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>
40 #define CREATE_TRACE_POINTS
43 #include <asm/uaccess.h>
49 #define MAX_IO_MSRS 256
50 #define CR0_RESERVED_BITS \
51 (~(unsigned long)(X86_CR0_PE | X86_CR0_MP | X86_CR0_EM | X86_CR0_TS \
52 | X86_CR0_ET | X86_CR0_NE | X86_CR0_WP | X86_CR0_AM \
53 | X86_CR0_NW | X86_CR0_CD | X86_CR0_PG))
54 #define CR4_RESERVED_BITS \
55 (~(unsigned long)(X86_CR4_VME | X86_CR4_PVI | X86_CR4_TSD | X86_CR4_DE\
56 | X86_CR4_PSE | X86_CR4_PAE | X86_CR4_MCE \
57 | X86_CR4_PGE | X86_CR4_PCE | X86_CR4_OSFXSR \
58 | X86_CR4_OSXMMEXCPT | X86_CR4_VMXE))
60 #define CR8_RESERVED_BITS (~(unsigned long)X86_CR8_TPR)
62 #define KVM_MAX_MCE_BANKS 32
63 #define KVM_MCE_CAP_SUPPORTED MCG_CTL_P
66 * - enable syscall per default because its emulated by KVM
67 * - enable LME and LMA per default on 64 bit KVM
70 static u64 __read_mostly efer_reserved_bits
= 0xfffffffffffffafeULL
;
72 static u64 __read_mostly efer_reserved_bits
= 0xfffffffffffffffeULL
;
75 #define VM_STAT(x) offsetof(struct kvm, stat.x), KVM_STAT_VM
76 #define VCPU_STAT(x) offsetof(struct kvm_vcpu, stat.x), KVM_STAT_VCPU
78 static int kvm_dev_ioctl_get_supported_cpuid(struct kvm_cpuid2
*cpuid
,
79 struct kvm_cpuid_entry2 __user
*entries
);
80 struct kvm_cpuid_entry2
*kvm_find_cpuid_entry(struct kvm_vcpu
*vcpu
,
81 u32 function
, u32 index
);
83 struct kvm_x86_ops
*kvm_x86_ops
;
84 EXPORT_SYMBOL_GPL(kvm_x86_ops
);
86 struct kvm_stats_debugfs_item debugfs_entries
[] = {
87 { "pf_fixed", VCPU_STAT(pf_fixed
) },
88 { "pf_guest", VCPU_STAT(pf_guest
) },
89 { "tlb_flush", VCPU_STAT(tlb_flush
) },
90 { "invlpg", VCPU_STAT(invlpg
) },
91 { "exits", VCPU_STAT(exits
) },
92 { "io_exits", VCPU_STAT(io_exits
) },
93 { "mmio_exits", VCPU_STAT(mmio_exits
) },
94 { "signal_exits", VCPU_STAT(signal_exits
) },
95 { "irq_window", VCPU_STAT(irq_window_exits
) },
96 { "nmi_window", VCPU_STAT(nmi_window_exits
) },
97 { "halt_exits", VCPU_STAT(halt_exits
) },
98 { "halt_wakeup", VCPU_STAT(halt_wakeup
) },
99 { "hypercalls", VCPU_STAT(hypercalls
) },
100 { "request_irq", VCPU_STAT(request_irq_exits
) },
101 { "irq_exits", VCPU_STAT(irq_exits
) },
102 { "host_state_reload", VCPU_STAT(host_state_reload
) },
103 { "efer_reload", VCPU_STAT(efer_reload
) },
104 { "fpu_reload", VCPU_STAT(fpu_reload
) },
105 { "insn_emulation", VCPU_STAT(insn_emulation
) },
106 { "insn_emulation_fail", VCPU_STAT(insn_emulation_fail
) },
107 { "irq_injections", VCPU_STAT(irq_injections
) },
108 { "nmi_injections", VCPU_STAT(nmi_injections
) },
109 { "mmu_shadow_zapped", VM_STAT(mmu_shadow_zapped
) },
110 { "mmu_pte_write", VM_STAT(mmu_pte_write
) },
111 { "mmu_pte_updated", VM_STAT(mmu_pte_updated
) },
112 { "mmu_pde_zapped", VM_STAT(mmu_pde_zapped
) },
113 { "mmu_flooded", VM_STAT(mmu_flooded
) },
114 { "mmu_recycled", VM_STAT(mmu_recycled
) },
115 { "mmu_cache_miss", VM_STAT(mmu_cache_miss
) },
116 { "mmu_unsync", VM_STAT(mmu_unsync
) },
117 { "remote_tlb_flush", VM_STAT(remote_tlb_flush
) },
118 { "largepages", VM_STAT(lpages
) },
122 unsigned long segment_base(u16 selector
)
124 struct descriptor_table gdt
;
125 struct desc_struct
*d
;
126 unsigned long table_base
;
132 asm("sgdt %0" : "=m"(gdt
));
133 table_base
= gdt
.base
;
135 if (selector
& 4) { /* from ldt */
138 asm("sldt %0" : "=g"(ldt_selector
));
139 table_base
= segment_base(ldt_selector
);
141 d
= (struct desc_struct
*)(table_base
+ (selector
& ~7));
142 v
= d
->base0
| ((unsigned long)d
->base1
<< 16) |
143 ((unsigned long)d
->base2
<< 24);
145 if (d
->s
== 0 && (d
->type
== 2 || d
->type
== 9 || d
->type
== 11))
146 v
|= ((unsigned long)((struct ldttss_desc64
*)d
)->base3
) << 32;
150 EXPORT_SYMBOL_GPL(segment_base
);
152 u64
kvm_get_apic_base(struct kvm_vcpu
*vcpu
)
154 if (irqchip_in_kernel(vcpu
->kvm
))
155 return vcpu
->arch
.apic_base
;
157 return vcpu
->arch
.apic_base
;
159 EXPORT_SYMBOL_GPL(kvm_get_apic_base
);
161 void kvm_set_apic_base(struct kvm_vcpu
*vcpu
, u64 data
)
163 /* TODO: reserve bits check */
164 if (irqchip_in_kernel(vcpu
->kvm
))
165 kvm_lapic_set_base(vcpu
, data
);
167 vcpu
->arch
.apic_base
= data
;
169 EXPORT_SYMBOL_GPL(kvm_set_apic_base
);
171 void kvm_queue_exception(struct kvm_vcpu
*vcpu
, unsigned nr
)
173 WARN_ON(vcpu
->arch
.exception
.pending
);
174 vcpu
->arch
.exception
.pending
= true;
175 vcpu
->arch
.exception
.has_error_code
= false;
176 vcpu
->arch
.exception
.nr
= nr
;
178 EXPORT_SYMBOL_GPL(kvm_queue_exception
);
180 void kvm_inject_page_fault(struct kvm_vcpu
*vcpu
, unsigned long addr
,
183 ++vcpu
->stat
.pf_guest
;
185 if (vcpu
->arch
.exception
.pending
) {
186 switch(vcpu
->arch
.exception
.nr
) {
188 /* triple fault -> shutdown */
189 set_bit(KVM_REQ_TRIPLE_FAULT
, &vcpu
->requests
);
192 vcpu
->arch
.exception
.nr
= DF_VECTOR
;
193 vcpu
->arch
.exception
.error_code
= 0;
196 /* replace previous exception with a new one in a hope
197 that instruction re-execution will regenerate lost
199 vcpu
->arch
.exception
.pending
= false;
203 vcpu
->arch
.cr2
= addr
;
204 kvm_queue_exception_e(vcpu
, PF_VECTOR
, error_code
);
207 void kvm_inject_nmi(struct kvm_vcpu
*vcpu
)
209 vcpu
->arch
.nmi_pending
= 1;
211 EXPORT_SYMBOL_GPL(kvm_inject_nmi
);
213 void kvm_queue_exception_e(struct kvm_vcpu
*vcpu
, unsigned nr
, u32 error_code
)
215 WARN_ON(vcpu
->arch
.exception
.pending
);
216 vcpu
->arch
.exception
.pending
= true;
217 vcpu
->arch
.exception
.has_error_code
= true;
218 vcpu
->arch
.exception
.nr
= nr
;
219 vcpu
->arch
.exception
.error_code
= error_code
;
221 EXPORT_SYMBOL_GPL(kvm_queue_exception_e
);
223 static void __queue_exception(struct kvm_vcpu
*vcpu
)
225 kvm_x86_ops
->queue_exception(vcpu
, vcpu
->arch
.exception
.nr
,
226 vcpu
->arch
.exception
.has_error_code
,
227 vcpu
->arch
.exception
.error_code
);
231 * Load the pae pdptrs. Return true is they are all valid.
233 int load_pdptrs(struct kvm_vcpu
*vcpu
, unsigned long cr3
)
235 gfn_t pdpt_gfn
= cr3
>> PAGE_SHIFT
;
236 unsigned offset
= ((cr3
& (PAGE_SIZE
-1)) >> 5) << 2;
239 u64 pdpte
[ARRAY_SIZE(vcpu
->arch
.pdptrs
)];
241 ret
= kvm_read_guest_page(vcpu
->kvm
, pdpt_gfn
, pdpte
,
242 offset
* sizeof(u64
), sizeof(pdpte
));
247 for (i
= 0; i
< ARRAY_SIZE(pdpte
); ++i
) {
248 if (is_present_gpte(pdpte
[i
]) &&
249 (pdpte
[i
] & vcpu
->arch
.mmu
.rsvd_bits_mask
[0][2])) {
256 memcpy(vcpu
->arch
.pdptrs
, pdpte
, sizeof(vcpu
->arch
.pdptrs
));
257 __set_bit(VCPU_EXREG_PDPTR
,
258 (unsigned long *)&vcpu
->arch
.regs_avail
);
259 __set_bit(VCPU_EXREG_PDPTR
,
260 (unsigned long *)&vcpu
->arch
.regs_dirty
);
265 EXPORT_SYMBOL_GPL(load_pdptrs
);
267 static bool pdptrs_changed(struct kvm_vcpu
*vcpu
)
269 u64 pdpte
[ARRAY_SIZE(vcpu
->arch
.pdptrs
)];
273 if (is_long_mode(vcpu
) || !is_pae(vcpu
))
276 if (!test_bit(VCPU_EXREG_PDPTR
,
277 (unsigned long *)&vcpu
->arch
.regs_avail
))
280 r
= kvm_read_guest(vcpu
->kvm
, vcpu
->arch
.cr3
& ~31u, pdpte
, sizeof(pdpte
));
283 changed
= memcmp(pdpte
, vcpu
->arch
.pdptrs
, sizeof(pdpte
)) != 0;
289 void kvm_set_cr0(struct kvm_vcpu
*vcpu
, unsigned long cr0
)
291 if (cr0
& CR0_RESERVED_BITS
) {
292 printk(KERN_DEBUG
"set_cr0: 0x%lx #GP, reserved bits 0x%lx\n",
293 cr0
, vcpu
->arch
.cr0
);
294 kvm_inject_gp(vcpu
, 0);
298 if ((cr0
& X86_CR0_NW
) && !(cr0
& X86_CR0_CD
)) {
299 printk(KERN_DEBUG
"set_cr0: #GP, CD == 0 && NW == 1\n");
300 kvm_inject_gp(vcpu
, 0);
304 if ((cr0
& X86_CR0_PG
) && !(cr0
& X86_CR0_PE
)) {
305 printk(KERN_DEBUG
"set_cr0: #GP, set PG flag "
306 "and a clear PE flag\n");
307 kvm_inject_gp(vcpu
, 0);
311 if (!is_paging(vcpu
) && (cr0
& X86_CR0_PG
)) {
313 if ((vcpu
->arch
.shadow_efer
& EFER_LME
)) {
317 printk(KERN_DEBUG
"set_cr0: #GP, start paging "
318 "in long mode while PAE is disabled\n");
319 kvm_inject_gp(vcpu
, 0);
322 kvm_x86_ops
->get_cs_db_l_bits(vcpu
, &cs_db
, &cs_l
);
324 printk(KERN_DEBUG
"set_cr0: #GP, start paging "
325 "in long mode while CS.L == 1\n");
326 kvm_inject_gp(vcpu
, 0);
332 if (is_pae(vcpu
) && !load_pdptrs(vcpu
, vcpu
->arch
.cr3
)) {
333 printk(KERN_DEBUG
"set_cr0: #GP, pdptrs "
335 kvm_inject_gp(vcpu
, 0);
341 kvm_x86_ops
->set_cr0(vcpu
, cr0
);
342 vcpu
->arch
.cr0
= cr0
;
344 kvm_mmu_reset_context(vcpu
);
347 EXPORT_SYMBOL_GPL(kvm_set_cr0
);
349 void kvm_lmsw(struct kvm_vcpu
*vcpu
, unsigned long msw
)
351 kvm_set_cr0(vcpu
, (vcpu
->arch
.cr0
& ~0x0ful
) | (msw
& 0x0f));
353 EXPORT_SYMBOL_GPL(kvm_lmsw
);
355 void kvm_set_cr4(struct kvm_vcpu
*vcpu
, unsigned long cr4
)
357 unsigned long old_cr4
= vcpu
->arch
.cr4
;
358 unsigned long pdptr_bits
= X86_CR4_PGE
| X86_CR4_PSE
| X86_CR4_PAE
;
360 if (cr4
& CR4_RESERVED_BITS
) {
361 printk(KERN_DEBUG
"set_cr4: #GP, reserved bits\n");
362 kvm_inject_gp(vcpu
, 0);
366 if (is_long_mode(vcpu
)) {
367 if (!(cr4
& X86_CR4_PAE
)) {
368 printk(KERN_DEBUG
"set_cr4: #GP, clearing PAE while "
370 kvm_inject_gp(vcpu
, 0);
373 } else if (is_paging(vcpu
) && (cr4
& X86_CR4_PAE
)
374 && ((cr4
^ old_cr4
) & pdptr_bits
)
375 && !load_pdptrs(vcpu
, vcpu
->arch
.cr3
)) {
376 printk(KERN_DEBUG
"set_cr4: #GP, pdptrs reserved bits\n");
377 kvm_inject_gp(vcpu
, 0);
381 if (cr4
& X86_CR4_VMXE
) {
382 printk(KERN_DEBUG
"set_cr4: #GP, setting VMXE\n");
383 kvm_inject_gp(vcpu
, 0);
386 kvm_x86_ops
->set_cr4(vcpu
, cr4
);
387 vcpu
->arch
.cr4
= cr4
;
388 vcpu
->arch
.mmu
.base_role
.cr4_pge
= (cr4
& X86_CR4_PGE
) && !tdp_enabled
;
389 kvm_mmu_reset_context(vcpu
);
391 EXPORT_SYMBOL_GPL(kvm_set_cr4
);
393 void kvm_set_cr3(struct kvm_vcpu
*vcpu
, unsigned long cr3
)
395 if (cr3
== vcpu
->arch
.cr3
&& !pdptrs_changed(vcpu
)) {
396 kvm_mmu_sync_roots(vcpu
);
397 kvm_mmu_flush_tlb(vcpu
);
401 if (is_long_mode(vcpu
)) {
402 if (cr3
& CR3_L_MODE_RESERVED_BITS
) {
403 printk(KERN_DEBUG
"set_cr3: #GP, reserved bits\n");
404 kvm_inject_gp(vcpu
, 0);
409 if (cr3
& CR3_PAE_RESERVED_BITS
) {
411 "set_cr3: #GP, reserved bits\n");
412 kvm_inject_gp(vcpu
, 0);
415 if (is_paging(vcpu
) && !load_pdptrs(vcpu
, cr3
)) {
416 printk(KERN_DEBUG
"set_cr3: #GP, pdptrs "
418 kvm_inject_gp(vcpu
, 0);
423 * We don't check reserved bits in nonpae mode, because
424 * this isn't enforced, and VMware depends on this.
429 * Does the new cr3 value map to physical memory? (Note, we
430 * catch an invalid cr3 even in real-mode, because it would
431 * cause trouble later on when we turn on paging anyway.)
433 * A real CPU would silently accept an invalid cr3 and would
434 * attempt to use it - with largely undefined (and often hard
435 * to debug) behavior on the guest side.
437 if (unlikely(!gfn_to_memslot(vcpu
->kvm
, cr3
>> PAGE_SHIFT
)))
438 kvm_inject_gp(vcpu
, 0);
440 vcpu
->arch
.cr3
= cr3
;
441 vcpu
->arch
.mmu
.new_cr3(vcpu
);
444 EXPORT_SYMBOL_GPL(kvm_set_cr3
);
446 void kvm_set_cr8(struct kvm_vcpu
*vcpu
, unsigned long cr8
)
448 if (cr8
& CR8_RESERVED_BITS
) {
449 printk(KERN_DEBUG
"set_cr8: #GP, reserved bits 0x%lx\n", cr8
);
450 kvm_inject_gp(vcpu
, 0);
453 if (irqchip_in_kernel(vcpu
->kvm
))
454 kvm_lapic_set_tpr(vcpu
, cr8
);
456 vcpu
->arch
.cr8
= cr8
;
458 EXPORT_SYMBOL_GPL(kvm_set_cr8
);
460 unsigned long kvm_get_cr8(struct kvm_vcpu
*vcpu
)
462 if (irqchip_in_kernel(vcpu
->kvm
))
463 return kvm_lapic_get_cr8(vcpu
);
465 return vcpu
->arch
.cr8
;
467 EXPORT_SYMBOL_GPL(kvm_get_cr8
);
469 static inline u32
bit(int bitno
)
471 return 1 << (bitno
& 31);
475 * List of msr numbers which we expose to userspace through KVM_GET_MSRS
476 * and KVM_SET_MSRS, and KVM_GET_MSR_INDEX_LIST.
478 * This list is modified at module load time to reflect the
479 * capabilities of the host cpu.
481 static u32 msrs_to_save
[] = {
482 MSR_IA32_SYSENTER_CS
, MSR_IA32_SYSENTER_ESP
, MSR_IA32_SYSENTER_EIP
,
485 MSR_CSTAR
, MSR_KERNEL_GS_BASE
, MSR_SYSCALL_MASK
, MSR_LSTAR
,
487 MSR_IA32_TSC
, MSR_KVM_SYSTEM_TIME
, MSR_KVM_WALL_CLOCK
,
488 MSR_IA32_PERF_STATUS
, MSR_IA32_CR_PAT
, MSR_VM_HSAVE_PA
491 static unsigned num_msrs_to_save
;
493 static u32 emulated_msrs
[] = {
494 MSR_IA32_MISC_ENABLE
,
497 static void set_efer(struct kvm_vcpu
*vcpu
, u64 efer
)
499 if (efer
& efer_reserved_bits
) {
500 printk(KERN_DEBUG
"set_efer: 0x%llx #GP, reserved bits\n",
502 kvm_inject_gp(vcpu
, 0);
507 && (vcpu
->arch
.shadow_efer
& EFER_LME
) != (efer
& EFER_LME
)) {
508 printk(KERN_DEBUG
"set_efer: #GP, change LME while paging\n");
509 kvm_inject_gp(vcpu
, 0);
513 if (efer
& EFER_FFXSR
) {
514 struct kvm_cpuid_entry2
*feat
;
516 feat
= kvm_find_cpuid_entry(vcpu
, 0x80000001, 0);
517 if (!feat
|| !(feat
->edx
& bit(X86_FEATURE_FXSR_OPT
))) {
518 printk(KERN_DEBUG
"set_efer: #GP, enable FFXSR w/o CPUID capability\n");
519 kvm_inject_gp(vcpu
, 0);
524 if (efer
& EFER_SVME
) {
525 struct kvm_cpuid_entry2
*feat
;
527 feat
= kvm_find_cpuid_entry(vcpu
, 0x80000001, 0);
528 if (!feat
|| !(feat
->ecx
& bit(X86_FEATURE_SVM
))) {
529 printk(KERN_DEBUG
"set_efer: #GP, enable SVM w/o SVM\n");
530 kvm_inject_gp(vcpu
, 0);
535 kvm_x86_ops
->set_efer(vcpu
, efer
);
538 efer
|= vcpu
->arch
.shadow_efer
& EFER_LMA
;
540 vcpu
->arch
.shadow_efer
= efer
;
542 vcpu
->arch
.mmu
.base_role
.nxe
= (efer
& EFER_NX
) && !tdp_enabled
;
543 kvm_mmu_reset_context(vcpu
);
546 void kvm_enable_efer_bits(u64 mask
)
548 efer_reserved_bits
&= ~mask
;
550 EXPORT_SYMBOL_GPL(kvm_enable_efer_bits
);
554 * Writes msr value into into the appropriate "register".
555 * Returns 0 on success, non-0 otherwise.
556 * Assumes vcpu_load() was already called.
558 int kvm_set_msr(struct kvm_vcpu
*vcpu
, u32 msr_index
, u64 data
)
560 return kvm_x86_ops
->set_msr(vcpu
, msr_index
, data
);
564 * Adapt set_msr() to msr_io()'s calling convention
566 static int do_set_msr(struct kvm_vcpu
*vcpu
, unsigned index
, u64
*data
)
568 return kvm_set_msr(vcpu
, index
, *data
);
571 static void kvm_write_wall_clock(struct kvm
*kvm
, gpa_t wall_clock
)
574 struct pvclock_wall_clock wc
;
575 struct timespec now
, sys
, boot
;
582 kvm_write_guest(kvm
, wall_clock
, &version
, sizeof(version
));
585 * The guest calculates current wall clock time by adding
586 * system time (updated by kvm_write_guest_time below) to the
587 * wall clock specified here. guest system time equals host
588 * system time for us, thus we must fill in host boot time here.
590 now
= current_kernel_time();
592 boot
= ns_to_timespec(timespec_to_ns(&now
) - timespec_to_ns(&sys
));
594 wc
.sec
= boot
.tv_sec
;
595 wc
.nsec
= boot
.tv_nsec
;
596 wc
.version
= version
;
598 kvm_write_guest(kvm
, wall_clock
, &wc
, sizeof(wc
));
601 kvm_write_guest(kvm
, wall_clock
, &version
, sizeof(version
));
604 static uint32_t div_frac(uint32_t dividend
, uint32_t divisor
)
606 uint32_t quotient
, remainder
;
608 /* Don't try to replace with do_div(), this one calculates
609 * "(dividend << 32) / divisor" */
611 : "=a" (quotient
), "=d" (remainder
)
612 : "0" (0), "1" (dividend
), "r" (divisor
) );
616 static void kvm_set_time_scale(uint32_t tsc_khz
, struct pvclock_vcpu_time_info
*hv_clock
)
618 uint64_t nsecs
= 1000000000LL;
623 tps64
= tsc_khz
* 1000LL;
624 while (tps64
> nsecs
*2) {
629 tps32
= (uint32_t)tps64
;
630 while (tps32
<= (uint32_t)nsecs
) {
635 hv_clock
->tsc_shift
= shift
;
636 hv_clock
->tsc_to_system_mul
= div_frac(nsecs
, tps32
);
638 pr_debug("%s: tsc_khz %u, tsc_shift %d, tsc_mul %u\n",
639 __func__
, tsc_khz
, hv_clock
->tsc_shift
,
640 hv_clock
->tsc_to_system_mul
);
643 static DEFINE_PER_CPU(unsigned long, cpu_tsc_khz
);
645 static void kvm_write_guest_time(struct kvm_vcpu
*v
)
649 struct kvm_vcpu_arch
*vcpu
= &v
->arch
;
651 unsigned long this_tsc_khz
;
653 if ((!vcpu
->time_page
))
656 this_tsc_khz
= get_cpu_var(cpu_tsc_khz
);
657 if (unlikely(vcpu
->hv_clock_tsc_khz
!= this_tsc_khz
)) {
658 kvm_set_time_scale(this_tsc_khz
, &vcpu
->hv_clock
);
659 vcpu
->hv_clock_tsc_khz
= this_tsc_khz
;
661 put_cpu_var(cpu_tsc_khz
);
663 /* Keep irq disabled to prevent changes to the clock */
664 local_irq_save(flags
);
665 kvm_get_msr(v
, MSR_IA32_TSC
, &vcpu
->hv_clock
.tsc_timestamp
);
667 local_irq_restore(flags
);
669 /* With all the info we got, fill in the values */
671 vcpu
->hv_clock
.system_time
= ts
.tv_nsec
+
672 (NSEC_PER_SEC
* (u64
)ts
.tv_sec
);
674 * The interface expects us to write an even number signaling that the
675 * update is finished. Since the guest won't see the intermediate
676 * state, we just increase by 2 at the end.
678 vcpu
->hv_clock
.version
+= 2;
680 shared_kaddr
= kmap_atomic(vcpu
->time_page
, KM_USER0
);
682 memcpy(shared_kaddr
+ vcpu
->time_offset
, &vcpu
->hv_clock
,
683 sizeof(vcpu
->hv_clock
));
685 kunmap_atomic(shared_kaddr
, KM_USER0
);
687 mark_page_dirty(v
->kvm
, vcpu
->time
>> PAGE_SHIFT
);
690 static int kvm_request_guest_time_update(struct kvm_vcpu
*v
)
692 struct kvm_vcpu_arch
*vcpu
= &v
->arch
;
694 if (!vcpu
->time_page
)
696 set_bit(KVM_REQ_KVMCLOCK_UPDATE
, &v
->requests
);
700 static bool msr_mtrr_valid(unsigned msr
)
703 case 0x200 ... 0x200 + 2 * KVM_NR_VAR_MTRR
- 1:
704 case MSR_MTRRfix64K_00000
:
705 case MSR_MTRRfix16K_80000
:
706 case MSR_MTRRfix16K_A0000
:
707 case MSR_MTRRfix4K_C0000
:
708 case MSR_MTRRfix4K_C8000
:
709 case MSR_MTRRfix4K_D0000
:
710 case MSR_MTRRfix4K_D8000
:
711 case MSR_MTRRfix4K_E0000
:
712 case MSR_MTRRfix4K_E8000
:
713 case MSR_MTRRfix4K_F0000
:
714 case MSR_MTRRfix4K_F8000
:
715 case MSR_MTRRdefType
:
716 case MSR_IA32_CR_PAT
:
724 static bool valid_pat_type(unsigned t
)
726 return t
< 8 && (1 << t
) & 0xf3; /* 0, 1, 4, 5, 6, 7 */
729 static bool valid_mtrr_type(unsigned t
)
731 return t
< 8 && (1 << t
) & 0x73; /* 0, 1, 4, 5, 6 */
734 static bool mtrr_valid(struct kvm_vcpu
*vcpu
, u32 msr
, u64 data
)
738 if (!msr_mtrr_valid(msr
))
741 if (msr
== MSR_IA32_CR_PAT
) {
742 for (i
= 0; i
< 8; i
++)
743 if (!valid_pat_type((data
>> (i
* 8)) & 0xff))
746 } else if (msr
== MSR_MTRRdefType
) {
749 return valid_mtrr_type(data
& 0xff);
750 } else if (msr
>= MSR_MTRRfix64K_00000
&& msr
<= MSR_MTRRfix4K_F8000
) {
751 for (i
= 0; i
< 8 ; i
++)
752 if (!valid_mtrr_type((data
>> (i
* 8)) & 0xff))
758 return valid_mtrr_type(data
& 0xff);
761 static int set_msr_mtrr(struct kvm_vcpu
*vcpu
, u32 msr
, u64 data
)
763 u64
*p
= (u64
*)&vcpu
->arch
.mtrr_state
.fixed_ranges
;
765 if (!mtrr_valid(vcpu
, msr
, data
))
768 if (msr
== MSR_MTRRdefType
) {
769 vcpu
->arch
.mtrr_state
.def_type
= data
;
770 vcpu
->arch
.mtrr_state
.enabled
= (data
& 0xc00) >> 10;
771 } else if (msr
== MSR_MTRRfix64K_00000
)
773 else if (msr
== MSR_MTRRfix16K_80000
|| msr
== MSR_MTRRfix16K_A0000
)
774 p
[1 + msr
- MSR_MTRRfix16K_80000
] = data
;
775 else if (msr
>= MSR_MTRRfix4K_C0000
&& msr
<= MSR_MTRRfix4K_F8000
)
776 p
[3 + msr
- MSR_MTRRfix4K_C0000
] = data
;
777 else if (msr
== MSR_IA32_CR_PAT
)
778 vcpu
->arch
.pat
= data
;
779 else { /* Variable MTRRs */
780 int idx
, is_mtrr_mask
;
783 idx
= (msr
- 0x200) / 2;
784 is_mtrr_mask
= msr
- 0x200 - 2 * idx
;
787 (u64
*)&vcpu
->arch
.mtrr_state
.var_ranges
[idx
].base_lo
;
790 (u64
*)&vcpu
->arch
.mtrr_state
.var_ranges
[idx
].mask_lo
;
794 kvm_mmu_reset_context(vcpu
);
798 static int set_msr_mce(struct kvm_vcpu
*vcpu
, u32 msr
, u64 data
)
800 u64 mcg_cap
= vcpu
->arch
.mcg_cap
;
801 unsigned bank_num
= mcg_cap
& 0xff;
804 case MSR_IA32_MCG_STATUS
:
805 vcpu
->arch
.mcg_status
= data
;
807 case MSR_IA32_MCG_CTL
:
808 if (!(mcg_cap
& MCG_CTL_P
))
810 if (data
!= 0 && data
!= ~(u64
)0)
812 vcpu
->arch
.mcg_ctl
= data
;
815 if (msr
>= MSR_IA32_MC0_CTL
&&
816 msr
< MSR_IA32_MC0_CTL
+ 4 * bank_num
) {
817 u32 offset
= msr
- MSR_IA32_MC0_CTL
;
818 /* only 0 or all 1s can be written to IA32_MCi_CTL */
819 if ((offset
& 0x3) == 0 &&
820 data
!= 0 && data
!= ~(u64
)0)
822 vcpu
->arch
.mce_banks
[offset
] = data
;
830 int kvm_set_msr_common(struct kvm_vcpu
*vcpu
, u32 msr
, u64 data
)
834 set_efer(vcpu
, data
);
837 data
&= ~(u64
)0x40; /* ignore flush filter disable */
839 pr_unimpl(vcpu
, "unimplemented HWCR wrmsr: 0x%llx\n",
844 case MSR_IA32_DEBUGCTLMSR
:
846 /* We support the non-activated case already */
848 } else if (data
& ~(DEBUGCTLMSR_LBR
| DEBUGCTLMSR_BTF
)) {
849 /* Values other than LBR and BTF are vendor-specific,
850 thus reserved and should throw a #GP */
853 pr_unimpl(vcpu
, "%s: MSR_IA32_DEBUGCTLMSR 0x%llx, nop\n",
856 case MSR_IA32_UCODE_REV
:
857 case MSR_IA32_UCODE_WRITE
:
858 case MSR_VM_HSAVE_PA
:
860 case 0x200 ... 0x2ff:
861 return set_msr_mtrr(vcpu
, msr
, data
);
862 case MSR_IA32_APICBASE
:
863 kvm_set_apic_base(vcpu
, data
);
865 case MSR_IA32_MISC_ENABLE
:
866 vcpu
->arch
.ia32_misc_enable_msr
= data
;
868 case MSR_KVM_WALL_CLOCK
:
869 vcpu
->kvm
->arch
.wall_clock
= data
;
870 kvm_write_wall_clock(vcpu
->kvm
, data
);
872 case MSR_KVM_SYSTEM_TIME
: {
873 if (vcpu
->arch
.time_page
) {
874 kvm_release_page_dirty(vcpu
->arch
.time_page
);
875 vcpu
->arch
.time_page
= NULL
;
878 vcpu
->arch
.time
= data
;
880 /* we verify if the enable bit is set... */
884 /* ...but clean it before doing the actual write */
885 vcpu
->arch
.time_offset
= data
& ~(PAGE_MASK
| 1);
887 vcpu
->arch
.time_page
=
888 gfn_to_page(vcpu
->kvm
, data
>> PAGE_SHIFT
);
890 if (is_error_page(vcpu
->arch
.time_page
)) {
891 kvm_release_page_clean(vcpu
->arch
.time_page
);
892 vcpu
->arch
.time_page
= NULL
;
895 kvm_request_guest_time_update(vcpu
);
898 case MSR_IA32_MCG_CTL
:
899 case MSR_IA32_MCG_STATUS
:
900 case MSR_IA32_MC0_CTL
... MSR_IA32_MC0_CTL
+ 4 * KVM_MAX_MCE_BANKS
- 1:
901 return set_msr_mce(vcpu
, msr
, data
);
903 /* Performance counters are not protected by a CPUID bit,
904 * so we should check all of them in the generic path for the sake of
905 * cross vendor migration.
906 * Writing a zero into the event select MSRs disables them,
907 * which we perfectly emulate ;-). Any other value should be at least
908 * reported, some guests depend on them.
910 case MSR_P6_EVNTSEL0
:
911 case MSR_P6_EVNTSEL1
:
912 case MSR_K7_EVNTSEL0
:
913 case MSR_K7_EVNTSEL1
:
914 case MSR_K7_EVNTSEL2
:
915 case MSR_K7_EVNTSEL3
:
917 pr_unimpl(vcpu
, "unimplemented perfctr wrmsr: "
918 "0x%x data 0x%llx\n", msr
, data
);
920 /* at least RHEL 4 unconditionally writes to the perfctr registers,
921 * so we ignore writes to make it happy.
923 case MSR_P6_PERFCTR0
:
924 case MSR_P6_PERFCTR1
:
925 case MSR_K7_PERFCTR0
:
926 case MSR_K7_PERFCTR1
:
927 case MSR_K7_PERFCTR2
:
928 case MSR_K7_PERFCTR3
:
929 pr_unimpl(vcpu
, "unimplemented perfctr wrmsr: "
930 "0x%x data 0x%llx\n", msr
, data
);
933 pr_unimpl(vcpu
, "unhandled wrmsr: 0x%x data %llx\n", msr
, data
);
938 EXPORT_SYMBOL_GPL(kvm_set_msr_common
);
942 * Reads an msr value (of 'msr_index') into 'pdata'.
943 * Returns 0 on success, non-0 otherwise.
944 * Assumes vcpu_load() was already called.
946 int kvm_get_msr(struct kvm_vcpu
*vcpu
, u32 msr_index
, u64
*pdata
)
948 return kvm_x86_ops
->get_msr(vcpu
, msr_index
, pdata
);
951 static int get_msr_mtrr(struct kvm_vcpu
*vcpu
, u32 msr
, u64
*pdata
)
953 u64
*p
= (u64
*)&vcpu
->arch
.mtrr_state
.fixed_ranges
;
955 if (!msr_mtrr_valid(msr
))
958 if (msr
== MSR_MTRRdefType
)
959 *pdata
= vcpu
->arch
.mtrr_state
.def_type
+
960 (vcpu
->arch
.mtrr_state
.enabled
<< 10);
961 else if (msr
== MSR_MTRRfix64K_00000
)
963 else if (msr
== MSR_MTRRfix16K_80000
|| msr
== MSR_MTRRfix16K_A0000
)
964 *pdata
= p
[1 + msr
- MSR_MTRRfix16K_80000
];
965 else if (msr
>= MSR_MTRRfix4K_C0000
&& msr
<= MSR_MTRRfix4K_F8000
)
966 *pdata
= p
[3 + msr
- MSR_MTRRfix4K_C0000
];
967 else if (msr
== MSR_IA32_CR_PAT
)
968 *pdata
= vcpu
->arch
.pat
;
969 else { /* Variable MTRRs */
970 int idx
, is_mtrr_mask
;
973 idx
= (msr
- 0x200) / 2;
974 is_mtrr_mask
= msr
- 0x200 - 2 * idx
;
977 (u64
*)&vcpu
->arch
.mtrr_state
.var_ranges
[idx
].base_lo
;
980 (u64
*)&vcpu
->arch
.mtrr_state
.var_ranges
[idx
].mask_lo
;
987 static int get_msr_mce(struct kvm_vcpu
*vcpu
, u32 msr
, u64
*pdata
)
990 u64 mcg_cap
= vcpu
->arch
.mcg_cap
;
991 unsigned bank_num
= mcg_cap
& 0xff;
994 case MSR_IA32_P5_MC_ADDR
:
995 case MSR_IA32_P5_MC_TYPE
:
998 case MSR_IA32_MCG_CAP
:
999 data
= vcpu
->arch
.mcg_cap
;
1001 case MSR_IA32_MCG_CTL
:
1002 if (!(mcg_cap
& MCG_CTL_P
))
1004 data
= vcpu
->arch
.mcg_ctl
;
1006 case MSR_IA32_MCG_STATUS
:
1007 data
= vcpu
->arch
.mcg_status
;
1010 if (msr
>= MSR_IA32_MC0_CTL
&&
1011 msr
< MSR_IA32_MC0_CTL
+ 4 * bank_num
) {
1012 u32 offset
= msr
- MSR_IA32_MC0_CTL
;
1013 data
= vcpu
->arch
.mce_banks
[offset
];
1022 int kvm_get_msr_common(struct kvm_vcpu
*vcpu
, u32 msr
, u64
*pdata
)
1027 case MSR_IA32_PLATFORM_ID
:
1028 case MSR_IA32_UCODE_REV
:
1029 case MSR_IA32_EBL_CR_POWERON
:
1030 case MSR_IA32_DEBUGCTLMSR
:
1031 case MSR_IA32_LASTBRANCHFROMIP
:
1032 case MSR_IA32_LASTBRANCHTOIP
:
1033 case MSR_IA32_LASTINTFROMIP
:
1034 case MSR_IA32_LASTINTTOIP
:
1037 case MSR_VM_HSAVE_PA
:
1038 case MSR_P6_EVNTSEL0
:
1039 case MSR_P6_EVNTSEL1
:
1040 case MSR_K7_EVNTSEL0
:
1041 case MSR_K8_INT_PENDING_MSG
:
1045 data
= 0x500 | KVM_NR_VAR_MTRR
;
1047 case 0x200 ... 0x2ff:
1048 return get_msr_mtrr(vcpu
, msr
, pdata
);
1049 case 0xcd: /* fsb frequency */
1052 case MSR_IA32_APICBASE
:
1053 data
= kvm_get_apic_base(vcpu
);
1055 case MSR_IA32_MISC_ENABLE
:
1056 data
= vcpu
->arch
.ia32_misc_enable_msr
;
1058 case MSR_IA32_PERF_STATUS
:
1059 /* TSC increment by tick */
1061 /* CPU multiplier */
1062 data
|= (((uint64_t)4ULL) << 40);
1065 data
= vcpu
->arch
.shadow_efer
;
1067 case MSR_KVM_WALL_CLOCK
:
1068 data
= vcpu
->kvm
->arch
.wall_clock
;
1070 case MSR_KVM_SYSTEM_TIME
:
1071 data
= vcpu
->arch
.time
;
1073 case MSR_IA32_P5_MC_ADDR
:
1074 case MSR_IA32_P5_MC_TYPE
:
1075 case MSR_IA32_MCG_CAP
:
1076 case MSR_IA32_MCG_CTL
:
1077 case MSR_IA32_MCG_STATUS
:
1078 case MSR_IA32_MC0_CTL
... MSR_IA32_MC0_CTL
+ 4 * KVM_MAX_MCE_BANKS
- 1:
1079 return get_msr_mce(vcpu
, msr
, pdata
);
1081 pr_unimpl(vcpu
, "unhandled rdmsr: 0x%x\n", msr
);
1087 EXPORT_SYMBOL_GPL(kvm_get_msr_common
);
1090 * Read or write a bunch of msrs. All parameters are kernel addresses.
1092 * @return number of msrs set successfully.
1094 static int __msr_io(struct kvm_vcpu
*vcpu
, struct kvm_msrs
*msrs
,
1095 struct kvm_msr_entry
*entries
,
1096 int (*do_msr
)(struct kvm_vcpu
*vcpu
,
1097 unsigned index
, u64
*data
))
1103 down_read(&vcpu
->kvm
->slots_lock
);
1104 for (i
= 0; i
< msrs
->nmsrs
; ++i
)
1105 if (do_msr(vcpu
, entries
[i
].index
, &entries
[i
].data
))
1107 up_read(&vcpu
->kvm
->slots_lock
);
1115 * Read or write a bunch of msrs. Parameters are user addresses.
1117 * @return number of msrs set successfully.
1119 static int msr_io(struct kvm_vcpu
*vcpu
, struct kvm_msrs __user
*user_msrs
,
1120 int (*do_msr
)(struct kvm_vcpu
*vcpu
,
1121 unsigned index
, u64
*data
),
1124 struct kvm_msrs msrs
;
1125 struct kvm_msr_entry
*entries
;
1130 if (copy_from_user(&msrs
, user_msrs
, sizeof msrs
))
1134 if (msrs
.nmsrs
>= MAX_IO_MSRS
)
1138 size
= sizeof(struct kvm_msr_entry
) * msrs
.nmsrs
;
1139 entries
= vmalloc(size
);
1144 if (copy_from_user(entries
, user_msrs
->entries
, size
))
1147 r
= n
= __msr_io(vcpu
, &msrs
, entries
, do_msr
);
1152 if (writeback
&& copy_to_user(user_msrs
->entries
, entries
, size
))
1163 int kvm_dev_ioctl_check_extension(long ext
)
1168 case KVM_CAP_IRQCHIP
:
1170 case KVM_CAP_MMU_SHADOW_CACHE_CONTROL
:
1171 case KVM_CAP_SET_TSS_ADDR
:
1172 case KVM_CAP_EXT_CPUID
:
1173 case KVM_CAP_CLOCKSOURCE
:
1175 case KVM_CAP_NOP_IO_DELAY
:
1176 case KVM_CAP_MP_STATE
:
1177 case KVM_CAP_SYNC_MMU
:
1178 case KVM_CAP_REINJECT_CONTROL
:
1179 case KVM_CAP_IRQ_INJECT_STATUS
:
1180 case KVM_CAP_ASSIGN_DEV_IRQ
:
1185 case KVM_CAP_COALESCED_MMIO
:
1186 r
= KVM_COALESCED_MMIO_PAGE_OFFSET
;
1189 r
= !kvm_x86_ops
->cpu_has_accelerated_tpr();
1191 case KVM_CAP_NR_VCPUS
:
1194 case KVM_CAP_NR_MEMSLOTS
:
1195 r
= KVM_MEMORY_SLOTS
;
1197 case KVM_CAP_PV_MMU
:
1204 r
= KVM_MAX_MCE_BANKS
;
1214 long kvm_arch_dev_ioctl(struct file
*filp
,
1215 unsigned int ioctl
, unsigned long arg
)
1217 void __user
*argp
= (void __user
*)arg
;
1221 case KVM_GET_MSR_INDEX_LIST
: {
1222 struct kvm_msr_list __user
*user_msr_list
= argp
;
1223 struct kvm_msr_list msr_list
;
1227 if (copy_from_user(&msr_list
, user_msr_list
, sizeof msr_list
))
1230 msr_list
.nmsrs
= num_msrs_to_save
+ ARRAY_SIZE(emulated_msrs
);
1231 if (copy_to_user(user_msr_list
, &msr_list
, sizeof msr_list
))
1234 if (n
< msr_list
.nmsrs
)
1237 if (copy_to_user(user_msr_list
->indices
, &msrs_to_save
,
1238 num_msrs_to_save
* sizeof(u32
)))
1240 if (copy_to_user(user_msr_list
->indices
+ num_msrs_to_save
,
1242 ARRAY_SIZE(emulated_msrs
) * sizeof(u32
)))
1247 case KVM_GET_SUPPORTED_CPUID
: {
1248 struct kvm_cpuid2 __user
*cpuid_arg
= argp
;
1249 struct kvm_cpuid2 cpuid
;
1252 if (copy_from_user(&cpuid
, cpuid_arg
, sizeof cpuid
))
1254 r
= kvm_dev_ioctl_get_supported_cpuid(&cpuid
,
1255 cpuid_arg
->entries
);
1260 if (copy_to_user(cpuid_arg
, &cpuid
, sizeof cpuid
))
1265 case KVM_X86_GET_MCE_CAP_SUPPORTED
: {
1268 mce_cap
= KVM_MCE_CAP_SUPPORTED
;
1270 if (copy_to_user(argp
, &mce_cap
, sizeof mce_cap
))
1282 void kvm_arch_vcpu_load(struct kvm_vcpu
*vcpu
, int cpu
)
1284 kvm_x86_ops
->vcpu_load(vcpu
, cpu
);
1285 kvm_request_guest_time_update(vcpu
);
1288 void kvm_arch_vcpu_put(struct kvm_vcpu
*vcpu
)
1290 kvm_x86_ops
->vcpu_put(vcpu
);
1291 kvm_put_guest_fpu(vcpu
);
1294 static int is_efer_nx(void)
1296 unsigned long long efer
= 0;
1298 rdmsrl_safe(MSR_EFER
, &efer
);
1299 return efer
& EFER_NX
;
1302 static void cpuid_fix_nx_cap(struct kvm_vcpu
*vcpu
)
1305 struct kvm_cpuid_entry2
*e
, *entry
;
1308 for (i
= 0; i
< vcpu
->arch
.cpuid_nent
; ++i
) {
1309 e
= &vcpu
->arch
.cpuid_entries
[i
];
1310 if (e
->function
== 0x80000001) {
1315 if (entry
&& (entry
->edx
& (1 << 20)) && !is_efer_nx()) {
1316 entry
->edx
&= ~(1 << 20);
1317 printk(KERN_INFO
"kvm: guest NX capability removed\n");
1321 /* when an old userspace process fills a new kernel module */
1322 static int kvm_vcpu_ioctl_set_cpuid(struct kvm_vcpu
*vcpu
,
1323 struct kvm_cpuid
*cpuid
,
1324 struct kvm_cpuid_entry __user
*entries
)
1327 struct kvm_cpuid_entry
*cpuid_entries
;
1330 if (cpuid
->nent
> KVM_MAX_CPUID_ENTRIES
)
1333 cpuid_entries
= vmalloc(sizeof(struct kvm_cpuid_entry
) * cpuid
->nent
);
1337 if (copy_from_user(cpuid_entries
, entries
,
1338 cpuid
->nent
* sizeof(struct kvm_cpuid_entry
)))
1340 for (i
= 0; i
< cpuid
->nent
; i
++) {
1341 vcpu
->arch
.cpuid_entries
[i
].function
= cpuid_entries
[i
].function
;
1342 vcpu
->arch
.cpuid_entries
[i
].eax
= cpuid_entries
[i
].eax
;
1343 vcpu
->arch
.cpuid_entries
[i
].ebx
= cpuid_entries
[i
].ebx
;
1344 vcpu
->arch
.cpuid_entries
[i
].ecx
= cpuid_entries
[i
].ecx
;
1345 vcpu
->arch
.cpuid_entries
[i
].edx
= cpuid_entries
[i
].edx
;
1346 vcpu
->arch
.cpuid_entries
[i
].index
= 0;
1347 vcpu
->arch
.cpuid_entries
[i
].flags
= 0;
1348 vcpu
->arch
.cpuid_entries
[i
].padding
[0] = 0;
1349 vcpu
->arch
.cpuid_entries
[i
].padding
[1] = 0;
1350 vcpu
->arch
.cpuid_entries
[i
].padding
[2] = 0;
1352 vcpu
->arch
.cpuid_nent
= cpuid
->nent
;
1353 cpuid_fix_nx_cap(vcpu
);
1357 vfree(cpuid_entries
);
1362 static int kvm_vcpu_ioctl_set_cpuid2(struct kvm_vcpu
*vcpu
,
1363 struct kvm_cpuid2
*cpuid
,
1364 struct kvm_cpuid_entry2 __user
*entries
)
1369 if (cpuid
->nent
> KVM_MAX_CPUID_ENTRIES
)
1372 if (copy_from_user(&vcpu
->arch
.cpuid_entries
, entries
,
1373 cpuid
->nent
* sizeof(struct kvm_cpuid_entry2
)))
1375 vcpu
->arch
.cpuid_nent
= cpuid
->nent
;
1382 static int kvm_vcpu_ioctl_get_cpuid2(struct kvm_vcpu
*vcpu
,
1383 struct kvm_cpuid2
*cpuid
,
1384 struct kvm_cpuid_entry2 __user
*entries
)
1389 if (cpuid
->nent
< vcpu
->arch
.cpuid_nent
)
1392 if (copy_to_user(entries
, &vcpu
->arch
.cpuid_entries
,
1393 vcpu
->arch
.cpuid_nent
* sizeof(struct kvm_cpuid_entry2
)))
1398 cpuid
->nent
= vcpu
->arch
.cpuid_nent
;
1402 static void do_cpuid_1_ent(struct kvm_cpuid_entry2
*entry
, u32 function
,
1405 entry
->function
= function
;
1406 entry
->index
= index
;
1407 cpuid_count(entry
->function
, entry
->index
,
1408 &entry
->eax
, &entry
->ebx
, &entry
->ecx
, &entry
->edx
);
1412 #define F(x) bit(X86_FEATURE_##x)
1414 static void do_cpuid_ent(struct kvm_cpuid_entry2
*entry
, u32 function
,
1415 u32 index
, int *nent
, int maxnent
)
1417 unsigned f_nx
= is_efer_nx() ? F(NX
) : 0;
1418 #ifdef CONFIG_X86_64
1419 unsigned f_lm
= F(LM
);
1425 const u32 kvm_supported_word0_x86_features
=
1426 F(FPU
) | F(VME
) | F(DE
) | F(PSE
) |
1427 F(TSC
) | F(MSR
) | F(PAE
) | F(MCE
) |
1428 F(CX8
) | F(APIC
) | 0 /* Reserved */ | F(SEP
) |
1429 F(MTRR
) | F(PGE
) | F(MCA
) | F(CMOV
) |
1430 F(PAT
) | F(PSE36
) | 0 /* PSN */ | F(CLFLSH
) |
1431 0 /* Reserved, DS, ACPI */ | F(MMX
) |
1432 F(FXSR
) | F(XMM
) | F(XMM2
) | F(SELFSNOOP
) |
1433 0 /* HTT, TM, Reserved, PBE */;
1434 /* cpuid 0x80000001.edx */
1435 const u32 kvm_supported_word1_x86_features
=
1436 F(FPU
) | F(VME
) | F(DE
) | F(PSE
) |
1437 F(TSC
) | F(MSR
) | F(PAE
) | F(MCE
) |
1438 F(CX8
) | F(APIC
) | 0 /* Reserved */ | F(SYSCALL
) |
1439 F(MTRR
) | F(PGE
) | F(MCA
) | F(CMOV
) |
1440 F(PAT
) | F(PSE36
) | 0 /* Reserved */ |
1441 f_nx
| 0 /* Reserved */ | F(MMXEXT
) | F(MMX
) |
1442 F(FXSR
) | F(FXSR_OPT
) | 0 /* GBPAGES */ | 0 /* RDTSCP */ |
1443 0 /* Reserved */ | f_lm
| F(3DNOWEXT
) | F(3DNOW
);
1445 const u32 kvm_supported_word4_x86_features
=
1446 F(XMM3
) | 0 /* Reserved, DTES64, MONITOR */ |
1447 0 /* DS-CPL, VMX, SMX, EST */ |
1448 0 /* TM2 */ | F(SSSE3
) | 0 /* CNXT-ID */ | 0 /* Reserved */ |
1449 0 /* Reserved */ | F(CX16
) | 0 /* xTPR Update, PDCM */ |
1450 0 /* Reserved, DCA */ | F(XMM4_1
) |
1451 F(XMM4_2
) | 0 /* x2APIC */ | F(MOVBE
) | F(POPCNT
) |
1452 0 /* Reserved, XSAVE, OSXSAVE */;
1453 /* cpuid 0x80000001.ecx */
1454 const u32 kvm_supported_word6_x86_features
=
1455 F(LAHF_LM
) | F(CMP_LEGACY
) | F(SVM
) | 0 /* ExtApicSpace */ |
1456 F(CR8_LEGACY
) | F(ABM
) | F(SSE4A
) | F(MISALIGNSSE
) |
1457 F(3DNOWPREFETCH
) | 0 /* OSVW */ | 0 /* IBS */ | F(SSE5
) |
1458 0 /* SKINIT */ | 0 /* WDT */;
1460 /* all calls to cpuid_count() should be made on the same cpu */
1462 do_cpuid_1_ent(entry
, function
, index
);
1467 entry
->eax
= min(entry
->eax
, (u32
)0xb);
1470 entry
->edx
&= kvm_supported_word0_x86_features
;
1471 entry
->ecx
&= kvm_supported_word4_x86_features
;
1473 /* function 2 entries are STATEFUL. That is, repeated cpuid commands
1474 * may return different values. This forces us to get_cpu() before
1475 * issuing the first command, and also to emulate this annoying behavior
1476 * in kvm_emulate_cpuid() using KVM_CPUID_FLAG_STATE_READ_NEXT */
1478 int t
, times
= entry
->eax
& 0xff;
1480 entry
->flags
|= KVM_CPUID_FLAG_STATEFUL_FUNC
;
1481 entry
->flags
|= KVM_CPUID_FLAG_STATE_READ_NEXT
;
1482 for (t
= 1; t
< times
&& *nent
< maxnent
; ++t
) {
1483 do_cpuid_1_ent(&entry
[t
], function
, 0);
1484 entry
[t
].flags
|= KVM_CPUID_FLAG_STATEFUL_FUNC
;
1489 /* function 4 and 0xb have additional index. */
1493 entry
->flags
|= KVM_CPUID_FLAG_SIGNIFCANT_INDEX
;
1494 /* read more entries until cache_type is zero */
1495 for (i
= 1; *nent
< maxnent
; ++i
) {
1496 cache_type
= entry
[i
- 1].eax
& 0x1f;
1499 do_cpuid_1_ent(&entry
[i
], function
, i
);
1501 KVM_CPUID_FLAG_SIGNIFCANT_INDEX
;
1509 entry
->flags
|= KVM_CPUID_FLAG_SIGNIFCANT_INDEX
;
1510 /* read more entries until level_type is zero */
1511 for (i
= 1; *nent
< maxnent
; ++i
) {
1512 level_type
= entry
[i
- 1].ecx
& 0xff00;
1515 do_cpuid_1_ent(&entry
[i
], function
, i
);
1517 KVM_CPUID_FLAG_SIGNIFCANT_INDEX
;
1523 entry
->eax
= min(entry
->eax
, 0x8000001a);
1526 entry
->edx
&= kvm_supported_word1_x86_features
;
1527 entry
->ecx
&= kvm_supported_word6_x86_features
;
1535 static int kvm_dev_ioctl_get_supported_cpuid(struct kvm_cpuid2
*cpuid
,
1536 struct kvm_cpuid_entry2 __user
*entries
)
1538 struct kvm_cpuid_entry2
*cpuid_entries
;
1539 int limit
, nent
= 0, r
= -E2BIG
;
1542 if (cpuid
->nent
< 1)
1545 cpuid_entries
= vmalloc(sizeof(struct kvm_cpuid_entry2
) * cpuid
->nent
);
1549 do_cpuid_ent(&cpuid_entries
[0], 0, 0, &nent
, cpuid
->nent
);
1550 limit
= cpuid_entries
[0].eax
;
1551 for (func
= 1; func
<= limit
&& nent
< cpuid
->nent
; ++func
)
1552 do_cpuid_ent(&cpuid_entries
[nent
], func
, 0,
1553 &nent
, cpuid
->nent
);
1555 if (nent
>= cpuid
->nent
)
1558 do_cpuid_ent(&cpuid_entries
[nent
], 0x80000000, 0, &nent
, cpuid
->nent
);
1559 limit
= cpuid_entries
[nent
- 1].eax
;
1560 for (func
= 0x80000001; func
<= limit
&& nent
< cpuid
->nent
; ++func
)
1561 do_cpuid_ent(&cpuid_entries
[nent
], func
, 0,
1562 &nent
, cpuid
->nent
);
1564 if (nent
>= cpuid
->nent
)
1568 if (copy_to_user(entries
, cpuid_entries
,
1569 nent
* sizeof(struct kvm_cpuid_entry2
)))
1575 vfree(cpuid_entries
);
1580 static int kvm_vcpu_ioctl_get_lapic(struct kvm_vcpu
*vcpu
,
1581 struct kvm_lapic_state
*s
)
1584 memcpy(s
->regs
, vcpu
->arch
.apic
->regs
, sizeof *s
);
1590 static int kvm_vcpu_ioctl_set_lapic(struct kvm_vcpu
*vcpu
,
1591 struct kvm_lapic_state
*s
)
1594 memcpy(vcpu
->arch
.apic
->regs
, s
->regs
, sizeof *s
);
1595 kvm_apic_post_state_restore(vcpu
);
1601 static int kvm_vcpu_ioctl_interrupt(struct kvm_vcpu
*vcpu
,
1602 struct kvm_interrupt
*irq
)
1604 if (irq
->irq
< 0 || irq
->irq
>= 256)
1606 if (irqchip_in_kernel(vcpu
->kvm
))
1610 kvm_queue_interrupt(vcpu
, irq
->irq
, false);
1617 static int kvm_vcpu_ioctl_nmi(struct kvm_vcpu
*vcpu
)
1620 kvm_inject_nmi(vcpu
);
1626 static int vcpu_ioctl_tpr_access_reporting(struct kvm_vcpu
*vcpu
,
1627 struct kvm_tpr_access_ctl
*tac
)
1631 vcpu
->arch
.tpr_access_reporting
= !!tac
->enabled
;
1635 static int kvm_vcpu_ioctl_x86_setup_mce(struct kvm_vcpu
*vcpu
,
1639 unsigned bank_num
= mcg_cap
& 0xff, bank
;
1644 if (mcg_cap
& ~(KVM_MCE_CAP_SUPPORTED
| 0xff | 0xff0000))
1647 vcpu
->arch
.mcg_cap
= mcg_cap
;
1648 /* Init IA32_MCG_CTL to all 1s */
1649 if (mcg_cap
& MCG_CTL_P
)
1650 vcpu
->arch
.mcg_ctl
= ~(u64
)0;
1651 /* Init IA32_MCi_CTL to all 1s */
1652 for (bank
= 0; bank
< bank_num
; bank
++)
1653 vcpu
->arch
.mce_banks
[bank
*4] = ~(u64
)0;
1658 static int kvm_vcpu_ioctl_x86_set_mce(struct kvm_vcpu
*vcpu
,
1659 struct kvm_x86_mce
*mce
)
1661 u64 mcg_cap
= vcpu
->arch
.mcg_cap
;
1662 unsigned bank_num
= mcg_cap
& 0xff;
1663 u64
*banks
= vcpu
->arch
.mce_banks
;
1665 if (mce
->bank
>= bank_num
|| !(mce
->status
& MCI_STATUS_VAL
))
1668 * if IA32_MCG_CTL is not all 1s, the uncorrected error
1669 * reporting is disabled
1671 if ((mce
->status
& MCI_STATUS_UC
) && (mcg_cap
& MCG_CTL_P
) &&
1672 vcpu
->arch
.mcg_ctl
!= ~(u64
)0)
1674 banks
+= 4 * mce
->bank
;
1676 * if IA32_MCi_CTL is not all 1s, the uncorrected error
1677 * reporting is disabled for the bank
1679 if ((mce
->status
& MCI_STATUS_UC
) && banks
[0] != ~(u64
)0)
1681 if (mce
->status
& MCI_STATUS_UC
) {
1682 if ((vcpu
->arch
.mcg_status
& MCG_STATUS_MCIP
) ||
1683 !(vcpu
->arch
.cr4
& X86_CR4_MCE
)) {
1684 printk(KERN_DEBUG
"kvm: set_mce: "
1685 "injects mce exception while "
1686 "previous one is in progress!\n");
1687 set_bit(KVM_REQ_TRIPLE_FAULT
, &vcpu
->requests
);
1690 if (banks
[1] & MCI_STATUS_VAL
)
1691 mce
->status
|= MCI_STATUS_OVER
;
1692 banks
[2] = mce
->addr
;
1693 banks
[3] = mce
->misc
;
1694 vcpu
->arch
.mcg_status
= mce
->mcg_status
;
1695 banks
[1] = mce
->status
;
1696 kvm_queue_exception(vcpu
, MC_VECTOR
);
1697 } else if (!(banks
[1] & MCI_STATUS_VAL
)
1698 || !(banks
[1] & MCI_STATUS_UC
)) {
1699 if (banks
[1] & MCI_STATUS_VAL
)
1700 mce
->status
|= MCI_STATUS_OVER
;
1701 banks
[2] = mce
->addr
;
1702 banks
[3] = mce
->misc
;
1703 banks
[1] = mce
->status
;
1705 banks
[1] |= MCI_STATUS_OVER
;
1709 long kvm_arch_vcpu_ioctl(struct file
*filp
,
1710 unsigned int ioctl
, unsigned long arg
)
1712 struct kvm_vcpu
*vcpu
= filp
->private_data
;
1713 void __user
*argp
= (void __user
*)arg
;
1715 struct kvm_lapic_state
*lapic
= NULL
;
1718 case KVM_GET_LAPIC
: {
1719 lapic
= kzalloc(sizeof(struct kvm_lapic_state
), GFP_KERNEL
);
1724 r
= kvm_vcpu_ioctl_get_lapic(vcpu
, lapic
);
1728 if (copy_to_user(argp
, lapic
, sizeof(struct kvm_lapic_state
)))
1733 case KVM_SET_LAPIC
: {
1734 lapic
= kmalloc(sizeof(struct kvm_lapic_state
), GFP_KERNEL
);
1739 if (copy_from_user(lapic
, argp
, sizeof(struct kvm_lapic_state
)))
1741 r
= kvm_vcpu_ioctl_set_lapic(vcpu
, lapic
);
1747 case KVM_INTERRUPT
: {
1748 struct kvm_interrupt irq
;
1751 if (copy_from_user(&irq
, argp
, sizeof irq
))
1753 r
= kvm_vcpu_ioctl_interrupt(vcpu
, &irq
);
1760 r
= kvm_vcpu_ioctl_nmi(vcpu
);
1766 case KVM_SET_CPUID
: {
1767 struct kvm_cpuid __user
*cpuid_arg
= argp
;
1768 struct kvm_cpuid cpuid
;
1771 if (copy_from_user(&cpuid
, cpuid_arg
, sizeof cpuid
))
1773 r
= kvm_vcpu_ioctl_set_cpuid(vcpu
, &cpuid
, cpuid_arg
->entries
);
1778 case KVM_SET_CPUID2
: {
1779 struct kvm_cpuid2 __user
*cpuid_arg
= argp
;
1780 struct kvm_cpuid2 cpuid
;
1783 if (copy_from_user(&cpuid
, cpuid_arg
, sizeof cpuid
))
1785 r
= kvm_vcpu_ioctl_set_cpuid2(vcpu
, &cpuid
,
1786 cpuid_arg
->entries
);
1791 case KVM_GET_CPUID2
: {
1792 struct kvm_cpuid2 __user
*cpuid_arg
= argp
;
1793 struct kvm_cpuid2 cpuid
;
1796 if (copy_from_user(&cpuid
, cpuid_arg
, sizeof cpuid
))
1798 r
= kvm_vcpu_ioctl_get_cpuid2(vcpu
, &cpuid
,
1799 cpuid_arg
->entries
);
1803 if (copy_to_user(cpuid_arg
, &cpuid
, sizeof cpuid
))
1809 r
= msr_io(vcpu
, argp
, kvm_get_msr
, 1);
1812 r
= msr_io(vcpu
, argp
, do_set_msr
, 0);
1814 case KVM_TPR_ACCESS_REPORTING
: {
1815 struct kvm_tpr_access_ctl tac
;
1818 if (copy_from_user(&tac
, argp
, sizeof tac
))
1820 r
= vcpu_ioctl_tpr_access_reporting(vcpu
, &tac
);
1824 if (copy_to_user(argp
, &tac
, sizeof tac
))
1829 case KVM_SET_VAPIC_ADDR
: {
1830 struct kvm_vapic_addr va
;
1833 if (!irqchip_in_kernel(vcpu
->kvm
))
1836 if (copy_from_user(&va
, argp
, sizeof va
))
1839 kvm_lapic_set_vapic_addr(vcpu
, va
.vapic_addr
);
1842 case KVM_X86_SETUP_MCE
: {
1846 if (copy_from_user(&mcg_cap
, argp
, sizeof mcg_cap
))
1848 r
= kvm_vcpu_ioctl_x86_setup_mce(vcpu
, mcg_cap
);
1851 case KVM_X86_SET_MCE
: {
1852 struct kvm_x86_mce mce
;
1855 if (copy_from_user(&mce
, argp
, sizeof mce
))
1857 r
= kvm_vcpu_ioctl_x86_set_mce(vcpu
, &mce
);
1868 static int kvm_vm_ioctl_set_tss_addr(struct kvm
*kvm
, unsigned long addr
)
1872 if (addr
> (unsigned int)(-3 * PAGE_SIZE
))
1874 ret
= kvm_x86_ops
->set_tss_addr(kvm
, addr
);
1878 static int kvm_vm_ioctl_set_nr_mmu_pages(struct kvm
*kvm
,
1879 u32 kvm_nr_mmu_pages
)
1881 if (kvm_nr_mmu_pages
< KVM_MIN_ALLOC_MMU_PAGES
)
1884 down_write(&kvm
->slots_lock
);
1885 spin_lock(&kvm
->mmu_lock
);
1887 kvm_mmu_change_mmu_pages(kvm
, kvm_nr_mmu_pages
);
1888 kvm
->arch
.n_requested_mmu_pages
= kvm_nr_mmu_pages
;
1890 spin_unlock(&kvm
->mmu_lock
);
1891 up_write(&kvm
->slots_lock
);
1895 static int kvm_vm_ioctl_get_nr_mmu_pages(struct kvm
*kvm
)
1897 return kvm
->arch
.n_alloc_mmu_pages
;
1900 gfn_t
unalias_gfn(struct kvm
*kvm
, gfn_t gfn
)
1903 struct kvm_mem_alias
*alias
;
1905 for (i
= 0; i
< kvm
->arch
.naliases
; ++i
) {
1906 alias
= &kvm
->arch
.aliases
[i
];
1907 if (gfn
>= alias
->base_gfn
1908 && gfn
< alias
->base_gfn
+ alias
->npages
)
1909 return alias
->target_gfn
+ gfn
- alias
->base_gfn
;
1915 * Set a new alias region. Aliases map a portion of physical memory into
1916 * another portion. This is useful for memory windows, for example the PC
1919 static int kvm_vm_ioctl_set_memory_alias(struct kvm
*kvm
,
1920 struct kvm_memory_alias
*alias
)
1923 struct kvm_mem_alias
*p
;
1926 /* General sanity checks */
1927 if (alias
->memory_size
& (PAGE_SIZE
- 1))
1929 if (alias
->guest_phys_addr
& (PAGE_SIZE
- 1))
1931 if (alias
->slot
>= KVM_ALIAS_SLOTS
)
1933 if (alias
->guest_phys_addr
+ alias
->memory_size
1934 < alias
->guest_phys_addr
)
1936 if (alias
->target_phys_addr
+ alias
->memory_size
1937 < alias
->target_phys_addr
)
1940 down_write(&kvm
->slots_lock
);
1941 spin_lock(&kvm
->mmu_lock
);
1943 p
= &kvm
->arch
.aliases
[alias
->slot
];
1944 p
->base_gfn
= alias
->guest_phys_addr
>> PAGE_SHIFT
;
1945 p
->npages
= alias
->memory_size
>> PAGE_SHIFT
;
1946 p
->target_gfn
= alias
->target_phys_addr
>> PAGE_SHIFT
;
1948 for (n
= KVM_ALIAS_SLOTS
; n
> 0; --n
)
1949 if (kvm
->arch
.aliases
[n
- 1].npages
)
1951 kvm
->arch
.naliases
= n
;
1953 spin_unlock(&kvm
->mmu_lock
);
1954 kvm_mmu_zap_all(kvm
);
1956 up_write(&kvm
->slots_lock
);
1964 static int kvm_vm_ioctl_get_irqchip(struct kvm
*kvm
, struct kvm_irqchip
*chip
)
1969 switch (chip
->chip_id
) {
1970 case KVM_IRQCHIP_PIC_MASTER
:
1971 memcpy(&chip
->chip
.pic
,
1972 &pic_irqchip(kvm
)->pics
[0],
1973 sizeof(struct kvm_pic_state
));
1975 case KVM_IRQCHIP_PIC_SLAVE
:
1976 memcpy(&chip
->chip
.pic
,
1977 &pic_irqchip(kvm
)->pics
[1],
1978 sizeof(struct kvm_pic_state
));
1980 case KVM_IRQCHIP_IOAPIC
:
1981 memcpy(&chip
->chip
.ioapic
,
1982 ioapic_irqchip(kvm
),
1983 sizeof(struct kvm_ioapic_state
));
1992 static int kvm_vm_ioctl_set_irqchip(struct kvm
*kvm
, struct kvm_irqchip
*chip
)
1997 switch (chip
->chip_id
) {
1998 case KVM_IRQCHIP_PIC_MASTER
:
1999 spin_lock(&pic_irqchip(kvm
)->lock
);
2000 memcpy(&pic_irqchip(kvm
)->pics
[0],
2002 sizeof(struct kvm_pic_state
));
2003 spin_unlock(&pic_irqchip(kvm
)->lock
);
2005 case KVM_IRQCHIP_PIC_SLAVE
:
2006 spin_lock(&pic_irqchip(kvm
)->lock
);
2007 memcpy(&pic_irqchip(kvm
)->pics
[1],
2009 sizeof(struct kvm_pic_state
));
2010 spin_unlock(&pic_irqchip(kvm
)->lock
);
2012 case KVM_IRQCHIP_IOAPIC
:
2013 mutex_lock(&kvm
->irq_lock
);
2014 memcpy(ioapic_irqchip(kvm
),
2016 sizeof(struct kvm_ioapic_state
));
2017 mutex_unlock(&kvm
->irq_lock
);
2023 kvm_pic_update_irq(pic_irqchip(kvm
));
2027 static int kvm_vm_ioctl_get_pit(struct kvm
*kvm
, struct kvm_pit_state
*ps
)
2031 mutex_lock(&kvm
->arch
.vpit
->pit_state
.lock
);
2032 memcpy(ps
, &kvm
->arch
.vpit
->pit_state
, sizeof(struct kvm_pit_state
));
2033 mutex_unlock(&kvm
->arch
.vpit
->pit_state
.lock
);
2037 static int kvm_vm_ioctl_set_pit(struct kvm
*kvm
, struct kvm_pit_state
*ps
)
2041 mutex_lock(&kvm
->arch
.vpit
->pit_state
.lock
);
2042 memcpy(&kvm
->arch
.vpit
->pit_state
, ps
, sizeof(struct kvm_pit_state
));
2043 kvm_pit_load_count(kvm
, 0, ps
->channels
[0].count
);
2044 mutex_unlock(&kvm
->arch
.vpit
->pit_state
.lock
);
2048 static int kvm_vm_ioctl_reinject(struct kvm
*kvm
,
2049 struct kvm_reinject_control
*control
)
2051 if (!kvm
->arch
.vpit
)
2053 mutex_lock(&kvm
->arch
.vpit
->pit_state
.lock
);
2054 kvm
->arch
.vpit
->pit_state
.pit_timer
.reinject
= control
->pit_reinject
;
2055 mutex_unlock(&kvm
->arch
.vpit
->pit_state
.lock
);
2060 * Get (and clear) the dirty memory log for a memory slot.
2062 int kvm_vm_ioctl_get_dirty_log(struct kvm
*kvm
,
2063 struct kvm_dirty_log
*log
)
2067 struct kvm_memory_slot
*memslot
;
2070 down_write(&kvm
->slots_lock
);
2072 r
= kvm_get_dirty_log(kvm
, log
, &is_dirty
);
2076 /* If nothing is dirty, don't bother messing with page tables. */
2078 spin_lock(&kvm
->mmu_lock
);
2079 kvm_mmu_slot_remove_write_access(kvm
, log
->slot
);
2080 spin_unlock(&kvm
->mmu_lock
);
2081 kvm_flush_remote_tlbs(kvm
);
2082 memslot
= &kvm
->memslots
[log
->slot
];
2083 n
= ALIGN(memslot
->npages
, BITS_PER_LONG
) / 8;
2084 memset(memslot
->dirty_bitmap
, 0, n
);
2088 up_write(&kvm
->slots_lock
);
2092 long kvm_arch_vm_ioctl(struct file
*filp
,
2093 unsigned int ioctl
, unsigned long arg
)
2095 struct kvm
*kvm
= filp
->private_data
;
2096 void __user
*argp
= (void __user
*)arg
;
2099 * This union makes it completely explicit to gcc-3.x
2100 * that these two variables' stack usage should be
2101 * combined, not added together.
2104 struct kvm_pit_state ps
;
2105 struct kvm_memory_alias alias
;
2106 struct kvm_pit_config pit_config
;
2110 case KVM_SET_TSS_ADDR
:
2111 r
= kvm_vm_ioctl_set_tss_addr(kvm
, arg
);
2115 case KVM_SET_MEMORY_REGION
: {
2116 struct kvm_memory_region kvm_mem
;
2117 struct kvm_userspace_memory_region kvm_userspace_mem
;
2120 if (copy_from_user(&kvm_mem
, argp
, sizeof kvm_mem
))
2122 kvm_userspace_mem
.slot
= kvm_mem
.slot
;
2123 kvm_userspace_mem
.flags
= kvm_mem
.flags
;
2124 kvm_userspace_mem
.guest_phys_addr
= kvm_mem
.guest_phys_addr
;
2125 kvm_userspace_mem
.memory_size
= kvm_mem
.memory_size
;
2126 r
= kvm_vm_ioctl_set_memory_region(kvm
, &kvm_userspace_mem
, 0);
2131 case KVM_SET_NR_MMU_PAGES
:
2132 r
= kvm_vm_ioctl_set_nr_mmu_pages(kvm
, arg
);
2136 case KVM_GET_NR_MMU_PAGES
:
2137 r
= kvm_vm_ioctl_get_nr_mmu_pages(kvm
);
2139 case KVM_SET_MEMORY_ALIAS
:
2141 if (copy_from_user(&u
.alias
, argp
, sizeof(struct kvm_memory_alias
)))
2143 r
= kvm_vm_ioctl_set_memory_alias(kvm
, &u
.alias
);
2147 case KVM_CREATE_IRQCHIP
:
2149 kvm
->arch
.vpic
= kvm_create_pic(kvm
);
2150 if (kvm
->arch
.vpic
) {
2151 r
= kvm_ioapic_init(kvm
);
2153 kfree(kvm
->arch
.vpic
);
2154 kvm
->arch
.vpic
= NULL
;
2159 r
= kvm_setup_default_irq_routing(kvm
);
2161 kfree(kvm
->arch
.vpic
);
2162 kfree(kvm
->arch
.vioapic
);
2166 case KVM_CREATE_PIT
:
2167 u
.pit_config
.flags
= KVM_PIT_SPEAKER_DUMMY
;
2169 case KVM_CREATE_PIT2
:
2171 if (copy_from_user(&u
.pit_config
, argp
,
2172 sizeof(struct kvm_pit_config
)))
2175 mutex_lock(&kvm
->lock
);
2178 goto create_pit_unlock
;
2180 kvm
->arch
.vpit
= kvm_create_pit(kvm
, u
.pit_config
.flags
);
2184 mutex_unlock(&kvm
->lock
);
2186 case KVM_IRQ_LINE_STATUS
:
2187 case KVM_IRQ_LINE
: {
2188 struct kvm_irq_level irq_event
;
2191 if (copy_from_user(&irq_event
, argp
, sizeof irq_event
))
2193 if (irqchip_in_kernel(kvm
)) {
2195 mutex_lock(&kvm
->irq_lock
);
2196 status
= kvm_set_irq(kvm
, KVM_USERSPACE_IRQ_SOURCE_ID
,
2197 irq_event
.irq
, irq_event
.level
);
2198 mutex_unlock(&kvm
->irq_lock
);
2199 if (ioctl
== KVM_IRQ_LINE_STATUS
) {
2200 irq_event
.status
= status
;
2201 if (copy_to_user(argp
, &irq_event
,
2209 case KVM_GET_IRQCHIP
: {
2210 /* 0: PIC master, 1: PIC slave, 2: IOAPIC */
2211 struct kvm_irqchip
*chip
= kmalloc(sizeof(*chip
), GFP_KERNEL
);
2217 if (copy_from_user(chip
, argp
, sizeof *chip
))
2218 goto get_irqchip_out
;
2220 if (!irqchip_in_kernel(kvm
))
2221 goto get_irqchip_out
;
2222 r
= kvm_vm_ioctl_get_irqchip(kvm
, chip
);
2224 goto get_irqchip_out
;
2226 if (copy_to_user(argp
, chip
, sizeof *chip
))
2227 goto get_irqchip_out
;
2235 case KVM_SET_IRQCHIP
: {
2236 /* 0: PIC master, 1: PIC slave, 2: IOAPIC */
2237 struct kvm_irqchip
*chip
= kmalloc(sizeof(*chip
), GFP_KERNEL
);
2243 if (copy_from_user(chip
, argp
, sizeof *chip
))
2244 goto set_irqchip_out
;
2246 if (!irqchip_in_kernel(kvm
))
2247 goto set_irqchip_out
;
2248 r
= kvm_vm_ioctl_set_irqchip(kvm
, chip
);
2250 goto set_irqchip_out
;
2260 if (copy_from_user(&u
.ps
, argp
, sizeof(struct kvm_pit_state
)))
2263 if (!kvm
->arch
.vpit
)
2265 r
= kvm_vm_ioctl_get_pit(kvm
, &u
.ps
);
2269 if (copy_to_user(argp
, &u
.ps
, sizeof(struct kvm_pit_state
)))
2276 if (copy_from_user(&u
.ps
, argp
, sizeof u
.ps
))
2279 if (!kvm
->arch
.vpit
)
2281 r
= kvm_vm_ioctl_set_pit(kvm
, &u
.ps
);
2287 case KVM_REINJECT_CONTROL
: {
2288 struct kvm_reinject_control control
;
2290 if (copy_from_user(&control
, argp
, sizeof(control
)))
2292 r
= kvm_vm_ioctl_reinject(kvm
, &control
);
2305 static void kvm_init_msr_list(void)
2310 for (i
= j
= 0; i
< ARRAY_SIZE(msrs_to_save
); i
++) {
2311 if (rdmsr_safe(msrs_to_save
[i
], &dummy
[0], &dummy
[1]) < 0)
2314 msrs_to_save
[j
] = msrs_to_save
[i
];
2317 num_msrs_to_save
= j
;
2321 * Only apic need an MMIO device hook, so shortcut now..
2323 static struct kvm_io_device
*vcpu_find_pervcpu_dev(struct kvm_vcpu
*vcpu
,
2324 gpa_t addr
, int len
,
2327 struct kvm_io_device
*dev
;
2329 if (vcpu
->arch
.apic
) {
2330 dev
= &vcpu
->arch
.apic
->dev
;
2331 if (kvm_iodevice_in_range(dev
, addr
, len
, is_write
))
2338 static struct kvm_io_device
*vcpu_find_mmio_dev(struct kvm_vcpu
*vcpu
,
2339 gpa_t addr
, int len
,
2342 struct kvm_io_device
*dev
;
2344 dev
= vcpu_find_pervcpu_dev(vcpu
, addr
, len
, is_write
);
2346 dev
= kvm_io_bus_find_dev(&vcpu
->kvm
->mmio_bus
, addr
, len
,
2351 static int kvm_read_guest_virt(gva_t addr
, void *val
, unsigned int bytes
,
2352 struct kvm_vcpu
*vcpu
)
2355 int r
= X86EMUL_CONTINUE
;
2358 gpa_t gpa
= vcpu
->arch
.mmu
.gva_to_gpa(vcpu
, addr
);
2359 unsigned offset
= addr
& (PAGE_SIZE
-1);
2360 unsigned toread
= min(bytes
, (unsigned)PAGE_SIZE
- offset
);
2363 if (gpa
== UNMAPPED_GVA
) {
2364 r
= X86EMUL_PROPAGATE_FAULT
;
2367 ret
= kvm_read_guest(vcpu
->kvm
, gpa
, data
, toread
);
2369 r
= X86EMUL_UNHANDLEABLE
;
2381 static int kvm_write_guest_virt(gva_t addr
, void *val
, unsigned int bytes
,
2382 struct kvm_vcpu
*vcpu
)
2385 int r
= X86EMUL_CONTINUE
;
2388 gpa_t gpa
= vcpu
->arch
.mmu
.gva_to_gpa(vcpu
, addr
);
2389 unsigned offset
= addr
& (PAGE_SIZE
-1);
2390 unsigned towrite
= min(bytes
, (unsigned)PAGE_SIZE
- offset
);
2393 if (gpa
== UNMAPPED_GVA
) {
2394 r
= X86EMUL_PROPAGATE_FAULT
;
2397 ret
= kvm_write_guest(vcpu
->kvm
, gpa
, data
, towrite
);
2399 r
= X86EMUL_UNHANDLEABLE
;
2412 static int emulator_read_emulated(unsigned long addr
,
2415 struct kvm_vcpu
*vcpu
)
2417 struct kvm_io_device
*mmio_dev
;
2420 if (vcpu
->mmio_read_completed
) {
2421 memcpy(val
, vcpu
->mmio_data
, bytes
);
2422 vcpu
->mmio_read_completed
= 0;
2423 return X86EMUL_CONTINUE
;
2426 gpa
= vcpu
->arch
.mmu
.gva_to_gpa(vcpu
, addr
);
2428 /* For APIC access vmexit */
2429 if ((gpa
& PAGE_MASK
) == APIC_DEFAULT_PHYS_BASE
)
2432 if (kvm_read_guest_virt(addr
, val
, bytes
, vcpu
)
2433 == X86EMUL_CONTINUE
)
2434 return X86EMUL_CONTINUE
;
2435 if (gpa
== UNMAPPED_GVA
)
2436 return X86EMUL_PROPAGATE_FAULT
;
2440 * Is this MMIO handled locally?
2442 mutex_lock(&vcpu
->kvm
->lock
);
2443 mmio_dev
= vcpu_find_mmio_dev(vcpu
, gpa
, bytes
, 0);
2444 mutex_unlock(&vcpu
->kvm
->lock
);
2446 kvm_iodevice_read(mmio_dev
, gpa
, bytes
, val
);
2447 return X86EMUL_CONTINUE
;
2450 vcpu
->mmio_needed
= 1;
2451 vcpu
->mmio_phys_addr
= gpa
;
2452 vcpu
->mmio_size
= bytes
;
2453 vcpu
->mmio_is_write
= 0;
2455 return X86EMUL_UNHANDLEABLE
;
2458 int emulator_write_phys(struct kvm_vcpu
*vcpu
, gpa_t gpa
,
2459 const void *val
, int bytes
)
2463 ret
= kvm_write_guest(vcpu
->kvm
, gpa
, val
, bytes
);
2466 kvm_mmu_pte_write(vcpu
, gpa
, val
, bytes
, 1);
2470 static int emulator_write_emulated_onepage(unsigned long addr
,
2473 struct kvm_vcpu
*vcpu
)
2475 struct kvm_io_device
*mmio_dev
;
2478 gpa
= vcpu
->arch
.mmu
.gva_to_gpa(vcpu
, addr
);
2480 if (gpa
== UNMAPPED_GVA
) {
2481 kvm_inject_page_fault(vcpu
, addr
, 2);
2482 return X86EMUL_PROPAGATE_FAULT
;
2485 /* For APIC access vmexit */
2486 if ((gpa
& PAGE_MASK
) == APIC_DEFAULT_PHYS_BASE
)
2489 if (emulator_write_phys(vcpu
, gpa
, val
, bytes
))
2490 return X86EMUL_CONTINUE
;
2494 * Is this MMIO handled locally?
2496 mutex_lock(&vcpu
->kvm
->lock
);
2497 mmio_dev
= vcpu_find_mmio_dev(vcpu
, gpa
, bytes
, 1);
2498 mutex_unlock(&vcpu
->kvm
->lock
);
2500 kvm_iodevice_write(mmio_dev
, gpa
, bytes
, val
);
2501 return X86EMUL_CONTINUE
;
2504 vcpu
->mmio_needed
= 1;
2505 vcpu
->mmio_phys_addr
= gpa
;
2506 vcpu
->mmio_size
= bytes
;
2507 vcpu
->mmio_is_write
= 1;
2508 memcpy(vcpu
->mmio_data
, val
, bytes
);
2510 return X86EMUL_CONTINUE
;
2513 int emulator_write_emulated(unsigned long addr
,
2516 struct kvm_vcpu
*vcpu
)
2518 /* Crossing a page boundary? */
2519 if (((addr
+ bytes
- 1) ^ addr
) & PAGE_MASK
) {
2522 now
= -addr
& ~PAGE_MASK
;
2523 rc
= emulator_write_emulated_onepage(addr
, val
, now
, vcpu
);
2524 if (rc
!= X86EMUL_CONTINUE
)
2530 return emulator_write_emulated_onepage(addr
, val
, bytes
, vcpu
);
2532 EXPORT_SYMBOL_GPL(emulator_write_emulated
);
2534 static int emulator_cmpxchg_emulated(unsigned long addr
,
2538 struct kvm_vcpu
*vcpu
)
2540 static int reported
;
2544 printk(KERN_WARNING
"kvm: emulating exchange as write\n");
2546 #ifndef CONFIG_X86_64
2547 /* guests cmpxchg8b have to be emulated atomically */
2554 gpa
= vcpu
->arch
.mmu
.gva_to_gpa(vcpu
, addr
);
2556 if (gpa
== UNMAPPED_GVA
||
2557 (gpa
& PAGE_MASK
) == APIC_DEFAULT_PHYS_BASE
)
2560 if (((gpa
+ bytes
- 1) & PAGE_MASK
) != (gpa
& PAGE_MASK
))
2565 page
= gfn_to_page(vcpu
->kvm
, gpa
>> PAGE_SHIFT
);
2567 kaddr
= kmap_atomic(page
, KM_USER0
);
2568 set_64bit((u64
*)(kaddr
+ offset_in_page(gpa
)), val
);
2569 kunmap_atomic(kaddr
, KM_USER0
);
2570 kvm_release_page_dirty(page
);
2575 return emulator_write_emulated(addr
, new, bytes
, vcpu
);
2578 static unsigned long get_segment_base(struct kvm_vcpu
*vcpu
, int seg
)
2580 return kvm_x86_ops
->get_segment_base(vcpu
, seg
);
2583 int emulate_invlpg(struct kvm_vcpu
*vcpu
, gva_t address
)
2585 kvm_mmu_invlpg(vcpu
, address
);
2586 return X86EMUL_CONTINUE
;
2589 int emulate_clts(struct kvm_vcpu
*vcpu
)
2591 kvm_x86_ops
->set_cr0(vcpu
, vcpu
->arch
.cr0
& ~X86_CR0_TS
);
2592 return X86EMUL_CONTINUE
;
2595 int emulator_get_dr(struct x86_emulate_ctxt
*ctxt
, int dr
, unsigned long *dest
)
2597 struct kvm_vcpu
*vcpu
= ctxt
->vcpu
;
2601 *dest
= kvm_x86_ops
->get_dr(vcpu
, dr
);
2602 return X86EMUL_CONTINUE
;
2604 pr_unimpl(vcpu
, "%s: unexpected dr %u\n", __func__
, dr
);
2605 return X86EMUL_UNHANDLEABLE
;
2609 int emulator_set_dr(struct x86_emulate_ctxt
*ctxt
, int dr
, unsigned long value
)
2611 unsigned long mask
= (ctxt
->mode
== X86EMUL_MODE_PROT64
) ? ~0ULL : ~0U;
2614 kvm_x86_ops
->set_dr(ctxt
->vcpu
, dr
, value
& mask
, &exception
);
2616 /* FIXME: better handling */
2617 return X86EMUL_UNHANDLEABLE
;
2619 return X86EMUL_CONTINUE
;
2622 void kvm_report_emulation_failure(struct kvm_vcpu
*vcpu
, const char *context
)
2625 unsigned long rip
= kvm_rip_read(vcpu
);
2626 unsigned long rip_linear
;
2628 if (!printk_ratelimit())
2631 rip_linear
= rip
+ get_segment_base(vcpu
, VCPU_SREG_CS
);
2633 kvm_read_guest_virt(rip_linear
, (void *)opcodes
, 4, vcpu
);
2635 printk(KERN_ERR
"emulation failed (%s) rip %lx %02x %02x %02x %02x\n",
2636 context
, rip
, opcodes
[0], opcodes
[1], opcodes
[2], opcodes
[3]);
2638 EXPORT_SYMBOL_GPL(kvm_report_emulation_failure
);
2640 static struct x86_emulate_ops emulate_ops
= {
2641 .read_std
= kvm_read_guest_virt
,
2642 .read_emulated
= emulator_read_emulated
,
2643 .write_emulated
= emulator_write_emulated
,
2644 .cmpxchg_emulated
= emulator_cmpxchg_emulated
,
2647 static void cache_all_regs(struct kvm_vcpu
*vcpu
)
2649 kvm_register_read(vcpu
, VCPU_REGS_RAX
);
2650 kvm_register_read(vcpu
, VCPU_REGS_RSP
);
2651 kvm_register_read(vcpu
, VCPU_REGS_RIP
);
2652 vcpu
->arch
.regs_dirty
= ~0;
2655 int emulate_instruction(struct kvm_vcpu
*vcpu
,
2656 struct kvm_run
*run
,
2662 struct decode_cache
*c
;
2664 kvm_clear_exception_queue(vcpu
);
2665 vcpu
->arch
.mmio_fault_cr2
= cr2
;
2667 * TODO: fix x86_emulate.c to use guest_read/write_register
2668 * instead of direct ->regs accesses, can save hundred cycles
2669 * on Intel for instructions that don't read/change RSP, for
2672 cache_all_regs(vcpu
);
2674 vcpu
->mmio_is_write
= 0;
2675 vcpu
->arch
.pio
.string
= 0;
2677 if (!(emulation_type
& EMULTYPE_NO_DECODE
)) {
2679 kvm_x86_ops
->get_cs_db_l_bits(vcpu
, &cs_db
, &cs_l
);
2681 vcpu
->arch
.emulate_ctxt
.vcpu
= vcpu
;
2682 vcpu
->arch
.emulate_ctxt
.eflags
= kvm_x86_ops
->get_rflags(vcpu
);
2683 vcpu
->arch
.emulate_ctxt
.mode
=
2684 (vcpu
->arch
.emulate_ctxt
.eflags
& X86_EFLAGS_VM
)
2685 ? X86EMUL_MODE_REAL
: cs_l
2686 ? X86EMUL_MODE_PROT64
: cs_db
2687 ? X86EMUL_MODE_PROT32
: X86EMUL_MODE_PROT16
;
2689 r
= x86_decode_insn(&vcpu
->arch
.emulate_ctxt
, &emulate_ops
);
2691 /* Only allow emulation of specific instructions on #UD
2692 * (namely VMMCALL, sysenter, sysexit, syscall)*/
2693 c
= &vcpu
->arch
.emulate_ctxt
.decode
;
2694 if (emulation_type
& EMULTYPE_TRAP_UD
) {
2696 return EMULATE_FAIL
;
2698 case 0x01: /* VMMCALL */
2699 if (c
->modrm_mod
!= 3 || c
->modrm_rm
!= 1)
2700 return EMULATE_FAIL
;
2702 case 0x34: /* sysenter */
2703 case 0x35: /* sysexit */
2704 if (c
->modrm_mod
!= 0 || c
->modrm_rm
!= 0)
2705 return EMULATE_FAIL
;
2707 case 0x05: /* syscall */
2708 if (c
->modrm_mod
!= 0 || c
->modrm_rm
!= 0)
2709 return EMULATE_FAIL
;
2712 return EMULATE_FAIL
;
2715 if (!(c
->modrm_reg
== 0 || c
->modrm_reg
== 3))
2716 return EMULATE_FAIL
;
2719 ++vcpu
->stat
.insn_emulation
;
2721 ++vcpu
->stat
.insn_emulation_fail
;
2722 if (kvm_mmu_unprotect_page_virt(vcpu
, cr2
))
2723 return EMULATE_DONE
;
2724 return EMULATE_FAIL
;
2728 if (emulation_type
& EMULTYPE_SKIP
) {
2729 kvm_rip_write(vcpu
, vcpu
->arch
.emulate_ctxt
.decode
.eip
);
2730 return EMULATE_DONE
;
2733 r
= x86_emulate_insn(&vcpu
->arch
.emulate_ctxt
, &emulate_ops
);
2734 shadow_mask
= vcpu
->arch
.emulate_ctxt
.interruptibility
;
2737 kvm_x86_ops
->set_interrupt_shadow(vcpu
, shadow_mask
);
2739 if (vcpu
->arch
.pio
.string
)
2740 return EMULATE_DO_MMIO
;
2742 if ((r
|| vcpu
->mmio_is_write
) && run
) {
2743 run
->exit_reason
= KVM_EXIT_MMIO
;
2744 run
->mmio
.phys_addr
= vcpu
->mmio_phys_addr
;
2745 memcpy(run
->mmio
.data
, vcpu
->mmio_data
, 8);
2746 run
->mmio
.len
= vcpu
->mmio_size
;
2747 run
->mmio
.is_write
= vcpu
->mmio_is_write
;
2751 if (kvm_mmu_unprotect_page_virt(vcpu
, cr2
))
2752 return EMULATE_DONE
;
2753 if (!vcpu
->mmio_needed
) {
2754 kvm_report_emulation_failure(vcpu
, "mmio");
2755 return EMULATE_FAIL
;
2757 return EMULATE_DO_MMIO
;
2760 kvm_x86_ops
->set_rflags(vcpu
, vcpu
->arch
.emulate_ctxt
.eflags
);
2762 if (vcpu
->mmio_is_write
) {
2763 vcpu
->mmio_needed
= 0;
2764 return EMULATE_DO_MMIO
;
2767 return EMULATE_DONE
;
2769 EXPORT_SYMBOL_GPL(emulate_instruction
);
2771 static int pio_copy_data(struct kvm_vcpu
*vcpu
)
2773 void *p
= vcpu
->arch
.pio_data
;
2774 gva_t q
= vcpu
->arch
.pio
.guest_gva
;
2778 bytes
= vcpu
->arch
.pio
.size
* vcpu
->arch
.pio
.cur_count
;
2779 if (vcpu
->arch
.pio
.in
)
2780 ret
= kvm_write_guest_virt(q
, p
, bytes
, vcpu
);
2782 ret
= kvm_read_guest_virt(q
, p
, bytes
, vcpu
);
2786 int complete_pio(struct kvm_vcpu
*vcpu
)
2788 struct kvm_pio_request
*io
= &vcpu
->arch
.pio
;
2795 val
= kvm_register_read(vcpu
, VCPU_REGS_RAX
);
2796 memcpy(&val
, vcpu
->arch
.pio_data
, io
->size
);
2797 kvm_register_write(vcpu
, VCPU_REGS_RAX
, val
);
2801 r
= pio_copy_data(vcpu
);
2808 delta
*= io
->cur_count
;
2810 * The size of the register should really depend on
2811 * current address size.
2813 val
= kvm_register_read(vcpu
, VCPU_REGS_RCX
);
2815 kvm_register_write(vcpu
, VCPU_REGS_RCX
, val
);
2821 val
= kvm_register_read(vcpu
, VCPU_REGS_RDI
);
2823 kvm_register_write(vcpu
, VCPU_REGS_RDI
, val
);
2825 val
= kvm_register_read(vcpu
, VCPU_REGS_RSI
);
2827 kvm_register_write(vcpu
, VCPU_REGS_RSI
, val
);
2831 io
->count
-= io
->cur_count
;
2837 static void kernel_pio(struct kvm_io_device
*pio_dev
,
2838 struct kvm_vcpu
*vcpu
,
2841 /* TODO: String I/O for in kernel device */
2843 if (vcpu
->arch
.pio
.in
)
2844 kvm_iodevice_read(pio_dev
, vcpu
->arch
.pio
.port
,
2845 vcpu
->arch
.pio
.size
,
2848 kvm_iodevice_write(pio_dev
, vcpu
->arch
.pio
.port
,
2849 vcpu
->arch
.pio
.size
,
2853 static void pio_string_write(struct kvm_io_device
*pio_dev
,
2854 struct kvm_vcpu
*vcpu
)
2856 struct kvm_pio_request
*io
= &vcpu
->arch
.pio
;
2857 void *pd
= vcpu
->arch
.pio_data
;
2860 for (i
= 0; i
< io
->cur_count
; i
++) {
2861 kvm_iodevice_write(pio_dev
, io
->port
,
2868 static struct kvm_io_device
*vcpu_find_pio_dev(struct kvm_vcpu
*vcpu
,
2869 gpa_t addr
, int len
,
2872 return kvm_io_bus_find_dev(&vcpu
->kvm
->pio_bus
, addr
, len
, is_write
);
2875 int kvm_emulate_pio(struct kvm_vcpu
*vcpu
, struct kvm_run
*run
, int in
,
2876 int size
, unsigned port
)
2878 struct kvm_io_device
*pio_dev
;
2881 vcpu
->run
->exit_reason
= KVM_EXIT_IO
;
2882 vcpu
->run
->io
.direction
= in
? KVM_EXIT_IO_IN
: KVM_EXIT_IO_OUT
;
2883 vcpu
->run
->io
.size
= vcpu
->arch
.pio
.size
= size
;
2884 vcpu
->run
->io
.data_offset
= KVM_PIO_PAGE_OFFSET
* PAGE_SIZE
;
2885 vcpu
->run
->io
.count
= vcpu
->arch
.pio
.count
= vcpu
->arch
.pio
.cur_count
= 1;
2886 vcpu
->run
->io
.port
= vcpu
->arch
.pio
.port
= port
;
2887 vcpu
->arch
.pio
.in
= in
;
2888 vcpu
->arch
.pio
.string
= 0;
2889 vcpu
->arch
.pio
.down
= 0;
2890 vcpu
->arch
.pio
.rep
= 0;
2892 trace_kvm_pio(vcpu
->run
->io
.direction
== KVM_EXIT_IO_OUT
, port
,
2895 val
= kvm_register_read(vcpu
, VCPU_REGS_RAX
);
2896 memcpy(vcpu
->arch
.pio_data
, &val
, 4);
2898 mutex_lock(&vcpu
->kvm
->lock
);
2899 pio_dev
= vcpu_find_pio_dev(vcpu
, port
, size
, !in
);
2900 mutex_unlock(&vcpu
->kvm
->lock
);
2902 kernel_pio(pio_dev
, vcpu
, vcpu
->arch
.pio_data
);
2908 EXPORT_SYMBOL_GPL(kvm_emulate_pio
);
2910 int kvm_emulate_pio_string(struct kvm_vcpu
*vcpu
, struct kvm_run
*run
, int in
,
2911 int size
, unsigned long count
, int down
,
2912 gva_t address
, int rep
, unsigned port
)
2914 unsigned now
, in_page
;
2916 struct kvm_io_device
*pio_dev
;
2918 vcpu
->run
->exit_reason
= KVM_EXIT_IO
;
2919 vcpu
->run
->io
.direction
= in
? KVM_EXIT_IO_IN
: KVM_EXIT_IO_OUT
;
2920 vcpu
->run
->io
.size
= vcpu
->arch
.pio
.size
= size
;
2921 vcpu
->run
->io
.data_offset
= KVM_PIO_PAGE_OFFSET
* PAGE_SIZE
;
2922 vcpu
->run
->io
.count
= vcpu
->arch
.pio
.count
= vcpu
->arch
.pio
.cur_count
= count
;
2923 vcpu
->run
->io
.port
= vcpu
->arch
.pio
.port
= port
;
2924 vcpu
->arch
.pio
.in
= in
;
2925 vcpu
->arch
.pio
.string
= 1;
2926 vcpu
->arch
.pio
.down
= down
;
2927 vcpu
->arch
.pio
.rep
= rep
;
2929 trace_kvm_pio(vcpu
->run
->io
.direction
== KVM_EXIT_IO_OUT
, port
,
2933 kvm_x86_ops
->skip_emulated_instruction(vcpu
);
2938 in_page
= PAGE_SIZE
- offset_in_page(address
);
2940 in_page
= offset_in_page(address
) + size
;
2941 now
= min(count
, (unsigned long)in_page
/ size
);
2946 * String I/O in reverse. Yuck. Kill the guest, fix later.
2948 pr_unimpl(vcpu
, "guest string pio down\n");
2949 kvm_inject_gp(vcpu
, 0);
2952 vcpu
->run
->io
.count
= now
;
2953 vcpu
->arch
.pio
.cur_count
= now
;
2955 if (vcpu
->arch
.pio
.cur_count
== vcpu
->arch
.pio
.count
)
2956 kvm_x86_ops
->skip_emulated_instruction(vcpu
);
2958 vcpu
->arch
.pio
.guest_gva
= address
;
2960 mutex_lock(&vcpu
->kvm
->lock
);
2961 pio_dev
= vcpu_find_pio_dev(vcpu
, port
,
2962 vcpu
->arch
.pio
.cur_count
,
2963 !vcpu
->arch
.pio
.in
);
2964 mutex_unlock(&vcpu
->kvm
->lock
);
2966 if (!vcpu
->arch
.pio
.in
) {
2967 /* string PIO write */
2968 ret
= pio_copy_data(vcpu
);
2969 if (ret
== X86EMUL_PROPAGATE_FAULT
) {
2970 kvm_inject_gp(vcpu
, 0);
2973 if (ret
== 0 && pio_dev
) {
2974 pio_string_write(pio_dev
, vcpu
);
2976 if (vcpu
->arch
.pio
.count
== 0)
2980 pr_unimpl(vcpu
, "no string pio read support yet, "
2981 "port %x size %d count %ld\n",
2986 EXPORT_SYMBOL_GPL(kvm_emulate_pio_string
);
2988 static void bounce_off(void *info
)
2993 static unsigned int ref_freq
;
2994 static unsigned long tsc_khz_ref
;
2996 static int kvmclock_cpufreq_notifier(struct notifier_block
*nb
, unsigned long val
,
2999 struct cpufreq_freqs
*freq
= data
;
3001 struct kvm_vcpu
*vcpu
;
3002 int i
, send_ipi
= 0;
3005 ref_freq
= freq
->old
;
3007 if (val
== CPUFREQ_PRECHANGE
&& freq
->old
> freq
->new)
3009 if (val
== CPUFREQ_POSTCHANGE
&& freq
->old
< freq
->new)
3011 per_cpu(cpu_tsc_khz
, freq
->cpu
) = cpufreq_scale(tsc_khz_ref
, ref_freq
, freq
->new);
3013 spin_lock(&kvm_lock
);
3014 list_for_each_entry(kvm
, &vm_list
, vm_list
) {
3015 kvm_for_each_vcpu(i
, vcpu
, kvm
) {
3016 if (vcpu
->cpu
!= freq
->cpu
)
3018 if (!kvm_request_guest_time_update(vcpu
))
3020 if (vcpu
->cpu
!= smp_processor_id())
3024 spin_unlock(&kvm_lock
);
3026 if (freq
->old
< freq
->new && send_ipi
) {
3028 * We upscale the frequency. Must make the guest
3029 * doesn't see old kvmclock values while running with
3030 * the new frequency, otherwise we risk the guest sees
3031 * time go backwards.
3033 * In case we update the frequency for another cpu
3034 * (which might be in guest context) send an interrupt
3035 * to kick the cpu out of guest context. Next time
3036 * guest context is entered kvmclock will be updated,
3037 * so the guest will not see stale values.
3039 smp_call_function_single(freq
->cpu
, bounce_off
, NULL
, 1);
3044 static struct notifier_block kvmclock_cpufreq_notifier_block
= {
3045 .notifier_call
= kvmclock_cpufreq_notifier
3048 int kvm_arch_init(void *opaque
)
3051 struct kvm_x86_ops
*ops
= (struct kvm_x86_ops
*)opaque
;
3054 printk(KERN_ERR
"kvm: already loaded the other module\n");
3059 if (!ops
->cpu_has_kvm_support()) {
3060 printk(KERN_ERR
"kvm: no hardware support\n");
3064 if (ops
->disabled_by_bios()) {
3065 printk(KERN_ERR
"kvm: disabled by bios\n");
3070 r
= kvm_mmu_module_init();
3074 kvm_init_msr_list();
3077 kvm_mmu_set_nonpresent_ptes(0ull, 0ull);
3078 kvm_mmu_set_base_ptes(PT_PRESENT_MASK
);
3079 kvm_mmu_set_mask_ptes(PT_USER_MASK
, PT_ACCESSED_MASK
,
3080 PT_DIRTY_MASK
, PT64_NX_MASK
, 0);
3082 for_each_possible_cpu(cpu
)
3083 per_cpu(cpu_tsc_khz
, cpu
) = tsc_khz
;
3084 if (!boot_cpu_has(X86_FEATURE_CONSTANT_TSC
)) {
3085 tsc_khz_ref
= tsc_khz
;
3086 cpufreq_register_notifier(&kvmclock_cpufreq_notifier_block
,
3087 CPUFREQ_TRANSITION_NOTIFIER
);
3096 void kvm_arch_exit(void)
3098 if (!boot_cpu_has(X86_FEATURE_CONSTANT_TSC
))
3099 cpufreq_unregister_notifier(&kvmclock_cpufreq_notifier_block
,
3100 CPUFREQ_TRANSITION_NOTIFIER
);
3102 kvm_mmu_module_exit();
3105 int kvm_emulate_halt(struct kvm_vcpu
*vcpu
)
3107 ++vcpu
->stat
.halt_exits
;
3108 if (irqchip_in_kernel(vcpu
->kvm
)) {
3109 vcpu
->arch
.mp_state
= KVM_MP_STATE_HALTED
;
3112 vcpu
->run
->exit_reason
= KVM_EXIT_HLT
;
3116 EXPORT_SYMBOL_GPL(kvm_emulate_halt
);
3118 static inline gpa_t
hc_gpa(struct kvm_vcpu
*vcpu
, unsigned long a0
,
3121 if (is_long_mode(vcpu
))
3124 return a0
| ((gpa_t
)a1
<< 32);
3127 int kvm_emulate_hypercall(struct kvm_vcpu
*vcpu
)
3129 unsigned long nr
, a0
, a1
, a2
, a3
, ret
;
3132 nr
= kvm_register_read(vcpu
, VCPU_REGS_RAX
);
3133 a0
= kvm_register_read(vcpu
, VCPU_REGS_RBX
);
3134 a1
= kvm_register_read(vcpu
, VCPU_REGS_RCX
);
3135 a2
= kvm_register_read(vcpu
, VCPU_REGS_RDX
);
3136 a3
= kvm_register_read(vcpu
, VCPU_REGS_RSI
);
3138 trace_kvm_hypercall(nr
, a0
, a1
, a2
, a3
);
3140 if (!is_long_mode(vcpu
)) {
3149 case KVM_HC_VAPIC_POLL_IRQ
:
3153 r
= kvm_pv_mmu_op(vcpu
, a0
, hc_gpa(vcpu
, a1
, a2
), &ret
);
3159 kvm_register_write(vcpu
, VCPU_REGS_RAX
, ret
);
3160 ++vcpu
->stat
.hypercalls
;
3163 EXPORT_SYMBOL_GPL(kvm_emulate_hypercall
);
3165 int kvm_fix_hypercall(struct kvm_vcpu
*vcpu
)
3167 char instruction
[3];
3169 unsigned long rip
= kvm_rip_read(vcpu
);
3173 * Blow out the MMU to ensure that no other VCPU has an active mapping
3174 * to ensure that the updated hypercall appears atomically across all
3177 kvm_mmu_zap_all(vcpu
->kvm
);
3179 kvm_x86_ops
->patch_hypercall(vcpu
, instruction
);
3180 if (emulator_write_emulated(rip
, instruction
, 3, vcpu
)
3181 != X86EMUL_CONTINUE
)
3187 static u64
mk_cr_64(u64 curr_cr
, u32 new_val
)
3189 return (curr_cr
& ~((1ULL << 32) - 1)) | new_val
;
3192 void realmode_lgdt(struct kvm_vcpu
*vcpu
, u16 limit
, unsigned long base
)
3194 struct descriptor_table dt
= { limit
, base
};
3196 kvm_x86_ops
->set_gdt(vcpu
, &dt
);
3199 void realmode_lidt(struct kvm_vcpu
*vcpu
, u16 limit
, unsigned long base
)
3201 struct descriptor_table dt
= { limit
, base
};
3203 kvm_x86_ops
->set_idt(vcpu
, &dt
);
3206 void realmode_lmsw(struct kvm_vcpu
*vcpu
, unsigned long msw
,
3207 unsigned long *rflags
)
3209 kvm_lmsw(vcpu
, msw
);
3210 *rflags
= kvm_x86_ops
->get_rflags(vcpu
);
3213 unsigned long realmode_get_cr(struct kvm_vcpu
*vcpu
, int cr
)
3215 unsigned long value
;
3217 kvm_x86_ops
->decache_cr4_guest_bits(vcpu
);
3220 value
= vcpu
->arch
.cr0
;
3223 value
= vcpu
->arch
.cr2
;
3226 value
= vcpu
->arch
.cr3
;
3229 value
= vcpu
->arch
.cr4
;
3232 value
= kvm_get_cr8(vcpu
);
3235 vcpu_printf(vcpu
, "%s: unexpected cr %u\n", __func__
, cr
);
3242 void realmode_set_cr(struct kvm_vcpu
*vcpu
, int cr
, unsigned long val
,
3243 unsigned long *rflags
)
3247 kvm_set_cr0(vcpu
, mk_cr_64(vcpu
->arch
.cr0
, val
));
3248 *rflags
= kvm_x86_ops
->get_rflags(vcpu
);
3251 vcpu
->arch
.cr2
= val
;
3254 kvm_set_cr3(vcpu
, val
);
3257 kvm_set_cr4(vcpu
, mk_cr_64(vcpu
->arch
.cr4
, val
));
3260 kvm_set_cr8(vcpu
, val
& 0xfUL
);
3263 vcpu_printf(vcpu
, "%s: unexpected cr %u\n", __func__
, cr
);
3267 static int move_to_next_stateful_cpuid_entry(struct kvm_vcpu
*vcpu
, int i
)
3269 struct kvm_cpuid_entry2
*e
= &vcpu
->arch
.cpuid_entries
[i
];
3270 int j
, nent
= vcpu
->arch
.cpuid_nent
;
3272 e
->flags
&= ~KVM_CPUID_FLAG_STATE_READ_NEXT
;
3273 /* when no next entry is found, the current entry[i] is reselected */
3274 for (j
= i
+ 1; ; j
= (j
+ 1) % nent
) {
3275 struct kvm_cpuid_entry2
*ej
= &vcpu
->arch
.cpuid_entries
[j
];
3276 if (ej
->function
== e
->function
) {
3277 ej
->flags
|= KVM_CPUID_FLAG_STATE_READ_NEXT
;
3281 return 0; /* silence gcc, even though control never reaches here */
3284 /* find an entry with matching function, matching index (if needed), and that
3285 * should be read next (if it's stateful) */
3286 static int is_matching_cpuid_entry(struct kvm_cpuid_entry2
*e
,
3287 u32 function
, u32 index
)
3289 if (e
->function
!= function
)
3291 if ((e
->flags
& KVM_CPUID_FLAG_SIGNIFCANT_INDEX
) && e
->index
!= index
)
3293 if ((e
->flags
& KVM_CPUID_FLAG_STATEFUL_FUNC
) &&
3294 !(e
->flags
& KVM_CPUID_FLAG_STATE_READ_NEXT
))
3299 struct kvm_cpuid_entry2
*kvm_find_cpuid_entry(struct kvm_vcpu
*vcpu
,
3300 u32 function
, u32 index
)
3303 struct kvm_cpuid_entry2
*best
= NULL
;
3305 for (i
= 0; i
< vcpu
->arch
.cpuid_nent
; ++i
) {
3306 struct kvm_cpuid_entry2
*e
;
3308 e
= &vcpu
->arch
.cpuid_entries
[i
];
3309 if (is_matching_cpuid_entry(e
, function
, index
)) {
3310 if (e
->flags
& KVM_CPUID_FLAG_STATEFUL_FUNC
)
3311 move_to_next_stateful_cpuid_entry(vcpu
, i
);
3316 * Both basic or both extended?
3318 if (((e
->function
^ function
) & 0x80000000) == 0)
3319 if (!best
|| e
->function
> best
->function
)
3325 int cpuid_maxphyaddr(struct kvm_vcpu
*vcpu
)
3327 struct kvm_cpuid_entry2
*best
;
3329 best
= kvm_find_cpuid_entry(vcpu
, 0x80000008, 0);
3331 return best
->eax
& 0xff;
3335 void kvm_emulate_cpuid(struct kvm_vcpu
*vcpu
)
3337 u32 function
, index
;
3338 struct kvm_cpuid_entry2
*best
;
3340 function
= kvm_register_read(vcpu
, VCPU_REGS_RAX
);
3341 index
= kvm_register_read(vcpu
, VCPU_REGS_RCX
);
3342 kvm_register_write(vcpu
, VCPU_REGS_RAX
, 0);
3343 kvm_register_write(vcpu
, VCPU_REGS_RBX
, 0);
3344 kvm_register_write(vcpu
, VCPU_REGS_RCX
, 0);
3345 kvm_register_write(vcpu
, VCPU_REGS_RDX
, 0);
3346 best
= kvm_find_cpuid_entry(vcpu
, function
, index
);
3348 kvm_register_write(vcpu
, VCPU_REGS_RAX
, best
->eax
);
3349 kvm_register_write(vcpu
, VCPU_REGS_RBX
, best
->ebx
);
3350 kvm_register_write(vcpu
, VCPU_REGS_RCX
, best
->ecx
);
3351 kvm_register_write(vcpu
, VCPU_REGS_RDX
, best
->edx
);
3353 kvm_x86_ops
->skip_emulated_instruction(vcpu
);
3354 trace_kvm_cpuid(function
,
3355 kvm_register_read(vcpu
, VCPU_REGS_RAX
),
3356 kvm_register_read(vcpu
, VCPU_REGS_RBX
),
3357 kvm_register_read(vcpu
, VCPU_REGS_RCX
),
3358 kvm_register_read(vcpu
, VCPU_REGS_RDX
));
3360 EXPORT_SYMBOL_GPL(kvm_emulate_cpuid
);
3363 * Check if userspace requested an interrupt window, and that the
3364 * interrupt window is open.
3366 * No need to exit to userspace if we already have an interrupt queued.
3368 static int dm_request_for_irq_injection(struct kvm_vcpu
*vcpu
,
3369 struct kvm_run
*kvm_run
)
3371 return (!irqchip_in_kernel(vcpu
->kvm
) && !kvm_cpu_has_interrupt(vcpu
) &&
3372 kvm_run
->request_interrupt_window
&&
3373 kvm_arch_interrupt_allowed(vcpu
));
3376 static void post_kvm_run_save(struct kvm_vcpu
*vcpu
,
3377 struct kvm_run
*kvm_run
)
3379 kvm_run
->if_flag
= (kvm_x86_ops
->get_rflags(vcpu
) & X86_EFLAGS_IF
) != 0;
3380 kvm_run
->cr8
= kvm_get_cr8(vcpu
);
3381 kvm_run
->apic_base
= kvm_get_apic_base(vcpu
);
3382 if (irqchip_in_kernel(vcpu
->kvm
))
3383 kvm_run
->ready_for_interrupt_injection
= 1;
3385 kvm_run
->ready_for_interrupt_injection
=
3386 kvm_arch_interrupt_allowed(vcpu
) &&
3387 !kvm_cpu_has_interrupt(vcpu
) &&
3388 !kvm_event_needs_reinjection(vcpu
);
3391 static void vapic_enter(struct kvm_vcpu
*vcpu
)
3393 struct kvm_lapic
*apic
= vcpu
->arch
.apic
;
3396 if (!apic
|| !apic
->vapic_addr
)
3399 page
= gfn_to_page(vcpu
->kvm
, apic
->vapic_addr
>> PAGE_SHIFT
);
3401 vcpu
->arch
.apic
->vapic_page
= page
;
3404 static void vapic_exit(struct kvm_vcpu
*vcpu
)
3406 struct kvm_lapic
*apic
= vcpu
->arch
.apic
;
3408 if (!apic
|| !apic
->vapic_addr
)
3411 down_read(&vcpu
->kvm
->slots_lock
);
3412 kvm_release_page_dirty(apic
->vapic_page
);
3413 mark_page_dirty(vcpu
->kvm
, apic
->vapic_addr
>> PAGE_SHIFT
);
3414 up_read(&vcpu
->kvm
->slots_lock
);
3417 static void update_cr8_intercept(struct kvm_vcpu
*vcpu
)
3421 if (!kvm_x86_ops
->update_cr8_intercept
)
3424 if (!vcpu
->arch
.apic
->vapic_addr
)
3425 max_irr
= kvm_lapic_find_highest_irr(vcpu
);
3432 tpr
= kvm_lapic_get_cr8(vcpu
);
3434 kvm_x86_ops
->update_cr8_intercept(vcpu
, tpr
, max_irr
);
3437 static void inject_pending_irq(struct kvm_vcpu
*vcpu
, struct kvm_run
*kvm_run
)
3439 /* try to reinject previous events if any */
3440 if (vcpu
->arch
.nmi_injected
) {
3441 kvm_x86_ops
->set_nmi(vcpu
);
3445 if (vcpu
->arch
.interrupt
.pending
) {
3446 kvm_x86_ops
->set_irq(vcpu
);
3450 /* try to inject new event if pending */
3451 if (vcpu
->arch
.nmi_pending
) {
3452 if (kvm_x86_ops
->nmi_allowed(vcpu
)) {
3453 vcpu
->arch
.nmi_pending
= false;
3454 vcpu
->arch
.nmi_injected
= true;
3455 kvm_x86_ops
->set_nmi(vcpu
);
3457 } else if (kvm_cpu_has_interrupt(vcpu
)) {
3458 if (kvm_x86_ops
->interrupt_allowed(vcpu
)) {
3459 kvm_queue_interrupt(vcpu
, kvm_cpu_get_interrupt(vcpu
),
3461 kvm_x86_ops
->set_irq(vcpu
);
3466 static int vcpu_enter_guest(struct kvm_vcpu
*vcpu
, struct kvm_run
*kvm_run
)
3469 bool req_int_win
= !irqchip_in_kernel(vcpu
->kvm
) &&
3470 kvm_run
->request_interrupt_window
;
3473 if (test_and_clear_bit(KVM_REQ_MMU_RELOAD
, &vcpu
->requests
))
3474 kvm_mmu_unload(vcpu
);
3476 r
= kvm_mmu_reload(vcpu
);
3480 if (vcpu
->requests
) {
3481 if (test_and_clear_bit(KVM_REQ_MIGRATE_TIMER
, &vcpu
->requests
))
3482 __kvm_migrate_timers(vcpu
);
3483 if (test_and_clear_bit(KVM_REQ_KVMCLOCK_UPDATE
, &vcpu
->requests
))
3484 kvm_write_guest_time(vcpu
);
3485 if (test_and_clear_bit(KVM_REQ_MMU_SYNC
, &vcpu
->requests
))
3486 kvm_mmu_sync_roots(vcpu
);
3487 if (test_and_clear_bit(KVM_REQ_TLB_FLUSH
, &vcpu
->requests
))
3488 kvm_x86_ops
->tlb_flush(vcpu
);
3489 if (test_and_clear_bit(KVM_REQ_REPORT_TPR_ACCESS
,
3491 kvm_run
->exit_reason
= KVM_EXIT_TPR_ACCESS
;
3495 if (test_and_clear_bit(KVM_REQ_TRIPLE_FAULT
, &vcpu
->requests
)) {
3496 kvm_run
->exit_reason
= KVM_EXIT_SHUTDOWN
;
3504 kvm_x86_ops
->prepare_guest_switch(vcpu
);
3505 kvm_load_guest_fpu(vcpu
);
3507 local_irq_disable();
3509 clear_bit(KVM_REQ_KICK
, &vcpu
->requests
);
3510 smp_mb__after_clear_bit();
3512 if (vcpu
->requests
|| need_resched() || signal_pending(current
)) {
3519 if (vcpu
->arch
.exception
.pending
)
3520 __queue_exception(vcpu
);
3522 inject_pending_irq(vcpu
, kvm_run
);
3524 /* enable NMI/IRQ window open exits if needed */
3525 if (vcpu
->arch
.nmi_pending
)
3526 kvm_x86_ops
->enable_nmi_window(vcpu
);
3527 else if (kvm_cpu_has_interrupt(vcpu
) || req_int_win
)
3528 kvm_x86_ops
->enable_irq_window(vcpu
);
3530 if (kvm_lapic_enabled(vcpu
)) {
3531 update_cr8_intercept(vcpu
);
3532 kvm_lapic_sync_to_vapic(vcpu
);
3535 up_read(&vcpu
->kvm
->slots_lock
);
3539 get_debugreg(vcpu
->arch
.host_dr6
, 6);
3540 get_debugreg(vcpu
->arch
.host_dr7
, 7);
3541 if (unlikely(vcpu
->arch
.switch_db_regs
)) {
3542 get_debugreg(vcpu
->arch
.host_db
[0], 0);
3543 get_debugreg(vcpu
->arch
.host_db
[1], 1);
3544 get_debugreg(vcpu
->arch
.host_db
[2], 2);
3545 get_debugreg(vcpu
->arch
.host_db
[3], 3);
3548 set_debugreg(vcpu
->arch
.eff_db
[0], 0);
3549 set_debugreg(vcpu
->arch
.eff_db
[1], 1);
3550 set_debugreg(vcpu
->arch
.eff_db
[2], 2);
3551 set_debugreg(vcpu
->arch
.eff_db
[3], 3);
3554 trace_kvm_entry(vcpu
->vcpu_id
);
3555 kvm_x86_ops
->run(vcpu
, kvm_run
);
3557 if (unlikely(vcpu
->arch
.switch_db_regs
)) {
3559 set_debugreg(vcpu
->arch
.host_db
[0], 0);
3560 set_debugreg(vcpu
->arch
.host_db
[1], 1);
3561 set_debugreg(vcpu
->arch
.host_db
[2], 2);
3562 set_debugreg(vcpu
->arch
.host_db
[3], 3);
3564 set_debugreg(vcpu
->arch
.host_dr6
, 6);
3565 set_debugreg(vcpu
->arch
.host_dr7
, 7);
3567 set_bit(KVM_REQ_KICK
, &vcpu
->requests
);
3573 * We must have an instruction between local_irq_enable() and
3574 * kvm_guest_exit(), so the timer interrupt isn't delayed by
3575 * the interrupt shadow. The stat.exits increment will do nicely.
3576 * But we need to prevent reordering, hence this barrier():
3584 down_read(&vcpu
->kvm
->slots_lock
);
3587 * Profile KVM exit RIPs:
3589 if (unlikely(prof_on
== KVM_PROFILING
)) {
3590 unsigned long rip
= kvm_rip_read(vcpu
);
3591 profile_hit(KVM_PROFILING
, (void *)rip
);
3595 kvm_lapic_sync_from_vapic(vcpu
);
3597 r
= kvm_x86_ops
->handle_exit(kvm_run
, vcpu
);
3603 static int __vcpu_run(struct kvm_vcpu
*vcpu
, struct kvm_run
*kvm_run
)
3607 if (unlikely(vcpu
->arch
.mp_state
== KVM_MP_STATE_SIPI_RECEIVED
)) {
3608 pr_debug("vcpu %d received sipi with vector # %x\n",
3609 vcpu
->vcpu_id
, vcpu
->arch
.sipi_vector
);
3610 kvm_lapic_reset(vcpu
);
3611 r
= kvm_arch_vcpu_reset(vcpu
);
3614 vcpu
->arch
.mp_state
= KVM_MP_STATE_RUNNABLE
;
3617 down_read(&vcpu
->kvm
->slots_lock
);
3622 if (vcpu
->arch
.mp_state
== KVM_MP_STATE_RUNNABLE
)
3623 r
= vcpu_enter_guest(vcpu
, kvm_run
);
3625 up_read(&vcpu
->kvm
->slots_lock
);
3626 kvm_vcpu_block(vcpu
);
3627 down_read(&vcpu
->kvm
->slots_lock
);
3628 if (test_and_clear_bit(KVM_REQ_UNHALT
, &vcpu
->requests
))
3630 switch(vcpu
->arch
.mp_state
) {
3631 case KVM_MP_STATE_HALTED
:
3632 vcpu
->arch
.mp_state
=
3633 KVM_MP_STATE_RUNNABLE
;
3634 case KVM_MP_STATE_RUNNABLE
:
3636 case KVM_MP_STATE_SIPI_RECEIVED
:
3647 clear_bit(KVM_REQ_PENDING_TIMER
, &vcpu
->requests
);
3648 if (kvm_cpu_has_pending_timer(vcpu
))
3649 kvm_inject_pending_timer_irqs(vcpu
);
3651 if (dm_request_for_irq_injection(vcpu
, kvm_run
)) {
3653 kvm_run
->exit_reason
= KVM_EXIT_INTR
;
3654 ++vcpu
->stat
.request_irq_exits
;
3656 if (signal_pending(current
)) {
3658 kvm_run
->exit_reason
= KVM_EXIT_INTR
;
3659 ++vcpu
->stat
.signal_exits
;
3661 if (need_resched()) {
3662 up_read(&vcpu
->kvm
->slots_lock
);
3664 down_read(&vcpu
->kvm
->slots_lock
);
3668 up_read(&vcpu
->kvm
->slots_lock
);
3669 post_kvm_run_save(vcpu
, kvm_run
);
3676 int kvm_arch_vcpu_ioctl_run(struct kvm_vcpu
*vcpu
, struct kvm_run
*kvm_run
)
3683 if (vcpu
->sigset_active
)
3684 sigprocmask(SIG_SETMASK
, &vcpu
->sigset
, &sigsaved
);
3686 if (unlikely(vcpu
->arch
.mp_state
== KVM_MP_STATE_UNINITIALIZED
)) {
3687 kvm_vcpu_block(vcpu
);
3688 clear_bit(KVM_REQ_UNHALT
, &vcpu
->requests
);
3693 /* re-sync apic's tpr */
3694 if (!irqchip_in_kernel(vcpu
->kvm
))
3695 kvm_set_cr8(vcpu
, kvm_run
->cr8
);
3697 if (vcpu
->arch
.pio
.cur_count
) {
3698 r
= complete_pio(vcpu
);
3702 #if CONFIG_HAS_IOMEM
3703 if (vcpu
->mmio_needed
) {
3704 memcpy(vcpu
->mmio_data
, kvm_run
->mmio
.data
, 8);
3705 vcpu
->mmio_read_completed
= 1;
3706 vcpu
->mmio_needed
= 0;
3708 down_read(&vcpu
->kvm
->slots_lock
);
3709 r
= emulate_instruction(vcpu
, kvm_run
,
3710 vcpu
->arch
.mmio_fault_cr2
, 0,
3711 EMULTYPE_NO_DECODE
);
3712 up_read(&vcpu
->kvm
->slots_lock
);
3713 if (r
== EMULATE_DO_MMIO
) {
3715 * Read-modify-write. Back to userspace.
3722 if (kvm_run
->exit_reason
== KVM_EXIT_HYPERCALL
)
3723 kvm_register_write(vcpu
, VCPU_REGS_RAX
,
3724 kvm_run
->hypercall
.ret
);
3726 r
= __vcpu_run(vcpu
, kvm_run
);
3729 if (vcpu
->sigset_active
)
3730 sigprocmask(SIG_SETMASK
, &sigsaved
, NULL
);
3736 int kvm_arch_vcpu_ioctl_get_regs(struct kvm_vcpu
*vcpu
, struct kvm_regs
*regs
)
3740 regs
->rax
= kvm_register_read(vcpu
, VCPU_REGS_RAX
);
3741 regs
->rbx
= kvm_register_read(vcpu
, VCPU_REGS_RBX
);
3742 regs
->rcx
= kvm_register_read(vcpu
, VCPU_REGS_RCX
);
3743 regs
->rdx
= kvm_register_read(vcpu
, VCPU_REGS_RDX
);
3744 regs
->rsi
= kvm_register_read(vcpu
, VCPU_REGS_RSI
);
3745 regs
->rdi
= kvm_register_read(vcpu
, VCPU_REGS_RDI
);
3746 regs
->rsp
= kvm_register_read(vcpu
, VCPU_REGS_RSP
);
3747 regs
->rbp
= kvm_register_read(vcpu
, VCPU_REGS_RBP
);
3748 #ifdef CONFIG_X86_64
3749 regs
->r8
= kvm_register_read(vcpu
, VCPU_REGS_R8
);
3750 regs
->r9
= kvm_register_read(vcpu
, VCPU_REGS_R9
);
3751 regs
->r10
= kvm_register_read(vcpu
, VCPU_REGS_R10
);
3752 regs
->r11
= kvm_register_read(vcpu
, VCPU_REGS_R11
);
3753 regs
->r12
= kvm_register_read(vcpu
, VCPU_REGS_R12
);
3754 regs
->r13
= kvm_register_read(vcpu
, VCPU_REGS_R13
);
3755 regs
->r14
= kvm_register_read(vcpu
, VCPU_REGS_R14
);
3756 regs
->r15
= kvm_register_read(vcpu
, VCPU_REGS_R15
);
3759 regs
->rip
= kvm_rip_read(vcpu
);
3760 regs
->rflags
= kvm_x86_ops
->get_rflags(vcpu
);
3763 * Don't leak debug flags in case they were set for guest debugging
3765 if (vcpu
->guest_debug
& KVM_GUESTDBG_SINGLESTEP
)
3766 regs
->rflags
&= ~(X86_EFLAGS_TF
| X86_EFLAGS_RF
);
3773 int kvm_arch_vcpu_ioctl_set_regs(struct kvm_vcpu
*vcpu
, struct kvm_regs
*regs
)
3777 kvm_register_write(vcpu
, VCPU_REGS_RAX
, regs
->rax
);
3778 kvm_register_write(vcpu
, VCPU_REGS_RBX
, regs
->rbx
);
3779 kvm_register_write(vcpu
, VCPU_REGS_RCX
, regs
->rcx
);
3780 kvm_register_write(vcpu
, VCPU_REGS_RDX
, regs
->rdx
);
3781 kvm_register_write(vcpu
, VCPU_REGS_RSI
, regs
->rsi
);
3782 kvm_register_write(vcpu
, VCPU_REGS_RDI
, regs
->rdi
);
3783 kvm_register_write(vcpu
, VCPU_REGS_RSP
, regs
->rsp
);
3784 kvm_register_write(vcpu
, VCPU_REGS_RBP
, regs
->rbp
);
3785 #ifdef CONFIG_X86_64
3786 kvm_register_write(vcpu
, VCPU_REGS_R8
, regs
->r8
);
3787 kvm_register_write(vcpu
, VCPU_REGS_R9
, regs
->r9
);
3788 kvm_register_write(vcpu
, VCPU_REGS_R10
, regs
->r10
);
3789 kvm_register_write(vcpu
, VCPU_REGS_R11
, regs
->r11
);
3790 kvm_register_write(vcpu
, VCPU_REGS_R12
, regs
->r12
);
3791 kvm_register_write(vcpu
, VCPU_REGS_R13
, regs
->r13
);
3792 kvm_register_write(vcpu
, VCPU_REGS_R14
, regs
->r14
);
3793 kvm_register_write(vcpu
, VCPU_REGS_R15
, regs
->r15
);
3797 kvm_rip_write(vcpu
, regs
->rip
);
3798 kvm_x86_ops
->set_rflags(vcpu
, regs
->rflags
);
3801 vcpu
->arch
.exception
.pending
= false;
3808 void kvm_get_segment(struct kvm_vcpu
*vcpu
,
3809 struct kvm_segment
*var
, int seg
)
3811 kvm_x86_ops
->get_segment(vcpu
, var
, seg
);
3814 void kvm_get_cs_db_l_bits(struct kvm_vcpu
*vcpu
, int *db
, int *l
)
3816 struct kvm_segment cs
;
3818 kvm_get_segment(vcpu
, &cs
, VCPU_SREG_CS
);
3822 EXPORT_SYMBOL_GPL(kvm_get_cs_db_l_bits
);
3824 int kvm_arch_vcpu_ioctl_get_sregs(struct kvm_vcpu
*vcpu
,
3825 struct kvm_sregs
*sregs
)
3827 struct descriptor_table dt
;
3831 kvm_get_segment(vcpu
, &sregs
->cs
, VCPU_SREG_CS
);
3832 kvm_get_segment(vcpu
, &sregs
->ds
, VCPU_SREG_DS
);
3833 kvm_get_segment(vcpu
, &sregs
->es
, VCPU_SREG_ES
);
3834 kvm_get_segment(vcpu
, &sregs
->fs
, VCPU_SREG_FS
);
3835 kvm_get_segment(vcpu
, &sregs
->gs
, VCPU_SREG_GS
);
3836 kvm_get_segment(vcpu
, &sregs
->ss
, VCPU_SREG_SS
);
3838 kvm_get_segment(vcpu
, &sregs
->tr
, VCPU_SREG_TR
);
3839 kvm_get_segment(vcpu
, &sregs
->ldt
, VCPU_SREG_LDTR
);
3841 kvm_x86_ops
->get_idt(vcpu
, &dt
);
3842 sregs
->idt
.limit
= dt
.limit
;
3843 sregs
->idt
.base
= dt
.base
;
3844 kvm_x86_ops
->get_gdt(vcpu
, &dt
);
3845 sregs
->gdt
.limit
= dt
.limit
;
3846 sregs
->gdt
.base
= dt
.base
;
3848 kvm_x86_ops
->decache_cr4_guest_bits(vcpu
);
3849 sregs
->cr0
= vcpu
->arch
.cr0
;
3850 sregs
->cr2
= vcpu
->arch
.cr2
;
3851 sregs
->cr3
= vcpu
->arch
.cr3
;
3852 sregs
->cr4
= vcpu
->arch
.cr4
;
3853 sregs
->cr8
= kvm_get_cr8(vcpu
);
3854 sregs
->efer
= vcpu
->arch
.shadow_efer
;
3855 sregs
->apic_base
= kvm_get_apic_base(vcpu
);
3857 memset(sregs
->interrupt_bitmap
, 0, sizeof sregs
->interrupt_bitmap
);
3859 if (vcpu
->arch
.interrupt
.pending
&& !vcpu
->arch
.interrupt
.soft
)
3860 set_bit(vcpu
->arch
.interrupt
.nr
,
3861 (unsigned long *)sregs
->interrupt_bitmap
);
3868 int kvm_arch_vcpu_ioctl_get_mpstate(struct kvm_vcpu
*vcpu
,
3869 struct kvm_mp_state
*mp_state
)
3872 mp_state
->mp_state
= vcpu
->arch
.mp_state
;
3877 int kvm_arch_vcpu_ioctl_set_mpstate(struct kvm_vcpu
*vcpu
,
3878 struct kvm_mp_state
*mp_state
)
3881 vcpu
->arch
.mp_state
= mp_state
->mp_state
;
3886 static void kvm_set_segment(struct kvm_vcpu
*vcpu
,
3887 struct kvm_segment
*var
, int seg
)
3889 kvm_x86_ops
->set_segment(vcpu
, var
, seg
);
3892 static void seg_desct_to_kvm_desct(struct desc_struct
*seg_desc
, u16 selector
,
3893 struct kvm_segment
*kvm_desct
)
3895 kvm_desct
->base
= seg_desc
->base0
;
3896 kvm_desct
->base
|= seg_desc
->base1
<< 16;
3897 kvm_desct
->base
|= seg_desc
->base2
<< 24;
3898 kvm_desct
->limit
= seg_desc
->limit0
;
3899 kvm_desct
->limit
|= seg_desc
->limit
<< 16;
3901 kvm_desct
->limit
<<= 12;
3902 kvm_desct
->limit
|= 0xfff;
3904 kvm_desct
->selector
= selector
;
3905 kvm_desct
->type
= seg_desc
->type
;
3906 kvm_desct
->present
= seg_desc
->p
;
3907 kvm_desct
->dpl
= seg_desc
->dpl
;
3908 kvm_desct
->db
= seg_desc
->d
;
3909 kvm_desct
->s
= seg_desc
->s
;
3910 kvm_desct
->l
= seg_desc
->l
;
3911 kvm_desct
->g
= seg_desc
->g
;
3912 kvm_desct
->avl
= seg_desc
->avl
;
3914 kvm_desct
->unusable
= 1;
3916 kvm_desct
->unusable
= 0;
3917 kvm_desct
->padding
= 0;
3920 static void get_segment_descriptor_dtable(struct kvm_vcpu
*vcpu
,
3922 struct descriptor_table
*dtable
)
3924 if (selector
& 1 << 2) {
3925 struct kvm_segment kvm_seg
;
3927 kvm_get_segment(vcpu
, &kvm_seg
, VCPU_SREG_LDTR
);
3929 if (kvm_seg
.unusable
)
3932 dtable
->limit
= kvm_seg
.limit
;
3933 dtable
->base
= kvm_seg
.base
;
3936 kvm_x86_ops
->get_gdt(vcpu
, dtable
);
3939 /* allowed just for 8 bytes segments */
3940 static int load_guest_segment_descriptor(struct kvm_vcpu
*vcpu
, u16 selector
,
3941 struct desc_struct
*seg_desc
)
3944 struct descriptor_table dtable
;
3945 u16 index
= selector
>> 3;
3947 get_segment_descriptor_dtable(vcpu
, selector
, &dtable
);
3949 if (dtable
.limit
< index
* 8 + 7) {
3950 kvm_queue_exception_e(vcpu
, GP_VECTOR
, selector
& 0xfffc);
3953 gpa
= vcpu
->arch
.mmu
.gva_to_gpa(vcpu
, dtable
.base
);
3955 return kvm_read_guest(vcpu
->kvm
, gpa
, seg_desc
, 8);
3958 /* allowed just for 8 bytes segments */
3959 static int save_guest_segment_descriptor(struct kvm_vcpu
*vcpu
, u16 selector
,
3960 struct desc_struct
*seg_desc
)
3963 struct descriptor_table dtable
;
3964 u16 index
= selector
>> 3;
3966 get_segment_descriptor_dtable(vcpu
, selector
, &dtable
);
3968 if (dtable
.limit
< index
* 8 + 7)
3970 gpa
= vcpu
->arch
.mmu
.gva_to_gpa(vcpu
, dtable
.base
);
3972 return kvm_write_guest(vcpu
->kvm
, gpa
, seg_desc
, 8);
3975 static u32
get_tss_base_addr(struct kvm_vcpu
*vcpu
,
3976 struct desc_struct
*seg_desc
)
3980 base_addr
= seg_desc
->base0
;
3981 base_addr
|= (seg_desc
->base1
<< 16);
3982 base_addr
|= (seg_desc
->base2
<< 24);
3984 return vcpu
->arch
.mmu
.gva_to_gpa(vcpu
, base_addr
);
3987 static u16
get_segment_selector(struct kvm_vcpu
*vcpu
, int seg
)
3989 struct kvm_segment kvm_seg
;
3991 kvm_get_segment(vcpu
, &kvm_seg
, seg
);
3992 return kvm_seg
.selector
;
3995 static int load_segment_descriptor_to_kvm_desct(struct kvm_vcpu
*vcpu
,
3997 struct kvm_segment
*kvm_seg
)
3999 struct desc_struct seg_desc
;
4001 if (load_guest_segment_descriptor(vcpu
, selector
, &seg_desc
))
4003 seg_desct_to_kvm_desct(&seg_desc
, selector
, kvm_seg
);
4007 static int kvm_load_realmode_segment(struct kvm_vcpu
*vcpu
, u16 selector
, int seg
)
4009 struct kvm_segment segvar
= {
4010 .base
= selector
<< 4,
4012 .selector
= selector
,
4023 kvm_x86_ops
->set_segment(vcpu
, &segvar
, seg
);
4027 int kvm_load_segment_descriptor(struct kvm_vcpu
*vcpu
, u16 selector
,
4028 int type_bits
, int seg
)
4030 struct kvm_segment kvm_seg
;
4032 if (!(vcpu
->arch
.cr0
& X86_CR0_PE
))
4033 return kvm_load_realmode_segment(vcpu
, selector
, seg
);
4034 if (load_segment_descriptor_to_kvm_desct(vcpu
, selector
, &kvm_seg
))
4036 kvm_seg
.type
|= type_bits
;
4038 if (seg
!= VCPU_SREG_SS
&& seg
!= VCPU_SREG_CS
&&
4039 seg
!= VCPU_SREG_LDTR
)
4041 kvm_seg
.unusable
= 1;
4043 kvm_set_segment(vcpu
, &kvm_seg
, seg
);
4047 static void save_state_to_tss32(struct kvm_vcpu
*vcpu
,
4048 struct tss_segment_32
*tss
)
4050 tss
->cr3
= vcpu
->arch
.cr3
;
4051 tss
->eip
= kvm_rip_read(vcpu
);
4052 tss
->eflags
= kvm_x86_ops
->get_rflags(vcpu
);
4053 tss
->eax
= kvm_register_read(vcpu
, VCPU_REGS_RAX
);
4054 tss
->ecx
= kvm_register_read(vcpu
, VCPU_REGS_RCX
);
4055 tss
->edx
= kvm_register_read(vcpu
, VCPU_REGS_RDX
);
4056 tss
->ebx
= kvm_register_read(vcpu
, VCPU_REGS_RBX
);
4057 tss
->esp
= kvm_register_read(vcpu
, VCPU_REGS_RSP
);
4058 tss
->ebp
= kvm_register_read(vcpu
, VCPU_REGS_RBP
);
4059 tss
->esi
= kvm_register_read(vcpu
, VCPU_REGS_RSI
);
4060 tss
->edi
= kvm_register_read(vcpu
, VCPU_REGS_RDI
);
4061 tss
->es
= get_segment_selector(vcpu
, VCPU_SREG_ES
);
4062 tss
->cs
= get_segment_selector(vcpu
, VCPU_SREG_CS
);
4063 tss
->ss
= get_segment_selector(vcpu
, VCPU_SREG_SS
);
4064 tss
->ds
= get_segment_selector(vcpu
, VCPU_SREG_DS
);
4065 tss
->fs
= get_segment_selector(vcpu
, VCPU_SREG_FS
);
4066 tss
->gs
= get_segment_selector(vcpu
, VCPU_SREG_GS
);
4067 tss
->ldt_selector
= get_segment_selector(vcpu
, VCPU_SREG_LDTR
);
4070 static int load_state_from_tss32(struct kvm_vcpu
*vcpu
,
4071 struct tss_segment_32
*tss
)
4073 kvm_set_cr3(vcpu
, tss
->cr3
);
4075 kvm_rip_write(vcpu
, tss
->eip
);
4076 kvm_x86_ops
->set_rflags(vcpu
, tss
->eflags
| 2);
4078 kvm_register_write(vcpu
, VCPU_REGS_RAX
, tss
->eax
);
4079 kvm_register_write(vcpu
, VCPU_REGS_RCX
, tss
->ecx
);
4080 kvm_register_write(vcpu
, VCPU_REGS_RDX
, tss
->edx
);
4081 kvm_register_write(vcpu
, VCPU_REGS_RBX
, tss
->ebx
);
4082 kvm_register_write(vcpu
, VCPU_REGS_RSP
, tss
->esp
);
4083 kvm_register_write(vcpu
, VCPU_REGS_RBP
, tss
->ebp
);
4084 kvm_register_write(vcpu
, VCPU_REGS_RSI
, tss
->esi
);
4085 kvm_register_write(vcpu
, VCPU_REGS_RDI
, tss
->edi
);
4087 if (kvm_load_segment_descriptor(vcpu
, tss
->ldt_selector
, 0, VCPU_SREG_LDTR
))
4090 if (kvm_load_segment_descriptor(vcpu
, tss
->es
, 1, VCPU_SREG_ES
))
4093 if (kvm_load_segment_descriptor(vcpu
, tss
->cs
, 9, VCPU_SREG_CS
))
4096 if (kvm_load_segment_descriptor(vcpu
, tss
->ss
, 1, VCPU_SREG_SS
))
4099 if (kvm_load_segment_descriptor(vcpu
, tss
->ds
, 1, VCPU_SREG_DS
))
4102 if (kvm_load_segment_descriptor(vcpu
, tss
->fs
, 1, VCPU_SREG_FS
))
4105 if (kvm_load_segment_descriptor(vcpu
, tss
->gs
, 1, VCPU_SREG_GS
))
4110 static void save_state_to_tss16(struct kvm_vcpu
*vcpu
,
4111 struct tss_segment_16
*tss
)
4113 tss
->ip
= kvm_rip_read(vcpu
);
4114 tss
->flag
= kvm_x86_ops
->get_rflags(vcpu
);
4115 tss
->ax
= kvm_register_read(vcpu
, VCPU_REGS_RAX
);
4116 tss
->cx
= kvm_register_read(vcpu
, VCPU_REGS_RCX
);
4117 tss
->dx
= kvm_register_read(vcpu
, VCPU_REGS_RDX
);
4118 tss
->bx
= kvm_register_read(vcpu
, VCPU_REGS_RBX
);
4119 tss
->sp
= kvm_register_read(vcpu
, VCPU_REGS_RSP
);
4120 tss
->bp
= kvm_register_read(vcpu
, VCPU_REGS_RBP
);
4121 tss
->si
= kvm_register_read(vcpu
, VCPU_REGS_RSI
);
4122 tss
->di
= kvm_register_read(vcpu
, VCPU_REGS_RDI
);
4124 tss
->es
= get_segment_selector(vcpu
, VCPU_SREG_ES
);
4125 tss
->cs
= get_segment_selector(vcpu
, VCPU_SREG_CS
);
4126 tss
->ss
= get_segment_selector(vcpu
, VCPU_SREG_SS
);
4127 tss
->ds
= get_segment_selector(vcpu
, VCPU_SREG_DS
);
4128 tss
->ldt
= get_segment_selector(vcpu
, VCPU_SREG_LDTR
);
4129 tss
->prev_task_link
= get_segment_selector(vcpu
, VCPU_SREG_TR
);
4132 static int load_state_from_tss16(struct kvm_vcpu
*vcpu
,
4133 struct tss_segment_16
*tss
)
4135 kvm_rip_write(vcpu
, tss
->ip
);
4136 kvm_x86_ops
->set_rflags(vcpu
, tss
->flag
| 2);
4137 kvm_register_write(vcpu
, VCPU_REGS_RAX
, tss
->ax
);
4138 kvm_register_write(vcpu
, VCPU_REGS_RCX
, tss
->cx
);
4139 kvm_register_write(vcpu
, VCPU_REGS_RDX
, tss
->dx
);
4140 kvm_register_write(vcpu
, VCPU_REGS_RBX
, tss
->bx
);
4141 kvm_register_write(vcpu
, VCPU_REGS_RSP
, tss
->sp
);
4142 kvm_register_write(vcpu
, VCPU_REGS_RBP
, tss
->bp
);
4143 kvm_register_write(vcpu
, VCPU_REGS_RSI
, tss
->si
);
4144 kvm_register_write(vcpu
, VCPU_REGS_RDI
, tss
->di
);
4146 if (kvm_load_segment_descriptor(vcpu
, tss
->ldt
, 0, VCPU_SREG_LDTR
))
4149 if (kvm_load_segment_descriptor(vcpu
, tss
->es
, 1, VCPU_SREG_ES
))
4152 if (kvm_load_segment_descriptor(vcpu
, tss
->cs
, 9, VCPU_SREG_CS
))
4155 if (kvm_load_segment_descriptor(vcpu
, tss
->ss
, 1, VCPU_SREG_SS
))
4158 if (kvm_load_segment_descriptor(vcpu
, tss
->ds
, 1, VCPU_SREG_DS
))
4163 static int kvm_task_switch_16(struct kvm_vcpu
*vcpu
, u16 tss_selector
,
4164 u16 old_tss_sel
, u32 old_tss_base
,
4165 struct desc_struct
*nseg_desc
)
4167 struct tss_segment_16 tss_segment_16
;
4170 if (kvm_read_guest(vcpu
->kvm
, old_tss_base
, &tss_segment_16
,
4171 sizeof tss_segment_16
))
4174 save_state_to_tss16(vcpu
, &tss_segment_16
);
4176 if (kvm_write_guest(vcpu
->kvm
, old_tss_base
, &tss_segment_16
,
4177 sizeof tss_segment_16
))
4180 if (kvm_read_guest(vcpu
->kvm
, get_tss_base_addr(vcpu
, nseg_desc
),
4181 &tss_segment_16
, sizeof tss_segment_16
))
4184 if (old_tss_sel
!= 0xffff) {
4185 tss_segment_16
.prev_task_link
= old_tss_sel
;
4187 if (kvm_write_guest(vcpu
->kvm
,
4188 get_tss_base_addr(vcpu
, nseg_desc
),
4189 &tss_segment_16
.prev_task_link
,
4190 sizeof tss_segment_16
.prev_task_link
))
4194 if (load_state_from_tss16(vcpu
, &tss_segment_16
))
4202 static int kvm_task_switch_32(struct kvm_vcpu
*vcpu
, u16 tss_selector
,
4203 u16 old_tss_sel
, u32 old_tss_base
,
4204 struct desc_struct
*nseg_desc
)
4206 struct tss_segment_32 tss_segment_32
;
4209 if (kvm_read_guest(vcpu
->kvm
, old_tss_base
, &tss_segment_32
,
4210 sizeof tss_segment_32
))
4213 save_state_to_tss32(vcpu
, &tss_segment_32
);
4215 if (kvm_write_guest(vcpu
->kvm
, old_tss_base
, &tss_segment_32
,
4216 sizeof tss_segment_32
))
4219 if (kvm_read_guest(vcpu
->kvm
, get_tss_base_addr(vcpu
, nseg_desc
),
4220 &tss_segment_32
, sizeof tss_segment_32
))
4223 if (old_tss_sel
!= 0xffff) {
4224 tss_segment_32
.prev_task_link
= old_tss_sel
;
4226 if (kvm_write_guest(vcpu
->kvm
,
4227 get_tss_base_addr(vcpu
, nseg_desc
),
4228 &tss_segment_32
.prev_task_link
,
4229 sizeof tss_segment_32
.prev_task_link
))
4233 if (load_state_from_tss32(vcpu
, &tss_segment_32
))
4241 int kvm_task_switch(struct kvm_vcpu
*vcpu
, u16 tss_selector
, int reason
)
4243 struct kvm_segment tr_seg
;
4244 struct desc_struct cseg_desc
;
4245 struct desc_struct nseg_desc
;
4247 u32 old_tss_base
= get_segment_base(vcpu
, VCPU_SREG_TR
);
4248 u16 old_tss_sel
= get_segment_selector(vcpu
, VCPU_SREG_TR
);
4250 old_tss_base
= vcpu
->arch
.mmu
.gva_to_gpa(vcpu
, old_tss_base
);
4252 /* FIXME: Handle errors. Failure to read either TSS or their
4253 * descriptors should generate a pagefault.
4255 if (load_guest_segment_descriptor(vcpu
, tss_selector
, &nseg_desc
))
4258 if (load_guest_segment_descriptor(vcpu
, old_tss_sel
, &cseg_desc
))
4261 if (reason
!= TASK_SWITCH_IRET
) {
4264 cpl
= kvm_x86_ops
->get_cpl(vcpu
);
4265 if ((tss_selector
& 3) > nseg_desc
.dpl
|| cpl
> nseg_desc
.dpl
) {
4266 kvm_queue_exception_e(vcpu
, GP_VECTOR
, 0);
4271 if (!nseg_desc
.p
|| (nseg_desc
.limit0
| nseg_desc
.limit
<< 16) < 0x67) {
4272 kvm_queue_exception_e(vcpu
, TS_VECTOR
, tss_selector
& 0xfffc);
4276 if (reason
== TASK_SWITCH_IRET
|| reason
== TASK_SWITCH_JMP
) {
4277 cseg_desc
.type
&= ~(1 << 1); //clear the B flag
4278 save_guest_segment_descriptor(vcpu
, old_tss_sel
, &cseg_desc
);
4281 if (reason
== TASK_SWITCH_IRET
) {
4282 u32 eflags
= kvm_x86_ops
->get_rflags(vcpu
);
4283 kvm_x86_ops
->set_rflags(vcpu
, eflags
& ~X86_EFLAGS_NT
);
4286 /* set back link to prev task only if NT bit is set in eflags
4287 note that old_tss_sel is not used afetr this point */
4288 if (reason
!= TASK_SWITCH_CALL
&& reason
!= TASK_SWITCH_GATE
)
4289 old_tss_sel
= 0xffff;
4291 /* set back link to prev task only if NT bit is set in eflags
4292 note that old_tss_sel is not used afetr this point */
4293 if (reason
!= TASK_SWITCH_CALL
&& reason
!= TASK_SWITCH_GATE
)
4294 old_tss_sel
= 0xffff;
4296 if (nseg_desc
.type
& 8)
4297 ret
= kvm_task_switch_32(vcpu
, tss_selector
, old_tss_sel
,
4298 old_tss_base
, &nseg_desc
);
4300 ret
= kvm_task_switch_16(vcpu
, tss_selector
, old_tss_sel
,
4301 old_tss_base
, &nseg_desc
);
4303 if (reason
== TASK_SWITCH_CALL
|| reason
== TASK_SWITCH_GATE
) {
4304 u32 eflags
= kvm_x86_ops
->get_rflags(vcpu
);
4305 kvm_x86_ops
->set_rflags(vcpu
, eflags
| X86_EFLAGS_NT
);
4308 if (reason
!= TASK_SWITCH_IRET
) {
4309 nseg_desc
.type
|= (1 << 1);
4310 save_guest_segment_descriptor(vcpu
, tss_selector
,
4314 kvm_x86_ops
->set_cr0(vcpu
, vcpu
->arch
.cr0
| X86_CR0_TS
);
4315 seg_desct_to_kvm_desct(&nseg_desc
, tss_selector
, &tr_seg
);
4317 kvm_set_segment(vcpu
, &tr_seg
, VCPU_SREG_TR
);
4321 EXPORT_SYMBOL_GPL(kvm_task_switch
);
4323 int kvm_arch_vcpu_ioctl_set_sregs(struct kvm_vcpu
*vcpu
,
4324 struct kvm_sregs
*sregs
)
4326 int mmu_reset_needed
= 0;
4327 int pending_vec
, max_bits
;
4328 struct descriptor_table dt
;
4332 dt
.limit
= sregs
->idt
.limit
;
4333 dt
.base
= sregs
->idt
.base
;
4334 kvm_x86_ops
->set_idt(vcpu
, &dt
);
4335 dt
.limit
= sregs
->gdt
.limit
;
4336 dt
.base
= sregs
->gdt
.base
;
4337 kvm_x86_ops
->set_gdt(vcpu
, &dt
);
4339 vcpu
->arch
.cr2
= sregs
->cr2
;
4340 mmu_reset_needed
|= vcpu
->arch
.cr3
!= sregs
->cr3
;
4342 down_read(&vcpu
->kvm
->slots_lock
);
4343 if (gfn_to_memslot(vcpu
->kvm
, sregs
->cr3
>> PAGE_SHIFT
))
4344 vcpu
->arch
.cr3
= sregs
->cr3
;
4346 set_bit(KVM_REQ_TRIPLE_FAULT
, &vcpu
->requests
);
4347 up_read(&vcpu
->kvm
->slots_lock
);
4349 kvm_set_cr8(vcpu
, sregs
->cr8
);
4351 mmu_reset_needed
|= vcpu
->arch
.shadow_efer
!= sregs
->efer
;
4352 kvm_x86_ops
->set_efer(vcpu
, sregs
->efer
);
4353 kvm_set_apic_base(vcpu
, sregs
->apic_base
);
4355 kvm_x86_ops
->decache_cr4_guest_bits(vcpu
);
4357 mmu_reset_needed
|= vcpu
->arch
.cr0
!= sregs
->cr0
;
4358 kvm_x86_ops
->set_cr0(vcpu
, sregs
->cr0
);
4359 vcpu
->arch
.cr0
= sregs
->cr0
;
4361 mmu_reset_needed
|= vcpu
->arch
.cr4
!= sregs
->cr4
;
4362 kvm_x86_ops
->set_cr4(vcpu
, sregs
->cr4
);
4363 if (!is_long_mode(vcpu
) && is_pae(vcpu
))
4364 load_pdptrs(vcpu
, vcpu
->arch
.cr3
);
4366 if (mmu_reset_needed
)
4367 kvm_mmu_reset_context(vcpu
);
4369 max_bits
= (sizeof sregs
->interrupt_bitmap
) << 3;
4370 pending_vec
= find_first_bit(
4371 (const unsigned long *)sregs
->interrupt_bitmap
, max_bits
);
4372 if (pending_vec
< max_bits
) {
4373 kvm_queue_interrupt(vcpu
, pending_vec
, false);
4374 pr_debug("Set back pending irq %d\n", pending_vec
);
4375 if (irqchip_in_kernel(vcpu
->kvm
))
4376 kvm_pic_clear_isr_ack(vcpu
->kvm
);
4379 kvm_set_segment(vcpu
, &sregs
->cs
, VCPU_SREG_CS
);
4380 kvm_set_segment(vcpu
, &sregs
->ds
, VCPU_SREG_DS
);
4381 kvm_set_segment(vcpu
, &sregs
->es
, VCPU_SREG_ES
);
4382 kvm_set_segment(vcpu
, &sregs
->fs
, VCPU_SREG_FS
);
4383 kvm_set_segment(vcpu
, &sregs
->gs
, VCPU_SREG_GS
);
4384 kvm_set_segment(vcpu
, &sregs
->ss
, VCPU_SREG_SS
);
4386 kvm_set_segment(vcpu
, &sregs
->tr
, VCPU_SREG_TR
);
4387 kvm_set_segment(vcpu
, &sregs
->ldt
, VCPU_SREG_LDTR
);
4389 /* Older userspace won't unhalt the vcpu on reset. */
4390 if (kvm_vcpu_is_bsp(vcpu
) && kvm_rip_read(vcpu
) == 0xfff0 &&
4391 sregs
->cs
.selector
== 0xf000 && sregs
->cs
.base
== 0xffff0000 &&
4392 !(vcpu
->arch
.cr0
& X86_CR0_PE
))
4393 vcpu
->arch
.mp_state
= KVM_MP_STATE_RUNNABLE
;
4400 int kvm_arch_vcpu_ioctl_set_guest_debug(struct kvm_vcpu
*vcpu
,
4401 struct kvm_guest_debug
*dbg
)
4407 if ((dbg
->control
& (KVM_GUESTDBG_ENABLE
| KVM_GUESTDBG_USE_HW_BP
)) ==
4408 (KVM_GUESTDBG_ENABLE
| KVM_GUESTDBG_USE_HW_BP
)) {
4409 for (i
= 0; i
< KVM_NR_DB_REGS
; ++i
)
4410 vcpu
->arch
.eff_db
[i
] = dbg
->arch
.debugreg
[i
];
4411 vcpu
->arch
.switch_db_regs
=
4412 (dbg
->arch
.debugreg
[7] & DR7_BP_EN_MASK
);
4414 for (i
= 0; i
< KVM_NR_DB_REGS
; i
++)
4415 vcpu
->arch
.eff_db
[i
] = vcpu
->arch
.db
[i
];
4416 vcpu
->arch
.switch_db_regs
= (vcpu
->arch
.dr7
& DR7_BP_EN_MASK
);
4419 r
= kvm_x86_ops
->set_guest_debug(vcpu
, dbg
);
4421 if (dbg
->control
& KVM_GUESTDBG_INJECT_DB
)
4422 kvm_queue_exception(vcpu
, DB_VECTOR
);
4423 else if (dbg
->control
& KVM_GUESTDBG_INJECT_BP
)
4424 kvm_queue_exception(vcpu
, BP_VECTOR
);
4432 * fxsave fpu state. Taken from x86_64/processor.h. To be killed when
4433 * we have asm/x86/processor.h
4444 u32 st_space
[32]; /* 8*16 bytes for each FP-reg = 128 bytes */
4445 #ifdef CONFIG_X86_64
4446 u32 xmm_space
[64]; /* 16*16 bytes for each XMM-reg = 256 bytes */
4448 u32 xmm_space
[32]; /* 8*16 bytes for each XMM-reg = 128 bytes */
4453 * Translate a guest virtual address to a guest physical address.
4455 int kvm_arch_vcpu_ioctl_translate(struct kvm_vcpu
*vcpu
,
4456 struct kvm_translation
*tr
)
4458 unsigned long vaddr
= tr
->linear_address
;
4462 down_read(&vcpu
->kvm
->slots_lock
);
4463 gpa
= vcpu
->arch
.mmu
.gva_to_gpa(vcpu
, vaddr
);
4464 up_read(&vcpu
->kvm
->slots_lock
);
4465 tr
->physical_address
= gpa
;
4466 tr
->valid
= gpa
!= UNMAPPED_GVA
;
4474 int kvm_arch_vcpu_ioctl_get_fpu(struct kvm_vcpu
*vcpu
, struct kvm_fpu
*fpu
)
4476 struct fxsave
*fxsave
= (struct fxsave
*)&vcpu
->arch
.guest_fx_image
;
4480 memcpy(fpu
->fpr
, fxsave
->st_space
, 128);
4481 fpu
->fcw
= fxsave
->cwd
;
4482 fpu
->fsw
= fxsave
->swd
;
4483 fpu
->ftwx
= fxsave
->twd
;
4484 fpu
->last_opcode
= fxsave
->fop
;
4485 fpu
->last_ip
= fxsave
->rip
;
4486 fpu
->last_dp
= fxsave
->rdp
;
4487 memcpy(fpu
->xmm
, fxsave
->xmm_space
, sizeof fxsave
->xmm_space
);
4494 int kvm_arch_vcpu_ioctl_set_fpu(struct kvm_vcpu
*vcpu
, struct kvm_fpu
*fpu
)
4496 struct fxsave
*fxsave
= (struct fxsave
*)&vcpu
->arch
.guest_fx_image
;
4500 memcpy(fxsave
->st_space
, fpu
->fpr
, 128);
4501 fxsave
->cwd
= fpu
->fcw
;
4502 fxsave
->swd
= fpu
->fsw
;
4503 fxsave
->twd
= fpu
->ftwx
;
4504 fxsave
->fop
= fpu
->last_opcode
;
4505 fxsave
->rip
= fpu
->last_ip
;
4506 fxsave
->rdp
= fpu
->last_dp
;
4507 memcpy(fxsave
->xmm_space
, fpu
->xmm
, sizeof fxsave
->xmm_space
);
4514 void fx_init(struct kvm_vcpu
*vcpu
)
4516 unsigned after_mxcsr_mask
;
4519 * Touch the fpu the first time in non atomic context as if
4520 * this is the first fpu instruction the exception handler
4521 * will fire before the instruction returns and it'll have to
4522 * allocate ram with GFP_KERNEL.
4525 kvm_fx_save(&vcpu
->arch
.host_fx_image
);
4527 /* Initialize guest FPU by resetting ours and saving into guest's */
4529 kvm_fx_save(&vcpu
->arch
.host_fx_image
);
4531 kvm_fx_save(&vcpu
->arch
.guest_fx_image
);
4532 kvm_fx_restore(&vcpu
->arch
.host_fx_image
);
4535 vcpu
->arch
.cr0
|= X86_CR0_ET
;
4536 after_mxcsr_mask
= offsetof(struct i387_fxsave_struct
, st_space
);
4537 vcpu
->arch
.guest_fx_image
.mxcsr
= 0x1f80;
4538 memset((void *)&vcpu
->arch
.guest_fx_image
+ after_mxcsr_mask
,
4539 0, sizeof(struct i387_fxsave_struct
) - after_mxcsr_mask
);
4541 EXPORT_SYMBOL_GPL(fx_init
);
4543 void kvm_load_guest_fpu(struct kvm_vcpu
*vcpu
)
4545 if (!vcpu
->fpu_active
|| vcpu
->guest_fpu_loaded
)
4548 vcpu
->guest_fpu_loaded
= 1;
4549 kvm_fx_save(&vcpu
->arch
.host_fx_image
);
4550 kvm_fx_restore(&vcpu
->arch
.guest_fx_image
);
4552 EXPORT_SYMBOL_GPL(kvm_load_guest_fpu
);
4554 void kvm_put_guest_fpu(struct kvm_vcpu
*vcpu
)
4556 if (!vcpu
->guest_fpu_loaded
)
4559 vcpu
->guest_fpu_loaded
= 0;
4560 kvm_fx_save(&vcpu
->arch
.guest_fx_image
);
4561 kvm_fx_restore(&vcpu
->arch
.host_fx_image
);
4562 ++vcpu
->stat
.fpu_reload
;
4564 EXPORT_SYMBOL_GPL(kvm_put_guest_fpu
);
4566 void kvm_arch_vcpu_free(struct kvm_vcpu
*vcpu
)
4568 if (vcpu
->arch
.time_page
) {
4569 kvm_release_page_dirty(vcpu
->arch
.time_page
);
4570 vcpu
->arch
.time_page
= NULL
;
4573 kvm_x86_ops
->vcpu_free(vcpu
);
4576 struct kvm_vcpu
*kvm_arch_vcpu_create(struct kvm
*kvm
,
4579 return kvm_x86_ops
->vcpu_create(kvm
, id
);
4582 int kvm_arch_vcpu_setup(struct kvm_vcpu
*vcpu
)
4586 /* We do fxsave: this must be aligned. */
4587 BUG_ON((unsigned long)&vcpu
->arch
.host_fx_image
& 0xF);
4589 vcpu
->arch
.mtrr_state
.have_fixed
= 1;
4591 r
= kvm_arch_vcpu_reset(vcpu
);
4593 r
= kvm_mmu_setup(vcpu
);
4600 kvm_x86_ops
->vcpu_free(vcpu
);
4604 void kvm_arch_vcpu_destroy(struct kvm_vcpu
*vcpu
)
4607 kvm_mmu_unload(vcpu
);
4610 kvm_x86_ops
->vcpu_free(vcpu
);
4613 int kvm_arch_vcpu_reset(struct kvm_vcpu
*vcpu
)
4615 vcpu
->arch
.nmi_pending
= false;
4616 vcpu
->arch
.nmi_injected
= false;
4618 vcpu
->arch
.switch_db_regs
= 0;
4619 memset(vcpu
->arch
.db
, 0, sizeof(vcpu
->arch
.db
));
4620 vcpu
->arch
.dr6
= DR6_FIXED_1
;
4621 vcpu
->arch
.dr7
= DR7_FIXED_1
;
4623 return kvm_x86_ops
->vcpu_reset(vcpu
);
4626 void kvm_arch_hardware_enable(void *garbage
)
4628 kvm_x86_ops
->hardware_enable(garbage
);
4631 void kvm_arch_hardware_disable(void *garbage
)
4633 kvm_x86_ops
->hardware_disable(garbage
);
4636 int kvm_arch_hardware_setup(void)
4638 return kvm_x86_ops
->hardware_setup();
4641 void kvm_arch_hardware_unsetup(void)
4643 kvm_x86_ops
->hardware_unsetup();
4646 void kvm_arch_check_processor_compat(void *rtn
)
4648 kvm_x86_ops
->check_processor_compatibility(rtn
);
4651 int kvm_arch_vcpu_init(struct kvm_vcpu
*vcpu
)
4657 BUG_ON(vcpu
->kvm
== NULL
);
4660 vcpu
->arch
.mmu
.root_hpa
= INVALID_PAGE
;
4661 if (!irqchip_in_kernel(kvm
) || kvm_vcpu_is_bsp(vcpu
))
4662 vcpu
->arch
.mp_state
= KVM_MP_STATE_RUNNABLE
;
4664 vcpu
->arch
.mp_state
= KVM_MP_STATE_UNINITIALIZED
;
4666 page
= alloc_page(GFP_KERNEL
| __GFP_ZERO
);
4671 vcpu
->arch
.pio_data
= page_address(page
);
4673 r
= kvm_mmu_create(vcpu
);
4675 goto fail_free_pio_data
;
4677 if (irqchip_in_kernel(kvm
)) {
4678 r
= kvm_create_lapic(vcpu
);
4680 goto fail_mmu_destroy
;
4683 vcpu
->arch
.mce_banks
= kzalloc(KVM_MAX_MCE_BANKS
* sizeof(u64
) * 4,
4685 if (!vcpu
->arch
.mce_banks
) {
4687 goto fail_mmu_destroy
;
4689 vcpu
->arch
.mcg_cap
= KVM_MAX_MCE_BANKS
;
4694 kvm_mmu_destroy(vcpu
);
4696 free_page((unsigned long)vcpu
->arch
.pio_data
);
4701 void kvm_arch_vcpu_uninit(struct kvm_vcpu
*vcpu
)
4703 kvm_free_lapic(vcpu
);
4704 down_read(&vcpu
->kvm
->slots_lock
);
4705 kvm_mmu_destroy(vcpu
);
4706 up_read(&vcpu
->kvm
->slots_lock
);
4707 free_page((unsigned long)vcpu
->arch
.pio_data
);
4710 struct kvm
*kvm_arch_create_vm(void)
4712 struct kvm
*kvm
= kzalloc(sizeof(struct kvm
), GFP_KERNEL
);
4715 return ERR_PTR(-ENOMEM
);
4717 INIT_LIST_HEAD(&kvm
->arch
.active_mmu_pages
);
4718 INIT_LIST_HEAD(&kvm
->arch
.assigned_dev_head
);
4720 /* Reserve bit 0 of irq_sources_bitmap for userspace irq source */
4721 set_bit(KVM_USERSPACE_IRQ_SOURCE_ID
, &kvm
->arch
.irq_sources_bitmap
);
4723 rdtscll(kvm
->arch
.vm_init_tsc
);
4728 static void kvm_unload_vcpu_mmu(struct kvm_vcpu
*vcpu
)
4731 kvm_mmu_unload(vcpu
);
4735 static void kvm_free_vcpus(struct kvm
*kvm
)
4738 struct kvm_vcpu
*vcpu
;
4741 * Unpin any mmu pages first.
4743 kvm_for_each_vcpu(i
, vcpu
, kvm
)
4744 kvm_unload_vcpu_mmu(vcpu
);
4745 kvm_for_each_vcpu(i
, vcpu
, kvm
)
4746 kvm_arch_vcpu_free(vcpu
);
4748 mutex_lock(&kvm
->lock
);
4749 for (i
= 0; i
< atomic_read(&kvm
->online_vcpus
); i
++)
4750 kvm
->vcpus
[i
] = NULL
;
4752 atomic_set(&kvm
->online_vcpus
, 0);
4753 mutex_unlock(&kvm
->lock
);
4756 void kvm_arch_sync_events(struct kvm
*kvm
)
4758 kvm_free_all_assigned_devices(kvm
);
4761 void kvm_arch_destroy_vm(struct kvm
*kvm
)
4763 kvm_iommu_unmap_guest(kvm
);
4765 kfree(kvm
->arch
.vpic
);
4766 kfree(kvm
->arch
.vioapic
);
4767 kvm_free_vcpus(kvm
);
4768 kvm_free_physmem(kvm
);
4769 if (kvm
->arch
.apic_access_page
)
4770 put_page(kvm
->arch
.apic_access_page
);
4771 if (kvm
->arch
.ept_identity_pagetable
)
4772 put_page(kvm
->arch
.ept_identity_pagetable
);
4776 int kvm_arch_set_memory_region(struct kvm
*kvm
,
4777 struct kvm_userspace_memory_region
*mem
,
4778 struct kvm_memory_slot old
,
4781 int npages
= mem
->memory_size
>> PAGE_SHIFT
;
4782 struct kvm_memory_slot
*memslot
= &kvm
->memslots
[mem
->slot
];
4784 /*To keep backward compatibility with older userspace,
4785 *x86 needs to hanlde !user_alloc case.
4788 if (npages
&& !old
.rmap
) {
4789 unsigned long userspace_addr
;
4791 down_write(¤t
->mm
->mmap_sem
);
4792 userspace_addr
= do_mmap(NULL
, 0,
4794 PROT_READ
| PROT_WRITE
,
4795 MAP_PRIVATE
| MAP_ANONYMOUS
,
4797 up_write(¤t
->mm
->mmap_sem
);
4799 if (IS_ERR((void *)userspace_addr
))
4800 return PTR_ERR((void *)userspace_addr
);
4802 /* set userspace_addr atomically for kvm_hva_to_rmapp */
4803 spin_lock(&kvm
->mmu_lock
);
4804 memslot
->userspace_addr
= userspace_addr
;
4805 spin_unlock(&kvm
->mmu_lock
);
4807 if (!old
.user_alloc
&& old
.rmap
) {
4810 down_write(¤t
->mm
->mmap_sem
);
4811 ret
= do_munmap(current
->mm
, old
.userspace_addr
,
4812 old
.npages
* PAGE_SIZE
);
4813 up_write(¤t
->mm
->mmap_sem
);
4816 "kvm_vm_ioctl_set_memory_region: "
4817 "failed to munmap memory\n");
4822 spin_lock(&kvm
->mmu_lock
);
4823 if (!kvm
->arch
.n_requested_mmu_pages
) {
4824 unsigned int nr_mmu_pages
= kvm_mmu_calculate_mmu_pages(kvm
);
4825 kvm_mmu_change_mmu_pages(kvm
, nr_mmu_pages
);
4828 kvm_mmu_slot_remove_write_access(kvm
, mem
->slot
);
4829 spin_unlock(&kvm
->mmu_lock
);
4830 kvm_flush_remote_tlbs(kvm
);
4835 void kvm_arch_flush_shadow(struct kvm
*kvm
)
4837 kvm_mmu_zap_all(kvm
);
4838 kvm_reload_remote_mmus(kvm
);
4841 int kvm_arch_vcpu_runnable(struct kvm_vcpu
*vcpu
)
4843 return vcpu
->arch
.mp_state
== KVM_MP_STATE_RUNNABLE
4844 || vcpu
->arch
.mp_state
== KVM_MP_STATE_SIPI_RECEIVED
4845 || vcpu
->arch
.nmi_pending
;
4848 void kvm_vcpu_kick(struct kvm_vcpu
*vcpu
)
4851 int cpu
= vcpu
->cpu
;
4853 if (waitqueue_active(&vcpu
->wq
)) {
4854 wake_up_interruptible(&vcpu
->wq
);
4855 ++vcpu
->stat
.halt_wakeup
;
4859 if (cpu
!= me
&& (unsigned)cpu
< nr_cpu_ids
&& cpu_online(cpu
))
4860 if (!test_and_set_bit(KVM_REQ_KICK
, &vcpu
->requests
))
4861 smp_send_reschedule(cpu
);
4865 int kvm_arch_interrupt_allowed(struct kvm_vcpu
*vcpu
)
4867 return kvm_x86_ops
->interrupt_allowed(vcpu
);
4870 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_exit
);
4871 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_inj_virq
);
4872 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_page_fault
);
4873 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_msr
);
4874 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_cr
);