Merge branch 'next' of git://git.kernel.org/pub/scm/linux/kernel/git/benh/powerpc
[pandora-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  * Copyright 2009 Red Hat, Inc. and/or its affilates.
9  *
10  * Authors:
11  *   Dor Laor <dor.laor@qumranet.com>
12  *   Gregory Haskins <ghaskins@novell.com>
13  *   Yaozu (Eddie) Dong <eddie.dong@intel.com>
14  *
15  * Based on Xen 3.1 code, Copyright (c) 2004, Intel Corporation.
16  *
17  * This work is licensed under the terms of the GNU GPL, version 2.  See
18  * the COPYING file in the top-level directory.
19  */
20
21 #include <linux/kvm_host.h>
22 #include <linux/kvm.h>
23 #include <linux/mm.h>
24 #include <linux/highmem.h>
25 #include <linux/smp.h>
26 #include <linux/hrtimer.h>
27 #include <linux/io.h>
28 #include <linux/module.h>
29 #include <linux/math64.h>
30 #include <linux/slab.h>
31 #include <asm/processor.h>
32 #include <asm/msr.h>
33 #include <asm/page.h>
34 #include <asm/current.h>
35 #include <asm/apicdef.h>
36 #include <asm/atomic.h>
37 #include "kvm_cache_regs.h"
38 #include "irq.h"
39 #include "trace.h"
40 #include "x86.h"
41
42 #ifndef CONFIG_X86_64
43 #define mod_64(x, y) ((x) - (y) * div64_u64(x, y))
44 #else
45 #define mod_64(x, y) ((x) % (y))
46 #endif
47
48 #define PRId64 "d"
49 #define PRIx64 "llx"
50 #define PRIu64 "u"
51 #define PRIo64 "o"
52
53 #define APIC_BUS_CYCLE_NS 1
54
55 /* #define apic_debug(fmt,arg...) printk(KERN_WARNING fmt,##arg) */
56 #define apic_debug(fmt, arg...)
57
58 #define APIC_LVT_NUM                    6
59 /* 14 is the version for Xeon and Pentium 8.4.8*/
60 #define APIC_VERSION                    (0x14UL | ((APIC_LVT_NUM - 1) << 16))
61 #define LAPIC_MMIO_LENGTH               (1 << 12)
62 /* followed define is not in apicdef.h */
63 #define APIC_SHORT_MASK                 0xc0000
64 #define APIC_DEST_NOSHORT               0x0
65 #define APIC_DEST_MASK                  0x800
66 #define MAX_APIC_VECTOR                 256
67
68 #define VEC_POS(v) ((v) & (32 - 1))
69 #define REG_POS(v) (((v) >> 5) << 4)
70
71 static inline u32 apic_get_reg(struct kvm_lapic *apic, int reg_off)
72 {
73         return *((u32 *) (apic->regs + reg_off));
74 }
75
76 static inline void apic_set_reg(struct kvm_lapic *apic, int reg_off, u32 val)
77 {
78         *((u32 *) (apic->regs + reg_off)) = val;
79 }
80
81 static inline int apic_test_and_set_vector(int vec, void *bitmap)
82 {
83         return test_and_set_bit(VEC_POS(vec), (bitmap) + REG_POS(vec));
84 }
85
86 static inline int apic_test_and_clear_vector(int vec, void *bitmap)
87 {
88         return test_and_clear_bit(VEC_POS(vec), (bitmap) + REG_POS(vec));
89 }
90
91 static inline void apic_set_vector(int vec, void *bitmap)
92 {
93         set_bit(VEC_POS(vec), (bitmap) + REG_POS(vec));
94 }
95
96 static inline void apic_clear_vector(int vec, void *bitmap)
97 {
98         clear_bit(VEC_POS(vec), (bitmap) + REG_POS(vec));
99 }
100
101 static inline int apic_hw_enabled(struct kvm_lapic *apic)
102 {
103         return (apic)->vcpu->arch.apic_base & MSR_IA32_APICBASE_ENABLE;
104 }
105
106 static inline int  apic_sw_enabled(struct kvm_lapic *apic)
107 {
108         return apic_get_reg(apic, APIC_SPIV) & APIC_SPIV_APIC_ENABLED;
109 }
110
111 static inline int apic_enabled(struct kvm_lapic *apic)
112 {
113         return apic_sw_enabled(apic) && apic_hw_enabled(apic);
114 }
115
116 #define LVT_MASK        \
117         (APIC_LVT_MASKED | APIC_SEND_PENDING | APIC_VECTOR_MASK)
118
119 #define LINT_MASK       \
120         (LVT_MASK | APIC_MODE_MASK | APIC_INPUT_POLARITY | \
121          APIC_LVT_REMOTE_IRR | APIC_LVT_LEVEL_TRIGGER)
122
123 static inline int kvm_apic_id(struct kvm_lapic *apic)
124 {
125         return (apic_get_reg(apic, APIC_ID) >> 24) & 0xff;
126 }
127
128 static inline int apic_lvt_enabled(struct kvm_lapic *apic, int lvt_type)
129 {
130         return !(apic_get_reg(apic, lvt_type) & APIC_LVT_MASKED);
131 }
132
133 static inline int apic_lvt_vector(struct kvm_lapic *apic, int lvt_type)
134 {
135         return apic_get_reg(apic, lvt_type) & APIC_VECTOR_MASK;
136 }
137
138 static inline int apic_lvtt_period(struct kvm_lapic *apic)
139 {
140         return apic_get_reg(apic, APIC_LVTT) & APIC_LVT_TIMER_PERIODIC;
141 }
142
143 static inline int apic_lvt_nmi_mode(u32 lvt_val)
144 {
145         return (lvt_val & (APIC_MODE_MASK | APIC_LVT_MASKED)) == APIC_DM_NMI;
146 }
147
148 void kvm_apic_set_version(struct kvm_vcpu *vcpu)
149 {
150         struct kvm_lapic *apic = vcpu->arch.apic;
151         struct kvm_cpuid_entry2 *feat;
152         u32 v = APIC_VERSION;
153
154         if (!irqchip_in_kernel(vcpu->kvm))
155                 return;
156
157         feat = kvm_find_cpuid_entry(apic->vcpu, 0x1, 0);
158         if (feat && (feat->ecx & (1 << (X86_FEATURE_X2APIC & 31))))
159                 v |= APIC_LVR_DIRECTED_EOI;
160         apic_set_reg(apic, APIC_LVR, v);
161 }
162
163 static inline int apic_x2apic_mode(struct kvm_lapic *apic)
164 {
165         return apic->vcpu->arch.apic_base & X2APIC_ENABLE;
166 }
167
168 static unsigned int apic_lvt_mask[APIC_LVT_NUM] = {
169         LVT_MASK | APIC_LVT_TIMER_PERIODIC,     /* LVTT */
170         LVT_MASK | APIC_MODE_MASK,      /* LVTTHMR */
171         LVT_MASK | APIC_MODE_MASK,      /* LVTPC */
172         LINT_MASK, LINT_MASK,   /* LVT0-1 */
173         LVT_MASK                /* LVTERR */
174 };
175
176 static int find_highest_vector(void *bitmap)
177 {
178         u32 *word = bitmap;
179         int word_offset = MAX_APIC_VECTOR >> 5;
180
181         while ((word_offset != 0) && (word[(--word_offset) << 2] == 0))
182                 continue;
183
184         if (likely(!word_offset && !word[0]))
185                 return -1;
186         else
187                 return fls(word[word_offset << 2]) - 1 + (word_offset << 5);
188 }
189
190 static inline int apic_test_and_set_irr(int vec, struct kvm_lapic *apic)
191 {
192         apic->irr_pending = true;
193         return apic_test_and_set_vector(vec, apic->regs + APIC_IRR);
194 }
195
196 static inline int apic_search_irr(struct kvm_lapic *apic)
197 {
198         return find_highest_vector(apic->regs + APIC_IRR);
199 }
200
201 static inline int apic_find_highest_irr(struct kvm_lapic *apic)
202 {
203         int result;
204
205         if (!apic->irr_pending)
206                 return -1;
207
208         result = apic_search_irr(apic);
209         ASSERT(result == -1 || result >= 16);
210
211         return result;
212 }
213
214 static inline void apic_clear_irr(int vec, struct kvm_lapic *apic)
215 {
216         apic->irr_pending = false;
217         apic_clear_vector(vec, apic->regs + APIC_IRR);
218         if (apic_search_irr(apic) != -1)
219                 apic->irr_pending = true;
220 }
221
222 int kvm_lapic_find_highest_irr(struct kvm_vcpu *vcpu)
223 {
224         struct kvm_lapic *apic = vcpu->arch.apic;
225         int highest_irr;
226
227         /* This may race with setting of irr in __apic_accept_irq() and
228          * value returned may be wrong, but kvm_vcpu_kick() in __apic_accept_irq
229          * will cause vmexit immediately and the value will be recalculated
230          * on the next vmentry.
231          */
232         if (!apic)
233                 return 0;
234         highest_irr = apic_find_highest_irr(apic);
235
236         return highest_irr;
237 }
238
239 static int __apic_accept_irq(struct kvm_lapic *apic, int delivery_mode,
240                              int vector, int level, int trig_mode);
241
242 int kvm_apic_set_irq(struct kvm_vcpu *vcpu, struct kvm_lapic_irq *irq)
243 {
244         struct kvm_lapic *apic = vcpu->arch.apic;
245
246         return __apic_accept_irq(apic, irq->delivery_mode, irq->vector,
247                         irq->level, irq->trig_mode);
248 }
249
250 static inline int apic_find_highest_isr(struct kvm_lapic *apic)
251 {
252         int result;
253
254         result = find_highest_vector(apic->regs + APIC_ISR);
255         ASSERT(result == -1 || result >= 16);
256
257         return result;
258 }
259
260 static void apic_update_ppr(struct kvm_lapic *apic)
261 {
262         u32 tpr, isrv, ppr;
263         int isr;
264
265         tpr = apic_get_reg(apic, APIC_TASKPRI);
266         isr = apic_find_highest_isr(apic);
267         isrv = (isr != -1) ? isr : 0;
268
269         if ((tpr & 0xf0) >= (isrv & 0xf0))
270                 ppr = tpr & 0xff;
271         else
272                 ppr = isrv & 0xf0;
273
274         apic_debug("vlapic %p, ppr 0x%x, isr 0x%x, isrv 0x%x",
275                    apic, ppr, isr, isrv);
276
277         apic_set_reg(apic, APIC_PROCPRI, ppr);
278 }
279
280 static void apic_set_tpr(struct kvm_lapic *apic, u32 tpr)
281 {
282         apic_set_reg(apic, APIC_TASKPRI, tpr);
283         apic_update_ppr(apic);
284 }
285
286 int kvm_apic_match_physical_addr(struct kvm_lapic *apic, u16 dest)
287 {
288         return dest == 0xff || kvm_apic_id(apic) == dest;
289 }
290
291 int kvm_apic_match_logical_addr(struct kvm_lapic *apic, u8 mda)
292 {
293         int result = 0;
294         u32 logical_id;
295
296         if (apic_x2apic_mode(apic)) {
297                 logical_id = apic_get_reg(apic, APIC_LDR);
298                 return logical_id & mda;
299         }
300
301         logical_id = GET_APIC_LOGICAL_ID(apic_get_reg(apic, APIC_LDR));
302
303         switch (apic_get_reg(apic, APIC_DFR)) {
304         case APIC_DFR_FLAT:
305                 if (logical_id & mda)
306                         result = 1;
307                 break;
308         case APIC_DFR_CLUSTER:
309                 if (((logical_id >> 4) == (mda >> 0x4))
310                     && (logical_id & mda & 0xf))
311                         result = 1;
312                 break;
313         default:
314                 printk(KERN_WARNING "Bad DFR vcpu %d: %08x\n",
315                        apic->vcpu->vcpu_id, apic_get_reg(apic, APIC_DFR));
316                 break;
317         }
318
319         return result;
320 }
321
322 int kvm_apic_match_dest(struct kvm_vcpu *vcpu, struct kvm_lapic *source,
323                            int short_hand, int dest, int dest_mode)
324 {
325         int result = 0;
326         struct kvm_lapic *target = vcpu->arch.apic;
327
328         apic_debug("target %p, source %p, dest 0x%x, "
329                    "dest_mode 0x%x, short_hand 0x%x\n",
330                    target, source, dest, dest_mode, short_hand);
331
332         ASSERT(target);
333         switch (short_hand) {
334         case APIC_DEST_NOSHORT:
335                 if (dest_mode == 0)
336                         /* Physical mode. */
337                         result = kvm_apic_match_physical_addr(target, dest);
338                 else
339                         /* Logical mode. */
340                         result = kvm_apic_match_logical_addr(target, dest);
341                 break;
342         case APIC_DEST_SELF:
343                 result = (target == source);
344                 break;
345         case APIC_DEST_ALLINC:
346                 result = 1;
347                 break;
348         case APIC_DEST_ALLBUT:
349                 result = (target != source);
350                 break;
351         default:
352                 printk(KERN_WARNING "Bad dest shorthand value %x\n",
353                        short_hand);
354                 break;
355         }
356
357         return result;
358 }
359
360 /*
361  * Add a pending IRQ into lapic.
362  * Return 1 if successfully added and 0 if discarded.
363  */
364 static int __apic_accept_irq(struct kvm_lapic *apic, int delivery_mode,
365                              int vector, int level, int trig_mode)
366 {
367         int result = 0;
368         struct kvm_vcpu *vcpu = apic->vcpu;
369
370         switch (delivery_mode) {
371         case APIC_DM_LOWEST:
372                 vcpu->arch.apic_arb_prio++;
373         case APIC_DM_FIXED:
374                 /* FIXME add logic for vcpu on reset */
375                 if (unlikely(!apic_enabled(apic)))
376                         break;
377
378                 if (trig_mode) {
379                         apic_debug("level trig mode for vector %d", vector);
380                         apic_set_vector(vector, apic->regs + APIC_TMR);
381                 } else
382                         apic_clear_vector(vector, apic->regs + APIC_TMR);
383
384                 result = !apic_test_and_set_irr(vector, apic);
385                 trace_kvm_apic_accept_irq(vcpu->vcpu_id, delivery_mode,
386                                           trig_mode, vector, !result);
387                 if (!result) {
388                         if (trig_mode)
389                                 apic_debug("level trig mode repeatedly for "
390                                                 "vector %d", vector);
391                         break;
392                 }
393
394                 kvm_vcpu_kick(vcpu);
395                 break;
396
397         case APIC_DM_REMRD:
398                 printk(KERN_DEBUG "Ignoring delivery mode 3\n");
399                 break;
400
401         case APIC_DM_SMI:
402                 printk(KERN_DEBUG "Ignoring guest SMI\n");
403                 break;
404
405         case APIC_DM_NMI:
406                 result = 1;
407                 kvm_inject_nmi(vcpu);
408                 kvm_vcpu_kick(vcpu);
409                 break;
410
411         case APIC_DM_INIT:
412                 if (level) {
413                         result = 1;
414                         if (vcpu->arch.mp_state == KVM_MP_STATE_RUNNABLE)
415                                 printk(KERN_DEBUG
416                                        "INIT on a runnable vcpu %d\n",
417                                        vcpu->vcpu_id);
418                         vcpu->arch.mp_state = KVM_MP_STATE_INIT_RECEIVED;
419                         kvm_vcpu_kick(vcpu);
420                 } else {
421                         apic_debug("Ignoring de-assert INIT to vcpu %d\n",
422                                    vcpu->vcpu_id);
423                 }
424                 break;
425
426         case APIC_DM_STARTUP:
427                 apic_debug("SIPI to vcpu %d vector 0x%02x\n",
428                            vcpu->vcpu_id, vector);
429                 if (vcpu->arch.mp_state == KVM_MP_STATE_INIT_RECEIVED) {
430                         result = 1;
431                         vcpu->arch.sipi_vector = vector;
432                         vcpu->arch.mp_state = KVM_MP_STATE_SIPI_RECEIVED;
433                         kvm_vcpu_kick(vcpu);
434                 }
435                 break;
436
437         case APIC_DM_EXTINT:
438                 /*
439                  * Should only be called by kvm_apic_local_deliver() with LVT0,
440                  * before NMI watchdog was enabled. Already handled by
441                  * kvm_apic_accept_pic_intr().
442                  */
443                 break;
444
445         default:
446                 printk(KERN_ERR "TODO: unsupported delivery mode %x\n",
447                        delivery_mode);
448                 break;
449         }
450         return result;
451 }
452
453 int kvm_apic_compare_prio(struct kvm_vcpu *vcpu1, struct kvm_vcpu *vcpu2)
454 {
455         return vcpu1->arch.apic_arb_prio - vcpu2->arch.apic_arb_prio;
456 }
457
458 static void apic_set_eoi(struct kvm_lapic *apic)
459 {
460         int vector = apic_find_highest_isr(apic);
461         int trigger_mode;
462         /*
463          * Not every write EOI will has corresponding ISR,
464          * one example is when Kernel check timer on setup_IO_APIC
465          */
466         if (vector == -1)
467                 return;
468
469         apic_clear_vector(vector, apic->regs + APIC_ISR);
470         apic_update_ppr(apic);
471
472         if (apic_test_and_clear_vector(vector, apic->regs + APIC_TMR))
473                 trigger_mode = IOAPIC_LEVEL_TRIG;
474         else
475                 trigger_mode = IOAPIC_EDGE_TRIG;
476         if (!(apic_get_reg(apic, APIC_SPIV) & APIC_SPIV_DIRECTED_EOI))
477                 kvm_ioapic_update_eoi(apic->vcpu->kvm, vector, trigger_mode);
478 }
479
480 static void apic_send_ipi(struct kvm_lapic *apic)
481 {
482         u32 icr_low = apic_get_reg(apic, APIC_ICR);
483         u32 icr_high = apic_get_reg(apic, APIC_ICR2);
484         struct kvm_lapic_irq irq;
485
486         irq.vector = icr_low & APIC_VECTOR_MASK;
487         irq.delivery_mode = icr_low & APIC_MODE_MASK;
488         irq.dest_mode = icr_low & APIC_DEST_MASK;
489         irq.level = icr_low & APIC_INT_ASSERT;
490         irq.trig_mode = icr_low & APIC_INT_LEVELTRIG;
491         irq.shorthand = icr_low & APIC_SHORT_MASK;
492         if (apic_x2apic_mode(apic))
493                 irq.dest_id = icr_high;
494         else
495                 irq.dest_id = GET_APIC_DEST_FIELD(icr_high);
496
497         trace_kvm_apic_ipi(icr_low, irq.dest_id);
498
499         apic_debug("icr_high 0x%x, icr_low 0x%x, "
500                    "short_hand 0x%x, dest 0x%x, trig_mode 0x%x, level 0x%x, "
501                    "dest_mode 0x%x, delivery_mode 0x%x, vector 0x%x\n",
502                    icr_high, icr_low, irq.shorthand, irq.dest_id,
503                    irq.trig_mode, irq.level, irq.dest_mode, irq.delivery_mode,
504                    irq.vector);
505
506         kvm_irq_delivery_to_apic(apic->vcpu->kvm, apic, &irq);
507 }
508
509 static u32 apic_get_tmcct(struct kvm_lapic *apic)
510 {
511         ktime_t remaining;
512         s64 ns;
513         u32 tmcct;
514
515         ASSERT(apic != NULL);
516
517         /* if initial count is 0, current count should also be 0 */
518         if (apic_get_reg(apic, APIC_TMICT) == 0)
519                 return 0;
520
521         remaining = hrtimer_get_remaining(&apic->lapic_timer.timer);
522         if (ktime_to_ns(remaining) < 0)
523                 remaining = ktime_set(0, 0);
524
525         ns = mod_64(ktime_to_ns(remaining), apic->lapic_timer.period);
526         tmcct = div64_u64(ns,
527                          (APIC_BUS_CYCLE_NS * apic->divide_count));
528
529         return tmcct;
530 }
531
532 static void __report_tpr_access(struct kvm_lapic *apic, bool write)
533 {
534         struct kvm_vcpu *vcpu = apic->vcpu;
535         struct kvm_run *run = vcpu->run;
536
537         kvm_make_request(KVM_REQ_REPORT_TPR_ACCESS, vcpu);
538         run->tpr_access.rip = kvm_rip_read(vcpu);
539         run->tpr_access.is_write = write;
540 }
541
542 static inline void report_tpr_access(struct kvm_lapic *apic, bool write)
543 {
544         if (apic->vcpu->arch.tpr_access_reporting)
545                 __report_tpr_access(apic, write);
546 }
547
548 static u32 __apic_read(struct kvm_lapic *apic, unsigned int offset)
549 {
550         u32 val = 0;
551
552         if (offset >= LAPIC_MMIO_LENGTH)
553                 return 0;
554
555         switch (offset) {
556         case APIC_ID:
557                 if (apic_x2apic_mode(apic))
558                         val = kvm_apic_id(apic);
559                 else
560                         val = kvm_apic_id(apic) << 24;
561                 break;
562         case APIC_ARBPRI:
563                 printk(KERN_WARNING "Access APIC ARBPRI register "
564                        "which is for P6\n");
565                 break;
566
567         case APIC_TMCCT:        /* Timer CCR */
568                 val = apic_get_tmcct(apic);
569                 break;
570
571         case APIC_TASKPRI:
572                 report_tpr_access(apic, false);
573                 /* fall thru */
574         default:
575                 apic_update_ppr(apic);
576                 val = apic_get_reg(apic, offset);
577                 break;
578         }
579
580         return val;
581 }
582
583 static inline struct kvm_lapic *to_lapic(struct kvm_io_device *dev)
584 {
585         return container_of(dev, struct kvm_lapic, dev);
586 }
587
588 static int apic_reg_read(struct kvm_lapic *apic, u32 offset, int len,
589                 void *data)
590 {
591         unsigned char alignment = offset & 0xf;
592         u32 result;
593         /* this bitmask has a bit cleared for each reserver register */
594         static const u64 rmask = 0x43ff01ffffffe70cULL;
595
596         if ((alignment + len) > 4) {
597                 apic_debug("KVM_APIC_READ: alignment error %x %d\n",
598                            offset, len);
599                 return 1;
600         }
601
602         if (offset > 0x3f0 || !(rmask & (1ULL << (offset >> 4)))) {
603                 apic_debug("KVM_APIC_READ: read reserved register %x\n",
604                            offset);
605                 return 1;
606         }
607
608         result = __apic_read(apic, offset & ~0xf);
609
610         trace_kvm_apic_read(offset, result);
611
612         switch (len) {
613         case 1:
614         case 2:
615         case 4:
616                 memcpy(data, (char *)&result + alignment, len);
617                 break;
618         default:
619                 printk(KERN_ERR "Local APIC read with len = %x, "
620                        "should be 1,2, or 4 instead\n", len);
621                 break;
622         }
623         return 0;
624 }
625
626 static int apic_mmio_in_range(struct kvm_lapic *apic, gpa_t addr)
627 {
628         return apic_hw_enabled(apic) &&
629             addr >= apic->base_address &&
630             addr < apic->base_address + LAPIC_MMIO_LENGTH;
631 }
632
633 static int apic_mmio_read(struct kvm_io_device *this,
634                            gpa_t address, int len, void *data)
635 {
636         struct kvm_lapic *apic = to_lapic(this);
637         u32 offset = address - apic->base_address;
638
639         if (!apic_mmio_in_range(apic, address))
640                 return -EOPNOTSUPP;
641
642         apic_reg_read(apic, offset, len, data);
643
644         return 0;
645 }
646
647 static void update_divide_count(struct kvm_lapic *apic)
648 {
649         u32 tmp1, tmp2, tdcr;
650
651         tdcr = apic_get_reg(apic, APIC_TDCR);
652         tmp1 = tdcr & 0xf;
653         tmp2 = ((tmp1 & 0x3) | ((tmp1 & 0x8) >> 1)) + 1;
654         apic->divide_count = 0x1 << (tmp2 & 0x7);
655
656         apic_debug("timer divide count is 0x%x\n",
657                                    apic->divide_count);
658 }
659
660 static void start_apic_timer(struct kvm_lapic *apic)
661 {
662         ktime_t now = apic->lapic_timer.timer.base->get_time();
663
664         apic->lapic_timer.period = (u64)apic_get_reg(apic, APIC_TMICT) *
665                     APIC_BUS_CYCLE_NS * apic->divide_count;
666         atomic_set(&apic->lapic_timer.pending, 0);
667
668         if (!apic->lapic_timer.period)
669                 return;
670         /*
671          * Do not allow the guest to program periodic timers with small
672          * interval, since the hrtimers are not throttled by the host
673          * scheduler.
674          */
675         if (apic_lvtt_period(apic)) {
676                 if (apic->lapic_timer.period < NSEC_PER_MSEC/2)
677                         apic->lapic_timer.period = NSEC_PER_MSEC/2;
678         }
679
680         hrtimer_start(&apic->lapic_timer.timer,
681                       ktime_add_ns(now, apic->lapic_timer.period),
682                       HRTIMER_MODE_ABS);
683
684         apic_debug("%s: bus cycle is %" PRId64 "ns, now 0x%016"
685                            PRIx64 ", "
686                            "timer initial count 0x%x, period %lldns, "
687                            "expire @ 0x%016" PRIx64 ".\n", __func__,
688                            APIC_BUS_CYCLE_NS, ktime_to_ns(now),
689                            apic_get_reg(apic, APIC_TMICT),
690                            apic->lapic_timer.period,
691                            ktime_to_ns(ktime_add_ns(now,
692                                         apic->lapic_timer.period)));
693 }
694
695 static void apic_manage_nmi_watchdog(struct kvm_lapic *apic, u32 lvt0_val)
696 {
697         int nmi_wd_enabled = apic_lvt_nmi_mode(apic_get_reg(apic, APIC_LVT0));
698
699         if (apic_lvt_nmi_mode(lvt0_val)) {
700                 if (!nmi_wd_enabled) {
701                         apic_debug("Receive NMI setting on APIC_LVT0 "
702                                    "for cpu %d\n", apic->vcpu->vcpu_id);
703                         apic->vcpu->kvm->arch.vapics_in_nmi_mode++;
704                 }
705         } else if (nmi_wd_enabled)
706                 apic->vcpu->kvm->arch.vapics_in_nmi_mode--;
707 }
708
709 static int apic_reg_write(struct kvm_lapic *apic, u32 reg, u32 val)
710 {
711         int ret = 0;
712
713         trace_kvm_apic_write(reg, val);
714
715         switch (reg) {
716         case APIC_ID:           /* Local APIC ID */
717                 if (!apic_x2apic_mode(apic))
718                         apic_set_reg(apic, APIC_ID, val);
719                 else
720                         ret = 1;
721                 break;
722
723         case APIC_TASKPRI:
724                 report_tpr_access(apic, true);
725                 apic_set_tpr(apic, val & 0xff);
726                 break;
727
728         case APIC_EOI:
729                 apic_set_eoi(apic);
730                 break;
731
732         case APIC_LDR:
733                 if (!apic_x2apic_mode(apic))
734                         apic_set_reg(apic, APIC_LDR, val & APIC_LDR_MASK);
735                 else
736                         ret = 1;
737                 break;
738
739         case APIC_DFR:
740                 if (!apic_x2apic_mode(apic))
741                         apic_set_reg(apic, APIC_DFR, val | 0x0FFFFFFF);
742                 else
743                         ret = 1;
744                 break;
745
746         case APIC_SPIV: {
747                 u32 mask = 0x3ff;
748                 if (apic_get_reg(apic, APIC_LVR) & APIC_LVR_DIRECTED_EOI)
749                         mask |= APIC_SPIV_DIRECTED_EOI;
750                 apic_set_reg(apic, APIC_SPIV, val & mask);
751                 if (!(val & APIC_SPIV_APIC_ENABLED)) {
752                         int i;
753                         u32 lvt_val;
754
755                         for (i = 0; i < APIC_LVT_NUM; i++) {
756                                 lvt_val = apic_get_reg(apic,
757                                                        APIC_LVTT + 0x10 * i);
758                                 apic_set_reg(apic, APIC_LVTT + 0x10 * i,
759                                              lvt_val | APIC_LVT_MASKED);
760                         }
761                         atomic_set(&apic->lapic_timer.pending, 0);
762
763                 }
764                 break;
765         }
766         case APIC_ICR:
767                 /* No delay here, so we always clear the pending bit */
768                 apic_set_reg(apic, APIC_ICR, val & ~(1 << 12));
769                 apic_send_ipi(apic);
770                 break;
771
772         case APIC_ICR2:
773                 if (!apic_x2apic_mode(apic))
774                         val &= 0xff000000;
775                 apic_set_reg(apic, APIC_ICR2, val);
776                 break;
777
778         case APIC_LVT0:
779                 apic_manage_nmi_watchdog(apic, val);
780         case APIC_LVTT:
781         case APIC_LVTTHMR:
782         case APIC_LVTPC:
783         case APIC_LVT1:
784         case APIC_LVTERR:
785                 /* TODO: Check vector */
786                 if (!apic_sw_enabled(apic))
787                         val |= APIC_LVT_MASKED;
788
789                 val &= apic_lvt_mask[(reg - APIC_LVTT) >> 4];
790                 apic_set_reg(apic, reg, val);
791
792                 break;
793
794         case APIC_TMICT:
795                 hrtimer_cancel(&apic->lapic_timer.timer);
796                 apic_set_reg(apic, APIC_TMICT, val);
797                 start_apic_timer(apic);
798                 break;
799
800         case APIC_TDCR:
801                 if (val & 4)
802                         printk(KERN_ERR "KVM_WRITE:TDCR %x\n", val);
803                 apic_set_reg(apic, APIC_TDCR, val);
804                 update_divide_count(apic);
805                 break;
806
807         case APIC_ESR:
808                 if (apic_x2apic_mode(apic) && val != 0) {
809                         printk(KERN_ERR "KVM_WRITE:ESR not zero %x\n", val);
810                         ret = 1;
811                 }
812                 break;
813
814         case APIC_SELF_IPI:
815                 if (apic_x2apic_mode(apic)) {
816                         apic_reg_write(apic, APIC_ICR, 0x40000 | (val & 0xff));
817                 } else
818                         ret = 1;
819                 break;
820         default:
821                 ret = 1;
822                 break;
823         }
824         if (ret)
825                 apic_debug("Local APIC Write to read-only register %x\n", reg);
826         return ret;
827 }
828
829 static int apic_mmio_write(struct kvm_io_device *this,
830                             gpa_t address, int len, const void *data)
831 {
832         struct kvm_lapic *apic = to_lapic(this);
833         unsigned int offset = address - apic->base_address;
834         u32 val;
835
836         if (!apic_mmio_in_range(apic, address))
837                 return -EOPNOTSUPP;
838
839         /*
840          * APIC register must be aligned on 128-bits boundary.
841          * 32/64/128 bits registers must be accessed thru 32 bits.
842          * Refer SDM 8.4.1
843          */
844         if (len != 4 || (offset & 0xf)) {
845                 /* Don't shout loud, $infamous_os would cause only noise. */
846                 apic_debug("apic write: bad size=%d %lx\n", len, (long)address);
847                 return 0;
848         }
849
850         val = *(u32*)data;
851
852         /* too common printing */
853         if (offset != APIC_EOI)
854                 apic_debug("%s: offset 0x%x with length 0x%x, and value is "
855                            "0x%x\n", __func__, offset, len, val);
856
857         apic_reg_write(apic, offset & 0xff0, val);
858
859         return 0;
860 }
861
862 void kvm_free_lapic(struct kvm_vcpu *vcpu)
863 {
864         if (!vcpu->arch.apic)
865                 return;
866
867         hrtimer_cancel(&vcpu->arch.apic->lapic_timer.timer);
868
869         if (vcpu->arch.apic->regs_page)
870                 __free_page(vcpu->arch.apic->regs_page);
871
872         kfree(vcpu->arch.apic);
873 }
874
875 /*
876  *----------------------------------------------------------------------
877  * LAPIC interface
878  *----------------------------------------------------------------------
879  */
880
881 void kvm_lapic_set_tpr(struct kvm_vcpu *vcpu, unsigned long cr8)
882 {
883         struct kvm_lapic *apic = vcpu->arch.apic;
884
885         if (!apic)
886                 return;
887         apic_set_tpr(apic, ((cr8 & 0x0f) << 4)
888                      | (apic_get_reg(apic, APIC_TASKPRI) & 4));
889 }
890
891 u64 kvm_lapic_get_cr8(struct kvm_vcpu *vcpu)
892 {
893         struct kvm_lapic *apic = vcpu->arch.apic;
894         u64 tpr;
895
896         if (!apic)
897                 return 0;
898         tpr = (u64) apic_get_reg(apic, APIC_TASKPRI);
899
900         return (tpr & 0xf0) >> 4;
901 }
902
903 void kvm_lapic_set_base(struct kvm_vcpu *vcpu, u64 value)
904 {
905         struct kvm_lapic *apic = vcpu->arch.apic;
906
907         if (!apic) {
908                 value |= MSR_IA32_APICBASE_BSP;
909                 vcpu->arch.apic_base = value;
910                 return;
911         }
912
913         if (!kvm_vcpu_is_bsp(apic->vcpu))
914                 value &= ~MSR_IA32_APICBASE_BSP;
915
916         vcpu->arch.apic_base = value;
917         if (apic_x2apic_mode(apic)) {
918                 u32 id = kvm_apic_id(apic);
919                 u32 ldr = ((id & ~0xf) << 16) | (1 << (id & 0xf));
920                 apic_set_reg(apic, APIC_LDR, ldr);
921         }
922         apic->base_address = apic->vcpu->arch.apic_base &
923                              MSR_IA32_APICBASE_BASE;
924
925         /* with FSB delivery interrupt, we can restart APIC functionality */
926         apic_debug("apic base msr is 0x%016" PRIx64 ", and base address is "
927                    "0x%lx.\n", apic->vcpu->arch.apic_base, apic->base_address);
928
929 }
930
931 void kvm_lapic_reset(struct kvm_vcpu *vcpu)
932 {
933         struct kvm_lapic *apic;
934         int i;
935
936         apic_debug("%s\n", __func__);
937
938         ASSERT(vcpu);
939         apic = vcpu->arch.apic;
940         ASSERT(apic != NULL);
941
942         /* Stop the timer in case it's a reset to an active apic */
943         hrtimer_cancel(&apic->lapic_timer.timer);
944
945         apic_set_reg(apic, APIC_ID, vcpu->vcpu_id << 24);
946         kvm_apic_set_version(apic->vcpu);
947
948         for (i = 0; i < APIC_LVT_NUM; i++)
949                 apic_set_reg(apic, APIC_LVTT + 0x10 * i, APIC_LVT_MASKED);
950         apic_set_reg(apic, APIC_LVT0,
951                      SET_APIC_DELIVERY_MODE(0, APIC_MODE_EXTINT));
952
953         apic_set_reg(apic, APIC_DFR, 0xffffffffU);
954         apic_set_reg(apic, APIC_SPIV, 0xff);
955         apic_set_reg(apic, APIC_TASKPRI, 0);
956         apic_set_reg(apic, APIC_LDR, 0);
957         apic_set_reg(apic, APIC_ESR, 0);
958         apic_set_reg(apic, APIC_ICR, 0);
959         apic_set_reg(apic, APIC_ICR2, 0);
960         apic_set_reg(apic, APIC_TDCR, 0);
961         apic_set_reg(apic, APIC_TMICT, 0);
962         for (i = 0; i < 8; i++) {
963                 apic_set_reg(apic, APIC_IRR + 0x10 * i, 0);
964                 apic_set_reg(apic, APIC_ISR + 0x10 * i, 0);
965                 apic_set_reg(apic, APIC_TMR + 0x10 * i, 0);
966         }
967         apic->irr_pending = false;
968         update_divide_count(apic);
969         atomic_set(&apic->lapic_timer.pending, 0);
970         if (kvm_vcpu_is_bsp(vcpu))
971                 vcpu->arch.apic_base |= MSR_IA32_APICBASE_BSP;
972         apic_update_ppr(apic);
973
974         vcpu->arch.apic_arb_prio = 0;
975
976         apic_debug(KERN_INFO "%s: vcpu=%p, id=%d, base_msr="
977                    "0x%016" PRIx64 ", base_address=0x%0lx.\n", __func__,
978                    vcpu, kvm_apic_id(apic),
979                    vcpu->arch.apic_base, apic->base_address);
980 }
981
982 bool kvm_apic_present(struct kvm_vcpu *vcpu)
983 {
984         return vcpu->arch.apic && apic_hw_enabled(vcpu->arch.apic);
985 }
986
987 int kvm_lapic_enabled(struct kvm_vcpu *vcpu)
988 {
989         return kvm_apic_present(vcpu) && apic_sw_enabled(vcpu->arch.apic);
990 }
991
992 /*
993  *----------------------------------------------------------------------
994  * timer interface
995  *----------------------------------------------------------------------
996  */
997
998 static bool lapic_is_periodic(struct kvm_timer *ktimer)
999 {
1000         struct kvm_lapic *apic = container_of(ktimer, struct kvm_lapic,
1001                                               lapic_timer);
1002         return apic_lvtt_period(apic);
1003 }
1004
1005 int apic_has_pending_timer(struct kvm_vcpu *vcpu)
1006 {
1007         struct kvm_lapic *lapic = vcpu->arch.apic;
1008
1009         if (lapic && apic_enabled(lapic) && apic_lvt_enabled(lapic, APIC_LVTT))
1010                 return atomic_read(&lapic->lapic_timer.pending);
1011
1012         return 0;
1013 }
1014
1015 static int kvm_apic_local_deliver(struct kvm_lapic *apic, int lvt_type)
1016 {
1017         u32 reg = apic_get_reg(apic, lvt_type);
1018         int vector, mode, trig_mode;
1019
1020         if (apic_hw_enabled(apic) && !(reg & APIC_LVT_MASKED)) {
1021                 vector = reg & APIC_VECTOR_MASK;
1022                 mode = reg & APIC_MODE_MASK;
1023                 trig_mode = reg & APIC_LVT_LEVEL_TRIGGER;
1024                 return __apic_accept_irq(apic, mode, vector, 1, trig_mode);
1025         }
1026         return 0;
1027 }
1028
1029 void kvm_apic_nmi_wd_deliver(struct kvm_vcpu *vcpu)
1030 {
1031         struct kvm_lapic *apic = vcpu->arch.apic;
1032
1033         if (apic)
1034                 kvm_apic_local_deliver(apic, APIC_LVT0);
1035 }
1036
1037 static struct kvm_timer_ops lapic_timer_ops = {
1038         .is_periodic = lapic_is_periodic,
1039 };
1040
1041 static const struct kvm_io_device_ops apic_mmio_ops = {
1042         .read     = apic_mmio_read,
1043         .write    = apic_mmio_write,
1044 };
1045
1046 int kvm_create_lapic(struct kvm_vcpu *vcpu)
1047 {
1048         struct kvm_lapic *apic;
1049
1050         ASSERT(vcpu != NULL);
1051         apic_debug("apic_init %d\n", vcpu->vcpu_id);
1052
1053         apic = kzalloc(sizeof(*apic), GFP_KERNEL);
1054         if (!apic)
1055                 goto nomem;
1056
1057         vcpu->arch.apic = apic;
1058
1059         apic->regs_page = alloc_page(GFP_KERNEL|__GFP_ZERO);
1060         if (apic->regs_page == NULL) {
1061                 printk(KERN_ERR "malloc apic regs error for vcpu %x\n",
1062                        vcpu->vcpu_id);
1063                 goto nomem_free_apic;
1064         }
1065         apic->regs = page_address(apic->regs_page);
1066         apic->vcpu = vcpu;
1067
1068         hrtimer_init(&apic->lapic_timer.timer, CLOCK_MONOTONIC,
1069                      HRTIMER_MODE_ABS);
1070         apic->lapic_timer.timer.function = kvm_timer_fn;
1071         apic->lapic_timer.t_ops = &lapic_timer_ops;
1072         apic->lapic_timer.kvm = vcpu->kvm;
1073         apic->lapic_timer.vcpu = vcpu;
1074
1075         apic->base_address = APIC_DEFAULT_PHYS_BASE;
1076         vcpu->arch.apic_base = APIC_DEFAULT_PHYS_BASE;
1077
1078         kvm_lapic_reset(vcpu);
1079         kvm_iodevice_init(&apic->dev, &apic_mmio_ops);
1080
1081         return 0;
1082 nomem_free_apic:
1083         kfree(apic);
1084 nomem:
1085         return -ENOMEM;
1086 }
1087
1088 int kvm_apic_has_interrupt(struct kvm_vcpu *vcpu)
1089 {
1090         struct kvm_lapic *apic = vcpu->arch.apic;
1091         int highest_irr;
1092
1093         if (!apic || !apic_enabled(apic))
1094                 return -1;
1095
1096         apic_update_ppr(apic);
1097         highest_irr = apic_find_highest_irr(apic);
1098         if ((highest_irr == -1) ||
1099             ((highest_irr & 0xF0) <= apic_get_reg(apic, APIC_PROCPRI)))
1100                 return -1;
1101         return highest_irr;
1102 }
1103
1104 int kvm_apic_accept_pic_intr(struct kvm_vcpu *vcpu)
1105 {
1106         u32 lvt0 = apic_get_reg(vcpu->arch.apic, APIC_LVT0);
1107         int r = 0;
1108
1109         if (!apic_hw_enabled(vcpu->arch.apic))
1110                 r = 1;
1111         if ((lvt0 & APIC_LVT_MASKED) == 0 &&
1112             GET_APIC_DELIVERY_MODE(lvt0) == APIC_MODE_EXTINT)
1113                 r = 1;
1114         return r;
1115 }
1116
1117 void kvm_inject_apic_timer_irqs(struct kvm_vcpu *vcpu)
1118 {
1119         struct kvm_lapic *apic = vcpu->arch.apic;
1120
1121         if (apic && atomic_read(&apic->lapic_timer.pending) > 0) {
1122                 if (kvm_apic_local_deliver(apic, APIC_LVTT))
1123                         atomic_dec(&apic->lapic_timer.pending);
1124         }
1125 }
1126
1127 int kvm_get_apic_interrupt(struct kvm_vcpu *vcpu)
1128 {
1129         int vector = kvm_apic_has_interrupt(vcpu);
1130         struct kvm_lapic *apic = vcpu->arch.apic;
1131
1132         if (vector == -1)
1133                 return -1;
1134
1135         apic_set_vector(vector, apic->regs + APIC_ISR);
1136         apic_update_ppr(apic);
1137         apic_clear_irr(vector, apic);
1138         return vector;
1139 }
1140
1141 void kvm_apic_post_state_restore(struct kvm_vcpu *vcpu)
1142 {
1143         struct kvm_lapic *apic = vcpu->arch.apic;
1144
1145         apic->base_address = vcpu->arch.apic_base &
1146                              MSR_IA32_APICBASE_BASE;
1147         kvm_apic_set_version(vcpu);
1148
1149         apic_update_ppr(apic);
1150         hrtimer_cancel(&apic->lapic_timer.timer);
1151         update_divide_count(apic);
1152         start_apic_timer(apic);
1153         apic->irr_pending = true;
1154 }
1155
1156 void __kvm_migrate_apic_timer(struct kvm_vcpu *vcpu)
1157 {
1158         struct kvm_lapic *apic = vcpu->arch.apic;
1159         struct hrtimer *timer;
1160
1161         if (!apic)
1162                 return;
1163
1164         timer = &apic->lapic_timer.timer;
1165         if (hrtimer_cancel(timer))
1166                 hrtimer_start_expires(timer, HRTIMER_MODE_ABS);
1167 }
1168
1169 void kvm_lapic_sync_from_vapic(struct kvm_vcpu *vcpu)
1170 {
1171         u32 data;
1172         void *vapic;
1173
1174         if (!irqchip_in_kernel(vcpu->kvm) || !vcpu->arch.apic->vapic_addr)
1175                 return;
1176
1177         vapic = kmap_atomic(vcpu->arch.apic->vapic_page, KM_USER0);
1178         data = *(u32 *)(vapic + offset_in_page(vcpu->arch.apic->vapic_addr));
1179         kunmap_atomic(vapic, KM_USER0);
1180
1181         apic_set_tpr(vcpu->arch.apic, data & 0xff);
1182 }
1183
1184 void kvm_lapic_sync_to_vapic(struct kvm_vcpu *vcpu)
1185 {
1186         u32 data, tpr;
1187         int max_irr, max_isr;
1188         struct kvm_lapic *apic;
1189         void *vapic;
1190
1191         if (!irqchip_in_kernel(vcpu->kvm) || !vcpu->arch.apic->vapic_addr)
1192                 return;
1193
1194         apic = vcpu->arch.apic;
1195         tpr = apic_get_reg(apic, APIC_TASKPRI) & 0xff;
1196         max_irr = apic_find_highest_irr(apic);
1197         if (max_irr < 0)
1198                 max_irr = 0;
1199         max_isr = apic_find_highest_isr(apic);
1200         if (max_isr < 0)
1201                 max_isr = 0;
1202         data = (tpr & 0xff) | ((max_isr & 0xf0) << 8) | (max_irr << 24);
1203
1204         vapic = kmap_atomic(vcpu->arch.apic->vapic_page, KM_USER0);
1205         *(u32 *)(vapic + offset_in_page(vcpu->arch.apic->vapic_addr)) = data;
1206         kunmap_atomic(vapic, KM_USER0);
1207 }
1208
1209 void kvm_lapic_set_vapic_addr(struct kvm_vcpu *vcpu, gpa_t vapic_addr)
1210 {
1211         if (!irqchip_in_kernel(vcpu->kvm))
1212                 return;
1213
1214         vcpu->arch.apic->vapic_addr = vapic_addr;
1215 }
1216
1217 int kvm_x2apic_msr_write(struct kvm_vcpu *vcpu, u32 msr, u64 data)
1218 {
1219         struct kvm_lapic *apic = vcpu->arch.apic;
1220         u32 reg = (msr - APIC_BASE_MSR) << 4;
1221
1222         if (!irqchip_in_kernel(vcpu->kvm) || !apic_x2apic_mode(apic))
1223                 return 1;
1224
1225         /* if this is ICR write vector before command */
1226         if (msr == 0x830)
1227                 apic_reg_write(apic, APIC_ICR2, (u32)(data >> 32));
1228         return apic_reg_write(apic, reg, (u32)data);
1229 }
1230
1231 int kvm_x2apic_msr_read(struct kvm_vcpu *vcpu, u32 msr, u64 *data)
1232 {
1233         struct kvm_lapic *apic = vcpu->arch.apic;
1234         u32 reg = (msr - APIC_BASE_MSR) << 4, low, high = 0;
1235
1236         if (!irqchip_in_kernel(vcpu->kvm) || !apic_x2apic_mode(apic))
1237                 return 1;
1238
1239         if (apic_reg_read(apic, reg, 4, &low))
1240                 return 1;
1241         if (msr == 0x830)
1242                 apic_reg_read(apic, APIC_ICR2, 4, &high);
1243
1244         *data = (((u64)high) << 32) | low;
1245
1246         return 0;
1247 }
1248
1249 int kvm_hv_vapic_msr_write(struct kvm_vcpu *vcpu, u32 reg, u64 data)
1250 {
1251         struct kvm_lapic *apic = vcpu->arch.apic;
1252
1253         if (!irqchip_in_kernel(vcpu->kvm))
1254                 return 1;
1255
1256         /* if this is ICR write vector before command */
1257         if (reg == APIC_ICR)
1258                 apic_reg_write(apic, APIC_ICR2, (u32)(data >> 32));
1259         return apic_reg_write(apic, reg, (u32)data);
1260 }
1261
1262 int kvm_hv_vapic_msr_read(struct kvm_vcpu *vcpu, u32 reg, u64 *data)
1263 {
1264         struct kvm_lapic *apic = vcpu->arch.apic;
1265         u32 low, high = 0;
1266
1267         if (!irqchip_in_kernel(vcpu->kvm))
1268                 return 1;
1269
1270         if (apic_reg_read(apic, reg, 4, &low))
1271                 return 1;
1272         if (reg == APIC_ICR)
1273                 apic_reg_read(apic, APIC_ICR2, 4, &high);
1274
1275         *data = (((u64)high) << 32) | low;
1276
1277         return 0;
1278 }