?  search:
summary | log | graphiclog1 | graphiclog2 | commit | commitdiff | tree | refs | edit | fork
blame | history | raw | HEAD
[PATCH] KVM: Move find_vmx_entry() to vmx.c
[linux-2.6.22.y-op.git] / drivers / kvm / kvm_main.c
blob088a445d1df7f16582bafd8cf19b8ea7df1e3092
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
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
41 MODULE_AUTHOR("Qumranet");
42 MODULE_LICENSE("GPL");
43
44 struct kvm_arch_ops *kvm_arch_ops;
45 struct kvm_stat kvm_stat;
46 EXPORT_SYMBOL_GPL(kvm_stat);
47
48 static 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
65 static 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
75 #ifdef CONFIG_X86_64
76 // LDT or TSS descriptor in the GDT. 16 bytes.
77 struct segment_descriptor_64 {
78 struct segment_descriptor s;
79 u32 base_higher;
80 u32 pad_zero;
81 };
82
83 #endif
84
85 unsigned long segment_base(u16 selector)
86 {
87 struct descriptor_table gdt;
88 struct segment_descriptor *d;
89 unsigned long table_base;
90 typedef unsigned long ul;
91 unsigned long v;
92
93 if (selector == 0)
94 return 0;
95
96 asm ("sgdt %0" : "=m"(gdt));
97 table_base = gdt.base;
98
99 if (selector & 4) { /* from ldt */
100 u16 ldt_selector;
101
102 asm ("sldt %0" : "=g"(ldt_selector));
103 table_base = segment_base(ldt_selector);
104 }
105 d = (struct segment_descriptor *)(table_base + (selector & ~7));
106 v = d->base_low | ((ul)d->base_mid << 16) | ((ul)d->base_high << 24);
107 #ifdef CONFIG_X86_64
108 if (d->system == 0
109 && (d->type == 2 || d->type == 9 || d->type == 11))
110 v |= ((ul)((struct segment_descriptor_64 *)d)->base_higher) << 32;
111 #endif
112 return v;
113 }
114 EXPORT_SYMBOL_GPL(segment_base);
115
116 int kvm_read_guest(struct kvm_vcpu *vcpu,
117 gva_t addr,
118 unsigned long size,
119 void *dest)
120 {
121 unsigned char *host_buf = dest;
122 unsigned long req_size = size;
123
124 while (size) {
125 hpa_t paddr;
126 unsigned now;
127 unsigned offset;
128 hva_t guest_buf;
129
130 paddr = gva_to_hpa(vcpu, addr);
131
132 if (is_error_hpa(paddr))
133 break;
134
135 guest_buf = (hva_t)kmap_atomic(
136 pfn_to_page(paddr >> PAGE_SHIFT),
137 KM_USER0);
138 offset = addr & ~PAGE_MASK;
139 guest_buf |= offset;
140 now = min(size, PAGE_SIZE - offset);
141 memcpy(host_buf, (void*)guest_buf, now);
142 host_buf += now;
143 addr += now;
144 size -= now;
145 kunmap_atomic((void *)(guest_buf & PAGE_MASK), KM_USER0);
146 }
147 return req_size - size;
148 }
149 EXPORT_SYMBOL_GPL(kvm_read_guest);
150
151 int kvm_write_guest(struct kvm_vcpu *vcpu,
152 gva_t addr,
153 unsigned long size,
154 void *data)
155 {
156 unsigned char *host_buf = data;
157 unsigned long req_size = size;
158
159 while (size) {
160 hpa_t paddr;
161 unsigned now;
162 unsigned offset;
163 hva_t guest_buf;
164
165 paddr = gva_to_hpa(vcpu, addr);
166
167 if (is_error_hpa(paddr))
168 break;
169
170 guest_buf = (hva_t)kmap_atomic(
171 pfn_to_page(paddr >> PAGE_SHIFT), KM_USER0);
172 offset = addr & ~PAGE_MASK;
173 guest_buf |= offset;
174 now = min(size, PAGE_SIZE - offset);
175 memcpy((void*)guest_buf, host_buf, now);
176 host_buf += now;
177 addr += now;
178 size -= now;
179 kunmap_atomic((void *)(guest_buf & PAGE_MASK), KM_USER0);
180 }
181 return req_size - size;
182 }
183 EXPORT_SYMBOL_GPL(kvm_write_guest);
184
185 static int vcpu_slot(struct kvm_vcpu *vcpu)
186 {
187 return vcpu - vcpu->kvm->vcpus;
188 }
189
190 /*
191 * Switches to specified vcpu, until a matching vcpu_put()
192 */
193 static struct kvm_vcpu *vcpu_load(struct kvm *kvm, int vcpu_slot)
194 {
195 struct kvm_vcpu *vcpu = &kvm->vcpus[vcpu_slot];
196
197 mutex_lock(&vcpu->mutex);
198 if (unlikely(!vcpu->vmcs)) {
199 mutex_unlock(&vcpu->mutex);
200 return 0;
201 }
202 return kvm_arch_ops->vcpu_load(vcpu);
203 }
204
205 static void vcpu_put(struct kvm_vcpu *vcpu)
206 {
207 kvm_arch_ops->vcpu_put(vcpu);
208 put_cpu();
209 mutex_unlock(&vcpu->mutex);
210 }
211
212 static int kvm_dev_open(struct inode *inode, struct file *filp)
213 {
214 struct kvm *kvm = kzalloc(sizeof(struct kvm), GFP_KERNEL);
215 int i;
216
217 if (!kvm)
218 return -ENOMEM;
219
220 spin_lock_init(&kvm->lock);
221 INIT_LIST_HEAD(&kvm->active_mmu_pages);
222 for (i = 0; i < KVM_MAX_VCPUS; ++i) {
223 struct kvm_vcpu *vcpu = &kvm->vcpus[i];
224
225 mutex_init(&vcpu->mutex);
226 vcpu->mmu.root_hpa = INVALID_PAGE;
227 INIT_LIST_HEAD(&vcpu->free_pages);
228 }
229 filp->private_data = kvm;
230 return 0;
231 }
232
233 /*
234 * Free any memory in @free but not in @dont.
235 */
236 static void kvm_free_physmem_slot(struct kvm_memory_slot *free,
237 struct kvm_memory_slot *dont)
238 {
239 int i;
240
241 if (!dont || free->phys_mem != dont->phys_mem)
242 if (free->phys_mem) {
243 for (i = 0; i < free->npages; ++i)
244 __free_page(free->phys_mem[i]);
245 vfree(free->phys_mem);
246 }
247
248 if (!dont || free->dirty_bitmap != dont->dirty_bitmap)
249 vfree(free->dirty_bitmap);
250
251 free->phys_mem = 0;
252 free->npages = 0;
253 free->dirty_bitmap = 0;
254 }
255
256 static void kvm_free_physmem(struct kvm *kvm)
257 {
258 int i;
259
260 for (i = 0; i < kvm->nmemslots; ++i)
261 kvm_free_physmem_slot(&kvm->memslots[i], 0);
262 }
263
264 static void kvm_free_vcpu(struct kvm_vcpu *vcpu)
265 {
266 kvm_arch_ops->vcpu_free(vcpu);
267 kvm_mmu_destroy(vcpu);
268 }
269
270 static void kvm_free_vcpus(struct kvm *kvm)
271 {
272 unsigned int i;
273
274 for (i = 0; i < KVM_MAX_VCPUS; ++i)
275 kvm_free_vcpu(&kvm->vcpus[i]);
276 }
277
278 static int kvm_dev_release(struct inode *inode, struct file *filp)
279 {
280 struct kvm *kvm = filp->private_data;
281
282 kvm_free_vcpus(kvm);
283 kvm_free_physmem(kvm);
284 kfree(kvm);
285 return 0;
286 }
287
288 static void inject_gp(struct kvm_vcpu *vcpu)
289 {
290 kvm_arch_ops->inject_gp(vcpu, 0);
291 }
292
293 static int pdptrs_have_reserved_bits_set(struct kvm_vcpu *vcpu,
294 unsigned long cr3)
295 {
296 gfn_t pdpt_gfn = cr3 >> PAGE_SHIFT;
297 unsigned offset = (cr3 & (PAGE_SIZE-1)) >> 5;
298 int i;
299 u64 pdpte;
300 u64 *pdpt;
301 struct kvm_memory_slot *memslot;
302
303 spin_lock(&vcpu->kvm->lock);
304 memslot = gfn_to_memslot(vcpu->kvm, pdpt_gfn);
305 /* FIXME: !memslot - emulate? 0xff? */
306 pdpt = kmap_atomic(gfn_to_page(memslot, pdpt_gfn), KM_USER0);
307
308 for (i = 0; i < 4; ++i) {
309 pdpte = pdpt[offset + i];
310 if ((pdpte & 1) && (pdpte & 0xfffffff0000001e6ull))
311 break;
312 }
313
314 kunmap_atomic(pdpt, KM_USER0);
315 spin_unlock(&vcpu->kvm->lock);
316
317 return i != 4;
318 }
319
320 void set_cr0(struct kvm_vcpu *vcpu, unsigned long cr0)
321 {
322 if (cr0 & CR0_RESEVED_BITS) {
323 printk(KERN_DEBUG "set_cr0: 0x%lx #GP, reserved bits 0x%lx\n",
324 cr0, vcpu->cr0);
325 inject_gp(vcpu);
326 return;
327 }
328
329 if ((cr0 & CR0_NW_MASK) && !(cr0 & CR0_CD_MASK)) {
330 printk(KERN_DEBUG "set_cr0: #GP, CD == 0 && NW == 1\n");
331 inject_gp(vcpu);
332 return;
333 }
334
335 if ((cr0 & CR0_PG_MASK) && !(cr0 & CR0_PE_MASK)) {
336 printk(KERN_DEBUG "set_cr0: #GP, set PG flag "
337 "and a clear PE flag\n");
338 inject_gp(vcpu);
339 return;
340 }
341
342 if (!is_paging(vcpu) && (cr0 & CR0_PG_MASK)) {
343 #ifdef CONFIG_X86_64
344 if ((vcpu->shadow_efer & EFER_LME)) {
345 int cs_db, cs_l;
346
347 if (!is_pae(vcpu)) {
348 printk(KERN_DEBUG "set_cr0: #GP, start paging "
349 "in long mode while PAE is disabled\n");
350 inject_gp(vcpu);
351 return;
352 }
353 kvm_arch_ops->get_cs_db_l_bits(vcpu, &cs_db, &cs_l);
354 if (cs_l) {
355 printk(KERN_DEBUG "set_cr0: #GP, start paging "
356 "in long mode while CS.L == 1\n");
357 inject_gp(vcpu);
358 return;
359
360 }
361 } else
362 #endif
363 if (is_pae(vcpu) &&
364 pdptrs_have_reserved_bits_set(vcpu, vcpu->cr3)) {
365 printk(KERN_DEBUG "set_cr0: #GP, pdptrs "
366 "reserved bits\n");
367 inject_gp(vcpu);
368 return;
369 }
370
371 }
372
373 kvm_arch_ops->set_cr0(vcpu, cr0);
374 vcpu->cr0 = cr0;
375
376 spin_lock(&vcpu->kvm->lock);
377 kvm_mmu_reset_context(vcpu);
378 spin_unlock(&vcpu->kvm->lock);
379 return;
380 }
381 EXPORT_SYMBOL_GPL(set_cr0);
382
383 void lmsw(struct kvm_vcpu *vcpu, unsigned long msw)
384 {
385 set_cr0(vcpu, (vcpu->cr0 & ~0x0ful) | (msw & 0x0f));
386 }
387 EXPORT_SYMBOL_GPL(lmsw);
388
389 void set_cr4(struct kvm_vcpu *vcpu, unsigned long cr4)
390 {
391 if (cr4 & CR4_RESEVED_BITS) {
392 printk(KERN_DEBUG "set_cr4: #GP, reserved bits\n");
393 inject_gp(vcpu);
394 return;
395 }
396
397 if (kvm_arch_ops->is_long_mode(vcpu)) {
398 if (!(cr4 & CR4_PAE_MASK)) {
399 printk(KERN_DEBUG "set_cr4: #GP, clearing PAE while "
400 "in long mode\n");
401 inject_gp(vcpu);
402 return;
403 }
404 } else if (is_paging(vcpu) && !is_pae(vcpu) && (cr4 & CR4_PAE_MASK)
405 && pdptrs_have_reserved_bits_set(vcpu, vcpu->cr3)) {
406 printk(KERN_DEBUG "set_cr4: #GP, pdptrs reserved bits\n");
407 inject_gp(vcpu);
408 }
409
410 if (cr4 & CR4_VMXE_MASK) {
411 printk(KERN_DEBUG "set_cr4: #GP, setting VMXE\n");
412 inject_gp(vcpu);
413 return;
414 }
415 kvm_arch_ops->set_cr4(vcpu, cr4);
416 spin_lock(&vcpu->kvm->lock);
417 kvm_mmu_reset_context(vcpu);
418 spin_unlock(&vcpu->kvm->lock);
419 }
420 EXPORT_SYMBOL_GPL(set_cr4);
421
422 void set_cr3(struct kvm_vcpu *vcpu, unsigned long cr3)
423 {
424 if (kvm_arch_ops->is_long_mode(vcpu)) {
425 if ( cr3 & CR3_L_MODE_RESEVED_BITS) {
426 printk(KERN_DEBUG "set_cr3: #GP, reserved bits\n");
427 inject_gp(vcpu);
428 return;
429 }
430 } else {
431 if (cr3 & CR3_RESEVED_BITS) {
432 printk(KERN_DEBUG "set_cr3: #GP, reserved bits\n");
433 inject_gp(vcpu);
434 return;
435 }
436 if (is_paging(vcpu) && is_pae(vcpu) &&
437 pdptrs_have_reserved_bits_set(vcpu, cr3)) {
438 printk(KERN_DEBUG "set_cr3: #GP, pdptrs "
439 "reserved bits\n");
440 inject_gp(vcpu);
441 return;
442 }
443 }
444
445 vcpu->cr3 = cr3;
446 spin_lock(&vcpu->kvm->lock);
447 vcpu->mmu.new_cr3(vcpu);
448 spin_unlock(&vcpu->kvm->lock);
449 }
450 EXPORT_SYMBOL_GPL(set_cr3);
451
452 void set_cr8(struct kvm_vcpu *vcpu, unsigned long cr8)
453 {
454 if ( cr8 & CR8_RESEVED_BITS) {
455 printk(KERN_DEBUG "set_cr8: #GP, reserved bits 0x%lx\n", cr8);
456 inject_gp(vcpu);
457 return;
458 }
459 vcpu->cr8 = cr8;
460 }
461 EXPORT_SYMBOL_GPL(set_cr8);
462
463 void fx_init(struct kvm_vcpu *vcpu)
464 {
465 struct __attribute__ ((__packed__)) fx_image_s {
466 u16 control; //fcw
467 u16 status; //fsw
468 u16 tag; // ftw
469 u16 opcode; //fop
470 u64 ip; // fpu ip
471 u64 operand;// fpu dp
472 u32 mxcsr;
473 u32 mxcsr_mask;
474
475 } *fx_image;
476
477 fx_save(vcpu->host_fx_image);
478 fpu_init();
479 fx_save(vcpu->guest_fx_image);
480 fx_restore(vcpu->host_fx_image);
481
482 fx_image = (struct fx_image_s *)vcpu->guest_fx_image;
483 fx_image->mxcsr = 0x1f80;
484 memset(vcpu->guest_fx_image + sizeof(struct fx_image_s),
485 0, FX_IMAGE_SIZE - sizeof(struct fx_image_s));
486 }
487 EXPORT_SYMBOL_GPL(fx_init);
488
489 /*
490 * Creates some virtual cpus. Good luck creating more than one.
491 */
492 static int kvm_dev_ioctl_create_vcpu(struct kvm *kvm, int n)
493 {
494 int r;
495 struct kvm_vcpu *vcpu;
496
497 r = -EINVAL;
498 if (n < 0 || n >= KVM_MAX_VCPUS)
499 goto out;
500
501 vcpu = &kvm->vcpus[n];
502
503 mutex_lock(&vcpu->mutex);
504
505 if (vcpu->vmcs) {
506 mutex_unlock(&vcpu->mutex);
507 return -EEXIST;
508 }
509
510 vcpu->host_fx_image = (char*)ALIGN((hva_t)vcpu->fx_buf,
511 FX_IMAGE_ALIGN);
512 vcpu->guest_fx_image = vcpu->host_fx_image + FX_IMAGE_SIZE;
513
514 vcpu->cpu = -1; /* First load will set up TR */
515 vcpu->kvm = kvm;
516 r = kvm_arch_ops->vcpu_create(vcpu);
517 if (r < 0)
518 goto out_free_vcpus;
519
520 kvm_arch_ops->vcpu_load(vcpu);
521
522 r = kvm_arch_ops->vcpu_setup(vcpu);
523 if (r >= 0)
524 r = kvm_mmu_init(vcpu);
525
526 vcpu_put(vcpu);
527
528 if (r < 0)
529 goto out_free_vcpus;
530
531 return 0;
532
533 out_free_vcpus:
534 kvm_free_vcpu(vcpu);
535 mutex_unlock(&vcpu->mutex);
536 out:
537 return r;
538 }
539
540 /*
541 * Allocate some memory and give it an address in the guest physical address
542 * space.
543 *
544 * Discontiguous memory is allowed, mostly for framebuffers.
545 */
546 static int kvm_dev_ioctl_set_memory_region(struct kvm *kvm,
547 struct kvm_memory_region *mem)
548 {
549 int r;
550 gfn_t base_gfn;
551 unsigned long npages;
552 unsigned long i;
553 struct kvm_memory_slot *memslot;
554 struct kvm_memory_slot old, new;
555 int memory_config_version;
556
557 r = -EINVAL;
558 /* General sanity checks */
559 if (mem->memory_size & (PAGE_SIZE - 1))
560 goto out;
561 if (mem->guest_phys_addr & (PAGE_SIZE - 1))
562 goto out;
563 if (mem->slot >= KVM_MEMORY_SLOTS)
564 goto out;
565 if (mem->guest_phys_addr + mem->memory_size < mem->guest_phys_addr)
566 goto out;
567
568 memslot = &kvm->memslots[mem->slot];
569 base_gfn = mem->guest_phys_addr >> PAGE_SHIFT;
570 npages = mem->memory_size >> PAGE_SHIFT;
571
572 if (!npages)
573 mem->flags &= ~KVM_MEM_LOG_DIRTY_PAGES;
574
575 raced:
576 spin_lock(&kvm->lock);
577
578 memory_config_version = kvm->memory_config_version;
579 new = old = *memslot;
580
581 new.base_gfn = base_gfn;
582 new.npages = npages;
583 new.flags = mem->flags;
584
585 /* Disallow changing a memory slot's size. */
586 r = -EINVAL;
587 if (npages && old.npages && npages != old.npages)
588 goto out_unlock;
589
590 /* Check for overlaps */
591 r = -EEXIST;
592 for (i = 0; i < KVM_MEMORY_SLOTS; ++i) {
593 struct kvm_memory_slot *s = &kvm->memslots[i];
594
595 if (s == memslot)
596 continue;
597 if (!((base_gfn + npages <= s->base_gfn) ||
598 (base_gfn >= s->base_gfn + s->npages)))
599 goto out_unlock;
600 }
601 /*
602 * Do memory allocations outside lock. memory_config_version will
603 * detect any races.
604 */
605 spin_unlock(&kvm->lock);
606
607 /* Deallocate if slot is being removed */
608 if (!npages)
609 new.phys_mem = 0;
610
611 /* Free page dirty bitmap if unneeded */
612 if (!(new.flags & KVM_MEM_LOG_DIRTY_PAGES))
613 new.dirty_bitmap = 0;
614
615 r = -ENOMEM;
616
617 /* Allocate if a slot is being created */
618 if (npages && !new.phys_mem) {
619 new.phys_mem = vmalloc(npages * sizeof(struct page *));
620
621 if (!new.phys_mem)
622 goto out_free;
623
624 memset(new.phys_mem, 0, npages * sizeof(struct page *));
625 for (i = 0; i < npages; ++i) {
626 new.phys_mem[i] = alloc_page(GFP_HIGHUSER
627 | __GFP_ZERO);
628 if (!new.phys_mem[i])
629 goto out_free;
630 }
631 }
632
633 /* Allocate page dirty bitmap if needed */
634 if ((new.flags & KVM_MEM_LOG_DIRTY_PAGES) && !new.dirty_bitmap) {
635 unsigned dirty_bytes = ALIGN(npages, BITS_PER_LONG) / 8;
636
637 new.dirty_bitmap = vmalloc(dirty_bytes);
638 if (!new.dirty_bitmap)
639 goto out_free;
640 memset(new.dirty_bitmap, 0, dirty_bytes);
641 }
642
643 spin_lock(&kvm->lock);
644
645 if (memory_config_version != kvm->memory_config_version) {
646 spin_unlock(&kvm->lock);
647 kvm_free_physmem_slot(&new, &old);
648 goto raced;
649 }
650
651 r = -EAGAIN;
652 if (kvm->busy)
653 goto out_unlock;
654
655 if (mem->slot >= kvm->nmemslots)
656 kvm->nmemslots = mem->slot + 1;
657
658 *memslot = new;
659 ++kvm->memory_config_version;
660
661 spin_unlock(&kvm->lock);
662
663 for (i = 0; i < KVM_MAX_VCPUS; ++i) {
664 struct kvm_vcpu *vcpu;
665
666 vcpu = vcpu_load(kvm, i);
667 if (!vcpu)
668 continue;
669 kvm_mmu_reset_context(vcpu);
670 vcpu_put(vcpu);
671 }
672
673 kvm_free_physmem_slot(&old, &new);
674 return 0;
675
676 out_unlock:
677 spin_unlock(&kvm->lock);
678 out_free:
679 kvm_free_physmem_slot(&new, &old);
680 out:
681 return r;
682 }
683
684 /*
685 * Get (and clear) the dirty memory log for a memory slot.
686 */
687 static int kvm_dev_ioctl_get_dirty_log(struct kvm *kvm,
688 struct kvm_dirty_log *log)
689 {
690 struct kvm_memory_slot *memslot;
691 int r, i;
692 int n;
693 unsigned long any = 0;
694
695 spin_lock(&kvm->lock);
696
697 /*
698 * Prevent changes to guest memory configuration even while the lock
699 * is not taken.
700 */
701 ++kvm->busy;
702 spin_unlock(&kvm->lock);
703 r = -EINVAL;
704 if (log->slot >= KVM_MEMORY_SLOTS)
705 goto out;
706
707 memslot = &kvm->memslots[log->slot];
708 r = -ENOENT;
709 if (!memslot->dirty_bitmap)
710 goto out;
711
712 n = ALIGN(memslot->npages, 8) / 8;
713
714 for (i = 0; !any && i < n; ++i)
715 any = memslot->dirty_bitmap[i];
716
717 r = -EFAULT;
718 if (copy_to_user(log->dirty_bitmap, memslot->dirty_bitmap, n))
719 goto out;
720
721
722 if (any) {
723 spin_lock(&kvm->lock);
724 kvm_mmu_slot_remove_write_access(kvm, log->slot);
725 spin_unlock(&kvm->lock);
726 memset(memslot->dirty_bitmap, 0, n);
727 for (i = 0; i < KVM_MAX_VCPUS; ++i) {
728 struct kvm_vcpu *vcpu = vcpu_load(kvm, i);
729
730 if (!vcpu)
731 continue;
732 kvm_arch_ops->tlb_flush(vcpu);
733 vcpu_put(vcpu);
734 }
735 }
736
737 r = 0;
738
739 out:
740 spin_lock(&kvm->lock);
741 --kvm->busy;
742 spin_unlock(&kvm->lock);
743 return r;
744 }
745
746 struct kvm_memory_slot *gfn_to_memslot(struct kvm *kvm, gfn_t gfn)
747 {
748 int i;
749
750 for (i = 0; i < kvm->nmemslots; ++i) {
751 struct kvm_memory_slot *memslot = &kvm->memslots[i];
752
753 if (gfn >= memslot->base_gfn
754 && gfn < memslot->base_gfn + memslot->npages)
755 return memslot;
756 }
757 return 0;
758 }
759 EXPORT_SYMBOL_GPL(gfn_to_memslot);
760
761 void mark_page_dirty(struct kvm *kvm, gfn_t gfn)
762 {
763 int i;
764 struct kvm_memory_slot *memslot = 0;
765 unsigned long rel_gfn;
766
767 for (i = 0; i < kvm->nmemslots; ++i) {
768 memslot = &kvm->memslots[i];
769
770 if (gfn >= memslot->base_gfn
771 && gfn < memslot->base_gfn + memslot->npages) {
772
773 if (!memslot || !memslot->dirty_bitmap)
774 return;
775
776 rel_gfn = gfn - memslot->base_gfn;
777
778 /* avoid RMW */
779 if (!test_bit(rel_gfn, memslot->dirty_bitmap))
780 set_bit(rel_gfn, memslot->dirty_bitmap);
781 return;
782 }
783 }
784 }
785
786 static int emulator_read_std(unsigned long addr,
787 unsigned long *val,
788 unsigned int bytes,
789 struct x86_emulate_ctxt *ctxt)
790 {
791 struct kvm_vcpu *vcpu = ctxt->vcpu;
792 void *data = val;
793
794 while (bytes) {
795 gpa_t gpa = vcpu->mmu.gva_to_gpa(vcpu, addr);
796 unsigned offset = addr & (PAGE_SIZE-1);
797 unsigned tocopy = min(bytes, (unsigned)PAGE_SIZE - offset);
798 unsigned long pfn;
799 struct kvm_memory_slot *memslot;
800 void *page;
801
802 if (gpa == UNMAPPED_GVA)
803 return X86EMUL_PROPAGATE_FAULT;
804 pfn = gpa >> PAGE_SHIFT;
805 memslot = gfn_to_memslot(vcpu->kvm, pfn);
806 if (!memslot)
807 return X86EMUL_UNHANDLEABLE;
808 page = kmap_atomic(gfn_to_page(memslot, pfn), KM_USER0);
809
810 memcpy(data, page + offset, tocopy);
811
812 kunmap_atomic(page, KM_USER0);
813
814 bytes -= tocopy;
815 data += tocopy;
816 addr += tocopy;
817 }
818
819 return X86EMUL_CONTINUE;
820 }
821
822 static int emulator_write_std(unsigned long addr,
823 unsigned long val,
824 unsigned int bytes,
825 struct x86_emulate_ctxt *ctxt)
826 {
827 printk(KERN_ERR "emulator_write_std: addr %lx n %d\n",
828 addr, bytes);
829 return X86EMUL_UNHANDLEABLE;
830 }
831
832 static int emulator_read_emulated(unsigned long addr,
833 unsigned long *val,
834 unsigned int bytes,
835 struct x86_emulate_ctxt *ctxt)
836 {
837 struct kvm_vcpu *vcpu = ctxt->vcpu;
838
839 if (vcpu->mmio_read_completed) {
840 memcpy(val, vcpu->mmio_data, bytes);
841 vcpu->mmio_read_completed = 0;
842 return X86EMUL_CONTINUE;
843 } else if (emulator_read_std(addr, val, bytes, ctxt)
844 == X86EMUL_CONTINUE)
845 return X86EMUL_CONTINUE;
846 else {
847 gpa_t gpa = vcpu->mmu.gva_to_gpa(vcpu, addr);
848 if (gpa == UNMAPPED_GVA)
849 return vcpu_printf(vcpu, "not present\n"), X86EMUL_PROPAGATE_FAULT;
850 vcpu->mmio_needed = 1;
851 vcpu->mmio_phys_addr = gpa;
852 vcpu->mmio_size = bytes;
853 vcpu->mmio_is_write = 0;
854
855 return X86EMUL_UNHANDLEABLE;
856 }
857 }
858
859 static int emulator_write_emulated(unsigned long addr,
860 unsigned long val,
861 unsigned int bytes,
862 struct x86_emulate_ctxt *ctxt)
863 {
864 struct kvm_vcpu *vcpu = ctxt->vcpu;
865 gpa_t gpa = vcpu->mmu.gva_to_gpa(vcpu, addr);
866
867 if (gpa == UNMAPPED_GVA)
868 return X86EMUL_PROPAGATE_FAULT;
869
870 vcpu->mmio_needed = 1;
871 vcpu->mmio_phys_addr = gpa;
872 vcpu->mmio_size = bytes;
873 vcpu->mmio_is_write = 1;
874 memcpy(vcpu->mmio_data, &val, bytes);
875
876 return X86EMUL_CONTINUE;
877 }
878
879 static int emulator_cmpxchg_emulated(unsigned long addr,
880 unsigned long old,
881 unsigned long new,
882 unsigned int bytes,
883 struct x86_emulate_ctxt *ctxt)
884 {
885 static int reported;
886
887 if (!reported) {
888 reported = 1;
889 printk(KERN_WARNING "kvm: emulating exchange as write\n");
890 }
891 return emulator_write_emulated(addr, new, bytes, ctxt);
892 }
893
894 static unsigned long get_segment_base(struct kvm_vcpu *vcpu, int seg)
895 {
896 return kvm_arch_ops->get_segment_base(vcpu, seg);
897 }
898
899 int emulate_invlpg(struct kvm_vcpu *vcpu, gva_t address)
900 {
901 spin_lock(&vcpu->kvm->lock);
902 vcpu->mmu.inval_page(vcpu, address);
903 spin_unlock(&vcpu->kvm->lock);
904 kvm_arch_ops->invlpg(vcpu, address);
905 return X86EMUL_CONTINUE;
906 }
907
908 int emulate_clts(struct kvm_vcpu *vcpu)
909 {
910 unsigned long cr0 = vcpu->cr0;
911
912 cr0 &= ~CR0_TS_MASK;
913 kvm_arch_ops->set_cr0(vcpu, cr0);
914 return X86EMUL_CONTINUE;
915 }
916
917 int emulator_get_dr(struct x86_emulate_ctxt* ctxt, int dr, unsigned long *dest)
918 {
919 struct kvm_vcpu *vcpu = ctxt->vcpu;
920
921 switch (dr) {
922 case 0 ... 3:
923 *dest = kvm_arch_ops->get_dr(vcpu, dr);
924 return X86EMUL_CONTINUE;
925 default:
926 printk(KERN_DEBUG "%s: unexpected dr %u\n",
927 __FUNCTION__, dr);
928 return X86EMUL_UNHANDLEABLE;
929 }
930 }
931
932 int emulator_set_dr(struct x86_emulate_ctxt *ctxt, int dr, unsigned long value)
933 {
934 unsigned long mask = (ctxt->mode == X86EMUL_MODE_PROT64) ? ~0ULL : ~0U;
935 int exception;
936
937 kvm_arch_ops->set_dr(ctxt->vcpu, dr, value & mask, &exception);
938 if (exception) {
939 /* FIXME: better handling */
940 return X86EMUL_UNHANDLEABLE;
941 }
942 return X86EMUL_CONTINUE;
943 }
944
945 static void report_emulation_failure(struct x86_emulate_ctxt *ctxt)
946 {
947 static int reported;
948 u8 opcodes[4];
949 unsigned long rip = ctxt->vcpu->rip;
950 unsigned long rip_linear;
951
952 rip_linear = rip + get_segment_base(ctxt->vcpu, VCPU_SREG_CS);
953
954 if (reported)
955 return;
956
957 emulator_read_std(rip_linear, (void *)opcodes, 4, ctxt);
958
959 printk(KERN_ERR "emulation failed but !mmio_needed?"
960 " rip %lx %02x %02x %02x %02x\n",
961 rip, opcodes[0], opcodes[1], opcodes[2], opcodes[3]);
962 reported = 1;
963 }
964
965 struct x86_emulate_ops emulate_ops = {
966 .read_std = emulator_read_std,
967 .write_std = emulator_write_std,
968 .read_emulated = emulator_read_emulated,
969 .write_emulated = emulator_write_emulated,
970 .cmpxchg_emulated = emulator_cmpxchg_emulated,
971 };
972
973 int emulate_instruction(struct kvm_vcpu *vcpu,
974 struct kvm_run *run,
975 unsigned long cr2,
976 u16 error_code)
977 {
978 struct x86_emulate_ctxt emulate_ctxt;
979 int r;
980 int cs_db, cs_l;
981
982 kvm_arch_ops->cache_regs(vcpu);
983
984 kvm_arch_ops->get_cs_db_l_bits(vcpu, &cs_db, &cs_l);
985
986 emulate_ctxt.vcpu = vcpu;
987 emulate_ctxt.eflags = kvm_arch_ops->get_rflags(vcpu);
988 emulate_ctxt.cr2 = cr2;
989 emulate_ctxt.mode = (emulate_ctxt.eflags & X86_EFLAGS_VM)
990 ? X86EMUL_MODE_REAL : cs_l
991 ? X86EMUL_MODE_PROT64 : cs_db
992 ? X86EMUL_MODE_PROT32 : X86EMUL_MODE_PROT16;
993
994 if (emulate_ctxt.mode == X86EMUL_MODE_PROT64) {
995 emulate_ctxt.cs_base = 0;
996 emulate_ctxt.ds_base = 0;
997 emulate_ctxt.es_base = 0;
998 emulate_ctxt.ss_base = 0;
999 } else {
1000 emulate_ctxt.cs_base = get_segment_base(vcpu, VCPU_SREG_CS);
1001 emulate_ctxt.ds_base = get_segment_base(vcpu, VCPU_SREG_DS);
1002 emulate_ctxt.es_base = get_segment_base(vcpu, VCPU_SREG_ES);
1003 emulate_ctxt.ss_base = get_segment_base(vcpu, VCPU_SREG_SS);
1004 }
1005
1006 emulate_ctxt.gs_base = get_segment_base(vcpu, VCPU_SREG_GS);
1007 emulate_ctxt.fs_base = get_segment_base(vcpu, VCPU_SREG_FS);
1008
1009 vcpu->mmio_is_write = 0;
1010 r = x86_emulate_memop(&emulate_ctxt, &emulate_ops);
1011
1012 if ((r || vcpu->mmio_is_write) && run) {
1013 run->mmio.phys_addr = vcpu->mmio_phys_addr;
1014 memcpy(run->mmio.data, vcpu->mmio_data, 8);
1015 run->mmio.len = vcpu->mmio_size;
1016 run->mmio.is_write = vcpu->mmio_is_write;
1017 }
1018
1019 if (r) {
1020 if (!vcpu->mmio_needed) {
1021 report_emulation_failure(&emulate_ctxt);
1022 return EMULATE_FAIL;
1023 }
1024 return EMULATE_DO_MMIO;
1025 }
1026
1027 kvm_arch_ops->decache_regs(vcpu);
1028 kvm_arch_ops->set_rflags(vcpu, emulate_ctxt.eflags);
1029
1030 if (vcpu->mmio_is_write)
1031 return EMULATE_DO_MMIO;
1032
1033 return EMULATE_DONE;
1034 }
1035 EXPORT_SYMBOL_GPL(emulate_instruction);
1036
1037 static u64 mk_cr_64(u64 curr_cr, u32 new_val)
1038 {
1039 return (curr_cr & ~((1ULL << 32) - 1)) | new_val;
1040 }
1041
1042 void realmode_lgdt(struct kvm_vcpu *vcpu, u16 limit, unsigned long base)
1043 {
1044 struct descriptor_table dt = { limit, base };
1045
1046 kvm_arch_ops->set_gdt(vcpu, &dt);
1047 }
1048
1049 void realmode_lidt(struct kvm_vcpu *vcpu, u16 limit, unsigned long base)
1050 {
1051 struct descriptor_table dt = { limit, base };
1052
1053 kvm_arch_ops->set_idt(vcpu, &dt);
1054 }
1055
1056 void realmode_lmsw(struct kvm_vcpu *vcpu, unsigned long msw,
1057 unsigned long *rflags)
1058 {
1059 lmsw(vcpu, msw);
1060 *rflags = kvm_arch_ops->get_rflags(vcpu);
1061 }
1062
1063 unsigned long realmode_get_cr(struct kvm_vcpu *vcpu, int cr)
1064 {
1065 switch (cr) {
1066 case 0:
1067 return vcpu->cr0;
1068 case 2:
1069 return vcpu->cr2;
1070 case 3:
1071 return vcpu->cr3;
1072 case 4:
1073 return vcpu->cr4;
1074 default:
1075 vcpu_printf(vcpu, "%s: unexpected cr %u\n", __FUNCTION__, cr);
1076 return 0;
1077 }
1078 }
1079
1080 void realmode_set_cr(struct kvm_vcpu *vcpu, int cr, unsigned long val,
1081 unsigned long *rflags)
1082 {
1083 switch (cr) {
1084 case 0:
1085 set_cr0(vcpu, mk_cr_64(vcpu->cr0, val));
1086 *rflags = kvm_arch_ops->get_rflags(vcpu);
1087 break;
1088 case 2:
1089 vcpu->cr2 = val;
1090 break;
1091 case 3:
1092 set_cr3(vcpu, val);
1093 break;
1094 case 4:
1095 set_cr4(vcpu, mk_cr_64(vcpu->cr4, val));
1096 break;
1097 default:
1098 vcpu_printf(vcpu, "%s: unexpected cr %u\n", __FUNCTION__, cr);
1099 }
1100 }
1101
1102 /*
1103 * Reads an msr value (of 'msr_index') into 'pdata'.
1104 * Returns 0 on success, non-0 otherwise.
1105 * Assumes vcpu_load() was already called.
1106 */
1107 static int get_msr(struct kvm_vcpu *vcpu, u32 msr_index, u64 *pdata)
1108 {
1109 return kvm_arch_ops->get_msr(vcpu, msr_index, pdata);
1110 }
1111
1112 #ifdef CONFIG_X86_64
1113
1114 void set_efer(struct kvm_vcpu *vcpu, u64 efer)
1115 {
1116 if (efer & EFER_RESERVED_BITS) {
1117 printk(KERN_DEBUG "set_efer: 0x%llx #GP, reserved bits\n",
1118 efer);
1119 inject_gp(vcpu);
1120 return;
1121 }
1122
1123 if (is_paging(vcpu)
1124 && (vcpu->shadow_efer & EFER_LME) != (efer & EFER_LME)) {
1125 printk(KERN_DEBUG "set_efer: #GP, change LME while paging\n");
1126 inject_gp(vcpu);
1127 return;
1128 }
1129
1130 kvm_arch_ops->set_efer(vcpu, efer);
1131
1132 efer &= ~EFER_LMA;
1133 efer |= vcpu->shadow_efer & EFER_LMA;
1134
1135 vcpu->shadow_efer = efer;
1136 }
1137 EXPORT_SYMBOL_GPL(set_efer);
1138
1139 #endif
1140
1141 /*
1142 * Writes msr value into into the appropriate "register".
1143 * Returns 0 on success, non-0 otherwise.
1144 * Assumes vcpu_load() was already called.
1145 */
1146 static int set_msr(struct kvm_vcpu *vcpu, u32 msr_index, u64 data)
1147 {
1148 return kvm_arch_ops->set_msr(vcpu, msr_index, data);
1149 }
1150
1151 void kvm_resched(struct kvm_vcpu *vcpu)
1152 {
1153 vcpu_put(vcpu);
1154 cond_resched();
1155 /* Cannot fail - no vcpu unplug yet. */
1156 vcpu_load(vcpu->kvm, vcpu_slot(vcpu));
1157 }
1158 EXPORT_SYMBOL_GPL(kvm_resched);
1159
1160 void load_msrs(struct vmx_msr_entry *e, int n)
1161 {
1162 int i;
1163
1164 for (i = 0; i < n; ++i)
1165 wrmsrl(e[i].index, e[i].data);
1166 }
1167 EXPORT_SYMBOL_GPL(load_msrs);
1168
1169 void save_msrs(struct vmx_msr_entry *e, int n)
1170 {
1171 int i;
1172
1173 for (i = 0; i < n; ++i)
1174 rdmsrl(e[i].index, e[i].data);
1175 }
1176 EXPORT_SYMBOL_GPL(save_msrs);
1177
1178 static int kvm_dev_ioctl_run(struct kvm *kvm, struct kvm_run *kvm_run)
1179 {
1180 struct kvm_vcpu *vcpu;
1181 int r;
1182
1183 if (kvm_run->vcpu < 0 || kvm_run->vcpu >= KVM_MAX_VCPUS)
1184 return -EINVAL;
1185
1186 vcpu = vcpu_load(kvm, kvm_run->vcpu);
1187 if (!vcpu)
1188 return -ENOENT;
1189
1190 if (kvm_run->emulated) {
1191 kvm_arch_ops->skip_emulated_instruction(vcpu);
1192 kvm_run->emulated = 0;
1193 }
1194
1195 if (kvm_run->mmio_completed) {
1196 memcpy(vcpu->mmio_data, kvm_run->mmio.data, 8);
1197 vcpu->mmio_read_completed = 1;
1198 }
1199
1200 vcpu->mmio_needed = 0;
1201
1202 r = kvm_arch_ops->run(vcpu, kvm_run);
1203
1204 vcpu_put(vcpu);
1205 return r;
1206 }
1207
1208 static int kvm_dev_ioctl_get_regs(struct kvm *kvm, struct kvm_regs *regs)
1209 {
1210 struct kvm_vcpu *vcpu;
1211
1212 if (regs->vcpu < 0 || regs->vcpu >= KVM_MAX_VCPUS)
1213 return -EINVAL;
1214
1215 vcpu = vcpu_load(kvm, regs->vcpu);
1216 if (!vcpu)
1217 return -ENOENT;
1218
1219 kvm_arch_ops->cache_regs(vcpu);
1220
1221 regs->rax = vcpu->regs[VCPU_REGS_RAX];
1222 regs->rbx = vcpu->regs[VCPU_REGS_RBX];
1223 regs->rcx = vcpu->regs[VCPU_REGS_RCX];
1224 regs->rdx = vcpu->regs[VCPU_REGS_RDX];
1225 regs->rsi = vcpu->regs[VCPU_REGS_RSI];
1226 regs->rdi = vcpu->regs[VCPU_REGS_RDI];
1227 regs->rsp = vcpu->regs[VCPU_REGS_RSP];
1228 regs->rbp = vcpu->regs[VCPU_REGS_RBP];
1229 #ifdef CONFIG_X86_64
1230 regs->r8 = vcpu->regs[VCPU_REGS_R8];
1231 regs->r9 = vcpu->regs[VCPU_REGS_R9];
1232 regs->r10 = vcpu->regs[VCPU_REGS_R10];
1233 regs->r11 = vcpu->regs[VCPU_REGS_R11];
1234 regs->r12 = vcpu->regs[VCPU_REGS_R12];
1235 regs->r13 = vcpu->regs[VCPU_REGS_R13];
1236 regs->r14 = vcpu->regs[VCPU_REGS_R14];
1237 regs->r15 = vcpu->regs[VCPU_REGS_R15];
1238 #endif
1239
1240 regs->rip = vcpu->rip;
1241 regs->rflags = kvm_arch_ops->get_rflags(vcpu);
1242
1243 /*
1244 * Don't leak debug flags in case they were set for guest debugging
1245 */
1246 if (vcpu->guest_debug.enabled && vcpu->guest_debug.singlestep)
1247 regs->rflags &= ~(X86_EFLAGS_TF | X86_EFLAGS_RF);
1248
1249 vcpu_put(vcpu);
1250
1251 return 0;
1252 }
1253
1254 static int kvm_dev_ioctl_set_regs(struct kvm *kvm, struct kvm_regs *regs)
1255 {
1256 struct kvm_vcpu *vcpu;
1257
1258 if (regs->vcpu < 0 || regs->vcpu >= KVM_MAX_VCPUS)
1259 return -EINVAL;
1260
1261 vcpu = vcpu_load(kvm, regs->vcpu);
1262 if (!vcpu)
1263 return -ENOENT;
1264
1265 vcpu->regs[VCPU_REGS_RAX] = regs->rax;
1266 vcpu->regs[VCPU_REGS_RBX] = regs->rbx;
1267 vcpu->regs[VCPU_REGS_RCX] = regs->rcx;
1268 vcpu->regs[VCPU_REGS_RDX] = regs->rdx;
1269 vcpu->regs[VCPU_REGS_RSI] = regs->rsi;
1270 vcpu->regs[VCPU_REGS_RDI] = regs->rdi;
1271 vcpu->regs[VCPU_REGS_RSP] = regs->rsp;
1272 vcpu->regs[VCPU_REGS_RBP] = regs->rbp;
1273 #ifdef CONFIG_X86_64
1274 vcpu->regs[VCPU_REGS_R8] = regs->r8;
1275 vcpu->regs[VCPU_REGS_R9] = regs->r9;
1276 vcpu->regs[VCPU_REGS_R10] = regs->r10;
1277 vcpu->regs[VCPU_REGS_R11] = regs->r11;
1278 vcpu->regs[VCPU_REGS_R12] = regs->r12;
1279 vcpu->regs[VCPU_REGS_R13] = regs->r13;
1280 vcpu->regs[VCPU_REGS_R14] = regs->r14;
1281 vcpu->regs[VCPU_REGS_R15] = regs->r15;
1282 #endif
1283
1284 vcpu->rip = regs->rip;
1285 kvm_arch_ops->set_rflags(vcpu, regs->rflags);
1286
1287 kvm_arch_ops->decache_regs(vcpu);
1288
1289 vcpu_put(vcpu);
1290
1291 return 0;
1292 }
1293
1294 static void get_segment(struct kvm_vcpu *vcpu,
1295 struct kvm_segment *var, int seg)
1296 {
1297 return kvm_arch_ops->get_segment(vcpu, var, seg);
1298 }
1299
1300 static int kvm_dev_ioctl_get_sregs(struct kvm *kvm, struct kvm_sregs *sregs)
1301 {
1302 struct kvm_vcpu *vcpu;
1303 struct descriptor_table dt;
1304
1305 if (sregs->vcpu < 0 || sregs->vcpu >= KVM_MAX_VCPUS)
1306 return -EINVAL;
1307 vcpu = vcpu_load(kvm, sregs->vcpu);
1308 if (!vcpu)
1309 return -ENOENT;
1310
1311 get_segment(vcpu, &sregs->cs, VCPU_SREG_CS);
1312 get_segment(vcpu, &sregs->ds, VCPU_SREG_DS);
1313 get_segment(vcpu, &sregs->es, VCPU_SREG_ES);
1314 get_segment(vcpu, &sregs->fs, VCPU_SREG_FS);
1315 get_segment(vcpu, &sregs->gs, VCPU_SREG_GS);
1316 get_segment(vcpu, &sregs->ss, VCPU_SREG_SS);
1317
1318 get_segment(vcpu, &sregs->tr, VCPU_SREG_TR);
1319 get_segment(vcpu, &sregs->ldt, VCPU_SREG_LDTR);
1320
1321 kvm_arch_ops->get_idt(vcpu, &dt);
1322 sregs->idt.limit = dt.limit;
1323 sregs->idt.base = dt.base;
1324 kvm_arch_ops->get_gdt(vcpu, &dt);
1325 sregs->gdt.limit = dt.limit;
1326 sregs->gdt.base = dt.base;
1327
1328 sregs->cr0 = vcpu->cr0;
1329 sregs->cr2 = vcpu->cr2;
1330 sregs->cr3 = vcpu->cr3;
1331 sregs->cr4 = vcpu->cr4;
1332 sregs->cr8 = vcpu->cr8;
1333 sregs->efer = vcpu->shadow_efer;
1334 sregs->apic_base = vcpu->apic_base;
1335
1336 memcpy(sregs->interrupt_bitmap, vcpu->irq_pending,
1337 sizeof sregs->interrupt_bitmap);
1338
1339 vcpu_put(vcpu);
1340
1341 return 0;
1342 }
1343
1344 static void set_segment(struct kvm_vcpu *vcpu,
1345 struct kvm_segment *var, int seg)
1346 {
1347 return kvm_arch_ops->set_segment(vcpu, var, seg);
1348 }
1349
1350 static int kvm_dev_ioctl_set_sregs(struct kvm *kvm, struct kvm_sregs *sregs)
1351 {
1352 struct kvm_vcpu *vcpu;
1353 int mmu_reset_needed = 0;
1354 int i;
1355 struct descriptor_table dt;
1356
1357 if (sregs->vcpu < 0 || sregs->vcpu >= KVM_MAX_VCPUS)
1358 return -EINVAL;
1359 vcpu = vcpu_load(kvm, sregs->vcpu);
1360 if (!vcpu)
1361 return -ENOENT;
1362
1363 set_segment(vcpu, &sregs->cs, VCPU_SREG_CS);
1364 set_segment(vcpu, &sregs->ds, VCPU_SREG_DS);
1365 set_segment(vcpu, &sregs->es, VCPU_SREG_ES);
1366 set_segment(vcpu, &sregs->fs, VCPU_SREG_FS);
1367 set_segment(vcpu, &sregs->gs, VCPU_SREG_GS);
1368 set_segment(vcpu, &sregs->ss, VCPU_SREG_SS);
1369
1370 set_segment(vcpu, &sregs->tr, VCPU_SREG_TR);
1371 set_segment(vcpu, &sregs->ldt, VCPU_SREG_LDTR);
1372
1373 dt.limit = sregs->idt.limit;
1374 dt.base = sregs->idt.base;
1375 kvm_arch_ops->set_idt(vcpu, &dt);
1376 dt.limit = sregs->gdt.limit;
1377 dt.base = sregs->gdt.base;
1378 kvm_arch_ops->set_gdt(vcpu, &dt);
1379
1380 vcpu->cr2 = sregs->cr2;
1381 mmu_reset_needed |= vcpu->cr3 != sregs->cr3;
1382 vcpu->cr3 = sregs->cr3;
1383
1384 vcpu->cr8 = sregs->cr8;
1385
1386 mmu_reset_needed |= vcpu->shadow_efer != sregs->efer;
1387 #ifdef CONFIG_X86_64
1388 kvm_arch_ops->set_efer(vcpu, sregs->efer);
1389 #endif
1390 vcpu->apic_base = sregs->apic_base;
1391
1392 mmu_reset_needed |= vcpu->cr0 != sregs->cr0;
1393 kvm_arch_ops->set_cr0_no_modeswitch(vcpu, sregs->cr0);
1394
1395 mmu_reset_needed |= vcpu->cr4 != sregs->cr4;
1396 kvm_arch_ops->set_cr4(vcpu, sregs->cr4);
1397
1398 if (mmu_reset_needed)
1399 kvm_mmu_reset_context(vcpu);
1400
1401 memcpy(vcpu->irq_pending, sregs->interrupt_bitmap,
1402 sizeof vcpu->irq_pending);
1403 vcpu->irq_summary = 0;
1404 for (i = 0; i < NR_IRQ_WORDS; ++i)
1405 if (vcpu->irq_pending[i])
1406 __set_bit(i, &vcpu->irq_summary);
1407
1408 vcpu_put(vcpu);
1409
1410 return 0;
1411 }
1412
1413 /*
1414 * List of msr numbers which we expose to userspace through KVM_GET_MSRS
1415 * and KVM_SET_MSRS, and KVM_GET_MSR_INDEX_LIST.
1416 */
1417 static u32 msrs_to_save[] = {
1418 MSR_IA32_SYSENTER_CS, MSR_IA32_SYSENTER_ESP, MSR_IA32_SYSENTER_EIP,
1419 MSR_K6_STAR,
1420 #ifdef CONFIG_X86_64
1421 MSR_CSTAR, MSR_KERNEL_GS_BASE, MSR_SYSCALL_MASK, MSR_LSTAR,
1422 #endif
1423 MSR_IA32_TIME_STAMP_COUNTER,
1424 };
1425
1426
1427 /*
1428 * Adapt set_msr() to msr_io()'s calling convention
1429 */
1430 static int do_set_msr(struct kvm_vcpu *vcpu, unsigned index, u64 *data)
1431 {
1432 return set_msr(vcpu, index, *data);
1433 }
1434
1435 /*
1436 * Read or write a bunch of msrs. All parameters are kernel addresses.
1437 *
1438 * @return number of msrs set successfully.
1439 */
1440 static int __msr_io(struct kvm *kvm, struct kvm_msrs *msrs,
1441 struct kvm_msr_entry *entries,
1442 int (*do_msr)(struct kvm_vcpu *vcpu,
1443 unsigned index, u64 *data))
1444 {
1445 struct kvm_vcpu *vcpu;
1446 int i;
1447
1448 if (msrs->vcpu < 0 || msrs->vcpu >= KVM_MAX_VCPUS)
1449 return -EINVAL;
1450
1451 vcpu = vcpu_load(kvm, msrs->vcpu);
1452 if (!vcpu)
1453 return -ENOENT;
1454
1455 for (i = 0; i < msrs->nmsrs; ++i)
1456 if (do_msr(vcpu, entries[i].index, &entries[i].data))
1457 break;
1458
1459 vcpu_put(vcpu);
1460
1461 return i;
1462 }
1463
1464 /*
1465 * Read or write a bunch of msrs. Parameters are user addresses.
1466 *
1467 * @return number of msrs set successfully.
1468 */
1469 static int msr_io(struct kvm *kvm, struct kvm_msrs __user *user_msrs,
1470 int (*do_msr)(struct kvm_vcpu *vcpu,
1471 unsigned index, u64 *data),
1472 int writeback)
1473 {
1474 struct kvm_msrs msrs;
1475 struct kvm_msr_entry *entries;
1476 int r, n;
1477 unsigned size;
1478
1479 r = -EFAULT;
1480 if (copy_from_user(&msrs, user_msrs, sizeof msrs))
1481 goto out;
1482
1483 r = -E2BIG;
1484 if (msrs.nmsrs >= MAX_IO_MSRS)
1485 goto out;
1486
1487 r = -ENOMEM;
1488 size = sizeof(struct kvm_msr_entry) * msrs.nmsrs;
1489 entries = vmalloc(size);
1490 if (!entries)
1491 goto out;
1492
1493 r = -EFAULT;
1494 if (copy_from_user(entries, user_msrs->entries, size))
1495 goto out_free;
1496
1497 r = n = __msr_io(kvm, &msrs, entries, do_msr);
1498 if (r < 0)
1499 goto out_free;
1500
1501 r = -EFAULT;
1502 if (writeback && copy_to_user(user_msrs->entries, entries, size))
1503 goto out_free;
1504
1505 r = n;
1506
1507 out_free:
1508 vfree(entries);
1509 out:
1510 return r;
1511 }
1512
1513 /*
1514 * Translate a guest virtual address to a guest physical address.
1515 */
1516 static int kvm_dev_ioctl_translate(struct kvm *kvm, struct kvm_translation *tr)
1517 {
1518 unsigned long vaddr = tr->linear_address;
1519 struct kvm_vcpu *vcpu;
1520 gpa_t gpa;
1521
1522 vcpu = vcpu_load(kvm, tr->vcpu);
1523 if (!vcpu)
1524 return -ENOENT;
1525 spin_lock(&kvm->lock);
1526 gpa = vcpu->mmu.gva_to_gpa(vcpu, vaddr);
1527 tr->physical_address = gpa;
1528 tr->valid = gpa != UNMAPPED_GVA;
1529 tr->writeable = 1;
1530 tr->usermode = 0;
1531 spin_unlock(&kvm->lock);
1532 vcpu_put(vcpu);
1533
1534 return 0;
1535 }
1536
1537 static int kvm_dev_ioctl_interrupt(struct kvm *kvm, struct kvm_interrupt *irq)
1538 {
1539 struct kvm_vcpu *vcpu;
1540
1541 if (irq->vcpu < 0 || irq->vcpu >= KVM_MAX_VCPUS)
1542 return -EINVAL;
1543 if (irq->irq < 0 || irq->irq >= 256)
1544 return -EINVAL;
1545 vcpu = vcpu_load(kvm, irq->vcpu);
1546 if (!vcpu)
1547 return -ENOENT;
1548
1549 set_bit(irq->irq, vcpu->irq_pending);
1550 set_bit(irq->irq / BITS_PER_LONG, &vcpu->irq_summary);
1551
1552 vcpu_put(vcpu);
1553
1554 return 0;
1555 }
1556
1557 static int kvm_dev_ioctl_debug_guest(struct kvm *kvm,
1558 struct kvm_debug_guest *dbg)
1559 {
1560 struct kvm_vcpu *vcpu;
1561 int r;
1562
1563 if (dbg->vcpu < 0 || dbg->vcpu >= KVM_MAX_VCPUS)
1564 return -EINVAL;
1565 vcpu = vcpu_load(kvm, dbg->vcpu);
1566 if (!vcpu)
1567 return -ENOENT;
1568
1569 r = kvm_arch_ops->set_guest_debug(vcpu, dbg);
1570
1571 vcpu_put(vcpu);
1572
1573 return r;
1574 }
1575
1576 static long kvm_dev_ioctl(struct file *filp,
1577 unsigned int ioctl, unsigned long arg)
1578 {
1579 struct kvm *kvm = filp->private_data;
1580 int r = -EINVAL;
1581
1582 switch (ioctl) {
1583 case KVM_CREATE_VCPU: {
1584 r = kvm_dev_ioctl_create_vcpu(kvm, arg);
1585 if (r)
1586 goto out;
1587 break;
1588 }
1589 case KVM_RUN: {
1590 struct kvm_run kvm_run;
1591
1592 r = -EFAULT;
1593 if (copy_from_user(&kvm_run, (void *)arg, sizeof kvm_run))
1594 goto out;
1595 r = kvm_dev_ioctl_run(kvm, &kvm_run);
1596 if (r < 0)
1597 goto out;
1598 r = -EFAULT;
1599 if (copy_to_user((void *)arg, &kvm_run, sizeof kvm_run))
1600 goto out;
1601 r = 0;
1602 break;
1603 }
1604 case KVM_GET_REGS: {
1605 struct kvm_regs kvm_regs;
1606
1607 r = -EFAULT;
1608 if (copy_from_user(&kvm_regs, (void *)arg, sizeof kvm_regs))
1609 goto out;
1610 r = kvm_dev_ioctl_get_regs(kvm, &kvm_regs);
1611 if (r)
1612 goto out;
1613 r = -EFAULT;
1614 if (copy_to_user((void *)arg, &kvm_regs, sizeof kvm_regs))
1615 goto out;
1616 r = 0;
1617 break;
1618 }
1619 case KVM_SET_REGS: {
1620 struct kvm_regs kvm_regs;
1621
1622 r = -EFAULT;
1623 if (copy_from_user(&kvm_regs, (void *)arg, sizeof kvm_regs))
1624 goto out;
1625 r = kvm_dev_ioctl_set_regs(kvm, &kvm_regs);
1626 if (r)
1627 goto out;
1628 r = 0;
1629 break;
1630 }
1631 case KVM_GET_SREGS: {
1632 struct kvm_sregs kvm_sregs;
1633
1634 r = -EFAULT;
1635 if (copy_from_user(&kvm_sregs, (void *)arg, sizeof kvm_sregs))
1636 goto out;
1637 r = kvm_dev_ioctl_get_sregs(kvm, &kvm_sregs);
1638 if (r)
1639 goto out;
1640 r = -EFAULT;
1641 if (copy_to_user((void *)arg, &kvm_sregs, sizeof kvm_sregs))
1642 goto out;
1643 r = 0;
1644 break;
1645 }
1646 case KVM_SET_SREGS: {
1647 struct kvm_sregs kvm_sregs;
1648
1649 r = -EFAULT;
1650 if (copy_from_user(&kvm_sregs, (void *)arg, sizeof kvm_sregs))
1651 goto out;
1652 r = kvm_dev_ioctl_set_sregs(kvm, &kvm_sregs);
1653 if (r)
1654 goto out;
1655 r = 0;
1656 break;
1657 }
1658 case KVM_TRANSLATE: {
1659 struct kvm_translation tr;
1660
1661 r = -EFAULT;
1662 if (copy_from_user(&tr, (void *)arg, sizeof tr))
1663 goto out;
1664 r = kvm_dev_ioctl_translate(kvm, &tr);
1665 if (r)
1666 goto out;
1667 r = -EFAULT;
1668 if (copy_to_user((void *)arg, &tr, sizeof tr))
1669 goto out;
1670 r = 0;
1671 break;
1672 }
1673 case KVM_INTERRUPT: {
1674 struct kvm_interrupt irq;
1675
1676 r = -EFAULT;
1677 if (copy_from_user(&irq, (void *)arg, sizeof irq))
1678 goto out;
1679 r = kvm_dev_ioctl_interrupt(kvm, &irq);
1680 if (r)
1681 goto out;
1682 r = 0;
1683 break;
1684 }
1685 case KVM_DEBUG_GUEST: {
1686 struct kvm_debug_guest dbg;
1687
1688 r = -EFAULT;
1689 if (copy_from_user(&dbg, (void *)arg, sizeof dbg))
1690 goto out;
1691 r = kvm_dev_ioctl_debug_guest(kvm, &dbg);
1692 if (r)
1693 goto out;
1694 r = 0;
1695 break;
1696 }
1697 case KVM_SET_MEMORY_REGION: {
1698 struct kvm_memory_region kvm_mem;
1699
1700 r = -EFAULT;
1701 if (copy_from_user(&kvm_mem, (void *)arg, sizeof kvm_mem))
1702 goto out;
1703 r = kvm_dev_ioctl_set_memory_region(kvm, &kvm_mem);
1704 if (r)
1705 goto out;
1706 break;
1707 }
1708 case KVM_GET_DIRTY_LOG: {
1709 struct kvm_dirty_log log;
1710
1711 r = -EFAULT;
1712 if (copy_from_user(&log, (void *)arg, sizeof log))
1713 goto out;
1714 r = kvm_dev_ioctl_get_dirty_log(kvm, &log);
1715 if (r)
1716 goto out;
1717 break;
1718 }
1719 case KVM_GET_MSRS:
1720 r = msr_io(kvm, (void __user *)arg, get_msr, 1);
1721 break;
1722 case KVM_SET_MSRS:
1723 r = msr_io(kvm, (void __user *)arg, do_set_msr, 0);
1724 break;
1725 case KVM_GET_MSR_INDEX_LIST: {
1726 struct kvm_msr_list __user *user_msr_list = (void __user *)arg;
1727 struct kvm_msr_list msr_list;
1728 unsigned n;
1729
1730 r = -EFAULT;
1731 if (copy_from_user(&msr_list, user_msr_list, sizeof msr_list))
1732 goto out;
1733 n = msr_list.nmsrs;
1734 msr_list.nmsrs = ARRAY_SIZE(msrs_to_save);
1735 if (copy_to_user(user_msr_list, &msr_list, sizeof msr_list))
1736 goto out;
1737 r = -E2BIG;
1738 if (n < ARRAY_SIZE(msrs_to_save))
1739 goto out;
1740 r = -EFAULT;
1741 if (copy_to_user(user_msr_list->indices, &msrs_to_save,
1742 sizeof msrs_to_save))
1743 goto out;
1744 r = 0;
1745 }
1746 default:
1747 ;
1748 }
1749 out:
1750 return r;
1751 }
1752
1753 static struct page *kvm_dev_nopage(struct vm_area_struct *vma,
1754 unsigned long address,
1755 int *type)
1756 {
1757 struct kvm *kvm = vma->vm_file->private_data;
1758 unsigned long pgoff;
1759 struct kvm_memory_slot *slot;
1760 struct page *page;
1761
1762 *type = VM_FAULT_MINOR;
1763 pgoff = ((address - vma->vm_start) >> PAGE_SHIFT) + vma->vm_pgoff;
1764 slot = gfn_to_memslot(kvm, pgoff);
1765 if (!slot)
1766 return NOPAGE_SIGBUS;
1767 page = gfn_to_page(slot, pgoff);
1768 if (!page)
1769 return NOPAGE_SIGBUS;
1770 get_page(page);
1771 return page;
1772 }
1773
1774 static struct vm_operations_struct kvm_dev_vm_ops = {
1775 .nopage = kvm_dev_nopage,
1776 };
1777
1778 static int kvm_dev_mmap(struct file *file, struct vm_area_struct *vma)
1779 {
1780 vma->vm_ops = &kvm_dev_vm_ops;
1781 return 0;
1782 }
1783
1784 static struct file_operations kvm_chardev_ops = {
1785 .open = kvm_dev_open,
1786 .release = kvm_dev_release,
1787 .unlocked_ioctl = kvm_dev_ioctl,
1788 .compat_ioctl = kvm_dev_ioctl,
1789 .mmap = kvm_dev_mmap,
1790 };
1791
1792 static struct miscdevice kvm_dev = {
1793 MISC_DYNAMIC_MINOR,
1794 "kvm",
1795 &kvm_chardev_ops,
1796 };
1797
1798 static int kvm_reboot(struct notifier_block *notifier, unsigned long val,
1799 void *v)
1800 {
1801 if (val == SYS_RESTART) {
1802 /*
1803 * Some (well, at least mine) BIOSes hang on reboot if
1804 * in vmx root mode.
1805 */
1806 printk(KERN_INFO "kvm: exiting hardware virtualization\n");
1807 on_each_cpu(kvm_arch_ops->hardware_disable, 0, 0, 1);
1808 }
1809 return NOTIFY_OK;
1810 }
1811
1812 static struct notifier_block kvm_reboot_notifier = {
1813 .notifier_call = kvm_reboot,
1814 .priority = 0,
1815 };
1816
1817 static __init void kvm_init_debug(void)
1818 {
1819 struct kvm_stats_debugfs_item *p;
1820
1821 debugfs_dir = debugfs_create_dir("kvm", 0);
1822 for (p = debugfs_entries; p->name; ++p)
1823 p->dentry = debugfs_create_u32(p->name, 0444, debugfs_dir,
1824 p->data);
1825 }
1826
1827 static void kvm_exit_debug(void)
1828 {
1829 struct kvm_stats_debugfs_item *p;
1830
1831 for (p = debugfs_entries; p->name; ++p)
1832 debugfs_remove(p->dentry);
1833 debugfs_remove(debugfs_dir);
1834 }
1835
1836 hpa_t bad_page_address;
1837
1838 int kvm_init_arch(struct kvm_arch_ops *ops, struct module *module)
1839 {
1840 int r;
1841
1842 kvm_arch_ops = ops;
1843
1844 if (!kvm_arch_ops->cpu_has_kvm_support()) {
1845 printk(KERN_ERR "kvm: no hardware support\n");
1846 return -EOPNOTSUPP;
1847 }
1848 if (kvm_arch_ops->disabled_by_bios()) {
1849 printk(KERN_ERR "kvm: disabled by bios\n");
1850 return -EOPNOTSUPP;
1851 }
1852
1853 r = kvm_arch_ops->hardware_setup();
1854 if (r < 0)
1855 return r;
1856
1857 on_each_cpu(kvm_arch_ops->hardware_enable, 0, 0, 1);
1858 register_reboot_notifier(&kvm_reboot_notifier);
1859
1860 kvm_chardev_ops.owner = module;
1861
1862 r = misc_register(&kvm_dev);
1863 if (r) {
1864 printk (KERN_ERR "kvm: misc device register failed\n");
1865 goto out_free;
1866 }
1867
1868 return r;
1869
1870 out_free:
1871 unregister_reboot_notifier(&kvm_reboot_notifier);
1872 on_each_cpu(kvm_arch_ops->hardware_disable, 0, 0, 1);
1873 kvm_arch_ops->hardware_unsetup();
1874 return r;
1875 }
1876
1877 void kvm_exit_arch(void)
1878 {
1879 misc_deregister(&kvm_dev);
1880
1881 unregister_reboot_notifier(&kvm_reboot_notifier);
1882 on_each_cpu(kvm_arch_ops->hardware_disable, 0, 0, 1);
1883 kvm_arch_ops->hardware_unsetup();
1884 }
1885
1886 static __init int kvm_init(void)
1887 {
1888 static struct page *bad_page;
1889 int r = 0;
1890
1891 kvm_init_debug();
1892
1893 if ((bad_page = alloc_page(GFP_KERNEL)) == NULL) {
1894 r = -ENOMEM;
1895 goto out;
1896 }
1897
1898 bad_page_address = page_to_pfn(bad_page) << PAGE_SHIFT;
1899 memset(__va(bad_page_address), 0, PAGE_SIZE);
1900
1901 return r;
1902
1903 out:
1904 kvm_exit_debug();
1905 return r;
1906 }
1907
1908 static __exit void kvm_exit(void)
1909 {
1910 kvm_exit_debug();
1911 __free_page(pfn_to_page(bad_page_address >> PAGE_SHIFT));
1912 }
1913
1914 module_init(kvm_init)
1915 module_exit(kvm_exit)
1916
1917 EXPORT_SYMBOL_GPL(kvm_init_arch);
1918 EXPORT_SYMBOL_GPL(kvm_exit_arch);
linux v2.6.22.y-op