spi-topcliff-pch: add recovery processing in case wait-event timeout
[zen-stable.git] / arch / x86 / kvm / lapic.c
blobcfdc6e0ef0025f4687c1a7b9d3c6dff42f0c8875
2 /*
3 * Local APIC virtualization
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.
10 * Authors:
11 * Dor Laor <dor.laor@qumranet.com>
12 * Gregory Haskins <ghaskins@novell.com>
13 * Yaozu (Eddie) Dong <eddie.dong@intel.com>
15 * Based on Xen 3.1 code, Copyright (c) 2004, Intel Corporation.
17 * This work is licensed under the terms of the GNU GPL, version 2. See
18 * the COPYING file in the top-level directory.
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 #include "cpuid.h"
43 #ifndef CONFIG_X86_64
44 #define mod_64(x, y) ((x) - (y) * div64_u64(x, y))
45 #else
46 #define mod_64(x, y) ((x) % (y))
47 #endif
49 #define PRId64 "d"
50 #define PRIx64 "llx"
51 #define PRIu64 "u"
52 #define PRIo64 "o"
54 #define APIC_BUS_CYCLE_NS 1
56 /* #define apic_debug(fmt,arg...) printk(KERN_WARNING fmt,##arg) */
57 #define apic_debug(fmt, arg...)
59 #define APIC_LVT_NUM 6
60 /* 14 is the version for Xeon and Pentium 8.4.8*/
61 #define APIC_VERSION (0x14UL | ((APIC_LVT_NUM - 1) << 16))
62 #define LAPIC_MMIO_LENGTH (1 << 12)
63 /* followed define is not in apicdef.h */
64 #define APIC_SHORT_MASK 0xc0000
65 #define APIC_DEST_NOSHORT 0x0
66 #define APIC_DEST_MASK 0x800
67 #define MAX_APIC_VECTOR 256
69 #define VEC_POS(v) ((v) & (32 - 1))
70 #define REG_POS(v) (((v) >> 5) << 4)
72 static unsigned int min_timer_period_us = 500;
73 module_param(min_timer_period_us, uint, S_IRUGO | S_IWUSR);
75 static inline u32 apic_get_reg(struct kvm_lapic *apic, int reg_off)
77 return *((u32 *) (apic->regs + reg_off));
80 static inline void apic_set_reg(struct kvm_lapic *apic, int reg_off, u32 val)
82 *((u32 *) (apic->regs + reg_off)) = val;
85 static inline int apic_test_and_set_vector(int vec, void *bitmap)
87 return test_and_set_bit(VEC_POS(vec), (bitmap) + REG_POS(vec));
90 static inline int apic_test_and_clear_vector(int vec, void *bitmap)
92 return test_and_clear_bit(VEC_POS(vec), (bitmap) + REG_POS(vec));
95 static inline void apic_set_vector(int vec, void *bitmap)
97 set_bit(VEC_POS(vec), (bitmap) + REG_POS(vec));
100 static inline void apic_clear_vector(int vec, void *bitmap)
102 clear_bit(VEC_POS(vec), (bitmap) + REG_POS(vec));
105 static inline int apic_hw_enabled(struct kvm_lapic *apic)
107 return (apic)->vcpu->arch.apic_base & MSR_IA32_APICBASE_ENABLE;
110 static inline int apic_sw_enabled(struct kvm_lapic *apic)
112 return apic_get_reg(apic, APIC_SPIV) & APIC_SPIV_APIC_ENABLED;
115 static inline int apic_enabled(struct kvm_lapic *apic)
117 return apic_sw_enabled(apic) && apic_hw_enabled(apic);
120 #define LVT_MASK \
121 (APIC_LVT_MASKED | APIC_SEND_PENDING | APIC_VECTOR_MASK)
123 #define LINT_MASK \
124 (LVT_MASK | APIC_MODE_MASK | APIC_INPUT_POLARITY | \
125 APIC_LVT_REMOTE_IRR | APIC_LVT_LEVEL_TRIGGER)
127 static inline int kvm_apic_id(struct kvm_lapic *apic)
129 return (apic_get_reg(apic, APIC_ID) >> 24) & 0xff;
132 static inline int apic_lvt_enabled(struct kvm_lapic *apic, int lvt_type)
134 return !(apic_get_reg(apic, lvt_type) & APIC_LVT_MASKED);
137 static inline int apic_lvt_vector(struct kvm_lapic *apic, int lvt_type)
139 return apic_get_reg(apic, lvt_type) & APIC_VECTOR_MASK;
142 static inline int apic_lvtt_oneshot(struct kvm_lapic *apic)
144 return ((apic_get_reg(apic, APIC_LVTT) &
145 apic->lapic_timer.timer_mode_mask) == APIC_LVT_TIMER_ONESHOT);
148 static inline int apic_lvtt_period(struct kvm_lapic *apic)
150 return ((apic_get_reg(apic, APIC_LVTT) &
151 apic->lapic_timer.timer_mode_mask) == APIC_LVT_TIMER_PERIODIC);
154 static inline int apic_lvtt_tscdeadline(struct kvm_lapic *apic)
156 return ((apic_get_reg(apic, APIC_LVTT) &
157 apic->lapic_timer.timer_mode_mask) ==
158 APIC_LVT_TIMER_TSCDEADLINE);
161 static inline int apic_lvt_nmi_mode(u32 lvt_val)
163 return (lvt_val & (APIC_MODE_MASK | APIC_LVT_MASKED)) == APIC_DM_NMI;
166 void kvm_apic_set_version(struct kvm_vcpu *vcpu)
168 struct kvm_lapic *apic = vcpu->arch.apic;
169 struct kvm_cpuid_entry2 *feat;
170 u32 v = APIC_VERSION;
172 if (!irqchip_in_kernel(vcpu->kvm))
173 return;
175 feat = kvm_find_cpuid_entry(apic->vcpu, 0x1, 0);
176 if (feat && (feat->ecx & (1 << (X86_FEATURE_X2APIC & 31))))
177 v |= APIC_LVR_DIRECTED_EOI;
178 apic_set_reg(apic, APIC_LVR, v);
181 static inline int apic_x2apic_mode(struct kvm_lapic *apic)
183 return apic->vcpu->arch.apic_base & X2APIC_ENABLE;
186 static unsigned int apic_lvt_mask[APIC_LVT_NUM] = {
187 LVT_MASK , /* part LVTT mask, timer mode mask added at runtime */
188 LVT_MASK | APIC_MODE_MASK, /* LVTTHMR */
189 LVT_MASK | APIC_MODE_MASK, /* LVTPC */
190 LINT_MASK, LINT_MASK, /* LVT0-1 */
191 LVT_MASK /* LVTERR */
194 static int find_highest_vector(void *bitmap)
196 u32 *word = bitmap;
197 int word_offset = MAX_APIC_VECTOR >> 5;
199 while ((word_offset != 0) && (word[(--word_offset) << 2] == 0))
200 continue;
202 if (likely(!word_offset && !word[0]))
203 return -1;
204 else
205 return fls(word[word_offset << 2]) - 1 + (word_offset << 5);
208 static inline int apic_test_and_set_irr(int vec, struct kvm_lapic *apic)
210 apic->irr_pending = true;
211 return apic_test_and_set_vector(vec, apic->regs + APIC_IRR);
214 static inline int apic_search_irr(struct kvm_lapic *apic)
216 return find_highest_vector(apic->regs + APIC_IRR);
219 static inline int apic_find_highest_irr(struct kvm_lapic *apic)
221 int result;
223 if (!apic->irr_pending)
224 return -1;
226 result = apic_search_irr(apic);
227 ASSERT(result == -1 || result >= 16);
229 return result;
232 static inline void apic_clear_irr(int vec, struct kvm_lapic *apic)
234 apic->irr_pending = false;
235 apic_clear_vector(vec, apic->regs + APIC_IRR);
236 if (apic_search_irr(apic) != -1)
237 apic->irr_pending = true;
240 int kvm_lapic_find_highest_irr(struct kvm_vcpu *vcpu)
242 struct kvm_lapic *apic = vcpu->arch.apic;
243 int highest_irr;
245 /* This may race with setting of irr in __apic_accept_irq() and
246 * value returned may be wrong, but kvm_vcpu_kick() in __apic_accept_irq
247 * will cause vmexit immediately and the value will be recalculated
248 * on the next vmentry.
250 if (!apic)
251 return 0;
252 highest_irr = apic_find_highest_irr(apic);
254 return highest_irr;
257 static int __apic_accept_irq(struct kvm_lapic *apic, int delivery_mode,
258 int vector, int level, int trig_mode);
260 int kvm_apic_set_irq(struct kvm_vcpu *vcpu, struct kvm_lapic_irq *irq)
262 struct kvm_lapic *apic = vcpu->arch.apic;
264 return __apic_accept_irq(apic, irq->delivery_mode, irq->vector,
265 irq->level, irq->trig_mode);
268 static inline int apic_find_highest_isr(struct kvm_lapic *apic)
270 int result;
272 result = find_highest_vector(apic->regs + APIC_ISR);
273 ASSERT(result == -1 || result >= 16);
275 return result;
278 static void apic_update_ppr(struct kvm_lapic *apic)
280 u32 tpr, isrv, ppr, old_ppr;
281 int isr;
283 old_ppr = apic_get_reg(apic, APIC_PROCPRI);
284 tpr = apic_get_reg(apic, APIC_TASKPRI);
285 isr = apic_find_highest_isr(apic);
286 isrv = (isr != -1) ? isr : 0;
288 if ((tpr & 0xf0) >= (isrv & 0xf0))
289 ppr = tpr & 0xff;
290 else
291 ppr = isrv & 0xf0;
293 apic_debug("vlapic %p, ppr 0x%x, isr 0x%x, isrv 0x%x",
294 apic, ppr, isr, isrv);
296 if (old_ppr != ppr) {
297 apic_set_reg(apic, APIC_PROCPRI, ppr);
298 if (ppr < old_ppr)
299 kvm_make_request(KVM_REQ_EVENT, apic->vcpu);
303 static void apic_set_tpr(struct kvm_lapic *apic, u32 tpr)
305 apic_set_reg(apic, APIC_TASKPRI, tpr);
306 apic_update_ppr(apic);
309 int kvm_apic_match_physical_addr(struct kvm_lapic *apic, u16 dest)
311 return dest == 0xff || kvm_apic_id(apic) == dest;
314 int kvm_apic_match_logical_addr(struct kvm_lapic *apic, u8 mda)
316 int result = 0;
317 u32 logical_id;
319 if (apic_x2apic_mode(apic)) {
320 logical_id = apic_get_reg(apic, APIC_LDR);
321 return logical_id & mda;
324 logical_id = GET_APIC_LOGICAL_ID(apic_get_reg(apic, APIC_LDR));
326 switch (apic_get_reg(apic, APIC_DFR)) {
327 case APIC_DFR_FLAT:
328 if (logical_id & mda)
329 result = 1;
330 break;
331 case APIC_DFR_CLUSTER:
332 if (((logical_id >> 4) == (mda >> 0x4))
333 && (logical_id & mda & 0xf))
334 result = 1;
335 break;
336 default:
337 apic_debug("Bad DFR vcpu %d: %08x\n",
338 apic->vcpu->vcpu_id, apic_get_reg(apic, APIC_DFR));
339 break;
342 return result;
345 int kvm_apic_match_dest(struct kvm_vcpu *vcpu, struct kvm_lapic *source,
346 int short_hand, int dest, int dest_mode)
348 int result = 0;
349 struct kvm_lapic *target = vcpu->arch.apic;
351 apic_debug("target %p, source %p, dest 0x%x, "
352 "dest_mode 0x%x, short_hand 0x%x\n",
353 target, source, dest, dest_mode, short_hand);
355 ASSERT(target);
356 switch (short_hand) {
357 case APIC_DEST_NOSHORT:
358 if (dest_mode == 0)
359 /* Physical mode. */
360 result = kvm_apic_match_physical_addr(target, dest);
361 else
362 /* Logical mode. */
363 result = kvm_apic_match_logical_addr(target, dest);
364 break;
365 case APIC_DEST_SELF:
366 result = (target == source);
367 break;
368 case APIC_DEST_ALLINC:
369 result = 1;
370 break;
371 case APIC_DEST_ALLBUT:
372 result = (target != source);
373 break;
374 default:
375 apic_debug("kvm: apic: Bad dest shorthand value %x\n",
376 short_hand);
377 break;
380 return result;
384 * Add a pending IRQ into lapic.
385 * Return 1 if successfully added and 0 if discarded.
387 static int __apic_accept_irq(struct kvm_lapic *apic, int delivery_mode,
388 int vector, int level, int trig_mode)
390 int result = 0;
391 struct kvm_vcpu *vcpu = apic->vcpu;
393 switch (delivery_mode) {
394 case APIC_DM_LOWEST:
395 vcpu->arch.apic_arb_prio++;
396 case APIC_DM_FIXED:
397 /* FIXME add logic for vcpu on reset */
398 if (unlikely(!apic_enabled(apic)))
399 break;
401 if (trig_mode) {
402 apic_debug("level trig mode for vector %d", vector);
403 apic_set_vector(vector, apic->regs + APIC_TMR);
404 } else
405 apic_clear_vector(vector, apic->regs + APIC_TMR);
407 result = !apic_test_and_set_irr(vector, apic);
408 trace_kvm_apic_accept_irq(vcpu->vcpu_id, delivery_mode,
409 trig_mode, vector, !result);
410 if (!result) {
411 if (trig_mode)
412 apic_debug("level trig mode repeatedly for "
413 "vector %d", vector);
414 break;
417 kvm_make_request(KVM_REQ_EVENT, vcpu);
418 kvm_vcpu_kick(vcpu);
419 break;
421 case APIC_DM_REMRD:
422 apic_debug("Ignoring delivery mode 3\n");
423 break;
425 case APIC_DM_SMI:
426 apic_debug("Ignoring guest SMI\n");
427 break;
429 case APIC_DM_NMI:
430 result = 1;
431 kvm_inject_nmi(vcpu);
432 kvm_vcpu_kick(vcpu);
433 break;
435 case APIC_DM_INIT:
436 if (level) {
437 result = 1;
438 vcpu->arch.mp_state = KVM_MP_STATE_INIT_RECEIVED;
439 kvm_make_request(KVM_REQ_EVENT, vcpu);
440 kvm_vcpu_kick(vcpu);
441 } else {
442 apic_debug("Ignoring de-assert INIT to vcpu %d\n",
443 vcpu->vcpu_id);
445 break;
447 case APIC_DM_STARTUP:
448 apic_debug("SIPI to vcpu %d vector 0x%02x\n",
449 vcpu->vcpu_id, vector);
450 if (vcpu->arch.mp_state == KVM_MP_STATE_INIT_RECEIVED) {
451 result = 1;
452 vcpu->arch.sipi_vector = vector;
453 vcpu->arch.mp_state = KVM_MP_STATE_SIPI_RECEIVED;
454 kvm_make_request(KVM_REQ_EVENT, vcpu);
455 kvm_vcpu_kick(vcpu);
457 break;
459 case APIC_DM_EXTINT:
461 * Should only be called by kvm_apic_local_deliver() with LVT0,
462 * before NMI watchdog was enabled. Already handled by
463 * kvm_apic_accept_pic_intr().
465 break;
467 default:
468 printk(KERN_ERR "TODO: unsupported delivery mode %x\n",
469 delivery_mode);
470 break;
472 return result;
475 int kvm_apic_compare_prio(struct kvm_vcpu *vcpu1, struct kvm_vcpu *vcpu2)
477 return vcpu1->arch.apic_arb_prio - vcpu2->arch.apic_arb_prio;
480 static void apic_set_eoi(struct kvm_lapic *apic)
482 int vector = apic_find_highest_isr(apic);
483 int trigger_mode;
485 * Not every write EOI will has corresponding ISR,
486 * one example is when Kernel check timer on setup_IO_APIC
488 if (vector == -1)
489 return;
491 apic_clear_vector(vector, apic->regs + APIC_ISR);
492 apic_update_ppr(apic);
494 if (apic_test_and_clear_vector(vector, apic->regs + APIC_TMR))
495 trigger_mode = IOAPIC_LEVEL_TRIG;
496 else
497 trigger_mode = IOAPIC_EDGE_TRIG;
498 if (!(apic_get_reg(apic, APIC_SPIV) & APIC_SPIV_DIRECTED_EOI))
499 kvm_ioapic_update_eoi(apic->vcpu->kvm, vector, trigger_mode);
500 kvm_make_request(KVM_REQ_EVENT, apic->vcpu);
503 static void apic_send_ipi(struct kvm_lapic *apic)
505 u32 icr_low = apic_get_reg(apic, APIC_ICR);
506 u32 icr_high = apic_get_reg(apic, APIC_ICR2);
507 struct kvm_lapic_irq irq;
509 irq.vector = icr_low & APIC_VECTOR_MASK;
510 irq.delivery_mode = icr_low & APIC_MODE_MASK;
511 irq.dest_mode = icr_low & APIC_DEST_MASK;
512 irq.level = icr_low & APIC_INT_ASSERT;
513 irq.trig_mode = icr_low & APIC_INT_LEVELTRIG;
514 irq.shorthand = icr_low & APIC_SHORT_MASK;
515 if (apic_x2apic_mode(apic))
516 irq.dest_id = icr_high;
517 else
518 irq.dest_id = GET_APIC_DEST_FIELD(icr_high);
520 trace_kvm_apic_ipi(icr_low, irq.dest_id);
522 apic_debug("icr_high 0x%x, icr_low 0x%x, "
523 "short_hand 0x%x, dest 0x%x, trig_mode 0x%x, level 0x%x, "
524 "dest_mode 0x%x, delivery_mode 0x%x, vector 0x%x\n",
525 icr_high, icr_low, irq.shorthand, irq.dest_id,
526 irq.trig_mode, irq.level, irq.dest_mode, irq.delivery_mode,
527 irq.vector);
529 kvm_irq_delivery_to_apic(apic->vcpu->kvm, apic, &irq);
532 static u32 apic_get_tmcct(struct kvm_lapic *apic)
534 ktime_t remaining;
535 s64 ns;
536 u32 tmcct;
538 ASSERT(apic != NULL);
540 /* if initial count is 0, current count should also be 0 */
541 if (apic_get_reg(apic, APIC_TMICT) == 0)
542 return 0;
544 remaining = hrtimer_get_remaining(&apic->lapic_timer.timer);
545 if (ktime_to_ns(remaining) < 0)
546 remaining = ktime_set(0, 0);
548 ns = mod_64(ktime_to_ns(remaining), apic->lapic_timer.period);
549 tmcct = div64_u64(ns,
550 (APIC_BUS_CYCLE_NS * apic->divide_count));
552 return tmcct;
555 static void __report_tpr_access(struct kvm_lapic *apic, bool write)
557 struct kvm_vcpu *vcpu = apic->vcpu;
558 struct kvm_run *run = vcpu->run;
560 kvm_make_request(KVM_REQ_REPORT_TPR_ACCESS, vcpu);
561 run->tpr_access.rip = kvm_rip_read(vcpu);
562 run->tpr_access.is_write = write;
565 static inline void report_tpr_access(struct kvm_lapic *apic, bool write)
567 if (apic->vcpu->arch.tpr_access_reporting)
568 __report_tpr_access(apic, write);
571 static u32 __apic_read(struct kvm_lapic *apic, unsigned int offset)
573 u32 val = 0;
575 if (offset >= LAPIC_MMIO_LENGTH)
576 return 0;
578 switch (offset) {
579 case APIC_ID:
580 if (apic_x2apic_mode(apic))
581 val = kvm_apic_id(apic);
582 else
583 val = kvm_apic_id(apic) << 24;
584 break;
585 case APIC_ARBPRI:
586 apic_debug("Access APIC ARBPRI register which is for P6\n");
587 break;
589 case APIC_TMCCT: /* Timer CCR */
590 if (apic_lvtt_tscdeadline(apic))
591 return 0;
593 val = apic_get_tmcct(apic);
594 break;
596 case APIC_TASKPRI:
597 report_tpr_access(apic, false);
598 /* fall thru */
599 default:
600 apic_update_ppr(apic);
601 val = apic_get_reg(apic, offset);
602 break;
605 return val;
608 static inline struct kvm_lapic *to_lapic(struct kvm_io_device *dev)
610 return container_of(dev, struct kvm_lapic, dev);
613 static int apic_reg_read(struct kvm_lapic *apic, u32 offset, int len,
614 void *data)
616 unsigned char alignment = offset & 0xf;
617 u32 result;
618 /* this bitmask has a bit cleared for each reserver register */
619 static const u64 rmask = 0x43ff01ffffffe70cULL;
621 if ((alignment + len) > 4) {
622 apic_debug("KVM_APIC_READ: alignment error %x %d\n",
623 offset, len);
624 return 1;
627 if (offset > 0x3f0 || !(rmask & (1ULL << (offset >> 4)))) {
628 apic_debug("KVM_APIC_READ: read reserved register %x\n",
629 offset);
630 return 1;
633 result = __apic_read(apic, offset & ~0xf);
635 trace_kvm_apic_read(offset, result);
637 switch (len) {
638 case 1:
639 case 2:
640 case 4:
641 memcpy(data, (char *)&result + alignment, len);
642 break;
643 default:
644 printk(KERN_ERR "Local APIC read with len = %x, "
645 "should be 1,2, or 4 instead\n", len);
646 break;
648 return 0;
651 static int apic_mmio_in_range(struct kvm_lapic *apic, gpa_t addr)
653 return apic_hw_enabled(apic) &&
654 addr >= apic->base_address &&
655 addr < apic->base_address + LAPIC_MMIO_LENGTH;
658 static int apic_mmio_read(struct kvm_io_device *this,
659 gpa_t address, int len, void *data)
661 struct kvm_lapic *apic = to_lapic(this);
662 u32 offset = address - apic->base_address;
664 if (!apic_mmio_in_range(apic, address))
665 return -EOPNOTSUPP;
667 apic_reg_read(apic, offset, len, data);
669 return 0;
672 static void update_divide_count(struct kvm_lapic *apic)
674 u32 tmp1, tmp2, tdcr;
676 tdcr = apic_get_reg(apic, APIC_TDCR);
677 tmp1 = tdcr & 0xf;
678 tmp2 = ((tmp1 & 0x3) | ((tmp1 & 0x8) >> 1)) + 1;
679 apic->divide_count = 0x1 << (tmp2 & 0x7);
681 apic_debug("timer divide count is 0x%x\n",
682 apic->divide_count);
685 static void start_apic_timer(struct kvm_lapic *apic)
687 ktime_t now;
688 atomic_set(&apic->lapic_timer.pending, 0);
690 if (apic_lvtt_period(apic) || apic_lvtt_oneshot(apic)) {
691 /* lapic timer in oneshot or peroidic mode */
692 now = apic->lapic_timer.timer.base->get_time();
693 apic->lapic_timer.period = (u64)apic_get_reg(apic, APIC_TMICT)
694 * APIC_BUS_CYCLE_NS * apic->divide_count;
696 if (!apic->lapic_timer.period)
697 return;
699 * Do not allow the guest to program periodic timers with small
700 * interval, since the hrtimers are not throttled by the host
701 * scheduler.
703 if (apic_lvtt_period(apic)) {
704 s64 min_period = min_timer_period_us * 1000LL;
706 if (apic->lapic_timer.period < min_period) {
707 pr_info_ratelimited(
708 "kvm: vcpu %i: requested %lld ns "
709 "lapic timer period limited to %lld ns\n",
710 apic->vcpu->vcpu_id,
711 apic->lapic_timer.period, min_period);
712 apic->lapic_timer.period = min_period;
716 hrtimer_start(&apic->lapic_timer.timer,
717 ktime_add_ns(now, apic->lapic_timer.period),
718 HRTIMER_MODE_ABS);
720 apic_debug("%s: bus cycle is %" PRId64 "ns, now 0x%016"
721 PRIx64 ", "
722 "timer initial count 0x%x, period %lldns, "
723 "expire @ 0x%016" PRIx64 ".\n", __func__,
724 APIC_BUS_CYCLE_NS, ktime_to_ns(now),
725 apic_get_reg(apic, APIC_TMICT),
726 apic->lapic_timer.period,
727 ktime_to_ns(ktime_add_ns(now,
728 apic->lapic_timer.period)));
729 } else if (apic_lvtt_tscdeadline(apic)) {
730 /* lapic timer in tsc deadline mode */
731 u64 guest_tsc, tscdeadline = apic->lapic_timer.tscdeadline;
732 u64 ns = 0;
733 struct kvm_vcpu *vcpu = apic->vcpu;
734 unsigned long this_tsc_khz = vcpu_tsc_khz(vcpu);
735 unsigned long flags;
737 if (unlikely(!tscdeadline || !this_tsc_khz))
738 return;
740 local_irq_save(flags);
742 now = apic->lapic_timer.timer.base->get_time();
743 guest_tsc = kvm_x86_ops->read_l1_tsc(vcpu);
744 if (likely(tscdeadline > guest_tsc)) {
745 ns = (tscdeadline - guest_tsc) * 1000000ULL;
746 do_div(ns, this_tsc_khz);
748 hrtimer_start(&apic->lapic_timer.timer,
749 ktime_add_ns(now, ns), HRTIMER_MODE_ABS);
751 local_irq_restore(flags);
755 static void apic_manage_nmi_watchdog(struct kvm_lapic *apic, u32 lvt0_val)
757 int nmi_wd_enabled = apic_lvt_nmi_mode(apic_get_reg(apic, APIC_LVT0));
759 if (apic_lvt_nmi_mode(lvt0_val)) {
760 if (!nmi_wd_enabled) {
761 apic_debug("Receive NMI setting on APIC_LVT0 "
762 "for cpu %d\n", apic->vcpu->vcpu_id);
763 apic->vcpu->kvm->arch.vapics_in_nmi_mode++;
765 } else if (nmi_wd_enabled)
766 apic->vcpu->kvm->arch.vapics_in_nmi_mode--;
769 static int apic_reg_write(struct kvm_lapic *apic, u32 reg, u32 val)
771 int ret = 0;
773 trace_kvm_apic_write(reg, val);
775 switch (reg) {
776 case APIC_ID: /* Local APIC ID */
777 if (!apic_x2apic_mode(apic))
778 apic_set_reg(apic, APIC_ID, val);
779 else
780 ret = 1;
781 break;
783 case APIC_TASKPRI:
784 report_tpr_access(apic, true);
785 apic_set_tpr(apic, val & 0xff);
786 break;
788 case APIC_EOI:
789 apic_set_eoi(apic);
790 break;
792 case APIC_LDR:
793 if (!apic_x2apic_mode(apic))
794 apic_set_reg(apic, APIC_LDR, val & APIC_LDR_MASK);
795 else
796 ret = 1;
797 break;
799 case APIC_DFR:
800 if (!apic_x2apic_mode(apic))
801 apic_set_reg(apic, APIC_DFR, val | 0x0FFFFFFF);
802 else
803 ret = 1;
804 break;
806 case APIC_SPIV: {
807 u32 mask = 0x3ff;
808 if (apic_get_reg(apic, APIC_LVR) & APIC_LVR_DIRECTED_EOI)
809 mask |= APIC_SPIV_DIRECTED_EOI;
810 apic_set_reg(apic, APIC_SPIV, val & mask);
811 if (!(val & APIC_SPIV_APIC_ENABLED)) {
812 int i;
813 u32 lvt_val;
815 for (i = 0; i < APIC_LVT_NUM; i++) {
816 lvt_val = apic_get_reg(apic,
817 APIC_LVTT + 0x10 * i);
818 apic_set_reg(apic, APIC_LVTT + 0x10 * i,
819 lvt_val | APIC_LVT_MASKED);
821 atomic_set(&apic->lapic_timer.pending, 0);
824 break;
826 case APIC_ICR:
827 /* No delay here, so we always clear the pending bit */
828 apic_set_reg(apic, APIC_ICR, val & ~(1 << 12));
829 apic_send_ipi(apic);
830 break;
832 case APIC_ICR2:
833 if (!apic_x2apic_mode(apic))
834 val &= 0xff000000;
835 apic_set_reg(apic, APIC_ICR2, val);
836 break;
838 case APIC_LVT0:
839 apic_manage_nmi_watchdog(apic, val);
840 case APIC_LVTTHMR:
841 case APIC_LVTPC:
842 case APIC_LVT1:
843 case APIC_LVTERR:
844 /* TODO: Check vector */
845 if (!apic_sw_enabled(apic))
846 val |= APIC_LVT_MASKED;
848 val &= apic_lvt_mask[(reg - APIC_LVTT) >> 4];
849 apic_set_reg(apic, reg, val);
851 break;
853 case APIC_LVTT:
854 if ((apic_get_reg(apic, APIC_LVTT) &
855 apic->lapic_timer.timer_mode_mask) !=
856 (val & apic->lapic_timer.timer_mode_mask))
857 hrtimer_cancel(&apic->lapic_timer.timer);
859 if (!apic_sw_enabled(apic))
860 val |= APIC_LVT_MASKED;
861 val &= (apic_lvt_mask[0] | apic->lapic_timer.timer_mode_mask);
862 apic_set_reg(apic, APIC_LVTT, val);
863 break;
865 case APIC_TMICT:
866 if (apic_lvtt_tscdeadline(apic))
867 break;
869 hrtimer_cancel(&apic->lapic_timer.timer);
870 apic_set_reg(apic, APIC_TMICT, val);
871 start_apic_timer(apic);
872 break;
874 case APIC_TDCR:
875 if (val & 4)
876 apic_debug("KVM_WRITE:TDCR %x\n", val);
877 apic_set_reg(apic, APIC_TDCR, val);
878 update_divide_count(apic);
879 break;
881 case APIC_ESR:
882 if (apic_x2apic_mode(apic) && val != 0) {
883 apic_debug("KVM_WRITE:ESR not zero %x\n", val);
884 ret = 1;
886 break;
888 case APIC_SELF_IPI:
889 if (apic_x2apic_mode(apic)) {
890 apic_reg_write(apic, APIC_ICR, 0x40000 | (val & 0xff));
891 } else
892 ret = 1;
893 break;
894 default:
895 ret = 1;
896 break;
898 if (ret)
899 apic_debug("Local APIC Write to read-only register %x\n", reg);
900 return ret;
903 static int apic_mmio_write(struct kvm_io_device *this,
904 gpa_t address, int len, const void *data)
906 struct kvm_lapic *apic = to_lapic(this);
907 unsigned int offset = address - apic->base_address;
908 u32 val;
910 if (!apic_mmio_in_range(apic, address))
911 return -EOPNOTSUPP;
914 * APIC register must be aligned on 128-bits boundary.
915 * 32/64/128 bits registers must be accessed thru 32 bits.
916 * Refer SDM 8.4.1
918 if (len != 4 || (offset & 0xf)) {
919 /* Don't shout loud, $infamous_os would cause only noise. */
920 apic_debug("apic write: bad size=%d %lx\n", len, (long)address);
921 return 0;
924 val = *(u32*)data;
926 /* too common printing */
927 if (offset != APIC_EOI)
928 apic_debug("%s: offset 0x%x with length 0x%x, and value is "
929 "0x%x\n", __func__, offset, len, val);
931 apic_reg_write(apic, offset & 0xff0, val);
933 return 0;
936 void kvm_lapic_set_eoi(struct kvm_vcpu *vcpu)
938 struct kvm_lapic *apic = vcpu->arch.apic;
940 if (apic)
941 apic_reg_write(vcpu->arch.apic, APIC_EOI, 0);
943 EXPORT_SYMBOL_GPL(kvm_lapic_set_eoi);
945 void kvm_free_lapic(struct kvm_vcpu *vcpu)
947 if (!vcpu->arch.apic)
948 return;
950 hrtimer_cancel(&vcpu->arch.apic->lapic_timer.timer);
952 if (vcpu->arch.apic->regs)
953 free_page((unsigned long)vcpu->arch.apic->regs);
955 kfree(vcpu->arch.apic);
959 *----------------------------------------------------------------------
960 * LAPIC interface
961 *----------------------------------------------------------------------
964 u64 kvm_get_lapic_tscdeadline_msr(struct kvm_vcpu *vcpu)
966 struct kvm_lapic *apic = vcpu->arch.apic;
967 if (!apic)
968 return 0;
970 if (apic_lvtt_oneshot(apic) || apic_lvtt_period(apic))
971 return 0;
973 return apic->lapic_timer.tscdeadline;
976 void kvm_set_lapic_tscdeadline_msr(struct kvm_vcpu *vcpu, u64 data)
978 struct kvm_lapic *apic = vcpu->arch.apic;
979 if (!apic)
980 return;
982 if (apic_lvtt_oneshot(apic) || apic_lvtt_period(apic))
983 return;
985 hrtimer_cancel(&apic->lapic_timer.timer);
986 apic->lapic_timer.tscdeadline = data;
987 start_apic_timer(apic);
990 void kvm_lapic_set_tpr(struct kvm_vcpu *vcpu, unsigned long cr8)
992 struct kvm_lapic *apic = vcpu->arch.apic;
994 if (!apic)
995 return;
996 apic_set_tpr(apic, ((cr8 & 0x0f) << 4)
997 | (apic_get_reg(apic, APIC_TASKPRI) & 4));
1000 u64 kvm_lapic_get_cr8(struct kvm_vcpu *vcpu)
1002 struct kvm_lapic *apic = vcpu->arch.apic;
1003 u64 tpr;
1005 if (!apic)
1006 return 0;
1007 tpr = (u64) apic_get_reg(apic, APIC_TASKPRI);
1009 return (tpr & 0xf0) >> 4;
1012 void kvm_lapic_set_base(struct kvm_vcpu *vcpu, u64 value)
1014 struct kvm_lapic *apic = vcpu->arch.apic;
1016 if (!apic) {
1017 value |= MSR_IA32_APICBASE_BSP;
1018 vcpu->arch.apic_base = value;
1019 return;
1022 if (!kvm_vcpu_is_bsp(apic->vcpu))
1023 value &= ~MSR_IA32_APICBASE_BSP;
1025 vcpu->arch.apic_base = value;
1026 if (apic_x2apic_mode(apic)) {
1027 u32 id = kvm_apic_id(apic);
1028 u32 ldr = ((id & ~0xf) << 16) | (1 << (id & 0xf));
1029 apic_set_reg(apic, APIC_LDR, ldr);
1031 apic->base_address = apic->vcpu->arch.apic_base &
1032 MSR_IA32_APICBASE_BASE;
1034 /* with FSB delivery interrupt, we can restart APIC functionality */
1035 apic_debug("apic base msr is 0x%016" PRIx64 ", and base address is "
1036 "0x%lx.\n", apic->vcpu->arch.apic_base, apic->base_address);
1040 void kvm_lapic_reset(struct kvm_vcpu *vcpu)
1042 struct kvm_lapic *apic;
1043 int i;
1045 apic_debug("%s\n", __func__);
1047 ASSERT(vcpu);
1048 apic = vcpu->arch.apic;
1049 ASSERT(apic != NULL);
1051 /* Stop the timer in case it's a reset to an active apic */
1052 hrtimer_cancel(&apic->lapic_timer.timer);
1054 apic_set_reg(apic, APIC_ID, vcpu->vcpu_id << 24);
1055 kvm_apic_set_version(apic->vcpu);
1057 for (i = 0; i < APIC_LVT_NUM; i++)
1058 apic_set_reg(apic, APIC_LVTT + 0x10 * i, APIC_LVT_MASKED);
1059 apic_set_reg(apic, APIC_LVT0,
1060 SET_APIC_DELIVERY_MODE(0, APIC_MODE_EXTINT));
1062 apic_set_reg(apic, APIC_DFR, 0xffffffffU);
1063 apic_set_reg(apic, APIC_SPIV, 0xff);
1064 apic_set_reg(apic, APIC_TASKPRI, 0);
1065 apic_set_reg(apic, APIC_LDR, 0);
1066 apic_set_reg(apic, APIC_ESR, 0);
1067 apic_set_reg(apic, APIC_ICR, 0);
1068 apic_set_reg(apic, APIC_ICR2, 0);
1069 apic_set_reg(apic, APIC_TDCR, 0);
1070 apic_set_reg(apic, APIC_TMICT, 0);
1071 for (i = 0; i < 8; i++) {
1072 apic_set_reg(apic, APIC_IRR + 0x10 * i, 0);
1073 apic_set_reg(apic, APIC_ISR + 0x10 * i, 0);
1074 apic_set_reg(apic, APIC_TMR + 0x10 * i, 0);
1076 apic->irr_pending = false;
1077 update_divide_count(apic);
1078 atomic_set(&apic->lapic_timer.pending, 0);
1079 if (kvm_vcpu_is_bsp(vcpu))
1080 vcpu->arch.apic_base |= MSR_IA32_APICBASE_BSP;
1081 apic_update_ppr(apic);
1083 vcpu->arch.apic_arb_prio = 0;
1085 apic_debug(KERN_INFO "%s: vcpu=%p, id=%d, base_msr="
1086 "0x%016" PRIx64 ", base_address=0x%0lx.\n", __func__,
1087 vcpu, kvm_apic_id(apic),
1088 vcpu->arch.apic_base, apic->base_address);
1091 bool kvm_apic_present(struct kvm_vcpu *vcpu)
1093 return vcpu->arch.apic && apic_hw_enabled(vcpu->arch.apic);
1096 int kvm_lapic_enabled(struct kvm_vcpu *vcpu)
1098 return kvm_apic_present(vcpu) && apic_sw_enabled(vcpu->arch.apic);
1102 *----------------------------------------------------------------------
1103 * timer interface
1104 *----------------------------------------------------------------------
1107 static bool lapic_is_periodic(struct kvm_timer *ktimer)
1109 struct kvm_lapic *apic = container_of(ktimer, struct kvm_lapic,
1110 lapic_timer);
1111 return apic_lvtt_period(apic);
1114 int apic_has_pending_timer(struct kvm_vcpu *vcpu)
1116 struct kvm_lapic *lapic = vcpu->arch.apic;
1118 if (lapic && apic_enabled(lapic) && apic_lvt_enabled(lapic, APIC_LVTT))
1119 return atomic_read(&lapic->lapic_timer.pending);
1121 return 0;
1124 int kvm_apic_local_deliver(struct kvm_lapic *apic, int lvt_type)
1126 u32 reg = apic_get_reg(apic, lvt_type);
1127 int vector, mode, trig_mode;
1129 if (apic_hw_enabled(apic) && !(reg & APIC_LVT_MASKED)) {
1130 vector = reg & APIC_VECTOR_MASK;
1131 mode = reg & APIC_MODE_MASK;
1132 trig_mode = reg & APIC_LVT_LEVEL_TRIGGER;
1133 return __apic_accept_irq(apic, mode, vector, 1, trig_mode);
1135 return 0;
1138 void kvm_apic_nmi_wd_deliver(struct kvm_vcpu *vcpu)
1140 struct kvm_lapic *apic = vcpu->arch.apic;
1142 if (apic)
1143 kvm_apic_local_deliver(apic, APIC_LVT0);
1146 static struct kvm_timer_ops lapic_timer_ops = {
1147 .is_periodic = lapic_is_periodic,
1150 static const struct kvm_io_device_ops apic_mmio_ops = {
1151 .read = apic_mmio_read,
1152 .write = apic_mmio_write,
1155 int kvm_create_lapic(struct kvm_vcpu *vcpu)
1157 struct kvm_lapic *apic;
1159 ASSERT(vcpu != NULL);
1160 apic_debug("apic_init %d\n", vcpu->vcpu_id);
1162 apic = kzalloc(sizeof(*apic), GFP_KERNEL);
1163 if (!apic)
1164 goto nomem;
1166 vcpu->arch.apic = apic;
1168 apic->regs = (void *)get_zeroed_page(GFP_KERNEL);
1169 if (!apic->regs) {
1170 printk(KERN_ERR "malloc apic regs error for vcpu %x\n",
1171 vcpu->vcpu_id);
1172 goto nomem_free_apic;
1174 apic->vcpu = vcpu;
1176 hrtimer_init(&apic->lapic_timer.timer, CLOCK_MONOTONIC,
1177 HRTIMER_MODE_ABS);
1178 apic->lapic_timer.timer.function = kvm_timer_fn;
1179 apic->lapic_timer.t_ops = &lapic_timer_ops;
1180 apic->lapic_timer.kvm = vcpu->kvm;
1181 apic->lapic_timer.vcpu = vcpu;
1183 apic->base_address = APIC_DEFAULT_PHYS_BASE;
1184 vcpu->arch.apic_base = APIC_DEFAULT_PHYS_BASE;
1186 kvm_lapic_reset(vcpu);
1187 kvm_iodevice_init(&apic->dev, &apic_mmio_ops);
1189 return 0;
1190 nomem_free_apic:
1191 kfree(apic);
1192 nomem:
1193 return -ENOMEM;
1196 int kvm_apic_has_interrupt(struct kvm_vcpu *vcpu)
1198 struct kvm_lapic *apic = vcpu->arch.apic;
1199 int highest_irr;
1201 if (!apic || !apic_enabled(apic))
1202 return -1;
1204 apic_update_ppr(apic);
1205 highest_irr = apic_find_highest_irr(apic);
1206 if ((highest_irr == -1) ||
1207 ((highest_irr & 0xF0) <= apic_get_reg(apic, APIC_PROCPRI)))
1208 return -1;
1209 return highest_irr;
1212 int kvm_apic_accept_pic_intr(struct kvm_vcpu *vcpu)
1214 u32 lvt0 = apic_get_reg(vcpu->arch.apic, APIC_LVT0);
1215 int r = 0;
1217 if (!apic_hw_enabled(vcpu->arch.apic))
1218 r = 1;
1219 if ((lvt0 & APIC_LVT_MASKED) == 0 &&
1220 GET_APIC_DELIVERY_MODE(lvt0) == APIC_MODE_EXTINT)
1221 r = 1;
1222 return r;
1225 void kvm_inject_apic_timer_irqs(struct kvm_vcpu *vcpu)
1227 struct kvm_lapic *apic = vcpu->arch.apic;
1229 if (apic && atomic_read(&apic->lapic_timer.pending) > 0) {
1230 if (kvm_apic_local_deliver(apic, APIC_LVTT))
1231 atomic_dec(&apic->lapic_timer.pending);
1235 int kvm_get_apic_interrupt(struct kvm_vcpu *vcpu)
1237 int vector = kvm_apic_has_interrupt(vcpu);
1238 struct kvm_lapic *apic = vcpu->arch.apic;
1240 if (vector == -1)
1241 return -1;
1243 apic_set_vector(vector, apic->regs + APIC_ISR);
1244 apic_update_ppr(apic);
1245 apic_clear_irr(vector, apic);
1246 return vector;
1249 void kvm_apic_post_state_restore(struct kvm_vcpu *vcpu)
1251 struct kvm_lapic *apic = vcpu->arch.apic;
1253 apic->base_address = vcpu->arch.apic_base &
1254 MSR_IA32_APICBASE_BASE;
1255 kvm_apic_set_version(vcpu);
1257 apic_update_ppr(apic);
1258 hrtimer_cancel(&apic->lapic_timer.timer);
1259 update_divide_count(apic);
1260 start_apic_timer(apic);
1261 apic->irr_pending = true;
1262 kvm_make_request(KVM_REQ_EVENT, vcpu);
1265 void __kvm_migrate_apic_timer(struct kvm_vcpu *vcpu)
1267 struct kvm_lapic *apic = vcpu->arch.apic;
1268 struct hrtimer *timer;
1270 if (!apic)
1271 return;
1273 timer = &apic->lapic_timer.timer;
1274 if (hrtimer_cancel(timer))
1275 hrtimer_start_expires(timer, HRTIMER_MODE_ABS);
1278 void kvm_lapic_sync_from_vapic(struct kvm_vcpu *vcpu)
1280 u32 data;
1281 void *vapic;
1283 if (!irqchip_in_kernel(vcpu->kvm) || !vcpu->arch.apic->vapic_addr)
1284 return;
1286 vapic = kmap_atomic(vcpu->arch.apic->vapic_page, KM_USER0);
1287 data = *(u32 *)(vapic + offset_in_page(vcpu->arch.apic->vapic_addr));
1288 kunmap_atomic(vapic, KM_USER0);
1290 apic_set_tpr(vcpu->arch.apic, data & 0xff);
1293 void kvm_lapic_sync_to_vapic(struct kvm_vcpu *vcpu)
1295 u32 data, tpr;
1296 int max_irr, max_isr;
1297 struct kvm_lapic *apic;
1298 void *vapic;
1300 if (!irqchip_in_kernel(vcpu->kvm) || !vcpu->arch.apic->vapic_addr)
1301 return;
1303 apic = vcpu->arch.apic;
1304 tpr = apic_get_reg(apic, APIC_TASKPRI) & 0xff;
1305 max_irr = apic_find_highest_irr(apic);
1306 if (max_irr < 0)
1307 max_irr = 0;
1308 max_isr = apic_find_highest_isr(apic);
1309 if (max_isr < 0)
1310 max_isr = 0;
1311 data = (tpr & 0xff) | ((max_isr & 0xf0) << 8) | (max_irr << 24);
1313 vapic = kmap_atomic(vcpu->arch.apic->vapic_page, KM_USER0);
1314 *(u32 *)(vapic + offset_in_page(vcpu->arch.apic->vapic_addr)) = data;
1315 kunmap_atomic(vapic, KM_USER0);
1318 void kvm_lapic_set_vapic_addr(struct kvm_vcpu *vcpu, gpa_t vapic_addr)
1320 if (!irqchip_in_kernel(vcpu->kvm))
1321 return;
1323 vcpu->arch.apic->vapic_addr = vapic_addr;
1326 int kvm_x2apic_msr_write(struct kvm_vcpu *vcpu, u32 msr, u64 data)
1328 struct kvm_lapic *apic = vcpu->arch.apic;
1329 u32 reg = (msr - APIC_BASE_MSR) << 4;
1331 if (!irqchip_in_kernel(vcpu->kvm) || !apic_x2apic_mode(apic))
1332 return 1;
1334 /* if this is ICR write vector before command */
1335 if (msr == 0x830)
1336 apic_reg_write(apic, APIC_ICR2, (u32)(data >> 32));
1337 return apic_reg_write(apic, reg, (u32)data);
1340 int kvm_x2apic_msr_read(struct kvm_vcpu *vcpu, u32 msr, u64 *data)
1342 struct kvm_lapic *apic = vcpu->arch.apic;
1343 u32 reg = (msr - APIC_BASE_MSR) << 4, low, high = 0;
1345 if (!irqchip_in_kernel(vcpu->kvm) || !apic_x2apic_mode(apic))
1346 return 1;
1348 if (apic_reg_read(apic, reg, 4, &low))
1349 return 1;
1350 if (msr == 0x830)
1351 apic_reg_read(apic, APIC_ICR2, 4, &high);
1353 *data = (((u64)high) << 32) | low;
1355 return 0;
1358 int kvm_hv_vapic_msr_write(struct kvm_vcpu *vcpu, u32 reg, u64 data)
1360 struct kvm_lapic *apic = vcpu->arch.apic;
1362 if (!irqchip_in_kernel(vcpu->kvm))
1363 return 1;
1365 /* if this is ICR write vector before command */
1366 if (reg == APIC_ICR)
1367 apic_reg_write(apic, APIC_ICR2, (u32)(data >> 32));
1368 return apic_reg_write(apic, reg, (u32)data);
1371 int kvm_hv_vapic_msr_read(struct kvm_vcpu *vcpu, u32 reg, u64 *data)
1373 struct kvm_lapic *apic = vcpu->arch.apic;
1374 u32 low, high = 0;
1376 if (!irqchip_in_kernel(vcpu->kvm))
1377 return 1;
1379 if (apic_reg_read(apic, reg, 4, &low))
1380 return 1;
1381 if (reg == APIC_ICR)
1382 apic_reg_read(apic, APIC_ICR2, 4, &high);
1384 *data = (((u64)high) << 32) | low;
1386 return 0;