2 * Kernel-based Virtual Machine driver for Linux
4 * derived from drivers/kvm/kvm_main.c
6 * Copyright (C) 2006 Qumranet, Inc.
7 * Copyright (C) 2008 Qumranet, Inc.
8 * Copyright IBM Corporation, 2008
11 * Avi Kivity <avi@qumranet.com>
12 * Yaniv Kamay <yaniv@qumranet.com>
13 * Amit Shah <amit.shah@qumranet.com>
14 * Ben-Ami Yassour <benami@il.ibm.com>
16 * This work is licensed under the terms of the GNU GPL, version 2. See
17 * the COPYING file in the top-level directory.
21 #include <linux/kvm_host.h>
26 #include "kvm_cache_regs.h"
29 #include <linux/clocksource.h>
30 #include <linux/interrupt.h>
31 #include <linux/kvm.h>
33 #include <linux/vmalloc.h>
34 #include <linux/module.h>
35 #include <linux/mman.h>
36 #include <linux/highmem.h>
37 #include <linux/iommu.h>
38 #include <linux/intel-iommu.h>
39 #include <linux/cpufreq.h>
41 #include <asm/uaccess.h>
47 #define MAX_IO_MSRS 256
48 #define CR0_RESERVED_BITS \
49 (~(unsigned long)(X86_CR0_PE | X86_CR0_MP | X86_CR0_EM | X86_CR0_TS \
50 | X86_CR0_ET | X86_CR0_NE | X86_CR0_WP | X86_CR0_AM \
51 | X86_CR0_NW | X86_CR0_CD | X86_CR0_PG))
52 #define CR4_RESERVED_BITS \
53 (~(unsigned long)(X86_CR4_VME | X86_CR4_PVI | X86_CR4_TSD | X86_CR4_DE\
54 | X86_CR4_PSE | X86_CR4_PAE | X86_CR4_MCE \
55 | X86_CR4_PGE | X86_CR4_PCE | X86_CR4_OSFXSR \
56 | X86_CR4_OSXMMEXCPT | X86_CR4_VMXE))
58 #define CR8_RESERVED_BITS (~(unsigned long)X86_CR8_TPR)
60 #define KVM_MAX_MCE_BANKS 32
61 #define KVM_MCE_CAP_SUPPORTED MCG_CTL_P
64 * - enable syscall per default because its emulated by KVM
65 * - enable LME and LMA per default on 64 bit KVM
68 static u64 __read_mostly efer_reserved_bits = 0xfffffffffffffafeULL;
70 static u64 __read_mostly efer_reserved_bits = 0xfffffffffffffffeULL;
73 #define VM_STAT(x) offsetof(struct kvm, stat.x), KVM_STAT_VM
74 #define VCPU_STAT(x) offsetof(struct kvm_vcpu, stat.x), KVM_STAT_VCPU
76 static int kvm_dev_ioctl_get_supported_cpuid(struct kvm_cpuid2 *cpuid,
77 struct kvm_cpuid_entry2 __user *entries);
78 struct kvm_cpuid_entry2 *kvm_find_cpuid_entry(struct kvm_vcpu *vcpu,
79 u32 function, u32 index);
81 struct kvm_x86_ops *kvm_x86_ops;
82 EXPORT_SYMBOL_GPL(kvm_x86_ops);
84 struct kvm_stats_debugfs_item debugfs_entries[] = {
85 { "pf_fixed", VCPU_STAT(pf_fixed) },
86 { "pf_guest", VCPU_STAT(pf_guest) },
87 { "tlb_flush", VCPU_STAT(tlb_flush) },
88 { "invlpg", VCPU_STAT(invlpg) },
89 { "exits", VCPU_STAT(exits) },
90 { "io_exits", VCPU_STAT(io_exits) },
91 { "mmio_exits", VCPU_STAT(mmio_exits) },
92 { "signal_exits", VCPU_STAT(signal_exits) },
93 { "irq_window", VCPU_STAT(irq_window_exits) },
94 { "nmi_window", VCPU_STAT(nmi_window_exits) },
95 { "halt_exits", VCPU_STAT(halt_exits) },
96 { "halt_wakeup", VCPU_STAT(halt_wakeup) },
97 { "hypercalls", VCPU_STAT(hypercalls) },
98 { "request_irq", VCPU_STAT(request_irq_exits) },
99 { "irq_exits", VCPU_STAT(irq_exits) },
100 { "host_state_reload", VCPU_STAT(host_state_reload) },
101 { "efer_reload", VCPU_STAT(efer_reload) },
102 { "fpu_reload", VCPU_STAT(fpu_reload) },
103 { "insn_emulation", VCPU_STAT(insn_emulation) },
104 { "insn_emulation_fail", VCPU_STAT(insn_emulation_fail) },
105 { "irq_injections", VCPU_STAT(irq_injections) },
106 { "nmi_injections", VCPU_STAT(nmi_injections) },
107 { "mmu_shadow_zapped", VM_STAT(mmu_shadow_zapped) },
108 { "mmu_pte_write", VM_STAT(mmu_pte_write) },
109 { "mmu_pte_updated", VM_STAT(mmu_pte_updated) },
110 { "mmu_pde_zapped", VM_STAT(mmu_pde_zapped) },
111 { "mmu_flooded", VM_STAT(mmu_flooded) },
112 { "mmu_recycled", VM_STAT(mmu_recycled) },
113 { "mmu_cache_miss", VM_STAT(mmu_cache_miss) },
114 { "mmu_unsync", VM_STAT(mmu_unsync) },
115 { "remote_tlb_flush", VM_STAT(remote_tlb_flush) },
116 { "largepages", VM_STAT(lpages) },
120 unsigned long segment_base(u16 selector)
122 struct descriptor_table gdt;
123 struct desc_struct *d;
124 unsigned long table_base;
130 asm("sgdt %0" : "=m"(gdt));
131 table_base = gdt.base;
133 if (selector & 4) { /* from ldt */
136 asm("sldt %0" : "=g"(ldt_selector));
137 table_base = segment_base(ldt_selector);
139 d = (struct desc_struct *)(table_base + (selector & ~7));
140 v = d->base0 | ((unsigned long)d->base1 << 16) |
141 ((unsigned long)d->base2 << 24);
143 if (d->s == 0 && (d->type == 2 || d->type == 9 || d->type == 11))
144 v |= ((unsigned long)((struct ldttss_desc64 *)d)->base3) << 32;
148 EXPORT_SYMBOL_GPL(segment_base);
150 u64 kvm_get_apic_base(struct kvm_vcpu *vcpu)
152 if (irqchip_in_kernel(vcpu->kvm))
153 return vcpu->arch.apic_base;
155 return vcpu->arch.apic_base;
157 EXPORT_SYMBOL_GPL(kvm_get_apic_base);
159 void kvm_set_apic_base(struct kvm_vcpu *vcpu, u64 data)
161 /* TODO: reserve bits check */
162 if (irqchip_in_kernel(vcpu->kvm))
163 kvm_lapic_set_base(vcpu, data);
165 vcpu->arch.apic_base = data;
167 EXPORT_SYMBOL_GPL(kvm_set_apic_base);
169 void kvm_queue_exception(struct kvm_vcpu *vcpu, unsigned nr)
171 WARN_ON(vcpu->arch.exception.pending);
172 vcpu->arch.exception.pending = true;
173 vcpu->arch.exception.has_error_code = false;
174 vcpu->arch.exception.nr = nr;
176 EXPORT_SYMBOL_GPL(kvm_queue_exception);
178 void kvm_inject_page_fault(struct kvm_vcpu *vcpu, unsigned long addr,
181 ++vcpu->stat.pf_guest;
183 if (vcpu->arch.exception.pending) {
184 if (vcpu->arch.exception.nr == PF_VECTOR) {
185 printk(KERN_DEBUG "kvm: inject_page_fault:"
186 " double fault 0x%lx\n", addr);
187 vcpu->arch.exception.nr = DF_VECTOR;
188 vcpu->arch.exception.error_code = 0;
189 } else if (vcpu->arch.exception.nr == DF_VECTOR) {
190 /* triple fault -> shutdown */
191 set_bit(KVM_REQ_TRIPLE_FAULT, &vcpu->requests);
195 vcpu->arch.cr2 = addr;
196 kvm_queue_exception_e(vcpu, PF_VECTOR, error_code);
199 void kvm_inject_nmi(struct kvm_vcpu *vcpu)
201 vcpu->arch.nmi_pending = 1;
203 EXPORT_SYMBOL_GPL(kvm_inject_nmi);
205 void kvm_queue_exception_e(struct kvm_vcpu *vcpu, unsigned nr, u32 error_code)
207 WARN_ON(vcpu->arch.exception.pending);
208 vcpu->arch.exception.pending = true;
209 vcpu->arch.exception.has_error_code = true;
210 vcpu->arch.exception.nr = nr;
211 vcpu->arch.exception.error_code = error_code;
213 EXPORT_SYMBOL_GPL(kvm_queue_exception_e);
215 static void __queue_exception(struct kvm_vcpu *vcpu)
217 kvm_x86_ops->queue_exception(vcpu, vcpu->arch.exception.nr,
218 vcpu->arch.exception.has_error_code,
219 vcpu->arch.exception.error_code);
223 * Load the pae pdptrs. Return true is they are all valid.
225 int load_pdptrs(struct kvm_vcpu *vcpu, unsigned long cr3)
227 gfn_t pdpt_gfn = cr3 >> PAGE_SHIFT;
228 unsigned offset = ((cr3 & (PAGE_SIZE-1)) >> 5) << 2;
231 u64 pdpte[ARRAY_SIZE(vcpu->arch.pdptrs)];
233 ret = kvm_read_guest_page(vcpu->kvm, pdpt_gfn, pdpte,
234 offset * sizeof(u64), sizeof(pdpte));
239 for (i = 0; i < ARRAY_SIZE(pdpte); ++i) {
240 if (is_present_pte(pdpte[i]) &&
241 (pdpte[i] & vcpu->arch.mmu.rsvd_bits_mask[0][2])) {
248 memcpy(vcpu->arch.pdptrs, pdpte, sizeof(vcpu->arch.pdptrs));
253 EXPORT_SYMBOL_GPL(load_pdptrs);
255 static bool pdptrs_changed(struct kvm_vcpu *vcpu)
257 u64 pdpte[ARRAY_SIZE(vcpu->arch.pdptrs)];
261 if (is_long_mode(vcpu) || !is_pae(vcpu))
264 r = kvm_read_guest(vcpu->kvm, vcpu->arch.cr3 & ~31u, pdpte, sizeof(pdpte));
267 changed = memcmp(pdpte, vcpu->arch.pdptrs, sizeof(pdpte)) != 0;
273 void kvm_set_cr0(struct kvm_vcpu *vcpu, unsigned long cr0)
275 if (cr0 & CR0_RESERVED_BITS) {
276 printk(KERN_DEBUG "set_cr0: 0x%lx #GP, reserved bits 0x%lx\n",
277 cr0, vcpu->arch.cr0);
278 kvm_inject_gp(vcpu, 0);
282 if ((cr0 & X86_CR0_NW) && !(cr0 & X86_CR0_CD)) {
283 printk(KERN_DEBUG "set_cr0: #GP, CD == 0 && NW == 1\n");
284 kvm_inject_gp(vcpu, 0);
288 if ((cr0 & X86_CR0_PG) && !(cr0 & X86_CR0_PE)) {
289 printk(KERN_DEBUG "set_cr0: #GP, set PG flag "
290 "and a clear PE flag\n");
291 kvm_inject_gp(vcpu, 0);
295 if (!is_paging(vcpu) && (cr0 & X86_CR0_PG)) {
297 if ((vcpu->arch.shadow_efer & EFER_LME)) {
301 printk(KERN_DEBUG "set_cr0: #GP, start paging "
302 "in long mode while PAE is disabled\n");
303 kvm_inject_gp(vcpu, 0);
306 kvm_x86_ops->get_cs_db_l_bits(vcpu, &cs_db, &cs_l);
308 printk(KERN_DEBUG "set_cr0: #GP, start paging "
309 "in long mode while CS.L == 1\n");
310 kvm_inject_gp(vcpu, 0);
316 if (is_pae(vcpu) && !load_pdptrs(vcpu, vcpu->arch.cr3)) {
317 printk(KERN_DEBUG "set_cr0: #GP, pdptrs "
319 kvm_inject_gp(vcpu, 0);
325 kvm_x86_ops->set_cr0(vcpu, cr0);
326 vcpu->arch.cr0 = cr0;
328 kvm_mmu_reset_context(vcpu);
331 EXPORT_SYMBOL_GPL(kvm_set_cr0);
333 void kvm_lmsw(struct kvm_vcpu *vcpu, unsigned long msw)
335 kvm_set_cr0(vcpu, (vcpu->arch.cr0 & ~0x0ful) | (msw & 0x0f));
336 KVMTRACE_1D(LMSW, vcpu,
337 (u32)((vcpu->arch.cr0 & ~0x0ful) | (msw & 0x0f)),
340 EXPORT_SYMBOL_GPL(kvm_lmsw);
342 void kvm_set_cr4(struct kvm_vcpu *vcpu, unsigned long cr4)
344 unsigned long old_cr4 = vcpu->arch.cr4;
345 unsigned long pdptr_bits = X86_CR4_PGE | X86_CR4_PSE | X86_CR4_PAE;
347 if (cr4 & CR4_RESERVED_BITS) {
348 printk(KERN_DEBUG "set_cr4: #GP, reserved bits\n");
349 kvm_inject_gp(vcpu, 0);
353 if (is_long_mode(vcpu)) {
354 if (!(cr4 & X86_CR4_PAE)) {
355 printk(KERN_DEBUG "set_cr4: #GP, clearing PAE while "
357 kvm_inject_gp(vcpu, 0);
360 } else if (is_paging(vcpu) && (cr4 & X86_CR4_PAE)
361 && ((cr4 ^ old_cr4) & pdptr_bits)
362 && !load_pdptrs(vcpu, vcpu->arch.cr3)) {
363 printk(KERN_DEBUG "set_cr4: #GP, pdptrs reserved bits\n");
364 kvm_inject_gp(vcpu, 0);
368 if (cr4 & X86_CR4_VMXE) {
369 printk(KERN_DEBUG "set_cr4: #GP, setting VMXE\n");
370 kvm_inject_gp(vcpu, 0);
373 kvm_x86_ops->set_cr4(vcpu, cr4);
374 vcpu->arch.cr4 = cr4;
375 vcpu->arch.mmu.base_role.cr4_pge = (cr4 & X86_CR4_PGE) && !tdp_enabled;
376 kvm_mmu_reset_context(vcpu);
378 EXPORT_SYMBOL_GPL(kvm_set_cr4);
380 void kvm_set_cr3(struct kvm_vcpu *vcpu, unsigned long cr3)
382 if (cr3 == vcpu->arch.cr3 && !pdptrs_changed(vcpu)) {
383 kvm_mmu_sync_roots(vcpu);
384 kvm_mmu_flush_tlb(vcpu);
388 if (is_long_mode(vcpu)) {
389 if (cr3 & CR3_L_MODE_RESERVED_BITS) {
390 printk(KERN_DEBUG "set_cr3: #GP, reserved bits\n");
391 kvm_inject_gp(vcpu, 0);
396 if (cr3 & CR3_PAE_RESERVED_BITS) {
398 "set_cr3: #GP, reserved bits\n");
399 kvm_inject_gp(vcpu, 0);
402 if (is_paging(vcpu) && !load_pdptrs(vcpu, cr3)) {
403 printk(KERN_DEBUG "set_cr3: #GP, pdptrs "
405 kvm_inject_gp(vcpu, 0);
410 * We don't check reserved bits in nonpae mode, because
411 * this isn't enforced, and VMware depends on this.
416 * Does the new cr3 value map to physical memory? (Note, we
417 * catch an invalid cr3 even in real-mode, because it would
418 * cause trouble later on when we turn on paging anyway.)
420 * A real CPU would silently accept an invalid cr3 and would
421 * attempt to use it - with largely undefined (and often hard
422 * to debug) behavior on the guest side.
424 if (unlikely(!gfn_to_memslot(vcpu->kvm, cr3 >> PAGE_SHIFT)))
425 kvm_inject_gp(vcpu, 0);
427 vcpu->arch.cr3 = cr3;
428 vcpu->arch.mmu.new_cr3(vcpu);
431 EXPORT_SYMBOL_GPL(kvm_set_cr3);
433 void kvm_set_cr8(struct kvm_vcpu *vcpu, unsigned long cr8)
435 if (cr8 & CR8_RESERVED_BITS) {
436 printk(KERN_DEBUG "set_cr8: #GP, reserved bits 0x%lx\n", cr8);
437 kvm_inject_gp(vcpu, 0);
440 if (irqchip_in_kernel(vcpu->kvm))
441 kvm_lapic_set_tpr(vcpu, cr8);
443 vcpu->arch.cr8 = cr8;
445 EXPORT_SYMBOL_GPL(kvm_set_cr8);
447 unsigned long kvm_get_cr8(struct kvm_vcpu *vcpu)
449 if (irqchip_in_kernel(vcpu->kvm))
450 return kvm_lapic_get_cr8(vcpu);
452 return vcpu->arch.cr8;
454 EXPORT_SYMBOL_GPL(kvm_get_cr8);
456 static inline u32 bit(int bitno)
458 return 1 << (bitno & 31);
462 * List of msr numbers which we expose to userspace through KVM_GET_MSRS
463 * and KVM_SET_MSRS, and KVM_GET_MSR_INDEX_LIST.
465 * This list is modified at module load time to reflect the
466 * capabilities of the host cpu.
468 static u32 msrs_to_save[] = {
469 MSR_IA32_SYSENTER_CS, MSR_IA32_SYSENTER_ESP, MSR_IA32_SYSENTER_EIP,
472 MSR_CSTAR, MSR_KERNEL_GS_BASE, MSR_SYSCALL_MASK, MSR_LSTAR,
474 MSR_IA32_TSC, MSR_KVM_SYSTEM_TIME, MSR_KVM_WALL_CLOCK,
475 MSR_IA32_PERF_STATUS, MSR_IA32_CR_PAT, MSR_VM_HSAVE_PA
478 static unsigned num_msrs_to_save;
480 static u32 emulated_msrs[] = {
481 MSR_IA32_MISC_ENABLE,
484 static void set_efer(struct kvm_vcpu *vcpu, u64 efer)
486 if (efer & efer_reserved_bits) {
487 printk(KERN_DEBUG "set_efer: 0x%llx #GP, reserved bits\n",
489 kvm_inject_gp(vcpu, 0);
494 && (vcpu->arch.shadow_efer & EFER_LME) != (efer & EFER_LME)) {
495 printk(KERN_DEBUG "set_efer: #GP, change LME while paging\n");
496 kvm_inject_gp(vcpu, 0);
500 if (efer & EFER_FFXSR) {
501 struct kvm_cpuid_entry2 *feat;
503 feat = kvm_find_cpuid_entry(vcpu, 0x80000001, 0);
504 if (!feat || !(feat->edx & bit(X86_FEATURE_FXSR_OPT))) {
505 printk(KERN_DEBUG "set_efer: #GP, enable FFXSR w/o CPUID capability\n");
506 kvm_inject_gp(vcpu, 0);
511 if (efer & EFER_SVME) {
512 struct kvm_cpuid_entry2 *feat;
514 feat = kvm_find_cpuid_entry(vcpu, 0x80000001, 0);
515 if (!feat || !(feat->ecx & bit(X86_FEATURE_SVM))) {
516 printk(KERN_DEBUG "set_efer: #GP, enable SVM w/o SVM\n");
517 kvm_inject_gp(vcpu, 0);
522 kvm_x86_ops->set_efer(vcpu, efer);
525 efer |= vcpu->arch.shadow_efer & EFER_LMA;
527 vcpu->arch.shadow_efer = efer;
529 vcpu->arch.mmu.base_role.nxe = (efer & EFER_NX) && !tdp_enabled;
530 kvm_mmu_reset_context(vcpu);
533 void kvm_enable_efer_bits(u64 mask)
535 efer_reserved_bits &= ~mask;
537 EXPORT_SYMBOL_GPL(kvm_enable_efer_bits);
541 * Writes msr value into into the appropriate "register".
542 * Returns 0 on success, non-0 otherwise.
543 * Assumes vcpu_load() was already called.
545 int kvm_set_msr(struct kvm_vcpu *vcpu, u32 msr_index, u64 data)
547 return kvm_x86_ops->set_msr(vcpu, msr_index, data);
551 * Adapt set_msr() to msr_io()'s calling convention
553 static int do_set_msr(struct kvm_vcpu *vcpu, unsigned index, u64 *data)
555 return kvm_set_msr(vcpu, index, *data);
558 static void kvm_write_wall_clock(struct kvm *kvm, gpa_t wall_clock)
561 struct pvclock_wall_clock wc;
562 struct timespec now, sys, boot;
569 kvm_write_guest(kvm, wall_clock, &version, sizeof(version));
572 * The guest calculates current wall clock time by adding
573 * system time (updated by kvm_write_guest_time below) to the
574 * wall clock specified here. guest system time equals host
575 * system time for us, thus we must fill in host boot time here.
577 now = current_kernel_time();
579 boot = ns_to_timespec(timespec_to_ns(&now) - timespec_to_ns(&sys));
581 wc.sec = boot.tv_sec;
582 wc.nsec = boot.tv_nsec;
583 wc.version = version;
585 kvm_write_guest(kvm, wall_clock, &wc, sizeof(wc));
588 kvm_write_guest(kvm, wall_clock, &version, sizeof(version));
591 static uint32_t div_frac(uint32_t dividend, uint32_t divisor)
593 uint32_t quotient, remainder;
595 /* Don't try to replace with do_div(), this one calculates
596 * "(dividend << 32) / divisor" */
598 : "=a" (quotient), "=d" (remainder)
599 : "0" (0), "1" (dividend), "r" (divisor) );
603 static void kvm_set_time_scale(uint32_t tsc_khz, struct pvclock_vcpu_time_info *hv_clock)
605 uint64_t nsecs = 1000000000LL;
610 tps64 = tsc_khz * 1000LL;
611 while (tps64 > nsecs*2) {
616 tps32 = (uint32_t)tps64;
617 while (tps32 <= (uint32_t)nsecs) {
622 hv_clock->tsc_shift = shift;
623 hv_clock->tsc_to_system_mul = div_frac(nsecs, tps32);
625 pr_debug("%s: tsc_khz %u, tsc_shift %d, tsc_mul %u\n",
626 __func__, tsc_khz, hv_clock->tsc_shift,
627 hv_clock->tsc_to_system_mul);
630 static DEFINE_PER_CPU(unsigned long, cpu_tsc_khz);
632 static void kvm_write_guest_time(struct kvm_vcpu *v)
636 struct kvm_vcpu_arch *vcpu = &v->arch;
638 unsigned long this_tsc_khz;
640 if ((!vcpu->time_page))
643 this_tsc_khz = get_cpu_var(cpu_tsc_khz);
644 if (unlikely(vcpu->hv_clock_tsc_khz != this_tsc_khz)) {
645 kvm_set_time_scale(this_tsc_khz, &vcpu->hv_clock);
646 vcpu->hv_clock_tsc_khz = this_tsc_khz;
648 put_cpu_var(cpu_tsc_khz);
650 /* Keep irq disabled to prevent changes to the clock */
651 local_irq_save(flags);
652 kvm_get_msr(v, MSR_IA32_TSC, &vcpu->hv_clock.tsc_timestamp);
654 local_irq_restore(flags);
656 /* With all the info we got, fill in the values */
658 vcpu->hv_clock.system_time = ts.tv_nsec +
659 (NSEC_PER_SEC * (u64)ts.tv_sec);
661 * The interface expects us to write an even number signaling that the
662 * update is finished. Since the guest won't see the intermediate
663 * state, we just increase by 2 at the end.
665 vcpu->hv_clock.version += 2;
667 shared_kaddr = kmap_atomic(vcpu->time_page, KM_USER0);
669 memcpy(shared_kaddr + vcpu->time_offset, &vcpu->hv_clock,
670 sizeof(vcpu->hv_clock));
672 kunmap_atomic(shared_kaddr, KM_USER0);
674 mark_page_dirty(v->kvm, vcpu->time >> PAGE_SHIFT);
677 static int kvm_request_guest_time_update(struct kvm_vcpu *v)
679 struct kvm_vcpu_arch *vcpu = &v->arch;
681 if (!vcpu->time_page)
683 set_bit(KVM_REQ_KVMCLOCK_UPDATE, &v->requests);
687 static bool msr_mtrr_valid(unsigned msr)
690 case 0x200 ... 0x200 + 2 * KVM_NR_VAR_MTRR - 1:
691 case MSR_MTRRfix64K_00000:
692 case MSR_MTRRfix16K_80000:
693 case MSR_MTRRfix16K_A0000:
694 case MSR_MTRRfix4K_C0000:
695 case MSR_MTRRfix4K_C8000:
696 case MSR_MTRRfix4K_D0000:
697 case MSR_MTRRfix4K_D8000:
698 case MSR_MTRRfix4K_E0000:
699 case MSR_MTRRfix4K_E8000:
700 case MSR_MTRRfix4K_F0000:
701 case MSR_MTRRfix4K_F8000:
702 case MSR_MTRRdefType:
703 case MSR_IA32_CR_PAT:
711 static bool valid_pat_type(unsigned t)
713 return t < 8 && (1 << t) & 0xf3; /* 0, 1, 4, 5, 6, 7 */
716 static bool valid_mtrr_type(unsigned t)
718 return t < 8 && (1 << t) & 0x73; /* 0, 1, 4, 5, 6 */
721 static bool mtrr_valid(struct kvm_vcpu *vcpu, u32 msr, u64 data)
725 if (!msr_mtrr_valid(msr))
728 if (msr == MSR_IA32_CR_PAT) {
729 for (i = 0; i < 8; i++)
730 if (!valid_pat_type((data >> (i * 8)) & 0xff))
733 } else if (msr == MSR_MTRRdefType) {
736 return valid_mtrr_type(data & 0xff);
737 } else if (msr >= MSR_MTRRfix64K_00000 && msr <= MSR_MTRRfix4K_F8000) {
738 for (i = 0; i < 8 ; i++)
739 if (!valid_mtrr_type((data >> (i * 8)) & 0xff))
745 return valid_mtrr_type(data & 0xff);
748 static int set_msr_mtrr(struct kvm_vcpu *vcpu, u32 msr, u64 data)
750 u64 *p = (u64 *)&vcpu->arch.mtrr_state.fixed_ranges;
752 if (!mtrr_valid(vcpu, msr, data))
755 if (msr == MSR_MTRRdefType) {
756 vcpu->arch.mtrr_state.def_type = data;
757 vcpu->arch.mtrr_state.enabled = (data & 0xc00) >> 10;
758 } else if (msr == MSR_MTRRfix64K_00000)
760 else if (msr == MSR_MTRRfix16K_80000 || msr == MSR_MTRRfix16K_A0000)
761 p[1 + msr - MSR_MTRRfix16K_80000] = data;
762 else if (msr >= MSR_MTRRfix4K_C0000 && msr <= MSR_MTRRfix4K_F8000)
763 p[3 + msr - MSR_MTRRfix4K_C0000] = data;
764 else if (msr == MSR_IA32_CR_PAT)
765 vcpu->arch.pat = data;
766 else { /* Variable MTRRs */
767 int idx, is_mtrr_mask;
770 idx = (msr - 0x200) / 2;
771 is_mtrr_mask = msr - 0x200 - 2 * idx;
774 (u64 *)&vcpu->arch.mtrr_state.var_ranges[idx].base_lo;
777 (u64 *)&vcpu->arch.mtrr_state.var_ranges[idx].mask_lo;
781 kvm_mmu_reset_context(vcpu);
785 static int set_msr_mce(struct kvm_vcpu *vcpu, u32 msr, u64 data)
787 u64 mcg_cap = vcpu->arch.mcg_cap;
788 unsigned bank_num = mcg_cap & 0xff;
791 case MSR_IA32_MCG_STATUS:
792 vcpu->arch.mcg_status = data;
794 case MSR_IA32_MCG_CTL:
795 if (!(mcg_cap & MCG_CTL_P))
797 if (data != 0 && data != ~(u64)0)
799 vcpu->arch.mcg_ctl = data;
802 if (msr >= MSR_IA32_MC0_CTL &&
803 msr < MSR_IA32_MC0_CTL + 4 * bank_num) {
804 u32 offset = msr - MSR_IA32_MC0_CTL;
805 /* only 0 or all 1s can be written to IA32_MCi_CTL */
806 if ((offset & 0x3) == 0 &&
807 data != 0 && data != ~(u64)0)
809 vcpu->arch.mce_banks[offset] = data;
817 int kvm_set_msr_common(struct kvm_vcpu *vcpu, u32 msr, u64 data)
821 set_efer(vcpu, data);
823 case MSR_IA32_DEBUGCTLMSR:
825 /* We support the non-activated case already */
827 } else if (data & ~(DEBUGCTLMSR_LBR | DEBUGCTLMSR_BTF)) {
828 /* Values other than LBR and BTF are vendor-specific,
829 thus reserved and should throw a #GP */
832 pr_unimpl(vcpu, "%s: MSR_IA32_DEBUGCTLMSR 0x%llx, nop\n",
835 case MSR_IA32_UCODE_REV:
836 case MSR_IA32_UCODE_WRITE:
837 case MSR_VM_HSAVE_PA:
839 case 0x200 ... 0x2ff:
840 return set_msr_mtrr(vcpu, msr, data);
841 case MSR_IA32_APICBASE:
842 kvm_set_apic_base(vcpu, data);
844 case MSR_IA32_MISC_ENABLE:
845 vcpu->arch.ia32_misc_enable_msr = data;
847 case MSR_KVM_WALL_CLOCK:
848 vcpu->kvm->arch.wall_clock = data;
849 kvm_write_wall_clock(vcpu->kvm, data);
851 case MSR_KVM_SYSTEM_TIME: {
852 if (vcpu->arch.time_page) {
853 kvm_release_page_dirty(vcpu->arch.time_page);
854 vcpu->arch.time_page = NULL;
857 vcpu->arch.time = data;
859 /* we verify if the enable bit is set... */
863 /* ...but clean it before doing the actual write */
864 vcpu->arch.time_offset = data & ~(PAGE_MASK | 1);
866 vcpu->arch.time_page =
867 gfn_to_page(vcpu->kvm, data >> PAGE_SHIFT);
869 if (is_error_page(vcpu->arch.time_page)) {
870 kvm_release_page_clean(vcpu->arch.time_page);
871 vcpu->arch.time_page = NULL;
874 kvm_request_guest_time_update(vcpu);
877 case MSR_IA32_MCG_CTL:
878 case MSR_IA32_MCG_STATUS:
879 case MSR_IA32_MC0_CTL ... MSR_IA32_MC0_CTL + 4 * KVM_MAX_MCE_BANKS - 1:
880 return set_msr_mce(vcpu, msr, data);
882 pr_unimpl(vcpu, "unhandled wrmsr: 0x%x data %llx\n", msr, data);
887 EXPORT_SYMBOL_GPL(kvm_set_msr_common);
891 * Reads an msr value (of 'msr_index') into 'pdata'.
892 * Returns 0 on success, non-0 otherwise.
893 * Assumes vcpu_load() was already called.
895 int kvm_get_msr(struct kvm_vcpu *vcpu, u32 msr_index, u64 *pdata)
897 return kvm_x86_ops->get_msr(vcpu, msr_index, pdata);
900 static int get_msr_mtrr(struct kvm_vcpu *vcpu, u32 msr, u64 *pdata)
902 u64 *p = (u64 *)&vcpu->arch.mtrr_state.fixed_ranges;
904 if (!msr_mtrr_valid(msr))
907 if (msr == MSR_MTRRdefType)
908 *pdata = vcpu->arch.mtrr_state.def_type +
909 (vcpu->arch.mtrr_state.enabled << 10);
910 else if (msr == MSR_MTRRfix64K_00000)
912 else if (msr == MSR_MTRRfix16K_80000 || msr == MSR_MTRRfix16K_A0000)
913 *pdata = p[1 + msr - MSR_MTRRfix16K_80000];
914 else if (msr >= MSR_MTRRfix4K_C0000 && msr <= MSR_MTRRfix4K_F8000)
915 *pdata = p[3 + msr - MSR_MTRRfix4K_C0000];
916 else if (msr == MSR_IA32_CR_PAT)
917 *pdata = vcpu->arch.pat;
918 else { /* Variable MTRRs */
919 int idx, is_mtrr_mask;
922 idx = (msr - 0x200) / 2;
923 is_mtrr_mask = msr - 0x200 - 2 * idx;
926 (u64 *)&vcpu->arch.mtrr_state.var_ranges[idx].base_lo;
929 (u64 *)&vcpu->arch.mtrr_state.var_ranges[idx].mask_lo;
936 static int get_msr_mce(struct kvm_vcpu *vcpu, u32 msr, u64 *pdata)
939 u64 mcg_cap = vcpu->arch.mcg_cap;
940 unsigned bank_num = mcg_cap & 0xff;
943 case MSR_IA32_P5_MC_ADDR:
944 case MSR_IA32_P5_MC_TYPE:
947 case MSR_IA32_MCG_CAP:
948 data = vcpu->arch.mcg_cap;
950 case MSR_IA32_MCG_CTL:
951 if (!(mcg_cap & MCG_CTL_P))
953 data = vcpu->arch.mcg_ctl;
955 case MSR_IA32_MCG_STATUS:
956 data = vcpu->arch.mcg_status;
959 if (msr >= MSR_IA32_MC0_CTL &&
960 msr < MSR_IA32_MC0_CTL + 4 * bank_num) {
961 u32 offset = msr - MSR_IA32_MC0_CTL;
962 data = vcpu->arch.mce_banks[offset];
971 int kvm_get_msr_common(struct kvm_vcpu *vcpu, u32 msr, u64 *pdata)
976 case MSR_IA32_PLATFORM_ID:
977 case MSR_IA32_UCODE_REV:
978 case MSR_IA32_EBL_CR_POWERON:
979 case MSR_IA32_DEBUGCTLMSR:
980 case MSR_IA32_LASTBRANCHFROMIP:
981 case MSR_IA32_LASTBRANCHTOIP:
982 case MSR_IA32_LASTINTFROMIP:
983 case MSR_IA32_LASTINTTOIP:
986 case MSR_VM_HSAVE_PA:
987 case MSR_P6_EVNTSEL0:
988 case MSR_P6_EVNTSEL1:
989 case MSR_K7_EVNTSEL0:
993 data = 0x500 | KVM_NR_VAR_MTRR;
995 case 0x200 ... 0x2ff:
996 return get_msr_mtrr(vcpu, msr, pdata);
997 case 0xcd: /* fsb frequency */
1000 case MSR_IA32_APICBASE:
1001 data = kvm_get_apic_base(vcpu);
1003 case MSR_IA32_MISC_ENABLE:
1004 data = vcpu->arch.ia32_misc_enable_msr;
1006 case MSR_IA32_PERF_STATUS:
1007 /* TSC increment by tick */
1009 /* CPU multiplier */
1010 data |= (((uint64_t)4ULL) << 40);
1013 data = vcpu->arch.shadow_efer;
1015 case MSR_KVM_WALL_CLOCK:
1016 data = vcpu->kvm->arch.wall_clock;
1018 case MSR_KVM_SYSTEM_TIME:
1019 data = vcpu->arch.time;
1021 case MSR_IA32_P5_MC_ADDR:
1022 case MSR_IA32_P5_MC_TYPE:
1023 case MSR_IA32_MCG_CAP:
1024 case MSR_IA32_MCG_CTL:
1025 case MSR_IA32_MCG_STATUS:
1026 case MSR_IA32_MC0_CTL ... MSR_IA32_MC0_CTL + 4 * KVM_MAX_MCE_BANKS - 1:
1027 return get_msr_mce(vcpu, msr, pdata);
1029 pr_unimpl(vcpu, "unhandled rdmsr: 0x%x\n", msr);
1035 EXPORT_SYMBOL_GPL(kvm_get_msr_common);
1038 * Read or write a bunch of msrs. All parameters are kernel addresses.
1040 * @return number of msrs set successfully.
1042 static int __msr_io(struct kvm_vcpu *vcpu, struct kvm_msrs *msrs,
1043 struct kvm_msr_entry *entries,
1044 int (*do_msr)(struct kvm_vcpu *vcpu,
1045 unsigned index, u64 *data))
1051 down_read(&vcpu->kvm->slots_lock);
1052 for (i = 0; i < msrs->nmsrs; ++i)
1053 if (do_msr(vcpu, entries[i].index, &entries[i].data))
1055 up_read(&vcpu->kvm->slots_lock);
1063 * Read or write a bunch of msrs. Parameters are user addresses.
1065 * @return number of msrs set successfully.
1067 static int msr_io(struct kvm_vcpu *vcpu, struct kvm_msrs __user *user_msrs,
1068 int (*do_msr)(struct kvm_vcpu *vcpu,
1069 unsigned index, u64 *data),
1072 struct kvm_msrs msrs;
1073 struct kvm_msr_entry *entries;
1078 if (copy_from_user(&msrs, user_msrs, sizeof msrs))
1082 if (msrs.nmsrs >= MAX_IO_MSRS)
1086 size = sizeof(struct kvm_msr_entry) * msrs.nmsrs;
1087 entries = vmalloc(size);
1092 if (copy_from_user(entries, user_msrs->entries, size))
1095 r = n = __msr_io(vcpu, &msrs, entries, do_msr);
1100 if (writeback && copy_to_user(user_msrs->entries, entries, size))
1111 int kvm_dev_ioctl_check_extension(long ext)
1116 case KVM_CAP_IRQCHIP:
1118 case KVM_CAP_MMU_SHADOW_CACHE_CONTROL:
1119 case KVM_CAP_SET_TSS_ADDR:
1120 case KVM_CAP_EXT_CPUID:
1121 case KVM_CAP_CLOCKSOURCE:
1123 case KVM_CAP_NOP_IO_DELAY:
1124 case KVM_CAP_MP_STATE:
1125 case KVM_CAP_SYNC_MMU:
1126 case KVM_CAP_REINJECT_CONTROL:
1127 case KVM_CAP_IRQ_INJECT_STATUS:
1128 case KVM_CAP_ASSIGN_DEV_IRQ:
1131 case KVM_CAP_COALESCED_MMIO:
1132 r = KVM_COALESCED_MMIO_PAGE_OFFSET;
1135 r = !kvm_x86_ops->cpu_has_accelerated_tpr();
1137 case KVM_CAP_NR_VCPUS:
1140 case KVM_CAP_NR_MEMSLOTS:
1141 r = KVM_MEMORY_SLOTS;
1143 case KVM_CAP_PV_MMU:
1150 r = KVM_MAX_MCE_BANKS;
1160 long kvm_arch_dev_ioctl(struct file *filp,
1161 unsigned int ioctl, unsigned long arg)
1163 void __user *argp = (void __user *)arg;
1167 case KVM_GET_MSR_INDEX_LIST: {
1168 struct kvm_msr_list __user *user_msr_list = argp;
1169 struct kvm_msr_list msr_list;
1173 if (copy_from_user(&msr_list, user_msr_list, sizeof msr_list))
1176 msr_list.nmsrs = num_msrs_to_save + ARRAY_SIZE(emulated_msrs);
1177 if (copy_to_user(user_msr_list, &msr_list, sizeof msr_list))
1180 if (n < msr_list.nmsrs)
1183 if (copy_to_user(user_msr_list->indices, &msrs_to_save,
1184 num_msrs_to_save * sizeof(u32)))
1186 if (copy_to_user(user_msr_list->indices + num_msrs_to_save,
1188 ARRAY_SIZE(emulated_msrs) * sizeof(u32)))
1193 case KVM_GET_SUPPORTED_CPUID: {
1194 struct kvm_cpuid2 __user *cpuid_arg = argp;
1195 struct kvm_cpuid2 cpuid;
1198 if (copy_from_user(&cpuid, cpuid_arg, sizeof cpuid))
1200 r = kvm_dev_ioctl_get_supported_cpuid(&cpuid,
1201 cpuid_arg->entries);
1206 if (copy_to_user(cpuid_arg, &cpuid, sizeof cpuid))
1211 case KVM_X86_GET_MCE_CAP_SUPPORTED: {
1214 mce_cap = KVM_MCE_CAP_SUPPORTED;
1216 if (copy_to_user(argp, &mce_cap, sizeof mce_cap))
1228 void kvm_arch_vcpu_load(struct kvm_vcpu *vcpu, int cpu)
1230 kvm_x86_ops->vcpu_load(vcpu, cpu);
1231 kvm_request_guest_time_update(vcpu);
1234 void kvm_arch_vcpu_put(struct kvm_vcpu *vcpu)
1236 kvm_x86_ops->vcpu_put(vcpu);
1237 kvm_put_guest_fpu(vcpu);
1240 static int is_efer_nx(void)
1242 unsigned long long efer = 0;
1244 rdmsrl_safe(MSR_EFER, &efer);
1245 return efer & EFER_NX;
1248 static void cpuid_fix_nx_cap(struct kvm_vcpu *vcpu)
1251 struct kvm_cpuid_entry2 *e, *entry;
1254 for (i = 0; i < vcpu->arch.cpuid_nent; ++i) {
1255 e = &vcpu->arch.cpuid_entries[i];
1256 if (e->function == 0x80000001) {
1261 if (entry && (entry->edx & (1 << 20)) && !is_efer_nx()) {
1262 entry->edx &= ~(1 << 20);
1263 printk(KERN_INFO "kvm: guest NX capability removed\n");
1267 /* when an old userspace process fills a new kernel module */
1268 static int kvm_vcpu_ioctl_set_cpuid(struct kvm_vcpu *vcpu,
1269 struct kvm_cpuid *cpuid,
1270 struct kvm_cpuid_entry __user *entries)
1273 struct kvm_cpuid_entry *cpuid_entries;
1276 if (cpuid->nent > KVM_MAX_CPUID_ENTRIES)
1279 cpuid_entries = vmalloc(sizeof(struct kvm_cpuid_entry) * cpuid->nent);
1283 if (copy_from_user(cpuid_entries, entries,
1284 cpuid->nent * sizeof(struct kvm_cpuid_entry)))
1286 for (i = 0; i < cpuid->nent; i++) {
1287 vcpu->arch.cpuid_entries[i].function = cpuid_entries[i].function;
1288 vcpu->arch.cpuid_entries[i].eax = cpuid_entries[i].eax;
1289 vcpu->arch.cpuid_entries[i].ebx = cpuid_entries[i].ebx;
1290 vcpu->arch.cpuid_entries[i].ecx = cpuid_entries[i].ecx;
1291 vcpu->arch.cpuid_entries[i].edx = cpuid_entries[i].edx;
1292 vcpu->arch.cpuid_entries[i].index = 0;
1293 vcpu->arch.cpuid_entries[i].flags = 0;
1294 vcpu->arch.cpuid_entries[i].padding[0] = 0;
1295 vcpu->arch.cpuid_entries[i].padding[1] = 0;
1296 vcpu->arch.cpuid_entries[i].padding[2] = 0;
1298 vcpu->arch.cpuid_nent = cpuid->nent;
1299 cpuid_fix_nx_cap(vcpu);
1303 vfree(cpuid_entries);
1308 static int kvm_vcpu_ioctl_set_cpuid2(struct kvm_vcpu *vcpu,
1309 struct kvm_cpuid2 *cpuid,
1310 struct kvm_cpuid_entry2 __user *entries)
1315 if (cpuid->nent > KVM_MAX_CPUID_ENTRIES)
1318 if (copy_from_user(&vcpu->arch.cpuid_entries, entries,
1319 cpuid->nent * sizeof(struct kvm_cpuid_entry2)))
1321 vcpu->arch.cpuid_nent = cpuid->nent;
1328 static int kvm_vcpu_ioctl_get_cpuid2(struct kvm_vcpu *vcpu,
1329 struct kvm_cpuid2 *cpuid,
1330 struct kvm_cpuid_entry2 __user *entries)
1335 if (cpuid->nent < vcpu->arch.cpuid_nent)
1338 if (copy_to_user(entries, &vcpu->arch.cpuid_entries,
1339 vcpu->arch.cpuid_nent * sizeof(struct kvm_cpuid_entry2)))
1344 cpuid->nent = vcpu->arch.cpuid_nent;
1348 static void do_cpuid_1_ent(struct kvm_cpuid_entry2 *entry, u32 function,
1351 entry->function = function;
1352 entry->index = index;
1353 cpuid_count(entry->function, entry->index,
1354 &entry->eax, &entry->ebx, &entry->ecx, &entry->edx);
1358 #define F(x) bit(X86_FEATURE_##x)
1360 static void do_cpuid_ent(struct kvm_cpuid_entry2 *entry, u32 function,
1361 u32 index, int *nent, int maxnent)
1363 unsigned f_nx = is_efer_nx() ? F(NX) : 0;
1364 #ifdef CONFIG_X86_64
1365 unsigned f_lm = F(LM);
1371 const u32 kvm_supported_word0_x86_features =
1372 F(FPU) | F(VME) | F(DE) | F(PSE) |
1373 F(TSC) | F(MSR) | F(PAE) | F(MCE) |
1374 F(CX8) | F(APIC) | 0 /* Reserved */ | F(SEP) |
1375 F(MTRR) | F(PGE) | F(MCA) | F(CMOV) |
1376 F(PAT) | F(PSE36) | 0 /* PSN */ | F(CLFLSH) |
1377 0 /* Reserved, DS, ACPI */ | F(MMX) |
1378 F(FXSR) | F(XMM) | F(XMM2) | F(SELFSNOOP) |
1379 0 /* HTT, TM, Reserved, PBE */;
1380 /* cpuid 0x80000001.edx */
1381 const u32 kvm_supported_word1_x86_features =
1382 F(FPU) | F(VME) | F(DE) | F(PSE) |
1383 F(TSC) | F(MSR) | F(PAE) | F(MCE) |
1384 F(CX8) | F(APIC) | 0 /* Reserved */ | F(SYSCALL) |
1385 F(MTRR) | F(PGE) | F(MCA) | F(CMOV) |
1386 F(PAT) | F(PSE36) | 0 /* Reserved */ |
1387 f_nx | 0 /* Reserved */ | F(MMXEXT) | F(MMX) |
1388 F(FXSR) | F(FXSR_OPT) | 0 /* GBPAGES */ | 0 /* RDTSCP */ |
1389 0 /* Reserved */ | f_lm | F(3DNOWEXT) | F(3DNOW);
1391 const u32 kvm_supported_word4_x86_features =
1392 F(XMM3) | 0 /* Reserved, DTES64, MONITOR */ |
1393 0 /* DS-CPL, VMX, SMX, EST */ |
1394 0 /* TM2 */ | F(SSSE3) | 0 /* CNXT-ID */ | 0 /* Reserved */ |
1395 0 /* Reserved */ | F(CX16) | 0 /* xTPR Update, PDCM */ |
1396 0 /* Reserved, DCA */ | F(XMM4_1) |
1397 F(XMM4_2) | 0 /* x2APIC */ | F(MOVBE) | F(POPCNT) |
1398 0 /* Reserved, XSAVE, OSXSAVE */;
1399 /* cpuid 0x80000001.ecx */
1400 const u32 kvm_supported_word6_x86_features =
1401 F(LAHF_LM) | F(CMP_LEGACY) | F(SVM) | 0 /* ExtApicSpace */ |
1402 F(CR8_LEGACY) | F(ABM) | F(SSE4A) | F(MISALIGNSSE) |
1403 F(3DNOWPREFETCH) | 0 /* OSVW */ | 0 /* IBS */ | F(SSE5) |
1404 0 /* SKINIT */ | 0 /* WDT */;
1406 /* all calls to cpuid_count() should be made on the same cpu */
1408 do_cpuid_1_ent(entry, function, index);
1413 entry->eax = min(entry->eax, (u32)0xb);
1416 entry->edx &= kvm_supported_word0_x86_features;
1417 entry->ecx &= kvm_supported_word4_x86_features;
1419 /* function 2 entries are STATEFUL. That is, repeated cpuid commands
1420 * may return different values. This forces us to get_cpu() before
1421 * issuing the first command, and also to emulate this annoying behavior
1422 * in kvm_emulate_cpuid() using KVM_CPUID_FLAG_STATE_READ_NEXT */
1424 int t, times = entry->eax & 0xff;
1426 entry->flags |= KVM_CPUID_FLAG_STATEFUL_FUNC;
1427 entry->flags |= KVM_CPUID_FLAG_STATE_READ_NEXT;
1428 for (t = 1; t < times && *nent < maxnent; ++t) {
1429 do_cpuid_1_ent(&entry[t], function, 0);
1430 entry[t].flags |= KVM_CPUID_FLAG_STATEFUL_FUNC;
1435 /* function 4 and 0xb have additional index. */
1439 entry->flags |= KVM_CPUID_FLAG_SIGNIFCANT_INDEX;
1440 /* read more entries until cache_type is zero */
1441 for (i = 1; *nent < maxnent; ++i) {
1442 cache_type = entry[i - 1].eax & 0x1f;
1445 do_cpuid_1_ent(&entry[i], function, i);
1447 KVM_CPUID_FLAG_SIGNIFCANT_INDEX;
1455 entry->flags |= KVM_CPUID_FLAG_SIGNIFCANT_INDEX;
1456 /* read more entries until level_type is zero */
1457 for (i = 1; *nent < maxnent; ++i) {
1458 level_type = entry[i - 1].ecx & 0xff00;
1461 do_cpuid_1_ent(&entry[i], function, i);
1463 KVM_CPUID_FLAG_SIGNIFCANT_INDEX;
1469 entry->eax = min(entry->eax, 0x8000001a);
1472 entry->edx &= kvm_supported_word1_x86_features;
1473 entry->ecx &= kvm_supported_word6_x86_features;
1481 static int kvm_dev_ioctl_get_supported_cpuid(struct kvm_cpuid2 *cpuid,
1482 struct kvm_cpuid_entry2 __user *entries)
1484 struct kvm_cpuid_entry2 *cpuid_entries;
1485 int limit, nent = 0, r = -E2BIG;
1488 if (cpuid->nent < 1)
1491 cpuid_entries = vmalloc(sizeof(struct kvm_cpuid_entry2) * cpuid->nent);
1495 do_cpuid_ent(&cpuid_entries[0], 0, 0, &nent, cpuid->nent);
1496 limit = cpuid_entries[0].eax;
1497 for (func = 1; func <= limit && nent < cpuid->nent; ++func)
1498 do_cpuid_ent(&cpuid_entries[nent], func, 0,
1499 &nent, cpuid->nent);
1501 if (nent >= cpuid->nent)
1504 do_cpuid_ent(&cpuid_entries[nent], 0x80000000, 0, &nent, cpuid->nent);
1505 limit = cpuid_entries[nent - 1].eax;
1506 for (func = 0x80000001; func <= limit && nent < cpuid->nent; ++func)
1507 do_cpuid_ent(&cpuid_entries[nent], func, 0,
1508 &nent, cpuid->nent);
1510 if (copy_to_user(entries, cpuid_entries,
1511 nent * sizeof(struct kvm_cpuid_entry2)))
1517 vfree(cpuid_entries);
1522 static int kvm_vcpu_ioctl_get_lapic(struct kvm_vcpu *vcpu,
1523 struct kvm_lapic_state *s)
1526 memcpy(s->regs, vcpu->arch.apic->regs, sizeof *s);
1532 static int kvm_vcpu_ioctl_set_lapic(struct kvm_vcpu *vcpu,
1533 struct kvm_lapic_state *s)
1536 memcpy(vcpu->arch.apic->regs, s->regs, sizeof *s);
1537 kvm_apic_post_state_restore(vcpu);
1543 static int kvm_vcpu_ioctl_interrupt(struct kvm_vcpu *vcpu,
1544 struct kvm_interrupt *irq)
1546 if (irq->irq < 0 || irq->irq >= 256)
1548 if (irqchip_in_kernel(vcpu->kvm))
1552 kvm_queue_interrupt(vcpu, irq->irq, false);
1559 static int kvm_vcpu_ioctl_nmi(struct kvm_vcpu *vcpu)
1562 kvm_inject_nmi(vcpu);
1568 static int vcpu_ioctl_tpr_access_reporting(struct kvm_vcpu *vcpu,
1569 struct kvm_tpr_access_ctl *tac)
1573 vcpu->arch.tpr_access_reporting = !!tac->enabled;
1577 static int kvm_vcpu_ioctl_x86_setup_mce(struct kvm_vcpu *vcpu,
1581 unsigned bank_num = mcg_cap & 0xff, bank;
1586 if (mcg_cap & ~(KVM_MCE_CAP_SUPPORTED | 0xff | 0xff0000))
1589 vcpu->arch.mcg_cap = mcg_cap;
1590 /* Init IA32_MCG_CTL to all 1s */
1591 if (mcg_cap & MCG_CTL_P)
1592 vcpu->arch.mcg_ctl = ~(u64)0;
1593 /* Init IA32_MCi_CTL to all 1s */
1594 for (bank = 0; bank < bank_num; bank++)
1595 vcpu->arch.mce_banks[bank*4] = ~(u64)0;
1600 static int kvm_vcpu_ioctl_x86_set_mce(struct kvm_vcpu *vcpu,
1601 struct kvm_x86_mce *mce)
1603 u64 mcg_cap = vcpu->arch.mcg_cap;
1604 unsigned bank_num = mcg_cap & 0xff;
1605 u64 *banks = vcpu->arch.mce_banks;
1607 if (mce->bank >= bank_num || !(mce->status & MCI_STATUS_VAL))
1610 * if IA32_MCG_CTL is not all 1s, the uncorrected error
1611 * reporting is disabled
1613 if ((mce->status & MCI_STATUS_UC) && (mcg_cap & MCG_CTL_P) &&
1614 vcpu->arch.mcg_ctl != ~(u64)0)
1616 banks += 4 * mce->bank;
1618 * if IA32_MCi_CTL is not all 1s, the uncorrected error
1619 * reporting is disabled for the bank
1621 if ((mce->status & MCI_STATUS_UC) && banks[0] != ~(u64)0)
1623 if (mce->status & MCI_STATUS_UC) {
1624 if ((vcpu->arch.mcg_status & MCG_STATUS_MCIP) ||
1625 !(vcpu->arch.cr4 & X86_CR4_MCE)) {
1626 printk(KERN_DEBUG "kvm: set_mce: "
1627 "injects mce exception while "
1628 "previous one is in progress!\n");
1629 set_bit(KVM_REQ_TRIPLE_FAULT, &vcpu->requests);
1632 if (banks[1] & MCI_STATUS_VAL)
1633 mce->status |= MCI_STATUS_OVER;
1634 banks[2] = mce->addr;
1635 banks[3] = mce->misc;
1636 vcpu->arch.mcg_status = mce->mcg_status;
1637 banks[1] = mce->status;
1638 kvm_queue_exception(vcpu, MC_VECTOR);
1639 } else if (!(banks[1] & MCI_STATUS_VAL)
1640 || !(banks[1] & MCI_STATUS_UC)) {
1641 if (banks[1] & MCI_STATUS_VAL)
1642 mce->status |= MCI_STATUS_OVER;
1643 banks[2] = mce->addr;
1644 banks[3] = mce->misc;
1645 banks[1] = mce->status;
1647 banks[1] |= MCI_STATUS_OVER;
1651 long kvm_arch_vcpu_ioctl(struct file *filp,
1652 unsigned int ioctl, unsigned long arg)
1654 struct kvm_vcpu *vcpu = filp->private_data;
1655 void __user *argp = (void __user *)arg;
1657 struct kvm_lapic_state *lapic = NULL;
1660 case KVM_GET_LAPIC: {
1661 lapic = kzalloc(sizeof(struct kvm_lapic_state), GFP_KERNEL);
1666 r = kvm_vcpu_ioctl_get_lapic(vcpu, lapic);
1670 if (copy_to_user(argp, lapic, sizeof(struct kvm_lapic_state)))
1675 case KVM_SET_LAPIC: {
1676 lapic = kmalloc(sizeof(struct kvm_lapic_state), GFP_KERNEL);
1681 if (copy_from_user(lapic, argp, sizeof(struct kvm_lapic_state)))
1683 r = kvm_vcpu_ioctl_set_lapic(vcpu, lapic);
1689 case KVM_INTERRUPT: {
1690 struct kvm_interrupt irq;
1693 if (copy_from_user(&irq, argp, sizeof irq))
1695 r = kvm_vcpu_ioctl_interrupt(vcpu, &irq);
1702 r = kvm_vcpu_ioctl_nmi(vcpu);
1708 case KVM_SET_CPUID: {
1709 struct kvm_cpuid __user *cpuid_arg = argp;
1710 struct kvm_cpuid cpuid;
1713 if (copy_from_user(&cpuid, cpuid_arg, sizeof cpuid))
1715 r = kvm_vcpu_ioctl_set_cpuid(vcpu, &cpuid, cpuid_arg->entries);
1720 case KVM_SET_CPUID2: {
1721 struct kvm_cpuid2 __user *cpuid_arg = argp;
1722 struct kvm_cpuid2 cpuid;
1725 if (copy_from_user(&cpuid, cpuid_arg, sizeof cpuid))
1727 r = kvm_vcpu_ioctl_set_cpuid2(vcpu, &cpuid,
1728 cpuid_arg->entries);
1733 case KVM_GET_CPUID2: {
1734 struct kvm_cpuid2 __user *cpuid_arg = argp;
1735 struct kvm_cpuid2 cpuid;
1738 if (copy_from_user(&cpuid, cpuid_arg, sizeof cpuid))
1740 r = kvm_vcpu_ioctl_get_cpuid2(vcpu, &cpuid,
1741 cpuid_arg->entries);
1745 if (copy_to_user(cpuid_arg, &cpuid, sizeof cpuid))
1751 r = msr_io(vcpu, argp, kvm_get_msr, 1);
1754 r = msr_io(vcpu, argp, do_set_msr, 0);
1756 case KVM_TPR_ACCESS_REPORTING: {
1757 struct kvm_tpr_access_ctl tac;
1760 if (copy_from_user(&tac, argp, sizeof tac))
1762 r = vcpu_ioctl_tpr_access_reporting(vcpu, &tac);
1766 if (copy_to_user(argp, &tac, sizeof tac))
1771 case KVM_SET_VAPIC_ADDR: {
1772 struct kvm_vapic_addr va;
1775 if (!irqchip_in_kernel(vcpu->kvm))
1778 if (copy_from_user(&va, argp, sizeof va))
1781 kvm_lapic_set_vapic_addr(vcpu, va.vapic_addr);
1784 case KVM_X86_SETUP_MCE: {
1788 if (copy_from_user(&mcg_cap, argp, sizeof mcg_cap))
1790 r = kvm_vcpu_ioctl_x86_setup_mce(vcpu, mcg_cap);
1793 case KVM_X86_SET_MCE: {
1794 struct kvm_x86_mce mce;
1797 if (copy_from_user(&mce, argp, sizeof mce))
1799 r = kvm_vcpu_ioctl_x86_set_mce(vcpu, &mce);
1810 static int kvm_vm_ioctl_set_tss_addr(struct kvm *kvm, unsigned long addr)
1814 if (addr > (unsigned int)(-3 * PAGE_SIZE))
1816 ret = kvm_x86_ops->set_tss_addr(kvm, addr);
1820 static int kvm_vm_ioctl_set_nr_mmu_pages(struct kvm *kvm,
1821 u32 kvm_nr_mmu_pages)
1823 if (kvm_nr_mmu_pages < KVM_MIN_ALLOC_MMU_PAGES)
1826 down_write(&kvm->slots_lock);
1827 spin_lock(&kvm->mmu_lock);
1829 kvm_mmu_change_mmu_pages(kvm, kvm_nr_mmu_pages);
1830 kvm->arch.n_requested_mmu_pages = kvm_nr_mmu_pages;
1832 spin_unlock(&kvm->mmu_lock);
1833 up_write(&kvm->slots_lock);
1837 static int kvm_vm_ioctl_get_nr_mmu_pages(struct kvm *kvm)
1839 return kvm->arch.n_alloc_mmu_pages;
1842 gfn_t unalias_gfn(struct kvm *kvm, gfn_t gfn)
1845 struct kvm_mem_alias *alias;
1847 for (i = 0; i < kvm->arch.naliases; ++i) {
1848 alias = &kvm->arch.aliases[i];
1849 if (gfn >= alias->base_gfn
1850 && gfn < alias->base_gfn + alias->npages)
1851 return alias->target_gfn + gfn - alias->base_gfn;
1857 * Set a new alias region. Aliases map a portion of physical memory into
1858 * another portion. This is useful for memory windows, for example the PC
1861 static int kvm_vm_ioctl_set_memory_alias(struct kvm *kvm,
1862 struct kvm_memory_alias *alias)
1865 struct kvm_mem_alias *p;
1868 /* General sanity checks */
1869 if (alias->memory_size & (PAGE_SIZE - 1))
1871 if (alias->guest_phys_addr & (PAGE_SIZE - 1))
1873 if (alias->slot >= KVM_ALIAS_SLOTS)
1875 if (alias->guest_phys_addr + alias->memory_size
1876 < alias->guest_phys_addr)
1878 if (alias->target_phys_addr + alias->memory_size
1879 < alias->target_phys_addr)
1882 down_write(&kvm->slots_lock);
1883 spin_lock(&kvm->mmu_lock);
1885 p = &kvm->arch.aliases[alias->slot];
1886 p->base_gfn = alias->guest_phys_addr >> PAGE_SHIFT;
1887 p->npages = alias->memory_size >> PAGE_SHIFT;
1888 p->target_gfn = alias->target_phys_addr >> PAGE_SHIFT;
1890 for (n = KVM_ALIAS_SLOTS; n > 0; --n)
1891 if (kvm->arch.aliases[n - 1].npages)
1893 kvm->arch.naliases = n;
1895 spin_unlock(&kvm->mmu_lock);
1896 kvm_mmu_zap_all(kvm);
1898 up_write(&kvm->slots_lock);
1906 static int kvm_vm_ioctl_get_irqchip(struct kvm *kvm, struct kvm_irqchip *chip)
1911 switch (chip->chip_id) {
1912 case KVM_IRQCHIP_PIC_MASTER:
1913 memcpy(&chip->chip.pic,
1914 &pic_irqchip(kvm)->pics[0],
1915 sizeof(struct kvm_pic_state));
1917 case KVM_IRQCHIP_PIC_SLAVE:
1918 memcpy(&chip->chip.pic,
1919 &pic_irqchip(kvm)->pics[1],
1920 sizeof(struct kvm_pic_state));
1922 case KVM_IRQCHIP_IOAPIC:
1923 memcpy(&chip->chip.ioapic,
1924 ioapic_irqchip(kvm),
1925 sizeof(struct kvm_ioapic_state));
1934 static int kvm_vm_ioctl_set_irqchip(struct kvm *kvm, struct kvm_irqchip *chip)
1939 switch (chip->chip_id) {
1940 case KVM_IRQCHIP_PIC_MASTER:
1941 memcpy(&pic_irqchip(kvm)->pics[0],
1943 sizeof(struct kvm_pic_state));
1945 case KVM_IRQCHIP_PIC_SLAVE:
1946 memcpy(&pic_irqchip(kvm)->pics[1],
1948 sizeof(struct kvm_pic_state));
1950 case KVM_IRQCHIP_IOAPIC:
1951 memcpy(ioapic_irqchip(kvm),
1953 sizeof(struct kvm_ioapic_state));
1959 kvm_pic_update_irq(pic_irqchip(kvm));
1963 static int kvm_vm_ioctl_get_pit(struct kvm *kvm, struct kvm_pit_state *ps)
1967 memcpy(ps, &kvm->arch.vpit->pit_state, sizeof(struct kvm_pit_state));
1971 static int kvm_vm_ioctl_set_pit(struct kvm *kvm, struct kvm_pit_state *ps)
1975 memcpy(&kvm->arch.vpit->pit_state, ps, sizeof(struct kvm_pit_state));
1976 kvm_pit_load_count(kvm, 0, ps->channels[0].count);
1980 static int kvm_vm_ioctl_reinject(struct kvm *kvm,
1981 struct kvm_reinject_control *control)
1983 if (!kvm->arch.vpit)
1985 kvm->arch.vpit->pit_state.pit_timer.reinject = control->pit_reinject;
1990 * Get (and clear) the dirty memory log for a memory slot.
1992 int kvm_vm_ioctl_get_dirty_log(struct kvm *kvm,
1993 struct kvm_dirty_log *log)
1997 struct kvm_memory_slot *memslot;
2000 down_write(&kvm->slots_lock);
2002 r = kvm_get_dirty_log(kvm, log, &is_dirty);
2006 /* If nothing is dirty, don't bother messing with page tables. */
2008 spin_lock(&kvm->mmu_lock);
2009 kvm_mmu_slot_remove_write_access(kvm, log->slot);
2010 spin_unlock(&kvm->mmu_lock);
2011 kvm_flush_remote_tlbs(kvm);
2012 memslot = &kvm->memslots[log->slot];
2013 n = ALIGN(memslot->npages, BITS_PER_LONG) / 8;
2014 memset(memslot->dirty_bitmap, 0, n);
2018 up_write(&kvm->slots_lock);
2022 long kvm_arch_vm_ioctl(struct file *filp,
2023 unsigned int ioctl, unsigned long arg)
2025 struct kvm *kvm = filp->private_data;
2026 void __user *argp = (void __user *)arg;
2029 * This union makes it completely explicit to gcc-3.x
2030 * that these two variables' stack usage should be
2031 * combined, not added together.
2034 struct kvm_pit_state ps;
2035 struct kvm_memory_alias alias;
2039 case KVM_SET_TSS_ADDR:
2040 r = kvm_vm_ioctl_set_tss_addr(kvm, arg);
2044 case KVM_SET_MEMORY_REGION: {
2045 struct kvm_memory_region kvm_mem;
2046 struct kvm_userspace_memory_region kvm_userspace_mem;
2049 if (copy_from_user(&kvm_mem, argp, sizeof kvm_mem))
2051 kvm_userspace_mem.slot = kvm_mem.slot;
2052 kvm_userspace_mem.flags = kvm_mem.flags;
2053 kvm_userspace_mem.guest_phys_addr = kvm_mem.guest_phys_addr;
2054 kvm_userspace_mem.memory_size = kvm_mem.memory_size;
2055 r = kvm_vm_ioctl_set_memory_region(kvm, &kvm_userspace_mem, 0);
2060 case KVM_SET_NR_MMU_PAGES:
2061 r = kvm_vm_ioctl_set_nr_mmu_pages(kvm, arg);
2065 case KVM_GET_NR_MMU_PAGES:
2066 r = kvm_vm_ioctl_get_nr_mmu_pages(kvm);
2068 case KVM_SET_MEMORY_ALIAS:
2070 if (copy_from_user(&u.alias, argp, sizeof(struct kvm_memory_alias)))
2072 r = kvm_vm_ioctl_set_memory_alias(kvm, &u.alias);
2076 case KVM_CREATE_IRQCHIP:
2078 kvm->arch.vpic = kvm_create_pic(kvm);
2079 if (kvm->arch.vpic) {
2080 r = kvm_ioapic_init(kvm);
2082 kfree(kvm->arch.vpic);
2083 kvm->arch.vpic = NULL;
2088 r = kvm_setup_default_irq_routing(kvm);
2090 kfree(kvm->arch.vpic);
2091 kfree(kvm->arch.vioapic);
2095 case KVM_CREATE_PIT:
2096 mutex_lock(&kvm->lock);
2099 goto create_pit_unlock;
2101 kvm->arch.vpit = kvm_create_pit(kvm);
2105 mutex_unlock(&kvm->lock);
2107 case KVM_IRQ_LINE_STATUS:
2108 case KVM_IRQ_LINE: {
2109 struct kvm_irq_level irq_event;
2112 if (copy_from_user(&irq_event, argp, sizeof irq_event))
2114 if (irqchip_in_kernel(kvm)) {
2116 mutex_lock(&kvm->lock);
2117 status = kvm_set_irq(kvm, KVM_USERSPACE_IRQ_SOURCE_ID,
2118 irq_event.irq, irq_event.level);
2119 mutex_unlock(&kvm->lock);
2120 if (ioctl == KVM_IRQ_LINE_STATUS) {
2121 irq_event.status = status;
2122 if (copy_to_user(argp, &irq_event,
2130 case KVM_GET_IRQCHIP: {
2131 /* 0: PIC master, 1: PIC slave, 2: IOAPIC */
2132 struct kvm_irqchip *chip = kmalloc(sizeof(*chip), GFP_KERNEL);
2138 if (copy_from_user(chip, argp, sizeof *chip))
2139 goto get_irqchip_out;
2141 if (!irqchip_in_kernel(kvm))
2142 goto get_irqchip_out;
2143 r = kvm_vm_ioctl_get_irqchip(kvm, chip);
2145 goto get_irqchip_out;
2147 if (copy_to_user(argp, chip, sizeof *chip))
2148 goto get_irqchip_out;
2156 case KVM_SET_IRQCHIP: {
2157 /* 0: PIC master, 1: PIC slave, 2: IOAPIC */
2158 struct kvm_irqchip *chip = kmalloc(sizeof(*chip), GFP_KERNEL);
2164 if (copy_from_user(chip, argp, sizeof *chip))
2165 goto set_irqchip_out;
2167 if (!irqchip_in_kernel(kvm))
2168 goto set_irqchip_out;
2169 r = kvm_vm_ioctl_set_irqchip(kvm, chip);
2171 goto set_irqchip_out;
2181 if (copy_from_user(&u.ps, argp, sizeof(struct kvm_pit_state)))
2184 if (!kvm->arch.vpit)
2186 r = kvm_vm_ioctl_get_pit(kvm, &u.ps);
2190 if (copy_to_user(argp, &u.ps, sizeof(struct kvm_pit_state)))
2197 if (copy_from_user(&u.ps, argp, sizeof u.ps))
2200 if (!kvm->arch.vpit)
2202 r = kvm_vm_ioctl_set_pit(kvm, &u.ps);
2208 case KVM_REINJECT_CONTROL: {
2209 struct kvm_reinject_control control;
2211 if (copy_from_user(&control, argp, sizeof(control)))
2213 r = kvm_vm_ioctl_reinject(kvm, &control);
2226 static void kvm_init_msr_list(void)
2231 for (i = j = 0; i < ARRAY_SIZE(msrs_to_save); i++) {
2232 if (rdmsr_safe(msrs_to_save[i], &dummy[0], &dummy[1]) < 0)
2235 msrs_to_save[j] = msrs_to_save[i];
2238 num_msrs_to_save = j;
2242 * Only apic need an MMIO device hook, so shortcut now..
2244 static struct kvm_io_device *vcpu_find_pervcpu_dev(struct kvm_vcpu *vcpu,
2245 gpa_t addr, int len,
2248 struct kvm_io_device *dev;
2250 if (vcpu->arch.apic) {
2251 dev = &vcpu->arch.apic->dev;
2252 if (dev->in_range(dev, addr, len, is_write))
2259 static struct kvm_io_device *vcpu_find_mmio_dev(struct kvm_vcpu *vcpu,
2260 gpa_t addr, int len,
2263 struct kvm_io_device *dev;
2265 dev = vcpu_find_pervcpu_dev(vcpu, addr, len, is_write);
2267 dev = kvm_io_bus_find_dev(&vcpu->kvm->mmio_bus, addr, len,
2272 static int kvm_read_guest_virt(gva_t addr, void *val, unsigned int bytes,
2273 struct kvm_vcpu *vcpu)
2276 int r = X86EMUL_CONTINUE;
2279 gpa_t gpa = vcpu->arch.mmu.gva_to_gpa(vcpu, addr);
2280 unsigned offset = addr & (PAGE_SIZE-1);
2281 unsigned toread = min(bytes, (unsigned)PAGE_SIZE - offset);
2284 if (gpa == UNMAPPED_GVA) {
2285 r = X86EMUL_PROPAGATE_FAULT;
2288 ret = kvm_read_guest(vcpu->kvm, gpa, data, toread);
2290 r = X86EMUL_UNHANDLEABLE;
2302 static int kvm_write_guest_virt(gva_t addr, void *val, unsigned int bytes,
2303 struct kvm_vcpu *vcpu)
2306 int r = X86EMUL_CONTINUE;
2309 gpa_t gpa = vcpu->arch.mmu.gva_to_gpa(vcpu, addr);
2310 unsigned offset = addr & (PAGE_SIZE-1);
2311 unsigned towrite = min(bytes, (unsigned)PAGE_SIZE - offset);
2314 if (gpa == UNMAPPED_GVA) {
2315 r = X86EMUL_PROPAGATE_FAULT;
2318 ret = kvm_write_guest(vcpu->kvm, gpa, data, towrite);
2320 r = X86EMUL_UNHANDLEABLE;
2333 static int emulator_read_emulated(unsigned long addr,
2336 struct kvm_vcpu *vcpu)
2338 struct kvm_io_device *mmio_dev;
2341 if (vcpu->mmio_read_completed) {
2342 memcpy(val, vcpu->mmio_data, bytes);
2343 vcpu->mmio_read_completed = 0;
2344 return X86EMUL_CONTINUE;
2347 gpa = vcpu->arch.mmu.gva_to_gpa(vcpu, addr);
2349 /* For APIC access vmexit */
2350 if ((gpa & PAGE_MASK) == APIC_DEFAULT_PHYS_BASE)
2353 if (kvm_read_guest_virt(addr, val, bytes, vcpu)
2354 == X86EMUL_CONTINUE)
2355 return X86EMUL_CONTINUE;
2356 if (gpa == UNMAPPED_GVA)
2357 return X86EMUL_PROPAGATE_FAULT;
2361 * Is this MMIO handled locally?
2363 mutex_lock(&vcpu->kvm->lock);
2364 mmio_dev = vcpu_find_mmio_dev(vcpu, gpa, bytes, 0);
2366 kvm_iodevice_read(mmio_dev, gpa, bytes, val);
2367 mutex_unlock(&vcpu->kvm->lock);
2368 return X86EMUL_CONTINUE;
2370 mutex_unlock(&vcpu->kvm->lock);
2372 vcpu->mmio_needed = 1;
2373 vcpu->mmio_phys_addr = gpa;
2374 vcpu->mmio_size = bytes;
2375 vcpu->mmio_is_write = 0;
2377 return X86EMUL_UNHANDLEABLE;
2380 int emulator_write_phys(struct kvm_vcpu *vcpu, gpa_t gpa,
2381 const void *val, int bytes)
2385 ret = kvm_write_guest(vcpu->kvm, gpa, val, bytes);
2388 kvm_mmu_pte_write(vcpu, gpa, val, bytes, 1);
2392 static int emulator_write_emulated_onepage(unsigned long addr,
2395 struct kvm_vcpu *vcpu)
2397 struct kvm_io_device *mmio_dev;
2400 gpa = vcpu->arch.mmu.gva_to_gpa(vcpu, addr);
2402 if (gpa == UNMAPPED_GVA) {
2403 kvm_inject_page_fault(vcpu, addr, 2);
2404 return X86EMUL_PROPAGATE_FAULT;
2407 /* For APIC access vmexit */
2408 if ((gpa & PAGE_MASK) == APIC_DEFAULT_PHYS_BASE)
2411 if (emulator_write_phys(vcpu, gpa, val, bytes))
2412 return X86EMUL_CONTINUE;
2416 * Is this MMIO handled locally?
2418 mutex_lock(&vcpu->kvm->lock);
2419 mmio_dev = vcpu_find_mmio_dev(vcpu, gpa, bytes, 1);
2421 kvm_iodevice_write(mmio_dev, gpa, bytes, val);
2422 mutex_unlock(&vcpu->kvm->lock);
2423 return X86EMUL_CONTINUE;
2425 mutex_unlock(&vcpu->kvm->lock);
2427 vcpu->mmio_needed = 1;
2428 vcpu->mmio_phys_addr = gpa;
2429 vcpu->mmio_size = bytes;
2430 vcpu->mmio_is_write = 1;
2431 memcpy(vcpu->mmio_data, val, bytes);
2433 return X86EMUL_CONTINUE;
2436 int emulator_write_emulated(unsigned long addr,
2439 struct kvm_vcpu *vcpu)
2441 /* Crossing a page boundary? */
2442 if (((addr + bytes - 1) ^ addr) & PAGE_MASK) {
2445 now = -addr & ~PAGE_MASK;
2446 rc = emulator_write_emulated_onepage(addr, val, now, vcpu);
2447 if (rc != X86EMUL_CONTINUE)
2453 return emulator_write_emulated_onepage(addr, val, bytes, vcpu);
2455 EXPORT_SYMBOL_GPL(emulator_write_emulated);
2457 static int emulator_cmpxchg_emulated(unsigned long addr,
2461 struct kvm_vcpu *vcpu)
2463 static int reported;
2467 printk(KERN_WARNING "kvm: emulating exchange as write\n");
2469 #ifndef CONFIG_X86_64
2470 /* guests cmpxchg8b have to be emulated atomically */
2477 gpa = vcpu->arch.mmu.gva_to_gpa(vcpu, addr);
2479 if (gpa == UNMAPPED_GVA ||
2480 (gpa & PAGE_MASK) == APIC_DEFAULT_PHYS_BASE)
2483 if (((gpa + bytes - 1) & PAGE_MASK) != (gpa & PAGE_MASK))
2488 page = gfn_to_page(vcpu->kvm, gpa >> PAGE_SHIFT);
2490 kaddr = kmap_atomic(page, KM_USER0);
2491 set_64bit((u64 *)(kaddr + offset_in_page(gpa)), val);
2492 kunmap_atomic(kaddr, KM_USER0);
2493 kvm_release_page_dirty(page);
2498 return emulator_write_emulated(addr, new, bytes, vcpu);
2501 static unsigned long get_segment_base(struct kvm_vcpu *vcpu, int seg)
2503 return kvm_x86_ops->get_segment_base(vcpu, seg);
2506 int emulate_invlpg(struct kvm_vcpu *vcpu, gva_t address)
2508 kvm_mmu_invlpg(vcpu, address);
2509 return X86EMUL_CONTINUE;
2512 int emulate_clts(struct kvm_vcpu *vcpu)
2514 KVMTRACE_0D(CLTS, vcpu, handler);
2515 kvm_x86_ops->set_cr0(vcpu, vcpu->arch.cr0 & ~X86_CR0_TS);
2516 return X86EMUL_CONTINUE;
2519 int emulator_get_dr(struct x86_emulate_ctxt *ctxt, int dr, unsigned long *dest)
2521 struct kvm_vcpu *vcpu = ctxt->vcpu;
2525 *dest = kvm_x86_ops->get_dr(vcpu, dr);
2526 return X86EMUL_CONTINUE;
2528 pr_unimpl(vcpu, "%s: unexpected dr %u\n", __func__, dr);
2529 return X86EMUL_UNHANDLEABLE;
2533 int emulator_set_dr(struct x86_emulate_ctxt *ctxt, int dr, unsigned long value)
2535 unsigned long mask = (ctxt->mode == X86EMUL_MODE_PROT64) ? ~0ULL : ~0U;
2538 kvm_x86_ops->set_dr(ctxt->vcpu, dr, value & mask, &exception);
2540 /* FIXME: better handling */
2541 return X86EMUL_UNHANDLEABLE;
2543 return X86EMUL_CONTINUE;
2546 void kvm_report_emulation_failure(struct kvm_vcpu *vcpu, const char *context)
2549 unsigned long rip = kvm_rip_read(vcpu);
2550 unsigned long rip_linear;
2552 if (!printk_ratelimit())
2555 rip_linear = rip + get_segment_base(vcpu, VCPU_SREG_CS);
2557 kvm_read_guest_virt(rip_linear, (void *)opcodes, 4, vcpu);
2559 printk(KERN_ERR "emulation failed (%s) rip %lx %02x %02x %02x %02x\n",
2560 context, rip, opcodes[0], opcodes[1], opcodes[2], opcodes[3]);
2562 EXPORT_SYMBOL_GPL(kvm_report_emulation_failure);
2564 static struct x86_emulate_ops emulate_ops = {
2565 .read_std = kvm_read_guest_virt,
2566 .read_emulated = emulator_read_emulated,
2567 .write_emulated = emulator_write_emulated,
2568 .cmpxchg_emulated = emulator_cmpxchg_emulated,
2571 static void cache_all_regs(struct kvm_vcpu *vcpu)
2573 kvm_register_read(vcpu, VCPU_REGS_RAX);
2574 kvm_register_read(vcpu, VCPU_REGS_RSP);
2575 kvm_register_read(vcpu, VCPU_REGS_RIP);
2576 vcpu->arch.regs_dirty = ~0;
2579 int emulate_instruction(struct kvm_vcpu *vcpu,
2580 struct kvm_run *run,
2586 struct decode_cache *c;
2588 kvm_clear_exception_queue(vcpu);
2589 vcpu->arch.mmio_fault_cr2 = cr2;
2591 * TODO: fix x86_emulate.c to use guest_read/write_register
2592 * instead of direct ->regs accesses, can save hundred cycles
2593 * on Intel for instructions that don't read/change RSP, for
2596 cache_all_regs(vcpu);
2598 vcpu->mmio_is_write = 0;
2599 vcpu->arch.pio.string = 0;
2601 if (!(emulation_type & EMULTYPE_NO_DECODE)) {
2603 kvm_x86_ops->get_cs_db_l_bits(vcpu, &cs_db, &cs_l);
2605 vcpu->arch.emulate_ctxt.vcpu = vcpu;
2606 vcpu->arch.emulate_ctxt.eflags = kvm_x86_ops->get_rflags(vcpu);
2607 vcpu->arch.emulate_ctxt.mode =
2608 (vcpu->arch.emulate_ctxt.eflags & X86_EFLAGS_VM)
2609 ? X86EMUL_MODE_REAL : cs_l
2610 ? X86EMUL_MODE_PROT64 : cs_db
2611 ? X86EMUL_MODE_PROT32 : X86EMUL_MODE_PROT16;
2613 r = x86_decode_insn(&vcpu->arch.emulate_ctxt, &emulate_ops);
2615 /* Reject the instructions other than VMCALL/VMMCALL when
2616 * try to emulate invalid opcode */
2617 c = &vcpu->arch.emulate_ctxt.decode;
2618 if ((emulation_type & EMULTYPE_TRAP_UD) &&
2619 (!(c->twobyte && c->b == 0x01 &&
2620 (c->modrm_reg == 0 || c->modrm_reg == 3) &&
2621 c->modrm_mod == 3 && c->modrm_rm == 1)))
2622 return EMULATE_FAIL;
2624 ++vcpu->stat.insn_emulation;
2626 ++vcpu->stat.insn_emulation_fail;
2627 if (kvm_mmu_unprotect_page_virt(vcpu, cr2))
2628 return EMULATE_DONE;
2629 return EMULATE_FAIL;
2633 if (emulation_type & EMULTYPE_SKIP) {
2634 kvm_rip_write(vcpu, vcpu->arch.emulate_ctxt.decode.eip);
2635 return EMULATE_DONE;
2638 r = x86_emulate_insn(&vcpu->arch.emulate_ctxt, &emulate_ops);
2639 shadow_mask = vcpu->arch.emulate_ctxt.interruptibility;
2642 kvm_x86_ops->set_interrupt_shadow(vcpu, shadow_mask);
2644 if (vcpu->arch.pio.string)
2645 return EMULATE_DO_MMIO;
2647 if ((r || vcpu->mmio_is_write) && run) {
2648 run->exit_reason = KVM_EXIT_MMIO;
2649 run->mmio.phys_addr = vcpu->mmio_phys_addr;
2650 memcpy(run->mmio.data, vcpu->mmio_data, 8);
2651 run->mmio.len = vcpu->mmio_size;
2652 run->mmio.is_write = vcpu->mmio_is_write;
2656 if (kvm_mmu_unprotect_page_virt(vcpu, cr2))
2657 return EMULATE_DONE;
2658 if (!vcpu->mmio_needed) {
2659 kvm_report_emulation_failure(vcpu, "mmio");
2660 return EMULATE_FAIL;
2662 return EMULATE_DO_MMIO;
2665 kvm_x86_ops->set_rflags(vcpu, vcpu->arch.emulate_ctxt.eflags);
2667 if (vcpu->mmio_is_write) {
2668 vcpu->mmio_needed = 0;
2669 return EMULATE_DO_MMIO;
2672 return EMULATE_DONE;
2674 EXPORT_SYMBOL_GPL(emulate_instruction);
2676 static int pio_copy_data(struct kvm_vcpu *vcpu)
2678 void *p = vcpu->arch.pio_data;
2679 gva_t q = vcpu->arch.pio.guest_gva;
2683 bytes = vcpu->arch.pio.size * vcpu->arch.pio.cur_count;
2684 if (vcpu->arch.pio.in)
2685 ret = kvm_write_guest_virt(q, p, bytes, vcpu);
2687 ret = kvm_read_guest_virt(q, p, bytes, vcpu);
2691 int complete_pio(struct kvm_vcpu *vcpu)
2693 struct kvm_pio_request *io = &vcpu->arch.pio;
2700 val = kvm_register_read(vcpu, VCPU_REGS_RAX);
2701 memcpy(&val, vcpu->arch.pio_data, io->size);
2702 kvm_register_write(vcpu, VCPU_REGS_RAX, val);
2706 r = pio_copy_data(vcpu);
2713 delta *= io->cur_count;
2715 * The size of the register should really depend on
2716 * current address size.
2718 val = kvm_register_read(vcpu, VCPU_REGS_RCX);
2720 kvm_register_write(vcpu, VCPU_REGS_RCX, val);
2726 val = kvm_register_read(vcpu, VCPU_REGS_RDI);
2728 kvm_register_write(vcpu, VCPU_REGS_RDI, val);
2730 val = kvm_register_read(vcpu, VCPU_REGS_RSI);
2732 kvm_register_write(vcpu, VCPU_REGS_RSI, val);
2736 io->count -= io->cur_count;
2742 static void kernel_pio(struct kvm_io_device *pio_dev,
2743 struct kvm_vcpu *vcpu,
2746 /* TODO: String I/O for in kernel device */
2748 mutex_lock(&vcpu->kvm->lock);
2749 if (vcpu->arch.pio.in)
2750 kvm_iodevice_read(pio_dev, vcpu->arch.pio.port,
2751 vcpu->arch.pio.size,
2754 kvm_iodevice_write(pio_dev, vcpu->arch.pio.port,
2755 vcpu->arch.pio.size,
2757 mutex_unlock(&vcpu->kvm->lock);
2760 static void pio_string_write(struct kvm_io_device *pio_dev,
2761 struct kvm_vcpu *vcpu)
2763 struct kvm_pio_request *io = &vcpu->arch.pio;
2764 void *pd = vcpu->arch.pio_data;
2767 mutex_lock(&vcpu->kvm->lock);
2768 for (i = 0; i < io->cur_count; i++) {
2769 kvm_iodevice_write(pio_dev, io->port,
2774 mutex_unlock(&vcpu->kvm->lock);
2777 static struct kvm_io_device *vcpu_find_pio_dev(struct kvm_vcpu *vcpu,
2778 gpa_t addr, int len,
2781 return kvm_io_bus_find_dev(&vcpu->kvm->pio_bus, addr, len, is_write);
2784 int kvm_emulate_pio(struct kvm_vcpu *vcpu, struct kvm_run *run, int in,
2785 int size, unsigned port)
2787 struct kvm_io_device *pio_dev;
2790 vcpu->run->exit_reason = KVM_EXIT_IO;
2791 vcpu->run->io.direction = in ? KVM_EXIT_IO_IN : KVM_EXIT_IO_OUT;
2792 vcpu->run->io.size = vcpu->arch.pio.size = size;
2793 vcpu->run->io.data_offset = KVM_PIO_PAGE_OFFSET * PAGE_SIZE;
2794 vcpu->run->io.count = vcpu->arch.pio.count = vcpu->arch.pio.cur_count = 1;
2795 vcpu->run->io.port = vcpu->arch.pio.port = port;
2796 vcpu->arch.pio.in = in;
2797 vcpu->arch.pio.string = 0;
2798 vcpu->arch.pio.down = 0;
2799 vcpu->arch.pio.rep = 0;
2801 if (vcpu->run->io.direction == KVM_EXIT_IO_IN)
2802 KVMTRACE_2D(IO_READ, vcpu, vcpu->run->io.port, (u32)size,
2805 KVMTRACE_2D(IO_WRITE, vcpu, vcpu->run->io.port, (u32)size,
2808 val = kvm_register_read(vcpu, VCPU_REGS_RAX);
2809 memcpy(vcpu->arch.pio_data, &val, 4);
2811 pio_dev = vcpu_find_pio_dev(vcpu, port, size, !in);
2813 kernel_pio(pio_dev, vcpu, vcpu->arch.pio_data);
2819 EXPORT_SYMBOL_GPL(kvm_emulate_pio);
2821 int kvm_emulate_pio_string(struct kvm_vcpu *vcpu, struct kvm_run *run, int in,
2822 int size, unsigned long count, int down,
2823 gva_t address, int rep, unsigned port)
2825 unsigned now, in_page;
2827 struct kvm_io_device *pio_dev;
2829 vcpu->run->exit_reason = KVM_EXIT_IO;
2830 vcpu->run->io.direction = in ? KVM_EXIT_IO_IN : KVM_EXIT_IO_OUT;
2831 vcpu->run->io.size = vcpu->arch.pio.size = size;
2832 vcpu->run->io.data_offset = KVM_PIO_PAGE_OFFSET * PAGE_SIZE;
2833 vcpu->run->io.count = vcpu->arch.pio.count = vcpu->arch.pio.cur_count = count;
2834 vcpu->run->io.port = vcpu->arch.pio.port = port;
2835 vcpu->arch.pio.in = in;
2836 vcpu->arch.pio.string = 1;
2837 vcpu->arch.pio.down = down;
2838 vcpu->arch.pio.rep = rep;
2840 if (vcpu->run->io.direction == KVM_EXIT_IO_IN)
2841 KVMTRACE_2D(IO_READ, vcpu, vcpu->run->io.port, (u32)size,
2844 KVMTRACE_2D(IO_WRITE, vcpu, vcpu->run->io.port, (u32)size,
2848 kvm_x86_ops->skip_emulated_instruction(vcpu);
2853 in_page = PAGE_SIZE - offset_in_page(address);
2855 in_page = offset_in_page(address) + size;
2856 now = min(count, (unsigned long)in_page / size);
2861 * String I/O in reverse. Yuck. Kill the guest, fix later.
2863 pr_unimpl(vcpu, "guest string pio down\n");
2864 kvm_inject_gp(vcpu, 0);
2867 vcpu->run->io.count = now;
2868 vcpu->arch.pio.cur_count = now;
2870 if (vcpu->arch.pio.cur_count == vcpu->arch.pio.count)
2871 kvm_x86_ops->skip_emulated_instruction(vcpu);
2873 vcpu->arch.pio.guest_gva = address;
2875 pio_dev = vcpu_find_pio_dev(vcpu, port,
2876 vcpu->arch.pio.cur_count,
2877 !vcpu->arch.pio.in);
2878 if (!vcpu->arch.pio.in) {
2879 /* string PIO write */
2880 ret = pio_copy_data(vcpu);
2881 if (ret == X86EMUL_PROPAGATE_FAULT) {
2882 kvm_inject_gp(vcpu, 0);
2885 if (ret == 0 && pio_dev) {
2886 pio_string_write(pio_dev, vcpu);
2888 if (vcpu->arch.pio.count == 0)
2892 pr_unimpl(vcpu, "no string pio read support yet, "
2893 "port %x size %d count %ld\n",
2898 EXPORT_SYMBOL_GPL(kvm_emulate_pio_string);
2900 static void bounce_off(void *info)
2905 static unsigned int ref_freq;
2906 static unsigned long tsc_khz_ref;
2908 static int kvmclock_cpufreq_notifier(struct notifier_block *nb, unsigned long val,
2911 struct cpufreq_freqs *freq = data;
2913 struct kvm_vcpu *vcpu;
2914 int i, send_ipi = 0;
2917 ref_freq = freq->old;
2919 if (val == CPUFREQ_PRECHANGE && freq->old > freq->new)
2921 if (val == CPUFREQ_POSTCHANGE && freq->old < freq->new)
2923 per_cpu(cpu_tsc_khz, freq->cpu) = cpufreq_scale(tsc_khz_ref, ref_freq, freq->new);
2925 spin_lock(&kvm_lock);
2926 list_for_each_entry(kvm, &vm_list, vm_list) {
2927 for (i = 0; i < KVM_MAX_VCPUS; ++i) {
2928 vcpu = kvm->vcpus[i];
2931 if (vcpu->cpu != freq->cpu)
2933 if (!kvm_request_guest_time_update(vcpu))
2935 if (vcpu->cpu != smp_processor_id())
2939 spin_unlock(&kvm_lock);
2941 if (freq->old < freq->new && send_ipi) {
2943 * We upscale the frequency. Must make the guest
2944 * doesn't see old kvmclock values while running with
2945 * the new frequency, otherwise we risk the guest sees
2946 * time go backwards.
2948 * In case we update the frequency for another cpu
2949 * (which might be in guest context) send an interrupt
2950 * to kick the cpu out of guest context. Next time
2951 * guest context is entered kvmclock will be updated,
2952 * so the guest will not see stale values.
2954 smp_call_function_single(freq->cpu, bounce_off, NULL, 1);
2959 static struct notifier_block kvmclock_cpufreq_notifier_block = {
2960 .notifier_call = kvmclock_cpufreq_notifier
2963 int kvm_arch_init(void *opaque)
2966 struct kvm_x86_ops *ops = (struct kvm_x86_ops *)opaque;
2969 printk(KERN_ERR "kvm: already loaded the other module\n");
2974 if (!ops->cpu_has_kvm_support()) {
2975 printk(KERN_ERR "kvm: no hardware support\n");
2979 if (ops->disabled_by_bios()) {
2980 printk(KERN_ERR "kvm: disabled by bios\n");
2985 r = kvm_mmu_module_init();
2989 kvm_init_msr_list();
2992 kvm_mmu_set_nonpresent_ptes(0ull, 0ull);
2993 kvm_mmu_set_base_ptes(PT_PRESENT_MASK);
2994 kvm_mmu_set_mask_ptes(PT_USER_MASK, PT_ACCESSED_MASK,
2995 PT_DIRTY_MASK, PT64_NX_MASK, 0);
2997 for_each_possible_cpu(cpu)
2998 per_cpu(cpu_tsc_khz, cpu) = tsc_khz;
2999 if (!boot_cpu_has(X86_FEATURE_CONSTANT_TSC)) {
3000 tsc_khz_ref = tsc_khz;
3001 cpufreq_register_notifier(&kvmclock_cpufreq_notifier_block,
3002 CPUFREQ_TRANSITION_NOTIFIER);
3011 void kvm_arch_exit(void)
3013 if (!boot_cpu_has(X86_FEATURE_CONSTANT_TSC))
3014 cpufreq_unregister_notifier(&kvmclock_cpufreq_notifier_block,
3015 CPUFREQ_TRANSITION_NOTIFIER);
3017 kvm_mmu_module_exit();
3020 int kvm_emulate_halt(struct kvm_vcpu *vcpu)
3022 ++vcpu->stat.halt_exits;
3023 KVMTRACE_0D(HLT, vcpu, handler);
3024 if (irqchip_in_kernel(vcpu->kvm)) {
3025 vcpu->arch.mp_state = KVM_MP_STATE_HALTED;
3028 vcpu->run->exit_reason = KVM_EXIT_HLT;
3032 EXPORT_SYMBOL_GPL(kvm_emulate_halt);
3034 static inline gpa_t hc_gpa(struct kvm_vcpu *vcpu, unsigned long a0,
3037 if (is_long_mode(vcpu))
3040 return a0 | ((gpa_t)a1 << 32);
3043 int kvm_emulate_hypercall(struct kvm_vcpu *vcpu)
3045 unsigned long nr, a0, a1, a2, a3, ret;
3048 nr = kvm_register_read(vcpu, VCPU_REGS_RAX);
3049 a0 = kvm_register_read(vcpu, VCPU_REGS_RBX);
3050 a1 = kvm_register_read(vcpu, VCPU_REGS_RCX);
3051 a2 = kvm_register_read(vcpu, VCPU_REGS_RDX);
3052 a3 = kvm_register_read(vcpu, VCPU_REGS_RSI);
3054 KVMTRACE_1D(VMMCALL, vcpu, (u32)nr, handler);
3056 if (!is_long_mode(vcpu)) {
3065 case KVM_HC_VAPIC_POLL_IRQ:
3069 r = kvm_pv_mmu_op(vcpu, a0, hc_gpa(vcpu, a1, a2), &ret);
3075 kvm_register_write(vcpu, VCPU_REGS_RAX, ret);
3076 ++vcpu->stat.hypercalls;
3079 EXPORT_SYMBOL_GPL(kvm_emulate_hypercall);
3081 int kvm_fix_hypercall(struct kvm_vcpu *vcpu)
3083 char instruction[3];
3085 unsigned long rip = kvm_rip_read(vcpu);
3089 * Blow out the MMU to ensure that no other VCPU has an active mapping
3090 * to ensure that the updated hypercall appears atomically across all
3093 kvm_mmu_zap_all(vcpu->kvm);
3095 kvm_x86_ops->patch_hypercall(vcpu, instruction);
3096 if (emulator_write_emulated(rip, instruction, 3, vcpu)
3097 != X86EMUL_CONTINUE)
3103 static u64 mk_cr_64(u64 curr_cr, u32 new_val)
3105 return (curr_cr & ~((1ULL << 32) - 1)) | new_val;
3108 void realmode_lgdt(struct kvm_vcpu *vcpu, u16 limit, unsigned long base)
3110 struct descriptor_table dt = { limit, base };
3112 kvm_x86_ops->set_gdt(vcpu, &dt);
3115 void realmode_lidt(struct kvm_vcpu *vcpu, u16 limit, unsigned long base)
3117 struct descriptor_table dt = { limit, base };
3119 kvm_x86_ops->set_idt(vcpu, &dt);
3122 void realmode_lmsw(struct kvm_vcpu *vcpu, unsigned long msw,
3123 unsigned long *rflags)
3125 kvm_lmsw(vcpu, msw);
3126 *rflags = kvm_x86_ops->get_rflags(vcpu);
3129 unsigned long realmode_get_cr(struct kvm_vcpu *vcpu, int cr)
3131 unsigned long value;
3133 kvm_x86_ops->decache_cr4_guest_bits(vcpu);
3136 value = vcpu->arch.cr0;
3139 value = vcpu->arch.cr2;
3142 value = vcpu->arch.cr3;
3145 value = vcpu->arch.cr4;
3148 value = kvm_get_cr8(vcpu);
3151 vcpu_printf(vcpu, "%s: unexpected cr %u\n", __func__, cr);
3154 KVMTRACE_3D(CR_READ, vcpu, (u32)cr, (u32)value,
3155 (u32)((u64)value >> 32), handler);
3160 void realmode_set_cr(struct kvm_vcpu *vcpu, int cr, unsigned long val,
3161 unsigned long *rflags)
3163 KVMTRACE_3D(CR_WRITE, vcpu, (u32)cr, (u32)val,
3164 (u32)((u64)val >> 32), handler);
3168 kvm_set_cr0(vcpu, mk_cr_64(vcpu->arch.cr0, val));
3169 *rflags = kvm_x86_ops->get_rflags(vcpu);
3172 vcpu->arch.cr2 = val;
3175 kvm_set_cr3(vcpu, val);
3178 kvm_set_cr4(vcpu, mk_cr_64(vcpu->arch.cr4, val));
3181 kvm_set_cr8(vcpu, val & 0xfUL);
3184 vcpu_printf(vcpu, "%s: unexpected cr %u\n", __func__, cr);
3188 static int move_to_next_stateful_cpuid_entry(struct kvm_vcpu *vcpu, int i)
3190 struct kvm_cpuid_entry2 *e = &vcpu->arch.cpuid_entries[i];
3191 int j, nent = vcpu->arch.cpuid_nent;
3193 e->flags &= ~KVM_CPUID_FLAG_STATE_READ_NEXT;
3194 /* when no next entry is found, the current entry[i] is reselected */
3195 for (j = i + 1; ; j = (j + 1) % nent) {
3196 struct kvm_cpuid_entry2 *ej = &vcpu->arch.cpuid_entries[j];
3197 if (ej->function == e->function) {
3198 ej->flags |= KVM_CPUID_FLAG_STATE_READ_NEXT;
3202 return 0; /* silence gcc, even though control never reaches here */
3205 /* find an entry with matching function, matching index (if needed), and that
3206 * should be read next (if it's stateful) */
3207 static int is_matching_cpuid_entry(struct kvm_cpuid_entry2 *e,
3208 u32 function, u32 index)
3210 if (e->function != function)
3212 if ((e->flags & KVM_CPUID_FLAG_SIGNIFCANT_INDEX) && e->index != index)
3214 if ((e->flags & KVM_CPUID_FLAG_STATEFUL_FUNC) &&
3215 !(e->flags & KVM_CPUID_FLAG_STATE_READ_NEXT))
3220 struct kvm_cpuid_entry2 *kvm_find_cpuid_entry(struct kvm_vcpu *vcpu,
3221 u32 function, u32 index)
3224 struct kvm_cpuid_entry2 *best = NULL;
3226 for (i = 0; i < vcpu->arch.cpuid_nent; ++i) {
3227 struct kvm_cpuid_entry2 *e;
3229 e = &vcpu->arch.cpuid_entries[i];
3230 if (is_matching_cpuid_entry(e, function, index)) {
3231 if (e->flags & KVM_CPUID_FLAG_STATEFUL_FUNC)
3232 move_to_next_stateful_cpuid_entry(vcpu, i);
3237 * Both basic or both extended?
3239 if (((e->function ^ function) & 0x80000000) == 0)
3240 if (!best || e->function > best->function)
3246 int cpuid_maxphyaddr(struct kvm_vcpu *vcpu)
3248 struct kvm_cpuid_entry2 *best;
3250 best = kvm_find_cpuid_entry(vcpu, 0x80000008, 0);
3252 return best->eax & 0xff;
3256 void kvm_emulate_cpuid(struct kvm_vcpu *vcpu)
3258 u32 function, index;
3259 struct kvm_cpuid_entry2 *best;
3261 function = kvm_register_read(vcpu, VCPU_REGS_RAX);
3262 index = kvm_register_read(vcpu, VCPU_REGS_RCX);
3263 kvm_register_write(vcpu, VCPU_REGS_RAX, 0);
3264 kvm_register_write(vcpu, VCPU_REGS_RBX, 0);
3265 kvm_register_write(vcpu, VCPU_REGS_RCX, 0);
3266 kvm_register_write(vcpu, VCPU_REGS_RDX, 0);
3267 best = kvm_find_cpuid_entry(vcpu, function, index);
3269 kvm_register_write(vcpu, VCPU_REGS_RAX, best->eax);
3270 kvm_register_write(vcpu, VCPU_REGS_RBX, best->ebx);
3271 kvm_register_write(vcpu, VCPU_REGS_RCX, best->ecx);
3272 kvm_register_write(vcpu, VCPU_REGS_RDX, best->edx);
3274 kvm_x86_ops->skip_emulated_instruction(vcpu);
3275 KVMTRACE_5D(CPUID, vcpu, function,
3276 (u32)kvm_register_read(vcpu, VCPU_REGS_RAX),
3277 (u32)kvm_register_read(vcpu, VCPU_REGS_RBX),
3278 (u32)kvm_register_read(vcpu, VCPU_REGS_RCX),
3279 (u32)kvm_register_read(vcpu, VCPU_REGS_RDX), handler);
3281 EXPORT_SYMBOL_GPL(kvm_emulate_cpuid);
3284 * Check if userspace requested an interrupt window, and that the
3285 * interrupt window is open.
3287 * No need to exit to userspace if we already have an interrupt queued.
3289 static int dm_request_for_irq_injection(struct kvm_vcpu *vcpu,
3290 struct kvm_run *kvm_run)
3292 return (!irqchip_in_kernel(vcpu->kvm) && !kvm_cpu_has_interrupt(vcpu) &&
3293 kvm_run->request_interrupt_window &&
3294 kvm_arch_interrupt_allowed(vcpu));
3297 static void post_kvm_run_save(struct kvm_vcpu *vcpu,
3298 struct kvm_run *kvm_run)
3300 kvm_run->if_flag = (kvm_x86_ops->get_rflags(vcpu) & X86_EFLAGS_IF) != 0;
3301 kvm_run->cr8 = kvm_get_cr8(vcpu);
3302 kvm_run->apic_base = kvm_get_apic_base(vcpu);
3303 if (irqchip_in_kernel(vcpu->kvm))
3304 kvm_run->ready_for_interrupt_injection = 1;
3306 kvm_run->ready_for_interrupt_injection =
3307 kvm_arch_interrupt_allowed(vcpu) &&
3308 !kvm_cpu_has_interrupt(vcpu) &&
3309 !kvm_event_needs_reinjection(vcpu);
3312 static void vapic_enter(struct kvm_vcpu *vcpu)
3314 struct kvm_lapic *apic = vcpu->arch.apic;
3317 if (!apic || !apic->vapic_addr)
3320 page = gfn_to_page(vcpu->kvm, apic->vapic_addr >> PAGE_SHIFT);
3322 vcpu->arch.apic->vapic_page = page;
3325 static void vapic_exit(struct kvm_vcpu *vcpu)
3327 struct kvm_lapic *apic = vcpu->arch.apic;
3329 if (!apic || !apic->vapic_addr)
3332 down_read(&vcpu->kvm->slots_lock);
3333 kvm_release_page_dirty(apic->vapic_page);
3334 mark_page_dirty(vcpu->kvm, apic->vapic_addr >> PAGE_SHIFT);
3335 up_read(&vcpu->kvm->slots_lock);
3338 static void update_cr8_intercept(struct kvm_vcpu *vcpu)
3342 if (!kvm_x86_ops->update_cr8_intercept)
3345 if (!vcpu->arch.apic->vapic_addr)
3346 max_irr = kvm_lapic_find_highest_irr(vcpu);
3353 tpr = kvm_lapic_get_cr8(vcpu);
3355 kvm_x86_ops->update_cr8_intercept(vcpu, tpr, max_irr);
3358 static void inject_pending_irq(struct kvm_vcpu *vcpu, struct kvm_run *kvm_run)
3360 if (vcpu->guest_debug & KVM_GUESTDBG_SINGLESTEP)
3361 kvm_x86_ops->set_interrupt_shadow(vcpu, 0);
3363 /* try to reinject previous events if any */
3364 if (vcpu->arch.nmi_injected) {
3365 kvm_x86_ops->set_nmi(vcpu);
3369 if (vcpu->arch.interrupt.pending) {
3370 kvm_x86_ops->set_irq(vcpu);
3374 /* try to inject new event if pending */
3375 if (vcpu->arch.nmi_pending) {
3376 if (kvm_x86_ops->nmi_allowed(vcpu)) {
3377 vcpu->arch.nmi_pending = false;
3378 vcpu->arch.nmi_injected = true;
3379 kvm_x86_ops->set_nmi(vcpu);
3381 } else if (kvm_cpu_has_interrupt(vcpu)) {
3382 if (kvm_x86_ops->interrupt_allowed(vcpu)) {
3383 kvm_queue_interrupt(vcpu, kvm_cpu_get_interrupt(vcpu),
3385 kvm_x86_ops->set_irq(vcpu);
3390 static int vcpu_enter_guest(struct kvm_vcpu *vcpu, struct kvm_run *kvm_run)
3393 bool req_int_win = !irqchip_in_kernel(vcpu->kvm) &&
3394 kvm_run->request_interrupt_window;
3397 if (test_and_clear_bit(KVM_REQ_MMU_RELOAD, &vcpu->requests))
3398 kvm_mmu_unload(vcpu);
3400 r = kvm_mmu_reload(vcpu);
3404 if (vcpu->requests) {
3405 if (test_and_clear_bit(KVM_REQ_MIGRATE_TIMER, &vcpu->requests))
3406 __kvm_migrate_timers(vcpu);
3407 if (test_and_clear_bit(KVM_REQ_KVMCLOCK_UPDATE, &vcpu->requests))
3408 kvm_write_guest_time(vcpu);
3409 if (test_and_clear_bit(KVM_REQ_MMU_SYNC, &vcpu->requests))
3410 kvm_mmu_sync_roots(vcpu);
3411 if (test_and_clear_bit(KVM_REQ_TLB_FLUSH, &vcpu->requests))
3412 kvm_x86_ops->tlb_flush(vcpu);
3413 if (test_and_clear_bit(KVM_REQ_REPORT_TPR_ACCESS,
3415 kvm_run->exit_reason = KVM_EXIT_TPR_ACCESS;
3419 if (test_and_clear_bit(KVM_REQ_TRIPLE_FAULT, &vcpu->requests)) {
3420 kvm_run->exit_reason = KVM_EXIT_SHUTDOWN;
3428 kvm_x86_ops->prepare_guest_switch(vcpu);
3429 kvm_load_guest_fpu(vcpu);
3431 local_irq_disable();
3433 clear_bit(KVM_REQ_KICK, &vcpu->requests);
3434 smp_mb__after_clear_bit();
3436 if (vcpu->requests || need_resched() || signal_pending(current)) {
3443 if (vcpu->arch.exception.pending)
3444 __queue_exception(vcpu);
3446 inject_pending_irq(vcpu, kvm_run);
3448 /* enable NMI/IRQ window open exits if needed */
3449 if (vcpu->arch.nmi_pending)
3450 kvm_x86_ops->enable_nmi_window(vcpu);
3451 else if (kvm_cpu_has_interrupt(vcpu) || req_int_win)
3452 kvm_x86_ops->enable_irq_window(vcpu);
3454 if (kvm_lapic_enabled(vcpu)) {
3455 update_cr8_intercept(vcpu);
3456 kvm_lapic_sync_to_vapic(vcpu);
3459 up_read(&vcpu->kvm->slots_lock);
3463 get_debugreg(vcpu->arch.host_dr6, 6);
3464 get_debugreg(vcpu->arch.host_dr7, 7);
3465 if (unlikely(vcpu->arch.switch_db_regs)) {
3466 get_debugreg(vcpu->arch.host_db[0], 0);
3467 get_debugreg(vcpu->arch.host_db[1], 1);
3468 get_debugreg(vcpu->arch.host_db[2], 2);
3469 get_debugreg(vcpu->arch.host_db[3], 3);
3472 set_debugreg(vcpu->arch.eff_db[0], 0);
3473 set_debugreg(vcpu->arch.eff_db[1], 1);
3474 set_debugreg(vcpu->arch.eff_db[2], 2);
3475 set_debugreg(vcpu->arch.eff_db[3], 3);
3478 KVMTRACE_0D(VMENTRY, vcpu, entryexit);
3479 kvm_x86_ops->run(vcpu, kvm_run);
3481 if (unlikely(vcpu->arch.switch_db_regs)) {
3483 set_debugreg(vcpu->arch.host_db[0], 0);
3484 set_debugreg(vcpu->arch.host_db[1], 1);
3485 set_debugreg(vcpu->arch.host_db[2], 2);
3486 set_debugreg(vcpu->arch.host_db[3], 3);
3488 set_debugreg(vcpu->arch.host_dr6, 6);
3489 set_debugreg(vcpu->arch.host_dr7, 7);
3491 set_bit(KVM_REQ_KICK, &vcpu->requests);
3497 * We must have an instruction between local_irq_enable() and
3498 * kvm_guest_exit(), so the timer interrupt isn't delayed by
3499 * the interrupt shadow. The stat.exits increment will do nicely.
3500 * But we need to prevent reordering, hence this barrier():
3508 down_read(&vcpu->kvm->slots_lock);
3511 * Profile KVM exit RIPs:
3513 if (unlikely(prof_on == KVM_PROFILING)) {
3514 unsigned long rip = kvm_rip_read(vcpu);
3515 profile_hit(KVM_PROFILING, (void *)rip);
3519 kvm_lapic_sync_from_vapic(vcpu);
3521 r = kvm_x86_ops->handle_exit(kvm_run, vcpu);
3527 static int __vcpu_run(struct kvm_vcpu *vcpu, struct kvm_run *kvm_run)
3531 if (unlikely(vcpu->arch.mp_state == KVM_MP_STATE_SIPI_RECEIVED)) {
3532 pr_debug("vcpu %d received sipi with vector # %x\n",
3533 vcpu->vcpu_id, vcpu->arch.sipi_vector);
3534 kvm_lapic_reset(vcpu);
3535 r = kvm_arch_vcpu_reset(vcpu);
3538 vcpu->arch.mp_state = KVM_MP_STATE_RUNNABLE;
3541 down_read(&vcpu->kvm->slots_lock);
3546 if (vcpu->arch.mp_state == KVM_MP_STATE_RUNNABLE)
3547 r = vcpu_enter_guest(vcpu, kvm_run);
3549 up_read(&vcpu->kvm->slots_lock);
3550 kvm_vcpu_block(vcpu);
3551 down_read(&vcpu->kvm->slots_lock);
3552 if (test_and_clear_bit(KVM_REQ_UNHALT, &vcpu->requests))
3554 switch(vcpu->arch.mp_state) {
3555 case KVM_MP_STATE_HALTED:
3556 vcpu->arch.mp_state =
3557 KVM_MP_STATE_RUNNABLE;
3558 case KVM_MP_STATE_RUNNABLE:
3560 case KVM_MP_STATE_SIPI_RECEIVED:
3571 clear_bit(KVM_REQ_PENDING_TIMER, &vcpu->requests);
3572 if (kvm_cpu_has_pending_timer(vcpu))
3573 kvm_inject_pending_timer_irqs(vcpu);
3575 if (dm_request_for_irq_injection(vcpu, kvm_run)) {
3577 kvm_run->exit_reason = KVM_EXIT_INTR;
3578 ++vcpu->stat.request_irq_exits;
3580 if (signal_pending(current)) {
3582 kvm_run->exit_reason = KVM_EXIT_INTR;
3583 ++vcpu->stat.signal_exits;
3585 if (need_resched()) {
3586 up_read(&vcpu->kvm->slots_lock);
3588 down_read(&vcpu->kvm->slots_lock);
3592 up_read(&vcpu->kvm->slots_lock);
3593 post_kvm_run_save(vcpu, kvm_run);
3600 int kvm_arch_vcpu_ioctl_run(struct kvm_vcpu *vcpu, struct kvm_run *kvm_run)
3607 if (vcpu->sigset_active)
3608 sigprocmask(SIG_SETMASK, &vcpu->sigset, &sigsaved);
3610 if (unlikely(vcpu->arch.mp_state == KVM_MP_STATE_UNINITIALIZED)) {
3611 kvm_vcpu_block(vcpu);
3612 clear_bit(KVM_REQ_UNHALT, &vcpu->requests);
3617 /* re-sync apic's tpr */
3618 if (!irqchip_in_kernel(vcpu->kvm))
3619 kvm_set_cr8(vcpu, kvm_run->cr8);
3621 if (vcpu->arch.pio.cur_count) {
3622 r = complete_pio(vcpu);
3626 #if CONFIG_HAS_IOMEM
3627 if (vcpu->mmio_needed) {
3628 memcpy(vcpu->mmio_data, kvm_run->mmio.data, 8);
3629 vcpu->mmio_read_completed = 1;
3630 vcpu->mmio_needed = 0;
3632 down_read(&vcpu->kvm->slots_lock);
3633 r = emulate_instruction(vcpu, kvm_run,
3634 vcpu->arch.mmio_fault_cr2, 0,
3635 EMULTYPE_NO_DECODE);
3636 up_read(&vcpu->kvm->slots_lock);
3637 if (r == EMULATE_DO_MMIO) {
3639 * Read-modify-write. Back to userspace.
3646 if (kvm_run->exit_reason == KVM_EXIT_HYPERCALL)
3647 kvm_register_write(vcpu, VCPU_REGS_RAX,
3648 kvm_run->hypercall.ret);
3650 r = __vcpu_run(vcpu, kvm_run);
3653 if (vcpu->sigset_active)
3654 sigprocmask(SIG_SETMASK, &sigsaved, NULL);
3660 int kvm_arch_vcpu_ioctl_get_regs(struct kvm_vcpu *vcpu, struct kvm_regs *regs)
3664 regs->rax = kvm_register_read(vcpu, VCPU_REGS_RAX);
3665 regs->rbx = kvm_register_read(vcpu, VCPU_REGS_RBX);
3666 regs->rcx = kvm_register_read(vcpu, VCPU_REGS_RCX);
3667 regs->rdx = kvm_register_read(vcpu, VCPU_REGS_RDX);
3668 regs->rsi = kvm_register_read(vcpu, VCPU_REGS_RSI);
3669 regs->rdi = kvm_register_read(vcpu, VCPU_REGS_RDI);
3670 regs->rsp = kvm_register_read(vcpu, VCPU_REGS_RSP);
3671 regs->rbp = kvm_register_read(vcpu, VCPU_REGS_RBP);
3672 #ifdef CONFIG_X86_64
3673 regs->r8 = kvm_register_read(vcpu, VCPU_REGS_R8);
3674 regs->r9 = kvm_register_read(vcpu, VCPU_REGS_R9);
3675 regs->r10 = kvm_register_read(vcpu, VCPU_REGS_R10);
3676 regs->r11 = kvm_register_read(vcpu, VCPU_REGS_R11);
3677 regs->r12 = kvm_register_read(vcpu, VCPU_REGS_R12);
3678 regs->r13 = kvm_register_read(vcpu, VCPU_REGS_R13);
3679 regs->r14 = kvm_register_read(vcpu, VCPU_REGS_R14);
3680 regs->r15 = kvm_register_read(vcpu, VCPU_REGS_R15);
3683 regs->rip = kvm_rip_read(vcpu);
3684 regs->rflags = kvm_x86_ops->get_rflags(vcpu);
3687 * Don't leak debug flags in case they were set for guest debugging
3689 if (vcpu->guest_debug & KVM_GUESTDBG_SINGLESTEP)
3690 regs->rflags &= ~(X86_EFLAGS_TF | X86_EFLAGS_RF);
3697 int kvm_arch_vcpu_ioctl_set_regs(struct kvm_vcpu *vcpu, struct kvm_regs *regs)
3701 kvm_register_write(vcpu, VCPU_REGS_RAX, regs->rax);
3702 kvm_register_write(vcpu, VCPU_REGS_RBX, regs->rbx);
3703 kvm_register_write(vcpu, VCPU_REGS_RCX, regs->rcx);
3704 kvm_register_write(vcpu, VCPU_REGS_RDX, regs->rdx);
3705 kvm_register_write(vcpu, VCPU_REGS_RSI, regs->rsi);
3706 kvm_register_write(vcpu, VCPU_REGS_RDI, regs->rdi);
3707 kvm_register_write(vcpu, VCPU_REGS_RSP, regs->rsp);
3708 kvm_register_write(vcpu, VCPU_REGS_RBP, regs->rbp);
3709 #ifdef CONFIG_X86_64
3710 kvm_register_write(vcpu, VCPU_REGS_R8, regs->r8);
3711 kvm_register_write(vcpu, VCPU_REGS_R9, regs->r9);
3712 kvm_register_write(vcpu, VCPU_REGS_R10, regs->r10);
3713 kvm_register_write(vcpu, VCPU_REGS_R11, regs->r11);
3714 kvm_register_write(vcpu, VCPU_REGS_R12, regs->r12);
3715 kvm_register_write(vcpu, VCPU_REGS_R13, regs->r13);
3716 kvm_register_write(vcpu, VCPU_REGS_R14, regs->r14);
3717 kvm_register_write(vcpu, VCPU_REGS_R15, regs->r15);
3721 kvm_rip_write(vcpu, regs->rip);
3722 kvm_x86_ops->set_rflags(vcpu, regs->rflags);
3725 vcpu->arch.exception.pending = false;
3732 void kvm_get_segment(struct kvm_vcpu *vcpu,
3733 struct kvm_segment *var, int seg)
3735 kvm_x86_ops->get_segment(vcpu, var, seg);
3738 void kvm_get_cs_db_l_bits(struct kvm_vcpu *vcpu, int *db, int *l)
3740 struct kvm_segment cs;
3742 kvm_get_segment(vcpu, &cs, VCPU_SREG_CS);
3746 EXPORT_SYMBOL_GPL(kvm_get_cs_db_l_bits);
3748 int kvm_arch_vcpu_ioctl_get_sregs(struct kvm_vcpu *vcpu,
3749 struct kvm_sregs *sregs)
3751 struct descriptor_table dt;
3755 kvm_get_segment(vcpu, &sregs->cs, VCPU_SREG_CS);
3756 kvm_get_segment(vcpu, &sregs->ds, VCPU_SREG_DS);
3757 kvm_get_segment(vcpu, &sregs->es, VCPU_SREG_ES);
3758 kvm_get_segment(vcpu, &sregs->fs, VCPU_SREG_FS);
3759 kvm_get_segment(vcpu, &sregs->gs, VCPU_SREG_GS);
3760 kvm_get_segment(vcpu, &sregs->ss, VCPU_SREG_SS);
3762 kvm_get_segment(vcpu, &sregs->tr, VCPU_SREG_TR);
3763 kvm_get_segment(vcpu, &sregs->ldt, VCPU_SREG_LDTR);
3765 kvm_x86_ops->get_idt(vcpu, &dt);
3766 sregs->idt.limit = dt.limit;
3767 sregs->idt.base = dt.base;
3768 kvm_x86_ops->get_gdt(vcpu, &dt);
3769 sregs->gdt.limit = dt.limit;
3770 sregs->gdt.base = dt.base;
3772 kvm_x86_ops->decache_cr4_guest_bits(vcpu);
3773 sregs->cr0 = vcpu->arch.cr0;
3774 sregs->cr2 = vcpu->arch.cr2;
3775 sregs->cr3 = vcpu->arch.cr3;
3776 sregs->cr4 = vcpu->arch.cr4;
3777 sregs->cr8 = kvm_get_cr8(vcpu);
3778 sregs->efer = vcpu->arch.shadow_efer;
3779 sregs->apic_base = kvm_get_apic_base(vcpu);
3781 memset(sregs->interrupt_bitmap, 0, sizeof sregs->interrupt_bitmap);
3783 if (vcpu->arch.interrupt.pending && !vcpu->arch.interrupt.soft)
3784 set_bit(vcpu->arch.interrupt.nr,
3785 (unsigned long *)sregs->interrupt_bitmap);
3792 int kvm_arch_vcpu_ioctl_get_mpstate(struct kvm_vcpu *vcpu,
3793 struct kvm_mp_state *mp_state)
3796 mp_state->mp_state = vcpu->arch.mp_state;
3801 int kvm_arch_vcpu_ioctl_set_mpstate(struct kvm_vcpu *vcpu,
3802 struct kvm_mp_state *mp_state)
3805 vcpu->arch.mp_state = mp_state->mp_state;
3810 static void kvm_set_segment(struct kvm_vcpu *vcpu,
3811 struct kvm_segment *var, int seg)
3813 kvm_x86_ops->set_segment(vcpu, var, seg);
3816 static void seg_desct_to_kvm_desct(struct desc_struct *seg_desc, u16 selector,
3817 struct kvm_segment *kvm_desct)
3819 kvm_desct->base = seg_desc->base0;
3820 kvm_desct->base |= seg_desc->base1 << 16;
3821 kvm_desct->base |= seg_desc->base2 << 24;
3822 kvm_desct->limit = seg_desc->limit0;
3823 kvm_desct->limit |= seg_desc->limit << 16;
3825 kvm_desct->limit <<= 12;
3826 kvm_desct->limit |= 0xfff;
3828 kvm_desct->selector = selector;
3829 kvm_desct->type = seg_desc->type;
3830 kvm_desct->present = seg_desc->p;
3831 kvm_desct->dpl = seg_desc->dpl;
3832 kvm_desct->db = seg_desc->d;
3833 kvm_desct->s = seg_desc->s;
3834 kvm_desct->l = seg_desc->l;
3835 kvm_desct->g = seg_desc->g;
3836 kvm_desct->avl = seg_desc->avl;
3838 kvm_desct->unusable = 1;
3840 kvm_desct->unusable = 0;
3841 kvm_desct->padding = 0;
3844 static void get_segment_descriptor_dtable(struct kvm_vcpu *vcpu,
3846 struct descriptor_table *dtable)
3848 if (selector & 1 << 2) {
3849 struct kvm_segment kvm_seg;
3851 kvm_get_segment(vcpu, &kvm_seg, VCPU_SREG_LDTR);
3853 if (kvm_seg.unusable)
3856 dtable->limit = kvm_seg.limit;
3857 dtable->base = kvm_seg.base;
3860 kvm_x86_ops->get_gdt(vcpu, dtable);
3863 /* allowed just for 8 bytes segments */
3864 static int load_guest_segment_descriptor(struct kvm_vcpu *vcpu, u16 selector,
3865 struct desc_struct *seg_desc)
3868 struct descriptor_table dtable;
3869 u16 index = selector >> 3;
3871 get_segment_descriptor_dtable(vcpu, selector, &dtable);
3873 if (dtable.limit < index * 8 + 7) {
3874 kvm_queue_exception_e(vcpu, GP_VECTOR, selector & 0xfffc);
3877 gpa = vcpu->arch.mmu.gva_to_gpa(vcpu, dtable.base);
3879 return kvm_read_guest(vcpu->kvm, gpa, seg_desc, 8);
3882 /* allowed just for 8 bytes segments */
3883 static int save_guest_segment_descriptor(struct kvm_vcpu *vcpu, u16 selector,
3884 struct desc_struct *seg_desc)
3887 struct descriptor_table dtable;
3888 u16 index = selector >> 3;
3890 get_segment_descriptor_dtable(vcpu, selector, &dtable);
3892 if (dtable.limit < index * 8 + 7)
3894 gpa = vcpu->arch.mmu.gva_to_gpa(vcpu, dtable.base);
3896 return kvm_write_guest(vcpu->kvm, gpa, seg_desc, 8);
3899 static u32 get_tss_base_addr(struct kvm_vcpu *vcpu,
3900 struct desc_struct *seg_desc)
3904 base_addr = seg_desc->base0;
3905 base_addr |= (seg_desc->base1 << 16);
3906 base_addr |= (seg_desc->base2 << 24);
3908 return vcpu->arch.mmu.gva_to_gpa(vcpu, base_addr);
3911 static u16 get_segment_selector(struct kvm_vcpu *vcpu, int seg)
3913 struct kvm_segment kvm_seg;
3915 kvm_get_segment(vcpu, &kvm_seg, seg);
3916 return kvm_seg.selector;
3919 static int load_segment_descriptor_to_kvm_desct(struct kvm_vcpu *vcpu,
3921 struct kvm_segment *kvm_seg)
3923 struct desc_struct seg_desc;
3925 if (load_guest_segment_descriptor(vcpu, selector, &seg_desc))
3927 seg_desct_to_kvm_desct(&seg_desc, selector, kvm_seg);
3931 static int kvm_load_realmode_segment(struct kvm_vcpu *vcpu, u16 selector, int seg)
3933 struct kvm_segment segvar = {
3934 .base = selector << 4,
3936 .selector = selector,
3947 kvm_x86_ops->set_segment(vcpu, &segvar, seg);
3951 int kvm_load_segment_descriptor(struct kvm_vcpu *vcpu, u16 selector,
3952 int type_bits, int seg)
3954 struct kvm_segment kvm_seg;
3956 if (!(vcpu->arch.cr0 & X86_CR0_PE))
3957 return kvm_load_realmode_segment(vcpu, selector, seg);
3958 if (load_segment_descriptor_to_kvm_desct(vcpu, selector, &kvm_seg))
3960 kvm_seg.type |= type_bits;
3962 if (seg != VCPU_SREG_SS && seg != VCPU_SREG_CS &&
3963 seg != VCPU_SREG_LDTR)
3965 kvm_seg.unusable = 1;
3967 kvm_set_segment(vcpu, &kvm_seg, seg);
3971 static void save_state_to_tss32(struct kvm_vcpu *vcpu,
3972 struct tss_segment_32 *tss)
3974 tss->cr3 = vcpu->arch.cr3;
3975 tss->eip = kvm_rip_read(vcpu);
3976 tss->eflags = kvm_x86_ops->get_rflags(vcpu);
3977 tss->eax = kvm_register_read(vcpu, VCPU_REGS_RAX);
3978 tss->ecx = kvm_register_read(vcpu, VCPU_REGS_RCX);
3979 tss->edx = kvm_register_read(vcpu, VCPU_REGS_RDX);
3980 tss->ebx = kvm_register_read(vcpu, VCPU_REGS_RBX);
3981 tss->esp = kvm_register_read(vcpu, VCPU_REGS_RSP);
3982 tss->ebp = kvm_register_read(vcpu, VCPU_REGS_RBP);
3983 tss->esi = kvm_register_read(vcpu, VCPU_REGS_RSI);
3984 tss->edi = kvm_register_read(vcpu, VCPU_REGS_RDI);
3985 tss->es = get_segment_selector(vcpu, VCPU_SREG_ES);
3986 tss->cs = get_segment_selector(vcpu, VCPU_SREG_CS);
3987 tss->ss = get_segment_selector(vcpu, VCPU_SREG_SS);
3988 tss->ds = get_segment_selector(vcpu, VCPU_SREG_DS);
3989 tss->fs = get_segment_selector(vcpu, VCPU_SREG_FS);
3990 tss->gs = get_segment_selector(vcpu, VCPU_SREG_GS);
3991 tss->ldt_selector = get_segment_selector(vcpu, VCPU_SREG_LDTR);
3994 static int load_state_from_tss32(struct kvm_vcpu *vcpu,
3995 struct tss_segment_32 *tss)
3997 kvm_set_cr3(vcpu, tss->cr3);
3999 kvm_rip_write(vcpu, tss->eip);
4000 kvm_x86_ops->set_rflags(vcpu, tss->eflags | 2);
4002 kvm_register_write(vcpu, VCPU_REGS_RAX, tss->eax);
4003 kvm_register_write(vcpu, VCPU_REGS_RCX, tss->ecx);
4004 kvm_register_write(vcpu, VCPU_REGS_RDX, tss->edx);
4005 kvm_register_write(vcpu, VCPU_REGS_RBX, tss->ebx);
4006 kvm_register_write(vcpu, VCPU_REGS_RSP, tss->esp);
4007 kvm_register_write(vcpu, VCPU_REGS_RBP, tss->ebp);
4008 kvm_register_write(vcpu, VCPU_REGS_RSI, tss->esi);
4009 kvm_register_write(vcpu, VCPU_REGS_RDI, tss->edi);
4011 if (kvm_load_segment_descriptor(vcpu, tss->ldt_selector, 0, VCPU_SREG_LDTR))
4014 if (kvm_load_segment_descriptor(vcpu, tss->es, 1, VCPU_SREG_ES))
4017 if (kvm_load_segment_descriptor(vcpu, tss->cs, 9, VCPU_SREG_CS))
4020 if (kvm_load_segment_descriptor(vcpu, tss->ss, 1, VCPU_SREG_SS))
4023 if (kvm_load_segment_descriptor(vcpu, tss->ds, 1, VCPU_SREG_DS))
4026 if (kvm_load_segment_descriptor(vcpu, tss->fs, 1, VCPU_SREG_FS))
4029 if (kvm_load_segment_descriptor(vcpu, tss->gs, 1, VCPU_SREG_GS))
4034 static void save_state_to_tss16(struct kvm_vcpu *vcpu,
4035 struct tss_segment_16 *tss)
4037 tss->ip = kvm_rip_read(vcpu);
4038 tss->flag = kvm_x86_ops->get_rflags(vcpu);
4039 tss->ax = kvm_register_read(vcpu, VCPU_REGS_RAX);
4040 tss->cx = kvm_register_read(vcpu, VCPU_REGS_RCX);
4041 tss->dx = kvm_register_read(vcpu, VCPU_REGS_RDX);
4042 tss->bx = kvm_register_read(vcpu, VCPU_REGS_RBX);
4043 tss->sp = kvm_register_read(vcpu, VCPU_REGS_RSP);
4044 tss->bp = kvm_register_read(vcpu, VCPU_REGS_RBP);
4045 tss->si = kvm_register_read(vcpu, VCPU_REGS_RSI);
4046 tss->di = kvm_register_read(vcpu, VCPU_REGS_RDI);
4048 tss->es = get_segment_selector(vcpu, VCPU_SREG_ES);
4049 tss->cs = get_segment_selector(vcpu, VCPU_SREG_CS);
4050 tss->ss = get_segment_selector(vcpu, VCPU_SREG_SS);
4051 tss->ds = get_segment_selector(vcpu, VCPU_SREG_DS);
4052 tss->ldt = get_segment_selector(vcpu, VCPU_SREG_LDTR);
4053 tss->prev_task_link = get_segment_selector(vcpu, VCPU_SREG_TR);
4056 static int load_state_from_tss16(struct kvm_vcpu *vcpu,
4057 struct tss_segment_16 *tss)
4059 kvm_rip_write(vcpu, tss->ip);
4060 kvm_x86_ops->set_rflags(vcpu, tss->flag | 2);
4061 kvm_register_write(vcpu, VCPU_REGS_RAX, tss->ax);
4062 kvm_register_write(vcpu, VCPU_REGS_RCX, tss->cx);
4063 kvm_register_write(vcpu, VCPU_REGS_RDX, tss->dx);
4064 kvm_register_write(vcpu, VCPU_REGS_RBX, tss->bx);
4065 kvm_register_write(vcpu, VCPU_REGS_RSP, tss->sp);
4066 kvm_register_write(vcpu, VCPU_REGS_RBP, tss->bp);
4067 kvm_register_write(vcpu, VCPU_REGS_RSI, tss->si);
4068 kvm_register_write(vcpu, VCPU_REGS_RDI, tss->di);
4070 if (kvm_load_segment_descriptor(vcpu, tss->ldt, 0, VCPU_SREG_LDTR))
4073 if (kvm_load_segment_descriptor(vcpu, tss->es, 1, VCPU_SREG_ES))
4076 if (kvm_load_segment_descriptor(vcpu, tss->cs, 9, VCPU_SREG_CS))
4079 if (kvm_load_segment_descriptor(vcpu, tss->ss, 1, VCPU_SREG_SS))
4082 if (kvm_load_segment_descriptor(vcpu, tss->ds, 1, VCPU_SREG_DS))
4087 static int kvm_task_switch_16(struct kvm_vcpu *vcpu, u16 tss_selector,
4088 u16 old_tss_sel, u32 old_tss_base,
4089 struct desc_struct *nseg_desc)
4091 struct tss_segment_16 tss_segment_16;
4094 if (kvm_read_guest(vcpu->kvm, old_tss_base, &tss_segment_16,
4095 sizeof tss_segment_16))
4098 save_state_to_tss16(vcpu, &tss_segment_16);
4100 if (kvm_write_guest(vcpu->kvm, old_tss_base, &tss_segment_16,
4101 sizeof tss_segment_16))
4104 if (kvm_read_guest(vcpu->kvm, get_tss_base_addr(vcpu, nseg_desc),
4105 &tss_segment_16, sizeof tss_segment_16))
4108 if (old_tss_sel != 0xffff) {
4109 tss_segment_16.prev_task_link = old_tss_sel;
4111 if (kvm_write_guest(vcpu->kvm,
4112 get_tss_base_addr(vcpu, nseg_desc),
4113 &tss_segment_16.prev_task_link,
4114 sizeof tss_segment_16.prev_task_link))
4118 if (load_state_from_tss16(vcpu, &tss_segment_16))
4126 static int kvm_task_switch_32(struct kvm_vcpu *vcpu, u16 tss_selector,
4127 u16 old_tss_sel, u32 old_tss_base,
4128 struct desc_struct *nseg_desc)
4130 struct tss_segment_32 tss_segment_32;
4133 if (kvm_read_guest(vcpu->kvm, old_tss_base, &tss_segment_32,
4134 sizeof tss_segment_32))
4137 save_state_to_tss32(vcpu, &tss_segment_32);
4139 if (kvm_write_guest(vcpu->kvm, old_tss_base, &tss_segment_32,
4140 sizeof tss_segment_32))
4143 if (kvm_read_guest(vcpu->kvm, get_tss_base_addr(vcpu, nseg_desc),
4144 &tss_segment_32, sizeof tss_segment_32))
4147 if (old_tss_sel != 0xffff) {
4148 tss_segment_32.prev_task_link = old_tss_sel;
4150 if (kvm_write_guest(vcpu->kvm,
4151 get_tss_base_addr(vcpu, nseg_desc),
4152 &tss_segment_32.prev_task_link,
4153 sizeof tss_segment_32.prev_task_link))
4157 if (load_state_from_tss32(vcpu, &tss_segment_32))
4165 int kvm_task_switch(struct kvm_vcpu *vcpu, u16 tss_selector, int reason)
4167 struct kvm_segment tr_seg;
4168 struct desc_struct cseg_desc;
4169 struct desc_struct nseg_desc;
4171 u32 old_tss_base = get_segment_base(vcpu, VCPU_SREG_TR);
4172 u16 old_tss_sel = get_segment_selector(vcpu, VCPU_SREG_TR);
4174 old_tss_base = vcpu->arch.mmu.gva_to_gpa(vcpu, old_tss_base);
4176 /* FIXME: Handle errors. Failure to read either TSS or their
4177 * descriptors should generate a pagefault.
4179 if (load_guest_segment_descriptor(vcpu, tss_selector, &nseg_desc))
4182 if (load_guest_segment_descriptor(vcpu, old_tss_sel, &cseg_desc))
4185 if (reason != TASK_SWITCH_IRET) {
4188 cpl = kvm_x86_ops->get_cpl(vcpu);
4189 if ((tss_selector & 3) > nseg_desc.dpl || cpl > nseg_desc.dpl) {
4190 kvm_queue_exception_e(vcpu, GP_VECTOR, 0);
4195 if (!nseg_desc.p || (nseg_desc.limit0 | nseg_desc.limit << 16) < 0x67) {
4196 kvm_queue_exception_e(vcpu, TS_VECTOR, tss_selector & 0xfffc);
4200 if (reason == TASK_SWITCH_IRET || reason == TASK_SWITCH_JMP) {
4201 cseg_desc.type &= ~(1 << 1); //clear the B flag
4202 save_guest_segment_descriptor(vcpu, old_tss_sel, &cseg_desc);
4205 if (reason == TASK_SWITCH_IRET) {
4206 u32 eflags = kvm_x86_ops->get_rflags(vcpu);
4207 kvm_x86_ops->set_rflags(vcpu, eflags & ~X86_EFLAGS_NT);
4210 /* set back link to prev task only if NT bit is set in eflags
4211 note that old_tss_sel is not used afetr this point */
4212 if (reason != TASK_SWITCH_CALL && reason != TASK_SWITCH_GATE)
4213 old_tss_sel = 0xffff;
4215 /* set back link to prev task only if NT bit is set in eflags
4216 note that old_tss_sel is not used afetr this point */
4217 if (reason != TASK_SWITCH_CALL && reason != TASK_SWITCH_GATE)
4218 old_tss_sel = 0xffff;
4220 if (nseg_desc.type & 8)
4221 ret = kvm_task_switch_32(vcpu, tss_selector, old_tss_sel,
4222 old_tss_base, &nseg_desc);
4224 ret = kvm_task_switch_16(vcpu, tss_selector, old_tss_sel,
4225 old_tss_base, &nseg_desc);
4227 if (reason == TASK_SWITCH_CALL || reason == TASK_SWITCH_GATE) {
4228 u32 eflags = kvm_x86_ops->get_rflags(vcpu);
4229 kvm_x86_ops->set_rflags(vcpu, eflags | X86_EFLAGS_NT);
4232 if (reason != TASK_SWITCH_IRET) {
4233 nseg_desc.type |= (1 << 1);
4234 save_guest_segment_descriptor(vcpu, tss_selector,
4238 kvm_x86_ops->set_cr0(vcpu, vcpu->arch.cr0 | X86_CR0_TS);
4239 seg_desct_to_kvm_desct(&nseg_desc, tss_selector, &tr_seg);
4241 kvm_set_segment(vcpu, &tr_seg, VCPU_SREG_TR);
4245 EXPORT_SYMBOL_GPL(kvm_task_switch);
4247 int kvm_arch_vcpu_ioctl_set_sregs(struct kvm_vcpu *vcpu,
4248 struct kvm_sregs *sregs)
4250 int mmu_reset_needed = 0;
4251 int pending_vec, max_bits;
4252 struct descriptor_table dt;
4256 dt.limit = sregs->idt.limit;
4257 dt.base = sregs->idt.base;
4258 kvm_x86_ops->set_idt(vcpu, &dt);
4259 dt.limit = sregs->gdt.limit;
4260 dt.base = sregs->gdt.base;
4261 kvm_x86_ops->set_gdt(vcpu, &dt);
4263 vcpu->arch.cr2 = sregs->cr2;
4264 mmu_reset_needed |= vcpu->arch.cr3 != sregs->cr3;
4266 down_read(&vcpu->kvm->slots_lock);
4267 if (gfn_to_memslot(vcpu->kvm, sregs->cr3 >> PAGE_SHIFT))
4268 vcpu->arch.cr3 = sregs->cr3;
4270 set_bit(KVM_REQ_TRIPLE_FAULT, &vcpu->requests);
4271 up_read(&vcpu->kvm->slots_lock);
4273 kvm_set_cr8(vcpu, sregs->cr8);
4275 mmu_reset_needed |= vcpu->arch.shadow_efer != sregs->efer;
4276 kvm_x86_ops->set_efer(vcpu, sregs->efer);
4277 kvm_set_apic_base(vcpu, sregs->apic_base);
4279 kvm_x86_ops->decache_cr4_guest_bits(vcpu);
4281 mmu_reset_needed |= vcpu->arch.cr0 != sregs->cr0;
4282 kvm_x86_ops->set_cr0(vcpu, sregs->cr0);
4283 vcpu->arch.cr0 = sregs->cr0;
4285 mmu_reset_needed |= vcpu->arch.cr4 != sregs->cr4;
4286 kvm_x86_ops->set_cr4(vcpu, sregs->cr4);
4287 if (!is_long_mode(vcpu) && is_pae(vcpu))
4288 load_pdptrs(vcpu, vcpu->arch.cr3);
4290 if (mmu_reset_needed)
4291 kvm_mmu_reset_context(vcpu);
4293 max_bits = (sizeof sregs->interrupt_bitmap) << 3;
4294 pending_vec = find_first_bit(
4295 (const unsigned long *)sregs->interrupt_bitmap, max_bits);
4296 if (pending_vec < max_bits) {
4297 kvm_queue_interrupt(vcpu, pending_vec, false);
4298 pr_debug("Set back pending irq %d\n", pending_vec);
4299 if (irqchip_in_kernel(vcpu->kvm))
4300 kvm_pic_clear_isr_ack(vcpu->kvm);
4303 kvm_set_segment(vcpu, &sregs->cs, VCPU_SREG_CS);
4304 kvm_set_segment(vcpu, &sregs->ds, VCPU_SREG_DS);
4305 kvm_set_segment(vcpu, &sregs->es, VCPU_SREG_ES);
4306 kvm_set_segment(vcpu, &sregs->fs, VCPU_SREG_FS);
4307 kvm_set_segment(vcpu, &sregs->gs, VCPU_SREG_GS);
4308 kvm_set_segment(vcpu, &sregs->ss, VCPU_SREG_SS);
4310 kvm_set_segment(vcpu, &sregs->tr, VCPU_SREG_TR);
4311 kvm_set_segment(vcpu, &sregs->ldt, VCPU_SREG_LDTR);
4313 /* Older userspace won't unhalt the vcpu on reset. */
4314 if (vcpu->vcpu_id == 0 && kvm_rip_read(vcpu) == 0xfff0 &&
4315 sregs->cs.selector == 0xf000 && sregs->cs.base == 0xffff0000 &&
4316 !(vcpu->arch.cr0 & X86_CR0_PE))
4317 vcpu->arch.mp_state = KVM_MP_STATE_RUNNABLE;
4324 int kvm_arch_vcpu_ioctl_set_guest_debug(struct kvm_vcpu *vcpu,
4325 struct kvm_guest_debug *dbg)
4331 if ((dbg->control & (KVM_GUESTDBG_ENABLE | KVM_GUESTDBG_USE_HW_BP)) ==
4332 (KVM_GUESTDBG_ENABLE | KVM_GUESTDBG_USE_HW_BP)) {
4333 for (i = 0; i < KVM_NR_DB_REGS; ++i)
4334 vcpu->arch.eff_db[i] = dbg->arch.debugreg[i];
4335 vcpu->arch.switch_db_regs =
4336 (dbg->arch.debugreg[7] & DR7_BP_EN_MASK);
4338 for (i = 0; i < KVM_NR_DB_REGS; i++)
4339 vcpu->arch.eff_db[i] = vcpu->arch.db[i];
4340 vcpu->arch.switch_db_regs = (vcpu->arch.dr7 & DR7_BP_EN_MASK);
4343 r = kvm_x86_ops->set_guest_debug(vcpu, dbg);
4345 if (dbg->control & KVM_GUESTDBG_INJECT_DB)
4346 kvm_queue_exception(vcpu, DB_VECTOR);
4347 else if (dbg->control & KVM_GUESTDBG_INJECT_BP)
4348 kvm_queue_exception(vcpu, BP_VECTOR);
4356 * fxsave fpu state. Taken from x86_64/processor.h. To be killed when
4357 * we have asm/x86/processor.h
4368 u32 st_space[32]; /* 8*16 bytes for each FP-reg = 128 bytes */
4369 #ifdef CONFIG_X86_64
4370 u32 xmm_space[64]; /* 16*16 bytes for each XMM-reg = 256 bytes */
4372 u32 xmm_space[32]; /* 8*16 bytes for each XMM-reg = 128 bytes */
4377 * Translate a guest virtual address to a guest physical address.
4379 int kvm_arch_vcpu_ioctl_translate(struct kvm_vcpu *vcpu,
4380 struct kvm_translation *tr)
4382 unsigned long vaddr = tr->linear_address;
4386 down_read(&vcpu->kvm->slots_lock);
4387 gpa = vcpu->arch.mmu.gva_to_gpa(vcpu, vaddr);
4388 up_read(&vcpu->kvm->slots_lock);
4389 tr->physical_address = gpa;
4390 tr->valid = gpa != UNMAPPED_GVA;
4398 int kvm_arch_vcpu_ioctl_get_fpu(struct kvm_vcpu *vcpu, struct kvm_fpu *fpu)
4400 struct fxsave *fxsave = (struct fxsave *)&vcpu->arch.guest_fx_image;
4404 memcpy(fpu->fpr, fxsave->st_space, 128);
4405 fpu->fcw = fxsave->cwd;
4406 fpu->fsw = fxsave->swd;
4407 fpu->ftwx = fxsave->twd;
4408 fpu->last_opcode = fxsave->fop;
4409 fpu->last_ip = fxsave->rip;
4410 fpu->last_dp = fxsave->rdp;
4411 memcpy(fpu->xmm, fxsave->xmm_space, sizeof fxsave->xmm_space);
4418 int kvm_arch_vcpu_ioctl_set_fpu(struct kvm_vcpu *vcpu, struct kvm_fpu *fpu)
4420 struct fxsave *fxsave = (struct fxsave *)&vcpu->arch.guest_fx_image;
4424 memcpy(fxsave->st_space, fpu->fpr, 128);
4425 fxsave->cwd = fpu->fcw;
4426 fxsave->swd = fpu->fsw;
4427 fxsave->twd = fpu->ftwx;
4428 fxsave->fop = fpu->last_opcode;
4429 fxsave->rip = fpu->last_ip;
4430 fxsave->rdp = fpu->last_dp;
4431 memcpy(fxsave->xmm_space, fpu->xmm, sizeof fxsave->xmm_space);
4438 void fx_init(struct kvm_vcpu *vcpu)
4440 unsigned after_mxcsr_mask;
4443 * Touch the fpu the first time in non atomic context as if
4444 * this is the first fpu instruction the exception handler
4445 * will fire before the instruction returns and it'll have to
4446 * allocate ram with GFP_KERNEL.
4449 kvm_fx_save(&vcpu->arch.host_fx_image);
4451 /* Initialize guest FPU by resetting ours and saving into guest's */
4453 kvm_fx_save(&vcpu->arch.host_fx_image);
4455 kvm_fx_save(&vcpu->arch.guest_fx_image);
4456 kvm_fx_restore(&vcpu->arch.host_fx_image);
4459 vcpu->arch.cr0 |= X86_CR0_ET;
4460 after_mxcsr_mask = offsetof(struct i387_fxsave_struct, st_space);
4461 vcpu->arch.guest_fx_image.mxcsr = 0x1f80;
4462 memset((void *)&vcpu->arch.guest_fx_image + after_mxcsr_mask,
4463 0, sizeof(struct i387_fxsave_struct) - after_mxcsr_mask);
4465 EXPORT_SYMBOL_GPL(fx_init);
4467 void kvm_load_guest_fpu(struct kvm_vcpu *vcpu)
4469 if (!vcpu->fpu_active || vcpu->guest_fpu_loaded)
4472 vcpu->guest_fpu_loaded = 1;
4473 kvm_fx_save(&vcpu->arch.host_fx_image);
4474 kvm_fx_restore(&vcpu->arch.guest_fx_image);
4476 EXPORT_SYMBOL_GPL(kvm_load_guest_fpu);
4478 void kvm_put_guest_fpu(struct kvm_vcpu *vcpu)
4480 if (!vcpu->guest_fpu_loaded)
4483 vcpu->guest_fpu_loaded = 0;
4484 kvm_fx_save(&vcpu->arch.guest_fx_image);
4485 kvm_fx_restore(&vcpu->arch.host_fx_image);
4486 ++vcpu->stat.fpu_reload;
4488 EXPORT_SYMBOL_GPL(kvm_put_guest_fpu);
4490 void kvm_arch_vcpu_free(struct kvm_vcpu *vcpu)
4492 if (vcpu->arch.time_page) {
4493 kvm_release_page_dirty(vcpu->arch.time_page);
4494 vcpu->arch.time_page = NULL;
4497 kvm_x86_ops->vcpu_free(vcpu);
4500 struct kvm_vcpu *kvm_arch_vcpu_create(struct kvm *kvm,
4503 return kvm_x86_ops->vcpu_create(kvm, id);
4506 int kvm_arch_vcpu_setup(struct kvm_vcpu *vcpu)
4510 /* We do fxsave: this must be aligned. */
4511 BUG_ON((unsigned long)&vcpu->arch.host_fx_image & 0xF);
4513 vcpu->arch.mtrr_state.have_fixed = 1;
4515 r = kvm_arch_vcpu_reset(vcpu);
4517 r = kvm_mmu_setup(vcpu);
4524 kvm_x86_ops->vcpu_free(vcpu);
4528 void kvm_arch_vcpu_destroy(struct kvm_vcpu *vcpu)
4531 kvm_mmu_unload(vcpu);
4534 kvm_x86_ops->vcpu_free(vcpu);
4537 int kvm_arch_vcpu_reset(struct kvm_vcpu *vcpu)
4539 vcpu->arch.nmi_pending = false;
4540 vcpu->arch.nmi_injected = false;
4542 vcpu->arch.switch_db_regs = 0;
4543 memset(vcpu->arch.db, 0, sizeof(vcpu->arch.db));
4544 vcpu->arch.dr6 = DR6_FIXED_1;
4545 vcpu->arch.dr7 = DR7_FIXED_1;
4547 return kvm_x86_ops->vcpu_reset(vcpu);
4550 void kvm_arch_hardware_enable(void *garbage)
4552 kvm_x86_ops->hardware_enable(garbage);
4555 void kvm_arch_hardware_disable(void *garbage)
4557 kvm_x86_ops->hardware_disable(garbage);
4560 int kvm_arch_hardware_setup(void)
4562 return kvm_x86_ops->hardware_setup();
4565 void kvm_arch_hardware_unsetup(void)
4567 kvm_x86_ops->hardware_unsetup();
4570 void kvm_arch_check_processor_compat(void *rtn)
4572 kvm_x86_ops->check_processor_compatibility(rtn);
4575 int kvm_arch_vcpu_init(struct kvm_vcpu *vcpu)
4581 BUG_ON(vcpu->kvm == NULL);
4584 vcpu->arch.mmu.root_hpa = INVALID_PAGE;
4585 if (!irqchip_in_kernel(kvm) || vcpu->vcpu_id == 0)
4586 vcpu->arch.mp_state = KVM_MP_STATE_RUNNABLE;
4588 vcpu->arch.mp_state = KVM_MP_STATE_UNINITIALIZED;
4590 page = alloc_page(GFP_KERNEL | __GFP_ZERO);
4595 vcpu->arch.pio_data = page_address(page);
4597 r = kvm_mmu_create(vcpu);
4599 goto fail_free_pio_data;
4601 if (irqchip_in_kernel(kvm)) {
4602 r = kvm_create_lapic(vcpu);
4604 goto fail_mmu_destroy;
4607 vcpu->arch.mce_banks = kzalloc(KVM_MAX_MCE_BANKS * sizeof(u64) * 4,
4609 if (!vcpu->arch.mce_banks) {
4611 goto fail_mmu_destroy;
4613 vcpu->arch.mcg_cap = KVM_MAX_MCE_BANKS;
4618 kvm_mmu_destroy(vcpu);
4620 free_page((unsigned long)vcpu->arch.pio_data);
4625 void kvm_arch_vcpu_uninit(struct kvm_vcpu *vcpu)
4627 kvm_free_lapic(vcpu);
4628 down_read(&vcpu->kvm->slots_lock);
4629 kvm_mmu_destroy(vcpu);
4630 up_read(&vcpu->kvm->slots_lock);
4631 free_page((unsigned long)vcpu->arch.pio_data);
4634 struct kvm *kvm_arch_create_vm(void)
4636 struct kvm *kvm = kzalloc(sizeof(struct kvm), GFP_KERNEL);
4639 return ERR_PTR(-ENOMEM);
4641 INIT_LIST_HEAD(&kvm->arch.active_mmu_pages);
4642 INIT_LIST_HEAD(&kvm->arch.assigned_dev_head);
4644 /* Reserve bit 0 of irq_sources_bitmap for userspace irq source */
4645 set_bit(KVM_USERSPACE_IRQ_SOURCE_ID, &kvm->arch.irq_sources_bitmap);
4647 rdtscll(kvm->arch.vm_init_tsc);
4652 static void kvm_unload_vcpu_mmu(struct kvm_vcpu *vcpu)
4655 kvm_mmu_unload(vcpu);
4659 static void kvm_free_vcpus(struct kvm *kvm)
4664 * Unpin any mmu pages first.
4666 for (i = 0; i < KVM_MAX_VCPUS; ++i)
4668 kvm_unload_vcpu_mmu(kvm->vcpus[i]);
4669 for (i = 0; i < KVM_MAX_VCPUS; ++i) {
4670 if (kvm->vcpus[i]) {
4671 kvm_arch_vcpu_free(kvm->vcpus[i]);
4672 kvm->vcpus[i] = NULL;
4678 void kvm_arch_sync_events(struct kvm *kvm)
4680 kvm_free_all_assigned_devices(kvm);
4683 void kvm_arch_destroy_vm(struct kvm *kvm)
4685 kvm_iommu_unmap_guest(kvm);
4687 kfree(kvm->arch.vpic);
4688 kfree(kvm->arch.vioapic);
4689 kvm_free_vcpus(kvm);
4690 kvm_free_physmem(kvm);
4691 if (kvm->arch.apic_access_page)
4692 put_page(kvm->arch.apic_access_page);
4693 if (kvm->arch.ept_identity_pagetable)
4694 put_page(kvm->arch.ept_identity_pagetable);
4698 int kvm_arch_set_memory_region(struct kvm *kvm,
4699 struct kvm_userspace_memory_region *mem,
4700 struct kvm_memory_slot old,
4703 int npages = mem->memory_size >> PAGE_SHIFT;
4704 struct kvm_memory_slot *memslot = &kvm->memslots[mem->slot];
4706 /*To keep backward compatibility with older userspace,
4707 *x86 needs to hanlde !user_alloc case.
4710 if (npages && !old.rmap) {
4711 unsigned long userspace_addr;
4713 down_write(¤t->mm->mmap_sem);
4714 userspace_addr = do_mmap(NULL, 0,
4716 PROT_READ | PROT_WRITE,
4717 MAP_PRIVATE | MAP_ANONYMOUS,
4719 up_write(¤t->mm->mmap_sem);
4721 if (IS_ERR((void *)userspace_addr))
4722 return PTR_ERR((void *)userspace_addr);
4724 /* set userspace_addr atomically for kvm_hva_to_rmapp */
4725 spin_lock(&kvm->mmu_lock);
4726 memslot->userspace_addr = userspace_addr;
4727 spin_unlock(&kvm->mmu_lock);
4729 if (!old.user_alloc && old.rmap) {
4732 down_write(¤t->mm->mmap_sem);
4733 ret = do_munmap(current->mm, old.userspace_addr,
4734 old.npages * PAGE_SIZE);
4735 up_write(¤t->mm->mmap_sem);
4738 "kvm_vm_ioctl_set_memory_region: "
4739 "failed to munmap memory\n");
4744 spin_lock(&kvm->mmu_lock);
4745 if (!kvm->arch.n_requested_mmu_pages) {
4746 unsigned int nr_mmu_pages = kvm_mmu_calculate_mmu_pages(kvm);
4747 kvm_mmu_change_mmu_pages(kvm, nr_mmu_pages);
4750 kvm_mmu_slot_remove_write_access(kvm, mem->slot);
4751 spin_unlock(&kvm->mmu_lock);
4752 kvm_flush_remote_tlbs(kvm);
4757 void kvm_arch_flush_shadow(struct kvm *kvm)
4759 kvm_mmu_zap_all(kvm);
4760 kvm_reload_remote_mmus(kvm);
4763 int kvm_arch_vcpu_runnable(struct kvm_vcpu *vcpu)
4765 return vcpu->arch.mp_state == KVM_MP_STATE_RUNNABLE
4766 || vcpu->arch.mp_state == KVM_MP_STATE_SIPI_RECEIVED
4767 || vcpu->arch.nmi_pending;
4770 void kvm_vcpu_kick(struct kvm_vcpu *vcpu)
4773 int cpu = vcpu->cpu;
4775 if (waitqueue_active(&vcpu->wq)) {
4776 wake_up_interruptible(&vcpu->wq);
4777 ++vcpu->stat.halt_wakeup;
4781 if (cpu != me && (unsigned)cpu < nr_cpu_ids && cpu_online(cpu))
4782 if (!test_and_set_bit(KVM_REQ_KICK, &vcpu->requests))
4783 smp_send_reschedule(cpu);
4787 int kvm_arch_interrupt_allowed(struct kvm_vcpu *vcpu)
4789 return kvm_x86_ops->interrupt_allowed(vcpu);