2 * Kernel-based Virtual Machine driver for Linux
4 * This module enables machines with Intel VT-x extensions to run virtual
5 * machines without emulation or binary translation.
7 * Copyright (C) 2006 Qumranet, Inc.
10 * Avi Kivity <avi@qumranet.com>
11 * Yaniv Kamay <yaniv@qumranet.com>
13 * This work is licensed under the terms of the GNU GPL, version 2. See
14 * the COPYING file in the top-level directory.
19 #include "x86_emulate.h"
20 #include "segment_descriptor.h"
22 #include <linux/kvm.h>
23 #include <linux/module.h>
24 #include <linux/errno.h>
25 #include <linux/percpu.h>
26 #include <linux/gfp.h>
28 #include <linux/miscdevice.h>
29 #include <linux/vmalloc.h>
30 #include <linux/reboot.h>
31 #include <linux/debugfs.h>
32 #include <linux/highmem.h>
33 #include <linux/file.h>
34 #include <linux/sysdev.h>
35 #include <linux/cpu.h>
36 #include <linux/sched.h>
37 #include <linux/cpumask.h>
38 #include <linux/smp.h>
39 #include <linux/anon_inodes.h>
41 #include <asm/processor.h>
44 #include <asm/uaccess.h>
47 MODULE_AUTHOR("Qumranet");
48 MODULE_LICENSE("GPL");
50 static DEFINE_SPINLOCK(kvm_lock);
51 static LIST_HEAD(vm_list);
53 static cpumask_t cpus_hardware_enabled;
55 struct kvm_arch_ops *kvm_arch_ops;
57 static void hardware_disable(void *ignored);
59 #define STAT_OFFSET(x) offsetof(struct kvm_vcpu, stat.x)
61 static struct kvm_stats_debugfs_item {
64 struct dentry *dentry;
65 } debugfs_entries[] = {
66 { "pf_fixed", STAT_OFFSET(pf_fixed) },
67 { "pf_guest", STAT_OFFSET(pf_guest) },
68 { "tlb_flush", STAT_OFFSET(tlb_flush) },
69 { "invlpg", STAT_OFFSET(invlpg) },
70 { "exits", STAT_OFFSET(exits) },
71 { "io_exits", STAT_OFFSET(io_exits) },
72 { "mmio_exits", STAT_OFFSET(mmio_exits) },
73 { "signal_exits", STAT_OFFSET(signal_exits) },
74 { "irq_window", STAT_OFFSET(irq_window_exits) },
75 { "halt_exits", STAT_OFFSET(halt_exits) },
76 { "request_irq", STAT_OFFSET(request_irq_exits) },
77 { "irq_exits", STAT_OFFSET(irq_exits) },
78 { "light_exits", STAT_OFFSET(light_exits) },
79 { "efer_reload", STAT_OFFSET(efer_reload) },
83 static struct dentry *debugfs_dir;
85 #define MAX_IO_MSRS 256
87 #define CR0_RESEVED_BITS 0xffffffff1ffaffc0ULL
88 #define LMSW_GUEST_MASK 0x0eULL
89 #define CR4_RESEVED_BITS (~((1ULL << 11) - 1))
90 #define CR8_RESEVED_BITS (~0x0fULL)
91 #define EFER_RESERVED_BITS 0xfffffffffffff2fe
94 // LDT or TSS descriptor in the GDT. 16 bytes.
95 struct segment_descriptor_64 {
96 struct segment_descriptor s;
103 static long kvm_vcpu_ioctl(struct file *file, unsigned int ioctl,
106 unsigned long segment_base(u16 selector)
108 struct descriptor_table gdt;
109 struct segment_descriptor *d;
110 unsigned long table_base;
111 typedef unsigned long ul;
117 asm ("sgdt %0" : "=m"(gdt));
118 table_base = gdt.base;
120 if (selector & 4) { /* from ldt */
123 asm ("sldt %0" : "=g"(ldt_selector));
124 table_base = segment_base(ldt_selector);
126 d = (struct segment_descriptor *)(table_base + (selector & ~7));
127 v = d->base_low | ((ul)d->base_mid << 16) | ((ul)d->base_high << 24);
130 && (d->type == 2 || d->type == 9 || d->type == 11))
131 v |= ((ul)((struct segment_descriptor_64 *)d)->base_higher) << 32;
135 EXPORT_SYMBOL_GPL(segment_base);
137 static inline int valid_vcpu(int n)
139 return likely(n >= 0 && n < KVM_MAX_VCPUS);
142 int kvm_read_guest(struct kvm_vcpu *vcpu, gva_t addr, unsigned long size,
145 unsigned char *host_buf = dest;
146 unsigned long req_size = size;
154 paddr = gva_to_hpa(vcpu, addr);
156 if (is_error_hpa(paddr))
159 guest_buf = (hva_t)kmap_atomic(
160 pfn_to_page(paddr >> PAGE_SHIFT),
162 offset = addr & ~PAGE_MASK;
164 now = min(size, PAGE_SIZE - offset);
165 memcpy(host_buf, (void*)guest_buf, now);
169 kunmap_atomic((void *)(guest_buf & PAGE_MASK), KM_USER0);
171 return req_size - size;
173 EXPORT_SYMBOL_GPL(kvm_read_guest);
175 int kvm_write_guest(struct kvm_vcpu *vcpu, gva_t addr, unsigned long size,
178 unsigned char *host_buf = data;
179 unsigned long req_size = size;
188 paddr = gva_to_hpa(vcpu, addr);
190 if (is_error_hpa(paddr))
193 gfn = vcpu->mmu.gva_to_gpa(vcpu, addr) >> PAGE_SHIFT;
194 mark_page_dirty(vcpu->kvm, gfn);
195 guest_buf = (hva_t)kmap_atomic(
196 pfn_to_page(paddr >> PAGE_SHIFT), KM_USER0);
197 offset = addr & ~PAGE_MASK;
199 now = min(size, PAGE_SIZE - offset);
200 memcpy((void*)guest_buf, host_buf, now);
204 kunmap_atomic((void *)(guest_buf & PAGE_MASK), KM_USER0);
206 return req_size - size;
208 EXPORT_SYMBOL_GPL(kvm_write_guest);
210 void kvm_load_guest_fpu(struct kvm_vcpu *vcpu)
212 if (!vcpu->fpu_active || vcpu->guest_fpu_loaded)
215 vcpu->guest_fpu_loaded = 1;
216 fx_save(vcpu->host_fx_image);
217 fx_restore(vcpu->guest_fx_image);
219 EXPORT_SYMBOL_GPL(kvm_load_guest_fpu);
221 void kvm_put_guest_fpu(struct kvm_vcpu *vcpu)
223 if (!vcpu->guest_fpu_loaded)
226 vcpu->guest_fpu_loaded = 0;
227 fx_save(vcpu->guest_fx_image);
228 fx_restore(vcpu->host_fx_image);
230 EXPORT_SYMBOL_GPL(kvm_put_guest_fpu);
233 * Switches to specified vcpu, until a matching vcpu_put()
235 static void vcpu_load(struct kvm_vcpu *vcpu)
237 mutex_lock(&vcpu->mutex);
238 kvm_arch_ops->vcpu_load(vcpu);
242 * Switches to specified vcpu, until a matching vcpu_put(). Will return NULL
243 * if the slot is not populated.
245 static struct kvm_vcpu *vcpu_load_slot(struct kvm *kvm, int slot)
247 struct kvm_vcpu *vcpu = &kvm->vcpus[slot];
249 mutex_lock(&vcpu->mutex);
251 mutex_unlock(&vcpu->mutex);
254 kvm_arch_ops->vcpu_load(vcpu);
258 static void vcpu_put(struct kvm_vcpu *vcpu)
260 kvm_arch_ops->vcpu_put(vcpu);
261 mutex_unlock(&vcpu->mutex);
264 static void ack_flush(void *_completed)
266 atomic_t *completed = _completed;
268 atomic_inc(completed);
271 void kvm_flush_remote_tlbs(struct kvm *kvm)
275 struct kvm_vcpu *vcpu;
278 atomic_set(&completed, 0);
281 for (i = 0; i < kvm->nvcpus; ++i) {
282 vcpu = &kvm->vcpus[i];
283 if (test_and_set_bit(KVM_TLB_FLUSH, &vcpu->requests))
286 if (cpu != -1 && cpu != raw_smp_processor_id())
287 if (!cpu_isset(cpu, cpus)) {
294 * We really want smp_call_function_mask() here. But that's not
295 * available, so ipi all cpus in parallel and wait for them
298 for (cpu = first_cpu(cpus); cpu != NR_CPUS; cpu = next_cpu(cpu, cpus))
299 smp_call_function_single(cpu, ack_flush, &completed, 1, 0);
300 while (atomic_read(&completed) != needed) {
306 static struct kvm *kvm_create_vm(void)
308 struct kvm *kvm = kzalloc(sizeof(struct kvm), GFP_KERNEL);
312 return ERR_PTR(-ENOMEM);
314 kvm_io_bus_init(&kvm->pio_bus);
315 spin_lock_init(&kvm->lock);
316 INIT_LIST_HEAD(&kvm->active_mmu_pages);
317 spin_lock(&kvm_lock);
318 list_add(&kvm->vm_list, &vm_list);
319 spin_unlock(&kvm_lock);
320 kvm_io_bus_init(&kvm->mmio_bus);
321 for (i = 0; i < KVM_MAX_VCPUS; ++i) {
322 struct kvm_vcpu *vcpu = &kvm->vcpus[i];
324 mutex_init(&vcpu->mutex);
327 vcpu->mmu.root_hpa = INVALID_PAGE;
332 static int kvm_dev_open(struct inode *inode, struct file *filp)
338 * Free any memory in @free but not in @dont.
340 static void kvm_free_physmem_slot(struct kvm_memory_slot *free,
341 struct kvm_memory_slot *dont)
345 if (!dont || free->phys_mem != dont->phys_mem)
346 if (free->phys_mem) {
347 for (i = 0; i < free->npages; ++i)
348 if (free->phys_mem[i])
349 __free_page(free->phys_mem[i]);
350 vfree(free->phys_mem);
353 if (!dont || free->dirty_bitmap != dont->dirty_bitmap)
354 vfree(free->dirty_bitmap);
356 free->phys_mem = NULL;
358 free->dirty_bitmap = NULL;
361 static void kvm_free_physmem(struct kvm *kvm)
365 for (i = 0; i < kvm->nmemslots; ++i)
366 kvm_free_physmem_slot(&kvm->memslots[i], NULL);
369 static void free_pio_guest_pages(struct kvm_vcpu *vcpu)
373 for (i = 0; i < 2; ++i)
374 if (vcpu->pio.guest_pages[i]) {
375 __free_page(vcpu->pio.guest_pages[i]);
376 vcpu->pio.guest_pages[i] = NULL;
380 static void kvm_unload_vcpu_mmu(struct kvm_vcpu *vcpu)
386 kvm_mmu_unload(vcpu);
390 static void kvm_free_vcpu(struct kvm_vcpu *vcpu)
396 kvm_mmu_destroy(vcpu);
398 kvm_arch_ops->vcpu_free(vcpu);
399 free_page((unsigned long)vcpu->run);
401 free_page((unsigned long)vcpu->pio_data);
402 vcpu->pio_data = NULL;
403 free_pio_guest_pages(vcpu);
406 static void kvm_free_vcpus(struct kvm *kvm)
411 * Unpin any mmu pages first.
413 for (i = 0; i < KVM_MAX_VCPUS; ++i)
414 kvm_unload_vcpu_mmu(&kvm->vcpus[i]);
415 for (i = 0; i < KVM_MAX_VCPUS; ++i)
416 kvm_free_vcpu(&kvm->vcpus[i]);
419 static int kvm_dev_release(struct inode *inode, struct file *filp)
424 static void kvm_destroy_vm(struct kvm *kvm)
426 spin_lock(&kvm_lock);
427 list_del(&kvm->vm_list);
428 spin_unlock(&kvm_lock);
429 kvm_io_bus_destroy(&kvm->pio_bus);
430 kvm_io_bus_destroy(&kvm->mmio_bus);
432 kvm_free_physmem(kvm);
436 static int kvm_vm_release(struct inode *inode, struct file *filp)
438 struct kvm *kvm = filp->private_data;
444 static void inject_gp(struct kvm_vcpu *vcpu)
446 kvm_arch_ops->inject_gp(vcpu, 0);
450 * Load the pae pdptrs. Return true is they are all valid.
452 static int load_pdptrs(struct kvm_vcpu *vcpu, unsigned long cr3)
454 gfn_t pdpt_gfn = cr3 >> PAGE_SHIFT;
455 unsigned offset = ((cr3 & (PAGE_SIZE-1)) >> 5) << 2;
462 spin_lock(&vcpu->kvm->lock);
463 page = gfn_to_page(vcpu->kvm, pdpt_gfn);
464 /* FIXME: !page - emulate? 0xff? */
465 pdpt = kmap_atomic(page, KM_USER0);
468 for (i = 0; i < 4; ++i) {
469 pdpte = pdpt[offset + i];
470 if ((pdpte & 1) && (pdpte & 0xfffffff0000001e6ull)) {
476 for (i = 0; i < 4; ++i)
477 vcpu->pdptrs[i] = pdpt[offset + i];
480 kunmap_atomic(pdpt, KM_USER0);
481 spin_unlock(&vcpu->kvm->lock);
486 void set_cr0(struct kvm_vcpu *vcpu, unsigned long cr0)
488 if (cr0 & CR0_RESEVED_BITS) {
489 printk(KERN_DEBUG "set_cr0: 0x%lx #GP, reserved bits 0x%lx\n",
495 if ((cr0 & CR0_NW_MASK) && !(cr0 & CR0_CD_MASK)) {
496 printk(KERN_DEBUG "set_cr0: #GP, CD == 0 && NW == 1\n");
501 if ((cr0 & CR0_PG_MASK) && !(cr0 & CR0_PE_MASK)) {
502 printk(KERN_DEBUG "set_cr0: #GP, set PG flag "
503 "and a clear PE flag\n");
508 if (!is_paging(vcpu) && (cr0 & CR0_PG_MASK)) {
510 if ((vcpu->shadow_efer & EFER_LME)) {
514 printk(KERN_DEBUG "set_cr0: #GP, start paging "
515 "in long mode while PAE is disabled\n");
519 kvm_arch_ops->get_cs_db_l_bits(vcpu, &cs_db, &cs_l);
521 printk(KERN_DEBUG "set_cr0: #GP, start paging "
522 "in long mode while CS.L == 1\n");
529 if (is_pae(vcpu) && !load_pdptrs(vcpu, vcpu->cr3)) {
530 printk(KERN_DEBUG "set_cr0: #GP, pdptrs "
538 kvm_arch_ops->set_cr0(vcpu, cr0);
541 spin_lock(&vcpu->kvm->lock);
542 kvm_mmu_reset_context(vcpu);
543 spin_unlock(&vcpu->kvm->lock);
546 EXPORT_SYMBOL_GPL(set_cr0);
548 void lmsw(struct kvm_vcpu *vcpu, unsigned long msw)
550 set_cr0(vcpu, (vcpu->cr0 & ~0x0ful) | (msw & 0x0f));
552 EXPORT_SYMBOL_GPL(lmsw);
554 void set_cr4(struct kvm_vcpu *vcpu, unsigned long cr4)
556 if (cr4 & CR4_RESEVED_BITS) {
557 printk(KERN_DEBUG "set_cr4: #GP, reserved bits\n");
562 if (is_long_mode(vcpu)) {
563 if (!(cr4 & CR4_PAE_MASK)) {
564 printk(KERN_DEBUG "set_cr4: #GP, clearing PAE while "
569 } else if (is_paging(vcpu) && !is_pae(vcpu) && (cr4 & CR4_PAE_MASK)
570 && !load_pdptrs(vcpu, vcpu->cr3)) {
571 printk(KERN_DEBUG "set_cr4: #GP, pdptrs reserved bits\n");
575 if (cr4 & CR4_VMXE_MASK) {
576 printk(KERN_DEBUG "set_cr4: #GP, setting VMXE\n");
580 kvm_arch_ops->set_cr4(vcpu, cr4);
581 spin_lock(&vcpu->kvm->lock);
582 kvm_mmu_reset_context(vcpu);
583 spin_unlock(&vcpu->kvm->lock);
585 EXPORT_SYMBOL_GPL(set_cr4);
587 void set_cr3(struct kvm_vcpu *vcpu, unsigned long cr3)
589 if (is_long_mode(vcpu)) {
590 if (cr3 & CR3_L_MODE_RESEVED_BITS) {
591 printk(KERN_DEBUG "set_cr3: #GP, reserved bits\n");
596 if (cr3 & CR3_RESEVED_BITS) {
597 printk(KERN_DEBUG "set_cr3: #GP, reserved bits\n");
601 if (is_paging(vcpu) && is_pae(vcpu) &&
602 !load_pdptrs(vcpu, cr3)) {
603 printk(KERN_DEBUG "set_cr3: #GP, pdptrs "
611 spin_lock(&vcpu->kvm->lock);
613 * Does the new cr3 value map to physical memory? (Note, we
614 * catch an invalid cr3 even in real-mode, because it would
615 * cause trouble later on when we turn on paging anyway.)
617 * A real CPU would silently accept an invalid cr3 and would
618 * attempt to use it - with largely undefined (and often hard
619 * to debug) behavior on the guest side.
621 if (unlikely(!gfn_to_memslot(vcpu->kvm, cr3 >> PAGE_SHIFT)))
624 vcpu->mmu.new_cr3(vcpu);
625 spin_unlock(&vcpu->kvm->lock);
627 EXPORT_SYMBOL_GPL(set_cr3);
629 void set_cr8(struct kvm_vcpu *vcpu, unsigned long cr8)
631 if ( cr8 & CR8_RESEVED_BITS) {
632 printk(KERN_DEBUG "set_cr8: #GP, reserved bits 0x%lx\n", cr8);
638 EXPORT_SYMBOL_GPL(set_cr8);
640 void fx_init(struct kvm_vcpu *vcpu)
642 struct __attribute__ ((__packed__)) fx_image_s {
648 u64 operand;// fpu dp
654 fx_save(vcpu->host_fx_image);
656 fx_save(vcpu->guest_fx_image);
657 fx_restore(vcpu->host_fx_image);
659 fx_image = (struct fx_image_s *)vcpu->guest_fx_image;
660 fx_image->mxcsr = 0x1f80;
661 memset(vcpu->guest_fx_image + sizeof(struct fx_image_s),
662 0, FX_IMAGE_SIZE - sizeof(struct fx_image_s));
664 EXPORT_SYMBOL_GPL(fx_init);
666 static void do_remove_write_access(struct kvm_vcpu *vcpu, int slot)
668 spin_lock(&vcpu->kvm->lock);
669 kvm_mmu_slot_remove_write_access(vcpu, slot);
670 spin_unlock(&vcpu->kvm->lock);
674 * Allocate some memory and give it an address in the guest physical address
677 * Discontiguous memory is allowed, mostly for framebuffers.
679 static int kvm_vm_ioctl_set_memory_region(struct kvm *kvm,
680 struct kvm_memory_region *mem)
684 unsigned long npages;
686 struct kvm_memory_slot *memslot;
687 struct kvm_memory_slot old, new;
688 int memory_config_version;
691 /* General sanity checks */
692 if (mem->memory_size & (PAGE_SIZE - 1))
694 if (mem->guest_phys_addr & (PAGE_SIZE - 1))
696 if (mem->slot >= KVM_MEMORY_SLOTS)
698 if (mem->guest_phys_addr + mem->memory_size < mem->guest_phys_addr)
701 memslot = &kvm->memslots[mem->slot];
702 base_gfn = mem->guest_phys_addr >> PAGE_SHIFT;
703 npages = mem->memory_size >> PAGE_SHIFT;
706 mem->flags &= ~KVM_MEM_LOG_DIRTY_PAGES;
709 spin_lock(&kvm->lock);
711 memory_config_version = kvm->memory_config_version;
712 new = old = *memslot;
714 new.base_gfn = base_gfn;
716 new.flags = mem->flags;
718 /* Disallow changing a memory slot's size. */
720 if (npages && old.npages && npages != old.npages)
723 /* Check for overlaps */
725 for (i = 0; i < KVM_MEMORY_SLOTS; ++i) {
726 struct kvm_memory_slot *s = &kvm->memslots[i];
730 if (!((base_gfn + npages <= s->base_gfn) ||
731 (base_gfn >= s->base_gfn + s->npages)))
735 * Do memory allocations outside lock. memory_config_version will
738 spin_unlock(&kvm->lock);
740 /* Deallocate if slot is being removed */
744 /* Free page dirty bitmap if unneeded */
745 if (!(new.flags & KVM_MEM_LOG_DIRTY_PAGES))
746 new.dirty_bitmap = NULL;
750 /* Allocate if a slot is being created */
751 if (npages && !new.phys_mem) {
752 new.phys_mem = vmalloc(npages * sizeof(struct page *));
757 memset(new.phys_mem, 0, npages * sizeof(struct page *));
758 for (i = 0; i < npages; ++i) {
759 new.phys_mem[i] = alloc_page(GFP_HIGHUSER
761 if (!new.phys_mem[i])
763 set_page_private(new.phys_mem[i],0);
767 /* Allocate page dirty bitmap if needed */
768 if ((new.flags & KVM_MEM_LOG_DIRTY_PAGES) && !new.dirty_bitmap) {
769 unsigned dirty_bytes = ALIGN(npages, BITS_PER_LONG) / 8;
771 new.dirty_bitmap = vmalloc(dirty_bytes);
772 if (!new.dirty_bitmap)
774 memset(new.dirty_bitmap, 0, dirty_bytes);
777 spin_lock(&kvm->lock);
779 if (memory_config_version != kvm->memory_config_version) {
780 spin_unlock(&kvm->lock);
781 kvm_free_physmem_slot(&new, &old);
789 if (mem->slot >= kvm->nmemslots)
790 kvm->nmemslots = mem->slot + 1;
793 ++kvm->memory_config_version;
795 spin_unlock(&kvm->lock);
797 for (i = 0; i < KVM_MAX_VCPUS; ++i) {
798 struct kvm_vcpu *vcpu;
800 vcpu = vcpu_load_slot(kvm, i);
803 if (new.flags & KVM_MEM_LOG_DIRTY_PAGES)
804 do_remove_write_access(vcpu, mem->slot);
805 kvm_mmu_reset_context(vcpu);
809 kvm_free_physmem_slot(&old, &new);
813 spin_unlock(&kvm->lock);
815 kvm_free_physmem_slot(&new, &old);
821 * Get (and clear) the dirty memory log for a memory slot.
823 static int kvm_vm_ioctl_get_dirty_log(struct kvm *kvm,
824 struct kvm_dirty_log *log)
826 struct kvm_memory_slot *memslot;
830 unsigned long any = 0;
832 spin_lock(&kvm->lock);
835 * Prevent changes to guest memory configuration even while the lock
839 spin_unlock(&kvm->lock);
841 if (log->slot >= KVM_MEMORY_SLOTS)
844 memslot = &kvm->memslots[log->slot];
846 if (!memslot->dirty_bitmap)
849 n = ALIGN(memslot->npages, BITS_PER_LONG) / 8;
851 for (i = 0; !any && i < n/sizeof(long); ++i)
852 any = memslot->dirty_bitmap[i];
855 if (copy_to_user(log->dirty_bitmap, memslot->dirty_bitmap, n))
860 for (i = 0; i < KVM_MAX_VCPUS; ++i) {
861 struct kvm_vcpu *vcpu;
863 vcpu = vcpu_load_slot(kvm, i);
867 do_remove_write_access(vcpu, log->slot);
868 memset(memslot->dirty_bitmap, 0, n);
871 kvm_arch_ops->tlb_flush(vcpu);
879 spin_lock(&kvm->lock);
881 spin_unlock(&kvm->lock);
886 * Set a new alias region. Aliases map a portion of physical memory into
887 * another portion. This is useful for memory windows, for example the PC
890 static int kvm_vm_ioctl_set_memory_alias(struct kvm *kvm,
891 struct kvm_memory_alias *alias)
894 struct kvm_mem_alias *p;
897 /* General sanity checks */
898 if (alias->memory_size & (PAGE_SIZE - 1))
900 if (alias->guest_phys_addr & (PAGE_SIZE - 1))
902 if (alias->slot >= KVM_ALIAS_SLOTS)
904 if (alias->guest_phys_addr + alias->memory_size
905 < alias->guest_phys_addr)
907 if (alias->target_phys_addr + alias->memory_size
908 < alias->target_phys_addr)
911 spin_lock(&kvm->lock);
913 p = &kvm->aliases[alias->slot];
914 p->base_gfn = alias->guest_phys_addr >> PAGE_SHIFT;
915 p->npages = alias->memory_size >> PAGE_SHIFT;
916 p->target_gfn = alias->target_phys_addr >> PAGE_SHIFT;
918 for (n = KVM_ALIAS_SLOTS; n > 0; --n)
919 if (kvm->aliases[n - 1].npages)
923 spin_unlock(&kvm->lock);
925 vcpu_load(&kvm->vcpus[0]);
926 spin_lock(&kvm->lock);
927 kvm_mmu_zap_all(&kvm->vcpus[0]);
928 spin_unlock(&kvm->lock);
929 vcpu_put(&kvm->vcpus[0]);
937 static gfn_t unalias_gfn(struct kvm *kvm, gfn_t gfn)
940 struct kvm_mem_alias *alias;
942 for (i = 0; i < kvm->naliases; ++i) {
943 alias = &kvm->aliases[i];
944 if (gfn >= alias->base_gfn
945 && gfn < alias->base_gfn + alias->npages)
946 return alias->target_gfn + gfn - alias->base_gfn;
951 static struct kvm_memory_slot *__gfn_to_memslot(struct kvm *kvm, gfn_t gfn)
955 for (i = 0; i < kvm->nmemslots; ++i) {
956 struct kvm_memory_slot *memslot = &kvm->memslots[i];
958 if (gfn >= memslot->base_gfn
959 && gfn < memslot->base_gfn + memslot->npages)
965 struct kvm_memory_slot *gfn_to_memslot(struct kvm *kvm, gfn_t gfn)
967 gfn = unalias_gfn(kvm, gfn);
968 return __gfn_to_memslot(kvm, gfn);
971 struct page *gfn_to_page(struct kvm *kvm, gfn_t gfn)
973 struct kvm_memory_slot *slot;
975 gfn = unalias_gfn(kvm, gfn);
976 slot = __gfn_to_memslot(kvm, gfn);
979 return slot->phys_mem[gfn - slot->base_gfn];
981 EXPORT_SYMBOL_GPL(gfn_to_page);
983 void mark_page_dirty(struct kvm *kvm, gfn_t gfn)
986 struct kvm_memory_slot *memslot;
987 unsigned long rel_gfn;
989 for (i = 0; i < kvm->nmemslots; ++i) {
990 memslot = &kvm->memslots[i];
992 if (gfn >= memslot->base_gfn
993 && gfn < memslot->base_gfn + memslot->npages) {
995 if (!memslot->dirty_bitmap)
998 rel_gfn = gfn - memslot->base_gfn;
1001 if (!test_bit(rel_gfn, memslot->dirty_bitmap))
1002 set_bit(rel_gfn, memslot->dirty_bitmap);
1008 static int emulator_read_std(unsigned long addr,
1011 struct x86_emulate_ctxt *ctxt)
1013 struct kvm_vcpu *vcpu = ctxt->vcpu;
1017 gpa_t gpa = vcpu->mmu.gva_to_gpa(vcpu, addr);
1018 unsigned offset = addr & (PAGE_SIZE-1);
1019 unsigned tocopy = min(bytes, (unsigned)PAGE_SIZE - offset);
1024 if (gpa == UNMAPPED_GVA)
1025 return X86EMUL_PROPAGATE_FAULT;
1026 pfn = gpa >> PAGE_SHIFT;
1027 page = gfn_to_page(vcpu->kvm, pfn);
1029 return X86EMUL_UNHANDLEABLE;
1030 page_virt = kmap_atomic(page, KM_USER0);
1032 memcpy(data, page_virt + offset, tocopy);
1034 kunmap_atomic(page_virt, KM_USER0);
1041 return X86EMUL_CONTINUE;
1044 static int emulator_write_std(unsigned long addr,
1047 struct x86_emulate_ctxt *ctxt)
1049 printk(KERN_ERR "emulator_write_std: addr %lx n %d\n",
1051 return X86EMUL_UNHANDLEABLE;
1054 static struct kvm_io_device *vcpu_find_mmio_dev(struct kvm_vcpu *vcpu,
1058 * Note that its important to have this wrapper function because
1059 * in the very near future we will be checking for MMIOs against
1060 * the LAPIC as well as the general MMIO bus
1062 return kvm_io_bus_find_dev(&vcpu->kvm->mmio_bus, addr);
1065 static struct kvm_io_device *vcpu_find_pio_dev(struct kvm_vcpu *vcpu,
1068 return kvm_io_bus_find_dev(&vcpu->kvm->pio_bus, addr);
1071 static int emulator_read_emulated(unsigned long addr,
1074 struct x86_emulate_ctxt *ctxt)
1076 struct kvm_vcpu *vcpu = ctxt->vcpu;
1077 struct kvm_io_device *mmio_dev;
1080 if (vcpu->mmio_read_completed) {
1081 memcpy(val, vcpu->mmio_data, bytes);
1082 vcpu->mmio_read_completed = 0;
1083 return X86EMUL_CONTINUE;
1084 } else if (emulator_read_std(addr, val, bytes, ctxt)
1085 == X86EMUL_CONTINUE)
1086 return X86EMUL_CONTINUE;
1088 gpa = vcpu->mmu.gva_to_gpa(vcpu, addr);
1089 if (gpa == UNMAPPED_GVA)
1090 return X86EMUL_PROPAGATE_FAULT;
1093 * Is this MMIO handled locally?
1095 mmio_dev = vcpu_find_mmio_dev(vcpu, gpa);
1097 kvm_iodevice_read(mmio_dev, gpa, bytes, val);
1098 return X86EMUL_CONTINUE;
1101 vcpu->mmio_needed = 1;
1102 vcpu->mmio_phys_addr = gpa;
1103 vcpu->mmio_size = bytes;
1104 vcpu->mmio_is_write = 0;
1106 return X86EMUL_UNHANDLEABLE;
1109 static int emulator_write_phys(struct kvm_vcpu *vcpu, gpa_t gpa,
1110 const void *val, int bytes)
1114 unsigned offset = offset_in_page(gpa);
1116 if (((gpa + bytes - 1) >> PAGE_SHIFT) != (gpa >> PAGE_SHIFT))
1118 page = gfn_to_page(vcpu->kvm, gpa >> PAGE_SHIFT);
1121 mark_page_dirty(vcpu->kvm, gpa >> PAGE_SHIFT);
1122 virt = kmap_atomic(page, KM_USER0);
1123 if (memcmp(virt + offset_in_page(gpa), val, bytes)) {
1124 kvm_mmu_pte_write(vcpu, gpa, virt + offset, val, bytes);
1125 memcpy(virt + offset_in_page(gpa), val, bytes);
1127 kunmap_atomic(virt, KM_USER0);
1131 static int emulator_write_emulated(unsigned long addr,
1134 struct x86_emulate_ctxt *ctxt)
1136 struct kvm_vcpu *vcpu = ctxt->vcpu;
1137 struct kvm_io_device *mmio_dev;
1138 gpa_t gpa = vcpu->mmu.gva_to_gpa(vcpu, addr);
1140 if (gpa == UNMAPPED_GVA) {
1141 kvm_arch_ops->inject_page_fault(vcpu, addr, 2);
1142 return X86EMUL_PROPAGATE_FAULT;
1145 if (emulator_write_phys(vcpu, gpa, val, bytes))
1146 return X86EMUL_CONTINUE;
1149 * Is this MMIO handled locally?
1151 mmio_dev = vcpu_find_mmio_dev(vcpu, gpa);
1153 kvm_iodevice_write(mmio_dev, gpa, bytes, val);
1154 return X86EMUL_CONTINUE;
1157 vcpu->mmio_needed = 1;
1158 vcpu->mmio_phys_addr = gpa;
1159 vcpu->mmio_size = bytes;
1160 vcpu->mmio_is_write = 1;
1161 memcpy(vcpu->mmio_data, val, bytes);
1163 return X86EMUL_CONTINUE;
1166 static int emulator_cmpxchg_emulated(unsigned long addr,
1170 struct x86_emulate_ctxt *ctxt)
1172 static int reported;
1176 printk(KERN_WARNING "kvm: emulating exchange as write\n");
1178 return emulator_write_emulated(addr, new, bytes, ctxt);
1181 static unsigned long get_segment_base(struct kvm_vcpu *vcpu, int seg)
1183 return kvm_arch_ops->get_segment_base(vcpu, seg);
1186 int emulate_invlpg(struct kvm_vcpu *vcpu, gva_t address)
1188 return X86EMUL_CONTINUE;
1191 int emulate_clts(struct kvm_vcpu *vcpu)
1195 cr0 = vcpu->cr0 & ~CR0_TS_MASK;
1196 kvm_arch_ops->set_cr0(vcpu, cr0);
1197 return X86EMUL_CONTINUE;
1200 int emulator_get_dr(struct x86_emulate_ctxt* ctxt, int dr, unsigned long *dest)
1202 struct kvm_vcpu *vcpu = ctxt->vcpu;
1206 *dest = kvm_arch_ops->get_dr(vcpu, dr);
1207 return X86EMUL_CONTINUE;
1209 printk(KERN_DEBUG "%s: unexpected dr %u\n",
1211 return X86EMUL_UNHANDLEABLE;
1215 int emulator_set_dr(struct x86_emulate_ctxt *ctxt, int dr, unsigned long value)
1217 unsigned long mask = (ctxt->mode == X86EMUL_MODE_PROT64) ? ~0ULL : ~0U;
1220 kvm_arch_ops->set_dr(ctxt->vcpu, dr, value & mask, &exception);
1222 /* FIXME: better handling */
1223 return X86EMUL_UNHANDLEABLE;
1225 return X86EMUL_CONTINUE;
1228 static void report_emulation_failure(struct x86_emulate_ctxt *ctxt)
1230 static int reported;
1232 unsigned long rip = ctxt->vcpu->rip;
1233 unsigned long rip_linear;
1235 rip_linear = rip + get_segment_base(ctxt->vcpu, VCPU_SREG_CS);
1240 emulator_read_std(rip_linear, (void *)opcodes, 4, ctxt);
1242 printk(KERN_ERR "emulation failed but !mmio_needed?"
1243 " rip %lx %02x %02x %02x %02x\n",
1244 rip, opcodes[0], opcodes[1], opcodes[2], opcodes[3]);
1248 struct x86_emulate_ops emulate_ops = {
1249 .read_std = emulator_read_std,
1250 .write_std = emulator_write_std,
1251 .read_emulated = emulator_read_emulated,
1252 .write_emulated = emulator_write_emulated,
1253 .cmpxchg_emulated = emulator_cmpxchg_emulated,
1256 int emulate_instruction(struct kvm_vcpu *vcpu,
1257 struct kvm_run *run,
1261 struct x86_emulate_ctxt emulate_ctxt;
1265 vcpu->mmio_fault_cr2 = cr2;
1266 kvm_arch_ops->cache_regs(vcpu);
1268 kvm_arch_ops->get_cs_db_l_bits(vcpu, &cs_db, &cs_l);
1270 emulate_ctxt.vcpu = vcpu;
1271 emulate_ctxt.eflags = kvm_arch_ops->get_rflags(vcpu);
1272 emulate_ctxt.cr2 = cr2;
1273 emulate_ctxt.mode = (emulate_ctxt.eflags & X86_EFLAGS_VM)
1274 ? X86EMUL_MODE_REAL : cs_l
1275 ? X86EMUL_MODE_PROT64 : cs_db
1276 ? X86EMUL_MODE_PROT32 : X86EMUL_MODE_PROT16;
1278 if (emulate_ctxt.mode == X86EMUL_MODE_PROT64) {
1279 emulate_ctxt.cs_base = 0;
1280 emulate_ctxt.ds_base = 0;
1281 emulate_ctxt.es_base = 0;
1282 emulate_ctxt.ss_base = 0;
1284 emulate_ctxt.cs_base = get_segment_base(vcpu, VCPU_SREG_CS);
1285 emulate_ctxt.ds_base = get_segment_base(vcpu, VCPU_SREG_DS);
1286 emulate_ctxt.es_base = get_segment_base(vcpu, VCPU_SREG_ES);
1287 emulate_ctxt.ss_base = get_segment_base(vcpu, VCPU_SREG_SS);
1290 emulate_ctxt.gs_base = get_segment_base(vcpu, VCPU_SREG_GS);
1291 emulate_ctxt.fs_base = get_segment_base(vcpu, VCPU_SREG_FS);
1293 vcpu->mmio_is_write = 0;
1294 r = x86_emulate_memop(&emulate_ctxt, &emulate_ops);
1296 if ((r || vcpu->mmio_is_write) && run) {
1297 run->mmio.phys_addr = vcpu->mmio_phys_addr;
1298 memcpy(run->mmio.data, vcpu->mmio_data, 8);
1299 run->mmio.len = vcpu->mmio_size;
1300 run->mmio.is_write = vcpu->mmio_is_write;
1304 if (kvm_mmu_unprotect_page_virt(vcpu, cr2))
1305 return EMULATE_DONE;
1306 if (!vcpu->mmio_needed) {
1307 report_emulation_failure(&emulate_ctxt);
1308 return EMULATE_FAIL;
1310 return EMULATE_DO_MMIO;
1313 kvm_arch_ops->decache_regs(vcpu);
1314 kvm_arch_ops->set_rflags(vcpu, emulate_ctxt.eflags);
1316 if (vcpu->mmio_is_write) {
1317 vcpu->mmio_needed = 0;
1318 return EMULATE_DO_MMIO;
1321 return EMULATE_DONE;
1323 EXPORT_SYMBOL_GPL(emulate_instruction);
1325 int kvm_emulate_halt(struct kvm_vcpu *vcpu)
1327 if (vcpu->irq_summary)
1330 vcpu->run->exit_reason = KVM_EXIT_HLT;
1331 ++vcpu->stat.halt_exits;
1334 EXPORT_SYMBOL_GPL(kvm_emulate_halt);
1336 int kvm_hypercall(struct kvm_vcpu *vcpu, struct kvm_run *run)
1338 unsigned long nr, a0, a1, a2, a3, a4, a5, ret;
1340 kvm_arch_ops->cache_regs(vcpu);
1342 #ifdef CONFIG_X86_64
1343 if (is_long_mode(vcpu)) {
1344 nr = vcpu->regs[VCPU_REGS_RAX];
1345 a0 = vcpu->regs[VCPU_REGS_RDI];
1346 a1 = vcpu->regs[VCPU_REGS_RSI];
1347 a2 = vcpu->regs[VCPU_REGS_RDX];
1348 a3 = vcpu->regs[VCPU_REGS_RCX];
1349 a4 = vcpu->regs[VCPU_REGS_R8];
1350 a5 = vcpu->regs[VCPU_REGS_R9];
1354 nr = vcpu->regs[VCPU_REGS_RBX] & -1u;
1355 a0 = vcpu->regs[VCPU_REGS_RAX] & -1u;
1356 a1 = vcpu->regs[VCPU_REGS_RCX] & -1u;
1357 a2 = vcpu->regs[VCPU_REGS_RDX] & -1u;
1358 a3 = vcpu->regs[VCPU_REGS_RSI] & -1u;
1359 a4 = vcpu->regs[VCPU_REGS_RDI] & -1u;
1360 a5 = vcpu->regs[VCPU_REGS_RBP] & -1u;
1364 run->hypercall.args[0] = a0;
1365 run->hypercall.args[1] = a1;
1366 run->hypercall.args[2] = a2;
1367 run->hypercall.args[3] = a3;
1368 run->hypercall.args[4] = a4;
1369 run->hypercall.args[5] = a5;
1370 run->hypercall.ret = ret;
1371 run->hypercall.longmode = is_long_mode(vcpu);
1372 kvm_arch_ops->decache_regs(vcpu);
1375 vcpu->regs[VCPU_REGS_RAX] = ret;
1376 kvm_arch_ops->decache_regs(vcpu);
1379 EXPORT_SYMBOL_GPL(kvm_hypercall);
1381 static u64 mk_cr_64(u64 curr_cr, u32 new_val)
1383 return (curr_cr & ~((1ULL << 32) - 1)) | new_val;
1386 void realmode_lgdt(struct kvm_vcpu *vcpu, u16 limit, unsigned long base)
1388 struct descriptor_table dt = { limit, base };
1390 kvm_arch_ops->set_gdt(vcpu, &dt);
1393 void realmode_lidt(struct kvm_vcpu *vcpu, u16 limit, unsigned long base)
1395 struct descriptor_table dt = { limit, base };
1397 kvm_arch_ops->set_idt(vcpu, &dt);
1400 void realmode_lmsw(struct kvm_vcpu *vcpu, unsigned long msw,
1401 unsigned long *rflags)
1404 *rflags = kvm_arch_ops->get_rflags(vcpu);
1407 unsigned long realmode_get_cr(struct kvm_vcpu *vcpu, int cr)
1409 kvm_arch_ops->decache_cr4_guest_bits(vcpu);
1420 vcpu_printf(vcpu, "%s: unexpected cr %u\n", __FUNCTION__, cr);
1425 void realmode_set_cr(struct kvm_vcpu *vcpu, int cr, unsigned long val,
1426 unsigned long *rflags)
1430 set_cr0(vcpu, mk_cr_64(vcpu->cr0, val));
1431 *rflags = kvm_arch_ops->get_rflags(vcpu);
1440 set_cr4(vcpu, mk_cr_64(vcpu->cr4, val));
1443 vcpu_printf(vcpu, "%s: unexpected cr %u\n", __FUNCTION__, cr);
1448 * Register the para guest with the host:
1450 static int vcpu_register_para(struct kvm_vcpu *vcpu, gpa_t para_state_gpa)
1452 struct kvm_vcpu_para_state *para_state;
1453 hpa_t para_state_hpa, hypercall_hpa;
1454 struct page *para_state_page;
1455 unsigned char *hypercall;
1456 gpa_t hypercall_gpa;
1458 printk(KERN_DEBUG "kvm: guest trying to enter paravirtual mode\n");
1459 printk(KERN_DEBUG ".... para_state_gpa: %08Lx\n", para_state_gpa);
1462 * Needs to be page aligned:
1464 if (para_state_gpa != PAGE_ALIGN(para_state_gpa))
1467 para_state_hpa = gpa_to_hpa(vcpu, para_state_gpa);
1468 printk(KERN_DEBUG ".... para_state_hpa: %08Lx\n", para_state_hpa);
1469 if (is_error_hpa(para_state_hpa))
1472 mark_page_dirty(vcpu->kvm, para_state_gpa >> PAGE_SHIFT);
1473 para_state_page = pfn_to_page(para_state_hpa >> PAGE_SHIFT);
1474 para_state = kmap_atomic(para_state_page, KM_USER0);
1476 printk(KERN_DEBUG ".... guest version: %d\n", para_state->guest_version);
1477 printk(KERN_DEBUG ".... size: %d\n", para_state->size);
1479 para_state->host_version = KVM_PARA_API_VERSION;
1481 * We cannot support guests that try to register themselves
1482 * with a newer API version than the host supports:
1484 if (para_state->guest_version > KVM_PARA_API_VERSION) {
1485 para_state->ret = -KVM_EINVAL;
1486 goto err_kunmap_skip;
1489 hypercall_gpa = para_state->hypercall_gpa;
1490 hypercall_hpa = gpa_to_hpa(vcpu, hypercall_gpa);
1491 printk(KERN_DEBUG ".... hypercall_hpa: %08Lx\n", hypercall_hpa);
1492 if (is_error_hpa(hypercall_hpa)) {
1493 para_state->ret = -KVM_EINVAL;
1494 goto err_kunmap_skip;
1497 printk(KERN_DEBUG "kvm: para guest successfully registered.\n");
1498 vcpu->para_state_page = para_state_page;
1499 vcpu->para_state_gpa = para_state_gpa;
1500 vcpu->hypercall_gpa = hypercall_gpa;
1502 mark_page_dirty(vcpu->kvm, hypercall_gpa >> PAGE_SHIFT);
1503 hypercall = kmap_atomic(pfn_to_page(hypercall_hpa >> PAGE_SHIFT),
1504 KM_USER1) + (hypercall_hpa & ~PAGE_MASK);
1505 kvm_arch_ops->patch_hypercall(vcpu, hypercall);
1506 kunmap_atomic(hypercall, KM_USER1);
1508 para_state->ret = 0;
1510 kunmap_atomic(para_state, KM_USER0);
1516 int kvm_get_msr_common(struct kvm_vcpu *vcpu, u32 msr, u64 *pdata)
1521 case 0xc0010010: /* SYSCFG */
1522 case 0xc0010015: /* HWCR */
1523 case MSR_IA32_PLATFORM_ID:
1524 case MSR_IA32_P5_MC_ADDR:
1525 case MSR_IA32_P5_MC_TYPE:
1526 case MSR_IA32_MC0_CTL:
1527 case MSR_IA32_MCG_STATUS:
1528 case MSR_IA32_MCG_CAP:
1529 case MSR_IA32_MC0_MISC:
1530 case MSR_IA32_MC0_MISC+4:
1531 case MSR_IA32_MC0_MISC+8:
1532 case MSR_IA32_MC0_MISC+12:
1533 case MSR_IA32_MC0_MISC+16:
1534 case MSR_IA32_UCODE_REV:
1535 case MSR_IA32_PERF_STATUS:
1536 case MSR_IA32_EBL_CR_POWERON:
1537 /* MTRR registers */
1539 case 0x200 ... 0x2ff:
1542 case 0xcd: /* fsb frequency */
1545 case MSR_IA32_APICBASE:
1546 data = vcpu->apic_base;
1548 case MSR_IA32_MISC_ENABLE:
1549 data = vcpu->ia32_misc_enable_msr;
1551 #ifdef CONFIG_X86_64
1553 data = vcpu->shadow_efer;
1557 printk(KERN_ERR "kvm: unhandled rdmsr: 0x%x\n", msr);
1563 EXPORT_SYMBOL_GPL(kvm_get_msr_common);
1566 * Reads an msr value (of 'msr_index') into 'pdata'.
1567 * Returns 0 on success, non-0 otherwise.
1568 * Assumes vcpu_load() was already called.
1570 static int get_msr(struct kvm_vcpu *vcpu, u32 msr_index, u64 *pdata)
1572 return kvm_arch_ops->get_msr(vcpu, msr_index, pdata);
1575 #ifdef CONFIG_X86_64
1577 static void set_efer(struct kvm_vcpu *vcpu, u64 efer)
1579 if (efer & EFER_RESERVED_BITS) {
1580 printk(KERN_DEBUG "set_efer: 0x%llx #GP, reserved bits\n",
1587 && (vcpu->shadow_efer & EFER_LME) != (efer & EFER_LME)) {
1588 printk(KERN_DEBUG "set_efer: #GP, change LME while paging\n");
1593 kvm_arch_ops->set_efer(vcpu, efer);
1596 efer |= vcpu->shadow_efer & EFER_LMA;
1598 vcpu->shadow_efer = efer;
1603 int kvm_set_msr_common(struct kvm_vcpu *vcpu, u32 msr, u64 data)
1606 #ifdef CONFIG_X86_64
1608 set_efer(vcpu, data);
1611 case MSR_IA32_MC0_STATUS:
1612 printk(KERN_WARNING "%s: MSR_IA32_MC0_STATUS 0x%llx, nop\n",
1613 __FUNCTION__, data);
1615 case MSR_IA32_MCG_STATUS:
1616 printk(KERN_WARNING "%s: MSR_IA32_MCG_STATUS 0x%llx, nop\n",
1617 __FUNCTION__, data);
1619 case MSR_IA32_UCODE_REV:
1620 case MSR_IA32_UCODE_WRITE:
1621 case 0x200 ... 0x2ff: /* MTRRs */
1623 case MSR_IA32_APICBASE:
1624 vcpu->apic_base = data;
1626 case MSR_IA32_MISC_ENABLE:
1627 vcpu->ia32_misc_enable_msr = data;
1630 * This is the 'probe whether the host is KVM' logic:
1632 case MSR_KVM_API_MAGIC:
1633 return vcpu_register_para(vcpu, data);
1636 printk(KERN_ERR "kvm: unhandled wrmsr: 0x%x\n", msr);
1641 EXPORT_SYMBOL_GPL(kvm_set_msr_common);
1644 * Writes msr value into into the appropriate "register".
1645 * Returns 0 on success, non-0 otherwise.
1646 * Assumes vcpu_load() was already called.
1648 static int set_msr(struct kvm_vcpu *vcpu, u32 msr_index, u64 data)
1650 return kvm_arch_ops->set_msr(vcpu, msr_index, data);
1653 void kvm_resched(struct kvm_vcpu *vcpu)
1655 if (!need_resched())
1661 EXPORT_SYMBOL_GPL(kvm_resched);
1663 void load_msrs(struct vmx_msr_entry *e, int n)
1667 for (i = 0; i < n; ++i)
1668 wrmsrl(e[i].index, e[i].data);
1670 EXPORT_SYMBOL_GPL(load_msrs);
1672 void save_msrs(struct vmx_msr_entry *e, int n)
1676 for (i = 0; i < n; ++i)
1677 rdmsrl(e[i].index, e[i].data);
1679 EXPORT_SYMBOL_GPL(save_msrs);
1681 void kvm_emulate_cpuid(struct kvm_vcpu *vcpu)
1685 struct kvm_cpuid_entry *e, *best;
1687 kvm_arch_ops->cache_regs(vcpu);
1688 function = vcpu->regs[VCPU_REGS_RAX];
1689 vcpu->regs[VCPU_REGS_RAX] = 0;
1690 vcpu->regs[VCPU_REGS_RBX] = 0;
1691 vcpu->regs[VCPU_REGS_RCX] = 0;
1692 vcpu->regs[VCPU_REGS_RDX] = 0;
1694 for (i = 0; i < vcpu->cpuid_nent; ++i) {
1695 e = &vcpu->cpuid_entries[i];
1696 if (e->function == function) {
1701 * Both basic or both extended?
1703 if (((e->function ^ function) & 0x80000000) == 0)
1704 if (!best || e->function > best->function)
1708 vcpu->regs[VCPU_REGS_RAX] = best->eax;
1709 vcpu->regs[VCPU_REGS_RBX] = best->ebx;
1710 vcpu->regs[VCPU_REGS_RCX] = best->ecx;
1711 vcpu->regs[VCPU_REGS_RDX] = best->edx;
1713 kvm_arch_ops->decache_regs(vcpu);
1714 kvm_arch_ops->skip_emulated_instruction(vcpu);
1716 EXPORT_SYMBOL_GPL(kvm_emulate_cpuid);
1718 static int pio_copy_data(struct kvm_vcpu *vcpu)
1720 void *p = vcpu->pio_data;
1723 int nr_pages = vcpu->pio.guest_pages[1] ? 2 : 1;
1725 kvm_arch_ops->vcpu_put(vcpu);
1726 q = vmap(vcpu->pio.guest_pages, nr_pages, VM_READ|VM_WRITE,
1729 kvm_arch_ops->vcpu_load(vcpu);
1730 free_pio_guest_pages(vcpu);
1733 q += vcpu->pio.guest_page_offset;
1734 bytes = vcpu->pio.size * vcpu->pio.cur_count;
1736 memcpy(q, p, bytes);
1738 memcpy(p, q, bytes);
1739 q -= vcpu->pio.guest_page_offset;
1741 kvm_arch_ops->vcpu_load(vcpu);
1742 free_pio_guest_pages(vcpu);
1746 static int complete_pio(struct kvm_vcpu *vcpu)
1748 struct kvm_pio_request *io = &vcpu->pio;
1752 kvm_arch_ops->cache_regs(vcpu);
1756 memcpy(&vcpu->regs[VCPU_REGS_RAX], vcpu->pio_data,
1760 r = pio_copy_data(vcpu);
1762 kvm_arch_ops->cache_regs(vcpu);
1769 delta *= io->cur_count;
1771 * The size of the register should really depend on
1772 * current address size.
1774 vcpu->regs[VCPU_REGS_RCX] -= delta;
1780 vcpu->regs[VCPU_REGS_RDI] += delta;
1782 vcpu->regs[VCPU_REGS_RSI] += delta;
1785 kvm_arch_ops->decache_regs(vcpu);
1787 io->count -= io->cur_count;
1791 kvm_arch_ops->skip_emulated_instruction(vcpu);
1795 void kernel_pio(struct kvm_io_device *pio_dev, struct kvm_vcpu *vcpu)
1797 /* TODO: String I/O for in kernel device */
1800 kvm_iodevice_read(pio_dev, vcpu->pio.port,
1804 kvm_iodevice_write(pio_dev, vcpu->pio.port,
1809 int kvm_setup_pio(struct kvm_vcpu *vcpu, struct kvm_run *run, int in,
1810 int size, unsigned long count, int string, int down,
1811 gva_t address, int rep, unsigned port)
1813 unsigned now, in_page;
1817 struct kvm_io_device *pio_dev;
1819 vcpu->run->exit_reason = KVM_EXIT_IO;
1820 vcpu->run->io.direction = in ? KVM_EXIT_IO_IN : KVM_EXIT_IO_OUT;
1821 vcpu->run->io.size = size;
1822 vcpu->run->io.data_offset = KVM_PIO_PAGE_OFFSET * PAGE_SIZE;
1823 vcpu->run->io.count = count;
1824 vcpu->run->io.port = port;
1825 vcpu->pio.count = count;
1826 vcpu->pio.cur_count = count;
1827 vcpu->pio.size = size;
1829 vcpu->pio.port = port;
1830 vcpu->pio.string = string;
1831 vcpu->pio.down = down;
1832 vcpu->pio.guest_page_offset = offset_in_page(address);
1833 vcpu->pio.rep = rep;
1835 pio_dev = vcpu_find_pio_dev(vcpu, port);
1837 kvm_arch_ops->cache_regs(vcpu);
1838 memcpy(vcpu->pio_data, &vcpu->regs[VCPU_REGS_RAX], 4);
1839 kvm_arch_ops->decache_regs(vcpu);
1841 kernel_pio(pio_dev, vcpu);
1847 /* TODO: String I/O for in kernel device */
1849 printk(KERN_ERR "kvm_setup_pio: no string io support\n");
1852 kvm_arch_ops->skip_emulated_instruction(vcpu);
1856 now = min(count, PAGE_SIZE / size);
1859 in_page = PAGE_SIZE - offset_in_page(address);
1861 in_page = offset_in_page(address) + size;
1862 now = min(count, (unsigned long)in_page / size);
1865 * String I/O straddles page boundary. Pin two guest pages
1866 * so that we satisfy atomicity constraints. Do just one
1867 * transaction to avoid complexity.
1874 * String I/O in reverse. Yuck. Kill the guest, fix later.
1876 printk(KERN_ERR "kvm: guest string pio down\n");
1880 vcpu->run->io.count = now;
1881 vcpu->pio.cur_count = now;
1883 for (i = 0; i < nr_pages; ++i) {
1884 spin_lock(&vcpu->kvm->lock);
1885 page = gva_to_page(vcpu, address + i * PAGE_SIZE);
1888 vcpu->pio.guest_pages[i] = page;
1889 spin_unlock(&vcpu->kvm->lock);
1892 free_pio_guest_pages(vcpu);
1898 return pio_copy_data(vcpu);
1901 EXPORT_SYMBOL_GPL(kvm_setup_pio);
1903 static int kvm_vcpu_ioctl_run(struct kvm_vcpu *vcpu, struct kvm_run *kvm_run)
1910 if (vcpu->sigset_active)
1911 sigprocmask(SIG_SETMASK, &vcpu->sigset, &sigsaved);
1913 /* re-sync apic's tpr */
1914 vcpu->cr8 = kvm_run->cr8;
1916 if (vcpu->pio.cur_count) {
1917 r = complete_pio(vcpu);
1922 if (vcpu->mmio_needed) {
1923 memcpy(vcpu->mmio_data, kvm_run->mmio.data, 8);
1924 vcpu->mmio_read_completed = 1;
1925 vcpu->mmio_needed = 0;
1926 r = emulate_instruction(vcpu, kvm_run,
1927 vcpu->mmio_fault_cr2, 0);
1928 if (r == EMULATE_DO_MMIO) {
1930 * Read-modify-write. Back to userspace.
1932 kvm_run->exit_reason = KVM_EXIT_MMIO;
1938 if (kvm_run->exit_reason == KVM_EXIT_HYPERCALL) {
1939 kvm_arch_ops->cache_regs(vcpu);
1940 vcpu->regs[VCPU_REGS_RAX] = kvm_run->hypercall.ret;
1941 kvm_arch_ops->decache_regs(vcpu);
1944 r = kvm_arch_ops->run(vcpu, kvm_run);
1947 if (vcpu->sigset_active)
1948 sigprocmask(SIG_SETMASK, &sigsaved, NULL);
1954 static int kvm_vcpu_ioctl_get_regs(struct kvm_vcpu *vcpu,
1955 struct kvm_regs *regs)
1959 kvm_arch_ops->cache_regs(vcpu);
1961 regs->rax = vcpu->regs[VCPU_REGS_RAX];
1962 regs->rbx = vcpu->regs[VCPU_REGS_RBX];
1963 regs->rcx = vcpu->regs[VCPU_REGS_RCX];
1964 regs->rdx = vcpu->regs[VCPU_REGS_RDX];
1965 regs->rsi = vcpu->regs[VCPU_REGS_RSI];
1966 regs->rdi = vcpu->regs[VCPU_REGS_RDI];
1967 regs->rsp = vcpu->regs[VCPU_REGS_RSP];
1968 regs->rbp = vcpu->regs[VCPU_REGS_RBP];
1969 #ifdef CONFIG_X86_64
1970 regs->r8 = vcpu->regs[VCPU_REGS_R8];
1971 regs->r9 = vcpu->regs[VCPU_REGS_R9];
1972 regs->r10 = vcpu->regs[VCPU_REGS_R10];
1973 regs->r11 = vcpu->regs[VCPU_REGS_R11];
1974 regs->r12 = vcpu->regs[VCPU_REGS_R12];
1975 regs->r13 = vcpu->regs[VCPU_REGS_R13];
1976 regs->r14 = vcpu->regs[VCPU_REGS_R14];
1977 regs->r15 = vcpu->regs[VCPU_REGS_R15];
1980 regs->rip = vcpu->rip;
1981 regs->rflags = kvm_arch_ops->get_rflags(vcpu);
1984 * Don't leak debug flags in case they were set for guest debugging
1986 if (vcpu->guest_debug.enabled && vcpu->guest_debug.singlestep)
1987 regs->rflags &= ~(X86_EFLAGS_TF | X86_EFLAGS_RF);
1994 static int kvm_vcpu_ioctl_set_regs(struct kvm_vcpu *vcpu,
1995 struct kvm_regs *regs)
1999 vcpu->regs[VCPU_REGS_RAX] = regs->rax;
2000 vcpu->regs[VCPU_REGS_RBX] = regs->rbx;
2001 vcpu->regs[VCPU_REGS_RCX] = regs->rcx;
2002 vcpu->regs[VCPU_REGS_RDX] = regs->rdx;
2003 vcpu->regs[VCPU_REGS_RSI] = regs->rsi;
2004 vcpu->regs[VCPU_REGS_RDI] = regs->rdi;
2005 vcpu->regs[VCPU_REGS_RSP] = regs->rsp;
2006 vcpu->regs[VCPU_REGS_RBP] = regs->rbp;
2007 #ifdef CONFIG_X86_64
2008 vcpu->regs[VCPU_REGS_R8] = regs->r8;
2009 vcpu->regs[VCPU_REGS_R9] = regs->r9;
2010 vcpu->regs[VCPU_REGS_R10] = regs->r10;
2011 vcpu->regs[VCPU_REGS_R11] = regs->r11;
2012 vcpu->regs[VCPU_REGS_R12] = regs->r12;
2013 vcpu->regs[VCPU_REGS_R13] = regs->r13;
2014 vcpu->regs[VCPU_REGS_R14] = regs->r14;
2015 vcpu->regs[VCPU_REGS_R15] = regs->r15;
2018 vcpu->rip = regs->rip;
2019 kvm_arch_ops->set_rflags(vcpu, regs->rflags);
2021 kvm_arch_ops->decache_regs(vcpu);
2028 static void get_segment(struct kvm_vcpu *vcpu,
2029 struct kvm_segment *var, int seg)
2031 return kvm_arch_ops->get_segment(vcpu, var, seg);
2034 static int kvm_vcpu_ioctl_get_sregs(struct kvm_vcpu *vcpu,
2035 struct kvm_sregs *sregs)
2037 struct descriptor_table dt;
2041 get_segment(vcpu, &sregs->cs, VCPU_SREG_CS);
2042 get_segment(vcpu, &sregs->ds, VCPU_SREG_DS);
2043 get_segment(vcpu, &sregs->es, VCPU_SREG_ES);
2044 get_segment(vcpu, &sregs->fs, VCPU_SREG_FS);
2045 get_segment(vcpu, &sregs->gs, VCPU_SREG_GS);
2046 get_segment(vcpu, &sregs->ss, VCPU_SREG_SS);
2048 get_segment(vcpu, &sregs->tr, VCPU_SREG_TR);
2049 get_segment(vcpu, &sregs->ldt, VCPU_SREG_LDTR);
2051 kvm_arch_ops->get_idt(vcpu, &dt);
2052 sregs->idt.limit = dt.limit;
2053 sregs->idt.base = dt.base;
2054 kvm_arch_ops->get_gdt(vcpu, &dt);
2055 sregs->gdt.limit = dt.limit;
2056 sregs->gdt.base = dt.base;
2058 kvm_arch_ops->decache_cr4_guest_bits(vcpu);
2059 sregs->cr0 = vcpu->cr0;
2060 sregs->cr2 = vcpu->cr2;
2061 sregs->cr3 = vcpu->cr3;
2062 sregs->cr4 = vcpu->cr4;
2063 sregs->cr8 = vcpu->cr8;
2064 sregs->efer = vcpu->shadow_efer;
2065 sregs->apic_base = vcpu->apic_base;
2067 memcpy(sregs->interrupt_bitmap, vcpu->irq_pending,
2068 sizeof sregs->interrupt_bitmap);
2075 static void set_segment(struct kvm_vcpu *vcpu,
2076 struct kvm_segment *var, int seg)
2078 return kvm_arch_ops->set_segment(vcpu, var, seg);
2081 static int kvm_vcpu_ioctl_set_sregs(struct kvm_vcpu *vcpu,
2082 struct kvm_sregs *sregs)
2084 int mmu_reset_needed = 0;
2086 struct descriptor_table dt;
2090 dt.limit = sregs->idt.limit;
2091 dt.base = sregs->idt.base;
2092 kvm_arch_ops->set_idt(vcpu, &dt);
2093 dt.limit = sregs->gdt.limit;
2094 dt.base = sregs->gdt.base;
2095 kvm_arch_ops->set_gdt(vcpu, &dt);
2097 vcpu->cr2 = sregs->cr2;
2098 mmu_reset_needed |= vcpu->cr3 != sregs->cr3;
2099 vcpu->cr3 = sregs->cr3;
2101 vcpu->cr8 = sregs->cr8;
2103 mmu_reset_needed |= vcpu->shadow_efer != sregs->efer;
2104 #ifdef CONFIG_X86_64
2105 kvm_arch_ops->set_efer(vcpu, sregs->efer);
2107 vcpu->apic_base = sregs->apic_base;
2109 kvm_arch_ops->decache_cr4_guest_bits(vcpu);
2111 mmu_reset_needed |= vcpu->cr0 != sregs->cr0;
2112 kvm_arch_ops->set_cr0(vcpu, sregs->cr0);
2114 mmu_reset_needed |= vcpu->cr4 != sregs->cr4;
2115 kvm_arch_ops->set_cr4(vcpu, sregs->cr4);
2116 if (!is_long_mode(vcpu) && is_pae(vcpu))
2117 load_pdptrs(vcpu, vcpu->cr3);
2119 if (mmu_reset_needed)
2120 kvm_mmu_reset_context(vcpu);
2122 memcpy(vcpu->irq_pending, sregs->interrupt_bitmap,
2123 sizeof vcpu->irq_pending);
2124 vcpu->irq_summary = 0;
2125 for (i = 0; i < NR_IRQ_WORDS; ++i)
2126 if (vcpu->irq_pending[i])
2127 __set_bit(i, &vcpu->irq_summary);
2129 set_segment(vcpu, &sregs->cs, VCPU_SREG_CS);
2130 set_segment(vcpu, &sregs->ds, VCPU_SREG_DS);
2131 set_segment(vcpu, &sregs->es, VCPU_SREG_ES);
2132 set_segment(vcpu, &sregs->fs, VCPU_SREG_FS);
2133 set_segment(vcpu, &sregs->gs, VCPU_SREG_GS);
2134 set_segment(vcpu, &sregs->ss, VCPU_SREG_SS);
2136 set_segment(vcpu, &sregs->tr, VCPU_SREG_TR);
2137 set_segment(vcpu, &sregs->ldt, VCPU_SREG_LDTR);
2145 * List of msr numbers which we expose to userspace through KVM_GET_MSRS
2146 * and KVM_SET_MSRS, and KVM_GET_MSR_INDEX_LIST.
2148 * This list is modified at module load time to reflect the
2149 * capabilities of the host cpu.
2151 static u32 msrs_to_save[] = {
2152 MSR_IA32_SYSENTER_CS, MSR_IA32_SYSENTER_ESP, MSR_IA32_SYSENTER_EIP,
2154 #ifdef CONFIG_X86_64
2155 MSR_CSTAR, MSR_KERNEL_GS_BASE, MSR_SYSCALL_MASK, MSR_LSTAR,
2157 MSR_IA32_TIME_STAMP_COUNTER,
2160 static unsigned num_msrs_to_save;
2162 static u32 emulated_msrs[] = {
2163 MSR_IA32_MISC_ENABLE,
2166 static __init void kvm_init_msr_list(void)
2171 for (i = j = 0; i < ARRAY_SIZE(msrs_to_save); i++) {
2172 if (rdmsr_safe(msrs_to_save[i], &dummy[0], &dummy[1]) < 0)
2175 msrs_to_save[j] = msrs_to_save[i];
2178 num_msrs_to_save = j;
2182 * Adapt set_msr() to msr_io()'s calling convention
2184 static int do_set_msr(struct kvm_vcpu *vcpu, unsigned index, u64 *data)
2186 return set_msr(vcpu, index, *data);
2190 * Read or write a bunch of msrs. All parameters are kernel addresses.
2192 * @return number of msrs set successfully.
2194 static int __msr_io(struct kvm_vcpu *vcpu, struct kvm_msrs *msrs,
2195 struct kvm_msr_entry *entries,
2196 int (*do_msr)(struct kvm_vcpu *vcpu,
2197 unsigned index, u64 *data))
2203 for (i = 0; i < msrs->nmsrs; ++i)
2204 if (do_msr(vcpu, entries[i].index, &entries[i].data))
2213 * Read or write a bunch of msrs. Parameters are user addresses.
2215 * @return number of msrs set successfully.
2217 static int msr_io(struct kvm_vcpu *vcpu, struct kvm_msrs __user *user_msrs,
2218 int (*do_msr)(struct kvm_vcpu *vcpu,
2219 unsigned index, u64 *data),
2222 struct kvm_msrs msrs;
2223 struct kvm_msr_entry *entries;
2228 if (copy_from_user(&msrs, user_msrs, sizeof msrs))
2232 if (msrs.nmsrs >= MAX_IO_MSRS)
2236 size = sizeof(struct kvm_msr_entry) * msrs.nmsrs;
2237 entries = vmalloc(size);
2242 if (copy_from_user(entries, user_msrs->entries, size))
2245 r = n = __msr_io(vcpu, &msrs, entries, do_msr);
2250 if (writeback && copy_to_user(user_msrs->entries, entries, size))
2262 * Translate a guest virtual address to a guest physical address.
2264 static int kvm_vcpu_ioctl_translate(struct kvm_vcpu *vcpu,
2265 struct kvm_translation *tr)
2267 unsigned long vaddr = tr->linear_address;
2271 spin_lock(&vcpu->kvm->lock);
2272 gpa = vcpu->mmu.gva_to_gpa(vcpu, vaddr);
2273 tr->physical_address = gpa;
2274 tr->valid = gpa != UNMAPPED_GVA;
2277 spin_unlock(&vcpu->kvm->lock);
2283 static int kvm_vcpu_ioctl_interrupt(struct kvm_vcpu *vcpu,
2284 struct kvm_interrupt *irq)
2286 if (irq->irq < 0 || irq->irq >= 256)
2290 set_bit(irq->irq, vcpu->irq_pending);
2291 set_bit(irq->irq / BITS_PER_LONG, &vcpu->irq_summary);
2298 static int kvm_vcpu_ioctl_debug_guest(struct kvm_vcpu *vcpu,
2299 struct kvm_debug_guest *dbg)
2305 r = kvm_arch_ops->set_guest_debug(vcpu, dbg);
2312 static struct page *kvm_vcpu_nopage(struct vm_area_struct *vma,
2313 unsigned long address,
2316 struct kvm_vcpu *vcpu = vma->vm_file->private_data;
2317 unsigned long pgoff;
2320 *type = VM_FAULT_MINOR;
2321 pgoff = ((address - vma->vm_start) >> PAGE_SHIFT) + vma->vm_pgoff;
2323 page = virt_to_page(vcpu->run);
2324 else if (pgoff == KVM_PIO_PAGE_OFFSET)
2325 page = virt_to_page(vcpu->pio_data);
2327 return NOPAGE_SIGBUS;
2332 static struct vm_operations_struct kvm_vcpu_vm_ops = {
2333 .nopage = kvm_vcpu_nopage,
2336 static int kvm_vcpu_mmap(struct file *file, struct vm_area_struct *vma)
2338 vma->vm_ops = &kvm_vcpu_vm_ops;
2342 static int kvm_vcpu_release(struct inode *inode, struct file *filp)
2344 struct kvm_vcpu *vcpu = filp->private_data;
2346 fput(vcpu->kvm->filp);
2350 static struct file_operations kvm_vcpu_fops = {
2351 .release = kvm_vcpu_release,
2352 .unlocked_ioctl = kvm_vcpu_ioctl,
2353 .compat_ioctl = kvm_vcpu_ioctl,
2354 .mmap = kvm_vcpu_mmap,
2358 * Allocates an inode for the vcpu.
2360 static int create_vcpu_fd(struct kvm_vcpu *vcpu)
2363 struct inode *inode;
2366 r = anon_inode_getfd(&fd, &inode, &file,
2367 "kvm-vcpu", &kvm_vcpu_fops, vcpu);
2370 atomic_inc(&vcpu->kvm->filp->f_count);
2375 * Creates some virtual cpus. Good luck creating more than one.
2377 static int kvm_vm_ioctl_create_vcpu(struct kvm *kvm, int n)
2380 struct kvm_vcpu *vcpu;
2387 vcpu = &kvm->vcpus[n];
2389 mutex_lock(&vcpu->mutex);
2392 mutex_unlock(&vcpu->mutex);
2396 page = alloc_page(GFP_KERNEL | __GFP_ZERO);
2400 vcpu->run = page_address(page);
2402 page = alloc_page(GFP_KERNEL | __GFP_ZERO);
2406 vcpu->pio_data = page_address(page);
2408 vcpu->host_fx_image = (char*)ALIGN((hva_t)vcpu->fx_buf,
2410 vcpu->guest_fx_image = vcpu->host_fx_image + FX_IMAGE_SIZE;
2413 r = kvm_arch_ops->vcpu_create(vcpu);
2415 goto out_free_vcpus;
2417 r = kvm_mmu_create(vcpu);
2419 goto out_free_vcpus;
2421 kvm_arch_ops->vcpu_load(vcpu);
2422 r = kvm_mmu_setup(vcpu);
2424 r = kvm_arch_ops->vcpu_setup(vcpu);
2428 goto out_free_vcpus;
2430 r = create_vcpu_fd(vcpu);
2432 goto out_free_vcpus;
2434 spin_lock(&kvm_lock);
2435 if (n >= kvm->nvcpus)
2436 kvm->nvcpus = n + 1;
2437 spin_unlock(&kvm_lock);
2442 kvm_free_vcpu(vcpu);
2444 free_page((unsigned long)vcpu->run);
2447 mutex_unlock(&vcpu->mutex);
2452 static void cpuid_fix_nx_cap(struct kvm_vcpu *vcpu)
2456 struct kvm_cpuid_entry *e, *entry;
2458 rdmsrl(MSR_EFER, efer);
2460 for (i = 0; i < vcpu->cpuid_nent; ++i) {
2461 e = &vcpu->cpuid_entries[i];
2462 if (e->function == 0x80000001) {
2467 if (entry && (entry->edx & EFER_NX) && !(efer & EFER_NX)) {
2468 entry->edx &= ~(1 << 20);
2469 printk(KERN_INFO ": guest NX capability removed\n");
2473 static int kvm_vcpu_ioctl_set_cpuid(struct kvm_vcpu *vcpu,
2474 struct kvm_cpuid *cpuid,
2475 struct kvm_cpuid_entry __user *entries)
2480 if (cpuid->nent > KVM_MAX_CPUID_ENTRIES)
2483 if (copy_from_user(&vcpu->cpuid_entries, entries,
2484 cpuid->nent * sizeof(struct kvm_cpuid_entry)))
2486 vcpu->cpuid_nent = cpuid->nent;
2487 cpuid_fix_nx_cap(vcpu);
2494 static int kvm_vcpu_ioctl_set_sigmask(struct kvm_vcpu *vcpu, sigset_t *sigset)
2497 sigdelsetmask(sigset, sigmask(SIGKILL)|sigmask(SIGSTOP));
2498 vcpu->sigset_active = 1;
2499 vcpu->sigset = *sigset;
2501 vcpu->sigset_active = 0;
2506 * fxsave fpu state. Taken from x86_64/processor.h. To be killed when
2507 * we have asm/x86/processor.h
2518 u32 st_space[32]; /* 8*16 bytes for each FP-reg = 128 bytes */
2519 #ifdef CONFIG_X86_64
2520 u32 xmm_space[64]; /* 16*16 bytes for each XMM-reg = 256 bytes */
2522 u32 xmm_space[32]; /* 8*16 bytes for each XMM-reg = 128 bytes */
2526 static int kvm_vcpu_ioctl_get_fpu(struct kvm_vcpu *vcpu, struct kvm_fpu *fpu)
2528 struct fxsave *fxsave = (struct fxsave *)vcpu->guest_fx_image;
2532 memcpy(fpu->fpr, fxsave->st_space, 128);
2533 fpu->fcw = fxsave->cwd;
2534 fpu->fsw = fxsave->swd;
2535 fpu->ftwx = fxsave->twd;
2536 fpu->last_opcode = fxsave->fop;
2537 fpu->last_ip = fxsave->rip;
2538 fpu->last_dp = fxsave->rdp;
2539 memcpy(fpu->xmm, fxsave->xmm_space, sizeof fxsave->xmm_space);
2546 static int kvm_vcpu_ioctl_set_fpu(struct kvm_vcpu *vcpu, struct kvm_fpu *fpu)
2548 struct fxsave *fxsave = (struct fxsave *)vcpu->guest_fx_image;
2552 memcpy(fxsave->st_space, fpu->fpr, 128);
2553 fxsave->cwd = fpu->fcw;
2554 fxsave->swd = fpu->fsw;
2555 fxsave->twd = fpu->ftwx;
2556 fxsave->fop = fpu->last_opcode;
2557 fxsave->rip = fpu->last_ip;
2558 fxsave->rdp = fpu->last_dp;
2559 memcpy(fxsave->xmm_space, fpu->xmm, sizeof fxsave->xmm_space);
2566 static long kvm_vcpu_ioctl(struct file *filp,
2567 unsigned int ioctl, unsigned long arg)
2569 struct kvm_vcpu *vcpu = filp->private_data;
2570 void __user *argp = (void __user *)arg;
2578 r = kvm_vcpu_ioctl_run(vcpu, vcpu->run);
2580 case KVM_GET_REGS: {
2581 struct kvm_regs kvm_regs;
2583 memset(&kvm_regs, 0, sizeof kvm_regs);
2584 r = kvm_vcpu_ioctl_get_regs(vcpu, &kvm_regs);
2588 if (copy_to_user(argp, &kvm_regs, sizeof kvm_regs))
2593 case KVM_SET_REGS: {
2594 struct kvm_regs kvm_regs;
2597 if (copy_from_user(&kvm_regs, argp, sizeof kvm_regs))
2599 r = kvm_vcpu_ioctl_set_regs(vcpu, &kvm_regs);
2605 case KVM_GET_SREGS: {
2606 struct kvm_sregs kvm_sregs;
2608 memset(&kvm_sregs, 0, sizeof kvm_sregs);
2609 r = kvm_vcpu_ioctl_get_sregs(vcpu, &kvm_sregs);
2613 if (copy_to_user(argp, &kvm_sregs, sizeof kvm_sregs))
2618 case KVM_SET_SREGS: {
2619 struct kvm_sregs kvm_sregs;
2622 if (copy_from_user(&kvm_sregs, argp, sizeof kvm_sregs))
2624 r = kvm_vcpu_ioctl_set_sregs(vcpu, &kvm_sregs);
2630 case KVM_TRANSLATE: {
2631 struct kvm_translation tr;
2634 if (copy_from_user(&tr, argp, sizeof tr))
2636 r = kvm_vcpu_ioctl_translate(vcpu, &tr);
2640 if (copy_to_user(argp, &tr, sizeof tr))
2645 case KVM_INTERRUPT: {
2646 struct kvm_interrupt irq;
2649 if (copy_from_user(&irq, argp, sizeof irq))
2651 r = kvm_vcpu_ioctl_interrupt(vcpu, &irq);
2657 case KVM_DEBUG_GUEST: {
2658 struct kvm_debug_guest dbg;
2661 if (copy_from_user(&dbg, argp, sizeof dbg))
2663 r = kvm_vcpu_ioctl_debug_guest(vcpu, &dbg);
2670 r = msr_io(vcpu, argp, get_msr, 1);
2673 r = msr_io(vcpu, argp, do_set_msr, 0);
2675 case KVM_SET_CPUID: {
2676 struct kvm_cpuid __user *cpuid_arg = argp;
2677 struct kvm_cpuid cpuid;
2680 if (copy_from_user(&cpuid, cpuid_arg, sizeof cpuid))
2682 r = kvm_vcpu_ioctl_set_cpuid(vcpu, &cpuid, cpuid_arg->entries);
2687 case KVM_SET_SIGNAL_MASK: {
2688 struct kvm_signal_mask __user *sigmask_arg = argp;
2689 struct kvm_signal_mask kvm_sigmask;
2690 sigset_t sigset, *p;
2695 if (copy_from_user(&kvm_sigmask, argp,
2696 sizeof kvm_sigmask))
2699 if (kvm_sigmask.len != sizeof sigset)
2702 if (copy_from_user(&sigset, sigmask_arg->sigset,
2707 r = kvm_vcpu_ioctl_set_sigmask(vcpu, &sigset);
2713 memset(&fpu, 0, sizeof fpu);
2714 r = kvm_vcpu_ioctl_get_fpu(vcpu, &fpu);
2718 if (copy_to_user(argp, &fpu, sizeof fpu))
2727 if (copy_from_user(&fpu, argp, sizeof fpu))
2729 r = kvm_vcpu_ioctl_set_fpu(vcpu, &fpu);
2742 static long kvm_vm_ioctl(struct file *filp,
2743 unsigned int ioctl, unsigned long arg)
2745 struct kvm *kvm = filp->private_data;
2746 void __user *argp = (void __user *)arg;
2750 case KVM_CREATE_VCPU:
2751 r = kvm_vm_ioctl_create_vcpu(kvm, arg);
2755 case KVM_SET_MEMORY_REGION: {
2756 struct kvm_memory_region kvm_mem;
2759 if (copy_from_user(&kvm_mem, argp, sizeof kvm_mem))
2761 r = kvm_vm_ioctl_set_memory_region(kvm, &kvm_mem);
2766 case KVM_GET_DIRTY_LOG: {
2767 struct kvm_dirty_log log;
2770 if (copy_from_user(&log, argp, sizeof log))
2772 r = kvm_vm_ioctl_get_dirty_log(kvm, &log);
2777 case KVM_SET_MEMORY_ALIAS: {
2778 struct kvm_memory_alias alias;
2781 if (copy_from_user(&alias, argp, sizeof alias))
2783 r = kvm_vm_ioctl_set_memory_alias(kvm, &alias);
2795 static struct page *kvm_vm_nopage(struct vm_area_struct *vma,
2796 unsigned long address,
2799 struct kvm *kvm = vma->vm_file->private_data;
2800 unsigned long pgoff;
2803 *type = VM_FAULT_MINOR;
2804 pgoff = ((address - vma->vm_start) >> PAGE_SHIFT) + vma->vm_pgoff;
2805 page = gfn_to_page(kvm, pgoff);
2807 return NOPAGE_SIGBUS;
2812 static struct vm_operations_struct kvm_vm_vm_ops = {
2813 .nopage = kvm_vm_nopage,
2816 static int kvm_vm_mmap(struct file *file, struct vm_area_struct *vma)
2818 vma->vm_ops = &kvm_vm_vm_ops;
2822 static struct file_operations kvm_vm_fops = {
2823 .release = kvm_vm_release,
2824 .unlocked_ioctl = kvm_vm_ioctl,
2825 .compat_ioctl = kvm_vm_ioctl,
2826 .mmap = kvm_vm_mmap,
2829 static int kvm_dev_ioctl_create_vm(void)
2832 struct inode *inode;
2836 kvm = kvm_create_vm();
2838 return PTR_ERR(kvm);
2839 r = anon_inode_getfd(&fd, &inode, &file, "kvm-vm", &kvm_vm_fops, kvm);
2841 kvm_destroy_vm(kvm);
2850 static long kvm_dev_ioctl(struct file *filp,
2851 unsigned int ioctl, unsigned long arg)
2853 void __user *argp = (void __user *)arg;
2857 case KVM_GET_API_VERSION:
2861 r = KVM_API_VERSION;
2867 r = kvm_dev_ioctl_create_vm();
2869 case KVM_GET_MSR_INDEX_LIST: {
2870 struct kvm_msr_list __user *user_msr_list = argp;
2871 struct kvm_msr_list msr_list;
2875 if (copy_from_user(&msr_list, user_msr_list, sizeof msr_list))
2878 msr_list.nmsrs = num_msrs_to_save + ARRAY_SIZE(emulated_msrs);
2879 if (copy_to_user(user_msr_list, &msr_list, sizeof msr_list))
2882 if (n < num_msrs_to_save)
2885 if (copy_to_user(user_msr_list->indices, &msrs_to_save,
2886 num_msrs_to_save * sizeof(u32)))
2888 if (copy_to_user(user_msr_list->indices
2889 + num_msrs_to_save * sizeof(u32),
2891 ARRAY_SIZE(emulated_msrs) * sizeof(u32)))
2896 case KVM_CHECK_EXTENSION:
2898 * No extensions defined at present.
2902 case KVM_GET_VCPU_MMAP_SIZE:
2915 static struct file_operations kvm_chardev_ops = {
2916 .open = kvm_dev_open,
2917 .release = kvm_dev_release,
2918 .unlocked_ioctl = kvm_dev_ioctl,
2919 .compat_ioctl = kvm_dev_ioctl,
2922 static struct miscdevice kvm_dev = {
2928 static int kvm_reboot(struct notifier_block *notifier, unsigned long val,
2931 if (val == SYS_RESTART) {
2933 * Some (well, at least mine) BIOSes hang on reboot if
2936 printk(KERN_INFO "kvm: exiting hardware virtualization\n");
2937 on_each_cpu(hardware_disable, NULL, 0, 1);
2942 static struct notifier_block kvm_reboot_notifier = {
2943 .notifier_call = kvm_reboot,
2948 * Make sure that a cpu that is being hot-unplugged does not have any vcpus
2951 static void decache_vcpus_on_cpu(int cpu)
2954 struct kvm_vcpu *vcpu;
2957 spin_lock(&kvm_lock);
2958 list_for_each_entry(vm, &vm_list, vm_list)
2959 for (i = 0; i < KVM_MAX_VCPUS; ++i) {
2960 vcpu = &vm->vcpus[i];
2962 * If the vcpu is locked, then it is running on some
2963 * other cpu and therefore it is not cached on the
2966 * If it's not locked, check the last cpu it executed
2969 if (mutex_trylock(&vcpu->mutex)) {
2970 if (vcpu->cpu == cpu) {
2971 kvm_arch_ops->vcpu_decache(vcpu);
2974 mutex_unlock(&vcpu->mutex);
2977 spin_unlock(&kvm_lock);
2980 static void hardware_enable(void *junk)
2982 int cpu = raw_smp_processor_id();
2984 if (cpu_isset(cpu, cpus_hardware_enabled))
2986 cpu_set(cpu, cpus_hardware_enabled);
2987 kvm_arch_ops->hardware_enable(NULL);
2990 static void hardware_disable(void *junk)
2992 int cpu = raw_smp_processor_id();
2994 if (!cpu_isset(cpu, cpus_hardware_enabled))
2996 cpu_clear(cpu, cpus_hardware_enabled);
2997 decache_vcpus_on_cpu(cpu);
2998 kvm_arch_ops->hardware_disable(NULL);
3001 static int kvm_cpu_hotplug(struct notifier_block *notifier, unsigned long val,
3008 case CPU_DYING_FROZEN:
3009 case CPU_UP_CANCELED:
3010 case CPU_UP_CANCELED_FROZEN:
3011 printk(KERN_INFO "kvm: disabling virtualization on CPU%d\n",
3013 smp_call_function_single(cpu, hardware_disable, NULL, 0, 1);
3016 case CPU_ONLINE_FROZEN:
3017 printk(KERN_INFO "kvm: enabling virtualization on CPU%d\n",
3019 smp_call_function_single(cpu, hardware_enable, NULL, 0, 1);
3025 void kvm_io_bus_init(struct kvm_io_bus *bus)
3027 memset(bus, 0, sizeof(*bus));
3030 void kvm_io_bus_destroy(struct kvm_io_bus *bus)
3034 for (i = 0; i < bus->dev_count; i++) {
3035 struct kvm_io_device *pos = bus->devs[i];
3037 kvm_iodevice_destructor(pos);
3041 struct kvm_io_device *kvm_io_bus_find_dev(struct kvm_io_bus *bus, gpa_t addr)
3045 for (i = 0; i < bus->dev_count; i++) {
3046 struct kvm_io_device *pos = bus->devs[i];
3048 if (pos->in_range(pos, addr))
3055 void kvm_io_bus_register_dev(struct kvm_io_bus *bus, struct kvm_io_device *dev)
3057 BUG_ON(bus->dev_count > (NR_IOBUS_DEVS-1));
3059 bus->devs[bus->dev_count++] = dev;
3062 static struct notifier_block kvm_cpu_notifier = {
3063 .notifier_call = kvm_cpu_hotplug,
3064 .priority = 20, /* must be > scheduler priority */
3067 static u64 stat_get(void *_offset)
3069 unsigned offset = (long)_offset;
3072 struct kvm_vcpu *vcpu;
3075 spin_lock(&kvm_lock);
3076 list_for_each_entry(kvm, &vm_list, vm_list)
3077 for (i = 0; i < KVM_MAX_VCPUS; ++i) {
3078 vcpu = &kvm->vcpus[i];
3079 total += *(u32 *)((void *)vcpu + offset);
3081 spin_unlock(&kvm_lock);
3085 static void stat_set(void *offset, u64 val)
3089 DEFINE_SIMPLE_ATTRIBUTE(stat_fops, stat_get, stat_set, "%llu\n");
3091 static __init void kvm_init_debug(void)
3093 struct kvm_stats_debugfs_item *p;
3095 debugfs_dir = debugfs_create_dir("kvm", NULL);
3096 for (p = debugfs_entries; p->name; ++p)
3097 p->dentry = debugfs_create_file(p->name, 0444, debugfs_dir,
3098 (void *)(long)p->offset,
3102 static void kvm_exit_debug(void)
3104 struct kvm_stats_debugfs_item *p;
3106 for (p = debugfs_entries; p->name; ++p)
3107 debugfs_remove(p->dentry);
3108 debugfs_remove(debugfs_dir);
3111 static int kvm_suspend(struct sys_device *dev, pm_message_t state)
3113 hardware_disable(NULL);
3117 static int kvm_resume(struct sys_device *dev)
3119 hardware_enable(NULL);
3123 static struct sysdev_class kvm_sysdev_class = {
3124 set_kset_name("kvm"),
3125 .suspend = kvm_suspend,
3126 .resume = kvm_resume,
3129 static struct sys_device kvm_sysdev = {
3131 .cls = &kvm_sysdev_class,
3134 hpa_t bad_page_address;
3136 int kvm_init_arch(struct kvm_arch_ops *ops, struct module *module)
3141 printk(KERN_ERR "kvm: already loaded the other module\n");
3145 if (!ops->cpu_has_kvm_support()) {
3146 printk(KERN_ERR "kvm: no hardware support\n");
3149 if (ops->disabled_by_bios()) {
3150 printk(KERN_ERR "kvm: disabled by bios\n");
3156 r = kvm_arch_ops->hardware_setup();
3160 on_each_cpu(hardware_enable, NULL, 0, 1);
3161 r = register_cpu_notifier(&kvm_cpu_notifier);
3164 register_reboot_notifier(&kvm_reboot_notifier);
3166 r = sysdev_class_register(&kvm_sysdev_class);
3170 r = sysdev_register(&kvm_sysdev);
3174 kvm_chardev_ops.owner = module;
3176 r = misc_register(&kvm_dev);
3178 printk (KERN_ERR "kvm: misc device register failed\n");
3185 sysdev_unregister(&kvm_sysdev);
3187 sysdev_class_unregister(&kvm_sysdev_class);
3189 unregister_reboot_notifier(&kvm_reboot_notifier);
3190 unregister_cpu_notifier(&kvm_cpu_notifier);
3192 on_each_cpu(hardware_disable, NULL, 0, 1);
3193 kvm_arch_ops->hardware_unsetup();
3195 kvm_arch_ops = NULL;
3199 void kvm_exit_arch(void)
3201 misc_deregister(&kvm_dev);
3202 sysdev_unregister(&kvm_sysdev);
3203 sysdev_class_unregister(&kvm_sysdev_class);
3204 unregister_reboot_notifier(&kvm_reboot_notifier);
3205 unregister_cpu_notifier(&kvm_cpu_notifier);
3206 on_each_cpu(hardware_disable, NULL, 0, 1);
3207 kvm_arch_ops->hardware_unsetup();
3208 kvm_arch_ops = NULL;
3211 static __init int kvm_init(void)
3213 static struct page *bad_page;
3216 r = kvm_mmu_module_init();
3222 kvm_init_msr_list();
3224 if ((bad_page = alloc_page(GFP_KERNEL)) == NULL) {
3229 bad_page_address = page_to_pfn(bad_page) << PAGE_SHIFT;
3230 memset(__va(bad_page_address), 0, PAGE_SIZE);
3236 kvm_mmu_module_exit();
3241 static __exit void kvm_exit(void)
3244 __free_page(pfn_to_page(bad_page_address >> PAGE_SHIFT));
3245 kvm_mmu_module_exit();
3248 module_init(kvm_init)
3249 module_exit(kvm_exit)
3251 EXPORT_SYMBOL_GPL(kvm_init_arch);
3252 EXPORT_SYMBOL_GPL(kvm_exit_arch);