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, KVM_MAX_VCPUS);
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 union kvm_ioapic_redirect_entry *e;
114 e = &ioapic->redirtbl[RTC_GSI];
115 if (!kvm_apic_match_dest(vcpu, NULL, 0, e->fields.dest_id,
116 e->fields.dest_mode))
119 new_val = kvm_apic_pending_eoi(vcpu, e->fields.vector);
120 old_val = test_bit(vcpu->vcpu_id, ioapic->rtc_status.dest_map);
122 if (new_val == old_val)
126 __set_bit(vcpu->vcpu_id, ioapic->rtc_status.dest_map);
127 ioapic->rtc_status.pending_eoi++;
129 __clear_bit(vcpu->vcpu_id, ioapic->rtc_status.dest_map);
130 ioapic->rtc_status.pending_eoi--;
131 rtc_status_pending_eoi_check_valid(ioapic);
135 void kvm_rtc_eoi_tracking_restore_one(struct kvm_vcpu *vcpu)
137 struct kvm_ioapic *ioapic = vcpu->kvm->arch.vioapic;
139 spin_lock(&ioapic->lock);
140 __rtc_irq_eoi_tracking_restore_one(vcpu);
141 spin_unlock(&ioapic->lock);
144 static void kvm_rtc_eoi_tracking_restore_all(struct kvm_ioapic *ioapic)
146 struct kvm_vcpu *vcpu;
149 if (RTC_GSI >= IOAPIC_NUM_PINS)
152 rtc_irq_eoi_tracking_reset(ioapic);
153 kvm_for_each_vcpu(i, vcpu, ioapic->kvm)
154 __rtc_irq_eoi_tracking_restore_one(vcpu);
157 static void rtc_irq_eoi(struct kvm_ioapic *ioapic, struct kvm_vcpu *vcpu)
159 if (test_and_clear_bit(vcpu->vcpu_id, ioapic->rtc_status.dest_map)) {
160 --ioapic->rtc_status.pending_eoi;
161 rtc_status_pending_eoi_check_valid(ioapic);
165 static bool rtc_irq_check_coalesced(struct kvm_ioapic *ioapic)
167 if (ioapic->rtc_status.pending_eoi > 0)
168 return true; /* coalesced */
173 static int ioapic_set_irq(struct kvm_ioapic *ioapic, unsigned int irq,
174 int irq_level, bool line_status)
176 union kvm_ioapic_redirect_entry entry;
181 entry = ioapic->redirtbl[irq];
182 edge = (entry.fields.trig_mode == IOAPIC_EDGE_TRIG);
185 ioapic->irr &= ~mask;
191 * Return 0 for coalesced interrupts; for edge-triggered interrupts,
192 * this only happens if a previous edge has not been delivered due
193 * do masking. For level interrupts, the remote_irr field tells
194 * us if the interrupt is waiting for an EOI.
196 * RTC is special: it is edge-triggered, but userspace likes to know
197 * if it has been already ack-ed via EOI because coalesced RTC
198 * interrupts lead to time drift in Windows guests. So we track
199 * EOI manually for the RTC interrupt.
201 if (irq == RTC_GSI && line_status &&
202 rtc_irq_check_coalesced(ioapic)) {
207 old_irr = ioapic->irr;
210 ioapic->irr_delivered &= ~mask;
211 if ((edge && old_irr == ioapic->irr) ||
212 (!edge && entry.fields.remote_irr)) {
217 ret = ioapic_service(ioapic, irq, line_status);
220 trace_kvm_ioapic_set_irq(entry.bits, irq, ret == 0);
224 static void kvm_ioapic_inject_all(struct kvm_ioapic *ioapic, unsigned long irr)
228 rtc_irq_eoi_tracking_reset(ioapic);
229 for_each_set_bit(idx, &irr, IOAPIC_NUM_PINS)
230 ioapic_set_irq(ioapic, idx, 1, true);
232 kvm_rtc_eoi_tracking_restore_all(ioapic);
236 static void update_handled_vectors(struct kvm_ioapic *ioapic)
238 DECLARE_BITMAP(handled_vectors, 256);
241 memset(handled_vectors, 0, sizeof(handled_vectors));
242 for (i = 0; i < IOAPIC_NUM_PINS; ++i)
243 __set_bit(ioapic->redirtbl[i].fields.vector, handled_vectors);
244 memcpy(ioapic->handled_vectors, handled_vectors,
245 sizeof(handled_vectors));
249 void kvm_ioapic_scan_entry(struct kvm_vcpu *vcpu, u64 *eoi_exit_bitmap,
252 struct kvm_ioapic *ioapic = vcpu->kvm->arch.vioapic;
253 union kvm_ioapic_redirect_entry *e;
256 spin_lock(&ioapic->lock);
257 for (index = 0; index < IOAPIC_NUM_PINS; index++) {
258 e = &ioapic->redirtbl[index];
259 if (e->fields.trig_mode == IOAPIC_LEVEL_TRIG ||
260 kvm_irq_has_notifier(ioapic->kvm, KVM_IRQCHIP_IOAPIC, index) ||
262 if (kvm_apic_match_dest(vcpu, NULL, 0,
263 e->fields.dest_id, e->fields.dest_mode)) {
264 __set_bit(e->fields.vector,
265 (unsigned long *)eoi_exit_bitmap);
266 if (e->fields.trig_mode == IOAPIC_LEVEL_TRIG)
267 __set_bit(e->fields.vector,
268 (unsigned long *)tmr);
272 spin_unlock(&ioapic->lock);
275 void kvm_vcpu_request_scan_ioapic(struct kvm *kvm)
277 struct kvm_ioapic *ioapic = kvm->arch.vioapic;
281 kvm_make_scan_ioapic_request(kvm);
284 static void ioapic_write_indirect(struct kvm_ioapic *ioapic, u32 val)
287 bool mask_before, mask_after;
288 union kvm_ioapic_redirect_entry *e;
290 switch (ioapic->ioregsel) {
291 case IOAPIC_REG_VERSION:
292 /* Writes are ignored. */
295 case IOAPIC_REG_APIC_ID:
296 ioapic->id = (val >> 24) & 0xf;
299 case IOAPIC_REG_ARB_ID:
303 index = (ioapic->ioregsel - 0x10) >> 1;
305 ioapic_debug("change redir index %x val %x\n", index, val);
306 if (index >= IOAPIC_NUM_PINS)
308 e = &ioapic->redirtbl[index];
309 mask_before = e->fields.mask;
310 if (ioapic->ioregsel & 1) {
311 e->bits &= 0xffffffff;
312 e->bits |= (u64) val << 32;
314 e->bits &= ~0xffffffffULL;
315 e->bits |= (u32) val;
316 e->fields.remote_irr = 0;
318 update_handled_vectors(ioapic);
319 mask_after = e->fields.mask;
320 if (mask_before != mask_after)
321 kvm_fire_mask_notifiers(ioapic->kvm, KVM_IRQCHIP_IOAPIC, index, mask_after);
322 if (e->fields.trig_mode == IOAPIC_LEVEL_TRIG
323 && ioapic->irr & (1 << index))
324 ioapic_service(ioapic, index, false);
325 kvm_vcpu_request_scan_ioapic(ioapic->kvm);
330 static int ioapic_service(struct kvm_ioapic *ioapic, int irq, bool line_status)
332 union kvm_ioapic_redirect_entry *entry = &ioapic->redirtbl[irq];
333 struct kvm_lapic_irq irqe;
336 if (entry->fields.mask)
339 ioapic_debug("dest=%x dest_mode=%x delivery_mode=%x "
340 "vector=%x trig_mode=%x\n",
341 entry->fields.dest_id, entry->fields.dest_mode,
342 entry->fields.delivery_mode, entry->fields.vector,
343 entry->fields.trig_mode);
345 irqe.dest_id = entry->fields.dest_id;
346 irqe.vector = entry->fields.vector;
347 irqe.dest_mode = entry->fields.dest_mode;
348 irqe.trig_mode = entry->fields.trig_mode;
349 irqe.delivery_mode = entry->fields.delivery_mode << 8;
352 irqe.msi_redir_hint = false;
354 if (irqe.trig_mode == IOAPIC_EDGE_TRIG)
355 ioapic->irr_delivered |= 1 << irq;
357 if (irq == RTC_GSI && line_status) {
359 * pending_eoi cannot ever become negative (see
360 * rtc_status_pending_eoi_check_valid) and the caller
361 * ensures that it is only called if it is >= zero, namely
362 * if rtc_irq_check_coalesced returns false).
364 BUG_ON(ioapic->rtc_status.pending_eoi != 0);
365 ret = kvm_irq_delivery_to_apic(ioapic->kvm, NULL, &irqe,
366 ioapic->rtc_status.dest_map);
367 ioapic->rtc_status.pending_eoi = (ret < 0 ? 0 : ret);
369 ret = kvm_irq_delivery_to_apic(ioapic->kvm, NULL, &irqe, NULL);
371 if (ret && irqe.trig_mode == IOAPIC_LEVEL_TRIG)
372 entry->fields.remote_irr = 1;
377 int kvm_ioapic_set_irq(struct kvm_ioapic *ioapic, int irq, int irq_source_id,
378 int level, bool line_status)
382 BUG_ON(irq < 0 || irq >= IOAPIC_NUM_PINS);
384 spin_lock(&ioapic->lock);
385 irq_level = __kvm_irq_line_state(&ioapic->irq_states[irq],
386 irq_source_id, level);
387 ret = ioapic_set_irq(ioapic, irq, irq_level, line_status);
389 spin_unlock(&ioapic->lock);
394 void kvm_ioapic_clear_all(struct kvm_ioapic *ioapic, int irq_source_id)
398 spin_lock(&ioapic->lock);
399 for (i = 0; i < KVM_IOAPIC_NUM_PINS; i++)
400 __clear_bit(irq_source_id, &ioapic->irq_states[i]);
401 spin_unlock(&ioapic->lock);
404 static void kvm_ioapic_eoi_inject_work(struct work_struct *work)
407 struct kvm_ioapic *ioapic = container_of(work, struct kvm_ioapic,
409 spin_lock(&ioapic->lock);
410 for (i = 0; i < IOAPIC_NUM_PINS; i++) {
411 union kvm_ioapic_redirect_entry *ent = &ioapic->redirtbl[i];
413 if (ent->fields.trig_mode != IOAPIC_LEVEL_TRIG)
416 if (ioapic->irr & (1 << i) && !ent->fields.remote_irr)
417 ioapic_service(ioapic, i, false);
419 spin_unlock(&ioapic->lock);
422 #define IOAPIC_SUCCESSIVE_IRQ_MAX_COUNT 10000
424 static void __kvm_ioapic_update_eoi(struct kvm_vcpu *vcpu,
425 struct kvm_ioapic *ioapic, int vector, int trigger_mode)
428 struct kvm_lapic *apic = vcpu->arch.apic;
430 for (i = 0; i < IOAPIC_NUM_PINS; i++) {
431 union kvm_ioapic_redirect_entry *ent = &ioapic->redirtbl[i];
433 if (ent->fields.vector != vector)
437 rtc_irq_eoi(ioapic, vcpu);
439 * We are dropping lock while calling ack notifiers because ack
440 * notifier callbacks for assigned devices call into IOAPIC
441 * recursively. Since remote_irr is cleared only after call
442 * to notifiers if the same vector will be delivered while lock
443 * is dropped it will be put into irr and will be delivered
444 * after ack notifier returns.
446 spin_unlock(&ioapic->lock);
447 kvm_notify_acked_irq(ioapic->kvm, KVM_IRQCHIP_IOAPIC, i);
448 spin_lock(&ioapic->lock);
450 if (trigger_mode != IOAPIC_LEVEL_TRIG ||
451 kvm_apic_get_reg(apic, APIC_SPIV) & APIC_SPIV_DIRECTED_EOI)
454 ASSERT(ent->fields.trig_mode == IOAPIC_LEVEL_TRIG);
455 ent->fields.remote_irr = 0;
456 if (!ent->fields.mask && (ioapic->irr & (1 << i))) {
457 ++ioapic->irq_eoi[i];
458 if (ioapic->irq_eoi[i] == IOAPIC_SUCCESSIVE_IRQ_MAX_COUNT) {
460 * Real hardware does not deliver the interrupt
461 * immediately during eoi broadcast, and this
462 * lets a buggy guest make slow progress
463 * even if it does not correctly handle a
464 * level-triggered interrupt. Emulate this
465 * behavior if we detect an interrupt storm.
467 schedule_delayed_work(&ioapic->eoi_inject, HZ / 100);
468 ioapic->irq_eoi[i] = 0;
469 trace_kvm_ioapic_delayed_eoi_inj(ent->bits);
471 ioapic_service(ioapic, i, false);
474 ioapic->irq_eoi[i] = 0;
479 void kvm_ioapic_update_eoi(struct kvm_vcpu *vcpu, int vector, int trigger_mode)
481 struct kvm_ioapic *ioapic = vcpu->kvm->arch.vioapic;
483 spin_lock(&ioapic->lock);
484 __kvm_ioapic_update_eoi(vcpu, ioapic, vector, trigger_mode);
485 spin_unlock(&ioapic->lock);
488 static inline struct kvm_ioapic *to_ioapic(struct kvm_io_device *dev)
490 return container_of(dev, struct kvm_ioapic, dev);
493 static inline int ioapic_in_range(struct kvm_ioapic *ioapic, gpa_t addr)
495 return ((addr >= ioapic->base_address &&
496 (addr < ioapic->base_address + IOAPIC_MEM_LENGTH)));
499 static int ioapic_mmio_read(struct kvm_vcpu *vcpu, struct kvm_io_device *this,
500 gpa_t addr, int len, void *val)
502 struct kvm_ioapic *ioapic = to_ioapic(this);
504 if (!ioapic_in_range(ioapic, addr))
507 ioapic_debug("addr %lx\n", (unsigned long)addr);
508 ASSERT(!(addr & 0xf)); /* check alignment */
511 spin_lock(&ioapic->lock);
513 case IOAPIC_REG_SELECT:
514 result = ioapic->ioregsel;
517 case IOAPIC_REG_WINDOW:
518 result = ioapic_read_indirect(ioapic, addr, len);
525 spin_unlock(&ioapic->lock);
529 *(u64 *) val = result;
534 memcpy(val, (char *)&result, len);
537 printk(KERN_WARNING "ioapic: wrong length %d\n", len);
542 static int ioapic_mmio_write(struct kvm_vcpu *vcpu, struct kvm_io_device *this,
543 gpa_t addr, int len, const void *val)
545 struct kvm_ioapic *ioapic = to_ioapic(this);
547 if (!ioapic_in_range(ioapic, addr))
550 ioapic_debug("ioapic_mmio_write addr=%p len=%d val=%p\n",
551 (void*)addr, len, val);
552 ASSERT(!(addr & 0xf)); /* check alignment */
566 printk(KERN_WARNING "ioapic: Unsupported size %d\n", len);
571 spin_lock(&ioapic->lock);
573 case IOAPIC_REG_SELECT:
574 ioapic->ioregsel = data & 0xFF; /* 8-bit register */
577 case IOAPIC_REG_WINDOW:
578 ioapic_write_indirect(ioapic, data);
584 spin_unlock(&ioapic->lock);
588 static void kvm_ioapic_reset(struct kvm_ioapic *ioapic)
592 cancel_delayed_work_sync(&ioapic->eoi_inject);
593 for (i = 0; i < IOAPIC_NUM_PINS; i++)
594 ioapic->redirtbl[i].fields.mask = 1;
595 ioapic->base_address = IOAPIC_DEFAULT_BASE_ADDRESS;
596 ioapic->ioregsel = 0;
598 ioapic->irr_delivered = 0;
600 memset(ioapic->irq_eoi, 0x00, IOAPIC_NUM_PINS);
601 rtc_irq_eoi_tracking_reset(ioapic);
602 update_handled_vectors(ioapic);
605 static const struct kvm_io_device_ops ioapic_mmio_ops = {
606 .read = ioapic_mmio_read,
607 .write = ioapic_mmio_write,
610 int kvm_ioapic_init(struct kvm *kvm)
612 struct kvm_ioapic *ioapic;
615 ioapic = kzalloc(sizeof(struct kvm_ioapic), GFP_KERNEL);
618 spin_lock_init(&ioapic->lock);
619 INIT_DELAYED_WORK(&ioapic->eoi_inject, kvm_ioapic_eoi_inject_work);
620 kvm->arch.vioapic = ioapic;
621 kvm_ioapic_reset(ioapic);
622 kvm_iodevice_init(&ioapic->dev, &ioapic_mmio_ops);
624 mutex_lock(&kvm->slots_lock);
625 ret = kvm_io_bus_register_dev(kvm, KVM_MMIO_BUS, ioapic->base_address,
626 IOAPIC_MEM_LENGTH, &ioapic->dev);
627 mutex_unlock(&kvm->slots_lock);
629 kvm->arch.vioapic = NULL;
636 void kvm_ioapic_destroy(struct kvm *kvm)
638 struct kvm_ioapic *ioapic = kvm->arch.vioapic;
640 cancel_delayed_work_sync(&ioapic->eoi_inject);
641 kvm_io_bus_unregister_dev(kvm, KVM_MMIO_BUS, &ioapic->dev);
642 kvm->arch.vioapic = NULL;
646 int kvm_get_ioapic(struct kvm *kvm, struct kvm_ioapic_state *state)
648 struct kvm_ioapic *ioapic = ioapic_irqchip(kvm);
652 spin_lock(&ioapic->lock);
653 memcpy(state, ioapic, sizeof(struct kvm_ioapic_state));
654 state->irr &= ~ioapic->irr_delivered;
655 spin_unlock(&ioapic->lock);
659 int kvm_set_ioapic(struct kvm *kvm, struct kvm_ioapic_state *state)
661 struct kvm_ioapic *ioapic = ioapic_irqchip(kvm);
665 spin_lock(&ioapic->lock);
666 memcpy(ioapic, state, sizeof(struct kvm_ioapic_state));
668 ioapic->irr_delivered = 0;
669 update_handled_vectors(ioapic);
670 kvm_vcpu_request_scan_ioapic(kvm);
671 kvm_ioapic_inject_all(ioapic, state->irr);
672 spin_unlock(&ioapic->lock);