OSDN Git Service

KVM: kvm_vm_ioctl_get_dirty_log restore "nothing dirty" optimization
[sagit-ice-cold/kernel_xiaomi_msm8998.git] / 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
22 #include <linux/kvm.h>
23 #include <linux/module.h>
24 #include <linux/errno.h>
25 #include <linux/percpu.h>
26 #include <linux/gfp.h>
27 #include <linux/mm.h>
28 #include <linux/miscdevice.h>
29 #include <linux/vmalloc.h>
30 #include <linux/reboot.h>
31 #include <linux/debugfs.h>
32 #include <linux/highmem.h>
33 #include <linux/file.h>
34 #include <linux/sysdev.h>
35 #include <linux/cpu.h>
36 #include <linux/sched.h>
37 #include <linux/cpumask.h>
38 #include <linux/smp.h>
39 #include <linux/anon_inodes.h>
40
41 #include <asm/processor.h>
42 #include <asm/msr.h>
43 #include <asm/io.h>
44 #include <asm/uaccess.h>
45 #include <asm/desc.h>
46
47 MODULE_AUTHOR("Qumranet");
48 MODULE_LICENSE("GPL");
49
50 static DEFINE_SPINLOCK(kvm_lock);
51 static LIST_HEAD(vm_list);
52
53 static cpumask_t cpus_hardware_enabled;
54
55 struct kvm_arch_ops *kvm_arch_ops;
56 struct kmem_cache *kvm_vcpu_cache;
57 EXPORT_SYMBOL_GPL(kvm_vcpu_cache);
58
59 static __read_mostly struct preempt_ops kvm_preempt_ops;
60
61 #define STAT_OFFSET(x) offsetof(struct kvm_vcpu, stat.x)
62
63 static struct kvm_stats_debugfs_item {
64         const char *name;
65         int offset;
66         struct dentry *dentry;
67 } debugfs_entries[] = {
68         { "pf_fixed", STAT_OFFSET(pf_fixed) },
69         { "pf_guest", STAT_OFFSET(pf_guest) },
70         { "tlb_flush", STAT_OFFSET(tlb_flush) },
71         { "invlpg", STAT_OFFSET(invlpg) },
72         { "exits", STAT_OFFSET(exits) },
73         { "io_exits", STAT_OFFSET(io_exits) },
74         { "mmio_exits", STAT_OFFSET(mmio_exits) },
75         { "signal_exits", STAT_OFFSET(signal_exits) },
76         { "irq_window", STAT_OFFSET(irq_window_exits) },
77         { "halt_exits", STAT_OFFSET(halt_exits) },
78         { "request_irq", STAT_OFFSET(request_irq_exits) },
79         { "irq_exits", STAT_OFFSET(irq_exits) },
80         { "light_exits", STAT_OFFSET(light_exits) },
81         { "efer_reload", STAT_OFFSET(efer_reload) },
82         { NULL }
83 };
84
85 static struct dentry *debugfs_dir;
86
87 #define MAX_IO_MSRS 256
88
89 #define CR0_RESERVED_BITS                                               \
90         (~(unsigned long)(X86_CR0_PE | X86_CR0_MP | X86_CR0_EM | X86_CR0_TS \
91                           | X86_CR0_ET | X86_CR0_NE | X86_CR0_WP | X86_CR0_AM \
92                           | X86_CR0_NW | X86_CR0_CD | X86_CR0_PG))
93 #define CR4_RESERVED_BITS                                               \
94         (~(unsigned long)(X86_CR4_VME | X86_CR4_PVI | X86_CR4_TSD | X86_CR4_DE\
95                           | X86_CR4_PSE | X86_CR4_PAE | X86_CR4_MCE     \
96                           | X86_CR4_PGE | X86_CR4_PCE | X86_CR4_OSFXSR  \
97                           | X86_CR4_OSXMMEXCPT | X86_CR4_VMXE))
98
99 #define CR8_RESERVED_BITS (~(unsigned long)X86_CR8_TPR)
100 #define EFER_RESERVED_BITS 0xfffffffffffff2fe
101
102 #ifdef CONFIG_X86_64
103 // LDT or TSS descriptor in the GDT. 16 bytes.
104 struct segment_descriptor_64 {
105         struct segment_descriptor s;
106         u32 base_higher;
107         u32 pad_zero;
108 };
109
110 #endif
111
112 static long kvm_vcpu_ioctl(struct file *file, unsigned int ioctl,
113                            unsigned long arg);
114
115 unsigned long segment_base(u16 selector)
116 {
117         struct descriptor_table gdt;
118         struct segment_descriptor *d;
119         unsigned long table_base;
120         typedef unsigned long ul;
121         unsigned long v;
122
123         if (selector == 0)
124                 return 0;
125
126         asm ("sgdt %0" : "=m"(gdt));
127         table_base = gdt.base;
128
129         if (selector & 4) {           /* from ldt */
130                 u16 ldt_selector;
131
132                 asm ("sldt %0" : "=g"(ldt_selector));
133                 table_base = segment_base(ldt_selector);
134         }
135         d = (struct segment_descriptor *)(table_base + (selector & ~7));
136         v = d->base_low | ((ul)d->base_mid << 16) | ((ul)d->base_high << 24);
137 #ifdef CONFIG_X86_64
138         if (d->system == 0
139             && (d->type == 2 || d->type == 9 || d->type == 11))
140                 v |= ((ul)((struct segment_descriptor_64 *)d)->base_higher) << 32;
141 #endif
142         return v;
143 }
144 EXPORT_SYMBOL_GPL(segment_base);
145
146 static inline int valid_vcpu(int n)
147 {
148         return likely(n >= 0 && n < KVM_MAX_VCPUS);
149 }
150
151 void kvm_load_guest_fpu(struct kvm_vcpu *vcpu)
152 {
153         if (!vcpu->fpu_active || vcpu->guest_fpu_loaded)
154                 return;
155
156         vcpu->guest_fpu_loaded = 1;
157         fx_save(&vcpu->host_fx_image);
158         fx_restore(&vcpu->guest_fx_image);
159 }
160 EXPORT_SYMBOL_GPL(kvm_load_guest_fpu);
161
162 void kvm_put_guest_fpu(struct kvm_vcpu *vcpu)
163 {
164         if (!vcpu->guest_fpu_loaded)
165                 return;
166
167         vcpu->guest_fpu_loaded = 0;
168         fx_save(&vcpu->guest_fx_image);
169         fx_restore(&vcpu->host_fx_image);
170 }
171 EXPORT_SYMBOL_GPL(kvm_put_guest_fpu);
172
173 /*
174  * Switches to specified vcpu, until a matching vcpu_put()
175  */
176 static void vcpu_load(struct kvm_vcpu *vcpu)
177 {
178         int cpu;
179
180         mutex_lock(&vcpu->mutex);
181         cpu = get_cpu();
182         preempt_notifier_register(&vcpu->preempt_notifier);
183         kvm_arch_ops->vcpu_load(vcpu, cpu);
184         put_cpu();
185 }
186
187 static void vcpu_put(struct kvm_vcpu *vcpu)
188 {
189         preempt_disable();
190         kvm_arch_ops->vcpu_put(vcpu);
191         preempt_notifier_unregister(&vcpu->preempt_notifier);
192         preempt_enable();
193         mutex_unlock(&vcpu->mutex);
194 }
195
196 static void ack_flush(void *_completed)
197 {
198         atomic_t *completed = _completed;
199
200         atomic_inc(completed);
201 }
202
203 void kvm_flush_remote_tlbs(struct kvm *kvm)
204 {
205         int i, cpu, needed;
206         cpumask_t cpus;
207         struct kvm_vcpu *vcpu;
208         atomic_t completed;
209
210         atomic_set(&completed, 0);
211         cpus_clear(cpus);
212         needed = 0;
213         for (i = 0; i < KVM_MAX_VCPUS; ++i) {
214                 vcpu = kvm->vcpus[i];
215                 if (!vcpu)
216                         continue;
217                 if (test_and_set_bit(KVM_TLB_FLUSH, &vcpu->requests))
218                         continue;
219                 cpu = vcpu->cpu;
220                 if (cpu != -1 && cpu != raw_smp_processor_id())
221                         if (!cpu_isset(cpu, cpus)) {
222                                 cpu_set(cpu, cpus);
223                                 ++needed;
224                         }
225         }
226
227         /*
228          * We really want smp_call_function_mask() here.  But that's not
229          * available, so ipi all cpus in parallel and wait for them
230          * to complete.
231          */
232         for (cpu = first_cpu(cpus); cpu != NR_CPUS; cpu = next_cpu(cpu, cpus))
233                 smp_call_function_single(cpu, ack_flush, &completed, 1, 0);
234         while (atomic_read(&completed) != needed) {
235                 cpu_relax();
236                 barrier();
237         }
238 }
239
240 int kvm_vcpu_init(struct kvm_vcpu *vcpu, struct kvm *kvm, unsigned id)
241 {
242         struct page *page;
243         int r;
244
245         mutex_init(&vcpu->mutex);
246         vcpu->cpu = -1;
247         vcpu->mmu.root_hpa = INVALID_PAGE;
248         vcpu->kvm = kvm;
249         vcpu->vcpu_id = id;
250
251         page = alloc_page(GFP_KERNEL | __GFP_ZERO);
252         if (!page) {
253                 r = -ENOMEM;
254                 goto fail;
255         }
256         vcpu->run = page_address(page);
257
258         page = alloc_page(GFP_KERNEL | __GFP_ZERO);
259         if (!page) {
260                 r = -ENOMEM;
261                 goto fail_free_run;
262         }
263         vcpu->pio_data = page_address(page);
264
265         r = kvm_mmu_create(vcpu);
266         if (r < 0)
267                 goto fail_free_pio_data;
268
269         return 0;
270
271 fail_free_pio_data:
272         free_page((unsigned long)vcpu->pio_data);
273 fail_free_run:
274         free_page((unsigned long)vcpu->run);
275 fail:
276         return -ENOMEM;
277 }
278 EXPORT_SYMBOL_GPL(kvm_vcpu_init);
279
280 void kvm_vcpu_uninit(struct kvm_vcpu *vcpu)
281 {
282         kvm_mmu_destroy(vcpu);
283         free_page((unsigned long)vcpu->pio_data);
284         free_page((unsigned long)vcpu->run);
285 }
286 EXPORT_SYMBOL_GPL(kvm_vcpu_uninit);
287
288 static struct kvm *kvm_create_vm(void)
289 {
290         struct kvm *kvm = kzalloc(sizeof(struct kvm), GFP_KERNEL);
291
292         if (!kvm)
293                 return ERR_PTR(-ENOMEM);
294
295         kvm_io_bus_init(&kvm->pio_bus);
296         mutex_init(&kvm->lock);
297         INIT_LIST_HEAD(&kvm->active_mmu_pages);
298         kvm_io_bus_init(&kvm->mmio_bus);
299         spin_lock(&kvm_lock);
300         list_add(&kvm->vm_list, &vm_list);
301         spin_unlock(&kvm_lock);
302         return kvm;
303 }
304
305 static int kvm_dev_open(struct inode *inode, struct file *filp)
306 {
307         return 0;
308 }
309
310 /*
311  * Free any memory in @free but not in @dont.
312  */
313 static void kvm_free_physmem_slot(struct kvm_memory_slot *free,
314                                   struct kvm_memory_slot *dont)
315 {
316         int i;
317
318         if (!dont || free->phys_mem != dont->phys_mem)
319                 if (free->phys_mem) {
320                         for (i = 0; i < free->npages; ++i)
321                                 if (free->phys_mem[i])
322                                         __free_page(free->phys_mem[i]);
323                         vfree(free->phys_mem);
324                 }
325
326         if (!dont || free->dirty_bitmap != dont->dirty_bitmap)
327                 vfree(free->dirty_bitmap);
328
329         free->phys_mem = NULL;
330         free->npages = 0;
331         free->dirty_bitmap = NULL;
332 }
333
334 static void kvm_free_physmem(struct kvm *kvm)
335 {
336         int i;
337
338         for (i = 0; i < kvm->nmemslots; ++i)
339                 kvm_free_physmem_slot(&kvm->memslots[i], NULL);
340 }
341
342 static void free_pio_guest_pages(struct kvm_vcpu *vcpu)
343 {
344         int i;
345
346         for (i = 0; i < ARRAY_SIZE(vcpu->pio.guest_pages); ++i)
347                 if (vcpu->pio.guest_pages[i]) {
348                         __free_page(vcpu->pio.guest_pages[i]);
349                         vcpu->pio.guest_pages[i] = NULL;
350                 }
351 }
352
353 static void kvm_unload_vcpu_mmu(struct kvm_vcpu *vcpu)
354 {
355         vcpu_load(vcpu);
356         kvm_mmu_unload(vcpu);
357         vcpu_put(vcpu);
358 }
359
360 static void kvm_free_vcpus(struct kvm *kvm)
361 {
362         unsigned int i;
363
364         /*
365          * Unpin any mmu pages first.
366          */
367         for (i = 0; i < KVM_MAX_VCPUS; ++i)
368                 if (kvm->vcpus[i])
369                         kvm_unload_vcpu_mmu(kvm->vcpus[i]);
370         for (i = 0; i < KVM_MAX_VCPUS; ++i) {
371                 if (kvm->vcpus[i]) {
372                         kvm_arch_ops->vcpu_free(kvm->vcpus[i]);
373                         kvm->vcpus[i] = NULL;
374                 }
375         }
376
377 }
378
379 static int kvm_dev_release(struct inode *inode, struct file *filp)
380 {
381         return 0;
382 }
383
384 static void kvm_destroy_vm(struct kvm *kvm)
385 {
386         spin_lock(&kvm_lock);
387         list_del(&kvm->vm_list);
388         spin_unlock(&kvm_lock);
389         kvm_io_bus_destroy(&kvm->pio_bus);
390         kvm_io_bus_destroy(&kvm->mmio_bus);
391         kvm_free_vcpus(kvm);
392         kvm_free_physmem(kvm);
393         kfree(kvm);
394 }
395
396 static int kvm_vm_release(struct inode *inode, struct file *filp)
397 {
398         struct kvm *kvm = filp->private_data;
399
400         kvm_destroy_vm(kvm);
401         return 0;
402 }
403
404 static void inject_gp(struct kvm_vcpu *vcpu)
405 {
406         kvm_arch_ops->inject_gp(vcpu, 0);
407 }
408
409 /*
410  * Load the pae pdptrs.  Return true is they are all valid.
411  */
412 static int load_pdptrs(struct kvm_vcpu *vcpu, unsigned long cr3)
413 {
414         gfn_t pdpt_gfn = cr3 >> PAGE_SHIFT;
415         unsigned offset = ((cr3 & (PAGE_SIZE-1)) >> 5) << 2;
416         int i;
417         u64 *pdpt;
418         int ret;
419         struct page *page;
420         u64 pdpte[ARRAY_SIZE(vcpu->pdptrs)];
421
422         mutex_lock(&vcpu->kvm->lock);
423         page = gfn_to_page(vcpu->kvm, pdpt_gfn);
424         if (!page) {
425                 ret = 0;
426                 goto out;
427         }
428
429         pdpt = kmap_atomic(page, KM_USER0);
430         memcpy(pdpte, pdpt+offset, sizeof(pdpte));
431         kunmap_atomic(pdpt, KM_USER0);
432
433         for (i = 0; i < ARRAY_SIZE(pdpte); ++i) {
434                 if ((pdpte[i] & 1) && (pdpte[i] & 0xfffffff0000001e6ull)) {
435                         ret = 0;
436                         goto out;
437                 }
438         }
439         ret = 1;
440
441         memcpy(vcpu->pdptrs, pdpte, sizeof(vcpu->pdptrs));
442 out:
443         mutex_unlock(&vcpu->kvm->lock);
444
445         return ret;
446 }
447
448 void set_cr0(struct kvm_vcpu *vcpu, unsigned long cr0)
449 {
450         if (cr0 & CR0_RESERVED_BITS) {
451                 printk(KERN_DEBUG "set_cr0: 0x%lx #GP, reserved bits 0x%lx\n",
452                        cr0, vcpu->cr0);
453                 inject_gp(vcpu);
454                 return;
455         }
456
457         if ((cr0 & X86_CR0_NW) && !(cr0 & X86_CR0_CD)) {
458                 printk(KERN_DEBUG "set_cr0: #GP, CD == 0 && NW == 1\n");
459                 inject_gp(vcpu);
460                 return;
461         }
462
463         if ((cr0 & X86_CR0_PG) && !(cr0 & X86_CR0_PE)) {
464                 printk(KERN_DEBUG "set_cr0: #GP, set PG flag "
465                        "and a clear PE flag\n");
466                 inject_gp(vcpu);
467                 return;
468         }
469
470         if (!is_paging(vcpu) && (cr0 & X86_CR0_PG)) {
471 #ifdef CONFIG_X86_64
472                 if ((vcpu->shadow_efer & EFER_LME)) {
473                         int cs_db, cs_l;
474
475                         if (!is_pae(vcpu)) {
476                                 printk(KERN_DEBUG "set_cr0: #GP, start paging "
477                                        "in long mode while PAE is disabled\n");
478                                 inject_gp(vcpu);
479                                 return;
480                         }
481                         kvm_arch_ops->get_cs_db_l_bits(vcpu, &cs_db, &cs_l);
482                         if (cs_l) {
483                                 printk(KERN_DEBUG "set_cr0: #GP, start paging "
484                                        "in long mode while CS.L == 1\n");
485                                 inject_gp(vcpu);
486                                 return;
487
488                         }
489                 } else
490 #endif
491                 if (is_pae(vcpu) && !load_pdptrs(vcpu, vcpu->cr3)) {
492                         printk(KERN_DEBUG "set_cr0: #GP, pdptrs "
493                                "reserved bits\n");
494                         inject_gp(vcpu);
495                         return;
496                 }
497
498         }
499
500         kvm_arch_ops->set_cr0(vcpu, cr0);
501         vcpu->cr0 = cr0;
502
503         mutex_lock(&vcpu->kvm->lock);
504         kvm_mmu_reset_context(vcpu);
505         mutex_unlock(&vcpu->kvm->lock);
506         return;
507 }
508 EXPORT_SYMBOL_GPL(set_cr0);
509
510 void lmsw(struct kvm_vcpu *vcpu, unsigned long msw)
511 {
512         set_cr0(vcpu, (vcpu->cr0 & ~0x0ful) | (msw & 0x0f));
513 }
514 EXPORT_SYMBOL_GPL(lmsw);
515
516 void set_cr4(struct kvm_vcpu *vcpu, unsigned long cr4)
517 {
518         if (cr4 & CR4_RESERVED_BITS) {
519                 printk(KERN_DEBUG "set_cr4: #GP, reserved bits\n");
520                 inject_gp(vcpu);
521                 return;
522         }
523
524         if (is_long_mode(vcpu)) {
525                 if (!(cr4 & X86_CR4_PAE)) {
526                         printk(KERN_DEBUG "set_cr4: #GP, clearing PAE while "
527                                "in long mode\n");
528                         inject_gp(vcpu);
529                         return;
530                 }
531         } else if (is_paging(vcpu) && !is_pae(vcpu) && (cr4 & X86_CR4_PAE)
532                    && !load_pdptrs(vcpu, vcpu->cr3)) {
533                 printk(KERN_DEBUG "set_cr4: #GP, pdptrs reserved bits\n");
534                 inject_gp(vcpu);
535                 return;
536         }
537
538         if (cr4 & X86_CR4_VMXE) {
539                 printk(KERN_DEBUG "set_cr4: #GP, setting VMXE\n");
540                 inject_gp(vcpu);
541                 return;
542         }
543         kvm_arch_ops->set_cr4(vcpu, cr4);
544         mutex_lock(&vcpu->kvm->lock);
545         kvm_mmu_reset_context(vcpu);
546         mutex_unlock(&vcpu->kvm->lock);
547 }
548 EXPORT_SYMBOL_GPL(set_cr4);
549
550 void set_cr3(struct kvm_vcpu *vcpu, unsigned long cr3)
551 {
552         if (is_long_mode(vcpu)) {
553                 if (cr3 & CR3_L_MODE_RESERVED_BITS) {
554                         printk(KERN_DEBUG "set_cr3: #GP, reserved bits\n");
555                         inject_gp(vcpu);
556                         return;
557                 }
558         } else {
559                 if (is_pae(vcpu)) {
560                         if (cr3 & CR3_PAE_RESERVED_BITS) {
561                                 printk(KERN_DEBUG
562                                        "set_cr3: #GP, reserved bits\n");
563                                 inject_gp(vcpu);
564                                 return;
565                         }
566                         if (is_paging(vcpu) && !load_pdptrs(vcpu, cr3)) {
567                                 printk(KERN_DEBUG "set_cr3: #GP, pdptrs "
568                                        "reserved bits\n");
569                                 inject_gp(vcpu);
570                                 return;
571                         }
572                 } else {
573                         if (cr3 & CR3_NONPAE_RESERVED_BITS) {
574                                 printk(KERN_DEBUG
575                                        "set_cr3: #GP, reserved bits\n");
576                                 inject_gp(vcpu);
577                                 return;
578                         }
579                 }
580         }
581
582         vcpu->cr3 = cr3;
583         mutex_lock(&vcpu->kvm->lock);
584         /*
585          * Does the new cr3 value map to physical memory? (Note, we
586          * catch an invalid cr3 even in real-mode, because it would
587          * cause trouble later on when we turn on paging anyway.)
588          *
589          * A real CPU would silently accept an invalid cr3 and would
590          * attempt to use it - with largely undefined (and often hard
591          * to debug) behavior on the guest side.
592          */
593         if (unlikely(!gfn_to_memslot(vcpu->kvm, cr3 >> PAGE_SHIFT)))
594                 inject_gp(vcpu);
595         else
596                 vcpu->mmu.new_cr3(vcpu);
597         mutex_unlock(&vcpu->kvm->lock);
598 }
599 EXPORT_SYMBOL_GPL(set_cr3);
600
601 void set_cr8(struct kvm_vcpu *vcpu, unsigned long cr8)
602 {
603         if (cr8 & CR8_RESERVED_BITS) {
604                 printk(KERN_DEBUG "set_cr8: #GP, reserved bits 0x%lx\n", cr8);
605                 inject_gp(vcpu);
606                 return;
607         }
608         vcpu->cr8 = cr8;
609 }
610 EXPORT_SYMBOL_GPL(set_cr8);
611
612 void fx_init(struct kvm_vcpu *vcpu)
613 {
614         unsigned after_mxcsr_mask;
615
616         /* Initialize guest FPU by resetting ours and saving into guest's */
617         preempt_disable();
618         fx_save(&vcpu->host_fx_image);
619         fpu_init();
620         fx_save(&vcpu->guest_fx_image);
621         fx_restore(&vcpu->host_fx_image);
622         preempt_enable();
623
624         after_mxcsr_mask = offsetof(struct i387_fxsave_struct, st_space);
625         vcpu->guest_fx_image.mxcsr = 0x1f80;
626         memset((void *)&vcpu->guest_fx_image + after_mxcsr_mask,
627                0, sizeof(struct i387_fxsave_struct) - after_mxcsr_mask);
628 }
629 EXPORT_SYMBOL_GPL(fx_init);
630
631 /*
632  * Allocate some memory and give it an address in the guest physical address
633  * space.
634  *
635  * Discontiguous memory is allowed, mostly for framebuffers.
636  */
637 static int kvm_vm_ioctl_set_memory_region(struct kvm *kvm,
638                                           struct kvm_memory_region *mem)
639 {
640         int r;
641         gfn_t base_gfn;
642         unsigned long npages;
643         unsigned long i;
644         struct kvm_memory_slot *memslot;
645         struct kvm_memory_slot old, new;
646         int memory_config_version;
647
648         r = -EINVAL;
649         /* General sanity checks */
650         if (mem->memory_size & (PAGE_SIZE - 1))
651                 goto out;
652         if (mem->guest_phys_addr & (PAGE_SIZE - 1))
653                 goto out;
654         if (mem->slot >= KVM_MEMORY_SLOTS)
655                 goto out;
656         if (mem->guest_phys_addr + mem->memory_size < mem->guest_phys_addr)
657                 goto out;
658
659         memslot = &kvm->memslots[mem->slot];
660         base_gfn = mem->guest_phys_addr >> PAGE_SHIFT;
661         npages = mem->memory_size >> PAGE_SHIFT;
662
663         if (!npages)
664                 mem->flags &= ~KVM_MEM_LOG_DIRTY_PAGES;
665
666 raced:
667         mutex_lock(&kvm->lock);
668
669         memory_config_version = kvm->memory_config_version;
670         new = old = *memslot;
671
672         new.base_gfn = base_gfn;
673         new.npages = npages;
674         new.flags = mem->flags;
675
676         /* Disallow changing a memory slot's size. */
677         r = -EINVAL;
678         if (npages && old.npages && npages != old.npages)
679                 goto out_unlock;
680
681         /* Check for overlaps */
682         r = -EEXIST;
683         for (i = 0; i < KVM_MEMORY_SLOTS; ++i) {
684                 struct kvm_memory_slot *s = &kvm->memslots[i];
685
686                 if (s == memslot)
687                         continue;
688                 if (!((base_gfn + npages <= s->base_gfn) ||
689                       (base_gfn >= s->base_gfn + s->npages)))
690                         goto out_unlock;
691         }
692         /*
693          * Do memory allocations outside lock.  memory_config_version will
694          * detect any races.
695          */
696         mutex_unlock(&kvm->lock);
697
698         /* Deallocate if slot is being removed */
699         if (!npages)
700                 new.phys_mem = NULL;
701
702         /* Free page dirty bitmap if unneeded */
703         if (!(new.flags & KVM_MEM_LOG_DIRTY_PAGES))
704                 new.dirty_bitmap = NULL;
705
706         r = -ENOMEM;
707
708         /* Allocate if a slot is being created */
709         if (npages && !new.phys_mem) {
710                 new.phys_mem = vmalloc(npages * sizeof(struct page *));
711
712                 if (!new.phys_mem)
713                         goto out_free;
714
715                 memset(new.phys_mem, 0, npages * sizeof(struct page *));
716                 for (i = 0; i < npages; ++i) {
717                         new.phys_mem[i] = alloc_page(GFP_HIGHUSER
718                                                      | __GFP_ZERO);
719                         if (!new.phys_mem[i])
720                                 goto out_free;
721                         set_page_private(new.phys_mem[i],0);
722                 }
723         }
724
725         /* Allocate page dirty bitmap if needed */
726         if ((new.flags & KVM_MEM_LOG_DIRTY_PAGES) && !new.dirty_bitmap) {
727                 unsigned dirty_bytes = ALIGN(npages, BITS_PER_LONG) / 8;
728
729                 new.dirty_bitmap = vmalloc(dirty_bytes);
730                 if (!new.dirty_bitmap)
731                         goto out_free;
732                 memset(new.dirty_bitmap, 0, dirty_bytes);
733         }
734
735         mutex_lock(&kvm->lock);
736
737         if (memory_config_version != kvm->memory_config_version) {
738                 mutex_unlock(&kvm->lock);
739                 kvm_free_physmem_slot(&new, &old);
740                 goto raced;
741         }
742
743         r = -EAGAIN;
744         if (kvm->busy)
745                 goto out_unlock;
746
747         if (mem->slot >= kvm->nmemslots)
748                 kvm->nmemslots = mem->slot + 1;
749
750         *memslot = new;
751         ++kvm->memory_config_version;
752
753         kvm_mmu_slot_remove_write_access(kvm, mem->slot);
754         kvm_flush_remote_tlbs(kvm);
755
756         mutex_unlock(&kvm->lock);
757
758         kvm_free_physmem_slot(&old, &new);
759         return 0;
760
761 out_unlock:
762         mutex_unlock(&kvm->lock);
763 out_free:
764         kvm_free_physmem_slot(&new, &old);
765 out:
766         return r;
767 }
768
769 /*
770  * Get (and clear) the dirty memory log for a memory slot.
771  */
772 static int kvm_vm_ioctl_get_dirty_log(struct kvm *kvm,
773                                       struct kvm_dirty_log *log)
774 {
775         struct kvm_memory_slot *memslot;
776         int r, i;
777         int n;
778         unsigned long any = 0;
779
780         mutex_lock(&kvm->lock);
781
782         /*
783          * Prevent changes to guest memory configuration even while the lock
784          * is not taken.
785          */
786         ++kvm->busy;
787         mutex_unlock(&kvm->lock);
788         r = -EINVAL;
789         if (log->slot >= KVM_MEMORY_SLOTS)
790                 goto out;
791
792         memslot = &kvm->memslots[log->slot];
793         r = -ENOENT;
794         if (!memslot->dirty_bitmap)
795                 goto out;
796
797         n = ALIGN(memslot->npages, BITS_PER_LONG) / 8;
798
799         for (i = 0; !any && i < n/sizeof(long); ++i)
800                 any = memslot->dirty_bitmap[i];
801
802         r = -EFAULT;
803         if (copy_to_user(log->dirty_bitmap, memslot->dirty_bitmap, n))
804                 goto out;
805
806         /* If nothing is dirty, don't bother messing with page tables. */
807         if (any) {
808                 mutex_lock(&kvm->lock);
809                 kvm_mmu_slot_remove_write_access(kvm, log->slot);
810                 kvm_flush_remote_tlbs(kvm);
811                 memset(memslot->dirty_bitmap, 0, n);
812                 mutex_unlock(&kvm->lock);
813         }
814
815         r = 0;
816
817 out:
818         mutex_lock(&kvm->lock);
819         --kvm->busy;
820         mutex_unlock(&kvm->lock);
821         return r;
822 }
823
824 /*
825  * Set a new alias region.  Aliases map a portion of physical memory into
826  * another portion.  This is useful for memory windows, for example the PC
827  * VGA region.
828  */
829 static int kvm_vm_ioctl_set_memory_alias(struct kvm *kvm,
830                                          struct kvm_memory_alias *alias)
831 {
832         int r, n;
833         struct kvm_mem_alias *p;
834
835         r = -EINVAL;
836         /* General sanity checks */
837         if (alias->memory_size & (PAGE_SIZE - 1))
838                 goto out;
839         if (alias->guest_phys_addr & (PAGE_SIZE - 1))
840                 goto out;
841         if (alias->slot >= KVM_ALIAS_SLOTS)
842                 goto out;
843         if (alias->guest_phys_addr + alias->memory_size
844             < alias->guest_phys_addr)
845                 goto out;
846         if (alias->target_phys_addr + alias->memory_size
847             < alias->target_phys_addr)
848                 goto out;
849
850         mutex_lock(&kvm->lock);
851
852         p = &kvm->aliases[alias->slot];
853         p->base_gfn = alias->guest_phys_addr >> PAGE_SHIFT;
854         p->npages = alias->memory_size >> PAGE_SHIFT;
855         p->target_gfn = alias->target_phys_addr >> PAGE_SHIFT;
856
857         for (n = KVM_ALIAS_SLOTS; n > 0; --n)
858                 if (kvm->aliases[n - 1].npages)
859                         break;
860         kvm->naliases = n;
861
862         kvm_mmu_zap_all(kvm);
863
864         mutex_unlock(&kvm->lock);
865
866         return 0;
867
868 out:
869         return r;
870 }
871
872 static gfn_t unalias_gfn(struct kvm *kvm, gfn_t gfn)
873 {
874         int i;
875         struct kvm_mem_alias *alias;
876
877         for (i = 0; i < kvm->naliases; ++i) {
878                 alias = &kvm->aliases[i];
879                 if (gfn >= alias->base_gfn
880                     && gfn < alias->base_gfn + alias->npages)
881                         return alias->target_gfn + gfn - alias->base_gfn;
882         }
883         return gfn;
884 }
885
886 static struct kvm_memory_slot *__gfn_to_memslot(struct kvm *kvm, gfn_t gfn)
887 {
888         int i;
889
890         for (i = 0; i < kvm->nmemslots; ++i) {
891                 struct kvm_memory_slot *memslot = &kvm->memslots[i];
892
893                 if (gfn >= memslot->base_gfn
894                     && gfn < memslot->base_gfn + memslot->npages)
895                         return memslot;
896         }
897         return NULL;
898 }
899
900 struct kvm_memory_slot *gfn_to_memslot(struct kvm *kvm, gfn_t gfn)
901 {
902         gfn = unalias_gfn(kvm, gfn);
903         return __gfn_to_memslot(kvm, gfn);
904 }
905
906 struct page *gfn_to_page(struct kvm *kvm, gfn_t gfn)
907 {
908         struct kvm_memory_slot *slot;
909
910         gfn = unalias_gfn(kvm, gfn);
911         slot = __gfn_to_memslot(kvm, gfn);
912         if (!slot)
913                 return NULL;
914         return slot->phys_mem[gfn - slot->base_gfn];
915 }
916 EXPORT_SYMBOL_GPL(gfn_to_page);
917
918 void mark_page_dirty(struct kvm *kvm, gfn_t gfn)
919 {
920         int i;
921         struct kvm_memory_slot *memslot;
922         unsigned long rel_gfn;
923
924         for (i = 0; i < kvm->nmemslots; ++i) {
925                 memslot = &kvm->memslots[i];
926
927                 if (gfn >= memslot->base_gfn
928                     && gfn < memslot->base_gfn + memslot->npages) {
929
930                         if (!memslot->dirty_bitmap)
931                                 return;
932
933                         rel_gfn = gfn - memslot->base_gfn;
934
935                         /* avoid RMW */
936                         if (!test_bit(rel_gfn, memslot->dirty_bitmap))
937                                 set_bit(rel_gfn, memslot->dirty_bitmap);
938                         return;
939                 }
940         }
941 }
942
943 int emulator_read_std(unsigned long addr,
944                              void *val,
945                              unsigned int bytes,
946                              struct kvm_vcpu *vcpu)
947 {
948         void *data = val;
949
950         while (bytes) {
951                 gpa_t gpa = vcpu->mmu.gva_to_gpa(vcpu, addr);
952                 unsigned offset = addr & (PAGE_SIZE-1);
953                 unsigned tocopy = min(bytes, (unsigned)PAGE_SIZE - offset);
954                 unsigned long pfn;
955                 struct page *page;
956                 void *page_virt;
957
958                 if (gpa == UNMAPPED_GVA)
959                         return X86EMUL_PROPAGATE_FAULT;
960                 pfn = gpa >> PAGE_SHIFT;
961                 page = gfn_to_page(vcpu->kvm, pfn);
962                 if (!page)
963                         return X86EMUL_UNHANDLEABLE;
964                 page_virt = kmap_atomic(page, KM_USER0);
965
966                 memcpy(data, page_virt + offset, tocopy);
967
968                 kunmap_atomic(page_virt, KM_USER0);
969
970                 bytes -= tocopy;
971                 data += tocopy;
972                 addr += tocopy;
973         }
974
975         return X86EMUL_CONTINUE;
976 }
977 EXPORT_SYMBOL_GPL(emulator_read_std);
978
979 static int emulator_write_std(unsigned long addr,
980                               const void *val,
981                               unsigned int bytes,
982                               struct kvm_vcpu *vcpu)
983 {
984         printk(KERN_ERR "emulator_write_std: addr %lx n %d\n",
985                addr, bytes);
986         return X86EMUL_UNHANDLEABLE;
987 }
988
989 static struct kvm_io_device *vcpu_find_mmio_dev(struct kvm_vcpu *vcpu,
990                                                 gpa_t addr)
991 {
992         /*
993          * Note that its important to have this wrapper function because
994          * in the very near future we will be checking for MMIOs against
995          * the LAPIC as well as the general MMIO bus
996          */
997         return kvm_io_bus_find_dev(&vcpu->kvm->mmio_bus, addr);
998 }
999
1000 static struct kvm_io_device *vcpu_find_pio_dev(struct kvm_vcpu *vcpu,
1001                                                gpa_t addr)
1002 {
1003         return kvm_io_bus_find_dev(&vcpu->kvm->pio_bus, addr);
1004 }
1005
1006 static int emulator_read_emulated(unsigned long addr,
1007                                   void *val,
1008                                   unsigned int bytes,
1009                                   struct kvm_vcpu *vcpu)
1010 {
1011         struct kvm_io_device *mmio_dev;
1012         gpa_t                 gpa;
1013
1014         if (vcpu->mmio_read_completed) {
1015                 memcpy(val, vcpu->mmio_data, bytes);
1016                 vcpu->mmio_read_completed = 0;
1017                 return X86EMUL_CONTINUE;
1018         } else if (emulator_read_std(addr, val, bytes, vcpu)
1019                    == X86EMUL_CONTINUE)
1020                 return X86EMUL_CONTINUE;
1021
1022         gpa = vcpu->mmu.gva_to_gpa(vcpu, addr);
1023         if (gpa == UNMAPPED_GVA)
1024                 return X86EMUL_PROPAGATE_FAULT;
1025
1026         /*
1027          * Is this MMIO handled locally?
1028          */
1029         mmio_dev = vcpu_find_mmio_dev(vcpu, gpa);
1030         if (mmio_dev) {
1031                 kvm_iodevice_read(mmio_dev, gpa, bytes, val);
1032                 return X86EMUL_CONTINUE;
1033         }
1034
1035         vcpu->mmio_needed = 1;
1036         vcpu->mmio_phys_addr = gpa;
1037         vcpu->mmio_size = bytes;
1038         vcpu->mmio_is_write = 0;
1039
1040         return X86EMUL_UNHANDLEABLE;
1041 }
1042
1043 static int emulator_write_phys(struct kvm_vcpu *vcpu, gpa_t gpa,
1044                                const void *val, int bytes)
1045 {
1046         struct page *page;
1047         void *virt;
1048
1049         if (((gpa + bytes - 1) >> PAGE_SHIFT) != (gpa >> PAGE_SHIFT))
1050                 return 0;
1051         page = gfn_to_page(vcpu->kvm, gpa >> PAGE_SHIFT);
1052         if (!page)
1053                 return 0;
1054         mark_page_dirty(vcpu->kvm, gpa >> PAGE_SHIFT);
1055         virt = kmap_atomic(page, KM_USER0);
1056         kvm_mmu_pte_write(vcpu, gpa, val, bytes);
1057         memcpy(virt + offset_in_page(gpa), val, bytes);
1058         kunmap_atomic(virt, KM_USER0);
1059         return 1;
1060 }
1061
1062 static int emulator_write_emulated_onepage(unsigned long addr,
1063                                            const void *val,
1064                                            unsigned int bytes,
1065                                            struct kvm_vcpu *vcpu)
1066 {
1067         struct kvm_io_device *mmio_dev;
1068         gpa_t                 gpa = vcpu->mmu.gva_to_gpa(vcpu, addr);
1069
1070         if (gpa == UNMAPPED_GVA) {
1071                 kvm_arch_ops->inject_page_fault(vcpu, addr, 2);
1072                 return X86EMUL_PROPAGATE_FAULT;
1073         }
1074
1075         if (emulator_write_phys(vcpu, gpa, val, bytes))
1076                 return X86EMUL_CONTINUE;
1077
1078         /*
1079          * Is this MMIO handled locally?
1080          */
1081         mmio_dev = vcpu_find_mmio_dev(vcpu, gpa);
1082         if (mmio_dev) {
1083                 kvm_iodevice_write(mmio_dev, gpa, bytes, val);
1084                 return X86EMUL_CONTINUE;
1085         }
1086
1087         vcpu->mmio_needed = 1;
1088         vcpu->mmio_phys_addr = gpa;
1089         vcpu->mmio_size = bytes;
1090         vcpu->mmio_is_write = 1;
1091         memcpy(vcpu->mmio_data, val, bytes);
1092
1093         return X86EMUL_CONTINUE;
1094 }
1095
1096 int emulator_write_emulated(unsigned long addr,
1097                                    const void *val,
1098                                    unsigned int bytes,
1099                                    struct kvm_vcpu *vcpu)
1100 {
1101         /* Crossing a page boundary? */
1102         if (((addr + bytes - 1) ^ addr) & PAGE_MASK) {
1103                 int rc, now;
1104
1105                 now = -addr & ~PAGE_MASK;
1106                 rc = emulator_write_emulated_onepage(addr, val, now, vcpu);
1107                 if (rc != X86EMUL_CONTINUE)
1108                         return rc;
1109                 addr += now;
1110                 val += now;
1111                 bytes -= now;
1112         }
1113         return emulator_write_emulated_onepage(addr, val, bytes, vcpu);
1114 }
1115 EXPORT_SYMBOL_GPL(emulator_write_emulated);
1116
1117 static int emulator_cmpxchg_emulated(unsigned long addr,
1118                                      const void *old,
1119                                      const void *new,
1120                                      unsigned int bytes,
1121                                      struct kvm_vcpu *vcpu)
1122 {
1123         static int reported;
1124
1125         if (!reported) {
1126                 reported = 1;
1127                 printk(KERN_WARNING "kvm: emulating exchange as write\n");
1128         }
1129         return emulator_write_emulated(addr, new, bytes, vcpu);
1130 }
1131
1132 static unsigned long get_segment_base(struct kvm_vcpu *vcpu, int seg)
1133 {
1134         return kvm_arch_ops->get_segment_base(vcpu, seg);
1135 }
1136
1137 int emulate_invlpg(struct kvm_vcpu *vcpu, gva_t address)
1138 {
1139         return X86EMUL_CONTINUE;
1140 }
1141
1142 int emulate_clts(struct kvm_vcpu *vcpu)
1143 {
1144         unsigned long cr0;
1145
1146         cr0 = vcpu->cr0 & ~X86_CR0_TS;
1147         kvm_arch_ops->set_cr0(vcpu, cr0);
1148         return X86EMUL_CONTINUE;
1149 }
1150
1151 int emulator_get_dr(struct x86_emulate_ctxt* ctxt, int dr, unsigned long *dest)
1152 {
1153         struct kvm_vcpu *vcpu = ctxt->vcpu;
1154
1155         switch (dr) {
1156         case 0 ... 3:
1157                 *dest = kvm_arch_ops->get_dr(vcpu, dr);
1158                 return X86EMUL_CONTINUE;
1159         default:
1160                 printk(KERN_DEBUG "%s: unexpected dr %u\n",
1161                        __FUNCTION__, dr);
1162                 return X86EMUL_UNHANDLEABLE;
1163         }
1164 }
1165
1166 int emulator_set_dr(struct x86_emulate_ctxt *ctxt, int dr, unsigned long value)
1167 {
1168         unsigned long mask = (ctxt->mode == X86EMUL_MODE_PROT64) ? ~0ULL : ~0U;
1169         int exception;
1170
1171         kvm_arch_ops->set_dr(ctxt->vcpu, dr, value & mask, &exception);
1172         if (exception) {
1173                 /* FIXME: better handling */
1174                 return X86EMUL_UNHANDLEABLE;
1175         }
1176         return X86EMUL_CONTINUE;
1177 }
1178
1179 static void report_emulation_failure(struct x86_emulate_ctxt *ctxt)
1180 {
1181         static int reported;
1182         u8 opcodes[4];
1183         unsigned long rip = ctxt->vcpu->rip;
1184         unsigned long rip_linear;
1185
1186         rip_linear = rip + get_segment_base(ctxt->vcpu, VCPU_SREG_CS);
1187
1188         if (reported)
1189                 return;
1190
1191         emulator_read_std(rip_linear, (void *)opcodes, 4, ctxt->vcpu);
1192
1193         printk(KERN_ERR "emulation failed but !mmio_needed?"
1194                " rip %lx %02x %02x %02x %02x\n",
1195                rip, opcodes[0], opcodes[1], opcodes[2], opcodes[3]);
1196         reported = 1;
1197 }
1198
1199 struct x86_emulate_ops emulate_ops = {
1200         .read_std            = emulator_read_std,
1201         .write_std           = emulator_write_std,
1202         .read_emulated       = emulator_read_emulated,
1203         .write_emulated      = emulator_write_emulated,
1204         .cmpxchg_emulated    = emulator_cmpxchg_emulated,
1205 };
1206
1207 int emulate_instruction(struct kvm_vcpu *vcpu,
1208                         struct kvm_run *run,
1209                         unsigned long cr2,
1210                         u16 error_code)
1211 {
1212         struct x86_emulate_ctxt emulate_ctxt;
1213         int r;
1214         int cs_db, cs_l;
1215
1216         vcpu->mmio_fault_cr2 = cr2;
1217         kvm_arch_ops->cache_regs(vcpu);
1218
1219         kvm_arch_ops->get_cs_db_l_bits(vcpu, &cs_db, &cs_l);
1220
1221         emulate_ctxt.vcpu = vcpu;
1222         emulate_ctxt.eflags = kvm_arch_ops->get_rflags(vcpu);
1223         emulate_ctxt.cr2 = cr2;
1224         emulate_ctxt.mode = (emulate_ctxt.eflags & X86_EFLAGS_VM)
1225                 ? X86EMUL_MODE_REAL : cs_l
1226                 ? X86EMUL_MODE_PROT64 : cs_db
1227                 ? X86EMUL_MODE_PROT32 : X86EMUL_MODE_PROT16;
1228
1229         if (emulate_ctxt.mode == X86EMUL_MODE_PROT64) {
1230                 emulate_ctxt.cs_base = 0;
1231                 emulate_ctxt.ds_base = 0;
1232                 emulate_ctxt.es_base = 0;
1233                 emulate_ctxt.ss_base = 0;
1234         } else {
1235                 emulate_ctxt.cs_base = get_segment_base(vcpu, VCPU_SREG_CS);
1236                 emulate_ctxt.ds_base = get_segment_base(vcpu, VCPU_SREG_DS);
1237                 emulate_ctxt.es_base = get_segment_base(vcpu, VCPU_SREG_ES);
1238                 emulate_ctxt.ss_base = get_segment_base(vcpu, VCPU_SREG_SS);
1239         }
1240
1241         emulate_ctxt.gs_base = get_segment_base(vcpu, VCPU_SREG_GS);
1242         emulate_ctxt.fs_base = get_segment_base(vcpu, VCPU_SREG_FS);
1243
1244         vcpu->mmio_is_write = 0;
1245         r = x86_emulate_memop(&emulate_ctxt, &emulate_ops);
1246
1247         if ((r || vcpu->mmio_is_write) && run) {
1248                 run->exit_reason = KVM_EXIT_MMIO;
1249                 run->mmio.phys_addr = vcpu->mmio_phys_addr;
1250                 memcpy(run->mmio.data, vcpu->mmio_data, 8);
1251                 run->mmio.len = vcpu->mmio_size;
1252                 run->mmio.is_write = vcpu->mmio_is_write;
1253         }
1254
1255         if (r) {
1256                 if (kvm_mmu_unprotect_page_virt(vcpu, cr2))
1257                         return EMULATE_DONE;
1258                 if (!vcpu->mmio_needed) {
1259                         report_emulation_failure(&emulate_ctxt);
1260                         return EMULATE_FAIL;
1261                 }
1262                 return EMULATE_DO_MMIO;
1263         }
1264
1265         kvm_arch_ops->decache_regs(vcpu);
1266         kvm_arch_ops->set_rflags(vcpu, emulate_ctxt.eflags);
1267
1268         if (vcpu->mmio_is_write) {
1269                 vcpu->mmio_needed = 0;
1270                 return EMULATE_DO_MMIO;
1271         }
1272
1273         return EMULATE_DONE;
1274 }
1275 EXPORT_SYMBOL_GPL(emulate_instruction);
1276
1277 int kvm_emulate_halt(struct kvm_vcpu *vcpu)
1278 {
1279         if (vcpu->irq_summary)
1280                 return 1;
1281
1282         vcpu->run->exit_reason = KVM_EXIT_HLT;
1283         ++vcpu->stat.halt_exits;
1284         return 0;
1285 }
1286 EXPORT_SYMBOL_GPL(kvm_emulate_halt);
1287
1288 int kvm_hypercall(struct kvm_vcpu *vcpu, struct kvm_run *run)
1289 {
1290         unsigned long nr, a0, a1, a2, a3, a4, a5, ret;
1291
1292         kvm_arch_ops->cache_regs(vcpu);
1293         ret = -KVM_EINVAL;
1294 #ifdef CONFIG_X86_64
1295         if (is_long_mode(vcpu)) {
1296                 nr = vcpu->regs[VCPU_REGS_RAX];
1297                 a0 = vcpu->regs[VCPU_REGS_RDI];
1298                 a1 = vcpu->regs[VCPU_REGS_RSI];
1299                 a2 = vcpu->regs[VCPU_REGS_RDX];
1300                 a3 = vcpu->regs[VCPU_REGS_RCX];
1301                 a4 = vcpu->regs[VCPU_REGS_R8];
1302                 a5 = vcpu->regs[VCPU_REGS_R9];
1303         } else
1304 #endif
1305         {
1306                 nr = vcpu->regs[VCPU_REGS_RBX] & -1u;
1307                 a0 = vcpu->regs[VCPU_REGS_RAX] & -1u;
1308                 a1 = vcpu->regs[VCPU_REGS_RCX] & -1u;
1309                 a2 = vcpu->regs[VCPU_REGS_RDX] & -1u;
1310                 a3 = vcpu->regs[VCPU_REGS_RSI] & -1u;
1311                 a4 = vcpu->regs[VCPU_REGS_RDI] & -1u;
1312                 a5 = vcpu->regs[VCPU_REGS_RBP] & -1u;
1313         }
1314         switch (nr) {
1315         default:
1316                 run->hypercall.nr = nr;
1317                 run->hypercall.args[0] = a0;
1318                 run->hypercall.args[1] = a1;
1319                 run->hypercall.args[2] = a2;
1320                 run->hypercall.args[3] = a3;
1321                 run->hypercall.args[4] = a4;
1322                 run->hypercall.args[5] = a5;
1323                 run->hypercall.ret = ret;
1324                 run->hypercall.longmode = is_long_mode(vcpu);
1325                 kvm_arch_ops->decache_regs(vcpu);
1326                 return 0;
1327         }
1328         vcpu->regs[VCPU_REGS_RAX] = ret;
1329         kvm_arch_ops->decache_regs(vcpu);
1330         return 1;
1331 }
1332 EXPORT_SYMBOL_GPL(kvm_hypercall);
1333
1334 static u64 mk_cr_64(u64 curr_cr, u32 new_val)
1335 {
1336         return (curr_cr & ~((1ULL << 32) - 1)) | new_val;
1337 }
1338
1339 void realmode_lgdt(struct kvm_vcpu *vcpu, u16 limit, unsigned long base)
1340 {
1341         struct descriptor_table dt = { limit, base };
1342
1343         kvm_arch_ops->set_gdt(vcpu, &dt);
1344 }
1345
1346 void realmode_lidt(struct kvm_vcpu *vcpu, u16 limit, unsigned long base)
1347 {
1348         struct descriptor_table dt = { limit, base };
1349
1350         kvm_arch_ops->set_idt(vcpu, &dt);
1351 }
1352
1353 void realmode_lmsw(struct kvm_vcpu *vcpu, unsigned long msw,
1354                    unsigned long *rflags)
1355 {
1356         lmsw(vcpu, msw);
1357         *rflags = kvm_arch_ops->get_rflags(vcpu);
1358 }
1359
1360 unsigned long realmode_get_cr(struct kvm_vcpu *vcpu, int cr)
1361 {
1362         kvm_arch_ops->decache_cr4_guest_bits(vcpu);
1363         switch (cr) {
1364         case 0:
1365                 return vcpu->cr0;
1366         case 2:
1367                 return vcpu->cr2;
1368         case 3:
1369                 return vcpu->cr3;
1370         case 4:
1371                 return vcpu->cr4;
1372         default:
1373                 vcpu_printf(vcpu, "%s: unexpected cr %u\n", __FUNCTION__, cr);
1374                 return 0;
1375         }
1376 }
1377
1378 void realmode_set_cr(struct kvm_vcpu *vcpu, int cr, unsigned long val,
1379                      unsigned long *rflags)
1380 {
1381         switch (cr) {
1382         case 0:
1383                 set_cr0(vcpu, mk_cr_64(vcpu->cr0, val));
1384                 *rflags = kvm_arch_ops->get_rflags(vcpu);
1385                 break;
1386         case 2:
1387                 vcpu->cr2 = val;
1388                 break;
1389         case 3:
1390                 set_cr3(vcpu, val);
1391                 break;
1392         case 4:
1393                 set_cr4(vcpu, mk_cr_64(vcpu->cr4, val));
1394                 break;
1395         default:
1396                 vcpu_printf(vcpu, "%s: unexpected cr %u\n", __FUNCTION__, cr);
1397         }
1398 }
1399
1400 /*
1401  * Register the para guest with the host:
1402  */
1403 static int vcpu_register_para(struct kvm_vcpu *vcpu, gpa_t para_state_gpa)
1404 {
1405         struct kvm_vcpu_para_state *para_state;
1406         hpa_t para_state_hpa, hypercall_hpa;
1407         struct page *para_state_page;
1408         unsigned char *hypercall;
1409         gpa_t hypercall_gpa;
1410
1411         printk(KERN_DEBUG "kvm: guest trying to enter paravirtual mode\n");
1412         printk(KERN_DEBUG ".... para_state_gpa: %08Lx\n", para_state_gpa);
1413
1414         /*
1415          * Needs to be page aligned:
1416          */
1417         if (para_state_gpa != PAGE_ALIGN(para_state_gpa))
1418                 goto err_gp;
1419
1420         para_state_hpa = gpa_to_hpa(vcpu, para_state_gpa);
1421         printk(KERN_DEBUG ".... para_state_hpa: %08Lx\n", para_state_hpa);
1422         if (is_error_hpa(para_state_hpa))
1423                 goto err_gp;
1424
1425         mark_page_dirty(vcpu->kvm, para_state_gpa >> PAGE_SHIFT);
1426         para_state_page = pfn_to_page(para_state_hpa >> PAGE_SHIFT);
1427         para_state = kmap(para_state_page);
1428
1429         printk(KERN_DEBUG "....  guest version: %d\n", para_state->guest_version);
1430         printk(KERN_DEBUG "....           size: %d\n", para_state->size);
1431
1432         para_state->host_version = KVM_PARA_API_VERSION;
1433         /*
1434          * We cannot support guests that try to register themselves
1435          * with a newer API version than the host supports:
1436          */
1437         if (para_state->guest_version > KVM_PARA_API_VERSION) {
1438                 para_state->ret = -KVM_EINVAL;
1439                 goto err_kunmap_skip;
1440         }
1441
1442         hypercall_gpa = para_state->hypercall_gpa;
1443         hypercall_hpa = gpa_to_hpa(vcpu, hypercall_gpa);
1444         printk(KERN_DEBUG ".... hypercall_hpa: %08Lx\n", hypercall_hpa);
1445         if (is_error_hpa(hypercall_hpa)) {
1446                 para_state->ret = -KVM_EINVAL;
1447                 goto err_kunmap_skip;
1448         }
1449
1450         printk(KERN_DEBUG "kvm: para guest successfully registered.\n");
1451         vcpu->para_state_page = para_state_page;
1452         vcpu->para_state_gpa = para_state_gpa;
1453         vcpu->hypercall_gpa = hypercall_gpa;
1454
1455         mark_page_dirty(vcpu->kvm, hypercall_gpa >> PAGE_SHIFT);
1456         hypercall = kmap_atomic(pfn_to_page(hypercall_hpa >> PAGE_SHIFT),
1457                                 KM_USER1) + (hypercall_hpa & ~PAGE_MASK);
1458         kvm_arch_ops->patch_hypercall(vcpu, hypercall);
1459         kunmap_atomic(hypercall, KM_USER1);
1460
1461         para_state->ret = 0;
1462 err_kunmap_skip:
1463         kunmap(para_state_page);
1464         return 0;
1465 err_gp:
1466         return 1;
1467 }
1468
1469 int kvm_get_msr_common(struct kvm_vcpu *vcpu, u32 msr, u64 *pdata)
1470 {
1471         u64 data;
1472
1473         switch (msr) {
1474         case 0xc0010010: /* SYSCFG */
1475         case 0xc0010015: /* HWCR */
1476         case MSR_IA32_PLATFORM_ID:
1477         case MSR_IA32_P5_MC_ADDR:
1478         case MSR_IA32_P5_MC_TYPE:
1479         case MSR_IA32_MC0_CTL:
1480         case MSR_IA32_MCG_STATUS:
1481         case MSR_IA32_MCG_CAP:
1482         case MSR_IA32_MC0_MISC:
1483         case MSR_IA32_MC0_MISC+4:
1484         case MSR_IA32_MC0_MISC+8:
1485         case MSR_IA32_MC0_MISC+12:
1486         case MSR_IA32_MC0_MISC+16:
1487         case MSR_IA32_UCODE_REV:
1488         case MSR_IA32_PERF_STATUS:
1489         case MSR_IA32_EBL_CR_POWERON:
1490                 /* MTRR registers */
1491         case 0xfe:
1492         case 0x200 ... 0x2ff:
1493                 data = 0;
1494                 break;
1495         case 0xcd: /* fsb frequency */
1496                 data = 3;
1497                 break;
1498         case MSR_IA32_APICBASE:
1499                 data = vcpu->apic_base;
1500                 break;
1501         case MSR_IA32_MISC_ENABLE:
1502                 data = vcpu->ia32_misc_enable_msr;
1503                 break;
1504 #ifdef CONFIG_X86_64
1505         case MSR_EFER:
1506                 data = vcpu->shadow_efer;
1507                 break;
1508 #endif
1509         default:
1510                 printk(KERN_ERR "kvm: unhandled rdmsr: 0x%x\n", msr);
1511                 return 1;
1512         }
1513         *pdata = data;
1514         return 0;
1515 }
1516 EXPORT_SYMBOL_GPL(kvm_get_msr_common);
1517
1518 /*
1519  * Reads an msr value (of 'msr_index') into 'pdata'.
1520  * Returns 0 on success, non-0 otherwise.
1521  * Assumes vcpu_load() was already called.
1522  */
1523 int kvm_get_msr(struct kvm_vcpu *vcpu, u32 msr_index, u64 *pdata)
1524 {
1525         return kvm_arch_ops->get_msr(vcpu, msr_index, pdata);
1526 }
1527
1528 #ifdef CONFIG_X86_64
1529
1530 static void set_efer(struct kvm_vcpu *vcpu, u64 efer)
1531 {
1532         if (efer & EFER_RESERVED_BITS) {
1533                 printk(KERN_DEBUG "set_efer: 0x%llx #GP, reserved bits\n",
1534                        efer);
1535                 inject_gp(vcpu);
1536                 return;
1537         }
1538
1539         if (is_paging(vcpu)
1540             && (vcpu->shadow_efer & EFER_LME) != (efer & EFER_LME)) {
1541                 printk(KERN_DEBUG "set_efer: #GP, change LME while paging\n");
1542                 inject_gp(vcpu);
1543                 return;
1544         }
1545
1546         kvm_arch_ops->set_efer(vcpu, efer);
1547
1548         efer &= ~EFER_LMA;
1549         efer |= vcpu->shadow_efer & EFER_LMA;
1550
1551         vcpu->shadow_efer = efer;
1552 }
1553
1554 #endif
1555
1556 int kvm_set_msr_common(struct kvm_vcpu *vcpu, u32 msr, u64 data)
1557 {
1558         switch (msr) {
1559 #ifdef CONFIG_X86_64
1560         case MSR_EFER:
1561                 set_efer(vcpu, data);
1562                 break;
1563 #endif
1564         case MSR_IA32_MC0_STATUS:
1565                 printk(KERN_WARNING "%s: MSR_IA32_MC0_STATUS 0x%llx, nop\n",
1566                        __FUNCTION__, data);
1567                 break;
1568         case MSR_IA32_MCG_STATUS:
1569                 printk(KERN_WARNING "%s: MSR_IA32_MCG_STATUS 0x%llx, nop\n",
1570                         __FUNCTION__, data);
1571                 break;
1572         case MSR_IA32_UCODE_REV:
1573         case MSR_IA32_UCODE_WRITE:
1574         case 0x200 ... 0x2ff: /* MTRRs */
1575                 break;
1576         case MSR_IA32_APICBASE:
1577                 vcpu->apic_base = data;
1578                 break;
1579         case MSR_IA32_MISC_ENABLE:
1580                 vcpu->ia32_misc_enable_msr = data;
1581                 break;
1582         /*
1583          * This is the 'probe whether the host is KVM' logic:
1584          */
1585         case MSR_KVM_API_MAGIC:
1586                 return vcpu_register_para(vcpu, data);
1587
1588         default:
1589                 printk(KERN_ERR "kvm: unhandled wrmsr: 0x%x\n", msr);
1590                 return 1;
1591         }
1592         return 0;
1593 }
1594 EXPORT_SYMBOL_GPL(kvm_set_msr_common);
1595
1596 /*
1597  * Writes msr value into into the appropriate "register".
1598  * Returns 0 on success, non-0 otherwise.
1599  * Assumes vcpu_load() was already called.
1600  */
1601 int kvm_set_msr(struct kvm_vcpu *vcpu, u32 msr_index, u64 data)
1602 {
1603         return kvm_arch_ops->set_msr(vcpu, msr_index, data);
1604 }
1605
1606 void kvm_resched(struct kvm_vcpu *vcpu)
1607 {
1608         if (!need_resched())
1609                 return;
1610         cond_resched();
1611 }
1612 EXPORT_SYMBOL_GPL(kvm_resched);
1613
1614 void kvm_emulate_cpuid(struct kvm_vcpu *vcpu)
1615 {
1616         int i;
1617         u32 function;
1618         struct kvm_cpuid_entry *e, *best;
1619
1620         kvm_arch_ops->cache_regs(vcpu);
1621         function = vcpu->regs[VCPU_REGS_RAX];
1622         vcpu->regs[VCPU_REGS_RAX] = 0;
1623         vcpu->regs[VCPU_REGS_RBX] = 0;
1624         vcpu->regs[VCPU_REGS_RCX] = 0;
1625         vcpu->regs[VCPU_REGS_RDX] = 0;
1626         best = NULL;
1627         for (i = 0; i < vcpu->cpuid_nent; ++i) {
1628                 e = &vcpu->cpuid_entries[i];
1629                 if (e->function == function) {
1630                         best = e;
1631                         break;
1632                 }
1633                 /*
1634                  * Both basic or both extended?
1635                  */
1636                 if (((e->function ^ function) & 0x80000000) == 0)
1637                         if (!best || e->function > best->function)
1638                                 best = e;
1639         }
1640         if (best) {
1641                 vcpu->regs[VCPU_REGS_RAX] = best->eax;
1642                 vcpu->regs[VCPU_REGS_RBX] = best->ebx;
1643                 vcpu->regs[VCPU_REGS_RCX] = best->ecx;
1644                 vcpu->regs[VCPU_REGS_RDX] = best->edx;
1645         }
1646         kvm_arch_ops->decache_regs(vcpu);
1647         kvm_arch_ops->skip_emulated_instruction(vcpu);
1648 }
1649 EXPORT_SYMBOL_GPL(kvm_emulate_cpuid);
1650
1651 static int pio_copy_data(struct kvm_vcpu *vcpu)
1652 {
1653         void *p = vcpu->pio_data;
1654         void *q;
1655         unsigned bytes;
1656         int nr_pages = vcpu->pio.guest_pages[1] ? 2 : 1;
1657
1658         q = vmap(vcpu->pio.guest_pages, nr_pages, VM_READ|VM_WRITE,
1659                  PAGE_KERNEL);
1660         if (!q) {
1661                 free_pio_guest_pages(vcpu);
1662                 return -ENOMEM;
1663         }
1664         q += vcpu->pio.guest_page_offset;
1665         bytes = vcpu->pio.size * vcpu->pio.cur_count;
1666         if (vcpu->pio.in)
1667                 memcpy(q, p, bytes);
1668         else
1669                 memcpy(p, q, bytes);
1670         q -= vcpu->pio.guest_page_offset;
1671         vunmap(q);
1672         free_pio_guest_pages(vcpu);
1673         return 0;
1674 }
1675
1676 static int complete_pio(struct kvm_vcpu *vcpu)
1677 {
1678         struct kvm_pio_request *io = &vcpu->pio;
1679         long delta;
1680         int r;
1681
1682         kvm_arch_ops->cache_regs(vcpu);
1683
1684         if (!io->string) {
1685                 if (io->in)
1686                         memcpy(&vcpu->regs[VCPU_REGS_RAX], vcpu->pio_data,
1687                                io->size);
1688         } else {
1689                 if (io->in) {
1690                         r = pio_copy_data(vcpu);
1691                         if (r) {
1692                                 kvm_arch_ops->cache_regs(vcpu);
1693                                 return r;
1694                         }
1695                 }
1696
1697                 delta = 1;
1698                 if (io->rep) {
1699                         delta *= io->cur_count;
1700                         /*
1701                          * The size of the register should really depend on
1702                          * current address size.
1703                          */
1704                         vcpu->regs[VCPU_REGS_RCX] -= delta;
1705                 }
1706                 if (io->down)
1707                         delta = -delta;
1708                 delta *= io->size;
1709                 if (io->in)
1710                         vcpu->regs[VCPU_REGS_RDI] += delta;
1711                 else
1712                         vcpu->regs[VCPU_REGS_RSI] += delta;
1713         }
1714
1715         kvm_arch_ops->decache_regs(vcpu);
1716
1717         io->count -= io->cur_count;
1718         io->cur_count = 0;
1719
1720         if (!io->count)
1721                 kvm_arch_ops->skip_emulated_instruction(vcpu);
1722         return 0;
1723 }
1724
1725 static void kernel_pio(struct kvm_io_device *pio_dev,
1726                        struct kvm_vcpu *vcpu,
1727                        void *pd)
1728 {
1729         /* TODO: String I/O for in kernel device */
1730
1731         if (vcpu->pio.in)
1732                 kvm_iodevice_read(pio_dev, vcpu->pio.port,
1733                                   vcpu->pio.size,
1734                                   pd);
1735         else
1736                 kvm_iodevice_write(pio_dev, vcpu->pio.port,
1737                                    vcpu->pio.size,
1738                                    pd);
1739 }
1740
1741 static void pio_string_write(struct kvm_io_device *pio_dev,
1742                              struct kvm_vcpu *vcpu)
1743 {
1744         struct kvm_pio_request *io = &vcpu->pio;
1745         void *pd = vcpu->pio_data;
1746         int i;
1747
1748         for (i = 0; i < io->cur_count; i++) {
1749                 kvm_iodevice_write(pio_dev, io->port,
1750                                    io->size,
1751                                    pd);
1752                 pd += io->size;
1753         }
1754 }
1755
1756 int kvm_setup_pio(struct kvm_vcpu *vcpu, struct kvm_run *run, int in,
1757                   int size, unsigned long count, int string, int down,
1758                   gva_t address, int rep, unsigned port)
1759 {
1760         unsigned now, in_page;
1761         int i, ret = 0;
1762         int nr_pages = 1;
1763         struct page *page;
1764         struct kvm_io_device *pio_dev;
1765
1766         vcpu->run->exit_reason = KVM_EXIT_IO;
1767         vcpu->run->io.direction = in ? KVM_EXIT_IO_IN : KVM_EXIT_IO_OUT;
1768         vcpu->run->io.size = size;
1769         vcpu->run->io.data_offset = KVM_PIO_PAGE_OFFSET * PAGE_SIZE;
1770         vcpu->run->io.count = count;
1771         vcpu->run->io.port = port;
1772         vcpu->pio.count = count;
1773         vcpu->pio.cur_count = count;
1774         vcpu->pio.size = size;
1775         vcpu->pio.in = in;
1776         vcpu->pio.port = port;
1777         vcpu->pio.string = string;
1778         vcpu->pio.down = down;
1779         vcpu->pio.guest_page_offset = offset_in_page(address);
1780         vcpu->pio.rep = rep;
1781
1782         pio_dev = vcpu_find_pio_dev(vcpu, port);
1783         if (!string) {
1784                 kvm_arch_ops->cache_regs(vcpu);
1785                 memcpy(vcpu->pio_data, &vcpu->regs[VCPU_REGS_RAX], 4);
1786                 kvm_arch_ops->decache_regs(vcpu);
1787                 if (pio_dev) {
1788                         kernel_pio(pio_dev, vcpu, vcpu->pio_data);
1789                         complete_pio(vcpu);
1790                         return 1;
1791                 }
1792                 return 0;
1793         }
1794
1795         if (!count) {
1796                 kvm_arch_ops->skip_emulated_instruction(vcpu);
1797                 return 1;
1798         }
1799
1800         now = min(count, PAGE_SIZE / size);
1801
1802         if (!down)
1803                 in_page = PAGE_SIZE - offset_in_page(address);
1804         else
1805                 in_page = offset_in_page(address) + size;
1806         now = min(count, (unsigned long)in_page / size);
1807         if (!now) {
1808                 /*
1809                  * String I/O straddles page boundary.  Pin two guest pages
1810                  * so that we satisfy atomicity constraints.  Do just one
1811                  * transaction to avoid complexity.
1812                  */
1813                 nr_pages = 2;
1814                 now = 1;
1815         }
1816         if (down) {
1817                 /*
1818                  * String I/O in reverse.  Yuck.  Kill the guest, fix later.
1819                  */
1820                 printk(KERN_ERR "kvm: guest string pio down\n");
1821                 inject_gp(vcpu);
1822                 return 1;
1823         }
1824         vcpu->run->io.count = now;
1825         vcpu->pio.cur_count = now;
1826
1827         for (i = 0; i < nr_pages; ++i) {
1828                 mutex_lock(&vcpu->kvm->lock);
1829                 page = gva_to_page(vcpu, address + i * PAGE_SIZE);
1830                 if (page)
1831                         get_page(page);
1832                 vcpu->pio.guest_pages[i] = page;
1833                 mutex_unlock(&vcpu->kvm->lock);
1834                 if (!page) {
1835                         inject_gp(vcpu);
1836                         free_pio_guest_pages(vcpu);
1837                         return 1;
1838                 }
1839         }
1840
1841         if (!vcpu->pio.in) {
1842                 /* string PIO write */
1843                 ret = pio_copy_data(vcpu);
1844                 if (ret >= 0 && pio_dev) {
1845                         pio_string_write(pio_dev, vcpu);
1846                         complete_pio(vcpu);
1847                         if (vcpu->pio.count == 0)
1848                                 ret = 1;
1849                 }
1850         } else if (pio_dev)
1851                 printk(KERN_ERR "no string pio read support yet, "
1852                        "port %x size %d count %ld\n",
1853                         port, size, count);
1854
1855         return ret;
1856 }
1857 EXPORT_SYMBOL_GPL(kvm_setup_pio);
1858
1859 static int kvm_vcpu_ioctl_run(struct kvm_vcpu *vcpu, struct kvm_run *kvm_run)
1860 {
1861         int r;
1862         sigset_t sigsaved;
1863
1864         vcpu_load(vcpu);
1865
1866         if (vcpu->sigset_active)
1867                 sigprocmask(SIG_SETMASK, &vcpu->sigset, &sigsaved);
1868
1869         /* re-sync apic's tpr */
1870         vcpu->cr8 = kvm_run->cr8;
1871
1872         if (vcpu->pio.cur_count) {
1873                 r = complete_pio(vcpu);
1874                 if (r)
1875                         goto out;
1876         }
1877
1878         if (vcpu->mmio_needed) {
1879                 memcpy(vcpu->mmio_data, kvm_run->mmio.data, 8);
1880                 vcpu->mmio_read_completed = 1;
1881                 vcpu->mmio_needed = 0;
1882                 r = emulate_instruction(vcpu, kvm_run,
1883                                         vcpu->mmio_fault_cr2, 0);
1884                 if (r == EMULATE_DO_MMIO) {
1885                         /*
1886                          * Read-modify-write.  Back to userspace.
1887                          */
1888                         r = 0;
1889                         goto out;
1890                 }
1891         }
1892
1893         if (kvm_run->exit_reason == KVM_EXIT_HYPERCALL) {
1894                 kvm_arch_ops->cache_regs(vcpu);
1895                 vcpu->regs[VCPU_REGS_RAX] = kvm_run->hypercall.ret;
1896                 kvm_arch_ops->decache_regs(vcpu);
1897         }
1898
1899         r = kvm_arch_ops->run(vcpu, kvm_run);
1900
1901 out:
1902         if (vcpu->sigset_active)
1903                 sigprocmask(SIG_SETMASK, &sigsaved, NULL);
1904
1905         vcpu_put(vcpu);
1906         return r;
1907 }
1908
1909 static int kvm_vcpu_ioctl_get_regs(struct kvm_vcpu *vcpu,
1910                                    struct kvm_regs *regs)
1911 {
1912         vcpu_load(vcpu);
1913
1914         kvm_arch_ops->cache_regs(vcpu);
1915
1916         regs->rax = vcpu->regs[VCPU_REGS_RAX];
1917         regs->rbx = vcpu->regs[VCPU_REGS_RBX];
1918         regs->rcx = vcpu->regs[VCPU_REGS_RCX];
1919         regs->rdx = vcpu->regs[VCPU_REGS_RDX];
1920         regs->rsi = vcpu->regs[VCPU_REGS_RSI];
1921         regs->rdi = vcpu->regs[VCPU_REGS_RDI];
1922         regs->rsp = vcpu->regs[VCPU_REGS_RSP];
1923         regs->rbp = vcpu->regs[VCPU_REGS_RBP];
1924 #ifdef CONFIG_X86_64
1925         regs->r8 = vcpu->regs[VCPU_REGS_R8];
1926         regs->r9 = vcpu->regs[VCPU_REGS_R9];
1927         regs->r10 = vcpu->regs[VCPU_REGS_R10];
1928         regs->r11 = vcpu->regs[VCPU_REGS_R11];
1929         regs->r12 = vcpu->regs[VCPU_REGS_R12];
1930         regs->r13 = vcpu->regs[VCPU_REGS_R13];
1931         regs->r14 = vcpu->regs[VCPU_REGS_R14];
1932         regs->r15 = vcpu->regs[VCPU_REGS_R15];
1933 #endif
1934
1935         regs->rip = vcpu->rip;
1936         regs->rflags = kvm_arch_ops->get_rflags(vcpu);
1937
1938         /*
1939          * Don't leak debug flags in case they were set for guest debugging
1940          */
1941         if (vcpu->guest_debug.enabled && vcpu->guest_debug.singlestep)
1942                 regs->rflags &= ~(X86_EFLAGS_TF | X86_EFLAGS_RF);
1943
1944         vcpu_put(vcpu);
1945
1946         return 0;
1947 }
1948
1949 static int kvm_vcpu_ioctl_set_regs(struct kvm_vcpu *vcpu,
1950                                    struct kvm_regs *regs)
1951 {
1952         vcpu_load(vcpu);
1953
1954         vcpu->regs[VCPU_REGS_RAX] = regs->rax;
1955         vcpu->regs[VCPU_REGS_RBX] = regs->rbx;
1956         vcpu->regs[VCPU_REGS_RCX] = regs->rcx;
1957         vcpu->regs[VCPU_REGS_RDX] = regs->rdx;
1958         vcpu->regs[VCPU_REGS_RSI] = regs->rsi;
1959         vcpu->regs[VCPU_REGS_RDI] = regs->rdi;
1960         vcpu->regs[VCPU_REGS_RSP] = regs->rsp;
1961         vcpu->regs[VCPU_REGS_RBP] = regs->rbp;
1962 #ifdef CONFIG_X86_64
1963         vcpu->regs[VCPU_REGS_R8] = regs->r8;
1964         vcpu->regs[VCPU_REGS_R9] = regs->r9;
1965         vcpu->regs[VCPU_REGS_R10] = regs->r10;
1966         vcpu->regs[VCPU_REGS_R11] = regs->r11;
1967         vcpu->regs[VCPU_REGS_R12] = regs->r12;
1968         vcpu->regs[VCPU_REGS_R13] = regs->r13;
1969         vcpu->regs[VCPU_REGS_R14] = regs->r14;
1970         vcpu->regs[VCPU_REGS_R15] = regs->r15;
1971 #endif
1972
1973         vcpu->rip = regs->rip;
1974         kvm_arch_ops->set_rflags(vcpu, regs->rflags);
1975
1976         kvm_arch_ops->decache_regs(vcpu);
1977
1978         vcpu_put(vcpu);
1979
1980         return 0;
1981 }
1982
1983 static void get_segment(struct kvm_vcpu *vcpu,
1984                         struct kvm_segment *var, int seg)
1985 {
1986         return kvm_arch_ops->get_segment(vcpu, var, seg);
1987 }
1988
1989 static int kvm_vcpu_ioctl_get_sregs(struct kvm_vcpu *vcpu,
1990                                     struct kvm_sregs *sregs)
1991 {
1992         struct descriptor_table dt;
1993
1994         vcpu_load(vcpu);
1995
1996         get_segment(vcpu, &sregs->cs, VCPU_SREG_CS);
1997         get_segment(vcpu, &sregs->ds, VCPU_SREG_DS);
1998         get_segment(vcpu, &sregs->es, VCPU_SREG_ES);
1999         get_segment(vcpu, &sregs->fs, VCPU_SREG_FS);
2000         get_segment(vcpu, &sregs->gs, VCPU_SREG_GS);
2001         get_segment(vcpu, &sregs->ss, VCPU_SREG_SS);
2002
2003         get_segment(vcpu, &sregs->tr, VCPU_SREG_TR);
2004         get_segment(vcpu, &sregs->ldt, VCPU_SREG_LDTR);
2005
2006         kvm_arch_ops->get_idt(vcpu, &dt);
2007         sregs->idt.limit = dt.limit;
2008         sregs->idt.base = dt.base;
2009         kvm_arch_ops->get_gdt(vcpu, &dt);
2010         sregs->gdt.limit = dt.limit;
2011         sregs->gdt.base = dt.base;
2012
2013         kvm_arch_ops->decache_cr4_guest_bits(vcpu);
2014         sregs->cr0 = vcpu->cr0;
2015         sregs->cr2 = vcpu->cr2;
2016         sregs->cr3 = vcpu->cr3;
2017         sregs->cr4 = vcpu->cr4;
2018         sregs->cr8 = vcpu->cr8;
2019         sregs->efer = vcpu->shadow_efer;
2020         sregs->apic_base = vcpu->apic_base;
2021
2022         memcpy(sregs->interrupt_bitmap, vcpu->irq_pending,
2023                sizeof sregs->interrupt_bitmap);
2024
2025         vcpu_put(vcpu);
2026
2027         return 0;
2028 }
2029
2030 static void set_segment(struct kvm_vcpu *vcpu,
2031                         struct kvm_segment *var, int seg)
2032 {
2033         return kvm_arch_ops->set_segment(vcpu, var, seg);
2034 }
2035
2036 static int kvm_vcpu_ioctl_set_sregs(struct kvm_vcpu *vcpu,
2037                                     struct kvm_sregs *sregs)
2038 {
2039         int mmu_reset_needed = 0;
2040         int i;
2041         struct descriptor_table dt;
2042
2043         vcpu_load(vcpu);
2044
2045         dt.limit = sregs->idt.limit;
2046         dt.base = sregs->idt.base;
2047         kvm_arch_ops->set_idt(vcpu, &dt);
2048         dt.limit = sregs->gdt.limit;
2049         dt.base = sregs->gdt.base;
2050         kvm_arch_ops->set_gdt(vcpu, &dt);
2051
2052         vcpu->cr2 = sregs->cr2;
2053         mmu_reset_needed |= vcpu->cr3 != sregs->cr3;
2054         vcpu->cr3 = sregs->cr3;
2055
2056         vcpu->cr8 = sregs->cr8;
2057
2058         mmu_reset_needed |= vcpu->shadow_efer != sregs->efer;
2059 #ifdef CONFIG_X86_64
2060         kvm_arch_ops->set_efer(vcpu, sregs->efer);
2061 #endif
2062         vcpu->apic_base = sregs->apic_base;
2063
2064         kvm_arch_ops->decache_cr4_guest_bits(vcpu);
2065
2066         mmu_reset_needed |= vcpu->cr0 != sregs->cr0;
2067         kvm_arch_ops->set_cr0(vcpu, sregs->cr0);
2068
2069         mmu_reset_needed |= vcpu->cr4 != sregs->cr4;
2070         kvm_arch_ops->set_cr4(vcpu, sregs->cr4);
2071         if (!is_long_mode(vcpu) && is_pae(vcpu))
2072                 load_pdptrs(vcpu, vcpu->cr3);
2073
2074         if (mmu_reset_needed)
2075                 kvm_mmu_reset_context(vcpu);
2076
2077         memcpy(vcpu->irq_pending, sregs->interrupt_bitmap,
2078                sizeof vcpu->irq_pending);
2079         vcpu->irq_summary = 0;
2080         for (i = 0; i < ARRAY_SIZE(vcpu->irq_pending); ++i)
2081                 if (vcpu->irq_pending[i])
2082                         __set_bit(i, &vcpu->irq_summary);
2083
2084         set_segment(vcpu, &sregs->cs, VCPU_SREG_CS);
2085         set_segment(vcpu, &sregs->ds, VCPU_SREG_DS);
2086         set_segment(vcpu, &sregs->es, VCPU_SREG_ES);
2087         set_segment(vcpu, &sregs->fs, VCPU_SREG_FS);
2088         set_segment(vcpu, &sregs->gs, VCPU_SREG_GS);
2089         set_segment(vcpu, &sregs->ss, VCPU_SREG_SS);
2090
2091         set_segment(vcpu, &sregs->tr, VCPU_SREG_TR);
2092         set_segment(vcpu, &sregs->ldt, VCPU_SREG_LDTR);
2093
2094         vcpu_put(vcpu);
2095
2096         return 0;
2097 }
2098
2099 /*
2100  * List of msr numbers which we expose to userspace through KVM_GET_MSRS
2101  * and KVM_SET_MSRS, and KVM_GET_MSR_INDEX_LIST.
2102  *
2103  * This list is modified at module load time to reflect the
2104  * capabilities of the host cpu.
2105  */
2106 static u32 msrs_to_save[] = {
2107         MSR_IA32_SYSENTER_CS, MSR_IA32_SYSENTER_ESP, MSR_IA32_SYSENTER_EIP,
2108         MSR_K6_STAR,
2109 #ifdef CONFIG_X86_64
2110         MSR_CSTAR, MSR_KERNEL_GS_BASE, MSR_SYSCALL_MASK, MSR_LSTAR,
2111 #endif
2112         MSR_IA32_TIME_STAMP_COUNTER,
2113 };
2114
2115 static unsigned num_msrs_to_save;
2116
2117 static u32 emulated_msrs[] = {
2118         MSR_IA32_MISC_ENABLE,
2119 };
2120
2121 static __init void kvm_init_msr_list(void)
2122 {
2123         u32 dummy[2];
2124         unsigned i, j;
2125
2126         for (i = j = 0; i < ARRAY_SIZE(msrs_to_save); i++) {
2127                 if (rdmsr_safe(msrs_to_save[i], &dummy[0], &dummy[1]) < 0)
2128                         continue;
2129                 if (j < i)
2130                         msrs_to_save[j] = msrs_to_save[i];
2131                 j++;
2132         }
2133         num_msrs_to_save = j;
2134 }
2135
2136 /*
2137  * Adapt set_msr() to msr_io()'s calling convention
2138  */
2139 static int do_set_msr(struct kvm_vcpu *vcpu, unsigned index, u64 *data)
2140 {
2141         return kvm_set_msr(vcpu, index, *data);
2142 }
2143
2144 /*
2145  * Read or write a bunch of msrs. All parameters are kernel addresses.
2146  *
2147  * @return number of msrs set successfully.
2148  */
2149 static int __msr_io(struct kvm_vcpu *vcpu, struct kvm_msrs *msrs,
2150                     struct kvm_msr_entry *entries,
2151                     int (*do_msr)(struct kvm_vcpu *vcpu,
2152                                   unsigned index, u64 *data))
2153 {
2154         int i;
2155
2156         vcpu_load(vcpu);
2157
2158         for (i = 0; i < msrs->nmsrs; ++i)
2159                 if (do_msr(vcpu, entries[i].index, &entries[i].data))
2160                         break;
2161
2162         vcpu_put(vcpu);
2163
2164         return i;
2165 }
2166
2167 /*
2168  * Read or write a bunch of msrs. Parameters are user addresses.
2169  *
2170  * @return number of msrs set successfully.
2171  */
2172 static int msr_io(struct kvm_vcpu *vcpu, struct kvm_msrs __user *user_msrs,
2173                   int (*do_msr)(struct kvm_vcpu *vcpu,
2174                                 unsigned index, u64 *data),
2175                   int writeback)
2176 {
2177         struct kvm_msrs msrs;
2178         struct kvm_msr_entry *entries;
2179         int r, n;
2180         unsigned size;
2181
2182         r = -EFAULT;
2183         if (copy_from_user(&msrs, user_msrs, sizeof msrs))
2184                 goto out;
2185
2186         r = -E2BIG;
2187         if (msrs.nmsrs >= MAX_IO_MSRS)
2188                 goto out;
2189
2190         r = -ENOMEM;
2191         size = sizeof(struct kvm_msr_entry) * msrs.nmsrs;
2192         entries = vmalloc(size);
2193         if (!entries)
2194                 goto out;
2195
2196         r = -EFAULT;
2197         if (copy_from_user(entries, user_msrs->entries, size))
2198                 goto out_free;
2199
2200         r = n = __msr_io(vcpu, &msrs, entries, do_msr);
2201         if (r < 0)
2202                 goto out_free;
2203
2204         r = -EFAULT;
2205         if (writeback && copy_to_user(user_msrs->entries, entries, size))
2206                 goto out_free;
2207
2208         r = n;
2209
2210 out_free:
2211         vfree(entries);
2212 out:
2213         return r;
2214 }
2215
2216 /*
2217  * Translate a guest virtual address to a guest physical address.
2218  */
2219 static int kvm_vcpu_ioctl_translate(struct kvm_vcpu *vcpu,
2220                                     struct kvm_translation *tr)
2221 {
2222         unsigned long vaddr = tr->linear_address;
2223         gpa_t gpa;
2224
2225         vcpu_load(vcpu);
2226         mutex_lock(&vcpu->kvm->lock);
2227         gpa = vcpu->mmu.gva_to_gpa(vcpu, vaddr);
2228         tr->physical_address = gpa;
2229         tr->valid = gpa != UNMAPPED_GVA;
2230         tr->writeable = 1;
2231         tr->usermode = 0;
2232         mutex_unlock(&vcpu->kvm->lock);
2233         vcpu_put(vcpu);
2234
2235         return 0;
2236 }
2237
2238 static int kvm_vcpu_ioctl_interrupt(struct kvm_vcpu *vcpu,
2239                                     struct kvm_interrupt *irq)
2240 {
2241         if (irq->irq < 0 || irq->irq >= 256)
2242                 return -EINVAL;
2243         vcpu_load(vcpu);
2244
2245         set_bit(irq->irq, vcpu->irq_pending);
2246         set_bit(irq->irq / BITS_PER_LONG, &vcpu->irq_summary);
2247
2248         vcpu_put(vcpu);
2249
2250         return 0;
2251 }
2252
2253 static int kvm_vcpu_ioctl_debug_guest(struct kvm_vcpu *vcpu,
2254                                       struct kvm_debug_guest *dbg)
2255 {
2256         int r;
2257
2258         vcpu_load(vcpu);
2259
2260         r = kvm_arch_ops->set_guest_debug(vcpu, dbg);
2261
2262         vcpu_put(vcpu);
2263
2264         return r;
2265 }
2266
2267 static struct page *kvm_vcpu_nopage(struct vm_area_struct *vma,
2268                                     unsigned long address,
2269                                     int *type)
2270 {
2271         struct kvm_vcpu *vcpu = vma->vm_file->private_data;
2272         unsigned long pgoff;
2273         struct page *page;
2274
2275         pgoff = ((address - vma->vm_start) >> PAGE_SHIFT) + vma->vm_pgoff;
2276         if (pgoff == 0)
2277                 page = virt_to_page(vcpu->run);
2278         else if (pgoff == KVM_PIO_PAGE_OFFSET)
2279                 page = virt_to_page(vcpu->pio_data);
2280         else
2281                 return NOPAGE_SIGBUS;
2282         get_page(page);
2283         if (type != NULL)
2284                 *type = VM_FAULT_MINOR;
2285
2286         return page;
2287 }
2288
2289 static struct vm_operations_struct kvm_vcpu_vm_ops = {
2290         .nopage = kvm_vcpu_nopage,
2291 };
2292
2293 static int kvm_vcpu_mmap(struct file *file, struct vm_area_struct *vma)
2294 {
2295         vma->vm_ops = &kvm_vcpu_vm_ops;
2296         return 0;
2297 }
2298
2299 static int kvm_vcpu_release(struct inode *inode, struct file *filp)
2300 {
2301         struct kvm_vcpu *vcpu = filp->private_data;
2302
2303         fput(vcpu->kvm->filp);
2304         return 0;
2305 }
2306
2307 static struct file_operations kvm_vcpu_fops = {
2308         .release        = kvm_vcpu_release,
2309         .unlocked_ioctl = kvm_vcpu_ioctl,
2310         .compat_ioctl   = kvm_vcpu_ioctl,
2311         .mmap           = kvm_vcpu_mmap,
2312 };
2313
2314 /*
2315  * Allocates an inode for the vcpu.
2316  */
2317 static int create_vcpu_fd(struct kvm_vcpu *vcpu)
2318 {
2319         int fd, r;
2320         struct inode *inode;
2321         struct file *file;
2322
2323         r = anon_inode_getfd(&fd, &inode, &file,
2324                              "kvm-vcpu", &kvm_vcpu_fops, vcpu);
2325         if (r)
2326                 return r;
2327         atomic_inc(&vcpu->kvm->filp->f_count);
2328         return fd;
2329 }
2330
2331 /*
2332  * Creates some virtual cpus.  Good luck creating more than one.
2333  */
2334 static int kvm_vm_ioctl_create_vcpu(struct kvm *kvm, int n)
2335 {
2336         int r;
2337         struct kvm_vcpu *vcpu;
2338
2339         if (!valid_vcpu(n))
2340                 return -EINVAL;
2341
2342         vcpu = kvm_arch_ops->vcpu_create(kvm, n);
2343         if (IS_ERR(vcpu))
2344                 return PTR_ERR(vcpu);
2345
2346         preempt_notifier_init(&vcpu->preempt_notifier, &kvm_preempt_ops);
2347
2348         /* We do fxsave: this must be aligned. */
2349         BUG_ON((unsigned long)&vcpu->host_fx_image & 0xF);
2350
2351         vcpu_load(vcpu);
2352         r = kvm_mmu_setup(vcpu);
2353         vcpu_put(vcpu);
2354         if (r < 0)
2355                 goto free_vcpu;
2356
2357         mutex_lock(&kvm->lock);
2358         if (kvm->vcpus[n]) {
2359                 r = -EEXIST;
2360                 mutex_unlock(&kvm->lock);
2361                 goto mmu_unload;
2362         }
2363         kvm->vcpus[n] = vcpu;
2364         mutex_unlock(&kvm->lock);
2365
2366         /* Now it's all set up, let userspace reach it */
2367         r = create_vcpu_fd(vcpu);
2368         if (r < 0)
2369                 goto unlink;
2370         return r;
2371
2372 unlink:
2373         mutex_lock(&kvm->lock);
2374         kvm->vcpus[n] = NULL;
2375         mutex_unlock(&kvm->lock);
2376
2377 mmu_unload:
2378         vcpu_load(vcpu);
2379         kvm_mmu_unload(vcpu);
2380         vcpu_put(vcpu);
2381
2382 free_vcpu:
2383         kvm_arch_ops->vcpu_free(vcpu);
2384         return r;
2385 }
2386
2387 static void cpuid_fix_nx_cap(struct kvm_vcpu *vcpu)
2388 {
2389         u64 efer;
2390         int i;
2391         struct kvm_cpuid_entry *e, *entry;
2392
2393         rdmsrl(MSR_EFER, efer);
2394         entry = NULL;
2395         for (i = 0; i < vcpu->cpuid_nent; ++i) {
2396                 e = &vcpu->cpuid_entries[i];
2397                 if (e->function == 0x80000001) {
2398                         entry = e;
2399                         break;
2400                 }
2401         }
2402         if (entry && (entry->edx & (1 << 20)) && !(efer & EFER_NX)) {
2403                 entry->edx &= ~(1 << 20);
2404                 printk(KERN_INFO "kvm: guest NX capability removed\n");
2405         }
2406 }
2407
2408 static int kvm_vcpu_ioctl_set_cpuid(struct kvm_vcpu *vcpu,
2409                                     struct kvm_cpuid *cpuid,
2410                                     struct kvm_cpuid_entry __user *entries)
2411 {
2412         int r;
2413
2414         r = -E2BIG;
2415         if (cpuid->nent > KVM_MAX_CPUID_ENTRIES)
2416                 goto out;
2417         r = -EFAULT;
2418         if (copy_from_user(&vcpu->cpuid_entries, entries,
2419                            cpuid->nent * sizeof(struct kvm_cpuid_entry)))
2420                 goto out;
2421         vcpu->cpuid_nent = cpuid->nent;
2422         cpuid_fix_nx_cap(vcpu);
2423         return 0;
2424
2425 out:
2426         return r;
2427 }
2428
2429 static int kvm_vcpu_ioctl_set_sigmask(struct kvm_vcpu *vcpu, sigset_t *sigset)
2430 {
2431         if (sigset) {
2432                 sigdelsetmask(sigset, sigmask(SIGKILL)|sigmask(SIGSTOP));
2433                 vcpu->sigset_active = 1;
2434                 vcpu->sigset = *sigset;
2435         } else
2436                 vcpu->sigset_active = 0;
2437         return 0;
2438 }
2439
2440 /*
2441  * fxsave fpu state.  Taken from x86_64/processor.h.  To be killed when
2442  * we have asm/x86/processor.h
2443  */
2444 struct fxsave {
2445         u16     cwd;
2446         u16     swd;
2447         u16     twd;
2448         u16     fop;
2449         u64     rip;
2450         u64     rdp;
2451         u32     mxcsr;
2452         u32     mxcsr_mask;
2453         u32     st_space[32];   /* 8*16 bytes for each FP-reg = 128 bytes */
2454 #ifdef CONFIG_X86_64
2455         u32     xmm_space[64];  /* 16*16 bytes for each XMM-reg = 256 bytes */
2456 #else
2457         u32     xmm_space[32];  /* 8*16 bytes for each XMM-reg = 128 bytes */
2458 #endif
2459 };
2460
2461 static int kvm_vcpu_ioctl_get_fpu(struct kvm_vcpu *vcpu, struct kvm_fpu *fpu)
2462 {
2463         struct fxsave *fxsave = (struct fxsave *)&vcpu->guest_fx_image;
2464
2465         vcpu_load(vcpu);
2466
2467         memcpy(fpu->fpr, fxsave->st_space, 128);
2468         fpu->fcw = fxsave->cwd;
2469         fpu->fsw = fxsave->swd;
2470         fpu->ftwx = fxsave->twd;
2471         fpu->last_opcode = fxsave->fop;
2472         fpu->last_ip = fxsave->rip;
2473         fpu->last_dp = fxsave->rdp;
2474         memcpy(fpu->xmm, fxsave->xmm_space, sizeof fxsave->xmm_space);
2475
2476         vcpu_put(vcpu);
2477
2478         return 0;
2479 }
2480
2481 static int kvm_vcpu_ioctl_set_fpu(struct kvm_vcpu *vcpu, struct kvm_fpu *fpu)
2482 {
2483         struct fxsave *fxsave = (struct fxsave *)&vcpu->guest_fx_image;
2484
2485         vcpu_load(vcpu);
2486
2487         memcpy(fxsave->st_space, fpu->fpr, 128);
2488         fxsave->cwd = fpu->fcw;
2489         fxsave->swd = fpu->fsw;
2490         fxsave->twd = fpu->ftwx;
2491         fxsave->fop = fpu->last_opcode;
2492         fxsave->rip = fpu->last_ip;
2493         fxsave->rdp = fpu->last_dp;
2494         memcpy(fxsave->xmm_space, fpu->xmm, sizeof fxsave->xmm_space);
2495
2496         vcpu_put(vcpu);
2497
2498         return 0;
2499 }
2500
2501 static long kvm_vcpu_ioctl(struct file *filp,
2502                            unsigned int ioctl, unsigned long arg)
2503 {
2504         struct kvm_vcpu *vcpu = filp->private_data;
2505         void __user *argp = (void __user *)arg;
2506         int r = -EINVAL;
2507
2508         switch (ioctl) {
2509         case KVM_RUN:
2510                 r = -EINVAL;
2511                 if (arg)
2512                         goto out;
2513                 r = kvm_vcpu_ioctl_run(vcpu, vcpu->run);
2514                 break;
2515         case KVM_GET_REGS: {
2516                 struct kvm_regs kvm_regs;
2517
2518                 memset(&kvm_regs, 0, sizeof kvm_regs);
2519                 r = kvm_vcpu_ioctl_get_regs(vcpu, &kvm_regs);
2520                 if (r)
2521                         goto out;
2522                 r = -EFAULT;
2523                 if (copy_to_user(argp, &kvm_regs, sizeof kvm_regs))
2524                         goto out;
2525                 r = 0;
2526                 break;
2527         }
2528         case KVM_SET_REGS: {
2529                 struct kvm_regs kvm_regs;
2530
2531                 r = -EFAULT;
2532                 if (copy_from_user(&kvm_regs, argp, sizeof kvm_regs))
2533                         goto out;
2534                 r = kvm_vcpu_ioctl_set_regs(vcpu, &kvm_regs);
2535                 if (r)
2536                         goto out;
2537                 r = 0;
2538                 break;
2539         }
2540         case KVM_GET_SREGS: {
2541                 struct kvm_sregs kvm_sregs;
2542
2543                 memset(&kvm_sregs, 0, sizeof kvm_sregs);
2544                 r = kvm_vcpu_ioctl_get_sregs(vcpu, &kvm_sregs);
2545                 if (r)
2546                         goto out;
2547                 r = -EFAULT;
2548                 if (copy_to_user(argp, &kvm_sregs, sizeof kvm_sregs))
2549                         goto out;
2550                 r = 0;
2551                 break;
2552         }
2553         case KVM_SET_SREGS: {
2554                 struct kvm_sregs kvm_sregs;
2555
2556                 r = -EFAULT;
2557                 if (copy_from_user(&kvm_sregs, argp, sizeof kvm_sregs))
2558                         goto out;
2559                 r = kvm_vcpu_ioctl_set_sregs(vcpu, &kvm_sregs);
2560                 if (r)
2561                         goto out;
2562                 r = 0;
2563                 break;
2564         }
2565         case KVM_TRANSLATE: {
2566                 struct kvm_translation tr;
2567
2568                 r = -EFAULT;
2569                 if (copy_from_user(&tr, argp, sizeof tr))
2570                         goto out;
2571                 r = kvm_vcpu_ioctl_translate(vcpu, &tr);
2572                 if (r)
2573                         goto out;
2574                 r = -EFAULT;
2575                 if (copy_to_user(argp, &tr, sizeof tr))
2576                         goto out;
2577                 r = 0;
2578                 break;
2579         }
2580         case KVM_INTERRUPT: {
2581                 struct kvm_interrupt irq;
2582
2583                 r = -EFAULT;
2584                 if (copy_from_user(&irq, argp, sizeof irq))
2585                         goto out;
2586                 r = kvm_vcpu_ioctl_interrupt(vcpu, &irq);
2587                 if (r)
2588                         goto out;
2589                 r = 0;
2590                 break;
2591         }
2592         case KVM_DEBUG_GUEST: {
2593                 struct kvm_debug_guest dbg;
2594
2595                 r = -EFAULT;
2596                 if (copy_from_user(&dbg, argp, sizeof dbg))
2597                         goto out;
2598                 r = kvm_vcpu_ioctl_debug_guest(vcpu, &dbg);
2599                 if (r)
2600                         goto out;
2601                 r = 0;
2602                 break;
2603         }
2604         case KVM_GET_MSRS:
2605                 r = msr_io(vcpu, argp, kvm_get_msr, 1);
2606                 break;
2607         case KVM_SET_MSRS:
2608                 r = msr_io(vcpu, argp, do_set_msr, 0);
2609                 break;
2610         case KVM_SET_CPUID: {
2611                 struct kvm_cpuid __user *cpuid_arg = argp;
2612                 struct kvm_cpuid cpuid;
2613
2614                 r = -EFAULT;
2615                 if (copy_from_user(&cpuid, cpuid_arg, sizeof cpuid))
2616                         goto out;
2617                 r = kvm_vcpu_ioctl_set_cpuid(vcpu, &cpuid, cpuid_arg->entries);
2618                 if (r)
2619                         goto out;
2620                 break;
2621         }
2622         case KVM_SET_SIGNAL_MASK: {
2623                 struct kvm_signal_mask __user *sigmask_arg = argp;
2624                 struct kvm_signal_mask kvm_sigmask;
2625                 sigset_t sigset, *p;
2626
2627                 p = NULL;
2628                 if (argp) {
2629                         r = -EFAULT;
2630                         if (copy_from_user(&kvm_sigmask, argp,
2631                                            sizeof kvm_sigmask))
2632                                 goto out;
2633                         r = -EINVAL;
2634                         if (kvm_sigmask.len != sizeof sigset)
2635                                 goto out;
2636                         r = -EFAULT;
2637                         if (copy_from_user(&sigset, sigmask_arg->sigset,
2638                                            sizeof sigset))
2639                                 goto out;
2640                         p = &sigset;
2641                 }
2642                 r = kvm_vcpu_ioctl_set_sigmask(vcpu, &sigset);
2643                 break;
2644         }
2645         case KVM_GET_FPU: {
2646                 struct kvm_fpu fpu;
2647
2648                 memset(&fpu, 0, sizeof fpu);
2649                 r = kvm_vcpu_ioctl_get_fpu(vcpu, &fpu);
2650                 if (r)
2651                         goto out;
2652                 r = -EFAULT;
2653                 if (copy_to_user(argp, &fpu, sizeof fpu))
2654                         goto out;
2655                 r = 0;
2656                 break;
2657         }
2658         case KVM_SET_FPU: {
2659                 struct kvm_fpu fpu;
2660
2661                 r = -EFAULT;
2662                 if (copy_from_user(&fpu, argp, sizeof fpu))
2663                         goto out;
2664                 r = kvm_vcpu_ioctl_set_fpu(vcpu, &fpu);
2665                 if (r)
2666                         goto out;
2667                 r = 0;
2668                 break;
2669         }
2670         default:
2671                 ;
2672         }
2673 out:
2674         return r;
2675 }
2676
2677 static long kvm_vm_ioctl(struct file *filp,
2678                            unsigned int ioctl, unsigned long arg)
2679 {
2680         struct kvm *kvm = filp->private_data;
2681         void __user *argp = (void __user *)arg;
2682         int r = -EINVAL;
2683
2684         switch (ioctl) {
2685         case KVM_CREATE_VCPU:
2686                 r = kvm_vm_ioctl_create_vcpu(kvm, arg);
2687                 if (r < 0)
2688                         goto out;
2689                 break;
2690         case KVM_SET_MEMORY_REGION: {
2691                 struct kvm_memory_region kvm_mem;
2692
2693                 r = -EFAULT;
2694                 if (copy_from_user(&kvm_mem, argp, sizeof kvm_mem))
2695                         goto out;
2696                 r = kvm_vm_ioctl_set_memory_region(kvm, &kvm_mem);
2697                 if (r)
2698                         goto out;
2699                 break;
2700         }
2701         case KVM_GET_DIRTY_LOG: {
2702                 struct kvm_dirty_log log;
2703
2704                 r = -EFAULT;
2705                 if (copy_from_user(&log, argp, sizeof log))
2706                         goto out;
2707                 r = kvm_vm_ioctl_get_dirty_log(kvm, &log);
2708                 if (r)
2709                         goto out;
2710                 break;
2711         }
2712         case KVM_SET_MEMORY_ALIAS: {
2713                 struct kvm_memory_alias alias;
2714
2715                 r = -EFAULT;
2716                 if (copy_from_user(&alias, argp, sizeof alias))
2717                         goto out;
2718                 r = kvm_vm_ioctl_set_memory_alias(kvm, &alias);
2719                 if (r)
2720                         goto out;
2721                 break;
2722         }
2723         default:
2724                 ;
2725         }
2726 out:
2727         return r;
2728 }
2729
2730 static struct page *kvm_vm_nopage(struct vm_area_struct *vma,
2731                                   unsigned long address,
2732                                   int *type)
2733 {
2734         struct kvm *kvm = vma->vm_file->private_data;
2735         unsigned long pgoff;
2736         struct page *page;
2737
2738         pgoff = ((address - vma->vm_start) >> PAGE_SHIFT) + vma->vm_pgoff;
2739         page = gfn_to_page(kvm, pgoff);
2740         if (!page)
2741                 return NOPAGE_SIGBUS;
2742         get_page(page);
2743         if (type != NULL)
2744                 *type = VM_FAULT_MINOR;
2745
2746         return page;
2747 }
2748
2749 static struct vm_operations_struct kvm_vm_vm_ops = {
2750         .nopage = kvm_vm_nopage,
2751 };
2752
2753 static int kvm_vm_mmap(struct file *file, struct vm_area_struct *vma)
2754 {
2755         vma->vm_ops = &kvm_vm_vm_ops;
2756         return 0;
2757 }
2758
2759 static struct file_operations kvm_vm_fops = {
2760         .release        = kvm_vm_release,
2761         .unlocked_ioctl = kvm_vm_ioctl,
2762         .compat_ioctl   = kvm_vm_ioctl,
2763         .mmap           = kvm_vm_mmap,
2764 };
2765
2766 static int kvm_dev_ioctl_create_vm(void)
2767 {
2768         int fd, r;
2769         struct inode *inode;
2770         struct file *file;
2771         struct kvm *kvm;
2772
2773         kvm = kvm_create_vm();
2774         if (IS_ERR(kvm))
2775                 return PTR_ERR(kvm);
2776         r = anon_inode_getfd(&fd, &inode, &file, "kvm-vm", &kvm_vm_fops, kvm);
2777         if (r) {
2778                 kvm_destroy_vm(kvm);
2779                 return r;
2780         }
2781
2782         kvm->filp = file;
2783
2784         return fd;
2785 }
2786
2787 static long kvm_dev_ioctl(struct file *filp,
2788                           unsigned int ioctl, unsigned long arg)
2789 {
2790         void __user *argp = (void __user *)arg;
2791         long r = -EINVAL;
2792
2793         switch (ioctl) {
2794         case KVM_GET_API_VERSION:
2795                 r = -EINVAL;
2796                 if (arg)
2797                         goto out;
2798                 r = KVM_API_VERSION;
2799                 break;
2800         case KVM_CREATE_VM:
2801                 r = -EINVAL;
2802                 if (arg)
2803                         goto out;
2804                 r = kvm_dev_ioctl_create_vm();
2805                 break;
2806         case KVM_GET_MSR_INDEX_LIST: {
2807                 struct kvm_msr_list __user *user_msr_list = argp;
2808                 struct kvm_msr_list msr_list;
2809                 unsigned n;
2810
2811                 r = -EFAULT;
2812                 if (copy_from_user(&msr_list, user_msr_list, sizeof msr_list))
2813                         goto out;
2814                 n = msr_list.nmsrs;
2815                 msr_list.nmsrs = num_msrs_to_save + ARRAY_SIZE(emulated_msrs);
2816                 if (copy_to_user(user_msr_list, &msr_list, sizeof msr_list))
2817                         goto out;
2818                 r = -E2BIG;
2819                 if (n < num_msrs_to_save)
2820                         goto out;
2821                 r = -EFAULT;
2822                 if (copy_to_user(user_msr_list->indices, &msrs_to_save,
2823                                  num_msrs_to_save * sizeof(u32)))
2824                         goto out;
2825                 if (copy_to_user(user_msr_list->indices
2826                                  + num_msrs_to_save * sizeof(u32),
2827                                  &emulated_msrs,
2828                                  ARRAY_SIZE(emulated_msrs) * sizeof(u32)))
2829                         goto out;
2830                 r = 0;
2831                 break;
2832         }
2833         case KVM_CHECK_EXTENSION:
2834                 /*
2835                  * No extensions defined at present.
2836                  */
2837                 r = 0;
2838                 break;
2839         case KVM_GET_VCPU_MMAP_SIZE:
2840                 r = -EINVAL;
2841                 if (arg)
2842                         goto out;
2843                 r = 2 * PAGE_SIZE;
2844                 break;
2845         default:
2846                 ;
2847         }
2848 out:
2849         return r;
2850 }
2851
2852 static struct file_operations kvm_chardev_ops = {
2853         .open           = kvm_dev_open,
2854         .release        = kvm_dev_release,
2855         .unlocked_ioctl = kvm_dev_ioctl,
2856         .compat_ioctl   = kvm_dev_ioctl,
2857 };
2858
2859 static struct miscdevice kvm_dev = {
2860         KVM_MINOR,
2861         "kvm",
2862         &kvm_chardev_ops,
2863 };
2864
2865 /*
2866  * Make sure that a cpu that is being hot-unplugged does not have any vcpus
2867  * cached on it.
2868  */
2869 static void decache_vcpus_on_cpu(int cpu)
2870 {
2871         struct kvm *vm;
2872         struct kvm_vcpu *vcpu;
2873         int i;
2874
2875         spin_lock(&kvm_lock);
2876         list_for_each_entry(vm, &vm_list, vm_list)
2877                 for (i = 0; i < KVM_MAX_VCPUS; ++i) {
2878                         vcpu = vm->vcpus[i];
2879                         if (!vcpu)
2880                                 continue;
2881                         /*
2882                          * If the vcpu is locked, then it is running on some
2883                          * other cpu and therefore it is not cached on the
2884                          * cpu in question.
2885                          *
2886                          * If it's not locked, check the last cpu it executed
2887                          * on.
2888                          */
2889                         if (mutex_trylock(&vcpu->mutex)) {
2890                                 if (vcpu->cpu == cpu) {
2891                                         kvm_arch_ops->vcpu_decache(vcpu);
2892                                         vcpu->cpu = -1;
2893                                 }
2894                                 mutex_unlock(&vcpu->mutex);
2895                         }
2896                 }
2897         spin_unlock(&kvm_lock);
2898 }
2899
2900 static void hardware_enable(void *junk)
2901 {
2902         int cpu = raw_smp_processor_id();
2903
2904         if (cpu_isset(cpu, cpus_hardware_enabled))
2905                 return;
2906         cpu_set(cpu, cpus_hardware_enabled);
2907         kvm_arch_ops->hardware_enable(NULL);
2908 }
2909
2910 static void hardware_disable(void *junk)
2911 {
2912         int cpu = raw_smp_processor_id();
2913
2914         if (!cpu_isset(cpu, cpus_hardware_enabled))
2915                 return;
2916         cpu_clear(cpu, cpus_hardware_enabled);
2917         decache_vcpus_on_cpu(cpu);
2918         kvm_arch_ops->hardware_disable(NULL);
2919 }
2920
2921 static int kvm_cpu_hotplug(struct notifier_block *notifier, unsigned long val,
2922                            void *v)
2923 {
2924         int cpu = (long)v;
2925
2926         switch (val) {
2927         case CPU_DYING:
2928         case CPU_DYING_FROZEN:
2929                 printk(KERN_INFO "kvm: disabling virtualization on CPU%d\n",
2930                        cpu);
2931                 hardware_disable(NULL);
2932                 break;
2933         case CPU_UP_CANCELED:
2934         case CPU_UP_CANCELED_FROZEN:
2935                 printk(KERN_INFO "kvm: disabling virtualization on CPU%d\n",
2936                        cpu);
2937                 smp_call_function_single(cpu, hardware_disable, NULL, 0, 1);
2938                 break;
2939         case CPU_ONLINE:
2940         case CPU_ONLINE_FROZEN:
2941                 printk(KERN_INFO "kvm: enabling virtualization on CPU%d\n",
2942                        cpu);
2943                 smp_call_function_single(cpu, hardware_enable, NULL, 0, 1);
2944                 break;
2945         }
2946         return NOTIFY_OK;
2947 }
2948
2949 static int kvm_reboot(struct notifier_block *notifier, unsigned long val,
2950                        void *v)
2951 {
2952         if (val == SYS_RESTART) {
2953                 /*
2954                  * Some (well, at least mine) BIOSes hang on reboot if
2955                  * in vmx root mode.
2956                  */
2957                 printk(KERN_INFO "kvm: exiting hardware virtualization\n");
2958                 on_each_cpu(hardware_disable, NULL, 0, 1);
2959         }
2960         return NOTIFY_OK;
2961 }
2962
2963 static struct notifier_block kvm_reboot_notifier = {
2964         .notifier_call = kvm_reboot,
2965         .priority = 0,
2966 };
2967
2968 void kvm_io_bus_init(struct kvm_io_bus *bus)
2969 {
2970         memset(bus, 0, sizeof(*bus));
2971 }
2972
2973 void kvm_io_bus_destroy(struct kvm_io_bus *bus)
2974 {
2975         int i;
2976
2977         for (i = 0; i < bus->dev_count; i++) {
2978                 struct kvm_io_device *pos = bus->devs[i];
2979
2980                 kvm_iodevice_destructor(pos);
2981         }
2982 }
2983
2984 struct kvm_io_device *kvm_io_bus_find_dev(struct kvm_io_bus *bus, gpa_t addr)
2985 {
2986         int i;
2987
2988         for (i = 0; i < bus->dev_count; i++) {
2989                 struct kvm_io_device *pos = bus->devs[i];
2990
2991                 if (pos->in_range(pos, addr))
2992                         return pos;
2993         }
2994
2995         return NULL;
2996 }
2997
2998 void kvm_io_bus_register_dev(struct kvm_io_bus *bus, struct kvm_io_device *dev)
2999 {
3000         BUG_ON(bus->dev_count > (NR_IOBUS_DEVS-1));
3001
3002         bus->devs[bus->dev_count++] = dev;
3003 }
3004
3005 static struct notifier_block kvm_cpu_notifier = {
3006         .notifier_call = kvm_cpu_hotplug,
3007         .priority = 20, /* must be > scheduler priority */
3008 };
3009
3010 static u64 stat_get(void *_offset)
3011 {
3012         unsigned offset = (long)_offset;
3013         u64 total = 0;
3014         struct kvm *kvm;
3015         struct kvm_vcpu *vcpu;
3016         int i;
3017
3018         spin_lock(&kvm_lock);
3019         list_for_each_entry(kvm, &vm_list, vm_list)
3020                 for (i = 0; i < KVM_MAX_VCPUS; ++i) {
3021                         vcpu = kvm->vcpus[i];
3022                         if (vcpu)
3023                                 total += *(u32 *)((void *)vcpu + offset);
3024                 }
3025         spin_unlock(&kvm_lock);
3026         return total;
3027 }
3028
3029 static void stat_set(void *offset, u64 val)
3030 {
3031 }
3032
3033 DEFINE_SIMPLE_ATTRIBUTE(stat_fops, stat_get, stat_set, "%llu\n");
3034
3035 static __init void kvm_init_debug(void)
3036 {
3037         struct kvm_stats_debugfs_item *p;
3038
3039         debugfs_dir = debugfs_create_dir("kvm", NULL);
3040         for (p = debugfs_entries; p->name; ++p)
3041                 p->dentry = debugfs_create_file(p->name, 0444, debugfs_dir,
3042                                                 (void *)(long)p->offset,
3043                                                 &stat_fops);
3044 }
3045
3046 static void kvm_exit_debug(void)
3047 {
3048         struct kvm_stats_debugfs_item *p;
3049
3050         for (p = debugfs_entries; p->name; ++p)
3051                 debugfs_remove(p->dentry);
3052         debugfs_remove(debugfs_dir);
3053 }
3054
3055 static int kvm_suspend(struct sys_device *dev, pm_message_t state)
3056 {
3057         hardware_disable(NULL);
3058         return 0;
3059 }
3060
3061 static int kvm_resume(struct sys_device *dev)
3062 {
3063         hardware_enable(NULL);
3064         return 0;
3065 }
3066
3067 static struct sysdev_class kvm_sysdev_class = {
3068         set_kset_name("kvm"),
3069         .suspend = kvm_suspend,
3070         .resume = kvm_resume,
3071 };
3072
3073 static struct sys_device kvm_sysdev = {
3074         .id = 0,
3075         .cls = &kvm_sysdev_class,
3076 };
3077
3078 hpa_t bad_page_address;
3079
3080 static inline
3081 struct kvm_vcpu *preempt_notifier_to_vcpu(struct preempt_notifier *pn)
3082 {
3083         return container_of(pn, struct kvm_vcpu, preempt_notifier);
3084 }
3085
3086 static void kvm_sched_in(struct preempt_notifier *pn, int cpu)
3087 {
3088         struct kvm_vcpu *vcpu = preempt_notifier_to_vcpu(pn);
3089
3090         kvm_arch_ops->vcpu_load(vcpu, cpu);
3091 }
3092
3093 static void kvm_sched_out(struct preempt_notifier *pn,
3094                           struct task_struct *next)
3095 {
3096         struct kvm_vcpu *vcpu = preempt_notifier_to_vcpu(pn);
3097
3098         kvm_arch_ops->vcpu_put(vcpu);
3099 }
3100
3101 int kvm_init_arch(struct kvm_arch_ops *ops, unsigned int vcpu_size,
3102                   struct module *module)
3103 {
3104         int r;
3105
3106         if (kvm_arch_ops) {
3107                 printk(KERN_ERR "kvm: already loaded the other module\n");
3108                 return -EEXIST;
3109         }
3110
3111         if (!ops->cpu_has_kvm_support()) {
3112                 printk(KERN_ERR "kvm: no hardware support\n");
3113                 return -EOPNOTSUPP;
3114         }
3115         if (ops->disabled_by_bios()) {
3116                 printk(KERN_ERR "kvm: disabled by bios\n");
3117                 return -EOPNOTSUPP;
3118         }
3119
3120         kvm_arch_ops = ops;
3121
3122         r = kvm_arch_ops->hardware_setup();
3123         if (r < 0)
3124                 goto out;
3125
3126         on_each_cpu(hardware_enable, NULL, 0, 1);
3127         r = register_cpu_notifier(&kvm_cpu_notifier);
3128         if (r)
3129                 goto out_free_1;
3130         register_reboot_notifier(&kvm_reboot_notifier);
3131
3132         r = sysdev_class_register(&kvm_sysdev_class);
3133         if (r)
3134                 goto out_free_2;
3135
3136         r = sysdev_register(&kvm_sysdev);
3137         if (r)
3138                 goto out_free_3;
3139
3140         /* A kmem cache lets us meet the alignment requirements of fx_save. */
3141         kvm_vcpu_cache = kmem_cache_create("kvm_vcpu", vcpu_size,
3142                                            __alignof__(struct kvm_vcpu), 0, 0);
3143         if (!kvm_vcpu_cache) {
3144                 r = -ENOMEM;
3145                 goto out_free_4;
3146         }
3147
3148         kvm_chardev_ops.owner = module;
3149
3150         r = misc_register(&kvm_dev);
3151         if (r) {
3152                 printk (KERN_ERR "kvm: misc device register failed\n");
3153                 goto out_free;
3154         }
3155
3156         kvm_preempt_ops.sched_in = kvm_sched_in;
3157         kvm_preempt_ops.sched_out = kvm_sched_out;
3158
3159         return r;
3160
3161 out_free:
3162         kmem_cache_destroy(kvm_vcpu_cache);
3163 out_free_4:
3164         sysdev_unregister(&kvm_sysdev);
3165 out_free_3:
3166         sysdev_class_unregister(&kvm_sysdev_class);
3167 out_free_2:
3168         unregister_reboot_notifier(&kvm_reboot_notifier);
3169         unregister_cpu_notifier(&kvm_cpu_notifier);
3170 out_free_1:
3171         on_each_cpu(hardware_disable, NULL, 0, 1);
3172         kvm_arch_ops->hardware_unsetup();
3173 out:
3174         kvm_arch_ops = NULL;
3175         return r;
3176 }
3177
3178 void kvm_exit_arch(void)
3179 {
3180         misc_deregister(&kvm_dev);
3181         kmem_cache_destroy(kvm_vcpu_cache);
3182         sysdev_unregister(&kvm_sysdev);
3183         sysdev_class_unregister(&kvm_sysdev_class);
3184         unregister_reboot_notifier(&kvm_reboot_notifier);
3185         unregister_cpu_notifier(&kvm_cpu_notifier);
3186         on_each_cpu(hardware_disable, NULL, 0, 1);
3187         kvm_arch_ops->hardware_unsetup();
3188         kvm_arch_ops = NULL;
3189 }
3190
3191 static __init int kvm_init(void)
3192 {
3193         static struct page *bad_page;
3194         int r;
3195
3196         r = kvm_mmu_module_init();
3197         if (r)
3198                 goto out4;
3199
3200         kvm_init_debug();
3201
3202         kvm_init_msr_list();
3203
3204         if ((bad_page = alloc_page(GFP_KERNEL)) == NULL) {
3205                 r = -ENOMEM;
3206                 goto out;
3207         }
3208
3209         bad_page_address = page_to_pfn(bad_page) << PAGE_SHIFT;
3210         memset(__va(bad_page_address), 0, PAGE_SIZE);
3211
3212         return 0;
3213
3214 out:
3215         kvm_exit_debug();
3216         kvm_mmu_module_exit();
3217 out4:
3218         return r;
3219 }
3220
3221 static __exit void kvm_exit(void)
3222 {
3223         kvm_exit_debug();
3224         __free_page(pfn_to_page(bad_page_address >> PAGE_SHIFT));
3225         kvm_mmu_module_exit();
3226 }
3227
3228 module_init(kvm_init)
3229 module_exit(kvm_exit)
3230
3231 EXPORT_SYMBOL_GPL(kvm_init_arch);
3232 EXPORT_SYMBOL_GPL(kvm_exit_arch);