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
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>
43
44#include "ioapic.h"
45#include "lapic.h"
46#include "irq.h"
47
48#if 0
49#define ioapic_debug(fmt,arg...) printk(KERN_WARNING fmt,##arg)
50#else
51#define ioapic_debug(fmt, arg...)
52#endif
53static int ioapic_service(struct kvm_ioapic *vioapic, int irq,
54 bool line_status);
55
56static unsigned long ioapic_read_indirect(struct kvm_ioapic *ioapic,
57 unsigned long addr,
58 unsigned long length)
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;
77
78 if (redir_index < IOAPIC_NUM_PINS)
79 redir_content =
80 ioapic->redirtbl[redir_index].bits;
81 else
82 redir_content = ~0ULL;
83
84 result = (ioapic->ioregsel & 0x1) ?
85 (redir_content >> 32) & 0xffffffff :
86 redir_content & 0xffffffff;
87 break;
88 }
89 }
90
91 return result;
92}
93
94static void rtc_irq_eoi_tracking_reset(struct kvm_ioapic *ioapic)
95{
96 ioapic->rtc_status.pending_eoi = 0;
97 bitmap_zero(ioapic->rtc_status.dest_map, KVM_MAX_VCPUS);
98}
99
100static void kvm_rtc_eoi_tracking_restore_all(struct kvm_ioapic *ioapic);
101
102static void rtc_status_pending_eoi_check_valid(struct kvm_ioapic *ioapic)
103{
104 if (WARN_ON(ioapic->rtc_status.pending_eoi < 0))
105 kvm_rtc_eoi_tracking_restore_all(ioapic);
106}
107
108static void __rtc_irq_eoi_tracking_restore_one(struct kvm_vcpu *vcpu)
109{
110 bool new_val, old_val;
111 struct kvm_ioapic *ioapic = vcpu->kvm->arch.vioapic;
112 union kvm_ioapic_redirect_entry *e;
113
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;
118
119 new_val = kvm_apic_pending_eoi(vcpu, e->fields.vector);
120 old_val = test_bit(vcpu->vcpu_id, ioapic->rtc_status.dest_map);
121
122 if (new_val == old_val)
123 return;
124
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);
132 }
133}
134
135void kvm_rtc_eoi_tracking_restore_one(struct kvm_vcpu *vcpu)
136{
137 struct kvm_ioapic *ioapic = vcpu->kvm->arch.vioapic;
138
139 spin_lock(&ioapic->lock);
140 __rtc_irq_eoi_tracking_restore_one(vcpu);
141 spin_unlock(&ioapic->lock);
142}
143
144static void kvm_rtc_eoi_tracking_restore_all(struct kvm_ioapic *ioapic)
145{
146 struct kvm_vcpu *vcpu;
147 int i;
148
149 if (RTC_GSI >= IOAPIC_NUM_PINS)
150 return;
151
152 rtc_irq_eoi_tracking_reset(ioapic);
153 kvm_for_each_vcpu(i, vcpu, ioapic->kvm)
154 __rtc_irq_eoi_tracking_restore_one(vcpu);
155}
156
157static void rtc_irq_eoi(struct kvm_ioapic *ioapic, struct kvm_vcpu *vcpu)
158{
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);
162 }
163}
164
165static bool rtc_irq_check_coalesced(struct kvm_ioapic *ioapic)
166{
167 if (ioapic->rtc_status.pending_eoi > 0)
168 return true; /* coalesced */
169
170 return false;
171}
172
173static int ioapic_set_irq(struct kvm_ioapic *ioapic, unsigned int irq,
174 int irq_level, bool line_status)
175{
176 union kvm_ioapic_redirect_entry entry;
177 u32 mask = 1 << irq;
178 u32 old_irr;
179 int edge, ret;
180
181 entry = ioapic->redirtbl[irq];
182 edge = (entry.fields.trig_mode == IOAPIC_EDGE_TRIG);
183
184 if (!irq_level) {
185 ioapic->irr &= ~mask;
186 ret = 1;
187 goto out;
188 }
189
190 /*
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.
195 *
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.
200 */
201 if (irq == RTC_GSI && line_status &&
202 rtc_irq_check_coalesced(ioapic)) {
203 ret = 0;
204 goto out;
205 }
206
207 old_irr = ioapic->irr;
208 ioapic->irr |= mask;
209 if ((edge && old_irr == ioapic->irr) ||
210 (!edge && entry.fields.remote_irr)) {
211 ret = 0;
212 goto out;
213 }
214
215 ret = ioapic_service(ioapic, irq, line_status);
216
217out:
218 trace_kvm_ioapic_set_irq(entry.bits, irq, ret == 0);
219 return ret;
220}
221
222static void kvm_ioapic_inject_all(struct kvm_ioapic *ioapic, unsigned long irr)
223{
224 u32 idx;
225
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);
229
230 kvm_rtc_eoi_tracking_restore_all(ioapic);
231}
232
233
234static void update_handled_vectors(struct kvm_ioapic *ioapic)
235{
236 DECLARE_BITMAP(handled_vectors, 256);
237 int i;
238
239 memset(handled_vectors, 0, sizeof(handled_vectors));
240 for (i = 0; i < IOAPIC_NUM_PINS; ++i)
241 __set_bit(ioapic->redirtbl[i].fields.vector, handled_vectors);
242 memcpy(ioapic->handled_vectors, handled_vectors,
243 sizeof(handled_vectors));
244 smp_wmb();
245}
246
247void kvm_ioapic_scan_entry(struct kvm_vcpu *vcpu, u64 *eoi_exit_bitmap,
248 u32 *tmr)
249{
250 struct kvm_ioapic *ioapic = vcpu->kvm->arch.vioapic;
251 union kvm_ioapic_redirect_entry *e;
252 int index;
253
254 spin_lock(&ioapic->lock);
255 for (index = 0; index < IOAPIC_NUM_PINS; index++) {
256 e = &ioapic->redirtbl[index];
257 if (!e->fields.mask &&
258 (e->fields.trig_mode == IOAPIC_LEVEL_TRIG ||
259 kvm_irq_has_notifier(ioapic->kvm, KVM_IRQCHIP_IOAPIC,
260 index) || index == RTC_GSI)) {
261 if (kvm_apic_match_dest(vcpu, NULL, 0,
262 e->fields.dest_id, e->fields.dest_mode)) {
263 __set_bit(e->fields.vector,
264 (unsigned long *)eoi_exit_bitmap);
265 if (e->fields.trig_mode == IOAPIC_LEVEL_TRIG)
266 __set_bit(e->fields.vector,
267 (unsigned long *)tmr);
268 }
269 }
270 }
271 spin_unlock(&ioapic->lock);
272}
273
274#ifdef CONFIG_X86
275void kvm_vcpu_request_scan_ioapic(struct kvm *kvm)
276{
277 struct kvm_ioapic *ioapic = kvm->arch.vioapic;
278
279 if (!ioapic)
280 return;
281 kvm_make_scan_ioapic_request(kvm);
282}
283#else
284void kvm_vcpu_request_scan_ioapic(struct kvm *kvm)
285{
286 return;
287}
288#endif
289
290static void ioapic_write_indirect(struct kvm_ioapic *ioapic, u32 val)
291{
292 unsigned index;
293 bool mask_before, mask_after;
294 union kvm_ioapic_redirect_entry *e;
295
296 switch (ioapic->ioregsel) {
297 case IOAPIC_REG_VERSION:
298 /* Writes are ignored. */
299 break;
300
301 case IOAPIC_REG_APIC_ID:
302 ioapic->id = (val >> 24) & 0xf;
303 break;
304
305 case IOAPIC_REG_ARB_ID:
306 break;
307
308 default:
309 index = (ioapic->ioregsel - 0x10) >> 1;
310
311 ioapic_debug("change redir index %x val %x\n", index, val);
312 if (index >= IOAPIC_NUM_PINS)
313 return;
314 e = &ioapic->redirtbl[index];
315 mask_before = e->fields.mask;
316 if (ioapic->ioregsel & 1) {
317 e->bits &= 0xffffffff;
318 e->bits |= (u64) val << 32;
319 } else {
320 e->bits &= ~0xffffffffULL;
321 e->bits |= (u32) val;
322 e->fields.remote_irr = 0;
323 }
324 update_handled_vectors(ioapic);
325 mask_after = e->fields.mask;
326 if (mask_before != mask_after)
327 kvm_fire_mask_notifiers(ioapic->kvm, KVM_IRQCHIP_IOAPIC, index, mask_after);
328 if (e->fields.trig_mode == IOAPIC_LEVEL_TRIG
329 && ioapic->irr & (1 << index))
330 ioapic_service(ioapic, index, false);
331 kvm_vcpu_request_scan_ioapic(ioapic->kvm);
332 break;
333 }
334}
335
336static int ioapic_service(struct kvm_ioapic *ioapic, int irq, bool line_status)
337{
338 union kvm_ioapic_redirect_entry *entry = &ioapic->redirtbl[irq];
339 struct kvm_lapic_irq irqe;
340 int ret;
341
342 if (entry->fields.mask)
343 return -1;
344
345 ioapic_debug("dest=%x dest_mode=%x delivery_mode=%x "
346 "vector=%x trig_mode=%x\n",
347 entry->fields.dest_id, entry->fields.dest_mode,
348 entry->fields.delivery_mode, entry->fields.vector,
349 entry->fields.trig_mode);
350
351 irqe.dest_id = entry->fields.dest_id;
352 irqe.vector = entry->fields.vector;
353 irqe.dest_mode = entry->fields.dest_mode;
354 irqe.trig_mode = entry->fields.trig_mode;
355 irqe.delivery_mode = entry->fields.delivery_mode << 8;
356 irqe.level = 1;
357 irqe.shorthand = 0;
358
359 if (irqe.trig_mode == IOAPIC_EDGE_TRIG)
360 ioapic->irr &= ~(1 << irq);
361
362 if (irq == RTC_GSI && line_status) {
363 /*
364 * pending_eoi cannot ever become negative (see
365 * rtc_status_pending_eoi_check_valid) and the caller
366 * ensures that it is only called if it is >= zero, namely
367 * if rtc_irq_check_coalesced returns false).
368 */
369 BUG_ON(ioapic->rtc_status.pending_eoi != 0);
370 ret = kvm_irq_delivery_to_apic(ioapic->kvm, NULL, &irqe,
371 ioapic->rtc_status.dest_map);
372 ioapic->rtc_status.pending_eoi = (ret < 0 ? 0 : ret);
373 } else
374 ret = kvm_irq_delivery_to_apic(ioapic->kvm, NULL, &irqe, NULL);
375
376 if (ret && irqe.trig_mode == IOAPIC_LEVEL_TRIG)
377 entry->fields.remote_irr = 1;
378
379 return ret;
380}
381
382int kvm_ioapic_set_irq(struct kvm_ioapic *ioapic, int irq, int irq_source_id,
383 int level, bool line_status)
384{
385 int ret, irq_level;
386
387 BUG_ON(irq < 0 || irq >= IOAPIC_NUM_PINS);
388
389 spin_lock(&ioapic->lock);
390 irq_level = __kvm_irq_line_state(&ioapic->irq_states[irq],
391 irq_source_id, level);
392 ret = ioapic_set_irq(ioapic, irq, irq_level, line_status);
393
394 spin_unlock(&ioapic->lock);
395
396 return ret;
397}
398
399void kvm_ioapic_clear_all(struct kvm_ioapic *ioapic, int irq_source_id)
400{
401 int i;
402
403 spin_lock(&ioapic->lock);
404 for (i = 0; i < KVM_IOAPIC_NUM_PINS; i++)
405 __clear_bit(irq_source_id, &ioapic->irq_states[i]);
406 spin_unlock(&ioapic->lock);
407}
408
409static void __kvm_ioapic_update_eoi(struct kvm_vcpu *vcpu,
410 struct kvm_ioapic *ioapic, int vector, int trigger_mode)
411{
412 int i;
413
414 for (i = 0; i < IOAPIC_NUM_PINS; i++) {
415 union kvm_ioapic_redirect_entry *ent = &ioapic->redirtbl[i];
416
417 if (ent->fields.vector != vector)
418 continue;
419
420 if (i == RTC_GSI)
421 rtc_irq_eoi(ioapic, vcpu);
422 /*
423 * We are dropping lock while calling ack notifiers because ack
424 * notifier callbacks for assigned devices call into IOAPIC
425 * recursively. Since remote_irr is cleared only after call
426 * to notifiers if the same vector will be delivered while lock
427 * is dropped it will be put into irr and will be delivered
428 * after ack notifier returns.
429 */
430 spin_unlock(&ioapic->lock);
431 kvm_notify_acked_irq(ioapic->kvm, KVM_IRQCHIP_IOAPIC, i);
432 spin_lock(&ioapic->lock);
433
434 if (trigger_mode != IOAPIC_LEVEL_TRIG)
435 continue;
436
437 ASSERT(ent->fields.trig_mode == IOAPIC_LEVEL_TRIG);
438 ent->fields.remote_irr = 0;
439 if (ioapic->irr & (1 << i))
440 ioapic_service(ioapic, i, false);
441 }
442}
443
444bool kvm_ioapic_handles_vector(struct kvm *kvm, int vector)
445{
446 struct kvm_ioapic *ioapic = kvm->arch.vioapic;
447 smp_rmb();
448 return test_bit(vector, ioapic->handled_vectors);
449}
450
451void kvm_ioapic_update_eoi(struct kvm_vcpu *vcpu, int vector, int trigger_mode)
452{
453 struct kvm_ioapic *ioapic = vcpu->kvm->arch.vioapic;
454
455 spin_lock(&ioapic->lock);
456 __kvm_ioapic_update_eoi(vcpu, ioapic, vector, trigger_mode);
457 spin_unlock(&ioapic->lock);
458}
459
460static inline struct kvm_ioapic *to_ioapic(struct kvm_io_device *dev)
461{
462 return container_of(dev, struct kvm_ioapic, dev);
463}
464
465static inline int ioapic_in_range(struct kvm_ioapic *ioapic, gpa_t addr)
466{
467 return ((addr >= ioapic->base_address &&
468 (addr < ioapic->base_address + IOAPIC_MEM_LENGTH)));
469}
470
471static int ioapic_mmio_read(struct kvm_io_device *this, gpa_t addr, int len,
472 void *val)
473{
474 struct kvm_ioapic *ioapic = to_ioapic(this);
475 u32 result;
476 if (!ioapic_in_range(ioapic, addr))
477 return -EOPNOTSUPP;
478
479 ioapic_debug("addr %lx\n", (unsigned long)addr);
480 ASSERT(!(addr & 0xf)); /* check alignment */
481
482 addr &= 0xff;
483 spin_lock(&ioapic->lock);
484 switch (addr) {
485 case IOAPIC_REG_SELECT:
486 result = ioapic->ioregsel;
487 break;
488
489 case IOAPIC_REG_WINDOW:
490 result = ioapic_read_indirect(ioapic, addr, len);
491 break;
492
493 default:
494 result = 0;
495 break;
496 }
497 spin_unlock(&ioapic->lock);
498
499 switch (len) {
500 case 8:
501 *(u64 *) val = result;
502 break;
503 case 1:
504 case 2:
505 case 4:
506 memcpy(val, (char *)&result, len);
507 break;
508 default:
509 printk(KERN_WARNING "ioapic: wrong length %d\n", len);
510 }
511 return 0;
512}
513
514static int ioapic_mmio_write(struct kvm_io_device *this, gpa_t addr, int len,
515 const void *val)
516{
517 struct kvm_ioapic *ioapic = to_ioapic(this);
518 u32 data;
519 if (!ioapic_in_range(ioapic, addr))
520 return -EOPNOTSUPP;
521
522 ioapic_debug("ioapic_mmio_write addr=%p len=%d val=%p\n",
523 (void*)addr, len, val);
524 ASSERT(!(addr & 0xf)); /* check alignment */
525
526 switch (len) {
527 case 8:
528 case 4:
529 data = *(u32 *) val;
530 break;
531 case 2:
532 data = *(u16 *) val;
533 break;
534 case 1:
535 data = *(u8 *) val;
536 break;
537 default:
538 printk(KERN_WARNING "ioapic: Unsupported size %d\n", len);
539 return 0;
540 }
541
542 addr &= 0xff;
543 spin_lock(&ioapic->lock);
544 switch (addr) {
545 case IOAPIC_REG_SELECT:
546 ioapic->ioregsel = data & 0xFF; /* 8-bit register */
547 break;
548
549 case IOAPIC_REG_WINDOW:
550 ioapic_write_indirect(ioapic, data);
551 break;
552#ifdef CONFIG_IA64
553 case IOAPIC_REG_EOI:
554 __kvm_ioapic_update_eoi(NULL, ioapic, data, IOAPIC_LEVEL_TRIG);
555 break;
556#endif
557
558 default:
559 break;
560 }
561 spin_unlock(&ioapic->lock);
562 return 0;
563}
564
565static void kvm_ioapic_reset(struct kvm_ioapic *ioapic)
566{
567 int i;
568
569 for (i = 0; i < IOAPIC_NUM_PINS; i++)
570 ioapic->redirtbl[i].fields.mask = 1;
571 ioapic->base_address = IOAPIC_DEFAULT_BASE_ADDRESS;
572 ioapic->ioregsel = 0;
573 ioapic->irr = 0;
574 ioapic->id = 0;
575 rtc_irq_eoi_tracking_reset(ioapic);
576 update_handled_vectors(ioapic);
577}
578
579static const struct kvm_io_device_ops ioapic_mmio_ops = {
580 .read = ioapic_mmio_read,
581 .write = ioapic_mmio_write,
582};
583
584int kvm_ioapic_init(struct kvm *kvm)
585{
586 struct kvm_ioapic *ioapic;
587 int ret;
588
589 ioapic = kzalloc(sizeof(struct kvm_ioapic), GFP_KERNEL);
590 if (!ioapic)
591 return -ENOMEM;
592 spin_lock_init(&ioapic->lock);
593 kvm->arch.vioapic = ioapic;
594 kvm_ioapic_reset(ioapic);
595 kvm_iodevice_init(&ioapic->dev, &ioapic_mmio_ops);
596 ioapic->kvm = kvm;
597 mutex_lock(&kvm->slots_lock);
598 ret = kvm_io_bus_register_dev(kvm, KVM_MMIO_BUS, ioapic->base_address,
599 IOAPIC_MEM_LENGTH, &ioapic->dev);
600 mutex_unlock(&kvm->slots_lock);
601 if (ret < 0) {
602 kvm->arch.vioapic = NULL;
603 kfree(ioapic);
604 }
605
606 return ret;
607}
608
609void kvm_ioapic_destroy(struct kvm *kvm)
610{
611 struct kvm_ioapic *ioapic = kvm->arch.vioapic;
612
613 if (ioapic) {
614 kvm_io_bus_unregister_dev(kvm, KVM_MMIO_BUS, &ioapic->dev);
615 kvm->arch.vioapic = NULL;
616 kfree(ioapic);
617 }
618}
619
620int kvm_get_ioapic(struct kvm *kvm, struct kvm_ioapic_state *state)
621{
622 struct kvm_ioapic *ioapic = ioapic_irqchip(kvm);
623 if (!ioapic)
624 return -EINVAL;
625
626 spin_lock(&ioapic->lock);
627 memcpy(state, ioapic, sizeof(struct kvm_ioapic_state));
628 spin_unlock(&ioapic->lock);
629 return 0;
630}
631
632int kvm_set_ioapic(struct kvm *kvm, struct kvm_ioapic_state *state)
633{
634 struct kvm_ioapic *ioapic = ioapic_irqchip(kvm);
635 if (!ioapic)
636 return -EINVAL;
637
638 spin_lock(&ioapic->lock);
639 memcpy(ioapic, state, sizeof(struct kvm_ioapic_state));
640 ioapic->irr = 0;
641 update_handled_vectors(ioapic);
642 kvm_vcpu_request_scan_ioapic(kvm);
643 kvm_ioapic_inject_all(ioapic, state->irr);
644 spin_unlock(&ioapic->lock);
645 return 0;
646}
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 <asm/processor.h>
39#include <asm/page.h>
40#include <asm/current.h>
41#include <trace/events/kvm.h>
42
43#include "ioapic.h"
44#include "lapic.h"
45#include "irq.h"
46
47#if 0
48#define ioapic_debug(fmt,arg...) printk(KERN_WARNING fmt,##arg)
49#else
50#define ioapic_debug(fmt, arg...)
51#endif
52static int ioapic_deliver(struct kvm_ioapic *vioapic, int irq);
53
54static unsigned long ioapic_read_indirect(struct kvm_ioapic *ioapic,
55 unsigned long addr,
56 unsigned long length)
57{
58 unsigned long result = 0;
59
60 switch (ioapic->ioregsel) {
61 case IOAPIC_REG_VERSION:
62 result = ((((IOAPIC_NUM_PINS - 1) & 0xff) << 16)
63 | (IOAPIC_VERSION_ID & 0xff));
64 break;
65
66 case IOAPIC_REG_APIC_ID:
67 case IOAPIC_REG_ARB_ID:
68 result = ((ioapic->id & 0xf) << 24);
69 break;
70
71 default:
72 {
73 u32 redir_index = (ioapic->ioregsel - 0x10) >> 1;
74 u64 redir_content;
75
76 ASSERT(redir_index < IOAPIC_NUM_PINS);
77
78 redir_content = ioapic->redirtbl[redir_index].bits;
79 result = (ioapic->ioregsel & 0x1) ?
80 (redir_content >> 32) & 0xffffffff :
81 redir_content & 0xffffffff;
82 break;
83 }
84 }
85
86 return result;
87}
88
89static int ioapic_service(struct kvm_ioapic *ioapic, unsigned int idx)
90{
91 union kvm_ioapic_redirect_entry *pent;
92 int injected = -1;
93
94 pent = &ioapic->redirtbl[idx];
95
96 if (!pent->fields.mask) {
97 injected = ioapic_deliver(ioapic, idx);
98 if (injected && pent->fields.trig_mode == IOAPIC_LEVEL_TRIG)
99 pent->fields.remote_irr = 1;
100 }
101
102 return injected;
103}
104
105static void update_handled_vectors(struct kvm_ioapic *ioapic)
106{
107 DECLARE_BITMAP(handled_vectors, 256);
108 int i;
109
110 memset(handled_vectors, 0, sizeof(handled_vectors));
111 for (i = 0; i < IOAPIC_NUM_PINS; ++i)
112 __set_bit(ioapic->redirtbl[i].fields.vector, handled_vectors);
113 memcpy(ioapic->handled_vectors, handled_vectors,
114 sizeof(handled_vectors));
115 smp_wmb();
116}
117
118static void ioapic_write_indirect(struct kvm_ioapic *ioapic, u32 val)
119{
120 unsigned index;
121 bool mask_before, mask_after;
122 union kvm_ioapic_redirect_entry *e;
123
124 switch (ioapic->ioregsel) {
125 case IOAPIC_REG_VERSION:
126 /* Writes are ignored. */
127 break;
128
129 case IOAPIC_REG_APIC_ID:
130 ioapic->id = (val >> 24) & 0xf;
131 break;
132
133 case IOAPIC_REG_ARB_ID:
134 break;
135
136 default:
137 index = (ioapic->ioregsel - 0x10) >> 1;
138
139 ioapic_debug("change redir index %x val %x\n", index, val);
140 if (index >= IOAPIC_NUM_PINS)
141 return;
142 e = &ioapic->redirtbl[index];
143 mask_before = e->fields.mask;
144 if (ioapic->ioregsel & 1) {
145 e->bits &= 0xffffffff;
146 e->bits |= (u64) val << 32;
147 } else {
148 e->bits &= ~0xffffffffULL;
149 e->bits |= (u32) val;
150 e->fields.remote_irr = 0;
151 }
152 update_handled_vectors(ioapic);
153 mask_after = e->fields.mask;
154 if (mask_before != mask_after)
155 kvm_fire_mask_notifiers(ioapic->kvm, KVM_IRQCHIP_IOAPIC, index, mask_after);
156 if (e->fields.trig_mode == IOAPIC_LEVEL_TRIG
157 && ioapic->irr & (1 << index))
158 ioapic_service(ioapic, index);
159 break;
160 }
161}
162
163static int ioapic_deliver(struct kvm_ioapic *ioapic, int irq)
164{
165 union kvm_ioapic_redirect_entry *entry = &ioapic->redirtbl[irq];
166 struct kvm_lapic_irq irqe;
167
168 ioapic_debug("dest=%x dest_mode=%x delivery_mode=%x "
169 "vector=%x trig_mode=%x\n",
170 entry->fields.dest_id, entry->fields.dest_mode,
171 entry->fields.delivery_mode, entry->fields.vector,
172 entry->fields.trig_mode);
173
174 irqe.dest_id = entry->fields.dest_id;
175 irqe.vector = entry->fields.vector;
176 irqe.dest_mode = entry->fields.dest_mode;
177 irqe.trig_mode = entry->fields.trig_mode;
178 irqe.delivery_mode = entry->fields.delivery_mode << 8;
179 irqe.level = 1;
180 irqe.shorthand = 0;
181
182#ifdef CONFIG_X86
183 /* Always delivery PIT interrupt to vcpu 0 */
184 if (irq == 0) {
185 irqe.dest_mode = 0; /* Physical mode. */
186 /* need to read apic_id from apic regiest since
187 * it can be rewritten */
188 irqe.dest_id = ioapic->kvm->bsp_vcpu_id;
189 }
190#endif
191 return kvm_irq_delivery_to_apic(ioapic->kvm, NULL, &irqe);
192}
193
194int kvm_ioapic_set_irq(struct kvm_ioapic *ioapic, int irq, int level)
195{
196 u32 old_irr;
197 u32 mask = 1 << irq;
198 union kvm_ioapic_redirect_entry entry;
199 int ret = 1;
200
201 spin_lock(&ioapic->lock);
202 old_irr = ioapic->irr;
203 if (irq >= 0 && irq < IOAPIC_NUM_PINS) {
204 entry = ioapic->redirtbl[irq];
205 level ^= entry.fields.polarity;
206 if (!level)
207 ioapic->irr &= ~mask;
208 else {
209 int edge = (entry.fields.trig_mode == IOAPIC_EDGE_TRIG);
210 ioapic->irr |= mask;
211 if ((edge && old_irr != ioapic->irr) ||
212 (!edge && !entry.fields.remote_irr))
213 ret = ioapic_service(ioapic, irq);
214 else
215 ret = 0; /* report coalesced interrupt */
216 }
217 trace_kvm_ioapic_set_irq(entry.bits, irq, ret == 0);
218 }
219 spin_unlock(&ioapic->lock);
220
221 return ret;
222}
223
224static void __kvm_ioapic_update_eoi(struct kvm_ioapic *ioapic, int vector,
225 int trigger_mode)
226{
227 int i;
228
229 for (i = 0; i < IOAPIC_NUM_PINS; i++) {
230 union kvm_ioapic_redirect_entry *ent = &ioapic->redirtbl[i];
231
232 if (ent->fields.vector != vector)
233 continue;
234
235 /*
236 * We are dropping lock while calling ack notifiers because ack
237 * notifier callbacks for assigned devices call into IOAPIC
238 * recursively. Since remote_irr is cleared only after call
239 * to notifiers if the same vector will be delivered while lock
240 * is dropped it will be put into irr and will be delivered
241 * after ack notifier returns.
242 */
243 spin_unlock(&ioapic->lock);
244 kvm_notify_acked_irq(ioapic->kvm, KVM_IRQCHIP_IOAPIC, i);
245 spin_lock(&ioapic->lock);
246
247 if (trigger_mode != IOAPIC_LEVEL_TRIG)
248 continue;
249
250 ASSERT(ent->fields.trig_mode == IOAPIC_LEVEL_TRIG);
251 ent->fields.remote_irr = 0;
252 if (!ent->fields.mask && (ioapic->irr & (1 << i)))
253 ioapic_service(ioapic, i);
254 }
255}
256
257bool kvm_ioapic_handles_vector(struct kvm *kvm, int vector)
258{
259 struct kvm_ioapic *ioapic = kvm->arch.vioapic;
260 smp_rmb();
261 return test_bit(vector, ioapic->handled_vectors);
262}
263
264void kvm_ioapic_update_eoi(struct kvm *kvm, int vector, int trigger_mode)
265{
266 struct kvm_ioapic *ioapic = kvm->arch.vioapic;
267
268 spin_lock(&ioapic->lock);
269 __kvm_ioapic_update_eoi(ioapic, vector, trigger_mode);
270 spin_unlock(&ioapic->lock);
271}
272
273static inline struct kvm_ioapic *to_ioapic(struct kvm_io_device *dev)
274{
275 return container_of(dev, struct kvm_ioapic, dev);
276}
277
278static inline int ioapic_in_range(struct kvm_ioapic *ioapic, gpa_t addr)
279{
280 return ((addr >= ioapic->base_address &&
281 (addr < ioapic->base_address + IOAPIC_MEM_LENGTH)));
282}
283
284static int ioapic_mmio_read(struct kvm_io_device *this, gpa_t addr, int len,
285 void *val)
286{
287 struct kvm_ioapic *ioapic = to_ioapic(this);
288 u32 result;
289 if (!ioapic_in_range(ioapic, addr))
290 return -EOPNOTSUPP;
291
292 ioapic_debug("addr %lx\n", (unsigned long)addr);
293 ASSERT(!(addr & 0xf)); /* check alignment */
294
295 addr &= 0xff;
296 spin_lock(&ioapic->lock);
297 switch (addr) {
298 case IOAPIC_REG_SELECT:
299 result = ioapic->ioregsel;
300 break;
301
302 case IOAPIC_REG_WINDOW:
303 result = ioapic_read_indirect(ioapic, addr, len);
304 break;
305
306 default:
307 result = 0;
308 break;
309 }
310 spin_unlock(&ioapic->lock);
311
312 switch (len) {
313 case 8:
314 *(u64 *) val = result;
315 break;
316 case 1:
317 case 2:
318 case 4:
319 memcpy(val, (char *)&result, len);
320 break;
321 default:
322 printk(KERN_WARNING "ioapic: wrong length %d\n", len);
323 }
324 return 0;
325}
326
327static int ioapic_mmio_write(struct kvm_io_device *this, gpa_t addr, int len,
328 const void *val)
329{
330 struct kvm_ioapic *ioapic = to_ioapic(this);
331 u32 data;
332 if (!ioapic_in_range(ioapic, addr))
333 return -EOPNOTSUPP;
334
335 ioapic_debug("ioapic_mmio_write addr=%p len=%d val=%p\n",
336 (void*)addr, len, val);
337 ASSERT(!(addr & 0xf)); /* check alignment */
338
339 switch (len) {
340 case 8:
341 case 4:
342 data = *(u32 *) val;
343 break;
344 case 2:
345 data = *(u16 *) val;
346 break;
347 case 1:
348 data = *(u8 *) val;
349 break;
350 default:
351 printk(KERN_WARNING "ioapic: Unsupported size %d\n", len);
352 return 0;
353 }
354
355 addr &= 0xff;
356 spin_lock(&ioapic->lock);
357 switch (addr) {
358 case IOAPIC_REG_SELECT:
359 ioapic->ioregsel = data & 0xFF; /* 8-bit register */
360 break;
361
362 case IOAPIC_REG_WINDOW:
363 ioapic_write_indirect(ioapic, data);
364 break;
365#ifdef CONFIG_IA64
366 case IOAPIC_REG_EOI:
367 __kvm_ioapic_update_eoi(ioapic, data, IOAPIC_LEVEL_TRIG);
368 break;
369#endif
370
371 default:
372 break;
373 }
374 spin_unlock(&ioapic->lock);
375 return 0;
376}
377
378void kvm_ioapic_reset(struct kvm_ioapic *ioapic)
379{
380 int i;
381
382 for (i = 0; i < IOAPIC_NUM_PINS; i++)
383 ioapic->redirtbl[i].fields.mask = 1;
384 ioapic->base_address = IOAPIC_DEFAULT_BASE_ADDRESS;
385 ioapic->ioregsel = 0;
386 ioapic->irr = 0;
387 ioapic->id = 0;
388 update_handled_vectors(ioapic);
389}
390
391static const struct kvm_io_device_ops ioapic_mmio_ops = {
392 .read = ioapic_mmio_read,
393 .write = ioapic_mmio_write,
394};
395
396int kvm_ioapic_init(struct kvm *kvm)
397{
398 struct kvm_ioapic *ioapic;
399 int ret;
400
401 ioapic = kzalloc(sizeof(struct kvm_ioapic), GFP_KERNEL);
402 if (!ioapic)
403 return -ENOMEM;
404 spin_lock_init(&ioapic->lock);
405 kvm->arch.vioapic = ioapic;
406 kvm_ioapic_reset(ioapic);
407 kvm_iodevice_init(&ioapic->dev, &ioapic_mmio_ops);
408 ioapic->kvm = kvm;
409 mutex_lock(&kvm->slots_lock);
410 ret = kvm_io_bus_register_dev(kvm, KVM_MMIO_BUS, ioapic->base_address,
411 IOAPIC_MEM_LENGTH, &ioapic->dev);
412 mutex_unlock(&kvm->slots_lock);
413 if (ret < 0) {
414 kvm->arch.vioapic = NULL;
415 kfree(ioapic);
416 }
417
418 return ret;
419}
420
421void kvm_ioapic_destroy(struct kvm *kvm)
422{
423 struct kvm_ioapic *ioapic = kvm->arch.vioapic;
424
425 if (ioapic) {
426 kvm_io_bus_unregister_dev(kvm, KVM_MMIO_BUS, &ioapic->dev);
427 kvm->arch.vioapic = NULL;
428 kfree(ioapic);
429 }
430}
431
432int kvm_get_ioapic(struct kvm *kvm, struct kvm_ioapic_state *state)
433{
434 struct kvm_ioapic *ioapic = ioapic_irqchip(kvm);
435 if (!ioapic)
436 return -EINVAL;
437
438 spin_lock(&ioapic->lock);
439 memcpy(state, ioapic, sizeof(struct kvm_ioapic_state));
440 spin_unlock(&ioapic->lock);
441 return 0;
442}
443
444int kvm_set_ioapic(struct kvm *kvm, struct kvm_ioapic_state *state)
445{
446 struct kvm_ioapic *ioapic = ioapic_irqchip(kvm);
447 if (!ioapic)
448 return -EINVAL;
449
450 spin_lock(&ioapic->lock);
451 memcpy(ioapic, state, sizeof(struct kvm_ioapic_state));
452 update_handled_vectors(ioapic);
453 spin_unlock(&ioapic->lock);
454 return 0;
455}