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));
923 static int kvm_vm_ioctl_set_irqchip(struct kvm *kvm, struct kvm_irqchip *chip)
928 switch (chip->chip_id) {
929 case KVM_IRQCHIP_PIC_MASTER:
930 memcpy (&pic_irqchip(kvm)->pics[0],
932 sizeof(struct kvm_pic_state));
934 case KVM_IRQCHIP_PIC_SLAVE:
935 memcpy (&pic_irqchip(kvm)->pics[1],
937 sizeof(struct kvm_pic_state));
943 kvm_pic_update_irq(pic_irqchip(kvm));
947 static gfn_t unalias_gfn(struct kvm *kvm, gfn_t gfn)
950 struct kvm_mem_alias *alias;
952 for (i = 0; i < kvm->naliases; ++i) {
953 alias = &kvm->aliases[i];
954 if (gfn >= alias->base_gfn
955 && gfn < alias->base_gfn + alias->npages)
956 return alias->target_gfn + gfn - alias->base_gfn;
961 static struct kvm_memory_slot *__gfn_to_memslot(struct kvm *kvm, gfn_t gfn)
965 for (i = 0; i < kvm->nmemslots; ++i) {
966 struct kvm_memory_slot *memslot = &kvm->memslots[i];
968 if (gfn >= memslot->base_gfn
969 && gfn < memslot->base_gfn + memslot->npages)
975 struct kvm_memory_slot *gfn_to_memslot(struct kvm *kvm, gfn_t gfn)
977 gfn = unalias_gfn(kvm, gfn);
978 return __gfn_to_memslot(kvm, gfn);
981 struct page *gfn_to_page(struct kvm *kvm, gfn_t gfn)
983 struct kvm_memory_slot *slot;
985 gfn = unalias_gfn(kvm, gfn);
986 slot = __gfn_to_memslot(kvm, gfn);
989 return slot->phys_mem[gfn - slot->base_gfn];
991 EXPORT_SYMBOL_GPL(gfn_to_page);
993 /* WARNING: Does not work on aliased pages. */
994 void mark_page_dirty(struct kvm *kvm, gfn_t gfn)
996 struct kvm_memory_slot *memslot;
998 memslot = __gfn_to_memslot(kvm, gfn);
999 if (memslot && memslot->dirty_bitmap) {
1000 unsigned long rel_gfn = gfn - memslot->base_gfn;
1003 if (!test_bit(rel_gfn, memslot->dirty_bitmap))
1004 set_bit(rel_gfn, memslot->dirty_bitmap);
1008 int emulator_read_std(unsigned long addr,
1011 struct kvm_vcpu *vcpu)
1016 gpa_t gpa = vcpu->mmu.gva_to_gpa(vcpu, addr);
1017 unsigned offset = addr & (PAGE_SIZE-1);
1018 unsigned tocopy = min(bytes, (unsigned)PAGE_SIZE - offset);
1023 if (gpa == UNMAPPED_GVA)
1024 return X86EMUL_PROPAGATE_FAULT;
1025 pfn = gpa >> PAGE_SHIFT;
1026 page = gfn_to_page(vcpu->kvm, pfn);
1028 return X86EMUL_UNHANDLEABLE;
1029 page_virt = kmap_atomic(page, KM_USER0);
1031 memcpy(data, page_virt + offset, tocopy);
1033 kunmap_atomic(page_virt, KM_USER0);
1040 return X86EMUL_CONTINUE;
1042 EXPORT_SYMBOL_GPL(emulator_read_std);
1044 static int emulator_write_std(unsigned long addr,
1047 struct kvm_vcpu *vcpu)
1049 pr_unimpl(vcpu, "emulator_write_std: addr %lx n %d\n", addr, bytes);
1050 return X86EMUL_UNHANDLEABLE;
1054 * Only apic need an MMIO device hook, so shortcut now..
1056 static struct kvm_io_device *vcpu_find_pervcpu_dev(struct kvm_vcpu *vcpu,
1059 struct kvm_io_device *dev;
1062 dev = &vcpu->apic->dev;
1063 if (dev->in_range(dev, addr))
1069 static struct kvm_io_device *vcpu_find_mmio_dev(struct kvm_vcpu *vcpu,
1072 struct kvm_io_device *dev;
1074 dev = vcpu_find_pervcpu_dev(vcpu, addr);
1076 dev = kvm_io_bus_find_dev(&vcpu->kvm->mmio_bus, addr);
1080 static struct kvm_io_device *vcpu_find_pio_dev(struct kvm_vcpu *vcpu,
1083 return kvm_io_bus_find_dev(&vcpu->kvm->pio_bus, addr);
1086 static int emulator_read_emulated(unsigned long addr,
1089 struct kvm_vcpu *vcpu)
1091 struct kvm_io_device *mmio_dev;
1094 if (vcpu->mmio_read_completed) {
1095 memcpy(val, vcpu->mmio_data, bytes);
1096 vcpu->mmio_read_completed = 0;
1097 return X86EMUL_CONTINUE;
1098 } else if (emulator_read_std(addr, val, bytes, vcpu)
1099 == X86EMUL_CONTINUE)
1100 return X86EMUL_CONTINUE;
1102 gpa = vcpu->mmu.gva_to_gpa(vcpu, addr);
1103 if (gpa == UNMAPPED_GVA)
1104 return X86EMUL_PROPAGATE_FAULT;
1107 * Is this MMIO handled locally?
1109 mmio_dev = vcpu_find_mmio_dev(vcpu, gpa);
1111 kvm_iodevice_read(mmio_dev, gpa, bytes, val);
1112 return X86EMUL_CONTINUE;
1115 vcpu->mmio_needed = 1;
1116 vcpu->mmio_phys_addr = gpa;
1117 vcpu->mmio_size = bytes;
1118 vcpu->mmio_is_write = 0;
1120 return X86EMUL_UNHANDLEABLE;
1123 static int emulator_write_phys(struct kvm_vcpu *vcpu, gpa_t gpa,
1124 const void *val, int bytes)
1129 if (((gpa + bytes - 1) >> PAGE_SHIFT) != (gpa >> PAGE_SHIFT))
1131 page = gfn_to_page(vcpu->kvm, gpa >> PAGE_SHIFT);
1134 mark_page_dirty(vcpu->kvm, gpa >> PAGE_SHIFT);
1135 virt = kmap_atomic(page, KM_USER0);
1136 kvm_mmu_pte_write(vcpu, gpa, val, bytes);
1137 memcpy(virt + offset_in_page(gpa), val, bytes);
1138 kunmap_atomic(virt, KM_USER0);
1142 static int emulator_write_emulated_onepage(unsigned long addr,
1145 struct kvm_vcpu *vcpu)
1147 struct kvm_io_device *mmio_dev;
1148 gpa_t gpa = vcpu->mmu.gva_to_gpa(vcpu, addr);
1150 if (gpa == UNMAPPED_GVA) {
1151 kvm_arch_ops->inject_page_fault(vcpu, addr, 2);
1152 return X86EMUL_PROPAGATE_FAULT;
1155 if (emulator_write_phys(vcpu, gpa, val, bytes))
1156 return X86EMUL_CONTINUE;
1159 * Is this MMIO handled locally?
1161 mmio_dev = vcpu_find_mmio_dev(vcpu, gpa);
1163 kvm_iodevice_write(mmio_dev, gpa, bytes, val);
1164 return X86EMUL_CONTINUE;
1167 vcpu->mmio_needed = 1;
1168 vcpu->mmio_phys_addr = gpa;
1169 vcpu->mmio_size = bytes;
1170 vcpu->mmio_is_write = 1;
1171 memcpy(vcpu->mmio_data, val, bytes);
1173 return X86EMUL_CONTINUE;
1176 int emulator_write_emulated(unsigned long addr,
1179 struct kvm_vcpu *vcpu)
1181 /* Crossing a page boundary? */
1182 if (((addr + bytes - 1) ^ addr) & PAGE_MASK) {
1185 now = -addr & ~PAGE_MASK;
1186 rc = emulator_write_emulated_onepage(addr, val, now, vcpu);
1187 if (rc != X86EMUL_CONTINUE)
1193 return emulator_write_emulated_onepage(addr, val, bytes, vcpu);
1195 EXPORT_SYMBOL_GPL(emulator_write_emulated);
1197 static int emulator_cmpxchg_emulated(unsigned long addr,
1201 struct kvm_vcpu *vcpu)
1203 static int reported;
1207 printk(KERN_WARNING "kvm: emulating exchange as write\n");
1209 return emulator_write_emulated(addr, new, bytes, vcpu);
1212 static unsigned long get_segment_base(struct kvm_vcpu *vcpu, int seg)
1214 return kvm_arch_ops->get_segment_base(vcpu, seg);
1217 int emulate_invlpg(struct kvm_vcpu *vcpu, gva_t address)
1219 return X86EMUL_CONTINUE;
1222 int emulate_clts(struct kvm_vcpu *vcpu)
1226 cr0 = vcpu->cr0 & ~X86_CR0_TS;
1227 kvm_arch_ops->set_cr0(vcpu, cr0);
1228 return X86EMUL_CONTINUE;
1231 int emulator_get_dr(struct x86_emulate_ctxt* ctxt, int dr, unsigned long *dest)
1233 struct kvm_vcpu *vcpu = ctxt->vcpu;
1237 *dest = kvm_arch_ops->get_dr(vcpu, dr);
1238 return X86EMUL_CONTINUE;
1240 pr_unimpl(vcpu, "%s: unexpected dr %u\n", __FUNCTION__, dr);
1241 return X86EMUL_UNHANDLEABLE;
1245 int emulator_set_dr(struct x86_emulate_ctxt *ctxt, int dr, unsigned long value)
1247 unsigned long mask = (ctxt->mode == X86EMUL_MODE_PROT64) ? ~0ULL : ~0U;
1250 kvm_arch_ops->set_dr(ctxt->vcpu, dr, value & mask, &exception);
1252 /* FIXME: better handling */
1253 return X86EMUL_UNHANDLEABLE;
1255 return X86EMUL_CONTINUE;
1258 static void report_emulation_failure(struct x86_emulate_ctxt *ctxt)
1260 static int reported;
1262 unsigned long rip = ctxt->vcpu->rip;
1263 unsigned long rip_linear;
1265 rip_linear = rip + get_segment_base(ctxt->vcpu, VCPU_SREG_CS);
1270 emulator_read_std(rip_linear, (void *)opcodes, 4, ctxt->vcpu);
1272 printk(KERN_ERR "emulation failed but !mmio_needed?"
1273 " rip %lx %02x %02x %02x %02x\n",
1274 rip, opcodes[0], opcodes[1], opcodes[2], opcodes[3]);
1278 struct x86_emulate_ops emulate_ops = {
1279 .read_std = emulator_read_std,
1280 .write_std = emulator_write_std,
1281 .read_emulated = emulator_read_emulated,
1282 .write_emulated = emulator_write_emulated,
1283 .cmpxchg_emulated = emulator_cmpxchg_emulated,
1286 int emulate_instruction(struct kvm_vcpu *vcpu,
1287 struct kvm_run *run,
1291 struct x86_emulate_ctxt emulate_ctxt;
1295 vcpu->mmio_fault_cr2 = cr2;
1296 kvm_arch_ops->cache_regs(vcpu);
1298 kvm_arch_ops->get_cs_db_l_bits(vcpu, &cs_db, &cs_l);
1300 emulate_ctxt.vcpu = vcpu;
1301 emulate_ctxt.eflags = kvm_arch_ops->get_rflags(vcpu);
1302 emulate_ctxt.cr2 = cr2;
1303 emulate_ctxt.mode = (emulate_ctxt.eflags & X86_EFLAGS_VM)
1304 ? X86EMUL_MODE_REAL : cs_l
1305 ? X86EMUL_MODE_PROT64 : cs_db
1306 ? X86EMUL_MODE_PROT32 : X86EMUL_MODE_PROT16;
1308 if (emulate_ctxt.mode == X86EMUL_MODE_PROT64) {
1309 emulate_ctxt.cs_base = 0;
1310 emulate_ctxt.ds_base = 0;
1311 emulate_ctxt.es_base = 0;
1312 emulate_ctxt.ss_base = 0;
1314 emulate_ctxt.cs_base = get_segment_base(vcpu, VCPU_SREG_CS);
1315 emulate_ctxt.ds_base = get_segment_base(vcpu, VCPU_SREG_DS);
1316 emulate_ctxt.es_base = get_segment_base(vcpu, VCPU_SREG_ES);
1317 emulate_ctxt.ss_base = get_segment_base(vcpu, VCPU_SREG_SS);
1320 emulate_ctxt.gs_base = get_segment_base(vcpu, VCPU_SREG_GS);
1321 emulate_ctxt.fs_base = get_segment_base(vcpu, VCPU_SREG_FS);
1323 vcpu->mmio_is_write = 0;
1324 vcpu->pio.string = 0;
1325 r = x86_emulate_memop(&emulate_ctxt, &emulate_ops);
1326 if (vcpu->pio.string)
1327 return EMULATE_DO_MMIO;
1329 if ((r || vcpu->mmio_is_write) && run) {
1330 run->exit_reason = KVM_EXIT_MMIO;
1331 run->mmio.phys_addr = vcpu->mmio_phys_addr;
1332 memcpy(run->mmio.data, vcpu->mmio_data, 8);
1333 run->mmio.len = vcpu->mmio_size;
1334 run->mmio.is_write = vcpu->mmio_is_write;
1338 if (kvm_mmu_unprotect_page_virt(vcpu, cr2))
1339 return EMULATE_DONE;
1340 if (!vcpu->mmio_needed) {
1341 report_emulation_failure(&emulate_ctxt);
1342 return EMULATE_FAIL;
1344 return EMULATE_DO_MMIO;
1347 kvm_arch_ops->decache_regs(vcpu);
1348 kvm_arch_ops->set_rflags(vcpu, emulate_ctxt.eflags);
1350 if (vcpu->mmio_is_write) {
1351 vcpu->mmio_needed = 0;
1352 return EMULATE_DO_MMIO;
1355 return EMULATE_DONE;
1357 EXPORT_SYMBOL_GPL(emulate_instruction);
1360 * The vCPU has executed a HLT instruction with in-kernel mode enabled.
1362 static void kvm_vcpu_kernel_halt(struct kvm_vcpu *vcpu)
1364 DECLARE_WAITQUEUE(wait, current);
1366 add_wait_queue(&vcpu->wq, &wait);
1369 * We will block until either an interrupt or a signal wakes us up
1371 while(!(irqchip_in_kernel(vcpu->kvm) && kvm_cpu_has_interrupt(vcpu))
1372 && !vcpu->irq_summary
1373 && !signal_pending(current)) {
1374 set_current_state(TASK_INTERRUPTIBLE);
1380 remove_wait_queue(&vcpu->wq, &wait);
1381 set_current_state(TASK_RUNNING);
1384 int kvm_emulate_halt(struct kvm_vcpu *vcpu)
1386 ++vcpu->stat.halt_exits;
1387 if (irqchip_in_kernel(vcpu->kvm)) {
1388 kvm_vcpu_kernel_halt(vcpu);
1391 vcpu->run->exit_reason = KVM_EXIT_HLT;
1395 EXPORT_SYMBOL_GPL(kvm_emulate_halt);
1397 int kvm_hypercall(struct kvm_vcpu *vcpu, struct kvm_run *run)
1399 unsigned long nr, a0, a1, a2, a3, a4, a5, ret;
1401 kvm_arch_ops->cache_regs(vcpu);
1403 #ifdef CONFIG_X86_64
1404 if (is_long_mode(vcpu)) {
1405 nr = vcpu->regs[VCPU_REGS_RAX];
1406 a0 = vcpu->regs[VCPU_REGS_RDI];
1407 a1 = vcpu->regs[VCPU_REGS_RSI];
1408 a2 = vcpu->regs[VCPU_REGS_RDX];
1409 a3 = vcpu->regs[VCPU_REGS_RCX];
1410 a4 = vcpu->regs[VCPU_REGS_R8];
1411 a5 = vcpu->regs[VCPU_REGS_R9];
1415 nr = vcpu->regs[VCPU_REGS_RBX] & -1u;
1416 a0 = vcpu->regs[VCPU_REGS_RAX] & -1u;
1417 a1 = vcpu->regs[VCPU_REGS_RCX] & -1u;
1418 a2 = vcpu->regs[VCPU_REGS_RDX] & -1u;
1419 a3 = vcpu->regs[VCPU_REGS_RSI] & -1u;
1420 a4 = vcpu->regs[VCPU_REGS_RDI] & -1u;
1421 a5 = vcpu->regs[VCPU_REGS_RBP] & -1u;
1425 run->hypercall.nr = nr;
1426 run->hypercall.args[0] = a0;
1427 run->hypercall.args[1] = a1;
1428 run->hypercall.args[2] = a2;
1429 run->hypercall.args[3] = a3;
1430 run->hypercall.args[4] = a4;
1431 run->hypercall.args[5] = a5;
1432 run->hypercall.ret = ret;
1433 run->hypercall.longmode = is_long_mode(vcpu);
1434 kvm_arch_ops->decache_regs(vcpu);
1437 vcpu->regs[VCPU_REGS_RAX] = ret;
1438 kvm_arch_ops->decache_regs(vcpu);
1441 EXPORT_SYMBOL_GPL(kvm_hypercall);
1443 static u64 mk_cr_64(u64 curr_cr, u32 new_val)
1445 return (curr_cr & ~((1ULL << 32) - 1)) | new_val;
1448 void realmode_lgdt(struct kvm_vcpu *vcpu, u16 limit, unsigned long base)
1450 struct descriptor_table dt = { limit, base };
1452 kvm_arch_ops->set_gdt(vcpu, &dt);
1455 void realmode_lidt(struct kvm_vcpu *vcpu, u16 limit, unsigned long base)
1457 struct descriptor_table dt = { limit, base };
1459 kvm_arch_ops->set_idt(vcpu, &dt);
1462 void realmode_lmsw(struct kvm_vcpu *vcpu, unsigned long msw,
1463 unsigned long *rflags)
1466 *rflags = kvm_arch_ops->get_rflags(vcpu);
1469 unsigned long realmode_get_cr(struct kvm_vcpu *vcpu, int cr)
1471 kvm_arch_ops->decache_cr4_guest_bits(vcpu);
1482 vcpu_printf(vcpu, "%s: unexpected cr %u\n", __FUNCTION__, cr);
1487 void realmode_set_cr(struct kvm_vcpu *vcpu, int cr, unsigned long val,
1488 unsigned long *rflags)
1492 set_cr0(vcpu, mk_cr_64(vcpu->cr0, val));
1493 *rflags = kvm_arch_ops->get_rflags(vcpu);
1502 set_cr4(vcpu, mk_cr_64(vcpu->cr4, val));
1505 vcpu_printf(vcpu, "%s: unexpected cr %u\n", __FUNCTION__, cr);
1510 * Register the para guest with the host:
1512 static int vcpu_register_para(struct kvm_vcpu *vcpu, gpa_t para_state_gpa)
1514 struct kvm_vcpu_para_state *para_state;
1515 hpa_t para_state_hpa, hypercall_hpa;
1516 struct page *para_state_page;
1517 unsigned char *hypercall;
1518 gpa_t hypercall_gpa;
1520 printk(KERN_DEBUG "kvm: guest trying to enter paravirtual mode\n");
1521 printk(KERN_DEBUG ".... para_state_gpa: %08Lx\n", para_state_gpa);
1524 * Needs to be page aligned:
1526 if (para_state_gpa != PAGE_ALIGN(para_state_gpa))
1529 para_state_hpa = gpa_to_hpa(vcpu, para_state_gpa);
1530 printk(KERN_DEBUG ".... para_state_hpa: %08Lx\n", para_state_hpa);
1531 if (is_error_hpa(para_state_hpa))
1534 mark_page_dirty(vcpu->kvm, para_state_gpa >> PAGE_SHIFT);
1535 para_state_page = pfn_to_page(para_state_hpa >> PAGE_SHIFT);
1536 para_state = kmap(para_state_page);
1538 printk(KERN_DEBUG ".... guest version: %d\n", para_state->guest_version);
1539 printk(KERN_DEBUG ".... size: %d\n", para_state->size);
1541 para_state->host_version = KVM_PARA_API_VERSION;
1543 * We cannot support guests that try to register themselves
1544 * with a newer API version than the host supports:
1546 if (para_state->guest_version > KVM_PARA_API_VERSION) {
1547 para_state->ret = -KVM_EINVAL;
1548 goto err_kunmap_skip;
1551 hypercall_gpa = para_state->hypercall_gpa;
1552 hypercall_hpa = gpa_to_hpa(vcpu, hypercall_gpa);
1553 printk(KERN_DEBUG ".... hypercall_hpa: %08Lx\n", hypercall_hpa);
1554 if (is_error_hpa(hypercall_hpa)) {
1555 para_state->ret = -KVM_EINVAL;
1556 goto err_kunmap_skip;
1559 printk(KERN_DEBUG "kvm: para guest successfully registered.\n");
1560 vcpu->para_state_page = para_state_page;
1561 vcpu->para_state_gpa = para_state_gpa;
1562 vcpu->hypercall_gpa = hypercall_gpa;
1564 mark_page_dirty(vcpu->kvm, hypercall_gpa >> PAGE_SHIFT);
1565 hypercall = kmap_atomic(pfn_to_page(hypercall_hpa >> PAGE_SHIFT),
1566 KM_USER1) + (hypercall_hpa & ~PAGE_MASK);
1567 kvm_arch_ops->patch_hypercall(vcpu, hypercall);
1568 kunmap_atomic(hypercall, KM_USER1);
1570 para_state->ret = 0;
1572 kunmap(para_state_page);
1578 int kvm_get_msr_common(struct kvm_vcpu *vcpu, u32 msr, u64 *pdata)
1583 case 0xc0010010: /* SYSCFG */
1584 case 0xc0010015: /* HWCR */
1585 case MSR_IA32_PLATFORM_ID:
1586 case MSR_IA32_P5_MC_ADDR:
1587 case MSR_IA32_P5_MC_TYPE:
1588 case MSR_IA32_MC0_CTL:
1589 case MSR_IA32_MCG_STATUS:
1590 case MSR_IA32_MCG_CAP:
1591 case MSR_IA32_MC0_MISC:
1592 case MSR_IA32_MC0_MISC+4:
1593 case MSR_IA32_MC0_MISC+8:
1594 case MSR_IA32_MC0_MISC+12:
1595 case MSR_IA32_MC0_MISC+16:
1596 case MSR_IA32_UCODE_REV:
1597 case MSR_IA32_PERF_STATUS:
1598 case MSR_IA32_EBL_CR_POWERON:
1599 /* MTRR registers */
1601 case 0x200 ... 0x2ff:
1604 case 0xcd: /* fsb frequency */
1607 case MSR_IA32_APICBASE:
1608 data = kvm_get_apic_base(vcpu);
1610 case MSR_IA32_MISC_ENABLE:
1611 data = vcpu->ia32_misc_enable_msr;
1613 #ifdef CONFIG_X86_64
1615 data = vcpu->shadow_efer;
1619 pr_unimpl(vcpu, "unhandled rdmsr: 0x%x\n", msr);
1625 EXPORT_SYMBOL_GPL(kvm_get_msr_common);
1628 * Reads an msr value (of 'msr_index') into 'pdata'.
1629 * Returns 0 on success, non-0 otherwise.
1630 * Assumes vcpu_load() was already called.
1632 int kvm_get_msr(struct kvm_vcpu *vcpu, u32 msr_index, u64 *pdata)
1634 return kvm_arch_ops->get_msr(vcpu, msr_index, pdata);
1637 #ifdef CONFIG_X86_64
1639 static void set_efer(struct kvm_vcpu *vcpu, u64 efer)
1641 if (efer & EFER_RESERVED_BITS) {
1642 printk(KERN_DEBUG "set_efer: 0x%llx #GP, reserved bits\n",
1649 && (vcpu->shadow_efer & EFER_LME) != (efer & EFER_LME)) {
1650 printk(KERN_DEBUG "set_efer: #GP, change LME while paging\n");
1655 kvm_arch_ops->set_efer(vcpu, efer);
1658 efer |= vcpu->shadow_efer & EFER_LMA;
1660 vcpu->shadow_efer = efer;
1665 int kvm_set_msr_common(struct kvm_vcpu *vcpu, u32 msr, u64 data)
1668 #ifdef CONFIG_X86_64
1670 set_efer(vcpu, data);
1673 case MSR_IA32_MC0_STATUS:
1674 pr_unimpl(vcpu, "%s: MSR_IA32_MC0_STATUS 0x%llx, nop\n",
1675 __FUNCTION__, data);
1677 case MSR_IA32_MCG_STATUS:
1678 pr_unimpl(vcpu, "%s: MSR_IA32_MCG_STATUS 0x%llx, nop\n",
1679 __FUNCTION__, data);
1681 case MSR_IA32_UCODE_REV:
1682 case MSR_IA32_UCODE_WRITE:
1683 case 0x200 ... 0x2ff: /* MTRRs */
1685 case MSR_IA32_APICBASE:
1686 kvm_set_apic_base(vcpu, data);
1688 case MSR_IA32_MISC_ENABLE:
1689 vcpu->ia32_misc_enable_msr = data;
1692 * This is the 'probe whether the host is KVM' logic:
1694 case MSR_KVM_API_MAGIC:
1695 return vcpu_register_para(vcpu, data);
1698 pr_unimpl(vcpu, "unhandled wrmsr: 0x%x\n", msr);
1703 EXPORT_SYMBOL_GPL(kvm_set_msr_common);
1706 * Writes msr value into into the appropriate "register".
1707 * Returns 0 on success, non-0 otherwise.
1708 * Assumes vcpu_load() was already called.
1710 int kvm_set_msr(struct kvm_vcpu *vcpu, u32 msr_index, u64 data)
1712 return kvm_arch_ops->set_msr(vcpu, msr_index, data);
1715 void kvm_resched(struct kvm_vcpu *vcpu)
1717 if (!need_resched())
1721 EXPORT_SYMBOL_GPL(kvm_resched);
1723 void kvm_emulate_cpuid(struct kvm_vcpu *vcpu)
1727 struct kvm_cpuid_entry *e, *best;
1729 kvm_arch_ops->cache_regs(vcpu);
1730 function = vcpu->regs[VCPU_REGS_RAX];
1731 vcpu->regs[VCPU_REGS_RAX] = 0;
1732 vcpu->regs[VCPU_REGS_RBX] = 0;
1733 vcpu->regs[VCPU_REGS_RCX] = 0;
1734 vcpu->regs[VCPU_REGS_RDX] = 0;
1736 for (i = 0; i < vcpu->cpuid_nent; ++i) {
1737 e = &vcpu->cpuid_entries[i];
1738 if (e->function == function) {
1743 * Both basic or both extended?
1745 if (((e->function ^ function) & 0x80000000) == 0)
1746 if (!best || e->function > best->function)
1750 vcpu->regs[VCPU_REGS_RAX] = best->eax;
1751 vcpu->regs[VCPU_REGS_RBX] = best->ebx;
1752 vcpu->regs[VCPU_REGS_RCX] = best->ecx;
1753 vcpu->regs[VCPU_REGS_RDX] = best->edx;
1755 kvm_arch_ops->decache_regs(vcpu);
1756 kvm_arch_ops->skip_emulated_instruction(vcpu);
1758 EXPORT_SYMBOL_GPL(kvm_emulate_cpuid);
1760 static int pio_copy_data(struct kvm_vcpu *vcpu)
1762 void *p = vcpu->pio_data;
1765 int nr_pages = vcpu->pio.guest_pages[1] ? 2 : 1;
1767 q = vmap(vcpu->pio.guest_pages, nr_pages, VM_READ|VM_WRITE,
1770 free_pio_guest_pages(vcpu);
1773 q += vcpu->pio.guest_page_offset;
1774 bytes = vcpu->pio.size * vcpu->pio.cur_count;
1776 memcpy(q, p, bytes);
1778 memcpy(p, q, bytes);
1779 q -= vcpu->pio.guest_page_offset;
1781 free_pio_guest_pages(vcpu);
1785 static int complete_pio(struct kvm_vcpu *vcpu)
1787 struct kvm_pio_request *io = &vcpu->pio;
1791 kvm_arch_ops->cache_regs(vcpu);
1795 memcpy(&vcpu->regs[VCPU_REGS_RAX], vcpu->pio_data,
1799 r = pio_copy_data(vcpu);
1801 kvm_arch_ops->cache_regs(vcpu);
1808 delta *= io->cur_count;
1810 * The size of the register should really depend on
1811 * current address size.
1813 vcpu->regs[VCPU_REGS_RCX] -= delta;
1819 vcpu->regs[VCPU_REGS_RDI] += delta;
1821 vcpu->regs[VCPU_REGS_RSI] += delta;
1824 kvm_arch_ops->decache_regs(vcpu);
1826 io->count -= io->cur_count;
1830 kvm_arch_ops->skip_emulated_instruction(vcpu);
1834 static void kernel_pio(struct kvm_io_device *pio_dev,
1835 struct kvm_vcpu *vcpu,
1838 /* TODO: String I/O for in kernel device */
1840 mutex_lock(&vcpu->kvm->lock);
1842 kvm_iodevice_read(pio_dev, vcpu->pio.port,
1846 kvm_iodevice_write(pio_dev, vcpu->pio.port,
1849 mutex_unlock(&vcpu->kvm->lock);
1852 static void pio_string_write(struct kvm_io_device *pio_dev,
1853 struct kvm_vcpu *vcpu)
1855 struct kvm_pio_request *io = &vcpu->pio;
1856 void *pd = vcpu->pio_data;
1859 mutex_lock(&vcpu->kvm->lock);
1860 for (i = 0; i < io->cur_count; i++) {
1861 kvm_iodevice_write(pio_dev, io->port,
1866 mutex_unlock(&vcpu->kvm->lock);
1869 int kvm_emulate_pio (struct kvm_vcpu *vcpu, struct kvm_run *run, int in,
1870 int size, unsigned port)
1872 struct kvm_io_device *pio_dev;
1874 vcpu->run->exit_reason = KVM_EXIT_IO;
1875 vcpu->run->io.direction = in ? KVM_EXIT_IO_IN : KVM_EXIT_IO_OUT;
1876 vcpu->run->io.size = vcpu->pio.size = size;
1877 vcpu->run->io.data_offset = KVM_PIO_PAGE_OFFSET * PAGE_SIZE;
1878 vcpu->run->io.count = vcpu->pio.count = vcpu->pio.cur_count = 1;
1879 vcpu->run->io.port = vcpu->pio.port = port;
1881 vcpu->pio.string = 0;
1883 vcpu->pio.guest_page_offset = 0;
1886 kvm_arch_ops->cache_regs(vcpu);
1887 memcpy(vcpu->pio_data, &vcpu->regs[VCPU_REGS_RAX], 4);
1888 kvm_arch_ops->decache_regs(vcpu);
1890 pio_dev = vcpu_find_pio_dev(vcpu, port);
1892 kernel_pio(pio_dev, vcpu, vcpu->pio_data);
1898 EXPORT_SYMBOL_GPL(kvm_emulate_pio);
1900 int kvm_emulate_pio_string(struct kvm_vcpu *vcpu, struct kvm_run *run, int in,
1901 int size, unsigned long count, int down,
1902 gva_t address, int rep, unsigned port)
1904 unsigned now, in_page;
1908 struct kvm_io_device *pio_dev;
1910 vcpu->run->exit_reason = KVM_EXIT_IO;
1911 vcpu->run->io.direction = in ? KVM_EXIT_IO_IN : KVM_EXIT_IO_OUT;
1912 vcpu->run->io.size = vcpu->pio.size = size;
1913 vcpu->run->io.data_offset = KVM_PIO_PAGE_OFFSET * PAGE_SIZE;
1914 vcpu->run->io.count = vcpu->pio.count = vcpu->pio.cur_count = count;
1915 vcpu->run->io.port = vcpu->pio.port = port;
1917 vcpu->pio.string = 1;
1918 vcpu->pio.down = down;
1919 vcpu->pio.guest_page_offset = offset_in_page(address);
1920 vcpu->pio.rep = rep;
1923 kvm_arch_ops->skip_emulated_instruction(vcpu);
1928 in_page = PAGE_SIZE - offset_in_page(address);
1930 in_page = offset_in_page(address) + size;
1931 now = min(count, (unsigned long)in_page / size);
1934 * String I/O straddles page boundary. Pin two guest pages
1935 * so that we satisfy atomicity constraints. Do just one
1936 * transaction to avoid complexity.
1943 * String I/O in reverse. Yuck. Kill the guest, fix later.
1945 pr_unimpl(vcpu, "guest string pio down\n");
1949 vcpu->run->io.count = now;
1950 vcpu->pio.cur_count = now;
1952 for (i = 0; i < nr_pages; ++i) {
1953 mutex_lock(&vcpu->kvm->lock);
1954 page = gva_to_page(vcpu, address + i * PAGE_SIZE);
1957 vcpu->pio.guest_pages[i] = page;
1958 mutex_unlock(&vcpu->kvm->lock);
1961 free_pio_guest_pages(vcpu);
1966 pio_dev = vcpu_find_pio_dev(vcpu, port);
1967 if (!vcpu->pio.in) {
1968 /* string PIO write */
1969 ret = pio_copy_data(vcpu);
1970 if (ret >= 0 && pio_dev) {
1971 pio_string_write(pio_dev, vcpu);
1973 if (vcpu->pio.count == 0)
1977 pr_unimpl(vcpu, "no string pio read support yet, "
1978 "port %x size %d count %ld\n",
1983 EXPORT_SYMBOL_GPL(kvm_emulate_pio_string);
1985 static int kvm_vcpu_ioctl_run(struct kvm_vcpu *vcpu, struct kvm_run *kvm_run)
1992 if (vcpu->sigset_active)
1993 sigprocmask(SIG_SETMASK, &vcpu->sigset, &sigsaved);
1995 /* re-sync apic's tpr */
1996 set_cr8(vcpu, kvm_run->cr8);
1998 if (vcpu->pio.cur_count) {
1999 r = complete_pio(vcpu);
2004 if (vcpu->mmio_needed) {
2005 memcpy(vcpu->mmio_data, kvm_run->mmio.data, 8);
2006 vcpu->mmio_read_completed = 1;
2007 vcpu->mmio_needed = 0;
2008 r = emulate_instruction(vcpu, kvm_run,
2009 vcpu->mmio_fault_cr2, 0);
2010 if (r == EMULATE_DO_MMIO) {
2012 * Read-modify-write. Back to userspace.
2019 if (kvm_run->exit_reason == KVM_EXIT_HYPERCALL) {
2020 kvm_arch_ops->cache_regs(vcpu);
2021 vcpu->regs[VCPU_REGS_RAX] = kvm_run->hypercall.ret;
2022 kvm_arch_ops->decache_regs(vcpu);
2025 r = kvm_arch_ops->run(vcpu, kvm_run);
2028 if (vcpu->sigset_active)
2029 sigprocmask(SIG_SETMASK, &sigsaved, NULL);
2035 static int kvm_vcpu_ioctl_get_regs(struct kvm_vcpu *vcpu,
2036 struct kvm_regs *regs)
2040 kvm_arch_ops->cache_regs(vcpu);
2042 regs->rax = vcpu->regs[VCPU_REGS_RAX];
2043 regs->rbx = vcpu->regs[VCPU_REGS_RBX];
2044 regs->rcx = vcpu->regs[VCPU_REGS_RCX];
2045 regs->rdx = vcpu->regs[VCPU_REGS_RDX];
2046 regs->rsi = vcpu->regs[VCPU_REGS_RSI];
2047 regs->rdi = vcpu->regs[VCPU_REGS_RDI];
2048 regs->rsp = vcpu->regs[VCPU_REGS_RSP];
2049 regs->rbp = vcpu->regs[VCPU_REGS_RBP];
2050 #ifdef CONFIG_X86_64
2051 regs->r8 = vcpu->regs[VCPU_REGS_R8];
2052 regs->r9 = vcpu->regs[VCPU_REGS_R9];
2053 regs->r10 = vcpu->regs[VCPU_REGS_R10];
2054 regs->r11 = vcpu->regs[VCPU_REGS_R11];
2055 regs->r12 = vcpu->regs[VCPU_REGS_R12];
2056 regs->r13 = vcpu->regs[VCPU_REGS_R13];
2057 regs->r14 = vcpu->regs[VCPU_REGS_R14];
2058 regs->r15 = vcpu->regs[VCPU_REGS_R15];
2061 regs->rip = vcpu->rip;
2062 regs->rflags = kvm_arch_ops->get_rflags(vcpu);
2065 * Don't leak debug flags in case they were set for guest debugging
2067 if (vcpu->guest_debug.enabled && vcpu->guest_debug.singlestep)
2068 regs->rflags &= ~(X86_EFLAGS_TF | X86_EFLAGS_RF);
2075 static int kvm_vcpu_ioctl_set_regs(struct kvm_vcpu *vcpu,
2076 struct kvm_regs *regs)
2080 vcpu->regs[VCPU_REGS_RAX] = regs->rax;
2081 vcpu->regs[VCPU_REGS_RBX] = regs->rbx;
2082 vcpu->regs[VCPU_REGS_RCX] = regs->rcx;
2083 vcpu->regs[VCPU_REGS_RDX] = regs->rdx;
2084 vcpu->regs[VCPU_REGS_RSI] = regs->rsi;
2085 vcpu->regs[VCPU_REGS_RDI] = regs->rdi;
2086 vcpu->regs[VCPU_REGS_RSP] = regs->rsp;
2087 vcpu->regs[VCPU_REGS_RBP] = regs->rbp;
2088 #ifdef CONFIG_X86_64
2089 vcpu->regs[VCPU_REGS_R8] = regs->r8;
2090 vcpu->regs[VCPU_REGS_R9] = regs->r9;
2091 vcpu->regs[VCPU_REGS_R10] = regs->r10;
2092 vcpu->regs[VCPU_REGS_R11] = regs->r11;
2093 vcpu->regs[VCPU_REGS_R12] = regs->r12;
2094 vcpu->regs[VCPU_REGS_R13] = regs->r13;
2095 vcpu->regs[VCPU_REGS_R14] = regs->r14;
2096 vcpu->regs[VCPU_REGS_R15] = regs->r15;
2099 vcpu->rip = regs->rip;
2100 kvm_arch_ops->set_rflags(vcpu, regs->rflags);
2102 kvm_arch_ops->decache_regs(vcpu);
2109 static void get_segment(struct kvm_vcpu *vcpu,
2110 struct kvm_segment *var, int seg)
2112 return kvm_arch_ops->get_segment(vcpu, var, seg);
2115 static int kvm_vcpu_ioctl_get_sregs(struct kvm_vcpu *vcpu,
2116 struct kvm_sregs *sregs)
2118 struct descriptor_table dt;
2122 get_segment(vcpu, &sregs->cs, VCPU_SREG_CS);
2123 get_segment(vcpu, &sregs->ds, VCPU_SREG_DS);
2124 get_segment(vcpu, &sregs->es, VCPU_SREG_ES);
2125 get_segment(vcpu, &sregs->fs, VCPU_SREG_FS);
2126 get_segment(vcpu, &sregs->gs, VCPU_SREG_GS);
2127 get_segment(vcpu, &sregs->ss, VCPU_SREG_SS);
2129 get_segment(vcpu, &sregs->tr, VCPU_SREG_TR);
2130 get_segment(vcpu, &sregs->ldt, VCPU_SREG_LDTR);
2132 kvm_arch_ops->get_idt(vcpu, &dt);
2133 sregs->idt.limit = dt.limit;
2134 sregs->idt.base = dt.base;
2135 kvm_arch_ops->get_gdt(vcpu, &dt);
2136 sregs->gdt.limit = dt.limit;
2137 sregs->gdt.base = dt.base;
2139 kvm_arch_ops->decache_cr4_guest_bits(vcpu);
2140 sregs->cr0 = vcpu->cr0;
2141 sregs->cr2 = vcpu->cr2;
2142 sregs->cr3 = vcpu->cr3;
2143 sregs->cr4 = vcpu->cr4;
2144 sregs->cr8 = get_cr8(vcpu);
2145 sregs->efer = vcpu->shadow_efer;
2146 sregs->apic_base = kvm_get_apic_base(vcpu);
2148 memcpy(sregs->interrupt_bitmap, vcpu->irq_pending,
2149 sizeof sregs->interrupt_bitmap);
2156 static void set_segment(struct kvm_vcpu *vcpu,
2157 struct kvm_segment *var, int seg)
2159 return kvm_arch_ops->set_segment(vcpu, var, seg);
2162 static int kvm_vcpu_ioctl_set_sregs(struct kvm_vcpu *vcpu,
2163 struct kvm_sregs *sregs)
2165 int mmu_reset_needed = 0;
2167 struct descriptor_table dt;
2171 dt.limit = sregs->idt.limit;
2172 dt.base = sregs->idt.base;
2173 kvm_arch_ops->set_idt(vcpu, &dt);
2174 dt.limit = sregs->gdt.limit;
2175 dt.base = sregs->gdt.base;
2176 kvm_arch_ops->set_gdt(vcpu, &dt);
2178 vcpu->cr2 = sregs->cr2;
2179 mmu_reset_needed |= vcpu->cr3 != sregs->cr3;
2180 vcpu->cr3 = sregs->cr3;
2182 set_cr8(vcpu, sregs->cr8);
2184 mmu_reset_needed |= vcpu->shadow_efer != sregs->efer;
2185 #ifdef CONFIG_X86_64
2186 kvm_arch_ops->set_efer(vcpu, sregs->efer);
2188 kvm_set_apic_base(vcpu, sregs->apic_base);
2190 kvm_arch_ops->decache_cr4_guest_bits(vcpu);
2192 mmu_reset_needed |= vcpu->cr0 != sregs->cr0;
2193 kvm_arch_ops->set_cr0(vcpu, sregs->cr0);
2195 mmu_reset_needed |= vcpu->cr4 != sregs->cr4;
2196 kvm_arch_ops->set_cr4(vcpu, sregs->cr4);
2197 if (!is_long_mode(vcpu) && is_pae(vcpu))
2198 load_pdptrs(vcpu, vcpu->cr3);
2200 if (mmu_reset_needed)
2201 kvm_mmu_reset_context(vcpu);
2203 memcpy(vcpu->irq_pending, sregs->interrupt_bitmap,
2204 sizeof vcpu->irq_pending);
2205 vcpu->irq_summary = 0;
2206 for (i = 0; i < ARRAY_SIZE(vcpu->irq_pending); ++i)
2207 if (vcpu->irq_pending[i])
2208 __set_bit(i, &vcpu->irq_summary);
2210 set_segment(vcpu, &sregs->cs, VCPU_SREG_CS);
2211 set_segment(vcpu, &sregs->ds, VCPU_SREG_DS);
2212 set_segment(vcpu, &sregs->es, VCPU_SREG_ES);
2213 set_segment(vcpu, &sregs->fs, VCPU_SREG_FS);
2214 set_segment(vcpu, &sregs->gs, VCPU_SREG_GS);
2215 set_segment(vcpu, &sregs->ss, VCPU_SREG_SS);
2217 set_segment(vcpu, &sregs->tr, VCPU_SREG_TR);
2218 set_segment(vcpu, &sregs->ldt, VCPU_SREG_LDTR);
2226 * List of msr numbers which we expose to userspace through KVM_GET_MSRS
2227 * and KVM_SET_MSRS, and KVM_GET_MSR_INDEX_LIST.
2229 * This list is modified at module load time to reflect the
2230 * capabilities of the host cpu.
2232 static u32 msrs_to_save[] = {
2233 MSR_IA32_SYSENTER_CS, MSR_IA32_SYSENTER_ESP, MSR_IA32_SYSENTER_EIP,
2235 #ifdef CONFIG_X86_64
2236 MSR_CSTAR, MSR_KERNEL_GS_BASE, MSR_SYSCALL_MASK, MSR_LSTAR,
2238 MSR_IA32_TIME_STAMP_COUNTER,
2241 static unsigned num_msrs_to_save;
2243 static u32 emulated_msrs[] = {
2244 MSR_IA32_MISC_ENABLE,
2247 static __init void kvm_init_msr_list(void)
2252 for (i = j = 0; i < ARRAY_SIZE(msrs_to_save); i++) {
2253 if (rdmsr_safe(msrs_to_save[i], &dummy[0], &dummy[1]) < 0)
2256 msrs_to_save[j] = msrs_to_save[i];
2259 num_msrs_to_save = j;
2263 * Adapt set_msr() to msr_io()'s calling convention
2265 static int do_set_msr(struct kvm_vcpu *vcpu, unsigned index, u64 *data)
2267 return kvm_set_msr(vcpu, index, *data);
2271 * Read or write a bunch of msrs. All parameters are kernel addresses.
2273 * @return number of msrs set successfully.
2275 static int __msr_io(struct kvm_vcpu *vcpu, struct kvm_msrs *msrs,
2276 struct kvm_msr_entry *entries,
2277 int (*do_msr)(struct kvm_vcpu *vcpu,
2278 unsigned index, u64 *data))
2284 for (i = 0; i < msrs->nmsrs; ++i)
2285 if (do_msr(vcpu, entries[i].index, &entries[i].data))
2294 * Read or write a bunch of msrs. Parameters are user addresses.
2296 * @return number of msrs set successfully.
2298 static int msr_io(struct kvm_vcpu *vcpu, struct kvm_msrs __user *user_msrs,
2299 int (*do_msr)(struct kvm_vcpu *vcpu,
2300 unsigned index, u64 *data),
2303 struct kvm_msrs msrs;
2304 struct kvm_msr_entry *entries;
2309 if (copy_from_user(&msrs, user_msrs, sizeof msrs))
2313 if (msrs.nmsrs >= MAX_IO_MSRS)
2317 size = sizeof(struct kvm_msr_entry) * msrs.nmsrs;
2318 entries = vmalloc(size);
2323 if (copy_from_user(entries, user_msrs->entries, size))
2326 r = n = __msr_io(vcpu, &msrs, entries, do_msr);
2331 if (writeback && copy_to_user(user_msrs->entries, entries, size))
2343 * Translate a guest virtual address to a guest physical address.
2345 static int kvm_vcpu_ioctl_translate(struct kvm_vcpu *vcpu,
2346 struct kvm_translation *tr)
2348 unsigned long vaddr = tr->linear_address;
2352 mutex_lock(&vcpu->kvm->lock);
2353 gpa = vcpu->mmu.gva_to_gpa(vcpu, vaddr);
2354 tr->physical_address = gpa;
2355 tr->valid = gpa != UNMAPPED_GVA;
2358 mutex_unlock(&vcpu->kvm->lock);
2364 static int kvm_vcpu_ioctl_interrupt(struct kvm_vcpu *vcpu,
2365 struct kvm_interrupt *irq)
2367 if (irq->irq < 0 || irq->irq >= 256)
2369 if (irqchip_in_kernel(vcpu->kvm))
2373 set_bit(irq->irq, vcpu->irq_pending);
2374 set_bit(irq->irq / BITS_PER_LONG, &vcpu->irq_summary);
2381 static int kvm_vcpu_ioctl_debug_guest(struct kvm_vcpu *vcpu,
2382 struct kvm_debug_guest *dbg)
2388 r = kvm_arch_ops->set_guest_debug(vcpu, dbg);
2395 static struct page *kvm_vcpu_nopage(struct vm_area_struct *vma,
2396 unsigned long address,
2399 struct kvm_vcpu *vcpu = vma->vm_file->private_data;
2400 unsigned long pgoff;
2403 pgoff = ((address - vma->vm_start) >> PAGE_SHIFT) + vma->vm_pgoff;
2405 page = virt_to_page(vcpu->run);
2406 else if (pgoff == KVM_PIO_PAGE_OFFSET)
2407 page = virt_to_page(vcpu->pio_data);
2409 return NOPAGE_SIGBUS;
2412 *type = VM_FAULT_MINOR;
2417 static struct vm_operations_struct kvm_vcpu_vm_ops = {
2418 .nopage = kvm_vcpu_nopage,
2421 static int kvm_vcpu_mmap(struct file *file, struct vm_area_struct *vma)
2423 vma->vm_ops = &kvm_vcpu_vm_ops;
2427 static int kvm_vcpu_release(struct inode *inode, struct file *filp)
2429 struct kvm_vcpu *vcpu = filp->private_data;
2431 fput(vcpu->kvm->filp);
2435 static struct file_operations kvm_vcpu_fops = {
2436 .release = kvm_vcpu_release,
2437 .unlocked_ioctl = kvm_vcpu_ioctl,
2438 .compat_ioctl = kvm_vcpu_ioctl,
2439 .mmap = kvm_vcpu_mmap,
2443 * Allocates an inode for the vcpu.
2445 static int create_vcpu_fd(struct kvm_vcpu *vcpu)
2448 struct inode *inode;
2451 r = anon_inode_getfd(&fd, &inode, &file,
2452 "kvm-vcpu", &kvm_vcpu_fops, vcpu);
2455 atomic_inc(&vcpu->kvm->filp->f_count);
2460 * Creates some virtual cpus. Good luck creating more than one.
2462 static int kvm_vm_ioctl_create_vcpu(struct kvm *kvm, int n)
2465 struct kvm_vcpu *vcpu;
2470 vcpu = kvm_arch_ops->vcpu_create(kvm, n);
2472 return PTR_ERR(vcpu);
2474 preempt_notifier_init(&vcpu->preempt_notifier, &kvm_preempt_ops);
2476 /* We do fxsave: this must be aligned. */
2477 BUG_ON((unsigned long)&vcpu->host_fx_image & 0xF);
2480 r = kvm_mmu_setup(vcpu);
2485 mutex_lock(&kvm->lock);
2486 if (kvm->vcpus[n]) {
2488 mutex_unlock(&kvm->lock);
2491 kvm->vcpus[n] = vcpu;
2492 mutex_unlock(&kvm->lock);
2494 /* Now it's all set up, let userspace reach it */
2495 r = create_vcpu_fd(vcpu);
2501 mutex_lock(&kvm->lock);
2502 kvm->vcpus[n] = NULL;
2503 mutex_unlock(&kvm->lock);
2507 kvm_mmu_unload(vcpu);
2511 kvm_arch_ops->vcpu_free(vcpu);
2515 static void cpuid_fix_nx_cap(struct kvm_vcpu *vcpu)
2519 struct kvm_cpuid_entry *e, *entry;
2521 rdmsrl(MSR_EFER, efer);
2523 for (i = 0; i < vcpu->cpuid_nent; ++i) {
2524 e = &vcpu->cpuid_entries[i];
2525 if (e->function == 0x80000001) {
2530 if (entry && (entry->edx & (1 << 20)) && !(efer & EFER_NX)) {
2531 entry->edx &= ~(1 << 20);
2532 printk(KERN_INFO "kvm: guest NX capability removed\n");
2536 static int kvm_vcpu_ioctl_set_cpuid(struct kvm_vcpu *vcpu,
2537 struct kvm_cpuid *cpuid,
2538 struct kvm_cpuid_entry __user *entries)
2543 if (cpuid->nent > KVM_MAX_CPUID_ENTRIES)
2546 if (copy_from_user(&vcpu->cpuid_entries, entries,
2547 cpuid->nent * sizeof(struct kvm_cpuid_entry)))
2549 vcpu->cpuid_nent = cpuid->nent;
2550 cpuid_fix_nx_cap(vcpu);
2557 static int kvm_vcpu_ioctl_set_sigmask(struct kvm_vcpu *vcpu, sigset_t *sigset)
2560 sigdelsetmask(sigset, sigmask(SIGKILL)|sigmask(SIGSTOP));
2561 vcpu->sigset_active = 1;
2562 vcpu->sigset = *sigset;
2564 vcpu->sigset_active = 0;
2569 * fxsave fpu state. Taken from x86_64/processor.h. To be killed when
2570 * we have asm/x86/processor.h
2581 u32 st_space[32]; /* 8*16 bytes for each FP-reg = 128 bytes */
2582 #ifdef CONFIG_X86_64
2583 u32 xmm_space[64]; /* 16*16 bytes for each XMM-reg = 256 bytes */
2585 u32 xmm_space[32]; /* 8*16 bytes for each XMM-reg = 128 bytes */
2589 static int kvm_vcpu_ioctl_get_fpu(struct kvm_vcpu *vcpu, struct kvm_fpu *fpu)
2591 struct fxsave *fxsave = (struct fxsave *)&vcpu->guest_fx_image;
2595 memcpy(fpu->fpr, fxsave->st_space, 128);
2596 fpu->fcw = fxsave->cwd;
2597 fpu->fsw = fxsave->swd;
2598 fpu->ftwx = fxsave->twd;
2599 fpu->last_opcode = fxsave->fop;
2600 fpu->last_ip = fxsave->rip;
2601 fpu->last_dp = fxsave->rdp;
2602 memcpy(fpu->xmm, fxsave->xmm_space, sizeof fxsave->xmm_space);
2609 static int kvm_vcpu_ioctl_set_fpu(struct kvm_vcpu *vcpu, struct kvm_fpu *fpu)
2611 struct fxsave *fxsave = (struct fxsave *)&vcpu->guest_fx_image;
2615 memcpy(fxsave->st_space, fpu->fpr, 128);
2616 fxsave->cwd = fpu->fcw;
2617 fxsave->swd = fpu->fsw;
2618 fxsave->twd = fpu->ftwx;
2619 fxsave->fop = fpu->last_opcode;
2620 fxsave->rip = fpu->last_ip;
2621 fxsave->rdp = fpu->last_dp;
2622 memcpy(fxsave->xmm_space, fpu->xmm, sizeof fxsave->xmm_space);
2629 static long kvm_vcpu_ioctl(struct file *filp,
2630 unsigned int ioctl, unsigned long arg)
2632 struct kvm_vcpu *vcpu = filp->private_data;
2633 void __user *argp = (void __user *)arg;
2641 r = kvm_vcpu_ioctl_run(vcpu, vcpu->run);
2643 case KVM_GET_REGS: {
2644 struct kvm_regs kvm_regs;
2646 memset(&kvm_regs, 0, sizeof kvm_regs);
2647 r = kvm_vcpu_ioctl_get_regs(vcpu, &kvm_regs);
2651 if (copy_to_user(argp, &kvm_regs, sizeof kvm_regs))
2656 case KVM_SET_REGS: {
2657 struct kvm_regs kvm_regs;
2660 if (copy_from_user(&kvm_regs, argp, sizeof kvm_regs))
2662 r = kvm_vcpu_ioctl_set_regs(vcpu, &kvm_regs);
2668 case KVM_GET_SREGS: {
2669 struct kvm_sregs kvm_sregs;
2671 memset(&kvm_sregs, 0, sizeof kvm_sregs);
2672 r = kvm_vcpu_ioctl_get_sregs(vcpu, &kvm_sregs);
2676 if (copy_to_user(argp, &kvm_sregs, sizeof kvm_sregs))
2681 case KVM_SET_SREGS: {
2682 struct kvm_sregs kvm_sregs;
2685 if (copy_from_user(&kvm_sregs, argp, sizeof kvm_sregs))
2687 r = kvm_vcpu_ioctl_set_sregs(vcpu, &kvm_sregs);
2693 case KVM_TRANSLATE: {
2694 struct kvm_translation tr;
2697 if (copy_from_user(&tr, argp, sizeof tr))
2699 r = kvm_vcpu_ioctl_translate(vcpu, &tr);
2703 if (copy_to_user(argp, &tr, sizeof tr))
2708 case KVM_INTERRUPT: {
2709 struct kvm_interrupt irq;
2712 if (copy_from_user(&irq, argp, sizeof irq))
2714 r = kvm_vcpu_ioctl_interrupt(vcpu, &irq);
2720 case KVM_DEBUG_GUEST: {
2721 struct kvm_debug_guest dbg;
2724 if (copy_from_user(&dbg, argp, sizeof dbg))
2726 r = kvm_vcpu_ioctl_debug_guest(vcpu, &dbg);
2733 r = msr_io(vcpu, argp, kvm_get_msr, 1);
2736 r = msr_io(vcpu, argp, do_set_msr, 0);
2738 case KVM_SET_CPUID: {
2739 struct kvm_cpuid __user *cpuid_arg = argp;
2740 struct kvm_cpuid cpuid;
2743 if (copy_from_user(&cpuid, cpuid_arg, sizeof cpuid))
2745 r = kvm_vcpu_ioctl_set_cpuid(vcpu, &cpuid, cpuid_arg->entries);
2750 case KVM_SET_SIGNAL_MASK: {
2751 struct kvm_signal_mask __user *sigmask_arg = argp;
2752 struct kvm_signal_mask kvm_sigmask;
2753 sigset_t sigset, *p;
2758 if (copy_from_user(&kvm_sigmask, argp,
2759 sizeof kvm_sigmask))
2762 if (kvm_sigmask.len != sizeof sigset)
2765 if (copy_from_user(&sigset, sigmask_arg->sigset,
2770 r = kvm_vcpu_ioctl_set_sigmask(vcpu, &sigset);
2776 memset(&fpu, 0, sizeof fpu);
2777 r = kvm_vcpu_ioctl_get_fpu(vcpu, &fpu);
2781 if (copy_to_user(argp, &fpu, sizeof fpu))
2790 if (copy_from_user(&fpu, argp, sizeof fpu))
2792 r = kvm_vcpu_ioctl_set_fpu(vcpu, &fpu);
2805 static long kvm_vm_ioctl(struct file *filp,
2806 unsigned int ioctl, unsigned long arg)
2808 struct kvm *kvm = filp->private_data;
2809 void __user *argp = (void __user *)arg;
2813 case KVM_CREATE_VCPU:
2814 r = kvm_vm_ioctl_create_vcpu(kvm, arg);
2818 case KVM_SET_MEMORY_REGION: {
2819 struct kvm_memory_region kvm_mem;
2822 if (copy_from_user(&kvm_mem, argp, sizeof kvm_mem))
2824 r = kvm_vm_ioctl_set_memory_region(kvm, &kvm_mem);
2829 case KVM_GET_DIRTY_LOG: {
2830 struct kvm_dirty_log log;
2833 if (copy_from_user(&log, argp, sizeof log))
2835 r = kvm_vm_ioctl_get_dirty_log(kvm, &log);
2840 case KVM_SET_MEMORY_ALIAS: {
2841 struct kvm_memory_alias alias;
2844 if (copy_from_user(&alias, argp, sizeof alias))
2846 r = kvm_vm_ioctl_set_memory_alias(kvm, &alias);
2851 case KVM_CREATE_IRQCHIP:
2853 kvm->vpic = kvm_create_pic(kvm);
2855 r = kvm_ioapic_init(kvm);
2865 case KVM_IRQ_LINE: {
2866 struct kvm_irq_level irq_event;
2869 if (copy_from_user(&irq_event, argp, sizeof irq_event))
2871 if (irqchip_in_kernel(kvm)) {
2872 mutex_lock(&kvm->lock);
2873 if (irq_event.irq < 16)
2874 kvm_pic_set_irq(pic_irqchip(kvm),
2877 kvm_ioapic_set_irq(kvm->vioapic,
2880 mutex_unlock(&kvm->lock);
2885 case KVM_GET_IRQCHIP: {
2886 /* 0: PIC master, 1: PIC slave, 2: IOAPIC */
2887 struct kvm_irqchip chip;
2890 if (copy_from_user(&chip, argp, sizeof chip))
2893 if (!irqchip_in_kernel(kvm))
2895 r = kvm_vm_ioctl_get_irqchip(kvm, &chip);
2899 if (copy_to_user(argp, &chip, sizeof chip))
2904 case KVM_SET_IRQCHIP: {
2905 /* 0: PIC master, 1: PIC slave, 2: IOAPIC */
2906 struct kvm_irqchip chip;
2909 if (copy_from_user(&chip, argp, sizeof chip))
2912 if (!irqchip_in_kernel(kvm))
2914 r = kvm_vm_ioctl_set_irqchip(kvm, &chip);
2927 static struct page *kvm_vm_nopage(struct vm_area_struct *vma,
2928 unsigned long address,
2931 struct kvm *kvm = vma->vm_file->private_data;
2932 unsigned long pgoff;
2935 pgoff = ((address - vma->vm_start) >> PAGE_SHIFT) + vma->vm_pgoff;
2936 page = gfn_to_page(kvm, pgoff);
2938 return NOPAGE_SIGBUS;
2941 *type = VM_FAULT_MINOR;
2946 static struct vm_operations_struct kvm_vm_vm_ops = {
2947 .nopage = kvm_vm_nopage,
2950 static int kvm_vm_mmap(struct file *file, struct vm_area_struct *vma)
2952 vma->vm_ops = &kvm_vm_vm_ops;
2956 static struct file_operations kvm_vm_fops = {
2957 .release = kvm_vm_release,
2958 .unlocked_ioctl = kvm_vm_ioctl,
2959 .compat_ioctl = kvm_vm_ioctl,
2960 .mmap = kvm_vm_mmap,
2963 static int kvm_dev_ioctl_create_vm(void)
2966 struct inode *inode;
2970 kvm = kvm_create_vm();
2972 return PTR_ERR(kvm);
2973 r = anon_inode_getfd(&fd, &inode, &file, "kvm-vm", &kvm_vm_fops, kvm);
2975 kvm_destroy_vm(kvm);
2984 static long kvm_dev_ioctl(struct file *filp,
2985 unsigned int ioctl, unsigned long arg)
2987 void __user *argp = (void __user *)arg;
2991 case KVM_GET_API_VERSION:
2995 r = KVM_API_VERSION;
3001 r = kvm_dev_ioctl_create_vm();
3003 case KVM_GET_MSR_INDEX_LIST: {
3004 struct kvm_msr_list __user *user_msr_list = argp;
3005 struct kvm_msr_list msr_list;
3009 if (copy_from_user(&msr_list, user_msr_list, sizeof msr_list))
3012 msr_list.nmsrs = num_msrs_to_save + ARRAY_SIZE(emulated_msrs);
3013 if (copy_to_user(user_msr_list, &msr_list, sizeof msr_list))
3016 if (n < num_msrs_to_save)
3019 if (copy_to_user(user_msr_list->indices, &msrs_to_save,
3020 num_msrs_to_save * sizeof(u32)))
3022 if (copy_to_user(user_msr_list->indices
3023 + num_msrs_to_save * sizeof(u32),
3025 ARRAY_SIZE(emulated_msrs) * sizeof(u32)))
3030 case KVM_CHECK_EXTENSION: {
3031 int ext = (long)argp;
3034 case KVM_CAP_IRQCHIP:
3044 case KVM_GET_VCPU_MMAP_SIZE:
3057 static struct file_operations kvm_chardev_ops = {
3058 .unlocked_ioctl = kvm_dev_ioctl,
3059 .compat_ioctl = kvm_dev_ioctl,
3062 static struct miscdevice kvm_dev = {
3069 * Make sure that a cpu that is being hot-unplugged does not have any vcpus
3072 static void decache_vcpus_on_cpu(int cpu)
3075 struct kvm_vcpu *vcpu;
3078 spin_lock(&kvm_lock);
3079 list_for_each_entry(vm, &vm_list, vm_list)
3080 for (i = 0; i < KVM_MAX_VCPUS; ++i) {
3081 vcpu = vm->vcpus[i];
3085 * If the vcpu is locked, then it is running on some
3086 * other cpu and therefore it is not cached on the
3089 * If it's not locked, check the last cpu it executed
3092 if (mutex_trylock(&vcpu->mutex)) {
3093 if (vcpu->cpu == cpu) {
3094 kvm_arch_ops->vcpu_decache(vcpu);
3097 mutex_unlock(&vcpu->mutex);
3100 spin_unlock(&kvm_lock);
3103 static void hardware_enable(void *junk)
3105 int cpu = raw_smp_processor_id();
3107 if (cpu_isset(cpu, cpus_hardware_enabled))
3109 cpu_set(cpu, cpus_hardware_enabled);
3110 kvm_arch_ops->hardware_enable(NULL);
3113 static void hardware_disable(void *junk)
3115 int cpu = raw_smp_processor_id();
3117 if (!cpu_isset(cpu, cpus_hardware_enabled))
3119 cpu_clear(cpu, cpus_hardware_enabled);
3120 decache_vcpus_on_cpu(cpu);
3121 kvm_arch_ops->hardware_disable(NULL);
3124 static int kvm_cpu_hotplug(struct notifier_block *notifier, unsigned long val,
3131 case CPU_DYING_FROZEN:
3132 printk(KERN_INFO "kvm: disabling virtualization on CPU%d\n",
3134 hardware_disable(NULL);
3136 case CPU_UP_CANCELED:
3137 case CPU_UP_CANCELED_FROZEN:
3138 printk(KERN_INFO "kvm: disabling virtualization on CPU%d\n",
3140 smp_call_function_single(cpu, hardware_disable, NULL, 0, 1);
3143 case CPU_ONLINE_FROZEN:
3144 printk(KERN_INFO "kvm: enabling virtualization on CPU%d\n",
3146 smp_call_function_single(cpu, hardware_enable, NULL, 0, 1);
3152 static int kvm_reboot(struct notifier_block *notifier, unsigned long val,
3155 if (val == SYS_RESTART) {
3157 * Some (well, at least mine) BIOSes hang on reboot if
3160 printk(KERN_INFO "kvm: exiting hardware virtualization\n");
3161 on_each_cpu(hardware_disable, NULL, 0, 1);
3166 static struct notifier_block kvm_reboot_notifier = {
3167 .notifier_call = kvm_reboot,
3171 void kvm_io_bus_init(struct kvm_io_bus *bus)
3173 memset(bus, 0, sizeof(*bus));
3176 void kvm_io_bus_destroy(struct kvm_io_bus *bus)
3180 for (i = 0; i < bus->dev_count; i++) {
3181 struct kvm_io_device *pos = bus->devs[i];
3183 kvm_iodevice_destructor(pos);
3187 struct kvm_io_device *kvm_io_bus_find_dev(struct kvm_io_bus *bus, gpa_t addr)
3191 for (i = 0; i < bus->dev_count; i++) {
3192 struct kvm_io_device *pos = bus->devs[i];
3194 if (pos->in_range(pos, addr))
3201 void kvm_io_bus_register_dev(struct kvm_io_bus *bus, struct kvm_io_device *dev)
3203 BUG_ON(bus->dev_count > (NR_IOBUS_DEVS-1));
3205 bus->devs[bus->dev_count++] = dev;
3208 static struct notifier_block kvm_cpu_notifier = {
3209 .notifier_call = kvm_cpu_hotplug,
3210 .priority = 20, /* must be > scheduler priority */
3213 static u64 stat_get(void *_offset)
3215 unsigned offset = (long)_offset;
3218 struct kvm_vcpu *vcpu;
3221 spin_lock(&kvm_lock);
3222 list_for_each_entry(kvm, &vm_list, vm_list)
3223 for (i = 0; i < KVM_MAX_VCPUS; ++i) {
3224 vcpu = kvm->vcpus[i];
3226 total += *(u32 *)((void *)vcpu + offset);
3228 spin_unlock(&kvm_lock);
3232 DEFINE_SIMPLE_ATTRIBUTE(stat_fops, stat_get, NULL, "%llu\n");
3234 static __init void kvm_init_debug(void)
3236 struct kvm_stats_debugfs_item *p;
3238 debugfs_dir = debugfs_create_dir("kvm", NULL);
3239 for (p = debugfs_entries; p->name; ++p)
3240 p->dentry = debugfs_create_file(p->name, 0444, debugfs_dir,
3241 (void *)(long)p->offset,
3245 static void kvm_exit_debug(void)
3247 struct kvm_stats_debugfs_item *p;
3249 for (p = debugfs_entries; p->name; ++p)
3250 debugfs_remove(p->dentry);
3251 debugfs_remove(debugfs_dir);
3254 static int kvm_suspend(struct sys_device *dev, pm_message_t state)
3256 hardware_disable(NULL);
3260 static int kvm_resume(struct sys_device *dev)
3262 hardware_enable(NULL);
3266 static struct sysdev_class kvm_sysdev_class = {
3267 set_kset_name("kvm"),
3268 .suspend = kvm_suspend,
3269 .resume = kvm_resume,
3272 static struct sys_device kvm_sysdev = {
3274 .cls = &kvm_sysdev_class,
3277 hpa_t bad_page_address;
3280 struct kvm_vcpu *preempt_notifier_to_vcpu(struct preempt_notifier *pn)
3282 return container_of(pn, struct kvm_vcpu, preempt_notifier);
3285 static void kvm_sched_in(struct preempt_notifier *pn, int cpu)
3287 struct kvm_vcpu *vcpu = preempt_notifier_to_vcpu(pn);
3289 kvm_arch_ops->vcpu_load(vcpu, cpu);
3292 static void kvm_sched_out(struct preempt_notifier *pn,
3293 struct task_struct *next)
3295 struct kvm_vcpu *vcpu = preempt_notifier_to_vcpu(pn);
3297 kvm_arch_ops->vcpu_put(vcpu);
3300 int kvm_init_arch(struct kvm_arch_ops *ops, unsigned int vcpu_size,
3301 struct module *module)
3307 printk(KERN_ERR "kvm: already loaded the other module\n");
3311 if (!ops->cpu_has_kvm_support()) {
3312 printk(KERN_ERR "kvm: no hardware support\n");
3315 if (ops->disabled_by_bios()) {
3316 printk(KERN_ERR "kvm: disabled by bios\n");
3322 r = kvm_arch_ops->hardware_setup();
3326 for_each_online_cpu(cpu) {
3327 smp_call_function_single(cpu,
3328 kvm_arch_ops->check_processor_compatibility,
3334 on_each_cpu(hardware_enable, NULL, 0, 1);
3335 r = register_cpu_notifier(&kvm_cpu_notifier);
3338 register_reboot_notifier(&kvm_reboot_notifier);
3340 r = sysdev_class_register(&kvm_sysdev_class);
3344 r = sysdev_register(&kvm_sysdev);
3348 /* A kmem cache lets us meet the alignment requirements of fx_save. */
3349 kvm_vcpu_cache = kmem_cache_create("kvm_vcpu", vcpu_size,
3350 __alignof__(struct kvm_vcpu), 0, 0);
3351 if (!kvm_vcpu_cache) {
3356 kvm_chardev_ops.owner = module;
3358 r = misc_register(&kvm_dev);
3360 printk (KERN_ERR "kvm: misc device register failed\n");
3364 kvm_preempt_ops.sched_in = kvm_sched_in;
3365 kvm_preempt_ops.sched_out = kvm_sched_out;
3370 kmem_cache_destroy(kvm_vcpu_cache);
3372 sysdev_unregister(&kvm_sysdev);
3374 sysdev_class_unregister(&kvm_sysdev_class);
3376 unregister_reboot_notifier(&kvm_reboot_notifier);
3377 unregister_cpu_notifier(&kvm_cpu_notifier);
3379 on_each_cpu(hardware_disable, NULL, 0, 1);
3381 kvm_arch_ops->hardware_unsetup();
3383 kvm_arch_ops = NULL;
3387 void kvm_exit_arch(void)
3389 misc_deregister(&kvm_dev);
3390 kmem_cache_destroy(kvm_vcpu_cache);
3391 sysdev_unregister(&kvm_sysdev);
3392 sysdev_class_unregister(&kvm_sysdev_class);
3393 unregister_reboot_notifier(&kvm_reboot_notifier);
3394 unregister_cpu_notifier(&kvm_cpu_notifier);
3395 on_each_cpu(hardware_disable, NULL, 0, 1);
3396 kvm_arch_ops->hardware_unsetup();
3397 kvm_arch_ops = NULL;
3400 static __init int kvm_init(void)
3402 static struct page *bad_page;
3405 r = kvm_mmu_module_init();
3411 kvm_init_msr_list();
3413 if ((bad_page = alloc_page(GFP_KERNEL)) == NULL) {
3418 bad_page_address = page_to_pfn(bad_page) << PAGE_SHIFT;
3419 memset(__va(bad_page_address), 0, PAGE_SIZE);
3425 kvm_mmu_module_exit();
3430 static __exit void kvm_exit(void)
3433 __free_page(pfn_to_page(bad_page_address >> PAGE_SHIFT));
3434 kvm_mmu_module_exit();
3437 module_init(kvm_init)
3438 module_exit(kvm_exit)
3440 EXPORT_SYMBOL_GPL(kvm_init_arch);
3441 EXPORT_SYMBOL_GPL(kvm_exit_arch);