]> git.proxmox.com Git - mirror_ubuntu-artful-kernel.git/blob - arch/x86/kvm/lapic.c
KVM: change the way how lowest priority vcpu is calculated
[mirror_ubuntu-artful-kernel.git] / arch / x86 / kvm / lapic.c
1
2 /*
3 * Local APIC virtualization
4 *
5 * Copyright (C) 2006 Qumranet, Inc.
6 * Copyright (C) 2007 Novell
7 * Copyright (C) 2007 Intel
8 *
9 * Authors:
10 * Dor Laor <dor.laor@qumranet.com>
11 * Gregory Haskins <ghaskins@novell.com>
12 * Yaozu (Eddie) Dong <eddie.dong@intel.com>
13 *
14 * Based on Xen 3.1 code, Copyright (c) 2004, Intel Corporation.
15 *
16 * This work is licensed under the terms of the GNU GPL, version 2. See
17 * the COPYING file in the top-level directory.
18 */
19
20 #include <linux/kvm_host.h>
21 #include <linux/kvm.h>
22 #include <linux/mm.h>
23 #include <linux/highmem.h>
24 #include <linux/smp.h>
25 #include <linux/hrtimer.h>
26 #include <linux/io.h>
27 #include <linux/module.h>
28 #include <linux/math64.h>
29 #include <asm/processor.h>
30 #include <asm/msr.h>
31 #include <asm/page.h>
32 #include <asm/current.h>
33 #include <asm/apicdef.h>
34 #include <asm/atomic.h>
35 #include "kvm_cache_regs.h"
36 #include "irq.h"
37
38 #ifndef CONFIG_X86_64
39 #define mod_64(x, y) ((x) - (y) * div64_u64(x, y))
40 #else
41 #define mod_64(x, y) ((x) % (y))
42 #endif
43
44 #define PRId64 "d"
45 #define PRIx64 "llx"
46 #define PRIu64 "u"
47 #define PRIo64 "o"
48
49 #define APIC_BUS_CYCLE_NS 1
50
51 /* #define apic_debug(fmt,arg...) printk(KERN_WARNING fmt,##arg) */
52 #define apic_debug(fmt, arg...)
53
54 #define APIC_LVT_NUM 6
55 /* 14 is the version for Xeon and Pentium 8.4.8*/
56 #define APIC_VERSION (0x14UL | ((APIC_LVT_NUM - 1) << 16))
57 #define LAPIC_MMIO_LENGTH (1 << 12)
58 /* followed define is not in apicdef.h */
59 #define APIC_SHORT_MASK 0xc0000
60 #define APIC_DEST_NOSHORT 0x0
61 #define APIC_DEST_MASK 0x800
62 #define MAX_APIC_VECTOR 256
63
64 #define VEC_POS(v) ((v) & (32 - 1))
65 #define REG_POS(v) (((v) >> 5) << 4)
66
67 static inline u32 apic_get_reg(struct kvm_lapic *apic, int reg_off)
68 {
69 return *((u32 *) (apic->regs + reg_off));
70 }
71
72 static inline void apic_set_reg(struct kvm_lapic *apic, int reg_off, u32 val)
73 {
74 *((u32 *) (apic->regs + reg_off)) = val;
75 }
76
77 static inline int apic_test_and_set_vector(int vec, void *bitmap)
78 {
79 return test_and_set_bit(VEC_POS(vec), (bitmap) + REG_POS(vec));
80 }
81
82 static inline int apic_test_and_clear_vector(int vec, void *bitmap)
83 {
84 return test_and_clear_bit(VEC_POS(vec), (bitmap) + REG_POS(vec));
85 }
86
87 static inline void apic_set_vector(int vec, void *bitmap)
88 {
89 set_bit(VEC_POS(vec), (bitmap) + REG_POS(vec));
90 }
91
92 static inline void apic_clear_vector(int vec, void *bitmap)
93 {
94 clear_bit(VEC_POS(vec), (bitmap) + REG_POS(vec));
95 }
96
97 static inline int apic_hw_enabled(struct kvm_lapic *apic)
98 {
99 return (apic)->vcpu->arch.apic_base & MSR_IA32_APICBASE_ENABLE;
100 }
101
102 static inline int apic_sw_enabled(struct kvm_lapic *apic)
103 {
104 return apic_get_reg(apic, APIC_SPIV) & APIC_SPIV_APIC_ENABLED;
105 }
106
107 static inline int apic_enabled(struct kvm_lapic *apic)
108 {
109 return apic_sw_enabled(apic) && apic_hw_enabled(apic);
110 }
111
112 #define LVT_MASK \
113 (APIC_LVT_MASKED | APIC_SEND_PENDING | APIC_VECTOR_MASK)
114
115 #define LINT_MASK \
116 (LVT_MASK | APIC_MODE_MASK | APIC_INPUT_POLARITY | \
117 APIC_LVT_REMOTE_IRR | APIC_LVT_LEVEL_TRIGGER)
118
119 static inline int kvm_apic_id(struct kvm_lapic *apic)
120 {
121 return (apic_get_reg(apic, APIC_ID) >> 24) & 0xff;
122 }
123
124 static inline int apic_lvt_enabled(struct kvm_lapic *apic, int lvt_type)
125 {
126 return !(apic_get_reg(apic, lvt_type) & APIC_LVT_MASKED);
127 }
128
129 static inline int apic_lvt_vector(struct kvm_lapic *apic, int lvt_type)
130 {
131 return apic_get_reg(apic, lvt_type) & APIC_VECTOR_MASK;
132 }
133
134 static inline int apic_lvtt_period(struct kvm_lapic *apic)
135 {
136 return apic_get_reg(apic, APIC_LVTT) & APIC_LVT_TIMER_PERIODIC;
137 }
138
139 static inline int apic_lvt_nmi_mode(u32 lvt_val)
140 {
141 return (lvt_val & (APIC_MODE_MASK | APIC_LVT_MASKED)) == APIC_DM_NMI;
142 }
143
144 static unsigned int apic_lvt_mask[APIC_LVT_NUM] = {
145 LVT_MASK | APIC_LVT_TIMER_PERIODIC, /* LVTT */
146 LVT_MASK | APIC_MODE_MASK, /* LVTTHMR */
147 LVT_MASK | APIC_MODE_MASK, /* LVTPC */
148 LINT_MASK, LINT_MASK, /* LVT0-1 */
149 LVT_MASK /* LVTERR */
150 };
151
152 static int find_highest_vector(void *bitmap)
153 {
154 u32 *word = bitmap;
155 int word_offset = MAX_APIC_VECTOR >> 5;
156
157 while ((word_offset != 0) && (word[(--word_offset) << 2] == 0))
158 continue;
159
160 if (likely(!word_offset && !word[0]))
161 return -1;
162 else
163 return fls(word[word_offset << 2]) - 1 + (word_offset << 5);
164 }
165
166 static inline int apic_test_and_set_irr(int vec, struct kvm_lapic *apic)
167 {
168 return apic_test_and_set_vector(vec, apic->regs + APIC_IRR);
169 }
170
171 static inline void apic_clear_irr(int vec, struct kvm_lapic *apic)
172 {
173 apic_clear_vector(vec, apic->regs + APIC_IRR);
174 }
175
176 static inline int apic_find_highest_irr(struct kvm_lapic *apic)
177 {
178 int result;
179
180 result = find_highest_vector(apic->regs + APIC_IRR);
181 ASSERT(result == -1 || result >= 16);
182
183 return result;
184 }
185
186 int kvm_lapic_find_highest_irr(struct kvm_vcpu *vcpu)
187 {
188 struct kvm_lapic *apic = vcpu->arch.apic;
189 int highest_irr;
190
191 if (!apic)
192 return 0;
193 highest_irr = apic_find_highest_irr(apic);
194
195 return highest_irr;
196 }
197 EXPORT_SYMBOL_GPL(kvm_lapic_find_highest_irr);
198
199 static int __apic_accept_irq(struct kvm_lapic *apic, int delivery_mode,
200 int vector, int level, int trig_mode);
201
202 int kvm_apic_set_irq(struct kvm_vcpu *vcpu, u8 vec, u8 dmode, u8 trig)
203 {
204 struct kvm_lapic *apic = vcpu->arch.apic;
205 int lapic_dmode;
206
207 switch (dmode) {
208 case IOAPIC_LOWEST_PRIORITY:
209 lapic_dmode = APIC_DM_LOWEST;
210 break;
211 case IOAPIC_FIXED:
212 lapic_dmode = APIC_DM_FIXED;
213 break;
214 case IOAPIC_NMI:
215 lapic_dmode = APIC_DM_NMI;
216 break;
217 default:
218 printk(KERN_DEBUG"Ignoring delivery mode %d\n", dmode);
219 return 0;
220 break;
221 }
222 return __apic_accept_irq(apic, lapic_dmode, vec, 1, trig);
223 }
224
225 static inline int apic_find_highest_isr(struct kvm_lapic *apic)
226 {
227 int result;
228
229 result = find_highest_vector(apic->regs + APIC_ISR);
230 ASSERT(result == -1 || result >= 16);
231
232 return result;
233 }
234
235 static void apic_update_ppr(struct kvm_lapic *apic)
236 {
237 u32 tpr, isrv, ppr;
238 int isr;
239
240 tpr = apic_get_reg(apic, APIC_TASKPRI);
241 isr = apic_find_highest_isr(apic);
242 isrv = (isr != -1) ? isr : 0;
243
244 if ((tpr & 0xf0) >= (isrv & 0xf0))
245 ppr = tpr & 0xff;
246 else
247 ppr = isrv & 0xf0;
248
249 apic_debug("vlapic %p, ppr 0x%x, isr 0x%x, isrv 0x%x",
250 apic, ppr, isr, isrv);
251
252 apic_set_reg(apic, APIC_PROCPRI, ppr);
253 }
254
255 static void apic_set_tpr(struct kvm_lapic *apic, u32 tpr)
256 {
257 apic_set_reg(apic, APIC_TASKPRI, tpr);
258 apic_update_ppr(apic);
259 }
260
261 int kvm_apic_match_physical_addr(struct kvm_lapic *apic, u16 dest)
262 {
263 return dest == 0xff || kvm_apic_id(apic) == dest;
264 }
265
266 int kvm_apic_match_logical_addr(struct kvm_lapic *apic, u8 mda)
267 {
268 int result = 0;
269 u8 logical_id;
270
271 logical_id = GET_APIC_LOGICAL_ID(apic_get_reg(apic, APIC_LDR));
272
273 switch (apic_get_reg(apic, APIC_DFR)) {
274 case APIC_DFR_FLAT:
275 if (logical_id & mda)
276 result = 1;
277 break;
278 case APIC_DFR_CLUSTER:
279 if (((logical_id >> 4) == (mda >> 0x4))
280 && (logical_id & mda & 0xf))
281 result = 1;
282 break;
283 default:
284 printk(KERN_WARNING "Bad DFR vcpu %d: %08x\n",
285 apic->vcpu->vcpu_id, apic_get_reg(apic, APIC_DFR));
286 break;
287 }
288
289 return result;
290 }
291
292 int kvm_apic_match_dest(struct kvm_vcpu *vcpu, struct kvm_lapic *source,
293 int short_hand, int dest, int dest_mode)
294 {
295 int result = 0;
296 struct kvm_lapic *target = vcpu->arch.apic;
297
298 apic_debug("target %p, source %p, dest 0x%x, "
299 "dest_mode 0x%x, short_hand 0x%x\n",
300 target, source, dest, dest_mode, short_hand);
301
302 ASSERT(!target);
303 switch (short_hand) {
304 case APIC_DEST_NOSHORT:
305 if (dest_mode == 0)
306 /* Physical mode. */
307 result = kvm_apic_match_physical_addr(target, dest);
308 else
309 /* Logical mode. */
310 result = kvm_apic_match_logical_addr(target, dest);
311 break;
312 case APIC_DEST_SELF:
313 result = (target == source);
314 break;
315 case APIC_DEST_ALLINC:
316 result = 1;
317 break;
318 case APIC_DEST_ALLBUT:
319 result = (target != source);
320 break;
321 default:
322 printk(KERN_WARNING "Bad dest shorthand value %x\n",
323 short_hand);
324 break;
325 }
326
327 return result;
328 }
329
330 /*
331 * Add a pending IRQ into lapic.
332 * Return 1 if successfully added and 0 if discarded.
333 */
334 static int __apic_accept_irq(struct kvm_lapic *apic, int delivery_mode,
335 int vector, int level, int trig_mode)
336 {
337 int result = 0;
338 struct kvm_vcpu *vcpu = apic->vcpu;
339
340 switch (delivery_mode) {
341 case APIC_DM_LOWEST:
342 vcpu->arch.apic_arb_prio++;
343 case APIC_DM_FIXED:
344 /* FIXME add logic for vcpu on reset */
345 if (unlikely(!apic_enabled(apic)))
346 break;
347
348 result = !apic_test_and_set_irr(vector, apic);
349 if (!result) {
350 if (trig_mode)
351 apic_debug("level trig mode repeatedly for "
352 "vector %d", vector);
353 break;
354 }
355
356 if (trig_mode) {
357 apic_debug("level trig mode for vector %d", vector);
358 apic_set_vector(vector, apic->regs + APIC_TMR);
359 } else
360 apic_clear_vector(vector, apic->regs + APIC_TMR);
361 kvm_vcpu_kick(vcpu);
362 break;
363
364 case APIC_DM_REMRD:
365 printk(KERN_DEBUG "Ignoring delivery mode 3\n");
366 break;
367
368 case APIC_DM_SMI:
369 printk(KERN_DEBUG "Ignoring guest SMI\n");
370 break;
371
372 case APIC_DM_NMI:
373 result = 1;
374 kvm_inject_nmi(vcpu);
375 kvm_vcpu_kick(vcpu);
376 break;
377
378 case APIC_DM_INIT:
379 if (level) {
380 result = 1;
381 if (vcpu->arch.mp_state == KVM_MP_STATE_RUNNABLE)
382 printk(KERN_DEBUG
383 "INIT on a runnable vcpu %d\n",
384 vcpu->vcpu_id);
385 vcpu->arch.mp_state = KVM_MP_STATE_INIT_RECEIVED;
386 kvm_vcpu_kick(vcpu);
387 } else {
388 apic_debug("Ignoring de-assert INIT to vcpu %d\n",
389 vcpu->vcpu_id);
390 }
391 break;
392
393 case APIC_DM_STARTUP:
394 apic_debug("SIPI to vcpu %d vector 0x%02x\n",
395 vcpu->vcpu_id, vector);
396 if (vcpu->arch.mp_state == KVM_MP_STATE_INIT_RECEIVED) {
397 result = 1;
398 vcpu->arch.sipi_vector = vector;
399 vcpu->arch.mp_state = KVM_MP_STATE_SIPI_RECEIVED;
400 kvm_vcpu_kick(vcpu);
401 }
402 break;
403
404 case APIC_DM_EXTINT:
405 /*
406 * Should only be called by kvm_apic_local_deliver() with LVT0,
407 * before NMI watchdog was enabled. Already handled by
408 * kvm_apic_accept_pic_intr().
409 */
410 break;
411
412 default:
413 printk(KERN_ERR "TODO: unsupported delivery mode %x\n",
414 delivery_mode);
415 break;
416 }
417 return result;
418 }
419
420 int kvm_apic_compare_prio(struct kvm_vcpu *vcpu1, struct kvm_vcpu *vcpu2)
421 {
422 return vcpu1->arch.apic_arb_prio - vcpu2->arch.apic_arb_prio;
423 }
424
425 static void apic_set_eoi(struct kvm_lapic *apic)
426 {
427 int vector = apic_find_highest_isr(apic);
428 int trigger_mode;
429 /*
430 * Not every write EOI will has corresponding ISR,
431 * one example is when Kernel check timer on setup_IO_APIC
432 */
433 if (vector == -1)
434 return;
435
436 apic_clear_vector(vector, apic->regs + APIC_ISR);
437 apic_update_ppr(apic);
438
439 if (apic_test_and_clear_vector(vector, apic->regs + APIC_TMR))
440 trigger_mode = IOAPIC_LEVEL_TRIG;
441 else
442 trigger_mode = IOAPIC_EDGE_TRIG;
443 kvm_ioapic_update_eoi(apic->vcpu->kvm, vector, trigger_mode);
444 }
445
446 static void apic_send_ipi(struct kvm_lapic *apic)
447 {
448 u32 icr_low = apic_get_reg(apic, APIC_ICR);
449 u32 icr_high = apic_get_reg(apic, APIC_ICR2);
450
451 unsigned int dest = GET_APIC_DEST_FIELD(icr_high);
452 unsigned int short_hand = icr_low & APIC_SHORT_MASK;
453 unsigned int trig_mode = icr_low & APIC_INT_LEVELTRIG;
454 unsigned int level = icr_low & APIC_INT_ASSERT;
455 unsigned int dest_mode = icr_low & APIC_DEST_MASK;
456 unsigned int delivery_mode = icr_low & APIC_MODE_MASK;
457 unsigned int vector = icr_low & APIC_VECTOR_MASK;
458
459 DECLARE_BITMAP(deliver_bitmask, KVM_MAX_VCPUS);
460 int i;
461
462 apic_debug("icr_high 0x%x, icr_low 0x%x, "
463 "short_hand 0x%x, dest 0x%x, trig_mode 0x%x, level 0x%x, "
464 "dest_mode 0x%x, delivery_mode 0x%x, vector 0x%x\n",
465 icr_high, icr_low, short_hand, dest,
466 trig_mode, level, dest_mode, delivery_mode, vector);
467
468 kvm_get_intr_delivery_bitmask(apic->vcpu->kvm, apic, dest, dest_mode,
469 delivery_mode == APIC_DM_LOWEST, short_hand,
470 deliver_bitmask);
471
472 while ((i = find_first_bit(deliver_bitmask, KVM_MAX_VCPUS))
473 < KVM_MAX_VCPUS) {
474 struct kvm_vcpu *vcpu = apic->vcpu->kvm->vcpus[i];
475 __clear_bit(i, deliver_bitmask);
476 if (vcpu)
477 __apic_accept_irq(vcpu->arch.apic, delivery_mode,
478 vector, level, trig_mode);
479 }
480 }
481
482 static u32 apic_get_tmcct(struct kvm_lapic *apic)
483 {
484 ktime_t remaining;
485 s64 ns;
486 u32 tmcct;
487
488 ASSERT(apic != NULL);
489
490 /* if initial count is 0, current count should also be 0 */
491 if (apic_get_reg(apic, APIC_TMICT) == 0)
492 return 0;
493
494 remaining = hrtimer_expires_remaining(&apic->lapic_timer.timer);
495 if (ktime_to_ns(remaining) < 0)
496 remaining = ktime_set(0, 0);
497
498 ns = mod_64(ktime_to_ns(remaining), apic->lapic_timer.period);
499 tmcct = div64_u64(ns,
500 (APIC_BUS_CYCLE_NS * apic->divide_count));
501
502 return tmcct;
503 }
504
505 static void __report_tpr_access(struct kvm_lapic *apic, bool write)
506 {
507 struct kvm_vcpu *vcpu = apic->vcpu;
508 struct kvm_run *run = vcpu->run;
509
510 set_bit(KVM_REQ_REPORT_TPR_ACCESS, &vcpu->requests);
511 run->tpr_access.rip = kvm_rip_read(vcpu);
512 run->tpr_access.is_write = write;
513 }
514
515 static inline void report_tpr_access(struct kvm_lapic *apic, bool write)
516 {
517 if (apic->vcpu->arch.tpr_access_reporting)
518 __report_tpr_access(apic, write);
519 }
520
521 static u32 __apic_read(struct kvm_lapic *apic, unsigned int offset)
522 {
523 u32 val = 0;
524
525 KVMTRACE_1D(APIC_ACCESS, apic->vcpu, (u32)offset, handler);
526
527 if (offset >= LAPIC_MMIO_LENGTH)
528 return 0;
529
530 switch (offset) {
531 case APIC_ARBPRI:
532 printk(KERN_WARNING "Access APIC ARBPRI register "
533 "which is for P6\n");
534 break;
535
536 case APIC_TMCCT: /* Timer CCR */
537 val = apic_get_tmcct(apic);
538 break;
539
540 case APIC_TASKPRI:
541 report_tpr_access(apic, false);
542 /* fall thru */
543 default:
544 apic_update_ppr(apic);
545 val = apic_get_reg(apic, offset);
546 break;
547 }
548
549 return val;
550 }
551
552 static void apic_mmio_read(struct kvm_io_device *this,
553 gpa_t address, int len, void *data)
554 {
555 struct kvm_lapic *apic = (struct kvm_lapic *)this->private;
556 unsigned int offset = address - apic->base_address;
557 unsigned char alignment = offset & 0xf;
558 u32 result;
559
560 if ((alignment + len) > 4) {
561 printk(KERN_ERR "KVM_APIC_READ: alignment error %lx %d",
562 (unsigned long)address, len);
563 return;
564 }
565 result = __apic_read(apic, offset & ~0xf);
566
567 switch (len) {
568 case 1:
569 case 2:
570 case 4:
571 memcpy(data, (char *)&result + alignment, len);
572 break;
573 default:
574 printk(KERN_ERR "Local APIC read with len = %x, "
575 "should be 1,2, or 4 instead\n", len);
576 break;
577 }
578 }
579
580 static void update_divide_count(struct kvm_lapic *apic)
581 {
582 u32 tmp1, tmp2, tdcr;
583
584 tdcr = apic_get_reg(apic, APIC_TDCR);
585 tmp1 = tdcr & 0xf;
586 tmp2 = ((tmp1 & 0x3) | ((tmp1 & 0x8) >> 1)) + 1;
587 apic->divide_count = 0x1 << (tmp2 & 0x7);
588
589 apic_debug("timer divide count is 0x%x\n",
590 apic->lapic_timer.divide_count);
591 }
592
593 static void start_apic_timer(struct kvm_lapic *apic)
594 {
595 ktime_t now = apic->lapic_timer.timer.base->get_time();
596
597 apic->lapic_timer.period = apic_get_reg(apic, APIC_TMICT) *
598 APIC_BUS_CYCLE_NS * apic->divide_count;
599 atomic_set(&apic->lapic_timer.pending, 0);
600
601 if (!apic->lapic_timer.period)
602 return;
603
604 hrtimer_start(&apic->lapic_timer.timer,
605 ktime_add_ns(now, apic->lapic_timer.period),
606 HRTIMER_MODE_ABS);
607
608 apic_debug("%s: bus cycle is %" PRId64 "ns, now 0x%016"
609 PRIx64 ", "
610 "timer initial count 0x%x, period %lldns, "
611 "expire @ 0x%016" PRIx64 ".\n", __func__,
612 APIC_BUS_CYCLE_NS, ktime_to_ns(now),
613 apic_get_reg(apic, APIC_TMICT),
614 apic->lapic_timer.period,
615 ktime_to_ns(ktime_add_ns(now,
616 apic->lapic_timer.period)));
617 }
618
619 static void apic_manage_nmi_watchdog(struct kvm_lapic *apic, u32 lvt0_val)
620 {
621 int nmi_wd_enabled = apic_lvt_nmi_mode(apic_get_reg(apic, APIC_LVT0));
622
623 if (apic_lvt_nmi_mode(lvt0_val)) {
624 if (!nmi_wd_enabled) {
625 apic_debug("Receive NMI setting on APIC_LVT0 "
626 "for cpu %d\n", apic->vcpu->vcpu_id);
627 apic->vcpu->kvm->arch.vapics_in_nmi_mode++;
628 }
629 } else if (nmi_wd_enabled)
630 apic->vcpu->kvm->arch.vapics_in_nmi_mode--;
631 }
632
633 static void apic_mmio_write(struct kvm_io_device *this,
634 gpa_t address, int len, const void *data)
635 {
636 struct kvm_lapic *apic = (struct kvm_lapic *)this->private;
637 unsigned int offset = address - apic->base_address;
638 unsigned char alignment = offset & 0xf;
639 u32 val;
640
641 /*
642 * APIC register must be aligned on 128-bits boundary.
643 * 32/64/128 bits registers must be accessed thru 32 bits.
644 * Refer SDM 8.4.1
645 */
646 if (len != 4 || alignment) {
647 /* Don't shout loud, $infamous_os would cause only noise. */
648 apic_debug("apic write: bad size=%d %lx\n",
649 len, (long)address);
650 return;
651 }
652
653 val = *(u32 *) data;
654
655 /* too common printing */
656 if (offset != APIC_EOI)
657 apic_debug("%s: offset 0x%x with length 0x%x, and value is "
658 "0x%x\n", __func__, offset, len, val);
659
660 offset &= 0xff0;
661
662 KVMTRACE_1D(APIC_ACCESS, apic->vcpu, (u32)offset, handler);
663
664 switch (offset) {
665 case APIC_ID: /* Local APIC ID */
666 apic_set_reg(apic, APIC_ID, val);
667 break;
668
669 case APIC_TASKPRI:
670 report_tpr_access(apic, true);
671 apic_set_tpr(apic, val & 0xff);
672 break;
673
674 case APIC_EOI:
675 apic_set_eoi(apic);
676 break;
677
678 case APIC_LDR:
679 apic_set_reg(apic, APIC_LDR, val & APIC_LDR_MASK);
680 break;
681
682 case APIC_DFR:
683 apic_set_reg(apic, APIC_DFR, val | 0x0FFFFFFF);
684 break;
685
686 case APIC_SPIV:
687 apic_set_reg(apic, APIC_SPIV, val & 0x3ff);
688 if (!(val & APIC_SPIV_APIC_ENABLED)) {
689 int i;
690 u32 lvt_val;
691
692 for (i = 0; i < APIC_LVT_NUM; i++) {
693 lvt_val = apic_get_reg(apic,
694 APIC_LVTT + 0x10 * i);
695 apic_set_reg(apic, APIC_LVTT + 0x10 * i,
696 lvt_val | APIC_LVT_MASKED);
697 }
698 atomic_set(&apic->lapic_timer.pending, 0);
699
700 }
701 break;
702
703 case APIC_ICR:
704 /* No delay here, so we always clear the pending bit */
705 apic_set_reg(apic, APIC_ICR, val & ~(1 << 12));
706 apic_send_ipi(apic);
707 break;
708
709 case APIC_ICR2:
710 apic_set_reg(apic, APIC_ICR2, val & 0xff000000);
711 break;
712
713 case APIC_LVT0:
714 apic_manage_nmi_watchdog(apic, val);
715 case APIC_LVTT:
716 case APIC_LVTTHMR:
717 case APIC_LVTPC:
718 case APIC_LVT1:
719 case APIC_LVTERR:
720 /* TODO: Check vector */
721 if (!apic_sw_enabled(apic))
722 val |= APIC_LVT_MASKED;
723
724 val &= apic_lvt_mask[(offset - APIC_LVTT) >> 4];
725 apic_set_reg(apic, offset, val);
726
727 break;
728
729 case APIC_TMICT:
730 hrtimer_cancel(&apic->lapic_timer.timer);
731 apic_set_reg(apic, APIC_TMICT, val);
732 start_apic_timer(apic);
733 return;
734
735 case APIC_TDCR:
736 if (val & 4)
737 printk(KERN_ERR "KVM_WRITE:TDCR %x\n", val);
738 apic_set_reg(apic, APIC_TDCR, val);
739 update_divide_count(apic);
740 break;
741
742 default:
743 apic_debug("Local APIC Write to read-only register %x\n",
744 offset);
745 break;
746 }
747
748 }
749
750 static int apic_mmio_range(struct kvm_io_device *this, gpa_t addr,
751 int len, int size)
752 {
753 struct kvm_lapic *apic = (struct kvm_lapic *)this->private;
754 int ret = 0;
755
756
757 if (apic_hw_enabled(apic) &&
758 (addr >= apic->base_address) &&
759 (addr < (apic->base_address + LAPIC_MMIO_LENGTH)))
760 ret = 1;
761
762 return ret;
763 }
764
765 void kvm_free_lapic(struct kvm_vcpu *vcpu)
766 {
767 if (!vcpu->arch.apic)
768 return;
769
770 hrtimer_cancel(&vcpu->arch.apic->lapic_timer.timer);
771
772 if (vcpu->arch.apic->regs_page)
773 __free_page(vcpu->arch.apic->regs_page);
774
775 kfree(vcpu->arch.apic);
776 }
777
778 /*
779 *----------------------------------------------------------------------
780 * LAPIC interface
781 *----------------------------------------------------------------------
782 */
783
784 void kvm_lapic_set_tpr(struct kvm_vcpu *vcpu, unsigned long cr8)
785 {
786 struct kvm_lapic *apic = vcpu->arch.apic;
787
788 if (!apic)
789 return;
790 apic_set_tpr(apic, ((cr8 & 0x0f) << 4)
791 | (apic_get_reg(apic, APIC_TASKPRI) & 4));
792 }
793 EXPORT_SYMBOL_GPL(kvm_lapic_set_tpr);
794
795 u64 kvm_lapic_get_cr8(struct kvm_vcpu *vcpu)
796 {
797 struct kvm_lapic *apic = vcpu->arch.apic;
798 u64 tpr;
799
800 if (!apic)
801 return 0;
802 tpr = (u64) apic_get_reg(apic, APIC_TASKPRI);
803
804 return (tpr & 0xf0) >> 4;
805 }
806 EXPORT_SYMBOL_GPL(kvm_lapic_get_cr8);
807
808 void kvm_lapic_set_base(struct kvm_vcpu *vcpu, u64 value)
809 {
810 struct kvm_lapic *apic = vcpu->arch.apic;
811
812 if (!apic) {
813 value |= MSR_IA32_APICBASE_BSP;
814 vcpu->arch.apic_base = value;
815 return;
816 }
817 if (apic->vcpu->vcpu_id)
818 value &= ~MSR_IA32_APICBASE_BSP;
819
820 vcpu->arch.apic_base = value;
821 apic->base_address = apic->vcpu->arch.apic_base &
822 MSR_IA32_APICBASE_BASE;
823
824 /* with FSB delivery interrupt, we can restart APIC functionality */
825 apic_debug("apic base msr is 0x%016" PRIx64 ", and base address is "
826 "0x%lx.\n", apic->vcpu->arch.apic_base, apic->base_address);
827
828 }
829
830 u64 kvm_lapic_get_base(struct kvm_vcpu *vcpu)
831 {
832 return vcpu->arch.apic_base;
833 }
834 EXPORT_SYMBOL_GPL(kvm_lapic_get_base);
835
836 void kvm_lapic_reset(struct kvm_vcpu *vcpu)
837 {
838 struct kvm_lapic *apic;
839 int i;
840
841 apic_debug("%s\n", __func__);
842
843 ASSERT(vcpu);
844 apic = vcpu->arch.apic;
845 ASSERT(apic != NULL);
846
847 /* Stop the timer in case it's a reset to an active apic */
848 hrtimer_cancel(&apic->lapic_timer.timer);
849
850 apic_set_reg(apic, APIC_ID, vcpu->vcpu_id << 24);
851 apic_set_reg(apic, APIC_LVR, APIC_VERSION);
852
853 for (i = 0; i < APIC_LVT_NUM; i++)
854 apic_set_reg(apic, APIC_LVTT + 0x10 * i, APIC_LVT_MASKED);
855 apic_set_reg(apic, APIC_LVT0,
856 SET_APIC_DELIVERY_MODE(0, APIC_MODE_EXTINT));
857
858 apic_set_reg(apic, APIC_DFR, 0xffffffffU);
859 apic_set_reg(apic, APIC_SPIV, 0xff);
860 apic_set_reg(apic, APIC_TASKPRI, 0);
861 apic_set_reg(apic, APIC_LDR, 0);
862 apic_set_reg(apic, APIC_ESR, 0);
863 apic_set_reg(apic, APIC_ICR, 0);
864 apic_set_reg(apic, APIC_ICR2, 0);
865 apic_set_reg(apic, APIC_TDCR, 0);
866 apic_set_reg(apic, APIC_TMICT, 0);
867 for (i = 0; i < 8; i++) {
868 apic_set_reg(apic, APIC_IRR + 0x10 * i, 0);
869 apic_set_reg(apic, APIC_ISR + 0x10 * i, 0);
870 apic_set_reg(apic, APIC_TMR + 0x10 * i, 0);
871 }
872 update_divide_count(apic);
873 atomic_set(&apic->lapic_timer.pending, 0);
874 if (vcpu->vcpu_id == 0)
875 vcpu->arch.apic_base |= MSR_IA32_APICBASE_BSP;
876 apic_update_ppr(apic);
877
878 vcpu->arch.apic_arb_prio = 0;
879
880 apic_debug(KERN_INFO "%s: vcpu=%p, id=%d, base_msr="
881 "0x%016" PRIx64 ", base_address=0x%0lx.\n", __func__,
882 vcpu, kvm_apic_id(apic),
883 vcpu->arch.apic_base, apic->base_address);
884 }
885 EXPORT_SYMBOL_GPL(kvm_lapic_reset);
886
887 bool kvm_apic_present(struct kvm_vcpu *vcpu)
888 {
889 return vcpu->arch.apic && apic_hw_enabled(vcpu->arch.apic);
890 }
891
892 int kvm_lapic_enabled(struct kvm_vcpu *vcpu)
893 {
894 return kvm_apic_present(vcpu) && apic_sw_enabled(vcpu->arch.apic);
895 }
896 EXPORT_SYMBOL_GPL(kvm_lapic_enabled);
897
898 /*
899 *----------------------------------------------------------------------
900 * timer interface
901 *----------------------------------------------------------------------
902 */
903
904 static bool lapic_is_periodic(struct kvm_timer *ktimer)
905 {
906 struct kvm_lapic *apic = container_of(ktimer, struct kvm_lapic,
907 lapic_timer);
908 return apic_lvtt_period(apic);
909 }
910
911 int apic_has_pending_timer(struct kvm_vcpu *vcpu)
912 {
913 struct kvm_lapic *lapic = vcpu->arch.apic;
914
915 if (lapic && apic_enabled(lapic) && apic_lvt_enabled(lapic, APIC_LVTT))
916 return atomic_read(&lapic->lapic_timer.pending);
917
918 return 0;
919 }
920
921 static int kvm_apic_local_deliver(struct kvm_lapic *apic, int lvt_type)
922 {
923 u32 reg = apic_get_reg(apic, lvt_type);
924 int vector, mode, trig_mode;
925
926 if (apic_hw_enabled(apic) && !(reg & APIC_LVT_MASKED)) {
927 vector = reg & APIC_VECTOR_MASK;
928 mode = reg & APIC_MODE_MASK;
929 trig_mode = reg & APIC_LVT_LEVEL_TRIGGER;
930 return __apic_accept_irq(apic, mode, vector, 1, trig_mode);
931 }
932 return 0;
933 }
934
935 void kvm_apic_nmi_wd_deliver(struct kvm_vcpu *vcpu)
936 {
937 struct kvm_lapic *apic = vcpu->arch.apic;
938
939 if (apic)
940 kvm_apic_local_deliver(apic, APIC_LVT0);
941 }
942
943 struct kvm_timer_ops lapic_timer_ops = {
944 .is_periodic = lapic_is_periodic,
945 };
946
947 int kvm_create_lapic(struct kvm_vcpu *vcpu)
948 {
949 struct kvm_lapic *apic;
950
951 ASSERT(vcpu != NULL);
952 apic_debug("apic_init %d\n", vcpu->vcpu_id);
953
954 apic = kzalloc(sizeof(*apic), GFP_KERNEL);
955 if (!apic)
956 goto nomem;
957
958 vcpu->arch.apic = apic;
959
960 apic->regs_page = alloc_page(GFP_KERNEL);
961 if (apic->regs_page == NULL) {
962 printk(KERN_ERR "malloc apic regs error for vcpu %x\n",
963 vcpu->vcpu_id);
964 goto nomem_free_apic;
965 }
966 apic->regs = page_address(apic->regs_page);
967 memset(apic->regs, 0, PAGE_SIZE);
968 apic->vcpu = vcpu;
969
970 hrtimer_init(&apic->lapic_timer.timer, CLOCK_MONOTONIC,
971 HRTIMER_MODE_ABS);
972 apic->lapic_timer.timer.function = kvm_timer_fn;
973 apic->lapic_timer.t_ops = &lapic_timer_ops;
974 apic->lapic_timer.kvm = vcpu->kvm;
975 apic->lapic_timer.vcpu_id = vcpu->vcpu_id;
976
977 apic->base_address = APIC_DEFAULT_PHYS_BASE;
978 vcpu->arch.apic_base = APIC_DEFAULT_PHYS_BASE;
979
980 kvm_lapic_reset(vcpu);
981 apic->dev.read = apic_mmio_read;
982 apic->dev.write = apic_mmio_write;
983 apic->dev.in_range = apic_mmio_range;
984 apic->dev.private = apic;
985
986 return 0;
987 nomem_free_apic:
988 kfree(apic);
989 nomem:
990 return -ENOMEM;
991 }
992 EXPORT_SYMBOL_GPL(kvm_create_lapic);
993
994 int kvm_apic_has_interrupt(struct kvm_vcpu *vcpu)
995 {
996 struct kvm_lapic *apic = vcpu->arch.apic;
997 int highest_irr;
998
999 if (!apic || !apic_enabled(apic))
1000 return -1;
1001
1002 apic_update_ppr(apic);
1003 highest_irr = apic_find_highest_irr(apic);
1004 if ((highest_irr == -1) ||
1005 ((highest_irr & 0xF0) <= apic_get_reg(apic, APIC_PROCPRI)))
1006 return -1;
1007 return highest_irr;
1008 }
1009
1010 int kvm_apic_accept_pic_intr(struct kvm_vcpu *vcpu)
1011 {
1012 u32 lvt0 = apic_get_reg(vcpu->arch.apic, APIC_LVT0);
1013 int r = 0;
1014
1015 if (vcpu->vcpu_id == 0) {
1016 if (!apic_hw_enabled(vcpu->arch.apic))
1017 r = 1;
1018 if ((lvt0 & APIC_LVT_MASKED) == 0 &&
1019 GET_APIC_DELIVERY_MODE(lvt0) == APIC_MODE_EXTINT)
1020 r = 1;
1021 }
1022 return r;
1023 }
1024
1025 void kvm_inject_apic_timer_irqs(struct kvm_vcpu *vcpu)
1026 {
1027 struct kvm_lapic *apic = vcpu->arch.apic;
1028
1029 if (apic && atomic_read(&apic->lapic_timer.pending) > 0) {
1030 if (kvm_apic_local_deliver(apic, APIC_LVTT))
1031 atomic_dec(&apic->lapic_timer.pending);
1032 }
1033 }
1034
1035 int kvm_get_apic_interrupt(struct kvm_vcpu *vcpu)
1036 {
1037 int vector = kvm_apic_has_interrupt(vcpu);
1038 struct kvm_lapic *apic = vcpu->arch.apic;
1039
1040 if (vector == -1)
1041 return -1;
1042
1043 apic_set_vector(vector, apic->regs + APIC_ISR);
1044 apic_update_ppr(apic);
1045 apic_clear_irr(vector, apic);
1046 return vector;
1047 }
1048
1049 void kvm_apic_post_state_restore(struct kvm_vcpu *vcpu)
1050 {
1051 struct kvm_lapic *apic = vcpu->arch.apic;
1052
1053 apic->base_address = vcpu->arch.apic_base &
1054 MSR_IA32_APICBASE_BASE;
1055 apic_set_reg(apic, APIC_LVR, APIC_VERSION);
1056 apic_update_ppr(apic);
1057 hrtimer_cancel(&apic->lapic_timer.timer);
1058 update_divide_count(apic);
1059 start_apic_timer(apic);
1060 }
1061
1062 void __kvm_migrate_apic_timer(struct kvm_vcpu *vcpu)
1063 {
1064 struct kvm_lapic *apic = vcpu->arch.apic;
1065 struct hrtimer *timer;
1066
1067 if (!apic)
1068 return;
1069
1070 timer = &apic->lapic_timer.timer;
1071 if (hrtimer_cancel(timer))
1072 hrtimer_start_expires(timer, HRTIMER_MODE_ABS);
1073 }
1074
1075 void kvm_lapic_sync_from_vapic(struct kvm_vcpu *vcpu)
1076 {
1077 u32 data;
1078 void *vapic;
1079
1080 if (!irqchip_in_kernel(vcpu->kvm) || !vcpu->arch.apic->vapic_addr)
1081 return;
1082
1083 vapic = kmap_atomic(vcpu->arch.apic->vapic_page, KM_USER0);
1084 data = *(u32 *)(vapic + offset_in_page(vcpu->arch.apic->vapic_addr));
1085 kunmap_atomic(vapic, KM_USER0);
1086
1087 apic_set_tpr(vcpu->arch.apic, data & 0xff);
1088 }
1089
1090 void kvm_lapic_sync_to_vapic(struct kvm_vcpu *vcpu)
1091 {
1092 u32 data, tpr;
1093 int max_irr, max_isr;
1094 struct kvm_lapic *apic;
1095 void *vapic;
1096
1097 if (!irqchip_in_kernel(vcpu->kvm) || !vcpu->arch.apic->vapic_addr)
1098 return;
1099
1100 apic = vcpu->arch.apic;
1101 tpr = apic_get_reg(apic, APIC_TASKPRI) & 0xff;
1102 max_irr = apic_find_highest_irr(apic);
1103 if (max_irr < 0)
1104 max_irr = 0;
1105 max_isr = apic_find_highest_isr(apic);
1106 if (max_isr < 0)
1107 max_isr = 0;
1108 data = (tpr & 0xff) | ((max_isr & 0xf0) << 8) | (max_irr << 24);
1109
1110 vapic = kmap_atomic(vcpu->arch.apic->vapic_page, KM_USER0);
1111 *(u32 *)(vapic + offset_in_page(vcpu->arch.apic->vapic_addr)) = data;
1112 kunmap_atomic(vapic, KM_USER0);
1113 }
1114
1115 void kvm_lapic_set_vapic_addr(struct kvm_vcpu *vcpu, gpa_t vapic_addr)
1116 {
1117 if (!irqchip_in_kernel(vcpu->kvm))
1118 return;
1119
1120 vcpu->arch.apic->vapic_addr = vapic_addr;
1121 }