]> bbs.cooldavid.org Git - net-next-2.6.git/blame - drivers/kvm/kvm_main.c
git://bbs.cooldavid.org / net-next-2.6.git / blame
? search:
summary | shortlog | log | commit | commitdiff | tree
blob | blame (incremental) | history | HEAD
KVM: CodingStyle cleanup
[net-next-2.6.git] / drivers / kvm / kvm_main.c
CommitLineData
6aa8b732
AK		 1/*
2 * Kernel-based Virtual Machine driver for Linux
3 *
4 * This module enables machines with Intel VT-x extensions to run virtual
5 * machines without emulation or binary translation.
6 *
7 * Copyright (C) 2006 Qumranet, Inc.
8 *
9 * Authors:
10 * Avi Kivity <avi@qumranet.com>
11 * Yaniv Kamay <yaniv@qumranet.com>
12 *
13 * This work is licensed under the terms of the GNU GPL, version 2. See
14 * the COPYING file in the top-level directory.
15 *
16 */
17
18#include "kvm.h"
e495606d
AK		 19#include "x86_emulate.h"
20#include "segment_descriptor.h"
85f455f7 21#include "irq.h"
6aa8b732
AK		 22
23#include <linux/kvm.h>
24#include <linux/module.h>
25#include <linux/errno.h>
6aa8b732
AK		 26#include <linux/percpu.h>
27#include <linux/gfp.h>
6aa8b732
AK		 28#include <linux/mm.h>
29#include <linux/miscdevice.h>
30#include <linux/vmalloc.h>
6aa8b732 31#include <linux/reboot.h>
6aa8b732
AK		 32#include <linux/debugfs.h>
33#include <linux/highmem.h>
34#include <linux/file.h>
59ae6c6b 35#include <linux/sysdev.h>
774c47f1 36#include <linux/cpu.h>
e8edc6e0 37#include <linux/sched.h>
d9e368d6
AK		 38#include <linux/cpumask.h>
39#include <linux/smp.h>
d6d28168 40#include <linux/anon_inodes.h>
04d2cc77 41#include <linux/profile.h>
7aa81cc0 42#include <linux/kvm_para.h>
6aa8b732 43
e495606d
AK		 44#include <asm/processor.h>
45#include <asm/msr.h>
46#include <asm/io.h>
47#include <asm/uaccess.h>
48#include <asm/desc.h>
6aa8b732
AK		 49
50MODULE_AUTHOR("Qumranet");
51MODULE_LICENSE("GPL");
52
133de902
AK		 53static DEFINE_SPINLOCK(kvm_lock);
54static LIST_HEAD(vm_list);
55
1b6c0168
AK		 56static cpumask_t cpus_hardware_enabled;
57
cbdd1bea 58struct kvm_x86_ops *kvm_x86_ops;
c16f862d
RR		 59struct kmem_cache *kvm_vcpu_cache;
60EXPORT_SYMBOL_GPL(kvm_vcpu_cache);
1165f5fe 61
15ad7146
AK		 62static __read_mostly struct preempt_ops kvm_preempt_ops;
63
1165f5fe 64#define STAT_OFFSET(x) offsetof(struct kvm_vcpu, stat.x)
6aa8b732
AK		 65
66static struct kvm_stats_debugfs_item {
67 const char *name;
1165f5fe 68 int offset;
6aa8b732
AK		 69 struct dentry *dentry;
70} debugfs_entries[] = {
1165f5fe
AK		 71 { "pf_fixed", STAT_OFFSET(pf_fixed) },
72 { "pf_guest", STAT_OFFSET(pf_guest) },
73 { "tlb_flush", STAT_OFFSET(tlb_flush) },
74 { "invlpg", STAT_OFFSET(invlpg) },
75 { "exits", STAT_OFFSET(exits) },
76 { "io_exits", STAT_OFFSET(io_exits) },
77 { "mmio_exits", STAT_OFFSET(mmio_exits) },
78 { "signal_exits", STAT_OFFSET(signal_exits) },
79 { "irq_window", STAT_OFFSET(irq_window_exits) },
80 { "halt_exits", STAT_OFFSET(halt_exits) },
b6958ce4 81 { "halt_wakeup", STAT_OFFSET(halt_wakeup) },
1165f5fe
AK		 82 { "request_irq", STAT_OFFSET(request_irq_exits) },
83 { "irq_exits", STAT_OFFSET(irq_exits) },
e6adf283 84 { "light_exits", STAT_OFFSET(light_exits) },
2cc51560 85 { "efer_reload", STAT_OFFSET(efer_reload) },
1165f5fe 86 { NULL }
6aa8b732
AK		 87};
88
89static struct dentry *debugfs_dir;
90
91#define MAX_IO_MSRS 256
92
707d92fa
RR		 93#define CR0_RESERVED_BITS \
94 (~(unsigned long)(X86_CR0_PE | X86_CR0_MP | X86_CR0_EM | X86_CR0_TS \
95 | X86_CR0_ET | X86_CR0_NE | X86_CR0_WP | X86_CR0_AM \
96 | X86_CR0_NW | X86_CR0_CD | X86_CR0_PG))
66aee91a
RR		 97#define CR4_RESERVED_BITS \
98 (~(unsigned long)(X86_CR4_VME | X86_CR4_PVI | X86_CR4_TSD | X86_CR4_DE\
99 | X86_CR4_PSE | X86_CR4_PAE | X86_CR4_MCE \
100 | X86_CR4_PGE | X86_CR4_PCE | X86_CR4_OSFXSR \
101 | X86_CR4_OSXMMEXCPT | X86_CR4_VMXE))
102
7075bc81 103#define CR8_RESERVED_BITS (~(unsigned long)X86_CR8_TPR)
6aa8b732
AK		 104#define EFER_RESERVED_BITS 0xfffffffffffff2fe
105
05b3e0c2 106#ifdef CONFIG_X86_64
d77c26fc 107/* LDT or TSS descriptor in the GDT. 16 bytes. */
6aa8b732
AK		 108struct segment_descriptor_64 {
109 struct segment_descriptor s;
110 u32 base_higher;
111 u32 pad_zero;
112};
113
114#endif
115
bccf2150
AK		 116static long kvm_vcpu_ioctl(struct file *file, unsigned int ioctl,
117 unsigned long arg);
118
6aa8b732
AK		 119unsigned long segment_base(u16 selector)
120{
121 struct descriptor_table gdt;
122 struct segment_descriptor *d;
123 unsigned long table_base;
6aa8b732
AK		 124 unsigned long v;
125
126 if (selector == 0)
127 return 0;
128
d77c26fc 129 asm("sgdt %0" : "=m"(gdt));
6aa8b732
AK		 130 table_base = gdt.base;
131
132 if (selector & 4) { /* from ldt */
133 u16 ldt_selector;
134
d77c26fc 135 asm("sldt %0" : "=g"(ldt_selector));
6aa8b732
AK		 136 table_base = segment_base(ldt_selector);
137 }
138 d = (struct segment_descriptor *)(table_base + (selector & ~7));
d77c26fc
MD		 139 v = d->base_low | ((unsigned long)d->base_mid << 16) |
140 ((unsigned long)d->base_high << 24);
05b3e0c2 141#ifdef CONFIG_X86_64
d77c26fc
MD		 142 if (d->system == 0 && (d->type == 2 || d->type == 9 || d->type == 11))
143 v |= ((unsigned long) \
144 ((struct segment_descriptor_64 *)d)->base_higher) << 32;
6aa8b732
AK		 145#endif
146 return v;
147}
148EXPORT_SYMBOL_GPL(segment_base);
149
5aacf0ca
JM		 150static inline int valid_vcpu(int n)
151{
152 return likely(n >= 0 && n < KVM_MAX_VCPUS);
153}
154
7702fd1f
AK		 155void kvm_load_guest_fpu(struct kvm_vcpu *vcpu)
156{
157 if (!vcpu->fpu_active || vcpu->guest_fpu_loaded)
158 return;
159
160 vcpu->guest_fpu_loaded = 1;
b114b080
RR		 161 fx_save(&vcpu->host_fx_image);
162 fx_restore(&vcpu->guest_fx_image);
7702fd1f
AK		 163}
164EXPORT_SYMBOL_GPL(kvm_load_guest_fpu);
165
166void kvm_put_guest_fpu(struct kvm_vcpu *vcpu)
167{
168 if (!vcpu->guest_fpu_loaded)
169 return;
170
171 vcpu->guest_fpu_loaded = 0;
b114b080
RR		 172 fx_save(&vcpu->guest_fx_image);
173 fx_restore(&vcpu->host_fx_image);
7702fd1f
AK		 174}
175EXPORT_SYMBOL_GPL(kvm_put_guest_fpu);
176
bccf2150
AK		 177/*
178 * Switches to specified vcpu, until a matching vcpu_put()
179 */
180static void vcpu_load(struct kvm_vcpu *vcpu)
6aa8b732 181{
15ad7146
AK		 182 int cpu;
183
bccf2150 184 mutex_lock(&vcpu->mutex);
15ad7146
AK		 185 cpu = get_cpu();
186 preempt_notifier_register(&vcpu->preempt_notifier);
cbdd1bea 187 kvm_x86_ops->vcpu_load(vcpu, cpu);
15ad7146 188 put_cpu();
6aa8b732
AK		 189}
190
6aa8b732
AK		 191static void vcpu_put(struct kvm_vcpu *vcpu)
192{
15ad7146 193 preempt_disable();
cbdd1bea 194 kvm_x86_ops->vcpu_put(vcpu);
15ad7146
AK		 195 preempt_notifier_unregister(&vcpu->preempt_notifier);
196 preempt_enable();
6aa8b732
AK		 197 mutex_unlock(&vcpu->mutex);
198}
199
d9e368d6
AK		 200static void ack_flush(void *_completed)
201{
d9e368d6
AK		 202}
203
204void kvm_flush_remote_tlbs(struct kvm *kvm)
205{
49d3bd7e 206 int i, cpu;
d9e368d6
AK		 207 cpumask_t cpus;
208 struct kvm_vcpu *vcpu;
d9e368d6 209
d9e368d6 210 cpus_clear(cpus);
fb3f0f51
RR		 211 for (i = 0; i < KVM_MAX_VCPUS; ++i) {
212 vcpu = kvm->vcpus[i];
213 if (!vcpu)
214 continue;
d9e368d6
AK		 215 if (test_and_set_bit(KVM_TLB_FLUSH, &vcpu->requests))
216 continue;
217 cpu = vcpu->cpu;
218 if (cpu != -1 && cpu != raw_smp_processor_id())
49d3bd7e 219 cpu_set(cpu, cpus);
d9e368d6 220 }
49d3bd7e 221 smp_call_function_mask(cpus, ack_flush, NULL, 1);
d9e368d6
AK		 222}
223
fb3f0f51
RR		 224int kvm_vcpu_init(struct kvm_vcpu *vcpu, struct kvm *kvm, unsigned id)
225{
226 struct page *page;
227 int r;
228
229 mutex_init(&vcpu->mutex);
230 vcpu->cpu = -1;
231 vcpu->mmu.root_hpa = INVALID_PAGE;
232 vcpu->kvm = kvm;
233 vcpu->vcpu_id = id;
c5ec1534
HQ		 234 if (!irqchip_in_kernel(kvm) || id == 0)
235 vcpu->mp_state = VCPU_MP_STATE_RUNNABLE;
236 else
237 vcpu->mp_state = VCPU_MP_STATE_UNINITIALIZED;
b6958ce4 238 init_waitqueue_head(&vcpu->wq);
fb3f0f51
RR		 239
240 page = alloc_page(GFP_KERNEL | __GFP_ZERO);
241 if (!page) {
242 r = -ENOMEM;
243 goto fail;
244 }
245 vcpu->run = page_address(page);
246
247 page = alloc_page(GFP_KERNEL | __GFP_ZERO);
248 if (!page) {
249 r = -ENOMEM;
250 goto fail_free_run;
251 }
252 vcpu->pio_data = page_address(page);
253
fb3f0f51
RR		 254 r = kvm_mmu_create(vcpu);
255 if (r < 0)
256 goto fail_free_pio_data;
257
76fafa5e
RR		 258 if (irqchip_in_kernel(kvm)) {
259 r = kvm_create_lapic(vcpu);
260 if (r < 0)
261 goto fail_mmu_destroy;
262 }
263
fb3f0f51
RR		 264 return 0;
265
76fafa5e
RR		 266fail_mmu_destroy:
267 kvm_mmu_destroy(vcpu);
fb3f0f51
RR		 268fail_free_pio_data:
269 free_page((unsigned long)vcpu->pio_data);
270fail_free_run:
271 free_page((unsigned long)vcpu->run);
272fail:
76fafa5e 273 return r;
fb3f0f51
RR		 274}
275EXPORT_SYMBOL_GPL(kvm_vcpu_init);
276
277void kvm_vcpu_uninit(struct kvm_vcpu *vcpu)
278{
d589444e 279 kvm_free_lapic(vcpu);
fb3f0f51
RR		 280 kvm_mmu_destroy(vcpu);
281 free_page((unsigned long)vcpu->pio_data);
282 free_page((unsigned long)vcpu->run);
283}
284EXPORT_SYMBOL_GPL(kvm_vcpu_uninit);
285
f17abe9a 286static struct kvm *kvm_create_vm(void)
6aa8b732
AK		 287{
288 struct kvm *kvm = kzalloc(sizeof(struct kvm), GFP_KERNEL);
6aa8b732
AK		 289
290 if (!kvm)
f17abe9a 291 return ERR_PTR(-ENOMEM);
6aa8b732 292
74906345 293 kvm_io_bus_init(&kvm->pio_bus);
11ec2804 294 mutex_init(&kvm->lock);
6aa8b732 295 INIT_LIST_HEAD(&kvm->active_mmu_pages);
2eeb2e94 296 kvm_io_bus_init(&kvm->mmio_bus);
5e58cfe4
RR		 297 spin_lock(&kvm_lock);
298 list_add(&kvm->vm_list, &vm_list);
299 spin_unlock(&kvm_lock);
f17abe9a
AK		 300 return kvm;
301}
302
6aa8b732
AK		 303/*
304 * Free any memory in @free but not in @dont.
305 */
306static void kvm_free_physmem_slot(struct kvm_memory_slot *free,
307 struct kvm_memory_slot *dont)
308{
309 int i;
310
311 if (!dont || free->phys_mem != dont->phys_mem)
312 if (free->phys_mem) {
313 for (i = 0; i < free->npages; ++i)
55a54f79
AK		 314 if (free->phys_mem[i])
315 __free_page(free->phys_mem[i]);
6aa8b732
AK		 316 vfree(free->phys_mem);
317 }
290fc38d
IE		 318 if (!dont || free->rmap != dont->rmap)
319 vfree(free->rmap);
6aa8b732
AK		 320
321 if (!dont || free->dirty_bitmap != dont->dirty_bitmap)
322 vfree(free->dirty_bitmap);
323
8b6d44c7 324 free->phys_mem = NULL;
6aa8b732 325 free->npages = 0;
8b6d44c7 326 free->dirty_bitmap = NULL;
6aa8b732
AK		 327}
328
329static void kvm_free_physmem(struct kvm *kvm)
330{
331 int i;
332
333 for (i = 0; i < kvm->nmemslots; ++i)
8b6d44c7 334 kvm_free_physmem_slot(&kvm->memslots[i], NULL);
6aa8b732
AK		 335}
336
039576c0
AK		 337static void free_pio_guest_pages(struct kvm_vcpu *vcpu)
338{
339 int i;
340
3077c451 341 for (i = 0; i < ARRAY_SIZE(vcpu->pio.guest_pages); ++i)
039576c0
AK		 342 if (vcpu->pio.guest_pages[i]) {
343 __free_page(vcpu->pio.guest_pages[i]);
344 vcpu->pio.guest_pages[i] = NULL;
345 }
346}
347
7b53aa56
AK		 348static void kvm_unload_vcpu_mmu(struct kvm_vcpu *vcpu)
349{
7b53aa56
AK		 350 vcpu_load(vcpu);
351 kvm_mmu_unload(vcpu);
352 vcpu_put(vcpu);
353}
354
6aa8b732
AK		 355static void kvm_free_vcpus(struct kvm *kvm)
356{
357 unsigned int i;
358
7b53aa56
AK		 359 /*
360 * Unpin any mmu pages first.
361 */
362 for (i = 0; i < KVM_MAX_VCPUS; ++i)
fb3f0f51
RR		 363 if (kvm->vcpus[i])
364 kvm_unload_vcpu_mmu(kvm->vcpus[i]);
365 for (i = 0; i < KVM_MAX_VCPUS; ++i) {
366 if (kvm->vcpus[i]) {
cbdd1bea 367 kvm_x86_ops->vcpu_free(kvm->vcpus[i]);
fb3f0f51
RR		 368 kvm->vcpus[i] = NULL;
369 }
370 }
371
6aa8b732
AK		 372}
373
f17abe9a
AK		 374static void kvm_destroy_vm(struct kvm *kvm)
375{
133de902
AK		 376 spin_lock(&kvm_lock);
377 list_del(&kvm->vm_list);
378 spin_unlock(&kvm_lock);
74906345 379 kvm_io_bus_destroy(&kvm->pio_bus);
2eeb2e94 380 kvm_io_bus_destroy(&kvm->mmio_bus);
85f455f7 381 kfree(kvm->vpic);
1fd4f2a5 382 kfree(kvm->vioapic);
6aa8b732
AK		 383 kvm_free_vcpus(kvm);
384 kvm_free_physmem(kvm);
385 kfree(kvm);
f17abe9a
AK		 386}
387
388static int kvm_vm_release(struct inode *inode, struct file *filp)
389{
390 struct kvm *kvm = filp->private_data;
391
392 kvm_destroy_vm(kvm);
6aa8b732
AK		 393 return 0;
394}
395
396static void inject_gp(struct kvm_vcpu *vcpu)
397{
cbdd1bea 398 kvm_x86_ops->inject_gp(vcpu, 0);
6aa8b732
AK		 399}
400
1342d353
AK		 401/*
402 * Load the pae pdptrs. Return true is they are all valid.
403 */
404static int load_pdptrs(struct kvm_vcpu *vcpu, unsigned long cr3)
6aa8b732
AK		 405{
406 gfn_t pdpt_gfn = cr3 >> PAGE_SHIFT;
1342d353 407 unsigned offset = ((cr3 & (PAGE_SIZE-1)) >> 5) << 2;
6aa8b732 408 int i;
1342d353 409 int ret;
c820c2aa 410 u64 pdpte[ARRAY_SIZE(vcpu->pdptrs)];
6aa8b732 411
11ec2804 412 mutex_lock(&vcpu->kvm->lock);
195aefde
IE		 413 ret = kvm_read_guest_page(vcpu->kvm, pdpt_gfn, pdpte,
414 offset * sizeof(u64), sizeof(pdpte));
415 if (ret < 0) {
c820c2aa
RR		 416 ret = 0;
417 goto out;
418 }
c820c2aa
RR		 419 for (i = 0; i < ARRAY_SIZE(pdpte); ++i) {
420 if ((pdpte[i] & 1) && (pdpte[i] & 0xfffffff0000001e6ull)) {
1342d353
AK		 421 ret = 0;
422 goto out;
423 }
6aa8b732 424 }
c820c2aa 425 ret = 1;
6aa8b732 426
c820c2aa 427 memcpy(vcpu->pdptrs, pdpte, sizeof(vcpu->pdptrs));
1342d353 428out:
11ec2804 429 mutex_unlock(&vcpu->kvm->lock);
6aa8b732 430
1342d353 431 return ret;
6aa8b732
AK		 432}
433
434void set_cr0(struct kvm_vcpu *vcpu, unsigned long cr0)
435{
707d92fa 436 if (cr0 & CR0_RESERVED_BITS) {
6aa8b732
AK		 437 printk(KERN_DEBUG "set_cr0: 0x%lx #GP, reserved bits 0x%lx\n",
438 cr0, vcpu->cr0);
439 inject_gp(vcpu);
440 return;
441 }
442
707d92fa 443 if ((cr0 & X86_CR0_NW) && !(cr0 & X86_CR0_CD)) {
6aa8b732
AK		 444 printk(KERN_DEBUG "set_cr0: #GP, CD == 0 && NW == 1\n");
445 inject_gp(vcpu);
446 return;
447 }
448
707d92fa 449 if ((cr0 & X86_CR0_PG) && !(cr0 & X86_CR0_PE)) {
6aa8b732
AK		 450 printk(KERN_DEBUG "set_cr0: #GP, set PG flag "
451 "and a clear PE flag\n");
452 inject_gp(vcpu);
453 return;
454 }
455
707d92fa 456 if (!is_paging(vcpu) && (cr0 & X86_CR0_PG)) {
05b3e0c2 457#ifdef CONFIG_X86_64
6aa8b732
AK		 458 if ((vcpu->shadow_efer & EFER_LME)) {
459 int cs_db, cs_l;
460
461 if (!is_pae(vcpu)) {
462 printk(KERN_DEBUG "set_cr0: #GP, start paging "
463 "in long mode while PAE is disabled\n");
464 inject_gp(vcpu);
465 return;
466 }
cbdd1bea 467 kvm_x86_ops->get_cs_db_l_bits(vcpu, &cs_db, &cs_l);
6aa8b732
AK		 468 if (cs_l) {
469 printk(KERN_DEBUG "set_cr0: #GP, start paging "
470 "in long mode while CS.L == 1\n");
471 inject_gp(vcpu);
472 return;
473
474 }
475 } else
476#endif
1342d353 477 if (is_pae(vcpu) && !load_pdptrs(vcpu, vcpu->cr3)) {
6aa8b732
AK		 478 printk(KERN_DEBUG "set_cr0: #GP, pdptrs "
479 "reserved bits\n");
480 inject_gp(vcpu);
481 return;
482 }
483
484 }
485
cbdd1bea 486 kvm_x86_ops->set_cr0(vcpu, cr0);
6aa8b732
AK		 487 vcpu->cr0 = cr0;
488
11ec2804 489 mutex_lock(&vcpu->kvm->lock);
6aa8b732 490 kvm_mmu_reset_context(vcpu);
11ec2804 491 mutex_unlock(&vcpu->kvm->lock);
6aa8b732
AK		 492 return;
493}
494EXPORT_SYMBOL_GPL(set_cr0);
495
496void lmsw(struct kvm_vcpu *vcpu, unsigned long msw)
497{
498 set_cr0(vcpu, (vcpu->cr0 & ~0x0ful) | (msw & 0x0f));
499}
500EXPORT_SYMBOL_GPL(lmsw);
501
502void set_cr4(struct kvm_vcpu *vcpu, unsigned long cr4)
503{
66aee91a 504 if (cr4 & CR4_RESERVED_BITS) {
6aa8b732
AK		 505 printk(KERN_DEBUG "set_cr4: #GP, reserved bits\n");
506 inject_gp(vcpu);
507 return;
508 }
509
a9058ecd 510 if (is_long_mode(vcpu)) {
66aee91a 511 if (!(cr4 & X86_CR4_PAE)) {
6aa8b732
AK		 512 printk(KERN_DEBUG "set_cr4: #GP, clearing PAE while "
513 "in long mode\n");
514 inject_gp(vcpu);
515 return;
516 }
66aee91a 517 } else if (is_paging(vcpu) && !is_pae(vcpu) && (cr4 & X86_CR4_PAE)
1342d353 518 && !load_pdptrs(vcpu, vcpu->cr3)) {
6aa8b732
AK		 519 printk(KERN_DEBUG "set_cr4: #GP, pdptrs reserved bits\n");
520 inject_gp(vcpu);
310bc76c 521 return;
6aa8b732
AK		 522 }
523
66aee91a 524 if (cr4 & X86_CR4_VMXE) {
6aa8b732
AK		 525 printk(KERN_DEBUG "set_cr4: #GP, setting VMXE\n");
526 inject_gp(vcpu);
527 return;
528 }
cbdd1bea 529 kvm_x86_ops->set_cr4(vcpu, cr4);
81f50e3b 530 vcpu->cr4 = cr4;
11ec2804 531 mutex_lock(&vcpu->kvm->lock);
6aa8b732 532 kvm_mmu_reset_context(vcpu);
11ec2804 533 mutex_unlock(&vcpu->kvm->lock);
6aa8b732
AK		 534}
535EXPORT_SYMBOL_GPL(set_cr4);
536
537void set_cr3(struct kvm_vcpu *vcpu, unsigned long cr3)
538{
a9058ecd 539 if (is_long_mode(vcpu)) {
f802a307 540 if (cr3 & CR3_L_MODE_RESERVED_BITS) {
6aa8b732
AK		 541 printk(KERN_DEBUG "set_cr3: #GP, reserved bits\n");
542 inject_gp(vcpu);
543 return;
544 }
545 } else {
f802a307
RR		 546 if (is_pae(vcpu)) {
547 if (cr3 & CR3_PAE_RESERVED_BITS) {
548 printk(KERN_DEBUG
549 "set_cr3: #GP, reserved bits\n");
550 inject_gp(vcpu);
551 return;
552 }
553 if (is_paging(vcpu) && !load_pdptrs(vcpu, cr3)) {
554 printk(KERN_DEBUG "set_cr3: #GP, pdptrs "
555 "reserved bits\n");
556 inject_gp(vcpu);
557 return;
558 }
6aa8b732 559 }
21764863
RH		 560 /*
561 * We don't check reserved bits in nonpae mode, because
562 * this isn't enforced, and VMware depends on this.
563 */
6aa8b732
AK		 564 }
565
11ec2804 566 mutex_lock(&vcpu->kvm->lock);
d21225ee
IM		 567 /*
568 * Does the new cr3 value map to physical memory? (Note, we
569 * catch an invalid cr3 even in real-mode, because it would
570 * cause trouble later on when we turn on paging anyway.)
571 *
572 * A real CPU would silently accept an invalid cr3 and would
573 * attempt to use it - with largely undefined (and often hard
574 * to debug) behavior on the guest side.
575 */
576 if (unlikely(!gfn_to_memslot(vcpu->kvm, cr3 >> PAGE_SHIFT)))
577 inject_gp(vcpu);
fb764416
RR		 578 else {
579 vcpu->cr3 = cr3;
d21225ee 580 vcpu->mmu.new_cr3(vcpu);
fb764416 581 }
11ec2804 582 mutex_unlock(&vcpu->kvm->lock);
6aa8b732
AK		 583}
584EXPORT_SYMBOL_GPL(set_cr3);
585
586void set_cr8(struct kvm_vcpu *vcpu, unsigned long cr8)
587{
7075bc81 588 if (cr8 & CR8_RESERVED_BITS) {
6aa8b732
AK		 589 printk(KERN_DEBUG "set_cr8: #GP, reserved bits 0x%lx\n", cr8);
590 inject_gp(vcpu);
591 return;
592 }
97222cc8
ED		 593 if (irqchip_in_kernel(vcpu->kvm))
594 kvm_lapic_set_tpr(vcpu, cr8);
595 else
596 vcpu->cr8 = cr8;
6aa8b732
AK		 597}
598EXPORT_SYMBOL_GPL(set_cr8);
599
7017fc3d
ED		 600unsigned long get_cr8(struct kvm_vcpu *vcpu)
601{
97222cc8
ED		 602 if (irqchip_in_kernel(vcpu->kvm))
603 return kvm_lapic_get_cr8(vcpu);
604 else
605 return vcpu->cr8;
7017fc3d
ED		 606}
607EXPORT_SYMBOL_GPL(get_cr8);
608
609u64 kvm_get_apic_base(struct kvm_vcpu *vcpu)
610{
97222cc8
ED		 611 if (irqchip_in_kernel(vcpu->kvm))
612 return vcpu->apic_base;
613 else
614 return vcpu->apic_base;
7017fc3d
ED		 615}
616EXPORT_SYMBOL_GPL(kvm_get_apic_base);
617
618void kvm_set_apic_base(struct kvm_vcpu *vcpu, u64 data)
619{
97222cc8
ED		 620 /* TODO: reserve bits check */
621 if (irqchip_in_kernel(vcpu->kvm))
622 kvm_lapic_set_base(vcpu, data);
623 else
624 vcpu->apic_base = data;
7017fc3d
ED		 625}
626EXPORT_SYMBOL_GPL(kvm_set_apic_base);
627
6aa8b732
AK		 628void fx_init(struct kvm_vcpu *vcpu)
629{
b114b080 630 unsigned after_mxcsr_mask;
6aa8b732 631
9bd01506
RR		 632 /* Initialize guest FPU by resetting ours and saving into guest's */
633 preempt_disable();
b114b080 634 fx_save(&vcpu->host_fx_image);
6aa8b732 635 fpu_init();
b114b080
RR		 636 fx_save(&vcpu->guest_fx_image);
637 fx_restore(&vcpu->host_fx_image);
9bd01506 638 preempt_enable();
6aa8b732 639
380102c8 640 vcpu->cr0 |= X86_CR0_ET;
b114b080
RR		 641 after_mxcsr_mask = offsetof(struct i387_fxsave_struct, st_space);
642 vcpu->guest_fx_image.mxcsr = 0x1f80;
643 memset((void *)&vcpu->guest_fx_image + after_mxcsr_mask,
644 0, sizeof(struct i387_fxsave_struct) - after_mxcsr_mask);
6aa8b732
AK		 645}
646EXPORT_SYMBOL_GPL(fx_init);
647
6aa8b732
AK		 648/*
649 * Allocate some memory and give it an address in the guest physical address
650 * space.
651 *
652 * Discontiguous memory is allowed, mostly for framebuffers.
653 */
2c6f5df9
AK		 654static int kvm_vm_ioctl_set_memory_region(struct kvm *kvm,
655 struct kvm_memory_region *mem)
6aa8b732
AK		 656{
657 int r;
658 gfn_t base_gfn;
659 unsigned long npages;
660 unsigned long i;
661 struct kvm_memory_slot *memslot;
662 struct kvm_memory_slot old, new;
6aa8b732
AK		 663
664 r = -EINVAL;
665 /* General sanity checks */
666 if (mem->memory_size & (PAGE_SIZE - 1))
667 goto out;
668 if (mem->guest_phys_addr & (PAGE_SIZE - 1))
669 goto out;
670 if (mem->slot >= KVM_MEMORY_SLOTS)
671 goto out;
672 if (mem->guest_phys_addr + mem->memory_size < mem->guest_phys_addr)
673 goto out;
674
675 memslot = &kvm->memslots[mem->slot];
676 base_gfn = mem->guest_phys_addr >> PAGE_SHIFT;
677 npages = mem->memory_size >> PAGE_SHIFT;
678
679 if (!npages)
680 mem->flags &= ~KVM_MEM_LOG_DIRTY_PAGES;
681
11ec2804 682 mutex_lock(&kvm->lock);
6aa8b732 683
6aa8b732
AK		 684 new = old = *memslot;
685
686 new.base_gfn = base_gfn;
687 new.npages = npages;
688 new.flags = mem->flags;
689
690 /* Disallow changing a memory slot's size. */
691 r = -EINVAL;
692 if (npages && old.npages && npages != old.npages)
693 goto out_unlock;
694
695 /* Check for overlaps */
696 r = -EEXIST;
697 for (i = 0; i < KVM_MEMORY_SLOTS; ++i) {
698 struct kvm_memory_slot *s = &kvm->memslots[i];
699
700 if (s == memslot)
701 continue;
702 if (!((base_gfn + npages <= s->base_gfn) ||
703 (base_gfn >= s->base_gfn + s->npages)))
704 goto out_unlock;
705 }
6aa8b732
AK		 706
707 /* Deallocate if slot is being removed */
708 if (!npages)
8b6d44c7 709 new.phys_mem = NULL;
6aa8b732
AK		 710
711 /* Free page dirty bitmap if unneeded */
712 if (!(new.flags & KVM_MEM_LOG_DIRTY_PAGES))
8b6d44c7 713 new.dirty_bitmap = NULL;
6aa8b732
AK		 714
715 r = -ENOMEM;
716
717 /* Allocate if a slot is being created */
718 if (npages && !new.phys_mem) {
719 new.phys_mem = vmalloc(npages * sizeof(struct page *));
720
721 if (!new.phys_mem)
0d8d2bd4 722 goto out_unlock;
6aa8b732 723
d77c26fc 724 new.rmap = vmalloc(npages * sizeof(struct page *));
290fc38d
IE		 725
726 if (!new.rmap)
727 goto out_unlock;
728
6aa8b732 729 memset(new.phys_mem, 0, npages * sizeof(struct page *));
290fc38d 730 memset(new.rmap, 0, npages * sizeof(*new.rmap));
6aa8b732
AK		 731 for (i = 0; i < npages; ++i) {
732 new.phys_mem[i] = alloc_page(GFP_HIGHUSER
733 | __GFP_ZERO);
734 if (!new.phys_mem[i])
0d8d2bd4 735 goto out_unlock;
6aa8b732
AK		 736 }
737 }
738
739 /* Allocate page dirty bitmap if needed */
740 if ((new.flags & KVM_MEM_LOG_DIRTY_PAGES) && !new.dirty_bitmap) {
741 unsigned dirty_bytes = ALIGN(npages, BITS_PER_LONG) / 8;
742
743 new.dirty_bitmap = vmalloc(dirty_bytes);
744 if (!new.dirty_bitmap)
0d8d2bd4 745 goto out_unlock;
6aa8b732
AK		 746 memset(new.dirty_bitmap, 0, dirty_bytes);
747 }
748
6aa8b732
AK		 749 if (mem->slot >= kvm->nmemslots)
750 kvm->nmemslots = mem->slot + 1;
751
82ce2c96
IE		 752 if (!kvm->n_requested_mmu_pages) {
753 unsigned int n_pages;
754
755 if (npages) {
756 n_pages = npages * KVM_PERMILLE_MMU_PAGES / 1000;
757 kvm_mmu_change_mmu_pages(kvm, kvm->n_alloc_mmu_pages +
758 n_pages);
759 } else {
760 unsigned int nr_mmu_pages;
761
762 n_pages = old.npages * KVM_PERMILLE_MMU_PAGES / 1000;
763 nr_mmu_pages = kvm->n_alloc_mmu_pages - n_pages;
764 nr_mmu_pages = max(nr_mmu_pages,
765 (unsigned int) KVM_MIN_ALLOC_MMU_PAGES);
766 kvm_mmu_change_mmu_pages(kvm, nr_mmu_pages);
767 }
768 }
769
6aa8b732 770 *memslot = new;
6aa8b732 771
90cb0529
AK		 772 kvm_mmu_slot_remove_write_access(kvm, mem->slot);
773 kvm_flush_remote_tlbs(kvm);
6aa8b732 774
11ec2804 775 mutex_unlock(&kvm->lock);
6aa8b732
AK		 776
777 kvm_free_physmem_slot(&old, &new);
778 return 0;
779
780out_unlock:
11ec2804 781 mutex_unlock(&kvm->lock);
6aa8b732
AK		 782 kvm_free_physmem_slot(&new, &old);
783out:
784 return r;
785}
786
82ce2c96
IE		 787static int kvm_vm_ioctl_set_nr_mmu_pages(struct kvm *kvm,
788 u32 kvm_nr_mmu_pages)
789{
790 if (kvm_nr_mmu_pages < KVM_MIN_ALLOC_MMU_PAGES)
791 return -EINVAL;
792
793 mutex_lock(&kvm->lock);
794
795 kvm_mmu_change_mmu_pages(kvm, kvm_nr_mmu_pages);
796 kvm->n_requested_mmu_pages = kvm_nr_mmu_pages;
797
798 mutex_unlock(&kvm->lock);
799 return 0;
800}
801
802static int kvm_vm_ioctl_get_nr_mmu_pages(struct kvm *kvm)
803{
804 return kvm->n_alloc_mmu_pages;
805}
806
6aa8b732
AK		 807/*
808 * Get (and clear) the dirty memory log for a memory slot.
809 */
2c6f5df9
AK		 810static int kvm_vm_ioctl_get_dirty_log(struct kvm *kvm,
811 struct kvm_dirty_log *log)
6aa8b732
AK		 812{
813 struct kvm_memory_slot *memslot;
814 int r, i;
815 int n;
816 unsigned long any = 0;
817
11ec2804 818 mutex_lock(&kvm->lock);
6aa8b732 819
6aa8b732
AK		 820 r = -EINVAL;
821 if (log->slot >= KVM_MEMORY_SLOTS)
822 goto out;
823
824 memslot = &kvm->memslots[log->slot];
825 r = -ENOENT;
826 if (!memslot->dirty_bitmap)
827 goto out;
828
cd1a4a98 829 n = ALIGN(memslot->npages, BITS_PER_LONG) / 8;
6aa8b732 830
cd1a4a98 831 for (i = 0; !any && i < n/sizeof(long); ++i)
6aa8b732
AK		 832 any = memslot->dirty_bitmap[i];
833
834 r = -EFAULT;
835 if (copy_to_user(log->dirty_bitmap, memslot->dirty_bitmap, n))
836 goto out;
837
39214915
RR		 838 /* If nothing is dirty, don't bother messing with page tables. */
839 if (any) {
39214915
RR		 840 kvm_mmu_slot_remove_write_access(kvm, log->slot);
841 kvm_flush_remote_tlbs(kvm);
842 memset(memslot->dirty_bitmap, 0, n);
39214915 843 }
6aa8b732
AK		 844
845 r = 0;
846
847out:
11ec2804 848 mutex_unlock(&kvm->lock);
6aa8b732
AK		 849 return r;
850}
851
e8207547
AK		 852/*
853 * Set a new alias region. Aliases map a portion of physical memory into
854 * another portion. This is useful for memory windows, for example the PC
855 * VGA region.
856 */
857static int kvm_vm_ioctl_set_memory_alias(struct kvm *kvm,
858 struct kvm_memory_alias *alias)
859{
860 int r, n;
861 struct kvm_mem_alias *p;
862
863 r = -EINVAL;
864 /* General sanity checks */
865 if (alias->memory_size & (PAGE_SIZE - 1))
866 goto out;
867 if (alias->guest_phys_addr & (PAGE_SIZE - 1))
868 goto out;
869 if (alias->slot >= KVM_ALIAS_SLOTS)
870 goto out;
871 if (alias->guest_phys_addr + alias->memory_size
872 < alias->guest_phys_addr)
873 goto out;
874 if (alias->target_phys_addr + alias->memory_size
875 < alias->target_phys_addr)
876 goto out;
877
11ec2804 878 mutex_lock(&kvm->lock);
e8207547
AK		 879
880 p = &kvm->aliases[alias->slot];
881 p->base_gfn = alias->guest_phys_addr >> PAGE_SHIFT;
882 p->npages = alias->memory_size >> PAGE_SHIFT;
883 p->target_gfn = alias->target_phys_addr >> PAGE_SHIFT;
884
885 for (n = KVM_ALIAS_SLOTS; n > 0; --n)
886 if (kvm->aliases[n - 1].npages)
887 break;
888 kvm->naliases = n;
889
90cb0529 890 kvm_mmu_zap_all(kvm);
e8207547 891
11ec2804 892 mutex_unlock(&kvm->lock);
e8207547
AK		 893
894 return 0;
895
896out:
897 return r;
898}
899
6ceb9d79
HQ		 900static int kvm_vm_ioctl_get_irqchip(struct kvm *kvm, struct kvm_irqchip *chip)
901{
902 int r;
903
904 r = 0;
905 switch (chip->chip_id) {
906 case KVM_IRQCHIP_PIC_MASTER:
d77c26fc 907 memcpy(&chip->chip.pic,
6ceb9d79
HQ		 908 &pic_irqchip(kvm)->pics[0],
909 sizeof(struct kvm_pic_state));
910 break;
911 case KVM_IRQCHIP_PIC_SLAVE:
d77c26fc 912 memcpy(&chip->chip.pic,
6ceb9d79
HQ		 913 &pic_irqchip(kvm)->pics[1],
914 sizeof(struct kvm_pic_state));
915 break;
6bf9e962 916 case KVM_IRQCHIP_IOAPIC:
d77c26fc 917 memcpy(&chip->chip.ioapic,
6bf9e962
HQ		 918 ioapic_irqchip(kvm),
919 sizeof(struct kvm_ioapic_state));
920 break;
6ceb9d79
HQ		 921 default:
922 r = -EINVAL;
923 break;
924 }
925 return r;
926}
927
928static int kvm_vm_ioctl_set_irqchip(struct kvm *kvm, struct kvm_irqchip *chip)
929{
930 int r;
931
932 r = 0;
933 switch (chip->chip_id) {
934 case KVM_IRQCHIP_PIC_MASTER:
d77c26fc 935 memcpy(&pic_irqchip(kvm)->pics[0],
6ceb9d79
HQ		 936 &chip->chip.pic,
937 sizeof(struct kvm_pic_state));
938 break;
939 case KVM_IRQCHIP_PIC_SLAVE:
d77c26fc 940 memcpy(&pic_irqchip(kvm)->pics[1],
6ceb9d79
HQ		 941 &chip->chip.pic,
942 sizeof(struct kvm_pic_state));
943 break;
6bf9e962 944 case KVM_IRQCHIP_IOAPIC:
d77c26fc 945 memcpy(ioapic_irqchip(kvm),
6bf9e962
HQ		 946 &chip->chip.ioapic,
947 sizeof(struct kvm_ioapic_state));
948 break;
6ceb9d79
HQ		 949 default:
950 r = -EINVAL;
951 break;
952 }
953 kvm_pic_update_irq(pic_irqchip(kvm));
954 return r;
955}
956
290fc38d 957gfn_t unalias_gfn(struct kvm *kvm, gfn_t gfn)
e8207547
AK		 958{
959 int i;
960 struct kvm_mem_alias *alias;
961
962 for (i = 0; i < kvm->naliases; ++i) {
963 alias = &kvm->aliases[i];
964 if (gfn >= alias->base_gfn
965 && gfn < alias->base_gfn + alias->npages)
966 return alias->target_gfn + gfn - alias->base_gfn;
967 }
968 return gfn;
969}
970
971static struct kvm_memory_slot *__gfn_to_memslot(struct kvm *kvm, gfn_t gfn)
6aa8b732
AK		 972{
973 int i;
974
975 for (i = 0; i < kvm->nmemslots; ++i) {
976 struct kvm_memory_slot *memslot = &kvm->memslots[i];
977
978 if (gfn >= memslot->base_gfn
979 && gfn < memslot->base_gfn + memslot->npages)
980 return memslot;
981 }
8b6d44c7 982 return NULL;
6aa8b732 983}
e8207547
AK		 984
985struct kvm_memory_slot *gfn_to_memslot(struct kvm *kvm, gfn_t gfn)
986{
987 gfn = unalias_gfn(kvm, gfn);
988 return __gfn_to_memslot(kvm, gfn);
989}
6aa8b732 990
954bbbc2
AK		 991struct page *gfn_to_page(struct kvm *kvm, gfn_t gfn)
992{
993 struct kvm_memory_slot *slot;
994
e8207547
AK		 995 gfn = unalias_gfn(kvm, gfn);
996 slot = __gfn_to_memslot(kvm, gfn);
954bbbc2
AK		 997 if (!slot)
998 return NULL;
999 return slot->phys_mem[gfn - slot->base_gfn];
1000}
1001EXPORT_SYMBOL_GPL(gfn_to_page);
1002
195aefde
IE		 1003static int next_segment(unsigned long len, int offset)
1004{
1005 if (len > PAGE_SIZE - offset)
1006 return PAGE_SIZE - offset;
1007 else
1008 return len;
1009}
1010
1011int kvm_read_guest_page(struct kvm *kvm, gfn_t gfn, void *data, int offset,
1012 int len)
1013{
1014 void *page_virt;
1015 struct page *page;
1016
1017 page = gfn_to_page(kvm, gfn);
1018 if (!page)
1019 return -EFAULT;
1020 page_virt = kmap_atomic(page, KM_USER0);
1021
1022 memcpy(data, page_virt + offset, len);
1023
1024 kunmap_atomic(page_virt, KM_USER0);
1025 return 0;
1026}
1027EXPORT_SYMBOL_GPL(kvm_read_guest_page);
1028
1029int kvm_read_guest(struct kvm *kvm, gpa_t gpa, void *data, unsigned long len)
1030{
1031 gfn_t gfn = gpa >> PAGE_SHIFT;
1032 int seg;
1033 int offset = offset_in_page(gpa);
1034 int ret;
1035
1036 while ((seg = next_segment(len, offset)) != 0) {
1037 ret = kvm_read_guest_page(kvm, gfn, data, offset, seg);
1038 if (ret < 0)
1039 return ret;
1040 offset = 0;
1041 len -= seg;
1042 data += seg;
1043 ++gfn;
1044 }
1045 return 0;
1046}
1047EXPORT_SYMBOL_GPL(kvm_read_guest);
1048
1049int kvm_write_guest_page(struct kvm *kvm, gfn_t gfn, const void *data,
1050 int offset, int len)
1051{
1052 void *page_virt;
1053 struct page *page;
1054
1055 page = gfn_to_page(kvm, gfn);
1056 if (!page)
1057 return -EFAULT;
1058 page_virt = kmap_atomic(page, KM_USER0);
1059
1060 memcpy(page_virt + offset, data, len);
1061
1062 kunmap_atomic(page_virt, KM_USER0);
1063 mark_page_dirty(kvm, gfn);
1064 return 0;
1065}
1066EXPORT_SYMBOL_GPL(kvm_write_guest_page);
1067
1068int kvm_write_guest(struct kvm *kvm, gpa_t gpa, const void *data,
1069 unsigned long len)
1070{
1071 gfn_t gfn = gpa >> PAGE_SHIFT;
1072 int seg;
1073 int offset = offset_in_page(gpa);
1074 int ret;
1075
1076 while ((seg = next_segment(len, offset)) != 0) {
1077 ret = kvm_write_guest_page(kvm, gfn, data, offset, seg);
1078 if (ret < 0)
1079 return ret;
1080 offset = 0;
1081 len -= seg;
1082 data += seg;
1083 ++gfn;
1084 }
1085 return 0;
1086}
1087
1088int kvm_clear_guest_page(struct kvm *kvm, gfn_t gfn, int offset, int len)
1089{
1090 void *page_virt;
1091 struct page *page;
1092
1093 page = gfn_to_page(kvm, gfn);
1094 if (!page)
1095 return -EFAULT;
1096 page_virt = kmap_atomic(page, KM_USER0);
1097
1098 memset(page_virt + offset, 0, len);
1099
1100 kunmap_atomic(page_virt, KM_USER0);
1101 return 0;
1102}
1103EXPORT_SYMBOL_GPL(kvm_clear_guest_page);
1104
1105int kvm_clear_guest(struct kvm *kvm, gpa_t gpa, unsigned long len)
1106{
1107 gfn_t gfn = gpa >> PAGE_SHIFT;
1108 int seg;
1109 int offset = offset_in_page(gpa);
1110 int ret;
1111
1112 while ((seg = next_segment(len, offset)) != 0) {
1113 ret = kvm_clear_guest_page(kvm, gfn, offset, seg);
1114 if (ret < 0)
1115 return ret;
1116 offset = 0;
1117 len -= seg;
1118 ++gfn;
1119 }
1120 return 0;
1121}
1122EXPORT_SYMBOL_GPL(kvm_clear_guest);
1123
7e9d619d 1124/* WARNING: Does not work on aliased pages. */
6aa8b732
AK		 1125void mark_page_dirty(struct kvm *kvm, gfn_t gfn)
1126{
31389947 1127 struct kvm_memory_slot *memslot;
6aa8b732 1128
7e9d619d
RR		 1129 memslot = __gfn_to_memslot(kvm, gfn);
1130 if (memslot && memslot->dirty_bitmap) {
1131 unsigned long rel_gfn = gfn - memslot->base_gfn;
6aa8b732 1132
7e9d619d
RR		 1133 /* avoid RMW */
1134 if (!test_bit(rel_gfn, memslot->dirty_bitmap))
1135 set_bit(rel_gfn, memslot->dirty_bitmap);
6aa8b732
AK		 1136 }
1137}
1138
e7d5d76c 1139int emulator_read_std(unsigned long addr,
4c690a1e 1140 void *val,
6aa8b732 1141 unsigned int bytes,
cebff02b 1142 struct kvm_vcpu *vcpu)
6aa8b732 1143{
6aa8b732
AK		 1144 void *data = val;
1145
1146 while (bytes) {
1147 gpa_t gpa = vcpu->mmu.gva_to_gpa(vcpu, addr);
1148 unsigned offset = addr & (PAGE_SIZE-1);
1149 unsigned tocopy = min(bytes, (unsigned)PAGE_SIZE - offset);
195aefde 1150 int ret;
6aa8b732
AK		 1151
1152 if (gpa == UNMAPPED_GVA)
1153 return X86EMUL_PROPAGATE_FAULT;
195aefde
IE		 1154 ret = kvm_read_guest(vcpu->kvm, gpa, data, tocopy);
1155 if (ret < 0)
6aa8b732 1156 return X86EMUL_UNHANDLEABLE;
6aa8b732
AK		 1157
1158 bytes -= tocopy;
1159 data += tocopy;
1160 addr += tocopy;
1161 }
1162
1163 return X86EMUL_CONTINUE;
1164}
e7d5d76c 1165EXPORT_SYMBOL_GPL(emulator_read_std);
6aa8b732
AK		 1166
1167static int emulator_write_std(unsigned long addr,
4c690a1e 1168 const void *val,
6aa8b732 1169 unsigned int bytes,
cebff02b 1170 struct kvm_vcpu *vcpu)
6aa8b732 1171{
f0242478 1172 pr_unimpl(vcpu, "emulator_write_std: addr %lx n %d\n", addr, bytes);
6aa8b732
AK		 1173 return X86EMUL_UNHANDLEABLE;
1174}
1175
97222cc8
ED		 1176/*
1177 * Only apic need an MMIO device hook, so shortcut now..
1178 */
1179static struct kvm_io_device *vcpu_find_pervcpu_dev(struct kvm_vcpu *vcpu,
1180 gpa_t addr)
1181{
1182 struct kvm_io_device *dev;
1183
1184 if (vcpu->apic) {
1185 dev = &vcpu->apic->dev;
1186 if (dev->in_range(dev, addr))
1187 return dev;
1188 }
1189 return NULL;
1190}
1191
2eeb2e94
GH		 1192static struct kvm_io_device *vcpu_find_mmio_dev(struct kvm_vcpu *vcpu,
1193 gpa_t addr)
1194{
97222cc8
ED		 1195 struct kvm_io_device *dev;
1196
1197 dev = vcpu_find_pervcpu_dev(vcpu, addr);
1198 if (dev == NULL)
1199 dev = kvm_io_bus_find_dev(&vcpu->kvm->mmio_bus, addr);
1200 return dev;
2eeb2e94
GH		 1201}
1202
74906345
ED		 1203static struct kvm_io_device *vcpu_find_pio_dev(struct kvm_vcpu *vcpu,
1204 gpa_t addr)
1205{
1206 return kvm_io_bus_find_dev(&vcpu->kvm->pio_bus, addr);
1207}
1208
6aa8b732 1209static int emulator_read_emulated(unsigned long addr,
4c690a1e 1210 void *val,
6aa8b732 1211 unsigned int bytes,
cebff02b 1212 struct kvm_vcpu *vcpu)
6aa8b732 1213{
2eeb2e94
GH		 1214 struct kvm_io_device *mmio_dev;
1215 gpa_t gpa;
6aa8b732
AK		 1216
1217 if (vcpu->mmio_read_completed) {
1218 memcpy(val, vcpu->mmio_data, bytes);
1219 vcpu->mmio_read_completed = 0;
1220 return X86EMUL_CONTINUE;
cebff02b 1221 } else if (emulator_read_std(addr, val, bytes, vcpu)
6aa8b732
AK		 1222 == X86EMUL_CONTINUE)
1223 return X86EMUL_CONTINUE;
d27d4aca 1224
2eeb2e94
GH		 1225 gpa = vcpu->mmu.gva_to_gpa(vcpu, addr);
1226 if (gpa == UNMAPPED_GVA)
1227 return X86EMUL_PROPAGATE_FAULT;
6aa8b732 1228
2eeb2e94
GH		 1229 /*
1230 * Is this MMIO handled locally?
1231 */
1232 mmio_dev = vcpu_find_mmio_dev(vcpu, gpa);
1233 if (mmio_dev) {
1234 kvm_iodevice_read(mmio_dev, gpa, bytes, val);
1235 return X86EMUL_CONTINUE;
6aa8b732 1236 }
2eeb2e94
GH		 1237
1238 vcpu->mmio_needed = 1;
1239 vcpu->mmio_phys_addr = gpa;
1240 vcpu->mmio_size = bytes;
1241 vcpu->mmio_is_write = 0;
1242
1243 return X86EMUL_UNHANDLEABLE;
6aa8b732
AK		 1244}
1245
da4a00f0 1246static int emulator_write_phys(struct kvm_vcpu *vcpu, gpa_t gpa,
4c690a1e 1247 const void *val, int bytes)
da4a00f0 1248{
195aefde 1249 int ret;
da4a00f0 1250
195aefde
IE		 1251 ret = kvm_write_guest(vcpu->kvm, gpa, val, bytes);
1252 if (ret < 0)
da4a00f0 1253 return 0;
fe551881 1254 kvm_mmu_pte_write(vcpu, gpa, val, bytes);
da4a00f0
AK		 1255 return 1;
1256}
1257
b0fcd903
AK		 1258static int emulator_write_emulated_onepage(unsigned long addr,
1259 const void *val,
1260 unsigned int bytes,
cebff02b 1261 struct kvm_vcpu *vcpu)
6aa8b732 1262{
2eeb2e94
GH		 1263 struct kvm_io_device *mmio_dev;
1264 gpa_t gpa = vcpu->mmu.gva_to_gpa(vcpu, addr);
6aa8b732 1265
c9047f53 1266 if (gpa == UNMAPPED_GVA) {
cbdd1bea 1267 kvm_x86_ops->inject_page_fault(vcpu, addr, 2);
6aa8b732 1268 return X86EMUL_PROPAGATE_FAULT;
c9047f53 1269 }
6aa8b732 1270
da4a00f0
AK		 1271 if (emulator_write_phys(vcpu, gpa, val, bytes))
1272 return X86EMUL_CONTINUE;
1273
2eeb2e94
GH		 1274 /*
1275 * Is this MMIO handled locally?
1276 */
1277 mmio_dev = vcpu_find_mmio_dev(vcpu, gpa);
1278 if (mmio_dev) {
1279 kvm_iodevice_write(mmio_dev, gpa, bytes, val);
1280 return X86EMUL_CONTINUE;
1281 }
1282
6aa8b732
AK		 1283 vcpu->mmio_needed = 1;
1284 vcpu->mmio_phys_addr = gpa;
1285 vcpu->mmio_size = bytes;
1286 vcpu->mmio_is_write = 1;
4c690a1e 1287 memcpy(vcpu->mmio_data, val, bytes);
6aa8b732
AK		 1288
1289 return X86EMUL_CONTINUE;
1290}
1291
e7d5d76c 1292int emulator_write_emulated(unsigned long addr,
b0fcd903
AK		 1293 const void *val,
1294 unsigned int bytes,
cebff02b 1295 struct kvm_vcpu *vcpu)
b0fcd903
AK		 1296{
1297 /* Crossing a page boundary? */
1298 if (((addr + bytes - 1) ^ addr) & PAGE_MASK) {
1299 int rc, now;
1300
1301 now = -addr & ~PAGE_MASK;
cebff02b 1302 rc = emulator_write_emulated_onepage(addr, val, now, vcpu);
b0fcd903
AK		 1303 if (rc != X86EMUL_CONTINUE)
1304 return rc;
1305 addr += now;
1306 val += now;
1307 bytes -= now;
1308 }
cebff02b 1309 return emulator_write_emulated_onepage(addr, val, bytes, vcpu);
b0fcd903 1310}
e7d5d76c 1311EXPORT_SYMBOL_GPL(emulator_write_emulated);
b0fcd903 1312
6aa8b732 1313static int emulator_cmpxchg_emulated(unsigned long addr,
4c690a1e
AK		 1314 const void *old,
1315 const void *new,
6aa8b732 1316 unsigned int bytes,
cebff02b 1317 struct kvm_vcpu *vcpu)
6aa8b732
AK		 1318{
1319 static int reported;
1320
1321 if (!reported) {
1322 reported = 1;
1323 printk(KERN_WARNING "kvm: emulating exchange as write\n");
1324 }
cebff02b 1325 return emulator_write_emulated(addr, new, bytes, vcpu);
6aa8b732
AK		 1326}
1327
1328static unsigned long get_segment_base(struct kvm_vcpu *vcpu, int seg)
1329{
cbdd1bea 1330 return kvm_x86_ops->get_segment_base(vcpu, seg);
6aa8b732
AK		 1331}
1332
1333int emulate_invlpg(struct kvm_vcpu *vcpu, gva_t address)
1334{
6aa8b732
AK		 1335 return X86EMUL_CONTINUE;
1336}
1337
1338int emulate_clts(struct kvm_vcpu *vcpu)
1339{
404fb881 1340 kvm_x86_ops->set_cr0(vcpu, vcpu->cr0 & ~X86_CR0_TS);
6aa8b732
AK		 1341 return X86EMUL_CONTINUE;
1342}
1343
d77c26fc 1344int emulator_get_dr(struct x86_emulate_ctxt *ctxt, int dr, unsigned long *dest)
6aa8b732
AK		 1345{
1346 struct kvm_vcpu *vcpu = ctxt->vcpu;
1347
1348 switch (dr) {
1349 case 0 ... 3:
cbdd1bea 1350 *dest = kvm_x86_ops->get_dr(vcpu, dr);
6aa8b732
AK		 1351 return X86EMUL_CONTINUE;
1352 default:
f0242478 1353 pr_unimpl(vcpu, "%s: unexpected dr %u\n", __FUNCTION__, dr);
6aa8b732
AK		 1354 return X86EMUL_UNHANDLEABLE;
1355 }
1356}
1357
1358int emulator_set_dr(struct x86_emulate_ctxt *ctxt, int dr, unsigned long value)
1359{
1360 unsigned long mask = (ctxt->mode == X86EMUL_MODE_PROT64) ? ~0ULL : ~0U;
1361 int exception;
1362
cbdd1bea 1363 kvm_x86_ops->set_dr(ctxt->vcpu, dr, value & mask, &exception);
6aa8b732
AK		 1364 if (exception) {
1365 /* FIXME: better handling */
1366 return X86EMUL_UNHANDLEABLE;
1367 }
1368 return X86EMUL_CONTINUE;
1369}
1370
054b1369 1371void kvm_report_emulation_failure(struct kvm_vcpu *vcpu, const char *context)
6aa8b732
AK		 1372{
1373 static int reported;
1374 u8 opcodes[4];
054b1369 1375 unsigned long rip = vcpu->rip;
6aa8b732
AK		 1376 unsigned long rip_linear;
1377
054b1369 1378 rip_linear = rip + get_segment_base(vcpu, VCPU_SREG_CS);
6aa8b732
AK		 1379
1380 if (reported)
1381 return;
1382
054b1369 1383 emulator_read_std(rip_linear, (void *)opcodes, 4, vcpu);
6aa8b732 1384
054b1369
AK		 1385 printk(KERN_ERR "emulation failed (%s) rip %lx %02x %02x %02x %02x\n",
1386 context, rip, opcodes[0], opcodes[1], opcodes[2], opcodes[3]);
6aa8b732
AK		 1387 reported = 1;
1388}
054b1369 1389EXPORT_SYMBOL_GPL(kvm_report_emulation_failure);
6aa8b732
AK		 1390
1391struct x86_emulate_ops emulate_ops = {
1392 .read_std = emulator_read_std,
1393 .write_std = emulator_write_std,
1394 .read_emulated = emulator_read_emulated,
1395 .write_emulated = emulator_write_emulated,
1396 .cmpxchg_emulated = emulator_cmpxchg_emulated,
1397};
1398
1399int emulate_instruction(struct kvm_vcpu *vcpu,
1400 struct kvm_run *run,
1401 unsigned long cr2,
3427318f
LV		 1402 u16 error_code,
1403 int no_decode)
6aa8b732 1404{
a22436b7 1405 int r;
6aa8b732 1406
e7df56e4 1407 vcpu->mmio_fault_cr2 = cr2;
cbdd1bea 1408 kvm_x86_ops->cache_regs(vcpu);
6aa8b732 1409
6aa8b732 1410 vcpu->mmio_is_write = 0;
e70669ab 1411 vcpu->pio.string = 0;
3427318f
LV		 1412
1413 if (!no_decode) {
1414 int cs_db, cs_l;
1415 kvm_x86_ops->get_cs_db_l_bits(vcpu, &cs_db, &cs_l);
1416
1417 vcpu->emulate_ctxt.vcpu = vcpu;
1418 vcpu->emulate_ctxt.eflags = kvm_x86_ops->get_rflags(vcpu);
1419 vcpu->emulate_ctxt.cr2 = cr2;
1420 vcpu->emulate_ctxt.mode =
1421 (vcpu->emulate_ctxt.eflags & X86_EFLAGS_VM)
1422 ? X86EMUL_MODE_REAL : cs_l
1423 ? X86EMUL_MODE_PROT64 : cs_db
1424 ? X86EMUL_MODE_PROT32 : X86EMUL_MODE_PROT16;
1425
1426 if (vcpu->emulate_ctxt.mode == X86EMUL_MODE_PROT64) {
1427 vcpu->emulate_ctxt.cs_base = 0;
1428 vcpu->emulate_ctxt.ds_base = 0;
1429 vcpu->emulate_ctxt.es_base = 0;
1430 vcpu->emulate_ctxt.ss_base = 0;
1431 } else {
1432 vcpu->emulate_ctxt.cs_base =
1433 get_segment_base(vcpu, VCPU_SREG_CS);
1434 vcpu->emulate_ctxt.ds_base =
1435 get_segment_base(vcpu, VCPU_SREG_DS);
1436 vcpu->emulate_ctxt.es_base =
1437 get_segment_base(vcpu, VCPU_SREG_ES);
1438 vcpu->emulate_ctxt.ss_base =
1439 get_segment_base(vcpu, VCPU_SREG_SS);
1440 }
1441
1442 vcpu->emulate_ctxt.gs_base =
1443 get_segment_base(vcpu, VCPU_SREG_GS);
1444 vcpu->emulate_ctxt.fs_base =
1445 get_segment_base(vcpu, VCPU_SREG_FS);
1446
1447 r = x86_decode_insn(&vcpu->emulate_ctxt, &emulate_ops);
a22436b7
LV		 1448 if (r) {
1449 if (kvm_mmu_unprotect_page_virt(vcpu, cr2))
1450 return EMULATE_DONE;
1451 return EMULATE_FAIL;
1452 }
3427318f
LV		 1453 }
1454
a22436b7 1455 r = x86_emulate_insn(&vcpu->emulate_ctxt, &emulate_ops);
1be3aa47 1456
e70669ab
LV		 1457 if (vcpu->pio.string)
1458 return EMULATE_DO_MMIO;
6aa8b732
AK		 1459
1460 if ((r || vcpu->mmio_is_write) && run) {
8fc0d085 1461 run->exit_reason = KVM_EXIT_MMIO;
6aa8b732
AK		 1462 run->mmio.phys_addr = vcpu->mmio_phys_addr;
1463 memcpy(run->mmio.data, vcpu->mmio_data, 8);
1464 run->mmio.len = vcpu->mmio_size;
1465 run->mmio.is_write = vcpu->mmio_is_write;
1466 }
1467
1468 if (r) {
a436036b
AK		 1469 if (kvm_mmu_unprotect_page_virt(vcpu, cr2))
1470 return EMULATE_DONE;
6aa8b732 1471 if (!vcpu->mmio_needed) {
054b1369 1472 kvm_report_emulation_failure(vcpu, "mmio");
6aa8b732
AK		 1473 return EMULATE_FAIL;
1474 }
1475 return EMULATE_DO_MMIO;
1476 }
1477
cbdd1bea 1478 kvm_x86_ops->decache_regs(vcpu);
3427318f 1479 kvm_x86_ops->set_rflags(vcpu, vcpu->emulate_ctxt.eflags);
6aa8b732 1480
02c83209
AK		 1481 if (vcpu->mmio_is_write) {
1482 vcpu->mmio_needed = 0;
6aa8b732 1483 return EMULATE_DO_MMIO;
02c83209 1484 }
6aa8b732
AK		 1485
1486 return EMULATE_DONE;
1487}
1488EXPORT_SYMBOL_GPL(emulate_instruction);
1489
b6958ce4
ED		 1490/*
1491 * The vCPU has executed a HLT instruction with in-kernel mode enabled.
1492 */
c5ec1534 1493static void kvm_vcpu_block(struct kvm_vcpu *vcpu)
d3bef15f 1494{
b6958ce4
ED		 1495 DECLARE_WAITQUEUE(wait, current);
1496
1497 add_wait_queue(&vcpu->wq, &wait);
1498
1499 /*
1500 * We will block until either an interrupt or a signal wakes us up
1501 */
c5ec1534
HQ		 1502 while (!kvm_cpu_has_interrupt(vcpu)
1503 && !signal_pending(current)
1504 && vcpu->mp_state != VCPU_MP_STATE_RUNNABLE
1505 && vcpu->mp_state != VCPU_MP_STATE_SIPI_RECEIVED) {
b6958ce4
ED		 1506 set_current_state(TASK_INTERRUPTIBLE);
1507 vcpu_put(vcpu);
1508 schedule();
1509 vcpu_load(vcpu);
1510 }
d3bef15f 1511
c5ec1534 1512 __set_current_state(TASK_RUNNING);
b6958ce4 1513 remove_wait_queue(&vcpu->wq, &wait);
b6958ce4
ED		 1514}
1515
1516int kvm_emulate_halt(struct kvm_vcpu *vcpu)
1517{
d3bef15f 1518 ++vcpu->stat.halt_exits;
b6958ce4 1519 if (irqchip_in_kernel(vcpu->kvm)) {
c5ec1534
HQ		 1520 vcpu->mp_state = VCPU_MP_STATE_HALTED;
1521 kvm_vcpu_block(vcpu);
1522 if (vcpu->mp_state != VCPU_MP_STATE_RUNNABLE)
1523 return -EINTR;
b6958ce4
ED		 1524 return 1;
1525 } else {
1526 vcpu->run->exit_reason = KVM_EXIT_HLT;
1527 return 0;
1528 }
d3bef15f
AK		 1529}
1530EXPORT_SYMBOL_GPL(kvm_emulate_halt);
1531
7aa81cc0 1532int kvm_emulate_hypercall(struct kvm_vcpu *vcpu)
270fd9b9 1533{
7aa81cc0 1534 unsigned long nr, a0, a1, a2, a3, ret;
270fd9b9 1535
cbdd1bea 1536 kvm_x86_ops->cache_regs(vcpu);
7aa81cc0
AL		 1537
1538 nr = vcpu->regs[VCPU_REGS_RAX];
1539 a0 = vcpu->regs[VCPU_REGS_RBX];
1540 a1 = vcpu->regs[VCPU_REGS_RCX];
1541 a2 = vcpu->regs[VCPU_REGS_RDX];
1542 a3 = vcpu->regs[VCPU_REGS_RSI];
1543
1544 if (!is_long_mode(vcpu)) {
1545 nr &= 0xFFFFFFFF;
1546 a0 &= 0xFFFFFFFF;
1547 a1 &= 0xFFFFFFFF;
1548 a2 &= 0xFFFFFFFF;
1549 a3 &= 0xFFFFFFFF;
270fd9b9 1550 }
7aa81cc0 1551
270fd9b9
AK		 1552 switch (nr) {
1553 default:
7aa81cc0
AL		 1554 ret = -KVM_ENOSYS;
1555 break;
270fd9b9
AK		 1556 }
1557 vcpu->regs[VCPU_REGS_RAX] = ret;
cbdd1bea 1558 kvm_x86_ops->decache_regs(vcpu);
7aa81cc0
AL		 1559 return 0;
1560}
1561EXPORT_SYMBOL_GPL(kvm_emulate_hypercall);
1562
1563int kvm_fix_hypercall(struct kvm_vcpu *vcpu)
1564{
1565 char instruction[3];
1566 int ret = 0;
1567
1568 mutex_lock(&vcpu->kvm->lock);
1569
1570 /*
1571 * Blow out the MMU to ensure that no other VCPU has an active mapping
1572 * to ensure that the updated hypercall appears atomically across all
1573 * VCPUs.
1574 */
1575 kvm_mmu_zap_all(vcpu->kvm);
1576
1577 kvm_x86_ops->cache_regs(vcpu);
1578 kvm_x86_ops->patch_hypercall(vcpu, instruction);
1579 if (emulator_write_emulated(vcpu->rip, instruction, 3, vcpu)
1580 != X86EMUL_CONTINUE)
1581 ret = -EFAULT;
1582
1583 mutex_unlock(&vcpu->kvm->lock);
1584
1585 return ret;
270fd9b9 1586}
270fd9b9 1587
6aa8b732
AK		 1588static u64 mk_cr_64(u64 curr_cr, u32 new_val)
1589{
1590 return (curr_cr & ~((1ULL << 32) - 1)) | new_val;
1591}
1592
1593void realmode_lgdt(struct kvm_vcpu *vcpu, u16 limit, unsigned long base)
1594{
1595 struct descriptor_table dt = { limit, base };
1596
cbdd1bea 1597 kvm_x86_ops->set_gdt(vcpu, &dt);
6aa8b732
AK		 1598}
1599
1600void realmode_lidt(struct kvm_vcpu *vcpu, u16 limit, unsigned long base)
1601{
1602 struct descriptor_table dt = { limit, base };
1603
cbdd1bea 1604 kvm_x86_ops->set_idt(vcpu, &dt);
6aa8b732
AK		 1605}
1606
1607void realmode_lmsw(struct kvm_vcpu *vcpu, unsigned long msw,
1608 unsigned long *rflags)
1609{
1610 lmsw(vcpu, msw);
cbdd1bea 1611 *rflags = kvm_x86_ops->get_rflags(vcpu);
6aa8b732
AK		 1612}
1613
1614unsigned long realmode_get_cr(struct kvm_vcpu *vcpu, int cr)
1615{
cbdd1bea 1616 kvm_x86_ops->decache_cr4_guest_bits(vcpu);
6aa8b732
AK		 1617 switch (cr) {
1618 case 0:
1619 return vcpu->cr0;
1620 case 2:
1621 return vcpu->cr2;
1622 case 3:
1623 return vcpu->cr3;
1624 case 4:
1625 return vcpu->cr4;
1626 default:
1627 vcpu_printf(vcpu, "%s: unexpected cr %u\n", __FUNCTION__, cr);
1628 return 0;
1629 }
1630}
1631
1632void realmode_set_cr(struct kvm_vcpu *vcpu, int cr, unsigned long val,
1633 unsigned long *rflags)
1634{
1635 switch (cr) {
1636 case 0:
1637 set_cr0(vcpu, mk_cr_64(vcpu->cr0, val));
cbdd1bea 1638 *rflags = kvm_x86_ops->get_rflags(vcpu);
6aa8b732
AK		 1639 break;
1640 case 2:
1641 vcpu->cr2 = val;
1642 break;
1643 case 3:
1644 set_cr3(vcpu, val);
1645 break;
1646 case 4:
1647 set_cr4(vcpu, mk_cr_64(vcpu->cr4, val));
1648 break;
1649 default:
1650 vcpu_printf(vcpu, "%s: unexpected cr %u\n", __FUNCTION__, cr);
1651 }
1652}
1653
3bab1f5d
AK		 1654int kvm_get_msr_common(struct kvm_vcpu *vcpu, u32 msr, u64 *pdata)
1655{
1656 u64 data;
1657
1658 switch (msr) {
1659 case 0xc0010010: /* SYSCFG */
1660 case 0xc0010015: /* HWCR */
1661 case MSR_IA32_PLATFORM_ID:
1662 case MSR_IA32_P5_MC_ADDR:
1663 case MSR_IA32_P5_MC_TYPE:
1664 case MSR_IA32_MC0_CTL:
1665 case MSR_IA32_MCG_STATUS:
1666 case MSR_IA32_MCG_CAP:
1667 case MSR_IA32_MC0_MISC:
1668 case MSR_IA32_MC0_MISC+4:
1669 case MSR_IA32_MC0_MISC+8:
1670 case MSR_IA32_MC0_MISC+12:
1671 case MSR_IA32_MC0_MISC+16:
1672 case MSR_IA32_UCODE_REV:
a8d13ea2 1673 case MSR_IA32_PERF_STATUS:
2dc7094b 1674 case MSR_IA32_EBL_CR_POWERON:
3bab1f5d
AK		 1675 /* MTRR registers */
1676 case 0xfe:
1677 case 0x200 ... 0x2ff:
1678 data = 0;
1679 break;
a8d13ea2
AK		 1680 case 0xcd: /* fsb frequency */
1681 data = 3;
1682 break;
3bab1f5d 1683 case MSR_IA32_APICBASE:
7017fc3d 1684 data = kvm_get_apic_base(vcpu);
3bab1f5d 1685 break;
6f00e68f
AK		 1686 case MSR_IA32_MISC_ENABLE:
1687 data = vcpu->ia32_misc_enable_msr;
1688 break;
3bab1f5d
AK		 1689#ifdef CONFIG_X86_64
1690 case MSR_EFER:
1691 data = vcpu->shadow_efer;
1692 break;
1693#endif
1694 default:
f0242478 1695 pr_unimpl(vcpu, "unhandled rdmsr: 0x%x\n", msr);
3bab1f5d
AK		 1696 return 1;
1697 }
1698 *pdata = data;
1699 return 0;
1700}
1701EXPORT_SYMBOL_GPL(kvm_get_msr_common);
1702
6aa8b732
AK		 1703/*
1704 * Reads an msr value (of 'msr_index') into 'pdata'.
1705 * Returns 0 on success, non-0 otherwise.
1706 * Assumes vcpu_load() was already called.
1707 */
35f3f286 1708int kvm_get_msr(struct kvm_vcpu *vcpu, u32 msr_index, u64 *pdata)
6aa8b732 1709{
cbdd1bea 1710 return kvm_x86_ops->get_msr(vcpu, msr_index, pdata);
6aa8b732
AK		 1711}
1712
05b3e0c2 1713#ifdef CONFIG_X86_64
6aa8b732 1714
3bab1f5d 1715static void set_efer(struct kvm_vcpu *vcpu, u64 efer)
6aa8b732 1716{
6aa8b732
AK		 1717 if (efer & EFER_RESERVED_BITS) {
1718 printk(KERN_DEBUG "set_efer: 0x%llx #GP, reserved bits\n",
1719 efer);
1720 inject_gp(vcpu);
1721 return;
1722 }
1723
1724 if (is_paging(vcpu)
1725 && (vcpu->shadow_efer & EFER_LME) != (efer & EFER_LME)) {
1726 printk(KERN_DEBUG "set_efer: #GP, change LME while paging\n");
1727 inject_gp(vcpu);
1728 return;
1729 }
1730
cbdd1bea 1731 kvm_x86_ops->set_efer(vcpu, efer);
7725f0ba 1732
6aa8b732
AK		 1733 efer &= ~EFER_LMA;
1734 efer |= vcpu->shadow_efer & EFER_LMA;
1735
1736 vcpu->shadow_efer = efer;
6aa8b732 1737}
6aa8b732
AK		 1738
1739#endif
1740
3bab1f5d
AK		 1741int kvm_set_msr_common(struct kvm_vcpu *vcpu, u32 msr, u64 data)
1742{
1743 switch (msr) {
1744#ifdef CONFIG_X86_64
1745 case MSR_EFER:
1746 set_efer(vcpu, data);
1747 break;
1748#endif
1749 case MSR_IA32_MC0_STATUS:
f0242478 1750 pr_unimpl(vcpu, "%s: MSR_IA32_MC0_STATUS 0x%llx, nop\n",
3bab1f5d
AK		 1751 __FUNCTION__, data);
1752 break;
0e5bf0d0 1753 case MSR_IA32_MCG_STATUS:
f0242478 1754 pr_unimpl(vcpu, "%s: MSR_IA32_MCG_STATUS 0x%llx, nop\n",
0e5bf0d0
SK		 1755 __FUNCTION__, data);
1756 break;
3bab1f5d
AK		 1757 case MSR_IA32_UCODE_REV:
1758 case MSR_IA32_UCODE_WRITE:
1759 case 0x200 ... 0x2ff: /* MTRRs */
1760 break;
1761 case MSR_IA32_APICBASE:
7017fc3d 1762 kvm_set_apic_base(vcpu, data);
3bab1f5d 1763 break;
6f00e68f
AK		 1764 case MSR_IA32_MISC_ENABLE:
1765 vcpu->ia32_misc_enable_msr = data;
1766 break;
3bab1f5d 1767 default:
f0242478 1768 pr_unimpl(vcpu, "unhandled wrmsr: 0x%x\n", msr);
3bab1f5d
AK		 1769 return 1;
1770 }
1771 return 0;
1772}
1773EXPORT_SYMBOL_GPL(kvm_set_msr_common);
1774
6aa8b732
AK		 1775/*
1776 * Writes msr value into into the appropriate "register".
1777 * Returns 0 on success, non-0 otherwise.
1778 * Assumes vcpu_load() was already called.
1779 */
35f3f286 1780int kvm_set_msr(struct kvm_vcpu *vcpu, u32 msr_index, u64 data)
6aa8b732 1781{
cbdd1bea 1782 return kvm_x86_ops->set_msr(vcpu, msr_index, data);
6aa8b732
AK		 1783}
1784
1785void kvm_resched(struct kvm_vcpu *vcpu)
1786{
3fca0365
YD		 1787 if (!need_resched())
1788 return;
6aa8b732 1789 cond_resched();
6aa8b732
AK		 1790}
1791EXPORT_SYMBOL_GPL(kvm_resched);
1792
06465c5a
AK		 1793void kvm_emulate_cpuid(struct kvm_vcpu *vcpu)
1794{
1795 int i;
1796 u32 function;
1797 struct kvm_cpuid_entry *e, *best;
1798
cbdd1bea 1799 kvm_x86_ops->cache_regs(vcpu);
06465c5a
AK		 1800 function = vcpu->regs[VCPU_REGS_RAX];
1801 vcpu->regs[VCPU_REGS_RAX] = 0;
1802 vcpu->regs[VCPU_REGS_RBX] = 0;
1803 vcpu->regs[VCPU_REGS_RCX] = 0;
1804 vcpu->regs[VCPU_REGS_RDX] = 0;
1805 best = NULL;
1806 for (i = 0; i < vcpu->cpuid_nent; ++i) {
1807 e = &vcpu->cpuid_entries[i];
1808 if (e->function == function) {
1809 best = e;
1810 break;
1811 }
1812 /*
1813 * Both basic or both extended?
1814 */
1815 if (((e->function ^ function) & 0x80000000) == 0)
1816 if (!best || e->function > best->function)
1817 best = e;
1818 }
1819 if (best) {
1820 vcpu->regs[VCPU_REGS_RAX] = best->eax;
1821 vcpu->regs[VCPU_REGS_RBX] = best->ebx;
1822 vcpu->regs[VCPU_REGS_RCX] = best->ecx;
1823 vcpu->regs[VCPU_REGS_RDX] = best->edx;
1824 }
cbdd1bea
CE		 1825 kvm_x86_ops->decache_regs(vcpu);
1826 kvm_x86_ops->skip_emulated_instruction(vcpu);
06465c5a
AK		 1827}
1828EXPORT_SYMBOL_GPL(kvm_emulate_cpuid);
1829
039576c0 1830static int pio_copy_data(struct kvm_vcpu *vcpu)
46fc1477 1831{
039576c0
AK		 1832 void *p = vcpu->pio_data;
1833 void *q;
1834 unsigned bytes;
1835 int nr_pages = vcpu->pio.guest_pages[1] ? 2 : 1;
1836
039576c0
AK		 1837 q = vmap(vcpu->pio.guest_pages, nr_pages, VM_READ|VM_WRITE,
1838 PAGE_KERNEL);
1839 if (!q) {
039576c0
AK		 1840 free_pio_guest_pages(vcpu);
1841 return -ENOMEM;
1842 }
1843 q += vcpu->pio.guest_page_offset;
1844 bytes = vcpu->pio.size * vcpu->pio.cur_count;
1845 if (vcpu->pio.in)
1846 memcpy(q, p, bytes);
1847 else
1848 memcpy(p, q, bytes);
1849 q -= vcpu->pio.guest_page_offset;
1850 vunmap(q);
039576c0
AK		 1851 free_pio_guest_pages(vcpu);
1852 return 0;
1853}
1854
1855static int complete_pio(struct kvm_vcpu *vcpu)
1856{
1857 struct kvm_pio_request *io = &vcpu->pio;
46fc1477 1858 long delta;
039576c0 1859 int r;
46fc1477 1860
cbdd1bea 1861 kvm_x86_ops->cache_regs(vcpu);
46fc1477
AK		 1862
1863 if (!io->string) {
039576c0
AK		 1864 if (io->in)
1865 memcpy(&vcpu->regs[VCPU_REGS_RAX], vcpu->pio_data,
46fc1477
AK		 1866 io->size);
1867 } else {
039576c0
AK		 1868 if (io->in) {
1869 r = pio_copy_data(vcpu);
1870 if (r) {
cbdd1bea 1871 kvm_x86_ops->cache_regs(vcpu);
039576c0
AK		 1872 return r;
1873 }
1874 }
1875
46fc1477
AK		 1876 delta = 1;
1877 if (io->rep) {
039576c0 1878 delta *= io->cur_count;
46fc1477
AK		 1879 /*
1880 * The size of the register should really depend on
1881 * current address size.
1882 */
1883 vcpu->regs[VCPU_REGS_RCX] -= delta;
1884 }
039576c0 1885 if (io->down)
46fc1477
AK		 1886 delta = -delta;
1887 delta *= io->size;
039576c0 1888 if (io->in)
46fc1477
AK		 1889 vcpu->regs[VCPU_REGS_RDI] += delta;
1890 else
1891 vcpu->regs[VCPU_REGS_RSI] += delta;
1892 }
1893
cbdd1bea 1894 kvm_x86_ops->decache_regs(vcpu);
46fc1477 1895
039576c0
AK		 1896 io->count -= io->cur_count;
1897 io->cur_count = 0;
1898
039576c0 1899 return 0;
46fc1477
AK		 1900}
1901
65619eb5
ED		 1902static void kernel_pio(struct kvm_io_device *pio_dev,
1903 struct kvm_vcpu *vcpu,
1904 void *pd)
74906345
ED		 1905{
1906 /* TODO: String I/O for in kernel device */
1907
9cf98828 1908 mutex_lock(&vcpu->kvm->lock);
74906345
ED		 1909 if (vcpu->pio.in)
1910 kvm_iodevice_read(pio_dev, vcpu->pio.port,
1911 vcpu->pio.size,
65619eb5 1912 pd);
74906345
ED		 1913 else
1914 kvm_iodevice_write(pio_dev, vcpu->pio.port,
1915 vcpu->pio.size,
65619eb5 1916 pd);
9cf98828 1917 mutex_unlock(&vcpu->kvm->lock);
65619eb5
ED		 1918}
1919
1920static void pio_string_write(struct kvm_io_device *pio_dev,
1921 struct kvm_vcpu *vcpu)
1922{
1923 struct kvm_pio_request *io = &vcpu->pio;
1924 void *pd = vcpu->pio_data;
1925 int i;
1926
9cf98828 1927 mutex_lock(&vcpu->kvm->lock);
65619eb5
ED		 1928 for (i = 0; i < io->cur_count; i++) {
1929 kvm_iodevice_write(pio_dev, io->port,
1930 io->size,
1931 pd);
1932 pd += io->size;
1933 }
9cf98828 1934 mutex_unlock(&vcpu->kvm->lock);
74906345
ED		 1935}
1936
d77c26fc 1937int kvm_emulate_pio(struct kvm_vcpu *vcpu, struct kvm_run *run, int in,
3090dd73
LV		 1938 int size, unsigned port)
1939{
1940 struct kvm_io_device *pio_dev;
1941
1942 vcpu->run->exit_reason = KVM_EXIT_IO;
1943 vcpu->run->io.direction = in ? KVM_EXIT_IO_IN : KVM_EXIT_IO_OUT;
1944 vcpu->run->io.size = vcpu->pio.size = size;
1945 vcpu->run->io.data_offset = KVM_PIO_PAGE_OFFSET * PAGE_SIZE;
1946 vcpu->run->io.count = vcpu->pio.count = vcpu->pio.cur_count = 1;
1947 vcpu->run->io.port = vcpu->pio.port = port;
1948 vcpu->pio.in = in;
1949 vcpu->pio.string = 0;
1950 vcpu->pio.down = 0;
1951 vcpu->pio.guest_page_offset = 0;
1952 vcpu->pio.rep = 0;
1953
cbdd1bea 1954 kvm_x86_ops->cache_regs(vcpu);
3090dd73 1955 memcpy(vcpu->pio_data, &vcpu->regs[VCPU_REGS_RAX], 4);
cbdd1bea 1956 kvm_x86_ops->decache_regs(vcpu);
3090dd73 1957
0967b7bf
AK		 1958 kvm_x86_ops->skip_emulated_instruction(vcpu);
1959
3090dd73
LV		 1960 pio_dev = vcpu_find_pio_dev(vcpu, port);
1961 if (pio_dev) {
1962 kernel_pio(pio_dev, vcpu, vcpu->pio_data);
1963 complete_pio(vcpu);
1964 return 1;
1965 }
1966 return 0;
1967}
1968EXPORT_SYMBOL_GPL(kvm_emulate_pio);
1969
1970int kvm_emulate_pio_string(struct kvm_vcpu *vcpu, struct kvm_run *run, int in,
1971 int size, unsigned long count, int down,
039576c0
AK		 1972 gva_t address, int rep, unsigned port)
1973{
1974 unsigned now, in_page;
65619eb5 1975 int i, ret = 0;
039576c0
AK		 1976 int nr_pages = 1;
1977 struct page *page;
74906345 1978 struct kvm_io_device *pio_dev;
039576c0
AK		 1979
1980 vcpu->run->exit_reason = KVM_EXIT_IO;
1981 vcpu->run->io.direction = in ? KVM_EXIT_IO_IN : KVM_EXIT_IO_OUT;
3090dd73 1982 vcpu->run->io.size = vcpu->pio.size = size;
039576c0 1983 vcpu->run->io.data_offset = KVM_PIO_PAGE_OFFSET * PAGE_SIZE;
3090dd73
LV		 1984 vcpu->run->io.count = vcpu->pio.count = vcpu->pio.cur_count = count;
1985 vcpu->run->io.port = vcpu->pio.port = port;
039576c0 1986 vcpu->pio.in = in;
3090dd73 1987 vcpu->pio.string = 1;
039576c0
AK		 1988 vcpu->pio.down = down;
1989 vcpu->pio.guest_page_offset = offset_in_page(address);
1990 vcpu->pio.rep = rep;
1991
039576c0 1992 if (!count) {
cbdd1bea 1993 kvm_x86_ops->skip_emulated_instruction(vcpu);
039576c0
AK		 1994 return 1;
1995 }
1996
039576c0
AK		 1997 if (!down)
1998 in_page = PAGE_SIZE - offset_in_page(address);
1999 else
2000 in_page = offset_in_page(address) + size;
2001 now = min(count, (unsigned long)in_page / size);
2002 if (!now) {
2003 /*
2004 * String I/O straddles page boundary. Pin two guest pages
2005 * so that we satisfy atomicity constraints. Do just one
2006 * transaction to avoid complexity.
2007 */
2008 nr_pages = 2;
2009 now = 1;
2010 }
2011 if (down) {
2012 /*
2013 * String I/O in reverse. Yuck. Kill the guest, fix later.
2014 */
f0242478 2015 pr_unimpl(vcpu, "guest string pio down\n");
039576c0
AK		 2016 inject_gp(vcpu);
2017 return 1;
2018 }
2019 vcpu->run->io.count = now;
2020 vcpu->pio.cur_count = now;
2021
0967b7bf
AK		 2022 if (vcpu->pio.cur_count == vcpu->pio.count)
2023 kvm_x86_ops->skip_emulated_instruction(vcpu);
2024
039576c0 2025 for (i = 0; i < nr_pages; ++i) {
11ec2804 2026 mutex_lock(&vcpu->kvm->lock);
039576c0
AK		 2027 page = gva_to_page(vcpu, address + i * PAGE_SIZE);
2028 if (page)
2029 get_page(page);
2030 vcpu->pio.guest_pages[i] = page;
11ec2804 2031 mutex_unlock(&vcpu->kvm->lock);
039576c0
AK		 2032 if (!page) {
2033 inject_gp(vcpu);
2034 free_pio_guest_pages(vcpu);
2035 return 1;
2036 }
2037 }
2038
3090dd73 2039 pio_dev = vcpu_find_pio_dev(vcpu, port);
65619eb5
ED		 2040 if (!vcpu->pio.in) {
2041 /* string PIO write */
2042 ret = pio_copy_data(vcpu);
2043 if (ret >= 0 && pio_dev) {
2044 pio_string_write(pio_dev, vcpu);
2045 complete_pio(vcpu);
2046 if (vcpu->pio.count == 0)
2047 ret = 1;
2048 }
2049 } else if (pio_dev)
f0242478 2050 pr_unimpl(vcpu, "no string pio read support yet, "
65619eb5
ED		 2051 "port %x size %d count %ld\n",
2052 port, size, count);
2053
2054 return ret;
039576c0 2055}
3090dd73 2056EXPORT_SYMBOL_GPL(kvm_emulate_pio_string);
039576c0 2057
04d2cc77
AK		 2058/*
2059 * Check if userspace requested an interrupt window, and that the
2060 * interrupt window is open.
2061 *
2062 * No need to exit to userspace if we already have an interrupt queued.
2063 */
2064static int dm_request_for_irq_injection(struct kvm_vcpu *vcpu,
2065 struct kvm_run *kvm_run)
2066{
2067 return (!vcpu->irq_summary &&
2068 kvm_run->request_interrupt_window &&
2069 vcpu->interrupt_window_open &&
2070 (kvm_x86_ops->get_rflags(vcpu) & X86_EFLAGS_IF));
2071}
2072
2073static void post_kvm_run_save(struct kvm_vcpu *vcpu,
2074 struct kvm_run *kvm_run)
2075{
2076 kvm_run->if_flag = (kvm_x86_ops->get_rflags(vcpu) & X86_EFLAGS_IF) != 0;
2077 kvm_run->cr8 = get_cr8(vcpu);
2078 kvm_run->apic_base = kvm_get_apic_base(vcpu);
2079 if (irqchip_in_kernel(vcpu->kvm))
2080 kvm_run->ready_for_interrupt_injection = 1;
2081 else
2082 kvm_run->ready_for_interrupt_injection =
2083 (vcpu->interrupt_window_open &&
2084 vcpu->irq_summary == 0);
2085}
2086
2087static int __vcpu_run(struct kvm_vcpu *vcpu, struct kvm_run *kvm_run)
2088{
2089 int r;
2090
2091 if (unlikely(vcpu->mp_state == VCPU_MP_STATE_SIPI_RECEIVED)) {
d77c26fc 2092 pr_debug("vcpu %d received sipi with vector # %x\n",
04d2cc77
AK		 2093 vcpu->vcpu_id, vcpu->sipi_vector);
2094 kvm_lapic_reset(vcpu);
2095 kvm_x86_ops->vcpu_reset(vcpu);
2096 vcpu->mp_state = VCPU_MP_STATE_RUNNABLE;
2097 }
2098
2099preempted:
2100 if (vcpu->guest_debug.enabled)
2101 kvm_x86_ops->guest_debug_pre(vcpu);
2102
2103again:
2104 r = kvm_mmu_reload(vcpu);
2105 if (unlikely(r))
2106 goto out;
2107
2108 preempt_disable();
2109
2110 kvm_x86_ops->prepare_guest_switch(vcpu);
2111 kvm_load_guest_fpu(vcpu);
2112
2113 local_irq_disable();
2114
2115 if (signal_pending(current)) {
2116 local_irq_enable();
2117 preempt_enable();
2118 r = -EINTR;
2119 kvm_run->exit_reason = KVM_EXIT_INTR;
2120 ++vcpu->stat.signal_exits;
2121 goto out;
2122 }
2123
2124 if (irqchip_in_kernel(vcpu->kvm))
2125 kvm_x86_ops->inject_pending_irq(vcpu);
2126 else if (!vcpu->mmio_read_completed)
2127 kvm_x86_ops->inject_pending_vectors(vcpu, kvm_run);
2128
2129 vcpu->guest_mode = 1;
d172fcd3 2130 kvm_guest_enter();
04d2cc77
AK		 2131
2132 if (vcpu->requests)
2133 if (test_and_clear_bit(KVM_TLB_FLUSH, &vcpu->requests))
2134 kvm_x86_ops->tlb_flush(vcpu);
2135
2136 kvm_x86_ops->run(vcpu, kvm_run);
2137
2138 vcpu->guest_mode = 0;
2139 local_irq_enable();
2140
2141 ++vcpu->stat.exits;
2142
0552f73b
LV		 2143 /*
2144 * We must have an instruction between local_irq_enable() and
2145 * kvm_guest_exit(), so the timer interrupt isn't delayed by
2146 * the interrupt shadow. The stat.exits increment will do nicely.
2147 * But we need to prevent reordering, hence this barrier():
2148 */
2149 barrier();
2150
2151 kvm_guest_exit();
2152
04d2cc77
AK		 2153 preempt_enable();
2154
2155 /*
2156 * Profile KVM exit RIPs:
2157 */
2158 if (unlikely(prof_on == KVM_PROFILING)) {
2159 kvm_x86_ops->cache_regs(vcpu);
2160 profile_hit(KVM_PROFILING, (void *)vcpu->rip);
2161 }
2162
2163 r = kvm_x86_ops->handle_exit(kvm_run, vcpu);
2164
2165 if (r > 0) {
2166 if (dm_request_for_irq_injection(vcpu, kvm_run)) {
2167 r = -EINTR;
2168 kvm_run->exit_reason = KVM_EXIT_INTR;
2169 ++vcpu->stat.request_irq_exits;
2170 goto out;
2171 }
2172 if (!need_resched()) {
2173 ++vcpu->stat.light_exits;
2174 goto again;
2175 }
2176 }
2177
2178out:
2179 if (r > 0) {
2180 kvm_resched(vcpu);
2181 goto preempted;
2182 }
2183
2184 post_kvm_run_save(vcpu, kvm_run);
2185
2186 return r;
2187}
2188
2189
bccf2150 2190static int kvm_vcpu_ioctl_run(struct kvm_vcpu *vcpu, struct kvm_run *kvm_run)
6aa8b732 2191{
6aa8b732 2192 int r;
1961d276 2193 sigset_t sigsaved;
6aa8b732 2194
bccf2150 2195 vcpu_load(vcpu);
6aa8b732 2196
c5ec1534
HQ		 2197 if (unlikely(vcpu->mp_state == VCPU_MP_STATE_UNINITIALIZED)) {
2198 kvm_vcpu_block(vcpu);
2199 vcpu_put(vcpu);
2200 return -EAGAIN;
2201 }
2202
1961d276
AK		 2203 if (vcpu->sigset_active)
2204 sigprocmask(SIG_SETMASK, &vcpu->sigset, &sigsaved);
2205
54810342 2206 /* re-sync apic's tpr */
5cd4f6fd
HQ		 2207 if (!irqchip_in_kernel(vcpu->kvm))
2208 set_cr8(vcpu, kvm_run->cr8);
54810342 2209
02c83209
AK		 2210 if (vcpu->pio.cur_count) {
2211 r = complete_pio(vcpu);
2212 if (r)
2213 goto out;
2214 }
2215
2216 if (vcpu->mmio_needed) {
2217 memcpy(vcpu->mmio_data, kvm_run->mmio.data, 8);
2218 vcpu->mmio_read_completed = 1;
2219 vcpu->mmio_needed = 0;
2220 r = emulate_instruction(vcpu, kvm_run,
3427318f 2221 vcpu->mmio_fault_cr2, 0, 1);
02c83209
AK		 2222 if (r == EMULATE_DO_MMIO) {
2223 /*
2224 * Read-modify-write. Back to userspace.
2225 */
02c83209
AK		 2226 r = 0;
2227 goto out;
46fc1477 2228 }
6aa8b732
AK		 2229 }
2230
8eb7d334 2231 if (kvm_run->exit_reason == KVM_EXIT_HYPERCALL) {
cbdd1bea 2232 kvm_x86_ops->cache_regs(vcpu);
b4e63f56 2233 vcpu->regs[VCPU_REGS_RAX] = kvm_run->hypercall.ret;
cbdd1bea 2234 kvm_x86_ops->decache_regs(vcpu);
b4e63f56
AK		 2235 }
2236
04d2cc77 2237 r = __vcpu_run(vcpu, kvm_run);
6aa8b732 2238
039576c0 2239out:
1961d276
AK		 2240 if (vcpu->sigset_active)
2241 sigprocmask(SIG_SETMASK, &sigsaved, NULL);
2242
6aa8b732
AK		 2243 vcpu_put(vcpu);
2244 return r;
2245}
2246
bccf2150
AK		 2247static int kvm_vcpu_ioctl_get_regs(struct kvm_vcpu *vcpu,
2248 struct kvm_regs *regs)
6aa8b732 2249{
bccf2150 2250 vcpu_load(vcpu);
6aa8b732 2251
cbdd1bea 2252 kvm_x86_ops->cache_regs(vcpu);
6aa8b732
AK		 2253
2254 regs->rax = vcpu->regs[VCPU_REGS_RAX];
2255 regs->rbx = vcpu->regs[VCPU_REGS_RBX];
2256 regs->rcx = vcpu->regs[VCPU_REGS_RCX];
2257 regs->rdx = vcpu->regs[VCPU_REGS_RDX];
2258 regs->rsi = vcpu->regs[VCPU_REGS_RSI];
2259 regs->rdi = vcpu->regs[VCPU_REGS_RDI];
2260 regs->rsp = vcpu->regs[VCPU_REGS_RSP];
2261 regs->rbp = vcpu->regs[VCPU_REGS_RBP];
05b3e0c2 2262#ifdef CONFIG_X86_64
6aa8b732
AK		 2263 regs->r8 = vcpu->regs[VCPU_REGS_R8];
2264 regs->r9 = vcpu->regs[VCPU_REGS_R9];
2265 regs->r10 = vcpu->regs[VCPU_REGS_R10];
2266 regs->r11 = vcpu->regs[VCPU_REGS_R11];
2267 regs->r12 = vcpu->regs[VCPU_REGS_R12];
2268 regs->r13 = vcpu->regs[VCPU_REGS_R13];
2269 regs->r14 = vcpu->regs[VCPU_REGS_R14];
2270 regs->r15 = vcpu->regs[VCPU_REGS_R15];
2271#endif
2272
2273 regs->rip = vcpu->rip;
cbdd1bea 2274 regs->rflags = kvm_x86_ops->get_rflags(vcpu);
6aa8b732
AK		 2275
2276 /*
2277 * Don't leak debug flags in case they were set for guest debugging
2278 */
2279 if (vcpu->guest_debug.enabled && vcpu->guest_debug.singlestep)
2280 regs->rflags &= ~(X86_EFLAGS_TF | X86_EFLAGS_RF);
2281
2282 vcpu_put(vcpu);
2283
2284 return 0;
2285}
2286
bccf2150
AK		 2287static int kvm_vcpu_ioctl_set_regs(struct kvm_vcpu *vcpu,
2288 struct kvm_regs *regs)
6aa8b732 2289{
bccf2150 2290 vcpu_load(vcpu);
6aa8b732
AK		 2291
2292 vcpu->regs[VCPU_REGS_RAX] = regs->rax;
2293 vcpu->regs[VCPU_REGS_RBX] = regs->rbx;
2294 vcpu->regs[VCPU_REGS_RCX] = regs->rcx;
2295 vcpu->regs[VCPU_REGS_RDX] = regs->rdx;
2296 vcpu->regs[VCPU_REGS_RSI] = regs->rsi;
2297 vcpu->regs[VCPU_REGS_RDI] = regs->rdi;
2298 vcpu->regs[VCPU_REGS_RSP] = regs->rsp;
2299 vcpu->regs[VCPU_REGS_RBP] = regs->rbp;
05b3e0c2 2300#ifdef CONFIG_X86_64
6aa8b732
AK		 2301 vcpu->regs[VCPU_REGS_R8] = regs->r8;
2302 vcpu->regs[VCPU_REGS_R9] = regs->r9;
2303 vcpu->regs[VCPU_REGS_R10] = regs->r10;
2304 vcpu->regs[VCPU_REGS_R11] = regs->r11;
2305 vcpu->regs[VCPU_REGS_R12] = regs->r12;
2306 vcpu->regs[VCPU_REGS_R13] = regs->r13;
2307 vcpu->regs[VCPU_REGS_R14] = regs->r14;
2308 vcpu->regs[VCPU_REGS_R15] = regs->r15;
2309#endif
2310
2311 vcpu->rip = regs->rip;
cbdd1bea 2312 kvm_x86_ops->set_rflags(vcpu, regs->rflags);
6aa8b732 2313
cbdd1bea 2314 kvm_x86_ops->decache_regs(vcpu);
6aa8b732
AK		 2315
2316 vcpu_put(vcpu);
2317
2318 return 0;
2319}
2320
2321static void get_segment(struct kvm_vcpu *vcpu,
2322 struct kvm_segment *var, int seg)
2323{
cbdd1bea 2324 return kvm_x86_ops->get_segment(vcpu, var, seg);
6aa8b732
AK		 2325}
2326
bccf2150
AK		 2327static int kvm_vcpu_ioctl_get_sregs(struct kvm_vcpu *vcpu,
2328 struct kvm_sregs *sregs)
6aa8b732 2329{
6aa8b732 2330 struct descriptor_table dt;
2a8067f1 2331 int pending_vec;
6aa8b732 2332
bccf2150 2333 vcpu_load(vcpu);
6aa8b732
AK		 2334
2335 get_segment(vcpu, &sregs->cs, VCPU_SREG_CS);
2336 get_segment(vcpu, &sregs->ds, VCPU_SREG_DS);
2337 get_segment(vcpu, &sregs->es, VCPU_SREG_ES);
2338 get_segment(vcpu, &sregs->fs, VCPU_SREG_FS);
2339 get_segment(vcpu, &sregs->gs, VCPU_SREG_GS);
2340 get_segment(vcpu, &sregs->ss, VCPU_SREG_SS);
2341
2342 get_segment(vcpu, &sregs->tr, VCPU_SREG_TR);
2343 get_segment(vcpu, &sregs->ldt, VCPU_SREG_LDTR);
2344
cbdd1bea 2345 kvm_x86_ops->get_idt(vcpu, &dt);
6aa8b732
AK		 2346 sregs->idt.limit = dt.limit;
2347 sregs->idt.base = dt.base;
cbdd1bea 2348 kvm_x86_ops->get_gdt(vcpu, &dt);
6aa8b732
AK		 2349 sregs->gdt.limit = dt.limit;
2350 sregs->gdt.base = dt.base;
2351
cbdd1bea 2352 kvm_x86_ops->decache_cr4_guest_bits(vcpu);
6aa8b732
AK		 2353 sregs->cr0 = vcpu->cr0;
2354 sregs->cr2 = vcpu->cr2;
2355 sregs->cr3 = vcpu->cr3;
2356 sregs->cr4 = vcpu->cr4;
7017fc3d 2357 sregs->cr8 = get_cr8(vcpu);
6aa8b732 2358 sregs->efer = vcpu->shadow_efer;
7017fc3d 2359 sregs->apic_base = kvm_get_apic_base(vcpu);
6aa8b732 2360
2a8067f1 2361 if (irqchip_in_kernel(vcpu->kvm)) {
c52fb35a
HQ		 2362 memset(sregs->interrupt_bitmap, 0,
2363 sizeof sregs->interrupt_bitmap);
cbdd1bea 2364 pending_vec = kvm_x86_ops->get_irq(vcpu);
2a8067f1 2365 if (pending_vec >= 0)
d77c26fc
MD		 2366 set_bit(pending_vec,
2367 (unsigned long *)sregs->interrupt_bitmap);
2a8067f1 2368 } else
c52fb35a
HQ		 2369 memcpy(sregs->interrupt_bitmap, vcpu->irq_pending,
2370 sizeof sregs->interrupt_bitmap);
6aa8b732
AK		 2371
2372 vcpu_put(vcpu);
2373
2374 return 0;
2375}
2376
2377static void set_segment(struct kvm_vcpu *vcpu,
2378 struct kvm_segment *var, int seg)
2379{
cbdd1bea 2380 return kvm_x86_ops->set_segment(vcpu, var, seg);
6aa8b732
AK		 2381}
2382
bccf2150
AK		 2383static int kvm_vcpu_ioctl_set_sregs(struct kvm_vcpu *vcpu,
2384 struct kvm_sregs *sregs)
6aa8b732 2385{
6aa8b732 2386 int mmu_reset_needed = 0;
2a8067f1 2387 int i, pending_vec, max_bits;
6aa8b732
AK		 2388 struct descriptor_table dt;
2389
bccf2150 2390 vcpu_load(vcpu);
6aa8b732 2391
6aa8b732
AK		 2392 dt.limit = sregs->idt.limit;
2393 dt.base = sregs->idt.base;
cbdd1bea 2394 kvm_x86_ops->set_idt(vcpu, &dt);
6aa8b732
AK		 2395 dt.limit = sregs->gdt.limit;
2396 dt.base = sregs->gdt.base;
cbdd1bea 2397 kvm_x86_ops->set_gdt(vcpu, &dt);
6aa8b732
AK		 2398
2399 vcpu->cr2 = sregs->cr2;
2400 mmu_reset_needed |= vcpu->cr3 != sregs->cr3;
2401 vcpu->cr3 = sregs->cr3;
2402
7017fc3d 2403 set_cr8(vcpu, sregs->cr8);
6aa8b732
AK		 2404
2405 mmu_reset_needed |= vcpu->shadow_efer != sregs->efer;
05b3e0c2 2406#ifdef CONFIG_X86_64
cbdd1bea 2407 kvm_x86_ops->set_efer(vcpu, sregs->efer);
6aa8b732 2408#endif
7017fc3d 2409 kvm_set_apic_base(vcpu, sregs->apic_base);
6aa8b732 2410
cbdd1bea 2411 kvm_x86_ops->decache_cr4_guest_bits(vcpu);
399badf3 2412
6aa8b732 2413 mmu_reset_needed |= vcpu->cr0 != sregs->cr0;
81f50e3b 2414 vcpu->cr0 = sregs->cr0;
cbdd1bea 2415 kvm_x86_ops->set_cr0(vcpu, sregs->cr0);
6aa8b732
AK		 2416
2417 mmu_reset_needed |= vcpu->cr4 != sregs->cr4;
cbdd1bea 2418 kvm_x86_ops->set_cr4(vcpu, sregs->cr4);
1b0973bd
AK		 2419 if (!is_long_mode(vcpu) && is_pae(vcpu))
2420 load_pdptrs(vcpu, vcpu->cr3);
6aa8b732
AK		 2421
2422 if (mmu_reset_needed)
2423 kvm_mmu_reset_context(vcpu);
2424
c52fb35a
HQ		 2425 if (!irqchip_in_kernel(vcpu->kvm)) {
2426 memcpy(vcpu->irq_pending, sregs->interrupt_bitmap,
2427 sizeof vcpu->irq_pending);
2428 vcpu->irq_summary = 0;
2429 for (i = 0; i < ARRAY_SIZE(vcpu->irq_pending); ++i)
2430 if (vcpu->irq_pending[i])
2431 __set_bit(i, &vcpu->irq_summary);
2a8067f1
ED		 2432 } else {
2433 max_bits = (sizeof sregs->interrupt_bitmap) << 3;
2434 pending_vec = find_first_bit(
2435 (const unsigned long *)sregs->interrupt_bitmap,
2436 max_bits);
2437 /* Only pending external irq is handled here */
2438 if (pending_vec < max_bits) {
cbdd1bea 2439 kvm_x86_ops->set_irq(vcpu, pending_vec);
d77c26fc
MD		 2440 pr_debug("Set back pending irq %d\n",
2441 pending_vec);
2a8067f1 2442 }
c52fb35a 2443 }
6aa8b732 2444
024aa1c0
AK		 2445 set_segment(vcpu, &sregs->cs, VCPU_SREG_CS);
2446 set_segment(vcpu, &sregs->ds, VCPU_SREG_DS);
2447 set_segment(vcpu, &sregs->es, VCPU_SREG_ES);
2448 set_segment(vcpu, &sregs->fs, VCPU_SREG_FS);
2449 set_segment(vcpu, &sregs->gs, VCPU_SREG_GS);
2450 set_segment(vcpu, &sregs->ss, VCPU_SREG_SS);
2451
2452 set_segment(vcpu, &sregs->tr, VCPU_SREG_TR);
2453 set_segment(vcpu, &sregs->ldt, VCPU_SREG_LDTR);
2454
6aa8b732
AK		 2455 vcpu_put(vcpu);
2456
2457 return 0;
2458}
2459
1747fb71
RR		 2460void kvm_get_cs_db_l_bits(struct kvm_vcpu *vcpu, int *db, int *l)
2461{
2462 struct kvm_segment cs;
2463
2464 get_segment(vcpu, &cs, VCPU_SREG_CS);
2465 *db = cs.db;
2466 *l = cs.l;
2467}
2468EXPORT_SYMBOL_GPL(kvm_get_cs_db_l_bits);
2469
6aa8b732
AK		 2470/*
2471 * List of msr numbers which we expose to userspace through KVM_GET_MSRS
2472 * and KVM_SET_MSRS, and KVM_GET_MSR_INDEX_LIST.
bf591b24
MR		 2473 *
2474 * This list is modified at module load time to reflect the
2475 * capabilities of the host cpu.
6aa8b732
AK		 2476 */
2477static u32 msrs_to_save[] = {
2478 MSR_IA32_SYSENTER_CS, MSR_IA32_SYSENTER_ESP, MSR_IA32_SYSENTER_EIP,
2479 MSR_K6_STAR,
05b3e0c2 2480#ifdef CONFIG_X86_64
6aa8b732
AK		 2481 MSR_CSTAR, MSR_KERNEL_GS_BASE, MSR_SYSCALL_MASK, MSR_LSTAR,
2482#endif
2483 MSR_IA32_TIME_STAMP_COUNTER,
2484};
2485
bf591b24
MR		 2486static unsigned num_msrs_to_save;
2487
6f00e68f
AK		 2488static u32 emulated_msrs[] = {
2489 MSR_IA32_MISC_ENABLE,
2490};
2491
bf591b24
MR		 2492static __init void kvm_init_msr_list(void)
2493{
2494 u32 dummy[2];
2495 unsigned i, j;
2496
2497 for (i = j = 0; i < ARRAY_SIZE(msrs_to_save); i++) {
2498 if (rdmsr_safe(msrs_to_save[i], &dummy[0], &dummy[1]) < 0)
2499 continue;
2500 if (j < i)
2501 msrs_to_save[j] = msrs_to_save[i];
2502 j++;
2503 }
2504 num_msrs_to_save = j;
2505}
6aa8b732
AK		 2506
2507/*
2508 * Adapt set_msr() to msr_io()'s calling convention
2509 */
2510static int do_set_msr(struct kvm_vcpu *vcpu, unsigned index, u64 *data)
2511{
35f3f286 2512 return kvm_set_msr(vcpu, index, *data);
6aa8b732
AK		 2513}
2514
2515/*
2516 * Read or write a bunch of msrs. All parameters are kernel addresses.
2517 *
2518 * @return number of msrs set successfully.
2519 */
bccf2150 2520static int __msr_io(struct kvm_vcpu *vcpu, struct kvm_msrs *msrs,
6aa8b732
AK		 2521 struct kvm_msr_entry *entries,
2522 int (*do_msr)(struct kvm_vcpu *vcpu,
2523 unsigned index, u64 *data))
2524{
6aa8b732
AK		 2525 int i;
2526
bccf2150 2527 vcpu_load(vcpu);
6aa8b732
AK		 2528
2529 for (i = 0; i < msrs->nmsrs; ++i)
2530 if (do_msr(vcpu, entries[i].index, &entries[i].data))
2531 break;
2532
2533 vcpu_put(vcpu);
2534
2535 return i;
2536}
2537
2538/*
2539 * Read or write a bunch of msrs. Parameters are user addresses.
2540 *
2541 * @return number of msrs set successfully.
2542 */
bccf2150 2543static int msr_io(struct kvm_vcpu *vcpu, struct kvm_msrs __user *user_msrs,
6aa8b732
AK		 2544 int (*do_msr)(struct kvm_vcpu *vcpu,
2545 unsigned index, u64 *data),
2546 int writeback)
2547{
2548 struct kvm_msrs msrs;
2549 struct kvm_msr_entry *entries;
2550 int r, n;
2551 unsigned size;
2552
2553 r = -EFAULT;
2554 if (copy_from_user(&msrs, user_msrs, sizeof msrs))
2555 goto out;
2556
2557 r = -E2BIG;
2558 if (msrs.nmsrs >= MAX_IO_MSRS)
2559 goto out;
2560
2561 r = -ENOMEM;
2562 size = sizeof(struct kvm_msr_entry) * msrs.nmsrs;
2563 entries = vmalloc(size);
2564 if (!entries)
2565 goto out;
2566
2567 r = -EFAULT;
2568 if (copy_from_user(entries, user_msrs->entries, size))
2569 goto out_free;
2570
bccf2150 2571 r = n = __msr_io(vcpu, &msrs, entries, do_msr);
6aa8b732
AK		 2572 if (r < 0)
2573 goto out_free;
2574
2575 r = -EFAULT;
2576 if (writeback && copy_to_user(user_msrs->entries, entries, size))
2577 goto out_free;
2578
2579 r = n;
2580
2581out_free:
2582 vfree(entries);
2583out:
2584 return r;
2585}
2586
2587/*
2588 * Translate a guest virtual address to a guest physical address.
2589 */
bccf2150
AK		 2590static int kvm_vcpu_ioctl_translate(struct kvm_vcpu *vcpu,
2591 struct kvm_translation *tr)
6aa8b732
AK		 2592{
2593 unsigned long vaddr = tr->linear_address;
6aa8b732
AK		 2594 gpa_t gpa;
2595
bccf2150 2596 vcpu_load(vcpu);
11ec2804 2597 mutex_lock(&vcpu->kvm->lock);
6aa8b732
AK		 2598 gpa = vcpu->mmu.gva_to_gpa(vcpu, vaddr);
2599 tr->physical_address = gpa;
2600 tr->valid = gpa != UNMAPPED_GVA;
2601 tr->writeable = 1;
2602 tr->usermode = 0;
11ec2804 2603 mutex_unlock(&vcpu->kvm->lock);
6aa8b732
AK		 2604 vcpu_put(vcpu);
2605
2606 return 0;
2607}
2608
bccf2150
AK		 2609static int kvm_vcpu_ioctl_interrupt(struct kvm_vcpu *vcpu,
2610 struct kvm_interrupt *irq)
6aa8b732 2611{
6aa8b732
AK		 2612 if (irq->irq < 0 || irq->irq >= 256)
2613 return -EINVAL;
97222cc8
ED		 2614 if (irqchip_in_kernel(vcpu->kvm))
2615 return -ENXIO;
bccf2150 2616 vcpu_load(vcpu);
6aa8b732
AK		 2617
2618 set_bit(irq->irq, vcpu->irq_pending);
2619 set_bit(irq->irq / BITS_PER_LONG, &vcpu->irq_summary);
2620
2621 vcpu_put(vcpu);
2622
2623 return 0;
2624}
2625
bccf2150
AK		 2626static int kvm_vcpu_ioctl_debug_guest(struct kvm_vcpu *vcpu,
2627 struct kvm_debug_guest *dbg)
6aa8b732 2628{
6aa8b732
AK		 2629 int r;
2630
bccf2150 2631 vcpu_load(vcpu);
6aa8b732 2632
cbdd1bea 2633 r = kvm_x86_ops->set_guest_debug(vcpu, dbg);
6aa8b732
AK		 2634
2635 vcpu_put(vcpu);
2636
2637 return r;
2638}
2639
9a2bb7f4
AK		 2640static struct page *kvm_vcpu_nopage(struct vm_area_struct *vma,
2641 unsigned long address,
2642 int *type)
2643{
2644 struct kvm_vcpu *vcpu = vma->vm_file->private_data;
2645 unsigned long pgoff;
2646 struct page *page;
2647
9a2bb7f4 2648 pgoff = ((address - vma->vm_start) >> PAGE_SHIFT) + vma->vm_pgoff;
039576c0
AK		 2649 if (pgoff == 0)
2650 page = virt_to_page(vcpu->run);
2651 else if (pgoff == KVM_PIO_PAGE_OFFSET)
2652 page = virt_to_page(vcpu->pio_data);
2653 else
9a2bb7f4 2654 return NOPAGE_SIGBUS;
9a2bb7f4 2655 get_page(page);
cd0d9137
NAQ		 2656 if (type != NULL)
2657 *type = VM_FAULT_MINOR;
2658
9a2bb7f4
AK		 2659 return page;
2660}
2661
2662static struct vm_operations_struct kvm_vcpu_vm_ops = {
2663 .nopage = kvm_vcpu_nopage,
2664};
2665
2666static int kvm_vcpu_mmap(struct file *file, struct vm_area_struct *vma)
2667{
2668 vma->vm_ops = &kvm_vcpu_vm_ops;
2669 return 0;
2670}
2671
bccf2150
AK		 2672static int kvm_vcpu_release(struct inode *inode, struct file *filp)
2673{
2674 struct kvm_vcpu *vcpu = filp->private_data;
2675
2676 fput(vcpu->kvm->filp);
2677 return 0;
2678}
2679
2680static struct file_operations kvm_vcpu_fops = {
2681 .release = kvm_vcpu_release,
2682 .unlocked_ioctl = kvm_vcpu_ioctl,
2683 .compat_ioctl = kvm_vcpu_ioctl,
9a2bb7f4 2684 .mmap = kvm_vcpu_mmap,
bccf2150
AK		 2685};
2686
2687/*
2688 * Allocates an inode for the vcpu.
2689 */
2690static int create_vcpu_fd(struct kvm_vcpu *vcpu)
2691{
2692 int fd, r;
2693 struct inode *inode;
2694 struct file *file;
2695
d6d28168
AK		 2696 r = anon_inode_getfd(&fd, &inode, &file,
2697 "kvm-vcpu", &kvm_vcpu_fops, vcpu);
2698 if (r)
2699 return r;
bccf2150 2700 atomic_inc(&vcpu->kvm->filp->f_count);
bccf2150 2701 return fd;
bccf2150
AK		 2702}
2703
c5ea7660
AK		 2704/*
2705 * Creates some virtual cpus. Good luck creating more than one.
2706 */
2707static int kvm_vm_ioctl_create_vcpu(struct kvm *kvm, int n)
2708{
2709 int r;
2710 struct kvm_vcpu *vcpu;
2711
c5ea7660 2712 if (!valid_vcpu(n))
fb3f0f51 2713 return -EINVAL;
c5ea7660 2714
cbdd1bea 2715 vcpu = kvm_x86_ops->vcpu_create(kvm, n);
fb3f0f51
RR		 2716 if (IS_ERR(vcpu))
2717 return PTR_ERR(vcpu);
c5ea7660 2718
15ad7146
AK		 2719 preempt_notifier_init(&vcpu->preempt_notifier, &kvm_preempt_ops);
2720
b114b080
RR		 2721 /* We do fxsave: this must be aligned. */
2722 BUG_ON((unsigned long)&vcpu->host_fx_image & 0xF);
2723
fb3f0f51 2724 vcpu_load(vcpu);
c5ea7660 2725 r = kvm_mmu_setup(vcpu);
c5ea7660 2726 vcpu_put(vcpu);
c5ea7660 2727 if (r < 0)
fb3f0f51
RR		 2728 goto free_vcpu;
2729
11ec2804 2730 mutex_lock(&kvm->lock);
fb3f0f51
RR		 2731 if (kvm->vcpus[n]) {
2732 r = -EEXIST;
11ec2804 2733 mutex_unlock(&kvm->lock);
fb3f0f51
RR		 2734 goto mmu_unload;
2735 }
2736 kvm->vcpus[n] = vcpu;
11ec2804 2737 mutex_unlock(&kvm->lock);
c5ea7660 2738
fb3f0f51 2739 /* Now it's all set up, let userspace reach it */
bccf2150
AK		 2740 r = create_vcpu_fd(vcpu);
2741 if (r < 0)
fb3f0f51
RR		 2742 goto unlink;
2743 return r;
39c3b86e 2744
fb3f0f51 2745unlink:
11ec2804 2746 mutex_lock(&kvm->lock);
fb3f0f51 2747 kvm->vcpus[n] = NULL;
11ec2804 2748 mutex_unlock(&kvm->lock);
a2fa3e9f 2749
fb3f0f51
RR		 2750mmu_unload:
2751 vcpu_load(vcpu);
2752 kvm_mmu_unload(vcpu);
2753 vcpu_put(vcpu);
c5ea7660 2754
fb3f0f51 2755free_vcpu:
cbdd1bea 2756 kvm_x86_ops->vcpu_free(vcpu);
c5ea7660
AK		 2757 return r;
2758}
2759
2cc51560
ED		 2760static void cpuid_fix_nx_cap(struct kvm_vcpu *vcpu)
2761{
2762 u64 efer;
2763 int i;
2764 struct kvm_cpuid_entry *e, *entry;
2765
2766 rdmsrl(MSR_EFER, efer);
2767 entry = NULL;
2768 for (i = 0; i < vcpu->cpuid_nent; ++i) {
2769 e = &vcpu->cpuid_entries[i];
2770 if (e->function == 0x80000001) {
2771 entry = e;
2772 break;
2773 }
2774 }
4c981b43 2775 if (entry && (entry->edx & (1 << 20)) && !(efer & EFER_NX)) {
2cc51560 2776 entry->edx &= ~(1 << 20);
4c981b43 2777 printk(KERN_INFO "kvm: guest NX capability removed\n");
2cc51560
ED		 2778 }
2779}
2780
06465c5a
AK		 2781static int kvm_vcpu_ioctl_set_cpuid(struct kvm_vcpu *vcpu,
2782 struct kvm_cpuid *cpuid,
2783 struct kvm_cpuid_entry __user *entries)
2784{
2785 int r;
2786
2787 r = -E2BIG;
2788 if (cpuid->nent > KVM_MAX_CPUID_ENTRIES)
2789 goto out;
2790 r = -EFAULT;
2791 if (copy_from_user(&vcpu->cpuid_entries, entries,
2792 cpuid->nent * sizeof(struct kvm_cpuid_entry)))
2793 goto out;
2794 vcpu->cpuid_nent = cpuid->nent;
2cc51560 2795 cpuid_fix_nx_cap(vcpu);
06465c5a
AK		 2796 return 0;
2797
2798out:
2799 return r;
2800}
2801
1961d276
AK		 2802static int kvm_vcpu_ioctl_set_sigmask(struct kvm_vcpu *vcpu, sigset_t *sigset)
2803{
2804 if (sigset) {
2805 sigdelsetmask(sigset, sigmask(SIGKILL)|sigmask(SIGSTOP));
2806 vcpu->sigset_active = 1;
2807 vcpu->sigset = *sigset;
2808 } else
2809 vcpu->sigset_active = 0;
2810 return 0;
2811}
2812
b8836737
AK		 2813/*
2814 * fxsave fpu state. Taken from x86_64/processor.h. To be killed when
2815 * we have asm/x86/processor.h
2816 */
2817struct fxsave {
2818 u16 cwd;
2819 u16 swd;
2820 u16 twd;
2821 u16 fop;
2822 u64 rip;
2823 u64 rdp;
2824 u32 mxcsr;
2825 u32 mxcsr_mask;
2826 u32 st_space[32]; /* 8*16 bytes for each FP-reg = 128 bytes */
2827#ifdef CONFIG_X86_64
2828 u32 xmm_space[64]; /* 16*16 bytes for each XMM-reg = 256 bytes */
2829#else
2830 u32 xmm_space[32]; /* 8*16 bytes for each XMM-reg = 128 bytes */
2831#endif
2832};
2833
2834static int kvm_vcpu_ioctl_get_fpu(struct kvm_vcpu *vcpu, struct kvm_fpu *fpu)
2835{
b114b080 2836 struct fxsave *fxsave = (struct fxsave *)&vcpu->guest_fx_image;
b8836737
AK		 2837
2838 vcpu_load(vcpu);
2839
2840 memcpy(fpu->fpr, fxsave->st_space, 128);
2841 fpu->fcw = fxsave->cwd;
2842 fpu->fsw = fxsave->swd;
2843 fpu->ftwx = fxsave->twd;
2844 fpu->last_opcode = fxsave->fop;
2845 fpu->last_ip = fxsave->rip;
2846 fpu->last_dp = fxsave->rdp;
2847 memcpy(fpu->xmm, fxsave->xmm_space, sizeof fxsave->xmm_space);
2848
2849 vcpu_put(vcpu);
2850
2851 return 0;
2852}
2853
2854static int kvm_vcpu_ioctl_set_fpu(struct kvm_vcpu *vcpu, struct kvm_fpu *fpu)
2855{
b114b080 2856 struct fxsave *fxsave = (struct fxsave *)&vcpu->guest_fx_image;
b8836737
AK		 2857
2858 vcpu_load(vcpu);
2859
2860 memcpy(fxsave->st_space, fpu->fpr, 128);
2861 fxsave->cwd = fpu->fcw;
2862 fxsave->swd = fpu->fsw;
2863 fxsave->twd = fpu->ftwx;
2864 fxsave->fop = fpu->last_opcode;
2865 fxsave->rip = fpu->last_ip;
2866 fxsave->rdp = fpu->last_dp;
2867 memcpy(fxsave->xmm_space, fpu->xmm, sizeof fxsave->xmm_space);
2868
2869 vcpu_put(vcpu);
2870
2871 return 0;
2872}
2873
96ad2cc6
ED		 2874static int kvm_vcpu_ioctl_get_lapic(struct kvm_vcpu *vcpu,
2875 struct kvm_lapic_state *s)
2876{
2877 vcpu_load(vcpu);
2878 memcpy(s->regs, vcpu->apic->regs, sizeof *s);
2879 vcpu_put(vcpu);
2880
2881 return 0;
2882}
2883
2884static int kvm_vcpu_ioctl_set_lapic(struct kvm_vcpu *vcpu,
2885 struct kvm_lapic_state *s)
2886{
2887 vcpu_load(vcpu);
2888 memcpy(vcpu->apic->regs, s->regs, sizeof *s);
2889 kvm_apic_post_state_restore(vcpu);
2890 vcpu_put(vcpu);
2891
2892 return 0;
2893}
2894
bccf2150
AK		 2895static long kvm_vcpu_ioctl(struct file *filp,
2896 unsigned int ioctl, unsigned long arg)
6aa8b732 2897{
bccf2150 2898 struct kvm_vcpu *vcpu = filp->private_data;
2f366987 2899 void __user *argp = (void __user *)arg;
6aa8b732
AK		 2900 int r = -EINVAL;
2901
2902 switch (ioctl) {
9a2bb7f4 2903 case KVM_RUN:
f0fe5108
AK		 2904 r = -EINVAL;
2905 if (arg)
2906 goto out;
9a2bb7f4 2907 r = kvm_vcpu_ioctl_run(vcpu, vcpu->run);
6aa8b732 2908 break;
6aa8b732
AK		 2909 case KVM_GET_REGS: {
2910 struct kvm_regs kvm_regs;
2911
bccf2150
AK		 2912 memset(&kvm_regs, 0, sizeof kvm_regs);
2913 r = kvm_vcpu_ioctl_get_regs(vcpu, &kvm_regs);
6aa8b732
AK		 2914 if (r)
2915 goto out;
2916 r = -EFAULT;
2f366987 2917 if (copy_to_user(argp, &kvm_regs, sizeof kvm_regs))
6aa8b732
AK		 2918 goto out;
2919 r = 0;
2920 break;
2921 }
2922 case KVM_SET_REGS: {
2923 struct kvm_regs kvm_regs;
2924
2925 r = -EFAULT;
2f366987 2926 if (copy_from_user(&kvm_regs, argp, sizeof kvm_regs))
6aa8b732 2927 goto out;
bccf2150 2928 r = kvm_vcpu_ioctl_set_regs(vcpu, &kvm_regs);
6aa8b732
AK		 2929 if (r)
2930 goto out;
2931 r = 0;
2932 break;
2933 }
2934 case KVM_GET_SREGS: {
2935 struct kvm_sregs kvm_sregs;
2936
bccf2150
AK		 2937 memset(&kvm_sregs, 0, sizeof kvm_sregs);
2938 r = kvm_vcpu_ioctl_get_sregs(vcpu, &kvm_sregs);
6aa8b732
AK		 2939 if (r)
2940 goto out;
2941 r = -EFAULT;
2f366987 2942 if (copy_to_user(argp, &kvm_sregs, sizeof kvm_sregs))
6aa8b732
AK		 2943 goto out;
2944 r = 0;
2945 break;
2946 }
2947 case KVM_SET_SREGS: {
2948 struct kvm_sregs kvm_sregs;
2949
2950 r = -EFAULT;
2f366987 2951 if (copy_from_user(&kvm_sregs, argp, sizeof kvm_sregs))
6aa8b732 2952 goto out;
bccf2150 2953 r = kvm_vcpu_ioctl_set_sregs(vcpu, &kvm_sregs);
6aa8b732
AK		 2954 if (r)
2955 goto out;
2956 r = 0;
2957 break;
2958 }
2959 case KVM_TRANSLATE: {
2960 struct kvm_translation tr;
2961
2962 r = -EFAULT;
2f366987 2963 if (copy_from_user(&tr, argp, sizeof tr))
6aa8b732 2964 goto out;
bccf2150 2965 r = kvm_vcpu_ioctl_translate(vcpu, &tr);
6aa8b732
AK		 2966 if (r)
2967 goto out;
2968 r = -EFAULT;
2f366987 2969 if (copy_to_user(argp, &tr, sizeof tr))
6aa8b732
AK		 2970 goto out;
2971 r = 0;
2972 break;
2973 }
2974 case KVM_INTERRUPT: {
2975 struct kvm_interrupt irq;
2976
2977 r = -EFAULT;
2f366987 2978 if (copy_from_user(&irq, argp, sizeof irq))
6aa8b732 2979 goto out;
bccf2150 2980 r = kvm_vcpu_ioctl_interrupt(vcpu, &irq);
6aa8b732
AK		 2981 if (r)
2982 goto out;
2983 r = 0;
2984 break;
2985 }
2986 case KVM_DEBUG_GUEST: {
2987 struct kvm_debug_guest dbg;
2988
2989 r = -EFAULT;
2f366987 2990 if (copy_from_user(&dbg, argp, sizeof dbg))
6aa8b732 2991 goto out;
bccf2150 2992 r = kvm_vcpu_ioctl_debug_guest(vcpu, &dbg);
6aa8b732
AK		 2993 if (r)
2994 goto out;
2995 r = 0;
2996 break;
2997 }
bccf2150 2998 case KVM_GET_MSRS:
35f3f286 2999 r = msr_io(vcpu, argp, kvm_get_msr, 1);
bccf2150
AK		 3000 break;
3001 case KVM_SET_MSRS:
3002 r = msr_io(vcpu, argp, do_set_msr, 0);
3003 break;
06465c5a
AK		 3004 case KVM_SET_CPUID: {
3005 struct kvm_cpuid __user *cpuid_arg = argp;
3006 struct kvm_cpuid cpuid;
3007
3008 r = -EFAULT;
3009 if (copy_from_user(&cpuid, cpuid_arg, sizeof cpuid))
3010 goto out;
3011 r = kvm_vcpu_ioctl_set_cpuid(vcpu, &cpuid, cpuid_arg->entries);
3012 if (r)
3013 goto out;
3014 break;
3015 }
1961d276
AK		 3016 case KVM_SET_SIGNAL_MASK: {
3017 struct kvm_signal_mask __user *sigmask_arg = argp;
3018 struct kvm_signal_mask kvm_sigmask;
3019 sigset_t sigset, *p;
3020
3021 p = NULL;
3022 if (argp) {
3023 r = -EFAULT;
3024 if (copy_from_user(&kvm_sigmask, argp,
3025 sizeof kvm_sigmask))
3026 goto out;
3027 r = -EINVAL;
3028 if (kvm_sigmask.len != sizeof sigset)
3029 goto out;
3030 r = -EFAULT;
3031 if (copy_from_user(&sigset, sigmask_arg->sigset,
3032 sizeof sigset))
3033 goto out;
3034 p = &sigset;
3035 }
3036 r = kvm_vcpu_ioctl_set_sigmask(vcpu, &sigset);
3037 break;
3038 }
b8836737
AK		 3039 case KVM_GET_FPU: {
3040 struct kvm_fpu fpu;
3041
3042 memset(&fpu, 0, sizeof fpu);
3043 r = kvm_vcpu_ioctl_get_fpu(vcpu, &fpu);
3044 if (r)
3045 goto out;
3046 r = -EFAULT;
3047 if (copy_to_user(argp, &fpu, sizeof fpu))
3048 goto out;
3049 r = 0;
3050 break;
3051 }
3052 case KVM_SET_FPU: {
3053 struct kvm_fpu fpu;
3054
3055 r = -EFAULT;
3056 if (copy_from_user(&fpu, argp, sizeof fpu))
3057 goto out;
3058 r = kvm_vcpu_ioctl_set_fpu(vcpu, &fpu);
3059 if (r)
3060 goto out;
3061 r = 0;
3062 break;
3063 }
96ad2cc6
ED		 3064 case KVM_GET_LAPIC: {
3065 struct kvm_lapic_state lapic;
3066
3067 memset(&lapic, 0, sizeof lapic);
3068 r = kvm_vcpu_ioctl_get_lapic(vcpu, &lapic);
3069 if (r)
3070 goto out;
3071 r = -EFAULT;
3072 if (copy_to_user(argp, &lapic, sizeof lapic))
3073 goto out;
3074 r = 0;
3075 break;
3076 }
3077 case KVM_SET_LAPIC: {
3078 struct kvm_lapic_state lapic;
3079
3080 r = -EFAULT;
3081 if (copy_from_user(&lapic, argp, sizeof lapic))
3082 goto out;
3083 r = kvm_vcpu_ioctl_set_lapic(vcpu, &lapic);;
3084 if (r)
3085 goto out;
3086 r = 0;
3087 break;
3088 }
bccf2150
AK		 3089 default:
3090 ;
3091 }
3092out:
3093 return r;
3094}
3095
3096static long kvm_vm_ioctl(struct file *filp,
3097 unsigned int ioctl, unsigned long arg)
3098{
3099 struct kvm *kvm = filp->private_data;
3100 void __user *argp = (void __user *)arg;
3101 int r = -EINVAL;
3102
3103 switch (ioctl) {
3104 case KVM_CREATE_VCPU:
3105 r = kvm_vm_ioctl_create_vcpu(kvm, arg);
3106 if (r < 0)
3107 goto out;
3108 break;
6aa8b732
AK		 3109 case KVM_SET_MEMORY_REGION: {
3110 struct kvm_memory_region kvm_mem;
3111
3112 r = -EFAULT;
2f366987 3113 if (copy_from_user(&kvm_mem, argp, sizeof kvm_mem))
6aa8b732 3114 goto out;
2c6f5df9 3115 r = kvm_vm_ioctl_set_memory_region(kvm, &kvm_mem);
6aa8b732
AK		 3116 if (r)
3117 goto out;
3118 break;
3119 }
82ce2c96
IE		 3120 case KVM_SET_NR_MMU_PAGES:
3121 r = kvm_vm_ioctl_set_nr_mmu_pages(kvm, arg);
3122 if (r)
3123 goto out;
3124 break;
3125 case KVM_GET_NR_MMU_PAGES:
3126 r = kvm_vm_ioctl_get_nr_mmu_pages(kvm);
3127 break;
6aa8b732
AK		 3128 case KVM_GET_DIRTY_LOG: {
3129 struct kvm_dirty_log log;
3130
3131 r = -EFAULT;
2f366987 3132 if (copy_from_user(&log, argp, sizeof log))
6aa8b732 3133 goto out;
2c6f5df9 3134 r = kvm_vm_ioctl_get_dirty_log(kvm, &log);
6aa8b732
AK		 3135 if (r)
3136 goto out;
3137 break;
3138 }
e8207547
AK		 3139 case KVM_SET_MEMORY_ALIAS: {
3140 struct kvm_memory_alias alias;
3141
3142 r = -EFAULT;
3143 if (copy_from_user(&alias, argp, sizeof alias))
3144 goto out;
3145 r = kvm_vm_ioctl_set_memory_alias(kvm, &alias);
3146 if (r)
3147 goto out;
3148 break;
3149 }
85f455f7
ED		 3150 case KVM_CREATE_IRQCHIP:
3151 r = -ENOMEM;
3152 kvm->vpic = kvm_create_pic(kvm);
1fd4f2a5
ED		 3153 if (kvm->vpic) {
3154 r = kvm_ioapic_init(kvm);
3155 if (r) {
3156 kfree(kvm->vpic);
3157 kvm->vpic = NULL;
3158 goto out;
3159 }
d77c26fc 3160 } else
85f455f7
ED		 3161 goto out;
3162 break;
3163 case KVM_IRQ_LINE: {
3164 struct kvm_irq_level irq_event;
3165
3166 r = -EFAULT;
3167 if (copy_from_user(&irq_event, argp, sizeof irq_event))
3168 goto out;
3169 if (irqchip_in_kernel(kvm)) {
9cf98828 3170 mutex_lock(&kvm->lock);
85f455f7
ED		 3171 if (irq_event.irq < 16)
3172 kvm_pic_set_irq(pic_irqchip(kvm),
3173 irq_event.irq,
3174 irq_event.level);
1fd4f2a5
ED		 3175 kvm_ioapic_set_irq(kvm->vioapic,
3176 irq_event.irq,
3177 irq_event.level);
9cf98828 3178 mutex_unlock(&kvm->lock);
85f455f7
ED		 3179 r = 0;
3180 }
3181 break;
3182 }
6ceb9d79
HQ		 3183 case KVM_GET_IRQCHIP: {
3184 /* 0: PIC master, 1: PIC slave, 2: IOAPIC */
3185 struct kvm_irqchip chip;
3186
3187 r = -EFAULT;
3188 if (copy_from_user(&chip, argp, sizeof chip))
3189 goto out;
3190 r = -ENXIO;
3191 if (!irqchip_in_kernel(kvm))
3192 goto out;
3193 r = kvm_vm_ioctl_get_irqchip(kvm, &chip);
3194 if (r)
3195 goto out;
3196 r = -EFAULT;
3197 if (copy_to_user(argp, &chip, sizeof chip))
3198 goto out;
3199 r = 0;
3200 break;
3201 }
3202 case KVM_SET_IRQCHIP: {
3203 /* 0: PIC master, 1: PIC slave, 2: IOAPIC */
3204 struct kvm_irqchip chip;
3205
3206 r = -EFAULT;
3207 if (copy_from_user(&chip, argp, sizeof chip))
3208 goto out;
3209 r = -ENXIO;
3210 if (!irqchip_in_kernel(kvm))
3211 goto out;
3212 r = kvm_vm_ioctl_set_irqchip(kvm, &chip);
3213 if (r)
3214 goto out;
3215 r = 0;
3216 break;
3217 }
f17abe9a
AK		 3218 default:
3219 ;
3220 }
3221out:
3222 return r;
3223}
3224
3225static struct page *kvm_vm_nopage(struct vm_area_struct *vma,
3226 unsigned long address,
3227 int *type)
3228{
3229 struct kvm *kvm = vma->vm_file->private_data;
3230 unsigned long pgoff;
f17abe9a
AK		 3231 struct page *page;
3232
f17abe9a 3233 pgoff = ((address - vma->vm_start) >> PAGE_SHIFT) + vma->vm_pgoff;
954bbbc2 3234 page = gfn_to_page(kvm, pgoff);
f17abe9a
AK		 3235 if (!page)
3236 return NOPAGE_SIGBUS;
3237 get_page(page);
cd0d9137
NAQ		 3238 if (type != NULL)
3239 *type = VM_FAULT_MINOR;
3240
f17abe9a
AK		 3241 return page;
3242}
3243
3244static struct vm_operations_struct kvm_vm_vm_ops = {
3245 .nopage = kvm_vm_nopage,
3246};
3247
3248static int kvm_vm_mmap(struct file *file, struct vm_area_struct *vma)
3249{
3250 vma->vm_ops = &kvm_vm_vm_ops;
3251 return 0;
3252}
3253
3254static struct file_operations kvm_vm_fops = {
3255 .release = kvm_vm_release,
3256 .unlocked_ioctl = kvm_vm_ioctl,
3257 .compat_ioctl = kvm_vm_ioctl,
3258 .mmap = kvm_vm_mmap,
3259};
3260
3261static int kvm_dev_ioctl_create_vm(void)
3262{
3263 int fd, r;
3264 struct inode *inode;
3265 struct file *file;
3266 struct kvm *kvm;
3267
f17abe9a 3268 kvm = kvm_create_vm();
d6d28168
AK		 3269 if (IS_ERR(kvm))
3270 return PTR_ERR(kvm);
3271 r = anon_inode_getfd(&fd, &inode, &file, "kvm-vm", &kvm_vm_fops, kvm);
3272 if (r) {
3273 kvm_destroy_vm(kvm);
3274 return r;
f17abe9a
AK		 3275 }
3276
bccf2150 3277 kvm->filp = file;
f17abe9a 3278
f17abe9a 3279 return fd;
f17abe9a
AK		 3280}
3281
3282static long kvm_dev_ioctl(struct file *filp,
3283 unsigned int ioctl, unsigned long arg)
3284{
3285 void __user *argp = (void __user *)arg;
07c45a36 3286 long r = -EINVAL;
f17abe9a
AK		 3287
3288 switch (ioctl) {
3289 case KVM_GET_API_VERSION:
f0fe5108
AK		 3290 r = -EINVAL;
3291 if (arg)
3292 goto out;
f17abe9a
AK		 3293 r = KVM_API_VERSION;
3294 break;
3295 case KVM_CREATE_VM:
f0fe5108
AK		 3296 r = -EINVAL;
3297 if (arg)
3298 goto out;
f17abe9a
AK		 3299 r = kvm_dev_ioctl_create_vm();
3300 break;
6aa8b732 3301 case KVM_GET_MSR_INDEX_LIST: {
2f366987 3302 struct kvm_msr_list __user *user_msr_list = argp;
6aa8b732
AK		 3303 struct kvm_msr_list msr_list;
3304 unsigned n;
3305
3306 r = -EFAULT;
3307 if (copy_from_user(&msr_list, user_msr_list, sizeof msr_list))
3308 goto out;
3309 n = msr_list.nmsrs;
6f00e68f 3310 msr_list.nmsrs = num_msrs_to_save + ARRAY_SIZE(emulated_msrs);
6aa8b732
AK		 3311 if (copy_to_user(user_msr_list, &msr_list, sizeof msr_list))
3312 goto out;
3313 r = -E2BIG;
bf591b24 3314 if (n < num_msrs_to_save)
6aa8b732
AK		 3315 goto out;
3316 r = -EFAULT;
3317 if (copy_to_user(user_msr_list->indices, &msrs_to_save,
bf591b24 3318 num_msrs_to_save * sizeof(u32)))
6aa8b732 3319 goto out;
6f00e68f
AK		 3320 if (copy_to_user(user_msr_list->indices
3321 + num_msrs_to_save * sizeof(u32),
3322 &emulated_msrs,
3323 ARRAY_SIZE(emulated_msrs) * sizeof(u32)))
3324 goto out;
6aa8b732 3325 r = 0;
cc1d8955 3326 break;
6aa8b732 3327 }
85f455f7
ED		 3328 case KVM_CHECK_EXTENSION: {
3329 int ext = (long)argp;
3330
3331 switch (ext) {
3332 case KVM_CAP_IRQCHIP:
b6958ce4 3333 case KVM_CAP_HLT:
82ce2c96 3334 case KVM_CAP_MMU_SHADOW_CACHE_CONTROL:
85f455f7
ED		 3335 r = 1;
3336 break;
3337 default:
3338 r = 0;
3339 break;
3340 }
5d308f45 3341 break;
85f455f7 3342 }
07c45a36
AK		 3343 case KVM_GET_VCPU_MMAP_SIZE:
3344 r = -EINVAL;
3345 if (arg)
3346 goto out;
039576c0 3347 r = 2 * PAGE_SIZE;
07c45a36 3348 break;
6aa8b732
AK		 3349 default:
3350 ;
3351 }
3352out:
3353 return r;
3354}
3355
6aa8b732 3356static struct file_operations kvm_chardev_ops = {
6aa8b732
AK		 3357 .unlocked_ioctl = kvm_dev_ioctl,
3358 .compat_ioctl = kvm_dev_ioctl,
6aa8b732
AK		 3359};
3360
3361static struct miscdevice kvm_dev = {
bbe4432e 3362 KVM_MINOR,
6aa8b732
AK		 3363 "kvm",
3364 &kvm_chardev_ops,
3365};
3366
774c47f1
AK		 3367/*
3368 * Make sure that a cpu that is being hot-unplugged does not have any vcpus
3369 * cached on it.
3370 */
3371static void decache_vcpus_on_cpu(int cpu)
3372{
3373 struct kvm *vm;
3374 struct kvm_vcpu *vcpu;
3375 int i;
3376
3377 spin_lock(&kvm_lock);
11ec2804 3378 list_for_each_entry(vm, &vm_list, vm_list)
774c47f1 3379 for (i = 0; i < KVM_MAX_VCPUS; ++i) {
fb3f0f51
RR		 3380 vcpu = vm->vcpus[i];
3381 if (!vcpu)
3382 continue;
774c47f1
AK		 3383 /*
3384 * If the vcpu is locked, then it is running on some
3385 * other cpu and therefore it is not cached on the
3386 * cpu in question.
3387 *
3388 * If it's not locked, check the last cpu it executed
3389 * on.
3390 */
3391 if (mutex_trylock(&vcpu->mutex)) {
3392 if (vcpu->cpu == cpu) {
cbdd1bea 3393 kvm_x86_ops->vcpu_decache(vcpu);
774c47f1
AK		 3394 vcpu->cpu = -1;
3395 }
3396 mutex_unlock(&vcpu->mutex);
3397 }
3398 }
3399 spin_unlock(&kvm_lock);
3400}
3401
1b6c0168
AK		 3402static void hardware_enable(void *junk)
3403{
3404 int cpu = raw_smp_processor_id();
3405
3406 if (cpu_isset(cpu, cpus_hardware_enabled))
3407 return;
3408 cpu_set(cpu, cpus_hardware_enabled);
cbdd1bea 3409 kvm_x86_ops->hardware_enable(NULL);
1b6c0168
AK		 3410}
3411
3412static void hardware_disable(void *junk)
3413{
3414 int cpu = raw_smp_processor_id();
3415
3416 if (!cpu_isset(cpu, cpus_hardware_enabled))
3417 return;
3418 cpu_clear(cpu, cpus_hardware_enabled);
3419 decache_vcpus_on_cpu(cpu);
cbdd1bea 3420 kvm_x86_ops->hardware_disable(NULL);
1b6c0168
AK		 3421}
3422
774c47f1
AK		 3423static int kvm_cpu_hotplug(struct notifier_block *notifier, unsigned long val,
3424 void *v)
3425{
3426 int cpu = (long)v;
3427
3428 switch (val) {
cec9ad27
AK		 3429 case CPU_DYING:
3430 case CPU_DYING_FROZEN:
6ec8a856
AK		 3431 printk(KERN_INFO "kvm: disabling virtualization on CPU%d\n",
3432 cpu);
3433 hardware_disable(NULL);
3434 break;
774c47f1 3435 case CPU_UP_CANCELED:
8bb78442 3436 case CPU_UP_CANCELED_FROZEN:
43934a38
JK		 3437 printk(KERN_INFO "kvm: disabling virtualization on CPU%d\n",
3438 cpu);
1b6c0168 3439 smp_call_function_single(cpu, hardware_disable, NULL, 0, 1);
774c47f1 3440 break;
43934a38 3441 case CPU_ONLINE:
8bb78442 3442 case CPU_ONLINE_FROZEN:
43934a38
JK		 3443 printk(KERN_INFO "kvm: enabling virtualization on CPU%d\n",
3444 cpu);
1b6c0168 3445 smp_call_function_single(cpu, hardware_enable, NULL, 0, 1);
774c47f1
AK		 3446 break;
3447 }
3448 return NOTIFY_OK;
3449}
3450
9a2b85c6 3451static int kvm_reboot(struct notifier_block *notifier, unsigned long val,
d77c26fc 3452 void *v)
9a2b85c6
RR		 3453{
3454 if (val == SYS_RESTART) {
3455 /*
3456 * Some (well, at least mine) BIOSes hang on reboot if
3457 * in vmx root mode.
3458 */
3459 printk(KERN_INFO "kvm: exiting hardware virtualization\n");
3460 on_each_cpu(hardware_disable, NULL, 0, 1);
3461 }
3462 return NOTIFY_OK;
3463}
3464
3465static struct notifier_block kvm_reboot_notifier = {
3466 .notifier_call = kvm_reboot,
3467 .priority = 0,
3468};
3469
2eeb2e94
GH		 3470void kvm_io_bus_init(struct kvm_io_bus *bus)
3471{
3472 memset(bus, 0, sizeof(*bus));
3473}
3474
3475void kvm_io_bus_destroy(struct kvm_io_bus *bus)
3476{
3477 int i;
3478
3479 for (i = 0; i < bus->dev_count; i++) {
3480 struct kvm_io_device *pos = bus->devs[i];
3481
3482 kvm_iodevice_destructor(pos);
3483 }
3484}
3485
3486struct kvm_io_device *kvm_io_bus_find_dev(struct kvm_io_bus *bus, gpa_t addr)
3487{
3488 int i;
3489
3490 for (i = 0; i < bus->dev_count; i++) {
3491 struct kvm_io_device *pos = bus->devs[i];
3492
3493 if (pos->in_range(pos, addr))
3494 return pos;
3495 }
3496
3497 return NULL;
3498}
3499
3500void kvm_io_bus_register_dev(struct kvm_io_bus *bus, struct kvm_io_device *dev)
3501{
3502 BUG_ON(bus->dev_count > (NR_IOBUS_DEVS-1));
3503
3504 bus->devs[bus->dev_count++] = dev;
3505}
3506
774c47f1
AK		 3507static struct notifier_block kvm_cpu_notifier = {
3508 .notifier_call = kvm_cpu_hotplug,
3509 .priority = 20, /* must be > scheduler priority */
3510};
3511
1165f5fe
AK		 3512static u64 stat_get(void *_offset)
3513{
3514 unsigned offset = (long)_offset;
3515 u64 total = 0;
3516 struct kvm *kvm;
3517 struct kvm_vcpu *vcpu;
3518 int i;
3519
3520 spin_lock(&kvm_lock);
3521 list_for_each_entry(kvm, &vm_list, vm_list)
3522 for (i = 0; i < KVM_MAX_VCPUS; ++i) {
fb3f0f51
RR		 3523 vcpu = kvm->vcpus[i];
3524 if (vcpu)
3525 total += *(u32 *)((void *)vcpu + offset);
1165f5fe
AK		 3526 }
3527 spin_unlock(&kvm_lock);
3528 return total;
3529}
3530
3dea7ca7 3531DEFINE_SIMPLE_ATTRIBUTE(stat_fops, stat_get, NULL, "%llu\n");
1165f5fe 3532
6aa8b732
AK		 3533static __init void kvm_init_debug(void)
3534{
3535 struct kvm_stats_debugfs_item *p;
3536
8b6d44c7 3537 debugfs_dir = debugfs_create_dir("kvm", NULL);
6aa8b732 3538 for (p = debugfs_entries; p->name; ++p)
1165f5fe
AK		 3539 p->dentry = debugfs_create_file(p->name, 0444, debugfs_dir,
3540 (void *)(long)p->offset,
3541 &stat_fops);
6aa8b732
AK		 3542}
3543
3544static void kvm_exit_debug(void)
3545{
3546 struct kvm_stats_debugfs_item *p;
3547
3548 for (p = debugfs_entries; p->name; ++p)
3549 debugfs_remove(p->dentry);
3550 debugfs_remove(debugfs_dir);
3551}
3552
59ae6c6b
AK		 3553static int kvm_suspend(struct sys_device *dev, pm_message_t state)
3554{
4267c41a 3555 hardware_disable(NULL);
59ae6c6b
AK		 3556 return 0;
3557}
3558
3559static int kvm_resume(struct sys_device *dev)
3560{
4267c41a 3561 hardware_enable(NULL);
59ae6c6b
AK		 3562 return 0;
3563}
3564
3565static struct sysdev_class kvm_sysdev_class = {
af5ca3f4 3566 .name = "kvm",
59ae6c6b
AK		 3567 .suspend = kvm_suspend,
3568 .resume = kvm_resume,
3569};
3570
3571static struct sys_device kvm_sysdev = {
3572 .id = 0,
3573 .cls = &kvm_sysdev_class,
3574};
3575
6aa8b732
AK		 3576hpa_t bad_page_address;
3577
15ad7146
AK		 3578static inline
3579struct kvm_vcpu *preempt_notifier_to_vcpu(struct preempt_notifier *pn)
3580{
3581 return container_of(pn, struct kvm_vcpu, preempt_notifier);
3582}
3583
3584static void kvm_sched_in(struct preempt_notifier *pn, int cpu)
3585{
3586 struct kvm_vcpu *vcpu = preempt_notifier_to_vcpu(pn);
3587
cbdd1bea 3588 kvm_x86_ops->vcpu_load(vcpu, cpu);
15ad7146
AK		 3589}
3590
3591static void kvm_sched_out(struct preempt_notifier *pn,
3592 struct task_struct *next)
3593{
3594 struct kvm_vcpu *vcpu = preempt_notifier_to_vcpu(pn);
3595
cbdd1bea 3596 kvm_x86_ops->vcpu_put(vcpu);
15ad7146
AK		 3597}
3598
cbdd1bea 3599int kvm_init_x86(struct kvm_x86_ops *ops, unsigned int vcpu_size,
c16f862d 3600 struct module *module)
6aa8b732
AK		 3601{
3602 int r;
002c7f7c 3603 int cpu;
6aa8b732 3604
cbdd1bea 3605 if (kvm_x86_ops) {
09db28b8
YI		 3606 printk(KERN_ERR "kvm: already loaded the other module\n");
3607 return -EEXIST;
3608 }
3609
e097f35c 3610 if (!ops->cpu_has_kvm_support()) {
6aa8b732
AK		 3611 printk(KERN_ERR "kvm: no hardware support\n");
3612 return -EOPNOTSUPP;
3613 }
e097f35c 3614 if (ops->disabled_by_bios()) {
6aa8b732
AK		 3615 printk(KERN_ERR "kvm: disabled by bios\n");
3616 return -EOPNOTSUPP;
3617 }
3618
cbdd1bea 3619 kvm_x86_ops = ops;
e097f35c 3620
cbdd1bea 3621 r = kvm_x86_ops->hardware_setup();
6aa8b732 3622 if (r < 0)
ca45aaae 3623 goto out;
6aa8b732 3624
002c7f7c
YS		 3625 for_each_online_cpu(cpu) {
3626 smp_call_function_single(cpu,
cbdd1bea 3627 kvm_x86_ops->check_processor_compatibility,
002c7f7c
YS		 3628 &r, 0, 1);
3629 if (r < 0)
3630 goto out_free_0;
3631 }
3632
1b6c0168 3633 on_each_cpu(hardware_enable, NULL, 0, 1);
774c47f1
AK		 3634 r = register_cpu_notifier(&kvm_cpu_notifier);
3635 if (r)
3636 goto out_free_1;
6aa8b732
AK		 3637 register_reboot_notifier(&kvm_reboot_notifier);
3638
59ae6c6b
AK		 3639 r = sysdev_class_register(&kvm_sysdev_class);
3640 if (r)
3641 goto out_free_2;
3642
3643 r = sysdev_register(&kvm_sysdev);
3644 if (r)
3645 goto out_free_3;
3646
c16f862d
RR		 3647 /* A kmem cache lets us meet the alignment requirements of fx_save. */
3648 kvm_vcpu_cache = kmem_cache_create("kvm_vcpu", vcpu_size,
3649 __alignof__(struct kvm_vcpu), 0, 0);
3650 if (!kvm_vcpu_cache) {
3651 r = -ENOMEM;
3652 goto out_free_4;
3653 }
3654
6aa8b732
AK		 3655 kvm_chardev_ops.owner = module;
3656
3657 r = misc_register(&kvm_dev);
3658 if (r) {
d77c26fc 3659 printk(KERN_ERR "kvm: misc device register failed\n");
6aa8b732
AK		 3660 goto out_free;
3661 }
3662
15ad7146
AK		 3663 kvm_preempt_ops.sched_in = kvm_sched_in;
3664 kvm_preempt_ops.sched_out = kvm_sched_out;
3665
c7addb90
AK		 3666 kvm_mmu_set_nonpresent_ptes(0ull, 0ull);
3667
3668 return 0;
6aa8b732
AK		 3669
3670out_free:
c16f862d
RR		 3671 kmem_cache_destroy(kvm_vcpu_cache);
3672out_free_4:
59ae6c6b
AK		 3673 sysdev_unregister(&kvm_sysdev);
3674out_free_3:
3675 sysdev_class_unregister(&kvm_sysdev_class);
3676out_free_2:
6aa8b732 3677 unregister_reboot_notifier(&kvm_reboot_notifier);
774c47f1
AK		 3678 unregister_cpu_notifier(&kvm_cpu_notifier);
3679out_free_1:
1b6c0168 3680 on_each_cpu(hardware_disable, NULL, 0, 1);
002c7f7c 3681out_free_0:
cbdd1bea 3682 kvm_x86_ops->hardware_unsetup();
ca45aaae 3683out:
cbdd1bea 3684 kvm_x86_ops = NULL;
6aa8b732
AK		 3685 return r;
3686}
d77c26fc 3687EXPORT_SYMBOL_GPL(kvm_init_x86);
6aa8b732 3688
cbdd1bea 3689void kvm_exit_x86(void)
6aa8b732
AK		 3690{
3691 misc_deregister(&kvm_dev);
c16f862d 3692 kmem_cache_destroy(kvm_vcpu_cache);
59ae6c6b
AK		 3693 sysdev_unregister(&kvm_sysdev);
3694 sysdev_class_unregister(&kvm_sysdev_class);
6aa8b732 3695 unregister_reboot_notifier(&kvm_reboot_notifier);
59ae6c6b 3696 unregister_cpu_notifier(&kvm_cpu_notifier);
1b6c0168 3697 on_each_cpu(hardware_disable, NULL, 0, 1);
cbdd1bea
CE		 3698 kvm_x86_ops->hardware_unsetup();
3699 kvm_x86_ops = NULL;
6aa8b732 3700}
d77c26fc 3701EXPORT_SYMBOL_GPL(kvm_exit_x86);
6aa8b732
AK		 3702
3703static __init int kvm_init(void)
3704{
3705 static struct page *bad_page;
37e29d90
AK		 3706 int r;
3707
b5a33a75
AK		 3708 r = kvm_mmu_module_init();
3709 if (r)
3710 goto out4;
3711
6aa8b732
AK		 3712 kvm_init_debug();
3713
bf591b24
MR		 3714 kvm_init_msr_list();
3715
d77c26fc
MD		 3716 bad_page = alloc_page(GFP_KERNEL);
3717
3718 if (bad_page == NULL) {
6aa8b732
AK		 3719 r = -ENOMEM;
3720 goto out;
3721 }
3722
3723 bad_page_address = page_to_pfn(bad_page) << PAGE_SHIFT;
3724 memset(__va(bad_page_address), 0, PAGE_SIZE);
3725
58e690e6 3726 return 0;
6aa8b732
AK		 3727
3728out:
3729 kvm_exit_debug();
b5a33a75
AK		 3730 kvm_mmu_module_exit();
3731out4:
6aa8b732
AK		 3732 return r;
3733}
3734
3735static __exit void kvm_exit(void)
3736{
3737 kvm_exit_debug();
3738 __free_page(pfn_to_page(bad_page_address >> PAGE_SHIFT));
b5a33a75 3739 kvm_mmu_module_exit();
6aa8b732
AK		 3740}
3741
3742module_init(kvm_init)
3743module_exit(kvm_exit)
Clone of davem's net-next-2.6 tree