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 <linux/user-return-notifier.h>
41 #include <linux/srcu.h>
42 #include <linux/slab.h>
43 #include <linux/perf_event.h>
44 #include <trace/events/kvm.h>
45 #undef TRACE_INCLUDE_FILE
46 #define CREATE_TRACE_POINTS
49 #include <asm/debugreg.h>
50 #include <asm/uaccess.h>
56 #define MAX_IO_MSRS 256
57 #define CR0_RESERVED_BITS \
58 (~(unsigned long)(X86_CR0_PE | X86_CR0_MP | X86_CR0_EM | X86_CR0_TS \
59 | X86_CR0_ET | X86_CR0_NE | X86_CR0_WP | X86_CR0_AM \
60 | X86_CR0_NW | X86_CR0_CD | X86_CR0_PG))
61 #define CR4_RESERVED_BITS \
62 (~(unsigned long)(X86_CR4_VME | X86_CR4_PVI | X86_CR4_TSD | X86_CR4_DE\
63 | X86_CR4_PSE | X86_CR4_PAE | X86_CR4_MCE \
64 | X86_CR4_PGE | X86_CR4_PCE | X86_CR4_OSFXSR \
65 | X86_CR4_OSXMMEXCPT | X86_CR4_VMXE))
67 #define CR8_RESERVED_BITS (~(unsigned long)X86_CR8_TPR)
69 #define KVM_MAX_MCE_BANKS 32
70 #define KVM_MCE_CAP_SUPPORTED MCG_CTL_P
73 * - enable syscall per default because its emulated by KVM
74 * - enable LME and LMA per default on 64 bit KVM
77 static u64 __read_mostly efer_reserved_bits = 0xfffffffffffffafeULL;
79 static u64 __read_mostly efer_reserved_bits = 0xfffffffffffffffeULL;
82 #define VM_STAT(x) offsetof(struct kvm, stat.x), KVM_STAT_VM
83 #define VCPU_STAT(x) offsetof(struct kvm_vcpu, stat.x), KVM_STAT_VCPU
85 static void update_cr8_intercept(struct kvm_vcpu *vcpu);
86 static int kvm_dev_ioctl_get_supported_cpuid(struct kvm_cpuid2 *cpuid,
87 struct kvm_cpuid_entry2 __user *entries);
89 struct kvm_x86_ops *kvm_x86_ops;
90 EXPORT_SYMBOL_GPL(kvm_x86_ops);
93 module_param_named(ignore_msrs, ignore_msrs, bool, S_IRUGO | S_IWUSR);
95 #define KVM_NR_SHARED_MSRS 16
97 struct kvm_shared_msrs_global {
99 u32 msrs[KVM_NR_SHARED_MSRS];
102 struct kvm_shared_msrs {
103 struct user_return_notifier urn;
105 struct kvm_shared_msr_values {
108 } values[KVM_NR_SHARED_MSRS];
111 static struct kvm_shared_msrs_global __read_mostly shared_msrs_global;
112 static DEFINE_PER_CPU(struct kvm_shared_msrs, shared_msrs);
114 struct kvm_stats_debugfs_item debugfs_entries[] = {
115 { "pf_fixed", VCPU_STAT(pf_fixed) },
116 { "pf_guest", VCPU_STAT(pf_guest) },
117 { "tlb_flush", VCPU_STAT(tlb_flush) },
118 { "invlpg", VCPU_STAT(invlpg) },
119 { "exits", VCPU_STAT(exits) },
120 { "io_exits", VCPU_STAT(io_exits) },
121 { "mmio_exits", VCPU_STAT(mmio_exits) },
122 { "signal_exits", VCPU_STAT(signal_exits) },
123 { "irq_window", VCPU_STAT(irq_window_exits) },
124 { "nmi_window", VCPU_STAT(nmi_window_exits) },
125 { "halt_exits", VCPU_STAT(halt_exits) },
126 { "halt_wakeup", VCPU_STAT(halt_wakeup) },
127 { "hypercalls", VCPU_STAT(hypercalls) },
128 { "request_irq", VCPU_STAT(request_irq_exits) },
129 { "irq_exits", VCPU_STAT(irq_exits) },
130 { "host_state_reload", VCPU_STAT(host_state_reload) },
131 { "efer_reload", VCPU_STAT(efer_reload) },
132 { "fpu_reload", VCPU_STAT(fpu_reload) },
133 { "insn_emulation", VCPU_STAT(insn_emulation) },
134 { "insn_emulation_fail", VCPU_STAT(insn_emulation_fail) },
135 { "irq_injections", VCPU_STAT(irq_injections) },
136 { "nmi_injections", VCPU_STAT(nmi_injections) },
137 { "mmu_shadow_zapped", VM_STAT(mmu_shadow_zapped) },
138 { "mmu_pte_write", VM_STAT(mmu_pte_write) },
139 { "mmu_pte_updated", VM_STAT(mmu_pte_updated) },
140 { "mmu_pde_zapped", VM_STAT(mmu_pde_zapped) },
141 { "mmu_flooded", VM_STAT(mmu_flooded) },
142 { "mmu_recycled", VM_STAT(mmu_recycled) },
143 { "mmu_cache_miss", VM_STAT(mmu_cache_miss) },
144 { "mmu_unsync", VM_STAT(mmu_unsync) },
145 { "remote_tlb_flush", VM_STAT(remote_tlb_flush) },
146 { "largepages", VM_STAT(lpages) },
150 static void kvm_on_user_return(struct user_return_notifier *urn)
153 struct kvm_shared_msrs *locals
154 = container_of(urn, struct kvm_shared_msrs, urn);
155 struct kvm_shared_msr_values *values;
157 for (slot = 0; slot < shared_msrs_global.nr; ++slot) {
158 values = &locals->values[slot];
159 if (values->host != values->curr) {
160 wrmsrl(shared_msrs_global.msrs[slot], values->host);
161 values->curr = values->host;
164 locals->registered = false;
165 user_return_notifier_unregister(urn);
168 static void shared_msr_update(unsigned slot, u32 msr)
170 struct kvm_shared_msrs *smsr;
173 smsr = &__get_cpu_var(shared_msrs);
174 /* only read, and nobody should modify it at this time,
175 * so don't need lock */
176 if (slot >= shared_msrs_global.nr) {
177 printk(KERN_ERR "kvm: invalid MSR slot!");
180 rdmsrl_safe(msr, &value);
181 smsr->values[slot].host = value;
182 smsr->values[slot].curr = value;
185 void kvm_define_shared_msr(unsigned slot, u32 msr)
187 if (slot >= shared_msrs_global.nr)
188 shared_msrs_global.nr = slot + 1;
189 shared_msrs_global.msrs[slot] = msr;
190 /* we need ensured the shared_msr_global have been updated */
193 EXPORT_SYMBOL_GPL(kvm_define_shared_msr);
195 static void kvm_shared_msr_cpu_online(void)
199 for (i = 0; i < shared_msrs_global.nr; ++i)
200 shared_msr_update(i, shared_msrs_global.msrs[i]);
203 void kvm_set_shared_msr(unsigned slot, u64 value, u64 mask)
205 struct kvm_shared_msrs *smsr = &__get_cpu_var(shared_msrs);
207 if (((value ^ smsr->values[slot].curr) & mask) == 0)
209 smsr->values[slot].curr = value;
210 wrmsrl(shared_msrs_global.msrs[slot], value);
211 if (!smsr->registered) {
212 smsr->urn.on_user_return = kvm_on_user_return;
213 user_return_notifier_register(&smsr->urn);
214 smsr->registered = true;
217 EXPORT_SYMBOL_GPL(kvm_set_shared_msr);
219 static void drop_user_return_notifiers(void *ignore)
221 struct kvm_shared_msrs *smsr = &__get_cpu_var(shared_msrs);
223 if (smsr->registered)
224 kvm_on_user_return(&smsr->urn);
227 unsigned long segment_base(u16 selector)
229 struct descriptor_table gdt;
230 struct desc_struct *d;
231 unsigned long table_base;
238 table_base = gdt.base;
240 if (selector & 4) { /* from ldt */
241 u16 ldt_selector = kvm_read_ldt();
243 table_base = segment_base(ldt_selector);
245 d = (struct desc_struct *)(table_base + (selector & ~7));
246 v = get_desc_base(d);
248 if (d->s == 0 && (d->type == 2 || d->type == 9 || d->type == 11))
249 v |= ((unsigned long)((struct ldttss_desc64 *)d)->base3) << 32;
253 EXPORT_SYMBOL_GPL(segment_base);
255 u64 kvm_get_apic_base(struct kvm_vcpu *vcpu)
257 if (irqchip_in_kernel(vcpu->kvm))
258 return vcpu->arch.apic_base;
260 return vcpu->arch.apic_base;
262 EXPORT_SYMBOL_GPL(kvm_get_apic_base);
264 void kvm_set_apic_base(struct kvm_vcpu *vcpu, u64 data)
266 /* TODO: reserve bits check */
267 if (irqchip_in_kernel(vcpu->kvm))
268 kvm_lapic_set_base(vcpu, data);
270 vcpu->arch.apic_base = data;
272 EXPORT_SYMBOL_GPL(kvm_set_apic_base);
274 #define EXCPT_BENIGN 0
275 #define EXCPT_CONTRIBUTORY 1
278 static int exception_class(int vector)
288 return EXCPT_CONTRIBUTORY;
295 static void kvm_multiple_exception(struct kvm_vcpu *vcpu,
296 unsigned nr, bool has_error, u32 error_code)
301 if (!vcpu->arch.exception.pending) {
303 vcpu->arch.exception.pending = true;
304 vcpu->arch.exception.has_error_code = has_error;
305 vcpu->arch.exception.nr = nr;
306 vcpu->arch.exception.error_code = error_code;
310 /* to check exception */
311 prev_nr = vcpu->arch.exception.nr;
312 if (prev_nr == DF_VECTOR) {
313 /* triple fault -> shutdown */
314 set_bit(KVM_REQ_TRIPLE_FAULT, &vcpu->requests);
317 class1 = exception_class(prev_nr);
318 class2 = exception_class(nr);
319 if ((class1 == EXCPT_CONTRIBUTORY && class2 == EXCPT_CONTRIBUTORY)
320 || (class1 == EXCPT_PF && class2 != EXCPT_BENIGN)) {
321 /* generate double fault per SDM Table 5-5 */
322 vcpu->arch.exception.pending = true;
323 vcpu->arch.exception.has_error_code = true;
324 vcpu->arch.exception.nr = DF_VECTOR;
325 vcpu->arch.exception.error_code = 0;
327 /* replace previous exception with a new one in a hope
328 that instruction re-execution will regenerate lost
333 void kvm_queue_exception(struct kvm_vcpu *vcpu, unsigned nr)
335 kvm_multiple_exception(vcpu, nr, false, 0);
337 EXPORT_SYMBOL_GPL(kvm_queue_exception);
339 void kvm_inject_page_fault(struct kvm_vcpu *vcpu, unsigned long addr,
342 ++vcpu->stat.pf_guest;
343 vcpu->arch.cr2 = addr;
344 kvm_queue_exception_e(vcpu, PF_VECTOR, error_code);
347 void kvm_inject_nmi(struct kvm_vcpu *vcpu)
349 vcpu->arch.nmi_pending = 1;
351 EXPORT_SYMBOL_GPL(kvm_inject_nmi);
353 void kvm_queue_exception_e(struct kvm_vcpu *vcpu, unsigned nr, u32 error_code)
355 kvm_multiple_exception(vcpu, nr, true, error_code);
357 EXPORT_SYMBOL_GPL(kvm_queue_exception_e);
360 * Checks if cpl <= required_cpl; if true, return true. Otherwise queue
361 * a #GP and return false.
363 bool kvm_require_cpl(struct kvm_vcpu *vcpu, int required_cpl)
365 if (kvm_x86_ops->get_cpl(vcpu) <= required_cpl)
367 kvm_queue_exception_e(vcpu, GP_VECTOR, 0);
370 EXPORT_SYMBOL_GPL(kvm_require_cpl);
373 * Load the pae pdptrs. Return true is they are all valid.
375 int load_pdptrs(struct kvm_vcpu *vcpu, unsigned long cr3)
377 gfn_t pdpt_gfn = cr3 >> PAGE_SHIFT;
378 unsigned offset = ((cr3 & (PAGE_SIZE-1)) >> 5) << 2;
381 u64 pdpte[ARRAY_SIZE(vcpu->arch.pdptrs)];
383 ret = kvm_read_guest_page(vcpu->kvm, pdpt_gfn, pdpte,
384 offset * sizeof(u64), sizeof(pdpte));
389 for (i = 0; i < ARRAY_SIZE(pdpte); ++i) {
390 if (is_present_gpte(pdpte[i]) &&
391 (pdpte[i] & vcpu->arch.mmu.rsvd_bits_mask[0][2])) {
398 memcpy(vcpu->arch.pdptrs, pdpte, sizeof(vcpu->arch.pdptrs));
399 __set_bit(VCPU_EXREG_PDPTR,
400 (unsigned long *)&vcpu->arch.regs_avail);
401 __set_bit(VCPU_EXREG_PDPTR,
402 (unsigned long *)&vcpu->arch.regs_dirty);
407 EXPORT_SYMBOL_GPL(load_pdptrs);
409 static bool pdptrs_changed(struct kvm_vcpu *vcpu)
411 u64 pdpte[ARRAY_SIZE(vcpu->arch.pdptrs)];
415 if (is_long_mode(vcpu) || !is_pae(vcpu))
418 if (!test_bit(VCPU_EXREG_PDPTR,
419 (unsigned long *)&vcpu->arch.regs_avail))
422 r = kvm_read_guest(vcpu->kvm, vcpu->arch.cr3 & ~31u, pdpte, sizeof(pdpte));
425 changed = memcmp(pdpte, vcpu->arch.pdptrs, sizeof(pdpte)) != 0;
431 void kvm_set_cr0(struct kvm_vcpu *vcpu, unsigned long cr0)
436 if (cr0 & 0xffffffff00000000UL) {
437 kvm_inject_gp(vcpu, 0);
442 cr0 &= ~CR0_RESERVED_BITS;
444 if ((cr0 & X86_CR0_NW) && !(cr0 & X86_CR0_CD)) {
445 kvm_inject_gp(vcpu, 0);
449 if ((cr0 & X86_CR0_PG) && !(cr0 & X86_CR0_PE)) {
450 kvm_inject_gp(vcpu, 0);
454 if (!is_paging(vcpu) && (cr0 & X86_CR0_PG)) {
456 if ((vcpu->arch.efer & EFER_LME)) {
460 kvm_inject_gp(vcpu, 0);
463 kvm_x86_ops->get_cs_db_l_bits(vcpu, &cs_db, &cs_l);
465 kvm_inject_gp(vcpu, 0);
471 if (is_pae(vcpu) && !load_pdptrs(vcpu, vcpu->arch.cr3)) {
472 kvm_inject_gp(vcpu, 0);
478 kvm_x86_ops->set_cr0(vcpu, cr0);
479 vcpu->arch.cr0 = cr0;
481 kvm_mmu_reset_context(vcpu);
484 EXPORT_SYMBOL_GPL(kvm_set_cr0);
486 void kvm_lmsw(struct kvm_vcpu *vcpu, unsigned long msw)
488 kvm_set_cr0(vcpu, kvm_read_cr0_bits(vcpu, ~0x0ful) | (msw & 0x0f));
490 EXPORT_SYMBOL_GPL(kvm_lmsw);
492 void kvm_set_cr4(struct kvm_vcpu *vcpu, unsigned long cr4)
494 unsigned long old_cr4 = kvm_read_cr4(vcpu);
495 unsigned long pdptr_bits = X86_CR4_PGE | X86_CR4_PSE | X86_CR4_PAE;
497 if (cr4 & CR4_RESERVED_BITS) {
498 kvm_inject_gp(vcpu, 0);
502 if (is_long_mode(vcpu)) {
503 if (!(cr4 & X86_CR4_PAE)) {
504 kvm_inject_gp(vcpu, 0);
507 } else if (is_paging(vcpu) && (cr4 & X86_CR4_PAE)
508 && ((cr4 ^ old_cr4) & pdptr_bits)
509 && !load_pdptrs(vcpu, vcpu->arch.cr3)) {
510 kvm_inject_gp(vcpu, 0);
514 if (cr4 & X86_CR4_VMXE) {
515 kvm_inject_gp(vcpu, 0);
518 kvm_x86_ops->set_cr4(vcpu, cr4);
519 vcpu->arch.cr4 = cr4;
520 vcpu->arch.mmu.base_role.cr4_pge = (cr4 & X86_CR4_PGE) && !tdp_enabled;
521 kvm_mmu_reset_context(vcpu);
523 EXPORT_SYMBOL_GPL(kvm_set_cr4);
525 void kvm_set_cr3(struct kvm_vcpu *vcpu, unsigned long cr3)
527 if (cr3 == vcpu->arch.cr3 && !pdptrs_changed(vcpu)) {
528 kvm_mmu_sync_roots(vcpu);
529 kvm_mmu_flush_tlb(vcpu);
533 if (is_long_mode(vcpu)) {
534 if (cr3 & CR3_L_MODE_RESERVED_BITS) {
535 kvm_inject_gp(vcpu, 0);
540 if (cr3 & CR3_PAE_RESERVED_BITS) {
541 kvm_inject_gp(vcpu, 0);
544 if (is_paging(vcpu) && !load_pdptrs(vcpu, cr3)) {
545 kvm_inject_gp(vcpu, 0);
550 * We don't check reserved bits in nonpae mode, because
551 * this isn't enforced, and VMware depends on this.
556 * Does the new cr3 value map to physical memory? (Note, we
557 * catch an invalid cr3 even in real-mode, because it would
558 * cause trouble later on when we turn on paging anyway.)
560 * A real CPU would silently accept an invalid cr3 and would
561 * attempt to use it - with largely undefined (and often hard
562 * to debug) behavior on the guest side.
564 if (unlikely(!gfn_to_memslot(vcpu->kvm, cr3 >> PAGE_SHIFT)))
565 kvm_inject_gp(vcpu, 0);
567 vcpu->arch.cr3 = cr3;
568 vcpu->arch.mmu.new_cr3(vcpu);
571 EXPORT_SYMBOL_GPL(kvm_set_cr3);
573 void kvm_set_cr8(struct kvm_vcpu *vcpu, unsigned long cr8)
575 if (cr8 & CR8_RESERVED_BITS) {
576 kvm_inject_gp(vcpu, 0);
579 if (irqchip_in_kernel(vcpu->kvm))
580 kvm_lapic_set_tpr(vcpu, cr8);
582 vcpu->arch.cr8 = cr8;
584 EXPORT_SYMBOL_GPL(kvm_set_cr8);
586 unsigned long kvm_get_cr8(struct kvm_vcpu *vcpu)
588 if (irqchip_in_kernel(vcpu->kvm))
589 return kvm_lapic_get_cr8(vcpu);
591 return vcpu->arch.cr8;
593 EXPORT_SYMBOL_GPL(kvm_get_cr8);
595 static inline u32 bit(int bitno)
597 return 1 << (bitno & 31);
601 * List of msr numbers which we expose to userspace through KVM_GET_MSRS
602 * and KVM_SET_MSRS, and KVM_GET_MSR_INDEX_LIST.
604 * This list is modified at module load time to reflect the
605 * capabilities of the host cpu. This capabilities test skips MSRs that are
606 * kvm-specific. Those are put in the beginning of the list.
609 #define KVM_SAVE_MSRS_BEGIN 5
610 static u32 msrs_to_save[] = {
611 MSR_KVM_SYSTEM_TIME, MSR_KVM_WALL_CLOCK,
612 HV_X64_MSR_GUEST_OS_ID, HV_X64_MSR_HYPERCALL,
613 HV_X64_MSR_APIC_ASSIST_PAGE,
614 MSR_IA32_SYSENTER_CS, MSR_IA32_SYSENTER_ESP, MSR_IA32_SYSENTER_EIP,
617 MSR_CSTAR, MSR_KERNEL_GS_BASE, MSR_SYSCALL_MASK, MSR_LSTAR,
619 MSR_IA32_TSC, MSR_IA32_PERF_STATUS, MSR_IA32_CR_PAT, MSR_VM_HSAVE_PA
622 static unsigned num_msrs_to_save;
624 static u32 emulated_msrs[] = {
625 MSR_IA32_MISC_ENABLE,
628 static void set_efer(struct kvm_vcpu *vcpu, u64 efer)
630 if (efer & efer_reserved_bits) {
631 kvm_inject_gp(vcpu, 0);
636 && (vcpu->arch.efer & EFER_LME) != (efer & EFER_LME)) {
637 kvm_inject_gp(vcpu, 0);
641 if (efer & EFER_FFXSR) {
642 struct kvm_cpuid_entry2 *feat;
644 feat = kvm_find_cpuid_entry(vcpu, 0x80000001, 0);
645 if (!feat || !(feat->edx & bit(X86_FEATURE_FXSR_OPT))) {
646 kvm_inject_gp(vcpu, 0);
651 if (efer & EFER_SVME) {
652 struct kvm_cpuid_entry2 *feat;
654 feat = kvm_find_cpuid_entry(vcpu, 0x80000001, 0);
655 if (!feat || !(feat->ecx & bit(X86_FEATURE_SVM))) {
656 kvm_inject_gp(vcpu, 0);
661 kvm_x86_ops->set_efer(vcpu, efer);
664 efer |= vcpu->arch.efer & EFER_LMA;
666 vcpu->arch.efer = efer;
668 vcpu->arch.mmu.base_role.nxe = (efer & EFER_NX) && !tdp_enabled;
669 kvm_mmu_reset_context(vcpu);
672 void kvm_enable_efer_bits(u64 mask)
674 efer_reserved_bits &= ~mask;
676 EXPORT_SYMBOL_GPL(kvm_enable_efer_bits);
680 * Writes msr value into into the appropriate "register".
681 * Returns 0 on success, non-0 otherwise.
682 * Assumes vcpu_load() was already called.
684 int kvm_set_msr(struct kvm_vcpu *vcpu, u32 msr_index, u64 data)
686 return kvm_x86_ops->set_msr(vcpu, msr_index, data);
690 * Adapt set_msr() to msr_io()'s calling convention
692 static int do_set_msr(struct kvm_vcpu *vcpu, unsigned index, u64 *data)
694 return kvm_set_msr(vcpu, index, *data);
697 static void kvm_write_wall_clock(struct kvm *kvm, gpa_t wall_clock)
700 struct pvclock_wall_clock wc;
701 struct timespec boot;
708 kvm_write_guest(kvm, wall_clock, &version, sizeof(version));
711 * The guest calculates current wall clock time by adding
712 * system time (updated by kvm_write_guest_time below) to the
713 * wall clock specified here. guest system time equals host
714 * system time for us, thus we must fill in host boot time here.
718 wc.sec = boot.tv_sec;
719 wc.nsec = boot.tv_nsec;
720 wc.version = version;
722 kvm_write_guest(kvm, wall_clock, &wc, sizeof(wc));
725 kvm_write_guest(kvm, wall_clock, &version, sizeof(version));
728 static uint32_t div_frac(uint32_t dividend, uint32_t divisor)
730 uint32_t quotient, remainder;
732 /* Don't try to replace with do_div(), this one calculates
733 * "(dividend << 32) / divisor" */
735 : "=a" (quotient), "=d" (remainder)
736 : "0" (0), "1" (dividend), "r" (divisor) );
740 static void kvm_set_time_scale(uint32_t tsc_khz, struct pvclock_vcpu_time_info *hv_clock)
742 uint64_t nsecs = 1000000000LL;
747 tps64 = tsc_khz * 1000LL;
748 while (tps64 > nsecs*2) {
753 tps32 = (uint32_t)tps64;
754 while (tps32 <= (uint32_t)nsecs) {
759 hv_clock->tsc_shift = shift;
760 hv_clock->tsc_to_system_mul = div_frac(nsecs, tps32);
762 pr_debug("%s: tsc_khz %u, tsc_shift %d, tsc_mul %u\n",
763 __func__, tsc_khz, hv_clock->tsc_shift,
764 hv_clock->tsc_to_system_mul);
767 static DEFINE_PER_CPU(unsigned long, cpu_tsc_khz);
769 static void kvm_write_guest_time(struct kvm_vcpu *v)
773 struct kvm_vcpu_arch *vcpu = &v->arch;
775 unsigned long this_tsc_khz;
777 if ((!vcpu->time_page))
780 this_tsc_khz = get_cpu_var(cpu_tsc_khz);
781 if (unlikely(vcpu->hv_clock_tsc_khz != this_tsc_khz)) {
782 kvm_set_time_scale(this_tsc_khz, &vcpu->hv_clock);
783 vcpu->hv_clock_tsc_khz = this_tsc_khz;
785 put_cpu_var(cpu_tsc_khz);
787 /* Keep irq disabled to prevent changes to the clock */
788 local_irq_save(flags);
789 kvm_get_msr(v, MSR_IA32_TSC, &vcpu->hv_clock.tsc_timestamp);
791 monotonic_to_bootbased(&ts);
792 local_irq_restore(flags);
794 /* With all the info we got, fill in the values */
796 vcpu->hv_clock.system_time = ts.tv_nsec +
797 (NSEC_PER_SEC * (u64)ts.tv_sec) + v->kvm->arch.kvmclock_offset;
800 * The interface expects us to write an even number signaling that the
801 * update is finished. Since the guest won't see the intermediate
802 * state, we just increase by 2 at the end.
804 vcpu->hv_clock.version += 2;
806 shared_kaddr = kmap_atomic(vcpu->time_page, KM_USER0);
808 memcpy(shared_kaddr + vcpu->time_offset, &vcpu->hv_clock,
809 sizeof(vcpu->hv_clock));
811 kunmap_atomic(shared_kaddr, KM_USER0);
813 mark_page_dirty(v->kvm, vcpu->time >> PAGE_SHIFT);
816 static int kvm_request_guest_time_update(struct kvm_vcpu *v)
818 struct kvm_vcpu_arch *vcpu = &v->arch;
820 if (!vcpu->time_page)
822 set_bit(KVM_REQ_KVMCLOCK_UPDATE, &v->requests);
826 static bool msr_mtrr_valid(unsigned msr)
829 case 0x200 ... 0x200 + 2 * KVM_NR_VAR_MTRR - 1:
830 case MSR_MTRRfix64K_00000:
831 case MSR_MTRRfix16K_80000:
832 case MSR_MTRRfix16K_A0000:
833 case MSR_MTRRfix4K_C0000:
834 case MSR_MTRRfix4K_C8000:
835 case MSR_MTRRfix4K_D0000:
836 case MSR_MTRRfix4K_D8000:
837 case MSR_MTRRfix4K_E0000:
838 case MSR_MTRRfix4K_E8000:
839 case MSR_MTRRfix4K_F0000:
840 case MSR_MTRRfix4K_F8000:
841 case MSR_MTRRdefType:
842 case MSR_IA32_CR_PAT:
850 static bool valid_pat_type(unsigned t)
852 return t < 8 && (1 << t) & 0xf3; /* 0, 1, 4, 5, 6, 7 */
855 static bool valid_mtrr_type(unsigned t)
857 return t < 8 && (1 << t) & 0x73; /* 0, 1, 4, 5, 6 */
860 static bool mtrr_valid(struct kvm_vcpu *vcpu, u32 msr, u64 data)
864 if (!msr_mtrr_valid(msr))
867 if (msr == MSR_IA32_CR_PAT) {
868 for (i = 0; i < 8; i++)
869 if (!valid_pat_type((data >> (i * 8)) & 0xff))
872 } else if (msr == MSR_MTRRdefType) {
875 return valid_mtrr_type(data & 0xff);
876 } else if (msr >= MSR_MTRRfix64K_00000 && msr <= MSR_MTRRfix4K_F8000) {
877 for (i = 0; i < 8 ; i++)
878 if (!valid_mtrr_type((data >> (i * 8)) & 0xff))
884 return valid_mtrr_type(data & 0xff);
887 static int set_msr_mtrr(struct kvm_vcpu *vcpu, u32 msr, u64 data)
889 u64 *p = (u64 *)&vcpu->arch.mtrr_state.fixed_ranges;
891 if (!mtrr_valid(vcpu, msr, data))
894 if (msr == MSR_MTRRdefType) {
895 vcpu->arch.mtrr_state.def_type = data;
896 vcpu->arch.mtrr_state.enabled = (data & 0xc00) >> 10;
897 } else if (msr == MSR_MTRRfix64K_00000)
899 else if (msr == MSR_MTRRfix16K_80000 || msr == MSR_MTRRfix16K_A0000)
900 p[1 + msr - MSR_MTRRfix16K_80000] = data;
901 else if (msr >= MSR_MTRRfix4K_C0000 && msr <= MSR_MTRRfix4K_F8000)
902 p[3 + msr - MSR_MTRRfix4K_C0000] = data;
903 else if (msr == MSR_IA32_CR_PAT)
904 vcpu->arch.pat = data;
905 else { /* Variable MTRRs */
906 int idx, is_mtrr_mask;
909 idx = (msr - 0x200) / 2;
910 is_mtrr_mask = msr - 0x200 - 2 * idx;
913 (u64 *)&vcpu->arch.mtrr_state.var_ranges[idx].base_lo;
916 (u64 *)&vcpu->arch.mtrr_state.var_ranges[idx].mask_lo;
920 kvm_mmu_reset_context(vcpu);
924 static int set_msr_mce(struct kvm_vcpu *vcpu, u32 msr, u64 data)
926 u64 mcg_cap = vcpu->arch.mcg_cap;
927 unsigned bank_num = mcg_cap & 0xff;
930 case MSR_IA32_MCG_STATUS:
931 vcpu->arch.mcg_status = data;
933 case MSR_IA32_MCG_CTL:
934 if (!(mcg_cap & MCG_CTL_P))
936 if (data != 0 && data != ~(u64)0)
938 vcpu->arch.mcg_ctl = data;
941 if (msr >= MSR_IA32_MC0_CTL &&
942 msr < MSR_IA32_MC0_CTL + 4 * bank_num) {
943 u32 offset = msr - MSR_IA32_MC0_CTL;
944 /* only 0 or all 1s can be written to IA32_MCi_CTL
945 * some Linux kernels though clear bit 10 in bank 4 to
946 * workaround a BIOS/GART TBL issue on AMD K8s, ignore
947 * this to avoid an uncatched #GP in the guest
949 if ((offset & 0x3) == 0 &&
950 data != 0 && (data | (1 << 10)) != ~(u64)0)
952 vcpu->arch.mce_banks[offset] = data;
960 static int xen_hvm_config(struct kvm_vcpu *vcpu, u64 data)
962 struct kvm *kvm = vcpu->kvm;
963 int lm = is_long_mode(vcpu);
964 u8 *blob_addr = lm ? (u8 *)(long)kvm->arch.xen_hvm_config.blob_addr_64
965 : (u8 *)(long)kvm->arch.xen_hvm_config.blob_addr_32;
966 u8 blob_size = lm ? kvm->arch.xen_hvm_config.blob_size_64
967 : kvm->arch.xen_hvm_config.blob_size_32;
968 u32 page_num = data & ~PAGE_MASK;
969 u64 page_addr = data & PAGE_MASK;
974 if (page_num >= blob_size)
977 page = kzalloc(PAGE_SIZE, GFP_KERNEL);
981 if (copy_from_user(page, blob_addr + (page_num * PAGE_SIZE), PAGE_SIZE))
983 if (kvm_write_guest(kvm, page_addr, page, PAGE_SIZE))
992 static bool kvm_hv_hypercall_enabled(struct kvm *kvm)
994 return kvm->arch.hv_hypercall & HV_X64_MSR_HYPERCALL_ENABLE;
997 static bool kvm_hv_msr_partition_wide(u32 msr)
1001 case HV_X64_MSR_GUEST_OS_ID:
1002 case HV_X64_MSR_HYPERCALL:
1010 static int set_msr_hyperv_pw(struct kvm_vcpu *vcpu, u32 msr, u64 data)
1012 struct kvm *kvm = vcpu->kvm;
1015 case HV_X64_MSR_GUEST_OS_ID:
1016 kvm->arch.hv_guest_os_id = data;
1017 /* setting guest os id to zero disables hypercall page */
1018 if (!kvm->arch.hv_guest_os_id)
1019 kvm->arch.hv_hypercall &= ~HV_X64_MSR_HYPERCALL_ENABLE;
1021 case HV_X64_MSR_HYPERCALL: {
1026 /* if guest os id is not set hypercall should remain disabled */
1027 if (!kvm->arch.hv_guest_os_id)
1029 if (!(data & HV_X64_MSR_HYPERCALL_ENABLE)) {
1030 kvm->arch.hv_hypercall = data;
1033 gfn = data >> HV_X64_MSR_HYPERCALL_PAGE_ADDRESS_SHIFT;
1034 addr = gfn_to_hva(kvm, gfn);
1035 if (kvm_is_error_hva(addr))
1037 kvm_x86_ops->patch_hypercall(vcpu, instructions);
1038 ((unsigned char *)instructions)[3] = 0xc3; /* ret */
1039 if (copy_to_user((void __user *)addr, instructions, 4))
1041 kvm->arch.hv_hypercall = data;
1045 pr_unimpl(vcpu, "HYPER-V unimplemented wrmsr: 0x%x "
1046 "data 0x%llx\n", msr, data);
1052 static int set_msr_hyperv(struct kvm_vcpu *vcpu, u32 msr, u64 data)
1055 case HV_X64_MSR_APIC_ASSIST_PAGE: {
1058 if (!(data & HV_X64_MSR_APIC_ASSIST_PAGE_ENABLE)) {
1059 vcpu->arch.hv_vapic = data;
1062 addr = gfn_to_hva(vcpu->kvm, data >>
1063 HV_X64_MSR_APIC_ASSIST_PAGE_ADDRESS_SHIFT);
1064 if (kvm_is_error_hva(addr))
1066 if (clear_user((void __user *)addr, PAGE_SIZE))
1068 vcpu->arch.hv_vapic = data;
1071 case HV_X64_MSR_EOI:
1072 return kvm_hv_vapic_msr_write(vcpu, APIC_EOI, data);
1073 case HV_X64_MSR_ICR:
1074 return kvm_hv_vapic_msr_write(vcpu, APIC_ICR, data);
1075 case HV_X64_MSR_TPR:
1076 return kvm_hv_vapic_msr_write(vcpu, APIC_TASKPRI, data);
1078 pr_unimpl(vcpu, "HYPER-V unimplemented wrmsr: 0x%x "
1079 "data 0x%llx\n", msr, data);
1086 int kvm_set_msr_common(struct kvm_vcpu *vcpu, u32 msr, u64 data)
1090 set_efer(vcpu, data);
1093 data &= ~(u64)0x40; /* ignore flush filter disable */
1095 pr_unimpl(vcpu, "unimplemented HWCR wrmsr: 0x%llx\n",
1100 case MSR_FAM10H_MMIO_CONF_BASE:
1102 pr_unimpl(vcpu, "unimplemented MMIO_CONF_BASE wrmsr: "
1107 case MSR_AMD64_NB_CFG:
1109 case MSR_IA32_DEBUGCTLMSR:
1111 /* We support the non-activated case already */
1113 } else if (data & ~(DEBUGCTLMSR_LBR | DEBUGCTLMSR_BTF)) {
1114 /* Values other than LBR and BTF are vendor-specific,
1115 thus reserved and should throw a #GP */
1118 pr_unimpl(vcpu, "%s: MSR_IA32_DEBUGCTLMSR 0x%llx, nop\n",
1121 case MSR_IA32_UCODE_REV:
1122 case MSR_IA32_UCODE_WRITE:
1123 case MSR_VM_HSAVE_PA:
1124 case MSR_AMD64_PATCH_LOADER:
1126 case 0x200 ... 0x2ff:
1127 return set_msr_mtrr(vcpu, msr, data);
1128 case MSR_IA32_APICBASE:
1129 kvm_set_apic_base(vcpu, data);
1131 case APIC_BASE_MSR ... APIC_BASE_MSR + 0x3ff:
1132 return kvm_x2apic_msr_write(vcpu, msr, data);
1133 case MSR_IA32_MISC_ENABLE:
1134 vcpu->arch.ia32_misc_enable_msr = data;
1136 case MSR_KVM_WALL_CLOCK:
1137 vcpu->kvm->arch.wall_clock = data;
1138 kvm_write_wall_clock(vcpu->kvm, data);
1140 case MSR_KVM_SYSTEM_TIME: {
1141 if (vcpu->arch.time_page) {
1142 kvm_release_page_dirty(vcpu->arch.time_page);
1143 vcpu->arch.time_page = NULL;
1146 vcpu->arch.time = data;
1148 /* we verify if the enable bit is set... */
1152 /* ...but clean it before doing the actual write */
1153 vcpu->arch.time_offset = data & ~(PAGE_MASK | 1);
1155 vcpu->arch.time_page =
1156 gfn_to_page(vcpu->kvm, data >> PAGE_SHIFT);
1158 if (is_error_page(vcpu->arch.time_page)) {
1159 kvm_release_page_clean(vcpu->arch.time_page);
1160 vcpu->arch.time_page = NULL;
1163 kvm_request_guest_time_update(vcpu);
1166 case MSR_IA32_MCG_CTL:
1167 case MSR_IA32_MCG_STATUS:
1168 case MSR_IA32_MC0_CTL ... MSR_IA32_MC0_CTL + 4 * KVM_MAX_MCE_BANKS - 1:
1169 return set_msr_mce(vcpu, msr, data);
1171 /* Performance counters are not protected by a CPUID bit,
1172 * so we should check all of them in the generic path for the sake of
1173 * cross vendor migration.
1174 * Writing a zero into the event select MSRs disables them,
1175 * which we perfectly emulate ;-). Any other value should be at least
1176 * reported, some guests depend on them.
1178 case MSR_P6_EVNTSEL0:
1179 case MSR_P6_EVNTSEL1:
1180 case MSR_K7_EVNTSEL0:
1181 case MSR_K7_EVNTSEL1:
1182 case MSR_K7_EVNTSEL2:
1183 case MSR_K7_EVNTSEL3:
1185 pr_unimpl(vcpu, "unimplemented perfctr wrmsr: "
1186 "0x%x data 0x%llx\n", msr, data);
1188 /* at least RHEL 4 unconditionally writes to the perfctr registers,
1189 * so we ignore writes to make it happy.
1191 case MSR_P6_PERFCTR0:
1192 case MSR_P6_PERFCTR1:
1193 case MSR_K7_PERFCTR0:
1194 case MSR_K7_PERFCTR1:
1195 case MSR_K7_PERFCTR2:
1196 case MSR_K7_PERFCTR3:
1197 pr_unimpl(vcpu, "unimplemented perfctr wrmsr: "
1198 "0x%x data 0x%llx\n", msr, data);
1200 case HV_X64_MSR_GUEST_OS_ID ... HV_X64_MSR_SINT15:
1201 if (kvm_hv_msr_partition_wide(msr)) {
1203 mutex_lock(&vcpu->kvm->lock);
1204 r = set_msr_hyperv_pw(vcpu, msr, data);
1205 mutex_unlock(&vcpu->kvm->lock);
1208 return set_msr_hyperv(vcpu, msr, data);
1211 if (msr && (msr == vcpu->kvm->arch.xen_hvm_config.msr))
1212 return xen_hvm_config(vcpu, data);
1214 pr_unimpl(vcpu, "unhandled wrmsr: 0x%x data %llx\n",
1218 pr_unimpl(vcpu, "ignored wrmsr: 0x%x data %llx\n",
1225 EXPORT_SYMBOL_GPL(kvm_set_msr_common);
1229 * Reads an msr value (of 'msr_index') into 'pdata'.
1230 * Returns 0 on success, non-0 otherwise.
1231 * Assumes vcpu_load() was already called.
1233 int kvm_get_msr(struct kvm_vcpu *vcpu, u32 msr_index, u64 *pdata)
1235 return kvm_x86_ops->get_msr(vcpu, msr_index, pdata);
1238 static int get_msr_mtrr(struct kvm_vcpu *vcpu, u32 msr, u64 *pdata)
1240 u64 *p = (u64 *)&vcpu->arch.mtrr_state.fixed_ranges;
1242 if (!msr_mtrr_valid(msr))
1245 if (msr == MSR_MTRRdefType)
1246 *pdata = vcpu->arch.mtrr_state.def_type +
1247 (vcpu->arch.mtrr_state.enabled << 10);
1248 else if (msr == MSR_MTRRfix64K_00000)
1250 else if (msr == MSR_MTRRfix16K_80000 || msr == MSR_MTRRfix16K_A0000)
1251 *pdata = p[1 + msr - MSR_MTRRfix16K_80000];
1252 else if (msr >= MSR_MTRRfix4K_C0000 && msr <= MSR_MTRRfix4K_F8000)
1253 *pdata = p[3 + msr - MSR_MTRRfix4K_C0000];
1254 else if (msr == MSR_IA32_CR_PAT)
1255 *pdata = vcpu->arch.pat;
1256 else { /* Variable MTRRs */
1257 int idx, is_mtrr_mask;
1260 idx = (msr - 0x200) / 2;
1261 is_mtrr_mask = msr - 0x200 - 2 * idx;
1264 (u64 *)&vcpu->arch.mtrr_state.var_ranges[idx].base_lo;
1267 (u64 *)&vcpu->arch.mtrr_state.var_ranges[idx].mask_lo;
1274 static int get_msr_mce(struct kvm_vcpu *vcpu, u32 msr, u64 *pdata)
1277 u64 mcg_cap = vcpu->arch.mcg_cap;
1278 unsigned bank_num = mcg_cap & 0xff;
1281 case MSR_IA32_P5_MC_ADDR:
1282 case MSR_IA32_P5_MC_TYPE:
1285 case MSR_IA32_MCG_CAP:
1286 data = vcpu->arch.mcg_cap;
1288 case MSR_IA32_MCG_CTL:
1289 if (!(mcg_cap & MCG_CTL_P))
1291 data = vcpu->arch.mcg_ctl;
1293 case MSR_IA32_MCG_STATUS:
1294 data = vcpu->arch.mcg_status;
1297 if (msr >= MSR_IA32_MC0_CTL &&
1298 msr < MSR_IA32_MC0_CTL + 4 * bank_num) {
1299 u32 offset = msr - MSR_IA32_MC0_CTL;
1300 data = vcpu->arch.mce_banks[offset];
1309 static int get_msr_hyperv_pw(struct kvm_vcpu *vcpu, u32 msr, u64 *pdata)
1312 struct kvm *kvm = vcpu->kvm;
1315 case HV_X64_MSR_GUEST_OS_ID:
1316 data = kvm->arch.hv_guest_os_id;
1318 case HV_X64_MSR_HYPERCALL:
1319 data = kvm->arch.hv_hypercall;
1322 pr_unimpl(vcpu, "Hyper-V unhandled rdmsr: 0x%x\n", msr);
1330 static int get_msr_hyperv(struct kvm_vcpu *vcpu, u32 msr, u64 *pdata)
1335 case HV_X64_MSR_VP_INDEX: {
1338 kvm_for_each_vcpu(r, v, vcpu->kvm)
1343 case HV_X64_MSR_EOI:
1344 return kvm_hv_vapic_msr_read(vcpu, APIC_EOI, pdata);
1345 case HV_X64_MSR_ICR:
1346 return kvm_hv_vapic_msr_read(vcpu, APIC_ICR, pdata);
1347 case HV_X64_MSR_TPR:
1348 return kvm_hv_vapic_msr_read(vcpu, APIC_TASKPRI, pdata);
1350 pr_unimpl(vcpu, "Hyper-V unhandled rdmsr: 0x%x\n", msr);
1357 int kvm_get_msr_common(struct kvm_vcpu *vcpu, u32 msr, u64 *pdata)
1362 case MSR_IA32_PLATFORM_ID:
1363 case MSR_IA32_UCODE_REV:
1364 case MSR_IA32_EBL_CR_POWERON:
1365 case MSR_IA32_DEBUGCTLMSR:
1366 case MSR_IA32_LASTBRANCHFROMIP:
1367 case MSR_IA32_LASTBRANCHTOIP:
1368 case MSR_IA32_LASTINTFROMIP:
1369 case MSR_IA32_LASTINTTOIP:
1372 case MSR_VM_HSAVE_PA:
1373 case MSR_P6_PERFCTR0:
1374 case MSR_P6_PERFCTR1:
1375 case MSR_P6_EVNTSEL0:
1376 case MSR_P6_EVNTSEL1:
1377 case MSR_K7_EVNTSEL0:
1378 case MSR_K7_PERFCTR0:
1379 case MSR_K8_INT_PENDING_MSG:
1380 case MSR_AMD64_NB_CFG:
1381 case MSR_FAM10H_MMIO_CONF_BASE:
1385 data = 0x500 | KVM_NR_VAR_MTRR;
1387 case 0x200 ... 0x2ff:
1388 return get_msr_mtrr(vcpu, msr, pdata);
1389 case 0xcd: /* fsb frequency */
1392 case MSR_IA32_APICBASE:
1393 data = kvm_get_apic_base(vcpu);
1395 case APIC_BASE_MSR ... APIC_BASE_MSR + 0x3ff:
1396 return kvm_x2apic_msr_read(vcpu, msr, pdata);
1398 case MSR_IA32_MISC_ENABLE:
1399 data = vcpu->arch.ia32_misc_enable_msr;
1401 case MSR_IA32_PERF_STATUS:
1402 /* TSC increment by tick */
1404 /* CPU multiplier */
1405 data |= (((uint64_t)4ULL) << 40);
1408 data = vcpu->arch.efer;
1410 case MSR_KVM_WALL_CLOCK:
1411 data = vcpu->kvm->arch.wall_clock;
1413 case MSR_KVM_SYSTEM_TIME:
1414 data = vcpu->arch.time;
1416 case MSR_IA32_P5_MC_ADDR:
1417 case MSR_IA32_P5_MC_TYPE:
1418 case MSR_IA32_MCG_CAP:
1419 case MSR_IA32_MCG_CTL:
1420 case MSR_IA32_MCG_STATUS:
1421 case MSR_IA32_MC0_CTL ... MSR_IA32_MC0_CTL + 4 * KVM_MAX_MCE_BANKS - 1:
1422 return get_msr_mce(vcpu, msr, pdata);
1423 case HV_X64_MSR_GUEST_OS_ID ... HV_X64_MSR_SINT15:
1424 if (kvm_hv_msr_partition_wide(msr)) {
1426 mutex_lock(&vcpu->kvm->lock);
1427 r = get_msr_hyperv_pw(vcpu, msr, pdata);
1428 mutex_unlock(&vcpu->kvm->lock);
1431 return get_msr_hyperv(vcpu, msr, pdata);
1435 pr_unimpl(vcpu, "unhandled rdmsr: 0x%x\n", msr);
1438 pr_unimpl(vcpu, "ignored rdmsr: 0x%x\n", msr);
1446 EXPORT_SYMBOL_GPL(kvm_get_msr_common);
1449 * Read or write a bunch of msrs. All parameters are kernel addresses.
1451 * @return number of msrs set successfully.
1453 static int __msr_io(struct kvm_vcpu *vcpu, struct kvm_msrs *msrs,
1454 struct kvm_msr_entry *entries,
1455 int (*do_msr)(struct kvm_vcpu *vcpu,
1456 unsigned index, u64 *data))
1462 idx = srcu_read_lock(&vcpu->kvm->srcu);
1463 for (i = 0; i < msrs->nmsrs; ++i)
1464 if (do_msr(vcpu, entries[i].index, &entries[i].data))
1466 srcu_read_unlock(&vcpu->kvm->srcu, idx);
1474 * Read or write a bunch of msrs. Parameters are user addresses.
1476 * @return number of msrs set successfully.
1478 static int msr_io(struct kvm_vcpu *vcpu, struct kvm_msrs __user *user_msrs,
1479 int (*do_msr)(struct kvm_vcpu *vcpu,
1480 unsigned index, u64 *data),
1483 struct kvm_msrs msrs;
1484 struct kvm_msr_entry *entries;
1489 if (copy_from_user(&msrs, user_msrs, sizeof msrs))
1493 if (msrs.nmsrs >= MAX_IO_MSRS)
1497 size = sizeof(struct kvm_msr_entry) * msrs.nmsrs;
1498 entries = vmalloc(size);
1503 if (copy_from_user(entries, user_msrs->entries, size))
1506 r = n = __msr_io(vcpu, &msrs, entries, do_msr);
1511 if (writeback && copy_to_user(user_msrs->entries, entries, size))
1522 int kvm_dev_ioctl_check_extension(long ext)
1527 case KVM_CAP_IRQCHIP:
1529 case KVM_CAP_MMU_SHADOW_CACHE_CONTROL:
1530 case KVM_CAP_SET_TSS_ADDR:
1531 case KVM_CAP_EXT_CPUID:
1532 case KVM_CAP_CLOCKSOURCE:
1534 case KVM_CAP_NOP_IO_DELAY:
1535 case KVM_CAP_MP_STATE:
1536 case KVM_CAP_SYNC_MMU:
1537 case KVM_CAP_REINJECT_CONTROL:
1538 case KVM_CAP_IRQ_INJECT_STATUS:
1539 case KVM_CAP_ASSIGN_DEV_IRQ:
1541 case KVM_CAP_IOEVENTFD:
1543 case KVM_CAP_PIT_STATE2:
1544 case KVM_CAP_SET_IDENTITY_MAP_ADDR:
1545 case KVM_CAP_XEN_HVM:
1546 case KVM_CAP_ADJUST_CLOCK:
1547 case KVM_CAP_VCPU_EVENTS:
1548 case KVM_CAP_HYPERV:
1549 case KVM_CAP_HYPERV_VAPIC:
1550 case KVM_CAP_HYPERV_SPIN:
1551 case KVM_CAP_PCI_SEGMENT:
1552 case KVM_CAP_X86_ROBUST_SINGLESTEP:
1555 case KVM_CAP_COALESCED_MMIO:
1556 r = KVM_COALESCED_MMIO_PAGE_OFFSET;
1559 r = !kvm_x86_ops->cpu_has_accelerated_tpr();
1561 case KVM_CAP_NR_VCPUS:
1564 case KVM_CAP_NR_MEMSLOTS:
1565 r = KVM_MEMORY_SLOTS;
1567 case KVM_CAP_PV_MMU: /* obsolete */
1574 r = KVM_MAX_MCE_BANKS;
1584 long kvm_arch_dev_ioctl(struct file *filp,
1585 unsigned int ioctl, unsigned long arg)
1587 void __user *argp = (void __user *)arg;
1591 case KVM_GET_MSR_INDEX_LIST: {
1592 struct kvm_msr_list __user *user_msr_list = argp;
1593 struct kvm_msr_list msr_list;
1597 if (copy_from_user(&msr_list, user_msr_list, sizeof msr_list))
1600 msr_list.nmsrs = num_msrs_to_save + ARRAY_SIZE(emulated_msrs);
1601 if (copy_to_user(user_msr_list, &msr_list, sizeof msr_list))
1604 if (n < msr_list.nmsrs)
1607 if (copy_to_user(user_msr_list->indices, &msrs_to_save,
1608 num_msrs_to_save * sizeof(u32)))
1610 if (copy_to_user(user_msr_list->indices + num_msrs_to_save,
1612 ARRAY_SIZE(emulated_msrs) * sizeof(u32)))
1617 case KVM_GET_SUPPORTED_CPUID: {
1618 struct kvm_cpuid2 __user *cpuid_arg = argp;
1619 struct kvm_cpuid2 cpuid;
1622 if (copy_from_user(&cpuid, cpuid_arg, sizeof cpuid))
1624 r = kvm_dev_ioctl_get_supported_cpuid(&cpuid,
1625 cpuid_arg->entries);
1630 if (copy_to_user(cpuid_arg, &cpuid, sizeof cpuid))
1635 case KVM_X86_GET_MCE_CAP_SUPPORTED: {
1638 mce_cap = KVM_MCE_CAP_SUPPORTED;
1640 if (copy_to_user(argp, &mce_cap, sizeof mce_cap))
1652 void kvm_arch_vcpu_load(struct kvm_vcpu *vcpu, int cpu)
1654 kvm_x86_ops->vcpu_load(vcpu, cpu);
1655 if (unlikely(per_cpu(cpu_tsc_khz, cpu) == 0)) {
1656 unsigned long khz = cpufreq_quick_get(cpu);
1659 per_cpu(cpu_tsc_khz, cpu) = khz;
1661 kvm_request_guest_time_update(vcpu);
1664 void kvm_arch_vcpu_put(struct kvm_vcpu *vcpu)
1666 kvm_put_guest_fpu(vcpu);
1667 kvm_x86_ops->vcpu_put(vcpu);
1670 static int is_efer_nx(void)
1672 unsigned long long efer = 0;
1674 rdmsrl_safe(MSR_EFER, &efer);
1675 return efer & EFER_NX;
1678 static void cpuid_fix_nx_cap(struct kvm_vcpu *vcpu)
1681 struct kvm_cpuid_entry2 *e, *entry;
1684 for (i = 0; i < vcpu->arch.cpuid_nent; ++i) {
1685 e = &vcpu->arch.cpuid_entries[i];
1686 if (e->function == 0x80000001) {
1691 if (entry && (entry->edx & (1 << 20)) && !is_efer_nx()) {
1692 entry->edx &= ~(1 << 20);
1693 printk(KERN_INFO "kvm: guest NX capability removed\n");
1697 /* when an old userspace process fills a new kernel module */
1698 static int kvm_vcpu_ioctl_set_cpuid(struct kvm_vcpu *vcpu,
1699 struct kvm_cpuid *cpuid,
1700 struct kvm_cpuid_entry __user *entries)
1703 struct kvm_cpuid_entry *cpuid_entries;
1706 if (cpuid->nent > KVM_MAX_CPUID_ENTRIES)
1709 cpuid_entries = vmalloc(sizeof(struct kvm_cpuid_entry) * cpuid->nent);
1713 if (copy_from_user(cpuid_entries, entries,
1714 cpuid->nent * sizeof(struct kvm_cpuid_entry)))
1717 for (i = 0; i < cpuid->nent; i++) {
1718 vcpu->arch.cpuid_entries[i].function = cpuid_entries[i].function;
1719 vcpu->arch.cpuid_entries[i].eax = cpuid_entries[i].eax;
1720 vcpu->arch.cpuid_entries[i].ebx = cpuid_entries[i].ebx;
1721 vcpu->arch.cpuid_entries[i].ecx = cpuid_entries[i].ecx;
1722 vcpu->arch.cpuid_entries[i].edx = cpuid_entries[i].edx;
1723 vcpu->arch.cpuid_entries[i].index = 0;
1724 vcpu->arch.cpuid_entries[i].flags = 0;
1725 vcpu->arch.cpuid_entries[i].padding[0] = 0;
1726 vcpu->arch.cpuid_entries[i].padding[1] = 0;
1727 vcpu->arch.cpuid_entries[i].padding[2] = 0;
1729 vcpu->arch.cpuid_nent = cpuid->nent;
1730 cpuid_fix_nx_cap(vcpu);
1732 kvm_apic_set_version(vcpu);
1733 kvm_x86_ops->cpuid_update(vcpu);
1737 vfree(cpuid_entries);
1742 static int kvm_vcpu_ioctl_set_cpuid2(struct kvm_vcpu *vcpu,
1743 struct kvm_cpuid2 *cpuid,
1744 struct kvm_cpuid_entry2 __user *entries)
1749 if (cpuid->nent > KVM_MAX_CPUID_ENTRIES)
1752 if (copy_from_user(&vcpu->arch.cpuid_entries, entries,
1753 cpuid->nent * sizeof(struct kvm_cpuid_entry2)))
1756 vcpu->arch.cpuid_nent = cpuid->nent;
1757 kvm_apic_set_version(vcpu);
1758 kvm_x86_ops->cpuid_update(vcpu);
1766 static int kvm_vcpu_ioctl_get_cpuid2(struct kvm_vcpu *vcpu,
1767 struct kvm_cpuid2 *cpuid,
1768 struct kvm_cpuid_entry2 __user *entries)
1773 if (cpuid->nent < vcpu->arch.cpuid_nent)
1776 if (copy_to_user(entries, &vcpu->arch.cpuid_entries,
1777 vcpu->arch.cpuid_nent * sizeof(struct kvm_cpuid_entry2)))
1782 cpuid->nent = vcpu->arch.cpuid_nent;
1786 static void do_cpuid_1_ent(struct kvm_cpuid_entry2 *entry, u32 function,
1789 entry->function = function;
1790 entry->index = index;
1791 cpuid_count(entry->function, entry->index,
1792 &entry->eax, &entry->ebx, &entry->ecx, &entry->edx);
1796 #define F(x) bit(X86_FEATURE_##x)
1798 static void do_cpuid_ent(struct kvm_cpuid_entry2 *entry, u32 function,
1799 u32 index, int *nent, int maxnent)
1801 unsigned f_nx = is_efer_nx() ? F(NX) : 0;
1802 #ifdef CONFIG_X86_64
1803 unsigned f_gbpages = (kvm_x86_ops->get_lpage_level() == PT_PDPE_LEVEL)
1805 unsigned f_lm = F(LM);
1807 unsigned f_gbpages = 0;
1810 unsigned f_rdtscp = kvm_x86_ops->rdtscp_supported() ? F(RDTSCP) : 0;
1813 const u32 kvm_supported_word0_x86_features =
1814 F(FPU) | F(VME) | F(DE) | F(PSE) |
1815 F(TSC) | F(MSR) | F(PAE) | F(MCE) |
1816 F(CX8) | F(APIC) | 0 /* Reserved */ | F(SEP) |
1817 F(MTRR) | F(PGE) | F(MCA) | F(CMOV) |
1818 F(PAT) | F(PSE36) | 0 /* PSN */ | F(CLFLSH) |
1819 0 /* Reserved, DS, ACPI */ | F(MMX) |
1820 F(FXSR) | F(XMM) | F(XMM2) | F(SELFSNOOP) |
1821 0 /* HTT, TM, Reserved, PBE */;
1822 /* cpuid 0x80000001.edx */
1823 const u32 kvm_supported_word1_x86_features =
1824 F(FPU) | F(VME) | F(DE) | F(PSE) |
1825 F(TSC) | F(MSR) | F(PAE) | F(MCE) |
1826 F(CX8) | F(APIC) | 0 /* Reserved */ | F(SYSCALL) |
1827 F(MTRR) | F(PGE) | F(MCA) | F(CMOV) |
1828 F(PAT) | F(PSE36) | 0 /* Reserved */ |
1829 f_nx | 0 /* Reserved */ | F(MMXEXT) | F(MMX) |
1830 F(FXSR) | F(FXSR_OPT) | f_gbpages | f_rdtscp |
1831 0 /* Reserved */ | f_lm | F(3DNOWEXT) | F(3DNOW);
1833 const u32 kvm_supported_word4_x86_features =
1834 F(XMM3) | 0 /* Reserved, DTES64, MONITOR */ |
1835 0 /* DS-CPL, VMX, SMX, EST */ |
1836 0 /* TM2 */ | F(SSSE3) | 0 /* CNXT-ID */ | 0 /* Reserved */ |
1837 0 /* Reserved */ | F(CX16) | 0 /* xTPR Update, PDCM */ |
1838 0 /* Reserved, DCA */ | F(XMM4_1) |
1839 F(XMM4_2) | F(X2APIC) | F(MOVBE) | F(POPCNT) |
1840 0 /* Reserved, XSAVE, OSXSAVE */;
1841 /* cpuid 0x80000001.ecx */
1842 const u32 kvm_supported_word6_x86_features =
1843 F(LAHF_LM) | F(CMP_LEGACY) | F(SVM) | 0 /* ExtApicSpace */ |
1844 F(CR8_LEGACY) | F(ABM) | F(SSE4A) | F(MISALIGNSSE) |
1845 F(3DNOWPREFETCH) | 0 /* OSVW */ | 0 /* IBS */ | F(SSE5) |
1846 0 /* SKINIT */ | 0 /* WDT */;
1848 /* all calls to cpuid_count() should be made on the same cpu */
1850 do_cpuid_1_ent(entry, function, index);
1855 entry->eax = min(entry->eax, (u32)0xb);
1858 entry->edx &= kvm_supported_word0_x86_features;
1859 entry->ecx &= kvm_supported_word4_x86_features;
1860 /* we support x2apic emulation even if host does not support
1861 * it since we emulate x2apic in software */
1862 entry->ecx |= F(X2APIC);
1864 /* function 2 entries are STATEFUL. That is, repeated cpuid commands
1865 * may return different values. This forces us to get_cpu() before
1866 * issuing the first command, and also to emulate this annoying behavior
1867 * in kvm_emulate_cpuid() using KVM_CPUID_FLAG_STATE_READ_NEXT */
1869 int t, times = entry->eax & 0xff;
1871 entry->flags |= KVM_CPUID_FLAG_STATEFUL_FUNC;
1872 entry->flags |= KVM_CPUID_FLAG_STATE_READ_NEXT;
1873 for (t = 1; t < times && *nent < maxnent; ++t) {
1874 do_cpuid_1_ent(&entry[t], function, 0);
1875 entry[t].flags |= KVM_CPUID_FLAG_STATEFUL_FUNC;
1880 /* function 4 and 0xb have additional index. */
1884 entry->flags |= KVM_CPUID_FLAG_SIGNIFCANT_INDEX;
1885 /* read more entries until cache_type is zero */
1886 for (i = 1; *nent < maxnent; ++i) {
1887 cache_type = entry[i - 1].eax & 0x1f;
1890 do_cpuid_1_ent(&entry[i], function, i);
1892 KVM_CPUID_FLAG_SIGNIFCANT_INDEX;
1900 entry->flags |= KVM_CPUID_FLAG_SIGNIFCANT_INDEX;
1901 /* read more entries until level_type is zero */
1902 for (i = 1; *nent < maxnent; ++i) {
1903 level_type = entry[i - 1].ecx & 0xff00;
1906 do_cpuid_1_ent(&entry[i], function, i);
1908 KVM_CPUID_FLAG_SIGNIFCANT_INDEX;
1914 entry->eax = min(entry->eax, 0x8000001a);
1917 entry->edx &= kvm_supported_word1_x86_features;
1918 entry->ecx &= kvm_supported_word6_x86_features;
1926 static int kvm_dev_ioctl_get_supported_cpuid(struct kvm_cpuid2 *cpuid,
1927 struct kvm_cpuid_entry2 __user *entries)
1929 struct kvm_cpuid_entry2 *cpuid_entries;
1930 int limit, nent = 0, r = -E2BIG;
1933 if (cpuid->nent < 1)
1935 if (cpuid->nent > KVM_MAX_CPUID_ENTRIES)
1936 cpuid->nent = KVM_MAX_CPUID_ENTRIES;
1938 cpuid_entries = vmalloc(sizeof(struct kvm_cpuid_entry2) * cpuid->nent);
1942 do_cpuid_ent(&cpuid_entries[0], 0, 0, &nent, cpuid->nent);
1943 limit = cpuid_entries[0].eax;
1944 for (func = 1; func <= limit && nent < cpuid->nent; ++func)
1945 do_cpuid_ent(&cpuid_entries[nent], func, 0,
1946 &nent, cpuid->nent);
1948 if (nent >= cpuid->nent)
1951 do_cpuid_ent(&cpuid_entries[nent], 0x80000000, 0, &nent, cpuid->nent);
1952 limit = cpuid_entries[nent - 1].eax;
1953 for (func = 0x80000001; func <= limit && nent < cpuid->nent; ++func)
1954 do_cpuid_ent(&cpuid_entries[nent], func, 0,
1955 &nent, cpuid->nent);
1957 if (nent >= cpuid->nent)
1961 if (copy_to_user(entries, cpuid_entries,
1962 nent * sizeof(struct kvm_cpuid_entry2)))
1968 vfree(cpuid_entries);
1973 static int kvm_vcpu_ioctl_get_lapic(struct kvm_vcpu *vcpu,
1974 struct kvm_lapic_state *s)
1977 memcpy(s->regs, vcpu->arch.apic->regs, sizeof *s);
1983 static int kvm_vcpu_ioctl_set_lapic(struct kvm_vcpu *vcpu,
1984 struct kvm_lapic_state *s)
1987 memcpy(vcpu->arch.apic->regs, s->regs, sizeof *s);
1988 kvm_apic_post_state_restore(vcpu);
1989 update_cr8_intercept(vcpu);
1995 static int kvm_vcpu_ioctl_interrupt(struct kvm_vcpu *vcpu,
1996 struct kvm_interrupt *irq)
1998 if (irq->irq < 0 || irq->irq >= 256)
2000 if (irqchip_in_kernel(vcpu->kvm))
2004 kvm_queue_interrupt(vcpu, irq->irq, false);
2011 static int kvm_vcpu_ioctl_nmi(struct kvm_vcpu *vcpu)
2014 kvm_inject_nmi(vcpu);
2020 static int vcpu_ioctl_tpr_access_reporting(struct kvm_vcpu *vcpu,
2021 struct kvm_tpr_access_ctl *tac)
2025 vcpu->arch.tpr_access_reporting = !!tac->enabled;
2029 static int kvm_vcpu_ioctl_x86_setup_mce(struct kvm_vcpu *vcpu,
2033 unsigned bank_num = mcg_cap & 0xff, bank;
2036 if (!bank_num || bank_num >= KVM_MAX_MCE_BANKS)
2038 if (mcg_cap & ~(KVM_MCE_CAP_SUPPORTED | 0xff | 0xff0000))
2041 vcpu->arch.mcg_cap = mcg_cap;
2042 /* Init IA32_MCG_CTL to all 1s */
2043 if (mcg_cap & MCG_CTL_P)
2044 vcpu->arch.mcg_ctl = ~(u64)0;
2045 /* Init IA32_MCi_CTL to all 1s */
2046 for (bank = 0; bank < bank_num; bank++)
2047 vcpu->arch.mce_banks[bank*4] = ~(u64)0;
2052 static int kvm_vcpu_ioctl_x86_set_mce(struct kvm_vcpu *vcpu,
2053 struct kvm_x86_mce *mce)
2055 u64 mcg_cap = vcpu->arch.mcg_cap;
2056 unsigned bank_num = mcg_cap & 0xff;
2057 u64 *banks = vcpu->arch.mce_banks;
2059 if (mce->bank >= bank_num || !(mce->status & MCI_STATUS_VAL))
2062 * if IA32_MCG_CTL is not all 1s, the uncorrected error
2063 * reporting is disabled
2065 if ((mce->status & MCI_STATUS_UC) && (mcg_cap & MCG_CTL_P) &&
2066 vcpu->arch.mcg_ctl != ~(u64)0)
2068 banks += 4 * mce->bank;
2070 * if IA32_MCi_CTL is not all 1s, the uncorrected error
2071 * reporting is disabled for the bank
2073 if ((mce->status & MCI_STATUS_UC) && banks[0] != ~(u64)0)
2075 if (mce->status & MCI_STATUS_UC) {
2076 if ((vcpu->arch.mcg_status & MCG_STATUS_MCIP) ||
2077 !kvm_read_cr4_bits(vcpu, X86_CR4_MCE)) {
2078 printk(KERN_DEBUG "kvm: set_mce: "
2079 "injects mce exception while "
2080 "previous one is in progress!\n");
2081 set_bit(KVM_REQ_TRIPLE_FAULT, &vcpu->requests);
2084 if (banks[1] & MCI_STATUS_VAL)
2085 mce->status |= MCI_STATUS_OVER;
2086 banks[2] = mce->addr;
2087 banks[3] = mce->misc;
2088 vcpu->arch.mcg_status = mce->mcg_status;
2089 banks[1] = mce->status;
2090 kvm_queue_exception(vcpu, MC_VECTOR);
2091 } else if (!(banks[1] & MCI_STATUS_VAL)
2092 || !(banks[1] & MCI_STATUS_UC)) {
2093 if (banks[1] & MCI_STATUS_VAL)
2094 mce->status |= MCI_STATUS_OVER;
2095 banks[2] = mce->addr;
2096 banks[3] = mce->misc;
2097 banks[1] = mce->status;
2099 banks[1] |= MCI_STATUS_OVER;
2103 static void kvm_vcpu_ioctl_x86_get_vcpu_events(struct kvm_vcpu *vcpu,
2104 struct kvm_vcpu_events *events)
2108 events->exception.injected = vcpu->arch.exception.pending;
2109 events->exception.nr = vcpu->arch.exception.nr;
2110 events->exception.has_error_code = vcpu->arch.exception.has_error_code;
2111 events->exception.error_code = vcpu->arch.exception.error_code;
2113 events->interrupt.injected = vcpu->arch.interrupt.pending;
2114 events->interrupt.nr = vcpu->arch.interrupt.nr;
2115 events->interrupt.soft = vcpu->arch.interrupt.soft;
2117 events->nmi.injected = vcpu->arch.nmi_injected;
2118 events->nmi.pending = vcpu->arch.nmi_pending;
2119 events->nmi.masked = kvm_x86_ops->get_nmi_mask(vcpu);
2121 events->sipi_vector = vcpu->arch.sipi_vector;
2123 events->flags = (KVM_VCPUEVENT_VALID_NMI_PENDING
2124 | KVM_VCPUEVENT_VALID_SIPI_VECTOR);
2129 static int kvm_vcpu_ioctl_x86_set_vcpu_events(struct kvm_vcpu *vcpu,
2130 struct kvm_vcpu_events *events)
2132 if (events->flags & ~(KVM_VCPUEVENT_VALID_NMI_PENDING
2133 | KVM_VCPUEVENT_VALID_SIPI_VECTOR))
2138 vcpu->arch.exception.pending = events->exception.injected;
2139 vcpu->arch.exception.nr = events->exception.nr;
2140 vcpu->arch.exception.has_error_code = events->exception.has_error_code;
2141 vcpu->arch.exception.error_code = events->exception.error_code;
2143 vcpu->arch.interrupt.pending = events->interrupt.injected;
2144 vcpu->arch.interrupt.nr = events->interrupt.nr;
2145 vcpu->arch.interrupt.soft = events->interrupt.soft;
2146 if (vcpu->arch.interrupt.pending && irqchip_in_kernel(vcpu->kvm))
2147 kvm_pic_clear_isr_ack(vcpu->kvm);
2149 vcpu->arch.nmi_injected = events->nmi.injected;
2150 if (events->flags & KVM_VCPUEVENT_VALID_NMI_PENDING)
2151 vcpu->arch.nmi_pending = events->nmi.pending;
2152 kvm_x86_ops->set_nmi_mask(vcpu, events->nmi.masked);
2154 if (events->flags & KVM_VCPUEVENT_VALID_SIPI_VECTOR)
2155 vcpu->arch.sipi_vector = events->sipi_vector;
2162 long kvm_arch_vcpu_ioctl(struct file *filp,
2163 unsigned int ioctl, unsigned long arg)
2165 struct kvm_vcpu *vcpu = filp->private_data;
2166 void __user *argp = (void __user *)arg;
2168 struct kvm_lapic_state *lapic = NULL;
2171 case KVM_GET_LAPIC: {
2173 if (!vcpu->arch.apic)
2175 lapic = kzalloc(sizeof(struct kvm_lapic_state), GFP_KERNEL);
2180 r = kvm_vcpu_ioctl_get_lapic(vcpu, lapic);
2184 if (copy_to_user(argp, lapic, sizeof(struct kvm_lapic_state)))
2189 case KVM_SET_LAPIC: {
2191 if (!vcpu->arch.apic)
2193 lapic = kmalloc(sizeof(struct kvm_lapic_state), GFP_KERNEL);
2198 if (copy_from_user(lapic, argp, sizeof(struct kvm_lapic_state)))
2200 r = kvm_vcpu_ioctl_set_lapic(vcpu, lapic);
2206 case KVM_INTERRUPT: {
2207 struct kvm_interrupt irq;
2210 if (copy_from_user(&irq, argp, sizeof irq))
2212 r = kvm_vcpu_ioctl_interrupt(vcpu, &irq);
2219 r = kvm_vcpu_ioctl_nmi(vcpu);
2225 case KVM_SET_CPUID: {
2226 struct kvm_cpuid __user *cpuid_arg = argp;
2227 struct kvm_cpuid cpuid;
2230 if (copy_from_user(&cpuid, cpuid_arg, sizeof cpuid))
2232 r = kvm_vcpu_ioctl_set_cpuid(vcpu, &cpuid, cpuid_arg->entries);
2237 case KVM_SET_CPUID2: {
2238 struct kvm_cpuid2 __user *cpuid_arg = argp;
2239 struct kvm_cpuid2 cpuid;
2242 if (copy_from_user(&cpuid, cpuid_arg, sizeof cpuid))
2244 r = kvm_vcpu_ioctl_set_cpuid2(vcpu, &cpuid,
2245 cpuid_arg->entries);
2250 case KVM_GET_CPUID2: {
2251 struct kvm_cpuid2 __user *cpuid_arg = argp;
2252 struct kvm_cpuid2 cpuid;
2255 if (copy_from_user(&cpuid, cpuid_arg, sizeof cpuid))
2257 r = kvm_vcpu_ioctl_get_cpuid2(vcpu, &cpuid,
2258 cpuid_arg->entries);
2262 if (copy_to_user(cpuid_arg, &cpuid, sizeof cpuid))
2268 r = msr_io(vcpu, argp, kvm_get_msr, 1);
2271 r = msr_io(vcpu, argp, do_set_msr, 0);
2273 case KVM_TPR_ACCESS_REPORTING: {
2274 struct kvm_tpr_access_ctl tac;
2277 if (copy_from_user(&tac, argp, sizeof tac))
2279 r = vcpu_ioctl_tpr_access_reporting(vcpu, &tac);
2283 if (copy_to_user(argp, &tac, sizeof tac))
2288 case KVM_SET_VAPIC_ADDR: {
2289 struct kvm_vapic_addr va;
2292 if (!irqchip_in_kernel(vcpu->kvm))
2295 if (copy_from_user(&va, argp, sizeof va))
2298 kvm_lapic_set_vapic_addr(vcpu, va.vapic_addr);
2301 case KVM_X86_SETUP_MCE: {
2305 if (copy_from_user(&mcg_cap, argp, sizeof mcg_cap))
2307 r = kvm_vcpu_ioctl_x86_setup_mce(vcpu, mcg_cap);
2310 case KVM_X86_SET_MCE: {
2311 struct kvm_x86_mce mce;
2314 if (copy_from_user(&mce, argp, sizeof mce))
2316 r = kvm_vcpu_ioctl_x86_set_mce(vcpu, &mce);
2319 case KVM_GET_VCPU_EVENTS: {
2320 struct kvm_vcpu_events events;
2322 kvm_vcpu_ioctl_x86_get_vcpu_events(vcpu, &events);
2325 if (copy_to_user(argp, &events, sizeof(struct kvm_vcpu_events)))
2330 case KVM_SET_VCPU_EVENTS: {
2331 struct kvm_vcpu_events events;
2334 if (copy_from_user(&events, argp, sizeof(struct kvm_vcpu_events)))
2337 r = kvm_vcpu_ioctl_x86_set_vcpu_events(vcpu, &events);
2348 static int kvm_vm_ioctl_set_tss_addr(struct kvm *kvm, unsigned long addr)
2352 if (addr > (unsigned int)(-3 * PAGE_SIZE))
2354 ret = kvm_x86_ops->set_tss_addr(kvm, addr);
2358 static int kvm_vm_ioctl_set_identity_map_addr(struct kvm *kvm,
2361 kvm->arch.ept_identity_map_addr = ident_addr;
2365 static int kvm_vm_ioctl_set_nr_mmu_pages(struct kvm *kvm,
2366 u32 kvm_nr_mmu_pages)
2368 if (kvm_nr_mmu_pages < KVM_MIN_ALLOC_MMU_PAGES)
2371 mutex_lock(&kvm->slots_lock);
2372 spin_lock(&kvm->mmu_lock);
2374 kvm_mmu_change_mmu_pages(kvm, kvm_nr_mmu_pages);
2375 kvm->arch.n_requested_mmu_pages = kvm_nr_mmu_pages;
2377 spin_unlock(&kvm->mmu_lock);
2378 mutex_unlock(&kvm->slots_lock);
2382 static int kvm_vm_ioctl_get_nr_mmu_pages(struct kvm *kvm)
2384 return kvm->arch.n_alloc_mmu_pages;
2387 gfn_t unalias_gfn_instantiation(struct kvm *kvm, gfn_t gfn)
2390 struct kvm_mem_alias *alias;
2391 struct kvm_mem_aliases *aliases;
2393 aliases = rcu_dereference(kvm->arch.aliases);
2395 for (i = 0; i < aliases->naliases; ++i) {
2396 alias = &aliases->aliases[i];
2397 if (alias->flags & KVM_ALIAS_INVALID)
2399 if (gfn >= alias->base_gfn
2400 && gfn < alias->base_gfn + alias->npages)
2401 return alias->target_gfn + gfn - alias->base_gfn;
2406 gfn_t unalias_gfn(struct kvm *kvm, gfn_t gfn)
2409 struct kvm_mem_alias *alias;
2410 struct kvm_mem_aliases *aliases;
2412 aliases = rcu_dereference(kvm->arch.aliases);
2414 for (i = 0; i < aliases->naliases; ++i) {
2415 alias = &aliases->aliases[i];
2416 if (gfn >= alias->base_gfn
2417 && gfn < alias->base_gfn + alias->npages)
2418 return alias->target_gfn + gfn - alias->base_gfn;
2424 * Set a new alias region. Aliases map a portion of physical memory into
2425 * another portion. This is useful for memory windows, for example the PC
2428 static int kvm_vm_ioctl_set_memory_alias(struct kvm *kvm,
2429 struct kvm_memory_alias *alias)
2432 struct kvm_mem_alias *p;
2433 struct kvm_mem_aliases *aliases, *old_aliases;
2436 /* General sanity checks */
2437 if (alias->memory_size & (PAGE_SIZE - 1))
2439 if (alias->guest_phys_addr & (PAGE_SIZE - 1))
2441 if (alias->slot >= KVM_ALIAS_SLOTS)
2443 if (alias->guest_phys_addr + alias->memory_size
2444 < alias->guest_phys_addr)
2446 if (alias->target_phys_addr + alias->memory_size
2447 < alias->target_phys_addr)
2451 aliases = kzalloc(sizeof(struct kvm_mem_aliases), GFP_KERNEL);
2455 mutex_lock(&kvm->slots_lock);
2457 /* invalidate any gfn reference in case of deletion/shrinking */
2458 memcpy(aliases, kvm->arch.aliases, sizeof(struct kvm_mem_aliases));
2459 aliases->aliases[alias->slot].flags |= KVM_ALIAS_INVALID;
2460 old_aliases = kvm->arch.aliases;
2461 rcu_assign_pointer(kvm->arch.aliases, aliases);
2462 synchronize_srcu_expedited(&kvm->srcu);
2463 kvm_mmu_zap_all(kvm);
2467 aliases = kzalloc(sizeof(struct kvm_mem_aliases), GFP_KERNEL);
2471 memcpy(aliases, kvm->arch.aliases, sizeof(struct kvm_mem_aliases));
2473 p = &aliases->aliases[alias->slot];
2474 p->base_gfn = alias->guest_phys_addr >> PAGE_SHIFT;
2475 p->npages = alias->memory_size >> PAGE_SHIFT;
2476 p->target_gfn = alias->target_phys_addr >> PAGE_SHIFT;
2477 p->flags &= ~(KVM_ALIAS_INVALID);
2479 for (n = KVM_ALIAS_SLOTS; n > 0; --n)
2480 if (aliases->aliases[n - 1].npages)
2482 aliases->naliases = n;
2484 old_aliases = kvm->arch.aliases;
2485 rcu_assign_pointer(kvm->arch.aliases, aliases);
2486 synchronize_srcu_expedited(&kvm->srcu);
2491 mutex_unlock(&kvm->slots_lock);
2496 static int kvm_vm_ioctl_get_irqchip(struct kvm *kvm, struct kvm_irqchip *chip)
2501 switch (chip->chip_id) {
2502 case KVM_IRQCHIP_PIC_MASTER:
2503 memcpy(&chip->chip.pic,
2504 &pic_irqchip(kvm)->pics[0],
2505 sizeof(struct kvm_pic_state));
2507 case KVM_IRQCHIP_PIC_SLAVE:
2508 memcpy(&chip->chip.pic,
2509 &pic_irqchip(kvm)->pics[1],
2510 sizeof(struct kvm_pic_state));
2512 case KVM_IRQCHIP_IOAPIC:
2513 r = kvm_get_ioapic(kvm, &chip->chip.ioapic);
2522 static int kvm_vm_ioctl_set_irqchip(struct kvm *kvm, struct kvm_irqchip *chip)
2527 switch (chip->chip_id) {
2528 case KVM_IRQCHIP_PIC_MASTER:
2529 raw_spin_lock(&pic_irqchip(kvm)->lock);
2530 memcpy(&pic_irqchip(kvm)->pics[0],
2532 sizeof(struct kvm_pic_state));
2533 raw_spin_unlock(&pic_irqchip(kvm)->lock);
2535 case KVM_IRQCHIP_PIC_SLAVE:
2536 raw_spin_lock(&pic_irqchip(kvm)->lock);
2537 memcpy(&pic_irqchip(kvm)->pics[1],
2539 sizeof(struct kvm_pic_state));
2540 raw_spin_unlock(&pic_irqchip(kvm)->lock);
2542 case KVM_IRQCHIP_IOAPIC:
2543 r = kvm_set_ioapic(kvm, &chip->chip.ioapic);
2549 kvm_pic_update_irq(pic_irqchip(kvm));
2553 static int kvm_vm_ioctl_get_pit(struct kvm *kvm, struct kvm_pit_state *ps)
2557 mutex_lock(&kvm->arch.vpit->pit_state.lock);
2558 memcpy(ps, &kvm->arch.vpit->pit_state, sizeof(struct kvm_pit_state));
2559 mutex_unlock(&kvm->arch.vpit->pit_state.lock);
2563 static int kvm_vm_ioctl_set_pit(struct kvm *kvm, struct kvm_pit_state *ps)
2567 mutex_lock(&kvm->arch.vpit->pit_state.lock);
2568 memcpy(&kvm->arch.vpit->pit_state, ps, sizeof(struct kvm_pit_state));
2569 kvm_pit_load_count(kvm, 0, ps->channels[0].count, 0);
2570 mutex_unlock(&kvm->arch.vpit->pit_state.lock);
2574 static int kvm_vm_ioctl_get_pit2(struct kvm *kvm, struct kvm_pit_state2 *ps)
2578 mutex_lock(&kvm->arch.vpit->pit_state.lock);
2579 memcpy(ps->channels, &kvm->arch.vpit->pit_state.channels,
2580 sizeof(ps->channels));
2581 ps->flags = kvm->arch.vpit->pit_state.flags;
2582 mutex_unlock(&kvm->arch.vpit->pit_state.lock);
2586 static int kvm_vm_ioctl_set_pit2(struct kvm *kvm, struct kvm_pit_state2 *ps)
2588 int r = 0, start = 0;
2589 u32 prev_legacy, cur_legacy;
2590 mutex_lock(&kvm->arch.vpit->pit_state.lock);
2591 prev_legacy = kvm->arch.vpit->pit_state.flags & KVM_PIT_FLAGS_HPET_LEGACY;
2592 cur_legacy = ps->flags & KVM_PIT_FLAGS_HPET_LEGACY;
2593 if (!prev_legacy && cur_legacy)
2595 memcpy(&kvm->arch.vpit->pit_state.channels, &ps->channels,
2596 sizeof(kvm->arch.vpit->pit_state.channels));
2597 kvm->arch.vpit->pit_state.flags = ps->flags;
2598 kvm_pit_load_count(kvm, 0, kvm->arch.vpit->pit_state.channels[0].count, start);
2599 mutex_unlock(&kvm->arch.vpit->pit_state.lock);
2603 static int kvm_vm_ioctl_reinject(struct kvm *kvm,
2604 struct kvm_reinject_control *control)
2606 if (!kvm->arch.vpit)
2608 mutex_lock(&kvm->arch.vpit->pit_state.lock);
2609 kvm->arch.vpit->pit_state.pit_timer.reinject = control->pit_reinject;
2610 mutex_unlock(&kvm->arch.vpit->pit_state.lock);
2615 * Get (and clear) the dirty memory log for a memory slot.
2617 int kvm_vm_ioctl_get_dirty_log(struct kvm *kvm,
2618 struct kvm_dirty_log *log)
2621 struct kvm_memory_slot *memslot;
2623 unsigned long is_dirty = 0;
2624 unsigned long *dirty_bitmap = NULL;
2626 mutex_lock(&kvm->slots_lock);
2629 if (log->slot >= KVM_MEMORY_SLOTS)
2632 memslot = &kvm->memslots->memslots[log->slot];
2634 if (!memslot->dirty_bitmap)
2637 n = kvm_dirty_bitmap_bytes(memslot);
2640 dirty_bitmap = vmalloc(n);
2643 memset(dirty_bitmap, 0, n);
2645 for (i = 0; !is_dirty && i < n/sizeof(long); i++)
2646 is_dirty = memslot->dirty_bitmap[i];
2648 /* If nothing is dirty, don't bother messing with page tables. */
2650 struct kvm_memslots *slots, *old_slots;
2652 spin_lock(&kvm->mmu_lock);
2653 kvm_mmu_slot_remove_write_access(kvm, log->slot);
2654 spin_unlock(&kvm->mmu_lock);
2656 slots = kzalloc(sizeof(struct kvm_memslots), GFP_KERNEL);
2660 memcpy(slots, kvm->memslots, sizeof(struct kvm_memslots));
2661 slots->memslots[log->slot].dirty_bitmap = dirty_bitmap;
2663 old_slots = kvm->memslots;
2664 rcu_assign_pointer(kvm->memslots, slots);
2665 synchronize_srcu_expedited(&kvm->srcu);
2666 dirty_bitmap = old_slots->memslots[log->slot].dirty_bitmap;
2671 if (copy_to_user(log->dirty_bitmap, dirty_bitmap, n))
2674 vfree(dirty_bitmap);
2676 mutex_unlock(&kvm->slots_lock);
2680 long kvm_arch_vm_ioctl(struct file *filp,
2681 unsigned int ioctl, unsigned long arg)
2683 struct kvm *kvm = filp->private_data;
2684 void __user *argp = (void __user *)arg;
2687 * This union makes it completely explicit to gcc-3.x
2688 * that these two variables' stack usage should be
2689 * combined, not added together.
2692 struct kvm_pit_state ps;
2693 struct kvm_pit_state2 ps2;
2694 struct kvm_memory_alias alias;
2695 struct kvm_pit_config pit_config;
2699 case KVM_SET_TSS_ADDR:
2700 r = kvm_vm_ioctl_set_tss_addr(kvm, arg);
2704 case KVM_SET_IDENTITY_MAP_ADDR: {
2708 if (copy_from_user(&ident_addr, argp, sizeof ident_addr))
2710 r = kvm_vm_ioctl_set_identity_map_addr(kvm, ident_addr);
2715 case KVM_SET_MEMORY_REGION: {
2716 struct kvm_memory_region kvm_mem;
2717 struct kvm_userspace_memory_region kvm_userspace_mem;
2720 if (copy_from_user(&kvm_mem, argp, sizeof kvm_mem))
2722 kvm_userspace_mem.slot = kvm_mem.slot;
2723 kvm_userspace_mem.flags = kvm_mem.flags;
2724 kvm_userspace_mem.guest_phys_addr = kvm_mem.guest_phys_addr;
2725 kvm_userspace_mem.memory_size = kvm_mem.memory_size;
2726 r = kvm_vm_ioctl_set_memory_region(kvm, &kvm_userspace_mem, 0);
2731 case KVM_SET_NR_MMU_PAGES:
2732 r = kvm_vm_ioctl_set_nr_mmu_pages(kvm, arg);
2736 case KVM_GET_NR_MMU_PAGES:
2737 r = kvm_vm_ioctl_get_nr_mmu_pages(kvm);
2739 case KVM_SET_MEMORY_ALIAS:
2741 if (copy_from_user(&u.alias, argp, sizeof(struct kvm_memory_alias)))
2743 r = kvm_vm_ioctl_set_memory_alias(kvm, &u.alias);
2747 case KVM_CREATE_IRQCHIP: {
2748 struct kvm_pic *vpic;
2750 mutex_lock(&kvm->lock);
2753 goto create_irqchip_unlock;
2755 vpic = kvm_create_pic(kvm);
2757 r = kvm_ioapic_init(kvm);
2759 kvm_io_bus_unregister_dev(kvm, KVM_PIO_BUS,
2762 goto create_irqchip_unlock;
2765 goto create_irqchip_unlock;
2767 kvm->arch.vpic = vpic;
2769 r = kvm_setup_default_irq_routing(kvm);
2771 mutex_lock(&kvm->irq_lock);
2772 kvm_ioapic_destroy(kvm);
2773 kvm_destroy_pic(kvm);
2774 mutex_unlock(&kvm->irq_lock);
2776 create_irqchip_unlock:
2777 mutex_unlock(&kvm->lock);
2780 case KVM_CREATE_PIT:
2781 u.pit_config.flags = KVM_PIT_SPEAKER_DUMMY;
2783 case KVM_CREATE_PIT2:
2785 if (copy_from_user(&u.pit_config, argp,
2786 sizeof(struct kvm_pit_config)))
2789 mutex_lock(&kvm->slots_lock);
2792 goto create_pit_unlock;
2794 kvm->arch.vpit = kvm_create_pit(kvm, u.pit_config.flags);
2798 mutex_unlock(&kvm->slots_lock);
2800 case KVM_IRQ_LINE_STATUS:
2801 case KVM_IRQ_LINE: {
2802 struct kvm_irq_level irq_event;
2805 if (copy_from_user(&irq_event, argp, sizeof irq_event))
2807 if (irqchip_in_kernel(kvm)) {
2809 status = kvm_set_irq(kvm, KVM_USERSPACE_IRQ_SOURCE_ID,
2810 irq_event.irq, irq_event.level);
2811 if (ioctl == KVM_IRQ_LINE_STATUS) {
2812 irq_event.status = status;
2813 if (copy_to_user(argp, &irq_event,
2821 case KVM_GET_IRQCHIP: {
2822 /* 0: PIC master, 1: PIC slave, 2: IOAPIC */
2823 struct kvm_irqchip *chip = kmalloc(sizeof(*chip), GFP_KERNEL);
2829 if (copy_from_user(chip, argp, sizeof *chip))
2830 goto get_irqchip_out;
2832 if (!irqchip_in_kernel(kvm))
2833 goto get_irqchip_out;
2834 r = kvm_vm_ioctl_get_irqchip(kvm, chip);
2836 goto get_irqchip_out;
2838 if (copy_to_user(argp, chip, sizeof *chip))
2839 goto get_irqchip_out;
2847 case KVM_SET_IRQCHIP: {
2848 /* 0: PIC master, 1: PIC slave, 2: IOAPIC */
2849 struct kvm_irqchip *chip = kmalloc(sizeof(*chip), GFP_KERNEL);
2855 if (copy_from_user(chip, argp, sizeof *chip))
2856 goto set_irqchip_out;
2858 if (!irqchip_in_kernel(kvm))
2859 goto set_irqchip_out;
2860 r = kvm_vm_ioctl_set_irqchip(kvm, chip);
2862 goto set_irqchip_out;
2872 if (copy_from_user(&u.ps, argp, sizeof(struct kvm_pit_state)))
2875 if (!kvm->arch.vpit)
2877 r = kvm_vm_ioctl_get_pit(kvm, &u.ps);
2881 if (copy_to_user(argp, &u.ps, sizeof(struct kvm_pit_state)))
2888 if (copy_from_user(&u.ps, argp, sizeof u.ps))
2891 if (!kvm->arch.vpit)
2893 r = kvm_vm_ioctl_set_pit(kvm, &u.ps);
2899 case KVM_GET_PIT2: {
2901 if (!kvm->arch.vpit)
2903 r = kvm_vm_ioctl_get_pit2(kvm, &u.ps2);
2907 if (copy_to_user(argp, &u.ps2, sizeof(u.ps2)))
2912 case KVM_SET_PIT2: {
2914 if (copy_from_user(&u.ps2, argp, sizeof(u.ps2)))
2917 if (!kvm->arch.vpit)
2919 r = kvm_vm_ioctl_set_pit2(kvm, &u.ps2);
2925 case KVM_REINJECT_CONTROL: {
2926 struct kvm_reinject_control control;
2928 if (copy_from_user(&control, argp, sizeof(control)))
2930 r = kvm_vm_ioctl_reinject(kvm, &control);
2936 case KVM_XEN_HVM_CONFIG: {
2938 if (copy_from_user(&kvm->arch.xen_hvm_config, argp,
2939 sizeof(struct kvm_xen_hvm_config)))
2942 if (kvm->arch.xen_hvm_config.flags)
2947 case KVM_SET_CLOCK: {
2948 struct timespec now;
2949 struct kvm_clock_data user_ns;
2954 if (copy_from_user(&user_ns, argp, sizeof(user_ns)))
2963 now_ns = timespec_to_ns(&now);
2964 delta = user_ns.clock - now_ns;
2965 kvm->arch.kvmclock_offset = delta;
2968 case KVM_GET_CLOCK: {
2969 struct timespec now;
2970 struct kvm_clock_data user_ns;
2974 now_ns = timespec_to_ns(&now);
2975 user_ns.clock = kvm->arch.kvmclock_offset + now_ns;
2979 if (copy_to_user(argp, &user_ns, sizeof(user_ns)))
2992 static void kvm_init_msr_list(void)
2997 /* skip the first msrs in the list. KVM-specific */
2998 for (i = j = KVM_SAVE_MSRS_BEGIN; i < ARRAY_SIZE(msrs_to_save); i++) {
2999 if (rdmsr_safe(msrs_to_save[i], &dummy[0], &dummy[1]) < 0)
3002 msrs_to_save[j] = msrs_to_save[i];
3005 num_msrs_to_save = j;
3008 static int vcpu_mmio_write(struct kvm_vcpu *vcpu, gpa_t addr, int len,
3011 if (vcpu->arch.apic &&
3012 !kvm_iodevice_write(&vcpu->arch.apic->dev, addr, len, v))
3015 return kvm_io_bus_write(vcpu->kvm, KVM_MMIO_BUS, addr, len, v);
3018 static int vcpu_mmio_read(struct kvm_vcpu *vcpu, gpa_t addr, int len, void *v)
3020 if (vcpu->arch.apic &&
3021 !kvm_iodevice_read(&vcpu->arch.apic->dev, addr, len, v))
3024 return kvm_io_bus_read(vcpu->kvm, KVM_MMIO_BUS, addr, len, v);
3027 gpa_t kvm_mmu_gva_to_gpa_read(struct kvm_vcpu *vcpu, gva_t gva, u32 *error)
3029 u32 access = (kvm_x86_ops->get_cpl(vcpu) == 3) ? PFERR_USER_MASK : 0;
3030 return vcpu->arch.mmu.gva_to_gpa(vcpu, gva, access, error);
3033 gpa_t kvm_mmu_gva_to_gpa_fetch(struct kvm_vcpu *vcpu, gva_t gva, u32 *error)
3035 u32 access = (kvm_x86_ops->get_cpl(vcpu) == 3) ? PFERR_USER_MASK : 0;
3036 access |= PFERR_FETCH_MASK;
3037 return vcpu->arch.mmu.gva_to_gpa(vcpu, gva, access, error);
3040 gpa_t kvm_mmu_gva_to_gpa_write(struct kvm_vcpu *vcpu, gva_t gva, u32 *error)
3042 u32 access = (kvm_x86_ops->get_cpl(vcpu) == 3) ? PFERR_USER_MASK : 0;
3043 access |= PFERR_WRITE_MASK;
3044 return vcpu->arch.mmu.gva_to_gpa(vcpu, gva, access, error);
3047 /* uses this to access any guest's mapped memory without checking CPL */
3048 gpa_t kvm_mmu_gva_to_gpa_system(struct kvm_vcpu *vcpu, gva_t gva, u32 *error)
3050 return vcpu->arch.mmu.gva_to_gpa(vcpu, gva, 0, error);
3053 static int kvm_read_guest_virt_helper(gva_t addr, void *val, unsigned int bytes,
3054 struct kvm_vcpu *vcpu, u32 access,
3058 int r = X86EMUL_CONTINUE;
3061 gpa_t gpa = vcpu->arch.mmu.gva_to_gpa(vcpu, addr, access, error);
3062 unsigned offset = addr & (PAGE_SIZE-1);
3063 unsigned toread = min(bytes, (unsigned)PAGE_SIZE - offset);
3066 if (gpa == UNMAPPED_GVA) {
3067 r = X86EMUL_PROPAGATE_FAULT;
3070 ret = kvm_read_guest(vcpu->kvm, gpa, data, toread);
3072 r = X86EMUL_UNHANDLEABLE;
3084 /* used for instruction fetching */
3085 static int kvm_fetch_guest_virt(gva_t addr, void *val, unsigned int bytes,
3086 struct kvm_vcpu *vcpu, u32 *error)
3088 u32 access = (kvm_x86_ops->get_cpl(vcpu) == 3) ? PFERR_USER_MASK : 0;
3089 return kvm_read_guest_virt_helper(addr, val, bytes, vcpu,
3090 access | PFERR_FETCH_MASK, error);
3093 static int kvm_read_guest_virt(gva_t addr, void *val, unsigned int bytes,
3094 struct kvm_vcpu *vcpu, u32 *error)
3096 u32 access = (kvm_x86_ops->get_cpl(vcpu) == 3) ? PFERR_USER_MASK : 0;
3097 return kvm_read_guest_virt_helper(addr, val, bytes, vcpu, access,
3101 static int kvm_read_guest_virt_system(gva_t addr, void *val, unsigned int bytes,
3102 struct kvm_vcpu *vcpu, u32 *error)
3104 return kvm_read_guest_virt_helper(addr, val, bytes, vcpu, 0, error);
3107 static int kvm_write_guest_virt(gva_t addr, void *val, unsigned int bytes,
3108 struct kvm_vcpu *vcpu, u32 *error)
3111 int r = X86EMUL_CONTINUE;
3114 gpa_t gpa = kvm_mmu_gva_to_gpa_write(vcpu, addr, error);
3115 unsigned offset = addr & (PAGE_SIZE-1);
3116 unsigned towrite = min(bytes, (unsigned)PAGE_SIZE - offset);
3119 if (gpa == UNMAPPED_GVA) {
3120 r = X86EMUL_PROPAGATE_FAULT;
3123 ret = kvm_write_guest(vcpu->kvm, gpa, data, towrite);
3125 r = X86EMUL_UNHANDLEABLE;
3138 static int emulator_read_emulated(unsigned long addr,
3141 struct kvm_vcpu *vcpu)
3146 if (vcpu->mmio_read_completed) {
3147 memcpy(val, vcpu->mmio_data, bytes);
3148 trace_kvm_mmio(KVM_TRACE_MMIO_READ, bytes,
3149 vcpu->mmio_phys_addr, *(u64 *)val);
3150 vcpu->mmio_read_completed = 0;
3151 return X86EMUL_CONTINUE;
3154 gpa = kvm_mmu_gva_to_gpa_read(vcpu, addr, &error_code);
3156 if (gpa == UNMAPPED_GVA) {
3157 kvm_inject_page_fault(vcpu, addr, error_code);
3158 return X86EMUL_PROPAGATE_FAULT;
3161 /* For APIC access vmexit */
3162 if ((gpa & PAGE_MASK) == APIC_DEFAULT_PHYS_BASE)
3165 if (kvm_read_guest_virt(addr, val, bytes, vcpu, NULL)
3166 == X86EMUL_CONTINUE)
3167 return X86EMUL_CONTINUE;
3171 * Is this MMIO handled locally?
3173 if (!vcpu_mmio_read(vcpu, gpa, bytes, val)) {
3174 trace_kvm_mmio(KVM_TRACE_MMIO_READ, bytes, gpa, *(u64 *)val);
3175 return X86EMUL_CONTINUE;
3178 trace_kvm_mmio(KVM_TRACE_MMIO_READ_UNSATISFIED, bytes, gpa, 0);
3180 vcpu->mmio_needed = 1;
3181 vcpu->mmio_phys_addr = gpa;
3182 vcpu->mmio_size = bytes;
3183 vcpu->mmio_is_write = 0;
3185 return X86EMUL_UNHANDLEABLE;
3188 int emulator_write_phys(struct kvm_vcpu *vcpu, gpa_t gpa,
3189 const void *val, int bytes)
3193 ret = kvm_write_guest(vcpu->kvm, gpa, val, bytes);
3196 kvm_mmu_pte_write(vcpu, gpa, val, bytes, 1);
3200 static int emulator_write_emulated_onepage(unsigned long addr,
3203 struct kvm_vcpu *vcpu)
3208 gpa = kvm_mmu_gva_to_gpa_write(vcpu, addr, &error_code);
3210 if (gpa == UNMAPPED_GVA) {
3211 kvm_inject_page_fault(vcpu, addr, error_code);
3212 return X86EMUL_PROPAGATE_FAULT;
3215 /* For APIC access vmexit */
3216 if ((gpa & PAGE_MASK) == APIC_DEFAULT_PHYS_BASE)
3219 if (emulator_write_phys(vcpu, gpa, val, bytes))
3220 return X86EMUL_CONTINUE;
3223 trace_kvm_mmio(KVM_TRACE_MMIO_WRITE, bytes, gpa, *(u64 *)val);
3225 * Is this MMIO handled locally?
3227 if (!vcpu_mmio_write(vcpu, gpa, bytes, val))
3228 return X86EMUL_CONTINUE;
3230 vcpu->mmio_needed = 1;
3231 vcpu->mmio_phys_addr = gpa;
3232 vcpu->mmio_size = bytes;
3233 vcpu->mmio_is_write = 1;
3234 memcpy(vcpu->mmio_data, val, bytes);
3236 return X86EMUL_CONTINUE;
3239 int emulator_write_emulated(unsigned long addr,
3242 struct kvm_vcpu *vcpu)
3244 /* Crossing a page boundary? */
3245 if (((addr + bytes - 1) ^ addr) & PAGE_MASK) {
3248 now = -addr & ~PAGE_MASK;
3249 rc = emulator_write_emulated_onepage(addr, val, now, vcpu);
3250 if (rc != X86EMUL_CONTINUE)
3256 return emulator_write_emulated_onepage(addr, val, bytes, vcpu);
3258 EXPORT_SYMBOL_GPL(emulator_write_emulated);
3260 static int emulator_cmpxchg_emulated(unsigned long addr,
3264 struct kvm_vcpu *vcpu)
3266 printk_once(KERN_WARNING "kvm: emulating exchange as write\n");
3267 #ifndef CONFIG_X86_64
3268 /* guests cmpxchg8b have to be emulated atomically */
3275 gpa = kvm_mmu_gva_to_gpa_write(vcpu, addr, NULL);
3277 if (gpa == UNMAPPED_GVA ||
3278 (gpa & PAGE_MASK) == APIC_DEFAULT_PHYS_BASE)
3281 if (((gpa + bytes - 1) & PAGE_MASK) != (gpa & PAGE_MASK))
3286 page = gfn_to_page(vcpu->kvm, gpa >> PAGE_SHIFT);
3288 kaddr = kmap_atomic(page, KM_USER0);
3289 set_64bit((u64 *)(kaddr + offset_in_page(gpa)), val);
3290 kunmap_atomic(kaddr, KM_USER0);
3291 kvm_release_page_dirty(page);
3296 return emulator_write_emulated(addr, new, bytes, vcpu);
3299 static unsigned long get_segment_base(struct kvm_vcpu *vcpu, int seg)
3301 return kvm_x86_ops->get_segment_base(vcpu, seg);
3304 int emulate_invlpg(struct kvm_vcpu *vcpu, gva_t address)
3306 kvm_mmu_invlpg(vcpu, address);
3307 return X86EMUL_CONTINUE;
3310 int emulate_clts(struct kvm_vcpu *vcpu)
3312 kvm_x86_ops->set_cr0(vcpu, kvm_read_cr0_bits(vcpu, ~X86_CR0_TS));
3313 kvm_x86_ops->fpu_activate(vcpu);
3314 return X86EMUL_CONTINUE;
3317 int emulator_get_dr(struct x86_emulate_ctxt *ctxt, int dr, unsigned long *dest)
3319 return kvm_x86_ops->get_dr(ctxt->vcpu, dr, dest);
3322 int emulator_set_dr(struct x86_emulate_ctxt *ctxt, int dr, unsigned long value)
3324 unsigned long mask = (ctxt->mode == X86EMUL_MODE_PROT64) ? ~0ULL : ~0U;
3326 return kvm_x86_ops->set_dr(ctxt->vcpu, dr, value & mask);
3329 void kvm_report_emulation_failure(struct kvm_vcpu *vcpu, const char *context)
3332 unsigned long rip = kvm_rip_read(vcpu);
3333 unsigned long rip_linear;
3335 if (!printk_ratelimit())
3338 rip_linear = rip + get_segment_base(vcpu, VCPU_SREG_CS);
3340 kvm_read_guest_virt(rip_linear, (void *)opcodes, 4, vcpu, NULL);
3342 printk(KERN_ERR "emulation failed (%s) rip %lx %02x %02x %02x %02x\n",
3343 context, rip, opcodes[0], opcodes[1], opcodes[2], opcodes[3]);
3345 EXPORT_SYMBOL_GPL(kvm_report_emulation_failure);
3347 static struct x86_emulate_ops emulate_ops = {
3348 .read_std = kvm_read_guest_virt_system,
3349 .fetch = kvm_fetch_guest_virt,
3350 .read_emulated = emulator_read_emulated,
3351 .write_emulated = emulator_write_emulated,
3352 .cmpxchg_emulated = emulator_cmpxchg_emulated,
3355 static void cache_all_regs(struct kvm_vcpu *vcpu)
3357 kvm_register_read(vcpu, VCPU_REGS_RAX);
3358 kvm_register_read(vcpu, VCPU_REGS_RSP);
3359 kvm_register_read(vcpu, VCPU_REGS_RIP);
3360 vcpu->arch.regs_dirty = ~0;
3363 int emulate_instruction(struct kvm_vcpu *vcpu,
3369 struct decode_cache *c;
3370 struct kvm_run *run = vcpu->run;
3372 kvm_clear_exception_queue(vcpu);
3373 vcpu->arch.mmio_fault_cr2 = cr2;
3375 * TODO: fix emulate.c to use guest_read/write_register
3376 * instead of direct ->regs accesses, can save hundred cycles
3377 * on Intel for instructions that don't read/change RSP, for
3380 cache_all_regs(vcpu);
3382 vcpu->mmio_is_write = 0;
3383 vcpu->arch.pio.string = 0;
3385 if (!(emulation_type & EMULTYPE_NO_DECODE)) {
3387 kvm_x86_ops->get_cs_db_l_bits(vcpu, &cs_db, &cs_l);
3389 vcpu->arch.emulate_ctxt.vcpu = vcpu;
3390 vcpu->arch.emulate_ctxt.eflags = kvm_get_rflags(vcpu);
3391 vcpu->arch.emulate_ctxt.mode =
3392 (!is_protmode(vcpu)) ? X86EMUL_MODE_REAL :
3393 (vcpu->arch.emulate_ctxt.eflags & X86_EFLAGS_VM)
3394 ? X86EMUL_MODE_VM86 : cs_l
3395 ? X86EMUL_MODE_PROT64 : cs_db
3396 ? X86EMUL_MODE_PROT32 : X86EMUL_MODE_PROT16;
3398 r = x86_decode_insn(&vcpu->arch.emulate_ctxt, &emulate_ops);
3400 /* Only allow emulation of specific instructions on #UD
3401 * (namely VMMCALL, sysenter, sysexit, syscall)*/
3402 c = &vcpu->arch.emulate_ctxt.decode;
3403 if (emulation_type & EMULTYPE_TRAP_UD) {
3405 return EMULATE_FAIL;
3407 case 0x01: /* VMMCALL */
3408 if (c->modrm_mod != 3 || c->modrm_rm != 1)
3409 return EMULATE_FAIL;
3411 case 0x34: /* sysenter */
3412 case 0x35: /* sysexit */
3413 if (c->modrm_mod != 0 || c->modrm_rm != 0)
3414 return EMULATE_FAIL;
3416 case 0x05: /* syscall */
3417 if (c->modrm_mod != 0 || c->modrm_rm != 0)
3418 return EMULATE_FAIL;
3421 return EMULATE_FAIL;
3424 if (!(c->modrm_reg == 0 || c->modrm_reg == 3))
3425 return EMULATE_FAIL;
3428 ++vcpu->stat.insn_emulation;
3430 ++vcpu->stat.insn_emulation_fail;
3431 if (kvm_mmu_unprotect_page_virt(vcpu, cr2))
3432 return EMULATE_DONE;
3433 return EMULATE_FAIL;
3437 if (emulation_type & EMULTYPE_SKIP) {
3438 kvm_rip_write(vcpu, vcpu->arch.emulate_ctxt.decode.eip);
3439 return EMULATE_DONE;
3442 r = x86_emulate_insn(&vcpu->arch.emulate_ctxt, &emulate_ops);
3443 shadow_mask = vcpu->arch.emulate_ctxt.interruptibility;
3446 kvm_x86_ops->set_interrupt_shadow(vcpu, shadow_mask);
3448 if (vcpu->arch.pio.string)
3449 return EMULATE_DO_MMIO;
3451 if ((r || vcpu->mmio_is_write) && run) {
3452 run->exit_reason = KVM_EXIT_MMIO;
3453 run->mmio.phys_addr = vcpu->mmio_phys_addr;
3454 memcpy(run->mmio.data, vcpu->mmio_data, 8);
3455 run->mmio.len = vcpu->mmio_size;
3456 run->mmio.is_write = vcpu->mmio_is_write;
3460 if (kvm_mmu_unprotect_page_virt(vcpu, cr2))
3461 return EMULATE_DONE;
3462 if (!vcpu->mmio_needed) {
3463 kvm_report_emulation_failure(vcpu, "mmio");
3464 return EMULATE_FAIL;
3466 return EMULATE_DO_MMIO;
3469 kvm_set_rflags(vcpu, vcpu->arch.emulate_ctxt.eflags);
3471 if (vcpu->mmio_is_write) {
3472 vcpu->mmio_needed = 0;
3473 return EMULATE_DO_MMIO;
3476 return EMULATE_DONE;
3478 EXPORT_SYMBOL_GPL(emulate_instruction);
3480 static int pio_copy_data(struct kvm_vcpu *vcpu)
3482 void *p = vcpu->arch.pio_data;
3483 gva_t q = vcpu->arch.pio.guest_gva;
3488 bytes = vcpu->arch.pio.size * vcpu->arch.pio.cur_count;
3489 if (vcpu->arch.pio.in)
3490 ret = kvm_write_guest_virt(q, p, bytes, vcpu, &error_code);
3492 ret = kvm_read_guest_virt(q, p, bytes, vcpu, &error_code);
3494 if (ret == X86EMUL_PROPAGATE_FAULT)
3495 kvm_inject_page_fault(vcpu, q, error_code);
3500 int complete_pio(struct kvm_vcpu *vcpu)
3502 struct kvm_pio_request *io = &vcpu->arch.pio;
3509 val = kvm_register_read(vcpu, VCPU_REGS_RAX);
3510 memcpy(&val, vcpu->arch.pio_data, io->size);
3511 kvm_register_write(vcpu, VCPU_REGS_RAX, val);
3515 r = pio_copy_data(vcpu);
3522 delta *= io->cur_count;
3524 * The size of the register should really depend on
3525 * current address size.
3527 val = kvm_register_read(vcpu, VCPU_REGS_RCX);
3529 kvm_register_write(vcpu, VCPU_REGS_RCX, val);
3535 val = kvm_register_read(vcpu, VCPU_REGS_RDI);
3537 kvm_register_write(vcpu, VCPU_REGS_RDI, val);
3539 val = kvm_register_read(vcpu, VCPU_REGS_RSI);
3541 kvm_register_write(vcpu, VCPU_REGS_RSI, val);
3545 io->count -= io->cur_count;
3551 static int kernel_pio(struct kvm_vcpu *vcpu, void *pd)
3553 /* TODO: String I/O for in kernel device */
3556 if (vcpu->arch.pio.in)
3557 r = kvm_io_bus_read(vcpu->kvm, KVM_PIO_BUS, vcpu->arch.pio.port,
3558 vcpu->arch.pio.size, pd);
3560 r = kvm_io_bus_write(vcpu->kvm, KVM_PIO_BUS,
3561 vcpu->arch.pio.port, vcpu->arch.pio.size,
3566 static int pio_string_write(struct kvm_vcpu *vcpu)
3568 struct kvm_pio_request *io = &vcpu->arch.pio;
3569 void *pd = vcpu->arch.pio_data;
3572 for (i = 0; i < io->cur_count; i++) {
3573 if (kvm_io_bus_write(vcpu->kvm, KVM_PIO_BUS,
3574 io->port, io->size, pd)) {
3583 int kvm_emulate_pio(struct kvm_vcpu *vcpu, int in, int size, unsigned port)
3587 trace_kvm_pio(!in, port, size, 1);
3589 vcpu->run->exit_reason = KVM_EXIT_IO;
3590 vcpu->run->io.direction = in ? KVM_EXIT_IO_IN : KVM_EXIT_IO_OUT;
3591 vcpu->run->io.size = vcpu->arch.pio.size = size;
3592 vcpu->run->io.data_offset = KVM_PIO_PAGE_OFFSET * PAGE_SIZE;
3593 vcpu->run->io.count = vcpu->arch.pio.count = vcpu->arch.pio.cur_count = 1;
3594 vcpu->run->io.port = vcpu->arch.pio.port = port;
3595 vcpu->arch.pio.in = in;
3596 vcpu->arch.pio.string = 0;
3597 vcpu->arch.pio.down = 0;
3598 vcpu->arch.pio.rep = 0;
3600 if (!vcpu->arch.pio.in) {
3601 val = kvm_register_read(vcpu, VCPU_REGS_RAX);
3602 memcpy(vcpu->arch.pio_data, &val, 4);
3605 if (!kernel_pio(vcpu, vcpu->arch.pio_data)) {
3611 EXPORT_SYMBOL_GPL(kvm_emulate_pio);
3613 int kvm_emulate_pio_string(struct kvm_vcpu *vcpu, int in,
3614 int size, unsigned long count, int down,
3615 gva_t address, int rep, unsigned port)
3617 unsigned now, in_page;
3620 trace_kvm_pio(!in, port, size, count);
3622 vcpu->run->exit_reason = KVM_EXIT_IO;
3623 vcpu->run->io.direction = in ? KVM_EXIT_IO_IN : KVM_EXIT_IO_OUT;
3624 vcpu->run->io.size = vcpu->arch.pio.size = size;
3625 vcpu->run->io.data_offset = KVM_PIO_PAGE_OFFSET * PAGE_SIZE;
3626 vcpu->run->io.count = vcpu->arch.pio.count = vcpu->arch.pio.cur_count = count;
3627 vcpu->run->io.port = vcpu->arch.pio.port = port;
3628 vcpu->arch.pio.in = in;
3629 vcpu->arch.pio.string = 1;
3630 vcpu->arch.pio.down = down;
3631 vcpu->arch.pio.rep = rep;
3634 kvm_x86_ops->skip_emulated_instruction(vcpu);
3639 in_page = PAGE_SIZE - offset_in_page(address);
3641 in_page = offset_in_page(address) + size;
3642 now = min(count, (unsigned long)in_page / size);
3647 * String I/O in reverse. Yuck. Kill the guest, fix later.
3649 pr_unimpl(vcpu, "guest string pio down\n");
3650 kvm_inject_gp(vcpu, 0);
3653 vcpu->run->io.count = now;
3654 vcpu->arch.pio.cur_count = now;
3656 if (vcpu->arch.pio.cur_count == vcpu->arch.pio.count)
3657 kvm_x86_ops->skip_emulated_instruction(vcpu);
3659 vcpu->arch.pio.guest_gva = address;
3661 if (!vcpu->arch.pio.in) {
3662 /* string PIO write */
3663 ret = pio_copy_data(vcpu);
3664 if (ret == X86EMUL_PROPAGATE_FAULT)
3666 if (ret == 0 && !pio_string_write(vcpu)) {
3668 if (vcpu->arch.pio.count == 0)
3672 /* no string PIO read support yet */
3676 EXPORT_SYMBOL_GPL(kvm_emulate_pio_string);
3678 static void bounce_off(void *info)
3683 static int kvmclock_cpufreq_notifier(struct notifier_block *nb, unsigned long val,
3686 struct cpufreq_freqs *freq = data;
3688 struct kvm_vcpu *vcpu;
3689 int i, send_ipi = 0;
3691 if (val == CPUFREQ_PRECHANGE && freq->old > freq->new)
3693 if (val == CPUFREQ_POSTCHANGE && freq->old < freq->new)
3695 per_cpu(cpu_tsc_khz, freq->cpu) = freq->new;
3697 spin_lock(&kvm_lock);
3698 list_for_each_entry(kvm, &vm_list, vm_list) {
3699 kvm_for_each_vcpu(i, vcpu, kvm) {
3700 if (vcpu->cpu != freq->cpu)
3702 if (!kvm_request_guest_time_update(vcpu))
3704 if (vcpu->cpu != smp_processor_id())
3708 spin_unlock(&kvm_lock);
3710 if (freq->old < freq->new && send_ipi) {
3712 * We upscale the frequency. Must make the guest
3713 * doesn't see old kvmclock values while running with
3714 * the new frequency, otherwise we risk the guest sees
3715 * time go backwards.
3717 * In case we update the frequency for another cpu
3718 * (which might be in guest context) send an interrupt
3719 * to kick the cpu out of guest context. Next time
3720 * guest context is entered kvmclock will be updated,
3721 * so the guest will not see stale values.
3723 smp_call_function_single(freq->cpu, bounce_off, NULL, 1);
3728 static struct notifier_block kvmclock_cpufreq_notifier_block = {
3729 .notifier_call = kvmclock_cpufreq_notifier
3732 static void kvm_timer_init(void)
3736 if (!boot_cpu_has(X86_FEATURE_CONSTANT_TSC)) {
3737 cpufreq_register_notifier(&kvmclock_cpufreq_notifier_block,
3738 CPUFREQ_TRANSITION_NOTIFIER);
3739 for_each_online_cpu(cpu) {
3740 unsigned long khz = cpufreq_get(cpu);
3743 per_cpu(cpu_tsc_khz, cpu) = khz;
3746 for_each_possible_cpu(cpu)
3747 per_cpu(cpu_tsc_khz, cpu) = tsc_khz;
3751 static DEFINE_PER_CPU(struct kvm_vcpu *, current_vcpu);
3753 static int kvm_is_in_guest(void)
3755 return percpu_read(current_vcpu) != NULL;
3758 static int kvm_is_user_mode(void)
3762 if (percpu_read(current_vcpu))
3763 user_mode = kvm_x86_ops->get_cpl(percpu_read(current_vcpu));
3765 return user_mode != 0;
3768 static unsigned long kvm_get_guest_ip(void)
3770 unsigned long ip = 0;
3772 if (percpu_read(current_vcpu))
3773 ip = kvm_rip_read(percpu_read(current_vcpu));
3778 static struct perf_guest_info_callbacks kvm_guest_cbs = {
3779 .is_in_guest = kvm_is_in_guest,
3780 .is_user_mode = kvm_is_user_mode,
3781 .get_guest_ip = kvm_get_guest_ip,
3784 void kvm_before_handle_nmi(struct kvm_vcpu *vcpu)
3786 percpu_write(current_vcpu, vcpu);
3788 EXPORT_SYMBOL_GPL(kvm_before_handle_nmi);
3790 void kvm_after_handle_nmi(struct kvm_vcpu *vcpu)
3792 percpu_write(current_vcpu, NULL);
3794 EXPORT_SYMBOL_GPL(kvm_after_handle_nmi);
3796 int kvm_arch_init(void *opaque)
3799 struct kvm_x86_ops *ops = (struct kvm_x86_ops *)opaque;
3802 printk(KERN_ERR "kvm: already loaded the other module\n");
3807 if (!ops->cpu_has_kvm_support()) {
3808 printk(KERN_ERR "kvm: no hardware support\n");
3812 if (ops->disabled_by_bios()) {
3813 printk(KERN_ERR "kvm: disabled by bios\n");
3818 r = kvm_mmu_module_init();
3822 kvm_init_msr_list();
3825 kvm_mmu_set_nonpresent_ptes(0ull, 0ull);
3826 kvm_mmu_set_base_ptes(PT_PRESENT_MASK);
3827 kvm_mmu_set_mask_ptes(PT_USER_MASK, PT_ACCESSED_MASK,
3828 PT_DIRTY_MASK, PT64_NX_MASK, 0);
3832 perf_register_guest_info_callbacks(&kvm_guest_cbs);
3840 void kvm_arch_exit(void)
3842 perf_unregister_guest_info_callbacks(&kvm_guest_cbs);
3844 if (!boot_cpu_has(X86_FEATURE_CONSTANT_TSC))
3845 cpufreq_unregister_notifier(&kvmclock_cpufreq_notifier_block,
3846 CPUFREQ_TRANSITION_NOTIFIER);
3848 kvm_mmu_module_exit();
3851 int kvm_emulate_halt(struct kvm_vcpu *vcpu)
3853 ++vcpu->stat.halt_exits;
3854 if (irqchip_in_kernel(vcpu->kvm)) {
3855 vcpu->arch.mp_state = KVM_MP_STATE_HALTED;
3858 vcpu->run->exit_reason = KVM_EXIT_HLT;
3862 EXPORT_SYMBOL_GPL(kvm_emulate_halt);
3864 static inline gpa_t hc_gpa(struct kvm_vcpu *vcpu, unsigned long a0,
3867 if (is_long_mode(vcpu))
3870 return a0 | ((gpa_t)a1 << 32);
3873 int kvm_hv_hypercall(struct kvm_vcpu *vcpu)
3875 u64 param, ingpa, outgpa, ret;
3876 uint16_t code, rep_idx, rep_cnt, res = HV_STATUS_SUCCESS, rep_done = 0;
3877 bool fast, longmode;
3881 * hypercall generates UD from non zero cpl and real mode
3884 if (kvm_x86_ops->get_cpl(vcpu) != 0 || !is_protmode(vcpu)) {
3885 kvm_queue_exception(vcpu, UD_VECTOR);
3889 kvm_x86_ops->get_cs_db_l_bits(vcpu, &cs_db, &cs_l);
3890 longmode = is_long_mode(vcpu) && cs_l == 1;
3893 param = ((u64)kvm_register_read(vcpu, VCPU_REGS_RDX) << 32) |
3894 (kvm_register_read(vcpu, VCPU_REGS_RAX) & 0xffffffff);
3895 ingpa = ((u64)kvm_register_read(vcpu, VCPU_REGS_RBX) << 32) |
3896 (kvm_register_read(vcpu, VCPU_REGS_RCX) & 0xffffffff);
3897 outgpa = ((u64)kvm_register_read(vcpu, VCPU_REGS_RDI) << 32) |
3898 (kvm_register_read(vcpu, VCPU_REGS_RSI) & 0xffffffff);
3900 #ifdef CONFIG_X86_64
3902 param = kvm_register_read(vcpu, VCPU_REGS_RCX);
3903 ingpa = kvm_register_read(vcpu, VCPU_REGS_RDX);
3904 outgpa = kvm_register_read(vcpu, VCPU_REGS_R8);
3908 code = param & 0xffff;
3909 fast = (param >> 16) & 0x1;
3910 rep_cnt = (param >> 32) & 0xfff;
3911 rep_idx = (param >> 48) & 0xfff;
3913 trace_kvm_hv_hypercall(code, fast, rep_cnt, rep_idx, ingpa, outgpa);
3916 case HV_X64_HV_NOTIFY_LONG_SPIN_WAIT:
3917 kvm_vcpu_on_spin(vcpu);
3920 res = HV_STATUS_INVALID_HYPERCALL_CODE;
3924 ret = res | (((u64)rep_done & 0xfff) << 32);
3926 kvm_register_write(vcpu, VCPU_REGS_RAX, ret);
3928 kvm_register_write(vcpu, VCPU_REGS_RDX, ret >> 32);
3929 kvm_register_write(vcpu, VCPU_REGS_RAX, ret & 0xffffffff);
3935 int kvm_emulate_hypercall(struct kvm_vcpu *vcpu)
3937 unsigned long nr, a0, a1, a2, a3, ret;
3940 if (kvm_hv_hypercall_enabled(vcpu->kvm))
3941 return kvm_hv_hypercall(vcpu);
3943 nr = kvm_register_read(vcpu, VCPU_REGS_RAX);
3944 a0 = kvm_register_read(vcpu, VCPU_REGS_RBX);
3945 a1 = kvm_register_read(vcpu, VCPU_REGS_RCX);
3946 a2 = kvm_register_read(vcpu, VCPU_REGS_RDX);
3947 a3 = kvm_register_read(vcpu, VCPU_REGS_RSI);
3949 trace_kvm_hypercall(nr, a0, a1, a2, a3);
3951 if (!is_long_mode(vcpu)) {
3959 if (kvm_x86_ops->get_cpl(vcpu) != 0) {
3965 case KVM_HC_VAPIC_POLL_IRQ:
3969 r = kvm_pv_mmu_op(vcpu, a0, hc_gpa(vcpu, a1, a2), &ret);
3976 kvm_register_write(vcpu, VCPU_REGS_RAX, ret);
3977 ++vcpu->stat.hypercalls;
3980 EXPORT_SYMBOL_GPL(kvm_emulate_hypercall);
3982 int kvm_fix_hypercall(struct kvm_vcpu *vcpu)
3984 char instruction[3];
3985 unsigned long rip = kvm_rip_read(vcpu);
3988 * Blow out the MMU to ensure that no other VCPU has an active mapping
3989 * to ensure that the updated hypercall appears atomically across all
3992 kvm_mmu_zap_all(vcpu->kvm);
3994 kvm_x86_ops->patch_hypercall(vcpu, instruction);
3996 return emulator_write_emulated(rip, instruction, 3, vcpu);
3999 static u64 mk_cr_64(u64 curr_cr, u32 new_val)
4001 return (curr_cr & ~((1ULL << 32) - 1)) | new_val;
4004 void realmode_lgdt(struct kvm_vcpu *vcpu, u16 limit, unsigned long base)
4006 struct descriptor_table dt = { limit, base };
4008 kvm_x86_ops->set_gdt(vcpu, &dt);
4011 void realmode_lidt(struct kvm_vcpu *vcpu, u16 limit, unsigned long base)
4013 struct descriptor_table dt = { limit, base };
4015 kvm_x86_ops->set_idt(vcpu, &dt);
4018 void realmode_lmsw(struct kvm_vcpu *vcpu, unsigned long msw,
4019 unsigned long *rflags)
4021 kvm_lmsw(vcpu, msw);
4022 *rflags = kvm_get_rflags(vcpu);
4025 unsigned long realmode_get_cr(struct kvm_vcpu *vcpu, int cr)
4027 unsigned long value;
4031 value = kvm_read_cr0(vcpu);
4034 value = vcpu->arch.cr2;
4037 value = vcpu->arch.cr3;
4040 value = kvm_read_cr4(vcpu);
4043 value = kvm_get_cr8(vcpu);
4046 vcpu_printf(vcpu, "%s: unexpected cr %u\n", __func__, cr);
4053 void realmode_set_cr(struct kvm_vcpu *vcpu, int cr, unsigned long val,
4054 unsigned long *rflags)
4058 kvm_set_cr0(vcpu, mk_cr_64(kvm_read_cr0(vcpu), val));
4059 *rflags = kvm_get_rflags(vcpu);
4062 vcpu->arch.cr2 = val;
4065 kvm_set_cr3(vcpu, val);
4068 kvm_set_cr4(vcpu, mk_cr_64(kvm_read_cr4(vcpu), val));
4071 kvm_set_cr8(vcpu, val & 0xfUL);
4074 vcpu_printf(vcpu, "%s: unexpected cr %u\n", __func__, cr);
4078 static int move_to_next_stateful_cpuid_entry(struct kvm_vcpu *vcpu, int i)
4080 struct kvm_cpuid_entry2 *e = &vcpu->arch.cpuid_entries[i];
4081 int j, nent = vcpu->arch.cpuid_nent;
4083 e->flags &= ~KVM_CPUID_FLAG_STATE_READ_NEXT;
4084 /* when no next entry is found, the current entry[i] is reselected */
4085 for (j = i + 1; ; j = (j + 1) % nent) {
4086 struct kvm_cpuid_entry2 *ej = &vcpu->arch.cpuid_entries[j];
4087 if (ej->function == e->function) {
4088 ej->flags |= KVM_CPUID_FLAG_STATE_READ_NEXT;
4092 return 0; /* silence gcc, even though control never reaches here */
4095 /* find an entry with matching function, matching index (if needed), and that
4096 * should be read next (if it's stateful) */
4097 static int is_matching_cpuid_entry(struct kvm_cpuid_entry2 *e,
4098 u32 function, u32 index)
4100 if (e->function != function)
4102 if ((e->flags & KVM_CPUID_FLAG_SIGNIFCANT_INDEX) && e->index != index)
4104 if ((e->flags & KVM_CPUID_FLAG_STATEFUL_FUNC) &&
4105 !(e->flags & KVM_CPUID_FLAG_STATE_READ_NEXT))
4110 struct kvm_cpuid_entry2 *kvm_find_cpuid_entry(struct kvm_vcpu *vcpu,
4111 u32 function, u32 index)
4114 struct kvm_cpuid_entry2 *best = NULL;
4116 for (i = 0; i < vcpu->arch.cpuid_nent; ++i) {
4117 struct kvm_cpuid_entry2 *e;
4119 e = &vcpu->arch.cpuid_entries[i];
4120 if (is_matching_cpuid_entry(e, function, index)) {
4121 if (e->flags & KVM_CPUID_FLAG_STATEFUL_FUNC)
4122 move_to_next_stateful_cpuid_entry(vcpu, i);
4127 * Both basic or both extended?
4129 if (((e->function ^ function) & 0x80000000) == 0)
4130 if (!best || e->function > best->function)
4135 EXPORT_SYMBOL_GPL(kvm_find_cpuid_entry);
4137 int cpuid_maxphyaddr(struct kvm_vcpu *vcpu)
4139 struct kvm_cpuid_entry2 *best;
4141 best = kvm_find_cpuid_entry(vcpu, 0x80000008, 0);
4143 return best->eax & 0xff;
4147 void kvm_emulate_cpuid(struct kvm_vcpu *vcpu)
4149 u32 function, index;
4150 struct kvm_cpuid_entry2 *best;
4152 function = kvm_register_read(vcpu, VCPU_REGS_RAX);
4153 index = kvm_register_read(vcpu, VCPU_REGS_RCX);
4154 kvm_register_write(vcpu, VCPU_REGS_RAX, 0);
4155 kvm_register_write(vcpu, VCPU_REGS_RBX, 0);
4156 kvm_register_write(vcpu, VCPU_REGS_RCX, 0);
4157 kvm_register_write(vcpu, VCPU_REGS_RDX, 0);
4158 best = kvm_find_cpuid_entry(vcpu, function, index);
4160 kvm_register_write(vcpu, VCPU_REGS_RAX, best->eax);
4161 kvm_register_write(vcpu, VCPU_REGS_RBX, best->ebx);
4162 kvm_register_write(vcpu, VCPU_REGS_RCX, best->ecx);
4163 kvm_register_write(vcpu, VCPU_REGS_RDX, best->edx);
4165 kvm_x86_ops->skip_emulated_instruction(vcpu);
4166 trace_kvm_cpuid(function,
4167 kvm_register_read(vcpu, VCPU_REGS_RAX),
4168 kvm_register_read(vcpu, VCPU_REGS_RBX),
4169 kvm_register_read(vcpu, VCPU_REGS_RCX),
4170 kvm_register_read(vcpu, VCPU_REGS_RDX));
4172 EXPORT_SYMBOL_GPL(kvm_emulate_cpuid);
4175 * Check if userspace requested an interrupt window, and that the
4176 * interrupt window is open.
4178 * No need to exit to userspace if we already have an interrupt queued.
4180 static int dm_request_for_irq_injection(struct kvm_vcpu *vcpu)
4182 return (!irqchip_in_kernel(vcpu->kvm) && !kvm_cpu_has_interrupt(vcpu) &&
4183 vcpu->run->request_interrupt_window &&
4184 kvm_arch_interrupt_allowed(vcpu));
4187 static void post_kvm_run_save(struct kvm_vcpu *vcpu)
4189 struct kvm_run *kvm_run = vcpu->run;
4191 kvm_run->if_flag = (kvm_get_rflags(vcpu) & X86_EFLAGS_IF) != 0;
4192 kvm_run->cr8 = kvm_get_cr8(vcpu);
4193 kvm_run->apic_base = kvm_get_apic_base(vcpu);
4194 if (irqchip_in_kernel(vcpu->kvm))
4195 kvm_run->ready_for_interrupt_injection = 1;
4197 kvm_run->ready_for_interrupt_injection =
4198 kvm_arch_interrupt_allowed(vcpu) &&
4199 !kvm_cpu_has_interrupt(vcpu) &&
4200 !kvm_event_needs_reinjection(vcpu);
4203 static void vapic_enter(struct kvm_vcpu *vcpu)
4205 struct kvm_lapic *apic = vcpu->arch.apic;
4208 if (!apic || !apic->vapic_addr)
4211 page = gfn_to_page(vcpu->kvm, apic->vapic_addr >> PAGE_SHIFT);
4213 vcpu->arch.apic->vapic_page = page;
4216 static void vapic_exit(struct kvm_vcpu *vcpu)
4218 struct kvm_lapic *apic = vcpu->arch.apic;
4221 if (!apic || !apic->vapic_addr)
4224 idx = srcu_read_lock(&vcpu->kvm->srcu);
4225 kvm_release_page_dirty(apic->vapic_page);
4226 mark_page_dirty(vcpu->kvm, apic->vapic_addr >> PAGE_SHIFT);
4227 srcu_read_unlock(&vcpu->kvm->srcu, idx);
4230 static void update_cr8_intercept(struct kvm_vcpu *vcpu)
4234 if (!kvm_x86_ops->update_cr8_intercept)
4237 if (!vcpu->arch.apic)
4240 if (!vcpu->arch.apic->vapic_addr)
4241 max_irr = kvm_lapic_find_highest_irr(vcpu);
4248 tpr = kvm_lapic_get_cr8(vcpu);
4250 kvm_x86_ops->update_cr8_intercept(vcpu, tpr, max_irr);
4253 static void inject_pending_event(struct kvm_vcpu *vcpu)
4255 /* try to reinject previous events if any */
4256 if (vcpu->arch.exception.pending) {
4257 kvm_x86_ops->queue_exception(vcpu, vcpu->arch.exception.nr,
4258 vcpu->arch.exception.has_error_code,
4259 vcpu->arch.exception.error_code);
4263 if (vcpu->arch.nmi_injected) {
4264 kvm_x86_ops->set_nmi(vcpu);
4268 if (vcpu->arch.interrupt.pending) {
4269 kvm_x86_ops->set_irq(vcpu);
4273 /* try to inject new event if pending */
4274 if (vcpu->arch.nmi_pending) {
4275 if (kvm_x86_ops->nmi_allowed(vcpu)) {
4276 vcpu->arch.nmi_pending = false;
4277 vcpu->arch.nmi_injected = true;
4278 kvm_x86_ops->set_nmi(vcpu);
4280 } else if (kvm_cpu_has_interrupt(vcpu)) {
4281 if (kvm_x86_ops->interrupt_allowed(vcpu)) {
4282 kvm_queue_interrupt(vcpu, kvm_cpu_get_interrupt(vcpu),
4284 kvm_x86_ops->set_irq(vcpu);
4289 static int vcpu_enter_guest(struct kvm_vcpu *vcpu)
4292 bool req_int_win = !irqchip_in_kernel(vcpu->kvm) &&
4293 vcpu->run->request_interrupt_window;
4296 if (test_and_clear_bit(KVM_REQ_MMU_RELOAD, &vcpu->requests))
4297 kvm_mmu_unload(vcpu);
4299 r = kvm_mmu_reload(vcpu);
4303 if (vcpu->requests) {
4304 if (test_and_clear_bit(KVM_REQ_MIGRATE_TIMER, &vcpu->requests))
4305 __kvm_migrate_timers(vcpu);
4306 if (test_and_clear_bit(KVM_REQ_KVMCLOCK_UPDATE, &vcpu->requests))
4307 kvm_write_guest_time(vcpu);
4308 if (test_and_clear_bit(KVM_REQ_MMU_SYNC, &vcpu->requests))
4309 kvm_mmu_sync_roots(vcpu);
4310 if (test_and_clear_bit(KVM_REQ_TLB_FLUSH, &vcpu->requests))
4311 kvm_x86_ops->tlb_flush(vcpu);
4312 if (test_and_clear_bit(KVM_REQ_REPORT_TPR_ACCESS,
4314 vcpu->run->exit_reason = KVM_EXIT_TPR_ACCESS;
4318 if (test_and_clear_bit(KVM_REQ_TRIPLE_FAULT, &vcpu->requests)) {
4319 vcpu->run->exit_reason = KVM_EXIT_SHUTDOWN;
4323 if (test_and_clear_bit(KVM_REQ_DEACTIVATE_FPU, &vcpu->requests)) {
4324 vcpu->fpu_active = 0;
4325 kvm_x86_ops->fpu_deactivate(vcpu);
4331 kvm_x86_ops->prepare_guest_switch(vcpu);
4332 if (vcpu->fpu_active)
4333 kvm_load_guest_fpu(vcpu);
4335 local_irq_disable();
4337 clear_bit(KVM_REQ_KICK, &vcpu->requests);
4338 smp_mb__after_clear_bit();
4340 if (vcpu->requests || need_resched() || signal_pending(current)) {
4341 set_bit(KVM_REQ_KICK, &vcpu->requests);
4348 inject_pending_event(vcpu);
4350 /* enable NMI/IRQ window open exits if needed */
4351 if (vcpu->arch.nmi_pending)
4352 kvm_x86_ops->enable_nmi_window(vcpu);
4353 else if (kvm_cpu_has_interrupt(vcpu) || req_int_win)
4354 kvm_x86_ops->enable_irq_window(vcpu);
4356 if (kvm_lapic_enabled(vcpu)) {
4357 update_cr8_intercept(vcpu);
4358 kvm_lapic_sync_to_vapic(vcpu);
4361 srcu_read_unlock(&vcpu->kvm->srcu, vcpu->srcu_idx);
4365 if (unlikely(vcpu->arch.switch_db_regs)) {
4367 set_debugreg(vcpu->arch.eff_db[0], 0);
4368 set_debugreg(vcpu->arch.eff_db[1], 1);
4369 set_debugreg(vcpu->arch.eff_db[2], 2);
4370 set_debugreg(vcpu->arch.eff_db[3], 3);
4373 trace_kvm_entry(vcpu->vcpu_id);
4374 kvm_x86_ops->run(vcpu);
4377 * If the guest has used debug registers, at least dr7
4378 * will be disabled while returning to the host.
4379 * If we don't have active breakpoints in the host, we don't
4380 * care about the messed up debug address registers. But if
4381 * we have some of them active, restore the old state.
4383 if (hw_breakpoint_active())
4384 hw_breakpoint_restore();
4386 set_bit(KVM_REQ_KICK, &vcpu->requests);
4392 * We must have an instruction between local_irq_enable() and
4393 * kvm_guest_exit(), so the timer interrupt isn't delayed by
4394 * the interrupt shadow. The stat.exits increment will do nicely.
4395 * But we need to prevent reordering, hence this barrier():
4403 vcpu->srcu_idx = srcu_read_lock(&vcpu->kvm->srcu);
4406 * Profile KVM exit RIPs:
4408 if (unlikely(prof_on == KVM_PROFILING)) {
4409 unsigned long rip = kvm_rip_read(vcpu);
4410 profile_hit(KVM_PROFILING, (void *)rip);
4414 kvm_lapic_sync_from_vapic(vcpu);
4416 r = kvm_x86_ops->handle_exit(vcpu);
4422 static int __vcpu_run(struct kvm_vcpu *vcpu)
4425 struct kvm *kvm = vcpu->kvm;
4427 if (unlikely(vcpu->arch.mp_state == KVM_MP_STATE_SIPI_RECEIVED)) {
4428 pr_debug("vcpu %d received sipi with vector # %x\n",
4429 vcpu->vcpu_id, vcpu->arch.sipi_vector);
4430 kvm_lapic_reset(vcpu);
4431 r = kvm_arch_vcpu_reset(vcpu);
4434 vcpu->arch.mp_state = KVM_MP_STATE_RUNNABLE;
4437 vcpu->srcu_idx = srcu_read_lock(&kvm->srcu);
4442 if (vcpu->arch.mp_state == KVM_MP_STATE_RUNNABLE)
4443 r = vcpu_enter_guest(vcpu);
4445 srcu_read_unlock(&kvm->srcu, vcpu->srcu_idx);
4446 kvm_vcpu_block(vcpu);
4447 vcpu->srcu_idx = srcu_read_lock(&kvm->srcu);
4448 if (test_and_clear_bit(KVM_REQ_UNHALT, &vcpu->requests))
4450 switch(vcpu->arch.mp_state) {
4451 case KVM_MP_STATE_HALTED:
4452 vcpu->arch.mp_state =
4453 KVM_MP_STATE_RUNNABLE;
4454 case KVM_MP_STATE_RUNNABLE:
4456 case KVM_MP_STATE_SIPI_RECEIVED:
4467 clear_bit(KVM_REQ_PENDING_TIMER, &vcpu->requests);
4468 if (kvm_cpu_has_pending_timer(vcpu))
4469 kvm_inject_pending_timer_irqs(vcpu);
4471 if (dm_request_for_irq_injection(vcpu)) {
4473 vcpu->run->exit_reason = KVM_EXIT_INTR;
4474 ++vcpu->stat.request_irq_exits;
4476 if (signal_pending(current)) {
4478 vcpu->run->exit_reason = KVM_EXIT_INTR;
4479 ++vcpu->stat.signal_exits;
4481 if (need_resched()) {
4482 srcu_read_unlock(&kvm->srcu, vcpu->srcu_idx);
4484 vcpu->srcu_idx = srcu_read_lock(&kvm->srcu);
4488 srcu_read_unlock(&kvm->srcu, vcpu->srcu_idx);
4489 post_kvm_run_save(vcpu);
4496 int kvm_arch_vcpu_ioctl_run(struct kvm_vcpu *vcpu, struct kvm_run *kvm_run)
4503 if (vcpu->sigset_active)
4504 sigprocmask(SIG_SETMASK, &vcpu->sigset, &sigsaved);
4506 if (unlikely(vcpu->arch.mp_state == KVM_MP_STATE_UNINITIALIZED)) {
4507 kvm_vcpu_block(vcpu);
4508 clear_bit(KVM_REQ_UNHALT, &vcpu->requests);
4513 /* re-sync apic's tpr */
4514 if (!irqchip_in_kernel(vcpu->kvm))
4515 kvm_set_cr8(vcpu, kvm_run->cr8);
4517 if (vcpu->arch.pio.cur_count) {
4518 vcpu->srcu_idx = srcu_read_lock(&vcpu->kvm->srcu);
4519 r = complete_pio(vcpu);
4520 srcu_read_unlock(&vcpu->kvm->srcu, vcpu->srcu_idx);
4524 if (vcpu->mmio_needed) {
4525 memcpy(vcpu->mmio_data, kvm_run->mmio.data, 8);
4526 vcpu->mmio_read_completed = 1;
4527 vcpu->mmio_needed = 0;
4529 vcpu->srcu_idx = srcu_read_lock(&vcpu->kvm->srcu);
4530 r = emulate_instruction(vcpu, vcpu->arch.mmio_fault_cr2, 0,
4531 EMULTYPE_NO_DECODE);
4532 srcu_read_unlock(&vcpu->kvm->srcu, vcpu->srcu_idx);
4533 if (r == EMULATE_DO_MMIO) {
4535 * Read-modify-write. Back to userspace.
4541 if (kvm_run->exit_reason == KVM_EXIT_HYPERCALL)
4542 kvm_register_write(vcpu, VCPU_REGS_RAX,
4543 kvm_run->hypercall.ret);
4545 r = __vcpu_run(vcpu);
4548 if (vcpu->sigset_active)
4549 sigprocmask(SIG_SETMASK, &sigsaved, NULL);
4555 int kvm_arch_vcpu_ioctl_get_regs(struct kvm_vcpu *vcpu, struct kvm_regs *regs)
4559 regs->rax = kvm_register_read(vcpu, VCPU_REGS_RAX);
4560 regs->rbx = kvm_register_read(vcpu, VCPU_REGS_RBX);
4561 regs->rcx = kvm_register_read(vcpu, VCPU_REGS_RCX);
4562 regs->rdx = kvm_register_read(vcpu, VCPU_REGS_RDX);
4563 regs->rsi = kvm_register_read(vcpu, VCPU_REGS_RSI);
4564 regs->rdi = kvm_register_read(vcpu, VCPU_REGS_RDI);
4565 regs->rsp = kvm_register_read(vcpu, VCPU_REGS_RSP);
4566 regs->rbp = kvm_register_read(vcpu, VCPU_REGS_RBP);
4567 #ifdef CONFIG_X86_64
4568 regs->r8 = kvm_register_read(vcpu, VCPU_REGS_R8);
4569 regs->r9 = kvm_register_read(vcpu, VCPU_REGS_R9);
4570 regs->r10 = kvm_register_read(vcpu, VCPU_REGS_R10);
4571 regs->r11 = kvm_register_read(vcpu, VCPU_REGS_R11);
4572 regs->r12 = kvm_register_read(vcpu, VCPU_REGS_R12);
4573 regs->r13 = kvm_register_read(vcpu, VCPU_REGS_R13);
4574 regs->r14 = kvm_register_read(vcpu, VCPU_REGS_R14);
4575 regs->r15 = kvm_register_read(vcpu, VCPU_REGS_R15);
4578 regs->rip = kvm_rip_read(vcpu);
4579 regs->rflags = kvm_get_rflags(vcpu);
4586 int kvm_arch_vcpu_ioctl_set_regs(struct kvm_vcpu *vcpu, struct kvm_regs *regs)
4590 kvm_register_write(vcpu, VCPU_REGS_RAX, regs->rax);
4591 kvm_register_write(vcpu, VCPU_REGS_RBX, regs->rbx);
4592 kvm_register_write(vcpu, VCPU_REGS_RCX, regs->rcx);
4593 kvm_register_write(vcpu, VCPU_REGS_RDX, regs->rdx);
4594 kvm_register_write(vcpu, VCPU_REGS_RSI, regs->rsi);
4595 kvm_register_write(vcpu, VCPU_REGS_RDI, regs->rdi);
4596 kvm_register_write(vcpu, VCPU_REGS_RSP, regs->rsp);
4597 kvm_register_write(vcpu, VCPU_REGS_RBP, regs->rbp);
4598 #ifdef CONFIG_X86_64
4599 kvm_register_write(vcpu, VCPU_REGS_R8, regs->r8);
4600 kvm_register_write(vcpu, VCPU_REGS_R9, regs->r9);
4601 kvm_register_write(vcpu, VCPU_REGS_R10, regs->r10);
4602 kvm_register_write(vcpu, VCPU_REGS_R11, regs->r11);
4603 kvm_register_write(vcpu, VCPU_REGS_R12, regs->r12);
4604 kvm_register_write(vcpu, VCPU_REGS_R13, regs->r13);
4605 kvm_register_write(vcpu, VCPU_REGS_R14, regs->r14);
4606 kvm_register_write(vcpu, VCPU_REGS_R15, regs->r15);
4609 kvm_rip_write(vcpu, regs->rip);
4610 kvm_set_rflags(vcpu, regs->rflags);
4612 vcpu->arch.exception.pending = false;
4619 void kvm_get_segment(struct kvm_vcpu *vcpu,
4620 struct kvm_segment *var, int seg)
4622 kvm_x86_ops->get_segment(vcpu, var, seg);
4625 void kvm_get_cs_db_l_bits(struct kvm_vcpu *vcpu, int *db, int *l)
4627 struct kvm_segment cs;
4629 kvm_get_segment(vcpu, &cs, VCPU_SREG_CS);
4633 EXPORT_SYMBOL_GPL(kvm_get_cs_db_l_bits);
4635 int kvm_arch_vcpu_ioctl_get_sregs(struct kvm_vcpu *vcpu,
4636 struct kvm_sregs *sregs)
4638 struct descriptor_table dt;
4642 kvm_get_segment(vcpu, &sregs->cs, VCPU_SREG_CS);
4643 kvm_get_segment(vcpu, &sregs->ds, VCPU_SREG_DS);
4644 kvm_get_segment(vcpu, &sregs->es, VCPU_SREG_ES);
4645 kvm_get_segment(vcpu, &sregs->fs, VCPU_SREG_FS);
4646 kvm_get_segment(vcpu, &sregs->gs, VCPU_SREG_GS);
4647 kvm_get_segment(vcpu, &sregs->ss, VCPU_SREG_SS);
4649 kvm_get_segment(vcpu, &sregs->tr, VCPU_SREG_TR);
4650 kvm_get_segment(vcpu, &sregs->ldt, VCPU_SREG_LDTR);
4652 kvm_x86_ops->get_idt(vcpu, &dt);
4653 sregs->idt.limit = dt.limit;
4654 sregs->idt.base = dt.base;
4655 kvm_x86_ops->get_gdt(vcpu, &dt);
4656 sregs->gdt.limit = dt.limit;
4657 sregs->gdt.base = dt.base;
4659 sregs->cr0 = kvm_read_cr0(vcpu);
4660 sregs->cr2 = vcpu->arch.cr2;
4661 sregs->cr3 = vcpu->arch.cr3;
4662 sregs->cr4 = kvm_read_cr4(vcpu);
4663 sregs->cr8 = kvm_get_cr8(vcpu);
4664 sregs->efer = vcpu->arch.efer;
4665 sregs->apic_base = kvm_get_apic_base(vcpu);
4667 memset(sregs->interrupt_bitmap, 0, sizeof sregs->interrupt_bitmap);
4669 if (vcpu->arch.interrupt.pending && !vcpu->arch.interrupt.soft)
4670 set_bit(vcpu->arch.interrupt.nr,
4671 (unsigned long *)sregs->interrupt_bitmap);
4678 int kvm_arch_vcpu_ioctl_get_mpstate(struct kvm_vcpu *vcpu,
4679 struct kvm_mp_state *mp_state)
4682 mp_state->mp_state = vcpu->arch.mp_state;
4687 int kvm_arch_vcpu_ioctl_set_mpstate(struct kvm_vcpu *vcpu,
4688 struct kvm_mp_state *mp_state)
4691 vcpu->arch.mp_state = mp_state->mp_state;
4696 static void kvm_set_segment(struct kvm_vcpu *vcpu,
4697 struct kvm_segment *var, int seg)
4699 kvm_x86_ops->set_segment(vcpu, var, seg);
4702 static void seg_desct_to_kvm_desct(struct desc_struct *seg_desc, u16 selector,
4703 struct kvm_segment *kvm_desct)
4705 kvm_desct->base = get_desc_base(seg_desc);
4706 kvm_desct->limit = get_desc_limit(seg_desc);
4708 kvm_desct->limit <<= 12;
4709 kvm_desct->limit |= 0xfff;
4711 kvm_desct->selector = selector;
4712 kvm_desct->type = seg_desc->type;
4713 kvm_desct->present = seg_desc->p;
4714 kvm_desct->dpl = seg_desc->dpl;
4715 kvm_desct->db = seg_desc->d;
4716 kvm_desct->s = seg_desc->s;
4717 kvm_desct->l = seg_desc->l;
4718 kvm_desct->g = seg_desc->g;
4719 kvm_desct->avl = seg_desc->avl;
4721 kvm_desct->unusable = 1;
4723 kvm_desct->unusable = 0;
4724 kvm_desct->padding = 0;
4727 static void get_segment_descriptor_dtable(struct kvm_vcpu *vcpu,
4729 struct descriptor_table *dtable)
4731 if (selector & 1 << 2) {
4732 struct kvm_segment kvm_seg;
4734 kvm_get_segment(vcpu, &kvm_seg, VCPU_SREG_LDTR);
4736 if (kvm_seg.unusable)
4739 dtable->limit = kvm_seg.limit;
4740 dtable->base = kvm_seg.base;
4743 kvm_x86_ops->get_gdt(vcpu, dtable);
4746 /* allowed just for 8 bytes segments */
4747 static int load_guest_segment_descriptor(struct kvm_vcpu *vcpu, u16 selector,
4748 struct desc_struct *seg_desc)
4750 struct descriptor_table dtable;
4751 u16 index = selector >> 3;
4756 get_segment_descriptor_dtable(vcpu, selector, &dtable);
4758 if (dtable.limit < index * 8 + 7) {
4759 kvm_queue_exception_e(vcpu, GP_VECTOR, selector & 0xfffc);
4760 return X86EMUL_PROPAGATE_FAULT;
4762 addr = dtable.base + index * 8;
4763 ret = kvm_read_guest_virt_system(addr, seg_desc, sizeof(*seg_desc),
4765 if (ret == X86EMUL_PROPAGATE_FAULT)
4766 kvm_inject_page_fault(vcpu, addr, err);
4771 /* allowed just for 8 bytes segments */
4772 static int save_guest_segment_descriptor(struct kvm_vcpu *vcpu, u16 selector,
4773 struct desc_struct *seg_desc)
4775 struct descriptor_table dtable;
4776 u16 index = selector >> 3;
4778 get_segment_descriptor_dtable(vcpu, selector, &dtable);
4780 if (dtable.limit < index * 8 + 7)
4782 return kvm_write_guest_virt(dtable.base + index*8, seg_desc, sizeof(*seg_desc), vcpu, NULL);
4785 static gpa_t get_tss_base_addr_write(struct kvm_vcpu *vcpu,
4786 struct desc_struct *seg_desc)
4788 u32 base_addr = get_desc_base(seg_desc);
4790 return kvm_mmu_gva_to_gpa_write(vcpu, base_addr, NULL);
4793 static gpa_t get_tss_base_addr_read(struct kvm_vcpu *vcpu,
4794 struct desc_struct *seg_desc)
4796 u32 base_addr = get_desc_base(seg_desc);
4798 return kvm_mmu_gva_to_gpa_read(vcpu, base_addr, NULL);
4801 static u16 get_segment_selector(struct kvm_vcpu *vcpu, int seg)
4803 struct kvm_segment kvm_seg;
4805 kvm_get_segment(vcpu, &kvm_seg, seg);
4806 return kvm_seg.selector;
4809 static int kvm_load_realmode_segment(struct kvm_vcpu *vcpu, u16 selector, int seg)
4811 struct kvm_segment segvar = {
4812 .base = selector << 4,
4814 .selector = selector,
4825 kvm_x86_ops->set_segment(vcpu, &segvar, seg);
4826 return X86EMUL_CONTINUE;
4829 static int is_vm86_segment(struct kvm_vcpu *vcpu, int seg)
4831 return (seg != VCPU_SREG_LDTR) &&
4832 (seg != VCPU_SREG_TR) &&
4833 (kvm_get_rflags(vcpu) & X86_EFLAGS_VM);
4836 int kvm_load_segment_descriptor(struct kvm_vcpu *vcpu, u16 selector, int seg)
4838 struct kvm_segment kvm_seg;
4839 struct desc_struct seg_desc;
4841 unsigned err_vec = GP_VECTOR;
4843 bool null_selector = !(selector & ~0x3); /* 0000-0003 are null */
4846 if (is_vm86_segment(vcpu, seg) || !is_protmode(vcpu))
4847 return kvm_load_realmode_segment(vcpu, selector, seg);
4849 /* NULL selector is not valid for TR, CS and SS */
4850 if ((seg == VCPU_SREG_CS || seg == VCPU_SREG_SS || seg == VCPU_SREG_TR)
4854 /* TR should be in GDT only */
4855 if (seg == VCPU_SREG_TR && (selector & (1 << 2)))
4858 ret = load_guest_segment_descriptor(vcpu, selector, &seg_desc);
4862 seg_desct_to_kvm_desct(&seg_desc, selector, &kvm_seg);
4864 if (null_selector) { /* for NULL selector skip all following checks */
4865 kvm_seg.unusable = 1;
4869 err_code = selector & 0xfffc;
4870 err_vec = GP_VECTOR;
4872 /* can't load system descriptor into segment selecor */
4873 if (seg <= VCPU_SREG_GS && !kvm_seg.s)
4876 if (!kvm_seg.present) {
4877 err_vec = (seg == VCPU_SREG_SS) ? SS_VECTOR : NP_VECTOR;
4883 cpl = kvm_x86_ops->get_cpl(vcpu);
4888 * segment is not a writable data segment or segment
4889 * selector's RPL != CPL or segment selector's RPL != CPL
4891 if (rpl != cpl || (kvm_seg.type & 0xa) != 0x2 || dpl != cpl)
4895 if (!(kvm_seg.type & 8))
4898 if (kvm_seg.type & 4) {
4904 if (rpl > cpl || dpl != cpl)
4907 /* CS(RPL) <- CPL */
4908 selector = (selector & 0xfffc) | cpl;
4911 if (kvm_seg.s || (kvm_seg.type != 1 && kvm_seg.type != 9))
4914 case VCPU_SREG_LDTR:
4915 if (kvm_seg.s || kvm_seg.type != 2)
4918 default: /* DS, ES, FS, or GS */
4920 * segment is not a data or readable code segment or
4921 * ((segment is a data or nonconforming code segment)
4922 * and (both RPL and CPL > DPL))
4924 if ((kvm_seg.type & 0xa) == 0x8 ||
4925 (((kvm_seg.type & 0xc) != 0xc) && (rpl > dpl && cpl > dpl)))
4930 if (!kvm_seg.unusable && kvm_seg.s) {
4931 /* mark segment as accessed */
4934 save_guest_segment_descriptor(vcpu, selector, &seg_desc);
4937 kvm_set_segment(vcpu, &kvm_seg, seg);
4938 return X86EMUL_CONTINUE;
4940 kvm_queue_exception_e(vcpu, err_vec, err_code);
4941 return X86EMUL_PROPAGATE_FAULT;
4944 static void save_state_to_tss32(struct kvm_vcpu *vcpu,
4945 struct tss_segment_32 *tss)
4947 tss->cr3 = vcpu->arch.cr3;
4948 tss->eip = kvm_rip_read(vcpu);
4949 tss->eflags = kvm_get_rflags(vcpu);
4950 tss->eax = kvm_register_read(vcpu, VCPU_REGS_RAX);
4951 tss->ecx = kvm_register_read(vcpu, VCPU_REGS_RCX);
4952 tss->edx = kvm_register_read(vcpu, VCPU_REGS_RDX);
4953 tss->ebx = kvm_register_read(vcpu, VCPU_REGS_RBX);
4954 tss->esp = kvm_register_read(vcpu, VCPU_REGS_RSP);
4955 tss->ebp = kvm_register_read(vcpu, VCPU_REGS_RBP);
4956 tss->esi = kvm_register_read(vcpu, VCPU_REGS_RSI);
4957 tss->edi = kvm_register_read(vcpu, VCPU_REGS_RDI);
4958 tss->es = get_segment_selector(vcpu, VCPU_SREG_ES);
4959 tss->cs = get_segment_selector(vcpu, VCPU_SREG_CS);
4960 tss->ss = get_segment_selector(vcpu, VCPU_SREG_SS);
4961 tss->ds = get_segment_selector(vcpu, VCPU_SREG_DS);
4962 tss->fs = get_segment_selector(vcpu, VCPU_SREG_FS);
4963 tss->gs = get_segment_selector(vcpu, VCPU_SREG_GS);
4964 tss->ldt_selector = get_segment_selector(vcpu, VCPU_SREG_LDTR);
4967 static void kvm_load_segment_selector(struct kvm_vcpu *vcpu, u16 sel, int seg)
4969 struct kvm_segment kvm_seg;
4970 kvm_get_segment(vcpu, &kvm_seg, seg);
4971 kvm_seg.selector = sel;
4972 kvm_set_segment(vcpu, &kvm_seg, seg);
4975 static int load_state_from_tss32(struct kvm_vcpu *vcpu,
4976 struct tss_segment_32 *tss)
4978 kvm_set_cr3(vcpu, tss->cr3);
4980 kvm_rip_write(vcpu, tss->eip);
4981 kvm_set_rflags(vcpu, tss->eflags | 2);
4983 kvm_register_write(vcpu, VCPU_REGS_RAX, tss->eax);
4984 kvm_register_write(vcpu, VCPU_REGS_RCX, tss->ecx);
4985 kvm_register_write(vcpu, VCPU_REGS_RDX, tss->edx);
4986 kvm_register_write(vcpu, VCPU_REGS_RBX, tss->ebx);
4987 kvm_register_write(vcpu, VCPU_REGS_RSP, tss->esp);
4988 kvm_register_write(vcpu, VCPU_REGS_RBP, tss->ebp);
4989 kvm_register_write(vcpu, VCPU_REGS_RSI, tss->esi);
4990 kvm_register_write(vcpu, VCPU_REGS_RDI, tss->edi);
4993 * SDM says that segment selectors are loaded before segment
4996 kvm_load_segment_selector(vcpu, tss->ldt_selector, VCPU_SREG_LDTR);
4997 kvm_load_segment_selector(vcpu, tss->es, VCPU_SREG_ES);
4998 kvm_load_segment_selector(vcpu, tss->cs, VCPU_SREG_CS);
4999 kvm_load_segment_selector(vcpu, tss->ss, VCPU_SREG_SS);
5000 kvm_load_segment_selector(vcpu, tss->ds, VCPU_SREG_DS);
5001 kvm_load_segment_selector(vcpu, tss->fs, VCPU_SREG_FS);
5002 kvm_load_segment_selector(vcpu, tss->gs, VCPU_SREG_GS);
5005 * Now load segment descriptors. If fault happenes at this stage
5006 * it is handled in a context of new task
5008 if (kvm_load_segment_descriptor(vcpu, tss->ldt_selector, VCPU_SREG_LDTR))
5011 if (kvm_load_segment_descriptor(vcpu, tss->es, VCPU_SREG_ES))
5014 if (kvm_load_segment_descriptor(vcpu, tss->cs, VCPU_SREG_CS))
5017 if (kvm_load_segment_descriptor(vcpu, tss->ss, VCPU_SREG_SS))
5020 if (kvm_load_segment_descriptor(vcpu, tss->ds, VCPU_SREG_DS))
5023 if (kvm_load_segment_descriptor(vcpu, tss->fs, VCPU_SREG_FS))
5026 if (kvm_load_segment_descriptor(vcpu, tss->gs, VCPU_SREG_GS))
5031 static void save_state_to_tss16(struct kvm_vcpu *vcpu,
5032 struct tss_segment_16 *tss)
5034 tss->ip = kvm_rip_read(vcpu);
5035 tss->flag = kvm_get_rflags(vcpu);
5036 tss->ax = kvm_register_read(vcpu, VCPU_REGS_RAX);
5037 tss->cx = kvm_register_read(vcpu, VCPU_REGS_RCX);
5038 tss->dx = kvm_register_read(vcpu, VCPU_REGS_RDX);
5039 tss->bx = kvm_register_read(vcpu, VCPU_REGS_RBX);
5040 tss->sp = kvm_register_read(vcpu, VCPU_REGS_RSP);
5041 tss->bp = kvm_register_read(vcpu, VCPU_REGS_RBP);
5042 tss->si = kvm_register_read(vcpu, VCPU_REGS_RSI);
5043 tss->di = kvm_register_read(vcpu, VCPU_REGS_RDI);
5045 tss->es = get_segment_selector(vcpu, VCPU_SREG_ES);
5046 tss->cs = get_segment_selector(vcpu, VCPU_SREG_CS);
5047 tss->ss = get_segment_selector(vcpu, VCPU_SREG_SS);
5048 tss->ds = get_segment_selector(vcpu, VCPU_SREG_DS);
5049 tss->ldt = get_segment_selector(vcpu, VCPU_SREG_LDTR);
5052 static int load_state_from_tss16(struct kvm_vcpu *vcpu,
5053 struct tss_segment_16 *tss)
5055 kvm_rip_write(vcpu, tss->ip);
5056 kvm_set_rflags(vcpu, tss->flag | 2);
5057 kvm_register_write(vcpu, VCPU_REGS_RAX, tss->ax);
5058 kvm_register_write(vcpu, VCPU_REGS_RCX, tss->cx);
5059 kvm_register_write(vcpu, VCPU_REGS_RDX, tss->dx);
5060 kvm_register_write(vcpu, VCPU_REGS_RBX, tss->bx);
5061 kvm_register_write(vcpu, VCPU_REGS_RSP, tss->sp);
5062 kvm_register_write(vcpu, VCPU_REGS_RBP, tss->bp);
5063 kvm_register_write(vcpu, VCPU_REGS_RSI, tss->si);
5064 kvm_register_write(vcpu, VCPU_REGS_RDI, tss->di);
5067 * SDM says that segment selectors are loaded before segment
5070 kvm_load_segment_selector(vcpu, tss->ldt, VCPU_SREG_LDTR);
5071 kvm_load_segment_selector(vcpu, tss->es, VCPU_SREG_ES);
5072 kvm_load_segment_selector(vcpu, tss->cs, VCPU_SREG_CS);
5073 kvm_load_segment_selector(vcpu, tss->ss, VCPU_SREG_SS);
5074 kvm_load_segment_selector(vcpu, tss->ds, VCPU_SREG_DS);
5077 * Now load segment descriptors. If fault happenes at this stage
5078 * it is handled in a context of new task
5080 if (kvm_load_segment_descriptor(vcpu, tss->ldt, VCPU_SREG_LDTR))
5083 if (kvm_load_segment_descriptor(vcpu, tss->es, VCPU_SREG_ES))
5086 if (kvm_load_segment_descriptor(vcpu, tss->cs, VCPU_SREG_CS))
5089 if (kvm_load_segment_descriptor(vcpu, tss->ss, VCPU_SREG_SS))
5092 if (kvm_load_segment_descriptor(vcpu, tss->ds, VCPU_SREG_DS))
5097 static int kvm_task_switch_16(struct kvm_vcpu *vcpu, u16 tss_selector,
5098 u16 old_tss_sel, u32 old_tss_base,
5099 struct desc_struct *nseg_desc)
5101 struct tss_segment_16 tss_segment_16;
5104 if (kvm_read_guest(vcpu->kvm, old_tss_base, &tss_segment_16,
5105 sizeof tss_segment_16))
5108 save_state_to_tss16(vcpu, &tss_segment_16);
5110 if (kvm_write_guest(vcpu->kvm, old_tss_base, &tss_segment_16,
5111 sizeof tss_segment_16))
5114 if (kvm_read_guest(vcpu->kvm, get_tss_base_addr_read(vcpu, nseg_desc),
5115 &tss_segment_16, sizeof tss_segment_16))
5118 if (old_tss_sel != 0xffff) {
5119 tss_segment_16.prev_task_link = old_tss_sel;
5121 if (kvm_write_guest(vcpu->kvm,
5122 get_tss_base_addr_write(vcpu, nseg_desc),
5123 &tss_segment_16.prev_task_link,
5124 sizeof tss_segment_16.prev_task_link))
5128 if (load_state_from_tss16(vcpu, &tss_segment_16))
5136 static int kvm_task_switch_32(struct kvm_vcpu *vcpu, u16 tss_selector,
5137 u16 old_tss_sel, u32 old_tss_base,
5138 struct desc_struct *nseg_desc)
5140 struct tss_segment_32 tss_segment_32;
5143 if (kvm_read_guest(vcpu->kvm, old_tss_base, &tss_segment_32,
5144 sizeof tss_segment_32))
5147 save_state_to_tss32(vcpu, &tss_segment_32);
5149 if (kvm_write_guest(vcpu->kvm, old_tss_base, &tss_segment_32,
5150 sizeof tss_segment_32))
5153 if (kvm_read_guest(vcpu->kvm, get_tss_base_addr_read(vcpu, nseg_desc),
5154 &tss_segment_32, sizeof tss_segment_32))
5157 if (old_tss_sel != 0xffff) {
5158 tss_segment_32.prev_task_link = old_tss_sel;
5160 if (kvm_write_guest(vcpu->kvm,
5161 get_tss_base_addr_write(vcpu, nseg_desc),
5162 &tss_segment_32.prev_task_link,
5163 sizeof tss_segment_32.prev_task_link))
5167 if (load_state_from_tss32(vcpu, &tss_segment_32))
5175 int kvm_task_switch(struct kvm_vcpu *vcpu, u16 tss_selector, int reason)
5177 struct kvm_segment tr_seg;
5178 struct desc_struct cseg_desc;
5179 struct desc_struct nseg_desc;
5181 u32 old_tss_base = get_segment_base(vcpu, VCPU_SREG_TR);
5182 u16 old_tss_sel = get_segment_selector(vcpu, VCPU_SREG_TR);
5185 old_tss_base = kvm_mmu_gva_to_gpa_write(vcpu, old_tss_base, NULL);
5187 /* FIXME: Handle errors. Failure to read either TSS or their
5188 * descriptors should generate a pagefault.
5190 if (load_guest_segment_descriptor(vcpu, tss_selector, &nseg_desc))
5193 if (load_guest_segment_descriptor(vcpu, old_tss_sel, &cseg_desc))
5196 if (reason != TASK_SWITCH_IRET) {
5199 cpl = kvm_x86_ops->get_cpl(vcpu);
5200 if ((tss_selector & 3) > nseg_desc.dpl || cpl > nseg_desc.dpl) {
5201 kvm_queue_exception_e(vcpu, GP_VECTOR, 0);
5206 desc_limit = get_desc_limit(&nseg_desc);
5208 ((desc_limit < 0x67 && (nseg_desc.type & 8)) ||
5209 desc_limit < 0x2b)) {
5210 kvm_queue_exception_e(vcpu, TS_VECTOR, tss_selector & 0xfffc);
5214 if (reason == TASK_SWITCH_IRET || reason == TASK_SWITCH_JMP) {
5215 cseg_desc.type &= ~(1 << 1); //clear the B flag
5216 save_guest_segment_descriptor(vcpu, old_tss_sel, &cseg_desc);
5219 if (reason == TASK_SWITCH_IRET) {
5220 u32 eflags = kvm_get_rflags(vcpu);
5221 kvm_set_rflags(vcpu, eflags & ~X86_EFLAGS_NT);
5224 /* set back link to prev task only if NT bit is set in eflags
5225 note that old_tss_sel is not used afetr this point */
5226 if (reason != TASK_SWITCH_CALL && reason != TASK_SWITCH_GATE)
5227 old_tss_sel = 0xffff;
5229 if (nseg_desc.type & 8)
5230 ret = kvm_task_switch_32(vcpu, tss_selector, old_tss_sel,
5231 old_tss_base, &nseg_desc);
5233 ret = kvm_task_switch_16(vcpu, tss_selector, old_tss_sel,
5234 old_tss_base, &nseg_desc);
5236 if (reason == TASK_SWITCH_CALL || reason == TASK_SWITCH_GATE) {
5237 u32 eflags = kvm_get_rflags(vcpu);
5238 kvm_set_rflags(vcpu, eflags | X86_EFLAGS_NT);
5241 if (reason != TASK_SWITCH_IRET) {
5242 nseg_desc.type |= (1 << 1);
5243 save_guest_segment_descriptor(vcpu, tss_selector,
5247 kvm_x86_ops->set_cr0(vcpu, kvm_read_cr0(vcpu) | X86_CR0_TS);
5248 seg_desct_to_kvm_desct(&nseg_desc, tss_selector, &tr_seg);
5250 kvm_set_segment(vcpu, &tr_seg, VCPU_SREG_TR);
5254 EXPORT_SYMBOL_GPL(kvm_task_switch);
5256 int kvm_arch_vcpu_ioctl_set_sregs(struct kvm_vcpu *vcpu,
5257 struct kvm_sregs *sregs)
5259 int mmu_reset_needed = 0;
5260 int pending_vec, max_bits;
5261 struct descriptor_table dt;
5265 dt.limit = sregs->idt.limit;
5266 dt.base = sregs->idt.base;
5267 kvm_x86_ops->set_idt(vcpu, &dt);
5268 dt.limit = sregs->gdt.limit;
5269 dt.base = sregs->gdt.base;
5270 kvm_x86_ops->set_gdt(vcpu, &dt);
5272 vcpu->arch.cr2 = sregs->cr2;
5273 mmu_reset_needed |= vcpu->arch.cr3 != sregs->cr3;
5274 vcpu->arch.cr3 = sregs->cr3;
5276 kvm_set_cr8(vcpu, sregs->cr8);
5278 mmu_reset_needed |= vcpu->arch.efer != sregs->efer;
5279 kvm_x86_ops->set_efer(vcpu, sregs->efer);
5280 kvm_set_apic_base(vcpu, sregs->apic_base);
5282 mmu_reset_needed |= kvm_read_cr0(vcpu) != sregs->cr0;
5283 kvm_x86_ops->set_cr0(vcpu, sregs->cr0);
5284 vcpu->arch.cr0 = sregs->cr0;
5286 mmu_reset_needed |= kvm_read_cr4(vcpu) != sregs->cr4;
5287 kvm_x86_ops->set_cr4(vcpu, sregs->cr4);
5288 if (!is_long_mode(vcpu) && is_pae(vcpu)) {
5289 load_pdptrs(vcpu, vcpu->arch.cr3);
5290 mmu_reset_needed = 1;
5293 if (mmu_reset_needed)
5294 kvm_mmu_reset_context(vcpu);
5296 max_bits = (sizeof sregs->interrupt_bitmap) << 3;
5297 pending_vec = find_first_bit(
5298 (const unsigned long *)sregs->interrupt_bitmap, max_bits);
5299 if (pending_vec < max_bits) {
5300 kvm_queue_interrupt(vcpu, pending_vec, false);
5301 pr_debug("Set back pending irq %d\n", pending_vec);
5302 if (irqchip_in_kernel(vcpu->kvm))
5303 kvm_pic_clear_isr_ack(vcpu->kvm);
5306 kvm_set_segment(vcpu, &sregs->cs, VCPU_SREG_CS);
5307 kvm_set_segment(vcpu, &sregs->ds, VCPU_SREG_DS);
5308 kvm_set_segment(vcpu, &sregs->es, VCPU_SREG_ES);
5309 kvm_set_segment(vcpu, &sregs->fs, VCPU_SREG_FS);
5310 kvm_set_segment(vcpu, &sregs->gs, VCPU_SREG_GS);
5311 kvm_set_segment(vcpu, &sregs->ss, VCPU_SREG_SS);
5313 kvm_set_segment(vcpu, &sregs->tr, VCPU_SREG_TR);
5314 kvm_set_segment(vcpu, &sregs->ldt, VCPU_SREG_LDTR);
5316 update_cr8_intercept(vcpu);
5318 /* Older userspace won't unhalt the vcpu on reset. */
5319 if (kvm_vcpu_is_bsp(vcpu) && kvm_rip_read(vcpu) == 0xfff0 &&
5320 sregs->cs.selector == 0xf000 && sregs->cs.base == 0xffff0000 &&
5322 vcpu->arch.mp_state = KVM_MP_STATE_RUNNABLE;
5329 int kvm_arch_vcpu_ioctl_set_guest_debug(struct kvm_vcpu *vcpu,
5330 struct kvm_guest_debug *dbg)
5332 unsigned long rflags;
5337 if (dbg->control & (KVM_GUESTDBG_INJECT_DB | KVM_GUESTDBG_INJECT_BP)) {
5339 if (vcpu->arch.exception.pending)
5341 if (dbg->control & KVM_GUESTDBG_INJECT_DB)
5342 kvm_queue_exception(vcpu, DB_VECTOR);
5344 kvm_queue_exception(vcpu, BP_VECTOR);
5348 * Read rflags as long as potentially injected trace flags are still
5351 rflags = kvm_get_rflags(vcpu);
5353 vcpu->guest_debug = dbg->control;
5354 if (!(vcpu->guest_debug & KVM_GUESTDBG_ENABLE))
5355 vcpu->guest_debug = 0;
5357 if (vcpu->guest_debug & KVM_GUESTDBG_USE_HW_BP) {
5358 for (i = 0; i < KVM_NR_DB_REGS; ++i)
5359 vcpu->arch.eff_db[i] = dbg->arch.debugreg[i];
5360 vcpu->arch.switch_db_regs =
5361 (dbg->arch.debugreg[7] & DR7_BP_EN_MASK);
5363 for (i = 0; i < KVM_NR_DB_REGS; i++)
5364 vcpu->arch.eff_db[i] = vcpu->arch.db[i];
5365 vcpu->arch.switch_db_regs = (vcpu->arch.dr7 & DR7_BP_EN_MASK);
5368 if (vcpu->guest_debug & KVM_GUESTDBG_SINGLESTEP) {
5369 vcpu->arch.singlestep_cs =
5370 get_segment_selector(vcpu, VCPU_SREG_CS);
5371 vcpu->arch.singlestep_rip = kvm_rip_read(vcpu);
5375 * Trigger an rflags update that will inject or remove the trace
5378 kvm_set_rflags(vcpu, rflags);
5380 kvm_x86_ops->set_guest_debug(vcpu, dbg);
5391 * fxsave fpu state. Taken from x86_64/processor.h. To be killed when
5392 * we have asm/x86/processor.h
5403 u32 st_space[32]; /* 8*16 bytes for each FP-reg = 128 bytes */
5404 #ifdef CONFIG_X86_64
5405 u32 xmm_space[64]; /* 16*16 bytes for each XMM-reg = 256 bytes */
5407 u32 xmm_space[32]; /* 8*16 bytes for each XMM-reg = 128 bytes */
5412 * Translate a guest virtual address to a guest physical address.
5414 int kvm_arch_vcpu_ioctl_translate(struct kvm_vcpu *vcpu,
5415 struct kvm_translation *tr)
5417 unsigned long vaddr = tr->linear_address;
5422 idx = srcu_read_lock(&vcpu->kvm->srcu);
5423 gpa = kvm_mmu_gva_to_gpa_system(vcpu, vaddr, NULL);
5424 srcu_read_unlock(&vcpu->kvm->srcu, idx);
5425 tr->physical_address = gpa;
5426 tr->valid = gpa != UNMAPPED_GVA;
5434 int kvm_arch_vcpu_ioctl_get_fpu(struct kvm_vcpu *vcpu, struct kvm_fpu *fpu)
5436 struct fxsave *fxsave = (struct fxsave *)&vcpu->arch.guest_fx_image;
5440 memcpy(fpu->fpr, fxsave->st_space, 128);
5441 fpu->fcw = fxsave->cwd;
5442 fpu->fsw = fxsave->swd;
5443 fpu->ftwx = fxsave->twd;
5444 fpu->last_opcode = fxsave->fop;
5445 fpu->last_ip = fxsave->rip;
5446 fpu->last_dp = fxsave->rdp;
5447 memcpy(fpu->xmm, fxsave->xmm_space, sizeof fxsave->xmm_space);
5454 int kvm_arch_vcpu_ioctl_set_fpu(struct kvm_vcpu *vcpu, struct kvm_fpu *fpu)
5456 struct fxsave *fxsave = (struct fxsave *)&vcpu->arch.guest_fx_image;
5460 memcpy(fxsave->st_space, fpu->fpr, 128);
5461 fxsave->cwd = fpu->fcw;
5462 fxsave->swd = fpu->fsw;
5463 fxsave->twd = fpu->ftwx;
5464 fxsave->fop = fpu->last_opcode;
5465 fxsave->rip = fpu->last_ip;
5466 fxsave->rdp = fpu->last_dp;
5467 memcpy(fxsave->xmm_space, fpu->xmm, sizeof fxsave->xmm_space);
5474 void fx_init(struct kvm_vcpu *vcpu)
5476 unsigned after_mxcsr_mask;
5479 * Touch the fpu the first time in non atomic context as if
5480 * this is the first fpu instruction the exception handler
5481 * will fire before the instruction returns and it'll have to
5482 * allocate ram with GFP_KERNEL.
5485 kvm_fx_save(&vcpu->arch.host_fx_image);
5487 /* Initialize guest FPU by resetting ours and saving into guest's */
5489 kvm_fx_save(&vcpu->arch.host_fx_image);
5491 kvm_fx_save(&vcpu->arch.guest_fx_image);
5492 kvm_fx_restore(&vcpu->arch.host_fx_image);
5495 vcpu->arch.cr0 |= X86_CR0_ET;
5496 after_mxcsr_mask = offsetof(struct i387_fxsave_struct, st_space);
5497 vcpu->arch.guest_fx_image.mxcsr = 0x1f80;
5498 memset((void *)&vcpu->arch.guest_fx_image + after_mxcsr_mask,
5499 0, sizeof(struct i387_fxsave_struct) - after_mxcsr_mask);
5501 EXPORT_SYMBOL_GPL(fx_init);
5503 void kvm_load_guest_fpu(struct kvm_vcpu *vcpu)
5505 if (vcpu->guest_fpu_loaded)
5508 vcpu->guest_fpu_loaded = 1;
5509 kvm_fx_save(&vcpu->arch.host_fx_image);
5510 kvm_fx_restore(&vcpu->arch.guest_fx_image);
5514 void kvm_put_guest_fpu(struct kvm_vcpu *vcpu)
5516 if (!vcpu->guest_fpu_loaded)
5519 vcpu->guest_fpu_loaded = 0;
5520 kvm_fx_save(&vcpu->arch.guest_fx_image);
5521 kvm_fx_restore(&vcpu->arch.host_fx_image);
5522 ++vcpu->stat.fpu_reload;
5523 set_bit(KVM_REQ_DEACTIVATE_FPU, &vcpu->requests);
5527 void kvm_arch_vcpu_free(struct kvm_vcpu *vcpu)
5529 if (vcpu->arch.time_page) {
5530 kvm_release_page_dirty(vcpu->arch.time_page);
5531 vcpu->arch.time_page = NULL;
5534 kvm_x86_ops->vcpu_free(vcpu);
5537 struct kvm_vcpu *kvm_arch_vcpu_create(struct kvm *kvm,
5540 return kvm_x86_ops->vcpu_create(kvm, id);
5543 int kvm_arch_vcpu_setup(struct kvm_vcpu *vcpu)
5547 /* We do fxsave: this must be aligned. */
5548 BUG_ON((unsigned long)&vcpu->arch.host_fx_image & 0xF);
5550 vcpu->arch.mtrr_state.have_fixed = 1;
5552 r = kvm_arch_vcpu_reset(vcpu);
5554 r = kvm_mmu_setup(vcpu);
5561 kvm_x86_ops->vcpu_free(vcpu);
5565 void kvm_arch_vcpu_destroy(struct kvm_vcpu *vcpu)
5568 kvm_mmu_unload(vcpu);
5571 kvm_x86_ops->vcpu_free(vcpu);
5574 int kvm_arch_vcpu_reset(struct kvm_vcpu *vcpu)
5576 vcpu->arch.nmi_pending = false;
5577 vcpu->arch.nmi_injected = false;
5579 vcpu->arch.switch_db_regs = 0;
5580 memset(vcpu->arch.db, 0, sizeof(vcpu->arch.db));
5581 vcpu->arch.dr6 = DR6_FIXED_1;
5582 vcpu->arch.dr7 = DR7_FIXED_1;
5584 return kvm_x86_ops->vcpu_reset(vcpu);
5587 int kvm_arch_hardware_enable(void *garbage)
5590 * Since this may be called from a hotplug notifcation,
5591 * we can't get the CPU frequency directly.
5593 if (!boot_cpu_has(X86_FEATURE_CONSTANT_TSC)) {
5594 int cpu = raw_smp_processor_id();
5595 per_cpu(cpu_tsc_khz, cpu) = 0;
5598 kvm_shared_msr_cpu_online();
5600 return kvm_x86_ops->hardware_enable(garbage);
5603 void kvm_arch_hardware_disable(void *garbage)
5605 kvm_x86_ops->hardware_disable(garbage);
5606 drop_user_return_notifiers(garbage);
5609 int kvm_arch_hardware_setup(void)
5611 return kvm_x86_ops->hardware_setup();
5614 void kvm_arch_hardware_unsetup(void)
5616 kvm_x86_ops->hardware_unsetup();
5619 void kvm_arch_check_processor_compat(void *rtn)
5621 kvm_x86_ops->check_processor_compatibility(rtn);
5624 int kvm_arch_vcpu_init(struct kvm_vcpu *vcpu)
5630 BUG_ON(vcpu->kvm == NULL);
5633 vcpu->arch.mmu.root_hpa = INVALID_PAGE;
5634 if (!irqchip_in_kernel(kvm) || kvm_vcpu_is_bsp(vcpu))
5635 vcpu->arch.mp_state = KVM_MP_STATE_RUNNABLE;
5637 vcpu->arch.mp_state = KVM_MP_STATE_UNINITIALIZED;
5639 page = alloc_page(GFP_KERNEL | __GFP_ZERO);
5644 vcpu->arch.pio_data = page_address(page);
5646 r = kvm_mmu_create(vcpu);
5648 goto fail_free_pio_data;
5650 if (irqchip_in_kernel(kvm)) {
5651 r = kvm_create_lapic(vcpu);
5653 goto fail_mmu_destroy;
5656 vcpu->arch.mce_banks = kzalloc(KVM_MAX_MCE_BANKS * sizeof(u64) * 4,
5658 if (!vcpu->arch.mce_banks) {
5660 goto fail_free_lapic;
5662 vcpu->arch.mcg_cap = KVM_MAX_MCE_BANKS;
5666 kvm_free_lapic(vcpu);
5668 kvm_mmu_destroy(vcpu);
5670 free_page((unsigned long)vcpu->arch.pio_data);
5675 void kvm_arch_vcpu_uninit(struct kvm_vcpu *vcpu)
5679 kfree(vcpu->arch.mce_banks);
5680 kvm_free_lapic(vcpu);
5681 idx = srcu_read_lock(&vcpu->kvm->srcu);
5682 kvm_mmu_destroy(vcpu);
5683 srcu_read_unlock(&vcpu->kvm->srcu, idx);
5684 free_page((unsigned long)vcpu->arch.pio_data);
5687 struct kvm *kvm_arch_create_vm(void)
5689 struct kvm *kvm = kzalloc(sizeof(struct kvm), GFP_KERNEL);
5692 return ERR_PTR(-ENOMEM);
5694 kvm->arch.aliases = kzalloc(sizeof(struct kvm_mem_aliases), GFP_KERNEL);
5695 if (!kvm->arch.aliases) {
5697 return ERR_PTR(-ENOMEM);
5700 INIT_LIST_HEAD(&kvm->arch.active_mmu_pages);
5701 INIT_LIST_HEAD(&kvm->arch.assigned_dev_head);
5703 /* Reserve bit 0 of irq_sources_bitmap for userspace irq source */
5704 set_bit(KVM_USERSPACE_IRQ_SOURCE_ID, &kvm->arch.irq_sources_bitmap);
5706 rdtscll(kvm->arch.vm_init_tsc);
5711 static void kvm_unload_vcpu_mmu(struct kvm_vcpu *vcpu)
5714 kvm_mmu_unload(vcpu);
5718 static void kvm_free_vcpus(struct kvm *kvm)
5721 struct kvm_vcpu *vcpu;
5724 * Unpin any mmu pages first.
5726 kvm_for_each_vcpu(i, vcpu, kvm)
5727 kvm_unload_vcpu_mmu(vcpu);
5728 kvm_for_each_vcpu(i, vcpu, kvm)
5729 kvm_arch_vcpu_free(vcpu);
5731 mutex_lock(&kvm->lock);
5732 for (i = 0; i < atomic_read(&kvm->online_vcpus); i++)
5733 kvm->vcpus[i] = NULL;
5735 atomic_set(&kvm->online_vcpus, 0);
5736 mutex_unlock(&kvm->lock);
5739 void kvm_arch_sync_events(struct kvm *kvm)
5741 kvm_free_all_assigned_devices(kvm);
5744 void kvm_arch_destroy_vm(struct kvm *kvm)
5746 kvm_iommu_unmap_guest(kvm);
5748 kfree(kvm->arch.vpic);
5749 kfree(kvm->arch.vioapic);
5750 kvm_free_vcpus(kvm);
5751 kvm_free_physmem(kvm);
5752 if (kvm->arch.apic_access_page)
5753 put_page(kvm->arch.apic_access_page);
5754 if (kvm->arch.ept_identity_pagetable)
5755 put_page(kvm->arch.ept_identity_pagetable);
5756 cleanup_srcu_struct(&kvm->srcu);
5757 kfree(kvm->arch.aliases);
5761 int kvm_arch_prepare_memory_region(struct kvm *kvm,
5762 struct kvm_memory_slot *memslot,
5763 struct kvm_memory_slot old,
5764 struct kvm_userspace_memory_region *mem,
5767 int npages = memslot->npages;
5769 /*To keep backward compatibility with older userspace,
5770 *x86 needs to hanlde !user_alloc case.
5773 if (npages && !old.rmap) {
5774 unsigned long userspace_addr;
5776 down_write(¤t->mm->mmap_sem);
5777 userspace_addr = do_mmap(NULL, 0,
5779 PROT_READ | PROT_WRITE,
5780 MAP_PRIVATE | MAP_ANONYMOUS,
5782 up_write(¤t->mm->mmap_sem);
5784 if (IS_ERR((void *)userspace_addr))
5785 return PTR_ERR((void *)userspace_addr);
5787 memslot->userspace_addr = userspace_addr;
5795 void kvm_arch_commit_memory_region(struct kvm *kvm,
5796 struct kvm_userspace_memory_region *mem,
5797 struct kvm_memory_slot old,
5801 int npages = mem->memory_size >> PAGE_SHIFT;
5803 if (!user_alloc && !old.user_alloc && old.rmap && !npages) {
5806 down_write(¤t->mm->mmap_sem);
5807 ret = do_munmap(current->mm, old.userspace_addr,
5808 old.npages * PAGE_SIZE);
5809 up_write(¤t->mm->mmap_sem);
5812 "kvm_vm_ioctl_set_memory_region: "
5813 "failed to munmap memory\n");
5816 spin_lock(&kvm->mmu_lock);
5817 if (!kvm->arch.n_requested_mmu_pages) {
5818 unsigned int nr_mmu_pages = kvm_mmu_calculate_mmu_pages(kvm);
5819 kvm_mmu_change_mmu_pages(kvm, nr_mmu_pages);
5822 kvm_mmu_slot_remove_write_access(kvm, mem->slot);
5823 spin_unlock(&kvm->mmu_lock);
5826 void kvm_arch_flush_shadow(struct kvm *kvm)
5828 kvm_mmu_zap_all(kvm);
5829 kvm_reload_remote_mmus(kvm);
5832 int kvm_arch_vcpu_runnable(struct kvm_vcpu *vcpu)
5834 return vcpu->arch.mp_state == KVM_MP_STATE_RUNNABLE
5835 || vcpu->arch.mp_state == KVM_MP_STATE_SIPI_RECEIVED
5836 || vcpu->arch.nmi_pending ||
5837 (kvm_arch_interrupt_allowed(vcpu) &&
5838 kvm_cpu_has_interrupt(vcpu));
5841 void kvm_vcpu_kick(struct kvm_vcpu *vcpu)
5844 int cpu = vcpu->cpu;
5846 if (waitqueue_active(&vcpu->wq)) {
5847 wake_up_interruptible(&vcpu->wq);
5848 ++vcpu->stat.halt_wakeup;
5852 if (cpu != me && (unsigned)cpu < nr_cpu_ids && cpu_online(cpu))
5853 if (!test_and_set_bit(KVM_REQ_KICK, &vcpu->requests))
5854 smp_send_reschedule(cpu);
5858 int kvm_arch_interrupt_allowed(struct kvm_vcpu *vcpu)
5860 return kvm_x86_ops->interrupt_allowed(vcpu);
5863 unsigned long kvm_get_rflags(struct kvm_vcpu *vcpu)
5865 unsigned long rflags;
5867 rflags = kvm_x86_ops->get_rflags(vcpu);
5868 if (vcpu->guest_debug & KVM_GUESTDBG_SINGLESTEP)
5869 rflags &= ~(unsigned long)(X86_EFLAGS_TF | X86_EFLAGS_RF);
5872 EXPORT_SYMBOL_GPL(kvm_get_rflags);
5874 void kvm_set_rflags(struct kvm_vcpu *vcpu, unsigned long rflags)
5876 if (vcpu->guest_debug & KVM_GUESTDBG_SINGLESTEP &&
5877 vcpu->arch.singlestep_cs ==
5878 get_segment_selector(vcpu, VCPU_SREG_CS) &&
5879 vcpu->arch.singlestep_rip == kvm_rip_read(vcpu))
5880 rflags |= X86_EFLAGS_TF | X86_EFLAGS_RF;
5881 kvm_x86_ops->set_rflags(vcpu, rflags);
5883 EXPORT_SYMBOL_GPL(kvm_set_rflags);
5885 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_exit);
5886 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_inj_virq);
5887 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_page_fault);
5888 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_msr);
5889 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_cr);
5890 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_nested_vmrun);
5891 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_nested_vmexit);
5892 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_nested_vmexit_inject);
5893 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_nested_intr_vmexit);
5894 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_invlpga);
5895 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_skinit);