KVM: Trivial: Use standard CR0 flags macros from asm/cpu-features.h
[linux-2.6] / drivers / kvm / kvm_main.c
1 /*
2  * Kernel-based Virtual Machine driver for Linux
3  *
4  * This module enables machines with Intel VT-x extensions to run virtual
5  * machines without emulation or binary translation.
6  *
7  * Copyright (C) 2006 Qumranet, Inc.
8  *
9  * Authors:
10  *   Avi Kivity   <avi@qumranet.com>
11  *   Yaniv Kamay  <yaniv@qumranet.com>
12  *
13  * This work is licensed under the terms of the GNU GPL, version 2.  See
14  * the COPYING file in the top-level directory.
15  *
16  */
17
18 #include "kvm.h"
19 #include "x86_emulate.h"
20 #include "segment_descriptor.h"
21
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
57 #define STAT_OFFSET(x) offsetof(struct kvm_vcpu, stat.x)
58
59 static struct kvm_stats_debugfs_item {
60         const char *name;
61         int offset;
62         struct dentry *dentry;
63 } debugfs_entries[] = {
64         { "pf_fixed", STAT_OFFSET(pf_fixed) },
65         { "pf_guest", STAT_OFFSET(pf_guest) },
66         { "tlb_flush", STAT_OFFSET(tlb_flush) },
67         { "invlpg", STAT_OFFSET(invlpg) },
68         { "exits", STAT_OFFSET(exits) },
69         { "io_exits", STAT_OFFSET(io_exits) },
70         { "mmio_exits", STAT_OFFSET(mmio_exits) },
71         { "signal_exits", STAT_OFFSET(signal_exits) },
72         { "irq_window", STAT_OFFSET(irq_window_exits) },
73         { "halt_exits", STAT_OFFSET(halt_exits) },
74         { "request_irq", STAT_OFFSET(request_irq_exits) },
75         { "irq_exits", STAT_OFFSET(irq_exits) },
76         { "light_exits", STAT_OFFSET(light_exits) },
77         { "efer_reload", STAT_OFFSET(efer_reload) },
78         { NULL }
79 };
80
81 static struct dentry *debugfs_dir;
82
83 #define MAX_IO_MSRS 256
84
85 #define CR0_RESERVED_BITS                                               \
86         (~(unsigned long)(X86_CR0_PE | X86_CR0_MP | X86_CR0_EM | X86_CR0_TS \
87                           | X86_CR0_ET | X86_CR0_NE | X86_CR0_WP | X86_CR0_AM \
88                           | X86_CR0_NW | X86_CR0_CD | X86_CR0_PG))
89 #define LMSW_GUEST_MASK 0x0eULL
90 #define CR4_RESEVED_BITS (~((1ULL << 11) - 1))
91 #define CR8_RESEVED_BITS (~0x0fULL)
92 #define EFER_RESERVED_BITS 0xfffffffffffff2fe
93
94 #ifdef CONFIG_X86_64
95 // LDT or TSS descriptor in the GDT. 16 bytes.
96 struct segment_descriptor_64 {
97         struct segment_descriptor s;
98         u32 base_higher;
99         u32 pad_zero;
100 };
101
102 #endif
103
104 static long kvm_vcpu_ioctl(struct file *file, unsigned int ioctl,
105                            unsigned long arg);
106
107 unsigned long segment_base(u16 selector)
108 {
109         struct descriptor_table gdt;
110         struct segment_descriptor *d;
111         unsigned long table_base;
112         typedef unsigned long ul;
113         unsigned long v;
114
115         if (selector == 0)
116                 return 0;
117
118         asm ("sgdt %0" : "=m"(gdt));
119         table_base = gdt.base;
120
121         if (selector & 4) {           /* from ldt */
122                 u16 ldt_selector;
123
124                 asm ("sldt %0" : "=g"(ldt_selector));
125                 table_base = segment_base(ldt_selector);
126         }
127         d = (struct segment_descriptor *)(table_base + (selector & ~7));
128         v = d->base_low | ((ul)d->base_mid << 16) | ((ul)d->base_high << 24);
129 #ifdef CONFIG_X86_64
130         if (d->system == 0
131             && (d->type == 2 || d->type == 9 || d->type == 11))
132                 v |= ((ul)((struct segment_descriptor_64 *)d)->base_higher) << 32;
133 #endif
134         return v;
135 }
136 EXPORT_SYMBOL_GPL(segment_base);
137
138 static inline int valid_vcpu(int n)
139 {
140         return likely(n >= 0 && n < KVM_MAX_VCPUS);
141 }
142
143 int kvm_read_guest(struct kvm_vcpu *vcpu, gva_t addr, unsigned long size,
144                    void *dest)
145 {
146         unsigned char *host_buf = dest;
147         unsigned long req_size = size;
148
149         while (size) {
150                 hpa_t paddr;
151                 unsigned now;
152                 unsigned offset;
153                 hva_t guest_buf;
154
155                 paddr = gva_to_hpa(vcpu, addr);
156
157                 if (is_error_hpa(paddr))
158                         break;
159
160                 guest_buf = (hva_t)kmap_atomic(
161                                         pfn_to_page(paddr >> PAGE_SHIFT),
162                                         KM_USER0);
163                 offset = addr & ~PAGE_MASK;
164                 guest_buf |= offset;
165                 now = min(size, PAGE_SIZE - offset);
166                 memcpy(host_buf, (void*)guest_buf, now);
167                 host_buf += now;
168                 addr += now;
169                 size -= now;
170                 kunmap_atomic((void *)(guest_buf & PAGE_MASK), KM_USER0);
171         }
172         return req_size - size;
173 }
174 EXPORT_SYMBOL_GPL(kvm_read_guest);
175
176 int kvm_write_guest(struct kvm_vcpu *vcpu, gva_t addr, unsigned long size,
177                     void *data)
178 {
179         unsigned char *host_buf = data;
180         unsigned long req_size = size;
181
182         while (size) {
183                 hpa_t paddr;
184                 unsigned now;
185                 unsigned offset;
186                 hva_t guest_buf;
187                 gfn_t gfn;
188
189                 paddr = gva_to_hpa(vcpu, addr);
190
191                 if (is_error_hpa(paddr))
192                         break;
193
194                 gfn = vcpu->mmu.gva_to_gpa(vcpu, addr) >> PAGE_SHIFT;
195                 mark_page_dirty(vcpu->kvm, gfn);
196                 guest_buf = (hva_t)kmap_atomic(
197                                 pfn_to_page(paddr >> PAGE_SHIFT), KM_USER0);
198                 offset = addr & ~PAGE_MASK;
199                 guest_buf |= offset;
200                 now = min(size, PAGE_SIZE - offset);
201                 memcpy((void*)guest_buf, host_buf, now);
202                 host_buf += now;
203                 addr += now;
204                 size -= now;
205                 kunmap_atomic((void *)(guest_buf & PAGE_MASK), KM_USER0);
206         }
207         return req_size - size;
208 }
209 EXPORT_SYMBOL_GPL(kvm_write_guest);
210
211 void kvm_load_guest_fpu(struct kvm_vcpu *vcpu)
212 {
213         if (!vcpu->fpu_active || vcpu->guest_fpu_loaded)
214                 return;
215
216         vcpu->guest_fpu_loaded = 1;
217         fx_save(vcpu->host_fx_image);
218         fx_restore(vcpu->guest_fx_image);
219 }
220 EXPORT_SYMBOL_GPL(kvm_load_guest_fpu);
221
222 void kvm_put_guest_fpu(struct kvm_vcpu *vcpu)
223 {
224         if (!vcpu->guest_fpu_loaded)
225                 return;
226
227         vcpu->guest_fpu_loaded = 0;
228         fx_save(vcpu->guest_fx_image);
229         fx_restore(vcpu->host_fx_image);
230 }
231 EXPORT_SYMBOL_GPL(kvm_put_guest_fpu);
232
233 /*
234  * Switches to specified vcpu, until a matching vcpu_put()
235  */
236 static void vcpu_load(struct kvm_vcpu *vcpu)
237 {
238         mutex_lock(&vcpu->mutex);
239         kvm_arch_ops->vcpu_load(vcpu);
240 }
241
242 static void vcpu_put(struct kvm_vcpu *vcpu)
243 {
244         kvm_arch_ops->vcpu_put(vcpu);
245         mutex_unlock(&vcpu->mutex);
246 }
247
248 static void ack_flush(void *_completed)
249 {
250         atomic_t *completed = _completed;
251
252         atomic_inc(completed);
253 }
254
255 void kvm_flush_remote_tlbs(struct kvm *kvm)
256 {
257         int i, cpu, needed;
258         cpumask_t cpus;
259         struct kvm_vcpu *vcpu;
260         atomic_t completed;
261
262         atomic_set(&completed, 0);
263         cpus_clear(cpus);
264         needed = 0;
265         for (i = 0; i < kvm->nvcpus; ++i) {
266                 vcpu = &kvm->vcpus[i];
267                 if (test_and_set_bit(KVM_TLB_FLUSH, &vcpu->requests))
268                         continue;
269                 cpu = vcpu->cpu;
270                 if (cpu != -1 && cpu != raw_smp_processor_id())
271                         if (!cpu_isset(cpu, cpus)) {
272                                 cpu_set(cpu, cpus);
273                                 ++needed;
274                         }
275         }
276
277         /*
278          * We really want smp_call_function_mask() here.  But that's not
279          * available, so ipi all cpus in parallel and wait for them
280          * to complete.
281          */
282         for (cpu = first_cpu(cpus); cpu != NR_CPUS; cpu = next_cpu(cpu, cpus))
283                 smp_call_function_single(cpu, ack_flush, &completed, 1, 0);
284         while (atomic_read(&completed) != needed) {
285                 cpu_relax();
286                 barrier();
287         }
288 }
289
290 static struct kvm *kvm_create_vm(void)
291 {
292         struct kvm *kvm = kzalloc(sizeof(struct kvm), GFP_KERNEL);
293         int i;
294
295         if (!kvm)
296                 return ERR_PTR(-ENOMEM);
297
298         kvm_io_bus_init(&kvm->pio_bus);
299         spin_lock_init(&kvm->lock);
300         INIT_LIST_HEAD(&kvm->active_mmu_pages);
301         kvm_io_bus_init(&kvm->mmio_bus);
302         for (i = 0; i < KVM_MAX_VCPUS; ++i) {
303                 struct kvm_vcpu *vcpu = &kvm->vcpus[i];
304
305                 mutex_init(&vcpu->mutex);
306                 vcpu->cpu = -1;
307                 vcpu->kvm = kvm;
308                 vcpu->mmu.root_hpa = INVALID_PAGE;
309         }
310         spin_lock(&kvm_lock);
311         list_add(&kvm->vm_list, &vm_list);
312         spin_unlock(&kvm_lock);
313         return kvm;
314 }
315
316 static int kvm_dev_open(struct inode *inode, struct file *filp)
317 {
318         return 0;
319 }
320
321 /*
322  * Free any memory in @free but not in @dont.
323  */
324 static void kvm_free_physmem_slot(struct kvm_memory_slot *free,
325                                   struct kvm_memory_slot *dont)
326 {
327         int i;
328
329         if (!dont || free->phys_mem != dont->phys_mem)
330                 if (free->phys_mem) {
331                         for (i = 0; i < free->npages; ++i)
332                                 if (free->phys_mem[i])
333                                         __free_page(free->phys_mem[i]);
334                         vfree(free->phys_mem);
335                 }
336
337         if (!dont || free->dirty_bitmap != dont->dirty_bitmap)
338                 vfree(free->dirty_bitmap);
339
340         free->phys_mem = NULL;
341         free->npages = 0;
342         free->dirty_bitmap = NULL;
343 }
344
345 static void kvm_free_physmem(struct kvm *kvm)
346 {
347         int i;
348
349         for (i = 0; i < kvm->nmemslots; ++i)
350                 kvm_free_physmem_slot(&kvm->memslots[i], NULL);
351 }
352
353 static void free_pio_guest_pages(struct kvm_vcpu *vcpu)
354 {
355         int i;
356
357         for (i = 0; i < 2; ++i)
358                 if (vcpu->pio.guest_pages[i]) {
359                         __free_page(vcpu->pio.guest_pages[i]);
360                         vcpu->pio.guest_pages[i] = NULL;
361                 }
362 }
363
364 static void kvm_unload_vcpu_mmu(struct kvm_vcpu *vcpu)
365 {
366         if (!vcpu->vmcs)
367                 return;
368
369         vcpu_load(vcpu);
370         kvm_mmu_unload(vcpu);
371         vcpu_put(vcpu);
372 }
373
374 static void kvm_free_vcpu(struct kvm_vcpu *vcpu)
375 {
376         if (!vcpu->vmcs)
377                 return;
378
379         vcpu_load(vcpu);
380         kvm_mmu_destroy(vcpu);
381         vcpu_put(vcpu);
382         kvm_arch_ops->vcpu_free(vcpu);
383         free_page((unsigned long)vcpu->run);
384         vcpu->run = NULL;
385         free_page((unsigned long)vcpu->pio_data);
386         vcpu->pio_data = NULL;
387         free_pio_guest_pages(vcpu);
388 }
389
390 static void kvm_free_vcpus(struct kvm *kvm)
391 {
392         unsigned int i;
393
394         /*
395          * Unpin any mmu pages first.
396          */
397         for (i = 0; i < KVM_MAX_VCPUS; ++i)
398                 kvm_unload_vcpu_mmu(&kvm->vcpus[i]);
399         for (i = 0; i < KVM_MAX_VCPUS; ++i)
400                 kvm_free_vcpu(&kvm->vcpus[i]);
401 }
402
403 static int kvm_dev_release(struct inode *inode, struct file *filp)
404 {
405         return 0;
406 }
407
408 static void kvm_destroy_vm(struct kvm *kvm)
409 {
410         spin_lock(&kvm_lock);
411         list_del(&kvm->vm_list);
412         spin_unlock(&kvm_lock);
413         kvm_io_bus_destroy(&kvm->pio_bus);
414         kvm_io_bus_destroy(&kvm->mmio_bus);
415         kvm_free_vcpus(kvm);
416         kvm_free_physmem(kvm);
417         kfree(kvm);
418 }
419
420 static int kvm_vm_release(struct inode *inode, struct file *filp)
421 {
422         struct kvm *kvm = filp->private_data;
423
424         kvm_destroy_vm(kvm);
425         return 0;
426 }
427
428 static void inject_gp(struct kvm_vcpu *vcpu)
429 {
430         kvm_arch_ops->inject_gp(vcpu, 0);
431 }
432
433 /*
434  * Load the pae pdptrs.  Return true is they are all valid.
435  */
436 static int load_pdptrs(struct kvm_vcpu *vcpu, unsigned long cr3)
437 {
438         gfn_t pdpt_gfn = cr3 >> PAGE_SHIFT;
439         unsigned offset = ((cr3 & (PAGE_SIZE-1)) >> 5) << 2;
440         int i;
441         u64 pdpte;
442         u64 *pdpt;
443         int ret;
444         struct page *page;
445
446         spin_lock(&vcpu->kvm->lock);
447         page = gfn_to_page(vcpu->kvm, pdpt_gfn);
448         /* FIXME: !page - emulate? 0xff? */
449         pdpt = kmap_atomic(page, KM_USER0);
450
451         ret = 1;
452         for (i = 0; i < 4; ++i) {
453                 pdpte = pdpt[offset + i];
454                 if ((pdpte & 1) && (pdpte & 0xfffffff0000001e6ull)) {
455                         ret = 0;
456                         goto out;
457                 }
458         }
459
460         for (i = 0; i < 4; ++i)
461                 vcpu->pdptrs[i] = pdpt[offset + i];
462
463 out:
464         kunmap_atomic(pdpt, KM_USER0);
465         spin_unlock(&vcpu->kvm->lock);
466
467         return ret;
468 }
469
470 void set_cr0(struct kvm_vcpu *vcpu, unsigned long cr0)
471 {
472         if (cr0 & CR0_RESERVED_BITS) {
473                 printk(KERN_DEBUG "set_cr0: 0x%lx #GP, reserved bits 0x%lx\n",
474                        cr0, vcpu->cr0);
475                 inject_gp(vcpu);
476                 return;
477         }
478
479         if ((cr0 & X86_CR0_NW) && !(cr0 & X86_CR0_CD)) {
480                 printk(KERN_DEBUG "set_cr0: #GP, CD == 0 && NW == 1\n");
481                 inject_gp(vcpu);
482                 return;
483         }
484
485         if ((cr0 & X86_CR0_PG) && !(cr0 & X86_CR0_PE)) {
486                 printk(KERN_DEBUG "set_cr0: #GP, set PG flag "
487                        "and a clear PE flag\n");
488                 inject_gp(vcpu);
489                 return;
490         }
491
492         if (!is_paging(vcpu) && (cr0 & X86_CR0_PG)) {
493 #ifdef CONFIG_X86_64
494                 if ((vcpu->shadow_efer & EFER_LME)) {
495                         int cs_db, cs_l;
496
497                         if (!is_pae(vcpu)) {
498                                 printk(KERN_DEBUG "set_cr0: #GP, start paging "
499                                        "in long mode while PAE is disabled\n");
500                                 inject_gp(vcpu);
501                                 return;
502                         }
503                         kvm_arch_ops->get_cs_db_l_bits(vcpu, &cs_db, &cs_l);
504                         if (cs_l) {
505                                 printk(KERN_DEBUG "set_cr0: #GP, start paging "
506                                        "in long mode while CS.L == 1\n");
507                                 inject_gp(vcpu);
508                                 return;
509
510                         }
511                 } else
512 #endif
513                 if (is_pae(vcpu) && !load_pdptrs(vcpu, vcpu->cr3)) {
514                         printk(KERN_DEBUG "set_cr0: #GP, pdptrs "
515                                "reserved bits\n");
516                         inject_gp(vcpu);
517                         return;
518                 }
519
520         }
521
522         kvm_arch_ops->set_cr0(vcpu, cr0);
523         vcpu->cr0 = cr0;
524
525         spin_lock(&vcpu->kvm->lock);
526         kvm_mmu_reset_context(vcpu);
527         spin_unlock(&vcpu->kvm->lock);
528         return;
529 }
530 EXPORT_SYMBOL_GPL(set_cr0);
531
532 void lmsw(struct kvm_vcpu *vcpu, unsigned long msw)
533 {
534         set_cr0(vcpu, (vcpu->cr0 & ~0x0ful) | (msw & 0x0f));
535 }
536 EXPORT_SYMBOL_GPL(lmsw);
537
538 void set_cr4(struct kvm_vcpu *vcpu, unsigned long cr4)
539 {
540         if (cr4 & CR4_RESEVED_BITS) {
541                 printk(KERN_DEBUG "set_cr4: #GP, reserved bits\n");
542                 inject_gp(vcpu);
543                 return;
544         }
545
546         if (is_long_mode(vcpu)) {
547                 if (!(cr4 & CR4_PAE_MASK)) {
548                         printk(KERN_DEBUG "set_cr4: #GP, clearing PAE while "
549                                "in long mode\n");
550                         inject_gp(vcpu);
551                         return;
552                 }
553         } else if (is_paging(vcpu) && !is_pae(vcpu) && (cr4 & CR4_PAE_MASK)
554                    && !load_pdptrs(vcpu, vcpu->cr3)) {
555                 printk(KERN_DEBUG "set_cr4: #GP, pdptrs reserved bits\n");
556                 inject_gp(vcpu);
557         }
558
559         if (cr4 & CR4_VMXE_MASK) {
560                 printk(KERN_DEBUG "set_cr4: #GP, setting VMXE\n");
561                 inject_gp(vcpu);
562                 return;
563         }
564         kvm_arch_ops->set_cr4(vcpu, cr4);
565         spin_lock(&vcpu->kvm->lock);
566         kvm_mmu_reset_context(vcpu);
567         spin_unlock(&vcpu->kvm->lock);
568 }
569 EXPORT_SYMBOL_GPL(set_cr4);
570
571 void set_cr3(struct kvm_vcpu *vcpu, unsigned long cr3)
572 {
573         if (is_long_mode(vcpu)) {
574                 if (cr3 & CR3_L_MODE_RESEVED_BITS) {
575                         printk(KERN_DEBUG "set_cr3: #GP, reserved bits\n");
576                         inject_gp(vcpu);
577                         return;
578                 }
579         } else {
580                 if (cr3 & CR3_RESEVED_BITS) {
581                         printk(KERN_DEBUG "set_cr3: #GP, reserved bits\n");
582                         inject_gp(vcpu);
583                         return;
584                 }
585                 if (is_paging(vcpu) && is_pae(vcpu) &&
586                     !load_pdptrs(vcpu, cr3)) {
587                         printk(KERN_DEBUG "set_cr3: #GP, pdptrs "
588                                "reserved bits\n");
589                         inject_gp(vcpu);
590                         return;
591                 }
592         }
593
594         vcpu->cr3 = cr3;
595         spin_lock(&vcpu->kvm->lock);
596         /*
597          * Does the new cr3 value map to physical memory? (Note, we
598          * catch an invalid cr3 even in real-mode, because it would
599          * cause trouble later on when we turn on paging anyway.)
600          *
601          * A real CPU would silently accept an invalid cr3 and would
602          * attempt to use it - with largely undefined (and often hard
603          * to debug) behavior on the guest side.
604          */
605         if (unlikely(!gfn_to_memslot(vcpu->kvm, cr3 >> PAGE_SHIFT)))
606                 inject_gp(vcpu);
607         else
608                 vcpu->mmu.new_cr3(vcpu);
609         spin_unlock(&vcpu->kvm->lock);
610 }
611 EXPORT_SYMBOL_GPL(set_cr3);
612
613 void set_cr8(struct kvm_vcpu *vcpu, unsigned long cr8)
614 {
615         if ( cr8 & CR8_RESEVED_BITS) {
616                 printk(KERN_DEBUG "set_cr8: #GP, reserved bits 0x%lx\n", cr8);
617                 inject_gp(vcpu);
618                 return;
619         }
620         vcpu->cr8 = cr8;
621 }
622 EXPORT_SYMBOL_GPL(set_cr8);
623
624 void fx_init(struct kvm_vcpu *vcpu)
625 {
626         struct __attribute__ ((__packed__)) fx_image_s {
627                 u16 control; //fcw
628                 u16 status; //fsw
629                 u16 tag; // ftw
630                 u16 opcode; //fop
631                 u64 ip; // fpu ip
632                 u64 operand;// fpu dp
633                 u32 mxcsr;
634                 u32 mxcsr_mask;
635
636         } *fx_image;
637
638         fx_save(vcpu->host_fx_image);
639         fpu_init();
640         fx_save(vcpu->guest_fx_image);
641         fx_restore(vcpu->host_fx_image);
642
643         fx_image = (struct fx_image_s *)vcpu->guest_fx_image;
644         fx_image->mxcsr = 0x1f80;
645         memset(vcpu->guest_fx_image + sizeof(struct fx_image_s),
646                0, FX_IMAGE_SIZE - sizeof(struct fx_image_s));
647 }
648 EXPORT_SYMBOL_GPL(fx_init);
649
650 /*
651  * Allocate some memory and give it an address in the guest physical address
652  * space.
653  *
654  * Discontiguous memory is allowed, mostly for framebuffers.
655  */
656 static int kvm_vm_ioctl_set_memory_region(struct kvm *kvm,
657                                           struct kvm_memory_region *mem)
658 {
659         int r;
660         gfn_t base_gfn;
661         unsigned long npages;
662         unsigned long i;
663         struct kvm_memory_slot *memslot;
664         struct kvm_memory_slot old, new;
665         int memory_config_version;
666
667         r = -EINVAL;
668         /* General sanity checks */
669         if (mem->memory_size & (PAGE_SIZE - 1))
670                 goto out;
671         if (mem->guest_phys_addr & (PAGE_SIZE - 1))
672                 goto out;
673         if (mem->slot >= KVM_MEMORY_SLOTS)
674                 goto out;
675         if (mem->guest_phys_addr + mem->memory_size < mem->guest_phys_addr)
676                 goto out;
677
678         memslot = &kvm->memslots[mem->slot];
679         base_gfn = mem->guest_phys_addr >> PAGE_SHIFT;
680         npages = mem->memory_size >> PAGE_SHIFT;
681
682         if (!npages)
683                 mem->flags &= ~KVM_MEM_LOG_DIRTY_PAGES;
684
685 raced:
686         spin_lock(&kvm->lock);
687
688         memory_config_version = kvm->memory_config_version;
689         new = old = *memslot;
690
691         new.base_gfn = base_gfn;
692         new.npages = npages;
693         new.flags = mem->flags;
694
695         /* Disallow changing a memory slot's size. */
696         r = -EINVAL;
697         if (npages && old.npages && npages != old.npages)
698                 goto out_unlock;
699
700         /* Check for overlaps */
701         r = -EEXIST;
702         for (i = 0; i < KVM_MEMORY_SLOTS; ++i) {
703                 struct kvm_memory_slot *s = &kvm->memslots[i];
704
705                 if (s == memslot)
706                         continue;
707                 if (!((base_gfn + npages <= s->base_gfn) ||
708                       (base_gfn >= s->base_gfn + s->npages)))
709                         goto out_unlock;
710         }
711         /*
712          * Do memory allocations outside lock.  memory_config_version will
713          * detect any races.
714          */
715         spin_unlock(&kvm->lock);
716
717         /* Deallocate if slot is being removed */
718         if (!npages)
719                 new.phys_mem = NULL;
720
721         /* Free page dirty bitmap if unneeded */
722         if (!(new.flags & KVM_MEM_LOG_DIRTY_PAGES))
723                 new.dirty_bitmap = NULL;
724
725         r = -ENOMEM;
726
727         /* Allocate if a slot is being created */
728         if (npages && !new.phys_mem) {
729                 new.phys_mem = vmalloc(npages * sizeof(struct page *));
730
731                 if (!new.phys_mem)
732                         goto out_free;
733
734                 memset(new.phys_mem, 0, npages * sizeof(struct page *));
735                 for (i = 0; i < npages; ++i) {
736                         new.phys_mem[i] = alloc_page(GFP_HIGHUSER
737                                                      | __GFP_ZERO);
738                         if (!new.phys_mem[i])
739                                 goto out_free;
740                         set_page_private(new.phys_mem[i],0);
741                 }
742         }
743
744         /* Allocate page dirty bitmap if needed */
745         if ((new.flags & KVM_MEM_LOG_DIRTY_PAGES) && !new.dirty_bitmap) {
746                 unsigned dirty_bytes = ALIGN(npages, BITS_PER_LONG) / 8;
747
748                 new.dirty_bitmap = vmalloc(dirty_bytes);
749                 if (!new.dirty_bitmap)
750                         goto out_free;
751                 memset(new.dirty_bitmap, 0, dirty_bytes);
752         }
753
754         spin_lock(&kvm->lock);
755
756         if (memory_config_version != kvm->memory_config_version) {
757                 spin_unlock(&kvm->lock);
758                 kvm_free_physmem_slot(&new, &old);
759                 goto raced;
760         }
761
762         r = -EAGAIN;
763         if (kvm->busy)
764                 goto out_unlock;
765
766         if (mem->slot >= kvm->nmemslots)
767                 kvm->nmemslots = mem->slot + 1;
768
769         *memslot = new;
770         ++kvm->memory_config_version;
771
772         kvm_mmu_slot_remove_write_access(kvm, mem->slot);
773         kvm_flush_remote_tlbs(kvm);
774
775         spin_unlock(&kvm->lock);
776
777         kvm_free_physmem_slot(&old, &new);
778         return 0;
779
780 out_unlock:
781         spin_unlock(&kvm->lock);
782 out_free:
783         kvm_free_physmem_slot(&new, &old);
784 out:
785         return r;
786 }
787
788 /*
789  * Get (and clear) the dirty memory log for a memory slot.
790  */
791 static int kvm_vm_ioctl_get_dirty_log(struct kvm *kvm,
792                                       struct kvm_dirty_log *log)
793 {
794         struct kvm_memory_slot *memslot;
795         int r, i;
796         int n;
797         unsigned long any = 0;
798
799         spin_lock(&kvm->lock);
800
801         /*
802          * Prevent changes to guest memory configuration even while the lock
803          * is not taken.
804          */
805         ++kvm->busy;
806         spin_unlock(&kvm->lock);
807         r = -EINVAL;
808         if (log->slot >= KVM_MEMORY_SLOTS)
809                 goto out;
810
811         memslot = &kvm->memslots[log->slot];
812         r = -ENOENT;
813         if (!memslot->dirty_bitmap)
814                 goto out;
815
816         n = ALIGN(memslot->npages, BITS_PER_LONG) / 8;
817
818         for (i = 0; !any && i < n/sizeof(long); ++i)
819                 any = memslot->dirty_bitmap[i];
820
821         r = -EFAULT;
822         if (copy_to_user(log->dirty_bitmap, memslot->dirty_bitmap, n))
823                 goto out;
824
825         spin_lock(&kvm->lock);
826         kvm_mmu_slot_remove_write_access(kvm, log->slot);
827         kvm_flush_remote_tlbs(kvm);
828         memset(memslot->dirty_bitmap, 0, n);
829         spin_unlock(&kvm->lock);
830
831         r = 0;
832
833 out:
834         spin_lock(&kvm->lock);
835         --kvm->busy;
836         spin_unlock(&kvm->lock);
837         return r;
838 }
839
840 /*
841  * Set a new alias region.  Aliases map a portion of physical memory into
842  * another portion.  This is useful for memory windows, for example the PC
843  * VGA region.
844  */
845 static int kvm_vm_ioctl_set_memory_alias(struct kvm *kvm,
846                                          struct kvm_memory_alias *alias)
847 {
848         int r, n;
849         struct kvm_mem_alias *p;
850
851         r = -EINVAL;
852         /* General sanity checks */
853         if (alias->memory_size & (PAGE_SIZE - 1))
854                 goto out;
855         if (alias->guest_phys_addr & (PAGE_SIZE - 1))
856                 goto out;
857         if (alias->slot >= KVM_ALIAS_SLOTS)
858                 goto out;
859         if (alias->guest_phys_addr + alias->memory_size
860             < alias->guest_phys_addr)
861                 goto out;
862         if (alias->target_phys_addr + alias->memory_size
863             < alias->target_phys_addr)
864                 goto out;
865
866         spin_lock(&kvm->lock);
867
868         p = &kvm->aliases[alias->slot];
869         p->base_gfn = alias->guest_phys_addr >> PAGE_SHIFT;
870         p->npages = alias->memory_size >> PAGE_SHIFT;
871         p->target_gfn = alias->target_phys_addr >> PAGE_SHIFT;
872
873         for (n = KVM_ALIAS_SLOTS; n > 0; --n)
874                 if (kvm->aliases[n - 1].npages)
875                         break;
876         kvm->naliases = n;
877
878         kvm_mmu_zap_all(kvm);
879
880         spin_unlock(&kvm->lock);
881
882         return 0;
883
884 out:
885         return r;
886 }
887
888 static gfn_t unalias_gfn(struct kvm *kvm, gfn_t gfn)
889 {
890         int i;
891         struct kvm_mem_alias *alias;
892
893         for (i = 0; i < kvm->naliases; ++i) {
894                 alias = &kvm->aliases[i];
895                 if (gfn >= alias->base_gfn
896                     && gfn < alias->base_gfn + alias->npages)
897                         return alias->target_gfn + gfn - alias->base_gfn;
898         }
899         return gfn;
900 }
901
902 static struct kvm_memory_slot *__gfn_to_memslot(struct kvm *kvm, gfn_t gfn)
903 {
904         int i;
905
906         for (i = 0; i < kvm->nmemslots; ++i) {
907                 struct kvm_memory_slot *memslot = &kvm->memslots[i];
908
909                 if (gfn >= memslot->base_gfn
910                     && gfn < memslot->base_gfn + memslot->npages)
911                         return memslot;
912         }
913         return NULL;
914 }
915
916 struct kvm_memory_slot *gfn_to_memslot(struct kvm *kvm, gfn_t gfn)
917 {
918         gfn = unalias_gfn(kvm, gfn);
919         return __gfn_to_memslot(kvm, gfn);
920 }
921
922 struct page *gfn_to_page(struct kvm *kvm, gfn_t gfn)
923 {
924         struct kvm_memory_slot *slot;
925
926         gfn = unalias_gfn(kvm, gfn);
927         slot = __gfn_to_memslot(kvm, gfn);
928         if (!slot)
929                 return NULL;
930         return slot->phys_mem[gfn - slot->base_gfn];
931 }
932 EXPORT_SYMBOL_GPL(gfn_to_page);
933
934 void mark_page_dirty(struct kvm *kvm, gfn_t gfn)
935 {
936         int i;
937         struct kvm_memory_slot *memslot;
938         unsigned long rel_gfn;
939
940         for (i = 0; i < kvm->nmemslots; ++i) {
941                 memslot = &kvm->memslots[i];
942
943                 if (gfn >= memslot->base_gfn
944                     && gfn < memslot->base_gfn + memslot->npages) {
945
946                         if (!memslot->dirty_bitmap)
947                                 return;
948
949                         rel_gfn = gfn - memslot->base_gfn;
950
951                         /* avoid RMW */
952                         if (!test_bit(rel_gfn, memslot->dirty_bitmap))
953                                 set_bit(rel_gfn, memslot->dirty_bitmap);
954                         return;
955                 }
956         }
957 }
958
959 static int emulator_read_std(unsigned long addr,
960                              void *val,
961                              unsigned int bytes,
962                              struct x86_emulate_ctxt *ctxt)
963 {
964         struct kvm_vcpu *vcpu = ctxt->vcpu;
965         void *data = val;
966
967         while (bytes) {
968                 gpa_t gpa = vcpu->mmu.gva_to_gpa(vcpu, addr);
969                 unsigned offset = addr & (PAGE_SIZE-1);
970                 unsigned tocopy = min(bytes, (unsigned)PAGE_SIZE - offset);
971                 unsigned long pfn;
972                 struct page *page;
973                 void *page_virt;
974
975                 if (gpa == UNMAPPED_GVA)
976                         return X86EMUL_PROPAGATE_FAULT;
977                 pfn = gpa >> PAGE_SHIFT;
978                 page = gfn_to_page(vcpu->kvm, pfn);
979                 if (!page)
980                         return X86EMUL_UNHANDLEABLE;
981                 page_virt = kmap_atomic(page, KM_USER0);
982
983                 memcpy(data, page_virt + offset, tocopy);
984
985                 kunmap_atomic(page_virt, KM_USER0);
986
987                 bytes -= tocopy;
988                 data += tocopy;
989                 addr += tocopy;
990         }
991
992         return X86EMUL_CONTINUE;
993 }
994
995 static int emulator_write_std(unsigned long addr,
996                               const void *val,
997                               unsigned int bytes,
998                               struct x86_emulate_ctxt *ctxt)
999 {
1000         printk(KERN_ERR "emulator_write_std: addr %lx n %d\n",
1001                addr, bytes);
1002         return X86EMUL_UNHANDLEABLE;
1003 }
1004
1005 static struct kvm_io_device *vcpu_find_mmio_dev(struct kvm_vcpu *vcpu,
1006                                                 gpa_t addr)
1007 {
1008         /*
1009          * Note that its important to have this wrapper function because
1010          * in the very near future we will be checking for MMIOs against
1011          * the LAPIC as well as the general MMIO bus
1012          */
1013         return kvm_io_bus_find_dev(&vcpu->kvm->mmio_bus, addr);
1014 }
1015
1016 static struct kvm_io_device *vcpu_find_pio_dev(struct kvm_vcpu *vcpu,
1017                                                gpa_t addr)
1018 {
1019         return kvm_io_bus_find_dev(&vcpu->kvm->pio_bus, addr);
1020 }
1021
1022 static int emulator_read_emulated(unsigned long addr,
1023                                   void *val,
1024                                   unsigned int bytes,
1025                                   struct x86_emulate_ctxt *ctxt)
1026 {
1027         struct kvm_vcpu      *vcpu = ctxt->vcpu;
1028         struct kvm_io_device *mmio_dev;
1029         gpa_t                 gpa;
1030
1031         if (vcpu->mmio_read_completed) {
1032                 memcpy(val, vcpu->mmio_data, bytes);
1033                 vcpu->mmio_read_completed = 0;
1034                 return X86EMUL_CONTINUE;
1035         } else if (emulator_read_std(addr, val, bytes, ctxt)
1036                    == X86EMUL_CONTINUE)
1037                 return X86EMUL_CONTINUE;
1038
1039         gpa = vcpu->mmu.gva_to_gpa(vcpu, addr);
1040         if (gpa == UNMAPPED_GVA)
1041                 return X86EMUL_PROPAGATE_FAULT;
1042
1043         /*
1044          * Is this MMIO handled locally?
1045          */
1046         mmio_dev = vcpu_find_mmio_dev(vcpu, gpa);
1047         if (mmio_dev) {
1048                 kvm_iodevice_read(mmio_dev, gpa, bytes, val);
1049                 return X86EMUL_CONTINUE;
1050         }
1051
1052         vcpu->mmio_needed = 1;
1053         vcpu->mmio_phys_addr = gpa;
1054         vcpu->mmio_size = bytes;
1055         vcpu->mmio_is_write = 0;
1056
1057         return X86EMUL_UNHANDLEABLE;
1058 }
1059
1060 static int emulator_write_phys(struct kvm_vcpu *vcpu, gpa_t gpa,
1061                                const void *val, int bytes)
1062 {
1063         struct page *page;
1064         void *virt;
1065         unsigned offset = offset_in_page(gpa);
1066
1067         if (((gpa + bytes - 1) >> PAGE_SHIFT) != (gpa >> PAGE_SHIFT))
1068                 return 0;
1069         page = gfn_to_page(vcpu->kvm, gpa >> PAGE_SHIFT);
1070         if (!page)
1071                 return 0;
1072         mark_page_dirty(vcpu->kvm, gpa >> PAGE_SHIFT);
1073         virt = kmap_atomic(page, KM_USER0);
1074         kvm_mmu_pte_write(vcpu, gpa, virt + offset, val, bytes);
1075         memcpy(virt + offset_in_page(gpa), val, bytes);
1076         kunmap_atomic(virt, KM_USER0);
1077         return 1;
1078 }
1079
1080 static int emulator_write_emulated_onepage(unsigned long addr,
1081                                            const void *val,
1082                                            unsigned int bytes,
1083                                            struct x86_emulate_ctxt *ctxt)
1084 {
1085         struct kvm_vcpu      *vcpu = ctxt->vcpu;
1086         struct kvm_io_device *mmio_dev;
1087         gpa_t                 gpa = vcpu->mmu.gva_to_gpa(vcpu, addr);
1088
1089         if (gpa == UNMAPPED_GVA) {
1090                 kvm_arch_ops->inject_page_fault(vcpu, addr, 2);
1091                 return X86EMUL_PROPAGATE_FAULT;
1092         }
1093
1094         if (emulator_write_phys(vcpu, gpa, val, bytes))
1095                 return X86EMUL_CONTINUE;
1096
1097         /*
1098          * Is this MMIO handled locally?
1099          */
1100         mmio_dev = vcpu_find_mmio_dev(vcpu, gpa);
1101         if (mmio_dev) {
1102                 kvm_iodevice_write(mmio_dev, gpa, bytes, val);
1103                 return X86EMUL_CONTINUE;
1104         }
1105
1106         vcpu->mmio_needed = 1;
1107         vcpu->mmio_phys_addr = gpa;
1108         vcpu->mmio_size = bytes;
1109         vcpu->mmio_is_write = 1;
1110         memcpy(vcpu->mmio_data, val, bytes);
1111
1112         return X86EMUL_CONTINUE;
1113 }
1114
1115 static int emulator_write_emulated(unsigned long addr,
1116                                    const void *val,
1117                                    unsigned int bytes,
1118                                    struct x86_emulate_ctxt *ctxt)
1119 {
1120         /* Crossing a page boundary? */
1121         if (((addr + bytes - 1) ^ addr) & PAGE_MASK) {
1122                 int rc, now;
1123
1124                 now = -addr & ~PAGE_MASK;
1125                 rc = emulator_write_emulated_onepage(addr, val, now, ctxt);
1126                 if (rc != X86EMUL_CONTINUE)
1127                         return rc;
1128                 addr += now;
1129                 val += now;
1130                 bytes -= now;
1131         }
1132         return emulator_write_emulated_onepage(addr, val, bytes, ctxt);
1133 }
1134
1135 static int emulator_cmpxchg_emulated(unsigned long addr,
1136                                      const void *old,
1137                                      const void *new,
1138                                      unsigned int bytes,
1139                                      struct x86_emulate_ctxt *ctxt)
1140 {
1141         static int reported;
1142
1143         if (!reported) {
1144                 reported = 1;
1145                 printk(KERN_WARNING "kvm: emulating exchange as write\n");
1146         }
1147         return emulator_write_emulated(addr, new, bytes, ctxt);
1148 }
1149
1150 static unsigned long get_segment_base(struct kvm_vcpu *vcpu, int seg)
1151 {
1152         return kvm_arch_ops->get_segment_base(vcpu, seg);
1153 }
1154
1155 int emulate_invlpg(struct kvm_vcpu *vcpu, gva_t address)
1156 {
1157         return X86EMUL_CONTINUE;
1158 }
1159
1160 int emulate_clts(struct kvm_vcpu *vcpu)
1161 {
1162         unsigned long cr0;
1163
1164         cr0 = vcpu->cr0 & ~X86_CR0_TS;
1165         kvm_arch_ops->set_cr0(vcpu, cr0);
1166         return X86EMUL_CONTINUE;
1167 }
1168
1169 int emulator_get_dr(struct x86_emulate_ctxt* ctxt, int dr, unsigned long *dest)
1170 {
1171         struct kvm_vcpu *vcpu = ctxt->vcpu;
1172
1173         switch (dr) {
1174         case 0 ... 3:
1175                 *dest = kvm_arch_ops->get_dr(vcpu, dr);
1176                 return X86EMUL_CONTINUE;
1177         default:
1178                 printk(KERN_DEBUG "%s: unexpected dr %u\n",
1179                        __FUNCTION__, dr);
1180                 return X86EMUL_UNHANDLEABLE;
1181         }
1182 }
1183
1184 int emulator_set_dr(struct x86_emulate_ctxt *ctxt, int dr, unsigned long value)
1185 {
1186         unsigned long mask = (ctxt->mode == X86EMUL_MODE_PROT64) ? ~0ULL : ~0U;
1187         int exception;
1188
1189         kvm_arch_ops->set_dr(ctxt->vcpu, dr, value & mask, &exception);
1190         if (exception) {
1191                 /* FIXME: better handling */
1192                 return X86EMUL_UNHANDLEABLE;
1193         }
1194         return X86EMUL_CONTINUE;
1195 }
1196
1197 static void report_emulation_failure(struct x86_emulate_ctxt *ctxt)
1198 {
1199         static int reported;
1200         u8 opcodes[4];
1201         unsigned long rip = ctxt->vcpu->rip;
1202         unsigned long rip_linear;
1203
1204         rip_linear = rip + get_segment_base(ctxt->vcpu, VCPU_SREG_CS);
1205
1206         if (reported)
1207                 return;
1208
1209         emulator_read_std(rip_linear, (void *)opcodes, 4, ctxt);
1210
1211         printk(KERN_ERR "emulation failed but !mmio_needed?"
1212                " rip %lx %02x %02x %02x %02x\n",
1213                rip, opcodes[0], opcodes[1], opcodes[2], opcodes[3]);
1214         reported = 1;
1215 }
1216
1217 struct x86_emulate_ops emulate_ops = {
1218         .read_std            = emulator_read_std,
1219         .write_std           = emulator_write_std,
1220         .read_emulated       = emulator_read_emulated,
1221         .write_emulated      = emulator_write_emulated,
1222         .cmpxchg_emulated    = emulator_cmpxchg_emulated,
1223 };
1224
1225 int emulate_instruction(struct kvm_vcpu *vcpu,
1226                         struct kvm_run *run,
1227                         unsigned long cr2,
1228                         u16 error_code)
1229 {
1230         struct x86_emulate_ctxt emulate_ctxt;
1231         int r;
1232         int cs_db, cs_l;
1233
1234         vcpu->mmio_fault_cr2 = cr2;
1235         kvm_arch_ops->cache_regs(vcpu);
1236
1237         kvm_arch_ops->get_cs_db_l_bits(vcpu, &cs_db, &cs_l);
1238
1239         emulate_ctxt.vcpu = vcpu;
1240         emulate_ctxt.eflags = kvm_arch_ops->get_rflags(vcpu);
1241         emulate_ctxt.cr2 = cr2;
1242         emulate_ctxt.mode = (emulate_ctxt.eflags & X86_EFLAGS_VM)
1243                 ? X86EMUL_MODE_REAL : cs_l
1244                 ? X86EMUL_MODE_PROT64 : cs_db
1245                 ? X86EMUL_MODE_PROT32 : X86EMUL_MODE_PROT16;
1246
1247         if (emulate_ctxt.mode == X86EMUL_MODE_PROT64) {
1248                 emulate_ctxt.cs_base = 0;
1249                 emulate_ctxt.ds_base = 0;
1250                 emulate_ctxt.es_base = 0;
1251                 emulate_ctxt.ss_base = 0;
1252         } else {
1253                 emulate_ctxt.cs_base = get_segment_base(vcpu, VCPU_SREG_CS);
1254                 emulate_ctxt.ds_base = get_segment_base(vcpu, VCPU_SREG_DS);
1255                 emulate_ctxt.es_base = get_segment_base(vcpu, VCPU_SREG_ES);
1256                 emulate_ctxt.ss_base = get_segment_base(vcpu, VCPU_SREG_SS);
1257         }
1258
1259         emulate_ctxt.gs_base = get_segment_base(vcpu, VCPU_SREG_GS);
1260         emulate_ctxt.fs_base = get_segment_base(vcpu, VCPU_SREG_FS);
1261
1262         vcpu->mmio_is_write = 0;
1263         r = x86_emulate_memop(&emulate_ctxt, &emulate_ops);
1264
1265         if ((r || vcpu->mmio_is_write) && run) {
1266                 run->mmio.phys_addr = vcpu->mmio_phys_addr;
1267                 memcpy(run->mmio.data, vcpu->mmio_data, 8);
1268                 run->mmio.len = vcpu->mmio_size;
1269                 run->mmio.is_write = vcpu->mmio_is_write;
1270         }
1271
1272         if (r) {
1273                 if (kvm_mmu_unprotect_page_virt(vcpu, cr2))
1274                         return EMULATE_DONE;
1275                 if (!vcpu->mmio_needed) {
1276                         report_emulation_failure(&emulate_ctxt);
1277                         return EMULATE_FAIL;
1278                 }
1279                 return EMULATE_DO_MMIO;
1280         }
1281
1282         kvm_arch_ops->decache_regs(vcpu);
1283         kvm_arch_ops->set_rflags(vcpu, emulate_ctxt.eflags);
1284
1285         if (vcpu->mmio_is_write) {
1286                 vcpu->mmio_needed = 0;
1287                 return EMULATE_DO_MMIO;
1288         }
1289
1290         return EMULATE_DONE;
1291 }
1292 EXPORT_SYMBOL_GPL(emulate_instruction);
1293
1294 int kvm_emulate_halt(struct kvm_vcpu *vcpu)
1295 {
1296         if (vcpu->irq_summary)
1297                 return 1;
1298
1299         vcpu->run->exit_reason = KVM_EXIT_HLT;
1300         ++vcpu->stat.halt_exits;
1301         return 0;
1302 }
1303 EXPORT_SYMBOL_GPL(kvm_emulate_halt);
1304
1305 int kvm_hypercall(struct kvm_vcpu *vcpu, struct kvm_run *run)
1306 {
1307         unsigned long nr, a0, a1, a2, a3, a4, a5, ret;
1308
1309         kvm_arch_ops->cache_regs(vcpu);
1310         ret = -KVM_EINVAL;
1311 #ifdef CONFIG_X86_64
1312         if (is_long_mode(vcpu)) {
1313                 nr = vcpu->regs[VCPU_REGS_RAX];
1314                 a0 = vcpu->regs[VCPU_REGS_RDI];
1315                 a1 = vcpu->regs[VCPU_REGS_RSI];
1316                 a2 = vcpu->regs[VCPU_REGS_RDX];
1317                 a3 = vcpu->regs[VCPU_REGS_RCX];
1318                 a4 = vcpu->regs[VCPU_REGS_R8];
1319                 a5 = vcpu->regs[VCPU_REGS_R9];
1320         } else
1321 #endif
1322         {
1323                 nr = vcpu->regs[VCPU_REGS_RBX] & -1u;
1324                 a0 = vcpu->regs[VCPU_REGS_RAX] & -1u;
1325                 a1 = vcpu->regs[VCPU_REGS_RCX] & -1u;
1326                 a2 = vcpu->regs[VCPU_REGS_RDX] & -1u;
1327                 a3 = vcpu->regs[VCPU_REGS_RSI] & -1u;
1328                 a4 = vcpu->regs[VCPU_REGS_RDI] & -1u;
1329                 a5 = vcpu->regs[VCPU_REGS_RBP] & -1u;
1330         }
1331         switch (nr) {
1332         default:
1333                 run->hypercall.args[0] = a0;
1334                 run->hypercall.args[1] = a1;
1335                 run->hypercall.args[2] = a2;
1336                 run->hypercall.args[3] = a3;
1337                 run->hypercall.args[4] = a4;
1338                 run->hypercall.args[5] = a5;
1339                 run->hypercall.ret = ret;
1340                 run->hypercall.longmode = is_long_mode(vcpu);
1341                 kvm_arch_ops->decache_regs(vcpu);
1342                 return 0;
1343         }
1344         vcpu->regs[VCPU_REGS_RAX] = ret;
1345         kvm_arch_ops->decache_regs(vcpu);
1346         return 1;
1347 }
1348 EXPORT_SYMBOL_GPL(kvm_hypercall);
1349
1350 static u64 mk_cr_64(u64 curr_cr, u32 new_val)
1351 {
1352         return (curr_cr & ~((1ULL << 32) - 1)) | new_val;
1353 }
1354
1355 void realmode_lgdt(struct kvm_vcpu *vcpu, u16 limit, unsigned long base)
1356 {
1357         struct descriptor_table dt = { limit, base };
1358
1359         kvm_arch_ops->set_gdt(vcpu, &dt);
1360 }
1361
1362 void realmode_lidt(struct kvm_vcpu *vcpu, u16 limit, unsigned long base)
1363 {
1364         struct descriptor_table dt = { limit, base };
1365
1366         kvm_arch_ops->set_idt(vcpu, &dt);
1367 }
1368
1369 void realmode_lmsw(struct kvm_vcpu *vcpu, unsigned long msw,
1370                    unsigned long *rflags)
1371 {
1372         lmsw(vcpu, msw);
1373         *rflags = kvm_arch_ops->get_rflags(vcpu);
1374 }
1375
1376 unsigned long realmode_get_cr(struct kvm_vcpu *vcpu, int cr)
1377 {
1378         kvm_arch_ops->decache_cr4_guest_bits(vcpu);
1379         switch (cr) {
1380         case 0:
1381                 return vcpu->cr0;
1382         case 2:
1383                 return vcpu->cr2;
1384         case 3:
1385                 return vcpu->cr3;
1386         case 4:
1387                 return vcpu->cr4;
1388         default:
1389                 vcpu_printf(vcpu, "%s: unexpected cr %u\n", __FUNCTION__, cr);
1390                 return 0;
1391         }
1392 }
1393
1394 void realmode_set_cr(struct kvm_vcpu *vcpu, int cr, unsigned long val,
1395                      unsigned long *rflags)
1396 {
1397         switch (cr) {
1398         case 0:
1399                 set_cr0(vcpu, mk_cr_64(vcpu->cr0, val));
1400                 *rflags = kvm_arch_ops->get_rflags(vcpu);
1401                 break;
1402         case 2:
1403                 vcpu->cr2 = val;
1404                 break;
1405         case 3:
1406                 set_cr3(vcpu, val);
1407                 break;
1408         case 4:
1409                 set_cr4(vcpu, mk_cr_64(vcpu->cr4, val));
1410                 break;
1411         default:
1412                 vcpu_printf(vcpu, "%s: unexpected cr %u\n", __FUNCTION__, cr);
1413         }
1414 }
1415
1416 /*
1417  * Register the para guest with the host:
1418  */
1419 static int vcpu_register_para(struct kvm_vcpu *vcpu, gpa_t para_state_gpa)
1420 {
1421         struct kvm_vcpu_para_state *para_state;
1422         hpa_t para_state_hpa, hypercall_hpa;
1423         struct page *para_state_page;
1424         unsigned char *hypercall;
1425         gpa_t hypercall_gpa;
1426
1427         printk(KERN_DEBUG "kvm: guest trying to enter paravirtual mode\n");
1428         printk(KERN_DEBUG ".... para_state_gpa: %08Lx\n", para_state_gpa);
1429
1430         /*
1431          * Needs to be page aligned:
1432          */
1433         if (para_state_gpa != PAGE_ALIGN(para_state_gpa))
1434                 goto err_gp;
1435
1436         para_state_hpa = gpa_to_hpa(vcpu, para_state_gpa);
1437         printk(KERN_DEBUG ".... para_state_hpa: %08Lx\n", para_state_hpa);
1438         if (is_error_hpa(para_state_hpa))
1439                 goto err_gp;
1440
1441         mark_page_dirty(vcpu->kvm, para_state_gpa >> PAGE_SHIFT);
1442         para_state_page = pfn_to_page(para_state_hpa >> PAGE_SHIFT);
1443         para_state = kmap_atomic(para_state_page, KM_USER0);
1444
1445         printk(KERN_DEBUG "....  guest version: %d\n", para_state->guest_version);
1446         printk(KERN_DEBUG "....           size: %d\n", para_state->size);
1447
1448         para_state->host_version = KVM_PARA_API_VERSION;
1449         /*
1450          * We cannot support guests that try to register themselves
1451          * with a newer API version than the host supports:
1452          */
1453         if (para_state->guest_version > KVM_PARA_API_VERSION) {
1454                 para_state->ret = -KVM_EINVAL;
1455                 goto err_kunmap_skip;
1456         }
1457
1458         hypercall_gpa = para_state->hypercall_gpa;
1459         hypercall_hpa = gpa_to_hpa(vcpu, hypercall_gpa);
1460         printk(KERN_DEBUG ".... hypercall_hpa: %08Lx\n", hypercall_hpa);
1461         if (is_error_hpa(hypercall_hpa)) {
1462                 para_state->ret = -KVM_EINVAL;
1463                 goto err_kunmap_skip;
1464         }
1465
1466         printk(KERN_DEBUG "kvm: para guest successfully registered.\n");
1467         vcpu->para_state_page = para_state_page;
1468         vcpu->para_state_gpa = para_state_gpa;
1469         vcpu->hypercall_gpa = hypercall_gpa;
1470
1471         mark_page_dirty(vcpu->kvm, hypercall_gpa >> PAGE_SHIFT);
1472         hypercall = kmap_atomic(pfn_to_page(hypercall_hpa >> PAGE_SHIFT),
1473                                 KM_USER1) + (hypercall_hpa & ~PAGE_MASK);
1474         kvm_arch_ops->patch_hypercall(vcpu, hypercall);
1475         kunmap_atomic(hypercall, KM_USER1);
1476
1477         para_state->ret = 0;
1478 err_kunmap_skip:
1479         kunmap_atomic(para_state, KM_USER0);
1480         return 0;
1481 err_gp:
1482         return 1;
1483 }
1484
1485 int kvm_get_msr_common(struct kvm_vcpu *vcpu, u32 msr, u64 *pdata)
1486 {
1487         u64 data;
1488
1489         switch (msr) {
1490         case 0xc0010010: /* SYSCFG */
1491         case 0xc0010015: /* HWCR */
1492         case MSR_IA32_PLATFORM_ID:
1493         case MSR_IA32_P5_MC_ADDR:
1494         case MSR_IA32_P5_MC_TYPE:
1495         case MSR_IA32_MC0_CTL:
1496         case MSR_IA32_MCG_STATUS:
1497         case MSR_IA32_MCG_CAP:
1498         case MSR_IA32_MC0_MISC:
1499         case MSR_IA32_MC0_MISC+4:
1500         case MSR_IA32_MC0_MISC+8:
1501         case MSR_IA32_MC0_MISC+12:
1502         case MSR_IA32_MC0_MISC+16:
1503         case MSR_IA32_UCODE_REV:
1504         case MSR_IA32_PERF_STATUS:
1505         case MSR_IA32_EBL_CR_POWERON:
1506                 /* MTRR registers */
1507         case 0xfe:
1508         case 0x200 ... 0x2ff:
1509                 data = 0;
1510                 break;
1511         case 0xcd: /* fsb frequency */
1512                 data = 3;
1513                 break;
1514         case MSR_IA32_APICBASE:
1515                 data = vcpu->apic_base;
1516                 break;
1517         case MSR_IA32_MISC_ENABLE:
1518                 data = vcpu->ia32_misc_enable_msr;
1519                 break;
1520 #ifdef CONFIG_X86_64
1521         case MSR_EFER:
1522                 data = vcpu->shadow_efer;
1523                 break;
1524 #endif
1525         default:
1526                 printk(KERN_ERR "kvm: unhandled rdmsr: 0x%x\n", msr);
1527                 return 1;
1528         }
1529         *pdata = data;
1530         return 0;
1531 }
1532 EXPORT_SYMBOL_GPL(kvm_get_msr_common);
1533
1534 /*
1535  * Reads an msr value (of 'msr_index') into 'pdata'.
1536  * Returns 0 on success, non-0 otherwise.
1537  * Assumes vcpu_load() was already called.
1538  */
1539 int kvm_get_msr(struct kvm_vcpu *vcpu, u32 msr_index, u64 *pdata)
1540 {
1541         return kvm_arch_ops->get_msr(vcpu, msr_index, pdata);
1542 }
1543
1544 #ifdef CONFIG_X86_64
1545
1546 static void set_efer(struct kvm_vcpu *vcpu, u64 efer)
1547 {
1548         if (efer & EFER_RESERVED_BITS) {
1549                 printk(KERN_DEBUG "set_efer: 0x%llx #GP, reserved bits\n",
1550                        efer);
1551                 inject_gp(vcpu);
1552                 return;
1553         }
1554
1555         if (is_paging(vcpu)
1556             && (vcpu->shadow_efer & EFER_LME) != (efer & EFER_LME)) {
1557                 printk(KERN_DEBUG "set_efer: #GP, change LME while paging\n");
1558                 inject_gp(vcpu);
1559                 return;
1560         }
1561
1562         kvm_arch_ops->set_efer(vcpu, efer);
1563
1564         efer &= ~EFER_LMA;
1565         efer |= vcpu->shadow_efer & EFER_LMA;
1566
1567         vcpu->shadow_efer = efer;
1568 }
1569
1570 #endif
1571
1572 int kvm_set_msr_common(struct kvm_vcpu *vcpu, u32 msr, u64 data)
1573 {
1574         switch (msr) {
1575 #ifdef CONFIG_X86_64
1576         case MSR_EFER:
1577                 set_efer(vcpu, data);
1578                 break;
1579 #endif
1580         case MSR_IA32_MC0_STATUS:
1581                 printk(KERN_WARNING "%s: MSR_IA32_MC0_STATUS 0x%llx, nop\n",
1582                        __FUNCTION__, data);
1583                 break;
1584         case MSR_IA32_MCG_STATUS:
1585                 printk(KERN_WARNING "%s: MSR_IA32_MCG_STATUS 0x%llx, nop\n",
1586                         __FUNCTION__, data);
1587                 break;
1588         case MSR_IA32_UCODE_REV:
1589         case MSR_IA32_UCODE_WRITE:
1590         case 0x200 ... 0x2ff: /* MTRRs */
1591                 break;
1592         case MSR_IA32_APICBASE:
1593                 vcpu->apic_base = data;
1594                 break;
1595         case MSR_IA32_MISC_ENABLE:
1596                 vcpu->ia32_misc_enable_msr = data;
1597                 break;
1598         /*
1599          * This is the 'probe whether the host is KVM' logic:
1600          */
1601         case MSR_KVM_API_MAGIC:
1602                 return vcpu_register_para(vcpu, data);
1603
1604         default:
1605                 printk(KERN_ERR "kvm: unhandled wrmsr: 0x%x\n", msr);
1606                 return 1;
1607         }
1608         return 0;
1609 }
1610 EXPORT_SYMBOL_GPL(kvm_set_msr_common);
1611
1612 /*
1613  * Writes msr value into into the appropriate "register".
1614  * Returns 0 on success, non-0 otherwise.
1615  * Assumes vcpu_load() was already called.
1616  */
1617 int kvm_set_msr(struct kvm_vcpu *vcpu, u32 msr_index, u64 data)
1618 {
1619         return kvm_arch_ops->set_msr(vcpu, msr_index, data);
1620 }
1621
1622 void kvm_resched(struct kvm_vcpu *vcpu)
1623 {
1624         if (!need_resched())
1625                 return;
1626         vcpu_put(vcpu);
1627         cond_resched();
1628         vcpu_load(vcpu);
1629 }
1630 EXPORT_SYMBOL_GPL(kvm_resched);
1631
1632 void load_msrs(struct vmx_msr_entry *e, int n)
1633 {
1634         int i;
1635
1636         for (i = 0; i < n; ++i)
1637                 wrmsrl(e[i].index, e[i].data);
1638 }
1639 EXPORT_SYMBOL_GPL(load_msrs);
1640
1641 void save_msrs(struct vmx_msr_entry *e, int n)
1642 {
1643         int i;
1644
1645         for (i = 0; i < n; ++i)
1646                 rdmsrl(e[i].index, e[i].data);
1647 }
1648 EXPORT_SYMBOL_GPL(save_msrs);
1649
1650 void kvm_emulate_cpuid(struct kvm_vcpu *vcpu)
1651 {
1652         int i;
1653         u32 function;
1654         struct kvm_cpuid_entry *e, *best;
1655
1656         kvm_arch_ops->cache_regs(vcpu);
1657         function = vcpu->regs[VCPU_REGS_RAX];
1658         vcpu->regs[VCPU_REGS_RAX] = 0;
1659         vcpu->regs[VCPU_REGS_RBX] = 0;
1660         vcpu->regs[VCPU_REGS_RCX] = 0;
1661         vcpu->regs[VCPU_REGS_RDX] = 0;
1662         best = NULL;
1663         for (i = 0; i < vcpu->cpuid_nent; ++i) {
1664                 e = &vcpu->cpuid_entries[i];
1665                 if (e->function == function) {
1666                         best = e;
1667                         break;
1668                 }
1669                 /*
1670                  * Both basic or both extended?
1671                  */
1672                 if (((e->function ^ function) & 0x80000000) == 0)
1673                         if (!best || e->function > best->function)
1674                                 best = e;
1675         }
1676         if (best) {
1677                 vcpu->regs[VCPU_REGS_RAX] = best->eax;
1678                 vcpu->regs[VCPU_REGS_RBX] = best->ebx;
1679                 vcpu->regs[VCPU_REGS_RCX] = best->ecx;
1680                 vcpu->regs[VCPU_REGS_RDX] = best->edx;
1681         }
1682         kvm_arch_ops->decache_regs(vcpu);
1683         kvm_arch_ops->skip_emulated_instruction(vcpu);
1684 }
1685 EXPORT_SYMBOL_GPL(kvm_emulate_cpuid);
1686
1687 static int pio_copy_data(struct kvm_vcpu *vcpu)
1688 {
1689         void *p = vcpu->pio_data;
1690         void *q;
1691         unsigned bytes;
1692         int nr_pages = vcpu->pio.guest_pages[1] ? 2 : 1;
1693
1694         kvm_arch_ops->vcpu_put(vcpu);
1695         q = vmap(vcpu->pio.guest_pages, nr_pages, VM_READ|VM_WRITE,
1696                  PAGE_KERNEL);
1697         if (!q) {
1698                 kvm_arch_ops->vcpu_load(vcpu);
1699                 free_pio_guest_pages(vcpu);
1700                 return -ENOMEM;
1701         }
1702         q += vcpu->pio.guest_page_offset;
1703         bytes = vcpu->pio.size * vcpu->pio.cur_count;
1704         if (vcpu->pio.in)
1705                 memcpy(q, p, bytes);
1706         else
1707                 memcpy(p, q, bytes);
1708         q -= vcpu->pio.guest_page_offset;
1709         vunmap(q);
1710         kvm_arch_ops->vcpu_load(vcpu);
1711         free_pio_guest_pages(vcpu);
1712         return 0;
1713 }
1714
1715 static int complete_pio(struct kvm_vcpu *vcpu)
1716 {
1717         struct kvm_pio_request *io = &vcpu->pio;
1718         long delta;
1719         int r;
1720
1721         kvm_arch_ops->cache_regs(vcpu);
1722
1723         if (!io->string) {
1724                 if (io->in)
1725                         memcpy(&vcpu->regs[VCPU_REGS_RAX], vcpu->pio_data,
1726                                io->size);
1727         } else {
1728                 if (io->in) {
1729                         r = pio_copy_data(vcpu);
1730                         if (r) {
1731                                 kvm_arch_ops->cache_regs(vcpu);
1732                                 return r;
1733                         }
1734                 }
1735
1736                 delta = 1;
1737                 if (io->rep) {
1738                         delta *= io->cur_count;
1739                         /*
1740                          * The size of the register should really depend on
1741                          * current address size.
1742                          */
1743                         vcpu->regs[VCPU_REGS_RCX] -= delta;
1744                 }
1745                 if (io->down)
1746                         delta = -delta;
1747                 delta *= io->size;
1748                 if (io->in)
1749                         vcpu->regs[VCPU_REGS_RDI] += delta;
1750                 else
1751                         vcpu->regs[VCPU_REGS_RSI] += delta;
1752         }
1753
1754         kvm_arch_ops->decache_regs(vcpu);
1755
1756         io->count -= io->cur_count;
1757         io->cur_count = 0;
1758
1759         if (!io->count)
1760                 kvm_arch_ops->skip_emulated_instruction(vcpu);
1761         return 0;
1762 }
1763
1764 static void kernel_pio(struct kvm_io_device *pio_dev,
1765                        struct kvm_vcpu *vcpu,
1766                        void *pd)
1767 {
1768         /* TODO: String I/O for in kernel device */
1769
1770         if (vcpu->pio.in)
1771                 kvm_iodevice_read(pio_dev, vcpu->pio.port,
1772                                   vcpu->pio.size,
1773                                   pd);
1774         else
1775                 kvm_iodevice_write(pio_dev, vcpu->pio.port,
1776                                    vcpu->pio.size,
1777                                    pd);
1778 }
1779
1780 static void pio_string_write(struct kvm_io_device *pio_dev,
1781                              struct kvm_vcpu *vcpu)
1782 {
1783         struct kvm_pio_request *io = &vcpu->pio;
1784         void *pd = vcpu->pio_data;
1785         int i;
1786
1787         for (i = 0; i < io->cur_count; i++) {
1788                 kvm_iodevice_write(pio_dev, io->port,
1789                                    io->size,
1790                                    pd);
1791                 pd += io->size;
1792         }
1793 }
1794
1795 int kvm_setup_pio(struct kvm_vcpu *vcpu, struct kvm_run *run, int in,
1796                   int size, unsigned long count, int string, int down,
1797                   gva_t address, int rep, unsigned port)
1798 {
1799         unsigned now, in_page;
1800         int i, ret = 0;
1801         int nr_pages = 1;
1802         struct page *page;
1803         struct kvm_io_device *pio_dev;
1804
1805         vcpu->run->exit_reason = KVM_EXIT_IO;
1806         vcpu->run->io.direction = in ? KVM_EXIT_IO_IN : KVM_EXIT_IO_OUT;
1807         vcpu->run->io.size = size;
1808         vcpu->run->io.data_offset = KVM_PIO_PAGE_OFFSET * PAGE_SIZE;
1809         vcpu->run->io.count = count;
1810         vcpu->run->io.port = port;
1811         vcpu->pio.count = count;
1812         vcpu->pio.cur_count = count;
1813         vcpu->pio.size = size;
1814         vcpu->pio.in = in;
1815         vcpu->pio.port = port;
1816         vcpu->pio.string = string;
1817         vcpu->pio.down = down;
1818         vcpu->pio.guest_page_offset = offset_in_page(address);
1819         vcpu->pio.rep = rep;
1820
1821         pio_dev = vcpu_find_pio_dev(vcpu, port);
1822         if (!string) {
1823                 kvm_arch_ops->cache_regs(vcpu);
1824                 memcpy(vcpu->pio_data, &vcpu->regs[VCPU_REGS_RAX], 4);
1825                 kvm_arch_ops->decache_regs(vcpu);
1826                 if (pio_dev) {
1827                         kernel_pio(pio_dev, vcpu, vcpu->pio_data);
1828                         complete_pio(vcpu);
1829                         return 1;
1830                 }
1831                 return 0;
1832         }
1833
1834         if (!count) {
1835                 kvm_arch_ops->skip_emulated_instruction(vcpu);
1836                 return 1;
1837         }
1838
1839         now = min(count, PAGE_SIZE / size);
1840
1841         if (!down)
1842                 in_page = PAGE_SIZE - offset_in_page(address);
1843         else
1844                 in_page = offset_in_page(address) + size;
1845         now = min(count, (unsigned long)in_page / size);
1846         if (!now) {
1847                 /*
1848                  * String I/O straddles page boundary.  Pin two guest pages
1849                  * so that we satisfy atomicity constraints.  Do just one
1850                  * transaction to avoid complexity.
1851                  */
1852                 nr_pages = 2;
1853                 now = 1;
1854         }
1855         if (down) {
1856                 /*
1857                  * String I/O in reverse.  Yuck.  Kill the guest, fix later.
1858                  */
1859                 printk(KERN_ERR "kvm: guest string pio down\n");
1860                 inject_gp(vcpu);
1861                 return 1;
1862         }
1863         vcpu->run->io.count = now;
1864         vcpu->pio.cur_count = now;
1865
1866         for (i = 0; i < nr_pages; ++i) {
1867                 spin_lock(&vcpu->kvm->lock);
1868                 page = gva_to_page(vcpu, address + i * PAGE_SIZE);
1869                 if (page)
1870                         get_page(page);
1871                 vcpu->pio.guest_pages[i] = page;
1872                 spin_unlock(&vcpu->kvm->lock);
1873                 if (!page) {
1874                         inject_gp(vcpu);
1875                         free_pio_guest_pages(vcpu);
1876                         return 1;
1877                 }
1878         }
1879
1880         if (!vcpu->pio.in) {
1881                 /* string PIO write */
1882                 ret = pio_copy_data(vcpu);
1883                 if (ret >= 0 && pio_dev) {
1884                         pio_string_write(pio_dev, vcpu);
1885                         complete_pio(vcpu);
1886                         if (vcpu->pio.count == 0)
1887                                 ret = 1;
1888                 }
1889         } else if (pio_dev)
1890                 printk(KERN_ERR "no string pio read support yet, "
1891                        "port %x size %d count %ld\n",
1892                         port, size, count);
1893
1894         return ret;
1895 }
1896 EXPORT_SYMBOL_GPL(kvm_setup_pio);
1897
1898 static int kvm_vcpu_ioctl_run(struct kvm_vcpu *vcpu, struct kvm_run *kvm_run)
1899 {
1900         int r;
1901         sigset_t sigsaved;
1902
1903         vcpu_load(vcpu);
1904
1905         if (vcpu->sigset_active)
1906                 sigprocmask(SIG_SETMASK, &vcpu->sigset, &sigsaved);
1907
1908         /* re-sync apic's tpr */
1909         vcpu->cr8 = kvm_run->cr8;
1910
1911         if (vcpu->pio.cur_count) {
1912                 r = complete_pio(vcpu);
1913                 if (r)
1914                         goto out;
1915         }
1916
1917         if (vcpu->mmio_needed) {
1918                 memcpy(vcpu->mmio_data, kvm_run->mmio.data, 8);
1919                 vcpu->mmio_read_completed = 1;
1920                 vcpu->mmio_needed = 0;
1921                 r = emulate_instruction(vcpu, kvm_run,
1922                                         vcpu->mmio_fault_cr2, 0);
1923                 if (r == EMULATE_DO_MMIO) {
1924                         /*
1925                          * Read-modify-write.  Back to userspace.
1926                          */
1927                         kvm_run->exit_reason = KVM_EXIT_MMIO;
1928                         r = 0;
1929                         goto out;
1930                 }
1931         }
1932
1933         if (kvm_run->exit_reason == KVM_EXIT_HYPERCALL) {
1934                 kvm_arch_ops->cache_regs(vcpu);
1935                 vcpu->regs[VCPU_REGS_RAX] = kvm_run->hypercall.ret;
1936                 kvm_arch_ops->decache_regs(vcpu);
1937         }
1938
1939         r = kvm_arch_ops->run(vcpu, kvm_run);
1940
1941 out:
1942         if (vcpu->sigset_active)
1943                 sigprocmask(SIG_SETMASK, &sigsaved, NULL);
1944
1945         vcpu_put(vcpu);
1946         return r;
1947 }
1948
1949 static int kvm_vcpu_ioctl_get_regs(struct kvm_vcpu *vcpu,
1950                                    struct kvm_regs *regs)
1951 {
1952         vcpu_load(vcpu);
1953
1954         kvm_arch_ops->cache_regs(vcpu);
1955
1956         regs->rax = vcpu->regs[VCPU_REGS_RAX];
1957         regs->rbx = vcpu->regs[VCPU_REGS_RBX];
1958         regs->rcx = vcpu->regs[VCPU_REGS_RCX];
1959         regs->rdx = vcpu->regs[VCPU_REGS_RDX];
1960         regs->rsi = vcpu->regs[VCPU_REGS_RSI];
1961         regs->rdi = vcpu->regs[VCPU_REGS_RDI];
1962         regs->rsp = vcpu->regs[VCPU_REGS_RSP];
1963         regs->rbp = vcpu->regs[VCPU_REGS_RBP];
1964 #ifdef CONFIG_X86_64
1965         regs->r8 = vcpu->regs[VCPU_REGS_R8];
1966         regs->r9 = vcpu->regs[VCPU_REGS_R9];
1967         regs->r10 = vcpu->regs[VCPU_REGS_R10];
1968         regs->r11 = vcpu->regs[VCPU_REGS_R11];
1969         regs->r12 = vcpu->regs[VCPU_REGS_R12];
1970         regs->r13 = vcpu->regs[VCPU_REGS_R13];
1971         regs->r14 = vcpu->regs[VCPU_REGS_R14];
1972         regs->r15 = vcpu->regs[VCPU_REGS_R15];
1973 #endif
1974
1975         regs->rip = vcpu->rip;
1976         regs->rflags = kvm_arch_ops->get_rflags(vcpu);
1977
1978         /*
1979          * Don't leak debug flags in case they were set for guest debugging
1980          */
1981         if (vcpu->guest_debug.enabled && vcpu->guest_debug.singlestep)
1982                 regs->rflags &= ~(X86_EFLAGS_TF | X86_EFLAGS_RF);
1983
1984         vcpu_put(vcpu);
1985
1986         return 0;
1987 }
1988
1989 static int kvm_vcpu_ioctl_set_regs(struct kvm_vcpu *vcpu,
1990                                    struct kvm_regs *regs)
1991 {
1992         vcpu_load(vcpu);
1993
1994         vcpu->regs[VCPU_REGS_RAX] = regs->rax;
1995         vcpu->regs[VCPU_REGS_RBX] = regs->rbx;
1996         vcpu->regs[VCPU_REGS_RCX] = regs->rcx;
1997         vcpu->regs[VCPU_REGS_RDX] = regs->rdx;
1998         vcpu->regs[VCPU_REGS_RSI] = regs->rsi;
1999         vcpu->regs[VCPU_REGS_RDI] = regs->rdi;
2000         vcpu->regs[VCPU_REGS_RSP] = regs->rsp;
2001         vcpu->regs[VCPU_REGS_RBP] = regs->rbp;
2002 #ifdef CONFIG_X86_64
2003         vcpu->regs[VCPU_REGS_R8] = regs->r8;
2004         vcpu->regs[VCPU_REGS_R9] = regs->r9;
2005         vcpu->regs[VCPU_REGS_R10] = regs->r10;
2006         vcpu->regs[VCPU_REGS_R11] = regs->r11;
2007         vcpu->regs[VCPU_REGS_R12] = regs->r12;
2008         vcpu->regs[VCPU_REGS_R13] = regs->r13;
2009         vcpu->regs[VCPU_REGS_R14] = regs->r14;
2010         vcpu->regs[VCPU_REGS_R15] = regs->r15;
2011 #endif
2012
2013         vcpu->rip = regs->rip;
2014         kvm_arch_ops->set_rflags(vcpu, regs->rflags);
2015
2016         kvm_arch_ops->decache_regs(vcpu);
2017
2018         vcpu_put(vcpu);
2019
2020         return 0;
2021 }
2022
2023 static void get_segment(struct kvm_vcpu *vcpu,
2024                         struct kvm_segment *var, int seg)
2025 {
2026         return kvm_arch_ops->get_segment(vcpu, var, seg);
2027 }
2028
2029 static int kvm_vcpu_ioctl_get_sregs(struct kvm_vcpu *vcpu,
2030                                     struct kvm_sregs *sregs)
2031 {
2032         struct descriptor_table dt;
2033
2034         vcpu_load(vcpu);
2035
2036         get_segment(vcpu, &sregs->cs, VCPU_SREG_CS);
2037         get_segment(vcpu, &sregs->ds, VCPU_SREG_DS);
2038         get_segment(vcpu, &sregs->es, VCPU_SREG_ES);
2039         get_segment(vcpu, &sregs->fs, VCPU_SREG_FS);
2040         get_segment(vcpu, &sregs->gs, VCPU_SREG_GS);
2041         get_segment(vcpu, &sregs->ss, VCPU_SREG_SS);
2042
2043         get_segment(vcpu, &sregs->tr, VCPU_SREG_TR);
2044         get_segment(vcpu, &sregs->ldt, VCPU_SREG_LDTR);
2045
2046         kvm_arch_ops->get_idt(vcpu, &dt);
2047         sregs->idt.limit = dt.limit;
2048         sregs->idt.base = dt.base;
2049         kvm_arch_ops->get_gdt(vcpu, &dt);
2050         sregs->gdt.limit = dt.limit;
2051         sregs->gdt.base = dt.base;
2052
2053         kvm_arch_ops->decache_cr4_guest_bits(vcpu);
2054         sregs->cr0 = vcpu->cr0;
2055         sregs->cr2 = vcpu->cr2;
2056         sregs->cr3 = vcpu->cr3;
2057         sregs->cr4 = vcpu->cr4;
2058         sregs->cr8 = vcpu->cr8;
2059         sregs->efer = vcpu->shadow_efer;
2060         sregs->apic_base = vcpu->apic_base;
2061
2062         memcpy(sregs->interrupt_bitmap, vcpu->irq_pending,
2063                sizeof sregs->interrupt_bitmap);
2064
2065         vcpu_put(vcpu);
2066
2067         return 0;
2068 }
2069
2070 static void set_segment(struct kvm_vcpu *vcpu,
2071                         struct kvm_segment *var, int seg)
2072 {
2073         return kvm_arch_ops->set_segment(vcpu, var, seg);
2074 }
2075
2076 static int kvm_vcpu_ioctl_set_sregs(struct kvm_vcpu *vcpu,
2077                                     struct kvm_sregs *sregs)
2078 {
2079         int mmu_reset_needed = 0;
2080         int i;
2081         struct descriptor_table dt;
2082
2083         vcpu_load(vcpu);
2084
2085         dt.limit = sregs->idt.limit;
2086         dt.base = sregs->idt.base;
2087         kvm_arch_ops->set_idt(vcpu, &dt);
2088         dt.limit = sregs->gdt.limit;
2089         dt.base = sregs->gdt.base;
2090         kvm_arch_ops->set_gdt(vcpu, &dt);
2091
2092         vcpu->cr2 = sregs->cr2;
2093         mmu_reset_needed |= vcpu->cr3 != sregs->cr3;
2094         vcpu->cr3 = sregs->cr3;
2095
2096         vcpu->cr8 = sregs->cr8;
2097
2098         mmu_reset_needed |= vcpu->shadow_efer != sregs->efer;
2099 #ifdef CONFIG_X86_64
2100         kvm_arch_ops->set_efer(vcpu, sregs->efer);
2101 #endif
2102         vcpu->apic_base = sregs->apic_base;
2103
2104         kvm_arch_ops->decache_cr4_guest_bits(vcpu);
2105
2106         mmu_reset_needed |= vcpu->cr0 != sregs->cr0;
2107         kvm_arch_ops->set_cr0(vcpu, sregs->cr0);
2108
2109         mmu_reset_needed |= vcpu->cr4 != sregs->cr4;
2110         kvm_arch_ops->set_cr4(vcpu, sregs->cr4);
2111         if (!is_long_mode(vcpu) && is_pae(vcpu))
2112                 load_pdptrs(vcpu, vcpu->cr3);
2113
2114         if (mmu_reset_needed)
2115                 kvm_mmu_reset_context(vcpu);
2116
2117         memcpy(vcpu->irq_pending, sregs->interrupt_bitmap,
2118                sizeof vcpu->irq_pending);
2119         vcpu->irq_summary = 0;
2120         for (i = 0; i < NR_IRQ_WORDS; ++i)
2121                 if (vcpu->irq_pending[i])
2122                         __set_bit(i, &vcpu->irq_summary);
2123
2124         set_segment(vcpu, &sregs->cs, VCPU_SREG_CS);
2125         set_segment(vcpu, &sregs->ds, VCPU_SREG_DS);
2126         set_segment(vcpu, &sregs->es, VCPU_SREG_ES);
2127         set_segment(vcpu, &sregs->fs, VCPU_SREG_FS);
2128         set_segment(vcpu, &sregs->gs, VCPU_SREG_GS);
2129         set_segment(vcpu, &sregs->ss, VCPU_SREG_SS);
2130
2131         set_segment(vcpu, &sregs->tr, VCPU_SREG_TR);
2132         set_segment(vcpu, &sregs->ldt, VCPU_SREG_LDTR);
2133
2134         vcpu_put(vcpu);
2135
2136         return 0;
2137 }
2138
2139 /*
2140  * List of msr numbers which we expose to userspace through KVM_GET_MSRS
2141  * and KVM_SET_MSRS, and KVM_GET_MSR_INDEX_LIST.
2142  *
2143  * This list is modified at module load time to reflect the
2144  * capabilities of the host cpu.
2145  */
2146 static u32 msrs_to_save[] = {
2147         MSR_IA32_SYSENTER_CS, MSR_IA32_SYSENTER_ESP, MSR_IA32_SYSENTER_EIP,
2148         MSR_K6_STAR,
2149 #ifdef CONFIG_X86_64
2150         MSR_CSTAR, MSR_KERNEL_GS_BASE, MSR_SYSCALL_MASK, MSR_LSTAR,
2151 #endif
2152         MSR_IA32_TIME_STAMP_COUNTER,
2153 };
2154
2155 static unsigned num_msrs_to_save;
2156
2157 static u32 emulated_msrs[] = {
2158         MSR_IA32_MISC_ENABLE,
2159 };
2160
2161 static __init void kvm_init_msr_list(void)
2162 {
2163         u32 dummy[2];
2164         unsigned i, j;
2165
2166         for (i = j = 0; i < ARRAY_SIZE(msrs_to_save); i++) {
2167                 if (rdmsr_safe(msrs_to_save[i], &dummy[0], &dummy[1]) < 0)
2168                         continue;
2169                 if (j < i)
2170                         msrs_to_save[j] = msrs_to_save[i];
2171                 j++;
2172         }
2173         num_msrs_to_save = j;
2174 }
2175
2176 /*
2177  * Adapt set_msr() to msr_io()'s calling convention
2178  */
2179 static int do_set_msr(struct kvm_vcpu *vcpu, unsigned index, u64 *data)
2180 {
2181         return kvm_set_msr(vcpu, index, *data);
2182 }
2183
2184 /*
2185  * Read or write a bunch of msrs. All parameters are kernel addresses.
2186  *
2187  * @return number of msrs set successfully.
2188  */
2189 static int __msr_io(struct kvm_vcpu *vcpu, struct kvm_msrs *msrs,
2190                     struct kvm_msr_entry *entries,
2191                     int (*do_msr)(struct kvm_vcpu *vcpu,
2192                                   unsigned index, u64 *data))
2193 {
2194         int i;
2195
2196         vcpu_load(vcpu);
2197
2198         for (i = 0; i < msrs->nmsrs; ++i)
2199                 if (do_msr(vcpu, entries[i].index, &entries[i].data))
2200                         break;
2201
2202         vcpu_put(vcpu);
2203
2204         return i;
2205 }
2206
2207 /*
2208  * Read or write a bunch of msrs. Parameters are user addresses.
2209  *
2210  * @return number of msrs set successfully.
2211  */
2212 static int msr_io(struct kvm_vcpu *vcpu, struct kvm_msrs __user *user_msrs,
2213                   int (*do_msr)(struct kvm_vcpu *vcpu,
2214                                 unsigned index, u64 *data),
2215                   int writeback)
2216 {
2217         struct kvm_msrs msrs;
2218         struct kvm_msr_entry *entries;
2219         int r, n;
2220         unsigned size;
2221
2222         r = -EFAULT;
2223         if (copy_from_user(&msrs, user_msrs, sizeof msrs))
2224                 goto out;
2225
2226         r = -E2BIG;
2227         if (msrs.nmsrs >= MAX_IO_MSRS)
2228                 goto out;
2229
2230         r = -ENOMEM;
2231         size = sizeof(struct kvm_msr_entry) * msrs.nmsrs;
2232         entries = vmalloc(size);
2233         if (!entries)
2234                 goto out;
2235
2236         r = -EFAULT;
2237         if (copy_from_user(entries, user_msrs->entries, size))
2238                 goto out_free;
2239
2240         r = n = __msr_io(vcpu, &msrs, entries, do_msr);
2241         if (r < 0)
2242                 goto out_free;
2243
2244         r = -EFAULT;
2245         if (writeback && copy_to_user(user_msrs->entries, entries, size))
2246                 goto out_free;
2247
2248         r = n;
2249
2250 out_free:
2251         vfree(entries);
2252 out:
2253         return r;
2254 }
2255
2256 /*
2257  * Translate a guest virtual address to a guest physical address.
2258  */
2259 static int kvm_vcpu_ioctl_translate(struct kvm_vcpu *vcpu,
2260                                     struct kvm_translation *tr)
2261 {
2262         unsigned long vaddr = tr->linear_address;
2263         gpa_t gpa;
2264
2265         vcpu_load(vcpu);
2266         spin_lock(&vcpu->kvm->lock);
2267         gpa = vcpu->mmu.gva_to_gpa(vcpu, vaddr);
2268         tr->physical_address = gpa;
2269         tr->valid = gpa != UNMAPPED_GVA;
2270         tr->writeable = 1;
2271         tr->usermode = 0;
2272         spin_unlock(&vcpu->kvm->lock);
2273         vcpu_put(vcpu);
2274
2275         return 0;
2276 }
2277
2278 static int kvm_vcpu_ioctl_interrupt(struct kvm_vcpu *vcpu,
2279                                     struct kvm_interrupt *irq)
2280 {
2281         if (irq->irq < 0 || irq->irq >= 256)
2282                 return -EINVAL;
2283         vcpu_load(vcpu);
2284
2285         set_bit(irq->irq, vcpu->irq_pending);
2286         set_bit(irq->irq / BITS_PER_LONG, &vcpu->irq_summary);
2287
2288         vcpu_put(vcpu);
2289
2290         return 0;
2291 }
2292
2293 static int kvm_vcpu_ioctl_debug_guest(struct kvm_vcpu *vcpu,
2294                                       struct kvm_debug_guest *dbg)
2295 {
2296         int r;
2297
2298         vcpu_load(vcpu);
2299
2300         r = kvm_arch_ops->set_guest_debug(vcpu, dbg);
2301
2302         vcpu_put(vcpu);
2303
2304         return r;
2305 }
2306
2307 static struct page *kvm_vcpu_nopage(struct vm_area_struct *vma,
2308                                     unsigned long address,
2309                                     int *type)
2310 {
2311         struct kvm_vcpu *vcpu = vma->vm_file->private_data;
2312         unsigned long pgoff;
2313         struct page *page;
2314
2315         pgoff = ((address - vma->vm_start) >> PAGE_SHIFT) + vma->vm_pgoff;
2316         if (pgoff == 0)
2317                 page = virt_to_page(vcpu->run);
2318         else if (pgoff == KVM_PIO_PAGE_OFFSET)
2319                 page = virt_to_page(vcpu->pio_data);
2320         else
2321                 return NOPAGE_SIGBUS;
2322         get_page(page);
2323         if (type != NULL)
2324                 *type = VM_FAULT_MINOR;
2325
2326         return page;
2327 }
2328
2329 static struct vm_operations_struct kvm_vcpu_vm_ops = {
2330         .nopage = kvm_vcpu_nopage,
2331 };
2332
2333 static int kvm_vcpu_mmap(struct file *file, struct vm_area_struct *vma)
2334 {
2335         vma->vm_ops = &kvm_vcpu_vm_ops;
2336         return 0;
2337 }
2338
2339 static int kvm_vcpu_release(struct inode *inode, struct file *filp)
2340 {
2341         struct kvm_vcpu *vcpu = filp->private_data;
2342
2343         fput(vcpu->kvm->filp);
2344         return 0;
2345 }
2346
2347 static struct file_operations kvm_vcpu_fops = {
2348         .release        = kvm_vcpu_release,
2349         .unlocked_ioctl = kvm_vcpu_ioctl,
2350         .compat_ioctl   = kvm_vcpu_ioctl,
2351         .mmap           = kvm_vcpu_mmap,
2352 };
2353
2354 /*
2355  * Allocates an inode for the vcpu.
2356  */
2357 static int create_vcpu_fd(struct kvm_vcpu *vcpu)
2358 {
2359         int fd, r;
2360         struct inode *inode;
2361         struct file *file;
2362
2363         r = anon_inode_getfd(&fd, &inode, &file,
2364                              "kvm-vcpu", &kvm_vcpu_fops, vcpu);
2365         if (r)
2366                 return r;
2367         atomic_inc(&vcpu->kvm->filp->f_count);
2368         return fd;
2369 }
2370
2371 /*
2372  * Creates some virtual cpus.  Good luck creating more than one.
2373  */
2374 static int kvm_vm_ioctl_create_vcpu(struct kvm *kvm, int n)
2375 {
2376         int r;
2377         struct kvm_vcpu *vcpu;
2378         struct page *page;
2379
2380         r = -EINVAL;
2381         if (!valid_vcpu(n))
2382                 goto out;
2383
2384         vcpu = &kvm->vcpus[n];
2385         vcpu->vcpu_id = n;
2386
2387         mutex_lock(&vcpu->mutex);
2388
2389         if (vcpu->vmcs) {
2390                 mutex_unlock(&vcpu->mutex);
2391                 return -EEXIST;
2392         }
2393
2394         page = alloc_page(GFP_KERNEL | __GFP_ZERO);
2395         r = -ENOMEM;
2396         if (!page)
2397                 goto out_unlock;
2398         vcpu->run = page_address(page);
2399
2400         page = alloc_page(GFP_KERNEL | __GFP_ZERO);
2401         r = -ENOMEM;
2402         if (!page)
2403                 goto out_free_run;
2404         vcpu->pio_data = page_address(page);
2405
2406         vcpu->host_fx_image = (char*)ALIGN((hva_t)vcpu->fx_buf,
2407                                            FX_IMAGE_ALIGN);
2408         vcpu->guest_fx_image = vcpu->host_fx_image + FX_IMAGE_SIZE;
2409         vcpu->cr0 = 0x10;
2410
2411         r = kvm_arch_ops->vcpu_create(vcpu);
2412         if (r < 0)
2413                 goto out_free_vcpus;
2414
2415         r = kvm_mmu_create(vcpu);
2416         if (r < 0)
2417                 goto out_free_vcpus;
2418
2419         kvm_arch_ops->vcpu_load(vcpu);
2420         r = kvm_mmu_setup(vcpu);
2421         if (r >= 0)
2422                 r = kvm_arch_ops->vcpu_setup(vcpu);
2423         vcpu_put(vcpu);
2424
2425         if (r < 0)
2426                 goto out_free_vcpus;
2427
2428         r = create_vcpu_fd(vcpu);
2429         if (r < 0)
2430                 goto out_free_vcpus;
2431
2432         spin_lock(&kvm_lock);
2433         if (n >= kvm->nvcpus)
2434                 kvm->nvcpus = n + 1;
2435         spin_unlock(&kvm_lock);
2436
2437         return r;
2438
2439 out_free_vcpus:
2440         kvm_free_vcpu(vcpu);
2441 out_free_run:
2442         free_page((unsigned long)vcpu->run);
2443         vcpu->run = NULL;
2444 out_unlock:
2445         mutex_unlock(&vcpu->mutex);
2446 out:
2447         return r;
2448 }
2449
2450 static void cpuid_fix_nx_cap(struct kvm_vcpu *vcpu)
2451 {
2452         u64 efer;
2453         int i;
2454         struct kvm_cpuid_entry *e, *entry;
2455
2456         rdmsrl(MSR_EFER, efer);
2457         entry = NULL;
2458         for (i = 0; i < vcpu->cpuid_nent; ++i) {
2459                 e = &vcpu->cpuid_entries[i];
2460                 if (e->function == 0x80000001) {
2461                         entry = e;
2462                         break;
2463                 }
2464         }
2465         if (entry && (entry->edx & (1 << 20)) && !(efer & EFER_NX)) {
2466                 entry->edx &= ~(1 << 20);
2467                 printk(KERN_INFO "kvm: guest NX capability removed\n");
2468         }
2469 }
2470
2471 static int kvm_vcpu_ioctl_set_cpuid(struct kvm_vcpu *vcpu,
2472                                     struct kvm_cpuid *cpuid,
2473                                     struct kvm_cpuid_entry __user *entries)
2474 {
2475         int r;
2476
2477         r = -E2BIG;
2478         if (cpuid->nent > KVM_MAX_CPUID_ENTRIES)
2479                 goto out;
2480         r = -EFAULT;
2481         if (copy_from_user(&vcpu->cpuid_entries, entries,
2482                            cpuid->nent * sizeof(struct kvm_cpuid_entry)))
2483                 goto out;
2484         vcpu->cpuid_nent = cpuid->nent;
2485         cpuid_fix_nx_cap(vcpu);
2486         return 0;
2487
2488 out:
2489         return r;
2490 }
2491
2492 static int kvm_vcpu_ioctl_set_sigmask(struct kvm_vcpu *vcpu, sigset_t *sigset)
2493 {
2494         if (sigset) {
2495                 sigdelsetmask(sigset, sigmask(SIGKILL)|sigmask(SIGSTOP));
2496                 vcpu->sigset_active = 1;
2497                 vcpu->sigset = *sigset;
2498         } else
2499                 vcpu->sigset_active = 0;
2500         return 0;
2501 }
2502
2503 /*
2504  * fxsave fpu state.  Taken from x86_64/processor.h.  To be killed when
2505  * we have asm/x86/processor.h
2506  */
2507 struct fxsave {
2508         u16     cwd;
2509         u16     swd;
2510         u16     twd;
2511         u16     fop;
2512         u64     rip;
2513         u64     rdp;
2514         u32     mxcsr;
2515         u32     mxcsr_mask;
2516         u32     st_space[32];   /* 8*16 bytes for each FP-reg = 128 bytes */
2517 #ifdef CONFIG_X86_64
2518         u32     xmm_space[64];  /* 16*16 bytes for each XMM-reg = 256 bytes */
2519 #else
2520         u32     xmm_space[32];  /* 8*16 bytes for each XMM-reg = 128 bytes */
2521 #endif
2522 };
2523
2524 static int kvm_vcpu_ioctl_get_fpu(struct kvm_vcpu *vcpu, struct kvm_fpu *fpu)
2525 {
2526         struct fxsave *fxsave = (struct fxsave *)vcpu->guest_fx_image;
2527
2528         vcpu_load(vcpu);
2529
2530         memcpy(fpu->fpr, fxsave->st_space, 128);
2531         fpu->fcw = fxsave->cwd;
2532         fpu->fsw = fxsave->swd;
2533         fpu->ftwx = fxsave->twd;
2534         fpu->last_opcode = fxsave->fop;
2535         fpu->last_ip = fxsave->rip;
2536         fpu->last_dp = fxsave->rdp;
2537         memcpy(fpu->xmm, fxsave->xmm_space, sizeof fxsave->xmm_space);
2538
2539         vcpu_put(vcpu);
2540
2541         return 0;
2542 }
2543
2544 static int kvm_vcpu_ioctl_set_fpu(struct kvm_vcpu *vcpu, struct kvm_fpu *fpu)
2545 {
2546         struct fxsave *fxsave = (struct fxsave *)vcpu->guest_fx_image;
2547
2548         vcpu_load(vcpu);
2549
2550         memcpy(fxsave->st_space, fpu->fpr, 128);
2551         fxsave->cwd = fpu->fcw;
2552         fxsave->swd = fpu->fsw;
2553         fxsave->twd = fpu->ftwx;
2554         fxsave->fop = fpu->last_opcode;
2555         fxsave->rip = fpu->last_ip;
2556         fxsave->rdp = fpu->last_dp;
2557         memcpy(fxsave->xmm_space, fpu->xmm, sizeof fxsave->xmm_space);
2558
2559         vcpu_put(vcpu);
2560
2561         return 0;
2562 }
2563
2564 static long kvm_vcpu_ioctl(struct file *filp,
2565                            unsigned int ioctl, unsigned long arg)
2566 {
2567         struct kvm_vcpu *vcpu = filp->private_data;
2568         void __user *argp = (void __user *)arg;
2569         int r = -EINVAL;
2570
2571         switch (ioctl) {
2572         case KVM_RUN:
2573                 r = -EINVAL;
2574                 if (arg)
2575                         goto out;
2576                 r = kvm_vcpu_ioctl_run(vcpu, vcpu->run);
2577                 break;
2578         case KVM_GET_REGS: {
2579                 struct kvm_regs kvm_regs;
2580
2581                 memset(&kvm_regs, 0, sizeof kvm_regs);
2582                 r = kvm_vcpu_ioctl_get_regs(vcpu, &kvm_regs);
2583                 if (r)
2584                         goto out;
2585                 r = -EFAULT;
2586                 if (copy_to_user(argp, &kvm_regs, sizeof kvm_regs))
2587                         goto out;
2588                 r = 0;
2589                 break;
2590         }
2591         case KVM_SET_REGS: {
2592                 struct kvm_regs kvm_regs;
2593
2594                 r = -EFAULT;
2595                 if (copy_from_user(&kvm_regs, argp, sizeof kvm_regs))
2596                         goto out;
2597                 r = kvm_vcpu_ioctl_set_regs(vcpu, &kvm_regs);
2598                 if (r)
2599                         goto out;
2600                 r = 0;
2601                 break;
2602         }
2603         case KVM_GET_SREGS: {
2604                 struct kvm_sregs kvm_sregs;
2605
2606                 memset(&kvm_sregs, 0, sizeof kvm_sregs);
2607                 r = kvm_vcpu_ioctl_get_sregs(vcpu, &kvm_sregs);
2608                 if (r)
2609                         goto out;
2610                 r = -EFAULT;
2611                 if (copy_to_user(argp, &kvm_sregs, sizeof kvm_sregs))
2612                         goto out;
2613                 r = 0;
2614                 break;
2615         }
2616         case KVM_SET_SREGS: {
2617                 struct kvm_sregs kvm_sregs;
2618
2619                 r = -EFAULT;
2620                 if (copy_from_user(&kvm_sregs, argp, sizeof kvm_sregs))
2621                         goto out;
2622                 r = kvm_vcpu_ioctl_set_sregs(vcpu, &kvm_sregs);
2623                 if (r)
2624                         goto out;
2625                 r = 0;
2626                 break;
2627         }
2628         case KVM_TRANSLATE: {
2629                 struct kvm_translation tr;
2630
2631                 r = -EFAULT;
2632                 if (copy_from_user(&tr, argp, sizeof tr))
2633                         goto out;
2634                 r = kvm_vcpu_ioctl_translate(vcpu, &tr);
2635                 if (r)
2636                         goto out;
2637                 r = -EFAULT;
2638                 if (copy_to_user(argp, &tr, sizeof tr))
2639                         goto out;
2640                 r = 0;
2641                 break;
2642         }
2643         case KVM_INTERRUPT: {
2644                 struct kvm_interrupt irq;
2645
2646                 r = -EFAULT;
2647                 if (copy_from_user(&irq, argp, sizeof irq))
2648                         goto out;
2649                 r = kvm_vcpu_ioctl_interrupt(vcpu, &irq);
2650                 if (r)
2651                         goto out;
2652                 r = 0;
2653                 break;
2654         }
2655         case KVM_DEBUG_GUEST: {
2656                 struct kvm_debug_guest dbg;
2657
2658                 r = -EFAULT;
2659                 if (copy_from_user(&dbg, argp, sizeof dbg))
2660                         goto out;
2661                 r = kvm_vcpu_ioctl_debug_guest(vcpu, &dbg);
2662                 if (r)
2663                         goto out;
2664                 r = 0;
2665                 break;
2666         }
2667         case KVM_GET_MSRS:
2668                 r = msr_io(vcpu, argp, kvm_get_msr, 1);
2669                 break;
2670         case KVM_SET_MSRS:
2671                 r = msr_io(vcpu, argp, do_set_msr, 0);
2672                 break;
2673         case KVM_SET_CPUID: {
2674                 struct kvm_cpuid __user *cpuid_arg = argp;
2675                 struct kvm_cpuid cpuid;
2676
2677                 r = -EFAULT;
2678                 if (copy_from_user(&cpuid, cpuid_arg, sizeof cpuid))
2679                         goto out;
2680                 r = kvm_vcpu_ioctl_set_cpuid(vcpu, &cpuid, cpuid_arg->entries);
2681                 if (r)
2682                         goto out;
2683                 break;
2684         }
2685         case KVM_SET_SIGNAL_MASK: {
2686                 struct kvm_signal_mask __user *sigmask_arg = argp;
2687                 struct kvm_signal_mask kvm_sigmask;
2688                 sigset_t sigset, *p;
2689
2690                 p = NULL;
2691                 if (argp) {
2692                         r = -EFAULT;
2693                         if (copy_from_user(&kvm_sigmask, argp,
2694                                            sizeof kvm_sigmask))
2695                                 goto out;
2696                         r = -EINVAL;
2697                         if (kvm_sigmask.len != sizeof sigset)
2698                                 goto out;
2699                         r = -EFAULT;
2700                         if (copy_from_user(&sigset, sigmask_arg->sigset,
2701                                            sizeof sigset))
2702                                 goto out;
2703                         p = &sigset;
2704                 }
2705                 r = kvm_vcpu_ioctl_set_sigmask(vcpu, &sigset);
2706                 break;
2707         }
2708         case KVM_GET_FPU: {
2709                 struct kvm_fpu fpu;
2710
2711                 memset(&fpu, 0, sizeof fpu);
2712                 r = kvm_vcpu_ioctl_get_fpu(vcpu, &fpu);
2713                 if (r)
2714                         goto out;
2715                 r = -EFAULT;
2716                 if (copy_to_user(argp, &fpu, sizeof fpu))
2717                         goto out;
2718                 r = 0;
2719                 break;
2720         }
2721         case KVM_SET_FPU: {
2722                 struct kvm_fpu fpu;
2723
2724                 r = -EFAULT;
2725                 if (copy_from_user(&fpu, argp, sizeof fpu))
2726                         goto out;
2727                 r = kvm_vcpu_ioctl_set_fpu(vcpu, &fpu);
2728                 if (r)
2729                         goto out;
2730                 r = 0;
2731                 break;
2732         }
2733         default:
2734                 ;
2735         }
2736 out:
2737         return r;
2738 }
2739
2740 static long kvm_vm_ioctl(struct file *filp,
2741                            unsigned int ioctl, unsigned long arg)
2742 {
2743         struct kvm *kvm = filp->private_data;
2744         void __user *argp = (void __user *)arg;
2745         int r = -EINVAL;
2746
2747         switch (ioctl) {
2748         case KVM_CREATE_VCPU:
2749                 r = kvm_vm_ioctl_create_vcpu(kvm, arg);
2750                 if (r < 0)
2751                         goto out;
2752                 break;
2753         case KVM_SET_MEMORY_REGION: {
2754                 struct kvm_memory_region kvm_mem;
2755
2756                 r = -EFAULT;
2757                 if (copy_from_user(&kvm_mem, argp, sizeof kvm_mem))
2758                         goto out;
2759                 r = kvm_vm_ioctl_set_memory_region(kvm, &kvm_mem);
2760                 if (r)
2761                         goto out;
2762                 break;
2763         }
2764         case KVM_GET_DIRTY_LOG: {
2765                 struct kvm_dirty_log log;
2766
2767                 r = -EFAULT;
2768                 if (copy_from_user(&log, argp, sizeof log))
2769                         goto out;
2770                 r = kvm_vm_ioctl_get_dirty_log(kvm, &log);
2771                 if (r)
2772                         goto out;
2773                 break;
2774         }
2775         case KVM_SET_MEMORY_ALIAS: {
2776                 struct kvm_memory_alias alias;
2777
2778                 r = -EFAULT;
2779                 if (copy_from_user(&alias, argp, sizeof alias))
2780                         goto out;
2781                 r = kvm_vm_ioctl_set_memory_alias(kvm, &alias);
2782                 if (r)
2783                         goto out;
2784                 break;
2785         }
2786         default:
2787                 ;
2788         }
2789 out:
2790         return r;
2791 }
2792
2793 static struct page *kvm_vm_nopage(struct vm_area_struct *vma,
2794                                   unsigned long address,
2795                                   int *type)
2796 {
2797         struct kvm *kvm = vma->vm_file->private_data;
2798         unsigned long pgoff;
2799         struct page *page;
2800
2801         pgoff = ((address - vma->vm_start) >> PAGE_SHIFT) + vma->vm_pgoff;
2802         page = gfn_to_page(kvm, pgoff);
2803         if (!page)
2804                 return NOPAGE_SIGBUS;
2805         get_page(page);
2806         if (type != NULL)
2807                 *type = VM_FAULT_MINOR;
2808
2809         return page;
2810 }
2811
2812 static struct vm_operations_struct kvm_vm_vm_ops = {
2813         .nopage = kvm_vm_nopage,
2814 };
2815
2816 static int kvm_vm_mmap(struct file *file, struct vm_area_struct *vma)
2817 {
2818         vma->vm_ops = &kvm_vm_vm_ops;
2819         return 0;
2820 }
2821
2822 static struct file_operations kvm_vm_fops = {
2823         .release        = kvm_vm_release,
2824         .unlocked_ioctl = kvm_vm_ioctl,
2825         .compat_ioctl   = kvm_vm_ioctl,
2826         .mmap           = kvm_vm_mmap,
2827 };
2828
2829 static int kvm_dev_ioctl_create_vm(void)
2830 {
2831         int fd, r;
2832         struct inode *inode;
2833         struct file *file;
2834         struct kvm *kvm;
2835
2836         kvm = kvm_create_vm();
2837         if (IS_ERR(kvm))
2838                 return PTR_ERR(kvm);
2839         r = anon_inode_getfd(&fd, &inode, &file, "kvm-vm", &kvm_vm_fops, kvm);
2840         if (r) {
2841                 kvm_destroy_vm(kvm);
2842                 return r;
2843         }
2844
2845         kvm->filp = file;
2846
2847         return fd;
2848 }
2849
2850 static long kvm_dev_ioctl(struct file *filp,
2851                           unsigned int ioctl, unsigned long arg)
2852 {
2853         void __user *argp = (void __user *)arg;
2854         long r = -EINVAL;
2855
2856         switch (ioctl) {
2857         case KVM_GET_API_VERSION:
2858                 r = -EINVAL;
2859                 if (arg)
2860                         goto out;
2861                 r = KVM_API_VERSION;
2862                 break;
2863         case KVM_CREATE_VM:
2864                 r = -EINVAL;
2865                 if (arg)
2866                         goto out;
2867                 r = kvm_dev_ioctl_create_vm();
2868                 break;
2869         case KVM_GET_MSR_INDEX_LIST: {
2870                 struct kvm_msr_list __user *user_msr_list = argp;
2871                 struct kvm_msr_list msr_list;
2872                 unsigned n;
2873
2874                 r = -EFAULT;
2875                 if (copy_from_user(&msr_list, user_msr_list, sizeof msr_list))
2876                         goto out;
2877                 n = msr_list.nmsrs;
2878                 msr_list.nmsrs = num_msrs_to_save + ARRAY_SIZE(emulated_msrs);
2879                 if (copy_to_user(user_msr_list, &msr_list, sizeof msr_list))
2880                         goto out;
2881                 r = -E2BIG;
2882                 if (n < num_msrs_to_save)
2883                         goto out;
2884                 r = -EFAULT;
2885                 if (copy_to_user(user_msr_list->indices, &msrs_to_save,
2886                                  num_msrs_to_save * sizeof(u32)))
2887                         goto out;
2888                 if (copy_to_user(user_msr_list->indices
2889                                  + num_msrs_to_save * sizeof(u32),
2890                                  &emulated_msrs,
2891                                  ARRAY_SIZE(emulated_msrs) * sizeof(u32)))
2892                         goto out;
2893                 r = 0;
2894                 break;
2895         }
2896         case KVM_CHECK_EXTENSION:
2897                 /*
2898                  * No extensions defined at present.
2899                  */
2900                 r = 0;
2901                 break;
2902         case KVM_GET_VCPU_MMAP_SIZE:
2903                 r = -EINVAL;
2904                 if (arg)
2905                         goto out;
2906                 r = 2 * PAGE_SIZE;
2907                 break;
2908         default:
2909                 ;
2910         }
2911 out:
2912         return r;
2913 }
2914
2915 static struct file_operations kvm_chardev_ops = {
2916         .open           = kvm_dev_open,
2917         .release        = kvm_dev_release,
2918         .unlocked_ioctl = kvm_dev_ioctl,
2919         .compat_ioctl   = kvm_dev_ioctl,
2920 };
2921
2922 static struct miscdevice kvm_dev = {
2923         KVM_MINOR,
2924         "kvm",
2925         &kvm_chardev_ops,
2926 };
2927
2928 /*
2929  * Make sure that a cpu that is being hot-unplugged does not have any vcpus
2930  * cached on it.
2931  */
2932 static void decache_vcpus_on_cpu(int cpu)
2933 {
2934         struct kvm *vm;
2935         struct kvm_vcpu *vcpu;
2936         int i;
2937
2938         spin_lock(&kvm_lock);
2939         list_for_each_entry(vm, &vm_list, vm_list)
2940                 for (i = 0; i < KVM_MAX_VCPUS; ++i) {
2941                         vcpu = &vm->vcpus[i];
2942                         /*
2943                          * If the vcpu is locked, then it is running on some
2944                          * other cpu and therefore it is not cached on the
2945                          * cpu in question.
2946                          *
2947                          * If it's not locked, check the last cpu it executed
2948                          * on.
2949                          */
2950                         if (mutex_trylock(&vcpu->mutex)) {
2951                                 if (vcpu->cpu == cpu) {
2952                                         kvm_arch_ops->vcpu_decache(vcpu);
2953                                         vcpu->cpu = -1;
2954                                 }
2955                                 mutex_unlock(&vcpu->mutex);
2956                         }
2957                 }
2958         spin_unlock(&kvm_lock);
2959 }
2960
2961 static void hardware_enable(void *junk)
2962 {
2963         int cpu = raw_smp_processor_id();
2964
2965         if (cpu_isset(cpu, cpus_hardware_enabled))
2966                 return;
2967         cpu_set(cpu, cpus_hardware_enabled);
2968         kvm_arch_ops->hardware_enable(NULL);
2969 }
2970
2971 static void hardware_disable(void *junk)
2972 {
2973         int cpu = raw_smp_processor_id();
2974
2975         if (!cpu_isset(cpu, cpus_hardware_enabled))
2976                 return;
2977         cpu_clear(cpu, cpus_hardware_enabled);
2978         decache_vcpus_on_cpu(cpu);
2979         kvm_arch_ops->hardware_disable(NULL);
2980 }
2981
2982 static int kvm_cpu_hotplug(struct notifier_block *notifier, unsigned long val,
2983                            void *v)
2984 {
2985         int cpu = (long)v;
2986
2987         switch (val) {
2988         case CPU_DYING:
2989         case CPU_DYING_FROZEN:
2990                 printk(KERN_INFO "kvm: disabling virtualization on CPU%d\n",
2991                        cpu);
2992                 hardware_disable(NULL);
2993                 break;
2994         case CPU_UP_CANCELED:
2995         case CPU_UP_CANCELED_FROZEN:
2996                 printk(KERN_INFO "kvm: disabling virtualization on CPU%d\n",
2997                        cpu);
2998                 smp_call_function_single(cpu, hardware_disable, NULL, 0, 1);
2999                 break;
3000         case CPU_ONLINE:
3001         case CPU_ONLINE_FROZEN:
3002                 printk(KERN_INFO "kvm: enabling virtualization on CPU%d\n",
3003                        cpu);
3004                 smp_call_function_single(cpu, hardware_enable, NULL, 0, 1);
3005                 break;
3006         }
3007         return NOTIFY_OK;
3008 }
3009
3010 static int kvm_reboot(struct notifier_block *notifier, unsigned long val,
3011                        void *v)
3012 {
3013         if (val == SYS_RESTART) {
3014                 /*
3015                  * Some (well, at least mine) BIOSes hang on reboot if
3016                  * in vmx root mode.
3017                  */
3018                 printk(KERN_INFO "kvm: exiting hardware virtualization\n");
3019                 on_each_cpu(hardware_disable, NULL, 0, 1);
3020         }
3021         return NOTIFY_OK;
3022 }
3023
3024 static struct notifier_block kvm_reboot_notifier = {
3025         .notifier_call = kvm_reboot,
3026         .priority = 0,
3027 };
3028
3029 void kvm_io_bus_init(struct kvm_io_bus *bus)
3030 {
3031         memset(bus, 0, sizeof(*bus));
3032 }
3033
3034 void kvm_io_bus_destroy(struct kvm_io_bus *bus)
3035 {
3036         int i;
3037
3038         for (i = 0; i < bus->dev_count; i++) {
3039                 struct kvm_io_device *pos = bus->devs[i];
3040
3041                 kvm_iodevice_destructor(pos);
3042         }
3043 }
3044
3045 struct kvm_io_device *kvm_io_bus_find_dev(struct kvm_io_bus *bus, gpa_t addr)
3046 {
3047         int i;
3048
3049         for (i = 0; i < bus->dev_count; i++) {
3050                 struct kvm_io_device *pos = bus->devs[i];
3051
3052                 if (pos->in_range(pos, addr))
3053                         return pos;
3054         }
3055
3056         return NULL;
3057 }
3058
3059 void kvm_io_bus_register_dev(struct kvm_io_bus *bus, struct kvm_io_device *dev)
3060 {
3061         BUG_ON(bus->dev_count > (NR_IOBUS_DEVS-1));
3062
3063         bus->devs[bus->dev_count++] = dev;
3064 }
3065
3066 static struct notifier_block kvm_cpu_notifier = {
3067         .notifier_call = kvm_cpu_hotplug,
3068         .priority = 20, /* must be > scheduler priority */
3069 };
3070
3071 static u64 stat_get(void *_offset)
3072 {
3073         unsigned offset = (long)_offset;
3074         u64 total = 0;
3075         struct kvm *kvm;
3076         struct kvm_vcpu *vcpu;
3077         int i;
3078
3079         spin_lock(&kvm_lock);
3080         list_for_each_entry(kvm, &vm_list, vm_list)
3081                 for (i = 0; i < KVM_MAX_VCPUS; ++i) {
3082                         vcpu = &kvm->vcpus[i];
3083                         total += *(u32 *)((void *)vcpu + offset);
3084                 }
3085         spin_unlock(&kvm_lock);
3086         return total;
3087 }
3088
3089 static void stat_set(void *offset, u64 val)
3090 {
3091 }
3092
3093 DEFINE_SIMPLE_ATTRIBUTE(stat_fops, stat_get, stat_set, "%llu\n");
3094
3095 static __init void kvm_init_debug(void)
3096 {
3097         struct kvm_stats_debugfs_item *p;
3098
3099         debugfs_dir = debugfs_create_dir("kvm", NULL);
3100         for (p = debugfs_entries; p->name; ++p)
3101                 p->dentry = debugfs_create_file(p->name, 0444, debugfs_dir,
3102                                                 (void *)(long)p->offset,
3103                                                 &stat_fops);
3104 }
3105
3106 static void kvm_exit_debug(void)
3107 {
3108         struct kvm_stats_debugfs_item *p;
3109
3110         for (p = debugfs_entries; p->name; ++p)
3111                 debugfs_remove(p->dentry);
3112         debugfs_remove(debugfs_dir);
3113 }
3114
3115 static int kvm_suspend(struct sys_device *dev, pm_message_t state)
3116 {
3117         hardware_disable(NULL);
3118         return 0;
3119 }
3120
3121 static int kvm_resume(struct sys_device *dev)
3122 {
3123         hardware_enable(NULL);
3124         return 0;
3125 }
3126
3127 static struct sysdev_class kvm_sysdev_class = {
3128         set_kset_name("kvm"),
3129         .suspend = kvm_suspend,
3130         .resume = kvm_resume,
3131 };
3132
3133 static struct sys_device kvm_sysdev = {
3134         .id = 0,
3135         .cls = &kvm_sysdev_class,
3136 };
3137
3138 hpa_t bad_page_address;
3139
3140 int kvm_init_arch(struct kvm_arch_ops *ops, struct module *module)
3141 {
3142         int r;
3143
3144         if (kvm_arch_ops) {
3145                 printk(KERN_ERR "kvm: already loaded the other module\n");
3146                 return -EEXIST;
3147         }
3148
3149         if (!ops->cpu_has_kvm_support()) {
3150                 printk(KERN_ERR "kvm: no hardware support\n");
3151                 return -EOPNOTSUPP;
3152         }
3153         if (ops->disabled_by_bios()) {
3154                 printk(KERN_ERR "kvm: disabled by bios\n");
3155                 return -EOPNOTSUPP;
3156         }
3157
3158         kvm_arch_ops = ops;
3159
3160         r = kvm_arch_ops->hardware_setup();
3161         if (r < 0)
3162                 goto out;
3163
3164         on_each_cpu(hardware_enable, NULL, 0, 1);
3165         r = register_cpu_notifier(&kvm_cpu_notifier);
3166         if (r)
3167                 goto out_free_1;
3168         register_reboot_notifier(&kvm_reboot_notifier);
3169
3170         r = sysdev_class_register(&kvm_sysdev_class);
3171         if (r)
3172                 goto out_free_2;
3173
3174         r = sysdev_register(&kvm_sysdev);
3175         if (r)
3176                 goto out_free_3;
3177
3178         kvm_chardev_ops.owner = module;
3179
3180         r = misc_register(&kvm_dev);
3181         if (r) {
3182                 printk (KERN_ERR "kvm: misc device register failed\n");
3183                 goto out_free;
3184         }
3185
3186         return r;
3187
3188 out_free:
3189         sysdev_unregister(&kvm_sysdev);
3190 out_free_3:
3191         sysdev_class_unregister(&kvm_sysdev_class);
3192 out_free_2:
3193         unregister_reboot_notifier(&kvm_reboot_notifier);
3194         unregister_cpu_notifier(&kvm_cpu_notifier);
3195 out_free_1:
3196         on_each_cpu(hardware_disable, NULL, 0, 1);
3197         kvm_arch_ops->hardware_unsetup();
3198 out:
3199         kvm_arch_ops = NULL;
3200         return r;
3201 }
3202
3203 void kvm_exit_arch(void)
3204 {
3205         misc_deregister(&kvm_dev);
3206         sysdev_unregister(&kvm_sysdev);
3207         sysdev_class_unregister(&kvm_sysdev_class);
3208         unregister_reboot_notifier(&kvm_reboot_notifier);
3209         unregister_cpu_notifier(&kvm_cpu_notifier);
3210         on_each_cpu(hardware_disable, NULL, 0, 1);
3211         kvm_arch_ops->hardware_unsetup();
3212         kvm_arch_ops = NULL;
3213 }
3214
3215 static __init int kvm_init(void)
3216 {
3217         static struct page *bad_page;
3218         int r;
3219
3220         r = kvm_mmu_module_init();
3221         if (r)
3222                 goto out4;
3223
3224         kvm_init_debug();
3225
3226         kvm_init_msr_list();
3227
3228         if ((bad_page = alloc_page(GFP_KERNEL)) == NULL) {
3229                 r = -ENOMEM;
3230                 goto out;
3231         }
3232
3233         bad_page_address = page_to_pfn(bad_page) << PAGE_SHIFT;
3234         memset(__va(bad_page_address), 0, PAGE_SIZE);
3235
3236         return 0;
3237
3238 out:
3239         kvm_exit_debug();
3240         kvm_mmu_module_exit();
3241 out4:
3242         return r;
3243 }
3244
3245 static __exit void kvm_exit(void)
3246 {
3247         kvm_exit_debug();
3248         __free_page(pfn_to_page(bad_page_address >> PAGE_SHIFT));
3249         kvm_mmu_module_exit();
3250 }
3251
3252 module_init(kvm_init)
3253 module_exit(kvm_exit)
3254
3255 EXPORT_SYMBOL_GPL(kvm_init_arch);
3256 EXPORT_SYMBOL_GPL(kvm_exit_arch);