2 * Kernel-based Virtual Machine driver for Linux
4 * derived from drivers/kvm/kvm_main.c
6 * Copyright (C) 2006 Qumranet, Inc.
7 * Copyright (C) 2008 Qumranet, Inc.
8 * Copyright IBM Corporation, 2008
11 * Avi Kivity <avi@qumranet.com>
12 * Yaniv Kamay <yaniv@qumranet.com>
13 * Amit Shah <amit.shah@qumranet.com>
14 * Ben-Ami Yassour <benami@il.ibm.com>
16 * This work is licensed under the terms of the GNU GPL, version 2. See
17 * the COPYING file in the top-level directory.
21 #include <linux/kvm_host.h>
26 #include "kvm_cache_regs.h"
29 #include <linux/clocksource.h>
30 #include <linux/interrupt.h>
31 #include <linux/kvm.h>
33 #include <linux/vmalloc.h>
34 #include <linux/module.h>
35 #include <linux/mman.h>
36 #include <linux/highmem.h>
37 #include <linux/iommu.h>
38 #include <linux/intel-iommu.h>
39 #include <linux/cpufreq.h>
41 #include <asm/uaccess.h>
46 #define MAX_IO_MSRS 256
47 #define CR0_RESERVED_BITS \
48 (~(unsigned long)(X86_CR0_PE | X86_CR0_MP | X86_CR0_EM | X86_CR0_TS \
49 | X86_CR0_ET | X86_CR0_NE | X86_CR0_WP | X86_CR0_AM \
50 | X86_CR0_NW | X86_CR0_CD | X86_CR0_PG))
51 #define CR4_RESERVED_BITS \
52 (~(unsigned long)(X86_CR4_VME | X86_CR4_PVI | X86_CR4_TSD | X86_CR4_DE\
53 | X86_CR4_PSE | X86_CR4_PAE | X86_CR4_MCE \
54 | X86_CR4_PGE | X86_CR4_PCE | X86_CR4_OSFXSR \
55 | X86_CR4_OSXMMEXCPT | X86_CR4_VMXE))
57 #define CR8_RESERVED_BITS (~(unsigned long)X86_CR8_TPR)
59 * - enable syscall per default because its emulated by KVM
60 * - enable LME and LMA per default on 64 bit KVM
63 static u64 __read_mostly efer_reserved_bits
= 0xfffffffffffffafeULL
;
65 static u64 __read_mostly efer_reserved_bits
= 0xfffffffffffffffeULL
;
68 #define VM_STAT(x) offsetof(struct kvm, stat.x), KVM_STAT_VM
69 #define VCPU_STAT(x) offsetof(struct kvm_vcpu, stat.x), KVM_STAT_VCPU
71 static int kvm_dev_ioctl_get_supported_cpuid(struct kvm_cpuid2
*cpuid
,
72 struct kvm_cpuid_entry2 __user
*entries
);
73 struct kvm_cpuid_entry2
*kvm_find_cpuid_entry(struct kvm_vcpu
*vcpu
,
74 u32 function
, u32 index
);
76 struct kvm_x86_ops
*kvm_x86_ops
;
77 EXPORT_SYMBOL_GPL(kvm_x86_ops
);
79 struct kvm_stats_debugfs_item debugfs_entries
[] = {
80 { "pf_fixed", VCPU_STAT(pf_fixed
) },
81 { "pf_guest", VCPU_STAT(pf_guest
) },
82 { "tlb_flush", VCPU_STAT(tlb_flush
) },
83 { "invlpg", VCPU_STAT(invlpg
) },
84 { "exits", VCPU_STAT(exits
) },
85 { "io_exits", VCPU_STAT(io_exits
) },
86 { "mmio_exits", VCPU_STAT(mmio_exits
) },
87 { "signal_exits", VCPU_STAT(signal_exits
) },
88 { "irq_window", VCPU_STAT(irq_window_exits
) },
89 { "nmi_window", VCPU_STAT(nmi_window_exits
) },
90 { "halt_exits", VCPU_STAT(halt_exits
) },
91 { "halt_wakeup", VCPU_STAT(halt_wakeup
) },
92 { "hypercalls", VCPU_STAT(hypercalls
) },
93 { "request_irq", VCPU_STAT(request_irq_exits
) },
94 { "request_nmi", VCPU_STAT(request_nmi_exits
) },
95 { "irq_exits", VCPU_STAT(irq_exits
) },
96 { "host_state_reload", VCPU_STAT(host_state_reload
) },
97 { "efer_reload", VCPU_STAT(efer_reload
) },
98 { "fpu_reload", VCPU_STAT(fpu_reload
) },
99 { "insn_emulation", VCPU_STAT(insn_emulation
) },
100 { "insn_emulation_fail", VCPU_STAT(insn_emulation_fail
) },
101 { "irq_injections", VCPU_STAT(irq_injections
) },
102 { "nmi_injections", VCPU_STAT(nmi_injections
) },
103 { "mmu_shadow_zapped", VM_STAT(mmu_shadow_zapped
) },
104 { "mmu_pte_write", VM_STAT(mmu_pte_write
) },
105 { "mmu_pte_updated", VM_STAT(mmu_pte_updated
) },
106 { "mmu_pde_zapped", VM_STAT(mmu_pde_zapped
) },
107 { "mmu_flooded", VM_STAT(mmu_flooded
) },
108 { "mmu_recycled", VM_STAT(mmu_recycled
) },
109 { "mmu_cache_miss", VM_STAT(mmu_cache_miss
) },
110 { "mmu_unsync", VM_STAT(mmu_unsync
) },
111 { "remote_tlb_flush", VM_STAT(remote_tlb_flush
) },
112 { "largepages", VM_STAT(lpages
) },
116 unsigned long segment_base(u16 selector
)
118 struct descriptor_table gdt
;
119 struct desc_struct
*d
;
120 unsigned long table_base
;
126 asm("sgdt %0" : "=m"(gdt
));
127 table_base
= gdt
.base
;
129 if (selector
& 4) { /* from ldt */
132 asm("sldt %0" : "=g"(ldt_selector
));
133 table_base
= segment_base(ldt_selector
);
135 d
= (struct desc_struct
*)(table_base
+ (selector
& ~7));
136 v
= d
->base0
| ((unsigned long)d
->base1
<< 16) |
137 ((unsigned long)d
->base2
<< 24);
139 if (d
->s
== 0 && (d
->type
== 2 || d
->type
== 9 || d
->type
== 11))
140 v
|= ((unsigned long)((struct ldttss_desc64
*)d
)->base3
) << 32;
144 EXPORT_SYMBOL_GPL(segment_base
);
146 u64
kvm_get_apic_base(struct kvm_vcpu
*vcpu
)
148 if (irqchip_in_kernel(vcpu
->kvm
))
149 return vcpu
->arch
.apic_base
;
151 return vcpu
->arch
.apic_base
;
153 EXPORT_SYMBOL_GPL(kvm_get_apic_base
);
155 void kvm_set_apic_base(struct kvm_vcpu
*vcpu
, u64 data
)
157 /* TODO: reserve bits check */
158 if (irqchip_in_kernel(vcpu
->kvm
))
159 kvm_lapic_set_base(vcpu
, data
);
161 vcpu
->arch
.apic_base
= data
;
163 EXPORT_SYMBOL_GPL(kvm_set_apic_base
);
165 void kvm_queue_exception(struct kvm_vcpu
*vcpu
, unsigned nr
)
167 WARN_ON(vcpu
->arch
.exception
.pending
);
168 vcpu
->arch
.exception
.pending
= true;
169 vcpu
->arch
.exception
.has_error_code
= false;
170 vcpu
->arch
.exception
.nr
= nr
;
172 EXPORT_SYMBOL_GPL(kvm_queue_exception
);
174 void kvm_inject_page_fault(struct kvm_vcpu
*vcpu
, unsigned long addr
,
177 ++vcpu
->stat
.pf_guest
;
179 if (vcpu
->arch
.exception
.pending
) {
180 if (vcpu
->arch
.exception
.nr
== PF_VECTOR
) {
181 printk(KERN_DEBUG
"kvm: inject_page_fault:"
182 " double fault 0x%lx\n", addr
);
183 vcpu
->arch
.exception
.nr
= DF_VECTOR
;
184 vcpu
->arch
.exception
.error_code
= 0;
185 } else if (vcpu
->arch
.exception
.nr
== DF_VECTOR
) {
186 /* triple fault -> shutdown */
187 set_bit(KVM_REQ_TRIPLE_FAULT
, &vcpu
->requests
);
191 vcpu
->arch
.cr2
= addr
;
192 kvm_queue_exception_e(vcpu
, PF_VECTOR
, error_code
);
195 void kvm_inject_nmi(struct kvm_vcpu
*vcpu
)
197 vcpu
->arch
.nmi_pending
= 1;
199 EXPORT_SYMBOL_GPL(kvm_inject_nmi
);
201 void kvm_queue_exception_e(struct kvm_vcpu
*vcpu
, unsigned nr
, u32 error_code
)
203 WARN_ON(vcpu
->arch
.exception
.pending
);
204 vcpu
->arch
.exception
.pending
= true;
205 vcpu
->arch
.exception
.has_error_code
= true;
206 vcpu
->arch
.exception
.nr
= nr
;
207 vcpu
->arch
.exception
.error_code
= error_code
;
209 EXPORT_SYMBOL_GPL(kvm_queue_exception_e
);
211 static void __queue_exception(struct kvm_vcpu
*vcpu
)
213 kvm_x86_ops
->queue_exception(vcpu
, vcpu
->arch
.exception
.nr
,
214 vcpu
->arch
.exception
.has_error_code
,
215 vcpu
->arch
.exception
.error_code
);
219 * Load the pae pdptrs. Return true is they are all valid.
221 int load_pdptrs(struct kvm_vcpu
*vcpu
, unsigned long cr3
)
223 gfn_t pdpt_gfn
= cr3
>> PAGE_SHIFT
;
224 unsigned offset
= ((cr3
& (PAGE_SIZE
-1)) >> 5) << 2;
227 u64 pdpte
[ARRAY_SIZE(vcpu
->arch
.pdptrs
)];
229 ret
= kvm_read_guest_page(vcpu
->kvm
, pdpt_gfn
, pdpte
,
230 offset
* sizeof(u64
), sizeof(pdpte
));
235 for (i
= 0; i
< ARRAY_SIZE(pdpte
); ++i
) {
236 if (is_present_pte(pdpte
[i
]) &&
237 (pdpte
[i
] & vcpu
->arch
.mmu
.rsvd_bits_mask
[0][2])) {
244 memcpy(vcpu
->arch
.pdptrs
, pdpte
, sizeof(vcpu
->arch
.pdptrs
));
249 EXPORT_SYMBOL_GPL(load_pdptrs
);
251 static bool pdptrs_changed(struct kvm_vcpu
*vcpu
)
253 u64 pdpte
[ARRAY_SIZE(vcpu
->arch
.pdptrs
)];
257 if (is_long_mode(vcpu
) || !is_pae(vcpu
))
260 r
= kvm_read_guest(vcpu
->kvm
, vcpu
->arch
.cr3
& ~31u, pdpte
, sizeof(pdpte
));
263 changed
= memcmp(pdpte
, vcpu
->arch
.pdptrs
, sizeof(pdpte
)) != 0;
269 void kvm_set_cr0(struct kvm_vcpu
*vcpu
, unsigned long cr0
)
271 if (cr0
& CR0_RESERVED_BITS
) {
272 printk(KERN_DEBUG
"set_cr0: 0x%lx #GP, reserved bits 0x%lx\n",
273 cr0
, vcpu
->arch
.cr0
);
274 kvm_inject_gp(vcpu
, 0);
278 if ((cr0
& X86_CR0_NW
) && !(cr0
& X86_CR0_CD
)) {
279 printk(KERN_DEBUG
"set_cr0: #GP, CD == 0 && NW == 1\n");
280 kvm_inject_gp(vcpu
, 0);
284 if ((cr0
& X86_CR0_PG
) && !(cr0
& X86_CR0_PE
)) {
285 printk(KERN_DEBUG
"set_cr0: #GP, set PG flag "
286 "and a clear PE flag\n");
287 kvm_inject_gp(vcpu
, 0);
291 if (!is_paging(vcpu
) && (cr0
& X86_CR0_PG
)) {
293 if ((vcpu
->arch
.shadow_efer
& EFER_LME
)) {
297 printk(KERN_DEBUG
"set_cr0: #GP, start paging "
298 "in long mode while PAE is disabled\n");
299 kvm_inject_gp(vcpu
, 0);
302 kvm_x86_ops
->get_cs_db_l_bits(vcpu
, &cs_db
, &cs_l
);
304 printk(KERN_DEBUG
"set_cr0: #GP, start paging "
305 "in long mode while CS.L == 1\n");
306 kvm_inject_gp(vcpu
, 0);
312 if (is_pae(vcpu
) && !load_pdptrs(vcpu
, vcpu
->arch
.cr3
)) {
313 printk(KERN_DEBUG
"set_cr0: #GP, pdptrs "
315 kvm_inject_gp(vcpu
, 0);
321 kvm_x86_ops
->set_cr0(vcpu
, cr0
);
322 vcpu
->arch
.cr0
= cr0
;
324 kvm_mmu_reset_context(vcpu
);
327 EXPORT_SYMBOL_GPL(kvm_set_cr0
);
329 void kvm_lmsw(struct kvm_vcpu
*vcpu
, unsigned long msw
)
331 kvm_set_cr0(vcpu
, (vcpu
->arch
.cr0
& ~0x0ful
) | (msw
& 0x0f));
332 KVMTRACE_1D(LMSW
, vcpu
,
333 (u32
)((vcpu
->arch
.cr0
& ~0x0ful
) | (msw
& 0x0f)),
336 EXPORT_SYMBOL_GPL(kvm_lmsw
);
338 void kvm_set_cr4(struct kvm_vcpu
*vcpu
, unsigned long cr4
)
340 unsigned long old_cr4
= vcpu
->arch
.cr4
;
341 unsigned long pdptr_bits
= X86_CR4_PGE
| X86_CR4_PSE
| X86_CR4_PAE
;
343 if (cr4
& CR4_RESERVED_BITS
) {
344 printk(KERN_DEBUG
"set_cr4: #GP, reserved bits\n");
345 kvm_inject_gp(vcpu
, 0);
349 if (is_long_mode(vcpu
)) {
350 if (!(cr4
& X86_CR4_PAE
)) {
351 printk(KERN_DEBUG
"set_cr4: #GP, clearing PAE while "
353 kvm_inject_gp(vcpu
, 0);
356 } else if (is_paging(vcpu
) && (cr4
& X86_CR4_PAE
)
357 && ((cr4
^ old_cr4
) & pdptr_bits
)
358 && !load_pdptrs(vcpu
, vcpu
->arch
.cr3
)) {
359 printk(KERN_DEBUG
"set_cr4: #GP, pdptrs reserved bits\n");
360 kvm_inject_gp(vcpu
, 0);
364 if (cr4
& X86_CR4_VMXE
) {
365 printk(KERN_DEBUG
"set_cr4: #GP, setting VMXE\n");
366 kvm_inject_gp(vcpu
, 0);
369 kvm_x86_ops
->set_cr4(vcpu
, cr4
);
370 vcpu
->arch
.cr4
= cr4
;
371 vcpu
->arch
.mmu
.base_role
.cr4_pge
= (cr4
& X86_CR4_PGE
) && !tdp_enabled
;
372 kvm_mmu_reset_context(vcpu
);
374 EXPORT_SYMBOL_GPL(kvm_set_cr4
);
376 void kvm_set_cr3(struct kvm_vcpu
*vcpu
, unsigned long cr3
)
378 if (cr3
== vcpu
->arch
.cr3
&& !pdptrs_changed(vcpu
)) {
379 kvm_mmu_sync_roots(vcpu
);
380 kvm_mmu_flush_tlb(vcpu
);
384 if (is_long_mode(vcpu
)) {
385 if (cr3
& CR3_L_MODE_RESERVED_BITS
) {
386 printk(KERN_DEBUG
"set_cr3: #GP, reserved bits\n");
387 kvm_inject_gp(vcpu
, 0);
392 if (cr3
& CR3_PAE_RESERVED_BITS
) {
394 "set_cr3: #GP, reserved bits\n");
395 kvm_inject_gp(vcpu
, 0);
398 if (is_paging(vcpu
) && !load_pdptrs(vcpu
, cr3
)) {
399 printk(KERN_DEBUG
"set_cr3: #GP, pdptrs "
401 kvm_inject_gp(vcpu
, 0);
406 * We don't check reserved bits in nonpae mode, because
407 * this isn't enforced, and VMware depends on this.
412 * Does the new cr3 value map to physical memory? (Note, we
413 * catch an invalid cr3 even in real-mode, because it would
414 * cause trouble later on when we turn on paging anyway.)
416 * A real CPU would silently accept an invalid cr3 and would
417 * attempt to use it - with largely undefined (and often hard
418 * to debug) behavior on the guest side.
420 if (unlikely(!gfn_to_memslot(vcpu
->kvm
, cr3
>> PAGE_SHIFT
)))
421 kvm_inject_gp(vcpu
, 0);
423 vcpu
->arch
.cr3
= cr3
;
424 vcpu
->arch
.mmu
.new_cr3(vcpu
);
427 EXPORT_SYMBOL_GPL(kvm_set_cr3
);
429 void kvm_set_cr8(struct kvm_vcpu
*vcpu
, unsigned long cr8
)
431 if (cr8
& CR8_RESERVED_BITS
) {
432 printk(KERN_DEBUG
"set_cr8: #GP, reserved bits 0x%lx\n", cr8
);
433 kvm_inject_gp(vcpu
, 0);
436 if (irqchip_in_kernel(vcpu
->kvm
))
437 kvm_lapic_set_tpr(vcpu
, cr8
);
439 vcpu
->arch
.cr8
= cr8
;
441 EXPORT_SYMBOL_GPL(kvm_set_cr8
);
443 unsigned long kvm_get_cr8(struct kvm_vcpu
*vcpu
)
445 if (irqchip_in_kernel(vcpu
->kvm
))
446 return kvm_lapic_get_cr8(vcpu
);
448 return vcpu
->arch
.cr8
;
450 EXPORT_SYMBOL_GPL(kvm_get_cr8
);
452 static inline u32
bit(int bitno
)
454 return 1 << (bitno
& 31);
458 * List of msr numbers which we expose to userspace through KVM_GET_MSRS
459 * and KVM_SET_MSRS, and KVM_GET_MSR_INDEX_LIST.
461 * This list is modified at module load time to reflect the
462 * capabilities of the host cpu.
464 static u32 msrs_to_save
[] = {
465 MSR_IA32_SYSENTER_CS
, MSR_IA32_SYSENTER_ESP
, MSR_IA32_SYSENTER_EIP
,
468 MSR_CSTAR
, MSR_KERNEL_GS_BASE
, MSR_SYSCALL_MASK
, MSR_LSTAR
,
470 MSR_IA32_TIME_STAMP_COUNTER
, MSR_KVM_SYSTEM_TIME
, MSR_KVM_WALL_CLOCK
,
471 MSR_IA32_PERF_STATUS
, MSR_IA32_CR_PAT
, MSR_VM_HSAVE_PA
474 static unsigned num_msrs_to_save
;
476 static u32 emulated_msrs
[] = {
477 MSR_IA32_MISC_ENABLE
,
480 static void set_efer(struct kvm_vcpu
*vcpu
, u64 efer
)
482 if (efer
& efer_reserved_bits
) {
483 printk(KERN_DEBUG
"set_efer: 0x%llx #GP, reserved bits\n",
485 kvm_inject_gp(vcpu
, 0);
490 && (vcpu
->arch
.shadow_efer
& EFER_LME
) != (efer
& EFER_LME
)) {
491 printk(KERN_DEBUG
"set_efer: #GP, change LME while paging\n");
492 kvm_inject_gp(vcpu
, 0);
496 if (efer
& EFER_FFXSR
) {
497 struct kvm_cpuid_entry2
*feat
;
499 feat
= kvm_find_cpuid_entry(vcpu
, 0x80000001, 0);
500 if (!feat
|| !(feat
->edx
& bit(X86_FEATURE_FXSR_OPT
))) {
501 printk(KERN_DEBUG
"set_efer: #GP, enable FFXSR w/o CPUID capability\n");
502 kvm_inject_gp(vcpu
, 0);
507 if (efer
& EFER_SVME
) {
508 struct kvm_cpuid_entry2
*feat
;
510 feat
= kvm_find_cpuid_entry(vcpu
, 0x80000001, 0);
511 if (!feat
|| !(feat
->ecx
& bit(X86_FEATURE_SVM
))) {
512 printk(KERN_DEBUG
"set_efer: #GP, enable SVM w/o SVM\n");
513 kvm_inject_gp(vcpu
, 0);
518 kvm_x86_ops
->set_efer(vcpu
, efer
);
521 efer
|= vcpu
->arch
.shadow_efer
& EFER_LMA
;
523 vcpu
->arch
.shadow_efer
= efer
;
525 vcpu
->arch
.mmu
.base_role
.nxe
= (efer
& EFER_NX
) && !tdp_enabled
;
526 kvm_mmu_reset_context(vcpu
);
529 void kvm_enable_efer_bits(u64 mask
)
531 efer_reserved_bits
&= ~mask
;
533 EXPORT_SYMBOL_GPL(kvm_enable_efer_bits
);
537 * Writes msr value into into the appropriate "register".
538 * Returns 0 on success, non-0 otherwise.
539 * Assumes vcpu_load() was already called.
541 int kvm_set_msr(struct kvm_vcpu
*vcpu
, u32 msr_index
, u64 data
)
543 return kvm_x86_ops
->set_msr(vcpu
, msr_index
, data
);
547 * Adapt set_msr() to msr_io()'s calling convention
549 static int do_set_msr(struct kvm_vcpu
*vcpu
, unsigned index
, u64
*data
)
551 return kvm_set_msr(vcpu
, index
, *data
);
554 static void kvm_write_wall_clock(struct kvm
*kvm
, gpa_t wall_clock
)
557 struct pvclock_wall_clock wc
;
558 struct timespec now
, sys
, boot
;
565 kvm_write_guest(kvm
, wall_clock
, &version
, sizeof(version
));
568 * The guest calculates current wall clock time by adding
569 * system time (updated by kvm_write_guest_time below) to the
570 * wall clock specified here. guest system time equals host
571 * system time for us, thus we must fill in host boot time here.
573 now
= current_kernel_time();
575 boot
= ns_to_timespec(timespec_to_ns(&now
) - timespec_to_ns(&sys
));
577 wc
.sec
= boot
.tv_sec
;
578 wc
.nsec
= boot
.tv_nsec
;
579 wc
.version
= version
;
581 kvm_write_guest(kvm
, wall_clock
, &wc
, sizeof(wc
));
584 kvm_write_guest(kvm
, wall_clock
, &version
, sizeof(version
));
587 static uint32_t div_frac(uint32_t dividend
, uint32_t divisor
)
589 uint32_t quotient
, remainder
;
591 /* Don't try to replace with do_div(), this one calculates
592 * "(dividend << 32) / divisor" */
594 : "=a" (quotient
), "=d" (remainder
)
595 : "0" (0), "1" (dividend
), "r" (divisor
) );
599 static void kvm_set_time_scale(uint32_t tsc_khz
, struct pvclock_vcpu_time_info
*hv_clock
)
601 uint64_t nsecs
= 1000000000LL;
606 tps64
= tsc_khz
* 1000LL;
607 while (tps64
> nsecs
*2) {
612 tps32
= (uint32_t)tps64
;
613 while (tps32
<= (uint32_t)nsecs
) {
618 hv_clock
->tsc_shift
= shift
;
619 hv_clock
->tsc_to_system_mul
= div_frac(nsecs
, tps32
);
621 pr_debug("%s: tsc_khz %u, tsc_shift %d, tsc_mul %u\n",
622 __func__
, tsc_khz
, hv_clock
->tsc_shift
,
623 hv_clock
->tsc_to_system_mul
);
626 static DEFINE_PER_CPU(unsigned long, cpu_tsc_khz
);
628 static void kvm_write_guest_time(struct kvm_vcpu
*v
)
632 struct kvm_vcpu_arch
*vcpu
= &v
->arch
;
634 unsigned long this_tsc_khz
;
636 if ((!vcpu
->time_page
))
639 this_tsc_khz
= get_cpu_var(cpu_tsc_khz
);
640 if (unlikely(vcpu
->hv_clock_tsc_khz
!= this_tsc_khz
)) {
641 kvm_set_time_scale(this_tsc_khz
, &vcpu
->hv_clock
);
642 vcpu
->hv_clock_tsc_khz
= this_tsc_khz
;
644 put_cpu_var(cpu_tsc_khz
);
646 /* Keep irq disabled to prevent changes to the clock */
647 local_irq_save(flags
);
648 kvm_get_msr(v
, MSR_IA32_TIME_STAMP_COUNTER
,
649 &vcpu
->hv_clock
.tsc_timestamp
);
651 local_irq_restore(flags
);
653 /* With all the info we got, fill in the values */
655 vcpu
->hv_clock
.system_time
= ts
.tv_nsec
+
656 (NSEC_PER_SEC
* (u64
)ts
.tv_sec
);
658 * The interface expects us to write an even number signaling that the
659 * update is finished. Since the guest won't see the intermediate
660 * state, we just increase by 2 at the end.
662 vcpu
->hv_clock
.version
+= 2;
664 shared_kaddr
= kmap_atomic(vcpu
->time_page
, KM_USER0
);
666 memcpy(shared_kaddr
+ vcpu
->time_offset
, &vcpu
->hv_clock
,
667 sizeof(vcpu
->hv_clock
));
669 kunmap_atomic(shared_kaddr
, KM_USER0
);
671 mark_page_dirty(v
->kvm
, vcpu
->time
>> PAGE_SHIFT
);
674 static int kvm_request_guest_time_update(struct kvm_vcpu
*v
)
676 struct kvm_vcpu_arch
*vcpu
= &v
->arch
;
678 if (!vcpu
->time_page
)
680 set_bit(KVM_REQ_KVMCLOCK_UPDATE
, &v
->requests
);
684 static bool msr_mtrr_valid(unsigned msr
)
687 case 0x200 ... 0x200 + 2 * KVM_NR_VAR_MTRR
- 1:
688 case MSR_MTRRfix64K_00000
:
689 case MSR_MTRRfix16K_80000
:
690 case MSR_MTRRfix16K_A0000
:
691 case MSR_MTRRfix4K_C0000
:
692 case MSR_MTRRfix4K_C8000
:
693 case MSR_MTRRfix4K_D0000
:
694 case MSR_MTRRfix4K_D8000
:
695 case MSR_MTRRfix4K_E0000
:
696 case MSR_MTRRfix4K_E8000
:
697 case MSR_MTRRfix4K_F0000
:
698 case MSR_MTRRfix4K_F8000
:
699 case MSR_MTRRdefType
:
700 case MSR_IA32_CR_PAT
:
708 static int set_msr_mtrr(struct kvm_vcpu
*vcpu
, u32 msr
, u64 data
)
710 u64
*p
= (u64
*)&vcpu
->arch
.mtrr_state
.fixed_ranges
;
712 if (!msr_mtrr_valid(msr
))
715 if (msr
== MSR_MTRRdefType
) {
716 vcpu
->arch
.mtrr_state
.def_type
= data
;
717 vcpu
->arch
.mtrr_state
.enabled
= (data
& 0xc00) >> 10;
718 } else if (msr
== MSR_MTRRfix64K_00000
)
720 else if (msr
== MSR_MTRRfix16K_80000
|| msr
== MSR_MTRRfix16K_A0000
)
721 p
[1 + msr
- MSR_MTRRfix16K_80000
] = data
;
722 else if (msr
>= MSR_MTRRfix4K_C0000
&& msr
<= MSR_MTRRfix4K_F8000
)
723 p
[3 + msr
- MSR_MTRRfix4K_C0000
] = data
;
724 else if (msr
== MSR_IA32_CR_PAT
)
725 vcpu
->arch
.pat
= data
;
726 else { /* Variable MTRRs */
727 int idx
, is_mtrr_mask
;
730 idx
= (msr
- 0x200) / 2;
731 is_mtrr_mask
= msr
- 0x200 - 2 * idx
;
734 (u64
*)&vcpu
->arch
.mtrr_state
.var_ranges
[idx
].base_lo
;
737 (u64
*)&vcpu
->arch
.mtrr_state
.var_ranges
[idx
].mask_lo
;
741 kvm_mmu_reset_context(vcpu
);
745 int kvm_set_msr_common(struct kvm_vcpu
*vcpu
, u32 msr
, u64 data
)
749 set_efer(vcpu
, data
);
751 case MSR_IA32_MC0_STATUS
:
752 pr_unimpl(vcpu
, "%s: MSR_IA32_MC0_STATUS 0x%llx, nop\n",
755 case MSR_IA32_MCG_STATUS
:
756 pr_unimpl(vcpu
, "%s: MSR_IA32_MCG_STATUS 0x%llx, nop\n",
759 case MSR_IA32_MCG_CTL
:
760 pr_unimpl(vcpu
, "%s: MSR_IA32_MCG_CTL 0x%llx, nop\n",
763 case MSR_IA32_DEBUGCTLMSR
:
765 /* We support the non-activated case already */
767 } else if (data
& ~(DEBUGCTLMSR_LBR
| DEBUGCTLMSR_BTF
)) {
768 /* Values other than LBR and BTF are vendor-specific,
769 thus reserved and should throw a #GP */
772 pr_unimpl(vcpu
, "%s: MSR_IA32_DEBUGCTLMSR 0x%llx, nop\n",
775 case MSR_IA32_UCODE_REV
:
776 case MSR_IA32_UCODE_WRITE
:
777 case MSR_VM_HSAVE_PA
:
779 case 0x200 ... 0x2ff:
780 return set_msr_mtrr(vcpu
, msr
, data
);
781 case MSR_IA32_APICBASE
:
782 kvm_set_apic_base(vcpu
, data
);
784 case MSR_IA32_MISC_ENABLE
:
785 vcpu
->arch
.ia32_misc_enable_msr
= data
;
787 case MSR_KVM_WALL_CLOCK
:
788 vcpu
->kvm
->arch
.wall_clock
= data
;
789 kvm_write_wall_clock(vcpu
->kvm
, data
);
791 case MSR_KVM_SYSTEM_TIME
: {
792 if (vcpu
->arch
.time_page
) {
793 kvm_release_page_dirty(vcpu
->arch
.time_page
);
794 vcpu
->arch
.time_page
= NULL
;
797 vcpu
->arch
.time
= data
;
799 /* we verify if the enable bit is set... */
803 /* ...but clean it before doing the actual write */
804 vcpu
->arch
.time_offset
= data
& ~(PAGE_MASK
| 1);
806 vcpu
->arch
.time_page
=
807 gfn_to_page(vcpu
->kvm
, data
>> PAGE_SHIFT
);
809 if (is_error_page(vcpu
->arch
.time_page
)) {
810 kvm_release_page_clean(vcpu
->arch
.time_page
);
811 vcpu
->arch
.time_page
= NULL
;
814 kvm_request_guest_time_update(vcpu
);
818 pr_unimpl(vcpu
, "unhandled wrmsr: 0x%x data %llx\n", msr
, data
);
823 EXPORT_SYMBOL_GPL(kvm_set_msr_common
);
827 * Reads an msr value (of 'msr_index') into 'pdata'.
828 * Returns 0 on success, non-0 otherwise.
829 * Assumes vcpu_load() was already called.
831 int kvm_get_msr(struct kvm_vcpu
*vcpu
, u32 msr_index
, u64
*pdata
)
833 return kvm_x86_ops
->get_msr(vcpu
, msr_index
, pdata
);
836 static int get_msr_mtrr(struct kvm_vcpu
*vcpu
, u32 msr
, u64
*pdata
)
838 u64
*p
= (u64
*)&vcpu
->arch
.mtrr_state
.fixed_ranges
;
840 if (!msr_mtrr_valid(msr
))
843 if (msr
== MSR_MTRRdefType
)
844 *pdata
= vcpu
->arch
.mtrr_state
.def_type
+
845 (vcpu
->arch
.mtrr_state
.enabled
<< 10);
846 else if (msr
== MSR_MTRRfix64K_00000
)
848 else if (msr
== MSR_MTRRfix16K_80000
|| msr
== MSR_MTRRfix16K_A0000
)
849 *pdata
= p
[1 + msr
- MSR_MTRRfix16K_80000
];
850 else if (msr
>= MSR_MTRRfix4K_C0000
&& msr
<= MSR_MTRRfix4K_F8000
)
851 *pdata
= p
[3 + msr
- MSR_MTRRfix4K_C0000
];
852 else if (msr
== MSR_IA32_CR_PAT
)
853 *pdata
= vcpu
->arch
.pat
;
854 else { /* Variable MTRRs */
855 int idx
, is_mtrr_mask
;
858 idx
= (msr
- 0x200) / 2;
859 is_mtrr_mask
= msr
- 0x200 - 2 * idx
;
862 (u64
*)&vcpu
->arch
.mtrr_state
.var_ranges
[idx
].base_lo
;
865 (u64
*)&vcpu
->arch
.mtrr_state
.var_ranges
[idx
].mask_lo
;
872 int kvm_get_msr_common(struct kvm_vcpu
*vcpu
, u32 msr
, u64
*pdata
)
877 case 0xc0010010: /* SYSCFG */
878 case 0xc0010015: /* HWCR */
879 case MSR_IA32_PLATFORM_ID
:
880 case MSR_IA32_P5_MC_ADDR
:
881 case MSR_IA32_P5_MC_TYPE
:
882 case MSR_IA32_MC0_CTL
:
883 case MSR_IA32_MCG_STATUS
:
884 case MSR_IA32_MCG_CAP
:
885 case MSR_IA32_MCG_CTL
:
886 case MSR_IA32_MC0_MISC
:
887 case MSR_IA32_MC0_MISC
+4:
888 case MSR_IA32_MC0_MISC
+8:
889 case MSR_IA32_MC0_MISC
+12:
890 case MSR_IA32_MC0_MISC
+16:
891 case MSR_IA32_MC0_MISC
+20:
892 case MSR_IA32_UCODE_REV
:
893 case MSR_IA32_EBL_CR_POWERON
:
894 case MSR_IA32_DEBUGCTLMSR
:
895 case MSR_IA32_LASTBRANCHFROMIP
:
896 case MSR_IA32_LASTBRANCHTOIP
:
897 case MSR_IA32_LASTINTFROMIP
:
898 case MSR_IA32_LASTINTTOIP
:
899 case MSR_VM_HSAVE_PA
:
900 case MSR_P6_EVNTSEL0
:
901 case MSR_P6_EVNTSEL1
:
905 data
= 0x500 | KVM_NR_VAR_MTRR
;
907 case 0x200 ... 0x2ff:
908 return get_msr_mtrr(vcpu
, msr
, pdata
);
909 case 0xcd: /* fsb frequency */
912 case MSR_IA32_APICBASE
:
913 data
= kvm_get_apic_base(vcpu
);
915 case MSR_IA32_MISC_ENABLE
:
916 data
= vcpu
->arch
.ia32_misc_enable_msr
;
918 case MSR_IA32_PERF_STATUS
:
919 /* TSC increment by tick */
922 data
|= (((uint64_t)4ULL) << 40);
925 data
= vcpu
->arch
.shadow_efer
;
927 case MSR_KVM_WALL_CLOCK
:
928 data
= vcpu
->kvm
->arch
.wall_clock
;
930 case MSR_KVM_SYSTEM_TIME
:
931 data
= vcpu
->arch
.time
;
934 pr_unimpl(vcpu
, "unhandled rdmsr: 0x%x\n", msr
);
940 EXPORT_SYMBOL_GPL(kvm_get_msr_common
);
943 * Read or write a bunch of msrs. All parameters are kernel addresses.
945 * @return number of msrs set successfully.
947 static int __msr_io(struct kvm_vcpu
*vcpu
, struct kvm_msrs
*msrs
,
948 struct kvm_msr_entry
*entries
,
949 int (*do_msr
)(struct kvm_vcpu
*vcpu
,
950 unsigned index
, u64
*data
))
956 down_read(&vcpu
->kvm
->slots_lock
);
957 for (i
= 0; i
< msrs
->nmsrs
; ++i
)
958 if (do_msr(vcpu
, entries
[i
].index
, &entries
[i
].data
))
960 up_read(&vcpu
->kvm
->slots_lock
);
968 * Read or write a bunch of msrs. Parameters are user addresses.
970 * @return number of msrs set successfully.
972 static int msr_io(struct kvm_vcpu
*vcpu
, struct kvm_msrs __user
*user_msrs
,
973 int (*do_msr
)(struct kvm_vcpu
*vcpu
,
974 unsigned index
, u64
*data
),
977 struct kvm_msrs msrs
;
978 struct kvm_msr_entry
*entries
;
983 if (copy_from_user(&msrs
, user_msrs
, sizeof msrs
))
987 if (msrs
.nmsrs
>= MAX_IO_MSRS
)
991 size
= sizeof(struct kvm_msr_entry
) * msrs
.nmsrs
;
992 entries
= vmalloc(size
);
997 if (copy_from_user(entries
, user_msrs
->entries
, size
))
1000 r
= n
= __msr_io(vcpu
, &msrs
, entries
, do_msr
);
1005 if (writeback
&& copy_to_user(user_msrs
->entries
, entries
, size
))
1016 int kvm_dev_ioctl_check_extension(long ext
)
1021 case KVM_CAP_IRQCHIP
:
1023 case KVM_CAP_MMU_SHADOW_CACHE_CONTROL
:
1024 case KVM_CAP_SET_TSS_ADDR
:
1025 case KVM_CAP_EXT_CPUID
:
1026 case KVM_CAP_CLOCKSOURCE
:
1028 case KVM_CAP_NOP_IO_DELAY
:
1029 case KVM_CAP_MP_STATE
:
1030 case KVM_CAP_SYNC_MMU
:
1031 case KVM_CAP_REINJECT_CONTROL
:
1032 case KVM_CAP_IRQ_INJECT_STATUS
:
1033 case KVM_CAP_ASSIGN_DEV_IRQ
:
1036 case KVM_CAP_COALESCED_MMIO
:
1037 r
= KVM_COALESCED_MMIO_PAGE_OFFSET
;
1040 r
= !kvm_x86_ops
->cpu_has_accelerated_tpr();
1042 case KVM_CAP_NR_VCPUS
:
1045 case KVM_CAP_NR_MEMSLOTS
:
1046 r
= KVM_MEMORY_SLOTS
;
1048 case KVM_CAP_PV_MMU
:
1062 long kvm_arch_dev_ioctl(struct file
*filp
,
1063 unsigned int ioctl
, unsigned long arg
)
1065 void __user
*argp
= (void __user
*)arg
;
1069 case KVM_GET_MSR_INDEX_LIST
: {
1070 struct kvm_msr_list __user
*user_msr_list
= argp
;
1071 struct kvm_msr_list msr_list
;
1075 if (copy_from_user(&msr_list
, user_msr_list
, sizeof msr_list
))
1078 msr_list
.nmsrs
= num_msrs_to_save
+ ARRAY_SIZE(emulated_msrs
);
1079 if (copy_to_user(user_msr_list
, &msr_list
, sizeof msr_list
))
1082 if (n
< num_msrs_to_save
)
1085 if (copy_to_user(user_msr_list
->indices
, &msrs_to_save
,
1086 num_msrs_to_save
* sizeof(u32
)))
1088 if (copy_to_user(user_msr_list
->indices
1089 + num_msrs_to_save
* sizeof(u32
),
1091 ARRAY_SIZE(emulated_msrs
) * sizeof(u32
)))
1096 case KVM_GET_SUPPORTED_CPUID
: {
1097 struct kvm_cpuid2 __user
*cpuid_arg
= argp
;
1098 struct kvm_cpuid2 cpuid
;
1101 if (copy_from_user(&cpuid
, cpuid_arg
, sizeof cpuid
))
1103 r
= kvm_dev_ioctl_get_supported_cpuid(&cpuid
,
1104 cpuid_arg
->entries
);
1109 if (copy_to_user(cpuid_arg
, &cpuid
, sizeof cpuid
))
1121 void kvm_arch_vcpu_load(struct kvm_vcpu
*vcpu
, int cpu
)
1123 kvm_x86_ops
->vcpu_load(vcpu
, cpu
);
1124 kvm_request_guest_time_update(vcpu
);
1127 void kvm_arch_vcpu_put(struct kvm_vcpu
*vcpu
)
1129 kvm_x86_ops
->vcpu_put(vcpu
);
1130 kvm_put_guest_fpu(vcpu
);
1133 static int is_efer_nx(void)
1135 unsigned long long efer
= 0;
1137 rdmsrl_safe(MSR_EFER
, &efer
);
1138 return efer
& EFER_NX
;
1141 static void cpuid_fix_nx_cap(struct kvm_vcpu
*vcpu
)
1144 struct kvm_cpuid_entry2
*e
, *entry
;
1147 for (i
= 0; i
< vcpu
->arch
.cpuid_nent
; ++i
) {
1148 e
= &vcpu
->arch
.cpuid_entries
[i
];
1149 if (e
->function
== 0x80000001) {
1154 if (entry
&& (entry
->edx
& (1 << 20)) && !is_efer_nx()) {
1155 entry
->edx
&= ~(1 << 20);
1156 printk(KERN_INFO
"kvm: guest NX capability removed\n");
1160 /* when an old userspace process fills a new kernel module */
1161 static int kvm_vcpu_ioctl_set_cpuid(struct kvm_vcpu
*vcpu
,
1162 struct kvm_cpuid
*cpuid
,
1163 struct kvm_cpuid_entry __user
*entries
)
1166 struct kvm_cpuid_entry
*cpuid_entries
;
1169 if (cpuid
->nent
> KVM_MAX_CPUID_ENTRIES
)
1172 cpuid_entries
= vmalloc(sizeof(struct kvm_cpuid_entry
) * cpuid
->nent
);
1176 if (copy_from_user(cpuid_entries
, entries
,
1177 cpuid
->nent
* sizeof(struct kvm_cpuid_entry
)))
1179 for (i
= 0; i
< cpuid
->nent
; i
++) {
1180 vcpu
->arch
.cpuid_entries
[i
].function
= cpuid_entries
[i
].function
;
1181 vcpu
->arch
.cpuid_entries
[i
].eax
= cpuid_entries
[i
].eax
;
1182 vcpu
->arch
.cpuid_entries
[i
].ebx
= cpuid_entries
[i
].ebx
;
1183 vcpu
->arch
.cpuid_entries
[i
].ecx
= cpuid_entries
[i
].ecx
;
1184 vcpu
->arch
.cpuid_entries
[i
].edx
= cpuid_entries
[i
].edx
;
1185 vcpu
->arch
.cpuid_entries
[i
].index
= 0;
1186 vcpu
->arch
.cpuid_entries
[i
].flags
= 0;
1187 vcpu
->arch
.cpuid_entries
[i
].padding
[0] = 0;
1188 vcpu
->arch
.cpuid_entries
[i
].padding
[1] = 0;
1189 vcpu
->arch
.cpuid_entries
[i
].padding
[2] = 0;
1191 vcpu
->arch
.cpuid_nent
= cpuid
->nent
;
1192 cpuid_fix_nx_cap(vcpu
);
1196 vfree(cpuid_entries
);
1201 static int kvm_vcpu_ioctl_set_cpuid2(struct kvm_vcpu
*vcpu
,
1202 struct kvm_cpuid2
*cpuid
,
1203 struct kvm_cpuid_entry2 __user
*entries
)
1208 if (cpuid
->nent
> KVM_MAX_CPUID_ENTRIES
)
1211 if (copy_from_user(&vcpu
->arch
.cpuid_entries
, entries
,
1212 cpuid
->nent
* sizeof(struct kvm_cpuid_entry2
)))
1214 vcpu
->arch
.cpuid_nent
= cpuid
->nent
;
1221 static int kvm_vcpu_ioctl_get_cpuid2(struct kvm_vcpu
*vcpu
,
1222 struct kvm_cpuid2
*cpuid
,
1223 struct kvm_cpuid_entry2 __user
*entries
)
1228 if (cpuid
->nent
< vcpu
->arch
.cpuid_nent
)
1231 if (copy_to_user(entries
, &vcpu
->arch
.cpuid_entries
,
1232 vcpu
->arch
.cpuid_nent
* sizeof(struct kvm_cpuid_entry2
)))
1237 cpuid
->nent
= vcpu
->arch
.cpuid_nent
;
1241 static void do_cpuid_1_ent(struct kvm_cpuid_entry2
*entry
, u32 function
,
1244 entry
->function
= function
;
1245 entry
->index
= index
;
1246 cpuid_count(entry
->function
, entry
->index
,
1247 &entry
->eax
, &entry
->ebx
, &entry
->ecx
, &entry
->edx
);
1251 static void do_cpuid_ent(struct kvm_cpuid_entry2
*entry
, u32 function
,
1252 u32 index
, int *nent
, int maxnent
)
1254 const u32 kvm_supported_word0_x86_features
= bit(X86_FEATURE_FPU
) |
1255 bit(X86_FEATURE_VME
) | bit(X86_FEATURE_DE
) |
1256 bit(X86_FEATURE_PSE
) | bit(X86_FEATURE_TSC
) |
1257 bit(X86_FEATURE_MSR
) | bit(X86_FEATURE_PAE
) |
1258 bit(X86_FEATURE_CX8
) | bit(X86_FEATURE_APIC
) |
1259 bit(X86_FEATURE_SEP
) | bit(X86_FEATURE_PGE
) |
1260 bit(X86_FEATURE_CMOV
) | bit(X86_FEATURE_PSE36
) |
1261 bit(X86_FEATURE_CLFLSH
) | bit(X86_FEATURE_MMX
) |
1262 bit(X86_FEATURE_FXSR
) | bit(X86_FEATURE_XMM
) |
1263 bit(X86_FEATURE_XMM2
) | bit(X86_FEATURE_SELFSNOOP
);
1264 const u32 kvm_supported_word1_x86_features
= bit(X86_FEATURE_FPU
) |
1265 bit(X86_FEATURE_VME
) | bit(X86_FEATURE_DE
) |
1266 bit(X86_FEATURE_PSE
) | bit(X86_FEATURE_TSC
) |
1267 bit(X86_FEATURE_MSR
) | bit(X86_FEATURE_PAE
) |
1268 bit(X86_FEATURE_CX8
) | bit(X86_FEATURE_APIC
) |
1269 bit(X86_FEATURE_PGE
) |
1270 bit(X86_FEATURE_CMOV
) | bit(X86_FEATURE_PSE36
) |
1271 bit(X86_FEATURE_MMX
) | bit(X86_FEATURE_FXSR
) |
1272 bit(X86_FEATURE_SYSCALL
) |
1273 (is_efer_nx() ? bit(X86_FEATURE_NX
) : 0) |
1274 #ifdef CONFIG_X86_64
1275 bit(X86_FEATURE_LM
) |
1277 bit(X86_FEATURE_FXSR_OPT
) |
1278 bit(X86_FEATURE_MMXEXT
) |
1279 bit(X86_FEATURE_3DNOWEXT
) |
1280 bit(X86_FEATURE_3DNOW
);
1281 const u32 kvm_supported_word3_x86_features
=
1282 bit(X86_FEATURE_XMM3
) | bit(X86_FEATURE_CX16
);
1283 const u32 kvm_supported_word6_x86_features
=
1284 bit(X86_FEATURE_LAHF_LM
) | bit(X86_FEATURE_CMP_LEGACY
) |
1285 bit(X86_FEATURE_SVM
);
1287 /* all calls to cpuid_count() should be made on the same cpu */
1289 do_cpuid_1_ent(entry
, function
, index
);
1294 entry
->eax
= min(entry
->eax
, (u32
)0xb);
1297 entry
->edx
&= kvm_supported_word0_x86_features
;
1298 entry
->ecx
&= kvm_supported_word3_x86_features
;
1300 /* function 2 entries are STATEFUL. That is, repeated cpuid commands
1301 * may return different values. This forces us to get_cpu() before
1302 * issuing the first command, and also to emulate this annoying behavior
1303 * in kvm_emulate_cpuid() using KVM_CPUID_FLAG_STATE_READ_NEXT */
1305 int t
, times
= entry
->eax
& 0xff;
1307 entry
->flags
|= KVM_CPUID_FLAG_STATEFUL_FUNC
;
1308 entry
->flags
|= KVM_CPUID_FLAG_STATE_READ_NEXT
;
1309 for (t
= 1; t
< times
&& *nent
< maxnent
; ++t
) {
1310 do_cpuid_1_ent(&entry
[t
], function
, 0);
1311 entry
[t
].flags
|= KVM_CPUID_FLAG_STATEFUL_FUNC
;
1316 /* function 4 and 0xb have additional index. */
1320 entry
->flags
|= KVM_CPUID_FLAG_SIGNIFCANT_INDEX
;
1321 /* read more entries until cache_type is zero */
1322 for (i
= 1; *nent
< maxnent
; ++i
) {
1323 cache_type
= entry
[i
- 1].eax
& 0x1f;
1326 do_cpuid_1_ent(&entry
[i
], function
, i
);
1328 KVM_CPUID_FLAG_SIGNIFCANT_INDEX
;
1336 entry
->flags
|= KVM_CPUID_FLAG_SIGNIFCANT_INDEX
;
1337 /* read more entries until level_type is zero */
1338 for (i
= 1; *nent
< maxnent
; ++i
) {
1339 level_type
= entry
[i
- 1].ecx
& 0xff00;
1342 do_cpuid_1_ent(&entry
[i
], function
, i
);
1344 KVM_CPUID_FLAG_SIGNIFCANT_INDEX
;
1350 entry
->eax
= min(entry
->eax
, 0x8000001a);
1353 entry
->edx
&= kvm_supported_word1_x86_features
;
1354 entry
->ecx
&= kvm_supported_word6_x86_features
;
1360 static int kvm_dev_ioctl_get_supported_cpuid(struct kvm_cpuid2
*cpuid
,
1361 struct kvm_cpuid_entry2 __user
*entries
)
1363 struct kvm_cpuid_entry2
*cpuid_entries
;
1364 int limit
, nent
= 0, r
= -E2BIG
;
1367 if (cpuid
->nent
< 1)
1370 cpuid_entries
= vmalloc(sizeof(struct kvm_cpuid_entry2
) * cpuid
->nent
);
1374 do_cpuid_ent(&cpuid_entries
[0], 0, 0, &nent
, cpuid
->nent
);
1375 limit
= cpuid_entries
[0].eax
;
1376 for (func
= 1; func
<= limit
&& nent
< cpuid
->nent
; ++func
)
1377 do_cpuid_ent(&cpuid_entries
[nent
], func
, 0,
1378 &nent
, cpuid
->nent
);
1380 if (nent
>= cpuid
->nent
)
1383 do_cpuid_ent(&cpuid_entries
[nent
], 0x80000000, 0, &nent
, cpuid
->nent
);
1384 limit
= cpuid_entries
[nent
- 1].eax
;
1385 for (func
= 0x80000001; func
<= limit
&& nent
< cpuid
->nent
; ++func
)
1386 do_cpuid_ent(&cpuid_entries
[nent
], func
, 0,
1387 &nent
, cpuid
->nent
);
1389 if (copy_to_user(entries
, cpuid_entries
,
1390 nent
* sizeof(struct kvm_cpuid_entry2
)))
1396 vfree(cpuid_entries
);
1401 static int kvm_vcpu_ioctl_get_lapic(struct kvm_vcpu
*vcpu
,
1402 struct kvm_lapic_state
*s
)
1405 memcpy(s
->regs
, vcpu
->arch
.apic
->regs
, sizeof *s
);
1411 static int kvm_vcpu_ioctl_set_lapic(struct kvm_vcpu
*vcpu
,
1412 struct kvm_lapic_state
*s
)
1415 memcpy(vcpu
->arch
.apic
->regs
, s
->regs
, sizeof *s
);
1416 kvm_apic_post_state_restore(vcpu
);
1422 static int kvm_vcpu_ioctl_interrupt(struct kvm_vcpu
*vcpu
,
1423 struct kvm_interrupt
*irq
)
1425 if (irq
->irq
< 0 || irq
->irq
>= 256)
1427 if (irqchip_in_kernel(vcpu
->kvm
))
1431 set_bit(irq
->irq
, vcpu
->arch
.irq_pending
);
1432 set_bit(irq
->irq
/ BITS_PER_LONG
, &vcpu
->arch
.irq_summary
);
1439 static int kvm_vcpu_ioctl_nmi(struct kvm_vcpu
*vcpu
)
1442 kvm_inject_nmi(vcpu
);
1448 static int vcpu_ioctl_tpr_access_reporting(struct kvm_vcpu
*vcpu
,
1449 struct kvm_tpr_access_ctl
*tac
)
1453 vcpu
->arch
.tpr_access_reporting
= !!tac
->enabled
;
1457 long kvm_arch_vcpu_ioctl(struct file
*filp
,
1458 unsigned int ioctl
, unsigned long arg
)
1460 struct kvm_vcpu
*vcpu
= filp
->private_data
;
1461 void __user
*argp
= (void __user
*)arg
;
1463 struct kvm_lapic_state
*lapic
= NULL
;
1466 case KVM_GET_LAPIC
: {
1467 lapic
= kzalloc(sizeof(struct kvm_lapic_state
), GFP_KERNEL
);
1472 r
= kvm_vcpu_ioctl_get_lapic(vcpu
, lapic
);
1476 if (copy_to_user(argp
, lapic
, sizeof(struct kvm_lapic_state
)))
1481 case KVM_SET_LAPIC
: {
1482 lapic
= kmalloc(sizeof(struct kvm_lapic_state
), GFP_KERNEL
);
1487 if (copy_from_user(lapic
, argp
, sizeof(struct kvm_lapic_state
)))
1489 r
= kvm_vcpu_ioctl_set_lapic(vcpu
, lapic
);
1495 case KVM_INTERRUPT
: {
1496 struct kvm_interrupt irq
;
1499 if (copy_from_user(&irq
, argp
, sizeof irq
))
1501 r
= kvm_vcpu_ioctl_interrupt(vcpu
, &irq
);
1508 r
= kvm_vcpu_ioctl_nmi(vcpu
);
1514 case KVM_SET_CPUID
: {
1515 struct kvm_cpuid __user
*cpuid_arg
= argp
;
1516 struct kvm_cpuid cpuid
;
1519 if (copy_from_user(&cpuid
, cpuid_arg
, sizeof cpuid
))
1521 r
= kvm_vcpu_ioctl_set_cpuid(vcpu
, &cpuid
, cpuid_arg
->entries
);
1526 case KVM_SET_CPUID2
: {
1527 struct kvm_cpuid2 __user
*cpuid_arg
= argp
;
1528 struct kvm_cpuid2 cpuid
;
1531 if (copy_from_user(&cpuid
, cpuid_arg
, sizeof cpuid
))
1533 r
= kvm_vcpu_ioctl_set_cpuid2(vcpu
, &cpuid
,
1534 cpuid_arg
->entries
);
1539 case KVM_GET_CPUID2
: {
1540 struct kvm_cpuid2 __user
*cpuid_arg
= argp
;
1541 struct kvm_cpuid2 cpuid
;
1544 if (copy_from_user(&cpuid
, cpuid_arg
, sizeof cpuid
))
1546 r
= kvm_vcpu_ioctl_get_cpuid2(vcpu
, &cpuid
,
1547 cpuid_arg
->entries
);
1551 if (copy_to_user(cpuid_arg
, &cpuid
, sizeof cpuid
))
1557 r
= msr_io(vcpu
, argp
, kvm_get_msr
, 1);
1560 r
= msr_io(vcpu
, argp
, do_set_msr
, 0);
1562 case KVM_TPR_ACCESS_REPORTING
: {
1563 struct kvm_tpr_access_ctl tac
;
1566 if (copy_from_user(&tac
, argp
, sizeof tac
))
1568 r
= vcpu_ioctl_tpr_access_reporting(vcpu
, &tac
);
1572 if (copy_to_user(argp
, &tac
, sizeof tac
))
1577 case KVM_SET_VAPIC_ADDR
: {
1578 struct kvm_vapic_addr va
;
1581 if (!irqchip_in_kernel(vcpu
->kvm
))
1584 if (copy_from_user(&va
, argp
, sizeof va
))
1587 kvm_lapic_set_vapic_addr(vcpu
, va
.vapic_addr
);
1598 static int kvm_vm_ioctl_set_tss_addr(struct kvm
*kvm
, unsigned long addr
)
1602 if (addr
> (unsigned int)(-3 * PAGE_SIZE
))
1604 ret
= kvm_x86_ops
->set_tss_addr(kvm
, addr
);
1608 static int kvm_vm_ioctl_set_nr_mmu_pages(struct kvm
*kvm
,
1609 u32 kvm_nr_mmu_pages
)
1611 if (kvm_nr_mmu_pages
< KVM_MIN_ALLOC_MMU_PAGES
)
1614 down_write(&kvm
->slots_lock
);
1616 kvm_mmu_change_mmu_pages(kvm
, kvm_nr_mmu_pages
);
1617 kvm
->arch
.n_requested_mmu_pages
= kvm_nr_mmu_pages
;
1619 up_write(&kvm
->slots_lock
);
1623 static int kvm_vm_ioctl_get_nr_mmu_pages(struct kvm
*kvm
)
1625 return kvm
->arch
.n_alloc_mmu_pages
;
1628 gfn_t
unalias_gfn(struct kvm
*kvm
, gfn_t gfn
)
1631 struct kvm_mem_alias
*alias
;
1633 for (i
= 0; i
< kvm
->arch
.naliases
; ++i
) {
1634 alias
= &kvm
->arch
.aliases
[i
];
1635 if (gfn
>= alias
->base_gfn
1636 && gfn
< alias
->base_gfn
+ alias
->npages
)
1637 return alias
->target_gfn
+ gfn
- alias
->base_gfn
;
1643 * Set a new alias region. Aliases map a portion of physical memory into
1644 * another portion. This is useful for memory windows, for example the PC
1647 static int kvm_vm_ioctl_set_memory_alias(struct kvm
*kvm
,
1648 struct kvm_memory_alias
*alias
)
1651 struct kvm_mem_alias
*p
;
1654 /* General sanity checks */
1655 if (alias
->memory_size
& (PAGE_SIZE
- 1))
1657 if (alias
->guest_phys_addr
& (PAGE_SIZE
- 1))
1659 if (alias
->slot
>= KVM_ALIAS_SLOTS
)
1661 if (alias
->guest_phys_addr
+ alias
->memory_size
1662 < alias
->guest_phys_addr
)
1664 if (alias
->target_phys_addr
+ alias
->memory_size
1665 < alias
->target_phys_addr
)
1668 down_write(&kvm
->slots_lock
);
1669 spin_lock(&kvm
->mmu_lock
);
1671 p
= &kvm
->arch
.aliases
[alias
->slot
];
1672 p
->base_gfn
= alias
->guest_phys_addr
>> PAGE_SHIFT
;
1673 p
->npages
= alias
->memory_size
>> PAGE_SHIFT
;
1674 p
->target_gfn
= alias
->target_phys_addr
>> PAGE_SHIFT
;
1676 for (n
= KVM_ALIAS_SLOTS
; n
> 0; --n
)
1677 if (kvm
->arch
.aliases
[n
- 1].npages
)
1679 kvm
->arch
.naliases
= n
;
1681 spin_unlock(&kvm
->mmu_lock
);
1682 kvm_mmu_zap_all(kvm
);
1684 up_write(&kvm
->slots_lock
);
1692 static int kvm_vm_ioctl_get_irqchip(struct kvm
*kvm
, struct kvm_irqchip
*chip
)
1697 switch (chip
->chip_id
) {
1698 case KVM_IRQCHIP_PIC_MASTER
:
1699 memcpy(&chip
->chip
.pic
,
1700 &pic_irqchip(kvm
)->pics
[0],
1701 sizeof(struct kvm_pic_state
));
1703 case KVM_IRQCHIP_PIC_SLAVE
:
1704 memcpy(&chip
->chip
.pic
,
1705 &pic_irqchip(kvm
)->pics
[1],
1706 sizeof(struct kvm_pic_state
));
1708 case KVM_IRQCHIP_IOAPIC
:
1709 memcpy(&chip
->chip
.ioapic
,
1710 ioapic_irqchip(kvm
),
1711 sizeof(struct kvm_ioapic_state
));
1720 static int kvm_vm_ioctl_set_irqchip(struct kvm
*kvm
, struct kvm_irqchip
*chip
)
1725 switch (chip
->chip_id
) {
1726 case KVM_IRQCHIP_PIC_MASTER
:
1727 memcpy(&pic_irqchip(kvm
)->pics
[0],
1729 sizeof(struct kvm_pic_state
));
1731 case KVM_IRQCHIP_PIC_SLAVE
:
1732 memcpy(&pic_irqchip(kvm
)->pics
[1],
1734 sizeof(struct kvm_pic_state
));
1736 case KVM_IRQCHIP_IOAPIC
:
1737 memcpy(ioapic_irqchip(kvm
),
1739 sizeof(struct kvm_ioapic_state
));
1745 kvm_pic_update_irq(pic_irqchip(kvm
));
1749 static int kvm_vm_ioctl_get_pit(struct kvm
*kvm
, struct kvm_pit_state
*ps
)
1753 memcpy(ps
, &kvm
->arch
.vpit
->pit_state
, sizeof(struct kvm_pit_state
));
1757 static int kvm_vm_ioctl_set_pit(struct kvm
*kvm
, struct kvm_pit_state
*ps
)
1761 memcpy(&kvm
->arch
.vpit
->pit_state
, ps
, sizeof(struct kvm_pit_state
));
1762 kvm_pit_load_count(kvm
, 0, ps
->channels
[0].count
);
1766 static int kvm_vm_ioctl_reinject(struct kvm
*kvm
,
1767 struct kvm_reinject_control
*control
)
1769 if (!kvm
->arch
.vpit
)
1771 kvm
->arch
.vpit
->pit_state
.pit_timer
.reinject
= control
->pit_reinject
;
1776 * Get (and clear) the dirty memory log for a memory slot.
1778 int kvm_vm_ioctl_get_dirty_log(struct kvm
*kvm
,
1779 struct kvm_dirty_log
*log
)
1783 struct kvm_memory_slot
*memslot
;
1786 down_write(&kvm
->slots_lock
);
1788 r
= kvm_get_dirty_log(kvm
, log
, &is_dirty
);
1792 /* If nothing is dirty, don't bother messing with page tables. */
1794 kvm_mmu_slot_remove_write_access(kvm
, log
->slot
);
1795 kvm_flush_remote_tlbs(kvm
);
1796 memslot
= &kvm
->memslots
[log
->slot
];
1797 n
= ALIGN(memslot
->npages
, BITS_PER_LONG
) / 8;
1798 memset(memslot
->dirty_bitmap
, 0, n
);
1802 up_write(&kvm
->slots_lock
);
1806 long kvm_arch_vm_ioctl(struct file
*filp
,
1807 unsigned int ioctl
, unsigned long arg
)
1809 struct kvm
*kvm
= filp
->private_data
;
1810 void __user
*argp
= (void __user
*)arg
;
1813 * This union makes it completely explicit to gcc-3.x
1814 * that these two variables' stack usage should be
1815 * combined, not added together.
1818 struct kvm_pit_state ps
;
1819 struct kvm_memory_alias alias
;
1823 case KVM_SET_TSS_ADDR
:
1824 r
= kvm_vm_ioctl_set_tss_addr(kvm
, arg
);
1828 case KVM_SET_MEMORY_REGION
: {
1829 struct kvm_memory_region kvm_mem
;
1830 struct kvm_userspace_memory_region kvm_userspace_mem
;
1833 if (copy_from_user(&kvm_mem
, argp
, sizeof kvm_mem
))
1835 kvm_userspace_mem
.slot
= kvm_mem
.slot
;
1836 kvm_userspace_mem
.flags
= kvm_mem
.flags
;
1837 kvm_userspace_mem
.guest_phys_addr
= kvm_mem
.guest_phys_addr
;
1838 kvm_userspace_mem
.memory_size
= kvm_mem
.memory_size
;
1839 r
= kvm_vm_ioctl_set_memory_region(kvm
, &kvm_userspace_mem
, 0);
1844 case KVM_SET_NR_MMU_PAGES
:
1845 r
= kvm_vm_ioctl_set_nr_mmu_pages(kvm
, arg
);
1849 case KVM_GET_NR_MMU_PAGES
:
1850 r
= kvm_vm_ioctl_get_nr_mmu_pages(kvm
);
1852 case KVM_SET_MEMORY_ALIAS
:
1854 if (copy_from_user(&u
.alias
, argp
, sizeof(struct kvm_memory_alias
)))
1856 r
= kvm_vm_ioctl_set_memory_alias(kvm
, &u
.alias
);
1860 case KVM_CREATE_IRQCHIP
:
1862 kvm
->arch
.vpic
= kvm_create_pic(kvm
);
1863 if (kvm
->arch
.vpic
) {
1864 r
= kvm_ioapic_init(kvm
);
1866 kfree(kvm
->arch
.vpic
);
1867 kvm
->arch
.vpic
= NULL
;
1872 r
= kvm_setup_default_irq_routing(kvm
);
1874 kfree(kvm
->arch
.vpic
);
1875 kfree(kvm
->arch
.vioapic
);
1879 case KVM_CREATE_PIT
:
1880 mutex_lock(&kvm
->lock
);
1883 goto create_pit_unlock
;
1885 kvm
->arch
.vpit
= kvm_create_pit(kvm
);
1889 mutex_unlock(&kvm
->lock
);
1891 case KVM_IRQ_LINE_STATUS
:
1892 case KVM_IRQ_LINE
: {
1893 struct kvm_irq_level irq_event
;
1896 if (copy_from_user(&irq_event
, argp
, sizeof irq_event
))
1898 if (irqchip_in_kernel(kvm
)) {
1900 mutex_lock(&kvm
->lock
);
1901 status
= kvm_set_irq(kvm
, KVM_USERSPACE_IRQ_SOURCE_ID
,
1902 irq_event
.irq
, irq_event
.level
);
1903 mutex_unlock(&kvm
->lock
);
1904 if (ioctl
== KVM_IRQ_LINE_STATUS
) {
1905 irq_event
.status
= status
;
1906 if (copy_to_user(argp
, &irq_event
,
1914 case KVM_GET_IRQCHIP
: {
1915 /* 0: PIC master, 1: PIC slave, 2: IOAPIC */
1916 struct kvm_irqchip
*chip
= kmalloc(sizeof(*chip
), GFP_KERNEL
);
1922 if (copy_from_user(chip
, argp
, sizeof *chip
))
1923 goto get_irqchip_out
;
1925 if (!irqchip_in_kernel(kvm
))
1926 goto get_irqchip_out
;
1927 r
= kvm_vm_ioctl_get_irqchip(kvm
, chip
);
1929 goto get_irqchip_out
;
1931 if (copy_to_user(argp
, chip
, sizeof *chip
))
1932 goto get_irqchip_out
;
1940 case KVM_SET_IRQCHIP
: {
1941 /* 0: PIC master, 1: PIC slave, 2: IOAPIC */
1942 struct kvm_irqchip
*chip
= kmalloc(sizeof(*chip
), GFP_KERNEL
);
1948 if (copy_from_user(chip
, argp
, sizeof *chip
))
1949 goto set_irqchip_out
;
1951 if (!irqchip_in_kernel(kvm
))
1952 goto set_irqchip_out
;
1953 r
= kvm_vm_ioctl_set_irqchip(kvm
, chip
);
1955 goto set_irqchip_out
;
1965 if (copy_from_user(&u
.ps
, argp
, sizeof(struct kvm_pit_state
)))
1968 if (!kvm
->arch
.vpit
)
1970 r
= kvm_vm_ioctl_get_pit(kvm
, &u
.ps
);
1974 if (copy_to_user(argp
, &u
.ps
, sizeof(struct kvm_pit_state
)))
1981 if (copy_from_user(&u
.ps
, argp
, sizeof u
.ps
))
1984 if (!kvm
->arch
.vpit
)
1986 r
= kvm_vm_ioctl_set_pit(kvm
, &u
.ps
);
1992 case KVM_REINJECT_CONTROL
: {
1993 struct kvm_reinject_control control
;
1995 if (copy_from_user(&control
, argp
, sizeof(control
)))
1997 r
= kvm_vm_ioctl_reinject(kvm
, &control
);
2010 static void kvm_init_msr_list(void)
2015 for (i
= j
= 0; i
< ARRAY_SIZE(msrs_to_save
); i
++) {
2016 if (rdmsr_safe(msrs_to_save
[i
], &dummy
[0], &dummy
[1]) < 0)
2019 msrs_to_save
[j
] = msrs_to_save
[i
];
2022 num_msrs_to_save
= j
;
2026 * Only apic need an MMIO device hook, so shortcut now..
2028 static struct kvm_io_device
*vcpu_find_pervcpu_dev(struct kvm_vcpu
*vcpu
,
2029 gpa_t addr
, int len
,
2032 struct kvm_io_device
*dev
;
2034 if (vcpu
->arch
.apic
) {
2035 dev
= &vcpu
->arch
.apic
->dev
;
2036 if (dev
->in_range(dev
, addr
, len
, is_write
))
2043 static struct kvm_io_device
*vcpu_find_mmio_dev(struct kvm_vcpu
*vcpu
,
2044 gpa_t addr
, int len
,
2047 struct kvm_io_device
*dev
;
2049 dev
= vcpu_find_pervcpu_dev(vcpu
, addr
, len
, is_write
);
2051 dev
= kvm_io_bus_find_dev(&vcpu
->kvm
->mmio_bus
, addr
, len
,
2056 static int kvm_read_guest_virt(gva_t addr
, void *val
, unsigned int bytes
,
2057 struct kvm_vcpu
*vcpu
)
2060 int r
= X86EMUL_CONTINUE
;
2063 gpa_t gpa
= vcpu
->arch
.mmu
.gva_to_gpa(vcpu
, addr
);
2064 unsigned offset
= addr
& (PAGE_SIZE
-1);
2065 unsigned toread
= min(bytes
, (unsigned)PAGE_SIZE
- offset
);
2068 if (gpa
== UNMAPPED_GVA
) {
2069 r
= X86EMUL_PROPAGATE_FAULT
;
2072 ret
= kvm_read_guest(vcpu
->kvm
, gpa
, data
, toread
);
2074 r
= X86EMUL_UNHANDLEABLE
;
2086 static int kvm_write_guest_virt(gva_t addr
, void *val
, unsigned int bytes
,
2087 struct kvm_vcpu
*vcpu
)
2090 int r
= X86EMUL_CONTINUE
;
2093 gpa_t gpa
= vcpu
->arch
.mmu
.gva_to_gpa(vcpu
, addr
);
2094 unsigned offset
= addr
& (PAGE_SIZE
-1);
2095 unsigned towrite
= min(bytes
, (unsigned)PAGE_SIZE
- offset
);
2098 if (gpa
== UNMAPPED_GVA
) {
2099 r
= X86EMUL_PROPAGATE_FAULT
;
2102 ret
= kvm_write_guest(vcpu
->kvm
, gpa
, data
, towrite
);
2104 r
= X86EMUL_UNHANDLEABLE
;
2117 static int emulator_read_emulated(unsigned long addr
,
2120 struct kvm_vcpu
*vcpu
)
2122 struct kvm_io_device
*mmio_dev
;
2125 if (vcpu
->mmio_read_completed
) {
2126 memcpy(val
, vcpu
->mmio_data
, bytes
);
2127 vcpu
->mmio_read_completed
= 0;
2128 return X86EMUL_CONTINUE
;
2131 gpa
= vcpu
->arch
.mmu
.gva_to_gpa(vcpu
, addr
);
2133 /* For APIC access vmexit */
2134 if ((gpa
& PAGE_MASK
) == APIC_DEFAULT_PHYS_BASE
)
2137 if (kvm_read_guest_virt(addr
, val
, bytes
, vcpu
)
2138 == X86EMUL_CONTINUE
)
2139 return X86EMUL_CONTINUE
;
2140 if (gpa
== UNMAPPED_GVA
)
2141 return X86EMUL_PROPAGATE_FAULT
;
2145 * Is this MMIO handled locally?
2147 mutex_lock(&vcpu
->kvm
->lock
);
2148 mmio_dev
= vcpu_find_mmio_dev(vcpu
, gpa
, bytes
, 0);
2150 kvm_iodevice_read(mmio_dev
, gpa
, bytes
, val
);
2151 mutex_unlock(&vcpu
->kvm
->lock
);
2152 return X86EMUL_CONTINUE
;
2154 mutex_unlock(&vcpu
->kvm
->lock
);
2156 vcpu
->mmio_needed
= 1;
2157 vcpu
->mmio_phys_addr
= gpa
;
2158 vcpu
->mmio_size
= bytes
;
2159 vcpu
->mmio_is_write
= 0;
2161 return X86EMUL_UNHANDLEABLE
;
2164 int emulator_write_phys(struct kvm_vcpu
*vcpu
, gpa_t gpa
,
2165 const void *val
, int bytes
)
2169 ret
= kvm_write_guest(vcpu
->kvm
, gpa
, val
, bytes
);
2172 kvm_mmu_pte_write(vcpu
, gpa
, val
, bytes
, 1);
2176 static int emulator_write_emulated_onepage(unsigned long addr
,
2179 struct kvm_vcpu
*vcpu
)
2181 struct kvm_io_device
*mmio_dev
;
2184 gpa
= vcpu
->arch
.mmu
.gva_to_gpa(vcpu
, addr
);
2186 if (gpa
== UNMAPPED_GVA
) {
2187 kvm_inject_page_fault(vcpu
, addr
, 2);
2188 return X86EMUL_PROPAGATE_FAULT
;
2191 /* For APIC access vmexit */
2192 if ((gpa
& PAGE_MASK
) == APIC_DEFAULT_PHYS_BASE
)
2195 if (emulator_write_phys(vcpu
, gpa
, val
, bytes
))
2196 return X86EMUL_CONTINUE
;
2200 * Is this MMIO handled locally?
2202 mutex_lock(&vcpu
->kvm
->lock
);
2203 mmio_dev
= vcpu_find_mmio_dev(vcpu
, gpa
, bytes
, 1);
2205 kvm_iodevice_write(mmio_dev
, gpa
, bytes
, val
);
2206 mutex_unlock(&vcpu
->kvm
->lock
);
2207 return X86EMUL_CONTINUE
;
2209 mutex_unlock(&vcpu
->kvm
->lock
);
2211 vcpu
->mmio_needed
= 1;
2212 vcpu
->mmio_phys_addr
= gpa
;
2213 vcpu
->mmio_size
= bytes
;
2214 vcpu
->mmio_is_write
= 1;
2215 memcpy(vcpu
->mmio_data
, val
, bytes
);
2217 return X86EMUL_CONTINUE
;
2220 int emulator_write_emulated(unsigned long addr
,
2223 struct kvm_vcpu
*vcpu
)
2225 /* Crossing a page boundary? */
2226 if (((addr
+ bytes
- 1) ^ addr
) & PAGE_MASK
) {
2229 now
= -addr
& ~PAGE_MASK
;
2230 rc
= emulator_write_emulated_onepage(addr
, val
, now
, vcpu
);
2231 if (rc
!= X86EMUL_CONTINUE
)
2237 return emulator_write_emulated_onepage(addr
, val
, bytes
, vcpu
);
2239 EXPORT_SYMBOL_GPL(emulator_write_emulated
);
2241 static int emulator_cmpxchg_emulated(unsigned long addr
,
2245 struct kvm_vcpu
*vcpu
)
2247 static int reported
;
2251 printk(KERN_WARNING
"kvm: emulating exchange as write\n");
2253 #ifndef CONFIG_X86_64
2254 /* guests cmpxchg8b have to be emulated atomically */
2261 gpa
= vcpu
->arch
.mmu
.gva_to_gpa(vcpu
, addr
);
2263 if (gpa
== UNMAPPED_GVA
||
2264 (gpa
& PAGE_MASK
) == APIC_DEFAULT_PHYS_BASE
)
2267 if (((gpa
+ bytes
- 1) & PAGE_MASK
) != (gpa
& PAGE_MASK
))
2272 page
= gfn_to_page(vcpu
->kvm
, gpa
>> PAGE_SHIFT
);
2274 kaddr
= kmap_atomic(page
, KM_USER0
);
2275 set_64bit((u64
*)(kaddr
+ offset_in_page(gpa
)), val
);
2276 kunmap_atomic(kaddr
, KM_USER0
);
2277 kvm_release_page_dirty(page
);
2282 return emulator_write_emulated(addr
, new, bytes
, vcpu
);
2285 static unsigned long get_segment_base(struct kvm_vcpu
*vcpu
, int seg
)
2287 return kvm_x86_ops
->get_segment_base(vcpu
, seg
);
2290 int emulate_invlpg(struct kvm_vcpu
*vcpu
, gva_t address
)
2292 kvm_mmu_invlpg(vcpu
, address
);
2293 return X86EMUL_CONTINUE
;
2296 int emulate_clts(struct kvm_vcpu
*vcpu
)
2298 KVMTRACE_0D(CLTS
, vcpu
, handler
);
2299 kvm_x86_ops
->set_cr0(vcpu
, vcpu
->arch
.cr0
& ~X86_CR0_TS
);
2300 return X86EMUL_CONTINUE
;
2303 int emulator_get_dr(struct x86_emulate_ctxt
*ctxt
, int dr
, unsigned long *dest
)
2305 struct kvm_vcpu
*vcpu
= ctxt
->vcpu
;
2309 *dest
= kvm_x86_ops
->get_dr(vcpu
, dr
);
2310 return X86EMUL_CONTINUE
;
2312 pr_unimpl(vcpu
, "%s: unexpected dr %u\n", __func__
, dr
);
2313 return X86EMUL_UNHANDLEABLE
;
2317 int emulator_set_dr(struct x86_emulate_ctxt
*ctxt
, int dr
, unsigned long value
)
2319 unsigned long mask
= (ctxt
->mode
== X86EMUL_MODE_PROT64
) ? ~0ULL : ~0U;
2322 kvm_x86_ops
->set_dr(ctxt
->vcpu
, dr
, value
& mask
, &exception
);
2324 /* FIXME: better handling */
2325 return X86EMUL_UNHANDLEABLE
;
2327 return X86EMUL_CONTINUE
;
2330 void kvm_report_emulation_failure(struct kvm_vcpu
*vcpu
, const char *context
)
2333 unsigned long rip
= kvm_rip_read(vcpu
);
2334 unsigned long rip_linear
;
2336 if (!printk_ratelimit())
2339 rip_linear
= rip
+ get_segment_base(vcpu
, VCPU_SREG_CS
);
2341 kvm_read_guest_virt(rip_linear
, (void *)opcodes
, 4, vcpu
);
2343 printk(KERN_ERR
"emulation failed (%s) rip %lx %02x %02x %02x %02x\n",
2344 context
, rip
, opcodes
[0], opcodes
[1], opcodes
[2], opcodes
[3]);
2346 EXPORT_SYMBOL_GPL(kvm_report_emulation_failure
);
2348 static struct x86_emulate_ops emulate_ops
= {
2349 .read_std
= kvm_read_guest_virt
,
2350 .read_emulated
= emulator_read_emulated
,
2351 .write_emulated
= emulator_write_emulated
,
2352 .cmpxchg_emulated
= emulator_cmpxchg_emulated
,
2355 static void cache_all_regs(struct kvm_vcpu
*vcpu
)
2357 kvm_register_read(vcpu
, VCPU_REGS_RAX
);
2358 kvm_register_read(vcpu
, VCPU_REGS_RSP
);
2359 kvm_register_read(vcpu
, VCPU_REGS_RIP
);
2360 vcpu
->arch
.regs_dirty
= ~0;
2363 int emulate_instruction(struct kvm_vcpu
*vcpu
,
2364 struct kvm_run
*run
,
2370 struct decode_cache
*c
;
2372 kvm_clear_exception_queue(vcpu
);
2373 vcpu
->arch
.mmio_fault_cr2
= cr2
;
2375 * TODO: fix x86_emulate.c to use guest_read/write_register
2376 * instead of direct ->regs accesses, can save hundred cycles
2377 * on Intel for instructions that don't read/change RSP, for
2380 cache_all_regs(vcpu
);
2382 vcpu
->mmio_is_write
= 0;
2383 vcpu
->arch
.pio
.string
= 0;
2385 if (!(emulation_type
& EMULTYPE_NO_DECODE
)) {
2387 kvm_x86_ops
->get_cs_db_l_bits(vcpu
, &cs_db
, &cs_l
);
2389 vcpu
->arch
.emulate_ctxt
.vcpu
= vcpu
;
2390 vcpu
->arch
.emulate_ctxt
.eflags
= kvm_x86_ops
->get_rflags(vcpu
);
2391 vcpu
->arch
.emulate_ctxt
.mode
=
2392 (vcpu
->arch
.emulate_ctxt
.eflags
& X86_EFLAGS_VM
)
2393 ? X86EMUL_MODE_REAL
: cs_l
2394 ? X86EMUL_MODE_PROT64
: cs_db
2395 ? X86EMUL_MODE_PROT32
: X86EMUL_MODE_PROT16
;
2397 r
= x86_decode_insn(&vcpu
->arch
.emulate_ctxt
, &emulate_ops
);
2399 /* Reject the instructions other than VMCALL/VMMCALL when
2400 * try to emulate invalid opcode */
2401 c
= &vcpu
->arch
.emulate_ctxt
.decode
;
2402 if ((emulation_type
& EMULTYPE_TRAP_UD
) &&
2403 (!(c
->twobyte
&& c
->b
== 0x01 &&
2404 (c
->modrm_reg
== 0 || c
->modrm_reg
== 3) &&
2405 c
->modrm_mod
== 3 && c
->modrm_rm
== 1)))
2406 return EMULATE_FAIL
;
2408 ++vcpu
->stat
.insn_emulation
;
2410 ++vcpu
->stat
.insn_emulation_fail
;
2411 if (kvm_mmu_unprotect_page_virt(vcpu
, cr2
))
2412 return EMULATE_DONE
;
2413 return EMULATE_FAIL
;
2417 if (emulation_type
& EMULTYPE_SKIP
) {
2418 kvm_rip_write(vcpu
, vcpu
->arch
.emulate_ctxt
.decode
.eip
);
2419 return EMULATE_DONE
;
2422 r
= x86_emulate_insn(&vcpu
->arch
.emulate_ctxt
, &emulate_ops
);
2424 if (vcpu
->arch
.pio
.string
)
2425 return EMULATE_DO_MMIO
;
2427 if ((r
|| vcpu
->mmio_is_write
) && run
) {
2428 run
->exit_reason
= KVM_EXIT_MMIO
;
2429 run
->mmio
.phys_addr
= vcpu
->mmio_phys_addr
;
2430 memcpy(run
->mmio
.data
, vcpu
->mmio_data
, 8);
2431 run
->mmio
.len
= vcpu
->mmio_size
;
2432 run
->mmio
.is_write
= vcpu
->mmio_is_write
;
2436 if (kvm_mmu_unprotect_page_virt(vcpu
, cr2
))
2437 return EMULATE_DONE
;
2438 if (!vcpu
->mmio_needed
) {
2439 kvm_report_emulation_failure(vcpu
, "mmio");
2440 return EMULATE_FAIL
;
2442 return EMULATE_DO_MMIO
;
2445 kvm_x86_ops
->set_rflags(vcpu
, vcpu
->arch
.emulate_ctxt
.eflags
);
2447 if (vcpu
->mmio_is_write
) {
2448 vcpu
->mmio_needed
= 0;
2449 return EMULATE_DO_MMIO
;
2452 return EMULATE_DONE
;
2454 EXPORT_SYMBOL_GPL(emulate_instruction
);
2456 static int pio_copy_data(struct kvm_vcpu
*vcpu
)
2458 void *p
= vcpu
->arch
.pio_data
;
2459 gva_t q
= vcpu
->arch
.pio
.guest_gva
;
2463 bytes
= vcpu
->arch
.pio
.size
* vcpu
->arch
.pio
.cur_count
;
2464 if (vcpu
->arch
.pio
.in
)
2465 ret
= kvm_write_guest_virt(q
, p
, bytes
, vcpu
);
2467 ret
= kvm_read_guest_virt(q
, p
, bytes
, vcpu
);
2471 int complete_pio(struct kvm_vcpu
*vcpu
)
2473 struct kvm_pio_request
*io
= &vcpu
->arch
.pio
;
2480 val
= kvm_register_read(vcpu
, VCPU_REGS_RAX
);
2481 memcpy(&val
, vcpu
->arch
.pio_data
, io
->size
);
2482 kvm_register_write(vcpu
, VCPU_REGS_RAX
, val
);
2486 r
= pio_copy_data(vcpu
);
2493 delta
*= io
->cur_count
;
2495 * The size of the register should really depend on
2496 * current address size.
2498 val
= kvm_register_read(vcpu
, VCPU_REGS_RCX
);
2500 kvm_register_write(vcpu
, VCPU_REGS_RCX
, val
);
2506 val
= kvm_register_read(vcpu
, VCPU_REGS_RDI
);
2508 kvm_register_write(vcpu
, VCPU_REGS_RDI
, val
);
2510 val
= kvm_register_read(vcpu
, VCPU_REGS_RSI
);
2512 kvm_register_write(vcpu
, VCPU_REGS_RSI
, val
);
2516 io
->count
-= io
->cur_count
;
2522 static void kernel_pio(struct kvm_io_device
*pio_dev
,
2523 struct kvm_vcpu
*vcpu
,
2526 /* TODO: String I/O for in kernel device */
2528 mutex_lock(&vcpu
->kvm
->lock
);
2529 if (vcpu
->arch
.pio
.in
)
2530 kvm_iodevice_read(pio_dev
, vcpu
->arch
.pio
.port
,
2531 vcpu
->arch
.pio
.size
,
2534 kvm_iodevice_write(pio_dev
, vcpu
->arch
.pio
.port
,
2535 vcpu
->arch
.pio
.size
,
2537 mutex_unlock(&vcpu
->kvm
->lock
);
2540 static void pio_string_write(struct kvm_io_device
*pio_dev
,
2541 struct kvm_vcpu
*vcpu
)
2543 struct kvm_pio_request
*io
= &vcpu
->arch
.pio
;
2544 void *pd
= vcpu
->arch
.pio_data
;
2547 mutex_lock(&vcpu
->kvm
->lock
);
2548 for (i
= 0; i
< io
->cur_count
; i
++) {
2549 kvm_iodevice_write(pio_dev
, io
->port
,
2554 mutex_unlock(&vcpu
->kvm
->lock
);
2557 static struct kvm_io_device
*vcpu_find_pio_dev(struct kvm_vcpu
*vcpu
,
2558 gpa_t addr
, int len
,
2561 return kvm_io_bus_find_dev(&vcpu
->kvm
->pio_bus
, addr
, len
, is_write
);
2564 int kvm_emulate_pio(struct kvm_vcpu
*vcpu
, struct kvm_run
*run
, int in
,
2565 int size
, unsigned port
)
2567 struct kvm_io_device
*pio_dev
;
2570 vcpu
->run
->exit_reason
= KVM_EXIT_IO
;
2571 vcpu
->run
->io
.direction
= in
? KVM_EXIT_IO_IN
: KVM_EXIT_IO_OUT
;
2572 vcpu
->run
->io
.size
= vcpu
->arch
.pio
.size
= size
;
2573 vcpu
->run
->io
.data_offset
= KVM_PIO_PAGE_OFFSET
* PAGE_SIZE
;
2574 vcpu
->run
->io
.count
= vcpu
->arch
.pio
.count
= vcpu
->arch
.pio
.cur_count
= 1;
2575 vcpu
->run
->io
.port
= vcpu
->arch
.pio
.port
= port
;
2576 vcpu
->arch
.pio
.in
= in
;
2577 vcpu
->arch
.pio
.string
= 0;
2578 vcpu
->arch
.pio
.down
= 0;
2579 vcpu
->arch
.pio
.rep
= 0;
2581 if (vcpu
->run
->io
.direction
== KVM_EXIT_IO_IN
)
2582 KVMTRACE_2D(IO_READ
, vcpu
, vcpu
->run
->io
.port
, (u32
)size
,
2585 KVMTRACE_2D(IO_WRITE
, vcpu
, vcpu
->run
->io
.port
, (u32
)size
,
2588 val
= kvm_register_read(vcpu
, VCPU_REGS_RAX
);
2589 memcpy(vcpu
->arch
.pio_data
, &val
, 4);
2591 pio_dev
= vcpu_find_pio_dev(vcpu
, port
, size
, !in
);
2593 kernel_pio(pio_dev
, vcpu
, vcpu
->arch
.pio_data
);
2599 EXPORT_SYMBOL_GPL(kvm_emulate_pio
);
2601 int kvm_emulate_pio_string(struct kvm_vcpu
*vcpu
, struct kvm_run
*run
, int in
,
2602 int size
, unsigned long count
, int down
,
2603 gva_t address
, int rep
, unsigned port
)
2605 unsigned now
, in_page
;
2607 struct kvm_io_device
*pio_dev
;
2609 vcpu
->run
->exit_reason
= KVM_EXIT_IO
;
2610 vcpu
->run
->io
.direction
= in
? KVM_EXIT_IO_IN
: KVM_EXIT_IO_OUT
;
2611 vcpu
->run
->io
.size
= vcpu
->arch
.pio
.size
= size
;
2612 vcpu
->run
->io
.data_offset
= KVM_PIO_PAGE_OFFSET
* PAGE_SIZE
;
2613 vcpu
->run
->io
.count
= vcpu
->arch
.pio
.count
= vcpu
->arch
.pio
.cur_count
= count
;
2614 vcpu
->run
->io
.port
= vcpu
->arch
.pio
.port
= port
;
2615 vcpu
->arch
.pio
.in
= in
;
2616 vcpu
->arch
.pio
.string
= 1;
2617 vcpu
->arch
.pio
.down
= down
;
2618 vcpu
->arch
.pio
.rep
= rep
;
2620 if (vcpu
->run
->io
.direction
== KVM_EXIT_IO_IN
)
2621 KVMTRACE_2D(IO_READ
, vcpu
, vcpu
->run
->io
.port
, (u32
)size
,
2624 KVMTRACE_2D(IO_WRITE
, vcpu
, vcpu
->run
->io
.port
, (u32
)size
,
2628 kvm_x86_ops
->skip_emulated_instruction(vcpu
);
2633 in_page
= PAGE_SIZE
- offset_in_page(address
);
2635 in_page
= offset_in_page(address
) + size
;
2636 now
= min(count
, (unsigned long)in_page
/ size
);
2641 * String I/O in reverse. Yuck. Kill the guest, fix later.
2643 pr_unimpl(vcpu
, "guest string pio down\n");
2644 kvm_inject_gp(vcpu
, 0);
2647 vcpu
->run
->io
.count
= now
;
2648 vcpu
->arch
.pio
.cur_count
= now
;
2650 if (vcpu
->arch
.pio
.cur_count
== vcpu
->arch
.pio
.count
)
2651 kvm_x86_ops
->skip_emulated_instruction(vcpu
);
2653 vcpu
->arch
.pio
.guest_gva
= address
;
2655 pio_dev
= vcpu_find_pio_dev(vcpu
, port
,
2656 vcpu
->arch
.pio
.cur_count
,
2657 !vcpu
->arch
.pio
.in
);
2658 if (!vcpu
->arch
.pio
.in
) {
2659 /* string PIO write */
2660 ret
= pio_copy_data(vcpu
);
2661 if (ret
== X86EMUL_PROPAGATE_FAULT
) {
2662 kvm_inject_gp(vcpu
, 0);
2665 if (ret
== 0 && pio_dev
) {
2666 pio_string_write(pio_dev
, vcpu
);
2668 if (vcpu
->arch
.pio
.count
== 0)
2672 pr_unimpl(vcpu
, "no string pio read support yet, "
2673 "port %x size %d count %ld\n",
2678 EXPORT_SYMBOL_GPL(kvm_emulate_pio_string
);
2680 static void bounce_off(void *info
)
2685 static unsigned int ref_freq
;
2686 static unsigned long tsc_khz_ref
;
2688 static int kvmclock_cpufreq_notifier(struct notifier_block
*nb
, unsigned long val
,
2691 struct cpufreq_freqs
*freq
= data
;
2693 struct kvm_vcpu
*vcpu
;
2694 int i
, send_ipi
= 0;
2697 ref_freq
= freq
->old
;
2699 if (val
== CPUFREQ_PRECHANGE
&& freq
->old
> freq
->new)
2701 if (val
== CPUFREQ_POSTCHANGE
&& freq
->old
< freq
->new)
2703 per_cpu(cpu_tsc_khz
, freq
->cpu
) = cpufreq_scale(tsc_khz_ref
, ref_freq
, freq
->new);
2705 spin_lock(&kvm_lock
);
2706 list_for_each_entry(kvm
, &vm_list
, vm_list
) {
2707 for (i
= 0; i
< KVM_MAX_VCPUS
; ++i
) {
2708 vcpu
= kvm
->vcpus
[i
];
2711 if (vcpu
->cpu
!= freq
->cpu
)
2713 if (!kvm_request_guest_time_update(vcpu
))
2715 if (vcpu
->cpu
!= smp_processor_id())
2719 spin_unlock(&kvm_lock
);
2721 if (freq
->old
< freq
->new && send_ipi
) {
2723 * We upscale the frequency. Must make the guest
2724 * doesn't see old kvmclock values while running with
2725 * the new frequency, otherwise we risk the guest sees
2726 * time go backwards.
2728 * In case we update the frequency for another cpu
2729 * (which might be in guest context) send an interrupt
2730 * to kick the cpu out of guest context. Next time
2731 * guest context is entered kvmclock will be updated,
2732 * so the guest will not see stale values.
2734 smp_call_function_single(freq
->cpu
, bounce_off
, NULL
, 1);
2739 static struct notifier_block kvmclock_cpufreq_notifier_block
= {
2740 .notifier_call
= kvmclock_cpufreq_notifier
2743 int kvm_arch_init(void *opaque
)
2746 struct kvm_x86_ops
*ops
= (struct kvm_x86_ops
*)opaque
;
2749 printk(KERN_ERR
"kvm: already loaded the other module\n");
2754 if (!ops
->cpu_has_kvm_support()) {
2755 printk(KERN_ERR
"kvm: no hardware support\n");
2759 if (ops
->disabled_by_bios()) {
2760 printk(KERN_ERR
"kvm: disabled by bios\n");
2765 r
= kvm_mmu_module_init();
2769 kvm_init_msr_list();
2772 kvm_mmu_set_nonpresent_ptes(0ull, 0ull);
2773 kvm_mmu_set_base_ptes(PT_PRESENT_MASK
);
2774 kvm_mmu_set_mask_ptes(PT_USER_MASK
, PT_ACCESSED_MASK
,
2775 PT_DIRTY_MASK
, PT64_NX_MASK
, 0, 0);
2777 for_each_possible_cpu(cpu
)
2778 per_cpu(cpu_tsc_khz
, cpu
) = tsc_khz
;
2779 if (!boot_cpu_has(X86_FEATURE_CONSTANT_TSC
)) {
2780 tsc_khz_ref
= tsc_khz
;
2781 cpufreq_register_notifier(&kvmclock_cpufreq_notifier_block
,
2782 CPUFREQ_TRANSITION_NOTIFIER
);
2791 void kvm_arch_exit(void)
2793 if (!boot_cpu_has(X86_FEATURE_CONSTANT_TSC
))
2794 cpufreq_unregister_notifier(&kvmclock_cpufreq_notifier_block
,
2795 CPUFREQ_TRANSITION_NOTIFIER
);
2797 kvm_mmu_module_exit();
2800 int kvm_emulate_halt(struct kvm_vcpu
*vcpu
)
2802 ++vcpu
->stat
.halt_exits
;
2803 KVMTRACE_0D(HLT
, vcpu
, handler
);
2804 if (irqchip_in_kernel(vcpu
->kvm
)) {
2805 vcpu
->arch
.mp_state
= KVM_MP_STATE_HALTED
;
2808 vcpu
->run
->exit_reason
= KVM_EXIT_HLT
;
2812 EXPORT_SYMBOL_GPL(kvm_emulate_halt
);
2814 static inline gpa_t
hc_gpa(struct kvm_vcpu
*vcpu
, unsigned long a0
,
2817 if (is_long_mode(vcpu
))
2820 return a0
| ((gpa_t
)a1
<< 32);
2823 int kvm_emulate_hypercall(struct kvm_vcpu
*vcpu
)
2825 unsigned long nr
, a0
, a1
, a2
, a3
, ret
;
2828 nr
= kvm_register_read(vcpu
, VCPU_REGS_RAX
);
2829 a0
= kvm_register_read(vcpu
, VCPU_REGS_RBX
);
2830 a1
= kvm_register_read(vcpu
, VCPU_REGS_RCX
);
2831 a2
= kvm_register_read(vcpu
, VCPU_REGS_RDX
);
2832 a3
= kvm_register_read(vcpu
, VCPU_REGS_RSI
);
2834 KVMTRACE_1D(VMMCALL
, vcpu
, (u32
)nr
, handler
);
2836 if (!is_long_mode(vcpu
)) {
2845 case KVM_HC_VAPIC_POLL_IRQ
:
2849 r
= kvm_pv_mmu_op(vcpu
, a0
, hc_gpa(vcpu
, a1
, a2
), &ret
);
2855 kvm_register_write(vcpu
, VCPU_REGS_RAX
, ret
);
2856 ++vcpu
->stat
.hypercalls
;
2859 EXPORT_SYMBOL_GPL(kvm_emulate_hypercall
);
2861 int kvm_fix_hypercall(struct kvm_vcpu
*vcpu
)
2863 char instruction
[3];
2865 unsigned long rip
= kvm_rip_read(vcpu
);
2869 * Blow out the MMU to ensure that no other VCPU has an active mapping
2870 * to ensure that the updated hypercall appears atomically across all
2873 kvm_mmu_zap_all(vcpu
->kvm
);
2875 kvm_x86_ops
->patch_hypercall(vcpu
, instruction
);
2876 if (emulator_write_emulated(rip
, instruction
, 3, vcpu
)
2877 != X86EMUL_CONTINUE
)
2883 static u64
mk_cr_64(u64 curr_cr
, u32 new_val
)
2885 return (curr_cr
& ~((1ULL << 32) - 1)) | new_val
;
2888 void realmode_lgdt(struct kvm_vcpu
*vcpu
, u16 limit
, unsigned long base
)
2890 struct descriptor_table dt
= { limit
, base
};
2892 kvm_x86_ops
->set_gdt(vcpu
, &dt
);
2895 void realmode_lidt(struct kvm_vcpu
*vcpu
, u16 limit
, unsigned long base
)
2897 struct descriptor_table dt
= { limit
, base
};
2899 kvm_x86_ops
->set_idt(vcpu
, &dt
);
2902 void realmode_lmsw(struct kvm_vcpu
*vcpu
, unsigned long msw
,
2903 unsigned long *rflags
)
2905 kvm_lmsw(vcpu
, msw
);
2906 *rflags
= kvm_x86_ops
->get_rflags(vcpu
);
2909 unsigned long realmode_get_cr(struct kvm_vcpu
*vcpu
, int cr
)
2911 unsigned long value
;
2913 kvm_x86_ops
->decache_cr4_guest_bits(vcpu
);
2916 value
= vcpu
->arch
.cr0
;
2919 value
= vcpu
->arch
.cr2
;
2922 value
= vcpu
->arch
.cr3
;
2925 value
= vcpu
->arch
.cr4
;
2928 value
= kvm_get_cr8(vcpu
);
2931 vcpu_printf(vcpu
, "%s: unexpected cr %u\n", __func__
, cr
);
2934 KVMTRACE_3D(CR_READ
, vcpu
, (u32
)cr
, (u32
)value
,
2935 (u32
)((u64
)value
>> 32), handler
);
2940 void realmode_set_cr(struct kvm_vcpu
*vcpu
, int cr
, unsigned long val
,
2941 unsigned long *rflags
)
2943 KVMTRACE_3D(CR_WRITE
, vcpu
, (u32
)cr
, (u32
)val
,
2944 (u32
)((u64
)val
>> 32), handler
);
2948 kvm_set_cr0(vcpu
, mk_cr_64(vcpu
->arch
.cr0
, val
));
2949 *rflags
= kvm_x86_ops
->get_rflags(vcpu
);
2952 vcpu
->arch
.cr2
= val
;
2955 kvm_set_cr3(vcpu
, val
);
2958 kvm_set_cr4(vcpu
, mk_cr_64(vcpu
->arch
.cr4
, val
));
2961 kvm_set_cr8(vcpu
, val
& 0xfUL
);
2964 vcpu_printf(vcpu
, "%s: unexpected cr %u\n", __func__
, cr
);
2968 static int move_to_next_stateful_cpuid_entry(struct kvm_vcpu
*vcpu
, int i
)
2970 struct kvm_cpuid_entry2
*e
= &vcpu
->arch
.cpuid_entries
[i
];
2971 int j
, nent
= vcpu
->arch
.cpuid_nent
;
2973 e
->flags
&= ~KVM_CPUID_FLAG_STATE_READ_NEXT
;
2974 /* when no next entry is found, the current entry[i] is reselected */
2975 for (j
= i
+ 1; ; j
= (j
+ 1) % nent
) {
2976 struct kvm_cpuid_entry2
*ej
= &vcpu
->arch
.cpuid_entries
[j
];
2977 if (ej
->function
== e
->function
) {
2978 ej
->flags
|= KVM_CPUID_FLAG_STATE_READ_NEXT
;
2982 return 0; /* silence gcc, even though control never reaches here */
2985 /* find an entry with matching function, matching index (if needed), and that
2986 * should be read next (if it's stateful) */
2987 static int is_matching_cpuid_entry(struct kvm_cpuid_entry2
*e
,
2988 u32 function
, u32 index
)
2990 if (e
->function
!= function
)
2992 if ((e
->flags
& KVM_CPUID_FLAG_SIGNIFCANT_INDEX
) && e
->index
!= index
)
2994 if ((e
->flags
& KVM_CPUID_FLAG_STATEFUL_FUNC
) &&
2995 !(e
->flags
& KVM_CPUID_FLAG_STATE_READ_NEXT
))
3000 struct kvm_cpuid_entry2
*kvm_find_cpuid_entry(struct kvm_vcpu
*vcpu
,
3001 u32 function
, u32 index
)
3004 struct kvm_cpuid_entry2
*best
= NULL
;
3006 for (i
= 0; i
< vcpu
->arch
.cpuid_nent
; ++i
) {
3007 struct kvm_cpuid_entry2
*e
;
3009 e
= &vcpu
->arch
.cpuid_entries
[i
];
3010 if (is_matching_cpuid_entry(e
, function
, index
)) {
3011 if (e
->flags
& KVM_CPUID_FLAG_STATEFUL_FUNC
)
3012 move_to_next_stateful_cpuid_entry(vcpu
, i
);
3017 * Both basic or both extended?
3019 if (((e
->function
^ function
) & 0x80000000) == 0)
3020 if (!best
|| e
->function
> best
->function
)
3026 int cpuid_maxphyaddr(struct kvm_vcpu
*vcpu
)
3028 struct kvm_cpuid_entry2
*best
;
3030 best
= kvm_find_cpuid_entry(vcpu
, 0x80000008, 0);
3032 return best
->eax
& 0xff;
3036 void kvm_emulate_cpuid(struct kvm_vcpu
*vcpu
)
3038 u32 function
, index
;
3039 struct kvm_cpuid_entry2
*best
;
3041 function
= kvm_register_read(vcpu
, VCPU_REGS_RAX
);
3042 index
= kvm_register_read(vcpu
, VCPU_REGS_RCX
);
3043 kvm_register_write(vcpu
, VCPU_REGS_RAX
, 0);
3044 kvm_register_write(vcpu
, VCPU_REGS_RBX
, 0);
3045 kvm_register_write(vcpu
, VCPU_REGS_RCX
, 0);
3046 kvm_register_write(vcpu
, VCPU_REGS_RDX
, 0);
3047 best
= kvm_find_cpuid_entry(vcpu
, function
, index
);
3049 kvm_register_write(vcpu
, VCPU_REGS_RAX
, best
->eax
);
3050 kvm_register_write(vcpu
, VCPU_REGS_RBX
, best
->ebx
);
3051 kvm_register_write(vcpu
, VCPU_REGS_RCX
, best
->ecx
);
3052 kvm_register_write(vcpu
, VCPU_REGS_RDX
, best
->edx
);
3054 kvm_x86_ops
->skip_emulated_instruction(vcpu
);
3055 KVMTRACE_5D(CPUID
, vcpu
, function
,
3056 (u32
)kvm_register_read(vcpu
, VCPU_REGS_RAX
),
3057 (u32
)kvm_register_read(vcpu
, VCPU_REGS_RBX
),
3058 (u32
)kvm_register_read(vcpu
, VCPU_REGS_RCX
),
3059 (u32
)kvm_register_read(vcpu
, VCPU_REGS_RDX
), handler
);
3061 EXPORT_SYMBOL_GPL(kvm_emulate_cpuid
);
3064 * Check if userspace requested an interrupt window, and that the
3065 * interrupt window is open.
3067 * No need to exit to userspace if we already have an interrupt queued.
3069 static int dm_request_for_irq_injection(struct kvm_vcpu
*vcpu
,
3070 struct kvm_run
*kvm_run
)
3072 return (!irqchip_in_kernel(vcpu
->kvm
) && !kvm_cpu_has_interrupt(vcpu
) &&
3073 kvm_run
->request_interrupt_window
&&
3074 kvm_arch_interrupt_allowed(vcpu
));
3077 static void post_kvm_run_save(struct kvm_vcpu
*vcpu
,
3078 struct kvm_run
*kvm_run
)
3080 kvm_run
->if_flag
= (kvm_x86_ops
->get_rflags(vcpu
) & X86_EFLAGS_IF
) != 0;
3081 kvm_run
->cr8
= kvm_get_cr8(vcpu
);
3082 kvm_run
->apic_base
= kvm_get_apic_base(vcpu
);
3083 if (irqchip_in_kernel(vcpu
->kvm
))
3084 kvm_run
->ready_for_interrupt_injection
= 1;
3086 kvm_run
->ready_for_interrupt_injection
=
3087 (kvm_arch_interrupt_allowed(vcpu
) &&
3088 !kvm_cpu_has_interrupt(vcpu
));
3091 static void vapic_enter(struct kvm_vcpu
*vcpu
)
3093 struct kvm_lapic
*apic
= vcpu
->arch
.apic
;
3096 if (!apic
|| !apic
->vapic_addr
)
3099 page
= gfn_to_page(vcpu
->kvm
, apic
->vapic_addr
>> PAGE_SHIFT
);
3101 vcpu
->arch
.apic
->vapic_page
= page
;
3104 static void vapic_exit(struct kvm_vcpu
*vcpu
)
3106 struct kvm_lapic
*apic
= vcpu
->arch
.apic
;
3108 if (!apic
|| !apic
->vapic_addr
)
3111 down_read(&vcpu
->kvm
->slots_lock
);
3112 kvm_release_page_dirty(apic
->vapic_page
);
3113 mark_page_dirty(vcpu
->kvm
, apic
->vapic_addr
>> PAGE_SHIFT
);
3114 up_read(&vcpu
->kvm
->slots_lock
);
3117 static int vcpu_enter_guest(struct kvm_vcpu
*vcpu
, struct kvm_run
*kvm_run
)
3122 if (test_and_clear_bit(KVM_REQ_MMU_RELOAD
, &vcpu
->requests
))
3123 kvm_mmu_unload(vcpu
);
3125 r
= kvm_mmu_reload(vcpu
);
3129 if (vcpu
->requests
) {
3130 if (test_and_clear_bit(KVM_REQ_MIGRATE_TIMER
, &vcpu
->requests
))
3131 __kvm_migrate_timers(vcpu
);
3132 if (test_and_clear_bit(KVM_REQ_KVMCLOCK_UPDATE
, &vcpu
->requests
))
3133 kvm_write_guest_time(vcpu
);
3134 if (test_and_clear_bit(KVM_REQ_MMU_SYNC
, &vcpu
->requests
))
3135 kvm_mmu_sync_roots(vcpu
);
3136 if (test_and_clear_bit(KVM_REQ_TLB_FLUSH
, &vcpu
->requests
))
3137 kvm_x86_ops
->tlb_flush(vcpu
);
3138 if (test_and_clear_bit(KVM_REQ_REPORT_TPR_ACCESS
,
3140 kvm_run
->exit_reason
= KVM_EXIT_TPR_ACCESS
;
3144 if (test_and_clear_bit(KVM_REQ_TRIPLE_FAULT
, &vcpu
->requests
)) {
3145 kvm_run
->exit_reason
= KVM_EXIT_SHUTDOWN
;
3153 kvm_x86_ops
->prepare_guest_switch(vcpu
);
3154 kvm_load_guest_fpu(vcpu
);
3156 local_irq_disable();
3158 if (vcpu
->requests
|| need_resched() || signal_pending(current
)) {
3165 vcpu
->guest_mode
= 1;
3167 * Make sure that guest_mode assignment won't happen after
3168 * testing the pending IRQ vector bitmap.
3172 if (vcpu
->arch
.exception
.pending
)
3173 __queue_exception(vcpu
);
3175 kvm_x86_ops
->inject_pending_irq(vcpu
, kvm_run
);
3177 kvm_lapic_sync_to_vapic(vcpu
);
3179 up_read(&vcpu
->kvm
->slots_lock
);
3183 get_debugreg(vcpu
->arch
.host_dr6
, 6);
3184 get_debugreg(vcpu
->arch
.host_dr7
, 7);
3185 if (unlikely(vcpu
->arch
.switch_db_regs
)) {
3186 get_debugreg(vcpu
->arch
.host_db
[0], 0);
3187 get_debugreg(vcpu
->arch
.host_db
[1], 1);
3188 get_debugreg(vcpu
->arch
.host_db
[2], 2);
3189 get_debugreg(vcpu
->arch
.host_db
[3], 3);
3192 set_debugreg(vcpu
->arch
.eff_db
[0], 0);
3193 set_debugreg(vcpu
->arch
.eff_db
[1], 1);
3194 set_debugreg(vcpu
->arch
.eff_db
[2], 2);
3195 set_debugreg(vcpu
->arch
.eff_db
[3], 3);
3198 KVMTRACE_0D(VMENTRY
, vcpu
, entryexit
);
3199 kvm_x86_ops
->run(vcpu
, kvm_run
);
3201 if (unlikely(vcpu
->arch
.switch_db_regs
)) {
3203 set_debugreg(vcpu
->arch
.host_db
[0], 0);
3204 set_debugreg(vcpu
->arch
.host_db
[1], 1);
3205 set_debugreg(vcpu
->arch
.host_db
[2], 2);
3206 set_debugreg(vcpu
->arch
.host_db
[3], 3);
3208 set_debugreg(vcpu
->arch
.host_dr6
, 6);
3209 set_debugreg(vcpu
->arch
.host_dr7
, 7);
3211 vcpu
->guest_mode
= 0;
3217 * We must have an instruction between local_irq_enable() and
3218 * kvm_guest_exit(), so the timer interrupt isn't delayed by
3219 * the interrupt shadow. The stat.exits increment will do nicely.
3220 * But we need to prevent reordering, hence this barrier():
3228 down_read(&vcpu
->kvm
->slots_lock
);
3231 * Profile KVM exit RIPs:
3233 if (unlikely(prof_on
== KVM_PROFILING
)) {
3234 unsigned long rip
= kvm_rip_read(vcpu
);
3235 profile_hit(KVM_PROFILING
, (void *)rip
);
3239 kvm_lapic_sync_from_vapic(vcpu
);
3241 r
= kvm_x86_ops
->handle_exit(kvm_run
, vcpu
);
3247 static int __vcpu_run(struct kvm_vcpu
*vcpu
, struct kvm_run
*kvm_run
)
3251 if (unlikely(vcpu
->arch
.mp_state
== KVM_MP_STATE_SIPI_RECEIVED
)) {
3252 pr_debug("vcpu %d received sipi with vector # %x\n",
3253 vcpu
->vcpu_id
, vcpu
->arch
.sipi_vector
);
3254 kvm_lapic_reset(vcpu
);
3255 r
= kvm_arch_vcpu_reset(vcpu
);
3258 vcpu
->arch
.mp_state
= KVM_MP_STATE_RUNNABLE
;
3261 down_read(&vcpu
->kvm
->slots_lock
);
3266 if (vcpu
->arch
.mp_state
== KVM_MP_STATE_RUNNABLE
)
3267 r
= vcpu_enter_guest(vcpu
, kvm_run
);
3269 up_read(&vcpu
->kvm
->slots_lock
);
3270 kvm_vcpu_block(vcpu
);
3271 down_read(&vcpu
->kvm
->slots_lock
);
3272 if (test_and_clear_bit(KVM_REQ_UNHALT
, &vcpu
->requests
))
3274 switch(vcpu
->arch
.mp_state
) {
3275 case KVM_MP_STATE_HALTED
:
3276 vcpu
->arch
.mp_state
=
3277 KVM_MP_STATE_RUNNABLE
;
3278 case KVM_MP_STATE_RUNNABLE
:
3280 case KVM_MP_STATE_SIPI_RECEIVED
:
3291 clear_bit(KVM_REQ_PENDING_TIMER
, &vcpu
->requests
);
3292 if (kvm_cpu_has_pending_timer(vcpu
))
3293 kvm_inject_pending_timer_irqs(vcpu
);
3295 if (dm_request_for_irq_injection(vcpu
, kvm_run
)) {
3297 kvm_run
->exit_reason
= KVM_EXIT_INTR
;
3298 ++vcpu
->stat
.request_irq_exits
;
3300 if (signal_pending(current
)) {
3302 kvm_run
->exit_reason
= KVM_EXIT_INTR
;
3303 ++vcpu
->stat
.signal_exits
;
3305 if (need_resched()) {
3306 up_read(&vcpu
->kvm
->slots_lock
);
3308 down_read(&vcpu
->kvm
->slots_lock
);
3312 up_read(&vcpu
->kvm
->slots_lock
);
3313 post_kvm_run_save(vcpu
, kvm_run
);
3320 int kvm_arch_vcpu_ioctl_run(struct kvm_vcpu
*vcpu
, struct kvm_run
*kvm_run
)
3327 if (vcpu
->sigset_active
)
3328 sigprocmask(SIG_SETMASK
, &vcpu
->sigset
, &sigsaved
);
3330 if (unlikely(vcpu
->arch
.mp_state
== KVM_MP_STATE_UNINITIALIZED
)) {
3331 kvm_vcpu_block(vcpu
);
3332 clear_bit(KVM_REQ_UNHALT
, &vcpu
->requests
);
3337 /* re-sync apic's tpr */
3338 if (!irqchip_in_kernel(vcpu
->kvm
))
3339 kvm_set_cr8(vcpu
, kvm_run
->cr8
);
3341 if (vcpu
->arch
.pio
.cur_count
) {
3342 r
= complete_pio(vcpu
);
3346 #if CONFIG_HAS_IOMEM
3347 if (vcpu
->mmio_needed
) {
3348 memcpy(vcpu
->mmio_data
, kvm_run
->mmio
.data
, 8);
3349 vcpu
->mmio_read_completed
= 1;
3350 vcpu
->mmio_needed
= 0;
3352 down_read(&vcpu
->kvm
->slots_lock
);
3353 r
= emulate_instruction(vcpu
, kvm_run
,
3354 vcpu
->arch
.mmio_fault_cr2
, 0,
3355 EMULTYPE_NO_DECODE
);
3356 up_read(&vcpu
->kvm
->slots_lock
);
3357 if (r
== EMULATE_DO_MMIO
) {
3359 * Read-modify-write. Back to userspace.
3366 if (kvm_run
->exit_reason
== KVM_EXIT_HYPERCALL
)
3367 kvm_register_write(vcpu
, VCPU_REGS_RAX
,
3368 kvm_run
->hypercall
.ret
);
3370 r
= __vcpu_run(vcpu
, kvm_run
);
3373 if (vcpu
->sigset_active
)
3374 sigprocmask(SIG_SETMASK
, &sigsaved
, NULL
);
3380 int kvm_arch_vcpu_ioctl_get_regs(struct kvm_vcpu
*vcpu
, struct kvm_regs
*regs
)
3384 regs
->rax
= kvm_register_read(vcpu
, VCPU_REGS_RAX
);
3385 regs
->rbx
= kvm_register_read(vcpu
, VCPU_REGS_RBX
);
3386 regs
->rcx
= kvm_register_read(vcpu
, VCPU_REGS_RCX
);
3387 regs
->rdx
= kvm_register_read(vcpu
, VCPU_REGS_RDX
);
3388 regs
->rsi
= kvm_register_read(vcpu
, VCPU_REGS_RSI
);
3389 regs
->rdi
= kvm_register_read(vcpu
, VCPU_REGS_RDI
);
3390 regs
->rsp
= kvm_register_read(vcpu
, VCPU_REGS_RSP
);
3391 regs
->rbp
= kvm_register_read(vcpu
, VCPU_REGS_RBP
);
3392 #ifdef CONFIG_X86_64
3393 regs
->r8
= kvm_register_read(vcpu
, VCPU_REGS_R8
);
3394 regs
->r9
= kvm_register_read(vcpu
, VCPU_REGS_R9
);
3395 regs
->r10
= kvm_register_read(vcpu
, VCPU_REGS_R10
);
3396 regs
->r11
= kvm_register_read(vcpu
, VCPU_REGS_R11
);
3397 regs
->r12
= kvm_register_read(vcpu
, VCPU_REGS_R12
);
3398 regs
->r13
= kvm_register_read(vcpu
, VCPU_REGS_R13
);
3399 regs
->r14
= kvm_register_read(vcpu
, VCPU_REGS_R14
);
3400 regs
->r15
= kvm_register_read(vcpu
, VCPU_REGS_R15
);
3403 regs
->rip
= kvm_rip_read(vcpu
);
3404 regs
->rflags
= kvm_x86_ops
->get_rflags(vcpu
);
3407 * Don't leak debug flags in case they were set for guest debugging
3409 if (vcpu
->guest_debug
& KVM_GUESTDBG_SINGLESTEP
)
3410 regs
->rflags
&= ~(X86_EFLAGS_TF
| X86_EFLAGS_RF
);
3417 int kvm_arch_vcpu_ioctl_set_regs(struct kvm_vcpu
*vcpu
, struct kvm_regs
*regs
)
3421 kvm_register_write(vcpu
, VCPU_REGS_RAX
, regs
->rax
);
3422 kvm_register_write(vcpu
, VCPU_REGS_RBX
, regs
->rbx
);
3423 kvm_register_write(vcpu
, VCPU_REGS_RCX
, regs
->rcx
);
3424 kvm_register_write(vcpu
, VCPU_REGS_RDX
, regs
->rdx
);
3425 kvm_register_write(vcpu
, VCPU_REGS_RSI
, regs
->rsi
);
3426 kvm_register_write(vcpu
, VCPU_REGS_RDI
, regs
->rdi
);
3427 kvm_register_write(vcpu
, VCPU_REGS_RSP
, regs
->rsp
);
3428 kvm_register_write(vcpu
, VCPU_REGS_RBP
, regs
->rbp
);
3429 #ifdef CONFIG_X86_64
3430 kvm_register_write(vcpu
, VCPU_REGS_R8
, regs
->r8
);
3431 kvm_register_write(vcpu
, VCPU_REGS_R9
, regs
->r9
);
3432 kvm_register_write(vcpu
, VCPU_REGS_R10
, regs
->r10
);
3433 kvm_register_write(vcpu
, VCPU_REGS_R11
, regs
->r11
);
3434 kvm_register_write(vcpu
, VCPU_REGS_R12
, regs
->r12
);
3435 kvm_register_write(vcpu
, VCPU_REGS_R13
, regs
->r13
);
3436 kvm_register_write(vcpu
, VCPU_REGS_R14
, regs
->r14
);
3437 kvm_register_write(vcpu
, VCPU_REGS_R15
, regs
->r15
);
3441 kvm_rip_write(vcpu
, regs
->rip
);
3442 kvm_x86_ops
->set_rflags(vcpu
, regs
->rflags
);
3445 vcpu
->arch
.exception
.pending
= false;
3452 void kvm_get_segment(struct kvm_vcpu
*vcpu
,
3453 struct kvm_segment
*var
, int seg
)
3455 kvm_x86_ops
->get_segment(vcpu
, var
, seg
);
3458 void kvm_get_cs_db_l_bits(struct kvm_vcpu
*vcpu
, int *db
, int *l
)
3460 struct kvm_segment cs
;
3462 kvm_get_segment(vcpu
, &cs
, VCPU_SREG_CS
);
3466 EXPORT_SYMBOL_GPL(kvm_get_cs_db_l_bits
);
3468 int kvm_arch_vcpu_ioctl_get_sregs(struct kvm_vcpu
*vcpu
,
3469 struct kvm_sregs
*sregs
)
3471 struct descriptor_table dt
;
3476 kvm_get_segment(vcpu
, &sregs
->cs
, VCPU_SREG_CS
);
3477 kvm_get_segment(vcpu
, &sregs
->ds
, VCPU_SREG_DS
);
3478 kvm_get_segment(vcpu
, &sregs
->es
, VCPU_SREG_ES
);
3479 kvm_get_segment(vcpu
, &sregs
->fs
, VCPU_SREG_FS
);
3480 kvm_get_segment(vcpu
, &sregs
->gs
, VCPU_SREG_GS
);
3481 kvm_get_segment(vcpu
, &sregs
->ss
, VCPU_SREG_SS
);
3483 kvm_get_segment(vcpu
, &sregs
->tr
, VCPU_SREG_TR
);
3484 kvm_get_segment(vcpu
, &sregs
->ldt
, VCPU_SREG_LDTR
);
3486 kvm_x86_ops
->get_idt(vcpu
, &dt
);
3487 sregs
->idt
.limit
= dt
.limit
;
3488 sregs
->idt
.base
= dt
.base
;
3489 kvm_x86_ops
->get_gdt(vcpu
, &dt
);
3490 sregs
->gdt
.limit
= dt
.limit
;
3491 sregs
->gdt
.base
= dt
.base
;
3493 kvm_x86_ops
->decache_cr4_guest_bits(vcpu
);
3494 sregs
->cr0
= vcpu
->arch
.cr0
;
3495 sregs
->cr2
= vcpu
->arch
.cr2
;
3496 sregs
->cr3
= vcpu
->arch
.cr3
;
3497 sregs
->cr4
= vcpu
->arch
.cr4
;
3498 sregs
->cr8
= kvm_get_cr8(vcpu
);
3499 sregs
->efer
= vcpu
->arch
.shadow_efer
;
3500 sregs
->apic_base
= kvm_get_apic_base(vcpu
);
3502 if (irqchip_in_kernel(vcpu
->kvm
)) {
3503 memset(sregs
->interrupt_bitmap
, 0,
3504 sizeof sregs
->interrupt_bitmap
);
3505 pending_vec
= kvm_x86_ops
->get_irq(vcpu
);
3506 if (pending_vec
>= 0)
3507 set_bit(pending_vec
,
3508 (unsigned long *)sregs
->interrupt_bitmap
);
3510 memcpy(sregs
->interrupt_bitmap
, vcpu
->arch
.irq_pending
,
3511 sizeof sregs
->interrupt_bitmap
);
3518 int kvm_arch_vcpu_ioctl_get_mpstate(struct kvm_vcpu
*vcpu
,
3519 struct kvm_mp_state
*mp_state
)
3522 mp_state
->mp_state
= vcpu
->arch
.mp_state
;
3527 int kvm_arch_vcpu_ioctl_set_mpstate(struct kvm_vcpu
*vcpu
,
3528 struct kvm_mp_state
*mp_state
)
3531 vcpu
->arch
.mp_state
= mp_state
->mp_state
;
3536 static void kvm_set_segment(struct kvm_vcpu
*vcpu
,
3537 struct kvm_segment
*var
, int seg
)
3539 kvm_x86_ops
->set_segment(vcpu
, var
, seg
);
3542 static void seg_desct_to_kvm_desct(struct desc_struct
*seg_desc
, u16 selector
,
3543 struct kvm_segment
*kvm_desct
)
3545 kvm_desct
->base
= seg_desc
->base0
;
3546 kvm_desct
->base
|= seg_desc
->base1
<< 16;
3547 kvm_desct
->base
|= seg_desc
->base2
<< 24;
3548 kvm_desct
->limit
= seg_desc
->limit0
;
3549 kvm_desct
->limit
|= seg_desc
->limit
<< 16;
3551 kvm_desct
->limit
<<= 12;
3552 kvm_desct
->limit
|= 0xfff;
3554 kvm_desct
->selector
= selector
;
3555 kvm_desct
->type
= seg_desc
->type
;
3556 kvm_desct
->present
= seg_desc
->p
;
3557 kvm_desct
->dpl
= seg_desc
->dpl
;
3558 kvm_desct
->db
= seg_desc
->d
;
3559 kvm_desct
->s
= seg_desc
->s
;
3560 kvm_desct
->l
= seg_desc
->l
;
3561 kvm_desct
->g
= seg_desc
->g
;
3562 kvm_desct
->avl
= seg_desc
->avl
;
3564 kvm_desct
->unusable
= 1;
3566 kvm_desct
->unusable
= 0;
3567 kvm_desct
->padding
= 0;
3570 static void get_segment_descriptor_dtable(struct kvm_vcpu
*vcpu
,
3572 struct descriptor_table
*dtable
)
3574 if (selector
& 1 << 2) {
3575 struct kvm_segment kvm_seg
;
3577 kvm_get_segment(vcpu
, &kvm_seg
, VCPU_SREG_LDTR
);
3579 if (kvm_seg
.unusable
)
3582 dtable
->limit
= kvm_seg
.limit
;
3583 dtable
->base
= kvm_seg
.base
;
3586 kvm_x86_ops
->get_gdt(vcpu
, dtable
);
3589 /* allowed just for 8 bytes segments */
3590 static int load_guest_segment_descriptor(struct kvm_vcpu
*vcpu
, u16 selector
,
3591 struct desc_struct
*seg_desc
)
3594 struct descriptor_table dtable
;
3595 u16 index
= selector
>> 3;
3597 get_segment_descriptor_dtable(vcpu
, selector
, &dtable
);
3599 if (dtable
.limit
< index
* 8 + 7) {
3600 kvm_queue_exception_e(vcpu
, GP_VECTOR
, selector
& 0xfffc);
3603 gpa
= vcpu
->arch
.mmu
.gva_to_gpa(vcpu
, dtable
.base
);
3605 return kvm_read_guest(vcpu
->kvm
, gpa
, seg_desc
, 8);
3608 /* allowed just for 8 bytes segments */
3609 static int save_guest_segment_descriptor(struct kvm_vcpu
*vcpu
, u16 selector
,
3610 struct desc_struct
*seg_desc
)
3613 struct descriptor_table dtable
;
3614 u16 index
= selector
>> 3;
3616 get_segment_descriptor_dtable(vcpu
, selector
, &dtable
);
3618 if (dtable
.limit
< index
* 8 + 7)
3620 gpa
= vcpu
->arch
.mmu
.gva_to_gpa(vcpu
, dtable
.base
);
3622 return kvm_write_guest(vcpu
->kvm
, gpa
, seg_desc
, 8);
3625 static u32
get_tss_base_addr(struct kvm_vcpu
*vcpu
,
3626 struct desc_struct
*seg_desc
)
3630 base_addr
= seg_desc
->base0
;
3631 base_addr
|= (seg_desc
->base1
<< 16);
3632 base_addr
|= (seg_desc
->base2
<< 24);
3634 return vcpu
->arch
.mmu
.gva_to_gpa(vcpu
, base_addr
);
3637 static u16
get_segment_selector(struct kvm_vcpu
*vcpu
, int seg
)
3639 struct kvm_segment kvm_seg
;
3641 kvm_get_segment(vcpu
, &kvm_seg
, seg
);
3642 return kvm_seg
.selector
;
3645 static int load_segment_descriptor_to_kvm_desct(struct kvm_vcpu
*vcpu
,
3647 struct kvm_segment
*kvm_seg
)
3649 struct desc_struct seg_desc
;
3651 if (load_guest_segment_descriptor(vcpu
, selector
, &seg_desc
))
3653 seg_desct_to_kvm_desct(&seg_desc
, selector
, kvm_seg
);
3657 static int kvm_load_realmode_segment(struct kvm_vcpu
*vcpu
, u16 selector
, int seg
)
3659 struct kvm_segment segvar
= {
3660 .base
= selector
<< 4,
3662 .selector
= selector
,
3673 kvm_x86_ops
->set_segment(vcpu
, &segvar
, seg
);
3677 int kvm_load_segment_descriptor(struct kvm_vcpu
*vcpu
, u16 selector
,
3678 int type_bits
, int seg
)
3680 struct kvm_segment kvm_seg
;
3682 if (!(vcpu
->arch
.cr0
& X86_CR0_PE
))
3683 return kvm_load_realmode_segment(vcpu
, selector
, seg
);
3684 if (load_segment_descriptor_to_kvm_desct(vcpu
, selector
, &kvm_seg
))
3686 kvm_seg
.type
|= type_bits
;
3688 if (seg
!= VCPU_SREG_SS
&& seg
!= VCPU_SREG_CS
&&
3689 seg
!= VCPU_SREG_LDTR
)
3691 kvm_seg
.unusable
= 1;
3693 kvm_set_segment(vcpu
, &kvm_seg
, seg
);
3697 static void save_state_to_tss32(struct kvm_vcpu
*vcpu
,
3698 struct tss_segment_32
*tss
)
3700 tss
->cr3
= vcpu
->arch
.cr3
;
3701 tss
->eip
= kvm_rip_read(vcpu
);
3702 tss
->eflags
= kvm_x86_ops
->get_rflags(vcpu
);
3703 tss
->eax
= kvm_register_read(vcpu
, VCPU_REGS_RAX
);
3704 tss
->ecx
= kvm_register_read(vcpu
, VCPU_REGS_RCX
);
3705 tss
->edx
= kvm_register_read(vcpu
, VCPU_REGS_RDX
);
3706 tss
->ebx
= kvm_register_read(vcpu
, VCPU_REGS_RBX
);
3707 tss
->esp
= kvm_register_read(vcpu
, VCPU_REGS_RSP
);
3708 tss
->ebp
= kvm_register_read(vcpu
, VCPU_REGS_RBP
);
3709 tss
->esi
= kvm_register_read(vcpu
, VCPU_REGS_RSI
);
3710 tss
->edi
= kvm_register_read(vcpu
, VCPU_REGS_RDI
);
3711 tss
->es
= get_segment_selector(vcpu
, VCPU_SREG_ES
);
3712 tss
->cs
= get_segment_selector(vcpu
, VCPU_SREG_CS
);
3713 tss
->ss
= get_segment_selector(vcpu
, VCPU_SREG_SS
);
3714 tss
->ds
= get_segment_selector(vcpu
, VCPU_SREG_DS
);
3715 tss
->fs
= get_segment_selector(vcpu
, VCPU_SREG_FS
);
3716 tss
->gs
= get_segment_selector(vcpu
, VCPU_SREG_GS
);
3717 tss
->ldt_selector
= get_segment_selector(vcpu
, VCPU_SREG_LDTR
);
3720 static int load_state_from_tss32(struct kvm_vcpu
*vcpu
,
3721 struct tss_segment_32
*tss
)
3723 kvm_set_cr3(vcpu
, tss
->cr3
);
3725 kvm_rip_write(vcpu
, tss
->eip
);
3726 kvm_x86_ops
->set_rflags(vcpu
, tss
->eflags
| 2);
3728 kvm_register_write(vcpu
, VCPU_REGS_RAX
, tss
->eax
);
3729 kvm_register_write(vcpu
, VCPU_REGS_RCX
, tss
->ecx
);
3730 kvm_register_write(vcpu
, VCPU_REGS_RDX
, tss
->edx
);
3731 kvm_register_write(vcpu
, VCPU_REGS_RBX
, tss
->ebx
);
3732 kvm_register_write(vcpu
, VCPU_REGS_RSP
, tss
->esp
);
3733 kvm_register_write(vcpu
, VCPU_REGS_RBP
, tss
->ebp
);
3734 kvm_register_write(vcpu
, VCPU_REGS_RSI
, tss
->esi
);
3735 kvm_register_write(vcpu
, VCPU_REGS_RDI
, tss
->edi
);
3737 if (kvm_load_segment_descriptor(vcpu
, tss
->ldt_selector
, 0, VCPU_SREG_LDTR
))
3740 if (kvm_load_segment_descriptor(vcpu
, tss
->es
, 1, VCPU_SREG_ES
))
3743 if (kvm_load_segment_descriptor(vcpu
, tss
->cs
, 9, VCPU_SREG_CS
))
3746 if (kvm_load_segment_descriptor(vcpu
, tss
->ss
, 1, VCPU_SREG_SS
))
3749 if (kvm_load_segment_descriptor(vcpu
, tss
->ds
, 1, VCPU_SREG_DS
))
3752 if (kvm_load_segment_descriptor(vcpu
, tss
->fs
, 1, VCPU_SREG_FS
))
3755 if (kvm_load_segment_descriptor(vcpu
, tss
->gs
, 1, VCPU_SREG_GS
))
3760 static void save_state_to_tss16(struct kvm_vcpu
*vcpu
,
3761 struct tss_segment_16
*tss
)
3763 tss
->ip
= kvm_rip_read(vcpu
);
3764 tss
->flag
= kvm_x86_ops
->get_rflags(vcpu
);
3765 tss
->ax
= kvm_register_read(vcpu
, VCPU_REGS_RAX
);
3766 tss
->cx
= kvm_register_read(vcpu
, VCPU_REGS_RCX
);
3767 tss
->dx
= kvm_register_read(vcpu
, VCPU_REGS_RDX
);
3768 tss
->bx
= kvm_register_read(vcpu
, VCPU_REGS_RBX
);
3769 tss
->sp
= kvm_register_read(vcpu
, VCPU_REGS_RSP
);
3770 tss
->bp
= kvm_register_read(vcpu
, VCPU_REGS_RBP
);
3771 tss
->si
= kvm_register_read(vcpu
, VCPU_REGS_RSI
);
3772 tss
->di
= kvm_register_read(vcpu
, VCPU_REGS_RDI
);
3774 tss
->es
= get_segment_selector(vcpu
, VCPU_SREG_ES
);
3775 tss
->cs
= get_segment_selector(vcpu
, VCPU_SREG_CS
);
3776 tss
->ss
= get_segment_selector(vcpu
, VCPU_SREG_SS
);
3777 tss
->ds
= get_segment_selector(vcpu
, VCPU_SREG_DS
);
3778 tss
->ldt
= get_segment_selector(vcpu
, VCPU_SREG_LDTR
);
3779 tss
->prev_task_link
= get_segment_selector(vcpu
, VCPU_SREG_TR
);
3782 static int load_state_from_tss16(struct kvm_vcpu
*vcpu
,
3783 struct tss_segment_16
*tss
)
3785 kvm_rip_write(vcpu
, tss
->ip
);
3786 kvm_x86_ops
->set_rflags(vcpu
, tss
->flag
| 2);
3787 kvm_register_write(vcpu
, VCPU_REGS_RAX
, tss
->ax
);
3788 kvm_register_write(vcpu
, VCPU_REGS_RCX
, tss
->cx
);
3789 kvm_register_write(vcpu
, VCPU_REGS_RDX
, tss
->dx
);
3790 kvm_register_write(vcpu
, VCPU_REGS_RBX
, tss
->bx
);
3791 kvm_register_write(vcpu
, VCPU_REGS_RSP
, tss
->sp
);
3792 kvm_register_write(vcpu
, VCPU_REGS_RBP
, tss
->bp
);
3793 kvm_register_write(vcpu
, VCPU_REGS_RSI
, tss
->si
);
3794 kvm_register_write(vcpu
, VCPU_REGS_RDI
, tss
->di
);
3796 if (kvm_load_segment_descriptor(vcpu
, tss
->ldt
, 0, VCPU_SREG_LDTR
))
3799 if (kvm_load_segment_descriptor(vcpu
, tss
->es
, 1, VCPU_SREG_ES
))
3802 if (kvm_load_segment_descriptor(vcpu
, tss
->cs
, 9, VCPU_SREG_CS
))
3805 if (kvm_load_segment_descriptor(vcpu
, tss
->ss
, 1, VCPU_SREG_SS
))
3808 if (kvm_load_segment_descriptor(vcpu
, tss
->ds
, 1, VCPU_SREG_DS
))
3813 static int kvm_task_switch_16(struct kvm_vcpu
*vcpu
, u16 tss_selector
,
3814 u16 old_tss_sel
, u32 old_tss_base
,
3815 struct desc_struct
*nseg_desc
)
3817 struct tss_segment_16 tss_segment_16
;
3820 if (kvm_read_guest(vcpu
->kvm
, old_tss_base
, &tss_segment_16
,
3821 sizeof tss_segment_16
))
3824 save_state_to_tss16(vcpu
, &tss_segment_16
);
3826 if (kvm_write_guest(vcpu
->kvm
, old_tss_base
, &tss_segment_16
,
3827 sizeof tss_segment_16
))
3830 if (kvm_read_guest(vcpu
->kvm
, get_tss_base_addr(vcpu
, nseg_desc
),
3831 &tss_segment_16
, sizeof tss_segment_16
))
3834 if (old_tss_sel
!= 0xffff) {
3835 tss_segment_16
.prev_task_link
= old_tss_sel
;
3837 if (kvm_write_guest(vcpu
->kvm
,
3838 get_tss_base_addr(vcpu
, nseg_desc
),
3839 &tss_segment_16
.prev_task_link
,
3840 sizeof tss_segment_16
.prev_task_link
))
3844 if (load_state_from_tss16(vcpu
, &tss_segment_16
))
3852 static int kvm_task_switch_32(struct kvm_vcpu
*vcpu
, u16 tss_selector
,
3853 u16 old_tss_sel
, u32 old_tss_base
,
3854 struct desc_struct
*nseg_desc
)
3856 struct tss_segment_32 tss_segment_32
;
3859 if (kvm_read_guest(vcpu
->kvm
, old_tss_base
, &tss_segment_32
,
3860 sizeof tss_segment_32
))
3863 save_state_to_tss32(vcpu
, &tss_segment_32
);
3865 if (kvm_write_guest(vcpu
->kvm
, old_tss_base
, &tss_segment_32
,
3866 sizeof tss_segment_32
))
3869 if (kvm_read_guest(vcpu
->kvm
, get_tss_base_addr(vcpu
, nseg_desc
),
3870 &tss_segment_32
, sizeof tss_segment_32
))
3873 if (old_tss_sel
!= 0xffff) {
3874 tss_segment_32
.prev_task_link
= old_tss_sel
;
3876 if (kvm_write_guest(vcpu
->kvm
,
3877 get_tss_base_addr(vcpu
, nseg_desc
),
3878 &tss_segment_32
.prev_task_link
,
3879 sizeof tss_segment_32
.prev_task_link
))
3883 if (load_state_from_tss32(vcpu
, &tss_segment_32
))
3891 int kvm_task_switch(struct kvm_vcpu
*vcpu
, u16 tss_selector
, int reason
)
3893 struct kvm_segment tr_seg
;
3894 struct desc_struct cseg_desc
;
3895 struct desc_struct nseg_desc
;
3897 u32 old_tss_base
= get_segment_base(vcpu
, VCPU_SREG_TR
);
3898 u16 old_tss_sel
= get_segment_selector(vcpu
, VCPU_SREG_TR
);
3900 old_tss_base
= vcpu
->arch
.mmu
.gva_to_gpa(vcpu
, old_tss_base
);
3902 /* FIXME: Handle errors. Failure to read either TSS or their
3903 * descriptors should generate a pagefault.
3905 if (load_guest_segment_descriptor(vcpu
, tss_selector
, &nseg_desc
))
3908 if (load_guest_segment_descriptor(vcpu
, old_tss_sel
, &cseg_desc
))
3911 if (reason
!= TASK_SWITCH_IRET
) {
3914 cpl
= kvm_x86_ops
->get_cpl(vcpu
);
3915 if ((tss_selector
& 3) > nseg_desc
.dpl
|| cpl
> nseg_desc
.dpl
) {
3916 kvm_queue_exception_e(vcpu
, GP_VECTOR
, 0);
3921 if (!nseg_desc
.p
|| (nseg_desc
.limit0
| nseg_desc
.limit
<< 16) < 0x67) {
3922 kvm_queue_exception_e(vcpu
, TS_VECTOR
, tss_selector
& 0xfffc);
3926 if (reason
== TASK_SWITCH_IRET
|| reason
== TASK_SWITCH_JMP
) {
3927 cseg_desc
.type
&= ~(1 << 1); //clear the B flag
3928 save_guest_segment_descriptor(vcpu
, old_tss_sel
, &cseg_desc
);
3931 if (reason
== TASK_SWITCH_IRET
) {
3932 u32 eflags
= kvm_x86_ops
->get_rflags(vcpu
);
3933 kvm_x86_ops
->set_rflags(vcpu
, eflags
& ~X86_EFLAGS_NT
);
3936 /* set back link to prev task only if NT bit is set in eflags
3937 note that old_tss_sel is not used afetr this point */
3938 if (reason
!= TASK_SWITCH_CALL
&& reason
!= TASK_SWITCH_GATE
)
3939 old_tss_sel
= 0xffff;
3941 /* set back link to prev task only if NT bit is set in eflags
3942 note that old_tss_sel is not used afetr this point */
3943 if (reason
!= TASK_SWITCH_CALL
&& reason
!= TASK_SWITCH_GATE
)
3944 old_tss_sel
= 0xffff;
3946 if (nseg_desc
.type
& 8)
3947 ret
= kvm_task_switch_32(vcpu
, tss_selector
, old_tss_sel
,
3948 old_tss_base
, &nseg_desc
);
3950 ret
= kvm_task_switch_16(vcpu
, tss_selector
, old_tss_sel
,
3951 old_tss_base
, &nseg_desc
);
3953 if (reason
== TASK_SWITCH_CALL
|| reason
== TASK_SWITCH_GATE
) {
3954 u32 eflags
= kvm_x86_ops
->get_rflags(vcpu
);
3955 kvm_x86_ops
->set_rflags(vcpu
, eflags
| X86_EFLAGS_NT
);
3958 if (reason
!= TASK_SWITCH_IRET
) {
3959 nseg_desc
.type
|= (1 << 1);
3960 save_guest_segment_descriptor(vcpu
, tss_selector
,
3964 kvm_x86_ops
->set_cr0(vcpu
, vcpu
->arch
.cr0
| X86_CR0_TS
);
3965 seg_desct_to_kvm_desct(&nseg_desc
, tss_selector
, &tr_seg
);
3967 kvm_set_segment(vcpu
, &tr_seg
, VCPU_SREG_TR
);
3971 EXPORT_SYMBOL_GPL(kvm_task_switch
);
3973 int kvm_arch_vcpu_ioctl_set_sregs(struct kvm_vcpu
*vcpu
,
3974 struct kvm_sregs
*sregs
)
3976 int mmu_reset_needed
= 0;
3977 int i
, pending_vec
, max_bits
;
3978 struct descriptor_table dt
;
3982 dt
.limit
= sregs
->idt
.limit
;
3983 dt
.base
= sregs
->idt
.base
;
3984 kvm_x86_ops
->set_idt(vcpu
, &dt
);
3985 dt
.limit
= sregs
->gdt
.limit
;
3986 dt
.base
= sregs
->gdt
.base
;
3987 kvm_x86_ops
->set_gdt(vcpu
, &dt
);
3989 vcpu
->arch
.cr2
= sregs
->cr2
;
3990 mmu_reset_needed
|= vcpu
->arch
.cr3
!= sregs
->cr3
;
3992 down_read(&vcpu
->kvm
->slots_lock
);
3993 if (gfn_to_memslot(vcpu
->kvm
, sregs
->cr3
>> PAGE_SHIFT
))
3994 vcpu
->arch
.cr3
= sregs
->cr3
;
3996 set_bit(KVM_REQ_TRIPLE_FAULT
, &vcpu
->requests
);
3997 up_read(&vcpu
->kvm
->slots_lock
);
3999 kvm_set_cr8(vcpu
, sregs
->cr8
);
4001 mmu_reset_needed
|= vcpu
->arch
.shadow_efer
!= sregs
->efer
;
4002 kvm_x86_ops
->set_efer(vcpu
, sregs
->efer
);
4003 kvm_set_apic_base(vcpu
, sregs
->apic_base
);
4005 kvm_x86_ops
->decache_cr4_guest_bits(vcpu
);
4007 mmu_reset_needed
|= vcpu
->arch
.cr0
!= sregs
->cr0
;
4008 kvm_x86_ops
->set_cr0(vcpu
, sregs
->cr0
);
4009 vcpu
->arch
.cr0
= sregs
->cr0
;
4011 mmu_reset_needed
|= vcpu
->arch
.cr4
!= sregs
->cr4
;
4012 kvm_x86_ops
->set_cr4(vcpu
, sregs
->cr4
);
4013 if (!is_long_mode(vcpu
) && is_pae(vcpu
))
4014 load_pdptrs(vcpu
, vcpu
->arch
.cr3
);
4016 if (mmu_reset_needed
)
4017 kvm_mmu_reset_context(vcpu
);
4019 if (!irqchip_in_kernel(vcpu
->kvm
)) {
4020 memcpy(vcpu
->arch
.irq_pending
, sregs
->interrupt_bitmap
,
4021 sizeof vcpu
->arch
.irq_pending
);
4022 vcpu
->arch
.irq_summary
= 0;
4023 for (i
= 0; i
< ARRAY_SIZE(vcpu
->arch
.irq_pending
); ++i
)
4024 if (vcpu
->arch
.irq_pending
[i
])
4025 __set_bit(i
, &vcpu
->arch
.irq_summary
);
4027 max_bits
= (sizeof sregs
->interrupt_bitmap
) << 3;
4028 pending_vec
= find_first_bit(
4029 (const unsigned long *)sregs
->interrupt_bitmap
,
4031 /* Only pending external irq is handled here */
4032 if (pending_vec
< max_bits
) {
4033 kvm_x86_ops
->set_irq(vcpu
, pending_vec
);
4034 pr_debug("Set back pending irq %d\n",
4037 kvm_pic_clear_isr_ack(vcpu
->kvm
);
4040 kvm_set_segment(vcpu
, &sregs
->cs
, VCPU_SREG_CS
);
4041 kvm_set_segment(vcpu
, &sregs
->ds
, VCPU_SREG_DS
);
4042 kvm_set_segment(vcpu
, &sregs
->es
, VCPU_SREG_ES
);
4043 kvm_set_segment(vcpu
, &sregs
->fs
, VCPU_SREG_FS
);
4044 kvm_set_segment(vcpu
, &sregs
->gs
, VCPU_SREG_GS
);
4045 kvm_set_segment(vcpu
, &sregs
->ss
, VCPU_SREG_SS
);
4047 kvm_set_segment(vcpu
, &sregs
->tr
, VCPU_SREG_TR
);
4048 kvm_set_segment(vcpu
, &sregs
->ldt
, VCPU_SREG_LDTR
);
4050 /* Older userspace won't unhalt the vcpu on reset. */
4051 if (vcpu
->vcpu_id
== 0 && kvm_rip_read(vcpu
) == 0xfff0 &&
4052 sregs
->cs
.selector
== 0xf000 && sregs
->cs
.base
== 0xffff0000 &&
4053 !(vcpu
->arch
.cr0
& X86_CR0_PE
))
4054 vcpu
->arch
.mp_state
= KVM_MP_STATE_RUNNABLE
;
4061 int kvm_arch_vcpu_ioctl_set_guest_debug(struct kvm_vcpu
*vcpu
,
4062 struct kvm_guest_debug
*dbg
)
4068 if ((dbg
->control
& (KVM_GUESTDBG_ENABLE
| KVM_GUESTDBG_USE_HW_BP
)) ==
4069 (KVM_GUESTDBG_ENABLE
| KVM_GUESTDBG_USE_HW_BP
)) {
4070 for (i
= 0; i
< KVM_NR_DB_REGS
; ++i
)
4071 vcpu
->arch
.eff_db
[i
] = dbg
->arch
.debugreg
[i
];
4072 vcpu
->arch
.switch_db_regs
=
4073 (dbg
->arch
.debugreg
[7] & DR7_BP_EN_MASK
);
4075 for (i
= 0; i
< KVM_NR_DB_REGS
; i
++)
4076 vcpu
->arch
.eff_db
[i
] = vcpu
->arch
.db
[i
];
4077 vcpu
->arch
.switch_db_regs
= (vcpu
->arch
.dr7
& DR7_BP_EN_MASK
);
4080 r
= kvm_x86_ops
->set_guest_debug(vcpu
, dbg
);
4082 if (dbg
->control
& KVM_GUESTDBG_INJECT_DB
)
4083 kvm_queue_exception(vcpu
, DB_VECTOR
);
4084 else if (dbg
->control
& KVM_GUESTDBG_INJECT_BP
)
4085 kvm_queue_exception(vcpu
, BP_VECTOR
);
4093 * fxsave fpu state. Taken from x86_64/processor.h. To be killed when
4094 * we have asm/x86/processor.h
4105 u32 st_space
[32]; /* 8*16 bytes for each FP-reg = 128 bytes */
4106 #ifdef CONFIG_X86_64
4107 u32 xmm_space
[64]; /* 16*16 bytes for each XMM-reg = 256 bytes */
4109 u32 xmm_space
[32]; /* 8*16 bytes for each XMM-reg = 128 bytes */
4114 * Translate a guest virtual address to a guest physical address.
4116 int kvm_arch_vcpu_ioctl_translate(struct kvm_vcpu
*vcpu
,
4117 struct kvm_translation
*tr
)
4119 unsigned long vaddr
= tr
->linear_address
;
4123 down_read(&vcpu
->kvm
->slots_lock
);
4124 gpa
= vcpu
->arch
.mmu
.gva_to_gpa(vcpu
, vaddr
);
4125 up_read(&vcpu
->kvm
->slots_lock
);
4126 tr
->physical_address
= gpa
;
4127 tr
->valid
= gpa
!= UNMAPPED_GVA
;
4135 int kvm_arch_vcpu_ioctl_get_fpu(struct kvm_vcpu
*vcpu
, struct kvm_fpu
*fpu
)
4137 struct fxsave
*fxsave
= (struct fxsave
*)&vcpu
->arch
.guest_fx_image
;
4141 memcpy(fpu
->fpr
, fxsave
->st_space
, 128);
4142 fpu
->fcw
= fxsave
->cwd
;
4143 fpu
->fsw
= fxsave
->swd
;
4144 fpu
->ftwx
= fxsave
->twd
;
4145 fpu
->last_opcode
= fxsave
->fop
;
4146 fpu
->last_ip
= fxsave
->rip
;
4147 fpu
->last_dp
= fxsave
->rdp
;
4148 memcpy(fpu
->xmm
, fxsave
->xmm_space
, sizeof fxsave
->xmm_space
);
4155 int kvm_arch_vcpu_ioctl_set_fpu(struct kvm_vcpu
*vcpu
, struct kvm_fpu
*fpu
)
4157 struct fxsave
*fxsave
= (struct fxsave
*)&vcpu
->arch
.guest_fx_image
;
4161 memcpy(fxsave
->st_space
, fpu
->fpr
, 128);
4162 fxsave
->cwd
= fpu
->fcw
;
4163 fxsave
->swd
= fpu
->fsw
;
4164 fxsave
->twd
= fpu
->ftwx
;
4165 fxsave
->fop
= fpu
->last_opcode
;
4166 fxsave
->rip
= fpu
->last_ip
;
4167 fxsave
->rdp
= fpu
->last_dp
;
4168 memcpy(fxsave
->xmm_space
, fpu
->xmm
, sizeof fxsave
->xmm_space
);
4175 void fx_init(struct kvm_vcpu
*vcpu
)
4177 unsigned after_mxcsr_mask
;
4180 * Touch the fpu the first time in non atomic context as if
4181 * this is the first fpu instruction the exception handler
4182 * will fire before the instruction returns and it'll have to
4183 * allocate ram with GFP_KERNEL.
4186 kvm_fx_save(&vcpu
->arch
.host_fx_image
);
4188 /* Initialize guest FPU by resetting ours and saving into guest's */
4190 kvm_fx_save(&vcpu
->arch
.host_fx_image
);
4192 kvm_fx_save(&vcpu
->arch
.guest_fx_image
);
4193 kvm_fx_restore(&vcpu
->arch
.host_fx_image
);
4196 vcpu
->arch
.cr0
|= X86_CR0_ET
;
4197 after_mxcsr_mask
= offsetof(struct i387_fxsave_struct
, st_space
);
4198 vcpu
->arch
.guest_fx_image
.mxcsr
= 0x1f80;
4199 memset((void *)&vcpu
->arch
.guest_fx_image
+ after_mxcsr_mask
,
4200 0, sizeof(struct i387_fxsave_struct
) - after_mxcsr_mask
);
4202 EXPORT_SYMBOL_GPL(fx_init
);
4204 void kvm_load_guest_fpu(struct kvm_vcpu
*vcpu
)
4206 if (!vcpu
->fpu_active
|| vcpu
->guest_fpu_loaded
)
4209 vcpu
->guest_fpu_loaded
= 1;
4210 kvm_fx_save(&vcpu
->arch
.host_fx_image
);
4211 kvm_fx_restore(&vcpu
->arch
.guest_fx_image
);
4213 EXPORT_SYMBOL_GPL(kvm_load_guest_fpu
);
4215 void kvm_put_guest_fpu(struct kvm_vcpu
*vcpu
)
4217 if (!vcpu
->guest_fpu_loaded
)
4220 vcpu
->guest_fpu_loaded
= 0;
4221 kvm_fx_save(&vcpu
->arch
.guest_fx_image
);
4222 kvm_fx_restore(&vcpu
->arch
.host_fx_image
);
4223 ++vcpu
->stat
.fpu_reload
;
4225 EXPORT_SYMBOL_GPL(kvm_put_guest_fpu
);
4227 void kvm_arch_vcpu_free(struct kvm_vcpu
*vcpu
)
4229 if (vcpu
->arch
.time_page
) {
4230 kvm_release_page_dirty(vcpu
->arch
.time_page
);
4231 vcpu
->arch
.time_page
= NULL
;
4234 kvm_x86_ops
->vcpu_free(vcpu
);
4237 struct kvm_vcpu
*kvm_arch_vcpu_create(struct kvm
*kvm
,
4240 return kvm_x86_ops
->vcpu_create(kvm
, id
);
4243 int kvm_arch_vcpu_setup(struct kvm_vcpu
*vcpu
)
4247 /* We do fxsave: this must be aligned. */
4248 BUG_ON((unsigned long)&vcpu
->arch
.host_fx_image
& 0xF);
4250 vcpu
->arch
.mtrr_state
.have_fixed
= 1;
4252 r
= kvm_arch_vcpu_reset(vcpu
);
4254 r
= kvm_mmu_setup(vcpu
);
4261 kvm_x86_ops
->vcpu_free(vcpu
);
4265 void kvm_arch_vcpu_destroy(struct kvm_vcpu
*vcpu
)
4268 kvm_mmu_unload(vcpu
);
4271 kvm_x86_ops
->vcpu_free(vcpu
);
4274 int kvm_arch_vcpu_reset(struct kvm_vcpu
*vcpu
)
4276 vcpu
->arch
.nmi_pending
= false;
4277 vcpu
->arch
.nmi_injected
= false;
4279 vcpu
->arch
.switch_db_regs
= 0;
4280 memset(vcpu
->arch
.db
, 0, sizeof(vcpu
->arch
.db
));
4281 vcpu
->arch
.dr6
= DR6_FIXED_1
;
4282 vcpu
->arch
.dr7
= DR7_FIXED_1
;
4284 return kvm_x86_ops
->vcpu_reset(vcpu
);
4287 void kvm_arch_hardware_enable(void *garbage
)
4289 kvm_x86_ops
->hardware_enable(garbage
);
4292 void kvm_arch_hardware_disable(void *garbage
)
4294 kvm_x86_ops
->hardware_disable(garbage
);
4297 int kvm_arch_hardware_setup(void)
4299 return kvm_x86_ops
->hardware_setup();
4302 void kvm_arch_hardware_unsetup(void)
4304 kvm_x86_ops
->hardware_unsetup();
4307 void kvm_arch_check_processor_compat(void *rtn
)
4309 kvm_x86_ops
->check_processor_compatibility(rtn
);
4312 int kvm_arch_vcpu_init(struct kvm_vcpu
*vcpu
)
4318 BUG_ON(vcpu
->kvm
== NULL
);
4321 vcpu
->arch
.mmu
.root_hpa
= INVALID_PAGE
;
4322 if (!irqchip_in_kernel(kvm
) || vcpu
->vcpu_id
== 0)
4323 vcpu
->arch
.mp_state
= KVM_MP_STATE_RUNNABLE
;
4325 vcpu
->arch
.mp_state
= KVM_MP_STATE_UNINITIALIZED
;
4327 page
= alloc_page(GFP_KERNEL
| __GFP_ZERO
);
4332 vcpu
->arch
.pio_data
= page_address(page
);
4334 r
= kvm_mmu_create(vcpu
);
4336 goto fail_free_pio_data
;
4338 if (irqchip_in_kernel(kvm
)) {
4339 r
= kvm_create_lapic(vcpu
);
4341 goto fail_mmu_destroy
;
4347 kvm_mmu_destroy(vcpu
);
4349 free_page((unsigned long)vcpu
->arch
.pio_data
);
4354 void kvm_arch_vcpu_uninit(struct kvm_vcpu
*vcpu
)
4356 kvm_free_lapic(vcpu
);
4357 down_read(&vcpu
->kvm
->slots_lock
);
4358 kvm_mmu_destroy(vcpu
);
4359 up_read(&vcpu
->kvm
->slots_lock
);
4360 free_page((unsigned long)vcpu
->arch
.pio_data
);
4363 struct kvm
*kvm_arch_create_vm(void)
4365 struct kvm
*kvm
= kzalloc(sizeof(struct kvm
), GFP_KERNEL
);
4368 return ERR_PTR(-ENOMEM
);
4370 INIT_LIST_HEAD(&kvm
->arch
.active_mmu_pages
);
4371 INIT_LIST_HEAD(&kvm
->arch
.assigned_dev_head
);
4373 /* Reserve bit 0 of irq_sources_bitmap for userspace irq source */
4374 set_bit(KVM_USERSPACE_IRQ_SOURCE_ID
, &kvm
->arch
.irq_sources_bitmap
);
4376 rdtscll(kvm
->arch
.vm_init_tsc
);
4381 static void kvm_unload_vcpu_mmu(struct kvm_vcpu
*vcpu
)
4384 kvm_mmu_unload(vcpu
);
4388 static void kvm_free_vcpus(struct kvm
*kvm
)
4393 * Unpin any mmu pages first.
4395 for (i
= 0; i
< KVM_MAX_VCPUS
; ++i
)
4397 kvm_unload_vcpu_mmu(kvm
->vcpus
[i
]);
4398 for (i
= 0; i
< KVM_MAX_VCPUS
; ++i
) {
4399 if (kvm
->vcpus
[i
]) {
4400 kvm_arch_vcpu_free(kvm
->vcpus
[i
]);
4401 kvm
->vcpus
[i
] = NULL
;
4407 void kvm_arch_sync_events(struct kvm
*kvm
)
4409 kvm_free_all_assigned_devices(kvm
);
4412 void kvm_arch_destroy_vm(struct kvm
*kvm
)
4414 kvm_iommu_unmap_guest(kvm
);
4416 kfree(kvm
->arch
.vpic
);
4417 kfree(kvm
->arch
.vioapic
);
4418 kvm_free_vcpus(kvm
);
4419 kvm_free_physmem(kvm
);
4420 if (kvm
->arch
.apic_access_page
)
4421 put_page(kvm
->arch
.apic_access_page
);
4422 if (kvm
->arch
.ept_identity_pagetable
)
4423 put_page(kvm
->arch
.ept_identity_pagetable
);
4427 int kvm_arch_set_memory_region(struct kvm
*kvm
,
4428 struct kvm_userspace_memory_region
*mem
,
4429 struct kvm_memory_slot old
,
4432 int npages
= mem
->memory_size
>> PAGE_SHIFT
;
4433 struct kvm_memory_slot
*memslot
= &kvm
->memslots
[mem
->slot
];
4435 /*To keep backward compatibility with older userspace,
4436 *x86 needs to hanlde !user_alloc case.
4439 if (npages
&& !old
.rmap
) {
4440 unsigned long userspace_addr
;
4442 down_write(¤t
->mm
->mmap_sem
);
4443 userspace_addr
= do_mmap(NULL
, 0,
4445 PROT_READ
| PROT_WRITE
,
4446 MAP_PRIVATE
| MAP_ANONYMOUS
,
4448 up_write(¤t
->mm
->mmap_sem
);
4450 if (IS_ERR((void *)userspace_addr
))
4451 return PTR_ERR((void *)userspace_addr
);
4453 /* set userspace_addr atomically for kvm_hva_to_rmapp */
4454 spin_lock(&kvm
->mmu_lock
);
4455 memslot
->userspace_addr
= userspace_addr
;
4456 spin_unlock(&kvm
->mmu_lock
);
4458 if (!old
.user_alloc
&& old
.rmap
) {
4461 down_write(¤t
->mm
->mmap_sem
);
4462 ret
= do_munmap(current
->mm
, old
.userspace_addr
,
4463 old
.npages
* PAGE_SIZE
);
4464 up_write(¤t
->mm
->mmap_sem
);
4467 "kvm_vm_ioctl_set_memory_region: "
4468 "failed to munmap memory\n");
4473 if (!kvm
->arch
.n_requested_mmu_pages
) {
4474 unsigned int nr_mmu_pages
= kvm_mmu_calculate_mmu_pages(kvm
);
4475 kvm_mmu_change_mmu_pages(kvm
, nr_mmu_pages
);
4478 kvm_mmu_slot_remove_write_access(kvm
, mem
->slot
);
4479 kvm_flush_remote_tlbs(kvm
);
4484 void kvm_arch_flush_shadow(struct kvm
*kvm
)
4486 kvm_mmu_zap_all(kvm
);
4489 int kvm_arch_vcpu_runnable(struct kvm_vcpu
*vcpu
)
4491 return vcpu
->arch
.mp_state
== KVM_MP_STATE_RUNNABLE
4492 || vcpu
->arch
.mp_state
== KVM_MP_STATE_SIPI_RECEIVED
4493 || vcpu
->arch
.nmi_pending
;
4496 static void vcpu_kick_intr(void *info
)
4499 struct kvm_vcpu
*vcpu
= (struct kvm_vcpu
*)info
;
4500 printk(KERN_DEBUG
"vcpu_kick_intr %p \n", vcpu
);
4504 void kvm_vcpu_kick(struct kvm_vcpu
*vcpu
)
4506 int ipi_pcpu
= vcpu
->cpu
;
4507 int cpu
= get_cpu();
4509 if (waitqueue_active(&vcpu
->wq
)) {
4510 wake_up_interruptible(&vcpu
->wq
);
4511 ++vcpu
->stat
.halt_wakeup
;
4514 * We may be called synchronously with irqs disabled in guest mode,
4515 * So need not to call smp_call_function_single() in that case.
4517 if (vcpu
->guest_mode
&& vcpu
->cpu
!= cpu
)
4518 smp_call_function_single(ipi_pcpu
, vcpu_kick_intr
, vcpu
, 0);
4522 int kvm_arch_interrupt_allowed(struct kvm_vcpu
*vcpu
)
4524 return kvm_x86_ops
->interrupt_allowed(vcpu
);