KVM: x2apic interface to lapic
[safe/jmp/linux-2.6] / 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 <asm/apicdef.h>
36 #include "kvm_cache_regs.h"
37 #include "irq.h"
38 #include "trace.h"
39 #include "x86.h"
40
41 #ifndef CONFIG_X86_64
42 #define mod_64(x, y) ((x) - (y) * div64_u64(x, y))
43 #else
44 #define mod_64(x, y) ((x) % (y))
45 #endif
46
47 #define PRId64 "d"
48 #define PRIx64 "llx"
49 #define PRIu64 "u"
50 #define PRIo64 "o"
51
52 #define APIC_BUS_CYCLE_NS 1
53
54 /* #define apic_debug(fmt,arg...) printk(KERN_WARNING fmt,##arg) */
55 #define apic_debug(fmt, arg...)
56
57 #define APIC_LVT_NUM                    6
58 /* 14 is the version for Xeon and Pentium 8.4.8*/
59 #define APIC_VERSION                    (0x14UL | ((APIC_LVT_NUM - 1) << 16))
60 #define LAPIC_MMIO_LENGTH               (1 << 12)
61 /* followed define is not in apicdef.h */
62 #define APIC_SHORT_MASK                 0xc0000
63 #define APIC_DEST_NOSHORT               0x0
64 #define APIC_DEST_MASK                  0x800
65 #define MAX_APIC_VECTOR                 256
66
67 #define VEC_POS(v) ((v) & (32 - 1))
68 #define REG_POS(v) (((v) >> 5) << 4)
69
70 static inline u32 apic_get_reg(struct kvm_lapic *apic, int reg_off)
71 {
72         return *((u32 *) (apic->regs + reg_off));
73 }
74
75 static inline void apic_set_reg(struct kvm_lapic *apic, int reg_off, u32 val)
76 {
77         *((u32 *) (apic->regs + reg_off)) = val;
78 }
79
80 static inline int apic_test_and_set_vector(int vec, void *bitmap)
81 {
82         return test_and_set_bit(VEC_POS(vec), (bitmap) + REG_POS(vec));
83 }
84
85 static inline int apic_test_and_clear_vector(int vec, void *bitmap)
86 {
87         return test_and_clear_bit(VEC_POS(vec), (bitmap) + REG_POS(vec));
88 }
89
90 static inline void apic_set_vector(int vec, void *bitmap)
91 {
92         set_bit(VEC_POS(vec), (bitmap) + REG_POS(vec));
93 }
94
95 static inline void apic_clear_vector(int vec, void *bitmap)
96 {
97         clear_bit(VEC_POS(vec), (bitmap) + REG_POS(vec));
98 }
99
100 static inline int apic_hw_enabled(struct kvm_lapic *apic)
101 {
102         return (apic)->vcpu->arch.apic_base & MSR_IA32_APICBASE_ENABLE;
103 }
104
105 static inline int  apic_sw_enabled(struct kvm_lapic *apic)
106 {
107         return apic_get_reg(apic, APIC_SPIV) & APIC_SPIV_APIC_ENABLED;
108 }
109
110 static inline int apic_enabled(struct kvm_lapic *apic)
111 {
112         return apic_sw_enabled(apic) && apic_hw_enabled(apic);
113 }
114
115 #define LVT_MASK        \
116         (APIC_LVT_MASKED | APIC_SEND_PENDING | APIC_VECTOR_MASK)
117
118 #define LINT_MASK       \
119         (LVT_MASK | APIC_MODE_MASK | APIC_INPUT_POLARITY | \
120          APIC_LVT_REMOTE_IRR | APIC_LVT_LEVEL_TRIGGER)
121
122 static inline int kvm_apic_id(struct kvm_lapic *apic)
123 {
124         return (apic_get_reg(apic, APIC_ID) >> 24) & 0xff;
125 }
126
127 static inline int apic_lvt_enabled(struct kvm_lapic *apic, int lvt_type)
128 {
129         return !(apic_get_reg(apic, lvt_type) & APIC_LVT_MASKED);
130 }
131
132 static inline int apic_lvt_vector(struct kvm_lapic *apic, int lvt_type)
133 {
134         return apic_get_reg(apic, lvt_type) & APIC_VECTOR_MASK;
135 }
136
137 static inline int apic_lvtt_period(struct kvm_lapic *apic)
138 {
139         return apic_get_reg(apic, APIC_LVTT) & APIC_LVT_TIMER_PERIODIC;
140 }
141
142 static inline int apic_lvt_nmi_mode(u32 lvt_val)
143 {
144         return (lvt_val & (APIC_MODE_MASK | APIC_LVT_MASKED)) == APIC_DM_NMI;
145 }
146
147 void kvm_apic_set_version(struct kvm_vcpu *vcpu)
148 {
149         struct kvm_lapic *apic = vcpu->arch.apic;
150         struct kvm_cpuid_entry2 *feat;
151         u32 v = APIC_VERSION;
152
153         if (!irqchip_in_kernel(vcpu->kvm))
154                 return;
155
156         feat = kvm_find_cpuid_entry(apic->vcpu, 0x1, 0);
157         if (feat && (feat->ecx & (1 << (X86_FEATURE_X2APIC & 31))))
158                 v |= APIC_LVR_DIRECTED_EOI;
159         apic_set_reg(apic, APIC_LVR, v);
160 }
161
162 static inline int apic_x2apic_mode(struct kvm_lapic *apic)
163 {
164         return apic->vcpu->arch.apic_base & X2APIC_ENABLE;
165 }
166
167 static unsigned int apic_lvt_mask[APIC_LVT_NUM] = {
168         LVT_MASK | APIC_LVT_TIMER_PERIODIC,     /* LVTT */
169         LVT_MASK | APIC_MODE_MASK,      /* LVTTHMR */
170         LVT_MASK | APIC_MODE_MASK,      /* LVTPC */
171         LINT_MASK, LINT_MASK,   /* LVT0-1 */
172         LVT_MASK                /* LVTERR */
173 };
174
175 static int find_highest_vector(void *bitmap)
176 {
177         u32 *word = bitmap;
178         int word_offset = MAX_APIC_VECTOR >> 5;
179
180         while ((word_offset != 0) && (word[(--word_offset) << 2] == 0))
181                 continue;
182
183         if (likely(!word_offset && !word[0]))
184                 return -1;
185         else
186                 return fls(word[word_offset << 2]) - 1 + (word_offset << 5);
187 }
188
189 static inline int apic_test_and_set_irr(int vec, struct kvm_lapic *apic)
190 {
191         apic->irr_pending = true;
192         return apic_test_and_set_vector(vec, apic->regs + APIC_IRR);
193 }
194
195 static inline int apic_search_irr(struct kvm_lapic *apic)
196 {
197         return find_highest_vector(apic->regs + APIC_IRR);
198 }
199
200 static inline int apic_find_highest_irr(struct kvm_lapic *apic)
201 {
202         int result;
203
204         if (!apic->irr_pending)
205                 return -1;
206
207         result = apic_search_irr(apic);
208         ASSERT(result == -1 || result >= 16);
209
210         return result;
211 }
212
213 static inline void apic_clear_irr(int vec, struct kvm_lapic *apic)
214 {
215         apic->irr_pending = false;
216         apic_clear_vector(vec, apic->regs + APIC_IRR);
217         if (apic_search_irr(apic) != -1)
218                 apic->irr_pending = true;
219 }
220
221 int kvm_lapic_find_highest_irr(struct kvm_vcpu *vcpu)
222 {
223         struct kvm_lapic *apic = vcpu->arch.apic;
224         int highest_irr;
225
226         /* This may race with setting of irr in __apic_accept_irq() and
227          * value returned may be wrong, but kvm_vcpu_kick() in __apic_accept_irq
228          * will cause vmexit immediately and the value will be recalculated
229          * on the next vmentry.
230          */
231         if (!apic)
232                 return 0;
233         highest_irr = apic_find_highest_irr(apic);
234
235         return highest_irr;
236 }
237
238 static int __apic_accept_irq(struct kvm_lapic *apic, int delivery_mode,
239                              int vector, int level, int trig_mode);
240
241 int kvm_apic_set_irq(struct kvm_vcpu *vcpu, struct kvm_lapic_irq *irq)
242 {
243         struct kvm_lapic *apic = vcpu->arch.apic;
244
245         return __apic_accept_irq(apic, irq->delivery_mode, irq->vector,
246                         irq->level, irq->trig_mode);
247 }
248
249 static inline int apic_find_highest_isr(struct kvm_lapic *apic)
250 {
251         int result;
252
253         result = find_highest_vector(apic->regs + APIC_ISR);
254         ASSERT(result == -1 || result >= 16);
255
256         return result;
257 }
258
259 static void apic_update_ppr(struct kvm_lapic *apic)
260 {
261         u32 tpr, isrv, ppr;
262         int isr;
263
264         tpr = apic_get_reg(apic, APIC_TASKPRI);
265         isr = apic_find_highest_isr(apic);
266         isrv = (isr != -1) ? isr : 0;
267
268         if ((tpr & 0xf0) >= (isrv & 0xf0))
269                 ppr = tpr & 0xff;
270         else
271                 ppr = isrv & 0xf0;
272
273         apic_debug("vlapic %p, ppr 0x%x, isr 0x%x, isrv 0x%x",
274                    apic, ppr, isr, isrv);
275
276         apic_set_reg(apic, APIC_PROCPRI, ppr);
277 }
278
279 static void apic_set_tpr(struct kvm_lapic *apic, u32 tpr)
280 {
281         apic_set_reg(apic, APIC_TASKPRI, tpr);
282         apic_update_ppr(apic);
283 }
284
285 int kvm_apic_match_physical_addr(struct kvm_lapic *apic, u16 dest)
286 {
287         return dest == 0xff || kvm_apic_id(apic) == dest;
288 }
289
290 int kvm_apic_match_logical_addr(struct kvm_lapic *apic, u8 mda)
291 {
292         int result = 0;
293         u32 logical_id;
294
295         if (apic_x2apic_mode(apic)) {
296                 logical_id = apic_get_reg(apic, APIC_LDR);
297                 return logical_id & mda;
298         }
299
300         logical_id = GET_APIC_LOGICAL_ID(apic_get_reg(apic, APIC_LDR));
301
302         switch (apic_get_reg(apic, APIC_DFR)) {
303         case APIC_DFR_FLAT:
304                 if (logical_id & mda)
305                         result = 1;
306                 break;
307         case APIC_DFR_CLUSTER:
308                 if (((logical_id >> 4) == (mda >> 0x4))
309                     && (logical_id & mda & 0xf))
310                         result = 1;
311                 break;
312         default:
313                 printk(KERN_WARNING "Bad DFR vcpu %d: %08x\n",
314                        apic->vcpu->vcpu_id, apic_get_reg(apic, APIC_DFR));
315                 break;
316         }
317
318         return result;
319 }
320
321 int kvm_apic_match_dest(struct kvm_vcpu *vcpu, struct kvm_lapic *source,
322                            int short_hand, int dest, int dest_mode)
323 {
324         int result = 0;
325         struct kvm_lapic *target = vcpu->arch.apic;
326
327         apic_debug("target %p, source %p, dest 0x%x, "
328                    "dest_mode 0x%x, short_hand 0x%x\n",
329                    target, source, dest, dest_mode, short_hand);
330
331         ASSERT(!target);
332         switch (short_hand) {
333         case APIC_DEST_NOSHORT:
334                 if (dest_mode == 0)
335                         /* Physical mode. */
336                         result = kvm_apic_match_physical_addr(target, dest);
337                 else
338                         /* Logical mode. */
339                         result = kvm_apic_match_logical_addr(target, dest);
340                 break;
341         case APIC_DEST_SELF:
342                 result = (target == source);
343                 break;
344         case APIC_DEST_ALLINC:
345                 result = 1;
346                 break;
347         case APIC_DEST_ALLBUT:
348                 result = (target != source);
349                 break;
350         default:
351                 printk(KERN_WARNING "Bad dest shorthand value %x\n",
352                        short_hand);
353                 break;
354         }
355
356         return result;
357 }
358
359 /*
360  * Add a pending IRQ into lapic.
361  * Return 1 if successfully added and 0 if discarded.
362  */
363 static int __apic_accept_irq(struct kvm_lapic *apic, int delivery_mode,
364                              int vector, int level, int trig_mode)
365 {
366         int result = 0;
367         struct kvm_vcpu *vcpu = apic->vcpu;
368
369         switch (delivery_mode) {
370         case APIC_DM_LOWEST:
371                 vcpu->arch.apic_arb_prio++;
372         case APIC_DM_FIXED:
373                 /* FIXME add logic for vcpu on reset */
374                 if (unlikely(!apic_enabled(apic)))
375                         break;
376
377                 result = !apic_test_and_set_irr(vector, apic);
378                 if (!result) {
379                         if (trig_mode)
380                                 apic_debug("level trig mode repeatedly for "
381                                                 "vector %d", vector);
382                         break;
383                 }
384
385                 if (trig_mode) {
386                         apic_debug("level trig mode for vector %d", vector);
387                         apic_set_vector(vector, apic->regs + APIC_TMR);
388                 } else
389                         apic_clear_vector(vector, apic->regs + APIC_TMR);
390                 kvm_vcpu_kick(vcpu);
391                 break;
392
393         case APIC_DM_REMRD:
394                 printk(KERN_DEBUG "Ignoring delivery mode 3\n");
395                 break;
396
397         case APIC_DM_SMI:
398                 printk(KERN_DEBUG "Ignoring guest SMI\n");
399                 break;
400
401         case APIC_DM_NMI:
402                 result = 1;
403                 kvm_inject_nmi(vcpu);
404                 kvm_vcpu_kick(vcpu);
405                 break;
406
407         case APIC_DM_INIT:
408                 if (level) {
409                         result = 1;
410                         if (vcpu->arch.mp_state == KVM_MP_STATE_RUNNABLE)
411                                 printk(KERN_DEBUG
412                                        "INIT on a runnable vcpu %d\n",
413                                        vcpu->vcpu_id);
414                         vcpu->arch.mp_state = KVM_MP_STATE_INIT_RECEIVED;
415                         kvm_vcpu_kick(vcpu);
416                 } else {
417                         apic_debug("Ignoring de-assert INIT to vcpu %d\n",
418                                    vcpu->vcpu_id);
419                 }
420                 break;
421
422         case APIC_DM_STARTUP:
423                 apic_debug("SIPI to vcpu %d vector 0x%02x\n",
424                            vcpu->vcpu_id, vector);
425                 if (vcpu->arch.mp_state == KVM_MP_STATE_INIT_RECEIVED) {
426                         result = 1;
427                         vcpu->arch.sipi_vector = vector;
428                         vcpu->arch.mp_state = KVM_MP_STATE_SIPI_RECEIVED;
429                         kvm_vcpu_kick(vcpu);
430                 }
431                 break;
432
433         case APIC_DM_EXTINT:
434                 /*
435                  * Should only be called by kvm_apic_local_deliver() with LVT0,
436                  * before NMI watchdog was enabled. Already handled by
437                  * kvm_apic_accept_pic_intr().
438                  */
439                 break;
440
441         default:
442                 printk(KERN_ERR "TODO: unsupported delivery mode %x\n",
443                        delivery_mode);
444                 break;
445         }
446         return result;
447 }
448
449 int kvm_apic_compare_prio(struct kvm_vcpu *vcpu1, struct kvm_vcpu *vcpu2)
450 {
451         return vcpu1->arch.apic_arb_prio - vcpu2->arch.apic_arb_prio;
452 }
453
454 static void apic_set_eoi(struct kvm_lapic *apic)
455 {
456         int vector = apic_find_highest_isr(apic);
457         int trigger_mode;
458         /*
459          * Not every write EOI will has corresponding ISR,
460          * one example is when Kernel check timer on setup_IO_APIC
461          */
462         if (vector == -1)
463                 return;
464
465         apic_clear_vector(vector, apic->regs + APIC_ISR);
466         apic_update_ppr(apic);
467
468         if (apic_test_and_clear_vector(vector, apic->regs + APIC_TMR))
469                 trigger_mode = IOAPIC_LEVEL_TRIG;
470         else
471                 trigger_mode = IOAPIC_EDGE_TRIG;
472         if (!(apic_get_reg(apic, APIC_SPIV) & APIC_SPIV_DIRECTED_EOI)) {
473                 mutex_lock(&apic->vcpu->kvm->irq_lock);
474                 kvm_ioapic_update_eoi(apic->vcpu->kvm, vector, trigger_mode);
475                 mutex_unlock(&apic->vcpu->kvm->irq_lock);
476         }
477 }
478
479 static void apic_send_ipi(struct kvm_lapic *apic)
480 {
481         u32 icr_low = apic_get_reg(apic, APIC_ICR);
482         u32 icr_high = apic_get_reg(apic, APIC_ICR2);
483         struct kvm_lapic_irq irq;
484
485         irq.vector = icr_low & APIC_VECTOR_MASK;
486         irq.delivery_mode = icr_low & APIC_MODE_MASK;
487         irq.dest_mode = icr_low & APIC_DEST_MASK;
488         irq.level = icr_low & APIC_INT_ASSERT;
489         irq.trig_mode = icr_low & APIC_INT_LEVELTRIG;
490         irq.shorthand = icr_low & APIC_SHORT_MASK;
491         if (apic_x2apic_mode(apic))
492                 irq.dest_id = icr_high;
493         else
494                 irq.dest_id = GET_APIC_DEST_FIELD(icr_high);
495
496         apic_debug("icr_high 0x%x, icr_low 0x%x, "
497                    "short_hand 0x%x, dest 0x%x, trig_mode 0x%x, level 0x%x, "
498                    "dest_mode 0x%x, delivery_mode 0x%x, vector 0x%x\n",
499                    icr_high, icr_low, irq.shorthand, irq.dest_id,
500                    irq.trig_mode, irq.level, irq.dest_mode, irq.delivery_mode,
501                    irq.vector);
502
503         mutex_lock(&apic->vcpu->kvm->irq_lock);
504         kvm_irq_delivery_to_apic(apic->vcpu->kvm, apic, &irq);
505         mutex_unlock(&apic->vcpu->kvm->irq_lock);
506 }
507
508 static u32 apic_get_tmcct(struct kvm_lapic *apic)
509 {
510         ktime_t remaining;
511         s64 ns;
512         u32 tmcct;
513
514         ASSERT(apic != NULL);
515
516         /* if initial count is 0, current count should also be 0 */
517         if (apic_get_reg(apic, APIC_TMICT) == 0)
518                 return 0;
519
520         remaining = hrtimer_expires_remaining(&apic->lapic_timer.timer);
521         if (ktime_to_ns(remaining) < 0)
522                 remaining = ktime_set(0, 0);
523
524         ns = mod_64(ktime_to_ns(remaining), apic->lapic_timer.period);
525         tmcct = div64_u64(ns,
526                          (APIC_BUS_CYCLE_NS * apic->divide_count));
527
528         return tmcct;
529 }
530
531 static void __report_tpr_access(struct kvm_lapic *apic, bool write)
532 {
533         struct kvm_vcpu *vcpu = apic->vcpu;
534         struct kvm_run *run = vcpu->run;
535
536         set_bit(KVM_REQ_REPORT_TPR_ACCESS, &vcpu->requests);
537         run->tpr_access.rip = kvm_rip_read(vcpu);
538         run->tpr_access.is_write = write;
539 }
540
541 static inline void report_tpr_access(struct kvm_lapic *apic, bool write)
542 {
543         if (apic->vcpu->arch.tpr_access_reporting)
544                 __report_tpr_access(apic, write);
545 }
546
547 static u32 __apic_read(struct kvm_lapic *apic, unsigned int offset)
548 {
549         u32 val = 0;
550
551         if (offset >= LAPIC_MMIO_LENGTH)
552                 return 0;
553
554         switch (offset) {
555         case APIC_ID:
556                 if (apic_x2apic_mode(apic))
557                         val = kvm_apic_id(apic);
558                 else
559                         val = kvm_apic_id(apic) << 24;
560                 break;
561         case APIC_ARBPRI:
562                 printk(KERN_WARNING "Access APIC ARBPRI register "
563                        "which is for P6\n");
564                 break;
565
566         case APIC_TMCCT:        /* Timer CCR */
567                 val = apic_get_tmcct(apic);
568                 break;
569
570         case APIC_TASKPRI:
571                 report_tpr_access(apic, false);
572                 /* fall thru */
573         default:
574                 apic_update_ppr(apic);
575                 val = apic_get_reg(apic, offset);
576                 break;
577         }
578
579         return val;
580 }
581
582 static inline struct kvm_lapic *to_lapic(struct kvm_io_device *dev)
583 {
584         return container_of(dev, struct kvm_lapic, dev);
585 }
586
587 static int apic_reg_read(struct kvm_lapic *apic, u32 offset, int len,
588                 void *data)
589 {
590         unsigned char alignment = offset & 0xf;
591         u32 result;
592         /* this bitmask has a bit cleared for each reserver register */
593         static const u64 rmask = 0x43ff01ffffffe70cULL;
594
595         if ((alignment + len) > 4) {
596                 printk(KERN_ERR "KVM_APIC_READ: alignment error %x %d\n",
597                                 offset, len);
598                 return 1;
599         }
600
601         if (offset > 0x3f0 || !(rmask & (1ULL << (offset >> 4)))) {
602                 printk(KERN_ERR "KVM_APIC_READ: read reserved register %x\n",
603                                 offset);
604                 return 1;
605         }
606
607         result = __apic_read(apic, offset & ~0xf);
608
609         trace_kvm_apic_read(offset, result);
610
611         switch (len) {
612         case 1:
613         case 2:
614         case 4:
615                 memcpy(data, (char *)&result + alignment, len);
616                 break;
617         default:
618                 printk(KERN_ERR "Local APIC read with len = %x, "
619                        "should be 1,2, or 4 instead\n", len);
620                 break;
621         }
622         return 0;
623 }
624
625 static int apic_mmio_in_range(struct kvm_lapic *apic, gpa_t addr)
626 {
627         return apic_hw_enabled(apic) &&
628             addr >= apic->base_address &&
629             addr < apic->base_address + LAPIC_MMIO_LENGTH;
630 }
631
632 static int apic_mmio_read(struct kvm_io_device *this,
633                            gpa_t address, int len, void *data)
634 {
635         struct kvm_lapic *apic = to_lapic(this);
636         u32 offset = address - apic->base_address;
637
638         if (!apic_mmio_in_range(apic, address))
639                 return -EOPNOTSUPP;
640
641         apic_reg_read(apic, offset, len, data);
642
643         return 0;
644 }
645
646 static void update_divide_count(struct kvm_lapic *apic)
647 {
648         u32 tmp1, tmp2, tdcr;
649
650         tdcr = apic_get_reg(apic, APIC_TDCR);
651         tmp1 = tdcr & 0xf;
652         tmp2 = ((tmp1 & 0x3) | ((tmp1 & 0x8) >> 1)) + 1;
653         apic->divide_count = 0x1 << (tmp2 & 0x7);
654
655         apic_debug("timer divide count is 0x%x\n",
656                                    apic->divide_count);
657 }
658
659 static void start_apic_timer(struct kvm_lapic *apic)
660 {
661         ktime_t now = apic->lapic_timer.timer.base->get_time();
662
663         apic->lapic_timer.period = apic_get_reg(apic, APIC_TMICT) *
664                     APIC_BUS_CYCLE_NS * apic->divide_count;
665         atomic_set(&apic->lapic_timer.pending, 0);
666
667         if (!apic->lapic_timer.period)
668                 return;
669
670         hrtimer_start(&apic->lapic_timer.timer,
671                       ktime_add_ns(now, apic->lapic_timer.period),
672                       HRTIMER_MODE_ABS);
673
674         apic_debug("%s: bus cycle is %" PRId64 "ns, now 0x%016"
675                            PRIx64 ", "
676                            "timer initial count 0x%x, period %lldns, "
677                            "expire @ 0x%016" PRIx64 ".\n", __func__,
678                            APIC_BUS_CYCLE_NS, ktime_to_ns(now),
679                            apic_get_reg(apic, APIC_TMICT),
680                            apic->lapic_timer.period,
681                            ktime_to_ns(ktime_add_ns(now,
682                                         apic->lapic_timer.period)));
683 }
684
685 static void apic_manage_nmi_watchdog(struct kvm_lapic *apic, u32 lvt0_val)
686 {
687         int nmi_wd_enabled = apic_lvt_nmi_mode(apic_get_reg(apic, APIC_LVT0));
688
689         if (apic_lvt_nmi_mode(lvt0_val)) {
690                 if (!nmi_wd_enabled) {
691                         apic_debug("Receive NMI setting on APIC_LVT0 "
692                                    "for cpu %d\n", apic->vcpu->vcpu_id);
693                         apic->vcpu->kvm->arch.vapics_in_nmi_mode++;
694                 }
695         } else if (nmi_wd_enabled)
696                 apic->vcpu->kvm->arch.vapics_in_nmi_mode--;
697 }
698
699 static int apic_reg_write(struct kvm_lapic *apic, u32 reg, u32 val)
700 {
701         int ret = 0;
702
703         trace_kvm_apic_write(reg, val);
704
705         switch (reg) {
706         case APIC_ID:           /* Local APIC ID */
707                 if (!apic_x2apic_mode(apic))
708                         apic_set_reg(apic, APIC_ID, val);
709                 else
710                         ret = 1;
711                 break;
712
713         case APIC_TASKPRI:
714                 report_tpr_access(apic, true);
715                 apic_set_tpr(apic, val & 0xff);
716                 break;
717
718         case APIC_EOI:
719                 apic_set_eoi(apic);
720                 break;
721
722         case APIC_LDR:
723                 if (!apic_x2apic_mode(apic))
724                         apic_set_reg(apic, APIC_LDR, val & APIC_LDR_MASK);
725                 else
726                         ret = 1;
727                 break;
728
729         case APIC_DFR:
730                 if (!apic_x2apic_mode(apic))
731                         apic_set_reg(apic, APIC_DFR, val | 0x0FFFFFFF);
732                 else
733                         ret = 1;
734                 break;
735
736         case APIC_SPIV: {
737                 u32 mask = 0x3ff;
738                 if (apic_get_reg(apic, APIC_LVR) & APIC_LVR_DIRECTED_EOI)
739                         mask |= APIC_SPIV_DIRECTED_EOI;
740                 apic_set_reg(apic, APIC_SPIV, val & mask);
741                 if (!(val & APIC_SPIV_APIC_ENABLED)) {
742                         int i;
743                         u32 lvt_val;
744
745                         for (i = 0; i < APIC_LVT_NUM; i++) {
746                                 lvt_val = apic_get_reg(apic,
747                                                        APIC_LVTT + 0x10 * i);
748                                 apic_set_reg(apic, APIC_LVTT + 0x10 * i,
749                                              lvt_val | APIC_LVT_MASKED);
750                         }
751                         atomic_set(&apic->lapic_timer.pending, 0);
752
753                 }
754                 break;
755         }
756         case APIC_ICR:
757                 /* No delay here, so we always clear the pending bit */
758                 apic_set_reg(apic, APIC_ICR, val & ~(1 << 12));
759                 apic_send_ipi(apic);
760                 break;
761
762         case APIC_ICR2:
763                 if (!apic_x2apic_mode(apic))
764                         val &= 0xff000000;
765                 apic_set_reg(apic, APIC_ICR2, val);
766                 break;
767
768         case APIC_LVT0:
769                 apic_manage_nmi_watchdog(apic, val);
770         case APIC_LVTT:
771         case APIC_LVTTHMR:
772         case APIC_LVTPC:
773         case APIC_LVT1:
774         case APIC_LVTERR:
775                 /* TODO: Check vector */
776                 if (!apic_sw_enabled(apic))
777                         val |= APIC_LVT_MASKED;
778
779                 val &= apic_lvt_mask[(reg - APIC_LVTT) >> 4];
780                 apic_set_reg(apic, reg, val);
781
782                 break;
783
784         case APIC_TMICT:
785                 hrtimer_cancel(&apic->lapic_timer.timer);
786                 apic_set_reg(apic, APIC_TMICT, val);
787                 start_apic_timer(apic);
788                 break;
789
790         case APIC_TDCR:
791                 if (val & 4)
792                         printk(KERN_ERR "KVM_WRITE:TDCR %x\n", val);
793                 apic_set_reg(apic, APIC_TDCR, val);
794                 update_divide_count(apic);
795                 break;
796
797         case APIC_ESR:
798                 if (apic_x2apic_mode(apic) && val != 0) {
799                         printk(KERN_ERR "KVM_WRITE:ESR not zero %x\n", val);
800                         ret = 1;
801                 }
802                 break;
803
804         case APIC_SELF_IPI:
805                 if (apic_x2apic_mode(apic)) {
806                         apic_reg_write(apic, APIC_ICR, 0x40000 | (val & 0xff));
807                 } else
808                         ret = 1;
809                 break;
810         default:
811                 ret = 1;
812                 break;
813         }
814         if (ret)
815                 apic_debug("Local APIC Write to read-only register %x\n", reg);
816         return ret;
817 }
818
819 static int apic_mmio_write(struct kvm_io_device *this,
820                             gpa_t address, int len, const void *data)
821 {
822         struct kvm_lapic *apic = to_lapic(this);
823         unsigned int offset = address - apic->base_address;
824         u32 val;
825
826         if (!apic_mmio_in_range(apic, address))
827                 return -EOPNOTSUPP;
828
829         /*
830          * APIC register must be aligned on 128-bits boundary.
831          * 32/64/128 bits registers must be accessed thru 32 bits.
832          * Refer SDM 8.4.1
833          */
834         if (len != 4 || (offset & 0xf)) {
835                 /* Don't shout loud, $infamous_os would cause only noise. */
836                 apic_debug("apic write: bad size=%d %lx\n", len, (long)address);
837                 return;
838         }
839
840         val = *(u32*)data;
841
842         /* too common printing */
843         if (offset != APIC_EOI)
844                 apic_debug("%s: offset 0x%x with length 0x%x, and value is "
845                            "0x%x\n", __func__, offset, len, val);
846
847         apic_reg_write(apic, offset & 0xff0, val);
848
849         return 0;
850 }
851
852 void kvm_free_lapic(struct kvm_vcpu *vcpu)
853 {
854         if (!vcpu->arch.apic)
855                 return;
856
857         hrtimer_cancel(&vcpu->arch.apic->lapic_timer.timer);
858
859         if (vcpu->arch.apic->regs_page)
860                 __free_page(vcpu->arch.apic->regs_page);
861
862         kfree(vcpu->arch.apic);
863 }
864
865 /*
866  *----------------------------------------------------------------------
867  * LAPIC interface
868  *----------------------------------------------------------------------
869  */
870
871 void kvm_lapic_set_tpr(struct kvm_vcpu *vcpu, unsigned long cr8)
872 {
873         struct kvm_lapic *apic = vcpu->arch.apic;
874
875         if (!apic)
876                 return;
877         apic_set_tpr(apic, ((cr8 & 0x0f) << 4)
878                      | (apic_get_reg(apic, APIC_TASKPRI) & 4));
879 }
880
881 u64 kvm_lapic_get_cr8(struct kvm_vcpu *vcpu)
882 {
883         struct kvm_lapic *apic = vcpu->arch.apic;
884         u64 tpr;
885
886         if (!apic)
887                 return 0;
888         tpr = (u64) apic_get_reg(apic, APIC_TASKPRI);
889
890         return (tpr & 0xf0) >> 4;
891 }
892
893 void kvm_lapic_set_base(struct kvm_vcpu *vcpu, u64 value)
894 {
895         struct kvm_lapic *apic = vcpu->arch.apic;
896
897         if (!apic) {
898                 value |= MSR_IA32_APICBASE_BSP;
899                 vcpu->arch.apic_base = value;
900                 return;
901         }
902
903         if (!kvm_vcpu_is_bsp(apic->vcpu))
904                 value &= ~MSR_IA32_APICBASE_BSP;
905
906         vcpu->arch.apic_base = value;
907         if (apic_x2apic_mode(apic)) {
908                 u32 id = kvm_apic_id(apic);
909                 u32 ldr = ((id & ~0xf) << 16) | (1 << (id & 0xf));
910                 apic_set_reg(apic, APIC_LDR, ldr);
911         }
912         apic->base_address = apic->vcpu->arch.apic_base &
913                              MSR_IA32_APICBASE_BASE;
914
915         /* with FSB delivery interrupt, we can restart APIC functionality */
916         apic_debug("apic base msr is 0x%016" PRIx64 ", and base address is "
917                    "0x%lx.\n", apic->vcpu->arch.apic_base, apic->base_address);
918
919 }
920
921 void kvm_lapic_reset(struct kvm_vcpu *vcpu)
922 {
923         struct kvm_lapic *apic;
924         int i;
925
926         apic_debug("%s\n", __func__);
927
928         ASSERT(vcpu);
929         apic = vcpu->arch.apic;
930         ASSERT(apic != NULL);
931
932         /* Stop the timer in case it's a reset to an active apic */
933         hrtimer_cancel(&apic->lapic_timer.timer);
934
935         apic_set_reg(apic, APIC_ID, vcpu->vcpu_id << 24);
936         kvm_apic_set_version(apic->vcpu);
937
938         for (i = 0; i < APIC_LVT_NUM; i++)
939                 apic_set_reg(apic, APIC_LVTT + 0x10 * i, APIC_LVT_MASKED);
940         apic_set_reg(apic, APIC_LVT0,
941                      SET_APIC_DELIVERY_MODE(0, APIC_MODE_EXTINT));
942
943         apic_set_reg(apic, APIC_DFR, 0xffffffffU);
944         apic_set_reg(apic, APIC_SPIV, 0xff);
945         apic_set_reg(apic, APIC_TASKPRI, 0);
946         apic_set_reg(apic, APIC_LDR, 0);
947         apic_set_reg(apic, APIC_ESR, 0);
948         apic_set_reg(apic, APIC_ICR, 0);
949         apic_set_reg(apic, APIC_ICR2, 0);
950         apic_set_reg(apic, APIC_TDCR, 0);
951         apic_set_reg(apic, APIC_TMICT, 0);
952         for (i = 0; i < 8; i++) {
953                 apic_set_reg(apic, APIC_IRR + 0x10 * i, 0);
954                 apic_set_reg(apic, APIC_ISR + 0x10 * i, 0);
955                 apic_set_reg(apic, APIC_TMR + 0x10 * i, 0);
956         }
957         apic->irr_pending = false;
958         update_divide_count(apic);
959         atomic_set(&apic->lapic_timer.pending, 0);
960         if (kvm_vcpu_is_bsp(vcpu))
961                 vcpu->arch.apic_base |= MSR_IA32_APICBASE_BSP;
962         apic_update_ppr(apic);
963
964         vcpu->arch.apic_arb_prio = 0;
965
966         apic_debug(KERN_INFO "%s: vcpu=%p, id=%d, base_msr="
967                    "0x%016" PRIx64 ", base_address=0x%0lx.\n", __func__,
968                    vcpu, kvm_apic_id(apic),
969                    vcpu->arch.apic_base, apic->base_address);
970 }
971
972 bool kvm_apic_present(struct kvm_vcpu *vcpu)
973 {
974         return vcpu->arch.apic && apic_hw_enabled(vcpu->arch.apic);
975 }
976
977 int kvm_lapic_enabled(struct kvm_vcpu *vcpu)
978 {
979         return kvm_apic_present(vcpu) && apic_sw_enabled(vcpu->arch.apic);
980 }
981
982 /*
983  *----------------------------------------------------------------------
984  * timer interface
985  *----------------------------------------------------------------------
986  */
987
988 static bool lapic_is_periodic(struct kvm_timer *ktimer)
989 {
990         struct kvm_lapic *apic = container_of(ktimer, struct kvm_lapic,
991                                               lapic_timer);
992         return apic_lvtt_period(apic);
993 }
994
995 int apic_has_pending_timer(struct kvm_vcpu *vcpu)
996 {
997         struct kvm_lapic *lapic = vcpu->arch.apic;
998
999         if (lapic && apic_enabled(lapic) && apic_lvt_enabled(lapic, APIC_LVTT))
1000                 return atomic_read(&lapic->lapic_timer.pending);
1001
1002         return 0;
1003 }
1004
1005 static int kvm_apic_local_deliver(struct kvm_lapic *apic, int lvt_type)
1006 {
1007         u32 reg = apic_get_reg(apic, lvt_type);
1008         int vector, mode, trig_mode;
1009
1010         if (apic_hw_enabled(apic) && !(reg & APIC_LVT_MASKED)) {
1011                 vector = reg & APIC_VECTOR_MASK;
1012                 mode = reg & APIC_MODE_MASK;
1013                 trig_mode = reg & APIC_LVT_LEVEL_TRIGGER;
1014                 return __apic_accept_irq(apic, mode, vector, 1, trig_mode);
1015         }
1016         return 0;
1017 }
1018
1019 void kvm_apic_nmi_wd_deliver(struct kvm_vcpu *vcpu)
1020 {
1021         struct kvm_lapic *apic = vcpu->arch.apic;
1022
1023         if (apic)
1024                 kvm_apic_local_deliver(apic, APIC_LVT0);
1025 }
1026
1027 static struct kvm_timer_ops lapic_timer_ops = {
1028         .is_periodic = lapic_is_periodic,
1029 };
1030
1031 static const struct kvm_io_device_ops apic_mmio_ops = {
1032         .read     = apic_mmio_read,
1033         .write    = apic_mmio_write,
1034 };
1035
1036 int kvm_create_lapic(struct kvm_vcpu *vcpu)
1037 {
1038         struct kvm_lapic *apic;
1039
1040         ASSERT(vcpu != NULL);
1041         apic_debug("apic_init %d\n", vcpu->vcpu_id);
1042
1043         apic = kzalloc(sizeof(*apic), GFP_KERNEL);
1044         if (!apic)
1045                 goto nomem;
1046
1047         vcpu->arch.apic = apic;
1048
1049         apic->regs_page = alloc_page(GFP_KERNEL);
1050         if (apic->regs_page == NULL) {
1051                 printk(KERN_ERR "malloc apic regs error for vcpu %x\n",
1052                        vcpu->vcpu_id);
1053                 goto nomem_free_apic;
1054         }
1055         apic->regs = page_address(apic->regs_page);
1056         memset(apic->regs, 0, PAGE_SIZE);
1057         apic->vcpu = vcpu;
1058
1059         hrtimer_init(&apic->lapic_timer.timer, CLOCK_MONOTONIC,
1060                      HRTIMER_MODE_ABS);
1061         apic->lapic_timer.timer.function = kvm_timer_fn;
1062         apic->lapic_timer.t_ops = &lapic_timer_ops;
1063         apic->lapic_timer.kvm = vcpu->kvm;
1064         apic->lapic_timer.vcpu = vcpu;
1065
1066         apic->base_address = APIC_DEFAULT_PHYS_BASE;
1067         vcpu->arch.apic_base = APIC_DEFAULT_PHYS_BASE;
1068
1069         kvm_lapic_reset(vcpu);
1070         kvm_iodevice_init(&apic->dev, &apic_mmio_ops);
1071
1072         return 0;
1073 nomem_free_apic:
1074         kfree(apic);
1075 nomem:
1076         return -ENOMEM;
1077 }
1078
1079 int kvm_apic_has_interrupt(struct kvm_vcpu *vcpu)
1080 {
1081         struct kvm_lapic *apic = vcpu->arch.apic;
1082         int highest_irr;
1083
1084         if (!apic || !apic_enabled(apic))
1085                 return -1;
1086
1087         apic_update_ppr(apic);
1088         highest_irr = apic_find_highest_irr(apic);
1089         if ((highest_irr == -1) ||
1090             ((highest_irr & 0xF0) <= apic_get_reg(apic, APIC_PROCPRI)))
1091                 return -1;
1092         return highest_irr;
1093 }
1094
1095 int kvm_apic_accept_pic_intr(struct kvm_vcpu *vcpu)
1096 {
1097         u32 lvt0 = apic_get_reg(vcpu->arch.apic, APIC_LVT0);
1098         int r = 0;
1099
1100         if (kvm_vcpu_is_bsp(vcpu)) {
1101                 if (!apic_hw_enabled(vcpu->arch.apic))
1102                         r = 1;
1103                 if ((lvt0 & APIC_LVT_MASKED) == 0 &&
1104                     GET_APIC_DELIVERY_MODE(lvt0) == APIC_MODE_EXTINT)
1105                         r = 1;
1106         }
1107         return r;
1108 }
1109
1110 void kvm_inject_apic_timer_irqs(struct kvm_vcpu *vcpu)
1111 {
1112         struct kvm_lapic *apic = vcpu->arch.apic;
1113
1114         if (apic && atomic_read(&apic->lapic_timer.pending) > 0) {
1115                 if (kvm_apic_local_deliver(apic, APIC_LVTT))
1116                         atomic_dec(&apic->lapic_timer.pending);
1117         }
1118 }
1119
1120 int kvm_get_apic_interrupt(struct kvm_vcpu *vcpu)
1121 {
1122         int vector = kvm_apic_has_interrupt(vcpu);
1123         struct kvm_lapic *apic = vcpu->arch.apic;
1124
1125         if (vector == -1)
1126                 return -1;
1127
1128         apic_set_vector(vector, apic->regs + APIC_ISR);
1129         apic_update_ppr(apic);
1130         apic_clear_irr(vector, apic);
1131         return vector;
1132 }
1133
1134 void kvm_apic_post_state_restore(struct kvm_vcpu *vcpu)
1135 {
1136         struct kvm_lapic *apic = vcpu->arch.apic;
1137
1138         apic->base_address = vcpu->arch.apic_base &
1139                              MSR_IA32_APICBASE_BASE;
1140         kvm_apic_set_version(vcpu);
1141
1142         apic_update_ppr(apic);
1143         hrtimer_cancel(&apic->lapic_timer.timer);
1144         update_divide_count(apic);
1145         start_apic_timer(apic);
1146 }
1147
1148 void __kvm_migrate_apic_timer(struct kvm_vcpu *vcpu)
1149 {
1150         struct kvm_lapic *apic = vcpu->arch.apic;
1151         struct hrtimer *timer;
1152
1153         if (!apic)
1154                 return;
1155
1156         timer = &apic->lapic_timer.timer;
1157         if (hrtimer_cancel(timer))
1158                 hrtimer_start_expires(timer, HRTIMER_MODE_ABS);
1159 }
1160
1161 void kvm_lapic_sync_from_vapic(struct kvm_vcpu *vcpu)
1162 {
1163         u32 data;
1164         void *vapic;
1165
1166         if (!irqchip_in_kernel(vcpu->kvm) || !vcpu->arch.apic->vapic_addr)
1167                 return;
1168
1169         vapic = kmap_atomic(vcpu->arch.apic->vapic_page, KM_USER0);
1170         data = *(u32 *)(vapic + offset_in_page(vcpu->arch.apic->vapic_addr));
1171         kunmap_atomic(vapic, KM_USER0);
1172
1173         apic_set_tpr(vcpu->arch.apic, data & 0xff);
1174 }
1175
1176 void kvm_lapic_sync_to_vapic(struct kvm_vcpu *vcpu)
1177 {
1178         u32 data, tpr;
1179         int max_irr, max_isr;
1180         struct kvm_lapic *apic;
1181         void *vapic;
1182
1183         if (!irqchip_in_kernel(vcpu->kvm) || !vcpu->arch.apic->vapic_addr)
1184                 return;
1185
1186         apic = vcpu->arch.apic;
1187         tpr = apic_get_reg(apic, APIC_TASKPRI) & 0xff;
1188         max_irr = apic_find_highest_irr(apic);
1189         if (max_irr < 0)
1190                 max_irr = 0;
1191         max_isr = apic_find_highest_isr(apic);
1192         if (max_isr < 0)
1193                 max_isr = 0;
1194         data = (tpr & 0xff) | ((max_isr & 0xf0) << 8) | (max_irr << 24);
1195
1196         vapic = kmap_atomic(vcpu->arch.apic->vapic_page, KM_USER0);
1197         *(u32 *)(vapic + offset_in_page(vcpu->arch.apic->vapic_addr)) = data;
1198         kunmap_atomic(vapic, KM_USER0);
1199 }
1200
1201 void kvm_lapic_set_vapic_addr(struct kvm_vcpu *vcpu, gpa_t vapic_addr)
1202 {
1203         if (!irqchip_in_kernel(vcpu->kvm))
1204                 return;
1205
1206         vcpu->arch.apic->vapic_addr = vapic_addr;
1207 }
1208
1209 int kvm_x2apic_msr_write(struct kvm_vcpu *vcpu, u32 msr, u64 data)
1210 {
1211         struct kvm_lapic *apic = vcpu->arch.apic;
1212         u32 reg = (msr - APIC_BASE_MSR) << 4;
1213
1214         if (!irqchip_in_kernel(vcpu->kvm) || !apic_x2apic_mode(apic))
1215                 return 1;
1216
1217         /* if this is ICR write vector before command */
1218         if (msr == 0x830)
1219                 apic_reg_write(apic, APIC_ICR2, (u32)(data >> 32));
1220         return apic_reg_write(apic, reg, (u32)data);
1221 }
1222
1223 int kvm_x2apic_msr_read(struct kvm_vcpu *vcpu, u32 msr, u64 *data)
1224 {
1225         struct kvm_lapic *apic = vcpu->arch.apic;
1226         u32 reg = (msr - APIC_BASE_MSR) << 4, low, high = 0;
1227
1228         if (!irqchip_in_kernel(vcpu->kvm) || !apic_x2apic_mode(apic))
1229                 return 1;
1230
1231         if (apic_reg_read(apic, reg, 4, &low))
1232                 return 1;
1233         if (msr == 0x830)
1234                 apic_reg_read(apic, APIC_ICR2, 4, &high);
1235
1236         *data = (((u64)high) << 32) | low;
1237
1238         return 0;
1239 }