xref: /linux/arch/x86/kvm/ioapic.c (revision 1d1997db870f4058676439ef7014390ba9e24eb2)
1 /*
2  *  Copyright (C) 2001  MandrakeSoft S.A.
3  *  Copyright 2010 Red Hat, Inc. and/or its affiliates.
4  *
5  *    MandrakeSoft S.A.
6  *    43, rue d'Aboukir
7  *    75002 Paris - France
8  *    http://www.linux-mandrake.com/
9  *    http://www.mandrakesoft.com/
10  *
11  *  This library is free software; you can redistribute it and/or
12  *  modify it under the terms of the GNU Lesser General Public
13  *  License as published by the Free Software Foundation; either
14  *  version 2 of the License, or (at your option) any later version.
15  *
16  *  This library is distributed in the hope that it will be useful,
17  *  but WITHOUT ANY WARRANTY; without even the implied warranty of
18  *  MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
19  *  Lesser General Public License for more details.
20  *
21  *  You should have received a copy of the GNU Lesser General Public
22  *  License along with this library; if not, write to the Free Software
23  *  Foundation, Inc., 59 Temple Place, Suite 330, Boston, MA  02111-1307 USA
24  *
25  *  Yunhong Jiang <yunhong.jiang@intel.com>
26  *  Yaozu (Eddie) Dong <eddie.dong@intel.com>
27  *  Based on Xen 3.1 code.
28  */
29 
30 #include <linux/kvm_host.h>
31 #include <linux/kvm.h>
32 #include <linux/mm.h>
33 #include <linux/highmem.h>
34 #include <linux/smp.h>
35 #include <linux/hrtimer.h>
36 #include <linux/io.h>
37 #include <linux/slab.h>
38 #include <linux/export.h>
39 #include <asm/processor.h>
40 #include <asm/page.h>
41 #include <asm/current.h>
42 #include <trace/events/kvm.h>
43 
44 #include "ioapic.h"
45 #include "lapic.h"
46 #include "irq.h"
47 
48 static int ioapic_service(struct kvm_ioapic *vioapic, int irq,
49 		bool line_status);
50 
51 static unsigned long ioapic_read_indirect(struct kvm_ioapic *ioapic,
52 					  unsigned long addr,
53 					  unsigned long length)
54 {
55 	unsigned long result = 0;
56 
57 	switch (ioapic->ioregsel) {
58 	case IOAPIC_REG_VERSION:
59 		result = ((((IOAPIC_NUM_PINS - 1) & 0xff) << 16)
60 			  | (IOAPIC_VERSION_ID & 0xff));
61 		break;
62 
63 	case IOAPIC_REG_APIC_ID:
64 	case IOAPIC_REG_ARB_ID:
65 		result = ((ioapic->id & 0xf) << 24);
66 		break;
67 
68 	default:
69 		{
70 			u32 redir_index = (ioapic->ioregsel - 0x10) >> 1;
71 			u64 redir_content;
72 
73 			if (redir_index < IOAPIC_NUM_PINS)
74 				redir_content =
75 					ioapic->redirtbl[redir_index].bits;
76 			else
77 				redir_content = ~0ULL;
78 
79 			result = (ioapic->ioregsel & 0x1) ?
80 			    (redir_content >> 32) & 0xffffffff :
81 			    redir_content & 0xffffffff;
82 			break;
83 		}
84 	}
85 
86 	return result;
87 }
88 
89 static void rtc_irq_eoi_tracking_reset(struct kvm_ioapic *ioapic)
90 {
91 	ioapic->rtc_status.pending_eoi = 0;
92 	bitmap_zero(ioapic->rtc_status.dest_map.map, KVM_MAX_VCPU_ID);
93 }
94 
95 static void kvm_rtc_eoi_tracking_restore_all(struct kvm_ioapic *ioapic);
96 
97 static void rtc_status_pending_eoi_check_valid(struct kvm_ioapic *ioapic)
98 {
99 	if (WARN_ON(ioapic->rtc_status.pending_eoi < 0))
100 		kvm_rtc_eoi_tracking_restore_all(ioapic);
101 }
102 
103 static void __rtc_irq_eoi_tracking_restore_one(struct kvm_vcpu *vcpu)
104 {
105 	bool new_val, old_val;
106 	struct kvm_ioapic *ioapic = vcpu->kvm->arch.vioapic;
107 	struct dest_map *dest_map = &ioapic->rtc_status.dest_map;
108 	union kvm_ioapic_redirect_entry *e;
109 
110 	e = &ioapic->redirtbl[RTC_GSI];
111 	if (!kvm_apic_match_dest(vcpu, NULL, 0,	e->fields.dest_id,
112 				e->fields.dest_mode))
113 		return;
114 
115 	new_val = kvm_apic_pending_eoi(vcpu, e->fields.vector);
116 	old_val = test_bit(vcpu->vcpu_id, dest_map->map);
117 
118 	if (new_val == old_val)
119 		return;
120 
121 	if (new_val) {
122 		__set_bit(vcpu->vcpu_id, dest_map->map);
123 		dest_map->vectors[vcpu->vcpu_id] = e->fields.vector;
124 		ioapic->rtc_status.pending_eoi++;
125 	} else {
126 		__clear_bit(vcpu->vcpu_id, dest_map->map);
127 		ioapic->rtc_status.pending_eoi--;
128 		rtc_status_pending_eoi_check_valid(ioapic);
129 	}
130 }
131 
132 void kvm_rtc_eoi_tracking_restore_one(struct kvm_vcpu *vcpu)
133 {
134 	struct kvm_ioapic *ioapic = vcpu->kvm->arch.vioapic;
135 
136 	spin_lock(&ioapic->lock);
137 	__rtc_irq_eoi_tracking_restore_one(vcpu);
138 	spin_unlock(&ioapic->lock);
139 }
140 
141 static void kvm_rtc_eoi_tracking_restore_all(struct kvm_ioapic *ioapic)
142 {
143 	struct kvm_vcpu *vcpu;
144 	int i;
145 
146 	if (RTC_GSI >= IOAPIC_NUM_PINS)
147 		return;
148 
149 	rtc_irq_eoi_tracking_reset(ioapic);
150 	kvm_for_each_vcpu(i, vcpu, ioapic->kvm)
151 	    __rtc_irq_eoi_tracking_restore_one(vcpu);
152 }
153 
154 static void rtc_irq_eoi(struct kvm_ioapic *ioapic, struct kvm_vcpu *vcpu)
155 {
156 	if (test_and_clear_bit(vcpu->vcpu_id,
157 			       ioapic->rtc_status.dest_map.map)) {
158 		--ioapic->rtc_status.pending_eoi;
159 		rtc_status_pending_eoi_check_valid(ioapic);
160 	}
161 }
162 
163 static bool rtc_irq_check_coalesced(struct kvm_ioapic *ioapic)
164 {
165 	if (ioapic->rtc_status.pending_eoi > 0)
166 		return true; /* coalesced */
167 
168 	return false;
169 }
170 
171 static int ioapic_set_irq(struct kvm_ioapic *ioapic, unsigned int irq,
172 		int irq_level, bool line_status)
173 {
174 	union kvm_ioapic_redirect_entry entry;
175 	u32 mask = 1 << irq;
176 	u32 old_irr;
177 	int edge, ret;
178 
179 	entry = ioapic->redirtbl[irq];
180 	edge = (entry.fields.trig_mode == IOAPIC_EDGE_TRIG);
181 
182 	if (!irq_level) {
183 		ioapic->irr &= ~mask;
184 		ret = 1;
185 		goto out;
186 	}
187 
188 	/*
189 	 * Return 0 for coalesced interrupts; for edge-triggered interrupts,
190 	 * this only happens if a previous edge has not been delivered due
191 	 * do masking.  For level interrupts, the remote_irr field tells
192 	 * us if the interrupt is waiting for an EOI.
193 	 *
194 	 * RTC is special: it is edge-triggered, but userspace likes to know
195 	 * if it has been already ack-ed via EOI because coalesced RTC
196 	 * interrupts lead to time drift in Windows guests.  So we track
197 	 * EOI manually for the RTC interrupt.
198 	 */
199 	if (irq == RTC_GSI && line_status &&
200 		rtc_irq_check_coalesced(ioapic)) {
201 		ret = 0;
202 		goto out;
203 	}
204 
205 	old_irr = ioapic->irr;
206 	ioapic->irr |= mask;
207 	if (edge) {
208 		ioapic->irr_delivered &= ~mask;
209 		if (old_irr == ioapic->irr) {
210 			ret = 0;
211 			goto out;
212 		}
213 	}
214 
215 	ret = ioapic_service(ioapic, irq, line_status);
216 
217 out:
218 	trace_kvm_ioapic_set_irq(entry.bits, irq, ret == 0);
219 	return ret;
220 }
221 
222 static void kvm_ioapic_inject_all(struct kvm_ioapic *ioapic, unsigned long irr)
223 {
224 	u32 idx;
225 
226 	rtc_irq_eoi_tracking_reset(ioapic);
227 	for_each_set_bit(idx, &irr, IOAPIC_NUM_PINS)
228 		ioapic_set_irq(ioapic, idx, 1, true);
229 
230 	kvm_rtc_eoi_tracking_restore_all(ioapic);
231 }
232 
233 
234 void kvm_ioapic_scan_entry(struct kvm_vcpu *vcpu, ulong *ioapic_handled_vectors)
235 {
236 	struct kvm_ioapic *ioapic = vcpu->kvm->arch.vioapic;
237 	struct dest_map *dest_map = &ioapic->rtc_status.dest_map;
238 	union kvm_ioapic_redirect_entry *e;
239 	int index;
240 
241 	spin_lock(&ioapic->lock);
242 
243 	/* Make sure we see any missing RTC EOI */
244 	if (test_bit(vcpu->vcpu_id, dest_map->map))
245 		__set_bit(dest_map->vectors[vcpu->vcpu_id],
246 			  ioapic_handled_vectors);
247 
248 	for (index = 0; index < IOAPIC_NUM_PINS; index++) {
249 		e = &ioapic->redirtbl[index];
250 		if (e->fields.trig_mode == IOAPIC_LEVEL_TRIG ||
251 		    kvm_irq_has_notifier(ioapic->kvm, KVM_IRQCHIP_IOAPIC, index) ||
252 		    index == RTC_GSI) {
253 			if (kvm_apic_match_dest(vcpu, NULL, 0,
254 			             e->fields.dest_id, e->fields.dest_mode) ||
255 			    kvm_apic_pending_eoi(vcpu, e->fields.vector))
256 				__set_bit(e->fields.vector,
257 					  ioapic_handled_vectors);
258 		}
259 	}
260 	spin_unlock(&ioapic->lock);
261 }
262 
263 void kvm_arch_post_irq_ack_notifier_list_update(struct kvm *kvm)
264 {
265 	if (!ioapic_in_kernel(kvm))
266 		return;
267 	kvm_make_scan_ioapic_request(kvm);
268 }
269 
270 static void ioapic_write_indirect(struct kvm_ioapic *ioapic, u32 val)
271 {
272 	unsigned index;
273 	bool mask_before, mask_after;
274 	union kvm_ioapic_redirect_entry *e;
275 	unsigned long vcpu_bitmap;
276 	int old_remote_irr, old_delivery_status, old_dest_id, old_dest_mode;
277 
278 	switch (ioapic->ioregsel) {
279 	case IOAPIC_REG_VERSION:
280 		/* Writes are ignored. */
281 		break;
282 
283 	case IOAPIC_REG_APIC_ID:
284 		ioapic->id = (val >> 24) & 0xf;
285 		break;
286 
287 	case IOAPIC_REG_ARB_ID:
288 		break;
289 
290 	default:
291 		index = (ioapic->ioregsel - 0x10) >> 1;
292 
293 		if (index >= IOAPIC_NUM_PINS)
294 			return;
295 		e = &ioapic->redirtbl[index];
296 		mask_before = e->fields.mask;
297 		/* Preserve read-only fields */
298 		old_remote_irr = e->fields.remote_irr;
299 		old_delivery_status = e->fields.delivery_status;
300 		old_dest_id = e->fields.dest_id;
301 		old_dest_mode = e->fields.dest_mode;
302 		if (ioapic->ioregsel & 1) {
303 			e->bits &= 0xffffffff;
304 			e->bits |= (u64) val << 32;
305 		} else {
306 			e->bits &= ~0xffffffffULL;
307 			e->bits |= (u32) val;
308 		}
309 		e->fields.remote_irr = old_remote_irr;
310 		e->fields.delivery_status = old_delivery_status;
311 
312 		/*
313 		 * Some OSes (Linux, Xen) assume that Remote IRR bit will
314 		 * be cleared by IOAPIC hardware when the entry is configured
315 		 * as edge-triggered. This behavior is used to simulate an
316 		 * explicit EOI on IOAPICs that don't have the EOI register.
317 		 */
318 		if (e->fields.trig_mode == IOAPIC_EDGE_TRIG)
319 			e->fields.remote_irr = 0;
320 
321 		mask_after = e->fields.mask;
322 		if (mask_before != mask_after)
323 			kvm_fire_mask_notifiers(ioapic->kvm, KVM_IRQCHIP_IOAPIC, index, mask_after);
324 		if (e->fields.trig_mode == IOAPIC_LEVEL_TRIG
325 		    && ioapic->irr & (1 << index))
326 			ioapic_service(ioapic, index, false);
327 		if (e->fields.delivery_mode == APIC_DM_FIXED) {
328 			struct kvm_lapic_irq irq;
329 
330 			irq.shorthand = 0;
331 			irq.vector = e->fields.vector;
332 			irq.delivery_mode = e->fields.delivery_mode << 8;
333 			irq.dest_id = e->fields.dest_id;
334 			irq.dest_mode = e->fields.dest_mode;
335 			bitmap_zero(&vcpu_bitmap, 16);
336 			kvm_bitmap_or_dest_vcpus(ioapic->kvm, &irq,
337 						 &vcpu_bitmap);
338 			if (old_dest_mode != e->fields.dest_mode ||
339 			    old_dest_id != e->fields.dest_id) {
340 				/*
341 				 * Update vcpu_bitmap with vcpus specified in
342 				 * the previous request as well. This is done to
343 				 * keep ioapic_handled_vectors synchronized.
344 				 */
345 				irq.dest_id = old_dest_id;
346 				irq.dest_mode = old_dest_mode;
347 				kvm_bitmap_or_dest_vcpus(ioapic->kvm, &irq,
348 							 &vcpu_bitmap);
349 			}
350 			kvm_make_scan_ioapic_request_mask(ioapic->kvm,
351 							  &vcpu_bitmap);
352 		} else {
353 			kvm_make_scan_ioapic_request(ioapic->kvm);
354 		}
355 		break;
356 	}
357 }
358 
359 static int ioapic_service(struct kvm_ioapic *ioapic, int irq, bool line_status)
360 {
361 	union kvm_ioapic_redirect_entry *entry = &ioapic->redirtbl[irq];
362 	struct kvm_lapic_irq irqe;
363 	int ret;
364 
365 	if (entry->fields.mask ||
366 	    (entry->fields.trig_mode == IOAPIC_LEVEL_TRIG &&
367 	    entry->fields.remote_irr))
368 		return -1;
369 
370 	irqe.dest_id = entry->fields.dest_id;
371 	irqe.vector = entry->fields.vector;
372 	irqe.dest_mode = entry->fields.dest_mode;
373 	irqe.trig_mode = entry->fields.trig_mode;
374 	irqe.delivery_mode = entry->fields.delivery_mode << 8;
375 	irqe.level = 1;
376 	irqe.shorthand = 0;
377 	irqe.msi_redir_hint = false;
378 
379 	if (irqe.trig_mode == IOAPIC_EDGE_TRIG)
380 		ioapic->irr_delivered |= 1 << irq;
381 
382 	if (irq == RTC_GSI && line_status) {
383 		/*
384 		 * pending_eoi cannot ever become negative (see
385 		 * rtc_status_pending_eoi_check_valid) and the caller
386 		 * ensures that it is only called if it is >= zero, namely
387 		 * if rtc_irq_check_coalesced returns false).
388 		 */
389 		BUG_ON(ioapic->rtc_status.pending_eoi != 0);
390 		ret = kvm_irq_delivery_to_apic(ioapic->kvm, NULL, &irqe,
391 					       &ioapic->rtc_status.dest_map);
392 		ioapic->rtc_status.pending_eoi = (ret < 0 ? 0 : ret);
393 	} else
394 		ret = kvm_irq_delivery_to_apic(ioapic->kvm, NULL, &irqe, NULL);
395 
396 	if (ret && irqe.trig_mode == IOAPIC_LEVEL_TRIG)
397 		entry->fields.remote_irr = 1;
398 
399 	return ret;
400 }
401 
402 int kvm_ioapic_set_irq(struct kvm_ioapic *ioapic, int irq, int irq_source_id,
403 		       int level, bool line_status)
404 {
405 	int ret, irq_level;
406 
407 	BUG_ON(irq < 0 || irq >= IOAPIC_NUM_PINS);
408 
409 	spin_lock(&ioapic->lock);
410 	irq_level = __kvm_irq_line_state(&ioapic->irq_states[irq],
411 					 irq_source_id, level);
412 	ret = ioapic_set_irq(ioapic, irq, irq_level, line_status);
413 
414 	spin_unlock(&ioapic->lock);
415 
416 	return ret;
417 }
418 
419 void kvm_ioapic_clear_all(struct kvm_ioapic *ioapic, int irq_source_id)
420 {
421 	int i;
422 
423 	spin_lock(&ioapic->lock);
424 	for (i = 0; i < KVM_IOAPIC_NUM_PINS; i++)
425 		__clear_bit(irq_source_id, &ioapic->irq_states[i]);
426 	spin_unlock(&ioapic->lock);
427 }
428 
429 static void kvm_ioapic_eoi_inject_work(struct work_struct *work)
430 {
431 	int i;
432 	struct kvm_ioapic *ioapic = container_of(work, struct kvm_ioapic,
433 						 eoi_inject.work);
434 	spin_lock(&ioapic->lock);
435 	for (i = 0; i < IOAPIC_NUM_PINS; i++) {
436 		union kvm_ioapic_redirect_entry *ent = &ioapic->redirtbl[i];
437 
438 		if (ent->fields.trig_mode != IOAPIC_LEVEL_TRIG)
439 			continue;
440 
441 		if (ioapic->irr & (1 << i) && !ent->fields.remote_irr)
442 			ioapic_service(ioapic, i, false);
443 	}
444 	spin_unlock(&ioapic->lock);
445 }
446 
447 #define IOAPIC_SUCCESSIVE_IRQ_MAX_COUNT 10000
448 
449 static void __kvm_ioapic_update_eoi(struct kvm_vcpu *vcpu,
450 			struct kvm_ioapic *ioapic, int vector, int trigger_mode)
451 {
452 	struct dest_map *dest_map = &ioapic->rtc_status.dest_map;
453 	struct kvm_lapic *apic = vcpu->arch.apic;
454 	int i;
455 
456 	/* RTC special handling */
457 	if (test_bit(vcpu->vcpu_id, dest_map->map) &&
458 	    vector == dest_map->vectors[vcpu->vcpu_id])
459 		rtc_irq_eoi(ioapic, vcpu);
460 
461 	for (i = 0; i < IOAPIC_NUM_PINS; i++) {
462 		union kvm_ioapic_redirect_entry *ent = &ioapic->redirtbl[i];
463 
464 		if (ent->fields.vector != vector)
465 			continue;
466 
467 		/*
468 		 * We are dropping lock while calling ack notifiers because ack
469 		 * notifier callbacks for assigned devices call into IOAPIC
470 		 * recursively. Since remote_irr is cleared only after call
471 		 * to notifiers if the same vector will be delivered while lock
472 		 * is dropped it will be put into irr and will be delivered
473 		 * after ack notifier returns.
474 		 */
475 		spin_unlock(&ioapic->lock);
476 		kvm_notify_acked_irq(ioapic->kvm, KVM_IRQCHIP_IOAPIC, i);
477 		spin_lock(&ioapic->lock);
478 
479 		if (trigger_mode != IOAPIC_LEVEL_TRIG ||
480 		    kvm_lapic_get_reg(apic, APIC_SPIV) & APIC_SPIV_DIRECTED_EOI)
481 			continue;
482 
483 		ASSERT(ent->fields.trig_mode == IOAPIC_LEVEL_TRIG);
484 		ent->fields.remote_irr = 0;
485 		if (!ent->fields.mask && (ioapic->irr & (1 << i))) {
486 			++ioapic->irq_eoi[i];
487 			if (ioapic->irq_eoi[i] == IOAPIC_SUCCESSIVE_IRQ_MAX_COUNT) {
488 				/*
489 				 * Real hardware does not deliver the interrupt
490 				 * immediately during eoi broadcast, and this
491 				 * lets a buggy guest make slow progress
492 				 * even if it does not correctly handle a
493 				 * level-triggered interrupt.  Emulate this
494 				 * behavior if we detect an interrupt storm.
495 				 */
496 				schedule_delayed_work(&ioapic->eoi_inject, HZ / 100);
497 				ioapic->irq_eoi[i] = 0;
498 				trace_kvm_ioapic_delayed_eoi_inj(ent->bits);
499 			} else {
500 				ioapic_service(ioapic, i, false);
501 			}
502 		} else {
503 			ioapic->irq_eoi[i] = 0;
504 		}
505 	}
506 }
507 
508 void kvm_ioapic_update_eoi(struct kvm_vcpu *vcpu, int vector, int trigger_mode)
509 {
510 	struct kvm_ioapic *ioapic = vcpu->kvm->arch.vioapic;
511 
512 	spin_lock(&ioapic->lock);
513 	__kvm_ioapic_update_eoi(vcpu, ioapic, vector, trigger_mode);
514 	spin_unlock(&ioapic->lock);
515 }
516 
517 static inline struct kvm_ioapic *to_ioapic(struct kvm_io_device *dev)
518 {
519 	return container_of(dev, struct kvm_ioapic, dev);
520 }
521 
522 static inline int ioapic_in_range(struct kvm_ioapic *ioapic, gpa_t addr)
523 {
524 	return ((addr >= ioapic->base_address &&
525 		 (addr < ioapic->base_address + IOAPIC_MEM_LENGTH)));
526 }
527 
528 static int ioapic_mmio_read(struct kvm_vcpu *vcpu, struct kvm_io_device *this,
529 				gpa_t addr, int len, void *val)
530 {
531 	struct kvm_ioapic *ioapic = to_ioapic(this);
532 	u32 result;
533 	if (!ioapic_in_range(ioapic, addr))
534 		return -EOPNOTSUPP;
535 
536 	ASSERT(!(addr & 0xf));	/* check alignment */
537 
538 	addr &= 0xff;
539 	spin_lock(&ioapic->lock);
540 	switch (addr) {
541 	case IOAPIC_REG_SELECT:
542 		result = ioapic->ioregsel;
543 		break;
544 
545 	case IOAPIC_REG_WINDOW:
546 		result = ioapic_read_indirect(ioapic, addr, len);
547 		break;
548 
549 	default:
550 		result = 0;
551 		break;
552 	}
553 	spin_unlock(&ioapic->lock);
554 
555 	switch (len) {
556 	case 8:
557 		*(u64 *) val = result;
558 		break;
559 	case 1:
560 	case 2:
561 	case 4:
562 		memcpy(val, (char *)&result, len);
563 		break;
564 	default:
565 		printk(KERN_WARNING "ioapic: wrong length %d\n", len);
566 	}
567 	return 0;
568 }
569 
570 static int ioapic_mmio_write(struct kvm_vcpu *vcpu, struct kvm_io_device *this,
571 				 gpa_t addr, int len, const void *val)
572 {
573 	struct kvm_ioapic *ioapic = to_ioapic(this);
574 	u32 data;
575 	if (!ioapic_in_range(ioapic, addr))
576 		return -EOPNOTSUPP;
577 
578 	ASSERT(!(addr & 0xf));	/* check alignment */
579 
580 	switch (len) {
581 	case 8:
582 	case 4:
583 		data = *(u32 *) val;
584 		break;
585 	case 2:
586 		data = *(u16 *) val;
587 		break;
588 	case 1:
589 		data = *(u8  *) val;
590 		break;
591 	default:
592 		printk(KERN_WARNING "ioapic: Unsupported size %d\n", len);
593 		return 0;
594 	}
595 
596 	addr &= 0xff;
597 	spin_lock(&ioapic->lock);
598 	switch (addr) {
599 	case IOAPIC_REG_SELECT:
600 		ioapic->ioregsel = data & 0xFF; /* 8-bit register */
601 		break;
602 
603 	case IOAPIC_REG_WINDOW:
604 		ioapic_write_indirect(ioapic, data);
605 		break;
606 
607 	default:
608 		break;
609 	}
610 	spin_unlock(&ioapic->lock);
611 	return 0;
612 }
613 
614 static void kvm_ioapic_reset(struct kvm_ioapic *ioapic)
615 {
616 	int i;
617 
618 	cancel_delayed_work_sync(&ioapic->eoi_inject);
619 	for (i = 0; i < IOAPIC_NUM_PINS; i++)
620 		ioapic->redirtbl[i].fields.mask = 1;
621 	ioapic->base_address = IOAPIC_DEFAULT_BASE_ADDRESS;
622 	ioapic->ioregsel = 0;
623 	ioapic->irr = 0;
624 	ioapic->irr_delivered = 0;
625 	ioapic->id = 0;
626 	memset(ioapic->irq_eoi, 0x00, sizeof(ioapic->irq_eoi));
627 	rtc_irq_eoi_tracking_reset(ioapic);
628 }
629 
630 static const struct kvm_io_device_ops ioapic_mmio_ops = {
631 	.read     = ioapic_mmio_read,
632 	.write    = ioapic_mmio_write,
633 };
634 
635 int kvm_ioapic_init(struct kvm *kvm)
636 {
637 	struct kvm_ioapic *ioapic;
638 	int ret;
639 
640 	ioapic = kzalloc(sizeof(struct kvm_ioapic), GFP_KERNEL_ACCOUNT);
641 	if (!ioapic)
642 		return -ENOMEM;
643 	spin_lock_init(&ioapic->lock);
644 	INIT_DELAYED_WORK(&ioapic->eoi_inject, kvm_ioapic_eoi_inject_work);
645 	kvm->arch.vioapic = ioapic;
646 	kvm_ioapic_reset(ioapic);
647 	kvm_iodevice_init(&ioapic->dev, &ioapic_mmio_ops);
648 	ioapic->kvm = kvm;
649 	mutex_lock(&kvm->slots_lock);
650 	ret = kvm_io_bus_register_dev(kvm, KVM_MMIO_BUS, ioapic->base_address,
651 				      IOAPIC_MEM_LENGTH, &ioapic->dev);
652 	mutex_unlock(&kvm->slots_lock);
653 	if (ret < 0) {
654 		kvm->arch.vioapic = NULL;
655 		kfree(ioapic);
656 	}
657 
658 	return ret;
659 }
660 
661 void kvm_ioapic_destroy(struct kvm *kvm)
662 {
663 	struct kvm_ioapic *ioapic = kvm->arch.vioapic;
664 
665 	if (!ioapic)
666 		return;
667 
668 	cancel_delayed_work_sync(&ioapic->eoi_inject);
669 	mutex_lock(&kvm->slots_lock);
670 	kvm_io_bus_unregister_dev(kvm, KVM_MMIO_BUS, &ioapic->dev);
671 	mutex_unlock(&kvm->slots_lock);
672 	kvm->arch.vioapic = NULL;
673 	kfree(ioapic);
674 }
675 
676 void kvm_get_ioapic(struct kvm *kvm, struct kvm_ioapic_state *state)
677 {
678 	struct kvm_ioapic *ioapic = kvm->arch.vioapic;
679 
680 	spin_lock(&ioapic->lock);
681 	memcpy(state, ioapic, sizeof(struct kvm_ioapic_state));
682 	state->irr &= ~ioapic->irr_delivered;
683 	spin_unlock(&ioapic->lock);
684 }
685 
686 void kvm_set_ioapic(struct kvm *kvm, struct kvm_ioapic_state *state)
687 {
688 	struct kvm_ioapic *ioapic = kvm->arch.vioapic;
689 
690 	spin_lock(&ioapic->lock);
691 	memcpy(ioapic, state, sizeof(struct kvm_ioapic_state));
692 	ioapic->irr = 0;
693 	ioapic->irr_delivered = 0;
694 	kvm_make_scan_ioapic_request(kvm);
695 	kvm_ioapic_inject_all(ioapic, state->irr);
696 	spin_unlock(&ioapic->lock);
697 }
698