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>
48 static int ioapic_service(struct kvm_ioapic *vioapic, int irq,
51 static unsigned long ioapic_read_indirect(struct kvm_ioapic *ioapic,
55 unsigned long result = 0;
57 switch (ioapic->ioregsel) {
58 case IOAPIC_REG_VERSION:
59 result = ((((IOAPIC_NUM_PINS - 1) & 0xff) << 16)
60 | (IOAPIC_VERSION_ID & 0xff));
63 case IOAPIC_REG_APIC_ID:
64 case IOAPIC_REG_ARB_ID:
65 result = ((ioapic->id & 0xf) << 24);
70 u32 redir_index = (ioapic->ioregsel - 0x10) >> 1;
73 if (redir_index < IOAPIC_NUM_PINS)
75 ioapic->redirtbl[redir_index].bits;
77 redir_content = ~0ULL;
79 result = (ioapic->ioregsel & 0x1) ?
80 (redir_content >> 32) & 0xffffffff :
81 redir_content & 0xffffffff;
89 static void rtc_irq_eoi_tracking_reset(struct kvm_ioapic *ioapic)
91 ioapic->rtc_status.pending_eoi = 0;
92 bitmap_zero(ioapic->rtc_status.dest_map.map, KVM_MAX_VCPU_ID);
95 static void kvm_rtc_eoi_tracking_restore_all(struct kvm_ioapic *ioapic);
97 static void rtc_status_pending_eoi_check_valid(struct kvm_ioapic *ioapic)
99 if (WARN_ON(ioapic->rtc_status.pending_eoi < 0))
100 kvm_rtc_eoi_tracking_restore_all(ioapic);
103 static void __rtc_irq_eoi_tracking_restore_one(struct kvm_vcpu *vcpu)
105 bool new_val, old_val;
106 struct kvm_ioapic *ioapic = vcpu->kvm->arch.vioapic;
107 struct dest_map *dest_map = &ioapic->rtc_status.dest_map;
108 union kvm_ioapic_redirect_entry *e;
110 e = &ioapic->redirtbl[RTC_GSI];
111 if (!kvm_apic_match_dest(vcpu, NULL, 0, e->fields.dest_id,
112 e->fields.dest_mode))
115 new_val = kvm_apic_pending_eoi(vcpu, e->fields.vector);
116 old_val = test_bit(vcpu->vcpu_id, dest_map->map);
118 if (new_val == old_val)
122 __set_bit(vcpu->vcpu_id, dest_map->map);
123 dest_map->vectors[vcpu->vcpu_id] = e->fields.vector;
124 ioapic->rtc_status.pending_eoi++;
126 __clear_bit(vcpu->vcpu_id, dest_map->map);
127 ioapic->rtc_status.pending_eoi--;
128 rtc_status_pending_eoi_check_valid(ioapic);
132 void kvm_rtc_eoi_tracking_restore_one(struct kvm_vcpu *vcpu)
134 struct kvm_ioapic *ioapic = vcpu->kvm->arch.vioapic;
136 spin_lock(&ioapic->lock);
137 __rtc_irq_eoi_tracking_restore_one(vcpu);
138 spin_unlock(&ioapic->lock);
141 static void kvm_rtc_eoi_tracking_restore_all(struct kvm_ioapic *ioapic)
143 struct kvm_vcpu *vcpu;
146 if (RTC_GSI >= IOAPIC_NUM_PINS)
149 rtc_irq_eoi_tracking_reset(ioapic);
150 kvm_for_each_vcpu(i, vcpu, ioapic->kvm)
151 __rtc_irq_eoi_tracking_restore_one(vcpu);
154 static void rtc_irq_eoi(struct kvm_ioapic *ioapic, struct kvm_vcpu *vcpu)
156 if (test_and_clear_bit(vcpu->vcpu_id,
157 ioapic->rtc_status.dest_map.map)) {
158 --ioapic->rtc_status.pending_eoi;
159 rtc_status_pending_eoi_check_valid(ioapic);
163 static bool rtc_irq_check_coalesced(struct kvm_ioapic *ioapic)
165 if (ioapic->rtc_status.pending_eoi > 0)
166 return true; /* coalesced */
171 static int ioapic_set_irq(struct kvm_ioapic *ioapic, unsigned int irq,
172 int irq_level, bool line_status)
174 union kvm_ioapic_redirect_entry entry;
179 entry = ioapic->redirtbl[irq];
180 edge = (entry.fields.trig_mode == IOAPIC_EDGE_TRIG);
183 ioapic->irr &= ~mask;
189 * Return 0 for coalesced interrupts; for edge-triggered interrupts,
190 * this only happens if a previous edge has not been delivered due
191 * do masking. For level interrupts, the remote_irr field tells
192 * us if the interrupt is waiting for an EOI.
194 * RTC is special: it is edge-triggered, but userspace likes to know
195 * if it has been already ack-ed via EOI because coalesced RTC
196 * interrupts lead to time drift in Windows guests. So we track
197 * EOI manually for the RTC interrupt.
199 if (irq == RTC_GSI && line_status &&
200 rtc_irq_check_coalesced(ioapic)) {
205 old_irr = ioapic->irr;
208 ioapic->irr_delivered &= ~mask;
209 if (old_irr == ioapic->irr) {
215 ret = ioapic_service(ioapic, irq, line_status);
218 trace_kvm_ioapic_set_irq(entry.bits, irq, ret == 0);
222 static void kvm_ioapic_inject_all(struct kvm_ioapic *ioapic, unsigned long irr)
226 rtc_irq_eoi_tracking_reset(ioapic);
227 for_each_set_bit(idx, &irr, IOAPIC_NUM_PINS)
228 ioapic_set_irq(ioapic, idx, 1, true);
230 kvm_rtc_eoi_tracking_restore_all(ioapic);
234 void kvm_ioapic_scan_entry(struct kvm_vcpu *vcpu, ulong *ioapic_handled_vectors)
236 struct kvm_ioapic *ioapic = vcpu->kvm->arch.vioapic;
237 struct dest_map *dest_map = &ioapic->rtc_status.dest_map;
238 union kvm_ioapic_redirect_entry *e;
241 spin_lock(&ioapic->lock);
243 /* Make sure we see any missing RTC EOI */
244 if (test_bit(vcpu->vcpu_id, dest_map->map))
245 __set_bit(dest_map->vectors[vcpu->vcpu_id],
246 ioapic_handled_vectors);
248 for (index = 0; index < IOAPIC_NUM_PINS; index++) {
249 e = &ioapic->redirtbl[index];
250 if (e->fields.trig_mode == IOAPIC_LEVEL_TRIG ||
251 kvm_irq_has_notifier(ioapic->kvm, KVM_IRQCHIP_IOAPIC, index) ||
253 if (kvm_apic_match_dest(vcpu, NULL, 0,
254 e->fields.dest_id, e->fields.dest_mode) ||
255 kvm_apic_pending_eoi(vcpu, e->fields.vector))
256 __set_bit(e->fields.vector,
257 ioapic_handled_vectors);
260 spin_unlock(&ioapic->lock);
263 void kvm_arch_post_irq_ack_notifier_list_update(struct kvm *kvm)
265 if (!ioapic_in_kernel(kvm))
267 kvm_make_scan_ioapic_request(kvm);
270 static void ioapic_write_indirect(struct kvm_ioapic *ioapic, u32 val)
273 bool mask_before, mask_after;
274 union kvm_ioapic_redirect_entry *e;
275 unsigned long vcpu_bitmap;
276 int old_remote_irr, old_delivery_status, old_dest_id, old_dest_mode;
278 switch (ioapic->ioregsel) {
279 case IOAPIC_REG_VERSION:
280 /* Writes are ignored. */
283 case IOAPIC_REG_APIC_ID:
284 ioapic->id = (val >> 24) & 0xf;
287 case IOAPIC_REG_ARB_ID:
291 index = (ioapic->ioregsel - 0x10) >> 1;
293 if (index >= IOAPIC_NUM_PINS)
295 e = &ioapic->redirtbl[index];
296 mask_before = e->fields.mask;
297 /* Preserve read-only fields */
298 old_remote_irr = e->fields.remote_irr;
299 old_delivery_status = e->fields.delivery_status;
300 old_dest_id = e->fields.dest_id;
301 old_dest_mode = e->fields.dest_mode;
302 if (ioapic->ioregsel & 1) {
303 e->bits &= 0xffffffff;
304 e->bits |= (u64) val << 32;
306 e->bits &= ~0xffffffffULL;
307 e->bits |= (u32) val;
309 e->fields.remote_irr = old_remote_irr;
310 e->fields.delivery_status = old_delivery_status;
313 * Some OSes (Linux, Xen) assume that Remote IRR bit will
314 * be cleared by IOAPIC hardware when the entry is configured
315 * as edge-triggered. This behavior is used to simulate an
316 * explicit EOI on IOAPICs that don't have the EOI register.
318 if (e->fields.trig_mode == IOAPIC_EDGE_TRIG)
319 e->fields.remote_irr = 0;
321 mask_after = e->fields.mask;
322 if (mask_before != mask_after)
323 kvm_fire_mask_notifiers(ioapic->kvm, KVM_IRQCHIP_IOAPIC, index, mask_after);
324 if (e->fields.trig_mode == IOAPIC_LEVEL_TRIG
325 && ioapic->irr & (1 << index))
326 ioapic_service(ioapic, index, false);
327 if (e->fields.delivery_mode == APIC_DM_FIXED) {
328 struct kvm_lapic_irq irq;
331 irq.vector = e->fields.vector;
332 irq.delivery_mode = e->fields.delivery_mode << 8;
333 irq.dest_id = e->fields.dest_id;
334 irq.dest_mode = e->fields.dest_mode;
335 bitmap_zero(&vcpu_bitmap, 16);
336 kvm_bitmap_or_dest_vcpus(ioapic->kvm, &irq,
338 if (old_dest_mode != e->fields.dest_mode ||
339 old_dest_id != e->fields.dest_id) {
341 * Update vcpu_bitmap with vcpus specified in
342 * the previous request as well. This is done to
343 * keep ioapic_handled_vectors synchronized.
345 irq.dest_id = old_dest_id;
346 irq.dest_mode = old_dest_mode;
347 kvm_bitmap_or_dest_vcpus(ioapic->kvm, &irq,
350 kvm_make_scan_ioapic_request_mask(ioapic->kvm,
353 kvm_make_scan_ioapic_request(ioapic->kvm);
359 static int ioapic_service(struct kvm_ioapic *ioapic, int irq, bool line_status)
361 union kvm_ioapic_redirect_entry *entry = &ioapic->redirtbl[irq];
362 struct kvm_lapic_irq irqe;
365 if (entry->fields.mask ||
366 (entry->fields.trig_mode == IOAPIC_LEVEL_TRIG &&
367 entry->fields.remote_irr))
370 irqe.dest_id = entry->fields.dest_id;
371 irqe.vector = entry->fields.vector;
372 irqe.dest_mode = entry->fields.dest_mode;
373 irqe.trig_mode = entry->fields.trig_mode;
374 irqe.delivery_mode = entry->fields.delivery_mode << 8;
377 irqe.msi_redir_hint = false;
379 if (irqe.trig_mode == IOAPIC_EDGE_TRIG)
380 ioapic->irr_delivered |= 1 << irq;
382 if (irq == RTC_GSI && line_status) {
384 * pending_eoi cannot ever become negative (see
385 * rtc_status_pending_eoi_check_valid) and the caller
386 * ensures that it is only called if it is >= zero, namely
387 * if rtc_irq_check_coalesced returns false).
389 BUG_ON(ioapic->rtc_status.pending_eoi != 0);
390 ret = kvm_irq_delivery_to_apic(ioapic->kvm, NULL, &irqe,
391 &ioapic->rtc_status.dest_map);
392 ioapic->rtc_status.pending_eoi = (ret < 0 ? 0 : ret);
394 ret = kvm_irq_delivery_to_apic(ioapic->kvm, NULL, &irqe, NULL);
396 if (ret && irqe.trig_mode == IOAPIC_LEVEL_TRIG)
397 entry->fields.remote_irr = 1;
402 int kvm_ioapic_set_irq(struct kvm_ioapic *ioapic, int irq, int irq_source_id,
403 int level, bool line_status)
407 BUG_ON(irq < 0 || irq >= IOAPIC_NUM_PINS);
409 spin_lock(&ioapic->lock);
410 irq_level = __kvm_irq_line_state(&ioapic->irq_states[irq],
411 irq_source_id, level);
412 ret = ioapic_set_irq(ioapic, irq, irq_level, line_status);
414 spin_unlock(&ioapic->lock);
419 void kvm_ioapic_clear_all(struct kvm_ioapic *ioapic, int irq_source_id)
423 spin_lock(&ioapic->lock);
424 for (i = 0; i < KVM_IOAPIC_NUM_PINS; i++)
425 __clear_bit(irq_source_id, &ioapic->irq_states[i]);
426 spin_unlock(&ioapic->lock);
429 static void kvm_ioapic_eoi_inject_work(struct work_struct *work)
432 struct kvm_ioapic *ioapic = container_of(work, struct kvm_ioapic,
434 spin_lock(&ioapic->lock);
435 for (i = 0; i < IOAPIC_NUM_PINS; i++) {
436 union kvm_ioapic_redirect_entry *ent = &ioapic->redirtbl[i];
438 if (ent->fields.trig_mode != IOAPIC_LEVEL_TRIG)
441 if (ioapic->irr & (1 << i) && !ent->fields.remote_irr)
442 ioapic_service(ioapic, i, false);
444 spin_unlock(&ioapic->lock);
447 #define IOAPIC_SUCCESSIVE_IRQ_MAX_COUNT 10000
449 static void __kvm_ioapic_update_eoi(struct kvm_vcpu *vcpu,
450 struct kvm_ioapic *ioapic, int vector, int trigger_mode)
452 struct dest_map *dest_map = &ioapic->rtc_status.dest_map;
453 struct kvm_lapic *apic = vcpu->arch.apic;
456 /* RTC special handling */
457 if (test_bit(vcpu->vcpu_id, dest_map->map) &&
458 vector == dest_map->vectors[vcpu->vcpu_id])
459 rtc_irq_eoi(ioapic, vcpu);
461 for (i = 0; i < IOAPIC_NUM_PINS; i++) {
462 union kvm_ioapic_redirect_entry *ent = &ioapic->redirtbl[i];
464 if (ent->fields.vector != vector)
468 * We are dropping lock while calling ack notifiers because ack
469 * notifier callbacks for assigned devices call into IOAPIC
470 * recursively. Since remote_irr is cleared only after call
471 * to notifiers if the same vector will be delivered while lock
472 * is dropped it will be put into irr and will be delivered
473 * after ack notifier returns.
475 spin_unlock(&ioapic->lock);
476 kvm_notify_acked_irq(ioapic->kvm, KVM_IRQCHIP_IOAPIC, i);
477 spin_lock(&ioapic->lock);
479 if (trigger_mode != IOAPIC_LEVEL_TRIG ||
480 kvm_lapic_get_reg(apic, APIC_SPIV) & APIC_SPIV_DIRECTED_EOI)
483 ASSERT(ent->fields.trig_mode == IOAPIC_LEVEL_TRIG);
484 ent->fields.remote_irr = 0;
485 if (!ent->fields.mask && (ioapic->irr & (1 << i))) {
486 ++ioapic->irq_eoi[i];
487 if (ioapic->irq_eoi[i] == IOAPIC_SUCCESSIVE_IRQ_MAX_COUNT) {
489 * Real hardware does not deliver the interrupt
490 * immediately during eoi broadcast, and this
491 * lets a buggy guest make slow progress
492 * even if it does not correctly handle a
493 * level-triggered interrupt. Emulate this
494 * behavior if we detect an interrupt storm.
496 schedule_delayed_work(&ioapic->eoi_inject, HZ / 100);
497 ioapic->irq_eoi[i] = 0;
498 trace_kvm_ioapic_delayed_eoi_inj(ent->bits);
500 ioapic_service(ioapic, i, false);
503 ioapic->irq_eoi[i] = 0;
508 void kvm_ioapic_update_eoi(struct kvm_vcpu *vcpu, int vector, int trigger_mode)
510 struct kvm_ioapic *ioapic = vcpu->kvm->arch.vioapic;
512 spin_lock(&ioapic->lock);
513 __kvm_ioapic_update_eoi(vcpu, ioapic, vector, trigger_mode);
514 spin_unlock(&ioapic->lock);
517 static inline struct kvm_ioapic *to_ioapic(struct kvm_io_device *dev)
519 return container_of(dev, struct kvm_ioapic, dev);
522 static inline int ioapic_in_range(struct kvm_ioapic *ioapic, gpa_t addr)
524 return ((addr >= ioapic->base_address &&
525 (addr < ioapic->base_address + IOAPIC_MEM_LENGTH)));
528 static int ioapic_mmio_read(struct kvm_vcpu *vcpu, struct kvm_io_device *this,
529 gpa_t addr, int len, void *val)
531 struct kvm_ioapic *ioapic = to_ioapic(this);
533 if (!ioapic_in_range(ioapic, addr))
536 ASSERT(!(addr & 0xf)); /* check alignment */
539 spin_lock(&ioapic->lock);
541 case IOAPIC_REG_SELECT:
542 result = ioapic->ioregsel;
545 case IOAPIC_REG_WINDOW:
546 result = ioapic_read_indirect(ioapic, addr, len);
553 spin_unlock(&ioapic->lock);
557 *(u64 *) val = result;
562 memcpy(val, (char *)&result, len);
565 printk(KERN_WARNING "ioapic: wrong length %d\n", len);
570 static int ioapic_mmio_write(struct kvm_vcpu *vcpu, struct kvm_io_device *this,
571 gpa_t addr, int len, const void *val)
573 struct kvm_ioapic *ioapic = to_ioapic(this);
575 if (!ioapic_in_range(ioapic, addr))
578 ASSERT(!(addr & 0xf)); /* check alignment */
592 printk(KERN_WARNING "ioapic: Unsupported size %d\n", len);
597 spin_lock(&ioapic->lock);
599 case IOAPIC_REG_SELECT:
600 ioapic->ioregsel = data & 0xFF; /* 8-bit register */
603 case IOAPIC_REG_WINDOW:
604 ioapic_write_indirect(ioapic, data);
610 spin_unlock(&ioapic->lock);
614 static void kvm_ioapic_reset(struct kvm_ioapic *ioapic)
618 cancel_delayed_work_sync(&ioapic->eoi_inject);
619 for (i = 0; i < IOAPIC_NUM_PINS; i++)
620 ioapic->redirtbl[i].fields.mask = 1;
621 ioapic->base_address = IOAPIC_DEFAULT_BASE_ADDRESS;
622 ioapic->ioregsel = 0;
624 ioapic->irr_delivered = 0;
626 memset(ioapic->irq_eoi, 0x00, sizeof(ioapic->irq_eoi));
627 rtc_irq_eoi_tracking_reset(ioapic);
630 static const struct kvm_io_device_ops ioapic_mmio_ops = {
631 .read = ioapic_mmio_read,
632 .write = ioapic_mmio_write,
635 int kvm_ioapic_init(struct kvm *kvm)
637 struct kvm_ioapic *ioapic;
640 ioapic = kzalloc(sizeof(struct kvm_ioapic), GFP_KERNEL_ACCOUNT);
643 spin_lock_init(&ioapic->lock);
644 INIT_DELAYED_WORK(&ioapic->eoi_inject, kvm_ioapic_eoi_inject_work);
645 kvm->arch.vioapic = ioapic;
646 kvm_ioapic_reset(ioapic);
647 kvm_iodevice_init(&ioapic->dev, &ioapic_mmio_ops);
649 mutex_lock(&kvm->slots_lock);
650 ret = kvm_io_bus_register_dev(kvm, KVM_MMIO_BUS, ioapic->base_address,
651 IOAPIC_MEM_LENGTH, &ioapic->dev);
652 mutex_unlock(&kvm->slots_lock);
654 kvm->arch.vioapic = NULL;
661 void kvm_ioapic_destroy(struct kvm *kvm)
663 struct kvm_ioapic *ioapic = kvm->arch.vioapic;
668 cancel_delayed_work_sync(&ioapic->eoi_inject);
669 mutex_lock(&kvm->slots_lock);
670 kvm_io_bus_unregister_dev(kvm, KVM_MMIO_BUS, &ioapic->dev);
671 mutex_unlock(&kvm->slots_lock);
672 kvm->arch.vioapic = NULL;
676 void kvm_get_ioapic(struct kvm *kvm, struct kvm_ioapic_state *state)
678 struct kvm_ioapic *ioapic = kvm->arch.vioapic;
680 spin_lock(&ioapic->lock);
681 memcpy(state, ioapic, sizeof(struct kvm_ioapic_state));
682 state->irr &= ~ioapic->irr_delivered;
683 spin_unlock(&ioapic->lock);
686 void kvm_set_ioapic(struct kvm *kvm, struct kvm_ioapic_state *state)
688 struct kvm_ioapic *ioapic = kvm->arch.vioapic;
690 spin_lock(&ioapic->lock);
691 memcpy(ioapic, state, sizeof(struct kvm_ioapic_state));
693 ioapic->irr_delivered = 0;
694 kvm_make_scan_ioapic_request(kvm);
695 kvm_ioapic_inject_all(ioapic, state->irr);
696 spin_unlock(&ioapic->lock);