]> bbs.cooldavid.org Git - net-next-2.6.git/blame - drivers/kvm/kvm_main.c
git://bbs.cooldavid.org / net-next-2.6.git / blame
? search:
summary | shortlog | log | commit | commitdiff | tree
blob | blame (incremental) | history | HEAD
KVM: VMX: Simplify vcpu_clear()
[net-next-2.6.git] / drivers / kvm / kvm_main.c
CommitLineData
6aa8b732
AK		 1/*
2 * Kernel-based Virtual Machine driver for Linux
3 *
4 * This module enables machines with Intel VT-x extensions to run virtual
5 * machines without emulation or binary translation.
6 *
7 * Copyright (C) 2006 Qumranet, Inc.
8 *
9 * Authors:
10 * Avi Kivity <avi@qumranet.com>
11 * Yaniv Kamay <yaniv@qumranet.com>
12 *
13 * This work is licensed under the terms of the GNU GPL, version 2. See
14 * the COPYING file in the top-level directory.
15 *
16 */
17
18#include "kvm.h"
e495606d
AK		 19#include "x86_emulate.h"
20#include "segment_descriptor.h"
85f455f7 21#include "irq.h"
6aa8b732
AK		 22
23#include <linux/kvm.h>
24#include <linux/module.h>
25#include <linux/errno.h>
6aa8b732
AK		 26#include <linux/percpu.h>
27#include <linux/gfp.h>
6aa8b732
AK		 28#include <linux/mm.h>
29#include <linux/miscdevice.h>
30#include <linux/vmalloc.h>
6aa8b732 31#include <linux/reboot.h>
6aa8b732
AK		 32#include <linux/debugfs.h>
33#include <linux/highmem.h>
34#include <linux/file.h>
59ae6c6b 35#include <linux/sysdev.h>
774c47f1 36#include <linux/cpu.h>
e8edc6e0 37#include <linux/sched.h>
d9e368d6
AK		 38#include <linux/cpumask.h>
39#include <linux/smp.h>
d6d28168 40#include <linux/anon_inodes.h>
04d2cc77 41#include <linux/profile.h>
7aa81cc0 42#include <linux/kvm_para.h>
6aa8b732 43
e495606d
AK		 44#include <asm/processor.h>
45#include <asm/msr.h>
46#include <asm/io.h>
47#include <asm/uaccess.h>
48#include <asm/desc.h>
6aa8b732
AK		 49
50MODULE_AUTHOR("Qumranet");
51MODULE_LICENSE("GPL");
52
133de902
AK		 53static DEFINE_SPINLOCK(kvm_lock);
54static LIST_HEAD(vm_list);
55
1b6c0168
AK		 56static cpumask_t cpus_hardware_enabled;
57
cbdd1bea 58struct kvm_x86_ops *kvm_x86_ops;
c16f862d
RR		 59struct kmem_cache *kvm_vcpu_cache;
60EXPORT_SYMBOL_GPL(kvm_vcpu_cache);
1165f5fe 61
15ad7146
AK		 62static __read_mostly struct preempt_ops kvm_preempt_ops;
63
1165f5fe 64#define STAT_OFFSET(x) offsetof(struct kvm_vcpu, stat.x)
6aa8b732
AK		 65
66static struct kvm_stats_debugfs_item {
67 const char *name;
1165f5fe 68 int offset;
6aa8b732
AK		 69 struct dentry *dentry;
70} debugfs_entries[] = {
1165f5fe
AK		 71 { "pf_fixed", STAT_OFFSET(pf_fixed) },
72 { "pf_guest", STAT_OFFSET(pf_guest) },
73 { "tlb_flush", STAT_OFFSET(tlb_flush) },
74 { "invlpg", STAT_OFFSET(invlpg) },
75 { "exits", STAT_OFFSET(exits) },
76 { "io_exits", STAT_OFFSET(io_exits) },
77 { "mmio_exits", STAT_OFFSET(mmio_exits) },
78 { "signal_exits", STAT_OFFSET(signal_exits) },
79 { "irq_window", STAT_OFFSET(irq_window_exits) },
80 { "halt_exits", STAT_OFFSET(halt_exits) },
b6958ce4 81 { "halt_wakeup", STAT_OFFSET(halt_wakeup) },
1165f5fe
AK		 82 { "request_irq", STAT_OFFSET(request_irq_exits) },
83 { "irq_exits", STAT_OFFSET(irq_exits) },
e6adf283 84 { "light_exits", STAT_OFFSET(light_exits) },
2cc51560 85 { "efer_reload", STAT_OFFSET(efer_reload) },
1165f5fe 86 { NULL }
6aa8b732
AK		 87};
88
89static struct dentry *debugfs_dir;
90
91#define MAX_IO_MSRS 256
92
707d92fa
RR		 93#define CR0_RESERVED_BITS \
94 (~(unsigned long)(X86_CR0_PE | X86_CR0_MP | X86_CR0_EM | X86_CR0_TS \
95 | X86_CR0_ET | X86_CR0_NE | X86_CR0_WP | X86_CR0_AM \
96 | X86_CR0_NW | X86_CR0_CD | X86_CR0_PG))
66aee91a
RR		 97#define CR4_RESERVED_BITS \
98 (~(unsigned long)(X86_CR4_VME | X86_CR4_PVI | X86_CR4_TSD | X86_CR4_DE\
99 | X86_CR4_PSE | X86_CR4_PAE | X86_CR4_MCE \
100 | X86_CR4_PGE | X86_CR4_PCE | X86_CR4_OSFXSR \
101 | X86_CR4_OSXMMEXCPT | X86_CR4_VMXE))
102
7075bc81 103#define CR8_RESERVED_BITS (~(unsigned long)X86_CR8_TPR)
6aa8b732
AK		 104#define EFER_RESERVED_BITS 0xfffffffffffff2fe
105
05b3e0c2 106#ifdef CONFIG_X86_64
6aa8b732
AK		 107// LDT or TSS descriptor in the GDT. 16 bytes.
108struct segment_descriptor_64 {
109 struct segment_descriptor s;
110 u32 base_higher;
111 u32 pad_zero;
112};
113
114#endif
115
bccf2150
AK		 116static long kvm_vcpu_ioctl(struct file *file, unsigned int ioctl,
117 unsigned long arg);
118
6aa8b732
AK		 119unsigned long segment_base(u16 selector)
120{
121 struct descriptor_table gdt;
122 struct segment_descriptor *d;
123 unsigned long table_base;
124 typedef unsigned long ul;
125 unsigned long v;
126
127 if (selector == 0)
128 return 0;
129
130 asm ("sgdt %0" : "=m"(gdt));
131 table_base = gdt.base;
132
133 if (selector & 4) { /* from ldt */
134 u16 ldt_selector;
135
136 asm ("sldt %0" : "=g"(ldt_selector));
137 table_base = segment_base(ldt_selector);
138 }
139 d = (struct segment_descriptor *)(table_base + (selector & ~7));
140 v = d->base_low | ((ul)d->base_mid << 16) | ((ul)d->base_high << 24);
05b3e0c2 141#ifdef CONFIG_X86_64
6aa8b732
AK		 142 if (d->system == 0
143 && (d->type == 2 || d->type == 9 || d->type == 11))
144 v |= ((ul)((struct segment_descriptor_64 *)d)->base_higher) << 32;
145#endif
146 return v;
147}
148EXPORT_SYMBOL_GPL(segment_base);
149
5aacf0ca
JM		 150static inline int valid_vcpu(int n)
151{
152 return likely(n >= 0 && n < KVM_MAX_VCPUS);
153}
154
7702fd1f
AK		 155void kvm_load_guest_fpu(struct kvm_vcpu *vcpu)
156{
157 if (!vcpu->fpu_active || vcpu->guest_fpu_loaded)
158 return;
159
160 vcpu->guest_fpu_loaded = 1;
b114b080
RR		 161 fx_save(&vcpu->host_fx_image);
162 fx_restore(&vcpu->guest_fx_image);
7702fd1f
AK		 163}
164EXPORT_SYMBOL_GPL(kvm_load_guest_fpu);
165
166void kvm_put_guest_fpu(struct kvm_vcpu *vcpu)
167{
168 if (!vcpu->guest_fpu_loaded)
169 return;
170
171 vcpu->guest_fpu_loaded = 0;
b114b080
RR		 172 fx_save(&vcpu->guest_fx_image);
173 fx_restore(&vcpu->host_fx_image);
7702fd1f
AK		 174}
175EXPORT_SYMBOL_GPL(kvm_put_guest_fpu);
176
bccf2150
AK		 177/*
178 * Switches to specified vcpu, until a matching vcpu_put()
179 */
180static void vcpu_load(struct kvm_vcpu *vcpu)
6aa8b732 181{
15ad7146
AK		 182 int cpu;
183
bccf2150 184 mutex_lock(&vcpu->mutex);
15ad7146
AK		 185 cpu = get_cpu();
186 preempt_notifier_register(&vcpu->preempt_notifier);
cbdd1bea 187 kvm_x86_ops->vcpu_load(vcpu, cpu);
15ad7146 188 put_cpu();
6aa8b732
AK		 189}
190
6aa8b732
AK		 191static void vcpu_put(struct kvm_vcpu *vcpu)
192{
15ad7146 193 preempt_disable();
cbdd1bea 194 kvm_x86_ops->vcpu_put(vcpu);
15ad7146
AK		 195 preempt_notifier_unregister(&vcpu->preempt_notifier);
196 preempt_enable();
6aa8b732
AK		 197 mutex_unlock(&vcpu->mutex);
198}
199
d9e368d6
AK		 200static void ack_flush(void *_completed)
201{
d9e368d6
AK		 202}
203
204void kvm_flush_remote_tlbs(struct kvm *kvm)
205{
49d3bd7e 206 int i, cpu;
d9e368d6
AK		 207 cpumask_t cpus;
208 struct kvm_vcpu *vcpu;
d9e368d6 209
d9e368d6 210 cpus_clear(cpus);
fb3f0f51
RR		 211 for (i = 0; i < KVM_MAX_VCPUS; ++i) {
212 vcpu = kvm->vcpus[i];
213 if (!vcpu)
214 continue;
d9e368d6
AK		 215 if (test_and_set_bit(KVM_TLB_FLUSH, &vcpu->requests))
216 continue;
217 cpu = vcpu->cpu;
218 if (cpu != -1 && cpu != raw_smp_processor_id())
49d3bd7e 219 cpu_set(cpu, cpus);
d9e368d6 220 }
49d3bd7e 221 smp_call_function_mask(cpus, ack_flush, NULL, 1);
d9e368d6
AK		 222}
223
fb3f0f51
RR		 224int kvm_vcpu_init(struct kvm_vcpu *vcpu, struct kvm *kvm, unsigned id)
225{
226 struct page *page;
227 int r;
228
229 mutex_init(&vcpu->mutex);
230 vcpu->cpu = -1;
231 vcpu->mmu.root_hpa = INVALID_PAGE;
232 vcpu->kvm = kvm;
233 vcpu->vcpu_id = id;
c5ec1534
HQ		 234 if (!irqchip_in_kernel(kvm) || id == 0)
235 vcpu->mp_state = VCPU_MP_STATE_RUNNABLE;
236 else
237 vcpu->mp_state = VCPU_MP_STATE_UNINITIALIZED;
b6958ce4 238 init_waitqueue_head(&vcpu->wq);
fb3f0f51
RR		 239
240 page = alloc_page(GFP_KERNEL | __GFP_ZERO);
241 if (!page) {
242 r = -ENOMEM;
243 goto fail;
244 }
245 vcpu->run = page_address(page);
246
247 page = alloc_page(GFP_KERNEL | __GFP_ZERO);
248 if (!page) {
249 r = -ENOMEM;
250 goto fail_free_run;
251 }
252 vcpu->pio_data = page_address(page);
253
fb3f0f51
RR		 254 r = kvm_mmu_create(vcpu);
255 if (r < 0)
256 goto fail_free_pio_data;
257
258 return 0;
259
260fail_free_pio_data:
261 free_page((unsigned long)vcpu->pio_data);
262fail_free_run:
263 free_page((unsigned long)vcpu->run);
264fail:
265 return -ENOMEM;
266}
267EXPORT_SYMBOL_GPL(kvm_vcpu_init);
268
269void kvm_vcpu_uninit(struct kvm_vcpu *vcpu)
270{
271 kvm_mmu_destroy(vcpu);
1b9778da
ED		 272 if (vcpu->apic)
273 hrtimer_cancel(&vcpu->apic->timer.dev);
97222cc8 274 kvm_free_apic(vcpu->apic);
fb3f0f51
RR		 275 free_page((unsigned long)vcpu->pio_data);
276 free_page((unsigned long)vcpu->run);
277}
278EXPORT_SYMBOL_GPL(kvm_vcpu_uninit);
279
f17abe9a 280static struct kvm *kvm_create_vm(void)
6aa8b732
AK		 281{
282 struct kvm *kvm = kzalloc(sizeof(struct kvm), GFP_KERNEL);
6aa8b732
AK		 283
284 if (!kvm)
f17abe9a 285 return ERR_PTR(-ENOMEM);
6aa8b732 286
74906345 287 kvm_io_bus_init(&kvm->pio_bus);
11ec2804 288 mutex_init(&kvm->lock);
6aa8b732 289 INIT_LIST_HEAD(&kvm->active_mmu_pages);
2eeb2e94 290 kvm_io_bus_init(&kvm->mmio_bus);
5e58cfe4
RR		 291 spin_lock(&kvm_lock);
292 list_add(&kvm->vm_list, &vm_list);
293 spin_unlock(&kvm_lock);
f17abe9a
AK		 294 return kvm;
295}
296
6aa8b732
AK		 297/*
298 * Free any memory in @free but not in @dont.
299 */
300static void kvm_free_physmem_slot(struct kvm_memory_slot *free,
301 struct kvm_memory_slot *dont)
302{
303 int i;
304
305 if (!dont || free->phys_mem != dont->phys_mem)
306 if (free->phys_mem) {
307 for (i = 0; i < free->npages; ++i)
55a54f79
AK		 308 if (free->phys_mem[i])
309 __free_page(free->phys_mem[i]);
6aa8b732
AK		 310 vfree(free->phys_mem);
311 }
312
313 if (!dont || free->dirty_bitmap != dont->dirty_bitmap)
314 vfree(free->dirty_bitmap);
315
8b6d44c7 316 free->phys_mem = NULL;
6aa8b732 317 free->npages = 0;
8b6d44c7 318 free->dirty_bitmap = 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
AK		 334 if (vcpu->pio.guest_pages[i]) {
335 __free_page(vcpu->pio.guest_pages[i]);
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;
6aa8b732 401 u64 *pdpt;
1342d353 402 int ret;
954bbbc2 403 struct page *page;
c820c2aa 404 u64 pdpte[ARRAY_SIZE(vcpu->pdptrs)];
6aa8b732 405
11ec2804 406 mutex_lock(&vcpu->kvm->lock);
954bbbc2 407 page = gfn_to_page(vcpu->kvm, pdpt_gfn);
c820c2aa
RR		 408 if (!page) {
409 ret = 0;
410 goto out;
411 }
412
954bbbc2 413 pdpt = kmap_atomic(page, KM_USER0);
c820c2aa
RR		 414 memcpy(pdpte, pdpt+offset, sizeof(pdpte));
415 kunmap_atomic(pdpt, KM_USER0);
6aa8b732 416
c820c2aa
RR		 417 for (i = 0; i < ARRAY_SIZE(pdpte); ++i) {
418 if ((pdpte[i] & 1) && (pdpte[i] & 0xfffffff0000001e6ull)) {
1342d353
AK		 419 ret = 0;
420 goto out;
421 }
6aa8b732 422 }
c820c2aa 423 ret = 1;
6aa8b732 424
c820c2aa 425 memcpy(vcpu->pdptrs, pdpte, sizeof(vcpu->pdptrs));
1342d353 426out:
11ec2804 427 mutex_unlock(&vcpu->kvm->lock);
6aa8b732 428
1342d353 429 return ret;
6aa8b732
AK		 430}
431
432void set_cr0(struct kvm_vcpu *vcpu, unsigned long cr0)
433{
707d92fa 434 if (cr0 & CR0_RESERVED_BITS) {
6aa8b732
AK		 435 printk(KERN_DEBUG "set_cr0: 0x%lx #GP, reserved bits 0x%lx\n",
436 cr0, vcpu->cr0);
437 inject_gp(vcpu);
438 return;
439 }
440
707d92fa 441 if ((cr0 & X86_CR0_NW) && !(cr0 & X86_CR0_CD)) {
6aa8b732
AK		 442 printk(KERN_DEBUG "set_cr0: #GP, CD == 0 && NW == 1\n");
443 inject_gp(vcpu);
444 return;
445 }
446
707d92fa 447 if ((cr0 & X86_CR0_PG) && !(cr0 & X86_CR0_PE)) {
6aa8b732
AK		 448 printk(KERN_DEBUG "set_cr0: #GP, set PG flag "
449 "and a clear PE flag\n");
450 inject_gp(vcpu);
451 return;
452 }
453
707d92fa 454 if (!is_paging(vcpu) && (cr0 & X86_CR0_PG)) {
05b3e0c2 455#ifdef CONFIG_X86_64
6aa8b732
AK		 456 if ((vcpu->shadow_efer & EFER_LME)) {
457 int cs_db, cs_l;
458
459 if (!is_pae(vcpu)) {
460 printk(KERN_DEBUG "set_cr0: #GP, start paging "
461 "in long mode while PAE is disabled\n");
462 inject_gp(vcpu);
463 return;
464 }
cbdd1bea 465 kvm_x86_ops->get_cs_db_l_bits(vcpu, &cs_db, &cs_l);
6aa8b732
AK		 466 if (cs_l) {
467 printk(KERN_DEBUG "set_cr0: #GP, start paging "
468 "in long mode while CS.L == 1\n");
469 inject_gp(vcpu);
470 return;
471
472 }
473 } else
474#endif
1342d353 475 if (is_pae(vcpu) && !load_pdptrs(vcpu, vcpu->cr3)) {
6aa8b732
AK		 476 printk(KERN_DEBUG "set_cr0: #GP, pdptrs "
477 "reserved bits\n");
478 inject_gp(vcpu);
479 return;
480 }
481
482 }
483
cbdd1bea 484 kvm_x86_ops->set_cr0(vcpu, cr0);
6aa8b732
AK		 485 vcpu->cr0 = cr0;
486
11ec2804 487 mutex_lock(&vcpu->kvm->lock);
6aa8b732 488 kvm_mmu_reset_context(vcpu);
11ec2804 489 mutex_unlock(&vcpu->kvm->lock);
6aa8b732
AK		 490 return;
491}
492EXPORT_SYMBOL_GPL(set_cr0);
493
494void lmsw(struct kvm_vcpu *vcpu, unsigned long msw)
495{
496 set_cr0(vcpu, (vcpu->cr0 & ~0x0ful) | (msw & 0x0f));
497}
498EXPORT_SYMBOL_GPL(lmsw);
499
500void set_cr4(struct kvm_vcpu *vcpu, unsigned long cr4)
501{
66aee91a 502 if (cr4 & CR4_RESERVED_BITS) {
6aa8b732
AK		 503 printk(KERN_DEBUG "set_cr4: #GP, reserved bits\n");
504 inject_gp(vcpu);
505 return;
506 }
507
a9058ecd 508 if (is_long_mode(vcpu)) {
66aee91a 509 if (!(cr4 & X86_CR4_PAE)) {
6aa8b732
AK		 510 printk(KERN_DEBUG "set_cr4: #GP, clearing PAE while "
511 "in long mode\n");
512 inject_gp(vcpu);
513 return;
514 }
66aee91a 515 } else if (is_paging(vcpu) && !is_pae(vcpu) && (cr4 & X86_CR4_PAE)
1342d353 516 && !load_pdptrs(vcpu, vcpu->cr3)) {
6aa8b732
AK		 517 printk(KERN_DEBUG "set_cr4: #GP, pdptrs reserved bits\n");
518 inject_gp(vcpu);
310bc76c 519 return;
6aa8b732
AK		 520 }
521
66aee91a 522 if (cr4 & X86_CR4_VMXE) {
6aa8b732
AK		 523 printk(KERN_DEBUG "set_cr4: #GP, setting VMXE\n");
524 inject_gp(vcpu);
525 return;
526 }
cbdd1bea 527 kvm_x86_ops->set_cr4(vcpu, cr4);
81f50e3b 528 vcpu->cr4 = cr4;
11ec2804 529 mutex_lock(&vcpu->kvm->lock);
6aa8b732 530 kvm_mmu_reset_context(vcpu);
11ec2804 531 mutex_unlock(&vcpu->kvm->lock);
6aa8b732
AK		 532}
533EXPORT_SYMBOL_GPL(set_cr4);
534
535void set_cr3(struct kvm_vcpu *vcpu, unsigned long cr3)
536{
a9058ecd 537 if (is_long_mode(vcpu)) {
f802a307 538 if (cr3 & CR3_L_MODE_RESERVED_BITS) {
6aa8b732
AK		 539 printk(KERN_DEBUG "set_cr3: #GP, reserved bits\n");
540 inject_gp(vcpu);
541 return;
542 }
543 } else {
f802a307
RR		 544 if (is_pae(vcpu)) {
545 if (cr3 & CR3_PAE_RESERVED_BITS) {
546 printk(KERN_DEBUG
547 "set_cr3: #GP, reserved bits\n");
548 inject_gp(vcpu);
549 return;
550 }
551 if (is_paging(vcpu) && !load_pdptrs(vcpu, cr3)) {
552 printk(KERN_DEBUG "set_cr3: #GP, pdptrs "
553 "reserved bits\n");
554 inject_gp(vcpu);
555 return;
556 }
6aa8b732 557 }
21764863
RH		 558 /*
559 * We don't check reserved bits in nonpae mode, because
560 * this isn't enforced, and VMware depends on this.
561 */
6aa8b732
AK		 562 }
563
11ec2804 564 mutex_lock(&vcpu->kvm->lock);
d21225ee
IM		 565 /*
566 * Does the new cr3 value map to physical memory? (Note, we
567 * catch an invalid cr3 even in real-mode, because it would
568 * cause trouble later on when we turn on paging anyway.)
569 *
570 * A real CPU would silently accept an invalid cr3 and would
571 * attempt to use it - with largely undefined (and often hard
572 * to debug) behavior on the guest side.
573 */
574 if (unlikely(!gfn_to_memslot(vcpu->kvm, cr3 >> PAGE_SHIFT)))
575 inject_gp(vcpu);
fb764416
RR		 576 else {
577 vcpu->cr3 = cr3;
d21225ee 578 vcpu->mmu.new_cr3(vcpu);
fb764416 579 }
11ec2804 580 mutex_unlock(&vcpu->kvm->lock);
6aa8b732
AK		 581}
582EXPORT_SYMBOL_GPL(set_cr3);
583
584void set_cr8(struct kvm_vcpu *vcpu, unsigned long cr8)
585{
7075bc81 586 if (cr8 & CR8_RESERVED_BITS) {
6aa8b732
AK		 587 printk(KERN_DEBUG "set_cr8: #GP, reserved bits 0x%lx\n", cr8);
588 inject_gp(vcpu);
589 return;
590 }
97222cc8
ED		 591 if (irqchip_in_kernel(vcpu->kvm))
592 kvm_lapic_set_tpr(vcpu, cr8);
593 else
594 vcpu->cr8 = cr8;
6aa8b732
AK		 595}
596EXPORT_SYMBOL_GPL(set_cr8);
597
7017fc3d
ED		 598unsigned long get_cr8(struct kvm_vcpu *vcpu)
599{
97222cc8
ED		 600 if (irqchip_in_kernel(vcpu->kvm))
601 return kvm_lapic_get_cr8(vcpu);
602 else
603 return vcpu->cr8;
7017fc3d
ED		 604}
605EXPORT_SYMBOL_GPL(get_cr8);
606
607u64 kvm_get_apic_base(struct kvm_vcpu *vcpu)
608{
97222cc8
ED		 609 if (irqchip_in_kernel(vcpu->kvm))
610 return vcpu->apic_base;
611 else
612 return vcpu->apic_base;
7017fc3d
ED		 613}
614EXPORT_SYMBOL_GPL(kvm_get_apic_base);
615
616void kvm_set_apic_base(struct kvm_vcpu *vcpu, u64 data)
617{
97222cc8
ED		 618 /* TODO: reserve bits check */
619 if (irqchip_in_kernel(vcpu->kvm))
620 kvm_lapic_set_base(vcpu, data);
621 else
622 vcpu->apic_base = data;
7017fc3d
ED		 623}
624EXPORT_SYMBOL_GPL(kvm_set_apic_base);
625
6aa8b732
AK		 626void fx_init(struct kvm_vcpu *vcpu)
627{
b114b080 628 unsigned after_mxcsr_mask;
6aa8b732 629
9bd01506
RR		 630 /* Initialize guest FPU by resetting ours and saving into guest's */
631 preempt_disable();
b114b080 632 fx_save(&vcpu->host_fx_image);
6aa8b732 633 fpu_init();
b114b080
RR		 634 fx_save(&vcpu->guest_fx_image);
635 fx_restore(&vcpu->host_fx_image);
9bd01506 636 preempt_enable();
6aa8b732 637
380102c8 638 vcpu->cr0 |= X86_CR0_ET;
b114b080
RR		 639 after_mxcsr_mask = offsetof(struct i387_fxsave_struct, st_space);
640 vcpu->guest_fx_image.mxcsr = 0x1f80;
641 memset((void *)&vcpu->guest_fx_image + after_mxcsr_mask,
642 0, sizeof(struct i387_fxsave_struct) - after_mxcsr_mask);
6aa8b732
AK		 643}
644EXPORT_SYMBOL_GPL(fx_init);
645
6aa8b732
AK		 646/*
647 * Allocate some memory and give it an address in the guest physical address
648 * space.
649 *
650 * Discontiguous memory is allowed, mostly for framebuffers.
651 */
2c6f5df9
AK		 652static int kvm_vm_ioctl_set_memory_region(struct kvm *kvm,
653 struct kvm_memory_region *mem)
6aa8b732
AK		 654{
655 int r;
656 gfn_t base_gfn;
657 unsigned long npages;
658 unsigned long i;
659 struct kvm_memory_slot *memslot;
660 struct kvm_memory_slot old, new;
6aa8b732
AK		 661
662 r = -EINVAL;
663 /* General sanity checks */
664 if (mem->memory_size & (PAGE_SIZE - 1))
665 goto out;
666 if (mem->guest_phys_addr & (PAGE_SIZE - 1))
667 goto out;
668 if (mem->slot >= KVM_MEMORY_SLOTS)
669 goto out;
670 if (mem->guest_phys_addr + mem->memory_size < mem->guest_phys_addr)
671 goto out;
672
673 memslot = &kvm->memslots[mem->slot];
674 base_gfn = mem->guest_phys_addr >> PAGE_SHIFT;
675 npages = mem->memory_size >> PAGE_SHIFT;
676
677 if (!npages)
678 mem->flags &= ~KVM_MEM_LOG_DIRTY_PAGES;
679
11ec2804 680 mutex_lock(&kvm->lock);
6aa8b732 681
6aa8b732
AK		 682 new = old = *memslot;
683
684 new.base_gfn = base_gfn;
685 new.npages = npages;
686 new.flags = mem->flags;
687
688 /* Disallow changing a memory slot's size. */
689 r = -EINVAL;
690 if (npages && old.npages && npages != old.npages)
691 goto out_unlock;
692
693 /* Check for overlaps */
694 r = -EEXIST;
695 for (i = 0; i < KVM_MEMORY_SLOTS; ++i) {
696 struct kvm_memory_slot *s = &kvm->memslots[i];
697
698 if (s == memslot)
699 continue;
700 if (!((base_gfn + npages <= s->base_gfn) ||
701 (base_gfn >= s->base_gfn + s->npages)))
702 goto out_unlock;
703 }
6aa8b732
AK		 704
705 /* Deallocate if slot is being removed */
706 if (!npages)
8b6d44c7 707 new.phys_mem = NULL;
6aa8b732
AK		 708
709 /* Free page dirty bitmap if unneeded */
710 if (!(new.flags & KVM_MEM_LOG_DIRTY_PAGES))
8b6d44c7 711 new.dirty_bitmap = NULL;
6aa8b732
AK		 712
713 r = -ENOMEM;
714
715 /* Allocate if a slot is being created */
716 if (npages && !new.phys_mem) {
717 new.phys_mem = vmalloc(npages * sizeof(struct page *));
718
719 if (!new.phys_mem)
0d8d2bd4 720 goto out_unlock;
6aa8b732
AK		 721
722 memset(new.phys_mem, 0, npages * sizeof(struct page *));
723 for (i = 0; i < npages; ++i) {
724 new.phys_mem[i] = alloc_page(GFP_HIGHUSER
725 | __GFP_ZERO);
726 if (!new.phys_mem[i])
0d8d2bd4 727 goto out_unlock;
5972e953 728 set_page_private(new.phys_mem[i],0);
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
745 *memslot = new;
6aa8b732 746
90cb0529
AK		 747 kvm_mmu_slot_remove_write_access(kvm, mem->slot);
748 kvm_flush_remote_tlbs(kvm);
6aa8b732 749
11ec2804 750 mutex_unlock(&kvm->lock);
6aa8b732
AK		 751
752 kvm_free_physmem_slot(&old, &new);
753 return 0;
754
755out_unlock:
11ec2804 756 mutex_unlock(&kvm->lock);
6aa8b732
AK		 757 kvm_free_physmem_slot(&new, &old);
758out:
759 return r;
760}
761
762/*
763 * Get (and clear) the dirty memory log for a memory slot.
764 */
2c6f5df9
AK		 765static int kvm_vm_ioctl_get_dirty_log(struct kvm *kvm,
766 struct kvm_dirty_log *log)
6aa8b732
AK		 767{
768 struct kvm_memory_slot *memslot;
769 int r, i;
770 int n;
771 unsigned long any = 0;
772
11ec2804 773 mutex_lock(&kvm->lock);
6aa8b732 774
6aa8b732
AK		 775 r = -EINVAL;
776 if (log->slot >= KVM_MEMORY_SLOTS)
777 goto out;
778
779 memslot = &kvm->memslots[log->slot];
780 r = -ENOENT;
781 if (!memslot->dirty_bitmap)
782 goto out;
783
cd1a4a98 784 n = ALIGN(memslot->npages, BITS_PER_LONG) / 8;
6aa8b732 785
cd1a4a98 786 for (i = 0; !any && i < n/sizeof(long); ++i)
6aa8b732
AK		 787 any = memslot->dirty_bitmap[i];
788
789 r = -EFAULT;
790 if (copy_to_user(log->dirty_bitmap, memslot->dirty_bitmap, n))
791 goto out;
792
39214915
RR		 793 /* If nothing is dirty, don't bother messing with page tables. */
794 if (any) {
39214915
RR		 795 kvm_mmu_slot_remove_write_access(kvm, log->slot);
796 kvm_flush_remote_tlbs(kvm);
797 memset(memslot->dirty_bitmap, 0, n);
39214915 798 }
6aa8b732
AK		 799
800 r = 0;
801
802out:
11ec2804 803 mutex_unlock(&kvm->lock);
6aa8b732
AK		 804 return r;
805}
806
e8207547
AK		 807/*
808 * Set a new alias region. Aliases map a portion of physical memory into
809 * another portion. This is useful for memory windows, for example the PC
810 * VGA region.
811 */
812static int kvm_vm_ioctl_set_memory_alias(struct kvm *kvm,
813 struct kvm_memory_alias *alias)
814{
815 int r, n;
816 struct kvm_mem_alias *p;
817
818 r = -EINVAL;
819 /* General sanity checks */
820 if (alias->memory_size & (PAGE_SIZE - 1))
821 goto out;
822 if (alias->guest_phys_addr & (PAGE_SIZE - 1))
823 goto out;
824 if (alias->slot >= KVM_ALIAS_SLOTS)
825 goto out;
826 if (alias->guest_phys_addr + alias->memory_size
827 < alias->guest_phys_addr)
828 goto out;
829 if (alias->target_phys_addr + alias->memory_size
830 < alias->target_phys_addr)
831 goto out;
832
11ec2804 833 mutex_lock(&kvm->lock);
e8207547
AK		 834
835 p = &kvm->aliases[alias->slot];
836 p->base_gfn = alias->guest_phys_addr >> PAGE_SHIFT;
837 p->npages = alias->memory_size >> PAGE_SHIFT;
838 p->target_gfn = alias->target_phys_addr >> PAGE_SHIFT;
839
840 for (n = KVM_ALIAS_SLOTS; n > 0; --n)
841 if (kvm->aliases[n - 1].npages)
842 break;
843 kvm->naliases = n;
844
90cb0529 845 kvm_mmu_zap_all(kvm);
e8207547 846
11ec2804 847 mutex_unlock(&kvm->lock);
e8207547
AK		 848
849 return 0;
850
851out:
852 return r;
853}
854
6ceb9d79
HQ		 855static int kvm_vm_ioctl_get_irqchip(struct kvm *kvm, struct kvm_irqchip *chip)
856{
857 int r;
858
859 r = 0;
860 switch (chip->chip_id) {
861 case KVM_IRQCHIP_PIC_MASTER:
862 memcpy (&chip->chip.pic,
863 &pic_irqchip(kvm)->pics[0],
864 sizeof(struct kvm_pic_state));
865 break;
866 case KVM_IRQCHIP_PIC_SLAVE:
867 memcpy (&chip->chip.pic,
868 &pic_irqchip(kvm)->pics[1],
869 sizeof(struct kvm_pic_state));
870 break;
6bf9e962
HQ		 871 case KVM_IRQCHIP_IOAPIC:
872 memcpy (&chip->chip.ioapic,
873 ioapic_irqchip(kvm),
874 sizeof(struct kvm_ioapic_state));
875 break;
6ceb9d79
HQ		 876 default:
877 r = -EINVAL;
878 break;
879 }
880 return r;
881}
882
883static int kvm_vm_ioctl_set_irqchip(struct kvm *kvm, struct kvm_irqchip *chip)
884{
885 int r;
886
887 r = 0;
888 switch (chip->chip_id) {
889 case KVM_IRQCHIP_PIC_MASTER:
890 memcpy (&pic_irqchip(kvm)->pics[0],
891 &chip->chip.pic,
892 sizeof(struct kvm_pic_state));
893 break;
894 case KVM_IRQCHIP_PIC_SLAVE:
895 memcpy (&pic_irqchip(kvm)->pics[1],
896 &chip->chip.pic,
897 sizeof(struct kvm_pic_state));
898 break;
6bf9e962
HQ		 899 case KVM_IRQCHIP_IOAPIC:
900 memcpy (ioapic_irqchip(kvm),
901 &chip->chip.ioapic,
902 sizeof(struct kvm_ioapic_state));
903 break;
6ceb9d79
HQ		 904 default:
905 r = -EINVAL;
906 break;
907 }
908 kvm_pic_update_irq(pic_irqchip(kvm));
909 return r;
910}
911
e8207547
AK		 912static gfn_t unalias_gfn(struct kvm *kvm, gfn_t gfn)
913{
914 int i;
915 struct kvm_mem_alias *alias;
916
917 for (i = 0; i < kvm->naliases; ++i) {
918 alias = &kvm->aliases[i];
919 if (gfn >= alias->base_gfn
920 && gfn < alias->base_gfn + alias->npages)
921 return alias->target_gfn + gfn - alias->base_gfn;
922 }
923 return gfn;
924}
925
926static struct kvm_memory_slot *__gfn_to_memslot(struct kvm *kvm, gfn_t gfn)
6aa8b732
AK		 927{
928 int i;
929
930 for (i = 0; i < kvm->nmemslots; ++i) {
931 struct kvm_memory_slot *memslot = &kvm->memslots[i];
932
933 if (gfn >= memslot->base_gfn
934 && gfn < memslot->base_gfn + memslot->npages)
935 return memslot;
936 }
8b6d44c7 937 return NULL;
6aa8b732 938}
e8207547
AK		 939
940struct kvm_memory_slot *gfn_to_memslot(struct kvm *kvm, gfn_t gfn)
941{
942 gfn = unalias_gfn(kvm, gfn);
943 return __gfn_to_memslot(kvm, gfn);
944}
6aa8b732 945
954bbbc2
AK		 946struct page *gfn_to_page(struct kvm *kvm, gfn_t gfn)
947{
948 struct kvm_memory_slot *slot;
949
e8207547
AK		 950 gfn = unalias_gfn(kvm, gfn);
951 slot = __gfn_to_memslot(kvm, gfn);
954bbbc2
AK		 952 if (!slot)
953 return NULL;
954 return slot->phys_mem[gfn - slot->base_gfn];
955}
956EXPORT_SYMBOL_GPL(gfn_to_page);
957
7e9d619d 958/* WARNING: Does not work on aliased pages. */
6aa8b732
AK		 959void mark_page_dirty(struct kvm *kvm, gfn_t gfn)
960{
31389947 961 struct kvm_memory_slot *memslot;
6aa8b732 962
7e9d619d
RR		 963 memslot = __gfn_to_memslot(kvm, gfn);
964 if (memslot && memslot->dirty_bitmap) {
965 unsigned long rel_gfn = gfn - memslot->base_gfn;
6aa8b732 966
7e9d619d
RR		 967 /* avoid RMW */
968 if (!test_bit(rel_gfn, memslot->dirty_bitmap))
969 set_bit(rel_gfn, memslot->dirty_bitmap);
6aa8b732
AK		 970 }
971}
972
e7d5d76c 973int emulator_read_std(unsigned long addr,
4c690a1e 974 void *val,
6aa8b732 975 unsigned int bytes,
cebff02b 976 struct kvm_vcpu *vcpu)
6aa8b732 977{
6aa8b732
AK		 978 void *data = val;
979
980 while (bytes) {
981 gpa_t gpa = vcpu->mmu.gva_to_gpa(vcpu, addr);
982 unsigned offset = addr & (PAGE_SIZE-1);
983 unsigned tocopy = min(bytes, (unsigned)PAGE_SIZE - offset);
984 unsigned long pfn;
954bbbc2
AK		 985 struct page *page;
986 void *page_virt;
6aa8b732
AK		 987
988 if (gpa == UNMAPPED_GVA)
989 return X86EMUL_PROPAGATE_FAULT;
990 pfn = gpa >> PAGE_SHIFT;
954bbbc2
AK		 991 page = gfn_to_page(vcpu->kvm, pfn);
992 if (!page)
6aa8b732 993 return X86EMUL_UNHANDLEABLE;
954bbbc2 994 page_virt = kmap_atomic(page, KM_USER0);
6aa8b732 995
954bbbc2 996 memcpy(data, page_virt + offset, tocopy);
6aa8b732 997
954bbbc2 998 kunmap_atomic(page_virt, KM_USER0);
6aa8b732
AK		 999
1000 bytes -= tocopy;
1001 data += tocopy;
1002 addr += tocopy;
1003 }
1004
1005 return X86EMUL_CONTINUE;
1006}
e7d5d76c 1007EXPORT_SYMBOL_GPL(emulator_read_std);
6aa8b732
AK		 1008
1009static int emulator_write_std(unsigned long addr,
4c690a1e 1010 const void *val,
6aa8b732 1011 unsigned int bytes,
cebff02b 1012 struct kvm_vcpu *vcpu)
6aa8b732 1013{
f0242478 1014 pr_unimpl(vcpu, "emulator_write_std: addr %lx n %d\n", addr, bytes);
6aa8b732
AK		 1015 return X86EMUL_UNHANDLEABLE;
1016}
1017
97222cc8
ED		 1018/*
1019 * Only apic need an MMIO device hook, so shortcut now..
1020 */
1021static struct kvm_io_device *vcpu_find_pervcpu_dev(struct kvm_vcpu *vcpu,
1022 gpa_t addr)
1023{
1024 struct kvm_io_device *dev;
1025
1026 if (vcpu->apic) {
1027 dev = &vcpu->apic->dev;
1028 if (dev->in_range(dev, addr))
1029 return dev;
1030 }
1031 return NULL;
1032}
1033
2eeb2e94
GH		 1034static struct kvm_io_device *vcpu_find_mmio_dev(struct kvm_vcpu *vcpu,
1035 gpa_t addr)
1036{
97222cc8
ED		 1037 struct kvm_io_device *dev;
1038
1039 dev = vcpu_find_pervcpu_dev(vcpu, addr);
1040 if (dev == NULL)
1041 dev = kvm_io_bus_find_dev(&vcpu->kvm->mmio_bus, addr);
1042 return dev;
2eeb2e94
GH		 1043}
1044
74906345
ED		 1045static struct kvm_io_device *vcpu_find_pio_dev(struct kvm_vcpu *vcpu,
1046 gpa_t addr)
1047{
1048 return kvm_io_bus_find_dev(&vcpu->kvm->pio_bus, addr);
1049}
1050
6aa8b732 1051static int emulator_read_emulated(unsigned long addr,
4c690a1e 1052 void *val,
6aa8b732 1053 unsigned int bytes,
cebff02b 1054 struct kvm_vcpu *vcpu)
6aa8b732 1055{
2eeb2e94
GH		 1056 struct kvm_io_device *mmio_dev;
1057 gpa_t gpa;
6aa8b732
AK		 1058
1059 if (vcpu->mmio_read_completed) {
1060 memcpy(val, vcpu->mmio_data, bytes);
1061 vcpu->mmio_read_completed = 0;
1062 return X86EMUL_CONTINUE;
cebff02b 1063 } else if (emulator_read_std(addr, val, bytes, vcpu)
6aa8b732
AK		 1064 == X86EMUL_CONTINUE)
1065 return X86EMUL_CONTINUE;
d27d4aca 1066
2eeb2e94
GH		 1067 gpa = vcpu->mmu.gva_to_gpa(vcpu, addr);
1068 if (gpa == UNMAPPED_GVA)
1069 return X86EMUL_PROPAGATE_FAULT;
6aa8b732 1070
2eeb2e94
GH		 1071 /*
1072 * Is this MMIO handled locally?
1073 */
1074 mmio_dev = vcpu_find_mmio_dev(vcpu, gpa);
1075 if (mmio_dev) {
1076 kvm_iodevice_read(mmio_dev, gpa, bytes, val);
1077 return X86EMUL_CONTINUE;
6aa8b732 1078 }
2eeb2e94
GH		 1079
1080 vcpu->mmio_needed = 1;
1081 vcpu->mmio_phys_addr = gpa;
1082 vcpu->mmio_size = bytes;
1083 vcpu->mmio_is_write = 0;
1084
1085 return X86EMUL_UNHANDLEABLE;
6aa8b732
AK		 1086}
1087
da4a00f0 1088static int emulator_write_phys(struct kvm_vcpu *vcpu, gpa_t gpa,
4c690a1e 1089 const void *val, int bytes)
da4a00f0 1090{
da4a00f0
AK		 1091 struct page *page;
1092 void *virt;
1093
1094 if (((gpa + bytes - 1) >> PAGE_SHIFT) != (gpa >> PAGE_SHIFT))
1095 return 0;
954bbbc2
AK		 1096 page = gfn_to_page(vcpu->kvm, gpa >> PAGE_SHIFT);
1097 if (!page)
da4a00f0 1098 return 0;
ab51a434 1099 mark_page_dirty(vcpu->kvm, gpa >> PAGE_SHIFT);
da4a00f0 1100 virt = kmap_atomic(page, KM_USER0);
fe551881 1101 kvm_mmu_pte_write(vcpu, gpa, val, bytes);
7cfa4b0a 1102 memcpy(virt + offset_in_page(gpa), val, bytes);
da4a00f0 1103 kunmap_atomic(virt, KM_USER0);
da4a00f0
AK		 1104 return 1;
1105}
1106
b0fcd903
AK		 1107static int emulator_write_emulated_onepage(unsigned long addr,
1108 const void *val,
1109 unsigned int bytes,
cebff02b 1110 struct kvm_vcpu *vcpu)
6aa8b732 1111{
2eeb2e94
GH		 1112 struct kvm_io_device *mmio_dev;
1113 gpa_t gpa = vcpu->mmu.gva_to_gpa(vcpu, addr);
6aa8b732 1114
c9047f53 1115 if (gpa == UNMAPPED_GVA) {
cbdd1bea 1116 kvm_x86_ops->inject_page_fault(vcpu, addr, 2);
6aa8b732 1117 return X86EMUL_PROPAGATE_FAULT;
c9047f53 1118 }
6aa8b732 1119
da4a00f0
AK		 1120 if (emulator_write_phys(vcpu, gpa, val, bytes))
1121 return X86EMUL_CONTINUE;
1122
2eeb2e94
GH		 1123 /*
1124 * Is this MMIO handled locally?
1125 */
1126 mmio_dev = vcpu_find_mmio_dev(vcpu, gpa);
1127 if (mmio_dev) {
1128 kvm_iodevice_write(mmio_dev, gpa, bytes, val);
1129 return X86EMUL_CONTINUE;
1130 }
1131
6aa8b732
AK		 1132 vcpu->mmio_needed = 1;
1133 vcpu->mmio_phys_addr = gpa;
1134 vcpu->mmio_size = bytes;
1135 vcpu->mmio_is_write = 1;
4c690a1e 1136 memcpy(vcpu->mmio_data, val, bytes);
6aa8b732
AK		 1137
1138 return X86EMUL_CONTINUE;
1139}
1140
e7d5d76c 1141int emulator_write_emulated(unsigned long addr,
b0fcd903
AK		 1142 const void *val,
1143 unsigned int bytes,
cebff02b 1144 struct kvm_vcpu *vcpu)
b0fcd903
AK		 1145{
1146 /* Crossing a page boundary? */
1147 if (((addr + bytes - 1) ^ addr) & PAGE_MASK) {
1148 int rc, now;
1149
1150 now = -addr & ~PAGE_MASK;
cebff02b 1151 rc = emulator_write_emulated_onepage(addr, val, now, vcpu);
b0fcd903
AK		 1152 if (rc != X86EMUL_CONTINUE)
1153 return rc;
1154 addr += now;
1155 val += now;
1156 bytes -= now;
1157 }
cebff02b 1158 return emulator_write_emulated_onepage(addr, val, bytes, vcpu);
b0fcd903 1159}
e7d5d76c 1160EXPORT_SYMBOL_GPL(emulator_write_emulated);
b0fcd903 1161
6aa8b732 1162static int emulator_cmpxchg_emulated(unsigned long addr,
4c690a1e
AK		 1163 const void *old,
1164 const void *new,
6aa8b732 1165 unsigned int bytes,
cebff02b 1166 struct kvm_vcpu *vcpu)
6aa8b732
AK		 1167{
1168 static int reported;
1169
1170 if (!reported) {
1171 reported = 1;
1172 printk(KERN_WARNING "kvm: emulating exchange as write\n");
1173 }
cebff02b 1174 return emulator_write_emulated(addr, new, bytes, vcpu);
6aa8b732
AK		 1175}
1176
1177static unsigned long get_segment_base(struct kvm_vcpu *vcpu, int seg)
1178{
cbdd1bea 1179 return kvm_x86_ops->get_segment_base(vcpu, seg);
6aa8b732
AK		 1180}
1181
1182int emulate_invlpg(struct kvm_vcpu *vcpu, gva_t address)
1183{
6aa8b732
AK		 1184 return X86EMUL_CONTINUE;
1185}
1186
1187int emulate_clts(struct kvm_vcpu *vcpu)
1188{
404fb881 1189 kvm_x86_ops->set_cr0(vcpu, vcpu->cr0 & ~X86_CR0_TS);
6aa8b732
AK		 1190 return X86EMUL_CONTINUE;
1191}
1192
1193int emulator_get_dr(struct x86_emulate_ctxt* ctxt, int dr, unsigned long *dest)
1194{
1195 struct kvm_vcpu *vcpu = ctxt->vcpu;
1196
1197 switch (dr) {
1198 case 0 ... 3:
cbdd1bea 1199 *dest = kvm_x86_ops->get_dr(vcpu, dr);
6aa8b732
AK		 1200 return X86EMUL_CONTINUE;
1201 default:
f0242478 1202 pr_unimpl(vcpu, "%s: unexpected dr %u\n", __FUNCTION__, dr);
6aa8b732
AK		 1203 return X86EMUL_UNHANDLEABLE;
1204 }
1205}
1206
1207int emulator_set_dr(struct x86_emulate_ctxt *ctxt, int dr, unsigned long value)
1208{
1209 unsigned long mask = (ctxt->mode == X86EMUL_MODE_PROT64) ? ~0ULL : ~0U;
1210 int exception;
1211
cbdd1bea 1212 kvm_x86_ops->set_dr(ctxt->vcpu, dr, value & mask, &exception);
6aa8b732
AK		 1213 if (exception) {
1214 /* FIXME: better handling */
1215 return X86EMUL_UNHANDLEABLE;
1216 }
1217 return X86EMUL_CONTINUE;
1218}
1219
054b1369 1220void kvm_report_emulation_failure(struct kvm_vcpu *vcpu, const char *context)
6aa8b732
AK		 1221{
1222 static int reported;
1223 u8 opcodes[4];
054b1369 1224 unsigned long rip = vcpu->rip;
6aa8b732
AK		 1225 unsigned long rip_linear;
1226
054b1369 1227 rip_linear = rip + get_segment_base(vcpu, VCPU_SREG_CS);
6aa8b732
AK		 1228
1229 if (reported)
1230 return;
1231
054b1369 1232 emulator_read_std(rip_linear, (void *)opcodes, 4, vcpu);
6aa8b732 1233
054b1369
AK		 1234 printk(KERN_ERR "emulation failed (%s) rip %lx %02x %02x %02x %02x\n",
1235 context, rip, opcodes[0], opcodes[1], opcodes[2], opcodes[3]);
6aa8b732
AK		 1236 reported = 1;
1237}
054b1369 1238EXPORT_SYMBOL_GPL(kvm_report_emulation_failure);
6aa8b732
AK		 1239
1240struct x86_emulate_ops emulate_ops = {
1241 .read_std = emulator_read_std,
1242 .write_std = emulator_write_std,
1243 .read_emulated = emulator_read_emulated,
1244 .write_emulated = emulator_write_emulated,
1245 .cmpxchg_emulated = emulator_cmpxchg_emulated,
1246};
1247
1248int emulate_instruction(struct kvm_vcpu *vcpu,
1249 struct kvm_run *run,
1250 unsigned long cr2,
3427318f
LV		 1251 u16 error_code,
1252 int no_decode)
6aa8b732 1253{
a22436b7 1254 int r;
6aa8b732 1255
e7df56e4 1256 vcpu->mmio_fault_cr2 = cr2;
cbdd1bea 1257 kvm_x86_ops->cache_regs(vcpu);
6aa8b732 1258
6aa8b732 1259 vcpu->mmio_is_write = 0;
e70669ab 1260 vcpu->pio.string = 0;
3427318f
LV		 1261
1262 if (!no_decode) {
1263 int cs_db, cs_l;
1264 kvm_x86_ops->get_cs_db_l_bits(vcpu, &cs_db, &cs_l);
1265
1266 vcpu->emulate_ctxt.vcpu = vcpu;
1267 vcpu->emulate_ctxt.eflags = kvm_x86_ops->get_rflags(vcpu);
1268 vcpu->emulate_ctxt.cr2 = cr2;
1269 vcpu->emulate_ctxt.mode =
1270 (vcpu->emulate_ctxt.eflags & X86_EFLAGS_VM)
1271 ? X86EMUL_MODE_REAL : cs_l
1272 ? X86EMUL_MODE_PROT64 : cs_db
1273 ? X86EMUL_MODE_PROT32 : X86EMUL_MODE_PROT16;
1274
1275 if (vcpu->emulate_ctxt.mode == X86EMUL_MODE_PROT64) {
1276 vcpu->emulate_ctxt.cs_base = 0;
1277 vcpu->emulate_ctxt.ds_base = 0;
1278 vcpu->emulate_ctxt.es_base = 0;
1279 vcpu->emulate_ctxt.ss_base = 0;
1280 } else {
1281 vcpu->emulate_ctxt.cs_base =
1282 get_segment_base(vcpu, VCPU_SREG_CS);
1283 vcpu->emulate_ctxt.ds_base =
1284 get_segment_base(vcpu, VCPU_SREG_DS);
1285 vcpu->emulate_ctxt.es_base =
1286 get_segment_base(vcpu, VCPU_SREG_ES);
1287 vcpu->emulate_ctxt.ss_base =
1288 get_segment_base(vcpu, VCPU_SREG_SS);
1289 }
1290
1291 vcpu->emulate_ctxt.gs_base =
1292 get_segment_base(vcpu, VCPU_SREG_GS);
1293 vcpu->emulate_ctxt.fs_base =
1294 get_segment_base(vcpu, VCPU_SREG_FS);
1295
1296 r = x86_decode_insn(&vcpu->emulate_ctxt, &emulate_ops);
a22436b7
LV		 1297 if (r) {
1298 if (kvm_mmu_unprotect_page_virt(vcpu, cr2))
1299 return EMULATE_DONE;
1300 return EMULATE_FAIL;
1301 }
3427318f
LV		 1302 }
1303
a22436b7 1304 r = x86_emulate_insn(&vcpu->emulate_ctxt, &emulate_ops);
1be3aa47 1305
e70669ab
LV		 1306 if (vcpu->pio.string)
1307 return EMULATE_DO_MMIO;
6aa8b732
AK		 1308
1309 if ((r || vcpu->mmio_is_write) && run) {
8fc0d085 1310 run->exit_reason = KVM_EXIT_MMIO;
6aa8b732
AK		 1311 run->mmio.phys_addr = vcpu->mmio_phys_addr;
1312 memcpy(run->mmio.data, vcpu->mmio_data, 8);
1313 run->mmio.len = vcpu->mmio_size;
1314 run->mmio.is_write = vcpu->mmio_is_write;
1315 }
1316
1317 if (r) {
a436036b
AK		 1318 if (kvm_mmu_unprotect_page_virt(vcpu, cr2))
1319 return EMULATE_DONE;
6aa8b732 1320 if (!vcpu->mmio_needed) {
054b1369 1321 kvm_report_emulation_failure(vcpu, "mmio");
6aa8b732
AK		 1322 return EMULATE_FAIL;
1323 }
1324 return EMULATE_DO_MMIO;
1325 }
1326
cbdd1bea 1327 kvm_x86_ops->decache_regs(vcpu);
3427318f 1328 kvm_x86_ops->set_rflags(vcpu, vcpu->emulate_ctxt.eflags);
6aa8b732 1329
02c83209
AK		 1330 if (vcpu->mmio_is_write) {
1331 vcpu->mmio_needed = 0;
6aa8b732 1332 return EMULATE_DO_MMIO;
02c83209 1333 }
6aa8b732
AK		 1334
1335 return EMULATE_DONE;
1336}
1337EXPORT_SYMBOL_GPL(emulate_instruction);
1338
b6958ce4
ED		 1339/*
1340 * The vCPU has executed a HLT instruction with in-kernel mode enabled.
1341 */
c5ec1534 1342static void kvm_vcpu_block(struct kvm_vcpu *vcpu)
d3bef15f 1343{
b6958ce4
ED		 1344 DECLARE_WAITQUEUE(wait, current);
1345
1346 add_wait_queue(&vcpu->wq, &wait);
1347
1348 /*
1349 * We will block until either an interrupt or a signal wakes us up
1350 */
c5ec1534
HQ		 1351 while (!kvm_cpu_has_interrupt(vcpu)
1352 && !signal_pending(current)
1353 && vcpu->mp_state != VCPU_MP_STATE_RUNNABLE
1354 && vcpu->mp_state != VCPU_MP_STATE_SIPI_RECEIVED) {
b6958ce4
ED		 1355 set_current_state(TASK_INTERRUPTIBLE);
1356 vcpu_put(vcpu);
1357 schedule();
1358 vcpu_load(vcpu);
1359 }
d3bef15f 1360
c5ec1534 1361 __set_current_state(TASK_RUNNING);
b6958ce4 1362 remove_wait_queue(&vcpu->wq, &wait);
b6958ce4
ED		 1363}
1364
1365int kvm_emulate_halt(struct kvm_vcpu *vcpu)
1366{
d3bef15f 1367 ++vcpu->stat.halt_exits;
b6958ce4 1368 if (irqchip_in_kernel(vcpu->kvm)) {
c5ec1534
HQ		 1369 vcpu->mp_state = VCPU_MP_STATE_HALTED;
1370 kvm_vcpu_block(vcpu);
1371 if (vcpu->mp_state != VCPU_MP_STATE_RUNNABLE)
1372 return -EINTR;
b6958ce4
ED		 1373 return 1;
1374 } else {
1375 vcpu->run->exit_reason = KVM_EXIT_HLT;
1376 return 0;
1377 }
d3bef15f
AK		 1378}
1379EXPORT_SYMBOL_GPL(kvm_emulate_halt);
1380
7aa81cc0 1381int kvm_emulate_hypercall(struct kvm_vcpu *vcpu)
270fd9b9 1382{
7aa81cc0 1383 unsigned long nr, a0, a1, a2, a3, ret;
270fd9b9 1384
cbdd1bea 1385 kvm_x86_ops->cache_regs(vcpu);
7aa81cc0
AL		 1386
1387 nr = vcpu->regs[VCPU_REGS_RAX];
1388 a0 = vcpu->regs[VCPU_REGS_RBX];
1389 a1 = vcpu->regs[VCPU_REGS_RCX];
1390 a2 = vcpu->regs[VCPU_REGS_RDX];
1391 a3 = vcpu->regs[VCPU_REGS_RSI];
1392
1393 if (!is_long_mode(vcpu)) {
1394 nr &= 0xFFFFFFFF;
1395 a0 &= 0xFFFFFFFF;
1396 a1 &= 0xFFFFFFFF;
1397 a2 &= 0xFFFFFFFF;
1398 a3 &= 0xFFFFFFFF;
270fd9b9 1399 }
7aa81cc0 1400
270fd9b9
AK		 1401 switch (nr) {
1402 default:
7aa81cc0
AL		 1403 ret = -KVM_ENOSYS;
1404 break;
270fd9b9
AK		 1405 }
1406 vcpu->regs[VCPU_REGS_RAX] = ret;
cbdd1bea 1407 kvm_x86_ops->decache_regs(vcpu);
7aa81cc0
AL		 1408 return 0;
1409}
1410EXPORT_SYMBOL_GPL(kvm_emulate_hypercall);
1411
1412int kvm_fix_hypercall(struct kvm_vcpu *vcpu)
1413{
1414 char instruction[3];
1415 int ret = 0;
1416
1417 mutex_lock(&vcpu->kvm->lock);
1418
1419 /*
1420 * Blow out the MMU to ensure that no other VCPU has an active mapping
1421 * to ensure that the updated hypercall appears atomically across all
1422 * VCPUs.
1423 */
1424 kvm_mmu_zap_all(vcpu->kvm);
1425
1426 kvm_x86_ops->cache_regs(vcpu);
1427 kvm_x86_ops->patch_hypercall(vcpu, instruction);
1428 if (emulator_write_emulated(vcpu->rip, instruction, 3, vcpu)
1429 != X86EMUL_CONTINUE)
1430 ret = -EFAULT;
1431
1432 mutex_unlock(&vcpu->kvm->lock);
1433
1434 return ret;
270fd9b9 1435}
270fd9b9 1436
6aa8b732
AK		 1437static u64 mk_cr_64(u64 curr_cr, u32 new_val)
1438{
1439 return (curr_cr & ~((1ULL << 32) - 1)) | new_val;
1440}
1441
1442void realmode_lgdt(struct kvm_vcpu *vcpu, u16 limit, unsigned long base)
1443{
1444 struct descriptor_table dt = { limit, base };
1445
cbdd1bea 1446 kvm_x86_ops->set_gdt(vcpu, &dt);
6aa8b732
AK		 1447}
1448
1449void realmode_lidt(struct kvm_vcpu *vcpu, u16 limit, unsigned long base)
1450{
1451 struct descriptor_table dt = { limit, base };
1452
cbdd1bea 1453 kvm_x86_ops->set_idt(vcpu, &dt);
6aa8b732
AK		 1454}
1455
1456void realmode_lmsw(struct kvm_vcpu *vcpu, unsigned long msw,
1457 unsigned long *rflags)
1458{
1459 lmsw(vcpu, msw);
cbdd1bea 1460 *rflags = kvm_x86_ops->get_rflags(vcpu);
6aa8b732
AK		 1461}
1462
1463unsigned long realmode_get_cr(struct kvm_vcpu *vcpu, int cr)
1464{
cbdd1bea 1465 kvm_x86_ops->decache_cr4_guest_bits(vcpu);
6aa8b732
AK		 1466 switch (cr) {
1467 case 0:
1468 return vcpu->cr0;
1469 case 2:
1470 return vcpu->cr2;
1471 case 3:
1472 return vcpu->cr3;
1473 case 4:
1474 return vcpu->cr4;
1475 default:
1476 vcpu_printf(vcpu, "%s: unexpected cr %u\n", __FUNCTION__, cr);
1477 return 0;
1478 }
1479}
1480
1481void realmode_set_cr(struct kvm_vcpu *vcpu, int cr, unsigned long val,
1482 unsigned long *rflags)
1483{
1484 switch (cr) {
1485 case 0:
1486 set_cr0(vcpu, mk_cr_64(vcpu->cr0, val));
cbdd1bea 1487 *rflags = kvm_x86_ops->get_rflags(vcpu);
6aa8b732
AK		 1488 break;
1489 case 2:
1490 vcpu->cr2 = val;
1491 break;
1492 case 3:
1493 set_cr3(vcpu, val);
1494 break;
1495 case 4:
1496 set_cr4(vcpu, mk_cr_64(vcpu->cr4, val));
1497 break;
1498 default:
1499 vcpu_printf(vcpu, "%s: unexpected cr %u\n", __FUNCTION__, cr);
1500 }
1501}
1502
3bab1f5d
AK		 1503int kvm_get_msr_common(struct kvm_vcpu *vcpu, u32 msr, u64 *pdata)
1504{
1505 u64 data;
1506
1507 switch (msr) {
1508 case 0xc0010010: /* SYSCFG */
1509 case 0xc0010015: /* HWCR */
1510 case MSR_IA32_PLATFORM_ID:
1511 case MSR_IA32_P5_MC_ADDR:
1512 case MSR_IA32_P5_MC_TYPE:
1513 case MSR_IA32_MC0_CTL:
1514 case MSR_IA32_MCG_STATUS:
1515 case MSR_IA32_MCG_CAP:
1516 case MSR_IA32_MC0_MISC:
1517 case MSR_IA32_MC0_MISC+4:
1518 case MSR_IA32_MC0_MISC+8:
1519 case MSR_IA32_MC0_MISC+12:
1520 case MSR_IA32_MC0_MISC+16:
1521 case MSR_IA32_UCODE_REV:
a8d13ea2 1522 case MSR_IA32_PERF_STATUS:
2dc7094b 1523 case MSR_IA32_EBL_CR_POWERON:
3bab1f5d
AK		 1524 /* MTRR registers */
1525 case 0xfe:
1526 case 0x200 ... 0x2ff:
1527 data = 0;
1528 break;
a8d13ea2
AK		 1529 case 0xcd: /* fsb frequency */
1530 data = 3;
1531 break;
3bab1f5d 1532 case MSR_IA32_APICBASE:
7017fc3d 1533 data = kvm_get_apic_base(vcpu);
3bab1f5d 1534 break;
6f00e68f
AK		 1535 case MSR_IA32_MISC_ENABLE:
1536 data = vcpu->ia32_misc_enable_msr;
1537 break;
3bab1f5d
AK		 1538#ifdef CONFIG_X86_64
1539 case MSR_EFER:
1540 data = vcpu->shadow_efer;
1541 break;
1542#endif
1543 default:
f0242478 1544 pr_unimpl(vcpu, "unhandled rdmsr: 0x%x\n", msr);
3bab1f5d
AK		 1545 return 1;
1546 }
1547 *pdata = data;
1548 return 0;
1549}
1550EXPORT_SYMBOL_GPL(kvm_get_msr_common);
1551
6aa8b732
AK		 1552/*
1553 * Reads an msr value (of 'msr_index') into 'pdata'.
1554 * Returns 0 on success, non-0 otherwise.
1555 * Assumes vcpu_load() was already called.
1556 */
35f3f286 1557int kvm_get_msr(struct kvm_vcpu *vcpu, u32 msr_index, u64 *pdata)
6aa8b732 1558{
cbdd1bea 1559 return kvm_x86_ops->get_msr(vcpu, msr_index, pdata);
6aa8b732
AK		 1560}
1561
05b3e0c2 1562#ifdef CONFIG_X86_64
6aa8b732 1563
3bab1f5d 1564static void set_efer(struct kvm_vcpu *vcpu, u64 efer)
6aa8b732 1565{
6aa8b732
AK		 1566 if (efer & EFER_RESERVED_BITS) {
1567 printk(KERN_DEBUG "set_efer: 0x%llx #GP, reserved bits\n",
1568 efer);
1569 inject_gp(vcpu);
1570 return;
1571 }
1572
1573 if (is_paging(vcpu)
1574 && (vcpu->shadow_efer & EFER_LME) != (efer & EFER_LME)) {
1575 printk(KERN_DEBUG "set_efer: #GP, change LME while paging\n");
1576 inject_gp(vcpu);
1577 return;
1578 }
1579
cbdd1bea 1580 kvm_x86_ops->set_efer(vcpu, efer);
7725f0ba 1581
6aa8b732
AK		 1582 efer &= ~EFER_LMA;
1583 efer |= vcpu->shadow_efer & EFER_LMA;
1584
1585 vcpu->shadow_efer = efer;
6aa8b732 1586}
6aa8b732
AK		 1587
1588#endif
1589
3bab1f5d
AK		 1590int kvm_set_msr_common(struct kvm_vcpu *vcpu, u32 msr, u64 data)
1591{
1592 switch (msr) {
1593#ifdef CONFIG_X86_64
1594 case MSR_EFER:
1595 set_efer(vcpu, data);
1596 break;
1597#endif
1598 case MSR_IA32_MC0_STATUS:
f0242478 1599 pr_unimpl(vcpu, "%s: MSR_IA32_MC0_STATUS 0x%llx, nop\n",
3bab1f5d
AK		 1600 __FUNCTION__, data);
1601 break;
0e5bf0d0 1602 case MSR_IA32_MCG_STATUS:
f0242478 1603 pr_unimpl(vcpu, "%s: MSR_IA32_MCG_STATUS 0x%llx, nop\n",
0e5bf0d0
SK		 1604 __FUNCTION__, data);
1605 break;
3bab1f5d
AK		 1606 case MSR_IA32_UCODE_REV:
1607 case MSR_IA32_UCODE_WRITE:
1608 case 0x200 ... 0x2ff: /* MTRRs */
1609 break;
1610 case MSR_IA32_APICBASE:
7017fc3d 1611 kvm_set_apic_base(vcpu, data);
3bab1f5d 1612 break;
6f00e68f
AK		 1613 case MSR_IA32_MISC_ENABLE:
1614 vcpu->ia32_misc_enable_msr = data;
1615 break;
3bab1f5d 1616 default:
f0242478 1617 pr_unimpl(vcpu, "unhandled wrmsr: 0x%x\n", msr);
3bab1f5d
AK		 1618 return 1;
1619 }
1620 return 0;
1621}
1622EXPORT_SYMBOL_GPL(kvm_set_msr_common);
1623
6aa8b732
AK		 1624/*
1625 * Writes msr value into into the appropriate "register".
1626 * Returns 0 on success, non-0 otherwise.
1627 * Assumes vcpu_load() was already called.
1628 */
35f3f286 1629int kvm_set_msr(struct kvm_vcpu *vcpu, u32 msr_index, u64 data)
6aa8b732 1630{
cbdd1bea 1631 return kvm_x86_ops->set_msr(vcpu, msr_index, data);
6aa8b732
AK		 1632}
1633
1634void kvm_resched(struct kvm_vcpu *vcpu)
1635{
3fca0365
YD		 1636 if (!need_resched())
1637 return;
6aa8b732 1638 cond_resched();
6aa8b732
AK		 1639}
1640EXPORT_SYMBOL_GPL(kvm_resched);
1641
06465c5a
AK		 1642void kvm_emulate_cpuid(struct kvm_vcpu *vcpu)
1643{
1644 int i;
1645 u32 function;
1646 struct kvm_cpuid_entry *e, *best;
1647
cbdd1bea 1648 kvm_x86_ops->cache_regs(vcpu);
06465c5a
AK		 1649 function = vcpu->regs[VCPU_REGS_RAX];
1650 vcpu->regs[VCPU_REGS_RAX] = 0;
1651 vcpu->regs[VCPU_REGS_RBX] = 0;
1652 vcpu->regs[VCPU_REGS_RCX] = 0;
1653 vcpu->regs[VCPU_REGS_RDX] = 0;
1654 best = NULL;
1655 for (i = 0; i < vcpu->cpuid_nent; ++i) {
1656 e = &vcpu->cpuid_entries[i];
1657 if (e->function == function) {
1658 best = e;
1659 break;
1660 }
1661 /*
1662 * Both basic or both extended?
1663 */
1664 if (((e->function ^ function) & 0x80000000) == 0)
1665 if (!best || e->function > best->function)
1666 best = e;
1667 }
1668 if (best) {
1669 vcpu->regs[VCPU_REGS_RAX] = best->eax;
1670 vcpu->regs[VCPU_REGS_RBX] = best->ebx;
1671 vcpu->regs[VCPU_REGS_RCX] = best->ecx;
1672 vcpu->regs[VCPU_REGS_RDX] = best->edx;
1673 }
cbdd1bea
CE		 1674 kvm_x86_ops->decache_regs(vcpu);
1675 kvm_x86_ops->skip_emulated_instruction(vcpu);
06465c5a
AK		 1676}
1677EXPORT_SYMBOL_GPL(kvm_emulate_cpuid);
1678
039576c0 1679static int pio_copy_data(struct kvm_vcpu *vcpu)
46fc1477 1680{
039576c0
AK		 1681 void *p = vcpu->pio_data;
1682 void *q;
1683 unsigned bytes;
1684 int nr_pages = vcpu->pio.guest_pages[1] ? 2 : 1;
1685
039576c0
AK		 1686 q = vmap(vcpu->pio.guest_pages, nr_pages, VM_READ|VM_WRITE,
1687 PAGE_KERNEL);
1688 if (!q) {
039576c0
AK		 1689 free_pio_guest_pages(vcpu);
1690 return -ENOMEM;
1691 }
1692 q += vcpu->pio.guest_page_offset;
1693 bytes = vcpu->pio.size * vcpu->pio.cur_count;
1694 if (vcpu->pio.in)
1695 memcpy(q, p, bytes);
1696 else
1697 memcpy(p, q, bytes);
1698 q -= vcpu->pio.guest_page_offset;
1699 vunmap(q);
039576c0
AK		 1700 free_pio_guest_pages(vcpu);
1701 return 0;
1702}
1703
1704static int complete_pio(struct kvm_vcpu *vcpu)
1705{
1706 struct kvm_pio_request *io = &vcpu->pio;
46fc1477 1707 long delta;
039576c0 1708 int r;
46fc1477 1709
cbdd1bea 1710 kvm_x86_ops->cache_regs(vcpu);
46fc1477
AK		 1711
1712 if (!io->string) {
039576c0
AK		 1713 if (io->in)
1714 memcpy(&vcpu->regs[VCPU_REGS_RAX], vcpu->pio_data,
46fc1477
AK		 1715 io->size);
1716 } else {
039576c0
AK		 1717 if (io->in) {
1718 r = pio_copy_data(vcpu);
1719 if (r) {
cbdd1bea 1720 kvm_x86_ops->cache_regs(vcpu);
039576c0
AK		 1721 return r;
1722 }
1723 }
1724
46fc1477
AK		 1725 delta = 1;
1726 if (io->rep) {
039576c0 1727 delta *= io->cur_count;
46fc1477
AK		 1728 /*
1729 * The size of the register should really depend on
1730 * current address size.
1731 */
1732 vcpu->regs[VCPU_REGS_RCX] -= delta;
1733 }
039576c0 1734 if (io->down)
46fc1477
AK		 1735 delta = -delta;
1736 delta *= io->size;
039576c0 1737 if (io->in)
46fc1477
AK		 1738 vcpu->regs[VCPU_REGS_RDI] += delta;
1739 else
1740 vcpu->regs[VCPU_REGS_RSI] += delta;
1741 }
1742
cbdd1bea 1743 kvm_x86_ops->decache_regs(vcpu);
46fc1477 1744
039576c0
AK		 1745 io->count -= io->cur_count;
1746 io->cur_count = 0;
1747
039576c0 1748 return 0;
46fc1477
AK		 1749}
1750
65619eb5
ED		 1751static void kernel_pio(struct kvm_io_device *pio_dev,
1752 struct kvm_vcpu *vcpu,
1753 void *pd)
74906345
ED		 1754{
1755 /* TODO: String I/O for in kernel device */
1756
9cf98828 1757 mutex_lock(&vcpu->kvm->lock);
74906345
ED		 1758 if (vcpu->pio.in)
1759 kvm_iodevice_read(pio_dev, vcpu->pio.port,
1760 vcpu->pio.size,
65619eb5 1761 pd);
74906345
ED		 1762 else
1763 kvm_iodevice_write(pio_dev, vcpu->pio.port,
1764 vcpu->pio.size,
65619eb5 1765 pd);
9cf98828 1766 mutex_unlock(&vcpu->kvm->lock);
65619eb5
ED		 1767}
1768
1769static void pio_string_write(struct kvm_io_device *pio_dev,
1770 struct kvm_vcpu *vcpu)
1771{
1772 struct kvm_pio_request *io = &vcpu->pio;
1773 void *pd = vcpu->pio_data;
1774 int i;
1775
9cf98828 1776 mutex_lock(&vcpu->kvm->lock);
65619eb5
ED		 1777 for (i = 0; i < io->cur_count; i++) {
1778 kvm_iodevice_write(pio_dev, io->port,
1779 io->size,
1780 pd);
1781 pd += io->size;
1782 }
9cf98828 1783 mutex_unlock(&vcpu->kvm->lock);
74906345
ED		 1784}
1785
3090dd73
LV		 1786int kvm_emulate_pio (struct kvm_vcpu *vcpu, struct kvm_run *run, int in,
1787 int size, unsigned port)
1788{
1789 struct kvm_io_device *pio_dev;
1790
1791 vcpu->run->exit_reason = KVM_EXIT_IO;
1792 vcpu->run->io.direction = in ? KVM_EXIT_IO_IN : KVM_EXIT_IO_OUT;
1793 vcpu->run->io.size = vcpu->pio.size = size;
1794 vcpu->run->io.data_offset = KVM_PIO_PAGE_OFFSET * PAGE_SIZE;
1795 vcpu->run->io.count = vcpu->pio.count = vcpu->pio.cur_count = 1;
1796 vcpu->run->io.port = vcpu->pio.port = port;
1797 vcpu->pio.in = in;
1798 vcpu->pio.string = 0;
1799 vcpu->pio.down = 0;
1800 vcpu->pio.guest_page_offset = 0;
1801 vcpu->pio.rep = 0;
1802
cbdd1bea 1803 kvm_x86_ops->cache_regs(vcpu);
3090dd73 1804 memcpy(vcpu->pio_data, &vcpu->regs[VCPU_REGS_RAX], 4);
cbdd1bea 1805 kvm_x86_ops->decache_regs(vcpu);
3090dd73 1806
0967b7bf
AK		 1807 kvm_x86_ops->skip_emulated_instruction(vcpu);
1808
3090dd73
LV		 1809 pio_dev = vcpu_find_pio_dev(vcpu, port);
1810 if (pio_dev) {
1811 kernel_pio(pio_dev, vcpu, vcpu->pio_data);
1812 complete_pio(vcpu);
1813 return 1;
1814 }
1815 return 0;
1816}
1817EXPORT_SYMBOL_GPL(kvm_emulate_pio);
1818
1819int kvm_emulate_pio_string(struct kvm_vcpu *vcpu, struct kvm_run *run, int in,
1820 int size, unsigned long count, int down,
039576c0
AK		 1821 gva_t address, int rep, unsigned port)
1822{
1823 unsigned now, in_page;
65619eb5 1824 int i, ret = 0;
039576c0
AK		 1825 int nr_pages = 1;
1826 struct page *page;
74906345 1827 struct kvm_io_device *pio_dev;
039576c0
AK		 1828
1829 vcpu->run->exit_reason = KVM_EXIT_IO;
1830 vcpu->run->io.direction = in ? KVM_EXIT_IO_IN : KVM_EXIT_IO_OUT;
3090dd73 1831 vcpu->run->io.size = vcpu->pio.size = size;
039576c0 1832 vcpu->run->io.data_offset = KVM_PIO_PAGE_OFFSET * PAGE_SIZE;
3090dd73
LV		 1833 vcpu->run->io.count = vcpu->pio.count = vcpu->pio.cur_count = count;
1834 vcpu->run->io.port = vcpu->pio.port = port;
039576c0 1835 vcpu->pio.in = in;
3090dd73 1836 vcpu->pio.string = 1;
039576c0
AK		 1837 vcpu->pio.down = down;
1838 vcpu->pio.guest_page_offset = offset_in_page(address);
1839 vcpu->pio.rep = rep;
1840
039576c0 1841 if (!count) {
cbdd1bea 1842 kvm_x86_ops->skip_emulated_instruction(vcpu);
039576c0
AK		 1843 return 1;
1844 }
1845
039576c0
AK		 1846 if (!down)
1847 in_page = PAGE_SIZE - offset_in_page(address);
1848 else
1849 in_page = offset_in_page(address) + size;
1850 now = min(count, (unsigned long)in_page / size);
1851 if (!now) {
1852 /*
1853 * String I/O straddles page boundary. Pin two guest pages
1854 * so that we satisfy atomicity constraints. Do just one
1855 * transaction to avoid complexity.
1856 */
1857 nr_pages = 2;
1858 now = 1;
1859 }
1860 if (down) {
1861 /*
1862 * String I/O in reverse. Yuck. Kill the guest, fix later.
1863 */
f0242478 1864 pr_unimpl(vcpu, "guest string pio down\n");
039576c0
AK		 1865 inject_gp(vcpu);
1866 return 1;
1867 }
1868 vcpu->run->io.count = now;
1869 vcpu->pio.cur_count = now;
1870
0967b7bf
AK		 1871 if (vcpu->pio.cur_count == vcpu->pio.count)
1872 kvm_x86_ops->skip_emulated_instruction(vcpu);
1873
039576c0 1874 for (i = 0; i < nr_pages; ++i) {
11ec2804 1875 mutex_lock(&vcpu->kvm->lock);
039576c0
AK		 1876 page = gva_to_page(vcpu, address + i * PAGE_SIZE);
1877 if (page)
1878 get_page(page);
1879 vcpu->pio.guest_pages[i] = page;
11ec2804 1880 mutex_unlock(&vcpu->kvm->lock);
039576c0
AK		 1881 if (!page) {
1882 inject_gp(vcpu);
1883 free_pio_guest_pages(vcpu);
1884 return 1;
1885 }
1886 }
1887
3090dd73 1888 pio_dev = vcpu_find_pio_dev(vcpu, port);
65619eb5
ED		 1889 if (!vcpu->pio.in) {
1890 /* string PIO write */
1891 ret = pio_copy_data(vcpu);
1892 if (ret >= 0 && pio_dev) {
1893 pio_string_write(pio_dev, vcpu);
1894 complete_pio(vcpu);
1895 if (vcpu->pio.count == 0)
1896 ret = 1;
1897 }
1898 } else if (pio_dev)
f0242478 1899 pr_unimpl(vcpu, "no string pio read support yet, "
65619eb5
ED		 1900 "port %x size %d count %ld\n",
1901 port, size, count);
1902
1903 return ret;
039576c0 1904}
3090dd73 1905EXPORT_SYMBOL_GPL(kvm_emulate_pio_string);
039576c0 1906
04d2cc77
AK		 1907/*
1908 * Check if userspace requested an interrupt window, and that the
1909 * interrupt window is open.
1910 *
1911 * No need to exit to userspace if we already have an interrupt queued.
1912 */
1913static int dm_request_for_irq_injection(struct kvm_vcpu *vcpu,
1914 struct kvm_run *kvm_run)
1915{
1916 return (!vcpu->irq_summary &&
1917 kvm_run->request_interrupt_window &&
1918 vcpu->interrupt_window_open &&
1919 (kvm_x86_ops->get_rflags(vcpu) & X86_EFLAGS_IF));
1920}
1921
1922static void post_kvm_run_save(struct kvm_vcpu *vcpu,
1923 struct kvm_run *kvm_run)
1924{
1925 kvm_run->if_flag = (kvm_x86_ops->get_rflags(vcpu) & X86_EFLAGS_IF) != 0;
1926 kvm_run->cr8 = get_cr8(vcpu);
1927 kvm_run->apic_base = kvm_get_apic_base(vcpu);
1928 if (irqchip_in_kernel(vcpu->kvm))
1929 kvm_run->ready_for_interrupt_injection = 1;
1930 else
1931 kvm_run->ready_for_interrupt_injection =
1932 (vcpu->interrupt_window_open &&
1933 vcpu->irq_summary == 0);
1934}
1935
1936static int __vcpu_run(struct kvm_vcpu *vcpu, struct kvm_run *kvm_run)
1937{
1938 int r;
1939
1940 if (unlikely(vcpu->mp_state == VCPU_MP_STATE_SIPI_RECEIVED)) {
1941 printk("vcpu %d received sipi with vector # %x\n",
1942 vcpu->vcpu_id, vcpu->sipi_vector);
1943 kvm_lapic_reset(vcpu);
1944 kvm_x86_ops->vcpu_reset(vcpu);
1945 vcpu->mp_state = VCPU_MP_STATE_RUNNABLE;
1946 }
1947
1948preempted:
1949 if (vcpu->guest_debug.enabled)
1950 kvm_x86_ops->guest_debug_pre(vcpu);
1951
1952again:
1953 r = kvm_mmu_reload(vcpu);
1954 if (unlikely(r))
1955 goto out;
1956
1957 preempt_disable();
1958
1959 kvm_x86_ops->prepare_guest_switch(vcpu);
1960 kvm_load_guest_fpu(vcpu);
1961
1962 local_irq_disable();
1963
1964 if (signal_pending(current)) {
1965 local_irq_enable();
1966 preempt_enable();
1967 r = -EINTR;
1968 kvm_run->exit_reason = KVM_EXIT_INTR;
1969 ++vcpu->stat.signal_exits;
1970 goto out;
1971 }
1972
1973 if (irqchip_in_kernel(vcpu->kvm))
1974 kvm_x86_ops->inject_pending_irq(vcpu);
1975 else if (!vcpu->mmio_read_completed)
1976 kvm_x86_ops->inject_pending_vectors(vcpu, kvm_run);
1977
1978 vcpu->guest_mode = 1;
d172fcd3 1979 kvm_guest_enter();
04d2cc77
AK		 1980
1981 if (vcpu->requests)
1982 if (test_and_clear_bit(KVM_TLB_FLUSH, &vcpu->requests))
1983 kvm_x86_ops->tlb_flush(vcpu);
1984
1985 kvm_x86_ops->run(vcpu, kvm_run);
1986
1987 vcpu->guest_mode = 0;
1988 local_irq_enable();
1989
1990 ++vcpu->stat.exits;
1991
0552f73b
LV		 1992 /*
1993 * We must have an instruction between local_irq_enable() and
1994 * kvm_guest_exit(), so the timer interrupt isn't delayed by
1995 * the interrupt shadow. The stat.exits increment will do nicely.
1996 * But we need to prevent reordering, hence this barrier():
1997 */
1998 barrier();
1999
2000 kvm_guest_exit();
2001
04d2cc77
AK		 2002 preempt_enable();
2003
2004 /*
2005 * Profile KVM exit RIPs:
2006 */
2007 if (unlikely(prof_on == KVM_PROFILING)) {
2008 kvm_x86_ops->cache_regs(vcpu);
2009 profile_hit(KVM_PROFILING, (void *)vcpu->rip);
2010 }
2011
2012 r = kvm_x86_ops->handle_exit(kvm_run, vcpu);
2013
2014 if (r > 0) {
2015 if (dm_request_for_irq_injection(vcpu, kvm_run)) {
2016 r = -EINTR;
2017 kvm_run->exit_reason = KVM_EXIT_INTR;
2018 ++vcpu->stat.request_irq_exits;
2019 goto out;
2020 }
2021 if (!need_resched()) {
2022 ++vcpu->stat.light_exits;
2023 goto again;
2024 }
2025 }
2026
2027out:
2028 if (r > 0) {
2029 kvm_resched(vcpu);
2030 goto preempted;
2031 }
2032
2033 post_kvm_run_save(vcpu, kvm_run);
2034
2035 return r;
2036}
2037
2038
bccf2150 2039static int kvm_vcpu_ioctl_run(struct kvm_vcpu *vcpu, struct kvm_run *kvm_run)
6aa8b732 2040{
6aa8b732 2041 int r;
1961d276 2042 sigset_t sigsaved;
6aa8b732 2043
bccf2150 2044 vcpu_load(vcpu);
6aa8b732 2045
c5ec1534
HQ		 2046 if (unlikely(vcpu->mp_state == VCPU_MP_STATE_UNINITIALIZED)) {
2047 kvm_vcpu_block(vcpu);
2048 vcpu_put(vcpu);
2049 return -EAGAIN;
2050 }
2051
1961d276
AK		 2052 if (vcpu->sigset_active)
2053 sigprocmask(SIG_SETMASK, &vcpu->sigset, &sigsaved);
2054
54810342 2055 /* re-sync apic's tpr */
5cd4f6fd
HQ		 2056 if (!irqchip_in_kernel(vcpu->kvm))
2057 set_cr8(vcpu, kvm_run->cr8);
54810342 2058
02c83209
AK		 2059 if (vcpu->pio.cur_count) {
2060 r = complete_pio(vcpu);
2061 if (r)
2062 goto out;
2063 }
2064
2065 if (vcpu->mmio_needed) {
2066 memcpy(vcpu->mmio_data, kvm_run->mmio.data, 8);
2067 vcpu->mmio_read_completed = 1;
2068 vcpu->mmio_needed = 0;
2069 r = emulate_instruction(vcpu, kvm_run,
3427318f 2070 vcpu->mmio_fault_cr2, 0, 1);
02c83209
AK		 2071 if (r == EMULATE_DO_MMIO) {
2072 /*
2073 * Read-modify-write. Back to userspace.
2074 */
02c83209
AK		 2075 r = 0;
2076 goto out;
46fc1477 2077 }
6aa8b732
AK		 2078 }
2079
8eb7d334 2080 if (kvm_run->exit_reason == KVM_EXIT_HYPERCALL) {
cbdd1bea 2081 kvm_x86_ops->cache_regs(vcpu);
b4e63f56 2082 vcpu->regs[VCPU_REGS_RAX] = kvm_run->hypercall.ret;
cbdd1bea 2083 kvm_x86_ops->decache_regs(vcpu);
b4e63f56
AK		 2084 }
2085
04d2cc77 2086 r = __vcpu_run(vcpu, kvm_run);
6aa8b732 2087
039576c0 2088out:
1961d276
AK		 2089 if (vcpu->sigset_active)
2090 sigprocmask(SIG_SETMASK, &sigsaved, NULL);
2091
6aa8b732
AK		 2092 vcpu_put(vcpu);
2093 return r;
2094}
2095
bccf2150
AK		 2096static int kvm_vcpu_ioctl_get_regs(struct kvm_vcpu *vcpu,
2097 struct kvm_regs *regs)
6aa8b732 2098{
bccf2150 2099 vcpu_load(vcpu);
6aa8b732 2100
cbdd1bea 2101 kvm_x86_ops->cache_regs(vcpu);
6aa8b732
AK		 2102
2103 regs->rax = vcpu->regs[VCPU_REGS_RAX];
2104 regs->rbx = vcpu->regs[VCPU_REGS_RBX];
2105 regs->rcx = vcpu->regs[VCPU_REGS_RCX];
2106 regs->rdx = vcpu->regs[VCPU_REGS_RDX];
2107 regs->rsi = vcpu->regs[VCPU_REGS_RSI];
2108 regs->rdi = vcpu->regs[VCPU_REGS_RDI];
2109 regs->rsp = vcpu->regs[VCPU_REGS_RSP];
2110 regs->rbp = vcpu->regs[VCPU_REGS_RBP];
05b3e0c2 2111#ifdef CONFIG_X86_64
6aa8b732
AK		 2112 regs->r8 = vcpu->regs[VCPU_REGS_R8];
2113 regs->r9 = vcpu->regs[VCPU_REGS_R9];
2114 regs->r10 = vcpu->regs[VCPU_REGS_R10];
2115 regs->r11 = vcpu->regs[VCPU_REGS_R11];
2116 regs->r12 = vcpu->regs[VCPU_REGS_R12];
2117 regs->r13 = vcpu->regs[VCPU_REGS_R13];
2118 regs->r14 = vcpu->regs[VCPU_REGS_R14];
2119 regs->r15 = vcpu->regs[VCPU_REGS_R15];
2120#endif
2121
2122 regs->rip = vcpu->rip;
cbdd1bea 2123 regs->rflags = kvm_x86_ops->get_rflags(vcpu);
6aa8b732
AK		 2124
2125 /*
2126 * Don't leak debug flags in case they were set for guest debugging
2127 */
2128 if (vcpu->guest_debug.enabled && vcpu->guest_debug.singlestep)
2129 regs->rflags &= ~(X86_EFLAGS_TF | X86_EFLAGS_RF);
2130
2131 vcpu_put(vcpu);
2132
2133 return 0;
2134}
2135
bccf2150
AK		 2136static int kvm_vcpu_ioctl_set_regs(struct kvm_vcpu *vcpu,
2137 struct kvm_regs *regs)
6aa8b732 2138{
bccf2150 2139 vcpu_load(vcpu);
6aa8b732
AK		 2140
2141 vcpu->regs[VCPU_REGS_RAX] = regs->rax;
2142 vcpu->regs[VCPU_REGS_RBX] = regs->rbx;
2143 vcpu->regs[VCPU_REGS_RCX] = regs->rcx;
2144 vcpu->regs[VCPU_REGS_RDX] = regs->rdx;
2145 vcpu->regs[VCPU_REGS_RSI] = regs->rsi;
2146 vcpu->regs[VCPU_REGS_RDI] = regs->rdi;
2147 vcpu->regs[VCPU_REGS_RSP] = regs->rsp;
2148 vcpu->regs[VCPU_REGS_RBP] = regs->rbp;
05b3e0c2 2149#ifdef CONFIG_X86_64
6aa8b732
AK		 2150 vcpu->regs[VCPU_REGS_R8] = regs->r8;
2151 vcpu->regs[VCPU_REGS_R9] = regs->r9;
2152 vcpu->regs[VCPU_REGS_R10] = regs->r10;
2153 vcpu->regs[VCPU_REGS_R11] = regs->r11;
2154 vcpu->regs[VCPU_REGS_R12] = regs->r12;
2155 vcpu->regs[VCPU_REGS_R13] = regs->r13;
2156 vcpu->regs[VCPU_REGS_R14] = regs->r14;
2157 vcpu->regs[VCPU_REGS_R15] = regs->r15;
2158#endif
2159
2160 vcpu->rip = regs->rip;
cbdd1bea 2161 kvm_x86_ops->set_rflags(vcpu, regs->rflags);
6aa8b732 2162
cbdd1bea 2163 kvm_x86_ops->decache_regs(vcpu);
6aa8b732
AK		 2164
2165 vcpu_put(vcpu);
2166
2167 return 0;
2168}
2169
2170static void get_segment(struct kvm_vcpu *vcpu,
2171 struct kvm_segment *var, int seg)
2172{
cbdd1bea 2173 return kvm_x86_ops->get_segment(vcpu, var, seg);
6aa8b732
AK		 2174}
2175
bccf2150
AK		 2176static int kvm_vcpu_ioctl_get_sregs(struct kvm_vcpu *vcpu,
2177 struct kvm_sregs *sregs)
6aa8b732 2178{
6aa8b732 2179 struct descriptor_table dt;
2a8067f1 2180 int pending_vec;
6aa8b732 2181
bccf2150 2182 vcpu_load(vcpu);
6aa8b732
AK		 2183
2184 get_segment(vcpu, &sregs->cs, VCPU_SREG_CS);
2185 get_segment(vcpu, &sregs->ds, VCPU_SREG_DS);
2186 get_segment(vcpu, &sregs->es, VCPU_SREG_ES);
2187 get_segment(vcpu, &sregs->fs, VCPU_SREG_FS);
2188 get_segment(vcpu, &sregs->gs, VCPU_SREG_GS);
2189 get_segment(vcpu, &sregs->ss, VCPU_SREG_SS);
2190
2191 get_segment(vcpu, &sregs->tr, VCPU_SREG_TR);
2192 get_segment(vcpu, &sregs->ldt, VCPU_SREG_LDTR);
2193
cbdd1bea 2194 kvm_x86_ops->get_idt(vcpu, &dt);
6aa8b732
AK		 2195 sregs->idt.limit = dt.limit;
2196 sregs->idt.base = dt.base;
cbdd1bea 2197 kvm_x86_ops->get_gdt(vcpu, &dt);
6aa8b732
AK		 2198 sregs->gdt.limit = dt.limit;
2199 sregs->gdt.base = dt.base;
2200
cbdd1bea 2201 kvm_x86_ops->decache_cr4_guest_bits(vcpu);
6aa8b732
AK		 2202 sregs->cr0 = vcpu->cr0;
2203 sregs->cr2 = vcpu->cr2;
2204 sregs->cr3 = vcpu->cr3;
2205 sregs->cr4 = vcpu->cr4;
7017fc3d 2206 sregs->cr8 = get_cr8(vcpu);
6aa8b732 2207 sregs->efer = vcpu->shadow_efer;
7017fc3d 2208 sregs->apic_base = kvm_get_apic_base(vcpu);
6aa8b732 2209
2a8067f1 2210 if (irqchip_in_kernel(vcpu->kvm)) {
c52fb35a
HQ		 2211 memset(sregs->interrupt_bitmap, 0,
2212 sizeof sregs->interrupt_bitmap);
cbdd1bea 2213 pending_vec = kvm_x86_ops->get_irq(vcpu);
2a8067f1
ED		 2214 if (pending_vec >= 0)
2215 set_bit(pending_vec, (unsigned long *)sregs->interrupt_bitmap);
2216 } else
c52fb35a
HQ		 2217 memcpy(sregs->interrupt_bitmap, vcpu->irq_pending,
2218 sizeof sregs->interrupt_bitmap);
6aa8b732
AK		 2219
2220 vcpu_put(vcpu);
2221
2222 return 0;
2223}
2224
2225static void set_segment(struct kvm_vcpu *vcpu,
2226 struct kvm_segment *var, int seg)
2227{
cbdd1bea 2228 return kvm_x86_ops->set_segment(vcpu, var, seg);
6aa8b732
AK		 2229}
2230
bccf2150
AK		 2231static int kvm_vcpu_ioctl_set_sregs(struct kvm_vcpu *vcpu,
2232 struct kvm_sregs *sregs)
6aa8b732 2233{
6aa8b732 2234 int mmu_reset_needed = 0;
2a8067f1 2235 int i, pending_vec, max_bits;
6aa8b732
AK		 2236 struct descriptor_table dt;
2237
bccf2150 2238 vcpu_load(vcpu);
6aa8b732 2239
6aa8b732
AK		 2240 dt.limit = sregs->idt.limit;
2241 dt.base = sregs->idt.base;
cbdd1bea 2242 kvm_x86_ops->set_idt(vcpu, &dt);
6aa8b732
AK		 2243 dt.limit = sregs->gdt.limit;
2244 dt.base = sregs->gdt.base;
cbdd1bea 2245 kvm_x86_ops->set_gdt(vcpu, &dt);
6aa8b732
AK		 2246
2247 vcpu->cr2 = sregs->cr2;
2248 mmu_reset_needed |= vcpu->cr3 != sregs->cr3;
2249 vcpu->cr3 = sregs->cr3;
2250
7017fc3d 2251 set_cr8(vcpu, sregs->cr8);
6aa8b732
AK		 2252
2253 mmu_reset_needed |= vcpu->shadow_efer != sregs->efer;
05b3e0c2 2254#ifdef CONFIG_X86_64
cbdd1bea 2255 kvm_x86_ops->set_efer(vcpu, sregs->efer);
6aa8b732 2256#endif
7017fc3d 2257 kvm_set_apic_base(vcpu, sregs->apic_base);
6aa8b732 2258
cbdd1bea 2259 kvm_x86_ops->decache_cr4_guest_bits(vcpu);
399badf3 2260
6aa8b732 2261 mmu_reset_needed |= vcpu->cr0 != sregs->cr0;
81f50e3b 2262 vcpu->cr0 = sregs->cr0;
cbdd1bea 2263 kvm_x86_ops->set_cr0(vcpu, sregs->cr0);
6aa8b732
AK		 2264
2265 mmu_reset_needed |= vcpu->cr4 != sregs->cr4;
cbdd1bea 2266 kvm_x86_ops->set_cr4(vcpu, sregs->cr4);
1b0973bd
AK		 2267 if (!is_long_mode(vcpu) && is_pae(vcpu))
2268 load_pdptrs(vcpu, vcpu->cr3);
6aa8b732
AK		 2269
2270 if (mmu_reset_needed)
2271 kvm_mmu_reset_context(vcpu);
2272
c52fb35a
HQ		 2273 if (!irqchip_in_kernel(vcpu->kvm)) {
2274 memcpy(vcpu->irq_pending, sregs->interrupt_bitmap,
2275 sizeof vcpu->irq_pending);
2276 vcpu->irq_summary = 0;
2277 for (i = 0; i < ARRAY_SIZE(vcpu->irq_pending); ++i)
2278 if (vcpu->irq_pending[i])
2279 __set_bit(i, &vcpu->irq_summary);
2a8067f1
ED		 2280 } else {
2281 max_bits = (sizeof sregs->interrupt_bitmap) << 3;
2282 pending_vec = find_first_bit(
2283 (const unsigned long *)sregs->interrupt_bitmap,
2284 max_bits);
2285 /* Only pending external irq is handled here */
2286 if (pending_vec < max_bits) {
cbdd1bea 2287 kvm_x86_ops->set_irq(vcpu, pending_vec);
2a8067f1
ED		 2288 printk("Set back pending irq %d\n", pending_vec);
2289 }
c52fb35a 2290 }
6aa8b732 2291
024aa1c0
AK		 2292 set_segment(vcpu, &sregs->cs, VCPU_SREG_CS);
2293 set_segment(vcpu, &sregs->ds, VCPU_SREG_DS);
2294 set_segment(vcpu, &sregs->es, VCPU_SREG_ES);
2295 set_segment(vcpu, &sregs->fs, VCPU_SREG_FS);
2296 set_segment(vcpu, &sregs->gs, VCPU_SREG_GS);
2297 set_segment(vcpu, &sregs->ss, VCPU_SREG_SS);
2298
2299 set_segment(vcpu, &sregs->tr, VCPU_SREG_TR);
2300 set_segment(vcpu, &sregs->ldt, VCPU_SREG_LDTR);
2301
6aa8b732
AK		 2302 vcpu_put(vcpu);
2303
2304 return 0;
2305}
2306
1747fb71
RR		 2307void kvm_get_cs_db_l_bits(struct kvm_vcpu *vcpu, int *db, int *l)
2308{
2309 struct kvm_segment cs;
2310
2311 get_segment(vcpu, &cs, VCPU_SREG_CS);
2312 *db = cs.db;
2313 *l = cs.l;
2314}
2315EXPORT_SYMBOL_GPL(kvm_get_cs_db_l_bits);
2316
6aa8b732
AK		 2317/*
2318 * List of msr numbers which we expose to userspace through KVM_GET_MSRS
2319 * and KVM_SET_MSRS, and KVM_GET_MSR_INDEX_LIST.
bf591b24
MR		 2320 *
2321 * This list is modified at module load time to reflect the
2322 * capabilities of the host cpu.
6aa8b732
AK		 2323 */
2324static u32 msrs_to_save[] = {
2325 MSR_IA32_SYSENTER_CS, MSR_IA32_SYSENTER_ESP, MSR_IA32_SYSENTER_EIP,
2326 MSR_K6_STAR,
05b3e0c2 2327#ifdef CONFIG_X86_64
6aa8b732
AK		 2328 MSR_CSTAR, MSR_KERNEL_GS_BASE, MSR_SYSCALL_MASK, MSR_LSTAR,
2329#endif
2330 MSR_IA32_TIME_STAMP_COUNTER,
2331};
2332
bf591b24
MR		 2333static unsigned num_msrs_to_save;
2334
6f00e68f
AK		 2335static u32 emulated_msrs[] = {
2336 MSR_IA32_MISC_ENABLE,
2337};
2338
bf591b24
MR		 2339static __init void kvm_init_msr_list(void)
2340{
2341 u32 dummy[2];
2342 unsigned i, j;
2343
2344 for (i = j = 0; i < ARRAY_SIZE(msrs_to_save); i++) {
2345 if (rdmsr_safe(msrs_to_save[i], &dummy[0], &dummy[1]) < 0)
2346 continue;
2347 if (j < i)
2348 msrs_to_save[j] = msrs_to_save[i];
2349 j++;
2350 }
2351 num_msrs_to_save = j;
2352}
6aa8b732
AK		 2353
2354/*
2355 * Adapt set_msr() to msr_io()'s calling convention
2356 */
2357static int do_set_msr(struct kvm_vcpu *vcpu, unsigned index, u64 *data)
2358{
35f3f286 2359 return kvm_set_msr(vcpu, index, *data);
6aa8b732
AK		 2360}
2361
2362/*
2363 * Read or write a bunch of msrs. All parameters are kernel addresses.
2364 *
2365 * @return number of msrs set successfully.
2366 */
bccf2150 2367static int __msr_io(struct kvm_vcpu *vcpu, struct kvm_msrs *msrs,
6aa8b732
AK		 2368 struct kvm_msr_entry *entries,
2369 int (*do_msr)(struct kvm_vcpu *vcpu,
2370 unsigned index, u64 *data))
2371{
6aa8b732
AK		 2372 int i;
2373
bccf2150 2374 vcpu_load(vcpu);
6aa8b732
AK		 2375
2376 for (i = 0; i < msrs->nmsrs; ++i)
2377 if (do_msr(vcpu, entries[i].index, &entries[i].data))
2378 break;
2379
2380 vcpu_put(vcpu);
2381
2382 return i;
2383}
2384
2385/*
2386 * Read or write a bunch of msrs. Parameters are user addresses.
2387 *
2388 * @return number of msrs set successfully.
2389 */
bccf2150 2390static int msr_io(struct kvm_vcpu *vcpu, struct kvm_msrs __user *user_msrs,
6aa8b732
AK		 2391 int (*do_msr)(struct kvm_vcpu *vcpu,
2392 unsigned index, u64 *data),
2393 int writeback)
2394{
2395 struct kvm_msrs msrs;
2396 struct kvm_msr_entry *entries;
2397 int r, n;
2398 unsigned size;
2399
2400 r = -EFAULT;
2401 if (copy_from_user(&msrs, user_msrs, sizeof msrs))
2402 goto out;
2403
2404 r = -E2BIG;
2405 if (msrs.nmsrs >= MAX_IO_MSRS)
2406 goto out;
2407
2408 r = -ENOMEM;
2409 size = sizeof(struct kvm_msr_entry) * msrs.nmsrs;
2410 entries = vmalloc(size);
2411 if (!entries)
2412 goto out;
2413
2414 r = -EFAULT;
2415 if (copy_from_user(entries, user_msrs->entries, size))
2416 goto out_free;
2417
bccf2150 2418 r = n = __msr_io(vcpu, &msrs, entries, do_msr);
6aa8b732
AK		 2419 if (r < 0)
2420 goto out_free;
2421
2422 r = -EFAULT;
2423 if (writeback && copy_to_user(user_msrs->entries, entries, size))
2424 goto out_free;
2425
2426 r = n;
2427
2428out_free:
2429 vfree(entries);
2430out:
2431 return r;
2432}
2433
2434/*
2435 * Translate a guest virtual address to a guest physical address.
2436 */
bccf2150
AK		 2437static int kvm_vcpu_ioctl_translate(struct kvm_vcpu *vcpu,
2438 struct kvm_translation *tr)
6aa8b732
AK		 2439{
2440 unsigned long vaddr = tr->linear_address;
6aa8b732
AK		 2441 gpa_t gpa;
2442
bccf2150 2443 vcpu_load(vcpu);
11ec2804 2444 mutex_lock(&vcpu->kvm->lock);
6aa8b732
AK		 2445 gpa = vcpu->mmu.gva_to_gpa(vcpu, vaddr);
2446 tr->physical_address = gpa;
2447 tr->valid = gpa != UNMAPPED_GVA;
2448 tr->writeable = 1;
2449 tr->usermode = 0;
11ec2804 2450 mutex_unlock(&vcpu->kvm->lock);
6aa8b732
AK		 2451 vcpu_put(vcpu);
2452
2453 return 0;
2454}
2455
bccf2150
AK		 2456static int kvm_vcpu_ioctl_interrupt(struct kvm_vcpu *vcpu,
2457 struct kvm_interrupt *irq)
6aa8b732 2458{
6aa8b732
AK		 2459 if (irq->irq < 0 || irq->irq >= 256)
2460 return -EINVAL;
97222cc8
ED		 2461 if (irqchip_in_kernel(vcpu->kvm))
2462 return -ENXIO;
bccf2150 2463 vcpu_load(vcpu);
6aa8b732
AK		 2464
2465 set_bit(irq->irq, vcpu->irq_pending);
2466 set_bit(irq->irq / BITS_PER_LONG, &vcpu->irq_summary);
2467
2468 vcpu_put(vcpu);
2469
2470 return 0;
2471}
2472
bccf2150
AK		 2473static int kvm_vcpu_ioctl_debug_guest(struct kvm_vcpu *vcpu,
2474 struct kvm_debug_guest *dbg)
6aa8b732 2475{
6aa8b732
AK		 2476 int r;
2477
bccf2150 2478 vcpu_load(vcpu);
6aa8b732 2479
cbdd1bea 2480 r = kvm_x86_ops->set_guest_debug(vcpu, dbg);
6aa8b732
AK		 2481
2482 vcpu_put(vcpu);
2483
2484 return r;
2485}
2486
9a2bb7f4
AK		 2487static struct page *kvm_vcpu_nopage(struct vm_area_struct *vma,
2488 unsigned long address,
2489 int *type)
2490{
2491 struct kvm_vcpu *vcpu = vma->vm_file->private_data;
2492 unsigned long pgoff;
2493 struct page *page;
2494
9a2bb7f4 2495 pgoff = ((address - vma->vm_start) >> PAGE_SHIFT) + vma->vm_pgoff;
039576c0
AK		 2496 if (pgoff == 0)
2497 page = virt_to_page(vcpu->run);
2498 else if (pgoff == KVM_PIO_PAGE_OFFSET)
2499 page = virt_to_page(vcpu->pio_data);
2500 else
9a2bb7f4 2501 return NOPAGE_SIGBUS;
9a2bb7f4 2502 get_page(page);
cd0d9137
NAQ		 2503 if (type != NULL)
2504 *type = VM_FAULT_MINOR;
2505
9a2bb7f4
AK		 2506 return page;
2507}
2508
2509static struct vm_operations_struct kvm_vcpu_vm_ops = {
2510 .nopage = kvm_vcpu_nopage,
2511};
2512
2513static int kvm_vcpu_mmap(struct file *file, struct vm_area_struct *vma)
2514{
2515 vma->vm_ops = &kvm_vcpu_vm_ops;
2516 return 0;
2517}
2518
bccf2150
AK		 2519static int kvm_vcpu_release(struct inode *inode, struct file *filp)
2520{
2521 struct kvm_vcpu *vcpu = filp->private_data;
2522
2523 fput(vcpu->kvm->filp);
2524 return 0;
2525}
2526
2527static struct file_operations kvm_vcpu_fops = {
2528 .release = kvm_vcpu_release,
2529 .unlocked_ioctl = kvm_vcpu_ioctl,
2530 .compat_ioctl = kvm_vcpu_ioctl,
9a2bb7f4 2531 .mmap = kvm_vcpu_mmap,
bccf2150
AK		 2532};
2533
2534/*
2535 * Allocates an inode for the vcpu.
2536 */
2537static int create_vcpu_fd(struct kvm_vcpu *vcpu)
2538{
2539 int fd, r;
2540 struct inode *inode;
2541 struct file *file;
2542
d6d28168
AK		 2543 r = anon_inode_getfd(&fd, &inode, &file,
2544 "kvm-vcpu", &kvm_vcpu_fops, vcpu);
2545 if (r)
2546 return r;
bccf2150 2547 atomic_inc(&vcpu->kvm->filp->f_count);
bccf2150 2548 return fd;
bccf2150
AK		 2549}
2550
c5ea7660
AK		 2551/*
2552 * Creates some virtual cpus. Good luck creating more than one.
2553 */
2554static int kvm_vm_ioctl_create_vcpu(struct kvm *kvm, int n)
2555{
2556 int r;
2557 struct kvm_vcpu *vcpu;
2558
c5ea7660 2559 if (!valid_vcpu(n))
fb3f0f51 2560 return -EINVAL;
c5ea7660 2561
cbdd1bea 2562 vcpu = kvm_x86_ops->vcpu_create(kvm, n);
fb3f0f51
RR		 2563 if (IS_ERR(vcpu))
2564 return PTR_ERR(vcpu);
c5ea7660 2565
15ad7146
AK		 2566 preempt_notifier_init(&vcpu->preempt_notifier, &kvm_preempt_ops);
2567
b114b080
RR		 2568 /* We do fxsave: this must be aligned. */
2569 BUG_ON((unsigned long)&vcpu->host_fx_image & 0xF);
2570
fb3f0f51 2571 vcpu_load(vcpu);
c5ea7660 2572 r = kvm_mmu_setup(vcpu);
c5ea7660 2573 vcpu_put(vcpu);
c5ea7660 2574 if (r < 0)
fb3f0f51
RR		 2575 goto free_vcpu;
2576
11ec2804 2577 mutex_lock(&kvm->lock);
fb3f0f51
RR		 2578 if (kvm->vcpus[n]) {
2579 r = -EEXIST;
11ec2804 2580 mutex_unlock(&kvm->lock);
fb3f0f51
RR		 2581 goto mmu_unload;
2582 }
2583 kvm->vcpus[n] = vcpu;
11ec2804 2584 mutex_unlock(&kvm->lock);
c5ea7660 2585
fb3f0f51 2586 /* Now it's all set up, let userspace reach it */
bccf2150
AK		 2587 r = create_vcpu_fd(vcpu);
2588 if (r < 0)
fb3f0f51
RR		 2589 goto unlink;
2590 return r;
39c3b86e 2591
fb3f0f51 2592unlink:
11ec2804 2593 mutex_lock(&kvm->lock);
fb3f0f51 2594 kvm->vcpus[n] = NULL;
11ec2804 2595 mutex_unlock(&kvm->lock);
a2fa3e9f 2596
fb3f0f51
RR		 2597mmu_unload:
2598 vcpu_load(vcpu);
2599 kvm_mmu_unload(vcpu);
2600 vcpu_put(vcpu);
c5ea7660 2601
fb3f0f51 2602free_vcpu:
cbdd1bea 2603 kvm_x86_ops->vcpu_free(vcpu);
c5ea7660
AK		 2604 return r;
2605}
2606
2cc51560
ED		 2607static void cpuid_fix_nx_cap(struct kvm_vcpu *vcpu)
2608{
2609 u64 efer;
2610 int i;
2611 struct kvm_cpuid_entry *e, *entry;
2612
2613 rdmsrl(MSR_EFER, efer);
2614 entry = NULL;
2615 for (i = 0; i < vcpu->cpuid_nent; ++i) {
2616 e = &vcpu->cpuid_entries[i];
2617 if (e->function == 0x80000001) {
2618 entry = e;
2619 break;
2620 }
2621 }
4c981b43 2622 if (entry && (entry->edx & (1 << 20)) && !(efer & EFER_NX)) {
2cc51560 2623 entry->edx &= ~(1 << 20);
4c981b43 2624 printk(KERN_INFO "kvm: guest NX capability removed\n");
2cc51560
ED		 2625 }
2626}
2627
06465c5a
AK		 2628static int kvm_vcpu_ioctl_set_cpuid(struct kvm_vcpu *vcpu,
2629 struct kvm_cpuid *cpuid,
2630 struct kvm_cpuid_entry __user *entries)
2631{
2632 int r;
2633
2634 r = -E2BIG;
2635 if (cpuid->nent > KVM_MAX_CPUID_ENTRIES)
2636 goto out;
2637 r = -EFAULT;
2638 if (copy_from_user(&vcpu->cpuid_entries, entries,
2639 cpuid->nent * sizeof(struct kvm_cpuid_entry)))
2640 goto out;
2641 vcpu->cpuid_nent = cpuid->nent;
2cc51560 2642 cpuid_fix_nx_cap(vcpu);
06465c5a
AK		 2643 return 0;
2644
2645out:
2646 return r;
2647}
2648
1961d276
AK		 2649static int kvm_vcpu_ioctl_set_sigmask(struct kvm_vcpu *vcpu, sigset_t *sigset)
2650{
2651 if (sigset) {
2652 sigdelsetmask(sigset, sigmask(SIGKILL)|sigmask(SIGSTOP));
2653 vcpu->sigset_active = 1;
2654 vcpu->sigset = *sigset;
2655 } else
2656 vcpu->sigset_active = 0;
2657 return 0;
2658}
2659
b8836737
AK		 2660/*
2661 * fxsave fpu state. Taken from x86_64/processor.h. To be killed when
2662 * we have asm/x86/processor.h
2663 */
2664struct fxsave {
2665 u16 cwd;
2666 u16 swd;
2667 u16 twd;
2668 u16 fop;
2669 u64 rip;
2670 u64 rdp;
2671 u32 mxcsr;
2672 u32 mxcsr_mask;
2673 u32 st_space[32]; /* 8*16 bytes for each FP-reg = 128 bytes */
2674#ifdef CONFIG_X86_64
2675 u32 xmm_space[64]; /* 16*16 bytes for each XMM-reg = 256 bytes */
2676#else
2677 u32 xmm_space[32]; /* 8*16 bytes for each XMM-reg = 128 bytes */
2678#endif
2679};
2680
2681static int kvm_vcpu_ioctl_get_fpu(struct kvm_vcpu *vcpu, struct kvm_fpu *fpu)
2682{
b114b080 2683 struct fxsave *fxsave = (struct fxsave *)&vcpu->guest_fx_image;
b8836737
AK		 2684
2685 vcpu_load(vcpu);
2686
2687 memcpy(fpu->fpr, fxsave->st_space, 128);
2688 fpu->fcw = fxsave->cwd;
2689 fpu->fsw = fxsave->swd;
2690 fpu->ftwx = fxsave->twd;
2691 fpu->last_opcode = fxsave->fop;
2692 fpu->last_ip = fxsave->rip;
2693 fpu->last_dp = fxsave->rdp;
2694 memcpy(fpu->xmm, fxsave->xmm_space, sizeof fxsave->xmm_space);
2695
2696 vcpu_put(vcpu);
2697
2698 return 0;
2699}
2700
2701static int kvm_vcpu_ioctl_set_fpu(struct kvm_vcpu *vcpu, struct kvm_fpu *fpu)
2702{
b114b080 2703 struct fxsave *fxsave = (struct fxsave *)&vcpu->guest_fx_image;
b8836737
AK		 2704
2705 vcpu_load(vcpu);
2706
2707 memcpy(fxsave->st_space, fpu->fpr, 128);
2708 fxsave->cwd = fpu->fcw;
2709 fxsave->swd = fpu->fsw;
2710 fxsave->twd = fpu->ftwx;
2711 fxsave->fop = fpu->last_opcode;
2712 fxsave->rip = fpu->last_ip;
2713 fxsave->rdp = fpu->last_dp;
2714 memcpy(fxsave->xmm_space, fpu->xmm, sizeof fxsave->xmm_space);
2715
2716 vcpu_put(vcpu);
2717
2718 return 0;
2719}
2720
96ad2cc6
ED		 2721static int kvm_vcpu_ioctl_get_lapic(struct kvm_vcpu *vcpu,
2722 struct kvm_lapic_state *s)
2723{
2724 vcpu_load(vcpu);
2725 memcpy(s->regs, vcpu->apic->regs, sizeof *s);
2726 vcpu_put(vcpu);
2727
2728 return 0;
2729}
2730
2731static int kvm_vcpu_ioctl_set_lapic(struct kvm_vcpu *vcpu,
2732 struct kvm_lapic_state *s)
2733{
2734 vcpu_load(vcpu);
2735 memcpy(vcpu->apic->regs, s->regs, sizeof *s);
2736 kvm_apic_post_state_restore(vcpu);
2737 vcpu_put(vcpu);
2738
2739 return 0;
2740}
2741
bccf2150
AK		 2742static long kvm_vcpu_ioctl(struct file *filp,
2743 unsigned int ioctl, unsigned long arg)
6aa8b732 2744{
bccf2150 2745 struct kvm_vcpu *vcpu = filp->private_data;
2f366987 2746 void __user *argp = (void __user *)arg;
6aa8b732
AK		 2747 int r = -EINVAL;
2748
2749 switch (ioctl) {
9a2bb7f4 2750 case KVM_RUN:
f0fe5108
AK		 2751 r = -EINVAL;
2752 if (arg)
2753 goto out;
9a2bb7f4 2754 r = kvm_vcpu_ioctl_run(vcpu, vcpu->run);
6aa8b732 2755 break;
6aa8b732
AK		 2756 case KVM_GET_REGS: {
2757 struct kvm_regs kvm_regs;
2758
bccf2150
AK		 2759 memset(&kvm_regs, 0, sizeof kvm_regs);
2760 r = kvm_vcpu_ioctl_get_regs(vcpu, &kvm_regs);
6aa8b732
AK		 2761 if (r)
2762 goto out;
2763 r = -EFAULT;
2f366987 2764 if (copy_to_user(argp, &kvm_regs, sizeof kvm_regs))
6aa8b732
AK		 2765 goto out;
2766 r = 0;
2767 break;
2768 }
2769 case KVM_SET_REGS: {
2770 struct kvm_regs kvm_regs;
2771
2772 r = -EFAULT;
2f366987 2773 if (copy_from_user(&kvm_regs, argp, sizeof kvm_regs))
6aa8b732 2774 goto out;
bccf2150 2775 r = kvm_vcpu_ioctl_set_regs(vcpu, &kvm_regs);
6aa8b732
AK		 2776 if (r)
2777 goto out;
2778 r = 0;
2779 break;
2780 }
2781 case KVM_GET_SREGS: {
2782 struct kvm_sregs kvm_sregs;
2783
bccf2150
AK		 2784 memset(&kvm_sregs, 0, sizeof kvm_sregs);
2785 r = kvm_vcpu_ioctl_get_sregs(vcpu, &kvm_sregs);
6aa8b732
AK		 2786 if (r)
2787 goto out;
2788 r = -EFAULT;
2f366987 2789 if (copy_to_user(argp, &kvm_sregs, sizeof kvm_sregs))
6aa8b732
AK		 2790 goto out;
2791 r = 0;
2792 break;
2793 }
2794 case KVM_SET_SREGS: {
2795 struct kvm_sregs kvm_sregs;
2796
2797 r = -EFAULT;
2f366987 2798 if (copy_from_user(&kvm_sregs, argp, sizeof kvm_sregs))
6aa8b732 2799 goto out;
bccf2150 2800 r = kvm_vcpu_ioctl_set_sregs(vcpu, &kvm_sregs);
6aa8b732
AK		 2801 if (r)
2802 goto out;
2803 r = 0;
2804 break;
2805 }
2806 case KVM_TRANSLATE: {
2807 struct kvm_translation tr;
2808
2809 r = -EFAULT;
2f366987 2810 if (copy_from_user(&tr, argp, sizeof tr))
6aa8b732 2811 goto out;
bccf2150 2812 r = kvm_vcpu_ioctl_translate(vcpu, &tr);
6aa8b732
AK		 2813 if (r)
2814 goto out;
2815 r = -EFAULT;
2f366987 2816 if (copy_to_user(argp, &tr, sizeof tr))
6aa8b732
AK		 2817 goto out;
2818 r = 0;
2819 break;
2820 }
2821 case KVM_INTERRUPT: {
2822 struct kvm_interrupt irq;
2823
2824 r = -EFAULT;
2f366987 2825 if (copy_from_user(&irq, argp, sizeof irq))
6aa8b732 2826 goto out;
bccf2150 2827 r = kvm_vcpu_ioctl_interrupt(vcpu, &irq);
6aa8b732
AK		 2828 if (r)
2829 goto out;
2830 r = 0;
2831 break;
2832 }
2833 case KVM_DEBUG_GUEST: {
2834 struct kvm_debug_guest dbg;
2835
2836 r = -EFAULT;
2f366987 2837 if (copy_from_user(&dbg, argp, sizeof dbg))
6aa8b732 2838 goto out;
bccf2150 2839 r = kvm_vcpu_ioctl_debug_guest(vcpu, &dbg);
6aa8b732
AK		 2840 if (r)
2841 goto out;
2842 r = 0;
2843 break;
2844 }
bccf2150 2845 case KVM_GET_MSRS:
35f3f286 2846 r = msr_io(vcpu, argp, kvm_get_msr, 1);
bccf2150
AK		 2847 break;
2848 case KVM_SET_MSRS:
2849 r = msr_io(vcpu, argp, do_set_msr, 0);
2850 break;
06465c5a
AK		 2851 case KVM_SET_CPUID: {
2852 struct kvm_cpuid __user *cpuid_arg = argp;
2853 struct kvm_cpuid cpuid;
2854
2855 r = -EFAULT;
2856 if (copy_from_user(&cpuid, cpuid_arg, sizeof cpuid))
2857 goto out;
2858 r = kvm_vcpu_ioctl_set_cpuid(vcpu, &cpuid, cpuid_arg->entries);
2859 if (r)
2860 goto out;
2861 break;
2862 }
1961d276
AK		 2863 case KVM_SET_SIGNAL_MASK: {
2864 struct kvm_signal_mask __user *sigmask_arg = argp;
2865 struct kvm_signal_mask kvm_sigmask;
2866 sigset_t sigset, *p;
2867
2868 p = NULL;
2869 if (argp) {
2870 r = -EFAULT;
2871 if (copy_from_user(&kvm_sigmask, argp,
2872 sizeof kvm_sigmask))
2873 goto out;
2874 r = -EINVAL;
2875 if (kvm_sigmask.len != sizeof sigset)
2876 goto out;
2877 r = -EFAULT;
2878 if (copy_from_user(&sigset, sigmask_arg->sigset,
2879 sizeof sigset))
2880 goto out;
2881 p = &sigset;
2882 }
2883 r = kvm_vcpu_ioctl_set_sigmask(vcpu, &sigset);
2884 break;
2885 }
b8836737
AK		 2886 case KVM_GET_FPU: {
2887 struct kvm_fpu fpu;
2888
2889 memset(&fpu, 0, sizeof fpu);
2890 r = kvm_vcpu_ioctl_get_fpu(vcpu, &fpu);
2891 if (r)
2892 goto out;
2893 r = -EFAULT;
2894 if (copy_to_user(argp, &fpu, sizeof fpu))
2895 goto out;
2896 r = 0;
2897 break;
2898 }
2899 case KVM_SET_FPU: {
2900 struct kvm_fpu fpu;
2901
2902 r = -EFAULT;
2903 if (copy_from_user(&fpu, argp, sizeof fpu))
2904 goto out;
2905 r = kvm_vcpu_ioctl_set_fpu(vcpu, &fpu);
2906 if (r)
2907 goto out;
2908 r = 0;
2909 break;
2910 }
96ad2cc6
ED		 2911 case KVM_GET_LAPIC: {
2912 struct kvm_lapic_state lapic;
2913
2914 memset(&lapic, 0, sizeof lapic);
2915 r = kvm_vcpu_ioctl_get_lapic(vcpu, &lapic);
2916 if (r)
2917 goto out;
2918 r = -EFAULT;
2919 if (copy_to_user(argp, &lapic, sizeof lapic))
2920 goto out;
2921 r = 0;
2922 break;
2923 }
2924 case KVM_SET_LAPIC: {
2925 struct kvm_lapic_state lapic;
2926
2927 r = -EFAULT;
2928 if (copy_from_user(&lapic, argp, sizeof lapic))
2929 goto out;
2930 r = kvm_vcpu_ioctl_set_lapic(vcpu, &lapic);;
2931 if (r)
2932 goto out;
2933 r = 0;
2934 break;
2935 }
bccf2150
AK		 2936 default:
2937 ;
2938 }
2939out:
2940 return r;
2941}
2942
2943static long kvm_vm_ioctl(struct file *filp,
2944 unsigned int ioctl, unsigned long arg)
2945{
2946 struct kvm *kvm = filp->private_data;
2947 void __user *argp = (void __user *)arg;
2948 int r = -EINVAL;
2949
2950 switch (ioctl) {
2951 case KVM_CREATE_VCPU:
2952 r = kvm_vm_ioctl_create_vcpu(kvm, arg);
2953 if (r < 0)
2954 goto out;
2955 break;
6aa8b732
AK		 2956 case KVM_SET_MEMORY_REGION: {
2957 struct kvm_memory_region kvm_mem;
2958
2959 r = -EFAULT;
2f366987 2960 if (copy_from_user(&kvm_mem, argp, sizeof kvm_mem))
6aa8b732 2961 goto out;
2c6f5df9 2962 r = kvm_vm_ioctl_set_memory_region(kvm, &kvm_mem);
6aa8b732
AK		 2963 if (r)
2964 goto out;
2965 break;
2966 }
2967 case KVM_GET_DIRTY_LOG: {
2968 struct kvm_dirty_log log;
2969
2970 r = -EFAULT;
2f366987 2971 if (copy_from_user(&log, argp, sizeof log))
6aa8b732 2972 goto out;
2c6f5df9 2973 r = kvm_vm_ioctl_get_dirty_log(kvm, &log);
6aa8b732
AK		 2974 if (r)
2975 goto out;
2976 break;
2977 }
e8207547
AK		 2978 case KVM_SET_MEMORY_ALIAS: {
2979 struct kvm_memory_alias alias;
2980
2981 r = -EFAULT;
2982 if (copy_from_user(&alias, argp, sizeof alias))
2983 goto out;
2984 r = kvm_vm_ioctl_set_memory_alias(kvm, &alias);
2985 if (r)
2986 goto out;
2987 break;
2988 }
85f455f7
ED		 2989 case KVM_CREATE_IRQCHIP:
2990 r = -ENOMEM;
2991 kvm->vpic = kvm_create_pic(kvm);
1fd4f2a5
ED		 2992 if (kvm->vpic) {
2993 r = kvm_ioapic_init(kvm);
2994 if (r) {
2995 kfree(kvm->vpic);
2996 kvm->vpic = NULL;
2997 goto out;
2998 }
2999 }
85f455f7
ED		 3000 else
3001 goto out;
3002 break;
3003 case KVM_IRQ_LINE: {
3004 struct kvm_irq_level irq_event;
3005
3006 r = -EFAULT;
3007 if (copy_from_user(&irq_event, argp, sizeof irq_event))
3008 goto out;
3009 if (irqchip_in_kernel(kvm)) {
9cf98828 3010 mutex_lock(&kvm->lock);
85f455f7
ED		 3011 if (irq_event.irq < 16)
3012 kvm_pic_set_irq(pic_irqchip(kvm),
3013 irq_event.irq,
3014 irq_event.level);
1fd4f2a5
ED		 3015 kvm_ioapic_set_irq(kvm->vioapic,
3016 irq_event.irq,
3017 irq_event.level);
9cf98828 3018 mutex_unlock(&kvm->lock);
85f455f7
ED		 3019 r = 0;
3020 }
3021 break;
3022 }
6ceb9d79
HQ		 3023 case KVM_GET_IRQCHIP: {
3024 /* 0: PIC master, 1: PIC slave, 2: IOAPIC */
3025 struct kvm_irqchip chip;
3026
3027 r = -EFAULT;
3028 if (copy_from_user(&chip, argp, sizeof chip))
3029 goto out;
3030 r = -ENXIO;
3031 if (!irqchip_in_kernel(kvm))
3032 goto out;
3033 r = kvm_vm_ioctl_get_irqchip(kvm, &chip);
3034 if (r)
3035 goto out;
3036 r = -EFAULT;
3037 if (copy_to_user(argp, &chip, sizeof chip))
3038 goto out;
3039 r = 0;
3040 break;
3041 }
3042 case KVM_SET_IRQCHIP: {
3043 /* 0: PIC master, 1: PIC slave, 2: IOAPIC */
3044 struct kvm_irqchip chip;
3045
3046 r = -EFAULT;
3047 if (copy_from_user(&chip, argp, sizeof chip))
3048 goto out;
3049 r = -ENXIO;
3050 if (!irqchip_in_kernel(kvm))
3051 goto out;
3052 r = kvm_vm_ioctl_set_irqchip(kvm, &chip);
3053 if (r)
3054 goto out;
3055 r = 0;
3056 break;
3057 }
f17abe9a
AK		 3058 default:
3059 ;
3060 }
3061out:
3062 return r;
3063}
3064
3065static struct page *kvm_vm_nopage(struct vm_area_struct *vma,
3066 unsigned long address,
3067 int *type)
3068{
3069 struct kvm *kvm = vma->vm_file->private_data;
3070 unsigned long pgoff;
f17abe9a
AK		 3071 struct page *page;
3072
f17abe9a 3073 pgoff = ((address - vma->vm_start) >> PAGE_SHIFT) + vma->vm_pgoff;
954bbbc2 3074 page = gfn_to_page(kvm, pgoff);
f17abe9a
AK		 3075 if (!page)
3076 return NOPAGE_SIGBUS;
3077 get_page(page);
cd0d9137
NAQ		 3078 if (type != NULL)
3079 *type = VM_FAULT_MINOR;
3080
f17abe9a
AK		 3081 return page;
3082}
3083
3084static struct vm_operations_struct kvm_vm_vm_ops = {
3085 .nopage = kvm_vm_nopage,
3086};
3087
3088static int kvm_vm_mmap(struct file *file, struct vm_area_struct *vma)
3089{
3090 vma->vm_ops = &kvm_vm_vm_ops;
3091 return 0;
3092}
3093
3094static struct file_operations kvm_vm_fops = {
3095 .release = kvm_vm_release,
3096 .unlocked_ioctl = kvm_vm_ioctl,
3097 .compat_ioctl = kvm_vm_ioctl,
3098 .mmap = kvm_vm_mmap,
3099};
3100
3101static int kvm_dev_ioctl_create_vm(void)
3102{
3103 int fd, r;
3104 struct inode *inode;
3105 struct file *file;
3106 struct kvm *kvm;
3107
f17abe9a 3108 kvm = kvm_create_vm();
d6d28168
AK		 3109 if (IS_ERR(kvm))
3110 return PTR_ERR(kvm);
3111 r = anon_inode_getfd(&fd, &inode, &file, "kvm-vm", &kvm_vm_fops, kvm);
3112 if (r) {
3113 kvm_destroy_vm(kvm);
3114 return r;
f17abe9a
AK		 3115 }
3116
bccf2150 3117 kvm->filp = file;
f17abe9a 3118
f17abe9a 3119 return fd;
f17abe9a
AK		 3120}
3121
3122static long kvm_dev_ioctl(struct file *filp,
3123 unsigned int ioctl, unsigned long arg)
3124{
3125 void __user *argp = (void __user *)arg;
07c45a36 3126 long r = -EINVAL;
f17abe9a
AK		 3127
3128 switch (ioctl) {
3129 case KVM_GET_API_VERSION:
f0fe5108
AK		 3130 r = -EINVAL;
3131 if (arg)
3132 goto out;
f17abe9a
AK		 3133 r = KVM_API_VERSION;
3134 break;
3135 case KVM_CREATE_VM:
f0fe5108
AK		 3136 r = -EINVAL;
3137 if (arg)
3138 goto out;
f17abe9a
AK		 3139 r = kvm_dev_ioctl_create_vm();
3140 break;
6aa8b732 3141 case KVM_GET_MSR_INDEX_LIST: {
2f366987 3142 struct kvm_msr_list __user *user_msr_list = argp;
6aa8b732
AK		 3143 struct kvm_msr_list msr_list;
3144 unsigned n;
3145
3146 r = -EFAULT;
3147 if (copy_from_user(&msr_list, user_msr_list, sizeof msr_list))
3148 goto out;
3149 n = msr_list.nmsrs;
6f00e68f 3150 msr_list.nmsrs = num_msrs_to_save + ARRAY_SIZE(emulated_msrs);
6aa8b732
AK		 3151 if (copy_to_user(user_msr_list, &msr_list, sizeof msr_list))
3152 goto out;
3153 r = -E2BIG;
bf591b24 3154 if (n < num_msrs_to_save)
6aa8b732
AK		 3155 goto out;
3156 r = -EFAULT;
3157 if (copy_to_user(user_msr_list->indices, &msrs_to_save,
bf591b24 3158 num_msrs_to_save * sizeof(u32)))
6aa8b732 3159 goto out;
6f00e68f
AK		 3160 if (copy_to_user(user_msr_list->indices
3161 + num_msrs_to_save * sizeof(u32),
3162 &emulated_msrs,
3163 ARRAY_SIZE(emulated_msrs) * sizeof(u32)))
3164 goto out;
6aa8b732 3165 r = 0;
cc1d8955 3166 break;
6aa8b732 3167 }
85f455f7
ED		 3168 case KVM_CHECK_EXTENSION: {
3169 int ext = (long)argp;
3170
3171 switch (ext) {
3172 case KVM_CAP_IRQCHIP:
b6958ce4 3173 case KVM_CAP_HLT:
85f455f7
ED		 3174 r = 1;
3175 break;
3176 default:
3177 r = 0;
3178 break;
3179 }
5d308f45 3180 break;
85f455f7 3181 }
07c45a36
AK		 3182 case KVM_GET_VCPU_MMAP_SIZE:
3183 r = -EINVAL;
3184 if (arg)
3185 goto out;
039576c0 3186 r = 2 * PAGE_SIZE;
07c45a36 3187 break;
6aa8b732
AK		 3188 default:
3189 ;
3190 }
3191out:
3192 return r;
3193}
3194
6aa8b732 3195static struct file_operations kvm_chardev_ops = {
6aa8b732
AK		 3196 .unlocked_ioctl = kvm_dev_ioctl,
3197 .compat_ioctl = kvm_dev_ioctl,
6aa8b732
AK		 3198};
3199
3200static struct miscdevice kvm_dev = {
bbe4432e 3201 KVM_MINOR,
6aa8b732
AK		 3202 "kvm",
3203 &kvm_chardev_ops,
3204};
3205
774c47f1
AK		 3206/*
3207 * Make sure that a cpu that is being hot-unplugged does not have any vcpus
3208 * cached on it.
3209 */
3210static void decache_vcpus_on_cpu(int cpu)
3211{
3212 struct kvm *vm;
3213 struct kvm_vcpu *vcpu;
3214 int i;
3215
3216 spin_lock(&kvm_lock);
11ec2804 3217 list_for_each_entry(vm, &vm_list, vm_list)
774c47f1 3218 for (i = 0; i < KVM_MAX_VCPUS; ++i) {
fb3f0f51
RR		 3219 vcpu = vm->vcpus[i];
3220 if (!vcpu)
3221 continue;
774c47f1
AK		 3222 /*
3223 * If the vcpu is locked, then it is running on some
3224 * other cpu and therefore it is not cached on the
3225 * cpu in question.
3226 *
3227 * If it's not locked, check the last cpu it executed
3228 * on.
3229 */
3230 if (mutex_trylock(&vcpu->mutex)) {
3231 if (vcpu->cpu == cpu) {
cbdd1bea 3232 kvm_x86_ops->vcpu_decache(vcpu);
774c47f1
AK		 3233 vcpu->cpu = -1;
3234 }
3235 mutex_unlock(&vcpu->mutex);
3236 }
3237 }
3238 spin_unlock(&kvm_lock);
3239}
3240
1b6c0168
AK		 3241static void hardware_enable(void *junk)
3242{
3243 int cpu = raw_smp_processor_id();
3244
3245 if (cpu_isset(cpu, cpus_hardware_enabled))
3246 return;
3247 cpu_set(cpu, cpus_hardware_enabled);
cbdd1bea 3248 kvm_x86_ops->hardware_enable(NULL);
1b6c0168
AK		 3249}
3250
3251static void hardware_disable(void *junk)
3252{
3253 int cpu = raw_smp_processor_id();
3254
3255 if (!cpu_isset(cpu, cpus_hardware_enabled))
3256 return;
3257 cpu_clear(cpu, cpus_hardware_enabled);
3258 decache_vcpus_on_cpu(cpu);
cbdd1bea 3259 kvm_x86_ops->hardware_disable(NULL);
1b6c0168
AK		 3260}
3261
774c47f1
AK		 3262static int kvm_cpu_hotplug(struct notifier_block *notifier, unsigned long val,
3263 void *v)
3264{
3265 int cpu = (long)v;
3266
3267 switch (val) {
cec9ad27
AK		 3268 case CPU_DYING:
3269 case CPU_DYING_FROZEN:
6ec8a856
AK		 3270 printk(KERN_INFO "kvm: disabling virtualization on CPU%d\n",
3271 cpu);
3272 hardware_disable(NULL);
3273 break;
774c47f1 3274 case CPU_UP_CANCELED:
8bb78442 3275 case CPU_UP_CANCELED_FROZEN:
43934a38
JK		 3276 printk(KERN_INFO "kvm: disabling virtualization on CPU%d\n",
3277 cpu);
1b6c0168 3278 smp_call_function_single(cpu, hardware_disable, NULL, 0, 1);
774c47f1 3279 break;
43934a38 3280 case CPU_ONLINE:
8bb78442 3281 case CPU_ONLINE_FROZEN:
43934a38
JK		 3282 printk(KERN_INFO "kvm: enabling virtualization on CPU%d\n",
3283 cpu);
1b6c0168 3284 smp_call_function_single(cpu, hardware_enable, NULL, 0, 1);
774c47f1
AK		 3285 break;
3286 }
3287 return NOTIFY_OK;
3288}
3289
9a2b85c6
RR		 3290static int kvm_reboot(struct notifier_block *notifier, unsigned long val,
3291 void *v)
3292{
3293 if (val == SYS_RESTART) {
3294 /*
3295 * Some (well, at least mine) BIOSes hang on reboot if
3296 * in vmx root mode.
3297 */
3298 printk(KERN_INFO "kvm: exiting hardware virtualization\n");
3299 on_each_cpu(hardware_disable, NULL, 0, 1);
3300 }
3301 return NOTIFY_OK;
3302}
3303
3304static struct notifier_block kvm_reboot_notifier = {
3305 .notifier_call = kvm_reboot,
3306 .priority = 0,
3307};
3308
2eeb2e94
GH		 3309void kvm_io_bus_init(struct kvm_io_bus *bus)
3310{
3311 memset(bus, 0, sizeof(*bus));
3312}
3313
3314void kvm_io_bus_destroy(struct kvm_io_bus *bus)
3315{
3316 int i;
3317
3318 for (i = 0; i < bus->dev_count; i++) {
3319 struct kvm_io_device *pos = bus->devs[i];
3320
3321 kvm_iodevice_destructor(pos);
3322 }
3323}
3324
3325struct kvm_io_device *kvm_io_bus_find_dev(struct kvm_io_bus *bus, gpa_t addr)
3326{
3327 int i;
3328
3329 for (i = 0; i < bus->dev_count; i++) {
3330 struct kvm_io_device *pos = bus->devs[i];
3331
3332 if (pos->in_range(pos, addr))
3333 return pos;
3334 }
3335
3336 return NULL;
3337}
3338
3339void kvm_io_bus_register_dev(struct kvm_io_bus *bus, struct kvm_io_device *dev)
3340{
3341 BUG_ON(bus->dev_count > (NR_IOBUS_DEVS-1));
3342
3343 bus->devs[bus->dev_count++] = dev;
3344}
3345
774c47f1
AK		 3346static struct notifier_block kvm_cpu_notifier = {
3347 .notifier_call = kvm_cpu_hotplug,
3348 .priority = 20, /* must be > scheduler priority */
3349};
3350
1165f5fe
AK		 3351static u64 stat_get(void *_offset)
3352{
3353 unsigned offset = (long)_offset;
3354 u64 total = 0;
3355 struct kvm *kvm;
3356 struct kvm_vcpu *vcpu;
3357 int i;
3358
3359 spin_lock(&kvm_lock);
3360 list_for_each_entry(kvm, &vm_list, vm_list)
3361 for (i = 0; i < KVM_MAX_VCPUS; ++i) {
fb3f0f51
RR		 3362 vcpu = kvm->vcpus[i];
3363 if (vcpu)
3364 total += *(u32 *)((void *)vcpu + offset);
1165f5fe
AK		 3365 }
3366 spin_unlock(&kvm_lock);
3367 return total;
3368}
3369
3dea7ca7 3370DEFINE_SIMPLE_ATTRIBUTE(stat_fops, stat_get, NULL, "%llu\n");
1165f5fe 3371
6aa8b732
AK		 3372static __init void kvm_init_debug(void)
3373{
3374 struct kvm_stats_debugfs_item *p;
3375
8b6d44c7 3376 debugfs_dir = debugfs_create_dir("kvm", NULL);
6aa8b732 3377 for (p = debugfs_entries; p->name; ++p)
1165f5fe
AK		 3378 p->dentry = debugfs_create_file(p->name, 0444, debugfs_dir,
3379 (void *)(long)p->offset,
3380 &stat_fops);
6aa8b732
AK		 3381}
3382
3383static void kvm_exit_debug(void)
3384{
3385 struct kvm_stats_debugfs_item *p;
3386
3387 for (p = debugfs_entries; p->name; ++p)
3388 debugfs_remove(p->dentry);
3389 debugfs_remove(debugfs_dir);
3390}
3391
59ae6c6b
AK		 3392static int kvm_suspend(struct sys_device *dev, pm_message_t state)
3393{
4267c41a 3394 hardware_disable(NULL);
59ae6c6b
AK		 3395 return 0;
3396}
3397
3398static int kvm_resume(struct sys_device *dev)
3399{
4267c41a 3400 hardware_enable(NULL);
59ae6c6b
AK		 3401 return 0;
3402}
3403
3404static struct sysdev_class kvm_sysdev_class = {
af5ca3f4 3405 .name = "kvm",
59ae6c6b
AK		 3406 .suspend = kvm_suspend,
3407 .resume = kvm_resume,
3408};
3409
3410static struct sys_device kvm_sysdev = {
3411 .id = 0,
3412 .cls = &kvm_sysdev_class,
3413};
3414
6aa8b732
AK		 3415hpa_t bad_page_address;
3416
15ad7146
AK		 3417static inline
3418struct kvm_vcpu *preempt_notifier_to_vcpu(struct preempt_notifier *pn)
3419{
3420 return container_of(pn, struct kvm_vcpu, preempt_notifier);
3421}
3422
3423static void kvm_sched_in(struct preempt_notifier *pn, int cpu)
3424{
3425 struct kvm_vcpu *vcpu = preempt_notifier_to_vcpu(pn);
3426
cbdd1bea 3427 kvm_x86_ops->vcpu_load(vcpu, cpu);
15ad7146
AK		 3428}
3429
3430static void kvm_sched_out(struct preempt_notifier *pn,
3431 struct task_struct *next)
3432{
3433 struct kvm_vcpu *vcpu = preempt_notifier_to_vcpu(pn);
3434
cbdd1bea 3435 kvm_x86_ops->vcpu_put(vcpu);
15ad7146
AK		 3436}
3437
cbdd1bea 3438int kvm_init_x86(struct kvm_x86_ops *ops, unsigned int vcpu_size,
c16f862d 3439 struct module *module)
6aa8b732
AK		 3440{
3441 int r;
002c7f7c 3442 int cpu;
6aa8b732 3443
cbdd1bea 3444 if (kvm_x86_ops) {
09db28b8
YI		 3445 printk(KERN_ERR "kvm: already loaded the other module\n");
3446 return -EEXIST;
3447 }
3448
e097f35c 3449 if (!ops->cpu_has_kvm_support()) {
6aa8b732
AK		 3450 printk(KERN_ERR "kvm: no hardware support\n");
3451 return -EOPNOTSUPP;
3452 }
e097f35c 3453 if (ops->disabled_by_bios()) {
6aa8b732
AK		 3454 printk(KERN_ERR "kvm: disabled by bios\n");
3455 return -EOPNOTSUPP;
3456 }
3457
cbdd1bea 3458 kvm_x86_ops = ops;
e097f35c 3459
cbdd1bea 3460 r = kvm_x86_ops->hardware_setup();
6aa8b732 3461 if (r < 0)
ca45aaae 3462 goto out;
6aa8b732 3463
002c7f7c
YS		 3464 for_each_online_cpu(cpu) {
3465 smp_call_function_single(cpu,
cbdd1bea 3466 kvm_x86_ops->check_processor_compatibility,
002c7f7c
YS		 3467 &r, 0, 1);
3468 if (r < 0)
3469 goto out_free_0;
3470 }
3471
1b6c0168 3472 on_each_cpu(hardware_enable, NULL, 0, 1);
774c47f1
AK		 3473 r = register_cpu_notifier(&kvm_cpu_notifier);
3474 if (r)
3475 goto out_free_1;
6aa8b732
AK		 3476 register_reboot_notifier(&kvm_reboot_notifier);
3477
59ae6c6b
AK		 3478 r = sysdev_class_register(&kvm_sysdev_class);
3479 if (r)
3480 goto out_free_2;
3481
3482 r = sysdev_register(&kvm_sysdev);
3483 if (r)
3484 goto out_free_3;
3485
c16f862d
RR		 3486 /* A kmem cache lets us meet the alignment requirements of fx_save. */
3487 kvm_vcpu_cache = kmem_cache_create("kvm_vcpu", vcpu_size,
3488 __alignof__(struct kvm_vcpu), 0, 0);
3489 if (!kvm_vcpu_cache) {
3490 r = -ENOMEM;
3491 goto out_free_4;
3492 }
3493
6aa8b732
AK		 3494 kvm_chardev_ops.owner = module;
3495
3496 r = misc_register(&kvm_dev);
3497 if (r) {
3498 printk (KERN_ERR "kvm: misc device register failed\n");
3499 goto out_free;
3500 }
3501
15ad7146
AK		 3502 kvm_preempt_ops.sched_in = kvm_sched_in;
3503 kvm_preempt_ops.sched_out = kvm_sched_out;
3504
c7addb90
AK		 3505 kvm_mmu_set_nonpresent_ptes(0ull, 0ull);
3506
3507 return 0;
6aa8b732
AK		 3508
3509out_free:
c16f862d
RR		 3510 kmem_cache_destroy(kvm_vcpu_cache);
3511out_free_4:
59ae6c6b
AK		 3512 sysdev_unregister(&kvm_sysdev);
3513out_free_3:
3514 sysdev_class_unregister(&kvm_sysdev_class);
3515out_free_2:
6aa8b732 3516 unregister_reboot_notifier(&kvm_reboot_notifier);
774c47f1
AK		 3517 unregister_cpu_notifier(&kvm_cpu_notifier);
3518out_free_1:
1b6c0168 3519 on_each_cpu(hardware_disable, NULL, 0, 1);
002c7f7c 3520out_free_0:
cbdd1bea 3521 kvm_x86_ops->hardware_unsetup();
ca45aaae 3522out:
cbdd1bea 3523 kvm_x86_ops = NULL;
6aa8b732
AK		 3524 return r;
3525}
3526
cbdd1bea 3527void kvm_exit_x86(void)
6aa8b732
AK		 3528{
3529 misc_deregister(&kvm_dev);
c16f862d 3530 kmem_cache_destroy(kvm_vcpu_cache);
59ae6c6b
AK		 3531 sysdev_unregister(&kvm_sysdev);
3532 sysdev_class_unregister(&kvm_sysdev_class);
6aa8b732 3533 unregister_reboot_notifier(&kvm_reboot_notifier);
59ae6c6b 3534 unregister_cpu_notifier(&kvm_cpu_notifier);
1b6c0168 3535 on_each_cpu(hardware_disable, NULL, 0, 1);
cbdd1bea
CE		 3536 kvm_x86_ops->hardware_unsetup();
3537 kvm_x86_ops = NULL;
6aa8b732
AK		 3538}
3539
3540static __init int kvm_init(void)
3541{
3542 static struct page *bad_page;
37e29d90
AK		 3543 int r;
3544
b5a33a75
AK		 3545 r = kvm_mmu_module_init();
3546 if (r)
3547 goto out4;
3548
6aa8b732
AK		 3549 kvm_init_debug();
3550
bf591b24
MR		 3551 kvm_init_msr_list();
3552
6aa8b732
AK		 3553 if ((bad_page = alloc_page(GFP_KERNEL)) == NULL) {
3554 r = -ENOMEM;
3555 goto out;
3556 }
3557
3558 bad_page_address = page_to_pfn(bad_page) << PAGE_SHIFT;
3559 memset(__va(bad_page_address), 0, PAGE_SIZE);
3560
58e690e6 3561 return 0;
6aa8b732
AK		 3562
3563out:
3564 kvm_exit_debug();
b5a33a75
AK		 3565 kvm_mmu_module_exit();
3566out4:
6aa8b732
AK		 3567 return r;
3568}
3569
3570static __exit void kvm_exit(void)
3571{
3572 kvm_exit_debug();
3573 __free_page(pfn_to_page(bad_page_address >> PAGE_SHIFT));
b5a33a75 3574 kvm_mmu_module_exit();
6aa8b732
AK		 3575}
3576
3577module_init(kvm_init)
3578module_exit(kvm_exit)
3579
cbdd1bea
CE		 3580EXPORT_SYMBOL_GPL(kvm_init_x86);
3581EXPORT_SYMBOL_GPL(kvm_exit_x86);
Clone of davem's net-next-2.6 tree