2 * Copyright (C) 2001 MandrakeSoft S.A.
3 * Copyright 2010 Red Hat, Inc. and/or its affiliates.
8 * http://www.linux-mandrake.com/
9 * http://www.mandrakesoft.com/
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.
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.
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
25 * Yunhong Jiang <yunhong.jiang@intel.com>
26 * Yaozu (Eddie) Dong <eddie.dong@intel.com>
27 * Based on Xen 3.1 code.
30 #include <linux/kvm_host.h>
31 #include <linux/kvm.h>
33 #include <linux/highmem.h>
34 #include <linux/smp.h>
35 #include <linux/hrtimer.h>
37 #include <linux/slab.h>
38 #include <linux/export.h>
39 #include <asm/processor.h>
41 #include <asm/current.h>
42 #include <trace/events/kvm.h>
49 #define ioapic_debug(fmt,arg...) printk(KERN_WARNING fmt,##arg)
51 #define ioapic_debug(fmt, arg...)
53 static int ioapic_service(struct kvm_ioapic *vioapic, int irq,
56 static unsigned long ioapic_read_indirect(struct kvm_ioapic *ioapic,
60 unsigned long result = 0;
62 switch (ioapic->ioregsel) {
63 case IOAPIC_REG_VERSION:
64 result = ((((IOAPIC_NUM_PINS - 1) & 0xff) << 16)
65 | (IOAPIC_VERSION_ID & 0xff));
68 case IOAPIC_REG_APIC_ID:
69 case IOAPIC_REG_ARB_ID:
70 result = ((ioapic->id & 0xf) << 24);
75 u32 redir_index = (ioapic->ioregsel - 0x10) >> 1;
78 if (redir_index < IOAPIC_NUM_PINS)
80 ioapic->redirtbl[redir_index].bits;
82 redir_content = ~0ULL;
84 result = (ioapic->ioregsel & 0x1) ?
85 (redir_content >> 32) & 0xffffffff :
86 redir_content & 0xffffffff;
94 static void rtc_irq_eoi_tracking_reset(struct kvm_ioapic *ioapic)
96 ioapic->rtc_status.pending_eoi = 0;
97 bitmap_zero(ioapic->rtc_status.dest_map.map, KVM_MAX_VCPU_ID);
100 static void kvm_rtc_eoi_tracking_restore_all(struct kvm_ioapic *ioapic);
102 static void rtc_status_pending_eoi_check_valid(struct kvm_ioapic *ioapic)
104 if (WARN_ON(ioapic->rtc_status.pending_eoi < 0))
105 kvm_rtc_eoi_tracking_restore_all(ioapic);
108 static void __rtc_irq_eoi_tracking_restore_one(struct kvm_vcpu *vcpu)
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;
115 e = &ioapic->redirtbl[RTC_GSI];
116 if (!kvm_apic_match_dest(vcpu, NULL, 0, e->fields.dest_id,
117 e->fields.dest_mode))
120 new_val = kvm_apic_pending_eoi(vcpu, e->fields.vector);
121 old_val = test_bit(vcpu->vcpu_id, dest_map->map);
123 if (new_val == old_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++;
131 __clear_bit(vcpu->vcpu_id, dest_map->map);
132 ioapic->rtc_status.pending_eoi--;
133 rtc_status_pending_eoi_check_valid(ioapic);
137 void kvm_rtc_eoi_tracking_restore_one(struct kvm_vcpu *vcpu)
139 struct kvm_ioapic *ioapic = vcpu->kvm->arch.vioapic;
141 spin_lock(&ioapic->lock);
142 __rtc_irq_eoi_tracking_restore_one(vcpu);
143 spin_unlock(&ioapic->lock);
146 static void kvm_rtc_eoi_tracking_restore_all(struct kvm_ioapic *ioapic)
148 struct kvm_vcpu *vcpu;
151 if (RTC_GSI >= IOAPIC_NUM_PINS)
154 rtc_irq_eoi_tracking_reset(ioapic);
155 kvm_for_each_vcpu(i, vcpu, ioapic->kvm)
156 __rtc_irq_eoi_tracking_restore_one(vcpu);
159 static void rtc_irq_eoi(struct kvm_ioapic *ioapic, struct kvm_vcpu *vcpu)
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);
168 static bool rtc_irq_check_coalesced(struct kvm_ioapic *ioapic)
170 if (ioapic->rtc_status.pending_eoi > 0)
171 return true; /* coalesced */
176 static int ioapic_set_irq(struct kvm_ioapic *ioapic, unsigned int irq,
177 int irq_level, bool line_status)
179 union kvm_ioapic_redirect_entry entry;
184 entry = ioapic->redirtbl[irq];
185 edge = (entry.fields.trig_mode == IOAPIC_EDGE_TRIG);
188 ioapic->irr &= ~mask;
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.
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.
204 if (irq == RTC_GSI && line_status &&
205 rtc_irq_check_coalesced(ioapic)) {
210 old_irr = ioapic->irr;
213 ioapic->irr_delivered &= ~mask;
214 if ((edge && old_irr == ioapic->irr) ||
215 (!edge && entry.fields.remote_irr)) {
220 ret = ioapic_service(ioapic, irq, line_status);
223 trace_kvm_ioapic_set_irq(entry.bits, irq, ret == 0);
227 static void kvm_ioapic_inject_all(struct kvm_ioapic *ioapic, unsigned long irr)
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);
235 kvm_rtc_eoi_tracking_restore_all(ioapic);
239 void kvm_ioapic_scan_entry(struct kvm_vcpu *vcpu, ulong *ioapic_handled_vectors)
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;
246 spin_lock(&ioapic->lock);
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);
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) ||
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);
265 spin_unlock(&ioapic->lock);
268 void kvm_arch_post_irq_ack_notifier_list_update(struct kvm *kvm)
270 if (!ioapic_in_kernel(kvm))
272 kvm_make_scan_ioapic_request(kvm);
275 static void ioapic_write_indirect(struct kvm_ioapic *ioapic, u32 val)
278 bool mask_before, mask_after;
279 union kvm_ioapic_redirect_entry *e;
281 switch (ioapic->ioregsel) {
282 case IOAPIC_REG_VERSION:
283 /* Writes are ignored. */
286 case IOAPIC_REG_APIC_ID:
287 ioapic->id = (val >> 24) & 0xf;
290 case IOAPIC_REG_ARB_ID:
294 index = (ioapic->ioregsel - 0x10) >> 1;
296 ioapic_debug("change redir index %x val %x\n", index, val);
297 if (index >= IOAPIC_NUM_PINS)
299 e = &ioapic->redirtbl[index];
300 mask_before = e->fields.mask;
301 if (ioapic->ioregsel & 1) {
302 e->bits &= 0xffffffff;
303 e->bits |= (u64) val << 32;
305 e->bits &= ~0xffffffffULL;
306 e->bits |= (u32) val;
307 e->fields.remote_irr = 0;
309 mask_after = e->fields.mask;
310 if (mask_before != mask_after)
311 kvm_fire_mask_notifiers(ioapic->kvm, KVM_IRQCHIP_IOAPIC, index, mask_after);
312 if (e->fields.trig_mode == IOAPIC_LEVEL_TRIG
313 && ioapic->irr & (1 << index))
314 ioapic_service(ioapic, index, false);
315 kvm_make_scan_ioapic_request(ioapic->kvm);
320 static int ioapic_service(struct kvm_ioapic *ioapic, int irq, bool line_status)
322 union kvm_ioapic_redirect_entry *entry = &ioapic->redirtbl[irq];
323 struct kvm_lapic_irq irqe;
326 if (entry->fields.mask)
329 ioapic_debug("dest=%x dest_mode=%x delivery_mode=%x "
330 "vector=%x trig_mode=%x\n",
331 entry->fields.dest_id, entry->fields.dest_mode,
332 entry->fields.delivery_mode, entry->fields.vector,
333 entry->fields.trig_mode);
335 irqe.dest_id = entry->fields.dest_id;
336 irqe.vector = entry->fields.vector;
337 irqe.dest_mode = entry->fields.dest_mode;
338 irqe.trig_mode = entry->fields.trig_mode;
339 irqe.delivery_mode = entry->fields.delivery_mode << 8;
342 irqe.msi_redir_hint = false;
344 if (irqe.trig_mode == IOAPIC_EDGE_TRIG)
345 ioapic->irr_delivered |= 1 << irq;
347 if (irq == RTC_GSI && line_status) {
349 * pending_eoi cannot ever become negative (see
350 * rtc_status_pending_eoi_check_valid) and the caller
351 * ensures that it is only called if it is >= zero, namely
352 * if rtc_irq_check_coalesced returns false).
354 BUG_ON(ioapic->rtc_status.pending_eoi != 0);
355 ret = kvm_irq_delivery_to_apic(ioapic->kvm, NULL, &irqe,
356 &ioapic->rtc_status.dest_map);
357 ioapic->rtc_status.pending_eoi = (ret < 0 ? 0 : ret);
359 ret = kvm_irq_delivery_to_apic(ioapic->kvm, NULL, &irqe, NULL);
361 if (ret && irqe.trig_mode == IOAPIC_LEVEL_TRIG)
362 entry->fields.remote_irr = 1;
367 int kvm_ioapic_set_irq(struct kvm_ioapic *ioapic, int irq, int irq_source_id,
368 int level, bool line_status)
372 BUG_ON(irq < 0 || irq >= IOAPIC_NUM_PINS);
374 spin_lock(&ioapic->lock);
375 irq_level = __kvm_irq_line_state(&ioapic->irq_states[irq],
376 irq_source_id, level);
377 ret = ioapic_set_irq(ioapic, irq, irq_level, line_status);
379 spin_unlock(&ioapic->lock);
384 void kvm_ioapic_clear_all(struct kvm_ioapic *ioapic, int irq_source_id)
388 spin_lock(&ioapic->lock);
389 for (i = 0; i < KVM_IOAPIC_NUM_PINS; i++)
390 __clear_bit(irq_source_id, &ioapic->irq_states[i]);
391 spin_unlock(&ioapic->lock);
394 static void kvm_ioapic_eoi_inject_work(struct work_struct *work)
397 struct kvm_ioapic *ioapic = container_of(work, struct kvm_ioapic,
399 spin_lock(&ioapic->lock);
400 for (i = 0; i < IOAPIC_NUM_PINS; i++) {
401 union kvm_ioapic_redirect_entry *ent = &ioapic->redirtbl[i];
403 if (ent->fields.trig_mode != IOAPIC_LEVEL_TRIG)
406 if (ioapic->irr & (1 << i) && !ent->fields.remote_irr)
407 ioapic_service(ioapic, i, false);
409 spin_unlock(&ioapic->lock);
412 #define IOAPIC_SUCCESSIVE_IRQ_MAX_COUNT 10000
414 static void __kvm_ioapic_update_eoi(struct kvm_vcpu *vcpu,
415 struct kvm_ioapic *ioapic, int vector, int trigger_mode)
417 struct dest_map *dest_map = &ioapic->rtc_status.dest_map;
418 struct kvm_lapic *apic = vcpu->arch.apic;
421 /* RTC special handling */
422 if (test_bit(vcpu->vcpu_id, dest_map->map) &&
423 vector == dest_map->vectors[vcpu->vcpu_id])
424 rtc_irq_eoi(ioapic, vcpu);
426 for (i = 0; i < IOAPIC_NUM_PINS; i++) {
427 union kvm_ioapic_redirect_entry *ent = &ioapic->redirtbl[i];
429 if (ent->fields.vector != vector)
433 * We are dropping lock while calling ack notifiers because ack
434 * notifier callbacks for assigned devices call into IOAPIC
435 * recursively. Since remote_irr is cleared only after call
436 * to notifiers if the same vector will be delivered while lock
437 * is dropped it will be put into irr and will be delivered
438 * after ack notifier returns.
440 spin_unlock(&ioapic->lock);
441 kvm_notify_acked_irq(ioapic->kvm, KVM_IRQCHIP_IOAPIC, i);
442 spin_lock(&ioapic->lock);
444 if (trigger_mode != IOAPIC_LEVEL_TRIG ||
445 kvm_lapic_get_reg(apic, APIC_SPIV) & APIC_SPIV_DIRECTED_EOI)
448 ASSERT(ent->fields.trig_mode == IOAPIC_LEVEL_TRIG);
449 ent->fields.remote_irr = 0;
450 if (!ent->fields.mask && (ioapic->irr & (1 << i))) {
451 ++ioapic->irq_eoi[i];
452 if (ioapic->irq_eoi[i] == IOAPIC_SUCCESSIVE_IRQ_MAX_COUNT) {
454 * Real hardware does not deliver the interrupt
455 * immediately during eoi broadcast, and this
456 * lets a buggy guest make slow progress
457 * even if it does not correctly handle a
458 * level-triggered interrupt. Emulate this
459 * behavior if we detect an interrupt storm.
461 schedule_delayed_work(&ioapic->eoi_inject, HZ / 100);
462 ioapic->irq_eoi[i] = 0;
463 trace_kvm_ioapic_delayed_eoi_inj(ent->bits);
465 ioapic_service(ioapic, i, false);
468 ioapic->irq_eoi[i] = 0;
473 void kvm_ioapic_update_eoi(struct kvm_vcpu *vcpu, int vector, int trigger_mode)
475 struct kvm_ioapic *ioapic = vcpu->kvm->arch.vioapic;
477 spin_lock(&ioapic->lock);
478 __kvm_ioapic_update_eoi(vcpu, ioapic, vector, trigger_mode);
479 spin_unlock(&ioapic->lock);
482 static inline struct kvm_ioapic *to_ioapic(struct kvm_io_device *dev)
484 return container_of(dev, struct kvm_ioapic, dev);
487 static inline int ioapic_in_range(struct kvm_ioapic *ioapic, gpa_t addr)
489 return ((addr >= ioapic->base_address &&
490 (addr < ioapic->base_address + IOAPIC_MEM_LENGTH)));
493 static int ioapic_mmio_read(struct kvm_vcpu *vcpu, struct kvm_io_device *this,
494 gpa_t addr, int len, void *val)
496 struct kvm_ioapic *ioapic = to_ioapic(this);
498 if (!ioapic_in_range(ioapic, addr))
501 ioapic_debug("addr %lx\n", (unsigned long)addr);
502 ASSERT(!(addr & 0xf)); /* check alignment */
505 spin_lock(&ioapic->lock);
507 case IOAPIC_REG_SELECT:
508 result = ioapic->ioregsel;
511 case IOAPIC_REG_WINDOW:
512 result = ioapic_read_indirect(ioapic, addr, len);
519 spin_unlock(&ioapic->lock);
523 *(u64 *) val = result;
528 memcpy(val, (char *)&result, len);
531 printk(KERN_WARNING "ioapic: wrong length %d\n", len);
536 static int ioapic_mmio_write(struct kvm_vcpu *vcpu, struct kvm_io_device *this,
537 gpa_t addr, int len, const void *val)
539 struct kvm_ioapic *ioapic = to_ioapic(this);
541 if (!ioapic_in_range(ioapic, addr))
544 ioapic_debug("ioapic_mmio_write addr=%p len=%d val=%p\n",
545 (void*)addr, len, val);
546 ASSERT(!(addr & 0xf)); /* check alignment */
560 printk(KERN_WARNING "ioapic: Unsupported size %d\n", len);
565 spin_lock(&ioapic->lock);
567 case IOAPIC_REG_SELECT:
568 ioapic->ioregsel = data & 0xFF; /* 8-bit register */
571 case IOAPIC_REG_WINDOW:
572 ioapic_write_indirect(ioapic, data);
578 spin_unlock(&ioapic->lock);
582 static void kvm_ioapic_reset(struct kvm_ioapic *ioapic)
586 cancel_delayed_work_sync(&ioapic->eoi_inject);
587 for (i = 0; i < IOAPIC_NUM_PINS; i++)
588 ioapic->redirtbl[i].fields.mask = 1;
589 ioapic->base_address = IOAPIC_DEFAULT_BASE_ADDRESS;
590 ioapic->ioregsel = 0;
592 ioapic->irr_delivered = 0;
594 memset(ioapic->irq_eoi, 0x00, sizeof(ioapic->irq_eoi));
595 rtc_irq_eoi_tracking_reset(ioapic);
598 static const struct kvm_io_device_ops ioapic_mmio_ops = {
599 .read = ioapic_mmio_read,
600 .write = ioapic_mmio_write,
603 int kvm_ioapic_init(struct kvm *kvm)
605 struct kvm_ioapic *ioapic;
608 ioapic = kzalloc(sizeof(struct kvm_ioapic), GFP_KERNEL);
611 spin_lock_init(&ioapic->lock);
612 INIT_DELAYED_WORK(&ioapic->eoi_inject, kvm_ioapic_eoi_inject_work);
613 kvm->arch.vioapic = ioapic;
614 kvm_ioapic_reset(ioapic);
615 kvm_iodevice_init(&ioapic->dev, &ioapic_mmio_ops);
617 mutex_lock(&kvm->slots_lock);
618 ret = kvm_io_bus_register_dev(kvm, KVM_MMIO_BUS, ioapic->base_address,
619 IOAPIC_MEM_LENGTH, &ioapic->dev);
620 mutex_unlock(&kvm->slots_lock);
622 kvm->arch.vioapic = NULL;
629 void kvm_ioapic_destroy(struct kvm *kvm)
631 struct kvm_ioapic *ioapic = kvm->arch.vioapic;
636 cancel_delayed_work_sync(&ioapic->eoi_inject);
637 mutex_lock(&kvm->slots_lock);
638 kvm_io_bus_unregister_dev(kvm, KVM_MMIO_BUS, &ioapic->dev);
639 mutex_unlock(&kvm->slots_lock);
640 kvm->arch.vioapic = NULL;
644 void kvm_get_ioapic(struct kvm *kvm, struct kvm_ioapic_state *state)
646 struct kvm_ioapic *ioapic = kvm->arch.vioapic;
648 spin_lock(&ioapic->lock);
649 memcpy(state, ioapic, sizeof(struct kvm_ioapic_state));
650 state->irr &= ~ioapic->irr_delivered;
651 spin_unlock(&ioapic->lock);
654 void kvm_set_ioapic(struct kvm *kvm, struct kvm_ioapic_state *state)
656 struct kvm_ioapic *ioapic = kvm->arch.vioapic;
658 spin_lock(&ioapic->lock);
659 memcpy(ioapic, state, sizeof(struct kvm_ioapic_state));
661 ioapic->irr_delivered = 0;
662 kvm_make_scan_ioapic_request(kvm);
663 kvm_ioapic_inject_all(ioapic, state->irr);
664 spin_unlock(&ioapic->lock);