aboutsummaryrefslogtreecommitdiffstats
path: root/drivers/kvm
diff options
context:
space:
mode:
Diffstat (limited to 'drivers/kvm')
-rw-r--r--drivers/kvm/Makefile2
-rw-r--r--drivers/kvm/irq.c53
-rw-r--r--drivers/kvm/irq.h41
-rw-r--r--drivers/kvm/kvm.h3
-rw-r--r--drivers/kvm/kvm_main.c52
-rw-r--r--drivers/kvm/lapic.c933
-rw-r--r--drivers/kvm/svm.c6
-rw-r--r--drivers/kvm/vmx.c6
8 files changed, 1065 insertions, 31 deletions
diff --git a/drivers/kvm/Makefile b/drivers/kvm/Makefile
index 952dff38eb6c..3bf7276b032d 100644
--- a/drivers/kvm/Makefile
+++ b/drivers/kvm/Makefile
@@ -2,7 +2,7 @@
2# Makefile for Kernel-based Virtual Machine module 2# Makefile for Kernel-based Virtual Machine module
3# 3#
4 4
5kvm-objs := kvm_main.o mmu.o x86_emulate.o i8259.o irq.o 5kvm-objs := kvm_main.o mmu.o x86_emulate.o i8259.o irq.o lapic.o
6obj-$(CONFIG_KVM) += kvm.o 6obj-$(CONFIG_KVM) += kvm.o
7kvm-intel-objs = vmx.o 7kvm-intel-objs = vmx.o
8obj-$(CONFIG_KVM_INTEL) += kvm-intel.o 8obj-$(CONFIG_KVM_INTEL) += kvm-intel.o
diff --git a/drivers/kvm/irq.c b/drivers/kvm/irq.c
index b08005ca7050..0b4430a0cae0 100644
--- a/drivers/kvm/irq.c
+++ b/drivers/kvm/irq.c
@@ -30,14 +30,13 @@
30 */ 30 */
31int kvm_cpu_has_interrupt(struct kvm_vcpu *v) 31int kvm_cpu_has_interrupt(struct kvm_vcpu *v)
32{ 32{
33 struct kvm_pic *s = pic_irqchip(v->kvm); 33 struct kvm_pic *s;
34 34
35 if (s->output) /* PIC */ 35 if (kvm_apic_has_interrupt(v) == -1) { /* LAPIC */
36 return 1; 36 s = pic_irqchip(v->kvm); /* PIC */
37 /* 37 return s->output;
38 * TODO: APIC 38 }
39 */ 39 return 1;
40 return 0;
41} 40}
42EXPORT_SYMBOL_GPL(kvm_cpu_has_interrupt); 41EXPORT_SYMBOL_GPL(kvm_cpu_has_interrupt);
43 42
@@ -46,16 +45,36 @@ EXPORT_SYMBOL_GPL(kvm_cpu_has_interrupt);
46 */ 45 */
47int kvm_cpu_get_interrupt(struct kvm_vcpu *v) 46int kvm_cpu_get_interrupt(struct kvm_vcpu *v)
48{ 47{
49 struct kvm_pic *s = pic_irqchip(v->kvm); 48 struct kvm_pic *s;
50 int vector; 49 int vector;
51 50
52 s->output = 0; 51 vector = kvm_get_apic_interrupt(v); /* APIC */
53 vector = kvm_pic_read_irq(s); 52 if (vector == -1) {
54 if (vector != -1) 53 s = pic_irqchip(v->kvm);
55 return vector; 54 s->output = 0; /* PIC */
56 /* 55 vector = kvm_pic_read_irq(s);
57 * TODO: APIC 56 }
58 */ 57 return vector;
59 return -1;
60} 58}
61EXPORT_SYMBOL_GPL(kvm_cpu_get_interrupt); 59EXPORT_SYMBOL_GPL(kvm_cpu_get_interrupt);
60
61static void vcpu_kick_intr(void *info)
62{
63#ifdef DEBUG
64 struct kvm_vcpu *vcpu = (struct kvm_vcpu *)info;
65 printk(KERN_DEBUG "vcpu_kick_intr %p \n", vcpu);
66#endif
67}
68
69void kvm_vcpu_kick(struct kvm_vcpu *vcpu)
70{
71 int ipi_pcpu = vcpu->cpu;
72
73 if (vcpu->guest_mode)
74 smp_call_function_single(ipi_pcpu, vcpu_kick_intr, vcpu, 0, 0);
75}
76
77void kvm_ioapic_update_eoi(struct kvm *kvm, int vector)
78{
79 /* TODO: for kernel IOAPIC */
80}
diff --git a/drivers/kvm/irq.h b/drivers/kvm/irq.h
index bdb2fc34804b..57e23bdac530 100644
--- a/drivers/kvm/irq.h
+++ b/drivers/kvm/irq.h
@@ -26,7 +26,6 @@
26 26
27typedef void irq_request_func(void *opaque, int level); 27typedef void irq_request_func(void *opaque, int level);
28 28
29struct kvm_pic;
30struct kvm_kpic_state { 29struct kvm_kpic_state {
31 u8 last_irr; /* edge detection */ 30 u8 last_irr; /* edge detection */
32 u8 irr; /* interrupt request register */ 31 u8 irr; /* interrupt request register */
@@ -61,4 +60,44 @@ int kvm_pic_read_irq(struct kvm_pic *s);
61int kvm_cpu_get_interrupt(struct kvm_vcpu *v); 60int kvm_cpu_get_interrupt(struct kvm_vcpu *v);
62int kvm_cpu_has_interrupt(struct kvm_vcpu *v); 61int kvm_cpu_has_interrupt(struct kvm_vcpu *v);
63 62
63struct kvm_lapic {
64 unsigned long base_address;
65 struct kvm_io_device dev;
66 struct {
67 atomic_t pending;
68 s64 period; /* unit: ns */
69 u32 divide_count;
70 ktime_t last_update;
71 struct hrtimer dev;
72 } timer;
73 struct kvm_vcpu *vcpu;
74 struct page *regs_page;
75 void *regs;
76};
77
78#ifdef DEBUG
79#define ASSERT(x) \
80do { \
81 if (!(x)) { \
82 printk(KERN_EMERG "assertion failed %s: %d: %s\n", \
83 __FILE__, __LINE__, #x); \
84 BUG(); \
85 } \
86} while (0)
87#else
88#define ASSERT(x) do { } while (0)
89#endif
90
91void kvm_vcpu_kick(struct kvm_vcpu *vcpu);
92int kvm_apic_has_interrupt(struct kvm_vcpu *vcpu);
93int kvm_get_apic_interrupt(struct kvm_vcpu *vcpu);
94int kvm_create_lapic(struct kvm_vcpu *vcpu);
95void kvm_free_apic(struct kvm_lapic *apic);
96u64 kvm_lapic_get_cr8(struct kvm_vcpu *vcpu);
97void kvm_lapic_set_tpr(struct kvm_vcpu *vcpu, unsigned long cr8);
98void kvm_lapic_set_base(struct kvm_vcpu *vcpu, u64 value);
99u64 kvm_get_apic_base(struct kvm_vcpu *vcpu);
100void kvm_set_apic_base(struct kvm_vcpu *vcpu, u64 data);
101void kvm_ioapic_update_eoi(struct kvm *kvm, int vector);
102
64#endif 103#endif
diff --git a/drivers/kvm/kvm.h b/drivers/kvm/kvm.h
index e0a2f13faf8b..a5790cb21ffc 100644
--- a/drivers/kvm/kvm.h
+++ b/drivers/kvm/kvm.h
@@ -324,6 +324,7 @@ struct kvm_vcpu {
324 u64 pdptrs[4]; /* pae */ 324 u64 pdptrs[4]; /* pae */
325 u64 shadow_efer; 325 u64 shadow_efer;
326 u64 apic_base; 326 u64 apic_base;
327 struct kvm_lapic *apic; /* kernel irqchip context */
327 u64 ia32_misc_enable_msr; 328 u64 ia32_misc_enable_msr;
328 329
329 struct kvm_mmu mmu; 330 struct kvm_mmu mmu;
@@ -569,8 +570,6 @@ void set_cr3(struct kvm_vcpu *vcpu, unsigned long cr0);
569void set_cr4(struct kvm_vcpu *vcpu, unsigned long cr0); 570void set_cr4(struct kvm_vcpu *vcpu, unsigned long cr0);
570void set_cr8(struct kvm_vcpu *vcpu, unsigned long cr0); 571void set_cr8(struct kvm_vcpu *vcpu, unsigned long cr0);
571unsigned long get_cr8(struct kvm_vcpu *vcpu); 572unsigned long get_cr8(struct kvm_vcpu *vcpu);
572u64 kvm_get_apic_base(struct kvm_vcpu *vcpu);
573void kvm_set_apic_base(struct kvm_vcpu *vcpu, u64 data);
574void lmsw(struct kvm_vcpu *vcpu, unsigned long msw); 573void lmsw(struct kvm_vcpu *vcpu, unsigned long msw);
575 574
576int kvm_get_msr_common(struct kvm_vcpu *vcpu, u32 msr, u64 *pdata); 575int kvm_get_msr_common(struct kvm_vcpu *vcpu, u32 msr, u64 *pdata);
diff --git a/drivers/kvm/kvm_main.c b/drivers/kvm/kvm_main.c
index f879efbefcdf..401e3cdc4607 100644
--- a/drivers/kvm/kvm_main.c
+++ b/drivers/kvm/kvm_main.c
@@ -281,6 +281,7 @@ EXPORT_SYMBOL_GPL(kvm_vcpu_init);
281void kvm_vcpu_uninit(struct kvm_vcpu *vcpu) 281void kvm_vcpu_uninit(struct kvm_vcpu *vcpu)
282{ 282{
283 kvm_mmu_destroy(vcpu); 283 kvm_mmu_destroy(vcpu);
284 kvm_free_apic(vcpu->apic);
284 free_page((unsigned long)vcpu->pio_data); 285 free_page((unsigned long)vcpu->pio_data);
285 free_page((unsigned long)vcpu->run); 286 free_page((unsigned long)vcpu->run);
286} 287}
@@ -598,25 +599,38 @@ void set_cr8(struct kvm_vcpu *vcpu, unsigned long cr8)
598 inject_gp(vcpu); 599 inject_gp(vcpu);
599 return; 600 return;
600 } 601 }
601 vcpu->cr8 = cr8; 602 if (irqchip_in_kernel(vcpu->kvm))
603 kvm_lapic_set_tpr(vcpu, cr8);
604 else
605 vcpu->cr8 = cr8;
602} 606}
603EXPORT_SYMBOL_GPL(set_cr8); 607EXPORT_SYMBOL_GPL(set_cr8);
604 608
605unsigned long get_cr8(struct kvm_vcpu *vcpu) 609unsigned long get_cr8(struct kvm_vcpu *vcpu)
606{ 610{
607 return vcpu->cr8; 611 if (irqchip_in_kernel(vcpu->kvm))
612 return kvm_lapic_get_cr8(vcpu);
613 else
614 return vcpu->cr8;
608} 615}
609EXPORT_SYMBOL_GPL(get_cr8); 616EXPORT_SYMBOL_GPL(get_cr8);
610 617
611u64 kvm_get_apic_base(struct kvm_vcpu *vcpu) 618u64 kvm_get_apic_base(struct kvm_vcpu *vcpu)
612{ 619{
613 return vcpu->apic_base; 620 if (irqchip_in_kernel(vcpu->kvm))
621 return vcpu->apic_base;
622 else
623 return vcpu->apic_base;
614} 624}
615EXPORT_SYMBOL_GPL(kvm_get_apic_base); 625EXPORT_SYMBOL_GPL(kvm_get_apic_base);
616 626
617void kvm_set_apic_base(struct kvm_vcpu *vcpu, u64 data) 627void kvm_set_apic_base(struct kvm_vcpu *vcpu, u64 data)
618{ 628{
619 vcpu->apic_base = data; 629 /* TODO: reserve bits check */
630 if (irqchip_in_kernel(vcpu->kvm))
631 kvm_lapic_set_base(vcpu, data);
632 else
633 vcpu->apic_base = data;
620} 634}
621EXPORT_SYMBOL_GPL(kvm_set_apic_base); 635EXPORT_SYMBOL_GPL(kvm_set_apic_base);
622 636
@@ -986,15 +1000,31 @@ static int emulator_write_std(unsigned long addr,
986 return X86EMUL_UNHANDLEABLE; 1000 return X86EMUL_UNHANDLEABLE;
987} 1001}
988 1002
1003/*
1004 * Only apic need an MMIO device hook, so shortcut now..
1005 */
1006static struct kvm_io_device *vcpu_find_pervcpu_dev(struct kvm_vcpu *vcpu,
1007 gpa_t addr)
1008{
1009 struct kvm_io_device *dev;
1010
1011 if (vcpu->apic) {
1012 dev = &vcpu->apic->dev;
1013 if (dev->in_range(dev, addr))
1014 return dev;
1015 }
1016 return NULL;
1017}
1018
989static struct kvm_io_device *vcpu_find_mmio_dev(struct kvm_vcpu *vcpu, 1019static struct kvm_io_device *vcpu_find_mmio_dev(struct kvm_vcpu *vcpu,
990 gpa_t addr) 1020 gpa_t addr)
991{ 1021{
992 /* 1022 struct kvm_io_device *dev;
993 * Note that its important to have this wrapper function because 1023
994 * in the very near future we will be checking for MMIOs against 1024 dev = vcpu_find_pervcpu_dev(vcpu, addr);
995 * the LAPIC as well as the general MMIO bus 1025 if (dev == NULL)
996 */ 1026 dev = kvm_io_bus_find_dev(&vcpu->kvm->mmio_bus, addr);
997 return kvm_io_bus_find_dev(&vcpu->kvm->mmio_bus, addr); 1027 return dev;
998} 1028}
999 1029
1000static struct kvm_io_device *vcpu_find_pio_dev(struct kvm_vcpu *vcpu, 1030static struct kvm_io_device *vcpu_find_pio_dev(struct kvm_vcpu *vcpu,
@@ -2256,6 +2286,8 @@ static int kvm_vcpu_ioctl_interrupt(struct kvm_vcpu *vcpu,
2256{ 2286{
2257 if (irq->irq < 0 || irq->irq >= 256) 2287 if (irq->irq < 0 || irq->irq >= 256)
2258 return -EINVAL; 2288 return -EINVAL;
2289 if (irqchip_in_kernel(vcpu->kvm))
2290 return -ENXIO;
2259 vcpu_load(vcpu); 2291 vcpu_load(vcpu);
2260 2292
2261 set_bit(irq->irq, vcpu->irq_pending); 2293 set_bit(irq->irq, vcpu->irq_pending);
diff --git a/drivers/kvm/lapic.c b/drivers/kvm/lapic.c
new file mode 100644
index 000000000000..4b5c77d8900d
--- /dev/null
+++ b/drivers/kvm/lapic.c
@@ -0,0 +1,933 @@
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 "kvm.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 <asm/processor.h>
29#include <asm/msr.h>
30#include <asm/page.h>
31#include <asm/current.h>
32#include <asm/apicdef.h>
33#include <asm/atomic.h>
34#include <asm/div64.h>
35#include "irq.h"
36
37#define PRId64 "d"
38#define PRIx64 "llx"
39#define PRIu64 "u"
40#define PRIo64 "o"
41
42#define APIC_BUS_CYCLE_NS 1
43
44/* #define apic_debug(fmt,arg...) printk(KERN_WARNING fmt,##arg) */
45#define apic_debug(fmt, arg...)
46
47#define APIC_LVT_NUM 6
48/* 14 is the version for Xeon and Pentium 8.4.8*/
49#define APIC_VERSION (0x14UL | ((APIC_LVT_NUM - 1) << 16))
50#define LAPIC_MMIO_LENGTH (1 << 12)
51/* followed define is not in apicdef.h */
52#define APIC_SHORT_MASK 0xc0000
53#define APIC_DEST_NOSHORT 0x0
54#define APIC_DEST_MASK 0x800
55#define MAX_APIC_VECTOR 256
56
57#define VEC_POS(v) ((v) & (32 - 1))
58#define REG_POS(v) (((v) >> 5) << 4)
59static inline u32 apic_get_reg(struct kvm_lapic *apic, int reg_off)
60{
61 return *((u32 *) (apic->regs + reg_off));
62}
63
64static inline void apic_set_reg(struct kvm_lapic *apic, int reg_off, u32 val)
65{
66 *((u32 *) (apic->regs + reg_off)) = val;
67}
68
69static inline int apic_test_and_set_vector(int vec, void *bitmap)
70{
71 return test_and_set_bit(VEC_POS(vec), (bitmap) + REG_POS(vec));
72}
73
74static inline int apic_test_and_clear_vector(int vec, void *bitmap)
75{
76 return test_and_clear_bit(VEC_POS(vec), (bitmap) + REG_POS(vec));
77}
78
79static inline void apic_set_vector(int vec, void *bitmap)
80{
81 set_bit(VEC_POS(vec), (bitmap) + REG_POS(vec));
82}
83
84static inline void apic_clear_vector(int vec, void *bitmap)
85{
86 clear_bit(VEC_POS(vec), (bitmap) + REG_POS(vec));
87}
88
89static inline int apic_hw_enabled(struct kvm_lapic *apic)
90{
91 return (apic)->vcpu->apic_base & MSR_IA32_APICBASE_ENABLE;
92}
93
94static inline int apic_sw_enabled(struct kvm_lapic *apic)
95{
96 return apic_get_reg(apic, APIC_SPIV) & APIC_SPIV_APIC_ENABLED;
97}
98
99static inline int apic_enabled(struct kvm_lapic *apic)
100{
101 return apic_sw_enabled(apic) && apic_hw_enabled(apic);
102}
103
104#define LVT_MASK \
105 (APIC_LVT_MASKED | APIC_SEND_PENDING | APIC_VECTOR_MASK)
106
107#define LINT_MASK \
108 (LVT_MASK | APIC_MODE_MASK | APIC_INPUT_POLARITY | \
109 APIC_LVT_REMOTE_IRR | APIC_LVT_LEVEL_TRIGGER)
110
111static inline int kvm_apic_id(struct kvm_lapic *apic)
112{
113 return (apic_get_reg(apic, APIC_ID) >> 24) & 0xff;
114}
115
116static inline int apic_lvt_enabled(struct kvm_lapic *apic, int lvt_type)
117{
118 return !(apic_get_reg(apic, lvt_type) & APIC_LVT_MASKED);
119}
120
121static inline int apic_lvt_vector(struct kvm_lapic *apic, int lvt_type)
122{
123 return apic_get_reg(apic, lvt_type) & APIC_VECTOR_MASK;
124}
125
126static inline int apic_lvtt_period(struct kvm_lapic *apic)
127{
128 return apic_get_reg(apic, APIC_LVTT) & APIC_LVT_TIMER_PERIODIC;
129}
130
131static unsigned int apic_lvt_mask[APIC_LVT_NUM] = {
132 LVT_MASK | APIC_LVT_TIMER_PERIODIC, /* LVTT */
133 LVT_MASK | APIC_MODE_MASK, /* LVTTHMR */
134 LVT_MASK | APIC_MODE_MASK, /* LVTPC */
135 LINT_MASK, LINT_MASK, /* LVT0-1 */
136 LVT_MASK /* LVTERR */
137};
138
139static int find_highest_vector(void *bitmap)
140{
141 u32 *word = bitmap;
142 int word_offset = MAX_APIC_VECTOR >> 5;
143
144 while ((word_offset != 0) && (word[(--word_offset) << 2] == 0))
145 continue;
146
147 if (likely(!word_offset && !word[0]))
148 return -1;
149 else
150 return fls(word[word_offset << 2]) - 1 + (word_offset << 5);
151}
152
153static inline int apic_test_and_set_irr(int vec, struct kvm_lapic *apic)
154{
155 return apic_test_and_set_vector(vec, apic->regs + APIC_IRR);
156}
157
158static inline void apic_clear_irr(int vec, struct kvm_lapic *apic)
159{
160 apic_clear_vector(vec, apic->regs + APIC_IRR);
161}
162
163static inline int apic_find_highest_irr(struct kvm_lapic *apic)
164{
165 int result;
166
167 result = find_highest_vector(apic->regs + APIC_IRR);
168 ASSERT(result == -1 || result >= 16);
169
170 return result;
171}
172
173int kvm_apic_set_irq(struct kvm_lapic *apic, u8 vec, u8 trig)
174{
175 if (!apic_test_and_set_irr(vec, apic)) {
176 /* a new pending irq is set in IRR */
177 if (trig)
178 apic_set_vector(vec, apic->regs + APIC_TMR);
179 else
180 apic_clear_vector(vec, apic->regs + APIC_TMR);
181 kvm_vcpu_kick(apic->vcpu);
182 return 1;
183 }
184 return 0;
185}
186
187static inline int apic_find_highest_isr(struct kvm_lapic *apic)
188{
189 int result;
190
191 result = find_highest_vector(apic->regs + APIC_ISR);
192 ASSERT(result == -1 || result >= 16);
193
194 return result;
195}
196
197static void apic_update_ppr(struct kvm_lapic *apic)
198{
199 u32 tpr, isrv, ppr;
200 int isr;
201
202 tpr = apic_get_reg(apic, APIC_TASKPRI);
203 isr = apic_find_highest_isr(apic);
204 isrv = (isr != -1) ? isr : 0;
205
206 if ((tpr & 0xf0) >= (isrv & 0xf0))
207 ppr = tpr & 0xff;
208 else
209 ppr = isrv & 0xf0;
210
211 apic_debug("vlapic %p, ppr 0x%x, isr 0x%x, isrv 0x%x",
212 apic, ppr, isr, isrv);
213
214 apic_set_reg(apic, APIC_PROCPRI, ppr);
215}
216
217static void apic_set_tpr(struct kvm_lapic *apic, u32 tpr)
218{
219 apic_set_reg(apic, APIC_TASKPRI, tpr);
220 apic_update_ppr(apic);
221}
222
223int kvm_apic_match_physical_addr(struct kvm_lapic *apic, u16 dest)
224{
225 return kvm_apic_id(apic) == dest;
226}
227
228int kvm_apic_match_logical_addr(struct kvm_lapic *apic, u8 mda)
229{
230 int result = 0;
231 u8 logical_id;
232
233 logical_id = GET_APIC_LOGICAL_ID(apic_get_reg(apic, APIC_LDR));
234
235 switch (apic_get_reg(apic, APIC_DFR)) {
236 case APIC_DFR_FLAT:
237 if (logical_id & mda)
238 result = 1;
239 break;
240 case APIC_DFR_CLUSTER:
241 if (((logical_id >> 4) == (mda >> 0x4))
242 && (logical_id & mda & 0xf))
243 result = 1;
244 break;
245 default:
246 printk(KERN_WARNING "Bad DFR vcpu %d: %08x\n",
247 apic->vcpu->vcpu_id, apic_get_reg(apic, APIC_DFR));
248 break;
249 }
250
251 return result;
252}
253
254static int apic_match_dest(struct kvm_vcpu *vcpu, struct kvm_lapic *source,
255 int short_hand, int dest, int dest_mode)
256{
257 int result = 0;
258 struct kvm_lapic *target = vcpu->apic;
259
260 apic_debug("target %p, source %p, dest 0x%x, "
261 "dest_mode 0x%x, short_hand 0x%x",
262 target, source, dest, dest_mode, short_hand);
263
264 ASSERT(!target);
265 switch (short_hand) {
266 case APIC_DEST_NOSHORT:
267 if (dest_mode == 0) {
268 /* Physical mode. */
269 if ((dest == 0xFF) || (dest == kvm_apic_id(target)))
270 result = 1;
271 } else
272 /* Logical mode. */
273 result = kvm_apic_match_logical_addr(target, dest);
274 break;
275 case APIC_DEST_SELF:
276 if (target == source)
277 result = 1;
278 break;
279 case APIC_DEST_ALLINC:
280 result = 1;
281 break;
282 case APIC_DEST_ALLBUT:
283 if (target != source)
284 result = 1;
285 break;
286 default:
287 printk(KERN_WARNING "Bad dest shorthand value %x\n",
288 short_hand);
289 break;
290 }
291
292 return result;
293}
294
295/*
296 * Add a pending IRQ into lapic.
297 * Return 1 if successfully added and 0 if discarded.
298 */
299static int __apic_accept_irq(struct kvm_lapic *apic, int delivery_mode,
300 int vector, int level, int trig_mode)
301{
302 int result = 0;
303
304 switch (delivery_mode) {
305 case APIC_DM_FIXED:
306 case APIC_DM_LOWEST:
307 /* FIXME add logic for vcpu on reset */
308 if (unlikely(!apic_enabled(apic)))
309 break;
310
311 if (apic_test_and_set_irr(vector, apic) && trig_mode) {
312 apic_debug("level trig mode repeatedly for vector %d",
313 vector);
314 break;
315 }
316
317 if (trig_mode) {
318 apic_debug("level trig mode for vector %d", vector);
319 apic_set_vector(vector, apic->regs + APIC_TMR);
320 } else
321 apic_clear_vector(vector, apic->regs + APIC_TMR);
322
323 kvm_vcpu_kick(apic->vcpu);
324
325 result = 1;
326 break;
327
328 case APIC_DM_REMRD:
329 printk(KERN_DEBUG "Ignoring delivery mode 3\n");
330 break;
331
332 case APIC_DM_SMI:
333 printk(KERN_DEBUG "Ignoring guest SMI\n");
334 break;
335 case APIC_DM_NMI:
336 printk(KERN_DEBUG "Ignoring guest NMI\n");
337 break;
338
339 case APIC_DM_INIT:
340 printk(KERN_DEBUG "Ignoring guest INIT\n");
341 break;
342
343 case APIC_DM_STARTUP:
344 printk(KERN_DEBUG "Ignoring guest STARTUP\n");
345 break;
346
347 default:
348 printk(KERN_ERR "TODO: unsupported delivery mode %x\n",
349 delivery_mode);
350 break;
351 }
352 return result;
353}
354
355struct kvm_lapic *kvm_apic_round_robin(struct kvm *kvm, u8 vector,
356 unsigned long bitmap)
357{
358 int vcpu_id;
359
360 /* TODO for real round robin */
361 vcpu_id = fls(bitmap) - 1;
362 if (vcpu_id < 0)
363 printk(KERN_DEBUG "vcpu not ready for apic_round_robin\n");
364 return kvm->vcpus[vcpu_id]->apic;
365}
366
367static void apic_set_eoi(struct kvm_lapic *apic)
368{
369 int vector = apic_find_highest_isr(apic);
370
371 /*
372 * Not every write EOI will has corresponding ISR,
373 * one example is when Kernel check timer on setup_IO_APIC
374 */
375 if (vector == -1)
376 return;
377
378 apic_clear_vector(vector, apic->regs + APIC_ISR);
379 apic_update_ppr(apic);
380
381 if (apic_test_and_clear_vector(vector, apic->regs + APIC_TMR))
382 kvm_ioapic_update_eoi(apic->vcpu->kvm, vector);
383}
384
385static void apic_send_ipi(struct kvm_lapic *apic)
386{
387 u32 icr_low = apic_get_reg(apic, APIC_ICR);
388 u32 icr_high = apic_get_reg(apic, APIC_ICR2);
389
390 unsigned int dest = GET_APIC_DEST_FIELD(icr_high);
391 unsigned int short_hand = icr_low & APIC_SHORT_MASK;
392 unsigned int trig_mode = icr_low & APIC_INT_LEVELTRIG;
393 unsigned int level = icr_low & APIC_INT_ASSERT;
394 unsigned int dest_mode = icr_low & APIC_DEST_MASK;
395 unsigned int delivery_mode = icr_low & APIC_MODE_MASK;
396 unsigned int vector = icr_low & APIC_VECTOR_MASK;
397
398 struct kvm_lapic *target;
399 struct kvm_vcpu *vcpu;
400 unsigned long lpr_map = 0;
401 int i;
402
403 apic_debug("icr_high 0x%x, icr_low 0x%x, "
404 "short_hand 0x%x, dest 0x%x, trig_mode 0x%x, level 0x%x, "
405 "dest_mode 0x%x, delivery_mode 0x%x, vector 0x%x\n",
406 icr_high, icr_low, short_hand, dest,
407 trig_mode, level, dest_mode, delivery_mode, vector);
408
409 for (i = 0; i < KVM_MAX_VCPUS; i++) {
410 vcpu = apic->vcpu->kvm->vcpus[i];
411 if (!vcpu)
412 continue;
413
414 if (vcpu->apic &&
415 apic_match_dest(vcpu, apic, short_hand, dest, dest_mode)) {
416 if (delivery_mode == APIC_DM_LOWEST)
417 set_bit(vcpu->vcpu_id, &lpr_map);
418 else
419 __apic_accept_irq(vcpu->apic, delivery_mode,
420 vector, level, trig_mode);
421 }
422 }
423
424 if (delivery_mode == APIC_DM_LOWEST) {
425 target = kvm_apic_round_robin(vcpu->kvm, vector, lpr_map);
426 if (target != NULL)
427 __apic_accept_irq(target, delivery_mode,
428 vector, level, trig_mode);
429 }
430}
431
432static u32 apic_get_tmcct(struct kvm_lapic *apic)
433{
434 u32 counter_passed;
435 ktime_t passed, now = apic->timer.dev.base->get_time();
436 u32 tmcct = apic_get_reg(apic, APIC_TMICT);
437
438 ASSERT(apic != NULL);
439
440 if (unlikely(ktime_to_ns(now) <=
441 ktime_to_ns(apic->timer.last_update))) {
442 /* Wrap around */
443 passed = ktime_add(( {
444 (ktime_t) {
445 .tv64 = KTIME_MAX -
446 (apic->timer.last_update).tv64}; }
447 ), now);
448 apic_debug("time elapsed\n");
449 } else
450 passed = ktime_sub(now, apic->timer.last_update);
451
452 counter_passed = div64_64(ktime_to_ns(passed),
453 (APIC_BUS_CYCLE_NS * apic->timer.divide_count));
454 tmcct -= counter_passed;
455
456 if (tmcct <= 0) {
457 if (unlikely(!apic_lvtt_period(apic)))
458 tmcct = 0;
459 else
460 do {
461 tmcct += apic_get_reg(apic, APIC_TMICT);
462 } while (tmcct <= 0);
463 }
464
465 return tmcct;
466}
467
468static u32 __apic_read(struct kvm_lapic *apic, unsigned int offset)
469{
470 u32 val = 0;
471
472 if (offset >= LAPIC_MMIO_LENGTH)
473 return 0;
474
475 switch (offset) {
476 case APIC_ARBPRI:
477 printk(KERN_WARNING "Access APIC ARBPRI register "
478 "which is for P6\n");
479 break;
480
481 case APIC_TMCCT: /* Timer CCR */
482 val = apic_get_tmcct(apic);
483 break;
484
485 default:
486 val = apic_get_reg(apic, offset);
487 break;
488 }
489
490 return val;
491}
492
493static void apic_mmio_read(struct kvm_io_device *this,
494 gpa_t address, int len, void *data)
495{
496 struct kvm_lapic *apic = (struct kvm_lapic *)this->private;
497 unsigned int offset = address - apic->base_address;
498 unsigned char alignment = offset & 0xf;
499 u32 result;
500
501 if ((alignment + len) > 4) {
502 printk(KERN_ERR "KVM_APIC_READ: alignment error %lx %d",
503 (unsigned long)address, len);
504 return;
505 }
506 result = __apic_read(apic, offset & ~0xf);
507
508 switch (len) {
509 case 1:
510 case 2:
511 case 4:
512 memcpy(data, (char *)&result + alignment, len);
513 break;
514 default:
515 printk(KERN_ERR "Local APIC read with len = %x, "
516 "should be 1,2, or 4 instead\n", len);
517 break;
518 }
519}
520
521static void update_divide_count(struct kvm_lapic *apic)
522{
523 u32 tmp1, tmp2, tdcr;
524
525 tdcr = apic_get_reg(apic, APIC_TDCR);
526 tmp1 = tdcr & 0xf;
527 tmp2 = ((tmp1 & 0x3) | ((tmp1 & 0x8) >> 1)) + 1;
528 apic->timer.divide_count = 0x1 << (tmp2 & 0x7);
529
530 apic_debug("timer divide count is 0x%x\n",
531 apic->timer.divide_count);
532}
533
534static void start_apic_timer(struct kvm_lapic *apic)
535{
536 ktime_t now = apic->timer.dev.base->get_time();
537
538 apic->timer.last_update = now;
539
540 apic->timer.period = apic_get_reg(apic, APIC_TMICT) *
541 APIC_BUS_CYCLE_NS * apic->timer.divide_count;
542 atomic_set(&apic->timer.pending, 0);
543 hrtimer_start(&apic->timer.dev,
544 ktime_add_ns(now, apic->timer.period),
545 HRTIMER_MODE_ABS);
546
547 apic_debug("%s: bus cycle is %" PRId64 "ns, now 0x%016"
548 PRIx64 ", "
549 "timer initial count 0x%x, period %lldns, "
550 "expire @ 0x%016" PRIx64 ".\n", __FUNCTION__,
551 APIC_BUS_CYCLE_NS, ktime_to_ns(now),
552 apic_get_reg(apic, APIC_TMICT),
553 apic->timer.period,
554 ktime_to_ns(ktime_add_ns(now,
555 apic->timer.period)));
556}
557
558static void apic_mmio_write(struct kvm_io_device *this,
559 gpa_t address, int len, const void *data)
560{
561 struct kvm_lapic *apic = (struct kvm_lapic *)this->private;
562 unsigned int offset = address - apic->base_address;
563 unsigned char alignment = offset & 0xf;
564 u32 val;
565
566 /*
567 * APIC register must be aligned on 128-bits boundary.
568 * 32/64/128 bits registers must be accessed thru 32 bits.
569 * Refer SDM 8.4.1
570 */
571 if (len != 4 || alignment) {
572 if (printk_ratelimit())
573 printk(KERN_ERR "apic write: bad size=%d %lx\n",
574 len, (long)address);
575 return;
576 }
577
578 val = *(u32 *) data;
579
580 /* too common printing */
581 if (offset != APIC_EOI)
582 apic_debug("%s: offset 0x%x with length 0x%x, and value is "
583 "0x%x\n", __FUNCTION__, offset, len, val);
584
585 offset &= 0xff0;
586
587 switch (offset) {
588 case APIC_ID: /* Local APIC ID */
589 apic_set_reg(apic, APIC_ID, val);
590 break;
591
592 case APIC_TASKPRI:
593 apic_set_tpr(apic, val & 0xff);
594 break;
595
596 case APIC_EOI:
597 apic_set_eoi(apic);
598 break;
599
600 case APIC_LDR:
601 apic_set_reg(apic, APIC_LDR, val & APIC_LDR_MASK);
602 break;
603
604 case APIC_DFR:
605 apic_set_reg(apic, APIC_DFR, val | 0x0FFFFFFF);
606 break;
607
608 case APIC_SPIV:
609 apic_set_reg(apic, APIC_SPIV, val & 0x3ff);
610 if (!(val & APIC_SPIV_APIC_ENABLED)) {
611 int i;
612 u32 lvt_val;
613
614 for (i = 0; i < APIC_LVT_NUM; i++) {
615 lvt_val = apic_get_reg(apic,
616 APIC_LVTT + 0x10 * i);
617 apic_set_reg(apic, APIC_LVTT + 0x10 * i,
618 lvt_val | APIC_LVT_MASKED);
619 }
620 atomic_set(&apic->timer.pending, 0);
621
622 }
623 break;
624
625 case APIC_ICR:
626 /* No delay here, so we always clear the pending bit */
627 apic_set_reg(apic, APIC_ICR, val & ~(1 << 12));
628 apic_send_ipi(apic);
629 break;
630
631 case APIC_ICR2:
632 apic_set_reg(apic, APIC_ICR2, val & 0xff000000);
633 break;
634
635 case APIC_LVTT:
636 case APIC_LVTTHMR:
637 case APIC_LVTPC:
638 case APIC_LVT0:
639 case APIC_LVT1:
640 case APIC_LVTERR:
641 /* TODO: Check vector */
642 if (!apic_sw_enabled(apic))
643 val |= APIC_LVT_MASKED;
644
645 val &= apic_lvt_mask[(offset - APIC_LVTT) >> 4];
646 apic_set_reg(apic, offset, val);
647
648 break;
649
650 case APIC_TMICT:
651 hrtimer_cancel(&apic->timer.dev);
652 apic_set_reg(apic, APIC_TMICT, val);
653 start_apic_timer(apic);
654 return;
655
656 case APIC_TDCR:
657 if (val & 4)
658 printk(KERN_ERR "KVM_WRITE:TDCR %x\n", val);
659 apic_set_reg(apic, APIC_TDCR, val);
660 update_divide_count(apic);
661 break;
662
663 default:
664 apic_debug("Local APIC Write to read-only register %x\n",
665 offset);
666 break;
667 }
668
669}
670
671static int apic_mmio_range(struct kvm_io_device *this, gpa_t addr)
672{
673 struct kvm_lapic *apic = (struct kvm_lapic *)this->private;
674 int ret = 0;
675
676
677 if (apic_hw_enabled(apic) &&
678 (addr >= apic->base_address) &&
679 (addr < (apic->base_address + LAPIC_MMIO_LENGTH)))
680 ret = 1;
681
682 return ret;
683}
684
685void kvm_free_apic(struct kvm_lapic *apic)
686{
687 if (!apic)
688 return;
689
690 hrtimer_cancel(&apic->timer.dev);
691
692 if (apic->regs_page) {
693 __free_page(apic->regs_page);
694 apic->regs_page = 0;
695 }
696
697 kfree(apic);
698}
699
700/*
701 *----------------------------------------------------------------------
702 * LAPIC interface
703 *----------------------------------------------------------------------
704 */
705
706void kvm_lapic_set_tpr(struct kvm_vcpu *vcpu, unsigned long cr8)
707{
708 struct kvm_lapic *apic = (struct kvm_lapic *)vcpu->apic;
709
710 if (!apic)
711 return;
712 apic_set_tpr(apic, ((cr8 & 0x0f) << 4));
713}
714
715u64 kvm_lapic_get_cr8(struct kvm_vcpu *vcpu)
716{
717 struct kvm_lapic *apic = (struct kvm_lapic *)vcpu->apic;
718 u64 tpr;
719
720 if (!apic)
721 return 0;
722 tpr = (u64) apic_get_reg(apic, APIC_TASKPRI);
723
724 return (tpr & 0xf0) >> 4;
725}
726
727void kvm_lapic_set_base(struct kvm_vcpu *vcpu, u64 value)
728{
729 struct kvm_lapic *apic = (struct kvm_lapic *)vcpu->apic;
730
731 if (!apic) {
732 value |= MSR_IA32_APICBASE_BSP;
733 vcpu->apic_base = value;
734 return;
735 }
736 if (apic->vcpu->vcpu_id)
737 value &= ~MSR_IA32_APICBASE_BSP;
738
739 vcpu->apic_base = value;
740 apic->base_address = apic->vcpu->apic_base &
741 MSR_IA32_APICBASE_BASE;
742
743 /* with FSB delivery interrupt, we can restart APIC functionality */
744 apic_debug("apic base msr is 0x%016" PRIx64 ", and base address is "
745 "0x%lx.\n", apic->apic_base, apic->base_address);
746
747}
748
749u64 kvm_lapic_get_base(struct kvm_vcpu *vcpu)
750{
751 return vcpu->apic_base;
752}
753EXPORT_SYMBOL_GPL(kvm_lapic_get_base);
754
755static void lapic_reset(struct kvm_vcpu *vcpu)
756{
757 struct kvm_lapic *apic;
758 int i;
759
760 apic_debug("%s\n", __FUNCTION__);
761
762 ASSERT(vcpu);
763 apic = vcpu->apic;
764 ASSERT(apic != NULL);
765
766 /* Stop the timer in case it's a reset to an active apic */
767 hrtimer_cancel(&apic->timer.dev);
768
769 apic_set_reg(apic, APIC_ID, vcpu->vcpu_id << 24);
770 apic_set_reg(apic, APIC_LVR, APIC_VERSION);
771
772 for (i = 0; i < APIC_LVT_NUM; i++)
773 apic_set_reg(apic, APIC_LVTT + 0x10 * i, APIC_LVT_MASKED);
774
775 apic_set_reg(apic, APIC_DFR, 0xffffffffU);
776 apic_set_reg(apic, APIC_SPIV, 0xff);
777 apic_set_reg(apic, APIC_TASKPRI, 0);
778 apic_set_reg(apic, APIC_LDR, 0);
779 apic_set_reg(apic, APIC_ESR, 0);
780 apic_set_reg(apic, APIC_ICR, 0);
781 apic_set_reg(apic, APIC_ICR2, 0);
782 apic_set_reg(apic, APIC_TDCR, 0);
783 apic_set_reg(apic, APIC_TMICT, 0);
784 for (i = 0; i < 8; i++) {
785 apic_set_reg(apic, APIC_IRR + 0x10 * i, 0);
786 apic_set_reg(apic, APIC_ISR + 0x10 * i, 0);
787 apic_set_reg(apic, APIC_TMR + 0x10 * i, 0);
788 }
789 apic->timer.divide_count = 0;
790 atomic_set(&apic->timer.pending, 0);
791 if (vcpu->vcpu_id == 0)
792 vcpu->apic_base |= MSR_IA32_APICBASE_BSP;
793 apic_update_ppr(apic);
794
795 apic_debug(KERN_INFO "%s: vcpu=%p, id=%d, base_msr="
796 "0x%016" PRIx64 ", base_address=0x%0lx.\n", __FUNCTION__,
797 vcpu, kvm_apic_id(apic),
798 vcpu->apic_base, apic->base_address);
799}
800
801int kvm_lapic_enabled(struct kvm_vcpu *vcpu)
802{
803 struct kvm_lapic *apic = (struct kvm_lapic *)vcpu->apic;
804 int ret = 0;
805
806 if (!apic)
807 return 0;
808 ret = apic_enabled(apic);
809
810 return ret;
811}
812
813/*
814 *----------------------------------------------------------------------
815 * timer interface
816 *----------------------------------------------------------------------
817 */
818static int __apic_timer_fn(struct kvm_lapic *apic)
819{
820 u32 vector;
821 int result = 0;
822
823 if (unlikely(!apic_enabled(apic) ||
824 !apic_lvt_enabled(apic, APIC_LVTT))) {
825 apic_debug("%s: time interrupt although apic is down\n",
826 __FUNCTION__);
827 return 0;
828 }
829
830 vector = apic_lvt_vector(apic, APIC_LVTT);
831 apic->timer.last_update = apic->timer.dev.expires;
832 atomic_inc(&apic->timer.pending);
833 __apic_accept_irq(apic, APIC_DM_FIXED, vector, 1, 0);
834
835 if (apic_lvtt_period(apic)) {
836 u32 offset;
837 u32 tmict = apic_get_reg(apic, APIC_TMICT);
838
839 offset = APIC_BUS_CYCLE_NS * apic->timer.divide_count * tmict;
840
841 result = 1;
842 apic->timer.dev.expires = ktime_add_ns(
843 apic->timer.dev.expires,
844 apic->timer.period);
845 }
846
847 return result;
848}
849
850static enum hrtimer_restart apic_timer_fn(struct hrtimer *data)
851{
852 struct kvm_lapic *apic;
853 int restart_timer = 0;
854
855 apic = container_of(data, struct kvm_lapic, timer.dev);
856
857 restart_timer = __apic_timer_fn(apic);
858
859 if (restart_timer)
860 return HRTIMER_RESTART;
861 else
862 return HRTIMER_NORESTART;
863}
864
865int kvm_create_lapic(struct kvm_vcpu *vcpu)
866{
867 struct kvm_lapic *apic;
868
869 ASSERT(vcpu != NULL);
870 apic_debug("apic_init %d\n", vcpu->vcpu_id);
871
872 apic = kzalloc(sizeof(*apic), GFP_KERNEL);
873 if (!apic)
874 goto nomem;
875
876 vcpu->apic = apic;
877
878 apic->regs_page = alloc_page(GFP_KERNEL);
879 if (apic->regs_page == NULL) {
880 printk(KERN_ERR "malloc apic regs error for vcpu %x\n",
881 vcpu->vcpu_id);
882 goto nomem;
883 }
884 apic->regs = page_address(apic->regs_page);
885 memset(apic->regs, 0, PAGE_SIZE);
886 apic->vcpu = vcpu;
887
888 hrtimer_init(&apic->timer.dev, CLOCK_MONOTONIC, HRTIMER_MODE_ABS);
889 apic->timer.dev.function = apic_timer_fn;
890 apic->base_address = APIC_DEFAULT_PHYS_BASE;
891 vcpu->apic_base = APIC_DEFAULT_PHYS_BASE;
892
893 lapic_reset(vcpu);
894 apic->dev.read = apic_mmio_read;
895 apic->dev.write = apic_mmio_write;
896 apic->dev.in_range = apic_mmio_range;
897 apic->dev.private = apic;
898
899 return 0;
900nomem:
901 kvm_free_apic(apic);
902 return -ENOMEM;
903}
904EXPORT_SYMBOL_GPL(kvm_create_lapic);
905
906int kvm_apic_has_interrupt(struct kvm_vcpu *vcpu)
907{
908 struct kvm_lapic *apic = vcpu->apic;
909 int highest_irr;
910
911 if (!apic || !apic_enabled(apic))
912 return -1;
913
914 highest_irr = apic_find_highest_irr(apic);
915 if ((highest_irr == -1) ||
916 ((highest_irr & 0xF0) <= apic_get_reg(apic, APIC_PROCPRI)))
917 return -1;
918 return highest_irr;
919}
920
921int kvm_get_apic_interrupt(struct kvm_vcpu *vcpu)
922{
923 int vector = kvm_apic_has_interrupt(vcpu);
924 struct kvm_lapic *apic = vcpu->apic;
925
926 if (vector == -1)
927 return -1;
928
929 apic_set_vector(vector, apic->regs + APIC_ISR);
930 apic_update_ppr(apic);
931 apic_clear_irr(vector, apic);
932 return vector;
933}
diff --git a/drivers/kvm/svm.c b/drivers/kvm/svm.c
index 57525e7ed28f..d576451827e7 100644
--- a/drivers/kvm/svm.c
+++ b/drivers/kvm/svm.c
@@ -573,6 +573,12 @@ static struct kvm_vcpu *svm_create_vcpu(struct kvm *kvm, unsigned int id)
573 if (err) 573 if (err)
574 goto free_svm; 574 goto free_svm;
575 575
576 if (irqchip_in_kernel(kvm)) {
577 err = kvm_create_lapic(&svm->vcpu);
578 if (err < 0)
579 goto free_svm;
580 }
581
576 page = alloc_page(GFP_KERNEL); 582 page = alloc_page(GFP_KERNEL);
577 if (!page) { 583 if (!page) {
578 err = -ENOMEM; 584 err = -ENOMEM;
diff --git a/drivers/kvm/vmx.c b/drivers/kvm/vmx.c
index 19676b5a6719..c4cc17cc00f7 100644
--- a/drivers/kvm/vmx.c
+++ b/drivers/kvm/vmx.c
@@ -2390,6 +2390,12 @@ static struct kvm_vcpu *vmx_create_vcpu(struct kvm *kvm, unsigned int id)
2390 if (err) 2390 if (err)
2391 goto free_vcpu; 2391 goto free_vcpu;
2392 2392
2393 if (irqchip_in_kernel(kvm)) {
2394 err = kvm_create_lapic(&vmx->vcpu);
2395 if (err < 0)
2396 goto free_vcpu;
2397 }
2398
2393 vmx->guest_msrs = kmalloc(PAGE_SIZE, GFP_KERNEL); 2399 vmx->guest_msrs = kmalloc(PAGE_SIZE, GFP_KERNEL);
2394 if (!vmx->guest_msrs) { 2400 if (!vmx->guest_msrs) {
2395 err = -ENOMEM; 2401 err = -ENOMEM;
generated by cgit v1.2.2 (git 2.25.0) at 2024-09-27 05:39:38 -0400