?  search:
summary | log | graphiclog1 | graphiclog2 | commit | commitdiff | tree | refs | edit | fork
blame | history | raw | HEAD
kvm: add MSR based hypercall API
[qemu-kvm/fedora.git] / kvm / kernel / kvm_main.c
blobdd2c581c4b7160fea44c365ed10ee0597c975104
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 #include <linux/sysdev.h>
38 #include <linux/cpu.h>
39
40 #include "x86_emulate.h"
41 #include "segment_descriptor.h"
42
43 #ifdef KVM_DEBUG
44 #include "debug.h"
45 #endif
46
47 MODULE_AUTHOR("Qumranet");
48 MODULE_LICENSE("GPL");
49
50 static spinlock_t kvm_lock = SPIN_LOCK_UNLOCKED;
51 static struct list_head vm_list = LIST_HEAD_INIT(vm_list);
52
53 struct kvm_arch_ops *kvm_arch_ops;
54 struct kvm_stat kvm_stat;
55 EXPORT_SYMBOL_GPL(kvm_stat);
56
57 static struct kvm_stats_debugfs_item {
58 const char *name;
59 u32 *data;
60 struct dentry *dentry;
61 } debugfs_entries[] = {
62 { "pf_fixed", &kvm_stat.pf_fixed },
63 { "pf_guest", &kvm_stat.pf_guest },
64 { "tlb_flush", &kvm_stat.tlb_flush },
65 { "invlpg", &kvm_stat.invlpg },
66 { "exits", &kvm_stat.exits },
67 { "io_exits", &kvm_stat.io_exits },
68 { "mmio_exits", &kvm_stat.mmio_exits },
69 { "signal_exits", &kvm_stat.signal_exits },
70 { "irq_window", &kvm_stat.irq_window_exits },
71 { "halt_exits", &kvm_stat.halt_exits },
72 { "request_irq", &kvm_stat.request_irq_exits },
73 { "irq_exits", &kvm_stat.irq_exits },
74 { 0, 0 }
75 };
76
77 static struct dentry *debugfs_dir;
78
79 #define MAX_IO_MSRS 256
80
81 #define CR0_RESEVED_BITS 0xffffffff1ffaffc0ULL
82 #define LMSW_GUEST_MASK 0x0eULL
83 #define CR4_RESEVED_BITS (~((1ULL << 11) - 1))
84 #define CR8_RESEVED_BITS (~0x0fULL)
85 #define EFER_RESERVED_BITS 0xfffffffffffff2fe
86
87 #ifdef CONFIG_X86_64
88 // LDT or TSS descriptor in the GDT. 16 bytes.
89 struct segment_descriptor_64 {
90 struct segment_descriptor s;
91 u32 base_higher;
92 u32 pad_zero;
93 };
94
95 #endif
96
97 unsigned long segment_base(u16 selector)
98 {
99 struct descriptor_table gdt;
100 struct segment_descriptor *d;
101 unsigned long table_base;
102 typedef unsigned long ul;
103 unsigned long v;
104
105 if (selector == 0)
106 return 0;
107
108 asm ("sgdt %0" : "=m"(gdt));
109 table_base = gdt.base;
110
111 if (selector & 4) { /* from ldt */
112 u16 ldt_selector;
113
114 asm ("sldt %0" : "=g"(ldt_selector));
115 table_base = segment_base(ldt_selector);
116 }
117 d = (struct segment_descriptor *)(table_base + (selector & ~7));
118 v = d->base_low | ((ul)d->base_mid << 16) | ((ul)d->base_high << 24);
119 #ifdef CONFIG_X86_64
120 if (d->system == 0
121 && (d->type == 2 || d->type == 9 || d->type == 11))
122 v |= ((ul)((struct segment_descriptor_64 *)d)->base_higher) << 32;
123 #endif
124 return v;
125 }
126 EXPORT_SYMBOL_GPL(segment_base);
127
128 static inline int valid_vcpu(int n)
129 {
130 return likely(n >= 0 && n < KVM_MAX_VCPUS);
131 }
132
133 int kvm_read_guest(struct kvm_vcpu *vcpu, gva_t addr, unsigned long size,
134 void *dest)
135 {
136 unsigned char *host_buf = dest;
137 unsigned long req_size = size;
138
139 while (size) {
140 hpa_t paddr;
141 unsigned now;
142 unsigned offset;
143 hva_t guest_buf;
144
145 paddr = gva_to_hpa(vcpu, addr);
146
147 if (is_error_hpa(paddr))
148 break;
149
150 guest_buf = (hva_t)kmap_atomic(
151 pfn_to_page(paddr >> PAGE_SHIFT),
152 KM_USER0);
153 offset = addr & ~PAGE_MASK;
154 guest_buf |= offset;
155 now = min(size, PAGE_SIZE - offset);
156 memcpy(host_buf, (void*)guest_buf, now);
157 host_buf += now;
158 addr += now;
159 size -= now;
160 kunmap_atomic((void *)(guest_buf & PAGE_MASK), KM_USER0);
161 }
162 return req_size - size;
163 }
164 EXPORT_SYMBOL_GPL(kvm_read_guest);
165
166 int kvm_write_guest(struct kvm_vcpu *vcpu, gva_t addr, unsigned long size,
167 void *data)
168 {
169 unsigned char *host_buf = data;
170 unsigned long req_size = size;
171
172 while (size) {
173 hpa_t paddr;
174 unsigned now;
175 unsigned offset;
176 hva_t guest_buf;
177
178 paddr = gva_to_hpa(vcpu, addr);
179
180 if (is_error_hpa(paddr))
181 break;
182
183 guest_buf = (hva_t)kmap_atomic(
184 pfn_to_page(paddr >> PAGE_SHIFT), KM_USER0);
185 offset = addr & ~PAGE_MASK;
186 guest_buf |= offset;
187 now = min(size, PAGE_SIZE - offset);
188 memcpy((void*)guest_buf, host_buf, now);
189 host_buf += now;
190 addr += now;
191 size -= now;
192 kunmap_atomic((void *)(guest_buf & PAGE_MASK), KM_USER0);
193 }
194 return req_size - size;
195 }
196 EXPORT_SYMBOL_GPL(kvm_write_guest);
197
198 static int vcpu_slot(struct kvm_vcpu *vcpu)
199 {
200 return vcpu - vcpu->kvm->vcpus;
201 }
202
203 /*
204 * Switches to specified vcpu, until a matching vcpu_put()
205 */
206 static struct kvm_vcpu *vcpu_load(struct kvm *kvm, int vcpu_slot)
207 {
208 struct kvm_vcpu *vcpu = &kvm->vcpus[vcpu_slot];
209
210 mutex_lock(&vcpu->mutex);
211 if (unlikely(!vcpu->vmcs)) {
212 mutex_unlock(&vcpu->mutex);
213 return 0;
214 }
215 return kvm_arch_ops->vcpu_load(vcpu);
216 }
217
218 static void vcpu_put(struct kvm_vcpu *vcpu)
219 {
220 kvm_arch_ops->vcpu_put(vcpu);
221 mutex_unlock(&vcpu->mutex);
222 }
223
224 static int kvm_dev_open(struct inode *inode, struct file *filp)
225 {
226 struct kvm *kvm = kzalloc(sizeof(struct kvm), GFP_KERNEL);
227 int i;
228
229 if (!kvm)
230 return -ENOMEM;
231
232 spin_lock_init(&kvm->lock);
233 INIT_LIST_HEAD(&kvm->active_mmu_pages);
234 for (i = 0; i < KVM_MAX_VCPUS; ++i) {
235 struct kvm_vcpu *vcpu = &kvm->vcpus[i];
236
237 mutex_init(&vcpu->mutex);
238 vcpu->cpu = -1;
239 vcpu->kvm = kvm;
240 vcpu->mmu.root_hpa = INVALID_PAGE;
241 INIT_LIST_HEAD(&vcpu->free_pages);
242 spin_lock(&kvm_lock);
243 list_add(&kvm->vm_list, &vm_list);
244 spin_unlock(&kvm_lock);
245 }
246 filp->private_data = kvm;
247 return 0;
248 }
249
250 /*
251 * Free any memory in @free but not in @dont.
252 */
253 static void kvm_free_physmem_slot(struct kvm_memory_slot *free,
254 struct kvm_memory_slot *dont)
255 {
256 int i;
257
258 if (!dont || free->phys_mem != dont->phys_mem)
259 if (free->phys_mem) {
260 for (i = 0; i < free->npages; ++i)
261 if (free->phys_mem[i])
262 __free_page(free->phys_mem[i]);
263 vfree(free->phys_mem);
264 }
265
266 if (!dont || free->dirty_bitmap != dont->dirty_bitmap)
267 vfree(free->dirty_bitmap);
268
269 free->phys_mem = 0;
270 free->npages = 0;
271 free->dirty_bitmap = 0;
272 }
273
274 static void kvm_free_physmem(struct kvm *kvm)
275 {
276 int i;
277
278 for (i = 0; i < kvm->nmemslots; ++i)
279 kvm_free_physmem_slot(&kvm->memslots[i], 0);
280 }
281
282 static void kvm_free_vcpu(struct kvm_vcpu *vcpu)
283 {
284 if (!vcpu_load(vcpu->kvm, vcpu_slot(vcpu)))
285 return;
286
287 kvm_mmu_destroy(vcpu);
288 vcpu_put(vcpu);
289 kvm_arch_ops->vcpu_free(vcpu);
290 }
291
292 static void kvm_free_vcpus(struct kvm *kvm)
293 {
294 unsigned int i;
295
296 for (i = 0; i < KVM_MAX_VCPUS; ++i)
297 kvm_free_vcpu(&kvm->vcpus[i]);
298 }
299
300 static int kvm_dev_release(struct inode *inode, struct file *filp)
301 {
302 struct kvm *kvm = filp->private_data;
303
304 spin_lock(&kvm_lock);
305 list_del(&kvm->vm_list);
306 spin_unlock(&kvm_lock);
307 kvm_free_vcpus(kvm);
308 kvm_free_physmem(kvm);
309 kfree(kvm);
310 return 0;
311 }
312
313 static void inject_gp(struct kvm_vcpu *vcpu)
314 {
315 kvm_arch_ops->inject_gp(vcpu, 0);
316 }
317
318 /*
319 * Load the pae pdptrs. Return true is they are all valid.
320 */
321 static int load_pdptrs(struct kvm_vcpu *vcpu, unsigned long cr3)
322 {
323 gfn_t pdpt_gfn = cr3 >> PAGE_SHIFT;
324 unsigned offset = ((cr3 & (PAGE_SIZE-1)) >> 5) << 2;
325 int i;
326 u64 pdpte;
327 u64 *pdpt;
328 int ret;
329 struct kvm_memory_slot *memslot;
330
331 spin_lock(&vcpu->kvm->lock);
332 memslot = gfn_to_memslot(vcpu->kvm, pdpt_gfn);
333 /* FIXME: !memslot - emulate? 0xff? */
334 pdpt = kmap_atomic(gfn_to_page(memslot, pdpt_gfn), KM_USER0);
335
336 ret = 1;
337 for (i = 0; i < 4; ++i) {
338 pdpte = pdpt[offset + i];
339 if ((pdpte & 1) && (pdpte & 0xfffffff0000001e6ull)) {
340 ret = 0;
341 goto out;
342 }
343 }
344
345 for (i = 0; i < 4; ++i)
346 vcpu->pdptrs[i] = pdpt[offset + i];
347
348 out:
349 kunmap_atomic(pdpt, KM_USER0);
350 spin_unlock(&vcpu->kvm->lock);
351
352 return ret;
353 }
354
355 void set_cr0(struct kvm_vcpu *vcpu, unsigned long cr0)
356 {
357 if (cr0 & CR0_RESEVED_BITS) {
358 printk(KERN_DEBUG "set_cr0: 0x%lx #GP, reserved bits 0x%lx\n",
359 cr0, vcpu->cr0);
360 inject_gp(vcpu);
361 return;
362 }
363
364 if ((cr0 & CR0_NW_MASK) && !(cr0 & CR0_CD_MASK)) {
365 printk(KERN_DEBUG "set_cr0: #GP, CD == 0 && NW == 1\n");
366 inject_gp(vcpu);
367 return;
368 }
369
370 if ((cr0 & CR0_PG_MASK) && !(cr0 & CR0_PE_MASK)) {
371 printk(KERN_DEBUG "set_cr0: #GP, set PG flag "
372 "and a clear PE flag\n");
373 inject_gp(vcpu);
374 return;
375 }
376
377 if (!is_paging(vcpu) && (cr0 & CR0_PG_MASK)) {
378 #ifdef CONFIG_X86_64
379 if ((vcpu->shadow_efer & EFER_LME)) {
380 int cs_db, cs_l;
381
382 if (!is_pae(vcpu)) {
383 printk(KERN_DEBUG "set_cr0: #GP, start paging "
384 "in long mode while PAE is disabled\n");
385 inject_gp(vcpu);
386 return;
387 }
388 kvm_arch_ops->get_cs_db_l_bits(vcpu, &cs_db, &cs_l);
389 if (cs_l) {
390 printk(KERN_DEBUG "set_cr0: #GP, start paging "
391 "in long mode while CS.L == 1\n");
392 inject_gp(vcpu);
393 return;
394
395 }
396 } else
397 #endif
398 if (is_pae(vcpu) && !load_pdptrs(vcpu, vcpu->cr3)) {
399 printk(KERN_DEBUG "set_cr0: #GP, pdptrs "
400 "reserved bits\n");
401 inject_gp(vcpu);
402 return;
403 }
404
405 }
406
407 kvm_arch_ops->set_cr0(vcpu, cr0);
408 vcpu->cr0 = cr0;
409
410 spin_lock(&vcpu->kvm->lock);
411 kvm_mmu_reset_context(vcpu);
412 spin_unlock(&vcpu->kvm->lock);
413 return;
414 }
415 EXPORT_SYMBOL_GPL(set_cr0);
416
417 void lmsw(struct kvm_vcpu *vcpu, unsigned long msw)
418 {
419 kvm_arch_ops->decache_cr0_cr4_guest_bits(vcpu);
420 set_cr0(vcpu, (vcpu->cr0 & ~0x0ful) | (msw & 0x0f));
421 }
422 EXPORT_SYMBOL_GPL(lmsw);
423
424 void set_cr4(struct kvm_vcpu *vcpu, unsigned long cr4)
425 {
426 if (cr4 & CR4_RESEVED_BITS) {
427 printk(KERN_DEBUG "set_cr4: #GP, reserved bits\n");
428 inject_gp(vcpu);
429 return;
430 }
431
432 if (is_long_mode(vcpu)) {
433 if (!(cr4 & CR4_PAE_MASK)) {
434 printk(KERN_DEBUG "set_cr4: #GP, clearing PAE while "
435 "in long mode\n");
436 inject_gp(vcpu);
437 return;
438 }
439 } else if (is_paging(vcpu) && !is_pae(vcpu) && (cr4 & CR4_PAE_MASK)
440 && !load_pdptrs(vcpu, vcpu->cr3)) {
441 printk(KERN_DEBUG "set_cr4: #GP, pdptrs reserved bits\n");
442 inject_gp(vcpu);
443 }
444
445 if (cr4 & CR4_VMXE_MASK) {
446 printk(KERN_DEBUG "set_cr4: #GP, setting VMXE\n");
447 inject_gp(vcpu);
448 return;
449 }
450 kvm_arch_ops->set_cr4(vcpu, cr4);
451 spin_lock(&vcpu->kvm->lock);
452 kvm_mmu_reset_context(vcpu);
453 spin_unlock(&vcpu->kvm->lock);
454 }
455 EXPORT_SYMBOL_GPL(set_cr4);
456
457 void set_cr3(struct kvm_vcpu *vcpu, unsigned long cr3)
458 {
459 if (is_long_mode(vcpu)) {
460 if (cr3 & CR3_L_MODE_RESEVED_BITS) {
461 printk(KERN_DEBUG "set_cr3: #GP, reserved bits\n");
462 inject_gp(vcpu);
463 return;
464 }
465 } else {
466 if (cr3 & CR3_RESEVED_BITS) {
467 printk(KERN_DEBUG "set_cr3: #GP, reserved bits\n");
468 inject_gp(vcpu);
469 return;
470 }
471 if (is_paging(vcpu) && is_pae(vcpu) &&
472 !load_pdptrs(vcpu, cr3)) {
473 printk(KERN_DEBUG "set_cr3: #GP, pdptrs "
474 "reserved bits\n");
475 inject_gp(vcpu);
476 return;
477 }
478 }
479
480 vcpu->cr3 = cr3;
481 spin_lock(&vcpu->kvm->lock);
482 /*
483 * Does the new cr3 value map to physical memory? (Note, we
484 * catch an invalid cr3 even in real-mode, because it would
485 * cause trouble later on when we turn on paging anyway.)
486 *
487 * A real CPU would silently accept an invalid cr3 and would
488 * attempt to use it - with largely undefined (and often hard
489 * to debug) behavior on the guest side.
490 */
491 if (unlikely(!gfn_to_memslot(vcpu->kvm, cr3 >> PAGE_SHIFT)))
492 inject_gp(vcpu);
493 else
494 vcpu->mmu.new_cr3(vcpu);
495 spin_unlock(&vcpu->kvm->lock);
496 }
497 EXPORT_SYMBOL_GPL(set_cr3);
498
499 void set_cr8(struct kvm_vcpu *vcpu, unsigned long cr8)
500 {
501 if ( cr8 & CR8_RESEVED_BITS) {
502 printk(KERN_DEBUG "set_cr8: #GP, reserved bits 0x%lx\n", cr8);
503 inject_gp(vcpu);
504 return;
505 }
506 vcpu->cr8 = cr8;
507 }
508 EXPORT_SYMBOL_GPL(set_cr8);
509
510 void fx_init(struct kvm_vcpu *vcpu)
511 {
512 struct __attribute__ ((__packed__)) fx_image_s {
513 u16 control; //fcw
514 u16 status; //fsw
515 u16 tag; // ftw
516 u16 opcode; //fop
517 u64 ip; // fpu ip
518 u64 operand;// fpu dp
519 u32 mxcsr;
520 u32 mxcsr_mask;
521
522 } *fx_image;
523
524 fx_save(vcpu->host_fx_image);
525 fpu_init();
526 fx_save(vcpu->guest_fx_image);
527 fx_restore(vcpu->host_fx_image);
528
529 fx_image = (struct fx_image_s *)vcpu->guest_fx_image;
530 fx_image->mxcsr = 0x1f80;
531 memset(vcpu->guest_fx_image + sizeof(struct fx_image_s),
532 0, FX_IMAGE_SIZE - sizeof(struct fx_image_s));
533 }
534 EXPORT_SYMBOL_GPL(fx_init);
535
536 /*
537 * Creates some virtual cpus. Good luck creating more than one.
538 */
539 static int kvm_dev_ioctl_create_vcpu(struct kvm *kvm, int n)
540 {
541 int r;
542 struct kvm_vcpu *vcpu;
543
544 r = -EINVAL;
545 if (!valid_vcpu(n))
546 goto out;
547
548 vcpu = &kvm->vcpus[n];
549
550 mutex_lock(&vcpu->mutex);
551
552 if (vcpu->vmcs) {
553 mutex_unlock(&vcpu->mutex);
554 return -EEXIST;
555 }
556
557 vcpu->host_fx_image = (char*)ALIGN((hva_t)vcpu->fx_buf,
558 FX_IMAGE_ALIGN);
559 vcpu->guest_fx_image = vcpu->host_fx_image + FX_IMAGE_SIZE;
560
561 r = kvm_arch_ops->vcpu_create(vcpu);
562 if (r < 0)
563 goto out_free_vcpus;
564
565 r = kvm_mmu_create(vcpu);
566 if (r < 0)
567 goto out_free_vcpus;
568
569 kvm_arch_ops->vcpu_load(vcpu);
570 r = kvm_mmu_setup(vcpu);
571 if (r >= 0)
572 r = kvm_arch_ops->vcpu_setup(vcpu);
573 vcpu_put(vcpu);
574
575 if (r < 0)
576 goto out_free_vcpus;
577
578 return 0;
579
580 out_free_vcpus:
581 kvm_free_vcpu(vcpu);
582 mutex_unlock(&vcpu->mutex);
583 out:
584 return r;
585 }
586
587 /*
588 * Allocate some memory and give it an address in the guest physical address
589 * space.
590 *
591 * Discontiguous memory is allowed, mostly for framebuffers.
592 */
593 static int kvm_dev_ioctl_set_memory_region(struct kvm *kvm,
594 struct kvm_memory_region *mem)
595 {
596 int r;
597 gfn_t base_gfn;
598 unsigned long npages;
599 unsigned long i;
600 struct kvm_memory_slot *memslot;
601 struct kvm_memory_slot old, new;
602 int memory_config_version;
603
604 r = -EINVAL;
605 /* General sanity checks */
606 if (mem->memory_size & (PAGE_SIZE - 1))
607 goto out;
608 if (mem->guest_phys_addr & (PAGE_SIZE - 1))
609 goto out;
610 if (mem->slot >= KVM_MEMORY_SLOTS)
611 goto out;
612 if (mem->guest_phys_addr + mem->memory_size < mem->guest_phys_addr)
613 goto out;
614
615 memslot = &kvm->memslots[mem->slot];
616 base_gfn = mem->guest_phys_addr >> PAGE_SHIFT;
617 npages = mem->memory_size >> PAGE_SHIFT;
618
619 if (!npages)
620 mem->flags &= ~KVM_MEM_LOG_DIRTY_PAGES;
621
622 raced:
623 spin_lock(&kvm->lock);
624
625 memory_config_version = kvm->memory_config_version;
626 new = old = *memslot;
627
628 new.base_gfn = base_gfn;
629 new.npages = npages;
630 new.flags = mem->flags;
631
632 /* Disallow changing a memory slot's size. */
633 r = -EINVAL;
634 if (npages && old.npages && npages != old.npages)
635 goto out_unlock;
636
637 /* Check for overlaps */
638 r = -EEXIST;
639 for (i = 0; i < KVM_MEMORY_SLOTS; ++i) {
640 struct kvm_memory_slot *s = &kvm->memslots[i];
641
642 if (s == memslot)
643 continue;
644 if (!((base_gfn + npages <= s->base_gfn) ||
645 (base_gfn >= s->base_gfn + s->npages)))
646 goto out_unlock;
647 }
648 /*
649 * Do memory allocations outside lock. memory_config_version will
650 * detect any races.
651 */
652 spin_unlock(&kvm->lock);
653
654 /* Deallocate if slot is being removed */
655 if (!npages)
656 new.phys_mem = 0;
657
658 /* Free page dirty bitmap if unneeded */
659 if (!(new.flags & KVM_MEM_LOG_DIRTY_PAGES))
660 new.dirty_bitmap = 0;
661
662 r = -ENOMEM;
663
664 /* Allocate if a slot is being created */
665 if (npages && !new.phys_mem) {
666 new.phys_mem = vmalloc(npages * sizeof(struct page *));
667
668 if (!new.phys_mem)
669 goto out_free;
670
671 memset(new.phys_mem, 0, npages * sizeof(struct page *));
672 for (i = 0; i < npages; ++i) {
673 new.phys_mem[i] = alloc_page(GFP_HIGHUSER
674 | __GFP_ZERO);
675 if (!new.phys_mem[i])
676 goto out_free;
677 new.phys_mem[i]->private = 0;
678 }
679 }
680
681 /* Allocate page dirty bitmap if needed */
682 if ((new.flags & KVM_MEM_LOG_DIRTY_PAGES) && !new.dirty_bitmap) {
683 unsigned dirty_bytes = ALIGN(npages, BITS_PER_LONG) / 8;
684
685 new.dirty_bitmap = vmalloc(dirty_bytes);
686 if (!new.dirty_bitmap)
687 goto out_free;
688 memset(new.dirty_bitmap, 0, dirty_bytes);
689 }
690
691 spin_lock(&kvm->lock);
692
693 if (memory_config_version != kvm->memory_config_version) {
694 spin_unlock(&kvm->lock);
695 kvm_free_physmem_slot(&new, &old);
696 goto raced;
697 }
698
699 r = -EAGAIN;
700 if (kvm->busy)
701 goto out_unlock;
702
703 if (mem->slot >= kvm->nmemslots)
704 kvm->nmemslots = mem->slot + 1;
705
706 *memslot = new;
707 ++kvm->memory_config_version;
708
709 spin_unlock(&kvm->lock);
710
711 for (i = 0; i < KVM_MAX_VCPUS; ++i) {
712 struct kvm_vcpu *vcpu;
713
714 vcpu = vcpu_load(kvm, i);
715 if (!vcpu)
716 continue;
717 kvm_mmu_reset_context(vcpu);
718 vcpu_put(vcpu);
719 }
720
721 kvm_free_physmem_slot(&old, &new);
722 return 0;
723
724 out_unlock:
725 spin_unlock(&kvm->lock);
726 out_free:
727 kvm_free_physmem_slot(&new, &old);
728 out:
729 return r;
730 }
731
732 static void do_remove_write_access(struct kvm_vcpu *vcpu, int slot)
733 {
734 spin_lock(&vcpu->kvm->lock);
735 kvm_mmu_slot_remove_write_access(vcpu, slot);
736 spin_unlock(&vcpu->kvm->lock);
737 }
738
739 /*
740 * Get (and clear) the dirty memory log for a memory slot.
741 */
742 static int kvm_dev_ioctl_get_dirty_log(struct kvm *kvm,
743 struct kvm_dirty_log *log)
744 {
745 struct kvm_memory_slot *memslot;
746 int r, i;
747 int n;
748 int cleared;
749 unsigned long any = 0;
750
751 spin_lock(&kvm->lock);
752
753 /*
754 * Prevent changes to guest memory configuration even while the lock
755 * is not taken.
756 */
757 ++kvm->busy;
758 spin_unlock(&kvm->lock);
759 r = -EINVAL;
760 if (log->slot >= KVM_MEMORY_SLOTS)
761 goto out;
762
763 memslot = &kvm->memslots[log->slot];
764 r = -ENOENT;
765 if (!memslot->dirty_bitmap)
766 goto out;
767
768 n = ALIGN(memslot->npages, 8) / 8;
769
770 for (i = 0; !any && i < n; ++i)
771 any = memslot->dirty_bitmap[i];
772
773 r = -EFAULT;
774 if (copy_to_user(log->dirty_bitmap, memslot->dirty_bitmap, n))
775 goto out;
776
777 if (any) {
778 cleared = 0;
779 for (i = 0; i < KVM_MAX_VCPUS; ++i) {
780 struct kvm_vcpu *vcpu = vcpu_load(kvm, i);
781
782 if (!vcpu)
783 continue;
784 if (!cleared) {
785 do_remove_write_access(vcpu, log->slot);
786 memset(memslot->dirty_bitmap, 0, n);
787 cleared = 1;
788 }
789 kvm_arch_ops->tlb_flush(vcpu);
790 vcpu_put(vcpu);
791 }
792 }
793
794 r = 0;
795
796 out:
797 spin_lock(&kvm->lock);
798 --kvm->busy;
799 spin_unlock(&kvm->lock);
800 return r;
801 }
802
803 struct kvm_memory_slot *gfn_to_memslot(struct kvm *kvm, gfn_t gfn)
804 {
805 int i;
806
807 for (i = 0; i < kvm->nmemslots; ++i) {
808 struct kvm_memory_slot *memslot = &kvm->memslots[i];
809
810 if (gfn >= memslot->base_gfn
811 && gfn < memslot->base_gfn + memslot->npages)
812 return memslot;
813 }
814 return 0;
815 }
816 EXPORT_SYMBOL_GPL(gfn_to_memslot);
817
818 void mark_page_dirty(struct kvm *kvm, gfn_t gfn)
819 {
820 int i;
821 struct kvm_memory_slot *memslot = 0;
822 unsigned long rel_gfn;
823
824 for (i = 0; i < kvm->nmemslots; ++i) {
825 memslot = &kvm->memslots[i];
826
827 if (gfn >= memslot->base_gfn
828 && gfn < memslot->base_gfn + memslot->npages) {
829
830 if (!memslot || !memslot->dirty_bitmap)
831 return;
832
833 rel_gfn = gfn - memslot->base_gfn;
834
835 /* avoid RMW */
836 if (!test_bit(rel_gfn, memslot->dirty_bitmap))
837 set_bit(rel_gfn, memslot->dirty_bitmap);
838 return;
839 }
840 }
841 }
842
843 static int emulator_read_std(unsigned long addr,
844 unsigned long *val,
845 unsigned int bytes,
846 struct x86_emulate_ctxt *ctxt)
847 {
848 struct kvm_vcpu *vcpu = ctxt->vcpu;
849 void *data = val;
850
851 while (bytes) {
852 gpa_t gpa = vcpu->mmu.gva_to_gpa(vcpu, addr);
853 unsigned offset = addr & (PAGE_SIZE-1);
854 unsigned tocopy = min(bytes, (unsigned)PAGE_SIZE - offset);
855 unsigned long pfn;
856 struct kvm_memory_slot *memslot;
857 void *page;
858
859 if (gpa == UNMAPPED_GVA)
860 return X86EMUL_PROPAGATE_FAULT;
861 pfn = gpa >> PAGE_SHIFT;
862 memslot = gfn_to_memslot(vcpu->kvm, pfn);
863 if (!memslot)
864 return X86EMUL_UNHANDLEABLE;
865 page = kmap_atomic(gfn_to_page(memslot, pfn), KM_USER0);
866
867 memcpy(data, page + offset, tocopy);
868
869 kunmap_atomic(page, KM_USER0);
870
871 bytes -= tocopy;
872 data += tocopy;
873 addr += tocopy;
874 }
875
876 return X86EMUL_CONTINUE;
877 }
878
879 static int emulator_write_std(unsigned long addr,
880 unsigned long val,
881 unsigned int bytes,
882 struct x86_emulate_ctxt *ctxt)
883 {
884 printk(KERN_ERR "emulator_write_std: addr %lx n %d\n",
885 addr, bytes);
886 return X86EMUL_UNHANDLEABLE;
887 }
888
889 static int emulator_read_emulated(unsigned long addr,
890 unsigned long *val,
891 unsigned int bytes,
892 struct x86_emulate_ctxt *ctxt)
893 {
894 struct kvm_vcpu *vcpu = ctxt->vcpu;
895
896 if (vcpu->mmio_read_completed) {
897 memcpy(val, vcpu->mmio_data, bytes);
898 vcpu->mmio_read_completed = 0;
899 return X86EMUL_CONTINUE;
900 } else if (emulator_read_std(addr, val, bytes, ctxt)
901 == X86EMUL_CONTINUE)
902 return X86EMUL_CONTINUE;
903 else {
904 gpa_t gpa = vcpu->mmu.gva_to_gpa(vcpu, addr);
905
906 if (gpa == UNMAPPED_GVA)
907 return X86EMUL_PROPAGATE_FAULT;
908 vcpu->mmio_needed = 1;
909 vcpu->mmio_phys_addr = gpa;
910 vcpu->mmio_size = bytes;
911 vcpu->mmio_is_write = 0;
912
913 return X86EMUL_UNHANDLEABLE;
914 }
915 }
916
917 static int emulator_write_phys(struct kvm_vcpu *vcpu, gpa_t gpa,
918 unsigned long val, int bytes)
919 {
920 struct kvm_memory_slot *m;
921 struct page *page;
922 void *virt;
923
924 if (((gpa + bytes - 1) >> PAGE_SHIFT) != (gpa >> PAGE_SHIFT))
925 return 0;
926 m = gfn_to_memslot(vcpu->kvm, gpa >> PAGE_SHIFT);
927 if (!m)
928 return 0;
929 page = gfn_to_page(m, gpa >> PAGE_SHIFT);
930 kvm_mmu_pre_write(vcpu, gpa, bytes);
931 virt = kmap_atomic(page, KM_USER0);
932 memcpy(virt + offset_in_page(gpa), &val, bytes);
933 kunmap_atomic(virt, KM_USER0);
934 kvm_mmu_post_write(vcpu, gpa, bytes);
935 return 1;
936 }
937
938 static int emulator_write_emulated(unsigned long addr,
939 unsigned long val,
940 unsigned int bytes,
941 struct x86_emulate_ctxt *ctxt)
942 {
943 struct kvm_vcpu *vcpu = ctxt->vcpu;
944 gpa_t gpa = vcpu->mmu.gva_to_gpa(vcpu, addr);
945
946 if (gpa == UNMAPPED_GVA)
947 return X86EMUL_PROPAGATE_FAULT;
948
949 if (emulator_write_phys(vcpu, gpa, val, bytes))
950 return X86EMUL_CONTINUE;
951
952 vcpu->mmio_needed = 1;
953 vcpu->mmio_phys_addr = gpa;
954 vcpu->mmio_size = bytes;
955 vcpu->mmio_is_write = 1;
956 memcpy(vcpu->mmio_data, &val, bytes);
957
958 return X86EMUL_CONTINUE;
959 }
960
961 static int emulator_cmpxchg_emulated(unsigned long addr,
962 unsigned long old,
963 unsigned long new,
964 unsigned int bytes,
965 struct x86_emulate_ctxt *ctxt)
966 {
967 static int reported;
968
969 if (!reported) {
970 reported = 1;
971 printk(KERN_WARNING "kvm: emulating exchange as write\n");
972 }
973 return emulator_write_emulated(addr, new, bytes, ctxt);
974 }
975
976 #ifdef CONFIG_X86_32
977
978 static int emulator_cmpxchg8b_emulated(unsigned long addr,
979 unsigned long old_lo,
980 unsigned long old_hi,
981 unsigned long new_lo,
982 unsigned long new_hi,
983 struct x86_emulate_ctxt *ctxt)
984 {
985 static int reported;
986 int r;
987
988 if (!reported) {
989 reported = 1;
990 printk(KERN_WARNING "kvm: emulating exchange8b as write\n");
991 }
992 r = emulator_write_emulated(addr, new_lo, 4, ctxt);
993 if (r != X86EMUL_CONTINUE)
994 return r;
995 return emulator_write_emulated(addr+4, new_hi, 4, ctxt);
996 }
997
998 #endif
999
1000 static unsigned long get_segment_base(struct kvm_vcpu *vcpu, int seg)
1001 {
1002 return kvm_arch_ops->get_segment_base(vcpu, seg);
1003 }
1004
1005 int emulate_invlpg(struct kvm_vcpu *vcpu, gva_t address)
1006 {
1007 return X86EMUL_CONTINUE;
1008 }
1009
1010 int emulate_clts(struct kvm_vcpu *vcpu)
1011 {
1012 unsigned long cr0;
1013
1014 kvm_arch_ops->decache_cr0_cr4_guest_bits(vcpu);
1015 cr0 = vcpu->cr0 & ~CR0_TS_MASK;
1016 kvm_arch_ops->set_cr0(vcpu, cr0);
1017 return X86EMUL_CONTINUE;
1018 }
1019
1020 int emulator_get_dr(struct x86_emulate_ctxt* ctxt, int dr, unsigned long *dest)
1021 {
1022 struct kvm_vcpu *vcpu = ctxt->vcpu;
1023
1024 switch (dr) {
1025 case 0 ... 3:
1026 *dest = kvm_arch_ops->get_dr(vcpu, dr);
1027 return X86EMUL_CONTINUE;
1028 default:
1029 printk(KERN_DEBUG "%s: unexpected dr %u\n",
1030 __FUNCTION__, dr);
1031 return X86EMUL_UNHANDLEABLE;
1032 }
1033 }
1034
1035 int emulator_set_dr(struct x86_emulate_ctxt *ctxt, int dr, unsigned long value)
1036 {
1037 unsigned long mask = (ctxt->mode == X86EMUL_MODE_PROT64) ? ~0ULL : ~0U;
1038 int exception;
1039
1040 kvm_arch_ops->set_dr(ctxt->vcpu, dr, value & mask, &exception);
1041 if (exception) {
1042 /* FIXME: better handling */
1043 return X86EMUL_UNHANDLEABLE;
1044 }
1045 return X86EMUL_CONTINUE;
1046 }
1047
1048 static void report_emulation_failure(struct x86_emulate_ctxt *ctxt)
1049 {
1050 static int reported;
1051 u8 opcodes[4];
1052 unsigned long rip = ctxt->vcpu->rip;
1053 unsigned long rip_linear;
1054
1055 rip_linear = rip + get_segment_base(ctxt->vcpu, VCPU_SREG_CS);
1056
1057 if (reported)
1058 return;
1059
1060 emulator_read_std(rip_linear, (void *)opcodes, 4, ctxt);
1061
1062 printk(KERN_ERR "emulation failed but !mmio_needed?"
1063 " rip %lx %02x %02x %02x %02x\n",
1064 rip, opcodes[0], opcodes[1], opcodes[2], opcodes[3]);
1065 reported = 1;
1066 }
1067
1068 struct x86_emulate_ops emulate_ops = {
1069 .read_std = emulator_read_std,
1070 .write_std = emulator_write_std,
1071 .read_emulated = emulator_read_emulated,
1072 .write_emulated = emulator_write_emulated,
1073 .cmpxchg_emulated = emulator_cmpxchg_emulated,
1074 #ifdef CONFIG_X86_32
1075 .cmpxchg8b_emulated = emulator_cmpxchg8b_emulated,
1076 #endif
1077 };
1078
1079 int emulate_instruction(struct kvm_vcpu *vcpu,
1080 struct kvm_run *run,
1081 unsigned long cr2,
1082 u16 error_code)
1083 {
1084 struct x86_emulate_ctxt emulate_ctxt;
1085 int r;
1086 int cs_db, cs_l;
1087
1088 kvm_arch_ops->cache_regs(vcpu);
1089
1090 kvm_arch_ops->get_cs_db_l_bits(vcpu, &cs_db, &cs_l);
1091
1092 emulate_ctxt.vcpu = vcpu;
1093 emulate_ctxt.eflags = kvm_arch_ops->get_rflags(vcpu);
1094 emulate_ctxt.cr2 = cr2;
1095 emulate_ctxt.mode = (emulate_ctxt.eflags & X86_EFLAGS_VM)
1096 ? X86EMUL_MODE_REAL : cs_l
1097 ? X86EMUL_MODE_PROT64 : cs_db
1098 ? X86EMUL_MODE_PROT32 : X86EMUL_MODE_PROT16;
1099
1100 if (emulate_ctxt.mode == X86EMUL_MODE_PROT64) {
1101 emulate_ctxt.cs_base = 0;
1102 emulate_ctxt.ds_base = 0;
1103 emulate_ctxt.es_base = 0;
1104 emulate_ctxt.ss_base = 0;
1105 } else {
1106 emulate_ctxt.cs_base = get_segment_base(vcpu, VCPU_SREG_CS);
1107 emulate_ctxt.ds_base = get_segment_base(vcpu, VCPU_SREG_DS);
1108 emulate_ctxt.es_base = get_segment_base(vcpu, VCPU_SREG_ES);
1109 emulate_ctxt.ss_base = get_segment_base(vcpu, VCPU_SREG_SS);
1110 }
1111
1112 emulate_ctxt.gs_base = get_segment_base(vcpu, VCPU_SREG_GS);
1113 emulate_ctxt.fs_base = get_segment_base(vcpu, VCPU_SREG_FS);
1114
1115 vcpu->mmio_is_write = 0;
1116 r = x86_emulate_memop(&emulate_ctxt, &emulate_ops);
1117
1118 if ((r || vcpu->mmio_is_write) && run) {
1119 run->mmio.phys_addr = vcpu->mmio_phys_addr;
1120 memcpy(run->mmio.data, vcpu->mmio_data, 8);
1121 run->mmio.len = vcpu->mmio_size;
1122 run->mmio.is_write = vcpu->mmio_is_write;
1123 }
1124
1125 if (r) {
1126 if (kvm_mmu_unprotect_page_virt(vcpu, cr2))
1127 return EMULATE_DONE;
1128 if (!vcpu->mmio_needed) {
1129 report_emulation_failure(&emulate_ctxt);
1130 return EMULATE_FAIL;
1131 }
1132 return EMULATE_DO_MMIO;
1133 }
1134
1135 kvm_arch_ops->decache_regs(vcpu);
1136 kvm_arch_ops->set_rflags(vcpu, emulate_ctxt.eflags);
1137
1138 if (vcpu->mmio_is_write)
1139 return EMULATE_DO_MMIO;
1140
1141 return EMULATE_DONE;
1142 }
1143 EXPORT_SYMBOL_GPL(emulate_instruction);
1144
1145 static u64 mk_cr_64(u64 curr_cr, u32 new_val)
1146 {
1147 return (curr_cr & ~((1ULL << 32) - 1)) | new_val;
1148 }
1149
1150 void realmode_lgdt(struct kvm_vcpu *vcpu, u16 limit, unsigned long base)
1151 {
1152 struct descriptor_table dt = { limit, base };
1153
1154 kvm_arch_ops->set_gdt(vcpu, &dt);
1155 }
1156
1157 void realmode_lidt(struct kvm_vcpu *vcpu, u16 limit, unsigned long base)
1158 {
1159 struct descriptor_table dt = { limit, base };
1160
1161 kvm_arch_ops->set_idt(vcpu, &dt);
1162 }
1163
1164 void realmode_lmsw(struct kvm_vcpu *vcpu, unsigned long msw,
1165 unsigned long *rflags)
1166 {
1167 lmsw(vcpu, msw);
1168 *rflags = kvm_arch_ops->get_rflags(vcpu);
1169 }
1170
1171 unsigned long realmode_get_cr(struct kvm_vcpu *vcpu, int cr)
1172 {
1173 kvm_arch_ops->decache_cr0_cr4_guest_bits(vcpu);
1174 switch (cr) {
1175 case 0:
1176 return vcpu->cr0;
1177 case 2:
1178 return vcpu->cr2;
1179 case 3:
1180 return vcpu->cr3;
1181 case 4:
1182 return vcpu->cr4;
1183 default:
1184 vcpu_printf(vcpu, "%s: unexpected cr %u\n", __FUNCTION__, cr);
1185 return 0;
1186 }
1187 }
1188
1189 void realmode_set_cr(struct kvm_vcpu *vcpu, int cr, unsigned long val,
1190 unsigned long *rflags)
1191 {
1192 switch (cr) {
1193 case 0:
1194 set_cr0(vcpu, mk_cr_64(vcpu->cr0, val));
1195 *rflags = kvm_arch_ops->get_rflags(vcpu);
1196 break;
1197 case 2:
1198 vcpu->cr2 = val;
1199 break;
1200 case 3:
1201 set_cr3(vcpu, val);
1202 break;
1203 case 4:
1204 set_cr4(vcpu, mk_cr_64(vcpu->cr4, val));
1205 break;
1206 default:
1207 vcpu_printf(vcpu, "%s: unexpected cr %u\n", __FUNCTION__, cr);
1208 }
1209 }
1210
1211 /*
1212 * Register the para guest with the host:
1213 */
1214 static int vcpu_register_para(struct kvm_vcpu *vcpu, gpa_t para_state_gpa)
1215 {
1216 struct kvm_vcpu_para_state *para_state;
1217 hpa_t para_state_hpa, hypercall_hpa;
1218 struct page *para_state_page;
1219 unsigned char *hypercall;
1220 gpa_t hypercall_gpa;
1221
1222 printk(KERN_DEBUG "kvm: guest trying to enter paravirtual mode\n");
1223 printk(KERN_DEBUG ".... para_state_gpa: %08Lx\n", para_state_gpa);
1224
1225 /*
1226 * Needs to be page aligned:
1227 */
1228 if (para_state_gpa != PAGE_ALIGN(para_state_gpa))
1229 goto err_gp;
1230
1231 para_state_hpa = gpa_to_hpa(vcpu, para_state_gpa);
1232 printk(KERN_DEBUG ".... para_state_hpa: %08Lx\n", para_state_hpa);
1233 if (is_error_hpa(para_state_hpa))
1234 goto err_gp;
1235
1236 para_state_page = pfn_to_page(para_state_hpa >> PAGE_SHIFT);
1237 para_state = kmap_atomic(para_state_page, KM_USER0);
1238
1239 printk(KERN_DEBUG ".... guest version: %d\n", para_state->guest_version);
1240 printk(KERN_DEBUG ".... size: %d\n", para_state->size);
1241
1242 para_state->host_version = KVM_PARA_API_VERSION;
1243 /*
1244 * We cannot support guests that try to register themselves
1245 * with a newer API version than the host supports:
1246 */
1247 if (para_state->guest_version > KVM_PARA_API_VERSION) {
1248 para_state->ret = -KVM_EINVAL;
1249 goto err_kunmap_skip;
1250 }
1251
1252 hypercall_gpa = para_state->hypercall_gpa;
1253 hypercall_hpa = gpa_to_hpa(vcpu, hypercall_gpa);
1254 printk(KERN_DEBUG ".... hypercall_hpa: %08Lx\n", hypercall_hpa);
1255 if (is_error_hpa(hypercall_hpa)) {
1256 para_state->ret = -KVM_EINVAL;
1257 goto err_kunmap_skip;
1258 }
1259
1260 printk(KERN_DEBUG "kvm: para guest successfully registered.\n");
1261 vcpu->para_state_page = para_state_page;
1262 vcpu->para_state_gpa = para_state_gpa;
1263 vcpu->hypercall_gpa = hypercall_gpa;
1264
1265 hypercall = kmap_atomic(pfn_to_page(hypercall_hpa >> PAGE_SHIFT),
1266 KM_USER1) + (hypercall_hpa & ~PAGE_MASK);
1267 kvm_arch_ops->patch_hypercall(vcpu, hypercall);
1268 kunmap_atomic(hypercall, KM_USER1);
1269
1270 para_state->ret = 0;
1271 err_kunmap_skip:
1272 kunmap_atomic(para_state, KM_USER0);
1273 return 0;
1274 err_gp:
1275 return 1;
1276 }
1277
1278 int kvm_get_msr_common(struct kvm_vcpu *vcpu, u32 msr, u64 *pdata)
1279 {
1280 u64 data;
1281
1282 switch (msr) {
1283 case 0xc0010010: /* SYSCFG */
1284 case 0xc0010015: /* HWCR */
1285 case MSR_IA32_PLATFORM_ID:
1286 case MSR_IA32_P5_MC_ADDR:
1287 case MSR_IA32_P5_MC_TYPE:
1288 case MSR_IA32_MC0_CTL:
1289 case MSR_IA32_MCG_STATUS:
1290 case MSR_IA32_MCG_CAP:
1291 case MSR_IA32_MC0_MISC:
1292 case MSR_IA32_MC0_MISC+4:
1293 case MSR_IA32_MC0_MISC+8:
1294 case MSR_IA32_MC0_MISC+12:
1295 case MSR_IA32_MC0_MISC+16:
1296 case MSR_IA32_UCODE_REV:
1297 case MSR_IA32_PERF_STATUS:
1298 /* MTRR registers */
1299 case 0xfe:
1300 case 0x200 ... 0x2ff:
1301 data = 0;
1302 break;
1303 case 0xcd: /* fsb frequency */
1304 data = 3;
1305 break;
1306 case MSR_IA32_APICBASE:
1307 data = vcpu->apic_base;
1308 break;
1309 case MSR_IA32_MISC_ENABLE:
1310 data = vcpu->ia32_misc_enable_msr;
1311 break;
1312 #ifdef CONFIG_X86_64
1313 case MSR_EFER:
1314 data = vcpu->shadow_efer;
1315 break;
1316 #endif
1317 default:
1318 printk(KERN_ERR "kvm: unhandled rdmsr: 0x%x\n", msr);
1319 return 1;
1320 }
1321 *pdata = data;
1322 return 0;
1323 }
1324 EXPORT_SYMBOL_GPL(kvm_get_msr_common);
1325
1326 /*
1327 * Reads an msr value (of 'msr_index') into 'pdata'.
1328 * Returns 0 on success, non-0 otherwise.
1329 * Assumes vcpu_load() was already called.
1330 */
1331 static int get_msr(struct kvm_vcpu *vcpu, u32 msr_index, u64 *pdata)
1332 {
1333 return kvm_arch_ops->get_msr(vcpu, msr_index, pdata);
1334 }
1335
1336 #ifdef CONFIG_X86_64
1337
1338 static void set_efer(struct kvm_vcpu *vcpu, u64 efer)
1339 {
1340 if (efer & EFER_RESERVED_BITS) {
1341 printk(KERN_DEBUG "set_efer: 0x%llx #GP, reserved bits\n",
1342 efer);
1343 inject_gp(vcpu);
1344 return;
1345 }
1346
1347 if (is_paging(vcpu)
1348 && (vcpu->shadow_efer & EFER_LME) != (efer & EFER_LME)) {
1349 printk(KERN_DEBUG "set_efer: #GP, change LME while paging\n");
1350 inject_gp(vcpu);
1351 return;
1352 }
1353
1354 kvm_arch_ops->set_efer(vcpu, efer);
1355
1356 efer &= ~EFER_LMA;
1357 efer |= vcpu->shadow_efer & EFER_LMA;
1358
1359 vcpu->shadow_efer = efer;
1360 }
1361
1362 #endif
1363
1364 int kvm_set_msr_common(struct kvm_vcpu *vcpu, u32 msr, u64 data)
1365 {
1366 switch (msr) {
1367 #ifdef CONFIG_X86_64
1368 case MSR_EFER:
1369 set_efer(vcpu, data);
1370 break;
1371 #endif
1372 case MSR_IA32_MC0_STATUS:
1373 printk(KERN_WARNING "%s: MSR_IA32_MC0_STATUS 0x%llx, nop\n",
1374 __FUNCTION__, data);
1375 break;
1376 case MSR_IA32_UCODE_REV:
1377 case MSR_IA32_UCODE_WRITE:
1378 case 0x200 ... 0x2ff: /* MTRRs */
1379 break;
1380 case MSR_IA32_APICBASE:
1381 vcpu->apic_base = data;
1382 break;
1383 case MSR_IA32_MISC_ENABLE:
1384 vcpu->ia32_misc_enable_msr = data;
1385 break;
1386 /*
1387 * This is the 'probe whether the host is KVM' logic:
1388 */
1389 case MSR_KVM_API_MAGIC:
1390 return vcpu_register_para(vcpu, data);
1391
1392 default:
1393 printk(KERN_ERR "kvm: unhandled wrmsr: 0x%x\n", msr);
1394 return 1;
1395 }
1396 return 0;
1397 }
1398 EXPORT_SYMBOL_GPL(kvm_set_msr_common);
1399
1400 /*
1401 * Writes msr value into into the appropriate "register".
1402 * Returns 0 on success, non-0 otherwise.
1403 * Assumes vcpu_load() was already called.
1404 */
1405 static int set_msr(struct kvm_vcpu *vcpu, u32 msr_index, u64 data)
1406 {
1407 return kvm_arch_ops->set_msr(vcpu, msr_index, data);
1408 }
1409
1410 void kvm_resched(struct kvm_vcpu *vcpu)
1411 {
1412 vcpu_put(vcpu);
1413 cond_resched();
1414 /* Cannot fail - no vcpu unplug yet. */
1415 vcpu_load(vcpu->kvm, vcpu_slot(vcpu));
1416 }
1417 EXPORT_SYMBOL_GPL(kvm_resched);
1418
1419 void load_msrs(struct vmx_msr_entry *e, int n)
1420 {
1421 int i;
1422
1423 for (i = 0; i < n; ++i)
1424 wrmsrl(e[i].index, e[i].data);
1425 }
1426 EXPORT_SYMBOL_GPL(load_msrs);
1427
1428 void save_msrs(struct vmx_msr_entry *e, int n)
1429 {
1430 int i;
1431
1432 for (i = 0; i < n; ++i)
1433 rdmsrl(e[i].index, e[i].data);
1434 }
1435 EXPORT_SYMBOL_GPL(save_msrs);
1436
1437 static int kvm_dev_ioctl_run(struct kvm *kvm, struct kvm_run *kvm_run)
1438 {
1439 struct kvm_vcpu *vcpu;
1440 int r;
1441
1442 if (!valid_vcpu(kvm_run->vcpu))
1443 return -EINVAL;
1444
1445 vcpu = vcpu_load(kvm, kvm_run->vcpu);
1446 if (!vcpu)
1447 return -ENOENT;
1448
1449 /* re-sync apic's tpr */
1450 vcpu->cr8 = kvm_run->cr8;
1451
1452 if (kvm_run->emulated) {
1453 kvm_arch_ops->skip_emulated_instruction(vcpu);
1454 kvm_run->emulated = 0;
1455 }
1456
1457 if (kvm_run->mmio_completed) {
1458 memcpy(vcpu->mmio_data, kvm_run->mmio.data, 8);
1459 vcpu->mmio_read_completed = 1;
1460 }
1461
1462 vcpu->mmio_needed = 0;
1463
1464 r = kvm_arch_ops->run(vcpu, kvm_run);
1465
1466 vcpu_put(vcpu);
1467 return r;
1468 }
1469
1470 static int kvm_dev_ioctl_get_regs(struct kvm *kvm, struct kvm_regs *regs)
1471 {
1472 struct kvm_vcpu *vcpu;
1473
1474 if (!valid_vcpu(regs->vcpu))
1475 return -EINVAL;
1476
1477 vcpu = vcpu_load(kvm, regs->vcpu);
1478 if (!vcpu)
1479 return -ENOENT;
1480
1481 kvm_arch_ops->cache_regs(vcpu);
1482
1483 regs->rax = vcpu->regs[VCPU_REGS_RAX];
1484 regs->rbx = vcpu->regs[VCPU_REGS_RBX];
1485 regs->rcx = vcpu->regs[VCPU_REGS_RCX];
1486 regs->rdx = vcpu->regs[VCPU_REGS_RDX];
1487 regs->rsi = vcpu->regs[VCPU_REGS_RSI];
1488 regs->rdi = vcpu->regs[VCPU_REGS_RDI];
1489 regs->rsp = vcpu->regs[VCPU_REGS_RSP];
1490 regs->rbp = vcpu->regs[VCPU_REGS_RBP];
1491 #ifdef CONFIG_X86_64
1492 regs->r8 = vcpu->regs[VCPU_REGS_R8];
1493 regs->r9 = vcpu->regs[VCPU_REGS_R9];
1494 regs->r10 = vcpu->regs[VCPU_REGS_R10];
1495 regs->r11 = vcpu->regs[VCPU_REGS_R11];
1496 regs->r12 = vcpu->regs[VCPU_REGS_R12];
1497 regs->r13 = vcpu->regs[VCPU_REGS_R13];
1498 regs->r14 = vcpu->regs[VCPU_REGS_R14];
1499 regs->r15 = vcpu->regs[VCPU_REGS_R15];
1500 #endif
1501
1502 regs->rip = vcpu->rip;
1503 regs->rflags = kvm_arch_ops->get_rflags(vcpu);
1504
1505 /*
1506 * Don't leak debug flags in case they were set for guest debugging
1507 */
1508 if (vcpu->guest_debug.enabled && vcpu->guest_debug.singlestep)
1509 regs->rflags &= ~(X86_EFLAGS_TF | X86_EFLAGS_RF);
1510
1511 vcpu_put(vcpu);
1512
1513 return 0;
1514 }
1515
1516 static int kvm_dev_ioctl_set_regs(struct kvm *kvm, struct kvm_regs *regs)
1517 {
1518 struct kvm_vcpu *vcpu;
1519
1520 if (!valid_vcpu(regs->vcpu))
1521 return -EINVAL;
1522
1523 vcpu = vcpu_load(kvm, regs->vcpu);
1524 if (!vcpu)
1525 return -ENOENT;
1526
1527 vcpu->regs[VCPU_REGS_RAX] = regs->rax;
1528 vcpu->regs[VCPU_REGS_RBX] = regs->rbx;
1529 vcpu->regs[VCPU_REGS_RCX] = regs->rcx;
1530 vcpu->regs[VCPU_REGS_RDX] = regs->rdx;
1531 vcpu->regs[VCPU_REGS_RSI] = regs->rsi;
1532 vcpu->regs[VCPU_REGS_RDI] = regs->rdi;
1533 vcpu->regs[VCPU_REGS_RSP] = regs->rsp;
1534 vcpu->regs[VCPU_REGS_RBP] = regs->rbp;
1535 #ifdef CONFIG_X86_64
1536 vcpu->regs[VCPU_REGS_R8] = regs->r8;
1537 vcpu->regs[VCPU_REGS_R9] = regs->r9;
1538 vcpu->regs[VCPU_REGS_R10] = regs->r10;
1539 vcpu->regs[VCPU_REGS_R11] = regs->r11;
1540 vcpu->regs[VCPU_REGS_R12] = regs->r12;
1541 vcpu->regs[VCPU_REGS_R13] = regs->r13;
1542 vcpu->regs[VCPU_REGS_R14] = regs->r14;
1543 vcpu->regs[VCPU_REGS_R15] = regs->r15;
1544 #endif
1545
1546 vcpu->rip = regs->rip;
1547 kvm_arch_ops->set_rflags(vcpu, regs->rflags);
1548
1549 kvm_arch_ops->decache_regs(vcpu);
1550
1551 vcpu_put(vcpu);
1552
1553 return 0;
1554 }
1555
1556 static void get_segment(struct kvm_vcpu *vcpu,
1557 struct kvm_segment *var, int seg)
1558 {
1559 return kvm_arch_ops->get_segment(vcpu, var, seg);
1560 }
1561
1562 static int kvm_dev_ioctl_get_sregs(struct kvm *kvm, struct kvm_sregs *sregs)
1563 {
1564 struct kvm_vcpu *vcpu;
1565 struct descriptor_table dt;
1566
1567 if (!valid_vcpu(sregs->vcpu))
1568 return -EINVAL;
1569 vcpu = vcpu_load(kvm, sregs->vcpu);
1570 if (!vcpu)
1571 return -ENOENT;
1572
1573 get_segment(vcpu, &sregs->cs, VCPU_SREG_CS);
1574 get_segment(vcpu, &sregs->ds, VCPU_SREG_DS);
1575 get_segment(vcpu, &sregs->es, VCPU_SREG_ES);
1576 get_segment(vcpu, &sregs->fs, VCPU_SREG_FS);
1577 get_segment(vcpu, &sregs->gs, VCPU_SREG_GS);
1578 get_segment(vcpu, &sregs->ss, VCPU_SREG_SS);
1579
1580 get_segment(vcpu, &sregs->tr, VCPU_SREG_TR);
1581 get_segment(vcpu, &sregs->ldt, VCPU_SREG_LDTR);
1582
1583 kvm_arch_ops->get_idt(vcpu, &dt);
1584 sregs->idt.limit = dt.limit;
1585 sregs->idt.base = dt.base;
1586 kvm_arch_ops->get_gdt(vcpu, &dt);
1587 sregs->gdt.limit = dt.limit;
1588 sregs->gdt.base = dt.base;
1589
1590 kvm_arch_ops->decache_cr0_cr4_guest_bits(vcpu);
1591 sregs->cr0 = vcpu->cr0;
1592 sregs->cr2 = vcpu->cr2;
1593 sregs->cr3 = vcpu->cr3;
1594 sregs->cr4 = vcpu->cr4;
1595 sregs->cr8 = vcpu->cr8;
1596 sregs->efer = vcpu->shadow_efer;
1597 sregs->apic_base = vcpu->apic_base;
1598
1599 memcpy(sregs->interrupt_bitmap, vcpu->irq_pending,
1600 sizeof sregs->interrupt_bitmap);
1601
1602 vcpu_put(vcpu);
1603
1604 return 0;
1605 }
1606
1607 static void set_segment(struct kvm_vcpu *vcpu,
1608 struct kvm_segment *var, int seg)
1609 {
1610 return kvm_arch_ops->set_segment(vcpu, var, seg);
1611 }
1612
1613 static int kvm_dev_ioctl_set_sregs(struct kvm *kvm, struct kvm_sregs *sregs)
1614 {
1615 struct kvm_vcpu *vcpu;
1616 int mmu_reset_needed = 0;
1617 int i;
1618 struct descriptor_table dt;
1619
1620 if (!valid_vcpu(sregs->vcpu))
1621 return -EINVAL;
1622 vcpu = vcpu_load(kvm, sregs->vcpu);
1623 if (!vcpu)
1624 return -ENOENT;
1625
1626 set_segment(vcpu, &sregs->cs, VCPU_SREG_CS);
1627 set_segment(vcpu, &sregs->ds, VCPU_SREG_DS);
1628 set_segment(vcpu, &sregs->es, VCPU_SREG_ES);
1629 set_segment(vcpu, &sregs->fs, VCPU_SREG_FS);
1630 set_segment(vcpu, &sregs->gs, VCPU_SREG_GS);
1631 set_segment(vcpu, &sregs->ss, VCPU_SREG_SS);
1632
1633 set_segment(vcpu, &sregs->tr, VCPU_SREG_TR);
1634 set_segment(vcpu, &sregs->ldt, VCPU_SREG_LDTR);
1635
1636 dt.limit = sregs->idt.limit;
1637 dt.base = sregs->idt.base;
1638 kvm_arch_ops->set_idt(vcpu, &dt);
1639 dt.limit = sregs->gdt.limit;
1640 dt.base = sregs->gdt.base;
1641 kvm_arch_ops->set_gdt(vcpu, &dt);
1642
1643 vcpu->cr2 = sregs->cr2;
1644 mmu_reset_needed |= vcpu->cr3 != sregs->cr3;
1645 vcpu->cr3 = sregs->cr3;
1646
1647 vcpu->cr8 = sregs->cr8;
1648
1649 mmu_reset_needed |= vcpu->shadow_efer != sregs->efer;
1650 #ifdef CONFIG_X86_64
1651 kvm_arch_ops->set_efer(vcpu, sregs->efer);
1652 #endif
1653 vcpu->apic_base = sregs->apic_base;
1654
1655 kvm_arch_ops->decache_cr0_cr4_guest_bits(vcpu);
1656
1657 mmu_reset_needed |= vcpu->cr0 != sregs->cr0;
1658 kvm_arch_ops->set_cr0_no_modeswitch(vcpu, sregs->cr0);
1659
1660 mmu_reset_needed |= vcpu->cr4 != sregs->cr4;
1661 kvm_arch_ops->set_cr4(vcpu, sregs->cr4);
1662 if (!is_long_mode(vcpu) && is_pae(vcpu))
1663 load_pdptrs(vcpu, vcpu->cr3);
1664
1665 if (mmu_reset_needed)
1666 kvm_mmu_reset_context(vcpu);
1667
1668 memcpy(vcpu->irq_pending, sregs->interrupt_bitmap,
1669 sizeof vcpu->irq_pending);
1670 vcpu->irq_summary = 0;
1671 for (i = 0; i < NR_IRQ_WORDS; ++i)
1672 if (vcpu->irq_pending[i])
1673 __set_bit(i, &vcpu->irq_summary);
1674
1675 vcpu_put(vcpu);
1676
1677 return 0;
1678 }
1679
1680 /*
1681 * List of msr numbers which we expose to userspace through KVM_GET_MSRS
1682 * and KVM_SET_MSRS, and KVM_GET_MSR_INDEX_LIST.
1683 *
1684 * This list is modified at module load time to reflect the
1685 * capabilities of the host cpu.
1686 */
1687 static u32 msrs_to_save[] = {
1688 MSR_IA32_SYSENTER_CS, MSR_IA32_SYSENTER_ESP, MSR_IA32_SYSENTER_EIP,
1689 MSR_K6_STAR,
1690 #ifdef CONFIG_X86_64
1691 MSR_CSTAR, MSR_KERNEL_GS_BASE, MSR_SYSCALL_MASK, MSR_LSTAR,
1692 #endif
1693 MSR_IA32_TIME_STAMP_COUNTER,
1694 };
1695
1696 static unsigned num_msrs_to_save;
1697
1698 static u32 emulated_msrs[] = {
1699 MSR_IA32_MISC_ENABLE,
1700 };
1701
1702 static __init void kvm_init_msr_list(void)
1703 {
1704 u32 dummy[2];
1705 unsigned i, j;
1706
1707 for (i = j = 0; i < ARRAY_SIZE(msrs_to_save); i++) {
1708 if (rdmsr_safe(msrs_to_save[i], &dummy[0], &dummy[1]) < 0)
1709 continue;
1710 if (j < i)
1711 msrs_to_save[j] = msrs_to_save[i];
1712 j++;
1713 }
1714 num_msrs_to_save = j;
1715 }
1716
1717 /*
1718 * Adapt set_msr() to msr_io()'s calling convention
1719 */
1720 static int do_set_msr(struct kvm_vcpu *vcpu, unsigned index, u64 *data)
1721 {
1722 return set_msr(vcpu, index, *data);
1723 }
1724
1725 /*
1726 * Read or write a bunch of msrs. All parameters are kernel addresses.
1727 *
1728 * @return number of msrs set successfully.
1729 */
1730 static int __msr_io(struct kvm *kvm, struct kvm_msrs *msrs,
1731 struct kvm_msr_entry *entries,
1732 int (*do_msr)(struct kvm_vcpu *vcpu,
1733 unsigned index, u64 *data))
1734 {
1735 struct kvm_vcpu *vcpu;
1736 int i;
1737
1738 if (!valid_vcpu(msrs->vcpu))
1739 return -EINVAL;
1740
1741 vcpu = vcpu_load(kvm, msrs->vcpu);
1742 if (!vcpu)
1743 return -ENOENT;
1744
1745 for (i = 0; i < msrs->nmsrs; ++i)
1746 if (do_msr(vcpu, entries[i].index, &entries[i].data))
1747 break;
1748
1749 vcpu_put(vcpu);
1750
1751 return i;
1752 }
1753
1754 /*
1755 * Read or write a bunch of msrs. Parameters are user addresses.
1756 *
1757 * @return number of msrs set successfully.
1758 */
1759 static int msr_io(struct kvm *kvm, struct kvm_msrs __user *user_msrs,
1760 int (*do_msr)(struct kvm_vcpu *vcpu,
1761 unsigned index, u64 *data),
1762 int writeback)
1763 {
1764 struct kvm_msrs msrs;
1765 struct kvm_msr_entry *entries;
1766 int r, n;
1767 unsigned size;
1768
1769 r = -EFAULT;
1770 if (copy_from_user(&msrs, user_msrs, sizeof msrs))
1771 goto out;
1772
1773 r = -E2BIG;
1774 if (msrs.nmsrs >= MAX_IO_MSRS)
1775 goto out;
1776
1777 r = -ENOMEM;
1778 size = sizeof(struct kvm_msr_entry) * msrs.nmsrs;
1779 entries = vmalloc(size);
1780 if (!entries)
1781 goto out;
1782
1783 r = -EFAULT;
1784 if (copy_from_user(entries, user_msrs->entries, size))
1785 goto out_free;
1786
1787 r = n = __msr_io(kvm, &msrs, entries, do_msr);
1788 if (r < 0)
1789 goto out_free;
1790
1791 r = -EFAULT;
1792 if (writeback && copy_to_user(user_msrs->entries, entries, size))
1793 goto out_free;
1794
1795 r = n;
1796
1797 out_free:
1798 vfree(entries);
1799 out:
1800 return r;
1801 }
1802
1803 #ifdef KVM_DEBUG
1804 static int kvm_dev_ioctl_dump_vcpu(struct kvm *kvm, int vcpu_slot)
1805 {
1806 struct kvm_vcpu *vcpu;
1807
1808 if (vcpu_slot < 0 || vcpu_slot >= KVM_MAX_VCPUS)
1809 return -EINVAL;
1810 vcpu = vcpu_load(kvm, vcpu_slot);
1811 if (!vcpu)
1812 return -ENOENT;
1813 vcpu_dump(vcpu);
1814 vcpu_put(vcpu);
1815 return 0;
1816 }
1817 #endif
1818
1819 /*
1820 * Translate a guest virtual address to a guest physical address.
1821 */
1822 static int kvm_dev_ioctl_translate(struct kvm *kvm, struct kvm_translation *tr)
1823 {
1824 unsigned long vaddr = tr->linear_address;
1825 struct kvm_vcpu *vcpu;
1826 gpa_t gpa;
1827
1828 vcpu = vcpu_load(kvm, tr->vcpu);
1829 if (!vcpu)
1830 return -ENOENT;
1831 spin_lock(&kvm->lock);
1832 gpa = vcpu->mmu.gva_to_gpa(vcpu, vaddr);
1833 tr->physical_address = gpa;
1834 tr->valid = gpa != UNMAPPED_GVA;
1835 tr->writeable = 1;
1836 tr->usermode = 0;
1837 spin_unlock(&kvm->lock);
1838 vcpu_put(vcpu);
1839
1840 return 0;
1841 }
1842
1843 static int kvm_dev_ioctl_interrupt(struct kvm *kvm, struct kvm_interrupt *irq)
1844 {
1845 struct kvm_vcpu *vcpu;
1846
1847 if (!valid_vcpu(irq->vcpu))
1848 return -EINVAL;
1849 if (irq->irq < 0 || irq->irq >= 256)
1850 return -EINVAL;
1851 vcpu = vcpu_load(kvm, irq->vcpu);
1852 if (!vcpu)
1853 return -ENOENT;
1854
1855 set_bit(irq->irq, vcpu->irq_pending);
1856 set_bit(irq->irq / BITS_PER_LONG, &vcpu->irq_summary);
1857
1858 vcpu_put(vcpu);
1859
1860 return 0;
1861 }
1862
1863 static int kvm_dev_ioctl_debug_guest(struct kvm *kvm,
1864 struct kvm_debug_guest *dbg)
1865 {
1866 struct kvm_vcpu *vcpu;
1867 int r;
1868
1869 if (!valid_vcpu(dbg->vcpu))
1870 return -EINVAL;
1871 vcpu = vcpu_load(kvm, dbg->vcpu);
1872 if (!vcpu)
1873 return -ENOENT;
1874
1875 r = kvm_arch_ops->set_guest_debug(vcpu, dbg);
1876
1877 vcpu_put(vcpu);
1878
1879 return r;
1880 }
1881
1882 static long kvm_dev_ioctl(struct file *filp,
1883 unsigned int ioctl, unsigned long arg)
1884 {
1885 struct kvm *kvm = filp->private_data;
1886 int r = -EINVAL;
1887
1888 switch (ioctl) {
1889 case KVM_GET_API_VERSION:
1890 r = KVM_API_VERSION;
1891 break;
1892 case KVM_CREATE_VCPU:
1893 r = kvm_dev_ioctl_create_vcpu(kvm, arg);
1894 if (r)
1895 goto out;
1896 break;
1897 case KVM_RUN: {
1898 struct kvm_run kvm_run;
1899
1900 r = -EFAULT;
1901 if (copy_from_user(&kvm_run, (void *)arg, sizeof kvm_run))
1902 goto out;
1903 r = kvm_dev_ioctl_run(kvm, &kvm_run);
1904 if (r < 0 && r != -EINTR)
1905 goto out;
1906 if (copy_to_user((void *)arg, &kvm_run, sizeof kvm_run)) {
1907 r = -EFAULT;
1908 goto out;
1909 }
1910 break;
1911 }
1912 case KVM_GET_REGS: {
1913 struct kvm_regs kvm_regs;
1914
1915 r = -EFAULT;
1916 if (copy_from_user(&kvm_regs, (void *)arg, sizeof kvm_regs))
1917 goto out;
1918 r = kvm_dev_ioctl_get_regs(kvm, &kvm_regs);
1919 if (r)
1920 goto out;
1921 r = -EFAULT;
1922 if (copy_to_user((void *)arg, &kvm_regs, sizeof kvm_regs))
1923 goto out;
1924 r = 0;
1925 break;
1926 }
1927 case KVM_SET_REGS: {
1928 struct kvm_regs kvm_regs;
1929
1930 r = -EFAULT;
1931 if (copy_from_user(&kvm_regs, (void *)arg, sizeof kvm_regs))
1932 goto out;
1933 r = kvm_dev_ioctl_set_regs(kvm, &kvm_regs);
1934 if (r)
1935 goto out;
1936 r = 0;
1937 break;
1938 }
1939 case KVM_GET_SREGS: {
1940 struct kvm_sregs kvm_sregs;
1941
1942 r = -EFAULT;
1943 if (copy_from_user(&kvm_sregs, (void *)arg, sizeof kvm_sregs))
1944 goto out;
1945 r = kvm_dev_ioctl_get_sregs(kvm, &kvm_sregs);
1946 if (r)
1947 goto out;
1948 r = -EFAULT;
1949 if (copy_to_user((void *)arg, &kvm_sregs, sizeof kvm_sregs))
1950 goto out;
1951 r = 0;
1952 break;
1953 }
1954 case KVM_SET_SREGS: {
1955 struct kvm_sregs kvm_sregs;
1956
1957 r = -EFAULT;
1958 if (copy_from_user(&kvm_sregs, (void *)arg, sizeof kvm_sregs))
1959 goto out;
1960 r = kvm_dev_ioctl_set_sregs(kvm, &kvm_sregs);
1961 if (r)
1962 goto out;
1963 r = 0;
1964 break;
1965 }
1966 case KVM_TRANSLATE: {
1967 struct kvm_translation tr;
1968
1969 r = -EFAULT;
1970 if (copy_from_user(&tr, (void *)arg, sizeof tr))
1971 goto out;
1972 r = kvm_dev_ioctl_translate(kvm, &tr);
1973 if (r)
1974 goto out;
1975 r = -EFAULT;
1976 if (copy_to_user((void *)arg, &tr, sizeof tr))
1977 goto out;
1978 r = 0;
1979 break;
1980 }
1981 case KVM_INTERRUPT: {
1982 struct kvm_interrupt irq;
1983
1984 r = -EFAULT;
1985 if (copy_from_user(&irq, (void *)arg, sizeof irq))
1986 goto out;
1987 r = kvm_dev_ioctl_interrupt(kvm, &irq);
1988 if (r)
1989 goto out;
1990 r = 0;
1991 break;
1992 }
1993 case KVM_DEBUG_GUEST: {
1994 struct kvm_debug_guest dbg;
1995
1996 r = -EFAULT;
1997 if (copy_from_user(&dbg, (void *)arg, sizeof dbg))
1998 goto out;
1999 r = kvm_dev_ioctl_debug_guest(kvm, &dbg);
2000 if (r)
2001 goto out;
2002 r = 0;
2003 break;
2004 }
2005 case KVM_SET_MEMORY_REGION: {
2006 struct kvm_memory_region kvm_mem;
2007
2008 r = -EFAULT;
2009 if (copy_from_user(&kvm_mem, (void *)arg, sizeof kvm_mem))
2010 goto out;
2011 r = kvm_dev_ioctl_set_memory_region(kvm, &kvm_mem);
2012 if (r)
2013 goto out;
2014 break;
2015 }
2016 case KVM_GET_DIRTY_LOG: {
2017 struct kvm_dirty_log log;
2018
2019 r = -EFAULT;
2020 if (copy_from_user(&log, (void *)arg, sizeof log))
2021 goto out;
2022 r = kvm_dev_ioctl_get_dirty_log(kvm, &log);
2023 if (r)
2024 goto out;
2025 break;
2026 }
2027 case KVM_GET_MSRS:
2028 r = msr_io(kvm, (void __user *)arg, get_msr, 1);
2029 break;
2030 case KVM_SET_MSRS:
2031 r = msr_io(kvm, (void __user *)arg, do_set_msr, 0);
2032 break;
2033 case KVM_GET_MSR_INDEX_LIST: {
2034 struct kvm_msr_list __user *user_msr_list = (void __user *)arg;
2035 struct kvm_msr_list msr_list;
2036 unsigned n;
2037
2038 r = -EFAULT;
2039 if (copy_from_user(&msr_list, user_msr_list, sizeof msr_list))
2040 goto out;
2041 n = msr_list.nmsrs;
2042 msr_list.nmsrs = num_msrs_to_save + ARRAY_SIZE(emulated_msrs);
2043 if (copy_to_user(user_msr_list, &msr_list, sizeof msr_list))
2044 goto out;
2045 r = -E2BIG;
2046 if (n < num_msrs_to_save)
2047 goto out;
2048 r = -EFAULT;
2049 if (copy_to_user(user_msr_list->indices, &msrs_to_save,
2050 num_msrs_to_save * sizeof(u32)))
2051 goto out;
2052 if (copy_to_user(user_msr_list->indices
2053 + num_msrs_to_save * sizeof(u32),
2054 &emulated_msrs,
2055 ARRAY_SIZE(emulated_msrs) * sizeof(u32)))
2056 goto out;
2057 r = 0;
2058 break;
2059 }
2060 #ifdef KVM_DEBUG
2061 case KVM_DUMP_VCPU:
2062 r = kvm_dev_ioctl_dump_vcpu(kvm, arg);
2063 if (r)
2064 goto out;
2065 break;
2066 #endif
2067 default:
2068 ;
2069 }
2070 out:
2071 return r;
2072 }
2073
2074 static struct page *kvm_dev_nopage(struct vm_area_struct *vma,
2075 unsigned long address,
2076 int *type)
2077 {
2078 struct kvm *kvm = vma->vm_file->private_data;
2079 unsigned long pgoff;
2080 struct kvm_memory_slot *slot;
2081 struct page *page;
2082
2083 *type = VM_FAULT_MINOR;
2084 pgoff = ((address - vma->vm_start) >> PAGE_SHIFT) + vma->vm_pgoff;
2085 slot = gfn_to_memslot(kvm, pgoff);
2086 if (!slot)
2087 return NOPAGE_SIGBUS;
2088 page = gfn_to_page(slot, pgoff);
2089 if (!page)
2090 return NOPAGE_SIGBUS;
2091 get_page(page);
2092 return page;
2093 }
2094
2095 static struct vm_operations_struct kvm_dev_vm_ops = {
2096 .nopage = kvm_dev_nopage,
2097 };
2098
2099 static int kvm_dev_mmap(struct file *file, struct vm_area_struct *vma)
2100 {
2101 vma->vm_ops = &kvm_dev_vm_ops;
2102 return 0;
2103 }
2104
2105 static struct file_operations kvm_chardev_ops = {
2106 .open = kvm_dev_open,
2107 .release = kvm_dev_release,
2108 .unlocked_ioctl = kvm_dev_ioctl,
2109 .compat_ioctl = kvm_dev_ioctl,
2110 .mmap = kvm_dev_mmap,
2111 };
2112
2113 static struct miscdevice kvm_dev = {
2114 MISC_DYNAMIC_MINOR,
2115 "kvm",
2116 &kvm_chardev_ops,
2117 };
2118
2119 static int kvm_reboot(struct notifier_block *notifier, unsigned long val,
2120 void *v)
2121 {
2122 if (val == SYS_RESTART) {
2123 /*
2124 * Some (well, at least mine) BIOSes hang on reboot if
2125 * in vmx root mode.
2126 */
2127 printk(KERN_INFO "kvm: exiting hardware virtualization\n");
2128 on_each_cpu(kvm_arch_ops->hardware_disable, 0, 0, 1);
2129 }
2130 return NOTIFY_OK;
2131 }
2132
2133 static struct notifier_block kvm_reboot_notifier = {
2134 .notifier_call = kvm_reboot,
2135 .priority = 0,
2136 };
2137
2138 /*
2139 * Make sure that a cpu that is being hot-unplugged does not have any vcpus
2140 * cached on it.
2141 */
2142 static void decache_vcpus_on_cpu(int cpu)
2143 {
2144 struct kvm *vm;
2145 struct kvm_vcpu *vcpu;
2146 int i;
2147
2148 spin_lock(&kvm_lock);
2149 list_for_each_entry(vm, &vm_list, vm_list)
2150 for (i = 0; i < KVM_MAX_VCPUS; ++i) {
2151 vcpu = &vm->vcpus[i];
2152 /*
2153 * If the vcpu is locked, then it is running on some
2154 * other cpu and therefore it is not cached on the
2155 * cpu in question.
2156 *
2157 * If it's not locked, check the last cpu it executed
2158 * on.
2159 */
2160 if (mutex_trylock(&vcpu->mutex)) {
2161 if (vcpu->cpu == cpu) {
2162 kvm_arch_ops->vcpu_decache(vcpu);
2163 vcpu->cpu = -1;
2164 }
2165 mutex_unlock(&vcpu->mutex);
2166 }
2167 }
2168 spin_unlock(&kvm_lock);
2169 }
2170
2171 static int kvm_cpu_hotplug(struct notifier_block *notifier, unsigned long val,
2172 void *v)
2173 {
2174 int cpu = (long)v;
2175
2176 switch (val) {
2177 case CPU_DOWN_PREPARE:
2178 case CPU_UP_CANCELED:
2179 printk(KERN_INFO "kvm: disabling virtualization on CPU%d\n",
2180 cpu);
2181 decache_vcpus_on_cpu(cpu);
2182 smp_call_function_single(cpu, kvm_arch_ops->hardware_disable,
2183 NULL, 0, 1);
2184 break;
2185 case CPU_ONLINE:
2186 printk(KERN_INFO "kvm: enabling virtualization on CPU%d\n",
2187 cpu);
2188 smp_call_function_single(cpu, kvm_arch_ops->hardware_enable,
2189 NULL, 0, 1);
2190 break;
2191 }
2192 return NOTIFY_OK;
2193 }
2194
2195 static struct notifier_block kvm_cpu_notifier = {
2196 .notifier_call = kvm_cpu_hotplug,
2197 .priority = 20, /* must be > scheduler priority */
2198 };
2199
2200 static __init void kvm_init_debug(void)
2201 {
2202 struct kvm_stats_debugfs_item *p;
2203
2204 debugfs_dir = debugfs_create_dir("kvm", 0);
2205 for (p = debugfs_entries; p->name; ++p)
2206 p->dentry = debugfs_create_u32(p->name, 0444, debugfs_dir,
2207 p->data);
2208 }
2209
2210 static void kvm_exit_debug(void)
2211 {
2212 struct kvm_stats_debugfs_item *p;
2213
2214 for (p = debugfs_entries; p->name; ++p)
2215 debugfs_remove(p->dentry);
2216 debugfs_remove(debugfs_dir);
2217 }
2218
2219 static int kvm_suspend(struct sys_device *dev, pm_message_t state)
2220 {
2221 decache_vcpus_on_cpu(raw_smp_processor_id());
2222 on_each_cpu(kvm_arch_ops->hardware_disable, 0, 0, 1);
2223 return 0;
2224 }
2225
2226 static int kvm_resume(struct sys_device *dev)
2227 {
2228 on_each_cpu(kvm_arch_ops->hardware_enable, 0, 0, 1);
2229 return 0;
2230 }
2231
2232 static struct sysdev_class kvm_sysdev_class = {
2233 set_kset_name("kvm"),
2234 .suspend = kvm_suspend,
2235 .resume = kvm_resume,
2236 };
2237
2238 static struct sys_device kvm_sysdev = {
2239 .id = 0,
2240 .cls = &kvm_sysdev_class,
2241 };
2242
2243 hpa_t bad_page_address;
2244
2245 int kvm_init_arch(struct kvm_arch_ops *ops, struct module *module)
2246 {
2247 int r;
2248
2249 if (kvm_arch_ops) {
2250 printk(KERN_ERR "kvm: already loaded the other module\n");
2251 return -EEXIST;
2252 }
2253
2254 if (!ops->cpu_has_kvm_support()) {
2255 printk(KERN_ERR "kvm: no hardware support\n");
2256 return -EOPNOTSUPP;
2257 }
2258 if (ops->disabled_by_bios()) {
2259 printk(KERN_ERR "kvm: disabled by bios\n");
2260 return -EOPNOTSUPP;
2261 }
2262
2263 kvm_arch_ops = ops;
2264
2265 r = kvm_arch_ops->hardware_setup();
2266 if (r < 0)
2267 return r;
2268
2269 on_each_cpu(kvm_arch_ops->hardware_enable, 0, 0, 1);
2270 r = register_cpu_notifier(&kvm_cpu_notifier);
2271 if (r)
2272 goto out_free_1;
2273 register_reboot_notifier(&kvm_reboot_notifier);
2274
2275 r = sysdev_class_register(&kvm_sysdev_class);
2276 if (r)
2277 goto out_free_2;
2278
2279 r = sysdev_register(&kvm_sysdev);
2280 if (r)
2281 goto out_free_3;
2282
2283 kvm_chardev_ops.owner = module;
2284
2285 r = misc_register(&kvm_dev);
2286 if (r) {
2287 printk (KERN_ERR "kvm: misc device register failed\n");
2288 goto out_free;
2289 }
2290
2291 return r;
2292
2293 out_free:
2294 sysdev_unregister(&kvm_sysdev);
2295 out_free_3:
2296 sysdev_class_unregister(&kvm_sysdev_class);
2297 out_free_2:
2298 unregister_reboot_notifier(&kvm_reboot_notifier);
2299 unregister_cpu_notifier(&kvm_cpu_notifier);
2300 out_free_1:
2301 on_each_cpu(kvm_arch_ops->hardware_disable, 0, 0, 1);
2302 kvm_arch_ops->hardware_unsetup();
2303 return r;
2304 }
2305
2306 void kvm_exit_arch(void)
2307 {
2308 misc_deregister(&kvm_dev);
2309 sysdev_unregister(&kvm_sysdev);
2310 sysdev_class_unregister(&kvm_sysdev_class);
2311 unregister_reboot_notifier(&kvm_reboot_notifier);
2312 unregister_cpu_notifier(&kvm_cpu_notifier);
2313 on_each_cpu(kvm_arch_ops->hardware_disable, 0, 0, 1);
2314 kvm_arch_ops->hardware_unsetup();
2315 kvm_arch_ops = NULL;
2316 }
2317
2318 static __init int kvm_init(void)
2319 {
2320 static struct page *bad_page;
2321 int r = 0;
2322
2323 kvm_init_debug();
2324
2325 kvm_init_msr_list();
2326
2327 if ((bad_page = alloc_page(GFP_KERNEL)) == NULL) {
2328 r = -ENOMEM;
2329 goto out;
2330 }
2331
2332 bad_page_address = page_to_pfn(bad_page) << PAGE_SHIFT;
2333 memset(__va(bad_page_address), 0, PAGE_SIZE);
2334
2335 return r;
2336
2337 out:
2338 kvm_exit_debug();
2339 return r;
2340 }
2341
2342 static __exit void kvm_exit(void)
2343 {
2344 kvm_exit_debug();
2345 __free_page(pfn_to_page(bad_page_address >> PAGE_SHIFT));
2346 }
2347
2348 module_init(kvm_init)
2349 module_exit(kvm_exit)
2350
2351 EXPORT_SYMBOL_GPL(kvm_init_arch);
2352 EXPORT_SYMBOL_GPL(kvm_exit_arch);
Patches shipped by Fedora