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 #include <trace/events/kvm.h>
41 #undef TRACE_INCLUDE_FILE
42 #define CREATE_TRACE_POINTS
45 #include <asm/uaccess.h>
51 #define MAX_IO_MSRS 256
52 #define CR0_RESERVED_BITS \
53 (~(unsigned long)(X86_CR0_PE | X86_CR0_MP | X86_CR0_EM | X86_CR0_TS \
54 | X86_CR0_ET | X86_CR0_NE | X86_CR0_WP | X86_CR0_AM \
55 | X86_CR0_NW | X86_CR0_CD | X86_CR0_PG))
56 #define CR4_RESERVED_BITS \
57 (~(unsigned long)(X86_CR4_VME | X86_CR4_PVI | X86_CR4_TSD | X86_CR4_DE\
58 | X86_CR4_PSE | X86_CR4_PAE | X86_CR4_MCE \
59 | X86_CR4_PGE | X86_CR4_PCE | X86_CR4_OSFXSR \
60 | X86_CR4_OSXMMEXCPT | X86_CR4_VMXE))
62 #define CR8_RESERVED_BITS (~(unsigned long)X86_CR8_TPR)
64 #define KVM_MAX_MCE_BANKS 32
65 #define KVM_MCE_CAP_SUPPORTED MCG_CTL_P
68 * - enable syscall per default because its emulated by KVM
69 * - enable LME and LMA per default on 64 bit KVM
72 static u64 __read_mostly efer_reserved_bits
= 0xfffffffffffffafeULL
;
74 static u64 __read_mostly efer_reserved_bits
= 0xfffffffffffffffeULL
;
77 #define VM_STAT(x) offsetof(struct kvm, stat.x), KVM_STAT_VM
78 #define VCPU_STAT(x) offsetof(struct kvm_vcpu, stat.x), KVM_STAT_VCPU
80 static int kvm_dev_ioctl_get_supported_cpuid(struct kvm_cpuid2
*cpuid
,
81 struct kvm_cpuid_entry2 __user
*entries
);
83 struct kvm_x86_ops
*kvm_x86_ops
;
84 EXPORT_SYMBOL_GPL(kvm_x86_ops
);
87 module_param_named(ignore_msrs
, ignore_msrs
, bool, S_IRUGO
| S_IWUSR
);
89 struct kvm_stats_debugfs_item debugfs_entries
[] = {
90 { "pf_fixed", VCPU_STAT(pf_fixed
) },
91 { "pf_guest", VCPU_STAT(pf_guest
) },
92 { "tlb_flush", VCPU_STAT(tlb_flush
) },
93 { "invlpg", VCPU_STAT(invlpg
) },
94 { "exits", VCPU_STAT(exits
) },
95 { "io_exits", VCPU_STAT(io_exits
) },
96 { "mmio_exits", VCPU_STAT(mmio_exits
) },
97 { "signal_exits", VCPU_STAT(signal_exits
) },
98 { "irq_window", VCPU_STAT(irq_window_exits
) },
99 { "nmi_window", VCPU_STAT(nmi_window_exits
) },
100 { "halt_exits", VCPU_STAT(halt_exits
) },
101 { "halt_wakeup", VCPU_STAT(halt_wakeup
) },
102 { "hypercalls", VCPU_STAT(hypercalls
) },
103 { "request_irq", VCPU_STAT(request_irq_exits
) },
104 { "irq_exits", VCPU_STAT(irq_exits
) },
105 { "host_state_reload", VCPU_STAT(host_state_reload
) },
106 { "efer_reload", VCPU_STAT(efer_reload
) },
107 { "fpu_reload", VCPU_STAT(fpu_reload
) },
108 { "insn_emulation", VCPU_STAT(insn_emulation
) },
109 { "insn_emulation_fail", VCPU_STAT(insn_emulation_fail
) },
110 { "irq_injections", VCPU_STAT(irq_injections
) },
111 { "nmi_injections", VCPU_STAT(nmi_injections
) },
112 { "mmu_shadow_zapped", VM_STAT(mmu_shadow_zapped
) },
113 { "mmu_pte_write", VM_STAT(mmu_pte_write
) },
114 { "mmu_pte_updated", VM_STAT(mmu_pte_updated
) },
115 { "mmu_pde_zapped", VM_STAT(mmu_pde_zapped
) },
116 { "mmu_flooded", VM_STAT(mmu_flooded
) },
117 { "mmu_recycled", VM_STAT(mmu_recycled
) },
118 { "mmu_cache_miss", VM_STAT(mmu_cache_miss
) },
119 { "mmu_unsync", VM_STAT(mmu_unsync
) },
120 { "remote_tlb_flush", VM_STAT(remote_tlb_flush
) },
121 { "largepages", VM_STAT(lpages
) },
125 unsigned long segment_base(u16 selector
)
127 struct descriptor_table gdt
;
128 struct desc_struct
*d
;
129 unsigned long table_base
;
135 asm("sgdt %0" : "=m"(gdt
));
136 table_base
= gdt
.base
;
138 if (selector
& 4) { /* from ldt */
141 asm("sldt %0" : "=g"(ldt_selector
));
142 table_base
= segment_base(ldt_selector
);
144 d
= (struct desc_struct
*)(table_base
+ (selector
& ~7));
145 v
= d
->base0
| ((unsigned long)d
->base1
<< 16) |
146 ((unsigned long)d
->base2
<< 24);
148 if (d
->s
== 0 && (d
->type
== 2 || d
->type
== 9 || d
->type
== 11))
149 v
|= ((unsigned long)((struct ldttss_desc64
*)d
)->base3
) << 32;
153 EXPORT_SYMBOL_GPL(segment_base
);
155 u64
kvm_get_apic_base(struct kvm_vcpu
*vcpu
)
157 if (irqchip_in_kernel(vcpu
->kvm
))
158 return vcpu
->arch
.apic_base
;
160 return vcpu
->arch
.apic_base
;
162 EXPORT_SYMBOL_GPL(kvm_get_apic_base
);
164 void kvm_set_apic_base(struct kvm_vcpu
*vcpu
, u64 data
)
166 /* TODO: reserve bits check */
167 if (irqchip_in_kernel(vcpu
->kvm
))
168 kvm_lapic_set_base(vcpu
, data
);
170 vcpu
->arch
.apic_base
= data
;
172 EXPORT_SYMBOL_GPL(kvm_set_apic_base
);
174 void kvm_queue_exception(struct kvm_vcpu
*vcpu
, unsigned nr
)
176 WARN_ON(vcpu
->arch
.exception
.pending
);
177 vcpu
->arch
.exception
.pending
= true;
178 vcpu
->arch
.exception
.has_error_code
= false;
179 vcpu
->arch
.exception
.nr
= nr
;
181 EXPORT_SYMBOL_GPL(kvm_queue_exception
);
183 void kvm_inject_page_fault(struct kvm_vcpu
*vcpu
, unsigned long addr
,
186 ++vcpu
->stat
.pf_guest
;
188 if (vcpu
->arch
.exception
.pending
) {
189 switch(vcpu
->arch
.exception
.nr
) {
191 /* triple fault -> shutdown */
192 set_bit(KVM_REQ_TRIPLE_FAULT
, &vcpu
->requests
);
195 vcpu
->arch
.exception
.nr
= DF_VECTOR
;
196 vcpu
->arch
.exception
.error_code
= 0;
199 /* replace previous exception with a new one in a hope
200 that instruction re-execution will regenerate lost
202 vcpu
->arch
.exception
.pending
= false;
206 vcpu
->arch
.cr2
= addr
;
207 kvm_queue_exception_e(vcpu
, PF_VECTOR
, error_code
);
210 void kvm_inject_nmi(struct kvm_vcpu
*vcpu
)
212 vcpu
->arch
.nmi_pending
= 1;
214 EXPORT_SYMBOL_GPL(kvm_inject_nmi
);
216 void kvm_queue_exception_e(struct kvm_vcpu
*vcpu
, unsigned nr
, u32 error_code
)
218 WARN_ON(vcpu
->arch
.exception
.pending
);
219 vcpu
->arch
.exception
.pending
= true;
220 vcpu
->arch
.exception
.has_error_code
= true;
221 vcpu
->arch
.exception
.nr
= nr
;
222 vcpu
->arch
.exception
.error_code
= error_code
;
224 EXPORT_SYMBOL_GPL(kvm_queue_exception_e
);
226 static void __queue_exception(struct kvm_vcpu
*vcpu
)
228 kvm_x86_ops
->queue_exception(vcpu
, vcpu
->arch
.exception
.nr
,
229 vcpu
->arch
.exception
.has_error_code
,
230 vcpu
->arch
.exception
.error_code
);
234 * Load the pae pdptrs. Return true is they are all valid.
236 int load_pdptrs(struct kvm_vcpu
*vcpu
, unsigned long cr3
)
238 gfn_t pdpt_gfn
= cr3
>> PAGE_SHIFT
;
239 unsigned offset
= ((cr3
& (PAGE_SIZE
-1)) >> 5) << 2;
242 u64 pdpte
[ARRAY_SIZE(vcpu
->arch
.pdptrs
)];
244 ret
= kvm_read_guest_page(vcpu
->kvm
, pdpt_gfn
, pdpte
,
245 offset
* sizeof(u64
), sizeof(pdpte
));
250 for (i
= 0; i
< ARRAY_SIZE(pdpte
); ++i
) {
251 if (is_present_gpte(pdpte
[i
]) &&
252 (pdpte
[i
] & vcpu
->arch
.mmu
.rsvd_bits_mask
[0][2])) {
259 memcpy(vcpu
->arch
.pdptrs
, pdpte
, sizeof(vcpu
->arch
.pdptrs
));
260 __set_bit(VCPU_EXREG_PDPTR
,
261 (unsigned long *)&vcpu
->arch
.regs_avail
);
262 __set_bit(VCPU_EXREG_PDPTR
,
263 (unsigned long *)&vcpu
->arch
.regs_dirty
);
268 EXPORT_SYMBOL_GPL(load_pdptrs
);
270 static bool pdptrs_changed(struct kvm_vcpu
*vcpu
)
272 u64 pdpte
[ARRAY_SIZE(vcpu
->arch
.pdptrs
)];
276 if (is_long_mode(vcpu
) || !is_pae(vcpu
))
279 if (!test_bit(VCPU_EXREG_PDPTR
,
280 (unsigned long *)&vcpu
->arch
.regs_avail
))
283 r
= kvm_read_guest(vcpu
->kvm
, vcpu
->arch
.cr3
& ~31u, pdpte
, sizeof(pdpte
));
286 changed
= memcmp(pdpte
, vcpu
->arch
.pdptrs
, sizeof(pdpte
)) != 0;
292 void kvm_set_cr0(struct kvm_vcpu
*vcpu
, unsigned long cr0
)
294 if (cr0
& CR0_RESERVED_BITS
) {
295 printk(KERN_DEBUG
"set_cr0: 0x%lx #GP, reserved bits 0x%lx\n",
296 cr0
, vcpu
->arch
.cr0
);
297 kvm_inject_gp(vcpu
, 0);
301 if ((cr0
& X86_CR0_NW
) && !(cr0
& X86_CR0_CD
)) {
302 printk(KERN_DEBUG
"set_cr0: #GP, CD == 0 && NW == 1\n");
303 kvm_inject_gp(vcpu
, 0);
307 if ((cr0
& X86_CR0_PG
) && !(cr0
& X86_CR0_PE
)) {
308 printk(KERN_DEBUG
"set_cr0: #GP, set PG flag "
309 "and a clear PE flag\n");
310 kvm_inject_gp(vcpu
, 0);
314 if (!is_paging(vcpu
) && (cr0
& X86_CR0_PG
)) {
316 if ((vcpu
->arch
.shadow_efer
& EFER_LME
)) {
320 printk(KERN_DEBUG
"set_cr0: #GP, start paging "
321 "in long mode while PAE is disabled\n");
322 kvm_inject_gp(vcpu
, 0);
325 kvm_x86_ops
->get_cs_db_l_bits(vcpu
, &cs_db
, &cs_l
);
327 printk(KERN_DEBUG
"set_cr0: #GP, start paging "
328 "in long mode while CS.L == 1\n");
329 kvm_inject_gp(vcpu
, 0);
335 if (is_pae(vcpu
) && !load_pdptrs(vcpu
, vcpu
->arch
.cr3
)) {
336 printk(KERN_DEBUG
"set_cr0: #GP, pdptrs "
338 kvm_inject_gp(vcpu
, 0);
344 kvm_x86_ops
->set_cr0(vcpu
, cr0
);
345 vcpu
->arch
.cr0
= cr0
;
347 kvm_mmu_reset_context(vcpu
);
350 EXPORT_SYMBOL_GPL(kvm_set_cr0
);
352 void kvm_lmsw(struct kvm_vcpu
*vcpu
, unsigned long msw
)
354 kvm_set_cr0(vcpu
, (vcpu
->arch
.cr0
& ~0x0ful
) | (msw
& 0x0f));
356 EXPORT_SYMBOL_GPL(kvm_lmsw
);
358 void kvm_set_cr4(struct kvm_vcpu
*vcpu
, unsigned long cr4
)
360 unsigned long old_cr4
= vcpu
->arch
.cr4
;
361 unsigned long pdptr_bits
= X86_CR4_PGE
| X86_CR4_PSE
| X86_CR4_PAE
;
363 if (cr4
& CR4_RESERVED_BITS
) {
364 printk(KERN_DEBUG
"set_cr4: #GP, reserved bits\n");
365 kvm_inject_gp(vcpu
, 0);
369 if (is_long_mode(vcpu
)) {
370 if (!(cr4
& X86_CR4_PAE
)) {
371 printk(KERN_DEBUG
"set_cr4: #GP, clearing PAE while "
373 kvm_inject_gp(vcpu
, 0);
376 } else if (is_paging(vcpu
) && (cr4
& X86_CR4_PAE
)
377 && ((cr4
^ old_cr4
) & pdptr_bits
)
378 && !load_pdptrs(vcpu
, vcpu
->arch
.cr3
)) {
379 printk(KERN_DEBUG
"set_cr4: #GP, pdptrs reserved bits\n");
380 kvm_inject_gp(vcpu
, 0);
384 if (cr4
& X86_CR4_VMXE
) {
385 printk(KERN_DEBUG
"set_cr4: #GP, setting VMXE\n");
386 kvm_inject_gp(vcpu
, 0);
389 kvm_x86_ops
->set_cr4(vcpu
, cr4
);
390 vcpu
->arch
.cr4
= cr4
;
391 vcpu
->arch
.mmu
.base_role
.cr4_pge
= (cr4
& X86_CR4_PGE
) && !tdp_enabled
;
392 kvm_mmu_reset_context(vcpu
);
394 EXPORT_SYMBOL_GPL(kvm_set_cr4
);
396 void kvm_set_cr3(struct kvm_vcpu
*vcpu
, unsigned long cr3
)
398 if (cr3
== vcpu
->arch
.cr3
&& !pdptrs_changed(vcpu
)) {
399 kvm_mmu_sync_roots(vcpu
);
400 kvm_mmu_flush_tlb(vcpu
);
404 if (is_long_mode(vcpu
)) {
405 if (cr3
& CR3_L_MODE_RESERVED_BITS
) {
406 printk(KERN_DEBUG
"set_cr3: #GP, reserved bits\n");
407 kvm_inject_gp(vcpu
, 0);
412 if (cr3
& CR3_PAE_RESERVED_BITS
) {
414 "set_cr3: #GP, reserved bits\n");
415 kvm_inject_gp(vcpu
, 0);
418 if (is_paging(vcpu
) && !load_pdptrs(vcpu
, cr3
)) {
419 printk(KERN_DEBUG
"set_cr3: #GP, pdptrs "
421 kvm_inject_gp(vcpu
, 0);
426 * We don't check reserved bits in nonpae mode, because
427 * this isn't enforced, and VMware depends on this.
432 * Does the new cr3 value map to physical memory? (Note, we
433 * catch an invalid cr3 even in real-mode, because it would
434 * cause trouble later on when we turn on paging anyway.)
436 * A real CPU would silently accept an invalid cr3 and would
437 * attempt to use it - with largely undefined (and often hard
438 * to debug) behavior on the guest side.
440 if (unlikely(!gfn_to_memslot(vcpu
->kvm
, cr3
>> PAGE_SHIFT
)))
441 kvm_inject_gp(vcpu
, 0);
443 vcpu
->arch
.cr3
= cr3
;
444 vcpu
->arch
.mmu
.new_cr3(vcpu
);
447 EXPORT_SYMBOL_GPL(kvm_set_cr3
);
449 void kvm_set_cr8(struct kvm_vcpu
*vcpu
, unsigned long cr8
)
451 if (cr8
& CR8_RESERVED_BITS
) {
452 printk(KERN_DEBUG
"set_cr8: #GP, reserved bits 0x%lx\n", cr8
);
453 kvm_inject_gp(vcpu
, 0);
456 if (irqchip_in_kernel(vcpu
->kvm
))
457 kvm_lapic_set_tpr(vcpu
, cr8
);
459 vcpu
->arch
.cr8
= cr8
;
461 EXPORT_SYMBOL_GPL(kvm_set_cr8
);
463 unsigned long kvm_get_cr8(struct kvm_vcpu
*vcpu
)
465 if (irqchip_in_kernel(vcpu
->kvm
))
466 return kvm_lapic_get_cr8(vcpu
);
468 return vcpu
->arch
.cr8
;
470 EXPORT_SYMBOL_GPL(kvm_get_cr8
);
472 static inline u32
bit(int bitno
)
474 return 1 << (bitno
& 31);
478 * List of msr numbers which we expose to userspace through KVM_GET_MSRS
479 * and KVM_SET_MSRS, and KVM_GET_MSR_INDEX_LIST.
481 * This list is modified at module load time to reflect the
482 * capabilities of the host cpu.
484 static u32 msrs_to_save
[] = {
485 MSR_IA32_SYSENTER_CS
, MSR_IA32_SYSENTER_ESP
, MSR_IA32_SYSENTER_EIP
,
488 MSR_CSTAR
, MSR_KERNEL_GS_BASE
, MSR_SYSCALL_MASK
, MSR_LSTAR
,
490 MSR_IA32_TSC
, MSR_KVM_SYSTEM_TIME
, MSR_KVM_WALL_CLOCK
,
491 MSR_IA32_PERF_STATUS
, MSR_IA32_CR_PAT
, MSR_VM_HSAVE_PA
494 static unsigned num_msrs_to_save
;
496 static u32 emulated_msrs
[] = {
497 MSR_IA32_MISC_ENABLE
,
500 static void set_efer(struct kvm_vcpu
*vcpu
, u64 efer
)
502 if (efer
& efer_reserved_bits
) {
503 printk(KERN_DEBUG
"set_efer: 0x%llx #GP, reserved bits\n",
505 kvm_inject_gp(vcpu
, 0);
510 && (vcpu
->arch
.shadow_efer
& EFER_LME
) != (efer
& EFER_LME
)) {
511 printk(KERN_DEBUG
"set_efer: #GP, change LME while paging\n");
512 kvm_inject_gp(vcpu
, 0);
516 if (efer
& EFER_FFXSR
) {
517 struct kvm_cpuid_entry2
*feat
;
519 feat
= kvm_find_cpuid_entry(vcpu
, 0x80000001, 0);
520 if (!feat
|| !(feat
->edx
& bit(X86_FEATURE_FXSR_OPT
))) {
521 printk(KERN_DEBUG
"set_efer: #GP, enable FFXSR w/o CPUID capability\n");
522 kvm_inject_gp(vcpu
, 0);
527 if (efer
& EFER_SVME
) {
528 struct kvm_cpuid_entry2
*feat
;
530 feat
= kvm_find_cpuid_entry(vcpu
, 0x80000001, 0);
531 if (!feat
|| !(feat
->ecx
& bit(X86_FEATURE_SVM
))) {
532 printk(KERN_DEBUG
"set_efer: #GP, enable SVM w/o SVM\n");
533 kvm_inject_gp(vcpu
, 0);
538 kvm_x86_ops
->set_efer(vcpu
, efer
);
541 efer
|= vcpu
->arch
.shadow_efer
& EFER_LMA
;
543 vcpu
->arch
.shadow_efer
= efer
;
545 vcpu
->arch
.mmu
.base_role
.nxe
= (efer
& EFER_NX
) && !tdp_enabled
;
546 kvm_mmu_reset_context(vcpu
);
549 void kvm_enable_efer_bits(u64 mask
)
551 efer_reserved_bits
&= ~mask
;
553 EXPORT_SYMBOL_GPL(kvm_enable_efer_bits
);
557 * Writes msr value into into the appropriate "register".
558 * Returns 0 on success, non-0 otherwise.
559 * Assumes vcpu_load() was already called.
561 int kvm_set_msr(struct kvm_vcpu
*vcpu
, u32 msr_index
, u64 data
)
563 return kvm_x86_ops
->set_msr(vcpu
, msr_index
, data
);
567 * Adapt set_msr() to msr_io()'s calling convention
569 static int do_set_msr(struct kvm_vcpu
*vcpu
, unsigned index
, u64
*data
)
571 return kvm_set_msr(vcpu
, index
, *data
);
574 static void kvm_write_wall_clock(struct kvm
*kvm
, gpa_t wall_clock
)
577 struct pvclock_wall_clock wc
;
578 struct timespec now
, sys
, boot
;
585 kvm_write_guest(kvm
, wall_clock
, &version
, sizeof(version
));
588 * The guest calculates current wall clock time by adding
589 * system time (updated by kvm_write_guest_time below) to the
590 * wall clock specified here. guest system time equals host
591 * system time for us, thus we must fill in host boot time here.
593 now
= current_kernel_time();
595 boot
= ns_to_timespec(timespec_to_ns(&now
) - timespec_to_ns(&sys
));
597 wc
.sec
= boot
.tv_sec
;
598 wc
.nsec
= boot
.tv_nsec
;
599 wc
.version
= version
;
601 kvm_write_guest(kvm
, wall_clock
, &wc
, sizeof(wc
));
604 kvm_write_guest(kvm
, wall_clock
, &version
, sizeof(version
));
607 static uint32_t div_frac(uint32_t dividend
, uint32_t divisor
)
609 uint32_t quotient
, remainder
;
611 /* Don't try to replace with do_div(), this one calculates
612 * "(dividend << 32) / divisor" */
614 : "=a" (quotient
), "=d" (remainder
)
615 : "0" (0), "1" (dividend
), "r" (divisor
) );
619 static void kvm_set_time_scale(uint32_t tsc_khz
, struct pvclock_vcpu_time_info
*hv_clock
)
621 uint64_t nsecs
= 1000000000LL;
626 tps64
= tsc_khz
* 1000LL;
627 while (tps64
> nsecs
*2) {
632 tps32
= (uint32_t)tps64
;
633 while (tps32
<= (uint32_t)nsecs
) {
638 hv_clock
->tsc_shift
= shift
;
639 hv_clock
->tsc_to_system_mul
= div_frac(nsecs
, tps32
);
641 pr_debug("%s: tsc_khz %u, tsc_shift %d, tsc_mul %u\n",
642 __func__
, tsc_khz
, hv_clock
->tsc_shift
,
643 hv_clock
->tsc_to_system_mul
);
646 static DEFINE_PER_CPU(unsigned long, cpu_tsc_khz
);
648 static void kvm_write_guest_time(struct kvm_vcpu
*v
)
652 struct kvm_vcpu_arch
*vcpu
= &v
->arch
;
654 unsigned long this_tsc_khz
;
656 if ((!vcpu
->time_page
))
659 this_tsc_khz
= get_cpu_var(cpu_tsc_khz
);
660 if (unlikely(vcpu
->hv_clock_tsc_khz
!= this_tsc_khz
)) {
661 kvm_set_time_scale(this_tsc_khz
, &vcpu
->hv_clock
);
662 vcpu
->hv_clock_tsc_khz
= this_tsc_khz
;
664 put_cpu_var(cpu_tsc_khz
);
666 /* Keep irq disabled to prevent changes to the clock */
667 local_irq_save(flags
);
668 kvm_get_msr(v
, MSR_IA32_TSC
, &vcpu
->hv_clock
.tsc_timestamp
);
670 local_irq_restore(flags
);
672 /* With all the info we got, fill in the values */
674 vcpu
->hv_clock
.system_time
= ts
.tv_nsec
+
675 (NSEC_PER_SEC
* (u64
)ts
.tv_sec
);
677 * The interface expects us to write an even number signaling that the
678 * update is finished. Since the guest won't see the intermediate
679 * state, we just increase by 2 at the end.
681 vcpu
->hv_clock
.version
+= 2;
683 shared_kaddr
= kmap_atomic(vcpu
->time_page
, KM_USER0
);
685 memcpy(shared_kaddr
+ vcpu
->time_offset
, &vcpu
->hv_clock
,
686 sizeof(vcpu
->hv_clock
));
688 kunmap_atomic(shared_kaddr
, KM_USER0
);
690 mark_page_dirty(v
->kvm
, vcpu
->time
>> PAGE_SHIFT
);
693 static int kvm_request_guest_time_update(struct kvm_vcpu
*v
)
695 struct kvm_vcpu_arch
*vcpu
= &v
->arch
;
697 if (!vcpu
->time_page
)
699 set_bit(KVM_REQ_KVMCLOCK_UPDATE
, &v
->requests
);
703 static bool msr_mtrr_valid(unsigned msr
)
706 case 0x200 ... 0x200 + 2 * KVM_NR_VAR_MTRR
- 1:
707 case MSR_MTRRfix64K_00000
:
708 case MSR_MTRRfix16K_80000
:
709 case MSR_MTRRfix16K_A0000
:
710 case MSR_MTRRfix4K_C0000
:
711 case MSR_MTRRfix4K_C8000
:
712 case MSR_MTRRfix4K_D0000
:
713 case MSR_MTRRfix4K_D8000
:
714 case MSR_MTRRfix4K_E0000
:
715 case MSR_MTRRfix4K_E8000
:
716 case MSR_MTRRfix4K_F0000
:
717 case MSR_MTRRfix4K_F8000
:
718 case MSR_MTRRdefType
:
719 case MSR_IA32_CR_PAT
:
727 static bool valid_pat_type(unsigned t
)
729 return t
< 8 && (1 << t
) & 0xf3; /* 0, 1, 4, 5, 6, 7 */
732 static bool valid_mtrr_type(unsigned t
)
734 return t
< 8 && (1 << t
) & 0x73; /* 0, 1, 4, 5, 6 */
737 static bool mtrr_valid(struct kvm_vcpu
*vcpu
, u32 msr
, u64 data
)
741 if (!msr_mtrr_valid(msr
))
744 if (msr
== MSR_IA32_CR_PAT
) {
745 for (i
= 0; i
< 8; i
++)
746 if (!valid_pat_type((data
>> (i
* 8)) & 0xff))
749 } else if (msr
== MSR_MTRRdefType
) {
752 return valid_mtrr_type(data
& 0xff);
753 } else if (msr
>= MSR_MTRRfix64K_00000
&& msr
<= MSR_MTRRfix4K_F8000
) {
754 for (i
= 0; i
< 8 ; i
++)
755 if (!valid_mtrr_type((data
>> (i
* 8)) & 0xff))
761 return valid_mtrr_type(data
& 0xff);
764 static int set_msr_mtrr(struct kvm_vcpu
*vcpu
, u32 msr
, u64 data
)
766 u64
*p
= (u64
*)&vcpu
->arch
.mtrr_state
.fixed_ranges
;
768 if (!mtrr_valid(vcpu
, msr
, data
))
771 if (msr
== MSR_MTRRdefType
) {
772 vcpu
->arch
.mtrr_state
.def_type
= data
;
773 vcpu
->arch
.mtrr_state
.enabled
= (data
& 0xc00) >> 10;
774 } else if (msr
== MSR_MTRRfix64K_00000
)
776 else if (msr
== MSR_MTRRfix16K_80000
|| msr
== MSR_MTRRfix16K_A0000
)
777 p
[1 + msr
- MSR_MTRRfix16K_80000
] = data
;
778 else if (msr
>= MSR_MTRRfix4K_C0000
&& msr
<= MSR_MTRRfix4K_F8000
)
779 p
[3 + msr
- MSR_MTRRfix4K_C0000
] = data
;
780 else if (msr
== MSR_IA32_CR_PAT
)
781 vcpu
->arch
.pat
= data
;
782 else { /* Variable MTRRs */
783 int idx
, is_mtrr_mask
;
786 idx
= (msr
- 0x200) / 2;
787 is_mtrr_mask
= msr
- 0x200 - 2 * idx
;
790 (u64
*)&vcpu
->arch
.mtrr_state
.var_ranges
[idx
].base_lo
;
793 (u64
*)&vcpu
->arch
.mtrr_state
.var_ranges
[idx
].mask_lo
;
797 kvm_mmu_reset_context(vcpu
);
801 static int set_msr_mce(struct kvm_vcpu
*vcpu
, u32 msr
, u64 data
)
803 u64 mcg_cap
= vcpu
->arch
.mcg_cap
;
804 unsigned bank_num
= mcg_cap
& 0xff;
807 case MSR_IA32_MCG_STATUS
:
808 vcpu
->arch
.mcg_status
= data
;
810 case MSR_IA32_MCG_CTL
:
811 if (!(mcg_cap
& MCG_CTL_P
))
813 if (data
!= 0 && data
!= ~(u64
)0)
815 vcpu
->arch
.mcg_ctl
= data
;
818 if (msr
>= MSR_IA32_MC0_CTL
&&
819 msr
< MSR_IA32_MC0_CTL
+ 4 * bank_num
) {
820 u32 offset
= msr
- MSR_IA32_MC0_CTL
;
821 /* only 0 or all 1s can be written to IA32_MCi_CTL */
822 if ((offset
& 0x3) == 0 &&
823 data
!= 0 && data
!= ~(u64
)0)
825 vcpu
->arch
.mce_banks
[offset
] = data
;
833 int kvm_set_msr_common(struct kvm_vcpu
*vcpu
, u32 msr
, u64 data
)
837 set_efer(vcpu
, data
);
840 data
&= ~(u64
)0x40; /* ignore flush filter disable */
842 pr_unimpl(vcpu
, "unimplemented HWCR wrmsr: 0x%llx\n",
847 case MSR_AMD64_NB_CFG
:
849 case MSR_IA32_DEBUGCTLMSR
:
851 /* We support the non-activated case already */
853 } else if (data
& ~(DEBUGCTLMSR_LBR
| DEBUGCTLMSR_BTF
)) {
854 /* Values other than LBR and BTF are vendor-specific,
855 thus reserved and should throw a #GP */
858 pr_unimpl(vcpu
, "%s: MSR_IA32_DEBUGCTLMSR 0x%llx, nop\n",
861 case MSR_IA32_UCODE_REV
:
862 case MSR_IA32_UCODE_WRITE
:
863 case MSR_VM_HSAVE_PA
:
865 case 0x200 ... 0x2ff:
866 return set_msr_mtrr(vcpu
, msr
, data
);
867 case MSR_IA32_APICBASE
:
868 kvm_set_apic_base(vcpu
, data
);
870 case APIC_BASE_MSR
... APIC_BASE_MSR
+ 0x3ff:
871 return kvm_x2apic_msr_write(vcpu
, msr
, data
);
872 case MSR_IA32_MISC_ENABLE
:
873 vcpu
->arch
.ia32_misc_enable_msr
= data
;
875 case MSR_KVM_WALL_CLOCK
:
876 vcpu
->kvm
->arch
.wall_clock
= data
;
877 kvm_write_wall_clock(vcpu
->kvm
, data
);
879 case MSR_KVM_SYSTEM_TIME
: {
880 if (vcpu
->arch
.time_page
) {
881 kvm_release_page_dirty(vcpu
->arch
.time_page
);
882 vcpu
->arch
.time_page
= NULL
;
885 vcpu
->arch
.time
= data
;
887 /* we verify if the enable bit is set... */
891 /* ...but clean it before doing the actual write */
892 vcpu
->arch
.time_offset
= data
& ~(PAGE_MASK
| 1);
894 vcpu
->arch
.time_page
=
895 gfn_to_page(vcpu
->kvm
, data
>> PAGE_SHIFT
);
897 if (is_error_page(vcpu
->arch
.time_page
)) {
898 kvm_release_page_clean(vcpu
->arch
.time_page
);
899 vcpu
->arch
.time_page
= NULL
;
902 kvm_request_guest_time_update(vcpu
);
905 case MSR_IA32_MCG_CTL
:
906 case MSR_IA32_MCG_STATUS
:
907 case MSR_IA32_MC0_CTL
... MSR_IA32_MC0_CTL
+ 4 * KVM_MAX_MCE_BANKS
- 1:
908 return set_msr_mce(vcpu
, msr
, data
);
910 /* Performance counters are not protected by a CPUID bit,
911 * so we should check all of them in the generic path for the sake of
912 * cross vendor migration.
913 * Writing a zero into the event select MSRs disables them,
914 * which we perfectly emulate ;-). Any other value should be at least
915 * reported, some guests depend on them.
917 case MSR_P6_EVNTSEL0
:
918 case MSR_P6_EVNTSEL1
:
919 case MSR_K7_EVNTSEL0
:
920 case MSR_K7_EVNTSEL1
:
921 case MSR_K7_EVNTSEL2
:
922 case MSR_K7_EVNTSEL3
:
924 pr_unimpl(vcpu
, "unimplemented perfctr wrmsr: "
925 "0x%x data 0x%llx\n", msr
, data
);
927 /* at least RHEL 4 unconditionally writes to the perfctr registers,
928 * so we ignore writes to make it happy.
930 case MSR_P6_PERFCTR0
:
931 case MSR_P6_PERFCTR1
:
932 case MSR_K7_PERFCTR0
:
933 case MSR_K7_PERFCTR1
:
934 case MSR_K7_PERFCTR2
:
935 case MSR_K7_PERFCTR3
:
936 pr_unimpl(vcpu
, "unimplemented perfctr wrmsr: "
937 "0x%x data 0x%llx\n", msr
, data
);
941 pr_unimpl(vcpu
, "unhandled wrmsr: 0x%x data %llx\n",
945 pr_unimpl(vcpu
, "ignored wrmsr: 0x%x data %llx\n",
952 EXPORT_SYMBOL_GPL(kvm_set_msr_common
);
956 * Reads an msr value (of 'msr_index') into 'pdata'.
957 * Returns 0 on success, non-0 otherwise.
958 * Assumes vcpu_load() was already called.
960 int kvm_get_msr(struct kvm_vcpu
*vcpu
, u32 msr_index
, u64
*pdata
)
962 return kvm_x86_ops
->get_msr(vcpu
, msr_index
, pdata
);
965 static int get_msr_mtrr(struct kvm_vcpu
*vcpu
, u32 msr
, u64
*pdata
)
967 u64
*p
= (u64
*)&vcpu
->arch
.mtrr_state
.fixed_ranges
;
969 if (!msr_mtrr_valid(msr
))
972 if (msr
== MSR_MTRRdefType
)
973 *pdata
= vcpu
->arch
.mtrr_state
.def_type
+
974 (vcpu
->arch
.mtrr_state
.enabled
<< 10);
975 else if (msr
== MSR_MTRRfix64K_00000
)
977 else if (msr
== MSR_MTRRfix16K_80000
|| msr
== MSR_MTRRfix16K_A0000
)
978 *pdata
= p
[1 + msr
- MSR_MTRRfix16K_80000
];
979 else if (msr
>= MSR_MTRRfix4K_C0000
&& msr
<= MSR_MTRRfix4K_F8000
)
980 *pdata
= p
[3 + msr
- MSR_MTRRfix4K_C0000
];
981 else if (msr
== MSR_IA32_CR_PAT
)
982 *pdata
= vcpu
->arch
.pat
;
983 else { /* Variable MTRRs */
984 int idx
, is_mtrr_mask
;
987 idx
= (msr
- 0x200) / 2;
988 is_mtrr_mask
= msr
- 0x200 - 2 * idx
;
991 (u64
*)&vcpu
->arch
.mtrr_state
.var_ranges
[idx
].base_lo
;
994 (u64
*)&vcpu
->arch
.mtrr_state
.var_ranges
[idx
].mask_lo
;
1001 static int get_msr_mce(struct kvm_vcpu
*vcpu
, u32 msr
, u64
*pdata
)
1004 u64 mcg_cap
= vcpu
->arch
.mcg_cap
;
1005 unsigned bank_num
= mcg_cap
& 0xff;
1008 case MSR_IA32_P5_MC_ADDR
:
1009 case MSR_IA32_P5_MC_TYPE
:
1012 case MSR_IA32_MCG_CAP
:
1013 data
= vcpu
->arch
.mcg_cap
;
1015 case MSR_IA32_MCG_CTL
:
1016 if (!(mcg_cap
& MCG_CTL_P
))
1018 data
= vcpu
->arch
.mcg_ctl
;
1020 case MSR_IA32_MCG_STATUS
:
1021 data
= vcpu
->arch
.mcg_status
;
1024 if (msr
>= MSR_IA32_MC0_CTL
&&
1025 msr
< MSR_IA32_MC0_CTL
+ 4 * bank_num
) {
1026 u32 offset
= msr
- MSR_IA32_MC0_CTL
;
1027 data
= vcpu
->arch
.mce_banks
[offset
];
1036 int kvm_get_msr_common(struct kvm_vcpu
*vcpu
, u32 msr
, u64
*pdata
)
1041 case MSR_IA32_PLATFORM_ID
:
1042 case MSR_IA32_UCODE_REV
:
1043 case MSR_IA32_EBL_CR_POWERON
:
1044 case MSR_IA32_DEBUGCTLMSR
:
1045 case MSR_IA32_LASTBRANCHFROMIP
:
1046 case MSR_IA32_LASTBRANCHTOIP
:
1047 case MSR_IA32_LASTINTFROMIP
:
1048 case MSR_IA32_LASTINTTOIP
:
1051 case MSR_VM_HSAVE_PA
:
1052 case MSR_P6_EVNTSEL0
:
1053 case MSR_P6_EVNTSEL1
:
1054 case MSR_K7_EVNTSEL0
:
1055 case MSR_K8_INT_PENDING_MSG
:
1056 case MSR_AMD64_NB_CFG
:
1060 data
= 0x500 | KVM_NR_VAR_MTRR
;
1062 case 0x200 ... 0x2ff:
1063 return get_msr_mtrr(vcpu
, msr
, pdata
);
1064 case 0xcd: /* fsb frequency */
1067 case MSR_IA32_APICBASE
:
1068 data
= kvm_get_apic_base(vcpu
);
1070 case APIC_BASE_MSR
... APIC_BASE_MSR
+ 0x3ff:
1071 return kvm_x2apic_msr_read(vcpu
, msr
, pdata
);
1073 case MSR_IA32_MISC_ENABLE
:
1074 data
= vcpu
->arch
.ia32_misc_enable_msr
;
1076 case MSR_IA32_PERF_STATUS
:
1077 /* TSC increment by tick */
1079 /* CPU multiplier */
1080 data
|= (((uint64_t)4ULL) << 40);
1083 data
= vcpu
->arch
.shadow_efer
;
1085 case MSR_KVM_WALL_CLOCK
:
1086 data
= vcpu
->kvm
->arch
.wall_clock
;
1088 case MSR_KVM_SYSTEM_TIME
:
1089 data
= vcpu
->arch
.time
;
1091 case MSR_IA32_P5_MC_ADDR
:
1092 case MSR_IA32_P5_MC_TYPE
:
1093 case MSR_IA32_MCG_CAP
:
1094 case MSR_IA32_MCG_CTL
:
1095 case MSR_IA32_MCG_STATUS
:
1096 case MSR_IA32_MC0_CTL
... MSR_IA32_MC0_CTL
+ 4 * KVM_MAX_MCE_BANKS
- 1:
1097 return get_msr_mce(vcpu
, msr
, pdata
);
1100 pr_unimpl(vcpu
, "unhandled rdmsr: 0x%x\n", msr
);
1103 pr_unimpl(vcpu
, "ignored rdmsr: 0x%x\n", msr
);
1111 EXPORT_SYMBOL_GPL(kvm_get_msr_common
);
1114 * Read or write a bunch of msrs. All parameters are kernel addresses.
1116 * @return number of msrs set successfully.
1118 static int __msr_io(struct kvm_vcpu
*vcpu
, struct kvm_msrs
*msrs
,
1119 struct kvm_msr_entry
*entries
,
1120 int (*do_msr
)(struct kvm_vcpu
*vcpu
,
1121 unsigned index
, u64
*data
))
1127 down_read(&vcpu
->kvm
->slots_lock
);
1128 for (i
= 0; i
< msrs
->nmsrs
; ++i
)
1129 if (do_msr(vcpu
, entries
[i
].index
, &entries
[i
].data
))
1131 up_read(&vcpu
->kvm
->slots_lock
);
1139 * Read or write a bunch of msrs. Parameters are user addresses.
1141 * @return number of msrs set successfully.
1143 static int msr_io(struct kvm_vcpu
*vcpu
, struct kvm_msrs __user
*user_msrs
,
1144 int (*do_msr
)(struct kvm_vcpu
*vcpu
,
1145 unsigned index
, u64
*data
),
1148 struct kvm_msrs msrs
;
1149 struct kvm_msr_entry
*entries
;
1154 if (copy_from_user(&msrs
, user_msrs
, sizeof msrs
))
1158 if (msrs
.nmsrs
>= MAX_IO_MSRS
)
1162 size
= sizeof(struct kvm_msr_entry
) * msrs
.nmsrs
;
1163 entries
= vmalloc(size
);
1168 if (copy_from_user(entries
, user_msrs
->entries
, size
))
1171 r
= n
= __msr_io(vcpu
, &msrs
, entries
, do_msr
);
1176 if (writeback
&& copy_to_user(user_msrs
->entries
, entries
, size
))
1187 int kvm_dev_ioctl_check_extension(long ext
)
1192 case KVM_CAP_IRQCHIP
:
1194 case KVM_CAP_MMU_SHADOW_CACHE_CONTROL
:
1195 case KVM_CAP_SET_TSS_ADDR
:
1196 case KVM_CAP_EXT_CPUID
:
1197 case KVM_CAP_CLOCKSOURCE
:
1199 case KVM_CAP_NOP_IO_DELAY
:
1200 case KVM_CAP_MP_STATE
:
1201 case KVM_CAP_SYNC_MMU
:
1202 case KVM_CAP_REINJECT_CONTROL
:
1203 case KVM_CAP_IRQ_INJECT_STATUS
:
1204 case KVM_CAP_ASSIGN_DEV_IRQ
:
1209 case KVM_CAP_COALESCED_MMIO
:
1210 r
= KVM_COALESCED_MMIO_PAGE_OFFSET
;
1213 r
= !kvm_x86_ops
->cpu_has_accelerated_tpr();
1215 case KVM_CAP_NR_VCPUS
:
1218 case KVM_CAP_NR_MEMSLOTS
:
1219 r
= KVM_MEMORY_SLOTS
;
1221 case KVM_CAP_PV_MMU
:
1228 r
= KVM_MAX_MCE_BANKS
;
1238 long kvm_arch_dev_ioctl(struct file
*filp
,
1239 unsigned int ioctl
, unsigned long arg
)
1241 void __user
*argp
= (void __user
*)arg
;
1245 case KVM_GET_MSR_INDEX_LIST
: {
1246 struct kvm_msr_list __user
*user_msr_list
= argp
;
1247 struct kvm_msr_list msr_list
;
1251 if (copy_from_user(&msr_list
, user_msr_list
, sizeof msr_list
))
1254 msr_list
.nmsrs
= num_msrs_to_save
+ ARRAY_SIZE(emulated_msrs
);
1255 if (copy_to_user(user_msr_list
, &msr_list
, sizeof msr_list
))
1258 if (n
< msr_list
.nmsrs
)
1261 if (copy_to_user(user_msr_list
->indices
, &msrs_to_save
,
1262 num_msrs_to_save
* sizeof(u32
)))
1264 if (copy_to_user(user_msr_list
->indices
+ num_msrs_to_save
,
1266 ARRAY_SIZE(emulated_msrs
) * sizeof(u32
)))
1271 case KVM_GET_SUPPORTED_CPUID
: {
1272 struct kvm_cpuid2 __user
*cpuid_arg
= argp
;
1273 struct kvm_cpuid2 cpuid
;
1276 if (copy_from_user(&cpuid
, cpuid_arg
, sizeof cpuid
))
1278 r
= kvm_dev_ioctl_get_supported_cpuid(&cpuid
,
1279 cpuid_arg
->entries
);
1284 if (copy_to_user(cpuid_arg
, &cpuid
, sizeof cpuid
))
1289 case KVM_X86_GET_MCE_CAP_SUPPORTED
: {
1292 mce_cap
= KVM_MCE_CAP_SUPPORTED
;
1294 if (copy_to_user(argp
, &mce_cap
, sizeof mce_cap
))
1306 void kvm_arch_vcpu_load(struct kvm_vcpu
*vcpu
, int cpu
)
1308 kvm_x86_ops
->vcpu_load(vcpu
, cpu
);
1309 kvm_request_guest_time_update(vcpu
);
1312 void kvm_arch_vcpu_put(struct kvm_vcpu
*vcpu
)
1314 kvm_x86_ops
->vcpu_put(vcpu
);
1315 kvm_put_guest_fpu(vcpu
);
1318 static int is_efer_nx(void)
1320 unsigned long long efer
= 0;
1322 rdmsrl_safe(MSR_EFER
, &efer
);
1323 return efer
& EFER_NX
;
1326 static void cpuid_fix_nx_cap(struct kvm_vcpu
*vcpu
)
1329 struct kvm_cpuid_entry2
*e
, *entry
;
1332 for (i
= 0; i
< vcpu
->arch
.cpuid_nent
; ++i
) {
1333 e
= &vcpu
->arch
.cpuid_entries
[i
];
1334 if (e
->function
== 0x80000001) {
1339 if (entry
&& (entry
->edx
& (1 << 20)) && !is_efer_nx()) {
1340 entry
->edx
&= ~(1 << 20);
1341 printk(KERN_INFO
"kvm: guest NX capability removed\n");
1345 /* when an old userspace process fills a new kernel module */
1346 static int kvm_vcpu_ioctl_set_cpuid(struct kvm_vcpu
*vcpu
,
1347 struct kvm_cpuid
*cpuid
,
1348 struct kvm_cpuid_entry __user
*entries
)
1351 struct kvm_cpuid_entry
*cpuid_entries
;
1354 if (cpuid
->nent
> KVM_MAX_CPUID_ENTRIES
)
1357 cpuid_entries
= vmalloc(sizeof(struct kvm_cpuid_entry
) * cpuid
->nent
);
1361 if (copy_from_user(cpuid_entries
, entries
,
1362 cpuid
->nent
* sizeof(struct kvm_cpuid_entry
)))
1364 for (i
= 0; i
< cpuid
->nent
; i
++) {
1365 vcpu
->arch
.cpuid_entries
[i
].function
= cpuid_entries
[i
].function
;
1366 vcpu
->arch
.cpuid_entries
[i
].eax
= cpuid_entries
[i
].eax
;
1367 vcpu
->arch
.cpuid_entries
[i
].ebx
= cpuid_entries
[i
].ebx
;
1368 vcpu
->arch
.cpuid_entries
[i
].ecx
= cpuid_entries
[i
].ecx
;
1369 vcpu
->arch
.cpuid_entries
[i
].edx
= cpuid_entries
[i
].edx
;
1370 vcpu
->arch
.cpuid_entries
[i
].index
= 0;
1371 vcpu
->arch
.cpuid_entries
[i
].flags
= 0;
1372 vcpu
->arch
.cpuid_entries
[i
].padding
[0] = 0;
1373 vcpu
->arch
.cpuid_entries
[i
].padding
[1] = 0;
1374 vcpu
->arch
.cpuid_entries
[i
].padding
[2] = 0;
1376 vcpu
->arch
.cpuid_nent
= cpuid
->nent
;
1377 cpuid_fix_nx_cap(vcpu
);
1379 kvm_apic_set_version(vcpu
);
1382 vfree(cpuid_entries
);
1387 static int kvm_vcpu_ioctl_set_cpuid2(struct kvm_vcpu
*vcpu
,
1388 struct kvm_cpuid2
*cpuid
,
1389 struct kvm_cpuid_entry2 __user
*entries
)
1394 if (cpuid
->nent
> KVM_MAX_CPUID_ENTRIES
)
1397 if (copy_from_user(&vcpu
->arch
.cpuid_entries
, entries
,
1398 cpuid
->nent
* sizeof(struct kvm_cpuid_entry2
)))
1400 vcpu
->arch
.cpuid_nent
= cpuid
->nent
;
1401 kvm_apic_set_version(vcpu
);
1408 static int kvm_vcpu_ioctl_get_cpuid2(struct kvm_vcpu
*vcpu
,
1409 struct kvm_cpuid2
*cpuid
,
1410 struct kvm_cpuid_entry2 __user
*entries
)
1415 if (cpuid
->nent
< vcpu
->arch
.cpuid_nent
)
1418 if (copy_to_user(entries
, &vcpu
->arch
.cpuid_entries
,
1419 vcpu
->arch
.cpuid_nent
* sizeof(struct kvm_cpuid_entry2
)))
1424 cpuid
->nent
= vcpu
->arch
.cpuid_nent
;
1428 static void do_cpuid_1_ent(struct kvm_cpuid_entry2
*entry
, u32 function
,
1431 entry
->function
= function
;
1432 entry
->index
= index
;
1433 cpuid_count(entry
->function
, entry
->index
,
1434 &entry
->eax
, &entry
->ebx
, &entry
->ecx
, &entry
->edx
);
1438 #define F(x) bit(X86_FEATURE_##x)
1440 static void do_cpuid_ent(struct kvm_cpuid_entry2
*entry
, u32 function
,
1441 u32 index
, int *nent
, int maxnent
)
1443 unsigned f_nx
= is_efer_nx() ? F(NX
) : 0;
1444 #ifdef CONFIG_X86_64
1445 unsigned f_lm
= F(LM
);
1451 const u32 kvm_supported_word0_x86_features
=
1452 F(FPU
) | F(VME
) | F(DE
) | F(PSE
) |
1453 F(TSC
) | F(MSR
) | F(PAE
) | F(MCE
) |
1454 F(CX8
) | F(APIC
) | 0 /* Reserved */ | F(SEP
) |
1455 F(MTRR
) | F(PGE
) | F(MCA
) | F(CMOV
) |
1456 F(PAT
) | F(PSE36
) | 0 /* PSN */ | F(CLFLSH
) |
1457 0 /* Reserved, DS, ACPI */ | F(MMX
) |
1458 F(FXSR
) | F(XMM
) | F(XMM2
) | F(SELFSNOOP
) |
1459 0 /* HTT, TM, Reserved, PBE */;
1460 /* cpuid 0x80000001.edx */
1461 const u32 kvm_supported_word1_x86_features
=
1462 F(FPU
) | F(VME
) | F(DE
) | F(PSE
) |
1463 F(TSC
) | F(MSR
) | F(PAE
) | F(MCE
) |
1464 F(CX8
) | F(APIC
) | 0 /* Reserved */ | F(SYSCALL
) |
1465 F(MTRR
) | F(PGE
) | F(MCA
) | F(CMOV
) |
1466 F(PAT
) | F(PSE36
) | 0 /* Reserved */ |
1467 f_nx
| 0 /* Reserved */ | F(MMXEXT
) | F(MMX
) |
1468 F(FXSR
) | F(FXSR_OPT
) | 0 /* GBPAGES */ | 0 /* RDTSCP */ |
1469 0 /* Reserved */ | f_lm
| F(3DNOWEXT
) | F(3DNOW
);
1471 const u32 kvm_supported_word4_x86_features
=
1472 F(XMM3
) | 0 /* Reserved, DTES64, MONITOR */ |
1473 0 /* DS-CPL, VMX, SMX, EST */ |
1474 0 /* TM2 */ | F(SSSE3
) | 0 /* CNXT-ID */ | 0 /* Reserved */ |
1475 0 /* Reserved */ | F(CX16
) | 0 /* xTPR Update, PDCM */ |
1476 0 /* Reserved, DCA */ | F(XMM4_1
) |
1477 F(XMM4_2
) | F(X2APIC
) | F(MOVBE
) | F(POPCNT
) |
1478 0 /* Reserved, XSAVE, OSXSAVE */;
1479 /* cpuid 0x80000001.ecx */
1480 const u32 kvm_supported_word6_x86_features
=
1481 F(LAHF_LM
) | F(CMP_LEGACY
) | F(SVM
) | 0 /* ExtApicSpace */ |
1482 F(CR8_LEGACY
) | F(ABM
) | F(SSE4A
) | F(MISALIGNSSE
) |
1483 F(3DNOWPREFETCH
) | 0 /* OSVW */ | 0 /* IBS */ | F(SSE5
) |
1484 0 /* SKINIT */ | 0 /* WDT */;
1486 /* all calls to cpuid_count() should be made on the same cpu */
1488 do_cpuid_1_ent(entry
, function
, index
);
1493 entry
->eax
= min(entry
->eax
, (u32
)0xb);
1496 entry
->edx
&= kvm_supported_word0_x86_features
;
1497 entry
->ecx
&= kvm_supported_word4_x86_features
;
1499 /* function 2 entries are STATEFUL. That is, repeated cpuid commands
1500 * may return different values. This forces us to get_cpu() before
1501 * issuing the first command, and also to emulate this annoying behavior
1502 * in kvm_emulate_cpuid() using KVM_CPUID_FLAG_STATE_READ_NEXT */
1504 int t
, times
= entry
->eax
& 0xff;
1506 entry
->flags
|= KVM_CPUID_FLAG_STATEFUL_FUNC
;
1507 entry
->flags
|= KVM_CPUID_FLAG_STATE_READ_NEXT
;
1508 for (t
= 1; t
< times
&& *nent
< maxnent
; ++t
) {
1509 do_cpuid_1_ent(&entry
[t
], function
, 0);
1510 entry
[t
].flags
|= KVM_CPUID_FLAG_STATEFUL_FUNC
;
1515 /* function 4 and 0xb have additional index. */
1519 entry
->flags
|= KVM_CPUID_FLAG_SIGNIFCANT_INDEX
;
1520 /* read more entries until cache_type is zero */
1521 for (i
= 1; *nent
< maxnent
; ++i
) {
1522 cache_type
= entry
[i
- 1].eax
& 0x1f;
1525 do_cpuid_1_ent(&entry
[i
], function
, i
);
1527 KVM_CPUID_FLAG_SIGNIFCANT_INDEX
;
1535 entry
->flags
|= KVM_CPUID_FLAG_SIGNIFCANT_INDEX
;
1536 /* read more entries until level_type is zero */
1537 for (i
= 1; *nent
< maxnent
; ++i
) {
1538 level_type
= entry
[i
- 1].ecx
& 0xff00;
1541 do_cpuid_1_ent(&entry
[i
], function
, i
);
1543 KVM_CPUID_FLAG_SIGNIFCANT_INDEX
;
1549 entry
->eax
= min(entry
->eax
, 0x8000001a);
1552 entry
->edx
&= kvm_supported_word1_x86_features
;
1553 entry
->ecx
&= kvm_supported_word6_x86_features
;
1561 static int kvm_dev_ioctl_get_supported_cpuid(struct kvm_cpuid2
*cpuid
,
1562 struct kvm_cpuid_entry2 __user
*entries
)
1564 struct kvm_cpuid_entry2
*cpuid_entries
;
1565 int limit
, nent
= 0, r
= -E2BIG
;
1568 if (cpuid
->nent
< 1)
1571 cpuid_entries
= vmalloc(sizeof(struct kvm_cpuid_entry2
) * cpuid
->nent
);
1575 do_cpuid_ent(&cpuid_entries
[0], 0, 0, &nent
, cpuid
->nent
);
1576 limit
= cpuid_entries
[0].eax
;
1577 for (func
= 1; func
<= limit
&& nent
< cpuid
->nent
; ++func
)
1578 do_cpuid_ent(&cpuid_entries
[nent
], func
, 0,
1579 &nent
, cpuid
->nent
);
1581 if (nent
>= cpuid
->nent
)
1584 do_cpuid_ent(&cpuid_entries
[nent
], 0x80000000, 0, &nent
, cpuid
->nent
);
1585 limit
= cpuid_entries
[nent
- 1].eax
;
1586 for (func
= 0x80000001; func
<= limit
&& nent
< cpuid
->nent
; ++func
)
1587 do_cpuid_ent(&cpuid_entries
[nent
], func
, 0,
1588 &nent
, cpuid
->nent
);
1590 if (nent
>= cpuid
->nent
)
1594 if (copy_to_user(entries
, cpuid_entries
,
1595 nent
* sizeof(struct kvm_cpuid_entry2
)))
1601 vfree(cpuid_entries
);
1606 static int kvm_vcpu_ioctl_get_lapic(struct kvm_vcpu
*vcpu
,
1607 struct kvm_lapic_state
*s
)
1610 memcpy(s
->regs
, vcpu
->arch
.apic
->regs
, sizeof *s
);
1616 static int kvm_vcpu_ioctl_set_lapic(struct kvm_vcpu
*vcpu
,
1617 struct kvm_lapic_state
*s
)
1620 memcpy(vcpu
->arch
.apic
->regs
, s
->regs
, sizeof *s
);
1621 kvm_apic_post_state_restore(vcpu
);
1627 static int kvm_vcpu_ioctl_interrupt(struct kvm_vcpu
*vcpu
,
1628 struct kvm_interrupt
*irq
)
1630 if (irq
->irq
< 0 || irq
->irq
>= 256)
1632 if (irqchip_in_kernel(vcpu
->kvm
))
1636 kvm_queue_interrupt(vcpu
, irq
->irq
, false);
1643 static int kvm_vcpu_ioctl_nmi(struct kvm_vcpu
*vcpu
)
1646 kvm_inject_nmi(vcpu
);
1652 static int vcpu_ioctl_tpr_access_reporting(struct kvm_vcpu
*vcpu
,
1653 struct kvm_tpr_access_ctl
*tac
)
1657 vcpu
->arch
.tpr_access_reporting
= !!tac
->enabled
;
1661 static int kvm_vcpu_ioctl_x86_setup_mce(struct kvm_vcpu
*vcpu
,
1665 unsigned bank_num
= mcg_cap
& 0xff, bank
;
1670 if (mcg_cap
& ~(KVM_MCE_CAP_SUPPORTED
| 0xff | 0xff0000))
1673 vcpu
->arch
.mcg_cap
= mcg_cap
;
1674 /* Init IA32_MCG_CTL to all 1s */
1675 if (mcg_cap
& MCG_CTL_P
)
1676 vcpu
->arch
.mcg_ctl
= ~(u64
)0;
1677 /* Init IA32_MCi_CTL to all 1s */
1678 for (bank
= 0; bank
< bank_num
; bank
++)
1679 vcpu
->arch
.mce_banks
[bank
*4] = ~(u64
)0;
1684 static int kvm_vcpu_ioctl_x86_set_mce(struct kvm_vcpu
*vcpu
,
1685 struct kvm_x86_mce
*mce
)
1687 u64 mcg_cap
= vcpu
->arch
.mcg_cap
;
1688 unsigned bank_num
= mcg_cap
& 0xff;
1689 u64
*banks
= vcpu
->arch
.mce_banks
;
1691 if (mce
->bank
>= bank_num
|| !(mce
->status
& MCI_STATUS_VAL
))
1694 * if IA32_MCG_CTL is not all 1s, the uncorrected error
1695 * reporting is disabled
1697 if ((mce
->status
& MCI_STATUS_UC
) && (mcg_cap
& MCG_CTL_P
) &&
1698 vcpu
->arch
.mcg_ctl
!= ~(u64
)0)
1700 banks
+= 4 * mce
->bank
;
1702 * if IA32_MCi_CTL is not all 1s, the uncorrected error
1703 * reporting is disabled for the bank
1705 if ((mce
->status
& MCI_STATUS_UC
) && banks
[0] != ~(u64
)0)
1707 if (mce
->status
& MCI_STATUS_UC
) {
1708 if ((vcpu
->arch
.mcg_status
& MCG_STATUS_MCIP
) ||
1709 !(vcpu
->arch
.cr4
& X86_CR4_MCE
)) {
1710 printk(KERN_DEBUG
"kvm: set_mce: "
1711 "injects mce exception while "
1712 "previous one is in progress!\n");
1713 set_bit(KVM_REQ_TRIPLE_FAULT
, &vcpu
->requests
);
1716 if (banks
[1] & MCI_STATUS_VAL
)
1717 mce
->status
|= MCI_STATUS_OVER
;
1718 banks
[2] = mce
->addr
;
1719 banks
[3] = mce
->misc
;
1720 vcpu
->arch
.mcg_status
= mce
->mcg_status
;
1721 banks
[1] = mce
->status
;
1722 kvm_queue_exception(vcpu
, MC_VECTOR
);
1723 } else if (!(banks
[1] & MCI_STATUS_VAL
)
1724 || !(banks
[1] & MCI_STATUS_UC
)) {
1725 if (banks
[1] & MCI_STATUS_VAL
)
1726 mce
->status
|= MCI_STATUS_OVER
;
1727 banks
[2] = mce
->addr
;
1728 banks
[3] = mce
->misc
;
1729 banks
[1] = mce
->status
;
1731 banks
[1] |= MCI_STATUS_OVER
;
1735 long kvm_arch_vcpu_ioctl(struct file
*filp
,
1736 unsigned int ioctl
, unsigned long arg
)
1738 struct kvm_vcpu
*vcpu
= filp
->private_data
;
1739 void __user
*argp
= (void __user
*)arg
;
1741 struct kvm_lapic_state
*lapic
= NULL
;
1744 case KVM_GET_LAPIC
: {
1745 lapic
= kzalloc(sizeof(struct kvm_lapic_state
), GFP_KERNEL
);
1750 r
= kvm_vcpu_ioctl_get_lapic(vcpu
, lapic
);
1754 if (copy_to_user(argp
, lapic
, sizeof(struct kvm_lapic_state
)))
1759 case KVM_SET_LAPIC
: {
1760 lapic
= kmalloc(sizeof(struct kvm_lapic_state
), GFP_KERNEL
);
1765 if (copy_from_user(lapic
, argp
, sizeof(struct kvm_lapic_state
)))
1767 r
= kvm_vcpu_ioctl_set_lapic(vcpu
, lapic
);
1773 case KVM_INTERRUPT
: {
1774 struct kvm_interrupt irq
;
1777 if (copy_from_user(&irq
, argp
, sizeof irq
))
1779 r
= kvm_vcpu_ioctl_interrupt(vcpu
, &irq
);
1786 r
= kvm_vcpu_ioctl_nmi(vcpu
);
1792 case KVM_SET_CPUID
: {
1793 struct kvm_cpuid __user
*cpuid_arg
= argp
;
1794 struct kvm_cpuid cpuid
;
1797 if (copy_from_user(&cpuid
, cpuid_arg
, sizeof cpuid
))
1799 r
= kvm_vcpu_ioctl_set_cpuid(vcpu
, &cpuid
, cpuid_arg
->entries
);
1804 case KVM_SET_CPUID2
: {
1805 struct kvm_cpuid2 __user
*cpuid_arg
= argp
;
1806 struct kvm_cpuid2 cpuid
;
1809 if (copy_from_user(&cpuid
, cpuid_arg
, sizeof cpuid
))
1811 r
= kvm_vcpu_ioctl_set_cpuid2(vcpu
, &cpuid
,
1812 cpuid_arg
->entries
);
1817 case KVM_GET_CPUID2
: {
1818 struct kvm_cpuid2 __user
*cpuid_arg
= argp
;
1819 struct kvm_cpuid2 cpuid
;
1822 if (copy_from_user(&cpuid
, cpuid_arg
, sizeof cpuid
))
1824 r
= kvm_vcpu_ioctl_get_cpuid2(vcpu
, &cpuid
,
1825 cpuid_arg
->entries
);
1829 if (copy_to_user(cpuid_arg
, &cpuid
, sizeof cpuid
))
1835 r
= msr_io(vcpu
, argp
, kvm_get_msr
, 1);
1838 r
= msr_io(vcpu
, argp
, do_set_msr
, 0);
1840 case KVM_TPR_ACCESS_REPORTING
: {
1841 struct kvm_tpr_access_ctl tac
;
1844 if (copy_from_user(&tac
, argp
, sizeof tac
))
1846 r
= vcpu_ioctl_tpr_access_reporting(vcpu
, &tac
);
1850 if (copy_to_user(argp
, &tac
, sizeof tac
))
1855 case KVM_SET_VAPIC_ADDR
: {
1856 struct kvm_vapic_addr va
;
1859 if (!irqchip_in_kernel(vcpu
->kvm
))
1862 if (copy_from_user(&va
, argp
, sizeof va
))
1865 kvm_lapic_set_vapic_addr(vcpu
, va
.vapic_addr
);
1868 case KVM_X86_SETUP_MCE
: {
1872 if (copy_from_user(&mcg_cap
, argp
, sizeof mcg_cap
))
1874 r
= kvm_vcpu_ioctl_x86_setup_mce(vcpu
, mcg_cap
);
1877 case KVM_X86_SET_MCE
: {
1878 struct kvm_x86_mce mce
;
1881 if (copy_from_user(&mce
, argp
, sizeof mce
))
1883 r
= kvm_vcpu_ioctl_x86_set_mce(vcpu
, &mce
);
1894 static int kvm_vm_ioctl_set_tss_addr(struct kvm
*kvm
, unsigned long addr
)
1898 if (addr
> (unsigned int)(-3 * PAGE_SIZE
))
1900 ret
= kvm_x86_ops
->set_tss_addr(kvm
, addr
);
1904 static int kvm_vm_ioctl_set_nr_mmu_pages(struct kvm
*kvm
,
1905 u32 kvm_nr_mmu_pages
)
1907 if (kvm_nr_mmu_pages
< KVM_MIN_ALLOC_MMU_PAGES
)
1910 down_write(&kvm
->slots_lock
);
1911 spin_lock(&kvm
->mmu_lock
);
1913 kvm_mmu_change_mmu_pages(kvm
, kvm_nr_mmu_pages
);
1914 kvm
->arch
.n_requested_mmu_pages
= kvm_nr_mmu_pages
;
1916 spin_unlock(&kvm
->mmu_lock
);
1917 up_write(&kvm
->slots_lock
);
1921 static int kvm_vm_ioctl_get_nr_mmu_pages(struct kvm
*kvm
)
1923 return kvm
->arch
.n_alloc_mmu_pages
;
1926 gfn_t
unalias_gfn(struct kvm
*kvm
, gfn_t gfn
)
1929 struct kvm_mem_alias
*alias
;
1931 for (i
= 0; i
< kvm
->arch
.naliases
; ++i
) {
1932 alias
= &kvm
->arch
.aliases
[i
];
1933 if (gfn
>= alias
->base_gfn
1934 && gfn
< alias
->base_gfn
+ alias
->npages
)
1935 return alias
->target_gfn
+ gfn
- alias
->base_gfn
;
1941 * Set a new alias region. Aliases map a portion of physical memory into
1942 * another portion. This is useful for memory windows, for example the PC
1945 static int kvm_vm_ioctl_set_memory_alias(struct kvm
*kvm
,
1946 struct kvm_memory_alias
*alias
)
1949 struct kvm_mem_alias
*p
;
1952 /* General sanity checks */
1953 if (alias
->memory_size
& (PAGE_SIZE
- 1))
1955 if (alias
->guest_phys_addr
& (PAGE_SIZE
- 1))
1957 if (alias
->slot
>= KVM_ALIAS_SLOTS
)
1959 if (alias
->guest_phys_addr
+ alias
->memory_size
1960 < alias
->guest_phys_addr
)
1962 if (alias
->target_phys_addr
+ alias
->memory_size
1963 < alias
->target_phys_addr
)
1966 down_write(&kvm
->slots_lock
);
1967 spin_lock(&kvm
->mmu_lock
);
1969 p
= &kvm
->arch
.aliases
[alias
->slot
];
1970 p
->base_gfn
= alias
->guest_phys_addr
>> PAGE_SHIFT
;
1971 p
->npages
= alias
->memory_size
>> PAGE_SHIFT
;
1972 p
->target_gfn
= alias
->target_phys_addr
>> PAGE_SHIFT
;
1974 for (n
= KVM_ALIAS_SLOTS
; n
> 0; --n
)
1975 if (kvm
->arch
.aliases
[n
- 1].npages
)
1977 kvm
->arch
.naliases
= n
;
1979 spin_unlock(&kvm
->mmu_lock
);
1980 kvm_mmu_zap_all(kvm
);
1982 up_write(&kvm
->slots_lock
);
1990 static int kvm_vm_ioctl_get_irqchip(struct kvm
*kvm
, struct kvm_irqchip
*chip
)
1995 switch (chip
->chip_id
) {
1996 case KVM_IRQCHIP_PIC_MASTER
:
1997 memcpy(&chip
->chip
.pic
,
1998 &pic_irqchip(kvm
)->pics
[0],
1999 sizeof(struct kvm_pic_state
));
2001 case KVM_IRQCHIP_PIC_SLAVE
:
2002 memcpy(&chip
->chip
.pic
,
2003 &pic_irqchip(kvm
)->pics
[1],
2004 sizeof(struct kvm_pic_state
));
2006 case KVM_IRQCHIP_IOAPIC
:
2007 memcpy(&chip
->chip
.ioapic
,
2008 ioapic_irqchip(kvm
),
2009 sizeof(struct kvm_ioapic_state
));
2018 static int kvm_vm_ioctl_set_irqchip(struct kvm
*kvm
, struct kvm_irqchip
*chip
)
2023 switch (chip
->chip_id
) {
2024 case KVM_IRQCHIP_PIC_MASTER
:
2025 spin_lock(&pic_irqchip(kvm
)->lock
);
2026 memcpy(&pic_irqchip(kvm
)->pics
[0],
2028 sizeof(struct kvm_pic_state
));
2029 spin_unlock(&pic_irqchip(kvm
)->lock
);
2031 case KVM_IRQCHIP_PIC_SLAVE
:
2032 spin_lock(&pic_irqchip(kvm
)->lock
);
2033 memcpy(&pic_irqchip(kvm
)->pics
[1],
2035 sizeof(struct kvm_pic_state
));
2036 spin_unlock(&pic_irqchip(kvm
)->lock
);
2038 case KVM_IRQCHIP_IOAPIC
:
2039 mutex_lock(&kvm
->irq_lock
);
2040 memcpy(ioapic_irqchip(kvm
),
2042 sizeof(struct kvm_ioapic_state
));
2043 mutex_unlock(&kvm
->irq_lock
);
2049 kvm_pic_update_irq(pic_irqchip(kvm
));
2053 static int kvm_vm_ioctl_get_pit(struct kvm
*kvm
, struct kvm_pit_state
*ps
)
2057 mutex_lock(&kvm
->arch
.vpit
->pit_state
.lock
);
2058 memcpy(ps
, &kvm
->arch
.vpit
->pit_state
, sizeof(struct kvm_pit_state
));
2059 mutex_unlock(&kvm
->arch
.vpit
->pit_state
.lock
);
2063 static int kvm_vm_ioctl_set_pit(struct kvm
*kvm
, struct kvm_pit_state
*ps
)
2067 mutex_lock(&kvm
->arch
.vpit
->pit_state
.lock
);
2068 memcpy(&kvm
->arch
.vpit
->pit_state
, ps
, sizeof(struct kvm_pit_state
));
2069 kvm_pit_load_count(kvm
, 0, ps
->channels
[0].count
);
2070 mutex_unlock(&kvm
->arch
.vpit
->pit_state
.lock
);
2074 static int kvm_vm_ioctl_reinject(struct kvm
*kvm
,
2075 struct kvm_reinject_control
*control
)
2077 if (!kvm
->arch
.vpit
)
2079 mutex_lock(&kvm
->arch
.vpit
->pit_state
.lock
);
2080 kvm
->arch
.vpit
->pit_state
.pit_timer
.reinject
= control
->pit_reinject
;
2081 mutex_unlock(&kvm
->arch
.vpit
->pit_state
.lock
);
2086 * Get (and clear) the dirty memory log for a memory slot.
2088 int kvm_vm_ioctl_get_dirty_log(struct kvm
*kvm
,
2089 struct kvm_dirty_log
*log
)
2093 struct kvm_memory_slot
*memslot
;
2096 down_write(&kvm
->slots_lock
);
2098 r
= kvm_get_dirty_log(kvm
, log
, &is_dirty
);
2102 /* If nothing is dirty, don't bother messing with page tables. */
2104 spin_lock(&kvm
->mmu_lock
);
2105 kvm_mmu_slot_remove_write_access(kvm
, log
->slot
);
2106 spin_unlock(&kvm
->mmu_lock
);
2107 kvm_flush_remote_tlbs(kvm
);
2108 memslot
= &kvm
->memslots
[log
->slot
];
2109 n
= ALIGN(memslot
->npages
, BITS_PER_LONG
) / 8;
2110 memset(memslot
->dirty_bitmap
, 0, n
);
2114 up_write(&kvm
->slots_lock
);
2118 long kvm_arch_vm_ioctl(struct file
*filp
,
2119 unsigned int ioctl
, unsigned long arg
)
2121 struct kvm
*kvm
= filp
->private_data
;
2122 void __user
*argp
= (void __user
*)arg
;
2125 * This union makes it completely explicit to gcc-3.x
2126 * that these two variables' stack usage should be
2127 * combined, not added together.
2130 struct kvm_pit_state ps
;
2131 struct kvm_memory_alias alias
;
2132 struct kvm_pit_config pit_config
;
2136 case KVM_SET_TSS_ADDR
:
2137 r
= kvm_vm_ioctl_set_tss_addr(kvm
, arg
);
2141 case KVM_SET_MEMORY_REGION
: {
2142 struct kvm_memory_region kvm_mem
;
2143 struct kvm_userspace_memory_region kvm_userspace_mem
;
2146 if (copy_from_user(&kvm_mem
, argp
, sizeof kvm_mem
))
2148 kvm_userspace_mem
.slot
= kvm_mem
.slot
;
2149 kvm_userspace_mem
.flags
= kvm_mem
.flags
;
2150 kvm_userspace_mem
.guest_phys_addr
= kvm_mem
.guest_phys_addr
;
2151 kvm_userspace_mem
.memory_size
= kvm_mem
.memory_size
;
2152 r
= kvm_vm_ioctl_set_memory_region(kvm
, &kvm_userspace_mem
, 0);
2157 case KVM_SET_NR_MMU_PAGES
:
2158 r
= kvm_vm_ioctl_set_nr_mmu_pages(kvm
, arg
);
2162 case KVM_GET_NR_MMU_PAGES
:
2163 r
= kvm_vm_ioctl_get_nr_mmu_pages(kvm
);
2165 case KVM_SET_MEMORY_ALIAS
:
2167 if (copy_from_user(&u
.alias
, argp
, sizeof(struct kvm_memory_alias
)))
2169 r
= kvm_vm_ioctl_set_memory_alias(kvm
, &u
.alias
);
2173 case KVM_CREATE_IRQCHIP
:
2175 kvm
->arch
.vpic
= kvm_create_pic(kvm
);
2176 if (kvm
->arch
.vpic
) {
2177 r
= kvm_ioapic_init(kvm
);
2179 kfree(kvm
->arch
.vpic
);
2180 kvm
->arch
.vpic
= NULL
;
2185 r
= kvm_setup_default_irq_routing(kvm
);
2187 kfree(kvm
->arch
.vpic
);
2188 kfree(kvm
->arch
.vioapic
);
2192 case KVM_CREATE_PIT
:
2193 u
.pit_config
.flags
= KVM_PIT_SPEAKER_DUMMY
;
2195 case KVM_CREATE_PIT2
:
2197 if (copy_from_user(&u
.pit_config
, argp
,
2198 sizeof(struct kvm_pit_config
)))
2201 down_write(&kvm
->slots_lock
);
2204 goto create_pit_unlock
;
2206 kvm
->arch
.vpit
= kvm_create_pit(kvm
, u
.pit_config
.flags
);
2210 up_write(&kvm
->slots_lock
);
2212 case KVM_IRQ_LINE_STATUS
:
2213 case KVM_IRQ_LINE
: {
2214 struct kvm_irq_level irq_event
;
2217 if (copy_from_user(&irq_event
, argp
, sizeof irq_event
))
2219 if (irqchip_in_kernel(kvm
)) {
2221 mutex_lock(&kvm
->irq_lock
);
2222 status
= kvm_set_irq(kvm
, KVM_USERSPACE_IRQ_SOURCE_ID
,
2223 irq_event
.irq
, irq_event
.level
);
2224 mutex_unlock(&kvm
->irq_lock
);
2225 if (ioctl
== KVM_IRQ_LINE_STATUS
) {
2226 irq_event
.status
= status
;
2227 if (copy_to_user(argp
, &irq_event
,
2235 case KVM_GET_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 get_irqchip_out
;
2246 if (!irqchip_in_kernel(kvm
))
2247 goto get_irqchip_out
;
2248 r
= kvm_vm_ioctl_get_irqchip(kvm
, chip
);
2250 goto get_irqchip_out
;
2252 if (copy_to_user(argp
, chip
, sizeof *chip
))
2253 goto get_irqchip_out
;
2261 case KVM_SET_IRQCHIP
: {
2262 /* 0: PIC master, 1: PIC slave, 2: IOAPIC */
2263 struct kvm_irqchip
*chip
= kmalloc(sizeof(*chip
), GFP_KERNEL
);
2269 if (copy_from_user(chip
, argp
, sizeof *chip
))
2270 goto set_irqchip_out
;
2272 if (!irqchip_in_kernel(kvm
))
2273 goto set_irqchip_out
;
2274 r
= kvm_vm_ioctl_set_irqchip(kvm
, chip
);
2276 goto set_irqchip_out
;
2286 if (copy_from_user(&u
.ps
, argp
, sizeof(struct kvm_pit_state
)))
2289 if (!kvm
->arch
.vpit
)
2291 r
= kvm_vm_ioctl_get_pit(kvm
, &u
.ps
);
2295 if (copy_to_user(argp
, &u
.ps
, sizeof(struct kvm_pit_state
)))
2302 if (copy_from_user(&u
.ps
, argp
, sizeof u
.ps
))
2305 if (!kvm
->arch
.vpit
)
2307 r
= kvm_vm_ioctl_set_pit(kvm
, &u
.ps
);
2313 case KVM_REINJECT_CONTROL
: {
2314 struct kvm_reinject_control control
;
2316 if (copy_from_user(&control
, argp
, sizeof(control
)))
2318 r
= kvm_vm_ioctl_reinject(kvm
, &control
);
2331 static void kvm_init_msr_list(void)
2336 for (i
= j
= 0; i
< ARRAY_SIZE(msrs_to_save
); i
++) {
2337 if (rdmsr_safe(msrs_to_save
[i
], &dummy
[0], &dummy
[1]) < 0)
2340 msrs_to_save
[j
] = msrs_to_save
[i
];
2343 num_msrs_to_save
= j
;
2346 static int vcpu_mmio_write(struct kvm_vcpu
*vcpu
, gpa_t addr
, int len
,
2349 if (vcpu
->arch
.apic
&&
2350 !kvm_iodevice_write(&vcpu
->arch
.apic
->dev
, addr
, len
, v
))
2353 return kvm_io_bus_write(&vcpu
->kvm
->mmio_bus
, addr
, len
, v
);
2356 static int vcpu_mmio_read(struct kvm_vcpu
*vcpu
, gpa_t addr
, int len
, void *v
)
2358 if (vcpu
->arch
.apic
&&
2359 !kvm_iodevice_read(&vcpu
->arch
.apic
->dev
, addr
, len
, v
))
2362 return kvm_io_bus_read(&vcpu
->kvm
->mmio_bus
, addr
, len
, v
);
2365 static int kvm_read_guest_virt(gva_t addr
, void *val
, unsigned int bytes
,
2366 struct kvm_vcpu
*vcpu
)
2369 int r
= X86EMUL_CONTINUE
;
2372 gpa_t gpa
= vcpu
->arch
.mmu
.gva_to_gpa(vcpu
, addr
);
2373 unsigned offset
= addr
& (PAGE_SIZE
-1);
2374 unsigned toread
= min(bytes
, (unsigned)PAGE_SIZE
- offset
);
2377 if (gpa
== UNMAPPED_GVA
) {
2378 r
= X86EMUL_PROPAGATE_FAULT
;
2381 ret
= kvm_read_guest(vcpu
->kvm
, gpa
, data
, toread
);
2383 r
= X86EMUL_UNHANDLEABLE
;
2395 static int kvm_write_guest_virt(gva_t addr
, void *val
, unsigned int bytes
,
2396 struct kvm_vcpu
*vcpu
)
2399 int r
= X86EMUL_CONTINUE
;
2402 gpa_t gpa
= vcpu
->arch
.mmu
.gva_to_gpa(vcpu
, addr
);
2403 unsigned offset
= addr
& (PAGE_SIZE
-1);
2404 unsigned towrite
= min(bytes
, (unsigned)PAGE_SIZE
- offset
);
2407 if (gpa
== UNMAPPED_GVA
) {
2408 r
= X86EMUL_PROPAGATE_FAULT
;
2411 ret
= kvm_write_guest(vcpu
->kvm
, gpa
, data
, towrite
);
2413 r
= X86EMUL_UNHANDLEABLE
;
2426 static int emulator_read_emulated(unsigned long addr
,
2429 struct kvm_vcpu
*vcpu
)
2433 if (vcpu
->mmio_read_completed
) {
2434 memcpy(val
, vcpu
->mmio_data
, bytes
);
2435 trace_kvm_mmio(KVM_TRACE_MMIO_READ
, bytes
,
2436 vcpu
->mmio_phys_addr
, *(u64
*)val
);
2437 vcpu
->mmio_read_completed
= 0;
2438 return X86EMUL_CONTINUE
;
2441 gpa
= vcpu
->arch
.mmu
.gva_to_gpa(vcpu
, addr
);
2443 /* For APIC access vmexit */
2444 if ((gpa
& PAGE_MASK
) == APIC_DEFAULT_PHYS_BASE
)
2447 if (kvm_read_guest_virt(addr
, val
, bytes
, vcpu
)
2448 == X86EMUL_CONTINUE
)
2449 return X86EMUL_CONTINUE
;
2450 if (gpa
== UNMAPPED_GVA
)
2451 return X86EMUL_PROPAGATE_FAULT
;
2455 * Is this MMIO handled locally?
2457 if (!vcpu_mmio_read(vcpu
, gpa
, bytes
, val
)) {
2458 trace_kvm_mmio(KVM_TRACE_MMIO_READ
, bytes
, gpa
, *(u64
*)val
);
2459 return X86EMUL_CONTINUE
;
2462 trace_kvm_mmio(KVM_TRACE_MMIO_READ_UNSATISFIED
, bytes
, gpa
, 0);
2464 vcpu
->mmio_needed
= 1;
2465 vcpu
->mmio_phys_addr
= gpa
;
2466 vcpu
->mmio_size
= bytes
;
2467 vcpu
->mmio_is_write
= 0;
2469 return X86EMUL_UNHANDLEABLE
;
2472 int emulator_write_phys(struct kvm_vcpu
*vcpu
, gpa_t gpa
,
2473 const void *val
, int bytes
)
2477 ret
= kvm_write_guest(vcpu
->kvm
, gpa
, val
, bytes
);
2480 kvm_mmu_pte_write(vcpu
, gpa
, val
, bytes
, 1);
2484 static int emulator_write_emulated_onepage(unsigned long addr
,
2487 struct kvm_vcpu
*vcpu
)
2491 gpa
= vcpu
->arch
.mmu
.gva_to_gpa(vcpu
, addr
);
2493 if (gpa
== UNMAPPED_GVA
) {
2494 kvm_inject_page_fault(vcpu
, addr
, 2);
2495 return X86EMUL_PROPAGATE_FAULT
;
2498 /* For APIC access vmexit */
2499 if ((gpa
& PAGE_MASK
) == APIC_DEFAULT_PHYS_BASE
)
2502 if (emulator_write_phys(vcpu
, gpa
, val
, bytes
))
2503 return X86EMUL_CONTINUE
;
2506 trace_kvm_mmio(KVM_TRACE_MMIO_WRITE
, bytes
, gpa
, *(u64
*)val
);
2508 * Is this MMIO handled locally?
2510 if (!vcpu_mmio_write(vcpu
, gpa
, bytes
, val
))
2511 return X86EMUL_CONTINUE
;
2513 vcpu
->mmio_needed
= 1;
2514 vcpu
->mmio_phys_addr
= gpa
;
2515 vcpu
->mmio_size
= bytes
;
2516 vcpu
->mmio_is_write
= 1;
2517 memcpy(vcpu
->mmio_data
, val
, bytes
);
2519 return X86EMUL_CONTINUE
;
2522 int emulator_write_emulated(unsigned long addr
,
2525 struct kvm_vcpu
*vcpu
)
2527 /* Crossing a page boundary? */
2528 if (((addr
+ bytes
- 1) ^ addr
) & PAGE_MASK
) {
2531 now
= -addr
& ~PAGE_MASK
;
2532 rc
= emulator_write_emulated_onepage(addr
, val
, now
, vcpu
);
2533 if (rc
!= X86EMUL_CONTINUE
)
2539 return emulator_write_emulated_onepage(addr
, val
, bytes
, vcpu
);
2541 EXPORT_SYMBOL_GPL(emulator_write_emulated
);
2543 static int emulator_cmpxchg_emulated(unsigned long addr
,
2547 struct kvm_vcpu
*vcpu
)
2549 static int reported
;
2553 printk(KERN_WARNING
"kvm: emulating exchange as write\n");
2555 #ifndef CONFIG_X86_64
2556 /* guests cmpxchg8b have to be emulated atomically */
2563 gpa
= vcpu
->arch
.mmu
.gva_to_gpa(vcpu
, addr
);
2565 if (gpa
== UNMAPPED_GVA
||
2566 (gpa
& PAGE_MASK
) == APIC_DEFAULT_PHYS_BASE
)
2569 if (((gpa
+ bytes
- 1) & PAGE_MASK
) != (gpa
& PAGE_MASK
))
2574 page
= gfn_to_page(vcpu
->kvm
, gpa
>> PAGE_SHIFT
);
2576 kaddr
= kmap_atomic(page
, KM_USER0
);
2577 set_64bit((u64
*)(kaddr
+ offset_in_page(gpa
)), val
);
2578 kunmap_atomic(kaddr
, KM_USER0
);
2579 kvm_release_page_dirty(page
);
2584 return emulator_write_emulated(addr
, new, bytes
, vcpu
);
2587 static unsigned long get_segment_base(struct kvm_vcpu
*vcpu
, int seg
)
2589 return kvm_x86_ops
->get_segment_base(vcpu
, seg
);
2592 int emulate_invlpg(struct kvm_vcpu
*vcpu
, gva_t address
)
2594 kvm_mmu_invlpg(vcpu
, address
);
2595 return X86EMUL_CONTINUE
;
2598 int emulate_clts(struct kvm_vcpu
*vcpu
)
2600 kvm_x86_ops
->set_cr0(vcpu
, vcpu
->arch
.cr0
& ~X86_CR0_TS
);
2601 return X86EMUL_CONTINUE
;
2604 int emulator_get_dr(struct x86_emulate_ctxt
*ctxt
, int dr
, unsigned long *dest
)
2606 struct kvm_vcpu
*vcpu
= ctxt
->vcpu
;
2610 *dest
= kvm_x86_ops
->get_dr(vcpu
, dr
);
2611 return X86EMUL_CONTINUE
;
2613 pr_unimpl(vcpu
, "%s: unexpected dr %u\n", __func__
, dr
);
2614 return X86EMUL_UNHANDLEABLE
;
2618 int emulator_set_dr(struct x86_emulate_ctxt
*ctxt
, int dr
, unsigned long value
)
2620 unsigned long mask
= (ctxt
->mode
== X86EMUL_MODE_PROT64
) ? ~0ULL : ~0U;
2623 kvm_x86_ops
->set_dr(ctxt
->vcpu
, dr
, value
& mask
, &exception
);
2625 /* FIXME: better handling */
2626 return X86EMUL_UNHANDLEABLE
;
2628 return X86EMUL_CONTINUE
;
2631 void kvm_report_emulation_failure(struct kvm_vcpu
*vcpu
, const char *context
)
2634 unsigned long rip
= kvm_rip_read(vcpu
);
2635 unsigned long rip_linear
;
2637 if (!printk_ratelimit())
2640 rip_linear
= rip
+ get_segment_base(vcpu
, VCPU_SREG_CS
);
2642 kvm_read_guest_virt(rip_linear
, (void *)opcodes
, 4, vcpu
);
2644 printk(KERN_ERR
"emulation failed (%s) rip %lx %02x %02x %02x %02x\n",
2645 context
, rip
, opcodes
[0], opcodes
[1], opcodes
[2], opcodes
[3]);
2647 EXPORT_SYMBOL_GPL(kvm_report_emulation_failure
);
2649 static struct x86_emulate_ops emulate_ops
= {
2650 .read_std
= kvm_read_guest_virt
,
2651 .read_emulated
= emulator_read_emulated
,
2652 .write_emulated
= emulator_write_emulated
,
2653 .cmpxchg_emulated
= emulator_cmpxchg_emulated
,
2656 static void cache_all_regs(struct kvm_vcpu
*vcpu
)
2658 kvm_register_read(vcpu
, VCPU_REGS_RAX
);
2659 kvm_register_read(vcpu
, VCPU_REGS_RSP
);
2660 kvm_register_read(vcpu
, VCPU_REGS_RIP
);
2661 vcpu
->arch
.regs_dirty
= ~0;
2664 int emulate_instruction(struct kvm_vcpu
*vcpu
,
2665 struct kvm_run
*run
,
2671 struct decode_cache
*c
;
2673 kvm_clear_exception_queue(vcpu
);
2674 vcpu
->arch
.mmio_fault_cr2
= cr2
;
2676 * TODO: fix x86_emulate.c to use guest_read/write_register
2677 * instead of direct ->regs accesses, can save hundred cycles
2678 * on Intel for instructions that don't read/change RSP, for
2681 cache_all_regs(vcpu
);
2683 vcpu
->mmio_is_write
= 0;
2684 vcpu
->arch
.pio
.string
= 0;
2686 if (!(emulation_type
& EMULTYPE_NO_DECODE
)) {
2688 kvm_x86_ops
->get_cs_db_l_bits(vcpu
, &cs_db
, &cs_l
);
2690 vcpu
->arch
.emulate_ctxt
.vcpu
= vcpu
;
2691 vcpu
->arch
.emulate_ctxt
.eflags
= kvm_x86_ops
->get_rflags(vcpu
);
2692 vcpu
->arch
.emulate_ctxt
.mode
=
2693 (vcpu
->arch
.emulate_ctxt
.eflags
& X86_EFLAGS_VM
)
2694 ? X86EMUL_MODE_REAL
: cs_l
2695 ? X86EMUL_MODE_PROT64
: cs_db
2696 ? X86EMUL_MODE_PROT32
: X86EMUL_MODE_PROT16
;
2698 r
= x86_decode_insn(&vcpu
->arch
.emulate_ctxt
, &emulate_ops
);
2700 /* Only allow emulation of specific instructions on #UD
2701 * (namely VMMCALL, sysenter, sysexit, syscall)*/
2702 c
= &vcpu
->arch
.emulate_ctxt
.decode
;
2703 if (emulation_type
& EMULTYPE_TRAP_UD
) {
2705 return EMULATE_FAIL
;
2707 case 0x01: /* VMMCALL */
2708 if (c
->modrm_mod
!= 3 || c
->modrm_rm
!= 1)
2709 return EMULATE_FAIL
;
2711 case 0x34: /* sysenter */
2712 case 0x35: /* sysexit */
2713 if (c
->modrm_mod
!= 0 || c
->modrm_rm
!= 0)
2714 return EMULATE_FAIL
;
2716 case 0x05: /* syscall */
2717 if (c
->modrm_mod
!= 0 || c
->modrm_rm
!= 0)
2718 return EMULATE_FAIL
;
2721 return EMULATE_FAIL
;
2724 if (!(c
->modrm_reg
== 0 || c
->modrm_reg
== 3))
2725 return EMULATE_FAIL
;
2728 ++vcpu
->stat
.insn_emulation
;
2730 ++vcpu
->stat
.insn_emulation_fail
;
2731 if (kvm_mmu_unprotect_page_virt(vcpu
, cr2
))
2732 return EMULATE_DONE
;
2733 return EMULATE_FAIL
;
2737 if (emulation_type
& EMULTYPE_SKIP
) {
2738 kvm_rip_write(vcpu
, vcpu
->arch
.emulate_ctxt
.decode
.eip
);
2739 return EMULATE_DONE
;
2742 r
= x86_emulate_insn(&vcpu
->arch
.emulate_ctxt
, &emulate_ops
);
2743 shadow_mask
= vcpu
->arch
.emulate_ctxt
.interruptibility
;
2746 kvm_x86_ops
->set_interrupt_shadow(vcpu
, shadow_mask
);
2748 if (vcpu
->arch
.pio
.string
)
2749 return EMULATE_DO_MMIO
;
2751 if ((r
|| vcpu
->mmio_is_write
) && run
) {
2752 run
->exit_reason
= KVM_EXIT_MMIO
;
2753 run
->mmio
.phys_addr
= vcpu
->mmio_phys_addr
;
2754 memcpy(run
->mmio
.data
, vcpu
->mmio_data
, 8);
2755 run
->mmio
.len
= vcpu
->mmio_size
;
2756 run
->mmio
.is_write
= vcpu
->mmio_is_write
;
2760 if (kvm_mmu_unprotect_page_virt(vcpu
, cr2
))
2761 return EMULATE_DONE
;
2762 if (!vcpu
->mmio_needed
) {
2763 kvm_report_emulation_failure(vcpu
, "mmio");
2764 return EMULATE_FAIL
;
2766 return EMULATE_DO_MMIO
;
2769 kvm_x86_ops
->set_rflags(vcpu
, vcpu
->arch
.emulate_ctxt
.eflags
);
2771 if (vcpu
->mmio_is_write
) {
2772 vcpu
->mmio_needed
= 0;
2773 return EMULATE_DO_MMIO
;
2776 return EMULATE_DONE
;
2778 EXPORT_SYMBOL_GPL(emulate_instruction
);
2780 static int pio_copy_data(struct kvm_vcpu
*vcpu
)
2782 void *p
= vcpu
->arch
.pio_data
;
2783 gva_t q
= vcpu
->arch
.pio
.guest_gva
;
2787 bytes
= vcpu
->arch
.pio
.size
* vcpu
->arch
.pio
.cur_count
;
2788 if (vcpu
->arch
.pio
.in
)
2789 ret
= kvm_write_guest_virt(q
, p
, bytes
, vcpu
);
2791 ret
= kvm_read_guest_virt(q
, p
, bytes
, vcpu
);
2795 int complete_pio(struct kvm_vcpu
*vcpu
)
2797 struct kvm_pio_request
*io
= &vcpu
->arch
.pio
;
2804 val
= kvm_register_read(vcpu
, VCPU_REGS_RAX
);
2805 memcpy(&val
, vcpu
->arch
.pio_data
, io
->size
);
2806 kvm_register_write(vcpu
, VCPU_REGS_RAX
, val
);
2810 r
= pio_copy_data(vcpu
);
2817 delta
*= io
->cur_count
;
2819 * The size of the register should really depend on
2820 * current address size.
2822 val
= kvm_register_read(vcpu
, VCPU_REGS_RCX
);
2824 kvm_register_write(vcpu
, VCPU_REGS_RCX
, val
);
2830 val
= kvm_register_read(vcpu
, VCPU_REGS_RDI
);
2832 kvm_register_write(vcpu
, VCPU_REGS_RDI
, val
);
2834 val
= kvm_register_read(vcpu
, VCPU_REGS_RSI
);
2836 kvm_register_write(vcpu
, VCPU_REGS_RSI
, val
);
2840 io
->count
-= io
->cur_count
;
2846 static int kernel_pio(struct kvm_vcpu
*vcpu
, void *pd
)
2848 /* TODO: String I/O for in kernel device */
2851 if (vcpu
->arch
.pio
.in
)
2852 r
= kvm_io_bus_read(&vcpu
->kvm
->pio_bus
, vcpu
->arch
.pio
.port
,
2853 vcpu
->arch
.pio
.size
, pd
);
2855 r
= kvm_io_bus_write(&vcpu
->kvm
->pio_bus
, vcpu
->arch
.pio
.port
,
2856 vcpu
->arch
.pio
.size
, pd
);
2860 static int pio_string_write(struct kvm_vcpu
*vcpu
)
2862 struct kvm_pio_request
*io
= &vcpu
->arch
.pio
;
2863 void *pd
= vcpu
->arch
.pio_data
;
2866 for (i
= 0; i
< io
->cur_count
; i
++) {
2867 if (kvm_io_bus_write(&vcpu
->kvm
->pio_bus
,
2868 io
->port
, io
->size
, pd
)) {
2877 int kvm_emulate_pio(struct kvm_vcpu
*vcpu
, struct kvm_run
*run
, int in
,
2878 int size
, unsigned port
)
2882 vcpu
->run
->exit_reason
= KVM_EXIT_IO
;
2883 vcpu
->run
->io
.direction
= in
? KVM_EXIT_IO_IN
: KVM_EXIT_IO_OUT
;
2884 vcpu
->run
->io
.size
= vcpu
->arch
.pio
.size
= size
;
2885 vcpu
->run
->io
.data_offset
= KVM_PIO_PAGE_OFFSET
* PAGE_SIZE
;
2886 vcpu
->run
->io
.count
= vcpu
->arch
.pio
.count
= vcpu
->arch
.pio
.cur_count
= 1;
2887 vcpu
->run
->io
.port
= vcpu
->arch
.pio
.port
= port
;
2888 vcpu
->arch
.pio
.in
= in
;
2889 vcpu
->arch
.pio
.string
= 0;
2890 vcpu
->arch
.pio
.down
= 0;
2891 vcpu
->arch
.pio
.rep
= 0;
2893 trace_kvm_pio(vcpu
->run
->io
.direction
== KVM_EXIT_IO_OUT
, port
,
2896 val
= kvm_register_read(vcpu
, VCPU_REGS_RAX
);
2897 memcpy(vcpu
->arch
.pio_data
, &val
, 4);
2899 if (!kernel_pio(vcpu
, vcpu
->arch
.pio_data
)) {
2905 EXPORT_SYMBOL_GPL(kvm_emulate_pio
);
2907 int kvm_emulate_pio_string(struct kvm_vcpu
*vcpu
, struct kvm_run
*run
, int in
,
2908 int size
, unsigned long count
, int down
,
2909 gva_t address
, int rep
, unsigned port
)
2911 unsigned now
, in_page
;
2914 vcpu
->run
->exit_reason
= KVM_EXIT_IO
;
2915 vcpu
->run
->io
.direction
= in
? KVM_EXIT_IO_IN
: KVM_EXIT_IO_OUT
;
2916 vcpu
->run
->io
.size
= vcpu
->arch
.pio
.size
= size
;
2917 vcpu
->run
->io
.data_offset
= KVM_PIO_PAGE_OFFSET
* PAGE_SIZE
;
2918 vcpu
->run
->io
.count
= vcpu
->arch
.pio
.count
= vcpu
->arch
.pio
.cur_count
= count
;
2919 vcpu
->run
->io
.port
= vcpu
->arch
.pio
.port
= port
;
2920 vcpu
->arch
.pio
.in
= in
;
2921 vcpu
->arch
.pio
.string
= 1;
2922 vcpu
->arch
.pio
.down
= down
;
2923 vcpu
->arch
.pio
.rep
= rep
;
2925 trace_kvm_pio(vcpu
->run
->io
.direction
== KVM_EXIT_IO_OUT
, port
,
2929 kvm_x86_ops
->skip_emulated_instruction(vcpu
);
2934 in_page
= PAGE_SIZE
- offset_in_page(address
);
2936 in_page
= offset_in_page(address
) + size
;
2937 now
= min(count
, (unsigned long)in_page
/ size
);
2942 * String I/O in reverse. Yuck. Kill the guest, fix later.
2944 pr_unimpl(vcpu
, "guest string pio down\n");
2945 kvm_inject_gp(vcpu
, 0);
2948 vcpu
->run
->io
.count
= now
;
2949 vcpu
->arch
.pio
.cur_count
= now
;
2951 if (vcpu
->arch
.pio
.cur_count
== vcpu
->arch
.pio
.count
)
2952 kvm_x86_ops
->skip_emulated_instruction(vcpu
);
2954 vcpu
->arch
.pio
.guest_gva
= address
;
2956 if (!vcpu
->arch
.pio
.in
) {
2957 /* string PIO write */
2958 ret
= pio_copy_data(vcpu
);
2959 if (ret
== X86EMUL_PROPAGATE_FAULT
) {
2960 kvm_inject_gp(vcpu
, 0);
2963 if (ret
== 0 && !pio_string_write(vcpu
)) {
2965 if (vcpu
->arch
.pio
.count
== 0)
2969 /* no string PIO read support yet */
2973 EXPORT_SYMBOL_GPL(kvm_emulate_pio_string
);
2975 static void bounce_off(void *info
)
2980 static unsigned int ref_freq
;
2981 static unsigned long tsc_khz_ref
;
2983 static int kvmclock_cpufreq_notifier(struct notifier_block
*nb
, unsigned long val
,
2986 struct cpufreq_freqs
*freq
= data
;
2988 struct kvm_vcpu
*vcpu
;
2989 int i
, send_ipi
= 0;
2992 ref_freq
= freq
->old
;
2994 if (val
== CPUFREQ_PRECHANGE
&& freq
->old
> freq
->new)
2996 if (val
== CPUFREQ_POSTCHANGE
&& freq
->old
< freq
->new)
2998 per_cpu(cpu_tsc_khz
, freq
->cpu
) = cpufreq_scale(tsc_khz_ref
, ref_freq
, freq
->new);
3000 spin_lock(&kvm_lock
);
3001 list_for_each_entry(kvm
, &vm_list
, vm_list
) {
3002 kvm_for_each_vcpu(i
, vcpu
, kvm
) {
3003 if (vcpu
->cpu
!= freq
->cpu
)
3005 if (!kvm_request_guest_time_update(vcpu
))
3007 if (vcpu
->cpu
!= smp_processor_id())
3011 spin_unlock(&kvm_lock
);
3013 if (freq
->old
< freq
->new && send_ipi
) {
3015 * We upscale the frequency. Must make the guest
3016 * doesn't see old kvmclock values while running with
3017 * the new frequency, otherwise we risk the guest sees
3018 * time go backwards.
3020 * In case we update the frequency for another cpu
3021 * (which might be in guest context) send an interrupt
3022 * to kick the cpu out of guest context. Next time
3023 * guest context is entered kvmclock will be updated,
3024 * so the guest will not see stale values.
3026 smp_call_function_single(freq
->cpu
, bounce_off
, NULL
, 1);
3031 static struct notifier_block kvmclock_cpufreq_notifier_block
= {
3032 .notifier_call
= kvmclock_cpufreq_notifier
3035 int kvm_arch_init(void *opaque
)
3038 struct kvm_x86_ops
*ops
= (struct kvm_x86_ops
*)opaque
;
3041 printk(KERN_ERR
"kvm: already loaded the other module\n");
3046 if (!ops
->cpu_has_kvm_support()) {
3047 printk(KERN_ERR
"kvm: no hardware support\n");
3051 if (ops
->disabled_by_bios()) {
3052 printk(KERN_ERR
"kvm: disabled by bios\n");
3057 r
= kvm_mmu_module_init();
3061 kvm_init_msr_list();
3064 kvm_mmu_set_nonpresent_ptes(0ull, 0ull);
3065 kvm_mmu_set_base_ptes(PT_PRESENT_MASK
);
3066 kvm_mmu_set_mask_ptes(PT_USER_MASK
, PT_ACCESSED_MASK
,
3067 PT_DIRTY_MASK
, PT64_NX_MASK
, 0);
3069 for_each_possible_cpu(cpu
)
3070 per_cpu(cpu_tsc_khz
, cpu
) = tsc_khz
;
3071 if (!boot_cpu_has(X86_FEATURE_CONSTANT_TSC
)) {
3072 tsc_khz_ref
= tsc_khz
;
3073 cpufreq_register_notifier(&kvmclock_cpufreq_notifier_block
,
3074 CPUFREQ_TRANSITION_NOTIFIER
);
3083 void kvm_arch_exit(void)
3085 if (!boot_cpu_has(X86_FEATURE_CONSTANT_TSC
))
3086 cpufreq_unregister_notifier(&kvmclock_cpufreq_notifier_block
,
3087 CPUFREQ_TRANSITION_NOTIFIER
);
3089 kvm_mmu_module_exit();
3092 int kvm_emulate_halt(struct kvm_vcpu
*vcpu
)
3094 ++vcpu
->stat
.halt_exits
;
3095 if (irqchip_in_kernel(vcpu
->kvm
)) {
3096 vcpu
->arch
.mp_state
= KVM_MP_STATE_HALTED
;
3099 vcpu
->run
->exit_reason
= KVM_EXIT_HLT
;
3103 EXPORT_SYMBOL_GPL(kvm_emulate_halt
);
3105 static inline gpa_t
hc_gpa(struct kvm_vcpu
*vcpu
, unsigned long a0
,
3108 if (is_long_mode(vcpu
))
3111 return a0
| ((gpa_t
)a1
<< 32);
3114 int kvm_emulate_hypercall(struct kvm_vcpu
*vcpu
)
3116 unsigned long nr
, a0
, a1
, a2
, a3
, ret
;
3119 nr
= kvm_register_read(vcpu
, VCPU_REGS_RAX
);
3120 a0
= kvm_register_read(vcpu
, VCPU_REGS_RBX
);
3121 a1
= kvm_register_read(vcpu
, VCPU_REGS_RCX
);
3122 a2
= kvm_register_read(vcpu
, VCPU_REGS_RDX
);
3123 a3
= kvm_register_read(vcpu
, VCPU_REGS_RSI
);
3125 trace_kvm_hypercall(nr
, a0
, a1
, a2
, a3
);
3127 if (!is_long_mode(vcpu
)) {
3136 case KVM_HC_VAPIC_POLL_IRQ
:
3140 r
= kvm_pv_mmu_op(vcpu
, a0
, hc_gpa(vcpu
, a1
, a2
), &ret
);
3146 kvm_register_write(vcpu
, VCPU_REGS_RAX
, ret
);
3147 ++vcpu
->stat
.hypercalls
;
3150 EXPORT_SYMBOL_GPL(kvm_emulate_hypercall
);
3152 int kvm_fix_hypercall(struct kvm_vcpu
*vcpu
)
3154 char instruction
[3];
3156 unsigned long rip
= kvm_rip_read(vcpu
);
3160 * Blow out the MMU to ensure that no other VCPU has an active mapping
3161 * to ensure that the updated hypercall appears atomically across all
3164 kvm_mmu_zap_all(vcpu
->kvm
);
3166 kvm_x86_ops
->patch_hypercall(vcpu
, instruction
);
3167 if (emulator_write_emulated(rip
, instruction
, 3, vcpu
)
3168 != X86EMUL_CONTINUE
)
3174 static u64
mk_cr_64(u64 curr_cr
, u32 new_val
)
3176 return (curr_cr
& ~((1ULL << 32) - 1)) | new_val
;
3179 void realmode_lgdt(struct kvm_vcpu
*vcpu
, u16 limit
, unsigned long base
)
3181 struct descriptor_table dt
= { limit
, base
};
3183 kvm_x86_ops
->set_gdt(vcpu
, &dt
);
3186 void realmode_lidt(struct kvm_vcpu
*vcpu
, u16 limit
, unsigned long base
)
3188 struct descriptor_table dt
= { limit
, base
};
3190 kvm_x86_ops
->set_idt(vcpu
, &dt
);
3193 void realmode_lmsw(struct kvm_vcpu
*vcpu
, unsigned long msw
,
3194 unsigned long *rflags
)
3196 kvm_lmsw(vcpu
, msw
);
3197 *rflags
= kvm_x86_ops
->get_rflags(vcpu
);
3200 unsigned long realmode_get_cr(struct kvm_vcpu
*vcpu
, int cr
)
3202 unsigned long value
;
3204 kvm_x86_ops
->decache_cr4_guest_bits(vcpu
);
3207 value
= vcpu
->arch
.cr0
;
3210 value
= vcpu
->arch
.cr2
;
3213 value
= vcpu
->arch
.cr3
;
3216 value
= vcpu
->arch
.cr4
;
3219 value
= kvm_get_cr8(vcpu
);
3222 vcpu_printf(vcpu
, "%s: unexpected cr %u\n", __func__
, cr
);
3229 void realmode_set_cr(struct kvm_vcpu
*vcpu
, int cr
, unsigned long val
,
3230 unsigned long *rflags
)
3234 kvm_set_cr0(vcpu
, mk_cr_64(vcpu
->arch
.cr0
, val
));
3235 *rflags
= kvm_x86_ops
->get_rflags(vcpu
);
3238 vcpu
->arch
.cr2
= val
;
3241 kvm_set_cr3(vcpu
, val
);
3244 kvm_set_cr4(vcpu
, mk_cr_64(vcpu
->arch
.cr4
, val
));
3247 kvm_set_cr8(vcpu
, val
& 0xfUL
);
3250 vcpu_printf(vcpu
, "%s: unexpected cr %u\n", __func__
, cr
);
3254 static int move_to_next_stateful_cpuid_entry(struct kvm_vcpu
*vcpu
, int i
)
3256 struct kvm_cpuid_entry2
*e
= &vcpu
->arch
.cpuid_entries
[i
];
3257 int j
, nent
= vcpu
->arch
.cpuid_nent
;
3259 e
->flags
&= ~KVM_CPUID_FLAG_STATE_READ_NEXT
;
3260 /* when no next entry is found, the current entry[i] is reselected */
3261 for (j
= i
+ 1; ; j
= (j
+ 1) % nent
) {
3262 struct kvm_cpuid_entry2
*ej
= &vcpu
->arch
.cpuid_entries
[j
];
3263 if (ej
->function
== e
->function
) {
3264 ej
->flags
|= KVM_CPUID_FLAG_STATE_READ_NEXT
;
3268 return 0; /* silence gcc, even though control never reaches here */
3271 /* find an entry with matching function, matching index (if needed), and that
3272 * should be read next (if it's stateful) */
3273 static int is_matching_cpuid_entry(struct kvm_cpuid_entry2
*e
,
3274 u32 function
, u32 index
)
3276 if (e
->function
!= function
)
3278 if ((e
->flags
& KVM_CPUID_FLAG_SIGNIFCANT_INDEX
) && e
->index
!= index
)
3280 if ((e
->flags
& KVM_CPUID_FLAG_STATEFUL_FUNC
) &&
3281 !(e
->flags
& KVM_CPUID_FLAG_STATE_READ_NEXT
))
3286 struct kvm_cpuid_entry2
*kvm_find_cpuid_entry(struct kvm_vcpu
*vcpu
,
3287 u32 function
, u32 index
)
3290 struct kvm_cpuid_entry2
*best
= NULL
;
3292 for (i
= 0; i
< vcpu
->arch
.cpuid_nent
; ++i
) {
3293 struct kvm_cpuid_entry2
*e
;
3295 e
= &vcpu
->arch
.cpuid_entries
[i
];
3296 if (is_matching_cpuid_entry(e
, function
, index
)) {
3297 if (e
->flags
& KVM_CPUID_FLAG_STATEFUL_FUNC
)
3298 move_to_next_stateful_cpuid_entry(vcpu
, i
);
3303 * Both basic or both extended?
3305 if (((e
->function
^ function
) & 0x80000000) == 0)
3306 if (!best
|| e
->function
> best
->function
)
3312 int cpuid_maxphyaddr(struct kvm_vcpu
*vcpu
)
3314 struct kvm_cpuid_entry2
*best
;
3316 best
= kvm_find_cpuid_entry(vcpu
, 0x80000008, 0);
3318 return best
->eax
& 0xff;
3322 void kvm_emulate_cpuid(struct kvm_vcpu
*vcpu
)
3324 u32 function
, index
;
3325 struct kvm_cpuid_entry2
*best
;
3327 function
= kvm_register_read(vcpu
, VCPU_REGS_RAX
);
3328 index
= kvm_register_read(vcpu
, VCPU_REGS_RCX
);
3329 kvm_register_write(vcpu
, VCPU_REGS_RAX
, 0);
3330 kvm_register_write(vcpu
, VCPU_REGS_RBX
, 0);
3331 kvm_register_write(vcpu
, VCPU_REGS_RCX
, 0);
3332 kvm_register_write(vcpu
, VCPU_REGS_RDX
, 0);
3333 best
= kvm_find_cpuid_entry(vcpu
, function
, index
);
3335 kvm_register_write(vcpu
, VCPU_REGS_RAX
, best
->eax
);
3336 kvm_register_write(vcpu
, VCPU_REGS_RBX
, best
->ebx
);
3337 kvm_register_write(vcpu
, VCPU_REGS_RCX
, best
->ecx
);
3338 kvm_register_write(vcpu
, VCPU_REGS_RDX
, best
->edx
);
3340 kvm_x86_ops
->skip_emulated_instruction(vcpu
);
3341 trace_kvm_cpuid(function
,
3342 kvm_register_read(vcpu
, VCPU_REGS_RAX
),
3343 kvm_register_read(vcpu
, VCPU_REGS_RBX
),
3344 kvm_register_read(vcpu
, VCPU_REGS_RCX
),
3345 kvm_register_read(vcpu
, VCPU_REGS_RDX
));
3347 EXPORT_SYMBOL_GPL(kvm_emulate_cpuid
);
3350 * Check if userspace requested an interrupt window, and that the
3351 * interrupt window is open.
3353 * No need to exit to userspace if we already have an interrupt queued.
3355 static int dm_request_for_irq_injection(struct kvm_vcpu
*vcpu
,
3356 struct kvm_run
*kvm_run
)
3358 return (!irqchip_in_kernel(vcpu
->kvm
) && !kvm_cpu_has_interrupt(vcpu
) &&
3359 kvm_run
->request_interrupt_window
&&
3360 kvm_arch_interrupt_allowed(vcpu
));
3363 static void post_kvm_run_save(struct kvm_vcpu
*vcpu
,
3364 struct kvm_run
*kvm_run
)
3366 kvm_run
->if_flag
= (kvm_x86_ops
->get_rflags(vcpu
) & X86_EFLAGS_IF
) != 0;
3367 kvm_run
->cr8
= kvm_get_cr8(vcpu
);
3368 kvm_run
->apic_base
= kvm_get_apic_base(vcpu
);
3369 if (irqchip_in_kernel(vcpu
->kvm
))
3370 kvm_run
->ready_for_interrupt_injection
= 1;
3372 kvm_run
->ready_for_interrupt_injection
=
3373 kvm_arch_interrupt_allowed(vcpu
) &&
3374 !kvm_cpu_has_interrupt(vcpu
) &&
3375 !kvm_event_needs_reinjection(vcpu
);
3378 static void vapic_enter(struct kvm_vcpu
*vcpu
)
3380 struct kvm_lapic
*apic
= vcpu
->arch
.apic
;
3383 if (!apic
|| !apic
->vapic_addr
)
3386 page
= gfn_to_page(vcpu
->kvm
, apic
->vapic_addr
>> PAGE_SHIFT
);
3388 vcpu
->arch
.apic
->vapic_page
= page
;
3391 static void vapic_exit(struct kvm_vcpu
*vcpu
)
3393 struct kvm_lapic
*apic
= vcpu
->arch
.apic
;
3395 if (!apic
|| !apic
->vapic_addr
)
3398 down_read(&vcpu
->kvm
->slots_lock
);
3399 kvm_release_page_dirty(apic
->vapic_page
);
3400 mark_page_dirty(vcpu
->kvm
, apic
->vapic_addr
>> PAGE_SHIFT
);
3401 up_read(&vcpu
->kvm
->slots_lock
);
3404 static void update_cr8_intercept(struct kvm_vcpu
*vcpu
)
3408 if (!kvm_x86_ops
->update_cr8_intercept
)
3411 if (!vcpu
->arch
.apic
->vapic_addr
)
3412 max_irr
= kvm_lapic_find_highest_irr(vcpu
);
3419 tpr
= kvm_lapic_get_cr8(vcpu
);
3421 kvm_x86_ops
->update_cr8_intercept(vcpu
, tpr
, max_irr
);
3424 static void inject_pending_irq(struct kvm_vcpu
*vcpu
, struct kvm_run
*kvm_run
)
3426 /* try to reinject previous events if any */
3427 if (vcpu
->arch
.nmi_injected
) {
3428 kvm_x86_ops
->set_nmi(vcpu
);
3432 if (vcpu
->arch
.interrupt
.pending
) {
3433 kvm_x86_ops
->set_irq(vcpu
);
3437 /* try to inject new event if pending */
3438 if (vcpu
->arch
.nmi_pending
) {
3439 if (kvm_x86_ops
->nmi_allowed(vcpu
)) {
3440 vcpu
->arch
.nmi_pending
= false;
3441 vcpu
->arch
.nmi_injected
= true;
3442 kvm_x86_ops
->set_nmi(vcpu
);
3444 } else if (kvm_cpu_has_interrupt(vcpu
)) {
3445 if (kvm_x86_ops
->interrupt_allowed(vcpu
)) {
3446 kvm_queue_interrupt(vcpu
, kvm_cpu_get_interrupt(vcpu
),
3448 kvm_x86_ops
->set_irq(vcpu
);
3453 static int vcpu_enter_guest(struct kvm_vcpu
*vcpu
, struct kvm_run
*kvm_run
)
3456 bool req_int_win
= !irqchip_in_kernel(vcpu
->kvm
) &&
3457 kvm_run
->request_interrupt_window
;
3460 if (test_and_clear_bit(KVM_REQ_MMU_RELOAD
, &vcpu
->requests
))
3461 kvm_mmu_unload(vcpu
);
3463 r
= kvm_mmu_reload(vcpu
);
3467 if (vcpu
->requests
) {
3468 if (test_and_clear_bit(KVM_REQ_MIGRATE_TIMER
, &vcpu
->requests
))
3469 __kvm_migrate_timers(vcpu
);
3470 if (test_and_clear_bit(KVM_REQ_KVMCLOCK_UPDATE
, &vcpu
->requests
))
3471 kvm_write_guest_time(vcpu
);
3472 if (test_and_clear_bit(KVM_REQ_MMU_SYNC
, &vcpu
->requests
))
3473 kvm_mmu_sync_roots(vcpu
);
3474 if (test_and_clear_bit(KVM_REQ_TLB_FLUSH
, &vcpu
->requests
))
3475 kvm_x86_ops
->tlb_flush(vcpu
);
3476 if (test_and_clear_bit(KVM_REQ_REPORT_TPR_ACCESS
,
3478 kvm_run
->exit_reason
= KVM_EXIT_TPR_ACCESS
;
3482 if (test_and_clear_bit(KVM_REQ_TRIPLE_FAULT
, &vcpu
->requests
)) {
3483 kvm_run
->exit_reason
= KVM_EXIT_SHUTDOWN
;
3491 kvm_x86_ops
->prepare_guest_switch(vcpu
);
3492 kvm_load_guest_fpu(vcpu
);
3494 local_irq_disable();
3496 clear_bit(KVM_REQ_KICK
, &vcpu
->requests
);
3497 smp_mb__after_clear_bit();
3499 if (vcpu
->requests
|| need_resched() || signal_pending(current
)) {
3506 if (vcpu
->arch
.exception
.pending
)
3507 __queue_exception(vcpu
);
3509 inject_pending_irq(vcpu
, kvm_run
);
3511 /* enable NMI/IRQ window open exits if needed */
3512 if (vcpu
->arch
.nmi_pending
)
3513 kvm_x86_ops
->enable_nmi_window(vcpu
);
3514 else if (kvm_cpu_has_interrupt(vcpu
) || req_int_win
)
3515 kvm_x86_ops
->enable_irq_window(vcpu
);
3517 if (kvm_lapic_enabled(vcpu
)) {
3518 update_cr8_intercept(vcpu
);
3519 kvm_lapic_sync_to_vapic(vcpu
);
3522 up_read(&vcpu
->kvm
->slots_lock
);
3526 get_debugreg(vcpu
->arch
.host_dr6
, 6);
3527 get_debugreg(vcpu
->arch
.host_dr7
, 7);
3528 if (unlikely(vcpu
->arch
.switch_db_regs
)) {
3529 get_debugreg(vcpu
->arch
.host_db
[0], 0);
3530 get_debugreg(vcpu
->arch
.host_db
[1], 1);
3531 get_debugreg(vcpu
->arch
.host_db
[2], 2);
3532 get_debugreg(vcpu
->arch
.host_db
[3], 3);
3535 set_debugreg(vcpu
->arch
.eff_db
[0], 0);
3536 set_debugreg(vcpu
->arch
.eff_db
[1], 1);
3537 set_debugreg(vcpu
->arch
.eff_db
[2], 2);
3538 set_debugreg(vcpu
->arch
.eff_db
[3], 3);
3541 trace_kvm_entry(vcpu
->vcpu_id
);
3542 kvm_x86_ops
->run(vcpu
, kvm_run
);
3544 if (unlikely(vcpu
->arch
.switch_db_regs
)) {
3546 set_debugreg(vcpu
->arch
.host_db
[0], 0);
3547 set_debugreg(vcpu
->arch
.host_db
[1], 1);
3548 set_debugreg(vcpu
->arch
.host_db
[2], 2);
3549 set_debugreg(vcpu
->arch
.host_db
[3], 3);
3551 set_debugreg(vcpu
->arch
.host_dr6
, 6);
3552 set_debugreg(vcpu
->arch
.host_dr7
, 7);
3554 set_bit(KVM_REQ_KICK
, &vcpu
->requests
);
3560 * We must have an instruction between local_irq_enable() and
3561 * kvm_guest_exit(), so the timer interrupt isn't delayed by
3562 * the interrupt shadow. The stat.exits increment will do nicely.
3563 * But we need to prevent reordering, hence this barrier():
3571 down_read(&vcpu
->kvm
->slots_lock
);
3574 * Profile KVM exit RIPs:
3576 if (unlikely(prof_on
== KVM_PROFILING
)) {
3577 unsigned long rip
= kvm_rip_read(vcpu
);
3578 profile_hit(KVM_PROFILING
, (void *)rip
);
3582 kvm_lapic_sync_from_vapic(vcpu
);
3584 r
= kvm_x86_ops
->handle_exit(kvm_run
, vcpu
);
3590 static int __vcpu_run(struct kvm_vcpu
*vcpu
, struct kvm_run
*kvm_run
)
3594 if (unlikely(vcpu
->arch
.mp_state
== KVM_MP_STATE_SIPI_RECEIVED
)) {
3595 pr_debug("vcpu %d received sipi with vector # %x\n",
3596 vcpu
->vcpu_id
, vcpu
->arch
.sipi_vector
);
3597 kvm_lapic_reset(vcpu
);
3598 r
= kvm_arch_vcpu_reset(vcpu
);
3601 vcpu
->arch
.mp_state
= KVM_MP_STATE_RUNNABLE
;
3604 down_read(&vcpu
->kvm
->slots_lock
);
3609 if (vcpu
->arch
.mp_state
== KVM_MP_STATE_RUNNABLE
)
3610 r
= vcpu_enter_guest(vcpu
, kvm_run
);
3612 up_read(&vcpu
->kvm
->slots_lock
);
3613 kvm_vcpu_block(vcpu
);
3614 down_read(&vcpu
->kvm
->slots_lock
);
3615 if (test_and_clear_bit(KVM_REQ_UNHALT
, &vcpu
->requests
))
3617 switch(vcpu
->arch
.mp_state
) {
3618 case KVM_MP_STATE_HALTED
:
3619 vcpu
->arch
.mp_state
=
3620 KVM_MP_STATE_RUNNABLE
;
3621 case KVM_MP_STATE_RUNNABLE
:
3623 case KVM_MP_STATE_SIPI_RECEIVED
:
3634 clear_bit(KVM_REQ_PENDING_TIMER
, &vcpu
->requests
);
3635 if (kvm_cpu_has_pending_timer(vcpu
))
3636 kvm_inject_pending_timer_irqs(vcpu
);
3638 if (dm_request_for_irq_injection(vcpu
, kvm_run
)) {
3640 kvm_run
->exit_reason
= KVM_EXIT_INTR
;
3641 ++vcpu
->stat
.request_irq_exits
;
3643 if (signal_pending(current
)) {
3645 kvm_run
->exit_reason
= KVM_EXIT_INTR
;
3646 ++vcpu
->stat
.signal_exits
;
3648 if (need_resched()) {
3649 up_read(&vcpu
->kvm
->slots_lock
);
3651 down_read(&vcpu
->kvm
->slots_lock
);
3655 up_read(&vcpu
->kvm
->slots_lock
);
3656 post_kvm_run_save(vcpu
, kvm_run
);
3663 int kvm_arch_vcpu_ioctl_run(struct kvm_vcpu
*vcpu
, struct kvm_run
*kvm_run
)
3670 if (vcpu
->sigset_active
)
3671 sigprocmask(SIG_SETMASK
, &vcpu
->sigset
, &sigsaved
);
3673 if (unlikely(vcpu
->arch
.mp_state
== KVM_MP_STATE_UNINITIALIZED
)) {
3674 kvm_vcpu_block(vcpu
);
3675 clear_bit(KVM_REQ_UNHALT
, &vcpu
->requests
);
3680 /* re-sync apic's tpr */
3681 if (!irqchip_in_kernel(vcpu
->kvm
))
3682 kvm_set_cr8(vcpu
, kvm_run
->cr8
);
3684 if (vcpu
->arch
.pio
.cur_count
) {
3685 r
= complete_pio(vcpu
);
3689 #if CONFIG_HAS_IOMEM
3690 if (vcpu
->mmio_needed
) {
3691 memcpy(vcpu
->mmio_data
, kvm_run
->mmio
.data
, 8);
3692 vcpu
->mmio_read_completed
= 1;
3693 vcpu
->mmio_needed
= 0;
3695 down_read(&vcpu
->kvm
->slots_lock
);
3696 r
= emulate_instruction(vcpu
, kvm_run
,
3697 vcpu
->arch
.mmio_fault_cr2
, 0,
3698 EMULTYPE_NO_DECODE
);
3699 up_read(&vcpu
->kvm
->slots_lock
);
3700 if (r
== EMULATE_DO_MMIO
) {
3702 * Read-modify-write. Back to userspace.
3709 if (kvm_run
->exit_reason
== KVM_EXIT_HYPERCALL
)
3710 kvm_register_write(vcpu
, VCPU_REGS_RAX
,
3711 kvm_run
->hypercall
.ret
);
3713 r
= __vcpu_run(vcpu
, kvm_run
);
3716 if (vcpu
->sigset_active
)
3717 sigprocmask(SIG_SETMASK
, &sigsaved
, NULL
);
3723 int kvm_arch_vcpu_ioctl_get_regs(struct kvm_vcpu
*vcpu
, struct kvm_regs
*regs
)
3727 regs
->rax
= kvm_register_read(vcpu
, VCPU_REGS_RAX
);
3728 regs
->rbx
= kvm_register_read(vcpu
, VCPU_REGS_RBX
);
3729 regs
->rcx
= kvm_register_read(vcpu
, VCPU_REGS_RCX
);
3730 regs
->rdx
= kvm_register_read(vcpu
, VCPU_REGS_RDX
);
3731 regs
->rsi
= kvm_register_read(vcpu
, VCPU_REGS_RSI
);
3732 regs
->rdi
= kvm_register_read(vcpu
, VCPU_REGS_RDI
);
3733 regs
->rsp
= kvm_register_read(vcpu
, VCPU_REGS_RSP
);
3734 regs
->rbp
= kvm_register_read(vcpu
, VCPU_REGS_RBP
);
3735 #ifdef CONFIG_X86_64
3736 regs
->r8
= kvm_register_read(vcpu
, VCPU_REGS_R8
);
3737 regs
->r9
= kvm_register_read(vcpu
, VCPU_REGS_R9
);
3738 regs
->r10
= kvm_register_read(vcpu
, VCPU_REGS_R10
);
3739 regs
->r11
= kvm_register_read(vcpu
, VCPU_REGS_R11
);
3740 regs
->r12
= kvm_register_read(vcpu
, VCPU_REGS_R12
);
3741 regs
->r13
= kvm_register_read(vcpu
, VCPU_REGS_R13
);
3742 regs
->r14
= kvm_register_read(vcpu
, VCPU_REGS_R14
);
3743 regs
->r15
= kvm_register_read(vcpu
, VCPU_REGS_R15
);
3746 regs
->rip
= kvm_rip_read(vcpu
);
3747 regs
->rflags
= kvm_x86_ops
->get_rflags(vcpu
);
3750 * Don't leak debug flags in case they were set for guest debugging
3752 if (vcpu
->guest_debug
& KVM_GUESTDBG_SINGLESTEP
)
3753 regs
->rflags
&= ~(X86_EFLAGS_TF
| X86_EFLAGS_RF
);
3760 int kvm_arch_vcpu_ioctl_set_regs(struct kvm_vcpu
*vcpu
, struct kvm_regs
*regs
)
3764 kvm_register_write(vcpu
, VCPU_REGS_RAX
, regs
->rax
);
3765 kvm_register_write(vcpu
, VCPU_REGS_RBX
, regs
->rbx
);
3766 kvm_register_write(vcpu
, VCPU_REGS_RCX
, regs
->rcx
);
3767 kvm_register_write(vcpu
, VCPU_REGS_RDX
, regs
->rdx
);
3768 kvm_register_write(vcpu
, VCPU_REGS_RSI
, regs
->rsi
);
3769 kvm_register_write(vcpu
, VCPU_REGS_RDI
, regs
->rdi
);
3770 kvm_register_write(vcpu
, VCPU_REGS_RSP
, regs
->rsp
);
3771 kvm_register_write(vcpu
, VCPU_REGS_RBP
, regs
->rbp
);
3772 #ifdef CONFIG_X86_64
3773 kvm_register_write(vcpu
, VCPU_REGS_R8
, regs
->r8
);
3774 kvm_register_write(vcpu
, VCPU_REGS_R9
, regs
->r9
);
3775 kvm_register_write(vcpu
, VCPU_REGS_R10
, regs
->r10
);
3776 kvm_register_write(vcpu
, VCPU_REGS_R11
, regs
->r11
);
3777 kvm_register_write(vcpu
, VCPU_REGS_R12
, regs
->r12
);
3778 kvm_register_write(vcpu
, VCPU_REGS_R13
, regs
->r13
);
3779 kvm_register_write(vcpu
, VCPU_REGS_R14
, regs
->r14
);
3780 kvm_register_write(vcpu
, VCPU_REGS_R15
, regs
->r15
);
3784 kvm_rip_write(vcpu
, regs
->rip
);
3785 kvm_x86_ops
->set_rflags(vcpu
, regs
->rflags
);
3788 vcpu
->arch
.exception
.pending
= false;
3795 void kvm_get_segment(struct kvm_vcpu
*vcpu
,
3796 struct kvm_segment
*var
, int seg
)
3798 kvm_x86_ops
->get_segment(vcpu
, var
, seg
);
3801 void kvm_get_cs_db_l_bits(struct kvm_vcpu
*vcpu
, int *db
, int *l
)
3803 struct kvm_segment cs
;
3805 kvm_get_segment(vcpu
, &cs
, VCPU_SREG_CS
);
3809 EXPORT_SYMBOL_GPL(kvm_get_cs_db_l_bits
);
3811 int kvm_arch_vcpu_ioctl_get_sregs(struct kvm_vcpu
*vcpu
,
3812 struct kvm_sregs
*sregs
)
3814 struct descriptor_table dt
;
3818 kvm_get_segment(vcpu
, &sregs
->cs
, VCPU_SREG_CS
);
3819 kvm_get_segment(vcpu
, &sregs
->ds
, VCPU_SREG_DS
);
3820 kvm_get_segment(vcpu
, &sregs
->es
, VCPU_SREG_ES
);
3821 kvm_get_segment(vcpu
, &sregs
->fs
, VCPU_SREG_FS
);
3822 kvm_get_segment(vcpu
, &sregs
->gs
, VCPU_SREG_GS
);
3823 kvm_get_segment(vcpu
, &sregs
->ss
, VCPU_SREG_SS
);
3825 kvm_get_segment(vcpu
, &sregs
->tr
, VCPU_SREG_TR
);
3826 kvm_get_segment(vcpu
, &sregs
->ldt
, VCPU_SREG_LDTR
);
3828 kvm_x86_ops
->get_idt(vcpu
, &dt
);
3829 sregs
->idt
.limit
= dt
.limit
;
3830 sregs
->idt
.base
= dt
.base
;
3831 kvm_x86_ops
->get_gdt(vcpu
, &dt
);
3832 sregs
->gdt
.limit
= dt
.limit
;
3833 sregs
->gdt
.base
= dt
.base
;
3835 kvm_x86_ops
->decache_cr4_guest_bits(vcpu
);
3836 sregs
->cr0
= vcpu
->arch
.cr0
;
3837 sregs
->cr2
= vcpu
->arch
.cr2
;
3838 sregs
->cr3
= vcpu
->arch
.cr3
;
3839 sregs
->cr4
= vcpu
->arch
.cr4
;
3840 sregs
->cr8
= kvm_get_cr8(vcpu
);
3841 sregs
->efer
= vcpu
->arch
.shadow_efer
;
3842 sregs
->apic_base
= kvm_get_apic_base(vcpu
);
3844 memset(sregs
->interrupt_bitmap
, 0, sizeof sregs
->interrupt_bitmap
);
3846 if (vcpu
->arch
.interrupt
.pending
&& !vcpu
->arch
.interrupt
.soft
)
3847 set_bit(vcpu
->arch
.interrupt
.nr
,
3848 (unsigned long *)sregs
->interrupt_bitmap
);
3855 int kvm_arch_vcpu_ioctl_get_mpstate(struct kvm_vcpu
*vcpu
,
3856 struct kvm_mp_state
*mp_state
)
3859 mp_state
->mp_state
= vcpu
->arch
.mp_state
;
3864 int kvm_arch_vcpu_ioctl_set_mpstate(struct kvm_vcpu
*vcpu
,
3865 struct kvm_mp_state
*mp_state
)
3868 vcpu
->arch
.mp_state
= mp_state
->mp_state
;
3873 static void kvm_set_segment(struct kvm_vcpu
*vcpu
,
3874 struct kvm_segment
*var
, int seg
)
3876 kvm_x86_ops
->set_segment(vcpu
, var
, seg
);
3879 static void seg_desct_to_kvm_desct(struct desc_struct
*seg_desc
, u16 selector
,
3880 struct kvm_segment
*kvm_desct
)
3882 kvm_desct
->base
= seg_desc
->base0
;
3883 kvm_desct
->base
|= seg_desc
->base1
<< 16;
3884 kvm_desct
->base
|= seg_desc
->base2
<< 24;
3885 kvm_desct
->limit
= seg_desc
->limit0
;
3886 kvm_desct
->limit
|= seg_desc
->limit
<< 16;
3888 kvm_desct
->limit
<<= 12;
3889 kvm_desct
->limit
|= 0xfff;
3891 kvm_desct
->selector
= selector
;
3892 kvm_desct
->type
= seg_desc
->type
;
3893 kvm_desct
->present
= seg_desc
->p
;
3894 kvm_desct
->dpl
= seg_desc
->dpl
;
3895 kvm_desct
->db
= seg_desc
->d
;
3896 kvm_desct
->s
= seg_desc
->s
;
3897 kvm_desct
->l
= seg_desc
->l
;
3898 kvm_desct
->g
= seg_desc
->g
;
3899 kvm_desct
->avl
= seg_desc
->avl
;
3901 kvm_desct
->unusable
= 1;
3903 kvm_desct
->unusable
= 0;
3904 kvm_desct
->padding
= 0;
3907 static void get_segment_descriptor_dtable(struct kvm_vcpu
*vcpu
,
3909 struct descriptor_table
*dtable
)
3911 if (selector
& 1 << 2) {
3912 struct kvm_segment kvm_seg
;
3914 kvm_get_segment(vcpu
, &kvm_seg
, VCPU_SREG_LDTR
);
3916 if (kvm_seg
.unusable
)
3919 dtable
->limit
= kvm_seg
.limit
;
3920 dtable
->base
= kvm_seg
.base
;
3923 kvm_x86_ops
->get_gdt(vcpu
, dtable
);
3926 /* allowed just for 8 bytes segments */
3927 static int load_guest_segment_descriptor(struct kvm_vcpu
*vcpu
, u16 selector
,
3928 struct desc_struct
*seg_desc
)
3931 struct descriptor_table dtable
;
3932 u16 index
= selector
>> 3;
3934 get_segment_descriptor_dtable(vcpu
, selector
, &dtable
);
3936 if (dtable
.limit
< index
* 8 + 7) {
3937 kvm_queue_exception_e(vcpu
, GP_VECTOR
, selector
& 0xfffc);
3940 gpa
= vcpu
->arch
.mmu
.gva_to_gpa(vcpu
, dtable
.base
);
3942 return kvm_read_guest(vcpu
->kvm
, gpa
, seg_desc
, 8);
3945 /* allowed just for 8 bytes segments */
3946 static int save_guest_segment_descriptor(struct kvm_vcpu
*vcpu
, u16 selector
,
3947 struct desc_struct
*seg_desc
)
3950 struct descriptor_table dtable
;
3951 u16 index
= selector
>> 3;
3953 get_segment_descriptor_dtable(vcpu
, selector
, &dtable
);
3955 if (dtable
.limit
< index
* 8 + 7)
3957 gpa
= vcpu
->arch
.mmu
.gva_to_gpa(vcpu
, dtable
.base
);
3959 return kvm_write_guest(vcpu
->kvm
, gpa
, seg_desc
, 8);
3962 static u32
get_tss_base_addr(struct kvm_vcpu
*vcpu
,
3963 struct desc_struct
*seg_desc
)
3967 base_addr
= seg_desc
->base0
;
3968 base_addr
|= (seg_desc
->base1
<< 16);
3969 base_addr
|= (seg_desc
->base2
<< 24);
3971 return vcpu
->arch
.mmu
.gva_to_gpa(vcpu
, base_addr
);
3974 static u16
get_segment_selector(struct kvm_vcpu
*vcpu
, int seg
)
3976 struct kvm_segment kvm_seg
;
3978 kvm_get_segment(vcpu
, &kvm_seg
, seg
);
3979 return kvm_seg
.selector
;
3982 static int load_segment_descriptor_to_kvm_desct(struct kvm_vcpu
*vcpu
,
3984 struct kvm_segment
*kvm_seg
)
3986 struct desc_struct seg_desc
;
3988 if (load_guest_segment_descriptor(vcpu
, selector
, &seg_desc
))
3990 seg_desct_to_kvm_desct(&seg_desc
, selector
, kvm_seg
);
3994 static int kvm_load_realmode_segment(struct kvm_vcpu
*vcpu
, u16 selector
, int seg
)
3996 struct kvm_segment segvar
= {
3997 .base
= selector
<< 4,
3999 .selector
= selector
,
4010 kvm_x86_ops
->set_segment(vcpu
, &segvar
, seg
);
4014 int kvm_load_segment_descriptor(struct kvm_vcpu
*vcpu
, u16 selector
,
4015 int type_bits
, int seg
)
4017 struct kvm_segment kvm_seg
;
4019 if (!(vcpu
->arch
.cr0
& X86_CR0_PE
))
4020 return kvm_load_realmode_segment(vcpu
, selector
, seg
);
4021 if (load_segment_descriptor_to_kvm_desct(vcpu
, selector
, &kvm_seg
))
4023 kvm_seg
.type
|= type_bits
;
4025 if (seg
!= VCPU_SREG_SS
&& seg
!= VCPU_SREG_CS
&&
4026 seg
!= VCPU_SREG_LDTR
)
4028 kvm_seg
.unusable
= 1;
4030 kvm_set_segment(vcpu
, &kvm_seg
, seg
);
4034 static void save_state_to_tss32(struct kvm_vcpu
*vcpu
,
4035 struct tss_segment_32
*tss
)
4037 tss
->cr3
= vcpu
->arch
.cr3
;
4038 tss
->eip
= kvm_rip_read(vcpu
);
4039 tss
->eflags
= kvm_x86_ops
->get_rflags(vcpu
);
4040 tss
->eax
= kvm_register_read(vcpu
, VCPU_REGS_RAX
);
4041 tss
->ecx
= kvm_register_read(vcpu
, VCPU_REGS_RCX
);
4042 tss
->edx
= kvm_register_read(vcpu
, VCPU_REGS_RDX
);
4043 tss
->ebx
= kvm_register_read(vcpu
, VCPU_REGS_RBX
);
4044 tss
->esp
= kvm_register_read(vcpu
, VCPU_REGS_RSP
);
4045 tss
->ebp
= kvm_register_read(vcpu
, VCPU_REGS_RBP
);
4046 tss
->esi
= kvm_register_read(vcpu
, VCPU_REGS_RSI
);
4047 tss
->edi
= kvm_register_read(vcpu
, VCPU_REGS_RDI
);
4048 tss
->es
= get_segment_selector(vcpu
, VCPU_SREG_ES
);
4049 tss
->cs
= get_segment_selector(vcpu
, VCPU_SREG_CS
);
4050 tss
->ss
= get_segment_selector(vcpu
, VCPU_SREG_SS
);
4051 tss
->ds
= get_segment_selector(vcpu
, VCPU_SREG_DS
);
4052 tss
->fs
= get_segment_selector(vcpu
, VCPU_SREG_FS
);
4053 tss
->gs
= get_segment_selector(vcpu
, VCPU_SREG_GS
);
4054 tss
->ldt_selector
= get_segment_selector(vcpu
, VCPU_SREG_LDTR
);
4057 static int load_state_from_tss32(struct kvm_vcpu
*vcpu
,
4058 struct tss_segment_32
*tss
)
4060 kvm_set_cr3(vcpu
, tss
->cr3
);
4062 kvm_rip_write(vcpu
, tss
->eip
);
4063 kvm_x86_ops
->set_rflags(vcpu
, tss
->eflags
| 2);
4065 kvm_register_write(vcpu
, VCPU_REGS_RAX
, tss
->eax
);
4066 kvm_register_write(vcpu
, VCPU_REGS_RCX
, tss
->ecx
);
4067 kvm_register_write(vcpu
, VCPU_REGS_RDX
, tss
->edx
);
4068 kvm_register_write(vcpu
, VCPU_REGS_RBX
, tss
->ebx
);
4069 kvm_register_write(vcpu
, VCPU_REGS_RSP
, tss
->esp
);
4070 kvm_register_write(vcpu
, VCPU_REGS_RBP
, tss
->ebp
);
4071 kvm_register_write(vcpu
, VCPU_REGS_RSI
, tss
->esi
);
4072 kvm_register_write(vcpu
, VCPU_REGS_RDI
, tss
->edi
);
4074 if (kvm_load_segment_descriptor(vcpu
, tss
->ldt_selector
, 0, VCPU_SREG_LDTR
))
4077 if (kvm_load_segment_descriptor(vcpu
, tss
->es
, 1, VCPU_SREG_ES
))
4080 if (kvm_load_segment_descriptor(vcpu
, tss
->cs
, 9, VCPU_SREG_CS
))
4083 if (kvm_load_segment_descriptor(vcpu
, tss
->ss
, 1, VCPU_SREG_SS
))
4086 if (kvm_load_segment_descriptor(vcpu
, tss
->ds
, 1, VCPU_SREG_DS
))
4089 if (kvm_load_segment_descriptor(vcpu
, tss
->fs
, 1, VCPU_SREG_FS
))
4092 if (kvm_load_segment_descriptor(vcpu
, tss
->gs
, 1, VCPU_SREG_GS
))
4097 static void save_state_to_tss16(struct kvm_vcpu
*vcpu
,
4098 struct tss_segment_16
*tss
)
4100 tss
->ip
= kvm_rip_read(vcpu
);
4101 tss
->flag
= kvm_x86_ops
->get_rflags(vcpu
);
4102 tss
->ax
= kvm_register_read(vcpu
, VCPU_REGS_RAX
);
4103 tss
->cx
= kvm_register_read(vcpu
, VCPU_REGS_RCX
);
4104 tss
->dx
= kvm_register_read(vcpu
, VCPU_REGS_RDX
);
4105 tss
->bx
= kvm_register_read(vcpu
, VCPU_REGS_RBX
);
4106 tss
->sp
= kvm_register_read(vcpu
, VCPU_REGS_RSP
);
4107 tss
->bp
= kvm_register_read(vcpu
, VCPU_REGS_RBP
);
4108 tss
->si
= kvm_register_read(vcpu
, VCPU_REGS_RSI
);
4109 tss
->di
= kvm_register_read(vcpu
, VCPU_REGS_RDI
);
4111 tss
->es
= get_segment_selector(vcpu
, VCPU_SREG_ES
);
4112 tss
->cs
= get_segment_selector(vcpu
, VCPU_SREG_CS
);
4113 tss
->ss
= get_segment_selector(vcpu
, VCPU_SREG_SS
);
4114 tss
->ds
= get_segment_selector(vcpu
, VCPU_SREG_DS
);
4115 tss
->ldt
= get_segment_selector(vcpu
, VCPU_SREG_LDTR
);
4116 tss
->prev_task_link
= get_segment_selector(vcpu
, VCPU_SREG_TR
);
4119 static int load_state_from_tss16(struct kvm_vcpu
*vcpu
,
4120 struct tss_segment_16
*tss
)
4122 kvm_rip_write(vcpu
, tss
->ip
);
4123 kvm_x86_ops
->set_rflags(vcpu
, tss
->flag
| 2);
4124 kvm_register_write(vcpu
, VCPU_REGS_RAX
, tss
->ax
);
4125 kvm_register_write(vcpu
, VCPU_REGS_RCX
, tss
->cx
);
4126 kvm_register_write(vcpu
, VCPU_REGS_RDX
, tss
->dx
);
4127 kvm_register_write(vcpu
, VCPU_REGS_RBX
, tss
->bx
);
4128 kvm_register_write(vcpu
, VCPU_REGS_RSP
, tss
->sp
);
4129 kvm_register_write(vcpu
, VCPU_REGS_RBP
, tss
->bp
);
4130 kvm_register_write(vcpu
, VCPU_REGS_RSI
, tss
->si
);
4131 kvm_register_write(vcpu
, VCPU_REGS_RDI
, tss
->di
);
4133 if (kvm_load_segment_descriptor(vcpu
, tss
->ldt
, 0, VCPU_SREG_LDTR
))
4136 if (kvm_load_segment_descriptor(vcpu
, tss
->es
, 1, VCPU_SREG_ES
))
4139 if (kvm_load_segment_descriptor(vcpu
, tss
->cs
, 9, VCPU_SREG_CS
))
4142 if (kvm_load_segment_descriptor(vcpu
, tss
->ss
, 1, VCPU_SREG_SS
))
4145 if (kvm_load_segment_descriptor(vcpu
, tss
->ds
, 1, VCPU_SREG_DS
))
4150 static int kvm_task_switch_16(struct kvm_vcpu
*vcpu
, u16 tss_selector
,
4151 u16 old_tss_sel
, u32 old_tss_base
,
4152 struct desc_struct
*nseg_desc
)
4154 struct tss_segment_16 tss_segment_16
;
4157 if (kvm_read_guest(vcpu
->kvm
, old_tss_base
, &tss_segment_16
,
4158 sizeof tss_segment_16
))
4161 save_state_to_tss16(vcpu
, &tss_segment_16
);
4163 if (kvm_write_guest(vcpu
->kvm
, old_tss_base
, &tss_segment_16
,
4164 sizeof tss_segment_16
))
4167 if (kvm_read_guest(vcpu
->kvm
, get_tss_base_addr(vcpu
, nseg_desc
),
4168 &tss_segment_16
, sizeof tss_segment_16
))
4171 if (old_tss_sel
!= 0xffff) {
4172 tss_segment_16
.prev_task_link
= old_tss_sel
;
4174 if (kvm_write_guest(vcpu
->kvm
,
4175 get_tss_base_addr(vcpu
, nseg_desc
),
4176 &tss_segment_16
.prev_task_link
,
4177 sizeof tss_segment_16
.prev_task_link
))
4181 if (load_state_from_tss16(vcpu
, &tss_segment_16
))
4189 static int kvm_task_switch_32(struct kvm_vcpu
*vcpu
, u16 tss_selector
,
4190 u16 old_tss_sel
, u32 old_tss_base
,
4191 struct desc_struct
*nseg_desc
)
4193 struct tss_segment_32 tss_segment_32
;
4196 if (kvm_read_guest(vcpu
->kvm
, old_tss_base
, &tss_segment_32
,
4197 sizeof tss_segment_32
))
4200 save_state_to_tss32(vcpu
, &tss_segment_32
);
4202 if (kvm_write_guest(vcpu
->kvm
, old_tss_base
, &tss_segment_32
,
4203 sizeof tss_segment_32
))
4206 if (kvm_read_guest(vcpu
->kvm
, get_tss_base_addr(vcpu
, nseg_desc
),
4207 &tss_segment_32
, sizeof tss_segment_32
))
4210 if (old_tss_sel
!= 0xffff) {
4211 tss_segment_32
.prev_task_link
= old_tss_sel
;
4213 if (kvm_write_guest(vcpu
->kvm
,
4214 get_tss_base_addr(vcpu
, nseg_desc
),
4215 &tss_segment_32
.prev_task_link
,
4216 sizeof tss_segment_32
.prev_task_link
))
4220 if (load_state_from_tss32(vcpu
, &tss_segment_32
))
4228 int kvm_task_switch(struct kvm_vcpu
*vcpu
, u16 tss_selector
, int reason
)
4230 struct kvm_segment tr_seg
;
4231 struct desc_struct cseg_desc
;
4232 struct desc_struct nseg_desc
;
4234 u32 old_tss_base
= get_segment_base(vcpu
, VCPU_SREG_TR
);
4235 u16 old_tss_sel
= get_segment_selector(vcpu
, VCPU_SREG_TR
);
4237 old_tss_base
= vcpu
->arch
.mmu
.gva_to_gpa(vcpu
, old_tss_base
);
4239 /* FIXME: Handle errors. Failure to read either TSS or their
4240 * descriptors should generate a pagefault.
4242 if (load_guest_segment_descriptor(vcpu
, tss_selector
, &nseg_desc
))
4245 if (load_guest_segment_descriptor(vcpu
, old_tss_sel
, &cseg_desc
))
4248 if (reason
!= TASK_SWITCH_IRET
) {
4251 cpl
= kvm_x86_ops
->get_cpl(vcpu
);
4252 if ((tss_selector
& 3) > nseg_desc
.dpl
|| cpl
> nseg_desc
.dpl
) {
4253 kvm_queue_exception_e(vcpu
, GP_VECTOR
, 0);
4258 if (!nseg_desc
.p
|| (nseg_desc
.limit0
| nseg_desc
.limit
<< 16) < 0x67) {
4259 kvm_queue_exception_e(vcpu
, TS_VECTOR
, tss_selector
& 0xfffc);
4263 if (reason
== TASK_SWITCH_IRET
|| reason
== TASK_SWITCH_JMP
) {
4264 cseg_desc
.type
&= ~(1 << 1); //clear the B flag
4265 save_guest_segment_descriptor(vcpu
, old_tss_sel
, &cseg_desc
);
4268 if (reason
== TASK_SWITCH_IRET
) {
4269 u32 eflags
= kvm_x86_ops
->get_rflags(vcpu
);
4270 kvm_x86_ops
->set_rflags(vcpu
, eflags
& ~X86_EFLAGS_NT
);
4273 /* set back link to prev task only if NT bit is set in eflags
4274 note that old_tss_sel is not used afetr this point */
4275 if (reason
!= TASK_SWITCH_CALL
&& reason
!= TASK_SWITCH_GATE
)
4276 old_tss_sel
= 0xffff;
4278 /* set back link to prev task only if NT bit is set in eflags
4279 note that old_tss_sel is not used afetr this point */
4280 if (reason
!= TASK_SWITCH_CALL
&& reason
!= TASK_SWITCH_GATE
)
4281 old_tss_sel
= 0xffff;
4283 if (nseg_desc
.type
& 8)
4284 ret
= kvm_task_switch_32(vcpu
, tss_selector
, old_tss_sel
,
4285 old_tss_base
, &nseg_desc
);
4287 ret
= kvm_task_switch_16(vcpu
, tss_selector
, old_tss_sel
,
4288 old_tss_base
, &nseg_desc
);
4290 if (reason
== TASK_SWITCH_CALL
|| reason
== TASK_SWITCH_GATE
) {
4291 u32 eflags
= kvm_x86_ops
->get_rflags(vcpu
);
4292 kvm_x86_ops
->set_rflags(vcpu
, eflags
| X86_EFLAGS_NT
);
4295 if (reason
!= TASK_SWITCH_IRET
) {
4296 nseg_desc
.type
|= (1 << 1);
4297 save_guest_segment_descriptor(vcpu
, tss_selector
,
4301 kvm_x86_ops
->set_cr0(vcpu
, vcpu
->arch
.cr0
| X86_CR0_TS
);
4302 seg_desct_to_kvm_desct(&nseg_desc
, tss_selector
, &tr_seg
);
4304 kvm_set_segment(vcpu
, &tr_seg
, VCPU_SREG_TR
);
4308 EXPORT_SYMBOL_GPL(kvm_task_switch
);
4310 int kvm_arch_vcpu_ioctl_set_sregs(struct kvm_vcpu
*vcpu
,
4311 struct kvm_sregs
*sregs
)
4313 int mmu_reset_needed
= 0;
4314 int pending_vec
, max_bits
;
4315 struct descriptor_table dt
;
4319 dt
.limit
= sregs
->idt
.limit
;
4320 dt
.base
= sregs
->idt
.base
;
4321 kvm_x86_ops
->set_idt(vcpu
, &dt
);
4322 dt
.limit
= sregs
->gdt
.limit
;
4323 dt
.base
= sregs
->gdt
.base
;
4324 kvm_x86_ops
->set_gdt(vcpu
, &dt
);
4326 vcpu
->arch
.cr2
= sregs
->cr2
;
4327 mmu_reset_needed
|= vcpu
->arch
.cr3
!= sregs
->cr3
;
4329 down_read(&vcpu
->kvm
->slots_lock
);
4330 if (gfn_to_memslot(vcpu
->kvm
, sregs
->cr3
>> PAGE_SHIFT
))
4331 vcpu
->arch
.cr3
= sregs
->cr3
;
4333 set_bit(KVM_REQ_TRIPLE_FAULT
, &vcpu
->requests
);
4334 up_read(&vcpu
->kvm
->slots_lock
);
4336 kvm_set_cr8(vcpu
, sregs
->cr8
);
4338 mmu_reset_needed
|= vcpu
->arch
.shadow_efer
!= sregs
->efer
;
4339 kvm_x86_ops
->set_efer(vcpu
, sregs
->efer
);
4340 kvm_set_apic_base(vcpu
, sregs
->apic_base
);
4342 kvm_x86_ops
->decache_cr4_guest_bits(vcpu
);
4344 mmu_reset_needed
|= vcpu
->arch
.cr0
!= sregs
->cr0
;
4345 kvm_x86_ops
->set_cr0(vcpu
, sregs
->cr0
);
4346 vcpu
->arch
.cr0
= sregs
->cr0
;
4348 mmu_reset_needed
|= vcpu
->arch
.cr4
!= sregs
->cr4
;
4349 kvm_x86_ops
->set_cr4(vcpu
, sregs
->cr4
);
4350 if (!is_long_mode(vcpu
) && is_pae(vcpu
))
4351 load_pdptrs(vcpu
, vcpu
->arch
.cr3
);
4353 if (mmu_reset_needed
)
4354 kvm_mmu_reset_context(vcpu
);
4356 max_bits
= (sizeof sregs
->interrupt_bitmap
) << 3;
4357 pending_vec
= find_first_bit(
4358 (const unsigned long *)sregs
->interrupt_bitmap
, max_bits
);
4359 if (pending_vec
< max_bits
) {
4360 kvm_queue_interrupt(vcpu
, pending_vec
, false);
4361 pr_debug("Set back pending irq %d\n", pending_vec
);
4362 if (irqchip_in_kernel(vcpu
->kvm
))
4363 kvm_pic_clear_isr_ack(vcpu
->kvm
);
4366 kvm_set_segment(vcpu
, &sregs
->cs
, VCPU_SREG_CS
);
4367 kvm_set_segment(vcpu
, &sregs
->ds
, VCPU_SREG_DS
);
4368 kvm_set_segment(vcpu
, &sregs
->es
, VCPU_SREG_ES
);
4369 kvm_set_segment(vcpu
, &sregs
->fs
, VCPU_SREG_FS
);
4370 kvm_set_segment(vcpu
, &sregs
->gs
, VCPU_SREG_GS
);
4371 kvm_set_segment(vcpu
, &sregs
->ss
, VCPU_SREG_SS
);
4373 kvm_set_segment(vcpu
, &sregs
->tr
, VCPU_SREG_TR
);
4374 kvm_set_segment(vcpu
, &sregs
->ldt
, VCPU_SREG_LDTR
);
4376 /* Older userspace won't unhalt the vcpu on reset. */
4377 if (kvm_vcpu_is_bsp(vcpu
) && kvm_rip_read(vcpu
) == 0xfff0 &&
4378 sregs
->cs
.selector
== 0xf000 && sregs
->cs
.base
== 0xffff0000 &&
4379 !(vcpu
->arch
.cr0
& X86_CR0_PE
))
4380 vcpu
->arch
.mp_state
= KVM_MP_STATE_RUNNABLE
;
4387 int kvm_arch_vcpu_ioctl_set_guest_debug(struct kvm_vcpu
*vcpu
,
4388 struct kvm_guest_debug
*dbg
)
4394 if ((dbg
->control
& (KVM_GUESTDBG_ENABLE
| KVM_GUESTDBG_USE_HW_BP
)) ==
4395 (KVM_GUESTDBG_ENABLE
| KVM_GUESTDBG_USE_HW_BP
)) {
4396 for (i
= 0; i
< KVM_NR_DB_REGS
; ++i
)
4397 vcpu
->arch
.eff_db
[i
] = dbg
->arch
.debugreg
[i
];
4398 vcpu
->arch
.switch_db_regs
=
4399 (dbg
->arch
.debugreg
[7] & DR7_BP_EN_MASK
);
4401 for (i
= 0; i
< KVM_NR_DB_REGS
; i
++)
4402 vcpu
->arch
.eff_db
[i
] = vcpu
->arch
.db
[i
];
4403 vcpu
->arch
.switch_db_regs
= (vcpu
->arch
.dr7
& DR7_BP_EN_MASK
);
4406 r
= kvm_x86_ops
->set_guest_debug(vcpu
, dbg
);
4408 if (dbg
->control
& KVM_GUESTDBG_INJECT_DB
)
4409 kvm_queue_exception(vcpu
, DB_VECTOR
);
4410 else if (dbg
->control
& KVM_GUESTDBG_INJECT_BP
)
4411 kvm_queue_exception(vcpu
, BP_VECTOR
);
4419 * fxsave fpu state. Taken from x86_64/processor.h. To be killed when
4420 * we have asm/x86/processor.h
4431 u32 st_space
[32]; /* 8*16 bytes for each FP-reg = 128 bytes */
4432 #ifdef CONFIG_X86_64
4433 u32 xmm_space
[64]; /* 16*16 bytes for each XMM-reg = 256 bytes */
4435 u32 xmm_space
[32]; /* 8*16 bytes for each XMM-reg = 128 bytes */
4440 * Translate a guest virtual address to a guest physical address.
4442 int kvm_arch_vcpu_ioctl_translate(struct kvm_vcpu
*vcpu
,
4443 struct kvm_translation
*tr
)
4445 unsigned long vaddr
= tr
->linear_address
;
4449 down_read(&vcpu
->kvm
->slots_lock
);
4450 gpa
= vcpu
->arch
.mmu
.gva_to_gpa(vcpu
, vaddr
);
4451 up_read(&vcpu
->kvm
->slots_lock
);
4452 tr
->physical_address
= gpa
;
4453 tr
->valid
= gpa
!= UNMAPPED_GVA
;
4461 int kvm_arch_vcpu_ioctl_get_fpu(struct kvm_vcpu
*vcpu
, struct kvm_fpu
*fpu
)
4463 struct fxsave
*fxsave
= (struct fxsave
*)&vcpu
->arch
.guest_fx_image
;
4467 memcpy(fpu
->fpr
, fxsave
->st_space
, 128);
4468 fpu
->fcw
= fxsave
->cwd
;
4469 fpu
->fsw
= fxsave
->swd
;
4470 fpu
->ftwx
= fxsave
->twd
;
4471 fpu
->last_opcode
= fxsave
->fop
;
4472 fpu
->last_ip
= fxsave
->rip
;
4473 fpu
->last_dp
= fxsave
->rdp
;
4474 memcpy(fpu
->xmm
, fxsave
->xmm_space
, sizeof fxsave
->xmm_space
);
4481 int kvm_arch_vcpu_ioctl_set_fpu(struct kvm_vcpu
*vcpu
, struct kvm_fpu
*fpu
)
4483 struct fxsave
*fxsave
= (struct fxsave
*)&vcpu
->arch
.guest_fx_image
;
4487 memcpy(fxsave
->st_space
, fpu
->fpr
, 128);
4488 fxsave
->cwd
= fpu
->fcw
;
4489 fxsave
->swd
= fpu
->fsw
;
4490 fxsave
->twd
= fpu
->ftwx
;
4491 fxsave
->fop
= fpu
->last_opcode
;
4492 fxsave
->rip
= fpu
->last_ip
;
4493 fxsave
->rdp
= fpu
->last_dp
;
4494 memcpy(fxsave
->xmm_space
, fpu
->xmm
, sizeof fxsave
->xmm_space
);
4501 void fx_init(struct kvm_vcpu
*vcpu
)
4503 unsigned after_mxcsr_mask
;
4506 * Touch the fpu the first time in non atomic context as if
4507 * this is the first fpu instruction the exception handler
4508 * will fire before the instruction returns and it'll have to
4509 * allocate ram with GFP_KERNEL.
4512 kvm_fx_save(&vcpu
->arch
.host_fx_image
);
4514 /* Initialize guest FPU by resetting ours and saving into guest's */
4516 kvm_fx_save(&vcpu
->arch
.host_fx_image
);
4518 kvm_fx_save(&vcpu
->arch
.guest_fx_image
);
4519 kvm_fx_restore(&vcpu
->arch
.host_fx_image
);
4522 vcpu
->arch
.cr0
|= X86_CR0_ET
;
4523 after_mxcsr_mask
= offsetof(struct i387_fxsave_struct
, st_space
);
4524 vcpu
->arch
.guest_fx_image
.mxcsr
= 0x1f80;
4525 memset((void *)&vcpu
->arch
.guest_fx_image
+ after_mxcsr_mask
,
4526 0, sizeof(struct i387_fxsave_struct
) - after_mxcsr_mask
);
4528 EXPORT_SYMBOL_GPL(fx_init
);
4530 void kvm_load_guest_fpu(struct kvm_vcpu
*vcpu
)
4532 if (!vcpu
->fpu_active
|| vcpu
->guest_fpu_loaded
)
4535 vcpu
->guest_fpu_loaded
= 1;
4536 kvm_fx_save(&vcpu
->arch
.host_fx_image
);
4537 kvm_fx_restore(&vcpu
->arch
.guest_fx_image
);
4539 EXPORT_SYMBOL_GPL(kvm_load_guest_fpu
);
4541 void kvm_put_guest_fpu(struct kvm_vcpu
*vcpu
)
4543 if (!vcpu
->guest_fpu_loaded
)
4546 vcpu
->guest_fpu_loaded
= 0;
4547 kvm_fx_save(&vcpu
->arch
.guest_fx_image
);
4548 kvm_fx_restore(&vcpu
->arch
.host_fx_image
);
4549 ++vcpu
->stat
.fpu_reload
;
4551 EXPORT_SYMBOL_GPL(kvm_put_guest_fpu
);
4553 void kvm_arch_vcpu_free(struct kvm_vcpu
*vcpu
)
4555 if (vcpu
->arch
.time_page
) {
4556 kvm_release_page_dirty(vcpu
->arch
.time_page
);
4557 vcpu
->arch
.time_page
= NULL
;
4560 kvm_x86_ops
->vcpu_free(vcpu
);
4563 struct kvm_vcpu
*kvm_arch_vcpu_create(struct kvm
*kvm
,
4566 return kvm_x86_ops
->vcpu_create(kvm
, id
);
4569 int kvm_arch_vcpu_setup(struct kvm_vcpu
*vcpu
)
4573 /* We do fxsave: this must be aligned. */
4574 BUG_ON((unsigned long)&vcpu
->arch
.host_fx_image
& 0xF);
4576 vcpu
->arch
.mtrr_state
.have_fixed
= 1;
4578 r
= kvm_arch_vcpu_reset(vcpu
);
4580 r
= kvm_mmu_setup(vcpu
);
4587 kvm_x86_ops
->vcpu_free(vcpu
);
4591 void kvm_arch_vcpu_destroy(struct kvm_vcpu
*vcpu
)
4594 kvm_mmu_unload(vcpu
);
4597 kvm_x86_ops
->vcpu_free(vcpu
);
4600 int kvm_arch_vcpu_reset(struct kvm_vcpu
*vcpu
)
4602 vcpu
->arch
.nmi_pending
= false;
4603 vcpu
->arch
.nmi_injected
= false;
4605 vcpu
->arch
.switch_db_regs
= 0;
4606 memset(vcpu
->arch
.db
, 0, sizeof(vcpu
->arch
.db
));
4607 vcpu
->arch
.dr6
= DR6_FIXED_1
;
4608 vcpu
->arch
.dr7
= DR7_FIXED_1
;
4610 return kvm_x86_ops
->vcpu_reset(vcpu
);
4613 void kvm_arch_hardware_enable(void *garbage
)
4615 kvm_x86_ops
->hardware_enable(garbage
);
4618 void kvm_arch_hardware_disable(void *garbage
)
4620 kvm_x86_ops
->hardware_disable(garbage
);
4623 int kvm_arch_hardware_setup(void)
4625 return kvm_x86_ops
->hardware_setup();
4628 void kvm_arch_hardware_unsetup(void)
4630 kvm_x86_ops
->hardware_unsetup();
4633 void kvm_arch_check_processor_compat(void *rtn
)
4635 kvm_x86_ops
->check_processor_compatibility(rtn
);
4638 int kvm_arch_vcpu_init(struct kvm_vcpu
*vcpu
)
4644 BUG_ON(vcpu
->kvm
== NULL
);
4647 vcpu
->arch
.mmu
.root_hpa
= INVALID_PAGE
;
4648 if (!irqchip_in_kernel(kvm
) || kvm_vcpu_is_bsp(vcpu
))
4649 vcpu
->arch
.mp_state
= KVM_MP_STATE_RUNNABLE
;
4651 vcpu
->arch
.mp_state
= KVM_MP_STATE_UNINITIALIZED
;
4653 page
= alloc_page(GFP_KERNEL
| __GFP_ZERO
);
4658 vcpu
->arch
.pio_data
= page_address(page
);
4660 r
= kvm_mmu_create(vcpu
);
4662 goto fail_free_pio_data
;
4664 if (irqchip_in_kernel(kvm
)) {
4665 r
= kvm_create_lapic(vcpu
);
4667 goto fail_mmu_destroy
;
4670 vcpu
->arch
.mce_banks
= kzalloc(KVM_MAX_MCE_BANKS
* sizeof(u64
) * 4,
4672 if (!vcpu
->arch
.mce_banks
) {
4674 goto fail_mmu_destroy
;
4676 vcpu
->arch
.mcg_cap
= KVM_MAX_MCE_BANKS
;
4681 kvm_mmu_destroy(vcpu
);
4683 free_page((unsigned long)vcpu
->arch
.pio_data
);
4688 void kvm_arch_vcpu_uninit(struct kvm_vcpu
*vcpu
)
4690 kvm_free_lapic(vcpu
);
4691 down_read(&vcpu
->kvm
->slots_lock
);
4692 kvm_mmu_destroy(vcpu
);
4693 up_read(&vcpu
->kvm
->slots_lock
);
4694 free_page((unsigned long)vcpu
->arch
.pio_data
);
4697 struct kvm
*kvm_arch_create_vm(void)
4699 struct kvm
*kvm
= kzalloc(sizeof(struct kvm
), GFP_KERNEL
);
4702 return ERR_PTR(-ENOMEM
);
4704 INIT_LIST_HEAD(&kvm
->arch
.active_mmu_pages
);
4705 INIT_LIST_HEAD(&kvm
->arch
.assigned_dev_head
);
4707 /* Reserve bit 0 of irq_sources_bitmap for userspace irq source */
4708 set_bit(KVM_USERSPACE_IRQ_SOURCE_ID
, &kvm
->arch
.irq_sources_bitmap
);
4710 rdtscll(kvm
->arch
.vm_init_tsc
);
4715 static void kvm_unload_vcpu_mmu(struct kvm_vcpu
*vcpu
)
4718 kvm_mmu_unload(vcpu
);
4722 static void kvm_free_vcpus(struct kvm
*kvm
)
4725 struct kvm_vcpu
*vcpu
;
4728 * Unpin any mmu pages first.
4730 kvm_for_each_vcpu(i
, vcpu
, kvm
)
4731 kvm_unload_vcpu_mmu(vcpu
);
4732 kvm_for_each_vcpu(i
, vcpu
, kvm
)
4733 kvm_arch_vcpu_free(vcpu
);
4735 mutex_lock(&kvm
->lock
);
4736 for (i
= 0; i
< atomic_read(&kvm
->online_vcpus
); i
++)
4737 kvm
->vcpus
[i
] = NULL
;
4739 atomic_set(&kvm
->online_vcpus
, 0);
4740 mutex_unlock(&kvm
->lock
);
4743 void kvm_arch_sync_events(struct kvm
*kvm
)
4745 kvm_free_all_assigned_devices(kvm
);
4748 void kvm_arch_destroy_vm(struct kvm
*kvm
)
4750 kvm_iommu_unmap_guest(kvm
);
4752 kfree(kvm
->arch
.vpic
);
4753 kfree(kvm
->arch
.vioapic
);
4754 kvm_free_vcpus(kvm
);
4755 kvm_free_physmem(kvm
);
4756 if (kvm
->arch
.apic_access_page
)
4757 put_page(kvm
->arch
.apic_access_page
);
4758 if (kvm
->arch
.ept_identity_pagetable
)
4759 put_page(kvm
->arch
.ept_identity_pagetable
);
4763 int kvm_arch_set_memory_region(struct kvm
*kvm
,
4764 struct kvm_userspace_memory_region
*mem
,
4765 struct kvm_memory_slot old
,
4768 int npages
= mem
->memory_size
>> PAGE_SHIFT
;
4769 struct kvm_memory_slot
*memslot
= &kvm
->memslots
[mem
->slot
];
4771 /*To keep backward compatibility with older userspace,
4772 *x86 needs to hanlde !user_alloc case.
4775 if (npages
&& !old
.rmap
) {
4776 unsigned long userspace_addr
;
4778 down_write(¤t
->mm
->mmap_sem
);
4779 userspace_addr
= do_mmap(NULL
, 0,
4781 PROT_READ
| PROT_WRITE
,
4782 MAP_PRIVATE
| MAP_ANONYMOUS
,
4784 up_write(¤t
->mm
->mmap_sem
);
4786 if (IS_ERR((void *)userspace_addr
))
4787 return PTR_ERR((void *)userspace_addr
);
4789 /* set userspace_addr atomically for kvm_hva_to_rmapp */
4790 spin_lock(&kvm
->mmu_lock
);
4791 memslot
->userspace_addr
= userspace_addr
;
4792 spin_unlock(&kvm
->mmu_lock
);
4794 if (!old
.user_alloc
&& old
.rmap
) {
4797 down_write(¤t
->mm
->mmap_sem
);
4798 ret
= do_munmap(current
->mm
, old
.userspace_addr
,
4799 old
.npages
* PAGE_SIZE
);
4800 up_write(¤t
->mm
->mmap_sem
);
4803 "kvm_vm_ioctl_set_memory_region: "
4804 "failed to munmap memory\n");
4809 spin_lock(&kvm
->mmu_lock
);
4810 if (!kvm
->arch
.n_requested_mmu_pages
) {
4811 unsigned int nr_mmu_pages
= kvm_mmu_calculate_mmu_pages(kvm
);
4812 kvm_mmu_change_mmu_pages(kvm
, nr_mmu_pages
);
4815 kvm_mmu_slot_remove_write_access(kvm
, mem
->slot
);
4816 spin_unlock(&kvm
->mmu_lock
);
4817 kvm_flush_remote_tlbs(kvm
);
4822 void kvm_arch_flush_shadow(struct kvm
*kvm
)
4824 kvm_mmu_zap_all(kvm
);
4825 kvm_reload_remote_mmus(kvm
);
4828 int kvm_arch_vcpu_runnable(struct kvm_vcpu
*vcpu
)
4830 return vcpu
->arch
.mp_state
== KVM_MP_STATE_RUNNABLE
4831 || vcpu
->arch
.mp_state
== KVM_MP_STATE_SIPI_RECEIVED
4832 || vcpu
->arch
.nmi_pending
;
4835 void kvm_vcpu_kick(struct kvm_vcpu
*vcpu
)
4838 int cpu
= vcpu
->cpu
;
4840 if (waitqueue_active(&vcpu
->wq
)) {
4841 wake_up_interruptible(&vcpu
->wq
);
4842 ++vcpu
->stat
.halt_wakeup
;
4846 if (cpu
!= me
&& (unsigned)cpu
< nr_cpu_ids
&& cpu_online(cpu
))
4847 if (!test_and_set_bit(KVM_REQ_KICK
, &vcpu
->requests
))
4848 smp_send_reschedule(cpu
);
4852 int kvm_arch_interrupt_allowed(struct kvm_vcpu
*vcpu
)
4854 return kvm_x86_ops
->interrupt_allowed(vcpu
);
4857 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_exit
);
4858 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_inj_virq
);
4859 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_page_fault
);
4860 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_msr
);
4861 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_cr
);