Merge git://git.kernel.org/pub/scm/linux/kernel/git/bart/ide-2.6
[linux-2.6] / drivers / kvm / kvm_main.c
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"
19 #include "x86_emulate.h"
20 #include "segment_descriptor.h"
21 #include "irq.h"
22
23 #include <linux/kvm.h>
24 #include <linux/module.h>
25 #include <linux/errno.h>
26 #include <linux/percpu.h>
27 #include <linux/gfp.h>
28 #include <linux/mm.h>
29 #include <linux/miscdevice.h>
30 #include <linux/vmalloc.h>
31 #include <linux/reboot.h>
32 #include <linux/debugfs.h>
33 #include <linux/highmem.h>
34 #include <linux/file.h>
35 #include <linux/sysdev.h>
36 #include <linux/cpu.h>
37 #include <linux/sched.h>
38 #include <linux/cpumask.h>
39 #include <linux/smp.h>
40 #include <linux/anon_inodes.h>
41 #include <linux/profile.h>
42
43 #include <asm/processor.h>
44 #include <asm/msr.h>
45 #include <asm/io.h>
46 #include <asm/uaccess.h>
47 #include <asm/desc.h>
48
49 MODULE_AUTHOR("Qumranet");
50 MODULE_LICENSE("GPL");
51
52 static DEFINE_SPINLOCK(kvm_lock);
53 static LIST_HEAD(vm_list);
54
55 static cpumask_t cpus_hardware_enabled;
56
57 struct kvm_x86_ops *kvm_x86_ops;
58 struct kmem_cache *kvm_vcpu_cache;
59 EXPORT_SYMBOL_GPL(kvm_vcpu_cache);
60
61 static __read_mostly struct preempt_ops kvm_preempt_ops;
62
63 #define STAT_OFFSET(x) offsetof(struct kvm_vcpu, stat.x)
64
65 static struct kvm_stats_debugfs_item {
66         const char *name;
67         int offset;
68         struct dentry *dentry;
69 } debugfs_entries[] = {
70         { "pf_fixed", STAT_OFFSET(pf_fixed) },
71         { "pf_guest", STAT_OFFSET(pf_guest) },
72         { "tlb_flush", STAT_OFFSET(tlb_flush) },
73         { "invlpg", STAT_OFFSET(invlpg) },
74         { "exits", STAT_OFFSET(exits) },
75         { "io_exits", STAT_OFFSET(io_exits) },
76         { "mmio_exits", STAT_OFFSET(mmio_exits) },
77         { "signal_exits", STAT_OFFSET(signal_exits) },
78         { "irq_window", STAT_OFFSET(irq_window_exits) },
79         { "halt_exits", STAT_OFFSET(halt_exits) },
80         { "halt_wakeup", STAT_OFFSET(halt_wakeup) },
81         { "request_irq", STAT_OFFSET(request_irq_exits) },
82         { "irq_exits", STAT_OFFSET(irq_exits) },
83         { "light_exits", STAT_OFFSET(light_exits) },
84         { "efer_reload", STAT_OFFSET(efer_reload) },
85         { NULL }
86 };
87
88 static struct dentry *debugfs_dir;
89
90 #define MAX_IO_MSRS 256
91
92 #define CR0_RESERVED_BITS                                               \
93         (~(unsigned long)(X86_CR0_PE | X86_CR0_MP | X86_CR0_EM | X86_CR0_TS \
94                           | X86_CR0_ET | X86_CR0_NE | X86_CR0_WP | X86_CR0_AM \
95                           | X86_CR0_NW | X86_CR0_CD | X86_CR0_PG))
96 #define CR4_RESERVED_BITS                                               \
97         (~(unsigned long)(X86_CR4_VME | X86_CR4_PVI | X86_CR4_TSD | X86_CR4_DE\
98                           | X86_CR4_PSE | X86_CR4_PAE | X86_CR4_MCE     \
99                           | X86_CR4_PGE | X86_CR4_PCE | X86_CR4_OSFXSR  \
100                           | X86_CR4_OSXMMEXCPT | X86_CR4_VMXE))
101
102 #define CR8_RESERVED_BITS (~(unsigned long)X86_CR8_TPR)
103 #define EFER_RESERVED_BITS 0xfffffffffffff2fe
104
105 #ifdef CONFIG_X86_64
106 // LDT or TSS descriptor in the GDT. 16 bytes.
107 struct segment_descriptor_64 {
108         struct segment_descriptor s;
109         u32 base_higher;
110         u32 pad_zero;
111 };
112
113 #endif
114
115 static long kvm_vcpu_ioctl(struct file *file, unsigned int ioctl,
116                            unsigned long arg);
117
118 unsigned long segment_base(u16 selector)
119 {
120         struct descriptor_table gdt;
121         struct segment_descriptor *d;
122         unsigned long table_base;
123         typedef unsigned long ul;
124         unsigned long v;
125
126         if (selector == 0)
127                 return 0;
128
129         asm ("sgdt %0" : "=m"(gdt));
130         table_base = gdt.base;
131
132         if (selector & 4) {           /* from ldt */
133                 u16 ldt_selector;
134
135                 asm ("sldt %0" : "=g"(ldt_selector));
136                 table_base = segment_base(ldt_selector);
137         }
138         d = (struct segment_descriptor *)(table_base + (selector & ~7));
139         v = d->base_low | ((ul)d->base_mid << 16) | ((ul)d->base_high << 24);
140 #ifdef CONFIG_X86_64
141         if (d->system == 0
142             && (d->type == 2 || d->type == 9 || d->type == 11))
143                 v |= ((ul)((struct segment_descriptor_64 *)d)->base_higher) << 32;
144 #endif
145         return v;
146 }
147 EXPORT_SYMBOL_GPL(segment_base);
148
149 static inline int valid_vcpu(int n)
150 {
151         return likely(n >= 0 && n < KVM_MAX_VCPUS);
152 }
153
154 void kvm_load_guest_fpu(struct kvm_vcpu *vcpu)
155 {
156         if (!vcpu->fpu_active || vcpu->guest_fpu_loaded)
157                 return;
158
159         vcpu->guest_fpu_loaded = 1;
160         fx_save(&vcpu->host_fx_image);
161         fx_restore(&vcpu->guest_fx_image);
162 }
163 EXPORT_SYMBOL_GPL(kvm_load_guest_fpu);
164
165 void kvm_put_guest_fpu(struct kvm_vcpu *vcpu)
166 {
167         if (!vcpu->guest_fpu_loaded)
168                 return;
169
170         vcpu->guest_fpu_loaded = 0;
171         fx_save(&vcpu->guest_fx_image);
172         fx_restore(&vcpu->host_fx_image);
173 }
174 EXPORT_SYMBOL_GPL(kvm_put_guest_fpu);
175
176 /*
177  * Switches to specified vcpu, until a matching vcpu_put()
178  */
179 static void vcpu_load(struct kvm_vcpu *vcpu)
180 {
181         int cpu;
182
183         mutex_lock(&vcpu->mutex);
184         cpu = get_cpu();
185         preempt_notifier_register(&vcpu->preempt_notifier);
186         kvm_x86_ops->vcpu_load(vcpu, cpu);
187         put_cpu();
188 }
189
190 static void vcpu_put(struct kvm_vcpu *vcpu)
191 {
192         preempt_disable();
193         kvm_x86_ops->vcpu_put(vcpu);
194         preempt_notifier_unregister(&vcpu->preempt_notifier);
195         preempt_enable();
196         mutex_unlock(&vcpu->mutex);
197 }
198
199 static void ack_flush(void *_completed)
200 {
201         atomic_t *completed = _completed;
202
203         atomic_inc(completed);
204 }
205
206 void kvm_flush_remote_tlbs(struct kvm *kvm)
207 {
208         int i, cpu, needed;
209         cpumask_t cpus;
210         struct kvm_vcpu *vcpu;
211         atomic_t completed;
212
213         atomic_set(&completed, 0);
214         cpus_clear(cpus);
215         needed = 0;
216         for (i = 0; i < KVM_MAX_VCPUS; ++i) {
217                 vcpu = kvm->vcpus[i];
218                 if (!vcpu)
219                         continue;
220                 if (test_and_set_bit(KVM_TLB_FLUSH, &vcpu->requests))
221                         continue;
222                 cpu = vcpu->cpu;
223                 if (cpu != -1 && cpu != raw_smp_processor_id())
224                         if (!cpu_isset(cpu, cpus)) {
225                                 cpu_set(cpu, cpus);
226                                 ++needed;
227                         }
228         }
229
230         /*
231          * We really want smp_call_function_mask() here.  But that's not
232          * available, so ipi all cpus in parallel and wait for them
233          * to complete.
234          */
235         for (cpu = first_cpu(cpus); cpu != NR_CPUS; cpu = next_cpu(cpu, cpus))
236                 smp_call_function_single(cpu, ack_flush, &completed, 1, 0);
237         while (atomic_read(&completed) != needed) {
238                 cpu_relax();
239                 barrier();
240         }
241 }
242
243 int kvm_vcpu_init(struct kvm_vcpu *vcpu, struct kvm *kvm, unsigned id)
244 {
245         struct page *page;
246         int r;
247
248         mutex_init(&vcpu->mutex);
249         vcpu->cpu = -1;
250         vcpu->mmu.root_hpa = INVALID_PAGE;
251         vcpu->kvm = kvm;
252         vcpu->vcpu_id = id;
253         if (!irqchip_in_kernel(kvm) || id == 0)
254                 vcpu->mp_state = VCPU_MP_STATE_RUNNABLE;
255         else
256                 vcpu->mp_state = VCPU_MP_STATE_UNINITIALIZED;
257         init_waitqueue_head(&vcpu->wq);
258
259         page = alloc_page(GFP_KERNEL | __GFP_ZERO);
260         if (!page) {
261                 r = -ENOMEM;
262                 goto fail;
263         }
264         vcpu->run = page_address(page);
265
266         page = alloc_page(GFP_KERNEL | __GFP_ZERO);
267         if (!page) {
268                 r = -ENOMEM;
269                 goto fail_free_run;
270         }
271         vcpu->pio_data = page_address(page);
272
273         r = kvm_mmu_create(vcpu);
274         if (r < 0)
275                 goto fail_free_pio_data;
276
277         return 0;
278
279 fail_free_pio_data:
280         free_page((unsigned long)vcpu->pio_data);
281 fail_free_run:
282         free_page((unsigned long)vcpu->run);
283 fail:
284         return -ENOMEM;
285 }
286 EXPORT_SYMBOL_GPL(kvm_vcpu_init);
287
288 void kvm_vcpu_uninit(struct kvm_vcpu *vcpu)
289 {
290         kvm_mmu_destroy(vcpu);
291         if (vcpu->apic)
292                 hrtimer_cancel(&vcpu->apic->timer.dev);
293         kvm_free_apic(vcpu->apic);
294         free_page((unsigned long)vcpu->pio_data);
295         free_page((unsigned long)vcpu->run);
296 }
297 EXPORT_SYMBOL_GPL(kvm_vcpu_uninit);
298
299 static struct kvm *kvm_create_vm(void)
300 {
301         struct kvm *kvm = kzalloc(sizeof(struct kvm), GFP_KERNEL);
302
303         if (!kvm)
304                 return ERR_PTR(-ENOMEM);
305
306         kvm_io_bus_init(&kvm->pio_bus);
307         mutex_init(&kvm->lock);
308         INIT_LIST_HEAD(&kvm->active_mmu_pages);
309         kvm_io_bus_init(&kvm->mmio_bus);
310         spin_lock(&kvm_lock);
311         list_add(&kvm->vm_list, &vm_list);
312         spin_unlock(&kvm_lock);
313         return kvm;
314 }
315
316 /*
317  * Free any memory in @free but not in @dont.
318  */
319 static void kvm_free_physmem_slot(struct kvm_memory_slot *free,
320                                   struct kvm_memory_slot *dont)
321 {
322         int i;
323
324         if (!dont || free->phys_mem != dont->phys_mem)
325                 if (free->phys_mem) {
326                         for (i = 0; i < free->npages; ++i)
327                                 if (free->phys_mem[i])
328                                         __free_page(free->phys_mem[i]);
329                         vfree(free->phys_mem);
330                 }
331
332         if (!dont || free->dirty_bitmap != dont->dirty_bitmap)
333                 vfree(free->dirty_bitmap);
334
335         free->phys_mem = NULL;
336         free->npages = 0;
337         free->dirty_bitmap = NULL;
338 }
339
340 static void kvm_free_physmem(struct kvm *kvm)
341 {
342         int i;
343
344         for (i = 0; i < kvm->nmemslots; ++i)
345                 kvm_free_physmem_slot(&kvm->memslots[i], NULL);
346 }
347
348 static void free_pio_guest_pages(struct kvm_vcpu *vcpu)
349 {
350         int i;
351
352         for (i = 0; i < ARRAY_SIZE(vcpu->pio.guest_pages); ++i)
353                 if (vcpu->pio.guest_pages[i]) {
354                         __free_page(vcpu->pio.guest_pages[i]);
355                         vcpu->pio.guest_pages[i] = NULL;
356                 }
357 }
358
359 static void kvm_unload_vcpu_mmu(struct kvm_vcpu *vcpu)
360 {
361         vcpu_load(vcpu);
362         kvm_mmu_unload(vcpu);
363         vcpu_put(vcpu);
364 }
365
366 static void kvm_free_vcpus(struct kvm *kvm)
367 {
368         unsigned int i;
369
370         /*
371          * Unpin any mmu pages first.
372          */
373         for (i = 0; i < KVM_MAX_VCPUS; ++i)
374                 if (kvm->vcpus[i])
375                         kvm_unload_vcpu_mmu(kvm->vcpus[i]);
376         for (i = 0; i < KVM_MAX_VCPUS; ++i) {
377                 if (kvm->vcpus[i]) {
378                         kvm_x86_ops->vcpu_free(kvm->vcpus[i]);
379                         kvm->vcpus[i] = NULL;
380                 }
381         }
382
383 }
384
385 static void kvm_destroy_vm(struct kvm *kvm)
386 {
387         spin_lock(&kvm_lock);
388         list_del(&kvm->vm_list);
389         spin_unlock(&kvm_lock);
390         kvm_io_bus_destroy(&kvm->pio_bus);
391         kvm_io_bus_destroy(&kvm->mmio_bus);
392         kfree(kvm->vpic);
393         kfree(kvm->vioapic);
394         kvm_free_vcpus(kvm);
395         kvm_free_physmem(kvm);
396         kfree(kvm);
397 }
398
399 static int kvm_vm_release(struct inode *inode, struct file *filp)
400 {
401         struct kvm *kvm = filp->private_data;
402
403         kvm_destroy_vm(kvm);
404         return 0;
405 }
406
407 static void inject_gp(struct kvm_vcpu *vcpu)
408 {
409         kvm_x86_ops->inject_gp(vcpu, 0);
410 }
411
412 /*
413  * Load the pae pdptrs.  Return true is they are all valid.
414  */
415 static int load_pdptrs(struct kvm_vcpu *vcpu, unsigned long cr3)
416 {
417         gfn_t pdpt_gfn = cr3 >> PAGE_SHIFT;
418         unsigned offset = ((cr3 & (PAGE_SIZE-1)) >> 5) << 2;
419         int i;
420         u64 *pdpt;
421         int ret;
422         struct page *page;
423         u64 pdpte[ARRAY_SIZE(vcpu->pdptrs)];
424
425         mutex_lock(&vcpu->kvm->lock);
426         page = gfn_to_page(vcpu->kvm, pdpt_gfn);
427         if (!page) {
428                 ret = 0;
429                 goto out;
430         }
431
432         pdpt = kmap_atomic(page, KM_USER0);
433         memcpy(pdpte, pdpt+offset, sizeof(pdpte));
434         kunmap_atomic(pdpt, KM_USER0);
435
436         for (i = 0; i < ARRAY_SIZE(pdpte); ++i) {
437                 if ((pdpte[i] & 1) && (pdpte[i] & 0xfffffff0000001e6ull)) {
438                         ret = 0;
439                         goto out;
440                 }
441         }
442         ret = 1;
443
444         memcpy(vcpu->pdptrs, pdpte, sizeof(vcpu->pdptrs));
445 out:
446         mutex_unlock(&vcpu->kvm->lock);
447
448         return ret;
449 }
450
451 void set_cr0(struct kvm_vcpu *vcpu, unsigned long cr0)
452 {
453         if (cr0 & CR0_RESERVED_BITS) {
454                 printk(KERN_DEBUG "set_cr0: 0x%lx #GP, reserved bits 0x%lx\n",
455                        cr0, vcpu->cr0);
456                 inject_gp(vcpu);
457                 return;
458         }
459
460         if ((cr0 & X86_CR0_NW) && !(cr0 & X86_CR0_CD)) {
461                 printk(KERN_DEBUG "set_cr0: #GP, CD == 0 && NW == 1\n");
462                 inject_gp(vcpu);
463                 return;
464         }
465
466         if ((cr0 & X86_CR0_PG) && !(cr0 & X86_CR0_PE)) {
467                 printk(KERN_DEBUG "set_cr0: #GP, set PG flag "
468                        "and a clear PE flag\n");
469                 inject_gp(vcpu);
470                 return;
471         }
472
473         if (!is_paging(vcpu) && (cr0 & X86_CR0_PG)) {
474 #ifdef CONFIG_X86_64
475                 if ((vcpu->shadow_efer & EFER_LME)) {
476                         int cs_db, cs_l;
477
478                         if (!is_pae(vcpu)) {
479                                 printk(KERN_DEBUG "set_cr0: #GP, start paging "
480                                        "in long mode while PAE is disabled\n");
481                                 inject_gp(vcpu);
482                                 return;
483                         }
484                         kvm_x86_ops->get_cs_db_l_bits(vcpu, &cs_db, &cs_l);
485                         if (cs_l) {
486                                 printk(KERN_DEBUG "set_cr0: #GP, start paging "
487                                        "in long mode while CS.L == 1\n");
488                                 inject_gp(vcpu);
489                                 return;
490
491                         }
492                 } else
493 #endif
494                 if (is_pae(vcpu) && !load_pdptrs(vcpu, vcpu->cr3)) {
495                         printk(KERN_DEBUG "set_cr0: #GP, pdptrs "
496                                "reserved bits\n");
497                         inject_gp(vcpu);
498                         return;
499                 }
500
501         }
502
503         kvm_x86_ops->set_cr0(vcpu, cr0);
504         vcpu->cr0 = cr0;
505
506         mutex_lock(&vcpu->kvm->lock);
507         kvm_mmu_reset_context(vcpu);
508         mutex_unlock(&vcpu->kvm->lock);
509         return;
510 }
511 EXPORT_SYMBOL_GPL(set_cr0);
512
513 void lmsw(struct kvm_vcpu *vcpu, unsigned long msw)
514 {
515         set_cr0(vcpu, (vcpu->cr0 & ~0x0ful) | (msw & 0x0f));
516 }
517 EXPORT_SYMBOL_GPL(lmsw);
518
519 void set_cr4(struct kvm_vcpu *vcpu, unsigned long cr4)
520 {
521         if (cr4 & CR4_RESERVED_BITS) {
522                 printk(KERN_DEBUG "set_cr4: #GP, reserved bits\n");
523                 inject_gp(vcpu);
524                 return;
525         }
526
527         if (is_long_mode(vcpu)) {
528                 if (!(cr4 & X86_CR4_PAE)) {
529                         printk(KERN_DEBUG "set_cr4: #GP, clearing PAE while "
530                                "in long mode\n");
531                         inject_gp(vcpu);
532                         return;
533                 }
534         } else if (is_paging(vcpu) && !is_pae(vcpu) && (cr4 & X86_CR4_PAE)
535                    && !load_pdptrs(vcpu, vcpu->cr3)) {
536                 printk(KERN_DEBUG "set_cr4: #GP, pdptrs reserved bits\n");
537                 inject_gp(vcpu);
538                 return;
539         }
540
541         if (cr4 & X86_CR4_VMXE) {
542                 printk(KERN_DEBUG "set_cr4: #GP, setting VMXE\n");
543                 inject_gp(vcpu);
544                 return;
545         }
546         kvm_x86_ops->set_cr4(vcpu, cr4);
547         vcpu->cr4 = cr4;
548         mutex_lock(&vcpu->kvm->lock);
549         kvm_mmu_reset_context(vcpu);
550         mutex_unlock(&vcpu->kvm->lock);
551 }
552 EXPORT_SYMBOL_GPL(set_cr4);
553
554 void set_cr3(struct kvm_vcpu *vcpu, unsigned long cr3)
555 {
556         if (is_long_mode(vcpu)) {
557                 if (cr3 & CR3_L_MODE_RESERVED_BITS) {
558                         printk(KERN_DEBUG "set_cr3: #GP, reserved bits\n");
559                         inject_gp(vcpu);
560                         return;
561                 }
562         } else {
563                 if (is_pae(vcpu)) {
564                         if (cr3 & CR3_PAE_RESERVED_BITS) {
565                                 printk(KERN_DEBUG
566                                        "set_cr3: #GP, reserved bits\n");
567                                 inject_gp(vcpu);
568                                 return;
569                         }
570                         if (is_paging(vcpu) && !load_pdptrs(vcpu, cr3)) {
571                                 printk(KERN_DEBUG "set_cr3: #GP, pdptrs "
572                                        "reserved bits\n");
573                                 inject_gp(vcpu);
574                                 return;
575                         }
576                 } else {
577                         if (cr3 & CR3_NONPAE_RESERVED_BITS) {
578                                 printk(KERN_DEBUG
579                                        "set_cr3: #GP, reserved bits\n");
580                                 inject_gp(vcpu);
581                                 return;
582                         }
583                 }
584         }
585
586         mutex_lock(&vcpu->kvm->lock);
587         /*
588          * Does the new cr3 value map to physical memory? (Note, we
589          * catch an invalid cr3 even in real-mode, because it would
590          * cause trouble later on when we turn on paging anyway.)
591          *
592          * A real CPU would silently accept an invalid cr3 and would
593          * attempt to use it - with largely undefined (and often hard
594          * to debug) behavior on the guest side.
595          */
596         if (unlikely(!gfn_to_memslot(vcpu->kvm, cr3 >> PAGE_SHIFT)))
597                 inject_gp(vcpu);
598         else {
599                 vcpu->cr3 = cr3;
600                 vcpu->mmu.new_cr3(vcpu);
601         }
602         mutex_unlock(&vcpu->kvm->lock);
603 }
604 EXPORT_SYMBOL_GPL(set_cr3);
605
606 void set_cr8(struct kvm_vcpu *vcpu, unsigned long cr8)
607 {
608         if (cr8 & CR8_RESERVED_BITS) {
609                 printk(KERN_DEBUG "set_cr8: #GP, reserved bits 0x%lx\n", cr8);
610                 inject_gp(vcpu);
611                 return;
612         }
613         if (irqchip_in_kernel(vcpu->kvm))
614                 kvm_lapic_set_tpr(vcpu, cr8);
615         else
616                 vcpu->cr8 = cr8;
617 }
618 EXPORT_SYMBOL_GPL(set_cr8);
619
620 unsigned long get_cr8(struct kvm_vcpu *vcpu)
621 {
622         if (irqchip_in_kernel(vcpu->kvm))
623                 return kvm_lapic_get_cr8(vcpu);
624         else
625                 return vcpu->cr8;
626 }
627 EXPORT_SYMBOL_GPL(get_cr8);
628
629 u64 kvm_get_apic_base(struct kvm_vcpu *vcpu)
630 {
631         if (irqchip_in_kernel(vcpu->kvm))
632                 return vcpu->apic_base;
633         else
634                 return vcpu->apic_base;
635 }
636 EXPORT_SYMBOL_GPL(kvm_get_apic_base);
637
638 void kvm_set_apic_base(struct kvm_vcpu *vcpu, u64 data)
639 {
640         /* TODO: reserve bits check */
641         if (irqchip_in_kernel(vcpu->kvm))
642                 kvm_lapic_set_base(vcpu, data);
643         else
644                 vcpu->apic_base = data;
645 }
646 EXPORT_SYMBOL_GPL(kvm_set_apic_base);
647
648 void fx_init(struct kvm_vcpu *vcpu)
649 {
650         unsigned after_mxcsr_mask;
651
652         /* Initialize guest FPU by resetting ours and saving into guest's */
653         preempt_disable();
654         fx_save(&vcpu->host_fx_image);
655         fpu_init();
656         fx_save(&vcpu->guest_fx_image);
657         fx_restore(&vcpu->host_fx_image);
658         preempt_enable();
659
660         vcpu->cr0 |= X86_CR0_ET;
661         after_mxcsr_mask = offsetof(struct i387_fxsave_struct, st_space);
662         vcpu->guest_fx_image.mxcsr = 0x1f80;
663         memset((void *)&vcpu->guest_fx_image + after_mxcsr_mask,
664                0, sizeof(struct i387_fxsave_struct) - after_mxcsr_mask);
665 }
666 EXPORT_SYMBOL_GPL(fx_init);
667
668 /*
669  * Allocate some memory and give it an address in the guest physical address
670  * space.
671  *
672  * Discontiguous memory is allowed, mostly for framebuffers.
673  */
674 static int kvm_vm_ioctl_set_memory_region(struct kvm *kvm,
675                                           struct kvm_memory_region *mem)
676 {
677         int r;
678         gfn_t base_gfn;
679         unsigned long npages;
680         unsigned long i;
681         struct kvm_memory_slot *memslot;
682         struct kvm_memory_slot old, new;
683
684         r = -EINVAL;
685         /* General sanity checks */
686         if (mem->memory_size & (PAGE_SIZE - 1))
687                 goto out;
688         if (mem->guest_phys_addr & (PAGE_SIZE - 1))
689                 goto out;
690         if (mem->slot >= KVM_MEMORY_SLOTS)
691                 goto out;
692         if (mem->guest_phys_addr + mem->memory_size < mem->guest_phys_addr)
693                 goto out;
694
695         memslot = &kvm->memslots[mem->slot];
696         base_gfn = mem->guest_phys_addr >> PAGE_SHIFT;
697         npages = mem->memory_size >> PAGE_SHIFT;
698
699         if (!npages)
700                 mem->flags &= ~KVM_MEM_LOG_DIRTY_PAGES;
701
702         mutex_lock(&kvm->lock);
703
704         new = old = *memslot;
705
706         new.base_gfn = base_gfn;
707         new.npages = npages;
708         new.flags = mem->flags;
709
710         /* Disallow changing a memory slot's size. */
711         r = -EINVAL;
712         if (npages && old.npages && npages != old.npages)
713                 goto out_unlock;
714
715         /* Check for overlaps */
716         r = -EEXIST;
717         for (i = 0; i < KVM_MEMORY_SLOTS; ++i) {
718                 struct kvm_memory_slot *s = &kvm->memslots[i];
719
720                 if (s == memslot)
721                         continue;
722                 if (!((base_gfn + npages <= s->base_gfn) ||
723                       (base_gfn >= s->base_gfn + s->npages)))
724                         goto out_unlock;
725         }
726
727         /* Deallocate if slot is being removed */
728         if (!npages)
729                 new.phys_mem = NULL;
730
731         /* Free page dirty bitmap if unneeded */
732         if (!(new.flags & KVM_MEM_LOG_DIRTY_PAGES))
733                 new.dirty_bitmap = NULL;
734
735         r = -ENOMEM;
736
737         /* Allocate if a slot is being created */
738         if (npages && !new.phys_mem) {
739                 new.phys_mem = vmalloc(npages * sizeof(struct page *));
740
741                 if (!new.phys_mem)
742                         goto out_unlock;
743
744                 memset(new.phys_mem, 0, npages * sizeof(struct page *));
745                 for (i = 0; i < npages; ++i) {
746                         new.phys_mem[i] = alloc_page(GFP_HIGHUSER
747                                                      | __GFP_ZERO);
748                         if (!new.phys_mem[i])
749                                 goto out_unlock;
750                         set_page_private(new.phys_mem[i],0);
751                 }
752         }
753
754         /* Allocate page dirty bitmap if needed */
755         if ((new.flags & KVM_MEM_LOG_DIRTY_PAGES) && !new.dirty_bitmap) {
756                 unsigned dirty_bytes = ALIGN(npages, BITS_PER_LONG) / 8;
757
758                 new.dirty_bitmap = vmalloc(dirty_bytes);
759                 if (!new.dirty_bitmap)
760                         goto out_unlock;
761                 memset(new.dirty_bitmap, 0, dirty_bytes);
762         }
763
764         if (mem->slot >= kvm->nmemslots)
765                 kvm->nmemslots = mem->slot + 1;
766
767         *memslot = new;
768
769         kvm_mmu_slot_remove_write_access(kvm, mem->slot);
770         kvm_flush_remote_tlbs(kvm);
771
772         mutex_unlock(&kvm->lock);
773
774         kvm_free_physmem_slot(&old, &new);
775         return 0;
776
777 out_unlock:
778         mutex_unlock(&kvm->lock);
779         kvm_free_physmem_slot(&new, &old);
780 out:
781         return r;
782 }
783
784 /*
785  * Get (and clear) the dirty memory log for a memory slot.
786  */
787 static int kvm_vm_ioctl_get_dirty_log(struct kvm *kvm,
788                                       struct kvm_dirty_log *log)
789 {
790         struct kvm_memory_slot *memslot;
791         int r, i;
792         int n;
793         unsigned long any = 0;
794
795         mutex_lock(&kvm->lock);
796
797         r = -EINVAL;
798         if (log->slot >= KVM_MEMORY_SLOTS)
799                 goto out;
800
801         memslot = &kvm->memslots[log->slot];
802         r = -ENOENT;
803         if (!memslot->dirty_bitmap)
804                 goto out;
805
806         n = ALIGN(memslot->npages, BITS_PER_LONG) / 8;
807
808         for (i = 0; !any && i < n/sizeof(long); ++i)
809                 any = memslot->dirty_bitmap[i];
810
811         r = -EFAULT;
812         if (copy_to_user(log->dirty_bitmap, memslot->dirty_bitmap, n))
813                 goto out;
814
815         /* If nothing is dirty, don't bother messing with page tables. */
816         if (any) {
817                 kvm_mmu_slot_remove_write_access(kvm, log->slot);
818                 kvm_flush_remote_tlbs(kvm);
819                 memset(memslot->dirty_bitmap, 0, n);
820         }
821
822         r = 0;
823
824 out:
825         mutex_unlock(&kvm->lock);
826         return r;
827 }
828
829 /*
830  * Set a new alias region.  Aliases map a portion of physical memory into
831  * another portion.  This is useful for memory windows, for example the PC
832  * VGA region.
833  */
834 static int kvm_vm_ioctl_set_memory_alias(struct kvm *kvm,
835                                          struct kvm_memory_alias *alias)
836 {
837         int r, n;
838         struct kvm_mem_alias *p;
839
840         r = -EINVAL;
841         /* General sanity checks */
842         if (alias->memory_size & (PAGE_SIZE - 1))
843                 goto out;
844         if (alias->guest_phys_addr & (PAGE_SIZE - 1))
845                 goto out;
846         if (alias->slot >= KVM_ALIAS_SLOTS)
847                 goto out;
848         if (alias->guest_phys_addr + alias->memory_size
849             < alias->guest_phys_addr)
850                 goto out;
851         if (alias->target_phys_addr + alias->memory_size
852             < alias->target_phys_addr)
853                 goto out;
854
855         mutex_lock(&kvm->lock);
856
857         p = &kvm->aliases[alias->slot];
858         p->base_gfn = alias->guest_phys_addr >> PAGE_SHIFT;
859         p->npages = alias->memory_size >> PAGE_SHIFT;
860         p->target_gfn = alias->target_phys_addr >> PAGE_SHIFT;
861
862         for (n = KVM_ALIAS_SLOTS; n > 0; --n)
863                 if (kvm->aliases[n - 1].npages)
864                         break;
865         kvm->naliases = n;
866
867         kvm_mmu_zap_all(kvm);
868
869         mutex_unlock(&kvm->lock);
870
871         return 0;
872
873 out:
874         return r;
875 }
876
877 static int kvm_vm_ioctl_get_irqchip(struct kvm *kvm, struct kvm_irqchip *chip)
878 {
879         int r;
880
881         r = 0;
882         switch (chip->chip_id) {
883         case KVM_IRQCHIP_PIC_MASTER:
884                 memcpy (&chip->chip.pic,
885                         &pic_irqchip(kvm)->pics[0],
886                         sizeof(struct kvm_pic_state));
887                 break;
888         case KVM_IRQCHIP_PIC_SLAVE:
889                 memcpy (&chip->chip.pic,
890                         &pic_irqchip(kvm)->pics[1],
891                         sizeof(struct kvm_pic_state));
892                 break;
893         case KVM_IRQCHIP_IOAPIC:
894                 memcpy (&chip->chip.ioapic,
895                         ioapic_irqchip(kvm),
896                         sizeof(struct kvm_ioapic_state));
897                 break;
898         default:
899                 r = -EINVAL;
900                 break;
901         }
902         return r;
903 }
904
905 static int kvm_vm_ioctl_set_irqchip(struct kvm *kvm, struct kvm_irqchip *chip)
906 {
907         int r;
908
909         r = 0;
910         switch (chip->chip_id) {
911         case KVM_IRQCHIP_PIC_MASTER:
912                 memcpy (&pic_irqchip(kvm)->pics[0],
913                         &chip->chip.pic,
914                         sizeof(struct kvm_pic_state));
915                 break;
916         case KVM_IRQCHIP_PIC_SLAVE:
917                 memcpy (&pic_irqchip(kvm)->pics[1],
918                         &chip->chip.pic,
919                         sizeof(struct kvm_pic_state));
920                 break;
921         case KVM_IRQCHIP_IOAPIC:
922                 memcpy (ioapic_irqchip(kvm),
923                         &chip->chip.ioapic,
924                         sizeof(struct kvm_ioapic_state));
925                 break;
926         default:
927                 r = -EINVAL;
928                 break;
929         }
930         kvm_pic_update_irq(pic_irqchip(kvm));
931         return r;
932 }
933
934 static gfn_t unalias_gfn(struct kvm *kvm, gfn_t gfn)
935 {
936         int i;
937         struct kvm_mem_alias *alias;
938
939         for (i = 0; i < kvm->naliases; ++i) {
940                 alias = &kvm->aliases[i];
941                 if (gfn >= alias->base_gfn
942                     && gfn < alias->base_gfn + alias->npages)
943                         return alias->target_gfn + gfn - alias->base_gfn;
944         }
945         return gfn;
946 }
947
948 static struct kvm_memory_slot *__gfn_to_memslot(struct kvm *kvm, gfn_t gfn)
949 {
950         int i;
951
952         for (i = 0; i < kvm->nmemslots; ++i) {
953                 struct kvm_memory_slot *memslot = &kvm->memslots[i];
954
955                 if (gfn >= memslot->base_gfn
956                     && gfn < memslot->base_gfn + memslot->npages)
957                         return memslot;
958         }
959         return NULL;
960 }
961
962 struct kvm_memory_slot *gfn_to_memslot(struct kvm *kvm, gfn_t gfn)
963 {
964         gfn = unalias_gfn(kvm, gfn);
965         return __gfn_to_memslot(kvm, gfn);
966 }
967
968 struct page *gfn_to_page(struct kvm *kvm, gfn_t gfn)
969 {
970         struct kvm_memory_slot *slot;
971
972         gfn = unalias_gfn(kvm, gfn);
973         slot = __gfn_to_memslot(kvm, gfn);
974         if (!slot)
975                 return NULL;
976         return slot->phys_mem[gfn - slot->base_gfn];
977 }
978 EXPORT_SYMBOL_GPL(gfn_to_page);
979
980 /* WARNING: Does not work on aliased pages. */
981 void mark_page_dirty(struct kvm *kvm, gfn_t gfn)
982 {
983         struct kvm_memory_slot *memslot;
984
985         memslot = __gfn_to_memslot(kvm, gfn);
986         if (memslot && memslot->dirty_bitmap) {
987                 unsigned long rel_gfn = gfn - memslot->base_gfn;
988
989                 /* avoid RMW */
990                 if (!test_bit(rel_gfn, memslot->dirty_bitmap))
991                         set_bit(rel_gfn, memslot->dirty_bitmap);
992         }
993 }
994
995 int emulator_read_std(unsigned long addr,
996                              void *val,
997                              unsigned int bytes,
998                              struct kvm_vcpu *vcpu)
999 {
1000         void *data = val;
1001
1002         while (bytes) {
1003                 gpa_t gpa = vcpu->mmu.gva_to_gpa(vcpu, addr);
1004                 unsigned offset = addr & (PAGE_SIZE-1);
1005                 unsigned tocopy = min(bytes, (unsigned)PAGE_SIZE - offset);
1006                 unsigned long pfn;
1007                 struct page *page;
1008                 void *page_virt;
1009
1010                 if (gpa == UNMAPPED_GVA)
1011                         return X86EMUL_PROPAGATE_FAULT;
1012                 pfn = gpa >> PAGE_SHIFT;
1013                 page = gfn_to_page(vcpu->kvm, pfn);
1014                 if (!page)
1015                         return X86EMUL_UNHANDLEABLE;
1016                 page_virt = kmap_atomic(page, KM_USER0);
1017
1018                 memcpy(data, page_virt + offset, tocopy);
1019
1020                 kunmap_atomic(page_virt, KM_USER0);
1021
1022                 bytes -= tocopy;
1023                 data += tocopy;
1024                 addr += tocopy;
1025         }
1026
1027         return X86EMUL_CONTINUE;
1028 }
1029 EXPORT_SYMBOL_GPL(emulator_read_std);
1030
1031 static int emulator_write_std(unsigned long addr,
1032                               const void *val,
1033                               unsigned int bytes,
1034                               struct kvm_vcpu *vcpu)
1035 {
1036         pr_unimpl(vcpu, "emulator_write_std: addr %lx n %d\n", addr, bytes);
1037         return X86EMUL_UNHANDLEABLE;
1038 }
1039
1040 /*
1041  * Only apic need an MMIO device hook, so shortcut now..
1042  */
1043 static struct kvm_io_device *vcpu_find_pervcpu_dev(struct kvm_vcpu *vcpu,
1044                                                 gpa_t addr)
1045 {
1046         struct kvm_io_device *dev;
1047
1048         if (vcpu->apic) {
1049                 dev = &vcpu->apic->dev;
1050                 if (dev->in_range(dev, addr))
1051                         return dev;
1052         }
1053         return NULL;
1054 }
1055
1056 static struct kvm_io_device *vcpu_find_mmio_dev(struct kvm_vcpu *vcpu,
1057                                                 gpa_t addr)
1058 {
1059         struct kvm_io_device *dev;
1060
1061         dev = vcpu_find_pervcpu_dev(vcpu, addr);
1062         if (dev == NULL)
1063                 dev = kvm_io_bus_find_dev(&vcpu->kvm->mmio_bus, addr);
1064         return dev;
1065 }
1066
1067 static struct kvm_io_device *vcpu_find_pio_dev(struct kvm_vcpu *vcpu,
1068                                                gpa_t addr)
1069 {
1070         return kvm_io_bus_find_dev(&vcpu->kvm->pio_bus, addr);
1071 }
1072
1073 static int emulator_read_emulated(unsigned long addr,
1074                                   void *val,
1075                                   unsigned int bytes,
1076                                   struct kvm_vcpu *vcpu)
1077 {
1078         struct kvm_io_device *mmio_dev;
1079         gpa_t                 gpa;
1080
1081         if (vcpu->mmio_read_completed) {
1082                 memcpy(val, vcpu->mmio_data, bytes);
1083                 vcpu->mmio_read_completed = 0;
1084                 return X86EMUL_CONTINUE;
1085         } else if (emulator_read_std(addr, val, bytes, vcpu)
1086                    == X86EMUL_CONTINUE)
1087                 return X86EMUL_CONTINUE;
1088
1089         gpa = vcpu->mmu.gva_to_gpa(vcpu, addr);
1090         if (gpa == UNMAPPED_GVA)
1091                 return X86EMUL_PROPAGATE_FAULT;
1092
1093         /*
1094          * Is this MMIO handled locally?
1095          */
1096         mmio_dev = vcpu_find_mmio_dev(vcpu, gpa);
1097         if (mmio_dev) {
1098                 kvm_iodevice_read(mmio_dev, gpa, bytes, val);
1099                 return X86EMUL_CONTINUE;
1100         }
1101
1102         vcpu->mmio_needed = 1;
1103         vcpu->mmio_phys_addr = gpa;
1104         vcpu->mmio_size = bytes;
1105         vcpu->mmio_is_write = 0;
1106
1107         return X86EMUL_UNHANDLEABLE;
1108 }
1109
1110 static int emulator_write_phys(struct kvm_vcpu *vcpu, gpa_t gpa,
1111                                const void *val, int bytes)
1112 {
1113         struct page *page;
1114         void *virt;
1115
1116         if (((gpa + bytes - 1) >> PAGE_SHIFT) != (gpa >> PAGE_SHIFT))
1117                 return 0;
1118         page = gfn_to_page(vcpu->kvm, gpa >> PAGE_SHIFT);
1119         if (!page)
1120                 return 0;
1121         mark_page_dirty(vcpu->kvm, gpa >> PAGE_SHIFT);
1122         virt = kmap_atomic(page, KM_USER0);
1123         kvm_mmu_pte_write(vcpu, gpa, val, bytes);
1124         memcpy(virt + offset_in_page(gpa), val, bytes);
1125         kunmap_atomic(virt, KM_USER0);
1126         return 1;
1127 }
1128
1129 static int emulator_write_emulated_onepage(unsigned long addr,
1130                                            const void *val,
1131                                            unsigned int bytes,
1132                                            struct kvm_vcpu *vcpu)
1133 {
1134         struct kvm_io_device *mmio_dev;
1135         gpa_t                 gpa = vcpu->mmu.gva_to_gpa(vcpu, addr);
1136
1137         if (gpa == UNMAPPED_GVA) {
1138                 kvm_x86_ops->inject_page_fault(vcpu, addr, 2);
1139                 return X86EMUL_PROPAGATE_FAULT;
1140         }
1141
1142         if (emulator_write_phys(vcpu, gpa, val, bytes))
1143                 return X86EMUL_CONTINUE;
1144
1145         /*
1146          * Is this MMIO handled locally?
1147          */
1148         mmio_dev = vcpu_find_mmio_dev(vcpu, gpa);
1149         if (mmio_dev) {
1150                 kvm_iodevice_write(mmio_dev, gpa, bytes, val);
1151                 return X86EMUL_CONTINUE;
1152         }
1153
1154         vcpu->mmio_needed = 1;
1155         vcpu->mmio_phys_addr = gpa;
1156         vcpu->mmio_size = bytes;
1157         vcpu->mmio_is_write = 1;
1158         memcpy(vcpu->mmio_data, val, bytes);
1159
1160         return X86EMUL_CONTINUE;
1161 }
1162
1163 int emulator_write_emulated(unsigned long addr,
1164                                    const void *val,
1165                                    unsigned int bytes,
1166                                    struct kvm_vcpu *vcpu)
1167 {
1168         /* Crossing a page boundary? */
1169         if (((addr + bytes - 1) ^ addr) & PAGE_MASK) {
1170                 int rc, now;
1171
1172                 now = -addr & ~PAGE_MASK;
1173                 rc = emulator_write_emulated_onepage(addr, val, now, vcpu);
1174                 if (rc != X86EMUL_CONTINUE)
1175                         return rc;
1176                 addr += now;
1177                 val += now;
1178                 bytes -= now;
1179         }
1180         return emulator_write_emulated_onepage(addr, val, bytes, vcpu);
1181 }
1182 EXPORT_SYMBOL_GPL(emulator_write_emulated);
1183
1184 static int emulator_cmpxchg_emulated(unsigned long addr,
1185                                      const void *old,
1186                                      const void *new,
1187                                      unsigned int bytes,
1188                                      struct kvm_vcpu *vcpu)
1189 {
1190         static int reported;
1191
1192         if (!reported) {
1193                 reported = 1;
1194                 printk(KERN_WARNING "kvm: emulating exchange as write\n");
1195         }
1196         return emulator_write_emulated(addr, new, bytes, vcpu);
1197 }
1198
1199 static unsigned long get_segment_base(struct kvm_vcpu *vcpu, int seg)
1200 {
1201         return kvm_x86_ops->get_segment_base(vcpu, seg);
1202 }
1203
1204 int emulate_invlpg(struct kvm_vcpu *vcpu, gva_t address)
1205 {
1206         return X86EMUL_CONTINUE;
1207 }
1208
1209 int emulate_clts(struct kvm_vcpu *vcpu)
1210 {
1211         vcpu->cr0 &= ~X86_CR0_TS;
1212         kvm_x86_ops->set_cr0(vcpu, vcpu->cr0);
1213         return X86EMUL_CONTINUE;
1214 }
1215
1216 int emulator_get_dr(struct x86_emulate_ctxt* ctxt, int dr, unsigned long *dest)
1217 {
1218         struct kvm_vcpu *vcpu = ctxt->vcpu;
1219
1220         switch (dr) {
1221         case 0 ... 3:
1222                 *dest = kvm_x86_ops->get_dr(vcpu, dr);
1223                 return X86EMUL_CONTINUE;
1224         default:
1225                 pr_unimpl(vcpu, "%s: unexpected dr %u\n", __FUNCTION__, dr);
1226                 return X86EMUL_UNHANDLEABLE;
1227         }
1228 }
1229
1230 int emulator_set_dr(struct x86_emulate_ctxt *ctxt, int dr, unsigned long value)
1231 {
1232         unsigned long mask = (ctxt->mode == X86EMUL_MODE_PROT64) ? ~0ULL : ~0U;
1233         int exception;
1234
1235         kvm_x86_ops->set_dr(ctxt->vcpu, dr, value & mask, &exception);
1236         if (exception) {
1237                 /* FIXME: better handling */
1238                 return X86EMUL_UNHANDLEABLE;
1239         }
1240         return X86EMUL_CONTINUE;
1241 }
1242
1243 void kvm_report_emulation_failure(struct kvm_vcpu *vcpu, const char *context)
1244 {
1245         static int reported;
1246         u8 opcodes[4];
1247         unsigned long rip = vcpu->rip;
1248         unsigned long rip_linear;
1249
1250         rip_linear = rip + get_segment_base(vcpu, VCPU_SREG_CS);
1251
1252         if (reported)
1253                 return;
1254
1255         emulator_read_std(rip_linear, (void *)opcodes, 4, vcpu);
1256
1257         printk(KERN_ERR "emulation failed (%s) rip %lx %02x %02x %02x %02x\n",
1258                context, rip, opcodes[0], opcodes[1], opcodes[2], opcodes[3]);
1259         reported = 1;
1260 }
1261 EXPORT_SYMBOL_GPL(kvm_report_emulation_failure);
1262
1263 struct x86_emulate_ops emulate_ops = {
1264         .read_std            = emulator_read_std,
1265         .write_std           = emulator_write_std,
1266         .read_emulated       = emulator_read_emulated,
1267         .write_emulated      = emulator_write_emulated,
1268         .cmpxchg_emulated    = emulator_cmpxchg_emulated,
1269 };
1270
1271 int emulate_instruction(struct kvm_vcpu *vcpu,
1272                         struct kvm_run *run,
1273                         unsigned long cr2,
1274                         u16 error_code)
1275 {
1276         struct x86_emulate_ctxt emulate_ctxt;
1277         int r;
1278         int cs_db, cs_l;
1279
1280         vcpu->mmio_fault_cr2 = cr2;
1281         kvm_x86_ops->cache_regs(vcpu);
1282
1283         kvm_x86_ops->get_cs_db_l_bits(vcpu, &cs_db, &cs_l);
1284
1285         emulate_ctxt.vcpu = vcpu;
1286         emulate_ctxt.eflags = kvm_x86_ops->get_rflags(vcpu);
1287         emulate_ctxt.cr2 = cr2;
1288         emulate_ctxt.mode = (emulate_ctxt.eflags & X86_EFLAGS_VM)
1289                 ? X86EMUL_MODE_REAL : cs_l
1290                 ? X86EMUL_MODE_PROT64 : cs_db
1291                 ? X86EMUL_MODE_PROT32 : X86EMUL_MODE_PROT16;
1292
1293         if (emulate_ctxt.mode == X86EMUL_MODE_PROT64) {
1294                 emulate_ctxt.cs_base = 0;
1295                 emulate_ctxt.ds_base = 0;
1296                 emulate_ctxt.es_base = 0;
1297                 emulate_ctxt.ss_base = 0;
1298         } else {
1299                 emulate_ctxt.cs_base = get_segment_base(vcpu, VCPU_SREG_CS);
1300                 emulate_ctxt.ds_base = get_segment_base(vcpu, VCPU_SREG_DS);
1301                 emulate_ctxt.es_base = get_segment_base(vcpu, VCPU_SREG_ES);
1302                 emulate_ctxt.ss_base = get_segment_base(vcpu, VCPU_SREG_SS);
1303         }
1304
1305         emulate_ctxt.gs_base = get_segment_base(vcpu, VCPU_SREG_GS);
1306         emulate_ctxt.fs_base = get_segment_base(vcpu, VCPU_SREG_FS);
1307
1308         vcpu->mmio_is_write = 0;
1309         vcpu->pio.string = 0;
1310         r = x86_emulate_memop(&emulate_ctxt, &emulate_ops);
1311         if (vcpu->pio.string)
1312                 return EMULATE_DO_MMIO;
1313
1314         if ((r || vcpu->mmio_is_write) && run) {
1315                 run->exit_reason = KVM_EXIT_MMIO;
1316                 run->mmio.phys_addr = vcpu->mmio_phys_addr;
1317                 memcpy(run->mmio.data, vcpu->mmio_data, 8);
1318                 run->mmio.len = vcpu->mmio_size;
1319                 run->mmio.is_write = vcpu->mmio_is_write;
1320         }
1321
1322         if (r) {
1323                 if (kvm_mmu_unprotect_page_virt(vcpu, cr2))
1324                         return EMULATE_DONE;
1325                 if (!vcpu->mmio_needed) {
1326                         kvm_report_emulation_failure(vcpu, "mmio");
1327                         return EMULATE_FAIL;
1328                 }
1329                 return EMULATE_DO_MMIO;
1330         }
1331
1332         kvm_x86_ops->decache_regs(vcpu);
1333         kvm_x86_ops->set_rflags(vcpu, emulate_ctxt.eflags);
1334
1335         if (vcpu->mmio_is_write) {
1336                 vcpu->mmio_needed = 0;
1337                 return EMULATE_DO_MMIO;
1338         }
1339
1340         return EMULATE_DONE;
1341 }
1342 EXPORT_SYMBOL_GPL(emulate_instruction);
1343
1344 /*
1345  * The vCPU has executed a HLT instruction with in-kernel mode enabled.
1346  */
1347 static void kvm_vcpu_block(struct kvm_vcpu *vcpu)
1348 {
1349         DECLARE_WAITQUEUE(wait, current);
1350
1351         add_wait_queue(&vcpu->wq, &wait);
1352
1353         /*
1354          * We will block until either an interrupt or a signal wakes us up
1355          */
1356         while (!kvm_cpu_has_interrupt(vcpu)
1357                && !signal_pending(current)
1358                && vcpu->mp_state != VCPU_MP_STATE_RUNNABLE
1359                && vcpu->mp_state != VCPU_MP_STATE_SIPI_RECEIVED) {
1360                 set_current_state(TASK_INTERRUPTIBLE);
1361                 vcpu_put(vcpu);
1362                 schedule();
1363                 vcpu_load(vcpu);
1364         }
1365
1366         __set_current_state(TASK_RUNNING);
1367         remove_wait_queue(&vcpu->wq, &wait);
1368 }
1369
1370 int kvm_emulate_halt(struct kvm_vcpu *vcpu)
1371 {
1372         ++vcpu->stat.halt_exits;
1373         if (irqchip_in_kernel(vcpu->kvm)) {
1374                 vcpu->mp_state = VCPU_MP_STATE_HALTED;
1375                 kvm_vcpu_block(vcpu);
1376                 if (vcpu->mp_state != VCPU_MP_STATE_RUNNABLE)
1377                         return -EINTR;
1378                 return 1;
1379         } else {
1380                 vcpu->run->exit_reason = KVM_EXIT_HLT;
1381                 return 0;
1382         }
1383 }
1384 EXPORT_SYMBOL_GPL(kvm_emulate_halt);
1385
1386 int kvm_hypercall(struct kvm_vcpu *vcpu, struct kvm_run *run)
1387 {
1388         unsigned long nr, a0, a1, a2, a3, a4, a5, ret;
1389
1390         kvm_x86_ops->cache_regs(vcpu);
1391         ret = -KVM_EINVAL;
1392 #ifdef CONFIG_X86_64
1393         if (is_long_mode(vcpu)) {
1394                 nr = vcpu->regs[VCPU_REGS_RAX];
1395                 a0 = vcpu->regs[VCPU_REGS_RDI];
1396                 a1 = vcpu->regs[VCPU_REGS_RSI];
1397                 a2 = vcpu->regs[VCPU_REGS_RDX];
1398                 a3 = vcpu->regs[VCPU_REGS_RCX];
1399                 a4 = vcpu->regs[VCPU_REGS_R8];
1400                 a5 = vcpu->regs[VCPU_REGS_R9];
1401         } else
1402 #endif
1403         {
1404                 nr = vcpu->regs[VCPU_REGS_RBX] & -1u;
1405                 a0 = vcpu->regs[VCPU_REGS_RAX] & -1u;
1406                 a1 = vcpu->regs[VCPU_REGS_RCX] & -1u;
1407                 a2 = vcpu->regs[VCPU_REGS_RDX] & -1u;
1408                 a3 = vcpu->regs[VCPU_REGS_RSI] & -1u;
1409                 a4 = vcpu->regs[VCPU_REGS_RDI] & -1u;
1410                 a5 = vcpu->regs[VCPU_REGS_RBP] & -1u;
1411         }
1412         switch (nr) {
1413         default:
1414                 run->hypercall.nr = nr;
1415                 run->hypercall.args[0] = a0;
1416                 run->hypercall.args[1] = a1;
1417                 run->hypercall.args[2] = a2;
1418                 run->hypercall.args[3] = a3;
1419                 run->hypercall.args[4] = a4;
1420                 run->hypercall.args[5] = a5;
1421                 run->hypercall.ret = ret;
1422                 run->hypercall.longmode = is_long_mode(vcpu);
1423                 kvm_x86_ops->decache_regs(vcpu);
1424                 return 0;
1425         }
1426         vcpu->regs[VCPU_REGS_RAX] = ret;
1427         kvm_x86_ops->decache_regs(vcpu);
1428         return 1;
1429 }
1430 EXPORT_SYMBOL_GPL(kvm_hypercall);
1431
1432 static u64 mk_cr_64(u64 curr_cr, u32 new_val)
1433 {
1434         return (curr_cr & ~((1ULL << 32) - 1)) | new_val;
1435 }
1436
1437 void realmode_lgdt(struct kvm_vcpu *vcpu, u16 limit, unsigned long base)
1438 {
1439         struct descriptor_table dt = { limit, base };
1440
1441         kvm_x86_ops->set_gdt(vcpu, &dt);
1442 }
1443
1444 void realmode_lidt(struct kvm_vcpu *vcpu, u16 limit, unsigned long base)
1445 {
1446         struct descriptor_table dt = { limit, base };
1447
1448         kvm_x86_ops->set_idt(vcpu, &dt);
1449 }
1450
1451 void realmode_lmsw(struct kvm_vcpu *vcpu, unsigned long msw,
1452                    unsigned long *rflags)
1453 {
1454         lmsw(vcpu, msw);
1455         *rflags = kvm_x86_ops->get_rflags(vcpu);
1456 }
1457
1458 unsigned long realmode_get_cr(struct kvm_vcpu *vcpu, int cr)
1459 {
1460         kvm_x86_ops->decache_cr4_guest_bits(vcpu);
1461         switch (cr) {
1462         case 0:
1463                 return vcpu->cr0;
1464         case 2:
1465                 return vcpu->cr2;
1466         case 3:
1467                 return vcpu->cr3;
1468         case 4:
1469                 return vcpu->cr4;
1470         default:
1471                 vcpu_printf(vcpu, "%s: unexpected cr %u\n", __FUNCTION__, cr);
1472                 return 0;
1473         }
1474 }
1475
1476 void realmode_set_cr(struct kvm_vcpu *vcpu, int cr, unsigned long val,
1477                      unsigned long *rflags)
1478 {
1479         switch (cr) {
1480         case 0:
1481                 set_cr0(vcpu, mk_cr_64(vcpu->cr0, val));
1482                 *rflags = kvm_x86_ops->get_rflags(vcpu);
1483                 break;
1484         case 2:
1485                 vcpu->cr2 = val;
1486                 break;
1487         case 3:
1488                 set_cr3(vcpu, val);
1489                 break;
1490         case 4:
1491                 set_cr4(vcpu, mk_cr_64(vcpu->cr4, val));
1492                 break;
1493         default:
1494                 vcpu_printf(vcpu, "%s: unexpected cr %u\n", __FUNCTION__, cr);
1495         }
1496 }
1497
1498 /*
1499  * Register the para guest with the host:
1500  */
1501 static int vcpu_register_para(struct kvm_vcpu *vcpu, gpa_t para_state_gpa)
1502 {
1503         struct kvm_vcpu_para_state *para_state;
1504         hpa_t para_state_hpa, hypercall_hpa;
1505         struct page *para_state_page;
1506         unsigned char *hypercall;
1507         gpa_t hypercall_gpa;
1508
1509         printk(KERN_DEBUG "kvm: guest trying to enter paravirtual mode\n");
1510         printk(KERN_DEBUG ".... para_state_gpa: %08Lx\n", para_state_gpa);
1511
1512         /*
1513          * Needs to be page aligned:
1514          */
1515         if (para_state_gpa != PAGE_ALIGN(para_state_gpa))
1516                 goto err_gp;
1517
1518         para_state_hpa = gpa_to_hpa(vcpu, para_state_gpa);
1519         printk(KERN_DEBUG ".... para_state_hpa: %08Lx\n", para_state_hpa);
1520         if (is_error_hpa(para_state_hpa))
1521                 goto err_gp;
1522
1523         mark_page_dirty(vcpu->kvm, para_state_gpa >> PAGE_SHIFT);
1524         para_state_page = pfn_to_page(para_state_hpa >> PAGE_SHIFT);
1525         para_state = kmap(para_state_page);
1526
1527         printk(KERN_DEBUG "....  guest version: %d\n", para_state->guest_version);
1528         printk(KERN_DEBUG "....           size: %d\n", para_state->size);
1529
1530         para_state->host_version = KVM_PARA_API_VERSION;
1531         /*
1532          * We cannot support guests that try to register themselves
1533          * with a newer API version than the host supports:
1534          */
1535         if (para_state->guest_version > KVM_PARA_API_VERSION) {
1536                 para_state->ret = -KVM_EINVAL;
1537                 goto err_kunmap_skip;
1538         }
1539
1540         hypercall_gpa = para_state->hypercall_gpa;
1541         hypercall_hpa = gpa_to_hpa(vcpu, hypercall_gpa);
1542         printk(KERN_DEBUG ".... hypercall_hpa: %08Lx\n", hypercall_hpa);
1543         if (is_error_hpa(hypercall_hpa)) {
1544                 para_state->ret = -KVM_EINVAL;
1545                 goto err_kunmap_skip;
1546         }
1547
1548         printk(KERN_DEBUG "kvm: para guest successfully registered.\n");
1549         vcpu->para_state_page = para_state_page;
1550         vcpu->para_state_gpa = para_state_gpa;
1551         vcpu->hypercall_gpa = hypercall_gpa;
1552
1553         mark_page_dirty(vcpu->kvm, hypercall_gpa >> PAGE_SHIFT);
1554         hypercall = kmap_atomic(pfn_to_page(hypercall_hpa >> PAGE_SHIFT),
1555                                 KM_USER1) + (hypercall_hpa & ~PAGE_MASK);
1556         kvm_x86_ops->patch_hypercall(vcpu, hypercall);
1557         kunmap_atomic(hypercall, KM_USER1);
1558
1559         para_state->ret = 0;
1560 err_kunmap_skip:
1561         kunmap(para_state_page);
1562         return 0;
1563 err_gp:
1564         return 1;
1565 }
1566
1567 int kvm_get_msr_common(struct kvm_vcpu *vcpu, u32 msr, u64 *pdata)
1568 {
1569         u64 data;
1570
1571         switch (msr) {
1572         case 0xc0010010: /* SYSCFG */
1573         case 0xc0010015: /* HWCR */
1574         case MSR_IA32_PLATFORM_ID:
1575         case MSR_IA32_P5_MC_ADDR:
1576         case MSR_IA32_P5_MC_TYPE:
1577         case MSR_IA32_MC0_CTL:
1578         case MSR_IA32_MCG_STATUS:
1579         case MSR_IA32_MCG_CAP:
1580         case MSR_IA32_MC0_MISC:
1581         case MSR_IA32_MC0_MISC+4:
1582         case MSR_IA32_MC0_MISC+8:
1583         case MSR_IA32_MC0_MISC+12:
1584         case MSR_IA32_MC0_MISC+16:
1585         case MSR_IA32_UCODE_REV:
1586         case MSR_IA32_PERF_STATUS:
1587         case MSR_IA32_EBL_CR_POWERON:
1588                 /* MTRR registers */
1589         case 0xfe:
1590         case 0x200 ... 0x2ff:
1591                 data = 0;
1592                 break;
1593         case 0xcd: /* fsb frequency */
1594                 data = 3;
1595                 break;
1596         case MSR_IA32_APICBASE:
1597                 data = kvm_get_apic_base(vcpu);
1598                 break;
1599         case MSR_IA32_MISC_ENABLE:
1600                 data = vcpu->ia32_misc_enable_msr;
1601                 break;
1602 #ifdef CONFIG_X86_64
1603         case MSR_EFER:
1604                 data = vcpu->shadow_efer;
1605                 break;
1606 #endif
1607         default:
1608                 pr_unimpl(vcpu, "unhandled rdmsr: 0x%x\n", msr);
1609                 return 1;
1610         }
1611         *pdata = data;
1612         return 0;
1613 }
1614 EXPORT_SYMBOL_GPL(kvm_get_msr_common);
1615
1616 /*
1617  * Reads an msr value (of 'msr_index') into 'pdata'.
1618  * Returns 0 on success, non-0 otherwise.
1619  * Assumes vcpu_load() was already called.
1620  */
1621 int kvm_get_msr(struct kvm_vcpu *vcpu, u32 msr_index, u64 *pdata)
1622 {
1623         return kvm_x86_ops->get_msr(vcpu, msr_index, pdata);
1624 }
1625
1626 #ifdef CONFIG_X86_64
1627
1628 static void set_efer(struct kvm_vcpu *vcpu, u64 efer)
1629 {
1630         if (efer & EFER_RESERVED_BITS) {
1631                 printk(KERN_DEBUG "set_efer: 0x%llx #GP, reserved bits\n",
1632                        efer);
1633                 inject_gp(vcpu);
1634                 return;
1635         }
1636
1637         if (is_paging(vcpu)
1638             && (vcpu->shadow_efer & EFER_LME) != (efer & EFER_LME)) {
1639                 printk(KERN_DEBUG "set_efer: #GP, change LME while paging\n");
1640                 inject_gp(vcpu);
1641                 return;
1642         }
1643
1644         kvm_x86_ops->set_efer(vcpu, efer);
1645
1646         efer &= ~EFER_LMA;
1647         efer |= vcpu->shadow_efer & EFER_LMA;
1648
1649         vcpu->shadow_efer = efer;
1650 }
1651
1652 #endif
1653
1654 int kvm_set_msr_common(struct kvm_vcpu *vcpu, u32 msr, u64 data)
1655 {
1656         switch (msr) {
1657 #ifdef CONFIG_X86_64
1658         case MSR_EFER:
1659                 set_efer(vcpu, data);
1660                 break;
1661 #endif
1662         case MSR_IA32_MC0_STATUS:
1663                 pr_unimpl(vcpu, "%s: MSR_IA32_MC0_STATUS 0x%llx, nop\n",
1664                        __FUNCTION__, data);
1665                 break;
1666         case MSR_IA32_MCG_STATUS:
1667                 pr_unimpl(vcpu, "%s: MSR_IA32_MCG_STATUS 0x%llx, nop\n",
1668                         __FUNCTION__, data);
1669                 break;
1670         case MSR_IA32_UCODE_REV:
1671         case MSR_IA32_UCODE_WRITE:
1672         case 0x200 ... 0x2ff: /* MTRRs */
1673                 break;
1674         case MSR_IA32_APICBASE:
1675                 kvm_set_apic_base(vcpu, data);
1676                 break;
1677         case MSR_IA32_MISC_ENABLE:
1678                 vcpu->ia32_misc_enable_msr = data;
1679                 break;
1680         /*
1681          * This is the 'probe whether the host is KVM' logic:
1682          */
1683         case MSR_KVM_API_MAGIC:
1684                 return vcpu_register_para(vcpu, data);
1685
1686         default:
1687                 pr_unimpl(vcpu, "unhandled wrmsr: 0x%x\n", msr);
1688                 return 1;
1689         }
1690         return 0;
1691 }
1692 EXPORT_SYMBOL_GPL(kvm_set_msr_common);
1693
1694 /*
1695  * Writes msr value into into the appropriate "register".
1696  * Returns 0 on success, non-0 otherwise.
1697  * Assumes vcpu_load() was already called.
1698  */
1699 int kvm_set_msr(struct kvm_vcpu *vcpu, u32 msr_index, u64 data)
1700 {
1701         return kvm_x86_ops->set_msr(vcpu, msr_index, data);
1702 }
1703
1704 void kvm_resched(struct kvm_vcpu *vcpu)
1705 {
1706         if (!need_resched())
1707                 return;
1708         cond_resched();
1709 }
1710 EXPORT_SYMBOL_GPL(kvm_resched);
1711
1712 void kvm_emulate_cpuid(struct kvm_vcpu *vcpu)
1713 {
1714         int i;
1715         u32 function;
1716         struct kvm_cpuid_entry *e, *best;
1717
1718         kvm_x86_ops->cache_regs(vcpu);
1719         function = vcpu->regs[VCPU_REGS_RAX];
1720         vcpu->regs[VCPU_REGS_RAX] = 0;
1721         vcpu->regs[VCPU_REGS_RBX] = 0;
1722         vcpu->regs[VCPU_REGS_RCX] = 0;
1723         vcpu->regs[VCPU_REGS_RDX] = 0;
1724         best = NULL;
1725         for (i = 0; i < vcpu->cpuid_nent; ++i) {
1726                 e = &vcpu->cpuid_entries[i];
1727                 if (e->function == function) {
1728                         best = e;
1729                         break;
1730                 }
1731                 /*
1732                  * Both basic or both extended?
1733                  */
1734                 if (((e->function ^ function) & 0x80000000) == 0)
1735                         if (!best || e->function > best->function)
1736                                 best = e;
1737         }
1738         if (best) {
1739                 vcpu->regs[VCPU_REGS_RAX] = best->eax;
1740                 vcpu->regs[VCPU_REGS_RBX] = best->ebx;
1741                 vcpu->regs[VCPU_REGS_RCX] = best->ecx;
1742                 vcpu->regs[VCPU_REGS_RDX] = best->edx;
1743         }
1744         kvm_x86_ops->decache_regs(vcpu);
1745         kvm_x86_ops->skip_emulated_instruction(vcpu);
1746 }
1747 EXPORT_SYMBOL_GPL(kvm_emulate_cpuid);
1748
1749 static int pio_copy_data(struct kvm_vcpu *vcpu)
1750 {
1751         void *p = vcpu->pio_data;
1752         void *q;
1753         unsigned bytes;
1754         int nr_pages = vcpu->pio.guest_pages[1] ? 2 : 1;
1755
1756         q = vmap(vcpu->pio.guest_pages, nr_pages, VM_READ|VM_WRITE,
1757                  PAGE_KERNEL);
1758         if (!q) {
1759                 free_pio_guest_pages(vcpu);
1760                 return -ENOMEM;
1761         }
1762         q += vcpu->pio.guest_page_offset;
1763         bytes = vcpu->pio.size * vcpu->pio.cur_count;
1764         if (vcpu->pio.in)
1765                 memcpy(q, p, bytes);
1766         else
1767                 memcpy(p, q, bytes);
1768         q -= vcpu->pio.guest_page_offset;
1769         vunmap(q);
1770         free_pio_guest_pages(vcpu);
1771         return 0;
1772 }
1773
1774 static int complete_pio(struct kvm_vcpu *vcpu)
1775 {
1776         struct kvm_pio_request *io = &vcpu->pio;
1777         long delta;
1778         int r;
1779
1780         kvm_x86_ops->cache_regs(vcpu);
1781
1782         if (!io->string) {
1783                 if (io->in)
1784                         memcpy(&vcpu->regs[VCPU_REGS_RAX], vcpu->pio_data,
1785                                io->size);
1786         } else {
1787                 if (io->in) {
1788                         r = pio_copy_data(vcpu);
1789                         if (r) {
1790                                 kvm_x86_ops->cache_regs(vcpu);
1791                                 return r;
1792                         }
1793                 }
1794
1795                 delta = 1;
1796                 if (io->rep) {
1797                         delta *= io->cur_count;
1798                         /*
1799                          * The size of the register should really depend on
1800                          * current address size.
1801                          */
1802                         vcpu->regs[VCPU_REGS_RCX] -= delta;
1803                 }
1804                 if (io->down)
1805                         delta = -delta;
1806                 delta *= io->size;
1807                 if (io->in)
1808                         vcpu->regs[VCPU_REGS_RDI] += delta;
1809                 else
1810                         vcpu->regs[VCPU_REGS_RSI] += delta;
1811         }
1812
1813         kvm_x86_ops->decache_regs(vcpu);
1814
1815         io->count -= io->cur_count;
1816         io->cur_count = 0;
1817
1818         return 0;
1819 }
1820
1821 static void kernel_pio(struct kvm_io_device *pio_dev,
1822                        struct kvm_vcpu *vcpu,
1823                        void *pd)
1824 {
1825         /* TODO: String I/O for in kernel device */
1826
1827         mutex_lock(&vcpu->kvm->lock);
1828         if (vcpu->pio.in)
1829                 kvm_iodevice_read(pio_dev, vcpu->pio.port,
1830                                   vcpu->pio.size,
1831                                   pd);
1832         else
1833                 kvm_iodevice_write(pio_dev, vcpu->pio.port,
1834                                    vcpu->pio.size,
1835                                    pd);
1836         mutex_unlock(&vcpu->kvm->lock);
1837 }
1838
1839 static void pio_string_write(struct kvm_io_device *pio_dev,
1840                              struct kvm_vcpu *vcpu)
1841 {
1842         struct kvm_pio_request *io = &vcpu->pio;
1843         void *pd = vcpu->pio_data;
1844         int i;
1845
1846         mutex_lock(&vcpu->kvm->lock);
1847         for (i = 0; i < io->cur_count; i++) {
1848                 kvm_iodevice_write(pio_dev, io->port,
1849                                    io->size,
1850                                    pd);
1851                 pd += io->size;
1852         }
1853         mutex_unlock(&vcpu->kvm->lock);
1854 }
1855
1856 int kvm_emulate_pio (struct kvm_vcpu *vcpu, struct kvm_run *run, int in,
1857                   int size, unsigned port)
1858 {
1859         struct kvm_io_device *pio_dev;
1860
1861         vcpu->run->exit_reason = KVM_EXIT_IO;
1862         vcpu->run->io.direction = in ? KVM_EXIT_IO_IN : KVM_EXIT_IO_OUT;
1863         vcpu->run->io.size = vcpu->pio.size = size;
1864         vcpu->run->io.data_offset = KVM_PIO_PAGE_OFFSET * PAGE_SIZE;
1865         vcpu->run->io.count = vcpu->pio.count = vcpu->pio.cur_count = 1;
1866         vcpu->run->io.port = vcpu->pio.port = port;
1867         vcpu->pio.in = in;
1868         vcpu->pio.string = 0;
1869         vcpu->pio.down = 0;
1870         vcpu->pio.guest_page_offset = 0;
1871         vcpu->pio.rep = 0;
1872
1873         kvm_x86_ops->cache_regs(vcpu);
1874         memcpy(vcpu->pio_data, &vcpu->regs[VCPU_REGS_RAX], 4);
1875         kvm_x86_ops->decache_regs(vcpu);
1876
1877         kvm_x86_ops->skip_emulated_instruction(vcpu);
1878
1879         pio_dev = vcpu_find_pio_dev(vcpu, port);
1880         if (pio_dev) {
1881                 kernel_pio(pio_dev, vcpu, vcpu->pio_data);
1882                 complete_pio(vcpu);
1883                 return 1;
1884         }
1885         return 0;
1886 }
1887 EXPORT_SYMBOL_GPL(kvm_emulate_pio);
1888
1889 int kvm_emulate_pio_string(struct kvm_vcpu *vcpu, struct kvm_run *run, int in,
1890                   int size, unsigned long count, int down,
1891                   gva_t address, int rep, unsigned port)
1892 {
1893         unsigned now, in_page;
1894         int i, ret = 0;
1895         int nr_pages = 1;
1896         struct page *page;
1897         struct kvm_io_device *pio_dev;
1898
1899         vcpu->run->exit_reason = KVM_EXIT_IO;
1900         vcpu->run->io.direction = in ? KVM_EXIT_IO_IN : KVM_EXIT_IO_OUT;
1901         vcpu->run->io.size = vcpu->pio.size = size;
1902         vcpu->run->io.data_offset = KVM_PIO_PAGE_OFFSET * PAGE_SIZE;
1903         vcpu->run->io.count = vcpu->pio.count = vcpu->pio.cur_count = count;
1904         vcpu->run->io.port = vcpu->pio.port = port;
1905         vcpu->pio.in = in;
1906         vcpu->pio.string = 1;
1907         vcpu->pio.down = down;
1908         vcpu->pio.guest_page_offset = offset_in_page(address);
1909         vcpu->pio.rep = rep;
1910
1911         if (!count) {
1912                 kvm_x86_ops->skip_emulated_instruction(vcpu);
1913                 return 1;
1914         }
1915
1916         if (!down)
1917                 in_page = PAGE_SIZE - offset_in_page(address);
1918         else
1919                 in_page = offset_in_page(address) + size;
1920         now = min(count, (unsigned long)in_page / size);
1921         if (!now) {
1922                 /*
1923                  * String I/O straddles page boundary.  Pin two guest pages
1924                  * so that we satisfy atomicity constraints.  Do just one
1925                  * transaction to avoid complexity.
1926                  */
1927                 nr_pages = 2;
1928                 now = 1;
1929         }
1930         if (down) {
1931                 /*
1932                  * String I/O in reverse.  Yuck.  Kill the guest, fix later.
1933                  */
1934                 pr_unimpl(vcpu, "guest string pio down\n");
1935                 inject_gp(vcpu);
1936                 return 1;
1937         }
1938         vcpu->run->io.count = now;
1939         vcpu->pio.cur_count = now;
1940
1941         if (vcpu->pio.cur_count == vcpu->pio.count)
1942                 kvm_x86_ops->skip_emulated_instruction(vcpu);
1943
1944         for (i = 0; i < nr_pages; ++i) {
1945                 mutex_lock(&vcpu->kvm->lock);
1946                 page = gva_to_page(vcpu, address + i * PAGE_SIZE);
1947                 if (page)
1948                         get_page(page);
1949                 vcpu->pio.guest_pages[i] = page;
1950                 mutex_unlock(&vcpu->kvm->lock);
1951                 if (!page) {
1952                         inject_gp(vcpu);
1953                         free_pio_guest_pages(vcpu);
1954                         return 1;
1955                 }
1956         }
1957
1958         pio_dev = vcpu_find_pio_dev(vcpu, port);
1959         if (!vcpu->pio.in) {
1960                 /* string PIO write */
1961                 ret = pio_copy_data(vcpu);
1962                 if (ret >= 0 && pio_dev) {
1963                         pio_string_write(pio_dev, vcpu);
1964                         complete_pio(vcpu);
1965                         if (vcpu->pio.count == 0)
1966                                 ret = 1;
1967                 }
1968         } else if (pio_dev)
1969                 pr_unimpl(vcpu, "no string pio read support yet, "
1970                        "port %x size %d count %ld\n",
1971                         port, size, count);
1972
1973         return ret;
1974 }
1975 EXPORT_SYMBOL_GPL(kvm_emulate_pio_string);
1976
1977 /*
1978  * Check if userspace requested an interrupt window, and that the
1979  * interrupt window is open.
1980  *
1981  * No need to exit to userspace if we already have an interrupt queued.
1982  */
1983 static int dm_request_for_irq_injection(struct kvm_vcpu *vcpu,
1984                                           struct kvm_run *kvm_run)
1985 {
1986         return (!vcpu->irq_summary &&
1987                 kvm_run->request_interrupt_window &&
1988                 vcpu->interrupt_window_open &&
1989                 (kvm_x86_ops->get_rflags(vcpu) & X86_EFLAGS_IF));
1990 }
1991
1992 static void post_kvm_run_save(struct kvm_vcpu *vcpu,
1993                               struct kvm_run *kvm_run)
1994 {
1995         kvm_run->if_flag = (kvm_x86_ops->get_rflags(vcpu) & X86_EFLAGS_IF) != 0;
1996         kvm_run->cr8 = get_cr8(vcpu);
1997         kvm_run->apic_base = kvm_get_apic_base(vcpu);
1998         if (irqchip_in_kernel(vcpu->kvm))
1999                 kvm_run->ready_for_interrupt_injection = 1;
2000         else
2001                 kvm_run->ready_for_interrupt_injection =
2002                                         (vcpu->interrupt_window_open &&
2003                                          vcpu->irq_summary == 0);
2004 }
2005
2006 static int __vcpu_run(struct kvm_vcpu *vcpu, struct kvm_run *kvm_run)
2007 {
2008         int r;
2009
2010         if (unlikely(vcpu->mp_state == VCPU_MP_STATE_SIPI_RECEIVED)) {
2011                 printk("vcpu %d received sipi with vector # %x\n",
2012                        vcpu->vcpu_id, vcpu->sipi_vector);
2013                 kvm_lapic_reset(vcpu);
2014                 kvm_x86_ops->vcpu_reset(vcpu);
2015                 vcpu->mp_state = VCPU_MP_STATE_RUNNABLE;
2016         }
2017
2018 preempted:
2019         if (vcpu->guest_debug.enabled)
2020                 kvm_x86_ops->guest_debug_pre(vcpu);
2021
2022 again:
2023         r = kvm_mmu_reload(vcpu);
2024         if (unlikely(r))
2025                 goto out;
2026
2027         preempt_disable();
2028
2029         kvm_x86_ops->prepare_guest_switch(vcpu);
2030         kvm_load_guest_fpu(vcpu);
2031
2032         local_irq_disable();
2033
2034         if (signal_pending(current)) {
2035                 local_irq_enable();
2036                 preempt_enable();
2037                 r = -EINTR;
2038                 kvm_run->exit_reason = KVM_EXIT_INTR;
2039                 ++vcpu->stat.signal_exits;
2040                 goto out;
2041         }
2042
2043         if (irqchip_in_kernel(vcpu->kvm))
2044                 kvm_x86_ops->inject_pending_irq(vcpu);
2045         else if (!vcpu->mmio_read_completed)
2046                 kvm_x86_ops->inject_pending_vectors(vcpu, kvm_run);
2047
2048         vcpu->guest_mode = 1;
2049
2050         if (vcpu->requests)
2051                 if (test_and_clear_bit(KVM_TLB_FLUSH, &vcpu->requests))
2052                         kvm_x86_ops->tlb_flush(vcpu);
2053
2054         kvm_x86_ops->run(vcpu, kvm_run);
2055
2056         vcpu->guest_mode = 0;
2057         local_irq_enable();
2058
2059         ++vcpu->stat.exits;
2060
2061         preempt_enable();
2062
2063         /*
2064          * Profile KVM exit RIPs:
2065          */
2066         if (unlikely(prof_on == KVM_PROFILING)) {
2067                 kvm_x86_ops->cache_regs(vcpu);
2068                 profile_hit(KVM_PROFILING, (void *)vcpu->rip);
2069         }
2070
2071         r = kvm_x86_ops->handle_exit(kvm_run, vcpu);
2072
2073         if (r > 0) {
2074                 if (dm_request_for_irq_injection(vcpu, kvm_run)) {
2075                         r = -EINTR;
2076                         kvm_run->exit_reason = KVM_EXIT_INTR;
2077                         ++vcpu->stat.request_irq_exits;
2078                         goto out;
2079                 }
2080                 if (!need_resched()) {
2081                         ++vcpu->stat.light_exits;
2082                         goto again;
2083                 }
2084         }
2085
2086 out:
2087         if (r > 0) {
2088                 kvm_resched(vcpu);
2089                 goto preempted;
2090         }
2091
2092         post_kvm_run_save(vcpu, kvm_run);
2093
2094         return r;
2095 }
2096
2097
2098 static int kvm_vcpu_ioctl_run(struct kvm_vcpu *vcpu, struct kvm_run *kvm_run)
2099 {
2100         int r;
2101         sigset_t sigsaved;
2102
2103         vcpu_load(vcpu);
2104
2105         if (unlikely(vcpu->mp_state == VCPU_MP_STATE_UNINITIALIZED)) {
2106                 kvm_vcpu_block(vcpu);
2107                 vcpu_put(vcpu);
2108                 return -EAGAIN;
2109         }
2110
2111         if (vcpu->sigset_active)
2112                 sigprocmask(SIG_SETMASK, &vcpu->sigset, &sigsaved);
2113
2114         /* re-sync apic's tpr */
2115         if (!irqchip_in_kernel(vcpu->kvm))
2116                 set_cr8(vcpu, kvm_run->cr8);
2117
2118         if (vcpu->pio.cur_count) {
2119                 r = complete_pio(vcpu);
2120                 if (r)
2121                         goto out;
2122         }
2123
2124         if (vcpu->mmio_needed) {
2125                 memcpy(vcpu->mmio_data, kvm_run->mmio.data, 8);
2126                 vcpu->mmio_read_completed = 1;
2127                 vcpu->mmio_needed = 0;
2128                 r = emulate_instruction(vcpu, kvm_run,
2129                                         vcpu->mmio_fault_cr2, 0);
2130                 if (r == EMULATE_DO_MMIO) {
2131                         /*
2132                          * Read-modify-write.  Back to userspace.
2133                          */
2134                         r = 0;
2135                         goto out;
2136                 }
2137         }
2138
2139         if (kvm_run->exit_reason == KVM_EXIT_HYPERCALL) {
2140                 kvm_x86_ops->cache_regs(vcpu);
2141                 vcpu->regs[VCPU_REGS_RAX] = kvm_run->hypercall.ret;
2142                 kvm_x86_ops->decache_regs(vcpu);
2143         }
2144
2145         r = __vcpu_run(vcpu, kvm_run);
2146
2147 out:
2148         if (vcpu->sigset_active)
2149                 sigprocmask(SIG_SETMASK, &sigsaved, NULL);
2150
2151         vcpu_put(vcpu);
2152         return r;
2153 }
2154
2155 static int kvm_vcpu_ioctl_get_regs(struct kvm_vcpu *vcpu,
2156                                    struct kvm_regs *regs)
2157 {
2158         vcpu_load(vcpu);
2159
2160         kvm_x86_ops->cache_regs(vcpu);
2161
2162         regs->rax = vcpu->regs[VCPU_REGS_RAX];
2163         regs->rbx = vcpu->regs[VCPU_REGS_RBX];
2164         regs->rcx = vcpu->regs[VCPU_REGS_RCX];
2165         regs->rdx = vcpu->regs[VCPU_REGS_RDX];
2166         regs->rsi = vcpu->regs[VCPU_REGS_RSI];
2167         regs->rdi = vcpu->regs[VCPU_REGS_RDI];
2168         regs->rsp = vcpu->regs[VCPU_REGS_RSP];
2169         regs->rbp = vcpu->regs[VCPU_REGS_RBP];
2170 #ifdef CONFIG_X86_64
2171         regs->r8 = vcpu->regs[VCPU_REGS_R8];
2172         regs->r9 = vcpu->regs[VCPU_REGS_R9];
2173         regs->r10 = vcpu->regs[VCPU_REGS_R10];
2174         regs->r11 = vcpu->regs[VCPU_REGS_R11];
2175         regs->r12 = vcpu->regs[VCPU_REGS_R12];
2176         regs->r13 = vcpu->regs[VCPU_REGS_R13];
2177         regs->r14 = vcpu->regs[VCPU_REGS_R14];
2178         regs->r15 = vcpu->regs[VCPU_REGS_R15];
2179 #endif
2180
2181         regs->rip = vcpu->rip;
2182         regs->rflags = kvm_x86_ops->get_rflags(vcpu);
2183
2184         /*
2185          * Don't leak debug flags in case they were set for guest debugging
2186          */
2187         if (vcpu->guest_debug.enabled && vcpu->guest_debug.singlestep)
2188                 regs->rflags &= ~(X86_EFLAGS_TF | X86_EFLAGS_RF);
2189
2190         vcpu_put(vcpu);
2191
2192         return 0;
2193 }
2194
2195 static int kvm_vcpu_ioctl_set_regs(struct kvm_vcpu *vcpu,
2196                                    struct kvm_regs *regs)
2197 {
2198         vcpu_load(vcpu);
2199
2200         vcpu->regs[VCPU_REGS_RAX] = regs->rax;
2201         vcpu->regs[VCPU_REGS_RBX] = regs->rbx;
2202         vcpu->regs[VCPU_REGS_RCX] = regs->rcx;
2203         vcpu->regs[VCPU_REGS_RDX] = regs->rdx;
2204         vcpu->regs[VCPU_REGS_RSI] = regs->rsi;
2205         vcpu->regs[VCPU_REGS_RDI] = regs->rdi;
2206         vcpu->regs[VCPU_REGS_RSP] = regs->rsp;
2207         vcpu->regs[VCPU_REGS_RBP] = regs->rbp;
2208 #ifdef CONFIG_X86_64
2209         vcpu->regs[VCPU_REGS_R8] = regs->r8;
2210         vcpu->regs[VCPU_REGS_R9] = regs->r9;
2211         vcpu->regs[VCPU_REGS_R10] = regs->r10;
2212         vcpu->regs[VCPU_REGS_R11] = regs->r11;
2213         vcpu->regs[VCPU_REGS_R12] = regs->r12;
2214         vcpu->regs[VCPU_REGS_R13] = regs->r13;
2215         vcpu->regs[VCPU_REGS_R14] = regs->r14;
2216         vcpu->regs[VCPU_REGS_R15] = regs->r15;
2217 #endif
2218
2219         vcpu->rip = regs->rip;
2220         kvm_x86_ops->set_rflags(vcpu, regs->rflags);
2221
2222         kvm_x86_ops->decache_regs(vcpu);
2223
2224         vcpu_put(vcpu);
2225
2226         return 0;
2227 }
2228
2229 static void get_segment(struct kvm_vcpu *vcpu,
2230                         struct kvm_segment *var, int seg)
2231 {
2232         return kvm_x86_ops->get_segment(vcpu, var, seg);
2233 }
2234
2235 static int kvm_vcpu_ioctl_get_sregs(struct kvm_vcpu *vcpu,
2236                                     struct kvm_sregs *sregs)
2237 {
2238         struct descriptor_table dt;
2239         int pending_vec;
2240
2241         vcpu_load(vcpu);
2242
2243         get_segment(vcpu, &sregs->cs, VCPU_SREG_CS);
2244         get_segment(vcpu, &sregs->ds, VCPU_SREG_DS);
2245         get_segment(vcpu, &sregs->es, VCPU_SREG_ES);
2246         get_segment(vcpu, &sregs->fs, VCPU_SREG_FS);
2247         get_segment(vcpu, &sregs->gs, VCPU_SREG_GS);
2248         get_segment(vcpu, &sregs->ss, VCPU_SREG_SS);
2249
2250         get_segment(vcpu, &sregs->tr, VCPU_SREG_TR);
2251         get_segment(vcpu, &sregs->ldt, VCPU_SREG_LDTR);
2252
2253         kvm_x86_ops->get_idt(vcpu, &dt);
2254         sregs->idt.limit = dt.limit;
2255         sregs->idt.base = dt.base;
2256         kvm_x86_ops->get_gdt(vcpu, &dt);
2257         sregs->gdt.limit = dt.limit;
2258         sregs->gdt.base = dt.base;
2259
2260         kvm_x86_ops->decache_cr4_guest_bits(vcpu);
2261         sregs->cr0 = vcpu->cr0;
2262         sregs->cr2 = vcpu->cr2;
2263         sregs->cr3 = vcpu->cr3;
2264         sregs->cr4 = vcpu->cr4;
2265         sregs->cr8 = get_cr8(vcpu);
2266         sregs->efer = vcpu->shadow_efer;
2267         sregs->apic_base = kvm_get_apic_base(vcpu);
2268
2269         if (irqchip_in_kernel(vcpu->kvm)) {
2270                 memset(sregs->interrupt_bitmap, 0,
2271                        sizeof sregs->interrupt_bitmap);
2272                 pending_vec = kvm_x86_ops->get_irq(vcpu);
2273                 if (pending_vec >= 0)
2274                         set_bit(pending_vec, (unsigned long *)sregs->interrupt_bitmap);
2275         } else
2276                 memcpy(sregs->interrupt_bitmap, vcpu->irq_pending,
2277                        sizeof sregs->interrupt_bitmap);
2278
2279         vcpu_put(vcpu);
2280
2281         return 0;
2282 }
2283
2284 static void set_segment(struct kvm_vcpu *vcpu,
2285                         struct kvm_segment *var, int seg)
2286 {
2287         return kvm_x86_ops->set_segment(vcpu, var, seg);
2288 }
2289
2290 static int kvm_vcpu_ioctl_set_sregs(struct kvm_vcpu *vcpu,
2291                                     struct kvm_sregs *sregs)
2292 {
2293         int mmu_reset_needed = 0;
2294         int i, pending_vec, max_bits;
2295         struct descriptor_table dt;
2296
2297         vcpu_load(vcpu);
2298
2299         dt.limit = sregs->idt.limit;
2300         dt.base = sregs->idt.base;
2301         kvm_x86_ops->set_idt(vcpu, &dt);
2302         dt.limit = sregs->gdt.limit;
2303         dt.base = sregs->gdt.base;
2304         kvm_x86_ops->set_gdt(vcpu, &dt);
2305
2306         vcpu->cr2 = sregs->cr2;
2307         mmu_reset_needed |= vcpu->cr3 != sregs->cr3;
2308         vcpu->cr3 = sregs->cr3;
2309
2310         set_cr8(vcpu, sregs->cr8);
2311
2312         mmu_reset_needed |= vcpu->shadow_efer != sregs->efer;
2313 #ifdef CONFIG_X86_64
2314         kvm_x86_ops->set_efer(vcpu, sregs->efer);
2315 #endif
2316         kvm_set_apic_base(vcpu, sregs->apic_base);
2317
2318         kvm_x86_ops->decache_cr4_guest_bits(vcpu);
2319
2320         mmu_reset_needed |= vcpu->cr0 != sregs->cr0;
2321         vcpu->cr0 = sregs->cr0;
2322         kvm_x86_ops->set_cr0(vcpu, sregs->cr0);
2323
2324         mmu_reset_needed |= vcpu->cr4 != sregs->cr4;
2325         kvm_x86_ops->set_cr4(vcpu, sregs->cr4);
2326         if (!is_long_mode(vcpu) && is_pae(vcpu))
2327                 load_pdptrs(vcpu, vcpu->cr3);
2328
2329         if (mmu_reset_needed)
2330                 kvm_mmu_reset_context(vcpu);
2331
2332         if (!irqchip_in_kernel(vcpu->kvm)) {
2333                 memcpy(vcpu->irq_pending, sregs->interrupt_bitmap,
2334                        sizeof vcpu->irq_pending);
2335                 vcpu->irq_summary = 0;
2336                 for (i = 0; i < ARRAY_SIZE(vcpu->irq_pending); ++i)
2337                         if (vcpu->irq_pending[i])
2338                                 __set_bit(i, &vcpu->irq_summary);
2339         } else {
2340                 max_bits = (sizeof sregs->interrupt_bitmap) << 3;
2341                 pending_vec = find_first_bit(
2342                         (const unsigned long *)sregs->interrupt_bitmap,
2343                         max_bits);
2344                 /* Only pending external irq is handled here */
2345                 if (pending_vec < max_bits) {
2346                         kvm_x86_ops->set_irq(vcpu, pending_vec);
2347                         printk("Set back pending irq %d\n", pending_vec);
2348                 }
2349         }
2350
2351         set_segment(vcpu, &sregs->cs, VCPU_SREG_CS);
2352         set_segment(vcpu, &sregs->ds, VCPU_SREG_DS);
2353         set_segment(vcpu, &sregs->es, VCPU_SREG_ES);
2354         set_segment(vcpu, &sregs->fs, VCPU_SREG_FS);
2355         set_segment(vcpu, &sregs->gs, VCPU_SREG_GS);
2356         set_segment(vcpu, &sregs->ss, VCPU_SREG_SS);
2357
2358         set_segment(vcpu, &sregs->tr, VCPU_SREG_TR);
2359         set_segment(vcpu, &sregs->ldt, VCPU_SREG_LDTR);
2360
2361         vcpu_put(vcpu);
2362
2363         return 0;
2364 }
2365
2366 void kvm_get_cs_db_l_bits(struct kvm_vcpu *vcpu, int *db, int *l)
2367 {
2368         struct kvm_segment cs;
2369
2370         get_segment(vcpu, &cs, VCPU_SREG_CS);
2371         *db = cs.db;
2372         *l = cs.l;
2373 }
2374 EXPORT_SYMBOL_GPL(kvm_get_cs_db_l_bits);
2375
2376 /*
2377  * List of msr numbers which we expose to userspace through KVM_GET_MSRS
2378  * and KVM_SET_MSRS, and KVM_GET_MSR_INDEX_LIST.
2379  *
2380  * This list is modified at module load time to reflect the
2381  * capabilities of the host cpu.
2382  */
2383 static u32 msrs_to_save[] = {
2384         MSR_IA32_SYSENTER_CS, MSR_IA32_SYSENTER_ESP, MSR_IA32_SYSENTER_EIP,
2385         MSR_K6_STAR,
2386 #ifdef CONFIG_X86_64
2387         MSR_CSTAR, MSR_KERNEL_GS_BASE, MSR_SYSCALL_MASK, MSR_LSTAR,
2388 #endif
2389         MSR_IA32_TIME_STAMP_COUNTER,
2390 };
2391
2392 static unsigned num_msrs_to_save;
2393
2394 static u32 emulated_msrs[] = {
2395         MSR_IA32_MISC_ENABLE,
2396 };
2397
2398 static __init void kvm_init_msr_list(void)
2399 {
2400         u32 dummy[2];
2401         unsigned i, j;
2402
2403         for (i = j = 0; i < ARRAY_SIZE(msrs_to_save); i++) {
2404                 if (rdmsr_safe(msrs_to_save[i], &dummy[0], &dummy[1]) < 0)
2405                         continue;
2406                 if (j < i)
2407                         msrs_to_save[j] = msrs_to_save[i];
2408                 j++;
2409         }
2410         num_msrs_to_save = j;
2411 }
2412
2413 /*
2414  * Adapt set_msr() to msr_io()'s calling convention
2415  */
2416 static int do_set_msr(struct kvm_vcpu *vcpu, unsigned index, u64 *data)
2417 {
2418         return kvm_set_msr(vcpu, index, *data);
2419 }
2420
2421 /*
2422  * Read or write a bunch of msrs. All parameters are kernel addresses.
2423  *
2424  * @return number of msrs set successfully.
2425  */
2426 static int __msr_io(struct kvm_vcpu *vcpu, struct kvm_msrs *msrs,
2427                     struct kvm_msr_entry *entries,
2428                     int (*do_msr)(struct kvm_vcpu *vcpu,
2429                                   unsigned index, u64 *data))
2430 {
2431         int i;
2432
2433         vcpu_load(vcpu);
2434
2435         for (i = 0; i < msrs->nmsrs; ++i)
2436                 if (do_msr(vcpu, entries[i].index, &entries[i].data))
2437                         break;
2438
2439         vcpu_put(vcpu);
2440
2441         return i;
2442 }
2443
2444 /*
2445  * Read or write a bunch of msrs. Parameters are user addresses.
2446  *
2447  * @return number of msrs set successfully.
2448  */
2449 static int msr_io(struct kvm_vcpu *vcpu, struct kvm_msrs __user *user_msrs,
2450                   int (*do_msr)(struct kvm_vcpu *vcpu,
2451                                 unsigned index, u64 *data),
2452                   int writeback)
2453 {
2454         struct kvm_msrs msrs;
2455         struct kvm_msr_entry *entries;
2456         int r, n;
2457         unsigned size;
2458
2459         r = -EFAULT;
2460         if (copy_from_user(&msrs, user_msrs, sizeof msrs))
2461                 goto out;
2462
2463         r = -E2BIG;
2464         if (msrs.nmsrs >= MAX_IO_MSRS)
2465                 goto out;
2466
2467         r = -ENOMEM;
2468         size = sizeof(struct kvm_msr_entry) * msrs.nmsrs;
2469         entries = vmalloc(size);
2470         if (!entries)
2471                 goto out;
2472
2473         r = -EFAULT;
2474         if (copy_from_user(entries, user_msrs->entries, size))
2475                 goto out_free;
2476
2477         r = n = __msr_io(vcpu, &msrs, entries, do_msr);
2478         if (r < 0)
2479                 goto out_free;
2480
2481         r = -EFAULT;
2482         if (writeback && copy_to_user(user_msrs->entries, entries, size))
2483                 goto out_free;
2484
2485         r = n;
2486
2487 out_free:
2488         vfree(entries);
2489 out:
2490         return r;
2491 }
2492
2493 /*
2494  * Translate a guest virtual address to a guest physical address.
2495  */
2496 static int kvm_vcpu_ioctl_translate(struct kvm_vcpu *vcpu,
2497                                     struct kvm_translation *tr)
2498 {
2499         unsigned long vaddr = tr->linear_address;
2500         gpa_t gpa;
2501
2502         vcpu_load(vcpu);
2503         mutex_lock(&vcpu->kvm->lock);
2504         gpa = vcpu->mmu.gva_to_gpa(vcpu, vaddr);
2505         tr->physical_address = gpa;
2506         tr->valid = gpa != UNMAPPED_GVA;
2507         tr->writeable = 1;
2508         tr->usermode = 0;
2509         mutex_unlock(&vcpu->kvm->lock);
2510         vcpu_put(vcpu);
2511
2512         return 0;
2513 }
2514
2515 static int kvm_vcpu_ioctl_interrupt(struct kvm_vcpu *vcpu,
2516                                     struct kvm_interrupt *irq)
2517 {
2518         if (irq->irq < 0 || irq->irq >= 256)
2519                 return -EINVAL;
2520         if (irqchip_in_kernel(vcpu->kvm))
2521                 return -ENXIO;
2522         vcpu_load(vcpu);
2523
2524         set_bit(irq->irq, vcpu->irq_pending);
2525         set_bit(irq->irq / BITS_PER_LONG, &vcpu->irq_summary);
2526
2527         vcpu_put(vcpu);
2528
2529         return 0;
2530 }
2531
2532 static int kvm_vcpu_ioctl_debug_guest(struct kvm_vcpu *vcpu,
2533                                       struct kvm_debug_guest *dbg)
2534 {
2535         int r;
2536
2537         vcpu_load(vcpu);
2538
2539         r = kvm_x86_ops->set_guest_debug(vcpu, dbg);
2540
2541         vcpu_put(vcpu);
2542
2543         return r;
2544 }
2545
2546 static struct page *kvm_vcpu_nopage(struct vm_area_struct *vma,
2547                                     unsigned long address,
2548                                     int *type)
2549 {
2550         struct kvm_vcpu *vcpu = vma->vm_file->private_data;
2551         unsigned long pgoff;
2552         struct page *page;
2553
2554         pgoff = ((address - vma->vm_start) >> PAGE_SHIFT) + vma->vm_pgoff;
2555         if (pgoff == 0)
2556                 page = virt_to_page(vcpu->run);
2557         else if (pgoff == KVM_PIO_PAGE_OFFSET)
2558                 page = virt_to_page(vcpu->pio_data);
2559         else
2560                 return NOPAGE_SIGBUS;
2561         get_page(page);
2562         if (type != NULL)
2563                 *type = VM_FAULT_MINOR;
2564
2565         return page;
2566 }
2567
2568 static struct vm_operations_struct kvm_vcpu_vm_ops = {
2569         .nopage = kvm_vcpu_nopage,
2570 };
2571
2572 static int kvm_vcpu_mmap(struct file *file, struct vm_area_struct *vma)
2573 {
2574         vma->vm_ops = &kvm_vcpu_vm_ops;
2575         return 0;
2576 }
2577
2578 static int kvm_vcpu_release(struct inode *inode, struct file *filp)
2579 {
2580         struct kvm_vcpu *vcpu = filp->private_data;
2581
2582         fput(vcpu->kvm->filp);
2583         return 0;
2584 }
2585
2586 static struct file_operations kvm_vcpu_fops = {
2587         .release        = kvm_vcpu_release,
2588         .unlocked_ioctl = kvm_vcpu_ioctl,
2589         .compat_ioctl   = kvm_vcpu_ioctl,
2590         .mmap           = kvm_vcpu_mmap,
2591 };
2592
2593 /*
2594  * Allocates an inode for the vcpu.
2595  */
2596 static int create_vcpu_fd(struct kvm_vcpu *vcpu)
2597 {
2598         int fd, r;
2599         struct inode *inode;
2600         struct file *file;
2601
2602         r = anon_inode_getfd(&fd, &inode, &file,
2603                              "kvm-vcpu", &kvm_vcpu_fops, vcpu);
2604         if (r)
2605                 return r;
2606         atomic_inc(&vcpu->kvm->filp->f_count);
2607         return fd;
2608 }
2609
2610 /*
2611  * Creates some virtual cpus.  Good luck creating more than one.
2612  */
2613 static int kvm_vm_ioctl_create_vcpu(struct kvm *kvm, int n)
2614 {
2615         int r;
2616         struct kvm_vcpu *vcpu;
2617
2618         if (!valid_vcpu(n))
2619                 return -EINVAL;
2620
2621         vcpu = kvm_x86_ops->vcpu_create(kvm, n);
2622         if (IS_ERR(vcpu))
2623                 return PTR_ERR(vcpu);
2624
2625         preempt_notifier_init(&vcpu->preempt_notifier, &kvm_preempt_ops);
2626
2627         /* We do fxsave: this must be aligned. */
2628         BUG_ON((unsigned long)&vcpu->host_fx_image & 0xF);
2629
2630         vcpu_load(vcpu);
2631         r = kvm_mmu_setup(vcpu);
2632         vcpu_put(vcpu);
2633         if (r < 0)
2634                 goto free_vcpu;
2635
2636         mutex_lock(&kvm->lock);
2637         if (kvm->vcpus[n]) {
2638                 r = -EEXIST;
2639                 mutex_unlock(&kvm->lock);
2640                 goto mmu_unload;
2641         }
2642         kvm->vcpus[n] = vcpu;
2643         mutex_unlock(&kvm->lock);
2644
2645         /* Now it's all set up, let userspace reach it */
2646         r = create_vcpu_fd(vcpu);
2647         if (r < 0)
2648                 goto unlink;
2649         return r;
2650
2651 unlink:
2652         mutex_lock(&kvm->lock);
2653         kvm->vcpus[n] = NULL;
2654         mutex_unlock(&kvm->lock);
2655
2656 mmu_unload:
2657         vcpu_load(vcpu);
2658         kvm_mmu_unload(vcpu);
2659         vcpu_put(vcpu);
2660
2661 free_vcpu:
2662         kvm_x86_ops->vcpu_free(vcpu);
2663         return r;
2664 }
2665
2666 static void cpuid_fix_nx_cap(struct kvm_vcpu *vcpu)
2667 {
2668         u64 efer;
2669         int i;
2670         struct kvm_cpuid_entry *e, *entry;
2671
2672         rdmsrl(MSR_EFER, efer);
2673         entry = NULL;
2674         for (i = 0; i < vcpu->cpuid_nent; ++i) {
2675                 e = &vcpu->cpuid_entries[i];
2676                 if (e->function == 0x80000001) {
2677                         entry = e;
2678                         break;
2679                 }
2680         }
2681         if (entry && (entry->edx & (1 << 20)) && !(efer & EFER_NX)) {
2682                 entry->edx &= ~(1 << 20);
2683                 printk(KERN_INFO "kvm: guest NX capability removed\n");
2684         }
2685 }
2686
2687 static int kvm_vcpu_ioctl_set_cpuid(struct kvm_vcpu *vcpu,
2688                                     struct kvm_cpuid *cpuid,
2689                                     struct kvm_cpuid_entry __user *entries)
2690 {
2691         int r;
2692
2693         r = -E2BIG;
2694         if (cpuid->nent > KVM_MAX_CPUID_ENTRIES)
2695                 goto out;
2696         r = -EFAULT;
2697         if (copy_from_user(&vcpu->cpuid_entries, entries,
2698                            cpuid->nent * sizeof(struct kvm_cpuid_entry)))
2699                 goto out;
2700         vcpu->cpuid_nent = cpuid->nent;
2701         cpuid_fix_nx_cap(vcpu);
2702         return 0;
2703
2704 out:
2705         return r;
2706 }
2707
2708 static int kvm_vcpu_ioctl_set_sigmask(struct kvm_vcpu *vcpu, sigset_t *sigset)
2709 {
2710         if (sigset) {
2711                 sigdelsetmask(sigset, sigmask(SIGKILL)|sigmask(SIGSTOP));
2712                 vcpu->sigset_active = 1;
2713                 vcpu->sigset = *sigset;
2714         } else
2715                 vcpu->sigset_active = 0;
2716         return 0;
2717 }
2718
2719 /*
2720  * fxsave fpu state.  Taken from x86_64/processor.h.  To be killed when
2721  * we have asm/x86/processor.h
2722  */
2723 struct fxsave {
2724         u16     cwd;
2725         u16     swd;
2726         u16     twd;
2727         u16     fop;
2728         u64     rip;
2729         u64     rdp;
2730         u32     mxcsr;
2731         u32     mxcsr_mask;
2732         u32     st_space[32];   /* 8*16 bytes for each FP-reg = 128 bytes */
2733 #ifdef CONFIG_X86_64
2734         u32     xmm_space[64];  /* 16*16 bytes for each XMM-reg = 256 bytes */
2735 #else
2736         u32     xmm_space[32];  /* 8*16 bytes for each XMM-reg = 128 bytes */
2737 #endif
2738 };
2739
2740 static int kvm_vcpu_ioctl_get_fpu(struct kvm_vcpu *vcpu, struct kvm_fpu *fpu)
2741 {
2742         struct fxsave *fxsave = (struct fxsave *)&vcpu->guest_fx_image;
2743
2744         vcpu_load(vcpu);
2745
2746         memcpy(fpu->fpr, fxsave->st_space, 128);
2747         fpu->fcw = fxsave->cwd;
2748         fpu->fsw = fxsave->swd;
2749         fpu->ftwx = fxsave->twd;
2750         fpu->last_opcode = fxsave->fop;
2751         fpu->last_ip = fxsave->rip;
2752         fpu->last_dp = fxsave->rdp;
2753         memcpy(fpu->xmm, fxsave->xmm_space, sizeof fxsave->xmm_space);
2754
2755         vcpu_put(vcpu);
2756
2757         return 0;
2758 }
2759
2760 static int kvm_vcpu_ioctl_set_fpu(struct kvm_vcpu *vcpu, struct kvm_fpu *fpu)
2761 {
2762         struct fxsave *fxsave = (struct fxsave *)&vcpu->guest_fx_image;
2763
2764         vcpu_load(vcpu);
2765
2766         memcpy(fxsave->st_space, fpu->fpr, 128);
2767         fxsave->cwd = fpu->fcw;
2768         fxsave->swd = fpu->fsw;
2769         fxsave->twd = fpu->ftwx;
2770         fxsave->fop = fpu->last_opcode;
2771         fxsave->rip = fpu->last_ip;
2772         fxsave->rdp = fpu->last_dp;
2773         memcpy(fxsave->xmm_space, fpu->xmm, sizeof fxsave->xmm_space);
2774
2775         vcpu_put(vcpu);
2776
2777         return 0;
2778 }
2779
2780 static int kvm_vcpu_ioctl_get_lapic(struct kvm_vcpu *vcpu,
2781                                     struct kvm_lapic_state *s)
2782 {
2783         vcpu_load(vcpu);
2784         memcpy(s->regs, vcpu->apic->regs, sizeof *s);
2785         vcpu_put(vcpu);
2786
2787         return 0;
2788 }
2789
2790 static int kvm_vcpu_ioctl_set_lapic(struct kvm_vcpu *vcpu,
2791                                     struct kvm_lapic_state *s)
2792 {
2793         vcpu_load(vcpu);
2794         memcpy(vcpu->apic->regs, s->regs, sizeof *s);
2795         kvm_apic_post_state_restore(vcpu);
2796         vcpu_put(vcpu);
2797
2798         return 0;
2799 }
2800
2801 static long kvm_vcpu_ioctl(struct file *filp,
2802                            unsigned int ioctl, unsigned long arg)
2803 {
2804         struct kvm_vcpu *vcpu = filp->private_data;
2805         void __user *argp = (void __user *)arg;
2806         int r = -EINVAL;
2807
2808         switch (ioctl) {
2809         case KVM_RUN:
2810                 r = -EINVAL;
2811                 if (arg)
2812                         goto out;
2813                 r = kvm_vcpu_ioctl_run(vcpu, vcpu->run);
2814                 break;
2815         case KVM_GET_REGS: {
2816                 struct kvm_regs kvm_regs;
2817
2818                 memset(&kvm_regs, 0, sizeof kvm_regs);
2819                 r = kvm_vcpu_ioctl_get_regs(vcpu, &kvm_regs);
2820                 if (r)
2821                         goto out;
2822                 r = -EFAULT;
2823                 if (copy_to_user(argp, &kvm_regs, sizeof kvm_regs))
2824                         goto out;
2825                 r = 0;
2826                 break;
2827         }
2828         case KVM_SET_REGS: {
2829                 struct kvm_regs kvm_regs;
2830
2831                 r = -EFAULT;
2832                 if (copy_from_user(&kvm_regs, argp, sizeof kvm_regs))
2833                         goto out;
2834                 r = kvm_vcpu_ioctl_set_regs(vcpu, &kvm_regs);
2835                 if (r)
2836                         goto out;
2837                 r = 0;
2838                 break;
2839         }
2840         case KVM_GET_SREGS: {
2841                 struct kvm_sregs kvm_sregs;
2842
2843                 memset(&kvm_sregs, 0, sizeof kvm_sregs);
2844                 r = kvm_vcpu_ioctl_get_sregs(vcpu, &kvm_sregs);
2845                 if (r)
2846                         goto out;
2847                 r = -EFAULT;
2848                 if (copy_to_user(argp, &kvm_sregs, sizeof kvm_sregs))
2849                         goto out;
2850                 r = 0;
2851                 break;
2852         }
2853         case KVM_SET_SREGS: {
2854                 struct kvm_sregs kvm_sregs;
2855
2856                 r = -EFAULT;
2857                 if (copy_from_user(&kvm_sregs, argp, sizeof kvm_sregs))
2858                         goto out;
2859                 r = kvm_vcpu_ioctl_set_sregs(vcpu, &kvm_sregs);
2860                 if (r)
2861                         goto out;
2862                 r = 0;
2863                 break;
2864         }
2865         case KVM_TRANSLATE: {
2866                 struct kvm_translation tr;
2867
2868                 r = -EFAULT;
2869                 if (copy_from_user(&tr, argp, sizeof tr))
2870                         goto out;
2871                 r = kvm_vcpu_ioctl_translate(vcpu, &tr);
2872                 if (r)
2873                         goto out;
2874                 r = -EFAULT;
2875                 if (copy_to_user(argp, &tr, sizeof tr))
2876                         goto out;
2877                 r = 0;
2878                 break;
2879         }
2880         case KVM_INTERRUPT: {
2881                 struct kvm_interrupt irq;
2882
2883                 r = -EFAULT;
2884                 if (copy_from_user(&irq, argp, sizeof irq))
2885                         goto out;
2886                 r = kvm_vcpu_ioctl_interrupt(vcpu, &irq);
2887                 if (r)
2888                         goto out;
2889                 r = 0;
2890                 break;
2891         }
2892         case KVM_DEBUG_GUEST: {
2893                 struct kvm_debug_guest dbg;
2894
2895                 r = -EFAULT;
2896                 if (copy_from_user(&dbg, argp, sizeof dbg))
2897                         goto out;
2898                 r = kvm_vcpu_ioctl_debug_guest(vcpu, &dbg);
2899                 if (r)
2900                         goto out;
2901                 r = 0;
2902                 break;
2903         }
2904         case KVM_GET_MSRS:
2905                 r = msr_io(vcpu, argp, kvm_get_msr, 1);
2906                 break;
2907         case KVM_SET_MSRS:
2908                 r = msr_io(vcpu, argp, do_set_msr, 0);
2909                 break;
2910         case KVM_SET_CPUID: {
2911                 struct kvm_cpuid __user *cpuid_arg = argp;
2912                 struct kvm_cpuid cpuid;
2913
2914                 r = -EFAULT;
2915                 if (copy_from_user(&cpuid, cpuid_arg, sizeof cpuid))
2916                         goto out;
2917                 r = kvm_vcpu_ioctl_set_cpuid(vcpu, &cpuid, cpuid_arg->entries);
2918                 if (r)
2919                         goto out;
2920                 break;
2921         }
2922         case KVM_SET_SIGNAL_MASK: {
2923                 struct kvm_signal_mask __user *sigmask_arg = argp;
2924                 struct kvm_signal_mask kvm_sigmask;
2925                 sigset_t sigset, *p;
2926
2927                 p = NULL;
2928                 if (argp) {
2929                         r = -EFAULT;
2930                         if (copy_from_user(&kvm_sigmask, argp,
2931                                            sizeof kvm_sigmask))
2932                                 goto out;
2933                         r = -EINVAL;
2934                         if (kvm_sigmask.len != sizeof sigset)
2935                                 goto out;
2936                         r = -EFAULT;
2937                         if (copy_from_user(&sigset, sigmask_arg->sigset,
2938                                            sizeof sigset))
2939                                 goto out;
2940                         p = &sigset;
2941                 }
2942                 r = kvm_vcpu_ioctl_set_sigmask(vcpu, &sigset);
2943                 break;
2944         }
2945         case KVM_GET_FPU: {
2946                 struct kvm_fpu fpu;
2947
2948                 memset(&fpu, 0, sizeof fpu);
2949                 r = kvm_vcpu_ioctl_get_fpu(vcpu, &fpu);
2950                 if (r)
2951                         goto out;
2952                 r = -EFAULT;
2953                 if (copy_to_user(argp, &fpu, sizeof fpu))
2954                         goto out;
2955                 r = 0;
2956                 break;
2957         }
2958         case KVM_SET_FPU: {
2959                 struct kvm_fpu fpu;
2960
2961                 r = -EFAULT;
2962                 if (copy_from_user(&fpu, argp, sizeof fpu))
2963                         goto out;
2964                 r = kvm_vcpu_ioctl_set_fpu(vcpu, &fpu);
2965                 if (r)
2966                         goto out;
2967                 r = 0;
2968                 break;
2969         }
2970         case KVM_GET_LAPIC: {
2971                 struct kvm_lapic_state lapic;
2972
2973                 memset(&lapic, 0, sizeof lapic);
2974                 r = kvm_vcpu_ioctl_get_lapic(vcpu, &lapic);
2975                 if (r)
2976                         goto out;
2977                 r = -EFAULT;
2978                 if (copy_to_user(argp, &lapic, sizeof lapic))
2979                         goto out;
2980                 r = 0;
2981                 break;
2982         }
2983         case KVM_SET_LAPIC: {
2984                 struct kvm_lapic_state lapic;
2985
2986                 r = -EFAULT;
2987                 if (copy_from_user(&lapic, argp, sizeof lapic))
2988                         goto out;
2989                 r = kvm_vcpu_ioctl_set_lapic(vcpu, &lapic);;
2990                 if (r)
2991                         goto out;
2992                 r = 0;
2993                 break;
2994         }
2995         default:
2996                 ;
2997         }
2998 out:
2999         return r;
3000 }
3001
3002 static long kvm_vm_ioctl(struct file *filp,
3003                            unsigned int ioctl, unsigned long arg)
3004 {
3005         struct kvm *kvm = filp->private_data;
3006         void __user *argp = (void __user *)arg;
3007         int r = -EINVAL;
3008
3009         switch (ioctl) {
3010         case KVM_CREATE_VCPU:
3011                 r = kvm_vm_ioctl_create_vcpu(kvm, arg);
3012                 if (r < 0)
3013                         goto out;
3014                 break;
3015         case KVM_SET_MEMORY_REGION: {
3016                 struct kvm_memory_region kvm_mem;
3017
3018                 r = -EFAULT;
3019                 if (copy_from_user(&kvm_mem, argp, sizeof kvm_mem))
3020                         goto out;
3021                 r = kvm_vm_ioctl_set_memory_region(kvm, &kvm_mem);
3022                 if (r)
3023                         goto out;
3024                 break;
3025         }
3026         case KVM_GET_DIRTY_LOG: {
3027                 struct kvm_dirty_log log;
3028
3029                 r = -EFAULT;
3030                 if (copy_from_user(&log, argp, sizeof log))
3031                         goto out;
3032                 r = kvm_vm_ioctl_get_dirty_log(kvm, &log);
3033                 if (r)
3034                         goto out;
3035                 break;
3036         }
3037         case KVM_SET_MEMORY_ALIAS: {
3038                 struct kvm_memory_alias alias;
3039
3040                 r = -EFAULT;
3041                 if (copy_from_user(&alias, argp, sizeof alias))
3042                         goto out;
3043                 r = kvm_vm_ioctl_set_memory_alias(kvm, &alias);
3044                 if (r)
3045                         goto out;
3046                 break;
3047         }
3048         case KVM_CREATE_IRQCHIP:
3049                 r = -ENOMEM;
3050                 kvm->vpic = kvm_create_pic(kvm);
3051                 if (kvm->vpic) {
3052                         r = kvm_ioapic_init(kvm);
3053                         if (r) {
3054                                 kfree(kvm->vpic);
3055                                 kvm->vpic = NULL;
3056                                 goto out;
3057                         }
3058                 }
3059                 else
3060                         goto out;
3061                 break;
3062         case KVM_IRQ_LINE: {
3063                 struct kvm_irq_level irq_event;
3064
3065                 r = -EFAULT;
3066                 if (copy_from_user(&irq_event, argp, sizeof irq_event))
3067                         goto out;
3068                 if (irqchip_in_kernel(kvm)) {
3069                         mutex_lock(&kvm->lock);
3070                         if (irq_event.irq < 16)
3071                                 kvm_pic_set_irq(pic_irqchip(kvm),
3072                                         irq_event.irq,
3073                                         irq_event.level);
3074                         kvm_ioapic_set_irq(kvm->vioapic,
3075                                         irq_event.irq,
3076                                         irq_event.level);
3077                         mutex_unlock(&kvm->lock);
3078                         r = 0;
3079                 }
3080                 break;
3081         }
3082         case KVM_GET_IRQCHIP: {
3083                 /* 0: PIC master, 1: PIC slave, 2: IOAPIC */
3084                 struct kvm_irqchip chip;
3085
3086                 r = -EFAULT;
3087                 if (copy_from_user(&chip, argp, sizeof chip))
3088                         goto out;
3089                 r = -ENXIO;
3090                 if (!irqchip_in_kernel(kvm))
3091                         goto out;
3092                 r = kvm_vm_ioctl_get_irqchip(kvm, &chip);
3093                 if (r)
3094                         goto out;
3095                 r = -EFAULT;
3096                 if (copy_to_user(argp, &chip, sizeof chip))
3097                         goto out;
3098                 r = 0;
3099                 break;
3100         }
3101         case KVM_SET_IRQCHIP: {
3102                 /* 0: PIC master, 1: PIC slave, 2: IOAPIC */
3103                 struct kvm_irqchip chip;
3104
3105                 r = -EFAULT;
3106                 if (copy_from_user(&chip, argp, sizeof chip))
3107                         goto out;
3108                 r = -ENXIO;
3109                 if (!irqchip_in_kernel(kvm))
3110                         goto out;
3111                 r = kvm_vm_ioctl_set_irqchip(kvm, &chip);
3112                 if (r)
3113                         goto out;
3114                 r = 0;
3115                 break;
3116         }
3117         default:
3118                 ;
3119         }
3120 out:
3121         return r;
3122 }
3123
3124 static struct page *kvm_vm_nopage(struct vm_area_struct *vma,
3125                                   unsigned long address,
3126                                   int *type)
3127 {
3128         struct kvm *kvm = vma->vm_file->private_data;
3129         unsigned long pgoff;
3130         struct page *page;
3131
3132         pgoff = ((address - vma->vm_start) >> PAGE_SHIFT) + vma->vm_pgoff;
3133         page = gfn_to_page(kvm, pgoff);
3134         if (!page)
3135                 return NOPAGE_SIGBUS;
3136         get_page(page);
3137         if (type != NULL)
3138                 *type = VM_FAULT_MINOR;
3139
3140         return page;
3141 }
3142
3143 static struct vm_operations_struct kvm_vm_vm_ops = {
3144         .nopage = kvm_vm_nopage,
3145 };
3146
3147 static int kvm_vm_mmap(struct file *file, struct vm_area_struct *vma)
3148 {
3149         vma->vm_ops = &kvm_vm_vm_ops;
3150         return 0;
3151 }
3152
3153 static struct file_operations kvm_vm_fops = {
3154         .release        = kvm_vm_release,
3155         .unlocked_ioctl = kvm_vm_ioctl,
3156         .compat_ioctl   = kvm_vm_ioctl,
3157         .mmap           = kvm_vm_mmap,
3158 };
3159
3160 static int kvm_dev_ioctl_create_vm(void)
3161 {
3162         int fd, r;
3163         struct inode *inode;
3164         struct file *file;
3165         struct kvm *kvm;
3166
3167         kvm = kvm_create_vm();
3168         if (IS_ERR(kvm))
3169                 return PTR_ERR(kvm);
3170         r = anon_inode_getfd(&fd, &inode, &file, "kvm-vm", &kvm_vm_fops, kvm);
3171         if (r) {
3172                 kvm_destroy_vm(kvm);
3173                 return r;
3174         }
3175
3176         kvm->filp = file;
3177
3178         return fd;
3179 }
3180
3181 static long kvm_dev_ioctl(struct file *filp,
3182                           unsigned int ioctl, unsigned long arg)
3183 {
3184         void __user *argp = (void __user *)arg;
3185         long r = -EINVAL;
3186
3187         switch (ioctl) {
3188         case KVM_GET_API_VERSION:
3189                 r = -EINVAL;
3190                 if (arg)
3191                         goto out;
3192                 r = KVM_API_VERSION;
3193                 break;
3194         case KVM_CREATE_VM:
3195                 r = -EINVAL;
3196                 if (arg)
3197                         goto out;
3198                 r = kvm_dev_ioctl_create_vm();
3199                 break;
3200         case KVM_GET_MSR_INDEX_LIST: {
3201                 struct kvm_msr_list __user *user_msr_list = argp;
3202                 struct kvm_msr_list msr_list;
3203                 unsigned n;
3204
3205                 r = -EFAULT;
3206                 if (copy_from_user(&msr_list, user_msr_list, sizeof msr_list))
3207                         goto out;
3208                 n = msr_list.nmsrs;
3209                 msr_list.nmsrs = num_msrs_to_save + ARRAY_SIZE(emulated_msrs);
3210                 if (copy_to_user(user_msr_list, &msr_list, sizeof msr_list))
3211                         goto out;
3212                 r = -E2BIG;
3213                 if (n < num_msrs_to_save)
3214                         goto out;
3215                 r = -EFAULT;
3216                 if (copy_to_user(user_msr_list->indices, &msrs_to_save,
3217                                  num_msrs_to_save * sizeof(u32)))
3218                         goto out;
3219                 if (copy_to_user(user_msr_list->indices
3220                                  + num_msrs_to_save * sizeof(u32),
3221                                  &emulated_msrs,
3222                                  ARRAY_SIZE(emulated_msrs) * sizeof(u32)))
3223                         goto out;
3224                 r = 0;
3225                 break;
3226         }
3227         case KVM_CHECK_EXTENSION: {
3228                 int ext = (long)argp;
3229
3230                 switch (ext) {
3231                 case KVM_CAP_IRQCHIP:
3232                 case KVM_CAP_HLT:
3233                         r = 1;
3234                         break;
3235                 default:
3236                         r = 0;
3237                         break;
3238                 }
3239                 break;
3240         }
3241         case KVM_GET_VCPU_MMAP_SIZE:
3242                 r = -EINVAL;
3243                 if (arg)
3244                         goto out;
3245                 r = 2 * PAGE_SIZE;
3246                 break;
3247         default:
3248                 ;
3249         }
3250 out:
3251         return r;
3252 }
3253
3254 static struct file_operations kvm_chardev_ops = {
3255         .unlocked_ioctl = kvm_dev_ioctl,
3256         .compat_ioctl   = kvm_dev_ioctl,
3257 };
3258
3259 static struct miscdevice kvm_dev = {
3260         KVM_MINOR,
3261         "kvm",
3262         &kvm_chardev_ops,
3263 };
3264
3265 /*
3266  * Make sure that a cpu that is being hot-unplugged does not have any vcpus
3267  * cached on it.
3268  */
3269 static void decache_vcpus_on_cpu(int cpu)
3270 {
3271         struct kvm *vm;
3272         struct kvm_vcpu *vcpu;
3273         int i;
3274
3275         spin_lock(&kvm_lock);
3276         list_for_each_entry(vm, &vm_list, vm_list)
3277                 for (i = 0; i < KVM_MAX_VCPUS; ++i) {
3278                         vcpu = vm->vcpus[i];
3279                         if (!vcpu)
3280                                 continue;
3281                         /*
3282                          * If the vcpu is locked, then it is running on some
3283                          * other cpu and therefore it is not cached on the
3284                          * cpu in question.
3285                          *
3286                          * If it's not locked, check the last cpu it executed
3287                          * on.
3288                          */
3289                         if (mutex_trylock(&vcpu->mutex)) {
3290                                 if (vcpu->cpu == cpu) {
3291                                         kvm_x86_ops->vcpu_decache(vcpu);
3292                                         vcpu->cpu = -1;
3293                                 }
3294                                 mutex_unlock(&vcpu->mutex);
3295                         }
3296                 }
3297         spin_unlock(&kvm_lock);
3298 }
3299
3300 static void hardware_enable(void *junk)
3301 {
3302         int cpu = raw_smp_processor_id();
3303
3304         if (cpu_isset(cpu, cpus_hardware_enabled))
3305                 return;
3306         cpu_set(cpu, cpus_hardware_enabled);
3307         kvm_x86_ops->hardware_enable(NULL);
3308 }
3309
3310 static void hardware_disable(void *junk)
3311 {
3312         int cpu = raw_smp_processor_id();
3313
3314         if (!cpu_isset(cpu, cpus_hardware_enabled))
3315                 return;
3316         cpu_clear(cpu, cpus_hardware_enabled);
3317         decache_vcpus_on_cpu(cpu);
3318         kvm_x86_ops->hardware_disable(NULL);
3319 }
3320
3321 static int kvm_cpu_hotplug(struct notifier_block *notifier, unsigned long val,
3322                            void *v)
3323 {
3324         int cpu = (long)v;
3325
3326         switch (val) {
3327         case CPU_DYING:
3328         case CPU_DYING_FROZEN:
3329                 printk(KERN_INFO "kvm: disabling virtualization on CPU%d\n",
3330                        cpu);
3331                 hardware_disable(NULL);
3332                 break;
3333         case CPU_UP_CANCELED:
3334         case CPU_UP_CANCELED_FROZEN:
3335                 printk(KERN_INFO "kvm: disabling virtualization on CPU%d\n",
3336                        cpu);
3337                 smp_call_function_single(cpu, hardware_disable, NULL, 0, 1);
3338                 break;
3339         case CPU_ONLINE:
3340         case CPU_ONLINE_FROZEN:
3341                 printk(KERN_INFO "kvm: enabling virtualization on CPU%d\n",
3342                        cpu);
3343                 smp_call_function_single(cpu, hardware_enable, NULL, 0, 1);
3344                 break;
3345         }
3346         return NOTIFY_OK;
3347 }
3348
3349 static int kvm_reboot(struct notifier_block *notifier, unsigned long val,
3350                        void *v)
3351 {
3352         if (val == SYS_RESTART) {
3353                 /*
3354                  * Some (well, at least mine) BIOSes hang on reboot if
3355                  * in vmx root mode.
3356                  */
3357                 printk(KERN_INFO "kvm: exiting hardware virtualization\n");
3358                 on_each_cpu(hardware_disable, NULL, 0, 1);
3359         }
3360         return NOTIFY_OK;
3361 }
3362
3363 static struct notifier_block kvm_reboot_notifier = {
3364         .notifier_call = kvm_reboot,
3365         .priority = 0,
3366 };
3367
3368 void kvm_io_bus_init(struct kvm_io_bus *bus)
3369 {
3370         memset(bus, 0, sizeof(*bus));
3371 }
3372
3373 void kvm_io_bus_destroy(struct kvm_io_bus *bus)
3374 {
3375         int i;
3376
3377         for (i = 0; i < bus->dev_count; i++) {
3378                 struct kvm_io_device *pos = bus->devs[i];
3379
3380                 kvm_iodevice_destructor(pos);
3381         }
3382 }
3383
3384 struct kvm_io_device *kvm_io_bus_find_dev(struct kvm_io_bus *bus, gpa_t addr)
3385 {
3386         int i;
3387
3388         for (i = 0; i < bus->dev_count; i++) {
3389                 struct kvm_io_device *pos = bus->devs[i];
3390
3391                 if (pos->in_range(pos, addr))
3392                         return pos;
3393         }
3394
3395         return NULL;
3396 }
3397
3398 void kvm_io_bus_register_dev(struct kvm_io_bus *bus, struct kvm_io_device *dev)
3399 {
3400         BUG_ON(bus->dev_count > (NR_IOBUS_DEVS-1));
3401
3402         bus->devs[bus->dev_count++] = dev;
3403 }
3404
3405 static struct notifier_block kvm_cpu_notifier = {
3406         .notifier_call = kvm_cpu_hotplug,
3407         .priority = 20, /* must be > scheduler priority */
3408 };
3409
3410 static u64 stat_get(void *_offset)
3411 {
3412         unsigned offset = (long)_offset;
3413         u64 total = 0;
3414         struct kvm *kvm;
3415         struct kvm_vcpu *vcpu;
3416         int i;
3417
3418         spin_lock(&kvm_lock);
3419         list_for_each_entry(kvm, &vm_list, vm_list)
3420                 for (i = 0; i < KVM_MAX_VCPUS; ++i) {
3421                         vcpu = kvm->vcpus[i];
3422                         if (vcpu)
3423                                 total += *(u32 *)((void *)vcpu + offset);
3424                 }
3425         spin_unlock(&kvm_lock);
3426         return total;
3427 }
3428
3429 DEFINE_SIMPLE_ATTRIBUTE(stat_fops, stat_get, NULL, "%llu\n");
3430
3431 static __init void kvm_init_debug(void)
3432 {
3433         struct kvm_stats_debugfs_item *p;
3434
3435         debugfs_dir = debugfs_create_dir("kvm", NULL);
3436         for (p = debugfs_entries; p->name; ++p)
3437                 p->dentry = debugfs_create_file(p->name, 0444, debugfs_dir,
3438                                                 (void *)(long)p->offset,
3439                                                 &stat_fops);
3440 }
3441
3442 static void kvm_exit_debug(void)
3443 {
3444         struct kvm_stats_debugfs_item *p;
3445
3446         for (p = debugfs_entries; p->name; ++p)
3447                 debugfs_remove(p->dentry);
3448         debugfs_remove(debugfs_dir);
3449 }
3450
3451 static int kvm_suspend(struct sys_device *dev, pm_message_t state)
3452 {
3453         hardware_disable(NULL);
3454         return 0;
3455 }
3456
3457 static int kvm_resume(struct sys_device *dev)
3458 {
3459         hardware_enable(NULL);
3460         return 0;
3461 }
3462
3463 static struct sysdev_class kvm_sysdev_class = {
3464         set_kset_name("kvm"),
3465         .suspend = kvm_suspend,
3466         .resume = kvm_resume,
3467 };
3468
3469 static struct sys_device kvm_sysdev = {
3470         .id = 0,
3471         .cls = &kvm_sysdev_class,
3472 };
3473
3474 hpa_t bad_page_address;
3475
3476 static inline
3477 struct kvm_vcpu *preempt_notifier_to_vcpu(struct preempt_notifier *pn)
3478 {
3479         return container_of(pn, struct kvm_vcpu, preempt_notifier);
3480 }
3481
3482 static void kvm_sched_in(struct preempt_notifier *pn, int cpu)
3483 {
3484         struct kvm_vcpu *vcpu = preempt_notifier_to_vcpu(pn);
3485
3486         kvm_x86_ops->vcpu_load(vcpu, cpu);
3487 }
3488
3489 static void kvm_sched_out(struct preempt_notifier *pn,
3490                           struct task_struct *next)
3491 {
3492         struct kvm_vcpu *vcpu = preempt_notifier_to_vcpu(pn);
3493
3494         kvm_x86_ops->vcpu_put(vcpu);
3495 }
3496
3497 int kvm_init_x86(struct kvm_x86_ops *ops, unsigned int vcpu_size,
3498                   struct module *module)
3499 {
3500         int r;
3501         int cpu;
3502
3503         if (kvm_x86_ops) {
3504                 printk(KERN_ERR "kvm: already loaded the other module\n");
3505                 return -EEXIST;
3506         }
3507
3508         if (!ops->cpu_has_kvm_support()) {
3509                 printk(KERN_ERR "kvm: no hardware support\n");
3510                 return -EOPNOTSUPP;
3511         }
3512         if (ops->disabled_by_bios()) {
3513                 printk(KERN_ERR "kvm: disabled by bios\n");
3514                 return -EOPNOTSUPP;
3515         }
3516
3517         kvm_x86_ops = ops;
3518
3519         r = kvm_x86_ops->hardware_setup();
3520         if (r < 0)
3521                 goto out;
3522
3523         for_each_online_cpu(cpu) {
3524                 smp_call_function_single(cpu,
3525                                 kvm_x86_ops->check_processor_compatibility,
3526                                 &r, 0, 1);
3527                 if (r < 0)
3528                         goto out_free_0;
3529         }
3530
3531         on_each_cpu(hardware_enable, NULL, 0, 1);
3532         r = register_cpu_notifier(&kvm_cpu_notifier);
3533         if (r)
3534                 goto out_free_1;
3535         register_reboot_notifier(&kvm_reboot_notifier);
3536
3537         r = sysdev_class_register(&kvm_sysdev_class);
3538         if (r)
3539                 goto out_free_2;
3540
3541         r = sysdev_register(&kvm_sysdev);
3542         if (r)
3543                 goto out_free_3;
3544
3545         /* A kmem cache lets us meet the alignment requirements of fx_save. */
3546         kvm_vcpu_cache = kmem_cache_create("kvm_vcpu", vcpu_size,
3547                                            __alignof__(struct kvm_vcpu), 0, 0);
3548         if (!kvm_vcpu_cache) {
3549                 r = -ENOMEM;
3550                 goto out_free_4;
3551         }
3552
3553         kvm_chardev_ops.owner = module;
3554
3555         r = misc_register(&kvm_dev);
3556         if (r) {
3557                 printk (KERN_ERR "kvm: misc device register failed\n");
3558                 goto out_free;
3559         }
3560
3561         kvm_preempt_ops.sched_in = kvm_sched_in;
3562         kvm_preempt_ops.sched_out = kvm_sched_out;
3563
3564         return r;
3565
3566 out_free:
3567         kmem_cache_destroy(kvm_vcpu_cache);
3568 out_free_4:
3569         sysdev_unregister(&kvm_sysdev);
3570 out_free_3:
3571         sysdev_class_unregister(&kvm_sysdev_class);
3572 out_free_2:
3573         unregister_reboot_notifier(&kvm_reboot_notifier);
3574         unregister_cpu_notifier(&kvm_cpu_notifier);
3575 out_free_1:
3576         on_each_cpu(hardware_disable, NULL, 0, 1);
3577 out_free_0:
3578         kvm_x86_ops->hardware_unsetup();
3579 out:
3580         kvm_x86_ops = NULL;
3581         return r;
3582 }
3583
3584 void kvm_exit_x86(void)
3585 {
3586         misc_deregister(&kvm_dev);
3587         kmem_cache_destroy(kvm_vcpu_cache);
3588         sysdev_unregister(&kvm_sysdev);
3589         sysdev_class_unregister(&kvm_sysdev_class);
3590         unregister_reboot_notifier(&kvm_reboot_notifier);
3591         unregister_cpu_notifier(&kvm_cpu_notifier);
3592         on_each_cpu(hardware_disable, NULL, 0, 1);
3593         kvm_x86_ops->hardware_unsetup();
3594         kvm_x86_ops = NULL;
3595 }
3596
3597 static __init int kvm_init(void)
3598 {
3599         static struct page *bad_page;
3600         int r;
3601
3602         r = kvm_mmu_module_init();
3603         if (r)
3604                 goto out4;
3605
3606         kvm_init_debug();
3607
3608         kvm_init_msr_list();
3609
3610         if ((bad_page = alloc_page(GFP_KERNEL)) == NULL) {
3611                 r = -ENOMEM;
3612                 goto out;
3613         }
3614
3615         bad_page_address = page_to_pfn(bad_page) << PAGE_SHIFT;
3616         memset(__va(bad_page_address), 0, PAGE_SIZE);
3617
3618         return 0;
3619
3620 out:
3621         kvm_exit_debug();
3622         kvm_mmu_module_exit();
3623 out4:
3624         return r;
3625 }
3626
3627 static __exit void kvm_exit(void)
3628 {
3629         kvm_exit_debug();
3630         __free_page(pfn_to_page(bad_page_address >> PAGE_SHIFT));
3631         kvm_mmu_module_exit();
3632 }
3633
3634 module_init(kvm_init)
3635 module_exit(kvm_exit)
3636
3637 EXPORT_SYMBOL_GPL(kvm_init_x86);
3638 EXPORT_SYMBOL_GPL(kvm_exit_x86);