]> git.proxmox.com Git - mirror_ubuntu-bionic-kernel.git/blame - drivers/kvm/kvm_main.c
projects / mirror_ubuntu-bionic-kernel.git / blame
? search:
summary | shortlog | log | commit | commitdiff | tree
blob | blame (incremental) | history | HEAD
KVM: Portability: Split kvm_vcpu into arch dependent and independent parts (part 1)
[mirror_ubuntu-bionic-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"
043405e1 19#include "x86.h"
e495606d
AK		 20#include "x86_emulate.h"
21#include "segment_descriptor.h"
85f455f7 22#include "irq.h"
6aa8b732
AK		 23
24#include <linux/kvm.h>
25#include <linux/module.h>
26#include <linux/errno.h>
6aa8b732
AK		 27#include <linux/percpu.h>
28#include <linux/gfp.h>
6aa8b732
AK		 29#include <linux/mm.h>
30#include <linux/miscdevice.h>
31#include <linux/vmalloc.h>
6aa8b732 32#include <linux/reboot.h>
6aa8b732
AK		 33#include <linux/debugfs.h>
34#include <linux/highmem.h>
35#include <linux/file.h>
59ae6c6b 36#include <linux/sysdev.h>
774c47f1 37#include <linux/cpu.h>
e8edc6e0 38#include <linux/sched.h>
d9e368d6
AK		 39#include <linux/cpumask.h>
40#include <linux/smp.h>
d6d28168 41#include <linux/anon_inodes.h>
04d2cc77 42#include <linux/profile.h>
7aa81cc0 43#include <linux/kvm_para.h>
6fc138d2 44#include <linux/pagemap.h>
8d4e1288 45#include <linux/mman.h>
6aa8b732 46
e495606d
AK		 47#include <asm/processor.h>
48#include <asm/msr.h>
49#include <asm/io.h>
50#include <asm/uaccess.h>
51#include <asm/desc.h>
6aa8b732
AK		 52
53MODULE_AUTHOR("Qumranet");
54MODULE_LICENSE("GPL");
55
133de902
AK		 56static DEFINE_SPINLOCK(kvm_lock);
57static LIST_HEAD(vm_list);
58
1b6c0168
AK		 59static cpumask_t cpus_hardware_enabled;
60
cbdd1bea 61struct kvm_x86_ops *kvm_x86_ops;
c16f862d
RR		 62struct kmem_cache *kvm_vcpu_cache;
63EXPORT_SYMBOL_GPL(kvm_vcpu_cache);
1165f5fe 64
15ad7146
AK		 65static __read_mostly struct preempt_ops kvm_preempt_ops;
66
1165f5fe 67#define STAT_OFFSET(x) offsetof(struct kvm_vcpu, stat.x)
6aa8b732
AK		 68
69static struct kvm_stats_debugfs_item {
70 const char *name;
1165f5fe 71 int offset;
6aa8b732
AK		 72 struct dentry *dentry;
73} debugfs_entries[] = {
1165f5fe
AK		 74 { "pf_fixed", STAT_OFFSET(pf_fixed) },
75 { "pf_guest", STAT_OFFSET(pf_guest) },
76 { "tlb_flush", STAT_OFFSET(tlb_flush) },
77 { "invlpg", STAT_OFFSET(invlpg) },
78 { "exits", STAT_OFFSET(exits) },
79 { "io_exits", STAT_OFFSET(io_exits) },
80 { "mmio_exits", STAT_OFFSET(mmio_exits) },
81 { "signal_exits", STAT_OFFSET(signal_exits) },
82 { "irq_window", STAT_OFFSET(irq_window_exits) },
83 { "halt_exits", STAT_OFFSET(halt_exits) },
b6958ce4 84 { "halt_wakeup", STAT_OFFSET(halt_wakeup) },
1165f5fe
AK		 85 { "request_irq", STAT_OFFSET(request_irq_exits) },
86 { "irq_exits", STAT_OFFSET(irq_exits) },
e6adf283 87 { "light_exits", STAT_OFFSET(light_exits) },
2cc51560 88 { "efer_reload", STAT_OFFSET(efer_reload) },
1165f5fe 89 { NULL }
6aa8b732
AK		 90};
91
92static struct dentry *debugfs_dir;
93
707d92fa
RR		 94#define CR0_RESERVED_BITS \
95 (~(unsigned long)(X86_CR0_PE | X86_CR0_MP | X86_CR0_EM | X86_CR0_TS \
96 | X86_CR0_ET | X86_CR0_NE | X86_CR0_WP | X86_CR0_AM \
97 | X86_CR0_NW | X86_CR0_CD | X86_CR0_PG))
66aee91a
RR		 98#define CR4_RESERVED_BITS \
99 (~(unsigned long)(X86_CR4_VME | X86_CR4_PVI | X86_CR4_TSD | X86_CR4_DE\
100 | X86_CR4_PSE | X86_CR4_PAE | X86_CR4_MCE \
101 | X86_CR4_PGE | X86_CR4_PCE | X86_CR4_OSFXSR \
102 | X86_CR4_OSXMMEXCPT | X86_CR4_VMXE))
103
7075bc81 104#define CR8_RESERVED_BITS (~(unsigned long)X86_CR8_TPR)
6aa8b732
AK		 105#define EFER_RESERVED_BITS 0xfffffffffffff2fe
106
05b3e0c2 107#ifdef CONFIG_X86_64
d77c26fc 108/* LDT or TSS descriptor in the GDT. 16 bytes. */
6aa8b732
AK		 109struct segment_descriptor_64 {
110 struct segment_descriptor s;
111 u32 base_higher;
112 u32 pad_zero;
113};
114
115#endif
116
bccf2150
AK		 117static long kvm_vcpu_ioctl(struct file *file, unsigned int ioctl,
118 unsigned long arg);
119
6aa8b732
AK		 120unsigned long segment_base(u16 selector)
121{
122 struct descriptor_table gdt;
123 struct segment_descriptor *d;
124 unsigned long table_base;
6aa8b732
AK		 125 unsigned long v;
126
127 if (selector == 0)
128 return 0;
129
d77c26fc 130 asm("sgdt %0" : "=m"(gdt));
6aa8b732
AK		 131 table_base = gdt.base;
132
133 if (selector & 4) { /* from ldt */
134 u16 ldt_selector;
135
d77c26fc 136 asm("sldt %0" : "=g"(ldt_selector));
6aa8b732
AK		 137 table_base = segment_base(ldt_selector);
138 }
139 d = (struct segment_descriptor *)(table_base + (selector & ~7));
d77c26fc
MD		 140 v = d->base_low | ((unsigned long)d->base_mid << 16) |
141 ((unsigned long)d->base_high << 24);
05b3e0c2 142#ifdef CONFIG_X86_64
d77c26fc
MD		 143 if (d->system == 0 && (d->type == 2 || d->type == 9 || d->type == 11))
144 v |= ((unsigned long) \
145 ((struct segment_descriptor_64 *)d)->base_higher) << 32;
6aa8b732
AK		 146#endif
147 return v;
148}
149EXPORT_SYMBOL_GPL(segment_base);
150
5aacf0ca
JM		 151static inline int valid_vcpu(int n)
152{
153 return likely(n >= 0 && n < KVM_MAX_VCPUS);
154}
155
7702fd1f
AK		 156void kvm_load_guest_fpu(struct kvm_vcpu *vcpu)
157{
158 if (!vcpu->fpu_active || vcpu->guest_fpu_loaded)
159 return;
160
161 vcpu->guest_fpu_loaded = 1;
b114b080
RR		 162 fx_save(&vcpu->host_fx_image);
163 fx_restore(&vcpu->guest_fx_image);
7702fd1f
AK		 164}
165EXPORT_SYMBOL_GPL(kvm_load_guest_fpu);
166
167void kvm_put_guest_fpu(struct kvm_vcpu *vcpu)
168{
169 if (!vcpu->guest_fpu_loaded)
170 return;
171
172 vcpu->guest_fpu_loaded = 0;
b114b080
RR		 173 fx_save(&vcpu->guest_fx_image);
174 fx_restore(&vcpu->host_fx_image);
7702fd1f
AK		 175}
176EXPORT_SYMBOL_GPL(kvm_put_guest_fpu);
177
bccf2150
AK		 178/*
179 * Switches to specified vcpu, until a matching vcpu_put()
180 */
313a3dc7 181void vcpu_load(struct kvm_vcpu *vcpu)
6aa8b732 182{
15ad7146
AK		 183 int cpu;
184
bccf2150 185 mutex_lock(&vcpu->mutex);
15ad7146
AK		 186 cpu = get_cpu();
187 preempt_notifier_register(&vcpu->preempt_notifier);
313a3dc7 188 kvm_arch_vcpu_load(vcpu, cpu);
15ad7146 189 put_cpu();
6aa8b732
AK		 190}
191
313a3dc7 192void vcpu_put(struct kvm_vcpu *vcpu)
6aa8b732 193{
15ad7146 194 preempt_disable();
313a3dc7 195 kvm_arch_vcpu_put(vcpu);
15ad7146
AK		 196 preempt_notifier_unregister(&vcpu->preempt_notifier);
197 preempt_enable();
6aa8b732
AK		 198 mutex_unlock(&vcpu->mutex);
199}
200
d9e368d6
AK		 201static void ack_flush(void *_completed)
202{
d9e368d6
AK		 203}
204
205void kvm_flush_remote_tlbs(struct kvm *kvm)
206{
49d3bd7e 207 int i, cpu;
d9e368d6
AK		 208 cpumask_t cpus;
209 struct kvm_vcpu *vcpu;
d9e368d6 210
d9e368d6 211 cpus_clear(cpus);
fb3f0f51
RR		 212 for (i = 0; i < KVM_MAX_VCPUS; ++i) {
213 vcpu = kvm->vcpus[i];
214 if (!vcpu)
215 continue;
3176bc3e 216 if (test_and_set_bit(KVM_REQ_TLB_FLUSH, &vcpu->requests))
d9e368d6
AK		 217 continue;
218 cpu = vcpu->cpu;
219 if (cpu != -1 && cpu != raw_smp_processor_id())
49d3bd7e 220 cpu_set(cpu, cpus);
d9e368d6 221 }
49d3bd7e 222 smp_call_function_mask(cpus, ack_flush, NULL, 1);
d9e368d6
AK		 223}
224
fb3f0f51
RR		 225int kvm_vcpu_init(struct kvm_vcpu *vcpu, struct kvm *kvm, unsigned id)
226{
227 struct page *page;
228 int r;
229
230 mutex_init(&vcpu->mutex);
231 vcpu->cpu = -1;
232 vcpu->mmu.root_hpa = INVALID_PAGE;
233 vcpu->kvm = kvm;
234 vcpu->vcpu_id = id;
c5ec1534
HQ		 235 if (!irqchip_in_kernel(kvm) || id == 0)
236 vcpu->mp_state = VCPU_MP_STATE_RUNNABLE;
237 else
238 vcpu->mp_state = VCPU_MP_STATE_UNINITIALIZED;
b6958ce4 239 init_waitqueue_head(&vcpu->wq);
fb3f0f51
RR		 240
241 page = alloc_page(GFP_KERNEL | __GFP_ZERO);
242 if (!page) {
243 r = -ENOMEM;
244 goto fail;
245 }
246 vcpu->run = page_address(page);
247
248 page = alloc_page(GFP_KERNEL | __GFP_ZERO);
249 if (!page) {
250 r = -ENOMEM;
251 goto fail_free_run;
252 }
253 vcpu->pio_data = page_address(page);
254
fb3f0f51
RR		 255 r = kvm_mmu_create(vcpu);
256 if (r < 0)
257 goto fail_free_pio_data;
258
76fafa5e
RR		 259 if (irqchip_in_kernel(kvm)) {
260 r = kvm_create_lapic(vcpu);
261 if (r < 0)
262 goto fail_mmu_destroy;
263 }
264
fb3f0f51
RR		 265 return 0;
266
76fafa5e
RR		 267fail_mmu_destroy:
268 kvm_mmu_destroy(vcpu);
fb3f0f51
RR		 269fail_free_pio_data:
270 free_page((unsigned long)vcpu->pio_data);
271fail_free_run:
272 free_page((unsigned long)vcpu->run);
273fail:
76fafa5e 274 return r;
fb3f0f51
RR		 275}
276EXPORT_SYMBOL_GPL(kvm_vcpu_init);
277
278void kvm_vcpu_uninit(struct kvm_vcpu *vcpu)
279{
d589444e 280 kvm_free_lapic(vcpu);
fb3f0f51
RR		 281 kvm_mmu_destroy(vcpu);
282 free_page((unsigned long)vcpu->pio_data);
283 free_page((unsigned long)vcpu->run);
284}
285EXPORT_SYMBOL_GPL(kvm_vcpu_uninit);
286
f17abe9a 287static struct kvm *kvm_create_vm(void)
6aa8b732
AK		 288{
289 struct kvm *kvm = kzalloc(sizeof(struct kvm), GFP_KERNEL);
6aa8b732
AK		 290
291 if (!kvm)
f17abe9a 292 return ERR_PTR(-ENOMEM);
6aa8b732 293
74906345 294 kvm_io_bus_init(&kvm->pio_bus);
11ec2804 295 mutex_init(&kvm->lock);
6aa8b732 296 INIT_LIST_HEAD(&kvm->active_mmu_pages);
2eeb2e94 297 kvm_io_bus_init(&kvm->mmio_bus);
5e58cfe4
RR		 298 spin_lock(&kvm_lock);
299 list_add(&kvm->vm_list, &vm_list);
300 spin_unlock(&kvm_lock);
f17abe9a
AK		 301 return kvm;
302}
303
6aa8b732
AK		 304/*
305 * Free any memory in @free but not in @dont.
306 */
307static void kvm_free_physmem_slot(struct kvm_memory_slot *free,
308 struct kvm_memory_slot *dont)
309{
290fc38d
IE		 310 if (!dont || free->rmap != dont->rmap)
311 vfree(free->rmap);
6aa8b732
AK		 312
313 if (!dont || free->dirty_bitmap != dont->dirty_bitmap)
314 vfree(free->dirty_bitmap);
315
6aa8b732 316 free->npages = 0;
8b6d44c7 317 free->dirty_bitmap = NULL;
8d4e1288 318 free->rmap = NULL;
6aa8b732
AK		 319}
320
321static void kvm_free_physmem(struct kvm *kvm)
322{
323 int i;
324
325 for (i = 0; i < kvm->nmemslots; ++i)
8b6d44c7 326 kvm_free_physmem_slot(&kvm->memslots[i], NULL);
6aa8b732
AK		 327}
328
039576c0
AK		 329static void free_pio_guest_pages(struct kvm_vcpu *vcpu)
330{
331 int i;
332
3077c451 333 for (i = 0; i < ARRAY_SIZE(vcpu->pio.guest_pages); ++i)
039576c0 334 if (vcpu->pio.guest_pages[i]) {
8a7ae055 335 kvm_release_page(vcpu->pio.guest_pages[i]);
039576c0
AK		 336 vcpu->pio.guest_pages[i] = NULL;
337 }
338}
339
7b53aa56
AK		 340static void kvm_unload_vcpu_mmu(struct kvm_vcpu *vcpu)
341{
7b53aa56
AK		 342 vcpu_load(vcpu);
343 kvm_mmu_unload(vcpu);
344 vcpu_put(vcpu);
345}
346
6aa8b732
AK		 347static void kvm_free_vcpus(struct kvm *kvm)
348{
349 unsigned int i;
350
7b53aa56
AK		 351 /*
352 * Unpin any mmu pages first.
353 */
354 for (i = 0; i < KVM_MAX_VCPUS; ++i)
fb3f0f51
RR		 355 if (kvm->vcpus[i])
356 kvm_unload_vcpu_mmu(kvm->vcpus[i]);
357 for (i = 0; i < KVM_MAX_VCPUS; ++i) {
358 if (kvm->vcpus[i]) {
cbdd1bea 359 kvm_x86_ops->vcpu_free(kvm->vcpus[i]);
fb3f0f51
RR		 360 kvm->vcpus[i] = NULL;
361 }
362 }
363
6aa8b732
AK		 364}
365
f17abe9a
AK		 366static void kvm_destroy_vm(struct kvm *kvm)
367{
133de902
AK		 368 spin_lock(&kvm_lock);
369 list_del(&kvm->vm_list);
370 spin_unlock(&kvm_lock);
74906345 371 kvm_io_bus_destroy(&kvm->pio_bus);
2eeb2e94 372 kvm_io_bus_destroy(&kvm->mmio_bus);
85f455f7 373 kfree(kvm->vpic);
1fd4f2a5 374 kfree(kvm->vioapic);
6aa8b732
AK		 375 kvm_free_vcpus(kvm);
376 kvm_free_physmem(kvm);
377 kfree(kvm);
f17abe9a
AK		 378}
379
380static int kvm_vm_release(struct inode *inode, struct file *filp)
381{
382 struct kvm *kvm = filp->private_data;
383
384 kvm_destroy_vm(kvm);
6aa8b732
AK		 385 return 0;
386}
387
388static void inject_gp(struct kvm_vcpu *vcpu)
389{
cbdd1bea 390 kvm_x86_ops->inject_gp(vcpu, 0);
6aa8b732
AK		 391}
392
1342d353
AK		 393/*
394 * Load the pae pdptrs. Return true is they are all valid.
395 */
396static int load_pdptrs(struct kvm_vcpu *vcpu, unsigned long cr3)
6aa8b732
AK		 397{
398 gfn_t pdpt_gfn = cr3 >> PAGE_SHIFT;
1342d353 399 unsigned offset = ((cr3 & (PAGE_SIZE-1)) >> 5) << 2;
6aa8b732 400 int i;
1342d353 401 int ret;
c820c2aa 402 u64 pdpte[ARRAY_SIZE(vcpu->pdptrs)];
6aa8b732 403
11ec2804 404 mutex_lock(&vcpu->kvm->lock);
195aefde
IE		 405 ret = kvm_read_guest_page(vcpu->kvm, pdpt_gfn, pdpte,
406 offset * sizeof(u64), sizeof(pdpte));
407 if (ret < 0) {
c820c2aa
RR		 408 ret = 0;
409 goto out;
410 }
c820c2aa
RR		 411 for (i = 0; i < ARRAY_SIZE(pdpte); ++i) {
412 if ((pdpte[i] & 1) && (pdpte[i] & 0xfffffff0000001e6ull)) {
1342d353
AK		 413 ret = 0;
414 goto out;
415 }
6aa8b732 416 }
c820c2aa 417 ret = 1;
6aa8b732 418
c820c2aa 419 memcpy(vcpu->pdptrs, pdpte, sizeof(vcpu->pdptrs));
1342d353 420out:
11ec2804 421 mutex_unlock(&vcpu->kvm->lock);
6aa8b732 422
1342d353 423 return ret;
6aa8b732
AK		 424}
425
426void set_cr0(struct kvm_vcpu *vcpu, unsigned long cr0)
427{
707d92fa 428 if (cr0 & CR0_RESERVED_BITS) {
6aa8b732
AK		 429 printk(KERN_DEBUG "set_cr0: 0x%lx #GP, reserved bits 0x%lx\n",
430 cr0, vcpu->cr0);
431 inject_gp(vcpu);
432 return;
433 }
434
707d92fa 435 if ((cr0 & X86_CR0_NW) && !(cr0 & X86_CR0_CD)) {
6aa8b732
AK		 436 printk(KERN_DEBUG "set_cr0: #GP, CD == 0 && NW == 1\n");
437 inject_gp(vcpu);
438 return;
439 }
440
707d92fa 441 if ((cr0 & X86_CR0_PG) && !(cr0 & X86_CR0_PE)) {
6aa8b732
AK		 442 printk(KERN_DEBUG "set_cr0: #GP, set PG flag "
443 "and a clear PE flag\n");
444 inject_gp(vcpu);
445 return;
446 }
447
707d92fa 448 if (!is_paging(vcpu) && (cr0 & X86_CR0_PG)) {
05b3e0c2 449#ifdef CONFIG_X86_64
6aa8b732
AK		 450 if ((vcpu->shadow_efer & EFER_LME)) {
451 int cs_db, cs_l;
452
453 if (!is_pae(vcpu)) {
454 printk(KERN_DEBUG "set_cr0: #GP, start paging "
455 "in long mode while PAE is disabled\n");
456 inject_gp(vcpu);
457 return;
458 }
cbdd1bea 459 kvm_x86_ops->get_cs_db_l_bits(vcpu, &cs_db, &cs_l);
6aa8b732
AK		 460 if (cs_l) {
461 printk(KERN_DEBUG "set_cr0: #GP, start paging "
462 "in long mode while CS.L == 1\n");
463 inject_gp(vcpu);
464 return;
465
466 }
467 } else
468#endif
1342d353 469 if (is_pae(vcpu) && !load_pdptrs(vcpu, vcpu->cr3)) {
6aa8b732
AK		 470 printk(KERN_DEBUG "set_cr0: #GP, pdptrs "
471 "reserved bits\n");
472 inject_gp(vcpu);
473 return;
474 }
475
476 }
477
cbdd1bea 478 kvm_x86_ops->set_cr0(vcpu, cr0);
6aa8b732
AK		 479 vcpu->cr0 = cr0;
480
11ec2804 481 mutex_lock(&vcpu->kvm->lock);
6aa8b732 482 kvm_mmu_reset_context(vcpu);
11ec2804 483 mutex_unlock(&vcpu->kvm->lock);
6aa8b732
AK		 484 return;
485}
486EXPORT_SYMBOL_GPL(set_cr0);
487
488void lmsw(struct kvm_vcpu *vcpu, unsigned long msw)
489{
490 set_cr0(vcpu, (vcpu->cr0 & ~0x0ful) | (msw & 0x0f));
491}
492EXPORT_SYMBOL_GPL(lmsw);
493
494void set_cr4(struct kvm_vcpu *vcpu, unsigned long cr4)
495{
66aee91a 496 if (cr4 & CR4_RESERVED_BITS) {
6aa8b732
AK		 497 printk(KERN_DEBUG "set_cr4: #GP, reserved bits\n");
498 inject_gp(vcpu);
499 return;
500 }
501
a9058ecd 502 if (is_long_mode(vcpu)) {
66aee91a 503 if (!(cr4 & X86_CR4_PAE)) {
6aa8b732
AK		 504 printk(KERN_DEBUG "set_cr4: #GP, clearing PAE while "
505 "in long mode\n");
506 inject_gp(vcpu);
507 return;
508 }
66aee91a 509 } else if (is_paging(vcpu) && !is_pae(vcpu) && (cr4 & X86_CR4_PAE)
1342d353 510 && !load_pdptrs(vcpu, vcpu->cr3)) {
6aa8b732
AK		 511 printk(KERN_DEBUG "set_cr4: #GP, pdptrs reserved bits\n");
512 inject_gp(vcpu);
310bc76c 513 return;
6aa8b732
AK		 514 }
515
66aee91a 516 if (cr4 & X86_CR4_VMXE) {
6aa8b732
AK		 517 printk(KERN_DEBUG "set_cr4: #GP, setting VMXE\n");
518 inject_gp(vcpu);
519 return;
520 }
cbdd1bea 521 kvm_x86_ops->set_cr4(vcpu, cr4);
81f50e3b 522 vcpu->cr4 = cr4;
11ec2804 523 mutex_lock(&vcpu->kvm->lock);
6aa8b732 524 kvm_mmu_reset_context(vcpu);
11ec2804 525 mutex_unlock(&vcpu->kvm->lock);
6aa8b732
AK		 526}
527EXPORT_SYMBOL_GPL(set_cr4);
528
529void set_cr3(struct kvm_vcpu *vcpu, unsigned long cr3)
530{
a9058ecd 531 if (is_long_mode(vcpu)) {
f802a307 532 if (cr3 & CR3_L_MODE_RESERVED_BITS) {
6aa8b732
AK		 533 printk(KERN_DEBUG "set_cr3: #GP, reserved bits\n");
534 inject_gp(vcpu);
535 return;
536 }
537 } else {
f802a307
RR		 538 if (is_pae(vcpu)) {
539 if (cr3 & CR3_PAE_RESERVED_BITS) {
540 printk(KERN_DEBUG
541 "set_cr3: #GP, reserved bits\n");
542 inject_gp(vcpu);
543 return;
544 }
545 if (is_paging(vcpu) && !load_pdptrs(vcpu, cr3)) {
546 printk(KERN_DEBUG "set_cr3: #GP, pdptrs "
547 "reserved bits\n");
548 inject_gp(vcpu);
549 return;
550 }
6aa8b732 551 }
21764863
RH		 552 /*
553 * We don't check reserved bits in nonpae mode, because
554 * this isn't enforced, and VMware depends on this.
555 */
6aa8b732
AK		 556 }
557
11ec2804 558 mutex_lock(&vcpu->kvm->lock);
d21225ee
IM		 559 /*
560 * Does the new cr3 value map to physical memory? (Note, we
561 * catch an invalid cr3 even in real-mode, because it would
562 * cause trouble later on when we turn on paging anyway.)
563 *
564 * A real CPU would silently accept an invalid cr3 and would
565 * attempt to use it - with largely undefined (and often hard
566 * to debug) behavior on the guest side.
567 */
568 if (unlikely(!gfn_to_memslot(vcpu->kvm, cr3 >> PAGE_SHIFT)))
569 inject_gp(vcpu);
fb764416
RR		 570 else {
571 vcpu->cr3 = cr3;
d21225ee 572 vcpu->mmu.new_cr3(vcpu);
fb764416 573 }
11ec2804 574 mutex_unlock(&vcpu->kvm->lock);
6aa8b732
AK		 575}
576EXPORT_SYMBOL_GPL(set_cr3);
577
578void set_cr8(struct kvm_vcpu *vcpu, unsigned long cr8)
579{
7075bc81 580 if (cr8 & CR8_RESERVED_BITS) {
6aa8b732
AK		 581 printk(KERN_DEBUG "set_cr8: #GP, reserved bits 0x%lx\n", cr8);
582 inject_gp(vcpu);
583 return;
584 }
97222cc8
ED		 585 if (irqchip_in_kernel(vcpu->kvm))
586 kvm_lapic_set_tpr(vcpu, cr8);
587 else
588 vcpu->cr8 = cr8;
6aa8b732
AK		 589}
590EXPORT_SYMBOL_GPL(set_cr8);
591
7017fc3d
ED		 592unsigned long get_cr8(struct kvm_vcpu *vcpu)
593{
97222cc8
ED		 594 if (irqchip_in_kernel(vcpu->kvm))
595 return kvm_lapic_get_cr8(vcpu);
596 else
597 return vcpu->cr8;
7017fc3d
ED		 598}
599EXPORT_SYMBOL_GPL(get_cr8);
600
601u64 kvm_get_apic_base(struct kvm_vcpu *vcpu)
602{
97222cc8
ED		 603 if (irqchip_in_kernel(vcpu->kvm))
604 return vcpu->apic_base;
605 else
606 return vcpu->apic_base;
7017fc3d
ED		 607}
608EXPORT_SYMBOL_GPL(kvm_get_apic_base);
609
610void kvm_set_apic_base(struct kvm_vcpu *vcpu, u64 data)
611{
97222cc8
ED		 612 /* TODO: reserve bits check */
613 if (irqchip_in_kernel(vcpu->kvm))
614 kvm_lapic_set_base(vcpu, data);
615 else
616 vcpu->apic_base = data;
7017fc3d
ED		 617}
618EXPORT_SYMBOL_GPL(kvm_set_apic_base);
619
6aa8b732
AK		 620void fx_init(struct kvm_vcpu *vcpu)
621{
b114b080 622 unsigned after_mxcsr_mask;
6aa8b732 623
9bd01506
RR		 624 /* Initialize guest FPU by resetting ours and saving into guest's */
625 preempt_disable();
b114b080 626 fx_save(&vcpu->host_fx_image);
6aa8b732 627 fpu_init();
b114b080
RR		 628 fx_save(&vcpu->guest_fx_image);
629 fx_restore(&vcpu->host_fx_image);
9bd01506 630 preempt_enable();
6aa8b732 631
380102c8 632 vcpu->cr0 |= X86_CR0_ET;
b114b080
RR		 633 after_mxcsr_mask = offsetof(struct i387_fxsave_struct, st_space);
634 vcpu->guest_fx_image.mxcsr = 0x1f80;
635 memset((void *)&vcpu->guest_fx_image + after_mxcsr_mask,
636 0, sizeof(struct i387_fxsave_struct) - after_mxcsr_mask);
6aa8b732
AK		 637}
638EXPORT_SYMBOL_GPL(fx_init);
639
6aa8b732
AK		 640/*
641 * Allocate some memory and give it an address in the guest physical address
642 * space.
643 *
644 * Discontiguous memory is allowed, mostly for framebuffers.
645 */
2c6f5df9 646static int kvm_vm_ioctl_set_memory_region(struct kvm *kvm,
6fc138d2
IE		 647 struct
648 kvm_userspace_memory_region *mem,
649 int user_alloc)
6aa8b732
AK		 650{
651 int r;
652 gfn_t base_gfn;
653 unsigned long npages;
654 unsigned long i;
655 struct kvm_memory_slot *memslot;
656 struct kvm_memory_slot old, new;
6aa8b732
AK		 657
658 r = -EINVAL;
659 /* General sanity checks */
660 if (mem->memory_size & (PAGE_SIZE - 1))
661 goto out;
662 if (mem->guest_phys_addr & (PAGE_SIZE - 1))
663 goto out;
664 if (mem->slot >= KVM_MEMORY_SLOTS)
665 goto out;
666 if (mem->guest_phys_addr + mem->memory_size < mem->guest_phys_addr)
667 goto out;
668
669 memslot = &kvm->memslots[mem->slot];
670 base_gfn = mem->guest_phys_addr >> PAGE_SHIFT;
671 npages = mem->memory_size >> PAGE_SHIFT;
672
673 if (!npages)
674 mem->flags &= ~KVM_MEM_LOG_DIRTY_PAGES;
675
11ec2804 676 mutex_lock(&kvm->lock);
6aa8b732 677
6aa8b732
AK		 678 new = old = *memslot;
679
680 new.base_gfn = base_gfn;
681 new.npages = npages;
682 new.flags = mem->flags;
683
684 /* Disallow changing a memory slot's size. */
685 r = -EINVAL;
686 if (npages && old.npages && npages != old.npages)
687 goto out_unlock;
688
689 /* Check for overlaps */
690 r = -EEXIST;
691 for (i = 0; i < KVM_MEMORY_SLOTS; ++i) {
692 struct kvm_memory_slot *s = &kvm->memslots[i];
693
694 if (s == memslot)
695 continue;
696 if (!((base_gfn + npages <= s->base_gfn) ||
697 (base_gfn >= s->base_gfn + s->npages)))
698 goto out_unlock;
699 }
6aa8b732 700
6aa8b732
AK		 701 /* Free page dirty bitmap if unneeded */
702 if (!(new.flags & KVM_MEM_LOG_DIRTY_PAGES))
8b6d44c7 703 new.dirty_bitmap = NULL;
6aa8b732
AK		 704
705 r = -ENOMEM;
706
707 /* Allocate if a slot is being created */
8d4e1288 708 if (npages && !new.rmap) {
d77c26fc 709 new.rmap = vmalloc(npages * sizeof(struct page *));
290fc38d
IE		 710
711 if (!new.rmap)
712 goto out_unlock;
713
290fc38d 714 memset(new.rmap, 0, npages * sizeof(*new.rmap));
8d4e1288
AL		 715
716 if (user_alloc)
8a7ae055 717 new.userspace_addr = mem->userspace_addr;
8d4e1288
AL		 718 else {
719 down_write(&current->mm->mmap_sem);
720 new.userspace_addr = do_mmap(NULL, 0,
721 npages * PAGE_SIZE,
722 PROT_READ | PROT_WRITE,
723 MAP_SHARED | MAP_ANONYMOUS,
724 0);
725 up_write(&current->mm->mmap_sem);
726
727 if (IS_ERR((void *)new.userspace_addr))
728 goto out_unlock;
6aa8b732
AK		 729 }
730 }
731
732 /* Allocate page dirty bitmap if needed */
733 if ((new.flags & KVM_MEM_LOG_DIRTY_PAGES) && !new.dirty_bitmap) {
734 unsigned dirty_bytes = ALIGN(npages, BITS_PER_LONG) / 8;
735
736 new.dirty_bitmap = vmalloc(dirty_bytes);
737 if (!new.dirty_bitmap)
0d8d2bd4 738 goto out_unlock;
6aa8b732
AK		 739 memset(new.dirty_bitmap, 0, dirty_bytes);
740 }
741
6aa8b732
AK		 742 if (mem->slot >= kvm->nmemslots)
743 kvm->nmemslots = mem->slot + 1;
744
82ce2c96
IE		 745 if (!kvm->n_requested_mmu_pages) {
746 unsigned int n_pages;
747
748 if (npages) {
749 n_pages = npages * KVM_PERMILLE_MMU_PAGES / 1000;
750 kvm_mmu_change_mmu_pages(kvm, kvm->n_alloc_mmu_pages +
751 n_pages);
752 } else {
753 unsigned int nr_mmu_pages;
754
755 n_pages = old.npages * KVM_PERMILLE_MMU_PAGES / 1000;
756 nr_mmu_pages = kvm->n_alloc_mmu_pages - n_pages;
757 nr_mmu_pages = max(nr_mmu_pages,
758 (unsigned int) KVM_MIN_ALLOC_MMU_PAGES);
759 kvm_mmu_change_mmu_pages(kvm, nr_mmu_pages);
760 }
761 }
762
6aa8b732 763 *memslot = new;
6aa8b732 764
90cb0529
AK		 765 kvm_mmu_slot_remove_write_access(kvm, mem->slot);
766 kvm_flush_remote_tlbs(kvm);
6aa8b732 767
11ec2804 768 mutex_unlock(&kvm->lock);
6aa8b732
AK		 769
770 kvm_free_physmem_slot(&old, &new);
771 return 0;
772
773out_unlock:
11ec2804 774 mutex_unlock(&kvm->lock);
6aa8b732
AK		 775 kvm_free_physmem_slot(&new, &old);
776out:
777 return r;
778}
779
82ce2c96
IE		 780static int kvm_vm_ioctl_set_nr_mmu_pages(struct kvm *kvm,
781 u32 kvm_nr_mmu_pages)
782{
783 if (kvm_nr_mmu_pages < KVM_MIN_ALLOC_MMU_PAGES)
784 return -EINVAL;
785
786 mutex_lock(&kvm->lock);
787
788 kvm_mmu_change_mmu_pages(kvm, kvm_nr_mmu_pages);
789 kvm->n_requested_mmu_pages = kvm_nr_mmu_pages;
790
791 mutex_unlock(&kvm->lock);
792 return 0;
793}
794
795static int kvm_vm_ioctl_get_nr_mmu_pages(struct kvm *kvm)
796{
797 return kvm->n_alloc_mmu_pages;
798}
799
6aa8b732
AK		 800/*
801 * Get (and clear) the dirty memory log for a memory slot.
802 */
2c6f5df9
AK		 803static int kvm_vm_ioctl_get_dirty_log(struct kvm *kvm,
804 struct kvm_dirty_log *log)
6aa8b732
AK		 805{
806 struct kvm_memory_slot *memslot;
807 int r, i;
808 int n;
809 unsigned long any = 0;
810
11ec2804 811 mutex_lock(&kvm->lock);
6aa8b732 812
6aa8b732
AK		 813 r = -EINVAL;
814 if (log->slot >= KVM_MEMORY_SLOTS)
815 goto out;
816
817 memslot = &kvm->memslots[log->slot];
818 r = -ENOENT;
819 if (!memslot->dirty_bitmap)
820 goto out;
821
cd1a4a98 822 n = ALIGN(memslot->npages, BITS_PER_LONG) / 8;
6aa8b732 823
cd1a4a98 824 for (i = 0; !any && i < n/sizeof(long); ++i)
6aa8b732
AK		 825 any = memslot->dirty_bitmap[i];
826
827 r = -EFAULT;
828 if (copy_to_user(log->dirty_bitmap, memslot->dirty_bitmap, n))
829 goto out;
830
39214915
RR		 831 /* If nothing is dirty, don't bother messing with page tables. */
832 if (any) {
39214915
RR		 833 kvm_mmu_slot_remove_write_access(kvm, log->slot);
834 kvm_flush_remote_tlbs(kvm);
835 memset(memslot->dirty_bitmap, 0, n);
39214915 836 }
6aa8b732
AK		 837
838 r = 0;
839
840out:
11ec2804 841 mutex_unlock(&kvm->lock);
6aa8b732
AK		 842 return r;
843}
844
e8207547
AK		 845/*
846 * Set a new alias region. Aliases map a portion of physical memory into
847 * another portion. This is useful for memory windows, for example the PC
848 * VGA region.
849 */
850static int kvm_vm_ioctl_set_memory_alias(struct kvm *kvm,
851 struct kvm_memory_alias *alias)
852{
853 int r, n;
854 struct kvm_mem_alias *p;
855
856 r = -EINVAL;
857 /* General sanity checks */
858 if (alias->memory_size & (PAGE_SIZE - 1))
859 goto out;
860 if (alias->guest_phys_addr & (PAGE_SIZE - 1))
861 goto out;
862 if (alias->slot >= KVM_ALIAS_SLOTS)
863 goto out;
864 if (alias->guest_phys_addr + alias->memory_size
865 < alias->guest_phys_addr)
866 goto out;
867 if (alias->target_phys_addr + alias->memory_size
868 < alias->target_phys_addr)
869 goto out;
870
11ec2804 871 mutex_lock(&kvm->lock);
e8207547
AK		 872
873 p = &kvm->aliases[alias->slot];
874 p->base_gfn = alias->guest_phys_addr >> PAGE_SHIFT;
875 p->npages = alias->memory_size >> PAGE_SHIFT;
876 p->target_gfn = alias->target_phys_addr >> PAGE_SHIFT;
877
878 for (n = KVM_ALIAS_SLOTS; n > 0; --n)
879 if (kvm->aliases[n - 1].npages)
880 break;
881 kvm->naliases = n;
882
90cb0529 883 kvm_mmu_zap_all(kvm);
e8207547 884
11ec2804 885 mutex_unlock(&kvm->lock);
e8207547
AK		 886
887 return 0;
888
889out:
890 return r;
891}
892
6ceb9d79
HQ		 893static int kvm_vm_ioctl_get_irqchip(struct kvm *kvm, struct kvm_irqchip *chip)
894{
895 int r;
896
897 r = 0;
898 switch (chip->chip_id) {
899 case KVM_IRQCHIP_PIC_MASTER:
d77c26fc 900 memcpy(&chip->chip.pic,
6ceb9d79
HQ		 901 &pic_irqchip(kvm)->pics[0],
902 sizeof(struct kvm_pic_state));
903 break;
904 case KVM_IRQCHIP_PIC_SLAVE:
d77c26fc 905 memcpy(&chip->chip.pic,
6ceb9d79
HQ		 906 &pic_irqchip(kvm)->pics[1],
907 sizeof(struct kvm_pic_state));
908 break;
6bf9e962 909 case KVM_IRQCHIP_IOAPIC:
d77c26fc 910 memcpy(&chip->chip.ioapic,
6bf9e962
HQ		 911 ioapic_irqchip(kvm),
912 sizeof(struct kvm_ioapic_state));
913 break;
6ceb9d79
HQ		 914 default:
915 r = -EINVAL;
916 break;
917 }
918 return r;
919}
920
921static int kvm_vm_ioctl_set_irqchip(struct kvm *kvm, struct kvm_irqchip *chip)
922{
923 int r;
924
925 r = 0;
926 switch (chip->chip_id) {
927 case KVM_IRQCHIP_PIC_MASTER:
d77c26fc 928 memcpy(&pic_irqchip(kvm)->pics[0],
6ceb9d79
HQ		 929 &chip->chip.pic,
930 sizeof(struct kvm_pic_state));
931 break;
932 case KVM_IRQCHIP_PIC_SLAVE:
d77c26fc 933 memcpy(&pic_irqchip(kvm)->pics[1],
6ceb9d79
HQ		 934 &chip->chip.pic,
935 sizeof(struct kvm_pic_state));
936 break;
6bf9e962 937 case KVM_IRQCHIP_IOAPIC:
d77c26fc 938 memcpy(ioapic_irqchip(kvm),
6bf9e962
HQ		 939 &chip->chip.ioapic,
940 sizeof(struct kvm_ioapic_state));
941 break;
6ceb9d79
HQ		 942 default:
943 r = -EINVAL;
944 break;
945 }
946 kvm_pic_update_irq(pic_irqchip(kvm));
947 return r;
948}
949
cea7bb21
IE		 950int is_error_page(struct page *page)
951{
952 return page == bad_page;
953}
954EXPORT_SYMBOL_GPL(is_error_page);
955
290fc38d 956gfn_t unalias_gfn(struct kvm *kvm, gfn_t gfn)
e8207547
AK		 957{
958 int i;
959 struct kvm_mem_alias *alias;
960
961 for (i = 0; i < kvm->naliases; ++i) {
962 alias = &kvm->aliases[i];
963 if (gfn >= alias->base_gfn
964 && gfn < alias->base_gfn + alias->npages)
965 return alias->target_gfn + gfn - alias->base_gfn;
966 }
967 return gfn;
968}
969
970static struct kvm_memory_slot *__gfn_to_memslot(struct kvm *kvm, gfn_t gfn)
6aa8b732
AK		 971{
972 int i;
973
974 for (i = 0; i < kvm->nmemslots; ++i) {
975 struct kvm_memory_slot *memslot = &kvm->memslots[i];
976
977 if (gfn >= memslot->base_gfn
978 && gfn < memslot->base_gfn + memslot->npages)
979 return memslot;
980 }
8b6d44c7 981 return NULL;
6aa8b732 982}
e8207547
AK		 983
984struct kvm_memory_slot *gfn_to_memslot(struct kvm *kvm, gfn_t gfn)
985{
986 gfn = unalias_gfn(kvm, gfn);
987 return __gfn_to_memslot(kvm, gfn);
988}
6aa8b732 989
954bbbc2
AK		 990struct page *gfn_to_page(struct kvm *kvm, gfn_t gfn)
991{
992 struct kvm_memory_slot *slot;
8d4e1288
AL		 993 struct page *page[1];
994 int npages;
954bbbc2 995
e8207547
AK		 996 gfn = unalias_gfn(kvm, gfn);
997 slot = __gfn_to_memslot(kvm, gfn);
8a7ae055
IE		 998 if (!slot) {
999 get_page(bad_page);
cea7bb21 1000 return bad_page;
8a7ae055 1001 }
8d4e1288
AL		 1002
1003 down_read(&current->mm->mmap_sem);
1004 npages = get_user_pages(current, current->mm,
1005 slot->userspace_addr
1006 + (gfn - slot->base_gfn) * PAGE_SIZE, 1,
1007 1, 1, page, NULL);
1008 up_read(&current->mm->mmap_sem);
1009 if (npages != 1) {
1010 get_page(bad_page);
1011 return bad_page;
8a7ae055 1012 }
8d4e1288
AL		 1013
1014 return page[0];
954bbbc2
AK		 1015}
1016EXPORT_SYMBOL_GPL(gfn_to_page);
1017
8a7ae055
IE		 1018void kvm_release_page(struct page *page)
1019{
1020 if (!PageReserved(page))
1021 SetPageDirty(page);
1022 put_page(page);
1023}
1024EXPORT_SYMBOL_GPL(kvm_release_page);
1025
195aefde
IE		 1026static int next_segment(unsigned long len, int offset)
1027{
1028 if (len > PAGE_SIZE - offset)
1029 return PAGE_SIZE - offset;
1030 else
1031 return len;
1032}
1033
1034int kvm_read_guest_page(struct kvm *kvm, gfn_t gfn, void *data, int offset,
1035 int len)
1036{
1037 void *page_virt;
1038 struct page *page;
1039
1040 page = gfn_to_page(kvm, gfn);
8a7ae055
IE		 1041 if (is_error_page(page)) {
1042 kvm_release_page(page);
195aefde 1043 return -EFAULT;
8a7ae055 1044 }
195aefde
IE		 1045 page_virt = kmap_atomic(page, KM_USER0);
1046
1047 memcpy(data, page_virt + offset, len);
1048
1049 kunmap_atomic(page_virt, KM_USER0);
8a7ae055 1050 kvm_release_page(page);
195aefde
IE		 1051 return 0;
1052}
1053EXPORT_SYMBOL_GPL(kvm_read_guest_page);
1054
1055int kvm_read_guest(struct kvm *kvm, gpa_t gpa, void *data, unsigned long len)
1056{
1057 gfn_t gfn = gpa >> PAGE_SHIFT;
1058 int seg;
1059 int offset = offset_in_page(gpa);
1060 int ret;
1061
1062 while ((seg = next_segment(len, offset)) != 0) {
1063 ret = kvm_read_guest_page(kvm, gfn, data, offset, seg);
1064 if (ret < 0)
1065 return ret;
1066 offset = 0;
1067 len -= seg;
1068 data += seg;
1069 ++gfn;
1070 }
1071 return 0;
1072}
1073EXPORT_SYMBOL_GPL(kvm_read_guest);
1074
1075int kvm_write_guest_page(struct kvm *kvm, gfn_t gfn, const void *data,
1076 int offset, int len)
1077{
1078 void *page_virt;
1079 struct page *page;
1080
1081 page = gfn_to_page(kvm, gfn);
8a7ae055
IE		 1082 if (is_error_page(page)) {
1083 kvm_release_page(page);
195aefde 1084 return -EFAULT;
8a7ae055 1085 }
195aefde
IE		 1086 page_virt = kmap_atomic(page, KM_USER0);
1087
1088 memcpy(page_virt + offset, data, len);
1089
1090 kunmap_atomic(page_virt, KM_USER0);
1091 mark_page_dirty(kvm, gfn);
8a7ae055 1092 kvm_release_page(page);
195aefde
IE		 1093 return 0;
1094}
1095EXPORT_SYMBOL_GPL(kvm_write_guest_page);
1096
1097int kvm_write_guest(struct kvm *kvm, gpa_t gpa, const void *data,
1098 unsigned long len)
1099{
1100 gfn_t gfn = gpa >> PAGE_SHIFT;
1101 int seg;
1102 int offset = offset_in_page(gpa);
1103 int ret;
1104
1105 while ((seg = next_segment(len, offset)) != 0) {
1106 ret = kvm_write_guest_page(kvm, gfn, data, offset, seg);
1107 if (ret < 0)
1108 return ret;
1109 offset = 0;
1110 len -= seg;
1111 data += seg;
1112 ++gfn;
1113 }
1114 return 0;
1115}
1116
1117int kvm_clear_guest_page(struct kvm *kvm, gfn_t gfn, int offset, int len)
1118{
1119 void *page_virt;
1120 struct page *page;
1121
1122 page = gfn_to_page(kvm, gfn);
8a7ae055
IE		 1123 if (is_error_page(page)) {
1124 kvm_release_page(page);
195aefde 1125 return -EFAULT;
8a7ae055 1126 }
195aefde
IE		 1127 page_virt = kmap_atomic(page, KM_USER0);
1128
1129 memset(page_virt + offset, 0, len);
1130
1131 kunmap_atomic(page_virt, KM_USER0);
8a7ae055 1132 kvm_release_page(page);
195aefde
IE		 1133 return 0;
1134}
1135EXPORT_SYMBOL_GPL(kvm_clear_guest_page);
1136
1137int kvm_clear_guest(struct kvm *kvm, gpa_t gpa, unsigned long len)
1138{
1139 gfn_t gfn = gpa >> PAGE_SHIFT;
1140 int seg;
1141 int offset = offset_in_page(gpa);
1142 int ret;
1143
1144 while ((seg = next_segment(len, offset)) != 0) {
1145 ret = kvm_clear_guest_page(kvm, gfn, offset, seg);
1146 if (ret < 0)
1147 return ret;
1148 offset = 0;
1149 len -= seg;
1150 ++gfn;
1151 }
1152 return 0;
1153}
1154EXPORT_SYMBOL_GPL(kvm_clear_guest);
1155
7e9d619d 1156/* WARNING: Does not work on aliased pages. */
6aa8b732
AK		 1157void mark_page_dirty(struct kvm *kvm, gfn_t gfn)
1158{
31389947 1159 struct kvm_memory_slot *memslot;
6aa8b732 1160
7e9d619d
RR		 1161 memslot = __gfn_to_memslot(kvm, gfn);
1162 if (memslot && memslot->dirty_bitmap) {
1163 unsigned long rel_gfn = gfn - memslot->base_gfn;
6aa8b732 1164
7e9d619d
RR		 1165 /* avoid RMW */
1166 if (!test_bit(rel_gfn, memslot->dirty_bitmap))
1167 set_bit(rel_gfn, memslot->dirty_bitmap);
6aa8b732
AK		 1168 }
1169}
1170
e7d5d76c 1171int emulator_read_std(unsigned long addr,
4c690a1e 1172 void *val,
6aa8b732 1173 unsigned int bytes,
cebff02b 1174 struct kvm_vcpu *vcpu)
6aa8b732 1175{
6aa8b732
AK		 1176 void *data = val;
1177
1178 while (bytes) {
1179 gpa_t gpa = vcpu->mmu.gva_to_gpa(vcpu, addr);
1180 unsigned offset = addr & (PAGE_SIZE-1);
1181 unsigned tocopy = min(bytes, (unsigned)PAGE_SIZE - offset);
195aefde 1182 int ret;
6aa8b732
AK		 1183
1184 if (gpa == UNMAPPED_GVA)
1185 return X86EMUL_PROPAGATE_FAULT;
195aefde
IE		 1186 ret = kvm_read_guest(vcpu->kvm, gpa, data, tocopy);
1187 if (ret < 0)
6aa8b732 1188 return X86EMUL_UNHANDLEABLE;
6aa8b732
AK		 1189
1190 bytes -= tocopy;
1191 data += tocopy;
1192 addr += tocopy;
1193 }
1194
1195 return X86EMUL_CONTINUE;
1196}
e7d5d76c 1197EXPORT_SYMBOL_GPL(emulator_read_std);
6aa8b732
AK		 1198
1199static int emulator_write_std(unsigned long addr,
4c690a1e 1200 const void *val,
6aa8b732 1201 unsigned int bytes,
cebff02b 1202 struct kvm_vcpu *vcpu)
6aa8b732 1203{
f0242478 1204 pr_unimpl(vcpu, "emulator_write_std: addr %lx n %d\n", addr, bytes);
6aa8b732
AK		 1205 return X86EMUL_UNHANDLEABLE;
1206}
1207
97222cc8
ED		 1208/*
1209 * Only apic need an MMIO device hook, so shortcut now..
1210 */
1211static struct kvm_io_device *vcpu_find_pervcpu_dev(struct kvm_vcpu *vcpu,
1212 gpa_t addr)
1213{
1214 struct kvm_io_device *dev;
1215
1216 if (vcpu->apic) {
1217 dev = &vcpu->apic->dev;
1218 if (dev->in_range(dev, addr))
1219 return dev;
1220 }
1221 return NULL;
1222}
1223
2eeb2e94
GH		 1224static struct kvm_io_device *vcpu_find_mmio_dev(struct kvm_vcpu *vcpu,
1225 gpa_t addr)
1226{
97222cc8
ED		 1227 struct kvm_io_device *dev;
1228
1229 dev = vcpu_find_pervcpu_dev(vcpu, addr);
1230 if (dev == NULL)
1231 dev = kvm_io_bus_find_dev(&vcpu->kvm->mmio_bus, addr);
1232 return dev;
2eeb2e94
GH		 1233}
1234
74906345
ED		 1235static struct kvm_io_device *vcpu_find_pio_dev(struct kvm_vcpu *vcpu,
1236 gpa_t addr)
1237{
1238 return kvm_io_bus_find_dev(&vcpu->kvm->pio_bus, addr);
1239}
1240
6aa8b732 1241static int emulator_read_emulated(unsigned long addr,
4c690a1e 1242 void *val,
6aa8b732 1243 unsigned int bytes,
cebff02b 1244 struct kvm_vcpu *vcpu)
6aa8b732 1245{
2eeb2e94
GH		 1246 struct kvm_io_device *mmio_dev;
1247 gpa_t gpa;
6aa8b732
AK		 1248
1249 if (vcpu->mmio_read_completed) {
1250 memcpy(val, vcpu->mmio_data, bytes);
1251 vcpu->mmio_read_completed = 0;
1252 return X86EMUL_CONTINUE;
cebff02b 1253 } else if (emulator_read_std(addr, val, bytes, vcpu)
6aa8b732
AK		 1254 == X86EMUL_CONTINUE)
1255 return X86EMUL_CONTINUE;
d27d4aca 1256
2eeb2e94
GH		 1257 gpa = vcpu->mmu.gva_to_gpa(vcpu, addr);
1258 if (gpa == UNMAPPED_GVA)
1259 return X86EMUL_PROPAGATE_FAULT;
6aa8b732 1260
2eeb2e94
GH		 1261 /*
1262 * Is this MMIO handled locally?
1263 */
1264 mmio_dev = vcpu_find_mmio_dev(vcpu, gpa);
1265 if (mmio_dev) {
1266 kvm_iodevice_read(mmio_dev, gpa, bytes, val);
1267 return X86EMUL_CONTINUE;
6aa8b732 1268 }
2eeb2e94
GH		 1269
1270 vcpu->mmio_needed = 1;
1271 vcpu->mmio_phys_addr = gpa;
1272 vcpu->mmio_size = bytes;
1273 vcpu->mmio_is_write = 0;
1274
1275 return X86EMUL_UNHANDLEABLE;
6aa8b732
AK		 1276}
1277
da4a00f0 1278static int emulator_write_phys(struct kvm_vcpu *vcpu, gpa_t gpa,
4c690a1e 1279 const void *val, int bytes)
da4a00f0 1280{
195aefde 1281 int ret;
da4a00f0 1282
195aefde
IE		 1283 ret = kvm_write_guest(vcpu->kvm, gpa, val, bytes);
1284 if (ret < 0)
da4a00f0 1285 return 0;
fe551881 1286 kvm_mmu_pte_write(vcpu, gpa, val, bytes);
da4a00f0
AK		 1287 return 1;
1288}
1289
b0fcd903
AK		 1290static int emulator_write_emulated_onepage(unsigned long addr,
1291 const void *val,
1292 unsigned int bytes,
cebff02b 1293 struct kvm_vcpu *vcpu)
6aa8b732 1294{
2eeb2e94
GH		 1295 struct kvm_io_device *mmio_dev;
1296 gpa_t gpa = vcpu->mmu.gva_to_gpa(vcpu, addr);
6aa8b732 1297
c9047f53 1298 if (gpa == UNMAPPED_GVA) {
cbdd1bea 1299 kvm_x86_ops->inject_page_fault(vcpu, addr, 2);
6aa8b732 1300 return X86EMUL_PROPAGATE_FAULT;
c9047f53 1301 }
6aa8b732 1302
da4a00f0
AK		 1303 if (emulator_write_phys(vcpu, gpa, val, bytes))
1304 return X86EMUL_CONTINUE;
1305
2eeb2e94
GH		 1306 /*
1307 * Is this MMIO handled locally?
1308 */
1309 mmio_dev = vcpu_find_mmio_dev(vcpu, gpa);
1310 if (mmio_dev) {
1311 kvm_iodevice_write(mmio_dev, gpa, bytes, val);
1312 return X86EMUL_CONTINUE;
1313 }
1314
6aa8b732
AK		 1315 vcpu->mmio_needed = 1;
1316 vcpu->mmio_phys_addr = gpa;
1317 vcpu->mmio_size = bytes;
1318 vcpu->mmio_is_write = 1;
4c690a1e 1319 memcpy(vcpu->mmio_data, val, bytes);
6aa8b732
AK		 1320
1321 return X86EMUL_CONTINUE;
1322}
1323
e7d5d76c 1324int emulator_write_emulated(unsigned long addr,
b0fcd903
AK		 1325 const void *val,
1326 unsigned int bytes,
cebff02b 1327 struct kvm_vcpu *vcpu)
b0fcd903
AK		 1328{
1329 /* Crossing a page boundary? */
1330 if (((addr + bytes - 1) ^ addr) & PAGE_MASK) {
1331 int rc, now;
1332
1333 now = -addr & ~PAGE_MASK;
cebff02b 1334 rc = emulator_write_emulated_onepage(addr, val, now, vcpu);
b0fcd903
AK		 1335 if (rc != X86EMUL_CONTINUE)
1336 return rc;
1337 addr += now;
1338 val += now;
1339 bytes -= now;
1340 }
cebff02b 1341 return emulator_write_emulated_onepage(addr, val, bytes, vcpu);
b0fcd903 1342}
e7d5d76c 1343EXPORT_SYMBOL_GPL(emulator_write_emulated);
b0fcd903 1344
6aa8b732 1345static int emulator_cmpxchg_emulated(unsigned long addr,
4c690a1e
AK		 1346 const void *old,
1347 const void *new,
6aa8b732 1348 unsigned int bytes,
cebff02b 1349 struct kvm_vcpu *vcpu)
6aa8b732
AK		 1350{
1351 static int reported;
1352
1353 if (!reported) {
1354 reported = 1;
1355 printk(KERN_WARNING "kvm: emulating exchange as write\n");
1356 }
cebff02b 1357 return emulator_write_emulated(addr, new, bytes, vcpu);
6aa8b732
AK		 1358}
1359
1360static unsigned long get_segment_base(struct kvm_vcpu *vcpu, int seg)
1361{
cbdd1bea 1362 return kvm_x86_ops->get_segment_base(vcpu, seg);
6aa8b732
AK		 1363}
1364
1365int emulate_invlpg(struct kvm_vcpu *vcpu, gva_t address)
1366{
6aa8b732
AK		 1367 return X86EMUL_CONTINUE;
1368}
1369
1370int emulate_clts(struct kvm_vcpu *vcpu)
1371{
404fb881 1372 kvm_x86_ops->set_cr0(vcpu, vcpu->cr0 & ~X86_CR0_TS);
6aa8b732
AK		 1373 return X86EMUL_CONTINUE;
1374}
1375
d77c26fc 1376int emulator_get_dr(struct x86_emulate_ctxt *ctxt, int dr, unsigned long *dest)
6aa8b732
AK		 1377{
1378 struct kvm_vcpu *vcpu = ctxt->vcpu;
1379
1380 switch (dr) {
1381 case 0 ... 3:
cbdd1bea 1382 *dest = kvm_x86_ops->get_dr(vcpu, dr);
6aa8b732
AK		 1383 return X86EMUL_CONTINUE;
1384 default:
f0242478 1385 pr_unimpl(vcpu, "%s: unexpected dr %u\n", __FUNCTION__, dr);
6aa8b732
AK		 1386 return X86EMUL_UNHANDLEABLE;
1387 }
1388}
1389
1390int emulator_set_dr(struct x86_emulate_ctxt *ctxt, int dr, unsigned long value)
1391{
1392 unsigned long mask = (ctxt->mode == X86EMUL_MODE_PROT64) ? ~0ULL : ~0U;
1393 int exception;
1394
cbdd1bea 1395 kvm_x86_ops->set_dr(ctxt->vcpu, dr, value & mask, &exception);
6aa8b732
AK		 1396 if (exception) {
1397 /* FIXME: better handling */
1398 return X86EMUL_UNHANDLEABLE;
1399 }
1400 return X86EMUL_CONTINUE;
1401}
1402
054b1369 1403void kvm_report_emulation_failure(struct kvm_vcpu *vcpu, const char *context)
6aa8b732
AK		 1404{
1405 static int reported;
1406 u8 opcodes[4];
054b1369 1407 unsigned long rip = vcpu->rip;
6aa8b732
AK		 1408 unsigned long rip_linear;
1409
054b1369 1410 rip_linear = rip + get_segment_base(vcpu, VCPU_SREG_CS);
6aa8b732
AK		 1411
1412 if (reported)
1413 return;
1414
054b1369 1415 emulator_read_std(rip_linear, (void *)opcodes, 4, vcpu);
6aa8b732 1416
054b1369
AK		 1417 printk(KERN_ERR "emulation failed (%s) rip %lx %02x %02x %02x %02x\n",
1418 context, rip, opcodes[0], opcodes[1], opcodes[2], opcodes[3]);
6aa8b732
AK		 1419 reported = 1;
1420}
054b1369 1421EXPORT_SYMBOL_GPL(kvm_report_emulation_failure);
6aa8b732
AK		 1422
1423struct x86_emulate_ops emulate_ops = {
1424 .read_std = emulator_read_std,
1425 .write_std = emulator_write_std,
1426 .read_emulated = emulator_read_emulated,
1427 .write_emulated = emulator_write_emulated,
1428 .cmpxchg_emulated = emulator_cmpxchg_emulated,
1429};
1430
1431int emulate_instruction(struct kvm_vcpu *vcpu,
1432 struct kvm_run *run,
1433 unsigned long cr2,
3427318f
LV		 1434 u16 error_code,
1435 int no_decode)
6aa8b732 1436{
a22436b7 1437 int r;
6aa8b732 1438
e7df56e4 1439 vcpu->mmio_fault_cr2 = cr2;
cbdd1bea 1440 kvm_x86_ops->cache_regs(vcpu);
6aa8b732 1441
6aa8b732 1442 vcpu->mmio_is_write = 0;
e70669ab 1443 vcpu->pio.string = 0;
3427318f
LV		 1444
1445 if (!no_decode) {
1446 int cs_db, cs_l;
1447 kvm_x86_ops->get_cs_db_l_bits(vcpu, &cs_db, &cs_l);
1448
1449 vcpu->emulate_ctxt.vcpu = vcpu;
1450 vcpu->emulate_ctxt.eflags = kvm_x86_ops->get_rflags(vcpu);
1451 vcpu->emulate_ctxt.cr2 = cr2;
1452 vcpu->emulate_ctxt.mode =
1453 (vcpu->emulate_ctxt.eflags & X86_EFLAGS_VM)
1454 ? X86EMUL_MODE_REAL : cs_l
1455 ? X86EMUL_MODE_PROT64 : cs_db
1456 ? X86EMUL_MODE_PROT32 : X86EMUL_MODE_PROT16;
1457
1458 if (vcpu->emulate_ctxt.mode == X86EMUL_MODE_PROT64) {
1459 vcpu->emulate_ctxt.cs_base = 0;
1460 vcpu->emulate_ctxt.ds_base = 0;
1461 vcpu->emulate_ctxt.es_base = 0;
1462 vcpu->emulate_ctxt.ss_base = 0;
1463 } else {
1464 vcpu->emulate_ctxt.cs_base =
1465 get_segment_base(vcpu, VCPU_SREG_CS);
1466 vcpu->emulate_ctxt.ds_base =
1467 get_segment_base(vcpu, VCPU_SREG_DS);
1468 vcpu->emulate_ctxt.es_base =
1469 get_segment_base(vcpu, VCPU_SREG_ES);
1470 vcpu->emulate_ctxt.ss_base =
1471 get_segment_base(vcpu, VCPU_SREG_SS);
1472 }
1473
1474 vcpu->emulate_ctxt.gs_base =
1475 get_segment_base(vcpu, VCPU_SREG_GS);
1476 vcpu->emulate_ctxt.fs_base =
1477 get_segment_base(vcpu, VCPU_SREG_FS);
1478
1479 r = x86_decode_insn(&vcpu->emulate_ctxt, &emulate_ops);
a22436b7
LV		 1480 if (r) {
1481 if (kvm_mmu_unprotect_page_virt(vcpu, cr2))
1482 return EMULATE_DONE;
1483 return EMULATE_FAIL;
1484 }
3427318f
LV		 1485 }
1486
a22436b7 1487 r = x86_emulate_insn(&vcpu->emulate_ctxt, &emulate_ops);
1be3aa47 1488
e70669ab
LV		 1489 if (vcpu->pio.string)
1490 return EMULATE_DO_MMIO;
6aa8b732
AK		 1491
1492 if ((r || vcpu->mmio_is_write) && run) {
8fc0d085 1493 run->exit_reason = KVM_EXIT_MMIO;
6aa8b732
AK		 1494 run->mmio.phys_addr = vcpu->mmio_phys_addr;
1495 memcpy(run->mmio.data, vcpu->mmio_data, 8);
1496 run->mmio.len = vcpu->mmio_size;
1497 run->mmio.is_write = vcpu->mmio_is_write;
1498 }
1499
1500 if (r) {
a436036b
AK		 1501 if (kvm_mmu_unprotect_page_virt(vcpu, cr2))
1502 return EMULATE_DONE;
6aa8b732 1503 if (!vcpu->mmio_needed) {
054b1369 1504 kvm_report_emulation_failure(vcpu, "mmio");
6aa8b732
AK		 1505 return EMULATE_FAIL;
1506 }
1507 return EMULATE_DO_MMIO;
1508 }
1509
cbdd1bea 1510 kvm_x86_ops->decache_regs(vcpu);
3427318f 1511 kvm_x86_ops->set_rflags(vcpu, vcpu->emulate_ctxt.eflags);
6aa8b732 1512
02c83209
AK		 1513 if (vcpu->mmio_is_write) {
1514 vcpu->mmio_needed = 0;
6aa8b732 1515 return EMULATE_DO_MMIO;
02c83209 1516 }
6aa8b732
AK		 1517
1518 return EMULATE_DONE;
1519}
1520EXPORT_SYMBOL_GPL(emulate_instruction);
1521
b6958ce4
ED		 1522/*
1523 * The vCPU has executed a HLT instruction with in-kernel mode enabled.
1524 */
c5ec1534 1525static void kvm_vcpu_block(struct kvm_vcpu *vcpu)
d3bef15f 1526{
b6958ce4
ED		 1527 DECLARE_WAITQUEUE(wait, current);
1528
1529 add_wait_queue(&vcpu->wq, &wait);
1530
1531 /*
1532 * We will block until either an interrupt or a signal wakes us up
1533 */
c5ec1534
HQ		 1534 while (!kvm_cpu_has_interrupt(vcpu)
1535 && !signal_pending(current)
1536 && vcpu->mp_state != VCPU_MP_STATE_RUNNABLE
1537 && vcpu->mp_state != VCPU_MP_STATE_SIPI_RECEIVED) {
b6958ce4
ED		 1538 set_current_state(TASK_INTERRUPTIBLE);
1539 vcpu_put(vcpu);
1540 schedule();
1541 vcpu_load(vcpu);
1542 }
d3bef15f 1543
c5ec1534 1544 __set_current_state(TASK_RUNNING);
b6958ce4 1545 remove_wait_queue(&vcpu->wq, &wait);
b6958ce4
ED		 1546}
1547
1548int kvm_emulate_halt(struct kvm_vcpu *vcpu)
1549{
d3bef15f 1550 ++vcpu->stat.halt_exits;
b6958ce4 1551 if (irqchip_in_kernel(vcpu->kvm)) {
c5ec1534
HQ		 1552 vcpu->mp_state = VCPU_MP_STATE_HALTED;
1553 kvm_vcpu_block(vcpu);
1554 if (vcpu->mp_state != VCPU_MP_STATE_RUNNABLE)
1555 return -EINTR;
b6958ce4
ED		 1556 return 1;
1557 } else {
1558 vcpu->run->exit_reason = KVM_EXIT_HLT;
1559 return 0;
1560 }
d3bef15f
AK		 1561}
1562EXPORT_SYMBOL_GPL(kvm_emulate_halt);
1563
7aa81cc0 1564int kvm_emulate_hypercall(struct kvm_vcpu *vcpu)
270fd9b9 1565{
7aa81cc0 1566 unsigned long nr, a0, a1, a2, a3, ret;
270fd9b9 1567
cbdd1bea 1568 kvm_x86_ops->cache_regs(vcpu);
7aa81cc0
AL		 1569
1570 nr = vcpu->regs[VCPU_REGS_RAX];
1571 a0 = vcpu->regs[VCPU_REGS_RBX];
1572 a1 = vcpu->regs[VCPU_REGS_RCX];
1573 a2 = vcpu->regs[VCPU_REGS_RDX];
1574 a3 = vcpu->regs[VCPU_REGS_RSI];
1575
1576 if (!is_long_mode(vcpu)) {
1577 nr &= 0xFFFFFFFF;
1578 a0 &= 0xFFFFFFFF;
1579 a1 &= 0xFFFFFFFF;
1580 a2 &= 0xFFFFFFFF;
1581 a3 &= 0xFFFFFFFF;
270fd9b9 1582 }
7aa81cc0 1583
270fd9b9
AK		 1584 switch (nr) {
1585 default:
7aa81cc0
AL		 1586 ret = -KVM_ENOSYS;
1587 break;
270fd9b9
AK		 1588 }
1589 vcpu->regs[VCPU_REGS_RAX] = ret;
cbdd1bea 1590 kvm_x86_ops->decache_regs(vcpu);
7aa81cc0
AL		 1591 return 0;
1592}
1593EXPORT_SYMBOL_GPL(kvm_emulate_hypercall);
1594
1595int kvm_fix_hypercall(struct kvm_vcpu *vcpu)
1596{
1597 char instruction[3];
1598 int ret = 0;
1599
1600 mutex_lock(&vcpu->kvm->lock);
1601
1602 /*
1603 * Blow out the MMU to ensure that no other VCPU has an active mapping
1604 * to ensure that the updated hypercall appears atomically across all
1605 * VCPUs.
1606 */
1607 kvm_mmu_zap_all(vcpu->kvm);
1608
1609 kvm_x86_ops->cache_regs(vcpu);
1610 kvm_x86_ops->patch_hypercall(vcpu, instruction);
1611 if (emulator_write_emulated(vcpu->rip, instruction, 3, vcpu)
1612 != X86EMUL_CONTINUE)
1613 ret = -EFAULT;
1614
1615 mutex_unlock(&vcpu->kvm->lock);
1616
1617 return ret;
270fd9b9 1618}
270fd9b9 1619
6aa8b732
AK		 1620static u64 mk_cr_64(u64 curr_cr, u32 new_val)
1621{
1622 return (curr_cr & ~((1ULL << 32) - 1)) | new_val;
1623}
1624
1625void realmode_lgdt(struct kvm_vcpu *vcpu, u16 limit, unsigned long base)
1626{
1627 struct descriptor_table dt = { limit, base };
1628
cbdd1bea 1629 kvm_x86_ops->set_gdt(vcpu, &dt);
6aa8b732
AK		 1630}
1631
1632void realmode_lidt(struct kvm_vcpu *vcpu, u16 limit, unsigned long base)
1633{
1634 struct descriptor_table dt = { limit, base };
1635
cbdd1bea 1636 kvm_x86_ops->set_idt(vcpu, &dt);
6aa8b732
AK		 1637}
1638
1639void realmode_lmsw(struct kvm_vcpu *vcpu, unsigned long msw,
1640 unsigned long *rflags)
1641{
1642 lmsw(vcpu, msw);
cbdd1bea 1643 *rflags = kvm_x86_ops->get_rflags(vcpu);
6aa8b732
AK		 1644}
1645
1646unsigned long realmode_get_cr(struct kvm_vcpu *vcpu, int cr)
1647{
cbdd1bea 1648 kvm_x86_ops->decache_cr4_guest_bits(vcpu);
6aa8b732
AK		 1649 switch (cr) {
1650 case 0:
1651 return vcpu->cr0;
1652 case 2:
1653 return vcpu->cr2;
1654 case 3:
1655 return vcpu->cr3;
1656 case 4:
1657 return vcpu->cr4;
1658 default:
1659 vcpu_printf(vcpu, "%s: unexpected cr %u\n", __FUNCTION__, cr);
1660 return 0;
1661 }
1662}
1663
1664void realmode_set_cr(struct kvm_vcpu *vcpu, int cr, unsigned long val,
1665 unsigned long *rflags)
1666{
1667 switch (cr) {
1668 case 0:
1669 set_cr0(vcpu, mk_cr_64(vcpu->cr0, val));
cbdd1bea 1670 *rflags = kvm_x86_ops->get_rflags(vcpu);
6aa8b732
AK		 1671 break;
1672 case 2:
1673 vcpu->cr2 = val;
1674 break;
1675 case 3:
1676 set_cr3(vcpu, val);
1677 break;
1678 case 4:
1679 set_cr4(vcpu, mk_cr_64(vcpu->cr4, val));
1680 break;
1681 default:
1682 vcpu_printf(vcpu, "%s: unexpected cr %u\n", __FUNCTION__, cr);
1683 }
1684}
1685
3bab1f5d
AK		 1686int kvm_get_msr_common(struct kvm_vcpu *vcpu, u32 msr, u64 *pdata)
1687{
1688 u64 data;
1689
1690 switch (msr) {
1691 case 0xc0010010: /* SYSCFG */
1692 case 0xc0010015: /* HWCR */
1693 case MSR_IA32_PLATFORM_ID:
1694 case MSR_IA32_P5_MC_ADDR:
1695 case MSR_IA32_P5_MC_TYPE:
1696 case MSR_IA32_MC0_CTL:
1697 case MSR_IA32_MCG_STATUS:
1698 case MSR_IA32_MCG_CAP:
1699 case MSR_IA32_MC0_MISC:
1700 case MSR_IA32_MC0_MISC+4:
1701 case MSR_IA32_MC0_MISC+8:
1702 case MSR_IA32_MC0_MISC+12:
1703 case MSR_IA32_MC0_MISC+16:
1704 case MSR_IA32_UCODE_REV:
a8d13ea2 1705 case MSR_IA32_PERF_STATUS:
2dc7094b 1706 case MSR_IA32_EBL_CR_POWERON:
3bab1f5d
AK		 1707 /* MTRR registers */
1708 case 0xfe:
1709 case 0x200 ... 0x2ff:
1710 data = 0;
1711 break;
a8d13ea2
AK		 1712 case 0xcd: /* fsb frequency */
1713 data = 3;
1714 break;
3bab1f5d 1715 case MSR_IA32_APICBASE:
7017fc3d 1716 data = kvm_get_apic_base(vcpu);
3bab1f5d 1717 break;
6f00e68f
AK		 1718 case MSR_IA32_MISC_ENABLE:
1719 data = vcpu->ia32_misc_enable_msr;
1720 break;
3bab1f5d
AK		 1721#ifdef CONFIG_X86_64
1722 case MSR_EFER:
1723 data = vcpu->shadow_efer;
1724 break;
1725#endif
1726 default:
f0242478 1727 pr_unimpl(vcpu, "unhandled rdmsr: 0x%x\n", msr);
3bab1f5d
AK		 1728 return 1;
1729 }
1730 *pdata = data;
1731 return 0;
1732}
1733EXPORT_SYMBOL_GPL(kvm_get_msr_common);
1734
6aa8b732
AK		 1735/*
1736 * Reads an msr value (of 'msr_index') into 'pdata'.
1737 * Returns 0 on success, non-0 otherwise.
1738 * Assumes vcpu_load() was already called.
1739 */
35f3f286 1740int kvm_get_msr(struct kvm_vcpu *vcpu, u32 msr_index, u64 *pdata)
6aa8b732 1741{
cbdd1bea 1742 return kvm_x86_ops->get_msr(vcpu, msr_index, pdata);
6aa8b732
AK		 1743}
1744
05b3e0c2 1745#ifdef CONFIG_X86_64
6aa8b732 1746
3bab1f5d 1747static void set_efer(struct kvm_vcpu *vcpu, u64 efer)
6aa8b732 1748{
6aa8b732
AK		 1749 if (efer & EFER_RESERVED_BITS) {
1750 printk(KERN_DEBUG "set_efer: 0x%llx #GP, reserved bits\n",
1751 efer);
1752 inject_gp(vcpu);
1753 return;
1754 }
1755
1756 if (is_paging(vcpu)
1757 && (vcpu->shadow_efer & EFER_LME) != (efer & EFER_LME)) {
1758 printk(KERN_DEBUG "set_efer: #GP, change LME while paging\n");
1759 inject_gp(vcpu);
1760 return;
1761 }
1762
cbdd1bea 1763 kvm_x86_ops->set_efer(vcpu, efer);
7725f0ba 1764
6aa8b732
AK		 1765 efer &= ~EFER_LMA;
1766 efer |= vcpu->shadow_efer & EFER_LMA;
1767
1768 vcpu->shadow_efer = efer;
6aa8b732 1769}
6aa8b732
AK		 1770
1771#endif
1772
3bab1f5d
AK		 1773int kvm_set_msr_common(struct kvm_vcpu *vcpu, u32 msr, u64 data)
1774{
1775 switch (msr) {
1776#ifdef CONFIG_X86_64
1777 case MSR_EFER:
1778 set_efer(vcpu, data);
1779 break;
1780#endif
1781 case MSR_IA32_MC0_STATUS:
f0242478 1782 pr_unimpl(vcpu, "%s: MSR_IA32_MC0_STATUS 0x%llx, nop\n",
3bab1f5d
AK		 1783 __FUNCTION__, data);
1784 break;
0e5bf0d0 1785 case MSR_IA32_MCG_STATUS:
f0242478 1786 pr_unimpl(vcpu, "%s: MSR_IA32_MCG_STATUS 0x%llx, nop\n",
0e5bf0d0
SK		 1787 __FUNCTION__, data);
1788 break;
3bab1f5d
AK		 1789 case MSR_IA32_UCODE_REV:
1790 case MSR_IA32_UCODE_WRITE:
1791 case 0x200 ... 0x2ff: /* MTRRs */
1792 break;
1793 case MSR_IA32_APICBASE:
7017fc3d 1794 kvm_set_apic_base(vcpu, data);
3bab1f5d 1795 break;
6f00e68f
AK		 1796 case MSR_IA32_MISC_ENABLE:
1797 vcpu->ia32_misc_enable_msr = data;
1798 break;
3bab1f5d 1799 default:
f0242478 1800 pr_unimpl(vcpu, "unhandled wrmsr: 0x%x\n", msr);
3bab1f5d
AK		 1801 return 1;
1802 }
1803 return 0;
1804}
1805EXPORT_SYMBOL_GPL(kvm_set_msr_common);
1806
6aa8b732
AK		 1807/*
1808 * Writes msr value into into the appropriate "register".
1809 * Returns 0 on success, non-0 otherwise.
1810 * Assumes vcpu_load() was already called.
1811 */
35f3f286 1812int kvm_set_msr(struct kvm_vcpu *vcpu, u32 msr_index, u64 data)
6aa8b732 1813{
cbdd1bea 1814 return kvm_x86_ops->set_msr(vcpu, msr_index, data);
6aa8b732
AK		 1815}
1816
1817void kvm_resched(struct kvm_vcpu *vcpu)
1818{
3fca0365
YD		 1819 if (!need_resched())
1820 return;
6aa8b732 1821 cond_resched();
6aa8b732
AK		 1822}
1823EXPORT_SYMBOL_GPL(kvm_resched);
1824
06465c5a
AK		 1825void kvm_emulate_cpuid(struct kvm_vcpu *vcpu)
1826{
1827 int i;
1828 u32 function;
1829 struct kvm_cpuid_entry *e, *best;
1830
cbdd1bea 1831 kvm_x86_ops->cache_regs(vcpu);
06465c5a
AK		 1832 function = vcpu->regs[VCPU_REGS_RAX];
1833 vcpu->regs[VCPU_REGS_RAX] = 0;
1834 vcpu->regs[VCPU_REGS_RBX] = 0;
1835 vcpu->regs[VCPU_REGS_RCX] = 0;
1836 vcpu->regs[VCPU_REGS_RDX] = 0;
1837 best = NULL;
1838 for (i = 0; i < vcpu->cpuid_nent; ++i) {
1839 e = &vcpu->cpuid_entries[i];
1840 if (e->function == function) {
1841 best = e;
1842 break;
1843 }
1844 /*
1845 * Both basic or both extended?
1846 */
1847 if (((e->function ^ function) & 0x80000000) == 0)
1848 if (!best || e->function > best->function)
1849 best = e;
1850 }
1851 if (best) {
1852 vcpu->regs[VCPU_REGS_RAX] = best->eax;
1853 vcpu->regs[VCPU_REGS_RBX] = best->ebx;
1854 vcpu->regs[VCPU_REGS_RCX] = best->ecx;
1855 vcpu->regs[VCPU_REGS_RDX] = best->edx;
1856 }
cbdd1bea
CE		 1857 kvm_x86_ops->decache_regs(vcpu);
1858 kvm_x86_ops->skip_emulated_instruction(vcpu);
06465c5a
AK		 1859}
1860EXPORT_SYMBOL_GPL(kvm_emulate_cpuid);
1861
039576c0 1862static int pio_copy_data(struct kvm_vcpu *vcpu)
46fc1477 1863{
039576c0
AK		 1864 void *p = vcpu->pio_data;
1865 void *q;
1866 unsigned bytes;
1867 int nr_pages = vcpu->pio.guest_pages[1] ? 2 : 1;
1868
039576c0
AK		 1869 q = vmap(vcpu->pio.guest_pages, nr_pages, VM_READ|VM_WRITE,
1870 PAGE_KERNEL);
1871 if (!q) {
039576c0
AK		 1872 free_pio_guest_pages(vcpu);
1873 return -ENOMEM;
1874 }
1875 q += vcpu->pio.guest_page_offset;
1876 bytes = vcpu->pio.size * vcpu->pio.cur_count;
1877 if (vcpu->pio.in)
1878 memcpy(q, p, bytes);
1879 else
1880 memcpy(p, q, bytes);
1881 q -= vcpu->pio.guest_page_offset;
1882 vunmap(q);
039576c0
AK		 1883 free_pio_guest_pages(vcpu);
1884 return 0;
1885}
1886
1887static int complete_pio(struct kvm_vcpu *vcpu)
1888{
1889 struct kvm_pio_request *io = &vcpu->pio;
46fc1477 1890 long delta;
039576c0 1891 int r;
46fc1477 1892
cbdd1bea 1893 kvm_x86_ops->cache_regs(vcpu);
46fc1477
AK		 1894
1895 if (!io->string) {
039576c0
AK		 1896 if (io->in)
1897 memcpy(&vcpu->regs[VCPU_REGS_RAX], vcpu->pio_data,
46fc1477
AK		 1898 io->size);
1899 } else {
039576c0
AK		 1900 if (io->in) {
1901 r = pio_copy_data(vcpu);
1902 if (r) {
cbdd1bea 1903 kvm_x86_ops->cache_regs(vcpu);
039576c0
AK		 1904 return r;
1905 }
1906 }
1907
46fc1477
AK		 1908 delta = 1;
1909 if (io->rep) {
039576c0 1910 delta *= io->cur_count;
46fc1477
AK		 1911 /*
1912 * The size of the register should really depend on
1913 * current address size.
1914 */
1915 vcpu->regs[VCPU_REGS_RCX] -= delta;
1916 }
039576c0 1917 if (io->down)
46fc1477
AK		 1918 delta = -delta;
1919 delta *= io->size;
039576c0 1920 if (io->in)
46fc1477
AK		 1921 vcpu->regs[VCPU_REGS_RDI] += delta;
1922 else
1923 vcpu->regs[VCPU_REGS_RSI] += delta;
1924 }
1925
cbdd1bea 1926 kvm_x86_ops->decache_regs(vcpu);
46fc1477 1927
039576c0
AK		 1928 io->count -= io->cur_count;
1929 io->cur_count = 0;
1930
039576c0 1931 return 0;
46fc1477
AK		 1932}
1933
65619eb5
ED		 1934static void kernel_pio(struct kvm_io_device *pio_dev,
1935 struct kvm_vcpu *vcpu,
1936 void *pd)
74906345
ED		 1937{
1938 /* TODO: String I/O for in kernel device */
1939
9cf98828 1940 mutex_lock(&vcpu->kvm->lock);
74906345
ED		 1941 if (vcpu->pio.in)
1942 kvm_iodevice_read(pio_dev, vcpu->pio.port,
1943 vcpu->pio.size,
65619eb5 1944 pd);
74906345
ED		 1945 else
1946 kvm_iodevice_write(pio_dev, vcpu->pio.port,
1947 vcpu->pio.size,
65619eb5 1948 pd);
9cf98828 1949 mutex_unlock(&vcpu->kvm->lock);
65619eb5
ED		 1950}
1951
1952static void pio_string_write(struct kvm_io_device *pio_dev,
1953 struct kvm_vcpu *vcpu)
1954{
1955 struct kvm_pio_request *io = &vcpu->pio;
1956 void *pd = vcpu->pio_data;
1957 int i;
1958
9cf98828 1959 mutex_lock(&vcpu->kvm->lock);
65619eb5
ED		 1960 for (i = 0; i < io->cur_count; i++) {
1961 kvm_iodevice_write(pio_dev, io->port,
1962 io->size,
1963 pd);
1964 pd += io->size;
1965 }
9cf98828 1966 mutex_unlock(&vcpu->kvm->lock);
74906345
ED		 1967}
1968
d77c26fc 1969int kvm_emulate_pio(struct kvm_vcpu *vcpu, struct kvm_run *run, int in,
3090dd73
LV		 1970 int size, unsigned port)
1971{
1972 struct kvm_io_device *pio_dev;
1973
1974 vcpu->run->exit_reason = KVM_EXIT_IO;
1975 vcpu->run->io.direction = in ? KVM_EXIT_IO_IN : KVM_EXIT_IO_OUT;
1976 vcpu->run->io.size = vcpu->pio.size = size;
1977 vcpu->run->io.data_offset = KVM_PIO_PAGE_OFFSET * PAGE_SIZE;
1978 vcpu->run->io.count = vcpu->pio.count = vcpu->pio.cur_count = 1;
1979 vcpu->run->io.port = vcpu->pio.port = port;
1980 vcpu->pio.in = in;
1981 vcpu->pio.string = 0;
1982 vcpu->pio.down = 0;
1983 vcpu->pio.guest_page_offset = 0;
1984 vcpu->pio.rep = 0;
1985
cbdd1bea 1986 kvm_x86_ops->cache_regs(vcpu);
3090dd73 1987 memcpy(vcpu->pio_data, &vcpu->regs[VCPU_REGS_RAX], 4);
cbdd1bea 1988 kvm_x86_ops->decache_regs(vcpu);
3090dd73 1989
0967b7bf
AK		 1990 kvm_x86_ops->skip_emulated_instruction(vcpu);
1991
3090dd73
LV		 1992 pio_dev = vcpu_find_pio_dev(vcpu, port);
1993 if (pio_dev) {
1994 kernel_pio(pio_dev, vcpu, vcpu->pio_data);
1995 complete_pio(vcpu);
1996 return 1;
1997 }
1998 return 0;
1999}
2000EXPORT_SYMBOL_GPL(kvm_emulate_pio);
2001
2002int kvm_emulate_pio_string(struct kvm_vcpu *vcpu, struct kvm_run *run, int in,
2003 int size, unsigned long count, int down,
039576c0
AK		 2004 gva_t address, int rep, unsigned port)
2005{
2006 unsigned now, in_page;
65619eb5 2007 int i, ret = 0;
039576c0
AK		 2008 int nr_pages = 1;
2009 struct page *page;
74906345 2010 struct kvm_io_device *pio_dev;
039576c0
AK		 2011
2012 vcpu->run->exit_reason = KVM_EXIT_IO;
2013 vcpu->run->io.direction = in ? KVM_EXIT_IO_IN : KVM_EXIT_IO_OUT;
3090dd73 2014 vcpu->run->io.size = vcpu->pio.size = size;
039576c0 2015 vcpu->run->io.data_offset = KVM_PIO_PAGE_OFFSET * PAGE_SIZE;
3090dd73
LV		 2016 vcpu->run->io.count = vcpu->pio.count = vcpu->pio.cur_count = count;
2017 vcpu->run->io.port = vcpu->pio.port = port;
039576c0 2018 vcpu->pio.in = in;
3090dd73 2019 vcpu->pio.string = 1;
039576c0
AK		 2020 vcpu->pio.down = down;
2021 vcpu->pio.guest_page_offset = offset_in_page(address);
2022 vcpu->pio.rep = rep;
2023
039576c0 2024 if (!count) {
cbdd1bea 2025 kvm_x86_ops->skip_emulated_instruction(vcpu);
039576c0
AK		 2026 return 1;
2027 }
2028
039576c0
AK		 2029 if (!down)
2030 in_page = PAGE_SIZE - offset_in_page(address);
2031 else
2032 in_page = offset_in_page(address) + size;
2033 now = min(count, (unsigned long)in_page / size);
2034 if (!now) {
2035 /*
2036 * String I/O straddles page boundary. Pin two guest pages
2037 * so that we satisfy atomicity constraints. Do just one
2038 * transaction to avoid complexity.
2039 */
2040 nr_pages = 2;
2041 now = 1;
2042 }
2043 if (down) {
2044 /*
2045 * String I/O in reverse. Yuck. Kill the guest, fix later.
2046 */
f0242478 2047 pr_unimpl(vcpu, "guest string pio down\n");
039576c0
AK		 2048 inject_gp(vcpu);
2049 return 1;
2050 }
2051 vcpu->run->io.count = now;
2052 vcpu->pio.cur_count = now;
2053
0967b7bf
AK		 2054 if (vcpu->pio.cur_count == vcpu->pio.count)
2055 kvm_x86_ops->skip_emulated_instruction(vcpu);
2056
039576c0 2057 for (i = 0; i < nr_pages; ++i) {
11ec2804 2058 mutex_lock(&vcpu->kvm->lock);
039576c0 2059 page = gva_to_page(vcpu, address + i * PAGE_SIZE);
039576c0 2060 vcpu->pio.guest_pages[i] = page;
11ec2804 2061 mutex_unlock(&vcpu->kvm->lock);
039576c0
AK		 2062 if (!page) {
2063 inject_gp(vcpu);
2064 free_pio_guest_pages(vcpu);
2065 return 1;
2066 }
2067 }
2068
3090dd73 2069 pio_dev = vcpu_find_pio_dev(vcpu, port);
65619eb5
ED		 2070 if (!vcpu->pio.in) {
2071 /* string PIO write */
2072 ret = pio_copy_data(vcpu);
2073 if (ret >= 0 && pio_dev) {
2074 pio_string_write(pio_dev, vcpu);
2075 complete_pio(vcpu);
2076 if (vcpu->pio.count == 0)
2077 ret = 1;
2078 }
2079 } else if (pio_dev)
f0242478 2080 pr_unimpl(vcpu, "no string pio read support yet, "
65619eb5
ED		 2081 "port %x size %d count %ld\n",
2082 port, size, count);
2083
2084 return ret;
039576c0 2085}
3090dd73 2086EXPORT_SYMBOL_GPL(kvm_emulate_pio_string);
039576c0 2087
04d2cc77
AK		 2088/*
2089 * Check if userspace requested an interrupt window, and that the
2090 * interrupt window is open.
2091 *
2092 * No need to exit to userspace if we already have an interrupt queued.
2093 */
2094static int dm_request_for_irq_injection(struct kvm_vcpu *vcpu,
2095 struct kvm_run *kvm_run)
2096{
2097 return (!vcpu->irq_summary &&
2098 kvm_run->request_interrupt_window &&
2099 vcpu->interrupt_window_open &&
2100 (kvm_x86_ops->get_rflags(vcpu) & X86_EFLAGS_IF));
2101}
2102
2103static void post_kvm_run_save(struct kvm_vcpu *vcpu,
2104 struct kvm_run *kvm_run)
2105{
2106 kvm_run->if_flag = (kvm_x86_ops->get_rflags(vcpu) & X86_EFLAGS_IF) != 0;
2107 kvm_run->cr8 = get_cr8(vcpu);
2108 kvm_run->apic_base = kvm_get_apic_base(vcpu);
2109 if (irqchip_in_kernel(vcpu->kvm))
2110 kvm_run->ready_for_interrupt_injection = 1;
2111 else
2112 kvm_run->ready_for_interrupt_injection =
2113 (vcpu->interrupt_window_open &&
2114 vcpu->irq_summary == 0);
2115}
2116
2117static int __vcpu_run(struct kvm_vcpu *vcpu, struct kvm_run *kvm_run)
2118{
2119 int r;
2120
2121 if (unlikely(vcpu->mp_state == VCPU_MP_STATE_SIPI_RECEIVED)) {
d77c26fc 2122 pr_debug("vcpu %d received sipi with vector # %x\n",
04d2cc77
AK		 2123 vcpu->vcpu_id, vcpu->sipi_vector);
2124 kvm_lapic_reset(vcpu);
2125 kvm_x86_ops->vcpu_reset(vcpu);
2126 vcpu->mp_state = VCPU_MP_STATE_RUNNABLE;
2127 }
2128
2129preempted:
2130 if (vcpu->guest_debug.enabled)
2131 kvm_x86_ops->guest_debug_pre(vcpu);
2132
2133again:
2134 r = kvm_mmu_reload(vcpu);
2135 if (unlikely(r))
2136 goto out;
2137
ab6ef34b
AK		 2138 kvm_inject_pending_timer_irqs(vcpu);
2139
04d2cc77
AK		 2140 preempt_disable();
2141
2142 kvm_x86_ops->prepare_guest_switch(vcpu);
2143 kvm_load_guest_fpu(vcpu);
2144
2145 local_irq_disable();
2146
2147 if (signal_pending(current)) {
2148 local_irq_enable();
2149 preempt_enable();
2150 r = -EINTR;
2151 kvm_run->exit_reason = KVM_EXIT_INTR;
2152 ++vcpu->stat.signal_exits;
2153 goto out;
2154 }
2155
2156 if (irqchip_in_kernel(vcpu->kvm))
2157 kvm_x86_ops->inject_pending_irq(vcpu);
2158 else if (!vcpu->mmio_read_completed)
2159 kvm_x86_ops->inject_pending_vectors(vcpu, kvm_run);
2160
2161 vcpu->guest_mode = 1;
d172fcd3 2162 kvm_guest_enter();
04d2cc77
AK		 2163
2164 if (vcpu->requests)
3176bc3e 2165 if (test_and_clear_bit(KVM_REQ_TLB_FLUSH, &vcpu->requests))
04d2cc77
AK		 2166 kvm_x86_ops->tlb_flush(vcpu);
2167
2168 kvm_x86_ops->run(vcpu, kvm_run);
2169
2170 vcpu->guest_mode = 0;
2171 local_irq_enable();
2172
2173 ++vcpu->stat.exits;
2174
0552f73b
LV		 2175 /*
2176 * We must have an instruction between local_irq_enable() and
2177 * kvm_guest_exit(), so the timer interrupt isn't delayed by
2178 * the interrupt shadow. The stat.exits increment will do nicely.
2179 * But we need to prevent reordering, hence this barrier():
2180 */
2181 barrier();
2182
2183 kvm_guest_exit();
2184
04d2cc77
AK		 2185 preempt_enable();
2186
2187 /*
2188 * Profile KVM exit RIPs:
2189 */
2190 if (unlikely(prof_on == KVM_PROFILING)) {
2191 kvm_x86_ops->cache_regs(vcpu);
2192 profile_hit(KVM_PROFILING, (void *)vcpu->rip);
2193 }
2194
2195 r = kvm_x86_ops->handle_exit(kvm_run, vcpu);
2196
2197 if (r > 0) {
2198 if (dm_request_for_irq_injection(vcpu, kvm_run)) {
2199 r = -EINTR;
2200 kvm_run->exit_reason = KVM_EXIT_INTR;
2201 ++vcpu->stat.request_irq_exits;
2202 goto out;
2203 }
2204 if (!need_resched()) {
2205 ++vcpu->stat.light_exits;
2206 goto again;
2207 }
2208 }
2209
2210out:
2211 if (r > 0) {
2212 kvm_resched(vcpu);
2213 goto preempted;
2214 }
2215
2216 post_kvm_run_save(vcpu, kvm_run);
2217
2218 return r;
2219}
2220
2221
bccf2150 2222static int kvm_vcpu_ioctl_run(struct kvm_vcpu *vcpu, struct kvm_run *kvm_run)
6aa8b732 2223{
6aa8b732 2224 int r;
1961d276 2225 sigset_t sigsaved;
6aa8b732 2226
bccf2150 2227 vcpu_load(vcpu);
6aa8b732 2228
c5ec1534
HQ		 2229 if (unlikely(vcpu->mp_state == VCPU_MP_STATE_UNINITIALIZED)) {
2230 kvm_vcpu_block(vcpu);
2231 vcpu_put(vcpu);
2232 return -EAGAIN;
2233 }
2234
1961d276
AK		 2235 if (vcpu->sigset_active)
2236 sigprocmask(SIG_SETMASK, &vcpu->sigset, &sigsaved);
2237
54810342 2238 /* re-sync apic's tpr */
5cd4f6fd
HQ		 2239 if (!irqchip_in_kernel(vcpu->kvm))
2240 set_cr8(vcpu, kvm_run->cr8);
54810342 2241
02c83209
AK		 2242 if (vcpu->pio.cur_count) {
2243 r = complete_pio(vcpu);
2244 if (r)
2245 goto out;
2246 }
34c16eec 2247#if CONFIG_HAS_IOMEM
02c83209
AK		 2248 if (vcpu->mmio_needed) {
2249 memcpy(vcpu->mmio_data, kvm_run->mmio.data, 8);
2250 vcpu->mmio_read_completed = 1;
2251 vcpu->mmio_needed = 0;
2252 r = emulate_instruction(vcpu, kvm_run,
3427318f 2253 vcpu->mmio_fault_cr2, 0, 1);
02c83209
AK		 2254 if (r == EMULATE_DO_MMIO) {
2255 /*
2256 * Read-modify-write. Back to userspace.
2257 */
02c83209
AK		 2258 r = 0;
2259 goto out;
46fc1477 2260 }
6aa8b732 2261 }
34c16eec 2262#endif
8eb7d334 2263 if (kvm_run->exit_reason == KVM_EXIT_HYPERCALL) {
cbdd1bea 2264 kvm_x86_ops->cache_regs(vcpu);
b4e63f56 2265 vcpu->regs[VCPU_REGS_RAX] = kvm_run->hypercall.ret;
cbdd1bea 2266 kvm_x86_ops->decache_regs(vcpu);
b4e63f56
AK		 2267 }
2268
04d2cc77 2269 r = __vcpu_run(vcpu, kvm_run);
6aa8b732 2270
039576c0 2271out:
1961d276
AK		 2272 if (vcpu->sigset_active)
2273 sigprocmask(SIG_SETMASK, &sigsaved, NULL);
2274
6aa8b732
AK		 2275 vcpu_put(vcpu);
2276 return r;
2277}
2278
bccf2150
AK		 2279static int kvm_vcpu_ioctl_get_regs(struct kvm_vcpu *vcpu,
2280 struct kvm_regs *regs)
6aa8b732 2281{
bccf2150 2282 vcpu_load(vcpu);
6aa8b732 2283
cbdd1bea 2284 kvm_x86_ops->cache_regs(vcpu);
6aa8b732
AK		 2285
2286 regs->rax = vcpu->regs[VCPU_REGS_RAX];
2287 regs->rbx = vcpu->regs[VCPU_REGS_RBX];
2288 regs->rcx = vcpu->regs[VCPU_REGS_RCX];
2289 regs->rdx = vcpu->regs[VCPU_REGS_RDX];
2290 regs->rsi = vcpu->regs[VCPU_REGS_RSI];
2291 regs->rdi = vcpu->regs[VCPU_REGS_RDI];
2292 regs->rsp = vcpu->regs[VCPU_REGS_RSP];
2293 regs->rbp = vcpu->regs[VCPU_REGS_RBP];
05b3e0c2 2294#ifdef CONFIG_X86_64
6aa8b732
AK		 2295 regs->r8 = vcpu->regs[VCPU_REGS_R8];
2296 regs->r9 = vcpu->regs[VCPU_REGS_R9];
2297 regs->r10 = vcpu->regs[VCPU_REGS_R10];
2298 regs->r11 = vcpu->regs[VCPU_REGS_R11];
2299 regs->r12 = vcpu->regs[VCPU_REGS_R12];
2300 regs->r13 = vcpu->regs[VCPU_REGS_R13];
2301 regs->r14 = vcpu->regs[VCPU_REGS_R14];
2302 regs->r15 = vcpu->regs[VCPU_REGS_R15];
2303#endif
2304
2305 regs->rip = vcpu->rip;
cbdd1bea 2306 regs->rflags = kvm_x86_ops->get_rflags(vcpu);
6aa8b732
AK		 2307
2308 /*
2309 * Don't leak debug flags in case they were set for guest debugging
2310 */
2311 if (vcpu->guest_debug.enabled && vcpu->guest_debug.singlestep)
2312 regs->rflags &= ~(X86_EFLAGS_TF | X86_EFLAGS_RF);
2313
2314 vcpu_put(vcpu);
2315
2316 return 0;
2317}
2318
bccf2150
AK		 2319static int kvm_vcpu_ioctl_set_regs(struct kvm_vcpu *vcpu,
2320 struct kvm_regs *regs)
6aa8b732 2321{
bccf2150 2322 vcpu_load(vcpu);
6aa8b732
AK		 2323
2324 vcpu->regs[VCPU_REGS_RAX] = regs->rax;
2325 vcpu->regs[VCPU_REGS_RBX] = regs->rbx;
2326 vcpu->regs[VCPU_REGS_RCX] = regs->rcx;
2327 vcpu->regs[VCPU_REGS_RDX] = regs->rdx;
2328 vcpu->regs[VCPU_REGS_RSI] = regs->rsi;
2329 vcpu->regs[VCPU_REGS_RDI] = regs->rdi;
2330 vcpu->regs[VCPU_REGS_RSP] = regs->rsp;
2331 vcpu->regs[VCPU_REGS_RBP] = regs->rbp;
05b3e0c2 2332#ifdef CONFIG_X86_64
6aa8b732
AK		 2333 vcpu->regs[VCPU_REGS_R8] = regs->r8;
2334 vcpu->regs[VCPU_REGS_R9] = regs->r9;
2335 vcpu->regs[VCPU_REGS_R10] = regs->r10;
2336 vcpu->regs[VCPU_REGS_R11] = regs->r11;
2337 vcpu->regs[VCPU_REGS_R12] = regs->r12;
2338 vcpu->regs[VCPU_REGS_R13] = regs->r13;
2339 vcpu->regs[VCPU_REGS_R14] = regs->r14;
2340 vcpu->regs[VCPU_REGS_R15] = regs->r15;
2341#endif
2342
2343 vcpu->rip = regs->rip;
cbdd1bea 2344 kvm_x86_ops->set_rflags(vcpu, regs->rflags);
6aa8b732 2345
cbdd1bea 2346 kvm_x86_ops->decache_regs(vcpu);
6aa8b732
AK		 2347
2348 vcpu_put(vcpu);
2349
2350 return 0;
2351}
2352
2353static void get_segment(struct kvm_vcpu *vcpu,
2354 struct kvm_segment *var, int seg)
2355{
cbdd1bea 2356 return kvm_x86_ops->get_segment(vcpu, var, seg);
6aa8b732
AK		 2357}
2358
bccf2150
AK		 2359static int kvm_vcpu_ioctl_get_sregs(struct kvm_vcpu *vcpu,
2360 struct kvm_sregs *sregs)
6aa8b732 2361{
6aa8b732 2362 struct descriptor_table dt;
2a8067f1 2363 int pending_vec;
6aa8b732 2364
bccf2150 2365 vcpu_load(vcpu);
6aa8b732
AK		 2366
2367 get_segment(vcpu, &sregs->cs, VCPU_SREG_CS);
2368 get_segment(vcpu, &sregs->ds, VCPU_SREG_DS);
2369 get_segment(vcpu, &sregs->es, VCPU_SREG_ES);
2370 get_segment(vcpu, &sregs->fs, VCPU_SREG_FS);
2371 get_segment(vcpu, &sregs->gs, VCPU_SREG_GS);
2372 get_segment(vcpu, &sregs->ss, VCPU_SREG_SS);
2373
2374 get_segment(vcpu, &sregs->tr, VCPU_SREG_TR);
2375 get_segment(vcpu, &sregs->ldt, VCPU_SREG_LDTR);
2376
cbdd1bea 2377 kvm_x86_ops->get_idt(vcpu, &dt);
6aa8b732
AK		 2378 sregs->idt.limit = dt.limit;
2379 sregs->idt.base = dt.base;
cbdd1bea 2380 kvm_x86_ops->get_gdt(vcpu, &dt);
6aa8b732
AK		 2381 sregs->gdt.limit = dt.limit;
2382 sregs->gdt.base = dt.base;
2383
cbdd1bea 2384 kvm_x86_ops->decache_cr4_guest_bits(vcpu);
6aa8b732
AK		 2385 sregs->cr0 = vcpu->cr0;
2386 sregs->cr2 = vcpu->cr2;
2387 sregs->cr3 = vcpu->cr3;
2388 sregs->cr4 = vcpu->cr4;
7017fc3d 2389 sregs->cr8 = get_cr8(vcpu);
6aa8b732 2390 sregs->efer = vcpu->shadow_efer;
7017fc3d 2391 sregs->apic_base = kvm_get_apic_base(vcpu);
6aa8b732 2392
2a8067f1 2393 if (irqchip_in_kernel(vcpu->kvm)) {
c52fb35a
HQ		 2394 memset(sregs->interrupt_bitmap, 0,
2395 sizeof sregs->interrupt_bitmap);
cbdd1bea 2396 pending_vec = kvm_x86_ops->get_irq(vcpu);
2a8067f1 2397 if (pending_vec >= 0)
d77c26fc
MD		 2398 set_bit(pending_vec,
2399 (unsigned long *)sregs->interrupt_bitmap);
2a8067f1 2400 } else
c52fb35a
HQ		 2401 memcpy(sregs->interrupt_bitmap, vcpu->irq_pending,
2402 sizeof sregs->interrupt_bitmap);
6aa8b732
AK		 2403
2404 vcpu_put(vcpu);
2405
2406 return 0;
2407}
2408
2409static void set_segment(struct kvm_vcpu *vcpu,
2410 struct kvm_segment *var, int seg)
2411{
cbdd1bea 2412 return kvm_x86_ops->set_segment(vcpu, var, seg);
6aa8b732
AK		 2413}
2414
bccf2150
AK		 2415static int kvm_vcpu_ioctl_set_sregs(struct kvm_vcpu *vcpu,
2416 struct kvm_sregs *sregs)
6aa8b732 2417{
6aa8b732 2418 int mmu_reset_needed = 0;
2a8067f1 2419 int i, pending_vec, max_bits;
6aa8b732
AK		 2420 struct descriptor_table dt;
2421
bccf2150 2422 vcpu_load(vcpu);
6aa8b732 2423
6aa8b732
AK		 2424 dt.limit = sregs->idt.limit;
2425 dt.base = sregs->idt.base;
cbdd1bea 2426 kvm_x86_ops->set_idt(vcpu, &dt);
6aa8b732
AK		 2427 dt.limit = sregs->gdt.limit;
2428 dt.base = sregs->gdt.base;
cbdd1bea 2429 kvm_x86_ops->set_gdt(vcpu, &dt);
6aa8b732
AK		 2430
2431 vcpu->cr2 = sregs->cr2;
2432 mmu_reset_needed |= vcpu->cr3 != sregs->cr3;
2433 vcpu->cr3 = sregs->cr3;
2434
7017fc3d 2435 set_cr8(vcpu, sregs->cr8);
6aa8b732
AK		 2436
2437 mmu_reset_needed |= vcpu->shadow_efer != sregs->efer;
05b3e0c2 2438#ifdef CONFIG_X86_64
cbdd1bea 2439 kvm_x86_ops->set_efer(vcpu, sregs->efer);
6aa8b732 2440#endif
7017fc3d 2441 kvm_set_apic_base(vcpu, sregs->apic_base);
6aa8b732 2442
cbdd1bea 2443 kvm_x86_ops->decache_cr4_guest_bits(vcpu);
399badf3 2444
6aa8b732 2445 mmu_reset_needed |= vcpu->cr0 != sregs->cr0;
81f50e3b 2446 vcpu->cr0 = sregs->cr0;
cbdd1bea 2447 kvm_x86_ops->set_cr0(vcpu, sregs->cr0);
6aa8b732
AK		 2448
2449 mmu_reset_needed |= vcpu->cr4 != sregs->cr4;
cbdd1bea 2450 kvm_x86_ops->set_cr4(vcpu, sregs->cr4);
1b0973bd
AK		 2451 if (!is_long_mode(vcpu) && is_pae(vcpu))
2452 load_pdptrs(vcpu, vcpu->cr3);
6aa8b732
AK		 2453
2454 if (mmu_reset_needed)
2455 kvm_mmu_reset_context(vcpu);
2456
c52fb35a
HQ		 2457 if (!irqchip_in_kernel(vcpu->kvm)) {
2458 memcpy(vcpu->irq_pending, sregs->interrupt_bitmap,
2459 sizeof vcpu->irq_pending);
2460 vcpu->irq_summary = 0;
2461 for (i = 0; i < ARRAY_SIZE(vcpu->irq_pending); ++i)
2462 if (vcpu->irq_pending[i])
2463 __set_bit(i, &vcpu->irq_summary);
2a8067f1
ED		 2464 } else {
2465 max_bits = (sizeof sregs->interrupt_bitmap) << 3;
2466 pending_vec = find_first_bit(
2467 (const unsigned long *)sregs->interrupt_bitmap,
2468 max_bits);
2469 /* Only pending external irq is handled here */
2470 if (pending_vec < max_bits) {
cbdd1bea 2471 kvm_x86_ops->set_irq(vcpu, pending_vec);
d77c26fc
MD		 2472 pr_debug("Set back pending irq %d\n",
2473 pending_vec);
2a8067f1 2474 }
c52fb35a 2475 }
6aa8b732 2476
024aa1c0
AK		 2477 set_segment(vcpu, &sregs->cs, VCPU_SREG_CS);
2478 set_segment(vcpu, &sregs->ds, VCPU_SREG_DS);
2479 set_segment(vcpu, &sregs->es, VCPU_SREG_ES);
2480 set_segment(vcpu, &sregs->fs, VCPU_SREG_FS);
2481 set_segment(vcpu, &sregs->gs, VCPU_SREG_GS);
2482 set_segment(vcpu, &sregs->ss, VCPU_SREG_SS);
2483
2484 set_segment(vcpu, &sregs->tr, VCPU_SREG_TR);
2485 set_segment(vcpu, &sregs->ldt, VCPU_SREG_LDTR);
2486
6aa8b732
AK		 2487 vcpu_put(vcpu);
2488
2489 return 0;
2490}
2491
1747fb71
RR		 2492void kvm_get_cs_db_l_bits(struct kvm_vcpu *vcpu, int *db, int *l)
2493{
2494 struct kvm_segment cs;
2495
2496 get_segment(vcpu, &cs, VCPU_SREG_CS);
2497 *db = cs.db;
2498 *l = cs.l;
2499}
2500EXPORT_SYMBOL_GPL(kvm_get_cs_db_l_bits);
2501
6aa8b732
AK		 2502/*
2503 * Translate a guest virtual address to a guest physical address.
2504 */
bccf2150
AK		 2505static int kvm_vcpu_ioctl_translate(struct kvm_vcpu *vcpu,
2506 struct kvm_translation *tr)
6aa8b732
AK		 2507{
2508 unsigned long vaddr = tr->linear_address;
6aa8b732
AK		 2509 gpa_t gpa;
2510
bccf2150 2511 vcpu_load(vcpu);
11ec2804 2512 mutex_lock(&vcpu->kvm->lock);
6aa8b732
AK		 2513 gpa = vcpu->mmu.gva_to_gpa(vcpu, vaddr);
2514 tr->physical_address = gpa;
2515 tr->valid = gpa != UNMAPPED_GVA;
2516 tr->writeable = 1;
2517 tr->usermode = 0;
11ec2804 2518 mutex_unlock(&vcpu->kvm->lock);
6aa8b732
AK		 2519 vcpu_put(vcpu);
2520
2521 return 0;
2522}
2523
bccf2150
AK		 2524static int kvm_vcpu_ioctl_interrupt(struct kvm_vcpu *vcpu,
2525 struct kvm_interrupt *irq)
6aa8b732 2526{
6aa8b732
AK		 2527 if (irq->irq < 0 || irq->irq >= 256)
2528 return -EINVAL;
97222cc8
ED		 2529 if (irqchip_in_kernel(vcpu->kvm))
2530 return -ENXIO;
bccf2150 2531 vcpu_load(vcpu);
6aa8b732
AK		 2532
2533 set_bit(irq->irq, vcpu->irq_pending);
2534 set_bit(irq->irq / BITS_PER_LONG, &vcpu->irq_summary);
2535
2536 vcpu_put(vcpu);
2537
2538 return 0;
2539}
2540
bccf2150
AK		 2541static int kvm_vcpu_ioctl_debug_guest(struct kvm_vcpu *vcpu,
2542 struct kvm_debug_guest *dbg)
6aa8b732 2543{
6aa8b732
AK		 2544 int r;
2545
bccf2150 2546 vcpu_load(vcpu);
6aa8b732 2547
cbdd1bea 2548 r = kvm_x86_ops->set_guest_debug(vcpu, dbg);
6aa8b732
AK		 2549
2550 vcpu_put(vcpu);
2551
2552 return r;
2553}
2554
9a2bb7f4
AK		 2555static struct page *kvm_vcpu_nopage(struct vm_area_struct *vma,
2556 unsigned long address,
2557 int *type)
2558{
2559 struct kvm_vcpu *vcpu = vma->vm_file->private_data;
2560 unsigned long pgoff;
2561 struct page *page;
2562
9a2bb7f4 2563 pgoff = ((address - vma->vm_start) >> PAGE_SHIFT) + vma->vm_pgoff;
039576c0
AK		 2564 if (pgoff == 0)
2565 page = virt_to_page(vcpu->run);
2566 else if (pgoff == KVM_PIO_PAGE_OFFSET)
2567 page = virt_to_page(vcpu->pio_data);
2568 else
9a2bb7f4 2569 return NOPAGE_SIGBUS;
9a2bb7f4 2570 get_page(page);
cd0d9137
NAQ		 2571 if (type != NULL)
2572 *type = VM_FAULT_MINOR;
2573
9a2bb7f4
AK		 2574 return page;
2575}
2576
2577static struct vm_operations_struct kvm_vcpu_vm_ops = {
2578 .nopage = kvm_vcpu_nopage,
2579};
2580
2581static int kvm_vcpu_mmap(struct file *file, struct vm_area_struct *vma)
2582{
2583 vma->vm_ops = &kvm_vcpu_vm_ops;
2584 return 0;
2585}
2586
bccf2150
AK		 2587static int kvm_vcpu_release(struct inode *inode, struct file *filp)
2588{
2589 struct kvm_vcpu *vcpu = filp->private_data;
2590
2591 fput(vcpu->kvm->filp);
2592 return 0;
2593}
2594
2595static struct file_operations kvm_vcpu_fops = {
2596 .release = kvm_vcpu_release,
2597 .unlocked_ioctl = kvm_vcpu_ioctl,
2598 .compat_ioctl = kvm_vcpu_ioctl,
9a2bb7f4 2599 .mmap = kvm_vcpu_mmap,
bccf2150
AK		 2600};
2601
2602/*
2603 * Allocates an inode for the vcpu.
2604 */
2605static int create_vcpu_fd(struct kvm_vcpu *vcpu)
2606{
2607 int fd, r;
2608 struct inode *inode;
2609 struct file *file;
2610
d6d28168
AK		 2611 r = anon_inode_getfd(&fd, &inode, &file,
2612 "kvm-vcpu", &kvm_vcpu_fops, vcpu);
2613 if (r)
2614 return r;
bccf2150 2615 atomic_inc(&vcpu->kvm->filp->f_count);
bccf2150 2616 return fd;
bccf2150
AK		 2617}
2618
c5ea7660
AK		 2619/*
2620 * Creates some virtual cpus. Good luck creating more than one.
2621 */
2622static int kvm_vm_ioctl_create_vcpu(struct kvm *kvm, int n)
2623{
2624 int r;
2625 struct kvm_vcpu *vcpu;
2626
c5ea7660 2627 if (!valid_vcpu(n))
fb3f0f51 2628 return -EINVAL;
c5ea7660 2629
cbdd1bea 2630 vcpu = kvm_x86_ops->vcpu_create(kvm, n);
fb3f0f51
RR		 2631 if (IS_ERR(vcpu))
2632 return PTR_ERR(vcpu);
c5ea7660 2633
15ad7146
AK		 2634 preempt_notifier_init(&vcpu->preempt_notifier, &kvm_preempt_ops);
2635
b114b080
RR		 2636 /* We do fxsave: this must be aligned. */
2637 BUG_ON((unsigned long)&vcpu->host_fx_image & 0xF);
2638
fb3f0f51 2639 vcpu_load(vcpu);
c5ea7660 2640 r = kvm_mmu_setup(vcpu);
c5ea7660 2641 vcpu_put(vcpu);
c5ea7660 2642 if (r < 0)
fb3f0f51
RR		 2643 goto free_vcpu;
2644
11ec2804 2645 mutex_lock(&kvm->lock);
fb3f0f51
RR		 2646 if (kvm->vcpus[n]) {
2647 r = -EEXIST;
11ec2804 2648 mutex_unlock(&kvm->lock);
fb3f0f51
RR		 2649 goto mmu_unload;
2650 }
2651 kvm->vcpus[n] = vcpu;
11ec2804 2652 mutex_unlock(&kvm->lock);
c5ea7660 2653
fb3f0f51 2654 /* Now it's all set up, let userspace reach it */
bccf2150
AK		 2655 r = create_vcpu_fd(vcpu);
2656 if (r < 0)
fb3f0f51
RR		 2657 goto unlink;
2658 return r;
39c3b86e 2659
fb3f0f51 2660unlink:
11ec2804 2661 mutex_lock(&kvm->lock);
fb3f0f51 2662 kvm->vcpus[n] = NULL;
11ec2804 2663 mutex_unlock(&kvm->lock);
a2fa3e9f 2664
fb3f0f51
RR		 2665mmu_unload:
2666 vcpu_load(vcpu);
2667 kvm_mmu_unload(vcpu);
2668 vcpu_put(vcpu);
c5ea7660 2669
fb3f0f51 2670free_vcpu:
cbdd1bea 2671 kvm_x86_ops->vcpu_free(vcpu);
c5ea7660
AK		 2672 return r;
2673}
2674
1961d276
AK		 2675static int kvm_vcpu_ioctl_set_sigmask(struct kvm_vcpu *vcpu, sigset_t *sigset)
2676{
2677 if (sigset) {
2678 sigdelsetmask(sigset, sigmask(SIGKILL)|sigmask(SIGSTOP));
2679 vcpu->sigset_active = 1;
2680 vcpu->sigset = *sigset;
2681 } else
2682 vcpu->sigset_active = 0;
2683 return 0;
2684}
2685
b8836737
AK		 2686/*
2687 * fxsave fpu state. Taken from x86_64/processor.h. To be killed when
2688 * we have asm/x86/processor.h
2689 */
2690struct fxsave {
2691 u16 cwd;
2692 u16 swd;
2693 u16 twd;
2694 u16 fop;
2695 u64 rip;
2696 u64 rdp;
2697 u32 mxcsr;
2698 u32 mxcsr_mask;
2699 u32 st_space[32]; /* 8*16 bytes for each FP-reg = 128 bytes */
2700#ifdef CONFIG_X86_64
2701 u32 xmm_space[64]; /* 16*16 bytes for each XMM-reg = 256 bytes */
2702#else
2703 u32 xmm_space[32]; /* 8*16 bytes for each XMM-reg = 128 bytes */
2704#endif
2705};
2706
2707static int kvm_vcpu_ioctl_get_fpu(struct kvm_vcpu *vcpu, struct kvm_fpu *fpu)
2708{
b114b080 2709 struct fxsave *fxsave = (struct fxsave *)&vcpu->guest_fx_image;
b8836737
AK		 2710
2711 vcpu_load(vcpu);
2712
2713 memcpy(fpu->fpr, fxsave->st_space, 128);
2714 fpu->fcw = fxsave->cwd;
2715 fpu->fsw = fxsave->swd;
2716 fpu->ftwx = fxsave->twd;
2717 fpu->last_opcode = fxsave->fop;
2718 fpu->last_ip = fxsave->rip;
2719 fpu->last_dp = fxsave->rdp;
2720 memcpy(fpu->xmm, fxsave->xmm_space, sizeof fxsave->xmm_space);
2721
2722 vcpu_put(vcpu);
2723
2724 return 0;
2725}
2726
2727static int kvm_vcpu_ioctl_set_fpu(struct kvm_vcpu *vcpu, struct kvm_fpu *fpu)
2728{
b114b080 2729 struct fxsave *fxsave = (struct fxsave *)&vcpu->guest_fx_image;
b8836737
AK		 2730
2731 vcpu_load(vcpu);
2732
2733 memcpy(fxsave->st_space, fpu->fpr, 128);
2734 fxsave->cwd = fpu->fcw;
2735 fxsave->swd = fpu->fsw;
2736 fxsave->twd = fpu->ftwx;
2737 fxsave->fop = fpu->last_opcode;
2738 fxsave->rip = fpu->last_ip;
2739 fxsave->rdp = fpu->last_dp;
2740 memcpy(fxsave->xmm_space, fpu->xmm, sizeof fxsave->xmm_space);
2741
2742 vcpu_put(vcpu);
2743
2744 return 0;
2745}
2746
bccf2150
AK		 2747static long kvm_vcpu_ioctl(struct file *filp,
2748 unsigned int ioctl, unsigned long arg)
6aa8b732 2749{
bccf2150 2750 struct kvm_vcpu *vcpu = filp->private_data;
2f366987 2751 void __user *argp = (void __user *)arg;
313a3dc7 2752 int r;
6aa8b732
AK		 2753
2754 switch (ioctl) {
9a2bb7f4 2755 case KVM_RUN:
f0fe5108
AK		 2756 r = -EINVAL;
2757 if (arg)
2758 goto out;
9a2bb7f4 2759 r = kvm_vcpu_ioctl_run(vcpu, vcpu->run);
6aa8b732 2760 break;
6aa8b732
AK		 2761 case KVM_GET_REGS: {
2762 struct kvm_regs kvm_regs;
2763
bccf2150
AK		 2764 memset(&kvm_regs, 0, sizeof kvm_regs);
2765 r = kvm_vcpu_ioctl_get_regs(vcpu, &kvm_regs);
6aa8b732
AK		 2766 if (r)
2767 goto out;
2768 r = -EFAULT;
2f366987 2769 if (copy_to_user(argp, &kvm_regs, sizeof kvm_regs))
6aa8b732
AK		 2770 goto out;
2771 r = 0;
2772 break;
2773 }
2774 case KVM_SET_REGS: {
2775 struct kvm_regs kvm_regs;
2776
2777 r = -EFAULT;
2f366987 2778 if (copy_from_user(&kvm_regs, argp, sizeof kvm_regs))
6aa8b732 2779 goto out;
bccf2150 2780 r = kvm_vcpu_ioctl_set_regs(vcpu, &kvm_regs);
6aa8b732
AK		 2781 if (r)
2782 goto out;
2783 r = 0;
2784 break;
2785 }
2786 case KVM_GET_SREGS: {
2787 struct kvm_sregs kvm_sregs;
2788
bccf2150
AK		 2789 memset(&kvm_sregs, 0, sizeof kvm_sregs);
2790 r = kvm_vcpu_ioctl_get_sregs(vcpu, &kvm_sregs);
6aa8b732
AK		 2791 if (r)
2792 goto out;
2793 r = -EFAULT;
2f366987 2794 if (copy_to_user(argp, &kvm_sregs, sizeof kvm_sregs))
6aa8b732
AK		 2795 goto out;
2796 r = 0;
2797 break;
2798 }
2799 case KVM_SET_SREGS: {
2800 struct kvm_sregs kvm_sregs;
2801
2802 r = -EFAULT;
2f366987 2803 if (copy_from_user(&kvm_sregs, argp, sizeof kvm_sregs))
6aa8b732 2804 goto out;
bccf2150 2805 r = kvm_vcpu_ioctl_set_sregs(vcpu, &kvm_sregs);
6aa8b732
AK		 2806 if (r)
2807 goto out;
2808 r = 0;
2809 break;
2810 }
2811 case KVM_TRANSLATE: {
2812 struct kvm_translation tr;
2813
2814 r = -EFAULT;
2f366987 2815 if (copy_from_user(&tr, argp, sizeof tr))
6aa8b732 2816 goto out;
bccf2150 2817 r = kvm_vcpu_ioctl_translate(vcpu, &tr);
6aa8b732
AK		 2818 if (r)
2819 goto out;
2820 r = -EFAULT;
2f366987 2821 if (copy_to_user(argp, &tr, sizeof tr))
6aa8b732
AK		 2822 goto out;
2823 r = 0;
2824 break;
2825 }
2826 case KVM_INTERRUPT: {
2827 struct kvm_interrupt irq;
2828
2829 r = -EFAULT;
2f366987 2830 if (copy_from_user(&irq, argp, sizeof irq))
6aa8b732 2831 goto out;
bccf2150 2832 r = kvm_vcpu_ioctl_interrupt(vcpu, &irq);
6aa8b732
AK		 2833 if (r)
2834 goto out;
2835 r = 0;
2836 break;
2837 }
2838 case KVM_DEBUG_GUEST: {
2839 struct kvm_debug_guest dbg;
2840
2841 r = -EFAULT;
2f366987 2842 if (copy_from_user(&dbg, argp, sizeof dbg))
6aa8b732 2843 goto out;
bccf2150 2844 r = kvm_vcpu_ioctl_debug_guest(vcpu, &dbg);
6aa8b732
AK		 2845 if (r)
2846 goto out;
2847 r = 0;
2848 break;
2849 }
1961d276
AK		 2850 case KVM_SET_SIGNAL_MASK: {
2851 struct kvm_signal_mask __user *sigmask_arg = argp;
2852 struct kvm_signal_mask kvm_sigmask;
2853 sigset_t sigset, *p;
2854
2855 p = NULL;
2856 if (argp) {
2857 r = -EFAULT;
2858 if (copy_from_user(&kvm_sigmask, argp,
2859 sizeof kvm_sigmask))
2860 goto out;
2861 r = -EINVAL;
2862 if (kvm_sigmask.len != sizeof sigset)
2863 goto out;
2864 r = -EFAULT;
2865 if (copy_from_user(&sigset, sigmask_arg->sigset,
2866 sizeof sigset))
2867 goto out;
2868 p = &sigset;
2869 }
2870 r = kvm_vcpu_ioctl_set_sigmask(vcpu, &sigset);
2871 break;
2872 }
b8836737
AK		 2873 case KVM_GET_FPU: {
2874 struct kvm_fpu fpu;
2875
2876 memset(&fpu, 0, sizeof fpu);
2877 r = kvm_vcpu_ioctl_get_fpu(vcpu, &fpu);
2878 if (r)
2879 goto out;
2880 r = -EFAULT;
2881 if (copy_to_user(argp, &fpu, sizeof fpu))
2882 goto out;
2883 r = 0;
2884 break;
2885 }
2886 case KVM_SET_FPU: {
2887 struct kvm_fpu fpu;
2888
2889 r = -EFAULT;
2890 if (copy_from_user(&fpu, argp, sizeof fpu))
2891 goto out;
2892 r = kvm_vcpu_ioctl_set_fpu(vcpu, &fpu);
2893 if (r)
2894 goto out;
2895 r = 0;
2896 break;
2897 }
bccf2150 2898 default:
313a3dc7 2899 r = kvm_arch_vcpu_ioctl(filp, ioctl, arg);
bccf2150
AK		 2900 }
2901out:
2902 return r;
2903}
2904
2905static long kvm_vm_ioctl(struct file *filp,
2906 unsigned int ioctl, unsigned long arg)
2907{
2908 struct kvm *kvm = filp->private_data;
2909 void __user *argp = (void __user *)arg;
2910 int r = -EINVAL;
2911
2912 switch (ioctl) {
2913 case KVM_CREATE_VCPU:
2914 r = kvm_vm_ioctl_create_vcpu(kvm, arg);
2915 if (r < 0)
2916 goto out;
2917 break;
6aa8b732
AK		 2918 case KVM_SET_MEMORY_REGION: {
2919 struct kvm_memory_region kvm_mem;
6fc138d2 2920 struct kvm_userspace_memory_region kvm_userspace_mem;
6aa8b732
AK		 2921
2922 r = -EFAULT;
2f366987 2923 if (copy_from_user(&kvm_mem, argp, sizeof kvm_mem))
6aa8b732 2924 goto out;
6fc138d2
IE		 2925 kvm_userspace_mem.slot = kvm_mem.slot;
2926 kvm_userspace_mem.flags = kvm_mem.flags;
2927 kvm_userspace_mem.guest_phys_addr = kvm_mem.guest_phys_addr;
2928 kvm_userspace_mem.memory_size = kvm_mem.memory_size;
2929 r = kvm_vm_ioctl_set_memory_region(kvm, &kvm_userspace_mem, 0);
2930 if (r)
2931 goto out;
2932 break;
2933 }
2934 case KVM_SET_USER_MEMORY_REGION: {
2935 struct kvm_userspace_memory_region kvm_userspace_mem;
2936
2937 r = -EFAULT;
2938 if (copy_from_user(&kvm_userspace_mem, argp,
2939 sizeof kvm_userspace_mem))
2940 goto out;
2941
2942 r = kvm_vm_ioctl_set_memory_region(kvm, &kvm_userspace_mem, 1);
6aa8b732
AK		 2943 if (r)
2944 goto out;
2945 break;
2946 }
82ce2c96
IE		 2947 case KVM_SET_NR_MMU_PAGES:
2948 r = kvm_vm_ioctl_set_nr_mmu_pages(kvm, arg);
2949 if (r)
2950 goto out;
2951 break;
2952 case KVM_GET_NR_MMU_PAGES:
2953 r = kvm_vm_ioctl_get_nr_mmu_pages(kvm);
2954 break;
6aa8b732
AK		 2955 case KVM_GET_DIRTY_LOG: {
2956 struct kvm_dirty_log log;
2957
2958 r = -EFAULT;
2f366987 2959 if (copy_from_user(&log, argp, sizeof log))
6aa8b732 2960 goto out;
2c6f5df9 2961 r = kvm_vm_ioctl_get_dirty_log(kvm, &log);
6aa8b732
AK		 2962 if (r)
2963 goto out;
2964 break;
2965 }
e8207547
AK		 2966 case KVM_SET_MEMORY_ALIAS: {
2967 struct kvm_memory_alias alias;
2968
2969 r = -EFAULT;
2970 if (copy_from_user(&alias, argp, sizeof alias))
2971 goto out;
2972 r = kvm_vm_ioctl_set_memory_alias(kvm, &alias);
2973 if (r)
2974 goto out;
2975 break;
2976 }
85f455f7
ED		 2977 case KVM_CREATE_IRQCHIP:
2978 r = -ENOMEM;
2979 kvm->vpic = kvm_create_pic(kvm);
1fd4f2a5
ED		 2980 if (kvm->vpic) {
2981 r = kvm_ioapic_init(kvm);
2982 if (r) {
2983 kfree(kvm->vpic);
2984 kvm->vpic = NULL;
2985 goto out;
2986 }
d77c26fc 2987 } else
85f455f7
ED		 2988 goto out;
2989 break;
2990 case KVM_IRQ_LINE: {
2991 struct kvm_irq_level irq_event;
2992
2993 r = -EFAULT;
2994 if (copy_from_user(&irq_event, argp, sizeof irq_event))
2995 goto out;
2996 if (irqchip_in_kernel(kvm)) {
9cf98828 2997 mutex_lock(&kvm->lock);
85f455f7
ED		 2998 if (irq_event.irq < 16)
2999 kvm_pic_set_irq(pic_irqchip(kvm),
3000 irq_event.irq,
3001 irq_event.level);
1fd4f2a5
ED		 3002 kvm_ioapic_set_irq(kvm->vioapic,
3003 irq_event.irq,
3004 irq_event.level);
9cf98828 3005 mutex_unlock(&kvm->lock);
85f455f7
ED		 3006 r = 0;
3007 }
3008 break;
3009 }
6ceb9d79
HQ		 3010 case KVM_GET_IRQCHIP: {
3011 /* 0: PIC master, 1: PIC slave, 2: IOAPIC */
3012 struct kvm_irqchip chip;
3013
3014 r = -EFAULT;
3015 if (copy_from_user(&chip, argp, sizeof chip))
3016 goto out;
3017 r = -ENXIO;
3018 if (!irqchip_in_kernel(kvm))
3019 goto out;
3020 r = kvm_vm_ioctl_get_irqchip(kvm, &chip);
3021 if (r)
3022 goto out;
3023 r = -EFAULT;
3024 if (copy_to_user(argp, &chip, sizeof chip))
3025 goto out;
3026 r = 0;
3027 break;
3028 }
3029 case KVM_SET_IRQCHIP: {
3030 /* 0: PIC master, 1: PIC slave, 2: IOAPIC */
3031 struct kvm_irqchip chip;
3032
3033 r = -EFAULT;
3034 if (copy_from_user(&chip, argp, sizeof chip))
3035 goto out;
3036 r = -ENXIO;
3037 if (!irqchip_in_kernel(kvm))
3038 goto out;
3039 r = kvm_vm_ioctl_set_irqchip(kvm, &chip);
3040 if (r)
3041 goto out;
3042 r = 0;
3043 break;
3044 }
f17abe9a
AK		 3045 default:
3046 ;
3047 }
3048out:
3049 return r;
3050}
3051
3052static struct page *kvm_vm_nopage(struct vm_area_struct *vma,
3053 unsigned long address,
3054 int *type)
3055{
3056 struct kvm *kvm = vma->vm_file->private_data;
3057 unsigned long pgoff;
f17abe9a
AK		 3058 struct page *page;
3059
f17abe9a 3060 pgoff = ((address - vma->vm_start) >> PAGE_SHIFT) + vma->vm_pgoff;
954bbbc2 3061 page = gfn_to_page(kvm, pgoff);
8a7ae055
IE		 3062 if (is_error_page(page)) {
3063 kvm_release_page(page);
f17abe9a 3064 return NOPAGE_SIGBUS;
8a7ae055 3065 }
cd0d9137
NAQ		 3066 if (type != NULL)
3067 *type = VM_FAULT_MINOR;
3068
f17abe9a
AK		 3069 return page;
3070}
3071
3072static struct vm_operations_struct kvm_vm_vm_ops = {
3073 .nopage = kvm_vm_nopage,
3074};
3075
3076static int kvm_vm_mmap(struct file *file, struct vm_area_struct *vma)
3077{
3078 vma->vm_ops = &kvm_vm_vm_ops;
3079 return 0;
3080}
3081
3082static struct file_operations kvm_vm_fops = {
3083 .release = kvm_vm_release,
3084 .unlocked_ioctl = kvm_vm_ioctl,
3085 .compat_ioctl = kvm_vm_ioctl,
3086 .mmap = kvm_vm_mmap,
3087};
3088
3089static int kvm_dev_ioctl_create_vm(void)
3090{
3091 int fd, r;
3092 struct inode *inode;
3093 struct file *file;
3094 struct kvm *kvm;
3095
f17abe9a 3096 kvm = kvm_create_vm();
d6d28168
AK		 3097 if (IS_ERR(kvm))
3098 return PTR_ERR(kvm);
3099 r = anon_inode_getfd(&fd, &inode, &file, "kvm-vm", &kvm_vm_fops, kvm);
3100 if (r) {
3101 kvm_destroy_vm(kvm);
3102 return r;
f17abe9a
AK		 3103 }
3104
bccf2150 3105 kvm->filp = file;
f17abe9a 3106
f17abe9a 3107 return fd;
f17abe9a
AK		 3108}
3109
3110static long kvm_dev_ioctl(struct file *filp,
3111 unsigned int ioctl, unsigned long arg)
3112{
3113 void __user *argp = (void __user *)arg;
07c45a36 3114 long r = -EINVAL;
f17abe9a
AK		 3115
3116 switch (ioctl) {
3117 case KVM_GET_API_VERSION:
f0fe5108
AK		 3118 r = -EINVAL;
3119 if (arg)
3120 goto out;
f17abe9a
AK		 3121 r = KVM_API_VERSION;
3122 break;
3123 case KVM_CREATE_VM:
f0fe5108
AK		 3124 r = -EINVAL;
3125 if (arg)
3126 goto out;
f17abe9a
AK		 3127 r = kvm_dev_ioctl_create_vm();
3128 break;
85f455f7
ED		 3129 case KVM_CHECK_EXTENSION: {
3130 int ext = (long)argp;
3131
3132 switch (ext) {
3133 case KVM_CAP_IRQCHIP:
b6958ce4 3134 case KVM_CAP_HLT:
82ce2c96 3135 case KVM_CAP_MMU_SHADOW_CACHE_CONTROL:
6fc138d2 3136 case KVM_CAP_USER_MEMORY:
85f455f7
ED		 3137 r = 1;
3138 break;
3139 default:
3140 r = 0;
3141 break;
3142 }
5d308f45 3143 break;
85f455f7 3144 }
07c45a36
AK		 3145 case KVM_GET_VCPU_MMAP_SIZE:
3146 r = -EINVAL;
3147 if (arg)
3148 goto out;
039576c0 3149 r = 2 * PAGE_SIZE;
07c45a36 3150 break;
6aa8b732 3151 default:
043405e1 3152 return kvm_arch_dev_ioctl(filp, ioctl, arg);
6aa8b732
AK		 3153 }
3154out:
3155 return r;
3156}
3157
6aa8b732 3158static struct file_operations kvm_chardev_ops = {
6aa8b732
AK		 3159 .unlocked_ioctl = kvm_dev_ioctl,
3160 .compat_ioctl = kvm_dev_ioctl,
6aa8b732
AK		 3161};
3162
3163static struct miscdevice kvm_dev = {
bbe4432e 3164 KVM_MINOR,
6aa8b732
AK		 3165 "kvm",
3166 &kvm_chardev_ops,
3167};
3168
774c47f1
AK		 3169/*
3170 * Make sure that a cpu that is being hot-unplugged does not have any vcpus
3171 * cached on it.
3172 */
3173static void decache_vcpus_on_cpu(int cpu)
3174{
3175 struct kvm *vm;
3176 struct kvm_vcpu *vcpu;
3177 int i;
3178
3179 spin_lock(&kvm_lock);
11ec2804 3180 list_for_each_entry(vm, &vm_list, vm_list)
774c47f1 3181 for (i = 0; i < KVM_MAX_VCPUS; ++i) {
fb3f0f51
RR		 3182 vcpu = vm->vcpus[i];
3183 if (!vcpu)
3184 continue;
774c47f1
AK		 3185 /*
3186 * If the vcpu is locked, then it is running on some
3187 * other cpu and therefore it is not cached on the
3188 * cpu in question.
3189 *
3190 * If it's not locked, check the last cpu it executed
3191 * on.
3192 */
3193 if (mutex_trylock(&vcpu->mutex)) {
3194 if (vcpu->cpu == cpu) {
cbdd1bea 3195 kvm_x86_ops->vcpu_decache(vcpu);
774c47f1
AK		 3196 vcpu->cpu = -1;
3197 }
3198 mutex_unlock(&vcpu->mutex);
3199 }
3200 }
3201 spin_unlock(&kvm_lock);
3202}
3203
1b6c0168
AK		 3204static void hardware_enable(void *junk)
3205{
3206 int cpu = raw_smp_processor_id();
3207
3208 if (cpu_isset(cpu, cpus_hardware_enabled))
3209 return;
3210 cpu_set(cpu, cpus_hardware_enabled);
cbdd1bea 3211 kvm_x86_ops->hardware_enable(NULL);
1b6c0168
AK		 3212}
3213
3214static void hardware_disable(void *junk)
3215{
3216 int cpu = raw_smp_processor_id();
3217
3218 if (!cpu_isset(cpu, cpus_hardware_enabled))
3219 return;
3220 cpu_clear(cpu, cpus_hardware_enabled);
3221 decache_vcpus_on_cpu(cpu);
cbdd1bea 3222 kvm_x86_ops->hardware_disable(NULL);
1b6c0168
AK		 3223}
3224
774c47f1
AK		 3225static int kvm_cpu_hotplug(struct notifier_block *notifier, unsigned long val,
3226 void *v)
3227{
3228 int cpu = (long)v;
3229
3230 switch (val) {
cec9ad27
AK		 3231 case CPU_DYING:
3232 case CPU_DYING_FROZEN:
6ec8a856
AK		 3233 printk(KERN_INFO "kvm: disabling virtualization on CPU%d\n",
3234 cpu);
3235 hardware_disable(NULL);
3236 break;
774c47f1 3237 case CPU_UP_CANCELED:
8bb78442 3238 case CPU_UP_CANCELED_FROZEN:
43934a38
JK		 3239 printk(KERN_INFO "kvm: disabling virtualization on CPU%d\n",
3240 cpu);
1b6c0168 3241 smp_call_function_single(cpu, hardware_disable, NULL, 0, 1);
774c47f1 3242 break;
43934a38 3243 case CPU_ONLINE:
8bb78442 3244 case CPU_ONLINE_FROZEN:
43934a38
JK		 3245 printk(KERN_INFO "kvm: enabling virtualization on CPU%d\n",
3246 cpu);
1b6c0168 3247 smp_call_function_single(cpu, hardware_enable, NULL, 0, 1);
774c47f1
AK		 3248 break;
3249 }
3250 return NOTIFY_OK;
3251}
3252
9a2b85c6 3253static int kvm_reboot(struct notifier_block *notifier, unsigned long val,
d77c26fc 3254 void *v)
9a2b85c6
RR		 3255{
3256 if (val == SYS_RESTART) {
3257 /*
3258 * Some (well, at least mine) BIOSes hang on reboot if
3259 * in vmx root mode.
3260 */
3261 printk(KERN_INFO "kvm: exiting hardware virtualization\n");
3262 on_each_cpu(hardware_disable, NULL, 0, 1);
3263 }
3264 return NOTIFY_OK;
3265}
3266
3267static struct notifier_block kvm_reboot_notifier = {
3268 .notifier_call = kvm_reboot,
3269 .priority = 0,
3270};
3271
2eeb2e94
GH		 3272void kvm_io_bus_init(struct kvm_io_bus *bus)
3273{
3274 memset(bus, 0, sizeof(*bus));
3275}
3276
3277void kvm_io_bus_destroy(struct kvm_io_bus *bus)
3278{
3279 int i;
3280
3281 for (i = 0; i < bus->dev_count; i++) {
3282 struct kvm_io_device *pos = bus->devs[i];
3283
3284 kvm_iodevice_destructor(pos);
3285 }
3286}
3287
3288struct kvm_io_device *kvm_io_bus_find_dev(struct kvm_io_bus *bus, gpa_t addr)
3289{
3290 int i;
3291
3292 for (i = 0; i < bus->dev_count; i++) {
3293 struct kvm_io_device *pos = bus->devs[i];
3294
3295 if (pos->in_range(pos, addr))
3296 return pos;
3297 }
3298
3299 return NULL;
3300}
3301
3302void kvm_io_bus_register_dev(struct kvm_io_bus *bus, struct kvm_io_device *dev)
3303{
3304 BUG_ON(bus->dev_count > (NR_IOBUS_DEVS-1));
3305
3306 bus->devs[bus->dev_count++] = dev;
3307}
3308
774c47f1
AK		 3309static struct notifier_block kvm_cpu_notifier = {
3310 .notifier_call = kvm_cpu_hotplug,
3311 .priority = 20, /* must be > scheduler priority */
3312};
3313
1165f5fe
AK		 3314static u64 stat_get(void *_offset)
3315{
3316 unsigned offset = (long)_offset;
3317 u64 total = 0;
3318 struct kvm *kvm;
3319 struct kvm_vcpu *vcpu;
3320 int i;
3321
3322 spin_lock(&kvm_lock);
3323 list_for_each_entry(kvm, &vm_list, vm_list)
3324 for (i = 0; i < KVM_MAX_VCPUS; ++i) {
fb3f0f51
RR		 3325 vcpu = kvm->vcpus[i];
3326 if (vcpu)
3327 total += *(u32 *)((void *)vcpu + offset);
1165f5fe
AK		 3328 }
3329 spin_unlock(&kvm_lock);
3330 return total;
3331}
3332
3dea7ca7 3333DEFINE_SIMPLE_ATTRIBUTE(stat_fops, stat_get, NULL, "%llu\n");
1165f5fe 3334
6aa8b732
AK		 3335static __init void kvm_init_debug(void)
3336{
3337 struct kvm_stats_debugfs_item *p;
3338
8b6d44c7 3339 debugfs_dir = debugfs_create_dir("kvm", NULL);
6aa8b732 3340 for (p = debugfs_entries; p->name; ++p)
1165f5fe
AK		 3341 p->dentry = debugfs_create_file(p->name, 0444, debugfs_dir,
3342 (void *)(long)p->offset,
3343 &stat_fops);
6aa8b732
AK		 3344}
3345
3346static void kvm_exit_debug(void)
3347{
3348 struct kvm_stats_debugfs_item *p;
3349
3350 for (p = debugfs_entries; p->name; ++p)
3351 debugfs_remove(p->dentry);
3352 debugfs_remove(debugfs_dir);
3353}
3354
59ae6c6b
AK		 3355static int kvm_suspend(struct sys_device *dev, pm_message_t state)
3356{
4267c41a 3357 hardware_disable(NULL);
59ae6c6b
AK		 3358 return 0;
3359}
3360
3361static int kvm_resume(struct sys_device *dev)
3362{
4267c41a 3363 hardware_enable(NULL);
59ae6c6b
AK		 3364 return 0;
3365}
3366
3367static struct sysdev_class kvm_sysdev_class = {
af5ca3f4 3368 .name = "kvm",
59ae6c6b
AK		 3369 .suspend = kvm_suspend,
3370 .resume = kvm_resume,
3371};
3372
3373static struct sys_device kvm_sysdev = {
3374 .id = 0,
3375 .cls = &kvm_sysdev_class,
3376};
3377
cea7bb21 3378struct page *bad_page;
6aa8b732 3379
15ad7146
AK		 3380static inline
3381struct kvm_vcpu *preempt_notifier_to_vcpu(struct preempt_notifier *pn)
3382{
3383 return container_of(pn, struct kvm_vcpu, preempt_notifier);
3384}
3385
3386static void kvm_sched_in(struct preempt_notifier *pn, int cpu)
3387{
3388 struct kvm_vcpu *vcpu = preempt_notifier_to_vcpu(pn);
3389
cbdd1bea 3390 kvm_x86_ops->vcpu_load(vcpu, cpu);
15ad7146
AK		 3391}
3392
3393static void kvm_sched_out(struct preempt_notifier *pn,
3394 struct task_struct *next)
3395{
3396 struct kvm_vcpu *vcpu = preempt_notifier_to_vcpu(pn);
3397
cbdd1bea 3398 kvm_x86_ops->vcpu_put(vcpu);
15ad7146
AK		 3399}
3400
cbdd1bea 3401int kvm_init_x86(struct kvm_x86_ops *ops, unsigned int vcpu_size,
c16f862d 3402 struct module *module)
6aa8b732
AK		 3403{
3404 int r;
002c7f7c 3405 int cpu;
6aa8b732 3406
cbdd1bea 3407 if (kvm_x86_ops) {
09db28b8
YI		 3408 printk(KERN_ERR "kvm: already loaded the other module\n");
3409 return -EEXIST;
3410 }
3411
e097f35c 3412 if (!ops->cpu_has_kvm_support()) {
6aa8b732
AK		 3413 printk(KERN_ERR "kvm: no hardware support\n");
3414 return -EOPNOTSUPP;
3415 }
e097f35c 3416 if (ops->disabled_by_bios()) {
6aa8b732
AK		 3417 printk(KERN_ERR "kvm: disabled by bios\n");
3418 return -EOPNOTSUPP;
3419 }
3420
cbdd1bea 3421 kvm_x86_ops = ops;
e097f35c 3422
cbdd1bea 3423 r = kvm_x86_ops->hardware_setup();
6aa8b732 3424 if (r < 0)
ca45aaae 3425 goto out;
6aa8b732 3426
002c7f7c
YS		 3427 for_each_online_cpu(cpu) {
3428 smp_call_function_single(cpu,
cbdd1bea 3429 kvm_x86_ops->check_processor_compatibility,
002c7f7c
YS		 3430 &r, 0, 1);
3431 if (r < 0)
3432 goto out_free_0;
3433 }
3434
1b6c0168 3435 on_each_cpu(hardware_enable, NULL, 0, 1);
774c47f1
AK		 3436 r = register_cpu_notifier(&kvm_cpu_notifier);
3437 if (r)
3438 goto out_free_1;
6aa8b732
AK		 3439 register_reboot_notifier(&kvm_reboot_notifier);
3440
59ae6c6b
AK		 3441 r = sysdev_class_register(&kvm_sysdev_class);
3442 if (r)
3443 goto out_free_2;
3444
3445 r = sysdev_register(&kvm_sysdev);
3446 if (r)
3447 goto out_free_3;
3448
c16f862d
RR		 3449 /* A kmem cache lets us meet the alignment requirements of fx_save. */
3450 kvm_vcpu_cache = kmem_cache_create("kvm_vcpu", vcpu_size,
3451 __alignof__(struct kvm_vcpu), 0, 0);
3452 if (!kvm_vcpu_cache) {
3453 r = -ENOMEM;
3454 goto out_free_4;
3455 }
3456
6aa8b732
AK		 3457 kvm_chardev_ops.owner = module;
3458
3459 r = misc_register(&kvm_dev);
3460 if (r) {
d77c26fc 3461 printk(KERN_ERR "kvm: misc device register failed\n");
6aa8b732
AK		 3462 goto out_free;
3463 }
3464
15ad7146
AK		 3465 kvm_preempt_ops.sched_in = kvm_sched_in;
3466 kvm_preempt_ops.sched_out = kvm_sched_out;
3467
c7addb90
AK		 3468 kvm_mmu_set_nonpresent_ptes(0ull, 0ull);
3469
3470 return 0;
6aa8b732
AK		 3471
3472out_free:
c16f862d
RR		 3473 kmem_cache_destroy(kvm_vcpu_cache);
3474out_free_4:
59ae6c6b
AK		 3475 sysdev_unregister(&kvm_sysdev);
3476out_free_3:
3477 sysdev_class_unregister(&kvm_sysdev_class);
3478out_free_2:
6aa8b732 3479 unregister_reboot_notifier(&kvm_reboot_notifier);
774c47f1
AK		 3480 unregister_cpu_notifier(&kvm_cpu_notifier);
3481out_free_1:
1b6c0168 3482 on_each_cpu(hardware_disable, NULL, 0, 1);
002c7f7c 3483out_free_0:
cbdd1bea 3484 kvm_x86_ops->hardware_unsetup();
ca45aaae 3485out:
cbdd1bea 3486 kvm_x86_ops = NULL;
6aa8b732
AK		 3487 return r;
3488}
d77c26fc 3489EXPORT_SYMBOL_GPL(kvm_init_x86);
6aa8b732 3490
cbdd1bea 3491void kvm_exit_x86(void)
6aa8b732
AK		 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);
cbdd1bea
CE		 3500 kvm_x86_ops->hardware_unsetup();
3501 kvm_x86_ops = NULL;
6aa8b732 3502}
d77c26fc 3503EXPORT_SYMBOL_GPL(kvm_exit_x86);
6aa8b732
AK		 3504
3505static __init int kvm_init(void)
3506{
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
043405e1 3515 kvm_arch_init();
bf591b24 3516
cea7bb21 3517 bad_page = alloc_page(GFP_KERNEL | __GFP_ZERO);
d77c26fc
MD		 3518
3519 if (bad_page == NULL) {
6aa8b732
AK		 3520 r = -ENOMEM;
3521 goto out;
3522 }
3523
58e690e6 3524 return 0;
6aa8b732
AK		 3525
3526out:
3527 kvm_exit_debug();
b5a33a75
AK		 3528 kvm_mmu_module_exit();
3529out4:
6aa8b732
AK		 3530 return r;
3531}
3532
3533static __exit void kvm_exit(void)
3534{
3535 kvm_exit_debug();
cea7bb21 3536 __free_page(bad_page);
b5a33a75 3537 kvm_mmu_module_exit();
6aa8b732
AK		 3538}
3539
3540module_init(kvm_init)
3541module_exit(kvm_exit)
ubuntu-bionic-kernel mirror