2 * Kernel-based Virtual Machine driver for Linux
4 * derived from drivers/kvm/kvm_main.c
6 * Copyright (C) 2006 Qumranet, Inc.
9 * Avi Kivity <avi@qumranet.com>
10 * Yaniv Kamay <yaniv@qumranet.com>
12 * This work is licensed under the terms of the GNU GPL, version 2. See
13 * the COPYING file in the top-level directory.
19 #include "x86_emulate.h"
20 #include "segment_descriptor.h"
23 #include <linux/kvm.h>
25 #include <linux/vmalloc.h>
26 #include <linux/module.h>
27 #include <linux/mman.h>
29 #include <asm/uaccess.h>
32 #define MAX_IO_MSRS 256
33 #define CR0_RESERVED_BITS \
34 (~(unsigned long)(X86_CR0_PE | X86_CR0_MP | X86_CR0_EM | X86_CR0_TS \
35 | X86_CR0_ET | X86_CR0_NE | X86_CR0_WP | X86_CR0_AM \
36 | X86_CR0_NW | X86_CR0_CD | X86_CR0_PG))
37 #define CR4_RESERVED_BITS \
38 (~(unsigned long)(X86_CR4_VME | X86_CR4_PVI | X86_CR4_TSD | X86_CR4_DE\
39 | X86_CR4_PSE | X86_CR4_PAE | X86_CR4_MCE \
40 | X86_CR4_PGE | X86_CR4_PCE | X86_CR4_OSFXSR \
41 | X86_CR4_OSXMMEXCPT | X86_CR4_VMXE))
43 #define CR8_RESERVED_BITS (~(unsigned long)X86_CR8_TPR)
44 #define EFER_RESERVED_BITS 0xfffffffffffff2fe
46 #define VM_STAT(x) offsetof(struct kvm, stat.x), KVM_STAT_VM
47 #define VCPU_STAT(x) offsetof(struct kvm_vcpu, stat.x), KVM_STAT_VCPU
49 struct kvm_x86_ops
*kvm_x86_ops
;
51 struct kvm_stats_debugfs_item debugfs_entries
[] = {
52 { "pf_fixed", VCPU_STAT(pf_fixed
) },
53 { "pf_guest", VCPU_STAT(pf_guest
) },
54 { "tlb_flush", VCPU_STAT(tlb_flush
) },
55 { "invlpg", VCPU_STAT(invlpg
) },
56 { "exits", VCPU_STAT(exits
) },
57 { "io_exits", VCPU_STAT(io_exits
) },
58 { "mmio_exits", VCPU_STAT(mmio_exits
) },
59 { "signal_exits", VCPU_STAT(signal_exits
) },
60 { "irq_window", VCPU_STAT(irq_window_exits
) },
61 { "halt_exits", VCPU_STAT(halt_exits
) },
62 { "halt_wakeup", VCPU_STAT(halt_wakeup
) },
63 { "request_irq", VCPU_STAT(request_irq_exits
) },
64 { "irq_exits", VCPU_STAT(irq_exits
) },
65 { "host_state_reload", VCPU_STAT(host_state_reload
) },
66 { "efer_reload", VCPU_STAT(efer_reload
) },
67 { "fpu_reload", VCPU_STAT(fpu_reload
) },
68 { "insn_emulation", VCPU_STAT(insn_emulation
) },
69 { "insn_emulation_fail", VCPU_STAT(insn_emulation_fail
) },
70 { "mmu_shadow_zapped", VM_STAT(mmu_shadow_zapped
) },
71 { "mmu_pte_write", VM_STAT(mmu_pte_write
) },
72 { "mmu_pte_updated", VM_STAT(mmu_pte_updated
) },
73 { "mmu_pde_zapped", VM_STAT(mmu_pde_zapped
) },
74 { "mmu_flooded", VM_STAT(mmu_flooded
) },
75 { "mmu_recycled", VM_STAT(mmu_recycled
) },
76 { "remote_tlb_flush", VM_STAT(remote_tlb_flush
) },
81 unsigned long segment_base(u16 selector
)
83 struct descriptor_table gdt
;
84 struct segment_descriptor
*d
;
85 unsigned long table_base
;
91 asm("sgdt %0" : "=m"(gdt
));
92 table_base
= gdt
.base
;
94 if (selector
& 4) { /* from ldt */
97 asm("sldt %0" : "=g"(ldt_selector
));
98 table_base
= segment_base(ldt_selector
);
100 d
= (struct segment_descriptor
*)(table_base
+ (selector
& ~7));
101 v
= d
->base_low
| ((unsigned long)d
->base_mid
<< 16) |
102 ((unsigned long)d
->base_high
<< 24);
104 if (d
->system
== 0 && (d
->type
== 2 || d
->type
== 9 || d
->type
== 11))
105 v
|= ((unsigned long) \
106 ((struct segment_descriptor_64
*)d
)->base_higher
) << 32;
110 EXPORT_SYMBOL_GPL(segment_base
);
112 u64
kvm_get_apic_base(struct kvm_vcpu
*vcpu
)
114 if (irqchip_in_kernel(vcpu
->kvm
))
115 return vcpu
->apic_base
;
117 return vcpu
->apic_base
;
119 EXPORT_SYMBOL_GPL(kvm_get_apic_base
);
121 void kvm_set_apic_base(struct kvm_vcpu
*vcpu
, u64 data
)
123 /* TODO: reserve bits check */
124 if (irqchip_in_kernel(vcpu
->kvm
))
125 kvm_lapic_set_base(vcpu
, data
);
127 vcpu
->apic_base
= data
;
129 EXPORT_SYMBOL_GPL(kvm_set_apic_base
);
131 void kvm_queue_exception(struct kvm_vcpu
*vcpu
, unsigned nr
)
133 WARN_ON(vcpu
->exception
.pending
);
134 vcpu
->exception
.pending
= true;
135 vcpu
->exception
.has_error_code
= false;
136 vcpu
->exception
.nr
= nr
;
138 EXPORT_SYMBOL_GPL(kvm_queue_exception
);
140 void kvm_inject_page_fault(struct kvm_vcpu
*vcpu
, unsigned long addr
,
143 ++vcpu
->stat
.pf_guest
;
144 if (vcpu
->exception
.pending
&& vcpu
->exception
.nr
== PF_VECTOR
) {
145 printk(KERN_DEBUG
"kvm: inject_page_fault:"
146 " double fault 0x%lx\n", addr
);
147 vcpu
->exception
.nr
= DF_VECTOR
;
148 vcpu
->exception
.error_code
= 0;
152 kvm_queue_exception_e(vcpu
, PF_VECTOR
, error_code
);
155 void kvm_queue_exception_e(struct kvm_vcpu
*vcpu
, unsigned nr
, u32 error_code
)
157 WARN_ON(vcpu
->exception
.pending
);
158 vcpu
->exception
.pending
= true;
159 vcpu
->exception
.has_error_code
= true;
160 vcpu
->exception
.nr
= nr
;
161 vcpu
->exception
.error_code
= error_code
;
163 EXPORT_SYMBOL_GPL(kvm_queue_exception_e
);
165 static void __queue_exception(struct kvm_vcpu
*vcpu
)
167 kvm_x86_ops
->queue_exception(vcpu
, vcpu
->exception
.nr
,
168 vcpu
->exception
.has_error_code
,
169 vcpu
->exception
.error_code
);
173 * Load the pae pdptrs. Return true is they are all valid.
175 int load_pdptrs(struct kvm_vcpu
*vcpu
, unsigned long cr3
)
177 gfn_t pdpt_gfn
= cr3
>> PAGE_SHIFT
;
178 unsigned offset
= ((cr3
& (PAGE_SIZE
-1)) >> 5) << 2;
181 u64 pdpte
[ARRAY_SIZE(vcpu
->pdptrs
)];
183 mutex_lock(&vcpu
->kvm
->lock
);
184 ret
= kvm_read_guest_page(vcpu
->kvm
, pdpt_gfn
, pdpte
,
185 offset
* sizeof(u64
), sizeof(pdpte
));
190 for (i
= 0; i
< ARRAY_SIZE(pdpte
); ++i
) {
191 if ((pdpte
[i
] & 1) && (pdpte
[i
] & 0xfffffff0000001e6ull
)) {
198 memcpy(vcpu
->pdptrs
, pdpte
, sizeof(vcpu
->pdptrs
));
200 mutex_unlock(&vcpu
->kvm
->lock
);
205 static bool pdptrs_changed(struct kvm_vcpu
*vcpu
)
207 u64 pdpte
[ARRAY_SIZE(vcpu
->pdptrs
)];
211 if (is_long_mode(vcpu
) || !is_pae(vcpu
))
214 mutex_lock(&vcpu
->kvm
->lock
);
215 r
= kvm_read_guest(vcpu
->kvm
, vcpu
->cr3
& ~31u, pdpte
, sizeof(pdpte
));
218 changed
= memcmp(pdpte
, vcpu
->pdptrs
, sizeof(pdpte
)) != 0;
220 mutex_unlock(&vcpu
->kvm
->lock
);
225 void set_cr0(struct kvm_vcpu
*vcpu
, unsigned long cr0
)
227 if (cr0
& CR0_RESERVED_BITS
) {
228 printk(KERN_DEBUG
"set_cr0: 0x%lx #GP, reserved bits 0x%lx\n",
230 kvm_inject_gp(vcpu
, 0);
234 if ((cr0
& X86_CR0_NW
) && !(cr0
& X86_CR0_CD
)) {
235 printk(KERN_DEBUG
"set_cr0: #GP, CD == 0 && NW == 1\n");
236 kvm_inject_gp(vcpu
, 0);
240 if ((cr0
& X86_CR0_PG
) && !(cr0
& X86_CR0_PE
)) {
241 printk(KERN_DEBUG
"set_cr0: #GP, set PG flag "
242 "and a clear PE flag\n");
243 kvm_inject_gp(vcpu
, 0);
247 if (!is_paging(vcpu
) && (cr0
& X86_CR0_PG
)) {
249 if ((vcpu
->shadow_efer
& EFER_LME
)) {
253 printk(KERN_DEBUG
"set_cr0: #GP, start paging "
254 "in long mode while PAE is disabled\n");
255 kvm_inject_gp(vcpu
, 0);
258 kvm_x86_ops
->get_cs_db_l_bits(vcpu
, &cs_db
, &cs_l
);
260 printk(KERN_DEBUG
"set_cr0: #GP, start paging "
261 "in long mode while CS.L == 1\n");
262 kvm_inject_gp(vcpu
, 0);
268 if (is_pae(vcpu
) && !load_pdptrs(vcpu
, vcpu
->cr3
)) {
269 printk(KERN_DEBUG
"set_cr0: #GP, pdptrs "
271 kvm_inject_gp(vcpu
, 0);
277 kvm_x86_ops
->set_cr0(vcpu
, cr0
);
280 mutex_lock(&vcpu
->kvm
->lock
);
281 kvm_mmu_reset_context(vcpu
);
282 mutex_unlock(&vcpu
->kvm
->lock
);
285 EXPORT_SYMBOL_GPL(set_cr0
);
287 void lmsw(struct kvm_vcpu
*vcpu
, unsigned long msw
)
289 set_cr0(vcpu
, (vcpu
->cr0
& ~0x0ful
) | (msw
& 0x0f));
291 EXPORT_SYMBOL_GPL(lmsw
);
293 void set_cr4(struct kvm_vcpu
*vcpu
, unsigned long cr4
)
295 if (cr4
& CR4_RESERVED_BITS
) {
296 printk(KERN_DEBUG
"set_cr4: #GP, reserved bits\n");
297 kvm_inject_gp(vcpu
, 0);
301 if (is_long_mode(vcpu
)) {
302 if (!(cr4
& X86_CR4_PAE
)) {
303 printk(KERN_DEBUG
"set_cr4: #GP, clearing PAE while "
305 kvm_inject_gp(vcpu
, 0);
308 } else if (is_paging(vcpu
) && !is_pae(vcpu
) && (cr4
& X86_CR4_PAE
)
309 && !load_pdptrs(vcpu
, vcpu
->cr3
)) {
310 printk(KERN_DEBUG
"set_cr4: #GP, pdptrs reserved bits\n");
311 kvm_inject_gp(vcpu
, 0);
315 if (cr4
& X86_CR4_VMXE
) {
316 printk(KERN_DEBUG
"set_cr4: #GP, setting VMXE\n");
317 kvm_inject_gp(vcpu
, 0);
320 kvm_x86_ops
->set_cr4(vcpu
, cr4
);
322 mutex_lock(&vcpu
->kvm
->lock
);
323 kvm_mmu_reset_context(vcpu
);
324 mutex_unlock(&vcpu
->kvm
->lock
);
326 EXPORT_SYMBOL_GPL(set_cr4
);
328 void set_cr3(struct kvm_vcpu
*vcpu
, unsigned long cr3
)
330 if (cr3
== vcpu
->cr3
&& !pdptrs_changed(vcpu
)) {
331 kvm_mmu_flush_tlb(vcpu
);
335 if (is_long_mode(vcpu
)) {
336 if (cr3
& CR3_L_MODE_RESERVED_BITS
) {
337 printk(KERN_DEBUG
"set_cr3: #GP, reserved bits\n");
338 kvm_inject_gp(vcpu
, 0);
343 if (cr3
& CR3_PAE_RESERVED_BITS
) {
345 "set_cr3: #GP, reserved bits\n");
346 kvm_inject_gp(vcpu
, 0);
349 if (is_paging(vcpu
) && !load_pdptrs(vcpu
, cr3
)) {
350 printk(KERN_DEBUG
"set_cr3: #GP, pdptrs "
352 kvm_inject_gp(vcpu
, 0);
357 * We don't check reserved bits in nonpae mode, because
358 * this isn't enforced, and VMware depends on this.
362 mutex_lock(&vcpu
->kvm
->lock
);
364 * Does the new cr3 value map to physical memory? (Note, we
365 * catch an invalid cr3 even in real-mode, because it would
366 * cause trouble later on when we turn on paging anyway.)
368 * A real CPU would silently accept an invalid cr3 and would
369 * attempt to use it - with largely undefined (and often hard
370 * to debug) behavior on the guest side.
372 if (unlikely(!gfn_to_memslot(vcpu
->kvm
, cr3
>> PAGE_SHIFT
)))
373 kvm_inject_gp(vcpu
, 0);
376 vcpu
->mmu
.new_cr3(vcpu
);
378 mutex_unlock(&vcpu
->kvm
->lock
);
380 EXPORT_SYMBOL_GPL(set_cr3
);
382 void set_cr8(struct kvm_vcpu
*vcpu
, unsigned long cr8
)
384 if (cr8
& CR8_RESERVED_BITS
) {
385 printk(KERN_DEBUG
"set_cr8: #GP, reserved bits 0x%lx\n", cr8
);
386 kvm_inject_gp(vcpu
, 0);
389 if (irqchip_in_kernel(vcpu
->kvm
))
390 kvm_lapic_set_tpr(vcpu
, cr8
);
394 EXPORT_SYMBOL_GPL(set_cr8
);
396 unsigned long get_cr8(struct kvm_vcpu
*vcpu
)
398 if (irqchip_in_kernel(vcpu
->kvm
))
399 return kvm_lapic_get_cr8(vcpu
);
403 EXPORT_SYMBOL_GPL(get_cr8
);
406 * List of msr numbers which we expose to userspace through KVM_GET_MSRS
407 * and KVM_SET_MSRS, and KVM_GET_MSR_INDEX_LIST.
409 * This list is modified at module load time to reflect the
410 * capabilities of the host cpu.
412 static u32 msrs_to_save
[] = {
413 MSR_IA32_SYSENTER_CS
, MSR_IA32_SYSENTER_ESP
, MSR_IA32_SYSENTER_EIP
,
416 MSR_CSTAR
, MSR_KERNEL_GS_BASE
, MSR_SYSCALL_MASK
, MSR_LSTAR
,
418 MSR_IA32_TIME_STAMP_COUNTER
,
421 static unsigned num_msrs_to_save
;
423 static u32 emulated_msrs
[] = {
424 MSR_IA32_MISC_ENABLE
,
429 static void set_efer(struct kvm_vcpu
*vcpu
, u64 efer
)
431 if (efer
& EFER_RESERVED_BITS
) {
432 printk(KERN_DEBUG
"set_efer: 0x%llx #GP, reserved bits\n",
434 kvm_inject_gp(vcpu
, 0);
439 && (vcpu
->shadow_efer
& EFER_LME
) != (efer
& EFER_LME
)) {
440 printk(KERN_DEBUG
"set_efer: #GP, change LME while paging\n");
441 kvm_inject_gp(vcpu
, 0);
445 kvm_x86_ops
->set_efer(vcpu
, efer
);
448 efer
|= vcpu
->shadow_efer
& EFER_LMA
;
450 vcpu
->shadow_efer
= efer
;
456 * Writes msr value into into the appropriate "register".
457 * Returns 0 on success, non-0 otherwise.
458 * Assumes vcpu_load() was already called.
460 int kvm_set_msr(struct kvm_vcpu
*vcpu
, u32 msr_index
, u64 data
)
462 return kvm_x86_ops
->set_msr(vcpu
, msr_index
, data
);
466 * Adapt set_msr() to msr_io()'s calling convention
468 static int do_set_msr(struct kvm_vcpu
*vcpu
, unsigned index
, u64
*data
)
470 return kvm_set_msr(vcpu
, index
, *data
);
474 int kvm_set_msr_common(struct kvm_vcpu
*vcpu
, u32 msr
, u64 data
)
479 set_efer(vcpu
, data
);
482 case MSR_IA32_MC0_STATUS
:
483 pr_unimpl(vcpu
, "%s: MSR_IA32_MC0_STATUS 0x%llx, nop\n",
486 case MSR_IA32_MCG_STATUS
:
487 pr_unimpl(vcpu
, "%s: MSR_IA32_MCG_STATUS 0x%llx, nop\n",
490 case MSR_IA32_UCODE_REV
:
491 case MSR_IA32_UCODE_WRITE
:
492 case 0x200 ... 0x2ff: /* MTRRs */
494 case MSR_IA32_APICBASE
:
495 kvm_set_apic_base(vcpu
, data
);
497 case MSR_IA32_MISC_ENABLE
:
498 vcpu
->ia32_misc_enable_msr
= data
;
501 pr_unimpl(vcpu
, "unhandled wrmsr: 0x%x\n", msr
);
506 EXPORT_SYMBOL_GPL(kvm_set_msr_common
);
510 * Reads an msr value (of 'msr_index') into 'pdata'.
511 * Returns 0 on success, non-0 otherwise.
512 * Assumes vcpu_load() was already called.
514 int kvm_get_msr(struct kvm_vcpu
*vcpu
, u32 msr_index
, u64
*pdata
)
516 return kvm_x86_ops
->get_msr(vcpu
, msr_index
, pdata
);
519 int kvm_get_msr_common(struct kvm_vcpu
*vcpu
, u32 msr
, u64
*pdata
)
524 case 0xc0010010: /* SYSCFG */
525 case 0xc0010015: /* HWCR */
526 case MSR_IA32_PLATFORM_ID
:
527 case MSR_IA32_P5_MC_ADDR
:
528 case MSR_IA32_P5_MC_TYPE
:
529 case MSR_IA32_MC0_CTL
:
530 case MSR_IA32_MCG_STATUS
:
531 case MSR_IA32_MCG_CAP
:
532 case MSR_IA32_MC0_MISC
:
533 case MSR_IA32_MC0_MISC
+4:
534 case MSR_IA32_MC0_MISC
+8:
535 case MSR_IA32_MC0_MISC
+12:
536 case MSR_IA32_MC0_MISC
+16:
537 case MSR_IA32_UCODE_REV
:
538 case MSR_IA32_PERF_STATUS
:
539 case MSR_IA32_EBL_CR_POWERON
:
542 case 0x200 ... 0x2ff:
545 case 0xcd: /* fsb frequency */
548 case MSR_IA32_APICBASE
:
549 data
= kvm_get_apic_base(vcpu
);
551 case MSR_IA32_MISC_ENABLE
:
552 data
= vcpu
->ia32_misc_enable_msr
;
556 data
= vcpu
->shadow_efer
;
560 pr_unimpl(vcpu
, "unhandled rdmsr: 0x%x\n", msr
);
566 EXPORT_SYMBOL_GPL(kvm_get_msr_common
);
569 * Read or write a bunch of msrs. All parameters are kernel addresses.
571 * @return number of msrs set successfully.
573 static int __msr_io(struct kvm_vcpu
*vcpu
, struct kvm_msrs
*msrs
,
574 struct kvm_msr_entry
*entries
,
575 int (*do_msr
)(struct kvm_vcpu
*vcpu
,
576 unsigned index
, u64
*data
))
582 for (i
= 0; i
< msrs
->nmsrs
; ++i
)
583 if (do_msr(vcpu
, entries
[i
].index
, &entries
[i
].data
))
592 * Read or write a bunch of msrs. Parameters are user addresses.
594 * @return number of msrs set successfully.
596 static int msr_io(struct kvm_vcpu
*vcpu
, struct kvm_msrs __user
*user_msrs
,
597 int (*do_msr
)(struct kvm_vcpu
*vcpu
,
598 unsigned index
, u64
*data
),
601 struct kvm_msrs msrs
;
602 struct kvm_msr_entry
*entries
;
607 if (copy_from_user(&msrs
, user_msrs
, sizeof msrs
))
611 if (msrs
.nmsrs
>= MAX_IO_MSRS
)
615 size
= sizeof(struct kvm_msr_entry
) * msrs
.nmsrs
;
616 entries
= vmalloc(size
);
621 if (copy_from_user(entries
, user_msrs
->entries
, size
))
624 r
= n
= __msr_io(vcpu
, &msrs
, entries
, do_msr
);
629 if (writeback
&& copy_to_user(user_msrs
->entries
, entries
, size
))
641 * Make sure that a cpu that is being hot-unplugged does not have any vcpus
644 void decache_vcpus_on_cpu(int cpu
)
647 struct kvm_vcpu
*vcpu
;
650 spin_lock(&kvm_lock
);
651 list_for_each_entry(vm
, &vm_list
, vm_list
)
652 for (i
= 0; i
< KVM_MAX_VCPUS
; ++i
) {
657 * If the vcpu is locked, then it is running on some
658 * other cpu and therefore it is not cached on the
661 * If it's not locked, check the last cpu it executed
664 if (mutex_trylock(&vcpu
->mutex
)) {
665 if (vcpu
->cpu
== cpu
) {
666 kvm_x86_ops
->vcpu_decache(vcpu
);
669 mutex_unlock(&vcpu
->mutex
);
672 spin_unlock(&kvm_lock
);
675 int kvm_dev_ioctl_check_extension(long ext
)
680 case KVM_CAP_IRQCHIP
:
682 case KVM_CAP_MMU_SHADOW_CACHE_CONTROL
:
683 case KVM_CAP_USER_MEMORY
:
684 case KVM_CAP_SET_TSS_ADDR
:
685 case KVM_CAP_EXT_CPUID
:
696 long kvm_arch_dev_ioctl(struct file
*filp
,
697 unsigned int ioctl
, unsigned long arg
)
699 void __user
*argp
= (void __user
*)arg
;
703 case KVM_GET_MSR_INDEX_LIST
: {
704 struct kvm_msr_list __user
*user_msr_list
= argp
;
705 struct kvm_msr_list msr_list
;
709 if (copy_from_user(&msr_list
, user_msr_list
, sizeof msr_list
))
712 msr_list
.nmsrs
= num_msrs_to_save
+ ARRAY_SIZE(emulated_msrs
);
713 if (copy_to_user(user_msr_list
, &msr_list
, sizeof msr_list
))
716 if (n
< num_msrs_to_save
)
719 if (copy_to_user(user_msr_list
->indices
, &msrs_to_save
,
720 num_msrs_to_save
* sizeof(u32
)))
722 if (copy_to_user(user_msr_list
->indices
723 + num_msrs_to_save
* sizeof(u32
),
725 ARRAY_SIZE(emulated_msrs
) * sizeof(u32
)))
737 void kvm_arch_vcpu_load(struct kvm_vcpu
*vcpu
, int cpu
)
739 kvm_x86_ops
->vcpu_load(vcpu
, cpu
);
742 void kvm_arch_vcpu_put(struct kvm_vcpu
*vcpu
)
744 kvm_x86_ops
->vcpu_put(vcpu
);
745 kvm_put_guest_fpu(vcpu
);
748 static int is_efer_nx(void)
752 rdmsrl(MSR_EFER
, efer
);
753 return efer
& EFER_NX
;
756 static void cpuid_fix_nx_cap(struct kvm_vcpu
*vcpu
)
759 struct kvm_cpuid_entry2
*e
, *entry
;
762 for (i
= 0; i
< vcpu
->cpuid_nent
; ++i
) {
763 e
= &vcpu
->cpuid_entries
[i
];
764 if (e
->function
== 0x80000001) {
769 if (entry
&& (entry
->edx
& (1 << 20)) && !is_efer_nx()) {
770 entry
->edx
&= ~(1 << 20);
771 printk(KERN_INFO
"kvm: guest NX capability removed\n");
775 /* when an old userspace process fills a new kernel module */
776 static int kvm_vcpu_ioctl_set_cpuid(struct kvm_vcpu
*vcpu
,
777 struct kvm_cpuid
*cpuid
,
778 struct kvm_cpuid_entry __user
*entries
)
781 struct kvm_cpuid_entry
*cpuid_entries
;
784 if (cpuid
->nent
> KVM_MAX_CPUID_ENTRIES
)
787 cpuid_entries
= vmalloc(sizeof(struct kvm_cpuid_entry
) * cpuid
->nent
);
791 if (copy_from_user(cpuid_entries
, entries
,
792 cpuid
->nent
* sizeof(struct kvm_cpuid_entry
)))
794 for (i
= 0; i
< cpuid
->nent
; i
++) {
795 vcpu
->cpuid_entries
[i
].function
= cpuid_entries
[i
].function
;
796 vcpu
->cpuid_entries
[i
].eax
= cpuid_entries
[i
].eax
;
797 vcpu
->cpuid_entries
[i
].ebx
= cpuid_entries
[i
].ebx
;
798 vcpu
->cpuid_entries
[i
].ecx
= cpuid_entries
[i
].ecx
;
799 vcpu
->cpuid_entries
[i
].edx
= cpuid_entries
[i
].edx
;
800 vcpu
->cpuid_entries
[i
].index
= 0;
801 vcpu
->cpuid_entries
[i
].flags
= 0;
802 vcpu
->cpuid_entries
[i
].padding
[0] = 0;
803 vcpu
->cpuid_entries
[i
].padding
[1] = 0;
804 vcpu
->cpuid_entries
[i
].padding
[2] = 0;
806 vcpu
->cpuid_nent
= cpuid
->nent
;
807 cpuid_fix_nx_cap(vcpu
);
811 vfree(cpuid_entries
);
816 static int kvm_vcpu_ioctl_set_cpuid2(struct kvm_vcpu
*vcpu
,
817 struct kvm_cpuid2
*cpuid
,
818 struct kvm_cpuid_entry2 __user
*entries
)
823 if (cpuid
->nent
> KVM_MAX_CPUID_ENTRIES
)
826 if (copy_from_user(&vcpu
->cpuid_entries
, entries
,
827 cpuid
->nent
* sizeof(struct kvm_cpuid_entry2
)))
829 vcpu
->cpuid_nent
= cpuid
->nent
;
836 static int kvm_vcpu_ioctl_get_cpuid2(struct kvm_vcpu
*vcpu
,
837 struct kvm_cpuid2
*cpuid
,
838 struct kvm_cpuid_entry2 __user
*entries
)
843 if (cpuid
->nent
< vcpu
->cpuid_nent
)
846 if (copy_to_user(entries
, &vcpu
->cpuid_entries
,
847 vcpu
->cpuid_nent
* sizeof(struct kvm_cpuid_entry2
)))
852 cpuid
->nent
= vcpu
->cpuid_nent
;
856 static inline u32
bit(int bitno
)
858 return 1 << (bitno
& 31);
861 static void do_cpuid_1_ent(struct kvm_cpuid_entry2
*entry
, u32 function
,
864 entry
->function
= function
;
865 entry
->index
= index
;
866 cpuid_count(entry
->function
, entry
->index
,
867 &entry
->eax
, &entry
->ebx
, &entry
->ecx
, &entry
->edx
);
871 static void do_cpuid_ent(struct kvm_cpuid_entry2
*entry
, u32 function
,
872 u32 index
, int *nent
, int maxnent
)
874 const u32 kvm_supported_word0_x86_features
= bit(X86_FEATURE_FPU
) |
875 bit(X86_FEATURE_VME
) | bit(X86_FEATURE_DE
) |
876 bit(X86_FEATURE_PSE
) | bit(X86_FEATURE_TSC
) |
877 bit(X86_FEATURE_MSR
) | bit(X86_FEATURE_PAE
) |
878 bit(X86_FEATURE_CX8
) | bit(X86_FEATURE_APIC
) |
879 bit(X86_FEATURE_SEP
) | bit(X86_FEATURE_PGE
) |
880 bit(X86_FEATURE_CMOV
) | bit(X86_FEATURE_PSE36
) |
881 bit(X86_FEATURE_CLFLSH
) | bit(X86_FEATURE_MMX
) |
882 bit(X86_FEATURE_FXSR
) | bit(X86_FEATURE_XMM
) |
883 bit(X86_FEATURE_XMM2
) | bit(X86_FEATURE_SELFSNOOP
);
884 const u32 kvm_supported_word1_x86_features
= bit(X86_FEATURE_FPU
) |
885 bit(X86_FEATURE_VME
) | bit(X86_FEATURE_DE
) |
886 bit(X86_FEATURE_PSE
) | bit(X86_FEATURE_TSC
) |
887 bit(X86_FEATURE_MSR
) | bit(X86_FEATURE_PAE
) |
888 bit(X86_FEATURE_CX8
) | bit(X86_FEATURE_APIC
) |
889 bit(X86_FEATURE_PGE
) |
890 bit(X86_FEATURE_CMOV
) | bit(X86_FEATURE_PSE36
) |
891 bit(X86_FEATURE_MMX
) | bit(X86_FEATURE_FXSR
) |
892 bit(X86_FEATURE_SYSCALL
) |
893 (bit(X86_FEATURE_NX
) && is_efer_nx()) |
895 bit(X86_FEATURE_LM
) |
897 bit(X86_FEATURE_MMXEXT
) |
898 bit(X86_FEATURE_3DNOWEXT
) |
899 bit(X86_FEATURE_3DNOW
);
900 const u32 kvm_supported_word3_x86_features
=
901 bit(X86_FEATURE_XMM3
) | bit(X86_FEATURE_CX16
);
902 const u32 kvm_supported_word6_x86_features
=
903 bit(X86_FEATURE_LAHF_LM
) | bit(X86_FEATURE_CMP_LEGACY
);
905 /* all func 2 cpuid_count() should be called on the same cpu */
907 do_cpuid_1_ent(entry
, function
, index
);
912 entry
->eax
= min(entry
->eax
, (u32
)0xb);
915 entry
->edx
&= kvm_supported_word0_x86_features
;
916 entry
->ecx
&= kvm_supported_word3_x86_features
;
918 /* function 2 entries are STATEFUL. That is, repeated cpuid commands
919 * may return different values. This forces us to get_cpu() before
920 * issuing the first command, and also to emulate this annoying behavior
921 * in kvm_emulate_cpuid() using KVM_CPUID_FLAG_STATE_READ_NEXT */
923 int t
, times
= entry
->eax
& 0xff;
925 entry
->flags
|= KVM_CPUID_FLAG_STATEFUL_FUNC
;
926 for (t
= 1; t
< times
&& *nent
< maxnent
; ++t
) {
927 do_cpuid_1_ent(&entry
[t
], function
, 0);
928 entry
[t
].flags
|= KVM_CPUID_FLAG_STATEFUL_FUNC
;
933 /* function 4 and 0xb have additional index. */
935 int index
, cache_type
;
937 entry
->flags
|= KVM_CPUID_FLAG_SIGNIFCANT_INDEX
;
938 /* read more entries until cache_type is zero */
939 for (index
= 1; *nent
< maxnent
; ++index
) {
940 cache_type
= entry
[index
- 1].eax
& 0x1f;
943 do_cpuid_1_ent(&entry
[index
], function
, index
);
944 entry
[index
].flags
|=
945 KVM_CPUID_FLAG_SIGNIFCANT_INDEX
;
951 int index
, level_type
;
953 entry
->flags
|= KVM_CPUID_FLAG_SIGNIFCANT_INDEX
;
954 /* read more entries until level_type is zero */
955 for (index
= 1; *nent
< maxnent
; ++index
) {
956 level_type
= entry
[index
- 1].ecx
& 0xff;
959 do_cpuid_1_ent(&entry
[index
], function
, index
);
960 entry
[index
].flags
|=
961 KVM_CPUID_FLAG_SIGNIFCANT_INDEX
;
967 entry
->eax
= min(entry
->eax
, 0x8000001a);
970 entry
->edx
&= kvm_supported_word1_x86_features
;
971 entry
->ecx
&= kvm_supported_word6_x86_features
;
977 static int kvm_vm_ioctl_get_supported_cpuid(struct kvm
*kvm
,
978 struct kvm_cpuid2
*cpuid
,
979 struct kvm_cpuid_entry2 __user
*entries
)
981 struct kvm_cpuid_entry2
*cpuid_entries
;
982 int limit
, nent
= 0, r
= -E2BIG
;
988 cpuid_entries
= vmalloc(sizeof(struct kvm_cpuid_entry2
) * cpuid
->nent
);
992 do_cpuid_ent(&cpuid_entries
[0], 0, 0, &nent
, cpuid
->nent
);
993 limit
= cpuid_entries
[0].eax
;
994 for (func
= 1; func
<= limit
&& nent
< cpuid
->nent
; ++func
)
995 do_cpuid_ent(&cpuid_entries
[nent
], func
, 0,
998 if (nent
>= cpuid
->nent
)
1001 do_cpuid_ent(&cpuid_entries
[nent
], 0x80000000, 0, &nent
, cpuid
->nent
);
1002 limit
= cpuid_entries
[nent
- 1].eax
;
1003 for (func
= 0x80000001; func
<= limit
&& nent
< cpuid
->nent
; ++func
)
1004 do_cpuid_ent(&cpuid_entries
[nent
], func
, 0,
1005 &nent
, cpuid
->nent
);
1007 if (copy_to_user(entries
, cpuid_entries
,
1008 nent
* sizeof(struct kvm_cpuid_entry2
)))
1014 vfree(cpuid_entries
);
1019 static int kvm_vcpu_ioctl_get_lapic(struct kvm_vcpu
*vcpu
,
1020 struct kvm_lapic_state
*s
)
1023 memcpy(s
->regs
, vcpu
->apic
->regs
, sizeof *s
);
1029 static int kvm_vcpu_ioctl_set_lapic(struct kvm_vcpu
*vcpu
,
1030 struct kvm_lapic_state
*s
)
1033 memcpy(vcpu
->apic
->regs
, s
->regs
, sizeof *s
);
1034 kvm_apic_post_state_restore(vcpu
);
1040 static int kvm_vcpu_ioctl_interrupt(struct kvm_vcpu
*vcpu
,
1041 struct kvm_interrupt
*irq
)
1043 if (irq
->irq
< 0 || irq
->irq
>= 256)
1045 if (irqchip_in_kernel(vcpu
->kvm
))
1049 set_bit(irq
->irq
, vcpu
->irq_pending
);
1050 set_bit(irq
->irq
/ BITS_PER_LONG
, &vcpu
->irq_summary
);
1057 long kvm_arch_vcpu_ioctl(struct file
*filp
,
1058 unsigned int ioctl
, unsigned long arg
)
1060 struct kvm_vcpu
*vcpu
= filp
->private_data
;
1061 void __user
*argp
= (void __user
*)arg
;
1065 case KVM_GET_LAPIC
: {
1066 struct kvm_lapic_state lapic
;
1068 memset(&lapic
, 0, sizeof lapic
);
1069 r
= kvm_vcpu_ioctl_get_lapic(vcpu
, &lapic
);
1073 if (copy_to_user(argp
, &lapic
, sizeof lapic
))
1078 case KVM_SET_LAPIC
: {
1079 struct kvm_lapic_state lapic
;
1082 if (copy_from_user(&lapic
, argp
, sizeof lapic
))
1084 r
= kvm_vcpu_ioctl_set_lapic(vcpu
, &lapic
);;
1090 case KVM_INTERRUPT
: {
1091 struct kvm_interrupt irq
;
1094 if (copy_from_user(&irq
, argp
, sizeof irq
))
1096 r
= kvm_vcpu_ioctl_interrupt(vcpu
, &irq
);
1102 case KVM_SET_CPUID
: {
1103 struct kvm_cpuid __user
*cpuid_arg
= argp
;
1104 struct kvm_cpuid cpuid
;
1107 if (copy_from_user(&cpuid
, cpuid_arg
, sizeof cpuid
))
1109 r
= kvm_vcpu_ioctl_set_cpuid(vcpu
, &cpuid
, cpuid_arg
->entries
);
1114 case KVM_SET_CPUID2
: {
1115 struct kvm_cpuid2 __user
*cpuid_arg
= argp
;
1116 struct kvm_cpuid2 cpuid
;
1119 if (copy_from_user(&cpuid
, cpuid_arg
, sizeof cpuid
))
1121 r
= kvm_vcpu_ioctl_set_cpuid2(vcpu
, &cpuid
,
1122 cpuid_arg
->entries
);
1127 case KVM_GET_CPUID2
: {
1128 struct kvm_cpuid2 __user
*cpuid_arg
= argp
;
1129 struct kvm_cpuid2 cpuid
;
1132 if (copy_from_user(&cpuid
, cpuid_arg
, sizeof cpuid
))
1134 r
= kvm_vcpu_ioctl_get_cpuid2(vcpu
, &cpuid
,
1135 cpuid_arg
->entries
);
1139 if (copy_to_user(cpuid_arg
, &cpuid
, sizeof cpuid
))
1145 r
= msr_io(vcpu
, argp
, kvm_get_msr
, 1);
1148 r
= msr_io(vcpu
, argp
, do_set_msr
, 0);
1157 static int kvm_vm_ioctl_set_tss_addr(struct kvm
*kvm
, unsigned long addr
)
1161 if (addr
> (unsigned int)(-3 * PAGE_SIZE
))
1163 ret
= kvm_x86_ops
->set_tss_addr(kvm
, addr
);
1167 static int kvm_vm_ioctl_set_nr_mmu_pages(struct kvm
*kvm
,
1168 u32 kvm_nr_mmu_pages
)
1170 if (kvm_nr_mmu_pages
< KVM_MIN_ALLOC_MMU_PAGES
)
1173 mutex_lock(&kvm
->lock
);
1175 kvm_mmu_change_mmu_pages(kvm
, kvm_nr_mmu_pages
);
1176 kvm
->n_requested_mmu_pages
= kvm_nr_mmu_pages
;
1178 mutex_unlock(&kvm
->lock
);
1182 static int kvm_vm_ioctl_get_nr_mmu_pages(struct kvm
*kvm
)
1184 return kvm
->n_alloc_mmu_pages
;
1187 gfn_t
unalias_gfn(struct kvm
*kvm
, gfn_t gfn
)
1190 struct kvm_mem_alias
*alias
;
1192 for (i
= 0; i
< kvm
->naliases
; ++i
) {
1193 alias
= &kvm
->aliases
[i
];
1194 if (gfn
>= alias
->base_gfn
1195 && gfn
< alias
->base_gfn
+ alias
->npages
)
1196 return alias
->target_gfn
+ gfn
- alias
->base_gfn
;
1202 * Set a new alias region. Aliases map a portion of physical memory into
1203 * another portion. This is useful for memory windows, for example the PC
1206 static int kvm_vm_ioctl_set_memory_alias(struct kvm
*kvm
,
1207 struct kvm_memory_alias
*alias
)
1210 struct kvm_mem_alias
*p
;
1213 /* General sanity checks */
1214 if (alias
->memory_size
& (PAGE_SIZE
- 1))
1216 if (alias
->guest_phys_addr
& (PAGE_SIZE
- 1))
1218 if (alias
->slot
>= KVM_ALIAS_SLOTS
)
1220 if (alias
->guest_phys_addr
+ alias
->memory_size
1221 < alias
->guest_phys_addr
)
1223 if (alias
->target_phys_addr
+ alias
->memory_size
1224 < alias
->target_phys_addr
)
1227 mutex_lock(&kvm
->lock
);
1229 p
= &kvm
->aliases
[alias
->slot
];
1230 p
->base_gfn
= alias
->guest_phys_addr
>> PAGE_SHIFT
;
1231 p
->npages
= alias
->memory_size
>> PAGE_SHIFT
;
1232 p
->target_gfn
= alias
->target_phys_addr
>> PAGE_SHIFT
;
1234 for (n
= KVM_ALIAS_SLOTS
; n
> 0; --n
)
1235 if (kvm
->aliases
[n
- 1].npages
)
1239 kvm_mmu_zap_all(kvm
);
1241 mutex_unlock(&kvm
->lock
);
1249 static int kvm_vm_ioctl_get_irqchip(struct kvm
*kvm
, struct kvm_irqchip
*chip
)
1254 switch (chip
->chip_id
) {
1255 case KVM_IRQCHIP_PIC_MASTER
:
1256 memcpy(&chip
->chip
.pic
,
1257 &pic_irqchip(kvm
)->pics
[0],
1258 sizeof(struct kvm_pic_state
));
1260 case KVM_IRQCHIP_PIC_SLAVE
:
1261 memcpy(&chip
->chip
.pic
,
1262 &pic_irqchip(kvm
)->pics
[1],
1263 sizeof(struct kvm_pic_state
));
1265 case KVM_IRQCHIP_IOAPIC
:
1266 memcpy(&chip
->chip
.ioapic
,
1267 ioapic_irqchip(kvm
),
1268 sizeof(struct kvm_ioapic_state
));
1277 static int kvm_vm_ioctl_set_irqchip(struct kvm
*kvm
, struct kvm_irqchip
*chip
)
1282 switch (chip
->chip_id
) {
1283 case KVM_IRQCHIP_PIC_MASTER
:
1284 memcpy(&pic_irqchip(kvm
)->pics
[0],
1286 sizeof(struct kvm_pic_state
));
1288 case KVM_IRQCHIP_PIC_SLAVE
:
1289 memcpy(&pic_irqchip(kvm
)->pics
[1],
1291 sizeof(struct kvm_pic_state
));
1293 case KVM_IRQCHIP_IOAPIC
:
1294 memcpy(ioapic_irqchip(kvm
),
1296 sizeof(struct kvm_ioapic_state
));
1302 kvm_pic_update_irq(pic_irqchip(kvm
));
1307 * Get (and clear) the dirty memory log for a memory slot.
1309 int kvm_vm_ioctl_get_dirty_log(struct kvm
*kvm
,
1310 struct kvm_dirty_log
*log
)
1314 struct kvm_memory_slot
*memslot
;
1317 mutex_lock(&kvm
->lock
);
1319 r
= kvm_get_dirty_log(kvm
, log
, &is_dirty
);
1323 /* If nothing is dirty, don't bother messing with page tables. */
1325 kvm_mmu_slot_remove_write_access(kvm
, log
->slot
);
1326 kvm_flush_remote_tlbs(kvm
);
1327 memslot
= &kvm
->memslots
[log
->slot
];
1328 n
= ALIGN(memslot
->npages
, BITS_PER_LONG
) / 8;
1329 memset(memslot
->dirty_bitmap
, 0, n
);
1333 mutex_unlock(&kvm
->lock
);
1337 long kvm_arch_vm_ioctl(struct file
*filp
,
1338 unsigned int ioctl
, unsigned long arg
)
1340 struct kvm
*kvm
= filp
->private_data
;
1341 void __user
*argp
= (void __user
*)arg
;
1345 case KVM_SET_TSS_ADDR
:
1346 r
= kvm_vm_ioctl_set_tss_addr(kvm
, arg
);
1350 case KVM_SET_MEMORY_REGION
: {
1351 struct kvm_memory_region kvm_mem
;
1352 struct kvm_userspace_memory_region kvm_userspace_mem
;
1355 if (copy_from_user(&kvm_mem
, argp
, sizeof kvm_mem
))
1357 kvm_userspace_mem
.slot
= kvm_mem
.slot
;
1358 kvm_userspace_mem
.flags
= kvm_mem
.flags
;
1359 kvm_userspace_mem
.guest_phys_addr
= kvm_mem
.guest_phys_addr
;
1360 kvm_userspace_mem
.memory_size
= kvm_mem
.memory_size
;
1361 r
= kvm_vm_ioctl_set_memory_region(kvm
, &kvm_userspace_mem
, 0);
1366 case KVM_SET_NR_MMU_PAGES
:
1367 r
= kvm_vm_ioctl_set_nr_mmu_pages(kvm
, arg
);
1371 case KVM_GET_NR_MMU_PAGES
:
1372 r
= kvm_vm_ioctl_get_nr_mmu_pages(kvm
);
1374 case KVM_SET_MEMORY_ALIAS
: {
1375 struct kvm_memory_alias alias
;
1378 if (copy_from_user(&alias
, argp
, sizeof alias
))
1380 r
= kvm_vm_ioctl_set_memory_alias(kvm
, &alias
);
1385 case KVM_CREATE_IRQCHIP
:
1387 kvm
->vpic
= kvm_create_pic(kvm
);
1389 r
= kvm_ioapic_init(kvm
);
1398 case KVM_IRQ_LINE
: {
1399 struct kvm_irq_level irq_event
;
1402 if (copy_from_user(&irq_event
, argp
, sizeof irq_event
))
1404 if (irqchip_in_kernel(kvm
)) {
1405 mutex_lock(&kvm
->lock
);
1406 if (irq_event
.irq
< 16)
1407 kvm_pic_set_irq(pic_irqchip(kvm
),
1410 kvm_ioapic_set_irq(kvm
->vioapic
,
1413 mutex_unlock(&kvm
->lock
);
1418 case KVM_GET_IRQCHIP
: {
1419 /* 0: PIC master, 1: PIC slave, 2: IOAPIC */
1420 struct kvm_irqchip chip
;
1423 if (copy_from_user(&chip
, argp
, sizeof chip
))
1426 if (!irqchip_in_kernel(kvm
))
1428 r
= kvm_vm_ioctl_get_irqchip(kvm
, &chip
);
1432 if (copy_to_user(argp
, &chip
, sizeof chip
))
1437 case KVM_SET_IRQCHIP
: {
1438 /* 0: PIC master, 1: PIC slave, 2: IOAPIC */
1439 struct kvm_irqchip chip
;
1442 if (copy_from_user(&chip
, argp
, sizeof chip
))
1445 if (!irqchip_in_kernel(kvm
))
1447 r
= kvm_vm_ioctl_set_irqchip(kvm
, &chip
);
1453 case KVM_GET_SUPPORTED_CPUID
: {
1454 struct kvm_cpuid2 __user
*cpuid_arg
= argp
;
1455 struct kvm_cpuid2 cpuid
;
1458 if (copy_from_user(&cpuid
, cpuid_arg
, sizeof cpuid
))
1460 r
= kvm_vm_ioctl_get_supported_cpuid(kvm
, &cpuid
,
1461 cpuid_arg
->entries
);
1466 if (copy_to_user(cpuid_arg
, &cpuid
, sizeof cpuid
))
1478 static void kvm_init_msr_list(void)
1483 for (i
= j
= 0; i
< ARRAY_SIZE(msrs_to_save
); i
++) {
1484 if (rdmsr_safe(msrs_to_save
[i
], &dummy
[0], &dummy
[1]) < 0)
1487 msrs_to_save
[j
] = msrs_to_save
[i
];
1490 num_msrs_to_save
= j
;
1494 * Only apic need an MMIO device hook, so shortcut now..
1496 static struct kvm_io_device
*vcpu_find_pervcpu_dev(struct kvm_vcpu
*vcpu
,
1499 struct kvm_io_device
*dev
;
1502 dev
= &vcpu
->apic
->dev
;
1503 if (dev
->in_range(dev
, addr
))
1510 static struct kvm_io_device
*vcpu_find_mmio_dev(struct kvm_vcpu
*vcpu
,
1513 struct kvm_io_device
*dev
;
1515 dev
= vcpu_find_pervcpu_dev(vcpu
, addr
);
1517 dev
= kvm_io_bus_find_dev(&vcpu
->kvm
->mmio_bus
, addr
);
1521 int emulator_read_std(unsigned long addr
,
1524 struct kvm_vcpu
*vcpu
)
1529 gpa_t gpa
= vcpu
->mmu
.gva_to_gpa(vcpu
, addr
);
1530 unsigned offset
= addr
& (PAGE_SIZE
-1);
1531 unsigned tocopy
= min(bytes
, (unsigned)PAGE_SIZE
- offset
);
1534 if (gpa
== UNMAPPED_GVA
)
1535 return X86EMUL_PROPAGATE_FAULT
;
1536 ret
= kvm_read_guest(vcpu
->kvm
, gpa
, data
, tocopy
);
1538 return X86EMUL_UNHANDLEABLE
;
1545 return X86EMUL_CONTINUE
;
1547 EXPORT_SYMBOL_GPL(emulator_read_std
);
1549 static int emulator_read_emulated(unsigned long addr
,
1552 struct kvm_vcpu
*vcpu
)
1554 struct kvm_io_device
*mmio_dev
;
1557 if (vcpu
->mmio_read_completed
) {
1558 memcpy(val
, vcpu
->mmio_data
, bytes
);
1559 vcpu
->mmio_read_completed
= 0;
1560 return X86EMUL_CONTINUE
;
1563 gpa
= vcpu
->mmu
.gva_to_gpa(vcpu
, addr
);
1565 /* For APIC access vmexit */
1566 if ((gpa
& PAGE_MASK
) == APIC_DEFAULT_PHYS_BASE
)
1569 if (emulator_read_std(addr
, val
, bytes
, vcpu
)
1570 == X86EMUL_CONTINUE
)
1571 return X86EMUL_CONTINUE
;
1572 if (gpa
== UNMAPPED_GVA
)
1573 return X86EMUL_PROPAGATE_FAULT
;
1577 * Is this MMIO handled locally?
1579 mmio_dev
= vcpu_find_mmio_dev(vcpu
, gpa
);
1581 kvm_iodevice_read(mmio_dev
, gpa
, bytes
, val
);
1582 return X86EMUL_CONTINUE
;
1585 vcpu
->mmio_needed
= 1;
1586 vcpu
->mmio_phys_addr
= gpa
;
1587 vcpu
->mmio_size
= bytes
;
1588 vcpu
->mmio_is_write
= 0;
1590 return X86EMUL_UNHANDLEABLE
;
1593 static int emulator_write_phys(struct kvm_vcpu
*vcpu
, gpa_t gpa
,
1594 const void *val
, int bytes
)
1598 ret
= kvm_write_guest(vcpu
->kvm
, gpa
, val
, bytes
);
1601 kvm_mmu_pte_write(vcpu
, gpa
, val
, bytes
);
1605 static int emulator_write_emulated_onepage(unsigned long addr
,
1608 struct kvm_vcpu
*vcpu
)
1610 struct kvm_io_device
*mmio_dev
;
1611 gpa_t gpa
= vcpu
->mmu
.gva_to_gpa(vcpu
, addr
);
1613 if (gpa
== UNMAPPED_GVA
) {
1614 kvm_inject_page_fault(vcpu
, addr
, 2);
1615 return X86EMUL_PROPAGATE_FAULT
;
1618 /* For APIC access vmexit */
1619 if ((gpa
& PAGE_MASK
) == APIC_DEFAULT_PHYS_BASE
)
1622 if (emulator_write_phys(vcpu
, gpa
, val
, bytes
))
1623 return X86EMUL_CONTINUE
;
1627 * Is this MMIO handled locally?
1629 mmio_dev
= vcpu_find_mmio_dev(vcpu
, gpa
);
1631 kvm_iodevice_write(mmio_dev
, gpa
, bytes
, val
);
1632 return X86EMUL_CONTINUE
;
1635 vcpu
->mmio_needed
= 1;
1636 vcpu
->mmio_phys_addr
= gpa
;
1637 vcpu
->mmio_size
= bytes
;
1638 vcpu
->mmio_is_write
= 1;
1639 memcpy(vcpu
->mmio_data
, val
, bytes
);
1641 return X86EMUL_CONTINUE
;
1644 int emulator_write_emulated(unsigned long addr
,
1647 struct kvm_vcpu
*vcpu
)
1649 /* Crossing a page boundary? */
1650 if (((addr
+ bytes
- 1) ^ addr
) & PAGE_MASK
) {
1653 now
= -addr
& ~PAGE_MASK
;
1654 rc
= emulator_write_emulated_onepage(addr
, val
, now
, vcpu
);
1655 if (rc
!= X86EMUL_CONTINUE
)
1661 return emulator_write_emulated_onepage(addr
, val
, bytes
, vcpu
);
1663 EXPORT_SYMBOL_GPL(emulator_write_emulated
);
1665 static int emulator_cmpxchg_emulated(unsigned long addr
,
1669 struct kvm_vcpu
*vcpu
)
1671 static int reported
;
1675 printk(KERN_WARNING
"kvm: emulating exchange as write\n");
1677 return emulator_write_emulated(addr
, new, bytes
, vcpu
);
1680 static unsigned long get_segment_base(struct kvm_vcpu
*vcpu
, int seg
)
1682 return kvm_x86_ops
->get_segment_base(vcpu
, seg
);
1685 int emulate_invlpg(struct kvm_vcpu
*vcpu
, gva_t address
)
1687 return X86EMUL_CONTINUE
;
1690 int emulate_clts(struct kvm_vcpu
*vcpu
)
1692 kvm_x86_ops
->set_cr0(vcpu
, vcpu
->cr0
& ~X86_CR0_TS
);
1693 return X86EMUL_CONTINUE
;
1696 int emulator_get_dr(struct x86_emulate_ctxt
*ctxt
, int dr
, unsigned long *dest
)
1698 struct kvm_vcpu
*vcpu
= ctxt
->vcpu
;
1702 *dest
= kvm_x86_ops
->get_dr(vcpu
, dr
);
1703 return X86EMUL_CONTINUE
;
1705 pr_unimpl(vcpu
, "%s: unexpected dr %u\n", __FUNCTION__
, dr
);
1706 return X86EMUL_UNHANDLEABLE
;
1710 int emulator_set_dr(struct x86_emulate_ctxt
*ctxt
, int dr
, unsigned long value
)
1712 unsigned long mask
= (ctxt
->mode
== X86EMUL_MODE_PROT64
) ? ~0ULL : ~0U;
1715 kvm_x86_ops
->set_dr(ctxt
->vcpu
, dr
, value
& mask
, &exception
);
1717 /* FIXME: better handling */
1718 return X86EMUL_UNHANDLEABLE
;
1720 return X86EMUL_CONTINUE
;
1723 void kvm_report_emulation_failure(struct kvm_vcpu
*vcpu
, const char *context
)
1725 static int reported
;
1727 unsigned long rip
= vcpu
->rip
;
1728 unsigned long rip_linear
;
1730 rip_linear
= rip
+ get_segment_base(vcpu
, VCPU_SREG_CS
);
1735 emulator_read_std(rip_linear
, (void *)opcodes
, 4, vcpu
);
1737 printk(KERN_ERR
"emulation failed (%s) rip %lx %02x %02x %02x %02x\n",
1738 context
, rip
, opcodes
[0], opcodes
[1], opcodes
[2], opcodes
[3]);
1741 EXPORT_SYMBOL_GPL(kvm_report_emulation_failure
);
1743 struct x86_emulate_ops emulate_ops
= {
1744 .read_std
= emulator_read_std
,
1745 .read_emulated
= emulator_read_emulated
,
1746 .write_emulated
= emulator_write_emulated
,
1747 .cmpxchg_emulated
= emulator_cmpxchg_emulated
,
1750 int emulate_instruction(struct kvm_vcpu
*vcpu
,
1751 struct kvm_run
*run
,
1758 vcpu
->mmio_fault_cr2
= cr2
;
1759 kvm_x86_ops
->cache_regs(vcpu
);
1761 vcpu
->mmio_is_write
= 0;
1762 vcpu
->pio
.string
= 0;
1766 kvm_x86_ops
->get_cs_db_l_bits(vcpu
, &cs_db
, &cs_l
);
1768 vcpu
->emulate_ctxt
.vcpu
= vcpu
;
1769 vcpu
->emulate_ctxt
.eflags
= kvm_x86_ops
->get_rflags(vcpu
);
1770 vcpu
->emulate_ctxt
.mode
=
1771 (vcpu
->emulate_ctxt
.eflags
& X86_EFLAGS_VM
)
1772 ? X86EMUL_MODE_REAL
: cs_l
1773 ? X86EMUL_MODE_PROT64
: cs_db
1774 ? X86EMUL_MODE_PROT32
: X86EMUL_MODE_PROT16
;
1776 if (vcpu
->emulate_ctxt
.mode
== X86EMUL_MODE_PROT64
) {
1777 vcpu
->emulate_ctxt
.cs_base
= 0;
1778 vcpu
->emulate_ctxt
.ds_base
= 0;
1779 vcpu
->emulate_ctxt
.es_base
= 0;
1780 vcpu
->emulate_ctxt
.ss_base
= 0;
1782 vcpu
->emulate_ctxt
.cs_base
=
1783 get_segment_base(vcpu
, VCPU_SREG_CS
);
1784 vcpu
->emulate_ctxt
.ds_base
=
1785 get_segment_base(vcpu
, VCPU_SREG_DS
);
1786 vcpu
->emulate_ctxt
.es_base
=
1787 get_segment_base(vcpu
, VCPU_SREG_ES
);
1788 vcpu
->emulate_ctxt
.ss_base
=
1789 get_segment_base(vcpu
, VCPU_SREG_SS
);
1792 vcpu
->emulate_ctxt
.gs_base
=
1793 get_segment_base(vcpu
, VCPU_SREG_GS
);
1794 vcpu
->emulate_ctxt
.fs_base
=
1795 get_segment_base(vcpu
, VCPU_SREG_FS
);
1797 r
= x86_decode_insn(&vcpu
->emulate_ctxt
, &emulate_ops
);
1798 ++vcpu
->stat
.insn_emulation
;
1800 ++vcpu
->stat
.insn_emulation_fail
;
1801 if (kvm_mmu_unprotect_page_virt(vcpu
, cr2
))
1802 return EMULATE_DONE
;
1803 return EMULATE_FAIL
;
1807 r
= x86_emulate_insn(&vcpu
->emulate_ctxt
, &emulate_ops
);
1809 if (vcpu
->pio
.string
)
1810 return EMULATE_DO_MMIO
;
1812 if ((r
|| vcpu
->mmio_is_write
) && run
) {
1813 run
->exit_reason
= KVM_EXIT_MMIO
;
1814 run
->mmio
.phys_addr
= vcpu
->mmio_phys_addr
;
1815 memcpy(run
->mmio
.data
, vcpu
->mmio_data
, 8);
1816 run
->mmio
.len
= vcpu
->mmio_size
;
1817 run
->mmio
.is_write
= vcpu
->mmio_is_write
;
1821 if (kvm_mmu_unprotect_page_virt(vcpu
, cr2
))
1822 return EMULATE_DONE
;
1823 if (!vcpu
->mmio_needed
) {
1824 kvm_report_emulation_failure(vcpu
, "mmio");
1825 return EMULATE_FAIL
;
1827 return EMULATE_DO_MMIO
;
1830 kvm_x86_ops
->decache_regs(vcpu
);
1831 kvm_x86_ops
->set_rflags(vcpu
, vcpu
->emulate_ctxt
.eflags
);
1833 if (vcpu
->mmio_is_write
) {
1834 vcpu
->mmio_needed
= 0;
1835 return EMULATE_DO_MMIO
;
1838 return EMULATE_DONE
;
1840 EXPORT_SYMBOL_GPL(emulate_instruction
);
1842 static void free_pio_guest_pages(struct kvm_vcpu
*vcpu
)
1846 for (i
= 0; i
< ARRAY_SIZE(vcpu
->pio
.guest_pages
); ++i
)
1847 if (vcpu
->pio
.guest_pages
[i
]) {
1848 kvm_release_page_dirty(vcpu
->pio
.guest_pages
[i
]);
1849 vcpu
->pio
.guest_pages
[i
] = NULL
;
1853 static int pio_copy_data(struct kvm_vcpu
*vcpu
)
1855 void *p
= vcpu
->pio_data
;
1858 int nr_pages
= vcpu
->pio
.guest_pages
[1] ? 2 : 1;
1860 q
= vmap(vcpu
->pio
.guest_pages
, nr_pages
, VM_READ
|VM_WRITE
,
1863 free_pio_guest_pages(vcpu
);
1866 q
+= vcpu
->pio
.guest_page_offset
;
1867 bytes
= vcpu
->pio
.size
* vcpu
->pio
.cur_count
;
1869 memcpy(q
, p
, bytes
);
1871 memcpy(p
, q
, bytes
);
1872 q
-= vcpu
->pio
.guest_page_offset
;
1874 free_pio_guest_pages(vcpu
);
1878 int complete_pio(struct kvm_vcpu
*vcpu
)
1880 struct kvm_pio_request
*io
= &vcpu
->pio
;
1884 kvm_x86_ops
->cache_regs(vcpu
);
1888 memcpy(&vcpu
->regs
[VCPU_REGS_RAX
], vcpu
->pio_data
,
1892 r
= pio_copy_data(vcpu
);
1894 kvm_x86_ops
->cache_regs(vcpu
);
1901 delta
*= io
->cur_count
;
1903 * The size of the register should really depend on
1904 * current address size.
1906 vcpu
->regs
[VCPU_REGS_RCX
] -= delta
;
1912 vcpu
->regs
[VCPU_REGS_RDI
] += delta
;
1914 vcpu
->regs
[VCPU_REGS_RSI
] += delta
;
1917 kvm_x86_ops
->decache_regs(vcpu
);
1919 io
->count
-= io
->cur_count
;
1925 static void kernel_pio(struct kvm_io_device
*pio_dev
,
1926 struct kvm_vcpu
*vcpu
,
1929 /* TODO: String I/O for in kernel device */
1931 mutex_lock(&vcpu
->kvm
->lock
);
1933 kvm_iodevice_read(pio_dev
, vcpu
->pio
.port
,
1937 kvm_iodevice_write(pio_dev
, vcpu
->pio
.port
,
1940 mutex_unlock(&vcpu
->kvm
->lock
);
1943 static void pio_string_write(struct kvm_io_device
*pio_dev
,
1944 struct kvm_vcpu
*vcpu
)
1946 struct kvm_pio_request
*io
= &vcpu
->pio
;
1947 void *pd
= vcpu
->pio_data
;
1950 mutex_lock(&vcpu
->kvm
->lock
);
1951 for (i
= 0; i
< io
->cur_count
; i
++) {
1952 kvm_iodevice_write(pio_dev
, io
->port
,
1957 mutex_unlock(&vcpu
->kvm
->lock
);
1960 static struct kvm_io_device
*vcpu_find_pio_dev(struct kvm_vcpu
*vcpu
,
1963 return kvm_io_bus_find_dev(&vcpu
->kvm
->pio_bus
, addr
);
1966 int kvm_emulate_pio(struct kvm_vcpu
*vcpu
, struct kvm_run
*run
, int in
,
1967 int size
, unsigned port
)
1969 struct kvm_io_device
*pio_dev
;
1971 vcpu
->run
->exit_reason
= KVM_EXIT_IO
;
1972 vcpu
->run
->io
.direction
= in
? KVM_EXIT_IO_IN
: KVM_EXIT_IO_OUT
;
1973 vcpu
->run
->io
.size
= vcpu
->pio
.size
= size
;
1974 vcpu
->run
->io
.data_offset
= KVM_PIO_PAGE_OFFSET
* PAGE_SIZE
;
1975 vcpu
->run
->io
.count
= vcpu
->pio
.count
= vcpu
->pio
.cur_count
= 1;
1976 vcpu
->run
->io
.port
= vcpu
->pio
.port
= port
;
1978 vcpu
->pio
.string
= 0;
1980 vcpu
->pio
.guest_page_offset
= 0;
1983 kvm_x86_ops
->cache_regs(vcpu
);
1984 memcpy(vcpu
->pio_data
, &vcpu
->regs
[VCPU_REGS_RAX
], 4);
1985 kvm_x86_ops
->decache_regs(vcpu
);
1987 kvm_x86_ops
->skip_emulated_instruction(vcpu
);
1989 pio_dev
= vcpu_find_pio_dev(vcpu
, port
);
1991 kernel_pio(pio_dev
, vcpu
, vcpu
->pio_data
);
1997 EXPORT_SYMBOL_GPL(kvm_emulate_pio
);
1999 int kvm_emulate_pio_string(struct kvm_vcpu
*vcpu
, struct kvm_run
*run
, int in
,
2000 int size
, unsigned long count
, int down
,
2001 gva_t address
, int rep
, unsigned port
)
2003 unsigned now
, in_page
;
2007 struct kvm_io_device
*pio_dev
;
2009 vcpu
->run
->exit_reason
= KVM_EXIT_IO
;
2010 vcpu
->run
->io
.direction
= in
? KVM_EXIT_IO_IN
: KVM_EXIT_IO_OUT
;
2011 vcpu
->run
->io
.size
= vcpu
->pio
.size
= size
;
2012 vcpu
->run
->io
.data_offset
= KVM_PIO_PAGE_OFFSET
* PAGE_SIZE
;
2013 vcpu
->run
->io
.count
= vcpu
->pio
.count
= vcpu
->pio
.cur_count
= count
;
2014 vcpu
->run
->io
.port
= vcpu
->pio
.port
= port
;
2016 vcpu
->pio
.string
= 1;
2017 vcpu
->pio
.down
= down
;
2018 vcpu
->pio
.guest_page_offset
= offset_in_page(address
);
2019 vcpu
->pio
.rep
= rep
;
2022 kvm_x86_ops
->skip_emulated_instruction(vcpu
);
2027 in_page
= PAGE_SIZE
- offset_in_page(address
);
2029 in_page
= offset_in_page(address
) + size
;
2030 now
= min(count
, (unsigned long)in_page
/ size
);
2033 * String I/O straddles page boundary. Pin two guest pages
2034 * so that we satisfy atomicity constraints. Do just one
2035 * transaction to avoid complexity.
2042 * String I/O in reverse. Yuck. Kill the guest, fix later.
2044 pr_unimpl(vcpu
, "guest string pio down\n");
2045 kvm_inject_gp(vcpu
, 0);
2048 vcpu
->run
->io
.count
= now
;
2049 vcpu
->pio
.cur_count
= now
;
2051 if (vcpu
->pio
.cur_count
== vcpu
->pio
.count
)
2052 kvm_x86_ops
->skip_emulated_instruction(vcpu
);
2054 for (i
= 0; i
< nr_pages
; ++i
) {
2055 mutex_lock(&vcpu
->kvm
->lock
);
2056 page
= gva_to_page(vcpu
, address
+ i
* PAGE_SIZE
);
2057 vcpu
->pio
.guest_pages
[i
] = page
;
2058 mutex_unlock(&vcpu
->kvm
->lock
);
2060 kvm_inject_gp(vcpu
, 0);
2061 free_pio_guest_pages(vcpu
);
2066 pio_dev
= vcpu_find_pio_dev(vcpu
, port
);
2067 if (!vcpu
->pio
.in
) {
2068 /* string PIO write */
2069 ret
= pio_copy_data(vcpu
);
2070 if (ret
>= 0 && pio_dev
) {
2071 pio_string_write(pio_dev
, vcpu
);
2073 if (vcpu
->pio
.count
== 0)
2077 pr_unimpl(vcpu
, "no string pio read support yet, "
2078 "port %x size %d count %ld\n",
2083 EXPORT_SYMBOL_GPL(kvm_emulate_pio_string
);
2085 int kvm_arch_init(void *opaque
)
2088 struct kvm_x86_ops
*ops
= (struct kvm_x86_ops
*)opaque
;
2090 r
= kvm_mmu_module_init();
2094 kvm_init_msr_list();
2097 printk(KERN_ERR
"kvm: already loaded the other module\n");
2102 if (!ops
->cpu_has_kvm_support()) {
2103 printk(KERN_ERR
"kvm: no hardware support\n");
2107 if (ops
->disabled_by_bios()) {
2108 printk(KERN_ERR
"kvm: disabled by bios\n");
2114 kvm_mmu_set_nonpresent_ptes(0ull, 0ull);
2118 kvm_mmu_module_exit();
2123 void kvm_arch_exit(void)
2126 kvm_mmu_module_exit();
2129 int kvm_emulate_halt(struct kvm_vcpu
*vcpu
)
2131 ++vcpu
->stat
.halt_exits
;
2132 if (irqchip_in_kernel(vcpu
->kvm
)) {
2133 vcpu
->mp_state
= VCPU_MP_STATE_HALTED
;
2134 kvm_vcpu_block(vcpu
);
2135 if (vcpu
->mp_state
!= VCPU_MP_STATE_RUNNABLE
)
2139 vcpu
->run
->exit_reason
= KVM_EXIT_HLT
;
2143 EXPORT_SYMBOL_GPL(kvm_emulate_halt
);
2145 int kvm_emulate_hypercall(struct kvm_vcpu
*vcpu
)
2147 unsigned long nr
, a0
, a1
, a2
, a3
, ret
;
2149 kvm_x86_ops
->cache_regs(vcpu
);
2151 nr
= vcpu
->regs
[VCPU_REGS_RAX
];
2152 a0
= vcpu
->regs
[VCPU_REGS_RBX
];
2153 a1
= vcpu
->regs
[VCPU_REGS_RCX
];
2154 a2
= vcpu
->regs
[VCPU_REGS_RDX
];
2155 a3
= vcpu
->regs
[VCPU_REGS_RSI
];
2157 if (!is_long_mode(vcpu
)) {
2170 vcpu
->regs
[VCPU_REGS_RAX
] = ret
;
2171 kvm_x86_ops
->decache_regs(vcpu
);
2174 EXPORT_SYMBOL_GPL(kvm_emulate_hypercall
);
2176 int kvm_fix_hypercall(struct kvm_vcpu
*vcpu
)
2178 char instruction
[3];
2181 mutex_lock(&vcpu
->kvm
->lock
);
2184 * Blow out the MMU to ensure that no other VCPU has an active mapping
2185 * to ensure that the updated hypercall appears atomically across all
2188 kvm_mmu_zap_all(vcpu
->kvm
);
2190 kvm_x86_ops
->cache_regs(vcpu
);
2191 kvm_x86_ops
->patch_hypercall(vcpu
, instruction
);
2192 if (emulator_write_emulated(vcpu
->rip
, instruction
, 3, vcpu
)
2193 != X86EMUL_CONTINUE
)
2196 mutex_unlock(&vcpu
->kvm
->lock
);
2201 static u64
mk_cr_64(u64 curr_cr
, u32 new_val
)
2203 return (curr_cr
& ~((1ULL << 32) - 1)) | new_val
;
2206 void realmode_lgdt(struct kvm_vcpu
*vcpu
, u16 limit
, unsigned long base
)
2208 struct descriptor_table dt
= { limit
, base
};
2210 kvm_x86_ops
->set_gdt(vcpu
, &dt
);
2213 void realmode_lidt(struct kvm_vcpu
*vcpu
, u16 limit
, unsigned long base
)
2215 struct descriptor_table dt
= { limit
, base
};
2217 kvm_x86_ops
->set_idt(vcpu
, &dt
);
2220 void realmode_lmsw(struct kvm_vcpu
*vcpu
, unsigned long msw
,
2221 unsigned long *rflags
)
2224 *rflags
= kvm_x86_ops
->get_rflags(vcpu
);
2227 unsigned long realmode_get_cr(struct kvm_vcpu
*vcpu
, int cr
)
2229 kvm_x86_ops
->decache_cr4_guest_bits(vcpu
);
2240 vcpu_printf(vcpu
, "%s: unexpected cr %u\n", __FUNCTION__
, cr
);
2245 void realmode_set_cr(struct kvm_vcpu
*vcpu
, int cr
, unsigned long val
,
2246 unsigned long *rflags
)
2250 set_cr0(vcpu
, mk_cr_64(vcpu
->cr0
, val
));
2251 *rflags
= kvm_x86_ops
->get_rflags(vcpu
);
2260 set_cr4(vcpu
, mk_cr_64(vcpu
->cr4
, val
));
2263 vcpu_printf(vcpu
, "%s: unexpected cr %u\n", __FUNCTION__
, cr
);
2267 static int move_to_next_stateful_cpuid_entry(struct kvm_vcpu
*vcpu
, int i
)
2269 struct kvm_cpuid_entry2
*e
= &vcpu
->cpuid_entries
[i
];
2270 int j
, nent
= vcpu
->cpuid_nent
;
2272 e
->flags
&= ~KVM_CPUID_FLAG_STATE_READ_NEXT
;
2273 /* when no next entry is found, the current entry[i] is reselected */
2274 for (j
= i
+ 1; j
== i
; j
= (j
+ 1) % nent
) {
2275 struct kvm_cpuid_entry2
*ej
= &vcpu
->cpuid_entries
[j
];
2276 if (ej
->function
== e
->function
) {
2277 ej
->flags
|= KVM_CPUID_FLAG_STATE_READ_NEXT
;
2281 return 0; /* silence gcc, even though control never reaches here */
2284 /* find an entry with matching function, matching index (if needed), and that
2285 * should be read next (if it's stateful) */
2286 static int is_matching_cpuid_entry(struct kvm_cpuid_entry2
*e
,
2287 u32 function
, u32 index
)
2289 if (e
->function
!= function
)
2291 if ((e
->flags
& KVM_CPUID_FLAG_SIGNIFCANT_INDEX
) && e
->index
!= index
)
2293 if ((e
->flags
& KVM_CPUID_FLAG_STATEFUL_FUNC
) &&
2294 !(e
->flags
& KVM_CPUID_FLAG_STATE_READ_NEXT
))
2299 void kvm_emulate_cpuid(struct kvm_vcpu
*vcpu
)
2302 u32 function
, index
;
2303 struct kvm_cpuid_entry2
*e
, *best
;
2305 kvm_x86_ops
->cache_regs(vcpu
);
2306 function
= vcpu
->regs
[VCPU_REGS_RAX
];
2307 index
= vcpu
->regs
[VCPU_REGS_RCX
];
2308 vcpu
->regs
[VCPU_REGS_RAX
] = 0;
2309 vcpu
->regs
[VCPU_REGS_RBX
] = 0;
2310 vcpu
->regs
[VCPU_REGS_RCX
] = 0;
2311 vcpu
->regs
[VCPU_REGS_RDX
] = 0;
2313 for (i
= 0; i
< vcpu
->cpuid_nent
; ++i
) {
2314 e
= &vcpu
->cpuid_entries
[i
];
2315 if (is_matching_cpuid_entry(e
, function
, index
)) {
2316 if (e
->flags
& KVM_CPUID_FLAG_STATEFUL_FUNC
)
2317 move_to_next_stateful_cpuid_entry(vcpu
, i
);
2322 * Both basic or both extended?
2324 if (((e
->function
^ function
) & 0x80000000) == 0)
2325 if (!best
|| e
->function
> best
->function
)
2329 vcpu
->regs
[VCPU_REGS_RAX
] = best
->eax
;
2330 vcpu
->regs
[VCPU_REGS_RBX
] = best
->ebx
;
2331 vcpu
->regs
[VCPU_REGS_RCX
] = best
->ecx
;
2332 vcpu
->regs
[VCPU_REGS_RDX
] = best
->edx
;
2334 kvm_x86_ops
->decache_regs(vcpu
);
2335 kvm_x86_ops
->skip_emulated_instruction(vcpu
);
2337 EXPORT_SYMBOL_GPL(kvm_emulate_cpuid
);
2340 * Check if userspace requested an interrupt window, and that the
2341 * interrupt window is open.
2343 * No need to exit to userspace if we already have an interrupt queued.
2345 static int dm_request_for_irq_injection(struct kvm_vcpu
*vcpu
,
2346 struct kvm_run
*kvm_run
)
2348 return (!vcpu
->irq_summary
&&
2349 kvm_run
->request_interrupt_window
&&
2350 vcpu
->interrupt_window_open
&&
2351 (kvm_x86_ops
->get_rflags(vcpu
) & X86_EFLAGS_IF
));
2354 static void post_kvm_run_save(struct kvm_vcpu
*vcpu
,
2355 struct kvm_run
*kvm_run
)
2357 kvm_run
->if_flag
= (kvm_x86_ops
->get_rflags(vcpu
) & X86_EFLAGS_IF
) != 0;
2358 kvm_run
->cr8
= get_cr8(vcpu
);
2359 kvm_run
->apic_base
= kvm_get_apic_base(vcpu
);
2360 if (irqchip_in_kernel(vcpu
->kvm
))
2361 kvm_run
->ready_for_interrupt_injection
= 1;
2363 kvm_run
->ready_for_interrupt_injection
=
2364 (vcpu
->interrupt_window_open
&&
2365 vcpu
->irq_summary
== 0);
2368 static int __vcpu_run(struct kvm_vcpu
*vcpu
, struct kvm_run
*kvm_run
)
2372 if (unlikely(vcpu
->mp_state
== VCPU_MP_STATE_SIPI_RECEIVED
)) {
2373 pr_debug("vcpu %d received sipi with vector # %x\n",
2374 vcpu
->vcpu_id
, vcpu
->sipi_vector
);
2375 kvm_lapic_reset(vcpu
);
2376 r
= kvm_x86_ops
->vcpu_reset(vcpu
);
2379 vcpu
->mp_state
= VCPU_MP_STATE_RUNNABLE
;
2383 if (vcpu
->guest_debug
.enabled
)
2384 kvm_x86_ops
->guest_debug_pre(vcpu
);
2387 r
= kvm_mmu_reload(vcpu
);
2391 kvm_inject_pending_timer_irqs(vcpu
);
2395 kvm_x86_ops
->prepare_guest_switch(vcpu
);
2396 kvm_load_guest_fpu(vcpu
);
2398 local_irq_disable();
2400 if (signal_pending(current
)) {
2404 kvm_run
->exit_reason
= KVM_EXIT_INTR
;
2405 ++vcpu
->stat
.signal_exits
;
2409 if (vcpu
->exception
.pending
)
2410 __queue_exception(vcpu
);
2411 else if (irqchip_in_kernel(vcpu
->kvm
))
2412 kvm_x86_ops
->inject_pending_irq(vcpu
);
2414 kvm_x86_ops
->inject_pending_vectors(vcpu
, kvm_run
);
2416 vcpu
->guest_mode
= 1;
2420 if (test_and_clear_bit(KVM_REQ_TLB_FLUSH
, &vcpu
->requests
))
2421 kvm_x86_ops
->tlb_flush(vcpu
);
2423 kvm_x86_ops
->run(vcpu
, kvm_run
);
2425 vcpu
->guest_mode
= 0;
2431 * We must have an instruction between local_irq_enable() and
2432 * kvm_guest_exit(), so the timer interrupt isn't delayed by
2433 * the interrupt shadow. The stat.exits increment will do nicely.
2434 * But we need to prevent reordering, hence this barrier():
2443 * Profile KVM exit RIPs:
2445 if (unlikely(prof_on
== KVM_PROFILING
)) {
2446 kvm_x86_ops
->cache_regs(vcpu
);
2447 profile_hit(KVM_PROFILING
, (void *)vcpu
->rip
);
2450 if (vcpu
->exception
.pending
&& kvm_x86_ops
->exception_injected(vcpu
))
2451 vcpu
->exception
.pending
= false;
2453 r
= kvm_x86_ops
->handle_exit(kvm_run
, vcpu
);
2456 if (dm_request_for_irq_injection(vcpu
, kvm_run
)) {
2458 kvm_run
->exit_reason
= KVM_EXIT_INTR
;
2459 ++vcpu
->stat
.request_irq_exits
;
2462 if (!need_resched())
2472 post_kvm_run_save(vcpu
, kvm_run
);
2477 int kvm_arch_vcpu_ioctl_run(struct kvm_vcpu
*vcpu
, struct kvm_run
*kvm_run
)
2484 if (unlikely(vcpu
->mp_state
== VCPU_MP_STATE_UNINITIALIZED
)) {
2485 kvm_vcpu_block(vcpu
);
2490 if (vcpu
->sigset_active
)
2491 sigprocmask(SIG_SETMASK
, &vcpu
->sigset
, &sigsaved
);
2493 /* re-sync apic's tpr */
2494 if (!irqchip_in_kernel(vcpu
->kvm
))
2495 set_cr8(vcpu
, kvm_run
->cr8
);
2497 if (vcpu
->pio
.cur_count
) {
2498 r
= complete_pio(vcpu
);
2502 #if CONFIG_HAS_IOMEM
2503 if (vcpu
->mmio_needed
) {
2504 memcpy(vcpu
->mmio_data
, kvm_run
->mmio
.data
, 8);
2505 vcpu
->mmio_read_completed
= 1;
2506 vcpu
->mmio_needed
= 0;
2507 r
= emulate_instruction(vcpu
, kvm_run
,
2508 vcpu
->mmio_fault_cr2
, 0, 1);
2509 if (r
== EMULATE_DO_MMIO
) {
2511 * Read-modify-write. Back to userspace.
2518 if (kvm_run
->exit_reason
== KVM_EXIT_HYPERCALL
) {
2519 kvm_x86_ops
->cache_regs(vcpu
);
2520 vcpu
->regs
[VCPU_REGS_RAX
] = kvm_run
->hypercall
.ret
;
2521 kvm_x86_ops
->decache_regs(vcpu
);
2524 r
= __vcpu_run(vcpu
, kvm_run
);
2527 if (vcpu
->sigset_active
)
2528 sigprocmask(SIG_SETMASK
, &sigsaved
, NULL
);
2534 int kvm_arch_vcpu_ioctl_get_regs(struct kvm_vcpu
*vcpu
, struct kvm_regs
*regs
)
2538 kvm_x86_ops
->cache_regs(vcpu
);
2540 regs
->rax
= vcpu
->regs
[VCPU_REGS_RAX
];
2541 regs
->rbx
= vcpu
->regs
[VCPU_REGS_RBX
];
2542 regs
->rcx
= vcpu
->regs
[VCPU_REGS_RCX
];
2543 regs
->rdx
= vcpu
->regs
[VCPU_REGS_RDX
];
2544 regs
->rsi
= vcpu
->regs
[VCPU_REGS_RSI
];
2545 regs
->rdi
= vcpu
->regs
[VCPU_REGS_RDI
];
2546 regs
->rsp
= vcpu
->regs
[VCPU_REGS_RSP
];
2547 regs
->rbp
= vcpu
->regs
[VCPU_REGS_RBP
];
2548 #ifdef CONFIG_X86_64
2549 regs
->r8
= vcpu
->regs
[VCPU_REGS_R8
];
2550 regs
->r9
= vcpu
->regs
[VCPU_REGS_R9
];
2551 regs
->r10
= vcpu
->regs
[VCPU_REGS_R10
];
2552 regs
->r11
= vcpu
->regs
[VCPU_REGS_R11
];
2553 regs
->r12
= vcpu
->regs
[VCPU_REGS_R12
];
2554 regs
->r13
= vcpu
->regs
[VCPU_REGS_R13
];
2555 regs
->r14
= vcpu
->regs
[VCPU_REGS_R14
];
2556 regs
->r15
= vcpu
->regs
[VCPU_REGS_R15
];
2559 regs
->rip
= vcpu
->rip
;
2560 regs
->rflags
= kvm_x86_ops
->get_rflags(vcpu
);
2563 * Don't leak debug flags in case they were set for guest debugging
2565 if (vcpu
->guest_debug
.enabled
&& vcpu
->guest_debug
.singlestep
)
2566 regs
->rflags
&= ~(X86_EFLAGS_TF
| X86_EFLAGS_RF
);
2573 int kvm_arch_vcpu_ioctl_set_regs(struct kvm_vcpu
*vcpu
, struct kvm_regs
*regs
)
2577 vcpu
->regs
[VCPU_REGS_RAX
] = regs
->rax
;
2578 vcpu
->regs
[VCPU_REGS_RBX
] = regs
->rbx
;
2579 vcpu
->regs
[VCPU_REGS_RCX
] = regs
->rcx
;
2580 vcpu
->regs
[VCPU_REGS_RDX
] = regs
->rdx
;
2581 vcpu
->regs
[VCPU_REGS_RSI
] = regs
->rsi
;
2582 vcpu
->regs
[VCPU_REGS_RDI
] = regs
->rdi
;
2583 vcpu
->regs
[VCPU_REGS_RSP
] = regs
->rsp
;
2584 vcpu
->regs
[VCPU_REGS_RBP
] = regs
->rbp
;
2585 #ifdef CONFIG_X86_64
2586 vcpu
->regs
[VCPU_REGS_R8
] = regs
->r8
;
2587 vcpu
->regs
[VCPU_REGS_R9
] = regs
->r9
;
2588 vcpu
->regs
[VCPU_REGS_R10
] = regs
->r10
;
2589 vcpu
->regs
[VCPU_REGS_R11
] = regs
->r11
;
2590 vcpu
->regs
[VCPU_REGS_R12
] = regs
->r12
;
2591 vcpu
->regs
[VCPU_REGS_R13
] = regs
->r13
;
2592 vcpu
->regs
[VCPU_REGS_R14
] = regs
->r14
;
2593 vcpu
->regs
[VCPU_REGS_R15
] = regs
->r15
;
2596 vcpu
->rip
= regs
->rip
;
2597 kvm_x86_ops
->set_rflags(vcpu
, regs
->rflags
);
2599 kvm_x86_ops
->decache_regs(vcpu
);
2606 static void get_segment(struct kvm_vcpu
*vcpu
,
2607 struct kvm_segment
*var
, int seg
)
2609 return kvm_x86_ops
->get_segment(vcpu
, var
, seg
);
2612 void kvm_get_cs_db_l_bits(struct kvm_vcpu
*vcpu
, int *db
, int *l
)
2614 struct kvm_segment cs
;
2616 get_segment(vcpu
, &cs
, VCPU_SREG_CS
);
2620 EXPORT_SYMBOL_GPL(kvm_get_cs_db_l_bits
);
2622 int kvm_arch_vcpu_ioctl_get_sregs(struct kvm_vcpu
*vcpu
,
2623 struct kvm_sregs
*sregs
)
2625 struct descriptor_table dt
;
2630 get_segment(vcpu
, &sregs
->cs
, VCPU_SREG_CS
);
2631 get_segment(vcpu
, &sregs
->ds
, VCPU_SREG_DS
);
2632 get_segment(vcpu
, &sregs
->es
, VCPU_SREG_ES
);
2633 get_segment(vcpu
, &sregs
->fs
, VCPU_SREG_FS
);
2634 get_segment(vcpu
, &sregs
->gs
, VCPU_SREG_GS
);
2635 get_segment(vcpu
, &sregs
->ss
, VCPU_SREG_SS
);
2637 get_segment(vcpu
, &sregs
->tr
, VCPU_SREG_TR
);
2638 get_segment(vcpu
, &sregs
->ldt
, VCPU_SREG_LDTR
);
2640 kvm_x86_ops
->get_idt(vcpu
, &dt
);
2641 sregs
->idt
.limit
= dt
.limit
;
2642 sregs
->idt
.base
= dt
.base
;
2643 kvm_x86_ops
->get_gdt(vcpu
, &dt
);
2644 sregs
->gdt
.limit
= dt
.limit
;
2645 sregs
->gdt
.base
= dt
.base
;
2647 kvm_x86_ops
->decache_cr4_guest_bits(vcpu
);
2648 sregs
->cr0
= vcpu
->cr0
;
2649 sregs
->cr2
= vcpu
->cr2
;
2650 sregs
->cr3
= vcpu
->cr3
;
2651 sregs
->cr4
= vcpu
->cr4
;
2652 sregs
->cr8
= get_cr8(vcpu
);
2653 sregs
->efer
= vcpu
->shadow_efer
;
2654 sregs
->apic_base
= kvm_get_apic_base(vcpu
);
2656 if (irqchip_in_kernel(vcpu
->kvm
)) {
2657 memset(sregs
->interrupt_bitmap
, 0,
2658 sizeof sregs
->interrupt_bitmap
);
2659 pending_vec
= kvm_x86_ops
->get_irq(vcpu
);
2660 if (pending_vec
>= 0)
2661 set_bit(pending_vec
,
2662 (unsigned long *)sregs
->interrupt_bitmap
);
2664 memcpy(sregs
->interrupt_bitmap
, vcpu
->irq_pending
,
2665 sizeof sregs
->interrupt_bitmap
);
2672 static void set_segment(struct kvm_vcpu
*vcpu
,
2673 struct kvm_segment
*var
, int seg
)
2675 return kvm_x86_ops
->set_segment(vcpu
, var
, seg
);
2678 int kvm_arch_vcpu_ioctl_set_sregs(struct kvm_vcpu
*vcpu
,
2679 struct kvm_sregs
*sregs
)
2681 int mmu_reset_needed
= 0;
2682 int i
, pending_vec
, max_bits
;
2683 struct descriptor_table dt
;
2687 dt
.limit
= sregs
->idt
.limit
;
2688 dt
.base
= sregs
->idt
.base
;
2689 kvm_x86_ops
->set_idt(vcpu
, &dt
);
2690 dt
.limit
= sregs
->gdt
.limit
;
2691 dt
.base
= sregs
->gdt
.base
;
2692 kvm_x86_ops
->set_gdt(vcpu
, &dt
);
2694 vcpu
->cr2
= sregs
->cr2
;
2695 mmu_reset_needed
|= vcpu
->cr3
!= sregs
->cr3
;
2696 vcpu
->cr3
= sregs
->cr3
;
2698 set_cr8(vcpu
, sregs
->cr8
);
2700 mmu_reset_needed
|= vcpu
->shadow_efer
!= sregs
->efer
;
2701 #ifdef CONFIG_X86_64
2702 kvm_x86_ops
->set_efer(vcpu
, sregs
->efer
);
2704 kvm_set_apic_base(vcpu
, sregs
->apic_base
);
2706 kvm_x86_ops
->decache_cr4_guest_bits(vcpu
);
2708 mmu_reset_needed
|= vcpu
->cr0
!= sregs
->cr0
;
2709 vcpu
->cr0
= sregs
->cr0
;
2710 kvm_x86_ops
->set_cr0(vcpu
, sregs
->cr0
);
2712 mmu_reset_needed
|= vcpu
->cr4
!= sregs
->cr4
;
2713 kvm_x86_ops
->set_cr4(vcpu
, sregs
->cr4
);
2714 if (!is_long_mode(vcpu
) && is_pae(vcpu
))
2715 load_pdptrs(vcpu
, vcpu
->cr3
);
2717 if (mmu_reset_needed
)
2718 kvm_mmu_reset_context(vcpu
);
2720 if (!irqchip_in_kernel(vcpu
->kvm
)) {
2721 memcpy(vcpu
->irq_pending
, sregs
->interrupt_bitmap
,
2722 sizeof vcpu
->irq_pending
);
2723 vcpu
->irq_summary
= 0;
2724 for (i
= 0; i
< ARRAY_SIZE(vcpu
->irq_pending
); ++i
)
2725 if (vcpu
->irq_pending
[i
])
2726 __set_bit(i
, &vcpu
->irq_summary
);
2728 max_bits
= (sizeof sregs
->interrupt_bitmap
) << 3;
2729 pending_vec
= find_first_bit(
2730 (const unsigned long *)sregs
->interrupt_bitmap
,
2732 /* Only pending external irq is handled here */
2733 if (pending_vec
< max_bits
) {
2734 kvm_x86_ops
->set_irq(vcpu
, pending_vec
);
2735 pr_debug("Set back pending irq %d\n",
2740 set_segment(vcpu
, &sregs
->cs
, VCPU_SREG_CS
);
2741 set_segment(vcpu
, &sregs
->ds
, VCPU_SREG_DS
);
2742 set_segment(vcpu
, &sregs
->es
, VCPU_SREG_ES
);
2743 set_segment(vcpu
, &sregs
->fs
, VCPU_SREG_FS
);
2744 set_segment(vcpu
, &sregs
->gs
, VCPU_SREG_GS
);
2745 set_segment(vcpu
, &sregs
->ss
, VCPU_SREG_SS
);
2747 set_segment(vcpu
, &sregs
->tr
, VCPU_SREG_TR
);
2748 set_segment(vcpu
, &sregs
->ldt
, VCPU_SREG_LDTR
);
2755 int kvm_arch_vcpu_ioctl_debug_guest(struct kvm_vcpu
*vcpu
,
2756 struct kvm_debug_guest
*dbg
)
2762 r
= kvm_x86_ops
->set_guest_debug(vcpu
, dbg
);
2770 * fxsave fpu state. Taken from x86_64/processor.h. To be killed when
2771 * we have asm/x86/processor.h
2782 u32 st_space
[32]; /* 8*16 bytes for each FP-reg = 128 bytes */
2783 #ifdef CONFIG_X86_64
2784 u32 xmm_space
[64]; /* 16*16 bytes for each XMM-reg = 256 bytes */
2786 u32 xmm_space
[32]; /* 8*16 bytes for each XMM-reg = 128 bytes */
2791 * Translate a guest virtual address to a guest physical address.
2793 int kvm_arch_vcpu_ioctl_translate(struct kvm_vcpu
*vcpu
,
2794 struct kvm_translation
*tr
)
2796 unsigned long vaddr
= tr
->linear_address
;
2800 mutex_lock(&vcpu
->kvm
->lock
);
2801 gpa
= vcpu
->mmu
.gva_to_gpa(vcpu
, vaddr
);
2802 tr
->physical_address
= gpa
;
2803 tr
->valid
= gpa
!= UNMAPPED_GVA
;
2806 mutex_unlock(&vcpu
->kvm
->lock
);
2812 int kvm_arch_vcpu_ioctl_get_fpu(struct kvm_vcpu
*vcpu
, struct kvm_fpu
*fpu
)
2814 struct fxsave
*fxsave
= (struct fxsave
*)&vcpu
->guest_fx_image
;
2818 memcpy(fpu
->fpr
, fxsave
->st_space
, 128);
2819 fpu
->fcw
= fxsave
->cwd
;
2820 fpu
->fsw
= fxsave
->swd
;
2821 fpu
->ftwx
= fxsave
->twd
;
2822 fpu
->last_opcode
= fxsave
->fop
;
2823 fpu
->last_ip
= fxsave
->rip
;
2824 fpu
->last_dp
= fxsave
->rdp
;
2825 memcpy(fpu
->xmm
, fxsave
->xmm_space
, sizeof fxsave
->xmm_space
);
2832 int kvm_arch_vcpu_ioctl_set_fpu(struct kvm_vcpu
*vcpu
, struct kvm_fpu
*fpu
)
2834 struct fxsave
*fxsave
= (struct fxsave
*)&vcpu
->guest_fx_image
;
2838 memcpy(fxsave
->st_space
, fpu
->fpr
, 128);
2839 fxsave
->cwd
= fpu
->fcw
;
2840 fxsave
->swd
= fpu
->fsw
;
2841 fxsave
->twd
= fpu
->ftwx
;
2842 fxsave
->fop
= fpu
->last_opcode
;
2843 fxsave
->rip
= fpu
->last_ip
;
2844 fxsave
->rdp
= fpu
->last_dp
;
2845 memcpy(fxsave
->xmm_space
, fpu
->xmm
, sizeof fxsave
->xmm_space
);
2852 void fx_init(struct kvm_vcpu
*vcpu
)
2854 unsigned after_mxcsr_mask
;
2856 /* Initialize guest FPU by resetting ours and saving into guest's */
2858 fx_save(&vcpu
->host_fx_image
);
2860 fx_save(&vcpu
->guest_fx_image
);
2861 fx_restore(&vcpu
->host_fx_image
);
2864 vcpu
->cr0
|= X86_CR0_ET
;
2865 after_mxcsr_mask
= offsetof(struct i387_fxsave_struct
, st_space
);
2866 vcpu
->guest_fx_image
.mxcsr
= 0x1f80;
2867 memset((void *)&vcpu
->guest_fx_image
+ after_mxcsr_mask
,
2868 0, sizeof(struct i387_fxsave_struct
) - after_mxcsr_mask
);
2870 EXPORT_SYMBOL_GPL(fx_init
);
2872 void kvm_load_guest_fpu(struct kvm_vcpu
*vcpu
)
2874 if (!vcpu
->fpu_active
|| vcpu
->guest_fpu_loaded
)
2877 vcpu
->guest_fpu_loaded
= 1;
2878 fx_save(&vcpu
->host_fx_image
);
2879 fx_restore(&vcpu
->guest_fx_image
);
2881 EXPORT_SYMBOL_GPL(kvm_load_guest_fpu
);
2883 void kvm_put_guest_fpu(struct kvm_vcpu
*vcpu
)
2885 if (!vcpu
->guest_fpu_loaded
)
2888 vcpu
->guest_fpu_loaded
= 0;
2889 fx_save(&vcpu
->guest_fx_image
);
2890 fx_restore(&vcpu
->host_fx_image
);
2891 ++vcpu
->stat
.fpu_reload
;
2893 EXPORT_SYMBOL_GPL(kvm_put_guest_fpu
);
2895 void kvm_arch_vcpu_free(struct kvm_vcpu
*vcpu
)
2897 kvm_x86_ops
->vcpu_free(vcpu
);
2900 struct kvm_vcpu
*kvm_arch_vcpu_create(struct kvm
*kvm
,
2903 return kvm_x86_ops
->vcpu_create(kvm
, id
);
2906 int kvm_arch_vcpu_setup(struct kvm_vcpu
*vcpu
)
2910 /* We do fxsave: this must be aligned. */
2911 BUG_ON((unsigned long)&vcpu
->host_fx_image
& 0xF);
2914 r
= kvm_arch_vcpu_reset(vcpu
);
2916 r
= kvm_mmu_setup(vcpu
);
2923 kvm_x86_ops
->vcpu_free(vcpu
);
2927 void kvm_arch_vcpu_destroy(struct kvm_vcpu
*vcpu
)
2930 kvm_mmu_unload(vcpu
);
2933 kvm_x86_ops
->vcpu_free(vcpu
);
2936 int kvm_arch_vcpu_reset(struct kvm_vcpu
*vcpu
)
2938 return kvm_x86_ops
->vcpu_reset(vcpu
);
2941 void kvm_arch_hardware_enable(void *garbage
)
2943 kvm_x86_ops
->hardware_enable(garbage
);
2946 void kvm_arch_hardware_disable(void *garbage
)
2948 kvm_x86_ops
->hardware_disable(garbage
);
2951 int kvm_arch_hardware_setup(void)
2953 return kvm_x86_ops
->hardware_setup();
2956 void kvm_arch_hardware_unsetup(void)
2958 kvm_x86_ops
->hardware_unsetup();
2961 void kvm_arch_check_processor_compat(void *rtn
)
2963 kvm_x86_ops
->check_processor_compatibility(rtn
);
2966 int kvm_arch_vcpu_init(struct kvm_vcpu
*vcpu
)
2972 BUG_ON(vcpu
->kvm
== NULL
);
2975 vcpu
->mmu
.root_hpa
= INVALID_PAGE
;
2976 if (!irqchip_in_kernel(kvm
) || vcpu
->vcpu_id
== 0)
2977 vcpu
->mp_state
= VCPU_MP_STATE_RUNNABLE
;
2979 vcpu
->mp_state
= VCPU_MP_STATE_UNINITIALIZED
;
2981 page
= alloc_page(GFP_KERNEL
| __GFP_ZERO
);
2986 vcpu
->pio_data
= page_address(page
);
2988 r
= kvm_mmu_create(vcpu
);
2990 goto fail_free_pio_data
;
2992 if (irqchip_in_kernel(kvm
)) {
2993 r
= kvm_create_lapic(vcpu
);
2995 goto fail_mmu_destroy
;
3001 kvm_mmu_destroy(vcpu
);
3003 free_page((unsigned long)vcpu
->pio_data
);
3008 void kvm_arch_vcpu_uninit(struct kvm_vcpu
*vcpu
)
3010 kvm_free_lapic(vcpu
);
3011 kvm_mmu_destroy(vcpu
);
3012 free_page((unsigned long)vcpu
->pio_data
);
3015 struct kvm
*kvm_arch_create_vm(void)
3017 struct kvm
*kvm
= kzalloc(sizeof(struct kvm
), GFP_KERNEL
);
3020 return ERR_PTR(-ENOMEM
);
3022 INIT_LIST_HEAD(&kvm
->active_mmu_pages
);
3027 static void kvm_unload_vcpu_mmu(struct kvm_vcpu
*vcpu
)
3030 kvm_mmu_unload(vcpu
);
3034 static void kvm_free_vcpus(struct kvm
*kvm
)
3039 * Unpin any mmu pages first.
3041 for (i
= 0; i
< KVM_MAX_VCPUS
; ++i
)
3043 kvm_unload_vcpu_mmu(kvm
->vcpus
[i
]);
3044 for (i
= 0; i
< KVM_MAX_VCPUS
; ++i
) {
3045 if (kvm
->vcpus
[i
]) {
3046 kvm_arch_vcpu_free(kvm
->vcpus
[i
]);
3047 kvm
->vcpus
[i
] = NULL
;
3053 void kvm_arch_destroy_vm(struct kvm
*kvm
)
3056 kfree(kvm
->vioapic
);
3057 kvm_free_vcpus(kvm
);
3058 kvm_free_physmem(kvm
);
3062 int kvm_arch_set_memory_region(struct kvm
*kvm
,
3063 struct kvm_userspace_memory_region
*mem
,
3064 struct kvm_memory_slot old
,
3067 int npages
= mem
->memory_size
>> PAGE_SHIFT
;
3068 struct kvm_memory_slot
*memslot
= &kvm
->memslots
[mem
->slot
];
3070 /*To keep backward compatibility with older userspace,
3071 *x86 needs to hanlde !user_alloc case.
3074 if (npages
&& !old
.rmap
) {
3075 down_write(¤t
->mm
->mmap_sem
);
3076 memslot
->userspace_addr
= do_mmap(NULL
, 0,
3078 PROT_READ
| PROT_WRITE
,
3079 MAP_SHARED
| MAP_ANONYMOUS
,
3081 up_write(¤t
->mm
->mmap_sem
);
3083 if (IS_ERR((void *)memslot
->userspace_addr
))
3084 return PTR_ERR((void *)memslot
->userspace_addr
);
3086 if (!old
.user_alloc
&& old
.rmap
) {
3089 down_write(¤t
->mm
->mmap_sem
);
3090 ret
= do_munmap(current
->mm
, old
.userspace_addr
,
3091 old
.npages
* PAGE_SIZE
);
3092 up_write(¤t
->mm
->mmap_sem
);
3095 "kvm_vm_ioctl_set_memory_region: "
3096 "failed to munmap memory\n");
3101 if (!kvm
->n_requested_mmu_pages
) {
3102 unsigned int nr_mmu_pages
= kvm_mmu_calculate_mmu_pages(kvm
);
3103 kvm_mmu_change_mmu_pages(kvm
, nr_mmu_pages
);
3106 kvm_mmu_slot_remove_write_access(kvm
, mem
->slot
);
3107 kvm_flush_remote_tlbs(kvm
);