2 * Kernel-based Virtual Machine driver for Linux
4 * derived from drivers/kvm/kvm_main.c
6 * Copyright (C) 2006 Qumranet, Inc.
7 * Copyright (C) 2008 Qumranet, Inc.
8 * Copyright IBM Corporation, 2008
11 * Avi Kivity <avi@qumranet.com>
12 * Yaniv Kamay <yaniv@qumranet.com>
13 * Amit Shah <amit.shah@qumranet.com>
14 * Ben-Ami Yassour <benami@il.ibm.com>
16 * This work is licensed under the terms of the GNU GPL, version 2. See
17 * the COPYING file in the top-level directory.
21 #include <linux/kvm_host.h>
26 #include "kvm_cache_regs.h"
29 #include <linux/clocksource.h>
30 #include <linux/interrupt.h>
31 #include <linux/kvm.h>
33 #include <linux/pci.h>
34 #include <linux/vmalloc.h>
35 #include <linux/module.h>
36 #include <linux/mman.h>
37 #include <linux/highmem.h>
38 #include <linux/intel-iommu.h>
40 #include <asm/uaccess.h>
44 #define MAX_IO_MSRS 256
45 #define CR0_RESERVED_BITS \
46 (~(unsigned long)(X86_CR0_PE | X86_CR0_MP | X86_CR0_EM | X86_CR0_TS \
47 | X86_CR0_ET | X86_CR0_NE | X86_CR0_WP | X86_CR0_AM \
48 | X86_CR0_NW | X86_CR0_CD | X86_CR0_PG))
49 #define CR4_RESERVED_BITS \
50 (~(unsigned long)(X86_CR4_VME | X86_CR4_PVI | X86_CR4_TSD | X86_CR4_DE\
51 | X86_CR4_PSE | X86_CR4_PAE | X86_CR4_MCE \
52 | X86_CR4_PGE | X86_CR4_PCE | X86_CR4_OSFXSR \
53 | X86_CR4_OSXMMEXCPT | X86_CR4_VMXE))
55 #define CR8_RESERVED_BITS (~(unsigned long)X86_CR8_TPR)
57 * - enable syscall per default because its emulated by KVM
58 * - enable LME and LMA per default on 64 bit KVM
61 static u64 __read_mostly efer_reserved_bits = 0xfffffffffffffafeULL;
63 static u64 __read_mostly efer_reserved_bits = 0xfffffffffffffffeULL;
66 #define VM_STAT(x) offsetof(struct kvm, stat.x), KVM_STAT_VM
67 #define VCPU_STAT(x) offsetof(struct kvm_vcpu, stat.x), KVM_STAT_VCPU
69 static int kvm_dev_ioctl_get_supported_cpuid(struct kvm_cpuid2 *cpuid,
70 struct kvm_cpuid_entry2 __user *entries);
72 struct kvm_x86_ops *kvm_x86_ops;
73 EXPORT_SYMBOL_GPL(kvm_x86_ops);
75 struct kvm_stats_debugfs_item debugfs_entries[] = {
76 { "pf_fixed", VCPU_STAT(pf_fixed) },
77 { "pf_guest", VCPU_STAT(pf_guest) },
78 { "tlb_flush", VCPU_STAT(tlb_flush) },
79 { "invlpg", VCPU_STAT(invlpg) },
80 { "exits", VCPU_STAT(exits) },
81 { "io_exits", VCPU_STAT(io_exits) },
82 { "mmio_exits", VCPU_STAT(mmio_exits) },
83 { "signal_exits", VCPU_STAT(signal_exits) },
84 { "irq_window", VCPU_STAT(irq_window_exits) },
85 { "nmi_window", VCPU_STAT(nmi_window_exits) },
86 { "halt_exits", VCPU_STAT(halt_exits) },
87 { "halt_wakeup", VCPU_STAT(halt_wakeup) },
88 { "hypercalls", VCPU_STAT(hypercalls) },
89 { "request_irq", VCPU_STAT(request_irq_exits) },
90 { "irq_exits", VCPU_STAT(irq_exits) },
91 { "host_state_reload", VCPU_STAT(host_state_reload) },
92 { "efer_reload", VCPU_STAT(efer_reload) },
93 { "fpu_reload", VCPU_STAT(fpu_reload) },
94 { "insn_emulation", VCPU_STAT(insn_emulation) },
95 { "insn_emulation_fail", VCPU_STAT(insn_emulation_fail) },
96 { "irq_injections", VCPU_STAT(irq_injections) },
97 { "mmu_shadow_zapped", VM_STAT(mmu_shadow_zapped) },
98 { "mmu_pte_write", VM_STAT(mmu_pte_write) },
99 { "mmu_pte_updated", VM_STAT(mmu_pte_updated) },
100 { "mmu_pde_zapped", VM_STAT(mmu_pde_zapped) },
101 { "mmu_flooded", VM_STAT(mmu_flooded) },
102 { "mmu_recycled", VM_STAT(mmu_recycled) },
103 { "mmu_cache_miss", VM_STAT(mmu_cache_miss) },
104 { "remote_tlb_flush", VM_STAT(remote_tlb_flush) },
105 { "largepages", VM_STAT(lpages) },
109 static struct kvm_assigned_dev_kernel *kvm_find_assigned_dev(struct list_head *head,
112 struct list_head *ptr;
113 struct kvm_assigned_dev_kernel *match;
115 list_for_each(ptr, head) {
116 match = list_entry(ptr, struct kvm_assigned_dev_kernel, list);
117 if (match->assigned_dev_id == assigned_dev_id)
123 static void kvm_assigned_dev_interrupt_work_handler(struct work_struct *work)
125 struct kvm_assigned_dev_kernel *assigned_dev;
127 assigned_dev = container_of(work, struct kvm_assigned_dev_kernel,
130 /* This is taken to safely inject irq inside the guest. When
131 * the interrupt injection (or the ioapic code) uses a
132 * finer-grained lock, update this
134 mutex_lock(&assigned_dev->kvm->lock);
135 kvm_set_irq(assigned_dev->kvm,
136 assigned_dev->guest_irq, 1);
137 mutex_unlock(&assigned_dev->kvm->lock);
138 kvm_put_kvm(assigned_dev->kvm);
141 /* FIXME: Implement the OR logic needed to make shared interrupts on
142 * this line behave properly
144 static irqreturn_t kvm_assigned_dev_intr(int irq, void *dev_id)
146 struct kvm_assigned_dev_kernel *assigned_dev =
147 (struct kvm_assigned_dev_kernel *) dev_id;
149 kvm_get_kvm(assigned_dev->kvm);
150 schedule_work(&assigned_dev->interrupt_work);
151 disable_irq_nosync(irq);
155 /* Ack the irq line for an assigned device */
156 static void kvm_assigned_dev_ack_irq(struct kvm_irq_ack_notifier *kian)
158 struct kvm_assigned_dev_kernel *dev;
163 dev = container_of(kian, struct kvm_assigned_dev_kernel,
165 kvm_set_irq(dev->kvm, dev->guest_irq, 0);
166 enable_irq(dev->host_irq);
169 static void kvm_free_assigned_device(struct kvm *kvm,
170 struct kvm_assigned_dev_kernel
173 if (irqchip_in_kernel(kvm) && assigned_dev->irq_requested)
174 free_irq(assigned_dev->host_irq, (void *)assigned_dev);
176 kvm_unregister_irq_ack_notifier(kvm, &assigned_dev->ack_notifier);
178 if (cancel_work_sync(&assigned_dev->interrupt_work))
179 /* We had pending work. That means we will have to take
180 * care of kvm_put_kvm.
184 pci_release_regions(assigned_dev->dev);
185 pci_disable_device(assigned_dev->dev);
186 pci_dev_put(assigned_dev->dev);
188 list_del(&assigned_dev->list);
192 static void kvm_free_all_assigned_devices(struct kvm *kvm)
194 struct list_head *ptr, *ptr2;
195 struct kvm_assigned_dev_kernel *assigned_dev;
197 list_for_each_safe(ptr, ptr2, &kvm->arch.assigned_dev_head) {
198 assigned_dev = list_entry(ptr,
199 struct kvm_assigned_dev_kernel,
202 kvm_free_assigned_device(kvm, assigned_dev);
206 static int kvm_vm_ioctl_assign_irq(struct kvm *kvm,
207 struct kvm_assigned_irq
211 struct kvm_assigned_dev_kernel *match;
213 mutex_lock(&kvm->lock);
215 match = kvm_find_assigned_dev(&kvm->arch.assigned_dev_head,
216 assigned_irq->assigned_dev_id);
218 mutex_unlock(&kvm->lock);
222 if (match->irq_requested) {
223 match->guest_irq = assigned_irq->guest_irq;
224 match->ack_notifier.gsi = assigned_irq->guest_irq;
225 mutex_unlock(&kvm->lock);
229 INIT_WORK(&match->interrupt_work,
230 kvm_assigned_dev_interrupt_work_handler);
232 if (irqchip_in_kernel(kvm)) {
233 if (!capable(CAP_SYS_RAWIO)) {
238 if (assigned_irq->host_irq)
239 match->host_irq = assigned_irq->host_irq;
241 match->host_irq = match->dev->irq;
242 match->guest_irq = assigned_irq->guest_irq;
243 match->ack_notifier.gsi = assigned_irq->guest_irq;
244 match->ack_notifier.irq_acked = kvm_assigned_dev_ack_irq;
245 kvm_register_irq_ack_notifier(kvm, &match->ack_notifier);
247 /* Even though this is PCI, we don't want to use shared
248 * interrupts. Sharing host devices with guest-assigned devices
249 * on the same interrupt line is not a happy situation: there
250 * are going to be long delays in accepting, acking, etc.
252 if (request_irq(match->host_irq, kvm_assigned_dev_intr, 0,
253 "kvm_assigned_device", (void *)match)) {
259 match->irq_requested = true;
260 mutex_unlock(&kvm->lock);
263 mutex_unlock(&kvm->lock);
264 kvm_free_assigned_device(kvm, match);
268 static int kvm_vm_ioctl_assign_device(struct kvm *kvm,
269 struct kvm_assigned_pci_dev *assigned_dev)
272 struct kvm_assigned_dev_kernel *match;
275 mutex_lock(&kvm->lock);
277 match = kvm_find_assigned_dev(&kvm->arch.assigned_dev_head,
278 assigned_dev->assigned_dev_id);
280 /* device already assigned */
285 match = kzalloc(sizeof(struct kvm_assigned_dev_kernel), GFP_KERNEL);
287 printk(KERN_INFO "%s: Couldn't allocate memory\n",
292 dev = pci_get_bus_and_slot(assigned_dev->busnr,
293 assigned_dev->devfn);
295 printk(KERN_INFO "%s: host device not found\n", __func__);
299 if (pci_enable_device(dev)) {
300 printk(KERN_INFO "%s: Could not enable PCI device\n", __func__);
304 r = pci_request_regions(dev, "kvm_assigned_device");
306 printk(KERN_INFO "%s: Could not get access to device regions\n",
310 match->assigned_dev_id = assigned_dev->assigned_dev_id;
311 match->host_busnr = assigned_dev->busnr;
312 match->host_devfn = assigned_dev->devfn;
317 list_add(&match->list, &kvm->arch.assigned_dev_head);
319 if (assigned_dev->flags & KVM_DEV_ASSIGN_ENABLE_IOMMU) {
320 r = kvm_iommu_map_guest(kvm, match);
326 mutex_unlock(&kvm->lock);
329 list_del(&match->list);
330 pci_release_regions(dev);
332 pci_disable_device(dev);
337 mutex_unlock(&kvm->lock);
341 unsigned long segment_base(u16 selector)
343 struct descriptor_table gdt;
344 struct desc_struct *d;
345 unsigned long table_base;
351 asm("sgdt %0" : "=m"(gdt));
352 table_base = gdt.base;
354 if (selector & 4) { /* from ldt */
357 asm("sldt %0" : "=g"(ldt_selector));
358 table_base = segment_base(ldt_selector);
360 d = (struct desc_struct *)(table_base + (selector & ~7));
361 v = d->base0 | ((unsigned long)d->base1 << 16) |
362 ((unsigned long)d->base2 << 24);
364 if (d->s == 0 && (d->type == 2 || d->type == 9 || d->type == 11))
365 v |= ((unsigned long)((struct ldttss_desc64 *)d)->base3) << 32;
369 EXPORT_SYMBOL_GPL(segment_base);
371 u64 kvm_get_apic_base(struct kvm_vcpu *vcpu)
373 if (irqchip_in_kernel(vcpu->kvm))
374 return vcpu->arch.apic_base;
376 return vcpu->arch.apic_base;
378 EXPORT_SYMBOL_GPL(kvm_get_apic_base);
380 void kvm_set_apic_base(struct kvm_vcpu *vcpu, u64 data)
382 /* TODO: reserve bits check */
383 if (irqchip_in_kernel(vcpu->kvm))
384 kvm_lapic_set_base(vcpu, data);
386 vcpu->arch.apic_base = data;
388 EXPORT_SYMBOL_GPL(kvm_set_apic_base);
390 void kvm_queue_exception(struct kvm_vcpu *vcpu, unsigned nr)
392 WARN_ON(vcpu->arch.exception.pending);
393 vcpu->arch.exception.pending = true;
394 vcpu->arch.exception.has_error_code = false;
395 vcpu->arch.exception.nr = nr;
397 EXPORT_SYMBOL_GPL(kvm_queue_exception);
399 void kvm_inject_page_fault(struct kvm_vcpu *vcpu, unsigned long addr,
402 ++vcpu->stat.pf_guest;
403 if (vcpu->arch.exception.pending) {
404 if (vcpu->arch.exception.nr == PF_VECTOR) {
405 printk(KERN_DEBUG "kvm: inject_page_fault:"
406 " double fault 0x%lx\n", addr);
407 vcpu->arch.exception.nr = DF_VECTOR;
408 vcpu->arch.exception.error_code = 0;
409 } else if (vcpu->arch.exception.nr == DF_VECTOR) {
410 /* triple fault -> shutdown */
411 set_bit(KVM_REQ_TRIPLE_FAULT, &vcpu->requests);
415 vcpu->arch.cr2 = addr;
416 kvm_queue_exception_e(vcpu, PF_VECTOR, error_code);
419 void kvm_inject_nmi(struct kvm_vcpu *vcpu)
421 vcpu->arch.nmi_pending = 1;
423 EXPORT_SYMBOL_GPL(kvm_inject_nmi);
425 void kvm_queue_exception_e(struct kvm_vcpu *vcpu, unsigned nr, u32 error_code)
427 WARN_ON(vcpu->arch.exception.pending);
428 vcpu->arch.exception.pending = true;
429 vcpu->arch.exception.has_error_code = true;
430 vcpu->arch.exception.nr = nr;
431 vcpu->arch.exception.error_code = error_code;
433 EXPORT_SYMBOL_GPL(kvm_queue_exception_e);
435 static void __queue_exception(struct kvm_vcpu *vcpu)
437 kvm_x86_ops->queue_exception(vcpu, vcpu->arch.exception.nr,
438 vcpu->arch.exception.has_error_code,
439 vcpu->arch.exception.error_code);
443 * Load the pae pdptrs. Return true is they are all valid.
445 int load_pdptrs(struct kvm_vcpu *vcpu, unsigned long cr3)
447 gfn_t pdpt_gfn = cr3 >> PAGE_SHIFT;
448 unsigned offset = ((cr3 & (PAGE_SIZE-1)) >> 5) << 2;
451 u64 pdpte[ARRAY_SIZE(vcpu->arch.pdptrs)];
453 ret = kvm_read_guest_page(vcpu->kvm, pdpt_gfn, pdpte,
454 offset * sizeof(u64), sizeof(pdpte));
459 for (i = 0; i < ARRAY_SIZE(pdpte); ++i) {
460 if ((pdpte[i] & 1) && (pdpte[i] & 0xfffffff0000001e6ull)) {
467 memcpy(vcpu->arch.pdptrs, pdpte, sizeof(vcpu->arch.pdptrs));
472 EXPORT_SYMBOL_GPL(load_pdptrs);
474 static bool pdptrs_changed(struct kvm_vcpu *vcpu)
476 u64 pdpte[ARRAY_SIZE(vcpu->arch.pdptrs)];
480 if (is_long_mode(vcpu) || !is_pae(vcpu))
483 r = kvm_read_guest(vcpu->kvm, vcpu->arch.cr3 & ~31u, pdpte, sizeof(pdpte));
486 changed = memcmp(pdpte, vcpu->arch.pdptrs, sizeof(pdpte)) != 0;
492 void kvm_set_cr0(struct kvm_vcpu *vcpu, unsigned long cr0)
494 if (cr0 & CR0_RESERVED_BITS) {
495 printk(KERN_DEBUG "set_cr0: 0x%lx #GP, reserved bits 0x%lx\n",
496 cr0, vcpu->arch.cr0);
497 kvm_inject_gp(vcpu, 0);
501 if ((cr0 & X86_CR0_NW) && !(cr0 & X86_CR0_CD)) {
502 printk(KERN_DEBUG "set_cr0: #GP, CD == 0 && NW == 1\n");
503 kvm_inject_gp(vcpu, 0);
507 if ((cr0 & X86_CR0_PG) && !(cr0 & X86_CR0_PE)) {
508 printk(KERN_DEBUG "set_cr0: #GP, set PG flag "
509 "and a clear PE flag\n");
510 kvm_inject_gp(vcpu, 0);
514 if (!is_paging(vcpu) && (cr0 & X86_CR0_PG)) {
516 if ((vcpu->arch.shadow_efer & EFER_LME)) {
520 printk(KERN_DEBUG "set_cr0: #GP, start paging "
521 "in long mode while PAE is disabled\n");
522 kvm_inject_gp(vcpu, 0);
525 kvm_x86_ops->get_cs_db_l_bits(vcpu, &cs_db, &cs_l);
527 printk(KERN_DEBUG "set_cr0: #GP, start paging "
528 "in long mode while CS.L == 1\n");
529 kvm_inject_gp(vcpu, 0);
535 if (is_pae(vcpu) && !load_pdptrs(vcpu, vcpu->arch.cr3)) {
536 printk(KERN_DEBUG "set_cr0: #GP, pdptrs "
538 kvm_inject_gp(vcpu, 0);
544 kvm_x86_ops->set_cr0(vcpu, cr0);
545 vcpu->arch.cr0 = cr0;
547 kvm_mmu_reset_context(vcpu);
550 EXPORT_SYMBOL_GPL(kvm_set_cr0);
552 void kvm_lmsw(struct kvm_vcpu *vcpu, unsigned long msw)
554 kvm_set_cr0(vcpu, (vcpu->arch.cr0 & ~0x0ful) | (msw & 0x0f));
555 KVMTRACE_1D(LMSW, vcpu,
556 (u32)((vcpu->arch.cr0 & ~0x0ful) | (msw & 0x0f)),
559 EXPORT_SYMBOL_GPL(kvm_lmsw);
561 void kvm_set_cr4(struct kvm_vcpu *vcpu, unsigned long cr4)
563 if (cr4 & CR4_RESERVED_BITS) {
564 printk(KERN_DEBUG "set_cr4: #GP, reserved bits\n");
565 kvm_inject_gp(vcpu, 0);
569 if (is_long_mode(vcpu)) {
570 if (!(cr4 & X86_CR4_PAE)) {
571 printk(KERN_DEBUG "set_cr4: #GP, clearing PAE while "
573 kvm_inject_gp(vcpu, 0);
576 } else if (is_paging(vcpu) && !is_pae(vcpu) && (cr4 & X86_CR4_PAE)
577 && !load_pdptrs(vcpu, vcpu->arch.cr3)) {
578 printk(KERN_DEBUG "set_cr4: #GP, pdptrs reserved bits\n");
579 kvm_inject_gp(vcpu, 0);
583 if (cr4 & X86_CR4_VMXE) {
584 printk(KERN_DEBUG "set_cr4: #GP, setting VMXE\n");
585 kvm_inject_gp(vcpu, 0);
588 kvm_x86_ops->set_cr4(vcpu, cr4);
589 vcpu->arch.cr4 = cr4;
590 kvm_mmu_reset_context(vcpu);
592 EXPORT_SYMBOL_GPL(kvm_set_cr4);
594 void kvm_set_cr3(struct kvm_vcpu *vcpu, unsigned long cr3)
596 if (cr3 == vcpu->arch.cr3 && !pdptrs_changed(vcpu)) {
597 kvm_mmu_sync_roots(vcpu);
598 kvm_mmu_flush_tlb(vcpu);
602 if (is_long_mode(vcpu)) {
603 if (cr3 & CR3_L_MODE_RESERVED_BITS) {
604 printk(KERN_DEBUG "set_cr3: #GP, reserved bits\n");
605 kvm_inject_gp(vcpu, 0);
610 if (cr3 & CR3_PAE_RESERVED_BITS) {
612 "set_cr3: #GP, reserved bits\n");
613 kvm_inject_gp(vcpu, 0);
616 if (is_paging(vcpu) && !load_pdptrs(vcpu, cr3)) {
617 printk(KERN_DEBUG "set_cr3: #GP, pdptrs "
619 kvm_inject_gp(vcpu, 0);
624 * We don't check reserved bits in nonpae mode, because
625 * this isn't enforced, and VMware depends on this.
630 * Does the new cr3 value map to physical memory? (Note, we
631 * catch an invalid cr3 even in real-mode, because it would
632 * cause trouble later on when we turn on paging anyway.)
634 * A real CPU would silently accept an invalid cr3 and would
635 * attempt to use it - with largely undefined (and often hard
636 * to debug) behavior on the guest side.
638 if (unlikely(!gfn_to_memslot(vcpu->kvm, cr3 >> PAGE_SHIFT)))
639 kvm_inject_gp(vcpu, 0);
641 vcpu->arch.cr3 = cr3;
642 vcpu->arch.mmu.new_cr3(vcpu);
645 EXPORT_SYMBOL_GPL(kvm_set_cr3);
647 void kvm_set_cr8(struct kvm_vcpu *vcpu, unsigned long cr8)
649 if (cr8 & CR8_RESERVED_BITS) {
650 printk(KERN_DEBUG "set_cr8: #GP, reserved bits 0x%lx\n", cr8);
651 kvm_inject_gp(vcpu, 0);
654 if (irqchip_in_kernel(vcpu->kvm))
655 kvm_lapic_set_tpr(vcpu, cr8);
657 vcpu->arch.cr8 = cr8;
659 EXPORT_SYMBOL_GPL(kvm_set_cr8);
661 unsigned long kvm_get_cr8(struct kvm_vcpu *vcpu)
663 if (irqchip_in_kernel(vcpu->kvm))
664 return kvm_lapic_get_cr8(vcpu);
666 return vcpu->arch.cr8;
668 EXPORT_SYMBOL_GPL(kvm_get_cr8);
671 * List of msr numbers which we expose to userspace through KVM_GET_MSRS
672 * and KVM_SET_MSRS, and KVM_GET_MSR_INDEX_LIST.
674 * This list is modified at module load time to reflect the
675 * capabilities of the host cpu.
677 static u32 msrs_to_save[] = {
678 MSR_IA32_SYSENTER_CS, MSR_IA32_SYSENTER_ESP, MSR_IA32_SYSENTER_EIP,
681 MSR_CSTAR, MSR_KERNEL_GS_BASE, MSR_SYSCALL_MASK, MSR_LSTAR,
683 MSR_IA32_TIME_STAMP_COUNTER, MSR_KVM_SYSTEM_TIME, MSR_KVM_WALL_CLOCK,
684 MSR_IA32_PERF_STATUS,
687 static unsigned num_msrs_to_save;
689 static u32 emulated_msrs[] = {
690 MSR_IA32_MISC_ENABLE,
693 static void set_efer(struct kvm_vcpu *vcpu, u64 efer)
695 if (efer & efer_reserved_bits) {
696 printk(KERN_DEBUG "set_efer: 0x%llx #GP, reserved bits\n",
698 kvm_inject_gp(vcpu, 0);
703 && (vcpu->arch.shadow_efer & EFER_LME) != (efer & EFER_LME)) {
704 printk(KERN_DEBUG "set_efer: #GP, change LME while paging\n");
705 kvm_inject_gp(vcpu, 0);
709 kvm_x86_ops->set_efer(vcpu, efer);
712 efer |= vcpu->arch.shadow_efer & EFER_LMA;
714 vcpu->arch.shadow_efer = efer;
717 void kvm_enable_efer_bits(u64 mask)
719 efer_reserved_bits &= ~mask;
721 EXPORT_SYMBOL_GPL(kvm_enable_efer_bits);
725 * Writes msr value into into the appropriate "register".
726 * Returns 0 on success, non-0 otherwise.
727 * Assumes vcpu_load() was already called.
729 int kvm_set_msr(struct kvm_vcpu *vcpu, u32 msr_index, u64 data)
731 return kvm_x86_ops->set_msr(vcpu, msr_index, data);
735 * Adapt set_msr() to msr_io()'s calling convention
737 static int do_set_msr(struct kvm_vcpu *vcpu, unsigned index, u64 *data)
739 return kvm_set_msr(vcpu, index, *data);
742 static void kvm_write_wall_clock(struct kvm *kvm, gpa_t wall_clock)
745 struct pvclock_wall_clock wc;
746 struct timespec now, sys, boot;
753 kvm_write_guest(kvm, wall_clock, &version, sizeof(version));
756 * The guest calculates current wall clock time by adding
757 * system time (updated by kvm_write_guest_time below) to the
758 * wall clock specified here. guest system time equals host
759 * system time for us, thus we must fill in host boot time here.
761 now = current_kernel_time();
763 boot = ns_to_timespec(timespec_to_ns(&now) - timespec_to_ns(&sys));
765 wc.sec = boot.tv_sec;
766 wc.nsec = boot.tv_nsec;
767 wc.version = version;
769 kvm_write_guest(kvm, wall_clock, &wc, sizeof(wc));
772 kvm_write_guest(kvm, wall_clock, &version, sizeof(version));
775 static uint32_t div_frac(uint32_t dividend, uint32_t divisor)
777 uint32_t quotient, remainder;
779 /* Don't try to replace with do_div(), this one calculates
780 * "(dividend << 32) / divisor" */
782 : "=a" (quotient), "=d" (remainder)
783 : "0" (0), "1" (dividend), "r" (divisor) );
787 static void kvm_set_time_scale(uint32_t tsc_khz, struct pvclock_vcpu_time_info *hv_clock)
789 uint64_t nsecs = 1000000000LL;
794 tps64 = tsc_khz * 1000LL;
795 while (tps64 > nsecs*2) {
800 tps32 = (uint32_t)tps64;
801 while (tps32 <= (uint32_t)nsecs) {
806 hv_clock->tsc_shift = shift;
807 hv_clock->tsc_to_system_mul = div_frac(nsecs, tps32);
809 pr_debug("%s: tsc_khz %u, tsc_shift %d, tsc_mul %u\n",
810 __FUNCTION__, tsc_khz, hv_clock->tsc_shift,
811 hv_clock->tsc_to_system_mul);
814 static void kvm_write_guest_time(struct kvm_vcpu *v)
818 struct kvm_vcpu_arch *vcpu = &v->arch;
821 if ((!vcpu->time_page))
824 if (unlikely(vcpu->hv_clock_tsc_khz != tsc_khz)) {
825 kvm_set_time_scale(tsc_khz, &vcpu->hv_clock);
826 vcpu->hv_clock_tsc_khz = tsc_khz;
829 /* Keep irq disabled to prevent changes to the clock */
830 local_irq_save(flags);
831 kvm_get_msr(v, MSR_IA32_TIME_STAMP_COUNTER,
832 &vcpu->hv_clock.tsc_timestamp);
834 local_irq_restore(flags);
836 /* With all the info we got, fill in the values */
838 vcpu->hv_clock.system_time = ts.tv_nsec +
839 (NSEC_PER_SEC * (u64)ts.tv_sec);
841 * The interface expects us to write an even number signaling that the
842 * update is finished. Since the guest won't see the intermediate
843 * state, we just increase by 2 at the end.
845 vcpu->hv_clock.version += 2;
847 shared_kaddr = kmap_atomic(vcpu->time_page, KM_USER0);
849 memcpy(shared_kaddr + vcpu->time_offset, &vcpu->hv_clock,
850 sizeof(vcpu->hv_clock));
852 kunmap_atomic(shared_kaddr, KM_USER0);
854 mark_page_dirty(v->kvm, vcpu->time >> PAGE_SHIFT);
857 static bool msr_mtrr_valid(unsigned msr)
860 case 0x200 ... 0x200 + 2 * KVM_NR_VAR_MTRR - 1:
861 case MSR_MTRRfix64K_00000:
862 case MSR_MTRRfix16K_80000:
863 case MSR_MTRRfix16K_A0000:
864 case MSR_MTRRfix4K_C0000:
865 case MSR_MTRRfix4K_C8000:
866 case MSR_MTRRfix4K_D0000:
867 case MSR_MTRRfix4K_D8000:
868 case MSR_MTRRfix4K_E0000:
869 case MSR_MTRRfix4K_E8000:
870 case MSR_MTRRfix4K_F0000:
871 case MSR_MTRRfix4K_F8000:
872 case MSR_MTRRdefType:
873 case MSR_IA32_CR_PAT:
881 static int set_msr_mtrr(struct kvm_vcpu *vcpu, u32 msr, u64 data)
883 if (!msr_mtrr_valid(msr))
886 vcpu->arch.mtrr[msr - 0x200] = data;
890 int kvm_set_msr_common(struct kvm_vcpu *vcpu, u32 msr, u64 data)
894 set_efer(vcpu, data);
896 case MSR_IA32_MC0_STATUS:
897 pr_unimpl(vcpu, "%s: MSR_IA32_MC0_STATUS 0x%llx, nop\n",
900 case MSR_IA32_MCG_STATUS:
901 pr_unimpl(vcpu, "%s: MSR_IA32_MCG_STATUS 0x%llx, nop\n",
904 case MSR_IA32_MCG_CTL:
905 pr_unimpl(vcpu, "%s: MSR_IA32_MCG_CTL 0x%llx, nop\n",
908 case MSR_IA32_DEBUGCTLMSR:
910 /* We support the non-activated case already */
912 } else if (data & ~(DEBUGCTLMSR_LBR | DEBUGCTLMSR_BTF)) {
913 /* Values other than LBR and BTF are vendor-specific,
914 thus reserved and should throw a #GP */
917 pr_unimpl(vcpu, "%s: MSR_IA32_DEBUGCTLMSR 0x%llx, nop\n",
920 case MSR_IA32_UCODE_REV:
921 case MSR_IA32_UCODE_WRITE:
923 case 0x200 ... 0x2ff:
924 return set_msr_mtrr(vcpu, msr, data);
925 case MSR_IA32_APICBASE:
926 kvm_set_apic_base(vcpu, data);
928 case MSR_IA32_MISC_ENABLE:
929 vcpu->arch.ia32_misc_enable_msr = data;
931 case MSR_KVM_WALL_CLOCK:
932 vcpu->kvm->arch.wall_clock = data;
933 kvm_write_wall_clock(vcpu->kvm, data);
935 case MSR_KVM_SYSTEM_TIME: {
936 if (vcpu->arch.time_page) {
937 kvm_release_page_dirty(vcpu->arch.time_page);
938 vcpu->arch.time_page = NULL;
941 vcpu->arch.time = data;
943 /* we verify if the enable bit is set... */
947 /* ...but clean it before doing the actual write */
948 vcpu->arch.time_offset = data & ~(PAGE_MASK | 1);
950 vcpu->arch.time_page =
951 gfn_to_page(vcpu->kvm, data >> PAGE_SHIFT);
953 if (is_error_page(vcpu->arch.time_page)) {
954 kvm_release_page_clean(vcpu->arch.time_page);
955 vcpu->arch.time_page = NULL;
958 kvm_write_guest_time(vcpu);
962 pr_unimpl(vcpu, "unhandled wrmsr: 0x%x data %llx\n", msr, data);
967 EXPORT_SYMBOL_GPL(kvm_set_msr_common);
971 * Reads an msr value (of 'msr_index') into 'pdata'.
972 * Returns 0 on success, non-0 otherwise.
973 * Assumes vcpu_load() was already called.
975 int kvm_get_msr(struct kvm_vcpu *vcpu, u32 msr_index, u64 *pdata)
977 return kvm_x86_ops->get_msr(vcpu, msr_index, pdata);
980 static int get_msr_mtrr(struct kvm_vcpu *vcpu, u32 msr, u64 *pdata)
982 if (!msr_mtrr_valid(msr))
985 *pdata = vcpu->arch.mtrr[msr - 0x200];
989 int kvm_get_msr_common(struct kvm_vcpu *vcpu, u32 msr, u64 *pdata)
994 case 0xc0010010: /* SYSCFG */
995 case 0xc0010015: /* HWCR */
996 case MSR_IA32_PLATFORM_ID:
997 case MSR_IA32_P5_MC_ADDR:
998 case MSR_IA32_P5_MC_TYPE:
999 case MSR_IA32_MC0_CTL:
1000 case MSR_IA32_MCG_STATUS:
1001 case MSR_IA32_MCG_CAP:
1002 case MSR_IA32_MCG_CTL:
1003 case MSR_IA32_MC0_MISC:
1004 case MSR_IA32_MC0_MISC+4:
1005 case MSR_IA32_MC0_MISC+8:
1006 case MSR_IA32_MC0_MISC+12:
1007 case MSR_IA32_MC0_MISC+16:
1008 case MSR_IA32_MC0_MISC+20:
1009 case MSR_IA32_UCODE_REV:
1010 case MSR_IA32_EBL_CR_POWERON:
1011 case MSR_IA32_DEBUGCTLMSR:
1012 case MSR_IA32_LASTBRANCHFROMIP:
1013 case MSR_IA32_LASTBRANCHTOIP:
1014 case MSR_IA32_LASTINTFROMIP:
1015 case MSR_IA32_LASTINTTOIP:
1019 data = 0x500 | KVM_NR_VAR_MTRR;
1021 case 0x200 ... 0x2ff:
1022 return get_msr_mtrr(vcpu, msr, pdata);
1023 case 0xcd: /* fsb frequency */
1026 case MSR_IA32_APICBASE:
1027 data = kvm_get_apic_base(vcpu);
1029 case MSR_IA32_MISC_ENABLE:
1030 data = vcpu->arch.ia32_misc_enable_msr;
1032 case MSR_IA32_PERF_STATUS:
1033 /* TSC increment by tick */
1035 /* CPU multiplier */
1036 data |= (((uint64_t)4ULL) << 40);
1039 data = vcpu->arch.shadow_efer;
1041 case MSR_KVM_WALL_CLOCK:
1042 data = vcpu->kvm->arch.wall_clock;
1044 case MSR_KVM_SYSTEM_TIME:
1045 data = vcpu->arch.time;
1048 pr_unimpl(vcpu, "unhandled rdmsr: 0x%x\n", msr);
1054 EXPORT_SYMBOL_GPL(kvm_get_msr_common);
1057 * Read or write a bunch of msrs. All parameters are kernel addresses.
1059 * @return number of msrs set successfully.
1061 static int __msr_io(struct kvm_vcpu *vcpu, struct kvm_msrs *msrs,
1062 struct kvm_msr_entry *entries,
1063 int (*do_msr)(struct kvm_vcpu *vcpu,
1064 unsigned index, u64 *data))
1070 down_read(&vcpu->kvm->slots_lock);
1071 for (i = 0; i < msrs->nmsrs; ++i)
1072 if (do_msr(vcpu, entries[i].index, &entries[i].data))
1074 up_read(&vcpu->kvm->slots_lock);
1082 * Read or write a bunch of msrs. Parameters are user addresses.
1084 * @return number of msrs set successfully.
1086 static int msr_io(struct kvm_vcpu *vcpu, struct kvm_msrs __user *user_msrs,
1087 int (*do_msr)(struct kvm_vcpu *vcpu,
1088 unsigned index, u64 *data),
1091 struct kvm_msrs msrs;
1092 struct kvm_msr_entry *entries;
1097 if (copy_from_user(&msrs, user_msrs, sizeof msrs))
1101 if (msrs.nmsrs >= MAX_IO_MSRS)
1105 size = sizeof(struct kvm_msr_entry) * msrs.nmsrs;
1106 entries = vmalloc(size);
1111 if (copy_from_user(entries, user_msrs->entries, size))
1114 r = n = __msr_io(vcpu, &msrs, entries, do_msr);
1119 if (writeback && copy_to_user(user_msrs->entries, entries, size))
1130 int kvm_dev_ioctl_check_extension(long ext)
1135 case KVM_CAP_IRQCHIP:
1137 case KVM_CAP_MMU_SHADOW_CACHE_CONTROL:
1138 case KVM_CAP_USER_MEMORY:
1139 case KVM_CAP_SET_TSS_ADDR:
1140 case KVM_CAP_EXT_CPUID:
1141 case KVM_CAP_CLOCKSOURCE:
1143 case KVM_CAP_NOP_IO_DELAY:
1144 case KVM_CAP_MP_STATE:
1145 case KVM_CAP_SYNC_MMU:
1148 case KVM_CAP_COALESCED_MMIO:
1149 r = KVM_COALESCED_MMIO_PAGE_OFFSET;
1152 r = !kvm_x86_ops->cpu_has_accelerated_tpr();
1154 case KVM_CAP_NR_VCPUS:
1157 case KVM_CAP_NR_MEMSLOTS:
1158 r = KVM_MEMORY_SLOTS;
1160 case KVM_CAP_PV_MMU:
1164 r = intel_iommu_found();
1174 long kvm_arch_dev_ioctl(struct file *filp,
1175 unsigned int ioctl, unsigned long arg)
1177 void __user *argp = (void __user *)arg;
1181 case KVM_GET_MSR_INDEX_LIST: {
1182 struct kvm_msr_list __user *user_msr_list = argp;
1183 struct kvm_msr_list msr_list;
1187 if (copy_from_user(&msr_list, user_msr_list, sizeof msr_list))
1190 msr_list.nmsrs = num_msrs_to_save + ARRAY_SIZE(emulated_msrs);
1191 if (copy_to_user(user_msr_list, &msr_list, sizeof msr_list))
1194 if (n < num_msrs_to_save)
1197 if (copy_to_user(user_msr_list->indices, &msrs_to_save,
1198 num_msrs_to_save * sizeof(u32)))
1200 if (copy_to_user(user_msr_list->indices
1201 + num_msrs_to_save * sizeof(u32),
1203 ARRAY_SIZE(emulated_msrs) * sizeof(u32)))
1208 case KVM_GET_SUPPORTED_CPUID: {
1209 struct kvm_cpuid2 __user *cpuid_arg = argp;
1210 struct kvm_cpuid2 cpuid;
1213 if (copy_from_user(&cpuid, cpuid_arg, sizeof cpuid))
1215 r = kvm_dev_ioctl_get_supported_cpuid(&cpuid,
1216 cpuid_arg->entries);
1221 if (copy_to_user(cpuid_arg, &cpuid, sizeof cpuid))
1233 void kvm_arch_vcpu_load(struct kvm_vcpu *vcpu, int cpu)
1235 kvm_x86_ops->vcpu_load(vcpu, cpu);
1236 kvm_write_guest_time(vcpu);
1239 void kvm_arch_vcpu_put(struct kvm_vcpu *vcpu)
1241 kvm_x86_ops->vcpu_put(vcpu);
1242 kvm_put_guest_fpu(vcpu);
1245 static int is_efer_nx(void)
1249 rdmsrl(MSR_EFER, efer);
1250 return efer & EFER_NX;
1253 static void cpuid_fix_nx_cap(struct kvm_vcpu *vcpu)
1256 struct kvm_cpuid_entry2 *e, *entry;
1259 for (i = 0; i < vcpu->arch.cpuid_nent; ++i) {
1260 e = &vcpu->arch.cpuid_entries[i];
1261 if (e->function == 0x80000001) {
1266 if (entry && (entry->edx & (1 << 20)) && !is_efer_nx()) {
1267 entry->edx &= ~(1 << 20);
1268 printk(KERN_INFO "kvm: guest NX capability removed\n");
1272 /* when an old userspace process fills a new kernel module */
1273 static int kvm_vcpu_ioctl_set_cpuid(struct kvm_vcpu *vcpu,
1274 struct kvm_cpuid *cpuid,
1275 struct kvm_cpuid_entry __user *entries)
1278 struct kvm_cpuid_entry *cpuid_entries;
1281 if (cpuid->nent > KVM_MAX_CPUID_ENTRIES)
1284 cpuid_entries = vmalloc(sizeof(struct kvm_cpuid_entry) * cpuid->nent);
1288 if (copy_from_user(cpuid_entries, entries,
1289 cpuid->nent * sizeof(struct kvm_cpuid_entry)))
1291 for (i = 0; i < cpuid->nent; i++) {
1292 vcpu->arch.cpuid_entries[i].function = cpuid_entries[i].function;
1293 vcpu->arch.cpuid_entries[i].eax = cpuid_entries[i].eax;
1294 vcpu->arch.cpuid_entries[i].ebx = cpuid_entries[i].ebx;
1295 vcpu->arch.cpuid_entries[i].ecx = cpuid_entries[i].ecx;
1296 vcpu->arch.cpuid_entries[i].edx = cpuid_entries[i].edx;
1297 vcpu->arch.cpuid_entries[i].index = 0;
1298 vcpu->arch.cpuid_entries[i].flags = 0;
1299 vcpu->arch.cpuid_entries[i].padding[0] = 0;
1300 vcpu->arch.cpuid_entries[i].padding[1] = 0;
1301 vcpu->arch.cpuid_entries[i].padding[2] = 0;
1303 vcpu->arch.cpuid_nent = cpuid->nent;
1304 cpuid_fix_nx_cap(vcpu);
1308 vfree(cpuid_entries);
1313 static int kvm_vcpu_ioctl_set_cpuid2(struct kvm_vcpu *vcpu,
1314 struct kvm_cpuid2 *cpuid,
1315 struct kvm_cpuid_entry2 __user *entries)
1320 if (cpuid->nent > KVM_MAX_CPUID_ENTRIES)
1323 if (copy_from_user(&vcpu->arch.cpuid_entries, entries,
1324 cpuid->nent * sizeof(struct kvm_cpuid_entry2)))
1326 vcpu->arch.cpuid_nent = cpuid->nent;
1333 static int kvm_vcpu_ioctl_get_cpuid2(struct kvm_vcpu *vcpu,
1334 struct kvm_cpuid2 *cpuid,
1335 struct kvm_cpuid_entry2 __user *entries)
1340 if (cpuid->nent < vcpu->arch.cpuid_nent)
1343 if (copy_to_user(entries, &vcpu->arch.cpuid_entries,
1344 vcpu->arch.cpuid_nent * sizeof(struct kvm_cpuid_entry2)))
1349 cpuid->nent = vcpu->arch.cpuid_nent;
1353 static inline u32 bit(int bitno)
1355 return 1 << (bitno & 31);
1358 static void do_cpuid_1_ent(struct kvm_cpuid_entry2 *entry, u32 function,
1361 entry->function = function;
1362 entry->index = index;
1363 cpuid_count(entry->function, entry->index,
1364 &entry->eax, &entry->ebx, &entry->ecx, &entry->edx);
1368 static void do_cpuid_ent(struct kvm_cpuid_entry2 *entry, u32 function,
1369 u32 index, int *nent, int maxnent)
1371 const u32 kvm_supported_word0_x86_features = bit(X86_FEATURE_FPU) |
1372 bit(X86_FEATURE_VME) | bit(X86_FEATURE_DE) |
1373 bit(X86_FEATURE_PSE) | bit(X86_FEATURE_TSC) |
1374 bit(X86_FEATURE_MSR) | bit(X86_FEATURE_PAE) |
1375 bit(X86_FEATURE_CX8) | bit(X86_FEATURE_APIC) |
1376 bit(X86_FEATURE_SEP) | bit(X86_FEATURE_PGE) |
1377 bit(X86_FEATURE_CMOV) | bit(X86_FEATURE_PSE36) |
1378 bit(X86_FEATURE_CLFLSH) | bit(X86_FEATURE_MMX) |
1379 bit(X86_FEATURE_FXSR) | bit(X86_FEATURE_XMM) |
1380 bit(X86_FEATURE_XMM2) | bit(X86_FEATURE_SELFSNOOP);
1381 const u32 kvm_supported_word1_x86_features = bit(X86_FEATURE_FPU) |
1382 bit(X86_FEATURE_VME) | bit(X86_FEATURE_DE) |
1383 bit(X86_FEATURE_PSE) | bit(X86_FEATURE_TSC) |
1384 bit(X86_FEATURE_MSR) | bit(X86_FEATURE_PAE) |
1385 bit(X86_FEATURE_CX8) | bit(X86_FEATURE_APIC) |
1386 bit(X86_FEATURE_PGE) |
1387 bit(X86_FEATURE_CMOV) | bit(X86_FEATURE_PSE36) |
1388 bit(X86_FEATURE_MMX) | bit(X86_FEATURE_FXSR) |
1389 bit(X86_FEATURE_SYSCALL) |
1390 (bit(X86_FEATURE_NX) && is_efer_nx()) |
1391 #ifdef CONFIG_X86_64
1392 bit(X86_FEATURE_LM) |
1394 bit(X86_FEATURE_MMXEXT) |
1395 bit(X86_FEATURE_3DNOWEXT) |
1396 bit(X86_FEATURE_3DNOW);
1397 const u32 kvm_supported_word3_x86_features =
1398 bit(X86_FEATURE_XMM3) | bit(X86_FEATURE_CX16);
1399 const u32 kvm_supported_word6_x86_features =
1400 bit(X86_FEATURE_LAHF_LM) | bit(X86_FEATURE_CMP_LEGACY);
1402 /* all func 2 cpuid_count() should be called on the same cpu */
1404 do_cpuid_1_ent(entry, function, index);
1409 entry->eax = min(entry->eax, (u32)0xb);
1412 entry->edx &= kvm_supported_word0_x86_features;
1413 entry->ecx &= kvm_supported_word3_x86_features;
1415 /* function 2 entries are STATEFUL. That is, repeated cpuid commands
1416 * may return different values. This forces us to get_cpu() before
1417 * issuing the first command, and also to emulate this annoying behavior
1418 * in kvm_emulate_cpuid() using KVM_CPUID_FLAG_STATE_READ_NEXT */
1420 int t, times = entry->eax & 0xff;
1422 entry->flags |= KVM_CPUID_FLAG_STATEFUL_FUNC;
1423 for (t = 1; t < times && *nent < maxnent; ++t) {
1424 do_cpuid_1_ent(&entry[t], function, 0);
1425 entry[t].flags |= KVM_CPUID_FLAG_STATEFUL_FUNC;
1430 /* function 4 and 0xb have additional index. */
1434 entry->flags |= KVM_CPUID_FLAG_SIGNIFCANT_INDEX;
1435 /* read more entries until cache_type is zero */
1436 for (i = 1; *nent < maxnent; ++i) {
1437 cache_type = entry[i - 1].eax & 0x1f;
1440 do_cpuid_1_ent(&entry[i], function, i);
1442 KVM_CPUID_FLAG_SIGNIFCANT_INDEX;
1450 entry->flags |= KVM_CPUID_FLAG_SIGNIFCANT_INDEX;
1451 /* read more entries until level_type is zero */
1452 for (i = 1; *nent < maxnent; ++i) {
1453 level_type = entry[i - 1].ecx & 0xff;
1456 do_cpuid_1_ent(&entry[i], function, i);
1458 KVM_CPUID_FLAG_SIGNIFCANT_INDEX;
1464 entry->eax = min(entry->eax, 0x8000001a);
1467 entry->edx &= kvm_supported_word1_x86_features;
1468 entry->ecx &= kvm_supported_word6_x86_features;
1474 static int kvm_dev_ioctl_get_supported_cpuid(struct kvm_cpuid2 *cpuid,
1475 struct kvm_cpuid_entry2 __user *entries)
1477 struct kvm_cpuid_entry2 *cpuid_entries;
1478 int limit, nent = 0, r = -E2BIG;
1481 if (cpuid->nent < 1)
1484 cpuid_entries = vmalloc(sizeof(struct kvm_cpuid_entry2) * cpuid->nent);
1488 do_cpuid_ent(&cpuid_entries[0], 0, 0, &nent, cpuid->nent);
1489 limit = cpuid_entries[0].eax;
1490 for (func = 1; func <= limit && nent < cpuid->nent; ++func)
1491 do_cpuid_ent(&cpuid_entries[nent], func, 0,
1492 &nent, cpuid->nent);
1494 if (nent >= cpuid->nent)
1497 do_cpuid_ent(&cpuid_entries[nent], 0x80000000, 0, &nent, cpuid->nent);
1498 limit = cpuid_entries[nent - 1].eax;
1499 for (func = 0x80000001; func <= limit && nent < cpuid->nent; ++func)
1500 do_cpuid_ent(&cpuid_entries[nent], func, 0,
1501 &nent, cpuid->nent);
1503 if (copy_to_user(entries, cpuid_entries,
1504 nent * sizeof(struct kvm_cpuid_entry2)))
1510 vfree(cpuid_entries);
1515 static int kvm_vcpu_ioctl_get_lapic(struct kvm_vcpu *vcpu,
1516 struct kvm_lapic_state *s)
1519 memcpy(s->regs, vcpu->arch.apic->regs, sizeof *s);
1525 static int kvm_vcpu_ioctl_set_lapic(struct kvm_vcpu *vcpu,
1526 struct kvm_lapic_state *s)
1529 memcpy(vcpu->arch.apic->regs, s->regs, sizeof *s);
1530 kvm_apic_post_state_restore(vcpu);
1536 static int kvm_vcpu_ioctl_interrupt(struct kvm_vcpu *vcpu,
1537 struct kvm_interrupt *irq)
1539 if (irq->irq < 0 || irq->irq >= 256)
1541 if (irqchip_in_kernel(vcpu->kvm))
1545 set_bit(irq->irq, vcpu->arch.irq_pending);
1546 set_bit(irq->irq / BITS_PER_LONG, &vcpu->arch.irq_summary);
1553 static int vcpu_ioctl_tpr_access_reporting(struct kvm_vcpu *vcpu,
1554 struct kvm_tpr_access_ctl *tac)
1558 vcpu->arch.tpr_access_reporting = !!tac->enabled;
1562 long kvm_arch_vcpu_ioctl(struct file *filp,
1563 unsigned int ioctl, unsigned long arg)
1565 struct kvm_vcpu *vcpu = filp->private_data;
1566 void __user *argp = (void __user *)arg;
1568 struct kvm_lapic_state *lapic = NULL;
1571 case KVM_GET_LAPIC: {
1572 lapic = kzalloc(sizeof(struct kvm_lapic_state), GFP_KERNEL);
1577 r = kvm_vcpu_ioctl_get_lapic(vcpu, lapic);
1581 if (copy_to_user(argp, lapic, sizeof(struct kvm_lapic_state)))
1586 case KVM_SET_LAPIC: {
1587 lapic = kmalloc(sizeof(struct kvm_lapic_state), GFP_KERNEL);
1592 if (copy_from_user(lapic, argp, sizeof(struct kvm_lapic_state)))
1594 r = kvm_vcpu_ioctl_set_lapic(vcpu, lapic);
1600 case KVM_INTERRUPT: {
1601 struct kvm_interrupt irq;
1604 if (copy_from_user(&irq, argp, sizeof irq))
1606 r = kvm_vcpu_ioctl_interrupt(vcpu, &irq);
1612 case KVM_SET_CPUID: {
1613 struct kvm_cpuid __user *cpuid_arg = argp;
1614 struct kvm_cpuid cpuid;
1617 if (copy_from_user(&cpuid, cpuid_arg, sizeof cpuid))
1619 r = kvm_vcpu_ioctl_set_cpuid(vcpu, &cpuid, cpuid_arg->entries);
1624 case KVM_SET_CPUID2: {
1625 struct kvm_cpuid2 __user *cpuid_arg = argp;
1626 struct kvm_cpuid2 cpuid;
1629 if (copy_from_user(&cpuid, cpuid_arg, sizeof cpuid))
1631 r = kvm_vcpu_ioctl_set_cpuid2(vcpu, &cpuid,
1632 cpuid_arg->entries);
1637 case KVM_GET_CPUID2: {
1638 struct kvm_cpuid2 __user *cpuid_arg = argp;
1639 struct kvm_cpuid2 cpuid;
1642 if (copy_from_user(&cpuid, cpuid_arg, sizeof cpuid))
1644 r = kvm_vcpu_ioctl_get_cpuid2(vcpu, &cpuid,
1645 cpuid_arg->entries);
1649 if (copy_to_user(cpuid_arg, &cpuid, sizeof cpuid))
1655 r = msr_io(vcpu, argp, kvm_get_msr, 1);
1658 r = msr_io(vcpu, argp, do_set_msr, 0);
1660 case KVM_TPR_ACCESS_REPORTING: {
1661 struct kvm_tpr_access_ctl tac;
1664 if (copy_from_user(&tac, argp, sizeof tac))
1666 r = vcpu_ioctl_tpr_access_reporting(vcpu, &tac);
1670 if (copy_to_user(argp, &tac, sizeof tac))
1675 case KVM_SET_VAPIC_ADDR: {
1676 struct kvm_vapic_addr va;
1679 if (!irqchip_in_kernel(vcpu->kvm))
1682 if (copy_from_user(&va, argp, sizeof va))
1685 kvm_lapic_set_vapic_addr(vcpu, va.vapic_addr);
1697 static int kvm_vm_ioctl_set_tss_addr(struct kvm *kvm, unsigned long addr)
1701 if (addr > (unsigned int)(-3 * PAGE_SIZE))
1703 ret = kvm_x86_ops->set_tss_addr(kvm, addr);
1707 static int kvm_vm_ioctl_set_nr_mmu_pages(struct kvm *kvm,
1708 u32 kvm_nr_mmu_pages)
1710 if (kvm_nr_mmu_pages < KVM_MIN_ALLOC_MMU_PAGES)
1713 down_write(&kvm->slots_lock);
1715 kvm_mmu_change_mmu_pages(kvm, kvm_nr_mmu_pages);
1716 kvm->arch.n_requested_mmu_pages = kvm_nr_mmu_pages;
1718 up_write(&kvm->slots_lock);
1722 static int kvm_vm_ioctl_get_nr_mmu_pages(struct kvm *kvm)
1724 return kvm->arch.n_alloc_mmu_pages;
1727 gfn_t unalias_gfn(struct kvm *kvm, gfn_t gfn)
1730 struct kvm_mem_alias *alias;
1732 for (i = 0; i < kvm->arch.naliases; ++i) {
1733 alias = &kvm->arch.aliases[i];
1734 if (gfn >= alias->base_gfn
1735 && gfn < alias->base_gfn + alias->npages)
1736 return alias->target_gfn + gfn - alias->base_gfn;
1742 * Set a new alias region. Aliases map a portion of physical memory into
1743 * another portion. This is useful for memory windows, for example the PC
1746 static int kvm_vm_ioctl_set_memory_alias(struct kvm *kvm,
1747 struct kvm_memory_alias *alias)
1750 struct kvm_mem_alias *p;
1753 /* General sanity checks */
1754 if (alias->memory_size & (PAGE_SIZE - 1))
1756 if (alias->guest_phys_addr & (PAGE_SIZE - 1))
1758 if (alias->slot >= KVM_ALIAS_SLOTS)
1760 if (alias->guest_phys_addr + alias->memory_size
1761 < alias->guest_phys_addr)
1763 if (alias->target_phys_addr + alias->memory_size
1764 < alias->target_phys_addr)
1767 down_write(&kvm->slots_lock);
1768 spin_lock(&kvm->mmu_lock);
1770 p = &kvm->arch.aliases[alias->slot];
1771 p->base_gfn = alias->guest_phys_addr >> PAGE_SHIFT;
1772 p->npages = alias->memory_size >> PAGE_SHIFT;
1773 p->target_gfn = alias->target_phys_addr >> PAGE_SHIFT;
1775 for (n = KVM_ALIAS_SLOTS; n > 0; --n)
1776 if (kvm->arch.aliases[n - 1].npages)
1778 kvm->arch.naliases = n;
1780 spin_unlock(&kvm->mmu_lock);
1781 kvm_mmu_zap_all(kvm);
1783 up_write(&kvm->slots_lock);
1791 static int kvm_vm_ioctl_get_irqchip(struct kvm *kvm, struct kvm_irqchip *chip)
1796 switch (chip->chip_id) {
1797 case KVM_IRQCHIP_PIC_MASTER:
1798 memcpy(&chip->chip.pic,
1799 &pic_irqchip(kvm)->pics[0],
1800 sizeof(struct kvm_pic_state));
1802 case KVM_IRQCHIP_PIC_SLAVE:
1803 memcpy(&chip->chip.pic,
1804 &pic_irqchip(kvm)->pics[1],
1805 sizeof(struct kvm_pic_state));
1807 case KVM_IRQCHIP_IOAPIC:
1808 memcpy(&chip->chip.ioapic,
1809 ioapic_irqchip(kvm),
1810 sizeof(struct kvm_ioapic_state));
1819 static int kvm_vm_ioctl_set_irqchip(struct kvm *kvm, struct kvm_irqchip *chip)
1824 switch (chip->chip_id) {
1825 case KVM_IRQCHIP_PIC_MASTER:
1826 memcpy(&pic_irqchip(kvm)->pics[0],
1828 sizeof(struct kvm_pic_state));
1830 case KVM_IRQCHIP_PIC_SLAVE:
1831 memcpy(&pic_irqchip(kvm)->pics[1],
1833 sizeof(struct kvm_pic_state));
1835 case KVM_IRQCHIP_IOAPIC:
1836 memcpy(ioapic_irqchip(kvm),
1838 sizeof(struct kvm_ioapic_state));
1844 kvm_pic_update_irq(pic_irqchip(kvm));
1848 static int kvm_vm_ioctl_get_pit(struct kvm *kvm, struct kvm_pit_state *ps)
1852 memcpy(ps, &kvm->arch.vpit->pit_state, sizeof(struct kvm_pit_state));
1856 static int kvm_vm_ioctl_set_pit(struct kvm *kvm, struct kvm_pit_state *ps)
1860 memcpy(&kvm->arch.vpit->pit_state, ps, sizeof(struct kvm_pit_state));
1861 kvm_pit_load_count(kvm, 0, ps->channels[0].count);
1866 * Get (and clear) the dirty memory log for a memory slot.
1868 int kvm_vm_ioctl_get_dirty_log(struct kvm *kvm,
1869 struct kvm_dirty_log *log)
1873 struct kvm_memory_slot *memslot;
1876 down_write(&kvm->slots_lock);
1878 r = kvm_get_dirty_log(kvm, log, &is_dirty);
1882 /* If nothing is dirty, don't bother messing with page tables. */
1884 kvm_mmu_slot_remove_write_access(kvm, log->slot);
1885 kvm_flush_remote_tlbs(kvm);
1886 memslot = &kvm->memslots[log->slot];
1887 n = ALIGN(memslot->npages, BITS_PER_LONG) / 8;
1888 memset(memslot->dirty_bitmap, 0, n);
1892 up_write(&kvm->slots_lock);
1896 long kvm_arch_vm_ioctl(struct file *filp,
1897 unsigned int ioctl, unsigned long arg)
1899 struct kvm *kvm = filp->private_data;
1900 void __user *argp = (void __user *)arg;
1903 * This union makes it completely explicit to gcc-3.x
1904 * that these two variables' stack usage should be
1905 * combined, not added together.
1908 struct kvm_pit_state ps;
1909 struct kvm_memory_alias alias;
1913 case KVM_SET_TSS_ADDR:
1914 r = kvm_vm_ioctl_set_tss_addr(kvm, arg);
1918 case KVM_SET_MEMORY_REGION: {
1919 struct kvm_memory_region kvm_mem;
1920 struct kvm_userspace_memory_region kvm_userspace_mem;
1923 if (copy_from_user(&kvm_mem, argp, sizeof kvm_mem))
1925 kvm_userspace_mem.slot = kvm_mem.slot;
1926 kvm_userspace_mem.flags = kvm_mem.flags;
1927 kvm_userspace_mem.guest_phys_addr = kvm_mem.guest_phys_addr;
1928 kvm_userspace_mem.memory_size = kvm_mem.memory_size;
1929 r = kvm_vm_ioctl_set_memory_region(kvm, &kvm_userspace_mem, 0);
1934 case KVM_SET_NR_MMU_PAGES:
1935 r = kvm_vm_ioctl_set_nr_mmu_pages(kvm, arg);
1939 case KVM_GET_NR_MMU_PAGES:
1940 r = kvm_vm_ioctl_get_nr_mmu_pages(kvm);
1942 case KVM_SET_MEMORY_ALIAS:
1944 if (copy_from_user(&u.alias, argp, sizeof(struct kvm_memory_alias)))
1946 r = kvm_vm_ioctl_set_memory_alias(kvm, &u.alias);
1950 case KVM_CREATE_IRQCHIP:
1952 kvm->arch.vpic = kvm_create_pic(kvm);
1953 if (kvm->arch.vpic) {
1954 r = kvm_ioapic_init(kvm);
1956 kfree(kvm->arch.vpic);
1957 kvm->arch.vpic = NULL;
1963 case KVM_CREATE_PIT:
1965 kvm->arch.vpit = kvm_create_pit(kvm);
1969 case KVM_IRQ_LINE: {
1970 struct kvm_irq_level irq_event;
1973 if (copy_from_user(&irq_event, argp, sizeof irq_event))
1975 if (irqchip_in_kernel(kvm)) {
1976 mutex_lock(&kvm->lock);
1977 kvm_set_irq(kvm, irq_event.irq, irq_event.level);
1978 mutex_unlock(&kvm->lock);
1983 case KVM_GET_IRQCHIP: {
1984 /* 0: PIC master, 1: PIC slave, 2: IOAPIC */
1985 struct kvm_irqchip *chip = kmalloc(sizeof(*chip), GFP_KERNEL);
1991 if (copy_from_user(chip, argp, sizeof *chip))
1992 goto get_irqchip_out;
1994 if (!irqchip_in_kernel(kvm))
1995 goto get_irqchip_out;
1996 r = kvm_vm_ioctl_get_irqchip(kvm, chip);
1998 goto get_irqchip_out;
2000 if (copy_to_user(argp, chip, sizeof *chip))
2001 goto get_irqchip_out;
2009 case KVM_SET_IRQCHIP: {
2010 /* 0: PIC master, 1: PIC slave, 2: IOAPIC */
2011 struct kvm_irqchip *chip = kmalloc(sizeof(*chip), GFP_KERNEL);
2017 if (copy_from_user(chip, argp, sizeof *chip))
2018 goto set_irqchip_out;
2020 if (!irqchip_in_kernel(kvm))
2021 goto set_irqchip_out;
2022 r = kvm_vm_ioctl_set_irqchip(kvm, chip);
2024 goto set_irqchip_out;
2032 case KVM_ASSIGN_PCI_DEVICE: {
2033 struct kvm_assigned_pci_dev assigned_dev;
2036 if (copy_from_user(&assigned_dev, argp, sizeof assigned_dev))
2038 r = kvm_vm_ioctl_assign_device(kvm, &assigned_dev);
2043 case KVM_ASSIGN_IRQ: {
2044 struct kvm_assigned_irq assigned_irq;
2047 if (copy_from_user(&assigned_irq, argp, sizeof assigned_irq))
2049 r = kvm_vm_ioctl_assign_irq(kvm, &assigned_irq);
2056 if (copy_from_user(&u.ps, argp, sizeof(struct kvm_pit_state)))
2059 if (!kvm->arch.vpit)
2061 r = kvm_vm_ioctl_get_pit(kvm, &u.ps);
2065 if (copy_to_user(argp, &u.ps, sizeof(struct kvm_pit_state)))
2072 if (copy_from_user(&u.ps, argp, sizeof u.ps))
2075 if (!kvm->arch.vpit)
2077 r = kvm_vm_ioctl_set_pit(kvm, &u.ps);
2090 static void kvm_init_msr_list(void)
2095 for (i = j = 0; i < ARRAY_SIZE(msrs_to_save); i++) {
2096 if (rdmsr_safe(msrs_to_save[i], &dummy[0], &dummy[1]) < 0)
2099 msrs_to_save[j] = msrs_to_save[i];
2102 num_msrs_to_save = j;
2106 * Only apic need an MMIO device hook, so shortcut now..
2108 static struct kvm_io_device *vcpu_find_pervcpu_dev(struct kvm_vcpu *vcpu,
2109 gpa_t addr, int len,
2112 struct kvm_io_device *dev;
2114 if (vcpu->arch.apic) {
2115 dev = &vcpu->arch.apic->dev;
2116 if (dev->in_range(dev, addr, len, is_write))
2123 static struct kvm_io_device *vcpu_find_mmio_dev(struct kvm_vcpu *vcpu,
2124 gpa_t addr, int len,
2127 struct kvm_io_device *dev;
2129 dev = vcpu_find_pervcpu_dev(vcpu, addr, len, is_write);
2131 dev = kvm_io_bus_find_dev(&vcpu->kvm->mmio_bus, addr, len,
2136 int emulator_read_std(unsigned long addr,
2139 struct kvm_vcpu *vcpu)
2142 int r = X86EMUL_CONTINUE;
2145 gpa_t gpa = vcpu->arch.mmu.gva_to_gpa(vcpu, addr);
2146 unsigned offset = addr & (PAGE_SIZE-1);
2147 unsigned tocopy = min(bytes, (unsigned)PAGE_SIZE - offset);
2150 if (gpa == UNMAPPED_GVA) {
2151 r = X86EMUL_PROPAGATE_FAULT;
2154 ret = kvm_read_guest(vcpu->kvm, gpa, data, tocopy);
2156 r = X86EMUL_UNHANDLEABLE;
2167 EXPORT_SYMBOL_GPL(emulator_read_std);
2169 static int emulator_read_emulated(unsigned long addr,
2172 struct kvm_vcpu *vcpu)
2174 struct kvm_io_device *mmio_dev;
2177 if (vcpu->mmio_read_completed) {
2178 memcpy(val, vcpu->mmio_data, bytes);
2179 vcpu->mmio_read_completed = 0;
2180 return X86EMUL_CONTINUE;
2183 gpa = vcpu->arch.mmu.gva_to_gpa(vcpu, addr);
2185 /* For APIC access vmexit */
2186 if ((gpa & PAGE_MASK) == APIC_DEFAULT_PHYS_BASE)
2189 if (emulator_read_std(addr, val, bytes, vcpu)
2190 == X86EMUL_CONTINUE)
2191 return X86EMUL_CONTINUE;
2192 if (gpa == UNMAPPED_GVA)
2193 return X86EMUL_PROPAGATE_FAULT;
2197 * Is this MMIO handled locally?
2199 mutex_lock(&vcpu->kvm->lock);
2200 mmio_dev = vcpu_find_mmio_dev(vcpu, gpa, bytes, 0);
2202 kvm_iodevice_read(mmio_dev, gpa, bytes, val);
2203 mutex_unlock(&vcpu->kvm->lock);
2204 return X86EMUL_CONTINUE;
2206 mutex_unlock(&vcpu->kvm->lock);
2208 vcpu->mmio_needed = 1;
2209 vcpu->mmio_phys_addr = gpa;
2210 vcpu->mmio_size = bytes;
2211 vcpu->mmio_is_write = 0;
2213 return X86EMUL_UNHANDLEABLE;
2216 int emulator_write_phys(struct kvm_vcpu *vcpu, gpa_t gpa,
2217 const void *val, int bytes)
2221 ret = kvm_write_guest(vcpu->kvm, gpa, val, bytes);
2224 kvm_mmu_pte_write(vcpu, gpa, val, bytes);
2228 static int emulator_write_emulated_onepage(unsigned long addr,
2231 struct kvm_vcpu *vcpu)
2233 struct kvm_io_device *mmio_dev;
2236 gpa = vcpu->arch.mmu.gva_to_gpa(vcpu, addr);
2238 if (gpa == UNMAPPED_GVA) {
2239 kvm_inject_page_fault(vcpu, addr, 2);
2240 return X86EMUL_PROPAGATE_FAULT;
2243 /* For APIC access vmexit */
2244 if ((gpa & PAGE_MASK) == APIC_DEFAULT_PHYS_BASE)
2247 if (emulator_write_phys(vcpu, gpa, val, bytes))
2248 return X86EMUL_CONTINUE;
2252 * Is this MMIO handled locally?
2254 mutex_lock(&vcpu->kvm->lock);
2255 mmio_dev = vcpu_find_mmio_dev(vcpu, gpa, bytes, 1);
2257 kvm_iodevice_write(mmio_dev, gpa, bytes, val);
2258 mutex_unlock(&vcpu->kvm->lock);
2259 return X86EMUL_CONTINUE;
2261 mutex_unlock(&vcpu->kvm->lock);
2263 vcpu->mmio_needed = 1;
2264 vcpu->mmio_phys_addr = gpa;
2265 vcpu->mmio_size = bytes;
2266 vcpu->mmio_is_write = 1;
2267 memcpy(vcpu->mmio_data, val, bytes);
2269 return X86EMUL_CONTINUE;
2272 int emulator_write_emulated(unsigned long addr,
2275 struct kvm_vcpu *vcpu)
2277 /* Crossing a page boundary? */
2278 if (((addr + bytes - 1) ^ addr) & PAGE_MASK) {
2281 now = -addr & ~PAGE_MASK;
2282 rc = emulator_write_emulated_onepage(addr, val, now, vcpu);
2283 if (rc != X86EMUL_CONTINUE)
2289 return emulator_write_emulated_onepage(addr, val, bytes, vcpu);
2291 EXPORT_SYMBOL_GPL(emulator_write_emulated);
2293 static int emulator_cmpxchg_emulated(unsigned long addr,
2297 struct kvm_vcpu *vcpu)
2299 static int reported;
2303 printk(KERN_WARNING "kvm: emulating exchange as write\n");
2305 #ifndef CONFIG_X86_64
2306 /* guests cmpxchg8b have to be emulated atomically */
2313 gpa = vcpu->arch.mmu.gva_to_gpa(vcpu, addr);
2315 if (gpa == UNMAPPED_GVA ||
2316 (gpa & PAGE_MASK) == APIC_DEFAULT_PHYS_BASE)
2319 if (((gpa + bytes - 1) & PAGE_MASK) != (gpa & PAGE_MASK))
2324 page = gfn_to_page(vcpu->kvm, gpa >> PAGE_SHIFT);
2326 kaddr = kmap_atomic(page, KM_USER0);
2327 set_64bit((u64 *)(kaddr + offset_in_page(gpa)), val);
2328 kunmap_atomic(kaddr, KM_USER0);
2329 kvm_release_page_dirty(page);
2334 return emulator_write_emulated(addr, new, bytes, vcpu);
2337 static unsigned long get_segment_base(struct kvm_vcpu *vcpu, int seg)
2339 return kvm_x86_ops->get_segment_base(vcpu, seg);
2342 int emulate_invlpg(struct kvm_vcpu *vcpu, gva_t address)
2344 kvm_mmu_invlpg(vcpu, address);
2345 return X86EMUL_CONTINUE;
2348 int emulate_clts(struct kvm_vcpu *vcpu)
2350 KVMTRACE_0D(CLTS, vcpu, handler);
2351 kvm_x86_ops->set_cr0(vcpu, vcpu->arch.cr0 & ~X86_CR0_TS);
2352 return X86EMUL_CONTINUE;
2355 int emulator_get_dr(struct x86_emulate_ctxt *ctxt, int dr, unsigned long *dest)
2357 struct kvm_vcpu *vcpu = ctxt->vcpu;
2361 *dest = kvm_x86_ops->get_dr(vcpu, dr);
2362 return X86EMUL_CONTINUE;
2364 pr_unimpl(vcpu, "%s: unexpected dr %u\n", __func__, dr);
2365 return X86EMUL_UNHANDLEABLE;
2369 int emulator_set_dr(struct x86_emulate_ctxt *ctxt, int dr, unsigned long value)
2371 unsigned long mask = (ctxt->mode == X86EMUL_MODE_PROT64) ? ~0ULL : ~0U;
2374 kvm_x86_ops->set_dr(ctxt->vcpu, dr, value & mask, &exception);
2376 /* FIXME: better handling */
2377 return X86EMUL_UNHANDLEABLE;
2379 return X86EMUL_CONTINUE;
2382 void kvm_report_emulation_failure(struct kvm_vcpu *vcpu, const char *context)
2385 unsigned long rip = kvm_rip_read(vcpu);
2386 unsigned long rip_linear;
2388 if (!printk_ratelimit())
2391 rip_linear = rip + get_segment_base(vcpu, VCPU_SREG_CS);
2393 emulator_read_std(rip_linear, (void *)opcodes, 4, vcpu);
2395 printk(KERN_ERR "emulation failed (%s) rip %lx %02x %02x %02x %02x\n",
2396 context, rip, opcodes[0], opcodes[1], opcodes[2], opcodes[3]);
2398 EXPORT_SYMBOL_GPL(kvm_report_emulation_failure);
2400 static struct x86_emulate_ops emulate_ops = {
2401 .read_std = emulator_read_std,
2402 .read_emulated = emulator_read_emulated,
2403 .write_emulated = emulator_write_emulated,
2404 .cmpxchg_emulated = emulator_cmpxchg_emulated,
2407 static void cache_all_regs(struct kvm_vcpu *vcpu)
2409 kvm_register_read(vcpu, VCPU_REGS_RAX);
2410 kvm_register_read(vcpu, VCPU_REGS_RSP);
2411 kvm_register_read(vcpu, VCPU_REGS_RIP);
2412 vcpu->arch.regs_dirty = ~0;
2415 int emulate_instruction(struct kvm_vcpu *vcpu,
2416 struct kvm_run *run,
2422 struct decode_cache *c;
2424 kvm_clear_exception_queue(vcpu);
2425 vcpu->arch.mmio_fault_cr2 = cr2;
2427 * TODO: fix x86_emulate.c to use guest_read/write_register
2428 * instead of direct ->regs accesses, can save hundred cycles
2429 * on Intel for instructions that don't read/change RSP, for
2432 cache_all_regs(vcpu);
2434 vcpu->mmio_is_write = 0;
2435 vcpu->arch.pio.string = 0;
2437 if (!(emulation_type & EMULTYPE_NO_DECODE)) {
2439 kvm_x86_ops->get_cs_db_l_bits(vcpu, &cs_db, &cs_l);
2441 vcpu->arch.emulate_ctxt.vcpu = vcpu;
2442 vcpu->arch.emulate_ctxt.eflags = kvm_x86_ops->get_rflags(vcpu);
2443 vcpu->arch.emulate_ctxt.mode =
2444 (vcpu->arch.emulate_ctxt.eflags & X86_EFLAGS_VM)
2445 ? X86EMUL_MODE_REAL : cs_l
2446 ? X86EMUL_MODE_PROT64 : cs_db
2447 ? X86EMUL_MODE_PROT32 : X86EMUL_MODE_PROT16;
2449 r = x86_decode_insn(&vcpu->arch.emulate_ctxt, &emulate_ops);
2451 /* Reject the instructions other than VMCALL/VMMCALL when
2452 * try to emulate invalid opcode */
2453 c = &vcpu->arch.emulate_ctxt.decode;
2454 if ((emulation_type & EMULTYPE_TRAP_UD) &&
2455 (!(c->twobyte && c->b == 0x01 &&
2456 (c->modrm_reg == 0 || c->modrm_reg == 3) &&
2457 c->modrm_mod == 3 && c->modrm_rm == 1)))
2458 return EMULATE_FAIL;
2460 ++vcpu->stat.insn_emulation;
2462 ++vcpu->stat.insn_emulation_fail;
2463 if (kvm_mmu_unprotect_page_virt(vcpu, cr2))
2464 return EMULATE_DONE;
2465 return EMULATE_FAIL;
2469 r = x86_emulate_insn(&vcpu->arch.emulate_ctxt, &emulate_ops);
2471 if (vcpu->arch.pio.string)
2472 return EMULATE_DO_MMIO;
2474 if ((r || vcpu->mmio_is_write) && run) {
2475 run->exit_reason = KVM_EXIT_MMIO;
2476 run->mmio.phys_addr = vcpu->mmio_phys_addr;
2477 memcpy(run->mmio.data, vcpu->mmio_data, 8);
2478 run->mmio.len = vcpu->mmio_size;
2479 run->mmio.is_write = vcpu->mmio_is_write;
2483 if (kvm_mmu_unprotect_page_virt(vcpu, cr2))
2484 return EMULATE_DONE;
2485 if (!vcpu->mmio_needed) {
2486 kvm_report_emulation_failure(vcpu, "mmio");
2487 return EMULATE_FAIL;
2489 return EMULATE_DO_MMIO;
2492 kvm_x86_ops->set_rflags(vcpu, vcpu->arch.emulate_ctxt.eflags);
2494 if (vcpu->mmio_is_write) {
2495 vcpu->mmio_needed = 0;
2496 return EMULATE_DO_MMIO;
2499 return EMULATE_DONE;
2501 EXPORT_SYMBOL_GPL(emulate_instruction);
2503 static void free_pio_guest_pages(struct kvm_vcpu *vcpu)
2507 for (i = 0; i < ARRAY_SIZE(vcpu->arch.pio.guest_pages); ++i)
2508 if (vcpu->arch.pio.guest_pages[i]) {
2509 kvm_release_page_dirty(vcpu->arch.pio.guest_pages[i]);
2510 vcpu->arch.pio.guest_pages[i] = NULL;
2514 static int pio_copy_data(struct kvm_vcpu *vcpu)
2516 void *p = vcpu->arch.pio_data;
2519 int nr_pages = vcpu->arch.pio.guest_pages[1] ? 2 : 1;
2521 q = vmap(vcpu->arch.pio.guest_pages, nr_pages, VM_READ|VM_WRITE,
2524 free_pio_guest_pages(vcpu);
2527 q += vcpu->arch.pio.guest_page_offset;
2528 bytes = vcpu->arch.pio.size * vcpu->arch.pio.cur_count;
2529 if (vcpu->arch.pio.in)
2530 memcpy(q, p, bytes);
2532 memcpy(p, q, bytes);
2533 q -= vcpu->arch.pio.guest_page_offset;
2535 free_pio_guest_pages(vcpu);
2539 int complete_pio(struct kvm_vcpu *vcpu)
2541 struct kvm_pio_request *io = &vcpu->arch.pio;
2548 val = kvm_register_read(vcpu, VCPU_REGS_RAX);
2549 memcpy(&val, vcpu->arch.pio_data, io->size);
2550 kvm_register_write(vcpu, VCPU_REGS_RAX, val);
2554 r = pio_copy_data(vcpu);
2561 delta *= io->cur_count;
2563 * The size of the register should really depend on
2564 * current address size.
2566 val = kvm_register_read(vcpu, VCPU_REGS_RCX);
2568 kvm_register_write(vcpu, VCPU_REGS_RCX, val);
2574 val = kvm_register_read(vcpu, VCPU_REGS_RDI);
2576 kvm_register_write(vcpu, VCPU_REGS_RDI, val);
2578 val = kvm_register_read(vcpu, VCPU_REGS_RSI);
2580 kvm_register_write(vcpu, VCPU_REGS_RSI, val);
2584 io->count -= io->cur_count;
2590 static void kernel_pio(struct kvm_io_device *pio_dev,
2591 struct kvm_vcpu *vcpu,
2594 /* TODO: String I/O for in kernel device */
2596 mutex_lock(&vcpu->kvm->lock);
2597 if (vcpu->arch.pio.in)
2598 kvm_iodevice_read(pio_dev, vcpu->arch.pio.port,
2599 vcpu->arch.pio.size,
2602 kvm_iodevice_write(pio_dev, vcpu->arch.pio.port,
2603 vcpu->arch.pio.size,
2605 mutex_unlock(&vcpu->kvm->lock);
2608 static void pio_string_write(struct kvm_io_device *pio_dev,
2609 struct kvm_vcpu *vcpu)
2611 struct kvm_pio_request *io = &vcpu->arch.pio;
2612 void *pd = vcpu->arch.pio_data;
2615 mutex_lock(&vcpu->kvm->lock);
2616 for (i = 0; i < io->cur_count; i++) {
2617 kvm_iodevice_write(pio_dev, io->port,
2622 mutex_unlock(&vcpu->kvm->lock);
2625 static struct kvm_io_device *vcpu_find_pio_dev(struct kvm_vcpu *vcpu,
2626 gpa_t addr, int len,
2629 return kvm_io_bus_find_dev(&vcpu->kvm->pio_bus, addr, len, is_write);
2632 int kvm_emulate_pio(struct kvm_vcpu *vcpu, struct kvm_run *run, int in,
2633 int size, unsigned port)
2635 struct kvm_io_device *pio_dev;
2638 vcpu->run->exit_reason = KVM_EXIT_IO;
2639 vcpu->run->io.direction = in ? KVM_EXIT_IO_IN : KVM_EXIT_IO_OUT;
2640 vcpu->run->io.size = vcpu->arch.pio.size = size;
2641 vcpu->run->io.data_offset = KVM_PIO_PAGE_OFFSET * PAGE_SIZE;
2642 vcpu->run->io.count = vcpu->arch.pio.count = vcpu->arch.pio.cur_count = 1;
2643 vcpu->run->io.port = vcpu->arch.pio.port = port;
2644 vcpu->arch.pio.in = in;
2645 vcpu->arch.pio.string = 0;
2646 vcpu->arch.pio.down = 0;
2647 vcpu->arch.pio.guest_page_offset = 0;
2648 vcpu->arch.pio.rep = 0;
2650 if (vcpu->run->io.direction == KVM_EXIT_IO_IN)
2651 KVMTRACE_2D(IO_READ, vcpu, vcpu->run->io.port, (u32)size,
2654 KVMTRACE_2D(IO_WRITE, vcpu, vcpu->run->io.port, (u32)size,
2657 val = kvm_register_read(vcpu, VCPU_REGS_RAX);
2658 memcpy(vcpu->arch.pio_data, &val, 4);
2660 kvm_x86_ops->skip_emulated_instruction(vcpu);
2662 pio_dev = vcpu_find_pio_dev(vcpu, port, size, !in);
2664 kernel_pio(pio_dev, vcpu, vcpu->arch.pio_data);
2670 EXPORT_SYMBOL_GPL(kvm_emulate_pio);
2672 int kvm_emulate_pio_string(struct kvm_vcpu *vcpu, struct kvm_run *run, int in,
2673 int size, unsigned long count, int down,
2674 gva_t address, int rep, unsigned port)
2676 unsigned now, in_page;
2680 struct kvm_io_device *pio_dev;
2682 vcpu->run->exit_reason = KVM_EXIT_IO;
2683 vcpu->run->io.direction = in ? KVM_EXIT_IO_IN : KVM_EXIT_IO_OUT;
2684 vcpu->run->io.size = vcpu->arch.pio.size = size;
2685 vcpu->run->io.data_offset = KVM_PIO_PAGE_OFFSET * PAGE_SIZE;
2686 vcpu->run->io.count = vcpu->arch.pio.count = vcpu->arch.pio.cur_count = count;
2687 vcpu->run->io.port = vcpu->arch.pio.port = port;
2688 vcpu->arch.pio.in = in;
2689 vcpu->arch.pio.string = 1;
2690 vcpu->arch.pio.down = down;
2691 vcpu->arch.pio.guest_page_offset = offset_in_page(address);
2692 vcpu->arch.pio.rep = rep;
2694 if (vcpu->run->io.direction == KVM_EXIT_IO_IN)
2695 KVMTRACE_2D(IO_READ, vcpu, vcpu->run->io.port, (u32)size,
2698 KVMTRACE_2D(IO_WRITE, vcpu, vcpu->run->io.port, (u32)size,
2702 kvm_x86_ops->skip_emulated_instruction(vcpu);
2707 in_page = PAGE_SIZE - offset_in_page(address);
2709 in_page = offset_in_page(address) + size;
2710 now = min(count, (unsigned long)in_page / size);
2713 * String I/O straddles page boundary. Pin two guest pages
2714 * so that we satisfy atomicity constraints. Do just one
2715 * transaction to avoid complexity.
2722 * String I/O in reverse. Yuck. Kill the guest, fix later.
2724 pr_unimpl(vcpu, "guest string pio down\n");
2725 kvm_inject_gp(vcpu, 0);
2728 vcpu->run->io.count = now;
2729 vcpu->arch.pio.cur_count = now;
2731 if (vcpu->arch.pio.cur_count == vcpu->arch.pio.count)
2732 kvm_x86_ops->skip_emulated_instruction(vcpu);
2734 for (i = 0; i < nr_pages; ++i) {
2735 page = gva_to_page(vcpu, address + i * PAGE_SIZE);
2736 vcpu->arch.pio.guest_pages[i] = page;
2738 kvm_inject_gp(vcpu, 0);
2739 free_pio_guest_pages(vcpu);
2744 pio_dev = vcpu_find_pio_dev(vcpu, port,
2745 vcpu->arch.pio.cur_count,
2746 !vcpu->arch.pio.in);
2747 if (!vcpu->arch.pio.in) {
2748 /* string PIO write */
2749 ret = pio_copy_data(vcpu);
2750 if (ret >= 0 && pio_dev) {
2751 pio_string_write(pio_dev, vcpu);
2753 if (vcpu->arch.pio.count == 0)
2757 pr_unimpl(vcpu, "no string pio read support yet, "
2758 "port %x size %d count %ld\n",
2763 EXPORT_SYMBOL_GPL(kvm_emulate_pio_string);
2765 int kvm_arch_init(void *opaque)
2768 struct kvm_x86_ops *ops = (struct kvm_x86_ops *)opaque;
2771 printk(KERN_ERR "kvm: already loaded the other module\n");
2776 if (!ops->cpu_has_kvm_support()) {
2777 printk(KERN_ERR "kvm: no hardware support\n");
2781 if (ops->disabled_by_bios()) {
2782 printk(KERN_ERR "kvm: disabled by bios\n");
2787 r = kvm_mmu_module_init();
2791 kvm_init_msr_list();
2794 kvm_mmu_set_nonpresent_ptes(0ull, 0ull);
2795 kvm_mmu_set_base_ptes(PT_PRESENT_MASK);
2796 kvm_mmu_set_mask_ptes(PT_USER_MASK, PT_ACCESSED_MASK,
2797 PT_DIRTY_MASK, PT64_NX_MASK, 0);
2804 void kvm_arch_exit(void)
2807 kvm_mmu_module_exit();
2810 int kvm_emulate_halt(struct kvm_vcpu *vcpu)
2812 ++vcpu->stat.halt_exits;
2813 KVMTRACE_0D(HLT, vcpu, handler);
2814 if (irqchip_in_kernel(vcpu->kvm)) {
2815 vcpu->arch.mp_state = KVM_MP_STATE_HALTED;
2818 vcpu->run->exit_reason = KVM_EXIT_HLT;
2822 EXPORT_SYMBOL_GPL(kvm_emulate_halt);
2824 static inline gpa_t hc_gpa(struct kvm_vcpu *vcpu, unsigned long a0,
2827 if (is_long_mode(vcpu))
2830 return a0 | ((gpa_t)a1 << 32);
2833 int kvm_emulate_hypercall(struct kvm_vcpu *vcpu)
2835 unsigned long nr, a0, a1, a2, a3, ret;
2838 nr = kvm_register_read(vcpu, VCPU_REGS_RAX);
2839 a0 = kvm_register_read(vcpu, VCPU_REGS_RBX);
2840 a1 = kvm_register_read(vcpu, VCPU_REGS_RCX);
2841 a2 = kvm_register_read(vcpu, VCPU_REGS_RDX);
2842 a3 = kvm_register_read(vcpu, VCPU_REGS_RSI);
2844 KVMTRACE_1D(VMMCALL, vcpu, (u32)nr, handler);
2846 if (!is_long_mode(vcpu)) {
2855 case KVM_HC_VAPIC_POLL_IRQ:
2859 r = kvm_pv_mmu_op(vcpu, a0, hc_gpa(vcpu, a1, a2), &ret);
2865 kvm_register_write(vcpu, VCPU_REGS_RAX, ret);
2866 ++vcpu->stat.hypercalls;
2869 EXPORT_SYMBOL_GPL(kvm_emulate_hypercall);
2871 int kvm_fix_hypercall(struct kvm_vcpu *vcpu)
2873 char instruction[3];
2875 unsigned long rip = kvm_rip_read(vcpu);
2879 * Blow out the MMU to ensure that no other VCPU has an active mapping
2880 * to ensure that the updated hypercall appears atomically across all
2883 kvm_mmu_zap_all(vcpu->kvm);
2885 kvm_x86_ops->patch_hypercall(vcpu, instruction);
2886 if (emulator_write_emulated(rip, instruction, 3, vcpu)
2887 != X86EMUL_CONTINUE)
2893 static u64 mk_cr_64(u64 curr_cr, u32 new_val)
2895 return (curr_cr & ~((1ULL << 32) - 1)) | new_val;
2898 void realmode_lgdt(struct kvm_vcpu *vcpu, u16 limit, unsigned long base)
2900 struct descriptor_table dt = { limit, base };
2902 kvm_x86_ops->set_gdt(vcpu, &dt);
2905 void realmode_lidt(struct kvm_vcpu *vcpu, u16 limit, unsigned long base)
2907 struct descriptor_table dt = { limit, base };
2909 kvm_x86_ops->set_idt(vcpu, &dt);
2912 void realmode_lmsw(struct kvm_vcpu *vcpu, unsigned long msw,
2913 unsigned long *rflags)
2915 kvm_lmsw(vcpu, msw);
2916 *rflags = kvm_x86_ops->get_rflags(vcpu);
2919 unsigned long realmode_get_cr(struct kvm_vcpu *vcpu, int cr)
2921 unsigned long value;
2923 kvm_x86_ops->decache_cr4_guest_bits(vcpu);
2926 value = vcpu->arch.cr0;
2929 value = vcpu->arch.cr2;
2932 value = vcpu->arch.cr3;
2935 value = vcpu->arch.cr4;
2938 value = kvm_get_cr8(vcpu);
2941 vcpu_printf(vcpu, "%s: unexpected cr %u\n", __func__, cr);
2944 KVMTRACE_3D(CR_READ, vcpu, (u32)cr, (u32)value,
2945 (u32)((u64)value >> 32), handler);
2950 void realmode_set_cr(struct kvm_vcpu *vcpu, int cr, unsigned long val,
2951 unsigned long *rflags)
2953 KVMTRACE_3D(CR_WRITE, vcpu, (u32)cr, (u32)val,
2954 (u32)((u64)val >> 32), handler);
2958 kvm_set_cr0(vcpu, mk_cr_64(vcpu->arch.cr0, val));
2959 *rflags = kvm_x86_ops->get_rflags(vcpu);
2962 vcpu->arch.cr2 = val;
2965 kvm_set_cr3(vcpu, val);
2968 kvm_set_cr4(vcpu, mk_cr_64(vcpu->arch.cr4, val));
2971 kvm_set_cr8(vcpu, val & 0xfUL);
2974 vcpu_printf(vcpu, "%s: unexpected cr %u\n", __func__, cr);
2978 static int move_to_next_stateful_cpuid_entry(struct kvm_vcpu *vcpu, int i)
2980 struct kvm_cpuid_entry2 *e = &vcpu->arch.cpuid_entries[i];
2981 int j, nent = vcpu->arch.cpuid_nent;
2983 e->flags &= ~KVM_CPUID_FLAG_STATE_READ_NEXT;
2984 /* when no next entry is found, the current entry[i] is reselected */
2985 for (j = i + 1; j == i; j = (j + 1) % nent) {
2986 struct kvm_cpuid_entry2 *ej = &vcpu->arch.cpuid_entries[j];
2987 if (ej->function == e->function) {
2988 ej->flags |= KVM_CPUID_FLAG_STATE_READ_NEXT;
2992 return 0; /* silence gcc, even though control never reaches here */
2995 /* find an entry with matching function, matching index (if needed), and that
2996 * should be read next (if it's stateful) */
2997 static int is_matching_cpuid_entry(struct kvm_cpuid_entry2 *e,
2998 u32 function, u32 index)
3000 if (e->function != function)
3002 if ((e->flags & KVM_CPUID_FLAG_SIGNIFCANT_INDEX) && e->index != index)
3004 if ((e->flags & KVM_CPUID_FLAG_STATEFUL_FUNC) &&
3005 !(e->flags & KVM_CPUID_FLAG_STATE_READ_NEXT))
3010 void kvm_emulate_cpuid(struct kvm_vcpu *vcpu)
3013 u32 function, index;
3014 struct kvm_cpuid_entry2 *e, *best;
3016 function = kvm_register_read(vcpu, VCPU_REGS_RAX);
3017 index = kvm_register_read(vcpu, VCPU_REGS_RCX);
3018 kvm_register_write(vcpu, VCPU_REGS_RAX, 0);
3019 kvm_register_write(vcpu, VCPU_REGS_RBX, 0);
3020 kvm_register_write(vcpu, VCPU_REGS_RCX, 0);
3021 kvm_register_write(vcpu, VCPU_REGS_RDX, 0);
3023 for (i = 0; i < vcpu->arch.cpuid_nent; ++i) {
3024 e = &vcpu->arch.cpuid_entries[i];
3025 if (is_matching_cpuid_entry(e, function, index)) {
3026 if (e->flags & KVM_CPUID_FLAG_STATEFUL_FUNC)
3027 move_to_next_stateful_cpuid_entry(vcpu, i);
3032 * Both basic or both extended?
3034 if (((e->function ^ function) & 0x80000000) == 0)
3035 if (!best || e->function > best->function)
3039 kvm_register_write(vcpu, VCPU_REGS_RAX, best->eax);
3040 kvm_register_write(vcpu, VCPU_REGS_RBX, best->ebx);
3041 kvm_register_write(vcpu, VCPU_REGS_RCX, best->ecx);
3042 kvm_register_write(vcpu, VCPU_REGS_RDX, best->edx);
3044 kvm_x86_ops->skip_emulated_instruction(vcpu);
3045 KVMTRACE_5D(CPUID, vcpu, function,
3046 (u32)kvm_register_read(vcpu, VCPU_REGS_RAX),
3047 (u32)kvm_register_read(vcpu, VCPU_REGS_RBX),
3048 (u32)kvm_register_read(vcpu, VCPU_REGS_RCX),
3049 (u32)kvm_register_read(vcpu, VCPU_REGS_RDX), handler);
3051 EXPORT_SYMBOL_GPL(kvm_emulate_cpuid);
3054 * Check if userspace requested an interrupt window, and that the
3055 * interrupt window is open.
3057 * No need to exit to userspace if we already have an interrupt queued.
3059 static int dm_request_for_irq_injection(struct kvm_vcpu *vcpu,
3060 struct kvm_run *kvm_run)
3062 return (!vcpu->arch.irq_summary &&
3063 kvm_run->request_interrupt_window &&
3064 vcpu->arch.interrupt_window_open &&
3065 (kvm_x86_ops->get_rflags(vcpu) & X86_EFLAGS_IF));
3068 static void post_kvm_run_save(struct kvm_vcpu *vcpu,
3069 struct kvm_run *kvm_run)
3071 kvm_run->if_flag = (kvm_x86_ops->get_rflags(vcpu) & X86_EFLAGS_IF) != 0;
3072 kvm_run->cr8 = kvm_get_cr8(vcpu);
3073 kvm_run->apic_base = kvm_get_apic_base(vcpu);
3074 if (irqchip_in_kernel(vcpu->kvm))
3075 kvm_run->ready_for_interrupt_injection = 1;
3077 kvm_run->ready_for_interrupt_injection =
3078 (vcpu->arch.interrupt_window_open &&
3079 vcpu->arch.irq_summary == 0);
3082 static void vapic_enter(struct kvm_vcpu *vcpu)
3084 struct kvm_lapic *apic = vcpu->arch.apic;
3087 if (!apic || !apic->vapic_addr)
3090 page = gfn_to_page(vcpu->kvm, apic->vapic_addr >> PAGE_SHIFT);
3092 vcpu->arch.apic->vapic_page = page;
3095 static void vapic_exit(struct kvm_vcpu *vcpu)
3097 struct kvm_lapic *apic = vcpu->arch.apic;
3099 if (!apic || !apic->vapic_addr)
3102 down_read(&vcpu->kvm->slots_lock);
3103 kvm_release_page_dirty(apic->vapic_page);
3104 mark_page_dirty(vcpu->kvm, apic->vapic_addr >> PAGE_SHIFT);
3105 up_read(&vcpu->kvm->slots_lock);
3108 static int vcpu_enter_guest(struct kvm_vcpu *vcpu, struct kvm_run *kvm_run)
3113 if (test_and_clear_bit(KVM_REQ_MMU_RELOAD, &vcpu->requests))
3114 kvm_mmu_unload(vcpu);
3116 r = kvm_mmu_reload(vcpu);
3120 if (vcpu->requests) {
3121 if (test_and_clear_bit(KVM_REQ_MIGRATE_TIMER, &vcpu->requests))
3122 __kvm_migrate_timers(vcpu);
3123 if (test_and_clear_bit(KVM_REQ_TLB_FLUSH, &vcpu->requests))
3124 kvm_x86_ops->tlb_flush(vcpu);
3125 if (test_and_clear_bit(KVM_REQ_REPORT_TPR_ACCESS,
3127 kvm_run->exit_reason = KVM_EXIT_TPR_ACCESS;
3131 if (test_and_clear_bit(KVM_REQ_TRIPLE_FAULT, &vcpu->requests)) {
3132 kvm_run->exit_reason = KVM_EXIT_SHUTDOWN;
3138 clear_bit(KVM_REQ_PENDING_TIMER, &vcpu->requests);
3139 kvm_inject_pending_timer_irqs(vcpu);
3143 kvm_x86_ops->prepare_guest_switch(vcpu);
3144 kvm_load_guest_fpu(vcpu);
3146 local_irq_disable();
3148 if (vcpu->requests || need_resched() || signal_pending(current)) {
3155 if (vcpu->guest_debug.enabled)
3156 kvm_x86_ops->guest_debug_pre(vcpu);
3158 vcpu->guest_mode = 1;
3160 * Make sure that guest_mode assignment won't happen after
3161 * testing the pending IRQ vector bitmap.
3165 if (vcpu->arch.exception.pending)
3166 __queue_exception(vcpu);
3167 else if (irqchip_in_kernel(vcpu->kvm))
3168 kvm_x86_ops->inject_pending_irq(vcpu);
3170 kvm_x86_ops->inject_pending_vectors(vcpu, kvm_run);
3172 kvm_lapic_sync_to_vapic(vcpu);
3174 up_read(&vcpu->kvm->slots_lock);
3179 KVMTRACE_0D(VMENTRY, vcpu, entryexit);
3180 kvm_x86_ops->run(vcpu, kvm_run);
3182 vcpu->guest_mode = 0;
3188 * We must have an instruction between local_irq_enable() and
3189 * kvm_guest_exit(), so the timer interrupt isn't delayed by
3190 * the interrupt shadow. The stat.exits increment will do nicely.
3191 * But we need to prevent reordering, hence this barrier():
3199 down_read(&vcpu->kvm->slots_lock);
3202 * Profile KVM exit RIPs:
3204 if (unlikely(prof_on == KVM_PROFILING)) {
3205 unsigned long rip = kvm_rip_read(vcpu);
3206 profile_hit(KVM_PROFILING, (void *)rip);
3209 if (vcpu->arch.exception.pending && kvm_x86_ops->exception_injected(vcpu))
3210 vcpu->arch.exception.pending = false;
3212 kvm_lapic_sync_from_vapic(vcpu);
3214 r = kvm_x86_ops->handle_exit(kvm_run, vcpu);
3219 static int __vcpu_run(struct kvm_vcpu *vcpu, struct kvm_run *kvm_run)
3223 if (unlikely(vcpu->arch.mp_state == KVM_MP_STATE_SIPI_RECEIVED)) {
3224 printk("vcpu %d received sipi with vector # %x\n",
3225 vcpu->vcpu_id, vcpu->arch.sipi_vector);
3226 kvm_lapic_reset(vcpu);
3227 r = kvm_x86_ops->vcpu_reset(vcpu);
3230 vcpu->arch.mp_state = KVM_MP_STATE_RUNNABLE;
3233 down_read(&vcpu->kvm->slots_lock);
3238 if (vcpu->arch.mp_state == KVM_MP_STATE_RUNNABLE)
3239 r = vcpu_enter_guest(vcpu, kvm_run);
3241 up_read(&vcpu->kvm->slots_lock);
3242 kvm_vcpu_block(vcpu);
3243 down_read(&vcpu->kvm->slots_lock);
3244 if (test_and_clear_bit(KVM_REQ_UNHALT, &vcpu->requests))
3245 if (vcpu->arch.mp_state == KVM_MP_STATE_HALTED)
3246 vcpu->arch.mp_state =
3247 KVM_MP_STATE_RUNNABLE;
3248 if (vcpu->arch.mp_state != KVM_MP_STATE_RUNNABLE)
3253 if (dm_request_for_irq_injection(vcpu, kvm_run)) {
3255 kvm_run->exit_reason = KVM_EXIT_INTR;
3256 ++vcpu->stat.request_irq_exits;
3258 if (signal_pending(current)) {
3260 kvm_run->exit_reason = KVM_EXIT_INTR;
3261 ++vcpu->stat.signal_exits;
3263 if (need_resched()) {
3264 up_read(&vcpu->kvm->slots_lock);
3266 down_read(&vcpu->kvm->slots_lock);
3271 up_read(&vcpu->kvm->slots_lock);
3272 post_kvm_run_save(vcpu, kvm_run);
3279 int kvm_arch_vcpu_ioctl_run(struct kvm_vcpu *vcpu, struct kvm_run *kvm_run)
3286 if (vcpu->sigset_active)
3287 sigprocmask(SIG_SETMASK, &vcpu->sigset, &sigsaved);
3289 if (unlikely(vcpu->arch.mp_state == KVM_MP_STATE_UNINITIALIZED)) {
3290 kvm_vcpu_block(vcpu);
3291 clear_bit(KVM_REQ_UNHALT, &vcpu->requests);
3296 /* re-sync apic's tpr */
3297 if (!irqchip_in_kernel(vcpu->kvm))
3298 kvm_set_cr8(vcpu, kvm_run->cr8);
3300 if (vcpu->arch.pio.cur_count) {
3301 r = complete_pio(vcpu);
3305 #if CONFIG_HAS_IOMEM
3306 if (vcpu->mmio_needed) {
3307 memcpy(vcpu->mmio_data, kvm_run->mmio.data, 8);
3308 vcpu->mmio_read_completed = 1;
3309 vcpu->mmio_needed = 0;
3311 down_read(&vcpu->kvm->slots_lock);
3312 r = emulate_instruction(vcpu, kvm_run,
3313 vcpu->arch.mmio_fault_cr2, 0,
3314 EMULTYPE_NO_DECODE);
3315 up_read(&vcpu->kvm->slots_lock);
3316 if (r == EMULATE_DO_MMIO) {
3318 * Read-modify-write. Back to userspace.
3325 if (kvm_run->exit_reason == KVM_EXIT_HYPERCALL)
3326 kvm_register_write(vcpu, VCPU_REGS_RAX,
3327 kvm_run->hypercall.ret);
3329 r = __vcpu_run(vcpu, kvm_run);
3332 if (vcpu->sigset_active)
3333 sigprocmask(SIG_SETMASK, &sigsaved, NULL);
3339 int kvm_arch_vcpu_ioctl_get_regs(struct kvm_vcpu *vcpu, struct kvm_regs *regs)
3343 regs->rax = kvm_register_read(vcpu, VCPU_REGS_RAX);
3344 regs->rbx = kvm_register_read(vcpu, VCPU_REGS_RBX);
3345 regs->rcx = kvm_register_read(vcpu, VCPU_REGS_RCX);
3346 regs->rdx = kvm_register_read(vcpu, VCPU_REGS_RDX);
3347 regs->rsi = kvm_register_read(vcpu, VCPU_REGS_RSI);
3348 regs->rdi = kvm_register_read(vcpu, VCPU_REGS_RDI);
3349 regs->rsp = kvm_register_read(vcpu, VCPU_REGS_RSP);
3350 regs->rbp = kvm_register_read(vcpu, VCPU_REGS_RBP);
3351 #ifdef CONFIG_X86_64
3352 regs->r8 = kvm_register_read(vcpu, VCPU_REGS_R8);
3353 regs->r9 = kvm_register_read(vcpu, VCPU_REGS_R9);
3354 regs->r10 = kvm_register_read(vcpu, VCPU_REGS_R10);
3355 regs->r11 = kvm_register_read(vcpu, VCPU_REGS_R11);
3356 regs->r12 = kvm_register_read(vcpu, VCPU_REGS_R12);
3357 regs->r13 = kvm_register_read(vcpu, VCPU_REGS_R13);
3358 regs->r14 = kvm_register_read(vcpu, VCPU_REGS_R14);
3359 regs->r15 = kvm_register_read(vcpu, VCPU_REGS_R15);
3362 regs->rip = kvm_rip_read(vcpu);
3363 regs->rflags = kvm_x86_ops->get_rflags(vcpu);
3366 * Don't leak debug flags in case they were set for guest debugging
3368 if (vcpu->guest_debug.enabled && vcpu->guest_debug.singlestep)
3369 regs->rflags &= ~(X86_EFLAGS_TF | X86_EFLAGS_RF);
3376 int kvm_arch_vcpu_ioctl_set_regs(struct kvm_vcpu *vcpu, struct kvm_regs *regs)
3380 kvm_register_write(vcpu, VCPU_REGS_RAX, regs->rax);
3381 kvm_register_write(vcpu, VCPU_REGS_RBX, regs->rbx);
3382 kvm_register_write(vcpu, VCPU_REGS_RCX, regs->rcx);
3383 kvm_register_write(vcpu, VCPU_REGS_RDX, regs->rdx);
3384 kvm_register_write(vcpu, VCPU_REGS_RSI, regs->rsi);
3385 kvm_register_write(vcpu, VCPU_REGS_RDI, regs->rdi);
3386 kvm_register_write(vcpu, VCPU_REGS_RSP, regs->rsp);
3387 kvm_register_write(vcpu, VCPU_REGS_RBP, regs->rbp);
3388 #ifdef CONFIG_X86_64
3389 kvm_register_write(vcpu, VCPU_REGS_R8, regs->r8);
3390 kvm_register_write(vcpu, VCPU_REGS_R9, regs->r9);
3391 kvm_register_write(vcpu, VCPU_REGS_R10, regs->r10);
3392 kvm_register_write(vcpu, VCPU_REGS_R11, regs->r11);
3393 kvm_register_write(vcpu, VCPU_REGS_R12, regs->r12);
3394 kvm_register_write(vcpu, VCPU_REGS_R13, regs->r13);
3395 kvm_register_write(vcpu, VCPU_REGS_R14, regs->r14);
3396 kvm_register_write(vcpu, VCPU_REGS_R15, regs->r15);
3400 kvm_rip_write(vcpu, regs->rip);
3401 kvm_x86_ops->set_rflags(vcpu, regs->rflags);
3404 vcpu->arch.exception.pending = false;
3411 void kvm_get_segment(struct kvm_vcpu *vcpu,
3412 struct kvm_segment *var, int seg)
3414 kvm_x86_ops->get_segment(vcpu, var, seg);
3417 void kvm_get_cs_db_l_bits(struct kvm_vcpu *vcpu, int *db, int *l)
3419 struct kvm_segment cs;
3421 kvm_get_segment(vcpu, &cs, VCPU_SREG_CS);
3425 EXPORT_SYMBOL_GPL(kvm_get_cs_db_l_bits);
3427 int kvm_arch_vcpu_ioctl_get_sregs(struct kvm_vcpu *vcpu,
3428 struct kvm_sregs *sregs)
3430 struct descriptor_table dt;
3435 kvm_get_segment(vcpu, &sregs->cs, VCPU_SREG_CS);
3436 kvm_get_segment(vcpu, &sregs->ds, VCPU_SREG_DS);
3437 kvm_get_segment(vcpu, &sregs->es, VCPU_SREG_ES);
3438 kvm_get_segment(vcpu, &sregs->fs, VCPU_SREG_FS);
3439 kvm_get_segment(vcpu, &sregs->gs, VCPU_SREG_GS);
3440 kvm_get_segment(vcpu, &sregs->ss, VCPU_SREG_SS);
3442 kvm_get_segment(vcpu, &sregs->tr, VCPU_SREG_TR);
3443 kvm_get_segment(vcpu, &sregs->ldt, VCPU_SREG_LDTR);
3445 kvm_x86_ops->get_idt(vcpu, &dt);
3446 sregs->idt.limit = dt.limit;
3447 sregs->idt.base = dt.base;
3448 kvm_x86_ops->get_gdt(vcpu, &dt);
3449 sregs->gdt.limit = dt.limit;
3450 sregs->gdt.base = dt.base;
3452 kvm_x86_ops->decache_cr4_guest_bits(vcpu);
3453 sregs->cr0 = vcpu->arch.cr0;
3454 sregs->cr2 = vcpu->arch.cr2;
3455 sregs->cr3 = vcpu->arch.cr3;
3456 sregs->cr4 = vcpu->arch.cr4;
3457 sregs->cr8 = kvm_get_cr8(vcpu);
3458 sregs->efer = vcpu->arch.shadow_efer;
3459 sregs->apic_base = kvm_get_apic_base(vcpu);
3461 if (irqchip_in_kernel(vcpu->kvm)) {
3462 memset(sregs->interrupt_bitmap, 0,
3463 sizeof sregs->interrupt_bitmap);
3464 pending_vec = kvm_x86_ops->get_irq(vcpu);
3465 if (pending_vec >= 0)
3466 set_bit(pending_vec,
3467 (unsigned long *)sregs->interrupt_bitmap);
3469 memcpy(sregs->interrupt_bitmap, vcpu->arch.irq_pending,
3470 sizeof sregs->interrupt_bitmap);
3477 int kvm_arch_vcpu_ioctl_get_mpstate(struct kvm_vcpu *vcpu,
3478 struct kvm_mp_state *mp_state)
3481 mp_state->mp_state = vcpu->arch.mp_state;
3486 int kvm_arch_vcpu_ioctl_set_mpstate(struct kvm_vcpu *vcpu,
3487 struct kvm_mp_state *mp_state)
3490 vcpu->arch.mp_state = mp_state->mp_state;
3495 static void kvm_set_segment(struct kvm_vcpu *vcpu,
3496 struct kvm_segment *var, int seg)
3498 kvm_x86_ops->set_segment(vcpu, var, seg);
3501 static void seg_desct_to_kvm_desct(struct desc_struct *seg_desc, u16 selector,
3502 struct kvm_segment *kvm_desct)
3504 kvm_desct->base = seg_desc->base0;
3505 kvm_desct->base |= seg_desc->base1 << 16;
3506 kvm_desct->base |= seg_desc->base2 << 24;
3507 kvm_desct->limit = seg_desc->limit0;
3508 kvm_desct->limit |= seg_desc->limit << 16;
3510 kvm_desct->limit <<= 12;
3511 kvm_desct->limit |= 0xfff;
3513 kvm_desct->selector = selector;
3514 kvm_desct->type = seg_desc->type;
3515 kvm_desct->present = seg_desc->p;
3516 kvm_desct->dpl = seg_desc->dpl;
3517 kvm_desct->db = seg_desc->d;
3518 kvm_desct->s = seg_desc->s;
3519 kvm_desct->l = seg_desc->l;
3520 kvm_desct->g = seg_desc->g;
3521 kvm_desct->avl = seg_desc->avl;
3523 kvm_desct->unusable = 1;
3525 kvm_desct->unusable = 0;
3526 kvm_desct->padding = 0;
3529 static void get_segment_descritptor_dtable(struct kvm_vcpu *vcpu,
3531 struct descriptor_table *dtable)
3533 if (selector & 1 << 2) {
3534 struct kvm_segment kvm_seg;
3536 kvm_get_segment(vcpu, &kvm_seg, VCPU_SREG_LDTR);
3538 if (kvm_seg.unusable)
3541 dtable->limit = kvm_seg.limit;
3542 dtable->base = kvm_seg.base;
3545 kvm_x86_ops->get_gdt(vcpu, dtable);
3548 /* allowed just for 8 bytes segments */
3549 static int load_guest_segment_descriptor(struct kvm_vcpu *vcpu, u16 selector,
3550 struct desc_struct *seg_desc)
3553 struct descriptor_table dtable;
3554 u16 index = selector >> 3;
3556 get_segment_descritptor_dtable(vcpu, selector, &dtable);
3558 if (dtable.limit < index * 8 + 7) {
3559 kvm_queue_exception_e(vcpu, GP_VECTOR, selector & 0xfffc);
3562 gpa = vcpu->arch.mmu.gva_to_gpa(vcpu, dtable.base);
3564 return kvm_read_guest(vcpu->kvm, gpa, seg_desc, 8);
3567 /* allowed just for 8 bytes segments */
3568 static int save_guest_segment_descriptor(struct kvm_vcpu *vcpu, u16 selector,
3569 struct desc_struct *seg_desc)
3572 struct descriptor_table dtable;
3573 u16 index = selector >> 3;
3575 get_segment_descritptor_dtable(vcpu, selector, &dtable);
3577 if (dtable.limit < index * 8 + 7)
3579 gpa = vcpu->arch.mmu.gva_to_gpa(vcpu, dtable.base);
3581 return kvm_write_guest(vcpu->kvm, gpa, seg_desc, 8);
3584 static u32 get_tss_base_addr(struct kvm_vcpu *vcpu,
3585 struct desc_struct *seg_desc)
3589 base_addr = seg_desc->base0;
3590 base_addr |= (seg_desc->base1 << 16);
3591 base_addr |= (seg_desc->base2 << 24);
3593 return vcpu->arch.mmu.gva_to_gpa(vcpu, base_addr);
3596 static u16 get_segment_selector(struct kvm_vcpu *vcpu, int seg)
3598 struct kvm_segment kvm_seg;
3600 kvm_get_segment(vcpu, &kvm_seg, seg);
3601 return kvm_seg.selector;
3604 static int load_segment_descriptor_to_kvm_desct(struct kvm_vcpu *vcpu,
3606 struct kvm_segment *kvm_seg)
3608 struct desc_struct seg_desc;
3610 if (load_guest_segment_descriptor(vcpu, selector, &seg_desc))
3612 seg_desct_to_kvm_desct(&seg_desc, selector, kvm_seg);
3616 static int kvm_load_realmode_segment(struct kvm_vcpu *vcpu, u16 selector, int seg)
3618 struct kvm_segment segvar = {
3619 .base = selector << 4,
3621 .selector = selector,
3632 kvm_x86_ops->set_segment(vcpu, &segvar, seg);
3636 int kvm_load_segment_descriptor(struct kvm_vcpu *vcpu, u16 selector,
3637 int type_bits, int seg)
3639 struct kvm_segment kvm_seg;
3641 if (!(vcpu->arch.cr0 & X86_CR0_PE))
3642 return kvm_load_realmode_segment(vcpu, selector, seg);
3643 if (load_segment_descriptor_to_kvm_desct(vcpu, selector, &kvm_seg))
3645 kvm_seg.type |= type_bits;
3647 if (seg != VCPU_SREG_SS && seg != VCPU_SREG_CS &&
3648 seg != VCPU_SREG_LDTR)
3650 kvm_seg.unusable = 1;
3652 kvm_set_segment(vcpu, &kvm_seg, seg);
3656 static void save_state_to_tss32(struct kvm_vcpu *vcpu,
3657 struct tss_segment_32 *tss)
3659 tss->cr3 = vcpu->arch.cr3;
3660 tss->eip = kvm_rip_read(vcpu);
3661 tss->eflags = kvm_x86_ops->get_rflags(vcpu);
3662 tss->eax = kvm_register_read(vcpu, VCPU_REGS_RAX);
3663 tss->ecx = kvm_register_read(vcpu, VCPU_REGS_RCX);
3664 tss->edx = kvm_register_read(vcpu, VCPU_REGS_RDX);
3665 tss->ebx = kvm_register_read(vcpu, VCPU_REGS_RBX);
3666 tss->esp = kvm_register_read(vcpu, VCPU_REGS_RSP);
3667 tss->ebp = kvm_register_read(vcpu, VCPU_REGS_RBP);
3668 tss->esi = kvm_register_read(vcpu, VCPU_REGS_RSI);
3669 tss->edi = kvm_register_read(vcpu, VCPU_REGS_RDI);
3670 tss->es = get_segment_selector(vcpu, VCPU_SREG_ES);
3671 tss->cs = get_segment_selector(vcpu, VCPU_SREG_CS);
3672 tss->ss = get_segment_selector(vcpu, VCPU_SREG_SS);
3673 tss->ds = get_segment_selector(vcpu, VCPU_SREG_DS);
3674 tss->fs = get_segment_selector(vcpu, VCPU_SREG_FS);
3675 tss->gs = get_segment_selector(vcpu, VCPU_SREG_GS);
3676 tss->ldt_selector = get_segment_selector(vcpu, VCPU_SREG_LDTR);
3677 tss->prev_task_link = get_segment_selector(vcpu, VCPU_SREG_TR);
3680 static int load_state_from_tss32(struct kvm_vcpu *vcpu,
3681 struct tss_segment_32 *tss)
3683 kvm_set_cr3(vcpu, tss->cr3);
3685 kvm_rip_write(vcpu, tss->eip);
3686 kvm_x86_ops->set_rflags(vcpu, tss->eflags | 2);
3688 kvm_register_write(vcpu, VCPU_REGS_RAX, tss->eax);
3689 kvm_register_write(vcpu, VCPU_REGS_RCX, tss->ecx);
3690 kvm_register_write(vcpu, VCPU_REGS_RDX, tss->edx);
3691 kvm_register_write(vcpu, VCPU_REGS_RBX, tss->ebx);
3692 kvm_register_write(vcpu, VCPU_REGS_RSP, tss->esp);
3693 kvm_register_write(vcpu, VCPU_REGS_RBP, tss->ebp);
3694 kvm_register_write(vcpu, VCPU_REGS_RSI, tss->esi);
3695 kvm_register_write(vcpu, VCPU_REGS_RDI, tss->edi);
3697 if (kvm_load_segment_descriptor(vcpu, tss->ldt_selector, 0, VCPU_SREG_LDTR))
3700 if (kvm_load_segment_descriptor(vcpu, tss->es, 1, VCPU_SREG_ES))
3703 if (kvm_load_segment_descriptor(vcpu, tss->cs, 9, VCPU_SREG_CS))
3706 if (kvm_load_segment_descriptor(vcpu, tss->ss, 1, VCPU_SREG_SS))
3709 if (kvm_load_segment_descriptor(vcpu, tss->ds, 1, VCPU_SREG_DS))
3712 if (kvm_load_segment_descriptor(vcpu, tss->fs, 1, VCPU_SREG_FS))
3715 if (kvm_load_segment_descriptor(vcpu, tss->gs, 1, VCPU_SREG_GS))
3720 static void save_state_to_tss16(struct kvm_vcpu *vcpu,
3721 struct tss_segment_16 *tss)
3723 tss->ip = kvm_rip_read(vcpu);
3724 tss->flag = kvm_x86_ops->get_rflags(vcpu);
3725 tss->ax = kvm_register_read(vcpu, VCPU_REGS_RAX);
3726 tss->cx = kvm_register_read(vcpu, VCPU_REGS_RCX);
3727 tss->dx = kvm_register_read(vcpu, VCPU_REGS_RDX);
3728 tss->bx = kvm_register_read(vcpu, VCPU_REGS_RBX);
3729 tss->sp = kvm_register_read(vcpu, VCPU_REGS_RSP);
3730 tss->bp = kvm_register_read(vcpu, VCPU_REGS_RBP);
3731 tss->si = kvm_register_read(vcpu, VCPU_REGS_RSI);
3732 tss->di = kvm_register_read(vcpu, VCPU_REGS_RDI);
3734 tss->es = get_segment_selector(vcpu, VCPU_SREG_ES);
3735 tss->cs = get_segment_selector(vcpu, VCPU_SREG_CS);
3736 tss->ss = get_segment_selector(vcpu, VCPU_SREG_SS);
3737 tss->ds = get_segment_selector(vcpu, VCPU_SREG_DS);
3738 tss->ldt = get_segment_selector(vcpu, VCPU_SREG_LDTR);
3739 tss->prev_task_link = get_segment_selector(vcpu, VCPU_SREG_TR);
3742 static int load_state_from_tss16(struct kvm_vcpu *vcpu,
3743 struct tss_segment_16 *tss)
3745 kvm_rip_write(vcpu, tss->ip);
3746 kvm_x86_ops->set_rflags(vcpu, tss->flag | 2);
3747 kvm_register_write(vcpu, VCPU_REGS_RAX, tss->ax);
3748 kvm_register_write(vcpu, VCPU_REGS_RCX, tss->cx);
3749 kvm_register_write(vcpu, VCPU_REGS_RDX, tss->dx);
3750 kvm_register_write(vcpu, VCPU_REGS_RBX, tss->bx);
3751 kvm_register_write(vcpu, VCPU_REGS_RSP, tss->sp);
3752 kvm_register_write(vcpu, VCPU_REGS_RBP, tss->bp);
3753 kvm_register_write(vcpu, VCPU_REGS_RSI, tss->si);
3754 kvm_register_write(vcpu, VCPU_REGS_RDI, tss->di);
3756 if (kvm_load_segment_descriptor(vcpu, tss->ldt, 0, VCPU_SREG_LDTR))
3759 if (kvm_load_segment_descriptor(vcpu, tss->es, 1, VCPU_SREG_ES))
3762 if (kvm_load_segment_descriptor(vcpu, tss->cs, 9, VCPU_SREG_CS))
3765 if (kvm_load_segment_descriptor(vcpu, tss->ss, 1, VCPU_SREG_SS))
3768 if (kvm_load_segment_descriptor(vcpu, tss->ds, 1, VCPU_SREG_DS))
3773 static int kvm_task_switch_16(struct kvm_vcpu *vcpu, u16 tss_selector,
3775 struct desc_struct *nseg_desc)
3777 struct tss_segment_16 tss_segment_16;
3780 if (kvm_read_guest(vcpu->kvm, old_tss_base, &tss_segment_16,
3781 sizeof tss_segment_16))
3784 save_state_to_tss16(vcpu, &tss_segment_16);
3786 if (kvm_write_guest(vcpu->kvm, old_tss_base, &tss_segment_16,
3787 sizeof tss_segment_16))
3790 if (kvm_read_guest(vcpu->kvm, get_tss_base_addr(vcpu, nseg_desc),
3791 &tss_segment_16, sizeof tss_segment_16))
3794 if (load_state_from_tss16(vcpu, &tss_segment_16))
3802 static int kvm_task_switch_32(struct kvm_vcpu *vcpu, u16 tss_selector,
3804 struct desc_struct *nseg_desc)
3806 struct tss_segment_32 tss_segment_32;
3809 if (kvm_read_guest(vcpu->kvm, old_tss_base, &tss_segment_32,
3810 sizeof tss_segment_32))
3813 save_state_to_tss32(vcpu, &tss_segment_32);
3815 if (kvm_write_guest(vcpu->kvm, old_tss_base, &tss_segment_32,
3816 sizeof tss_segment_32))
3819 if (kvm_read_guest(vcpu->kvm, get_tss_base_addr(vcpu, nseg_desc),
3820 &tss_segment_32, sizeof tss_segment_32))
3823 if (load_state_from_tss32(vcpu, &tss_segment_32))
3831 int kvm_task_switch(struct kvm_vcpu *vcpu, u16 tss_selector, int reason)
3833 struct kvm_segment tr_seg;
3834 struct desc_struct cseg_desc;
3835 struct desc_struct nseg_desc;
3837 u32 old_tss_base = get_segment_base(vcpu, VCPU_SREG_TR);
3838 u16 old_tss_sel = get_segment_selector(vcpu, VCPU_SREG_TR);
3840 old_tss_base = vcpu->arch.mmu.gva_to_gpa(vcpu, old_tss_base);
3842 /* FIXME: Handle errors. Failure to read either TSS or their
3843 * descriptors should generate a pagefault.
3845 if (load_guest_segment_descriptor(vcpu, tss_selector, &nseg_desc))
3848 if (load_guest_segment_descriptor(vcpu, old_tss_sel, &cseg_desc))
3851 if (reason != TASK_SWITCH_IRET) {
3854 cpl = kvm_x86_ops->get_cpl(vcpu);
3855 if ((tss_selector & 3) > nseg_desc.dpl || cpl > nseg_desc.dpl) {
3856 kvm_queue_exception_e(vcpu, GP_VECTOR, 0);
3861 if (!nseg_desc.p || (nseg_desc.limit0 | nseg_desc.limit << 16) < 0x67) {
3862 kvm_queue_exception_e(vcpu, TS_VECTOR, tss_selector & 0xfffc);
3866 if (reason == TASK_SWITCH_IRET || reason == TASK_SWITCH_JMP) {
3867 cseg_desc.type &= ~(1 << 1); //clear the B flag
3868 save_guest_segment_descriptor(vcpu, old_tss_sel, &cseg_desc);
3871 if (reason == TASK_SWITCH_IRET) {
3872 u32 eflags = kvm_x86_ops->get_rflags(vcpu);
3873 kvm_x86_ops->set_rflags(vcpu, eflags & ~X86_EFLAGS_NT);
3876 kvm_x86_ops->skip_emulated_instruction(vcpu);
3878 if (nseg_desc.type & 8)
3879 ret = kvm_task_switch_32(vcpu, tss_selector, old_tss_base,
3882 ret = kvm_task_switch_16(vcpu, tss_selector, old_tss_base,
3885 if (reason == TASK_SWITCH_CALL || reason == TASK_SWITCH_GATE) {
3886 u32 eflags = kvm_x86_ops->get_rflags(vcpu);
3887 kvm_x86_ops->set_rflags(vcpu, eflags | X86_EFLAGS_NT);
3890 if (reason != TASK_SWITCH_IRET) {
3891 nseg_desc.type |= (1 << 1);
3892 save_guest_segment_descriptor(vcpu, tss_selector,
3896 kvm_x86_ops->set_cr0(vcpu, vcpu->arch.cr0 | X86_CR0_TS);
3897 seg_desct_to_kvm_desct(&nseg_desc, tss_selector, &tr_seg);
3899 kvm_set_segment(vcpu, &tr_seg, VCPU_SREG_TR);
3903 EXPORT_SYMBOL_GPL(kvm_task_switch);
3905 int kvm_arch_vcpu_ioctl_set_sregs(struct kvm_vcpu *vcpu,
3906 struct kvm_sregs *sregs)
3908 int mmu_reset_needed = 0;
3909 int i, pending_vec, max_bits;
3910 struct descriptor_table dt;
3914 dt.limit = sregs->idt.limit;
3915 dt.base = sregs->idt.base;
3916 kvm_x86_ops->set_idt(vcpu, &dt);
3917 dt.limit = sregs->gdt.limit;
3918 dt.base = sregs->gdt.base;
3919 kvm_x86_ops->set_gdt(vcpu, &dt);
3921 vcpu->arch.cr2 = sregs->cr2;
3922 mmu_reset_needed |= vcpu->arch.cr3 != sregs->cr3;
3923 vcpu->arch.cr3 = sregs->cr3;
3925 kvm_set_cr8(vcpu, sregs->cr8);
3927 mmu_reset_needed |= vcpu->arch.shadow_efer != sregs->efer;
3928 kvm_x86_ops->set_efer(vcpu, sregs->efer);
3929 kvm_set_apic_base(vcpu, sregs->apic_base);
3931 kvm_x86_ops->decache_cr4_guest_bits(vcpu);
3933 mmu_reset_needed |= vcpu->arch.cr0 != sregs->cr0;
3934 kvm_x86_ops->set_cr0(vcpu, sregs->cr0);
3935 vcpu->arch.cr0 = sregs->cr0;
3937 mmu_reset_needed |= vcpu->arch.cr4 != sregs->cr4;
3938 kvm_x86_ops->set_cr4(vcpu, sregs->cr4);
3939 if (!is_long_mode(vcpu) && is_pae(vcpu))
3940 load_pdptrs(vcpu, vcpu->arch.cr3);
3942 if (mmu_reset_needed)
3943 kvm_mmu_reset_context(vcpu);
3945 if (!irqchip_in_kernel(vcpu->kvm)) {
3946 memcpy(vcpu->arch.irq_pending, sregs->interrupt_bitmap,
3947 sizeof vcpu->arch.irq_pending);
3948 vcpu->arch.irq_summary = 0;
3949 for (i = 0; i < ARRAY_SIZE(vcpu->arch.irq_pending); ++i)
3950 if (vcpu->arch.irq_pending[i])
3951 __set_bit(i, &vcpu->arch.irq_summary);
3953 max_bits = (sizeof sregs->interrupt_bitmap) << 3;
3954 pending_vec = find_first_bit(
3955 (const unsigned long *)sregs->interrupt_bitmap,
3957 /* Only pending external irq is handled here */
3958 if (pending_vec < max_bits) {
3959 kvm_x86_ops->set_irq(vcpu, pending_vec);
3960 pr_debug("Set back pending irq %d\n",
3965 kvm_set_segment(vcpu, &sregs->cs, VCPU_SREG_CS);
3966 kvm_set_segment(vcpu, &sregs->ds, VCPU_SREG_DS);
3967 kvm_set_segment(vcpu, &sregs->es, VCPU_SREG_ES);
3968 kvm_set_segment(vcpu, &sregs->fs, VCPU_SREG_FS);
3969 kvm_set_segment(vcpu, &sregs->gs, VCPU_SREG_GS);
3970 kvm_set_segment(vcpu, &sregs->ss, VCPU_SREG_SS);
3972 kvm_set_segment(vcpu, &sregs->tr, VCPU_SREG_TR);
3973 kvm_set_segment(vcpu, &sregs->ldt, VCPU_SREG_LDTR);
3975 /* Older userspace won't unhalt the vcpu on reset. */
3976 if (vcpu->vcpu_id == 0 && kvm_rip_read(vcpu) == 0xfff0 &&
3977 sregs->cs.selector == 0xf000 && sregs->cs.base == 0xffff0000 &&
3978 !(vcpu->arch.cr0 & X86_CR0_PE))
3979 vcpu->arch.mp_state = KVM_MP_STATE_RUNNABLE;
3986 int kvm_arch_vcpu_ioctl_debug_guest(struct kvm_vcpu *vcpu,
3987 struct kvm_debug_guest *dbg)
3993 r = kvm_x86_ops->set_guest_debug(vcpu, dbg);
4001 * fxsave fpu state. Taken from x86_64/processor.h. To be killed when
4002 * we have asm/x86/processor.h
4013 u32 st_space[32]; /* 8*16 bytes for each FP-reg = 128 bytes */
4014 #ifdef CONFIG_X86_64
4015 u32 xmm_space[64]; /* 16*16 bytes for each XMM-reg = 256 bytes */
4017 u32 xmm_space[32]; /* 8*16 bytes for each XMM-reg = 128 bytes */
4022 * Translate a guest virtual address to a guest physical address.
4024 int kvm_arch_vcpu_ioctl_translate(struct kvm_vcpu *vcpu,
4025 struct kvm_translation *tr)
4027 unsigned long vaddr = tr->linear_address;
4031 down_read(&vcpu->kvm->slots_lock);
4032 gpa = vcpu->arch.mmu.gva_to_gpa(vcpu, vaddr);
4033 up_read(&vcpu->kvm->slots_lock);
4034 tr->physical_address = gpa;
4035 tr->valid = gpa != UNMAPPED_GVA;
4043 int kvm_arch_vcpu_ioctl_get_fpu(struct kvm_vcpu *vcpu, struct kvm_fpu *fpu)
4045 struct fxsave *fxsave = (struct fxsave *)&vcpu->arch.guest_fx_image;
4049 memcpy(fpu->fpr, fxsave->st_space, 128);
4050 fpu->fcw = fxsave->cwd;
4051 fpu->fsw = fxsave->swd;
4052 fpu->ftwx = fxsave->twd;
4053 fpu->last_opcode = fxsave->fop;
4054 fpu->last_ip = fxsave->rip;
4055 fpu->last_dp = fxsave->rdp;
4056 memcpy(fpu->xmm, fxsave->xmm_space, sizeof fxsave->xmm_space);
4063 int kvm_arch_vcpu_ioctl_set_fpu(struct kvm_vcpu *vcpu, struct kvm_fpu *fpu)
4065 struct fxsave *fxsave = (struct fxsave *)&vcpu->arch.guest_fx_image;
4069 memcpy(fxsave->st_space, fpu->fpr, 128);
4070 fxsave->cwd = fpu->fcw;
4071 fxsave->swd = fpu->fsw;
4072 fxsave->twd = fpu->ftwx;
4073 fxsave->fop = fpu->last_opcode;
4074 fxsave->rip = fpu->last_ip;
4075 fxsave->rdp = fpu->last_dp;
4076 memcpy(fxsave->xmm_space, fpu->xmm, sizeof fxsave->xmm_space);
4083 void fx_init(struct kvm_vcpu *vcpu)
4085 unsigned after_mxcsr_mask;
4088 * Touch the fpu the first time in non atomic context as if
4089 * this is the first fpu instruction the exception handler
4090 * will fire before the instruction returns and it'll have to
4091 * allocate ram with GFP_KERNEL.
4094 kvm_fx_save(&vcpu->arch.host_fx_image);
4096 /* Initialize guest FPU by resetting ours and saving into guest's */
4098 kvm_fx_save(&vcpu->arch.host_fx_image);
4100 kvm_fx_save(&vcpu->arch.guest_fx_image);
4101 kvm_fx_restore(&vcpu->arch.host_fx_image);
4104 vcpu->arch.cr0 |= X86_CR0_ET;
4105 after_mxcsr_mask = offsetof(struct i387_fxsave_struct, st_space);
4106 vcpu->arch.guest_fx_image.mxcsr = 0x1f80;
4107 memset((void *)&vcpu->arch.guest_fx_image + after_mxcsr_mask,
4108 0, sizeof(struct i387_fxsave_struct) - after_mxcsr_mask);
4110 EXPORT_SYMBOL_GPL(fx_init);
4112 void kvm_load_guest_fpu(struct kvm_vcpu *vcpu)
4114 if (!vcpu->fpu_active || vcpu->guest_fpu_loaded)
4117 vcpu->guest_fpu_loaded = 1;
4118 kvm_fx_save(&vcpu->arch.host_fx_image);
4119 kvm_fx_restore(&vcpu->arch.guest_fx_image);
4121 EXPORT_SYMBOL_GPL(kvm_load_guest_fpu);
4123 void kvm_put_guest_fpu(struct kvm_vcpu *vcpu)
4125 if (!vcpu->guest_fpu_loaded)
4128 vcpu->guest_fpu_loaded = 0;
4129 kvm_fx_save(&vcpu->arch.guest_fx_image);
4130 kvm_fx_restore(&vcpu->arch.host_fx_image);
4131 ++vcpu->stat.fpu_reload;
4133 EXPORT_SYMBOL_GPL(kvm_put_guest_fpu);
4135 void kvm_arch_vcpu_free(struct kvm_vcpu *vcpu)
4137 kvm_x86_ops->vcpu_free(vcpu);
4140 struct kvm_vcpu *kvm_arch_vcpu_create(struct kvm *kvm,
4143 return kvm_x86_ops->vcpu_create(kvm, id);
4146 int kvm_arch_vcpu_setup(struct kvm_vcpu *vcpu)
4150 /* We do fxsave: this must be aligned. */
4151 BUG_ON((unsigned long)&vcpu->arch.host_fx_image & 0xF);
4154 r = kvm_arch_vcpu_reset(vcpu);
4156 r = kvm_mmu_setup(vcpu);
4163 kvm_x86_ops->vcpu_free(vcpu);
4167 void kvm_arch_vcpu_destroy(struct kvm_vcpu *vcpu)
4170 kvm_mmu_unload(vcpu);
4173 kvm_x86_ops->vcpu_free(vcpu);
4176 int kvm_arch_vcpu_reset(struct kvm_vcpu *vcpu)
4178 return kvm_x86_ops->vcpu_reset(vcpu);
4181 void kvm_arch_hardware_enable(void *garbage)
4183 kvm_x86_ops->hardware_enable(garbage);
4186 void kvm_arch_hardware_disable(void *garbage)
4188 kvm_x86_ops->hardware_disable(garbage);
4191 int kvm_arch_hardware_setup(void)
4193 return kvm_x86_ops->hardware_setup();
4196 void kvm_arch_hardware_unsetup(void)
4198 kvm_x86_ops->hardware_unsetup();
4201 void kvm_arch_check_processor_compat(void *rtn)
4203 kvm_x86_ops->check_processor_compatibility(rtn);
4206 int kvm_arch_vcpu_init(struct kvm_vcpu *vcpu)
4212 BUG_ON(vcpu->kvm == NULL);
4215 vcpu->arch.mmu.root_hpa = INVALID_PAGE;
4216 if (!irqchip_in_kernel(kvm) || vcpu->vcpu_id == 0)
4217 vcpu->arch.mp_state = KVM_MP_STATE_RUNNABLE;
4219 vcpu->arch.mp_state = KVM_MP_STATE_UNINITIALIZED;
4221 page = alloc_page(GFP_KERNEL | __GFP_ZERO);
4226 vcpu->arch.pio_data = page_address(page);
4228 r = kvm_mmu_create(vcpu);
4230 goto fail_free_pio_data;
4232 if (irqchip_in_kernel(kvm)) {
4233 r = kvm_create_lapic(vcpu);
4235 goto fail_mmu_destroy;
4241 kvm_mmu_destroy(vcpu);
4243 free_page((unsigned long)vcpu->arch.pio_data);
4248 void kvm_arch_vcpu_uninit(struct kvm_vcpu *vcpu)
4250 kvm_free_lapic(vcpu);
4251 down_read(&vcpu->kvm->slots_lock);
4252 kvm_mmu_destroy(vcpu);
4253 up_read(&vcpu->kvm->slots_lock);
4254 free_page((unsigned long)vcpu->arch.pio_data);
4257 struct kvm *kvm_arch_create_vm(void)
4259 struct kvm *kvm = kzalloc(sizeof(struct kvm), GFP_KERNEL);
4262 return ERR_PTR(-ENOMEM);
4264 INIT_LIST_HEAD(&kvm->arch.active_mmu_pages);
4265 INIT_LIST_HEAD(&kvm->arch.assigned_dev_head);
4270 static void kvm_unload_vcpu_mmu(struct kvm_vcpu *vcpu)
4273 kvm_mmu_unload(vcpu);
4277 static void kvm_free_vcpus(struct kvm *kvm)
4282 * Unpin any mmu pages first.
4284 for (i = 0; i < KVM_MAX_VCPUS; ++i)
4286 kvm_unload_vcpu_mmu(kvm->vcpus[i]);
4287 for (i = 0; i < KVM_MAX_VCPUS; ++i) {
4288 if (kvm->vcpus[i]) {
4289 kvm_arch_vcpu_free(kvm->vcpus[i]);
4290 kvm->vcpus[i] = NULL;
4296 void kvm_arch_destroy_vm(struct kvm *kvm)
4298 kvm_iommu_unmap_guest(kvm);
4299 kvm_free_all_assigned_devices(kvm);
4301 kfree(kvm->arch.vpic);
4302 kfree(kvm->arch.vioapic);
4303 kvm_free_vcpus(kvm);
4304 kvm_free_physmem(kvm);
4305 if (kvm->arch.apic_access_page)
4306 put_page(kvm->arch.apic_access_page);
4307 if (kvm->arch.ept_identity_pagetable)
4308 put_page(kvm->arch.ept_identity_pagetable);
4312 int kvm_arch_set_memory_region(struct kvm *kvm,
4313 struct kvm_userspace_memory_region *mem,
4314 struct kvm_memory_slot old,
4317 int npages = mem->memory_size >> PAGE_SHIFT;
4318 struct kvm_memory_slot *memslot = &kvm->memslots[mem->slot];
4320 /*To keep backward compatibility with older userspace,
4321 *x86 needs to hanlde !user_alloc case.
4324 if (npages && !old.rmap) {
4325 unsigned long userspace_addr;
4327 down_write(¤t->mm->mmap_sem);
4328 userspace_addr = do_mmap(NULL, 0,
4330 PROT_READ | PROT_WRITE,
4331 MAP_PRIVATE | MAP_ANONYMOUS,
4333 up_write(¤t->mm->mmap_sem);
4335 if (IS_ERR((void *)userspace_addr))
4336 return PTR_ERR((void *)userspace_addr);
4338 /* set userspace_addr atomically for kvm_hva_to_rmapp */
4339 spin_lock(&kvm->mmu_lock);
4340 memslot->userspace_addr = userspace_addr;
4341 spin_unlock(&kvm->mmu_lock);
4343 if (!old.user_alloc && old.rmap) {
4346 down_write(¤t->mm->mmap_sem);
4347 ret = do_munmap(current->mm, old.userspace_addr,
4348 old.npages * PAGE_SIZE);
4349 up_write(¤t->mm->mmap_sem);
4352 "kvm_vm_ioctl_set_memory_region: "
4353 "failed to munmap memory\n");
4358 if (!kvm->arch.n_requested_mmu_pages) {
4359 unsigned int nr_mmu_pages = kvm_mmu_calculate_mmu_pages(kvm);
4360 kvm_mmu_change_mmu_pages(kvm, nr_mmu_pages);
4363 kvm_mmu_slot_remove_write_access(kvm, mem->slot);
4364 kvm_flush_remote_tlbs(kvm);
4369 void kvm_arch_flush_shadow(struct kvm *kvm)
4371 kvm_mmu_zap_all(kvm);
4374 int kvm_arch_vcpu_runnable(struct kvm_vcpu *vcpu)
4376 return vcpu->arch.mp_state == KVM_MP_STATE_RUNNABLE
4377 || vcpu->arch.mp_state == KVM_MP_STATE_SIPI_RECEIVED;
4380 static void vcpu_kick_intr(void *info)
4383 struct kvm_vcpu *vcpu = (struct kvm_vcpu *)info;
4384 printk(KERN_DEBUG "vcpu_kick_intr %p \n", vcpu);
4388 void kvm_vcpu_kick(struct kvm_vcpu *vcpu)
4390 int ipi_pcpu = vcpu->cpu;
4391 int cpu = get_cpu();
4393 if (waitqueue_active(&vcpu->wq)) {
4394 wake_up_interruptible(&vcpu->wq);
4395 ++vcpu->stat.halt_wakeup;
4398 * We may be called synchronously with irqs disabled in guest mode,
4399 * So need not to call smp_call_function_single() in that case.
4401 if (vcpu->guest_mode && vcpu->cpu != cpu)
4402 smp_call_function_single(ipi_pcpu, vcpu_kick_intr, vcpu, 0);