Loading...
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#define pr_fmt(fmt) KBUILD_MODNAME ": " fmt
30
31#include <linux/kvm_host.h>
32#include <linux/kvm.h>
33#include <linux/mm.h>
34#include <linux/highmem.h>
35#include <linux/smp.h>
36#include <linux/hrtimer.h>
37#include <linux/io.h>
38#include <linux/slab.h>
39#include <linux/export.h>
40#include <linux/nospec.h>
41#include <asm/processor.h>
42#include <asm/page.h>
43#include <asm/current.h>
44#include <trace/events/kvm.h>
45
46#include "ioapic.h"
47#include "lapic.h"
48#include "irq.h"
49
50static int ioapic_service(struct kvm_ioapic *vioapic, int irq,
51 bool line_status);
52
53static void kvm_ioapic_update_eoi_one(struct kvm_vcpu *vcpu,
54 struct kvm_ioapic *ioapic,
55 int trigger_mode,
56 int pin);
57
58static unsigned long ioapic_read_indirect(struct kvm_ioapic *ioapic)
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 = ~0ULL;
77
78 if (redir_index < IOAPIC_NUM_PINS) {
79 u32 index = array_index_nospec(
80 redir_index, IOAPIC_NUM_PINS);
81
82 redir_content = ioapic->redirtbl[index].bits;
83 }
84
85 result = (ioapic->ioregsel & 0x1) ?
86 (redir_content >> 32) & 0xffffffff :
87 redir_content & 0xffffffff;
88 break;
89 }
90 }
91
92 return result;
93}
94
95static void rtc_irq_eoi_tracking_reset(struct kvm_ioapic *ioapic)
96{
97 ioapic->rtc_status.pending_eoi = 0;
98 bitmap_zero(ioapic->rtc_status.dest_map.map, KVM_MAX_VCPU_IDS);
99}
100
101static void kvm_rtc_eoi_tracking_restore_all(struct kvm_ioapic *ioapic);
102
103static void rtc_status_pending_eoi_check_valid(struct kvm_ioapic *ioapic)
104{
105 if (WARN_ON(ioapic->rtc_status.pending_eoi < 0))
106 kvm_rtc_eoi_tracking_restore_all(ioapic);
107}
108
109static void __rtc_irq_eoi_tracking_restore_one(struct kvm_vcpu *vcpu)
110{
111 bool new_val, old_val;
112 struct kvm_ioapic *ioapic = vcpu->kvm->arch.vioapic;
113 struct dest_map *dest_map = &ioapic->rtc_status.dest_map;
114 union kvm_ioapic_redirect_entry *e;
115
116 e = &ioapic->redirtbl[RTC_GSI];
117 if (!kvm_apic_match_dest(vcpu, NULL, APIC_DEST_NOSHORT,
118 e->fields.dest_id,
119 kvm_lapic_irq_dest_mode(!!e->fields.dest_mode)))
120 return;
121
122 new_val = kvm_apic_pending_eoi(vcpu, e->fields.vector);
123 old_val = test_bit(vcpu->vcpu_id, dest_map->map);
124
125 if (new_val == old_val)
126 return;
127
128 if (new_val) {
129 __set_bit(vcpu->vcpu_id, dest_map->map);
130 dest_map->vectors[vcpu->vcpu_id] = e->fields.vector;
131 ioapic->rtc_status.pending_eoi++;
132 } else {
133 __clear_bit(vcpu->vcpu_id, dest_map->map);
134 ioapic->rtc_status.pending_eoi--;
135 rtc_status_pending_eoi_check_valid(ioapic);
136 }
137}
138
139void kvm_rtc_eoi_tracking_restore_one(struct kvm_vcpu *vcpu)
140{
141 struct kvm_ioapic *ioapic = vcpu->kvm->arch.vioapic;
142
143 spin_lock(&ioapic->lock);
144 __rtc_irq_eoi_tracking_restore_one(vcpu);
145 spin_unlock(&ioapic->lock);
146}
147
148static void kvm_rtc_eoi_tracking_restore_all(struct kvm_ioapic *ioapic)
149{
150 struct kvm_vcpu *vcpu;
151 unsigned long i;
152
153 if (RTC_GSI >= IOAPIC_NUM_PINS)
154 return;
155
156 rtc_irq_eoi_tracking_reset(ioapic);
157 kvm_for_each_vcpu(i, vcpu, ioapic->kvm)
158 __rtc_irq_eoi_tracking_restore_one(vcpu);
159}
160
161static void rtc_irq_eoi(struct kvm_ioapic *ioapic, struct kvm_vcpu *vcpu,
162 int vector)
163{
164 struct dest_map *dest_map = &ioapic->rtc_status.dest_map;
165
166 /* RTC special handling */
167 if (test_bit(vcpu->vcpu_id, dest_map->map) &&
168 (vector == dest_map->vectors[vcpu->vcpu_id]) &&
169 (test_and_clear_bit(vcpu->vcpu_id,
170 ioapic->rtc_status.dest_map.map))) {
171 --ioapic->rtc_status.pending_eoi;
172 rtc_status_pending_eoi_check_valid(ioapic);
173 }
174}
175
176static bool rtc_irq_check_coalesced(struct kvm_ioapic *ioapic)
177{
178 if (ioapic->rtc_status.pending_eoi > 0)
179 return true; /* coalesced */
180
181 return false;
182}
183
184static void ioapic_lazy_update_eoi(struct kvm_ioapic *ioapic, int irq)
185{
186 unsigned long i;
187 struct kvm_vcpu *vcpu;
188 union kvm_ioapic_redirect_entry *entry = &ioapic->redirtbl[irq];
189
190 kvm_for_each_vcpu(i, vcpu, ioapic->kvm) {
191 if (!kvm_apic_match_dest(vcpu, NULL, APIC_DEST_NOSHORT,
192 entry->fields.dest_id,
193 entry->fields.dest_mode) ||
194 kvm_apic_pending_eoi(vcpu, entry->fields.vector))
195 continue;
196
197 /*
198 * If no longer has pending EOI in LAPICs, update
199 * EOI for this vector.
200 */
201 rtc_irq_eoi(ioapic, vcpu, entry->fields.vector);
202 break;
203 }
204}
205
206static int ioapic_set_irq(struct kvm_ioapic *ioapic, unsigned int irq,
207 int irq_level, bool line_status)
208{
209 union kvm_ioapic_redirect_entry entry;
210 u32 mask = 1 << irq;
211 u32 old_irr;
212 int edge, ret;
213
214 entry = ioapic->redirtbl[irq];
215 edge = (entry.fields.trig_mode == IOAPIC_EDGE_TRIG);
216
217 if (!irq_level) {
218 ioapic->irr &= ~mask;
219 ret = 1;
220 goto out;
221 }
222
223 /*
224 * AMD SVM AVIC accelerate EOI write iff the interrupt is edge
225 * triggered, in which case the in-kernel IOAPIC will not be able
226 * to receive the EOI. In this case, we do a lazy update of the
227 * pending EOI when trying to set IOAPIC irq.
228 */
229 if (edge && kvm_apicv_activated(ioapic->kvm))
230 ioapic_lazy_update_eoi(ioapic, irq);
231
232 /*
233 * Return 0 for coalesced interrupts; for edge-triggered interrupts,
234 * this only happens if a previous edge has not been delivered due
235 * to masking. For level interrupts, the remote_irr field tells
236 * us if the interrupt is waiting for an EOI.
237 *
238 * RTC is special: it is edge-triggered, but userspace likes to know
239 * if it has been already ack-ed via EOI because coalesced RTC
240 * interrupts lead to time drift in Windows guests. So we track
241 * EOI manually for the RTC interrupt.
242 */
243 if (irq == RTC_GSI && line_status &&
244 rtc_irq_check_coalesced(ioapic)) {
245 ret = 0;
246 goto out;
247 }
248
249 old_irr = ioapic->irr;
250 ioapic->irr |= mask;
251 if (edge) {
252 ioapic->irr_delivered &= ~mask;
253 if (old_irr == ioapic->irr) {
254 ret = 0;
255 goto out;
256 }
257 }
258
259 ret = ioapic_service(ioapic, irq, line_status);
260
261out:
262 trace_kvm_ioapic_set_irq(entry.bits, irq, ret == 0);
263 return ret;
264}
265
266static void kvm_ioapic_inject_all(struct kvm_ioapic *ioapic, unsigned long irr)
267{
268 u32 idx;
269
270 rtc_irq_eoi_tracking_reset(ioapic);
271 for_each_set_bit(idx, &irr, IOAPIC_NUM_PINS)
272 ioapic_set_irq(ioapic, idx, 1, true);
273
274 kvm_rtc_eoi_tracking_restore_all(ioapic);
275}
276
277
278void kvm_ioapic_scan_entry(struct kvm_vcpu *vcpu, ulong *ioapic_handled_vectors)
279{
280 struct kvm_ioapic *ioapic = vcpu->kvm->arch.vioapic;
281 struct dest_map *dest_map = &ioapic->rtc_status.dest_map;
282 union kvm_ioapic_redirect_entry *e;
283 int index;
284
285 spin_lock(&ioapic->lock);
286
287 /* Make sure we see any missing RTC EOI */
288 if (test_bit(vcpu->vcpu_id, dest_map->map))
289 __set_bit(dest_map->vectors[vcpu->vcpu_id],
290 ioapic_handled_vectors);
291
292 for (index = 0; index < IOAPIC_NUM_PINS; index++) {
293 e = &ioapic->redirtbl[index];
294 if (e->fields.trig_mode == IOAPIC_LEVEL_TRIG ||
295 kvm_irq_has_notifier(ioapic->kvm, KVM_IRQCHIP_IOAPIC, index) ||
296 index == RTC_GSI) {
297 u16 dm = kvm_lapic_irq_dest_mode(!!e->fields.dest_mode);
298
299 if (kvm_apic_match_dest(vcpu, NULL, APIC_DEST_NOSHORT,
300 e->fields.dest_id, dm) ||
301 kvm_apic_pending_eoi(vcpu, e->fields.vector))
302 __set_bit(e->fields.vector,
303 ioapic_handled_vectors);
304 }
305 }
306 spin_unlock(&ioapic->lock);
307}
308
309void kvm_arch_post_irq_ack_notifier_list_update(struct kvm *kvm)
310{
311 if (!ioapic_in_kernel(kvm))
312 return;
313 kvm_make_scan_ioapic_request(kvm);
314}
315
316static void ioapic_write_indirect(struct kvm_ioapic *ioapic, u32 val)
317{
318 unsigned index;
319 bool mask_before, mask_after;
320 union kvm_ioapic_redirect_entry *e;
321 int old_remote_irr, old_delivery_status, old_dest_id, old_dest_mode;
322 DECLARE_BITMAP(vcpu_bitmap, KVM_MAX_VCPUS);
323
324 switch (ioapic->ioregsel) {
325 case IOAPIC_REG_VERSION:
326 /* Writes are ignored. */
327 break;
328
329 case IOAPIC_REG_APIC_ID:
330 ioapic->id = (val >> 24) & 0xf;
331 break;
332
333 case IOAPIC_REG_ARB_ID:
334 break;
335
336 default:
337 index = (ioapic->ioregsel - 0x10) >> 1;
338
339 if (index >= IOAPIC_NUM_PINS)
340 return;
341 index = array_index_nospec(index, IOAPIC_NUM_PINS);
342 e = &ioapic->redirtbl[index];
343 mask_before = e->fields.mask;
344 /* Preserve read-only fields */
345 old_remote_irr = e->fields.remote_irr;
346 old_delivery_status = e->fields.delivery_status;
347 old_dest_id = e->fields.dest_id;
348 old_dest_mode = e->fields.dest_mode;
349 if (ioapic->ioregsel & 1) {
350 e->bits &= 0xffffffff;
351 e->bits |= (u64) val << 32;
352 } else {
353 e->bits &= ~0xffffffffULL;
354 e->bits |= (u32) val;
355 }
356 e->fields.remote_irr = old_remote_irr;
357 e->fields.delivery_status = old_delivery_status;
358
359 /*
360 * Some OSes (Linux, Xen) assume that Remote IRR bit will
361 * be cleared by IOAPIC hardware when the entry is configured
362 * as edge-triggered. This behavior is used to simulate an
363 * explicit EOI on IOAPICs that don't have the EOI register.
364 */
365 if (e->fields.trig_mode == IOAPIC_EDGE_TRIG)
366 e->fields.remote_irr = 0;
367
368 mask_after = e->fields.mask;
369 if (mask_before != mask_after)
370 kvm_fire_mask_notifiers(ioapic->kvm, KVM_IRQCHIP_IOAPIC, index, mask_after);
371 if (e->fields.trig_mode == IOAPIC_LEVEL_TRIG &&
372 ioapic->irr & (1 << index) && !e->fields.mask && !e->fields.remote_irr) {
373 /*
374 * Pending status in irr may be outdated: the IRQ line may have
375 * already been deasserted by a device while the IRQ was masked.
376 * This occurs, for instance, if the interrupt is handled in a
377 * Linux guest as a oneshot interrupt (IRQF_ONESHOT). In this
378 * case the guest acknowledges the interrupt to the device in
379 * its threaded irq handler, i.e. after the EOI but before
380 * unmasking, so at the time of unmasking the IRQ line is
381 * already down but our pending irr bit is still set. In such
382 * cases, injecting this pending interrupt to the guest is
383 * buggy: the guest will receive an extra unwanted interrupt.
384 *
385 * So we need to check here if the IRQ is actually still pending.
386 * As we are generally not able to probe the IRQ line status
387 * directly, we do it through irqfd resampler. Namely, we clear
388 * the pending status and notify the resampler that this interrupt
389 * is done, without actually injecting it into the guest. If the
390 * IRQ line is actually already deasserted, we are done. If it is
391 * still asserted, a new interrupt will be shortly triggered
392 * through irqfd and injected into the guest.
393 *
394 * If, however, it's not possible to resample (no irqfd resampler
395 * registered for this irq), then unconditionally inject this
396 * pending interrupt into the guest, so the guest will not miss
397 * an interrupt, although may get an extra unwanted interrupt.
398 */
399 if (kvm_notify_irqfd_resampler(ioapic->kvm, KVM_IRQCHIP_IOAPIC, index))
400 ioapic->irr &= ~(1 << index);
401 else
402 ioapic_service(ioapic, index, false);
403 }
404 if (e->fields.delivery_mode == APIC_DM_FIXED) {
405 struct kvm_lapic_irq irq;
406
407 irq.vector = e->fields.vector;
408 irq.delivery_mode = e->fields.delivery_mode << 8;
409 irq.dest_mode =
410 kvm_lapic_irq_dest_mode(!!e->fields.dest_mode);
411 irq.level = false;
412 irq.trig_mode = e->fields.trig_mode;
413 irq.shorthand = APIC_DEST_NOSHORT;
414 irq.dest_id = e->fields.dest_id;
415 irq.msi_redir_hint = false;
416 bitmap_zero(vcpu_bitmap, KVM_MAX_VCPUS);
417 kvm_bitmap_or_dest_vcpus(ioapic->kvm, &irq,
418 vcpu_bitmap);
419 if (old_dest_mode != e->fields.dest_mode ||
420 old_dest_id != e->fields.dest_id) {
421 /*
422 * Update vcpu_bitmap with vcpus specified in
423 * the previous request as well. This is done to
424 * keep ioapic_handled_vectors synchronized.
425 */
426 irq.dest_id = old_dest_id;
427 irq.dest_mode =
428 kvm_lapic_irq_dest_mode(
429 !!e->fields.dest_mode);
430 kvm_bitmap_or_dest_vcpus(ioapic->kvm, &irq,
431 vcpu_bitmap);
432 }
433 kvm_make_scan_ioapic_request_mask(ioapic->kvm,
434 vcpu_bitmap);
435 } else {
436 kvm_make_scan_ioapic_request(ioapic->kvm);
437 }
438 break;
439 }
440}
441
442static int ioapic_service(struct kvm_ioapic *ioapic, int irq, bool line_status)
443{
444 union kvm_ioapic_redirect_entry *entry = &ioapic->redirtbl[irq];
445 struct kvm_lapic_irq irqe;
446 int ret;
447
448 if (entry->fields.mask ||
449 (entry->fields.trig_mode == IOAPIC_LEVEL_TRIG &&
450 entry->fields.remote_irr))
451 return -1;
452
453 irqe.dest_id = entry->fields.dest_id;
454 irqe.vector = entry->fields.vector;
455 irqe.dest_mode = kvm_lapic_irq_dest_mode(!!entry->fields.dest_mode);
456 irqe.trig_mode = entry->fields.trig_mode;
457 irqe.delivery_mode = entry->fields.delivery_mode << 8;
458 irqe.level = 1;
459 irqe.shorthand = APIC_DEST_NOSHORT;
460 irqe.msi_redir_hint = false;
461
462 if (irqe.trig_mode == IOAPIC_EDGE_TRIG)
463 ioapic->irr_delivered |= 1 << irq;
464
465 if (irq == RTC_GSI && line_status) {
466 /*
467 * pending_eoi cannot ever become negative (see
468 * rtc_status_pending_eoi_check_valid) and the caller
469 * ensures that it is only called if it is >= zero, namely
470 * if rtc_irq_check_coalesced returns false).
471 */
472 BUG_ON(ioapic->rtc_status.pending_eoi != 0);
473 ret = kvm_irq_delivery_to_apic(ioapic->kvm, NULL, &irqe,
474 &ioapic->rtc_status.dest_map);
475 ioapic->rtc_status.pending_eoi = (ret < 0 ? 0 : ret);
476 } else
477 ret = kvm_irq_delivery_to_apic(ioapic->kvm, NULL, &irqe, NULL);
478
479 if (ret && irqe.trig_mode == IOAPIC_LEVEL_TRIG)
480 entry->fields.remote_irr = 1;
481
482 return ret;
483}
484
485int kvm_ioapic_set_irq(struct kvm_ioapic *ioapic, int irq, int irq_source_id,
486 int level, bool line_status)
487{
488 int ret, irq_level;
489
490 BUG_ON(irq < 0 || irq >= IOAPIC_NUM_PINS);
491
492 spin_lock(&ioapic->lock);
493 irq_level = __kvm_irq_line_state(&ioapic->irq_states[irq],
494 irq_source_id, level);
495 ret = ioapic_set_irq(ioapic, irq, irq_level, line_status);
496
497 spin_unlock(&ioapic->lock);
498
499 return ret;
500}
501
502void kvm_ioapic_clear_all(struct kvm_ioapic *ioapic, int irq_source_id)
503{
504 int i;
505
506 spin_lock(&ioapic->lock);
507 for (i = 0; i < KVM_IOAPIC_NUM_PINS; i++)
508 __clear_bit(irq_source_id, &ioapic->irq_states[i]);
509 spin_unlock(&ioapic->lock);
510}
511
512static void kvm_ioapic_eoi_inject_work(struct work_struct *work)
513{
514 int i;
515 struct kvm_ioapic *ioapic = container_of(work, struct kvm_ioapic,
516 eoi_inject.work);
517 spin_lock(&ioapic->lock);
518 for (i = 0; i < IOAPIC_NUM_PINS; i++) {
519 union kvm_ioapic_redirect_entry *ent = &ioapic->redirtbl[i];
520
521 if (ent->fields.trig_mode != IOAPIC_LEVEL_TRIG)
522 continue;
523
524 if (ioapic->irr & (1 << i) && !ent->fields.remote_irr)
525 ioapic_service(ioapic, i, false);
526 }
527 spin_unlock(&ioapic->lock);
528}
529
530#define IOAPIC_SUCCESSIVE_IRQ_MAX_COUNT 10000
531static void kvm_ioapic_update_eoi_one(struct kvm_vcpu *vcpu,
532 struct kvm_ioapic *ioapic,
533 int trigger_mode,
534 int pin)
535{
536 struct kvm_lapic *apic = vcpu->arch.apic;
537 union kvm_ioapic_redirect_entry *ent = &ioapic->redirtbl[pin];
538
539 /*
540 * We are dropping lock while calling ack notifiers because ack
541 * notifier callbacks for assigned devices call into IOAPIC
542 * recursively. Since remote_irr is cleared only after call
543 * to notifiers if the same vector will be delivered while lock
544 * is dropped it will be put into irr and will be delivered
545 * after ack notifier returns.
546 */
547 spin_unlock(&ioapic->lock);
548 kvm_notify_acked_irq(ioapic->kvm, KVM_IRQCHIP_IOAPIC, pin);
549 spin_lock(&ioapic->lock);
550
551 if (trigger_mode != IOAPIC_LEVEL_TRIG ||
552 kvm_lapic_get_reg(apic, APIC_SPIV) & APIC_SPIV_DIRECTED_EOI)
553 return;
554
555 ASSERT(ent->fields.trig_mode == IOAPIC_LEVEL_TRIG);
556 ent->fields.remote_irr = 0;
557 if (!ent->fields.mask && (ioapic->irr & (1 << pin))) {
558 ++ioapic->irq_eoi[pin];
559 if (ioapic->irq_eoi[pin] == IOAPIC_SUCCESSIVE_IRQ_MAX_COUNT) {
560 /*
561 * Real hardware does not deliver the interrupt
562 * immediately during eoi broadcast, and this
563 * lets a buggy guest make slow progress
564 * even if it does not correctly handle a
565 * level-triggered interrupt. Emulate this
566 * behavior if we detect an interrupt storm.
567 */
568 schedule_delayed_work(&ioapic->eoi_inject, HZ / 100);
569 ioapic->irq_eoi[pin] = 0;
570 trace_kvm_ioapic_delayed_eoi_inj(ent->bits);
571 } else {
572 ioapic_service(ioapic, pin, false);
573 }
574 } else {
575 ioapic->irq_eoi[pin] = 0;
576 }
577}
578
579void kvm_ioapic_update_eoi(struct kvm_vcpu *vcpu, int vector, int trigger_mode)
580{
581 int i;
582 struct kvm_ioapic *ioapic = vcpu->kvm->arch.vioapic;
583
584 spin_lock(&ioapic->lock);
585 rtc_irq_eoi(ioapic, vcpu, vector);
586 for (i = 0; i < IOAPIC_NUM_PINS; i++) {
587 union kvm_ioapic_redirect_entry *ent = &ioapic->redirtbl[i];
588
589 if (ent->fields.vector != vector)
590 continue;
591 kvm_ioapic_update_eoi_one(vcpu, ioapic, trigger_mode, i);
592 }
593 spin_unlock(&ioapic->lock);
594}
595
596static inline struct kvm_ioapic *to_ioapic(struct kvm_io_device *dev)
597{
598 return container_of(dev, struct kvm_ioapic, dev);
599}
600
601static inline int ioapic_in_range(struct kvm_ioapic *ioapic, gpa_t addr)
602{
603 return ((addr >= ioapic->base_address &&
604 (addr < ioapic->base_address + IOAPIC_MEM_LENGTH)));
605}
606
607static int ioapic_mmio_read(struct kvm_vcpu *vcpu, struct kvm_io_device *this,
608 gpa_t addr, int len, void *val)
609{
610 struct kvm_ioapic *ioapic = to_ioapic(this);
611 u32 result;
612 if (!ioapic_in_range(ioapic, addr))
613 return -EOPNOTSUPP;
614
615 ASSERT(!(addr & 0xf)); /* check alignment */
616
617 addr &= 0xff;
618 spin_lock(&ioapic->lock);
619 switch (addr) {
620 case IOAPIC_REG_SELECT:
621 result = ioapic->ioregsel;
622 break;
623
624 case IOAPIC_REG_WINDOW:
625 result = ioapic_read_indirect(ioapic);
626 break;
627
628 default:
629 result = 0;
630 break;
631 }
632 spin_unlock(&ioapic->lock);
633
634 switch (len) {
635 case 8:
636 *(u64 *) val = result;
637 break;
638 case 1:
639 case 2:
640 case 4:
641 memcpy(val, (char *)&result, len);
642 break;
643 default:
644 printk(KERN_WARNING "ioapic: wrong length %d\n", len);
645 }
646 return 0;
647}
648
649static int ioapic_mmio_write(struct kvm_vcpu *vcpu, struct kvm_io_device *this,
650 gpa_t addr, int len, const void *val)
651{
652 struct kvm_ioapic *ioapic = to_ioapic(this);
653 u32 data;
654 if (!ioapic_in_range(ioapic, addr))
655 return -EOPNOTSUPP;
656
657 ASSERT(!(addr & 0xf)); /* check alignment */
658
659 switch (len) {
660 case 8:
661 case 4:
662 data = *(u32 *) val;
663 break;
664 case 2:
665 data = *(u16 *) val;
666 break;
667 case 1:
668 data = *(u8 *) val;
669 break;
670 default:
671 printk(KERN_WARNING "ioapic: Unsupported size %d\n", len);
672 return 0;
673 }
674
675 addr &= 0xff;
676 spin_lock(&ioapic->lock);
677 switch (addr) {
678 case IOAPIC_REG_SELECT:
679 ioapic->ioregsel = data & 0xFF; /* 8-bit register */
680 break;
681
682 case IOAPIC_REG_WINDOW:
683 ioapic_write_indirect(ioapic, data);
684 break;
685
686 default:
687 break;
688 }
689 spin_unlock(&ioapic->lock);
690 return 0;
691}
692
693static void kvm_ioapic_reset(struct kvm_ioapic *ioapic)
694{
695 int i;
696
697 cancel_delayed_work_sync(&ioapic->eoi_inject);
698 for (i = 0; i < IOAPIC_NUM_PINS; i++)
699 ioapic->redirtbl[i].fields.mask = 1;
700 ioapic->base_address = IOAPIC_DEFAULT_BASE_ADDRESS;
701 ioapic->ioregsel = 0;
702 ioapic->irr = 0;
703 ioapic->irr_delivered = 0;
704 ioapic->id = 0;
705 memset(ioapic->irq_eoi, 0x00, sizeof(ioapic->irq_eoi));
706 rtc_irq_eoi_tracking_reset(ioapic);
707}
708
709static const struct kvm_io_device_ops ioapic_mmio_ops = {
710 .read = ioapic_mmio_read,
711 .write = ioapic_mmio_write,
712};
713
714int kvm_ioapic_init(struct kvm *kvm)
715{
716 struct kvm_ioapic *ioapic;
717 int ret;
718
719 ioapic = kzalloc(sizeof(struct kvm_ioapic), GFP_KERNEL_ACCOUNT);
720 if (!ioapic)
721 return -ENOMEM;
722 spin_lock_init(&ioapic->lock);
723 INIT_DELAYED_WORK(&ioapic->eoi_inject, kvm_ioapic_eoi_inject_work);
724 kvm->arch.vioapic = ioapic;
725 kvm_ioapic_reset(ioapic);
726 kvm_iodevice_init(&ioapic->dev, &ioapic_mmio_ops);
727 ioapic->kvm = kvm;
728 mutex_lock(&kvm->slots_lock);
729 ret = kvm_io_bus_register_dev(kvm, KVM_MMIO_BUS, ioapic->base_address,
730 IOAPIC_MEM_LENGTH, &ioapic->dev);
731 mutex_unlock(&kvm->slots_lock);
732 if (ret < 0) {
733 kvm->arch.vioapic = NULL;
734 kfree(ioapic);
735 }
736
737 return ret;
738}
739
740void kvm_ioapic_destroy(struct kvm *kvm)
741{
742 struct kvm_ioapic *ioapic = kvm->arch.vioapic;
743
744 if (!ioapic)
745 return;
746
747 cancel_delayed_work_sync(&ioapic->eoi_inject);
748 mutex_lock(&kvm->slots_lock);
749 kvm_io_bus_unregister_dev(kvm, KVM_MMIO_BUS, &ioapic->dev);
750 mutex_unlock(&kvm->slots_lock);
751 kvm->arch.vioapic = NULL;
752 kfree(ioapic);
753}
754
755void kvm_get_ioapic(struct kvm *kvm, struct kvm_ioapic_state *state)
756{
757 struct kvm_ioapic *ioapic = kvm->arch.vioapic;
758
759 spin_lock(&ioapic->lock);
760 memcpy(state, ioapic, sizeof(struct kvm_ioapic_state));
761 state->irr &= ~ioapic->irr_delivered;
762 spin_unlock(&ioapic->lock);
763}
764
765void kvm_set_ioapic(struct kvm *kvm, struct kvm_ioapic_state *state)
766{
767 struct kvm_ioapic *ioapic = kvm->arch.vioapic;
768
769 spin_lock(&ioapic->lock);
770 memcpy(ioapic, state, sizeof(struct kvm_ioapic_state));
771 ioapic->irr = 0;
772 ioapic->irr_delivered = 0;
773 kvm_make_scan_ioapic_request(kvm);
774 kvm_ioapic_inject_all(ioapic, state->irr);
775 spin_unlock(&ioapic->lock);
776}
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
49static int ioapic_service(struct kvm_ioapic *vioapic, int irq,
50 bool line_status);
51
52static void kvm_ioapic_update_eoi_one(struct kvm_vcpu *vcpu,
53 struct kvm_ioapic *ioapic,
54 int trigger_mode,
55 int pin);
56
57static 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
96static 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 + 1);
100}
101
102static void kvm_rtc_eoi_tracking_restore_all(struct kvm_ioapic *ioapic);
103
104static 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
110static 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
140void 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
149static 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
162static 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
177static 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
185static 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 vector.
201 */
202 rtc_irq_eoi(ioapic, vcpu, entry->fields.vector);
203 break;
204 }
205}
206
207static int ioapic_set_irq(struct kvm_ioapic *ioapic, unsigned int irq,
208 int irq_level, bool line_status)
209{
210 union kvm_ioapic_redirect_entry entry;
211 u32 mask = 1 << irq;
212 u32 old_irr;
213 int edge, ret;
214
215 entry = ioapic->redirtbl[irq];
216 edge = (entry.fields.trig_mode == IOAPIC_EDGE_TRIG);
217
218 if (!irq_level) {
219 ioapic->irr &= ~mask;
220 ret = 1;
221 goto out;
222 }
223
224 /*
225 * AMD SVM AVIC accelerate EOI write iff the interrupt is edge
226 * triggered, in which case the in-kernel IOAPIC will not be able
227 * to receive the EOI. In this case, we do a lazy update of the
228 * pending EOI when trying to set IOAPIC irq.
229 */
230 if (edge && kvm_apicv_activated(ioapic->kvm))
231 ioapic_lazy_update_eoi(ioapic, irq);
232
233 /*
234 * Return 0 for coalesced interrupts; for edge-triggered interrupts,
235 * this only happens if a previous edge has not been delivered due
236 * to masking. For level interrupts, the remote_irr field tells
237 * us if the interrupt is waiting for an EOI.
238 *
239 * RTC is special: it is edge-triggered, but userspace likes to know
240 * if it has been already ack-ed via EOI because coalesced RTC
241 * interrupts lead to time drift in Windows guests. So we track
242 * EOI manually for the RTC interrupt.
243 */
244 if (irq == RTC_GSI && line_status &&
245 rtc_irq_check_coalesced(ioapic)) {
246 ret = 0;
247 goto out;
248 }
249
250 old_irr = ioapic->irr;
251 ioapic->irr |= mask;
252 if (edge) {
253 ioapic->irr_delivered &= ~mask;
254 if (old_irr == ioapic->irr) {
255 ret = 0;
256 goto out;
257 }
258 }
259
260 ret = ioapic_service(ioapic, irq, line_status);
261
262out:
263 trace_kvm_ioapic_set_irq(entry.bits, irq, ret == 0);
264 return ret;
265}
266
267static void kvm_ioapic_inject_all(struct kvm_ioapic *ioapic, unsigned long irr)
268{
269 u32 idx;
270
271 rtc_irq_eoi_tracking_reset(ioapic);
272 for_each_set_bit(idx, &irr, IOAPIC_NUM_PINS)
273 ioapic_set_irq(ioapic, idx, 1, true);
274
275 kvm_rtc_eoi_tracking_restore_all(ioapic);
276}
277
278
279void kvm_ioapic_scan_entry(struct kvm_vcpu *vcpu, ulong *ioapic_handled_vectors)
280{
281 struct kvm_ioapic *ioapic = vcpu->kvm->arch.vioapic;
282 struct dest_map *dest_map = &ioapic->rtc_status.dest_map;
283 union kvm_ioapic_redirect_entry *e;
284 int index;
285
286 spin_lock(&ioapic->lock);
287
288 /* Make sure we see any missing RTC EOI */
289 if (test_bit(vcpu->vcpu_id, dest_map->map))
290 __set_bit(dest_map->vectors[vcpu->vcpu_id],
291 ioapic_handled_vectors);
292
293 for (index = 0; index < IOAPIC_NUM_PINS; index++) {
294 e = &ioapic->redirtbl[index];
295 if (e->fields.trig_mode == IOAPIC_LEVEL_TRIG ||
296 kvm_irq_has_notifier(ioapic->kvm, KVM_IRQCHIP_IOAPIC, index) ||
297 index == RTC_GSI) {
298 u16 dm = kvm_lapic_irq_dest_mode(!!e->fields.dest_mode);
299
300 if (kvm_apic_match_dest(vcpu, NULL, APIC_DEST_NOSHORT,
301 e->fields.dest_id, dm) ||
302 kvm_apic_pending_eoi(vcpu, e->fields.vector))
303 __set_bit(e->fields.vector,
304 ioapic_handled_vectors);
305 }
306 }
307 spin_unlock(&ioapic->lock);
308}
309
310void kvm_arch_post_irq_ack_notifier_list_update(struct kvm *kvm)
311{
312 if (!ioapic_in_kernel(kvm))
313 return;
314 kvm_make_scan_ioapic_request(kvm);
315}
316
317static void ioapic_write_indirect(struct kvm_ioapic *ioapic, u32 val)
318{
319 unsigned index;
320 bool mask_before, mask_after;
321 union kvm_ioapic_redirect_entry *e;
322 int old_remote_irr, old_delivery_status, old_dest_id, old_dest_mode;
323 DECLARE_BITMAP(vcpu_bitmap, KVM_MAX_VCPUS);
324
325 switch (ioapic->ioregsel) {
326 case IOAPIC_REG_VERSION:
327 /* Writes are ignored. */
328 break;
329
330 case IOAPIC_REG_APIC_ID:
331 ioapic->id = (val >> 24) & 0xf;
332 break;
333
334 case IOAPIC_REG_ARB_ID:
335 break;
336
337 default:
338 index = (ioapic->ioregsel - 0x10) >> 1;
339
340 if (index >= IOAPIC_NUM_PINS)
341 return;
342 index = array_index_nospec(index, IOAPIC_NUM_PINS);
343 e = &ioapic->redirtbl[index];
344 mask_before = e->fields.mask;
345 /* Preserve read-only fields */
346 old_remote_irr = e->fields.remote_irr;
347 old_delivery_status = e->fields.delivery_status;
348 old_dest_id = e->fields.dest_id;
349 old_dest_mode = e->fields.dest_mode;
350 if (ioapic->ioregsel & 1) {
351 e->bits &= 0xffffffff;
352 e->bits |= (u64) val << 32;
353 } else {
354 e->bits &= ~0xffffffffULL;
355 e->bits |= (u32) val;
356 }
357 e->fields.remote_irr = old_remote_irr;
358 e->fields.delivery_status = old_delivery_status;
359
360 /*
361 * Some OSes (Linux, Xen) assume that Remote IRR bit will
362 * be cleared by IOAPIC hardware when the entry is configured
363 * as edge-triggered. This behavior is used to simulate an
364 * explicit EOI on IOAPICs that don't have the EOI register.
365 */
366 if (e->fields.trig_mode == IOAPIC_EDGE_TRIG)
367 e->fields.remote_irr = 0;
368
369 mask_after = e->fields.mask;
370 if (mask_before != mask_after)
371 kvm_fire_mask_notifiers(ioapic->kvm, KVM_IRQCHIP_IOAPIC, index, mask_after);
372 if (e->fields.trig_mode == IOAPIC_LEVEL_TRIG
373 && ioapic->irr & (1 << index))
374 ioapic_service(ioapic, index, false);
375 if (e->fields.delivery_mode == APIC_DM_FIXED) {
376 struct kvm_lapic_irq irq;
377
378 irq.vector = e->fields.vector;
379 irq.delivery_mode = e->fields.delivery_mode << 8;
380 irq.dest_mode =
381 kvm_lapic_irq_dest_mode(!!e->fields.dest_mode);
382 irq.level = false;
383 irq.trig_mode = e->fields.trig_mode;
384 irq.shorthand = APIC_DEST_NOSHORT;
385 irq.dest_id = e->fields.dest_id;
386 irq.msi_redir_hint = false;
387 bitmap_zero(vcpu_bitmap, KVM_MAX_VCPUS);
388 kvm_bitmap_or_dest_vcpus(ioapic->kvm, &irq,
389 vcpu_bitmap);
390 if (old_dest_mode != e->fields.dest_mode ||
391 old_dest_id != e->fields.dest_id) {
392 /*
393 * Update vcpu_bitmap with vcpus specified in
394 * the previous request as well. This is done to
395 * keep ioapic_handled_vectors synchronized.
396 */
397 irq.dest_id = old_dest_id;
398 irq.dest_mode =
399 kvm_lapic_irq_dest_mode(
400 !!e->fields.dest_mode);
401 kvm_bitmap_or_dest_vcpus(ioapic->kvm, &irq,
402 vcpu_bitmap);
403 }
404 kvm_make_scan_ioapic_request_mask(ioapic->kvm,
405 vcpu_bitmap);
406 } else {
407 kvm_make_scan_ioapic_request(ioapic->kvm);
408 }
409 break;
410 }
411}
412
413static int ioapic_service(struct kvm_ioapic *ioapic, int irq, bool line_status)
414{
415 union kvm_ioapic_redirect_entry *entry = &ioapic->redirtbl[irq];
416 struct kvm_lapic_irq irqe;
417 int ret;
418
419 if (entry->fields.mask ||
420 (entry->fields.trig_mode == IOAPIC_LEVEL_TRIG &&
421 entry->fields.remote_irr))
422 return -1;
423
424 irqe.dest_id = entry->fields.dest_id;
425 irqe.vector = entry->fields.vector;
426 irqe.dest_mode = kvm_lapic_irq_dest_mode(!!entry->fields.dest_mode);
427 irqe.trig_mode = entry->fields.trig_mode;
428 irqe.delivery_mode = entry->fields.delivery_mode << 8;
429 irqe.level = 1;
430 irqe.shorthand = APIC_DEST_NOSHORT;
431 irqe.msi_redir_hint = false;
432
433 if (irqe.trig_mode == IOAPIC_EDGE_TRIG)
434 ioapic->irr_delivered |= 1 << irq;
435
436 if (irq == RTC_GSI && line_status) {
437 /*
438 * pending_eoi cannot ever become negative (see
439 * rtc_status_pending_eoi_check_valid) and the caller
440 * ensures that it is only called if it is >= zero, namely
441 * if rtc_irq_check_coalesced returns false).
442 */
443 BUG_ON(ioapic->rtc_status.pending_eoi != 0);
444 ret = kvm_irq_delivery_to_apic(ioapic->kvm, NULL, &irqe,
445 &ioapic->rtc_status.dest_map);
446 ioapic->rtc_status.pending_eoi = (ret < 0 ? 0 : ret);
447 } else
448 ret = kvm_irq_delivery_to_apic(ioapic->kvm, NULL, &irqe, NULL);
449
450 if (ret && irqe.trig_mode == IOAPIC_LEVEL_TRIG)
451 entry->fields.remote_irr = 1;
452
453 return ret;
454}
455
456int kvm_ioapic_set_irq(struct kvm_ioapic *ioapic, int irq, int irq_source_id,
457 int level, bool line_status)
458{
459 int ret, irq_level;
460
461 BUG_ON(irq < 0 || irq >= IOAPIC_NUM_PINS);
462
463 spin_lock(&ioapic->lock);
464 irq_level = __kvm_irq_line_state(&ioapic->irq_states[irq],
465 irq_source_id, level);
466 ret = ioapic_set_irq(ioapic, irq, irq_level, line_status);
467
468 spin_unlock(&ioapic->lock);
469
470 return ret;
471}
472
473void kvm_ioapic_clear_all(struct kvm_ioapic *ioapic, int irq_source_id)
474{
475 int i;
476
477 spin_lock(&ioapic->lock);
478 for (i = 0; i < KVM_IOAPIC_NUM_PINS; i++)
479 __clear_bit(irq_source_id, &ioapic->irq_states[i]);
480 spin_unlock(&ioapic->lock);
481}
482
483static void kvm_ioapic_eoi_inject_work(struct work_struct *work)
484{
485 int i;
486 struct kvm_ioapic *ioapic = container_of(work, struct kvm_ioapic,
487 eoi_inject.work);
488 spin_lock(&ioapic->lock);
489 for (i = 0; i < IOAPIC_NUM_PINS; i++) {
490 union kvm_ioapic_redirect_entry *ent = &ioapic->redirtbl[i];
491
492 if (ent->fields.trig_mode != IOAPIC_LEVEL_TRIG)
493 continue;
494
495 if (ioapic->irr & (1 << i) && !ent->fields.remote_irr)
496 ioapic_service(ioapic, i, false);
497 }
498 spin_unlock(&ioapic->lock);
499}
500
501#define IOAPIC_SUCCESSIVE_IRQ_MAX_COUNT 10000
502static void kvm_ioapic_update_eoi_one(struct kvm_vcpu *vcpu,
503 struct kvm_ioapic *ioapic,
504 int trigger_mode,
505 int pin)
506{
507 struct kvm_lapic *apic = vcpu->arch.apic;
508 union kvm_ioapic_redirect_entry *ent = &ioapic->redirtbl[pin];
509
510 /*
511 * We are dropping lock while calling ack notifiers because ack
512 * notifier callbacks for assigned devices call into IOAPIC
513 * recursively. Since remote_irr is cleared only after call
514 * to notifiers if the same vector will be delivered while lock
515 * is dropped it will be put into irr and will be delivered
516 * after ack notifier returns.
517 */
518 spin_unlock(&ioapic->lock);
519 kvm_notify_acked_irq(ioapic->kvm, KVM_IRQCHIP_IOAPIC, pin);
520 spin_lock(&ioapic->lock);
521
522 if (trigger_mode != IOAPIC_LEVEL_TRIG ||
523 kvm_lapic_get_reg(apic, APIC_SPIV) & APIC_SPIV_DIRECTED_EOI)
524 return;
525
526 ASSERT(ent->fields.trig_mode == IOAPIC_LEVEL_TRIG);
527 ent->fields.remote_irr = 0;
528 if (!ent->fields.mask && (ioapic->irr & (1 << pin))) {
529 ++ioapic->irq_eoi[pin];
530 if (ioapic->irq_eoi[pin] == IOAPIC_SUCCESSIVE_IRQ_MAX_COUNT) {
531 /*
532 * Real hardware does not deliver the interrupt
533 * immediately during eoi broadcast, and this
534 * lets a buggy guest make slow progress
535 * even if it does not correctly handle a
536 * level-triggered interrupt. Emulate this
537 * behavior if we detect an interrupt storm.
538 */
539 schedule_delayed_work(&ioapic->eoi_inject, HZ / 100);
540 ioapic->irq_eoi[pin] = 0;
541 trace_kvm_ioapic_delayed_eoi_inj(ent->bits);
542 } else {
543 ioapic_service(ioapic, pin, false);
544 }
545 } else {
546 ioapic->irq_eoi[pin] = 0;
547 }
548}
549
550void kvm_ioapic_update_eoi(struct kvm_vcpu *vcpu, int vector, int trigger_mode)
551{
552 int i;
553 struct kvm_ioapic *ioapic = vcpu->kvm->arch.vioapic;
554
555 spin_lock(&ioapic->lock);
556 rtc_irq_eoi(ioapic, vcpu, vector);
557 for (i = 0; i < IOAPIC_NUM_PINS; i++) {
558 union kvm_ioapic_redirect_entry *ent = &ioapic->redirtbl[i];
559
560 if (ent->fields.vector != vector)
561 continue;
562 kvm_ioapic_update_eoi_one(vcpu, ioapic, trigger_mode, i);
563 }
564 spin_unlock(&ioapic->lock);
565}
566
567static inline struct kvm_ioapic *to_ioapic(struct kvm_io_device *dev)
568{
569 return container_of(dev, struct kvm_ioapic, dev);
570}
571
572static inline int ioapic_in_range(struct kvm_ioapic *ioapic, gpa_t addr)
573{
574 return ((addr >= ioapic->base_address &&
575 (addr < ioapic->base_address + IOAPIC_MEM_LENGTH)));
576}
577
578static int ioapic_mmio_read(struct kvm_vcpu *vcpu, struct kvm_io_device *this,
579 gpa_t addr, int len, void *val)
580{
581 struct kvm_ioapic *ioapic = to_ioapic(this);
582 u32 result;
583 if (!ioapic_in_range(ioapic, addr))
584 return -EOPNOTSUPP;
585
586 ASSERT(!(addr & 0xf)); /* check alignment */
587
588 addr &= 0xff;
589 spin_lock(&ioapic->lock);
590 switch (addr) {
591 case IOAPIC_REG_SELECT:
592 result = ioapic->ioregsel;
593 break;
594
595 case IOAPIC_REG_WINDOW:
596 result = ioapic_read_indirect(ioapic, addr, len);
597 break;
598
599 default:
600 result = 0;
601 break;
602 }
603 spin_unlock(&ioapic->lock);
604
605 switch (len) {
606 case 8:
607 *(u64 *) val = result;
608 break;
609 case 1:
610 case 2:
611 case 4:
612 memcpy(val, (char *)&result, len);
613 break;
614 default:
615 printk(KERN_WARNING "ioapic: wrong length %d\n", len);
616 }
617 return 0;
618}
619
620static int ioapic_mmio_write(struct kvm_vcpu *vcpu, struct kvm_io_device *this,
621 gpa_t addr, int len, const void *val)
622{
623 struct kvm_ioapic *ioapic = to_ioapic(this);
624 u32 data;
625 if (!ioapic_in_range(ioapic, addr))
626 return -EOPNOTSUPP;
627
628 ASSERT(!(addr & 0xf)); /* check alignment */
629
630 switch (len) {
631 case 8:
632 case 4:
633 data = *(u32 *) val;
634 break;
635 case 2:
636 data = *(u16 *) val;
637 break;
638 case 1:
639 data = *(u8 *) val;
640 break;
641 default:
642 printk(KERN_WARNING "ioapic: Unsupported size %d\n", len);
643 return 0;
644 }
645
646 addr &= 0xff;
647 spin_lock(&ioapic->lock);
648 switch (addr) {
649 case IOAPIC_REG_SELECT:
650 ioapic->ioregsel = data & 0xFF; /* 8-bit register */
651 break;
652
653 case IOAPIC_REG_WINDOW:
654 ioapic_write_indirect(ioapic, data);
655 break;
656
657 default:
658 break;
659 }
660 spin_unlock(&ioapic->lock);
661 return 0;
662}
663
664static void kvm_ioapic_reset(struct kvm_ioapic *ioapic)
665{
666 int i;
667
668 cancel_delayed_work_sync(&ioapic->eoi_inject);
669 for (i = 0; i < IOAPIC_NUM_PINS; i++)
670 ioapic->redirtbl[i].fields.mask = 1;
671 ioapic->base_address = IOAPIC_DEFAULT_BASE_ADDRESS;
672 ioapic->ioregsel = 0;
673 ioapic->irr = 0;
674 ioapic->irr_delivered = 0;
675 ioapic->id = 0;
676 memset(ioapic->irq_eoi, 0x00, sizeof(ioapic->irq_eoi));
677 rtc_irq_eoi_tracking_reset(ioapic);
678}
679
680static const struct kvm_io_device_ops ioapic_mmio_ops = {
681 .read = ioapic_mmio_read,
682 .write = ioapic_mmio_write,
683};
684
685int kvm_ioapic_init(struct kvm *kvm)
686{
687 struct kvm_ioapic *ioapic;
688 int ret;
689
690 ioapic = kzalloc(sizeof(struct kvm_ioapic), GFP_KERNEL_ACCOUNT);
691 if (!ioapic)
692 return -ENOMEM;
693 spin_lock_init(&ioapic->lock);
694 INIT_DELAYED_WORK(&ioapic->eoi_inject, kvm_ioapic_eoi_inject_work);
695 kvm->arch.vioapic = ioapic;
696 kvm_ioapic_reset(ioapic);
697 kvm_iodevice_init(&ioapic->dev, &ioapic_mmio_ops);
698 ioapic->kvm = kvm;
699 mutex_lock(&kvm->slots_lock);
700 ret = kvm_io_bus_register_dev(kvm, KVM_MMIO_BUS, ioapic->base_address,
701 IOAPIC_MEM_LENGTH, &ioapic->dev);
702 mutex_unlock(&kvm->slots_lock);
703 if (ret < 0) {
704 kvm->arch.vioapic = NULL;
705 kfree(ioapic);
706 }
707
708 return ret;
709}
710
711void kvm_ioapic_destroy(struct kvm *kvm)
712{
713 struct kvm_ioapic *ioapic = kvm->arch.vioapic;
714
715 if (!ioapic)
716 return;
717
718 cancel_delayed_work_sync(&ioapic->eoi_inject);
719 mutex_lock(&kvm->slots_lock);
720 kvm_io_bus_unregister_dev(kvm, KVM_MMIO_BUS, &ioapic->dev);
721 mutex_unlock(&kvm->slots_lock);
722 kvm->arch.vioapic = NULL;
723 kfree(ioapic);
724}
725
726void kvm_get_ioapic(struct kvm *kvm, struct kvm_ioapic_state *state)
727{
728 struct kvm_ioapic *ioapic = kvm->arch.vioapic;
729
730 spin_lock(&ioapic->lock);
731 memcpy(state, ioapic, sizeof(struct kvm_ioapic_state));
732 state->irr &= ~ioapic->irr_delivered;
733 spin_unlock(&ioapic->lock);
734}
735
736void kvm_set_ioapic(struct kvm *kvm, struct kvm_ioapic_state *state)
737{
738 struct kvm_ioapic *ioapic = kvm->arch.vioapic;
739
740 spin_lock(&ioapic->lock);
741 memcpy(ioapic, state, sizeof(struct kvm_ioapic_state));
742 ioapic->irr = 0;
743 ioapic->irr_delivered = 0;
744 kvm_make_scan_ioapic_request(kvm);
745 kvm_ioapic_inject_all(ioapic, state->irr);
746 spin_unlock(&ioapic->lock);
747}