]> git.proxmox.com Git - mirror_ubuntu-artful-kernel.git/blame - drivers/kvm/kvm_main.c
projects / mirror_ubuntu-artful-kernel.git / blame
? search:
summary | shortlog | log | commit | commitdiff | tree
blob | blame (incremental) | history | HEAD
KVM: Hoist SVM's get_cs_db_l_bits into core code.
[mirror_ubuntu-artful-kernel.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>
6aa8b732 41
e495606d
AK		 42#include <asm/processor.h>
43#include <asm/msr.h>
44#include <asm/io.h>
45#include <asm/uaccess.h>
46#include <asm/desc.h>
6aa8b732
AK		 47
48MODULE_AUTHOR("Qumranet");
49MODULE_LICENSE("GPL");
50
133de902
AK		 51static DEFINE_SPINLOCK(kvm_lock);
52static LIST_HEAD(vm_list);
53
1b6c0168
AK		 54static cpumask_t cpus_hardware_enabled;
55
6aa8b732 56struct kvm_arch_ops *kvm_arch_ops;
c16f862d
RR		 57struct kmem_cache *kvm_vcpu_cache;
58EXPORT_SYMBOL_GPL(kvm_vcpu_cache);
1165f5fe 59
15ad7146
AK		 60static __read_mostly struct preempt_ops kvm_preempt_ops;
61
1165f5fe 62#define STAT_OFFSET(x) offsetof(struct kvm_vcpu, stat.x)
6aa8b732
AK		 63
64static struct kvm_stats_debugfs_item {
65 const char *name;
1165f5fe 66 int offset;
6aa8b732
AK		 67 struct dentry *dentry;
68} debugfs_entries[] = {
1165f5fe
AK		 69 { "pf_fixed", STAT_OFFSET(pf_fixed) },
70 { "pf_guest", STAT_OFFSET(pf_guest) },
71 { "tlb_flush", STAT_OFFSET(tlb_flush) },
72 { "invlpg", STAT_OFFSET(invlpg) },
73 { "exits", STAT_OFFSET(exits) },
74 { "io_exits", STAT_OFFSET(io_exits) },
75 { "mmio_exits", STAT_OFFSET(mmio_exits) },
76 { "signal_exits", STAT_OFFSET(signal_exits) },
77 { "irq_window", STAT_OFFSET(irq_window_exits) },
78 { "halt_exits", STAT_OFFSET(halt_exits) },
b6958ce4 79 { "halt_wakeup", STAT_OFFSET(halt_wakeup) },
1165f5fe
AK		 80 { "request_irq", STAT_OFFSET(request_irq_exits) },
81 { "irq_exits", STAT_OFFSET(irq_exits) },
e6adf283 82 { "light_exits", STAT_OFFSET(light_exits) },
2cc51560 83 { "efer_reload", STAT_OFFSET(efer_reload) },
1165f5fe 84 { NULL }
6aa8b732
AK		 85};
86
87static struct dentry *debugfs_dir;
88
89#define MAX_IO_MSRS 256
90
707d92fa
RR		 91#define CR0_RESERVED_BITS \
92 (~(unsigned long)(X86_CR0_PE | X86_CR0_MP | X86_CR0_EM | X86_CR0_TS \
93 | X86_CR0_ET | X86_CR0_NE | X86_CR0_WP | X86_CR0_AM \
94 | X86_CR0_NW | X86_CR0_CD | X86_CR0_PG))
66aee91a
RR		 95#define CR4_RESERVED_BITS \
96 (~(unsigned long)(X86_CR4_VME | X86_CR4_PVI | X86_CR4_TSD | X86_CR4_DE\
97 | X86_CR4_PSE | X86_CR4_PAE | X86_CR4_MCE \
98 | X86_CR4_PGE | X86_CR4_PCE | X86_CR4_OSFXSR \
99 | X86_CR4_OSXMMEXCPT | X86_CR4_VMXE))
100
7075bc81 101#define CR8_RESERVED_BITS (~(unsigned long)X86_CR8_TPR)
6aa8b732
AK		 102#define EFER_RESERVED_BITS 0xfffffffffffff2fe
103
05b3e0c2 104#ifdef CONFIG_X86_64
6aa8b732
AK		 105// LDT or TSS descriptor in the GDT. 16 bytes.
106struct segment_descriptor_64 {
107 struct segment_descriptor s;
108 u32 base_higher;
109 u32 pad_zero;
110};
111
112#endif
113
bccf2150
AK		 114static long kvm_vcpu_ioctl(struct file *file, unsigned int ioctl,
115 unsigned long arg);
116
6aa8b732
AK		 117unsigned long segment_base(u16 selector)
118{
119 struct descriptor_table gdt;
120 struct segment_descriptor *d;
121 unsigned long table_base;
122 typedef unsigned long ul;
123 unsigned long v;
124
125 if (selector == 0)
126 return 0;
127
128 asm ("sgdt %0" : "=m"(gdt));
129 table_base = gdt.base;
130
131 if (selector & 4) { /* from ldt */
132 u16 ldt_selector;
133
134 asm ("sldt %0" : "=g"(ldt_selector));
135 table_base = segment_base(ldt_selector);
136 }
137 d = (struct segment_descriptor *)(table_base + (selector & ~7));
138 v = d->base_low | ((ul)d->base_mid << 16) | ((ul)d->base_high << 24);
05b3e0c2 139#ifdef CONFIG_X86_64
6aa8b732
AK		 140 if (d->system == 0
141 && (d->type == 2 || d->type == 9 || d->type == 11))
142 v |= ((ul)((struct segment_descriptor_64 *)d)->base_higher) << 32;
143#endif
144 return v;
145}
146EXPORT_SYMBOL_GPL(segment_base);
147
5aacf0ca
JM		 148static inline int valid_vcpu(int n)
149{
150 return likely(n >= 0 && n < KVM_MAX_VCPUS);
151}
152
7702fd1f
AK		 153void kvm_load_guest_fpu(struct kvm_vcpu *vcpu)
154{
155 if (!vcpu->fpu_active || vcpu->guest_fpu_loaded)
156 return;
157
158 vcpu->guest_fpu_loaded = 1;
b114b080
RR		 159 fx_save(&vcpu->host_fx_image);
160 fx_restore(&vcpu->guest_fx_image);
7702fd1f
AK		 161}
162EXPORT_SYMBOL_GPL(kvm_load_guest_fpu);
163
164void kvm_put_guest_fpu(struct kvm_vcpu *vcpu)
165{
166 if (!vcpu->guest_fpu_loaded)
167 return;
168
169 vcpu->guest_fpu_loaded = 0;
b114b080
RR		 170 fx_save(&vcpu->guest_fx_image);
171 fx_restore(&vcpu->host_fx_image);
7702fd1f
AK		 172}
173EXPORT_SYMBOL_GPL(kvm_put_guest_fpu);
174
bccf2150
AK		 175/*
176 * Switches to specified vcpu, until a matching vcpu_put()
177 */
178static void vcpu_load(struct kvm_vcpu *vcpu)
6aa8b732 179{
15ad7146
AK		 180 int cpu;
181
bccf2150 182 mutex_lock(&vcpu->mutex);
15ad7146
AK		 183 cpu = get_cpu();
184 preempt_notifier_register(&vcpu->preempt_notifier);
185 kvm_arch_ops->vcpu_load(vcpu, cpu);
186 put_cpu();
6aa8b732
AK		 187}
188
6aa8b732
AK		 189static void vcpu_put(struct kvm_vcpu *vcpu)
190{
15ad7146 191 preempt_disable();
6aa8b732 192 kvm_arch_ops->vcpu_put(vcpu);
15ad7146
AK		 193 preempt_notifier_unregister(&vcpu->preempt_notifier);
194 preempt_enable();
6aa8b732
AK		 195 mutex_unlock(&vcpu->mutex);
196}
197
d9e368d6
AK		 198static void ack_flush(void *_completed)
199{
200 atomic_t *completed = _completed;
201
202 atomic_inc(completed);
203}
204
205void kvm_flush_remote_tlbs(struct kvm *kvm)
206{
207 int i, cpu, needed;
208 cpumask_t cpus;
209 struct kvm_vcpu *vcpu;
210 atomic_t completed;
211
212 atomic_set(&completed, 0);
213 cpus_clear(cpus);
214 needed = 0;
fb3f0f51
RR		 215 for (i = 0; i < KVM_MAX_VCPUS; ++i) {
216 vcpu = kvm->vcpus[i];
217 if (!vcpu)
218 continue;
d9e368d6
AK		 219 if (test_and_set_bit(KVM_TLB_FLUSH, &vcpu->requests))
220 continue;
221 cpu = vcpu->cpu;
222 if (cpu != -1 && cpu != raw_smp_processor_id())
223 if (!cpu_isset(cpu, cpus)) {
224 cpu_set(cpu, cpus);
225 ++needed;
226 }
227 }
228
229 /*
230 * We really want smp_call_function_mask() here. But that's not
231 * available, so ipi all cpus in parallel and wait for them
232 * to complete.
233 */
234 for (cpu = first_cpu(cpus); cpu != NR_CPUS; cpu = next_cpu(cpu, cpus))
235 smp_call_function_single(cpu, ack_flush, &completed, 1, 0);
236 while (atomic_read(&completed) != needed) {
237 cpu_relax();
238 barrier();
239 }
240}
241
fb3f0f51
RR		 242int kvm_vcpu_init(struct kvm_vcpu *vcpu, struct kvm *kvm, unsigned id)
243{
244 struct page *page;
245 int r;
246
247 mutex_init(&vcpu->mutex);
248 vcpu->cpu = -1;
249 vcpu->mmu.root_hpa = INVALID_PAGE;
250 vcpu->kvm = kvm;
251 vcpu->vcpu_id = id;
c5ec1534
HQ		 252 if (!irqchip_in_kernel(kvm) || id == 0)
253 vcpu->mp_state = VCPU_MP_STATE_RUNNABLE;
254 else
255 vcpu->mp_state = VCPU_MP_STATE_UNINITIALIZED;
b6958ce4 256 init_waitqueue_head(&vcpu->wq);
fb3f0f51
RR		 257
258 page = alloc_page(GFP_KERNEL | __GFP_ZERO);
259 if (!page) {
260 r = -ENOMEM;
261 goto fail;
262 }
263 vcpu->run = page_address(page);
264
265 page = alloc_page(GFP_KERNEL | __GFP_ZERO);
266 if (!page) {
267 r = -ENOMEM;
268 goto fail_free_run;
269 }
270 vcpu->pio_data = page_address(page);
271
fb3f0f51
RR		 272 r = kvm_mmu_create(vcpu);
273 if (r < 0)
274 goto fail_free_pio_data;
275
276 return 0;
277
278fail_free_pio_data:
279 free_page((unsigned long)vcpu->pio_data);
280fail_free_run:
281 free_page((unsigned long)vcpu->run);
282fail:
283 return -ENOMEM;
284}
285EXPORT_SYMBOL_GPL(kvm_vcpu_init);
286
287void kvm_vcpu_uninit(struct kvm_vcpu *vcpu)
288{
289 kvm_mmu_destroy(vcpu);
1b9778da
ED		 290 if (vcpu->apic)
291 hrtimer_cancel(&vcpu->apic->timer.dev);
97222cc8 292 kvm_free_apic(vcpu->apic);
fb3f0f51
RR		 293 free_page((unsigned long)vcpu->pio_data);
294 free_page((unsigned long)vcpu->run);
295}
296EXPORT_SYMBOL_GPL(kvm_vcpu_uninit);
297
f17abe9a 298static struct kvm *kvm_create_vm(void)
6aa8b732
AK		 299{
300 struct kvm *kvm = kzalloc(sizeof(struct kvm), GFP_KERNEL);
6aa8b732
AK		 301
302 if (!kvm)
f17abe9a 303 return ERR_PTR(-ENOMEM);
6aa8b732 304
74906345 305 kvm_io_bus_init(&kvm->pio_bus);
11ec2804 306 mutex_init(&kvm->lock);
6aa8b732 307 INIT_LIST_HEAD(&kvm->active_mmu_pages);
2eeb2e94 308 kvm_io_bus_init(&kvm->mmio_bus);
5e58cfe4
RR		 309 spin_lock(&kvm_lock);
310 list_add(&kvm->vm_list, &vm_list);
311 spin_unlock(&kvm_lock);
f17abe9a
AK		 312 return kvm;
313}
314
6aa8b732
AK		 315/*
316 * Free any memory in @free but not in @dont.
317 */
318static void kvm_free_physmem_slot(struct kvm_memory_slot *free,
319 struct kvm_memory_slot *dont)
320{
321 int i;
322
323 if (!dont || free->phys_mem != dont->phys_mem)
324 if (free->phys_mem) {
325 for (i = 0; i < free->npages; ++i)
55a54f79
AK		 326 if (free->phys_mem[i])
327 __free_page(free->phys_mem[i]);
6aa8b732
AK		 328 vfree(free->phys_mem);
329 }
330
331 if (!dont || free->dirty_bitmap != dont->dirty_bitmap)
332 vfree(free->dirty_bitmap);
333
8b6d44c7 334 free->phys_mem = NULL;
6aa8b732 335 free->npages = 0;
8b6d44c7 336 free->dirty_bitmap = NULL;
6aa8b732
AK		 337}
338
339static void kvm_free_physmem(struct kvm *kvm)
340{
341 int i;
342
343 for (i = 0; i < kvm->nmemslots; ++i)
8b6d44c7 344 kvm_free_physmem_slot(&kvm->memslots[i], NULL);
6aa8b732
AK		 345}
346
039576c0
AK		 347static void free_pio_guest_pages(struct kvm_vcpu *vcpu)
348{
349 int i;
350
3077c451 351 for (i = 0; i < ARRAY_SIZE(vcpu->pio.guest_pages); ++i)
039576c0
AK		 352 if (vcpu->pio.guest_pages[i]) {
353 __free_page(vcpu->pio.guest_pages[i]);
354 vcpu->pio.guest_pages[i] = NULL;
355 }
356}
357
7b53aa56
AK		 358static void kvm_unload_vcpu_mmu(struct kvm_vcpu *vcpu)
359{
7b53aa56
AK		 360 vcpu_load(vcpu);
361 kvm_mmu_unload(vcpu);
362 vcpu_put(vcpu);
363}
364
6aa8b732
AK		 365static void kvm_free_vcpus(struct kvm *kvm)
366{
367 unsigned int i;
368
7b53aa56
AK		 369 /*
370 * Unpin any mmu pages first.
371 */
372 for (i = 0; i < KVM_MAX_VCPUS; ++i)
fb3f0f51
RR		 373 if (kvm->vcpus[i])
374 kvm_unload_vcpu_mmu(kvm->vcpus[i]);
375 for (i = 0; i < KVM_MAX_VCPUS; ++i) {
376 if (kvm->vcpus[i]) {
377 kvm_arch_ops->vcpu_free(kvm->vcpus[i]);
378 kvm->vcpus[i] = NULL;
379 }
380 }
381
6aa8b732
AK		 382}
383
f17abe9a
AK		 384static void kvm_destroy_vm(struct kvm *kvm)
385{
133de902
AK		 386 spin_lock(&kvm_lock);
387 list_del(&kvm->vm_list);
388 spin_unlock(&kvm_lock);
74906345 389 kvm_io_bus_destroy(&kvm->pio_bus);
2eeb2e94 390 kvm_io_bus_destroy(&kvm->mmio_bus);
85f455f7 391 kfree(kvm->vpic);
1fd4f2a5 392 kfree(kvm->vioapic);
6aa8b732
AK		 393 kvm_free_vcpus(kvm);
394 kvm_free_physmem(kvm);
395 kfree(kvm);
f17abe9a
AK		 396}
397
398static int kvm_vm_release(struct inode *inode, struct file *filp)
399{
400 struct kvm *kvm = filp->private_data;
401
402 kvm_destroy_vm(kvm);
6aa8b732
AK		 403 return 0;
404}
405
406static void inject_gp(struct kvm_vcpu *vcpu)
407{
408 kvm_arch_ops->inject_gp(vcpu, 0);
409}
410
1342d353
AK		 411/*
412 * Load the pae pdptrs. Return true is they are all valid.
413 */
414static int load_pdptrs(struct kvm_vcpu *vcpu, unsigned long cr3)
6aa8b732
AK		 415{
416 gfn_t pdpt_gfn = cr3 >> PAGE_SHIFT;
1342d353 417 unsigned offset = ((cr3 & (PAGE_SIZE-1)) >> 5) << 2;
6aa8b732 418 int i;
6aa8b732 419 u64 *pdpt;
1342d353 420 int ret;
954bbbc2 421 struct page *page;
c820c2aa 422 u64 pdpte[ARRAY_SIZE(vcpu->pdptrs)];
6aa8b732 423
11ec2804 424 mutex_lock(&vcpu->kvm->lock);
954bbbc2 425 page = gfn_to_page(vcpu->kvm, pdpt_gfn);
c820c2aa
RR		 426 if (!page) {
427 ret = 0;
428 goto out;
429 }
430
954bbbc2 431 pdpt = kmap_atomic(page, KM_USER0);
c820c2aa
RR		 432 memcpy(pdpte, pdpt+offset, sizeof(pdpte));
433 kunmap_atomic(pdpt, KM_USER0);
6aa8b732 434
c820c2aa
RR		 435 for (i = 0; i < ARRAY_SIZE(pdpte); ++i) {
436 if ((pdpte[i] & 1) && (pdpte[i] & 0xfffffff0000001e6ull)) {
1342d353
AK		 437 ret = 0;
438 goto out;
439 }
6aa8b732 440 }
c820c2aa 441 ret = 1;
6aa8b732 442
c820c2aa 443 memcpy(vcpu->pdptrs, pdpte, sizeof(vcpu->pdptrs));
1342d353 444out:
11ec2804 445 mutex_unlock(&vcpu->kvm->lock);
6aa8b732 446
1342d353 447 return ret;
6aa8b732
AK		 448}
449
450void set_cr0(struct kvm_vcpu *vcpu, unsigned long cr0)
451{
707d92fa 452 if (cr0 & CR0_RESERVED_BITS) {
6aa8b732
AK		 453 printk(KERN_DEBUG "set_cr0: 0x%lx #GP, reserved bits 0x%lx\n",
454 cr0, vcpu->cr0);
455 inject_gp(vcpu);
456 return;
457 }
458
707d92fa 459 if ((cr0 & X86_CR0_NW) && !(cr0 & X86_CR0_CD)) {
6aa8b732
AK		 460 printk(KERN_DEBUG "set_cr0: #GP, CD == 0 && NW == 1\n");
461 inject_gp(vcpu);
462 return;
463 }
464
707d92fa 465 if ((cr0 & X86_CR0_PG) && !(cr0 & X86_CR0_PE)) {
6aa8b732
AK		 466 printk(KERN_DEBUG "set_cr0: #GP, set PG flag "
467 "and a clear PE flag\n");
468 inject_gp(vcpu);
469 return;
470 }
471
707d92fa 472 if (!is_paging(vcpu) && (cr0 & X86_CR0_PG)) {
05b3e0c2 473#ifdef CONFIG_X86_64
6aa8b732
AK		 474 if ((vcpu->shadow_efer & EFER_LME)) {
475 int cs_db, cs_l;
476
477 if (!is_pae(vcpu)) {
478 printk(KERN_DEBUG "set_cr0: #GP, start paging "
479 "in long mode while PAE is disabled\n");
480 inject_gp(vcpu);
481 return;
482 }
483 kvm_arch_ops->get_cs_db_l_bits(vcpu, &cs_db, &cs_l);
484 if (cs_l) {
485 printk(KERN_DEBUG "set_cr0: #GP, start paging "
486 "in long mode while CS.L == 1\n");
487 inject_gp(vcpu);
488 return;
489
490 }
491 } else
492#endif
1342d353 493 if (is_pae(vcpu) && !load_pdptrs(vcpu, vcpu->cr3)) {
6aa8b732
AK		 494 printk(KERN_DEBUG "set_cr0: #GP, pdptrs "
495 "reserved bits\n");
496 inject_gp(vcpu);
497 return;
498 }
499
500 }
501
502 kvm_arch_ops->set_cr0(vcpu, cr0);
503 vcpu->cr0 = cr0;
504
11ec2804 505 mutex_lock(&vcpu->kvm->lock);
6aa8b732 506 kvm_mmu_reset_context(vcpu);
11ec2804 507 mutex_unlock(&vcpu->kvm->lock);
6aa8b732
AK		 508 return;
509}
510EXPORT_SYMBOL_GPL(set_cr0);
511
512void lmsw(struct kvm_vcpu *vcpu, unsigned long msw)
513{
514 set_cr0(vcpu, (vcpu->cr0 & ~0x0ful) | (msw & 0x0f));
515}
516EXPORT_SYMBOL_GPL(lmsw);
517
518void set_cr4(struct kvm_vcpu *vcpu, unsigned long cr4)
519{
66aee91a 520 if (cr4 & CR4_RESERVED_BITS) {
6aa8b732
AK		 521 printk(KERN_DEBUG "set_cr4: #GP, reserved bits\n");
522 inject_gp(vcpu);
523 return;
524 }
525
a9058ecd 526 if (is_long_mode(vcpu)) {
66aee91a 527 if (!(cr4 & X86_CR4_PAE)) {
6aa8b732
AK		 528 printk(KERN_DEBUG "set_cr4: #GP, clearing PAE while "
529 "in long mode\n");
530 inject_gp(vcpu);
531 return;
532 }
66aee91a 533 } else if (is_paging(vcpu) && !is_pae(vcpu) && (cr4 & X86_CR4_PAE)
1342d353 534 && !load_pdptrs(vcpu, vcpu->cr3)) {
6aa8b732
AK		 535 printk(KERN_DEBUG "set_cr4: #GP, pdptrs reserved bits\n");
536 inject_gp(vcpu);
310bc76c 537 return;
6aa8b732
AK		 538 }
539
66aee91a 540 if (cr4 & X86_CR4_VMXE) {
6aa8b732
AK		 541 printk(KERN_DEBUG "set_cr4: #GP, setting VMXE\n");
542 inject_gp(vcpu);
543 return;
544 }
545 kvm_arch_ops->set_cr4(vcpu, cr4);
81f50e3b 546 vcpu->cr4 = cr4;
11ec2804 547 mutex_lock(&vcpu->kvm->lock);
6aa8b732 548 kvm_mmu_reset_context(vcpu);
11ec2804 549 mutex_unlock(&vcpu->kvm->lock);
6aa8b732
AK		 550}
551EXPORT_SYMBOL_GPL(set_cr4);
552
553void set_cr3(struct kvm_vcpu *vcpu, unsigned long cr3)
554{
a9058ecd 555 if (is_long_mode(vcpu)) {
f802a307 556 if (cr3 & CR3_L_MODE_RESERVED_BITS) {
6aa8b732
AK		 557 printk(KERN_DEBUG "set_cr3: #GP, reserved bits\n");
558 inject_gp(vcpu);
559 return;
560 }
561 } else {
f802a307
RR		 562 if (is_pae(vcpu)) {
563 if (cr3 & CR3_PAE_RESERVED_BITS) {
564 printk(KERN_DEBUG
565 "set_cr3: #GP, reserved bits\n");
566 inject_gp(vcpu);
567 return;
568 }
569 if (is_paging(vcpu) && !load_pdptrs(vcpu, cr3)) {
570 printk(KERN_DEBUG "set_cr3: #GP, pdptrs "
571 "reserved bits\n");
572 inject_gp(vcpu);
573 return;
574 }
575 } else {
576 if (cr3 & CR3_NONPAE_RESERVED_BITS) {
577 printk(KERN_DEBUG
578 "set_cr3: #GP, reserved bits\n");
579 inject_gp(vcpu);
580 return;
581 }
6aa8b732
AK		 582 }
583 }
584
11ec2804 585 mutex_lock(&vcpu->kvm->lock);
d21225ee
IM		 586 /*
587 * Does the new cr3 value map to physical memory? (Note, we
588 * catch an invalid cr3 even in real-mode, because it would
589 * cause trouble later on when we turn on paging anyway.)
590 *
591 * A real CPU would silently accept an invalid cr3 and would
592 * attempt to use it - with largely undefined (and often hard
593 * to debug) behavior on the guest side.
594 */
595 if (unlikely(!gfn_to_memslot(vcpu->kvm, cr3 >> PAGE_SHIFT)))
596 inject_gp(vcpu);
fb764416
RR		 597 else {
598 vcpu->cr3 = cr3;
d21225ee 599 vcpu->mmu.new_cr3(vcpu);
fb764416 600 }
11ec2804 601 mutex_unlock(&vcpu->kvm->lock);
6aa8b732
AK		 602}
603EXPORT_SYMBOL_GPL(set_cr3);
604
605void set_cr8(struct kvm_vcpu *vcpu, unsigned long cr8)
606{
7075bc81 607 if (cr8 & CR8_RESERVED_BITS) {
6aa8b732
AK		 608 printk(KERN_DEBUG "set_cr8: #GP, reserved bits 0x%lx\n", cr8);
609 inject_gp(vcpu);
610 return;
611 }
97222cc8
ED		 612 if (irqchip_in_kernel(vcpu->kvm))
613 kvm_lapic_set_tpr(vcpu, cr8);
614 else
615 vcpu->cr8 = cr8;
6aa8b732
AK		 616}
617EXPORT_SYMBOL_GPL(set_cr8);
618
7017fc3d
ED		 619unsigned long get_cr8(struct kvm_vcpu *vcpu)
620{
97222cc8
ED		 621 if (irqchip_in_kernel(vcpu->kvm))
622 return kvm_lapic_get_cr8(vcpu);
623 else
624 return vcpu->cr8;
7017fc3d
ED		 625}
626EXPORT_SYMBOL_GPL(get_cr8);
627
628u64 kvm_get_apic_base(struct kvm_vcpu *vcpu)
629{
97222cc8
ED		 630 if (irqchip_in_kernel(vcpu->kvm))
631 return vcpu->apic_base;
632 else
633 return vcpu->apic_base;
7017fc3d
ED		 634}
635EXPORT_SYMBOL_GPL(kvm_get_apic_base);
636
637void kvm_set_apic_base(struct kvm_vcpu *vcpu, u64 data)
638{
97222cc8
ED		 639 /* TODO: reserve bits check */
640 if (irqchip_in_kernel(vcpu->kvm))
641 kvm_lapic_set_base(vcpu, data);
642 else
643 vcpu->apic_base = data;
7017fc3d
ED		 644}
645EXPORT_SYMBOL_GPL(kvm_set_apic_base);
646
6aa8b732
AK		 647void fx_init(struct kvm_vcpu *vcpu)
648{
b114b080 649 unsigned after_mxcsr_mask;
6aa8b732 650
9bd01506
RR		 651 /* Initialize guest FPU by resetting ours and saving into guest's */
652 preempt_disable();
b114b080 653 fx_save(&vcpu->host_fx_image);
6aa8b732 654 fpu_init();
b114b080
RR		 655 fx_save(&vcpu->guest_fx_image);
656 fx_restore(&vcpu->host_fx_image);
9bd01506 657 preempt_enable();
6aa8b732 658
380102c8 659 vcpu->cr0 |= X86_CR0_ET;
b114b080
RR		 660 after_mxcsr_mask = offsetof(struct i387_fxsave_struct, st_space);
661 vcpu->guest_fx_image.mxcsr = 0x1f80;
662 memset((void *)&vcpu->guest_fx_image + after_mxcsr_mask,
663 0, sizeof(struct i387_fxsave_struct) - after_mxcsr_mask);
6aa8b732
AK		 664}
665EXPORT_SYMBOL_GPL(fx_init);
666
6aa8b732
AK		 667/*
668 * Allocate some memory and give it an address in the guest physical address
669 * space.
670 *
671 * Discontiguous memory is allowed, mostly for framebuffers.
672 */
2c6f5df9
AK		 673static int kvm_vm_ioctl_set_memory_region(struct kvm *kvm,
674 struct kvm_memory_region *mem)
6aa8b732
AK		 675{
676 int r;
677 gfn_t base_gfn;
678 unsigned long npages;
679 unsigned long i;
680 struct kvm_memory_slot *memslot;
681 struct kvm_memory_slot old, new;
682 int memory_config_version;
683
684 r = -EINVAL;
685 /* General sanity checks */
686 if (mem->memory_size & (PAGE_SIZE - 1))
687 goto out;
688 if (mem->guest_phys_addr & (PAGE_SIZE - 1))
689 goto out;
690 if (mem->slot >= KVM_MEMORY_SLOTS)
691 goto out;
692 if (mem->guest_phys_addr + mem->memory_size < mem->guest_phys_addr)
693 goto out;
694
695 memslot = &kvm->memslots[mem->slot];
696 base_gfn = mem->guest_phys_addr >> PAGE_SHIFT;
697 npages = mem->memory_size >> PAGE_SHIFT;
698
699 if (!npages)
700 mem->flags &= ~KVM_MEM_LOG_DIRTY_PAGES;
701
702raced:
11ec2804 703 mutex_lock(&kvm->lock);
6aa8b732
AK		 704
705 memory_config_version = kvm->memory_config_version;
706 new = old = *memslot;
707
708 new.base_gfn = base_gfn;
709 new.npages = npages;
710 new.flags = mem->flags;
711
712 /* Disallow changing a memory slot's size. */
713 r = -EINVAL;
714 if (npages && old.npages && npages != old.npages)
715 goto out_unlock;
716
717 /* Check for overlaps */
718 r = -EEXIST;
719 for (i = 0; i < KVM_MEMORY_SLOTS; ++i) {
720 struct kvm_memory_slot *s = &kvm->memslots[i];
721
722 if (s == memslot)
723 continue;
724 if (!((base_gfn + npages <= s->base_gfn) ||
725 (base_gfn >= s->base_gfn + s->npages)))
726 goto out_unlock;
727 }
728 /*
729 * Do memory allocations outside lock. memory_config_version will
730 * detect any races.
731 */
11ec2804 732 mutex_unlock(&kvm->lock);
6aa8b732
AK		 733
734 /* Deallocate if slot is being removed */
735 if (!npages)
8b6d44c7 736 new.phys_mem = NULL;
6aa8b732
AK		 737
738 /* Free page dirty bitmap if unneeded */
739 if (!(new.flags & KVM_MEM_LOG_DIRTY_PAGES))
8b6d44c7 740 new.dirty_bitmap = NULL;
6aa8b732
AK		 741
742 r = -ENOMEM;
743
744 /* Allocate if a slot is being created */
745 if (npages && !new.phys_mem) {
746 new.phys_mem = vmalloc(npages * sizeof(struct page *));
747
748 if (!new.phys_mem)
749 goto out_free;
750
751 memset(new.phys_mem, 0, npages * sizeof(struct page *));
752 for (i = 0; i < npages; ++i) {
753 new.phys_mem[i] = alloc_page(GFP_HIGHUSER
754 | __GFP_ZERO);
755 if (!new.phys_mem[i])
756 goto out_free;
5972e953 757 set_page_private(new.phys_mem[i],0);
6aa8b732
AK		 758 }
759 }
760
761 /* Allocate page dirty bitmap if needed */
762 if ((new.flags & KVM_MEM_LOG_DIRTY_PAGES) && !new.dirty_bitmap) {
763 unsigned dirty_bytes = ALIGN(npages, BITS_PER_LONG) / 8;
764
765 new.dirty_bitmap = vmalloc(dirty_bytes);
766 if (!new.dirty_bitmap)
767 goto out_free;
768 memset(new.dirty_bitmap, 0, dirty_bytes);
769 }
770
11ec2804 771 mutex_lock(&kvm->lock);
6aa8b732
AK		 772
773 if (memory_config_version != kvm->memory_config_version) {
11ec2804 774 mutex_unlock(&kvm->lock);
6aa8b732
AK		 775 kvm_free_physmem_slot(&new, &old);
776 goto raced;
777 }
778
779 r = -EAGAIN;
780 if (kvm->busy)
781 goto out_unlock;
782
783 if (mem->slot >= kvm->nmemslots)
784 kvm->nmemslots = mem->slot + 1;
785
786 *memslot = new;
787 ++kvm->memory_config_version;
788
90cb0529
AK		 789 kvm_mmu_slot_remove_write_access(kvm, mem->slot);
790 kvm_flush_remote_tlbs(kvm);
6aa8b732 791
11ec2804 792 mutex_unlock(&kvm->lock);
6aa8b732
AK		 793
794 kvm_free_physmem_slot(&old, &new);
795 return 0;
796
797out_unlock:
11ec2804 798 mutex_unlock(&kvm->lock);
6aa8b732
AK		 799out_free:
800 kvm_free_physmem_slot(&new, &old);
801out:
802 return r;
803}
804
805/*
806 * Get (and clear) the dirty memory log for a memory slot.
807 */
2c6f5df9
AK		 808static int kvm_vm_ioctl_get_dirty_log(struct kvm *kvm,
809 struct kvm_dirty_log *log)
6aa8b732
AK		 810{
811 struct kvm_memory_slot *memslot;
812 int r, i;
813 int n;
814 unsigned long any = 0;
815
11ec2804 816 mutex_lock(&kvm->lock);
6aa8b732
AK		 817
818 /*
819 * Prevent changes to guest memory configuration even while the lock
820 * is not taken.
821 */
822 ++kvm->busy;
11ec2804 823 mutex_unlock(&kvm->lock);
6aa8b732
AK		 824 r = -EINVAL;
825 if (log->slot >= KVM_MEMORY_SLOTS)
826 goto out;
827
828 memslot = &kvm->memslots[log->slot];
829 r = -ENOENT;
830 if (!memslot->dirty_bitmap)
831 goto out;
832
cd1a4a98 833 n = ALIGN(memslot->npages, BITS_PER_LONG) / 8;
6aa8b732 834
cd1a4a98 835 for (i = 0; !any && i < n/sizeof(long); ++i)
6aa8b732
AK		 836 any = memslot->dirty_bitmap[i];
837
838 r = -EFAULT;
839 if (copy_to_user(log->dirty_bitmap, memslot->dirty_bitmap, n))
840 goto out;
841
39214915
RR		 842 /* If nothing is dirty, don't bother messing with page tables. */
843 if (any) {
844 mutex_lock(&kvm->lock);
845 kvm_mmu_slot_remove_write_access(kvm, log->slot);
846 kvm_flush_remote_tlbs(kvm);
847 memset(memslot->dirty_bitmap, 0, n);
848 mutex_unlock(&kvm->lock);
849 }
6aa8b732
AK		 850
851 r = 0;
852
853out:
11ec2804 854 mutex_lock(&kvm->lock);
6aa8b732 855 --kvm->busy;
11ec2804 856 mutex_unlock(&kvm->lock);
6aa8b732
AK		 857 return r;
858}
859
e8207547
AK		 860/*
861 * Set a new alias region. Aliases map a portion of physical memory into
862 * another portion. This is useful for memory windows, for example the PC
863 * VGA region.
864 */
865static int kvm_vm_ioctl_set_memory_alias(struct kvm *kvm,
866 struct kvm_memory_alias *alias)
867{
868 int r, n;
869 struct kvm_mem_alias *p;
870
871 r = -EINVAL;
872 /* General sanity checks */
873 if (alias->memory_size & (PAGE_SIZE - 1))
874 goto out;
875 if (alias->guest_phys_addr & (PAGE_SIZE - 1))
876 goto out;
877 if (alias->slot >= KVM_ALIAS_SLOTS)
878 goto out;
879 if (alias->guest_phys_addr + alias->memory_size
880 < alias->guest_phys_addr)
881 goto out;
882 if (alias->target_phys_addr + alias->memory_size
883 < alias->target_phys_addr)
884 goto out;
885
11ec2804 886 mutex_lock(&kvm->lock);
e8207547
AK		 887
888 p = &kvm->aliases[alias->slot];
889 p->base_gfn = alias->guest_phys_addr >> PAGE_SHIFT;
890 p->npages = alias->memory_size >> PAGE_SHIFT;
891 p->target_gfn = alias->target_phys_addr >> PAGE_SHIFT;
892
893 for (n = KVM_ALIAS_SLOTS; n > 0; --n)
894 if (kvm->aliases[n - 1].npages)
895 break;
896 kvm->naliases = n;
897
90cb0529 898 kvm_mmu_zap_all(kvm);
e8207547 899
11ec2804 900 mutex_unlock(&kvm->lock);
e8207547
AK		 901
902 return 0;
903
904out:
905 return r;
906}
907
6ceb9d79
HQ		 908static int kvm_vm_ioctl_get_irqchip(struct kvm *kvm, struct kvm_irqchip *chip)
909{
910 int r;
911
912 r = 0;
913 switch (chip->chip_id) {
914 case KVM_IRQCHIP_PIC_MASTER:
915 memcpy (&chip->chip.pic,
916 &pic_irqchip(kvm)->pics[0],
917 sizeof(struct kvm_pic_state));
918 break;
919 case KVM_IRQCHIP_PIC_SLAVE:
920 memcpy (&chip->chip.pic,
921 &pic_irqchip(kvm)->pics[1],
922 sizeof(struct kvm_pic_state));
923 break;
6bf9e962
HQ		 924 case KVM_IRQCHIP_IOAPIC:
925 memcpy (&chip->chip.ioapic,
926 ioapic_irqchip(kvm),
927 sizeof(struct kvm_ioapic_state));
928 break;
6ceb9d79
HQ		 929 default:
930 r = -EINVAL;
931 break;
932 }
933 return r;
934}
935
936static int kvm_vm_ioctl_set_irqchip(struct kvm *kvm, struct kvm_irqchip *chip)
937{
938 int r;
939
940 r = 0;
941 switch (chip->chip_id) {
942 case KVM_IRQCHIP_PIC_MASTER:
943 memcpy (&pic_irqchip(kvm)->pics[0],
944 &chip->chip.pic,
945 sizeof(struct kvm_pic_state));
946 break;
947 case KVM_IRQCHIP_PIC_SLAVE:
948 memcpy (&pic_irqchip(kvm)->pics[1],
949 &chip->chip.pic,
950 sizeof(struct kvm_pic_state));
951 break;
6bf9e962
HQ		 952 case KVM_IRQCHIP_IOAPIC:
953 memcpy (ioapic_irqchip(kvm),
954 &chip->chip.ioapic,
955 sizeof(struct kvm_ioapic_state));
956 break;
6ceb9d79
HQ		 957 default:
958 r = -EINVAL;
959 break;
960 }
961 kvm_pic_update_irq(pic_irqchip(kvm));
962 return r;
963}
964
e8207547
AK		 965static gfn_t unalias_gfn(struct kvm *kvm, gfn_t gfn)
966{
967 int i;
968 struct kvm_mem_alias *alias;
969
970 for (i = 0; i < kvm->naliases; ++i) {
971 alias = &kvm->aliases[i];
972 if (gfn >= alias->base_gfn
973 && gfn < alias->base_gfn + alias->npages)
974 return alias->target_gfn + gfn - alias->base_gfn;
975 }
976 return gfn;
977}
978
979static struct kvm_memory_slot *__gfn_to_memslot(struct kvm *kvm, gfn_t gfn)
6aa8b732
AK		 980{
981 int i;
982
983 for (i = 0; i < kvm->nmemslots; ++i) {
984 struct kvm_memory_slot *memslot = &kvm->memslots[i];
985
986 if (gfn >= memslot->base_gfn
987 && gfn < memslot->base_gfn + memslot->npages)
988 return memslot;
989 }
8b6d44c7 990 return NULL;
6aa8b732 991}
e8207547
AK		 992
993struct kvm_memory_slot *gfn_to_memslot(struct kvm *kvm, gfn_t gfn)
994{
995 gfn = unalias_gfn(kvm, gfn);
996 return __gfn_to_memslot(kvm, gfn);
997}
6aa8b732 998
954bbbc2
AK		 999struct page *gfn_to_page(struct kvm *kvm, gfn_t gfn)
1000{
1001 struct kvm_memory_slot *slot;
1002
e8207547
AK		 1003 gfn = unalias_gfn(kvm, gfn);
1004 slot = __gfn_to_memslot(kvm, gfn);
954bbbc2
AK		 1005 if (!slot)
1006 return NULL;
1007 return slot->phys_mem[gfn - slot->base_gfn];
1008}
1009EXPORT_SYMBOL_GPL(gfn_to_page);
1010
7e9d619d 1011/* WARNING: Does not work on aliased pages. */
6aa8b732
AK		 1012void mark_page_dirty(struct kvm *kvm, gfn_t gfn)
1013{
31389947 1014 struct kvm_memory_slot *memslot;
6aa8b732 1015
7e9d619d
RR		 1016 memslot = __gfn_to_memslot(kvm, gfn);
1017 if (memslot && memslot->dirty_bitmap) {
1018 unsigned long rel_gfn = gfn - memslot->base_gfn;
6aa8b732 1019
7e9d619d
RR		 1020 /* avoid RMW */
1021 if (!test_bit(rel_gfn, memslot->dirty_bitmap))
1022 set_bit(rel_gfn, memslot->dirty_bitmap);
6aa8b732
AK		 1023 }
1024}
1025
e7d5d76c 1026int emulator_read_std(unsigned long addr,
4c690a1e 1027 void *val,
6aa8b732 1028 unsigned int bytes,
cebff02b 1029 struct kvm_vcpu *vcpu)
6aa8b732 1030{
6aa8b732
AK		 1031 void *data = val;
1032
1033 while (bytes) {
1034 gpa_t gpa = vcpu->mmu.gva_to_gpa(vcpu, addr);
1035 unsigned offset = addr & (PAGE_SIZE-1);
1036 unsigned tocopy = min(bytes, (unsigned)PAGE_SIZE - offset);
1037 unsigned long pfn;
954bbbc2
AK		 1038 struct page *page;
1039 void *page_virt;
6aa8b732
AK		 1040
1041 if (gpa == UNMAPPED_GVA)
1042 return X86EMUL_PROPAGATE_FAULT;
1043 pfn = gpa >> PAGE_SHIFT;
954bbbc2
AK		 1044 page = gfn_to_page(vcpu->kvm, pfn);
1045 if (!page)
6aa8b732 1046 return X86EMUL_UNHANDLEABLE;
954bbbc2 1047 page_virt = kmap_atomic(page, KM_USER0);
6aa8b732 1048
954bbbc2 1049 memcpy(data, page_virt + offset, tocopy);
6aa8b732 1050
954bbbc2 1051 kunmap_atomic(page_virt, KM_USER0);
6aa8b732
AK		 1052
1053 bytes -= tocopy;
1054 data += tocopy;
1055 addr += tocopy;
1056 }
1057
1058 return X86EMUL_CONTINUE;
1059}
e7d5d76c 1060EXPORT_SYMBOL_GPL(emulator_read_std);
6aa8b732
AK		 1061
1062static int emulator_write_std(unsigned long addr,
4c690a1e 1063 const void *val,
6aa8b732 1064 unsigned int bytes,
cebff02b 1065 struct kvm_vcpu *vcpu)
6aa8b732 1066{
f0242478 1067 pr_unimpl(vcpu, "emulator_write_std: addr %lx n %d\n", addr, bytes);
6aa8b732
AK		 1068 return X86EMUL_UNHANDLEABLE;
1069}
1070
97222cc8
ED		 1071/*
1072 * Only apic need an MMIO device hook, so shortcut now..
1073 */
1074static struct kvm_io_device *vcpu_find_pervcpu_dev(struct kvm_vcpu *vcpu,
1075 gpa_t addr)
1076{
1077 struct kvm_io_device *dev;
1078
1079 if (vcpu->apic) {
1080 dev = &vcpu->apic->dev;
1081 if (dev->in_range(dev, addr))
1082 return dev;
1083 }
1084 return NULL;
1085}
1086
2eeb2e94
GH		 1087static struct kvm_io_device *vcpu_find_mmio_dev(struct kvm_vcpu *vcpu,
1088 gpa_t addr)
1089{
97222cc8
ED		 1090 struct kvm_io_device *dev;
1091
1092 dev = vcpu_find_pervcpu_dev(vcpu, addr);
1093 if (dev == NULL)
1094 dev = kvm_io_bus_find_dev(&vcpu->kvm->mmio_bus, addr);
1095 return dev;
2eeb2e94
GH		 1096}
1097
74906345
ED		 1098static struct kvm_io_device *vcpu_find_pio_dev(struct kvm_vcpu *vcpu,
1099 gpa_t addr)
1100{
1101 return kvm_io_bus_find_dev(&vcpu->kvm->pio_bus, addr);
1102}
1103
6aa8b732 1104static int emulator_read_emulated(unsigned long addr,
4c690a1e 1105 void *val,
6aa8b732 1106 unsigned int bytes,
cebff02b 1107 struct kvm_vcpu *vcpu)
6aa8b732 1108{
2eeb2e94
GH		 1109 struct kvm_io_device *mmio_dev;
1110 gpa_t gpa;
6aa8b732
AK		 1111
1112 if (vcpu->mmio_read_completed) {
1113 memcpy(val, vcpu->mmio_data, bytes);
1114 vcpu->mmio_read_completed = 0;
1115 return X86EMUL_CONTINUE;
cebff02b 1116 } else if (emulator_read_std(addr, val, bytes, vcpu)
6aa8b732
AK		 1117 == X86EMUL_CONTINUE)
1118 return X86EMUL_CONTINUE;
d27d4aca 1119
2eeb2e94
GH		 1120 gpa = vcpu->mmu.gva_to_gpa(vcpu, addr);
1121 if (gpa == UNMAPPED_GVA)
1122 return X86EMUL_PROPAGATE_FAULT;
6aa8b732 1123
2eeb2e94
GH		 1124 /*
1125 * Is this MMIO handled locally?
1126 */
1127 mmio_dev = vcpu_find_mmio_dev(vcpu, gpa);
1128 if (mmio_dev) {
1129 kvm_iodevice_read(mmio_dev, gpa, bytes, val);
1130 return X86EMUL_CONTINUE;
6aa8b732 1131 }
2eeb2e94
GH		 1132
1133 vcpu->mmio_needed = 1;
1134 vcpu->mmio_phys_addr = gpa;
1135 vcpu->mmio_size = bytes;
1136 vcpu->mmio_is_write = 0;
1137
1138 return X86EMUL_UNHANDLEABLE;
6aa8b732
AK		 1139}
1140
da4a00f0 1141static int emulator_write_phys(struct kvm_vcpu *vcpu, gpa_t gpa,
4c690a1e 1142 const void *val, int bytes)
da4a00f0 1143{
da4a00f0
AK		 1144 struct page *page;
1145 void *virt;
1146
1147 if (((gpa + bytes - 1) >> PAGE_SHIFT) != (gpa >> PAGE_SHIFT))
1148 return 0;
954bbbc2
AK		 1149 page = gfn_to_page(vcpu->kvm, gpa >> PAGE_SHIFT);
1150 if (!page)
da4a00f0 1151 return 0;
ab51a434 1152 mark_page_dirty(vcpu->kvm, gpa >> PAGE_SHIFT);
da4a00f0 1153 virt = kmap_atomic(page, KM_USER0);
fe551881 1154 kvm_mmu_pte_write(vcpu, gpa, val, bytes);
7cfa4b0a 1155 memcpy(virt + offset_in_page(gpa), val, bytes);
da4a00f0 1156 kunmap_atomic(virt, KM_USER0);
da4a00f0
AK		 1157 return 1;
1158}
1159
b0fcd903
AK		 1160static int emulator_write_emulated_onepage(unsigned long addr,
1161 const void *val,
1162 unsigned int bytes,
cebff02b 1163 struct kvm_vcpu *vcpu)
6aa8b732 1164{
2eeb2e94
GH		 1165 struct kvm_io_device *mmio_dev;
1166 gpa_t gpa = vcpu->mmu.gva_to_gpa(vcpu, addr);
6aa8b732 1167
c9047f53
AK		 1168 if (gpa == UNMAPPED_GVA) {
1169 kvm_arch_ops->inject_page_fault(vcpu, addr, 2);
6aa8b732 1170 return X86EMUL_PROPAGATE_FAULT;
c9047f53 1171 }
6aa8b732 1172
da4a00f0
AK		 1173 if (emulator_write_phys(vcpu, gpa, val, bytes))
1174 return X86EMUL_CONTINUE;
1175
2eeb2e94
GH		 1176 /*
1177 * Is this MMIO handled locally?
1178 */
1179 mmio_dev = vcpu_find_mmio_dev(vcpu, gpa);
1180 if (mmio_dev) {
1181 kvm_iodevice_write(mmio_dev, gpa, bytes, val);
1182 return X86EMUL_CONTINUE;
1183 }
1184
6aa8b732
AK		 1185 vcpu->mmio_needed = 1;
1186 vcpu->mmio_phys_addr = gpa;
1187 vcpu->mmio_size = bytes;
1188 vcpu->mmio_is_write = 1;
4c690a1e 1189 memcpy(vcpu->mmio_data, val, bytes);
6aa8b732
AK		 1190
1191 return X86EMUL_CONTINUE;
1192}
1193
e7d5d76c 1194int emulator_write_emulated(unsigned long addr,
b0fcd903
AK		 1195 const void *val,
1196 unsigned int bytes,
cebff02b 1197 struct kvm_vcpu *vcpu)
b0fcd903
AK		 1198{
1199 /* Crossing a page boundary? */
1200 if (((addr + bytes - 1) ^ addr) & PAGE_MASK) {
1201 int rc, now;
1202
1203 now = -addr & ~PAGE_MASK;
cebff02b 1204 rc = emulator_write_emulated_onepage(addr, val, now, vcpu);
b0fcd903
AK		 1205 if (rc != X86EMUL_CONTINUE)
1206 return rc;
1207 addr += now;
1208 val += now;
1209 bytes -= now;
1210 }
cebff02b 1211 return emulator_write_emulated_onepage(addr, val, bytes, vcpu);
b0fcd903 1212}
e7d5d76c 1213EXPORT_SYMBOL_GPL(emulator_write_emulated);
b0fcd903 1214
6aa8b732 1215static int emulator_cmpxchg_emulated(unsigned long addr,
4c690a1e
AK		 1216 const void *old,
1217 const void *new,
6aa8b732 1218 unsigned int bytes,
cebff02b 1219 struct kvm_vcpu *vcpu)
6aa8b732
AK		 1220{
1221 static int reported;
1222
1223 if (!reported) {
1224 reported = 1;
1225 printk(KERN_WARNING "kvm: emulating exchange as write\n");
1226 }
cebff02b 1227 return emulator_write_emulated(addr, new, bytes, vcpu);
6aa8b732
AK		 1228}
1229
1230static unsigned long get_segment_base(struct kvm_vcpu *vcpu, int seg)
1231{
1232 return kvm_arch_ops->get_segment_base(vcpu, seg);
1233}
1234
1235int emulate_invlpg(struct kvm_vcpu *vcpu, gva_t address)
1236{
6aa8b732
AK		 1237 return X86EMUL_CONTINUE;
1238}
1239
1240int emulate_clts(struct kvm_vcpu *vcpu)
1241{
81f50e3b
RR		 1242 vcpu->cr0 &= ~X86_CR0_TS;
1243 kvm_arch_ops->set_cr0(vcpu, vcpu->cr0);
6aa8b732
AK		 1244 return X86EMUL_CONTINUE;
1245}
1246
1247int emulator_get_dr(struct x86_emulate_ctxt* ctxt, int dr, unsigned long *dest)
1248{
1249 struct kvm_vcpu *vcpu = ctxt->vcpu;
1250
1251 switch (dr) {
1252 case 0 ... 3:
1253 *dest = kvm_arch_ops->get_dr(vcpu, dr);
1254 return X86EMUL_CONTINUE;
1255 default:
f0242478 1256 pr_unimpl(vcpu, "%s: unexpected dr %u\n", __FUNCTION__, dr);
6aa8b732
AK		 1257 return X86EMUL_UNHANDLEABLE;
1258 }
1259}
1260
1261int emulator_set_dr(struct x86_emulate_ctxt *ctxt, int dr, unsigned long value)
1262{
1263 unsigned long mask = (ctxt->mode == X86EMUL_MODE_PROT64) ? ~0ULL : ~0U;
1264 int exception;
1265
1266 kvm_arch_ops->set_dr(ctxt->vcpu, dr, value & mask, &exception);
1267 if (exception) {
1268 /* FIXME: better handling */
1269 return X86EMUL_UNHANDLEABLE;
1270 }
1271 return X86EMUL_CONTINUE;
1272}
1273
1274static void report_emulation_failure(struct x86_emulate_ctxt *ctxt)
1275{
1276 static int reported;
1277 u8 opcodes[4];
1278 unsigned long rip = ctxt->vcpu->rip;
1279 unsigned long rip_linear;
1280
1281 rip_linear = rip + get_segment_base(ctxt->vcpu, VCPU_SREG_CS);
1282
1283 if (reported)
1284 return;
1285
cebff02b 1286 emulator_read_std(rip_linear, (void *)opcodes, 4, ctxt->vcpu);
6aa8b732
AK		 1287
1288 printk(KERN_ERR "emulation failed but !mmio_needed?"
1289 " rip %lx %02x %02x %02x %02x\n",
1290 rip, opcodes[0], opcodes[1], opcodes[2], opcodes[3]);
1291 reported = 1;
1292}
1293
1294struct x86_emulate_ops emulate_ops = {
1295 .read_std = emulator_read_std,
1296 .write_std = emulator_write_std,
1297 .read_emulated = emulator_read_emulated,
1298 .write_emulated = emulator_write_emulated,
1299 .cmpxchg_emulated = emulator_cmpxchg_emulated,
1300};
1301
1302int emulate_instruction(struct kvm_vcpu *vcpu,
1303 struct kvm_run *run,
1304 unsigned long cr2,
1305 u16 error_code)
1306{
1307 struct x86_emulate_ctxt emulate_ctxt;
1308 int r;
1309 int cs_db, cs_l;
1310
e7df56e4 1311 vcpu->mmio_fault_cr2 = cr2;
6aa8b732
AK		 1312 kvm_arch_ops->cache_regs(vcpu);
1313
1314 kvm_arch_ops->get_cs_db_l_bits(vcpu, &cs_db, &cs_l);
1315
1316 emulate_ctxt.vcpu = vcpu;
1317 emulate_ctxt.eflags = kvm_arch_ops->get_rflags(vcpu);
1318 emulate_ctxt.cr2 = cr2;
1319 emulate_ctxt.mode = (emulate_ctxt.eflags & X86_EFLAGS_VM)
1320 ? X86EMUL_MODE_REAL : cs_l
1321 ? X86EMUL_MODE_PROT64 : cs_db
1322 ? X86EMUL_MODE_PROT32 : X86EMUL_MODE_PROT16;
1323
1324 if (emulate_ctxt.mode == X86EMUL_MODE_PROT64) {
1325 emulate_ctxt.cs_base = 0;
1326 emulate_ctxt.ds_base = 0;
1327 emulate_ctxt.es_base = 0;
1328 emulate_ctxt.ss_base = 0;
1329 } else {
1330 emulate_ctxt.cs_base = get_segment_base(vcpu, VCPU_SREG_CS);
1331 emulate_ctxt.ds_base = get_segment_base(vcpu, VCPU_SREG_DS);
1332 emulate_ctxt.es_base = get_segment_base(vcpu, VCPU_SREG_ES);
1333 emulate_ctxt.ss_base = get_segment_base(vcpu, VCPU_SREG_SS);
1334 }
1335
1336 emulate_ctxt.gs_base = get_segment_base(vcpu, VCPU_SREG_GS);
1337 emulate_ctxt.fs_base = get_segment_base(vcpu, VCPU_SREG_FS);
1338
1339 vcpu->mmio_is_write = 0;
e70669ab 1340 vcpu->pio.string = 0;
6aa8b732 1341 r = x86_emulate_memop(&emulate_ctxt, &emulate_ops);
e70669ab
LV		 1342 if (vcpu->pio.string)
1343 return EMULATE_DO_MMIO;
6aa8b732
AK		 1344
1345 if ((r || vcpu->mmio_is_write) && run) {
8fc0d085 1346 run->exit_reason = KVM_EXIT_MMIO;
6aa8b732
AK		 1347 run->mmio.phys_addr = vcpu->mmio_phys_addr;
1348 memcpy(run->mmio.data, vcpu->mmio_data, 8);
1349 run->mmio.len = vcpu->mmio_size;
1350 run->mmio.is_write = vcpu->mmio_is_write;
1351 }
1352
1353 if (r) {
a436036b
AK		 1354 if (kvm_mmu_unprotect_page_virt(vcpu, cr2))
1355 return EMULATE_DONE;
6aa8b732
AK		 1356 if (!vcpu->mmio_needed) {
1357 report_emulation_failure(&emulate_ctxt);
1358 return EMULATE_FAIL;
1359 }
1360 return EMULATE_DO_MMIO;
1361 }
1362
1363 kvm_arch_ops->decache_regs(vcpu);
1364 kvm_arch_ops->set_rflags(vcpu, emulate_ctxt.eflags);
1365
02c83209
AK		 1366 if (vcpu->mmio_is_write) {
1367 vcpu->mmio_needed = 0;
6aa8b732 1368 return EMULATE_DO_MMIO;
02c83209 1369 }
6aa8b732
AK		 1370
1371 return EMULATE_DONE;
1372}
1373EXPORT_SYMBOL_GPL(emulate_instruction);
1374
b6958ce4
ED		 1375/*
1376 * The vCPU has executed a HLT instruction with in-kernel mode enabled.
1377 */
c5ec1534 1378static void kvm_vcpu_block(struct kvm_vcpu *vcpu)
d3bef15f 1379{
b6958ce4
ED		 1380 DECLARE_WAITQUEUE(wait, current);
1381
1382 add_wait_queue(&vcpu->wq, &wait);
1383
1384 /*
1385 * We will block until either an interrupt or a signal wakes us up
1386 */
c5ec1534
HQ		 1387 while (!kvm_cpu_has_interrupt(vcpu)
1388 && !signal_pending(current)
1389 && vcpu->mp_state != VCPU_MP_STATE_RUNNABLE
1390 && vcpu->mp_state != VCPU_MP_STATE_SIPI_RECEIVED) {
b6958ce4
ED		 1391 set_current_state(TASK_INTERRUPTIBLE);
1392 vcpu_put(vcpu);
1393 schedule();
1394 vcpu_load(vcpu);
1395 }
d3bef15f 1396
c5ec1534 1397 __set_current_state(TASK_RUNNING);
b6958ce4 1398 remove_wait_queue(&vcpu->wq, &wait);
b6958ce4
ED		 1399}
1400
1401int kvm_emulate_halt(struct kvm_vcpu *vcpu)
1402{
d3bef15f 1403 ++vcpu->stat.halt_exits;
b6958ce4 1404 if (irqchip_in_kernel(vcpu->kvm)) {
c5ec1534
HQ		 1405 vcpu->mp_state = VCPU_MP_STATE_HALTED;
1406 kvm_vcpu_block(vcpu);
1407 if (vcpu->mp_state != VCPU_MP_STATE_RUNNABLE)
1408 return -EINTR;
b6958ce4
ED		 1409 return 1;
1410 } else {
1411 vcpu->run->exit_reason = KVM_EXIT_HLT;
1412 return 0;
1413 }
d3bef15f
AK		 1414}
1415EXPORT_SYMBOL_GPL(kvm_emulate_halt);
1416
270fd9b9
AK		 1417int kvm_hypercall(struct kvm_vcpu *vcpu, struct kvm_run *run)
1418{
1419 unsigned long nr, a0, a1, a2, a3, a4, a5, ret;
1420
9b22bf57 1421 kvm_arch_ops->cache_regs(vcpu);
270fd9b9
AK		 1422 ret = -KVM_EINVAL;
1423#ifdef CONFIG_X86_64
1424 if (is_long_mode(vcpu)) {
1425 nr = vcpu->regs[VCPU_REGS_RAX];
1426 a0 = vcpu->regs[VCPU_REGS_RDI];
1427 a1 = vcpu->regs[VCPU_REGS_RSI];
1428 a2 = vcpu->regs[VCPU_REGS_RDX];
1429 a3 = vcpu->regs[VCPU_REGS_RCX];
1430 a4 = vcpu->regs[VCPU_REGS_R8];
1431 a5 = vcpu->regs[VCPU_REGS_R9];
1432 } else
1433#endif
1434 {
1435 nr = vcpu->regs[VCPU_REGS_RBX] & -1u;
1436 a0 = vcpu->regs[VCPU_REGS_RAX] & -1u;
1437 a1 = vcpu->regs[VCPU_REGS_RCX] & -1u;
1438 a2 = vcpu->regs[VCPU_REGS_RDX] & -1u;
1439 a3 = vcpu->regs[VCPU_REGS_RSI] & -1u;
1440 a4 = vcpu->regs[VCPU_REGS_RDI] & -1u;
1441 a5 = vcpu->regs[VCPU_REGS_RBP] & -1u;
1442 }
1443 switch (nr) {
1444 default:
519ef353 1445 run->hypercall.nr = nr;
b4e63f56
AK		 1446 run->hypercall.args[0] = a0;
1447 run->hypercall.args[1] = a1;
1448 run->hypercall.args[2] = a2;
1449 run->hypercall.args[3] = a3;
1450 run->hypercall.args[4] = a4;
1451 run->hypercall.args[5] = a5;
1452 run->hypercall.ret = ret;
1453 run->hypercall.longmode = is_long_mode(vcpu);
1454 kvm_arch_ops->decache_regs(vcpu);
1455 return 0;
270fd9b9
AK		 1456 }
1457 vcpu->regs[VCPU_REGS_RAX] = ret;
9b22bf57 1458 kvm_arch_ops->decache_regs(vcpu);
270fd9b9
AK		 1459 return 1;
1460}
1461EXPORT_SYMBOL_GPL(kvm_hypercall);
1462
6aa8b732
AK		 1463static u64 mk_cr_64(u64 curr_cr, u32 new_val)
1464{
1465 return (curr_cr & ~((1ULL << 32) - 1)) | new_val;
1466}
1467
1468void realmode_lgdt(struct kvm_vcpu *vcpu, u16 limit, unsigned long base)
1469{
1470 struct descriptor_table dt = { limit, base };
1471
1472 kvm_arch_ops->set_gdt(vcpu, &dt);
1473}
1474
1475void realmode_lidt(struct kvm_vcpu *vcpu, u16 limit, unsigned long base)
1476{
1477 struct descriptor_table dt = { limit, base };
1478
1479 kvm_arch_ops->set_idt(vcpu, &dt);
1480}
1481
1482void realmode_lmsw(struct kvm_vcpu *vcpu, unsigned long msw,
1483 unsigned long *rflags)
1484{
1485 lmsw(vcpu, msw);
1486 *rflags = kvm_arch_ops->get_rflags(vcpu);
1487}
1488
1489unsigned long realmode_get_cr(struct kvm_vcpu *vcpu, int cr)
1490{
25c4c276 1491 kvm_arch_ops->decache_cr4_guest_bits(vcpu);
6aa8b732
AK		 1492 switch (cr) {
1493 case 0:
1494 return vcpu->cr0;
1495 case 2:
1496 return vcpu->cr2;
1497 case 3:
1498 return vcpu->cr3;
1499 case 4:
1500 return vcpu->cr4;
1501 default:
1502 vcpu_printf(vcpu, "%s: unexpected cr %u\n", __FUNCTION__, cr);
1503 return 0;
1504 }
1505}
1506
1507void realmode_set_cr(struct kvm_vcpu *vcpu, int cr, unsigned long val,
1508 unsigned long *rflags)
1509{
1510 switch (cr) {
1511 case 0:
1512 set_cr0(vcpu, mk_cr_64(vcpu->cr0, val));
1513 *rflags = kvm_arch_ops->get_rflags(vcpu);
1514 break;
1515 case 2:
1516 vcpu->cr2 = val;
1517 break;
1518 case 3:
1519 set_cr3(vcpu, val);
1520 break;
1521 case 4:
1522 set_cr4(vcpu, mk_cr_64(vcpu->cr4, val));
1523 break;
1524 default:
1525 vcpu_printf(vcpu, "%s: unexpected cr %u\n", __FUNCTION__, cr);
1526 }
1527}
1528
102d8325
IM		 1529/*
1530 * Register the para guest with the host:
1531 */
1532static int vcpu_register_para(struct kvm_vcpu *vcpu, gpa_t para_state_gpa)
1533{
1534 struct kvm_vcpu_para_state *para_state;
1535 hpa_t para_state_hpa, hypercall_hpa;
1536 struct page *para_state_page;
1537 unsigned char *hypercall;
1538 gpa_t hypercall_gpa;
1539
1540 printk(KERN_DEBUG "kvm: guest trying to enter paravirtual mode\n");
1541 printk(KERN_DEBUG ".... para_state_gpa: %08Lx\n", para_state_gpa);
1542
1543 /*
1544 * Needs to be page aligned:
1545 */
1546 if (para_state_gpa != PAGE_ALIGN(para_state_gpa))
1547 goto err_gp;
1548
1549 para_state_hpa = gpa_to_hpa(vcpu, para_state_gpa);
1550 printk(KERN_DEBUG ".... para_state_hpa: %08Lx\n", para_state_hpa);
1551 if (is_error_hpa(para_state_hpa))
1552 goto err_gp;
1553
ab51a434 1554 mark_page_dirty(vcpu->kvm, para_state_gpa >> PAGE_SHIFT);
102d8325 1555 para_state_page = pfn_to_page(para_state_hpa >> PAGE_SHIFT);
fe551881 1556 para_state = kmap(para_state_page);
102d8325
IM		 1557
1558 printk(KERN_DEBUG ".... guest version: %d\n", para_state->guest_version);
1559 printk(KERN_DEBUG ".... size: %d\n", para_state->size);
1560
1561 para_state->host_version = KVM_PARA_API_VERSION;
1562 /*
1563 * We cannot support guests that try to register themselves
1564 * with a newer API version than the host supports:
1565 */
1566 if (para_state->guest_version > KVM_PARA_API_VERSION) {
1567 para_state->ret = -KVM_EINVAL;
1568 goto err_kunmap_skip;
1569 }
1570
1571 hypercall_gpa = para_state->hypercall_gpa;
1572 hypercall_hpa = gpa_to_hpa(vcpu, hypercall_gpa);
1573 printk(KERN_DEBUG ".... hypercall_hpa: %08Lx\n", hypercall_hpa);
1574 if (is_error_hpa(hypercall_hpa)) {
1575 para_state->ret = -KVM_EINVAL;
1576 goto err_kunmap_skip;
1577 }
1578
1579 printk(KERN_DEBUG "kvm: para guest successfully registered.\n");
1580 vcpu->para_state_page = para_state_page;
1581 vcpu->para_state_gpa = para_state_gpa;
1582 vcpu->hypercall_gpa = hypercall_gpa;
1583
ab51a434 1584 mark_page_dirty(vcpu->kvm, hypercall_gpa >> PAGE_SHIFT);
102d8325
IM		 1585 hypercall = kmap_atomic(pfn_to_page(hypercall_hpa >> PAGE_SHIFT),
1586 KM_USER1) + (hypercall_hpa & ~PAGE_MASK);
1587 kvm_arch_ops->patch_hypercall(vcpu, hypercall);
1588 kunmap_atomic(hypercall, KM_USER1);
1589
1590 para_state->ret = 0;
1591err_kunmap_skip:
fe551881 1592 kunmap(para_state_page);
102d8325
IM		 1593 return 0;
1594err_gp:
1595 return 1;
1596}
1597
3bab1f5d
AK		 1598int kvm_get_msr_common(struct kvm_vcpu *vcpu, u32 msr, u64 *pdata)
1599{
1600 u64 data;
1601
1602 switch (msr) {
1603 case 0xc0010010: /* SYSCFG */
1604 case 0xc0010015: /* HWCR */
1605 case MSR_IA32_PLATFORM_ID:
1606 case MSR_IA32_P5_MC_ADDR:
1607 case MSR_IA32_P5_MC_TYPE:
1608 case MSR_IA32_MC0_CTL:
1609 case MSR_IA32_MCG_STATUS:
1610 case MSR_IA32_MCG_CAP:
1611 case MSR_IA32_MC0_MISC:
1612 case MSR_IA32_MC0_MISC+4:
1613 case MSR_IA32_MC0_MISC+8:
1614 case MSR_IA32_MC0_MISC+12:
1615 case MSR_IA32_MC0_MISC+16:
1616 case MSR_IA32_UCODE_REV:
a8d13ea2 1617 case MSR_IA32_PERF_STATUS:
2dc7094b 1618 case MSR_IA32_EBL_CR_POWERON:
3bab1f5d
AK		 1619 /* MTRR registers */
1620 case 0xfe:
1621 case 0x200 ... 0x2ff:
1622 data = 0;
1623 break;
a8d13ea2
AK		 1624 case 0xcd: /* fsb frequency */
1625 data = 3;
1626 break;
3bab1f5d 1627 case MSR_IA32_APICBASE:
7017fc3d 1628 data = kvm_get_apic_base(vcpu);
3bab1f5d 1629 break;
6f00e68f
AK		 1630 case MSR_IA32_MISC_ENABLE:
1631 data = vcpu->ia32_misc_enable_msr;
1632 break;
3bab1f5d
AK		 1633#ifdef CONFIG_X86_64
1634 case MSR_EFER:
1635 data = vcpu->shadow_efer;
1636 break;
1637#endif
1638 default:
f0242478 1639 pr_unimpl(vcpu, "unhandled rdmsr: 0x%x\n", msr);
3bab1f5d
AK		 1640 return 1;
1641 }
1642 *pdata = data;
1643 return 0;
1644}
1645EXPORT_SYMBOL_GPL(kvm_get_msr_common);
1646
6aa8b732
AK		 1647/*
1648 * Reads an msr value (of 'msr_index') into 'pdata'.
1649 * Returns 0 on success, non-0 otherwise.
1650 * Assumes vcpu_load() was already called.
1651 */
35f3f286 1652int kvm_get_msr(struct kvm_vcpu *vcpu, u32 msr_index, u64 *pdata)
6aa8b732
AK		 1653{
1654 return kvm_arch_ops->get_msr(vcpu, msr_index, pdata);
1655}
1656
05b3e0c2 1657#ifdef CONFIG_X86_64
6aa8b732 1658
3bab1f5d 1659static void set_efer(struct kvm_vcpu *vcpu, u64 efer)
6aa8b732 1660{
6aa8b732
AK		 1661 if (efer & EFER_RESERVED_BITS) {
1662 printk(KERN_DEBUG "set_efer: 0x%llx #GP, reserved bits\n",
1663 efer);
1664 inject_gp(vcpu);
1665 return;
1666 }
1667
1668 if (is_paging(vcpu)
1669 && (vcpu->shadow_efer & EFER_LME) != (efer & EFER_LME)) {
1670 printk(KERN_DEBUG "set_efer: #GP, change LME while paging\n");
1671 inject_gp(vcpu);
1672 return;
1673 }
1674
7725f0ba
AK		 1675 kvm_arch_ops->set_efer(vcpu, efer);
1676
6aa8b732
AK		 1677 efer &= ~EFER_LMA;
1678 efer |= vcpu->shadow_efer & EFER_LMA;
1679
1680 vcpu->shadow_efer = efer;
6aa8b732 1681}
6aa8b732
AK		 1682
1683#endif
1684
3bab1f5d
AK		 1685int kvm_set_msr_common(struct kvm_vcpu *vcpu, u32 msr, u64 data)
1686{
1687 switch (msr) {
1688#ifdef CONFIG_X86_64
1689 case MSR_EFER:
1690 set_efer(vcpu, data);
1691 break;
1692#endif
1693 case MSR_IA32_MC0_STATUS:
f0242478 1694 pr_unimpl(vcpu, "%s: MSR_IA32_MC0_STATUS 0x%llx, nop\n",
3bab1f5d
AK		 1695 __FUNCTION__, data);
1696 break;
0e5bf0d0 1697 case MSR_IA32_MCG_STATUS:
f0242478 1698 pr_unimpl(vcpu, "%s: MSR_IA32_MCG_STATUS 0x%llx, nop\n",
0e5bf0d0
SK		 1699 __FUNCTION__, data);
1700 break;
3bab1f5d
AK		 1701 case MSR_IA32_UCODE_REV:
1702 case MSR_IA32_UCODE_WRITE:
1703 case 0x200 ... 0x2ff: /* MTRRs */
1704 break;
1705 case MSR_IA32_APICBASE:
7017fc3d 1706 kvm_set_apic_base(vcpu, data);
3bab1f5d 1707 break;
6f00e68f
AK		 1708 case MSR_IA32_MISC_ENABLE:
1709 vcpu->ia32_misc_enable_msr = data;
1710 break;
102d8325
IM		 1711 /*
1712 * This is the 'probe whether the host is KVM' logic:
1713 */
1714 case MSR_KVM_API_MAGIC:
1715 return vcpu_register_para(vcpu, data);
1716
3bab1f5d 1717 default:
f0242478 1718 pr_unimpl(vcpu, "unhandled wrmsr: 0x%x\n", msr);
3bab1f5d
AK		 1719 return 1;
1720 }
1721 return 0;
1722}
1723EXPORT_SYMBOL_GPL(kvm_set_msr_common);
1724
6aa8b732
AK		 1725/*
1726 * Writes msr value into into the appropriate "register".
1727 * Returns 0 on success, non-0 otherwise.
1728 * Assumes vcpu_load() was already called.
1729 */
35f3f286 1730int kvm_set_msr(struct kvm_vcpu *vcpu, u32 msr_index, u64 data)
6aa8b732
AK		 1731{
1732 return kvm_arch_ops->set_msr(vcpu, msr_index, data);
1733}
1734
1735void kvm_resched(struct kvm_vcpu *vcpu)
1736{
3fca0365
YD		 1737 if (!need_resched())
1738 return;
6aa8b732 1739 cond_resched();
6aa8b732
AK		 1740}
1741EXPORT_SYMBOL_GPL(kvm_resched);
1742
06465c5a
AK		 1743void kvm_emulate_cpuid(struct kvm_vcpu *vcpu)
1744{
1745 int i;
1746 u32 function;
1747 struct kvm_cpuid_entry *e, *best;
1748
1749 kvm_arch_ops->cache_regs(vcpu);
1750 function = vcpu->regs[VCPU_REGS_RAX];
1751 vcpu->regs[VCPU_REGS_RAX] = 0;
1752 vcpu->regs[VCPU_REGS_RBX] = 0;
1753 vcpu->regs[VCPU_REGS_RCX] = 0;
1754 vcpu->regs[VCPU_REGS_RDX] = 0;
1755 best = NULL;
1756 for (i = 0; i < vcpu->cpuid_nent; ++i) {
1757 e = &vcpu->cpuid_entries[i];
1758 if (e->function == function) {
1759 best = e;
1760 break;
1761 }
1762 /*
1763 * Both basic or both extended?
1764 */
1765 if (((e->function ^ function) & 0x80000000) == 0)
1766 if (!best || e->function > best->function)
1767 best = e;
1768 }
1769 if (best) {
1770 vcpu->regs[VCPU_REGS_RAX] = best->eax;
1771 vcpu->regs[VCPU_REGS_RBX] = best->ebx;
1772 vcpu->regs[VCPU_REGS_RCX] = best->ecx;
1773 vcpu->regs[VCPU_REGS_RDX] = best->edx;
1774 }
1775 kvm_arch_ops->decache_regs(vcpu);
1776 kvm_arch_ops->skip_emulated_instruction(vcpu);
1777}
1778EXPORT_SYMBOL_GPL(kvm_emulate_cpuid);
1779
039576c0 1780static int pio_copy_data(struct kvm_vcpu *vcpu)
46fc1477 1781{
039576c0
AK		 1782 void *p = vcpu->pio_data;
1783 void *q;
1784 unsigned bytes;
1785 int nr_pages = vcpu->pio.guest_pages[1] ? 2 : 1;
1786
039576c0
AK		 1787 q = vmap(vcpu->pio.guest_pages, nr_pages, VM_READ|VM_WRITE,
1788 PAGE_KERNEL);
1789 if (!q) {
039576c0
AK		 1790 free_pio_guest_pages(vcpu);
1791 return -ENOMEM;
1792 }
1793 q += vcpu->pio.guest_page_offset;
1794 bytes = vcpu->pio.size * vcpu->pio.cur_count;
1795 if (vcpu->pio.in)
1796 memcpy(q, p, bytes);
1797 else
1798 memcpy(p, q, bytes);
1799 q -= vcpu->pio.guest_page_offset;
1800 vunmap(q);
039576c0
AK		 1801 free_pio_guest_pages(vcpu);
1802 return 0;
1803}
1804
1805static int complete_pio(struct kvm_vcpu *vcpu)
1806{
1807 struct kvm_pio_request *io = &vcpu->pio;
46fc1477 1808 long delta;
039576c0 1809 int r;
46fc1477
AK		 1810
1811 kvm_arch_ops->cache_regs(vcpu);
1812
1813 if (!io->string) {
039576c0
AK		 1814 if (io->in)
1815 memcpy(&vcpu->regs[VCPU_REGS_RAX], vcpu->pio_data,
46fc1477
AK		 1816 io->size);
1817 } else {
039576c0
AK		 1818 if (io->in) {
1819 r = pio_copy_data(vcpu);
1820 if (r) {
1821 kvm_arch_ops->cache_regs(vcpu);
1822 return r;
1823 }
1824 }
1825
46fc1477
AK		 1826 delta = 1;
1827 if (io->rep) {
039576c0 1828 delta *= io->cur_count;
46fc1477
AK		 1829 /*
1830 * The size of the register should really depend on
1831 * current address size.
1832 */
1833 vcpu->regs[VCPU_REGS_RCX] -= delta;
1834 }
039576c0 1835 if (io->down)
46fc1477
AK		 1836 delta = -delta;
1837 delta *= io->size;
039576c0 1838 if (io->in)
46fc1477
AK		 1839 vcpu->regs[VCPU_REGS_RDI] += delta;
1840 else
1841 vcpu->regs[VCPU_REGS_RSI] += delta;
1842 }
1843
46fc1477
AK		 1844 kvm_arch_ops->decache_regs(vcpu);
1845
039576c0
AK		 1846 io->count -= io->cur_count;
1847 io->cur_count = 0;
1848
1849 if (!io->count)
1850 kvm_arch_ops->skip_emulated_instruction(vcpu);
1851 return 0;
46fc1477
AK		 1852}
1853
65619eb5
ED		 1854static void kernel_pio(struct kvm_io_device *pio_dev,
1855 struct kvm_vcpu *vcpu,
1856 void *pd)
74906345
ED		 1857{
1858 /* TODO: String I/O for in kernel device */
1859
9cf98828 1860 mutex_lock(&vcpu->kvm->lock);
74906345
ED		 1861 if (vcpu->pio.in)
1862 kvm_iodevice_read(pio_dev, vcpu->pio.port,
1863 vcpu->pio.size,
65619eb5 1864 pd);
74906345
ED		 1865 else
1866 kvm_iodevice_write(pio_dev, vcpu->pio.port,
1867 vcpu->pio.size,
65619eb5 1868 pd);
9cf98828 1869 mutex_unlock(&vcpu->kvm->lock);
65619eb5
ED		 1870}
1871
1872static void pio_string_write(struct kvm_io_device *pio_dev,
1873 struct kvm_vcpu *vcpu)
1874{
1875 struct kvm_pio_request *io = &vcpu->pio;
1876 void *pd = vcpu->pio_data;
1877 int i;
1878
9cf98828 1879 mutex_lock(&vcpu->kvm->lock);
65619eb5
ED		 1880 for (i = 0; i < io->cur_count; i++) {
1881 kvm_iodevice_write(pio_dev, io->port,
1882 io->size,
1883 pd);
1884 pd += io->size;
1885 }
9cf98828 1886 mutex_unlock(&vcpu->kvm->lock);
74906345
ED		 1887}
1888
3090dd73
LV		 1889int kvm_emulate_pio (struct kvm_vcpu *vcpu, struct kvm_run *run, int in,
1890 int size, unsigned port)
1891{
1892 struct kvm_io_device *pio_dev;
1893
1894 vcpu->run->exit_reason = KVM_EXIT_IO;
1895 vcpu->run->io.direction = in ? KVM_EXIT_IO_IN : KVM_EXIT_IO_OUT;
1896 vcpu->run->io.size = vcpu->pio.size = size;
1897 vcpu->run->io.data_offset = KVM_PIO_PAGE_OFFSET * PAGE_SIZE;
1898 vcpu->run->io.count = vcpu->pio.count = vcpu->pio.cur_count = 1;
1899 vcpu->run->io.port = vcpu->pio.port = port;
1900 vcpu->pio.in = in;
1901 vcpu->pio.string = 0;
1902 vcpu->pio.down = 0;
1903 vcpu->pio.guest_page_offset = 0;
1904 vcpu->pio.rep = 0;
1905
1906 kvm_arch_ops->cache_regs(vcpu);
1907 memcpy(vcpu->pio_data, &vcpu->regs[VCPU_REGS_RAX], 4);
1908 kvm_arch_ops->decache_regs(vcpu);
1909
1910 pio_dev = vcpu_find_pio_dev(vcpu, port);
1911 if (pio_dev) {
1912 kernel_pio(pio_dev, vcpu, vcpu->pio_data);
1913 complete_pio(vcpu);
1914 return 1;
1915 }
1916 return 0;
1917}
1918EXPORT_SYMBOL_GPL(kvm_emulate_pio);
1919
1920int kvm_emulate_pio_string(struct kvm_vcpu *vcpu, struct kvm_run *run, int in,
1921 int size, unsigned long count, int down,
039576c0
AK		 1922 gva_t address, int rep, unsigned port)
1923{
1924 unsigned now, in_page;
65619eb5 1925 int i, ret = 0;
039576c0
AK		 1926 int nr_pages = 1;
1927 struct page *page;
74906345 1928 struct kvm_io_device *pio_dev;
039576c0
AK		 1929
1930 vcpu->run->exit_reason = KVM_EXIT_IO;
1931 vcpu->run->io.direction = in ? KVM_EXIT_IO_IN : KVM_EXIT_IO_OUT;
3090dd73 1932 vcpu->run->io.size = vcpu->pio.size = size;
039576c0 1933 vcpu->run->io.data_offset = KVM_PIO_PAGE_OFFSET * PAGE_SIZE;
3090dd73
LV		 1934 vcpu->run->io.count = vcpu->pio.count = vcpu->pio.cur_count = count;
1935 vcpu->run->io.port = vcpu->pio.port = port;
039576c0 1936 vcpu->pio.in = in;
3090dd73 1937 vcpu->pio.string = 1;
039576c0
AK		 1938 vcpu->pio.down = down;
1939 vcpu->pio.guest_page_offset = offset_in_page(address);
1940 vcpu->pio.rep = rep;
1941
039576c0
AK		 1942 if (!count) {
1943 kvm_arch_ops->skip_emulated_instruction(vcpu);
1944 return 1;
1945 }
1946
039576c0
AK		 1947 if (!down)
1948 in_page = PAGE_SIZE - offset_in_page(address);
1949 else
1950 in_page = offset_in_page(address) + size;
1951 now = min(count, (unsigned long)in_page / size);
1952 if (!now) {
1953 /*
1954 * String I/O straddles page boundary. Pin two guest pages
1955 * so that we satisfy atomicity constraints. Do just one
1956 * transaction to avoid complexity.
1957 */
1958 nr_pages = 2;
1959 now = 1;
1960 }
1961 if (down) {
1962 /*
1963 * String I/O in reverse. Yuck. Kill the guest, fix later.
1964 */
f0242478 1965 pr_unimpl(vcpu, "guest string pio down\n");
039576c0
AK		 1966 inject_gp(vcpu);
1967 return 1;
1968 }
1969 vcpu->run->io.count = now;
1970 vcpu->pio.cur_count = now;
1971
1972 for (i = 0; i < nr_pages; ++i) {
11ec2804 1973 mutex_lock(&vcpu->kvm->lock);
039576c0
AK		 1974 page = gva_to_page(vcpu, address + i * PAGE_SIZE);
1975 if (page)
1976 get_page(page);
1977 vcpu->pio.guest_pages[i] = page;
11ec2804 1978 mutex_unlock(&vcpu->kvm->lock);
039576c0
AK		 1979 if (!page) {
1980 inject_gp(vcpu);
1981 free_pio_guest_pages(vcpu);
1982 return 1;
1983 }
1984 }
1985
3090dd73 1986 pio_dev = vcpu_find_pio_dev(vcpu, port);
65619eb5
ED		 1987 if (!vcpu->pio.in) {
1988 /* string PIO write */
1989 ret = pio_copy_data(vcpu);
1990 if (ret >= 0 && pio_dev) {
1991 pio_string_write(pio_dev, vcpu);
1992 complete_pio(vcpu);
1993 if (vcpu->pio.count == 0)
1994 ret = 1;
1995 }
1996 } else if (pio_dev)
f0242478 1997 pr_unimpl(vcpu, "no string pio read support yet, "
65619eb5
ED		 1998 "port %x size %d count %ld\n",
1999 port, size, count);
2000
2001 return ret;
039576c0 2002}
3090dd73 2003EXPORT_SYMBOL_GPL(kvm_emulate_pio_string);
039576c0 2004
bccf2150 2005static int kvm_vcpu_ioctl_run(struct kvm_vcpu *vcpu, struct kvm_run *kvm_run)
6aa8b732 2006{
6aa8b732 2007 int r;
1961d276 2008 sigset_t sigsaved;
6aa8b732 2009
bccf2150 2010 vcpu_load(vcpu);
6aa8b732 2011
c5ec1534
HQ		 2012 if (unlikely(vcpu->mp_state == VCPU_MP_STATE_UNINITIALIZED)) {
2013 kvm_vcpu_block(vcpu);
2014 vcpu_put(vcpu);
2015 return -EAGAIN;
2016 }
2017
1961d276
AK		 2018 if (vcpu->sigset_active)
2019 sigprocmask(SIG_SETMASK, &vcpu->sigset, &sigsaved);
2020
54810342 2021 /* re-sync apic's tpr */
5cd4f6fd
HQ		 2022 if (!irqchip_in_kernel(vcpu->kvm))
2023 set_cr8(vcpu, kvm_run->cr8);
54810342 2024
02c83209
AK		 2025 if (vcpu->pio.cur_count) {
2026 r = complete_pio(vcpu);
2027 if (r)
2028 goto out;
2029 }
2030
2031 if (vcpu->mmio_needed) {
2032 memcpy(vcpu->mmio_data, kvm_run->mmio.data, 8);
2033 vcpu->mmio_read_completed = 1;
2034 vcpu->mmio_needed = 0;
2035 r = emulate_instruction(vcpu, kvm_run,
2036 vcpu->mmio_fault_cr2, 0);
2037 if (r == EMULATE_DO_MMIO) {
2038 /*
2039 * Read-modify-write. Back to userspace.
2040 */
02c83209
AK		 2041 r = 0;
2042 goto out;
46fc1477 2043 }
6aa8b732
AK		 2044 }
2045
8eb7d334 2046 if (kvm_run->exit_reason == KVM_EXIT_HYPERCALL) {
b4e63f56
AK		 2047 kvm_arch_ops->cache_regs(vcpu);
2048 vcpu->regs[VCPU_REGS_RAX] = kvm_run->hypercall.ret;
2049 kvm_arch_ops->decache_regs(vcpu);
2050 }
2051
6aa8b732
AK		 2052 r = kvm_arch_ops->run(vcpu, kvm_run);
2053
039576c0 2054out:
1961d276
AK		 2055 if (vcpu->sigset_active)
2056 sigprocmask(SIG_SETMASK, &sigsaved, NULL);
2057
6aa8b732
AK		 2058 vcpu_put(vcpu);
2059 return r;
2060}
2061
bccf2150
AK		 2062static int kvm_vcpu_ioctl_get_regs(struct kvm_vcpu *vcpu,
2063 struct kvm_regs *regs)
6aa8b732 2064{
bccf2150 2065 vcpu_load(vcpu);
6aa8b732
AK		 2066
2067 kvm_arch_ops->cache_regs(vcpu);
2068
2069 regs->rax = vcpu->regs[VCPU_REGS_RAX];
2070 regs->rbx = vcpu->regs[VCPU_REGS_RBX];
2071 regs->rcx = vcpu->regs[VCPU_REGS_RCX];
2072 regs->rdx = vcpu->regs[VCPU_REGS_RDX];
2073 regs->rsi = vcpu->regs[VCPU_REGS_RSI];
2074 regs->rdi = vcpu->regs[VCPU_REGS_RDI];
2075 regs->rsp = vcpu->regs[VCPU_REGS_RSP];
2076 regs->rbp = vcpu->regs[VCPU_REGS_RBP];
05b3e0c2 2077#ifdef CONFIG_X86_64
6aa8b732
AK		 2078 regs->r8 = vcpu->regs[VCPU_REGS_R8];
2079 regs->r9 = vcpu->regs[VCPU_REGS_R9];
2080 regs->r10 = vcpu->regs[VCPU_REGS_R10];
2081 regs->r11 = vcpu->regs[VCPU_REGS_R11];
2082 regs->r12 = vcpu->regs[VCPU_REGS_R12];
2083 regs->r13 = vcpu->regs[VCPU_REGS_R13];
2084 regs->r14 = vcpu->regs[VCPU_REGS_R14];
2085 regs->r15 = vcpu->regs[VCPU_REGS_R15];
2086#endif
2087
2088 regs->rip = vcpu->rip;
2089 regs->rflags = kvm_arch_ops->get_rflags(vcpu);
2090
2091 /*
2092 * Don't leak debug flags in case they were set for guest debugging
2093 */
2094 if (vcpu->guest_debug.enabled && vcpu->guest_debug.singlestep)
2095 regs->rflags &= ~(X86_EFLAGS_TF | X86_EFLAGS_RF);
2096
2097 vcpu_put(vcpu);
2098
2099 return 0;
2100}
2101
bccf2150
AK		 2102static int kvm_vcpu_ioctl_set_regs(struct kvm_vcpu *vcpu,
2103 struct kvm_regs *regs)
6aa8b732 2104{
bccf2150 2105 vcpu_load(vcpu);
6aa8b732
AK		 2106
2107 vcpu->regs[VCPU_REGS_RAX] = regs->rax;
2108 vcpu->regs[VCPU_REGS_RBX] = regs->rbx;
2109 vcpu->regs[VCPU_REGS_RCX] = regs->rcx;
2110 vcpu->regs[VCPU_REGS_RDX] = regs->rdx;
2111 vcpu->regs[VCPU_REGS_RSI] = regs->rsi;
2112 vcpu->regs[VCPU_REGS_RDI] = regs->rdi;
2113 vcpu->regs[VCPU_REGS_RSP] = regs->rsp;
2114 vcpu->regs[VCPU_REGS_RBP] = regs->rbp;
05b3e0c2 2115#ifdef CONFIG_X86_64
6aa8b732
AK		 2116 vcpu->regs[VCPU_REGS_R8] = regs->r8;
2117 vcpu->regs[VCPU_REGS_R9] = regs->r9;
2118 vcpu->regs[VCPU_REGS_R10] = regs->r10;
2119 vcpu->regs[VCPU_REGS_R11] = regs->r11;
2120 vcpu->regs[VCPU_REGS_R12] = regs->r12;
2121 vcpu->regs[VCPU_REGS_R13] = regs->r13;
2122 vcpu->regs[VCPU_REGS_R14] = regs->r14;
2123 vcpu->regs[VCPU_REGS_R15] = regs->r15;
2124#endif
2125
2126 vcpu->rip = regs->rip;
2127 kvm_arch_ops->set_rflags(vcpu, regs->rflags);
2128
2129 kvm_arch_ops->decache_regs(vcpu);
2130
2131 vcpu_put(vcpu);
2132
2133 return 0;
2134}
2135
2136static void get_segment(struct kvm_vcpu *vcpu,
2137 struct kvm_segment *var, int seg)
2138{
2139 return kvm_arch_ops->get_segment(vcpu, var, seg);
2140}
2141
bccf2150
AK		 2142static int kvm_vcpu_ioctl_get_sregs(struct kvm_vcpu *vcpu,
2143 struct kvm_sregs *sregs)
6aa8b732 2144{
6aa8b732 2145 struct descriptor_table dt;
2a8067f1 2146 int pending_vec;
6aa8b732 2147
bccf2150 2148 vcpu_load(vcpu);
6aa8b732
AK		 2149
2150 get_segment(vcpu, &sregs->cs, VCPU_SREG_CS);
2151 get_segment(vcpu, &sregs->ds, VCPU_SREG_DS);
2152 get_segment(vcpu, &sregs->es, VCPU_SREG_ES);
2153 get_segment(vcpu, &sregs->fs, VCPU_SREG_FS);
2154 get_segment(vcpu, &sregs->gs, VCPU_SREG_GS);
2155 get_segment(vcpu, &sregs->ss, VCPU_SREG_SS);
2156
2157 get_segment(vcpu, &sregs->tr, VCPU_SREG_TR);
2158 get_segment(vcpu, &sregs->ldt, VCPU_SREG_LDTR);
2159
2160 kvm_arch_ops->get_idt(vcpu, &dt);
2161 sregs->idt.limit = dt.limit;
2162 sregs->idt.base = dt.base;
2163 kvm_arch_ops->get_gdt(vcpu, &dt);
2164 sregs->gdt.limit = dt.limit;
2165 sregs->gdt.base = dt.base;
2166
25c4c276 2167 kvm_arch_ops->decache_cr4_guest_bits(vcpu);
6aa8b732
AK		 2168 sregs->cr0 = vcpu->cr0;
2169 sregs->cr2 = vcpu->cr2;
2170 sregs->cr3 = vcpu->cr3;
2171 sregs->cr4 = vcpu->cr4;
7017fc3d 2172 sregs->cr8 = get_cr8(vcpu);
6aa8b732 2173 sregs->efer = vcpu->shadow_efer;
7017fc3d 2174 sregs->apic_base = kvm_get_apic_base(vcpu);
6aa8b732 2175
2a8067f1 2176 if (irqchip_in_kernel(vcpu->kvm)) {
c52fb35a
HQ		 2177 memset(sregs->interrupt_bitmap, 0,
2178 sizeof sregs->interrupt_bitmap);
2a8067f1
ED		 2179 pending_vec = kvm_arch_ops->get_irq(vcpu);
2180 if (pending_vec >= 0)
2181 set_bit(pending_vec, (unsigned long *)sregs->interrupt_bitmap);
2182 } else
c52fb35a
HQ		 2183 memcpy(sregs->interrupt_bitmap, vcpu->irq_pending,
2184 sizeof sregs->interrupt_bitmap);
6aa8b732
AK		 2185
2186 vcpu_put(vcpu);
2187
2188 return 0;
2189}
2190
2191static void set_segment(struct kvm_vcpu *vcpu,
2192 struct kvm_segment *var, int seg)
2193{
2194 return kvm_arch_ops->set_segment(vcpu, var, seg);
2195}
2196
bccf2150
AK		 2197static int kvm_vcpu_ioctl_set_sregs(struct kvm_vcpu *vcpu,
2198 struct kvm_sregs *sregs)
6aa8b732 2199{
6aa8b732 2200 int mmu_reset_needed = 0;
2a8067f1 2201 int i, pending_vec, max_bits;
6aa8b732
AK		 2202 struct descriptor_table dt;
2203
bccf2150 2204 vcpu_load(vcpu);
6aa8b732 2205
6aa8b732
AK		 2206 dt.limit = sregs->idt.limit;
2207 dt.base = sregs->idt.base;
2208 kvm_arch_ops->set_idt(vcpu, &dt);
2209 dt.limit = sregs->gdt.limit;
2210 dt.base = sregs->gdt.base;
2211 kvm_arch_ops->set_gdt(vcpu, &dt);
2212
2213 vcpu->cr2 = sregs->cr2;
2214 mmu_reset_needed |= vcpu->cr3 != sregs->cr3;
2215 vcpu->cr3 = sregs->cr3;
2216
7017fc3d 2217 set_cr8(vcpu, sregs->cr8);
6aa8b732
AK		 2218
2219 mmu_reset_needed |= vcpu->shadow_efer != sregs->efer;
05b3e0c2 2220#ifdef CONFIG_X86_64
6aa8b732
AK		 2221 kvm_arch_ops->set_efer(vcpu, sregs->efer);
2222#endif
7017fc3d 2223 kvm_set_apic_base(vcpu, sregs->apic_base);
6aa8b732 2224
25c4c276 2225 kvm_arch_ops->decache_cr4_guest_bits(vcpu);
399badf3 2226
6aa8b732 2227 mmu_reset_needed |= vcpu->cr0 != sregs->cr0;
81f50e3b 2228 vcpu->cr0 = sregs->cr0;
f6528b03 2229 kvm_arch_ops->set_cr0(vcpu, sregs->cr0);
6aa8b732
AK		 2230
2231 mmu_reset_needed |= vcpu->cr4 != sregs->cr4;
2232 kvm_arch_ops->set_cr4(vcpu, sregs->cr4);
1b0973bd
AK		 2233 if (!is_long_mode(vcpu) && is_pae(vcpu))
2234 load_pdptrs(vcpu, vcpu->cr3);
6aa8b732
AK		 2235
2236 if (mmu_reset_needed)
2237 kvm_mmu_reset_context(vcpu);
2238
c52fb35a
HQ		 2239 if (!irqchip_in_kernel(vcpu->kvm)) {
2240 memcpy(vcpu->irq_pending, sregs->interrupt_bitmap,
2241 sizeof vcpu->irq_pending);
2242 vcpu->irq_summary = 0;
2243 for (i = 0; i < ARRAY_SIZE(vcpu->irq_pending); ++i)
2244 if (vcpu->irq_pending[i])
2245 __set_bit(i, &vcpu->irq_summary);
2a8067f1
ED		 2246 } else {
2247 max_bits = (sizeof sregs->interrupt_bitmap) << 3;
2248 pending_vec = find_first_bit(
2249 (const unsigned long *)sregs->interrupt_bitmap,
2250 max_bits);
2251 /* Only pending external irq is handled here */
2252 if (pending_vec < max_bits) {
2253 kvm_arch_ops->set_irq(vcpu, pending_vec);
2254 printk("Set back pending irq %d\n", pending_vec);
2255 }
c52fb35a 2256 }
6aa8b732 2257
024aa1c0
AK		 2258 set_segment(vcpu, &sregs->cs, VCPU_SREG_CS);
2259 set_segment(vcpu, &sregs->ds, VCPU_SREG_DS);
2260 set_segment(vcpu, &sregs->es, VCPU_SREG_ES);
2261 set_segment(vcpu, &sregs->fs, VCPU_SREG_FS);
2262 set_segment(vcpu, &sregs->gs, VCPU_SREG_GS);
2263 set_segment(vcpu, &sregs->ss, VCPU_SREG_SS);
2264
2265 set_segment(vcpu, &sregs->tr, VCPU_SREG_TR);
2266 set_segment(vcpu, &sregs->ldt, VCPU_SREG_LDTR);
2267
6aa8b732
AK		 2268 vcpu_put(vcpu);
2269
2270 return 0;
2271}
2272
1747fb71
RR		 2273void kvm_get_cs_db_l_bits(struct kvm_vcpu *vcpu, int *db, int *l)
2274{
2275 struct kvm_segment cs;
2276
2277 get_segment(vcpu, &cs, VCPU_SREG_CS);
2278 *db = cs.db;
2279 *l = cs.l;
2280}
2281EXPORT_SYMBOL_GPL(kvm_get_cs_db_l_bits);
2282
6aa8b732
AK		 2283/*
2284 * List of msr numbers which we expose to userspace through KVM_GET_MSRS
2285 * and KVM_SET_MSRS, and KVM_GET_MSR_INDEX_LIST.
bf591b24
MR		 2286 *
2287 * This list is modified at module load time to reflect the
2288 * capabilities of the host cpu.
6aa8b732
AK		 2289 */
2290static u32 msrs_to_save[] = {
2291 MSR_IA32_SYSENTER_CS, MSR_IA32_SYSENTER_ESP, MSR_IA32_SYSENTER_EIP,
2292 MSR_K6_STAR,
05b3e0c2 2293#ifdef CONFIG_X86_64
6aa8b732
AK		 2294 MSR_CSTAR, MSR_KERNEL_GS_BASE, MSR_SYSCALL_MASK, MSR_LSTAR,
2295#endif
2296 MSR_IA32_TIME_STAMP_COUNTER,
2297};
2298
bf591b24
MR		 2299static unsigned num_msrs_to_save;
2300
6f00e68f
AK		 2301static u32 emulated_msrs[] = {
2302 MSR_IA32_MISC_ENABLE,
2303};
2304
bf591b24
MR		 2305static __init void kvm_init_msr_list(void)
2306{
2307 u32 dummy[2];
2308 unsigned i, j;
2309
2310 for (i = j = 0; i < ARRAY_SIZE(msrs_to_save); i++) {
2311 if (rdmsr_safe(msrs_to_save[i], &dummy[0], &dummy[1]) < 0)
2312 continue;
2313 if (j < i)
2314 msrs_to_save[j] = msrs_to_save[i];
2315 j++;
2316 }
2317 num_msrs_to_save = j;
2318}
6aa8b732
AK		 2319
2320/*
2321 * Adapt set_msr() to msr_io()'s calling convention
2322 */
2323static int do_set_msr(struct kvm_vcpu *vcpu, unsigned index, u64 *data)
2324{
35f3f286 2325 return kvm_set_msr(vcpu, index, *data);
6aa8b732
AK		 2326}
2327
2328/*
2329 * Read or write a bunch of msrs. All parameters are kernel addresses.
2330 *
2331 * @return number of msrs set successfully.
2332 */
bccf2150 2333static int __msr_io(struct kvm_vcpu *vcpu, struct kvm_msrs *msrs,
6aa8b732
AK		 2334 struct kvm_msr_entry *entries,
2335 int (*do_msr)(struct kvm_vcpu *vcpu,
2336 unsigned index, u64 *data))
2337{
6aa8b732
AK		 2338 int i;
2339
bccf2150 2340 vcpu_load(vcpu);
6aa8b732
AK		 2341
2342 for (i = 0; i < msrs->nmsrs; ++i)
2343 if (do_msr(vcpu, entries[i].index, &entries[i].data))
2344 break;
2345
2346 vcpu_put(vcpu);
2347
2348 return i;
2349}
2350
2351/*
2352 * Read or write a bunch of msrs. Parameters are user addresses.
2353 *
2354 * @return number of msrs set successfully.
2355 */
bccf2150 2356static int msr_io(struct kvm_vcpu *vcpu, struct kvm_msrs __user *user_msrs,
6aa8b732
AK		 2357 int (*do_msr)(struct kvm_vcpu *vcpu,
2358 unsigned index, u64 *data),
2359 int writeback)
2360{
2361 struct kvm_msrs msrs;
2362 struct kvm_msr_entry *entries;
2363 int r, n;
2364 unsigned size;
2365
2366 r = -EFAULT;
2367 if (copy_from_user(&msrs, user_msrs, sizeof msrs))
2368 goto out;
2369
2370 r = -E2BIG;
2371 if (msrs.nmsrs >= MAX_IO_MSRS)
2372 goto out;
2373
2374 r = -ENOMEM;
2375 size = sizeof(struct kvm_msr_entry) * msrs.nmsrs;
2376 entries = vmalloc(size);
2377 if (!entries)
2378 goto out;
2379
2380 r = -EFAULT;
2381 if (copy_from_user(entries, user_msrs->entries, size))
2382 goto out_free;
2383
bccf2150 2384 r = n = __msr_io(vcpu, &msrs, entries, do_msr);
6aa8b732
AK		 2385 if (r < 0)
2386 goto out_free;
2387
2388 r = -EFAULT;
2389 if (writeback && copy_to_user(user_msrs->entries, entries, size))
2390 goto out_free;
2391
2392 r = n;
2393
2394out_free:
2395 vfree(entries);
2396out:
2397 return r;
2398}
2399
2400/*
2401 * Translate a guest virtual address to a guest physical address.
2402 */
bccf2150
AK		 2403static int kvm_vcpu_ioctl_translate(struct kvm_vcpu *vcpu,
2404 struct kvm_translation *tr)
6aa8b732
AK		 2405{
2406 unsigned long vaddr = tr->linear_address;
6aa8b732
AK		 2407 gpa_t gpa;
2408
bccf2150 2409 vcpu_load(vcpu);
11ec2804 2410 mutex_lock(&vcpu->kvm->lock);
6aa8b732
AK		 2411 gpa = vcpu->mmu.gva_to_gpa(vcpu, vaddr);
2412 tr->physical_address = gpa;
2413 tr->valid = gpa != UNMAPPED_GVA;
2414 tr->writeable = 1;
2415 tr->usermode = 0;
11ec2804 2416 mutex_unlock(&vcpu->kvm->lock);
6aa8b732
AK		 2417 vcpu_put(vcpu);
2418
2419 return 0;
2420}
2421
bccf2150
AK		 2422static int kvm_vcpu_ioctl_interrupt(struct kvm_vcpu *vcpu,
2423 struct kvm_interrupt *irq)
6aa8b732 2424{
6aa8b732
AK		 2425 if (irq->irq < 0 || irq->irq >= 256)
2426 return -EINVAL;
97222cc8
ED		 2427 if (irqchip_in_kernel(vcpu->kvm))
2428 return -ENXIO;
bccf2150 2429 vcpu_load(vcpu);
6aa8b732
AK		 2430
2431 set_bit(irq->irq, vcpu->irq_pending);
2432 set_bit(irq->irq / BITS_PER_LONG, &vcpu->irq_summary);
2433
2434 vcpu_put(vcpu);
2435
2436 return 0;
2437}
2438
bccf2150
AK		 2439static int kvm_vcpu_ioctl_debug_guest(struct kvm_vcpu *vcpu,
2440 struct kvm_debug_guest *dbg)
6aa8b732 2441{
6aa8b732
AK		 2442 int r;
2443
bccf2150 2444 vcpu_load(vcpu);
6aa8b732
AK		 2445
2446 r = kvm_arch_ops->set_guest_debug(vcpu, dbg);
2447
2448 vcpu_put(vcpu);
2449
2450 return r;
2451}
2452
9a2bb7f4
AK		 2453static struct page *kvm_vcpu_nopage(struct vm_area_struct *vma,
2454 unsigned long address,
2455 int *type)
2456{
2457 struct kvm_vcpu *vcpu = vma->vm_file->private_data;
2458 unsigned long pgoff;
2459 struct page *page;
2460
9a2bb7f4 2461 pgoff = ((address - vma->vm_start) >> PAGE_SHIFT) + vma->vm_pgoff;
039576c0
AK		 2462 if (pgoff == 0)
2463 page = virt_to_page(vcpu->run);
2464 else if (pgoff == KVM_PIO_PAGE_OFFSET)
2465 page = virt_to_page(vcpu->pio_data);
2466 else
9a2bb7f4 2467 return NOPAGE_SIGBUS;
9a2bb7f4 2468 get_page(page);
cd0d9137
NAQ		 2469 if (type != NULL)
2470 *type = VM_FAULT_MINOR;
2471
9a2bb7f4
AK		 2472 return page;
2473}
2474
2475static struct vm_operations_struct kvm_vcpu_vm_ops = {
2476 .nopage = kvm_vcpu_nopage,
2477};
2478
2479static int kvm_vcpu_mmap(struct file *file, struct vm_area_struct *vma)
2480{
2481 vma->vm_ops = &kvm_vcpu_vm_ops;
2482 return 0;
2483}
2484
bccf2150
AK		 2485static int kvm_vcpu_release(struct inode *inode, struct file *filp)
2486{
2487 struct kvm_vcpu *vcpu = filp->private_data;
2488
2489 fput(vcpu->kvm->filp);
2490 return 0;
2491}
2492
2493static struct file_operations kvm_vcpu_fops = {
2494 .release = kvm_vcpu_release,
2495 .unlocked_ioctl = kvm_vcpu_ioctl,
2496 .compat_ioctl = kvm_vcpu_ioctl,
9a2bb7f4 2497 .mmap = kvm_vcpu_mmap,
bccf2150
AK		 2498};
2499
2500/*
2501 * Allocates an inode for the vcpu.
2502 */
2503static int create_vcpu_fd(struct kvm_vcpu *vcpu)
2504{
2505 int fd, r;
2506 struct inode *inode;
2507 struct file *file;
2508
d6d28168
AK		 2509 r = anon_inode_getfd(&fd, &inode, &file,
2510 "kvm-vcpu", &kvm_vcpu_fops, vcpu);
2511 if (r)
2512 return r;
bccf2150 2513 atomic_inc(&vcpu->kvm->filp->f_count);
bccf2150 2514 return fd;
bccf2150
AK		 2515}
2516
c5ea7660
AK		 2517/*
2518 * Creates some virtual cpus. Good luck creating more than one.
2519 */
2520static int kvm_vm_ioctl_create_vcpu(struct kvm *kvm, int n)
2521{
2522 int r;
2523 struct kvm_vcpu *vcpu;
2524
c5ea7660 2525 if (!valid_vcpu(n))
fb3f0f51 2526 return -EINVAL;
c5ea7660 2527
fb3f0f51
RR		 2528 vcpu = kvm_arch_ops->vcpu_create(kvm, n);
2529 if (IS_ERR(vcpu))
2530 return PTR_ERR(vcpu);
c5ea7660 2531
15ad7146
AK		 2532 preempt_notifier_init(&vcpu->preempt_notifier, &kvm_preempt_ops);
2533
b114b080
RR		 2534 /* We do fxsave: this must be aligned. */
2535 BUG_ON((unsigned long)&vcpu->host_fx_image & 0xF);
2536
fb3f0f51 2537 vcpu_load(vcpu);
c5ea7660 2538 r = kvm_mmu_setup(vcpu);
c5ea7660 2539 vcpu_put(vcpu);
c5ea7660 2540 if (r < 0)
fb3f0f51
RR		 2541 goto free_vcpu;
2542
11ec2804 2543 mutex_lock(&kvm->lock);
fb3f0f51
RR		 2544 if (kvm->vcpus[n]) {
2545 r = -EEXIST;
11ec2804 2546 mutex_unlock(&kvm->lock);
fb3f0f51
RR		 2547 goto mmu_unload;
2548 }
2549 kvm->vcpus[n] = vcpu;
11ec2804 2550 mutex_unlock(&kvm->lock);
c5ea7660 2551
fb3f0f51 2552 /* Now it's all set up, let userspace reach it */
bccf2150
AK		 2553 r = create_vcpu_fd(vcpu);
2554 if (r < 0)
fb3f0f51
RR		 2555 goto unlink;
2556 return r;
39c3b86e 2557
fb3f0f51 2558unlink:
11ec2804 2559 mutex_lock(&kvm->lock);
fb3f0f51 2560 kvm->vcpus[n] = NULL;
11ec2804 2561 mutex_unlock(&kvm->lock);
a2fa3e9f 2562
fb3f0f51
RR		 2563mmu_unload:
2564 vcpu_load(vcpu);
2565 kvm_mmu_unload(vcpu);
2566 vcpu_put(vcpu);
c5ea7660 2567
fb3f0f51
RR		 2568free_vcpu:
2569 kvm_arch_ops->vcpu_free(vcpu);
c5ea7660
AK		 2570 return r;
2571}
2572
2cc51560
ED		 2573static void cpuid_fix_nx_cap(struct kvm_vcpu *vcpu)
2574{
2575 u64 efer;
2576 int i;
2577 struct kvm_cpuid_entry *e, *entry;
2578
2579 rdmsrl(MSR_EFER, efer);
2580 entry = NULL;
2581 for (i = 0; i < vcpu->cpuid_nent; ++i) {
2582 e = &vcpu->cpuid_entries[i];
2583 if (e->function == 0x80000001) {
2584 entry = e;
2585 break;
2586 }
2587 }
4c981b43 2588 if (entry && (entry->edx & (1 << 20)) && !(efer & EFER_NX)) {
2cc51560 2589 entry->edx &= ~(1 << 20);
4c981b43 2590 printk(KERN_INFO "kvm: guest NX capability removed\n");
2cc51560
ED		 2591 }
2592}
2593
06465c5a
AK		 2594static int kvm_vcpu_ioctl_set_cpuid(struct kvm_vcpu *vcpu,
2595 struct kvm_cpuid *cpuid,
2596 struct kvm_cpuid_entry __user *entries)
2597{
2598 int r;
2599
2600 r = -E2BIG;
2601 if (cpuid->nent > KVM_MAX_CPUID_ENTRIES)
2602 goto out;
2603 r = -EFAULT;
2604 if (copy_from_user(&vcpu->cpuid_entries, entries,
2605 cpuid->nent * sizeof(struct kvm_cpuid_entry)))
2606 goto out;
2607 vcpu->cpuid_nent = cpuid->nent;
2cc51560 2608 cpuid_fix_nx_cap(vcpu);
06465c5a
AK		 2609 return 0;
2610
2611out:
2612 return r;
2613}
2614
1961d276
AK		 2615static int kvm_vcpu_ioctl_set_sigmask(struct kvm_vcpu *vcpu, sigset_t *sigset)
2616{
2617 if (sigset) {
2618 sigdelsetmask(sigset, sigmask(SIGKILL)|sigmask(SIGSTOP));
2619 vcpu->sigset_active = 1;
2620 vcpu->sigset = *sigset;
2621 } else
2622 vcpu->sigset_active = 0;
2623 return 0;
2624}
2625
b8836737
AK		 2626/*
2627 * fxsave fpu state. Taken from x86_64/processor.h. To be killed when
2628 * we have asm/x86/processor.h
2629 */
2630struct fxsave {
2631 u16 cwd;
2632 u16 swd;
2633 u16 twd;
2634 u16 fop;
2635 u64 rip;
2636 u64 rdp;
2637 u32 mxcsr;
2638 u32 mxcsr_mask;
2639 u32 st_space[32]; /* 8*16 bytes for each FP-reg = 128 bytes */
2640#ifdef CONFIG_X86_64
2641 u32 xmm_space[64]; /* 16*16 bytes for each XMM-reg = 256 bytes */
2642#else
2643 u32 xmm_space[32]; /* 8*16 bytes for each XMM-reg = 128 bytes */
2644#endif
2645};
2646
2647static int kvm_vcpu_ioctl_get_fpu(struct kvm_vcpu *vcpu, struct kvm_fpu *fpu)
2648{
b114b080 2649 struct fxsave *fxsave = (struct fxsave *)&vcpu->guest_fx_image;
b8836737
AK		 2650
2651 vcpu_load(vcpu);
2652
2653 memcpy(fpu->fpr, fxsave->st_space, 128);
2654 fpu->fcw = fxsave->cwd;
2655 fpu->fsw = fxsave->swd;
2656 fpu->ftwx = fxsave->twd;
2657 fpu->last_opcode = fxsave->fop;
2658 fpu->last_ip = fxsave->rip;
2659 fpu->last_dp = fxsave->rdp;
2660 memcpy(fpu->xmm, fxsave->xmm_space, sizeof fxsave->xmm_space);
2661
2662 vcpu_put(vcpu);
2663
2664 return 0;
2665}
2666
2667static int kvm_vcpu_ioctl_set_fpu(struct kvm_vcpu *vcpu, struct kvm_fpu *fpu)
2668{
b114b080 2669 struct fxsave *fxsave = (struct fxsave *)&vcpu->guest_fx_image;
b8836737
AK		 2670
2671 vcpu_load(vcpu);
2672
2673 memcpy(fxsave->st_space, fpu->fpr, 128);
2674 fxsave->cwd = fpu->fcw;
2675 fxsave->swd = fpu->fsw;
2676 fxsave->twd = fpu->ftwx;
2677 fxsave->fop = fpu->last_opcode;
2678 fxsave->rip = fpu->last_ip;
2679 fxsave->rdp = fpu->last_dp;
2680 memcpy(fxsave->xmm_space, fpu->xmm, sizeof fxsave->xmm_space);
2681
2682 vcpu_put(vcpu);
2683
2684 return 0;
2685}
2686
96ad2cc6
ED		 2687static int kvm_vcpu_ioctl_get_lapic(struct kvm_vcpu *vcpu,
2688 struct kvm_lapic_state *s)
2689{
2690 vcpu_load(vcpu);
2691 memcpy(s->regs, vcpu->apic->regs, sizeof *s);
2692 vcpu_put(vcpu);
2693
2694 return 0;
2695}
2696
2697static int kvm_vcpu_ioctl_set_lapic(struct kvm_vcpu *vcpu,
2698 struct kvm_lapic_state *s)
2699{
2700 vcpu_load(vcpu);
2701 memcpy(vcpu->apic->regs, s->regs, sizeof *s);
2702 kvm_apic_post_state_restore(vcpu);
2703 vcpu_put(vcpu);
2704
2705 return 0;
2706}
2707
bccf2150
AK		 2708static long kvm_vcpu_ioctl(struct file *filp,
2709 unsigned int ioctl, unsigned long arg)
6aa8b732 2710{
bccf2150 2711 struct kvm_vcpu *vcpu = filp->private_data;
2f366987 2712 void __user *argp = (void __user *)arg;
6aa8b732
AK		 2713 int r = -EINVAL;
2714
2715 switch (ioctl) {
9a2bb7f4 2716 case KVM_RUN:
f0fe5108
AK		 2717 r = -EINVAL;
2718 if (arg)
2719 goto out;
9a2bb7f4 2720 r = kvm_vcpu_ioctl_run(vcpu, vcpu->run);
6aa8b732 2721 break;
6aa8b732
AK		 2722 case KVM_GET_REGS: {
2723 struct kvm_regs kvm_regs;
2724
bccf2150
AK		 2725 memset(&kvm_regs, 0, sizeof kvm_regs);
2726 r = kvm_vcpu_ioctl_get_regs(vcpu, &kvm_regs);
6aa8b732
AK		 2727 if (r)
2728 goto out;
2729 r = -EFAULT;
2f366987 2730 if (copy_to_user(argp, &kvm_regs, sizeof kvm_regs))
6aa8b732
AK		 2731 goto out;
2732 r = 0;
2733 break;
2734 }
2735 case KVM_SET_REGS: {
2736 struct kvm_regs kvm_regs;
2737
2738 r = -EFAULT;
2f366987 2739 if (copy_from_user(&kvm_regs, argp, sizeof kvm_regs))
6aa8b732 2740 goto out;
bccf2150 2741 r = kvm_vcpu_ioctl_set_regs(vcpu, &kvm_regs);
6aa8b732
AK		 2742 if (r)
2743 goto out;
2744 r = 0;
2745 break;
2746 }
2747 case KVM_GET_SREGS: {
2748 struct kvm_sregs kvm_sregs;
2749
bccf2150
AK		 2750 memset(&kvm_sregs, 0, sizeof kvm_sregs);
2751 r = kvm_vcpu_ioctl_get_sregs(vcpu, &kvm_sregs);
6aa8b732
AK		 2752 if (r)
2753 goto out;
2754 r = -EFAULT;
2f366987 2755 if (copy_to_user(argp, &kvm_sregs, sizeof kvm_sregs))
6aa8b732
AK		 2756 goto out;
2757 r = 0;
2758 break;
2759 }
2760 case KVM_SET_SREGS: {
2761 struct kvm_sregs kvm_sregs;
2762
2763 r = -EFAULT;
2f366987 2764 if (copy_from_user(&kvm_sregs, argp, sizeof kvm_sregs))
6aa8b732 2765 goto out;
bccf2150 2766 r = kvm_vcpu_ioctl_set_sregs(vcpu, &kvm_sregs);
6aa8b732
AK		 2767 if (r)
2768 goto out;
2769 r = 0;
2770 break;
2771 }
2772 case KVM_TRANSLATE: {
2773 struct kvm_translation tr;
2774
2775 r = -EFAULT;
2f366987 2776 if (copy_from_user(&tr, argp, sizeof tr))
6aa8b732 2777 goto out;
bccf2150 2778 r = kvm_vcpu_ioctl_translate(vcpu, &tr);
6aa8b732
AK		 2779 if (r)
2780 goto out;
2781 r = -EFAULT;
2f366987 2782 if (copy_to_user(argp, &tr, sizeof tr))
6aa8b732
AK		 2783 goto out;
2784 r = 0;
2785 break;
2786 }
2787 case KVM_INTERRUPT: {
2788 struct kvm_interrupt irq;
2789
2790 r = -EFAULT;
2f366987 2791 if (copy_from_user(&irq, argp, sizeof irq))
6aa8b732 2792 goto out;
bccf2150 2793 r = kvm_vcpu_ioctl_interrupt(vcpu, &irq);
6aa8b732
AK		 2794 if (r)
2795 goto out;
2796 r = 0;
2797 break;
2798 }
2799 case KVM_DEBUG_GUEST: {
2800 struct kvm_debug_guest dbg;
2801
2802 r = -EFAULT;
2f366987 2803 if (copy_from_user(&dbg, argp, sizeof dbg))
6aa8b732 2804 goto out;
bccf2150 2805 r = kvm_vcpu_ioctl_debug_guest(vcpu, &dbg);
6aa8b732
AK		 2806 if (r)
2807 goto out;
2808 r = 0;
2809 break;
2810 }
bccf2150 2811 case KVM_GET_MSRS:
35f3f286 2812 r = msr_io(vcpu, argp, kvm_get_msr, 1);
bccf2150
AK		 2813 break;
2814 case KVM_SET_MSRS:
2815 r = msr_io(vcpu, argp, do_set_msr, 0);
2816 break;
06465c5a
AK		 2817 case KVM_SET_CPUID: {
2818 struct kvm_cpuid __user *cpuid_arg = argp;
2819 struct kvm_cpuid cpuid;
2820
2821 r = -EFAULT;
2822 if (copy_from_user(&cpuid, cpuid_arg, sizeof cpuid))
2823 goto out;
2824 r = kvm_vcpu_ioctl_set_cpuid(vcpu, &cpuid, cpuid_arg->entries);
2825 if (r)
2826 goto out;
2827 break;
2828 }
1961d276
AK		 2829 case KVM_SET_SIGNAL_MASK: {
2830 struct kvm_signal_mask __user *sigmask_arg = argp;
2831 struct kvm_signal_mask kvm_sigmask;
2832 sigset_t sigset, *p;
2833
2834 p = NULL;
2835 if (argp) {
2836 r = -EFAULT;
2837 if (copy_from_user(&kvm_sigmask, argp,
2838 sizeof kvm_sigmask))
2839 goto out;
2840 r = -EINVAL;
2841 if (kvm_sigmask.len != sizeof sigset)
2842 goto out;
2843 r = -EFAULT;
2844 if (copy_from_user(&sigset, sigmask_arg->sigset,
2845 sizeof sigset))
2846 goto out;
2847 p = &sigset;
2848 }
2849 r = kvm_vcpu_ioctl_set_sigmask(vcpu, &sigset);
2850 break;
2851 }
b8836737
AK		 2852 case KVM_GET_FPU: {
2853 struct kvm_fpu fpu;
2854
2855 memset(&fpu, 0, sizeof fpu);
2856 r = kvm_vcpu_ioctl_get_fpu(vcpu, &fpu);
2857 if (r)
2858 goto out;
2859 r = -EFAULT;
2860 if (copy_to_user(argp, &fpu, sizeof fpu))
2861 goto out;
2862 r = 0;
2863 break;
2864 }
2865 case KVM_SET_FPU: {
2866 struct kvm_fpu fpu;
2867
2868 r = -EFAULT;
2869 if (copy_from_user(&fpu, argp, sizeof fpu))
2870 goto out;
2871 r = kvm_vcpu_ioctl_set_fpu(vcpu, &fpu);
2872 if (r)
2873 goto out;
2874 r = 0;
2875 break;
2876 }
96ad2cc6
ED		 2877 case KVM_GET_LAPIC: {
2878 struct kvm_lapic_state lapic;
2879
2880 memset(&lapic, 0, sizeof lapic);
2881 r = kvm_vcpu_ioctl_get_lapic(vcpu, &lapic);
2882 if (r)
2883 goto out;
2884 r = -EFAULT;
2885 if (copy_to_user(argp, &lapic, sizeof lapic))
2886 goto out;
2887 r = 0;
2888 break;
2889 }
2890 case KVM_SET_LAPIC: {
2891 struct kvm_lapic_state lapic;
2892
2893 r = -EFAULT;
2894 if (copy_from_user(&lapic, argp, sizeof lapic))
2895 goto out;
2896 r = kvm_vcpu_ioctl_set_lapic(vcpu, &lapic);;
2897 if (r)
2898 goto out;
2899 r = 0;
2900 break;
2901 }
bccf2150
AK		 2902 default:
2903 ;
2904 }
2905out:
2906 return r;
2907}
2908
2909static long kvm_vm_ioctl(struct file *filp,
2910 unsigned int ioctl, unsigned long arg)
2911{
2912 struct kvm *kvm = filp->private_data;
2913 void __user *argp = (void __user *)arg;
2914 int r = -EINVAL;
2915
2916 switch (ioctl) {
2917 case KVM_CREATE_VCPU:
2918 r = kvm_vm_ioctl_create_vcpu(kvm, arg);
2919 if (r < 0)
2920 goto out;
2921 break;
6aa8b732
AK		 2922 case KVM_SET_MEMORY_REGION: {
2923 struct kvm_memory_region kvm_mem;
2924
2925 r = -EFAULT;
2f366987 2926 if (copy_from_user(&kvm_mem, argp, sizeof kvm_mem))
6aa8b732 2927 goto out;
2c6f5df9 2928 r = kvm_vm_ioctl_set_memory_region(kvm, &kvm_mem);
6aa8b732
AK		 2929 if (r)
2930 goto out;
2931 break;
2932 }
2933 case KVM_GET_DIRTY_LOG: {
2934 struct kvm_dirty_log log;
2935
2936 r = -EFAULT;
2f366987 2937 if (copy_from_user(&log, argp, sizeof log))
6aa8b732 2938 goto out;
2c6f5df9 2939 r = kvm_vm_ioctl_get_dirty_log(kvm, &log);
6aa8b732
AK		 2940 if (r)
2941 goto out;
2942 break;
2943 }
e8207547
AK		 2944 case KVM_SET_MEMORY_ALIAS: {
2945 struct kvm_memory_alias alias;
2946
2947 r = -EFAULT;
2948 if (copy_from_user(&alias, argp, sizeof alias))
2949 goto out;
2950 r = kvm_vm_ioctl_set_memory_alias(kvm, &alias);
2951 if (r)
2952 goto out;
2953 break;
2954 }
85f455f7
ED		 2955 case KVM_CREATE_IRQCHIP:
2956 r = -ENOMEM;
2957 kvm->vpic = kvm_create_pic(kvm);
1fd4f2a5
ED		 2958 if (kvm->vpic) {
2959 r = kvm_ioapic_init(kvm);
2960 if (r) {
2961 kfree(kvm->vpic);
2962 kvm->vpic = NULL;
2963 goto out;
2964 }
2965 }
85f455f7
ED		 2966 else
2967 goto out;
2968 break;
2969 case KVM_IRQ_LINE: {
2970 struct kvm_irq_level irq_event;
2971
2972 r = -EFAULT;
2973 if (copy_from_user(&irq_event, argp, sizeof irq_event))
2974 goto out;
2975 if (irqchip_in_kernel(kvm)) {
9cf98828 2976 mutex_lock(&kvm->lock);
85f455f7
ED		 2977 if (irq_event.irq < 16)
2978 kvm_pic_set_irq(pic_irqchip(kvm),
2979 irq_event.irq,
2980 irq_event.level);
1fd4f2a5
ED		 2981 kvm_ioapic_set_irq(kvm->vioapic,
2982 irq_event.irq,
2983 irq_event.level);
9cf98828 2984 mutex_unlock(&kvm->lock);
85f455f7
ED		 2985 r = 0;
2986 }
2987 break;
2988 }
6ceb9d79
HQ		 2989 case KVM_GET_IRQCHIP: {
2990 /* 0: PIC master, 1: PIC slave, 2: IOAPIC */
2991 struct kvm_irqchip chip;
2992
2993 r = -EFAULT;
2994 if (copy_from_user(&chip, argp, sizeof chip))
2995 goto out;
2996 r = -ENXIO;
2997 if (!irqchip_in_kernel(kvm))
2998 goto out;
2999 r = kvm_vm_ioctl_get_irqchip(kvm, &chip);
3000 if (r)
3001 goto out;
3002 r = -EFAULT;
3003 if (copy_to_user(argp, &chip, sizeof chip))
3004 goto out;
3005 r = 0;
3006 break;
3007 }
3008 case KVM_SET_IRQCHIP: {
3009 /* 0: PIC master, 1: PIC slave, 2: IOAPIC */
3010 struct kvm_irqchip chip;
3011
3012 r = -EFAULT;
3013 if (copy_from_user(&chip, argp, sizeof chip))
3014 goto out;
3015 r = -ENXIO;
3016 if (!irqchip_in_kernel(kvm))
3017 goto out;
3018 r = kvm_vm_ioctl_set_irqchip(kvm, &chip);
3019 if (r)
3020 goto out;
3021 r = 0;
3022 break;
3023 }
f17abe9a
AK		 3024 default:
3025 ;
3026 }
3027out:
3028 return r;
3029}
3030
3031static struct page *kvm_vm_nopage(struct vm_area_struct *vma,
3032 unsigned long address,
3033 int *type)
3034{
3035 struct kvm *kvm = vma->vm_file->private_data;
3036 unsigned long pgoff;
f17abe9a
AK		 3037 struct page *page;
3038
f17abe9a 3039 pgoff = ((address - vma->vm_start) >> PAGE_SHIFT) + vma->vm_pgoff;
954bbbc2 3040 page = gfn_to_page(kvm, pgoff);
f17abe9a
AK		 3041 if (!page)
3042 return NOPAGE_SIGBUS;
3043 get_page(page);
cd0d9137
NAQ		 3044 if (type != NULL)
3045 *type = VM_FAULT_MINOR;
3046
f17abe9a
AK		 3047 return page;
3048}
3049
3050static struct vm_operations_struct kvm_vm_vm_ops = {
3051 .nopage = kvm_vm_nopage,
3052};
3053
3054static int kvm_vm_mmap(struct file *file, struct vm_area_struct *vma)
3055{
3056 vma->vm_ops = &kvm_vm_vm_ops;
3057 return 0;
3058}
3059
3060static struct file_operations kvm_vm_fops = {
3061 .release = kvm_vm_release,
3062 .unlocked_ioctl = kvm_vm_ioctl,
3063 .compat_ioctl = kvm_vm_ioctl,
3064 .mmap = kvm_vm_mmap,
3065};
3066
3067static int kvm_dev_ioctl_create_vm(void)
3068{
3069 int fd, r;
3070 struct inode *inode;
3071 struct file *file;
3072 struct kvm *kvm;
3073
f17abe9a 3074 kvm = kvm_create_vm();
d6d28168
AK		 3075 if (IS_ERR(kvm))
3076 return PTR_ERR(kvm);
3077 r = anon_inode_getfd(&fd, &inode, &file, "kvm-vm", &kvm_vm_fops, kvm);
3078 if (r) {
3079 kvm_destroy_vm(kvm);
3080 return r;
f17abe9a
AK		 3081 }
3082
bccf2150 3083 kvm->filp = file;
f17abe9a 3084
f17abe9a 3085 return fd;
f17abe9a
AK		 3086}
3087
3088static long kvm_dev_ioctl(struct file *filp,
3089 unsigned int ioctl, unsigned long arg)
3090{
3091 void __user *argp = (void __user *)arg;
07c45a36 3092 long r = -EINVAL;
f17abe9a
AK		 3093
3094 switch (ioctl) {
3095 case KVM_GET_API_VERSION:
f0fe5108
AK		 3096 r = -EINVAL;
3097 if (arg)
3098 goto out;
f17abe9a
AK		 3099 r = KVM_API_VERSION;
3100 break;
3101 case KVM_CREATE_VM:
f0fe5108
AK		 3102 r = -EINVAL;
3103 if (arg)
3104 goto out;
f17abe9a
AK		 3105 r = kvm_dev_ioctl_create_vm();
3106 break;
6aa8b732 3107 case KVM_GET_MSR_INDEX_LIST: {
2f366987 3108 struct kvm_msr_list __user *user_msr_list = argp;
6aa8b732
AK		 3109 struct kvm_msr_list msr_list;
3110 unsigned n;
3111
3112 r = -EFAULT;
3113 if (copy_from_user(&msr_list, user_msr_list, sizeof msr_list))
3114 goto out;
3115 n = msr_list.nmsrs;
6f00e68f 3116 msr_list.nmsrs = num_msrs_to_save + ARRAY_SIZE(emulated_msrs);
6aa8b732
AK		 3117 if (copy_to_user(user_msr_list, &msr_list, sizeof msr_list))
3118 goto out;
3119 r = -E2BIG;
bf591b24 3120 if (n < num_msrs_to_save)
6aa8b732
AK		 3121 goto out;
3122 r = -EFAULT;
3123 if (copy_to_user(user_msr_list->indices, &msrs_to_save,
bf591b24 3124 num_msrs_to_save * sizeof(u32)))
6aa8b732 3125 goto out;
6f00e68f
AK		 3126 if (copy_to_user(user_msr_list->indices
3127 + num_msrs_to_save * sizeof(u32),
3128 &emulated_msrs,
3129 ARRAY_SIZE(emulated_msrs) * sizeof(u32)))
3130 goto out;
6aa8b732 3131 r = 0;
cc1d8955 3132 break;
6aa8b732 3133 }
85f455f7
ED		 3134 case KVM_CHECK_EXTENSION: {
3135 int ext = (long)argp;
3136
3137 switch (ext) {
3138 case KVM_CAP_IRQCHIP:
b6958ce4 3139 case KVM_CAP_HLT:
85f455f7
ED		 3140 r = 1;
3141 break;
3142 default:
3143 r = 0;
3144 break;
3145 }
5d308f45 3146 break;
85f455f7 3147 }
07c45a36
AK		 3148 case KVM_GET_VCPU_MMAP_SIZE:
3149 r = -EINVAL;
3150 if (arg)
3151 goto out;
039576c0 3152 r = 2 * PAGE_SIZE;
07c45a36 3153 break;
6aa8b732
AK		 3154 default:
3155 ;
3156 }
3157out:
3158 return r;
3159}
3160
6aa8b732 3161static struct file_operations kvm_chardev_ops = {
6aa8b732
AK		 3162 .unlocked_ioctl = kvm_dev_ioctl,
3163 .compat_ioctl = kvm_dev_ioctl,
6aa8b732
AK		 3164};
3165
3166static struct miscdevice kvm_dev = {
bbe4432e 3167 KVM_MINOR,
6aa8b732
AK		 3168 "kvm",
3169 &kvm_chardev_ops,
3170};
3171
774c47f1
AK		 3172/*
3173 * Make sure that a cpu that is being hot-unplugged does not have any vcpus
3174 * cached on it.
3175 */
3176static void decache_vcpus_on_cpu(int cpu)
3177{
3178 struct kvm *vm;
3179 struct kvm_vcpu *vcpu;
3180 int i;
3181
3182 spin_lock(&kvm_lock);
11ec2804 3183 list_for_each_entry(vm, &vm_list, vm_list)
774c47f1 3184 for (i = 0; i < KVM_MAX_VCPUS; ++i) {
fb3f0f51
RR		 3185 vcpu = vm->vcpus[i];
3186 if (!vcpu)
3187 continue;
774c47f1
AK		 3188 /*
3189 * If the vcpu is locked, then it is running on some
3190 * other cpu and therefore it is not cached on the
3191 * cpu in question.
3192 *
3193 * If it's not locked, check the last cpu it executed
3194 * on.
3195 */
3196 if (mutex_trylock(&vcpu->mutex)) {
3197 if (vcpu->cpu == cpu) {
3198 kvm_arch_ops->vcpu_decache(vcpu);
3199 vcpu->cpu = -1;
3200 }
3201 mutex_unlock(&vcpu->mutex);
3202 }
3203 }
3204 spin_unlock(&kvm_lock);
3205}
3206
1b6c0168
AK		 3207static void hardware_enable(void *junk)
3208{
3209 int cpu = raw_smp_processor_id();
3210
3211 if (cpu_isset(cpu, cpus_hardware_enabled))
3212 return;
3213 cpu_set(cpu, cpus_hardware_enabled);
3214 kvm_arch_ops->hardware_enable(NULL);
3215}
3216
3217static void hardware_disable(void *junk)
3218{
3219 int cpu = raw_smp_processor_id();
3220
3221 if (!cpu_isset(cpu, cpus_hardware_enabled))
3222 return;
3223 cpu_clear(cpu, cpus_hardware_enabled);
3224 decache_vcpus_on_cpu(cpu);
3225 kvm_arch_ops->hardware_disable(NULL);
3226}
3227
774c47f1
AK		 3228static int kvm_cpu_hotplug(struct notifier_block *notifier, unsigned long val,
3229 void *v)
3230{
3231 int cpu = (long)v;
3232
3233 switch (val) {
cec9ad27
AK		 3234 case CPU_DYING:
3235 case CPU_DYING_FROZEN:
6ec8a856
AK		 3236 printk(KERN_INFO "kvm: disabling virtualization on CPU%d\n",
3237 cpu);
3238 hardware_disable(NULL);
3239 break;
774c47f1 3240 case CPU_UP_CANCELED:
8bb78442 3241 case CPU_UP_CANCELED_FROZEN:
43934a38
JK		 3242 printk(KERN_INFO "kvm: disabling virtualization on CPU%d\n",
3243 cpu);
1b6c0168 3244 smp_call_function_single(cpu, hardware_disable, NULL, 0, 1);
774c47f1 3245 break;
43934a38 3246 case CPU_ONLINE:
8bb78442 3247 case CPU_ONLINE_FROZEN:
43934a38
JK		 3248 printk(KERN_INFO "kvm: enabling virtualization on CPU%d\n",
3249 cpu);
1b6c0168 3250 smp_call_function_single(cpu, hardware_enable, NULL, 0, 1);
774c47f1
AK		 3251 break;
3252 }
3253 return NOTIFY_OK;
3254}
3255
9a2b85c6
RR		 3256static int kvm_reboot(struct notifier_block *notifier, unsigned long val,
3257 void *v)
3258{
3259 if (val == SYS_RESTART) {
3260 /*
3261 * Some (well, at least mine) BIOSes hang on reboot if
3262 * in vmx root mode.
3263 */
3264 printk(KERN_INFO "kvm: exiting hardware virtualization\n");
3265 on_each_cpu(hardware_disable, NULL, 0, 1);
3266 }
3267 return NOTIFY_OK;
3268}
3269
3270static struct notifier_block kvm_reboot_notifier = {
3271 .notifier_call = kvm_reboot,
3272 .priority = 0,
3273};
3274
2eeb2e94
GH		 3275void kvm_io_bus_init(struct kvm_io_bus *bus)
3276{
3277 memset(bus, 0, sizeof(*bus));
3278}
3279
3280void kvm_io_bus_destroy(struct kvm_io_bus *bus)
3281{
3282 int i;
3283
3284 for (i = 0; i < bus->dev_count; i++) {
3285 struct kvm_io_device *pos = bus->devs[i];
3286
3287 kvm_iodevice_destructor(pos);
3288 }
3289}
3290
3291struct kvm_io_device *kvm_io_bus_find_dev(struct kvm_io_bus *bus, gpa_t addr)
3292{
3293 int i;
3294
3295 for (i = 0; i < bus->dev_count; i++) {
3296 struct kvm_io_device *pos = bus->devs[i];
3297
3298 if (pos->in_range(pos, addr))
3299 return pos;
3300 }
3301
3302 return NULL;
3303}
3304
3305void kvm_io_bus_register_dev(struct kvm_io_bus *bus, struct kvm_io_device *dev)
3306{
3307 BUG_ON(bus->dev_count > (NR_IOBUS_DEVS-1));
3308
3309 bus->devs[bus->dev_count++] = dev;
3310}
3311
774c47f1
AK		 3312static struct notifier_block kvm_cpu_notifier = {
3313 .notifier_call = kvm_cpu_hotplug,
3314 .priority = 20, /* must be > scheduler priority */
3315};
3316
1165f5fe
AK		 3317static u64 stat_get(void *_offset)
3318{
3319 unsigned offset = (long)_offset;
3320 u64 total = 0;
3321 struct kvm *kvm;
3322 struct kvm_vcpu *vcpu;
3323 int i;
3324
3325 spin_lock(&kvm_lock);
3326 list_for_each_entry(kvm, &vm_list, vm_list)
3327 for (i = 0; i < KVM_MAX_VCPUS; ++i) {
fb3f0f51
RR		 3328 vcpu = kvm->vcpus[i];
3329 if (vcpu)
3330 total += *(u32 *)((void *)vcpu + offset);
1165f5fe
AK		 3331 }
3332 spin_unlock(&kvm_lock);
3333 return total;
3334}
3335
3dea7ca7 3336DEFINE_SIMPLE_ATTRIBUTE(stat_fops, stat_get, NULL, "%llu\n");
1165f5fe 3337
6aa8b732
AK		 3338static __init void kvm_init_debug(void)
3339{
3340 struct kvm_stats_debugfs_item *p;
3341
8b6d44c7 3342 debugfs_dir = debugfs_create_dir("kvm", NULL);
6aa8b732 3343 for (p = debugfs_entries; p->name; ++p)
1165f5fe
AK		 3344 p->dentry = debugfs_create_file(p->name, 0444, debugfs_dir,
3345 (void *)(long)p->offset,
3346 &stat_fops);
6aa8b732
AK		 3347}
3348
3349static void kvm_exit_debug(void)
3350{
3351 struct kvm_stats_debugfs_item *p;
3352
3353 for (p = debugfs_entries; p->name; ++p)
3354 debugfs_remove(p->dentry);
3355 debugfs_remove(debugfs_dir);
3356}
3357
59ae6c6b
AK		 3358static int kvm_suspend(struct sys_device *dev, pm_message_t state)
3359{
4267c41a 3360 hardware_disable(NULL);
59ae6c6b
AK		 3361 return 0;
3362}
3363
3364static int kvm_resume(struct sys_device *dev)
3365{
4267c41a 3366 hardware_enable(NULL);
59ae6c6b
AK		 3367 return 0;
3368}
3369
3370static struct sysdev_class kvm_sysdev_class = {
3371 set_kset_name("kvm"),
3372 .suspend = kvm_suspend,
3373 .resume = kvm_resume,
3374};
3375
3376static struct sys_device kvm_sysdev = {
3377 .id = 0,
3378 .cls = &kvm_sysdev_class,
3379};
3380
6aa8b732
AK		 3381hpa_t bad_page_address;
3382
15ad7146
AK		 3383static inline
3384struct kvm_vcpu *preempt_notifier_to_vcpu(struct preempt_notifier *pn)
3385{
3386 return container_of(pn, struct kvm_vcpu, preempt_notifier);
3387}
3388
3389static void kvm_sched_in(struct preempt_notifier *pn, int cpu)
3390{
3391 struct kvm_vcpu *vcpu = preempt_notifier_to_vcpu(pn);
3392
3393 kvm_arch_ops->vcpu_load(vcpu, cpu);
3394}
3395
3396static void kvm_sched_out(struct preempt_notifier *pn,
3397 struct task_struct *next)
3398{
3399 struct kvm_vcpu *vcpu = preempt_notifier_to_vcpu(pn);
3400
3401 kvm_arch_ops->vcpu_put(vcpu);
3402}
3403
c16f862d
RR		 3404int kvm_init_arch(struct kvm_arch_ops *ops, unsigned int vcpu_size,
3405 struct module *module)
6aa8b732
AK		 3406{
3407 int r;
002c7f7c 3408 int cpu;
6aa8b732 3409
09db28b8
YI		 3410 if (kvm_arch_ops) {
3411 printk(KERN_ERR "kvm: already loaded the other module\n");
3412 return -EEXIST;
3413 }
3414
e097f35c 3415 if (!ops->cpu_has_kvm_support()) {
6aa8b732
AK		 3416 printk(KERN_ERR "kvm: no hardware support\n");
3417 return -EOPNOTSUPP;
3418 }
e097f35c 3419 if (ops->disabled_by_bios()) {
6aa8b732
AK		 3420 printk(KERN_ERR "kvm: disabled by bios\n");
3421 return -EOPNOTSUPP;
3422 }
3423
e097f35c
YI		 3424 kvm_arch_ops = ops;
3425
6aa8b732
AK		 3426 r = kvm_arch_ops->hardware_setup();
3427 if (r < 0)
ca45aaae 3428 goto out;
6aa8b732 3429
002c7f7c
YS		 3430 for_each_online_cpu(cpu) {
3431 smp_call_function_single(cpu,
3432 kvm_arch_ops->check_processor_compatibility,
3433 &r, 0, 1);
3434 if (r < 0)
3435 goto out_free_0;
3436 }
3437
1b6c0168 3438 on_each_cpu(hardware_enable, NULL, 0, 1);
774c47f1
AK		 3439 r = register_cpu_notifier(&kvm_cpu_notifier);
3440 if (r)
3441 goto out_free_1;
6aa8b732
AK		 3442 register_reboot_notifier(&kvm_reboot_notifier);
3443
59ae6c6b
AK		 3444 r = sysdev_class_register(&kvm_sysdev_class);
3445 if (r)
3446 goto out_free_2;
3447
3448 r = sysdev_register(&kvm_sysdev);
3449 if (r)
3450 goto out_free_3;
3451
c16f862d
RR		 3452 /* A kmem cache lets us meet the alignment requirements of fx_save. */
3453 kvm_vcpu_cache = kmem_cache_create("kvm_vcpu", vcpu_size,
3454 __alignof__(struct kvm_vcpu), 0, 0);
3455 if (!kvm_vcpu_cache) {
3456 r = -ENOMEM;
3457 goto out_free_4;
3458 }
3459
6aa8b732
AK		 3460 kvm_chardev_ops.owner = module;
3461
3462 r = misc_register(&kvm_dev);
3463 if (r) {
3464 printk (KERN_ERR "kvm: misc device register failed\n");
3465 goto out_free;
3466 }
3467
15ad7146
AK		 3468 kvm_preempt_ops.sched_in = kvm_sched_in;
3469 kvm_preempt_ops.sched_out = kvm_sched_out;
3470
6aa8b732
AK		 3471 return r;
3472
3473out_free:
c16f862d
RR		 3474 kmem_cache_destroy(kvm_vcpu_cache);
3475out_free_4:
59ae6c6b
AK		 3476 sysdev_unregister(&kvm_sysdev);
3477out_free_3:
3478 sysdev_class_unregister(&kvm_sysdev_class);
3479out_free_2:
6aa8b732 3480 unregister_reboot_notifier(&kvm_reboot_notifier);
774c47f1
AK		 3481 unregister_cpu_notifier(&kvm_cpu_notifier);
3482out_free_1:
1b6c0168 3483 on_each_cpu(hardware_disable, NULL, 0, 1);
002c7f7c 3484out_free_0:
6aa8b732 3485 kvm_arch_ops->hardware_unsetup();
ca45aaae
AK		 3486out:
3487 kvm_arch_ops = NULL;
6aa8b732
AK		 3488 return r;
3489}
3490
3491void kvm_exit_arch(void)
3492{
3493 misc_deregister(&kvm_dev);
c16f862d 3494 kmem_cache_destroy(kvm_vcpu_cache);
59ae6c6b
AK		 3495 sysdev_unregister(&kvm_sysdev);
3496 sysdev_class_unregister(&kvm_sysdev_class);
6aa8b732 3497 unregister_reboot_notifier(&kvm_reboot_notifier);
59ae6c6b 3498 unregister_cpu_notifier(&kvm_cpu_notifier);
1b6c0168 3499 on_each_cpu(hardware_disable, NULL, 0, 1);
6aa8b732 3500 kvm_arch_ops->hardware_unsetup();
09db28b8 3501 kvm_arch_ops = NULL;
6aa8b732
AK		 3502}
3503
3504static __init int kvm_init(void)
3505{
3506 static struct page *bad_page;
37e29d90
AK		 3507 int r;
3508
b5a33a75
AK		 3509 r = kvm_mmu_module_init();
3510 if (r)
3511 goto out4;
3512
6aa8b732
AK		 3513 kvm_init_debug();
3514
bf591b24
MR		 3515 kvm_init_msr_list();
3516
6aa8b732
AK		 3517 if ((bad_page = alloc_page(GFP_KERNEL)) == NULL) {
3518 r = -ENOMEM;
3519 goto out;
3520 }
3521
3522 bad_page_address = page_to_pfn(bad_page) << PAGE_SHIFT;
3523 memset(__va(bad_page_address), 0, PAGE_SIZE);
3524
58e690e6 3525 return 0;
6aa8b732
AK		 3526
3527out:
3528 kvm_exit_debug();
b5a33a75
AK		 3529 kvm_mmu_module_exit();
3530out4:
6aa8b732
AK		 3531 return r;
3532}
3533
3534static __exit void kvm_exit(void)
3535{
3536 kvm_exit_debug();
3537 __free_page(pfn_to_page(bad_page_address >> PAGE_SHIFT));
b5a33a75 3538 kvm_mmu_module_exit();
6aa8b732
AK		 3539}
3540
3541module_init(kvm_init)
3542module_exit(kvm_exit)
3543
3544EXPORT_SYMBOL_GPL(kvm_init_arch);
3545EXPORT_SYMBOL_GPL(kvm_exit_arch);
Ubuntu Artful kernel mirror