blob: b6b8a41b5ec80b9a56bf39355ea3add4893d991b [file] [log] [blame]
Avi Kivity6aa8b732006-12-10 02:21:36 -08001/*
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
20#include <linux/kvm.h>
21#include <linux/module.h>
22#include <linux/errno.h>
23#include <asm/processor.h>
24#include <linux/percpu.h>
25#include <linux/gfp.h>
26#include <asm/msr.h>
27#include <linux/mm.h>
28#include <linux/miscdevice.h>
29#include <linux/vmalloc.h>
30#include <asm/uaccess.h>
31#include <linux/reboot.h>
32#include <asm/io.h>
33#include <linux/debugfs.h>
34#include <linux/highmem.h>
35#include <linux/file.h>
36#include <asm/desc.h>
37
38#include "x86_emulate.h"
39#include "segment_descriptor.h"
40
41MODULE_AUTHOR("Qumranet");
42MODULE_LICENSE("GPL");
43
44struct kvm_arch_ops *kvm_arch_ops;
45struct kvm_stat kvm_stat;
46EXPORT_SYMBOL_GPL(kvm_stat);
47
48static struct kvm_stats_debugfs_item {
49 const char *name;
50 u32 *data;
51 struct dentry *dentry;
52} debugfs_entries[] = {
53 { "pf_fixed", &kvm_stat.pf_fixed },
54 { "pf_guest", &kvm_stat.pf_guest },
55 { "tlb_flush", &kvm_stat.tlb_flush },
56 { "invlpg", &kvm_stat.invlpg },
57 { "exits", &kvm_stat.exits },
58 { "io_exits", &kvm_stat.io_exits },
59 { "mmio_exits", &kvm_stat.mmio_exits },
60 { "signal_exits", &kvm_stat.signal_exits },
61 { "irq_exits", &kvm_stat.irq_exits },
62 { 0, 0 }
63};
64
65static struct dentry *debugfs_dir;
66
67#define MAX_IO_MSRS 256
68
69#define CR0_RESEVED_BITS 0xffffffff1ffaffc0ULL
70#define LMSW_GUEST_MASK 0x0eULL
71#define CR4_RESEVED_BITS (~((1ULL << 11) - 1))
72#define CR8_RESEVED_BITS (~0x0fULL)
73#define EFER_RESERVED_BITS 0xfffffffffffff2fe
74
75struct vmx_msr_entry *find_msr_entry(struct kvm_vcpu *vcpu, u32 msr)
76{
77 int i;
78
79 for (i = 0; i < vcpu->nmsrs; ++i)
80 if (vcpu->guest_msrs[i].index == msr)
81 return &vcpu->guest_msrs[i];
82 return 0;
83}
84EXPORT_SYMBOL_GPL(find_msr_entry);
85
86#ifdef __x86_64__
87// LDT or TSS descriptor in the GDT. 16 bytes.
88struct segment_descriptor_64 {
89 struct segment_descriptor s;
90 u32 base_higher;
91 u32 pad_zero;
92};
93
94#endif
95
96unsigned long segment_base(u16 selector)
97{
98 struct descriptor_table gdt;
99 struct segment_descriptor *d;
100 unsigned long table_base;
101 typedef unsigned long ul;
102 unsigned long v;
103
104 if (selector == 0)
105 return 0;
106
107 asm ("sgdt %0" : "=m"(gdt));
108 table_base = gdt.base;
109
110 if (selector & 4) { /* from ldt */
111 u16 ldt_selector;
112
113 asm ("sldt %0" : "=g"(ldt_selector));
114 table_base = segment_base(ldt_selector);
115 }
116 d = (struct segment_descriptor *)(table_base + (selector & ~7));
117 v = d->base_low | ((ul)d->base_mid << 16) | ((ul)d->base_high << 24);
118#ifdef __x86_64__
119 if (d->system == 0
120 && (d->type == 2 || d->type == 9 || d->type == 11))
121 v |= ((ul)((struct segment_descriptor_64 *)d)->base_higher) << 32;
122#endif
123 return v;
124}
125EXPORT_SYMBOL_GPL(segment_base);
126
127int kvm_read_guest(struct kvm_vcpu *vcpu,
128 gva_t addr,
129 unsigned long size,
130 void *dest)
131{
132 unsigned char *host_buf = dest;
133 unsigned long req_size = size;
134
135 while (size) {
136 hpa_t paddr;
137 unsigned now;
138 unsigned offset;
139 hva_t guest_buf;
140
141 paddr = gva_to_hpa(vcpu, addr);
142
143 if (is_error_hpa(paddr))
144 break;
145
146 guest_buf = (hva_t)kmap_atomic(
147 pfn_to_page(paddr >> PAGE_SHIFT),
148 KM_USER0);
149 offset = addr & ~PAGE_MASK;
150 guest_buf |= offset;
151 now = min(size, PAGE_SIZE - offset);
152 memcpy(host_buf, (void*)guest_buf, now);
153 host_buf += now;
154 addr += now;
155 size -= now;
156 kunmap_atomic((void *)(guest_buf & PAGE_MASK), KM_USER0);
157 }
158 return req_size - size;
159}
160EXPORT_SYMBOL_GPL(kvm_read_guest);
161
162int kvm_write_guest(struct kvm_vcpu *vcpu,
163 gva_t addr,
164 unsigned long size,
165 void *data)
166{
167 unsigned char *host_buf = data;
168 unsigned long req_size = size;
169
170 while (size) {
171 hpa_t paddr;
172 unsigned now;
173 unsigned offset;
174 hva_t guest_buf;
175
176 paddr = gva_to_hpa(vcpu, addr);
177
178 if (is_error_hpa(paddr))
179 break;
180
181 guest_buf = (hva_t)kmap_atomic(
182 pfn_to_page(paddr >> PAGE_SHIFT), KM_USER0);
183 offset = addr & ~PAGE_MASK;
184 guest_buf |= offset;
185 now = min(size, PAGE_SIZE - offset);
186 memcpy((void*)guest_buf, host_buf, now);
187 host_buf += now;
188 addr += now;
189 size -= now;
190 kunmap_atomic((void *)(guest_buf & PAGE_MASK), KM_USER0);
191 }
192 return req_size - size;
193}
194EXPORT_SYMBOL_GPL(kvm_write_guest);
195
196static int vcpu_slot(struct kvm_vcpu *vcpu)
197{
198 return vcpu - vcpu->kvm->vcpus;
199}
200
201/*
202 * Switches to specified vcpu, until a matching vcpu_put()
203 */
204static struct kvm_vcpu *vcpu_load(struct kvm *kvm, int vcpu_slot)
205{
206 struct kvm_vcpu *vcpu = &kvm->vcpus[vcpu_slot];
207
208 mutex_lock(&vcpu->mutex);
209 if (unlikely(!vcpu->vmcs)) {
210 mutex_unlock(&vcpu->mutex);
211 return 0;
212 }
213 return kvm_arch_ops->vcpu_load(vcpu);
214}
215
216static void vcpu_put(struct kvm_vcpu *vcpu)
217{
218 kvm_arch_ops->vcpu_put(vcpu);
219 put_cpu();
220 mutex_unlock(&vcpu->mutex);
221}
222
223static int kvm_dev_open(struct inode *inode, struct file *filp)
224{
225 struct kvm *kvm = kzalloc(sizeof(struct kvm), GFP_KERNEL);
226 int i;
227
228 if (!kvm)
229 return -ENOMEM;
230
231 spin_lock_init(&kvm->lock);
232 INIT_LIST_HEAD(&kvm->active_mmu_pages);
233 for (i = 0; i < KVM_MAX_VCPUS; ++i) {
234 struct kvm_vcpu *vcpu = &kvm->vcpus[i];
235
236 mutex_init(&vcpu->mutex);
237 vcpu->mmu.root_hpa = INVALID_PAGE;
238 INIT_LIST_HEAD(&vcpu->free_pages);
239 }
240 filp->private_data = kvm;
241 return 0;
242}
243
244/*
245 * Free any memory in @free but not in @dont.
246 */
247static void kvm_free_physmem_slot(struct kvm_memory_slot *free,
248 struct kvm_memory_slot *dont)
249{
250 int i;
251
252 if (!dont || free->phys_mem != dont->phys_mem)
253 if (free->phys_mem) {
254 for (i = 0; i < free->npages; ++i)
255 __free_page(free->phys_mem[i]);
256 vfree(free->phys_mem);
257 }
258
259 if (!dont || free->dirty_bitmap != dont->dirty_bitmap)
260 vfree(free->dirty_bitmap);
261
262 free->phys_mem = 0;
263 free->npages = 0;
264 free->dirty_bitmap = 0;
265}
266
267static void kvm_free_physmem(struct kvm *kvm)
268{
269 int i;
270
271 for (i = 0; i < kvm->nmemslots; ++i)
272 kvm_free_physmem_slot(&kvm->memslots[i], 0);
273}
274
275static void kvm_free_vcpu(struct kvm_vcpu *vcpu)
276{
277 kvm_arch_ops->vcpu_free(vcpu);
278 kvm_mmu_destroy(vcpu);
279}
280
281static void kvm_free_vcpus(struct kvm *kvm)
282{
283 unsigned int i;
284
285 for (i = 0; i < KVM_MAX_VCPUS; ++i)
286 kvm_free_vcpu(&kvm->vcpus[i]);
287}
288
289static int kvm_dev_release(struct inode *inode, struct file *filp)
290{
291 struct kvm *kvm = filp->private_data;
292
293 kvm_free_vcpus(kvm);
294 kvm_free_physmem(kvm);
295 kfree(kvm);
296 return 0;
297}
298
299static void inject_gp(struct kvm_vcpu *vcpu)
300{
301 kvm_arch_ops->inject_gp(vcpu, 0);
302}
303
304static int pdptrs_have_reserved_bits_set(struct kvm_vcpu *vcpu,
305 unsigned long cr3)
306{
307 gfn_t pdpt_gfn = cr3 >> PAGE_SHIFT;
308 unsigned offset = (cr3 & (PAGE_SIZE-1)) >> 5;
309 int i;
310 u64 pdpte;
311 u64 *pdpt;
312 struct kvm_memory_slot *memslot;
313
314 spin_lock(&vcpu->kvm->lock);
315 memslot = gfn_to_memslot(vcpu->kvm, pdpt_gfn);
316 /* FIXME: !memslot - emulate? 0xff? */
317 pdpt = kmap_atomic(gfn_to_page(memslot, pdpt_gfn), KM_USER0);
318
319 for (i = 0; i < 4; ++i) {
320 pdpte = pdpt[offset + i];
321 if ((pdpte & 1) && (pdpte & 0xfffffff0000001e6ull))
322 break;
323 }
324
325 kunmap_atomic(pdpt, KM_USER0);
326 spin_unlock(&vcpu->kvm->lock);
327
328 return i != 4;
329}
330
331void set_cr0(struct kvm_vcpu *vcpu, unsigned long cr0)
332{
333 if (cr0 & CR0_RESEVED_BITS) {
334 printk(KERN_DEBUG "set_cr0: 0x%lx #GP, reserved bits 0x%lx\n",
335 cr0, vcpu->cr0);
336 inject_gp(vcpu);
337 return;
338 }
339
340 if ((cr0 & CR0_NW_MASK) && !(cr0 & CR0_CD_MASK)) {
341 printk(KERN_DEBUG "set_cr0: #GP, CD == 0 && NW == 1\n");
342 inject_gp(vcpu);
343 return;
344 }
345
346 if ((cr0 & CR0_PG_MASK) && !(cr0 & CR0_PE_MASK)) {
347 printk(KERN_DEBUG "set_cr0: #GP, set PG flag "
348 "and a clear PE flag\n");
349 inject_gp(vcpu);
350 return;
351 }
352
353 if (!is_paging(vcpu) && (cr0 & CR0_PG_MASK)) {
354#ifdef __x86_64__
355 if ((vcpu->shadow_efer & EFER_LME)) {
356 int cs_db, cs_l;
357
358 if (!is_pae(vcpu)) {
359 printk(KERN_DEBUG "set_cr0: #GP, start paging "
360 "in long mode while PAE is disabled\n");
361 inject_gp(vcpu);
362 return;
363 }
364 kvm_arch_ops->get_cs_db_l_bits(vcpu, &cs_db, &cs_l);
365 if (cs_l) {
366 printk(KERN_DEBUG "set_cr0: #GP, start paging "
367 "in long mode while CS.L == 1\n");
368 inject_gp(vcpu);
369 return;
370
371 }
372 } else
373#endif
374 if (is_pae(vcpu) &&
375 pdptrs_have_reserved_bits_set(vcpu, vcpu->cr3)) {
376 printk(KERN_DEBUG "set_cr0: #GP, pdptrs "
377 "reserved bits\n");
378 inject_gp(vcpu);
379 return;
380 }
381
382 }
383
384 kvm_arch_ops->set_cr0(vcpu, cr0);
385 vcpu->cr0 = cr0;
386
387 spin_lock(&vcpu->kvm->lock);
388 kvm_mmu_reset_context(vcpu);
389 spin_unlock(&vcpu->kvm->lock);
390 return;
391}
392EXPORT_SYMBOL_GPL(set_cr0);
393
394void lmsw(struct kvm_vcpu *vcpu, unsigned long msw)
395{
396 set_cr0(vcpu, (vcpu->cr0 & ~0x0ful) | (msw & 0x0f));
397}
398EXPORT_SYMBOL_GPL(lmsw);
399
400void set_cr4(struct kvm_vcpu *vcpu, unsigned long cr4)
401{
402 if (cr4 & CR4_RESEVED_BITS) {
403 printk(KERN_DEBUG "set_cr4: #GP, reserved bits\n");
404 inject_gp(vcpu);
405 return;
406 }
407
408 if (kvm_arch_ops->is_long_mode(vcpu)) {
409 if (!(cr4 & CR4_PAE_MASK)) {
410 printk(KERN_DEBUG "set_cr4: #GP, clearing PAE while "
411 "in long mode\n");
412 inject_gp(vcpu);
413 return;
414 }
415 } else if (is_paging(vcpu) && !is_pae(vcpu) && (cr4 & CR4_PAE_MASK)
416 && pdptrs_have_reserved_bits_set(vcpu, vcpu->cr3)) {
417 printk(KERN_DEBUG "set_cr4: #GP, pdptrs reserved bits\n");
418 inject_gp(vcpu);
419 }
420
421 if (cr4 & CR4_VMXE_MASK) {
422 printk(KERN_DEBUG "set_cr4: #GP, setting VMXE\n");
423 inject_gp(vcpu);
424 return;
425 }
426 kvm_arch_ops->set_cr4(vcpu, cr4);
427 spin_lock(&vcpu->kvm->lock);
428 kvm_mmu_reset_context(vcpu);
429 spin_unlock(&vcpu->kvm->lock);
430}
431EXPORT_SYMBOL_GPL(set_cr4);
432
433void set_cr3(struct kvm_vcpu *vcpu, unsigned long cr3)
434{
435 if (kvm_arch_ops->is_long_mode(vcpu)) {
436 if ( cr3 & CR3_L_MODE_RESEVED_BITS) {
437 printk(KERN_DEBUG "set_cr3: #GP, reserved bits\n");
438 inject_gp(vcpu);
439 return;
440 }
441 } else {
442 if (cr3 & CR3_RESEVED_BITS) {
443 printk(KERN_DEBUG "set_cr3: #GP, reserved bits\n");
444 inject_gp(vcpu);
445 return;
446 }
447 if (is_paging(vcpu) && is_pae(vcpu) &&
448 pdptrs_have_reserved_bits_set(vcpu, cr3)) {
449 printk(KERN_DEBUG "set_cr3: #GP, pdptrs "
450 "reserved bits\n");
451 inject_gp(vcpu);
452 return;
453 }
454 }
455
456 vcpu->cr3 = cr3;
457 spin_lock(&vcpu->kvm->lock);
458 vcpu->mmu.new_cr3(vcpu);
459 spin_unlock(&vcpu->kvm->lock);
460}
461EXPORT_SYMBOL_GPL(set_cr3);
462
463void set_cr8(struct kvm_vcpu *vcpu, unsigned long cr8)
464{
465 if ( cr8 & CR8_RESEVED_BITS) {
466 printk(KERN_DEBUG "set_cr8: #GP, reserved bits 0x%lx\n", cr8);
467 inject_gp(vcpu);
468 return;
469 }
470 vcpu->cr8 = cr8;
471}
472EXPORT_SYMBOL_GPL(set_cr8);
473
474void fx_init(struct kvm_vcpu *vcpu)
475{
476 struct __attribute__ ((__packed__)) fx_image_s {
477 u16 control; //fcw
478 u16 status; //fsw
479 u16 tag; // ftw
480 u16 opcode; //fop
481 u64 ip; // fpu ip
482 u64 operand;// fpu dp
483 u32 mxcsr;
484 u32 mxcsr_mask;
485
486 } *fx_image;
487
488 fx_save(vcpu->host_fx_image);
489 fpu_init();
490 fx_save(vcpu->guest_fx_image);
491 fx_restore(vcpu->host_fx_image);
492
493 fx_image = (struct fx_image_s *)vcpu->guest_fx_image;
494 fx_image->mxcsr = 0x1f80;
495 memset(vcpu->guest_fx_image + sizeof(struct fx_image_s),
496 0, FX_IMAGE_SIZE - sizeof(struct fx_image_s));
497}
498EXPORT_SYMBOL_GPL(fx_init);
499
500/*
501 * Creates some virtual cpus. Good luck creating more than one.
502 */
503static int kvm_dev_ioctl_create_vcpu(struct kvm *kvm, int n)
504{
505 int r;
506 struct kvm_vcpu *vcpu;
507
508 r = -EINVAL;
509 if (n < 0 || n >= KVM_MAX_VCPUS)
510 goto out;
511
512 vcpu = &kvm->vcpus[n];
513
514 mutex_lock(&vcpu->mutex);
515
516 if (vcpu->vmcs) {
517 mutex_unlock(&vcpu->mutex);
518 return -EEXIST;
519 }
520
521 vcpu->host_fx_image = (char*)ALIGN((hva_t)vcpu->fx_buf,
522 FX_IMAGE_ALIGN);
523 vcpu->guest_fx_image = vcpu->host_fx_image + FX_IMAGE_SIZE;
524
525 vcpu->cpu = -1; /* First load will set up TR */
526 vcpu->kvm = kvm;
527 r = kvm_arch_ops->vcpu_create(vcpu);
528 if (r < 0)
529 goto out_free_vcpus;
530
531 kvm_arch_ops->vcpu_load(vcpu);
532
533 r = kvm_arch_ops->vcpu_setup(vcpu);
534 if (r >= 0)
535 r = kvm_mmu_init(vcpu);
536
537 vcpu_put(vcpu);
538
539 if (r < 0)
540 goto out_free_vcpus;
541
542 return 0;
543
544out_free_vcpus:
545 kvm_free_vcpu(vcpu);
546 mutex_unlock(&vcpu->mutex);
547out:
548 return r;
549}
550
551/*
552 * Allocate some memory and give it an address in the guest physical address
553 * space.
554 *
555 * Discontiguous memory is allowed, mostly for framebuffers.
556 */
557static int kvm_dev_ioctl_set_memory_region(struct kvm *kvm,
558 struct kvm_memory_region *mem)
559{
560 int r;
561 gfn_t base_gfn;
562 unsigned long npages;
563 unsigned long i;
564 struct kvm_memory_slot *memslot;
565 struct kvm_memory_slot old, new;
566 int memory_config_version;
567
568 r = -EINVAL;
569 /* General sanity checks */
570 if (mem->memory_size & (PAGE_SIZE - 1))
571 goto out;
572 if (mem->guest_phys_addr & (PAGE_SIZE - 1))
573 goto out;
574 if (mem->slot >= KVM_MEMORY_SLOTS)
575 goto out;
576 if (mem->guest_phys_addr + mem->memory_size < mem->guest_phys_addr)
577 goto out;
578
579 memslot = &kvm->memslots[mem->slot];
580 base_gfn = mem->guest_phys_addr >> PAGE_SHIFT;
581 npages = mem->memory_size >> PAGE_SHIFT;
582
583 if (!npages)
584 mem->flags &= ~KVM_MEM_LOG_DIRTY_PAGES;
585
586raced:
587 spin_lock(&kvm->lock);
588
589 memory_config_version = kvm->memory_config_version;
590 new = old = *memslot;
591
592 new.base_gfn = base_gfn;
593 new.npages = npages;
594 new.flags = mem->flags;
595
596 /* Disallow changing a memory slot's size. */
597 r = -EINVAL;
598 if (npages && old.npages && npages != old.npages)
599 goto out_unlock;
600
601 /* Check for overlaps */
602 r = -EEXIST;
603 for (i = 0; i < KVM_MEMORY_SLOTS; ++i) {
604 struct kvm_memory_slot *s = &kvm->memslots[i];
605
606 if (s == memslot)
607 continue;
608 if (!((base_gfn + npages <= s->base_gfn) ||
609 (base_gfn >= s->base_gfn + s->npages)))
610 goto out_unlock;
611 }
612 /*
613 * Do memory allocations outside lock. memory_config_version will
614 * detect any races.
615 */
616 spin_unlock(&kvm->lock);
617
618 /* Deallocate if slot is being removed */
619 if (!npages)
620 new.phys_mem = 0;
621
622 /* Free page dirty bitmap if unneeded */
623 if (!(new.flags & KVM_MEM_LOG_DIRTY_PAGES))
624 new.dirty_bitmap = 0;
625
626 r = -ENOMEM;
627
628 /* Allocate if a slot is being created */
629 if (npages && !new.phys_mem) {
630 new.phys_mem = vmalloc(npages * sizeof(struct page *));
631
632 if (!new.phys_mem)
633 goto out_free;
634
635 memset(new.phys_mem, 0, npages * sizeof(struct page *));
636 for (i = 0; i < npages; ++i) {
637 new.phys_mem[i] = alloc_page(GFP_HIGHUSER
638 | __GFP_ZERO);
639 if (!new.phys_mem[i])
640 goto out_free;
641 }
642 }
643
644 /* Allocate page dirty bitmap if needed */
645 if ((new.flags & KVM_MEM_LOG_DIRTY_PAGES) && !new.dirty_bitmap) {
646 unsigned dirty_bytes = ALIGN(npages, BITS_PER_LONG) / 8;
647
648 new.dirty_bitmap = vmalloc(dirty_bytes);
649 if (!new.dirty_bitmap)
650 goto out_free;
651 memset(new.dirty_bitmap, 0, dirty_bytes);
652 }
653
654 spin_lock(&kvm->lock);
655
656 if (memory_config_version != kvm->memory_config_version) {
657 spin_unlock(&kvm->lock);
658 kvm_free_physmem_slot(&new, &old);
659 goto raced;
660 }
661
662 r = -EAGAIN;
663 if (kvm->busy)
664 goto out_unlock;
665
666 if (mem->slot >= kvm->nmemslots)
667 kvm->nmemslots = mem->slot + 1;
668
669 *memslot = new;
670 ++kvm->memory_config_version;
671
672 spin_unlock(&kvm->lock);
673
674 for (i = 0; i < KVM_MAX_VCPUS; ++i) {
675 struct kvm_vcpu *vcpu;
676
677 vcpu = vcpu_load(kvm, i);
678 if (!vcpu)
679 continue;
680 kvm_mmu_reset_context(vcpu);
681 vcpu_put(vcpu);
682 }
683
684 kvm_free_physmem_slot(&old, &new);
685 return 0;
686
687out_unlock:
688 spin_unlock(&kvm->lock);
689out_free:
690 kvm_free_physmem_slot(&new, &old);
691out:
692 return r;
693}
694
695/*
696 * Get (and clear) the dirty memory log for a memory slot.
697 */
698static int kvm_dev_ioctl_get_dirty_log(struct kvm *kvm,
699 struct kvm_dirty_log *log)
700{
701 struct kvm_memory_slot *memslot;
702 int r, i;
703 int n;
704 unsigned long any = 0;
705
706 spin_lock(&kvm->lock);
707
708 /*
709 * Prevent changes to guest memory configuration even while the lock
710 * is not taken.
711 */
712 ++kvm->busy;
713 spin_unlock(&kvm->lock);
714 r = -EINVAL;
715 if (log->slot >= KVM_MEMORY_SLOTS)
716 goto out;
717
718 memslot = &kvm->memslots[log->slot];
719 r = -ENOENT;
720 if (!memslot->dirty_bitmap)
721 goto out;
722
723 n = ALIGN(memslot->npages, 8) / 8;
724
725 for (i = 0; !any && i < n; ++i)
726 any = memslot->dirty_bitmap[i];
727
728 r = -EFAULT;
729 if (copy_to_user(log->dirty_bitmap, memslot->dirty_bitmap, n))
730 goto out;
731
732
733 if (any) {
734 spin_lock(&kvm->lock);
735 kvm_mmu_slot_remove_write_access(kvm, log->slot);
736 spin_unlock(&kvm->lock);
737 memset(memslot->dirty_bitmap, 0, n);
738 for (i = 0; i < KVM_MAX_VCPUS; ++i) {
739 struct kvm_vcpu *vcpu = vcpu_load(kvm, i);
740
741 if (!vcpu)
742 continue;
743 kvm_arch_ops->tlb_flush(vcpu);
744 vcpu_put(vcpu);
745 }
746 }
747
748 r = 0;
749
750out:
751 spin_lock(&kvm->lock);
752 --kvm->busy;
753 spin_unlock(&kvm->lock);
754 return r;
755}
756
757struct kvm_memory_slot *gfn_to_memslot(struct kvm *kvm, gfn_t gfn)
758{
759 int i;
760
761 for (i = 0; i < kvm->nmemslots; ++i) {
762 struct kvm_memory_slot *memslot = &kvm->memslots[i];
763
764 if (gfn >= memslot->base_gfn
765 && gfn < memslot->base_gfn + memslot->npages)
766 return memslot;
767 }
768 return 0;
769}
770EXPORT_SYMBOL_GPL(gfn_to_memslot);
771
772void mark_page_dirty(struct kvm *kvm, gfn_t gfn)
773{
774 int i;
775 struct kvm_memory_slot *memslot = 0;
776 unsigned long rel_gfn;
777
778 for (i = 0; i < kvm->nmemslots; ++i) {
779 memslot = &kvm->memslots[i];
780
781 if (gfn >= memslot->base_gfn
782 && gfn < memslot->base_gfn + memslot->npages) {
783
784 if (!memslot || !memslot->dirty_bitmap)
785 return;
786
787 rel_gfn = gfn - memslot->base_gfn;
788
789 /* avoid RMW */
790 if (!test_bit(rel_gfn, memslot->dirty_bitmap))
791 set_bit(rel_gfn, memslot->dirty_bitmap);
792 return;
793 }
794 }
795}
796
797static int emulator_read_std(unsigned long addr,
798 unsigned long *val,
799 unsigned int bytes,
800 struct x86_emulate_ctxt *ctxt)
801{
802 struct kvm_vcpu *vcpu = ctxt->vcpu;
803 void *data = val;
804
805 while (bytes) {
806 gpa_t gpa = vcpu->mmu.gva_to_gpa(vcpu, addr);
807 unsigned offset = addr & (PAGE_SIZE-1);
808 unsigned tocopy = min(bytes, (unsigned)PAGE_SIZE - offset);
809 unsigned long pfn;
810 struct kvm_memory_slot *memslot;
811 void *page;
812
813 if (gpa == UNMAPPED_GVA)
814 return X86EMUL_PROPAGATE_FAULT;
815 pfn = gpa >> PAGE_SHIFT;
816 memslot = gfn_to_memslot(vcpu->kvm, pfn);
817 if (!memslot)
818 return X86EMUL_UNHANDLEABLE;
819 page = kmap_atomic(gfn_to_page(memslot, pfn), KM_USER0);
820
821 memcpy(data, page + offset, tocopy);
822
823 kunmap_atomic(page, KM_USER0);
824
825 bytes -= tocopy;
826 data += tocopy;
827 addr += tocopy;
828 }
829
830 return X86EMUL_CONTINUE;
831}
832
833static int emulator_write_std(unsigned long addr,
834 unsigned long val,
835 unsigned int bytes,
836 struct x86_emulate_ctxt *ctxt)
837{
838 printk(KERN_ERR "emulator_write_std: addr %lx n %d\n",
839 addr, bytes);
840 return X86EMUL_UNHANDLEABLE;
841}
842
843static int emulator_read_emulated(unsigned long addr,
844 unsigned long *val,
845 unsigned int bytes,
846 struct x86_emulate_ctxt *ctxt)
847{
848 struct kvm_vcpu *vcpu = ctxt->vcpu;
849
850 if (vcpu->mmio_read_completed) {
851 memcpy(val, vcpu->mmio_data, bytes);
852 vcpu->mmio_read_completed = 0;
853 return X86EMUL_CONTINUE;
854 } else if (emulator_read_std(addr, val, bytes, ctxt)
855 == X86EMUL_CONTINUE)
856 return X86EMUL_CONTINUE;
857 else {
858 gpa_t gpa = vcpu->mmu.gva_to_gpa(vcpu, addr);
859 if (gpa == UNMAPPED_GVA)
860 return vcpu_printf(vcpu, "not present\n"), X86EMUL_PROPAGATE_FAULT;
861 vcpu->mmio_needed = 1;
862 vcpu->mmio_phys_addr = gpa;
863 vcpu->mmio_size = bytes;
864 vcpu->mmio_is_write = 0;
865
866 return X86EMUL_UNHANDLEABLE;
867 }
868}
869
870static int emulator_write_emulated(unsigned long addr,
871 unsigned long val,
872 unsigned int bytes,
873 struct x86_emulate_ctxt *ctxt)
874{
875 struct kvm_vcpu *vcpu = ctxt->vcpu;
876 gpa_t gpa = vcpu->mmu.gva_to_gpa(vcpu, addr);
877
878 if (gpa == UNMAPPED_GVA)
879 return X86EMUL_PROPAGATE_FAULT;
880
881 vcpu->mmio_needed = 1;
882 vcpu->mmio_phys_addr = gpa;
883 vcpu->mmio_size = bytes;
884 vcpu->mmio_is_write = 1;
885 memcpy(vcpu->mmio_data, &val, bytes);
886
887 return X86EMUL_CONTINUE;
888}
889
890static int emulator_cmpxchg_emulated(unsigned long addr,
891 unsigned long old,
892 unsigned long new,
893 unsigned int bytes,
894 struct x86_emulate_ctxt *ctxt)
895{
896 static int reported;
897
898 if (!reported) {
899 reported = 1;
900 printk(KERN_WARNING "kvm: emulating exchange as write\n");
901 }
902 return emulator_write_emulated(addr, new, bytes, ctxt);
903}
904
905static unsigned long get_segment_base(struct kvm_vcpu *vcpu, int seg)
906{
907 return kvm_arch_ops->get_segment_base(vcpu, seg);
908}
909
910int emulate_invlpg(struct kvm_vcpu *vcpu, gva_t address)
911{
912 spin_lock(&vcpu->kvm->lock);
913 vcpu->mmu.inval_page(vcpu, address);
914 spin_unlock(&vcpu->kvm->lock);
915 kvm_arch_ops->invlpg(vcpu, address);
916 return X86EMUL_CONTINUE;
917}
918
919int emulate_clts(struct kvm_vcpu *vcpu)
920{
921 unsigned long cr0 = vcpu->cr0;
922
923 cr0 &= ~CR0_TS_MASK;
924 kvm_arch_ops->set_cr0(vcpu, cr0);
925 return X86EMUL_CONTINUE;
926}
927
928int emulator_get_dr(struct x86_emulate_ctxt* ctxt, int dr, unsigned long *dest)
929{
930 struct kvm_vcpu *vcpu = ctxt->vcpu;
931
932 switch (dr) {
933 case 0 ... 3:
934 *dest = kvm_arch_ops->get_dr(vcpu, dr);
935 return X86EMUL_CONTINUE;
936 default:
937 printk(KERN_DEBUG "%s: unexpected dr %u\n",
938 __FUNCTION__, dr);
939 return X86EMUL_UNHANDLEABLE;
940 }
941}
942
943int emulator_set_dr(struct x86_emulate_ctxt *ctxt, int dr, unsigned long value)
944{
945 unsigned long mask = (ctxt->mode == X86EMUL_MODE_PROT64) ? ~0ULL : ~0U;
946 int exception;
947
948 kvm_arch_ops->set_dr(ctxt->vcpu, dr, value & mask, &exception);
949 if (exception) {
950 /* FIXME: better handling */
951 return X86EMUL_UNHANDLEABLE;
952 }
953 return X86EMUL_CONTINUE;
954}
955
956static void report_emulation_failure(struct x86_emulate_ctxt *ctxt)
957{
958 static int reported;
959 u8 opcodes[4];
960 unsigned long rip = ctxt->vcpu->rip;
961 unsigned long rip_linear;
962
963 rip_linear = rip + get_segment_base(ctxt->vcpu, VCPU_SREG_CS);
964
965 if (reported)
966 return;
967
968 emulator_read_std(rip_linear, (void *)opcodes, 4, ctxt);
969
970 printk(KERN_ERR "emulation failed but !mmio_needed?"
971 " rip %lx %02x %02x %02x %02x\n",
972 rip, opcodes[0], opcodes[1], opcodes[2], opcodes[3]);
973 reported = 1;
974}
975
976struct x86_emulate_ops emulate_ops = {
977 .read_std = emulator_read_std,
978 .write_std = emulator_write_std,
979 .read_emulated = emulator_read_emulated,
980 .write_emulated = emulator_write_emulated,
981 .cmpxchg_emulated = emulator_cmpxchg_emulated,
982};
983
984int emulate_instruction(struct kvm_vcpu *vcpu,
985 struct kvm_run *run,
986 unsigned long cr2,
987 u16 error_code)
988{
989 struct x86_emulate_ctxt emulate_ctxt;
990 int r;
991 int cs_db, cs_l;
992
993 kvm_arch_ops->cache_regs(vcpu);
994
995 kvm_arch_ops->get_cs_db_l_bits(vcpu, &cs_db, &cs_l);
996
997 emulate_ctxt.vcpu = vcpu;
998 emulate_ctxt.eflags = kvm_arch_ops->get_rflags(vcpu);
999 emulate_ctxt.cr2 = cr2;
1000 emulate_ctxt.mode = (emulate_ctxt.eflags & X86_EFLAGS_VM)
1001 ? X86EMUL_MODE_REAL : cs_l
1002 ? X86EMUL_MODE_PROT64 : cs_db
1003 ? X86EMUL_MODE_PROT32 : X86EMUL_MODE_PROT16;
1004
1005 if (emulate_ctxt.mode == X86EMUL_MODE_PROT64) {
1006 emulate_ctxt.cs_base = 0;
1007 emulate_ctxt.ds_base = 0;
1008 emulate_ctxt.es_base = 0;
1009 emulate_ctxt.ss_base = 0;
1010 } else {
1011 emulate_ctxt.cs_base = get_segment_base(vcpu, VCPU_SREG_CS);
1012 emulate_ctxt.ds_base = get_segment_base(vcpu, VCPU_SREG_DS);
1013 emulate_ctxt.es_base = get_segment_base(vcpu, VCPU_SREG_ES);
1014 emulate_ctxt.ss_base = get_segment_base(vcpu, VCPU_SREG_SS);
1015 }
1016
1017 emulate_ctxt.gs_base = get_segment_base(vcpu, VCPU_SREG_GS);
1018 emulate_ctxt.fs_base = get_segment_base(vcpu, VCPU_SREG_FS);
1019
1020 vcpu->mmio_is_write = 0;
1021 r = x86_emulate_memop(&emulate_ctxt, &emulate_ops);
1022
1023 if ((r || vcpu->mmio_is_write) && run) {
1024 run->mmio.phys_addr = vcpu->mmio_phys_addr;
1025 memcpy(run->mmio.data, vcpu->mmio_data, 8);
1026 run->mmio.len = vcpu->mmio_size;
1027 run->mmio.is_write = vcpu->mmio_is_write;
1028 }
1029
1030 if (r) {
1031 if (!vcpu->mmio_needed) {
1032 report_emulation_failure(&emulate_ctxt);
1033 return EMULATE_FAIL;
1034 }
1035 return EMULATE_DO_MMIO;
1036 }
1037
1038 kvm_arch_ops->decache_regs(vcpu);
1039 kvm_arch_ops->set_rflags(vcpu, emulate_ctxt.eflags);
1040
1041 if (vcpu->mmio_is_write)
1042 return EMULATE_DO_MMIO;
1043
1044 return EMULATE_DONE;
1045}
1046EXPORT_SYMBOL_GPL(emulate_instruction);
1047
1048static u64 mk_cr_64(u64 curr_cr, u32 new_val)
1049{
1050 return (curr_cr & ~((1ULL << 32) - 1)) | new_val;
1051}
1052
1053void realmode_lgdt(struct kvm_vcpu *vcpu, u16 limit, unsigned long base)
1054{
1055 struct descriptor_table dt = { limit, base };
1056
1057 kvm_arch_ops->set_gdt(vcpu, &dt);
1058}
1059
1060void realmode_lidt(struct kvm_vcpu *vcpu, u16 limit, unsigned long base)
1061{
1062 struct descriptor_table dt = { limit, base };
1063
1064 kvm_arch_ops->set_idt(vcpu, &dt);
1065}
1066
1067void realmode_lmsw(struct kvm_vcpu *vcpu, unsigned long msw,
1068 unsigned long *rflags)
1069{
1070 lmsw(vcpu, msw);
1071 *rflags = kvm_arch_ops->get_rflags(vcpu);
1072}
1073
1074unsigned long realmode_get_cr(struct kvm_vcpu *vcpu, int cr)
1075{
1076 switch (cr) {
1077 case 0:
1078 return vcpu->cr0;
1079 case 2:
1080 return vcpu->cr2;
1081 case 3:
1082 return vcpu->cr3;
1083 case 4:
1084 return vcpu->cr4;
1085 default:
1086 vcpu_printf(vcpu, "%s: unexpected cr %u\n", __FUNCTION__, cr);
1087 return 0;
1088 }
1089}
1090
1091void realmode_set_cr(struct kvm_vcpu *vcpu, int cr, unsigned long val,
1092 unsigned long *rflags)
1093{
1094 switch (cr) {
1095 case 0:
1096 set_cr0(vcpu, mk_cr_64(vcpu->cr0, val));
1097 *rflags = kvm_arch_ops->get_rflags(vcpu);
1098 break;
1099 case 2:
1100 vcpu->cr2 = val;
1101 break;
1102 case 3:
1103 set_cr3(vcpu, val);
1104 break;
1105 case 4:
1106 set_cr4(vcpu, mk_cr_64(vcpu->cr4, val));
1107 break;
1108 default:
1109 vcpu_printf(vcpu, "%s: unexpected cr %u\n", __FUNCTION__, cr);
1110 }
1111}
1112
1113/*
1114 * Reads an msr value (of 'msr_index') into 'pdata'.
1115 * Returns 0 on success, non-0 otherwise.
1116 * Assumes vcpu_load() was already called.
1117 */
1118static int get_msr(struct kvm_vcpu *vcpu, u32 msr_index, u64 *pdata)
1119{
1120 return kvm_arch_ops->get_msr(vcpu, msr_index, pdata);
1121}
1122
1123#ifdef __x86_64__
1124
1125void set_efer(struct kvm_vcpu *vcpu, u64 efer)
1126{
1127 struct vmx_msr_entry *msr;
1128
1129 if (efer & EFER_RESERVED_BITS) {
1130 printk(KERN_DEBUG "set_efer: 0x%llx #GP, reserved bits\n",
1131 efer);
1132 inject_gp(vcpu);
1133 return;
1134 }
1135
1136 if (is_paging(vcpu)
1137 && (vcpu->shadow_efer & EFER_LME) != (efer & EFER_LME)) {
1138 printk(KERN_DEBUG "set_efer: #GP, change LME while paging\n");
1139 inject_gp(vcpu);
1140 return;
1141 }
1142
1143 efer &= ~EFER_LMA;
1144 efer |= vcpu->shadow_efer & EFER_LMA;
1145
1146 vcpu->shadow_efer = efer;
1147
1148 msr = find_msr_entry(vcpu, MSR_EFER);
1149
1150 if (!(efer & EFER_LMA))
1151 efer &= ~EFER_LME;
1152 msr->data = efer;
1153}
1154EXPORT_SYMBOL_GPL(set_efer);
1155
1156#endif
1157
1158/*
1159 * Writes msr value into into the appropriate "register".
1160 * Returns 0 on success, non-0 otherwise.
1161 * Assumes vcpu_load() was already called.
1162 */
1163static int set_msr(struct kvm_vcpu *vcpu, u32 msr_index, u64 data)
1164{
1165 return kvm_arch_ops->set_msr(vcpu, msr_index, data);
1166}
1167
1168void kvm_resched(struct kvm_vcpu *vcpu)
1169{
1170 vcpu_put(vcpu);
1171 cond_resched();
1172 /* Cannot fail - no vcpu unplug yet. */
1173 vcpu_load(vcpu->kvm, vcpu_slot(vcpu));
1174}
1175EXPORT_SYMBOL_GPL(kvm_resched);
1176
1177void load_msrs(struct vmx_msr_entry *e, int n)
1178{
1179 int i;
1180
1181 for (i = 0; i < n; ++i)
1182 wrmsrl(e[i].index, e[i].data);
1183}
1184EXPORT_SYMBOL_GPL(load_msrs);
1185
1186void save_msrs(struct vmx_msr_entry *e, int n)
1187{
1188 int i;
1189
1190 for (i = 0; i < n; ++i)
1191 rdmsrl(e[i].index, e[i].data);
1192}
1193EXPORT_SYMBOL_GPL(save_msrs);
1194
1195static int kvm_dev_ioctl_run(struct kvm *kvm, struct kvm_run *kvm_run)
1196{
1197 struct kvm_vcpu *vcpu;
1198 int r;
1199
1200 if (kvm_run->vcpu < 0 || kvm_run->vcpu >= KVM_MAX_VCPUS)
1201 return -EINVAL;
1202
1203 vcpu = vcpu_load(kvm, kvm_run->vcpu);
1204 if (!vcpu)
1205 return -ENOENT;
1206
1207 if (kvm_run->emulated) {
1208 kvm_arch_ops->skip_emulated_instruction(vcpu);
1209 kvm_run->emulated = 0;
1210 }
1211
1212 if (kvm_run->mmio_completed) {
1213 memcpy(vcpu->mmio_data, kvm_run->mmio.data, 8);
1214 vcpu->mmio_read_completed = 1;
1215 }
1216
1217 vcpu->mmio_needed = 0;
1218
1219 r = kvm_arch_ops->run(vcpu, kvm_run);
1220
1221 vcpu_put(vcpu);
1222 return r;
1223}
1224
1225static int kvm_dev_ioctl_get_regs(struct kvm *kvm, struct kvm_regs *regs)
1226{
1227 struct kvm_vcpu *vcpu;
1228
1229 if (regs->vcpu < 0 || regs->vcpu >= KVM_MAX_VCPUS)
1230 return -EINVAL;
1231
1232 vcpu = vcpu_load(kvm, regs->vcpu);
1233 if (!vcpu)
1234 return -ENOENT;
1235
1236 kvm_arch_ops->cache_regs(vcpu);
1237
1238 regs->rax = vcpu->regs[VCPU_REGS_RAX];
1239 regs->rbx = vcpu->regs[VCPU_REGS_RBX];
1240 regs->rcx = vcpu->regs[VCPU_REGS_RCX];
1241 regs->rdx = vcpu->regs[VCPU_REGS_RDX];
1242 regs->rsi = vcpu->regs[VCPU_REGS_RSI];
1243 regs->rdi = vcpu->regs[VCPU_REGS_RDI];
1244 regs->rsp = vcpu->regs[VCPU_REGS_RSP];
1245 regs->rbp = vcpu->regs[VCPU_REGS_RBP];
1246#ifdef __x86_64__
1247 regs->r8 = vcpu->regs[VCPU_REGS_R8];
1248 regs->r9 = vcpu->regs[VCPU_REGS_R9];
1249 regs->r10 = vcpu->regs[VCPU_REGS_R10];
1250 regs->r11 = vcpu->regs[VCPU_REGS_R11];
1251 regs->r12 = vcpu->regs[VCPU_REGS_R12];
1252 regs->r13 = vcpu->regs[VCPU_REGS_R13];
1253 regs->r14 = vcpu->regs[VCPU_REGS_R14];
1254 regs->r15 = vcpu->regs[VCPU_REGS_R15];
1255#endif
1256
1257 regs->rip = vcpu->rip;
1258 regs->rflags = kvm_arch_ops->get_rflags(vcpu);
1259
1260 /*
1261 * Don't leak debug flags in case they were set for guest debugging
1262 */
1263 if (vcpu->guest_debug.enabled && vcpu->guest_debug.singlestep)
1264 regs->rflags &= ~(X86_EFLAGS_TF | X86_EFLAGS_RF);
1265
1266 vcpu_put(vcpu);
1267
1268 return 0;
1269}
1270
1271static int kvm_dev_ioctl_set_regs(struct kvm *kvm, struct kvm_regs *regs)
1272{
1273 struct kvm_vcpu *vcpu;
1274
1275 if (regs->vcpu < 0 || regs->vcpu >= KVM_MAX_VCPUS)
1276 return -EINVAL;
1277
1278 vcpu = vcpu_load(kvm, regs->vcpu);
1279 if (!vcpu)
1280 return -ENOENT;
1281
1282 vcpu->regs[VCPU_REGS_RAX] = regs->rax;
1283 vcpu->regs[VCPU_REGS_RBX] = regs->rbx;
1284 vcpu->regs[VCPU_REGS_RCX] = regs->rcx;
1285 vcpu->regs[VCPU_REGS_RDX] = regs->rdx;
1286 vcpu->regs[VCPU_REGS_RSI] = regs->rsi;
1287 vcpu->regs[VCPU_REGS_RDI] = regs->rdi;
1288 vcpu->regs[VCPU_REGS_RSP] = regs->rsp;
1289 vcpu->regs[VCPU_REGS_RBP] = regs->rbp;
1290#ifdef __x86_64__
1291 vcpu->regs[VCPU_REGS_R8] = regs->r8;
1292 vcpu->regs[VCPU_REGS_R9] = regs->r9;
1293 vcpu->regs[VCPU_REGS_R10] = regs->r10;
1294 vcpu->regs[VCPU_REGS_R11] = regs->r11;
1295 vcpu->regs[VCPU_REGS_R12] = regs->r12;
1296 vcpu->regs[VCPU_REGS_R13] = regs->r13;
1297 vcpu->regs[VCPU_REGS_R14] = regs->r14;
1298 vcpu->regs[VCPU_REGS_R15] = regs->r15;
1299#endif
1300
1301 vcpu->rip = regs->rip;
1302 kvm_arch_ops->set_rflags(vcpu, regs->rflags);
1303
1304 kvm_arch_ops->decache_regs(vcpu);
1305
1306 vcpu_put(vcpu);
1307
1308 return 0;
1309}
1310
1311static void get_segment(struct kvm_vcpu *vcpu,
1312 struct kvm_segment *var, int seg)
1313{
1314 return kvm_arch_ops->get_segment(vcpu, var, seg);
1315}
1316
1317static int kvm_dev_ioctl_get_sregs(struct kvm *kvm, struct kvm_sregs *sregs)
1318{
1319 struct kvm_vcpu *vcpu;
1320 struct descriptor_table dt;
1321
1322 if (sregs->vcpu < 0 || sregs->vcpu >= KVM_MAX_VCPUS)
1323 return -EINVAL;
1324 vcpu = vcpu_load(kvm, sregs->vcpu);
1325 if (!vcpu)
1326 return -ENOENT;
1327
1328 get_segment(vcpu, &sregs->cs, VCPU_SREG_CS);
1329 get_segment(vcpu, &sregs->ds, VCPU_SREG_DS);
1330 get_segment(vcpu, &sregs->es, VCPU_SREG_ES);
1331 get_segment(vcpu, &sregs->fs, VCPU_SREG_FS);
1332 get_segment(vcpu, &sregs->gs, VCPU_SREG_GS);
1333 get_segment(vcpu, &sregs->ss, VCPU_SREG_SS);
1334
1335 get_segment(vcpu, &sregs->tr, VCPU_SREG_TR);
1336 get_segment(vcpu, &sregs->ldt, VCPU_SREG_LDTR);
1337
1338 kvm_arch_ops->get_idt(vcpu, &dt);
1339 sregs->idt.limit = dt.limit;
1340 sregs->idt.base = dt.base;
1341 kvm_arch_ops->get_gdt(vcpu, &dt);
1342 sregs->gdt.limit = dt.limit;
1343 sregs->gdt.base = dt.base;
1344
1345 sregs->cr0 = vcpu->cr0;
1346 sregs->cr2 = vcpu->cr2;
1347 sregs->cr3 = vcpu->cr3;
1348 sregs->cr4 = vcpu->cr4;
1349 sregs->cr8 = vcpu->cr8;
1350 sregs->efer = vcpu->shadow_efer;
1351 sregs->apic_base = vcpu->apic_base;
1352
1353 memcpy(sregs->interrupt_bitmap, vcpu->irq_pending,
1354 sizeof sregs->interrupt_bitmap);
1355
1356 vcpu_put(vcpu);
1357
1358 return 0;
1359}
1360
1361static void set_segment(struct kvm_vcpu *vcpu,
1362 struct kvm_segment *var, int seg)
1363{
1364 return kvm_arch_ops->set_segment(vcpu, var, seg);
1365}
1366
1367static int kvm_dev_ioctl_set_sregs(struct kvm *kvm, struct kvm_sregs *sregs)
1368{
1369 struct kvm_vcpu *vcpu;
1370 int mmu_reset_needed = 0;
1371 int i;
1372 struct descriptor_table dt;
1373
1374 if (sregs->vcpu < 0 || sregs->vcpu >= KVM_MAX_VCPUS)
1375 return -EINVAL;
1376 vcpu = vcpu_load(kvm, sregs->vcpu);
1377 if (!vcpu)
1378 return -ENOENT;
1379
1380 set_segment(vcpu, &sregs->cs, VCPU_SREG_CS);
1381 set_segment(vcpu, &sregs->ds, VCPU_SREG_DS);
1382 set_segment(vcpu, &sregs->es, VCPU_SREG_ES);
1383 set_segment(vcpu, &sregs->fs, VCPU_SREG_FS);
1384 set_segment(vcpu, &sregs->gs, VCPU_SREG_GS);
1385 set_segment(vcpu, &sregs->ss, VCPU_SREG_SS);
1386
1387 set_segment(vcpu, &sregs->tr, VCPU_SREG_TR);
1388 set_segment(vcpu, &sregs->ldt, VCPU_SREG_LDTR);
1389
1390 dt.limit = sregs->idt.limit;
1391 dt.base = sregs->idt.base;
1392 kvm_arch_ops->set_idt(vcpu, &dt);
1393 dt.limit = sregs->gdt.limit;
1394 dt.base = sregs->gdt.base;
1395 kvm_arch_ops->set_gdt(vcpu, &dt);
1396
1397 vcpu->cr2 = sregs->cr2;
1398 mmu_reset_needed |= vcpu->cr3 != sregs->cr3;
1399 vcpu->cr3 = sregs->cr3;
1400
1401 vcpu->cr8 = sregs->cr8;
1402
1403 mmu_reset_needed |= vcpu->shadow_efer != sregs->efer;
1404#ifdef __x86_64__
1405 kvm_arch_ops->set_efer(vcpu, sregs->efer);
1406#endif
1407 vcpu->apic_base = sregs->apic_base;
1408
1409 mmu_reset_needed |= vcpu->cr0 != sregs->cr0;
1410 kvm_arch_ops->set_cr0_no_modeswitch(vcpu, sregs->cr0);
1411
1412 mmu_reset_needed |= vcpu->cr4 != sregs->cr4;
1413 kvm_arch_ops->set_cr4(vcpu, sregs->cr4);
1414
1415 if (mmu_reset_needed)
1416 kvm_mmu_reset_context(vcpu);
1417
1418 memcpy(vcpu->irq_pending, sregs->interrupt_bitmap,
1419 sizeof vcpu->irq_pending);
1420 vcpu->irq_summary = 0;
1421 for (i = 0; i < NR_IRQ_WORDS; ++i)
1422 if (vcpu->irq_pending[i])
1423 __set_bit(i, &vcpu->irq_summary);
1424
1425 vcpu_put(vcpu);
1426
1427 return 0;
1428}
1429
1430/*
1431 * List of msr numbers which we expose to userspace through KVM_GET_MSRS
1432 * and KVM_SET_MSRS, and KVM_GET_MSR_INDEX_LIST.
1433 */
1434static u32 msrs_to_save[] = {
1435 MSR_IA32_SYSENTER_CS, MSR_IA32_SYSENTER_ESP, MSR_IA32_SYSENTER_EIP,
1436 MSR_K6_STAR,
1437#ifdef __x86_64__
1438 MSR_CSTAR, MSR_KERNEL_GS_BASE, MSR_SYSCALL_MASK, MSR_LSTAR,
1439#endif
1440 MSR_IA32_TIME_STAMP_COUNTER,
1441};
1442
1443
1444/*
1445 * Adapt set_msr() to msr_io()'s calling convention
1446 */
1447static int do_set_msr(struct kvm_vcpu *vcpu, unsigned index, u64 *data)
1448{
1449 return set_msr(vcpu, index, *data);
1450}
1451
1452/*
1453 * Read or write a bunch of msrs. All parameters are kernel addresses.
1454 *
1455 * @return number of msrs set successfully.
1456 */
1457static int __msr_io(struct kvm *kvm, struct kvm_msrs *msrs,
1458 struct kvm_msr_entry *entries,
1459 int (*do_msr)(struct kvm_vcpu *vcpu,
1460 unsigned index, u64 *data))
1461{
1462 struct kvm_vcpu *vcpu;
1463 int i;
1464
1465 if (msrs->vcpu < 0 || msrs->vcpu >= KVM_MAX_VCPUS)
1466 return -EINVAL;
1467
1468 vcpu = vcpu_load(kvm, msrs->vcpu);
1469 if (!vcpu)
1470 return -ENOENT;
1471
1472 for (i = 0; i < msrs->nmsrs; ++i)
1473 if (do_msr(vcpu, entries[i].index, &entries[i].data))
1474 break;
1475
1476 vcpu_put(vcpu);
1477
1478 return i;
1479}
1480
1481/*
1482 * Read or write a bunch of msrs. Parameters are user addresses.
1483 *
1484 * @return number of msrs set successfully.
1485 */
1486static int msr_io(struct kvm *kvm, struct kvm_msrs __user *user_msrs,
1487 int (*do_msr)(struct kvm_vcpu *vcpu,
1488 unsigned index, u64 *data),
1489 int writeback)
1490{
1491 struct kvm_msrs msrs;
1492 struct kvm_msr_entry *entries;
1493 int r, n;
1494 unsigned size;
1495
1496 r = -EFAULT;
1497 if (copy_from_user(&msrs, user_msrs, sizeof msrs))
1498 goto out;
1499
1500 r = -E2BIG;
1501 if (msrs.nmsrs >= MAX_IO_MSRS)
1502 goto out;
1503
1504 r = -ENOMEM;
1505 size = sizeof(struct kvm_msr_entry) * msrs.nmsrs;
1506 entries = vmalloc(size);
1507 if (!entries)
1508 goto out;
1509
1510 r = -EFAULT;
1511 if (copy_from_user(entries, user_msrs->entries, size))
1512 goto out_free;
1513
1514 r = n = __msr_io(kvm, &msrs, entries, do_msr);
1515 if (r < 0)
1516 goto out_free;
1517
1518 r = -EFAULT;
1519 if (writeback && copy_to_user(user_msrs->entries, entries, size))
1520 goto out_free;
1521
1522 r = n;
1523
1524out_free:
1525 vfree(entries);
1526out:
1527 return r;
1528}
1529
1530/*
1531 * Translate a guest virtual address to a guest physical address.
1532 */
1533static int kvm_dev_ioctl_translate(struct kvm *kvm, struct kvm_translation *tr)
1534{
1535 unsigned long vaddr = tr->linear_address;
1536 struct kvm_vcpu *vcpu;
1537 gpa_t gpa;
1538
1539 vcpu = vcpu_load(kvm, tr->vcpu);
1540 if (!vcpu)
1541 return -ENOENT;
1542 spin_lock(&kvm->lock);
1543 gpa = vcpu->mmu.gva_to_gpa(vcpu, vaddr);
1544 tr->physical_address = gpa;
1545 tr->valid = gpa != UNMAPPED_GVA;
1546 tr->writeable = 1;
1547 tr->usermode = 0;
1548 spin_unlock(&kvm->lock);
1549 vcpu_put(vcpu);
1550
1551 return 0;
1552}
1553
1554static int kvm_dev_ioctl_interrupt(struct kvm *kvm, struct kvm_interrupt *irq)
1555{
1556 struct kvm_vcpu *vcpu;
1557
1558 if (irq->vcpu < 0 || irq->vcpu >= KVM_MAX_VCPUS)
1559 return -EINVAL;
1560 if (irq->irq < 0 || irq->irq >= 256)
1561 return -EINVAL;
1562 vcpu = vcpu_load(kvm, irq->vcpu);
1563 if (!vcpu)
1564 return -ENOENT;
1565
1566 set_bit(irq->irq, vcpu->irq_pending);
1567 set_bit(irq->irq / BITS_PER_LONG, &vcpu->irq_summary);
1568
1569 vcpu_put(vcpu);
1570
1571 return 0;
1572}
1573
1574static int kvm_dev_ioctl_debug_guest(struct kvm *kvm,
1575 struct kvm_debug_guest *dbg)
1576{
1577 struct kvm_vcpu *vcpu;
1578 int r;
1579
1580 if (dbg->vcpu < 0 || dbg->vcpu >= KVM_MAX_VCPUS)
1581 return -EINVAL;
1582 vcpu = vcpu_load(kvm, dbg->vcpu);
1583 if (!vcpu)
1584 return -ENOENT;
1585
1586 r = kvm_arch_ops->set_guest_debug(vcpu, dbg);
1587
1588 vcpu_put(vcpu);
1589
1590 return r;
1591}
1592
1593static long kvm_dev_ioctl(struct file *filp,
1594 unsigned int ioctl, unsigned long arg)
1595{
1596 struct kvm *kvm = filp->private_data;
1597 int r = -EINVAL;
1598
1599 switch (ioctl) {
1600 case KVM_CREATE_VCPU: {
1601 r = kvm_dev_ioctl_create_vcpu(kvm, arg);
1602 if (r)
1603 goto out;
1604 break;
1605 }
1606 case KVM_RUN: {
1607 struct kvm_run kvm_run;
1608
1609 r = -EFAULT;
1610 if (copy_from_user(&kvm_run, (void *)arg, sizeof kvm_run))
1611 goto out;
1612 r = kvm_dev_ioctl_run(kvm, &kvm_run);
1613 if (r < 0)
1614 goto out;
1615 r = -EFAULT;
1616 if (copy_to_user((void *)arg, &kvm_run, sizeof kvm_run))
1617 goto out;
1618 r = 0;
1619 break;
1620 }
1621 case KVM_GET_REGS: {
1622 struct kvm_regs kvm_regs;
1623
1624 r = -EFAULT;
1625 if (copy_from_user(&kvm_regs, (void *)arg, sizeof kvm_regs))
1626 goto out;
1627 r = kvm_dev_ioctl_get_regs(kvm, &kvm_regs);
1628 if (r)
1629 goto out;
1630 r = -EFAULT;
1631 if (copy_to_user((void *)arg, &kvm_regs, sizeof kvm_regs))
1632 goto out;
1633 r = 0;
1634 break;
1635 }
1636 case KVM_SET_REGS: {
1637 struct kvm_regs kvm_regs;
1638
1639 r = -EFAULT;
1640 if (copy_from_user(&kvm_regs, (void *)arg, sizeof kvm_regs))
1641 goto out;
1642 r = kvm_dev_ioctl_set_regs(kvm, &kvm_regs);
1643 if (r)
1644 goto out;
1645 r = 0;
1646 break;
1647 }
1648 case KVM_GET_SREGS: {
1649 struct kvm_sregs kvm_sregs;
1650
1651 r = -EFAULT;
1652 if (copy_from_user(&kvm_sregs, (void *)arg, sizeof kvm_sregs))
1653 goto out;
1654 r = kvm_dev_ioctl_get_sregs(kvm, &kvm_sregs);
1655 if (r)
1656 goto out;
1657 r = -EFAULT;
1658 if (copy_to_user((void *)arg, &kvm_sregs, sizeof kvm_sregs))
1659 goto out;
1660 r = 0;
1661 break;
1662 }
1663 case KVM_SET_SREGS: {
1664 struct kvm_sregs kvm_sregs;
1665
1666 r = -EFAULT;
1667 if (copy_from_user(&kvm_sregs, (void *)arg, sizeof kvm_sregs))
1668 goto out;
1669 r = kvm_dev_ioctl_set_sregs(kvm, &kvm_sregs);
1670 if (r)
1671 goto out;
1672 r = 0;
1673 break;
1674 }
1675 case KVM_TRANSLATE: {
1676 struct kvm_translation tr;
1677
1678 r = -EFAULT;
1679 if (copy_from_user(&tr, (void *)arg, sizeof tr))
1680 goto out;
1681 r = kvm_dev_ioctl_translate(kvm, &tr);
1682 if (r)
1683 goto out;
1684 r = -EFAULT;
1685 if (copy_to_user((void *)arg, &tr, sizeof tr))
1686 goto out;
1687 r = 0;
1688 break;
1689 }
1690 case KVM_INTERRUPT: {
1691 struct kvm_interrupt irq;
1692
1693 r = -EFAULT;
1694 if (copy_from_user(&irq, (void *)arg, sizeof irq))
1695 goto out;
1696 r = kvm_dev_ioctl_interrupt(kvm, &irq);
1697 if (r)
1698 goto out;
1699 r = 0;
1700 break;
1701 }
1702 case KVM_DEBUG_GUEST: {
1703 struct kvm_debug_guest dbg;
1704
1705 r = -EFAULT;
1706 if (copy_from_user(&dbg, (void *)arg, sizeof dbg))
1707 goto out;
1708 r = kvm_dev_ioctl_debug_guest(kvm, &dbg);
1709 if (r)
1710 goto out;
1711 r = 0;
1712 break;
1713 }
1714 case KVM_SET_MEMORY_REGION: {
1715 struct kvm_memory_region kvm_mem;
1716
1717 r = -EFAULT;
1718 if (copy_from_user(&kvm_mem, (void *)arg, sizeof kvm_mem))
1719 goto out;
1720 r = kvm_dev_ioctl_set_memory_region(kvm, &kvm_mem);
1721 if (r)
1722 goto out;
1723 break;
1724 }
1725 case KVM_GET_DIRTY_LOG: {
1726 struct kvm_dirty_log log;
1727
1728 r = -EFAULT;
1729 if (copy_from_user(&log, (void *)arg, sizeof log))
1730 goto out;
1731 r = kvm_dev_ioctl_get_dirty_log(kvm, &log);
1732 if (r)
1733 goto out;
1734 break;
1735 }
1736 case KVM_GET_MSRS:
1737 r = msr_io(kvm, (void __user *)arg, get_msr, 1);
1738 break;
1739 case KVM_SET_MSRS:
1740 r = msr_io(kvm, (void __user *)arg, do_set_msr, 0);
1741 break;
1742 case KVM_GET_MSR_INDEX_LIST: {
1743 struct kvm_msr_list __user *user_msr_list = (void __user *)arg;
1744 struct kvm_msr_list msr_list;
1745 unsigned n;
1746
1747 r = -EFAULT;
1748 if (copy_from_user(&msr_list, user_msr_list, sizeof msr_list))
1749 goto out;
1750 n = msr_list.nmsrs;
1751 msr_list.nmsrs = ARRAY_SIZE(msrs_to_save);
1752 if (copy_to_user(user_msr_list, &msr_list, sizeof msr_list))
1753 goto out;
1754 r = -E2BIG;
1755 if (n < ARRAY_SIZE(msrs_to_save))
1756 goto out;
1757 r = -EFAULT;
1758 if (copy_to_user(user_msr_list->indices, &msrs_to_save,
1759 sizeof msrs_to_save))
1760 goto out;
1761 r = 0;
1762 }
1763 default:
1764 ;
1765 }
1766out:
1767 return r;
1768}
1769
1770static struct page *kvm_dev_nopage(struct vm_area_struct *vma,
1771 unsigned long address,
1772 int *type)
1773{
1774 struct kvm *kvm = vma->vm_file->private_data;
1775 unsigned long pgoff;
1776 struct kvm_memory_slot *slot;
1777 struct page *page;
1778
1779 *type = VM_FAULT_MINOR;
1780 pgoff = ((address - vma->vm_start) >> PAGE_SHIFT) + vma->vm_pgoff;
1781 slot = gfn_to_memslot(kvm, pgoff);
1782 if (!slot)
1783 return NOPAGE_SIGBUS;
1784 page = gfn_to_page(slot, pgoff);
1785 if (!page)
1786 return NOPAGE_SIGBUS;
1787 get_page(page);
1788 return page;
1789}
1790
1791static struct vm_operations_struct kvm_dev_vm_ops = {
1792 .nopage = kvm_dev_nopage,
1793};
1794
1795static int kvm_dev_mmap(struct file *file, struct vm_area_struct *vma)
1796{
1797 vma->vm_ops = &kvm_dev_vm_ops;
1798 return 0;
1799}
1800
1801static struct file_operations kvm_chardev_ops = {
1802 .open = kvm_dev_open,
1803 .release = kvm_dev_release,
1804 .unlocked_ioctl = kvm_dev_ioctl,
1805 .compat_ioctl = kvm_dev_ioctl,
1806 .mmap = kvm_dev_mmap,
1807};
1808
1809static struct miscdevice kvm_dev = {
1810 MISC_DYNAMIC_MINOR,
1811 "kvm",
1812 &kvm_chardev_ops,
1813};
1814
1815static int kvm_reboot(struct notifier_block *notifier, unsigned long val,
1816 void *v)
1817{
1818 if (val == SYS_RESTART) {
1819 /*
1820 * Some (well, at least mine) BIOSes hang on reboot if
1821 * in vmx root mode.
1822 */
1823 printk(KERN_INFO "kvm: exiting hardware virtualization\n");
1824 on_each_cpu(kvm_arch_ops->hardware_disable, 0, 0, 1);
1825 }
1826 return NOTIFY_OK;
1827}
1828
1829static struct notifier_block kvm_reboot_notifier = {
1830 .notifier_call = kvm_reboot,
1831 .priority = 0,
1832};
1833
1834static __init void kvm_init_debug(void)
1835{
1836 struct kvm_stats_debugfs_item *p;
1837
1838 debugfs_dir = debugfs_create_dir("kvm", 0);
1839 for (p = debugfs_entries; p->name; ++p)
1840 p->dentry = debugfs_create_u32(p->name, 0444, debugfs_dir,
1841 p->data);
1842}
1843
1844static void kvm_exit_debug(void)
1845{
1846 struct kvm_stats_debugfs_item *p;
1847
1848 for (p = debugfs_entries; p->name; ++p)
1849 debugfs_remove(p->dentry);
1850 debugfs_remove(debugfs_dir);
1851}
1852
1853hpa_t bad_page_address;
1854
1855int kvm_init_arch(struct kvm_arch_ops *ops, struct module *module)
1856{
1857 int r;
1858
1859 kvm_arch_ops = ops;
1860
1861 if (!kvm_arch_ops->cpu_has_kvm_support()) {
1862 printk(KERN_ERR "kvm: no hardware support\n");
1863 return -EOPNOTSUPP;
1864 }
1865 if (kvm_arch_ops->disabled_by_bios()) {
1866 printk(KERN_ERR "kvm: disabled by bios\n");
1867 return -EOPNOTSUPP;
1868 }
1869
1870 r = kvm_arch_ops->hardware_setup();
1871 if (r < 0)
1872 return r;
1873
1874 on_each_cpu(kvm_arch_ops->hardware_enable, 0, 0, 1);
1875 register_reboot_notifier(&kvm_reboot_notifier);
1876
1877 kvm_chardev_ops.owner = module;
1878
1879 r = misc_register(&kvm_dev);
1880 if (r) {
1881 printk (KERN_ERR "kvm: misc device register failed\n");
1882 goto out_free;
1883 }
1884
1885 return r;
1886
1887out_free:
1888 unregister_reboot_notifier(&kvm_reboot_notifier);
1889 on_each_cpu(kvm_arch_ops->hardware_disable, 0, 0, 1);
1890 kvm_arch_ops->hardware_unsetup();
1891 return r;
1892}
1893
1894void kvm_exit_arch(void)
1895{
1896 misc_deregister(&kvm_dev);
1897
1898 unregister_reboot_notifier(&kvm_reboot_notifier);
1899 on_each_cpu(kvm_arch_ops->hardware_disable, 0, 0, 1);
1900 kvm_arch_ops->hardware_unsetup();
1901}
1902
1903static __init int kvm_init(void)
1904{
1905 static struct page *bad_page;
1906 int r = 0;
1907
1908 kvm_init_debug();
1909
1910 if ((bad_page = alloc_page(GFP_KERNEL)) == NULL) {
1911 r = -ENOMEM;
1912 goto out;
1913 }
1914
1915 bad_page_address = page_to_pfn(bad_page) << PAGE_SHIFT;
1916 memset(__va(bad_page_address), 0, PAGE_SIZE);
1917
1918 return r;
1919
1920out:
1921 kvm_exit_debug();
1922 return r;
1923}
1924
1925static __exit void kvm_exit(void)
1926{
1927 kvm_exit_debug();
1928 __free_page(pfn_to_page(bad_page_address >> PAGE_SHIFT));
1929}
1930
1931module_init(kvm_init)
1932module_exit(kvm_exit)
1933
1934EXPORT_SYMBOL_GPL(kvm_init_arch);
1935EXPORT_SYMBOL_GPL(kvm_exit_arch);