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.
17 #include <linux/kvm_host.h>
23 #include <linux/clocksource.h>
24 #include <linux/kvm.h>
26 #include <linux/vmalloc.h>
27 #include <linux/module.h>
28 #include <linux/mman.h>
29 #include <linux/highmem.h>
31 #include <asm/uaccess.h>
35 #define MAX_IO_MSRS 256
36 #define CR0_RESERVED_BITS \
37 (~(unsigned long)(X86_CR0_PE | X86_CR0_MP | X86_CR0_EM | X86_CR0_TS \
38 | X86_CR0_ET | X86_CR0_NE | X86_CR0_WP | X86_CR0_AM \
39 | X86_CR0_NW | X86_CR0_CD | X86_CR0_PG))
40 #define CR4_RESERVED_BITS \
41 (~(unsigned long)(X86_CR4_VME | X86_CR4_PVI | X86_CR4_TSD | X86_CR4_DE\
42 | X86_CR4_PSE | X86_CR4_PAE | X86_CR4_MCE \
43 | X86_CR4_PGE | X86_CR4_PCE | X86_CR4_OSFXSR \
44 | X86_CR4_OSXMMEXCPT | X86_CR4_VMXE))
46 #define CR8_RESERVED_BITS (~(unsigned long)X86_CR8_TPR)
48 * - enable syscall per default because its emulated by KVM
49 * - enable LME and LMA per default on 64 bit KVM
52 static u64 __read_mostly efer_reserved_bits
= 0xfffffffffffffafeULL
;
54 static u64 __read_mostly efer_reserved_bits
= 0xfffffffffffffffeULL
;
57 #define VM_STAT(x) offsetof(struct kvm, stat.x), KVM_STAT_VM
58 #define VCPU_STAT(x) offsetof(struct kvm_vcpu, stat.x), KVM_STAT_VCPU
60 static int kvm_dev_ioctl_get_supported_cpuid(struct kvm_cpuid2
*cpuid
,
61 struct kvm_cpuid_entry2 __user
*entries
);
63 struct kvm_x86_ops
*kvm_x86_ops
;
65 struct kvm_stats_debugfs_item debugfs_entries
[] = {
66 { "pf_fixed", VCPU_STAT(pf_fixed
) },
67 { "pf_guest", VCPU_STAT(pf_guest
) },
68 { "tlb_flush", VCPU_STAT(tlb_flush
) },
69 { "invlpg", VCPU_STAT(invlpg
) },
70 { "exits", VCPU_STAT(exits
) },
71 { "io_exits", VCPU_STAT(io_exits
) },
72 { "mmio_exits", VCPU_STAT(mmio_exits
) },
73 { "signal_exits", VCPU_STAT(signal_exits
) },
74 { "irq_window", VCPU_STAT(irq_window_exits
) },
75 { "halt_exits", VCPU_STAT(halt_exits
) },
76 { "halt_wakeup", VCPU_STAT(halt_wakeup
) },
77 { "hypercalls", VCPU_STAT(hypercalls
) },
78 { "request_irq", VCPU_STAT(request_irq_exits
) },
79 { "irq_exits", VCPU_STAT(irq_exits
) },
80 { "host_state_reload", VCPU_STAT(host_state_reload
) },
81 { "efer_reload", VCPU_STAT(efer_reload
) },
82 { "fpu_reload", VCPU_STAT(fpu_reload
) },
83 { "insn_emulation", VCPU_STAT(insn_emulation
) },
84 { "insn_emulation_fail", VCPU_STAT(insn_emulation_fail
) },
85 { "mmu_shadow_zapped", VM_STAT(mmu_shadow_zapped
) },
86 { "mmu_pte_write", VM_STAT(mmu_pte_write
) },
87 { "mmu_pte_updated", VM_STAT(mmu_pte_updated
) },
88 { "mmu_pde_zapped", VM_STAT(mmu_pde_zapped
) },
89 { "mmu_flooded", VM_STAT(mmu_flooded
) },
90 { "mmu_recycled", VM_STAT(mmu_recycled
) },
91 { "mmu_cache_miss", VM_STAT(mmu_cache_miss
) },
92 { "remote_tlb_flush", VM_STAT(remote_tlb_flush
) },
93 { "largepages", VM_STAT(lpages
) },
98 unsigned long segment_base(u16 selector
)
100 struct descriptor_table gdt
;
101 struct desc_struct
*d
;
102 unsigned long table_base
;
108 asm("sgdt %0" : "=m"(gdt
));
109 table_base
= gdt
.base
;
111 if (selector
& 4) { /* from ldt */
114 asm("sldt %0" : "=g"(ldt_selector
));
115 table_base
= segment_base(ldt_selector
);
117 d
= (struct desc_struct
*)(table_base
+ (selector
& ~7));
118 v
= d
->base0
| ((unsigned long)d
->base1
<< 16) |
119 ((unsigned long)d
->base2
<< 24);
121 if (d
->s
== 0 && (d
->type
== 2 || d
->type
== 9 || d
->type
== 11))
122 v
|= ((unsigned long)((struct ldttss_desc64
*)d
)->base3
) << 32;
126 EXPORT_SYMBOL_GPL(segment_base
);
128 u64
kvm_get_apic_base(struct kvm_vcpu
*vcpu
)
130 if (irqchip_in_kernel(vcpu
->kvm
))
131 return vcpu
->arch
.apic_base
;
133 return vcpu
->arch
.apic_base
;
135 EXPORT_SYMBOL_GPL(kvm_get_apic_base
);
137 void kvm_set_apic_base(struct kvm_vcpu
*vcpu
, u64 data
)
139 /* TODO: reserve bits check */
140 if (irqchip_in_kernel(vcpu
->kvm
))
141 kvm_lapic_set_base(vcpu
, data
);
143 vcpu
->arch
.apic_base
= data
;
145 EXPORT_SYMBOL_GPL(kvm_set_apic_base
);
147 void kvm_queue_exception(struct kvm_vcpu
*vcpu
, unsigned nr
)
149 WARN_ON(vcpu
->arch
.exception
.pending
);
150 vcpu
->arch
.exception
.pending
= true;
151 vcpu
->arch
.exception
.has_error_code
= false;
152 vcpu
->arch
.exception
.nr
= nr
;
154 EXPORT_SYMBOL_GPL(kvm_queue_exception
);
156 void kvm_inject_page_fault(struct kvm_vcpu
*vcpu
, unsigned long addr
,
159 ++vcpu
->stat
.pf_guest
;
160 if (vcpu
->arch
.exception
.pending
) {
161 if (vcpu
->arch
.exception
.nr
== PF_VECTOR
) {
162 printk(KERN_DEBUG
"kvm: inject_page_fault:"
163 " double fault 0x%lx\n", addr
);
164 vcpu
->arch
.exception
.nr
= DF_VECTOR
;
165 vcpu
->arch
.exception
.error_code
= 0;
166 } else if (vcpu
->arch
.exception
.nr
== DF_VECTOR
) {
167 /* triple fault -> shutdown */
168 set_bit(KVM_REQ_TRIPLE_FAULT
, &vcpu
->requests
);
172 vcpu
->arch
.cr2
= addr
;
173 kvm_queue_exception_e(vcpu
, PF_VECTOR
, error_code
);
176 void kvm_queue_exception_e(struct kvm_vcpu
*vcpu
, unsigned nr
, u32 error_code
)
178 WARN_ON(vcpu
->arch
.exception
.pending
);
179 vcpu
->arch
.exception
.pending
= true;
180 vcpu
->arch
.exception
.has_error_code
= true;
181 vcpu
->arch
.exception
.nr
= nr
;
182 vcpu
->arch
.exception
.error_code
= error_code
;
184 EXPORT_SYMBOL_GPL(kvm_queue_exception_e
);
186 static void __queue_exception(struct kvm_vcpu
*vcpu
)
188 kvm_x86_ops
->queue_exception(vcpu
, vcpu
->arch
.exception
.nr
,
189 vcpu
->arch
.exception
.has_error_code
,
190 vcpu
->arch
.exception
.error_code
);
194 * Load the pae pdptrs. Return true is they are all valid.
196 int load_pdptrs(struct kvm_vcpu
*vcpu
, unsigned long cr3
)
198 gfn_t pdpt_gfn
= cr3
>> PAGE_SHIFT
;
199 unsigned offset
= ((cr3
& (PAGE_SIZE
-1)) >> 5) << 2;
202 u64 pdpte
[ARRAY_SIZE(vcpu
->arch
.pdptrs
)];
204 down_read(&vcpu
->kvm
->slots_lock
);
205 ret
= kvm_read_guest_page(vcpu
->kvm
, pdpt_gfn
, pdpte
,
206 offset
* sizeof(u64
), sizeof(pdpte
));
211 for (i
= 0; i
< ARRAY_SIZE(pdpte
); ++i
) {
212 if ((pdpte
[i
] & 1) && (pdpte
[i
] & 0xfffffff0000001e6ull
)) {
219 memcpy(vcpu
->arch
.pdptrs
, pdpte
, sizeof(vcpu
->arch
.pdptrs
));
221 up_read(&vcpu
->kvm
->slots_lock
);
225 EXPORT_SYMBOL_GPL(load_pdptrs
);
227 static bool pdptrs_changed(struct kvm_vcpu
*vcpu
)
229 u64 pdpte
[ARRAY_SIZE(vcpu
->arch
.pdptrs
)];
233 if (is_long_mode(vcpu
) || !is_pae(vcpu
))
236 down_read(&vcpu
->kvm
->slots_lock
);
237 r
= kvm_read_guest(vcpu
->kvm
, vcpu
->arch
.cr3
& ~31u, pdpte
, sizeof(pdpte
));
240 changed
= memcmp(pdpte
, vcpu
->arch
.pdptrs
, sizeof(pdpte
)) != 0;
242 up_read(&vcpu
->kvm
->slots_lock
);
247 void kvm_set_cr0(struct kvm_vcpu
*vcpu
, unsigned long cr0
)
249 if (cr0
& CR0_RESERVED_BITS
) {
250 printk(KERN_DEBUG
"set_cr0: 0x%lx #GP, reserved bits 0x%lx\n",
251 cr0
, vcpu
->arch
.cr0
);
252 kvm_inject_gp(vcpu
, 0);
256 if ((cr0
& X86_CR0_NW
) && !(cr0
& X86_CR0_CD
)) {
257 printk(KERN_DEBUG
"set_cr0: #GP, CD == 0 && NW == 1\n");
258 kvm_inject_gp(vcpu
, 0);
262 if ((cr0
& X86_CR0_PG
) && !(cr0
& X86_CR0_PE
)) {
263 printk(KERN_DEBUG
"set_cr0: #GP, set PG flag "
264 "and a clear PE flag\n");
265 kvm_inject_gp(vcpu
, 0);
269 if (!is_paging(vcpu
) && (cr0
& X86_CR0_PG
)) {
271 if ((vcpu
->arch
.shadow_efer
& EFER_LME
)) {
275 printk(KERN_DEBUG
"set_cr0: #GP, start paging "
276 "in long mode while PAE is disabled\n");
277 kvm_inject_gp(vcpu
, 0);
280 kvm_x86_ops
->get_cs_db_l_bits(vcpu
, &cs_db
, &cs_l
);
282 printk(KERN_DEBUG
"set_cr0: #GP, start paging "
283 "in long mode while CS.L == 1\n");
284 kvm_inject_gp(vcpu
, 0);
290 if (is_pae(vcpu
) && !load_pdptrs(vcpu
, vcpu
->arch
.cr3
)) {
291 printk(KERN_DEBUG
"set_cr0: #GP, pdptrs "
293 kvm_inject_gp(vcpu
, 0);
299 kvm_x86_ops
->set_cr0(vcpu
, cr0
);
300 vcpu
->arch
.cr0
= cr0
;
302 kvm_mmu_reset_context(vcpu
);
305 EXPORT_SYMBOL_GPL(kvm_set_cr0
);
307 void kvm_lmsw(struct kvm_vcpu
*vcpu
, unsigned long msw
)
309 kvm_set_cr0(vcpu
, (vcpu
->arch
.cr0
& ~0x0ful
) | (msw
& 0x0f));
311 EXPORT_SYMBOL_GPL(kvm_lmsw
);
313 void kvm_set_cr4(struct kvm_vcpu
*vcpu
, unsigned long cr4
)
315 if (cr4
& CR4_RESERVED_BITS
) {
316 printk(KERN_DEBUG
"set_cr4: #GP, reserved bits\n");
317 kvm_inject_gp(vcpu
, 0);
321 if (is_long_mode(vcpu
)) {
322 if (!(cr4
& X86_CR4_PAE
)) {
323 printk(KERN_DEBUG
"set_cr4: #GP, clearing PAE while "
325 kvm_inject_gp(vcpu
, 0);
328 } else if (is_paging(vcpu
) && !is_pae(vcpu
) && (cr4
& X86_CR4_PAE
)
329 && !load_pdptrs(vcpu
, vcpu
->arch
.cr3
)) {
330 printk(KERN_DEBUG
"set_cr4: #GP, pdptrs reserved bits\n");
331 kvm_inject_gp(vcpu
, 0);
335 if (cr4
& X86_CR4_VMXE
) {
336 printk(KERN_DEBUG
"set_cr4: #GP, setting VMXE\n");
337 kvm_inject_gp(vcpu
, 0);
340 kvm_x86_ops
->set_cr4(vcpu
, cr4
);
341 vcpu
->arch
.cr4
= cr4
;
342 kvm_mmu_reset_context(vcpu
);
344 EXPORT_SYMBOL_GPL(kvm_set_cr4
);
346 void kvm_set_cr3(struct kvm_vcpu
*vcpu
, unsigned long cr3
)
348 if (cr3
== vcpu
->arch
.cr3
&& !pdptrs_changed(vcpu
)) {
349 kvm_mmu_flush_tlb(vcpu
);
353 if (is_long_mode(vcpu
)) {
354 if (cr3
& CR3_L_MODE_RESERVED_BITS
) {
355 printk(KERN_DEBUG
"set_cr3: #GP, reserved bits\n");
356 kvm_inject_gp(vcpu
, 0);
361 if (cr3
& CR3_PAE_RESERVED_BITS
) {
363 "set_cr3: #GP, reserved bits\n");
364 kvm_inject_gp(vcpu
, 0);
367 if (is_paging(vcpu
) && !load_pdptrs(vcpu
, cr3
)) {
368 printk(KERN_DEBUG
"set_cr3: #GP, pdptrs "
370 kvm_inject_gp(vcpu
, 0);
375 * We don't check reserved bits in nonpae mode, because
376 * this isn't enforced, and VMware depends on this.
380 down_read(&vcpu
->kvm
->slots_lock
);
382 * Does the new cr3 value map to physical memory? (Note, we
383 * catch an invalid cr3 even in real-mode, because it would
384 * cause trouble later on when we turn on paging anyway.)
386 * A real CPU would silently accept an invalid cr3 and would
387 * attempt to use it - with largely undefined (and often hard
388 * to debug) behavior on the guest side.
390 if (unlikely(!gfn_to_memslot(vcpu
->kvm
, cr3
>> PAGE_SHIFT
)))
391 kvm_inject_gp(vcpu
, 0);
393 vcpu
->arch
.cr3
= cr3
;
394 vcpu
->arch
.mmu
.new_cr3(vcpu
);
396 up_read(&vcpu
->kvm
->slots_lock
);
398 EXPORT_SYMBOL_GPL(kvm_set_cr3
);
400 void kvm_set_cr8(struct kvm_vcpu
*vcpu
, unsigned long cr8
)
402 if (cr8
& CR8_RESERVED_BITS
) {
403 printk(KERN_DEBUG
"set_cr8: #GP, reserved bits 0x%lx\n", cr8
);
404 kvm_inject_gp(vcpu
, 0);
407 if (irqchip_in_kernel(vcpu
->kvm
))
408 kvm_lapic_set_tpr(vcpu
, cr8
);
410 vcpu
->arch
.cr8
= cr8
;
412 EXPORT_SYMBOL_GPL(kvm_set_cr8
);
414 unsigned long kvm_get_cr8(struct kvm_vcpu
*vcpu
)
416 if (irqchip_in_kernel(vcpu
->kvm
))
417 return kvm_lapic_get_cr8(vcpu
);
419 return vcpu
->arch
.cr8
;
421 EXPORT_SYMBOL_GPL(kvm_get_cr8
);
424 * List of msr numbers which we expose to userspace through KVM_GET_MSRS
425 * and KVM_SET_MSRS, and KVM_GET_MSR_INDEX_LIST.
427 * This list is modified at module load time to reflect the
428 * capabilities of the host cpu.
430 static u32 msrs_to_save
[] = {
431 MSR_IA32_SYSENTER_CS
, MSR_IA32_SYSENTER_ESP
, MSR_IA32_SYSENTER_EIP
,
434 MSR_CSTAR
, MSR_KERNEL_GS_BASE
, MSR_SYSCALL_MASK
, MSR_LSTAR
,
436 MSR_IA32_TIME_STAMP_COUNTER
, MSR_KVM_SYSTEM_TIME
, MSR_KVM_WALL_CLOCK
,
437 MSR_IA32_PERF_STATUS
,
440 static unsigned num_msrs_to_save
;
442 static u32 emulated_msrs
[] = {
443 MSR_IA32_MISC_ENABLE
,
446 static void set_efer(struct kvm_vcpu
*vcpu
, u64 efer
)
448 if (efer
& efer_reserved_bits
) {
449 printk(KERN_DEBUG
"set_efer: 0x%llx #GP, reserved bits\n",
451 kvm_inject_gp(vcpu
, 0);
456 && (vcpu
->arch
.shadow_efer
& EFER_LME
) != (efer
& EFER_LME
)) {
457 printk(KERN_DEBUG
"set_efer: #GP, change LME while paging\n");
458 kvm_inject_gp(vcpu
, 0);
462 kvm_x86_ops
->set_efer(vcpu
, efer
);
465 efer
|= vcpu
->arch
.shadow_efer
& EFER_LMA
;
467 vcpu
->arch
.shadow_efer
= efer
;
470 void kvm_enable_efer_bits(u64 mask
)
472 efer_reserved_bits
&= ~mask
;
474 EXPORT_SYMBOL_GPL(kvm_enable_efer_bits
);
478 * Writes msr value into into the appropriate "register".
479 * Returns 0 on success, non-0 otherwise.
480 * Assumes vcpu_load() was already called.
482 int kvm_set_msr(struct kvm_vcpu
*vcpu
, u32 msr_index
, u64 data
)
484 return kvm_x86_ops
->set_msr(vcpu
, msr_index
, data
);
488 * Adapt set_msr() to msr_io()'s calling convention
490 static int do_set_msr(struct kvm_vcpu
*vcpu
, unsigned index
, u64
*data
)
492 return kvm_set_msr(vcpu
, index
, *data
);
495 static void kvm_write_wall_clock(struct kvm
*kvm
, gpa_t wall_clock
)
498 struct kvm_wall_clock wc
;
499 struct timespec wc_ts
;
506 down_read(&kvm
->slots_lock
);
507 kvm_write_guest(kvm
, wall_clock
, &version
, sizeof(version
));
509 wc_ts
= current_kernel_time();
510 wc
.wc_sec
= wc_ts
.tv_sec
;
511 wc
.wc_nsec
= wc_ts
.tv_nsec
;
512 wc
.wc_version
= version
;
514 kvm_write_guest(kvm
, wall_clock
, &wc
, sizeof(wc
));
517 kvm_write_guest(kvm
, wall_clock
, &version
, sizeof(version
));
518 up_read(&kvm
->slots_lock
);
521 static void kvm_write_guest_time(struct kvm_vcpu
*v
)
525 struct kvm_vcpu_arch
*vcpu
= &v
->arch
;
528 if ((!vcpu
->time_page
))
531 /* Keep irq disabled to prevent changes to the clock */
532 local_irq_save(flags
);
533 kvm_get_msr(v
, MSR_IA32_TIME_STAMP_COUNTER
,
534 &vcpu
->hv_clock
.tsc_timestamp
);
536 local_irq_restore(flags
);
538 /* With all the info we got, fill in the values */
540 vcpu
->hv_clock
.system_time
= ts
.tv_nsec
+
541 (NSEC_PER_SEC
* (u64
)ts
.tv_sec
);
543 * The interface expects us to write an even number signaling that the
544 * update is finished. Since the guest won't see the intermediate
545 * state, we just write "2" at the end
547 vcpu
->hv_clock
.version
= 2;
549 shared_kaddr
= kmap_atomic(vcpu
->time_page
, KM_USER0
);
551 memcpy(shared_kaddr
+ vcpu
->time_offset
, &vcpu
->hv_clock
,
552 sizeof(vcpu
->hv_clock
));
554 kunmap_atomic(shared_kaddr
, KM_USER0
);
556 mark_page_dirty(v
->kvm
, vcpu
->time
>> PAGE_SHIFT
);
560 int kvm_set_msr_common(struct kvm_vcpu
*vcpu
, u32 msr
, u64 data
)
564 set_efer(vcpu
, data
);
566 case MSR_IA32_MC0_STATUS
:
567 pr_unimpl(vcpu
, "%s: MSR_IA32_MC0_STATUS 0x%llx, nop\n",
570 case MSR_IA32_MCG_STATUS
:
571 pr_unimpl(vcpu
, "%s: MSR_IA32_MCG_STATUS 0x%llx, nop\n",
574 case MSR_IA32_MCG_CTL
:
575 pr_unimpl(vcpu
, "%s: MSR_IA32_MCG_CTL 0x%llx, nop\n",
578 case MSR_IA32_UCODE_REV
:
579 case MSR_IA32_UCODE_WRITE
:
580 case 0x200 ... 0x2ff: /* MTRRs */
582 case MSR_IA32_APICBASE
:
583 kvm_set_apic_base(vcpu
, data
);
585 case MSR_IA32_MISC_ENABLE
:
586 vcpu
->arch
.ia32_misc_enable_msr
= data
;
588 case MSR_KVM_WALL_CLOCK
:
589 vcpu
->kvm
->arch
.wall_clock
= data
;
590 kvm_write_wall_clock(vcpu
->kvm
, data
);
592 case MSR_KVM_SYSTEM_TIME
: {
593 if (vcpu
->arch
.time_page
) {
594 kvm_release_page_dirty(vcpu
->arch
.time_page
);
595 vcpu
->arch
.time_page
= NULL
;
598 vcpu
->arch
.time
= data
;
600 /* we verify if the enable bit is set... */
604 /* ...but clean it before doing the actual write */
605 vcpu
->arch
.time_offset
= data
& ~(PAGE_MASK
| 1);
607 vcpu
->arch
.hv_clock
.tsc_to_system_mul
=
608 clocksource_khz2mult(tsc_khz
, 22);
609 vcpu
->arch
.hv_clock
.tsc_shift
= 22;
611 down_read(¤t
->mm
->mmap_sem
);
612 down_read(&vcpu
->kvm
->slots_lock
);
613 vcpu
->arch
.time_page
=
614 gfn_to_page(vcpu
->kvm
, data
>> PAGE_SHIFT
);
615 up_read(&vcpu
->kvm
->slots_lock
);
616 up_read(¤t
->mm
->mmap_sem
);
618 if (is_error_page(vcpu
->arch
.time_page
)) {
619 kvm_release_page_clean(vcpu
->arch
.time_page
);
620 vcpu
->arch
.time_page
= NULL
;
623 kvm_write_guest_time(vcpu
);
627 pr_unimpl(vcpu
, "unhandled wrmsr: 0x%x data %llx\n", msr
, data
);
632 EXPORT_SYMBOL_GPL(kvm_set_msr_common
);
636 * Reads an msr value (of 'msr_index') into 'pdata'.
637 * Returns 0 on success, non-0 otherwise.
638 * Assumes vcpu_load() was already called.
640 int kvm_get_msr(struct kvm_vcpu
*vcpu
, u32 msr_index
, u64
*pdata
)
642 return kvm_x86_ops
->get_msr(vcpu
, msr_index
, pdata
);
645 int kvm_get_msr_common(struct kvm_vcpu
*vcpu
, u32 msr
, u64
*pdata
)
650 case 0xc0010010: /* SYSCFG */
651 case 0xc0010015: /* HWCR */
652 case MSR_IA32_PLATFORM_ID
:
653 case MSR_IA32_P5_MC_ADDR
:
654 case MSR_IA32_P5_MC_TYPE
:
655 case MSR_IA32_MC0_CTL
:
656 case MSR_IA32_MCG_STATUS
:
657 case MSR_IA32_MCG_CAP
:
658 case MSR_IA32_MCG_CTL
:
659 case MSR_IA32_MC0_MISC
:
660 case MSR_IA32_MC0_MISC
+4:
661 case MSR_IA32_MC0_MISC
+8:
662 case MSR_IA32_MC0_MISC
+12:
663 case MSR_IA32_MC0_MISC
+16:
664 case MSR_IA32_UCODE_REV
:
665 case MSR_IA32_EBL_CR_POWERON
:
668 case 0x200 ... 0x2ff:
671 case 0xcd: /* fsb frequency */
674 case MSR_IA32_APICBASE
:
675 data
= kvm_get_apic_base(vcpu
);
677 case MSR_IA32_MISC_ENABLE
:
678 data
= vcpu
->arch
.ia32_misc_enable_msr
;
680 case MSR_IA32_PERF_STATUS
:
681 /* TSC increment by tick */
684 data
|= (((uint64_t)4ULL) << 40);
687 data
= vcpu
->arch
.shadow_efer
;
689 case MSR_KVM_WALL_CLOCK
:
690 data
= vcpu
->kvm
->arch
.wall_clock
;
692 case MSR_KVM_SYSTEM_TIME
:
693 data
= vcpu
->arch
.time
;
696 pr_unimpl(vcpu
, "unhandled rdmsr: 0x%x\n", msr
);
702 EXPORT_SYMBOL_GPL(kvm_get_msr_common
);
705 * Read or write a bunch of msrs. All parameters are kernel addresses.
707 * @return number of msrs set successfully.
709 static int __msr_io(struct kvm_vcpu
*vcpu
, struct kvm_msrs
*msrs
,
710 struct kvm_msr_entry
*entries
,
711 int (*do_msr
)(struct kvm_vcpu
*vcpu
,
712 unsigned index
, u64
*data
))
718 for (i
= 0; i
< msrs
->nmsrs
; ++i
)
719 if (do_msr(vcpu
, entries
[i
].index
, &entries
[i
].data
))
728 * Read or write a bunch of msrs. Parameters are user addresses.
730 * @return number of msrs set successfully.
732 static int msr_io(struct kvm_vcpu
*vcpu
, struct kvm_msrs __user
*user_msrs
,
733 int (*do_msr
)(struct kvm_vcpu
*vcpu
,
734 unsigned index
, u64
*data
),
737 struct kvm_msrs msrs
;
738 struct kvm_msr_entry
*entries
;
743 if (copy_from_user(&msrs
, user_msrs
, sizeof msrs
))
747 if (msrs
.nmsrs
>= MAX_IO_MSRS
)
751 size
= sizeof(struct kvm_msr_entry
) * msrs
.nmsrs
;
752 entries
= vmalloc(size
);
757 if (copy_from_user(entries
, user_msrs
->entries
, size
))
760 r
= n
= __msr_io(vcpu
, &msrs
, entries
, do_msr
);
765 if (writeback
&& copy_to_user(user_msrs
->entries
, entries
, size
))
777 * Make sure that a cpu that is being hot-unplugged does not have any vcpus
780 void decache_vcpus_on_cpu(int cpu
)
783 struct kvm_vcpu
*vcpu
;
786 spin_lock(&kvm_lock
);
787 list_for_each_entry(vm
, &vm_list
, vm_list
)
788 for (i
= 0; i
< KVM_MAX_VCPUS
; ++i
) {
793 * If the vcpu is locked, then it is running on some
794 * other cpu and therefore it is not cached on the
797 * If it's not locked, check the last cpu it executed
800 if (mutex_trylock(&vcpu
->mutex
)) {
801 if (vcpu
->cpu
== cpu
) {
802 kvm_x86_ops
->vcpu_decache(vcpu
);
805 mutex_unlock(&vcpu
->mutex
);
808 spin_unlock(&kvm_lock
);
811 int kvm_dev_ioctl_check_extension(long ext
)
816 case KVM_CAP_IRQCHIP
:
818 case KVM_CAP_MMU_SHADOW_CACHE_CONTROL
:
819 case KVM_CAP_USER_MEMORY
:
820 case KVM_CAP_SET_TSS_ADDR
:
821 case KVM_CAP_EXT_CPUID
:
822 case KVM_CAP_CLOCKSOURCE
:
824 case KVM_CAP_NOP_IO_DELAY
:
828 r
= !kvm_x86_ops
->cpu_has_accelerated_tpr();
830 case KVM_CAP_NR_VCPUS
:
833 case KVM_CAP_NR_MEMSLOTS
:
834 r
= KVM_MEMORY_SLOTS
;
847 long kvm_arch_dev_ioctl(struct file
*filp
,
848 unsigned int ioctl
, unsigned long arg
)
850 void __user
*argp
= (void __user
*)arg
;
854 case KVM_GET_MSR_INDEX_LIST
: {
855 struct kvm_msr_list __user
*user_msr_list
= argp
;
856 struct kvm_msr_list msr_list
;
860 if (copy_from_user(&msr_list
, user_msr_list
, sizeof msr_list
))
863 msr_list
.nmsrs
= num_msrs_to_save
+ ARRAY_SIZE(emulated_msrs
);
864 if (copy_to_user(user_msr_list
, &msr_list
, sizeof msr_list
))
867 if (n
< num_msrs_to_save
)
870 if (copy_to_user(user_msr_list
->indices
, &msrs_to_save
,
871 num_msrs_to_save
* sizeof(u32
)))
873 if (copy_to_user(user_msr_list
->indices
874 + num_msrs_to_save
* sizeof(u32
),
876 ARRAY_SIZE(emulated_msrs
) * sizeof(u32
)))
881 case KVM_GET_SUPPORTED_CPUID
: {
882 struct kvm_cpuid2 __user
*cpuid_arg
= argp
;
883 struct kvm_cpuid2 cpuid
;
886 if (copy_from_user(&cpuid
, cpuid_arg
, sizeof cpuid
))
888 r
= kvm_dev_ioctl_get_supported_cpuid(&cpuid
,
894 if (copy_to_user(cpuid_arg
, &cpuid
, sizeof cpuid
))
906 void kvm_arch_vcpu_load(struct kvm_vcpu
*vcpu
, int cpu
)
908 kvm_x86_ops
->vcpu_load(vcpu
, cpu
);
909 kvm_write_guest_time(vcpu
);
912 void kvm_arch_vcpu_put(struct kvm_vcpu
*vcpu
)
914 kvm_x86_ops
->vcpu_put(vcpu
);
915 kvm_put_guest_fpu(vcpu
);
918 static int is_efer_nx(void)
922 rdmsrl(MSR_EFER
, efer
);
923 return efer
& EFER_NX
;
926 static void cpuid_fix_nx_cap(struct kvm_vcpu
*vcpu
)
929 struct kvm_cpuid_entry2
*e
, *entry
;
932 for (i
= 0; i
< vcpu
->arch
.cpuid_nent
; ++i
) {
933 e
= &vcpu
->arch
.cpuid_entries
[i
];
934 if (e
->function
== 0x80000001) {
939 if (entry
&& (entry
->edx
& (1 << 20)) && !is_efer_nx()) {
940 entry
->edx
&= ~(1 << 20);
941 printk(KERN_INFO
"kvm: guest NX capability removed\n");
945 /* when an old userspace process fills a new kernel module */
946 static int kvm_vcpu_ioctl_set_cpuid(struct kvm_vcpu
*vcpu
,
947 struct kvm_cpuid
*cpuid
,
948 struct kvm_cpuid_entry __user
*entries
)
951 struct kvm_cpuid_entry
*cpuid_entries
;
954 if (cpuid
->nent
> KVM_MAX_CPUID_ENTRIES
)
957 cpuid_entries
= vmalloc(sizeof(struct kvm_cpuid_entry
) * cpuid
->nent
);
961 if (copy_from_user(cpuid_entries
, entries
,
962 cpuid
->nent
* sizeof(struct kvm_cpuid_entry
)))
964 for (i
= 0; i
< cpuid
->nent
; i
++) {
965 vcpu
->arch
.cpuid_entries
[i
].function
= cpuid_entries
[i
].function
;
966 vcpu
->arch
.cpuid_entries
[i
].eax
= cpuid_entries
[i
].eax
;
967 vcpu
->arch
.cpuid_entries
[i
].ebx
= cpuid_entries
[i
].ebx
;
968 vcpu
->arch
.cpuid_entries
[i
].ecx
= cpuid_entries
[i
].ecx
;
969 vcpu
->arch
.cpuid_entries
[i
].edx
= cpuid_entries
[i
].edx
;
970 vcpu
->arch
.cpuid_entries
[i
].index
= 0;
971 vcpu
->arch
.cpuid_entries
[i
].flags
= 0;
972 vcpu
->arch
.cpuid_entries
[i
].padding
[0] = 0;
973 vcpu
->arch
.cpuid_entries
[i
].padding
[1] = 0;
974 vcpu
->arch
.cpuid_entries
[i
].padding
[2] = 0;
976 vcpu
->arch
.cpuid_nent
= cpuid
->nent
;
977 cpuid_fix_nx_cap(vcpu
);
981 vfree(cpuid_entries
);
986 static int kvm_vcpu_ioctl_set_cpuid2(struct kvm_vcpu
*vcpu
,
987 struct kvm_cpuid2
*cpuid
,
988 struct kvm_cpuid_entry2 __user
*entries
)
993 if (cpuid
->nent
> KVM_MAX_CPUID_ENTRIES
)
996 if (copy_from_user(&vcpu
->arch
.cpuid_entries
, entries
,
997 cpuid
->nent
* sizeof(struct kvm_cpuid_entry2
)))
999 vcpu
->arch
.cpuid_nent
= cpuid
->nent
;
1006 static int kvm_vcpu_ioctl_get_cpuid2(struct kvm_vcpu
*vcpu
,
1007 struct kvm_cpuid2
*cpuid
,
1008 struct kvm_cpuid_entry2 __user
*entries
)
1013 if (cpuid
->nent
< vcpu
->arch
.cpuid_nent
)
1016 if (copy_to_user(entries
, &vcpu
->arch
.cpuid_entries
,
1017 vcpu
->arch
.cpuid_nent
* sizeof(struct kvm_cpuid_entry2
)))
1022 cpuid
->nent
= vcpu
->arch
.cpuid_nent
;
1026 static inline u32
bit(int bitno
)
1028 return 1 << (bitno
& 31);
1031 static void do_cpuid_1_ent(struct kvm_cpuid_entry2
*entry
, u32 function
,
1034 entry
->function
= function
;
1035 entry
->index
= index
;
1036 cpuid_count(entry
->function
, entry
->index
,
1037 &entry
->eax
, &entry
->ebx
, &entry
->ecx
, &entry
->edx
);
1041 static void do_cpuid_ent(struct kvm_cpuid_entry2
*entry
, u32 function
,
1042 u32 index
, int *nent
, int maxnent
)
1044 const u32 kvm_supported_word0_x86_features
= bit(X86_FEATURE_FPU
) |
1045 bit(X86_FEATURE_VME
) | bit(X86_FEATURE_DE
) |
1046 bit(X86_FEATURE_PSE
) | bit(X86_FEATURE_TSC
) |
1047 bit(X86_FEATURE_MSR
) | bit(X86_FEATURE_PAE
) |
1048 bit(X86_FEATURE_CX8
) | bit(X86_FEATURE_APIC
) |
1049 bit(X86_FEATURE_SEP
) | bit(X86_FEATURE_PGE
) |
1050 bit(X86_FEATURE_CMOV
) | bit(X86_FEATURE_PSE36
) |
1051 bit(X86_FEATURE_CLFLSH
) | bit(X86_FEATURE_MMX
) |
1052 bit(X86_FEATURE_FXSR
) | bit(X86_FEATURE_XMM
) |
1053 bit(X86_FEATURE_XMM2
) | bit(X86_FEATURE_SELFSNOOP
);
1054 const u32 kvm_supported_word1_x86_features
= bit(X86_FEATURE_FPU
) |
1055 bit(X86_FEATURE_VME
) | bit(X86_FEATURE_DE
) |
1056 bit(X86_FEATURE_PSE
) | bit(X86_FEATURE_TSC
) |
1057 bit(X86_FEATURE_MSR
) | bit(X86_FEATURE_PAE
) |
1058 bit(X86_FEATURE_CX8
) | bit(X86_FEATURE_APIC
) |
1059 bit(X86_FEATURE_PGE
) |
1060 bit(X86_FEATURE_CMOV
) | bit(X86_FEATURE_PSE36
) |
1061 bit(X86_FEATURE_MMX
) | bit(X86_FEATURE_FXSR
) |
1062 bit(X86_FEATURE_SYSCALL
) |
1063 (bit(X86_FEATURE_NX
) && is_efer_nx()) |
1064 #ifdef CONFIG_X86_64
1065 bit(X86_FEATURE_LM
) |
1067 bit(X86_FEATURE_MMXEXT
) |
1068 bit(X86_FEATURE_3DNOWEXT
) |
1069 bit(X86_FEATURE_3DNOW
);
1070 const u32 kvm_supported_word3_x86_features
=
1071 bit(X86_FEATURE_XMM3
) | bit(X86_FEATURE_CX16
);
1072 const u32 kvm_supported_word6_x86_features
=
1073 bit(X86_FEATURE_LAHF_LM
) | bit(X86_FEATURE_CMP_LEGACY
);
1075 /* all func 2 cpuid_count() should be called on the same cpu */
1077 do_cpuid_1_ent(entry
, function
, index
);
1082 entry
->eax
= min(entry
->eax
, (u32
)0xb);
1085 entry
->edx
&= kvm_supported_word0_x86_features
;
1086 entry
->ecx
&= kvm_supported_word3_x86_features
;
1088 /* function 2 entries are STATEFUL. That is, repeated cpuid commands
1089 * may return different values. This forces us to get_cpu() before
1090 * issuing the first command, and also to emulate this annoying behavior
1091 * in kvm_emulate_cpuid() using KVM_CPUID_FLAG_STATE_READ_NEXT */
1093 int t
, times
= entry
->eax
& 0xff;
1095 entry
->flags
|= KVM_CPUID_FLAG_STATEFUL_FUNC
;
1096 for (t
= 1; t
< times
&& *nent
< maxnent
; ++t
) {
1097 do_cpuid_1_ent(&entry
[t
], function
, 0);
1098 entry
[t
].flags
|= KVM_CPUID_FLAG_STATEFUL_FUNC
;
1103 /* function 4 and 0xb have additional index. */
1107 entry
->flags
|= KVM_CPUID_FLAG_SIGNIFCANT_INDEX
;
1108 /* read more entries until cache_type is zero */
1109 for (i
= 1; *nent
< maxnent
; ++i
) {
1110 cache_type
= entry
[i
- 1].eax
& 0x1f;
1113 do_cpuid_1_ent(&entry
[i
], function
, i
);
1115 KVM_CPUID_FLAG_SIGNIFCANT_INDEX
;
1123 entry
->flags
|= KVM_CPUID_FLAG_SIGNIFCANT_INDEX
;
1124 /* read more entries until level_type is zero */
1125 for (i
= 1; *nent
< maxnent
; ++i
) {
1126 level_type
= entry
[i
- 1].ecx
& 0xff;
1129 do_cpuid_1_ent(&entry
[i
], function
, i
);
1131 KVM_CPUID_FLAG_SIGNIFCANT_INDEX
;
1137 entry
->eax
= min(entry
->eax
, 0x8000001a);
1140 entry
->edx
&= kvm_supported_word1_x86_features
;
1141 entry
->ecx
&= kvm_supported_word6_x86_features
;
1147 static int kvm_dev_ioctl_get_supported_cpuid(struct kvm_cpuid2
*cpuid
,
1148 struct kvm_cpuid_entry2 __user
*entries
)
1150 struct kvm_cpuid_entry2
*cpuid_entries
;
1151 int limit
, nent
= 0, r
= -E2BIG
;
1154 if (cpuid
->nent
< 1)
1157 cpuid_entries
= vmalloc(sizeof(struct kvm_cpuid_entry2
) * cpuid
->nent
);
1161 do_cpuid_ent(&cpuid_entries
[0], 0, 0, &nent
, cpuid
->nent
);
1162 limit
= cpuid_entries
[0].eax
;
1163 for (func
= 1; func
<= limit
&& nent
< cpuid
->nent
; ++func
)
1164 do_cpuid_ent(&cpuid_entries
[nent
], func
, 0,
1165 &nent
, cpuid
->nent
);
1167 if (nent
>= cpuid
->nent
)
1170 do_cpuid_ent(&cpuid_entries
[nent
], 0x80000000, 0, &nent
, cpuid
->nent
);
1171 limit
= cpuid_entries
[nent
- 1].eax
;
1172 for (func
= 0x80000001; func
<= limit
&& nent
< cpuid
->nent
; ++func
)
1173 do_cpuid_ent(&cpuid_entries
[nent
], func
, 0,
1174 &nent
, cpuid
->nent
);
1176 if (copy_to_user(entries
, cpuid_entries
,
1177 nent
* sizeof(struct kvm_cpuid_entry2
)))
1183 vfree(cpuid_entries
);
1188 static int kvm_vcpu_ioctl_get_lapic(struct kvm_vcpu
*vcpu
,
1189 struct kvm_lapic_state
*s
)
1192 memcpy(s
->regs
, vcpu
->arch
.apic
->regs
, sizeof *s
);
1198 static int kvm_vcpu_ioctl_set_lapic(struct kvm_vcpu
*vcpu
,
1199 struct kvm_lapic_state
*s
)
1202 memcpy(vcpu
->arch
.apic
->regs
, s
->regs
, sizeof *s
);
1203 kvm_apic_post_state_restore(vcpu
);
1209 static int kvm_vcpu_ioctl_interrupt(struct kvm_vcpu
*vcpu
,
1210 struct kvm_interrupt
*irq
)
1212 if (irq
->irq
< 0 || irq
->irq
>= 256)
1214 if (irqchip_in_kernel(vcpu
->kvm
))
1218 set_bit(irq
->irq
, vcpu
->arch
.irq_pending
);
1219 set_bit(irq
->irq
/ BITS_PER_LONG
, &vcpu
->arch
.irq_summary
);
1226 static int vcpu_ioctl_tpr_access_reporting(struct kvm_vcpu
*vcpu
,
1227 struct kvm_tpr_access_ctl
*tac
)
1231 vcpu
->arch
.tpr_access_reporting
= !!tac
->enabled
;
1235 long kvm_arch_vcpu_ioctl(struct file
*filp
,
1236 unsigned int ioctl
, unsigned long arg
)
1238 struct kvm_vcpu
*vcpu
= filp
->private_data
;
1239 void __user
*argp
= (void __user
*)arg
;
1243 case KVM_GET_LAPIC
: {
1244 struct kvm_lapic_state lapic
;
1246 memset(&lapic
, 0, sizeof lapic
);
1247 r
= kvm_vcpu_ioctl_get_lapic(vcpu
, &lapic
);
1251 if (copy_to_user(argp
, &lapic
, sizeof lapic
))
1256 case KVM_SET_LAPIC
: {
1257 struct kvm_lapic_state lapic
;
1260 if (copy_from_user(&lapic
, argp
, sizeof lapic
))
1262 r
= kvm_vcpu_ioctl_set_lapic(vcpu
, &lapic
);;
1268 case KVM_INTERRUPT
: {
1269 struct kvm_interrupt irq
;
1272 if (copy_from_user(&irq
, argp
, sizeof irq
))
1274 r
= kvm_vcpu_ioctl_interrupt(vcpu
, &irq
);
1280 case KVM_SET_CPUID
: {
1281 struct kvm_cpuid __user
*cpuid_arg
= argp
;
1282 struct kvm_cpuid cpuid
;
1285 if (copy_from_user(&cpuid
, cpuid_arg
, sizeof cpuid
))
1287 r
= kvm_vcpu_ioctl_set_cpuid(vcpu
, &cpuid
, cpuid_arg
->entries
);
1292 case KVM_SET_CPUID2
: {
1293 struct kvm_cpuid2 __user
*cpuid_arg
= argp
;
1294 struct kvm_cpuid2 cpuid
;
1297 if (copy_from_user(&cpuid
, cpuid_arg
, sizeof cpuid
))
1299 r
= kvm_vcpu_ioctl_set_cpuid2(vcpu
, &cpuid
,
1300 cpuid_arg
->entries
);
1305 case KVM_GET_CPUID2
: {
1306 struct kvm_cpuid2 __user
*cpuid_arg
= argp
;
1307 struct kvm_cpuid2 cpuid
;
1310 if (copy_from_user(&cpuid
, cpuid_arg
, sizeof cpuid
))
1312 r
= kvm_vcpu_ioctl_get_cpuid2(vcpu
, &cpuid
,
1313 cpuid_arg
->entries
);
1317 if (copy_to_user(cpuid_arg
, &cpuid
, sizeof cpuid
))
1323 r
= msr_io(vcpu
, argp
, kvm_get_msr
, 1);
1326 r
= msr_io(vcpu
, argp
, do_set_msr
, 0);
1328 case KVM_TPR_ACCESS_REPORTING
: {
1329 struct kvm_tpr_access_ctl tac
;
1332 if (copy_from_user(&tac
, argp
, sizeof tac
))
1334 r
= vcpu_ioctl_tpr_access_reporting(vcpu
, &tac
);
1338 if (copy_to_user(argp
, &tac
, sizeof tac
))
1343 case KVM_SET_VAPIC_ADDR
: {
1344 struct kvm_vapic_addr va
;
1347 if (!irqchip_in_kernel(vcpu
->kvm
))
1350 if (copy_from_user(&va
, argp
, sizeof va
))
1353 kvm_lapic_set_vapic_addr(vcpu
, va
.vapic_addr
);
1363 static int kvm_vm_ioctl_set_tss_addr(struct kvm
*kvm
, unsigned long addr
)
1367 if (addr
> (unsigned int)(-3 * PAGE_SIZE
))
1369 ret
= kvm_x86_ops
->set_tss_addr(kvm
, addr
);
1373 static int kvm_vm_ioctl_set_nr_mmu_pages(struct kvm
*kvm
,
1374 u32 kvm_nr_mmu_pages
)
1376 if (kvm_nr_mmu_pages
< KVM_MIN_ALLOC_MMU_PAGES
)
1379 down_write(&kvm
->slots_lock
);
1381 kvm_mmu_change_mmu_pages(kvm
, kvm_nr_mmu_pages
);
1382 kvm
->arch
.n_requested_mmu_pages
= kvm_nr_mmu_pages
;
1384 up_write(&kvm
->slots_lock
);
1388 static int kvm_vm_ioctl_get_nr_mmu_pages(struct kvm
*kvm
)
1390 return kvm
->arch
.n_alloc_mmu_pages
;
1393 gfn_t
unalias_gfn(struct kvm
*kvm
, gfn_t gfn
)
1396 struct kvm_mem_alias
*alias
;
1398 for (i
= 0; i
< kvm
->arch
.naliases
; ++i
) {
1399 alias
= &kvm
->arch
.aliases
[i
];
1400 if (gfn
>= alias
->base_gfn
1401 && gfn
< alias
->base_gfn
+ alias
->npages
)
1402 return alias
->target_gfn
+ gfn
- alias
->base_gfn
;
1408 * Set a new alias region. Aliases map a portion of physical memory into
1409 * another portion. This is useful for memory windows, for example the PC
1412 static int kvm_vm_ioctl_set_memory_alias(struct kvm
*kvm
,
1413 struct kvm_memory_alias
*alias
)
1416 struct kvm_mem_alias
*p
;
1419 /* General sanity checks */
1420 if (alias
->memory_size
& (PAGE_SIZE
- 1))
1422 if (alias
->guest_phys_addr
& (PAGE_SIZE
- 1))
1424 if (alias
->slot
>= KVM_ALIAS_SLOTS
)
1426 if (alias
->guest_phys_addr
+ alias
->memory_size
1427 < alias
->guest_phys_addr
)
1429 if (alias
->target_phys_addr
+ alias
->memory_size
1430 < alias
->target_phys_addr
)
1433 down_write(&kvm
->slots_lock
);
1435 p
= &kvm
->arch
.aliases
[alias
->slot
];
1436 p
->base_gfn
= alias
->guest_phys_addr
>> PAGE_SHIFT
;
1437 p
->npages
= alias
->memory_size
>> PAGE_SHIFT
;
1438 p
->target_gfn
= alias
->target_phys_addr
>> PAGE_SHIFT
;
1440 for (n
= KVM_ALIAS_SLOTS
; n
> 0; --n
)
1441 if (kvm
->arch
.aliases
[n
- 1].npages
)
1443 kvm
->arch
.naliases
= n
;
1445 kvm_mmu_zap_all(kvm
);
1447 up_write(&kvm
->slots_lock
);
1455 static int kvm_vm_ioctl_get_irqchip(struct kvm
*kvm
, struct kvm_irqchip
*chip
)
1460 switch (chip
->chip_id
) {
1461 case KVM_IRQCHIP_PIC_MASTER
:
1462 memcpy(&chip
->chip
.pic
,
1463 &pic_irqchip(kvm
)->pics
[0],
1464 sizeof(struct kvm_pic_state
));
1466 case KVM_IRQCHIP_PIC_SLAVE
:
1467 memcpy(&chip
->chip
.pic
,
1468 &pic_irqchip(kvm
)->pics
[1],
1469 sizeof(struct kvm_pic_state
));
1471 case KVM_IRQCHIP_IOAPIC
:
1472 memcpy(&chip
->chip
.ioapic
,
1473 ioapic_irqchip(kvm
),
1474 sizeof(struct kvm_ioapic_state
));
1483 static int kvm_vm_ioctl_set_irqchip(struct kvm
*kvm
, struct kvm_irqchip
*chip
)
1488 switch (chip
->chip_id
) {
1489 case KVM_IRQCHIP_PIC_MASTER
:
1490 memcpy(&pic_irqchip(kvm
)->pics
[0],
1492 sizeof(struct kvm_pic_state
));
1494 case KVM_IRQCHIP_PIC_SLAVE
:
1495 memcpy(&pic_irqchip(kvm
)->pics
[1],
1497 sizeof(struct kvm_pic_state
));
1499 case KVM_IRQCHIP_IOAPIC
:
1500 memcpy(ioapic_irqchip(kvm
),
1502 sizeof(struct kvm_ioapic_state
));
1508 kvm_pic_update_irq(pic_irqchip(kvm
));
1512 static int kvm_vm_ioctl_get_pit(struct kvm
*kvm
, struct kvm_pit_state
*ps
)
1516 memcpy(ps
, &kvm
->arch
.vpit
->pit_state
, sizeof(struct kvm_pit_state
));
1520 static int kvm_vm_ioctl_set_pit(struct kvm
*kvm
, struct kvm_pit_state
*ps
)
1524 memcpy(&kvm
->arch
.vpit
->pit_state
, ps
, sizeof(struct kvm_pit_state
));
1525 kvm_pit_load_count(kvm
, 0, ps
->channels
[0].count
);
1530 * Get (and clear) the dirty memory log for a memory slot.
1532 int kvm_vm_ioctl_get_dirty_log(struct kvm
*kvm
,
1533 struct kvm_dirty_log
*log
)
1537 struct kvm_memory_slot
*memslot
;
1540 down_write(&kvm
->slots_lock
);
1542 r
= kvm_get_dirty_log(kvm
, log
, &is_dirty
);
1546 /* If nothing is dirty, don't bother messing with page tables. */
1548 kvm_mmu_slot_remove_write_access(kvm
, log
->slot
);
1549 kvm_flush_remote_tlbs(kvm
);
1550 memslot
= &kvm
->memslots
[log
->slot
];
1551 n
= ALIGN(memslot
->npages
, BITS_PER_LONG
) / 8;
1552 memset(memslot
->dirty_bitmap
, 0, n
);
1556 up_write(&kvm
->slots_lock
);
1560 long kvm_arch_vm_ioctl(struct file
*filp
,
1561 unsigned int ioctl
, unsigned long arg
)
1563 struct kvm
*kvm
= filp
->private_data
;
1564 void __user
*argp
= (void __user
*)arg
;
1568 case KVM_SET_TSS_ADDR
:
1569 r
= kvm_vm_ioctl_set_tss_addr(kvm
, arg
);
1573 case KVM_SET_MEMORY_REGION
: {
1574 struct kvm_memory_region kvm_mem
;
1575 struct kvm_userspace_memory_region kvm_userspace_mem
;
1578 if (copy_from_user(&kvm_mem
, argp
, sizeof kvm_mem
))
1580 kvm_userspace_mem
.slot
= kvm_mem
.slot
;
1581 kvm_userspace_mem
.flags
= kvm_mem
.flags
;
1582 kvm_userspace_mem
.guest_phys_addr
= kvm_mem
.guest_phys_addr
;
1583 kvm_userspace_mem
.memory_size
= kvm_mem
.memory_size
;
1584 r
= kvm_vm_ioctl_set_memory_region(kvm
, &kvm_userspace_mem
, 0);
1589 case KVM_SET_NR_MMU_PAGES
:
1590 r
= kvm_vm_ioctl_set_nr_mmu_pages(kvm
, arg
);
1594 case KVM_GET_NR_MMU_PAGES
:
1595 r
= kvm_vm_ioctl_get_nr_mmu_pages(kvm
);
1597 case KVM_SET_MEMORY_ALIAS
: {
1598 struct kvm_memory_alias alias
;
1601 if (copy_from_user(&alias
, argp
, sizeof alias
))
1603 r
= kvm_vm_ioctl_set_memory_alias(kvm
, &alias
);
1608 case KVM_CREATE_IRQCHIP
:
1610 kvm
->arch
.vpic
= kvm_create_pic(kvm
);
1611 if (kvm
->arch
.vpic
) {
1612 r
= kvm_ioapic_init(kvm
);
1614 kfree(kvm
->arch
.vpic
);
1615 kvm
->arch
.vpic
= NULL
;
1621 case KVM_CREATE_PIT
:
1623 kvm
->arch
.vpit
= kvm_create_pit(kvm
);
1627 case KVM_IRQ_LINE
: {
1628 struct kvm_irq_level irq_event
;
1631 if (copy_from_user(&irq_event
, argp
, sizeof irq_event
))
1633 if (irqchip_in_kernel(kvm
)) {
1634 mutex_lock(&kvm
->lock
);
1635 if (irq_event
.irq
< 16)
1636 kvm_pic_set_irq(pic_irqchip(kvm
),
1639 kvm_ioapic_set_irq(kvm
->arch
.vioapic
,
1642 mutex_unlock(&kvm
->lock
);
1647 case KVM_GET_IRQCHIP
: {
1648 /* 0: PIC master, 1: PIC slave, 2: IOAPIC */
1649 struct kvm_irqchip chip
;
1652 if (copy_from_user(&chip
, argp
, sizeof chip
))
1655 if (!irqchip_in_kernel(kvm
))
1657 r
= kvm_vm_ioctl_get_irqchip(kvm
, &chip
);
1661 if (copy_to_user(argp
, &chip
, sizeof chip
))
1666 case KVM_SET_IRQCHIP
: {
1667 /* 0: PIC master, 1: PIC slave, 2: IOAPIC */
1668 struct kvm_irqchip chip
;
1671 if (copy_from_user(&chip
, argp
, sizeof chip
))
1674 if (!irqchip_in_kernel(kvm
))
1676 r
= kvm_vm_ioctl_set_irqchip(kvm
, &chip
);
1683 struct kvm_pit_state ps
;
1685 if (copy_from_user(&ps
, argp
, sizeof ps
))
1688 if (!kvm
->arch
.vpit
)
1690 r
= kvm_vm_ioctl_get_pit(kvm
, &ps
);
1694 if (copy_to_user(argp
, &ps
, sizeof ps
))
1700 struct kvm_pit_state ps
;
1702 if (copy_from_user(&ps
, argp
, sizeof ps
))
1705 if (!kvm
->arch
.vpit
)
1707 r
= kvm_vm_ioctl_set_pit(kvm
, &ps
);
1720 static void kvm_init_msr_list(void)
1725 for (i
= j
= 0; i
< ARRAY_SIZE(msrs_to_save
); i
++) {
1726 if (rdmsr_safe(msrs_to_save
[i
], &dummy
[0], &dummy
[1]) < 0)
1729 msrs_to_save
[j
] = msrs_to_save
[i
];
1732 num_msrs_to_save
= j
;
1736 * Only apic need an MMIO device hook, so shortcut now..
1738 static struct kvm_io_device
*vcpu_find_pervcpu_dev(struct kvm_vcpu
*vcpu
,
1741 struct kvm_io_device
*dev
;
1743 if (vcpu
->arch
.apic
) {
1744 dev
= &vcpu
->arch
.apic
->dev
;
1745 if (dev
->in_range(dev
, addr
))
1752 static struct kvm_io_device
*vcpu_find_mmio_dev(struct kvm_vcpu
*vcpu
,
1755 struct kvm_io_device
*dev
;
1757 dev
= vcpu_find_pervcpu_dev(vcpu
, addr
);
1759 dev
= kvm_io_bus_find_dev(&vcpu
->kvm
->mmio_bus
, addr
);
1763 int emulator_read_std(unsigned long addr
,
1766 struct kvm_vcpu
*vcpu
)
1769 int r
= X86EMUL_CONTINUE
;
1771 down_read(&vcpu
->kvm
->slots_lock
);
1773 gpa_t gpa
= vcpu
->arch
.mmu
.gva_to_gpa(vcpu
, addr
);
1774 unsigned offset
= addr
& (PAGE_SIZE
-1);
1775 unsigned tocopy
= min(bytes
, (unsigned)PAGE_SIZE
- offset
);
1778 if (gpa
== UNMAPPED_GVA
) {
1779 r
= X86EMUL_PROPAGATE_FAULT
;
1782 ret
= kvm_read_guest(vcpu
->kvm
, gpa
, data
, tocopy
);
1784 r
= X86EMUL_UNHANDLEABLE
;
1793 up_read(&vcpu
->kvm
->slots_lock
);
1796 EXPORT_SYMBOL_GPL(emulator_read_std
);
1798 static int emulator_read_emulated(unsigned long addr
,
1801 struct kvm_vcpu
*vcpu
)
1803 struct kvm_io_device
*mmio_dev
;
1806 if (vcpu
->mmio_read_completed
) {
1807 memcpy(val
, vcpu
->mmio_data
, bytes
);
1808 vcpu
->mmio_read_completed
= 0;
1809 return X86EMUL_CONTINUE
;
1812 down_read(&vcpu
->kvm
->slots_lock
);
1813 gpa
= vcpu
->arch
.mmu
.gva_to_gpa(vcpu
, addr
);
1814 up_read(&vcpu
->kvm
->slots_lock
);
1816 /* For APIC access vmexit */
1817 if ((gpa
& PAGE_MASK
) == APIC_DEFAULT_PHYS_BASE
)
1820 if (emulator_read_std(addr
, val
, bytes
, vcpu
)
1821 == X86EMUL_CONTINUE
)
1822 return X86EMUL_CONTINUE
;
1823 if (gpa
== UNMAPPED_GVA
)
1824 return X86EMUL_PROPAGATE_FAULT
;
1828 * Is this MMIO handled locally?
1830 mutex_lock(&vcpu
->kvm
->lock
);
1831 mmio_dev
= vcpu_find_mmio_dev(vcpu
, gpa
);
1833 kvm_iodevice_read(mmio_dev
, gpa
, bytes
, val
);
1834 mutex_unlock(&vcpu
->kvm
->lock
);
1835 return X86EMUL_CONTINUE
;
1837 mutex_unlock(&vcpu
->kvm
->lock
);
1839 vcpu
->mmio_needed
= 1;
1840 vcpu
->mmio_phys_addr
= gpa
;
1841 vcpu
->mmio_size
= bytes
;
1842 vcpu
->mmio_is_write
= 0;
1844 return X86EMUL_UNHANDLEABLE
;
1847 int __emulator_write_phys(struct kvm_vcpu
*vcpu
, gpa_t gpa
,
1848 const void *val
, int bytes
)
1852 ret
= kvm_write_guest(vcpu
->kvm
, gpa
, val
, bytes
);
1855 kvm_mmu_pte_write(vcpu
, gpa
, val
, bytes
);
1859 static int emulator_write_phys(struct kvm_vcpu
*vcpu
, gpa_t gpa
,
1860 const void *val
, int bytes
)
1864 down_read(&vcpu
->kvm
->slots_lock
);
1865 ret
=__emulator_write_phys(vcpu
, gpa
, val
, bytes
);
1866 up_read(&vcpu
->kvm
->slots_lock
);
1870 static int emulator_write_emulated_onepage(unsigned long addr
,
1873 struct kvm_vcpu
*vcpu
)
1875 struct kvm_io_device
*mmio_dev
;
1878 down_read(&vcpu
->kvm
->slots_lock
);
1879 gpa
= vcpu
->arch
.mmu
.gva_to_gpa(vcpu
, addr
);
1880 up_read(&vcpu
->kvm
->slots_lock
);
1882 if (gpa
== UNMAPPED_GVA
) {
1883 kvm_inject_page_fault(vcpu
, addr
, 2);
1884 return X86EMUL_PROPAGATE_FAULT
;
1887 /* For APIC access vmexit */
1888 if ((gpa
& PAGE_MASK
) == APIC_DEFAULT_PHYS_BASE
)
1891 if (emulator_write_phys(vcpu
, gpa
, val
, bytes
))
1892 return X86EMUL_CONTINUE
;
1896 * Is this MMIO handled locally?
1898 mutex_lock(&vcpu
->kvm
->lock
);
1899 mmio_dev
= vcpu_find_mmio_dev(vcpu
, gpa
);
1901 kvm_iodevice_write(mmio_dev
, gpa
, bytes
, val
);
1902 mutex_unlock(&vcpu
->kvm
->lock
);
1903 return X86EMUL_CONTINUE
;
1905 mutex_unlock(&vcpu
->kvm
->lock
);
1907 vcpu
->mmio_needed
= 1;
1908 vcpu
->mmio_phys_addr
= gpa
;
1909 vcpu
->mmio_size
= bytes
;
1910 vcpu
->mmio_is_write
= 1;
1911 memcpy(vcpu
->mmio_data
, val
, bytes
);
1913 return X86EMUL_CONTINUE
;
1916 int emulator_write_emulated(unsigned long addr
,
1919 struct kvm_vcpu
*vcpu
)
1921 /* Crossing a page boundary? */
1922 if (((addr
+ bytes
- 1) ^ addr
) & PAGE_MASK
) {
1925 now
= -addr
& ~PAGE_MASK
;
1926 rc
= emulator_write_emulated_onepage(addr
, val
, now
, vcpu
);
1927 if (rc
!= X86EMUL_CONTINUE
)
1933 return emulator_write_emulated_onepage(addr
, val
, bytes
, vcpu
);
1935 EXPORT_SYMBOL_GPL(emulator_write_emulated
);
1937 static int emulator_cmpxchg_emulated(unsigned long addr
,
1941 struct kvm_vcpu
*vcpu
)
1943 static int reported
;
1947 printk(KERN_WARNING
"kvm: emulating exchange as write\n");
1949 #ifndef CONFIG_X86_64
1950 /* guests cmpxchg8b have to be emulated atomically */
1957 down_read(&vcpu
->kvm
->slots_lock
);
1958 gpa
= vcpu
->arch
.mmu
.gva_to_gpa(vcpu
, addr
);
1960 if (gpa
== UNMAPPED_GVA
||
1961 (gpa
& PAGE_MASK
) == APIC_DEFAULT_PHYS_BASE
)
1964 if (((gpa
+ bytes
- 1) & PAGE_MASK
) != (gpa
& PAGE_MASK
))
1969 down_read(¤t
->mm
->mmap_sem
);
1970 page
= gfn_to_page(vcpu
->kvm
, gpa
>> PAGE_SHIFT
);
1971 up_read(¤t
->mm
->mmap_sem
);
1973 kaddr
= kmap_atomic(page
, KM_USER0
);
1974 set_64bit((u64
*)(kaddr
+ offset_in_page(gpa
)), val
);
1975 kunmap_atomic(kaddr
, KM_USER0
);
1976 kvm_release_page_dirty(page
);
1978 up_read(&vcpu
->kvm
->slots_lock
);
1982 return emulator_write_emulated(addr
, new, bytes
, vcpu
);
1985 static unsigned long get_segment_base(struct kvm_vcpu
*vcpu
, int seg
)
1987 return kvm_x86_ops
->get_segment_base(vcpu
, seg
);
1990 int emulate_invlpg(struct kvm_vcpu
*vcpu
, gva_t address
)
1992 return X86EMUL_CONTINUE
;
1995 int emulate_clts(struct kvm_vcpu
*vcpu
)
1997 kvm_x86_ops
->set_cr0(vcpu
, vcpu
->arch
.cr0
& ~X86_CR0_TS
);
1998 return X86EMUL_CONTINUE
;
2001 int emulator_get_dr(struct x86_emulate_ctxt
*ctxt
, int dr
, unsigned long *dest
)
2003 struct kvm_vcpu
*vcpu
= ctxt
->vcpu
;
2007 *dest
= kvm_x86_ops
->get_dr(vcpu
, dr
);
2008 return X86EMUL_CONTINUE
;
2010 pr_unimpl(vcpu
, "%s: unexpected dr %u\n", __func__
, dr
);
2011 return X86EMUL_UNHANDLEABLE
;
2015 int emulator_set_dr(struct x86_emulate_ctxt
*ctxt
, int dr
, unsigned long value
)
2017 unsigned long mask
= (ctxt
->mode
== X86EMUL_MODE_PROT64
) ? ~0ULL : ~0U;
2020 kvm_x86_ops
->set_dr(ctxt
->vcpu
, dr
, value
& mask
, &exception
);
2022 /* FIXME: better handling */
2023 return X86EMUL_UNHANDLEABLE
;
2025 return X86EMUL_CONTINUE
;
2028 void kvm_report_emulation_failure(struct kvm_vcpu
*vcpu
, const char *context
)
2030 static int reported
;
2032 unsigned long rip
= vcpu
->arch
.rip
;
2033 unsigned long rip_linear
;
2035 rip_linear
= rip
+ get_segment_base(vcpu
, VCPU_SREG_CS
);
2040 emulator_read_std(rip_linear
, (void *)opcodes
, 4, vcpu
);
2042 printk(KERN_ERR
"emulation failed (%s) rip %lx %02x %02x %02x %02x\n",
2043 context
, rip
, opcodes
[0], opcodes
[1], opcodes
[2], opcodes
[3]);
2046 EXPORT_SYMBOL_GPL(kvm_report_emulation_failure
);
2048 static struct x86_emulate_ops emulate_ops
= {
2049 .read_std
= emulator_read_std
,
2050 .read_emulated
= emulator_read_emulated
,
2051 .write_emulated
= emulator_write_emulated
,
2052 .cmpxchg_emulated
= emulator_cmpxchg_emulated
,
2055 int emulate_instruction(struct kvm_vcpu
*vcpu
,
2056 struct kvm_run
*run
,
2062 struct decode_cache
*c
;
2064 vcpu
->arch
.mmio_fault_cr2
= cr2
;
2065 kvm_x86_ops
->cache_regs(vcpu
);
2067 vcpu
->mmio_is_write
= 0;
2068 vcpu
->arch
.pio
.string
= 0;
2070 if (!(emulation_type
& EMULTYPE_NO_DECODE
)) {
2072 kvm_x86_ops
->get_cs_db_l_bits(vcpu
, &cs_db
, &cs_l
);
2074 vcpu
->arch
.emulate_ctxt
.vcpu
= vcpu
;
2075 vcpu
->arch
.emulate_ctxt
.eflags
= kvm_x86_ops
->get_rflags(vcpu
);
2076 vcpu
->arch
.emulate_ctxt
.mode
=
2077 (vcpu
->arch
.emulate_ctxt
.eflags
& X86_EFLAGS_VM
)
2078 ? X86EMUL_MODE_REAL
: cs_l
2079 ? X86EMUL_MODE_PROT64
: cs_db
2080 ? X86EMUL_MODE_PROT32
: X86EMUL_MODE_PROT16
;
2082 if (vcpu
->arch
.emulate_ctxt
.mode
== X86EMUL_MODE_PROT64
) {
2083 vcpu
->arch
.emulate_ctxt
.cs_base
= 0;
2084 vcpu
->arch
.emulate_ctxt
.ds_base
= 0;
2085 vcpu
->arch
.emulate_ctxt
.es_base
= 0;
2086 vcpu
->arch
.emulate_ctxt
.ss_base
= 0;
2088 vcpu
->arch
.emulate_ctxt
.cs_base
=
2089 get_segment_base(vcpu
, VCPU_SREG_CS
);
2090 vcpu
->arch
.emulate_ctxt
.ds_base
=
2091 get_segment_base(vcpu
, VCPU_SREG_DS
);
2092 vcpu
->arch
.emulate_ctxt
.es_base
=
2093 get_segment_base(vcpu
, VCPU_SREG_ES
);
2094 vcpu
->arch
.emulate_ctxt
.ss_base
=
2095 get_segment_base(vcpu
, VCPU_SREG_SS
);
2098 vcpu
->arch
.emulate_ctxt
.gs_base
=
2099 get_segment_base(vcpu
, VCPU_SREG_GS
);
2100 vcpu
->arch
.emulate_ctxt
.fs_base
=
2101 get_segment_base(vcpu
, VCPU_SREG_FS
);
2103 r
= x86_decode_insn(&vcpu
->arch
.emulate_ctxt
, &emulate_ops
);
2105 /* Reject the instructions other than VMCALL/VMMCALL when
2106 * try to emulate invalid opcode */
2107 c
= &vcpu
->arch
.emulate_ctxt
.decode
;
2108 if ((emulation_type
& EMULTYPE_TRAP_UD
) &&
2109 (!(c
->twobyte
&& c
->b
== 0x01 &&
2110 (c
->modrm_reg
== 0 || c
->modrm_reg
== 3) &&
2111 c
->modrm_mod
== 3 && c
->modrm_rm
== 1)))
2112 return EMULATE_FAIL
;
2114 ++vcpu
->stat
.insn_emulation
;
2116 ++vcpu
->stat
.insn_emulation_fail
;
2117 if (kvm_mmu_unprotect_page_virt(vcpu
, cr2
))
2118 return EMULATE_DONE
;
2119 return EMULATE_FAIL
;
2123 r
= x86_emulate_insn(&vcpu
->arch
.emulate_ctxt
, &emulate_ops
);
2125 if (vcpu
->arch
.pio
.string
)
2126 return EMULATE_DO_MMIO
;
2128 if ((r
|| vcpu
->mmio_is_write
) && run
) {
2129 run
->exit_reason
= KVM_EXIT_MMIO
;
2130 run
->mmio
.phys_addr
= vcpu
->mmio_phys_addr
;
2131 memcpy(run
->mmio
.data
, vcpu
->mmio_data
, 8);
2132 run
->mmio
.len
= vcpu
->mmio_size
;
2133 run
->mmio
.is_write
= vcpu
->mmio_is_write
;
2137 if (kvm_mmu_unprotect_page_virt(vcpu
, cr2
))
2138 return EMULATE_DONE
;
2139 if (!vcpu
->mmio_needed
) {
2140 kvm_report_emulation_failure(vcpu
, "mmio");
2141 return EMULATE_FAIL
;
2143 return EMULATE_DO_MMIO
;
2146 kvm_x86_ops
->decache_regs(vcpu
);
2147 kvm_x86_ops
->set_rflags(vcpu
, vcpu
->arch
.emulate_ctxt
.eflags
);
2149 if (vcpu
->mmio_is_write
) {
2150 vcpu
->mmio_needed
= 0;
2151 return EMULATE_DO_MMIO
;
2154 return EMULATE_DONE
;
2156 EXPORT_SYMBOL_GPL(emulate_instruction
);
2158 static void free_pio_guest_pages(struct kvm_vcpu
*vcpu
)
2162 for (i
= 0; i
< ARRAY_SIZE(vcpu
->arch
.pio
.guest_pages
); ++i
)
2163 if (vcpu
->arch
.pio
.guest_pages
[i
]) {
2164 kvm_release_page_dirty(vcpu
->arch
.pio
.guest_pages
[i
]);
2165 vcpu
->arch
.pio
.guest_pages
[i
] = NULL
;
2169 static int pio_copy_data(struct kvm_vcpu
*vcpu
)
2171 void *p
= vcpu
->arch
.pio_data
;
2174 int nr_pages
= vcpu
->arch
.pio
.guest_pages
[1] ? 2 : 1;
2176 q
= vmap(vcpu
->arch
.pio
.guest_pages
, nr_pages
, VM_READ
|VM_WRITE
,
2179 free_pio_guest_pages(vcpu
);
2182 q
+= vcpu
->arch
.pio
.guest_page_offset
;
2183 bytes
= vcpu
->arch
.pio
.size
* vcpu
->arch
.pio
.cur_count
;
2184 if (vcpu
->arch
.pio
.in
)
2185 memcpy(q
, p
, bytes
);
2187 memcpy(p
, q
, bytes
);
2188 q
-= vcpu
->arch
.pio
.guest_page_offset
;
2190 free_pio_guest_pages(vcpu
);
2194 int complete_pio(struct kvm_vcpu
*vcpu
)
2196 struct kvm_pio_request
*io
= &vcpu
->arch
.pio
;
2200 kvm_x86_ops
->cache_regs(vcpu
);
2204 memcpy(&vcpu
->arch
.regs
[VCPU_REGS_RAX
], vcpu
->arch
.pio_data
,
2208 r
= pio_copy_data(vcpu
);
2210 kvm_x86_ops
->cache_regs(vcpu
);
2217 delta
*= io
->cur_count
;
2219 * The size of the register should really depend on
2220 * current address size.
2222 vcpu
->arch
.regs
[VCPU_REGS_RCX
] -= delta
;
2228 vcpu
->arch
.regs
[VCPU_REGS_RDI
] += delta
;
2230 vcpu
->arch
.regs
[VCPU_REGS_RSI
] += delta
;
2233 kvm_x86_ops
->decache_regs(vcpu
);
2235 io
->count
-= io
->cur_count
;
2241 static void kernel_pio(struct kvm_io_device
*pio_dev
,
2242 struct kvm_vcpu
*vcpu
,
2245 /* TODO: String I/O for in kernel device */
2247 mutex_lock(&vcpu
->kvm
->lock
);
2248 if (vcpu
->arch
.pio
.in
)
2249 kvm_iodevice_read(pio_dev
, vcpu
->arch
.pio
.port
,
2250 vcpu
->arch
.pio
.size
,
2253 kvm_iodevice_write(pio_dev
, vcpu
->arch
.pio
.port
,
2254 vcpu
->arch
.pio
.size
,
2256 mutex_unlock(&vcpu
->kvm
->lock
);
2259 static void pio_string_write(struct kvm_io_device
*pio_dev
,
2260 struct kvm_vcpu
*vcpu
)
2262 struct kvm_pio_request
*io
= &vcpu
->arch
.pio
;
2263 void *pd
= vcpu
->arch
.pio_data
;
2266 mutex_lock(&vcpu
->kvm
->lock
);
2267 for (i
= 0; i
< io
->cur_count
; i
++) {
2268 kvm_iodevice_write(pio_dev
, io
->port
,
2273 mutex_unlock(&vcpu
->kvm
->lock
);
2276 static struct kvm_io_device
*vcpu_find_pio_dev(struct kvm_vcpu
*vcpu
,
2279 return kvm_io_bus_find_dev(&vcpu
->kvm
->pio_bus
, addr
);
2282 int kvm_emulate_pio(struct kvm_vcpu
*vcpu
, struct kvm_run
*run
, int in
,
2283 int size
, unsigned port
)
2285 struct kvm_io_device
*pio_dev
;
2287 vcpu
->run
->exit_reason
= KVM_EXIT_IO
;
2288 vcpu
->run
->io
.direction
= in
? KVM_EXIT_IO_IN
: KVM_EXIT_IO_OUT
;
2289 vcpu
->run
->io
.size
= vcpu
->arch
.pio
.size
= size
;
2290 vcpu
->run
->io
.data_offset
= KVM_PIO_PAGE_OFFSET
* PAGE_SIZE
;
2291 vcpu
->run
->io
.count
= vcpu
->arch
.pio
.count
= vcpu
->arch
.pio
.cur_count
= 1;
2292 vcpu
->run
->io
.port
= vcpu
->arch
.pio
.port
= port
;
2293 vcpu
->arch
.pio
.in
= in
;
2294 vcpu
->arch
.pio
.string
= 0;
2295 vcpu
->arch
.pio
.down
= 0;
2296 vcpu
->arch
.pio
.guest_page_offset
= 0;
2297 vcpu
->arch
.pio
.rep
= 0;
2299 kvm_x86_ops
->cache_regs(vcpu
);
2300 memcpy(vcpu
->arch
.pio_data
, &vcpu
->arch
.regs
[VCPU_REGS_RAX
], 4);
2301 kvm_x86_ops
->decache_regs(vcpu
);
2303 kvm_x86_ops
->skip_emulated_instruction(vcpu
);
2305 pio_dev
= vcpu_find_pio_dev(vcpu
, port
);
2307 kernel_pio(pio_dev
, vcpu
, vcpu
->arch
.pio_data
);
2313 EXPORT_SYMBOL_GPL(kvm_emulate_pio
);
2315 int kvm_emulate_pio_string(struct kvm_vcpu
*vcpu
, struct kvm_run
*run
, int in
,
2316 int size
, unsigned long count
, int down
,
2317 gva_t address
, int rep
, unsigned port
)
2319 unsigned now
, in_page
;
2323 struct kvm_io_device
*pio_dev
;
2325 vcpu
->run
->exit_reason
= KVM_EXIT_IO
;
2326 vcpu
->run
->io
.direction
= in
? KVM_EXIT_IO_IN
: KVM_EXIT_IO_OUT
;
2327 vcpu
->run
->io
.size
= vcpu
->arch
.pio
.size
= size
;
2328 vcpu
->run
->io
.data_offset
= KVM_PIO_PAGE_OFFSET
* PAGE_SIZE
;
2329 vcpu
->run
->io
.count
= vcpu
->arch
.pio
.count
= vcpu
->arch
.pio
.cur_count
= count
;
2330 vcpu
->run
->io
.port
= vcpu
->arch
.pio
.port
= port
;
2331 vcpu
->arch
.pio
.in
= in
;
2332 vcpu
->arch
.pio
.string
= 1;
2333 vcpu
->arch
.pio
.down
= down
;
2334 vcpu
->arch
.pio
.guest_page_offset
= offset_in_page(address
);
2335 vcpu
->arch
.pio
.rep
= rep
;
2338 kvm_x86_ops
->skip_emulated_instruction(vcpu
);
2343 in_page
= PAGE_SIZE
- offset_in_page(address
);
2345 in_page
= offset_in_page(address
) + size
;
2346 now
= min(count
, (unsigned long)in_page
/ size
);
2349 * String I/O straddles page boundary. Pin two guest pages
2350 * so that we satisfy atomicity constraints. Do just one
2351 * transaction to avoid complexity.
2358 * String I/O in reverse. Yuck. Kill the guest, fix later.
2360 pr_unimpl(vcpu
, "guest string pio down\n");
2361 kvm_inject_gp(vcpu
, 0);
2364 vcpu
->run
->io
.count
= now
;
2365 vcpu
->arch
.pio
.cur_count
= now
;
2367 if (vcpu
->arch
.pio
.cur_count
== vcpu
->arch
.pio
.count
)
2368 kvm_x86_ops
->skip_emulated_instruction(vcpu
);
2370 for (i
= 0; i
< nr_pages
; ++i
) {
2371 down_read(&vcpu
->kvm
->slots_lock
);
2372 page
= gva_to_page(vcpu
, address
+ i
* PAGE_SIZE
);
2373 vcpu
->arch
.pio
.guest_pages
[i
] = page
;
2374 up_read(&vcpu
->kvm
->slots_lock
);
2376 kvm_inject_gp(vcpu
, 0);
2377 free_pio_guest_pages(vcpu
);
2382 pio_dev
= vcpu_find_pio_dev(vcpu
, port
);
2383 if (!vcpu
->arch
.pio
.in
) {
2384 /* string PIO write */
2385 ret
= pio_copy_data(vcpu
);
2386 if (ret
>= 0 && pio_dev
) {
2387 pio_string_write(pio_dev
, vcpu
);
2389 if (vcpu
->arch
.pio
.count
== 0)
2393 pr_unimpl(vcpu
, "no string pio read support yet, "
2394 "port %x size %d count %ld\n",
2399 EXPORT_SYMBOL_GPL(kvm_emulate_pio_string
);
2401 int kvm_arch_init(void *opaque
)
2404 struct kvm_x86_ops
*ops
= (struct kvm_x86_ops
*)opaque
;
2407 printk(KERN_ERR
"kvm: already loaded the other module\n");
2412 if (!ops
->cpu_has_kvm_support()) {
2413 printk(KERN_ERR
"kvm: no hardware support\n");
2417 if (ops
->disabled_by_bios()) {
2418 printk(KERN_ERR
"kvm: disabled by bios\n");
2423 r
= kvm_mmu_module_init();
2427 kvm_init_msr_list();
2430 kvm_mmu_set_nonpresent_ptes(0ull, 0ull);
2437 void kvm_arch_exit(void)
2440 kvm_mmu_module_exit();
2443 int kvm_emulate_halt(struct kvm_vcpu
*vcpu
)
2445 ++vcpu
->stat
.halt_exits
;
2446 if (irqchip_in_kernel(vcpu
->kvm
)) {
2447 vcpu
->arch
.mp_state
= VCPU_MP_STATE_HALTED
;
2448 kvm_vcpu_block(vcpu
);
2449 if (vcpu
->arch
.mp_state
!= VCPU_MP_STATE_RUNNABLE
)
2453 vcpu
->run
->exit_reason
= KVM_EXIT_HLT
;
2457 EXPORT_SYMBOL_GPL(kvm_emulate_halt
);
2459 static inline gpa_t
hc_gpa(struct kvm_vcpu
*vcpu
, unsigned long a0
,
2462 if (is_long_mode(vcpu
))
2465 return a0
| ((gpa_t
)a1
<< 32);
2468 int kvm_emulate_hypercall(struct kvm_vcpu
*vcpu
)
2470 unsigned long nr
, a0
, a1
, a2
, a3
, ret
;
2473 kvm_x86_ops
->cache_regs(vcpu
);
2475 nr
= vcpu
->arch
.regs
[VCPU_REGS_RAX
];
2476 a0
= vcpu
->arch
.regs
[VCPU_REGS_RBX
];
2477 a1
= vcpu
->arch
.regs
[VCPU_REGS_RCX
];
2478 a2
= vcpu
->arch
.regs
[VCPU_REGS_RDX
];
2479 a3
= vcpu
->arch
.regs
[VCPU_REGS_RSI
];
2481 if (!is_long_mode(vcpu
)) {
2490 case KVM_HC_VAPIC_POLL_IRQ
:
2494 r
= kvm_pv_mmu_op(vcpu
, a0
, hc_gpa(vcpu
, a1
, a2
), &ret
);
2500 vcpu
->arch
.regs
[VCPU_REGS_RAX
] = ret
;
2501 kvm_x86_ops
->decache_regs(vcpu
);
2502 ++vcpu
->stat
.hypercalls
;
2505 EXPORT_SYMBOL_GPL(kvm_emulate_hypercall
);
2507 int kvm_fix_hypercall(struct kvm_vcpu
*vcpu
)
2509 char instruction
[3];
2514 * Blow out the MMU to ensure that no other VCPU has an active mapping
2515 * to ensure that the updated hypercall appears atomically across all
2518 kvm_mmu_zap_all(vcpu
->kvm
);
2520 kvm_x86_ops
->cache_regs(vcpu
);
2521 kvm_x86_ops
->patch_hypercall(vcpu
, instruction
);
2522 if (emulator_write_emulated(vcpu
->arch
.rip
, instruction
, 3, vcpu
)
2523 != X86EMUL_CONTINUE
)
2529 static u64
mk_cr_64(u64 curr_cr
, u32 new_val
)
2531 return (curr_cr
& ~((1ULL << 32) - 1)) | new_val
;
2534 void realmode_lgdt(struct kvm_vcpu
*vcpu
, u16 limit
, unsigned long base
)
2536 struct descriptor_table dt
= { limit
, base
};
2538 kvm_x86_ops
->set_gdt(vcpu
, &dt
);
2541 void realmode_lidt(struct kvm_vcpu
*vcpu
, u16 limit
, unsigned long base
)
2543 struct descriptor_table dt
= { limit
, base
};
2545 kvm_x86_ops
->set_idt(vcpu
, &dt
);
2548 void realmode_lmsw(struct kvm_vcpu
*vcpu
, unsigned long msw
,
2549 unsigned long *rflags
)
2551 kvm_lmsw(vcpu
, msw
);
2552 *rflags
= kvm_x86_ops
->get_rflags(vcpu
);
2555 unsigned long realmode_get_cr(struct kvm_vcpu
*vcpu
, int cr
)
2557 kvm_x86_ops
->decache_cr4_guest_bits(vcpu
);
2560 return vcpu
->arch
.cr0
;
2562 return vcpu
->arch
.cr2
;
2564 return vcpu
->arch
.cr3
;
2566 return vcpu
->arch
.cr4
;
2568 return kvm_get_cr8(vcpu
);
2570 vcpu_printf(vcpu
, "%s: unexpected cr %u\n", __func__
, cr
);
2575 void realmode_set_cr(struct kvm_vcpu
*vcpu
, int cr
, unsigned long val
,
2576 unsigned long *rflags
)
2580 kvm_set_cr0(vcpu
, mk_cr_64(vcpu
->arch
.cr0
, val
));
2581 *rflags
= kvm_x86_ops
->get_rflags(vcpu
);
2584 vcpu
->arch
.cr2
= val
;
2587 kvm_set_cr3(vcpu
, val
);
2590 kvm_set_cr4(vcpu
, mk_cr_64(vcpu
->arch
.cr4
, val
));
2593 kvm_set_cr8(vcpu
, val
& 0xfUL
);
2596 vcpu_printf(vcpu
, "%s: unexpected cr %u\n", __func__
, cr
);
2600 static int move_to_next_stateful_cpuid_entry(struct kvm_vcpu
*vcpu
, int i
)
2602 struct kvm_cpuid_entry2
*e
= &vcpu
->arch
.cpuid_entries
[i
];
2603 int j
, nent
= vcpu
->arch
.cpuid_nent
;
2605 e
->flags
&= ~KVM_CPUID_FLAG_STATE_READ_NEXT
;
2606 /* when no next entry is found, the current entry[i] is reselected */
2607 for (j
= i
+ 1; j
== i
; j
= (j
+ 1) % nent
) {
2608 struct kvm_cpuid_entry2
*ej
= &vcpu
->arch
.cpuid_entries
[j
];
2609 if (ej
->function
== e
->function
) {
2610 ej
->flags
|= KVM_CPUID_FLAG_STATE_READ_NEXT
;
2614 return 0; /* silence gcc, even though control never reaches here */
2617 /* find an entry with matching function, matching index (if needed), and that
2618 * should be read next (if it's stateful) */
2619 static int is_matching_cpuid_entry(struct kvm_cpuid_entry2
*e
,
2620 u32 function
, u32 index
)
2622 if (e
->function
!= function
)
2624 if ((e
->flags
& KVM_CPUID_FLAG_SIGNIFCANT_INDEX
) && e
->index
!= index
)
2626 if ((e
->flags
& KVM_CPUID_FLAG_STATEFUL_FUNC
) &&
2627 !(e
->flags
& KVM_CPUID_FLAG_STATE_READ_NEXT
))
2632 void kvm_emulate_cpuid(struct kvm_vcpu
*vcpu
)
2635 u32 function
, index
;
2636 struct kvm_cpuid_entry2
*e
, *best
;
2638 kvm_x86_ops
->cache_regs(vcpu
);
2639 function
= vcpu
->arch
.regs
[VCPU_REGS_RAX
];
2640 index
= vcpu
->arch
.regs
[VCPU_REGS_RCX
];
2641 vcpu
->arch
.regs
[VCPU_REGS_RAX
] = 0;
2642 vcpu
->arch
.regs
[VCPU_REGS_RBX
] = 0;
2643 vcpu
->arch
.regs
[VCPU_REGS_RCX
] = 0;
2644 vcpu
->arch
.regs
[VCPU_REGS_RDX
] = 0;
2646 for (i
= 0; i
< vcpu
->arch
.cpuid_nent
; ++i
) {
2647 e
= &vcpu
->arch
.cpuid_entries
[i
];
2648 if (is_matching_cpuid_entry(e
, function
, index
)) {
2649 if (e
->flags
& KVM_CPUID_FLAG_STATEFUL_FUNC
)
2650 move_to_next_stateful_cpuid_entry(vcpu
, i
);
2655 * Both basic or both extended?
2657 if (((e
->function
^ function
) & 0x80000000) == 0)
2658 if (!best
|| e
->function
> best
->function
)
2662 vcpu
->arch
.regs
[VCPU_REGS_RAX
] = best
->eax
;
2663 vcpu
->arch
.regs
[VCPU_REGS_RBX
] = best
->ebx
;
2664 vcpu
->arch
.regs
[VCPU_REGS_RCX
] = best
->ecx
;
2665 vcpu
->arch
.regs
[VCPU_REGS_RDX
] = best
->edx
;
2667 kvm_x86_ops
->decache_regs(vcpu
);
2668 kvm_x86_ops
->skip_emulated_instruction(vcpu
);
2670 EXPORT_SYMBOL_GPL(kvm_emulate_cpuid
);
2673 * Check if userspace requested an interrupt window, and that the
2674 * interrupt window is open.
2676 * No need to exit to userspace if we already have an interrupt queued.
2678 static int dm_request_for_irq_injection(struct kvm_vcpu
*vcpu
,
2679 struct kvm_run
*kvm_run
)
2681 return (!vcpu
->arch
.irq_summary
&&
2682 kvm_run
->request_interrupt_window
&&
2683 vcpu
->arch
.interrupt_window_open
&&
2684 (kvm_x86_ops
->get_rflags(vcpu
) & X86_EFLAGS_IF
));
2687 static void post_kvm_run_save(struct kvm_vcpu
*vcpu
,
2688 struct kvm_run
*kvm_run
)
2690 kvm_run
->if_flag
= (kvm_x86_ops
->get_rflags(vcpu
) & X86_EFLAGS_IF
) != 0;
2691 kvm_run
->cr8
= kvm_get_cr8(vcpu
);
2692 kvm_run
->apic_base
= kvm_get_apic_base(vcpu
);
2693 if (irqchip_in_kernel(vcpu
->kvm
))
2694 kvm_run
->ready_for_interrupt_injection
= 1;
2696 kvm_run
->ready_for_interrupt_injection
=
2697 (vcpu
->arch
.interrupt_window_open
&&
2698 vcpu
->arch
.irq_summary
== 0);
2701 static void vapic_enter(struct kvm_vcpu
*vcpu
)
2703 struct kvm_lapic
*apic
= vcpu
->arch
.apic
;
2706 if (!apic
|| !apic
->vapic_addr
)
2709 down_read(¤t
->mm
->mmap_sem
);
2710 page
= gfn_to_page(vcpu
->kvm
, apic
->vapic_addr
>> PAGE_SHIFT
);
2711 up_read(¤t
->mm
->mmap_sem
);
2713 vcpu
->arch
.apic
->vapic_page
= page
;
2716 static void vapic_exit(struct kvm_vcpu
*vcpu
)
2718 struct kvm_lapic
*apic
= vcpu
->arch
.apic
;
2720 if (!apic
|| !apic
->vapic_addr
)
2723 kvm_release_page_dirty(apic
->vapic_page
);
2724 mark_page_dirty(vcpu
->kvm
, apic
->vapic_addr
>> PAGE_SHIFT
);
2727 static int __vcpu_run(struct kvm_vcpu
*vcpu
, struct kvm_run
*kvm_run
)
2731 if (unlikely(vcpu
->arch
.mp_state
== VCPU_MP_STATE_SIPI_RECEIVED
)) {
2732 pr_debug("vcpu %d received sipi with vector # %x\n",
2733 vcpu
->vcpu_id
, vcpu
->arch
.sipi_vector
);
2734 kvm_lapic_reset(vcpu
);
2735 r
= kvm_x86_ops
->vcpu_reset(vcpu
);
2738 vcpu
->arch
.mp_state
= VCPU_MP_STATE_RUNNABLE
;
2744 if (vcpu
->guest_debug
.enabled
)
2745 kvm_x86_ops
->guest_debug_pre(vcpu
);
2749 if (test_and_clear_bit(KVM_REQ_MMU_RELOAD
, &vcpu
->requests
))
2750 kvm_mmu_unload(vcpu
);
2752 r
= kvm_mmu_reload(vcpu
);
2756 if (vcpu
->requests
) {
2757 if (test_and_clear_bit(KVM_REQ_MIGRATE_TIMER
, &vcpu
->requests
))
2758 __kvm_migrate_apic_timer(vcpu
);
2759 if (test_and_clear_bit(KVM_REQ_REPORT_TPR_ACCESS
,
2761 kvm_run
->exit_reason
= KVM_EXIT_TPR_ACCESS
;
2765 if (test_and_clear_bit(KVM_REQ_TRIPLE_FAULT
, &vcpu
->requests
)) {
2766 kvm_run
->exit_reason
= KVM_EXIT_SHUTDOWN
;
2772 kvm_inject_pending_timer_irqs(vcpu
);
2776 kvm_x86_ops
->prepare_guest_switch(vcpu
);
2777 kvm_load_guest_fpu(vcpu
);
2779 local_irq_disable();
2781 if (need_resched()) {
2789 if (test_bit(KVM_REQ_MMU_RELOAD
, &vcpu
->requests
)) {
2796 if (signal_pending(current
)) {
2800 kvm_run
->exit_reason
= KVM_EXIT_INTR
;
2801 ++vcpu
->stat
.signal_exits
;
2805 if (vcpu
->arch
.exception
.pending
)
2806 __queue_exception(vcpu
);
2807 else if (irqchip_in_kernel(vcpu
->kvm
))
2808 kvm_x86_ops
->inject_pending_irq(vcpu
);
2810 kvm_x86_ops
->inject_pending_vectors(vcpu
, kvm_run
);
2812 kvm_lapic_sync_to_vapic(vcpu
);
2814 vcpu
->guest_mode
= 1;
2818 if (test_and_clear_bit(KVM_REQ_TLB_FLUSH
, &vcpu
->requests
))
2819 kvm_x86_ops
->tlb_flush(vcpu
);
2821 kvm_x86_ops
->run(vcpu
, kvm_run
);
2823 vcpu
->guest_mode
= 0;
2829 * We must have an instruction between local_irq_enable() and
2830 * kvm_guest_exit(), so the timer interrupt isn't delayed by
2831 * the interrupt shadow. The stat.exits increment will do nicely.
2832 * But we need to prevent reordering, hence this barrier():
2841 * Profile KVM exit RIPs:
2843 if (unlikely(prof_on
== KVM_PROFILING
)) {
2844 kvm_x86_ops
->cache_regs(vcpu
);
2845 profile_hit(KVM_PROFILING
, (void *)vcpu
->arch
.rip
);
2848 if (vcpu
->arch
.exception
.pending
&& kvm_x86_ops
->exception_injected(vcpu
))
2849 vcpu
->arch
.exception
.pending
= false;
2851 kvm_lapic_sync_from_vapic(vcpu
);
2853 r
= kvm_x86_ops
->handle_exit(kvm_run
, vcpu
);
2856 if (dm_request_for_irq_injection(vcpu
, kvm_run
)) {
2858 kvm_run
->exit_reason
= KVM_EXIT_INTR
;
2859 ++vcpu
->stat
.request_irq_exits
;
2862 if (!need_resched())
2872 post_kvm_run_save(vcpu
, kvm_run
);
2879 int kvm_arch_vcpu_ioctl_run(struct kvm_vcpu
*vcpu
, struct kvm_run
*kvm_run
)
2886 if (unlikely(vcpu
->arch
.mp_state
== VCPU_MP_STATE_UNINITIALIZED
)) {
2887 kvm_vcpu_block(vcpu
);
2892 if (vcpu
->sigset_active
)
2893 sigprocmask(SIG_SETMASK
, &vcpu
->sigset
, &sigsaved
);
2895 /* re-sync apic's tpr */
2896 if (!irqchip_in_kernel(vcpu
->kvm
))
2897 kvm_set_cr8(vcpu
, kvm_run
->cr8
);
2899 if (vcpu
->arch
.pio
.cur_count
) {
2900 r
= complete_pio(vcpu
);
2904 #if CONFIG_HAS_IOMEM
2905 if (vcpu
->mmio_needed
) {
2906 memcpy(vcpu
->mmio_data
, kvm_run
->mmio
.data
, 8);
2907 vcpu
->mmio_read_completed
= 1;
2908 vcpu
->mmio_needed
= 0;
2909 r
= emulate_instruction(vcpu
, kvm_run
,
2910 vcpu
->arch
.mmio_fault_cr2
, 0,
2911 EMULTYPE_NO_DECODE
);
2912 if (r
== EMULATE_DO_MMIO
) {
2914 * Read-modify-write. Back to userspace.
2921 if (kvm_run
->exit_reason
== KVM_EXIT_HYPERCALL
) {
2922 kvm_x86_ops
->cache_regs(vcpu
);
2923 vcpu
->arch
.regs
[VCPU_REGS_RAX
] = kvm_run
->hypercall
.ret
;
2924 kvm_x86_ops
->decache_regs(vcpu
);
2927 r
= __vcpu_run(vcpu
, kvm_run
);
2930 if (vcpu
->sigset_active
)
2931 sigprocmask(SIG_SETMASK
, &sigsaved
, NULL
);
2937 int kvm_arch_vcpu_ioctl_get_regs(struct kvm_vcpu
*vcpu
, struct kvm_regs
*regs
)
2941 kvm_x86_ops
->cache_regs(vcpu
);
2943 regs
->rax
= vcpu
->arch
.regs
[VCPU_REGS_RAX
];
2944 regs
->rbx
= vcpu
->arch
.regs
[VCPU_REGS_RBX
];
2945 regs
->rcx
= vcpu
->arch
.regs
[VCPU_REGS_RCX
];
2946 regs
->rdx
= vcpu
->arch
.regs
[VCPU_REGS_RDX
];
2947 regs
->rsi
= vcpu
->arch
.regs
[VCPU_REGS_RSI
];
2948 regs
->rdi
= vcpu
->arch
.regs
[VCPU_REGS_RDI
];
2949 regs
->rsp
= vcpu
->arch
.regs
[VCPU_REGS_RSP
];
2950 regs
->rbp
= vcpu
->arch
.regs
[VCPU_REGS_RBP
];
2951 #ifdef CONFIG_X86_64
2952 regs
->r8
= vcpu
->arch
.regs
[VCPU_REGS_R8
];
2953 regs
->r9
= vcpu
->arch
.regs
[VCPU_REGS_R9
];
2954 regs
->r10
= vcpu
->arch
.regs
[VCPU_REGS_R10
];
2955 regs
->r11
= vcpu
->arch
.regs
[VCPU_REGS_R11
];
2956 regs
->r12
= vcpu
->arch
.regs
[VCPU_REGS_R12
];
2957 regs
->r13
= vcpu
->arch
.regs
[VCPU_REGS_R13
];
2958 regs
->r14
= vcpu
->arch
.regs
[VCPU_REGS_R14
];
2959 regs
->r15
= vcpu
->arch
.regs
[VCPU_REGS_R15
];
2962 regs
->rip
= vcpu
->arch
.rip
;
2963 regs
->rflags
= kvm_x86_ops
->get_rflags(vcpu
);
2966 * Don't leak debug flags in case they were set for guest debugging
2968 if (vcpu
->guest_debug
.enabled
&& vcpu
->guest_debug
.singlestep
)
2969 regs
->rflags
&= ~(X86_EFLAGS_TF
| X86_EFLAGS_RF
);
2976 int kvm_arch_vcpu_ioctl_set_regs(struct kvm_vcpu
*vcpu
, struct kvm_regs
*regs
)
2980 vcpu
->arch
.regs
[VCPU_REGS_RAX
] = regs
->rax
;
2981 vcpu
->arch
.regs
[VCPU_REGS_RBX
] = regs
->rbx
;
2982 vcpu
->arch
.regs
[VCPU_REGS_RCX
] = regs
->rcx
;
2983 vcpu
->arch
.regs
[VCPU_REGS_RDX
] = regs
->rdx
;
2984 vcpu
->arch
.regs
[VCPU_REGS_RSI
] = regs
->rsi
;
2985 vcpu
->arch
.regs
[VCPU_REGS_RDI
] = regs
->rdi
;
2986 vcpu
->arch
.regs
[VCPU_REGS_RSP
] = regs
->rsp
;
2987 vcpu
->arch
.regs
[VCPU_REGS_RBP
] = regs
->rbp
;
2988 #ifdef CONFIG_X86_64
2989 vcpu
->arch
.regs
[VCPU_REGS_R8
] = regs
->r8
;
2990 vcpu
->arch
.regs
[VCPU_REGS_R9
] = regs
->r9
;
2991 vcpu
->arch
.regs
[VCPU_REGS_R10
] = regs
->r10
;
2992 vcpu
->arch
.regs
[VCPU_REGS_R11
] = regs
->r11
;
2993 vcpu
->arch
.regs
[VCPU_REGS_R12
] = regs
->r12
;
2994 vcpu
->arch
.regs
[VCPU_REGS_R13
] = regs
->r13
;
2995 vcpu
->arch
.regs
[VCPU_REGS_R14
] = regs
->r14
;
2996 vcpu
->arch
.regs
[VCPU_REGS_R15
] = regs
->r15
;
2999 vcpu
->arch
.rip
= regs
->rip
;
3000 kvm_x86_ops
->set_rflags(vcpu
, regs
->rflags
);
3002 kvm_x86_ops
->decache_regs(vcpu
);
3009 static void get_segment(struct kvm_vcpu
*vcpu
,
3010 struct kvm_segment
*var
, int seg
)
3012 kvm_x86_ops
->get_segment(vcpu
, var
, seg
);
3015 void kvm_get_cs_db_l_bits(struct kvm_vcpu
*vcpu
, int *db
, int *l
)
3017 struct kvm_segment cs
;
3019 get_segment(vcpu
, &cs
, VCPU_SREG_CS
);
3023 EXPORT_SYMBOL_GPL(kvm_get_cs_db_l_bits
);
3025 int kvm_arch_vcpu_ioctl_get_sregs(struct kvm_vcpu
*vcpu
,
3026 struct kvm_sregs
*sregs
)
3028 struct descriptor_table dt
;
3033 get_segment(vcpu
, &sregs
->cs
, VCPU_SREG_CS
);
3034 get_segment(vcpu
, &sregs
->ds
, VCPU_SREG_DS
);
3035 get_segment(vcpu
, &sregs
->es
, VCPU_SREG_ES
);
3036 get_segment(vcpu
, &sregs
->fs
, VCPU_SREG_FS
);
3037 get_segment(vcpu
, &sregs
->gs
, VCPU_SREG_GS
);
3038 get_segment(vcpu
, &sregs
->ss
, VCPU_SREG_SS
);
3040 get_segment(vcpu
, &sregs
->tr
, VCPU_SREG_TR
);
3041 get_segment(vcpu
, &sregs
->ldt
, VCPU_SREG_LDTR
);
3043 kvm_x86_ops
->get_idt(vcpu
, &dt
);
3044 sregs
->idt
.limit
= dt
.limit
;
3045 sregs
->idt
.base
= dt
.base
;
3046 kvm_x86_ops
->get_gdt(vcpu
, &dt
);
3047 sregs
->gdt
.limit
= dt
.limit
;
3048 sregs
->gdt
.base
= dt
.base
;
3050 kvm_x86_ops
->decache_cr4_guest_bits(vcpu
);
3051 sregs
->cr0
= vcpu
->arch
.cr0
;
3052 sregs
->cr2
= vcpu
->arch
.cr2
;
3053 sregs
->cr3
= vcpu
->arch
.cr3
;
3054 sregs
->cr4
= vcpu
->arch
.cr4
;
3055 sregs
->cr8
= kvm_get_cr8(vcpu
);
3056 sregs
->efer
= vcpu
->arch
.shadow_efer
;
3057 sregs
->apic_base
= kvm_get_apic_base(vcpu
);
3059 if (irqchip_in_kernel(vcpu
->kvm
)) {
3060 memset(sregs
->interrupt_bitmap
, 0,
3061 sizeof sregs
->interrupt_bitmap
);
3062 pending_vec
= kvm_x86_ops
->get_irq(vcpu
);
3063 if (pending_vec
>= 0)
3064 set_bit(pending_vec
,
3065 (unsigned long *)sregs
->interrupt_bitmap
);
3067 memcpy(sregs
->interrupt_bitmap
, vcpu
->arch
.irq_pending
,
3068 sizeof sregs
->interrupt_bitmap
);
3075 static void set_segment(struct kvm_vcpu
*vcpu
,
3076 struct kvm_segment
*var
, int seg
)
3078 kvm_x86_ops
->set_segment(vcpu
, var
, seg
);
3081 static void seg_desct_to_kvm_desct(struct desc_struct
*seg_desc
, u16 selector
,
3082 struct kvm_segment
*kvm_desct
)
3084 kvm_desct
->base
= seg_desc
->base0
;
3085 kvm_desct
->base
|= seg_desc
->base1
<< 16;
3086 kvm_desct
->base
|= seg_desc
->base2
<< 24;
3087 kvm_desct
->limit
= seg_desc
->limit0
;
3088 kvm_desct
->limit
|= seg_desc
->limit
<< 16;
3089 kvm_desct
->selector
= selector
;
3090 kvm_desct
->type
= seg_desc
->type
;
3091 kvm_desct
->present
= seg_desc
->p
;
3092 kvm_desct
->dpl
= seg_desc
->dpl
;
3093 kvm_desct
->db
= seg_desc
->d
;
3094 kvm_desct
->s
= seg_desc
->s
;
3095 kvm_desct
->l
= seg_desc
->l
;
3096 kvm_desct
->g
= seg_desc
->g
;
3097 kvm_desct
->avl
= seg_desc
->avl
;
3099 kvm_desct
->unusable
= 1;
3101 kvm_desct
->unusable
= 0;
3102 kvm_desct
->padding
= 0;
3105 static void get_segment_descritptor_dtable(struct kvm_vcpu
*vcpu
,
3107 struct descriptor_table
*dtable
)
3109 if (selector
& 1 << 2) {
3110 struct kvm_segment kvm_seg
;
3112 get_segment(vcpu
, &kvm_seg
, VCPU_SREG_LDTR
);
3114 if (kvm_seg
.unusable
)
3117 dtable
->limit
= kvm_seg
.limit
;
3118 dtable
->base
= kvm_seg
.base
;
3121 kvm_x86_ops
->get_gdt(vcpu
, dtable
);
3124 /* allowed just for 8 bytes segments */
3125 static int load_guest_segment_descriptor(struct kvm_vcpu
*vcpu
, u16 selector
,
3126 struct desc_struct
*seg_desc
)
3128 struct descriptor_table dtable
;
3129 u16 index
= selector
>> 3;
3131 get_segment_descritptor_dtable(vcpu
, selector
, &dtable
);
3133 if (dtable
.limit
< index
* 8 + 7) {
3134 kvm_queue_exception_e(vcpu
, GP_VECTOR
, selector
& 0xfffc);
3137 return kvm_read_guest(vcpu
->kvm
, dtable
.base
+ index
* 8, seg_desc
, 8);
3140 /* allowed just for 8 bytes segments */
3141 static int save_guest_segment_descriptor(struct kvm_vcpu
*vcpu
, u16 selector
,
3142 struct desc_struct
*seg_desc
)
3144 struct descriptor_table dtable
;
3145 u16 index
= selector
>> 3;
3147 get_segment_descritptor_dtable(vcpu
, selector
, &dtable
);
3149 if (dtable
.limit
< index
* 8 + 7)
3151 return kvm_write_guest(vcpu
->kvm
, dtable
.base
+ index
* 8, seg_desc
, 8);
3154 static u32
get_tss_base_addr(struct kvm_vcpu
*vcpu
,
3155 struct desc_struct
*seg_desc
)
3159 base_addr
= seg_desc
->base0
;
3160 base_addr
|= (seg_desc
->base1
<< 16);
3161 base_addr
|= (seg_desc
->base2
<< 24);
3166 static int load_tss_segment32(struct kvm_vcpu
*vcpu
,
3167 struct desc_struct
*seg_desc
,
3168 struct tss_segment_32
*tss
)
3172 base_addr
= get_tss_base_addr(vcpu
, seg_desc
);
3174 return kvm_read_guest(vcpu
->kvm
, base_addr
, tss
,
3175 sizeof(struct tss_segment_32
));
3178 static int save_tss_segment32(struct kvm_vcpu
*vcpu
,
3179 struct desc_struct
*seg_desc
,
3180 struct tss_segment_32
*tss
)
3184 base_addr
= get_tss_base_addr(vcpu
, seg_desc
);
3186 return kvm_write_guest(vcpu
->kvm
, base_addr
, tss
,
3187 sizeof(struct tss_segment_32
));
3190 static int load_tss_segment16(struct kvm_vcpu
*vcpu
,
3191 struct desc_struct
*seg_desc
,
3192 struct tss_segment_16
*tss
)
3196 base_addr
= get_tss_base_addr(vcpu
, seg_desc
);
3198 return kvm_read_guest(vcpu
->kvm
, base_addr
, tss
,
3199 sizeof(struct tss_segment_16
));
3202 static int save_tss_segment16(struct kvm_vcpu
*vcpu
,
3203 struct desc_struct
*seg_desc
,
3204 struct tss_segment_16
*tss
)
3208 base_addr
= get_tss_base_addr(vcpu
, seg_desc
);
3210 return kvm_write_guest(vcpu
->kvm
, base_addr
, tss
,
3211 sizeof(struct tss_segment_16
));
3214 static u16
get_segment_selector(struct kvm_vcpu
*vcpu
, int seg
)
3216 struct kvm_segment kvm_seg
;
3218 get_segment(vcpu
, &kvm_seg
, seg
);
3219 return kvm_seg
.selector
;
3222 static int load_segment_descriptor_to_kvm_desct(struct kvm_vcpu
*vcpu
,
3224 struct kvm_segment
*kvm_seg
)
3226 struct desc_struct seg_desc
;
3228 if (load_guest_segment_descriptor(vcpu
, selector
, &seg_desc
))
3230 seg_desct_to_kvm_desct(&seg_desc
, selector
, kvm_seg
);
3234 static int load_segment_descriptor(struct kvm_vcpu
*vcpu
, u16 selector
,
3235 int type_bits
, int seg
)
3237 struct kvm_segment kvm_seg
;
3239 if (load_segment_descriptor_to_kvm_desct(vcpu
, selector
, &kvm_seg
))
3241 kvm_seg
.type
|= type_bits
;
3243 if (seg
!= VCPU_SREG_SS
&& seg
!= VCPU_SREG_CS
&&
3244 seg
!= VCPU_SREG_LDTR
)
3246 kvm_seg
.unusable
= 1;
3248 set_segment(vcpu
, &kvm_seg
, seg
);
3252 static void save_state_to_tss32(struct kvm_vcpu
*vcpu
,
3253 struct tss_segment_32
*tss
)
3255 tss
->cr3
= vcpu
->arch
.cr3
;
3256 tss
->eip
= vcpu
->arch
.rip
;
3257 tss
->eflags
= kvm_x86_ops
->get_rflags(vcpu
);
3258 tss
->eax
= vcpu
->arch
.regs
[VCPU_REGS_RAX
];
3259 tss
->ecx
= vcpu
->arch
.regs
[VCPU_REGS_RCX
];
3260 tss
->edx
= vcpu
->arch
.regs
[VCPU_REGS_RDX
];
3261 tss
->ebx
= vcpu
->arch
.regs
[VCPU_REGS_RBX
];
3262 tss
->esp
= vcpu
->arch
.regs
[VCPU_REGS_RSP
];
3263 tss
->ebp
= vcpu
->arch
.regs
[VCPU_REGS_RBP
];
3264 tss
->esi
= vcpu
->arch
.regs
[VCPU_REGS_RSI
];
3265 tss
->edi
= vcpu
->arch
.regs
[VCPU_REGS_RDI
];
3267 tss
->es
= get_segment_selector(vcpu
, VCPU_SREG_ES
);
3268 tss
->cs
= get_segment_selector(vcpu
, VCPU_SREG_CS
);
3269 tss
->ss
= get_segment_selector(vcpu
, VCPU_SREG_SS
);
3270 tss
->ds
= get_segment_selector(vcpu
, VCPU_SREG_DS
);
3271 tss
->fs
= get_segment_selector(vcpu
, VCPU_SREG_FS
);
3272 tss
->gs
= get_segment_selector(vcpu
, VCPU_SREG_GS
);
3273 tss
->ldt_selector
= get_segment_selector(vcpu
, VCPU_SREG_LDTR
);
3274 tss
->prev_task_link
= get_segment_selector(vcpu
, VCPU_SREG_TR
);
3277 static int load_state_from_tss32(struct kvm_vcpu
*vcpu
,
3278 struct tss_segment_32
*tss
)
3280 kvm_set_cr3(vcpu
, tss
->cr3
);
3282 vcpu
->arch
.rip
= tss
->eip
;
3283 kvm_x86_ops
->set_rflags(vcpu
, tss
->eflags
| 2);
3285 vcpu
->arch
.regs
[VCPU_REGS_RAX
] = tss
->eax
;
3286 vcpu
->arch
.regs
[VCPU_REGS_RCX
] = tss
->ecx
;
3287 vcpu
->arch
.regs
[VCPU_REGS_RDX
] = tss
->edx
;
3288 vcpu
->arch
.regs
[VCPU_REGS_RBX
] = tss
->ebx
;
3289 vcpu
->arch
.regs
[VCPU_REGS_RSP
] = tss
->esp
;
3290 vcpu
->arch
.regs
[VCPU_REGS_RBP
] = tss
->ebp
;
3291 vcpu
->arch
.regs
[VCPU_REGS_RSI
] = tss
->esi
;
3292 vcpu
->arch
.regs
[VCPU_REGS_RDI
] = tss
->edi
;
3294 if (load_segment_descriptor(vcpu
, tss
->ldt_selector
, 0, VCPU_SREG_LDTR
))
3297 if (load_segment_descriptor(vcpu
, tss
->es
, 1, VCPU_SREG_ES
))
3300 if (load_segment_descriptor(vcpu
, tss
->cs
, 9, VCPU_SREG_CS
))
3303 if (load_segment_descriptor(vcpu
, tss
->ss
, 1, VCPU_SREG_SS
))
3306 if (load_segment_descriptor(vcpu
, tss
->ds
, 1, VCPU_SREG_DS
))
3309 if (load_segment_descriptor(vcpu
, tss
->fs
, 1, VCPU_SREG_FS
))
3312 if (load_segment_descriptor(vcpu
, tss
->gs
, 1, VCPU_SREG_GS
))
3317 static void save_state_to_tss16(struct kvm_vcpu
*vcpu
,
3318 struct tss_segment_16
*tss
)
3320 tss
->ip
= vcpu
->arch
.rip
;
3321 tss
->flag
= kvm_x86_ops
->get_rflags(vcpu
);
3322 tss
->ax
= vcpu
->arch
.regs
[VCPU_REGS_RAX
];
3323 tss
->cx
= vcpu
->arch
.regs
[VCPU_REGS_RCX
];
3324 tss
->dx
= vcpu
->arch
.regs
[VCPU_REGS_RDX
];
3325 tss
->bx
= vcpu
->arch
.regs
[VCPU_REGS_RBX
];
3326 tss
->sp
= vcpu
->arch
.regs
[VCPU_REGS_RSP
];
3327 tss
->bp
= vcpu
->arch
.regs
[VCPU_REGS_RBP
];
3328 tss
->si
= vcpu
->arch
.regs
[VCPU_REGS_RSI
];
3329 tss
->di
= vcpu
->arch
.regs
[VCPU_REGS_RDI
];
3331 tss
->es
= get_segment_selector(vcpu
, VCPU_SREG_ES
);
3332 tss
->cs
= get_segment_selector(vcpu
, VCPU_SREG_CS
);
3333 tss
->ss
= get_segment_selector(vcpu
, VCPU_SREG_SS
);
3334 tss
->ds
= get_segment_selector(vcpu
, VCPU_SREG_DS
);
3335 tss
->ldt
= get_segment_selector(vcpu
, VCPU_SREG_LDTR
);
3336 tss
->prev_task_link
= get_segment_selector(vcpu
, VCPU_SREG_TR
);
3339 static int load_state_from_tss16(struct kvm_vcpu
*vcpu
,
3340 struct tss_segment_16
*tss
)
3342 vcpu
->arch
.rip
= tss
->ip
;
3343 kvm_x86_ops
->set_rflags(vcpu
, tss
->flag
| 2);
3344 vcpu
->arch
.regs
[VCPU_REGS_RAX
] = tss
->ax
;
3345 vcpu
->arch
.regs
[VCPU_REGS_RCX
] = tss
->cx
;
3346 vcpu
->arch
.regs
[VCPU_REGS_RDX
] = tss
->dx
;
3347 vcpu
->arch
.regs
[VCPU_REGS_RBX
] = tss
->bx
;
3348 vcpu
->arch
.regs
[VCPU_REGS_RSP
] = tss
->sp
;
3349 vcpu
->arch
.regs
[VCPU_REGS_RBP
] = tss
->bp
;
3350 vcpu
->arch
.regs
[VCPU_REGS_RSI
] = tss
->si
;
3351 vcpu
->arch
.regs
[VCPU_REGS_RDI
] = tss
->di
;
3353 if (load_segment_descriptor(vcpu
, tss
->ldt
, 0, VCPU_SREG_LDTR
))
3356 if (load_segment_descriptor(vcpu
, tss
->es
, 1, VCPU_SREG_ES
))
3359 if (load_segment_descriptor(vcpu
, tss
->cs
, 9, VCPU_SREG_CS
))
3362 if (load_segment_descriptor(vcpu
, tss
->ss
, 1, VCPU_SREG_SS
))
3365 if (load_segment_descriptor(vcpu
, tss
->ds
, 1, VCPU_SREG_DS
))
3370 int kvm_task_switch_16(struct kvm_vcpu
*vcpu
, u16 tss_selector
,
3371 struct desc_struct
*cseg_desc
,
3372 struct desc_struct
*nseg_desc
)
3374 struct tss_segment_16 tss_segment_16
;
3377 if (load_tss_segment16(vcpu
, cseg_desc
, &tss_segment_16
))
3380 save_state_to_tss16(vcpu
, &tss_segment_16
);
3381 save_tss_segment16(vcpu
, cseg_desc
, &tss_segment_16
);
3383 if (load_tss_segment16(vcpu
, nseg_desc
, &tss_segment_16
))
3385 if (load_state_from_tss16(vcpu
, &tss_segment_16
))
3393 int kvm_task_switch_32(struct kvm_vcpu
*vcpu
, u16 tss_selector
,
3394 struct desc_struct
*cseg_desc
,
3395 struct desc_struct
*nseg_desc
)
3397 struct tss_segment_32 tss_segment_32
;
3400 if (load_tss_segment32(vcpu
, cseg_desc
, &tss_segment_32
))
3403 save_state_to_tss32(vcpu
, &tss_segment_32
);
3404 save_tss_segment32(vcpu
, cseg_desc
, &tss_segment_32
);
3406 if (load_tss_segment32(vcpu
, nseg_desc
, &tss_segment_32
))
3408 if (load_state_from_tss32(vcpu
, &tss_segment_32
))
3416 int kvm_task_switch(struct kvm_vcpu
*vcpu
, u16 tss_selector
, int reason
)
3418 struct kvm_segment tr_seg
;
3419 struct desc_struct cseg_desc
;
3420 struct desc_struct nseg_desc
;
3423 get_segment(vcpu
, &tr_seg
, VCPU_SREG_TR
);
3425 if (load_guest_segment_descriptor(vcpu
, tss_selector
, &nseg_desc
))
3428 if (load_guest_segment_descriptor(vcpu
, tr_seg
.selector
, &cseg_desc
))
3432 if (reason
!= TASK_SWITCH_IRET
) {
3435 cpl
= kvm_x86_ops
->get_cpl(vcpu
);
3436 if ((tss_selector
& 3) > nseg_desc
.dpl
|| cpl
> nseg_desc
.dpl
) {
3437 kvm_queue_exception_e(vcpu
, GP_VECTOR
, 0);
3442 if (!nseg_desc
.p
|| (nseg_desc
.limit0
| nseg_desc
.limit
<< 16) < 0x67) {
3443 kvm_queue_exception_e(vcpu
, TS_VECTOR
, tss_selector
& 0xfffc);
3447 if (reason
== TASK_SWITCH_IRET
|| reason
== TASK_SWITCH_JMP
) {
3448 cseg_desc
.type
&= ~(1 << 8); //clear the B flag
3449 save_guest_segment_descriptor(vcpu
, tr_seg
.selector
,
3453 if (reason
== TASK_SWITCH_IRET
) {
3454 u32 eflags
= kvm_x86_ops
->get_rflags(vcpu
);
3455 kvm_x86_ops
->set_rflags(vcpu
, eflags
& ~X86_EFLAGS_NT
);
3458 kvm_x86_ops
->skip_emulated_instruction(vcpu
);
3459 kvm_x86_ops
->cache_regs(vcpu
);
3461 if (nseg_desc
.type
& 8)
3462 ret
= kvm_task_switch_32(vcpu
, tss_selector
, &cseg_desc
,
3465 ret
= kvm_task_switch_16(vcpu
, tss_selector
, &cseg_desc
,
3468 if (reason
== TASK_SWITCH_CALL
|| reason
== TASK_SWITCH_GATE
) {
3469 u32 eflags
= kvm_x86_ops
->get_rflags(vcpu
);
3470 kvm_x86_ops
->set_rflags(vcpu
, eflags
| X86_EFLAGS_NT
);
3473 if (reason
!= TASK_SWITCH_IRET
) {
3474 nseg_desc
.type
|= (1 << 8);
3475 save_guest_segment_descriptor(vcpu
, tss_selector
,
3479 kvm_x86_ops
->set_cr0(vcpu
, vcpu
->arch
.cr0
| X86_CR0_TS
);
3480 seg_desct_to_kvm_desct(&nseg_desc
, tss_selector
, &tr_seg
);
3482 set_segment(vcpu
, &tr_seg
, VCPU_SREG_TR
);
3484 kvm_x86_ops
->decache_regs(vcpu
);
3487 EXPORT_SYMBOL_GPL(kvm_task_switch
);
3489 int kvm_arch_vcpu_ioctl_set_sregs(struct kvm_vcpu
*vcpu
,
3490 struct kvm_sregs
*sregs
)
3492 int mmu_reset_needed
= 0;
3493 int i
, pending_vec
, max_bits
;
3494 struct descriptor_table dt
;
3498 dt
.limit
= sregs
->idt
.limit
;
3499 dt
.base
= sregs
->idt
.base
;
3500 kvm_x86_ops
->set_idt(vcpu
, &dt
);
3501 dt
.limit
= sregs
->gdt
.limit
;
3502 dt
.base
= sregs
->gdt
.base
;
3503 kvm_x86_ops
->set_gdt(vcpu
, &dt
);
3505 vcpu
->arch
.cr2
= sregs
->cr2
;
3506 mmu_reset_needed
|= vcpu
->arch
.cr3
!= sregs
->cr3
;
3507 vcpu
->arch
.cr3
= sregs
->cr3
;
3509 kvm_set_cr8(vcpu
, sregs
->cr8
);
3511 mmu_reset_needed
|= vcpu
->arch
.shadow_efer
!= sregs
->efer
;
3512 kvm_x86_ops
->set_efer(vcpu
, sregs
->efer
);
3513 kvm_set_apic_base(vcpu
, sregs
->apic_base
);
3515 kvm_x86_ops
->decache_cr4_guest_bits(vcpu
);
3517 mmu_reset_needed
|= vcpu
->arch
.cr0
!= sregs
->cr0
;
3518 kvm_x86_ops
->set_cr0(vcpu
, sregs
->cr0
);
3519 vcpu
->arch
.cr0
= sregs
->cr0
;
3521 mmu_reset_needed
|= vcpu
->arch
.cr4
!= sregs
->cr4
;
3522 kvm_x86_ops
->set_cr4(vcpu
, sregs
->cr4
);
3523 if (!is_long_mode(vcpu
) && is_pae(vcpu
))
3524 load_pdptrs(vcpu
, vcpu
->arch
.cr3
);
3526 if (mmu_reset_needed
)
3527 kvm_mmu_reset_context(vcpu
);
3529 if (!irqchip_in_kernel(vcpu
->kvm
)) {
3530 memcpy(vcpu
->arch
.irq_pending
, sregs
->interrupt_bitmap
,
3531 sizeof vcpu
->arch
.irq_pending
);
3532 vcpu
->arch
.irq_summary
= 0;
3533 for (i
= 0; i
< ARRAY_SIZE(vcpu
->arch
.irq_pending
); ++i
)
3534 if (vcpu
->arch
.irq_pending
[i
])
3535 __set_bit(i
, &vcpu
->arch
.irq_summary
);
3537 max_bits
= (sizeof sregs
->interrupt_bitmap
) << 3;
3538 pending_vec
= find_first_bit(
3539 (const unsigned long *)sregs
->interrupt_bitmap
,
3541 /* Only pending external irq is handled here */
3542 if (pending_vec
< max_bits
) {
3543 kvm_x86_ops
->set_irq(vcpu
, pending_vec
);
3544 pr_debug("Set back pending irq %d\n",
3549 set_segment(vcpu
, &sregs
->cs
, VCPU_SREG_CS
);
3550 set_segment(vcpu
, &sregs
->ds
, VCPU_SREG_DS
);
3551 set_segment(vcpu
, &sregs
->es
, VCPU_SREG_ES
);
3552 set_segment(vcpu
, &sregs
->fs
, VCPU_SREG_FS
);
3553 set_segment(vcpu
, &sregs
->gs
, VCPU_SREG_GS
);
3554 set_segment(vcpu
, &sregs
->ss
, VCPU_SREG_SS
);
3556 set_segment(vcpu
, &sregs
->tr
, VCPU_SREG_TR
);
3557 set_segment(vcpu
, &sregs
->ldt
, VCPU_SREG_LDTR
);
3564 int kvm_arch_vcpu_ioctl_debug_guest(struct kvm_vcpu
*vcpu
,
3565 struct kvm_debug_guest
*dbg
)
3571 r
= kvm_x86_ops
->set_guest_debug(vcpu
, dbg
);
3579 * fxsave fpu state. Taken from x86_64/processor.h. To be killed when
3580 * we have asm/x86/processor.h
3591 u32 st_space
[32]; /* 8*16 bytes for each FP-reg = 128 bytes */
3592 #ifdef CONFIG_X86_64
3593 u32 xmm_space
[64]; /* 16*16 bytes for each XMM-reg = 256 bytes */
3595 u32 xmm_space
[32]; /* 8*16 bytes for each XMM-reg = 128 bytes */
3600 * Translate a guest virtual address to a guest physical address.
3602 int kvm_arch_vcpu_ioctl_translate(struct kvm_vcpu
*vcpu
,
3603 struct kvm_translation
*tr
)
3605 unsigned long vaddr
= tr
->linear_address
;
3609 down_read(&vcpu
->kvm
->slots_lock
);
3610 gpa
= vcpu
->arch
.mmu
.gva_to_gpa(vcpu
, vaddr
);
3611 up_read(&vcpu
->kvm
->slots_lock
);
3612 tr
->physical_address
= gpa
;
3613 tr
->valid
= gpa
!= UNMAPPED_GVA
;
3621 int kvm_arch_vcpu_ioctl_get_fpu(struct kvm_vcpu
*vcpu
, struct kvm_fpu
*fpu
)
3623 struct fxsave
*fxsave
= (struct fxsave
*)&vcpu
->arch
.guest_fx_image
;
3627 memcpy(fpu
->fpr
, fxsave
->st_space
, 128);
3628 fpu
->fcw
= fxsave
->cwd
;
3629 fpu
->fsw
= fxsave
->swd
;
3630 fpu
->ftwx
= fxsave
->twd
;
3631 fpu
->last_opcode
= fxsave
->fop
;
3632 fpu
->last_ip
= fxsave
->rip
;
3633 fpu
->last_dp
= fxsave
->rdp
;
3634 memcpy(fpu
->xmm
, fxsave
->xmm_space
, sizeof fxsave
->xmm_space
);
3641 int kvm_arch_vcpu_ioctl_set_fpu(struct kvm_vcpu
*vcpu
, struct kvm_fpu
*fpu
)
3643 struct fxsave
*fxsave
= (struct fxsave
*)&vcpu
->arch
.guest_fx_image
;
3647 memcpy(fxsave
->st_space
, fpu
->fpr
, 128);
3648 fxsave
->cwd
= fpu
->fcw
;
3649 fxsave
->swd
= fpu
->fsw
;
3650 fxsave
->twd
= fpu
->ftwx
;
3651 fxsave
->fop
= fpu
->last_opcode
;
3652 fxsave
->rip
= fpu
->last_ip
;
3653 fxsave
->rdp
= fpu
->last_dp
;
3654 memcpy(fxsave
->xmm_space
, fpu
->xmm
, sizeof fxsave
->xmm_space
);
3661 void fx_init(struct kvm_vcpu
*vcpu
)
3663 unsigned after_mxcsr_mask
;
3665 /* Initialize guest FPU by resetting ours and saving into guest's */
3667 fx_save(&vcpu
->arch
.host_fx_image
);
3669 fx_save(&vcpu
->arch
.guest_fx_image
);
3670 fx_restore(&vcpu
->arch
.host_fx_image
);
3673 vcpu
->arch
.cr0
|= X86_CR0_ET
;
3674 after_mxcsr_mask
= offsetof(struct i387_fxsave_struct
, st_space
);
3675 vcpu
->arch
.guest_fx_image
.mxcsr
= 0x1f80;
3676 memset((void *)&vcpu
->arch
.guest_fx_image
+ after_mxcsr_mask
,
3677 0, sizeof(struct i387_fxsave_struct
) - after_mxcsr_mask
);
3679 EXPORT_SYMBOL_GPL(fx_init
);
3681 void kvm_load_guest_fpu(struct kvm_vcpu
*vcpu
)
3683 if (!vcpu
->fpu_active
|| vcpu
->guest_fpu_loaded
)
3686 vcpu
->guest_fpu_loaded
= 1;
3687 fx_save(&vcpu
->arch
.host_fx_image
);
3688 fx_restore(&vcpu
->arch
.guest_fx_image
);
3690 EXPORT_SYMBOL_GPL(kvm_load_guest_fpu
);
3692 void kvm_put_guest_fpu(struct kvm_vcpu
*vcpu
)
3694 if (!vcpu
->guest_fpu_loaded
)
3697 vcpu
->guest_fpu_loaded
= 0;
3698 fx_save(&vcpu
->arch
.guest_fx_image
);
3699 fx_restore(&vcpu
->arch
.host_fx_image
);
3700 ++vcpu
->stat
.fpu_reload
;
3702 EXPORT_SYMBOL_GPL(kvm_put_guest_fpu
);
3704 void kvm_arch_vcpu_free(struct kvm_vcpu
*vcpu
)
3706 kvm_x86_ops
->vcpu_free(vcpu
);
3709 struct kvm_vcpu
*kvm_arch_vcpu_create(struct kvm
*kvm
,
3712 return kvm_x86_ops
->vcpu_create(kvm
, id
);
3715 int kvm_arch_vcpu_setup(struct kvm_vcpu
*vcpu
)
3719 /* We do fxsave: this must be aligned. */
3720 BUG_ON((unsigned long)&vcpu
->arch
.host_fx_image
& 0xF);
3723 r
= kvm_arch_vcpu_reset(vcpu
);
3725 r
= kvm_mmu_setup(vcpu
);
3732 kvm_x86_ops
->vcpu_free(vcpu
);
3736 void kvm_arch_vcpu_destroy(struct kvm_vcpu
*vcpu
)
3739 kvm_mmu_unload(vcpu
);
3742 kvm_x86_ops
->vcpu_free(vcpu
);
3745 int kvm_arch_vcpu_reset(struct kvm_vcpu
*vcpu
)
3747 return kvm_x86_ops
->vcpu_reset(vcpu
);
3750 void kvm_arch_hardware_enable(void *garbage
)
3752 kvm_x86_ops
->hardware_enable(garbage
);
3755 void kvm_arch_hardware_disable(void *garbage
)
3757 kvm_x86_ops
->hardware_disable(garbage
);
3760 int kvm_arch_hardware_setup(void)
3762 return kvm_x86_ops
->hardware_setup();
3765 void kvm_arch_hardware_unsetup(void)
3767 kvm_x86_ops
->hardware_unsetup();
3770 void kvm_arch_check_processor_compat(void *rtn
)
3772 kvm_x86_ops
->check_processor_compatibility(rtn
);
3775 int kvm_arch_vcpu_init(struct kvm_vcpu
*vcpu
)
3781 BUG_ON(vcpu
->kvm
== NULL
);
3784 vcpu
->arch
.mmu
.root_hpa
= INVALID_PAGE
;
3785 if (!irqchip_in_kernel(kvm
) || vcpu
->vcpu_id
== 0)
3786 vcpu
->arch
.mp_state
= VCPU_MP_STATE_RUNNABLE
;
3788 vcpu
->arch
.mp_state
= VCPU_MP_STATE_UNINITIALIZED
;
3790 page
= alloc_page(GFP_KERNEL
| __GFP_ZERO
);
3795 vcpu
->arch
.pio_data
= page_address(page
);
3797 r
= kvm_mmu_create(vcpu
);
3799 goto fail_free_pio_data
;
3801 if (irqchip_in_kernel(kvm
)) {
3802 r
= kvm_create_lapic(vcpu
);
3804 goto fail_mmu_destroy
;
3810 kvm_mmu_destroy(vcpu
);
3812 free_page((unsigned long)vcpu
->arch
.pio_data
);
3817 void kvm_arch_vcpu_uninit(struct kvm_vcpu
*vcpu
)
3819 kvm_free_lapic(vcpu
);
3820 kvm_mmu_destroy(vcpu
);
3821 free_page((unsigned long)vcpu
->arch
.pio_data
);
3824 struct kvm
*kvm_arch_create_vm(void)
3826 struct kvm
*kvm
= kzalloc(sizeof(struct kvm
), GFP_KERNEL
);
3829 return ERR_PTR(-ENOMEM
);
3831 INIT_LIST_HEAD(&kvm
->arch
.active_mmu_pages
);
3836 static void kvm_unload_vcpu_mmu(struct kvm_vcpu
*vcpu
)
3839 kvm_mmu_unload(vcpu
);
3843 static void kvm_free_vcpus(struct kvm
*kvm
)
3848 * Unpin any mmu pages first.
3850 for (i
= 0; i
< KVM_MAX_VCPUS
; ++i
)
3852 kvm_unload_vcpu_mmu(kvm
->vcpus
[i
]);
3853 for (i
= 0; i
< KVM_MAX_VCPUS
; ++i
) {
3854 if (kvm
->vcpus
[i
]) {
3855 kvm_arch_vcpu_free(kvm
->vcpus
[i
]);
3856 kvm
->vcpus
[i
] = NULL
;
3862 void kvm_arch_destroy_vm(struct kvm
*kvm
)
3865 kfree(kvm
->arch
.vpic
);
3866 kfree(kvm
->arch
.vioapic
);
3867 kvm_free_vcpus(kvm
);
3868 kvm_free_physmem(kvm
);
3872 int kvm_arch_set_memory_region(struct kvm
*kvm
,
3873 struct kvm_userspace_memory_region
*mem
,
3874 struct kvm_memory_slot old
,
3877 int npages
= mem
->memory_size
>> PAGE_SHIFT
;
3878 struct kvm_memory_slot
*memslot
= &kvm
->memslots
[mem
->slot
];
3880 /*To keep backward compatibility with older userspace,
3881 *x86 needs to hanlde !user_alloc case.
3884 if (npages
&& !old
.rmap
) {
3885 down_write(¤t
->mm
->mmap_sem
);
3886 memslot
->userspace_addr
= do_mmap(NULL
, 0,
3888 PROT_READ
| PROT_WRITE
,
3889 MAP_SHARED
| MAP_ANONYMOUS
,
3891 up_write(¤t
->mm
->mmap_sem
);
3893 if (IS_ERR((void *)memslot
->userspace_addr
))
3894 return PTR_ERR((void *)memslot
->userspace_addr
);
3896 if (!old
.user_alloc
&& old
.rmap
) {
3899 down_write(¤t
->mm
->mmap_sem
);
3900 ret
= do_munmap(current
->mm
, old
.userspace_addr
,
3901 old
.npages
* PAGE_SIZE
);
3902 up_write(¤t
->mm
->mmap_sem
);
3905 "kvm_vm_ioctl_set_memory_region: "
3906 "failed to munmap memory\n");
3911 if (!kvm
->arch
.n_requested_mmu_pages
) {
3912 unsigned int nr_mmu_pages
= kvm_mmu_calculate_mmu_pages(kvm
);
3913 kvm_mmu_change_mmu_pages(kvm
, nr_mmu_pages
);
3916 kvm_mmu_slot_remove_write_access(kvm
, mem
->slot
);
3917 kvm_flush_remote_tlbs(kvm
);
3922 int kvm_arch_vcpu_runnable(struct kvm_vcpu
*vcpu
)
3924 return vcpu
->arch
.mp_state
== VCPU_MP_STATE_RUNNABLE
3925 || vcpu
->arch
.mp_state
== VCPU_MP_STATE_SIPI_RECEIVED
;
3928 static void vcpu_kick_intr(void *info
)
3931 struct kvm_vcpu
*vcpu
= (struct kvm_vcpu
*)info
;
3932 printk(KERN_DEBUG
"vcpu_kick_intr %p \n", vcpu
);
3936 void kvm_vcpu_kick(struct kvm_vcpu
*vcpu
)
3938 int ipi_pcpu
= vcpu
->cpu
;
3940 if (waitqueue_active(&vcpu
->wq
)) {
3941 wake_up_interruptible(&vcpu
->wq
);
3942 ++vcpu
->stat
.halt_wakeup
;
3944 if (vcpu
->guest_mode
)
3945 smp_call_function_single(ipi_pcpu
, vcpu_kick_intr
, vcpu
, 0, 0);