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"
23 #include <linux/kvm.h>
24 #include <linux/module.h>
25 #include <linux/errno.h>
26 #include <linux/percpu.h>
27 #include <linux/gfp.h>
29 #include <linux/miscdevice.h>
30 #include <linux/vmalloc.h>
31 #include <linux/reboot.h>
32 #include <linux/debugfs.h>
33 #include <linux/highmem.h>
34 #include <linux/file.h>
35 #include <linux/sysdev.h>
36 #include <linux/cpu.h>
37 #include <linux/sched.h>
38 #include <linux/cpumask.h>
39 #include <linux/smp.h>
40 #include <linux/anon_inodes.h>
42 #include <asm/processor.h>
45 #include <asm/uaccess.h>
48 MODULE_AUTHOR("Qumranet");
49 MODULE_LICENSE("GPL");
51 static DEFINE_SPINLOCK(kvm_lock);
52 static LIST_HEAD(vm_list);
54 static cpumask_t cpus_hardware_enabled;
56 struct kvm_arch_ops *kvm_arch_ops;
57 struct kmem_cache *kvm_vcpu_cache;
58 EXPORT_SYMBOL_GPL(kvm_vcpu_cache);
60 static __read_mostly struct preempt_ops kvm_preempt_ops;
62 #define STAT_OFFSET(x) offsetof(struct kvm_vcpu, stat.x)
64 static struct kvm_stats_debugfs_item {
67 struct dentry *dentry;
68 } debugfs_entries[] = {
69 { "pf_fixed", STAT_OFFSET(pf_fixed) },
70 { "pf_guest", STAT_OFFSET(pf_guest) },
71 { "tlb_flush", STAT_OFFSET(tlb_flush) },
72 { "invlpg", STAT_OFFSET(invlpg) },
73 { "exits", STAT_OFFSET(exits) },
74 { "io_exits", STAT_OFFSET(io_exits) },
75 { "mmio_exits", STAT_OFFSET(mmio_exits) },
76 { "signal_exits", STAT_OFFSET(signal_exits) },
77 { "irq_window", STAT_OFFSET(irq_window_exits) },
78 { "halt_exits", STAT_OFFSET(halt_exits) },
79 { "halt_wakeup", STAT_OFFSET(halt_wakeup) },
80 { "request_irq", STAT_OFFSET(request_irq_exits) },
81 { "irq_exits", STAT_OFFSET(irq_exits) },
82 { "light_exits", STAT_OFFSET(light_exits) },
83 { "efer_reload", STAT_OFFSET(efer_reload) },
87 static struct dentry *debugfs_dir;
89 #define MAX_IO_MSRS 256
91 #define CR0_RESERVED_BITS \
92 (~(unsigned long)(X86_CR0_PE | X86_CR0_MP | X86_CR0_EM | X86_CR0_TS \
93 | X86_CR0_ET | X86_CR0_NE | X86_CR0_WP | X86_CR0_AM \
94 | X86_CR0_NW | X86_CR0_CD | X86_CR0_PG))
95 #define CR4_RESERVED_BITS \
96 (~(unsigned long)(X86_CR4_VME | X86_CR4_PVI | X86_CR4_TSD | X86_CR4_DE\
97 | X86_CR4_PSE | X86_CR4_PAE | X86_CR4_MCE \
98 | X86_CR4_PGE | X86_CR4_PCE | X86_CR4_OSFXSR \
99 | X86_CR4_OSXMMEXCPT | X86_CR4_VMXE))
101 #define CR8_RESERVED_BITS (~(unsigned long)X86_CR8_TPR)
102 #define EFER_RESERVED_BITS 0xfffffffffffff2fe
105 // LDT or TSS descriptor in the GDT. 16 bytes.
106 struct segment_descriptor_64 {
107 struct segment_descriptor s;
114 static long kvm_vcpu_ioctl(struct file *file, unsigned int ioctl,
117 unsigned long segment_base(u16 selector)
119 struct descriptor_table gdt;
120 struct segment_descriptor *d;
121 unsigned long table_base;
122 typedef unsigned long ul;
128 asm ("sgdt %0" : "=m"(gdt));
129 table_base = gdt.base;
131 if (selector & 4) { /* from ldt */
134 asm ("sldt %0" : "=g"(ldt_selector));
135 table_base = segment_base(ldt_selector);
137 d = (struct segment_descriptor *)(table_base + (selector & ~7));
138 v = d->base_low | ((ul)d->base_mid << 16) | ((ul)d->base_high << 24);
141 && (d->type == 2 || d->type == 9 || d->type == 11))
142 v |= ((ul)((struct segment_descriptor_64 *)d)->base_higher) << 32;
146 EXPORT_SYMBOL_GPL(segment_base);
148 static inline int valid_vcpu(int n)
150 return likely(n >= 0 && n < KVM_MAX_VCPUS);
153 void kvm_load_guest_fpu(struct kvm_vcpu *vcpu)
155 if (!vcpu->fpu_active || vcpu->guest_fpu_loaded)
158 vcpu->guest_fpu_loaded = 1;
159 fx_save(&vcpu->host_fx_image);
160 fx_restore(&vcpu->guest_fx_image);
162 EXPORT_SYMBOL_GPL(kvm_load_guest_fpu);
164 void kvm_put_guest_fpu(struct kvm_vcpu *vcpu)
166 if (!vcpu->guest_fpu_loaded)
169 vcpu->guest_fpu_loaded = 0;
170 fx_save(&vcpu->guest_fx_image);
171 fx_restore(&vcpu->host_fx_image);
173 EXPORT_SYMBOL_GPL(kvm_put_guest_fpu);
176 * Switches to specified vcpu, until a matching vcpu_put()
178 static void vcpu_load(struct kvm_vcpu *vcpu)
182 mutex_lock(&vcpu->mutex);
184 preempt_notifier_register(&vcpu->preempt_notifier);
185 kvm_arch_ops->vcpu_load(vcpu, cpu);
189 static void vcpu_put(struct kvm_vcpu *vcpu)
192 kvm_arch_ops->vcpu_put(vcpu);
193 preempt_notifier_unregister(&vcpu->preempt_notifier);
195 mutex_unlock(&vcpu->mutex);
198 static void ack_flush(void *_completed)
200 atomic_t *completed = _completed;
202 atomic_inc(completed);
205 void kvm_flush_remote_tlbs(struct kvm *kvm)
209 struct kvm_vcpu *vcpu;
212 atomic_set(&completed, 0);
215 for (i = 0; i < KVM_MAX_VCPUS; ++i) {
216 vcpu = kvm->vcpus[i];
219 if (test_and_set_bit(KVM_TLB_FLUSH, &vcpu->requests))
222 if (cpu != -1 && cpu != raw_smp_processor_id())
223 if (!cpu_isset(cpu, cpus)) {
230 * We really want smp_call_function_mask() here. But that's not
231 * available, so ipi all cpus in parallel and wait for them
234 for (cpu = first_cpu(cpus); cpu != NR_CPUS; cpu = next_cpu(cpu, cpus))
235 smp_call_function_single(cpu, ack_flush, &completed, 1, 0);
236 while (atomic_read(&completed) != needed) {
242 int kvm_vcpu_init(struct kvm_vcpu *vcpu, struct kvm *kvm, unsigned id)
247 mutex_init(&vcpu->mutex);
249 vcpu->mmu.root_hpa = INVALID_PAGE;
252 init_waitqueue_head(&vcpu->wq);
254 page = alloc_page(GFP_KERNEL | __GFP_ZERO);
259 vcpu->run = page_address(page);
261 page = alloc_page(GFP_KERNEL | __GFP_ZERO);
266 vcpu->pio_data = page_address(page);
268 r = kvm_mmu_create(vcpu);
270 goto fail_free_pio_data;
275 free_page((unsigned long)vcpu->pio_data);
277 free_page((unsigned long)vcpu->run);
281 EXPORT_SYMBOL_GPL(kvm_vcpu_init);
283 void kvm_vcpu_uninit(struct kvm_vcpu *vcpu)
285 kvm_mmu_destroy(vcpu);
286 kvm_free_apic(vcpu->apic);
287 free_page((unsigned long)vcpu->pio_data);
288 free_page((unsigned long)vcpu->run);
290 EXPORT_SYMBOL_GPL(kvm_vcpu_uninit);
292 static struct kvm *kvm_create_vm(void)
294 struct kvm *kvm = kzalloc(sizeof(struct kvm), GFP_KERNEL);
297 return ERR_PTR(-ENOMEM);
299 kvm_io_bus_init(&kvm->pio_bus);
300 mutex_init(&kvm->lock);
301 INIT_LIST_HEAD(&kvm->active_mmu_pages);
302 kvm_io_bus_init(&kvm->mmio_bus);
303 spin_lock(&kvm_lock);
304 list_add(&kvm->vm_list, &vm_list);
305 spin_unlock(&kvm_lock);
310 * Free any memory in @free but not in @dont.
312 static void kvm_free_physmem_slot(struct kvm_memory_slot *free,
313 struct kvm_memory_slot *dont)
317 if (!dont || free->phys_mem != dont->phys_mem)
318 if (free->phys_mem) {
319 for (i = 0; i < free->npages; ++i)
320 if (free->phys_mem[i])
321 __free_page(free->phys_mem[i]);
322 vfree(free->phys_mem);
325 if (!dont || free->dirty_bitmap != dont->dirty_bitmap)
326 vfree(free->dirty_bitmap);
328 free->phys_mem = NULL;
330 free->dirty_bitmap = NULL;
333 static void kvm_free_physmem(struct kvm *kvm)
337 for (i = 0; i < kvm->nmemslots; ++i)
338 kvm_free_physmem_slot(&kvm->memslots[i], NULL);
341 static void free_pio_guest_pages(struct kvm_vcpu *vcpu)
345 for (i = 0; i < ARRAY_SIZE(vcpu->pio.guest_pages); ++i)
346 if (vcpu->pio.guest_pages[i]) {
347 __free_page(vcpu->pio.guest_pages[i]);
348 vcpu->pio.guest_pages[i] = NULL;
352 static void kvm_unload_vcpu_mmu(struct kvm_vcpu *vcpu)
355 kvm_mmu_unload(vcpu);
359 static void kvm_free_vcpus(struct kvm *kvm)
364 * Unpin any mmu pages first.
366 for (i = 0; i < KVM_MAX_VCPUS; ++i)
368 kvm_unload_vcpu_mmu(kvm->vcpus[i]);
369 for (i = 0; i < KVM_MAX_VCPUS; ++i) {
371 kvm_arch_ops->vcpu_free(kvm->vcpus[i]);
372 kvm->vcpus[i] = NULL;
378 static void kvm_destroy_vm(struct kvm *kvm)
380 spin_lock(&kvm_lock);
381 list_del(&kvm->vm_list);
382 spin_unlock(&kvm_lock);
383 kvm_io_bus_destroy(&kvm->pio_bus);
384 kvm_io_bus_destroy(&kvm->mmio_bus);
388 kvm_free_physmem(kvm);
392 static int kvm_vm_release(struct inode *inode, struct file *filp)
394 struct kvm *kvm = filp->private_data;
400 static void inject_gp(struct kvm_vcpu *vcpu)
402 kvm_arch_ops->inject_gp(vcpu, 0);
406 * Load the pae pdptrs. Return true is they are all valid.
408 static int load_pdptrs(struct kvm_vcpu *vcpu, unsigned long cr3)
410 gfn_t pdpt_gfn = cr3 >> PAGE_SHIFT;
411 unsigned offset = ((cr3 & (PAGE_SIZE-1)) >> 5) << 2;
416 u64 pdpte[ARRAY_SIZE(vcpu->pdptrs)];
418 mutex_lock(&vcpu->kvm->lock);
419 page = gfn_to_page(vcpu->kvm, pdpt_gfn);
425 pdpt = kmap_atomic(page, KM_USER0);
426 memcpy(pdpte, pdpt+offset, sizeof(pdpte));
427 kunmap_atomic(pdpt, KM_USER0);
429 for (i = 0; i < ARRAY_SIZE(pdpte); ++i) {
430 if ((pdpte[i] & 1) && (pdpte[i] & 0xfffffff0000001e6ull)) {
437 memcpy(vcpu->pdptrs, pdpte, sizeof(vcpu->pdptrs));
439 mutex_unlock(&vcpu->kvm->lock);
444 void set_cr0(struct kvm_vcpu *vcpu, unsigned long cr0)
446 if (cr0 & CR0_RESERVED_BITS) {
447 printk(KERN_DEBUG "set_cr0: 0x%lx #GP, reserved bits 0x%lx\n",
453 if ((cr0 & X86_CR0_NW) && !(cr0 & X86_CR0_CD)) {
454 printk(KERN_DEBUG "set_cr0: #GP, CD == 0 && NW == 1\n");
459 if ((cr0 & X86_CR0_PG) && !(cr0 & X86_CR0_PE)) {
460 printk(KERN_DEBUG "set_cr0: #GP, set PG flag "
461 "and a clear PE flag\n");
466 if (!is_paging(vcpu) && (cr0 & X86_CR0_PG)) {
468 if ((vcpu->shadow_efer & EFER_LME)) {
472 printk(KERN_DEBUG "set_cr0: #GP, start paging "
473 "in long mode while PAE is disabled\n");
477 kvm_arch_ops->get_cs_db_l_bits(vcpu, &cs_db, &cs_l);
479 printk(KERN_DEBUG "set_cr0: #GP, start paging "
480 "in long mode while CS.L == 1\n");
487 if (is_pae(vcpu) && !load_pdptrs(vcpu, vcpu->cr3)) {
488 printk(KERN_DEBUG "set_cr0: #GP, pdptrs "
496 kvm_arch_ops->set_cr0(vcpu, cr0);
499 mutex_lock(&vcpu->kvm->lock);
500 kvm_mmu_reset_context(vcpu);
501 mutex_unlock(&vcpu->kvm->lock);
504 EXPORT_SYMBOL_GPL(set_cr0);
506 void lmsw(struct kvm_vcpu *vcpu, unsigned long msw)
508 set_cr0(vcpu, (vcpu->cr0 & ~0x0ful) | (msw & 0x0f));
510 EXPORT_SYMBOL_GPL(lmsw);
512 void set_cr4(struct kvm_vcpu *vcpu, unsigned long cr4)
514 if (cr4 & CR4_RESERVED_BITS) {
515 printk(KERN_DEBUG "set_cr4: #GP, reserved bits\n");
520 if (is_long_mode(vcpu)) {
521 if (!(cr4 & X86_CR4_PAE)) {
522 printk(KERN_DEBUG "set_cr4: #GP, clearing PAE while "
527 } else if (is_paging(vcpu) && !is_pae(vcpu) && (cr4 & X86_CR4_PAE)
528 && !load_pdptrs(vcpu, vcpu->cr3)) {
529 printk(KERN_DEBUG "set_cr4: #GP, pdptrs reserved bits\n");
534 if (cr4 & X86_CR4_VMXE) {
535 printk(KERN_DEBUG "set_cr4: #GP, setting VMXE\n");
539 kvm_arch_ops->set_cr4(vcpu, cr4);
540 mutex_lock(&vcpu->kvm->lock);
541 kvm_mmu_reset_context(vcpu);
542 mutex_unlock(&vcpu->kvm->lock);
544 EXPORT_SYMBOL_GPL(set_cr4);
546 void set_cr3(struct kvm_vcpu *vcpu, unsigned long cr3)
548 if (is_long_mode(vcpu)) {
549 if (cr3 & CR3_L_MODE_RESERVED_BITS) {
550 printk(KERN_DEBUG "set_cr3: #GP, reserved bits\n");
556 if (cr3 & CR3_PAE_RESERVED_BITS) {
558 "set_cr3: #GP, reserved bits\n");
562 if (is_paging(vcpu) && !load_pdptrs(vcpu, cr3)) {
563 printk(KERN_DEBUG "set_cr3: #GP, pdptrs "
569 if (cr3 & CR3_NONPAE_RESERVED_BITS) {
571 "set_cr3: #GP, reserved bits\n");
578 mutex_lock(&vcpu->kvm->lock);
580 * Does the new cr3 value map to physical memory? (Note, we
581 * catch an invalid cr3 even in real-mode, because it would
582 * cause trouble later on when we turn on paging anyway.)
584 * A real CPU would silently accept an invalid cr3 and would
585 * attempt to use it - with largely undefined (and often hard
586 * to debug) behavior on the guest side.
588 if (unlikely(!gfn_to_memslot(vcpu->kvm, cr3 >> PAGE_SHIFT)))
592 vcpu->mmu.new_cr3(vcpu);
594 mutex_unlock(&vcpu->kvm->lock);
596 EXPORT_SYMBOL_GPL(set_cr3);
598 void set_cr8(struct kvm_vcpu *vcpu, unsigned long cr8)
600 if (cr8 & CR8_RESERVED_BITS) {
601 printk(KERN_DEBUG "set_cr8: #GP, reserved bits 0x%lx\n", cr8);
605 if (irqchip_in_kernel(vcpu->kvm))
606 kvm_lapic_set_tpr(vcpu, cr8);
610 EXPORT_SYMBOL_GPL(set_cr8);
612 unsigned long get_cr8(struct kvm_vcpu *vcpu)
614 if (irqchip_in_kernel(vcpu->kvm))
615 return kvm_lapic_get_cr8(vcpu);
619 EXPORT_SYMBOL_GPL(get_cr8);
621 u64 kvm_get_apic_base(struct kvm_vcpu *vcpu)
623 if (irqchip_in_kernel(vcpu->kvm))
624 return vcpu->apic_base;
626 return vcpu->apic_base;
628 EXPORT_SYMBOL_GPL(kvm_get_apic_base);
630 void kvm_set_apic_base(struct kvm_vcpu *vcpu, u64 data)
632 /* TODO: reserve bits check */
633 if (irqchip_in_kernel(vcpu->kvm))
634 kvm_lapic_set_base(vcpu, data);
636 vcpu->apic_base = data;
638 EXPORT_SYMBOL_GPL(kvm_set_apic_base);
640 void fx_init(struct kvm_vcpu *vcpu)
642 unsigned after_mxcsr_mask;
644 /* Initialize guest FPU by resetting ours and saving into guest's */
646 fx_save(&vcpu->host_fx_image);
648 fx_save(&vcpu->guest_fx_image);
649 fx_restore(&vcpu->host_fx_image);
652 after_mxcsr_mask = offsetof(struct i387_fxsave_struct, st_space);
653 vcpu->guest_fx_image.mxcsr = 0x1f80;
654 memset((void *)&vcpu->guest_fx_image + after_mxcsr_mask,
655 0, sizeof(struct i387_fxsave_struct) - after_mxcsr_mask);
657 EXPORT_SYMBOL_GPL(fx_init);
660 * Allocate some memory and give it an address in the guest physical address
663 * Discontiguous memory is allowed, mostly for framebuffers.
665 static int kvm_vm_ioctl_set_memory_region(struct kvm *kvm,
666 struct kvm_memory_region *mem)
670 unsigned long npages;
672 struct kvm_memory_slot *memslot;
673 struct kvm_memory_slot old, new;
674 int memory_config_version;
677 /* General sanity checks */
678 if (mem->memory_size & (PAGE_SIZE - 1))
680 if (mem->guest_phys_addr & (PAGE_SIZE - 1))
682 if (mem->slot >= KVM_MEMORY_SLOTS)
684 if (mem->guest_phys_addr + mem->memory_size < mem->guest_phys_addr)
687 memslot = &kvm->memslots[mem->slot];
688 base_gfn = mem->guest_phys_addr >> PAGE_SHIFT;
689 npages = mem->memory_size >> PAGE_SHIFT;
692 mem->flags &= ~KVM_MEM_LOG_DIRTY_PAGES;
695 mutex_lock(&kvm->lock);
697 memory_config_version = kvm->memory_config_version;
698 new = old = *memslot;
700 new.base_gfn = base_gfn;
702 new.flags = mem->flags;
704 /* Disallow changing a memory slot's size. */
706 if (npages && old.npages && npages != old.npages)
709 /* Check for overlaps */
711 for (i = 0; i < KVM_MEMORY_SLOTS; ++i) {
712 struct kvm_memory_slot *s = &kvm->memslots[i];
716 if (!((base_gfn + npages <= s->base_gfn) ||
717 (base_gfn >= s->base_gfn + s->npages)))
721 * Do memory allocations outside lock. memory_config_version will
724 mutex_unlock(&kvm->lock);
726 /* Deallocate if slot is being removed */
730 /* Free page dirty bitmap if unneeded */
731 if (!(new.flags & KVM_MEM_LOG_DIRTY_PAGES))
732 new.dirty_bitmap = NULL;
736 /* Allocate if a slot is being created */
737 if (npages && !new.phys_mem) {
738 new.phys_mem = vmalloc(npages * sizeof(struct page *));
743 memset(new.phys_mem, 0, npages * sizeof(struct page *));
744 for (i = 0; i < npages; ++i) {
745 new.phys_mem[i] = alloc_page(GFP_HIGHUSER
747 if (!new.phys_mem[i])
749 set_page_private(new.phys_mem[i],0);
753 /* Allocate page dirty bitmap if needed */
754 if ((new.flags & KVM_MEM_LOG_DIRTY_PAGES) && !new.dirty_bitmap) {
755 unsigned dirty_bytes = ALIGN(npages, BITS_PER_LONG) / 8;
757 new.dirty_bitmap = vmalloc(dirty_bytes);
758 if (!new.dirty_bitmap)
760 memset(new.dirty_bitmap, 0, dirty_bytes);
763 mutex_lock(&kvm->lock);
765 if (memory_config_version != kvm->memory_config_version) {
766 mutex_unlock(&kvm->lock);
767 kvm_free_physmem_slot(&new, &old);
775 if (mem->slot >= kvm->nmemslots)
776 kvm->nmemslots = mem->slot + 1;
779 ++kvm->memory_config_version;
781 kvm_mmu_slot_remove_write_access(kvm, mem->slot);
782 kvm_flush_remote_tlbs(kvm);
784 mutex_unlock(&kvm->lock);
786 kvm_free_physmem_slot(&old, &new);
790 mutex_unlock(&kvm->lock);
792 kvm_free_physmem_slot(&new, &old);
798 * Get (and clear) the dirty memory log for a memory slot.
800 static int kvm_vm_ioctl_get_dirty_log(struct kvm *kvm,
801 struct kvm_dirty_log *log)
803 struct kvm_memory_slot *memslot;
806 unsigned long any = 0;
808 mutex_lock(&kvm->lock);
811 * Prevent changes to guest memory configuration even while the lock
815 mutex_unlock(&kvm->lock);
817 if (log->slot >= KVM_MEMORY_SLOTS)
820 memslot = &kvm->memslots[log->slot];
822 if (!memslot->dirty_bitmap)
825 n = ALIGN(memslot->npages, BITS_PER_LONG) / 8;
827 for (i = 0; !any && i < n/sizeof(long); ++i)
828 any = memslot->dirty_bitmap[i];
831 if (copy_to_user(log->dirty_bitmap, memslot->dirty_bitmap, n))
834 /* If nothing is dirty, don't bother messing with page tables. */
836 mutex_lock(&kvm->lock);
837 kvm_mmu_slot_remove_write_access(kvm, log->slot);
838 kvm_flush_remote_tlbs(kvm);
839 memset(memslot->dirty_bitmap, 0, n);
840 mutex_unlock(&kvm->lock);
846 mutex_lock(&kvm->lock);
848 mutex_unlock(&kvm->lock);
853 * Set a new alias region. Aliases map a portion of physical memory into
854 * another portion. This is useful for memory windows, for example the PC
857 static int kvm_vm_ioctl_set_memory_alias(struct kvm *kvm,
858 struct kvm_memory_alias *alias)
861 struct kvm_mem_alias *p;
864 /* General sanity checks */
865 if (alias->memory_size & (PAGE_SIZE - 1))
867 if (alias->guest_phys_addr & (PAGE_SIZE - 1))
869 if (alias->slot >= KVM_ALIAS_SLOTS)
871 if (alias->guest_phys_addr + alias->memory_size
872 < alias->guest_phys_addr)
874 if (alias->target_phys_addr + alias->memory_size
875 < alias->target_phys_addr)
878 mutex_lock(&kvm->lock);
880 p = &kvm->aliases[alias->slot];
881 p->base_gfn = alias->guest_phys_addr >> PAGE_SHIFT;
882 p->npages = alias->memory_size >> PAGE_SHIFT;
883 p->target_gfn = alias->target_phys_addr >> PAGE_SHIFT;
885 for (n = KVM_ALIAS_SLOTS; n > 0; --n)
886 if (kvm->aliases[n - 1].npages)
890 kvm_mmu_zap_all(kvm);
892 mutex_unlock(&kvm->lock);
900 static int kvm_vm_ioctl_get_irqchip(struct kvm *kvm, struct kvm_irqchip *chip)
905 switch (chip->chip_id) {
906 case KVM_IRQCHIP_PIC_MASTER:
907 memcpy (&chip->chip.pic,
908 &pic_irqchip(kvm)->pics[0],
909 sizeof(struct kvm_pic_state));
911 case KVM_IRQCHIP_PIC_SLAVE:
912 memcpy (&chip->chip.pic,
913 &pic_irqchip(kvm)->pics[1],
914 sizeof(struct kvm_pic_state));
916 case KVM_IRQCHIP_IOAPIC:
917 memcpy (&chip->chip.ioapic,
919 sizeof(struct kvm_ioapic_state));
928 static int kvm_vm_ioctl_set_irqchip(struct kvm *kvm, struct kvm_irqchip *chip)
933 switch (chip->chip_id) {
934 case KVM_IRQCHIP_PIC_MASTER:
935 memcpy (&pic_irqchip(kvm)->pics[0],
937 sizeof(struct kvm_pic_state));
939 case KVM_IRQCHIP_PIC_SLAVE:
940 memcpy (&pic_irqchip(kvm)->pics[1],
942 sizeof(struct kvm_pic_state));
944 case KVM_IRQCHIP_IOAPIC:
945 memcpy (ioapic_irqchip(kvm),
947 sizeof(struct kvm_ioapic_state));
953 kvm_pic_update_irq(pic_irqchip(kvm));
957 static gfn_t unalias_gfn(struct kvm *kvm, gfn_t gfn)
960 struct kvm_mem_alias *alias;
962 for (i = 0; i < kvm->naliases; ++i) {
963 alias = &kvm->aliases[i];
964 if (gfn >= alias->base_gfn
965 && gfn < alias->base_gfn + alias->npages)
966 return alias->target_gfn + gfn - alias->base_gfn;
971 static struct kvm_memory_slot *__gfn_to_memslot(struct kvm *kvm, gfn_t gfn)
975 for (i = 0; i < kvm->nmemslots; ++i) {
976 struct kvm_memory_slot *memslot = &kvm->memslots[i];
978 if (gfn >= memslot->base_gfn
979 && gfn < memslot->base_gfn + memslot->npages)
985 struct kvm_memory_slot *gfn_to_memslot(struct kvm *kvm, gfn_t gfn)
987 gfn = unalias_gfn(kvm, gfn);
988 return __gfn_to_memslot(kvm, gfn);
991 struct page *gfn_to_page(struct kvm *kvm, gfn_t gfn)
993 struct kvm_memory_slot *slot;
995 gfn = unalias_gfn(kvm, gfn);
996 slot = __gfn_to_memslot(kvm, gfn);
999 return slot->phys_mem[gfn - slot->base_gfn];
1001 EXPORT_SYMBOL_GPL(gfn_to_page);
1003 /* WARNING: Does not work on aliased pages. */
1004 void mark_page_dirty(struct kvm *kvm, gfn_t gfn)
1006 struct kvm_memory_slot *memslot;
1008 memslot = __gfn_to_memslot(kvm, gfn);
1009 if (memslot && memslot->dirty_bitmap) {
1010 unsigned long rel_gfn = gfn - memslot->base_gfn;
1013 if (!test_bit(rel_gfn, memslot->dirty_bitmap))
1014 set_bit(rel_gfn, memslot->dirty_bitmap);
1018 int emulator_read_std(unsigned long addr,
1021 struct kvm_vcpu *vcpu)
1026 gpa_t gpa = vcpu->mmu.gva_to_gpa(vcpu, addr);
1027 unsigned offset = addr & (PAGE_SIZE-1);
1028 unsigned tocopy = min(bytes, (unsigned)PAGE_SIZE - offset);
1033 if (gpa == UNMAPPED_GVA)
1034 return X86EMUL_PROPAGATE_FAULT;
1035 pfn = gpa >> PAGE_SHIFT;
1036 page = gfn_to_page(vcpu->kvm, pfn);
1038 return X86EMUL_UNHANDLEABLE;
1039 page_virt = kmap_atomic(page, KM_USER0);
1041 memcpy(data, page_virt + offset, tocopy);
1043 kunmap_atomic(page_virt, KM_USER0);
1050 return X86EMUL_CONTINUE;
1052 EXPORT_SYMBOL_GPL(emulator_read_std);
1054 static int emulator_write_std(unsigned long addr,
1057 struct kvm_vcpu *vcpu)
1059 pr_unimpl(vcpu, "emulator_write_std: addr %lx n %d\n", addr, bytes);
1060 return X86EMUL_UNHANDLEABLE;
1064 * Only apic need an MMIO device hook, so shortcut now..
1066 static struct kvm_io_device *vcpu_find_pervcpu_dev(struct kvm_vcpu *vcpu,
1069 struct kvm_io_device *dev;
1072 dev = &vcpu->apic->dev;
1073 if (dev->in_range(dev, addr))
1079 static struct kvm_io_device *vcpu_find_mmio_dev(struct kvm_vcpu *vcpu,
1082 struct kvm_io_device *dev;
1084 dev = vcpu_find_pervcpu_dev(vcpu, addr);
1086 dev = kvm_io_bus_find_dev(&vcpu->kvm->mmio_bus, addr);
1090 static struct kvm_io_device *vcpu_find_pio_dev(struct kvm_vcpu *vcpu,
1093 return kvm_io_bus_find_dev(&vcpu->kvm->pio_bus, addr);
1096 static int emulator_read_emulated(unsigned long addr,
1099 struct kvm_vcpu *vcpu)
1101 struct kvm_io_device *mmio_dev;
1104 if (vcpu->mmio_read_completed) {
1105 memcpy(val, vcpu->mmio_data, bytes);
1106 vcpu->mmio_read_completed = 0;
1107 return X86EMUL_CONTINUE;
1108 } else if (emulator_read_std(addr, val, bytes, vcpu)
1109 == X86EMUL_CONTINUE)
1110 return X86EMUL_CONTINUE;
1112 gpa = vcpu->mmu.gva_to_gpa(vcpu, addr);
1113 if (gpa == UNMAPPED_GVA)
1114 return X86EMUL_PROPAGATE_FAULT;
1117 * Is this MMIO handled locally?
1119 mmio_dev = vcpu_find_mmio_dev(vcpu, gpa);
1121 kvm_iodevice_read(mmio_dev, gpa, bytes, val);
1122 return X86EMUL_CONTINUE;
1125 vcpu->mmio_needed = 1;
1126 vcpu->mmio_phys_addr = gpa;
1127 vcpu->mmio_size = bytes;
1128 vcpu->mmio_is_write = 0;
1130 return X86EMUL_UNHANDLEABLE;
1133 static int emulator_write_phys(struct kvm_vcpu *vcpu, gpa_t gpa,
1134 const void *val, int bytes)
1139 if (((gpa + bytes - 1) >> PAGE_SHIFT) != (gpa >> PAGE_SHIFT))
1141 page = gfn_to_page(vcpu->kvm, gpa >> PAGE_SHIFT);
1144 mark_page_dirty(vcpu->kvm, gpa >> PAGE_SHIFT);
1145 virt = kmap_atomic(page, KM_USER0);
1146 kvm_mmu_pte_write(vcpu, gpa, val, bytes);
1147 memcpy(virt + offset_in_page(gpa), val, bytes);
1148 kunmap_atomic(virt, KM_USER0);
1152 static int emulator_write_emulated_onepage(unsigned long addr,
1155 struct kvm_vcpu *vcpu)
1157 struct kvm_io_device *mmio_dev;
1158 gpa_t gpa = vcpu->mmu.gva_to_gpa(vcpu, addr);
1160 if (gpa == UNMAPPED_GVA) {
1161 kvm_arch_ops->inject_page_fault(vcpu, addr, 2);
1162 return X86EMUL_PROPAGATE_FAULT;
1165 if (emulator_write_phys(vcpu, gpa, val, bytes))
1166 return X86EMUL_CONTINUE;
1169 * Is this MMIO handled locally?
1171 mmio_dev = vcpu_find_mmio_dev(vcpu, gpa);
1173 kvm_iodevice_write(mmio_dev, gpa, bytes, val);
1174 return X86EMUL_CONTINUE;
1177 vcpu->mmio_needed = 1;
1178 vcpu->mmio_phys_addr = gpa;
1179 vcpu->mmio_size = bytes;
1180 vcpu->mmio_is_write = 1;
1181 memcpy(vcpu->mmio_data, val, bytes);
1183 return X86EMUL_CONTINUE;
1186 int emulator_write_emulated(unsigned long addr,
1189 struct kvm_vcpu *vcpu)
1191 /* Crossing a page boundary? */
1192 if (((addr + bytes - 1) ^ addr) & PAGE_MASK) {
1195 now = -addr & ~PAGE_MASK;
1196 rc = emulator_write_emulated_onepage(addr, val, now, vcpu);
1197 if (rc != X86EMUL_CONTINUE)
1203 return emulator_write_emulated_onepage(addr, val, bytes, vcpu);
1205 EXPORT_SYMBOL_GPL(emulator_write_emulated);
1207 static int emulator_cmpxchg_emulated(unsigned long addr,
1211 struct kvm_vcpu *vcpu)
1213 static int reported;
1217 printk(KERN_WARNING "kvm: emulating exchange as write\n");
1219 return emulator_write_emulated(addr, new, bytes, vcpu);
1222 static unsigned long get_segment_base(struct kvm_vcpu *vcpu, int seg)
1224 return kvm_arch_ops->get_segment_base(vcpu, seg);
1227 int emulate_invlpg(struct kvm_vcpu *vcpu, gva_t address)
1229 return X86EMUL_CONTINUE;
1232 int emulate_clts(struct kvm_vcpu *vcpu)
1236 cr0 = vcpu->cr0 & ~X86_CR0_TS;
1237 kvm_arch_ops->set_cr0(vcpu, cr0);
1238 return X86EMUL_CONTINUE;
1241 int emulator_get_dr(struct x86_emulate_ctxt* ctxt, int dr, unsigned long *dest)
1243 struct kvm_vcpu *vcpu = ctxt->vcpu;
1247 *dest = kvm_arch_ops->get_dr(vcpu, dr);
1248 return X86EMUL_CONTINUE;
1250 pr_unimpl(vcpu, "%s: unexpected dr %u\n", __FUNCTION__, dr);
1251 return X86EMUL_UNHANDLEABLE;
1255 int emulator_set_dr(struct x86_emulate_ctxt *ctxt, int dr, unsigned long value)
1257 unsigned long mask = (ctxt->mode == X86EMUL_MODE_PROT64) ? ~0ULL : ~0U;
1260 kvm_arch_ops->set_dr(ctxt->vcpu, dr, value & mask, &exception);
1262 /* FIXME: better handling */
1263 return X86EMUL_UNHANDLEABLE;
1265 return X86EMUL_CONTINUE;
1268 static void report_emulation_failure(struct x86_emulate_ctxt *ctxt)
1270 static int reported;
1272 unsigned long rip = ctxt->vcpu->rip;
1273 unsigned long rip_linear;
1275 rip_linear = rip + get_segment_base(ctxt->vcpu, VCPU_SREG_CS);
1280 emulator_read_std(rip_linear, (void *)opcodes, 4, ctxt->vcpu);
1282 printk(KERN_ERR "emulation failed but !mmio_needed?"
1283 " rip %lx %02x %02x %02x %02x\n",
1284 rip, opcodes[0], opcodes[1], opcodes[2], opcodes[3]);
1288 struct x86_emulate_ops emulate_ops = {
1289 .read_std = emulator_read_std,
1290 .write_std = emulator_write_std,
1291 .read_emulated = emulator_read_emulated,
1292 .write_emulated = emulator_write_emulated,
1293 .cmpxchg_emulated = emulator_cmpxchg_emulated,
1296 int emulate_instruction(struct kvm_vcpu *vcpu,
1297 struct kvm_run *run,
1301 struct x86_emulate_ctxt emulate_ctxt;
1305 vcpu->mmio_fault_cr2 = cr2;
1306 kvm_arch_ops->cache_regs(vcpu);
1308 kvm_arch_ops->get_cs_db_l_bits(vcpu, &cs_db, &cs_l);
1310 emulate_ctxt.vcpu = vcpu;
1311 emulate_ctxt.eflags = kvm_arch_ops->get_rflags(vcpu);
1312 emulate_ctxt.cr2 = cr2;
1313 emulate_ctxt.mode = (emulate_ctxt.eflags & X86_EFLAGS_VM)
1314 ? X86EMUL_MODE_REAL : cs_l
1315 ? X86EMUL_MODE_PROT64 : cs_db
1316 ? X86EMUL_MODE_PROT32 : X86EMUL_MODE_PROT16;
1318 if (emulate_ctxt.mode == X86EMUL_MODE_PROT64) {
1319 emulate_ctxt.cs_base = 0;
1320 emulate_ctxt.ds_base = 0;
1321 emulate_ctxt.es_base = 0;
1322 emulate_ctxt.ss_base = 0;
1324 emulate_ctxt.cs_base = get_segment_base(vcpu, VCPU_SREG_CS);
1325 emulate_ctxt.ds_base = get_segment_base(vcpu, VCPU_SREG_DS);
1326 emulate_ctxt.es_base = get_segment_base(vcpu, VCPU_SREG_ES);
1327 emulate_ctxt.ss_base = get_segment_base(vcpu, VCPU_SREG_SS);
1330 emulate_ctxt.gs_base = get_segment_base(vcpu, VCPU_SREG_GS);
1331 emulate_ctxt.fs_base = get_segment_base(vcpu, VCPU_SREG_FS);
1333 vcpu->mmio_is_write = 0;
1334 vcpu->pio.string = 0;
1335 r = x86_emulate_memop(&emulate_ctxt, &emulate_ops);
1336 if (vcpu->pio.string)
1337 return EMULATE_DO_MMIO;
1339 if ((r || vcpu->mmio_is_write) && run) {
1340 run->exit_reason = KVM_EXIT_MMIO;
1341 run->mmio.phys_addr = vcpu->mmio_phys_addr;
1342 memcpy(run->mmio.data, vcpu->mmio_data, 8);
1343 run->mmio.len = vcpu->mmio_size;
1344 run->mmio.is_write = vcpu->mmio_is_write;
1348 if (kvm_mmu_unprotect_page_virt(vcpu, cr2))
1349 return EMULATE_DONE;
1350 if (!vcpu->mmio_needed) {
1351 report_emulation_failure(&emulate_ctxt);
1352 return EMULATE_FAIL;
1354 return EMULATE_DO_MMIO;
1357 kvm_arch_ops->decache_regs(vcpu);
1358 kvm_arch_ops->set_rflags(vcpu, emulate_ctxt.eflags);
1360 if (vcpu->mmio_is_write) {
1361 vcpu->mmio_needed = 0;
1362 return EMULATE_DO_MMIO;
1365 return EMULATE_DONE;
1367 EXPORT_SYMBOL_GPL(emulate_instruction);
1370 * The vCPU has executed a HLT instruction with in-kernel mode enabled.
1372 static void kvm_vcpu_kernel_halt(struct kvm_vcpu *vcpu)
1374 DECLARE_WAITQUEUE(wait, current);
1376 add_wait_queue(&vcpu->wq, &wait);
1379 * We will block until either an interrupt or a signal wakes us up
1381 while(!(irqchip_in_kernel(vcpu->kvm) && kvm_cpu_has_interrupt(vcpu))
1382 && !vcpu->irq_summary
1383 && !signal_pending(current)) {
1384 set_current_state(TASK_INTERRUPTIBLE);
1390 remove_wait_queue(&vcpu->wq, &wait);
1391 set_current_state(TASK_RUNNING);
1394 int kvm_emulate_halt(struct kvm_vcpu *vcpu)
1396 ++vcpu->stat.halt_exits;
1397 if (irqchip_in_kernel(vcpu->kvm)) {
1398 kvm_vcpu_kernel_halt(vcpu);
1401 vcpu->run->exit_reason = KVM_EXIT_HLT;
1405 EXPORT_SYMBOL_GPL(kvm_emulate_halt);
1407 int kvm_hypercall(struct kvm_vcpu *vcpu, struct kvm_run *run)
1409 unsigned long nr, a0, a1, a2, a3, a4, a5, ret;
1411 kvm_arch_ops->cache_regs(vcpu);
1413 #ifdef CONFIG_X86_64
1414 if (is_long_mode(vcpu)) {
1415 nr = vcpu->regs[VCPU_REGS_RAX];
1416 a0 = vcpu->regs[VCPU_REGS_RDI];
1417 a1 = vcpu->regs[VCPU_REGS_RSI];
1418 a2 = vcpu->regs[VCPU_REGS_RDX];
1419 a3 = vcpu->regs[VCPU_REGS_RCX];
1420 a4 = vcpu->regs[VCPU_REGS_R8];
1421 a5 = vcpu->regs[VCPU_REGS_R9];
1425 nr = vcpu->regs[VCPU_REGS_RBX] & -1u;
1426 a0 = vcpu->regs[VCPU_REGS_RAX] & -1u;
1427 a1 = vcpu->regs[VCPU_REGS_RCX] & -1u;
1428 a2 = vcpu->regs[VCPU_REGS_RDX] & -1u;
1429 a3 = vcpu->regs[VCPU_REGS_RSI] & -1u;
1430 a4 = vcpu->regs[VCPU_REGS_RDI] & -1u;
1431 a5 = vcpu->regs[VCPU_REGS_RBP] & -1u;
1435 run->hypercall.nr = nr;
1436 run->hypercall.args[0] = a0;
1437 run->hypercall.args[1] = a1;
1438 run->hypercall.args[2] = a2;
1439 run->hypercall.args[3] = a3;
1440 run->hypercall.args[4] = a4;
1441 run->hypercall.args[5] = a5;
1442 run->hypercall.ret = ret;
1443 run->hypercall.longmode = is_long_mode(vcpu);
1444 kvm_arch_ops->decache_regs(vcpu);
1447 vcpu->regs[VCPU_REGS_RAX] = ret;
1448 kvm_arch_ops->decache_regs(vcpu);
1451 EXPORT_SYMBOL_GPL(kvm_hypercall);
1453 static u64 mk_cr_64(u64 curr_cr, u32 new_val)
1455 return (curr_cr & ~((1ULL << 32) - 1)) | new_val;
1458 void realmode_lgdt(struct kvm_vcpu *vcpu, u16 limit, unsigned long base)
1460 struct descriptor_table dt = { limit, base };
1462 kvm_arch_ops->set_gdt(vcpu, &dt);
1465 void realmode_lidt(struct kvm_vcpu *vcpu, u16 limit, unsigned long base)
1467 struct descriptor_table dt = { limit, base };
1469 kvm_arch_ops->set_idt(vcpu, &dt);
1472 void realmode_lmsw(struct kvm_vcpu *vcpu, unsigned long msw,
1473 unsigned long *rflags)
1476 *rflags = kvm_arch_ops->get_rflags(vcpu);
1479 unsigned long realmode_get_cr(struct kvm_vcpu *vcpu, int cr)
1481 kvm_arch_ops->decache_cr4_guest_bits(vcpu);
1492 vcpu_printf(vcpu, "%s: unexpected cr %u\n", __FUNCTION__, cr);
1497 void realmode_set_cr(struct kvm_vcpu *vcpu, int cr, unsigned long val,
1498 unsigned long *rflags)
1502 set_cr0(vcpu, mk_cr_64(vcpu->cr0, val));
1503 *rflags = kvm_arch_ops->get_rflags(vcpu);
1512 set_cr4(vcpu, mk_cr_64(vcpu->cr4, val));
1515 vcpu_printf(vcpu, "%s: unexpected cr %u\n", __FUNCTION__, cr);
1520 * Register the para guest with the host:
1522 static int vcpu_register_para(struct kvm_vcpu *vcpu, gpa_t para_state_gpa)
1524 struct kvm_vcpu_para_state *para_state;
1525 hpa_t para_state_hpa, hypercall_hpa;
1526 struct page *para_state_page;
1527 unsigned char *hypercall;
1528 gpa_t hypercall_gpa;
1530 printk(KERN_DEBUG "kvm: guest trying to enter paravirtual mode\n");
1531 printk(KERN_DEBUG ".... para_state_gpa: %08Lx\n", para_state_gpa);
1534 * Needs to be page aligned:
1536 if (para_state_gpa != PAGE_ALIGN(para_state_gpa))
1539 para_state_hpa = gpa_to_hpa(vcpu, para_state_gpa);
1540 printk(KERN_DEBUG ".... para_state_hpa: %08Lx\n", para_state_hpa);
1541 if (is_error_hpa(para_state_hpa))
1544 mark_page_dirty(vcpu->kvm, para_state_gpa >> PAGE_SHIFT);
1545 para_state_page = pfn_to_page(para_state_hpa >> PAGE_SHIFT);
1546 para_state = kmap(para_state_page);
1548 printk(KERN_DEBUG ".... guest version: %d\n", para_state->guest_version);
1549 printk(KERN_DEBUG ".... size: %d\n", para_state->size);
1551 para_state->host_version = KVM_PARA_API_VERSION;
1553 * We cannot support guests that try to register themselves
1554 * with a newer API version than the host supports:
1556 if (para_state->guest_version > KVM_PARA_API_VERSION) {
1557 para_state->ret = -KVM_EINVAL;
1558 goto err_kunmap_skip;
1561 hypercall_gpa = para_state->hypercall_gpa;
1562 hypercall_hpa = gpa_to_hpa(vcpu, hypercall_gpa);
1563 printk(KERN_DEBUG ".... hypercall_hpa: %08Lx\n", hypercall_hpa);
1564 if (is_error_hpa(hypercall_hpa)) {
1565 para_state->ret = -KVM_EINVAL;
1566 goto err_kunmap_skip;
1569 printk(KERN_DEBUG "kvm: para guest successfully registered.\n");
1570 vcpu->para_state_page = para_state_page;
1571 vcpu->para_state_gpa = para_state_gpa;
1572 vcpu->hypercall_gpa = hypercall_gpa;
1574 mark_page_dirty(vcpu->kvm, hypercall_gpa >> PAGE_SHIFT);
1575 hypercall = kmap_atomic(pfn_to_page(hypercall_hpa >> PAGE_SHIFT),
1576 KM_USER1) + (hypercall_hpa & ~PAGE_MASK);
1577 kvm_arch_ops->patch_hypercall(vcpu, hypercall);
1578 kunmap_atomic(hypercall, KM_USER1);
1580 para_state->ret = 0;
1582 kunmap(para_state_page);
1588 int kvm_get_msr_common(struct kvm_vcpu *vcpu, u32 msr, u64 *pdata)
1593 case 0xc0010010: /* SYSCFG */
1594 case 0xc0010015: /* HWCR */
1595 case MSR_IA32_PLATFORM_ID:
1596 case MSR_IA32_P5_MC_ADDR:
1597 case MSR_IA32_P5_MC_TYPE:
1598 case MSR_IA32_MC0_CTL:
1599 case MSR_IA32_MCG_STATUS:
1600 case MSR_IA32_MCG_CAP:
1601 case MSR_IA32_MC0_MISC:
1602 case MSR_IA32_MC0_MISC+4:
1603 case MSR_IA32_MC0_MISC+8:
1604 case MSR_IA32_MC0_MISC+12:
1605 case MSR_IA32_MC0_MISC+16:
1606 case MSR_IA32_UCODE_REV:
1607 case MSR_IA32_PERF_STATUS:
1608 case MSR_IA32_EBL_CR_POWERON:
1609 /* MTRR registers */
1611 case 0x200 ... 0x2ff:
1614 case 0xcd: /* fsb frequency */
1617 case MSR_IA32_APICBASE:
1618 data = kvm_get_apic_base(vcpu);
1620 case MSR_IA32_MISC_ENABLE:
1621 data = vcpu->ia32_misc_enable_msr;
1623 #ifdef CONFIG_X86_64
1625 data = vcpu->shadow_efer;
1629 pr_unimpl(vcpu, "unhandled rdmsr: 0x%x\n", msr);
1635 EXPORT_SYMBOL_GPL(kvm_get_msr_common);
1638 * Reads an msr value (of 'msr_index') into 'pdata'.
1639 * Returns 0 on success, non-0 otherwise.
1640 * Assumes vcpu_load() was already called.
1642 int kvm_get_msr(struct kvm_vcpu *vcpu, u32 msr_index, u64 *pdata)
1644 return kvm_arch_ops->get_msr(vcpu, msr_index, pdata);
1647 #ifdef CONFIG_X86_64
1649 static void set_efer(struct kvm_vcpu *vcpu, u64 efer)
1651 if (efer & EFER_RESERVED_BITS) {
1652 printk(KERN_DEBUG "set_efer: 0x%llx #GP, reserved bits\n",
1659 && (vcpu->shadow_efer & EFER_LME) != (efer & EFER_LME)) {
1660 printk(KERN_DEBUG "set_efer: #GP, change LME while paging\n");
1665 kvm_arch_ops->set_efer(vcpu, efer);
1668 efer |= vcpu->shadow_efer & EFER_LMA;
1670 vcpu->shadow_efer = efer;
1675 int kvm_set_msr_common(struct kvm_vcpu *vcpu, u32 msr, u64 data)
1678 #ifdef CONFIG_X86_64
1680 set_efer(vcpu, data);
1683 case MSR_IA32_MC0_STATUS:
1684 pr_unimpl(vcpu, "%s: MSR_IA32_MC0_STATUS 0x%llx, nop\n",
1685 __FUNCTION__, data);
1687 case MSR_IA32_MCG_STATUS:
1688 pr_unimpl(vcpu, "%s: MSR_IA32_MCG_STATUS 0x%llx, nop\n",
1689 __FUNCTION__, data);
1691 case MSR_IA32_UCODE_REV:
1692 case MSR_IA32_UCODE_WRITE:
1693 case 0x200 ... 0x2ff: /* MTRRs */
1695 case MSR_IA32_APICBASE:
1696 kvm_set_apic_base(vcpu, data);
1698 case MSR_IA32_MISC_ENABLE:
1699 vcpu->ia32_misc_enable_msr = data;
1702 * This is the 'probe whether the host is KVM' logic:
1704 case MSR_KVM_API_MAGIC:
1705 return vcpu_register_para(vcpu, data);
1708 pr_unimpl(vcpu, "unhandled wrmsr: 0x%x\n", msr);
1713 EXPORT_SYMBOL_GPL(kvm_set_msr_common);
1716 * Writes msr value into into the appropriate "register".
1717 * Returns 0 on success, non-0 otherwise.
1718 * Assumes vcpu_load() was already called.
1720 int kvm_set_msr(struct kvm_vcpu *vcpu, u32 msr_index, u64 data)
1722 return kvm_arch_ops->set_msr(vcpu, msr_index, data);
1725 void kvm_resched(struct kvm_vcpu *vcpu)
1727 if (!need_resched())
1731 EXPORT_SYMBOL_GPL(kvm_resched);
1733 void kvm_emulate_cpuid(struct kvm_vcpu *vcpu)
1737 struct kvm_cpuid_entry *e, *best;
1739 kvm_arch_ops->cache_regs(vcpu);
1740 function = vcpu->regs[VCPU_REGS_RAX];
1741 vcpu->regs[VCPU_REGS_RAX] = 0;
1742 vcpu->regs[VCPU_REGS_RBX] = 0;
1743 vcpu->regs[VCPU_REGS_RCX] = 0;
1744 vcpu->regs[VCPU_REGS_RDX] = 0;
1746 for (i = 0; i < vcpu->cpuid_nent; ++i) {
1747 e = &vcpu->cpuid_entries[i];
1748 if (e->function == function) {
1753 * Both basic or both extended?
1755 if (((e->function ^ function) & 0x80000000) == 0)
1756 if (!best || e->function > best->function)
1760 vcpu->regs[VCPU_REGS_RAX] = best->eax;
1761 vcpu->regs[VCPU_REGS_RBX] = best->ebx;
1762 vcpu->regs[VCPU_REGS_RCX] = best->ecx;
1763 vcpu->regs[VCPU_REGS_RDX] = best->edx;
1765 kvm_arch_ops->decache_regs(vcpu);
1766 kvm_arch_ops->skip_emulated_instruction(vcpu);
1768 EXPORT_SYMBOL_GPL(kvm_emulate_cpuid);
1770 static int pio_copy_data(struct kvm_vcpu *vcpu)
1772 void *p = vcpu->pio_data;
1775 int nr_pages = vcpu->pio.guest_pages[1] ? 2 : 1;
1777 q = vmap(vcpu->pio.guest_pages, nr_pages, VM_READ|VM_WRITE,
1780 free_pio_guest_pages(vcpu);
1783 q += vcpu->pio.guest_page_offset;
1784 bytes = vcpu->pio.size * vcpu->pio.cur_count;
1786 memcpy(q, p, bytes);
1788 memcpy(p, q, bytes);
1789 q -= vcpu->pio.guest_page_offset;
1791 free_pio_guest_pages(vcpu);
1795 static int complete_pio(struct kvm_vcpu *vcpu)
1797 struct kvm_pio_request *io = &vcpu->pio;
1801 kvm_arch_ops->cache_regs(vcpu);
1805 memcpy(&vcpu->regs[VCPU_REGS_RAX], vcpu->pio_data,
1809 r = pio_copy_data(vcpu);
1811 kvm_arch_ops->cache_regs(vcpu);
1818 delta *= io->cur_count;
1820 * The size of the register should really depend on
1821 * current address size.
1823 vcpu->regs[VCPU_REGS_RCX] -= delta;
1829 vcpu->regs[VCPU_REGS_RDI] += delta;
1831 vcpu->regs[VCPU_REGS_RSI] += delta;
1834 kvm_arch_ops->decache_regs(vcpu);
1836 io->count -= io->cur_count;
1840 kvm_arch_ops->skip_emulated_instruction(vcpu);
1844 static void kernel_pio(struct kvm_io_device *pio_dev,
1845 struct kvm_vcpu *vcpu,
1848 /* TODO: String I/O for in kernel device */
1850 mutex_lock(&vcpu->kvm->lock);
1852 kvm_iodevice_read(pio_dev, vcpu->pio.port,
1856 kvm_iodevice_write(pio_dev, vcpu->pio.port,
1859 mutex_unlock(&vcpu->kvm->lock);
1862 static void pio_string_write(struct kvm_io_device *pio_dev,
1863 struct kvm_vcpu *vcpu)
1865 struct kvm_pio_request *io = &vcpu->pio;
1866 void *pd = vcpu->pio_data;
1869 mutex_lock(&vcpu->kvm->lock);
1870 for (i = 0; i < io->cur_count; i++) {
1871 kvm_iodevice_write(pio_dev, io->port,
1876 mutex_unlock(&vcpu->kvm->lock);
1879 int kvm_emulate_pio (struct kvm_vcpu *vcpu, struct kvm_run *run, int in,
1880 int size, unsigned port)
1882 struct kvm_io_device *pio_dev;
1884 vcpu->run->exit_reason = KVM_EXIT_IO;
1885 vcpu->run->io.direction = in ? KVM_EXIT_IO_IN : KVM_EXIT_IO_OUT;
1886 vcpu->run->io.size = vcpu->pio.size = size;
1887 vcpu->run->io.data_offset = KVM_PIO_PAGE_OFFSET * PAGE_SIZE;
1888 vcpu->run->io.count = vcpu->pio.count = vcpu->pio.cur_count = 1;
1889 vcpu->run->io.port = vcpu->pio.port = port;
1891 vcpu->pio.string = 0;
1893 vcpu->pio.guest_page_offset = 0;
1896 kvm_arch_ops->cache_regs(vcpu);
1897 memcpy(vcpu->pio_data, &vcpu->regs[VCPU_REGS_RAX], 4);
1898 kvm_arch_ops->decache_regs(vcpu);
1900 pio_dev = vcpu_find_pio_dev(vcpu, port);
1902 kernel_pio(pio_dev, vcpu, vcpu->pio_data);
1908 EXPORT_SYMBOL_GPL(kvm_emulate_pio);
1910 int kvm_emulate_pio_string(struct kvm_vcpu *vcpu, struct kvm_run *run, int in,
1911 int size, unsigned long count, int down,
1912 gva_t address, int rep, unsigned port)
1914 unsigned now, in_page;
1918 struct kvm_io_device *pio_dev;
1920 vcpu->run->exit_reason = KVM_EXIT_IO;
1921 vcpu->run->io.direction = in ? KVM_EXIT_IO_IN : KVM_EXIT_IO_OUT;
1922 vcpu->run->io.size = vcpu->pio.size = size;
1923 vcpu->run->io.data_offset = KVM_PIO_PAGE_OFFSET * PAGE_SIZE;
1924 vcpu->run->io.count = vcpu->pio.count = vcpu->pio.cur_count = count;
1925 vcpu->run->io.port = vcpu->pio.port = port;
1927 vcpu->pio.string = 1;
1928 vcpu->pio.down = down;
1929 vcpu->pio.guest_page_offset = offset_in_page(address);
1930 vcpu->pio.rep = rep;
1933 kvm_arch_ops->skip_emulated_instruction(vcpu);
1938 in_page = PAGE_SIZE - offset_in_page(address);
1940 in_page = offset_in_page(address) + size;
1941 now = min(count, (unsigned long)in_page / size);
1944 * String I/O straddles page boundary. Pin two guest pages
1945 * so that we satisfy atomicity constraints. Do just one
1946 * transaction to avoid complexity.
1953 * String I/O in reverse. Yuck. Kill the guest, fix later.
1955 pr_unimpl(vcpu, "guest string pio down\n");
1959 vcpu->run->io.count = now;
1960 vcpu->pio.cur_count = now;
1962 for (i = 0; i < nr_pages; ++i) {
1963 mutex_lock(&vcpu->kvm->lock);
1964 page = gva_to_page(vcpu, address + i * PAGE_SIZE);
1967 vcpu->pio.guest_pages[i] = page;
1968 mutex_unlock(&vcpu->kvm->lock);
1971 free_pio_guest_pages(vcpu);
1976 pio_dev = vcpu_find_pio_dev(vcpu, port);
1977 if (!vcpu->pio.in) {
1978 /* string PIO write */
1979 ret = pio_copy_data(vcpu);
1980 if (ret >= 0 && pio_dev) {
1981 pio_string_write(pio_dev, vcpu);
1983 if (vcpu->pio.count == 0)
1987 pr_unimpl(vcpu, "no string pio read support yet, "
1988 "port %x size %d count %ld\n",
1993 EXPORT_SYMBOL_GPL(kvm_emulate_pio_string);
1995 static int kvm_vcpu_ioctl_run(struct kvm_vcpu *vcpu, struct kvm_run *kvm_run)
2002 if (vcpu->sigset_active)
2003 sigprocmask(SIG_SETMASK, &vcpu->sigset, &sigsaved);
2005 /* re-sync apic's tpr */
2006 set_cr8(vcpu, kvm_run->cr8);
2008 if (vcpu->pio.cur_count) {
2009 r = complete_pio(vcpu);
2014 if (vcpu->mmio_needed) {
2015 memcpy(vcpu->mmio_data, kvm_run->mmio.data, 8);
2016 vcpu->mmio_read_completed = 1;
2017 vcpu->mmio_needed = 0;
2018 r = emulate_instruction(vcpu, kvm_run,
2019 vcpu->mmio_fault_cr2, 0);
2020 if (r == EMULATE_DO_MMIO) {
2022 * Read-modify-write. Back to userspace.
2029 if (kvm_run->exit_reason == KVM_EXIT_HYPERCALL) {
2030 kvm_arch_ops->cache_regs(vcpu);
2031 vcpu->regs[VCPU_REGS_RAX] = kvm_run->hypercall.ret;
2032 kvm_arch_ops->decache_regs(vcpu);
2035 r = kvm_arch_ops->run(vcpu, kvm_run);
2038 if (vcpu->sigset_active)
2039 sigprocmask(SIG_SETMASK, &sigsaved, NULL);
2045 static int kvm_vcpu_ioctl_get_regs(struct kvm_vcpu *vcpu,
2046 struct kvm_regs *regs)
2050 kvm_arch_ops->cache_regs(vcpu);
2052 regs->rax = vcpu->regs[VCPU_REGS_RAX];
2053 regs->rbx = vcpu->regs[VCPU_REGS_RBX];
2054 regs->rcx = vcpu->regs[VCPU_REGS_RCX];
2055 regs->rdx = vcpu->regs[VCPU_REGS_RDX];
2056 regs->rsi = vcpu->regs[VCPU_REGS_RSI];
2057 regs->rdi = vcpu->regs[VCPU_REGS_RDI];
2058 regs->rsp = vcpu->regs[VCPU_REGS_RSP];
2059 regs->rbp = vcpu->regs[VCPU_REGS_RBP];
2060 #ifdef CONFIG_X86_64
2061 regs->r8 = vcpu->regs[VCPU_REGS_R8];
2062 regs->r9 = vcpu->regs[VCPU_REGS_R9];
2063 regs->r10 = vcpu->regs[VCPU_REGS_R10];
2064 regs->r11 = vcpu->regs[VCPU_REGS_R11];
2065 regs->r12 = vcpu->regs[VCPU_REGS_R12];
2066 regs->r13 = vcpu->regs[VCPU_REGS_R13];
2067 regs->r14 = vcpu->regs[VCPU_REGS_R14];
2068 regs->r15 = vcpu->regs[VCPU_REGS_R15];
2071 regs->rip = vcpu->rip;
2072 regs->rflags = kvm_arch_ops->get_rflags(vcpu);
2075 * Don't leak debug flags in case they were set for guest debugging
2077 if (vcpu->guest_debug.enabled && vcpu->guest_debug.singlestep)
2078 regs->rflags &= ~(X86_EFLAGS_TF | X86_EFLAGS_RF);
2085 static int kvm_vcpu_ioctl_set_regs(struct kvm_vcpu *vcpu,
2086 struct kvm_regs *regs)
2090 vcpu->regs[VCPU_REGS_RAX] = regs->rax;
2091 vcpu->regs[VCPU_REGS_RBX] = regs->rbx;
2092 vcpu->regs[VCPU_REGS_RCX] = regs->rcx;
2093 vcpu->regs[VCPU_REGS_RDX] = regs->rdx;
2094 vcpu->regs[VCPU_REGS_RSI] = regs->rsi;
2095 vcpu->regs[VCPU_REGS_RDI] = regs->rdi;
2096 vcpu->regs[VCPU_REGS_RSP] = regs->rsp;
2097 vcpu->regs[VCPU_REGS_RBP] = regs->rbp;
2098 #ifdef CONFIG_X86_64
2099 vcpu->regs[VCPU_REGS_R8] = regs->r8;
2100 vcpu->regs[VCPU_REGS_R9] = regs->r9;
2101 vcpu->regs[VCPU_REGS_R10] = regs->r10;
2102 vcpu->regs[VCPU_REGS_R11] = regs->r11;
2103 vcpu->regs[VCPU_REGS_R12] = regs->r12;
2104 vcpu->regs[VCPU_REGS_R13] = regs->r13;
2105 vcpu->regs[VCPU_REGS_R14] = regs->r14;
2106 vcpu->regs[VCPU_REGS_R15] = regs->r15;
2109 vcpu->rip = regs->rip;
2110 kvm_arch_ops->set_rflags(vcpu, regs->rflags);
2112 kvm_arch_ops->decache_regs(vcpu);
2119 static void get_segment(struct kvm_vcpu *vcpu,
2120 struct kvm_segment *var, int seg)
2122 return kvm_arch_ops->get_segment(vcpu, var, seg);
2125 static int kvm_vcpu_ioctl_get_sregs(struct kvm_vcpu *vcpu,
2126 struct kvm_sregs *sregs)
2128 struct descriptor_table dt;
2132 get_segment(vcpu, &sregs->cs, VCPU_SREG_CS);
2133 get_segment(vcpu, &sregs->ds, VCPU_SREG_DS);
2134 get_segment(vcpu, &sregs->es, VCPU_SREG_ES);
2135 get_segment(vcpu, &sregs->fs, VCPU_SREG_FS);
2136 get_segment(vcpu, &sregs->gs, VCPU_SREG_GS);
2137 get_segment(vcpu, &sregs->ss, VCPU_SREG_SS);
2139 get_segment(vcpu, &sregs->tr, VCPU_SREG_TR);
2140 get_segment(vcpu, &sregs->ldt, VCPU_SREG_LDTR);
2142 kvm_arch_ops->get_idt(vcpu, &dt);
2143 sregs->idt.limit = dt.limit;
2144 sregs->idt.base = dt.base;
2145 kvm_arch_ops->get_gdt(vcpu, &dt);
2146 sregs->gdt.limit = dt.limit;
2147 sregs->gdt.base = dt.base;
2149 kvm_arch_ops->decache_cr4_guest_bits(vcpu);
2150 sregs->cr0 = vcpu->cr0;
2151 sregs->cr2 = vcpu->cr2;
2152 sregs->cr3 = vcpu->cr3;
2153 sregs->cr4 = vcpu->cr4;
2154 sregs->cr8 = get_cr8(vcpu);
2155 sregs->efer = vcpu->shadow_efer;
2156 sregs->apic_base = kvm_get_apic_base(vcpu);
2158 if (irqchip_in_kernel(vcpu->kvm))
2159 memset(sregs->interrupt_bitmap, 0,
2160 sizeof sregs->interrupt_bitmap);
2162 memcpy(sregs->interrupt_bitmap, vcpu->irq_pending,
2163 sizeof sregs->interrupt_bitmap);
2170 static void set_segment(struct kvm_vcpu *vcpu,
2171 struct kvm_segment *var, int seg)
2173 return kvm_arch_ops->set_segment(vcpu, var, seg);
2176 static int kvm_vcpu_ioctl_set_sregs(struct kvm_vcpu *vcpu,
2177 struct kvm_sregs *sregs)
2179 int mmu_reset_needed = 0;
2181 struct descriptor_table dt;
2185 dt.limit = sregs->idt.limit;
2186 dt.base = sregs->idt.base;
2187 kvm_arch_ops->set_idt(vcpu, &dt);
2188 dt.limit = sregs->gdt.limit;
2189 dt.base = sregs->gdt.base;
2190 kvm_arch_ops->set_gdt(vcpu, &dt);
2192 vcpu->cr2 = sregs->cr2;
2193 mmu_reset_needed |= vcpu->cr3 != sregs->cr3;
2194 vcpu->cr3 = sregs->cr3;
2196 set_cr8(vcpu, sregs->cr8);
2198 mmu_reset_needed |= vcpu->shadow_efer != sregs->efer;
2199 #ifdef CONFIG_X86_64
2200 kvm_arch_ops->set_efer(vcpu, sregs->efer);
2202 kvm_set_apic_base(vcpu, sregs->apic_base);
2204 kvm_arch_ops->decache_cr4_guest_bits(vcpu);
2206 mmu_reset_needed |= vcpu->cr0 != sregs->cr0;
2207 kvm_arch_ops->set_cr0(vcpu, sregs->cr0);
2209 mmu_reset_needed |= vcpu->cr4 != sregs->cr4;
2210 kvm_arch_ops->set_cr4(vcpu, sregs->cr4);
2211 if (!is_long_mode(vcpu) && is_pae(vcpu))
2212 load_pdptrs(vcpu, vcpu->cr3);
2214 if (mmu_reset_needed)
2215 kvm_mmu_reset_context(vcpu);
2217 if (!irqchip_in_kernel(vcpu->kvm)) {
2218 memcpy(vcpu->irq_pending, sregs->interrupt_bitmap,
2219 sizeof vcpu->irq_pending);
2220 vcpu->irq_summary = 0;
2221 for (i = 0; i < ARRAY_SIZE(vcpu->irq_pending); ++i)
2222 if (vcpu->irq_pending[i])
2223 __set_bit(i, &vcpu->irq_summary);
2226 set_segment(vcpu, &sregs->cs, VCPU_SREG_CS);
2227 set_segment(vcpu, &sregs->ds, VCPU_SREG_DS);
2228 set_segment(vcpu, &sregs->es, VCPU_SREG_ES);
2229 set_segment(vcpu, &sregs->fs, VCPU_SREG_FS);
2230 set_segment(vcpu, &sregs->gs, VCPU_SREG_GS);
2231 set_segment(vcpu, &sregs->ss, VCPU_SREG_SS);
2233 set_segment(vcpu, &sregs->tr, VCPU_SREG_TR);
2234 set_segment(vcpu, &sregs->ldt, VCPU_SREG_LDTR);
2242 * List of msr numbers which we expose to userspace through KVM_GET_MSRS
2243 * and KVM_SET_MSRS, and KVM_GET_MSR_INDEX_LIST.
2245 * This list is modified at module load time to reflect the
2246 * capabilities of the host cpu.
2248 static u32 msrs_to_save[] = {
2249 MSR_IA32_SYSENTER_CS, MSR_IA32_SYSENTER_ESP, MSR_IA32_SYSENTER_EIP,
2251 #ifdef CONFIG_X86_64
2252 MSR_CSTAR, MSR_KERNEL_GS_BASE, MSR_SYSCALL_MASK, MSR_LSTAR,
2254 MSR_IA32_TIME_STAMP_COUNTER,
2257 static unsigned num_msrs_to_save;
2259 static u32 emulated_msrs[] = {
2260 MSR_IA32_MISC_ENABLE,
2263 static __init void kvm_init_msr_list(void)
2268 for (i = j = 0; i < ARRAY_SIZE(msrs_to_save); i++) {
2269 if (rdmsr_safe(msrs_to_save[i], &dummy[0], &dummy[1]) < 0)
2272 msrs_to_save[j] = msrs_to_save[i];
2275 num_msrs_to_save = j;
2279 * Adapt set_msr() to msr_io()'s calling convention
2281 static int do_set_msr(struct kvm_vcpu *vcpu, unsigned index, u64 *data)
2283 return kvm_set_msr(vcpu, index, *data);
2287 * Read or write a bunch of msrs. All parameters are kernel addresses.
2289 * @return number of msrs set successfully.
2291 static int __msr_io(struct kvm_vcpu *vcpu, struct kvm_msrs *msrs,
2292 struct kvm_msr_entry *entries,
2293 int (*do_msr)(struct kvm_vcpu *vcpu,
2294 unsigned index, u64 *data))
2300 for (i = 0; i < msrs->nmsrs; ++i)
2301 if (do_msr(vcpu, entries[i].index, &entries[i].data))
2310 * Read or write a bunch of msrs. Parameters are user addresses.
2312 * @return number of msrs set successfully.
2314 static int msr_io(struct kvm_vcpu *vcpu, struct kvm_msrs __user *user_msrs,
2315 int (*do_msr)(struct kvm_vcpu *vcpu,
2316 unsigned index, u64 *data),
2319 struct kvm_msrs msrs;
2320 struct kvm_msr_entry *entries;
2325 if (copy_from_user(&msrs, user_msrs, sizeof msrs))
2329 if (msrs.nmsrs >= MAX_IO_MSRS)
2333 size = sizeof(struct kvm_msr_entry) * msrs.nmsrs;
2334 entries = vmalloc(size);
2339 if (copy_from_user(entries, user_msrs->entries, size))
2342 r = n = __msr_io(vcpu, &msrs, entries, do_msr);
2347 if (writeback && copy_to_user(user_msrs->entries, entries, size))
2359 * Translate a guest virtual address to a guest physical address.
2361 static int kvm_vcpu_ioctl_translate(struct kvm_vcpu *vcpu,
2362 struct kvm_translation *tr)
2364 unsigned long vaddr = tr->linear_address;
2368 mutex_lock(&vcpu->kvm->lock);
2369 gpa = vcpu->mmu.gva_to_gpa(vcpu, vaddr);
2370 tr->physical_address = gpa;
2371 tr->valid = gpa != UNMAPPED_GVA;
2374 mutex_unlock(&vcpu->kvm->lock);
2380 static int kvm_vcpu_ioctl_interrupt(struct kvm_vcpu *vcpu,
2381 struct kvm_interrupt *irq)
2383 if (irq->irq < 0 || irq->irq >= 256)
2385 if (irqchip_in_kernel(vcpu->kvm))
2389 set_bit(irq->irq, vcpu->irq_pending);
2390 set_bit(irq->irq / BITS_PER_LONG, &vcpu->irq_summary);
2397 static int kvm_vcpu_ioctl_debug_guest(struct kvm_vcpu *vcpu,
2398 struct kvm_debug_guest *dbg)
2404 r = kvm_arch_ops->set_guest_debug(vcpu, dbg);
2411 static struct page *kvm_vcpu_nopage(struct vm_area_struct *vma,
2412 unsigned long address,
2415 struct kvm_vcpu *vcpu = vma->vm_file->private_data;
2416 unsigned long pgoff;
2419 pgoff = ((address - vma->vm_start) >> PAGE_SHIFT) + vma->vm_pgoff;
2421 page = virt_to_page(vcpu->run);
2422 else if (pgoff == KVM_PIO_PAGE_OFFSET)
2423 page = virt_to_page(vcpu->pio_data);
2425 return NOPAGE_SIGBUS;
2428 *type = VM_FAULT_MINOR;
2433 static struct vm_operations_struct kvm_vcpu_vm_ops = {
2434 .nopage = kvm_vcpu_nopage,
2437 static int kvm_vcpu_mmap(struct file *file, struct vm_area_struct *vma)
2439 vma->vm_ops = &kvm_vcpu_vm_ops;
2443 static int kvm_vcpu_release(struct inode *inode, struct file *filp)
2445 struct kvm_vcpu *vcpu = filp->private_data;
2447 fput(vcpu->kvm->filp);
2451 static struct file_operations kvm_vcpu_fops = {
2452 .release = kvm_vcpu_release,
2453 .unlocked_ioctl = kvm_vcpu_ioctl,
2454 .compat_ioctl = kvm_vcpu_ioctl,
2455 .mmap = kvm_vcpu_mmap,
2459 * Allocates an inode for the vcpu.
2461 static int create_vcpu_fd(struct kvm_vcpu *vcpu)
2464 struct inode *inode;
2467 r = anon_inode_getfd(&fd, &inode, &file,
2468 "kvm-vcpu", &kvm_vcpu_fops, vcpu);
2471 atomic_inc(&vcpu->kvm->filp->f_count);
2476 * Creates some virtual cpus. Good luck creating more than one.
2478 static int kvm_vm_ioctl_create_vcpu(struct kvm *kvm, int n)
2481 struct kvm_vcpu *vcpu;
2486 vcpu = kvm_arch_ops->vcpu_create(kvm, n);
2488 return PTR_ERR(vcpu);
2490 preempt_notifier_init(&vcpu->preempt_notifier, &kvm_preempt_ops);
2492 /* We do fxsave: this must be aligned. */
2493 BUG_ON((unsigned long)&vcpu->host_fx_image & 0xF);
2496 r = kvm_mmu_setup(vcpu);
2501 mutex_lock(&kvm->lock);
2502 if (kvm->vcpus[n]) {
2504 mutex_unlock(&kvm->lock);
2507 kvm->vcpus[n] = vcpu;
2508 mutex_unlock(&kvm->lock);
2510 /* Now it's all set up, let userspace reach it */
2511 r = create_vcpu_fd(vcpu);
2517 mutex_lock(&kvm->lock);
2518 kvm->vcpus[n] = NULL;
2519 mutex_unlock(&kvm->lock);
2523 kvm_mmu_unload(vcpu);
2527 kvm_arch_ops->vcpu_free(vcpu);
2531 static void cpuid_fix_nx_cap(struct kvm_vcpu *vcpu)
2535 struct kvm_cpuid_entry *e, *entry;
2537 rdmsrl(MSR_EFER, efer);
2539 for (i = 0; i < vcpu->cpuid_nent; ++i) {
2540 e = &vcpu->cpuid_entries[i];
2541 if (e->function == 0x80000001) {
2546 if (entry && (entry->edx & (1 << 20)) && !(efer & EFER_NX)) {
2547 entry->edx &= ~(1 << 20);
2548 printk(KERN_INFO "kvm: guest NX capability removed\n");
2552 static int kvm_vcpu_ioctl_set_cpuid(struct kvm_vcpu *vcpu,
2553 struct kvm_cpuid *cpuid,
2554 struct kvm_cpuid_entry __user *entries)
2559 if (cpuid->nent > KVM_MAX_CPUID_ENTRIES)
2562 if (copy_from_user(&vcpu->cpuid_entries, entries,
2563 cpuid->nent * sizeof(struct kvm_cpuid_entry)))
2565 vcpu->cpuid_nent = cpuid->nent;
2566 cpuid_fix_nx_cap(vcpu);
2573 static int kvm_vcpu_ioctl_set_sigmask(struct kvm_vcpu *vcpu, sigset_t *sigset)
2576 sigdelsetmask(sigset, sigmask(SIGKILL)|sigmask(SIGSTOP));
2577 vcpu->sigset_active = 1;
2578 vcpu->sigset = *sigset;
2580 vcpu->sigset_active = 0;
2585 * fxsave fpu state. Taken from x86_64/processor.h. To be killed when
2586 * we have asm/x86/processor.h
2597 u32 st_space[32]; /* 8*16 bytes for each FP-reg = 128 bytes */
2598 #ifdef CONFIG_X86_64
2599 u32 xmm_space[64]; /* 16*16 bytes for each XMM-reg = 256 bytes */
2601 u32 xmm_space[32]; /* 8*16 bytes for each XMM-reg = 128 bytes */
2605 static int kvm_vcpu_ioctl_get_fpu(struct kvm_vcpu *vcpu, struct kvm_fpu *fpu)
2607 struct fxsave *fxsave = (struct fxsave *)&vcpu->guest_fx_image;
2611 memcpy(fpu->fpr, fxsave->st_space, 128);
2612 fpu->fcw = fxsave->cwd;
2613 fpu->fsw = fxsave->swd;
2614 fpu->ftwx = fxsave->twd;
2615 fpu->last_opcode = fxsave->fop;
2616 fpu->last_ip = fxsave->rip;
2617 fpu->last_dp = fxsave->rdp;
2618 memcpy(fpu->xmm, fxsave->xmm_space, sizeof fxsave->xmm_space);
2625 static int kvm_vcpu_ioctl_set_fpu(struct kvm_vcpu *vcpu, struct kvm_fpu *fpu)
2627 struct fxsave *fxsave = (struct fxsave *)&vcpu->guest_fx_image;
2631 memcpy(fxsave->st_space, fpu->fpr, 128);
2632 fxsave->cwd = fpu->fcw;
2633 fxsave->swd = fpu->fsw;
2634 fxsave->twd = fpu->ftwx;
2635 fxsave->fop = fpu->last_opcode;
2636 fxsave->rip = fpu->last_ip;
2637 fxsave->rdp = fpu->last_dp;
2638 memcpy(fxsave->xmm_space, fpu->xmm, sizeof fxsave->xmm_space);
2645 static long kvm_vcpu_ioctl(struct file *filp,
2646 unsigned int ioctl, unsigned long arg)
2648 struct kvm_vcpu *vcpu = filp->private_data;
2649 void __user *argp = (void __user *)arg;
2657 r = kvm_vcpu_ioctl_run(vcpu, vcpu->run);
2659 case KVM_GET_REGS: {
2660 struct kvm_regs kvm_regs;
2662 memset(&kvm_regs, 0, sizeof kvm_regs);
2663 r = kvm_vcpu_ioctl_get_regs(vcpu, &kvm_regs);
2667 if (copy_to_user(argp, &kvm_regs, sizeof kvm_regs))
2672 case KVM_SET_REGS: {
2673 struct kvm_regs kvm_regs;
2676 if (copy_from_user(&kvm_regs, argp, sizeof kvm_regs))
2678 r = kvm_vcpu_ioctl_set_regs(vcpu, &kvm_regs);
2684 case KVM_GET_SREGS: {
2685 struct kvm_sregs kvm_sregs;
2687 memset(&kvm_sregs, 0, sizeof kvm_sregs);
2688 r = kvm_vcpu_ioctl_get_sregs(vcpu, &kvm_sregs);
2692 if (copy_to_user(argp, &kvm_sregs, sizeof kvm_sregs))
2697 case KVM_SET_SREGS: {
2698 struct kvm_sregs kvm_sregs;
2701 if (copy_from_user(&kvm_sregs, argp, sizeof kvm_sregs))
2703 r = kvm_vcpu_ioctl_set_sregs(vcpu, &kvm_sregs);
2709 case KVM_TRANSLATE: {
2710 struct kvm_translation tr;
2713 if (copy_from_user(&tr, argp, sizeof tr))
2715 r = kvm_vcpu_ioctl_translate(vcpu, &tr);
2719 if (copy_to_user(argp, &tr, sizeof tr))
2724 case KVM_INTERRUPT: {
2725 struct kvm_interrupt irq;
2728 if (copy_from_user(&irq, argp, sizeof irq))
2730 r = kvm_vcpu_ioctl_interrupt(vcpu, &irq);
2736 case KVM_DEBUG_GUEST: {
2737 struct kvm_debug_guest dbg;
2740 if (copy_from_user(&dbg, argp, sizeof dbg))
2742 r = kvm_vcpu_ioctl_debug_guest(vcpu, &dbg);
2749 r = msr_io(vcpu, argp, kvm_get_msr, 1);
2752 r = msr_io(vcpu, argp, do_set_msr, 0);
2754 case KVM_SET_CPUID: {
2755 struct kvm_cpuid __user *cpuid_arg = argp;
2756 struct kvm_cpuid cpuid;
2759 if (copy_from_user(&cpuid, cpuid_arg, sizeof cpuid))
2761 r = kvm_vcpu_ioctl_set_cpuid(vcpu, &cpuid, cpuid_arg->entries);
2766 case KVM_SET_SIGNAL_MASK: {
2767 struct kvm_signal_mask __user *sigmask_arg = argp;
2768 struct kvm_signal_mask kvm_sigmask;
2769 sigset_t sigset, *p;
2774 if (copy_from_user(&kvm_sigmask, argp,
2775 sizeof kvm_sigmask))
2778 if (kvm_sigmask.len != sizeof sigset)
2781 if (copy_from_user(&sigset, sigmask_arg->sigset,
2786 r = kvm_vcpu_ioctl_set_sigmask(vcpu, &sigset);
2792 memset(&fpu, 0, sizeof fpu);
2793 r = kvm_vcpu_ioctl_get_fpu(vcpu, &fpu);
2797 if (copy_to_user(argp, &fpu, sizeof fpu))
2806 if (copy_from_user(&fpu, argp, sizeof fpu))
2808 r = kvm_vcpu_ioctl_set_fpu(vcpu, &fpu);
2821 static long kvm_vm_ioctl(struct file *filp,
2822 unsigned int ioctl, unsigned long arg)
2824 struct kvm *kvm = filp->private_data;
2825 void __user *argp = (void __user *)arg;
2829 case KVM_CREATE_VCPU:
2830 r = kvm_vm_ioctl_create_vcpu(kvm, arg);
2834 case KVM_SET_MEMORY_REGION: {
2835 struct kvm_memory_region kvm_mem;
2838 if (copy_from_user(&kvm_mem, argp, sizeof kvm_mem))
2840 r = kvm_vm_ioctl_set_memory_region(kvm, &kvm_mem);
2845 case KVM_GET_DIRTY_LOG: {
2846 struct kvm_dirty_log log;
2849 if (copy_from_user(&log, argp, sizeof log))
2851 r = kvm_vm_ioctl_get_dirty_log(kvm, &log);
2856 case KVM_SET_MEMORY_ALIAS: {
2857 struct kvm_memory_alias alias;
2860 if (copy_from_user(&alias, argp, sizeof alias))
2862 r = kvm_vm_ioctl_set_memory_alias(kvm, &alias);
2867 case KVM_CREATE_IRQCHIP:
2869 kvm->vpic = kvm_create_pic(kvm);
2871 r = kvm_ioapic_init(kvm);
2881 case KVM_IRQ_LINE: {
2882 struct kvm_irq_level irq_event;
2885 if (copy_from_user(&irq_event, argp, sizeof irq_event))
2887 if (irqchip_in_kernel(kvm)) {
2888 mutex_lock(&kvm->lock);
2889 if (irq_event.irq < 16)
2890 kvm_pic_set_irq(pic_irqchip(kvm),
2893 kvm_ioapic_set_irq(kvm->vioapic,
2896 mutex_unlock(&kvm->lock);
2901 case KVM_GET_IRQCHIP: {
2902 /* 0: PIC master, 1: PIC slave, 2: IOAPIC */
2903 struct kvm_irqchip chip;
2906 if (copy_from_user(&chip, argp, sizeof chip))
2909 if (!irqchip_in_kernel(kvm))
2911 r = kvm_vm_ioctl_get_irqchip(kvm, &chip);
2915 if (copy_to_user(argp, &chip, sizeof chip))
2920 case KVM_SET_IRQCHIP: {
2921 /* 0: PIC master, 1: PIC slave, 2: IOAPIC */
2922 struct kvm_irqchip chip;
2925 if (copy_from_user(&chip, argp, sizeof chip))
2928 if (!irqchip_in_kernel(kvm))
2930 r = kvm_vm_ioctl_set_irqchip(kvm, &chip);
2943 static struct page *kvm_vm_nopage(struct vm_area_struct *vma,
2944 unsigned long address,
2947 struct kvm *kvm = vma->vm_file->private_data;
2948 unsigned long pgoff;
2951 pgoff = ((address - vma->vm_start) >> PAGE_SHIFT) + vma->vm_pgoff;
2952 page = gfn_to_page(kvm, pgoff);
2954 return NOPAGE_SIGBUS;
2957 *type = VM_FAULT_MINOR;
2962 static struct vm_operations_struct kvm_vm_vm_ops = {
2963 .nopage = kvm_vm_nopage,
2966 static int kvm_vm_mmap(struct file *file, struct vm_area_struct *vma)
2968 vma->vm_ops = &kvm_vm_vm_ops;
2972 static struct file_operations kvm_vm_fops = {
2973 .release = kvm_vm_release,
2974 .unlocked_ioctl = kvm_vm_ioctl,
2975 .compat_ioctl = kvm_vm_ioctl,
2976 .mmap = kvm_vm_mmap,
2979 static int kvm_dev_ioctl_create_vm(void)
2982 struct inode *inode;
2986 kvm = kvm_create_vm();
2988 return PTR_ERR(kvm);
2989 r = anon_inode_getfd(&fd, &inode, &file, "kvm-vm", &kvm_vm_fops, kvm);
2991 kvm_destroy_vm(kvm);
3000 static long kvm_dev_ioctl(struct file *filp,
3001 unsigned int ioctl, unsigned long arg)
3003 void __user *argp = (void __user *)arg;
3007 case KVM_GET_API_VERSION:
3011 r = KVM_API_VERSION;
3017 r = kvm_dev_ioctl_create_vm();
3019 case KVM_GET_MSR_INDEX_LIST: {
3020 struct kvm_msr_list __user *user_msr_list = argp;
3021 struct kvm_msr_list msr_list;
3025 if (copy_from_user(&msr_list, user_msr_list, sizeof msr_list))
3028 msr_list.nmsrs = num_msrs_to_save + ARRAY_SIZE(emulated_msrs);
3029 if (copy_to_user(user_msr_list, &msr_list, sizeof msr_list))
3032 if (n < num_msrs_to_save)
3035 if (copy_to_user(user_msr_list->indices, &msrs_to_save,
3036 num_msrs_to_save * sizeof(u32)))
3038 if (copy_to_user(user_msr_list->indices
3039 + num_msrs_to_save * sizeof(u32),
3041 ARRAY_SIZE(emulated_msrs) * sizeof(u32)))
3046 case KVM_CHECK_EXTENSION: {
3047 int ext = (long)argp;
3050 case KVM_CAP_IRQCHIP:
3060 case KVM_GET_VCPU_MMAP_SIZE:
3073 static struct file_operations kvm_chardev_ops = {
3074 .unlocked_ioctl = kvm_dev_ioctl,
3075 .compat_ioctl = kvm_dev_ioctl,
3078 static struct miscdevice kvm_dev = {
3085 * Make sure that a cpu that is being hot-unplugged does not have any vcpus
3088 static void decache_vcpus_on_cpu(int cpu)
3091 struct kvm_vcpu *vcpu;
3094 spin_lock(&kvm_lock);
3095 list_for_each_entry(vm, &vm_list, vm_list)
3096 for (i = 0; i < KVM_MAX_VCPUS; ++i) {
3097 vcpu = vm->vcpus[i];
3101 * If the vcpu is locked, then it is running on some
3102 * other cpu and therefore it is not cached on the
3105 * If it's not locked, check the last cpu it executed
3108 if (mutex_trylock(&vcpu->mutex)) {
3109 if (vcpu->cpu == cpu) {
3110 kvm_arch_ops->vcpu_decache(vcpu);
3113 mutex_unlock(&vcpu->mutex);
3116 spin_unlock(&kvm_lock);
3119 static void hardware_enable(void *junk)
3121 int cpu = raw_smp_processor_id();
3123 if (cpu_isset(cpu, cpus_hardware_enabled))
3125 cpu_set(cpu, cpus_hardware_enabled);
3126 kvm_arch_ops->hardware_enable(NULL);
3129 static void hardware_disable(void *junk)
3131 int cpu = raw_smp_processor_id();
3133 if (!cpu_isset(cpu, cpus_hardware_enabled))
3135 cpu_clear(cpu, cpus_hardware_enabled);
3136 decache_vcpus_on_cpu(cpu);
3137 kvm_arch_ops->hardware_disable(NULL);
3140 static int kvm_cpu_hotplug(struct notifier_block *notifier, unsigned long val,
3147 case CPU_DYING_FROZEN:
3148 printk(KERN_INFO "kvm: disabling virtualization on CPU%d\n",
3150 hardware_disable(NULL);
3152 case CPU_UP_CANCELED:
3153 case CPU_UP_CANCELED_FROZEN:
3154 printk(KERN_INFO "kvm: disabling virtualization on CPU%d\n",
3156 smp_call_function_single(cpu, hardware_disable, NULL, 0, 1);
3159 case CPU_ONLINE_FROZEN:
3160 printk(KERN_INFO "kvm: enabling virtualization on CPU%d\n",
3162 smp_call_function_single(cpu, hardware_enable, NULL, 0, 1);
3168 static int kvm_reboot(struct notifier_block *notifier, unsigned long val,
3171 if (val == SYS_RESTART) {
3173 * Some (well, at least mine) BIOSes hang on reboot if
3176 printk(KERN_INFO "kvm: exiting hardware virtualization\n");
3177 on_each_cpu(hardware_disable, NULL, 0, 1);
3182 static struct notifier_block kvm_reboot_notifier = {
3183 .notifier_call = kvm_reboot,
3187 void kvm_io_bus_init(struct kvm_io_bus *bus)
3189 memset(bus, 0, sizeof(*bus));
3192 void kvm_io_bus_destroy(struct kvm_io_bus *bus)
3196 for (i = 0; i < bus->dev_count; i++) {
3197 struct kvm_io_device *pos = bus->devs[i];
3199 kvm_iodevice_destructor(pos);
3203 struct kvm_io_device *kvm_io_bus_find_dev(struct kvm_io_bus *bus, gpa_t addr)
3207 for (i = 0; i < bus->dev_count; i++) {
3208 struct kvm_io_device *pos = bus->devs[i];
3210 if (pos->in_range(pos, addr))
3217 void kvm_io_bus_register_dev(struct kvm_io_bus *bus, struct kvm_io_device *dev)
3219 BUG_ON(bus->dev_count > (NR_IOBUS_DEVS-1));
3221 bus->devs[bus->dev_count++] = dev;
3224 static struct notifier_block kvm_cpu_notifier = {
3225 .notifier_call = kvm_cpu_hotplug,
3226 .priority = 20, /* must be > scheduler priority */
3229 static u64 stat_get(void *_offset)
3231 unsigned offset = (long)_offset;
3234 struct kvm_vcpu *vcpu;
3237 spin_lock(&kvm_lock);
3238 list_for_each_entry(kvm, &vm_list, vm_list)
3239 for (i = 0; i < KVM_MAX_VCPUS; ++i) {
3240 vcpu = kvm->vcpus[i];
3242 total += *(u32 *)((void *)vcpu + offset);
3244 spin_unlock(&kvm_lock);
3248 DEFINE_SIMPLE_ATTRIBUTE(stat_fops, stat_get, NULL, "%llu\n");
3250 static __init void kvm_init_debug(void)
3252 struct kvm_stats_debugfs_item *p;
3254 debugfs_dir = debugfs_create_dir("kvm", NULL);
3255 for (p = debugfs_entries; p->name; ++p)
3256 p->dentry = debugfs_create_file(p->name, 0444, debugfs_dir,
3257 (void *)(long)p->offset,
3261 static void kvm_exit_debug(void)
3263 struct kvm_stats_debugfs_item *p;
3265 for (p = debugfs_entries; p->name; ++p)
3266 debugfs_remove(p->dentry);
3267 debugfs_remove(debugfs_dir);
3270 static int kvm_suspend(struct sys_device *dev, pm_message_t state)
3272 hardware_disable(NULL);
3276 static int kvm_resume(struct sys_device *dev)
3278 hardware_enable(NULL);
3282 static struct sysdev_class kvm_sysdev_class = {
3283 set_kset_name("kvm"),
3284 .suspend = kvm_suspend,
3285 .resume = kvm_resume,
3288 static struct sys_device kvm_sysdev = {
3290 .cls = &kvm_sysdev_class,
3293 hpa_t bad_page_address;
3296 struct kvm_vcpu *preempt_notifier_to_vcpu(struct preempt_notifier *pn)
3298 return container_of(pn, struct kvm_vcpu, preempt_notifier);
3301 static void kvm_sched_in(struct preempt_notifier *pn, int cpu)
3303 struct kvm_vcpu *vcpu = preempt_notifier_to_vcpu(pn);
3305 kvm_arch_ops->vcpu_load(vcpu, cpu);
3308 static void kvm_sched_out(struct preempt_notifier *pn,
3309 struct task_struct *next)
3311 struct kvm_vcpu *vcpu = preempt_notifier_to_vcpu(pn);
3313 kvm_arch_ops->vcpu_put(vcpu);
3316 int kvm_init_arch(struct kvm_arch_ops *ops, unsigned int vcpu_size,
3317 struct module *module)
3323 printk(KERN_ERR "kvm: already loaded the other module\n");
3327 if (!ops->cpu_has_kvm_support()) {
3328 printk(KERN_ERR "kvm: no hardware support\n");
3331 if (ops->disabled_by_bios()) {
3332 printk(KERN_ERR "kvm: disabled by bios\n");
3338 r = kvm_arch_ops->hardware_setup();
3342 for_each_online_cpu(cpu) {
3343 smp_call_function_single(cpu,
3344 kvm_arch_ops->check_processor_compatibility,
3350 on_each_cpu(hardware_enable, NULL, 0, 1);
3351 r = register_cpu_notifier(&kvm_cpu_notifier);
3354 register_reboot_notifier(&kvm_reboot_notifier);
3356 r = sysdev_class_register(&kvm_sysdev_class);
3360 r = sysdev_register(&kvm_sysdev);
3364 /* A kmem cache lets us meet the alignment requirements of fx_save. */
3365 kvm_vcpu_cache = kmem_cache_create("kvm_vcpu", vcpu_size,
3366 __alignof__(struct kvm_vcpu), 0, 0);
3367 if (!kvm_vcpu_cache) {
3372 kvm_chardev_ops.owner = module;
3374 r = misc_register(&kvm_dev);
3376 printk (KERN_ERR "kvm: misc device register failed\n");
3380 kvm_preempt_ops.sched_in = kvm_sched_in;
3381 kvm_preempt_ops.sched_out = kvm_sched_out;
3386 kmem_cache_destroy(kvm_vcpu_cache);
3388 sysdev_unregister(&kvm_sysdev);
3390 sysdev_class_unregister(&kvm_sysdev_class);
3392 unregister_reboot_notifier(&kvm_reboot_notifier);
3393 unregister_cpu_notifier(&kvm_cpu_notifier);
3395 on_each_cpu(hardware_disable, NULL, 0, 1);
3397 kvm_arch_ops->hardware_unsetup();
3399 kvm_arch_ops = NULL;
3403 void kvm_exit_arch(void)
3405 misc_deregister(&kvm_dev);
3406 kmem_cache_destroy(kvm_vcpu_cache);
3407 sysdev_unregister(&kvm_sysdev);
3408 sysdev_class_unregister(&kvm_sysdev_class);
3409 unregister_reboot_notifier(&kvm_reboot_notifier);
3410 unregister_cpu_notifier(&kvm_cpu_notifier);
3411 on_each_cpu(hardware_disable, NULL, 0, 1);
3412 kvm_arch_ops->hardware_unsetup();
3413 kvm_arch_ops = NULL;
3416 static __init int kvm_init(void)
3418 static struct page *bad_page;
3421 r = kvm_mmu_module_init();
3427 kvm_init_msr_list();
3429 if ((bad_page = alloc_page(GFP_KERNEL)) == NULL) {
3434 bad_page_address = page_to_pfn(bad_page) << PAGE_SHIFT;
3435 memset(__va(bad_page_address), 0, PAGE_SIZE);
3441 kvm_mmu_module_exit();
3446 static __exit void kvm_exit(void)
3449 __free_page(pfn_to_page(bad_page_address >> PAGE_SHIFT));
3450 kvm_mmu_module_exit();
3453 module_init(kvm_init)
3454 module_exit(kvm_exit)
3456 EXPORT_SYMBOL_GPL(kvm_init_arch);
3457 EXPORT_SYMBOL_GPL(kvm_exit_arch);