xref: /openbmc/linux/arch/x86/kvm/ioapic.c (revision 52cdded0)
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 <linux/nospec.h>
40 #include <asm/processor.h>
41 #include <asm/page.h>
42 #include <asm/current.h>
43 #include <trace/events/kvm.h>
44 
45 #include "ioapic.h"
46 #include "lapic.h"
47 #include "irq.h"
48 
49 static int ioapic_service(struct kvm_ioapic *vioapic, int irq,
50 		bool line_status);
51 
52 static void kvm_ioapic_update_eoi_one(struct kvm_vcpu *vcpu,
53 				      struct kvm_ioapic *ioapic,
54 				      int trigger_mode,
55 				      int pin);
56 
57 static unsigned long ioapic_read_indirect(struct kvm_ioapic *ioapic,
58 					  unsigned long addr,
59 					  unsigned long length)
60 {
61 	unsigned long result = 0;
62 
63 	switch (ioapic->ioregsel) {
64 	case IOAPIC_REG_VERSION:
65 		result = ((((IOAPIC_NUM_PINS - 1) & 0xff) << 16)
66 			  | (IOAPIC_VERSION_ID & 0xff));
67 		break;
68 
69 	case IOAPIC_REG_APIC_ID:
70 	case IOAPIC_REG_ARB_ID:
71 		result = ((ioapic->id & 0xf) << 24);
72 		break;
73 
74 	default:
75 		{
76 			u32 redir_index = (ioapic->ioregsel - 0x10) >> 1;
77 			u64 redir_content = ~0ULL;
78 
79 			if (redir_index < IOAPIC_NUM_PINS) {
80 				u32 index = array_index_nospec(
81 					redir_index, IOAPIC_NUM_PINS);
82 
83 				redir_content = ioapic->redirtbl[index].bits;
84 			}
85 
86 			result = (ioapic->ioregsel & 0x1) ?
87 			    (redir_content >> 32) & 0xffffffff :
88 			    redir_content & 0xffffffff;
89 			break;
90 		}
91 	}
92 
93 	return result;
94 }
95 
96 static void rtc_irq_eoi_tracking_reset(struct kvm_ioapic *ioapic)
97 {
98 	ioapic->rtc_status.pending_eoi = 0;
99 	bitmap_zero(ioapic->rtc_status.dest_map.map, KVM_MAX_VCPU_ID);
100 }
101 
102 static void kvm_rtc_eoi_tracking_restore_all(struct kvm_ioapic *ioapic);
103 
104 static void rtc_status_pending_eoi_check_valid(struct kvm_ioapic *ioapic)
105 {
106 	if (WARN_ON(ioapic->rtc_status.pending_eoi < 0))
107 		kvm_rtc_eoi_tracking_restore_all(ioapic);
108 }
109 
110 static void __rtc_irq_eoi_tracking_restore_one(struct kvm_vcpu *vcpu)
111 {
112 	bool new_val, old_val;
113 	struct kvm_ioapic *ioapic = vcpu->kvm->arch.vioapic;
114 	struct dest_map *dest_map = &ioapic->rtc_status.dest_map;
115 	union kvm_ioapic_redirect_entry *e;
116 
117 	e = &ioapic->redirtbl[RTC_GSI];
118 	if (!kvm_apic_match_dest(vcpu, NULL, APIC_DEST_NOSHORT,
119 				 e->fields.dest_id,
120 				 kvm_lapic_irq_dest_mode(!!e->fields.dest_mode)))
121 		return;
122 
123 	new_val = kvm_apic_pending_eoi(vcpu, e->fields.vector);
124 	old_val = test_bit(vcpu->vcpu_id, dest_map->map);
125 
126 	if (new_val == old_val)
127 		return;
128 
129 	if (new_val) {
130 		__set_bit(vcpu->vcpu_id, dest_map->map);
131 		dest_map->vectors[vcpu->vcpu_id] = e->fields.vector;
132 		ioapic->rtc_status.pending_eoi++;
133 	} else {
134 		__clear_bit(vcpu->vcpu_id, dest_map->map);
135 		ioapic->rtc_status.pending_eoi--;
136 		rtc_status_pending_eoi_check_valid(ioapic);
137 	}
138 }
139 
140 void kvm_rtc_eoi_tracking_restore_one(struct kvm_vcpu *vcpu)
141 {
142 	struct kvm_ioapic *ioapic = vcpu->kvm->arch.vioapic;
143 
144 	spin_lock(&ioapic->lock);
145 	__rtc_irq_eoi_tracking_restore_one(vcpu);
146 	spin_unlock(&ioapic->lock);
147 }
148 
149 static void kvm_rtc_eoi_tracking_restore_all(struct kvm_ioapic *ioapic)
150 {
151 	struct kvm_vcpu *vcpu;
152 	int i;
153 
154 	if (RTC_GSI >= IOAPIC_NUM_PINS)
155 		return;
156 
157 	rtc_irq_eoi_tracking_reset(ioapic);
158 	kvm_for_each_vcpu(i, vcpu, ioapic->kvm)
159 	    __rtc_irq_eoi_tracking_restore_one(vcpu);
160 }
161 
162 static void rtc_irq_eoi(struct kvm_ioapic *ioapic, struct kvm_vcpu *vcpu,
163 			int vector)
164 {
165 	struct dest_map *dest_map = &ioapic->rtc_status.dest_map;
166 
167 	/* RTC special handling */
168 	if (test_bit(vcpu->vcpu_id, dest_map->map) &&
169 	    (vector == dest_map->vectors[vcpu->vcpu_id]) &&
170 	    (test_and_clear_bit(vcpu->vcpu_id,
171 				ioapic->rtc_status.dest_map.map))) {
172 		--ioapic->rtc_status.pending_eoi;
173 		rtc_status_pending_eoi_check_valid(ioapic);
174 	}
175 }
176 
177 static bool rtc_irq_check_coalesced(struct kvm_ioapic *ioapic)
178 {
179 	if (ioapic->rtc_status.pending_eoi > 0)
180 		return true; /* coalesced */
181 
182 	return false;
183 }
184 
185 static void ioapic_lazy_update_eoi(struct kvm_ioapic *ioapic, int irq)
186 {
187 	int i;
188 	struct kvm_vcpu *vcpu;
189 	union kvm_ioapic_redirect_entry *entry = &ioapic->redirtbl[irq];
190 
191 	kvm_for_each_vcpu(i, vcpu, ioapic->kvm) {
192 		if (!kvm_apic_match_dest(vcpu, NULL, APIC_DEST_NOSHORT,
193 					 entry->fields.dest_id,
194 					 entry->fields.dest_mode) ||
195 		    kvm_apic_pending_eoi(vcpu, entry->fields.vector))
196 			continue;
197 
198 		/*
199 		 * If no longer has pending EOI in LAPICs, update
200 		 * EOI for this vetor.
201 		 */
202 		rtc_irq_eoi(ioapic, vcpu, entry->fields.vector);
203 		kvm_ioapic_update_eoi_one(vcpu, ioapic,
204 					  entry->fields.trig_mode,
205 					  irq);
206 		break;
207 	}
208 }
209 
210 static int ioapic_set_irq(struct kvm_ioapic *ioapic, unsigned int irq,
211 		int irq_level, bool line_status)
212 {
213 	union kvm_ioapic_redirect_entry entry;
214 	u32 mask = 1 << irq;
215 	u32 old_irr;
216 	int edge, ret;
217 
218 	entry = ioapic->redirtbl[irq];
219 	edge = (entry.fields.trig_mode == IOAPIC_EDGE_TRIG);
220 
221 	if (!irq_level) {
222 		ioapic->irr &= ~mask;
223 		ret = 1;
224 		goto out;
225 	}
226 
227 	/*
228 	 * AMD SVM AVIC accelerate EOI write iff the interrupt is edge
229 	 * triggered, in which case the in-kernel IOAPIC will not be able
230 	 * to receive the EOI.  In this case, we do a lazy update of the
231 	 * pending EOI when trying to set IOAPIC irq.
232 	 */
233 	if (edge && kvm_apicv_activated(ioapic->kvm))
234 		ioapic_lazy_update_eoi(ioapic, irq);
235 
236 	/*
237 	 * Return 0 for coalesced interrupts; for edge-triggered interrupts,
238 	 * this only happens if a previous edge has not been delivered due
239 	 * to masking.  For level interrupts, the remote_irr field tells
240 	 * us if the interrupt is waiting for an EOI.
241 	 *
242 	 * RTC is special: it is edge-triggered, but userspace likes to know
243 	 * if it has been already ack-ed via EOI because coalesced RTC
244 	 * interrupts lead to time drift in Windows guests.  So we track
245 	 * EOI manually for the RTC interrupt.
246 	 */
247 	if (irq == RTC_GSI && line_status &&
248 		rtc_irq_check_coalesced(ioapic)) {
249 		ret = 0;
250 		goto out;
251 	}
252 
253 	old_irr = ioapic->irr;
254 	ioapic->irr |= mask;
255 	if (edge) {
256 		ioapic->irr_delivered &= ~mask;
257 		if (old_irr == ioapic->irr) {
258 			ret = 0;
259 			goto out;
260 		}
261 	}
262 
263 	ret = ioapic_service(ioapic, irq, line_status);
264 
265 out:
266 	trace_kvm_ioapic_set_irq(entry.bits, irq, ret == 0);
267 	return ret;
268 }
269 
270 static void kvm_ioapic_inject_all(struct kvm_ioapic *ioapic, unsigned long irr)
271 {
272 	u32 idx;
273 
274 	rtc_irq_eoi_tracking_reset(ioapic);
275 	for_each_set_bit(idx, &irr, IOAPIC_NUM_PINS)
276 		ioapic_set_irq(ioapic, idx, 1, true);
277 
278 	kvm_rtc_eoi_tracking_restore_all(ioapic);
279 }
280 
281 
282 void kvm_ioapic_scan_entry(struct kvm_vcpu *vcpu, ulong *ioapic_handled_vectors)
283 {
284 	struct kvm_ioapic *ioapic = vcpu->kvm->arch.vioapic;
285 	struct dest_map *dest_map = &ioapic->rtc_status.dest_map;
286 	union kvm_ioapic_redirect_entry *e;
287 	int index;
288 
289 	spin_lock(&ioapic->lock);
290 
291 	/* Make sure we see any missing RTC EOI */
292 	if (test_bit(vcpu->vcpu_id, dest_map->map))
293 		__set_bit(dest_map->vectors[vcpu->vcpu_id],
294 			  ioapic_handled_vectors);
295 
296 	for (index = 0; index < IOAPIC_NUM_PINS; index++) {
297 		e = &ioapic->redirtbl[index];
298 		if (e->fields.trig_mode == IOAPIC_LEVEL_TRIG ||
299 		    kvm_irq_has_notifier(ioapic->kvm, KVM_IRQCHIP_IOAPIC, index) ||
300 		    index == RTC_GSI) {
301 			u16 dm = kvm_lapic_irq_dest_mode(!!e->fields.dest_mode);
302 
303 			if (kvm_apic_match_dest(vcpu, NULL, APIC_DEST_NOSHORT,
304 						e->fields.dest_id, dm) ||
305 			    kvm_apic_pending_eoi(vcpu, e->fields.vector))
306 				__set_bit(e->fields.vector,
307 					  ioapic_handled_vectors);
308 		}
309 	}
310 	spin_unlock(&ioapic->lock);
311 }
312 
313 void kvm_arch_post_irq_ack_notifier_list_update(struct kvm *kvm)
314 {
315 	if (!ioapic_in_kernel(kvm))
316 		return;
317 	kvm_make_scan_ioapic_request(kvm);
318 }
319 
320 static void ioapic_write_indirect(struct kvm_ioapic *ioapic, u32 val)
321 {
322 	unsigned index;
323 	bool mask_before, mask_after;
324 	union kvm_ioapic_redirect_entry *e;
325 	unsigned long vcpu_bitmap;
326 	int old_remote_irr, old_delivery_status, old_dest_id, old_dest_mode;
327 
328 	switch (ioapic->ioregsel) {
329 	case IOAPIC_REG_VERSION:
330 		/* Writes are ignored. */
331 		break;
332 
333 	case IOAPIC_REG_APIC_ID:
334 		ioapic->id = (val >> 24) & 0xf;
335 		break;
336 
337 	case IOAPIC_REG_ARB_ID:
338 		break;
339 
340 	default:
341 		index = (ioapic->ioregsel - 0x10) >> 1;
342 
343 		if (index >= IOAPIC_NUM_PINS)
344 			return;
345 		index = array_index_nospec(index, IOAPIC_NUM_PINS);
346 		e = &ioapic->redirtbl[index];
347 		mask_before = e->fields.mask;
348 		/* Preserve read-only fields */
349 		old_remote_irr = e->fields.remote_irr;
350 		old_delivery_status = e->fields.delivery_status;
351 		old_dest_id = e->fields.dest_id;
352 		old_dest_mode = e->fields.dest_mode;
353 		if (ioapic->ioregsel & 1) {
354 			e->bits &= 0xffffffff;
355 			e->bits |= (u64) val << 32;
356 		} else {
357 			e->bits &= ~0xffffffffULL;
358 			e->bits |= (u32) val;
359 		}
360 		e->fields.remote_irr = old_remote_irr;
361 		e->fields.delivery_status = old_delivery_status;
362 
363 		/*
364 		 * Some OSes (Linux, Xen) assume that Remote IRR bit will
365 		 * be cleared by IOAPIC hardware when the entry is configured
366 		 * as edge-triggered. This behavior is used to simulate an
367 		 * explicit EOI on IOAPICs that don't have the EOI register.
368 		 */
369 		if (e->fields.trig_mode == IOAPIC_EDGE_TRIG)
370 			e->fields.remote_irr = 0;
371 
372 		mask_after = e->fields.mask;
373 		if (mask_before != mask_after)
374 			kvm_fire_mask_notifiers(ioapic->kvm, KVM_IRQCHIP_IOAPIC, index, mask_after);
375 		if (e->fields.trig_mode == IOAPIC_LEVEL_TRIG
376 		    && ioapic->irr & (1 << index))
377 			ioapic_service(ioapic, index, false);
378 		if (e->fields.delivery_mode == APIC_DM_FIXED) {
379 			struct kvm_lapic_irq irq;
380 
381 			irq.vector = e->fields.vector;
382 			irq.delivery_mode = e->fields.delivery_mode << 8;
383 			irq.dest_mode =
384 			    kvm_lapic_irq_dest_mode(!!e->fields.dest_mode);
385 			irq.level = false;
386 			irq.trig_mode = e->fields.trig_mode;
387 			irq.shorthand = APIC_DEST_NOSHORT;
388 			irq.dest_id = e->fields.dest_id;
389 			irq.msi_redir_hint = false;
390 			bitmap_zero(&vcpu_bitmap, 16);
391 			kvm_bitmap_or_dest_vcpus(ioapic->kvm, &irq,
392 						 &vcpu_bitmap);
393 			if (old_dest_mode != e->fields.dest_mode ||
394 			    old_dest_id != e->fields.dest_id) {
395 				/*
396 				 * Update vcpu_bitmap with vcpus specified in
397 				 * the previous request as well. This is done to
398 				 * keep ioapic_handled_vectors synchronized.
399 				 */
400 				irq.dest_id = old_dest_id;
401 				irq.dest_mode =
402 				    kvm_lapic_irq_dest_mode(
403 					!!e->fields.dest_mode);
404 				kvm_bitmap_or_dest_vcpus(ioapic->kvm, &irq,
405 							 &vcpu_bitmap);
406 			}
407 			kvm_make_scan_ioapic_request_mask(ioapic->kvm,
408 							  &vcpu_bitmap);
409 		} else {
410 			kvm_make_scan_ioapic_request(ioapic->kvm);
411 		}
412 		break;
413 	}
414 }
415 
416 static int ioapic_service(struct kvm_ioapic *ioapic, int irq, bool line_status)
417 {
418 	union kvm_ioapic_redirect_entry *entry = &ioapic->redirtbl[irq];
419 	struct kvm_lapic_irq irqe;
420 	int ret;
421 
422 	if (entry->fields.mask ||
423 	    (entry->fields.trig_mode == IOAPIC_LEVEL_TRIG &&
424 	    entry->fields.remote_irr))
425 		return -1;
426 
427 	irqe.dest_id = entry->fields.dest_id;
428 	irqe.vector = entry->fields.vector;
429 	irqe.dest_mode = kvm_lapic_irq_dest_mode(!!entry->fields.dest_mode);
430 	irqe.trig_mode = entry->fields.trig_mode;
431 	irqe.delivery_mode = entry->fields.delivery_mode << 8;
432 	irqe.level = 1;
433 	irqe.shorthand = APIC_DEST_NOSHORT;
434 	irqe.msi_redir_hint = false;
435 
436 	if (irqe.trig_mode == IOAPIC_EDGE_TRIG)
437 		ioapic->irr_delivered |= 1 << irq;
438 
439 	if (irq == RTC_GSI && line_status) {
440 		/*
441 		 * pending_eoi cannot ever become negative (see
442 		 * rtc_status_pending_eoi_check_valid) and the caller
443 		 * ensures that it is only called if it is >= zero, namely
444 		 * if rtc_irq_check_coalesced returns false).
445 		 */
446 		BUG_ON(ioapic->rtc_status.pending_eoi != 0);
447 		ret = kvm_irq_delivery_to_apic(ioapic->kvm, NULL, &irqe,
448 					       &ioapic->rtc_status.dest_map);
449 		ioapic->rtc_status.pending_eoi = (ret < 0 ? 0 : ret);
450 	} else
451 		ret = kvm_irq_delivery_to_apic(ioapic->kvm, NULL, &irqe, NULL);
452 
453 	if (ret && irqe.trig_mode == IOAPIC_LEVEL_TRIG)
454 		entry->fields.remote_irr = 1;
455 
456 	return ret;
457 }
458 
459 int kvm_ioapic_set_irq(struct kvm_ioapic *ioapic, int irq, int irq_source_id,
460 		       int level, bool line_status)
461 {
462 	int ret, irq_level;
463 
464 	BUG_ON(irq < 0 || irq >= IOAPIC_NUM_PINS);
465 
466 	spin_lock(&ioapic->lock);
467 	irq_level = __kvm_irq_line_state(&ioapic->irq_states[irq],
468 					 irq_source_id, level);
469 	ret = ioapic_set_irq(ioapic, irq, irq_level, line_status);
470 
471 	spin_unlock(&ioapic->lock);
472 
473 	return ret;
474 }
475 
476 void kvm_ioapic_clear_all(struct kvm_ioapic *ioapic, int irq_source_id)
477 {
478 	int i;
479 
480 	spin_lock(&ioapic->lock);
481 	for (i = 0; i < KVM_IOAPIC_NUM_PINS; i++)
482 		__clear_bit(irq_source_id, &ioapic->irq_states[i]);
483 	spin_unlock(&ioapic->lock);
484 }
485 
486 static void kvm_ioapic_eoi_inject_work(struct work_struct *work)
487 {
488 	int i;
489 	struct kvm_ioapic *ioapic = container_of(work, struct kvm_ioapic,
490 						 eoi_inject.work);
491 	spin_lock(&ioapic->lock);
492 	for (i = 0; i < IOAPIC_NUM_PINS; i++) {
493 		union kvm_ioapic_redirect_entry *ent = &ioapic->redirtbl[i];
494 
495 		if (ent->fields.trig_mode != IOAPIC_LEVEL_TRIG)
496 			continue;
497 
498 		if (ioapic->irr & (1 << i) && !ent->fields.remote_irr)
499 			ioapic_service(ioapic, i, false);
500 	}
501 	spin_unlock(&ioapic->lock);
502 }
503 
504 #define IOAPIC_SUCCESSIVE_IRQ_MAX_COUNT 10000
505 static void kvm_ioapic_update_eoi_one(struct kvm_vcpu *vcpu,
506 				      struct kvm_ioapic *ioapic,
507 				      int trigger_mode,
508 				      int pin)
509 {
510 	struct kvm_lapic *apic = vcpu->arch.apic;
511 	union kvm_ioapic_redirect_entry *ent = &ioapic->redirtbl[pin];
512 
513 	/*
514 	 * We are dropping lock while calling ack notifiers because ack
515 	 * notifier callbacks for assigned devices call into IOAPIC
516 	 * recursively. Since remote_irr is cleared only after call
517 	 * to notifiers if the same vector will be delivered while lock
518 	 * is dropped it will be put into irr and will be delivered
519 	 * after ack notifier returns.
520 	 */
521 	spin_unlock(&ioapic->lock);
522 	kvm_notify_acked_irq(ioapic->kvm, KVM_IRQCHIP_IOAPIC, pin);
523 	spin_lock(&ioapic->lock);
524 
525 	if (trigger_mode != IOAPIC_LEVEL_TRIG ||
526 	    kvm_lapic_get_reg(apic, APIC_SPIV) & APIC_SPIV_DIRECTED_EOI)
527 		return;
528 
529 	ASSERT(ent->fields.trig_mode == IOAPIC_LEVEL_TRIG);
530 	ent->fields.remote_irr = 0;
531 	if (!ent->fields.mask && (ioapic->irr & (1 << pin))) {
532 		++ioapic->irq_eoi[pin];
533 		if (ioapic->irq_eoi[pin] == IOAPIC_SUCCESSIVE_IRQ_MAX_COUNT) {
534 			/*
535 			 * Real hardware does not deliver the interrupt
536 			 * immediately during eoi broadcast, and this
537 			 * lets a buggy guest make slow progress
538 			 * even if it does not correctly handle a
539 			 * level-triggered interrupt.  Emulate this
540 			 * behavior if we detect an interrupt storm.
541 			 */
542 			schedule_delayed_work(&ioapic->eoi_inject, HZ / 100);
543 			ioapic->irq_eoi[pin] = 0;
544 			trace_kvm_ioapic_delayed_eoi_inj(ent->bits);
545 		} else {
546 			ioapic_service(ioapic, pin, false);
547 		}
548 	} else {
549 		ioapic->irq_eoi[pin] = 0;
550 	}
551 }
552 
553 void kvm_ioapic_update_eoi(struct kvm_vcpu *vcpu, int vector, int trigger_mode)
554 {
555 	int i;
556 	struct kvm_ioapic *ioapic = vcpu->kvm->arch.vioapic;
557 
558 	spin_lock(&ioapic->lock);
559 	rtc_irq_eoi(ioapic, vcpu, vector);
560 	for (i = 0; i < IOAPIC_NUM_PINS; i++) {
561 		union kvm_ioapic_redirect_entry *ent = &ioapic->redirtbl[i];
562 
563 		if (ent->fields.vector != vector)
564 			continue;
565 		kvm_ioapic_update_eoi_one(vcpu, ioapic, trigger_mode, i);
566 	}
567 	spin_unlock(&ioapic->lock);
568 }
569 
570 static inline struct kvm_ioapic *to_ioapic(struct kvm_io_device *dev)
571 {
572 	return container_of(dev, struct kvm_ioapic, dev);
573 }
574 
575 static inline int ioapic_in_range(struct kvm_ioapic *ioapic, gpa_t addr)
576 {
577 	return ((addr >= ioapic->base_address &&
578 		 (addr < ioapic->base_address + IOAPIC_MEM_LENGTH)));
579 }
580 
581 static int ioapic_mmio_read(struct kvm_vcpu *vcpu, struct kvm_io_device *this,
582 				gpa_t addr, int len, void *val)
583 {
584 	struct kvm_ioapic *ioapic = to_ioapic(this);
585 	u32 result;
586 	if (!ioapic_in_range(ioapic, addr))
587 		return -EOPNOTSUPP;
588 
589 	ASSERT(!(addr & 0xf));	/* check alignment */
590 
591 	addr &= 0xff;
592 	spin_lock(&ioapic->lock);
593 	switch (addr) {
594 	case IOAPIC_REG_SELECT:
595 		result = ioapic->ioregsel;
596 		break;
597 
598 	case IOAPIC_REG_WINDOW:
599 		result = ioapic_read_indirect(ioapic, addr, len);
600 		break;
601 
602 	default:
603 		result = 0;
604 		break;
605 	}
606 	spin_unlock(&ioapic->lock);
607 
608 	switch (len) {
609 	case 8:
610 		*(u64 *) val = result;
611 		break;
612 	case 1:
613 	case 2:
614 	case 4:
615 		memcpy(val, (char *)&result, len);
616 		break;
617 	default:
618 		printk(KERN_WARNING "ioapic: wrong length %d\n", len);
619 	}
620 	return 0;
621 }
622 
623 static int ioapic_mmio_write(struct kvm_vcpu *vcpu, struct kvm_io_device *this,
624 				 gpa_t addr, int len, const void *val)
625 {
626 	struct kvm_ioapic *ioapic = to_ioapic(this);
627 	u32 data;
628 	if (!ioapic_in_range(ioapic, addr))
629 		return -EOPNOTSUPP;
630 
631 	ASSERT(!(addr & 0xf));	/* check alignment */
632 
633 	switch (len) {
634 	case 8:
635 	case 4:
636 		data = *(u32 *) val;
637 		break;
638 	case 2:
639 		data = *(u16 *) val;
640 		break;
641 	case 1:
642 		data = *(u8  *) val;
643 		break;
644 	default:
645 		printk(KERN_WARNING "ioapic: Unsupported size %d\n", len);
646 		return 0;
647 	}
648 
649 	addr &= 0xff;
650 	spin_lock(&ioapic->lock);
651 	switch (addr) {
652 	case IOAPIC_REG_SELECT:
653 		ioapic->ioregsel = data & 0xFF; /* 8-bit register */
654 		break;
655 
656 	case IOAPIC_REG_WINDOW:
657 		ioapic_write_indirect(ioapic, data);
658 		break;
659 
660 	default:
661 		break;
662 	}
663 	spin_unlock(&ioapic->lock);
664 	return 0;
665 }
666 
667 static void kvm_ioapic_reset(struct kvm_ioapic *ioapic)
668 {
669 	int i;
670 
671 	cancel_delayed_work_sync(&ioapic->eoi_inject);
672 	for (i = 0; i < IOAPIC_NUM_PINS; i++)
673 		ioapic->redirtbl[i].fields.mask = 1;
674 	ioapic->base_address = IOAPIC_DEFAULT_BASE_ADDRESS;
675 	ioapic->ioregsel = 0;
676 	ioapic->irr = 0;
677 	ioapic->irr_delivered = 0;
678 	ioapic->id = 0;
679 	memset(ioapic->irq_eoi, 0x00, sizeof(ioapic->irq_eoi));
680 	rtc_irq_eoi_tracking_reset(ioapic);
681 }
682 
683 static const struct kvm_io_device_ops ioapic_mmio_ops = {
684 	.read     = ioapic_mmio_read,
685 	.write    = ioapic_mmio_write,
686 };
687 
688 int kvm_ioapic_init(struct kvm *kvm)
689 {
690 	struct kvm_ioapic *ioapic;
691 	int ret;
692 
693 	ioapic = kzalloc(sizeof(struct kvm_ioapic), GFP_KERNEL_ACCOUNT);
694 	if (!ioapic)
695 		return -ENOMEM;
696 	spin_lock_init(&ioapic->lock);
697 	INIT_DELAYED_WORK(&ioapic->eoi_inject, kvm_ioapic_eoi_inject_work);
698 	kvm->arch.vioapic = ioapic;
699 	kvm_ioapic_reset(ioapic);
700 	kvm_iodevice_init(&ioapic->dev, &ioapic_mmio_ops);
701 	ioapic->kvm = kvm;
702 	mutex_lock(&kvm->slots_lock);
703 	ret = kvm_io_bus_register_dev(kvm, KVM_MMIO_BUS, ioapic->base_address,
704 				      IOAPIC_MEM_LENGTH, &ioapic->dev);
705 	mutex_unlock(&kvm->slots_lock);
706 	if (ret < 0) {
707 		kvm->arch.vioapic = NULL;
708 		kfree(ioapic);
709 	}
710 
711 	return ret;
712 }
713 
714 void kvm_ioapic_destroy(struct kvm *kvm)
715 {
716 	struct kvm_ioapic *ioapic = kvm->arch.vioapic;
717 
718 	if (!ioapic)
719 		return;
720 
721 	cancel_delayed_work_sync(&ioapic->eoi_inject);
722 	mutex_lock(&kvm->slots_lock);
723 	kvm_io_bus_unregister_dev(kvm, KVM_MMIO_BUS, &ioapic->dev);
724 	mutex_unlock(&kvm->slots_lock);
725 	kvm->arch.vioapic = NULL;
726 	kfree(ioapic);
727 }
728 
729 void kvm_get_ioapic(struct kvm *kvm, struct kvm_ioapic_state *state)
730 {
731 	struct kvm_ioapic *ioapic = kvm->arch.vioapic;
732 
733 	spin_lock(&ioapic->lock);
734 	memcpy(state, ioapic, sizeof(struct kvm_ioapic_state));
735 	state->irr &= ~ioapic->irr_delivered;
736 	spin_unlock(&ioapic->lock);
737 }
738 
739 void kvm_set_ioapic(struct kvm *kvm, struct kvm_ioapic_state *state)
740 {
741 	struct kvm_ioapic *ioapic = kvm->arch.vioapic;
742 
743 	spin_lock(&ioapic->lock);
744 	memcpy(ioapic, state, sizeof(struct kvm_ioapic_state));
745 	ioapic->irr = 0;
746 	ioapic->irr_delivered = 0;
747 	kvm_make_scan_ioapic_request(kvm);
748 	kvm_ioapic_inject_all(ioapic, state->irr);
749 	spin_unlock(&ioapic->lock);
750 }
751