xref: /openbmc/linux/arch/x86/kvm/ioapic.c (revision 3557b3fd)
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 #if 0
49 #define ioapic_debug(fmt,arg...) printk(KERN_WARNING fmt,##arg)
50 #else
51 #define ioapic_debug(fmt, arg...)
52 #endif
53 static int ioapic_service(struct kvm_ioapic *vioapic, int irq,
54 		bool line_status);
55 
56 static unsigned long ioapic_read_indirect(struct kvm_ioapic *ioapic,
57 					  unsigned long addr,
58 					  unsigned long length)
59 {
60 	unsigned long result = 0;
61 
62 	switch (ioapic->ioregsel) {
63 	case IOAPIC_REG_VERSION:
64 		result = ((((IOAPIC_NUM_PINS - 1) & 0xff) << 16)
65 			  | (IOAPIC_VERSION_ID & 0xff));
66 		break;
67 
68 	case IOAPIC_REG_APIC_ID:
69 	case IOAPIC_REG_ARB_ID:
70 		result = ((ioapic->id & 0xf) << 24);
71 		break;
72 
73 	default:
74 		{
75 			u32 redir_index = (ioapic->ioregsel - 0x10) >> 1;
76 			u64 redir_content;
77 
78 			if (redir_index < IOAPIC_NUM_PINS)
79 				redir_content =
80 					ioapic->redirtbl[redir_index].bits;
81 			else
82 				redir_content = ~0ULL;
83 
84 			result = (ioapic->ioregsel & 0x1) ?
85 			    (redir_content >> 32) & 0xffffffff :
86 			    redir_content & 0xffffffff;
87 			break;
88 		}
89 	}
90 
91 	return result;
92 }
93 
94 static void rtc_irq_eoi_tracking_reset(struct kvm_ioapic *ioapic)
95 {
96 	ioapic->rtc_status.pending_eoi = 0;
97 	bitmap_zero(ioapic->rtc_status.dest_map.map, KVM_MAX_VCPU_ID);
98 }
99 
100 static void kvm_rtc_eoi_tracking_restore_all(struct kvm_ioapic *ioapic);
101 
102 static void rtc_status_pending_eoi_check_valid(struct kvm_ioapic *ioapic)
103 {
104 	if (WARN_ON(ioapic->rtc_status.pending_eoi < 0))
105 		kvm_rtc_eoi_tracking_restore_all(ioapic);
106 }
107 
108 static void __rtc_irq_eoi_tracking_restore_one(struct kvm_vcpu *vcpu)
109 {
110 	bool new_val, old_val;
111 	struct kvm_ioapic *ioapic = vcpu->kvm->arch.vioapic;
112 	struct dest_map *dest_map = &ioapic->rtc_status.dest_map;
113 	union kvm_ioapic_redirect_entry *e;
114 
115 	e = &ioapic->redirtbl[RTC_GSI];
116 	if (!kvm_apic_match_dest(vcpu, NULL, 0,	e->fields.dest_id,
117 				e->fields.dest_mode))
118 		return;
119 
120 	new_val = kvm_apic_pending_eoi(vcpu, e->fields.vector);
121 	old_val = test_bit(vcpu->vcpu_id, dest_map->map);
122 
123 	if (new_val == old_val)
124 		return;
125 
126 	if (new_val) {
127 		__set_bit(vcpu->vcpu_id, dest_map->map);
128 		dest_map->vectors[vcpu->vcpu_id] = e->fields.vector;
129 		ioapic->rtc_status.pending_eoi++;
130 	} else {
131 		__clear_bit(vcpu->vcpu_id, dest_map->map);
132 		ioapic->rtc_status.pending_eoi--;
133 		rtc_status_pending_eoi_check_valid(ioapic);
134 	}
135 }
136 
137 void kvm_rtc_eoi_tracking_restore_one(struct kvm_vcpu *vcpu)
138 {
139 	struct kvm_ioapic *ioapic = vcpu->kvm->arch.vioapic;
140 
141 	spin_lock(&ioapic->lock);
142 	__rtc_irq_eoi_tracking_restore_one(vcpu);
143 	spin_unlock(&ioapic->lock);
144 }
145 
146 static void kvm_rtc_eoi_tracking_restore_all(struct kvm_ioapic *ioapic)
147 {
148 	struct kvm_vcpu *vcpu;
149 	int i;
150 
151 	if (RTC_GSI >= IOAPIC_NUM_PINS)
152 		return;
153 
154 	rtc_irq_eoi_tracking_reset(ioapic);
155 	kvm_for_each_vcpu(i, vcpu, ioapic->kvm)
156 	    __rtc_irq_eoi_tracking_restore_one(vcpu);
157 }
158 
159 static void rtc_irq_eoi(struct kvm_ioapic *ioapic, struct kvm_vcpu *vcpu)
160 {
161 	if (test_and_clear_bit(vcpu->vcpu_id,
162 			       ioapic->rtc_status.dest_map.map)) {
163 		--ioapic->rtc_status.pending_eoi;
164 		rtc_status_pending_eoi_check_valid(ioapic);
165 	}
166 }
167 
168 static bool rtc_irq_check_coalesced(struct kvm_ioapic *ioapic)
169 {
170 	if (ioapic->rtc_status.pending_eoi > 0)
171 		return true; /* coalesced */
172 
173 	return false;
174 }
175 
176 static int ioapic_set_irq(struct kvm_ioapic *ioapic, unsigned int irq,
177 		int irq_level, bool line_status)
178 {
179 	union kvm_ioapic_redirect_entry entry;
180 	u32 mask = 1 << irq;
181 	u32 old_irr;
182 	int edge, ret;
183 
184 	entry = ioapic->redirtbl[irq];
185 	edge = (entry.fields.trig_mode == IOAPIC_EDGE_TRIG);
186 
187 	if (!irq_level) {
188 		ioapic->irr &= ~mask;
189 		ret = 1;
190 		goto out;
191 	}
192 
193 	/*
194 	 * Return 0 for coalesced interrupts; for edge-triggered interrupts,
195 	 * this only happens if a previous edge has not been delivered due
196 	 * do masking.  For level interrupts, the remote_irr field tells
197 	 * us if the interrupt is waiting for an EOI.
198 	 *
199 	 * RTC is special: it is edge-triggered, but userspace likes to know
200 	 * if it has been already ack-ed via EOI because coalesced RTC
201 	 * interrupts lead to time drift in Windows guests.  So we track
202 	 * EOI manually for the RTC interrupt.
203 	 */
204 	if (irq == RTC_GSI && line_status &&
205 		rtc_irq_check_coalesced(ioapic)) {
206 		ret = 0;
207 		goto out;
208 	}
209 
210 	old_irr = ioapic->irr;
211 	ioapic->irr |= mask;
212 	if (edge) {
213 		ioapic->irr_delivered &= ~mask;
214 		if (old_irr == ioapic->irr) {
215 			ret = 0;
216 			goto out;
217 		}
218 	}
219 
220 	ret = ioapic_service(ioapic, irq, line_status);
221 
222 out:
223 	trace_kvm_ioapic_set_irq(entry.bits, irq, ret == 0);
224 	return ret;
225 }
226 
227 static void kvm_ioapic_inject_all(struct kvm_ioapic *ioapic, unsigned long irr)
228 {
229 	u32 idx;
230 
231 	rtc_irq_eoi_tracking_reset(ioapic);
232 	for_each_set_bit(idx, &irr, IOAPIC_NUM_PINS)
233 		ioapic_set_irq(ioapic, idx, 1, true);
234 
235 	kvm_rtc_eoi_tracking_restore_all(ioapic);
236 }
237 
238 
239 void kvm_ioapic_scan_entry(struct kvm_vcpu *vcpu, ulong *ioapic_handled_vectors)
240 {
241 	struct kvm_ioapic *ioapic = vcpu->kvm->arch.vioapic;
242 	struct dest_map *dest_map = &ioapic->rtc_status.dest_map;
243 	union kvm_ioapic_redirect_entry *e;
244 	int index;
245 
246 	spin_lock(&ioapic->lock);
247 
248 	/* Make sure we see any missing RTC EOI */
249 	if (test_bit(vcpu->vcpu_id, dest_map->map))
250 		__set_bit(dest_map->vectors[vcpu->vcpu_id],
251 			  ioapic_handled_vectors);
252 
253 	for (index = 0; index < IOAPIC_NUM_PINS; index++) {
254 		e = &ioapic->redirtbl[index];
255 		if (e->fields.trig_mode == IOAPIC_LEVEL_TRIG ||
256 		    kvm_irq_has_notifier(ioapic->kvm, KVM_IRQCHIP_IOAPIC, index) ||
257 		    index == RTC_GSI) {
258 			if (kvm_apic_match_dest(vcpu, NULL, 0,
259 			             e->fields.dest_id, e->fields.dest_mode) ||
260 			    kvm_apic_pending_eoi(vcpu, e->fields.vector))
261 				__set_bit(e->fields.vector,
262 					  ioapic_handled_vectors);
263 		}
264 	}
265 	spin_unlock(&ioapic->lock);
266 }
267 
268 void kvm_arch_post_irq_ack_notifier_list_update(struct kvm *kvm)
269 {
270 	if (!ioapic_in_kernel(kvm))
271 		return;
272 	kvm_make_scan_ioapic_request(kvm);
273 }
274 
275 static void ioapic_write_indirect(struct kvm_ioapic *ioapic, u32 val)
276 {
277 	unsigned index;
278 	bool mask_before, mask_after;
279 	int old_remote_irr, old_delivery_status;
280 	union kvm_ioapic_redirect_entry *e;
281 
282 	switch (ioapic->ioregsel) {
283 	case IOAPIC_REG_VERSION:
284 		/* Writes are ignored. */
285 		break;
286 
287 	case IOAPIC_REG_APIC_ID:
288 		ioapic->id = (val >> 24) & 0xf;
289 		break;
290 
291 	case IOAPIC_REG_ARB_ID:
292 		break;
293 
294 	default:
295 		index = (ioapic->ioregsel - 0x10) >> 1;
296 
297 		ioapic_debug("change redir index %x val %x\n", index, val);
298 		if (index >= IOAPIC_NUM_PINS)
299 			return;
300 		e = &ioapic->redirtbl[index];
301 		mask_before = e->fields.mask;
302 		/* Preserve read-only fields */
303 		old_remote_irr = e->fields.remote_irr;
304 		old_delivery_status = e->fields.delivery_status;
305 		if (ioapic->ioregsel & 1) {
306 			e->bits &= 0xffffffff;
307 			e->bits |= (u64) val << 32;
308 		} else {
309 			e->bits &= ~0xffffffffULL;
310 			e->bits |= (u32) val;
311 		}
312 		e->fields.remote_irr = old_remote_irr;
313 		e->fields.delivery_status = old_delivery_status;
314 
315 		/*
316 		 * Some OSes (Linux, Xen) assume that Remote IRR bit will
317 		 * be cleared by IOAPIC hardware when the entry is configured
318 		 * as edge-triggered. This behavior is used to simulate an
319 		 * explicit EOI on IOAPICs that don't have the EOI register.
320 		 */
321 		if (e->fields.trig_mode == IOAPIC_EDGE_TRIG)
322 			e->fields.remote_irr = 0;
323 
324 		mask_after = e->fields.mask;
325 		if (mask_before != mask_after)
326 			kvm_fire_mask_notifiers(ioapic->kvm, KVM_IRQCHIP_IOAPIC, index, mask_after);
327 		if (e->fields.trig_mode == IOAPIC_LEVEL_TRIG
328 		    && ioapic->irr & (1 << index))
329 			ioapic_service(ioapic, index, false);
330 		kvm_make_scan_ioapic_request(ioapic->kvm);
331 		break;
332 	}
333 }
334 
335 static int ioapic_service(struct kvm_ioapic *ioapic, int irq, bool line_status)
336 {
337 	union kvm_ioapic_redirect_entry *entry = &ioapic->redirtbl[irq];
338 	struct kvm_lapic_irq irqe;
339 	int ret;
340 
341 	if (entry->fields.mask ||
342 	    (entry->fields.trig_mode == IOAPIC_LEVEL_TRIG &&
343 	    entry->fields.remote_irr))
344 		return -1;
345 
346 	ioapic_debug("dest=%x dest_mode=%x delivery_mode=%x "
347 		     "vector=%x trig_mode=%x\n",
348 		     entry->fields.dest_id, entry->fields.dest_mode,
349 		     entry->fields.delivery_mode, entry->fields.vector,
350 		     entry->fields.trig_mode);
351 
352 	irqe.dest_id = entry->fields.dest_id;
353 	irqe.vector = entry->fields.vector;
354 	irqe.dest_mode = entry->fields.dest_mode;
355 	irqe.trig_mode = entry->fields.trig_mode;
356 	irqe.delivery_mode = entry->fields.delivery_mode << 8;
357 	irqe.level = 1;
358 	irqe.shorthand = 0;
359 	irqe.msi_redir_hint = false;
360 
361 	if (irqe.trig_mode == IOAPIC_EDGE_TRIG)
362 		ioapic->irr_delivered |= 1 << irq;
363 
364 	if (irq == RTC_GSI && line_status) {
365 		/*
366 		 * pending_eoi cannot ever become negative (see
367 		 * rtc_status_pending_eoi_check_valid) and the caller
368 		 * ensures that it is only called if it is >= zero, namely
369 		 * if rtc_irq_check_coalesced returns false).
370 		 */
371 		BUG_ON(ioapic->rtc_status.pending_eoi != 0);
372 		ret = kvm_irq_delivery_to_apic(ioapic->kvm, NULL, &irqe,
373 					       &ioapic->rtc_status.dest_map);
374 		ioapic->rtc_status.pending_eoi = (ret < 0 ? 0 : ret);
375 	} else
376 		ret = kvm_irq_delivery_to_apic(ioapic->kvm, NULL, &irqe, NULL);
377 
378 	if (ret && irqe.trig_mode == IOAPIC_LEVEL_TRIG)
379 		entry->fields.remote_irr = 1;
380 
381 	return ret;
382 }
383 
384 int kvm_ioapic_set_irq(struct kvm_ioapic *ioapic, int irq, int irq_source_id,
385 		       int level, bool line_status)
386 {
387 	int ret, irq_level;
388 
389 	BUG_ON(irq < 0 || irq >= IOAPIC_NUM_PINS);
390 
391 	spin_lock(&ioapic->lock);
392 	irq_level = __kvm_irq_line_state(&ioapic->irq_states[irq],
393 					 irq_source_id, level);
394 	ret = ioapic_set_irq(ioapic, irq, irq_level, line_status);
395 
396 	spin_unlock(&ioapic->lock);
397 
398 	return ret;
399 }
400 
401 void kvm_ioapic_clear_all(struct kvm_ioapic *ioapic, int irq_source_id)
402 {
403 	int i;
404 
405 	spin_lock(&ioapic->lock);
406 	for (i = 0; i < KVM_IOAPIC_NUM_PINS; i++)
407 		__clear_bit(irq_source_id, &ioapic->irq_states[i]);
408 	spin_unlock(&ioapic->lock);
409 }
410 
411 static void kvm_ioapic_eoi_inject_work(struct work_struct *work)
412 {
413 	int i;
414 	struct kvm_ioapic *ioapic = container_of(work, struct kvm_ioapic,
415 						 eoi_inject.work);
416 	spin_lock(&ioapic->lock);
417 	for (i = 0; i < IOAPIC_NUM_PINS; i++) {
418 		union kvm_ioapic_redirect_entry *ent = &ioapic->redirtbl[i];
419 
420 		if (ent->fields.trig_mode != IOAPIC_LEVEL_TRIG)
421 			continue;
422 
423 		if (ioapic->irr & (1 << i) && !ent->fields.remote_irr)
424 			ioapic_service(ioapic, i, false);
425 	}
426 	spin_unlock(&ioapic->lock);
427 }
428 
429 #define IOAPIC_SUCCESSIVE_IRQ_MAX_COUNT 10000
430 
431 static void __kvm_ioapic_update_eoi(struct kvm_vcpu *vcpu,
432 			struct kvm_ioapic *ioapic, int vector, int trigger_mode)
433 {
434 	struct dest_map *dest_map = &ioapic->rtc_status.dest_map;
435 	struct kvm_lapic *apic = vcpu->arch.apic;
436 	int i;
437 
438 	/* RTC special handling */
439 	if (test_bit(vcpu->vcpu_id, dest_map->map) &&
440 	    vector == dest_map->vectors[vcpu->vcpu_id])
441 		rtc_irq_eoi(ioapic, vcpu);
442 
443 	for (i = 0; i < IOAPIC_NUM_PINS; i++) {
444 		union kvm_ioapic_redirect_entry *ent = &ioapic->redirtbl[i];
445 
446 		if (ent->fields.vector != vector)
447 			continue;
448 
449 		/*
450 		 * We are dropping lock while calling ack notifiers because ack
451 		 * notifier callbacks for assigned devices call into IOAPIC
452 		 * recursively. Since remote_irr is cleared only after call
453 		 * to notifiers if the same vector will be delivered while lock
454 		 * is dropped it will be put into irr and will be delivered
455 		 * after ack notifier returns.
456 		 */
457 		spin_unlock(&ioapic->lock);
458 		kvm_notify_acked_irq(ioapic->kvm, KVM_IRQCHIP_IOAPIC, i);
459 		spin_lock(&ioapic->lock);
460 
461 		if (trigger_mode != IOAPIC_LEVEL_TRIG ||
462 		    kvm_lapic_get_reg(apic, APIC_SPIV) & APIC_SPIV_DIRECTED_EOI)
463 			continue;
464 
465 		ASSERT(ent->fields.trig_mode == IOAPIC_LEVEL_TRIG);
466 		ent->fields.remote_irr = 0;
467 		if (!ent->fields.mask && (ioapic->irr & (1 << i))) {
468 			++ioapic->irq_eoi[i];
469 			if (ioapic->irq_eoi[i] == IOAPIC_SUCCESSIVE_IRQ_MAX_COUNT) {
470 				/*
471 				 * Real hardware does not deliver the interrupt
472 				 * immediately during eoi broadcast, and this
473 				 * lets a buggy guest make slow progress
474 				 * even if it does not correctly handle a
475 				 * level-triggered interrupt.  Emulate this
476 				 * behavior if we detect an interrupt storm.
477 				 */
478 				schedule_delayed_work(&ioapic->eoi_inject, HZ / 100);
479 				ioapic->irq_eoi[i] = 0;
480 				trace_kvm_ioapic_delayed_eoi_inj(ent->bits);
481 			} else {
482 				ioapic_service(ioapic, i, false);
483 			}
484 		} else {
485 			ioapic->irq_eoi[i] = 0;
486 		}
487 	}
488 }
489 
490 void kvm_ioapic_update_eoi(struct kvm_vcpu *vcpu, int vector, int trigger_mode)
491 {
492 	struct kvm_ioapic *ioapic = vcpu->kvm->arch.vioapic;
493 
494 	spin_lock(&ioapic->lock);
495 	__kvm_ioapic_update_eoi(vcpu, ioapic, vector, trigger_mode);
496 	spin_unlock(&ioapic->lock);
497 }
498 
499 static inline struct kvm_ioapic *to_ioapic(struct kvm_io_device *dev)
500 {
501 	return container_of(dev, struct kvm_ioapic, dev);
502 }
503 
504 static inline int ioapic_in_range(struct kvm_ioapic *ioapic, gpa_t addr)
505 {
506 	return ((addr >= ioapic->base_address &&
507 		 (addr < ioapic->base_address + IOAPIC_MEM_LENGTH)));
508 }
509 
510 static int ioapic_mmio_read(struct kvm_vcpu *vcpu, struct kvm_io_device *this,
511 				gpa_t addr, int len, void *val)
512 {
513 	struct kvm_ioapic *ioapic = to_ioapic(this);
514 	u32 result;
515 	if (!ioapic_in_range(ioapic, addr))
516 		return -EOPNOTSUPP;
517 
518 	ioapic_debug("addr %lx\n", (unsigned long)addr);
519 	ASSERT(!(addr & 0xf));	/* check alignment */
520 
521 	addr &= 0xff;
522 	spin_lock(&ioapic->lock);
523 	switch (addr) {
524 	case IOAPIC_REG_SELECT:
525 		result = ioapic->ioregsel;
526 		break;
527 
528 	case IOAPIC_REG_WINDOW:
529 		result = ioapic_read_indirect(ioapic, addr, len);
530 		break;
531 
532 	default:
533 		result = 0;
534 		break;
535 	}
536 	spin_unlock(&ioapic->lock);
537 
538 	switch (len) {
539 	case 8:
540 		*(u64 *) val = result;
541 		break;
542 	case 1:
543 	case 2:
544 	case 4:
545 		memcpy(val, (char *)&result, len);
546 		break;
547 	default:
548 		printk(KERN_WARNING "ioapic: wrong length %d\n", len);
549 	}
550 	return 0;
551 }
552 
553 static int ioapic_mmio_write(struct kvm_vcpu *vcpu, struct kvm_io_device *this,
554 				 gpa_t addr, int len, const void *val)
555 {
556 	struct kvm_ioapic *ioapic = to_ioapic(this);
557 	u32 data;
558 	if (!ioapic_in_range(ioapic, addr))
559 		return -EOPNOTSUPP;
560 
561 	ioapic_debug("ioapic_mmio_write addr=%p len=%d val=%p\n",
562 		     (void*)addr, len, val);
563 	ASSERT(!(addr & 0xf));	/* check alignment */
564 
565 	switch (len) {
566 	case 8:
567 	case 4:
568 		data = *(u32 *) val;
569 		break;
570 	case 2:
571 		data = *(u16 *) val;
572 		break;
573 	case 1:
574 		data = *(u8  *) val;
575 		break;
576 	default:
577 		printk(KERN_WARNING "ioapic: Unsupported size %d\n", len);
578 		return 0;
579 	}
580 
581 	addr &= 0xff;
582 	spin_lock(&ioapic->lock);
583 	switch (addr) {
584 	case IOAPIC_REG_SELECT:
585 		ioapic->ioregsel = data & 0xFF; /* 8-bit register */
586 		break;
587 
588 	case IOAPIC_REG_WINDOW:
589 		ioapic_write_indirect(ioapic, data);
590 		break;
591 
592 	default:
593 		break;
594 	}
595 	spin_unlock(&ioapic->lock);
596 	return 0;
597 }
598 
599 static void kvm_ioapic_reset(struct kvm_ioapic *ioapic)
600 {
601 	int i;
602 
603 	cancel_delayed_work_sync(&ioapic->eoi_inject);
604 	for (i = 0; i < IOAPIC_NUM_PINS; i++)
605 		ioapic->redirtbl[i].fields.mask = 1;
606 	ioapic->base_address = IOAPIC_DEFAULT_BASE_ADDRESS;
607 	ioapic->ioregsel = 0;
608 	ioapic->irr = 0;
609 	ioapic->irr_delivered = 0;
610 	ioapic->id = 0;
611 	memset(ioapic->irq_eoi, 0x00, sizeof(ioapic->irq_eoi));
612 	rtc_irq_eoi_tracking_reset(ioapic);
613 }
614 
615 static const struct kvm_io_device_ops ioapic_mmio_ops = {
616 	.read     = ioapic_mmio_read,
617 	.write    = ioapic_mmio_write,
618 };
619 
620 int kvm_ioapic_init(struct kvm *kvm)
621 {
622 	struct kvm_ioapic *ioapic;
623 	int ret;
624 
625 	ioapic = kzalloc(sizeof(struct kvm_ioapic), GFP_KERNEL_ACCOUNT);
626 	if (!ioapic)
627 		return -ENOMEM;
628 	spin_lock_init(&ioapic->lock);
629 	INIT_DELAYED_WORK(&ioapic->eoi_inject, kvm_ioapic_eoi_inject_work);
630 	kvm->arch.vioapic = ioapic;
631 	kvm_ioapic_reset(ioapic);
632 	kvm_iodevice_init(&ioapic->dev, &ioapic_mmio_ops);
633 	ioapic->kvm = kvm;
634 	mutex_lock(&kvm->slots_lock);
635 	ret = kvm_io_bus_register_dev(kvm, KVM_MMIO_BUS, ioapic->base_address,
636 				      IOAPIC_MEM_LENGTH, &ioapic->dev);
637 	mutex_unlock(&kvm->slots_lock);
638 	if (ret < 0) {
639 		kvm->arch.vioapic = NULL;
640 		kfree(ioapic);
641 	}
642 
643 	return ret;
644 }
645 
646 void kvm_ioapic_destroy(struct kvm *kvm)
647 {
648 	struct kvm_ioapic *ioapic = kvm->arch.vioapic;
649 
650 	if (!ioapic)
651 		return;
652 
653 	cancel_delayed_work_sync(&ioapic->eoi_inject);
654 	mutex_lock(&kvm->slots_lock);
655 	kvm_io_bus_unregister_dev(kvm, KVM_MMIO_BUS, &ioapic->dev);
656 	mutex_unlock(&kvm->slots_lock);
657 	kvm->arch.vioapic = NULL;
658 	kfree(ioapic);
659 }
660 
661 void kvm_get_ioapic(struct kvm *kvm, struct kvm_ioapic_state *state)
662 {
663 	struct kvm_ioapic *ioapic = kvm->arch.vioapic;
664 
665 	spin_lock(&ioapic->lock);
666 	memcpy(state, ioapic, sizeof(struct kvm_ioapic_state));
667 	state->irr &= ~ioapic->irr_delivered;
668 	spin_unlock(&ioapic->lock);
669 }
670 
671 void kvm_set_ioapic(struct kvm *kvm, struct kvm_ioapic_state *state)
672 {
673 	struct kvm_ioapic *ioapic = kvm->arch.vioapic;
674 
675 	spin_lock(&ioapic->lock);
676 	memcpy(ioapic, state, sizeof(struct kvm_ioapic_state));
677 	ioapic->irr = 0;
678 	ioapic->irr_delivered = 0;
679 	kvm_make_scan_ioapic_request(kvm);
680 	kvm_ioapic_inject_all(ioapic, state->irr);
681 	spin_unlock(&ioapic->lock);
682 }
683