FOREACH(ioapic, &ioapic_list, io_next)
1136 		db_show_ioapic_one(ioapic->io_addr);
1137 }
1138 #endif
1139