projects / deliverable / linux.git / blob
? search:
summary | shortlog | log | commit | commitdiff | tree
blame | history | raw | HEAD
Merge branch 'for-linus' of master.kernel.org:/home/rmk/linux-2.6-arm
[deliverable/linux.git] / drivers / kvm / kvm_main.c
1 /*
2 * Kernel-based Virtual Machine driver for Linux
3 *
4 * This module enables machines with Intel VT-x extensions to run virtual
5 * machines without emulation or binary translation.
6 *
7 * Copyright (C) 2006 Qumranet, Inc.
8 *
9 * Authors:
10 * Avi Kivity <avi@qumranet.com>
11 * Yaniv Kamay <yaniv@qumranet.com>
12 *
13 * This work is licensed under the terms of the GNU GPL, version 2. See
14 * the COPYING file in the top-level directory.
15 *
16 */
17
18 #include "kvm.h"
19
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 mutex_unlock(&vcpu->mutex);
209 }
210
211 static int kvm_dev_open(struct inode *inode, struct file *filp)
212 {
213 struct kvm *kvm = kzalloc(sizeof(struct kvm), GFP_KERNEL);
214 int i;
215
216 if (!kvm)
217 return -ENOMEM;
218
219 spin_lock_init(&kvm->lock);
220 INIT_LIST_HEAD(&kvm->active_mmu_pages);
221 for (i = 0; i < KVM_MAX_VCPUS; ++i) {
222 struct kvm_vcpu *vcpu = &kvm->vcpus[i];
223
224 mutex_init(&vcpu->mutex);
225 vcpu->mmu.root_hpa = INVALID_PAGE;
226 INIT_LIST_HEAD(&vcpu->free_pages);
227 }
228 filp->private_data = kvm;
229 return 0;
230 }
231
232 /*
233 * Free any memory in @free but not in @dont.
234 */
235 static void kvm_free_physmem_slot(struct kvm_memory_slot *free,
236 struct kvm_memory_slot *dont)
237 {
238 int i;
239
240 if (!dont || free->phys_mem != dont->phys_mem)
241 if (free->phys_mem) {
242 for (i = 0; i < free->npages; ++i)
243 __free_page(free->phys_mem[i]);
244 vfree(free->phys_mem);
245 }
246
247 if (!dont || free->dirty_bitmap != dont->dirty_bitmap)
248 vfree(free->dirty_bitmap);
249
250 free->phys_mem = 0;
251 free->npages = 0;
252 free->dirty_bitmap = 0;
253 }
254
255 static void kvm_free_physmem(struct kvm *kvm)
256 {
257 int i;
258
259 for (i = 0; i < kvm->nmemslots; ++i)
260 kvm_free_physmem_slot(&kvm->memslots[i], 0);
261 }
262
263 static void kvm_free_vcpu(struct kvm_vcpu *vcpu)
264 {
265 kvm_arch_ops->vcpu_free(vcpu);
266 kvm_mmu_destroy(vcpu);
267 }
268
269 static void kvm_free_vcpus(struct kvm *kvm)
270 {
271 unsigned int i;
272
273 for (i = 0; i < KVM_MAX_VCPUS; ++i)
274 kvm_free_vcpu(&kvm->vcpus[i]);
275 }
276
277 static int kvm_dev_release(struct inode *inode, struct file *filp)
278 {
279 struct kvm *kvm = filp->private_data;
280
281 kvm_free_vcpus(kvm);
282 kvm_free_physmem(kvm);
283 kfree(kvm);
284 return 0;
285 }
286
287 static void inject_gp(struct kvm_vcpu *vcpu)
288 {
289 kvm_arch_ops->inject_gp(vcpu, 0);
290 }
291
292 static int pdptrs_have_reserved_bits_set(struct kvm_vcpu *vcpu,
293 unsigned long cr3)
294 {
295 gfn_t pdpt_gfn = cr3 >> PAGE_SHIFT;
296 unsigned offset = (cr3 & (PAGE_SIZE-1)) >> 5;
297 int i;
298 u64 pdpte;
299 u64 *pdpt;
300 struct kvm_memory_slot *memslot;
301
302 spin_lock(&vcpu->kvm->lock);
303 memslot = gfn_to_memslot(vcpu->kvm, pdpt_gfn);
304 /* FIXME: !memslot - emulate? 0xff? */
305 pdpt = kmap_atomic(gfn_to_page(memslot, pdpt_gfn), KM_USER0);
306
307 for (i = 0; i < 4; ++i) {
308 pdpte = pdpt[offset + i];
309 if ((pdpte & 1) && (pdpte & 0xfffffff0000001e6ull))
310 break;
311 }
312
313 kunmap_atomic(pdpt, KM_USER0);
314 spin_unlock(&vcpu->kvm->lock);
315
316 return i != 4;
317 }
318
319 void set_cr0(struct kvm_vcpu *vcpu, unsigned long cr0)
320 {
321 if (cr0 & CR0_RESEVED_BITS) {
322 printk(KERN_DEBUG "set_cr0: 0x%lx #GP, reserved bits 0x%lx\n",
323 cr0, vcpu->cr0);
324 inject_gp(vcpu);
325 return;
326 }
327
328 if ((cr0 & CR0_NW_MASK) && !(cr0 & CR0_CD_MASK)) {
329 printk(KERN_DEBUG "set_cr0: #GP, CD == 0 && NW == 1\n");
330 inject_gp(vcpu);
331 return;
332 }
333
334 if ((cr0 & CR0_PG_MASK) && !(cr0 & CR0_PE_MASK)) {
335 printk(KERN_DEBUG "set_cr0: #GP, set PG flag "
336 "and a clear PE flag\n");
337 inject_gp(vcpu);
338 return;
339 }
340
341 if (!is_paging(vcpu) && (cr0 & CR0_PG_MASK)) {
342 #ifdef CONFIG_X86_64
343 if ((vcpu->shadow_efer & EFER_LME)) {
344 int cs_db, cs_l;
345
346 if (!is_pae(vcpu)) {
347 printk(KERN_DEBUG "set_cr0: #GP, start paging "
348 "in long mode while PAE is disabled\n");
349 inject_gp(vcpu);
350 return;
351 }
352 kvm_arch_ops->get_cs_db_l_bits(vcpu, &cs_db, &cs_l);
353 if (cs_l) {
354 printk(KERN_DEBUG "set_cr0: #GP, start paging "
355 "in long mode while CS.L == 1\n");
356 inject_gp(vcpu);
357 return;
358
359 }
360 } else
361 #endif
362 if (is_pae(vcpu) &&
363 pdptrs_have_reserved_bits_set(vcpu, vcpu->cr3)) {
364 printk(KERN_DEBUG "set_cr0: #GP, pdptrs "
365 "reserved bits\n");
366 inject_gp(vcpu);
367 return;
368 }
369
370 }
371
372 kvm_arch_ops->set_cr0(vcpu, cr0);
373 vcpu->cr0 = cr0;
374
375 spin_lock(&vcpu->kvm->lock);
376 kvm_mmu_reset_context(vcpu);
377 spin_unlock(&vcpu->kvm->lock);
378 return;
379 }
380 EXPORT_SYMBOL_GPL(set_cr0);
381
382 void lmsw(struct kvm_vcpu *vcpu, unsigned long msw)
383 {
384 set_cr0(vcpu, (vcpu->cr0 & ~0x0ful) | (msw & 0x0f));
385 }
386 EXPORT_SYMBOL_GPL(lmsw);
387
388 void set_cr4(struct kvm_vcpu *vcpu, unsigned long cr4)
389 {
390 if (cr4 & CR4_RESEVED_BITS) {
391 printk(KERN_DEBUG "set_cr4: #GP, reserved bits\n");
392 inject_gp(vcpu);
393 return;
394 }
395
396 if (kvm_arch_ops->is_long_mode(vcpu)) {
397 if (!(cr4 & CR4_PAE_MASK)) {
398 printk(KERN_DEBUG "set_cr4: #GP, clearing PAE while "
399 "in long mode\n");
400 inject_gp(vcpu);
401 return;
402 }
403 } else if (is_paging(vcpu) && !is_pae(vcpu) && (cr4 & CR4_PAE_MASK)
404 && pdptrs_have_reserved_bits_set(vcpu, vcpu->cr3)) {
405 printk(KERN_DEBUG "set_cr4: #GP, pdptrs reserved bits\n");
406 inject_gp(vcpu);
407 }
408
409 if (cr4 & CR4_VMXE_MASK) {
410 printk(KERN_DEBUG "set_cr4: #GP, setting VMXE\n");
411 inject_gp(vcpu);
412 return;
413 }
414 kvm_arch_ops->set_cr4(vcpu, cr4);
415 spin_lock(&vcpu->kvm->lock);
416 kvm_mmu_reset_context(vcpu);
417 spin_unlock(&vcpu->kvm->lock);
418 }
419 EXPORT_SYMBOL_GPL(set_cr4);
420
421 void set_cr3(struct kvm_vcpu *vcpu, unsigned long cr3)
422 {
423 if (kvm_arch_ops->is_long_mode(vcpu)) {
424 if ( cr3 & CR3_L_MODE_RESEVED_BITS) {
425 printk(KERN_DEBUG "set_cr3: #GP, reserved bits\n");
426 inject_gp(vcpu);
427 return;
428 }
429 } else {
430 if (cr3 & CR3_RESEVED_BITS) {
431 printk(KERN_DEBUG "set_cr3: #GP, reserved bits\n");
432 inject_gp(vcpu);
433 return;
434 }
435 if (is_paging(vcpu) && is_pae(vcpu) &&
436 pdptrs_have_reserved_bits_set(vcpu, cr3)) {
437 printk(KERN_DEBUG "set_cr3: #GP, pdptrs "
438 "reserved bits\n");
439 inject_gp(vcpu);
440 return;
441 }
442 }
443
444 vcpu->cr3 = cr3;
445 spin_lock(&vcpu->kvm->lock);
446 vcpu->mmu.new_cr3(vcpu);
447 spin_unlock(&vcpu->kvm->lock);
448 }
449 EXPORT_SYMBOL_GPL(set_cr3);
450
451 void set_cr8(struct kvm_vcpu *vcpu, unsigned long cr8)
452 {
453 if ( cr8 & CR8_RESEVED_BITS) {
454 printk(KERN_DEBUG "set_cr8: #GP, reserved bits 0x%lx\n", cr8);
455 inject_gp(vcpu);
456 return;
457 }
458 vcpu->cr8 = cr8;
459 }
460 EXPORT_SYMBOL_GPL(set_cr8);
461
462 void fx_init(struct kvm_vcpu *vcpu)
463 {
464 struct __attribute__ ((__packed__)) fx_image_s {
465 u16 control; //fcw
466 u16 status; //fsw
467 u16 tag; // ftw
468 u16 opcode; //fop
469 u64 ip; // fpu ip
470 u64 operand;// fpu dp
471 u32 mxcsr;
472 u32 mxcsr_mask;
473
474 } *fx_image;
475
476 fx_save(vcpu->host_fx_image);
477 fpu_init();
478 fx_save(vcpu->guest_fx_image);
479 fx_restore(vcpu->host_fx_image);
480
481 fx_image = (struct fx_image_s *)vcpu->guest_fx_image;
482 fx_image->mxcsr = 0x1f80;
483 memset(vcpu->guest_fx_image + sizeof(struct fx_image_s),
484 0, FX_IMAGE_SIZE - sizeof(struct fx_image_s));
485 }
486 EXPORT_SYMBOL_GPL(fx_init);
487
488 /*
489 * Creates some virtual cpus. Good luck creating more than one.
490 */
491 static int kvm_dev_ioctl_create_vcpu(struct kvm *kvm, int n)
492 {
493 int r;
494 struct kvm_vcpu *vcpu;
495
496 r = -EINVAL;
497 if (n < 0 || n >= KVM_MAX_VCPUS)
498 goto out;
499
500 vcpu = &kvm->vcpus[n];
501
502 mutex_lock(&vcpu->mutex);
503
504 if (vcpu->vmcs) {
505 mutex_unlock(&vcpu->mutex);
506 return -EEXIST;
507 }
508
509 vcpu->host_fx_image = (char*)ALIGN((hva_t)vcpu->fx_buf,
510 FX_IMAGE_ALIGN);
511 vcpu->guest_fx_image = vcpu->host_fx_image + FX_IMAGE_SIZE;
512
513 vcpu->cpu = -1; /* First load will set up TR */
514 vcpu->kvm = kvm;
515 r = kvm_arch_ops->vcpu_create(vcpu);
516 if (r < 0)
517 goto out_free_vcpus;
518
519 kvm_arch_ops->vcpu_load(vcpu);
520
521 r = kvm_arch_ops->vcpu_setup(vcpu);
522 if (r >= 0)
523 r = kvm_mmu_init(vcpu);
524
525 vcpu_put(vcpu);
526
527 if (r < 0)
528 goto out_free_vcpus;
529
530 return 0;
531
532 out_free_vcpus:
533 kvm_free_vcpu(vcpu);
534 mutex_unlock(&vcpu->mutex);
535 out:
536 return r;
537 }
538
539 /*
540 * Allocate some memory and give it an address in the guest physical address
541 * space.
542 *
543 * Discontiguous memory is allowed, mostly for framebuffers.
544 */
545 static int kvm_dev_ioctl_set_memory_region(struct kvm *kvm,
546 struct kvm_memory_region *mem)
547 {
548 int r;
549 gfn_t base_gfn;
550 unsigned long npages;
551 unsigned long i;
552 struct kvm_memory_slot *memslot;
553 struct kvm_memory_slot old, new;
554 int memory_config_version;
555
556 r = -EINVAL;
557 /* General sanity checks */
558 if (mem->memory_size & (PAGE_SIZE - 1))
559 goto out;
560 if (mem->guest_phys_addr & (PAGE_SIZE - 1))
561 goto out;
562 if (mem->slot >= KVM_MEMORY_SLOTS)
563 goto out;
564 if (mem->guest_phys_addr + mem->memory_size < mem->guest_phys_addr)
565 goto out;
566
567 memslot = &kvm->memslots[mem->slot];
568 base_gfn = mem->guest_phys_addr >> PAGE_SHIFT;
569 npages = mem->memory_size >> PAGE_SHIFT;
570
571 if (!npages)
572 mem->flags &= ~KVM_MEM_LOG_DIRTY_PAGES;
573
574 raced:
575 spin_lock(&kvm->lock);
576
577 memory_config_version = kvm->memory_config_version;
578 new = old = *memslot;
579
580 new.base_gfn = base_gfn;
581 new.npages = npages;
582 new.flags = mem->flags;
583
584 /* Disallow changing a memory slot's size. */
585 r = -EINVAL;
586 if (npages && old.npages && npages != old.npages)
587 goto out_unlock;
588
589 /* Check for overlaps */
590 r = -EEXIST;
591 for (i = 0; i < KVM_MEMORY_SLOTS; ++i) {
592 struct kvm_memory_slot *s = &kvm->memslots[i];
593
594 if (s == memslot)
595 continue;
596 if (!((base_gfn + npages <= s->base_gfn) ||
597 (base_gfn >= s->base_gfn + s->npages)))
598 goto out_unlock;
599 }
600 /*
601 * Do memory allocations outside lock. memory_config_version will
602 * detect any races.
603 */
604 spin_unlock(&kvm->lock);
605
606 /* Deallocate if slot is being removed */
607 if (!npages)
608 new.phys_mem = 0;
609
610 /* Free page dirty bitmap if unneeded */
611 if (!(new.flags & KVM_MEM_LOG_DIRTY_PAGES))
612 new.dirty_bitmap = 0;
613
614 r = -ENOMEM;
615
616 /* Allocate if a slot is being created */
617 if (npages && !new.phys_mem) {
618 new.phys_mem = vmalloc(npages * sizeof(struct page *));
619
620 if (!new.phys_mem)
621 goto out_free;
622
623 memset(new.phys_mem, 0, npages * sizeof(struct page *));
624 for (i = 0; i < npages; ++i) {
625 new.phys_mem[i] = alloc_page(GFP_HIGHUSER
626 | __GFP_ZERO);
627 if (!new.phys_mem[i])
628 goto out_free;
629 }
630 }
631
632 /* Allocate page dirty bitmap if needed */
633 if ((new.flags & KVM_MEM_LOG_DIRTY_PAGES) && !new.dirty_bitmap) {
634 unsigned dirty_bytes = ALIGN(npages, BITS_PER_LONG) / 8;
635
636 new.dirty_bitmap = vmalloc(dirty_bytes);
637 if (!new.dirty_bitmap)
638 goto out_free;
639 memset(new.dirty_bitmap, 0, dirty_bytes);
640 }
641
642 spin_lock(&kvm->lock);
643
644 if (memory_config_version != kvm->memory_config_version) {
645 spin_unlock(&kvm->lock);
646 kvm_free_physmem_slot(&new, &old);
647 goto raced;
648 }
649
650 r = -EAGAIN;
651 if (kvm->busy)
652 goto out_unlock;
653
654 if (mem->slot >= kvm->nmemslots)
655 kvm->nmemslots = mem->slot + 1;
656
657 *memslot = new;
658 ++kvm->memory_config_version;
659
660 spin_unlock(&kvm->lock);
661
662 for (i = 0; i < KVM_MAX_VCPUS; ++i) {
663 struct kvm_vcpu *vcpu;
664
665 vcpu = vcpu_load(kvm, i);
666 if (!vcpu)
667 continue;
668 kvm_mmu_reset_context(vcpu);
669 vcpu_put(vcpu);
670 }
671
672 kvm_free_physmem_slot(&old, &new);
673 return 0;
674
675 out_unlock:
676 spin_unlock(&kvm->lock);
677 out_free:
678 kvm_free_physmem_slot(&new, &old);
679 out:
680 return r;
681 }
682
683 /*
684 * Get (and clear) the dirty memory log for a memory slot.
685 */
686 static int kvm_dev_ioctl_get_dirty_log(struct kvm *kvm,
687 struct kvm_dirty_log *log)
688 {
689 struct kvm_memory_slot *memslot;
690 int r, i;
691 int n;
692 unsigned long any = 0;
693
694 spin_lock(&kvm->lock);
695
696 /*
697 * Prevent changes to guest memory configuration even while the lock
698 * is not taken.
699 */
700 ++kvm->busy;
701 spin_unlock(&kvm->lock);
702 r = -EINVAL;
703 if (log->slot >= KVM_MEMORY_SLOTS)
704 goto out;
705
706 memslot = &kvm->memslots[log->slot];
707 r = -ENOENT;
708 if (!memslot->dirty_bitmap)
709 goto out;
710
711 n = ALIGN(memslot->npages, 8) / 8;
712
713 for (i = 0; !any && i < n; ++i)
714 any = memslot->dirty_bitmap[i];
715
716 r = -EFAULT;
717 if (copy_to_user(log->dirty_bitmap, memslot->dirty_bitmap, n))
718 goto out;
719
720
721 if (any) {
722 spin_lock(&kvm->lock);
723 kvm_mmu_slot_remove_write_access(kvm, log->slot);
724 spin_unlock(&kvm->lock);
725 memset(memslot->dirty_bitmap, 0, n);
726 for (i = 0; i < KVM_MAX_VCPUS; ++i) {
727 struct kvm_vcpu *vcpu = vcpu_load(kvm, i);
728
729 if (!vcpu)
730 continue;
731 kvm_arch_ops->tlb_flush(vcpu);
732 vcpu_put(vcpu);
733 }
734 }
735
736 r = 0;
737
738 out:
739 spin_lock(&kvm->lock);
740 --kvm->busy;
741 spin_unlock(&kvm->lock);
742 return r;
743 }
744
745 struct kvm_memory_slot *gfn_to_memslot(struct kvm *kvm, gfn_t gfn)
746 {
747 int i;
748
749 for (i = 0; i < kvm->nmemslots; ++i) {
750 struct kvm_memory_slot *memslot = &kvm->memslots[i];
751
752 if (gfn >= memslot->base_gfn
753 && gfn < memslot->base_gfn + memslot->npages)
754 return memslot;
755 }
756 return 0;
757 }
758 EXPORT_SYMBOL_GPL(gfn_to_memslot);
759
760 void mark_page_dirty(struct kvm *kvm, gfn_t gfn)
761 {
762 int i;
763 struct kvm_memory_slot *memslot = 0;
764 unsigned long rel_gfn;
765
766 for (i = 0; i < kvm->nmemslots; ++i) {
767 memslot = &kvm->memslots[i];
768
769 if (gfn >= memslot->base_gfn
770 && gfn < memslot->base_gfn + memslot->npages) {
771
772 if (!memslot || !memslot->dirty_bitmap)
773 return;
774
775 rel_gfn = gfn - memslot->base_gfn;
776
777 /* avoid RMW */
778 if (!test_bit(rel_gfn, memslot->dirty_bitmap))
779 set_bit(rel_gfn, memslot->dirty_bitmap);
780 return;
781 }
782 }
783 }
784
785 static int emulator_read_std(unsigned long addr,
786 unsigned long *val,
787 unsigned int bytes,
788 struct x86_emulate_ctxt *ctxt)
789 {
790 struct kvm_vcpu *vcpu = ctxt->vcpu;
791 void *data = val;
792
793 while (bytes) {
794 gpa_t gpa = vcpu->mmu.gva_to_gpa(vcpu, addr);
795 unsigned offset = addr & (PAGE_SIZE-1);
796 unsigned tocopy = min(bytes, (unsigned)PAGE_SIZE - offset);
797 unsigned long pfn;
798 struct kvm_memory_slot *memslot;
799 void *page;
800
801 if (gpa == UNMAPPED_GVA)
802 return X86EMUL_PROPAGATE_FAULT;
803 pfn = gpa >> PAGE_SHIFT;
804 memslot = gfn_to_memslot(vcpu->kvm, pfn);
805 if (!memslot)
806 return X86EMUL_UNHANDLEABLE;
807 page = kmap_atomic(gfn_to_page(memslot, pfn), KM_USER0);
808
809 memcpy(data, page + offset, tocopy);
810
811 kunmap_atomic(page, KM_USER0);
812
813 bytes -= tocopy;
814 data += tocopy;
815 addr += tocopy;
816 }
817
818 return X86EMUL_CONTINUE;
819 }
820
821 static int emulator_write_std(unsigned long addr,
822 unsigned long val,
823 unsigned int bytes,
824 struct x86_emulate_ctxt *ctxt)
825 {
826 printk(KERN_ERR "emulator_write_std: addr %lx n %d\n",
827 addr, bytes);
828 return X86EMUL_UNHANDLEABLE;
829 }
830
831 static int emulator_read_emulated(unsigned long addr,
832 unsigned long *val,
833 unsigned int bytes,
834 struct x86_emulate_ctxt *ctxt)
835 {
836 struct kvm_vcpu *vcpu = ctxt->vcpu;
837
838 if (vcpu->mmio_read_completed) {
839 memcpy(val, vcpu->mmio_data, bytes);
840 vcpu->mmio_read_completed = 0;
841 return X86EMUL_CONTINUE;
842 } else if (emulator_read_std(addr, val, bytes, ctxt)
843 == X86EMUL_CONTINUE)
844 return X86EMUL_CONTINUE;
845 else {
846 gpa_t gpa = vcpu->mmu.gva_to_gpa(vcpu, addr);
847 if (gpa == UNMAPPED_GVA)
848 return vcpu_printf(vcpu, "not present\n"), X86EMUL_PROPAGATE_FAULT;
849 vcpu->mmio_needed = 1;
850 vcpu->mmio_phys_addr = gpa;
851 vcpu->mmio_size = bytes;
852 vcpu->mmio_is_write = 0;
853
854 return X86EMUL_UNHANDLEABLE;
855 }
856 }
857
858 static int emulator_write_emulated(unsigned long addr,
859 unsigned long val,
860 unsigned int bytes,
861 struct x86_emulate_ctxt *ctxt)
862 {
863 struct kvm_vcpu *vcpu = ctxt->vcpu;
864 gpa_t gpa = vcpu->mmu.gva_to_gpa(vcpu, addr);
865
866 if (gpa == UNMAPPED_GVA)
867 return X86EMUL_PROPAGATE_FAULT;
868
869 vcpu->mmio_needed = 1;
870 vcpu->mmio_phys_addr = gpa;
871 vcpu->mmio_size = bytes;
872 vcpu->mmio_is_write = 1;
873 memcpy(vcpu->mmio_data, &val, bytes);
874
875 return X86EMUL_CONTINUE;
876 }
877
878 static int emulator_cmpxchg_emulated(unsigned long addr,
879 unsigned long old,
880 unsigned long new,
881 unsigned int bytes,
882 struct x86_emulate_ctxt *ctxt)
883 {
884 static int reported;
885
886 if (!reported) {
887 reported = 1;
888 printk(KERN_WARNING "kvm: emulating exchange as write\n");
889 }
890 return emulator_write_emulated(addr, new, bytes, ctxt);
891 }
892
893 static unsigned long get_segment_base(struct kvm_vcpu *vcpu, int seg)
894 {
895 return kvm_arch_ops->get_segment_base(vcpu, seg);
896 }
897
898 int emulate_invlpg(struct kvm_vcpu *vcpu, gva_t address)
899 {
900 spin_lock(&vcpu->kvm->lock);
901 vcpu->mmu.inval_page(vcpu, address);
902 spin_unlock(&vcpu->kvm->lock);
903 kvm_arch_ops->invlpg(vcpu, address);
904 return X86EMUL_CONTINUE;
905 }
906
907 int emulate_clts(struct kvm_vcpu *vcpu)
908 {
909 unsigned long cr0 = vcpu->cr0;
910
911 cr0 &= ~CR0_TS_MASK;
912 kvm_arch_ops->set_cr0(vcpu, cr0);
913 return X86EMUL_CONTINUE;
914 }
915
916 int emulator_get_dr(struct x86_emulate_ctxt* ctxt, int dr, unsigned long *dest)
917 {
918 struct kvm_vcpu *vcpu = ctxt->vcpu;
919
920 switch (dr) {
921 case 0 ... 3:
922 *dest = kvm_arch_ops->get_dr(vcpu, dr);
923 return X86EMUL_CONTINUE;
924 default:
925 printk(KERN_DEBUG "%s: unexpected dr %u\n",
926 __FUNCTION__, dr);
927 return X86EMUL_UNHANDLEABLE;
928 }
929 }
930
931 int emulator_set_dr(struct x86_emulate_ctxt *ctxt, int dr, unsigned long value)
932 {
933 unsigned long mask = (ctxt->mode == X86EMUL_MODE_PROT64) ? ~0ULL : ~0U;
934 int exception;
935
936 kvm_arch_ops->set_dr(ctxt->vcpu, dr, value & mask, &exception);
937 if (exception) {
938 /* FIXME: better handling */
939 return X86EMUL_UNHANDLEABLE;
940 }
941 return X86EMUL_CONTINUE;
942 }
943
944 static void report_emulation_failure(struct x86_emulate_ctxt *ctxt)
945 {
946 static int reported;
947 u8 opcodes[4];
948 unsigned long rip = ctxt->vcpu->rip;
949 unsigned long rip_linear;
950
951 rip_linear = rip + get_segment_base(ctxt->vcpu, VCPU_SREG_CS);
952
953 if (reported)
954 return;
955
956 emulator_read_std(rip_linear, (void *)opcodes, 4, ctxt);
957
958 printk(KERN_ERR "emulation failed but !mmio_needed?"
959 " rip %lx %02x %02x %02x %02x\n",
960 rip, opcodes[0], opcodes[1], opcodes[2], opcodes[3]);
961 reported = 1;
962 }
963
964 struct x86_emulate_ops emulate_ops = {
965 .read_std = emulator_read_std,
966 .write_std = emulator_write_std,
967 .read_emulated = emulator_read_emulated,
968 .write_emulated = emulator_write_emulated,
969 .cmpxchg_emulated = emulator_cmpxchg_emulated,
970 };
971
972 int emulate_instruction(struct kvm_vcpu *vcpu,
973 struct kvm_run *run,
974 unsigned long cr2,
975 u16 error_code)
976 {
977 struct x86_emulate_ctxt emulate_ctxt;
978 int r;
979 int cs_db, cs_l;
980
981 kvm_arch_ops->cache_regs(vcpu);
982
983 kvm_arch_ops->get_cs_db_l_bits(vcpu, &cs_db, &cs_l);
984
985 emulate_ctxt.vcpu = vcpu;
986 emulate_ctxt.eflags = kvm_arch_ops->get_rflags(vcpu);
987 emulate_ctxt.cr2 = cr2;
988 emulate_ctxt.mode = (emulate_ctxt.eflags & X86_EFLAGS_VM)
989 ? X86EMUL_MODE_REAL : cs_l
990 ? X86EMUL_MODE_PROT64 : cs_db
991 ? X86EMUL_MODE_PROT32 : X86EMUL_MODE_PROT16;
992
993 if (emulate_ctxt.mode == X86EMUL_MODE_PROT64) {
994 emulate_ctxt.cs_base = 0;
995 emulate_ctxt.ds_base = 0;
996 emulate_ctxt.es_base = 0;
997 emulate_ctxt.ss_base = 0;
998 } else {
999 emulate_ctxt.cs_base = get_segment_base(vcpu, VCPU_SREG_CS);
1000 emulate_ctxt.ds_base = get_segment_base(vcpu, VCPU_SREG_DS);
1001 emulate_ctxt.es_base = get_segment_base(vcpu, VCPU_SREG_ES);
1002 emulate_ctxt.ss_base = get_segment_base(vcpu, VCPU_SREG_SS);
1003 }
1004
1005 emulate_ctxt.gs_base = get_segment_base(vcpu, VCPU_SREG_GS);
1006 emulate_ctxt.fs_base = get_segment_base(vcpu, VCPU_SREG_FS);
1007
1008 vcpu->mmio_is_write = 0;
1009 r = x86_emulate_memop(&emulate_ctxt, &emulate_ops);
1010
1011 if ((r || vcpu->mmio_is_write) && run) {
1012 run->mmio.phys_addr = vcpu->mmio_phys_addr;
1013 memcpy(run->mmio.data, vcpu->mmio_data, 8);
1014 run->mmio.len = vcpu->mmio_size;
1015 run->mmio.is_write = vcpu->mmio_is_write;
1016 }
1017
1018 if (r) {
1019 if (!vcpu->mmio_needed) {
1020 report_emulation_failure(&emulate_ctxt);
1021 return EMULATE_FAIL;
1022 }
1023 return EMULATE_DO_MMIO;
1024 }
1025
1026 kvm_arch_ops->decache_regs(vcpu);
1027 kvm_arch_ops->set_rflags(vcpu, emulate_ctxt.eflags);
1028
1029 if (vcpu->mmio_is_write)
1030 return EMULATE_DO_MMIO;
1031
1032 return EMULATE_DONE;
1033 }
1034 EXPORT_SYMBOL_GPL(emulate_instruction);
1035
1036 static u64 mk_cr_64(u64 curr_cr, u32 new_val)
1037 {
1038 return (curr_cr & ~((1ULL << 32) - 1)) | new_val;
1039 }
1040
1041 void realmode_lgdt(struct kvm_vcpu *vcpu, u16 limit, unsigned long base)
1042 {
1043 struct descriptor_table dt = { limit, base };
1044
1045 kvm_arch_ops->set_gdt(vcpu, &dt);
1046 }
1047
1048 void realmode_lidt(struct kvm_vcpu *vcpu, u16 limit, unsigned long base)
1049 {
1050 struct descriptor_table dt = { limit, base };
1051
1052 kvm_arch_ops->set_idt(vcpu, &dt);
1053 }
1054
1055 void realmode_lmsw(struct kvm_vcpu *vcpu, unsigned long msw,
1056 unsigned long *rflags)
1057 {
1058 lmsw(vcpu, msw);
1059 *rflags = kvm_arch_ops->get_rflags(vcpu);
1060 }
1061
1062 unsigned long realmode_get_cr(struct kvm_vcpu *vcpu, int cr)
1063 {
1064 switch (cr) {
1065 case 0:
1066 return vcpu->cr0;
1067 case 2:
1068 return vcpu->cr2;
1069 case 3:
1070 return vcpu->cr3;
1071 case 4:
1072 return vcpu->cr4;
1073 default:
1074 vcpu_printf(vcpu, "%s: unexpected cr %u\n", __FUNCTION__, cr);
1075 return 0;
1076 }
1077 }
1078
1079 void realmode_set_cr(struct kvm_vcpu *vcpu, int cr, unsigned long val,
1080 unsigned long *rflags)
1081 {
1082 switch (cr) {
1083 case 0:
1084 set_cr0(vcpu, mk_cr_64(vcpu->cr0, val));
1085 *rflags = kvm_arch_ops->get_rflags(vcpu);
1086 break;
1087 case 2:
1088 vcpu->cr2 = val;
1089 break;
1090 case 3:
1091 set_cr3(vcpu, val);
1092 break;
1093 case 4:
1094 set_cr4(vcpu, mk_cr_64(vcpu->cr4, val));
1095 break;
1096 default:
1097 vcpu_printf(vcpu, "%s: unexpected cr %u\n", __FUNCTION__, cr);
1098 }
1099 }
1100
1101 /*
1102 * Reads an msr value (of 'msr_index') into 'pdata'.
1103 * Returns 0 on success, non-0 otherwise.
1104 * Assumes vcpu_load() was already called.
1105 */
1106 static int get_msr(struct kvm_vcpu *vcpu, u32 msr_index, u64 *pdata)
1107 {
1108 return kvm_arch_ops->get_msr(vcpu, msr_index, pdata);
1109 }
1110
1111 #ifdef CONFIG_X86_64
1112
1113 void set_efer(struct kvm_vcpu *vcpu, u64 efer)
1114 {
1115 if (efer & EFER_RESERVED_BITS) {
1116 printk(KERN_DEBUG "set_efer: 0x%llx #GP, reserved bits\n",
1117 efer);
1118 inject_gp(vcpu);
1119 return;
1120 }
1121
1122 if (is_paging(vcpu)
1123 && (vcpu->shadow_efer & EFER_LME) != (efer & EFER_LME)) {
1124 printk(KERN_DEBUG "set_efer: #GP, change LME while paging\n");
1125 inject_gp(vcpu);
1126 return;
1127 }
1128
1129 kvm_arch_ops->set_efer(vcpu, efer);
1130
1131 efer &= ~EFER_LMA;
1132 efer |= vcpu->shadow_efer & EFER_LMA;
1133
1134 vcpu->shadow_efer = efer;
1135 }
1136 EXPORT_SYMBOL_GPL(set_efer);
1137
1138 #endif
1139
1140 /*
1141 * Writes msr value into into the appropriate "register".
1142 * Returns 0 on success, non-0 otherwise.
1143 * Assumes vcpu_load() was already called.
1144 */
1145 static int set_msr(struct kvm_vcpu *vcpu, u32 msr_index, u64 data)
1146 {
1147 return kvm_arch_ops->set_msr(vcpu, msr_index, data);
1148 }
1149
1150 void kvm_resched(struct kvm_vcpu *vcpu)
1151 {
1152 vcpu_put(vcpu);
1153 cond_resched();
1154 /* Cannot fail - no vcpu unplug yet. */
1155 vcpu_load(vcpu->kvm, vcpu_slot(vcpu));
1156 }
1157 EXPORT_SYMBOL_GPL(kvm_resched);
1158
1159 void load_msrs(struct vmx_msr_entry *e, int n)
1160 {
1161 int i;
1162
1163 for (i = 0; i < n; ++i)
1164 wrmsrl(e[i].index, e[i].data);
1165 }
1166 EXPORT_SYMBOL_GPL(load_msrs);
1167
1168 void save_msrs(struct vmx_msr_entry *e, int n)
1169 {
1170 int i;
1171
1172 for (i = 0; i < n; ++i)
1173 rdmsrl(e[i].index, e[i].data);
1174 }
1175 EXPORT_SYMBOL_GPL(save_msrs);
1176
1177 static int kvm_dev_ioctl_run(struct kvm *kvm, struct kvm_run *kvm_run)
1178 {
1179 struct kvm_vcpu *vcpu;
1180 int r;
1181
1182 if (kvm_run->vcpu < 0 || kvm_run->vcpu >= KVM_MAX_VCPUS)
1183 return -EINVAL;
1184
1185 vcpu = vcpu_load(kvm, kvm_run->vcpu);
1186 if (!vcpu)
1187 return -ENOENT;
1188
1189 if (kvm_run->emulated) {
1190 kvm_arch_ops->skip_emulated_instruction(vcpu);
1191 kvm_run->emulated = 0;
1192 }
1193
1194 if (kvm_run->mmio_completed) {
1195 memcpy(vcpu->mmio_data, kvm_run->mmio.data, 8);
1196 vcpu->mmio_read_completed = 1;
1197 }
1198
1199 vcpu->mmio_needed = 0;
1200
1201 r = kvm_arch_ops->run(vcpu, kvm_run);
1202
1203 vcpu_put(vcpu);
1204 return r;
1205 }
1206
1207 static int kvm_dev_ioctl_get_regs(struct kvm *kvm, struct kvm_regs *regs)
1208 {
1209 struct kvm_vcpu *vcpu;
1210
1211 if (regs->vcpu < 0 || regs->vcpu >= KVM_MAX_VCPUS)
1212 return -EINVAL;
1213
1214 vcpu = vcpu_load(kvm, regs->vcpu);
1215 if (!vcpu)
1216 return -ENOENT;
1217
1218 kvm_arch_ops->cache_regs(vcpu);
1219
1220 regs->rax = vcpu->regs[VCPU_REGS_RAX];
1221 regs->rbx = vcpu->regs[VCPU_REGS_RBX];
1222 regs->rcx = vcpu->regs[VCPU_REGS_RCX];
1223 regs->rdx = vcpu->regs[VCPU_REGS_RDX];
1224 regs->rsi = vcpu->regs[VCPU_REGS_RSI];
1225 regs->rdi = vcpu->regs[VCPU_REGS_RDI];
1226 regs->rsp = vcpu->regs[VCPU_REGS_RSP];
1227 regs->rbp = vcpu->regs[VCPU_REGS_RBP];
1228 #ifdef CONFIG_X86_64
1229 regs->r8 = vcpu->regs[VCPU_REGS_R8];
1230 regs->r9 = vcpu->regs[VCPU_REGS_R9];
1231 regs->r10 = vcpu->regs[VCPU_REGS_R10];
1232 regs->r11 = vcpu->regs[VCPU_REGS_R11];
1233 regs->r12 = vcpu->regs[VCPU_REGS_R12];
1234 regs->r13 = vcpu->regs[VCPU_REGS_R13];
1235 regs->r14 = vcpu->regs[VCPU_REGS_R14];
1236 regs->r15 = vcpu->regs[VCPU_REGS_R15];
1237 #endif
1238
1239 regs->rip = vcpu->rip;
1240 regs->rflags = kvm_arch_ops->get_rflags(vcpu);
1241
1242 /*
1243 * Don't leak debug flags in case they were set for guest debugging
1244 */
1245 if (vcpu->guest_debug.enabled && vcpu->guest_debug.singlestep)
1246 regs->rflags &= ~(X86_EFLAGS_TF | X86_EFLAGS_RF);
1247
1248 vcpu_put(vcpu);
1249
1250 return 0;
1251 }
1252
1253 static int kvm_dev_ioctl_set_regs(struct kvm *kvm, struct kvm_regs *regs)
1254 {
1255 struct kvm_vcpu *vcpu;
1256
1257 if (regs->vcpu < 0 || regs->vcpu >= KVM_MAX_VCPUS)
1258 return -EINVAL;
1259
1260 vcpu = vcpu_load(kvm, regs->vcpu);
1261 if (!vcpu)
1262 return -ENOENT;
1263
1264 vcpu->regs[VCPU_REGS_RAX] = regs->rax;
1265 vcpu->regs[VCPU_REGS_RBX] = regs->rbx;
1266 vcpu->regs[VCPU_REGS_RCX] = regs->rcx;
1267 vcpu->regs[VCPU_REGS_RDX] = regs->rdx;
1268 vcpu->regs[VCPU_REGS_RSI] = regs->rsi;
1269 vcpu->regs[VCPU_REGS_RDI] = regs->rdi;
1270 vcpu->regs[VCPU_REGS_RSP] = regs->rsp;
1271 vcpu->regs[VCPU_REGS_RBP] = regs->rbp;
1272 #ifdef CONFIG_X86_64
1273 vcpu->regs[VCPU_REGS_R8] = regs->r8;
1274 vcpu->regs[VCPU_REGS_R9] = regs->r9;
1275 vcpu->regs[VCPU_REGS_R10] = regs->r10;
1276 vcpu->regs[VCPU_REGS_R11] = regs->r11;
1277 vcpu->regs[VCPU_REGS_R12] = regs->r12;
1278 vcpu->regs[VCPU_REGS_R13] = regs->r13;
1279 vcpu->regs[VCPU_REGS_R14] = regs->r14;
1280 vcpu->regs[VCPU_REGS_R15] = regs->r15;
1281 #endif
1282
1283 vcpu->rip = regs->rip;
1284 kvm_arch_ops->set_rflags(vcpu, regs->rflags);
1285
1286 kvm_arch_ops->decache_regs(vcpu);
1287
1288 vcpu_put(vcpu);
1289
1290 return 0;
1291 }
1292
1293 static void get_segment(struct kvm_vcpu *vcpu,
1294 struct kvm_segment *var, int seg)
1295 {
1296 return kvm_arch_ops->get_segment(vcpu, var, seg);
1297 }
1298
1299 static int kvm_dev_ioctl_get_sregs(struct kvm *kvm, struct kvm_sregs *sregs)
1300 {
1301 struct kvm_vcpu *vcpu;
1302 struct descriptor_table dt;
1303
1304 if (sregs->vcpu < 0 || sregs->vcpu >= KVM_MAX_VCPUS)
1305 return -EINVAL;
1306 vcpu = vcpu_load(kvm, sregs->vcpu);
1307 if (!vcpu)
1308 return -ENOENT;
1309
1310 get_segment(vcpu, &sregs->cs, VCPU_SREG_CS);
1311 get_segment(vcpu, &sregs->ds, VCPU_SREG_DS);
1312 get_segment(vcpu, &sregs->es, VCPU_SREG_ES);
1313 get_segment(vcpu, &sregs->fs, VCPU_SREG_FS);
1314 get_segment(vcpu, &sregs->gs, VCPU_SREG_GS);
1315 get_segment(vcpu, &sregs->ss, VCPU_SREG_SS);
1316
1317 get_segment(vcpu, &sregs->tr, VCPU_SREG_TR);
1318 get_segment(vcpu, &sregs->ldt, VCPU_SREG_LDTR);
1319
1320 kvm_arch_ops->get_idt(vcpu, &dt);
1321 sregs->idt.limit = dt.limit;
1322 sregs->idt.base = dt.base;
1323 kvm_arch_ops->get_gdt(vcpu, &dt);
1324 sregs->gdt.limit = dt.limit;
1325 sregs->gdt.base = dt.base;
1326
1327 sregs->cr0 = vcpu->cr0;
1328 sregs->cr2 = vcpu->cr2;
1329 sregs->cr3 = vcpu->cr3;
1330 sregs->cr4 = vcpu->cr4;
1331 sregs->cr8 = vcpu->cr8;
1332 sregs->efer = vcpu->shadow_efer;
1333 sregs->apic_base = vcpu->apic_base;
1334
1335 memcpy(sregs->interrupt_bitmap, vcpu->irq_pending,
1336 sizeof sregs->interrupt_bitmap);
1337
1338 vcpu_put(vcpu);
1339
1340 return 0;
1341 }
1342
1343 static void set_segment(struct kvm_vcpu *vcpu,
1344 struct kvm_segment *var, int seg)
1345 {
1346 return kvm_arch_ops->set_segment(vcpu, var, seg);
1347 }
1348
1349 static int kvm_dev_ioctl_set_sregs(struct kvm *kvm, struct kvm_sregs *sregs)
1350 {
1351 struct kvm_vcpu *vcpu;
1352 int mmu_reset_needed = 0;
1353 int i;
1354 struct descriptor_table dt;
1355
1356 if (sregs->vcpu < 0 || sregs->vcpu >= KVM_MAX_VCPUS)
1357 return -EINVAL;
1358 vcpu = vcpu_load(kvm, sregs->vcpu);
1359 if (!vcpu)
1360 return -ENOENT;
1361
1362 set_segment(vcpu, &sregs->cs, VCPU_SREG_CS);
1363 set_segment(vcpu, &sregs->ds, VCPU_SREG_DS);
1364 set_segment(vcpu, &sregs->es, VCPU_SREG_ES);
1365 set_segment(vcpu, &sregs->fs, VCPU_SREG_FS);
1366 set_segment(vcpu, &sregs->gs, VCPU_SREG_GS);
1367 set_segment(vcpu, &sregs->ss, VCPU_SREG_SS);
1368
1369 set_segment(vcpu, &sregs->tr, VCPU_SREG_TR);
1370 set_segment(vcpu, &sregs->ldt, VCPU_SREG_LDTR);
1371
1372 dt.limit = sregs->idt.limit;
1373 dt.base = sregs->idt.base;
1374 kvm_arch_ops->set_idt(vcpu, &dt);
1375 dt.limit = sregs->gdt.limit;
1376 dt.base = sregs->gdt.base;
1377 kvm_arch_ops->set_gdt(vcpu, &dt);
1378
1379 vcpu->cr2 = sregs->cr2;
1380 mmu_reset_needed |= vcpu->cr3 != sregs->cr3;
1381 vcpu->cr3 = sregs->cr3;
1382
1383 vcpu->cr8 = sregs->cr8;
1384
1385 mmu_reset_needed |= vcpu->shadow_efer != sregs->efer;
1386 #ifdef CONFIG_X86_64
1387 kvm_arch_ops->set_efer(vcpu, sregs->efer);
1388 #endif
1389 vcpu->apic_base = sregs->apic_base;
1390
1391 mmu_reset_needed |= vcpu->cr0 != sregs->cr0;
1392 kvm_arch_ops->set_cr0_no_modeswitch(vcpu, sregs->cr0);
1393
1394 mmu_reset_needed |= vcpu->cr4 != sregs->cr4;
1395 kvm_arch_ops->set_cr4(vcpu, sregs->cr4);
1396
1397 if (mmu_reset_needed)
1398 kvm_mmu_reset_context(vcpu);
1399
1400 memcpy(vcpu->irq_pending, sregs->interrupt_bitmap,
1401 sizeof vcpu->irq_pending);
1402 vcpu->irq_summary = 0;
1403 for (i = 0; i < NR_IRQ_WORDS; ++i)
1404 if (vcpu->irq_pending[i])
1405 __set_bit(i, &vcpu->irq_summary);
1406
1407 vcpu_put(vcpu);
1408
1409 return 0;
1410 }
1411
1412 /*
1413 * List of msr numbers which we expose to userspace through KVM_GET_MSRS
1414 * and KVM_SET_MSRS, and KVM_GET_MSR_INDEX_LIST.
1415 */
1416 static u32 msrs_to_save[] = {
1417 MSR_IA32_SYSENTER_CS, MSR_IA32_SYSENTER_ESP, MSR_IA32_SYSENTER_EIP,
1418 MSR_K6_STAR,
1419 #ifdef CONFIG_X86_64
1420 MSR_CSTAR, MSR_KERNEL_GS_BASE, MSR_SYSCALL_MASK, MSR_LSTAR,
1421 #endif
1422 MSR_IA32_TIME_STAMP_COUNTER,
1423 };
1424
1425
1426 /*
1427 * Adapt set_msr() to msr_io()'s calling convention
1428 */
1429 static int do_set_msr(struct kvm_vcpu *vcpu, unsigned index, u64 *data)
1430 {
1431 return set_msr(vcpu, index, *data);
1432 }
1433
1434 /*
1435 * Read or write a bunch of msrs. All parameters are kernel addresses.
1436 *
1437 * @return number of msrs set successfully.
1438 */
1439 static int __msr_io(struct kvm *kvm, struct kvm_msrs *msrs,
1440 struct kvm_msr_entry *entries,
1441 int (*do_msr)(struct kvm_vcpu *vcpu,
1442 unsigned index, u64 *data))
1443 {
1444 struct kvm_vcpu *vcpu;
1445 int i;
1446
1447 if (msrs->vcpu < 0 || msrs->vcpu >= KVM_MAX_VCPUS)
1448 return -EINVAL;
1449
1450 vcpu = vcpu_load(kvm, msrs->vcpu);
1451 if (!vcpu)
1452 return -ENOENT;
1453
1454 for (i = 0; i < msrs->nmsrs; ++i)
1455 if (do_msr(vcpu, entries[i].index, &entries[i].data))
1456 break;
1457
1458 vcpu_put(vcpu);
1459
1460 return i;
1461 }
1462
1463 /*
1464 * Read or write a bunch of msrs. Parameters are user addresses.
1465 *
1466 * @return number of msrs set successfully.
1467 */
1468 static int msr_io(struct kvm *kvm, struct kvm_msrs __user *user_msrs,
1469 int (*do_msr)(struct kvm_vcpu *vcpu,
1470 unsigned index, u64 *data),
1471 int writeback)
1472 {
1473 struct kvm_msrs msrs;
1474 struct kvm_msr_entry *entries;
1475 int r, n;
1476 unsigned size;
1477
1478 r = -EFAULT;
1479 if (copy_from_user(&msrs, user_msrs, sizeof msrs))
1480 goto out;
1481
1482 r = -E2BIG;
1483 if (msrs.nmsrs >= MAX_IO_MSRS)
1484 goto out;
1485
1486 r = -ENOMEM;
1487 size = sizeof(struct kvm_msr_entry) * msrs.nmsrs;
1488 entries = vmalloc(size);
1489 if (!entries)
1490 goto out;
1491
1492 r = -EFAULT;
1493 if (copy_from_user(entries, user_msrs->entries, size))
1494 goto out_free;
1495
1496 r = n = __msr_io(kvm, &msrs, entries, do_msr);
1497 if (r < 0)
1498 goto out_free;
1499
1500 r = -EFAULT;
1501 if (writeback && copy_to_user(user_msrs->entries, entries, size))
1502 goto out_free;
1503
1504 r = n;
1505
1506 out_free:
1507 vfree(entries);
1508 out:
1509 return r;
1510 }
1511
1512 /*
1513 * Translate a guest virtual address to a guest physical address.
1514 */
1515 static int kvm_dev_ioctl_translate(struct kvm *kvm, struct kvm_translation *tr)
1516 {
1517 unsigned long vaddr = tr->linear_address;
1518 struct kvm_vcpu *vcpu;
1519 gpa_t gpa;
1520
1521 vcpu = vcpu_load(kvm, tr->vcpu);
1522 if (!vcpu)
1523 return -ENOENT;
1524 spin_lock(&kvm->lock);
1525 gpa = vcpu->mmu.gva_to_gpa(vcpu, vaddr);
1526 tr->physical_address = gpa;
1527 tr->valid = gpa != UNMAPPED_GVA;
1528 tr->writeable = 1;
1529 tr->usermode = 0;
1530 spin_unlock(&kvm->lock);
1531 vcpu_put(vcpu);
1532
1533 return 0;
1534 }
1535
1536 static int kvm_dev_ioctl_interrupt(struct kvm *kvm, struct kvm_interrupt *irq)
1537 {
1538 struct kvm_vcpu *vcpu;
1539
1540 if (irq->vcpu < 0 || irq->vcpu >= KVM_MAX_VCPUS)
1541 return -EINVAL;
1542 if (irq->irq < 0 || irq->irq >= 256)
1543 return -EINVAL;
1544 vcpu = vcpu_load(kvm, irq->vcpu);
1545 if (!vcpu)
1546 return -ENOENT;
1547
1548 set_bit(irq->irq, vcpu->irq_pending);
1549 set_bit(irq->irq / BITS_PER_LONG, &vcpu->irq_summary);
1550
1551 vcpu_put(vcpu);
1552
1553 return 0;
1554 }
1555
1556 static int kvm_dev_ioctl_debug_guest(struct kvm *kvm,
1557 struct kvm_debug_guest *dbg)
1558 {
1559 struct kvm_vcpu *vcpu;
1560 int r;
1561
1562 if (dbg->vcpu < 0 || dbg->vcpu >= KVM_MAX_VCPUS)
1563 return -EINVAL;
1564 vcpu = vcpu_load(kvm, dbg->vcpu);
1565 if (!vcpu)
1566 return -ENOENT;
1567
1568 r = kvm_arch_ops->set_guest_debug(vcpu, dbg);
1569
1570 vcpu_put(vcpu);
1571
1572 return r;
1573 }
1574
1575 static long kvm_dev_ioctl(struct file *filp,
1576 unsigned int ioctl, unsigned long arg)
1577 {
1578 struct kvm *kvm = filp->private_data;
1579 int r = -EINVAL;
1580
1581 switch (ioctl) {
1582 case KVM_CREATE_VCPU: {
1583 r = kvm_dev_ioctl_create_vcpu(kvm, arg);
1584 if (r)
1585 goto out;
1586 break;
1587 }
1588 case KVM_RUN: {
1589 struct kvm_run kvm_run;
1590
1591 r = -EFAULT;
1592 if (copy_from_user(&kvm_run, (void *)arg, sizeof kvm_run))
1593 goto out;
1594 r = kvm_dev_ioctl_run(kvm, &kvm_run);
1595 if (r < 0)
1596 goto out;
1597 r = -EFAULT;
1598 if (copy_to_user((void *)arg, &kvm_run, sizeof kvm_run))
1599 goto out;
1600 r = 0;
1601 break;
1602 }
1603 case KVM_GET_REGS: {
1604 struct kvm_regs kvm_regs;
1605
1606 r = -EFAULT;
1607 if (copy_from_user(&kvm_regs, (void *)arg, sizeof kvm_regs))
1608 goto out;
1609 r = kvm_dev_ioctl_get_regs(kvm, &kvm_regs);
1610 if (r)
1611 goto out;
1612 r = -EFAULT;
1613 if (copy_to_user((void *)arg, &kvm_regs, sizeof kvm_regs))
1614 goto out;
1615 r = 0;
1616 break;
1617 }
1618 case KVM_SET_REGS: {
1619 struct kvm_regs kvm_regs;
1620
1621 r = -EFAULT;
1622 if (copy_from_user(&kvm_regs, (void *)arg, sizeof kvm_regs))
1623 goto out;
1624 r = kvm_dev_ioctl_set_regs(kvm, &kvm_regs);
1625 if (r)
1626 goto out;
1627 r = 0;
1628 break;
1629 }
1630 case KVM_GET_SREGS: {
1631 struct kvm_sregs kvm_sregs;
1632
1633 r = -EFAULT;
1634 if (copy_from_user(&kvm_sregs, (void *)arg, sizeof kvm_sregs))
1635 goto out;
1636 r = kvm_dev_ioctl_get_sregs(kvm, &kvm_sregs);
1637 if (r)
1638 goto out;
1639 r = -EFAULT;
1640 if (copy_to_user((void *)arg, &kvm_sregs, sizeof kvm_sregs))
1641 goto out;
1642 r = 0;
1643 break;
1644 }
1645 case KVM_SET_SREGS: {
1646 struct kvm_sregs kvm_sregs;
1647
1648 r = -EFAULT;
1649 if (copy_from_user(&kvm_sregs, (void *)arg, sizeof kvm_sregs))
1650 goto out;
1651 r = kvm_dev_ioctl_set_sregs(kvm, &kvm_sregs);
1652 if (r)
1653 goto out;
1654 r = 0;
1655 break;
1656 }
1657 case KVM_TRANSLATE: {
1658 struct kvm_translation tr;
1659
1660 r = -EFAULT;
1661 if (copy_from_user(&tr, (void *)arg, sizeof tr))
1662 goto out;
1663 r = kvm_dev_ioctl_translate(kvm, &tr);
1664 if (r)
1665 goto out;
1666 r = -EFAULT;
1667 if (copy_to_user((void *)arg, &tr, sizeof tr))
1668 goto out;
1669 r = 0;
1670 break;
1671 }
1672 case KVM_INTERRUPT: {
1673 struct kvm_interrupt irq;
1674
1675 r = -EFAULT;
1676 if (copy_from_user(&irq, (void *)arg, sizeof irq))
1677 goto out;
1678 r = kvm_dev_ioctl_interrupt(kvm, &irq);
1679 if (r)
1680 goto out;
1681 r = 0;
1682 break;
1683 }
1684 case KVM_DEBUG_GUEST: {
1685 struct kvm_debug_guest dbg;
1686
1687 r = -EFAULT;
1688 if (copy_from_user(&dbg, (void *)arg, sizeof dbg))
1689 goto out;
1690 r = kvm_dev_ioctl_debug_guest(kvm, &dbg);
1691 if (r)
1692 goto out;
1693 r = 0;
1694 break;
1695 }
1696 case KVM_SET_MEMORY_REGION: {
1697 struct kvm_memory_region kvm_mem;
1698
1699 r = -EFAULT;
1700 if (copy_from_user(&kvm_mem, (void *)arg, sizeof kvm_mem))
1701 goto out;
1702 r = kvm_dev_ioctl_set_memory_region(kvm, &kvm_mem);
1703 if (r)
1704 goto out;
1705 break;
1706 }
1707 case KVM_GET_DIRTY_LOG: {
1708 struct kvm_dirty_log log;
1709
1710 r = -EFAULT;
1711 if (copy_from_user(&log, (void *)arg, sizeof log))
1712 goto out;
1713 r = kvm_dev_ioctl_get_dirty_log(kvm, &log);
1714 if (r)
1715 goto out;
1716 break;
1717 }
1718 case KVM_GET_MSRS:
1719 r = msr_io(kvm, (void __user *)arg, get_msr, 1);
1720 break;
1721 case KVM_SET_MSRS:
1722 r = msr_io(kvm, (void __user *)arg, do_set_msr, 0);
1723 break;
1724 case KVM_GET_MSR_INDEX_LIST: {
1725 struct kvm_msr_list __user *user_msr_list = (void __user *)arg;
1726 struct kvm_msr_list msr_list;
1727 unsigned n;
1728
1729 r = -EFAULT;
1730 if (copy_from_user(&msr_list, user_msr_list, sizeof msr_list))
1731 goto out;
1732 n = msr_list.nmsrs;
1733 msr_list.nmsrs = ARRAY_SIZE(msrs_to_save);
1734 if (copy_to_user(user_msr_list, &msr_list, sizeof msr_list))
1735 goto out;
1736 r = -E2BIG;
1737 if (n < ARRAY_SIZE(msrs_to_save))
1738 goto out;
1739 r = -EFAULT;
1740 if (copy_to_user(user_msr_list->indices, &msrs_to_save,
1741 sizeof msrs_to_save))
1742 goto out;
1743 r = 0;
1744 }
1745 default:
1746 ;
1747 }
1748 out:
1749 return r;
1750 }
1751
1752 static struct page *kvm_dev_nopage(struct vm_area_struct *vma,
1753 unsigned long address,
1754 int *type)
1755 {
1756 struct kvm *kvm = vma->vm_file->private_data;
1757 unsigned long pgoff;
1758 struct kvm_memory_slot *slot;
1759 struct page *page;
1760
1761 *type = VM_FAULT_MINOR;
1762 pgoff = ((address - vma->vm_start) >> PAGE_SHIFT) + vma->vm_pgoff;
1763 slot = gfn_to_memslot(kvm, pgoff);
1764 if (!slot)
1765 return NOPAGE_SIGBUS;
1766 page = gfn_to_page(slot, pgoff);
1767 if (!page)
1768 return NOPAGE_SIGBUS;
1769 get_page(page);
1770 return page;
1771 }
1772
1773 static struct vm_operations_struct kvm_dev_vm_ops = {
1774 .nopage = kvm_dev_nopage,
1775 };
1776
1777 static int kvm_dev_mmap(struct file *file, struct vm_area_struct *vma)
1778 {
1779 vma->vm_ops = &kvm_dev_vm_ops;
1780 return 0;
1781 }
1782
1783 static struct file_operations kvm_chardev_ops = {
1784 .open = kvm_dev_open,
1785 .release = kvm_dev_release,
1786 .unlocked_ioctl = kvm_dev_ioctl,
1787 .compat_ioctl = kvm_dev_ioctl,
1788 .mmap = kvm_dev_mmap,
1789 };
1790
1791 static struct miscdevice kvm_dev = {
1792 MISC_DYNAMIC_MINOR,
1793 "kvm",
1794 &kvm_chardev_ops,
1795 };
1796
1797 static int kvm_reboot(struct notifier_block *notifier, unsigned long val,
1798 void *v)
1799 {
1800 if (val == SYS_RESTART) {
1801 /*
1802 * Some (well, at least mine) BIOSes hang on reboot if
1803 * in vmx root mode.
1804 */
1805 printk(KERN_INFO "kvm: exiting hardware virtualization\n");
1806 on_each_cpu(kvm_arch_ops->hardware_disable, 0, 0, 1);
1807 }
1808 return NOTIFY_OK;
1809 }
1810
1811 static struct notifier_block kvm_reboot_notifier = {
1812 .notifier_call = kvm_reboot,
1813 .priority = 0,
1814 };
1815
1816 static __init void kvm_init_debug(void)
1817 {
1818 struct kvm_stats_debugfs_item *p;
1819
1820 debugfs_dir = debugfs_create_dir("kvm", 0);
1821 for (p = debugfs_entries; p->name; ++p)
1822 p->dentry = debugfs_create_u32(p->name, 0444, debugfs_dir,
1823 p->data);
1824 }
1825
1826 static void kvm_exit_debug(void)
1827 {
1828 struct kvm_stats_debugfs_item *p;
1829
1830 for (p = debugfs_entries; p->name; ++p)
1831 debugfs_remove(p->dentry);
1832 debugfs_remove(debugfs_dir);
1833 }
1834
1835 hpa_t bad_page_address;
1836
1837 int kvm_init_arch(struct kvm_arch_ops *ops, struct module *module)
1838 {
1839 int r;
1840
1841 kvm_arch_ops = ops;
1842
1843 if (!kvm_arch_ops->cpu_has_kvm_support()) {
1844 printk(KERN_ERR "kvm: no hardware support\n");
1845 return -EOPNOTSUPP;
1846 }
1847 if (kvm_arch_ops->disabled_by_bios()) {
1848 printk(KERN_ERR "kvm: disabled by bios\n");
1849 return -EOPNOTSUPP;
1850 }
1851
1852 r = kvm_arch_ops->hardware_setup();
1853 if (r < 0)
1854 return r;
1855
1856 on_each_cpu(kvm_arch_ops->hardware_enable, 0, 0, 1);
1857 register_reboot_notifier(&kvm_reboot_notifier);
1858
1859 kvm_chardev_ops.owner = module;
1860
1861 r = misc_register(&kvm_dev);
1862 if (r) {
1863 printk (KERN_ERR "kvm: misc device register failed\n");
1864 goto out_free;
1865 }
1866
1867 return r;
1868
1869 out_free:
1870 unregister_reboot_notifier(&kvm_reboot_notifier);
1871 on_each_cpu(kvm_arch_ops->hardware_disable, 0, 0, 1);
1872 kvm_arch_ops->hardware_unsetup();
1873 return r;
1874 }
1875
1876 void kvm_exit_arch(void)
1877 {
1878 misc_deregister(&kvm_dev);
1879
1880 unregister_reboot_notifier(&kvm_reboot_notifier);
1881 on_each_cpu(kvm_arch_ops->hardware_disable, 0, 0, 1);
1882 kvm_arch_ops->hardware_unsetup();
1883 }
1884
1885 static __init int kvm_init(void)
1886 {
1887 static struct page *bad_page;
1888 int r = 0;
1889
1890 kvm_init_debug();
1891
1892 if ((bad_page = alloc_page(GFP_KERNEL)) == NULL) {
1893 r = -ENOMEM;
1894 goto out;
1895 }
1896
1897 bad_page_address = page_to_pfn(bad_page) << PAGE_SHIFT;
1898 memset(__va(bad_page_address), 0, PAGE_SIZE);
1899
1900 return r;
1901
1902 out:
1903 kvm_exit_debug();
1904 return r;
1905 }
1906
1907 static __exit void kvm_exit(void)
1908 {
1909 kvm_exit_debug();
1910 __free_page(pfn_to_page(bad_page_address >> PAGE_SHIFT));
1911 }
1912
1913 module_init(kvm_init)
1914 module_exit(kvm_exit)
1915
1916 EXPORT_SYMBOL_GPL(kvm_init_arch);
1917 EXPORT_SYMBOL_GPL(kvm_exit_arch);
Linux kernel - Deliverables
This page took 0.123316 seconds and 6 git commands to generate.