Linux 4.1.18
[linux/fpc-iii.git] / arch / x86 / kvm / ioapic.c
blob28146f03c51421ce12f728d69613ded0a65699fd
1 /*
2 * Copyright (C) 2001 MandrakeSoft S.A.
3 * Copyright 2010 Red Hat, Inc. and/or its affiliates.
5 * MandrakeSoft S.A.
6 * 43, rue d'Aboukir
7 * 75002 Paris - France
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>
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 <asm/processor.h>
40 #include <asm/page.h>
41 #include <asm/current.h>
42 #include <trace/events/kvm.h>
44 #include "ioapic.h"
45 #include "lapic.h"
46 #include "irq.h"
48 #if 0
49 #define ioapic_debug(fmt,arg...) printk(KERN_WARNING fmt,##arg)
50 #else
51 #define ioapic_debug(fmt, arg...)
52 #endif
53 static int ioapic_service(struct kvm_ioapic *vioapic, int irq,
54 bool line_status);
56 static unsigned long ioapic_read_indirect(struct kvm_ioapic *ioapic,
57 unsigned long addr,
58 unsigned long length)
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));
66 break;
68 case IOAPIC_REG_APIC_ID:
69 case IOAPIC_REG_ARB_ID:
70 result = ((ioapic->id & 0xf) << 24);
71 break;
73 default:
75 u32 redir_index = (ioapic->ioregsel - 0x10) >> 1;
76 u64 redir_content;
78 if (redir_index < IOAPIC_NUM_PINS)
79 redir_content =
80 ioapic->redirtbl[redir_index].bits;
81 else
82 redir_content = ~0ULL;
84 result = (ioapic->ioregsel & 0x1) ?
85 (redir_content >> 32) & 0xffffffff :
86 redir_content & 0xffffffff;
87 break;
91 return result;
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))
117 return;
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)
123 return;
125 if (new_val) {
126 __set_bit(vcpu->vcpu_id, ioapic->rtc_status.dest_map);
127 ioapic->rtc_status.pending_eoi++;
128 } else {
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;
147 int i;
149 if (RTC_GSI >= IOAPIC_NUM_PINS)
150 return;
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 */
170 return false;
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;
177 u32 mask = 1 << irq;
178 u32 old_irr;
179 int edge, ret;
181 entry = ioapic->redirtbl[irq];
182 edge = (entry.fields.trig_mode == IOAPIC_EDGE_TRIG);
184 if (!irq_level) {
185 ioapic->irr &= ~mask;
186 ret = 1;
187 goto out;
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)) {
203 ret = 0;
204 goto out;
207 old_irr = ioapic->irr;
208 ioapic->irr |= mask;
209 if (edge)
210 ioapic->irr_delivered &= ~mask;
211 if ((edge && old_irr == ioapic->irr) ||
212 (!edge && entry.fields.remote_irr)) {
213 ret = 0;
214 goto out;
217 ret = ioapic_service(ioapic, irq, line_status);
219 out:
220 trace_kvm_ioapic_set_irq(entry.bits, irq, ret == 0);
221 return ret;
224 static void kvm_ioapic_inject_all(struct kvm_ioapic *ioapic, unsigned long irr)
226 u32 idx;
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);
239 int i;
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));
246 smp_wmb();
249 void kvm_ioapic_scan_entry(struct kvm_vcpu *vcpu, u64 *eoi_exit_bitmap,
250 u32 *tmr)
252 struct kvm_ioapic *ioapic = vcpu->kvm->arch.vioapic;
253 union kvm_ioapic_redirect_entry *e;
254 int index;
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) ||
261 index == RTC_GSI) {
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;
279 if (!ioapic)
280 return;
281 kvm_make_scan_ioapic_request(kvm);
284 static void ioapic_write_indirect(struct kvm_ioapic *ioapic, u32 val)
286 unsigned index;
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. */
293 break;
295 case IOAPIC_REG_APIC_ID:
296 ioapic->id = (val >> 24) & 0xf;
297 break;
299 case IOAPIC_REG_ARB_ID:
300 break;
302 default:
303 index = (ioapic->ioregsel - 0x10) >> 1;
305 ioapic_debug("change redir index %x val %x\n", index, val);
306 if (index >= IOAPIC_NUM_PINS)
307 return;
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;
313 } else {
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);
326 break;
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;
334 int ret;
336 if (entry->fields.mask)
337 return -1;
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;
350 irqe.level = 1;
351 irqe.shorthand = 0;
353 if (irqe.trig_mode == IOAPIC_EDGE_TRIG)
354 ioapic->irr_delivered |= 1 << irq;
356 if (irq == RTC_GSI && line_status) {
358 * pending_eoi cannot ever become negative (see
359 * rtc_status_pending_eoi_check_valid) and the caller
360 * ensures that it is only called if it is >= zero, namely
361 * if rtc_irq_check_coalesced returns false).
363 BUG_ON(ioapic->rtc_status.pending_eoi != 0);
364 ret = kvm_irq_delivery_to_apic(ioapic->kvm, NULL, &irqe,
365 ioapic->rtc_status.dest_map);
366 ioapic->rtc_status.pending_eoi = (ret < 0 ? 0 : ret);
367 } else
368 ret = kvm_irq_delivery_to_apic(ioapic->kvm, NULL, &irqe, NULL);
370 if (ret && irqe.trig_mode == IOAPIC_LEVEL_TRIG)
371 entry->fields.remote_irr = 1;
373 return ret;
376 int kvm_ioapic_set_irq(struct kvm_ioapic *ioapic, int irq, int irq_source_id,
377 int level, bool line_status)
379 int ret, irq_level;
381 BUG_ON(irq < 0 || irq >= IOAPIC_NUM_PINS);
383 spin_lock(&ioapic->lock);
384 irq_level = __kvm_irq_line_state(&ioapic->irq_states[irq],
385 irq_source_id, level);
386 ret = ioapic_set_irq(ioapic, irq, irq_level, line_status);
388 spin_unlock(&ioapic->lock);
390 return ret;
393 void kvm_ioapic_clear_all(struct kvm_ioapic *ioapic, int irq_source_id)
395 int i;
397 spin_lock(&ioapic->lock);
398 for (i = 0; i < KVM_IOAPIC_NUM_PINS; i++)
399 __clear_bit(irq_source_id, &ioapic->irq_states[i]);
400 spin_unlock(&ioapic->lock);
403 static void kvm_ioapic_eoi_inject_work(struct work_struct *work)
405 int i;
406 struct kvm_ioapic *ioapic = container_of(work, struct kvm_ioapic,
407 eoi_inject.work);
408 spin_lock(&ioapic->lock);
409 for (i = 0; i < IOAPIC_NUM_PINS; i++) {
410 union kvm_ioapic_redirect_entry *ent = &ioapic->redirtbl[i];
412 if (ent->fields.trig_mode != IOAPIC_LEVEL_TRIG)
413 continue;
415 if (ioapic->irr & (1 << i) && !ent->fields.remote_irr)
416 ioapic_service(ioapic, i, false);
418 spin_unlock(&ioapic->lock);
421 #define IOAPIC_SUCCESSIVE_IRQ_MAX_COUNT 10000
423 static void __kvm_ioapic_update_eoi(struct kvm_vcpu *vcpu,
424 struct kvm_ioapic *ioapic, int vector, int trigger_mode)
426 int i;
427 struct kvm_lapic *apic = vcpu->arch.apic;
429 for (i = 0; i < IOAPIC_NUM_PINS; i++) {
430 union kvm_ioapic_redirect_entry *ent = &ioapic->redirtbl[i];
432 if (ent->fields.vector != vector)
433 continue;
435 if (i == RTC_GSI)
436 rtc_irq_eoi(ioapic, vcpu);
438 * We are dropping lock while calling ack notifiers because ack
439 * notifier callbacks for assigned devices call into IOAPIC
440 * recursively. Since remote_irr is cleared only after call
441 * to notifiers if the same vector will be delivered while lock
442 * is dropped it will be put into irr and will be delivered
443 * after ack notifier returns.
445 spin_unlock(&ioapic->lock);
446 kvm_notify_acked_irq(ioapic->kvm, KVM_IRQCHIP_IOAPIC, i);
447 spin_lock(&ioapic->lock);
449 if (trigger_mode != IOAPIC_LEVEL_TRIG ||
450 kvm_apic_get_reg(apic, APIC_SPIV) & APIC_SPIV_DIRECTED_EOI)
451 continue;
453 ASSERT(ent->fields.trig_mode == IOAPIC_LEVEL_TRIG);
454 ent->fields.remote_irr = 0;
455 if (!ent->fields.mask && (ioapic->irr & (1 << i))) {
456 ++ioapic->irq_eoi[i];
457 if (ioapic->irq_eoi[i] == IOAPIC_SUCCESSIVE_IRQ_MAX_COUNT) {
459 * Real hardware does not deliver the interrupt
460 * immediately during eoi broadcast, and this
461 * lets a buggy guest make slow progress
462 * even if it does not correctly handle a
463 * level-triggered interrupt. Emulate this
464 * behavior if we detect an interrupt storm.
466 schedule_delayed_work(&ioapic->eoi_inject, HZ / 100);
467 ioapic->irq_eoi[i] = 0;
468 trace_kvm_ioapic_delayed_eoi_inj(ent->bits);
469 } else {
470 ioapic_service(ioapic, i, false);
472 } else {
473 ioapic->irq_eoi[i] = 0;
478 void kvm_ioapic_update_eoi(struct kvm_vcpu *vcpu, int vector, int trigger_mode)
480 struct kvm_ioapic *ioapic = vcpu->kvm->arch.vioapic;
482 spin_lock(&ioapic->lock);
483 __kvm_ioapic_update_eoi(vcpu, ioapic, vector, trigger_mode);
484 spin_unlock(&ioapic->lock);
487 static inline struct kvm_ioapic *to_ioapic(struct kvm_io_device *dev)
489 return container_of(dev, struct kvm_ioapic, dev);
492 static inline int ioapic_in_range(struct kvm_ioapic *ioapic, gpa_t addr)
494 return ((addr >= ioapic->base_address &&
495 (addr < ioapic->base_address + IOAPIC_MEM_LENGTH)));
498 static int ioapic_mmio_read(struct kvm_vcpu *vcpu, struct kvm_io_device *this,
499 gpa_t addr, int len, void *val)
501 struct kvm_ioapic *ioapic = to_ioapic(this);
502 u32 result;
503 if (!ioapic_in_range(ioapic, addr))
504 return -EOPNOTSUPP;
506 ioapic_debug("addr %lx\n", (unsigned long)addr);
507 ASSERT(!(addr & 0xf)); /* check alignment */
509 addr &= 0xff;
510 spin_lock(&ioapic->lock);
511 switch (addr) {
512 case IOAPIC_REG_SELECT:
513 result = ioapic->ioregsel;
514 break;
516 case IOAPIC_REG_WINDOW:
517 result = ioapic_read_indirect(ioapic, addr, len);
518 break;
520 default:
521 result = 0;
522 break;
524 spin_unlock(&ioapic->lock);
526 switch (len) {
527 case 8:
528 *(u64 *) val = result;
529 break;
530 case 1:
531 case 2:
532 case 4:
533 memcpy(val, (char *)&result, len);
534 break;
535 default:
536 printk(KERN_WARNING "ioapic: wrong length %d\n", len);
538 return 0;
541 static int ioapic_mmio_write(struct kvm_vcpu *vcpu, struct kvm_io_device *this,
542 gpa_t addr, int len, const void *val)
544 struct kvm_ioapic *ioapic = to_ioapic(this);
545 u32 data;
546 if (!ioapic_in_range(ioapic, addr))
547 return -EOPNOTSUPP;
549 ioapic_debug("ioapic_mmio_write addr=%p len=%d val=%p\n",
550 (void*)addr, len, val);
551 ASSERT(!(addr & 0xf)); /* check alignment */
553 switch (len) {
554 case 8:
555 case 4:
556 data = *(u32 *) val;
557 break;
558 case 2:
559 data = *(u16 *) val;
560 break;
561 case 1:
562 data = *(u8 *) val;
563 break;
564 default:
565 printk(KERN_WARNING "ioapic: Unsupported size %d\n", len);
566 return 0;
569 addr &= 0xff;
570 spin_lock(&ioapic->lock);
571 switch (addr) {
572 case IOAPIC_REG_SELECT:
573 ioapic->ioregsel = data & 0xFF; /* 8-bit register */
574 break;
576 case IOAPIC_REG_WINDOW:
577 ioapic_write_indirect(ioapic, data);
578 break;
580 default:
581 break;
583 spin_unlock(&ioapic->lock);
584 return 0;
587 static void kvm_ioapic_reset(struct kvm_ioapic *ioapic)
589 int i;
591 cancel_delayed_work_sync(&ioapic->eoi_inject);
592 for (i = 0; i < IOAPIC_NUM_PINS; i++)
593 ioapic->redirtbl[i].fields.mask = 1;
594 ioapic->base_address = IOAPIC_DEFAULT_BASE_ADDRESS;
595 ioapic->ioregsel = 0;
596 ioapic->irr = 0;
597 ioapic->irr_delivered = 0;
598 ioapic->id = 0;
599 memset(ioapic->irq_eoi, 0x00, IOAPIC_NUM_PINS);
600 rtc_irq_eoi_tracking_reset(ioapic);
601 update_handled_vectors(ioapic);
604 static const struct kvm_io_device_ops ioapic_mmio_ops = {
605 .read = ioapic_mmio_read,
606 .write = ioapic_mmio_write,
609 int kvm_ioapic_init(struct kvm *kvm)
611 struct kvm_ioapic *ioapic;
612 int ret;
614 ioapic = kzalloc(sizeof(struct kvm_ioapic), GFP_KERNEL);
615 if (!ioapic)
616 return -ENOMEM;
617 spin_lock_init(&ioapic->lock);
618 INIT_DELAYED_WORK(&ioapic->eoi_inject, kvm_ioapic_eoi_inject_work);
619 kvm->arch.vioapic = ioapic;
620 kvm_ioapic_reset(ioapic);
621 kvm_iodevice_init(&ioapic->dev, &ioapic_mmio_ops);
622 ioapic->kvm = kvm;
623 mutex_lock(&kvm->slots_lock);
624 ret = kvm_io_bus_register_dev(kvm, KVM_MMIO_BUS, ioapic->base_address,
625 IOAPIC_MEM_LENGTH, &ioapic->dev);
626 mutex_unlock(&kvm->slots_lock);
627 if (ret < 0) {
628 kvm->arch.vioapic = NULL;
629 kfree(ioapic);
632 return ret;
635 void kvm_ioapic_destroy(struct kvm *kvm)
637 struct kvm_ioapic *ioapic = kvm->arch.vioapic;
639 cancel_delayed_work_sync(&ioapic->eoi_inject);
640 if (ioapic) {
641 kvm_io_bus_unregister_dev(kvm, KVM_MMIO_BUS, &ioapic->dev);
642 kvm->arch.vioapic = NULL;
643 kfree(ioapic);
647 int kvm_get_ioapic(struct kvm *kvm, struct kvm_ioapic_state *state)
649 struct kvm_ioapic *ioapic = ioapic_irqchip(kvm);
650 if (!ioapic)
651 return -EINVAL;
653 spin_lock(&ioapic->lock);
654 memcpy(state, ioapic, sizeof(struct kvm_ioapic_state));
655 state->irr &= ~ioapic->irr_delivered;
656 spin_unlock(&ioapic->lock);
657 return 0;
660 int kvm_set_ioapic(struct kvm *kvm, struct kvm_ioapic_state *state)
662 struct kvm_ioapic *ioapic = ioapic_irqchip(kvm);
663 if (!ioapic)
664 return -EINVAL;
666 spin_lock(&ioapic->lock);
667 memcpy(ioapic, state, sizeof(struct kvm_ioapic_state));
668 ioapic->irr = 0;
669 ioapic->irr_delivered = 0;
670 update_handled_vectors(ioapic);
671 kvm_vcpu_request_scan_ioapic(kvm);
672 kvm_ioapic_inject_all(ioapic, state->irr);
673 spin_unlock(&ioapic->lock);
674 return 0;