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 __read_mostly struct preempt_ops kvm_preempt_ops;
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_RESERVED_BITS \
88 (~(unsigned long)(X86_CR0_PE | X86_CR0_MP | X86_CR0_EM | X86_CR0_TS \
89 | X86_CR0_ET | X86_CR0_NE | X86_CR0_WP | X86_CR0_AM \
90 | X86_CR0_NW | X86_CR0_CD | X86_CR0_PG))
91 #define CR4_RESERVED_BITS \
92 (~(unsigned long)(X86_CR4_VME | X86_CR4_PVI | X86_CR4_TSD | X86_CR4_DE\
93 | X86_CR4_PSE | X86_CR4_PAE | X86_CR4_MCE \
94 | X86_CR4_PGE | X86_CR4_PCE | X86_CR4_OSFXSR \
95 | X86_CR4_OSXMMEXCPT | X86_CR4_VMXE))
97 #define CR8_RESERVED_BITS (~(unsigned long)X86_CR8_TPR)
98 #define EFER_RESERVED_BITS 0xfffffffffffff2fe
101 // LDT or TSS descriptor in the GDT. 16 bytes.
102 struct segment_descriptor_64 {
103 struct segment_descriptor s;
110 static long kvm_vcpu_ioctl(struct file *file, unsigned int ioctl,
113 unsigned long segment_base(u16 selector)
115 struct descriptor_table gdt;
116 struct segment_descriptor *d;
117 unsigned long table_base;
118 typedef unsigned long ul;
124 asm ("sgdt %0" : "=m"(gdt));
125 table_base = gdt.base;
127 if (selector & 4) { /* from ldt */
130 asm ("sldt %0" : "=g"(ldt_selector));
131 table_base = segment_base(ldt_selector);
133 d = (struct segment_descriptor *)(table_base + (selector & ~7));
134 v = d->base_low | ((ul)d->base_mid << 16) | ((ul)d->base_high << 24);
137 && (d->type == 2 || d->type == 9 || d->type == 11))
138 v |= ((ul)((struct segment_descriptor_64 *)d)->base_higher) << 32;
142 EXPORT_SYMBOL_GPL(segment_base);
144 static inline int valid_vcpu(int n)
146 return likely(n >= 0 && n < KVM_MAX_VCPUS);
149 int kvm_read_guest(struct kvm_vcpu *vcpu, gva_t addr, unsigned long size,
152 unsigned char *host_buf = dest;
153 unsigned long req_size = size;
161 paddr = gva_to_hpa(vcpu, addr);
163 if (is_error_hpa(paddr))
166 guest_buf = (hva_t)kmap_atomic(
167 pfn_to_page(paddr >> PAGE_SHIFT),
169 offset = addr & ~PAGE_MASK;
171 now = min(size, PAGE_SIZE - offset);
172 memcpy(host_buf, (void*)guest_buf, now);
176 kunmap_atomic((void *)(guest_buf & PAGE_MASK), KM_USER0);
178 return req_size - size;
180 EXPORT_SYMBOL_GPL(kvm_read_guest);
182 int kvm_write_guest(struct kvm_vcpu *vcpu, gva_t addr, unsigned long size,
185 unsigned char *host_buf = data;
186 unsigned long req_size = size;
195 paddr = gva_to_hpa(vcpu, addr);
197 if (is_error_hpa(paddr))
200 gfn = vcpu->mmu.gva_to_gpa(vcpu, addr) >> PAGE_SHIFT;
201 mark_page_dirty(vcpu->kvm, gfn);
202 guest_buf = (hva_t)kmap_atomic(
203 pfn_to_page(paddr >> PAGE_SHIFT), KM_USER0);
204 offset = addr & ~PAGE_MASK;
206 now = min(size, PAGE_SIZE - offset);
207 memcpy((void*)guest_buf, host_buf, now);
211 kunmap_atomic((void *)(guest_buf & PAGE_MASK), KM_USER0);
213 return req_size - size;
215 EXPORT_SYMBOL_GPL(kvm_write_guest);
217 void kvm_load_guest_fpu(struct kvm_vcpu *vcpu)
219 if (!vcpu->fpu_active || vcpu->guest_fpu_loaded)
222 vcpu->guest_fpu_loaded = 1;
223 fx_save(vcpu->host_fx_image);
224 fx_restore(vcpu->guest_fx_image);
226 EXPORT_SYMBOL_GPL(kvm_load_guest_fpu);
228 void kvm_put_guest_fpu(struct kvm_vcpu *vcpu)
230 if (!vcpu->guest_fpu_loaded)
233 vcpu->guest_fpu_loaded = 0;
234 fx_save(vcpu->guest_fx_image);
235 fx_restore(vcpu->host_fx_image);
237 EXPORT_SYMBOL_GPL(kvm_put_guest_fpu);
240 * Switches to specified vcpu, until a matching vcpu_put()
242 static void vcpu_load(struct kvm_vcpu *vcpu)
246 mutex_lock(&vcpu->mutex);
248 preempt_notifier_register(&vcpu->preempt_notifier);
249 kvm_arch_ops->vcpu_load(vcpu, cpu);
253 static void vcpu_put(struct kvm_vcpu *vcpu)
256 kvm_arch_ops->vcpu_put(vcpu);
257 preempt_notifier_unregister(&vcpu->preempt_notifier);
259 mutex_unlock(&vcpu->mutex);
262 static void ack_flush(void *_completed)
264 atomic_t *completed = _completed;
266 atomic_inc(completed);
269 void kvm_flush_remote_tlbs(struct kvm *kvm)
273 struct kvm_vcpu *vcpu;
276 atomic_set(&completed, 0);
279 for (i = 0; i < KVM_MAX_VCPUS; ++i) {
280 vcpu = kvm->vcpus[i];
283 if (test_and_set_bit(KVM_TLB_FLUSH, &vcpu->requests))
286 if (cpu != -1 && cpu != raw_smp_processor_id())
287 if (!cpu_isset(cpu, cpus)) {
294 * We really want smp_call_function_mask() here. But that's not
295 * available, so ipi all cpus in parallel and wait for them
298 for (cpu = first_cpu(cpus); cpu != NR_CPUS; cpu = next_cpu(cpu, cpus))
299 smp_call_function_single(cpu, ack_flush, &completed, 1, 0);
300 while (atomic_read(&completed) != needed) {
306 int kvm_vcpu_init(struct kvm_vcpu *vcpu, struct kvm *kvm, unsigned id)
311 mutex_init(&vcpu->mutex);
313 vcpu->mmu.root_hpa = INVALID_PAGE;
317 page = alloc_page(GFP_KERNEL | __GFP_ZERO);
322 vcpu->run = page_address(page);
324 page = alloc_page(GFP_KERNEL | __GFP_ZERO);
329 vcpu->pio_data = page_address(page);
331 vcpu->host_fx_image = (char*)ALIGN((hva_t)vcpu->fx_buf,
333 vcpu->guest_fx_image = vcpu->host_fx_image + FX_IMAGE_SIZE;
335 r = kvm_mmu_create(vcpu);
337 goto fail_free_pio_data;
342 free_page((unsigned long)vcpu->pio_data);
344 free_page((unsigned long)vcpu->run);
348 EXPORT_SYMBOL_GPL(kvm_vcpu_init);
350 void kvm_vcpu_uninit(struct kvm_vcpu *vcpu)
352 kvm_mmu_destroy(vcpu);
353 free_page((unsigned long)vcpu->pio_data);
354 free_page((unsigned long)vcpu->run);
356 EXPORT_SYMBOL_GPL(kvm_vcpu_uninit);
358 static struct kvm *kvm_create_vm(void)
360 struct kvm *kvm = kzalloc(sizeof(struct kvm), GFP_KERNEL);
363 return ERR_PTR(-ENOMEM);
365 kvm_io_bus_init(&kvm->pio_bus);
366 mutex_init(&kvm->lock);
367 INIT_LIST_HEAD(&kvm->active_mmu_pages);
368 kvm_io_bus_init(&kvm->mmio_bus);
369 spin_lock(&kvm_lock);
370 list_add(&kvm->vm_list, &vm_list);
371 spin_unlock(&kvm_lock);
375 static int kvm_dev_open(struct inode *inode, struct file *filp)
381 * Free any memory in @free but not in @dont.
383 static void kvm_free_physmem_slot(struct kvm_memory_slot *free,
384 struct kvm_memory_slot *dont)
388 if (!dont || free->phys_mem != dont->phys_mem)
389 if (free->phys_mem) {
390 for (i = 0; i < free->npages; ++i)
391 if (free->phys_mem[i])
392 __free_page(free->phys_mem[i]);
393 vfree(free->phys_mem);
396 if (!dont || free->dirty_bitmap != dont->dirty_bitmap)
397 vfree(free->dirty_bitmap);
399 free->phys_mem = NULL;
401 free->dirty_bitmap = NULL;
404 static void kvm_free_physmem(struct kvm *kvm)
408 for (i = 0; i < kvm->nmemslots; ++i)
409 kvm_free_physmem_slot(&kvm->memslots[i], NULL);
412 static void free_pio_guest_pages(struct kvm_vcpu *vcpu)
416 for (i = 0; i < 2; ++i)
417 if (vcpu->pio.guest_pages[i]) {
418 __free_page(vcpu->pio.guest_pages[i]);
419 vcpu->pio.guest_pages[i] = NULL;
423 static void kvm_unload_vcpu_mmu(struct kvm_vcpu *vcpu)
426 kvm_mmu_unload(vcpu);
430 static void kvm_free_vcpus(struct kvm *kvm)
435 * Unpin any mmu pages first.
437 for (i = 0; i < KVM_MAX_VCPUS; ++i)
439 kvm_unload_vcpu_mmu(kvm->vcpus[i]);
440 for (i = 0; i < KVM_MAX_VCPUS; ++i) {
442 kvm_arch_ops->vcpu_free(kvm->vcpus[i]);
443 kvm->vcpus[i] = NULL;
449 static int kvm_dev_release(struct inode *inode, struct file *filp)
454 static void kvm_destroy_vm(struct kvm *kvm)
456 spin_lock(&kvm_lock);
457 list_del(&kvm->vm_list);
458 spin_unlock(&kvm_lock);
459 kvm_io_bus_destroy(&kvm->pio_bus);
460 kvm_io_bus_destroy(&kvm->mmio_bus);
462 kvm_free_physmem(kvm);
466 static int kvm_vm_release(struct inode *inode, struct file *filp)
468 struct kvm *kvm = filp->private_data;
474 static void inject_gp(struct kvm_vcpu *vcpu)
476 kvm_arch_ops->inject_gp(vcpu, 0);
480 * Load the pae pdptrs. Return true is they are all valid.
482 static int load_pdptrs(struct kvm_vcpu *vcpu, unsigned long cr3)
484 gfn_t pdpt_gfn = cr3 >> PAGE_SHIFT;
485 unsigned offset = ((cr3 & (PAGE_SIZE-1)) >> 5) << 2;
490 u64 pdpte[ARRAY_SIZE(vcpu->pdptrs)];
492 mutex_lock(&vcpu->kvm->lock);
493 page = gfn_to_page(vcpu->kvm, pdpt_gfn);
499 pdpt = kmap_atomic(page, KM_USER0);
500 memcpy(pdpte, pdpt+offset, sizeof(pdpte));
501 kunmap_atomic(pdpt, KM_USER0);
503 for (i = 0; i < ARRAY_SIZE(pdpte); ++i) {
504 if ((pdpte[i] & 1) && (pdpte[i] & 0xfffffff0000001e6ull)) {
511 memcpy(vcpu->pdptrs, pdpte, sizeof(vcpu->pdptrs));
513 mutex_unlock(&vcpu->kvm->lock);
518 void set_cr0(struct kvm_vcpu *vcpu, unsigned long cr0)
520 if (cr0 & CR0_RESERVED_BITS) {
521 printk(KERN_DEBUG "set_cr0: 0x%lx #GP, reserved bits 0x%lx\n",
527 if ((cr0 & X86_CR0_NW) && !(cr0 & X86_CR0_CD)) {
528 printk(KERN_DEBUG "set_cr0: #GP, CD == 0 && NW == 1\n");
533 if ((cr0 & X86_CR0_PG) && !(cr0 & X86_CR0_PE)) {
534 printk(KERN_DEBUG "set_cr0: #GP, set PG flag "
535 "and a clear PE flag\n");
540 if (!is_paging(vcpu) && (cr0 & X86_CR0_PG)) {
542 if ((vcpu->shadow_efer & EFER_LME)) {
546 printk(KERN_DEBUG "set_cr0: #GP, start paging "
547 "in long mode while PAE is disabled\n");
551 kvm_arch_ops->get_cs_db_l_bits(vcpu, &cs_db, &cs_l);
553 printk(KERN_DEBUG "set_cr0: #GP, start paging "
554 "in long mode while CS.L == 1\n");
561 if (is_pae(vcpu) && !load_pdptrs(vcpu, vcpu->cr3)) {
562 printk(KERN_DEBUG "set_cr0: #GP, pdptrs "
570 kvm_arch_ops->set_cr0(vcpu, cr0);
573 mutex_lock(&vcpu->kvm->lock);
574 kvm_mmu_reset_context(vcpu);
575 mutex_unlock(&vcpu->kvm->lock);
578 EXPORT_SYMBOL_GPL(set_cr0);
580 void lmsw(struct kvm_vcpu *vcpu, unsigned long msw)
582 set_cr0(vcpu, (vcpu->cr0 & ~0x0ful) | (msw & 0x0f));
584 EXPORT_SYMBOL_GPL(lmsw);
586 void set_cr4(struct kvm_vcpu *vcpu, unsigned long cr4)
588 if (cr4 & CR4_RESERVED_BITS) {
589 printk(KERN_DEBUG "set_cr4: #GP, reserved bits\n");
594 if (is_long_mode(vcpu)) {
595 if (!(cr4 & X86_CR4_PAE)) {
596 printk(KERN_DEBUG "set_cr4: #GP, clearing PAE while "
601 } else if (is_paging(vcpu) && !is_pae(vcpu) && (cr4 & X86_CR4_PAE)
602 && !load_pdptrs(vcpu, vcpu->cr3)) {
603 printk(KERN_DEBUG "set_cr4: #GP, pdptrs reserved bits\n");
608 if (cr4 & X86_CR4_VMXE) {
609 printk(KERN_DEBUG "set_cr4: #GP, setting VMXE\n");
613 kvm_arch_ops->set_cr4(vcpu, cr4);
614 mutex_lock(&vcpu->kvm->lock);
615 kvm_mmu_reset_context(vcpu);
616 mutex_unlock(&vcpu->kvm->lock);
618 EXPORT_SYMBOL_GPL(set_cr4);
620 void set_cr3(struct kvm_vcpu *vcpu, unsigned long cr3)
622 if (is_long_mode(vcpu)) {
623 if (cr3 & CR3_L_MODE_RESERVED_BITS) {
624 printk(KERN_DEBUG "set_cr3: #GP, reserved bits\n");
630 if (cr3 & CR3_PAE_RESERVED_BITS) {
632 "set_cr3: #GP, reserved bits\n");
636 if (is_paging(vcpu) && !load_pdptrs(vcpu, cr3)) {
637 printk(KERN_DEBUG "set_cr3: #GP, pdptrs "
643 if (cr3 & CR3_NONPAE_RESERVED_BITS) {
645 "set_cr3: #GP, reserved bits\n");
653 mutex_lock(&vcpu->kvm->lock);
655 * Does the new cr3 value map to physical memory? (Note, we
656 * catch an invalid cr3 even in real-mode, because it would
657 * cause trouble later on when we turn on paging anyway.)
659 * A real CPU would silently accept an invalid cr3 and would
660 * attempt to use it - with largely undefined (and often hard
661 * to debug) behavior on the guest side.
663 if (unlikely(!gfn_to_memslot(vcpu->kvm, cr3 >> PAGE_SHIFT)))
666 vcpu->mmu.new_cr3(vcpu);
667 mutex_unlock(&vcpu->kvm->lock);
669 EXPORT_SYMBOL_GPL(set_cr3);
671 void set_cr8(struct kvm_vcpu *vcpu, unsigned long cr8)
673 if (cr8 & CR8_RESERVED_BITS) {
674 printk(KERN_DEBUG "set_cr8: #GP, reserved bits 0x%lx\n", cr8);
680 EXPORT_SYMBOL_GPL(set_cr8);
682 void fx_init(struct kvm_vcpu *vcpu)
684 struct __attribute__ ((__packed__)) fx_image_s {
690 u64 operand;// fpu dp
696 /* Initialize guest FPU by resetting ours and saving into guest's */
698 fx_save(vcpu->host_fx_image);
700 fx_save(vcpu->guest_fx_image);
701 fx_restore(vcpu->host_fx_image);
704 fx_image = (struct fx_image_s *)vcpu->guest_fx_image;
705 fx_image->mxcsr = 0x1f80;
706 memset(vcpu->guest_fx_image + sizeof(struct fx_image_s),
707 0, FX_IMAGE_SIZE - sizeof(struct fx_image_s));
709 EXPORT_SYMBOL_GPL(fx_init);
712 * Allocate some memory and give it an address in the guest physical address
715 * Discontiguous memory is allowed, mostly for framebuffers.
717 static int kvm_vm_ioctl_set_memory_region(struct kvm *kvm,
718 struct kvm_memory_region *mem)
722 unsigned long npages;
724 struct kvm_memory_slot *memslot;
725 struct kvm_memory_slot old, new;
726 int memory_config_version;
729 /* General sanity checks */
730 if (mem->memory_size & (PAGE_SIZE - 1))
732 if (mem->guest_phys_addr & (PAGE_SIZE - 1))
734 if (mem->slot >= KVM_MEMORY_SLOTS)
736 if (mem->guest_phys_addr + mem->memory_size < mem->guest_phys_addr)
739 memslot = &kvm->memslots[mem->slot];
740 base_gfn = mem->guest_phys_addr >> PAGE_SHIFT;
741 npages = mem->memory_size >> PAGE_SHIFT;
744 mem->flags &= ~KVM_MEM_LOG_DIRTY_PAGES;
747 mutex_lock(&kvm->lock);
749 memory_config_version = kvm->memory_config_version;
750 new = old = *memslot;
752 new.base_gfn = base_gfn;
754 new.flags = mem->flags;
756 /* Disallow changing a memory slot's size. */
758 if (npages && old.npages && npages != old.npages)
761 /* Check for overlaps */
763 for (i = 0; i < KVM_MEMORY_SLOTS; ++i) {
764 struct kvm_memory_slot *s = &kvm->memslots[i];
768 if (!((base_gfn + npages <= s->base_gfn) ||
769 (base_gfn >= s->base_gfn + s->npages)))
773 * Do memory allocations outside lock. memory_config_version will
776 mutex_unlock(&kvm->lock);
778 /* Deallocate if slot is being removed */
782 /* Free page dirty bitmap if unneeded */
783 if (!(new.flags & KVM_MEM_LOG_DIRTY_PAGES))
784 new.dirty_bitmap = NULL;
788 /* Allocate if a slot is being created */
789 if (npages && !new.phys_mem) {
790 new.phys_mem = vmalloc(npages * sizeof(struct page *));
795 memset(new.phys_mem, 0, npages * sizeof(struct page *));
796 for (i = 0; i < npages; ++i) {
797 new.phys_mem[i] = alloc_page(GFP_HIGHUSER
799 if (!new.phys_mem[i])
801 set_page_private(new.phys_mem[i],0);
805 /* Allocate page dirty bitmap if needed */
806 if ((new.flags & KVM_MEM_LOG_DIRTY_PAGES) && !new.dirty_bitmap) {
807 unsigned dirty_bytes = ALIGN(npages, BITS_PER_LONG) / 8;
809 new.dirty_bitmap = vmalloc(dirty_bytes);
810 if (!new.dirty_bitmap)
812 memset(new.dirty_bitmap, 0, dirty_bytes);
815 mutex_lock(&kvm->lock);
817 if (memory_config_version != kvm->memory_config_version) {
818 mutex_unlock(&kvm->lock);
819 kvm_free_physmem_slot(&new, &old);
827 if (mem->slot >= kvm->nmemslots)
828 kvm->nmemslots = mem->slot + 1;
831 ++kvm->memory_config_version;
833 kvm_mmu_slot_remove_write_access(kvm, mem->slot);
834 kvm_flush_remote_tlbs(kvm);
836 mutex_unlock(&kvm->lock);
838 kvm_free_physmem_slot(&old, &new);
842 mutex_unlock(&kvm->lock);
844 kvm_free_physmem_slot(&new, &old);
850 * Get (and clear) the dirty memory log for a memory slot.
852 static int kvm_vm_ioctl_get_dirty_log(struct kvm *kvm,
853 struct kvm_dirty_log *log)
855 struct kvm_memory_slot *memslot;
858 unsigned long any = 0;
860 mutex_lock(&kvm->lock);
863 * Prevent changes to guest memory configuration even while the lock
867 mutex_unlock(&kvm->lock);
869 if (log->slot >= KVM_MEMORY_SLOTS)
872 memslot = &kvm->memslots[log->slot];
874 if (!memslot->dirty_bitmap)
877 n = ALIGN(memslot->npages, BITS_PER_LONG) / 8;
879 for (i = 0; !any && i < n/sizeof(long); ++i)
880 any = memslot->dirty_bitmap[i];
883 if (copy_to_user(log->dirty_bitmap, memslot->dirty_bitmap, n))
886 mutex_lock(&kvm->lock);
887 kvm_mmu_slot_remove_write_access(kvm, log->slot);
888 kvm_flush_remote_tlbs(kvm);
889 memset(memslot->dirty_bitmap, 0, n);
890 mutex_unlock(&kvm->lock);
895 mutex_lock(&kvm->lock);
897 mutex_unlock(&kvm->lock);
902 * Set a new alias region. Aliases map a portion of physical memory into
903 * another portion. This is useful for memory windows, for example the PC
906 static int kvm_vm_ioctl_set_memory_alias(struct kvm *kvm,
907 struct kvm_memory_alias *alias)
910 struct kvm_mem_alias *p;
913 /* General sanity checks */
914 if (alias->memory_size & (PAGE_SIZE - 1))
916 if (alias->guest_phys_addr & (PAGE_SIZE - 1))
918 if (alias->slot >= KVM_ALIAS_SLOTS)
920 if (alias->guest_phys_addr + alias->memory_size
921 < alias->guest_phys_addr)
923 if (alias->target_phys_addr + alias->memory_size
924 < alias->target_phys_addr)
927 mutex_lock(&kvm->lock);
929 p = &kvm->aliases[alias->slot];
930 p->base_gfn = alias->guest_phys_addr >> PAGE_SHIFT;
931 p->npages = alias->memory_size >> PAGE_SHIFT;
932 p->target_gfn = alias->target_phys_addr >> PAGE_SHIFT;
934 for (n = KVM_ALIAS_SLOTS; n > 0; --n)
935 if (kvm->aliases[n - 1].npages)
939 kvm_mmu_zap_all(kvm);
941 mutex_unlock(&kvm->lock);
949 static gfn_t unalias_gfn(struct kvm *kvm, gfn_t gfn)
952 struct kvm_mem_alias *alias;
954 for (i = 0; i < kvm->naliases; ++i) {
955 alias = &kvm->aliases[i];
956 if (gfn >= alias->base_gfn
957 && gfn < alias->base_gfn + alias->npages)
958 return alias->target_gfn + gfn - alias->base_gfn;
963 static struct kvm_memory_slot *__gfn_to_memslot(struct kvm *kvm, gfn_t gfn)
967 for (i = 0; i < kvm->nmemslots; ++i) {
968 struct kvm_memory_slot *memslot = &kvm->memslots[i];
970 if (gfn >= memslot->base_gfn
971 && gfn < memslot->base_gfn + memslot->npages)
977 struct kvm_memory_slot *gfn_to_memslot(struct kvm *kvm, gfn_t gfn)
979 gfn = unalias_gfn(kvm, gfn);
980 return __gfn_to_memslot(kvm, gfn);
983 struct page *gfn_to_page(struct kvm *kvm, gfn_t gfn)
985 struct kvm_memory_slot *slot;
987 gfn = unalias_gfn(kvm, gfn);
988 slot = __gfn_to_memslot(kvm, gfn);
991 return slot->phys_mem[gfn - slot->base_gfn];
993 EXPORT_SYMBOL_GPL(gfn_to_page);
995 void mark_page_dirty(struct kvm *kvm, gfn_t gfn)
998 struct kvm_memory_slot *memslot;
999 unsigned long rel_gfn;
1001 for (i = 0; i < kvm->nmemslots; ++i) {
1002 memslot = &kvm->memslots[i];
1004 if (gfn >= memslot->base_gfn
1005 && gfn < memslot->base_gfn + memslot->npages) {
1007 if (!memslot->dirty_bitmap)
1010 rel_gfn = gfn - memslot->base_gfn;
1013 if (!test_bit(rel_gfn, memslot->dirty_bitmap))
1014 set_bit(rel_gfn, memslot->dirty_bitmap);
1020 static int emulator_read_std(unsigned long addr,
1023 struct x86_emulate_ctxt *ctxt)
1025 struct kvm_vcpu *vcpu = ctxt->vcpu;
1029 gpa_t gpa = vcpu->mmu.gva_to_gpa(vcpu, addr);
1030 unsigned offset = addr & (PAGE_SIZE-1);
1031 unsigned tocopy = min(bytes, (unsigned)PAGE_SIZE - offset);
1036 if (gpa == UNMAPPED_GVA)
1037 return X86EMUL_PROPAGATE_FAULT;
1038 pfn = gpa >> PAGE_SHIFT;
1039 page = gfn_to_page(vcpu->kvm, pfn);
1041 return X86EMUL_UNHANDLEABLE;
1042 page_virt = kmap_atomic(page, KM_USER0);
1044 memcpy(data, page_virt + offset, tocopy);
1046 kunmap_atomic(page_virt, KM_USER0);
1053 return X86EMUL_CONTINUE;
1056 static int emulator_write_std(unsigned long addr,
1059 struct x86_emulate_ctxt *ctxt)
1061 printk(KERN_ERR "emulator_write_std: addr %lx n %d\n",
1063 return X86EMUL_UNHANDLEABLE;
1066 static struct kvm_io_device *vcpu_find_mmio_dev(struct kvm_vcpu *vcpu,
1070 * Note that its important to have this wrapper function because
1071 * in the very near future we will be checking for MMIOs against
1072 * the LAPIC as well as the general MMIO bus
1074 return kvm_io_bus_find_dev(&vcpu->kvm->mmio_bus, addr);
1077 static struct kvm_io_device *vcpu_find_pio_dev(struct kvm_vcpu *vcpu,
1080 return kvm_io_bus_find_dev(&vcpu->kvm->pio_bus, addr);
1083 static int emulator_read_emulated(unsigned long addr,
1086 struct x86_emulate_ctxt *ctxt)
1088 struct kvm_vcpu *vcpu = ctxt->vcpu;
1089 struct kvm_io_device *mmio_dev;
1092 if (vcpu->mmio_read_completed) {
1093 memcpy(val, vcpu->mmio_data, bytes);
1094 vcpu->mmio_read_completed = 0;
1095 return X86EMUL_CONTINUE;
1096 } else if (emulator_read_std(addr, val, bytes, ctxt)
1097 == X86EMUL_CONTINUE)
1098 return X86EMUL_CONTINUE;
1100 gpa = vcpu->mmu.gva_to_gpa(vcpu, addr);
1101 if (gpa == UNMAPPED_GVA)
1102 return X86EMUL_PROPAGATE_FAULT;
1105 * Is this MMIO handled locally?
1107 mmio_dev = vcpu_find_mmio_dev(vcpu, gpa);
1109 kvm_iodevice_read(mmio_dev, gpa, bytes, val);
1110 return X86EMUL_CONTINUE;
1113 vcpu->mmio_needed = 1;
1114 vcpu->mmio_phys_addr = gpa;
1115 vcpu->mmio_size = bytes;
1116 vcpu->mmio_is_write = 0;
1118 return X86EMUL_UNHANDLEABLE;
1121 static int emulator_write_phys(struct kvm_vcpu *vcpu, gpa_t gpa,
1122 const void *val, int bytes)
1127 if (((gpa + bytes - 1) >> PAGE_SHIFT) != (gpa >> PAGE_SHIFT))
1129 page = gfn_to_page(vcpu->kvm, gpa >> PAGE_SHIFT);
1132 mark_page_dirty(vcpu->kvm, gpa >> PAGE_SHIFT);
1133 virt = kmap_atomic(page, KM_USER0);
1134 kvm_mmu_pte_write(vcpu, gpa, val, bytes);
1135 memcpy(virt + offset_in_page(gpa), val, bytes);
1136 kunmap_atomic(virt, KM_USER0);
1140 static int emulator_write_emulated_onepage(unsigned long addr,
1143 struct x86_emulate_ctxt *ctxt)
1145 struct kvm_vcpu *vcpu = ctxt->vcpu;
1146 struct kvm_io_device *mmio_dev;
1147 gpa_t gpa = vcpu->mmu.gva_to_gpa(vcpu, addr);
1149 if (gpa == UNMAPPED_GVA) {
1150 kvm_arch_ops->inject_page_fault(vcpu, addr, 2);
1151 return X86EMUL_PROPAGATE_FAULT;
1154 if (emulator_write_phys(vcpu, gpa, val, bytes))
1155 return X86EMUL_CONTINUE;
1158 * Is this MMIO handled locally?
1160 mmio_dev = vcpu_find_mmio_dev(vcpu, gpa);
1162 kvm_iodevice_write(mmio_dev, gpa, bytes, val);
1163 return X86EMUL_CONTINUE;
1166 vcpu->mmio_needed = 1;
1167 vcpu->mmio_phys_addr = gpa;
1168 vcpu->mmio_size = bytes;
1169 vcpu->mmio_is_write = 1;
1170 memcpy(vcpu->mmio_data, val, bytes);
1172 return X86EMUL_CONTINUE;
1175 static int emulator_write_emulated(unsigned long addr,
1178 struct x86_emulate_ctxt *ctxt)
1180 /* Crossing a page boundary? */
1181 if (((addr + bytes - 1) ^ addr) & PAGE_MASK) {
1184 now = -addr & ~PAGE_MASK;
1185 rc = emulator_write_emulated_onepage(addr, val, now, ctxt);
1186 if (rc != X86EMUL_CONTINUE)
1192 return emulator_write_emulated_onepage(addr, val, bytes, ctxt);
1195 static int emulator_cmpxchg_emulated(unsigned long addr,
1199 struct x86_emulate_ctxt *ctxt)
1201 static int reported;
1205 printk(KERN_WARNING "kvm: emulating exchange as write\n");
1207 return emulator_write_emulated(addr, new, bytes, ctxt);
1210 static unsigned long get_segment_base(struct kvm_vcpu *vcpu, int seg)
1212 return kvm_arch_ops->get_segment_base(vcpu, seg);
1215 int emulate_invlpg(struct kvm_vcpu *vcpu, gva_t address)
1217 return X86EMUL_CONTINUE;
1220 int emulate_clts(struct kvm_vcpu *vcpu)
1224 cr0 = vcpu->cr0 & ~X86_CR0_TS;
1225 kvm_arch_ops->set_cr0(vcpu, cr0);
1226 return X86EMUL_CONTINUE;
1229 int emulator_get_dr(struct x86_emulate_ctxt* ctxt, int dr, unsigned long *dest)
1231 struct kvm_vcpu *vcpu = ctxt->vcpu;
1235 *dest = kvm_arch_ops->get_dr(vcpu, dr);
1236 return X86EMUL_CONTINUE;
1238 printk(KERN_DEBUG "%s: unexpected dr %u\n",
1240 return X86EMUL_UNHANDLEABLE;
1244 int emulator_set_dr(struct x86_emulate_ctxt *ctxt, int dr, unsigned long value)
1246 unsigned long mask = (ctxt->mode == X86EMUL_MODE_PROT64) ? ~0ULL : ~0U;
1249 kvm_arch_ops->set_dr(ctxt->vcpu, dr, value & mask, &exception);
1251 /* FIXME: better handling */
1252 return X86EMUL_UNHANDLEABLE;
1254 return X86EMUL_CONTINUE;
1257 static void report_emulation_failure(struct x86_emulate_ctxt *ctxt)
1259 static int reported;
1261 unsigned long rip = ctxt->vcpu->rip;
1262 unsigned long rip_linear;
1264 rip_linear = rip + get_segment_base(ctxt->vcpu, VCPU_SREG_CS);
1269 emulator_read_std(rip_linear, (void *)opcodes, 4, ctxt);
1271 printk(KERN_ERR "emulation failed but !mmio_needed?"
1272 " rip %lx %02x %02x %02x %02x\n",
1273 rip, opcodes[0], opcodes[1], opcodes[2], opcodes[3]);
1277 struct x86_emulate_ops emulate_ops = {
1278 .read_std = emulator_read_std,
1279 .write_std = emulator_write_std,
1280 .read_emulated = emulator_read_emulated,
1281 .write_emulated = emulator_write_emulated,
1282 .cmpxchg_emulated = emulator_cmpxchg_emulated,
1285 int emulate_instruction(struct kvm_vcpu *vcpu,
1286 struct kvm_run *run,
1290 struct x86_emulate_ctxt emulate_ctxt;
1294 vcpu->mmio_fault_cr2 = cr2;
1295 kvm_arch_ops->cache_regs(vcpu);
1297 kvm_arch_ops->get_cs_db_l_bits(vcpu, &cs_db, &cs_l);
1299 emulate_ctxt.vcpu = vcpu;
1300 emulate_ctxt.eflags = kvm_arch_ops->get_rflags(vcpu);
1301 emulate_ctxt.cr2 = cr2;
1302 emulate_ctxt.mode = (emulate_ctxt.eflags & X86_EFLAGS_VM)
1303 ? X86EMUL_MODE_REAL : cs_l
1304 ? X86EMUL_MODE_PROT64 : cs_db
1305 ? X86EMUL_MODE_PROT32 : X86EMUL_MODE_PROT16;
1307 if (emulate_ctxt.mode == X86EMUL_MODE_PROT64) {
1308 emulate_ctxt.cs_base = 0;
1309 emulate_ctxt.ds_base = 0;
1310 emulate_ctxt.es_base = 0;
1311 emulate_ctxt.ss_base = 0;
1313 emulate_ctxt.cs_base = get_segment_base(vcpu, VCPU_SREG_CS);
1314 emulate_ctxt.ds_base = get_segment_base(vcpu, VCPU_SREG_DS);
1315 emulate_ctxt.es_base = get_segment_base(vcpu, VCPU_SREG_ES);
1316 emulate_ctxt.ss_base = get_segment_base(vcpu, VCPU_SREG_SS);
1319 emulate_ctxt.gs_base = get_segment_base(vcpu, VCPU_SREG_GS);
1320 emulate_ctxt.fs_base = get_segment_base(vcpu, VCPU_SREG_FS);
1322 vcpu->mmio_is_write = 0;
1323 r = x86_emulate_memop(&emulate_ctxt, &emulate_ops);
1325 if ((r || vcpu->mmio_is_write) && run) {
1326 run->exit_reason = KVM_EXIT_MMIO;
1327 run->mmio.phys_addr = vcpu->mmio_phys_addr;
1328 memcpy(run->mmio.data, vcpu->mmio_data, 8);
1329 run->mmio.len = vcpu->mmio_size;
1330 run->mmio.is_write = vcpu->mmio_is_write;
1334 if (kvm_mmu_unprotect_page_virt(vcpu, cr2))
1335 return EMULATE_DONE;
1336 if (!vcpu->mmio_needed) {
1337 report_emulation_failure(&emulate_ctxt);
1338 return EMULATE_FAIL;
1340 return EMULATE_DO_MMIO;
1343 kvm_arch_ops->decache_regs(vcpu);
1344 kvm_arch_ops->set_rflags(vcpu, emulate_ctxt.eflags);
1346 if (vcpu->mmio_is_write) {
1347 vcpu->mmio_needed = 0;
1348 return EMULATE_DO_MMIO;
1351 return EMULATE_DONE;
1353 EXPORT_SYMBOL_GPL(emulate_instruction);
1355 int kvm_emulate_halt(struct kvm_vcpu *vcpu)
1357 if (vcpu->irq_summary)
1360 vcpu->run->exit_reason = KVM_EXIT_HLT;
1361 ++vcpu->stat.halt_exits;
1364 EXPORT_SYMBOL_GPL(kvm_emulate_halt);
1366 int kvm_hypercall(struct kvm_vcpu *vcpu, struct kvm_run *run)
1368 unsigned long nr, a0, a1, a2, a3, a4, a5, ret;
1370 kvm_arch_ops->cache_regs(vcpu);
1372 #ifdef CONFIG_X86_64
1373 if (is_long_mode(vcpu)) {
1374 nr = vcpu->regs[VCPU_REGS_RAX];
1375 a0 = vcpu->regs[VCPU_REGS_RDI];
1376 a1 = vcpu->regs[VCPU_REGS_RSI];
1377 a2 = vcpu->regs[VCPU_REGS_RDX];
1378 a3 = vcpu->regs[VCPU_REGS_RCX];
1379 a4 = vcpu->regs[VCPU_REGS_R8];
1380 a5 = vcpu->regs[VCPU_REGS_R9];
1384 nr = vcpu->regs[VCPU_REGS_RBX] & -1u;
1385 a0 = vcpu->regs[VCPU_REGS_RAX] & -1u;
1386 a1 = vcpu->regs[VCPU_REGS_RCX] & -1u;
1387 a2 = vcpu->regs[VCPU_REGS_RDX] & -1u;
1388 a3 = vcpu->regs[VCPU_REGS_RSI] & -1u;
1389 a4 = vcpu->regs[VCPU_REGS_RDI] & -1u;
1390 a5 = vcpu->regs[VCPU_REGS_RBP] & -1u;
1394 run->hypercall.nr = nr;
1395 run->hypercall.args[0] = a0;
1396 run->hypercall.args[1] = a1;
1397 run->hypercall.args[2] = a2;
1398 run->hypercall.args[3] = a3;
1399 run->hypercall.args[4] = a4;
1400 run->hypercall.args[5] = a5;
1401 run->hypercall.ret = ret;
1402 run->hypercall.longmode = is_long_mode(vcpu);
1403 kvm_arch_ops->decache_regs(vcpu);
1406 vcpu->regs[VCPU_REGS_RAX] = ret;
1407 kvm_arch_ops->decache_regs(vcpu);
1410 EXPORT_SYMBOL_GPL(kvm_hypercall);
1412 static u64 mk_cr_64(u64 curr_cr, u32 new_val)
1414 return (curr_cr & ~((1ULL << 32) - 1)) | new_val;
1417 void realmode_lgdt(struct kvm_vcpu *vcpu, u16 limit, unsigned long base)
1419 struct descriptor_table dt = { limit, base };
1421 kvm_arch_ops->set_gdt(vcpu, &dt);
1424 void realmode_lidt(struct kvm_vcpu *vcpu, u16 limit, unsigned long base)
1426 struct descriptor_table dt = { limit, base };
1428 kvm_arch_ops->set_idt(vcpu, &dt);
1431 void realmode_lmsw(struct kvm_vcpu *vcpu, unsigned long msw,
1432 unsigned long *rflags)
1435 *rflags = kvm_arch_ops->get_rflags(vcpu);
1438 unsigned long realmode_get_cr(struct kvm_vcpu *vcpu, int cr)
1440 kvm_arch_ops->decache_cr4_guest_bits(vcpu);
1451 vcpu_printf(vcpu, "%s: unexpected cr %u\n", __FUNCTION__, cr);
1456 void realmode_set_cr(struct kvm_vcpu *vcpu, int cr, unsigned long val,
1457 unsigned long *rflags)
1461 set_cr0(vcpu, mk_cr_64(vcpu->cr0, val));
1462 *rflags = kvm_arch_ops->get_rflags(vcpu);
1471 set_cr4(vcpu, mk_cr_64(vcpu->cr4, val));
1474 vcpu_printf(vcpu, "%s: unexpected cr %u\n", __FUNCTION__, cr);
1479 * Register the para guest with the host:
1481 static int vcpu_register_para(struct kvm_vcpu *vcpu, gpa_t para_state_gpa)
1483 struct kvm_vcpu_para_state *para_state;
1484 hpa_t para_state_hpa, hypercall_hpa;
1485 struct page *para_state_page;
1486 unsigned char *hypercall;
1487 gpa_t hypercall_gpa;
1489 printk(KERN_DEBUG "kvm: guest trying to enter paravirtual mode\n");
1490 printk(KERN_DEBUG ".... para_state_gpa: %08Lx\n", para_state_gpa);
1493 * Needs to be page aligned:
1495 if (para_state_gpa != PAGE_ALIGN(para_state_gpa))
1498 para_state_hpa = gpa_to_hpa(vcpu, para_state_gpa);
1499 printk(KERN_DEBUG ".... para_state_hpa: %08Lx\n", para_state_hpa);
1500 if (is_error_hpa(para_state_hpa))
1503 mark_page_dirty(vcpu->kvm, para_state_gpa >> PAGE_SHIFT);
1504 para_state_page = pfn_to_page(para_state_hpa >> PAGE_SHIFT);
1505 para_state = kmap(para_state_page);
1507 printk(KERN_DEBUG ".... guest version: %d\n", para_state->guest_version);
1508 printk(KERN_DEBUG ".... size: %d\n", para_state->size);
1510 para_state->host_version = KVM_PARA_API_VERSION;
1512 * We cannot support guests that try to register themselves
1513 * with a newer API version than the host supports:
1515 if (para_state->guest_version > KVM_PARA_API_VERSION) {
1516 para_state->ret = -KVM_EINVAL;
1517 goto err_kunmap_skip;
1520 hypercall_gpa = para_state->hypercall_gpa;
1521 hypercall_hpa = gpa_to_hpa(vcpu, hypercall_gpa);
1522 printk(KERN_DEBUG ".... hypercall_hpa: %08Lx\n", hypercall_hpa);
1523 if (is_error_hpa(hypercall_hpa)) {
1524 para_state->ret = -KVM_EINVAL;
1525 goto err_kunmap_skip;
1528 printk(KERN_DEBUG "kvm: para guest successfully registered.\n");
1529 vcpu->para_state_page = para_state_page;
1530 vcpu->para_state_gpa = para_state_gpa;
1531 vcpu->hypercall_gpa = hypercall_gpa;
1533 mark_page_dirty(vcpu->kvm, hypercall_gpa >> PAGE_SHIFT);
1534 hypercall = kmap_atomic(pfn_to_page(hypercall_hpa >> PAGE_SHIFT),
1535 KM_USER1) + (hypercall_hpa & ~PAGE_MASK);
1536 kvm_arch_ops->patch_hypercall(vcpu, hypercall);
1537 kunmap_atomic(hypercall, KM_USER1);
1539 para_state->ret = 0;
1541 kunmap(para_state_page);
1547 int kvm_get_msr_common(struct kvm_vcpu *vcpu, u32 msr, u64 *pdata)
1552 case 0xc0010010: /* SYSCFG */
1553 case 0xc0010015: /* HWCR */
1554 case MSR_IA32_PLATFORM_ID:
1555 case MSR_IA32_P5_MC_ADDR:
1556 case MSR_IA32_P5_MC_TYPE:
1557 case MSR_IA32_MC0_CTL:
1558 case MSR_IA32_MCG_STATUS:
1559 case MSR_IA32_MCG_CAP:
1560 case MSR_IA32_MC0_MISC:
1561 case MSR_IA32_MC0_MISC+4:
1562 case MSR_IA32_MC0_MISC+8:
1563 case MSR_IA32_MC0_MISC+12:
1564 case MSR_IA32_MC0_MISC+16:
1565 case MSR_IA32_UCODE_REV:
1566 case MSR_IA32_PERF_STATUS:
1567 case MSR_IA32_EBL_CR_POWERON:
1568 /* MTRR registers */
1570 case 0x200 ... 0x2ff:
1573 case 0xcd: /* fsb frequency */
1576 case MSR_IA32_APICBASE:
1577 data = vcpu->apic_base;
1579 case MSR_IA32_MISC_ENABLE:
1580 data = vcpu->ia32_misc_enable_msr;
1582 #ifdef CONFIG_X86_64
1584 data = vcpu->shadow_efer;
1588 printk(KERN_ERR "kvm: unhandled rdmsr: 0x%x\n", msr);
1594 EXPORT_SYMBOL_GPL(kvm_get_msr_common);
1597 * Reads an msr value (of 'msr_index') into 'pdata'.
1598 * Returns 0 on success, non-0 otherwise.
1599 * Assumes vcpu_load() was already called.
1601 int kvm_get_msr(struct kvm_vcpu *vcpu, u32 msr_index, u64 *pdata)
1603 return kvm_arch_ops->get_msr(vcpu, msr_index, pdata);
1606 #ifdef CONFIG_X86_64
1608 static void set_efer(struct kvm_vcpu *vcpu, u64 efer)
1610 if (efer & EFER_RESERVED_BITS) {
1611 printk(KERN_DEBUG "set_efer: 0x%llx #GP, reserved bits\n",
1618 && (vcpu->shadow_efer & EFER_LME) != (efer & EFER_LME)) {
1619 printk(KERN_DEBUG "set_efer: #GP, change LME while paging\n");
1624 kvm_arch_ops->set_efer(vcpu, efer);
1627 efer |= vcpu->shadow_efer & EFER_LMA;
1629 vcpu->shadow_efer = efer;
1634 int kvm_set_msr_common(struct kvm_vcpu *vcpu, u32 msr, u64 data)
1637 #ifdef CONFIG_X86_64
1639 set_efer(vcpu, data);
1642 case MSR_IA32_MC0_STATUS:
1643 printk(KERN_WARNING "%s: MSR_IA32_MC0_STATUS 0x%llx, nop\n",
1644 __FUNCTION__, data);
1646 case MSR_IA32_MCG_STATUS:
1647 printk(KERN_WARNING "%s: MSR_IA32_MCG_STATUS 0x%llx, nop\n",
1648 __FUNCTION__, data);
1650 case MSR_IA32_UCODE_REV:
1651 case MSR_IA32_UCODE_WRITE:
1652 case 0x200 ... 0x2ff: /* MTRRs */
1654 case MSR_IA32_APICBASE:
1655 vcpu->apic_base = data;
1657 case MSR_IA32_MISC_ENABLE:
1658 vcpu->ia32_misc_enable_msr = data;
1661 * This is the 'probe whether the host is KVM' logic:
1663 case MSR_KVM_API_MAGIC:
1664 return vcpu_register_para(vcpu, data);
1667 printk(KERN_ERR "kvm: unhandled wrmsr: 0x%x\n", msr);
1672 EXPORT_SYMBOL_GPL(kvm_set_msr_common);
1675 * Writes msr value into into the appropriate "register".
1676 * Returns 0 on success, non-0 otherwise.
1677 * Assumes vcpu_load() was already called.
1679 int kvm_set_msr(struct kvm_vcpu *vcpu, u32 msr_index, u64 data)
1681 return kvm_arch_ops->set_msr(vcpu, msr_index, data);
1684 void kvm_resched(struct kvm_vcpu *vcpu)
1686 if (!need_resched())
1690 EXPORT_SYMBOL_GPL(kvm_resched);
1692 void kvm_emulate_cpuid(struct kvm_vcpu *vcpu)
1696 struct kvm_cpuid_entry *e, *best;
1698 kvm_arch_ops->cache_regs(vcpu);
1699 function = vcpu->regs[VCPU_REGS_RAX];
1700 vcpu->regs[VCPU_REGS_RAX] = 0;
1701 vcpu->regs[VCPU_REGS_RBX] = 0;
1702 vcpu->regs[VCPU_REGS_RCX] = 0;
1703 vcpu->regs[VCPU_REGS_RDX] = 0;
1705 for (i = 0; i < vcpu->cpuid_nent; ++i) {
1706 e = &vcpu->cpuid_entries[i];
1707 if (e->function == function) {
1712 * Both basic or both extended?
1714 if (((e->function ^ function) & 0x80000000) == 0)
1715 if (!best || e->function > best->function)
1719 vcpu->regs[VCPU_REGS_RAX] = best->eax;
1720 vcpu->regs[VCPU_REGS_RBX] = best->ebx;
1721 vcpu->regs[VCPU_REGS_RCX] = best->ecx;
1722 vcpu->regs[VCPU_REGS_RDX] = best->edx;
1724 kvm_arch_ops->decache_regs(vcpu);
1725 kvm_arch_ops->skip_emulated_instruction(vcpu);
1727 EXPORT_SYMBOL_GPL(kvm_emulate_cpuid);
1729 static int pio_copy_data(struct kvm_vcpu *vcpu)
1731 void *p = vcpu->pio_data;
1734 int nr_pages = vcpu->pio.guest_pages[1] ? 2 : 1;
1736 q = vmap(vcpu->pio.guest_pages, nr_pages, VM_READ|VM_WRITE,
1739 free_pio_guest_pages(vcpu);
1742 q += vcpu->pio.guest_page_offset;
1743 bytes = vcpu->pio.size * vcpu->pio.cur_count;
1745 memcpy(q, p, bytes);
1747 memcpy(p, q, bytes);
1748 q -= vcpu->pio.guest_page_offset;
1750 free_pio_guest_pages(vcpu);
1754 static int complete_pio(struct kvm_vcpu *vcpu)
1756 struct kvm_pio_request *io = &vcpu->pio;
1760 kvm_arch_ops->cache_regs(vcpu);
1764 memcpy(&vcpu->regs[VCPU_REGS_RAX], vcpu->pio_data,
1768 r = pio_copy_data(vcpu);
1770 kvm_arch_ops->cache_regs(vcpu);
1777 delta *= io->cur_count;
1779 * The size of the register should really depend on
1780 * current address size.
1782 vcpu->regs[VCPU_REGS_RCX] -= delta;
1788 vcpu->regs[VCPU_REGS_RDI] += delta;
1790 vcpu->regs[VCPU_REGS_RSI] += delta;
1793 kvm_arch_ops->decache_regs(vcpu);
1795 io->count -= io->cur_count;
1799 kvm_arch_ops->skip_emulated_instruction(vcpu);
1803 static void kernel_pio(struct kvm_io_device *pio_dev,
1804 struct kvm_vcpu *vcpu,
1807 /* TODO: String I/O for in kernel device */
1810 kvm_iodevice_read(pio_dev, vcpu->pio.port,
1814 kvm_iodevice_write(pio_dev, vcpu->pio.port,
1819 static void pio_string_write(struct kvm_io_device *pio_dev,
1820 struct kvm_vcpu *vcpu)
1822 struct kvm_pio_request *io = &vcpu->pio;
1823 void *pd = vcpu->pio_data;
1826 for (i = 0; i < io->cur_count; i++) {
1827 kvm_iodevice_write(pio_dev, io->port,
1834 int kvm_setup_pio(struct kvm_vcpu *vcpu, struct kvm_run *run, int in,
1835 int size, unsigned long count, int string, int down,
1836 gva_t address, int rep, unsigned port)
1838 unsigned now, in_page;
1842 struct kvm_io_device *pio_dev;
1844 vcpu->run->exit_reason = KVM_EXIT_IO;
1845 vcpu->run->io.direction = in ? KVM_EXIT_IO_IN : KVM_EXIT_IO_OUT;
1846 vcpu->run->io.size = size;
1847 vcpu->run->io.data_offset = KVM_PIO_PAGE_OFFSET * PAGE_SIZE;
1848 vcpu->run->io.count = count;
1849 vcpu->run->io.port = port;
1850 vcpu->pio.count = count;
1851 vcpu->pio.cur_count = count;
1852 vcpu->pio.size = size;
1854 vcpu->pio.port = port;
1855 vcpu->pio.string = string;
1856 vcpu->pio.down = down;
1857 vcpu->pio.guest_page_offset = offset_in_page(address);
1858 vcpu->pio.rep = rep;
1860 pio_dev = vcpu_find_pio_dev(vcpu, port);
1862 kvm_arch_ops->cache_regs(vcpu);
1863 memcpy(vcpu->pio_data, &vcpu->regs[VCPU_REGS_RAX], 4);
1864 kvm_arch_ops->decache_regs(vcpu);
1866 kernel_pio(pio_dev, vcpu, vcpu->pio_data);
1874 kvm_arch_ops->skip_emulated_instruction(vcpu);
1878 now = min(count, PAGE_SIZE / size);
1881 in_page = PAGE_SIZE - offset_in_page(address);
1883 in_page = offset_in_page(address) + size;
1884 now = min(count, (unsigned long)in_page / size);
1887 * String I/O straddles page boundary. Pin two guest pages
1888 * so that we satisfy atomicity constraints. Do just one
1889 * transaction to avoid complexity.
1896 * String I/O in reverse. Yuck. Kill the guest, fix later.
1898 printk(KERN_ERR "kvm: guest string pio down\n");
1902 vcpu->run->io.count = now;
1903 vcpu->pio.cur_count = now;
1905 for (i = 0; i < nr_pages; ++i) {
1906 mutex_lock(&vcpu->kvm->lock);
1907 page = gva_to_page(vcpu, address + i * PAGE_SIZE);
1910 vcpu->pio.guest_pages[i] = page;
1911 mutex_unlock(&vcpu->kvm->lock);
1914 free_pio_guest_pages(vcpu);
1919 if (!vcpu->pio.in) {
1920 /* string PIO write */
1921 ret = pio_copy_data(vcpu);
1922 if (ret >= 0 && pio_dev) {
1923 pio_string_write(pio_dev, vcpu);
1925 if (vcpu->pio.count == 0)
1929 printk(KERN_ERR "no string pio read support yet, "
1930 "port %x size %d count %ld\n",
1935 EXPORT_SYMBOL_GPL(kvm_setup_pio);
1937 static int kvm_vcpu_ioctl_run(struct kvm_vcpu *vcpu, struct kvm_run *kvm_run)
1944 if (vcpu->sigset_active)
1945 sigprocmask(SIG_SETMASK, &vcpu->sigset, &sigsaved);
1947 /* re-sync apic's tpr */
1948 vcpu->cr8 = kvm_run->cr8;
1950 if (vcpu->pio.cur_count) {
1951 r = complete_pio(vcpu);
1956 if (vcpu->mmio_needed) {
1957 memcpy(vcpu->mmio_data, kvm_run->mmio.data, 8);
1958 vcpu->mmio_read_completed = 1;
1959 vcpu->mmio_needed = 0;
1960 r = emulate_instruction(vcpu, kvm_run,
1961 vcpu->mmio_fault_cr2, 0);
1962 if (r == EMULATE_DO_MMIO) {
1964 * Read-modify-write. Back to userspace.
1971 if (kvm_run->exit_reason == KVM_EXIT_HYPERCALL) {
1972 kvm_arch_ops->cache_regs(vcpu);
1973 vcpu->regs[VCPU_REGS_RAX] = kvm_run->hypercall.ret;
1974 kvm_arch_ops->decache_regs(vcpu);
1977 r = kvm_arch_ops->run(vcpu, kvm_run);
1980 if (vcpu->sigset_active)
1981 sigprocmask(SIG_SETMASK, &sigsaved, NULL);
1987 static int kvm_vcpu_ioctl_get_regs(struct kvm_vcpu *vcpu,
1988 struct kvm_regs *regs)
1992 kvm_arch_ops->cache_regs(vcpu);
1994 regs->rax = vcpu->regs[VCPU_REGS_RAX];
1995 regs->rbx = vcpu->regs[VCPU_REGS_RBX];
1996 regs->rcx = vcpu->regs[VCPU_REGS_RCX];
1997 regs->rdx = vcpu->regs[VCPU_REGS_RDX];
1998 regs->rsi = vcpu->regs[VCPU_REGS_RSI];
1999 regs->rdi = vcpu->regs[VCPU_REGS_RDI];
2000 regs->rsp = vcpu->regs[VCPU_REGS_RSP];
2001 regs->rbp = vcpu->regs[VCPU_REGS_RBP];
2002 #ifdef CONFIG_X86_64
2003 regs->r8 = vcpu->regs[VCPU_REGS_R8];
2004 regs->r9 = vcpu->regs[VCPU_REGS_R9];
2005 regs->r10 = vcpu->regs[VCPU_REGS_R10];
2006 regs->r11 = vcpu->regs[VCPU_REGS_R11];
2007 regs->r12 = vcpu->regs[VCPU_REGS_R12];
2008 regs->r13 = vcpu->regs[VCPU_REGS_R13];
2009 regs->r14 = vcpu->regs[VCPU_REGS_R14];
2010 regs->r15 = vcpu->regs[VCPU_REGS_R15];
2013 regs->rip = vcpu->rip;
2014 regs->rflags = kvm_arch_ops->get_rflags(vcpu);
2017 * Don't leak debug flags in case they were set for guest debugging
2019 if (vcpu->guest_debug.enabled && vcpu->guest_debug.singlestep)
2020 regs->rflags &= ~(X86_EFLAGS_TF | X86_EFLAGS_RF);
2027 static int kvm_vcpu_ioctl_set_regs(struct kvm_vcpu *vcpu,
2028 struct kvm_regs *regs)
2032 vcpu->regs[VCPU_REGS_RAX] = regs->rax;
2033 vcpu->regs[VCPU_REGS_RBX] = regs->rbx;
2034 vcpu->regs[VCPU_REGS_RCX] = regs->rcx;
2035 vcpu->regs[VCPU_REGS_RDX] = regs->rdx;
2036 vcpu->regs[VCPU_REGS_RSI] = regs->rsi;
2037 vcpu->regs[VCPU_REGS_RDI] = regs->rdi;
2038 vcpu->regs[VCPU_REGS_RSP] = regs->rsp;
2039 vcpu->regs[VCPU_REGS_RBP] = regs->rbp;
2040 #ifdef CONFIG_X86_64
2041 vcpu->regs[VCPU_REGS_R8] = regs->r8;
2042 vcpu->regs[VCPU_REGS_R9] = regs->r9;
2043 vcpu->regs[VCPU_REGS_R10] = regs->r10;
2044 vcpu->regs[VCPU_REGS_R11] = regs->r11;
2045 vcpu->regs[VCPU_REGS_R12] = regs->r12;
2046 vcpu->regs[VCPU_REGS_R13] = regs->r13;
2047 vcpu->regs[VCPU_REGS_R14] = regs->r14;
2048 vcpu->regs[VCPU_REGS_R15] = regs->r15;
2051 vcpu->rip = regs->rip;
2052 kvm_arch_ops->set_rflags(vcpu, regs->rflags);
2054 kvm_arch_ops->decache_regs(vcpu);
2061 static void get_segment(struct kvm_vcpu *vcpu,
2062 struct kvm_segment *var, int seg)
2064 return kvm_arch_ops->get_segment(vcpu, var, seg);
2067 static int kvm_vcpu_ioctl_get_sregs(struct kvm_vcpu *vcpu,
2068 struct kvm_sregs *sregs)
2070 struct descriptor_table dt;
2074 get_segment(vcpu, &sregs->cs, VCPU_SREG_CS);
2075 get_segment(vcpu, &sregs->ds, VCPU_SREG_DS);
2076 get_segment(vcpu, &sregs->es, VCPU_SREG_ES);
2077 get_segment(vcpu, &sregs->fs, VCPU_SREG_FS);
2078 get_segment(vcpu, &sregs->gs, VCPU_SREG_GS);
2079 get_segment(vcpu, &sregs->ss, VCPU_SREG_SS);
2081 get_segment(vcpu, &sregs->tr, VCPU_SREG_TR);
2082 get_segment(vcpu, &sregs->ldt, VCPU_SREG_LDTR);
2084 kvm_arch_ops->get_idt(vcpu, &dt);
2085 sregs->idt.limit = dt.limit;
2086 sregs->idt.base = dt.base;
2087 kvm_arch_ops->get_gdt(vcpu, &dt);
2088 sregs->gdt.limit = dt.limit;
2089 sregs->gdt.base = dt.base;
2091 kvm_arch_ops->decache_cr4_guest_bits(vcpu);
2092 sregs->cr0 = vcpu->cr0;
2093 sregs->cr2 = vcpu->cr2;
2094 sregs->cr3 = vcpu->cr3;
2095 sregs->cr4 = vcpu->cr4;
2096 sregs->cr8 = vcpu->cr8;
2097 sregs->efer = vcpu->shadow_efer;
2098 sregs->apic_base = vcpu->apic_base;
2100 memcpy(sregs->interrupt_bitmap, vcpu->irq_pending,
2101 sizeof sregs->interrupt_bitmap);
2108 static void set_segment(struct kvm_vcpu *vcpu,
2109 struct kvm_segment *var, int seg)
2111 return kvm_arch_ops->set_segment(vcpu, var, seg);
2114 static int kvm_vcpu_ioctl_set_sregs(struct kvm_vcpu *vcpu,
2115 struct kvm_sregs *sregs)
2117 int mmu_reset_needed = 0;
2119 struct descriptor_table dt;
2123 dt.limit = sregs->idt.limit;
2124 dt.base = sregs->idt.base;
2125 kvm_arch_ops->set_idt(vcpu, &dt);
2126 dt.limit = sregs->gdt.limit;
2127 dt.base = sregs->gdt.base;
2128 kvm_arch_ops->set_gdt(vcpu, &dt);
2130 vcpu->cr2 = sregs->cr2;
2131 mmu_reset_needed |= vcpu->cr3 != sregs->cr3;
2132 vcpu->cr3 = sregs->cr3;
2134 vcpu->cr8 = sregs->cr8;
2136 mmu_reset_needed |= vcpu->shadow_efer != sregs->efer;
2137 #ifdef CONFIG_X86_64
2138 kvm_arch_ops->set_efer(vcpu, sregs->efer);
2140 vcpu->apic_base = sregs->apic_base;
2142 kvm_arch_ops->decache_cr4_guest_bits(vcpu);
2144 mmu_reset_needed |= vcpu->cr0 != sregs->cr0;
2145 kvm_arch_ops->set_cr0(vcpu, sregs->cr0);
2147 mmu_reset_needed |= vcpu->cr4 != sregs->cr4;
2148 kvm_arch_ops->set_cr4(vcpu, sregs->cr4);
2149 if (!is_long_mode(vcpu) && is_pae(vcpu))
2150 load_pdptrs(vcpu, vcpu->cr3);
2152 if (mmu_reset_needed)
2153 kvm_mmu_reset_context(vcpu);
2155 memcpy(vcpu->irq_pending, sregs->interrupt_bitmap,
2156 sizeof vcpu->irq_pending);
2157 vcpu->irq_summary = 0;
2158 for (i = 0; i < ARRAY_SIZE(vcpu->irq_pending); ++i)
2159 if (vcpu->irq_pending[i])
2160 __set_bit(i, &vcpu->irq_summary);
2162 set_segment(vcpu, &sregs->cs, VCPU_SREG_CS);
2163 set_segment(vcpu, &sregs->ds, VCPU_SREG_DS);
2164 set_segment(vcpu, &sregs->es, VCPU_SREG_ES);
2165 set_segment(vcpu, &sregs->fs, VCPU_SREG_FS);
2166 set_segment(vcpu, &sregs->gs, VCPU_SREG_GS);
2167 set_segment(vcpu, &sregs->ss, VCPU_SREG_SS);
2169 set_segment(vcpu, &sregs->tr, VCPU_SREG_TR);
2170 set_segment(vcpu, &sregs->ldt, VCPU_SREG_LDTR);
2178 * List of msr numbers which we expose to userspace through KVM_GET_MSRS
2179 * and KVM_SET_MSRS, and KVM_GET_MSR_INDEX_LIST.
2181 * This list is modified at module load time to reflect the
2182 * capabilities of the host cpu.
2184 static u32 msrs_to_save[] = {
2185 MSR_IA32_SYSENTER_CS, MSR_IA32_SYSENTER_ESP, MSR_IA32_SYSENTER_EIP,
2187 #ifdef CONFIG_X86_64
2188 MSR_CSTAR, MSR_KERNEL_GS_BASE, MSR_SYSCALL_MASK, MSR_LSTAR,
2190 MSR_IA32_TIME_STAMP_COUNTER,
2193 static unsigned num_msrs_to_save;
2195 static u32 emulated_msrs[] = {
2196 MSR_IA32_MISC_ENABLE,
2199 static __init void kvm_init_msr_list(void)
2204 for (i = j = 0; i < ARRAY_SIZE(msrs_to_save); i++) {
2205 if (rdmsr_safe(msrs_to_save[i], &dummy[0], &dummy[1]) < 0)
2208 msrs_to_save[j] = msrs_to_save[i];
2211 num_msrs_to_save = j;
2215 * Adapt set_msr() to msr_io()'s calling convention
2217 static int do_set_msr(struct kvm_vcpu *vcpu, unsigned index, u64 *data)
2219 return kvm_set_msr(vcpu, index, *data);
2223 * Read or write a bunch of msrs. All parameters are kernel addresses.
2225 * @return number of msrs set successfully.
2227 static int __msr_io(struct kvm_vcpu *vcpu, struct kvm_msrs *msrs,
2228 struct kvm_msr_entry *entries,
2229 int (*do_msr)(struct kvm_vcpu *vcpu,
2230 unsigned index, u64 *data))
2236 for (i = 0; i < msrs->nmsrs; ++i)
2237 if (do_msr(vcpu, entries[i].index, &entries[i].data))
2246 * Read or write a bunch of msrs. Parameters are user addresses.
2248 * @return number of msrs set successfully.
2250 static int msr_io(struct kvm_vcpu *vcpu, struct kvm_msrs __user *user_msrs,
2251 int (*do_msr)(struct kvm_vcpu *vcpu,
2252 unsigned index, u64 *data),
2255 struct kvm_msrs msrs;
2256 struct kvm_msr_entry *entries;
2261 if (copy_from_user(&msrs, user_msrs, sizeof msrs))
2265 if (msrs.nmsrs >= MAX_IO_MSRS)
2269 size = sizeof(struct kvm_msr_entry) * msrs.nmsrs;
2270 entries = vmalloc(size);
2275 if (copy_from_user(entries, user_msrs->entries, size))
2278 r = n = __msr_io(vcpu, &msrs, entries, do_msr);
2283 if (writeback && copy_to_user(user_msrs->entries, entries, size))
2295 * Translate a guest virtual address to a guest physical address.
2297 static int kvm_vcpu_ioctl_translate(struct kvm_vcpu *vcpu,
2298 struct kvm_translation *tr)
2300 unsigned long vaddr = tr->linear_address;
2304 mutex_lock(&vcpu->kvm->lock);
2305 gpa = vcpu->mmu.gva_to_gpa(vcpu, vaddr);
2306 tr->physical_address = gpa;
2307 tr->valid = gpa != UNMAPPED_GVA;
2310 mutex_unlock(&vcpu->kvm->lock);
2316 static int kvm_vcpu_ioctl_interrupt(struct kvm_vcpu *vcpu,
2317 struct kvm_interrupt *irq)
2319 if (irq->irq < 0 || irq->irq >= 256)
2323 set_bit(irq->irq, vcpu->irq_pending);
2324 set_bit(irq->irq / BITS_PER_LONG, &vcpu->irq_summary);
2331 static int kvm_vcpu_ioctl_debug_guest(struct kvm_vcpu *vcpu,
2332 struct kvm_debug_guest *dbg)
2338 r = kvm_arch_ops->set_guest_debug(vcpu, dbg);
2345 static struct page *kvm_vcpu_nopage(struct vm_area_struct *vma,
2346 unsigned long address,
2349 struct kvm_vcpu *vcpu = vma->vm_file->private_data;
2350 unsigned long pgoff;
2353 pgoff = ((address - vma->vm_start) >> PAGE_SHIFT) + vma->vm_pgoff;
2355 page = virt_to_page(vcpu->run);
2356 else if (pgoff == KVM_PIO_PAGE_OFFSET)
2357 page = virt_to_page(vcpu->pio_data);
2359 return NOPAGE_SIGBUS;
2362 *type = VM_FAULT_MINOR;
2367 static struct vm_operations_struct kvm_vcpu_vm_ops = {
2368 .nopage = kvm_vcpu_nopage,
2371 static int kvm_vcpu_mmap(struct file *file, struct vm_area_struct *vma)
2373 vma->vm_ops = &kvm_vcpu_vm_ops;
2377 static int kvm_vcpu_release(struct inode *inode, struct file *filp)
2379 struct kvm_vcpu *vcpu = filp->private_data;
2381 fput(vcpu->kvm->filp);
2385 static struct file_operations kvm_vcpu_fops = {
2386 .release = kvm_vcpu_release,
2387 .unlocked_ioctl = kvm_vcpu_ioctl,
2388 .compat_ioctl = kvm_vcpu_ioctl,
2389 .mmap = kvm_vcpu_mmap,
2393 * Allocates an inode for the vcpu.
2395 static int create_vcpu_fd(struct kvm_vcpu *vcpu)
2398 struct inode *inode;
2401 r = anon_inode_getfd(&fd, &inode, &file,
2402 "kvm-vcpu", &kvm_vcpu_fops, vcpu);
2405 atomic_inc(&vcpu->kvm->filp->f_count);
2410 * Creates some virtual cpus. Good luck creating more than one.
2412 static int kvm_vm_ioctl_create_vcpu(struct kvm *kvm, int n)
2415 struct kvm_vcpu *vcpu;
2420 vcpu = kvm_arch_ops->vcpu_create(kvm, n);
2422 return PTR_ERR(vcpu);
2424 preempt_notifier_init(&vcpu->preempt_notifier, &kvm_preempt_ops);
2427 r = kvm_mmu_setup(vcpu);
2432 mutex_lock(&kvm->lock);
2433 if (kvm->vcpus[n]) {
2435 mutex_unlock(&kvm->lock);
2438 kvm->vcpus[n] = vcpu;
2439 mutex_unlock(&kvm->lock);
2441 /* Now it's all set up, let userspace reach it */
2442 r = create_vcpu_fd(vcpu);
2448 mutex_lock(&kvm->lock);
2449 kvm->vcpus[n] = NULL;
2450 mutex_unlock(&kvm->lock);
2454 kvm_mmu_unload(vcpu);
2458 kvm_arch_ops->vcpu_free(vcpu);
2462 static void cpuid_fix_nx_cap(struct kvm_vcpu *vcpu)
2466 struct kvm_cpuid_entry *e, *entry;
2468 rdmsrl(MSR_EFER, efer);
2470 for (i = 0; i < vcpu->cpuid_nent; ++i) {
2471 e = &vcpu->cpuid_entries[i];
2472 if (e->function == 0x80000001) {
2477 if (entry && (entry->edx & (1 << 20)) && !(efer & EFER_NX)) {
2478 entry->edx &= ~(1 << 20);
2479 printk(KERN_INFO "kvm: guest NX capability removed\n");
2483 static int kvm_vcpu_ioctl_set_cpuid(struct kvm_vcpu *vcpu,
2484 struct kvm_cpuid *cpuid,
2485 struct kvm_cpuid_entry __user *entries)
2490 if (cpuid->nent > KVM_MAX_CPUID_ENTRIES)
2493 if (copy_from_user(&vcpu->cpuid_entries, entries,
2494 cpuid->nent * sizeof(struct kvm_cpuid_entry)))
2496 vcpu->cpuid_nent = cpuid->nent;
2497 cpuid_fix_nx_cap(vcpu);
2504 static int kvm_vcpu_ioctl_set_sigmask(struct kvm_vcpu *vcpu, sigset_t *sigset)
2507 sigdelsetmask(sigset, sigmask(SIGKILL)|sigmask(SIGSTOP));
2508 vcpu->sigset_active = 1;
2509 vcpu->sigset = *sigset;
2511 vcpu->sigset_active = 0;
2516 * fxsave fpu state. Taken from x86_64/processor.h. To be killed when
2517 * we have asm/x86/processor.h
2528 u32 st_space[32]; /* 8*16 bytes for each FP-reg = 128 bytes */
2529 #ifdef CONFIG_X86_64
2530 u32 xmm_space[64]; /* 16*16 bytes for each XMM-reg = 256 bytes */
2532 u32 xmm_space[32]; /* 8*16 bytes for each XMM-reg = 128 bytes */
2536 static int kvm_vcpu_ioctl_get_fpu(struct kvm_vcpu *vcpu, struct kvm_fpu *fpu)
2538 struct fxsave *fxsave = (struct fxsave *)vcpu->guest_fx_image;
2542 memcpy(fpu->fpr, fxsave->st_space, 128);
2543 fpu->fcw = fxsave->cwd;
2544 fpu->fsw = fxsave->swd;
2545 fpu->ftwx = fxsave->twd;
2546 fpu->last_opcode = fxsave->fop;
2547 fpu->last_ip = fxsave->rip;
2548 fpu->last_dp = fxsave->rdp;
2549 memcpy(fpu->xmm, fxsave->xmm_space, sizeof fxsave->xmm_space);
2556 static int kvm_vcpu_ioctl_set_fpu(struct kvm_vcpu *vcpu, struct kvm_fpu *fpu)
2558 struct fxsave *fxsave = (struct fxsave *)vcpu->guest_fx_image;
2562 memcpy(fxsave->st_space, fpu->fpr, 128);
2563 fxsave->cwd = fpu->fcw;
2564 fxsave->swd = fpu->fsw;
2565 fxsave->twd = fpu->ftwx;
2566 fxsave->fop = fpu->last_opcode;
2567 fxsave->rip = fpu->last_ip;
2568 fxsave->rdp = fpu->last_dp;
2569 memcpy(fxsave->xmm_space, fpu->xmm, sizeof fxsave->xmm_space);
2576 static long kvm_vcpu_ioctl(struct file *filp,
2577 unsigned int ioctl, unsigned long arg)
2579 struct kvm_vcpu *vcpu = filp->private_data;
2580 void __user *argp = (void __user *)arg;
2588 r = kvm_vcpu_ioctl_run(vcpu, vcpu->run);
2590 case KVM_GET_REGS: {
2591 struct kvm_regs kvm_regs;
2593 memset(&kvm_regs, 0, sizeof kvm_regs);
2594 r = kvm_vcpu_ioctl_get_regs(vcpu, &kvm_regs);
2598 if (copy_to_user(argp, &kvm_regs, sizeof kvm_regs))
2603 case KVM_SET_REGS: {
2604 struct kvm_regs kvm_regs;
2607 if (copy_from_user(&kvm_regs, argp, sizeof kvm_regs))
2609 r = kvm_vcpu_ioctl_set_regs(vcpu, &kvm_regs);
2615 case KVM_GET_SREGS: {
2616 struct kvm_sregs kvm_sregs;
2618 memset(&kvm_sregs, 0, sizeof kvm_sregs);
2619 r = kvm_vcpu_ioctl_get_sregs(vcpu, &kvm_sregs);
2623 if (copy_to_user(argp, &kvm_sregs, sizeof kvm_sregs))
2628 case KVM_SET_SREGS: {
2629 struct kvm_sregs kvm_sregs;
2632 if (copy_from_user(&kvm_sregs, argp, sizeof kvm_sregs))
2634 r = kvm_vcpu_ioctl_set_sregs(vcpu, &kvm_sregs);
2640 case KVM_TRANSLATE: {
2641 struct kvm_translation tr;
2644 if (copy_from_user(&tr, argp, sizeof tr))
2646 r = kvm_vcpu_ioctl_translate(vcpu, &tr);
2650 if (copy_to_user(argp, &tr, sizeof tr))
2655 case KVM_INTERRUPT: {
2656 struct kvm_interrupt irq;
2659 if (copy_from_user(&irq, argp, sizeof irq))
2661 r = kvm_vcpu_ioctl_interrupt(vcpu, &irq);
2667 case KVM_DEBUG_GUEST: {
2668 struct kvm_debug_guest dbg;
2671 if (copy_from_user(&dbg, argp, sizeof dbg))
2673 r = kvm_vcpu_ioctl_debug_guest(vcpu, &dbg);
2680 r = msr_io(vcpu, argp, kvm_get_msr, 1);
2683 r = msr_io(vcpu, argp, do_set_msr, 0);
2685 case KVM_SET_CPUID: {
2686 struct kvm_cpuid __user *cpuid_arg = argp;
2687 struct kvm_cpuid cpuid;
2690 if (copy_from_user(&cpuid, cpuid_arg, sizeof cpuid))
2692 r = kvm_vcpu_ioctl_set_cpuid(vcpu, &cpuid, cpuid_arg->entries);
2697 case KVM_SET_SIGNAL_MASK: {
2698 struct kvm_signal_mask __user *sigmask_arg = argp;
2699 struct kvm_signal_mask kvm_sigmask;
2700 sigset_t sigset, *p;
2705 if (copy_from_user(&kvm_sigmask, argp,
2706 sizeof kvm_sigmask))
2709 if (kvm_sigmask.len != sizeof sigset)
2712 if (copy_from_user(&sigset, sigmask_arg->sigset,
2717 r = kvm_vcpu_ioctl_set_sigmask(vcpu, &sigset);
2723 memset(&fpu, 0, sizeof fpu);
2724 r = kvm_vcpu_ioctl_get_fpu(vcpu, &fpu);
2728 if (copy_to_user(argp, &fpu, sizeof fpu))
2737 if (copy_from_user(&fpu, argp, sizeof fpu))
2739 r = kvm_vcpu_ioctl_set_fpu(vcpu, &fpu);
2752 static long kvm_vm_ioctl(struct file *filp,
2753 unsigned int ioctl, unsigned long arg)
2755 struct kvm *kvm = filp->private_data;
2756 void __user *argp = (void __user *)arg;
2760 case KVM_CREATE_VCPU:
2761 r = kvm_vm_ioctl_create_vcpu(kvm, arg);
2765 case KVM_SET_MEMORY_REGION: {
2766 struct kvm_memory_region kvm_mem;
2769 if (copy_from_user(&kvm_mem, argp, sizeof kvm_mem))
2771 r = kvm_vm_ioctl_set_memory_region(kvm, &kvm_mem);
2776 case KVM_GET_DIRTY_LOG: {
2777 struct kvm_dirty_log log;
2780 if (copy_from_user(&log, argp, sizeof log))
2782 r = kvm_vm_ioctl_get_dirty_log(kvm, &log);
2787 case KVM_SET_MEMORY_ALIAS: {
2788 struct kvm_memory_alias alias;
2791 if (copy_from_user(&alias, argp, sizeof alias))
2793 r = kvm_vm_ioctl_set_memory_alias(kvm, &alias);
2805 static struct page *kvm_vm_nopage(struct vm_area_struct *vma,
2806 unsigned long address,
2809 struct kvm *kvm = vma->vm_file->private_data;
2810 unsigned long pgoff;
2813 pgoff = ((address - vma->vm_start) >> PAGE_SHIFT) + vma->vm_pgoff;
2814 page = gfn_to_page(kvm, pgoff);
2816 return NOPAGE_SIGBUS;
2819 *type = VM_FAULT_MINOR;
2824 static struct vm_operations_struct kvm_vm_vm_ops = {
2825 .nopage = kvm_vm_nopage,
2828 static int kvm_vm_mmap(struct file *file, struct vm_area_struct *vma)
2830 vma->vm_ops = &kvm_vm_vm_ops;
2834 static struct file_operations kvm_vm_fops = {
2835 .release = kvm_vm_release,
2836 .unlocked_ioctl = kvm_vm_ioctl,
2837 .compat_ioctl = kvm_vm_ioctl,
2838 .mmap = kvm_vm_mmap,
2841 static int kvm_dev_ioctl_create_vm(void)
2844 struct inode *inode;
2848 kvm = kvm_create_vm();
2850 return PTR_ERR(kvm);
2851 r = anon_inode_getfd(&fd, &inode, &file, "kvm-vm", &kvm_vm_fops, kvm);
2853 kvm_destroy_vm(kvm);
2862 static long kvm_dev_ioctl(struct file *filp,
2863 unsigned int ioctl, unsigned long arg)
2865 void __user *argp = (void __user *)arg;
2869 case KVM_GET_API_VERSION:
2873 r = KVM_API_VERSION;
2879 r = kvm_dev_ioctl_create_vm();
2881 case KVM_GET_MSR_INDEX_LIST: {
2882 struct kvm_msr_list __user *user_msr_list = argp;
2883 struct kvm_msr_list msr_list;
2887 if (copy_from_user(&msr_list, user_msr_list, sizeof msr_list))
2890 msr_list.nmsrs = num_msrs_to_save + ARRAY_SIZE(emulated_msrs);
2891 if (copy_to_user(user_msr_list, &msr_list, sizeof msr_list))
2894 if (n < num_msrs_to_save)
2897 if (copy_to_user(user_msr_list->indices, &msrs_to_save,
2898 num_msrs_to_save * sizeof(u32)))
2900 if (copy_to_user(user_msr_list->indices
2901 + num_msrs_to_save * sizeof(u32),
2903 ARRAY_SIZE(emulated_msrs) * sizeof(u32)))
2908 case KVM_CHECK_EXTENSION:
2910 * No extensions defined at present.
2914 case KVM_GET_VCPU_MMAP_SIZE:
2927 static struct file_operations kvm_chardev_ops = {
2928 .open = kvm_dev_open,
2929 .release = kvm_dev_release,
2930 .unlocked_ioctl = kvm_dev_ioctl,
2931 .compat_ioctl = kvm_dev_ioctl,
2934 static struct miscdevice kvm_dev = {
2941 * Make sure that a cpu that is being hot-unplugged does not have any vcpus
2944 static void decache_vcpus_on_cpu(int cpu)
2947 struct kvm_vcpu *vcpu;
2950 spin_lock(&kvm_lock);
2951 list_for_each_entry(vm, &vm_list, vm_list)
2952 for (i = 0; i < KVM_MAX_VCPUS; ++i) {
2953 vcpu = vm->vcpus[i];
2957 * If the vcpu is locked, then it is running on some
2958 * other cpu and therefore it is not cached on the
2961 * If it's not locked, check the last cpu it executed
2964 if (mutex_trylock(&vcpu->mutex)) {
2965 if (vcpu->cpu == cpu) {
2966 kvm_arch_ops->vcpu_decache(vcpu);
2969 mutex_unlock(&vcpu->mutex);
2972 spin_unlock(&kvm_lock);
2975 static void hardware_enable(void *junk)
2977 int cpu = raw_smp_processor_id();
2979 if (cpu_isset(cpu, cpus_hardware_enabled))
2981 cpu_set(cpu, cpus_hardware_enabled);
2982 kvm_arch_ops->hardware_enable(NULL);
2985 static void hardware_disable(void *junk)
2987 int cpu = raw_smp_processor_id();
2989 if (!cpu_isset(cpu, cpus_hardware_enabled))
2991 cpu_clear(cpu, cpus_hardware_enabled);
2992 decache_vcpus_on_cpu(cpu);
2993 kvm_arch_ops->hardware_disable(NULL);
2996 static int kvm_cpu_hotplug(struct notifier_block *notifier, unsigned long val,
3003 case CPU_DYING_FROZEN:
3004 printk(KERN_INFO "kvm: disabling virtualization on CPU%d\n",
3006 hardware_disable(NULL);
3008 case CPU_UP_CANCELED:
3009 case CPU_UP_CANCELED_FROZEN:
3010 printk(KERN_INFO "kvm: disabling virtualization on CPU%d\n",
3012 smp_call_function_single(cpu, hardware_disable, NULL, 0, 1);
3015 case CPU_ONLINE_FROZEN:
3016 printk(KERN_INFO "kvm: enabling virtualization on CPU%d\n",
3018 smp_call_function_single(cpu, hardware_enable, NULL, 0, 1);
3024 static int kvm_reboot(struct notifier_block *notifier, unsigned long val,
3027 if (val == SYS_RESTART) {
3029 * Some (well, at least mine) BIOSes hang on reboot if
3032 printk(KERN_INFO "kvm: exiting hardware virtualization\n");
3033 on_each_cpu(hardware_disable, NULL, 0, 1);
3038 static struct notifier_block kvm_reboot_notifier = {
3039 .notifier_call = kvm_reboot,
3043 void kvm_io_bus_init(struct kvm_io_bus *bus)
3045 memset(bus, 0, sizeof(*bus));
3048 void kvm_io_bus_destroy(struct kvm_io_bus *bus)
3052 for (i = 0; i < bus->dev_count; i++) {
3053 struct kvm_io_device *pos = bus->devs[i];
3055 kvm_iodevice_destructor(pos);
3059 struct kvm_io_device *kvm_io_bus_find_dev(struct kvm_io_bus *bus, gpa_t addr)
3063 for (i = 0; i < bus->dev_count; i++) {
3064 struct kvm_io_device *pos = bus->devs[i];
3066 if (pos->in_range(pos, addr))
3073 void kvm_io_bus_register_dev(struct kvm_io_bus *bus, struct kvm_io_device *dev)
3075 BUG_ON(bus->dev_count > (NR_IOBUS_DEVS-1));
3077 bus->devs[bus->dev_count++] = dev;
3080 static struct notifier_block kvm_cpu_notifier = {
3081 .notifier_call = kvm_cpu_hotplug,
3082 .priority = 20, /* must be > scheduler priority */
3085 static u64 stat_get(void *_offset)
3087 unsigned offset = (long)_offset;
3090 struct kvm_vcpu *vcpu;
3093 spin_lock(&kvm_lock);
3094 list_for_each_entry(kvm, &vm_list, vm_list)
3095 for (i = 0; i < KVM_MAX_VCPUS; ++i) {
3096 vcpu = kvm->vcpus[i];
3098 total += *(u32 *)((void *)vcpu + offset);
3100 spin_unlock(&kvm_lock);
3104 static void stat_set(void *offset, u64 val)
3108 DEFINE_SIMPLE_ATTRIBUTE(stat_fops, stat_get, stat_set, "%llu\n");
3110 static __init void kvm_init_debug(void)
3112 struct kvm_stats_debugfs_item *p;
3114 debugfs_dir = debugfs_create_dir("kvm", NULL);
3115 for (p = debugfs_entries; p->name; ++p)
3116 p->dentry = debugfs_create_file(p->name, 0444, debugfs_dir,
3117 (void *)(long)p->offset,
3121 static void kvm_exit_debug(void)
3123 struct kvm_stats_debugfs_item *p;
3125 for (p = debugfs_entries; p->name; ++p)
3126 debugfs_remove(p->dentry);
3127 debugfs_remove(debugfs_dir);
3130 static int kvm_suspend(struct sys_device *dev, pm_message_t state)
3132 hardware_disable(NULL);
3136 static int kvm_resume(struct sys_device *dev)
3138 hardware_enable(NULL);
3142 static struct sysdev_class kvm_sysdev_class = {
3143 set_kset_name("kvm"),
3144 .suspend = kvm_suspend,
3145 .resume = kvm_resume,
3148 static struct sys_device kvm_sysdev = {
3150 .cls = &kvm_sysdev_class,
3153 hpa_t bad_page_address;
3156 struct kvm_vcpu *preempt_notifier_to_vcpu(struct preempt_notifier *pn)
3158 return container_of(pn, struct kvm_vcpu, preempt_notifier);
3161 static void kvm_sched_in(struct preempt_notifier *pn, int cpu)
3163 struct kvm_vcpu *vcpu = preempt_notifier_to_vcpu(pn);
3165 kvm_arch_ops->vcpu_load(vcpu, cpu);
3168 static void kvm_sched_out(struct preempt_notifier *pn,
3169 struct task_struct *next)
3171 struct kvm_vcpu *vcpu = preempt_notifier_to_vcpu(pn);
3173 kvm_arch_ops->vcpu_put(vcpu);
3176 int kvm_init_arch(struct kvm_arch_ops *ops, struct module *module)
3181 printk(KERN_ERR "kvm: already loaded the other module\n");
3185 if (!ops->cpu_has_kvm_support()) {
3186 printk(KERN_ERR "kvm: no hardware support\n");
3189 if (ops->disabled_by_bios()) {
3190 printk(KERN_ERR "kvm: disabled by bios\n");
3196 r = kvm_arch_ops->hardware_setup();
3200 on_each_cpu(hardware_enable, NULL, 0, 1);
3201 r = register_cpu_notifier(&kvm_cpu_notifier);
3204 register_reboot_notifier(&kvm_reboot_notifier);
3206 r = sysdev_class_register(&kvm_sysdev_class);
3210 r = sysdev_register(&kvm_sysdev);
3214 kvm_chardev_ops.owner = module;
3216 r = misc_register(&kvm_dev);
3218 printk (KERN_ERR "kvm: misc device register failed\n");
3222 kvm_preempt_ops.sched_in = kvm_sched_in;
3223 kvm_preempt_ops.sched_out = kvm_sched_out;
3228 sysdev_unregister(&kvm_sysdev);
3230 sysdev_class_unregister(&kvm_sysdev_class);
3232 unregister_reboot_notifier(&kvm_reboot_notifier);
3233 unregister_cpu_notifier(&kvm_cpu_notifier);
3235 on_each_cpu(hardware_disable, NULL, 0, 1);
3236 kvm_arch_ops->hardware_unsetup();
3238 kvm_arch_ops = NULL;
3242 void kvm_exit_arch(void)
3244 misc_deregister(&kvm_dev);
3245 sysdev_unregister(&kvm_sysdev);
3246 sysdev_class_unregister(&kvm_sysdev_class);
3247 unregister_reboot_notifier(&kvm_reboot_notifier);
3248 unregister_cpu_notifier(&kvm_cpu_notifier);
3249 on_each_cpu(hardware_disable, NULL, 0, 1);
3250 kvm_arch_ops->hardware_unsetup();
3251 kvm_arch_ops = NULL;
3254 static __init int kvm_init(void)
3256 static struct page *bad_page;
3259 r = kvm_mmu_module_init();
3265 kvm_init_msr_list();
3267 if ((bad_page = alloc_page(GFP_KERNEL)) == NULL) {
3272 bad_page_address = page_to_pfn(bad_page) << PAGE_SHIFT;
3273 memset(__va(bad_page_address), 0, PAGE_SIZE);
3279 kvm_mmu_module_exit();
3284 static __exit void kvm_exit(void)
3287 __free_page(pfn_to_page(bad_page_address >> PAGE_SHIFT));
3288 kvm_mmu_module_exit();
3291 module_init(kvm_init)
3292 module_exit(kvm_exit)
3294 EXPORT_SYMBOL_GPL(kvm_init_arch);
3295 EXPORT_SYMBOL_GPL(kvm_exit_arch);