2 * Kernel-based Virtual Machine driver for Linux
4 * This module enables machines with Intel VT-x extensions to run virtual
5 * machines without emulation or binary translation.
7 * Copyright (C) 2006 Qumranet, Inc.
10 * Avi Kivity <avi@qumranet.com>
11 * Yaniv Kamay <yaniv@qumranet.com>
13 * This work is licensed under the terms of the GNU GPL, version 2. See
14 * the COPYING file in the top-level directory.
19 #include "x86_emulate.h"
20 #include "segment_descriptor.h"
22 #include <linux/kvm.h>
23 #include <linux/module.h>
24 #include <linux/errno.h>
25 #include <linux/percpu.h>
26 #include <linux/gfp.h>
28 #include <linux/miscdevice.h>
29 #include <linux/vmalloc.h>
30 #include <linux/reboot.h>
31 #include <linux/debugfs.h>
32 #include <linux/highmem.h>
33 #include <linux/file.h>
34 #include <linux/sysdev.h>
35 #include <linux/cpu.h>
36 #include <linux/sched.h>
37 #include <linux/cpumask.h>
38 #include <linux/smp.h>
39 #include <linux/anon_inodes.h>
41 #include <asm/processor.h>
44 #include <asm/uaccess.h>
47 MODULE_AUTHOR("Qumranet");
48 MODULE_LICENSE("GPL");
50 static DEFINE_SPINLOCK(kvm_lock);
51 static LIST_HEAD(vm_list);
53 static cpumask_t cpus_hardware_enabled;
55 struct kvm_arch_ops *kvm_arch_ops;
57 static void hardware_disable(void *ignored);
59 #define STAT_OFFSET(x) offsetof(struct kvm_vcpu, stat.x)
61 static struct kvm_stats_debugfs_item {
64 struct dentry *dentry;
65 } debugfs_entries[] = {
66 { "pf_fixed", STAT_OFFSET(pf_fixed) },
67 { "pf_guest", STAT_OFFSET(pf_guest) },
68 { "tlb_flush", STAT_OFFSET(tlb_flush) },
69 { "invlpg", STAT_OFFSET(invlpg) },
70 { "exits", STAT_OFFSET(exits) },
71 { "io_exits", STAT_OFFSET(io_exits) },
72 { "mmio_exits", STAT_OFFSET(mmio_exits) },
73 { "signal_exits", STAT_OFFSET(signal_exits) },
74 { "irq_window", STAT_OFFSET(irq_window_exits) },
75 { "halt_exits", STAT_OFFSET(halt_exits) },
76 { "request_irq", STAT_OFFSET(request_irq_exits) },
77 { "irq_exits", STAT_OFFSET(irq_exits) },
78 { "light_exits", STAT_OFFSET(light_exits) },
79 { "efer_reload", STAT_OFFSET(efer_reload) },
83 static struct dentry *debugfs_dir;
85 #define MAX_IO_MSRS 256
87 #define CR0_RESEVED_BITS 0xffffffff1ffaffc0ULL
88 #define LMSW_GUEST_MASK 0x0eULL
89 #define CR4_RESEVED_BITS (~((1ULL << 11) - 1))
90 #define CR8_RESEVED_BITS (~0x0fULL)
91 #define EFER_RESERVED_BITS 0xfffffffffffff2fe
94 // LDT or TSS descriptor in the GDT. 16 bytes.
95 struct segment_descriptor_64 {
96 struct segment_descriptor s;
103 static long kvm_vcpu_ioctl(struct file *file, unsigned int ioctl,
106 unsigned long segment_base(u16 selector)
108 struct descriptor_table gdt;
109 struct segment_descriptor *d;
110 unsigned long table_base;
111 typedef unsigned long ul;
117 asm ("sgdt %0" : "=m"(gdt));
118 table_base = gdt.base;
120 if (selector & 4) { /* from ldt */
123 asm ("sldt %0" : "=g"(ldt_selector));
124 table_base = segment_base(ldt_selector);
126 d = (struct segment_descriptor *)(table_base + (selector & ~7));
127 v = d->base_low | ((ul)d->base_mid << 16) | ((ul)d->base_high << 24);
130 && (d->type == 2 || d->type == 9 || d->type == 11))
131 v |= ((ul)((struct segment_descriptor_64 *)d)->base_higher) << 32;
135 EXPORT_SYMBOL_GPL(segment_base);
137 static inline int valid_vcpu(int n)
139 return likely(n >= 0 && n < KVM_MAX_VCPUS);
142 int kvm_read_guest(struct kvm_vcpu *vcpu, gva_t addr, unsigned long size,
145 unsigned char *host_buf = dest;
146 unsigned long req_size = size;
154 paddr = gva_to_hpa(vcpu, addr);
156 if (is_error_hpa(paddr))
159 guest_buf = (hva_t)kmap_atomic(
160 pfn_to_page(paddr >> PAGE_SHIFT),
162 offset = addr & ~PAGE_MASK;
164 now = min(size, PAGE_SIZE - offset);
165 memcpy(host_buf, (void*)guest_buf, now);
169 kunmap_atomic((void *)(guest_buf & PAGE_MASK), KM_USER0);
171 return req_size - size;
173 EXPORT_SYMBOL_GPL(kvm_read_guest);
175 int kvm_write_guest(struct kvm_vcpu *vcpu, gva_t addr, unsigned long size,
178 unsigned char *host_buf = data;
179 unsigned long req_size = size;
188 paddr = gva_to_hpa(vcpu, addr);
190 if (is_error_hpa(paddr))
193 gfn = vcpu->mmu.gva_to_gpa(vcpu, addr) >> PAGE_SHIFT;
194 mark_page_dirty(vcpu->kvm, gfn);
195 guest_buf = (hva_t)kmap_atomic(
196 pfn_to_page(paddr >> PAGE_SHIFT), KM_USER0);
197 offset = addr & ~PAGE_MASK;
199 now = min(size, PAGE_SIZE - offset);
200 memcpy((void*)guest_buf, host_buf, now);
204 kunmap_atomic((void *)(guest_buf & PAGE_MASK), KM_USER0);
206 return req_size - size;
208 EXPORT_SYMBOL_GPL(kvm_write_guest);
210 void kvm_load_guest_fpu(struct kvm_vcpu *vcpu)
212 if (!vcpu->fpu_active || vcpu->guest_fpu_loaded)
215 vcpu->guest_fpu_loaded = 1;
216 fx_save(vcpu->host_fx_image);
217 fx_restore(vcpu->guest_fx_image);
219 EXPORT_SYMBOL_GPL(kvm_load_guest_fpu);
221 void kvm_put_guest_fpu(struct kvm_vcpu *vcpu)
223 if (!vcpu->guest_fpu_loaded)
226 vcpu->guest_fpu_loaded = 0;
227 fx_save(vcpu->guest_fx_image);
228 fx_restore(vcpu->host_fx_image);
230 EXPORT_SYMBOL_GPL(kvm_put_guest_fpu);
233 * Switches to specified vcpu, until a matching vcpu_put()
235 static void vcpu_load(struct kvm_vcpu *vcpu)
237 mutex_lock(&vcpu->mutex);
238 kvm_arch_ops->vcpu_load(vcpu);
241 static void vcpu_put(struct kvm_vcpu *vcpu)
243 kvm_arch_ops->vcpu_put(vcpu);
244 mutex_unlock(&vcpu->mutex);
247 static void ack_flush(void *_completed)
249 atomic_t *completed = _completed;
251 atomic_inc(completed);
254 void kvm_flush_remote_tlbs(struct kvm *kvm)
258 struct kvm_vcpu *vcpu;
261 atomic_set(&completed, 0);
264 for (i = 0; i < kvm->nvcpus; ++i) {
265 vcpu = &kvm->vcpus[i];
266 if (test_and_set_bit(KVM_TLB_FLUSH, &vcpu->requests))
269 if (cpu != -1 && cpu != raw_smp_processor_id())
270 if (!cpu_isset(cpu, cpus)) {
277 * We really want smp_call_function_mask() here. But that's not
278 * available, so ipi all cpus in parallel and wait for them
281 for (cpu = first_cpu(cpus); cpu != NR_CPUS; cpu = next_cpu(cpu, cpus))
282 smp_call_function_single(cpu, ack_flush, &completed, 1, 0);
283 while (atomic_read(&completed) != needed) {
289 static struct kvm *kvm_create_vm(void)
291 struct kvm *kvm = kzalloc(sizeof(struct kvm), GFP_KERNEL);
295 return ERR_PTR(-ENOMEM);
297 kvm_io_bus_init(&kvm->pio_bus);
298 spin_lock_init(&kvm->lock);
299 INIT_LIST_HEAD(&kvm->active_mmu_pages);
300 spin_lock(&kvm_lock);
301 list_add(&kvm->vm_list, &vm_list);
302 spin_unlock(&kvm_lock);
303 kvm_io_bus_init(&kvm->mmio_bus);
304 for (i = 0; i < KVM_MAX_VCPUS; ++i) {
305 struct kvm_vcpu *vcpu = &kvm->vcpus[i];
307 mutex_init(&vcpu->mutex);
310 vcpu->mmu.root_hpa = INVALID_PAGE;
315 static int kvm_dev_open(struct inode *inode, struct file *filp)
321 * Free any memory in @free but not in @dont.
323 static void kvm_free_physmem_slot(struct kvm_memory_slot *free,
324 struct kvm_memory_slot *dont)
328 if (!dont || free->phys_mem != dont->phys_mem)
329 if (free->phys_mem) {
330 for (i = 0; i < free->npages; ++i)
331 if (free->phys_mem[i])
332 __free_page(free->phys_mem[i]);
333 vfree(free->phys_mem);
336 if (!dont || free->dirty_bitmap != dont->dirty_bitmap)
337 vfree(free->dirty_bitmap);
339 free->phys_mem = NULL;
341 free->dirty_bitmap = NULL;
344 static void kvm_free_physmem(struct kvm *kvm)
348 for (i = 0; i < kvm->nmemslots; ++i)
349 kvm_free_physmem_slot(&kvm->memslots[i], NULL);
352 static void free_pio_guest_pages(struct kvm_vcpu *vcpu)
356 for (i = 0; i < 2; ++i)
357 if (vcpu->pio.guest_pages[i]) {
358 __free_page(vcpu->pio.guest_pages[i]);
359 vcpu->pio.guest_pages[i] = NULL;
363 static void kvm_unload_vcpu_mmu(struct kvm_vcpu *vcpu)
369 kvm_mmu_unload(vcpu);
373 static void kvm_free_vcpu(struct kvm_vcpu *vcpu)
379 kvm_mmu_destroy(vcpu);
381 kvm_arch_ops->vcpu_free(vcpu);
382 free_page((unsigned long)vcpu->run);
384 free_page((unsigned long)vcpu->pio_data);
385 vcpu->pio_data = NULL;
386 free_pio_guest_pages(vcpu);
389 static void kvm_free_vcpus(struct kvm *kvm)
394 * Unpin any mmu pages first.
396 for (i = 0; i < KVM_MAX_VCPUS; ++i)
397 kvm_unload_vcpu_mmu(&kvm->vcpus[i]);
398 for (i = 0; i < KVM_MAX_VCPUS; ++i)
399 kvm_free_vcpu(&kvm->vcpus[i]);
402 static int kvm_dev_release(struct inode *inode, struct file *filp)
407 static void kvm_destroy_vm(struct kvm *kvm)
409 spin_lock(&kvm_lock);
410 list_del(&kvm->vm_list);
411 spin_unlock(&kvm_lock);
412 kvm_io_bus_destroy(&kvm->pio_bus);
413 kvm_io_bus_destroy(&kvm->mmio_bus);
415 kvm_free_physmem(kvm);
419 static int kvm_vm_release(struct inode *inode, struct file *filp)
421 struct kvm *kvm = filp->private_data;
427 static void inject_gp(struct kvm_vcpu *vcpu)
429 kvm_arch_ops->inject_gp(vcpu, 0);
433 * Load the pae pdptrs. Return true is they are all valid.
435 static int load_pdptrs(struct kvm_vcpu *vcpu, unsigned long cr3)
437 gfn_t pdpt_gfn = cr3 >> PAGE_SHIFT;
438 unsigned offset = ((cr3 & (PAGE_SIZE-1)) >> 5) << 2;
445 spin_lock(&vcpu->kvm->lock);
446 page = gfn_to_page(vcpu->kvm, pdpt_gfn);
447 /* FIXME: !page - emulate? 0xff? */
448 pdpt = kmap_atomic(page, KM_USER0);
451 for (i = 0; i < 4; ++i) {
452 pdpte = pdpt[offset + i];
453 if ((pdpte & 1) && (pdpte & 0xfffffff0000001e6ull)) {
459 for (i = 0; i < 4; ++i)
460 vcpu->pdptrs[i] = pdpt[offset + i];
463 kunmap_atomic(pdpt, KM_USER0);
464 spin_unlock(&vcpu->kvm->lock);
469 void set_cr0(struct kvm_vcpu *vcpu, unsigned long cr0)
471 if (cr0 & CR0_RESEVED_BITS) {
472 printk(KERN_DEBUG "set_cr0: 0x%lx #GP, reserved bits 0x%lx\n",
478 if ((cr0 & CR0_NW_MASK) && !(cr0 & CR0_CD_MASK)) {
479 printk(KERN_DEBUG "set_cr0: #GP, CD == 0 && NW == 1\n");
484 if ((cr0 & CR0_PG_MASK) && !(cr0 & CR0_PE_MASK)) {
485 printk(KERN_DEBUG "set_cr0: #GP, set PG flag "
486 "and a clear PE flag\n");
491 if (!is_paging(vcpu) && (cr0 & CR0_PG_MASK)) {
493 if ((vcpu->shadow_efer & EFER_LME)) {
497 printk(KERN_DEBUG "set_cr0: #GP, start paging "
498 "in long mode while PAE is disabled\n");
502 kvm_arch_ops->get_cs_db_l_bits(vcpu, &cs_db, &cs_l);
504 printk(KERN_DEBUG "set_cr0: #GP, start paging "
505 "in long mode while CS.L == 1\n");
512 if (is_pae(vcpu) && !load_pdptrs(vcpu, vcpu->cr3)) {
513 printk(KERN_DEBUG "set_cr0: #GP, pdptrs "
521 kvm_arch_ops->set_cr0(vcpu, cr0);
524 spin_lock(&vcpu->kvm->lock);
525 kvm_mmu_reset_context(vcpu);
526 spin_unlock(&vcpu->kvm->lock);
529 EXPORT_SYMBOL_GPL(set_cr0);
531 void lmsw(struct kvm_vcpu *vcpu, unsigned long msw)
533 set_cr0(vcpu, (vcpu->cr0 & ~0x0ful) | (msw & 0x0f));
535 EXPORT_SYMBOL_GPL(lmsw);
537 void set_cr4(struct kvm_vcpu *vcpu, unsigned long cr4)
539 if (cr4 & CR4_RESEVED_BITS) {
540 printk(KERN_DEBUG "set_cr4: #GP, reserved bits\n");
545 if (is_long_mode(vcpu)) {
546 if (!(cr4 & CR4_PAE_MASK)) {
547 printk(KERN_DEBUG "set_cr4: #GP, clearing PAE while "
552 } else if (is_paging(vcpu) && !is_pae(vcpu) && (cr4 & CR4_PAE_MASK)
553 && !load_pdptrs(vcpu, vcpu->cr3)) {
554 printk(KERN_DEBUG "set_cr4: #GP, pdptrs reserved bits\n");
558 if (cr4 & CR4_VMXE_MASK) {
559 printk(KERN_DEBUG "set_cr4: #GP, setting VMXE\n");
563 kvm_arch_ops->set_cr4(vcpu, cr4);
564 spin_lock(&vcpu->kvm->lock);
565 kvm_mmu_reset_context(vcpu);
566 spin_unlock(&vcpu->kvm->lock);
568 EXPORT_SYMBOL_GPL(set_cr4);
570 void set_cr3(struct kvm_vcpu *vcpu, unsigned long cr3)
572 if (is_long_mode(vcpu)) {
573 if (cr3 & CR3_L_MODE_RESEVED_BITS) {
574 printk(KERN_DEBUG "set_cr3: #GP, reserved bits\n");
579 if (cr3 & CR3_RESEVED_BITS) {
580 printk(KERN_DEBUG "set_cr3: #GP, reserved bits\n");
584 if (is_paging(vcpu) && is_pae(vcpu) &&
585 !load_pdptrs(vcpu, cr3)) {
586 printk(KERN_DEBUG "set_cr3: #GP, pdptrs "
594 spin_lock(&vcpu->kvm->lock);
596 * Does the new cr3 value map to physical memory? (Note, we
597 * catch an invalid cr3 even in real-mode, because it would
598 * cause trouble later on when we turn on paging anyway.)
600 * A real CPU would silently accept an invalid cr3 and would
601 * attempt to use it - with largely undefined (and often hard
602 * to debug) behavior on the guest side.
604 if (unlikely(!gfn_to_memslot(vcpu->kvm, cr3 >> PAGE_SHIFT)))
607 vcpu->mmu.new_cr3(vcpu);
608 spin_unlock(&vcpu->kvm->lock);
610 EXPORT_SYMBOL_GPL(set_cr3);
612 void set_cr8(struct kvm_vcpu *vcpu, unsigned long cr8)
614 if ( cr8 & CR8_RESEVED_BITS) {
615 printk(KERN_DEBUG "set_cr8: #GP, reserved bits 0x%lx\n", cr8);
621 EXPORT_SYMBOL_GPL(set_cr8);
623 void fx_init(struct kvm_vcpu *vcpu)
625 struct __attribute__ ((__packed__)) fx_image_s {
631 u64 operand;// fpu dp
637 fx_save(vcpu->host_fx_image);
639 fx_save(vcpu->guest_fx_image);
640 fx_restore(vcpu->host_fx_image);
642 fx_image = (struct fx_image_s *)vcpu->guest_fx_image;
643 fx_image->mxcsr = 0x1f80;
644 memset(vcpu->guest_fx_image + sizeof(struct fx_image_s),
645 0, FX_IMAGE_SIZE - sizeof(struct fx_image_s));
647 EXPORT_SYMBOL_GPL(fx_init);
650 * Allocate some memory and give it an address in the guest physical address
653 * Discontiguous memory is allowed, mostly for framebuffers.
655 static int kvm_vm_ioctl_set_memory_region(struct kvm *kvm,
656 struct kvm_memory_region *mem)
660 unsigned long npages;
662 struct kvm_memory_slot *memslot;
663 struct kvm_memory_slot old, new;
664 int memory_config_version;
667 /* General sanity checks */
668 if (mem->memory_size & (PAGE_SIZE - 1))
670 if (mem->guest_phys_addr & (PAGE_SIZE - 1))
672 if (mem->slot >= KVM_MEMORY_SLOTS)
674 if (mem->guest_phys_addr + mem->memory_size < mem->guest_phys_addr)
677 memslot = &kvm->memslots[mem->slot];
678 base_gfn = mem->guest_phys_addr >> PAGE_SHIFT;
679 npages = mem->memory_size >> PAGE_SHIFT;
682 mem->flags &= ~KVM_MEM_LOG_DIRTY_PAGES;
685 spin_lock(&kvm->lock);
687 memory_config_version = kvm->memory_config_version;
688 new = old = *memslot;
690 new.base_gfn = base_gfn;
692 new.flags = mem->flags;
694 /* Disallow changing a memory slot's size. */
696 if (npages && old.npages && npages != old.npages)
699 /* Check for overlaps */
701 for (i = 0; i < KVM_MEMORY_SLOTS; ++i) {
702 struct kvm_memory_slot *s = &kvm->memslots[i];
706 if (!((base_gfn + npages <= s->base_gfn) ||
707 (base_gfn >= s->base_gfn + s->npages)))
711 * Do memory allocations outside lock. memory_config_version will
714 spin_unlock(&kvm->lock);
716 /* Deallocate if slot is being removed */
720 /* Free page dirty bitmap if unneeded */
721 if (!(new.flags & KVM_MEM_LOG_DIRTY_PAGES))
722 new.dirty_bitmap = NULL;
726 /* Allocate if a slot is being created */
727 if (npages && !new.phys_mem) {
728 new.phys_mem = vmalloc(npages * sizeof(struct page *));
733 memset(new.phys_mem, 0, npages * sizeof(struct page *));
734 for (i = 0; i < npages; ++i) {
735 new.phys_mem[i] = alloc_page(GFP_HIGHUSER
737 if (!new.phys_mem[i])
739 set_page_private(new.phys_mem[i],0);
743 /* Allocate page dirty bitmap if needed */
744 if ((new.flags & KVM_MEM_LOG_DIRTY_PAGES) && !new.dirty_bitmap) {
745 unsigned dirty_bytes = ALIGN(npages, BITS_PER_LONG) / 8;
747 new.dirty_bitmap = vmalloc(dirty_bytes);
748 if (!new.dirty_bitmap)
750 memset(new.dirty_bitmap, 0, dirty_bytes);
753 spin_lock(&kvm->lock);
755 if (memory_config_version != kvm->memory_config_version) {
756 spin_unlock(&kvm->lock);
757 kvm_free_physmem_slot(&new, &old);
765 if (mem->slot >= kvm->nmemslots)
766 kvm->nmemslots = mem->slot + 1;
769 ++kvm->memory_config_version;
771 kvm_mmu_slot_remove_write_access(kvm, mem->slot);
772 kvm_flush_remote_tlbs(kvm);
774 spin_unlock(&kvm->lock);
776 kvm_free_physmem_slot(&old, &new);
780 spin_unlock(&kvm->lock);
782 kvm_free_physmem_slot(&new, &old);
788 * Get (and clear) the dirty memory log for a memory slot.
790 static int kvm_vm_ioctl_get_dirty_log(struct kvm *kvm,
791 struct kvm_dirty_log *log)
793 struct kvm_memory_slot *memslot;
796 unsigned long any = 0;
798 spin_lock(&kvm->lock);
801 * Prevent changes to guest memory configuration even while the lock
805 spin_unlock(&kvm->lock);
807 if (log->slot >= KVM_MEMORY_SLOTS)
810 memslot = &kvm->memslots[log->slot];
812 if (!memslot->dirty_bitmap)
815 n = ALIGN(memslot->npages, BITS_PER_LONG) / 8;
817 for (i = 0; !any && i < n/sizeof(long); ++i)
818 any = memslot->dirty_bitmap[i];
821 if (copy_to_user(log->dirty_bitmap, memslot->dirty_bitmap, n))
824 spin_lock(&kvm->lock);
825 kvm_mmu_slot_remove_write_access(kvm, log->slot);
826 kvm_flush_remote_tlbs(kvm);
827 memset(memslot->dirty_bitmap, 0, n);
828 spin_unlock(&kvm->lock);
833 spin_lock(&kvm->lock);
835 spin_unlock(&kvm->lock);
840 * Set a new alias region. Aliases map a portion of physical memory into
841 * another portion. This is useful for memory windows, for example the PC
844 static int kvm_vm_ioctl_set_memory_alias(struct kvm *kvm,
845 struct kvm_memory_alias *alias)
848 struct kvm_mem_alias *p;
851 /* General sanity checks */
852 if (alias->memory_size & (PAGE_SIZE - 1))
854 if (alias->guest_phys_addr & (PAGE_SIZE - 1))
856 if (alias->slot >= KVM_ALIAS_SLOTS)
858 if (alias->guest_phys_addr + alias->memory_size
859 < alias->guest_phys_addr)
861 if (alias->target_phys_addr + alias->memory_size
862 < alias->target_phys_addr)
865 spin_lock(&kvm->lock);
867 p = &kvm->aliases[alias->slot];
868 p->base_gfn = alias->guest_phys_addr >> PAGE_SHIFT;
869 p->npages = alias->memory_size >> PAGE_SHIFT;
870 p->target_gfn = alias->target_phys_addr >> PAGE_SHIFT;
872 for (n = KVM_ALIAS_SLOTS; n > 0; --n)
873 if (kvm->aliases[n - 1].npages)
877 kvm_mmu_zap_all(kvm);
879 spin_unlock(&kvm->lock);
887 static gfn_t unalias_gfn(struct kvm *kvm, gfn_t gfn)
890 struct kvm_mem_alias *alias;
892 for (i = 0; i < kvm->naliases; ++i) {
893 alias = &kvm->aliases[i];
894 if (gfn >= alias->base_gfn
895 && gfn < alias->base_gfn + alias->npages)
896 return alias->target_gfn + gfn - alias->base_gfn;
901 static struct kvm_memory_slot *__gfn_to_memslot(struct kvm *kvm, gfn_t gfn)
905 for (i = 0; i < kvm->nmemslots; ++i) {
906 struct kvm_memory_slot *memslot = &kvm->memslots[i];
908 if (gfn >= memslot->base_gfn
909 && gfn < memslot->base_gfn + memslot->npages)
915 struct kvm_memory_slot *gfn_to_memslot(struct kvm *kvm, gfn_t gfn)
917 gfn = unalias_gfn(kvm, gfn);
918 return __gfn_to_memslot(kvm, gfn);
921 struct page *gfn_to_page(struct kvm *kvm, gfn_t gfn)
923 struct kvm_memory_slot *slot;
925 gfn = unalias_gfn(kvm, gfn);
926 slot = __gfn_to_memslot(kvm, gfn);
929 return slot->phys_mem[gfn - slot->base_gfn];
931 EXPORT_SYMBOL_GPL(gfn_to_page);
933 void mark_page_dirty(struct kvm *kvm, gfn_t gfn)
936 struct kvm_memory_slot *memslot;
937 unsigned long rel_gfn;
939 for (i = 0; i < kvm->nmemslots; ++i) {
940 memslot = &kvm->memslots[i];
942 if (gfn >= memslot->base_gfn
943 && gfn < memslot->base_gfn + memslot->npages) {
945 if (!memslot->dirty_bitmap)
948 rel_gfn = gfn - memslot->base_gfn;
951 if (!test_bit(rel_gfn, memslot->dirty_bitmap))
952 set_bit(rel_gfn, memslot->dirty_bitmap);
958 static int emulator_read_std(unsigned long addr,
961 struct x86_emulate_ctxt *ctxt)
963 struct kvm_vcpu *vcpu = ctxt->vcpu;
967 gpa_t gpa = vcpu->mmu.gva_to_gpa(vcpu, addr);
968 unsigned offset = addr & (PAGE_SIZE-1);
969 unsigned tocopy = min(bytes, (unsigned)PAGE_SIZE - offset);
974 if (gpa == UNMAPPED_GVA)
975 return X86EMUL_PROPAGATE_FAULT;
976 pfn = gpa >> PAGE_SHIFT;
977 page = gfn_to_page(vcpu->kvm, pfn);
979 return X86EMUL_UNHANDLEABLE;
980 page_virt = kmap_atomic(page, KM_USER0);
982 memcpy(data, page_virt + offset, tocopy);
984 kunmap_atomic(page_virt, KM_USER0);
991 return X86EMUL_CONTINUE;
994 static int emulator_write_std(unsigned long addr,
997 struct x86_emulate_ctxt *ctxt)
999 printk(KERN_ERR "emulator_write_std: addr %lx n %d\n",
1001 return X86EMUL_UNHANDLEABLE;
1004 static struct kvm_io_device *vcpu_find_mmio_dev(struct kvm_vcpu *vcpu,
1008 * Note that its important to have this wrapper function because
1009 * in the very near future we will be checking for MMIOs against
1010 * the LAPIC as well as the general MMIO bus
1012 return kvm_io_bus_find_dev(&vcpu->kvm->mmio_bus, addr);
1015 static struct kvm_io_device *vcpu_find_pio_dev(struct kvm_vcpu *vcpu,
1018 return kvm_io_bus_find_dev(&vcpu->kvm->pio_bus, addr);
1021 static int emulator_read_emulated(unsigned long addr,
1024 struct x86_emulate_ctxt *ctxt)
1026 struct kvm_vcpu *vcpu = ctxt->vcpu;
1027 struct kvm_io_device *mmio_dev;
1030 if (vcpu->mmio_read_completed) {
1031 memcpy(val, vcpu->mmio_data, bytes);
1032 vcpu->mmio_read_completed = 0;
1033 return X86EMUL_CONTINUE;
1034 } else if (emulator_read_std(addr, val, bytes, ctxt)
1035 == X86EMUL_CONTINUE)
1036 return X86EMUL_CONTINUE;
1038 gpa = vcpu->mmu.gva_to_gpa(vcpu, addr);
1039 if (gpa == UNMAPPED_GVA)
1040 return X86EMUL_PROPAGATE_FAULT;
1043 * Is this MMIO handled locally?
1045 mmio_dev = vcpu_find_mmio_dev(vcpu, gpa);
1047 kvm_iodevice_read(mmio_dev, gpa, bytes, val);
1048 return X86EMUL_CONTINUE;
1051 vcpu->mmio_needed = 1;
1052 vcpu->mmio_phys_addr = gpa;
1053 vcpu->mmio_size = bytes;
1054 vcpu->mmio_is_write = 0;
1056 return X86EMUL_UNHANDLEABLE;
1059 static int emulator_write_phys(struct kvm_vcpu *vcpu, gpa_t gpa,
1060 const void *val, int bytes)
1064 unsigned offset = offset_in_page(gpa);
1066 if (((gpa + bytes - 1) >> PAGE_SHIFT) != (gpa >> PAGE_SHIFT))
1068 page = gfn_to_page(vcpu->kvm, gpa >> PAGE_SHIFT);
1071 mark_page_dirty(vcpu->kvm, gpa >> PAGE_SHIFT);
1072 virt = kmap_atomic(page, KM_USER0);
1073 if (memcmp(virt + offset_in_page(gpa), val, bytes)) {
1074 kvm_mmu_pte_write(vcpu, gpa, virt + offset, val, bytes);
1075 memcpy(virt + offset_in_page(gpa), val, bytes);
1077 kunmap_atomic(virt, KM_USER0);
1081 static int emulator_write_emulated_onepage(unsigned long addr,
1084 struct x86_emulate_ctxt *ctxt)
1086 struct kvm_vcpu *vcpu = ctxt->vcpu;
1087 struct kvm_io_device *mmio_dev;
1088 gpa_t gpa = vcpu->mmu.gva_to_gpa(vcpu, addr);
1090 if (gpa == UNMAPPED_GVA) {
1091 kvm_arch_ops->inject_page_fault(vcpu, addr, 2);
1092 return X86EMUL_PROPAGATE_FAULT;
1095 if (emulator_write_phys(vcpu, gpa, val, bytes))
1096 return X86EMUL_CONTINUE;
1099 * Is this MMIO handled locally?
1101 mmio_dev = vcpu_find_mmio_dev(vcpu, gpa);
1103 kvm_iodevice_write(mmio_dev, gpa, bytes, val);
1104 return X86EMUL_CONTINUE;
1107 vcpu->mmio_needed = 1;
1108 vcpu->mmio_phys_addr = gpa;
1109 vcpu->mmio_size = bytes;
1110 vcpu->mmio_is_write = 1;
1111 memcpy(vcpu->mmio_data, val, bytes);
1113 return X86EMUL_CONTINUE;
1116 static int emulator_write_emulated(unsigned long addr,
1119 struct x86_emulate_ctxt *ctxt)
1121 /* Crossing a page boundary? */
1122 if (((addr + bytes - 1) ^ addr) & PAGE_MASK) {
1125 now = -addr & ~PAGE_MASK;
1126 rc = emulator_write_emulated_onepage(addr, val, now, ctxt);
1127 if (rc != X86EMUL_CONTINUE)
1133 return emulator_write_emulated_onepage(addr, val, bytes, ctxt);
1136 static int emulator_cmpxchg_emulated(unsigned long addr,
1140 struct x86_emulate_ctxt *ctxt)
1142 static int reported;
1146 printk(KERN_WARNING "kvm: emulating exchange as write\n");
1148 return emulator_write_emulated(addr, new, bytes, ctxt);
1151 static unsigned long get_segment_base(struct kvm_vcpu *vcpu, int seg)
1153 return kvm_arch_ops->get_segment_base(vcpu, seg);
1156 int emulate_invlpg(struct kvm_vcpu *vcpu, gva_t address)
1158 return X86EMUL_CONTINUE;
1161 int emulate_clts(struct kvm_vcpu *vcpu)
1165 cr0 = vcpu->cr0 & ~CR0_TS_MASK;
1166 kvm_arch_ops->set_cr0(vcpu, cr0);
1167 return X86EMUL_CONTINUE;
1170 int emulator_get_dr(struct x86_emulate_ctxt* ctxt, int dr, unsigned long *dest)
1172 struct kvm_vcpu *vcpu = ctxt->vcpu;
1176 *dest = kvm_arch_ops->get_dr(vcpu, dr);
1177 return X86EMUL_CONTINUE;
1179 printk(KERN_DEBUG "%s: unexpected dr %u\n",
1181 return X86EMUL_UNHANDLEABLE;
1185 int emulator_set_dr(struct x86_emulate_ctxt *ctxt, int dr, unsigned long value)
1187 unsigned long mask = (ctxt->mode == X86EMUL_MODE_PROT64) ? ~0ULL : ~0U;
1190 kvm_arch_ops->set_dr(ctxt->vcpu, dr, value & mask, &exception);
1192 /* FIXME: better handling */
1193 return X86EMUL_UNHANDLEABLE;
1195 return X86EMUL_CONTINUE;
1198 static void report_emulation_failure(struct x86_emulate_ctxt *ctxt)
1200 static int reported;
1202 unsigned long rip = ctxt->vcpu->rip;
1203 unsigned long rip_linear;
1205 rip_linear = rip + get_segment_base(ctxt->vcpu, VCPU_SREG_CS);
1210 emulator_read_std(rip_linear, (void *)opcodes, 4, ctxt);
1212 printk(KERN_ERR "emulation failed but !mmio_needed?"
1213 " rip %lx %02x %02x %02x %02x\n",
1214 rip, opcodes[0], opcodes[1], opcodes[2], opcodes[3]);
1218 struct x86_emulate_ops emulate_ops = {
1219 .read_std = emulator_read_std,
1220 .write_std = emulator_write_std,
1221 .read_emulated = emulator_read_emulated,
1222 .write_emulated = emulator_write_emulated,
1223 .cmpxchg_emulated = emulator_cmpxchg_emulated,
1226 int emulate_instruction(struct kvm_vcpu *vcpu,
1227 struct kvm_run *run,
1231 struct x86_emulate_ctxt emulate_ctxt;
1235 vcpu->mmio_fault_cr2 = cr2;
1236 kvm_arch_ops->cache_regs(vcpu);
1238 kvm_arch_ops->get_cs_db_l_bits(vcpu, &cs_db, &cs_l);
1240 emulate_ctxt.vcpu = vcpu;
1241 emulate_ctxt.eflags = kvm_arch_ops->get_rflags(vcpu);
1242 emulate_ctxt.cr2 = cr2;
1243 emulate_ctxt.mode = (emulate_ctxt.eflags & X86_EFLAGS_VM)
1244 ? X86EMUL_MODE_REAL : cs_l
1245 ? X86EMUL_MODE_PROT64 : cs_db
1246 ? X86EMUL_MODE_PROT32 : X86EMUL_MODE_PROT16;
1248 if (emulate_ctxt.mode == X86EMUL_MODE_PROT64) {
1249 emulate_ctxt.cs_base = 0;
1250 emulate_ctxt.ds_base = 0;
1251 emulate_ctxt.es_base = 0;
1252 emulate_ctxt.ss_base = 0;
1254 emulate_ctxt.cs_base = get_segment_base(vcpu, VCPU_SREG_CS);
1255 emulate_ctxt.ds_base = get_segment_base(vcpu, VCPU_SREG_DS);
1256 emulate_ctxt.es_base = get_segment_base(vcpu, VCPU_SREG_ES);
1257 emulate_ctxt.ss_base = get_segment_base(vcpu, VCPU_SREG_SS);
1260 emulate_ctxt.gs_base = get_segment_base(vcpu, VCPU_SREG_GS);
1261 emulate_ctxt.fs_base = get_segment_base(vcpu, VCPU_SREG_FS);
1263 vcpu->mmio_is_write = 0;
1264 r = x86_emulate_memop(&emulate_ctxt, &emulate_ops);
1266 if ((r || vcpu->mmio_is_write) && run) {
1267 run->mmio.phys_addr = vcpu->mmio_phys_addr;
1268 memcpy(run->mmio.data, vcpu->mmio_data, 8);
1269 run->mmio.len = vcpu->mmio_size;
1270 run->mmio.is_write = vcpu->mmio_is_write;
1274 if (kvm_mmu_unprotect_page_virt(vcpu, cr2))
1275 return EMULATE_DONE;
1276 if (!vcpu->mmio_needed) {
1277 report_emulation_failure(&emulate_ctxt);
1278 return EMULATE_FAIL;
1280 return EMULATE_DO_MMIO;
1283 kvm_arch_ops->decache_regs(vcpu);
1284 kvm_arch_ops->set_rflags(vcpu, emulate_ctxt.eflags);
1286 if (vcpu->mmio_is_write) {
1287 vcpu->mmio_needed = 0;
1288 return EMULATE_DO_MMIO;
1291 return EMULATE_DONE;
1293 EXPORT_SYMBOL_GPL(emulate_instruction);
1295 int kvm_emulate_halt(struct kvm_vcpu *vcpu)
1297 if (vcpu->irq_summary)
1300 vcpu->run->exit_reason = KVM_EXIT_HLT;
1301 ++vcpu->stat.halt_exits;
1304 EXPORT_SYMBOL_GPL(kvm_emulate_halt);
1306 int kvm_hypercall(struct kvm_vcpu *vcpu, struct kvm_run *run)
1308 unsigned long nr, a0, a1, a2, a3, a4, a5, ret;
1310 kvm_arch_ops->cache_regs(vcpu);
1312 #ifdef CONFIG_X86_64
1313 if (is_long_mode(vcpu)) {
1314 nr = vcpu->regs[VCPU_REGS_RAX];
1315 a0 = vcpu->regs[VCPU_REGS_RDI];
1316 a1 = vcpu->regs[VCPU_REGS_RSI];
1317 a2 = vcpu->regs[VCPU_REGS_RDX];
1318 a3 = vcpu->regs[VCPU_REGS_RCX];
1319 a4 = vcpu->regs[VCPU_REGS_R8];
1320 a5 = vcpu->regs[VCPU_REGS_R9];
1324 nr = vcpu->regs[VCPU_REGS_RBX] & -1u;
1325 a0 = vcpu->regs[VCPU_REGS_RAX] & -1u;
1326 a1 = vcpu->regs[VCPU_REGS_RCX] & -1u;
1327 a2 = vcpu->regs[VCPU_REGS_RDX] & -1u;
1328 a3 = vcpu->regs[VCPU_REGS_RSI] & -1u;
1329 a4 = vcpu->regs[VCPU_REGS_RDI] & -1u;
1330 a5 = vcpu->regs[VCPU_REGS_RBP] & -1u;
1334 run->hypercall.args[0] = a0;
1335 run->hypercall.args[1] = a1;
1336 run->hypercall.args[2] = a2;
1337 run->hypercall.args[3] = a3;
1338 run->hypercall.args[4] = a4;
1339 run->hypercall.args[5] = a5;
1340 run->hypercall.ret = ret;
1341 run->hypercall.longmode = is_long_mode(vcpu);
1342 kvm_arch_ops->decache_regs(vcpu);
1345 vcpu->regs[VCPU_REGS_RAX] = ret;
1346 kvm_arch_ops->decache_regs(vcpu);
1349 EXPORT_SYMBOL_GPL(kvm_hypercall);
1351 static u64 mk_cr_64(u64 curr_cr, u32 new_val)
1353 return (curr_cr & ~((1ULL << 32) - 1)) | new_val;
1356 void realmode_lgdt(struct kvm_vcpu *vcpu, u16 limit, unsigned long base)
1358 struct descriptor_table dt = { limit, base };
1360 kvm_arch_ops->set_gdt(vcpu, &dt);
1363 void realmode_lidt(struct kvm_vcpu *vcpu, u16 limit, unsigned long base)
1365 struct descriptor_table dt = { limit, base };
1367 kvm_arch_ops->set_idt(vcpu, &dt);
1370 void realmode_lmsw(struct kvm_vcpu *vcpu, unsigned long msw,
1371 unsigned long *rflags)
1374 *rflags = kvm_arch_ops->get_rflags(vcpu);
1377 unsigned long realmode_get_cr(struct kvm_vcpu *vcpu, int cr)
1379 kvm_arch_ops->decache_cr4_guest_bits(vcpu);
1390 vcpu_printf(vcpu, "%s: unexpected cr %u\n", __FUNCTION__, cr);
1395 void realmode_set_cr(struct kvm_vcpu *vcpu, int cr, unsigned long val,
1396 unsigned long *rflags)
1400 set_cr0(vcpu, mk_cr_64(vcpu->cr0, val));
1401 *rflags = kvm_arch_ops->get_rflags(vcpu);
1410 set_cr4(vcpu, mk_cr_64(vcpu->cr4, val));
1413 vcpu_printf(vcpu, "%s: unexpected cr %u\n", __FUNCTION__, cr);
1418 * Register the para guest with the host:
1420 static int vcpu_register_para(struct kvm_vcpu *vcpu, gpa_t para_state_gpa)
1422 struct kvm_vcpu_para_state *para_state;
1423 hpa_t para_state_hpa, hypercall_hpa;
1424 struct page *para_state_page;
1425 unsigned char *hypercall;
1426 gpa_t hypercall_gpa;
1428 printk(KERN_DEBUG "kvm: guest trying to enter paravirtual mode\n");
1429 printk(KERN_DEBUG ".... para_state_gpa: %08Lx\n", para_state_gpa);
1432 * Needs to be page aligned:
1434 if (para_state_gpa != PAGE_ALIGN(para_state_gpa))
1437 para_state_hpa = gpa_to_hpa(vcpu, para_state_gpa);
1438 printk(KERN_DEBUG ".... para_state_hpa: %08Lx\n", para_state_hpa);
1439 if (is_error_hpa(para_state_hpa))
1442 mark_page_dirty(vcpu->kvm, para_state_gpa >> PAGE_SHIFT);
1443 para_state_page = pfn_to_page(para_state_hpa >> PAGE_SHIFT);
1444 para_state = kmap_atomic(para_state_page, KM_USER0);
1446 printk(KERN_DEBUG ".... guest version: %d\n", para_state->guest_version);
1447 printk(KERN_DEBUG ".... size: %d\n", para_state->size);
1449 para_state->host_version = KVM_PARA_API_VERSION;
1451 * We cannot support guests that try to register themselves
1452 * with a newer API version than the host supports:
1454 if (para_state->guest_version > KVM_PARA_API_VERSION) {
1455 para_state->ret = -KVM_EINVAL;
1456 goto err_kunmap_skip;
1459 hypercall_gpa = para_state->hypercall_gpa;
1460 hypercall_hpa = gpa_to_hpa(vcpu, hypercall_gpa);
1461 printk(KERN_DEBUG ".... hypercall_hpa: %08Lx\n", hypercall_hpa);
1462 if (is_error_hpa(hypercall_hpa)) {
1463 para_state->ret = -KVM_EINVAL;
1464 goto err_kunmap_skip;
1467 printk(KERN_DEBUG "kvm: para guest successfully registered.\n");
1468 vcpu->para_state_page = para_state_page;
1469 vcpu->para_state_gpa = para_state_gpa;
1470 vcpu->hypercall_gpa = hypercall_gpa;
1472 mark_page_dirty(vcpu->kvm, hypercall_gpa >> PAGE_SHIFT);
1473 hypercall = kmap_atomic(pfn_to_page(hypercall_hpa >> PAGE_SHIFT),
1474 KM_USER1) + (hypercall_hpa & ~PAGE_MASK);
1475 kvm_arch_ops->patch_hypercall(vcpu, hypercall);
1476 kunmap_atomic(hypercall, KM_USER1);
1478 para_state->ret = 0;
1480 kunmap_atomic(para_state, KM_USER0);
1486 int kvm_get_msr_common(struct kvm_vcpu *vcpu, u32 msr, u64 *pdata)
1491 case 0xc0010010: /* SYSCFG */
1492 case 0xc0010015: /* HWCR */
1493 case MSR_IA32_PLATFORM_ID:
1494 case MSR_IA32_P5_MC_ADDR:
1495 case MSR_IA32_P5_MC_TYPE:
1496 case MSR_IA32_MC0_CTL:
1497 case MSR_IA32_MCG_STATUS:
1498 case MSR_IA32_MCG_CAP:
1499 case MSR_IA32_MC0_MISC:
1500 case MSR_IA32_MC0_MISC+4:
1501 case MSR_IA32_MC0_MISC+8:
1502 case MSR_IA32_MC0_MISC+12:
1503 case MSR_IA32_MC0_MISC+16:
1504 case MSR_IA32_UCODE_REV:
1505 case MSR_IA32_PERF_STATUS:
1506 case MSR_IA32_EBL_CR_POWERON:
1507 /* MTRR registers */
1509 case 0x200 ... 0x2ff:
1512 case 0xcd: /* fsb frequency */
1515 case MSR_IA32_APICBASE:
1516 data = vcpu->apic_base;
1518 case MSR_IA32_MISC_ENABLE:
1519 data = vcpu->ia32_misc_enable_msr;
1521 #ifdef CONFIG_X86_64
1523 data = vcpu->shadow_efer;
1527 printk(KERN_ERR "kvm: unhandled rdmsr: 0x%x\n", msr);
1533 EXPORT_SYMBOL_GPL(kvm_get_msr_common);
1536 * Reads an msr value (of 'msr_index') into 'pdata'.
1537 * Returns 0 on success, non-0 otherwise.
1538 * Assumes vcpu_load() was already called.
1540 int kvm_get_msr(struct kvm_vcpu *vcpu, u32 msr_index, u64 *pdata)
1542 return kvm_arch_ops->get_msr(vcpu, msr_index, pdata);
1545 #ifdef CONFIG_X86_64
1547 static void set_efer(struct kvm_vcpu *vcpu, u64 efer)
1549 if (efer & EFER_RESERVED_BITS) {
1550 printk(KERN_DEBUG "set_efer: 0x%llx #GP, reserved bits\n",
1557 && (vcpu->shadow_efer & EFER_LME) != (efer & EFER_LME)) {
1558 printk(KERN_DEBUG "set_efer: #GP, change LME while paging\n");
1563 kvm_arch_ops->set_efer(vcpu, efer);
1566 efer |= vcpu->shadow_efer & EFER_LMA;
1568 vcpu->shadow_efer = efer;
1573 int kvm_set_msr_common(struct kvm_vcpu *vcpu, u32 msr, u64 data)
1576 #ifdef CONFIG_X86_64
1578 set_efer(vcpu, data);
1581 case MSR_IA32_MC0_STATUS:
1582 printk(KERN_WARNING "%s: MSR_IA32_MC0_STATUS 0x%llx, nop\n",
1583 __FUNCTION__, data);
1585 case MSR_IA32_MCG_STATUS:
1586 printk(KERN_WARNING "%s: MSR_IA32_MCG_STATUS 0x%llx, nop\n",
1587 __FUNCTION__, data);
1589 case MSR_IA32_UCODE_REV:
1590 case MSR_IA32_UCODE_WRITE:
1591 case 0x200 ... 0x2ff: /* MTRRs */
1593 case MSR_IA32_APICBASE:
1594 vcpu->apic_base = data;
1596 case MSR_IA32_MISC_ENABLE:
1597 vcpu->ia32_misc_enable_msr = data;
1600 * This is the 'probe whether the host is KVM' logic:
1602 case MSR_KVM_API_MAGIC:
1603 return vcpu_register_para(vcpu, data);
1606 printk(KERN_ERR "kvm: unhandled wrmsr: 0x%x\n", msr);
1611 EXPORT_SYMBOL_GPL(kvm_set_msr_common);
1614 * Writes msr value into into the appropriate "register".
1615 * Returns 0 on success, non-0 otherwise.
1616 * Assumes vcpu_load() was already called.
1618 int kvm_set_msr(struct kvm_vcpu *vcpu, u32 msr_index, u64 data)
1620 return kvm_arch_ops->set_msr(vcpu, msr_index, data);
1623 void kvm_resched(struct kvm_vcpu *vcpu)
1625 if (!need_resched())
1631 EXPORT_SYMBOL_GPL(kvm_resched);
1633 void load_msrs(struct vmx_msr_entry *e, int n)
1637 for (i = 0; i < n; ++i)
1638 wrmsrl(e[i].index, e[i].data);
1640 EXPORT_SYMBOL_GPL(load_msrs);
1642 void save_msrs(struct vmx_msr_entry *e, int n)
1646 for (i = 0; i < n; ++i)
1647 rdmsrl(e[i].index, e[i].data);
1649 EXPORT_SYMBOL_GPL(save_msrs);
1651 void kvm_emulate_cpuid(struct kvm_vcpu *vcpu)
1655 struct kvm_cpuid_entry *e, *best;
1657 kvm_arch_ops->cache_regs(vcpu);
1658 function = vcpu->regs[VCPU_REGS_RAX];
1659 vcpu->regs[VCPU_REGS_RAX] = 0;
1660 vcpu->regs[VCPU_REGS_RBX] = 0;
1661 vcpu->regs[VCPU_REGS_RCX] = 0;
1662 vcpu->regs[VCPU_REGS_RDX] = 0;
1664 for (i = 0; i < vcpu->cpuid_nent; ++i) {
1665 e = &vcpu->cpuid_entries[i];
1666 if (e->function == function) {
1671 * Both basic or both extended?
1673 if (((e->function ^ function) & 0x80000000) == 0)
1674 if (!best || e->function > best->function)
1678 vcpu->regs[VCPU_REGS_RAX] = best->eax;
1679 vcpu->regs[VCPU_REGS_RBX] = best->ebx;
1680 vcpu->regs[VCPU_REGS_RCX] = best->ecx;
1681 vcpu->regs[VCPU_REGS_RDX] = best->edx;
1683 kvm_arch_ops->decache_regs(vcpu);
1684 kvm_arch_ops->skip_emulated_instruction(vcpu);
1686 EXPORT_SYMBOL_GPL(kvm_emulate_cpuid);
1688 static int pio_copy_data(struct kvm_vcpu *vcpu)
1690 void *p = vcpu->pio_data;
1693 int nr_pages = vcpu->pio.guest_pages[1] ? 2 : 1;
1695 kvm_arch_ops->vcpu_put(vcpu);
1696 q = vmap(vcpu->pio.guest_pages, nr_pages, VM_READ|VM_WRITE,
1699 kvm_arch_ops->vcpu_load(vcpu);
1700 free_pio_guest_pages(vcpu);
1703 q += vcpu->pio.guest_page_offset;
1704 bytes = vcpu->pio.size * vcpu->pio.cur_count;
1706 memcpy(q, p, bytes);
1708 memcpy(p, q, bytes);
1709 q -= vcpu->pio.guest_page_offset;
1711 kvm_arch_ops->vcpu_load(vcpu);
1712 free_pio_guest_pages(vcpu);
1716 static int complete_pio(struct kvm_vcpu *vcpu)
1718 struct kvm_pio_request *io = &vcpu->pio;
1722 kvm_arch_ops->cache_regs(vcpu);
1726 memcpy(&vcpu->regs[VCPU_REGS_RAX], vcpu->pio_data,
1730 r = pio_copy_data(vcpu);
1732 kvm_arch_ops->cache_regs(vcpu);
1739 delta *= io->cur_count;
1741 * The size of the register should really depend on
1742 * current address size.
1744 vcpu->regs[VCPU_REGS_RCX] -= delta;
1750 vcpu->regs[VCPU_REGS_RDI] += delta;
1752 vcpu->regs[VCPU_REGS_RSI] += delta;
1755 kvm_arch_ops->decache_regs(vcpu);
1757 io->count -= io->cur_count;
1761 kvm_arch_ops->skip_emulated_instruction(vcpu);
1765 void kernel_pio(struct kvm_io_device *pio_dev, struct kvm_vcpu *vcpu)
1767 /* TODO: String I/O for in kernel device */
1770 kvm_iodevice_read(pio_dev, vcpu->pio.port,
1774 kvm_iodevice_write(pio_dev, vcpu->pio.port,
1779 int kvm_setup_pio(struct kvm_vcpu *vcpu, struct kvm_run *run, int in,
1780 int size, unsigned long count, int string, int down,
1781 gva_t address, int rep, unsigned port)
1783 unsigned now, in_page;
1787 struct kvm_io_device *pio_dev;
1789 vcpu->run->exit_reason = KVM_EXIT_IO;
1790 vcpu->run->io.direction = in ? KVM_EXIT_IO_IN : KVM_EXIT_IO_OUT;
1791 vcpu->run->io.size = size;
1792 vcpu->run->io.data_offset = KVM_PIO_PAGE_OFFSET * PAGE_SIZE;
1793 vcpu->run->io.count = count;
1794 vcpu->run->io.port = port;
1795 vcpu->pio.count = count;
1796 vcpu->pio.cur_count = count;
1797 vcpu->pio.size = size;
1799 vcpu->pio.port = port;
1800 vcpu->pio.string = string;
1801 vcpu->pio.down = down;
1802 vcpu->pio.guest_page_offset = offset_in_page(address);
1803 vcpu->pio.rep = rep;
1805 pio_dev = vcpu_find_pio_dev(vcpu, port);
1807 kvm_arch_ops->cache_regs(vcpu);
1808 memcpy(vcpu->pio_data, &vcpu->regs[VCPU_REGS_RAX], 4);
1809 kvm_arch_ops->decache_regs(vcpu);
1811 kernel_pio(pio_dev, vcpu);
1817 /* TODO: String I/O for in kernel device */
1819 printk(KERN_ERR "kvm_setup_pio: no string io support\n");
1822 kvm_arch_ops->skip_emulated_instruction(vcpu);
1826 now = min(count, PAGE_SIZE / size);
1829 in_page = PAGE_SIZE - offset_in_page(address);
1831 in_page = offset_in_page(address) + size;
1832 now = min(count, (unsigned long)in_page / size);
1835 * String I/O straddles page boundary. Pin two guest pages
1836 * so that we satisfy atomicity constraints. Do just one
1837 * transaction to avoid complexity.
1844 * String I/O in reverse. Yuck. Kill the guest, fix later.
1846 printk(KERN_ERR "kvm: guest string pio down\n");
1850 vcpu->run->io.count = now;
1851 vcpu->pio.cur_count = now;
1853 for (i = 0; i < nr_pages; ++i) {
1854 spin_lock(&vcpu->kvm->lock);
1855 page = gva_to_page(vcpu, address + i * PAGE_SIZE);
1858 vcpu->pio.guest_pages[i] = page;
1859 spin_unlock(&vcpu->kvm->lock);
1862 free_pio_guest_pages(vcpu);
1868 return pio_copy_data(vcpu);
1871 EXPORT_SYMBOL_GPL(kvm_setup_pio);
1873 static int kvm_vcpu_ioctl_run(struct kvm_vcpu *vcpu, struct kvm_run *kvm_run)
1880 if (vcpu->sigset_active)
1881 sigprocmask(SIG_SETMASK, &vcpu->sigset, &sigsaved);
1883 /* re-sync apic's tpr */
1884 vcpu->cr8 = kvm_run->cr8;
1886 if (vcpu->pio.cur_count) {
1887 r = complete_pio(vcpu);
1892 if (vcpu->mmio_needed) {
1893 memcpy(vcpu->mmio_data, kvm_run->mmio.data, 8);
1894 vcpu->mmio_read_completed = 1;
1895 vcpu->mmio_needed = 0;
1896 r = emulate_instruction(vcpu, kvm_run,
1897 vcpu->mmio_fault_cr2, 0);
1898 if (r == EMULATE_DO_MMIO) {
1900 * Read-modify-write. Back to userspace.
1902 kvm_run->exit_reason = KVM_EXIT_MMIO;
1908 if (kvm_run->exit_reason == KVM_EXIT_HYPERCALL) {
1909 kvm_arch_ops->cache_regs(vcpu);
1910 vcpu->regs[VCPU_REGS_RAX] = kvm_run->hypercall.ret;
1911 kvm_arch_ops->decache_regs(vcpu);
1914 r = kvm_arch_ops->run(vcpu, kvm_run);
1917 if (vcpu->sigset_active)
1918 sigprocmask(SIG_SETMASK, &sigsaved, NULL);
1924 static int kvm_vcpu_ioctl_get_regs(struct kvm_vcpu *vcpu,
1925 struct kvm_regs *regs)
1929 kvm_arch_ops->cache_regs(vcpu);
1931 regs->rax = vcpu->regs[VCPU_REGS_RAX];
1932 regs->rbx = vcpu->regs[VCPU_REGS_RBX];
1933 regs->rcx = vcpu->regs[VCPU_REGS_RCX];
1934 regs->rdx = vcpu->regs[VCPU_REGS_RDX];
1935 regs->rsi = vcpu->regs[VCPU_REGS_RSI];
1936 regs->rdi = vcpu->regs[VCPU_REGS_RDI];
1937 regs->rsp = vcpu->regs[VCPU_REGS_RSP];
1938 regs->rbp = vcpu->regs[VCPU_REGS_RBP];
1939 #ifdef CONFIG_X86_64
1940 regs->r8 = vcpu->regs[VCPU_REGS_R8];
1941 regs->r9 = vcpu->regs[VCPU_REGS_R9];
1942 regs->r10 = vcpu->regs[VCPU_REGS_R10];
1943 regs->r11 = vcpu->regs[VCPU_REGS_R11];
1944 regs->r12 = vcpu->regs[VCPU_REGS_R12];
1945 regs->r13 = vcpu->regs[VCPU_REGS_R13];
1946 regs->r14 = vcpu->regs[VCPU_REGS_R14];
1947 regs->r15 = vcpu->regs[VCPU_REGS_R15];
1950 regs->rip = vcpu->rip;
1951 regs->rflags = kvm_arch_ops->get_rflags(vcpu);
1954 * Don't leak debug flags in case they were set for guest debugging
1956 if (vcpu->guest_debug.enabled && vcpu->guest_debug.singlestep)
1957 regs->rflags &= ~(X86_EFLAGS_TF | X86_EFLAGS_RF);
1964 static int kvm_vcpu_ioctl_set_regs(struct kvm_vcpu *vcpu,
1965 struct kvm_regs *regs)
1969 vcpu->regs[VCPU_REGS_RAX] = regs->rax;
1970 vcpu->regs[VCPU_REGS_RBX] = regs->rbx;
1971 vcpu->regs[VCPU_REGS_RCX] = regs->rcx;
1972 vcpu->regs[VCPU_REGS_RDX] = regs->rdx;
1973 vcpu->regs[VCPU_REGS_RSI] = regs->rsi;
1974 vcpu->regs[VCPU_REGS_RDI] = regs->rdi;
1975 vcpu->regs[VCPU_REGS_RSP] = regs->rsp;
1976 vcpu->regs[VCPU_REGS_RBP] = regs->rbp;
1977 #ifdef CONFIG_X86_64
1978 vcpu->regs[VCPU_REGS_R8] = regs->r8;
1979 vcpu->regs[VCPU_REGS_R9] = regs->r9;
1980 vcpu->regs[VCPU_REGS_R10] = regs->r10;
1981 vcpu->regs[VCPU_REGS_R11] = regs->r11;
1982 vcpu->regs[VCPU_REGS_R12] = regs->r12;
1983 vcpu->regs[VCPU_REGS_R13] = regs->r13;
1984 vcpu->regs[VCPU_REGS_R14] = regs->r14;
1985 vcpu->regs[VCPU_REGS_R15] = regs->r15;
1988 vcpu->rip = regs->rip;
1989 kvm_arch_ops->set_rflags(vcpu, regs->rflags);
1991 kvm_arch_ops->decache_regs(vcpu);
1998 static void get_segment(struct kvm_vcpu *vcpu,
1999 struct kvm_segment *var, int seg)
2001 return kvm_arch_ops->get_segment(vcpu, var, seg);
2004 static int kvm_vcpu_ioctl_get_sregs(struct kvm_vcpu *vcpu,
2005 struct kvm_sregs *sregs)
2007 struct descriptor_table dt;
2011 get_segment(vcpu, &sregs->cs, VCPU_SREG_CS);
2012 get_segment(vcpu, &sregs->ds, VCPU_SREG_DS);
2013 get_segment(vcpu, &sregs->es, VCPU_SREG_ES);
2014 get_segment(vcpu, &sregs->fs, VCPU_SREG_FS);
2015 get_segment(vcpu, &sregs->gs, VCPU_SREG_GS);
2016 get_segment(vcpu, &sregs->ss, VCPU_SREG_SS);
2018 get_segment(vcpu, &sregs->tr, VCPU_SREG_TR);
2019 get_segment(vcpu, &sregs->ldt, VCPU_SREG_LDTR);
2021 kvm_arch_ops->get_idt(vcpu, &dt);
2022 sregs->idt.limit = dt.limit;
2023 sregs->idt.base = dt.base;
2024 kvm_arch_ops->get_gdt(vcpu, &dt);
2025 sregs->gdt.limit = dt.limit;
2026 sregs->gdt.base = dt.base;
2028 kvm_arch_ops->decache_cr4_guest_bits(vcpu);
2029 sregs->cr0 = vcpu->cr0;
2030 sregs->cr2 = vcpu->cr2;
2031 sregs->cr3 = vcpu->cr3;
2032 sregs->cr4 = vcpu->cr4;
2033 sregs->cr8 = vcpu->cr8;
2034 sregs->efer = vcpu->shadow_efer;
2035 sregs->apic_base = vcpu->apic_base;
2037 memcpy(sregs->interrupt_bitmap, vcpu->irq_pending,
2038 sizeof sregs->interrupt_bitmap);
2045 static void set_segment(struct kvm_vcpu *vcpu,
2046 struct kvm_segment *var, int seg)
2048 return kvm_arch_ops->set_segment(vcpu, var, seg);
2051 static int kvm_vcpu_ioctl_set_sregs(struct kvm_vcpu *vcpu,
2052 struct kvm_sregs *sregs)
2054 int mmu_reset_needed = 0;
2056 struct descriptor_table dt;
2060 dt.limit = sregs->idt.limit;
2061 dt.base = sregs->idt.base;
2062 kvm_arch_ops->set_idt(vcpu, &dt);
2063 dt.limit = sregs->gdt.limit;
2064 dt.base = sregs->gdt.base;
2065 kvm_arch_ops->set_gdt(vcpu, &dt);
2067 vcpu->cr2 = sregs->cr2;
2068 mmu_reset_needed |= vcpu->cr3 != sregs->cr3;
2069 vcpu->cr3 = sregs->cr3;
2071 vcpu->cr8 = sregs->cr8;
2073 mmu_reset_needed |= vcpu->shadow_efer != sregs->efer;
2074 #ifdef CONFIG_X86_64
2075 kvm_arch_ops->set_efer(vcpu, sregs->efer);
2077 vcpu->apic_base = sregs->apic_base;
2079 kvm_arch_ops->decache_cr4_guest_bits(vcpu);
2081 mmu_reset_needed |= vcpu->cr0 != sregs->cr0;
2082 kvm_arch_ops->set_cr0(vcpu, sregs->cr0);
2084 mmu_reset_needed |= vcpu->cr4 != sregs->cr4;
2085 kvm_arch_ops->set_cr4(vcpu, sregs->cr4);
2086 if (!is_long_mode(vcpu) && is_pae(vcpu))
2087 load_pdptrs(vcpu, vcpu->cr3);
2089 if (mmu_reset_needed)
2090 kvm_mmu_reset_context(vcpu);
2092 memcpy(vcpu->irq_pending, sregs->interrupt_bitmap,
2093 sizeof vcpu->irq_pending);
2094 vcpu->irq_summary = 0;
2095 for (i = 0; i < NR_IRQ_WORDS; ++i)
2096 if (vcpu->irq_pending[i])
2097 __set_bit(i, &vcpu->irq_summary);
2099 set_segment(vcpu, &sregs->cs, VCPU_SREG_CS);
2100 set_segment(vcpu, &sregs->ds, VCPU_SREG_DS);
2101 set_segment(vcpu, &sregs->es, VCPU_SREG_ES);
2102 set_segment(vcpu, &sregs->fs, VCPU_SREG_FS);
2103 set_segment(vcpu, &sregs->gs, VCPU_SREG_GS);
2104 set_segment(vcpu, &sregs->ss, VCPU_SREG_SS);
2106 set_segment(vcpu, &sregs->tr, VCPU_SREG_TR);
2107 set_segment(vcpu, &sregs->ldt, VCPU_SREG_LDTR);
2115 * List of msr numbers which we expose to userspace through KVM_GET_MSRS
2116 * and KVM_SET_MSRS, and KVM_GET_MSR_INDEX_LIST.
2118 * This list is modified at module load time to reflect the
2119 * capabilities of the host cpu.
2121 static u32 msrs_to_save[] = {
2122 MSR_IA32_SYSENTER_CS, MSR_IA32_SYSENTER_ESP, MSR_IA32_SYSENTER_EIP,
2124 #ifdef CONFIG_X86_64
2125 MSR_CSTAR, MSR_KERNEL_GS_BASE, MSR_SYSCALL_MASK, MSR_LSTAR,
2127 MSR_IA32_TIME_STAMP_COUNTER,
2130 static unsigned num_msrs_to_save;
2132 static u32 emulated_msrs[] = {
2133 MSR_IA32_MISC_ENABLE,
2136 static __init void kvm_init_msr_list(void)
2141 for (i = j = 0; i < ARRAY_SIZE(msrs_to_save); i++) {
2142 if (rdmsr_safe(msrs_to_save[i], &dummy[0], &dummy[1]) < 0)
2145 msrs_to_save[j] = msrs_to_save[i];
2148 num_msrs_to_save = j;
2152 * Adapt set_msr() to msr_io()'s calling convention
2154 static int do_set_msr(struct kvm_vcpu *vcpu, unsigned index, u64 *data)
2156 return kvm_set_msr(vcpu, index, *data);
2160 * Read or write a bunch of msrs. All parameters are kernel addresses.
2162 * @return number of msrs set successfully.
2164 static int __msr_io(struct kvm_vcpu *vcpu, struct kvm_msrs *msrs,
2165 struct kvm_msr_entry *entries,
2166 int (*do_msr)(struct kvm_vcpu *vcpu,
2167 unsigned index, u64 *data))
2173 for (i = 0; i < msrs->nmsrs; ++i)
2174 if (do_msr(vcpu, entries[i].index, &entries[i].data))
2183 * Read or write a bunch of msrs. Parameters are user addresses.
2185 * @return number of msrs set successfully.
2187 static int msr_io(struct kvm_vcpu *vcpu, struct kvm_msrs __user *user_msrs,
2188 int (*do_msr)(struct kvm_vcpu *vcpu,
2189 unsigned index, u64 *data),
2192 struct kvm_msrs msrs;
2193 struct kvm_msr_entry *entries;
2198 if (copy_from_user(&msrs, user_msrs, sizeof msrs))
2202 if (msrs.nmsrs >= MAX_IO_MSRS)
2206 size = sizeof(struct kvm_msr_entry) * msrs.nmsrs;
2207 entries = vmalloc(size);
2212 if (copy_from_user(entries, user_msrs->entries, size))
2215 r = n = __msr_io(vcpu, &msrs, entries, do_msr);
2220 if (writeback && copy_to_user(user_msrs->entries, entries, size))
2232 * Translate a guest virtual address to a guest physical address.
2234 static int kvm_vcpu_ioctl_translate(struct kvm_vcpu *vcpu,
2235 struct kvm_translation *tr)
2237 unsigned long vaddr = tr->linear_address;
2241 spin_lock(&vcpu->kvm->lock);
2242 gpa = vcpu->mmu.gva_to_gpa(vcpu, vaddr);
2243 tr->physical_address = gpa;
2244 tr->valid = gpa != UNMAPPED_GVA;
2247 spin_unlock(&vcpu->kvm->lock);
2253 static int kvm_vcpu_ioctl_interrupt(struct kvm_vcpu *vcpu,
2254 struct kvm_interrupt *irq)
2256 if (irq->irq < 0 || irq->irq >= 256)
2260 set_bit(irq->irq, vcpu->irq_pending);
2261 set_bit(irq->irq / BITS_PER_LONG, &vcpu->irq_summary);
2268 static int kvm_vcpu_ioctl_debug_guest(struct kvm_vcpu *vcpu,
2269 struct kvm_debug_guest *dbg)
2275 r = kvm_arch_ops->set_guest_debug(vcpu, dbg);
2282 static struct page *kvm_vcpu_nopage(struct vm_area_struct *vma,
2283 unsigned long address,
2286 struct kvm_vcpu *vcpu = vma->vm_file->private_data;
2287 unsigned long pgoff;
2290 *type = VM_FAULT_MINOR;
2291 pgoff = ((address - vma->vm_start) >> PAGE_SHIFT) + vma->vm_pgoff;
2293 page = virt_to_page(vcpu->run);
2294 else if (pgoff == KVM_PIO_PAGE_OFFSET)
2295 page = virt_to_page(vcpu->pio_data);
2297 return NOPAGE_SIGBUS;
2302 static struct vm_operations_struct kvm_vcpu_vm_ops = {
2303 .nopage = kvm_vcpu_nopage,
2306 static int kvm_vcpu_mmap(struct file *file, struct vm_area_struct *vma)
2308 vma->vm_ops = &kvm_vcpu_vm_ops;
2312 static int kvm_vcpu_release(struct inode *inode, struct file *filp)
2314 struct kvm_vcpu *vcpu = filp->private_data;
2316 fput(vcpu->kvm->filp);
2320 static struct file_operations kvm_vcpu_fops = {
2321 .release = kvm_vcpu_release,
2322 .unlocked_ioctl = kvm_vcpu_ioctl,
2323 .compat_ioctl = kvm_vcpu_ioctl,
2324 .mmap = kvm_vcpu_mmap,
2328 * Allocates an inode for the vcpu.
2330 static int create_vcpu_fd(struct kvm_vcpu *vcpu)
2333 struct inode *inode;
2336 r = anon_inode_getfd(&fd, &inode, &file,
2337 "kvm-vcpu", &kvm_vcpu_fops, vcpu);
2340 atomic_inc(&vcpu->kvm->filp->f_count);
2345 * Creates some virtual cpus. Good luck creating more than one.
2347 static int kvm_vm_ioctl_create_vcpu(struct kvm *kvm, int n)
2350 struct kvm_vcpu *vcpu;
2357 vcpu = &kvm->vcpus[n];
2359 mutex_lock(&vcpu->mutex);
2362 mutex_unlock(&vcpu->mutex);
2366 page = alloc_page(GFP_KERNEL | __GFP_ZERO);
2370 vcpu->run = page_address(page);
2372 page = alloc_page(GFP_KERNEL | __GFP_ZERO);
2376 vcpu->pio_data = page_address(page);
2378 vcpu->host_fx_image = (char*)ALIGN((hva_t)vcpu->fx_buf,
2380 vcpu->guest_fx_image = vcpu->host_fx_image + FX_IMAGE_SIZE;
2383 r = kvm_arch_ops->vcpu_create(vcpu);
2385 goto out_free_vcpus;
2387 r = kvm_mmu_create(vcpu);
2389 goto out_free_vcpus;
2391 kvm_arch_ops->vcpu_load(vcpu);
2392 r = kvm_mmu_setup(vcpu);
2394 r = kvm_arch_ops->vcpu_setup(vcpu);
2398 goto out_free_vcpus;
2400 r = create_vcpu_fd(vcpu);
2402 goto out_free_vcpus;
2404 spin_lock(&kvm_lock);
2405 if (n >= kvm->nvcpus)
2406 kvm->nvcpus = n + 1;
2407 spin_unlock(&kvm_lock);
2412 kvm_free_vcpu(vcpu);
2414 free_page((unsigned long)vcpu->run);
2417 mutex_unlock(&vcpu->mutex);
2422 static void cpuid_fix_nx_cap(struct kvm_vcpu *vcpu)
2426 struct kvm_cpuid_entry *e, *entry;
2428 rdmsrl(MSR_EFER, efer);
2430 for (i = 0; i < vcpu->cpuid_nent; ++i) {
2431 e = &vcpu->cpuid_entries[i];
2432 if (e->function == 0x80000001) {
2437 if (entry && (entry->edx & EFER_NX) && !(efer & EFER_NX)) {
2438 entry->edx &= ~(1 << 20);
2439 printk(KERN_INFO ": guest NX capability removed\n");
2443 static int kvm_vcpu_ioctl_set_cpuid(struct kvm_vcpu *vcpu,
2444 struct kvm_cpuid *cpuid,
2445 struct kvm_cpuid_entry __user *entries)
2450 if (cpuid->nent > KVM_MAX_CPUID_ENTRIES)
2453 if (copy_from_user(&vcpu->cpuid_entries, entries,
2454 cpuid->nent * sizeof(struct kvm_cpuid_entry)))
2456 vcpu->cpuid_nent = cpuid->nent;
2457 cpuid_fix_nx_cap(vcpu);
2464 static int kvm_vcpu_ioctl_set_sigmask(struct kvm_vcpu *vcpu, sigset_t *sigset)
2467 sigdelsetmask(sigset, sigmask(SIGKILL)|sigmask(SIGSTOP));
2468 vcpu->sigset_active = 1;
2469 vcpu->sigset = *sigset;
2471 vcpu->sigset_active = 0;
2476 * fxsave fpu state. Taken from x86_64/processor.h. To be killed when
2477 * we have asm/x86/processor.h
2488 u32 st_space[32]; /* 8*16 bytes for each FP-reg = 128 bytes */
2489 #ifdef CONFIG_X86_64
2490 u32 xmm_space[64]; /* 16*16 bytes for each XMM-reg = 256 bytes */
2492 u32 xmm_space[32]; /* 8*16 bytes for each XMM-reg = 128 bytes */
2496 static int kvm_vcpu_ioctl_get_fpu(struct kvm_vcpu *vcpu, struct kvm_fpu *fpu)
2498 struct fxsave *fxsave = (struct fxsave *)vcpu->guest_fx_image;
2502 memcpy(fpu->fpr, fxsave->st_space, 128);
2503 fpu->fcw = fxsave->cwd;
2504 fpu->fsw = fxsave->swd;
2505 fpu->ftwx = fxsave->twd;
2506 fpu->last_opcode = fxsave->fop;
2507 fpu->last_ip = fxsave->rip;
2508 fpu->last_dp = fxsave->rdp;
2509 memcpy(fpu->xmm, fxsave->xmm_space, sizeof fxsave->xmm_space);
2516 static int kvm_vcpu_ioctl_set_fpu(struct kvm_vcpu *vcpu, struct kvm_fpu *fpu)
2518 struct fxsave *fxsave = (struct fxsave *)vcpu->guest_fx_image;
2522 memcpy(fxsave->st_space, fpu->fpr, 128);
2523 fxsave->cwd = fpu->fcw;
2524 fxsave->swd = fpu->fsw;
2525 fxsave->twd = fpu->ftwx;
2526 fxsave->fop = fpu->last_opcode;
2527 fxsave->rip = fpu->last_ip;
2528 fxsave->rdp = fpu->last_dp;
2529 memcpy(fxsave->xmm_space, fpu->xmm, sizeof fxsave->xmm_space);
2536 static long kvm_vcpu_ioctl(struct file *filp,
2537 unsigned int ioctl, unsigned long arg)
2539 struct kvm_vcpu *vcpu = filp->private_data;
2540 void __user *argp = (void __user *)arg;
2548 r = kvm_vcpu_ioctl_run(vcpu, vcpu->run);
2550 case KVM_GET_REGS: {
2551 struct kvm_regs kvm_regs;
2553 memset(&kvm_regs, 0, sizeof kvm_regs);
2554 r = kvm_vcpu_ioctl_get_regs(vcpu, &kvm_regs);
2558 if (copy_to_user(argp, &kvm_regs, sizeof kvm_regs))
2563 case KVM_SET_REGS: {
2564 struct kvm_regs kvm_regs;
2567 if (copy_from_user(&kvm_regs, argp, sizeof kvm_regs))
2569 r = kvm_vcpu_ioctl_set_regs(vcpu, &kvm_regs);
2575 case KVM_GET_SREGS: {
2576 struct kvm_sregs kvm_sregs;
2578 memset(&kvm_sregs, 0, sizeof kvm_sregs);
2579 r = kvm_vcpu_ioctl_get_sregs(vcpu, &kvm_sregs);
2583 if (copy_to_user(argp, &kvm_sregs, sizeof kvm_sregs))
2588 case KVM_SET_SREGS: {
2589 struct kvm_sregs kvm_sregs;
2592 if (copy_from_user(&kvm_sregs, argp, sizeof kvm_sregs))
2594 r = kvm_vcpu_ioctl_set_sregs(vcpu, &kvm_sregs);
2600 case KVM_TRANSLATE: {
2601 struct kvm_translation tr;
2604 if (copy_from_user(&tr, argp, sizeof tr))
2606 r = kvm_vcpu_ioctl_translate(vcpu, &tr);
2610 if (copy_to_user(argp, &tr, sizeof tr))
2615 case KVM_INTERRUPT: {
2616 struct kvm_interrupt irq;
2619 if (copy_from_user(&irq, argp, sizeof irq))
2621 r = kvm_vcpu_ioctl_interrupt(vcpu, &irq);
2627 case KVM_DEBUG_GUEST: {
2628 struct kvm_debug_guest dbg;
2631 if (copy_from_user(&dbg, argp, sizeof dbg))
2633 r = kvm_vcpu_ioctl_debug_guest(vcpu, &dbg);
2640 r = msr_io(vcpu, argp, kvm_get_msr, 1);
2643 r = msr_io(vcpu, argp, do_set_msr, 0);
2645 case KVM_SET_CPUID: {
2646 struct kvm_cpuid __user *cpuid_arg = argp;
2647 struct kvm_cpuid cpuid;
2650 if (copy_from_user(&cpuid, cpuid_arg, sizeof cpuid))
2652 r = kvm_vcpu_ioctl_set_cpuid(vcpu, &cpuid, cpuid_arg->entries);
2657 case KVM_SET_SIGNAL_MASK: {
2658 struct kvm_signal_mask __user *sigmask_arg = argp;
2659 struct kvm_signal_mask kvm_sigmask;
2660 sigset_t sigset, *p;
2665 if (copy_from_user(&kvm_sigmask, argp,
2666 sizeof kvm_sigmask))
2669 if (kvm_sigmask.len != sizeof sigset)
2672 if (copy_from_user(&sigset, sigmask_arg->sigset,
2677 r = kvm_vcpu_ioctl_set_sigmask(vcpu, &sigset);
2683 memset(&fpu, 0, sizeof fpu);
2684 r = kvm_vcpu_ioctl_get_fpu(vcpu, &fpu);
2688 if (copy_to_user(argp, &fpu, sizeof fpu))
2697 if (copy_from_user(&fpu, argp, sizeof fpu))
2699 r = kvm_vcpu_ioctl_set_fpu(vcpu, &fpu);
2712 static long kvm_vm_ioctl(struct file *filp,
2713 unsigned int ioctl, unsigned long arg)
2715 struct kvm *kvm = filp->private_data;
2716 void __user *argp = (void __user *)arg;
2720 case KVM_CREATE_VCPU:
2721 r = kvm_vm_ioctl_create_vcpu(kvm, arg);
2725 case KVM_SET_MEMORY_REGION: {
2726 struct kvm_memory_region kvm_mem;
2729 if (copy_from_user(&kvm_mem, argp, sizeof kvm_mem))
2731 r = kvm_vm_ioctl_set_memory_region(kvm, &kvm_mem);
2736 case KVM_GET_DIRTY_LOG: {
2737 struct kvm_dirty_log log;
2740 if (copy_from_user(&log, argp, sizeof log))
2742 r = kvm_vm_ioctl_get_dirty_log(kvm, &log);
2747 case KVM_SET_MEMORY_ALIAS: {
2748 struct kvm_memory_alias alias;
2751 if (copy_from_user(&alias, argp, sizeof alias))
2753 r = kvm_vm_ioctl_set_memory_alias(kvm, &alias);
2765 static struct page *kvm_vm_nopage(struct vm_area_struct *vma,
2766 unsigned long address,
2769 struct kvm *kvm = vma->vm_file->private_data;
2770 unsigned long pgoff;
2773 *type = VM_FAULT_MINOR;
2774 pgoff = ((address - vma->vm_start) >> PAGE_SHIFT) + vma->vm_pgoff;
2775 page = gfn_to_page(kvm, pgoff);
2777 return NOPAGE_SIGBUS;
2782 static struct vm_operations_struct kvm_vm_vm_ops = {
2783 .nopage = kvm_vm_nopage,
2786 static int kvm_vm_mmap(struct file *file, struct vm_area_struct *vma)
2788 vma->vm_ops = &kvm_vm_vm_ops;
2792 static struct file_operations kvm_vm_fops = {
2793 .release = kvm_vm_release,
2794 .unlocked_ioctl = kvm_vm_ioctl,
2795 .compat_ioctl = kvm_vm_ioctl,
2796 .mmap = kvm_vm_mmap,
2799 static int kvm_dev_ioctl_create_vm(void)
2802 struct inode *inode;
2806 kvm = kvm_create_vm();
2808 return PTR_ERR(kvm);
2809 r = anon_inode_getfd(&fd, &inode, &file, "kvm-vm", &kvm_vm_fops, kvm);
2811 kvm_destroy_vm(kvm);
2820 static long kvm_dev_ioctl(struct file *filp,
2821 unsigned int ioctl, unsigned long arg)
2823 void __user *argp = (void __user *)arg;
2827 case KVM_GET_API_VERSION:
2831 r = KVM_API_VERSION;
2837 r = kvm_dev_ioctl_create_vm();
2839 case KVM_GET_MSR_INDEX_LIST: {
2840 struct kvm_msr_list __user *user_msr_list = argp;
2841 struct kvm_msr_list msr_list;
2845 if (copy_from_user(&msr_list, user_msr_list, sizeof msr_list))
2848 msr_list.nmsrs = num_msrs_to_save + ARRAY_SIZE(emulated_msrs);
2849 if (copy_to_user(user_msr_list, &msr_list, sizeof msr_list))
2852 if (n < num_msrs_to_save)
2855 if (copy_to_user(user_msr_list->indices, &msrs_to_save,
2856 num_msrs_to_save * sizeof(u32)))
2858 if (copy_to_user(user_msr_list->indices
2859 + num_msrs_to_save * sizeof(u32),
2861 ARRAY_SIZE(emulated_msrs) * sizeof(u32)))
2866 case KVM_CHECK_EXTENSION:
2868 * No extensions defined at present.
2872 case KVM_GET_VCPU_MMAP_SIZE:
2885 static struct file_operations kvm_chardev_ops = {
2886 .open = kvm_dev_open,
2887 .release = kvm_dev_release,
2888 .unlocked_ioctl = kvm_dev_ioctl,
2889 .compat_ioctl = kvm_dev_ioctl,
2892 static struct miscdevice kvm_dev = {
2898 static int kvm_reboot(struct notifier_block *notifier, unsigned long val,
2901 if (val == SYS_RESTART) {
2903 * Some (well, at least mine) BIOSes hang on reboot if
2906 printk(KERN_INFO "kvm: exiting hardware virtualization\n");
2907 on_each_cpu(hardware_disable, NULL, 0, 1);
2912 static struct notifier_block kvm_reboot_notifier = {
2913 .notifier_call = kvm_reboot,
2918 * Make sure that a cpu that is being hot-unplugged does not have any vcpus
2921 static void decache_vcpus_on_cpu(int cpu)
2924 struct kvm_vcpu *vcpu;
2927 spin_lock(&kvm_lock);
2928 list_for_each_entry(vm, &vm_list, vm_list)
2929 for (i = 0; i < KVM_MAX_VCPUS; ++i) {
2930 vcpu = &vm->vcpus[i];
2932 * If the vcpu is locked, then it is running on some
2933 * other cpu and therefore it is not cached on the
2936 * If it's not locked, check the last cpu it executed
2939 if (mutex_trylock(&vcpu->mutex)) {
2940 if (vcpu->cpu == cpu) {
2941 kvm_arch_ops->vcpu_decache(vcpu);
2944 mutex_unlock(&vcpu->mutex);
2947 spin_unlock(&kvm_lock);
2950 static void hardware_enable(void *junk)
2952 int cpu = raw_smp_processor_id();
2954 if (cpu_isset(cpu, cpus_hardware_enabled))
2956 cpu_set(cpu, cpus_hardware_enabled);
2957 kvm_arch_ops->hardware_enable(NULL);
2960 static void hardware_disable(void *junk)
2962 int cpu = raw_smp_processor_id();
2964 if (!cpu_isset(cpu, cpus_hardware_enabled))
2966 cpu_clear(cpu, cpus_hardware_enabled);
2967 decache_vcpus_on_cpu(cpu);
2968 kvm_arch_ops->hardware_disable(NULL);
2971 static int kvm_cpu_hotplug(struct notifier_block *notifier, unsigned long val,
2978 case CPU_DYING_FROZEN:
2979 case CPU_UP_CANCELED:
2980 case CPU_UP_CANCELED_FROZEN:
2981 printk(KERN_INFO "kvm: disabling virtualization on CPU%d\n",
2983 smp_call_function_single(cpu, hardware_disable, NULL, 0, 1);
2986 case CPU_ONLINE_FROZEN:
2987 printk(KERN_INFO "kvm: enabling virtualization on CPU%d\n",
2989 smp_call_function_single(cpu, hardware_enable, NULL, 0, 1);
2995 void kvm_io_bus_init(struct kvm_io_bus *bus)
2997 memset(bus, 0, sizeof(*bus));
3000 void kvm_io_bus_destroy(struct kvm_io_bus *bus)
3004 for (i = 0; i < bus->dev_count; i++) {
3005 struct kvm_io_device *pos = bus->devs[i];
3007 kvm_iodevice_destructor(pos);
3011 struct kvm_io_device *kvm_io_bus_find_dev(struct kvm_io_bus *bus, gpa_t addr)
3015 for (i = 0; i < bus->dev_count; i++) {
3016 struct kvm_io_device *pos = bus->devs[i];
3018 if (pos->in_range(pos, addr))
3025 void kvm_io_bus_register_dev(struct kvm_io_bus *bus, struct kvm_io_device *dev)
3027 BUG_ON(bus->dev_count > (NR_IOBUS_DEVS-1));
3029 bus->devs[bus->dev_count++] = dev;
3032 static struct notifier_block kvm_cpu_notifier = {
3033 .notifier_call = kvm_cpu_hotplug,
3034 .priority = 20, /* must be > scheduler priority */
3037 static u64 stat_get(void *_offset)
3039 unsigned offset = (long)_offset;
3042 struct kvm_vcpu *vcpu;
3045 spin_lock(&kvm_lock);
3046 list_for_each_entry(kvm, &vm_list, vm_list)
3047 for (i = 0; i < KVM_MAX_VCPUS; ++i) {
3048 vcpu = &kvm->vcpus[i];
3049 total += *(u32 *)((void *)vcpu + offset);
3051 spin_unlock(&kvm_lock);
3055 static void stat_set(void *offset, u64 val)
3059 DEFINE_SIMPLE_ATTRIBUTE(stat_fops, stat_get, stat_set, "%llu\n");
3061 static __init void kvm_init_debug(void)
3063 struct kvm_stats_debugfs_item *p;
3065 debugfs_dir = debugfs_create_dir("kvm", NULL);
3066 for (p = debugfs_entries; p->name; ++p)
3067 p->dentry = debugfs_create_file(p->name, 0444, debugfs_dir,
3068 (void *)(long)p->offset,
3072 static void kvm_exit_debug(void)
3074 struct kvm_stats_debugfs_item *p;
3076 for (p = debugfs_entries; p->name; ++p)
3077 debugfs_remove(p->dentry);
3078 debugfs_remove(debugfs_dir);
3081 static int kvm_suspend(struct sys_device *dev, pm_message_t state)
3083 hardware_disable(NULL);
3087 static int kvm_resume(struct sys_device *dev)
3089 hardware_enable(NULL);
3093 static struct sysdev_class kvm_sysdev_class = {
3094 set_kset_name("kvm"),
3095 .suspend = kvm_suspend,
3096 .resume = kvm_resume,
3099 static struct sys_device kvm_sysdev = {
3101 .cls = &kvm_sysdev_class,
3104 hpa_t bad_page_address;
3106 int kvm_init_arch(struct kvm_arch_ops *ops, struct module *module)
3111 printk(KERN_ERR "kvm: already loaded the other module\n");
3115 if (!ops->cpu_has_kvm_support()) {
3116 printk(KERN_ERR "kvm: no hardware support\n");
3119 if (ops->disabled_by_bios()) {
3120 printk(KERN_ERR "kvm: disabled by bios\n");
3126 r = kvm_arch_ops->hardware_setup();
3130 on_each_cpu(hardware_enable, NULL, 0, 1);
3131 r = register_cpu_notifier(&kvm_cpu_notifier);
3134 register_reboot_notifier(&kvm_reboot_notifier);
3136 r = sysdev_class_register(&kvm_sysdev_class);
3140 r = sysdev_register(&kvm_sysdev);
3144 kvm_chardev_ops.owner = module;
3146 r = misc_register(&kvm_dev);
3148 printk (KERN_ERR "kvm: misc device register failed\n");
3155 sysdev_unregister(&kvm_sysdev);
3157 sysdev_class_unregister(&kvm_sysdev_class);
3159 unregister_reboot_notifier(&kvm_reboot_notifier);
3160 unregister_cpu_notifier(&kvm_cpu_notifier);
3162 on_each_cpu(hardware_disable, NULL, 0, 1);
3163 kvm_arch_ops->hardware_unsetup();
3165 kvm_arch_ops = NULL;
3169 void kvm_exit_arch(void)
3171 misc_deregister(&kvm_dev);
3172 sysdev_unregister(&kvm_sysdev);
3173 sysdev_class_unregister(&kvm_sysdev_class);
3174 unregister_reboot_notifier(&kvm_reboot_notifier);
3175 unregister_cpu_notifier(&kvm_cpu_notifier);
3176 on_each_cpu(hardware_disable, NULL, 0, 1);
3177 kvm_arch_ops->hardware_unsetup();
3178 kvm_arch_ops = NULL;
3181 static __init int kvm_init(void)
3183 static struct page *bad_page;
3186 r = kvm_mmu_module_init();
3192 kvm_init_msr_list();
3194 if ((bad_page = alloc_page(GFP_KERNEL)) == NULL) {
3199 bad_page_address = page_to_pfn(bad_page) << PAGE_SHIFT;
3200 memset(__va(bad_page_address), 0, PAGE_SIZE);
3206 kvm_mmu_module_exit();
3211 static __exit void kvm_exit(void)
3214 __free_page(pfn_to_page(bad_page_address >> PAGE_SHIFT));
3215 kvm_mmu_module_exit();
3218 module_init(kvm_init)
3219 module_exit(kvm_exit)
3221 EXPORT_SYMBOL_GPL(kvm_init_arch);
3222 EXPORT_SYMBOL_GPL(kvm_exit_arch);