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>
39 #include <asm/uaccess.h>
43 #define MAX_IO_MSRS 256
44 #define CR0_RESERVED_BITS \
45 (~(unsigned long)(X86_CR0_PE | X86_CR0_MP | X86_CR0_EM | X86_CR0_TS \
46 | X86_CR0_ET | X86_CR0_NE | X86_CR0_WP | X86_CR0_AM \
47 | X86_CR0_NW | X86_CR0_CD | X86_CR0_PG))
48 #define CR4_RESERVED_BITS \
49 (~(unsigned long)(X86_CR4_VME | X86_CR4_PVI | X86_CR4_TSD | X86_CR4_DE\
50 | X86_CR4_PSE | X86_CR4_PAE | X86_CR4_MCE \
51 | X86_CR4_PGE | X86_CR4_PCE | X86_CR4_OSFXSR \
52 | X86_CR4_OSXMMEXCPT | X86_CR4_VMXE))
54 #define CR8_RESERVED_BITS (~(unsigned long)X86_CR8_TPR)
56 * - enable syscall per default because its emulated by KVM
57 * - enable LME and LMA per default on 64 bit KVM
60 static u64 __read_mostly efer_reserved_bits
= 0xfffffffffffffafeULL
;
62 static u64 __read_mostly efer_reserved_bits
= 0xfffffffffffffffeULL
;
65 #define VM_STAT(x) offsetof(struct kvm, stat.x), KVM_STAT_VM
66 #define VCPU_STAT(x) offsetof(struct kvm_vcpu, stat.x), KVM_STAT_VCPU
68 static int kvm_dev_ioctl_get_supported_cpuid(struct kvm_cpuid2
*cpuid
,
69 struct kvm_cpuid_entry2 __user
*entries
);
71 struct kvm_x86_ops
*kvm_x86_ops
;
72 EXPORT_SYMBOL_GPL(kvm_x86_ops
);
74 struct kvm_stats_debugfs_item debugfs_entries
[] = {
75 { "pf_fixed", VCPU_STAT(pf_fixed
) },
76 { "pf_guest", VCPU_STAT(pf_guest
) },
77 { "tlb_flush", VCPU_STAT(tlb_flush
) },
78 { "invlpg", VCPU_STAT(invlpg
) },
79 { "exits", VCPU_STAT(exits
) },
80 { "io_exits", VCPU_STAT(io_exits
) },
81 { "mmio_exits", VCPU_STAT(mmio_exits
) },
82 { "signal_exits", VCPU_STAT(signal_exits
) },
83 { "irq_window", VCPU_STAT(irq_window_exits
) },
84 { "nmi_window", VCPU_STAT(nmi_window_exits
) },
85 { "halt_exits", VCPU_STAT(halt_exits
) },
86 { "halt_wakeup", VCPU_STAT(halt_wakeup
) },
87 { "hypercalls", VCPU_STAT(hypercalls
) },
88 { "request_irq", VCPU_STAT(request_irq_exits
) },
89 { "irq_exits", VCPU_STAT(irq_exits
) },
90 { "host_state_reload", VCPU_STAT(host_state_reload
) },
91 { "efer_reload", VCPU_STAT(efer_reload
) },
92 { "fpu_reload", VCPU_STAT(fpu_reload
) },
93 { "insn_emulation", VCPU_STAT(insn_emulation
) },
94 { "insn_emulation_fail", VCPU_STAT(insn_emulation_fail
) },
95 { "mmu_shadow_zapped", VM_STAT(mmu_shadow_zapped
) },
96 { "mmu_pte_write", VM_STAT(mmu_pte_write
) },
97 { "mmu_pte_updated", VM_STAT(mmu_pte_updated
) },
98 { "mmu_pde_zapped", VM_STAT(mmu_pde_zapped
) },
99 { "mmu_flooded", VM_STAT(mmu_flooded
) },
100 { "mmu_recycled", VM_STAT(mmu_recycled
) },
101 { "mmu_cache_miss", VM_STAT(mmu_cache_miss
) },
102 { "remote_tlb_flush", VM_STAT(remote_tlb_flush
) },
103 { "largepages", VM_STAT(lpages
) },
107 static struct kvm_assigned_dev_kernel
*kvm_find_assigned_dev(struct list_head
*head
,
110 struct list_head
*ptr
;
111 struct kvm_assigned_dev_kernel
*match
;
113 list_for_each(ptr
, head
) {
114 match
= list_entry(ptr
, struct kvm_assigned_dev_kernel
, list
);
115 if (match
->assigned_dev_id
== assigned_dev_id
)
121 static void kvm_assigned_dev_interrupt_work_handler(struct work_struct
*work
)
123 struct kvm_assigned_dev_kernel
*assigned_dev
;
125 assigned_dev
= container_of(work
, struct kvm_assigned_dev_kernel
,
128 /* This is taken to safely inject irq inside the guest. When
129 * the interrupt injection (or the ioapic code) uses a
130 * finer-grained lock, update this
132 mutex_lock(&assigned_dev
->kvm
->lock
);
133 kvm_set_irq(assigned_dev
->kvm
,
134 assigned_dev
->guest_irq
, 1);
135 mutex_unlock(&assigned_dev
->kvm
->lock
);
136 kvm_put_kvm(assigned_dev
->kvm
);
139 /* FIXME: Implement the OR logic needed to make shared interrupts on
140 * this line behave properly
142 static irqreturn_t
kvm_assigned_dev_intr(int irq
, void *dev_id
)
144 struct kvm_assigned_dev_kernel
*assigned_dev
=
145 (struct kvm_assigned_dev_kernel
*) dev_id
;
147 kvm_get_kvm(assigned_dev
->kvm
);
148 schedule_work(&assigned_dev
->interrupt_work
);
149 disable_irq_nosync(irq
);
153 /* Ack the irq line for an assigned device */
154 static void kvm_assigned_dev_ack_irq(struct kvm_irq_ack_notifier
*kian
)
156 struct kvm_assigned_dev_kernel
*dev
;
161 dev
= container_of(kian
, struct kvm_assigned_dev_kernel
,
163 kvm_set_irq(dev
->kvm
, dev
->guest_irq
, 0);
164 enable_irq(dev
->host_irq
);
167 static int kvm_vm_ioctl_assign_irq(struct kvm
*kvm
,
168 struct kvm_assigned_irq
172 struct kvm_assigned_dev_kernel
*match
;
174 mutex_lock(&kvm
->lock
);
176 match
= kvm_find_assigned_dev(&kvm
->arch
.assigned_dev_head
,
177 assigned_irq
->assigned_dev_id
);
179 mutex_unlock(&kvm
->lock
);
183 if (match
->irq_requested
) {
184 match
->guest_irq
= assigned_irq
->guest_irq
;
185 match
->ack_notifier
.gsi
= assigned_irq
->guest_irq
;
186 mutex_unlock(&kvm
->lock
);
190 INIT_WORK(&match
->interrupt_work
,
191 kvm_assigned_dev_interrupt_work_handler
);
193 if (irqchip_in_kernel(kvm
)) {
194 if (!capable(CAP_SYS_RAWIO
)) {
199 if (assigned_irq
->host_irq
)
200 match
->host_irq
= assigned_irq
->host_irq
;
202 match
->host_irq
= match
->dev
->irq
;
203 match
->guest_irq
= assigned_irq
->guest_irq
;
204 match
->ack_notifier
.gsi
= assigned_irq
->guest_irq
;
205 match
->ack_notifier
.irq_acked
= kvm_assigned_dev_ack_irq
;
206 kvm_register_irq_ack_notifier(kvm
, &match
->ack_notifier
);
208 /* Even though this is PCI, we don't want to use shared
209 * interrupts. Sharing host devices with guest-assigned devices
210 * on the same interrupt line is not a happy situation: there
211 * are going to be long delays in accepting, acking, etc.
213 if (request_irq(match
->host_irq
, kvm_assigned_dev_intr
, 0,
214 "kvm_assigned_device", (void *)match
)) {
215 printk(KERN_INFO
"%s: couldn't allocate irq for pv "
216 "device\n", __func__
);
222 match
->irq_requested
= true;
224 mutex_unlock(&kvm
->lock
);
228 static int kvm_vm_ioctl_assign_device(struct kvm
*kvm
,
229 struct kvm_assigned_pci_dev
*assigned_dev
)
232 struct kvm_assigned_dev_kernel
*match
;
235 mutex_lock(&kvm
->lock
);
237 match
= kvm_find_assigned_dev(&kvm
->arch
.assigned_dev_head
,
238 assigned_dev
->assigned_dev_id
);
240 /* device already assigned */
245 match
= kzalloc(sizeof(struct kvm_assigned_dev_kernel
), GFP_KERNEL
);
247 printk(KERN_INFO
"%s: Couldn't allocate memory\n",
252 dev
= pci_get_bus_and_slot(assigned_dev
->busnr
,
253 assigned_dev
->devfn
);
255 printk(KERN_INFO
"%s: host device not found\n", __func__
);
259 if (pci_enable_device(dev
)) {
260 printk(KERN_INFO
"%s: Could not enable PCI device\n", __func__
);
264 r
= pci_request_regions(dev
, "kvm_assigned_device");
266 printk(KERN_INFO
"%s: Could not get access to device regions\n",
270 match
->assigned_dev_id
= assigned_dev
->assigned_dev_id
;
271 match
->host_busnr
= assigned_dev
->busnr
;
272 match
->host_devfn
= assigned_dev
->devfn
;
277 list_add(&match
->list
, &kvm
->arch
.assigned_dev_head
);
280 mutex_unlock(&kvm
->lock
);
283 pci_disable_device(dev
);
288 mutex_unlock(&kvm
->lock
);
292 static void kvm_free_assigned_devices(struct kvm
*kvm
)
294 struct list_head
*ptr
, *ptr2
;
295 struct kvm_assigned_dev_kernel
*assigned_dev
;
297 list_for_each_safe(ptr
, ptr2
, &kvm
->arch
.assigned_dev_head
) {
298 assigned_dev
= list_entry(ptr
,
299 struct kvm_assigned_dev_kernel
,
302 if (irqchip_in_kernel(kvm
) && assigned_dev
->irq_requested
) {
303 free_irq(assigned_dev
->host_irq
,
304 (void *)assigned_dev
);
306 kvm_unregister_irq_ack_notifier(kvm
,
311 if (cancel_work_sync(&assigned_dev
->interrupt_work
))
312 /* We had pending work. That means we will have to take
313 * care of kvm_put_kvm.
317 pci_release_regions(assigned_dev
->dev
);
318 pci_disable_device(assigned_dev
->dev
);
319 pci_dev_put(assigned_dev
->dev
);
321 list_del(&assigned_dev
->list
);
326 unsigned long segment_base(u16 selector
)
328 struct descriptor_table gdt
;
329 struct desc_struct
*d
;
330 unsigned long table_base
;
336 asm("sgdt %0" : "=m"(gdt
));
337 table_base
= gdt
.base
;
339 if (selector
& 4) { /* from ldt */
342 asm("sldt %0" : "=g"(ldt_selector
));
343 table_base
= segment_base(ldt_selector
);
345 d
= (struct desc_struct
*)(table_base
+ (selector
& ~7));
346 v
= d
->base0
| ((unsigned long)d
->base1
<< 16) |
347 ((unsigned long)d
->base2
<< 24);
349 if (d
->s
== 0 && (d
->type
== 2 || d
->type
== 9 || d
->type
== 11))
350 v
|= ((unsigned long)((struct ldttss_desc64
*)d
)->base3
) << 32;
354 EXPORT_SYMBOL_GPL(segment_base
);
356 u64
kvm_get_apic_base(struct kvm_vcpu
*vcpu
)
358 if (irqchip_in_kernel(vcpu
->kvm
))
359 return vcpu
->arch
.apic_base
;
361 return vcpu
->arch
.apic_base
;
363 EXPORT_SYMBOL_GPL(kvm_get_apic_base
);
365 void kvm_set_apic_base(struct kvm_vcpu
*vcpu
, u64 data
)
367 /* TODO: reserve bits check */
368 if (irqchip_in_kernel(vcpu
->kvm
))
369 kvm_lapic_set_base(vcpu
, data
);
371 vcpu
->arch
.apic_base
= data
;
373 EXPORT_SYMBOL_GPL(kvm_set_apic_base
);
375 void kvm_queue_exception(struct kvm_vcpu
*vcpu
, unsigned nr
)
377 WARN_ON(vcpu
->arch
.exception
.pending
);
378 vcpu
->arch
.exception
.pending
= true;
379 vcpu
->arch
.exception
.has_error_code
= false;
380 vcpu
->arch
.exception
.nr
= nr
;
382 EXPORT_SYMBOL_GPL(kvm_queue_exception
);
384 void kvm_inject_page_fault(struct kvm_vcpu
*vcpu
, unsigned long addr
,
387 ++vcpu
->stat
.pf_guest
;
388 if (vcpu
->arch
.exception
.pending
) {
389 if (vcpu
->arch
.exception
.nr
== PF_VECTOR
) {
390 printk(KERN_DEBUG
"kvm: inject_page_fault:"
391 " double fault 0x%lx\n", addr
);
392 vcpu
->arch
.exception
.nr
= DF_VECTOR
;
393 vcpu
->arch
.exception
.error_code
= 0;
394 } else if (vcpu
->arch
.exception
.nr
== DF_VECTOR
) {
395 /* triple fault -> shutdown */
396 set_bit(KVM_REQ_TRIPLE_FAULT
, &vcpu
->requests
);
400 vcpu
->arch
.cr2
= addr
;
401 kvm_queue_exception_e(vcpu
, PF_VECTOR
, error_code
);
404 void kvm_inject_nmi(struct kvm_vcpu
*vcpu
)
406 vcpu
->arch
.nmi_pending
= 1;
408 EXPORT_SYMBOL_GPL(kvm_inject_nmi
);
410 void kvm_queue_exception_e(struct kvm_vcpu
*vcpu
, unsigned nr
, u32 error_code
)
412 WARN_ON(vcpu
->arch
.exception
.pending
);
413 vcpu
->arch
.exception
.pending
= true;
414 vcpu
->arch
.exception
.has_error_code
= true;
415 vcpu
->arch
.exception
.nr
= nr
;
416 vcpu
->arch
.exception
.error_code
= error_code
;
418 EXPORT_SYMBOL_GPL(kvm_queue_exception_e
);
420 static void __queue_exception(struct kvm_vcpu
*vcpu
)
422 kvm_x86_ops
->queue_exception(vcpu
, vcpu
->arch
.exception
.nr
,
423 vcpu
->arch
.exception
.has_error_code
,
424 vcpu
->arch
.exception
.error_code
);
428 * Load the pae pdptrs. Return true is they are all valid.
430 int load_pdptrs(struct kvm_vcpu
*vcpu
, unsigned long cr3
)
432 gfn_t pdpt_gfn
= cr3
>> PAGE_SHIFT
;
433 unsigned offset
= ((cr3
& (PAGE_SIZE
-1)) >> 5) << 2;
436 u64 pdpte
[ARRAY_SIZE(vcpu
->arch
.pdptrs
)];
438 ret
= kvm_read_guest_page(vcpu
->kvm
, pdpt_gfn
, pdpte
,
439 offset
* sizeof(u64
), sizeof(pdpte
));
444 for (i
= 0; i
< ARRAY_SIZE(pdpte
); ++i
) {
445 if ((pdpte
[i
] & 1) && (pdpte
[i
] & 0xfffffff0000001e6ull
)) {
452 memcpy(vcpu
->arch
.pdptrs
, pdpte
, sizeof(vcpu
->arch
.pdptrs
));
457 EXPORT_SYMBOL_GPL(load_pdptrs
);
459 static bool pdptrs_changed(struct kvm_vcpu
*vcpu
)
461 u64 pdpte
[ARRAY_SIZE(vcpu
->arch
.pdptrs
)];
465 if (is_long_mode(vcpu
) || !is_pae(vcpu
))
468 r
= kvm_read_guest(vcpu
->kvm
, vcpu
->arch
.cr3
& ~31u, pdpte
, sizeof(pdpte
));
471 changed
= memcmp(pdpte
, vcpu
->arch
.pdptrs
, sizeof(pdpte
)) != 0;
477 void kvm_set_cr0(struct kvm_vcpu
*vcpu
, unsigned long cr0
)
479 if (cr0
& CR0_RESERVED_BITS
) {
480 printk(KERN_DEBUG
"set_cr0: 0x%lx #GP, reserved bits 0x%lx\n",
481 cr0
, vcpu
->arch
.cr0
);
482 kvm_inject_gp(vcpu
, 0);
486 if ((cr0
& X86_CR0_NW
) && !(cr0
& X86_CR0_CD
)) {
487 printk(KERN_DEBUG
"set_cr0: #GP, CD == 0 && NW == 1\n");
488 kvm_inject_gp(vcpu
, 0);
492 if ((cr0
& X86_CR0_PG
) && !(cr0
& X86_CR0_PE
)) {
493 printk(KERN_DEBUG
"set_cr0: #GP, set PG flag "
494 "and a clear PE flag\n");
495 kvm_inject_gp(vcpu
, 0);
499 if (!is_paging(vcpu
) && (cr0
& X86_CR0_PG
)) {
501 if ((vcpu
->arch
.shadow_efer
& EFER_LME
)) {
505 printk(KERN_DEBUG
"set_cr0: #GP, start paging "
506 "in long mode while PAE is disabled\n");
507 kvm_inject_gp(vcpu
, 0);
510 kvm_x86_ops
->get_cs_db_l_bits(vcpu
, &cs_db
, &cs_l
);
512 printk(KERN_DEBUG
"set_cr0: #GP, start paging "
513 "in long mode while CS.L == 1\n");
514 kvm_inject_gp(vcpu
, 0);
520 if (is_pae(vcpu
) && !load_pdptrs(vcpu
, vcpu
->arch
.cr3
)) {
521 printk(KERN_DEBUG
"set_cr0: #GP, pdptrs "
523 kvm_inject_gp(vcpu
, 0);
529 kvm_x86_ops
->set_cr0(vcpu
, cr0
);
530 vcpu
->arch
.cr0
= cr0
;
532 kvm_mmu_reset_context(vcpu
);
535 EXPORT_SYMBOL_GPL(kvm_set_cr0
);
537 void kvm_lmsw(struct kvm_vcpu
*vcpu
, unsigned long msw
)
539 kvm_set_cr0(vcpu
, (vcpu
->arch
.cr0
& ~0x0ful
) | (msw
& 0x0f));
540 KVMTRACE_1D(LMSW
, vcpu
,
541 (u32
)((vcpu
->arch
.cr0
& ~0x0ful
) | (msw
& 0x0f)),
544 EXPORT_SYMBOL_GPL(kvm_lmsw
);
546 void kvm_set_cr4(struct kvm_vcpu
*vcpu
, unsigned long cr4
)
548 if (cr4
& CR4_RESERVED_BITS
) {
549 printk(KERN_DEBUG
"set_cr4: #GP, reserved bits\n");
550 kvm_inject_gp(vcpu
, 0);
554 if (is_long_mode(vcpu
)) {
555 if (!(cr4
& X86_CR4_PAE
)) {
556 printk(KERN_DEBUG
"set_cr4: #GP, clearing PAE while "
558 kvm_inject_gp(vcpu
, 0);
561 } else if (is_paging(vcpu
) && !is_pae(vcpu
) && (cr4
& X86_CR4_PAE
)
562 && !load_pdptrs(vcpu
, vcpu
->arch
.cr3
)) {
563 printk(KERN_DEBUG
"set_cr4: #GP, pdptrs reserved bits\n");
564 kvm_inject_gp(vcpu
, 0);
568 if (cr4
& X86_CR4_VMXE
) {
569 printk(KERN_DEBUG
"set_cr4: #GP, setting VMXE\n");
570 kvm_inject_gp(vcpu
, 0);
573 kvm_x86_ops
->set_cr4(vcpu
, cr4
);
574 vcpu
->arch
.cr4
= cr4
;
575 kvm_mmu_reset_context(vcpu
);
577 EXPORT_SYMBOL_GPL(kvm_set_cr4
);
579 void kvm_set_cr3(struct kvm_vcpu
*vcpu
, unsigned long cr3
)
581 if (cr3
== vcpu
->arch
.cr3
&& !pdptrs_changed(vcpu
)) {
582 kvm_mmu_flush_tlb(vcpu
);
586 if (is_long_mode(vcpu
)) {
587 if (cr3
& CR3_L_MODE_RESERVED_BITS
) {
588 printk(KERN_DEBUG
"set_cr3: #GP, reserved bits\n");
589 kvm_inject_gp(vcpu
, 0);
594 if (cr3
& CR3_PAE_RESERVED_BITS
) {
596 "set_cr3: #GP, reserved bits\n");
597 kvm_inject_gp(vcpu
, 0);
600 if (is_paging(vcpu
) && !load_pdptrs(vcpu
, cr3
)) {
601 printk(KERN_DEBUG
"set_cr3: #GP, pdptrs "
603 kvm_inject_gp(vcpu
, 0);
608 * We don't check reserved bits in nonpae mode, because
609 * this isn't enforced, and VMware depends on this.
614 * Does the new cr3 value map to physical memory? (Note, we
615 * catch an invalid cr3 even in real-mode, because it would
616 * cause trouble later on when we turn on paging anyway.)
618 * A real CPU would silently accept an invalid cr3 and would
619 * attempt to use it - with largely undefined (and often hard
620 * to debug) behavior on the guest side.
622 if (unlikely(!gfn_to_memslot(vcpu
->kvm
, cr3
>> PAGE_SHIFT
)))
623 kvm_inject_gp(vcpu
, 0);
625 vcpu
->arch
.cr3
= cr3
;
626 vcpu
->arch
.mmu
.new_cr3(vcpu
);
629 EXPORT_SYMBOL_GPL(kvm_set_cr3
);
631 void kvm_set_cr8(struct kvm_vcpu
*vcpu
, unsigned long cr8
)
633 if (cr8
& CR8_RESERVED_BITS
) {
634 printk(KERN_DEBUG
"set_cr8: #GP, reserved bits 0x%lx\n", cr8
);
635 kvm_inject_gp(vcpu
, 0);
638 if (irqchip_in_kernel(vcpu
->kvm
))
639 kvm_lapic_set_tpr(vcpu
, cr8
);
641 vcpu
->arch
.cr8
= cr8
;
643 EXPORT_SYMBOL_GPL(kvm_set_cr8
);
645 unsigned long kvm_get_cr8(struct kvm_vcpu
*vcpu
)
647 if (irqchip_in_kernel(vcpu
->kvm
))
648 return kvm_lapic_get_cr8(vcpu
);
650 return vcpu
->arch
.cr8
;
652 EXPORT_SYMBOL_GPL(kvm_get_cr8
);
655 * List of msr numbers which we expose to userspace through KVM_GET_MSRS
656 * and KVM_SET_MSRS, and KVM_GET_MSR_INDEX_LIST.
658 * This list is modified at module load time to reflect the
659 * capabilities of the host cpu.
661 static u32 msrs_to_save
[] = {
662 MSR_IA32_SYSENTER_CS
, MSR_IA32_SYSENTER_ESP
, MSR_IA32_SYSENTER_EIP
,
665 MSR_CSTAR
, MSR_KERNEL_GS_BASE
, MSR_SYSCALL_MASK
, MSR_LSTAR
,
667 MSR_IA32_TIME_STAMP_COUNTER
, MSR_KVM_SYSTEM_TIME
, MSR_KVM_WALL_CLOCK
,
668 MSR_IA32_PERF_STATUS
,
671 static unsigned num_msrs_to_save
;
673 static u32 emulated_msrs
[] = {
674 MSR_IA32_MISC_ENABLE
,
677 static void set_efer(struct kvm_vcpu
*vcpu
, u64 efer
)
679 if (efer
& efer_reserved_bits
) {
680 printk(KERN_DEBUG
"set_efer: 0x%llx #GP, reserved bits\n",
682 kvm_inject_gp(vcpu
, 0);
687 && (vcpu
->arch
.shadow_efer
& EFER_LME
) != (efer
& EFER_LME
)) {
688 printk(KERN_DEBUG
"set_efer: #GP, change LME while paging\n");
689 kvm_inject_gp(vcpu
, 0);
693 kvm_x86_ops
->set_efer(vcpu
, efer
);
696 efer
|= vcpu
->arch
.shadow_efer
& EFER_LMA
;
698 vcpu
->arch
.shadow_efer
= efer
;
701 void kvm_enable_efer_bits(u64 mask
)
703 efer_reserved_bits
&= ~mask
;
705 EXPORT_SYMBOL_GPL(kvm_enable_efer_bits
);
709 * Writes msr value into into the appropriate "register".
710 * Returns 0 on success, non-0 otherwise.
711 * Assumes vcpu_load() was already called.
713 int kvm_set_msr(struct kvm_vcpu
*vcpu
, u32 msr_index
, u64 data
)
715 return kvm_x86_ops
->set_msr(vcpu
, msr_index
, data
);
719 * Adapt set_msr() to msr_io()'s calling convention
721 static int do_set_msr(struct kvm_vcpu
*vcpu
, unsigned index
, u64
*data
)
723 return kvm_set_msr(vcpu
, index
, *data
);
726 static void kvm_write_wall_clock(struct kvm
*kvm
, gpa_t wall_clock
)
729 struct pvclock_wall_clock wc
;
730 struct timespec now
, sys
, boot
;
737 kvm_write_guest(kvm
, wall_clock
, &version
, sizeof(version
));
740 * The guest calculates current wall clock time by adding
741 * system time (updated by kvm_write_guest_time below) to the
742 * wall clock specified here. guest system time equals host
743 * system time for us, thus we must fill in host boot time here.
745 now
= current_kernel_time();
747 boot
= ns_to_timespec(timespec_to_ns(&now
) - timespec_to_ns(&sys
));
749 wc
.sec
= boot
.tv_sec
;
750 wc
.nsec
= boot
.tv_nsec
;
751 wc
.version
= version
;
753 kvm_write_guest(kvm
, wall_clock
, &wc
, sizeof(wc
));
756 kvm_write_guest(kvm
, wall_clock
, &version
, sizeof(version
));
759 static uint32_t div_frac(uint32_t dividend
, uint32_t divisor
)
761 uint32_t quotient
, remainder
;
763 /* Don't try to replace with do_div(), this one calculates
764 * "(dividend << 32) / divisor" */
766 : "=a" (quotient
), "=d" (remainder
)
767 : "0" (0), "1" (dividend
), "r" (divisor
) );
771 static void kvm_set_time_scale(uint32_t tsc_khz
, struct pvclock_vcpu_time_info
*hv_clock
)
773 uint64_t nsecs
= 1000000000LL;
778 tps64
= tsc_khz
* 1000LL;
779 while (tps64
> nsecs
*2) {
784 tps32
= (uint32_t)tps64
;
785 while (tps32
<= (uint32_t)nsecs
) {
790 hv_clock
->tsc_shift
= shift
;
791 hv_clock
->tsc_to_system_mul
= div_frac(nsecs
, tps32
);
793 pr_debug("%s: tsc_khz %u, tsc_shift %d, tsc_mul %u\n",
794 __FUNCTION__
, tsc_khz
, hv_clock
->tsc_shift
,
795 hv_clock
->tsc_to_system_mul
);
798 static void kvm_write_guest_time(struct kvm_vcpu
*v
)
802 struct kvm_vcpu_arch
*vcpu
= &v
->arch
;
805 if ((!vcpu
->time_page
))
808 if (unlikely(vcpu
->hv_clock_tsc_khz
!= tsc_khz
)) {
809 kvm_set_time_scale(tsc_khz
, &vcpu
->hv_clock
);
810 vcpu
->hv_clock_tsc_khz
= tsc_khz
;
813 /* Keep irq disabled to prevent changes to the clock */
814 local_irq_save(flags
);
815 kvm_get_msr(v
, MSR_IA32_TIME_STAMP_COUNTER
,
816 &vcpu
->hv_clock
.tsc_timestamp
);
818 local_irq_restore(flags
);
820 /* With all the info we got, fill in the values */
822 vcpu
->hv_clock
.system_time
= ts
.tv_nsec
+
823 (NSEC_PER_SEC
* (u64
)ts
.tv_sec
);
825 * The interface expects us to write an even number signaling that the
826 * update is finished. Since the guest won't see the intermediate
827 * state, we just increase by 2 at the end.
829 vcpu
->hv_clock
.version
+= 2;
831 shared_kaddr
= kmap_atomic(vcpu
->time_page
, KM_USER0
);
833 memcpy(shared_kaddr
+ vcpu
->time_offset
, &vcpu
->hv_clock
,
834 sizeof(vcpu
->hv_clock
));
836 kunmap_atomic(shared_kaddr
, KM_USER0
);
838 mark_page_dirty(v
->kvm
, vcpu
->time
>> PAGE_SHIFT
);
841 static bool msr_mtrr_valid(unsigned msr
)
844 case 0x200 ... 0x200 + 2 * KVM_NR_VAR_MTRR
- 1:
845 case MSR_MTRRfix64K_00000
:
846 case MSR_MTRRfix16K_80000
:
847 case MSR_MTRRfix16K_A0000
:
848 case MSR_MTRRfix4K_C0000
:
849 case MSR_MTRRfix4K_C8000
:
850 case MSR_MTRRfix4K_D0000
:
851 case MSR_MTRRfix4K_D8000
:
852 case MSR_MTRRfix4K_E0000
:
853 case MSR_MTRRfix4K_E8000
:
854 case MSR_MTRRfix4K_F0000
:
855 case MSR_MTRRfix4K_F8000
:
856 case MSR_MTRRdefType
:
857 case MSR_IA32_CR_PAT
:
865 static int set_msr_mtrr(struct kvm_vcpu
*vcpu
, u32 msr
, u64 data
)
867 if (!msr_mtrr_valid(msr
))
870 vcpu
->arch
.mtrr
[msr
- 0x200] = data
;
874 int kvm_set_msr_common(struct kvm_vcpu
*vcpu
, u32 msr
, u64 data
)
878 set_efer(vcpu
, data
);
880 case MSR_IA32_MC0_STATUS
:
881 pr_unimpl(vcpu
, "%s: MSR_IA32_MC0_STATUS 0x%llx, nop\n",
884 case MSR_IA32_MCG_STATUS
:
885 pr_unimpl(vcpu
, "%s: MSR_IA32_MCG_STATUS 0x%llx, nop\n",
888 case MSR_IA32_MCG_CTL
:
889 pr_unimpl(vcpu
, "%s: MSR_IA32_MCG_CTL 0x%llx, nop\n",
892 case MSR_IA32_DEBUGCTLMSR
:
894 /* We support the non-activated case already */
896 } else if (data
& ~(DEBUGCTLMSR_LBR
| DEBUGCTLMSR_BTF
)) {
897 /* Values other than LBR and BTF are vendor-specific,
898 thus reserved and should throw a #GP */
901 pr_unimpl(vcpu
, "%s: MSR_IA32_DEBUGCTLMSR 0x%llx, nop\n",
904 case MSR_IA32_UCODE_REV
:
905 case MSR_IA32_UCODE_WRITE
:
907 case 0x200 ... 0x2ff:
908 return set_msr_mtrr(vcpu
, msr
, data
);
909 case MSR_IA32_APICBASE
:
910 kvm_set_apic_base(vcpu
, data
);
912 case MSR_IA32_MISC_ENABLE
:
913 vcpu
->arch
.ia32_misc_enable_msr
= data
;
915 case MSR_KVM_WALL_CLOCK
:
916 vcpu
->kvm
->arch
.wall_clock
= data
;
917 kvm_write_wall_clock(vcpu
->kvm
, data
);
919 case MSR_KVM_SYSTEM_TIME
: {
920 if (vcpu
->arch
.time_page
) {
921 kvm_release_page_dirty(vcpu
->arch
.time_page
);
922 vcpu
->arch
.time_page
= NULL
;
925 vcpu
->arch
.time
= data
;
927 /* we verify if the enable bit is set... */
931 /* ...but clean it before doing the actual write */
932 vcpu
->arch
.time_offset
= data
& ~(PAGE_MASK
| 1);
934 down_read(¤t
->mm
->mmap_sem
);
935 vcpu
->arch
.time_page
=
936 gfn_to_page(vcpu
->kvm
, data
>> PAGE_SHIFT
);
937 up_read(¤t
->mm
->mmap_sem
);
939 if (is_error_page(vcpu
->arch
.time_page
)) {
940 kvm_release_page_clean(vcpu
->arch
.time_page
);
941 vcpu
->arch
.time_page
= NULL
;
944 kvm_write_guest_time(vcpu
);
948 pr_unimpl(vcpu
, "unhandled wrmsr: 0x%x data %llx\n", msr
, data
);
953 EXPORT_SYMBOL_GPL(kvm_set_msr_common
);
957 * Reads an msr value (of 'msr_index') into 'pdata'.
958 * Returns 0 on success, non-0 otherwise.
959 * Assumes vcpu_load() was already called.
961 int kvm_get_msr(struct kvm_vcpu
*vcpu
, u32 msr_index
, u64
*pdata
)
963 return kvm_x86_ops
->get_msr(vcpu
, msr_index
, pdata
);
966 static int get_msr_mtrr(struct kvm_vcpu
*vcpu
, u32 msr
, u64
*pdata
)
968 if (!msr_mtrr_valid(msr
))
971 *pdata
= vcpu
->arch
.mtrr
[msr
- 0x200];
975 int kvm_get_msr_common(struct kvm_vcpu
*vcpu
, u32 msr
, u64
*pdata
)
980 case 0xc0010010: /* SYSCFG */
981 case 0xc0010015: /* HWCR */
982 case MSR_IA32_PLATFORM_ID
:
983 case MSR_IA32_P5_MC_ADDR
:
984 case MSR_IA32_P5_MC_TYPE
:
985 case MSR_IA32_MC0_CTL
:
986 case MSR_IA32_MCG_STATUS
:
987 case MSR_IA32_MCG_CAP
:
988 case MSR_IA32_MCG_CTL
:
989 case MSR_IA32_MC0_MISC
:
990 case MSR_IA32_MC0_MISC
+4:
991 case MSR_IA32_MC0_MISC
+8:
992 case MSR_IA32_MC0_MISC
+12:
993 case MSR_IA32_MC0_MISC
+16:
994 case MSR_IA32_UCODE_REV
:
995 case MSR_IA32_EBL_CR_POWERON
:
996 case MSR_IA32_DEBUGCTLMSR
:
997 case MSR_IA32_LASTBRANCHFROMIP
:
998 case MSR_IA32_LASTBRANCHTOIP
:
999 case MSR_IA32_LASTINTFROMIP
:
1000 case MSR_IA32_LASTINTTOIP
:
1004 data
= 0x500 | KVM_NR_VAR_MTRR
;
1006 case 0x200 ... 0x2ff:
1007 return get_msr_mtrr(vcpu
, msr
, pdata
);
1008 case 0xcd: /* fsb frequency */
1011 case MSR_IA32_APICBASE
:
1012 data
= kvm_get_apic_base(vcpu
);
1014 case MSR_IA32_MISC_ENABLE
:
1015 data
= vcpu
->arch
.ia32_misc_enable_msr
;
1017 case MSR_IA32_PERF_STATUS
:
1018 /* TSC increment by tick */
1020 /* CPU multiplier */
1021 data
|= (((uint64_t)4ULL) << 40);
1024 data
= vcpu
->arch
.shadow_efer
;
1026 case MSR_KVM_WALL_CLOCK
:
1027 data
= vcpu
->kvm
->arch
.wall_clock
;
1029 case MSR_KVM_SYSTEM_TIME
:
1030 data
= vcpu
->arch
.time
;
1033 pr_unimpl(vcpu
, "unhandled rdmsr: 0x%x\n", msr
);
1039 EXPORT_SYMBOL_GPL(kvm_get_msr_common
);
1042 * Read or write a bunch of msrs. All parameters are kernel addresses.
1044 * @return number of msrs set successfully.
1046 static int __msr_io(struct kvm_vcpu
*vcpu
, struct kvm_msrs
*msrs
,
1047 struct kvm_msr_entry
*entries
,
1048 int (*do_msr
)(struct kvm_vcpu
*vcpu
,
1049 unsigned index
, u64
*data
))
1055 down_read(&vcpu
->kvm
->slots_lock
);
1056 for (i
= 0; i
< msrs
->nmsrs
; ++i
)
1057 if (do_msr(vcpu
, entries
[i
].index
, &entries
[i
].data
))
1059 up_read(&vcpu
->kvm
->slots_lock
);
1067 * Read or write a bunch of msrs. Parameters are user addresses.
1069 * @return number of msrs set successfully.
1071 static int msr_io(struct kvm_vcpu
*vcpu
, struct kvm_msrs __user
*user_msrs
,
1072 int (*do_msr
)(struct kvm_vcpu
*vcpu
,
1073 unsigned index
, u64
*data
),
1076 struct kvm_msrs msrs
;
1077 struct kvm_msr_entry
*entries
;
1082 if (copy_from_user(&msrs
, user_msrs
, sizeof msrs
))
1086 if (msrs
.nmsrs
>= MAX_IO_MSRS
)
1090 size
= sizeof(struct kvm_msr_entry
) * msrs
.nmsrs
;
1091 entries
= vmalloc(size
);
1096 if (copy_from_user(entries
, user_msrs
->entries
, size
))
1099 r
= n
= __msr_io(vcpu
, &msrs
, entries
, do_msr
);
1104 if (writeback
&& copy_to_user(user_msrs
->entries
, entries
, size
))
1115 int kvm_dev_ioctl_check_extension(long ext
)
1120 case KVM_CAP_IRQCHIP
:
1122 case KVM_CAP_MMU_SHADOW_CACHE_CONTROL
:
1123 case KVM_CAP_USER_MEMORY
:
1124 case KVM_CAP_SET_TSS_ADDR
:
1125 case KVM_CAP_EXT_CPUID
:
1126 case KVM_CAP_CLOCKSOURCE
:
1128 case KVM_CAP_NOP_IO_DELAY
:
1129 case KVM_CAP_MP_STATE
:
1130 case KVM_CAP_SYNC_MMU
:
1133 case KVM_CAP_COALESCED_MMIO
:
1134 r
= KVM_COALESCED_MMIO_PAGE_OFFSET
;
1137 r
= !kvm_x86_ops
->cpu_has_accelerated_tpr();
1139 case KVM_CAP_NR_VCPUS
:
1142 case KVM_CAP_NR_MEMSLOTS
:
1143 r
= KVM_MEMORY_SLOTS
;
1145 case KVM_CAP_PV_MMU
:
1156 long kvm_arch_dev_ioctl(struct file
*filp
,
1157 unsigned int ioctl
, unsigned long arg
)
1159 void __user
*argp
= (void __user
*)arg
;
1163 case KVM_GET_MSR_INDEX_LIST
: {
1164 struct kvm_msr_list __user
*user_msr_list
= argp
;
1165 struct kvm_msr_list msr_list
;
1169 if (copy_from_user(&msr_list
, user_msr_list
, sizeof msr_list
))
1172 msr_list
.nmsrs
= num_msrs_to_save
+ ARRAY_SIZE(emulated_msrs
);
1173 if (copy_to_user(user_msr_list
, &msr_list
, sizeof msr_list
))
1176 if (n
< num_msrs_to_save
)
1179 if (copy_to_user(user_msr_list
->indices
, &msrs_to_save
,
1180 num_msrs_to_save
* sizeof(u32
)))
1182 if (copy_to_user(user_msr_list
->indices
1183 + num_msrs_to_save
* sizeof(u32
),
1185 ARRAY_SIZE(emulated_msrs
) * sizeof(u32
)))
1190 case KVM_GET_SUPPORTED_CPUID
: {
1191 struct kvm_cpuid2 __user
*cpuid_arg
= argp
;
1192 struct kvm_cpuid2 cpuid
;
1195 if (copy_from_user(&cpuid
, cpuid_arg
, sizeof cpuid
))
1197 r
= kvm_dev_ioctl_get_supported_cpuid(&cpuid
,
1198 cpuid_arg
->entries
);
1203 if (copy_to_user(cpuid_arg
, &cpuid
, sizeof cpuid
))
1215 void kvm_arch_vcpu_load(struct kvm_vcpu
*vcpu
, int cpu
)
1217 kvm_x86_ops
->vcpu_load(vcpu
, cpu
);
1218 kvm_write_guest_time(vcpu
);
1221 void kvm_arch_vcpu_put(struct kvm_vcpu
*vcpu
)
1223 kvm_x86_ops
->vcpu_put(vcpu
);
1224 kvm_put_guest_fpu(vcpu
);
1227 static int is_efer_nx(void)
1231 rdmsrl(MSR_EFER
, efer
);
1232 return efer
& EFER_NX
;
1235 static void cpuid_fix_nx_cap(struct kvm_vcpu
*vcpu
)
1238 struct kvm_cpuid_entry2
*e
, *entry
;
1241 for (i
= 0; i
< vcpu
->arch
.cpuid_nent
; ++i
) {
1242 e
= &vcpu
->arch
.cpuid_entries
[i
];
1243 if (e
->function
== 0x80000001) {
1248 if (entry
&& (entry
->edx
& (1 << 20)) && !is_efer_nx()) {
1249 entry
->edx
&= ~(1 << 20);
1250 printk(KERN_INFO
"kvm: guest NX capability removed\n");
1254 /* when an old userspace process fills a new kernel module */
1255 static int kvm_vcpu_ioctl_set_cpuid(struct kvm_vcpu
*vcpu
,
1256 struct kvm_cpuid
*cpuid
,
1257 struct kvm_cpuid_entry __user
*entries
)
1260 struct kvm_cpuid_entry
*cpuid_entries
;
1263 if (cpuid
->nent
> KVM_MAX_CPUID_ENTRIES
)
1266 cpuid_entries
= vmalloc(sizeof(struct kvm_cpuid_entry
) * cpuid
->nent
);
1270 if (copy_from_user(cpuid_entries
, entries
,
1271 cpuid
->nent
* sizeof(struct kvm_cpuid_entry
)))
1273 for (i
= 0; i
< cpuid
->nent
; i
++) {
1274 vcpu
->arch
.cpuid_entries
[i
].function
= cpuid_entries
[i
].function
;
1275 vcpu
->arch
.cpuid_entries
[i
].eax
= cpuid_entries
[i
].eax
;
1276 vcpu
->arch
.cpuid_entries
[i
].ebx
= cpuid_entries
[i
].ebx
;
1277 vcpu
->arch
.cpuid_entries
[i
].ecx
= cpuid_entries
[i
].ecx
;
1278 vcpu
->arch
.cpuid_entries
[i
].edx
= cpuid_entries
[i
].edx
;
1279 vcpu
->arch
.cpuid_entries
[i
].index
= 0;
1280 vcpu
->arch
.cpuid_entries
[i
].flags
= 0;
1281 vcpu
->arch
.cpuid_entries
[i
].padding
[0] = 0;
1282 vcpu
->arch
.cpuid_entries
[i
].padding
[1] = 0;
1283 vcpu
->arch
.cpuid_entries
[i
].padding
[2] = 0;
1285 vcpu
->arch
.cpuid_nent
= cpuid
->nent
;
1286 cpuid_fix_nx_cap(vcpu
);
1290 vfree(cpuid_entries
);
1295 static int kvm_vcpu_ioctl_set_cpuid2(struct kvm_vcpu
*vcpu
,
1296 struct kvm_cpuid2
*cpuid
,
1297 struct kvm_cpuid_entry2 __user
*entries
)
1302 if (cpuid
->nent
> KVM_MAX_CPUID_ENTRIES
)
1305 if (copy_from_user(&vcpu
->arch
.cpuid_entries
, entries
,
1306 cpuid
->nent
* sizeof(struct kvm_cpuid_entry2
)))
1308 vcpu
->arch
.cpuid_nent
= cpuid
->nent
;
1315 static int kvm_vcpu_ioctl_get_cpuid2(struct kvm_vcpu
*vcpu
,
1316 struct kvm_cpuid2
*cpuid
,
1317 struct kvm_cpuid_entry2 __user
*entries
)
1322 if (cpuid
->nent
< vcpu
->arch
.cpuid_nent
)
1325 if (copy_to_user(entries
, &vcpu
->arch
.cpuid_entries
,
1326 vcpu
->arch
.cpuid_nent
* sizeof(struct kvm_cpuid_entry2
)))
1331 cpuid
->nent
= vcpu
->arch
.cpuid_nent
;
1335 static inline u32
bit(int bitno
)
1337 return 1 << (bitno
& 31);
1340 static void do_cpuid_1_ent(struct kvm_cpuid_entry2
*entry
, u32 function
,
1343 entry
->function
= function
;
1344 entry
->index
= index
;
1345 cpuid_count(entry
->function
, entry
->index
,
1346 &entry
->eax
, &entry
->ebx
, &entry
->ecx
, &entry
->edx
);
1350 static void do_cpuid_ent(struct kvm_cpuid_entry2
*entry
, u32 function
,
1351 u32 index
, int *nent
, int maxnent
)
1353 const u32 kvm_supported_word0_x86_features
= bit(X86_FEATURE_FPU
) |
1354 bit(X86_FEATURE_VME
) | bit(X86_FEATURE_DE
) |
1355 bit(X86_FEATURE_PSE
) | bit(X86_FEATURE_TSC
) |
1356 bit(X86_FEATURE_MSR
) | bit(X86_FEATURE_PAE
) |
1357 bit(X86_FEATURE_CX8
) | bit(X86_FEATURE_APIC
) |
1358 bit(X86_FEATURE_SEP
) | bit(X86_FEATURE_PGE
) |
1359 bit(X86_FEATURE_CMOV
) | bit(X86_FEATURE_PSE36
) |
1360 bit(X86_FEATURE_CLFLSH
) | bit(X86_FEATURE_MMX
) |
1361 bit(X86_FEATURE_FXSR
) | bit(X86_FEATURE_XMM
) |
1362 bit(X86_FEATURE_XMM2
) | bit(X86_FEATURE_SELFSNOOP
);
1363 const u32 kvm_supported_word1_x86_features
= bit(X86_FEATURE_FPU
) |
1364 bit(X86_FEATURE_VME
) | bit(X86_FEATURE_DE
) |
1365 bit(X86_FEATURE_PSE
) | bit(X86_FEATURE_TSC
) |
1366 bit(X86_FEATURE_MSR
) | bit(X86_FEATURE_PAE
) |
1367 bit(X86_FEATURE_CX8
) | bit(X86_FEATURE_APIC
) |
1368 bit(X86_FEATURE_PGE
) |
1369 bit(X86_FEATURE_CMOV
) | bit(X86_FEATURE_PSE36
) |
1370 bit(X86_FEATURE_MMX
) | bit(X86_FEATURE_FXSR
) |
1371 bit(X86_FEATURE_SYSCALL
) |
1372 (bit(X86_FEATURE_NX
) && is_efer_nx()) |
1373 #ifdef CONFIG_X86_64
1374 bit(X86_FEATURE_LM
) |
1376 bit(X86_FEATURE_MMXEXT
) |
1377 bit(X86_FEATURE_3DNOWEXT
) |
1378 bit(X86_FEATURE_3DNOW
);
1379 const u32 kvm_supported_word3_x86_features
=
1380 bit(X86_FEATURE_XMM3
) | bit(X86_FEATURE_CX16
);
1381 const u32 kvm_supported_word6_x86_features
=
1382 bit(X86_FEATURE_LAHF_LM
) | bit(X86_FEATURE_CMP_LEGACY
);
1384 /* all func 2 cpuid_count() should be called on the same cpu */
1386 do_cpuid_1_ent(entry
, function
, index
);
1391 entry
->eax
= min(entry
->eax
, (u32
)0xb);
1394 entry
->edx
&= kvm_supported_word0_x86_features
;
1395 entry
->ecx
&= kvm_supported_word3_x86_features
;
1397 /* function 2 entries are STATEFUL. That is, repeated cpuid commands
1398 * may return different values. This forces us to get_cpu() before
1399 * issuing the first command, and also to emulate this annoying behavior
1400 * in kvm_emulate_cpuid() using KVM_CPUID_FLAG_STATE_READ_NEXT */
1402 int t
, times
= entry
->eax
& 0xff;
1404 entry
->flags
|= KVM_CPUID_FLAG_STATEFUL_FUNC
;
1405 for (t
= 1; t
< times
&& *nent
< maxnent
; ++t
) {
1406 do_cpuid_1_ent(&entry
[t
], function
, 0);
1407 entry
[t
].flags
|= KVM_CPUID_FLAG_STATEFUL_FUNC
;
1412 /* function 4 and 0xb have additional index. */
1416 entry
->flags
|= KVM_CPUID_FLAG_SIGNIFCANT_INDEX
;
1417 /* read more entries until cache_type is zero */
1418 for (i
= 1; *nent
< maxnent
; ++i
) {
1419 cache_type
= entry
[i
- 1].eax
& 0x1f;
1422 do_cpuid_1_ent(&entry
[i
], function
, i
);
1424 KVM_CPUID_FLAG_SIGNIFCANT_INDEX
;
1432 entry
->flags
|= KVM_CPUID_FLAG_SIGNIFCANT_INDEX
;
1433 /* read more entries until level_type is zero */
1434 for (i
= 1; *nent
< maxnent
; ++i
) {
1435 level_type
= entry
[i
- 1].ecx
& 0xff;
1438 do_cpuid_1_ent(&entry
[i
], function
, i
);
1440 KVM_CPUID_FLAG_SIGNIFCANT_INDEX
;
1446 entry
->eax
= min(entry
->eax
, 0x8000001a);
1449 entry
->edx
&= kvm_supported_word1_x86_features
;
1450 entry
->ecx
&= kvm_supported_word6_x86_features
;
1456 static int kvm_dev_ioctl_get_supported_cpuid(struct kvm_cpuid2
*cpuid
,
1457 struct kvm_cpuid_entry2 __user
*entries
)
1459 struct kvm_cpuid_entry2
*cpuid_entries
;
1460 int limit
, nent
= 0, r
= -E2BIG
;
1463 if (cpuid
->nent
< 1)
1466 cpuid_entries
= vmalloc(sizeof(struct kvm_cpuid_entry2
) * cpuid
->nent
);
1470 do_cpuid_ent(&cpuid_entries
[0], 0, 0, &nent
, cpuid
->nent
);
1471 limit
= cpuid_entries
[0].eax
;
1472 for (func
= 1; func
<= limit
&& nent
< cpuid
->nent
; ++func
)
1473 do_cpuid_ent(&cpuid_entries
[nent
], func
, 0,
1474 &nent
, cpuid
->nent
);
1476 if (nent
>= cpuid
->nent
)
1479 do_cpuid_ent(&cpuid_entries
[nent
], 0x80000000, 0, &nent
, cpuid
->nent
);
1480 limit
= cpuid_entries
[nent
- 1].eax
;
1481 for (func
= 0x80000001; func
<= limit
&& nent
< cpuid
->nent
; ++func
)
1482 do_cpuid_ent(&cpuid_entries
[nent
], func
, 0,
1483 &nent
, cpuid
->nent
);
1485 if (copy_to_user(entries
, cpuid_entries
,
1486 nent
* sizeof(struct kvm_cpuid_entry2
)))
1492 vfree(cpuid_entries
);
1497 static int kvm_vcpu_ioctl_get_lapic(struct kvm_vcpu
*vcpu
,
1498 struct kvm_lapic_state
*s
)
1501 memcpy(s
->regs
, vcpu
->arch
.apic
->regs
, sizeof *s
);
1507 static int kvm_vcpu_ioctl_set_lapic(struct kvm_vcpu
*vcpu
,
1508 struct kvm_lapic_state
*s
)
1511 memcpy(vcpu
->arch
.apic
->regs
, s
->regs
, sizeof *s
);
1512 kvm_apic_post_state_restore(vcpu
);
1518 static int kvm_vcpu_ioctl_interrupt(struct kvm_vcpu
*vcpu
,
1519 struct kvm_interrupt
*irq
)
1521 if (irq
->irq
< 0 || irq
->irq
>= 256)
1523 if (irqchip_in_kernel(vcpu
->kvm
))
1527 set_bit(irq
->irq
, vcpu
->arch
.irq_pending
);
1528 set_bit(irq
->irq
/ BITS_PER_LONG
, &vcpu
->arch
.irq_summary
);
1535 static int vcpu_ioctl_tpr_access_reporting(struct kvm_vcpu
*vcpu
,
1536 struct kvm_tpr_access_ctl
*tac
)
1540 vcpu
->arch
.tpr_access_reporting
= !!tac
->enabled
;
1544 long kvm_arch_vcpu_ioctl(struct file
*filp
,
1545 unsigned int ioctl
, unsigned long arg
)
1547 struct kvm_vcpu
*vcpu
= filp
->private_data
;
1548 void __user
*argp
= (void __user
*)arg
;
1550 struct kvm_lapic_state
*lapic
= NULL
;
1553 case KVM_GET_LAPIC
: {
1554 lapic
= kzalloc(sizeof(struct kvm_lapic_state
), GFP_KERNEL
);
1559 r
= kvm_vcpu_ioctl_get_lapic(vcpu
, lapic
);
1563 if (copy_to_user(argp
, lapic
, sizeof(struct kvm_lapic_state
)))
1568 case KVM_SET_LAPIC
: {
1569 lapic
= kmalloc(sizeof(struct kvm_lapic_state
), GFP_KERNEL
);
1574 if (copy_from_user(lapic
, argp
, sizeof(struct kvm_lapic_state
)))
1576 r
= kvm_vcpu_ioctl_set_lapic(vcpu
, lapic
);
1582 case KVM_INTERRUPT
: {
1583 struct kvm_interrupt irq
;
1586 if (copy_from_user(&irq
, argp
, sizeof irq
))
1588 r
= kvm_vcpu_ioctl_interrupt(vcpu
, &irq
);
1594 case KVM_SET_CPUID
: {
1595 struct kvm_cpuid __user
*cpuid_arg
= argp
;
1596 struct kvm_cpuid cpuid
;
1599 if (copy_from_user(&cpuid
, cpuid_arg
, sizeof cpuid
))
1601 r
= kvm_vcpu_ioctl_set_cpuid(vcpu
, &cpuid
, cpuid_arg
->entries
);
1606 case KVM_SET_CPUID2
: {
1607 struct kvm_cpuid2 __user
*cpuid_arg
= argp
;
1608 struct kvm_cpuid2 cpuid
;
1611 if (copy_from_user(&cpuid
, cpuid_arg
, sizeof cpuid
))
1613 r
= kvm_vcpu_ioctl_set_cpuid2(vcpu
, &cpuid
,
1614 cpuid_arg
->entries
);
1619 case KVM_GET_CPUID2
: {
1620 struct kvm_cpuid2 __user
*cpuid_arg
= argp
;
1621 struct kvm_cpuid2 cpuid
;
1624 if (copy_from_user(&cpuid
, cpuid_arg
, sizeof cpuid
))
1626 r
= kvm_vcpu_ioctl_get_cpuid2(vcpu
, &cpuid
,
1627 cpuid_arg
->entries
);
1631 if (copy_to_user(cpuid_arg
, &cpuid
, sizeof cpuid
))
1637 r
= msr_io(vcpu
, argp
, kvm_get_msr
, 1);
1640 r
= msr_io(vcpu
, argp
, do_set_msr
, 0);
1642 case KVM_TPR_ACCESS_REPORTING
: {
1643 struct kvm_tpr_access_ctl tac
;
1646 if (copy_from_user(&tac
, argp
, sizeof tac
))
1648 r
= vcpu_ioctl_tpr_access_reporting(vcpu
, &tac
);
1652 if (copy_to_user(argp
, &tac
, sizeof tac
))
1657 case KVM_SET_VAPIC_ADDR
: {
1658 struct kvm_vapic_addr va
;
1661 if (!irqchip_in_kernel(vcpu
->kvm
))
1664 if (copy_from_user(&va
, argp
, sizeof va
))
1667 kvm_lapic_set_vapic_addr(vcpu
, va
.vapic_addr
);
1679 static int kvm_vm_ioctl_set_tss_addr(struct kvm
*kvm
, unsigned long addr
)
1683 if (addr
> (unsigned int)(-3 * PAGE_SIZE
))
1685 ret
= kvm_x86_ops
->set_tss_addr(kvm
, addr
);
1689 static int kvm_vm_ioctl_set_nr_mmu_pages(struct kvm
*kvm
,
1690 u32 kvm_nr_mmu_pages
)
1692 if (kvm_nr_mmu_pages
< KVM_MIN_ALLOC_MMU_PAGES
)
1695 down_write(&kvm
->slots_lock
);
1697 kvm_mmu_change_mmu_pages(kvm
, kvm_nr_mmu_pages
);
1698 kvm
->arch
.n_requested_mmu_pages
= kvm_nr_mmu_pages
;
1700 up_write(&kvm
->slots_lock
);
1704 static int kvm_vm_ioctl_get_nr_mmu_pages(struct kvm
*kvm
)
1706 return kvm
->arch
.n_alloc_mmu_pages
;
1709 gfn_t
unalias_gfn(struct kvm
*kvm
, gfn_t gfn
)
1712 struct kvm_mem_alias
*alias
;
1714 for (i
= 0; i
< kvm
->arch
.naliases
; ++i
) {
1715 alias
= &kvm
->arch
.aliases
[i
];
1716 if (gfn
>= alias
->base_gfn
1717 && gfn
< alias
->base_gfn
+ alias
->npages
)
1718 return alias
->target_gfn
+ gfn
- alias
->base_gfn
;
1724 * Set a new alias region. Aliases map a portion of physical memory into
1725 * another portion. This is useful for memory windows, for example the PC
1728 static int kvm_vm_ioctl_set_memory_alias(struct kvm
*kvm
,
1729 struct kvm_memory_alias
*alias
)
1732 struct kvm_mem_alias
*p
;
1735 /* General sanity checks */
1736 if (alias
->memory_size
& (PAGE_SIZE
- 1))
1738 if (alias
->guest_phys_addr
& (PAGE_SIZE
- 1))
1740 if (alias
->slot
>= KVM_ALIAS_SLOTS
)
1742 if (alias
->guest_phys_addr
+ alias
->memory_size
1743 < alias
->guest_phys_addr
)
1745 if (alias
->target_phys_addr
+ alias
->memory_size
1746 < alias
->target_phys_addr
)
1749 down_write(&kvm
->slots_lock
);
1750 spin_lock(&kvm
->mmu_lock
);
1752 p
= &kvm
->arch
.aliases
[alias
->slot
];
1753 p
->base_gfn
= alias
->guest_phys_addr
>> PAGE_SHIFT
;
1754 p
->npages
= alias
->memory_size
>> PAGE_SHIFT
;
1755 p
->target_gfn
= alias
->target_phys_addr
>> PAGE_SHIFT
;
1757 for (n
= KVM_ALIAS_SLOTS
; n
> 0; --n
)
1758 if (kvm
->arch
.aliases
[n
- 1].npages
)
1760 kvm
->arch
.naliases
= n
;
1762 spin_unlock(&kvm
->mmu_lock
);
1763 kvm_mmu_zap_all(kvm
);
1765 up_write(&kvm
->slots_lock
);
1773 static int kvm_vm_ioctl_get_irqchip(struct kvm
*kvm
, struct kvm_irqchip
*chip
)
1778 switch (chip
->chip_id
) {
1779 case KVM_IRQCHIP_PIC_MASTER
:
1780 memcpy(&chip
->chip
.pic
,
1781 &pic_irqchip(kvm
)->pics
[0],
1782 sizeof(struct kvm_pic_state
));
1784 case KVM_IRQCHIP_PIC_SLAVE
:
1785 memcpy(&chip
->chip
.pic
,
1786 &pic_irqchip(kvm
)->pics
[1],
1787 sizeof(struct kvm_pic_state
));
1789 case KVM_IRQCHIP_IOAPIC
:
1790 memcpy(&chip
->chip
.ioapic
,
1791 ioapic_irqchip(kvm
),
1792 sizeof(struct kvm_ioapic_state
));
1801 static int kvm_vm_ioctl_set_irqchip(struct kvm
*kvm
, struct kvm_irqchip
*chip
)
1806 switch (chip
->chip_id
) {
1807 case KVM_IRQCHIP_PIC_MASTER
:
1808 memcpy(&pic_irqchip(kvm
)->pics
[0],
1810 sizeof(struct kvm_pic_state
));
1812 case KVM_IRQCHIP_PIC_SLAVE
:
1813 memcpy(&pic_irqchip(kvm
)->pics
[1],
1815 sizeof(struct kvm_pic_state
));
1817 case KVM_IRQCHIP_IOAPIC
:
1818 memcpy(ioapic_irqchip(kvm
),
1820 sizeof(struct kvm_ioapic_state
));
1826 kvm_pic_update_irq(pic_irqchip(kvm
));
1830 static int kvm_vm_ioctl_get_pit(struct kvm
*kvm
, struct kvm_pit_state
*ps
)
1834 memcpy(ps
, &kvm
->arch
.vpit
->pit_state
, sizeof(struct kvm_pit_state
));
1838 static int kvm_vm_ioctl_set_pit(struct kvm
*kvm
, struct kvm_pit_state
*ps
)
1842 memcpy(&kvm
->arch
.vpit
->pit_state
, ps
, sizeof(struct kvm_pit_state
));
1843 kvm_pit_load_count(kvm
, 0, ps
->channels
[0].count
);
1848 * Get (and clear) the dirty memory log for a memory slot.
1850 int kvm_vm_ioctl_get_dirty_log(struct kvm
*kvm
,
1851 struct kvm_dirty_log
*log
)
1855 struct kvm_memory_slot
*memslot
;
1858 down_write(&kvm
->slots_lock
);
1860 r
= kvm_get_dirty_log(kvm
, log
, &is_dirty
);
1864 /* If nothing is dirty, don't bother messing with page tables. */
1866 kvm_mmu_slot_remove_write_access(kvm
, log
->slot
);
1867 kvm_flush_remote_tlbs(kvm
);
1868 memslot
= &kvm
->memslots
[log
->slot
];
1869 n
= ALIGN(memslot
->npages
, BITS_PER_LONG
) / 8;
1870 memset(memslot
->dirty_bitmap
, 0, n
);
1874 up_write(&kvm
->slots_lock
);
1878 long kvm_arch_vm_ioctl(struct file
*filp
,
1879 unsigned int ioctl
, unsigned long arg
)
1881 struct kvm
*kvm
= filp
->private_data
;
1882 void __user
*argp
= (void __user
*)arg
;
1885 * This union makes it completely explicit to gcc-3.x
1886 * that these two variables' stack usage should be
1887 * combined, not added together.
1890 struct kvm_pit_state ps
;
1891 struct kvm_memory_alias alias
;
1895 case KVM_SET_TSS_ADDR
:
1896 r
= kvm_vm_ioctl_set_tss_addr(kvm
, arg
);
1900 case KVM_SET_MEMORY_REGION
: {
1901 struct kvm_memory_region kvm_mem
;
1902 struct kvm_userspace_memory_region kvm_userspace_mem
;
1905 if (copy_from_user(&kvm_mem
, argp
, sizeof kvm_mem
))
1907 kvm_userspace_mem
.slot
= kvm_mem
.slot
;
1908 kvm_userspace_mem
.flags
= kvm_mem
.flags
;
1909 kvm_userspace_mem
.guest_phys_addr
= kvm_mem
.guest_phys_addr
;
1910 kvm_userspace_mem
.memory_size
= kvm_mem
.memory_size
;
1911 r
= kvm_vm_ioctl_set_memory_region(kvm
, &kvm_userspace_mem
, 0);
1916 case KVM_SET_NR_MMU_PAGES
:
1917 r
= kvm_vm_ioctl_set_nr_mmu_pages(kvm
, arg
);
1921 case KVM_GET_NR_MMU_PAGES
:
1922 r
= kvm_vm_ioctl_get_nr_mmu_pages(kvm
);
1924 case KVM_SET_MEMORY_ALIAS
:
1926 if (copy_from_user(&u
.alias
, argp
, sizeof(struct kvm_memory_alias
)))
1928 r
= kvm_vm_ioctl_set_memory_alias(kvm
, &u
.alias
);
1932 case KVM_CREATE_IRQCHIP
:
1934 kvm
->arch
.vpic
= kvm_create_pic(kvm
);
1935 if (kvm
->arch
.vpic
) {
1936 r
= kvm_ioapic_init(kvm
);
1938 kfree(kvm
->arch
.vpic
);
1939 kvm
->arch
.vpic
= NULL
;
1945 case KVM_CREATE_PIT
:
1947 kvm
->arch
.vpit
= kvm_create_pit(kvm
);
1951 case KVM_IRQ_LINE
: {
1952 struct kvm_irq_level irq_event
;
1955 if (copy_from_user(&irq_event
, argp
, sizeof irq_event
))
1957 if (irqchip_in_kernel(kvm
)) {
1958 mutex_lock(&kvm
->lock
);
1959 kvm_set_irq(kvm
, irq_event
.irq
, irq_event
.level
);
1960 mutex_unlock(&kvm
->lock
);
1965 case KVM_GET_IRQCHIP
: {
1966 /* 0: PIC master, 1: PIC slave, 2: IOAPIC */
1967 struct kvm_irqchip
*chip
= kmalloc(sizeof(*chip
), GFP_KERNEL
);
1973 if (copy_from_user(chip
, argp
, sizeof *chip
))
1974 goto get_irqchip_out
;
1976 if (!irqchip_in_kernel(kvm
))
1977 goto get_irqchip_out
;
1978 r
= kvm_vm_ioctl_get_irqchip(kvm
, chip
);
1980 goto get_irqchip_out
;
1982 if (copy_to_user(argp
, chip
, sizeof *chip
))
1983 goto get_irqchip_out
;
1991 case KVM_SET_IRQCHIP
: {
1992 /* 0: PIC master, 1: PIC slave, 2: IOAPIC */
1993 struct kvm_irqchip
*chip
= kmalloc(sizeof(*chip
), GFP_KERNEL
);
1999 if (copy_from_user(chip
, argp
, sizeof *chip
))
2000 goto set_irqchip_out
;
2002 if (!irqchip_in_kernel(kvm
))
2003 goto set_irqchip_out
;
2004 r
= kvm_vm_ioctl_set_irqchip(kvm
, chip
);
2006 goto set_irqchip_out
;
2014 case KVM_ASSIGN_PCI_DEVICE
: {
2015 struct kvm_assigned_pci_dev assigned_dev
;
2018 if (copy_from_user(&assigned_dev
, argp
, sizeof assigned_dev
))
2020 r
= kvm_vm_ioctl_assign_device(kvm
, &assigned_dev
);
2025 case KVM_ASSIGN_IRQ
: {
2026 struct kvm_assigned_irq assigned_irq
;
2029 if (copy_from_user(&assigned_irq
, argp
, sizeof assigned_irq
))
2031 r
= kvm_vm_ioctl_assign_irq(kvm
, &assigned_irq
);
2038 if (copy_from_user(&u
.ps
, argp
, sizeof(struct kvm_pit_state
)))
2041 if (!kvm
->arch
.vpit
)
2043 r
= kvm_vm_ioctl_get_pit(kvm
, &u
.ps
);
2047 if (copy_to_user(argp
, &u
.ps
, sizeof(struct kvm_pit_state
)))
2054 if (copy_from_user(&u
.ps
, argp
, sizeof u
.ps
))
2057 if (!kvm
->arch
.vpit
)
2059 r
= kvm_vm_ioctl_set_pit(kvm
, &u
.ps
);
2072 static void kvm_init_msr_list(void)
2077 for (i
= j
= 0; i
< ARRAY_SIZE(msrs_to_save
); i
++) {
2078 if (rdmsr_safe(msrs_to_save
[i
], &dummy
[0], &dummy
[1]) < 0)
2081 msrs_to_save
[j
] = msrs_to_save
[i
];
2084 num_msrs_to_save
= j
;
2088 * Only apic need an MMIO device hook, so shortcut now..
2090 static struct kvm_io_device
*vcpu_find_pervcpu_dev(struct kvm_vcpu
*vcpu
,
2091 gpa_t addr
, int len
,
2094 struct kvm_io_device
*dev
;
2096 if (vcpu
->arch
.apic
) {
2097 dev
= &vcpu
->arch
.apic
->dev
;
2098 if (dev
->in_range(dev
, addr
, len
, is_write
))
2105 static struct kvm_io_device
*vcpu_find_mmio_dev(struct kvm_vcpu
*vcpu
,
2106 gpa_t addr
, int len
,
2109 struct kvm_io_device
*dev
;
2111 dev
= vcpu_find_pervcpu_dev(vcpu
, addr
, len
, is_write
);
2113 dev
= kvm_io_bus_find_dev(&vcpu
->kvm
->mmio_bus
, addr
, len
,
2118 int emulator_read_std(unsigned long addr
,
2121 struct kvm_vcpu
*vcpu
)
2124 int r
= X86EMUL_CONTINUE
;
2127 gpa_t gpa
= vcpu
->arch
.mmu
.gva_to_gpa(vcpu
, addr
);
2128 unsigned offset
= addr
& (PAGE_SIZE
-1);
2129 unsigned tocopy
= min(bytes
, (unsigned)PAGE_SIZE
- offset
);
2132 if (gpa
== UNMAPPED_GVA
) {
2133 r
= X86EMUL_PROPAGATE_FAULT
;
2136 ret
= kvm_read_guest(vcpu
->kvm
, gpa
, data
, tocopy
);
2138 r
= X86EMUL_UNHANDLEABLE
;
2149 EXPORT_SYMBOL_GPL(emulator_read_std
);
2151 static int emulator_read_emulated(unsigned long addr
,
2154 struct kvm_vcpu
*vcpu
)
2156 struct kvm_io_device
*mmio_dev
;
2159 if (vcpu
->mmio_read_completed
) {
2160 memcpy(val
, vcpu
->mmio_data
, bytes
);
2161 vcpu
->mmio_read_completed
= 0;
2162 return X86EMUL_CONTINUE
;
2165 gpa
= vcpu
->arch
.mmu
.gva_to_gpa(vcpu
, addr
);
2167 /* For APIC access vmexit */
2168 if ((gpa
& PAGE_MASK
) == APIC_DEFAULT_PHYS_BASE
)
2171 if (emulator_read_std(addr
, val
, bytes
, vcpu
)
2172 == X86EMUL_CONTINUE
)
2173 return X86EMUL_CONTINUE
;
2174 if (gpa
== UNMAPPED_GVA
)
2175 return X86EMUL_PROPAGATE_FAULT
;
2179 * Is this MMIO handled locally?
2181 mutex_lock(&vcpu
->kvm
->lock
);
2182 mmio_dev
= vcpu_find_mmio_dev(vcpu
, gpa
, bytes
, 0);
2184 kvm_iodevice_read(mmio_dev
, gpa
, bytes
, val
);
2185 mutex_unlock(&vcpu
->kvm
->lock
);
2186 return X86EMUL_CONTINUE
;
2188 mutex_unlock(&vcpu
->kvm
->lock
);
2190 vcpu
->mmio_needed
= 1;
2191 vcpu
->mmio_phys_addr
= gpa
;
2192 vcpu
->mmio_size
= bytes
;
2193 vcpu
->mmio_is_write
= 0;
2195 return X86EMUL_UNHANDLEABLE
;
2198 int emulator_write_phys(struct kvm_vcpu
*vcpu
, gpa_t gpa
,
2199 const void *val
, int bytes
)
2203 ret
= kvm_write_guest(vcpu
->kvm
, gpa
, val
, bytes
);
2206 kvm_mmu_pte_write(vcpu
, gpa
, val
, bytes
);
2210 static int emulator_write_emulated_onepage(unsigned long addr
,
2213 struct kvm_vcpu
*vcpu
)
2215 struct kvm_io_device
*mmio_dev
;
2218 gpa
= vcpu
->arch
.mmu
.gva_to_gpa(vcpu
, addr
);
2220 if (gpa
== UNMAPPED_GVA
) {
2221 kvm_inject_page_fault(vcpu
, addr
, 2);
2222 return X86EMUL_PROPAGATE_FAULT
;
2225 /* For APIC access vmexit */
2226 if ((gpa
& PAGE_MASK
) == APIC_DEFAULT_PHYS_BASE
)
2229 if (emulator_write_phys(vcpu
, gpa
, val
, bytes
))
2230 return X86EMUL_CONTINUE
;
2234 * Is this MMIO handled locally?
2236 mutex_lock(&vcpu
->kvm
->lock
);
2237 mmio_dev
= vcpu_find_mmio_dev(vcpu
, gpa
, bytes
, 1);
2239 kvm_iodevice_write(mmio_dev
, gpa
, bytes
, val
);
2240 mutex_unlock(&vcpu
->kvm
->lock
);
2241 return X86EMUL_CONTINUE
;
2243 mutex_unlock(&vcpu
->kvm
->lock
);
2245 vcpu
->mmio_needed
= 1;
2246 vcpu
->mmio_phys_addr
= gpa
;
2247 vcpu
->mmio_size
= bytes
;
2248 vcpu
->mmio_is_write
= 1;
2249 memcpy(vcpu
->mmio_data
, val
, bytes
);
2251 return X86EMUL_CONTINUE
;
2254 int emulator_write_emulated(unsigned long addr
,
2257 struct kvm_vcpu
*vcpu
)
2259 /* Crossing a page boundary? */
2260 if (((addr
+ bytes
- 1) ^ addr
) & PAGE_MASK
) {
2263 now
= -addr
& ~PAGE_MASK
;
2264 rc
= emulator_write_emulated_onepage(addr
, val
, now
, vcpu
);
2265 if (rc
!= X86EMUL_CONTINUE
)
2271 return emulator_write_emulated_onepage(addr
, val
, bytes
, vcpu
);
2273 EXPORT_SYMBOL_GPL(emulator_write_emulated
);
2275 static int emulator_cmpxchg_emulated(unsigned long addr
,
2279 struct kvm_vcpu
*vcpu
)
2281 static int reported
;
2285 printk(KERN_WARNING
"kvm: emulating exchange as write\n");
2287 #ifndef CONFIG_X86_64
2288 /* guests cmpxchg8b have to be emulated atomically */
2295 gpa
= vcpu
->arch
.mmu
.gva_to_gpa(vcpu
, addr
);
2297 if (gpa
== UNMAPPED_GVA
||
2298 (gpa
& PAGE_MASK
) == APIC_DEFAULT_PHYS_BASE
)
2301 if (((gpa
+ bytes
- 1) & PAGE_MASK
) != (gpa
& PAGE_MASK
))
2306 down_read(¤t
->mm
->mmap_sem
);
2307 page
= gfn_to_page(vcpu
->kvm
, gpa
>> PAGE_SHIFT
);
2308 up_read(¤t
->mm
->mmap_sem
);
2310 kaddr
= kmap_atomic(page
, KM_USER0
);
2311 set_64bit((u64
*)(kaddr
+ offset_in_page(gpa
)), val
);
2312 kunmap_atomic(kaddr
, KM_USER0
);
2313 kvm_release_page_dirty(page
);
2318 return emulator_write_emulated(addr
, new, bytes
, vcpu
);
2321 static unsigned long get_segment_base(struct kvm_vcpu
*vcpu
, int seg
)
2323 return kvm_x86_ops
->get_segment_base(vcpu
, seg
);
2326 int emulate_invlpg(struct kvm_vcpu
*vcpu
, gva_t address
)
2328 return X86EMUL_CONTINUE
;
2331 int emulate_clts(struct kvm_vcpu
*vcpu
)
2333 KVMTRACE_0D(CLTS
, vcpu
, handler
);
2334 kvm_x86_ops
->set_cr0(vcpu
, vcpu
->arch
.cr0
& ~X86_CR0_TS
);
2335 return X86EMUL_CONTINUE
;
2338 int emulator_get_dr(struct x86_emulate_ctxt
*ctxt
, int dr
, unsigned long *dest
)
2340 struct kvm_vcpu
*vcpu
= ctxt
->vcpu
;
2344 *dest
= kvm_x86_ops
->get_dr(vcpu
, dr
);
2345 return X86EMUL_CONTINUE
;
2347 pr_unimpl(vcpu
, "%s: unexpected dr %u\n", __func__
, dr
);
2348 return X86EMUL_UNHANDLEABLE
;
2352 int emulator_set_dr(struct x86_emulate_ctxt
*ctxt
, int dr
, unsigned long value
)
2354 unsigned long mask
= (ctxt
->mode
== X86EMUL_MODE_PROT64
) ? ~0ULL : ~0U;
2357 kvm_x86_ops
->set_dr(ctxt
->vcpu
, dr
, value
& mask
, &exception
);
2359 /* FIXME: better handling */
2360 return X86EMUL_UNHANDLEABLE
;
2362 return X86EMUL_CONTINUE
;
2365 void kvm_report_emulation_failure(struct kvm_vcpu
*vcpu
, const char *context
)
2368 unsigned long rip
= kvm_rip_read(vcpu
);
2369 unsigned long rip_linear
;
2371 if (!printk_ratelimit())
2374 rip_linear
= rip
+ get_segment_base(vcpu
, VCPU_SREG_CS
);
2376 emulator_read_std(rip_linear
, (void *)opcodes
, 4, vcpu
);
2378 printk(KERN_ERR
"emulation failed (%s) rip %lx %02x %02x %02x %02x\n",
2379 context
, rip
, opcodes
[0], opcodes
[1], opcodes
[2], opcodes
[3]);
2381 EXPORT_SYMBOL_GPL(kvm_report_emulation_failure
);
2383 static struct x86_emulate_ops emulate_ops
= {
2384 .read_std
= emulator_read_std
,
2385 .read_emulated
= emulator_read_emulated
,
2386 .write_emulated
= emulator_write_emulated
,
2387 .cmpxchg_emulated
= emulator_cmpxchg_emulated
,
2390 static void cache_all_regs(struct kvm_vcpu
*vcpu
)
2392 kvm_register_read(vcpu
, VCPU_REGS_RAX
);
2393 kvm_register_read(vcpu
, VCPU_REGS_RSP
);
2394 kvm_register_read(vcpu
, VCPU_REGS_RIP
);
2395 vcpu
->arch
.regs_dirty
= ~0;
2398 int emulate_instruction(struct kvm_vcpu
*vcpu
,
2399 struct kvm_run
*run
,
2405 struct decode_cache
*c
;
2407 kvm_clear_exception_queue(vcpu
);
2408 vcpu
->arch
.mmio_fault_cr2
= cr2
;
2410 * TODO: fix x86_emulate.c to use guest_read/write_register
2411 * instead of direct ->regs accesses, can save hundred cycles
2412 * on Intel for instructions that don't read/change RSP, for
2415 cache_all_regs(vcpu
);
2417 vcpu
->mmio_is_write
= 0;
2418 vcpu
->arch
.pio
.string
= 0;
2420 if (!(emulation_type
& EMULTYPE_NO_DECODE
)) {
2422 kvm_x86_ops
->get_cs_db_l_bits(vcpu
, &cs_db
, &cs_l
);
2424 vcpu
->arch
.emulate_ctxt
.vcpu
= vcpu
;
2425 vcpu
->arch
.emulate_ctxt
.eflags
= kvm_x86_ops
->get_rflags(vcpu
);
2426 vcpu
->arch
.emulate_ctxt
.mode
=
2427 (vcpu
->arch
.emulate_ctxt
.eflags
& X86_EFLAGS_VM
)
2428 ? X86EMUL_MODE_REAL
: cs_l
2429 ? X86EMUL_MODE_PROT64
: cs_db
2430 ? X86EMUL_MODE_PROT32
: X86EMUL_MODE_PROT16
;
2432 r
= x86_decode_insn(&vcpu
->arch
.emulate_ctxt
, &emulate_ops
);
2434 /* Reject the instructions other than VMCALL/VMMCALL when
2435 * try to emulate invalid opcode */
2436 c
= &vcpu
->arch
.emulate_ctxt
.decode
;
2437 if ((emulation_type
& EMULTYPE_TRAP_UD
) &&
2438 (!(c
->twobyte
&& c
->b
== 0x01 &&
2439 (c
->modrm_reg
== 0 || c
->modrm_reg
== 3) &&
2440 c
->modrm_mod
== 3 && c
->modrm_rm
== 1)))
2441 return EMULATE_FAIL
;
2443 ++vcpu
->stat
.insn_emulation
;
2445 ++vcpu
->stat
.insn_emulation_fail
;
2446 if (kvm_mmu_unprotect_page_virt(vcpu
, cr2
))
2447 return EMULATE_DONE
;
2448 return EMULATE_FAIL
;
2452 r
= x86_emulate_insn(&vcpu
->arch
.emulate_ctxt
, &emulate_ops
);
2454 if (vcpu
->arch
.pio
.string
)
2455 return EMULATE_DO_MMIO
;
2457 if ((r
|| vcpu
->mmio_is_write
) && run
) {
2458 run
->exit_reason
= KVM_EXIT_MMIO
;
2459 run
->mmio
.phys_addr
= vcpu
->mmio_phys_addr
;
2460 memcpy(run
->mmio
.data
, vcpu
->mmio_data
, 8);
2461 run
->mmio
.len
= vcpu
->mmio_size
;
2462 run
->mmio
.is_write
= vcpu
->mmio_is_write
;
2466 if (kvm_mmu_unprotect_page_virt(vcpu
, cr2
))
2467 return EMULATE_DONE
;
2468 if (!vcpu
->mmio_needed
) {
2469 kvm_report_emulation_failure(vcpu
, "mmio");
2470 return EMULATE_FAIL
;
2472 return EMULATE_DO_MMIO
;
2475 kvm_x86_ops
->set_rflags(vcpu
, vcpu
->arch
.emulate_ctxt
.eflags
);
2477 if (vcpu
->mmio_is_write
) {
2478 vcpu
->mmio_needed
= 0;
2479 return EMULATE_DO_MMIO
;
2482 return EMULATE_DONE
;
2484 EXPORT_SYMBOL_GPL(emulate_instruction
);
2486 static void free_pio_guest_pages(struct kvm_vcpu
*vcpu
)
2490 for (i
= 0; i
< ARRAY_SIZE(vcpu
->arch
.pio
.guest_pages
); ++i
)
2491 if (vcpu
->arch
.pio
.guest_pages
[i
]) {
2492 kvm_release_page_dirty(vcpu
->arch
.pio
.guest_pages
[i
]);
2493 vcpu
->arch
.pio
.guest_pages
[i
] = NULL
;
2497 static int pio_copy_data(struct kvm_vcpu
*vcpu
)
2499 void *p
= vcpu
->arch
.pio_data
;
2502 int nr_pages
= vcpu
->arch
.pio
.guest_pages
[1] ? 2 : 1;
2504 q
= vmap(vcpu
->arch
.pio
.guest_pages
, nr_pages
, VM_READ
|VM_WRITE
,
2507 free_pio_guest_pages(vcpu
);
2510 q
+= vcpu
->arch
.pio
.guest_page_offset
;
2511 bytes
= vcpu
->arch
.pio
.size
* vcpu
->arch
.pio
.cur_count
;
2512 if (vcpu
->arch
.pio
.in
)
2513 memcpy(q
, p
, bytes
);
2515 memcpy(p
, q
, bytes
);
2516 q
-= vcpu
->arch
.pio
.guest_page_offset
;
2518 free_pio_guest_pages(vcpu
);
2522 int complete_pio(struct kvm_vcpu
*vcpu
)
2524 struct kvm_pio_request
*io
= &vcpu
->arch
.pio
;
2531 val
= kvm_register_read(vcpu
, VCPU_REGS_RAX
);
2532 memcpy(&val
, vcpu
->arch
.pio_data
, io
->size
);
2533 kvm_register_write(vcpu
, VCPU_REGS_RAX
, val
);
2537 r
= pio_copy_data(vcpu
);
2544 delta
*= io
->cur_count
;
2546 * The size of the register should really depend on
2547 * current address size.
2549 val
= kvm_register_read(vcpu
, VCPU_REGS_RCX
);
2551 kvm_register_write(vcpu
, VCPU_REGS_RCX
, val
);
2557 val
= kvm_register_read(vcpu
, VCPU_REGS_RDI
);
2559 kvm_register_write(vcpu
, VCPU_REGS_RDI
, val
);
2561 val
= kvm_register_read(vcpu
, VCPU_REGS_RSI
);
2563 kvm_register_write(vcpu
, VCPU_REGS_RSI
, val
);
2567 io
->count
-= io
->cur_count
;
2573 static void kernel_pio(struct kvm_io_device
*pio_dev
,
2574 struct kvm_vcpu
*vcpu
,
2577 /* TODO: String I/O for in kernel device */
2579 mutex_lock(&vcpu
->kvm
->lock
);
2580 if (vcpu
->arch
.pio
.in
)
2581 kvm_iodevice_read(pio_dev
, vcpu
->arch
.pio
.port
,
2582 vcpu
->arch
.pio
.size
,
2585 kvm_iodevice_write(pio_dev
, vcpu
->arch
.pio
.port
,
2586 vcpu
->arch
.pio
.size
,
2588 mutex_unlock(&vcpu
->kvm
->lock
);
2591 static void pio_string_write(struct kvm_io_device
*pio_dev
,
2592 struct kvm_vcpu
*vcpu
)
2594 struct kvm_pio_request
*io
= &vcpu
->arch
.pio
;
2595 void *pd
= vcpu
->arch
.pio_data
;
2598 mutex_lock(&vcpu
->kvm
->lock
);
2599 for (i
= 0; i
< io
->cur_count
; i
++) {
2600 kvm_iodevice_write(pio_dev
, io
->port
,
2605 mutex_unlock(&vcpu
->kvm
->lock
);
2608 static struct kvm_io_device
*vcpu_find_pio_dev(struct kvm_vcpu
*vcpu
,
2609 gpa_t addr
, int len
,
2612 return kvm_io_bus_find_dev(&vcpu
->kvm
->pio_bus
, addr
, len
, is_write
);
2615 int kvm_emulate_pio(struct kvm_vcpu
*vcpu
, struct kvm_run
*run
, int in
,
2616 int size
, unsigned port
)
2618 struct kvm_io_device
*pio_dev
;
2621 vcpu
->run
->exit_reason
= KVM_EXIT_IO
;
2622 vcpu
->run
->io
.direction
= in
? KVM_EXIT_IO_IN
: KVM_EXIT_IO_OUT
;
2623 vcpu
->run
->io
.size
= vcpu
->arch
.pio
.size
= size
;
2624 vcpu
->run
->io
.data_offset
= KVM_PIO_PAGE_OFFSET
* PAGE_SIZE
;
2625 vcpu
->run
->io
.count
= vcpu
->arch
.pio
.count
= vcpu
->arch
.pio
.cur_count
= 1;
2626 vcpu
->run
->io
.port
= vcpu
->arch
.pio
.port
= port
;
2627 vcpu
->arch
.pio
.in
= in
;
2628 vcpu
->arch
.pio
.string
= 0;
2629 vcpu
->arch
.pio
.down
= 0;
2630 vcpu
->arch
.pio
.guest_page_offset
= 0;
2631 vcpu
->arch
.pio
.rep
= 0;
2633 if (vcpu
->run
->io
.direction
== KVM_EXIT_IO_IN
)
2634 KVMTRACE_2D(IO_READ
, vcpu
, vcpu
->run
->io
.port
, (u32
)size
,
2637 KVMTRACE_2D(IO_WRITE
, vcpu
, vcpu
->run
->io
.port
, (u32
)size
,
2640 val
= kvm_register_read(vcpu
, VCPU_REGS_RAX
);
2641 memcpy(vcpu
->arch
.pio_data
, &val
, 4);
2643 kvm_x86_ops
->skip_emulated_instruction(vcpu
);
2645 pio_dev
= vcpu_find_pio_dev(vcpu
, port
, size
, !in
);
2647 kernel_pio(pio_dev
, vcpu
, vcpu
->arch
.pio_data
);
2653 EXPORT_SYMBOL_GPL(kvm_emulate_pio
);
2655 int kvm_emulate_pio_string(struct kvm_vcpu
*vcpu
, struct kvm_run
*run
, int in
,
2656 int size
, unsigned long count
, int down
,
2657 gva_t address
, int rep
, unsigned port
)
2659 unsigned now
, in_page
;
2663 struct kvm_io_device
*pio_dev
;
2665 vcpu
->run
->exit_reason
= KVM_EXIT_IO
;
2666 vcpu
->run
->io
.direction
= in
? KVM_EXIT_IO_IN
: KVM_EXIT_IO_OUT
;
2667 vcpu
->run
->io
.size
= vcpu
->arch
.pio
.size
= size
;
2668 vcpu
->run
->io
.data_offset
= KVM_PIO_PAGE_OFFSET
* PAGE_SIZE
;
2669 vcpu
->run
->io
.count
= vcpu
->arch
.pio
.count
= vcpu
->arch
.pio
.cur_count
= count
;
2670 vcpu
->run
->io
.port
= vcpu
->arch
.pio
.port
= port
;
2671 vcpu
->arch
.pio
.in
= in
;
2672 vcpu
->arch
.pio
.string
= 1;
2673 vcpu
->arch
.pio
.down
= down
;
2674 vcpu
->arch
.pio
.guest_page_offset
= offset_in_page(address
);
2675 vcpu
->arch
.pio
.rep
= rep
;
2677 if (vcpu
->run
->io
.direction
== KVM_EXIT_IO_IN
)
2678 KVMTRACE_2D(IO_READ
, vcpu
, vcpu
->run
->io
.port
, (u32
)size
,
2681 KVMTRACE_2D(IO_WRITE
, vcpu
, vcpu
->run
->io
.port
, (u32
)size
,
2685 kvm_x86_ops
->skip_emulated_instruction(vcpu
);
2690 in_page
= PAGE_SIZE
- offset_in_page(address
);
2692 in_page
= offset_in_page(address
) + size
;
2693 now
= min(count
, (unsigned long)in_page
/ size
);
2696 * String I/O straddles page boundary. Pin two guest pages
2697 * so that we satisfy atomicity constraints. Do just one
2698 * transaction to avoid complexity.
2705 * String I/O in reverse. Yuck. Kill the guest, fix later.
2707 pr_unimpl(vcpu
, "guest string pio down\n");
2708 kvm_inject_gp(vcpu
, 0);
2711 vcpu
->run
->io
.count
= now
;
2712 vcpu
->arch
.pio
.cur_count
= now
;
2714 if (vcpu
->arch
.pio
.cur_count
== vcpu
->arch
.pio
.count
)
2715 kvm_x86_ops
->skip_emulated_instruction(vcpu
);
2717 for (i
= 0; i
< nr_pages
; ++i
) {
2718 page
= gva_to_page(vcpu
, address
+ i
* PAGE_SIZE
);
2719 vcpu
->arch
.pio
.guest_pages
[i
] = page
;
2721 kvm_inject_gp(vcpu
, 0);
2722 free_pio_guest_pages(vcpu
);
2727 pio_dev
= vcpu_find_pio_dev(vcpu
, port
,
2728 vcpu
->arch
.pio
.cur_count
,
2729 !vcpu
->arch
.pio
.in
);
2730 if (!vcpu
->arch
.pio
.in
) {
2731 /* string PIO write */
2732 ret
= pio_copy_data(vcpu
);
2733 if (ret
>= 0 && pio_dev
) {
2734 pio_string_write(pio_dev
, vcpu
);
2736 if (vcpu
->arch
.pio
.count
== 0)
2740 pr_unimpl(vcpu
, "no string pio read support yet, "
2741 "port %x size %d count %ld\n",
2746 EXPORT_SYMBOL_GPL(kvm_emulate_pio_string
);
2748 int kvm_arch_init(void *opaque
)
2751 struct kvm_x86_ops
*ops
= (struct kvm_x86_ops
*)opaque
;
2754 printk(KERN_ERR
"kvm: already loaded the other module\n");
2759 if (!ops
->cpu_has_kvm_support()) {
2760 printk(KERN_ERR
"kvm: no hardware support\n");
2764 if (ops
->disabled_by_bios()) {
2765 printk(KERN_ERR
"kvm: disabled by bios\n");
2770 r
= kvm_mmu_module_init();
2774 kvm_init_msr_list();
2777 kvm_mmu_set_nonpresent_ptes(0ull, 0ull);
2778 kvm_mmu_set_base_ptes(PT_PRESENT_MASK
);
2779 kvm_mmu_set_mask_ptes(PT_USER_MASK
, PT_ACCESSED_MASK
,
2780 PT_DIRTY_MASK
, PT64_NX_MASK
, 0);
2787 void kvm_arch_exit(void)
2790 kvm_mmu_module_exit();
2793 int kvm_emulate_halt(struct kvm_vcpu
*vcpu
)
2795 ++vcpu
->stat
.halt_exits
;
2796 KVMTRACE_0D(HLT
, vcpu
, handler
);
2797 if (irqchip_in_kernel(vcpu
->kvm
)) {
2798 vcpu
->arch
.mp_state
= KVM_MP_STATE_HALTED
;
2799 up_read(&vcpu
->kvm
->slots_lock
);
2800 kvm_vcpu_block(vcpu
);
2801 down_read(&vcpu
->kvm
->slots_lock
);
2802 if (vcpu
->arch
.mp_state
!= KVM_MP_STATE_RUNNABLE
)
2806 vcpu
->run
->exit_reason
= KVM_EXIT_HLT
;
2810 EXPORT_SYMBOL_GPL(kvm_emulate_halt
);
2812 static inline gpa_t
hc_gpa(struct kvm_vcpu
*vcpu
, unsigned long a0
,
2815 if (is_long_mode(vcpu
))
2818 return a0
| ((gpa_t
)a1
<< 32);
2821 int kvm_emulate_hypercall(struct kvm_vcpu
*vcpu
)
2823 unsigned long nr
, a0
, a1
, a2
, a3
, ret
;
2826 nr
= kvm_register_read(vcpu
, VCPU_REGS_RAX
);
2827 a0
= kvm_register_read(vcpu
, VCPU_REGS_RBX
);
2828 a1
= kvm_register_read(vcpu
, VCPU_REGS_RCX
);
2829 a2
= kvm_register_read(vcpu
, VCPU_REGS_RDX
);
2830 a3
= kvm_register_read(vcpu
, VCPU_REGS_RSI
);
2832 KVMTRACE_1D(VMMCALL
, vcpu
, (u32
)nr
, handler
);
2834 if (!is_long_mode(vcpu
)) {
2843 case KVM_HC_VAPIC_POLL_IRQ
:
2847 r
= kvm_pv_mmu_op(vcpu
, a0
, hc_gpa(vcpu
, a1
, a2
), &ret
);
2853 kvm_register_write(vcpu
, VCPU_REGS_RAX
, ret
);
2854 ++vcpu
->stat
.hypercalls
;
2857 EXPORT_SYMBOL_GPL(kvm_emulate_hypercall
);
2859 int kvm_fix_hypercall(struct kvm_vcpu
*vcpu
)
2861 char instruction
[3];
2863 unsigned long rip
= kvm_rip_read(vcpu
);
2867 * Blow out the MMU to ensure that no other VCPU has an active mapping
2868 * to ensure that the updated hypercall appears atomically across all
2871 kvm_mmu_zap_all(vcpu
->kvm
);
2873 kvm_x86_ops
->patch_hypercall(vcpu
, instruction
);
2874 if (emulator_write_emulated(rip
, instruction
, 3, vcpu
)
2875 != X86EMUL_CONTINUE
)
2881 static u64
mk_cr_64(u64 curr_cr
, u32 new_val
)
2883 return (curr_cr
& ~((1ULL << 32) - 1)) | new_val
;
2886 void realmode_lgdt(struct kvm_vcpu
*vcpu
, u16 limit
, unsigned long base
)
2888 struct descriptor_table dt
= { limit
, base
};
2890 kvm_x86_ops
->set_gdt(vcpu
, &dt
);
2893 void realmode_lidt(struct kvm_vcpu
*vcpu
, u16 limit
, unsigned long base
)
2895 struct descriptor_table dt
= { limit
, base
};
2897 kvm_x86_ops
->set_idt(vcpu
, &dt
);
2900 void realmode_lmsw(struct kvm_vcpu
*vcpu
, unsigned long msw
,
2901 unsigned long *rflags
)
2903 kvm_lmsw(vcpu
, msw
);
2904 *rflags
= kvm_x86_ops
->get_rflags(vcpu
);
2907 unsigned long realmode_get_cr(struct kvm_vcpu
*vcpu
, int cr
)
2909 unsigned long value
;
2911 kvm_x86_ops
->decache_cr4_guest_bits(vcpu
);
2914 value
= vcpu
->arch
.cr0
;
2917 value
= vcpu
->arch
.cr2
;
2920 value
= vcpu
->arch
.cr3
;
2923 value
= vcpu
->arch
.cr4
;
2926 value
= kvm_get_cr8(vcpu
);
2929 vcpu_printf(vcpu
, "%s: unexpected cr %u\n", __func__
, cr
);
2932 KVMTRACE_3D(CR_READ
, vcpu
, (u32
)cr
, (u32
)value
,
2933 (u32
)((u64
)value
>> 32), handler
);
2938 void realmode_set_cr(struct kvm_vcpu
*vcpu
, int cr
, unsigned long val
,
2939 unsigned long *rflags
)
2941 KVMTRACE_3D(CR_WRITE
, vcpu
, (u32
)cr
, (u32
)val
,
2942 (u32
)((u64
)val
>> 32), handler
);
2946 kvm_set_cr0(vcpu
, mk_cr_64(vcpu
->arch
.cr0
, val
));
2947 *rflags
= kvm_x86_ops
->get_rflags(vcpu
);
2950 vcpu
->arch
.cr2
= val
;
2953 kvm_set_cr3(vcpu
, val
);
2956 kvm_set_cr4(vcpu
, mk_cr_64(vcpu
->arch
.cr4
, val
));
2959 kvm_set_cr8(vcpu
, val
& 0xfUL
);
2962 vcpu_printf(vcpu
, "%s: unexpected cr %u\n", __func__
, cr
);
2966 static int move_to_next_stateful_cpuid_entry(struct kvm_vcpu
*vcpu
, int i
)
2968 struct kvm_cpuid_entry2
*e
= &vcpu
->arch
.cpuid_entries
[i
];
2969 int j
, nent
= vcpu
->arch
.cpuid_nent
;
2971 e
->flags
&= ~KVM_CPUID_FLAG_STATE_READ_NEXT
;
2972 /* when no next entry is found, the current entry[i] is reselected */
2973 for (j
= i
+ 1; j
== i
; j
= (j
+ 1) % nent
) {
2974 struct kvm_cpuid_entry2
*ej
= &vcpu
->arch
.cpuid_entries
[j
];
2975 if (ej
->function
== e
->function
) {
2976 ej
->flags
|= KVM_CPUID_FLAG_STATE_READ_NEXT
;
2980 return 0; /* silence gcc, even though control never reaches here */
2983 /* find an entry with matching function, matching index (if needed), and that
2984 * should be read next (if it's stateful) */
2985 static int is_matching_cpuid_entry(struct kvm_cpuid_entry2
*e
,
2986 u32 function
, u32 index
)
2988 if (e
->function
!= function
)
2990 if ((e
->flags
& KVM_CPUID_FLAG_SIGNIFCANT_INDEX
) && e
->index
!= index
)
2992 if ((e
->flags
& KVM_CPUID_FLAG_STATEFUL_FUNC
) &&
2993 !(e
->flags
& KVM_CPUID_FLAG_STATE_READ_NEXT
))
2998 void kvm_emulate_cpuid(struct kvm_vcpu
*vcpu
)
3001 u32 function
, index
;
3002 struct kvm_cpuid_entry2
*e
, *best
;
3004 function
= kvm_register_read(vcpu
, VCPU_REGS_RAX
);
3005 index
= kvm_register_read(vcpu
, VCPU_REGS_RCX
);
3006 kvm_register_write(vcpu
, VCPU_REGS_RAX
, 0);
3007 kvm_register_write(vcpu
, VCPU_REGS_RBX
, 0);
3008 kvm_register_write(vcpu
, VCPU_REGS_RCX
, 0);
3009 kvm_register_write(vcpu
, VCPU_REGS_RDX
, 0);
3011 for (i
= 0; i
< vcpu
->arch
.cpuid_nent
; ++i
) {
3012 e
= &vcpu
->arch
.cpuid_entries
[i
];
3013 if (is_matching_cpuid_entry(e
, function
, index
)) {
3014 if (e
->flags
& KVM_CPUID_FLAG_STATEFUL_FUNC
)
3015 move_to_next_stateful_cpuid_entry(vcpu
, i
);
3020 * Both basic or both extended?
3022 if (((e
->function
^ function
) & 0x80000000) == 0)
3023 if (!best
|| e
->function
> best
->function
)
3027 kvm_register_write(vcpu
, VCPU_REGS_RAX
, best
->eax
);
3028 kvm_register_write(vcpu
, VCPU_REGS_RBX
, best
->ebx
);
3029 kvm_register_write(vcpu
, VCPU_REGS_RCX
, best
->ecx
);
3030 kvm_register_write(vcpu
, VCPU_REGS_RDX
, best
->edx
);
3032 kvm_x86_ops
->skip_emulated_instruction(vcpu
);
3033 KVMTRACE_5D(CPUID
, vcpu
, function
,
3034 (u32
)kvm_register_read(vcpu
, VCPU_REGS_RAX
),
3035 (u32
)kvm_register_read(vcpu
, VCPU_REGS_RBX
),
3036 (u32
)kvm_register_read(vcpu
, VCPU_REGS_RCX
),
3037 (u32
)kvm_register_read(vcpu
, VCPU_REGS_RDX
), handler
);
3039 EXPORT_SYMBOL_GPL(kvm_emulate_cpuid
);
3042 * Check if userspace requested an interrupt window, and that the
3043 * interrupt window is open.
3045 * No need to exit to userspace if we already have an interrupt queued.
3047 static int dm_request_for_irq_injection(struct kvm_vcpu
*vcpu
,
3048 struct kvm_run
*kvm_run
)
3050 return (!vcpu
->arch
.irq_summary
&&
3051 kvm_run
->request_interrupt_window
&&
3052 vcpu
->arch
.interrupt_window_open
&&
3053 (kvm_x86_ops
->get_rflags(vcpu
) & X86_EFLAGS_IF
));
3056 static void post_kvm_run_save(struct kvm_vcpu
*vcpu
,
3057 struct kvm_run
*kvm_run
)
3059 kvm_run
->if_flag
= (kvm_x86_ops
->get_rflags(vcpu
) & X86_EFLAGS_IF
) != 0;
3060 kvm_run
->cr8
= kvm_get_cr8(vcpu
);
3061 kvm_run
->apic_base
= kvm_get_apic_base(vcpu
);
3062 if (irqchip_in_kernel(vcpu
->kvm
))
3063 kvm_run
->ready_for_interrupt_injection
= 1;
3065 kvm_run
->ready_for_interrupt_injection
=
3066 (vcpu
->arch
.interrupt_window_open
&&
3067 vcpu
->arch
.irq_summary
== 0);
3070 static void vapic_enter(struct kvm_vcpu
*vcpu
)
3072 struct kvm_lapic
*apic
= vcpu
->arch
.apic
;
3075 if (!apic
|| !apic
->vapic_addr
)
3078 down_read(¤t
->mm
->mmap_sem
);
3079 page
= gfn_to_page(vcpu
->kvm
, apic
->vapic_addr
>> PAGE_SHIFT
);
3080 up_read(¤t
->mm
->mmap_sem
);
3082 vcpu
->arch
.apic
->vapic_page
= page
;
3085 static void vapic_exit(struct kvm_vcpu
*vcpu
)
3087 struct kvm_lapic
*apic
= vcpu
->arch
.apic
;
3089 if (!apic
|| !apic
->vapic_addr
)
3092 down_read(&vcpu
->kvm
->slots_lock
);
3093 kvm_release_page_dirty(apic
->vapic_page
);
3094 mark_page_dirty(vcpu
->kvm
, apic
->vapic_addr
>> PAGE_SHIFT
);
3095 up_read(&vcpu
->kvm
->slots_lock
);
3098 static int __vcpu_run(struct kvm_vcpu
*vcpu
, struct kvm_run
*kvm_run
)
3102 if (unlikely(vcpu
->arch
.mp_state
== KVM_MP_STATE_SIPI_RECEIVED
)) {
3103 pr_debug("vcpu %d received sipi with vector # %x\n",
3104 vcpu
->vcpu_id
, vcpu
->arch
.sipi_vector
);
3105 kvm_lapic_reset(vcpu
);
3106 r
= kvm_x86_ops
->vcpu_reset(vcpu
);
3109 vcpu
->arch
.mp_state
= KVM_MP_STATE_RUNNABLE
;
3112 down_read(&vcpu
->kvm
->slots_lock
);
3117 if (test_and_clear_bit(KVM_REQ_MMU_RELOAD
, &vcpu
->requests
))
3118 kvm_mmu_unload(vcpu
);
3120 r
= kvm_mmu_reload(vcpu
);
3124 if (vcpu
->requests
) {
3125 if (test_and_clear_bit(KVM_REQ_MIGRATE_TIMER
, &vcpu
->requests
))
3126 __kvm_migrate_timers(vcpu
);
3127 if (test_and_clear_bit(KVM_REQ_TLB_FLUSH
, &vcpu
->requests
))
3128 kvm_x86_ops
->tlb_flush(vcpu
);
3129 if (test_and_clear_bit(KVM_REQ_REPORT_TPR_ACCESS
,
3131 kvm_run
->exit_reason
= KVM_EXIT_TPR_ACCESS
;
3135 if (test_and_clear_bit(KVM_REQ_TRIPLE_FAULT
, &vcpu
->requests
)) {
3136 kvm_run
->exit_reason
= KVM_EXIT_SHUTDOWN
;
3142 clear_bit(KVM_REQ_PENDING_TIMER
, &vcpu
->requests
);
3143 kvm_inject_pending_timer_irqs(vcpu
);
3147 kvm_x86_ops
->prepare_guest_switch(vcpu
);
3148 kvm_load_guest_fpu(vcpu
);
3150 local_irq_disable();
3152 if (vcpu
->requests
|| need_resched()) {
3159 if (signal_pending(current
)) {
3163 kvm_run
->exit_reason
= KVM_EXIT_INTR
;
3164 ++vcpu
->stat
.signal_exits
;
3168 if (vcpu
->guest_debug
.enabled
)
3169 kvm_x86_ops
->guest_debug_pre(vcpu
);
3171 vcpu
->guest_mode
= 1;
3173 * Make sure that guest_mode assignment won't happen after
3174 * testing the pending IRQ vector bitmap.
3178 if (vcpu
->arch
.exception
.pending
)
3179 __queue_exception(vcpu
);
3180 else if (irqchip_in_kernel(vcpu
->kvm
))
3181 kvm_x86_ops
->inject_pending_irq(vcpu
);
3183 kvm_x86_ops
->inject_pending_vectors(vcpu
, kvm_run
);
3185 kvm_lapic_sync_to_vapic(vcpu
);
3187 up_read(&vcpu
->kvm
->slots_lock
);
3192 KVMTRACE_0D(VMENTRY
, vcpu
, entryexit
);
3193 kvm_x86_ops
->run(vcpu
, kvm_run
);
3195 vcpu
->guest_mode
= 0;
3201 * We must have an instruction between local_irq_enable() and
3202 * kvm_guest_exit(), so the timer interrupt isn't delayed by
3203 * the interrupt shadow. The stat.exits increment will do nicely.
3204 * But we need to prevent reordering, hence this barrier():
3212 down_read(&vcpu
->kvm
->slots_lock
);
3215 * Profile KVM exit RIPs:
3217 if (unlikely(prof_on
== KVM_PROFILING
)) {
3218 unsigned long rip
= kvm_rip_read(vcpu
);
3219 profile_hit(KVM_PROFILING
, (void *)rip
);
3222 if (vcpu
->arch
.exception
.pending
&& kvm_x86_ops
->exception_injected(vcpu
))
3223 vcpu
->arch
.exception
.pending
= false;
3225 kvm_lapic_sync_from_vapic(vcpu
);
3227 r
= kvm_x86_ops
->handle_exit(kvm_run
, vcpu
);
3230 if (dm_request_for_irq_injection(vcpu
, kvm_run
)) {
3232 kvm_run
->exit_reason
= KVM_EXIT_INTR
;
3233 ++vcpu
->stat
.request_irq_exits
;
3236 if (!need_resched())
3241 up_read(&vcpu
->kvm
->slots_lock
);
3244 down_read(&vcpu
->kvm
->slots_lock
);
3248 post_kvm_run_save(vcpu
, kvm_run
);
3255 int kvm_arch_vcpu_ioctl_run(struct kvm_vcpu
*vcpu
, struct kvm_run
*kvm_run
)
3262 if (vcpu
->sigset_active
)
3263 sigprocmask(SIG_SETMASK
, &vcpu
->sigset
, &sigsaved
);
3265 if (unlikely(vcpu
->arch
.mp_state
== KVM_MP_STATE_UNINITIALIZED
)) {
3266 kvm_vcpu_block(vcpu
);
3271 /* re-sync apic's tpr */
3272 if (!irqchip_in_kernel(vcpu
->kvm
))
3273 kvm_set_cr8(vcpu
, kvm_run
->cr8
);
3275 if (vcpu
->arch
.pio
.cur_count
) {
3276 r
= complete_pio(vcpu
);
3280 #if CONFIG_HAS_IOMEM
3281 if (vcpu
->mmio_needed
) {
3282 memcpy(vcpu
->mmio_data
, kvm_run
->mmio
.data
, 8);
3283 vcpu
->mmio_read_completed
= 1;
3284 vcpu
->mmio_needed
= 0;
3286 down_read(&vcpu
->kvm
->slots_lock
);
3287 r
= emulate_instruction(vcpu
, kvm_run
,
3288 vcpu
->arch
.mmio_fault_cr2
, 0,
3289 EMULTYPE_NO_DECODE
);
3290 up_read(&vcpu
->kvm
->slots_lock
);
3291 if (r
== EMULATE_DO_MMIO
) {
3293 * Read-modify-write. Back to userspace.
3300 if (kvm_run
->exit_reason
== KVM_EXIT_HYPERCALL
)
3301 kvm_register_write(vcpu
, VCPU_REGS_RAX
,
3302 kvm_run
->hypercall
.ret
);
3304 r
= __vcpu_run(vcpu
, kvm_run
);
3307 if (vcpu
->sigset_active
)
3308 sigprocmask(SIG_SETMASK
, &sigsaved
, NULL
);
3314 int kvm_arch_vcpu_ioctl_get_regs(struct kvm_vcpu
*vcpu
, struct kvm_regs
*regs
)
3318 regs
->rax
= kvm_register_read(vcpu
, VCPU_REGS_RAX
);
3319 regs
->rbx
= kvm_register_read(vcpu
, VCPU_REGS_RBX
);
3320 regs
->rcx
= kvm_register_read(vcpu
, VCPU_REGS_RCX
);
3321 regs
->rdx
= kvm_register_read(vcpu
, VCPU_REGS_RDX
);
3322 regs
->rsi
= kvm_register_read(vcpu
, VCPU_REGS_RSI
);
3323 regs
->rdi
= kvm_register_read(vcpu
, VCPU_REGS_RDI
);
3324 regs
->rsp
= kvm_register_read(vcpu
, VCPU_REGS_RSP
);
3325 regs
->rbp
= kvm_register_read(vcpu
, VCPU_REGS_RBP
);
3326 #ifdef CONFIG_X86_64
3327 regs
->r8
= kvm_register_read(vcpu
, VCPU_REGS_R8
);
3328 regs
->r9
= kvm_register_read(vcpu
, VCPU_REGS_R9
);
3329 regs
->r10
= kvm_register_read(vcpu
, VCPU_REGS_R10
);
3330 regs
->r11
= kvm_register_read(vcpu
, VCPU_REGS_R11
);
3331 regs
->r12
= kvm_register_read(vcpu
, VCPU_REGS_R12
);
3332 regs
->r13
= kvm_register_read(vcpu
, VCPU_REGS_R13
);
3333 regs
->r14
= kvm_register_read(vcpu
, VCPU_REGS_R14
);
3334 regs
->r15
= kvm_register_read(vcpu
, VCPU_REGS_R15
);
3337 regs
->rip
= kvm_rip_read(vcpu
);
3338 regs
->rflags
= kvm_x86_ops
->get_rflags(vcpu
);
3341 * Don't leak debug flags in case they were set for guest debugging
3343 if (vcpu
->guest_debug
.enabled
&& vcpu
->guest_debug
.singlestep
)
3344 regs
->rflags
&= ~(X86_EFLAGS_TF
| X86_EFLAGS_RF
);
3351 int kvm_arch_vcpu_ioctl_set_regs(struct kvm_vcpu
*vcpu
, struct kvm_regs
*regs
)
3355 kvm_register_write(vcpu
, VCPU_REGS_RAX
, regs
->rax
);
3356 kvm_register_write(vcpu
, VCPU_REGS_RBX
, regs
->rbx
);
3357 kvm_register_write(vcpu
, VCPU_REGS_RCX
, regs
->rcx
);
3358 kvm_register_write(vcpu
, VCPU_REGS_RDX
, regs
->rdx
);
3359 kvm_register_write(vcpu
, VCPU_REGS_RSI
, regs
->rsi
);
3360 kvm_register_write(vcpu
, VCPU_REGS_RDI
, regs
->rdi
);
3361 kvm_register_write(vcpu
, VCPU_REGS_RSP
, regs
->rsp
);
3362 kvm_register_write(vcpu
, VCPU_REGS_RBP
, regs
->rbp
);
3363 #ifdef CONFIG_X86_64
3364 kvm_register_write(vcpu
, VCPU_REGS_R8
, regs
->r8
);
3365 kvm_register_write(vcpu
, VCPU_REGS_R9
, regs
->r9
);
3366 kvm_register_write(vcpu
, VCPU_REGS_R10
, regs
->r10
);
3367 kvm_register_write(vcpu
, VCPU_REGS_R11
, regs
->r11
);
3368 kvm_register_write(vcpu
, VCPU_REGS_R12
, regs
->r12
);
3369 kvm_register_write(vcpu
, VCPU_REGS_R13
, regs
->r13
);
3370 kvm_register_write(vcpu
, VCPU_REGS_R14
, regs
->r14
);
3371 kvm_register_write(vcpu
, VCPU_REGS_R15
, regs
->r15
);
3375 kvm_rip_write(vcpu
, regs
->rip
);
3376 kvm_x86_ops
->set_rflags(vcpu
, regs
->rflags
);
3379 vcpu
->arch
.exception
.pending
= false;
3386 void kvm_get_segment(struct kvm_vcpu
*vcpu
,
3387 struct kvm_segment
*var
, int seg
)
3389 kvm_x86_ops
->get_segment(vcpu
, var
, seg
);
3392 void kvm_get_cs_db_l_bits(struct kvm_vcpu
*vcpu
, int *db
, int *l
)
3394 struct kvm_segment cs
;
3396 kvm_get_segment(vcpu
, &cs
, VCPU_SREG_CS
);
3400 EXPORT_SYMBOL_GPL(kvm_get_cs_db_l_bits
);
3402 int kvm_arch_vcpu_ioctl_get_sregs(struct kvm_vcpu
*vcpu
,
3403 struct kvm_sregs
*sregs
)
3405 struct descriptor_table dt
;
3410 kvm_get_segment(vcpu
, &sregs
->cs
, VCPU_SREG_CS
);
3411 kvm_get_segment(vcpu
, &sregs
->ds
, VCPU_SREG_DS
);
3412 kvm_get_segment(vcpu
, &sregs
->es
, VCPU_SREG_ES
);
3413 kvm_get_segment(vcpu
, &sregs
->fs
, VCPU_SREG_FS
);
3414 kvm_get_segment(vcpu
, &sregs
->gs
, VCPU_SREG_GS
);
3415 kvm_get_segment(vcpu
, &sregs
->ss
, VCPU_SREG_SS
);
3417 kvm_get_segment(vcpu
, &sregs
->tr
, VCPU_SREG_TR
);
3418 kvm_get_segment(vcpu
, &sregs
->ldt
, VCPU_SREG_LDTR
);
3420 kvm_x86_ops
->get_idt(vcpu
, &dt
);
3421 sregs
->idt
.limit
= dt
.limit
;
3422 sregs
->idt
.base
= dt
.base
;
3423 kvm_x86_ops
->get_gdt(vcpu
, &dt
);
3424 sregs
->gdt
.limit
= dt
.limit
;
3425 sregs
->gdt
.base
= dt
.base
;
3427 kvm_x86_ops
->decache_cr4_guest_bits(vcpu
);
3428 sregs
->cr0
= vcpu
->arch
.cr0
;
3429 sregs
->cr2
= vcpu
->arch
.cr2
;
3430 sregs
->cr3
= vcpu
->arch
.cr3
;
3431 sregs
->cr4
= vcpu
->arch
.cr4
;
3432 sregs
->cr8
= kvm_get_cr8(vcpu
);
3433 sregs
->efer
= vcpu
->arch
.shadow_efer
;
3434 sregs
->apic_base
= kvm_get_apic_base(vcpu
);
3436 if (irqchip_in_kernel(vcpu
->kvm
)) {
3437 memset(sregs
->interrupt_bitmap
, 0,
3438 sizeof sregs
->interrupt_bitmap
);
3439 pending_vec
= kvm_x86_ops
->get_irq(vcpu
);
3440 if (pending_vec
>= 0)
3441 set_bit(pending_vec
,
3442 (unsigned long *)sregs
->interrupt_bitmap
);
3444 memcpy(sregs
->interrupt_bitmap
, vcpu
->arch
.irq_pending
,
3445 sizeof sregs
->interrupt_bitmap
);
3452 int kvm_arch_vcpu_ioctl_get_mpstate(struct kvm_vcpu
*vcpu
,
3453 struct kvm_mp_state
*mp_state
)
3456 mp_state
->mp_state
= vcpu
->arch
.mp_state
;
3461 int kvm_arch_vcpu_ioctl_set_mpstate(struct kvm_vcpu
*vcpu
,
3462 struct kvm_mp_state
*mp_state
)
3465 vcpu
->arch
.mp_state
= mp_state
->mp_state
;
3470 static void kvm_set_segment(struct kvm_vcpu
*vcpu
,
3471 struct kvm_segment
*var
, int seg
)
3473 kvm_x86_ops
->set_segment(vcpu
, var
, seg
);
3476 static void seg_desct_to_kvm_desct(struct desc_struct
*seg_desc
, u16 selector
,
3477 struct kvm_segment
*kvm_desct
)
3479 kvm_desct
->base
= seg_desc
->base0
;
3480 kvm_desct
->base
|= seg_desc
->base1
<< 16;
3481 kvm_desct
->base
|= seg_desc
->base2
<< 24;
3482 kvm_desct
->limit
= seg_desc
->limit0
;
3483 kvm_desct
->limit
|= seg_desc
->limit
<< 16;
3485 kvm_desct
->limit
<<= 12;
3486 kvm_desct
->limit
|= 0xfff;
3488 kvm_desct
->selector
= selector
;
3489 kvm_desct
->type
= seg_desc
->type
;
3490 kvm_desct
->present
= seg_desc
->p
;
3491 kvm_desct
->dpl
= seg_desc
->dpl
;
3492 kvm_desct
->db
= seg_desc
->d
;
3493 kvm_desct
->s
= seg_desc
->s
;
3494 kvm_desct
->l
= seg_desc
->l
;
3495 kvm_desct
->g
= seg_desc
->g
;
3496 kvm_desct
->avl
= seg_desc
->avl
;
3498 kvm_desct
->unusable
= 1;
3500 kvm_desct
->unusable
= 0;
3501 kvm_desct
->padding
= 0;
3504 static void get_segment_descritptor_dtable(struct kvm_vcpu
*vcpu
,
3506 struct descriptor_table
*dtable
)
3508 if (selector
& 1 << 2) {
3509 struct kvm_segment kvm_seg
;
3511 kvm_get_segment(vcpu
, &kvm_seg
, VCPU_SREG_LDTR
);
3513 if (kvm_seg
.unusable
)
3516 dtable
->limit
= kvm_seg
.limit
;
3517 dtable
->base
= kvm_seg
.base
;
3520 kvm_x86_ops
->get_gdt(vcpu
, dtable
);
3523 /* allowed just for 8 bytes segments */
3524 static int load_guest_segment_descriptor(struct kvm_vcpu
*vcpu
, u16 selector
,
3525 struct desc_struct
*seg_desc
)
3528 struct descriptor_table dtable
;
3529 u16 index
= selector
>> 3;
3531 get_segment_descritptor_dtable(vcpu
, selector
, &dtable
);
3533 if (dtable
.limit
< index
* 8 + 7) {
3534 kvm_queue_exception_e(vcpu
, GP_VECTOR
, selector
& 0xfffc);
3537 gpa
= vcpu
->arch
.mmu
.gva_to_gpa(vcpu
, dtable
.base
);
3539 return kvm_read_guest(vcpu
->kvm
, gpa
, seg_desc
, 8);
3542 /* allowed just for 8 bytes segments */
3543 static int save_guest_segment_descriptor(struct kvm_vcpu
*vcpu
, u16 selector
,
3544 struct desc_struct
*seg_desc
)
3547 struct descriptor_table dtable
;
3548 u16 index
= selector
>> 3;
3550 get_segment_descritptor_dtable(vcpu
, selector
, &dtable
);
3552 if (dtable
.limit
< index
* 8 + 7)
3554 gpa
= vcpu
->arch
.mmu
.gva_to_gpa(vcpu
, dtable
.base
);
3556 return kvm_write_guest(vcpu
->kvm
, gpa
, seg_desc
, 8);
3559 static u32
get_tss_base_addr(struct kvm_vcpu
*vcpu
,
3560 struct desc_struct
*seg_desc
)
3564 base_addr
= seg_desc
->base0
;
3565 base_addr
|= (seg_desc
->base1
<< 16);
3566 base_addr
|= (seg_desc
->base2
<< 24);
3568 return vcpu
->arch
.mmu
.gva_to_gpa(vcpu
, base_addr
);
3571 static u16
get_segment_selector(struct kvm_vcpu
*vcpu
, int seg
)
3573 struct kvm_segment kvm_seg
;
3575 kvm_get_segment(vcpu
, &kvm_seg
, seg
);
3576 return kvm_seg
.selector
;
3579 static int load_segment_descriptor_to_kvm_desct(struct kvm_vcpu
*vcpu
,
3581 struct kvm_segment
*kvm_seg
)
3583 struct desc_struct seg_desc
;
3585 if (load_guest_segment_descriptor(vcpu
, selector
, &seg_desc
))
3587 seg_desct_to_kvm_desct(&seg_desc
, selector
, kvm_seg
);
3591 int kvm_load_realmode_segment(struct kvm_vcpu
*vcpu
, u16 selector
, int seg
)
3593 struct kvm_segment segvar
= {
3594 .base
= selector
<< 4,
3596 .selector
= selector
,
3607 kvm_x86_ops
->set_segment(vcpu
, &segvar
, seg
);
3611 int kvm_load_segment_descriptor(struct kvm_vcpu
*vcpu
, u16 selector
,
3612 int type_bits
, int seg
)
3614 struct kvm_segment kvm_seg
;
3616 if (!(vcpu
->arch
.cr0
& X86_CR0_PE
))
3617 return kvm_load_realmode_segment(vcpu
, selector
, seg
);
3618 if (load_segment_descriptor_to_kvm_desct(vcpu
, selector
, &kvm_seg
))
3620 kvm_seg
.type
|= type_bits
;
3622 if (seg
!= VCPU_SREG_SS
&& seg
!= VCPU_SREG_CS
&&
3623 seg
!= VCPU_SREG_LDTR
)
3625 kvm_seg
.unusable
= 1;
3627 kvm_set_segment(vcpu
, &kvm_seg
, seg
);
3631 static void save_state_to_tss32(struct kvm_vcpu
*vcpu
,
3632 struct tss_segment_32
*tss
)
3634 tss
->cr3
= vcpu
->arch
.cr3
;
3635 tss
->eip
= kvm_rip_read(vcpu
);
3636 tss
->eflags
= kvm_x86_ops
->get_rflags(vcpu
);
3637 tss
->eax
= kvm_register_read(vcpu
, VCPU_REGS_RAX
);
3638 tss
->ecx
= kvm_register_read(vcpu
, VCPU_REGS_RCX
);
3639 tss
->edx
= kvm_register_read(vcpu
, VCPU_REGS_RDX
);
3640 tss
->ebx
= kvm_register_read(vcpu
, VCPU_REGS_RBX
);
3641 tss
->esp
= kvm_register_read(vcpu
, VCPU_REGS_RSP
);
3642 tss
->ebp
= kvm_register_read(vcpu
, VCPU_REGS_RBP
);
3643 tss
->esi
= kvm_register_read(vcpu
, VCPU_REGS_RSI
);
3644 tss
->edi
= kvm_register_read(vcpu
, VCPU_REGS_RDI
);
3645 tss
->es
= get_segment_selector(vcpu
, VCPU_SREG_ES
);
3646 tss
->cs
= get_segment_selector(vcpu
, VCPU_SREG_CS
);
3647 tss
->ss
= get_segment_selector(vcpu
, VCPU_SREG_SS
);
3648 tss
->ds
= get_segment_selector(vcpu
, VCPU_SREG_DS
);
3649 tss
->fs
= get_segment_selector(vcpu
, VCPU_SREG_FS
);
3650 tss
->gs
= get_segment_selector(vcpu
, VCPU_SREG_GS
);
3651 tss
->ldt_selector
= get_segment_selector(vcpu
, VCPU_SREG_LDTR
);
3652 tss
->prev_task_link
= get_segment_selector(vcpu
, VCPU_SREG_TR
);
3655 static int load_state_from_tss32(struct kvm_vcpu
*vcpu
,
3656 struct tss_segment_32
*tss
)
3658 kvm_set_cr3(vcpu
, tss
->cr3
);
3660 kvm_rip_write(vcpu
, tss
->eip
);
3661 kvm_x86_ops
->set_rflags(vcpu
, tss
->eflags
| 2);
3663 kvm_register_write(vcpu
, VCPU_REGS_RAX
, tss
->eax
);
3664 kvm_register_write(vcpu
, VCPU_REGS_RCX
, tss
->ecx
);
3665 kvm_register_write(vcpu
, VCPU_REGS_RDX
, tss
->edx
);
3666 kvm_register_write(vcpu
, VCPU_REGS_RBX
, tss
->ebx
);
3667 kvm_register_write(vcpu
, VCPU_REGS_RSP
, tss
->esp
);
3668 kvm_register_write(vcpu
, VCPU_REGS_RBP
, tss
->ebp
);
3669 kvm_register_write(vcpu
, VCPU_REGS_RSI
, tss
->esi
);
3670 kvm_register_write(vcpu
, VCPU_REGS_RDI
, tss
->edi
);
3672 if (kvm_load_segment_descriptor(vcpu
, tss
->ldt_selector
, 0, VCPU_SREG_LDTR
))
3675 if (kvm_load_segment_descriptor(vcpu
, tss
->es
, 1, VCPU_SREG_ES
))
3678 if (kvm_load_segment_descriptor(vcpu
, tss
->cs
, 9, VCPU_SREG_CS
))
3681 if (kvm_load_segment_descriptor(vcpu
, tss
->ss
, 1, VCPU_SREG_SS
))
3684 if (kvm_load_segment_descriptor(vcpu
, tss
->ds
, 1, VCPU_SREG_DS
))
3687 if (kvm_load_segment_descriptor(vcpu
, tss
->fs
, 1, VCPU_SREG_FS
))
3690 if (kvm_load_segment_descriptor(vcpu
, tss
->gs
, 1, VCPU_SREG_GS
))
3695 static void save_state_to_tss16(struct kvm_vcpu
*vcpu
,
3696 struct tss_segment_16
*tss
)
3698 tss
->ip
= kvm_rip_read(vcpu
);
3699 tss
->flag
= kvm_x86_ops
->get_rflags(vcpu
);
3700 tss
->ax
= kvm_register_read(vcpu
, VCPU_REGS_RAX
);
3701 tss
->cx
= kvm_register_read(vcpu
, VCPU_REGS_RCX
);
3702 tss
->dx
= kvm_register_read(vcpu
, VCPU_REGS_RDX
);
3703 tss
->bx
= kvm_register_read(vcpu
, VCPU_REGS_RBX
);
3704 tss
->sp
= kvm_register_read(vcpu
, VCPU_REGS_RSP
);
3705 tss
->bp
= kvm_register_read(vcpu
, VCPU_REGS_RBP
);
3706 tss
->si
= kvm_register_read(vcpu
, VCPU_REGS_RSI
);
3707 tss
->di
= kvm_register_read(vcpu
, VCPU_REGS_RDI
);
3709 tss
->es
= get_segment_selector(vcpu
, VCPU_SREG_ES
);
3710 tss
->cs
= get_segment_selector(vcpu
, VCPU_SREG_CS
);
3711 tss
->ss
= get_segment_selector(vcpu
, VCPU_SREG_SS
);
3712 tss
->ds
= get_segment_selector(vcpu
, VCPU_SREG_DS
);
3713 tss
->ldt
= get_segment_selector(vcpu
, VCPU_SREG_LDTR
);
3714 tss
->prev_task_link
= get_segment_selector(vcpu
, VCPU_SREG_TR
);
3717 static int load_state_from_tss16(struct kvm_vcpu
*vcpu
,
3718 struct tss_segment_16
*tss
)
3720 kvm_rip_write(vcpu
, tss
->ip
);
3721 kvm_x86_ops
->set_rflags(vcpu
, tss
->flag
| 2);
3722 kvm_register_write(vcpu
, VCPU_REGS_RAX
, tss
->ax
);
3723 kvm_register_write(vcpu
, VCPU_REGS_RCX
, tss
->cx
);
3724 kvm_register_write(vcpu
, VCPU_REGS_RDX
, tss
->dx
);
3725 kvm_register_write(vcpu
, VCPU_REGS_RBX
, tss
->bx
);
3726 kvm_register_write(vcpu
, VCPU_REGS_RSP
, tss
->sp
);
3727 kvm_register_write(vcpu
, VCPU_REGS_RBP
, tss
->bp
);
3728 kvm_register_write(vcpu
, VCPU_REGS_RSI
, tss
->si
);
3729 kvm_register_write(vcpu
, VCPU_REGS_RDI
, tss
->di
);
3731 if (kvm_load_segment_descriptor(vcpu
, tss
->ldt
, 0, VCPU_SREG_LDTR
))
3734 if (kvm_load_segment_descriptor(vcpu
, tss
->es
, 1, VCPU_SREG_ES
))
3737 if (kvm_load_segment_descriptor(vcpu
, tss
->cs
, 9, VCPU_SREG_CS
))
3740 if (kvm_load_segment_descriptor(vcpu
, tss
->ss
, 1, VCPU_SREG_SS
))
3743 if (kvm_load_segment_descriptor(vcpu
, tss
->ds
, 1, VCPU_SREG_DS
))
3748 static int kvm_task_switch_16(struct kvm_vcpu
*vcpu
, u16 tss_selector
,
3750 struct desc_struct
*nseg_desc
)
3752 struct tss_segment_16 tss_segment_16
;
3755 if (kvm_read_guest(vcpu
->kvm
, old_tss_base
, &tss_segment_16
,
3756 sizeof tss_segment_16
))
3759 save_state_to_tss16(vcpu
, &tss_segment_16
);
3761 if (kvm_write_guest(vcpu
->kvm
, old_tss_base
, &tss_segment_16
,
3762 sizeof tss_segment_16
))
3765 if (kvm_read_guest(vcpu
->kvm
, get_tss_base_addr(vcpu
, nseg_desc
),
3766 &tss_segment_16
, sizeof tss_segment_16
))
3769 if (load_state_from_tss16(vcpu
, &tss_segment_16
))
3777 static int kvm_task_switch_32(struct kvm_vcpu
*vcpu
, u16 tss_selector
,
3779 struct desc_struct
*nseg_desc
)
3781 struct tss_segment_32 tss_segment_32
;
3784 if (kvm_read_guest(vcpu
->kvm
, old_tss_base
, &tss_segment_32
,
3785 sizeof tss_segment_32
))
3788 save_state_to_tss32(vcpu
, &tss_segment_32
);
3790 if (kvm_write_guest(vcpu
->kvm
, old_tss_base
, &tss_segment_32
,
3791 sizeof tss_segment_32
))
3794 if (kvm_read_guest(vcpu
->kvm
, get_tss_base_addr(vcpu
, nseg_desc
),
3795 &tss_segment_32
, sizeof tss_segment_32
))
3798 if (load_state_from_tss32(vcpu
, &tss_segment_32
))
3806 int kvm_task_switch(struct kvm_vcpu
*vcpu
, u16 tss_selector
, int reason
)
3808 struct kvm_segment tr_seg
;
3809 struct desc_struct cseg_desc
;
3810 struct desc_struct nseg_desc
;
3812 u32 old_tss_base
= get_segment_base(vcpu
, VCPU_SREG_TR
);
3813 u16 old_tss_sel
= get_segment_selector(vcpu
, VCPU_SREG_TR
);
3815 old_tss_base
= vcpu
->arch
.mmu
.gva_to_gpa(vcpu
, old_tss_base
);
3817 /* FIXME: Handle errors. Failure to read either TSS or their
3818 * descriptors should generate a pagefault.
3820 if (load_guest_segment_descriptor(vcpu
, tss_selector
, &nseg_desc
))
3823 if (load_guest_segment_descriptor(vcpu
, old_tss_sel
, &cseg_desc
))
3826 if (reason
!= TASK_SWITCH_IRET
) {
3829 cpl
= kvm_x86_ops
->get_cpl(vcpu
);
3830 if ((tss_selector
& 3) > nseg_desc
.dpl
|| cpl
> nseg_desc
.dpl
) {
3831 kvm_queue_exception_e(vcpu
, GP_VECTOR
, 0);
3836 if (!nseg_desc
.p
|| (nseg_desc
.limit0
| nseg_desc
.limit
<< 16) < 0x67) {
3837 kvm_queue_exception_e(vcpu
, TS_VECTOR
, tss_selector
& 0xfffc);
3841 if (reason
== TASK_SWITCH_IRET
|| reason
== TASK_SWITCH_JMP
) {
3842 cseg_desc
.type
&= ~(1 << 1); //clear the B flag
3843 save_guest_segment_descriptor(vcpu
, old_tss_sel
, &cseg_desc
);
3846 if (reason
== TASK_SWITCH_IRET
) {
3847 u32 eflags
= kvm_x86_ops
->get_rflags(vcpu
);
3848 kvm_x86_ops
->set_rflags(vcpu
, eflags
& ~X86_EFLAGS_NT
);
3851 kvm_x86_ops
->skip_emulated_instruction(vcpu
);
3853 if (nseg_desc
.type
& 8)
3854 ret
= kvm_task_switch_32(vcpu
, tss_selector
, old_tss_base
,
3857 ret
= kvm_task_switch_16(vcpu
, tss_selector
, old_tss_base
,
3860 if (reason
== TASK_SWITCH_CALL
|| reason
== TASK_SWITCH_GATE
) {
3861 u32 eflags
= kvm_x86_ops
->get_rflags(vcpu
);
3862 kvm_x86_ops
->set_rflags(vcpu
, eflags
| X86_EFLAGS_NT
);
3865 if (reason
!= TASK_SWITCH_IRET
) {
3866 nseg_desc
.type
|= (1 << 1);
3867 save_guest_segment_descriptor(vcpu
, tss_selector
,
3871 kvm_x86_ops
->set_cr0(vcpu
, vcpu
->arch
.cr0
| X86_CR0_TS
);
3872 seg_desct_to_kvm_desct(&nseg_desc
, tss_selector
, &tr_seg
);
3874 kvm_set_segment(vcpu
, &tr_seg
, VCPU_SREG_TR
);
3878 EXPORT_SYMBOL_GPL(kvm_task_switch
);
3880 int kvm_arch_vcpu_ioctl_set_sregs(struct kvm_vcpu
*vcpu
,
3881 struct kvm_sregs
*sregs
)
3883 int mmu_reset_needed
= 0;
3884 int i
, pending_vec
, max_bits
;
3885 struct descriptor_table dt
;
3889 dt
.limit
= sregs
->idt
.limit
;
3890 dt
.base
= sregs
->idt
.base
;
3891 kvm_x86_ops
->set_idt(vcpu
, &dt
);
3892 dt
.limit
= sregs
->gdt
.limit
;
3893 dt
.base
= sregs
->gdt
.base
;
3894 kvm_x86_ops
->set_gdt(vcpu
, &dt
);
3896 vcpu
->arch
.cr2
= sregs
->cr2
;
3897 mmu_reset_needed
|= vcpu
->arch
.cr3
!= sregs
->cr3
;
3898 vcpu
->arch
.cr3
= sregs
->cr3
;
3900 kvm_set_cr8(vcpu
, sregs
->cr8
);
3902 mmu_reset_needed
|= vcpu
->arch
.shadow_efer
!= sregs
->efer
;
3903 kvm_x86_ops
->set_efer(vcpu
, sregs
->efer
);
3904 kvm_set_apic_base(vcpu
, sregs
->apic_base
);
3906 kvm_x86_ops
->decache_cr4_guest_bits(vcpu
);
3908 mmu_reset_needed
|= vcpu
->arch
.cr0
!= sregs
->cr0
;
3909 kvm_x86_ops
->set_cr0(vcpu
, sregs
->cr0
);
3910 vcpu
->arch
.cr0
= sregs
->cr0
;
3912 mmu_reset_needed
|= vcpu
->arch
.cr4
!= sregs
->cr4
;
3913 kvm_x86_ops
->set_cr4(vcpu
, sregs
->cr4
);
3914 if (!is_long_mode(vcpu
) && is_pae(vcpu
))
3915 load_pdptrs(vcpu
, vcpu
->arch
.cr3
);
3917 if (mmu_reset_needed
)
3918 kvm_mmu_reset_context(vcpu
);
3920 if (!irqchip_in_kernel(vcpu
->kvm
)) {
3921 memcpy(vcpu
->arch
.irq_pending
, sregs
->interrupt_bitmap
,
3922 sizeof vcpu
->arch
.irq_pending
);
3923 vcpu
->arch
.irq_summary
= 0;
3924 for (i
= 0; i
< ARRAY_SIZE(vcpu
->arch
.irq_pending
); ++i
)
3925 if (vcpu
->arch
.irq_pending
[i
])
3926 __set_bit(i
, &vcpu
->arch
.irq_summary
);
3928 max_bits
= (sizeof sregs
->interrupt_bitmap
) << 3;
3929 pending_vec
= find_first_bit(
3930 (const unsigned long *)sregs
->interrupt_bitmap
,
3932 /* Only pending external irq is handled here */
3933 if (pending_vec
< max_bits
) {
3934 kvm_x86_ops
->set_irq(vcpu
, pending_vec
);
3935 pr_debug("Set back pending irq %d\n",
3940 kvm_set_segment(vcpu
, &sregs
->cs
, VCPU_SREG_CS
);
3941 kvm_set_segment(vcpu
, &sregs
->ds
, VCPU_SREG_DS
);
3942 kvm_set_segment(vcpu
, &sregs
->es
, VCPU_SREG_ES
);
3943 kvm_set_segment(vcpu
, &sregs
->fs
, VCPU_SREG_FS
);
3944 kvm_set_segment(vcpu
, &sregs
->gs
, VCPU_SREG_GS
);
3945 kvm_set_segment(vcpu
, &sregs
->ss
, VCPU_SREG_SS
);
3947 kvm_set_segment(vcpu
, &sregs
->tr
, VCPU_SREG_TR
);
3948 kvm_set_segment(vcpu
, &sregs
->ldt
, VCPU_SREG_LDTR
);
3955 int kvm_arch_vcpu_ioctl_debug_guest(struct kvm_vcpu
*vcpu
,
3956 struct kvm_debug_guest
*dbg
)
3962 r
= kvm_x86_ops
->set_guest_debug(vcpu
, dbg
);
3970 * fxsave fpu state. Taken from x86_64/processor.h. To be killed when
3971 * we have asm/x86/processor.h
3982 u32 st_space
[32]; /* 8*16 bytes for each FP-reg = 128 bytes */
3983 #ifdef CONFIG_X86_64
3984 u32 xmm_space
[64]; /* 16*16 bytes for each XMM-reg = 256 bytes */
3986 u32 xmm_space
[32]; /* 8*16 bytes for each XMM-reg = 128 bytes */
3991 * Translate a guest virtual address to a guest physical address.
3993 int kvm_arch_vcpu_ioctl_translate(struct kvm_vcpu
*vcpu
,
3994 struct kvm_translation
*tr
)
3996 unsigned long vaddr
= tr
->linear_address
;
4000 down_read(&vcpu
->kvm
->slots_lock
);
4001 gpa
= vcpu
->arch
.mmu
.gva_to_gpa(vcpu
, vaddr
);
4002 up_read(&vcpu
->kvm
->slots_lock
);
4003 tr
->physical_address
= gpa
;
4004 tr
->valid
= gpa
!= UNMAPPED_GVA
;
4012 int kvm_arch_vcpu_ioctl_get_fpu(struct kvm_vcpu
*vcpu
, struct kvm_fpu
*fpu
)
4014 struct fxsave
*fxsave
= (struct fxsave
*)&vcpu
->arch
.guest_fx_image
;
4018 memcpy(fpu
->fpr
, fxsave
->st_space
, 128);
4019 fpu
->fcw
= fxsave
->cwd
;
4020 fpu
->fsw
= fxsave
->swd
;
4021 fpu
->ftwx
= fxsave
->twd
;
4022 fpu
->last_opcode
= fxsave
->fop
;
4023 fpu
->last_ip
= fxsave
->rip
;
4024 fpu
->last_dp
= fxsave
->rdp
;
4025 memcpy(fpu
->xmm
, fxsave
->xmm_space
, sizeof fxsave
->xmm_space
);
4032 int kvm_arch_vcpu_ioctl_set_fpu(struct kvm_vcpu
*vcpu
, struct kvm_fpu
*fpu
)
4034 struct fxsave
*fxsave
= (struct fxsave
*)&vcpu
->arch
.guest_fx_image
;
4038 memcpy(fxsave
->st_space
, fpu
->fpr
, 128);
4039 fxsave
->cwd
= fpu
->fcw
;
4040 fxsave
->swd
= fpu
->fsw
;
4041 fxsave
->twd
= fpu
->ftwx
;
4042 fxsave
->fop
= fpu
->last_opcode
;
4043 fxsave
->rip
= fpu
->last_ip
;
4044 fxsave
->rdp
= fpu
->last_dp
;
4045 memcpy(fxsave
->xmm_space
, fpu
->xmm
, sizeof fxsave
->xmm_space
);
4052 void fx_init(struct kvm_vcpu
*vcpu
)
4054 unsigned after_mxcsr_mask
;
4057 * Touch the fpu the first time in non atomic context as if
4058 * this is the first fpu instruction the exception handler
4059 * will fire before the instruction returns and it'll have to
4060 * allocate ram with GFP_KERNEL.
4063 kvm_fx_save(&vcpu
->arch
.host_fx_image
);
4065 /* Initialize guest FPU by resetting ours and saving into guest's */
4067 kvm_fx_save(&vcpu
->arch
.host_fx_image
);
4069 kvm_fx_save(&vcpu
->arch
.guest_fx_image
);
4070 kvm_fx_restore(&vcpu
->arch
.host_fx_image
);
4073 vcpu
->arch
.cr0
|= X86_CR0_ET
;
4074 after_mxcsr_mask
= offsetof(struct i387_fxsave_struct
, st_space
);
4075 vcpu
->arch
.guest_fx_image
.mxcsr
= 0x1f80;
4076 memset((void *)&vcpu
->arch
.guest_fx_image
+ after_mxcsr_mask
,
4077 0, sizeof(struct i387_fxsave_struct
) - after_mxcsr_mask
);
4079 EXPORT_SYMBOL_GPL(fx_init
);
4081 void kvm_load_guest_fpu(struct kvm_vcpu
*vcpu
)
4083 if (!vcpu
->fpu_active
|| vcpu
->guest_fpu_loaded
)
4086 vcpu
->guest_fpu_loaded
= 1;
4087 kvm_fx_save(&vcpu
->arch
.host_fx_image
);
4088 kvm_fx_restore(&vcpu
->arch
.guest_fx_image
);
4090 EXPORT_SYMBOL_GPL(kvm_load_guest_fpu
);
4092 void kvm_put_guest_fpu(struct kvm_vcpu
*vcpu
)
4094 if (!vcpu
->guest_fpu_loaded
)
4097 vcpu
->guest_fpu_loaded
= 0;
4098 kvm_fx_save(&vcpu
->arch
.guest_fx_image
);
4099 kvm_fx_restore(&vcpu
->arch
.host_fx_image
);
4100 ++vcpu
->stat
.fpu_reload
;
4102 EXPORT_SYMBOL_GPL(kvm_put_guest_fpu
);
4104 void kvm_arch_vcpu_free(struct kvm_vcpu
*vcpu
)
4106 kvm_x86_ops
->vcpu_free(vcpu
);
4109 struct kvm_vcpu
*kvm_arch_vcpu_create(struct kvm
*kvm
,
4112 return kvm_x86_ops
->vcpu_create(kvm
, id
);
4115 int kvm_arch_vcpu_setup(struct kvm_vcpu
*vcpu
)
4119 /* We do fxsave: this must be aligned. */
4120 BUG_ON((unsigned long)&vcpu
->arch
.host_fx_image
& 0xF);
4123 r
= kvm_arch_vcpu_reset(vcpu
);
4125 r
= kvm_mmu_setup(vcpu
);
4132 kvm_x86_ops
->vcpu_free(vcpu
);
4136 void kvm_arch_vcpu_destroy(struct kvm_vcpu
*vcpu
)
4139 kvm_mmu_unload(vcpu
);
4142 kvm_x86_ops
->vcpu_free(vcpu
);
4145 int kvm_arch_vcpu_reset(struct kvm_vcpu
*vcpu
)
4147 return kvm_x86_ops
->vcpu_reset(vcpu
);
4150 void kvm_arch_hardware_enable(void *garbage
)
4152 kvm_x86_ops
->hardware_enable(garbage
);
4155 void kvm_arch_hardware_disable(void *garbage
)
4157 kvm_x86_ops
->hardware_disable(garbage
);
4160 int kvm_arch_hardware_setup(void)
4162 return kvm_x86_ops
->hardware_setup();
4165 void kvm_arch_hardware_unsetup(void)
4167 kvm_x86_ops
->hardware_unsetup();
4170 void kvm_arch_check_processor_compat(void *rtn
)
4172 kvm_x86_ops
->check_processor_compatibility(rtn
);
4175 int kvm_arch_vcpu_init(struct kvm_vcpu
*vcpu
)
4181 BUG_ON(vcpu
->kvm
== NULL
);
4184 vcpu
->arch
.mmu
.root_hpa
= INVALID_PAGE
;
4185 if (!irqchip_in_kernel(kvm
) || vcpu
->vcpu_id
== 0)
4186 vcpu
->arch
.mp_state
= KVM_MP_STATE_RUNNABLE
;
4188 vcpu
->arch
.mp_state
= KVM_MP_STATE_UNINITIALIZED
;
4190 page
= alloc_page(GFP_KERNEL
| __GFP_ZERO
);
4195 vcpu
->arch
.pio_data
= page_address(page
);
4197 r
= kvm_mmu_create(vcpu
);
4199 goto fail_free_pio_data
;
4201 if (irqchip_in_kernel(kvm
)) {
4202 r
= kvm_create_lapic(vcpu
);
4204 goto fail_mmu_destroy
;
4210 kvm_mmu_destroy(vcpu
);
4212 free_page((unsigned long)vcpu
->arch
.pio_data
);
4217 void kvm_arch_vcpu_uninit(struct kvm_vcpu
*vcpu
)
4219 kvm_free_lapic(vcpu
);
4220 down_read(&vcpu
->kvm
->slots_lock
);
4221 kvm_mmu_destroy(vcpu
);
4222 up_read(&vcpu
->kvm
->slots_lock
);
4223 free_page((unsigned long)vcpu
->arch
.pio_data
);
4226 struct kvm
*kvm_arch_create_vm(void)
4228 struct kvm
*kvm
= kzalloc(sizeof(struct kvm
), GFP_KERNEL
);
4231 return ERR_PTR(-ENOMEM
);
4233 INIT_LIST_HEAD(&kvm
->arch
.active_mmu_pages
);
4234 INIT_LIST_HEAD(&kvm
->arch
.assigned_dev_head
);
4239 static void kvm_unload_vcpu_mmu(struct kvm_vcpu
*vcpu
)
4242 kvm_mmu_unload(vcpu
);
4246 static void kvm_free_vcpus(struct kvm
*kvm
)
4251 * Unpin any mmu pages first.
4253 for (i
= 0; i
< KVM_MAX_VCPUS
; ++i
)
4255 kvm_unload_vcpu_mmu(kvm
->vcpus
[i
]);
4256 for (i
= 0; i
< KVM_MAX_VCPUS
; ++i
) {
4257 if (kvm
->vcpus
[i
]) {
4258 kvm_arch_vcpu_free(kvm
->vcpus
[i
]);
4259 kvm
->vcpus
[i
] = NULL
;
4265 void kvm_arch_destroy_vm(struct kvm
*kvm
)
4267 kvm_free_assigned_devices(kvm
);
4269 kfree(kvm
->arch
.vpic
);
4270 kfree(kvm
->arch
.vioapic
);
4271 kvm_free_vcpus(kvm
);
4272 kvm_free_physmem(kvm
);
4273 if (kvm
->arch
.apic_access_page
)
4274 put_page(kvm
->arch
.apic_access_page
);
4275 if (kvm
->arch
.ept_identity_pagetable
)
4276 put_page(kvm
->arch
.ept_identity_pagetable
);
4280 int kvm_arch_set_memory_region(struct kvm
*kvm
,
4281 struct kvm_userspace_memory_region
*mem
,
4282 struct kvm_memory_slot old
,
4285 int npages
= mem
->memory_size
>> PAGE_SHIFT
;
4286 struct kvm_memory_slot
*memslot
= &kvm
->memslots
[mem
->slot
];
4288 /*To keep backward compatibility with older userspace,
4289 *x86 needs to hanlde !user_alloc case.
4292 if (npages
&& !old
.rmap
) {
4293 unsigned long userspace_addr
;
4295 down_write(¤t
->mm
->mmap_sem
);
4296 userspace_addr
= do_mmap(NULL
, 0,
4298 PROT_READ
| PROT_WRITE
,
4299 MAP_PRIVATE
| MAP_ANONYMOUS
,
4301 up_write(¤t
->mm
->mmap_sem
);
4303 if (IS_ERR((void *)userspace_addr
))
4304 return PTR_ERR((void *)userspace_addr
);
4306 /* set userspace_addr atomically for kvm_hva_to_rmapp */
4307 spin_lock(&kvm
->mmu_lock
);
4308 memslot
->userspace_addr
= userspace_addr
;
4309 spin_unlock(&kvm
->mmu_lock
);
4311 if (!old
.user_alloc
&& old
.rmap
) {
4314 down_write(¤t
->mm
->mmap_sem
);
4315 ret
= do_munmap(current
->mm
, old
.userspace_addr
,
4316 old
.npages
* PAGE_SIZE
);
4317 up_write(¤t
->mm
->mmap_sem
);
4320 "kvm_vm_ioctl_set_memory_region: "
4321 "failed to munmap memory\n");
4326 if (!kvm
->arch
.n_requested_mmu_pages
) {
4327 unsigned int nr_mmu_pages
= kvm_mmu_calculate_mmu_pages(kvm
);
4328 kvm_mmu_change_mmu_pages(kvm
, nr_mmu_pages
);
4331 kvm_mmu_slot_remove_write_access(kvm
, mem
->slot
);
4332 kvm_flush_remote_tlbs(kvm
);
4337 void kvm_arch_flush_shadow(struct kvm
*kvm
)
4339 kvm_mmu_zap_all(kvm
);
4342 int kvm_arch_vcpu_runnable(struct kvm_vcpu
*vcpu
)
4344 return vcpu
->arch
.mp_state
== KVM_MP_STATE_RUNNABLE
4345 || vcpu
->arch
.mp_state
== KVM_MP_STATE_SIPI_RECEIVED
;
4348 static void vcpu_kick_intr(void *info
)
4351 struct kvm_vcpu
*vcpu
= (struct kvm_vcpu
*)info
;
4352 printk(KERN_DEBUG
"vcpu_kick_intr %p \n", vcpu
);
4356 void kvm_vcpu_kick(struct kvm_vcpu
*vcpu
)
4358 int ipi_pcpu
= vcpu
->cpu
;
4359 int cpu
= get_cpu();
4361 if (waitqueue_active(&vcpu
->wq
)) {
4362 wake_up_interruptible(&vcpu
->wq
);
4363 ++vcpu
->stat
.halt_wakeup
;
4366 * We may be called synchronously with irqs disabled in guest mode,
4367 * So need not to call smp_call_function_single() in that case.
4369 if (vcpu
->guest_mode
&& vcpu
->cpu
!= cpu
)
4370 smp_call_function_single(ipi_pcpu
, vcpu_kick_intr
, vcpu
, 0);