Merge branch 'fbdev-next' of git://github.com/schandinat/linux-2.6
[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 affiliates.
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 <linux/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, old_ppr;
263         int isr;
264
265         old_ppr = apic_get_reg(apic, APIC_PROCPRI);
266         tpr = apic_get_reg(apic, APIC_TASKPRI);
267         isr = apic_find_highest_isr(apic);
268         isrv = (isr != -1) ? isr : 0;
269
270         if ((tpr & 0xf0) >= (isrv & 0xf0))
271                 ppr = tpr & 0xff;
272         else
273                 ppr = isrv & 0xf0;
274
275         apic_debug("vlapic %p, ppr 0x%x, isr 0x%x, isrv 0x%x",
276                    apic, ppr, isr, isrv);
277
278         if (old_ppr != ppr) {
279                 apic_set_reg(apic, APIC_PROCPRI, ppr);
280                 if (ppr < old_ppr)
281                         kvm_make_request(KVM_REQ_EVENT, apic->vcpu);
282         }
283 }
284
285 static void apic_set_tpr(struct kvm_lapic *apic, u32 tpr)
286 {
287         apic_set_reg(apic, APIC_TASKPRI, tpr);
288         apic_update_ppr(apic);
289 }
290
291 int kvm_apic_match_physical_addr(struct kvm_lapic *apic, u16 dest)
292 {
293         return dest == 0xff || kvm_apic_id(apic) == dest;
294 }
295
296 int kvm_apic_match_logical_addr(struct kvm_lapic *apic, u8 mda)
297 {
298         int result = 0;
299         u32 logical_id;
300
301         if (apic_x2apic_mode(apic)) {
302                 logical_id = apic_get_reg(apic, APIC_LDR);
303                 return logical_id & mda;
304         }
305
306         logical_id = GET_APIC_LOGICAL_ID(apic_get_reg(apic, APIC_LDR));
307
308         switch (apic_get_reg(apic, APIC_DFR)) {
309         case APIC_DFR_FLAT:
310                 if (logical_id & mda)
311                         result = 1;
312                 break;
313         case APIC_DFR_CLUSTER:
314                 if (((logical_id >> 4) == (mda >> 0x4))
315                     && (logical_id & mda & 0xf))
316                         result = 1;
317                 break;
318         default:
319                 printk(KERN_WARNING "Bad DFR vcpu %d: %08x\n",
320                        apic->vcpu->vcpu_id, apic_get_reg(apic, APIC_DFR));
321                 break;
322         }
323
324         return result;
325 }
326
327 int kvm_apic_match_dest(struct kvm_vcpu *vcpu, struct kvm_lapic *source,
328                            int short_hand, int dest, int dest_mode)
329 {
330         int result = 0;
331         struct kvm_lapic *target = vcpu->arch.apic;
332
333         apic_debug("target %p, source %p, dest 0x%x, "
334                    "dest_mode 0x%x, short_hand 0x%x\n",
335                    target, source, dest, dest_mode, short_hand);
336
337         ASSERT(target);
338         switch (short_hand) {
339         case APIC_DEST_NOSHORT:
340                 if (dest_mode == 0)
341                         /* Physical mode. */
342                         result = kvm_apic_match_physical_addr(target, dest);
343                 else
344                         /* Logical mode. */
345                         result = kvm_apic_match_logical_addr(target, dest);
346                 break;
347         case APIC_DEST_SELF:
348                 result = (target == source);
349                 break;
350         case APIC_DEST_ALLINC:
351                 result = 1;
352                 break;
353         case APIC_DEST_ALLBUT:
354                 result = (target != source);
355                 break;
356         default:
357                 printk(KERN_WARNING "Bad dest shorthand value %x\n",
358                        short_hand);
359                 break;
360         }
361
362         return result;
363 }
364
365 /*
366  * Add a pending IRQ into lapic.
367  * Return 1 if successfully added and 0 if discarded.
368  */
369 static int __apic_accept_irq(struct kvm_lapic *apic, int delivery_mode,
370                              int vector, int level, int trig_mode)
371 {
372         int result = 0;
373         struct kvm_vcpu *vcpu = apic->vcpu;
374
375         switch (delivery_mode) {
376         case APIC_DM_LOWEST:
377                 vcpu->arch.apic_arb_prio++;
378         case APIC_DM_FIXED:
379                 /* FIXME add logic for vcpu on reset */
380                 if (unlikely(!apic_enabled(apic)))
381                         break;
382
383                 if (trig_mode) {
384                         apic_debug("level trig mode for vector %d", vector);
385                         apic_set_vector(vector, apic->regs + APIC_TMR);
386                 } else
387                         apic_clear_vector(vector, apic->regs + APIC_TMR);
388
389                 result = !apic_test_and_set_irr(vector, apic);
390                 trace_kvm_apic_accept_irq(vcpu->vcpu_id, delivery_mode,
391                                           trig_mode, vector, !result);
392                 if (!result) {
393                         if (trig_mode)
394                                 apic_debug("level trig mode repeatedly for "
395                                                 "vector %d", vector);
396                         break;
397                 }
398
399                 kvm_make_request(KVM_REQ_EVENT, vcpu);
400                 kvm_vcpu_kick(vcpu);
401                 break;
402
403         case APIC_DM_REMRD:
404                 printk(KERN_DEBUG "Ignoring delivery mode 3\n");
405                 break;
406
407         case APIC_DM_SMI:
408                 printk(KERN_DEBUG "Ignoring guest SMI\n");
409                 break;
410
411         case APIC_DM_NMI:
412                 result = 1;
413                 kvm_inject_nmi(vcpu);
414                 kvm_vcpu_kick(vcpu);
415                 break;
416
417         case APIC_DM_INIT:
418                 if (level) {
419                         result = 1;
420                         vcpu->arch.mp_state = KVM_MP_STATE_INIT_RECEIVED;
421                         kvm_make_request(KVM_REQ_EVENT, vcpu);
422                         kvm_vcpu_kick(vcpu);
423                 } else {
424                         apic_debug("Ignoring de-assert INIT to vcpu %d\n",
425                                    vcpu->vcpu_id);
426                 }
427                 break;
428
429         case APIC_DM_STARTUP:
430                 apic_debug("SIPI to vcpu %d vector 0x%02x\n",
431                            vcpu->vcpu_id, vector);
432                 if (vcpu->arch.mp_state == KVM_MP_STATE_INIT_RECEIVED) {
433                         result = 1;
434                         vcpu->arch.sipi_vector = vector;
435                         vcpu->arch.mp_state = KVM_MP_STATE_SIPI_RECEIVED;
436                         kvm_make_request(KVM_REQ_EVENT, vcpu);
437                         kvm_vcpu_kick(vcpu);
438                 }
439                 break;
440
441         case APIC_DM_EXTINT:
442                 /*
443                  * Should only be called by kvm_apic_local_deliver() with LVT0,
444                  * before NMI watchdog was enabled. Already handled by
445                  * kvm_apic_accept_pic_intr().
446                  */
447                 break;
448
449         default:
450                 printk(KERN_ERR "TODO: unsupported delivery mode %x\n",
451                        delivery_mode);
452                 break;
453         }
454         return result;
455 }
456
457 int kvm_apic_compare_prio(struct kvm_vcpu *vcpu1, struct kvm_vcpu *vcpu2)
458 {
459         return vcpu1->arch.apic_arb_prio - vcpu2->arch.apic_arb_prio;
460 }
461
462 static void apic_set_eoi(struct kvm_lapic *apic)
463 {
464         int vector = apic_find_highest_isr(apic);
465         int trigger_mode;
466         /*
467          * Not every write EOI will has corresponding ISR,
468          * one example is when Kernel check timer on setup_IO_APIC
469          */
470         if (vector == -1)
471                 return;
472
473         apic_clear_vector(vector, apic->regs + APIC_ISR);
474         apic_update_ppr(apic);
475
476         if (apic_test_and_clear_vector(vector, apic->regs + APIC_TMR))
477                 trigger_mode = IOAPIC_LEVEL_TRIG;
478         else
479                 trigger_mode = IOAPIC_EDGE_TRIG;
480         if (!(apic_get_reg(apic, APIC_SPIV) & APIC_SPIV_DIRECTED_EOI))
481                 kvm_ioapic_update_eoi(apic->vcpu->kvm, vector, trigger_mode);
482         kvm_make_request(KVM_REQ_EVENT, apic->vcpu);
483 }
484
485 static void apic_send_ipi(struct kvm_lapic *apic)
486 {
487         u32 icr_low = apic_get_reg(apic, APIC_ICR);
488         u32 icr_high = apic_get_reg(apic, APIC_ICR2);
489         struct kvm_lapic_irq irq;
490
491         irq.vector = icr_low & APIC_VECTOR_MASK;
492         irq.delivery_mode = icr_low & APIC_MODE_MASK;
493         irq.dest_mode = icr_low & APIC_DEST_MASK;
494         irq.level = icr_low & APIC_INT_ASSERT;
495         irq.trig_mode = icr_low & APIC_INT_LEVELTRIG;
496         irq.shorthand = icr_low & APIC_SHORT_MASK;
497         if (apic_x2apic_mode(apic))
498                 irq.dest_id = icr_high;
499         else
500                 irq.dest_id = GET_APIC_DEST_FIELD(icr_high);
501
502         trace_kvm_apic_ipi(icr_low, irq.dest_id);
503
504         apic_debug("icr_high 0x%x, icr_low 0x%x, "
505                    "short_hand 0x%x, dest 0x%x, trig_mode 0x%x, level 0x%x, "
506                    "dest_mode 0x%x, delivery_mode 0x%x, vector 0x%x\n",
507                    icr_high, icr_low, irq.shorthand, irq.dest_id,
508                    irq.trig_mode, irq.level, irq.dest_mode, irq.delivery_mode,
509                    irq.vector);
510
511         kvm_irq_delivery_to_apic(apic->vcpu->kvm, apic, &irq);
512 }
513
514 static u32 apic_get_tmcct(struct kvm_lapic *apic)
515 {
516         ktime_t remaining;
517         s64 ns;
518         u32 tmcct;
519
520         ASSERT(apic != NULL);
521
522         /* if initial count is 0, current count should also be 0 */
523         if (apic_get_reg(apic, APIC_TMICT) == 0)
524                 return 0;
525
526         remaining = hrtimer_get_remaining(&apic->lapic_timer.timer);
527         if (ktime_to_ns(remaining) < 0)
528                 remaining = ktime_set(0, 0);
529
530         ns = mod_64(ktime_to_ns(remaining), apic->lapic_timer.period);
531         tmcct = div64_u64(ns,
532                          (APIC_BUS_CYCLE_NS * apic->divide_count));
533
534         return tmcct;
535 }
536
537 static void __report_tpr_access(struct kvm_lapic *apic, bool write)
538 {
539         struct kvm_vcpu *vcpu = apic->vcpu;
540         struct kvm_run *run = vcpu->run;
541
542         kvm_make_request(KVM_REQ_REPORT_TPR_ACCESS, vcpu);
543         run->tpr_access.rip = kvm_rip_read(vcpu);
544         run->tpr_access.is_write = write;
545 }
546
547 static inline void report_tpr_access(struct kvm_lapic *apic, bool write)
548 {
549         if (apic->vcpu->arch.tpr_access_reporting)
550                 __report_tpr_access(apic, write);
551 }
552
553 static u32 __apic_read(struct kvm_lapic *apic, unsigned int offset)
554 {
555         u32 val = 0;
556
557         if (offset >= LAPIC_MMIO_LENGTH)
558                 return 0;
559
560         switch (offset) {
561         case APIC_ID:
562                 if (apic_x2apic_mode(apic))
563                         val = kvm_apic_id(apic);
564                 else
565                         val = kvm_apic_id(apic) << 24;
566                 break;
567         case APIC_ARBPRI:
568                 printk(KERN_WARNING "Access APIC ARBPRI register "
569                        "which is for P6\n");
570                 break;
571
572         case APIC_TMCCT:        /* Timer CCR */
573                 val = apic_get_tmcct(apic);
574                 break;
575
576         case APIC_TASKPRI:
577                 report_tpr_access(apic, false);
578                 /* fall thru */
579         default:
580                 apic_update_ppr(apic);
581                 val = apic_get_reg(apic, offset);
582                 break;
583         }
584
585         return val;
586 }
587
588 static inline struct kvm_lapic *to_lapic(struct kvm_io_device *dev)
589 {
590         return container_of(dev, struct kvm_lapic, dev);
591 }
592
593 static int apic_reg_read(struct kvm_lapic *apic, u32 offset, int len,
594                 void *data)
595 {
596         unsigned char alignment = offset & 0xf;
597         u32 result;
598         /* this bitmask has a bit cleared for each reserver register */
599         static const u64 rmask = 0x43ff01ffffffe70cULL;
600
601         if ((alignment + len) > 4) {
602                 apic_debug("KVM_APIC_READ: alignment error %x %d\n",
603                            offset, len);
604                 return 1;
605         }
606
607         if (offset > 0x3f0 || !(rmask & (1ULL << (offset >> 4)))) {
608                 apic_debug("KVM_APIC_READ: read reserved register %x\n",
609                            offset);
610                 return 1;
611         }
612
613         result = __apic_read(apic, offset & ~0xf);
614
615         trace_kvm_apic_read(offset, result);
616
617         switch (len) {
618         case 1:
619         case 2:
620         case 4:
621                 memcpy(data, (char *)&result + alignment, len);
622                 break;
623         default:
624                 printk(KERN_ERR "Local APIC read with len = %x, "
625                        "should be 1,2, or 4 instead\n", len);
626                 break;
627         }
628         return 0;
629 }
630
631 static int apic_mmio_in_range(struct kvm_lapic *apic, gpa_t addr)
632 {
633         return apic_hw_enabled(apic) &&
634             addr >= apic->base_address &&
635             addr < apic->base_address + LAPIC_MMIO_LENGTH;
636 }
637
638 static int apic_mmio_read(struct kvm_io_device *this,
639                            gpa_t address, int len, void *data)
640 {
641         struct kvm_lapic *apic = to_lapic(this);
642         u32 offset = address - apic->base_address;
643
644         if (!apic_mmio_in_range(apic, address))
645                 return -EOPNOTSUPP;
646
647         apic_reg_read(apic, offset, len, data);
648
649         return 0;
650 }
651
652 static void update_divide_count(struct kvm_lapic *apic)
653 {
654         u32 tmp1, tmp2, tdcr;
655
656         tdcr = apic_get_reg(apic, APIC_TDCR);
657         tmp1 = tdcr & 0xf;
658         tmp2 = ((tmp1 & 0x3) | ((tmp1 & 0x8) >> 1)) + 1;
659         apic->divide_count = 0x1 << (tmp2 & 0x7);
660
661         apic_debug("timer divide count is 0x%x\n",
662                                    apic->divide_count);
663 }
664
665 static void start_apic_timer(struct kvm_lapic *apic)
666 {
667         ktime_t now = apic->lapic_timer.timer.base->get_time();
668
669         apic->lapic_timer.period = (u64)apic_get_reg(apic, APIC_TMICT) *
670                     APIC_BUS_CYCLE_NS * apic->divide_count;
671         atomic_set(&apic->lapic_timer.pending, 0);
672
673         if (!apic->lapic_timer.period)
674                 return;
675         /*
676          * Do not allow the guest to program periodic timers with small
677          * interval, since the hrtimers are not throttled by the host
678          * scheduler.
679          */
680         if (apic_lvtt_period(apic)) {
681                 if (apic->lapic_timer.period < NSEC_PER_MSEC/2)
682                         apic->lapic_timer.period = NSEC_PER_MSEC/2;
683         }
684
685         hrtimer_start(&apic->lapic_timer.timer,
686                       ktime_add_ns(now, apic->lapic_timer.period),
687                       HRTIMER_MODE_ABS);
688
689         apic_debug("%s: bus cycle is %" PRId64 "ns, now 0x%016"
690                            PRIx64 ", "
691                            "timer initial count 0x%x, period %lldns, "
692                            "expire @ 0x%016" PRIx64 ".\n", __func__,
693                            APIC_BUS_CYCLE_NS, ktime_to_ns(now),
694                            apic_get_reg(apic, APIC_TMICT),
695                            apic->lapic_timer.period,
696                            ktime_to_ns(ktime_add_ns(now,
697                                         apic->lapic_timer.period)));
698 }
699
700 static void apic_manage_nmi_watchdog(struct kvm_lapic *apic, u32 lvt0_val)
701 {
702         int nmi_wd_enabled = apic_lvt_nmi_mode(apic_get_reg(apic, APIC_LVT0));
703
704         if (apic_lvt_nmi_mode(lvt0_val)) {
705                 if (!nmi_wd_enabled) {
706                         apic_debug("Receive NMI setting on APIC_LVT0 "
707                                    "for cpu %d\n", apic->vcpu->vcpu_id);
708                         apic->vcpu->kvm->arch.vapics_in_nmi_mode++;
709                 }
710         } else if (nmi_wd_enabled)
711                 apic->vcpu->kvm->arch.vapics_in_nmi_mode--;
712 }
713
714 static int apic_reg_write(struct kvm_lapic *apic, u32 reg, u32 val)
715 {
716         int ret = 0;
717
718         trace_kvm_apic_write(reg, val);
719
720         switch (reg) {
721         case APIC_ID:           /* Local APIC ID */
722                 if (!apic_x2apic_mode(apic))
723                         apic_set_reg(apic, APIC_ID, val);
724                 else
725                         ret = 1;
726                 break;
727
728         case APIC_TASKPRI:
729                 report_tpr_access(apic, true);
730                 apic_set_tpr(apic, val & 0xff);
731                 break;
732
733         case APIC_EOI:
734                 apic_set_eoi(apic);
735                 break;
736
737         case APIC_LDR:
738                 if (!apic_x2apic_mode(apic))
739                         apic_set_reg(apic, APIC_LDR, val & APIC_LDR_MASK);
740                 else
741                         ret = 1;
742                 break;
743
744         case APIC_DFR:
745                 if (!apic_x2apic_mode(apic))
746                         apic_set_reg(apic, APIC_DFR, val | 0x0FFFFFFF);
747                 else
748                         ret = 1;
749                 break;
750
751         case APIC_SPIV: {
752                 u32 mask = 0x3ff;
753                 if (apic_get_reg(apic, APIC_LVR) & APIC_LVR_DIRECTED_EOI)
754                         mask |= APIC_SPIV_DIRECTED_EOI;
755                 apic_set_reg(apic, APIC_SPIV, val & mask);
756                 if (!(val & APIC_SPIV_APIC_ENABLED)) {
757                         int i;
758                         u32 lvt_val;
759
760                         for (i = 0; i < APIC_LVT_NUM; i++) {
761                                 lvt_val = apic_get_reg(apic,
762                                                        APIC_LVTT + 0x10 * i);
763                                 apic_set_reg(apic, APIC_LVTT + 0x10 * i,
764                                              lvt_val | APIC_LVT_MASKED);
765                         }
766                         atomic_set(&apic->lapic_timer.pending, 0);
767
768                 }
769                 break;
770         }
771         case APIC_ICR:
772                 /* No delay here, so we always clear the pending bit */
773                 apic_set_reg(apic, APIC_ICR, val & ~(1 << 12));
774                 apic_send_ipi(apic);
775                 break;
776
777         case APIC_ICR2:
778                 if (!apic_x2apic_mode(apic))
779                         val &= 0xff000000;
780                 apic_set_reg(apic, APIC_ICR2, val);
781                 break;
782
783         case APIC_LVT0:
784                 apic_manage_nmi_watchdog(apic, val);
785         case APIC_LVTT:
786         case APIC_LVTTHMR:
787         case APIC_LVTPC:
788         case APIC_LVT1:
789         case APIC_LVTERR:
790                 /* TODO: Check vector */
791                 if (!apic_sw_enabled(apic))
792                         val |= APIC_LVT_MASKED;
793
794                 val &= apic_lvt_mask[(reg - APIC_LVTT) >> 4];
795                 apic_set_reg(apic, reg, val);
796
797                 break;
798
799         case APIC_TMICT:
800                 hrtimer_cancel(&apic->lapic_timer.timer);
801                 apic_set_reg(apic, APIC_TMICT, val);
802                 start_apic_timer(apic);
803                 break;
804
805         case APIC_TDCR:
806                 if (val & 4)
807                         printk(KERN_ERR "KVM_WRITE:TDCR %x\n", val);
808                 apic_set_reg(apic, APIC_TDCR, val);
809                 update_divide_count(apic);
810                 break;
811
812         case APIC_ESR:
813                 if (apic_x2apic_mode(apic) && val != 0) {
814                         printk(KERN_ERR "KVM_WRITE:ESR not zero %x\n", val);
815                         ret = 1;
816                 }
817                 break;
818
819         case APIC_SELF_IPI:
820                 if (apic_x2apic_mode(apic)) {
821                         apic_reg_write(apic, APIC_ICR, 0x40000 | (val & 0xff));
822                 } else
823                         ret = 1;
824                 break;
825         default:
826                 ret = 1;
827                 break;
828         }
829         if (ret)
830                 apic_debug("Local APIC Write to read-only register %x\n", reg);
831         return ret;
832 }
833
834 static int apic_mmio_write(struct kvm_io_device *this,
835                             gpa_t address, int len, const void *data)
836 {
837         struct kvm_lapic *apic = to_lapic(this);
838         unsigned int offset = address - apic->base_address;
839         u32 val;
840
841         if (!apic_mmio_in_range(apic, address))
842                 return -EOPNOTSUPP;
843
844         /*
845          * APIC register must be aligned on 128-bits boundary.
846          * 32/64/128 bits registers must be accessed thru 32 bits.
847          * Refer SDM 8.4.1
848          */
849         if (len != 4 || (offset & 0xf)) {
850                 /* Don't shout loud, $infamous_os would cause only noise. */
851                 apic_debug("apic write: bad size=%d %lx\n", len, (long)address);
852                 return 0;
853         }
854
855         val = *(u32*)data;
856
857         /* too common printing */
858         if (offset != APIC_EOI)
859                 apic_debug("%s: offset 0x%x with length 0x%x, and value is "
860                            "0x%x\n", __func__, offset, len, val);
861
862         apic_reg_write(apic, offset & 0xff0, val);
863
864         return 0;
865 }
866
867 void kvm_free_lapic(struct kvm_vcpu *vcpu)
868 {
869         if (!vcpu->arch.apic)
870                 return;
871
872         hrtimer_cancel(&vcpu->arch.apic->lapic_timer.timer);
873
874         if (vcpu->arch.apic->regs)
875                 free_page((unsigned long)vcpu->arch.apic->regs);
876
877         kfree(vcpu->arch.apic);
878 }
879
880 /*
881  *----------------------------------------------------------------------
882  * LAPIC interface
883  *----------------------------------------------------------------------
884  */
885
886 void kvm_lapic_set_tpr(struct kvm_vcpu *vcpu, unsigned long cr8)
887 {
888         struct kvm_lapic *apic = vcpu->arch.apic;
889
890         if (!apic)
891                 return;
892         apic_set_tpr(apic, ((cr8 & 0x0f) << 4)
893                      | (apic_get_reg(apic, APIC_TASKPRI) & 4));
894 }
895
896 u64 kvm_lapic_get_cr8(struct kvm_vcpu *vcpu)
897 {
898         struct kvm_lapic *apic = vcpu->arch.apic;
899         u64 tpr;
900
901         if (!apic)
902                 return 0;
903         tpr = (u64) apic_get_reg(apic, APIC_TASKPRI);
904
905         return (tpr & 0xf0) >> 4;
906 }
907
908 void kvm_lapic_set_base(struct kvm_vcpu *vcpu, u64 value)
909 {
910         struct kvm_lapic *apic = vcpu->arch.apic;
911
912         if (!apic) {
913                 value |= MSR_IA32_APICBASE_BSP;
914                 vcpu->arch.apic_base = value;
915                 return;
916         }
917
918         if (!kvm_vcpu_is_bsp(apic->vcpu))
919                 value &= ~MSR_IA32_APICBASE_BSP;
920
921         vcpu->arch.apic_base = value;
922         if (apic_x2apic_mode(apic)) {
923                 u32 id = kvm_apic_id(apic);
924                 u32 ldr = ((id & ~0xf) << 16) | (1 << (id & 0xf));
925                 apic_set_reg(apic, APIC_LDR, ldr);
926         }
927         apic->base_address = apic->vcpu->arch.apic_base &
928                              MSR_IA32_APICBASE_BASE;
929
930         /* with FSB delivery interrupt, we can restart APIC functionality */
931         apic_debug("apic base msr is 0x%016" PRIx64 ", and base address is "
932                    "0x%lx.\n", apic->vcpu->arch.apic_base, apic->base_address);
933
934 }
935
936 void kvm_lapic_reset(struct kvm_vcpu *vcpu)
937 {
938         struct kvm_lapic *apic;
939         int i;
940
941         apic_debug("%s\n", __func__);
942
943         ASSERT(vcpu);
944         apic = vcpu->arch.apic;
945         ASSERT(apic != NULL);
946
947         /* Stop the timer in case it's a reset to an active apic */
948         hrtimer_cancel(&apic->lapic_timer.timer);
949
950         apic_set_reg(apic, APIC_ID, vcpu->vcpu_id << 24);
951         kvm_apic_set_version(apic->vcpu);
952
953         for (i = 0; i < APIC_LVT_NUM; i++)
954                 apic_set_reg(apic, APIC_LVTT + 0x10 * i, APIC_LVT_MASKED);
955         apic_set_reg(apic, APIC_LVT0,
956                      SET_APIC_DELIVERY_MODE(0, APIC_MODE_EXTINT));
957
958         apic_set_reg(apic, APIC_DFR, 0xffffffffU);
959         apic_set_reg(apic, APIC_SPIV, 0xff);
960         apic_set_reg(apic, APIC_TASKPRI, 0);
961         apic_set_reg(apic, APIC_LDR, 0);
962         apic_set_reg(apic, APIC_ESR, 0);
963         apic_set_reg(apic, APIC_ICR, 0);
964         apic_set_reg(apic, APIC_ICR2, 0);
965         apic_set_reg(apic, APIC_TDCR, 0);
966         apic_set_reg(apic, APIC_TMICT, 0);
967         for (i = 0; i < 8; i++) {
968                 apic_set_reg(apic, APIC_IRR + 0x10 * i, 0);
969                 apic_set_reg(apic, APIC_ISR + 0x10 * i, 0);
970                 apic_set_reg(apic, APIC_TMR + 0x10 * i, 0);
971         }
972         apic->irr_pending = false;
973         update_divide_count(apic);
974         atomic_set(&apic->lapic_timer.pending, 0);
975         if (kvm_vcpu_is_bsp(vcpu))
976                 vcpu->arch.apic_base |= MSR_IA32_APICBASE_BSP;
977         apic_update_ppr(apic);
978
979         vcpu->arch.apic_arb_prio = 0;
980
981         apic_debug(KERN_INFO "%s: vcpu=%p, id=%d, base_msr="
982                    "0x%016" PRIx64 ", base_address=0x%0lx.\n", __func__,
983                    vcpu, kvm_apic_id(apic),
984                    vcpu->arch.apic_base, apic->base_address);
985 }
986
987 bool kvm_apic_present(struct kvm_vcpu *vcpu)
988 {
989         return vcpu->arch.apic && apic_hw_enabled(vcpu->arch.apic);
990 }
991
992 int kvm_lapic_enabled(struct kvm_vcpu *vcpu)
993 {
994         return kvm_apic_present(vcpu) && apic_sw_enabled(vcpu->arch.apic);
995 }
996
997 /*
998  *----------------------------------------------------------------------
999  * timer interface
1000  *----------------------------------------------------------------------
1001  */
1002
1003 static bool lapic_is_periodic(struct kvm_timer *ktimer)
1004 {
1005         struct kvm_lapic *apic = container_of(ktimer, struct kvm_lapic,
1006                                               lapic_timer);
1007         return apic_lvtt_period(apic);
1008 }
1009
1010 int apic_has_pending_timer(struct kvm_vcpu *vcpu)
1011 {
1012         struct kvm_lapic *lapic = vcpu->arch.apic;
1013
1014         if (lapic && apic_enabled(lapic) && apic_lvt_enabled(lapic, APIC_LVTT))
1015                 return atomic_read(&lapic->lapic_timer.pending);
1016
1017         return 0;
1018 }
1019
1020 static int kvm_apic_local_deliver(struct kvm_lapic *apic, int lvt_type)
1021 {
1022         u32 reg = apic_get_reg(apic, lvt_type);
1023         int vector, mode, trig_mode;
1024
1025         if (apic_hw_enabled(apic) && !(reg & APIC_LVT_MASKED)) {
1026                 vector = reg & APIC_VECTOR_MASK;
1027                 mode = reg & APIC_MODE_MASK;
1028                 trig_mode = reg & APIC_LVT_LEVEL_TRIGGER;
1029                 return __apic_accept_irq(apic, mode, vector, 1, trig_mode);
1030         }
1031         return 0;
1032 }
1033
1034 void kvm_apic_nmi_wd_deliver(struct kvm_vcpu *vcpu)
1035 {
1036         struct kvm_lapic *apic = vcpu->arch.apic;
1037
1038         if (apic)
1039                 kvm_apic_local_deliver(apic, APIC_LVT0);
1040 }
1041
1042 static struct kvm_timer_ops lapic_timer_ops = {
1043         .is_periodic = lapic_is_periodic,
1044 };
1045
1046 static const struct kvm_io_device_ops apic_mmio_ops = {
1047         .read     = apic_mmio_read,
1048         .write    = apic_mmio_write,
1049 };
1050
1051 int kvm_create_lapic(struct kvm_vcpu *vcpu)
1052 {
1053         struct kvm_lapic *apic;
1054
1055         ASSERT(vcpu != NULL);
1056         apic_debug("apic_init %d\n", vcpu->vcpu_id);
1057
1058         apic = kzalloc(sizeof(*apic), GFP_KERNEL);
1059         if (!apic)
1060                 goto nomem;
1061
1062         vcpu->arch.apic = apic;
1063
1064         apic->regs = (void *)get_zeroed_page(GFP_KERNEL);
1065         if (!apic->regs) {
1066                 printk(KERN_ERR "malloc apic regs error for vcpu %x\n",
1067                        vcpu->vcpu_id);
1068                 goto nomem_free_apic;
1069         }
1070         apic->vcpu = vcpu;
1071
1072         hrtimer_init(&apic->lapic_timer.timer, CLOCK_MONOTONIC,
1073                      HRTIMER_MODE_ABS);
1074         apic->lapic_timer.timer.function = kvm_timer_fn;
1075         apic->lapic_timer.t_ops = &lapic_timer_ops;
1076         apic->lapic_timer.kvm = vcpu->kvm;
1077         apic->lapic_timer.vcpu = vcpu;
1078
1079         apic->base_address = APIC_DEFAULT_PHYS_BASE;
1080         vcpu->arch.apic_base = APIC_DEFAULT_PHYS_BASE;
1081
1082         kvm_lapic_reset(vcpu);
1083         kvm_iodevice_init(&apic->dev, &apic_mmio_ops);
1084
1085         return 0;
1086 nomem_free_apic:
1087         kfree(apic);
1088 nomem:
1089         return -ENOMEM;
1090 }
1091
1092 int kvm_apic_has_interrupt(struct kvm_vcpu *vcpu)
1093 {
1094         struct kvm_lapic *apic = vcpu->arch.apic;
1095         int highest_irr;
1096
1097         if (!apic || !apic_enabled(apic))
1098                 return -1;
1099
1100         apic_update_ppr(apic);
1101         highest_irr = apic_find_highest_irr(apic);
1102         if ((highest_irr == -1) ||
1103             ((highest_irr & 0xF0) <= apic_get_reg(apic, APIC_PROCPRI)))
1104                 return -1;
1105         return highest_irr;
1106 }
1107
1108 int kvm_apic_accept_pic_intr(struct kvm_vcpu *vcpu)
1109 {
1110         u32 lvt0 = apic_get_reg(vcpu->arch.apic, APIC_LVT0);
1111         int r = 0;
1112
1113         if (!apic_hw_enabled(vcpu->arch.apic))
1114                 r = 1;
1115         if ((lvt0 & APIC_LVT_MASKED) == 0 &&
1116             GET_APIC_DELIVERY_MODE(lvt0) == APIC_MODE_EXTINT)
1117                 r = 1;
1118         return r;
1119 }
1120
1121 void kvm_inject_apic_timer_irqs(struct kvm_vcpu *vcpu)
1122 {
1123         struct kvm_lapic *apic = vcpu->arch.apic;
1124
1125         if (apic && atomic_read(&apic->lapic_timer.pending) > 0) {
1126                 if (kvm_apic_local_deliver(apic, APIC_LVTT))
1127                         atomic_dec(&apic->lapic_timer.pending);
1128         }
1129 }
1130
1131 int kvm_get_apic_interrupt(struct kvm_vcpu *vcpu)
1132 {
1133         int vector = kvm_apic_has_interrupt(vcpu);
1134         struct kvm_lapic *apic = vcpu->arch.apic;
1135
1136         if (vector == -1)
1137                 return -1;
1138
1139         apic_set_vector(vector, apic->regs + APIC_ISR);
1140         apic_update_ppr(apic);
1141         apic_clear_irr(vector, apic);
1142         return vector;
1143 }
1144
1145 void kvm_apic_post_state_restore(struct kvm_vcpu *vcpu)
1146 {
1147         struct kvm_lapic *apic = vcpu->arch.apic;
1148
1149         apic->base_address = vcpu->arch.apic_base &
1150                              MSR_IA32_APICBASE_BASE;
1151         kvm_apic_set_version(vcpu);
1152
1153         apic_update_ppr(apic);
1154         hrtimer_cancel(&apic->lapic_timer.timer);
1155         update_divide_count(apic);
1156         start_apic_timer(apic);
1157         apic->irr_pending = true;
1158         kvm_make_request(KVM_REQ_EVENT, vcpu);
1159 }
1160
1161 void __kvm_migrate_apic_timer(struct kvm_vcpu *vcpu)
1162 {
1163         struct kvm_lapic *apic = vcpu->arch.apic;
1164         struct hrtimer *timer;
1165
1166         if (!apic)
1167                 return;
1168
1169         timer = &apic->lapic_timer.timer;
1170         if (hrtimer_cancel(timer))
1171                 hrtimer_start_expires(timer, HRTIMER_MODE_ABS);
1172 }
1173
1174 void kvm_lapic_sync_from_vapic(struct kvm_vcpu *vcpu)
1175 {
1176         u32 data;
1177         void *vapic;
1178
1179         if (!irqchip_in_kernel(vcpu->kvm) || !vcpu->arch.apic->vapic_addr)
1180                 return;
1181
1182         vapic = kmap_atomic(vcpu->arch.apic->vapic_page, KM_USER0);
1183         data = *(u32 *)(vapic + offset_in_page(vcpu->arch.apic->vapic_addr));
1184         kunmap_atomic(vapic, KM_USER0);
1185
1186         apic_set_tpr(vcpu->arch.apic, data & 0xff);
1187 }
1188
1189 void kvm_lapic_sync_to_vapic(struct kvm_vcpu *vcpu)
1190 {
1191         u32 data, tpr;
1192         int max_irr, max_isr;
1193         struct kvm_lapic *apic;
1194         void *vapic;
1195
1196         if (!irqchip_in_kernel(vcpu->kvm) || !vcpu->arch.apic->vapic_addr)
1197                 return;
1198
1199         apic = vcpu->arch.apic;
1200         tpr = apic_get_reg(apic, APIC_TASKPRI) & 0xff;
1201         max_irr = apic_find_highest_irr(apic);
1202         if (max_irr < 0)
1203                 max_irr = 0;
1204         max_isr = apic_find_highest_isr(apic);
1205         if (max_isr < 0)
1206                 max_isr = 0;
1207         data = (tpr & 0xff) | ((max_isr & 0xf0) << 8) | (max_irr << 24);
1208
1209         vapic = kmap_atomic(vcpu->arch.apic->vapic_page, KM_USER0);
1210         *(u32 *)(vapic + offset_in_page(vcpu->arch.apic->vapic_addr)) = data;
1211         kunmap_atomic(vapic, KM_USER0);
1212 }
1213
1214 void kvm_lapic_set_vapic_addr(struct kvm_vcpu *vcpu, gpa_t vapic_addr)
1215 {
1216         if (!irqchip_in_kernel(vcpu->kvm))
1217                 return;
1218
1219         vcpu->arch.apic->vapic_addr = vapic_addr;
1220 }
1221
1222 int kvm_x2apic_msr_write(struct kvm_vcpu *vcpu, u32 msr, u64 data)
1223 {
1224         struct kvm_lapic *apic = vcpu->arch.apic;
1225         u32 reg = (msr - APIC_BASE_MSR) << 4;
1226
1227         if (!irqchip_in_kernel(vcpu->kvm) || !apic_x2apic_mode(apic))
1228                 return 1;
1229
1230         /* if this is ICR write vector before command */
1231         if (msr == 0x830)
1232                 apic_reg_write(apic, APIC_ICR2, (u32)(data >> 32));
1233         return apic_reg_write(apic, reg, (u32)data);
1234 }
1235
1236 int kvm_x2apic_msr_read(struct kvm_vcpu *vcpu, u32 msr, u64 *data)
1237 {
1238         struct kvm_lapic *apic = vcpu->arch.apic;
1239         u32 reg = (msr - APIC_BASE_MSR) << 4, low, high = 0;
1240
1241         if (!irqchip_in_kernel(vcpu->kvm) || !apic_x2apic_mode(apic))
1242                 return 1;
1243
1244         if (apic_reg_read(apic, reg, 4, &low))
1245                 return 1;
1246         if (msr == 0x830)
1247                 apic_reg_read(apic, APIC_ICR2, 4, &high);
1248
1249         *data = (((u64)high) << 32) | low;
1250
1251         return 0;
1252 }
1253
1254 int kvm_hv_vapic_msr_write(struct kvm_vcpu *vcpu, u32 reg, u64 data)
1255 {
1256         struct kvm_lapic *apic = vcpu->arch.apic;
1257
1258         if (!irqchip_in_kernel(vcpu->kvm))
1259                 return 1;
1260
1261         /* if this is ICR write vector before command */
1262         if (reg == APIC_ICR)
1263                 apic_reg_write(apic, APIC_ICR2, (u32)(data >> 32));
1264         return apic_reg_write(apic, reg, (u32)data);
1265 }
1266
1267 int kvm_hv_vapic_msr_read(struct kvm_vcpu *vcpu, u32 reg, u64 *data)
1268 {
1269         struct kvm_lapic *apic = vcpu->arch.apic;
1270         u32 low, high = 0;
1271
1272         if (!irqchip_in_kernel(vcpu->kvm))
1273                 return 1;
1274
1275         if (apic_reg_read(apic, reg, 4, &low))
1276                 return 1;
1277         if (reg == APIC_ICR)
1278                 apic_reg_read(apic, APIC_ICR2, 4, &high);
1279
1280         *data = (((u64)high) << 32) | low;
1281
1282         return 0;
1283 }