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 int kvm_vm_ioctl_assign_irq(struct kvm *kvm,
170 struct kvm_assigned_irq
174 struct kvm_assigned_dev_kernel *match;
176 mutex_lock(&kvm->lock);
178 match = kvm_find_assigned_dev(&kvm->arch.assigned_dev_head,
179 assigned_irq->assigned_dev_id);
181 mutex_unlock(&kvm->lock);
185 if (match->irq_requested) {
186 match->guest_irq = assigned_irq->guest_irq;
187 match->ack_notifier.gsi = assigned_irq->guest_irq;
188 mutex_unlock(&kvm->lock);
192 INIT_WORK(&match->interrupt_work,
193 kvm_assigned_dev_interrupt_work_handler);
195 if (irqchip_in_kernel(kvm)) {
196 if (!capable(CAP_SYS_RAWIO)) {
201 if (assigned_irq->host_irq)
202 match->host_irq = assigned_irq->host_irq;
204 match->host_irq = match->dev->irq;
205 match->guest_irq = assigned_irq->guest_irq;
206 match->ack_notifier.gsi = assigned_irq->guest_irq;
207 match->ack_notifier.irq_acked = kvm_assigned_dev_ack_irq;
208 kvm_register_irq_ack_notifier(kvm, &match->ack_notifier);
210 /* Even though this is PCI, we don't want to use shared
211 * interrupts. Sharing host devices with guest-assigned devices
212 * on the same interrupt line is not a happy situation: there
213 * are going to be long delays in accepting, acking, etc.
215 if (request_irq(match->host_irq, kvm_assigned_dev_intr, 0,
216 "kvm_assigned_device", (void *)match)) {
217 printk(KERN_INFO "%s: couldn't allocate irq for pv "
218 "device\n", __func__);
224 match->irq_requested = true;
226 mutex_unlock(&kvm->lock);
230 static int kvm_vm_ioctl_assign_device(struct kvm *kvm,
231 struct kvm_assigned_pci_dev *assigned_dev)
234 struct kvm_assigned_dev_kernel *match;
237 mutex_lock(&kvm->lock);
239 match = kvm_find_assigned_dev(&kvm->arch.assigned_dev_head,
240 assigned_dev->assigned_dev_id);
242 /* device already assigned */
247 match = kzalloc(sizeof(struct kvm_assigned_dev_kernel), GFP_KERNEL);
249 printk(KERN_INFO "%s: Couldn't allocate memory\n",
254 dev = pci_get_bus_and_slot(assigned_dev->busnr,
255 assigned_dev->devfn);
257 printk(KERN_INFO "%s: host device not found\n", __func__);
261 if (pci_enable_device(dev)) {
262 printk(KERN_INFO "%s: Could not enable PCI device\n", __func__);
266 r = pci_request_regions(dev, "kvm_assigned_device");
268 printk(KERN_INFO "%s: Could not get access to device regions\n",
272 match->assigned_dev_id = assigned_dev->assigned_dev_id;
273 match->host_busnr = assigned_dev->busnr;
274 match->host_devfn = assigned_dev->devfn;
279 list_add(&match->list, &kvm->arch.assigned_dev_head);
281 if (assigned_dev->flags & KVM_DEV_ASSIGN_ENABLE_IOMMU) {
282 r = kvm_iommu_map_guest(kvm, match);
288 mutex_unlock(&kvm->lock);
291 list_del(&match->list);
292 pci_release_regions(dev);
294 pci_disable_device(dev);
299 mutex_unlock(&kvm->lock);
303 static void kvm_free_assigned_devices(struct kvm *kvm)
305 struct list_head *ptr, *ptr2;
306 struct kvm_assigned_dev_kernel *assigned_dev;
308 list_for_each_safe(ptr, ptr2, &kvm->arch.assigned_dev_head) {
309 assigned_dev = list_entry(ptr,
310 struct kvm_assigned_dev_kernel,
313 if (irqchip_in_kernel(kvm) && assigned_dev->irq_requested) {
314 free_irq(assigned_dev->host_irq,
315 (void *)assigned_dev);
317 kvm_unregister_irq_ack_notifier(kvm,
322 if (cancel_work_sync(&assigned_dev->interrupt_work))
323 /* We had pending work. That means we will have to take
324 * care of kvm_put_kvm.
328 pci_release_regions(assigned_dev->dev);
329 pci_disable_device(assigned_dev->dev);
330 pci_dev_put(assigned_dev->dev);
332 list_del(&assigned_dev->list);
337 unsigned long segment_base(u16 selector)
339 struct descriptor_table gdt;
340 struct desc_struct *d;
341 unsigned long table_base;
347 asm("sgdt %0" : "=m"(gdt));
348 table_base = gdt.base;
350 if (selector & 4) { /* from ldt */
353 asm("sldt %0" : "=g"(ldt_selector));
354 table_base = segment_base(ldt_selector);
356 d = (struct desc_struct *)(table_base + (selector & ~7));
357 v = d->base0 | ((unsigned long)d->base1 << 16) |
358 ((unsigned long)d->base2 << 24);
360 if (d->s == 0 && (d->type == 2 || d->type == 9 || d->type == 11))
361 v |= ((unsigned long)((struct ldttss_desc64 *)d)->base3) << 32;
365 EXPORT_SYMBOL_GPL(segment_base);
367 u64 kvm_get_apic_base(struct kvm_vcpu *vcpu)
369 if (irqchip_in_kernel(vcpu->kvm))
370 return vcpu->arch.apic_base;
372 return vcpu->arch.apic_base;
374 EXPORT_SYMBOL_GPL(kvm_get_apic_base);
376 void kvm_set_apic_base(struct kvm_vcpu *vcpu, u64 data)
378 /* TODO: reserve bits check */
379 if (irqchip_in_kernel(vcpu->kvm))
380 kvm_lapic_set_base(vcpu, data);
382 vcpu->arch.apic_base = data;
384 EXPORT_SYMBOL_GPL(kvm_set_apic_base);
386 void kvm_queue_exception(struct kvm_vcpu *vcpu, unsigned nr)
388 WARN_ON(vcpu->arch.exception.pending);
389 vcpu->arch.exception.pending = true;
390 vcpu->arch.exception.has_error_code = false;
391 vcpu->arch.exception.nr = nr;
393 EXPORT_SYMBOL_GPL(kvm_queue_exception);
395 void kvm_inject_page_fault(struct kvm_vcpu *vcpu, unsigned long addr,
398 ++vcpu->stat.pf_guest;
399 if (vcpu->arch.exception.pending) {
400 if (vcpu->arch.exception.nr == PF_VECTOR) {
401 printk(KERN_DEBUG "kvm: inject_page_fault:"
402 " double fault 0x%lx\n", addr);
403 vcpu->arch.exception.nr = DF_VECTOR;
404 vcpu->arch.exception.error_code = 0;
405 } else if (vcpu->arch.exception.nr == DF_VECTOR) {
406 /* triple fault -> shutdown */
407 set_bit(KVM_REQ_TRIPLE_FAULT, &vcpu->requests);
411 vcpu->arch.cr2 = addr;
412 kvm_queue_exception_e(vcpu, PF_VECTOR, error_code);
415 void kvm_inject_nmi(struct kvm_vcpu *vcpu)
417 vcpu->arch.nmi_pending = 1;
419 EXPORT_SYMBOL_GPL(kvm_inject_nmi);
421 void kvm_queue_exception_e(struct kvm_vcpu *vcpu, unsigned nr, u32 error_code)
423 WARN_ON(vcpu->arch.exception.pending);
424 vcpu->arch.exception.pending = true;
425 vcpu->arch.exception.has_error_code = true;
426 vcpu->arch.exception.nr = nr;
427 vcpu->arch.exception.error_code = error_code;
429 EXPORT_SYMBOL_GPL(kvm_queue_exception_e);
431 static void __queue_exception(struct kvm_vcpu *vcpu)
433 kvm_x86_ops->queue_exception(vcpu, vcpu->arch.exception.nr,
434 vcpu->arch.exception.has_error_code,
435 vcpu->arch.exception.error_code);
439 * Load the pae pdptrs. Return true is they are all valid.
441 int load_pdptrs(struct kvm_vcpu *vcpu, unsigned long cr3)
443 gfn_t pdpt_gfn = cr3 >> PAGE_SHIFT;
444 unsigned offset = ((cr3 & (PAGE_SIZE-1)) >> 5) << 2;
447 u64 pdpte[ARRAY_SIZE(vcpu->arch.pdptrs)];
449 ret = kvm_read_guest_page(vcpu->kvm, pdpt_gfn, pdpte,
450 offset * sizeof(u64), sizeof(pdpte));
455 for (i = 0; i < ARRAY_SIZE(pdpte); ++i) {
456 if ((pdpte[i] & 1) && (pdpte[i] & 0xfffffff0000001e6ull)) {
463 memcpy(vcpu->arch.pdptrs, pdpte, sizeof(vcpu->arch.pdptrs));
468 EXPORT_SYMBOL_GPL(load_pdptrs);
470 static bool pdptrs_changed(struct kvm_vcpu *vcpu)
472 u64 pdpte[ARRAY_SIZE(vcpu->arch.pdptrs)];
476 if (is_long_mode(vcpu) || !is_pae(vcpu))
479 r = kvm_read_guest(vcpu->kvm, vcpu->arch.cr3 & ~31u, pdpte, sizeof(pdpte));
482 changed = memcmp(pdpte, vcpu->arch.pdptrs, sizeof(pdpte)) != 0;
488 void kvm_set_cr0(struct kvm_vcpu *vcpu, unsigned long cr0)
490 if (cr0 & CR0_RESERVED_BITS) {
491 printk(KERN_DEBUG "set_cr0: 0x%lx #GP, reserved bits 0x%lx\n",
492 cr0, vcpu->arch.cr0);
493 kvm_inject_gp(vcpu, 0);
497 if ((cr0 & X86_CR0_NW) && !(cr0 & X86_CR0_CD)) {
498 printk(KERN_DEBUG "set_cr0: #GP, CD == 0 && NW == 1\n");
499 kvm_inject_gp(vcpu, 0);
503 if ((cr0 & X86_CR0_PG) && !(cr0 & X86_CR0_PE)) {
504 printk(KERN_DEBUG "set_cr0: #GP, set PG flag "
505 "and a clear PE flag\n");
506 kvm_inject_gp(vcpu, 0);
510 if (!is_paging(vcpu) && (cr0 & X86_CR0_PG)) {
512 if ((vcpu->arch.shadow_efer & EFER_LME)) {
516 printk(KERN_DEBUG "set_cr0: #GP, start paging "
517 "in long mode while PAE is disabled\n");
518 kvm_inject_gp(vcpu, 0);
521 kvm_x86_ops->get_cs_db_l_bits(vcpu, &cs_db, &cs_l);
523 printk(KERN_DEBUG "set_cr0: #GP, start paging "
524 "in long mode while CS.L == 1\n");
525 kvm_inject_gp(vcpu, 0);
531 if (is_pae(vcpu) && !load_pdptrs(vcpu, vcpu->arch.cr3)) {
532 printk(KERN_DEBUG "set_cr0: #GP, pdptrs "
534 kvm_inject_gp(vcpu, 0);
540 kvm_x86_ops->set_cr0(vcpu, cr0);
541 vcpu->arch.cr0 = cr0;
543 kvm_mmu_reset_context(vcpu);
546 EXPORT_SYMBOL_GPL(kvm_set_cr0);
548 void kvm_lmsw(struct kvm_vcpu *vcpu, unsigned long msw)
550 kvm_set_cr0(vcpu, (vcpu->arch.cr0 & ~0x0ful) | (msw & 0x0f));
551 KVMTRACE_1D(LMSW, vcpu,
552 (u32)((vcpu->arch.cr0 & ~0x0ful) | (msw & 0x0f)),
555 EXPORT_SYMBOL_GPL(kvm_lmsw);
557 void kvm_set_cr4(struct kvm_vcpu *vcpu, unsigned long cr4)
559 if (cr4 & CR4_RESERVED_BITS) {
560 printk(KERN_DEBUG "set_cr4: #GP, reserved bits\n");
561 kvm_inject_gp(vcpu, 0);
565 if (is_long_mode(vcpu)) {
566 if (!(cr4 & X86_CR4_PAE)) {
567 printk(KERN_DEBUG "set_cr4: #GP, clearing PAE while "
569 kvm_inject_gp(vcpu, 0);
572 } else if (is_paging(vcpu) && !is_pae(vcpu) && (cr4 & X86_CR4_PAE)
573 && !load_pdptrs(vcpu, vcpu->arch.cr3)) {
574 printk(KERN_DEBUG "set_cr4: #GP, pdptrs reserved bits\n");
575 kvm_inject_gp(vcpu, 0);
579 if (cr4 & X86_CR4_VMXE) {
580 printk(KERN_DEBUG "set_cr4: #GP, setting VMXE\n");
581 kvm_inject_gp(vcpu, 0);
584 kvm_x86_ops->set_cr4(vcpu, cr4);
585 vcpu->arch.cr4 = cr4;
586 kvm_mmu_reset_context(vcpu);
588 EXPORT_SYMBOL_GPL(kvm_set_cr4);
590 void kvm_set_cr3(struct kvm_vcpu *vcpu, unsigned long cr3)
592 if (cr3 == vcpu->arch.cr3 && !pdptrs_changed(vcpu)) {
593 kvm_mmu_flush_tlb(vcpu);
597 if (is_long_mode(vcpu)) {
598 if (cr3 & CR3_L_MODE_RESERVED_BITS) {
599 printk(KERN_DEBUG "set_cr3: #GP, reserved bits\n");
600 kvm_inject_gp(vcpu, 0);
605 if (cr3 & CR3_PAE_RESERVED_BITS) {
607 "set_cr3: #GP, reserved bits\n");
608 kvm_inject_gp(vcpu, 0);
611 if (is_paging(vcpu) && !load_pdptrs(vcpu, cr3)) {
612 printk(KERN_DEBUG "set_cr3: #GP, pdptrs "
614 kvm_inject_gp(vcpu, 0);
619 * We don't check reserved bits in nonpae mode, because
620 * this isn't enforced, and VMware depends on this.
625 * Does the new cr3 value map to physical memory? (Note, we
626 * catch an invalid cr3 even in real-mode, because it would
627 * cause trouble later on when we turn on paging anyway.)
629 * A real CPU would silently accept an invalid cr3 and would
630 * attempt to use it - with largely undefined (and often hard
631 * to debug) behavior on the guest side.
633 if (unlikely(!gfn_to_memslot(vcpu->kvm, cr3 >> PAGE_SHIFT)))
634 kvm_inject_gp(vcpu, 0);
636 vcpu->arch.cr3 = cr3;
637 vcpu->arch.mmu.new_cr3(vcpu);
640 EXPORT_SYMBOL_GPL(kvm_set_cr3);
642 void kvm_set_cr8(struct kvm_vcpu *vcpu, unsigned long cr8)
644 if (cr8 & CR8_RESERVED_BITS) {
645 printk(KERN_DEBUG "set_cr8: #GP, reserved bits 0x%lx\n", cr8);
646 kvm_inject_gp(vcpu, 0);
649 if (irqchip_in_kernel(vcpu->kvm))
650 kvm_lapic_set_tpr(vcpu, cr8);
652 vcpu->arch.cr8 = cr8;
654 EXPORT_SYMBOL_GPL(kvm_set_cr8);
656 unsigned long kvm_get_cr8(struct kvm_vcpu *vcpu)
658 if (irqchip_in_kernel(vcpu->kvm))
659 return kvm_lapic_get_cr8(vcpu);
661 return vcpu->arch.cr8;
663 EXPORT_SYMBOL_GPL(kvm_get_cr8);
666 * List of msr numbers which we expose to userspace through KVM_GET_MSRS
667 * and KVM_SET_MSRS, and KVM_GET_MSR_INDEX_LIST.
669 * This list is modified at module load time to reflect the
670 * capabilities of the host cpu.
672 static u32 msrs_to_save[] = {
673 MSR_IA32_SYSENTER_CS, MSR_IA32_SYSENTER_ESP, MSR_IA32_SYSENTER_EIP,
676 MSR_CSTAR, MSR_KERNEL_GS_BASE, MSR_SYSCALL_MASK, MSR_LSTAR,
678 MSR_IA32_TIME_STAMP_COUNTER, MSR_KVM_SYSTEM_TIME, MSR_KVM_WALL_CLOCK,
679 MSR_IA32_PERF_STATUS,
682 static unsigned num_msrs_to_save;
684 static u32 emulated_msrs[] = {
685 MSR_IA32_MISC_ENABLE,
688 static void set_efer(struct kvm_vcpu *vcpu, u64 efer)
690 if (efer & efer_reserved_bits) {
691 printk(KERN_DEBUG "set_efer: 0x%llx #GP, reserved bits\n",
693 kvm_inject_gp(vcpu, 0);
698 && (vcpu->arch.shadow_efer & EFER_LME) != (efer & EFER_LME)) {
699 printk(KERN_DEBUG "set_efer: #GP, change LME while paging\n");
700 kvm_inject_gp(vcpu, 0);
704 kvm_x86_ops->set_efer(vcpu, efer);
707 efer |= vcpu->arch.shadow_efer & EFER_LMA;
709 vcpu->arch.shadow_efer = efer;
712 void kvm_enable_efer_bits(u64 mask)
714 efer_reserved_bits &= ~mask;
716 EXPORT_SYMBOL_GPL(kvm_enable_efer_bits);
720 * Writes msr value into into the appropriate "register".
721 * Returns 0 on success, non-0 otherwise.
722 * Assumes vcpu_load() was already called.
724 int kvm_set_msr(struct kvm_vcpu *vcpu, u32 msr_index, u64 data)
726 return kvm_x86_ops->set_msr(vcpu, msr_index, data);
730 * Adapt set_msr() to msr_io()'s calling convention
732 static int do_set_msr(struct kvm_vcpu *vcpu, unsigned index, u64 *data)
734 return kvm_set_msr(vcpu, index, *data);
737 static void kvm_write_wall_clock(struct kvm *kvm, gpa_t wall_clock)
740 struct pvclock_wall_clock wc;
741 struct timespec now, sys, boot;
748 kvm_write_guest(kvm, wall_clock, &version, sizeof(version));
751 * The guest calculates current wall clock time by adding
752 * system time (updated by kvm_write_guest_time below) to the
753 * wall clock specified here. guest system time equals host
754 * system time for us, thus we must fill in host boot time here.
756 now = current_kernel_time();
758 boot = ns_to_timespec(timespec_to_ns(&now) - timespec_to_ns(&sys));
760 wc.sec = boot.tv_sec;
761 wc.nsec = boot.tv_nsec;
762 wc.version = version;
764 kvm_write_guest(kvm, wall_clock, &wc, sizeof(wc));
767 kvm_write_guest(kvm, wall_clock, &version, sizeof(version));
770 static uint32_t div_frac(uint32_t dividend, uint32_t divisor)
772 uint32_t quotient, remainder;
774 /* Don't try to replace with do_div(), this one calculates
775 * "(dividend << 32) / divisor" */
777 : "=a" (quotient), "=d" (remainder)
778 : "0" (0), "1" (dividend), "r" (divisor) );
782 static void kvm_set_time_scale(uint32_t tsc_khz, struct pvclock_vcpu_time_info *hv_clock)
784 uint64_t nsecs = 1000000000LL;
789 tps64 = tsc_khz * 1000LL;
790 while (tps64 > nsecs*2) {
795 tps32 = (uint32_t)tps64;
796 while (tps32 <= (uint32_t)nsecs) {
801 hv_clock->tsc_shift = shift;
802 hv_clock->tsc_to_system_mul = div_frac(nsecs, tps32);
804 pr_debug("%s: tsc_khz %u, tsc_shift %d, tsc_mul %u\n",
805 __FUNCTION__, tsc_khz, hv_clock->tsc_shift,
806 hv_clock->tsc_to_system_mul);
809 static void kvm_write_guest_time(struct kvm_vcpu *v)
813 struct kvm_vcpu_arch *vcpu = &v->arch;
816 if ((!vcpu->time_page))
819 if (unlikely(vcpu->hv_clock_tsc_khz != tsc_khz)) {
820 kvm_set_time_scale(tsc_khz, &vcpu->hv_clock);
821 vcpu->hv_clock_tsc_khz = tsc_khz;
824 /* Keep irq disabled to prevent changes to the clock */
825 local_irq_save(flags);
826 kvm_get_msr(v, MSR_IA32_TIME_STAMP_COUNTER,
827 &vcpu->hv_clock.tsc_timestamp);
829 local_irq_restore(flags);
831 /* With all the info we got, fill in the values */
833 vcpu->hv_clock.system_time = ts.tv_nsec +
834 (NSEC_PER_SEC * (u64)ts.tv_sec);
836 * The interface expects us to write an even number signaling that the
837 * update is finished. Since the guest won't see the intermediate
838 * state, we just increase by 2 at the end.
840 vcpu->hv_clock.version += 2;
842 shared_kaddr = kmap_atomic(vcpu->time_page, KM_USER0);
844 memcpy(shared_kaddr + vcpu->time_offset, &vcpu->hv_clock,
845 sizeof(vcpu->hv_clock));
847 kunmap_atomic(shared_kaddr, KM_USER0);
849 mark_page_dirty(v->kvm, vcpu->time >> PAGE_SHIFT);
852 static bool msr_mtrr_valid(unsigned msr)
855 case 0x200 ... 0x200 + 2 * KVM_NR_VAR_MTRR - 1:
856 case MSR_MTRRfix64K_00000:
857 case MSR_MTRRfix16K_80000:
858 case MSR_MTRRfix16K_A0000:
859 case MSR_MTRRfix4K_C0000:
860 case MSR_MTRRfix4K_C8000:
861 case MSR_MTRRfix4K_D0000:
862 case MSR_MTRRfix4K_D8000:
863 case MSR_MTRRfix4K_E0000:
864 case MSR_MTRRfix4K_E8000:
865 case MSR_MTRRfix4K_F0000:
866 case MSR_MTRRfix4K_F8000:
867 case MSR_MTRRdefType:
868 case MSR_IA32_CR_PAT:
876 static int set_msr_mtrr(struct kvm_vcpu *vcpu, u32 msr, u64 data)
878 if (!msr_mtrr_valid(msr))
881 vcpu->arch.mtrr[msr - 0x200] = data;
885 int kvm_set_msr_common(struct kvm_vcpu *vcpu, u32 msr, u64 data)
889 set_efer(vcpu, data);
891 case MSR_IA32_MC0_STATUS:
892 pr_unimpl(vcpu, "%s: MSR_IA32_MC0_STATUS 0x%llx, nop\n",
895 case MSR_IA32_MCG_STATUS:
896 pr_unimpl(vcpu, "%s: MSR_IA32_MCG_STATUS 0x%llx, nop\n",
899 case MSR_IA32_MCG_CTL:
900 pr_unimpl(vcpu, "%s: MSR_IA32_MCG_CTL 0x%llx, nop\n",
903 case MSR_IA32_DEBUGCTLMSR:
905 /* We support the non-activated case already */
907 } else if (data & ~(DEBUGCTLMSR_LBR | DEBUGCTLMSR_BTF)) {
908 /* Values other than LBR and BTF are vendor-specific,
909 thus reserved and should throw a #GP */
912 pr_unimpl(vcpu, "%s: MSR_IA32_DEBUGCTLMSR 0x%llx, nop\n",
915 case MSR_IA32_UCODE_REV:
916 case MSR_IA32_UCODE_WRITE:
918 case 0x200 ... 0x2ff:
919 return set_msr_mtrr(vcpu, msr, data);
920 case MSR_IA32_APICBASE:
921 kvm_set_apic_base(vcpu, data);
923 case MSR_IA32_MISC_ENABLE:
924 vcpu->arch.ia32_misc_enable_msr = data;
926 case MSR_KVM_WALL_CLOCK:
927 vcpu->kvm->arch.wall_clock = data;
928 kvm_write_wall_clock(vcpu->kvm, data);
930 case MSR_KVM_SYSTEM_TIME: {
931 if (vcpu->arch.time_page) {
932 kvm_release_page_dirty(vcpu->arch.time_page);
933 vcpu->arch.time_page = NULL;
936 vcpu->arch.time = data;
938 /* we verify if the enable bit is set... */
942 /* ...but clean it before doing the actual write */
943 vcpu->arch.time_offset = data & ~(PAGE_MASK | 1);
945 down_read(¤t->mm->mmap_sem);
946 vcpu->arch.time_page =
947 gfn_to_page(vcpu->kvm, data >> PAGE_SHIFT);
948 up_read(¤t->mm->mmap_sem);
950 if (is_error_page(vcpu->arch.time_page)) {
951 kvm_release_page_clean(vcpu->arch.time_page);
952 vcpu->arch.time_page = NULL;
955 kvm_write_guest_time(vcpu);
959 pr_unimpl(vcpu, "unhandled wrmsr: 0x%x data %llx\n", msr, data);
964 EXPORT_SYMBOL_GPL(kvm_set_msr_common);
968 * Reads an msr value (of 'msr_index') into 'pdata'.
969 * Returns 0 on success, non-0 otherwise.
970 * Assumes vcpu_load() was already called.
972 int kvm_get_msr(struct kvm_vcpu *vcpu, u32 msr_index, u64 *pdata)
974 return kvm_x86_ops->get_msr(vcpu, msr_index, pdata);
977 static int get_msr_mtrr(struct kvm_vcpu *vcpu, u32 msr, u64 *pdata)
979 if (!msr_mtrr_valid(msr))
982 *pdata = vcpu->arch.mtrr[msr - 0x200];
986 int kvm_get_msr_common(struct kvm_vcpu *vcpu, u32 msr, u64 *pdata)
991 case 0xc0010010: /* SYSCFG */
992 case 0xc0010015: /* HWCR */
993 case MSR_IA32_PLATFORM_ID:
994 case MSR_IA32_P5_MC_ADDR:
995 case MSR_IA32_P5_MC_TYPE:
996 case MSR_IA32_MC0_CTL:
997 case MSR_IA32_MCG_STATUS:
998 case MSR_IA32_MCG_CAP:
999 case MSR_IA32_MCG_CTL:
1000 case MSR_IA32_MC0_MISC:
1001 case MSR_IA32_MC0_MISC+4:
1002 case MSR_IA32_MC0_MISC+8:
1003 case MSR_IA32_MC0_MISC+12:
1004 case MSR_IA32_MC0_MISC+16:
1005 case MSR_IA32_MC0_MISC+20:
1006 case MSR_IA32_UCODE_REV:
1007 case MSR_IA32_EBL_CR_POWERON:
1008 case MSR_IA32_DEBUGCTLMSR:
1009 case MSR_IA32_LASTBRANCHFROMIP:
1010 case MSR_IA32_LASTBRANCHTOIP:
1011 case MSR_IA32_LASTINTFROMIP:
1012 case MSR_IA32_LASTINTTOIP:
1016 data = 0x500 | KVM_NR_VAR_MTRR;
1018 case 0x200 ... 0x2ff:
1019 return get_msr_mtrr(vcpu, msr, pdata);
1020 case 0xcd: /* fsb frequency */
1023 case MSR_IA32_APICBASE:
1024 data = kvm_get_apic_base(vcpu);
1026 case MSR_IA32_MISC_ENABLE:
1027 data = vcpu->arch.ia32_misc_enable_msr;
1029 case MSR_IA32_PERF_STATUS:
1030 /* TSC increment by tick */
1032 /* CPU multiplier */
1033 data |= (((uint64_t)4ULL) << 40);
1036 data = vcpu->arch.shadow_efer;
1038 case MSR_KVM_WALL_CLOCK:
1039 data = vcpu->kvm->arch.wall_clock;
1041 case MSR_KVM_SYSTEM_TIME:
1042 data = vcpu->arch.time;
1045 pr_unimpl(vcpu, "unhandled rdmsr: 0x%x\n", msr);
1051 EXPORT_SYMBOL_GPL(kvm_get_msr_common);
1054 * Read or write a bunch of msrs. All parameters are kernel addresses.
1056 * @return number of msrs set successfully.
1058 static int __msr_io(struct kvm_vcpu *vcpu, struct kvm_msrs *msrs,
1059 struct kvm_msr_entry *entries,
1060 int (*do_msr)(struct kvm_vcpu *vcpu,
1061 unsigned index, u64 *data))
1067 down_read(&vcpu->kvm->slots_lock);
1068 for (i = 0; i < msrs->nmsrs; ++i)
1069 if (do_msr(vcpu, entries[i].index, &entries[i].data))
1071 up_read(&vcpu->kvm->slots_lock);
1079 * Read or write a bunch of msrs. Parameters are user addresses.
1081 * @return number of msrs set successfully.
1083 static int msr_io(struct kvm_vcpu *vcpu, struct kvm_msrs __user *user_msrs,
1084 int (*do_msr)(struct kvm_vcpu *vcpu,
1085 unsigned index, u64 *data),
1088 struct kvm_msrs msrs;
1089 struct kvm_msr_entry *entries;
1094 if (copy_from_user(&msrs, user_msrs, sizeof msrs))
1098 if (msrs.nmsrs >= MAX_IO_MSRS)
1102 size = sizeof(struct kvm_msr_entry) * msrs.nmsrs;
1103 entries = vmalloc(size);
1108 if (copy_from_user(entries, user_msrs->entries, size))
1111 r = n = __msr_io(vcpu, &msrs, entries, do_msr);
1116 if (writeback && copy_to_user(user_msrs->entries, entries, size))
1127 int kvm_dev_ioctl_check_extension(long ext)
1132 case KVM_CAP_IRQCHIP:
1134 case KVM_CAP_MMU_SHADOW_CACHE_CONTROL:
1135 case KVM_CAP_USER_MEMORY:
1136 case KVM_CAP_SET_TSS_ADDR:
1137 case KVM_CAP_EXT_CPUID:
1138 case KVM_CAP_CLOCKSOURCE:
1140 case KVM_CAP_NOP_IO_DELAY:
1141 case KVM_CAP_MP_STATE:
1142 case KVM_CAP_SYNC_MMU:
1145 case KVM_CAP_COALESCED_MMIO:
1146 r = KVM_COALESCED_MMIO_PAGE_OFFSET;
1149 r = !kvm_x86_ops->cpu_has_accelerated_tpr();
1151 case KVM_CAP_NR_VCPUS:
1154 case KVM_CAP_NR_MEMSLOTS:
1155 r = KVM_MEMORY_SLOTS;
1157 case KVM_CAP_PV_MMU:
1161 r = intel_iommu_found();
1171 long kvm_arch_dev_ioctl(struct file *filp,
1172 unsigned int ioctl, unsigned long arg)
1174 void __user *argp = (void __user *)arg;
1178 case KVM_GET_MSR_INDEX_LIST: {
1179 struct kvm_msr_list __user *user_msr_list = argp;
1180 struct kvm_msr_list msr_list;
1184 if (copy_from_user(&msr_list, user_msr_list, sizeof msr_list))
1187 msr_list.nmsrs = num_msrs_to_save + ARRAY_SIZE(emulated_msrs);
1188 if (copy_to_user(user_msr_list, &msr_list, sizeof msr_list))
1191 if (n < num_msrs_to_save)
1194 if (copy_to_user(user_msr_list->indices, &msrs_to_save,
1195 num_msrs_to_save * sizeof(u32)))
1197 if (copy_to_user(user_msr_list->indices
1198 + num_msrs_to_save * sizeof(u32),
1200 ARRAY_SIZE(emulated_msrs) * sizeof(u32)))
1205 case KVM_GET_SUPPORTED_CPUID: {
1206 struct kvm_cpuid2 __user *cpuid_arg = argp;
1207 struct kvm_cpuid2 cpuid;
1210 if (copy_from_user(&cpuid, cpuid_arg, sizeof cpuid))
1212 r = kvm_dev_ioctl_get_supported_cpuid(&cpuid,
1213 cpuid_arg->entries);
1218 if (copy_to_user(cpuid_arg, &cpuid, sizeof cpuid))
1230 void kvm_arch_vcpu_load(struct kvm_vcpu *vcpu, int cpu)
1232 kvm_x86_ops->vcpu_load(vcpu, cpu);
1233 kvm_write_guest_time(vcpu);
1236 void kvm_arch_vcpu_put(struct kvm_vcpu *vcpu)
1238 kvm_x86_ops->vcpu_put(vcpu);
1239 kvm_put_guest_fpu(vcpu);
1242 static int is_efer_nx(void)
1246 rdmsrl(MSR_EFER, efer);
1247 return efer & EFER_NX;
1250 static void cpuid_fix_nx_cap(struct kvm_vcpu *vcpu)
1253 struct kvm_cpuid_entry2 *e, *entry;
1256 for (i = 0; i < vcpu->arch.cpuid_nent; ++i) {
1257 e = &vcpu->arch.cpuid_entries[i];
1258 if (e->function == 0x80000001) {
1263 if (entry && (entry->edx & (1 << 20)) && !is_efer_nx()) {
1264 entry->edx &= ~(1 << 20);
1265 printk(KERN_INFO "kvm: guest NX capability removed\n");
1269 /* when an old userspace process fills a new kernel module */
1270 static int kvm_vcpu_ioctl_set_cpuid(struct kvm_vcpu *vcpu,
1271 struct kvm_cpuid *cpuid,
1272 struct kvm_cpuid_entry __user *entries)
1275 struct kvm_cpuid_entry *cpuid_entries;
1278 if (cpuid->nent > KVM_MAX_CPUID_ENTRIES)
1281 cpuid_entries = vmalloc(sizeof(struct kvm_cpuid_entry) * cpuid->nent);
1285 if (copy_from_user(cpuid_entries, entries,
1286 cpuid->nent * sizeof(struct kvm_cpuid_entry)))
1288 for (i = 0; i < cpuid->nent; i++) {
1289 vcpu->arch.cpuid_entries[i].function = cpuid_entries[i].function;
1290 vcpu->arch.cpuid_entries[i].eax = cpuid_entries[i].eax;
1291 vcpu->arch.cpuid_entries[i].ebx = cpuid_entries[i].ebx;
1292 vcpu->arch.cpuid_entries[i].ecx = cpuid_entries[i].ecx;
1293 vcpu->arch.cpuid_entries[i].edx = cpuid_entries[i].edx;
1294 vcpu->arch.cpuid_entries[i].index = 0;
1295 vcpu->arch.cpuid_entries[i].flags = 0;
1296 vcpu->arch.cpuid_entries[i].padding[0] = 0;
1297 vcpu->arch.cpuid_entries[i].padding[1] = 0;
1298 vcpu->arch.cpuid_entries[i].padding[2] = 0;
1300 vcpu->arch.cpuid_nent = cpuid->nent;
1301 cpuid_fix_nx_cap(vcpu);
1305 vfree(cpuid_entries);
1310 static int kvm_vcpu_ioctl_set_cpuid2(struct kvm_vcpu *vcpu,
1311 struct kvm_cpuid2 *cpuid,
1312 struct kvm_cpuid_entry2 __user *entries)
1317 if (cpuid->nent > KVM_MAX_CPUID_ENTRIES)
1320 if (copy_from_user(&vcpu->arch.cpuid_entries, entries,
1321 cpuid->nent * sizeof(struct kvm_cpuid_entry2)))
1323 vcpu->arch.cpuid_nent = cpuid->nent;
1330 static int kvm_vcpu_ioctl_get_cpuid2(struct kvm_vcpu *vcpu,
1331 struct kvm_cpuid2 *cpuid,
1332 struct kvm_cpuid_entry2 __user *entries)
1337 if (cpuid->nent < vcpu->arch.cpuid_nent)
1340 if (copy_to_user(entries, &vcpu->arch.cpuid_entries,
1341 vcpu->arch.cpuid_nent * sizeof(struct kvm_cpuid_entry2)))
1346 cpuid->nent = vcpu->arch.cpuid_nent;
1350 static inline u32 bit(int bitno)
1352 return 1 << (bitno & 31);
1355 static void do_cpuid_1_ent(struct kvm_cpuid_entry2 *entry, u32 function,
1358 entry->function = function;
1359 entry->index = index;
1360 cpuid_count(entry->function, entry->index,
1361 &entry->eax, &entry->ebx, &entry->ecx, &entry->edx);
1365 static void do_cpuid_ent(struct kvm_cpuid_entry2 *entry, u32 function,
1366 u32 index, int *nent, int maxnent)
1368 const u32 kvm_supported_word0_x86_features = bit(X86_FEATURE_FPU) |
1369 bit(X86_FEATURE_VME) | bit(X86_FEATURE_DE) |
1370 bit(X86_FEATURE_PSE) | bit(X86_FEATURE_TSC) |
1371 bit(X86_FEATURE_MSR) | bit(X86_FEATURE_PAE) |
1372 bit(X86_FEATURE_CX8) | bit(X86_FEATURE_APIC) |
1373 bit(X86_FEATURE_SEP) | bit(X86_FEATURE_PGE) |
1374 bit(X86_FEATURE_CMOV) | bit(X86_FEATURE_PSE36) |
1375 bit(X86_FEATURE_CLFLSH) | bit(X86_FEATURE_MMX) |
1376 bit(X86_FEATURE_FXSR) | bit(X86_FEATURE_XMM) |
1377 bit(X86_FEATURE_XMM2) | bit(X86_FEATURE_SELFSNOOP);
1378 const u32 kvm_supported_word1_x86_features = bit(X86_FEATURE_FPU) |
1379 bit(X86_FEATURE_VME) | bit(X86_FEATURE_DE) |
1380 bit(X86_FEATURE_PSE) | bit(X86_FEATURE_TSC) |
1381 bit(X86_FEATURE_MSR) | bit(X86_FEATURE_PAE) |
1382 bit(X86_FEATURE_CX8) | bit(X86_FEATURE_APIC) |
1383 bit(X86_FEATURE_PGE) |
1384 bit(X86_FEATURE_CMOV) | bit(X86_FEATURE_PSE36) |
1385 bit(X86_FEATURE_MMX) | bit(X86_FEATURE_FXSR) |
1386 bit(X86_FEATURE_SYSCALL) |
1387 (bit(X86_FEATURE_NX) && is_efer_nx()) |
1388 #ifdef CONFIG_X86_64
1389 bit(X86_FEATURE_LM) |
1391 bit(X86_FEATURE_MMXEXT) |
1392 bit(X86_FEATURE_3DNOWEXT) |
1393 bit(X86_FEATURE_3DNOW);
1394 const u32 kvm_supported_word3_x86_features =
1395 bit(X86_FEATURE_XMM3) | bit(X86_FEATURE_CX16);
1396 const u32 kvm_supported_word6_x86_features =
1397 bit(X86_FEATURE_LAHF_LM) | bit(X86_FEATURE_CMP_LEGACY);
1399 /* all func 2 cpuid_count() should be called on the same cpu */
1401 do_cpuid_1_ent(entry, function, index);
1406 entry->eax = min(entry->eax, (u32)0xb);
1409 entry->edx &= kvm_supported_word0_x86_features;
1410 entry->ecx &= kvm_supported_word3_x86_features;
1412 /* function 2 entries are STATEFUL. That is, repeated cpuid commands
1413 * may return different values. This forces us to get_cpu() before
1414 * issuing the first command, and also to emulate this annoying behavior
1415 * in kvm_emulate_cpuid() using KVM_CPUID_FLAG_STATE_READ_NEXT */
1417 int t, times = entry->eax & 0xff;
1419 entry->flags |= KVM_CPUID_FLAG_STATEFUL_FUNC;
1420 for (t = 1; t < times && *nent < maxnent; ++t) {
1421 do_cpuid_1_ent(&entry[t], function, 0);
1422 entry[t].flags |= KVM_CPUID_FLAG_STATEFUL_FUNC;
1427 /* function 4 and 0xb have additional index. */
1431 entry->flags |= KVM_CPUID_FLAG_SIGNIFCANT_INDEX;
1432 /* read more entries until cache_type is zero */
1433 for (i = 1; *nent < maxnent; ++i) {
1434 cache_type = entry[i - 1].eax & 0x1f;
1437 do_cpuid_1_ent(&entry[i], function, i);
1439 KVM_CPUID_FLAG_SIGNIFCANT_INDEX;
1447 entry->flags |= KVM_CPUID_FLAG_SIGNIFCANT_INDEX;
1448 /* read more entries until level_type is zero */
1449 for (i = 1; *nent < maxnent; ++i) {
1450 level_type = entry[i - 1].ecx & 0xff;
1453 do_cpuid_1_ent(&entry[i], function, i);
1455 KVM_CPUID_FLAG_SIGNIFCANT_INDEX;
1461 entry->eax = min(entry->eax, 0x8000001a);
1464 entry->edx &= kvm_supported_word1_x86_features;
1465 entry->ecx &= kvm_supported_word6_x86_features;
1471 static int kvm_dev_ioctl_get_supported_cpuid(struct kvm_cpuid2 *cpuid,
1472 struct kvm_cpuid_entry2 __user *entries)
1474 struct kvm_cpuid_entry2 *cpuid_entries;
1475 int limit, nent = 0, r = -E2BIG;
1478 if (cpuid->nent < 1)
1481 cpuid_entries = vmalloc(sizeof(struct kvm_cpuid_entry2) * cpuid->nent);
1485 do_cpuid_ent(&cpuid_entries[0], 0, 0, &nent, cpuid->nent);
1486 limit = cpuid_entries[0].eax;
1487 for (func = 1; func <= limit && nent < cpuid->nent; ++func)
1488 do_cpuid_ent(&cpuid_entries[nent], func, 0,
1489 &nent, cpuid->nent);
1491 if (nent >= cpuid->nent)
1494 do_cpuid_ent(&cpuid_entries[nent], 0x80000000, 0, &nent, cpuid->nent);
1495 limit = cpuid_entries[nent - 1].eax;
1496 for (func = 0x80000001; func <= limit && nent < cpuid->nent; ++func)
1497 do_cpuid_ent(&cpuid_entries[nent], func, 0,
1498 &nent, cpuid->nent);
1500 if (copy_to_user(entries, cpuid_entries,
1501 nent * sizeof(struct kvm_cpuid_entry2)))
1507 vfree(cpuid_entries);
1512 static int kvm_vcpu_ioctl_get_lapic(struct kvm_vcpu *vcpu,
1513 struct kvm_lapic_state *s)
1516 memcpy(s->regs, vcpu->arch.apic->regs, sizeof *s);
1522 static int kvm_vcpu_ioctl_set_lapic(struct kvm_vcpu *vcpu,
1523 struct kvm_lapic_state *s)
1526 memcpy(vcpu->arch.apic->regs, s->regs, sizeof *s);
1527 kvm_apic_post_state_restore(vcpu);
1533 static int kvm_vcpu_ioctl_interrupt(struct kvm_vcpu *vcpu,
1534 struct kvm_interrupt *irq)
1536 if (irq->irq < 0 || irq->irq >= 256)
1538 if (irqchip_in_kernel(vcpu->kvm))
1542 set_bit(irq->irq, vcpu->arch.irq_pending);
1543 set_bit(irq->irq / BITS_PER_LONG, &vcpu->arch.irq_summary);
1550 static int vcpu_ioctl_tpr_access_reporting(struct kvm_vcpu *vcpu,
1551 struct kvm_tpr_access_ctl *tac)
1555 vcpu->arch.tpr_access_reporting = !!tac->enabled;
1559 long kvm_arch_vcpu_ioctl(struct file *filp,
1560 unsigned int ioctl, unsigned long arg)
1562 struct kvm_vcpu *vcpu = filp->private_data;
1563 void __user *argp = (void __user *)arg;
1565 struct kvm_lapic_state *lapic = NULL;
1568 case KVM_GET_LAPIC: {
1569 lapic = kzalloc(sizeof(struct kvm_lapic_state), GFP_KERNEL);
1574 r = kvm_vcpu_ioctl_get_lapic(vcpu, lapic);
1578 if (copy_to_user(argp, lapic, sizeof(struct kvm_lapic_state)))
1583 case KVM_SET_LAPIC: {
1584 lapic = kmalloc(sizeof(struct kvm_lapic_state), GFP_KERNEL);
1589 if (copy_from_user(lapic, argp, sizeof(struct kvm_lapic_state)))
1591 r = kvm_vcpu_ioctl_set_lapic(vcpu, lapic);
1597 case KVM_INTERRUPT: {
1598 struct kvm_interrupt irq;
1601 if (copy_from_user(&irq, argp, sizeof irq))
1603 r = kvm_vcpu_ioctl_interrupt(vcpu, &irq);
1609 case KVM_SET_CPUID: {
1610 struct kvm_cpuid __user *cpuid_arg = argp;
1611 struct kvm_cpuid cpuid;
1614 if (copy_from_user(&cpuid, cpuid_arg, sizeof cpuid))
1616 r = kvm_vcpu_ioctl_set_cpuid(vcpu, &cpuid, cpuid_arg->entries);
1621 case KVM_SET_CPUID2: {
1622 struct kvm_cpuid2 __user *cpuid_arg = argp;
1623 struct kvm_cpuid2 cpuid;
1626 if (copy_from_user(&cpuid, cpuid_arg, sizeof cpuid))
1628 r = kvm_vcpu_ioctl_set_cpuid2(vcpu, &cpuid,
1629 cpuid_arg->entries);
1634 case KVM_GET_CPUID2: {
1635 struct kvm_cpuid2 __user *cpuid_arg = argp;
1636 struct kvm_cpuid2 cpuid;
1639 if (copy_from_user(&cpuid, cpuid_arg, sizeof cpuid))
1641 r = kvm_vcpu_ioctl_get_cpuid2(vcpu, &cpuid,
1642 cpuid_arg->entries);
1646 if (copy_to_user(cpuid_arg, &cpuid, sizeof cpuid))
1652 r = msr_io(vcpu, argp, kvm_get_msr, 1);
1655 r = msr_io(vcpu, argp, do_set_msr, 0);
1657 case KVM_TPR_ACCESS_REPORTING: {
1658 struct kvm_tpr_access_ctl tac;
1661 if (copy_from_user(&tac, argp, sizeof tac))
1663 r = vcpu_ioctl_tpr_access_reporting(vcpu, &tac);
1667 if (copy_to_user(argp, &tac, sizeof tac))
1672 case KVM_SET_VAPIC_ADDR: {
1673 struct kvm_vapic_addr va;
1676 if (!irqchip_in_kernel(vcpu->kvm))
1679 if (copy_from_user(&va, argp, sizeof va))
1682 kvm_lapic_set_vapic_addr(vcpu, va.vapic_addr);
1694 static int kvm_vm_ioctl_set_tss_addr(struct kvm *kvm, unsigned long addr)
1698 if (addr > (unsigned int)(-3 * PAGE_SIZE))
1700 ret = kvm_x86_ops->set_tss_addr(kvm, addr);
1704 static int kvm_vm_ioctl_set_nr_mmu_pages(struct kvm *kvm,
1705 u32 kvm_nr_mmu_pages)
1707 if (kvm_nr_mmu_pages < KVM_MIN_ALLOC_MMU_PAGES)
1710 down_write(&kvm->slots_lock);
1712 kvm_mmu_change_mmu_pages(kvm, kvm_nr_mmu_pages);
1713 kvm->arch.n_requested_mmu_pages = kvm_nr_mmu_pages;
1715 up_write(&kvm->slots_lock);
1719 static int kvm_vm_ioctl_get_nr_mmu_pages(struct kvm *kvm)
1721 return kvm->arch.n_alloc_mmu_pages;
1724 gfn_t unalias_gfn(struct kvm *kvm, gfn_t gfn)
1727 struct kvm_mem_alias *alias;
1729 for (i = 0; i < kvm->arch.naliases; ++i) {
1730 alias = &kvm->arch.aliases[i];
1731 if (gfn >= alias->base_gfn
1732 && gfn < alias->base_gfn + alias->npages)
1733 return alias->target_gfn + gfn - alias->base_gfn;
1739 * Set a new alias region. Aliases map a portion of physical memory into
1740 * another portion. This is useful for memory windows, for example the PC
1743 static int kvm_vm_ioctl_set_memory_alias(struct kvm *kvm,
1744 struct kvm_memory_alias *alias)
1747 struct kvm_mem_alias *p;
1750 /* General sanity checks */
1751 if (alias->memory_size & (PAGE_SIZE - 1))
1753 if (alias->guest_phys_addr & (PAGE_SIZE - 1))
1755 if (alias->slot >= KVM_ALIAS_SLOTS)
1757 if (alias->guest_phys_addr + alias->memory_size
1758 < alias->guest_phys_addr)
1760 if (alias->target_phys_addr + alias->memory_size
1761 < alias->target_phys_addr)
1764 down_write(&kvm->slots_lock);
1765 spin_lock(&kvm->mmu_lock);
1767 p = &kvm->arch.aliases[alias->slot];
1768 p->base_gfn = alias->guest_phys_addr >> PAGE_SHIFT;
1769 p->npages = alias->memory_size >> PAGE_SHIFT;
1770 p->target_gfn = alias->target_phys_addr >> PAGE_SHIFT;
1772 for (n = KVM_ALIAS_SLOTS; n > 0; --n)
1773 if (kvm->arch.aliases[n - 1].npages)
1775 kvm->arch.naliases = n;
1777 spin_unlock(&kvm->mmu_lock);
1778 kvm_mmu_zap_all(kvm);
1780 up_write(&kvm->slots_lock);
1788 static int kvm_vm_ioctl_get_irqchip(struct kvm *kvm, struct kvm_irqchip *chip)
1793 switch (chip->chip_id) {
1794 case KVM_IRQCHIP_PIC_MASTER:
1795 memcpy(&chip->chip.pic,
1796 &pic_irqchip(kvm)->pics[0],
1797 sizeof(struct kvm_pic_state));
1799 case KVM_IRQCHIP_PIC_SLAVE:
1800 memcpy(&chip->chip.pic,
1801 &pic_irqchip(kvm)->pics[1],
1802 sizeof(struct kvm_pic_state));
1804 case KVM_IRQCHIP_IOAPIC:
1805 memcpy(&chip->chip.ioapic,
1806 ioapic_irqchip(kvm),
1807 sizeof(struct kvm_ioapic_state));
1816 static int kvm_vm_ioctl_set_irqchip(struct kvm *kvm, struct kvm_irqchip *chip)
1821 switch (chip->chip_id) {
1822 case KVM_IRQCHIP_PIC_MASTER:
1823 memcpy(&pic_irqchip(kvm)->pics[0],
1825 sizeof(struct kvm_pic_state));
1827 case KVM_IRQCHIP_PIC_SLAVE:
1828 memcpy(&pic_irqchip(kvm)->pics[1],
1830 sizeof(struct kvm_pic_state));
1832 case KVM_IRQCHIP_IOAPIC:
1833 memcpy(ioapic_irqchip(kvm),
1835 sizeof(struct kvm_ioapic_state));
1841 kvm_pic_update_irq(pic_irqchip(kvm));
1845 static int kvm_vm_ioctl_get_pit(struct kvm *kvm, struct kvm_pit_state *ps)
1849 memcpy(ps, &kvm->arch.vpit->pit_state, sizeof(struct kvm_pit_state));
1853 static int kvm_vm_ioctl_set_pit(struct kvm *kvm, struct kvm_pit_state *ps)
1857 memcpy(&kvm->arch.vpit->pit_state, ps, sizeof(struct kvm_pit_state));
1858 kvm_pit_load_count(kvm, 0, ps->channels[0].count);
1863 * Get (and clear) the dirty memory log for a memory slot.
1865 int kvm_vm_ioctl_get_dirty_log(struct kvm *kvm,
1866 struct kvm_dirty_log *log)
1870 struct kvm_memory_slot *memslot;
1873 down_write(&kvm->slots_lock);
1875 r = kvm_get_dirty_log(kvm, log, &is_dirty);
1879 /* If nothing is dirty, don't bother messing with page tables. */
1881 kvm_mmu_slot_remove_write_access(kvm, log->slot);
1882 kvm_flush_remote_tlbs(kvm);
1883 memslot = &kvm->memslots[log->slot];
1884 n = ALIGN(memslot->npages, BITS_PER_LONG) / 8;
1885 memset(memslot->dirty_bitmap, 0, n);
1889 up_write(&kvm->slots_lock);
1893 long kvm_arch_vm_ioctl(struct file *filp,
1894 unsigned int ioctl, unsigned long arg)
1896 struct kvm *kvm = filp->private_data;
1897 void __user *argp = (void __user *)arg;
1900 * This union makes it completely explicit to gcc-3.x
1901 * that these two variables' stack usage should be
1902 * combined, not added together.
1905 struct kvm_pit_state ps;
1906 struct kvm_memory_alias alias;
1910 case KVM_SET_TSS_ADDR:
1911 r = kvm_vm_ioctl_set_tss_addr(kvm, arg);
1915 case KVM_SET_MEMORY_REGION: {
1916 struct kvm_memory_region kvm_mem;
1917 struct kvm_userspace_memory_region kvm_userspace_mem;
1920 if (copy_from_user(&kvm_mem, argp, sizeof kvm_mem))
1922 kvm_userspace_mem.slot = kvm_mem.slot;
1923 kvm_userspace_mem.flags = kvm_mem.flags;
1924 kvm_userspace_mem.guest_phys_addr = kvm_mem.guest_phys_addr;
1925 kvm_userspace_mem.memory_size = kvm_mem.memory_size;
1926 r = kvm_vm_ioctl_set_memory_region(kvm, &kvm_userspace_mem, 0);
1931 case KVM_SET_NR_MMU_PAGES:
1932 r = kvm_vm_ioctl_set_nr_mmu_pages(kvm, arg);
1936 case KVM_GET_NR_MMU_PAGES:
1937 r = kvm_vm_ioctl_get_nr_mmu_pages(kvm);
1939 case KVM_SET_MEMORY_ALIAS:
1941 if (copy_from_user(&u.alias, argp, sizeof(struct kvm_memory_alias)))
1943 r = kvm_vm_ioctl_set_memory_alias(kvm, &u.alias);
1947 case KVM_CREATE_IRQCHIP:
1949 kvm->arch.vpic = kvm_create_pic(kvm);
1950 if (kvm->arch.vpic) {
1951 r = kvm_ioapic_init(kvm);
1953 kfree(kvm->arch.vpic);
1954 kvm->arch.vpic = NULL;
1960 case KVM_CREATE_PIT:
1962 kvm->arch.vpit = kvm_create_pit(kvm);
1966 case KVM_IRQ_LINE: {
1967 struct kvm_irq_level irq_event;
1970 if (copy_from_user(&irq_event, argp, sizeof irq_event))
1972 if (irqchip_in_kernel(kvm)) {
1973 mutex_lock(&kvm->lock);
1974 kvm_set_irq(kvm, irq_event.irq, irq_event.level);
1975 mutex_unlock(&kvm->lock);
1980 case KVM_GET_IRQCHIP: {
1981 /* 0: PIC master, 1: PIC slave, 2: IOAPIC */
1982 struct kvm_irqchip *chip = kmalloc(sizeof(*chip), GFP_KERNEL);
1988 if (copy_from_user(chip, argp, sizeof *chip))
1989 goto get_irqchip_out;
1991 if (!irqchip_in_kernel(kvm))
1992 goto get_irqchip_out;
1993 r = kvm_vm_ioctl_get_irqchip(kvm, chip);
1995 goto get_irqchip_out;
1997 if (copy_to_user(argp, chip, sizeof *chip))
1998 goto get_irqchip_out;
2006 case KVM_SET_IRQCHIP: {
2007 /* 0: PIC master, 1: PIC slave, 2: IOAPIC */
2008 struct kvm_irqchip *chip = kmalloc(sizeof(*chip), GFP_KERNEL);
2014 if (copy_from_user(chip, argp, sizeof *chip))
2015 goto set_irqchip_out;
2017 if (!irqchip_in_kernel(kvm))
2018 goto set_irqchip_out;
2019 r = kvm_vm_ioctl_set_irqchip(kvm, chip);
2021 goto set_irqchip_out;
2029 case KVM_ASSIGN_PCI_DEVICE: {
2030 struct kvm_assigned_pci_dev assigned_dev;
2033 if (copy_from_user(&assigned_dev, argp, sizeof assigned_dev))
2035 r = kvm_vm_ioctl_assign_device(kvm, &assigned_dev);
2040 case KVM_ASSIGN_IRQ: {
2041 struct kvm_assigned_irq assigned_irq;
2044 if (copy_from_user(&assigned_irq, argp, sizeof assigned_irq))
2046 r = kvm_vm_ioctl_assign_irq(kvm, &assigned_irq);
2053 if (copy_from_user(&u.ps, argp, sizeof(struct kvm_pit_state)))
2056 if (!kvm->arch.vpit)
2058 r = kvm_vm_ioctl_get_pit(kvm, &u.ps);
2062 if (copy_to_user(argp, &u.ps, sizeof(struct kvm_pit_state)))
2069 if (copy_from_user(&u.ps, argp, sizeof u.ps))
2072 if (!kvm->arch.vpit)
2074 r = kvm_vm_ioctl_set_pit(kvm, &u.ps);
2087 static void kvm_init_msr_list(void)
2092 for (i = j = 0; i < ARRAY_SIZE(msrs_to_save); i++) {
2093 if (rdmsr_safe(msrs_to_save[i], &dummy[0], &dummy[1]) < 0)
2096 msrs_to_save[j] = msrs_to_save[i];
2099 num_msrs_to_save = j;
2103 * Only apic need an MMIO device hook, so shortcut now..
2105 static struct kvm_io_device *vcpu_find_pervcpu_dev(struct kvm_vcpu *vcpu,
2106 gpa_t addr, int len,
2109 struct kvm_io_device *dev;
2111 if (vcpu->arch.apic) {
2112 dev = &vcpu->arch.apic->dev;
2113 if (dev->in_range(dev, addr, len, is_write))
2120 static struct kvm_io_device *vcpu_find_mmio_dev(struct kvm_vcpu *vcpu,
2121 gpa_t addr, int len,
2124 struct kvm_io_device *dev;
2126 dev = vcpu_find_pervcpu_dev(vcpu, addr, len, is_write);
2128 dev = kvm_io_bus_find_dev(&vcpu->kvm->mmio_bus, addr, len,
2133 int emulator_read_std(unsigned long addr,
2136 struct kvm_vcpu *vcpu)
2139 int r = X86EMUL_CONTINUE;
2142 gpa_t gpa = vcpu->arch.mmu.gva_to_gpa(vcpu, addr);
2143 unsigned offset = addr & (PAGE_SIZE-1);
2144 unsigned tocopy = min(bytes, (unsigned)PAGE_SIZE - offset);
2147 if (gpa == UNMAPPED_GVA) {
2148 r = X86EMUL_PROPAGATE_FAULT;
2151 ret = kvm_read_guest(vcpu->kvm, gpa, data, tocopy);
2153 r = X86EMUL_UNHANDLEABLE;
2164 EXPORT_SYMBOL_GPL(emulator_read_std);
2166 static int emulator_read_emulated(unsigned long addr,
2169 struct kvm_vcpu *vcpu)
2171 struct kvm_io_device *mmio_dev;
2174 if (vcpu->mmio_read_completed) {
2175 memcpy(val, vcpu->mmio_data, bytes);
2176 vcpu->mmio_read_completed = 0;
2177 return X86EMUL_CONTINUE;
2180 gpa = vcpu->arch.mmu.gva_to_gpa(vcpu, addr);
2182 /* For APIC access vmexit */
2183 if ((gpa & PAGE_MASK) == APIC_DEFAULT_PHYS_BASE)
2186 if (emulator_read_std(addr, val, bytes, vcpu)
2187 == X86EMUL_CONTINUE)
2188 return X86EMUL_CONTINUE;
2189 if (gpa == UNMAPPED_GVA)
2190 return X86EMUL_PROPAGATE_FAULT;
2194 * Is this MMIO handled locally?
2196 mutex_lock(&vcpu->kvm->lock);
2197 mmio_dev = vcpu_find_mmio_dev(vcpu, gpa, bytes, 0);
2199 kvm_iodevice_read(mmio_dev, gpa, bytes, val);
2200 mutex_unlock(&vcpu->kvm->lock);
2201 return X86EMUL_CONTINUE;
2203 mutex_unlock(&vcpu->kvm->lock);
2205 vcpu->mmio_needed = 1;
2206 vcpu->mmio_phys_addr = gpa;
2207 vcpu->mmio_size = bytes;
2208 vcpu->mmio_is_write = 0;
2210 return X86EMUL_UNHANDLEABLE;
2213 int emulator_write_phys(struct kvm_vcpu *vcpu, gpa_t gpa,
2214 const void *val, int bytes)
2218 ret = kvm_write_guest(vcpu->kvm, gpa, val, bytes);
2221 kvm_mmu_pte_write(vcpu, gpa, val, bytes);
2225 static int emulator_write_emulated_onepage(unsigned long addr,
2228 struct kvm_vcpu *vcpu)
2230 struct kvm_io_device *mmio_dev;
2233 gpa = vcpu->arch.mmu.gva_to_gpa(vcpu, addr);
2235 if (gpa == UNMAPPED_GVA) {
2236 kvm_inject_page_fault(vcpu, addr, 2);
2237 return X86EMUL_PROPAGATE_FAULT;
2240 /* For APIC access vmexit */
2241 if ((gpa & PAGE_MASK) == APIC_DEFAULT_PHYS_BASE)
2244 if (emulator_write_phys(vcpu, gpa, val, bytes))
2245 return X86EMUL_CONTINUE;
2249 * Is this MMIO handled locally?
2251 mutex_lock(&vcpu->kvm->lock);
2252 mmio_dev = vcpu_find_mmio_dev(vcpu, gpa, bytes, 1);
2254 kvm_iodevice_write(mmio_dev, gpa, bytes, val);
2255 mutex_unlock(&vcpu->kvm->lock);
2256 return X86EMUL_CONTINUE;
2258 mutex_unlock(&vcpu->kvm->lock);
2260 vcpu->mmio_needed = 1;
2261 vcpu->mmio_phys_addr = gpa;
2262 vcpu->mmio_size = bytes;
2263 vcpu->mmio_is_write = 1;
2264 memcpy(vcpu->mmio_data, val, bytes);
2266 return X86EMUL_CONTINUE;
2269 int emulator_write_emulated(unsigned long addr,
2272 struct kvm_vcpu *vcpu)
2274 /* Crossing a page boundary? */
2275 if (((addr + bytes - 1) ^ addr) & PAGE_MASK) {
2278 now = -addr & ~PAGE_MASK;
2279 rc = emulator_write_emulated_onepage(addr, val, now, vcpu);
2280 if (rc != X86EMUL_CONTINUE)
2286 return emulator_write_emulated_onepage(addr, val, bytes, vcpu);
2288 EXPORT_SYMBOL_GPL(emulator_write_emulated);
2290 static int emulator_cmpxchg_emulated(unsigned long addr,
2294 struct kvm_vcpu *vcpu)
2296 static int reported;
2300 printk(KERN_WARNING "kvm: emulating exchange as write\n");
2302 #ifndef CONFIG_X86_64
2303 /* guests cmpxchg8b have to be emulated atomically */
2310 gpa = vcpu->arch.mmu.gva_to_gpa(vcpu, addr);
2312 if (gpa == UNMAPPED_GVA ||
2313 (gpa & PAGE_MASK) == APIC_DEFAULT_PHYS_BASE)
2316 if (((gpa + bytes - 1) & PAGE_MASK) != (gpa & PAGE_MASK))
2321 down_read(¤t->mm->mmap_sem);
2322 page = gfn_to_page(vcpu->kvm, gpa >> PAGE_SHIFT);
2323 up_read(¤t->mm->mmap_sem);
2325 kaddr = kmap_atomic(page, KM_USER0);
2326 set_64bit((u64 *)(kaddr + offset_in_page(gpa)), val);
2327 kunmap_atomic(kaddr, KM_USER0);
2328 kvm_release_page_dirty(page);
2333 return emulator_write_emulated(addr, new, bytes, vcpu);
2336 static unsigned long get_segment_base(struct kvm_vcpu *vcpu, int seg)
2338 return kvm_x86_ops->get_segment_base(vcpu, seg);
2341 int emulate_invlpg(struct kvm_vcpu *vcpu, gva_t address)
2343 return X86EMUL_CONTINUE;
2346 int emulate_clts(struct kvm_vcpu *vcpu)
2348 KVMTRACE_0D(CLTS, vcpu, handler);
2349 kvm_x86_ops->set_cr0(vcpu, vcpu->arch.cr0 & ~X86_CR0_TS);
2350 return X86EMUL_CONTINUE;
2353 int emulator_get_dr(struct x86_emulate_ctxt *ctxt, int dr, unsigned long *dest)
2355 struct kvm_vcpu *vcpu = ctxt->vcpu;
2359 *dest = kvm_x86_ops->get_dr(vcpu, dr);
2360 return X86EMUL_CONTINUE;
2362 pr_unimpl(vcpu, "%s: unexpected dr %u\n", __func__, dr);
2363 return X86EMUL_UNHANDLEABLE;
2367 int emulator_set_dr(struct x86_emulate_ctxt *ctxt, int dr, unsigned long value)
2369 unsigned long mask = (ctxt->mode == X86EMUL_MODE_PROT64) ? ~0ULL : ~0U;
2372 kvm_x86_ops->set_dr(ctxt->vcpu, dr, value & mask, &exception);
2374 /* FIXME: better handling */
2375 return X86EMUL_UNHANDLEABLE;
2377 return X86EMUL_CONTINUE;
2380 void kvm_report_emulation_failure(struct kvm_vcpu *vcpu, const char *context)
2383 unsigned long rip = kvm_rip_read(vcpu);
2384 unsigned long rip_linear;
2386 if (!printk_ratelimit())
2389 rip_linear = rip + get_segment_base(vcpu, VCPU_SREG_CS);
2391 emulator_read_std(rip_linear, (void *)opcodes, 4, vcpu);
2393 printk(KERN_ERR "emulation failed (%s) rip %lx %02x %02x %02x %02x\n",
2394 context, rip, opcodes[0], opcodes[1], opcodes[2], opcodes[3]);
2396 EXPORT_SYMBOL_GPL(kvm_report_emulation_failure);
2398 static struct x86_emulate_ops emulate_ops = {
2399 .read_std = emulator_read_std,
2400 .read_emulated = emulator_read_emulated,
2401 .write_emulated = emulator_write_emulated,
2402 .cmpxchg_emulated = emulator_cmpxchg_emulated,
2405 static void cache_all_regs(struct kvm_vcpu *vcpu)
2407 kvm_register_read(vcpu, VCPU_REGS_RAX);
2408 kvm_register_read(vcpu, VCPU_REGS_RSP);
2409 kvm_register_read(vcpu, VCPU_REGS_RIP);
2410 vcpu->arch.regs_dirty = ~0;
2413 int emulate_instruction(struct kvm_vcpu *vcpu,
2414 struct kvm_run *run,
2420 struct decode_cache *c;
2422 kvm_clear_exception_queue(vcpu);
2423 vcpu->arch.mmio_fault_cr2 = cr2;
2425 * TODO: fix x86_emulate.c to use guest_read/write_register
2426 * instead of direct ->regs accesses, can save hundred cycles
2427 * on Intel for instructions that don't read/change RSP, for
2430 cache_all_regs(vcpu);
2432 vcpu->mmio_is_write = 0;
2433 vcpu->arch.pio.string = 0;
2435 if (!(emulation_type & EMULTYPE_NO_DECODE)) {
2437 kvm_x86_ops->get_cs_db_l_bits(vcpu, &cs_db, &cs_l);
2439 vcpu->arch.emulate_ctxt.vcpu = vcpu;
2440 vcpu->arch.emulate_ctxt.eflags = kvm_x86_ops->get_rflags(vcpu);
2441 vcpu->arch.emulate_ctxt.mode =
2442 (vcpu->arch.emulate_ctxt.eflags & X86_EFLAGS_VM)
2443 ? X86EMUL_MODE_REAL : cs_l
2444 ? X86EMUL_MODE_PROT64 : cs_db
2445 ? X86EMUL_MODE_PROT32 : X86EMUL_MODE_PROT16;
2447 r = x86_decode_insn(&vcpu->arch.emulate_ctxt, &emulate_ops);
2449 /* Reject the instructions other than VMCALL/VMMCALL when
2450 * try to emulate invalid opcode */
2451 c = &vcpu->arch.emulate_ctxt.decode;
2452 if ((emulation_type & EMULTYPE_TRAP_UD) &&
2453 (!(c->twobyte && c->b == 0x01 &&
2454 (c->modrm_reg == 0 || c->modrm_reg == 3) &&
2455 c->modrm_mod == 3 && c->modrm_rm == 1)))
2456 return EMULATE_FAIL;
2458 ++vcpu->stat.insn_emulation;
2460 ++vcpu->stat.insn_emulation_fail;
2461 if (kvm_mmu_unprotect_page_virt(vcpu, cr2))
2462 return EMULATE_DONE;
2463 return EMULATE_FAIL;
2467 r = x86_emulate_insn(&vcpu->arch.emulate_ctxt, &emulate_ops);
2469 if (vcpu->arch.pio.string)
2470 return EMULATE_DO_MMIO;
2472 if ((r || vcpu->mmio_is_write) && run) {
2473 run->exit_reason = KVM_EXIT_MMIO;
2474 run->mmio.phys_addr = vcpu->mmio_phys_addr;
2475 memcpy(run->mmio.data, vcpu->mmio_data, 8);
2476 run->mmio.len = vcpu->mmio_size;
2477 run->mmio.is_write = vcpu->mmio_is_write;
2481 if (kvm_mmu_unprotect_page_virt(vcpu, cr2))
2482 return EMULATE_DONE;
2483 if (!vcpu->mmio_needed) {
2484 kvm_report_emulation_failure(vcpu, "mmio");
2485 return EMULATE_FAIL;
2487 return EMULATE_DO_MMIO;
2490 kvm_x86_ops->set_rflags(vcpu, vcpu->arch.emulate_ctxt.eflags);
2492 if (vcpu->mmio_is_write) {
2493 vcpu->mmio_needed = 0;
2494 return EMULATE_DO_MMIO;
2497 return EMULATE_DONE;
2499 EXPORT_SYMBOL_GPL(emulate_instruction);
2501 static void free_pio_guest_pages(struct kvm_vcpu *vcpu)
2505 for (i = 0; i < ARRAY_SIZE(vcpu->arch.pio.guest_pages); ++i)
2506 if (vcpu->arch.pio.guest_pages[i]) {
2507 kvm_release_page_dirty(vcpu->arch.pio.guest_pages[i]);
2508 vcpu->arch.pio.guest_pages[i] = NULL;
2512 static int pio_copy_data(struct kvm_vcpu *vcpu)
2514 void *p = vcpu->arch.pio_data;
2517 int nr_pages = vcpu->arch.pio.guest_pages[1] ? 2 : 1;
2519 q = vmap(vcpu->arch.pio.guest_pages, nr_pages, VM_READ|VM_WRITE,
2522 free_pio_guest_pages(vcpu);
2525 q += vcpu->arch.pio.guest_page_offset;
2526 bytes = vcpu->arch.pio.size * vcpu->arch.pio.cur_count;
2527 if (vcpu->arch.pio.in)
2528 memcpy(q, p, bytes);
2530 memcpy(p, q, bytes);
2531 q -= vcpu->arch.pio.guest_page_offset;
2533 free_pio_guest_pages(vcpu);
2537 int complete_pio(struct kvm_vcpu *vcpu)
2539 struct kvm_pio_request *io = &vcpu->arch.pio;
2546 val = kvm_register_read(vcpu, VCPU_REGS_RAX);
2547 memcpy(&val, vcpu->arch.pio_data, io->size);
2548 kvm_register_write(vcpu, VCPU_REGS_RAX, val);
2552 r = pio_copy_data(vcpu);
2559 delta *= io->cur_count;
2561 * The size of the register should really depend on
2562 * current address size.
2564 val = kvm_register_read(vcpu, VCPU_REGS_RCX);
2566 kvm_register_write(vcpu, VCPU_REGS_RCX, val);
2572 val = kvm_register_read(vcpu, VCPU_REGS_RDI);
2574 kvm_register_write(vcpu, VCPU_REGS_RDI, val);
2576 val = kvm_register_read(vcpu, VCPU_REGS_RSI);
2578 kvm_register_write(vcpu, VCPU_REGS_RSI, val);
2582 io->count -= io->cur_count;
2588 static void kernel_pio(struct kvm_io_device *pio_dev,
2589 struct kvm_vcpu *vcpu,
2592 /* TODO: String I/O for in kernel device */
2594 mutex_lock(&vcpu->kvm->lock);
2595 if (vcpu->arch.pio.in)
2596 kvm_iodevice_read(pio_dev, vcpu->arch.pio.port,
2597 vcpu->arch.pio.size,
2600 kvm_iodevice_write(pio_dev, vcpu->arch.pio.port,
2601 vcpu->arch.pio.size,
2603 mutex_unlock(&vcpu->kvm->lock);
2606 static void pio_string_write(struct kvm_io_device *pio_dev,
2607 struct kvm_vcpu *vcpu)
2609 struct kvm_pio_request *io = &vcpu->arch.pio;
2610 void *pd = vcpu->arch.pio_data;
2613 mutex_lock(&vcpu->kvm->lock);
2614 for (i = 0; i < io->cur_count; i++) {
2615 kvm_iodevice_write(pio_dev, io->port,
2620 mutex_unlock(&vcpu->kvm->lock);
2623 static struct kvm_io_device *vcpu_find_pio_dev(struct kvm_vcpu *vcpu,
2624 gpa_t addr, int len,
2627 return kvm_io_bus_find_dev(&vcpu->kvm->pio_bus, addr, len, is_write);
2630 int kvm_emulate_pio(struct kvm_vcpu *vcpu, struct kvm_run *run, int in,
2631 int size, unsigned port)
2633 struct kvm_io_device *pio_dev;
2636 vcpu->run->exit_reason = KVM_EXIT_IO;
2637 vcpu->run->io.direction = in ? KVM_EXIT_IO_IN : KVM_EXIT_IO_OUT;
2638 vcpu->run->io.size = vcpu->arch.pio.size = size;
2639 vcpu->run->io.data_offset = KVM_PIO_PAGE_OFFSET * PAGE_SIZE;
2640 vcpu->run->io.count = vcpu->arch.pio.count = vcpu->arch.pio.cur_count = 1;
2641 vcpu->run->io.port = vcpu->arch.pio.port = port;
2642 vcpu->arch.pio.in = in;
2643 vcpu->arch.pio.string = 0;
2644 vcpu->arch.pio.down = 0;
2645 vcpu->arch.pio.guest_page_offset = 0;
2646 vcpu->arch.pio.rep = 0;
2648 if (vcpu->run->io.direction == KVM_EXIT_IO_IN)
2649 KVMTRACE_2D(IO_READ, vcpu, vcpu->run->io.port, (u32)size,
2652 KVMTRACE_2D(IO_WRITE, vcpu, vcpu->run->io.port, (u32)size,
2655 val = kvm_register_read(vcpu, VCPU_REGS_RAX);
2656 memcpy(vcpu->arch.pio_data, &val, 4);
2658 kvm_x86_ops->skip_emulated_instruction(vcpu);
2660 pio_dev = vcpu_find_pio_dev(vcpu, port, size, !in);
2662 kernel_pio(pio_dev, vcpu, vcpu->arch.pio_data);
2668 EXPORT_SYMBOL_GPL(kvm_emulate_pio);
2670 int kvm_emulate_pio_string(struct kvm_vcpu *vcpu, struct kvm_run *run, int in,
2671 int size, unsigned long count, int down,
2672 gva_t address, int rep, unsigned port)
2674 unsigned now, in_page;
2678 struct kvm_io_device *pio_dev;
2680 vcpu->run->exit_reason = KVM_EXIT_IO;
2681 vcpu->run->io.direction = in ? KVM_EXIT_IO_IN : KVM_EXIT_IO_OUT;
2682 vcpu->run->io.size = vcpu->arch.pio.size = size;
2683 vcpu->run->io.data_offset = KVM_PIO_PAGE_OFFSET * PAGE_SIZE;
2684 vcpu->run->io.count = vcpu->arch.pio.count = vcpu->arch.pio.cur_count = count;
2685 vcpu->run->io.port = vcpu->arch.pio.port = port;
2686 vcpu->arch.pio.in = in;
2687 vcpu->arch.pio.string = 1;
2688 vcpu->arch.pio.down = down;
2689 vcpu->arch.pio.guest_page_offset = offset_in_page(address);
2690 vcpu->arch.pio.rep = rep;
2692 if (vcpu->run->io.direction == KVM_EXIT_IO_IN)
2693 KVMTRACE_2D(IO_READ, vcpu, vcpu->run->io.port, (u32)size,
2696 KVMTRACE_2D(IO_WRITE, vcpu, vcpu->run->io.port, (u32)size,
2700 kvm_x86_ops->skip_emulated_instruction(vcpu);
2705 in_page = PAGE_SIZE - offset_in_page(address);
2707 in_page = offset_in_page(address) + size;
2708 now = min(count, (unsigned long)in_page / size);
2711 * String I/O straddles page boundary. Pin two guest pages
2712 * so that we satisfy atomicity constraints. Do just one
2713 * transaction to avoid complexity.
2720 * String I/O in reverse. Yuck. Kill the guest, fix later.
2722 pr_unimpl(vcpu, "guest string pio down\n");
2723 kvm_inject_gp(vcpu, 0);
2726 vcpu->run->io.count = now;
2727 vcpu->arch.pio.cur_count = now;
2729 if (vcpu->arch.pio.cur_count == vcpu->arch.pio.count)
2730 kvm_x86_ops->skip_emulated_instruction(vcpu);
2732 for (i = 0; i < nr_pages; ++i) {
2733 page = gva_to_page(vcpu, address + i * PAGE_SIZE);
2734 vcpu->arch.pio.guest_pages[i] = page;
2736 kvm_inject_gp(vcpu, 0);
2737 free_pio_guest_pages(vcpu);
2742 pio_dev = vcpu_find_pio_dev(vcpu, port,
2743 vcpu->arch.pio.cur_count,
2744 !vcpu->arch.pio.in);
2745 if (!vcpu->arch.pio.in) {
2746 /* string PIO write */
2747 ret = pio_copy_data(vcpu);
2748 if (ret >= 0 && pio_dev) {
2749 pio_string_write(pio_dev, vcpu);
2751 if (vcpu->arch.pio.count == 0)
2755 pr_unimpl(vcpu, "no string pio read support yet, "
2756 "port %x size %d count %ld\n",
2761 EXPORT_SYMBOL_GPL(kvm_emulate_pio_string);
2763 int kvm_arch_init(void *opaque)
2766 struct kvm_x86_ops *ops = (struct kvm_x86_ops *)opaque;
2769 printk(KERN_ERR "kvm: already loaded the other module\n");
2774 if (!ops->cpu_has_kvm_support()) {
2775 printk(KERN_ERR "kvm: no hardware support\n");
2779 if (ops->disabled_by_bios()) {
2780 printk(KERN_ERR "kvm: disabled by bios\n");
2785 r = kvm_mmu_module_init();
2789 kvm_init_msr_list();
2792 kvm_mmu_set_nonpresent_ptes(0ull, 0ull);
2793 kvm_mmu_set_base_ptes(PT_PRESENT_MASK);
2794 kvm_mmu_set_mask_ptes(PT_USER_MASK, PT_ACCESSED_MASK,
2795 PT_DIRTY_MASK, PT64_NX_MASK, 0);
2802 void kvm_arch_exit(void)
2805 kvm_mmu_module_exit();
2808 int kvm_emulate_halt(struct kvm_vcpu *vcpu)
2810 ++vcpu->stat.halt_exits;
2811 KVMTRACE_0D(HLT, vcpu, handler);
2812 if (irqchip_in_kernel(vcpu->kvm)) {
2813 vcpu->arch.mp_state = KVM_MP_STATE_HALTED;
2816 vcpu->run->exit_reason = KVM_EXIT_HLT;
2820 EXPORT_SYMBOL_GPL(kvm_emulate_halt);
2822 static inline gpa_t hc_gpa(struct kvm_vcpu *vcpu, unsigned long a0,
2825 if (is_long_mode(vcpu))
2828 return a0 | ((gpa_t)a1 << 32);
2831 int kvm_emulate_hypercall(struct kvm_vcpu *vcpu)
2833 unsigned long nr, a0, a1, a2, a3, ret;
2836 nr = kvm_register_read(vcpu, VCPU_REGS_RAX);
2837 a0 = kvm_register_read(vcpu, VCPU_REGS_RBX);
2838 a1 = kvm_register_read(vcpu, VCPU_REGS_RCX);
2839 a2 = kvm_register_read(vcpu, VCPU_REGS_RDX);
2840 a3 = kvm_register_read(vcpu, VCPU_REGS_RSI);
2842 KVMTRACE_1D(VMMCALL, vcpu, (u32)nr, handler);
2844 if (!is_long_mode(vcpu)) {
2853 case KVM_HC_VAPIC_POLL_IRQ:
2857 r = kvm_pv_mmu_op(vcpu, a0, hc_gpa(vcpu, a1, a2), &ret);
2863 kvm_register_write(vcpu, VCPU_REGS_RAX, ret);
2864 ++vcpu->stat.hypercalls;
2867 EXPORT_SYMBOL_GPL(kvm_emulate_hypercall);
2869 int kvm_fix_hypercall(struct kvm_vcpu *vcpu)
2871 char instruction[3];
2873 unsigned long rip = kvm_rip_read(vcpu);
2877 * Blow out the MMU to ensure that no other VCPU has an active mapping
2878 * to ensure that the updated hypercall appears atomically across all
2881 kvm_mmu_zap_all(vcpu->kvm);
2883 kvm_x86_ops->patch_hypercall(vcpu, instruction);
2884 if (emulator_write_emulated(rip, instruction, 3, vcpu)
2885 != X86EMUL_CONTINUE)
2891 static u64 mk_cr_64(u64 curr_cr, u32 new_val)
2893 return (curr_cr & ~((1ULL << 32) - 1)) | new_val;
2896 void realmode_lgdt(struct kvm_vcpu *vcpu, u16 limit, unsigned long base)
2898 struct descriptor_table dt = { limit, base };
2900 kvm_x86_ops->set_gdt(vcpu, &dt);
2903 void realmode_lidt(struct kvm_vcpu *vcpu, u16 limit, unsigned long base)
2905 struct descriptor_table dt = { limit, base };
2907 kvm_x86_ops->set_idt(vcpu, &dt);
2910 void realmode_lmsw(struct kvm_vcpu *vcpu, unsigned long msw,
2911 unsigned long *rflags)
2913 kvm_lmsw(vcpu, msw);
2914 *rflags = kvm_x86_ops->get_rflags(vcpu);
2917 unsigned long realmode_get_cr(struct kvm_vcpu *vcpu, int cr)
2919 unsigned long value;
2921 kvm_x86_ops->decache_cr4_guest_bits(vcpu);
2924 value = vcpu->arch.cr0;
2927 value = vcpu->arch.cr2;
2930 value = vcpu->arch.cr3;
2933 value = vcpu->arch.cr4;
2936 value = kvm_get_cr8(vcpu);
2939 vcpu_printf(vcpu, "%s: unexpected cr %u\n", __func__, cr);
2942 KVMTRACE_3D(CR_READ, vcpu, (u32)cr, (u32)value,
2943 (u32)((u64)value >> 32), handler);
2948 void realmode_set_cr(struct kvm_vcpu *vcpu, int cr, unsigned long val,
2949 unsigned long *rflags)
2951 KVMTRACE_3D(CR_WRITE, vcpu, (u32)cr, (u32)val,
2952 (u32)((u64)val >> 32), handler);
2956 kvm_set_cr0(vcpu, mk_cr_64(vcpu->arch.cr0, val));
2957 *rflags = kvm_x86_ops->get_rflags(vcpu);
2960 vcpu->arch.cr2 = val;
2963 kvm_set_cr3(vcpu, val);
2966 kvm_set_cr4(vcpu, mk_cr_64(vcpu->arch.cr4, val));
2969 kvm_set_cr8(vcpu, val & 0xfUL);
2972 vcpu_printf(vcpu, "%s: unexpected cr %u\n", __func__, cr);
2976 static int move_to_next_stateful_cpuid_entry(struct kvm_vcpu *vcpu, int i)
2978 struct kvm_cpuid_entry2 *e = &vcpu->arch.cpuid_entries[i];
2979 int j, nent = vcpu->arch.cpuid_nent;
2981 e->flags &= ~KVM_CPUID_FLAG_STATE_READ_NEXT;
2982 /* when no next entry is found, the current entry[i] is reselected */
2983 for (j = i + 1; j == i; j = (j + 1) % nent) {
2984 struct kvm_cpuid_entry2 *ej = &vcpu->arch.cpuid_entries[j];
2985 if (ej->function == e->function) {
2986 ej->flags |= KVM_CPUID_FLAG_STATE_READ_NEXT;
2990 return 0; /* silence gcc, even though control never reaches here */
2993 /* find an entry with matching function, matching index (if needed), and that
2994 * should be read next (if it's stateful) */
2995 static int is_matching_cpuid_entry(struct kvm_cpuid_entry2 *e,
2996 u32 function, u32 index)
2998 if (e->function != function)
3000 if ((e->flags & KVM_CPUID_FLAG_SIGNIFCANT_INDEX) && e->index != index)
3002 if ((e->flags & KVM_CPUID_FLAG_STATEFUL_FUNC) &&
3003 !(e->flags & KVM_CPUID_FLAG_STATE_READ_NEXT))
3008 void kvm_emulate_cpuid(struct kvm_vcpu *vcpu)
3011 u32 function, index;
3012 struct kvm_cpuid_entry2 *e, *best;
3014 function = kvm_register_read(vcpu, VCPU_REGS_RAX);
3015 index = kvm_register_read(vcpu, VCPU_REGS_RCX);
3016 kvm_register_write(vcpu, VCPU_REGS_RAX, 0);
3017 kvm_register_write(vcpu, VCPU_REGS_RBX, 0);
3018 kvm_register_write(vcpu, VCPU_REGS_RCX, 0);
3019 kvm_register_write(vcpu, VCPU_REGS_RDX, 0);
3021 for (i = 0; i < vcpu->arch.cpuid_nent; ++i) {
3022 e = &vcpu->arch.cpuid_entries[i];
3023 if (is_matching_cpuid_entry(e, function, index)) {
3024 if (e->flags & KVM_CPUID_FLAG_STATEFUL_FUNC)
3025 move_to_next_stateful_cpuid_entry(vcpu, i);
3030 * Both basic or both extended?
3032 if (((e->function ^ function) & 0x80000000) == 0)
3033 if (!best || e->function > best->function)
3037 kvm_register_write(vcpu, VCPU_REGS_RAX, best->eax);
3038 kvm_register_write(vcpu, VCPU_REGS_RBX, best->ebx);
3039 kvm_register_write(vcpu, VCPU_REGS_RCX, best->ecx);
3040 kvm_register_write(vcpu, VCPU_REGS_RDX, best->edx);
3042 kvm_x86_ops->skip_emulated_instruction(vcpu);
3043 KVMTRACE_5D(CPUID, vcpu, function,
3044 (u32)kvm_register_read(vcpu, VCPU_REGS_RAX),
3045 (u32)kvm_register_read(vcpu, VCPU_REGS_RBX),
3046 (u32)kvm_register_read(vcpu, VCPU_REGS_RCX),
3047 (u32)kvm_register_read(vcpu, VCPU_REGS_RDX), handler);
3049 EXPORT_SYMBOL_GPL(kvm_emulate_cpuid);
3052 * Check if userspace requested an interrupt window, and that the
3053 * interrupt window is open.
3055 * No need to exit to userspace if we already have an interrupt queued.
3057 static int dm_request_for_irq_injection(struct kvm_vcpu *vcpu,
3058 struct kvm_run *kvm_run)
3060 return (!vcpu->arch.irq_summary &&
3061 kvm_run->request_interrupt_window &&
3062 vcpu->arch.interrupt_window_open &&
3063 (kvm_x86_ops->get_rflags(vcpu) & X86_EFLAGS_IF));
3066 static void post_kvm_run_save(struct kvm_vcpu *vcpu,
3067 struct kvm_run *kvm_run)
3069 kvm_run->if_flag = (kvm_x86_ops->get_rflags(vcpu) & X86_EFLAGS_IF) != 0;
3070 kvm_run->cr8 = kvm_get_cr8(vcpu);
3071 kvm_run->apic_base = kvm_get_apic_base(vcpu);
3072 if (irqchip_in_kernel(vcpu->kvm))
3073 kvm_run->ready_for_interrupt_injection = 1;
3075 kvm_run->ready_for_interrupt_injection =
3076 (vcpu->arch.interrupt_window_open &&
3077 vcpu->arch.irq_summary == 0);
3080 static void vapic_enter(struct kvm_vcpu *vcpu)
3082 struct kvm_lapic *apic = vcpu->arch.apic;
3085 if (!apic || !apic->vapic_addr)
3088 down_read(¤t->mm->mmap_sem);
3089 page = gfn_to_page(vcpu->kvm, apic->vapic_addr >> PAGE_SHIFT);
3090 up_read(¤t->mm->mmap_sem);
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 (kvm_arch_vcpu_runnable(vcpu))
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 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_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);