2 * Kernel-based Virtual Machine driver for Linux
4 * derived from drivers/kvm/kvm_main.c
6 * Copyright (C) 2006 Qumranet, Inc.
7 * Copyright (C) 2008 Qumranet, Inc.
8 * Copyright IBM Corporation, 2008
11 * Avi Kivity <avi@qumranet.com>
12 * Yaniv Kamay <yaniv@qumranet.com>
13 * Amit Shah <amit.shah@qumranet.com>
14 * Ben-Ami Yassour <benami@il.ibm.com>
16 * This work is licensed under the terms of the GNU GPL, version 2. See
17 * the COPYING file in the top-level directory.
21 #include <linux/kvm_host.h>
26 #include "kvm_cache_regs.h"
29 #include <linux/clocksource.h>
30 #include <linux/interrupt.h>
31 #include <linux/kvm.h>
33 #include <linux/vmalloc.h>
34 #include <linux/module.h>
35 #include <linux/mman.h>
36 #include <linux/highmem.h>
37 #include <linux/iommu.h>
38 #include <linux/intel-iommu.h>
39 #include <linux/cpufreq.h>
40 #include <trace/events/kvm.h>
41 #undef TRACE_INCLUDE_FILE
42 #define CREATE_TRACE_POINTS
45 #include <asm/uaccess.h>
51 #define MAX_IO_MSRS 256
52 #define CR0_RESERVED_BITS \
53 (~(unsigned long)(X86_CR0_PE | X86_CR0_MP | X86_CR0_EM | X86_CR0_TS \
54 | X86_CR0_ET | X86_CR0_NE | X86_CR0_WP | X86_CR0_AM \
55 | X86_CR0_NW | X86_CR0_CD | X86_CR0_PG))
56 #define CR4_RESERVED_BITS \
57 (~(unsigned long)(X86_CR4_VME | X86_CR4_PVI | X86_CR4_TSD | X86_CR4_DE\
58 | X86_CR4_PSE | X86_CR4_PAE | X86_CR4_MCE \
59 | X86_CR4_PGE | X86_CR4_PCE | X86_CR4_OSFXSR \
60 | X86_CR4_OSXMMEXCPT | X86_CR4_VMXE))
62 #define CR8_RESERVED_BITS (~(unsigned long)X86_CR8_TPR)
64 #define KVM_MAX_MCE_BANKS 32
65 #define KVM_MCE_CAP_SUPPORTED MCG_CTL_P
68 * - enable syscall per default because its emulated by KVM
69 * - enable LME and LMA per default on 64 bit KVM
72 static u64 __read_mostly efer_reserved_bits = 0xfffffffffffffafeULL;
74 static u64 __read_mostly efer_reserved_bits = 0xfffffffffffffffeULL;
77 #define VM_STAT(x) offsetof(struct kvm, stat.x), KVM_STAT_VM
78 #define VCPU_STAT(x) offsetof(struct kvm_vcpu, stat.x), KVM_STAT_VCPU
80 static void update_cr8_intercept(struct kvm_vcpu *vcpu);
81 static int kvm_dev_ioctl_get_supported_cpuid(struct kvm_cpuid2 *cpuid,
82 struct kvm_cpuid_entry2 __user *entries);
84 struct kvm_x86_ops *kvm_x86_ops;
85 EXPORT_SYMBOL_GPL(kvm_x86_ops);
88 module_param_named(ignore_msrs, ignore_msrs, bool, S_IRUGO | S_IWUSR);
90 struct kvm_stats_debugfs_item debugfs_entries[] = {
91 { "pf_fixed", VCPU_STAT(pf_fixed) },
92 { "pf_guest", VCPU_STAT(pf_guest) },
93 { "tlb_flush", VCPU_STAT(tlb_flush) },
94 { "invlpg", VCPU_STAT(invlpg) },
95 { "exits", VCPU_STAT(exits) },
96 { "io_exits", VCPU_STAT(io_exits) },
97 { "mmio_exits", VCPU_STAT(mmio_exits) },
98 { "signal_exits", VCPU_STAT(signal_exits) },
99 { "irq_window", VCPU_STAT(irq_window_exits) },
100 { "nmi_window", VCPU_STAT(nmi_window_exits) },
101 { "halt_exits", VCPU_STAT(halt_exits) },
102 { "halt_wakeup", VCPU_STAT(halt_wakeup) },
103 { "hypercalls", VCPU_STAT(hypercalls) },
104 { "request_irq", VCPU_STAT(request_irq_exits) },
105 { "irq_exits", VCPU_STAT(irq_exits) },
106 { "host_state_reload", VCPU_STAT(host_state_reload) },
107 { "efer_reload", VCPU_STAT(efer_reload) },
108 { "fpu_reload", VCPU_STAT(fpu_reload) },
109 { "insn_emulation", VCPU_STAT(insn_emulation) },
110 { "insn_emulation_fail", VCPU_STAT(insn_emulation_fail) },
111 { "irq_injections", VCPU_STAT(irq_injections) },
112 { "nmi_injections", VCPU_STAT(nmi_injections) },
113 { "mmu_shadow_zapped", VM_STAT(mmu_shadow_zapped) },
114 { "mmu_pte_write", VM_STAT(mmu_pte_write) },
115 { "mmu_pte_updated", VM_STAT(mmu_pte_updated) },
116 { "mmu_pde_zapped", VM_STAT(mmu_pde_zapped) },
117 { "mmu_flooded", VM_STAT(mmu_flooded) },
118 { "mmu_recycled", VM_STAT(mmu_recycled) },
119 { "mmu_cache_miss", VM_STAT(mmu_cache_miss) },
120 { "mmu_unsync", VM_STAT(mmu_unsync) },
121 { "remote_tlb_flush", VM_STAT(remote_tlb_flush) },
122 { "largepages", VM_STAT(lpages) },
126 unsigned long segment_base(u16 selector)
128 struct descriptor_table gdt;
129 struct desc_struct *d;
130 unsigned long table_base;
137 table_base = gdt.base;
139 if (selector & 4) { /* from ldt */
140 u16 ldt_selector = kvm_read_ldt();
142 table_base = segment_base(ldt_selector);
144 d = (struct desc_struct *)(table_base + (selector & ~7));
145 v = get_desc_base(d);
147 if (d->s == 0 && (d->type == 2 || d->type == 9 || d->type == 11))
148 v |= ((unsigned long)((struct ldttss_desc64 *)d)->base3) << 32;
152 EXPORT_SYMBOL_GPL(segment_base);
154 u64 kvm_get_apic_base(struct kvm_vcpu *vcpu)
156 if (irqchip_in_kernel(vcpu->kvm))
157 return vcpu->arch.apic_base;
159 return vcpu->arch.apic_base;
161 EXPORT_SYMBOL_GPL(kvm_get_apic_base);
163 void kvm_set_apic_base(struct kvm_vcpu *vcpu, u64 data)
165 /* TODO: reserve bits check */
166 if (irqchip_in_kernel(vcpu->kvm))
167 kvm_lapic_set_base(vcpu, data);
169 vcpu->arch.apic_base = data;
171 EXPORT_SYMBOL_GPL(kvm_set_apic_base);
173 void kvm_queue_exception(struct kvm_vcpu *vcpu, unsigned nr)
175 WARN_ON(vcpu->arch.exception.pending);
176 vcpu->arch.exception.pending = true;
177 vcpu->arch.exception.has_error_code = false;
178 vcpu->arch.exception.nr = nr;
180 EXPORT_SYMBOL_GPL(kvm_queue_exception);
182 void kvm_inject_page_fault(struct kvm_vcpu *vcpu, unsigned long addr,
185 ++vcpu->stat.pf_guest;
187 if (vcpu->arch.exception.pending) {
188 switch(vcpu->arch.exception.nr) {
190 /* triple fault -> shutdown */
191 set_bit(KVM_REQ_TRIPLE_FAULT, &vcpu->requests);
194 vcpu->arch.exception.nr = DF_VECTOR;
195 vcpu->arch.exception.error_code = 0;
198 /* replace previous exception with a new one in a hope
199 that instruction re-execution will regenerate lost
201 vcpu->arch.exception.pending = false;
205 vcpu->arch.cr2 = addr;
206 kvm_queue_exception_e(vcpu, PF_VECTOR, error_code);
209 void kvm_inject_nmi(struct kvm_vcpu *vcpu)
211 vcpu->arch.nmi_pending = 1;
213 EXPORT_SYMBOL_GPL(kvm_inject_nmi);
215 void kvm_queue_exception_e(struct kvm_vcpu *vcpu, unsigned nr, u32 error_code)
217 WARN_ON(vcpu->arch.exception.pending);
218 vcpu->arch.exception.pending = true;
219 vcpu->arch.exception.has_error_code = true;
220 vcpu->arch.exception.nr = nr;
221 vcpu->arch.exception.error_code = error_code;
223 EXPORT_SYMBOL_GPL(kvm_queue_exception_e);
226 * Checks if cpl <= required_cpl; if true, return true. Otherwise queue
227 * a #GP and return false.
229 bool kvm_require_cpl(struct kvm_vcpu *vcpu, int required_cpl)
231 if (kvm_x86_ops->get_cpl(vcpu) <= required_cpl)
233 kvm_queue_exception_e(vcpu, GP_VECTOR, 0);
236 EXPORT_SYMBOL_GPL(kvm_require_cpl);
239 * Load the pae pdptrs. Return true is they are all valid.
241 int load_pdptrs(struct kvm_vcpu *vcpu, unsigned long cr3)
243 gfn_t pdpt_gfn = cr3 >> PAGE_SHIFT;
244 unsigned offset = ((cr3 & (PAGE_SIZE-1)) >> 5) << 2;
247 u64 pdpte[ARRAY_SIZE(vcpu->arch.pdptrs)];
249 ret = kvm_read_guest_page(vcpu->kvm, pdpt_gfn, pdpte,
250 offset * sizeof(u64), sizeof(pdpte));
255 for (i = 0; i < ARRAY_SIZE(pdpte); ++i) {
256 if (is_present_gpte(pdpte[i]) &&
257 (pdpte[i] & vcpu->arch.mmu.rsvd_bits_mask[0][2])) {
264 memcpy(vcpu->arch.pdptrs, pdpte, sizeof(vcpu->arch.pdptrs));
265 __set_bit(VCPU_EXREG_PDPTR,
266 (unsigned long *)&vcpu->arch.regs_avail);
267 __set_bit(VCPU_EXREG_PDPTR,
268 (unsigned long *)&vcpu->arch.regs_dirty);
273 EXPORT_SYMBOL_GPL(load_pdptrs);
275 static bool pdptrs_changed(struct kvm_vcpu *vcpu)
277 u64 pdpte[ARRAY_SIZE(vcpu->arch.pdptrs)];
281 if (is_long_mode(vcpu) || !is_pae(vcpu))
284 if (!test_bit(VCPU_EXREG_PDPTR,
285 (unsigned long *)&vcpu->arch.regs_avail))
288 r = kvm_read_guest(vcpu->kvm, vcpu->arch.cr3 & ~31u, pdpte, sizeof(pdpte));
291 changed = memcmp(pdpte, vcpu->arch.pdptrs, sizeof(pdpte)) != 0;
297 void kvm_set_cr0(struct kvm_vcpu *vcpu, unsigned long cr0)
299 if (cr0 & CR0_RESERVED_BITS) {
300 printk(KERN_DEBUG "set_cr0: 0x%lx #GP, reserved bits 0x%lx\n",
301 cr0, vcpu->arch.cr0);
302 kvm_inject_gp(vcpu, 0);
306 if ((cr0 & X86_CR0_NW) && !(cr0 & X86_CR0_CD)) {
307 printk(KERN_DEBUG "set_cr0: #GP, CD == 0 && NW == 1\n");
308 kvm_inject_gp(vcpu, 0);
312 if ((cr0 & X86_CR0_PG) && !(cr0 & X86_CR0_PE)) {
313 printk(KERN_DEBUG "set_cr0: #GP, set PG flag "
314 "and a clear PE flag\n");
315 kvm_inject_gp(vcpu, 0);
319 if (!is_paging(vcpu) && (cr0 & X86_CR0_PG)) {
321 if ((vcpu->arch.shadow_efer & EFER_LME)) {
325 printk(KERN_DEBUG "set_cr0: #GP, start paging "
326 "in long mode while PAE is disabled\n");
327 kvm_inject_gp(vcpu, 0);
330 kvm_x86_ops->get_cs_db_l_bits(vcpu, &cs_db, &cs_l);
332 printk(KERN_DEBUG "set_cr0: #GP, start paging "
333 "in long mode while CS.L == 1\n");
334 kvm_inject_gp(vcpu, 0);
340 if (is_pae(vcpu) && !load_pdptrs(vcpu, vcpu->arch.cr3)) {
341 printk(KERN_DEBUG "set_cr0: #GP, pdptrs "
343 kvm_inject_gp(vcpu, 0);
349 kvm_x86_ops->set_cr0(vcpu, cr0);
350 vcpu->arch.cr0 = cr0;
352 kvm_mmu_reset_context(vcpu);
355 EXPORT_SYMBOL_GPL(kvm_set_cr0);
357 void kvm_lmsw(struct kvm_vcpu *vcpu, unsigned long msw)
359 kvm_set_cr0(vcpu, (vcpu->arch.cr0 & ~0x0ful) | (msw & 0x0f));
361 EXPORT_SYMBOL_GPL(kvm_lmsw);
363 void kvm_set_cr4(struct kvm_vcpu *vcpu, unsigned long cr4)
365 unsigned long old_cr4 = vcpu->arch.cr4;
366 unsigned long pdptr_bits = X86_CR4_PGE | X86_CR4_PSE | X86_CR4_PAE;
368 if (cr4 & CR4_RESERVED_BITS) {
369 printk(KERN_DEBUG "set_cr4: #GP, reserved bits\n");
370 kvm_inject_gp(vcpu, 0);
374 if (is_long_mode(vcpu)) {
375 if (!(cr4 & X86_CR4_PAE)) {
376 printk(KERN_DEBUG "set_cr4: #GP, clearing PAE while "
378 kvm_inject_gp(vcpu, 0);
381 } else if (is_paging(vcpu) && (cr4 & X86_CR4_PAE)
382 && ((cr4 ^ old_cr4) & pdptr_bits)
383 && !load_pdptrs(vcpu, vcpu->arch.cr3)) {
384 printk(KERN_DEBUG "set_cr4: #GP, pdptrs reserved bits\n");
385 kvm_inject_gp(vcpu, 0);
389 if (cr4 & X86_CR4_VMXE) {
390 printk(KERN_DEBUG "set_cr4: #GP, setting VMXE\n");
391 kvm_inject_gp(vcpu, 0);
394 kvm_x86_ops->set_cr4(vcpu, cr4);
395 vcpu->arch.cr4 = cr4;
396 vcpu->arch.mmu.base_role.cr4_pge = (cr4 & X86_CR4_PGE) && !tdp_enabled;
397 kvm_mmu_reset_context(vcpu);
399 EXPORT_SYMBOL_GPL(kvm_set_cr4);
401 void kvm_set_cr3(struct kvm_vcpu *vcpu, unsigned long cr3)
403 if (cr3 == vcpu->arch.cr3 && !pdptrs_changed(vcpu)) {
404 kvm_mmu_sync_roots(vcpu);
405 kvm_mmu_flush_tlb(vcpu);
409 if (is_long_mode(vcpu)) {
410 if (cr3 & CR3_L_MODE_RESERVED_BITS) {
411 printk(KERN_DEBUG "set_cr3: #GP, reserved bits\n");
412 kvm_inject_gp(vcpu, 0);
417 if (cr3 & CR3_PAE_RESERVED_BITS) {
419 "set_cr3: #GP, reserved bits\n");
420 kvm_inject_gp(vcpu, 0);
423 if (is_paging(vcpu) && !load_pdptrs(vcpu, cr3)) {
424 printk(KERN_DEBUG "set_cr3: #GP, pdptrs "
426 kvm_inject_gp(vcpu, 0);
431 * We don't check reserved bits in nonpae mode, because
432 * this isn't enforced, and VMware depends on this.
437 * Does the new cr3 value map to physical memory? (Note, we
438 * catch an invalid cr3 even in real-mode, because it would
439 * cause trouble later on when we turn on paging anyway.)
441 * A real CPU would silently accept an invalid cr3 and would
442 * attempt to use it - with largely undefined (and often hard
443 * to debug) behavior on the guest side.
445 if (unlikely(!gfn_to_memslot(vcpu->kvm, cr3 >> PAGE_SHIFT)))
446 kvm_inject_gp(vcpu, 0);
448 vcpu->arch.cr3 = cr3;
449 vcpu->arch.mmu.new_cr3(vcpu);
452 EXPORT_SYMBOL_GPL(kvm_set_cr3);
454 void kvm_set_cr8(struct kvm_vcpu *vcpu, unsigned long cr8)
456 if (cr8 & CR8_RESERVED_BITS) {
457 printk(KERN_DEBUG "set_cr8: #GP, reserved bits 0x%lx\n", cr8);
458 kvm_inject_gp(vcpu, 0);
461 if (irqchip_in_kernel(vcpu->kvm))
462 kvm_lapic_set_tpr(vcpu, cr8);
464 vcpu->arch.cr8 = cr8;
466 EXPORT_SYMBOL_GPL(kvm_set_cr8);
468 unsigned long kvm_get_cr8(struct kvm_vcpu *vcpu)
470 if (irqchip_in_kernel(vcpu->kvm))
471 return kvm_lapic_get_cr8(vcpu);
473 return vcpu->arch.cr8;
475 EXPORT_SYMBOL_GPL(kvm_get_cr8);
477 static inline u32 bit(int bitno)
479 return 1 << (bitno & 31);
483 * List of msr numbers which we expose to userspace through KVM_GET_MSRS
484 * and KVM_SET_MSRS, and KVM_GET_MSR_INDEX_LIST.
486 * This list is modified at module load time to reflect the
487 * capabilities of the host cpu.
489 static u32 msrs_to_save[] = {
490 MSR_IA32_SYSENTER_CS, MSR_IA32_SYSENTER_ESP, MSR_IA32_SYSENTER_EIP,
493 MSR_CSTAR, MSR_KERNEL_GS_BASE, MSR_SYSCALL_MASK, MSR_LSTAR,
495 MSR_IA32_TSC, MSR_KVM_SYSTEM_TIME, MSR_KVM_WALL_CLOCK,
496 MSR_IA32_PERF_STATUS, MSR_IA32_CR_PAT, MSR_VM_HSAVE_PA
499 static unsigned num_msrs_to_save;
501 static u32 emulated_msrs[] = {
502 MSR_IA32_MISC_ENABLE,
505 static void set_efer(struct kvm_vcpu *vcpu, u64 efer)
507 if (efer & efer_reserved_bits) {
508 printk(KERN_DEBUG "set_efer: 0x%llx #GP, reserved bits\n",
510 kvm_inject_gp(vcpu, 0);
515 && (vcpu->arch.shadow_efer & EFER_LME) != (efer & EFER_LME)) {
516 printk(KERN_DEBUG "set_efer: #GP, change LME while paging\n");
517 kvm_inject_gp(vcpu, 0);
521 if (efer & EFER_FFXSR) {
522 struct kvm_cpuid_entry2 *feat;
524 feat = kvm_find_cpuid_entry(vcpu, 0x80000001, 0);
525 if (!feat || !(feat->edx & bit(X86_FEATURE_FXSR_OPT))) {
526 printk(KERN_DEBUG "set_efer: #GP, enable FFXSR w/o CPUID capability\n");
527 kvm_inject_gp(vcpu, 0);
532 if (efer & EFER_SVME) {
533 struct kvm_cpuid_entry2 *feat;
535 feat = kvm_find_cpuid_entry(vcpu, 0x80000001, 0);
536 if (!feat || !(feat->ecx & bit(X86_FEATURE_SVM))) {
537 printk(KERN_DEBUG "set_efer: #GP, enable SVM w/o SVM\n");
538 kvm_inject_gp(vcpu, 0);
543 kvm_x86_ops->set_efer(vcpu, efer);
546 efer |= vcpu->arch.shadow_efer & EFER_LMA;
548 vcpu->arch.shadow_efer = efer;
550 vcpu->arch.mmu.base_role.nxe = (efer & EFER_NX) && !tdp_enabled;
551 kvm_mmu_reset_context(vcpu);
554 void kvm_enable_efer_bits(u64 mask)
556 efer_reserved_bits &= ~mask;
558 EXPORT_SYMBOL_GPL(kvm_enable_efer_bits);
562 * Writes msr value into into the appropriate "register".
563 * Returns 0 on success, non-0 otherwise.
564 * Assumes vcpu_load() was already called.
566 int kvm_set_msr(struct kvm_vcpu *vcpu, u32 msr_index, u64 data)
568 return kvm_x86_ops->set_msr(vcpu, msr_index, data);
572 * Adapt set_msr() to msr_io()'s calling convention
574 static int do_set_msr(struct kvm_vcpu *vcpu, unsigned index, u64 *data)
576 return kvm_set_msr(vcpu, index, *data);
579 static void kvm_write_wall_clock(struct kvm *kvm, gpa_t wall_clock)
582 struct pvclock_wall_clock wc;
583 struct timespec now, sys, boot;
590 kvm_write_guest(kvm, wall_clock, &version, sizeof(version));
593 * The guest calculates current wall clock time by adding
594 * system time (updated by kvm_write_guest_time below) to the
595 * wall clock specified here. guest system time equals host
596 * system time for us, thus we must fill in host boot time here.
598 now = current_kernel_time();
600 boot = ns_to_timespec(timespec_to_ns(&now) - timespec_to_ns(&sys));
602 wc.sec = boot.tv_sec;
603 wc.nsec = boot.tv_nsec;
604 wc.version = version;
606 kvm_write_guest(kvm, wall_clock, &wc, sizeof(wc));
609 kvm_write_guest(kvm, wall_clock, &version, sizeof(version));
612 static uint32_t div_frac(uint32_t dividend, uint32_t divisor)
614 uint32_t quotient, remainder;
616 /* Don't try to replace with do_div(), this one calculates
617 * "(dividend << 32) / divisor" */
619 : "=a" (quotient), "=d" (remainder)
620 : "0" (0), "1" (dividend), "r" (divisor) );
624 static void kvm_set_time_scale(uint32_t tsc_khz, struct pvclock_vcpu_time_info *hv_clock)
626 uint64_t nsecs = 1000000000LL;
631 tps64 = tsc_khz * 1000LL;
632 while (tps64 > nsecs*2) {
637 tps32 = (uint32_t)tps64;
638 while (tps32 <= (uint32_t)nsecs) {
643 hv_clock->tsc_shift = shift;
644 hv_clock->tsc_to_system_mul = div_frac(nsecs, tps32);
646 pr_debug("%s: tsc_khz %u, tsc_shift %d, tsc_mul %u\n",
647 __func__, tsc_khz, hv_clock->tsc_shift,
648 hv_clock->tsc_to_system_mul);
651 static DEFINE_PER_CPU(unsigned long, cpu_tsc_khz);
653 static void kvm_write_guest_time(struct kvm_vcpu *v)
657 struct kvm_vcpu_arch *vcpu = &v->arch;
659 unsigned long this_tsc_khz;
661 if ((!vcpu->time_page))
664 this_tsc_khz = get_cpu_var(cpu_tsc_khz);
665 if (unlikely(vcpu->hv_clock_tsc_khz != this_tsc_khz)) {
666 kvm_set_time_scale(this_tsc_khz, &vcpu->hv_clock);
667 vcpu->hv_clock_tsc_khz = this_tsc_khz;
669 put_cpu_var(cpu_tsc_khz);
671 /* Keep irq disabled to prevent changes to the clock */
672 local_irq_save(flags);
673 kvm_get_msr(v, MSR_IA32_TSC, &vcpu->hv_clock.tsc_timestamp);
675 local_irq_restore(flags);
677 /* With all the info we got, fill in the values */
679 vcpu->hv_clock.system_time = ts.tv_nsec +
680 (NSEC_PER_SEC * (u64)ts.tv_sec);
682 * The interface expects us to write an even number signaling that the
683 * update is finished. Since the guest won't see the intermediate
684 * state, we just increase by 2 at the end.
686 vcpu->hv_clock.version += 2;
688 shared_kaddr = kmap_atomic(vcpu->time_page, KM_USER0);
690 memcpy(shared_kaddr + vcpu->time_offset, &vcpu->hv_clock,
691 sizeof(vcpu->hv_clock));
693 kunmap_atomic(shared_kaddr, KM_USER0);
695 mark_page_dirty(v->kvm, vcpu->time >> PAGE_SHIFT);
698 static int kvm_request_guest_time_update(struct kvm_vcpu *v)
700 struct kvm_vcpu_arch *vcpu = &v->arch;
702 if (!vcpu->time_page)
704 set_bit(KVM_REQ_KVMCLOCK_UPDATE, &v->requests);
708 static bool msr_mtrr_valid(unsigned msr)
711 case 0x200 ... 0x200 + 2 * KVM_NR_VAR_MTRR - 1:
712 case MSR_MTRRfix64K_00000:
713 case MSR_MTRRfix16K_80000:
714 case MSR_MTRRfix16K_A0000:
715 case MSR_MTRRfix4K_C0000:
716 case MSR_MTRRfix4K_C8000:
717 case MSR_MTRRfix4K_D0000:
718 case MSR_MTRRfix4K_D8000:
719 case MSR_MTRRfix4K_E0000:
720 case MSR_MTRRfix4K_E8000:
721 case MSR_MTRRfix4K_F0000:
722 case MSR_MTRRfix4K_F8000:
723 case MSR_MTRRdefType:
724 case MSR_IA32_CR_PAT:
732 static bool valid_pat_type(unsigned t)
734 return t < 8 && (1 << t) & 0xf3; /* 0, 1, 4, 5, 6, 7 */
737 static bool valid_mtrr_type(unsigned t)
739 return t < 8 && (1 << t) & 0x73; /* 0, 1, 4, 5, 6 */
742 static bool mtrr_valid(struct kvm_vcpu *vcpu, u32 msr, u64 data)
746 if (!msr_mtrr_valid(msr))
749 if (msr == MSR_IA32_CR_PAT) {
750 for (i = 0; i < 8; i++)
751 if (!valid_pat_type((data >> (i * 8)) & 0xff))
754 } else if (msr == MSR_MTRRdefType) {
757 return valid_mtrr_type(data & 0xff);
758 } else if (msr >= MSR_MTRRfix64K_00000 && msr <= MSR_MTRRfix4K_F8000) {
759 for (i = 0; i < 8 ; i++)
760 if (!valid_mtrr_type((data >> (i * 8)) & 0xff))
766 return valid_mtrr_type(data & 0xff);
769 static int set_msr_mtrr(struct kvm_vcpu *vcpu, u32 msr, u64 data)
771 u64 *p = (u64 *)&vcpu->arch.mtrr_state.fixed_ranges;
773 if (!mtrr_valid(vcpu, msr, data))
776 if (msr == MSR_MTRRdefType) {
777 vcpu->arch.mtrr_state.def_type = data;
778 vcpu->arch.mtrr_state.enabled = (data & 0xc00) >> 10;
779 } else if (msr == MSR_MTRRfix64K_00000)
781 else if (msr == MSR_MTRRfix16K_80000 || msr == MSR_MTRRfix16K_A0000)
782 p[1 + msr - MSR_MTRRfix16K_80000] = data;
783 else if (msr >= MSR_MTRRfix4K_C0000 && msr <= MSR_MTRRfix4K_F8000)
784 p[3 + msr - MSR_MTRRfix4K_C0000] = data;
785 else if (msr == MSR_IA32_CR_PAT)
786 vcpu->arch.pat = data;
787 else { /* Variable MTRRs */
788 int idx, is_mtrr_mask;
791 idx = (msr - 0x200) / 2;
792 is_mtrr_mask = msr - 0x200 - 2 * idx;
795 (u64 *)&vcpu->arch.mtrr_state.var_ranges[idx].base_lo;
798 (u64 *)&vcpu->arch.mtrr_state.var_ranges[idx].mask_lo;
802 kvm_mmu_reset_context(vcpu);
806 static int set_msr_mce(struct kvm_vcpu *vcpu, u32 msr, u64 data)
808 u64 mcg_cap = vcpu->arch.mcg_cap;
809 unsigned bank_num = mcg_cap & 0xff;
812 case MSR_IA32_MCG_STATUS:
813 vcpu->arch.mcg_status = data;
815 case MSR_IA32_MCG_CTL:
816 if (!(mcg_cap & MCG_CTL_P))
818 if (data != 0 && data != ~(u64)0)
820 vcpu->arch.mcg_ctl = data;
823 if (msr >= MSR_IA32_MC0_CTL &&
824 msr < MSR_IA32_MC0_CTL + 4 * bank_num) {
825 u32 offset = msr - MSR_IA32_MC0_CTL;
826 /* only 0 or all 1s can be written to IA32_MCi_CTL */
827 if ((offset & 0x3) == 0 &&
828 data != 0 && data != ~(u64)0)
830 vcpu->arch.mce_banks[offset] = data;
838 int kvm_set_msr_common(struct kvm_vcpu *vcpu, u32 msr, u64 data)
842 set_efer(vcpu, data);
845 data &= ~(u64)0x40; /* ignore flush filter disable */
847 pr_unimpl(vcpu, "unimplemented HWCR wrmsr: 0x%llx\n",
852 case MSR_FAM10H_MMIO_CONF_BASE:
854 pr_unimpl(vcpu, "unimplemented MMIO_CONF_BASE wrmsr: "
859 case MSR_AMD64_NB_CFG:
861 case MSR_IA32_DEBUGCTLMSR:
863 /* We support the non-activated case already */
865 } else if (data & ~(DEBUGCTLMSR_LBR | DEBUGCTLMSR_BTF)) {
866 /* Values other than LBR and BTF are vendor-specific,
867 thus reserved and should throw a #GP */
870 pr_unimpl(vcpu, "%s: MSR_IA32_DEBUGCTLMSR 0x%llx, nop\n",
873 case MSR_IA32_UCODE_REV:
874 case MSR_IA32_UCODE_WRITE:
875 case MSR_VM_HSAVE_PA:
876 case MSR_AMD64_PATCH_LOADER:
878 case 0x200 ... 0x2ff:
879 return set_msr_mtrr(vcpu, msr, data);
880 case MSR_IA32_APICBASE:
881 kvm_set_apic_base(vcpu, data);
883 case APIC_BASE_MSR ... APIC_BASE_MSR + 0x3ff:
884 return kvm_x2apic_msr_write(vcpu, msr, data);
885 case MSR_IA32_MISC_ENABLE:
886 vcpu->arch.ia32_misc_enable_msr = data;
888 case MSR_KVM_WALL_CLOCK:
889 vcpu->kvm->arch.wall_clock = data;
890 kvm_write_wall_clock(vcpu->kvm, data);
892 case MSR_KVM_SYSTEM_TIME: {
893 if (vcpu->arch.time_page) {
894 kvm_release_page_dirty(vcpu->arch.time_page);
895 vcpu->arch.time_page = NULL;
898 vcpu->arch.time = data;
900 /* we verify if the enable bit is set... */
904 /* ...but clean it before doing the actual write */
905 vcpu->arch.time_offset = data & ~(PAGE_MASK | 1);
907 vcpu->arch.time_page =
908 gfn_to_page(vcpu->kvm, data >> PAGE_SHIFT);
910 if (is_error_page(vcpu->arch.time_page)) {
911 kvm_release_page_clean(vcpu->arch.time_page);
912 vcpu->arch.time_page = NULL;
915 kvm_request_guest_time_update(vcpu);
918 case MSR_IA32_MCG_CTL:
919 case MSR_IA32_MCG_STATUS:
920 case MSR_IA32_MC0_CTL ... MSR_IA32_MC0_CTL + 4 * KVM_MAX_MCE_BANKS - 1:
921 return set_msr_mce(vcpu, msr, data);
923 /* Performance counters are not protected by a CPUID bit,
924 * so we should check all of them in the generic path for the sake of
925 * cross vendor migration.
926 * Writing a zero into the event select MSRs disables them,
927 * which we perfectly emulate ;-). Any other value should be at least
928 * reported, some guests depend on them.
930 case MSR_P6_EVNTSEL0:
931 case MSR_P6_EVNTSEL1:
932 case MSR_K7_EVNTSEL0:
933 case MSR_K7_EVNTSEL1:
934 case MSR_K7_EVNTSEL2:
935 case MSR_K7_EVNTSEL3:
937 pr_unimpl(vcpu, "unimplemented perfctr wrmsr: "
938 "0x%x data 0x%llx\n", msr, data);
940 /* at least RHEL 4 unconditionally writes to the perfctr registers,
941 * so we ignore writes to make it happy.
943 case MSR_P6_PERFCTR0:
944 case MSR_P6_PERFCTR1:
945 case MSR_K7_PERFCTR0:
946 case MSR_K7_PERFCTR1:
947 case MSR_K7_PERFCTR2:
948 case MSR_K7_PERFCTR3:
949 pr_unimpl(vcpu, "unimplemented perfctr wrmsr: "
950 "0x%x data 0x%llx\n", msr, data);
954 pr_unimpl(vcpu, "unhandled wrmsr: 0x%x data %llx\n",
958 pr_unimpl(vcpu, "ignored wrmsr: 0x%x data %llx\n",
965 EXPORT_SYMBOL_GPL(kvm_set_msr_common);
969 * Reads an msr value (of 'msr_index') into 'pdata'.
970 * Returns 0 on success, non-0 otherwise.
971 * Assumes vcpu_load() was already called.
973 int kvm_get_msr(struct kvm_vcpu *vcpu, u32 msr_index, u64 *pdata)
975 return kvm_x86_ops->get_msr(vcpu, msr_index, pdata);
978 static int get_msr_mtrr(struct kvm_vcpu *vcpu, u32 msr, u64 *pdata)
980 u64 *p = (u64 *)&vcpu->arch.mtrr_state.fixed_ranges;
982 if (!msr_mtrr_valid(msr))
985 if (msr == MSR_MTRRdefType)
986 *pdata = vcpu->arch.mtrr_state.def_type +
987 (vcpu->arch.mtrr_state.enabled << 10);
988 else if (msr == MSR_MTRRfix64K_00000)
990 else if (msr == MSR_MTRRfix16K_80000 || msr == MSR_MTRRfix16K_A0000)
991 *pdata = p[1 + msr - MSR_MTRRfix16K_80000];
992 else if (msr >= MSR_MTRRfix4K_C0000 && msr <= MSR_MTRRfix4K_F8000)
993 *pdata = p[3 + msr - MSR_MTRRfix4K_C0000];
994 else if (msr == MSR_IA32_CR_PAT)
995 *pdata = vcpu->arch.pat;
996 else { /* Variable MTRRs */
997 int idx, is_mtrr_mask;
1000 idx = (msr - 0x200) / 2;
1001 is_mtrr_mask = msr - 0x200 - 2 * idx;
1004 (u64 *)&vcpu->arch.mtrr_state.var_ranges[idx].base_lo;
1007 (u64 *)&vcpu->arch.mtrr_state.var_ranges[idx].mask_lo;
1014 static int get_msr_mce(struct kvm_vcpu *vcpu, u32 msr, u64 *pdata)
1017 u64 mcg_cap = vcpu->arch.mcg_cap;
1018 unsigned bank_num = mcg_cap & 0xff;
1021 case MSR_IA32_P5_MC_ADDR:
1022 case MSR_IA32_P5_MC_TYPE:
1025 case MSR_IA32_MCG_CAP:
1026 data = vcpu->arch.mcg_cap;
1028 case MSR_IA32_MCG_CTL:
1029 if (!(mcg_cap & MCG_CTL_P))
1031 data = vcpu->arch.mcg_ctl;
1033 case MSR_IA32_MCG_STATUS:
1034 data = vcpu->arch.mcg_status;
1037 if (msr >= MSR_IA32_MC0_CTL &&
1038 msr < MSR_IA32_MC0_CTL + 4 * bank_num) {
1039 u32 offset = msr - MSR_IA32_MC0_CTL;
1040 data = vcpu->arch.mce_banks[offset];
1049 int kvm_get_msr_common(struct kvm_vcpu *vcpu, u32 msr, u64 *pdata)
1054 case MSR_IA32_PLATFORM_ID:
1055 case MSR_IA32_UCODE_REV:
1056 case MSR_IA32_EBL_CR_POWERON:
1057 case MSR_IA32_DEBUGCTLMSR:
1058 case MSR_IA32_LASTBRANCHFROMIP:
1059 case MSR_IA32_LASTBRANCHTOIP:
1060 case MSR_IA32_LASTINTFROMIP:
1061 case MSR_IA32_LASTINTTOIP:
1064 case MSR_VM_HSAVE_PA:
1065 case MSR_P6_PERFCTR0:
1066 case MSR_P6_PERFCTR1:
1067 case MSR_P6_EVNTSEL0:
1068 case MSR_P6_EVNTSEL1:
1069 case MSR_K7_EVNTSEL0:
1070 case MSR_K7_PERFCTR0:
1071 case MSR_K8_INT_PENDING_MSG:
1072 case MSR_AMD64_NB_CFG:
1073 case MSR_FAM10H_MMIO_CONF_BASE:
1077 data = 0x500 | KVM_NR_VAR_MTRR;
1079 case 0x200 ... 0x2ff:
1080 return get_msr_mtrr(vcpu, msr, pdata);
1081 case 0xcd: /* fsb frequency */
1084 case MSR_IA32_APICBASE:
1085 data = kvm_get_apic_base(vcpu);
1087 case APIC_BASE_MSR ... APIC_BASE_MSR + 0x3ff:
1088 return kvm_x2apic_msr_read(vcpu, msr, pdata);
1090 case MSR_IA32_MISC_ENABLE:
1091 data = vcpu->arch.ia32_misc_enable_msr;
1093 case MSR_IA32_PERF_STATUS:
1094 /* TSC increment by tick */
1096 /* CPU multiplier */
1097 data |= (((uint64_t)4ULL) << 40);
1100 data = vcpu->arch.shadow_efer;
1102 case MSR_KVM_WALL_CLOCK:
1103 data = vcpu->kvm->arch.wall_clock;
1105 case MSR_KVM_SYSTEM_TIME:
1106 data = vcpu->arch.time;
1108 case MSR_IA32_P5_MC_ADDR:
1109 case MSR_IA32_P5_MC_TYPE:
1110 case MSR_IA32_MCG_CAP:
1111 case MSR_IA32_MCG_CTL:
1112 case MSR_IA32_MCG_STATUS:
1113 case MSR_IA32_MC0_CTL ... MSR_IA32_MC0_CTL + 4 * KVM_MAX_MCE_BANKS - 1:
1114 return get_msr_mce(vcpu, msr, pdata);
1117 pr_unimpl(vcpu, "unhandled rdmsr: 0x%x\n", msr);
1120 pr_unimpl(vcpu, "ignored rdmsr: 0x%x\n", msr);
1128 EXPORT_SYMBOL_GPL(kvm_get_msr_common);
1131 * Read or write a bunch of msrs. All parameters are kernel addresses.
1133 * @return number of msrs set successfully.
1135 static int __msr_io(struct kvm_vcpu *vcpu, struct kvm_msrs *msrs,
1136 struct kvm_msr_entry *entries,
1137 int (*do_msr)(struct kvm_vcpu *vcpu,
1138 unsigned index, u64 *data))
1144 down_read(&vcpu->kvm->slots_lock);
1145 for (i = 0; i < msrs->nmsrs; ++i)
1146 if (do_msr(vcpu, entries[i].index, &entries[i].data))
1148 up_read(&vcpu->kvm->slots_lock);
1156 * Read or write a bunch of msrs. Parameters are user addresses.
1158 * @return number of msrs set successfully.
1160 static int msr_io(struct kvm_vcpu *vcpu, struct kvm_msrs __user *user_msrs,
1161 int (*do_msr)(struct kvm_vcpu *vcpu,
1162 unsigned index, u64 *data),
1165 struct kvm_msrs msrs;
1166 struct kvm_msr_entry *entries;
1171 if (copy_from_user(&msrs, user_msrs, sizeof msrs))
1175 if (msrs.nmsrs >= MAX_IO_MSRS)
1179 size = sizeof(struct kvm_msr_entry) * msrs.nmsrs;
1180 entries = vmalloc(size);
1185 if (copy_from_user(entries, user_msrs->entries, size))
1188 r = n = __msr_io(vcpu, &msrs, entries, do_msr);
1193 if (writeback && copy_to_user(user_msrs->entries, entries, size))
1204 int kvm_dev_ioctl_check_extension(long ext)
1209 case KVM_CAP_IRQCHIP:
1211 case KVM_CAP_MMU_SHADOW_CACHE_CONTROL:
1212 case KVM_CAP_SET_TSS_ADDR:
1213 case KVM_CAP_EXT_CPUID:
1214 case KVM_CAP_CLOCKSOURCE:
1216 case KVM_CAP_NOP_IO_DELAY:
1217 case KVM_CAP_MP_STATE:
1218 case KVM_CAP_SYNC_MMU:
1219 case KVM_CAP_REINJECT_CONTROL:
1220 case KVM_CAP_IRQ_INJECT_STATUS:
1221 case KVM_CAP_ASSIGN_DEV_IRQ:
1223 case KVM_CAP_IOEVENTFD:
1225 case KVM_CAP_PIT_STATE2:
1226 case KVM_CAP_SET_IDENTITY_MAP_ADDR:
1229 case KVM_CAP_COALESCED_MMIO:
1230 r = KVM_COALESCED_MMIO_PAGE_OFFSET;
1233 r = !kvm_x86_ops->cpu_has_accelerated_tpr();
1235 case KVM_CAP_NR_VCPUS:
1238 case KVM_CAP_NR_MEMSLOTS:
1239 r = KVM_MEMORY_SLOTS;
1241 case KVM_CAP_PV_MMU:
1248 r = KVM_MAX_MCE_BANKS;
1258 long kvm_arch_dev_ioctl(struct file *filp,
1259 unsigned int ioctl, unsigned long arg)
1261 void __user *argp = (void __user *)arg;
1265 case KVM_GET_MSR_INDEX_LIST: {
1266 struct kvm_msr_list __user *user_msr_list = argp;
1267 struct kvm_msr_list msr_list;
1271 if (copy_from_user(&msr_list, user_msr_list, sizeof msr_list))
1274 msr_list.nmsrs = num_msrs_to_save + ARRAY_SIZE(emulated_msrs);
1275 if (copy_to_user(user_msr_list, &msr_list, sizeof msr_list))
1278 if (n < msr_list.nmsrs)
1281 if (copy_to_user(user_msr_list->indices, &msrs_to_save,
1282 num_msrs_to_save * sizeof(u32)))
1284 if (copy_to_user(user_msr_list->indices + num_msrs_to_save,
1286 ARRAY_SIZE(emulated_msrs) * sizeof(u32)))
1291 case KVM_GET_SUPPORTED_CPUID: {
1292 struct kvm_cpuid2 __user *cpuid_arg = argp;
1293 struct kvm_cpuid2 cpuid;
1296 if (copy_from_user(&cpuid, cpuid_arg, sizeof cpuid))
1298 r = kvm_dev_ioctl_get_supported_cpuid(&cpuid,
1299 cpuid_arg->entries);
1304 if (copy_to_user(cpuid_arg, &cpuid, sizeof cpuid))
1309 case KVM_X86_GET_MCE_CAP_SUPPORTED: {
1312 mce_cap = KVM_MCE_CAP_SUPPORTED;
1314 if (copy_to_user(argp, &mce_cap, sizeof mce_cap))
1326 void kvm_arch_vcpu_load(struct kvm_vcpu *vcpu, int cpu)
1328 kvm_x86_ops->vcpu_load(vcpu, cpu);
1329 kvm_request_guest_time_update(vcpu);
1332 void kvm_arch_vcpu_put(struct kvm_vcpu *vcpu)
1334 kvm_x86_ops->vcpu_put(vcpu);
1335 kvm_put_guest_fpu(vcpu);
1338 static int is_efer_nx(void)
1340 unsigned long long efer = 0;
1342 rdmsrl_safe(MSR_EFER, &efer);
1343 return efer & EFER_NX;
1346 static void cpuid_fix_nx_cap(struct kvm_vcpu *vcpu)
1349 struct kvm_cpuid_entry2 *e, *entry;
1352 for (i = 0; i < vcpu->arch.cpuid_nent; ++i) {
1353 e = &vcpu->arch.cpuid_entries[i];
1354 if (e->function == 0x80000001) {
1359 if (entry && (entry->edx & (1 << 20)) && !is_efer_nx()) {
1360 entry->edx &= ~(1 << 20);
1361 printk(KERN_INFO "kvm: guest NX capability removed\n");
1365 /* when an old userspace process fills a new kernel module */
1366 static int kvm_vcpu_ioctl_set_cpuid(struct kvm_vcpu *vcpu,
1367 struct kvm_cpuid *cpuid,
1368 struct kvm_cpuid_entry __user *entries)
1371 struct kvm_cpuid_entry *cpuid_entries;
1374 if (cpuid->nent > KVM_MAX_CPUID_ENTRIES)
1377 cpuid_entries = vmalloc(sizeof(struct kvm_cpuid_entry) * cpuid->nent);
1381 if (copy_from_user(cpuid_entries, entries,
1382 cpuid->nent * sizeof(struct kvm_cpuid_entry)))
1384 for (i = 0; i < cpuid->nent; i++) {
1385 vcpu->arch.cpuid_entries[i].function = cpuid_entries[i].function;
1386 vcpu->arch.cpuid_entries[i].eax = cpuid_entries[i].eax;
1387 vcpu->arch.cpuid_entries[i].ebx = cpuid_entries[i].ebx;
1388 vcpu->arch.cpuid_entries[i].ecx = cpuid_entries[i].ecx;
1389 vcpu->arch.cpuid_entries[i].edx = cpuid_entries[i].edx;
1390 vcpu->arch.cpuid_entries[i].index = 0;
1391 vcpu->arch.cpuid_entries[i].flags = 0;
1392 vcpu->arch.cpuid_entries[i].padding[0] = 0;
1393 vcpu->arch.cpuid_entries[i].padding[1] = 0;
1394 vcpu->arch.cpuid_entries[i].padding[2] = 0;
1396 vcpu->arch.cpuid_nent = cpuid->nent;
1397 cpuid_fix_nx_cap(vcpu);
1399 kvm_apic_set_version(vcpu);
1402 vfree(cpuid_entries);
1407 static int kvm_vcpu_ioctl_set_cpuid2(struct kvm_vcpu *vcpu,
1408 struct kvm_cpuid2 *cpuid,
1409 struct kvm_cpuid_entry2 __user *entries)
1414 if (cpuid->nent > KVM_MAX_CPUID_ENTRIES)
1417 if (copy_from_user(&vcpu->arch.cpuid_entries, entries,
1418 cpuid->nent * sizeof(struct kvm_cpuid_entry2)))
1420 vcpu->arch.cpuid_nent = cpuid->nent;
1421 kvm_apic_set_version(vcpu);
1428 static int kvm_vcpu_ioctl_get_cpuid2(struct kvm_vcpu *vcpu,
1429 struct kvm_cpuid2 *cpuid,
1430 struct kvm_cpuid_entry2 __user *entries)
1435 if (cpuid->nent < vcpu->arch.cpuid_nent)
1438 if (copy_to_user(entries, &vcpu->arch.cpuid_entries,
1439 vcpu->arch.cpuid_nent * sizeof(struct kvm_cpuid_entry2)))
1444 cpuid->nent = vcpu->arch.cpuid_nent;
1448 static void do_cpuid_1_ent(struct kvm_cpuid_entry2 *entry, u32 function,
1451 entry->function = function;
1452 entry->index = index;
1453 cpuid_count(entry->function, entry->index,
1454 &entry->eax, &entry->ebx, &entry->ecx, &entry->edx);
1458 #define F(x) bit(X86_FEATURE_##x)
1460 static void do_cpuid_ent(struct kvm_cpuid_entry2 *entry, u32 function,
1461 u32 index, int *nent, int maxnent)
1463 unsigned f_nx = is_efer_nx() ? F(NX) : 0;
1464 unsigned f_gbpages = kvm_x86_ops->gb_page_enable() ? F(GBPAGES) : 0;
1465 #ifdef CONFIG_X86_64
1466 unsigned f_lm = F(LM);
1472 const u32 kvm_supported_word0_x86_features =
1473 F(FPU) | F(VME) | F(DE) | F(PSE) |
1474 F(TSC) | F(MSR) | F(PAE) | F(MCE) |
1475 F(CX8) | F(APIC) | 0 /* Reserved */ | F(SEP) |
1476 F(MTRR) | F(PGE) | F(MCA) | F(CMOV) |
1477 F(PAT) | F(PSE36) | 0 /* PSN */ | F(CLFLSH) |
1478 0 /* Reserved, DS, ACPI */ | F(MMX) |
1479 F(FXSR) | F(XMM) | F(XMM2) | F(SELFSNOOP) |
1480 0 /* HTT, TM, Reserved, PBE */;
1481 /* cpuid 0x80000001.edx */
1482 const u32 kvm_supported_word1_x86_features =
1483 F(FPU) | F(VME) | F(DE) | F(PSE) |
1484 F(TSC) | F(MSR) | F(PAE) | F(MCE) |
1485 F(CX8) | F(APIC) | 0 /* Reserved */ | F(SYSCALL) |
1486 F(MTRR) | F(PGE) | F(MCA) | F(CMOV) |
1487 F(PAT) | F(PSE36) | 0 /* Reserved */ |
1488 f_nx | 0 /* Reserved */ | F(MMXEXT) | F(MMX) |
1489 F(FXSR) | F(FXSR_OPT) | f_gbpages | 0 /* RDTSCP */ |
1490 0 /* Reserved */ | f_lm | F(3DNOWEXT) | F(3DNOW);
1492 const u32 kvm_supported_word4_x86_features =
1493 F(XMM3) | 0 /* Reserved, DTES64, MONITOR */ |
1494 0 /* DS-CPL, VMX, SMX, EST */ |
1495 0 /* TM2 */ | F(SSSE3) | 0 /* CNXT-ID */ | 0 /* Reserved */ |
1496 0 /* Reserved */ | F(CX16) | 0 /* xTPR Update, PDCM */ |
1497 0 /* Reserved, DCA */ | F(XMM4_1) |
1498 F(XMM4_2) | F(X2APIC) | F(MOVBE) | F(POPCNT) |
1499 0 /* Reserved, XSAVE, OSXSAVE */;
1500 /* cpuid 0x80000001.ecx */
1501 const u32 kvm_supported_word6_x86_features =
1502 F(LAHF_LM) | F(CMP_LEGACY) | F(SVM) | 0 /* ExtApicSpace */ |
1503 F(CR8_LEGACY) | F(ABM) | F(SSE4A) | F(MISALIGNSSE) |
1504 F(3DNOWPREFETCH) | 0 /* OSVW */ | 0 /* IBS */ | F(SSE5) |
1505 0 /* SKINIT */ | 0 /* WDT */;
1507 /* all calls to cpuid_count() should be made on the same cpu */
1509 do_cpuid_1_ent(entry, function, index);
1514 entry->eax = min(entry->eax, (u32)0xb);
1517 entry->edx &= kvm_supported_word0_x86_features;
1518 entry->ecx &= kvm_supported_word4_x86_features;
1519 /* we support x2apic emulation even if host does not support
1520 * it since we emulate x2apic in software */
1521 entry->ecx |= F(X2APIC);
1523 /* function 2 entries are STATEFUL. That is, repeated cpuid commands
1524 * may return different values. This forces us to get_cpu() before
1525 * issuing the first command, and also to emulate this annoying behavior
1526 * in kvm_emulate_cpuid() using KVM_CPUID_FLAG_STATE_READ_NEXT */
1528 int t, times = entry->eax & 0xff;
1530 entry->flags |= KVM_CPUID_FLAG_STATEFUL_FUNC;
1531 entry->flags |= KVM_CPUID_FLAG_STATE_READ_NEXT;
1532 for (t = 1; t < times && *nent < maxnent; ++t) {
1533 do_cpuid_1_ent(&entry[t], function, 0);
1534 entry[t].flags |= KVM_CPUID_FLAG_STATEFUL_FUNC;
1539 /* function 4 and 0xb have additional index. */
1543 entry->flags |= KVM_CPUID_FLAG_SIGNIFCANT_INDEX;
1544 /* read more entries until cache_type is zero */
1545 for (i = 1; *nent < maxnent; ++i) {
1546 cache_type = entry[i - 1].eax & 0x1f;
1549 do_cpuid_1_ent(&entry[i], function, i);
1551 KVM_CPUID_FLAG_SIGNIFCANT_INDEX;
1559 entry->flags |= KVM_CPUID_FLAG_SIGNIFCANT_INDEX;
1560 /* read more entries until level_type is zero */
1561 for (i = 1; *nent < maxnent; ++i) {
1562 level_type = entry[i - 1].ecx & 0xff00;
1565 do_cpuid_1_ent(&entry[i], function, i);
1567 KVM_CPUID_FLAG_SIGNIFCANT_INDEX;
1573 entry->eax = min(entry->eax, 0x8000001a);
1576 entry->edx &= kvm_supported_word1_x86_features;
1577 entry->ecx &= kvm_supported_word6_x86_features;
1585 static int kvm_dev_ioctl_get_supported_cpuid(struct kvm_cpuid2 *cpuid,
1586 struct kvm_cpuid_entry2 __user *entries)
1588 struct kvm_cpuid_entry2 *cpuid_entries;
1589 int limit, nent = 0, r = -E2BIG;
1592 if (cpuid->nent < 1)
1594 if (cpuid->nent > KVM_MAX_CPUID_ENTRIES)
1595 cpuid->nent = KVM_MAX_CPUID_ENTRIES;
1597 cpuid_entries = vmalloc(sizeof(struct kvm_cpuid_entry2) * cpuid->nent);
1601 do_cpuid_ent(&cpuid_entries[0], 0, 0, &nent, cpuid->nent);
1602 limit = cpuid_entries[0].eax;
1603 for (func = 1; func <= limit && nent < cpuid->nent; ++func)
1604 do_cpuid_ent(&cpuid_entries[nent], func, 0,
1605 &nent, cpuid->nent);
1607 if (nent >= cpuid->nent)
1610 do_cpuid_ent(&cpuid_entries[nent], 0x80000000, 0, &nent, cpuid->nent);
1611 limit = cpuid_entries[nent - 1].eax;
1612 for (func = 0x80000001; func <= limit && nent < cpuid->nent; ++func)
1613 do_cpuid_ent(&cpuid_entries[nent], func, 0,
1614 &nent, cpuid->nent);
1616 if (nent >= cpuid->nent)
1620 if (copy_to_user(entries, cpuid_entries,
1621 nent * sizeof(struct kvm_cpuid_entry2)))
1627 vfree(cpuid_entries);
1632 static int kvm_vcpu_ioctl_get_lapic(struct kvm_vcpu *vcpu,
1633 struct kvm_lapic_state *s)
1636 memcpy(s->regs, vcpu->arch.apic->regs, sizeof *s);
1642 static int kvm_vcpu_ioctl_set_lapic(struct kvm_vcpu *vcpu,
1643 struct kvm_lapic_state *s)
1646 memcpy(vcpu->arch.apic->regs, s->regs, sizeof *s);
1647 kvm_apic_post_state_restore(vcpu);
1648 update_cr8_intercept(vcpu);
1654 static int kvm_vcpu_ioctl_interrupt(struct kvm_vcpu *vcpu,
1655 struct kvm_interrupt *irq)
1657 if (irq->irq < 0 || irq->irq >= 256)
1659 if (irqchip_in_kernel(vcpu->kvm))
1663 kvm_queue_interrupt(vcpu, irq->irq, false);
1670 static int kvm_vcpu_ioctl_nmi(struct kvm_vcpu *vcpu)
1673 kvm_inject_nmi(vcpu);
1679 static int vcpu_ioctl_tpr_access_reporting(struct kvm_vcpu *vcpu,
1680 struct kvm_tpr_access_ctl *tac)
1684 vcpu->arch.tpr_access_reporting = !!tac->enabled;
1688 static int kvm_vcpu_ioctl_x86_setup_mce(struct kvm_vcpu *vcpu,
1692 unsigned bank_num = mcg_cap & 0xff, bank;
1695 if (!bank_num || bank_num >= KVM_MAX_MCE_BANKS)
1697 if (mcg_cap & ~(KVM_MCE_CAP_SUPPORTED | 0xff | 0xff0000))
1700 vcpu->arch.mcg_cap = mcg_cap;
1701 /* Init IA32_MCG_CTL to all 1s */
1702 if (mcg_cap & MCG_CTL_P)
1703 vcpu->arch.mcg_ctl = ~(u64)0;
1704 /* Init IA32_MCi_CTL to all 1s */
1705 for (bank = 0; bank < bank_num; bank++)
1706 vcpu->arch.mce_banks[bank*4] = ~(u64)0;
1711 static int kvm_vcpu_ioctl_x86_set_mce(struct kvm_vcpu *vcpu,
1712 struct kvm_x86_mce *mce)
1714 u64 mcg_cap = vcpu->arch.mcg_cap;
1715 unsigned bank_num = mcg_cap & 0xff;
1716 u64 *banks = vcpu->arch.mce_banks;
1718 if (mce->bank >= bank_num || !(mce->status & MCI_STATUS_VAL))
1721 * if IA32_MCG_CTL is not all 1s, the uncorrected error
1722 * reporting is disabled
1724 if ((mce->status & MCI_STATUS_UC) && (mcg_cap & MCG_CTL_P) &&
1725 vcpu->arch.mcg_ctl != ~(u64)0)
1727 banks += 4 * mce->bank;
1729 * if IA32_MCi_CTL is not all 1s, the uncorrected error
1730 * reporting is disabled for the bank
1732 if ((mce->status & MCI_STATUS_UC) && banks[0] != ~(u64)0)
1734 if (mce->status & MCI_STATUS_UC) {
1735 if ((vcpu->arch.mcg_status & MCG_STATUS_MCIP) ||
1736 !(vcpu->arch.cr4 & X86_CR4_MCE)) {
1737 printk(KERN_DEBUG "kvm: set_mce: "
1738 "injects mce exception while "
1739 "previous one is in progress!\n");
1740 set_bit(KVM_REQ_TRIPLE_FAULT, &vcpu->requests);
1743 if (banks[1] & MCI_STATUS_VAL)
1744 mce->status |= MCI_STATUS_OVER;
1745 banks[2] = mce->addr;
1746 banks[3] = mce->misc;
1747 vcpu->arch.mcg_status = mce->mcg_status;
1748 banks[1] = mce->status;
1749 kvm_queue_exception(vcpu, MC_VECTOR);
1750 } else if (!(banks[1] & MCI_STATUS_VAL)
1751 || !(banks[1] & MCI_STATUS_UC)) {
1752 if (banks[1] & MCI_STATUS_VAL)
1753 mce->status |= MCI_STATUS_OVER;
1754 banks[2] = mce->addr;
1755 banks[3] = mce->misc;
1756 banks[1] = mce->status;
1758 banks[1] |= MCI_STATUS_OVER;
1762 long kvm_arch_vcpu_ioctl(struct file *filp,
1763 unsigned int ioctl, unsigned long arg)
1765 struct kvm_vcpu *vcpu = filp->private_data;
1766 void __user *argp = (void __user *)arg;
1768 struct kvm_lapic_state *lapic = NULL;
1771 case KVM_GET_LAPIC: {
1772 lapic = kzalloc(sizeof(struct kvm_lapic_state), GFP_KERNEL);
1777 r = kvm_vcpu_ioctl_get_lapic(vcpu, lapic);
1781 if (copy_to_user(argp, lapic, sizeof(struct kvm_lapic_state)))
1786 case KVM_SET_LAPIC: {
1787 lapic = kmalloc(sizeof(struct kvm_lapic_state), GFP_KERNEL);
1792 if (copy_from_user(lapic, argp, sizeof(struct kvm_lapic_state)))
1794 r = kvm_vcpu_ioctl_set_lapic(vcpu, lapic);
1800 case KVM_INTERRUPT: {
1801 struct kvm_interrupt irq;
1804 if (copy_from_user(&irq, argp, sizeof irq))
1806 r = kvm_vcpu_ioctl_interrupt(vcpu, &irq);
1813 r = kvm_vcpu_ioctl_nmi(vcpu);
1819 case KVM_SET_CPUID: {
1820 struct kvm_cpuid __user *cpuid_arg = argp;
1821 struct kvm_cpuid cpuid;
1824 if (copy_from_user(&cpuid, cpuid_arg, sizeof cpuid))
1826 r = kvm_vcpu_ioctl_set_cpuid(vcpu, &cpuid, cpuid_arg->entries);
1831 case KVM_SET_CPUID2: {
1832 struct kvm_cpuid2 __user *cpuid_arg = argp;
1833 struct kvm_cpuid2 cpuid;
1836 if (copy_from_user(&cpuid, cpuid_arg, sizeof cpuid))
1838 r = kvm_vcpu_ioctl_set_cpuid2(vcpu, &cpuid,
1839 cpuid_arg->entries);
1844 case KVM_GET_CPUID2: {
1845 struct kvm_cpuid2 __user *cpuid_arg = argp;
1846 struct kvm_cpuid2 cpuid;
1849 if (copy_from_user(&cpuid, cpuid_arg, sizeof cpuid))
1851 r = kvm_vcpu_ioctl_get_cpuid2(vcpu, &cpuid,
1852 cpuid_arg->entries);
1856 if (copy_to_user(cpuid_arg, &cpuid, sizeof cpuid))
1862 r = msr_io(vcpu, argp, kvm_get_msr, 1);
1865 r = msr_io(vcpu, argp, do_set_msr, 0);
1867 case KVM_TPR_ACCESS_REPORTING: {
1868 struct kvm_tpr_access_ctl tac;
1871 if (copy_from_user(&tac, argp, sizeof tac))
1873 r = vcpu_ioctl_tpr_access_reporting(vcpu, &tac);
1877 if (copy_to_user(argp, &tac, sizeof tac))
1882 case KVM_SET_VAPIC_ADDR: {
1883 struct kvm_vapic_addr va;
1886 if (!irqchip_in_kernel(vcpu->kvm))
1889 if (copy_from_user(&va, argp, sizeof va))
1892 kvm_lapic_set_vapic_addr(vcpu, va.vapic_addr);
1895 case KVM_X86_SETUP_MCE: {
1899 if (copy_from_user(&mcg_cap, argp, sizeof mcg_cap))
1901 r = kvm_vcpu_ioctl_x86_setup_mce(vcpu, mcg_cap);
1904 case KVM_X86_SET_MCE: {
1905 struct kvm_x86_mce mce;
1908 if (copy_from_user(&mce, argp, sizeof mce))
1910 r = kvm_vcpu_ioctl_x86_set_mce(vcpu, &mce);
1921 static int kvm_vm_ioctl_set_tss_addr(struct kvm *kvm, unsigned long addr)
1925 if (addr > (unsigned int)(-3 * PAGE_SIZE))
1927 ret = kvm_x86_ops->set_tss_addr(kvm, addr);
1931 static int kvm_vm_ioctl_set_identity_map_addr(struct kvm *kvm,
1934 kvm->arch.ept_identity_map_addr = ident_addr;
1938 static int kvm_vm_ioctl_set_nr_mmu_pages(struct kvm *kvm,
1939 u32 kvm_nr_mmu_pages)
1941 if (kvm_nr_mmu_pages < KVM_MIN_ALLOC_MMU_PAGES)
1944 down_write(&kvm->slots_lock);
1945 spin_lock(&kvm->mmu_lock);
1947 kvm_mmu_change_mmu_pages(kvm, kvm_nr_mmu_pages);
1948 kvm->arch.n_requested_mmu_pages = kvm_nr_mmu_pages;
1950 spin_unlock(&kvm->mmu_lock);
1951 up_write(&kvm->slots_lock);
1955 static int kvm_vm_ioctl_get_nr_mmu_pages(struct kvm *kvm)
1957 return kvm->arch.n_alloc_mmu_pages;
1960 gfn_t unalias_gfn(struct kvm *kvm, gfn_t gfn)
1963 struct kvm_mem_alias *alias;
1965 for (i = 0; i < kvm->arch.naliases; ++i) {
1966 alias = &kvm->arch.aliases[i];
1967 if (gfn >= alias->base_gfn
1968 && gfn < alias->base_gfn + alias->npages)
1969 return alias->target_gfn + gfn - alias->base_gfn;
1975 * Set a new alias region. Aliases map a portion of physical memory into
1976 * another portion. This is useful for memory windows, for example the PC
1979 static int kvm_vm_ioctl_set_memory_alias(struct kvm *kvm,
1980 struct kvm_memory_alias *alias)
1983 struct kvm_mem_alias *p;
1986 /* General sanity checks */
1987 if (alias->memory_size & (PAGE_SIZE - 1))
1989 if (alias->guest_phys_addr & (PAGE_SIZE - 1))
1991 if (alias->slot >= KVM_ALIAS_SLOTS)
1993 if (alias->guest_phys_addr + alias->memory_size
1994 < alias->guest_phys_addr)
1996 if (alias->target_phys_addr + alias->memory_size
1997 < alias->target_phys_addr)
2000 down_write(&kvm->slots_lock);
2001 spin_lock(&kvm->mmu_lock);
2003 p = &kvm->arch.aliases[alias->slot];
2004 p->base_gfn = alias->guest_phys_addr >> PAGE_SHIFT;
2005 p->npages = alias->memory_size >> PAGE_SHIFT;
2006 p->target_gfn = alias->target_phys_addr >> PAGE_SHIFT;
2008 for (n = KVM_ALIAS_SLOTS; n > 0; --n)
2009 if (kvm->arch.aliases[n - 1].npages)
2011 kvm->arch.naliases = n;
2013 spin_unlock(&kvm->mmu_lock);
2014 kvm_mmu_zap_all(kvm);
2016 up_write(&kvm->slots_lock);
2024 static int kvm_vm_ioctl_get_irqchip(struct kvm *kvm, struct kvm_irqchip *chip)
2029 switch (chip->chip_id) {
2030 case KVM_IRQCHIP_PIC_MASTER:
2031 memcpy(&chip->chip.pic,
2032 &pic_irqchip(kvm)->pics[0],
2033 sizeof(struct kvm_pic_state));
2035 case KVM_IRQCHIP_PIC_SLAVE:
2036 memcpy(&chip->chip.pic,
2037 &pic_irqchip(kvm)->pics[1],
2038 sizeof(struct kvm_pic_state));
2040 case KVM_IRQCHIP_IOAPIC:
2041 r = kvm_get_ioapic(kvm, &chip->chip.ioapic);
2050 static int kvm_vm_ioctl_set_irqchip(struct kvm *kvm, struct kvm_irqchip *chip)
2055 switch (chip->chip_id) {
2056 case KVM_IRQCHIP_PIC_MASTER:
2057 spin_lock(&pic_irqchip(kvm)->lock);
2058 memcpy(&pic_irqchip(kvm)->pics[0],
2060 sizeof(struct kvm_pic_state));
2061 spin_unlock(&pic_irqchip(kvm)->lock);
2063 case KVM_IRQCHIP_PIC_SLAVE:
2064 spin_lock(&pic_irqchip(kvm)->lock);
2065 memcpy(&pic_irqchip(kvm)->pics[1],
2067 sizeof(struct kvm_pic_state));
2068 spin_unlock(&pic_irqchip(kvm)->lock);
2070 case KVM_IRQCHIP_IOAPIC:
2071 r = kvm_set_ioapic(kvm, &chip->chip.ioapic);
2077 kvm_pic_update_irq(pic_irqchip(kvm));
2081 static int kvm_vm_ioctl_get_pit(struct kvm *kvm, struct kvm_pit_state *ps)
2085 mutex_lock(&kvm->arch.vpit->pit_state.lock);
2086 memcpy(ps, &kvm->arch.vpit->pit_state, sizeof(struct kvm_pit_state));
2087 mutex_unlock(&kvm->arch.vpit->pit_state.lock);
2091 static int kvm_vm_ioctl_set_pit(struct kvm *kvm, struct kvm_pit_state *ps)
2095 mutex_lock(&kvm->arch.vpit->pit_state.lock);
2096 memcpy(&kvm->arch.vpit->pit_state, ps, sizeof(struct kvm_pit_state));
2097 kvm_pit_load_count(kvm, 0, ps->channels[0].count, 0);
2098 mutex_unlock(&kvm->arch.vpit->pit_state.lock);
2102 static int kvm_vm_ioctl_get_pit2(struct kvm *kvm, struct kvm_pit_state2 *ps)
2106 mutex_lock(&kvm->arch.vpit->pit_state.lock);
2107 memcpy(ps->channels, &kvm->arch.vpit->pit_state.channels,
2108 sizeof(ps->channels));
2109 ps->flags = kvm->arch.vpit->pit_state.flags;
2110 mutex_unlock(&kvm->arch.vpit->pit_state.lock);
2114 static int kvm_vm_ioctl_set_pit2(struct kvm *kvm, struct kvm_pit_state2 *ps)
2116 int r = 0, start = 0;
2117 u32 prev_legacy, cur_legacy;
2118 mutex_lock(&kvm->arch.vpit->pit_state.lock);
2119 prev_legacy = kvm->arch.vpit->pit_state.flags & KVM_PIT_FLAGS_HPET_LEGACY;
2120 cur_legacy = ps->flags & KVM_PIT_FLAGS_HPET_LEGACY;
2121 if (!prev_legacy && cur_legacy)
2123 memcpy(&kvm->arch.vpit->pit_state.channels, &ps->channels,
2124 sizeof(kvm->arch.vpit->pit_state.channels));
2125 kvm->arch.vpit->pit_state.flags = ps->flags;
2126 kvm_pit_load_count(kvm, 0, kvm->arch.vpit->pit_state.channels[0].count, start);
2127 mutex_unlock(&kvm->arch.vpit->pit_state.lock);
2131 static int kvm_vm_ioctl_reinject(struct kvm *kvm,
2132 struct kvm_reinject_control *control)
2134 if (!kvm->arch.vpit)
2136 mutex_lock(&kvm->arch.vpit->pit_state.lock);
2137 kvm->arch.vpit->pit_state.pit_timer.reinject = control->pit_reinject;
2138 mutex_unlock(&kvm->arch.vpit->pit_state.lock);
2143 * Get (and clear) the dirty memory log for a memory slot.
2145 int kvm_vm_ioctl_get_dirty_log(struct kvm *kvm,
2146 struct kvm_dirty_log *log)
2150 struct kvm_memory_slot *memslot;
2153 down_write(&kvm->slots_lock);
2155 r = kvm_get_dirty_log(kvm, log, &is_dirty);
2159 /* If nothing is dirty, don't bother messing with page tables. */
2161 spin_lock(&kvm->mmu_lock);
2162 kvm_mmu_slot_remove_write_access(kvm, log->slot);
2163 spin_unlock(&kvm->mmu_lock);
2164 memslot = &kvm->memslots[log->slot];
2165 n = ALIGN(memslot->npages, BITS_PER_LONG) / 8;
2166 memset(memslot->dirty_bitmap, 0, n);
2170 up_write(&kvm->slots_lock);
2174 long kvm_arch_vm_ioctl(struct file *filp,
2175 unsigned int ioctl, unsigned long arg)
2177 struct kvm *kvm = filp->private_data;
2178 void __user *argp = (void __user *)arg;
2181 * This union makes it completely explicit to gcc-3.x
2182 * that these two variables' stack usage should be
2183 * combined, not added together.
2186 struct kvm_pit_state ps;
2187 struct kvm_pit_state2 ps2;
2188 struct kvm_memory_alias alias;
2189 struct kvm_pit_config pit_config;
2193 case KVM_SET_TSS_ADDR:
2194 r = kvm_vm_ioctl_set_tss_addr(kvm, arg);
2198 case KVM_SET_IDENTITY_MAP_ADDR: {
2202 if (copy_from_user(&ident_addr, argp, sizeof ident_addr))
2204 r = kvm_vm_ioctl_set_identity_map_addr(kvm, ident_addr);
2209 case KVM_SET_MEMORY_REGION: {
2210 struct kvm_memory_region kvm_mem;
2211 struct kvm_userspace_memory_region kvm_userspace_mem;
2214 if (copy_from_user(&kvm_mem, argp, sizeof kvm_mem))
2216 kvm_userspace_mem.slot = kvm_mem.slot;
2217 kvm_userspace_mem.flags = kvm_mem.flags;
2218 kvm_userspace_mem.guest_phys_addr = kvm_mem.guest_phys_addr;
2219 kvm_userspace_mem.memory_size = kvm_mem.memory_size;
2220 r = kvm_vm_ioctl_set_memory_region(kvm, &kvm_userspace_mem, 0);
2225 case KVM_SET_NR_MMU_PAGES:
2226 r = kvm_vm_ioctl_set_nr_mmu_pages(kvm, arg);
2230 case KVM_GET_NR_MMU_PAGES:
2231 r = kvm_vm_ioctl_get_nr_mmu_pages(kvm);
2233 case KVM_SET_MEMORY_ALIAS:
2235 if (copy_from_user(&u.alias, argp, sizeof(struct kvm_memory_alias)))
2237 r = kvm_vm_ioctl_set_memory_alias(kvm, &u.alias);
2241 case KVM_CREATE_IRQCHIP:
2243 kvm->arch.vpic = kvm_create_pic(kvm);
2244 if (kvm->arch.vpic) {
2245 r = kvm_ioapic_init(kvm);
2247 kfree(kvm->arch.vpic);
2248 kvm->arch.vpic = NULL;
2253 r = kvm_setup_default_irq_routing(kvm);
2255 kfree(kvm->arch.vpic);
2256 kfree(kvm->arch.vioapic);
2260 case KVM_CREATE_PIT:
2261 u.pit_config.flags = KVM_PIT_SPEAKER_DUMMY;
2263 case KVM_CREATE_PIT2:
2265 if (copy_from_user(&u.pit_config, argp,
2266 sizeof(struct kvm_pit_config)))
2269 down_write(&kvm->slots_lock);
2272 goto create_pit_unlock;
2274 kvm->arch.vpit = kvm_create_pit(kvm, u.pit_config.flags);
2278 up_write(&kvm->slots_lock);
2280 case KVM_IRQ_LINE_STATUS:
2281 case KVM_IRQ_LINE: {
2282 struct kvm_irq_level irq_event;
2285 if (copy_from_user(&irq_event, argp, sizeof irq_event))
2287 if (irqchip_in_kernel(kvm)) {
2289 mutex_lock(&kvm->irq_lock);
2290 status = kvm_set_irq(kvm, KVM_USERSPACE_IRQ_SOURCE_ID,
2291 irq_event.irq, irq_event.level);
2292 mutex_unlock(&kvm->irq_lock);
2293 if (ioctl == KVM_IRQ_LINE_STATUS) {
2294 irq_event.status = status;
2295 if (copy_to_user(argp, &irq_event,
2303 case KVM_GET_IRQCHIP: {
2304 /* 0: PIC master, 1: PIC slave, 2: IOAPIC */
2305 struct kvm_irqchip *chip = kmalloc(sizeof(*chip), GFP_KERNEL);
2311 if (copy_from_user(chip, argp, sizeof *chip))
2312 goto get_irqchip_out;
2314 if (!irqchip_in_kernel(kvm))
2315 goto get_irqchip_out;
2316 r = kvm_vm_ioctl_get_irqchip(kvm, chip);
2318 goto get_irqchip_out;
2320 if (copy_to_user(argp, chip, sizeof *chip))
2321 goto get_irqchip_out;
2329 case KVM_SET_IRQCHIP: {
2330 /* 0: PIC master, 1: PIC slave, 2: IOAPIC */
2331 struct kvm_irqchip *chip = kmalloc(sizeof(*chip), GFP_KERNEL);
2337 if (copy_from_user(chip, argp, sizeof *chip))
2338 goto set_irqchip_out;
2340 if (!irqchip_in_kernel(kvm))
2341 goto set_irqchip_out;
2342 r = kvm_vm_ioctl_set_irqchip(kvm, chip);
2344 goto set_irqchip_out;
2354 if (copy_from_user(&u.ps, argp, sizeof(struct kvm_pit_state)))
2357 if (!kvm->arch.vpit)
2359 r = kvm_vm_ioctl_get_pit(kvm, &u.ps);
2363 if (copy_to_user(argp, &u.ps, sizeof(struct kvm_pit_state)))
2370 if (copy_from_user(&u.ps, argp, sizeof u.ps))
2373 if (!kvm->arch.vpit)
2375 r = kvm_vm_ioctl_set_pit(kvm, &u.ps);
2381 case KVM_GET_PIT2: {
2383 if (!kvm->arch.vpit)
2385 r = kvm_vm_ioctl_get_pit2(kvm, &u.ps2);
2389 if (copy_to_user(argp, &u.ps2, sizeof(u.ps2)))
2394 case KVM_SET_PIT2: {
2396 if (copy_from_user(&u.ps2, argp, sizeof(u.ps2)))
2399 if (!kvm->arch.vpit)
2401 r = kvm_vm_ioctl_set_pit2(kvm, &u.ps2);
2407 case KVM_REINJECT_CONTROL: {
2408 struct kvm_reinject_control control;
2410 if (copy_from_user(&control, argp, sizeof(control)))
2412 r = kvm_vm_ioctl_reinject(kvm, &control);
2425 static void kvm_init_msr_list(void)
2430 for (i = j = 0; i < ARRAY_SIZE(msrs_to_save); i++) {
2431 if (rdmsr_safe(msrs_to_save[i], &dummy[0], &dummy[1]) < 0)
2434 msrs_to_save[j] = msrs_to_save[i];
2437 num_msrs_to_save = j;
2440 static int vcpu_mmio_write(struct kvm_vcpu *vcpu, gpa_t addr, int len,
2443 if (vcpu->arch.apic &&
2444 !kvm_iodevice_write(&vcpu->arch.apic->dev, addr, len, v))
2447 return kvm_io_bus_write(&vcpu->kvm->mmio_bus, addr, len, v);
2450 static int vcpu_mmio_read(struct kvm_vcpu *vcpu, gpa_t addr, int len, void *v)
2452 if (vcpu->arch.apic &&
2453 !kvm_iodevice_read(&vcpu->arch.apic->dev, addr, len, v))
2456 return kvm_io_bus_read(&vcpu->kvm->mmio_bus, addr, len, v);
2459 static int kvm_read_guest_virt(gva_t addr, void *val, unsigned int bytes,
2460 struct kvm_vcpu *vcpu)
2463 int r = X86EMUL_CONTINUE;
2466 gpa_t gpa = vcpu->arch.mmu.gva_to_gpa(vcpu, addr);
2467 unsigned offset = addr & (PAGE_SIZE-1);
2468 unsigned toread = min(bytes, (unsigned)PAGE_SIZE - offset);
2471 if (gpa == UNMAPPED_GVA) {
2472 r = X86EMUL_PROPAGATE_FAULT;
2475 ret = kvm_read_guest(vcpu->kvm, gpa, data, toread);
2477 r = X86EMUL_UNHANDLEABLE;
2489 static int kvm_write_guest_virt(gva_t addr, void *val, unsigned int bytes,
2490 struct kvm_vcpu *vcpu)
2493 int r = X86EMUL_CONTINUE;
2496 gpa_t gpa = vcpu->arch.mmu.gva_to_gpa(vcpu, addr);
2497 unsigned offset = addr & (PAGE_SIZE-1);
2498 unsigned towrite = min(bytes, (unsigned)PAGE_SIZE - offset);
2501 if (gpa == UNMAPPED_GVA) {
2502 r = X86EMUL_PROPAGATE_FAULT;
2505 ret = kvm_write_guest(vcpu->kvm, gpa, data, towrite);
2507 r = X86EMUL_UNHANDLEABLE;
2520 static int emulator_read_emulated(unsigned long addr,
2523 struct kvm_vcpu *vcpu)
2527 if (vcpu->mmio_read_completed) {
2528 memcpy(val, vcpu->mmio_data, bytes);
2529 trace_kvm_mmio(KVM_TRACE_MMIO_READ, bytes,
2530 vcpu->mmio_phys_addr, *(u64 *)val);
2531 vcpu->mmio_read_completed = 0;
2532 return X86EMUL_CONTINUE;
2535 gpa = vcpu->arch.mmu.gva_to_gpa(vcpu, addr);
2537 /* For APIC access vmexit */
2538 if ((gpa & PAGE_MASK) == APIC_DEFAULT_PHYS_BASE)
2541 if (kvm_read_guest_virt(addr, val, bytes, vcpu)
2542 == X86EMUL_CONTINUE)
2543 return X86EMUL_CONTINUE;
2544 if (gpa == UNMAPPED_GVA)
2545 return X86EMUL_PROPAGATE_FAULT;
2549 * Is this MMIO handled locally?
2551 if (!vcpu_mmio_read(vcpu, gpa, bytes, val)) {
2552 trace_kvm_mmio(KVM_TRACE_MMIO_READ, bytes, gpa, *(u64 *)val);
2553 return X86EMUL_CONTINUE;
2556 trace_kvm_mmio(KVM_TRACE_MMIO_READ_UNSATISFIED, bytes, gpa, 0);
2558 vcpu->mmio_needed = 1;
2559 vcpu->mmio_phys_addr = gpa;
2560 vcpu->mmio_size = bytes;
2561 vcpu->mmio_is_write = 0;
2563 return X86EMUL_UNHANDLEABLE;
2566 int emulator_write_phys(struct kvm_vcpu *vcpu, gpa_t gpa,
2567 const void *val, int bytes)
2571 ret = kvm_write_guest(vcpu->kvm, gpa, val, bytes);
2574 kvm_mmu_pte_write(vcpu, gpa, val, bytes, 1);
2578 static int emulator_write_emulated_onepage(unsigned long addr,
2581 struct kvm_vcpu *vcpu)
2585 gpa = vcpu->arch.mmu.gva_to_gpa(vcpu, addr);
2587 if (gpa == UNMAPPED_GVA) {
2588 kvm_inject_page_fault(vcpu, addr, 2);
2589 return X86EMUL_PROPAGATE_FAULT;
2592 /* For APIC access vmexit */
2593 if ((gpa & PAGE_MASK) == APIC_DEFAULT_PHYS_BASE)
2596 if (emulator_write_phys(vcpu, gpa, val, bytes))
2597 return X86EMUL_CONTINUE;
2600 trace_kvm_mmio(KVM_TRACE_MMIO_WRITE, bytes, gpa, *(u64 *)val);
2602 * Is this MMIO handled locally?
2604 if (!vcpu_mmio_write(vcpu, gpa, bytes, val))
2605 return X86EMUL_CONTINUE;
2607 vcpu->mmio_needed = 1;
2608 vcpu->mmio_phys_addr = gpa;
2609 vcpu->mmio_size = bytes;
2610 vcpu->mmio_is_write = 1;
2611 memcpy(vcpu->mmio_data, val, bytes);
2613 return X86EMUL_CONTINUE;
2616 int emulator_write_emulated(unsigned long addr,
2619 struct kvm_vcpu *vcpu)
2621 /* Crossing a page boundary? */
2622 if (((addr + bytes - 1) ^ addr) & PAGE_MASK) {
2625 now = -addr & ~PAGE_MASK;
2626 rc = emulator_write_emulated_onepage(addr, val, now, vcpu);
2627 if (rc != X86EMUL_CONTINUE)
2633 return emulator_write_emulated_onepage(addr, val, bytes, vcpu);
2635 EXPORT_SYMBOL_GPL(emulator_write_emulated);
2637 static int emulator_cmpxchg_emulated(unsigned long addr,
2641 struct kvm_vcpu *vcpu)
2643 printk_once(KERN_WARNING "kvm: emulating exchange as write\n");
2644 #ifndef CONFIG_X86_64
2645 /* guests cmpxchg8b have to be emulated atomically */
2652 gpa = vcpu->arch.mmu.gva_to_gpa(vcpu, addr);
2654 if (gpa == UNMAPPED_GVA ||
2655 (gpa & PAGE_MASK) == APIC_DEFAULT_PHYS_BASE)
2658 if (((gpa + bytes - 1) & PAGE_MASK) != (gpa & PAGE_MASK))
2663 page = gfn_to_page(vcpu->kvm, gpa >> PAGE_SHIFT);
2665 kaddr = kmap_atomic(page, KM_USER0);
2666 set_64bit((u64 *)(kaddr + offset_in_page(gpa)), val);
2667 kunmap_atomic(kaddr, KM_USER0);
2668 kvm_release_page_dirty(page);
2673 return emulator_write_emulated(addr, new, bytes, vcpu);
2676 static unsigned long get_segment_base(struct kvm_vcpu *vcpu, int seg)
2678 return kvm_x86_ops->get_segment_base(vcpu, seg);
2681 int emulate_invlpg(struct kvm_vcpu *vcpu, gva_t address)
2683 kvm_mmu_invlpg(vcpu, address);
2684 return X86EMUL_CONTINUE;
2687 int emulate_clts(struct kvm_vcpu *vcpu)
2689 kvm_x86_ops->set_cr0(vcpu, vcpu->arch.cr0 & ~X86_CR0_TS);
2690 return X86EMUL_CONTINUE;
2693 int emulator_get_dr(struct x86_emulate_ctxt *ctxt, int dr, unsigned long *dest)
2695 struct kvm_vcpu *vcpu = ctxt->vcpu;
2699 *dest = kvm_x86_ops->get_dr(vcpu, dr);
2700 return X86EMUL_CONTINUE;
2702 pr_unimpl(vcpu, "%s: unexpected dr %u\n", __func__, dr);
2703 return X86EMUL_UNHANDLEABLE;
2707 int emulator_set_dr(struct x86_emulate_ctxt *ctxt, int dr, unsigned long value)
2709 unsigned long mask = (ctxt->mode == X86EMUL_MODE_PROT64) ? ~0ULL : ~0U;
2712 kvm_x86_ops->set_dr(ctxt->vcpu, dr, value & mask, &exception);
2714 /* FIXME: better handling */
2715 return X86EMUL_UNHANDLEABLE;
2717 return X86EMUL_CONTINUE;
2720 void kvm_report_emulation_failure(struct kvm_vcpu *vcpu, const char *context)
2723 unsigned long rip = kvm_rip_read(vcpu);
2724 unsigned long rip_linear;
2726 if (!printk_ratelimit())
2729 rip_linear = rip + get_segment_base(vcpu, VCPU_SREG_CS);
2731 kvm_read_guest_virt(rip_linear, (void *)opcodes, 4, vcpu);
2733 printk(KERN_ERR "emulation failed (%s) rip %lx %02x %02x %02x %02x\n",
2734 context, rip, opcodes[0], opcodes[1], opcodes[2], opcodes[3]);
2736 EXPORT_SYMBOL_GPL(kvm_report_emulation_failure);
2738 static struct x86_emulate_ops emulate_ops = {
2739 .read_std = kvm_read_guest_virt,
2740 .read_emulated = emulator_read_emulated,
2741 .write_emulated = emulator_write_emulated,
2742 .cmpxchg_emulated = emulator_cmpxchg_emulated,
2745 static void cache_all_regs(struct kvm_vcpu *vcpu)
2747 kvm_register_read(vcpu, VCPU_REGS_RAX);
2748 kvm_register_read(vcpu, VCPU_REGS_RSP);
2749 kvm_register_read(vcpu, VCPU_REGS_RIP);
2750 vcpu->arch.regs_dirty = ~0;
2753 int emulate_instruction(struct kvm_vcpu *vcpu,
2759 struct decode_cache *c;
2760 struct kvm_run *run = vcpu->run;
2762 kvm_clear_exception_queue(vcpu);
2763 vcpu->arch.mmio_fault_cr2 = cr2;
2765 * TODO: fix emulate.c to use guest_read/write_register
2766 * instead of direct ->regs accesses, can save hundred cycles
2767 * on Intel for instructions that don't read/change RSP, for
2770 cache_all_regs(vcpu);
2772 vcpu->mmio_is_write = 0;
2773 vcpu->arch.pio.string = 0;
2775 if (!(emulation_type & EMULTYPE_NO_DECODE)) {
2777 kvm_x86_ops->get_cs_db_l_bits(vcpu, &cs_db, &cs_l);
2779 vcpu->arch.emulate_ctxt.vcpu = vcpu;
2780 vcpu->arch.emulate_ctxt.eflags = kvm_x86_ops->get_rflags(vcpu);
2781 vcpu->arch.emulate_ctxt.mode =
2782 (vcpu->arch.emulate_ctxt.eflags & X86_EFLAGS_VM)
2783 ? X86EMUL_MODE_REAL : cs_l
2784 ? X86EMUL_MODE_PROT64 : cs_db
2785 ? X86EMUL_MODE_PROT32 : X86EMUL_MODE_PROT16;
2787 r = x86_decode_insn(&vcpu->arch.emulate_ctxt, &emulate_ops);
2789 /* Only allow emulation of specific instructions on #UD
2790 * (namely VMMCALL, sysenter, sysexit, syscall)*/
2791 c = &vcpu->arch.emulate_ctxt.decode;
2792 if (emulation_type & EMULTYPE_TRAP_UD) {
2794 return EMULATE_FAIL;
2796 case 0x01: /* VMMCALL */
2797 if (c->modrm_mod != 3 || c->modrm_rm != 1)
2798 return EMULATE_FAIL;
2800 case 0x34: /* sysenter */
2801 case 0x35: /* sysexit */
2802 if (c->modrm_mod != 0 || c->modrm_rm != 0)
2803 return EMULATE_FAIL;
2805 case 0x05: /* syscall */
2806 if (c->modrm_mod != 0 || c->modrm_rm != 0)
2807 return EMULATE_FAIL;
2810 return EMULATE_FAIL;
2813 if (!(c->modrm_reg == 0 || c->modrm_reg == 3))
2814 return EMULATE_FAIL;
2817 ++vcpu->stat.insn_emulation;
2819 ++vcpu->stat.insn_emulation_fail;
2820 if (kvm_mmu_unprotect_page_virt(vcpu, cr2))
2821 return EMULATE_DONE;
2822 return EMULATE_FAIL;
2826 if (emulation_type & EMULTYPE_SKIP) {
2827 kvm_rip_write(vcpu, vcpu->arch.emulate_ctxt.decode.eip);
2828 return EMULATE_DONE;
2831 r = x86_emulate_insn(&vcpu->arch.emulate_ctxt, &emulate_ops);
2832 shadow_mask = vcpu->arch.emulate_ctxt.interruptibility;
2835 kvm_x86_ops->set_interrupt_shadow(vcpu, shadow_mask);
2837 if (vcpu->arch.pio.string)
2838 return EMULATE_DO_MMIO;
2840 if ((r || vcpu->mmio_is_write) && run) {
2841 run->exit_reason = KVM_EXIT_MMIO;
2842 run->mmio.phys_addr = vcpu->mmio_phys_addr;
2843 memcpy(run->mmio.data, vcpu->mmio_data, 8);
2844 run->mmio.len = vcpu->mmio_size;
2845 run->mmio.is_write = vcpu->mmio_is_write;
2849 if (kvm_mmu_unprotect_page_virt(vcpu, cr2))
2850 return EMULATE_DONE;
2851 if (!vcpu->mmio_needed) {
2852 kvm_report_emulation_failure(vcpu, "mmio");
2853 return EMULATE_FAIL;
2855 return EMULATE_DO_MMIO;
2858 kvm_x86_ops->set_rflags(vcpu, vcpu->arch.emulate_ctxt.eflags);
2860 if (vcpu->mmio_is_write) {
2861 vcpu->mmio_needed = 0;
2862 return EMULATE_DO_MMIO;
2865 return EMULATE_DONE;
2867 EXPORT_SYMBOL_GPL(emulate_instruction);
2869 static int pio_copy_data(struct kvm_vcpu *vcpu)
2871 void *p = vcpu->arch.pio_data;
2872 gva_t q = vcpu->arch.pio.guest_gva;
2876 bytes = vcpu->arch.pio.size * vcpu->arch.pio.cur_count;
2877 if (vcpu->arch.pio.in)
2878 ret = kvm_write_guest_virt(q, p, bytes, vcpu);
2880 ret = kvm_read_guest_virt(q, p, bytes, vcpu);
2884 int complete_pio(struct kvm_vcpu *vcpu)
2886 struct kvm_pio_request *io = &vcpu->arch.pio;
2893 val = kvm_register_read(vcpu, VCPU_REGS_RAX);
2894 memcpy(&val, vcpu->arch.pio_data, io->size);
2895 kvm_register_write(vcpu, VCPU_REGS_RAX, val);
2899 r = pio_copy_data(vcpu);
2906 delta *= io->cur_count;
2908 * The size of the register should really depend on
2909 * current address size.
2911 val = kvm_register_read(vcpu, VCPU_REGS_RCX);
2913 kvm_register_write(vcpu, VCPU_REGS_RCX, val);
2919 val = kvm_register_read(vcpu, VCPU_REGS_RDI);
2921 kvm_register_write(vcpu, VCPU_REGS_RDI, val);
2923 val = kvm_register_read(vcpu, VCPU_REGS_RSI);
2925 kvm_register_write(vcpu, VCPU_REGS_RSI, val);
2929 io->count -= io->cur_count;
2935 static int kernel_pio(struct kvm_vcpu *vcpu, void *pd)
2937 /* TODO: String I/O for in kernel device */
2940 if (vcpu->arch.pio.in)
2941 r = kvm_io_bus_read(&vcpu->kvm->pio_bus, vcpu->arch.pio.port,
2942 vcpu->arch.pio.size, pd);
2944 r = kvm_io_bus_write(&vcpu->kvm->pio_bus, vcpu->arch.pio.port,
2945 vcpu->arch.pio.size, pd);
2949 static int pio_string_write(struct kvm_vcpu *vcpu)
2951 struct kvm_pio_request *io = &vcpu->arch.pio;
2952 void *pd = vcpu->arch.pio_data;
2955 for (i = 0; i < io->cur_count; i++) {
2956 if (kvm_io_bus_write(&vcpu->kvm->pio_bus,
2957 io->port, io->size, pd)) {
2966 int kvm_emulate_pio(struct kvm_vcpu *vcpu, int in, int size, unsigned port)
2970 vcpu->run->exit_reason = KVM_EXIT_IO;
2971 vcpu->run->io.direction = in ? KVM_EXIT_IO_IN : KVM_EXIT_IO_OUT;
2972 vcpu->run->io.size = vcpu->arch.pio.size = size;
2973 vcpu->run->io.data_offset = KVM_PIO_PAGE_OFFSET * PAGE_SIZE;
2974 vcpu->run->io.count = vcpu->arch.pio.count = vcpu->arch.pio.cur_count = 1;
2975 vcpu->run->io.port = vcpu->arch.pio.port = port;
2976 vcpu->arch.pio.in = in;
2977 vcpu->arch.pio.string = 0;
2978 vcpu->arch.pio.down = 0;
2979 vcpu->arch.pio.rep = 0;
2981 trace_kvm_pio(vcpu->run->io.direction == KVM_EXIT_IO_OUT, port,
2984 val = kvm_register_read(vcpu, VCPU_REGS_RAX);
2985 memcpy(vcpu->arch.pio_data, &val, 4);
2987 if (!kernel_pio(vcpu, vcpu->arch.pio_data)) {
2993 EXPORT_SYMBOL_GPL(kvm_emulate_pio);
2995 int kvm_emulate_pio_string(struct kvm_vcpu *vcpu, int in,
2996 int size, unsigned long count, int down,
2997 gva_t address, int rep, unsigned port)
2999 unsigned now, in_page;
3002 vcpu->run->exit_reason = KVM_EXIT_IO;
3003 vcpu->run->io.direction = in ? KVM_EXIT_IO_IN : KVM_EXIT_IO_OUT;
3004 vcpu->run->io.size = vcpu->arch.pio.size = size;
3005 vcpu->run->io.data_offset = KVM_PIO_PAGE_OFFSET * PAGE_SIZE;
3006 vcpu->run->io.count = vcpu->arch.pio.count = vcpu->arch.pio.cur_count = count;
3007 vcpu->run->io.port = vcpu->arch.pio.port = port;
3008 vcpu->arch.pio.in = in;
3009 vcpu->arch.pio.string = 1;
3010 vcpu->arch.pio.down = down;
3011 vcpu->arch.pio.rep = rep;
3013 trace_kvm_pio(vcpu->run->io.direction == KVM_EXIT_IO_OUT, port,
3017 kvm_x86_ops->skip_emulated_instruction(vcpu);
3022 in_page = PAGE_SIZE - offset_in_page(address);
3024 in_page = offset_in_page(address) + size;
3025 now = min(count, (unsigned long)in_page / size);
3030 * String I/O in reverse. Yuck. Kill the guest, fix later.
3032 pr_unimpl(vcpu, "guest string pio down\n");
3033 kvm_inject_gp(vcpu, 0);
3036 vcpu->run->io.count = now;
3037 vcpu->arch.pio.cur_count = now;
3039 if (vcpu->arch.pio.cur_count == vcpu->arch.pio.count)
3040 kvm_x86_ops->skip_emulated_instruction(vcpu);
3042 vcpu->arch.pio.guest_gva = address;
3044 if (!vcpu->arch.pio.in) {
3045 /* string PIO write */
3046 ret = pio_copy_data(vcpu);
3047 if (ret == X86EMUL_PROPAGATE_FAULT) {
3048 kvm_inject_gp(vcpu, 0);
3051 if (ret == 0 && !pio_string_write(vcpu)) {
3053 if (vcpu->arch.pio.count == 0)
3057 /* no string PIO read support yet */
3061 EXPORT_SYMBOL_GPL(kvm_emulate_pio_string);
3063 static void bounce_off(void *info)
3068 static unsigned int ref_freq;
3069 static unsigned long tsc_khz_ref;
3071 static int kvmclock_cpufreq_notifier(struct notifier_block *nb, unsigned long val,
3074 struct cpufreq_freqs *freq = data;
3076 struct kvm_vcpu *vcpu;
3077 int i, send_ipi = 0;
3080 ref_freq = freq->old;
3082 if (val == CPUFREQ_PRECHANGE && freq->old > freq->new)
3084 if (val == CPUFREQ_POSTCHANGE && freq->old < freq->new)
3086 per_cpu(cpu_tsc_khz, freq->cpu) = cpufreq_scale(tsc_khz_ref, ref_freq, freq->new);
3088 spin_lock(&kvm_lock);
3089 list_for_each_entry(kvm, &vm_list, vm_list) {
3090 kvm_for_each_vcpu(i, vcpu, kvm) {
3091 if (vcpu->cpu != freq->cpu)
3093 if (!kvm_request_guest_time_update(vcpu))
3095 if (vcpu->cpu != smp_processor_id())
3099 spin_unlock(&kvm_lock);
3101 if (freq->old < freq->new && send_ipi) {
3103 * We upscale the frequency. Must make the guest
3104 * doesn't see old kvmclock values while running with
3105 * the new frequency, otherwise we risk the guest sees
3106 * time go backwards.
3108 * In case we update the frequency for another cpu
3109 * (which might be in guest context) send an interrupt
3110 * to kick the cpu out of guest context. Next time
3111 * guest context is entered kvmclock will be updated,
3112 * so the guest will not see stale values.
3114 smp_call_function_single(freq->cpu, bounce_off, NULL, 1);
3119 static struct notifier_block kvmclock_cpufreq_notifier_block = {
3120 .notifier_call = kvmclock_cpufreq_notifier
3123 int kvm_arch_init(void *opaque)
3126 struct kvm_x86_ops *ops = (struct kvm_x86_ops *)opaque;
3129 printk(KERN_ERR "kvm: already loaded the other module\n");
3134 if (!ops->cpu_has_kvm_support()) {
3135 printk(KERN_ERR "kvm: no hardware support\n");
3139 if (ops->disabled_by_bios()) {
3140 printk(KERN_ERR "kvm: disabled by bios\n");
3145 r = kvm_mmu_module_init();
3149 kvm_init_msr_list();
3152 kvm_mmu_set_nonpresent_ptes(0ull, 0ull);
3153 kvm_mmu_set_base_ptes(PT_PRESENT_MASK);
3154 kvm_mmu_set_mask_ptes(PT_USER_MASK, PT_ACCESSED_MASK,
3155 PT_DIRTY_MASK, PT64_NX_MASK, 0);
3157 for_each_possible_cpu(cpu)
3158 per_cpu(cpu_tsc_khz, cpu) = tsc_khz;
3159 if (!boot_cpu_has(X86_FEATURE_CONSTANT_TSC)) {
3160 tsc_khz_ref = tsc_khz;
3161 cpufreq_register_notifier(&kvmclock_cpufreq_notifier_block,
3162 CPUFREQ_TRANSITION_NOTIFIER);
3171 void kvm_arch_exit(void)
3173 if (!boot_cpu_has(X86_FEATURE_CONSTANT_TSC))
3174 cpufreq_unregister_notifier(&kvmclock_cpufreq_notifier_block,
3175 CPUFREQ_TRANSITION_NOTIFIER);
3177 kvm_mmu_module_exit();
3180 int kvm_emulate_halt(struct kvm_vcpu *vcpu)
3182 ++vcpu->stat.halt_exits;
3183 if (irqchip_in_kernel(vcpu->kvm)) {
3184 vcpu->arch.mp_state = KVM_MP_STATE_HALTED;
3187 vcpu->run->exit_reason = KVM_EXIT_HLT;
3191 EXPORT_SYMBOL_GPL(kvm_emulate_halt);
3193 static inline gpa_t hc_gpa(struct kvm_vcpu *vcpu, unsigned long a0,
3196 if (is_long_mode(vcpu))
3199 return a0 | ((gpa_t)a1 << 32);
3202 int kvm_emulate_hypercall(struct kvm_vcpu *vcpu)
3204 unsigned long nr, a0, a1, a2, a3, ret;
3207 nr = kvm_register_read(vcpu, VCPU_REGS_RAX);
3208 a0 = kvm_register_read(vcpu, VCPU_REGS_RBX);
3209 a1 = kvm_register_read(vcpu, VCPU_REGS_RCX);
3210 a2 = kvm_register_read(vcpu, VCPU_REGS_RDX);
3211 a3 = kvm_register_read(vcpu, VCPU_REGS_RSI);
3213 trace_kvm_hypercall(nr, a0, a1, a2, a3);
3215 if (!is_long_mode(vcpu)) {
3223 if (kvm_x86_ops->get_cpl(vcpu) != 0) {
3229 case KVM_HC_VAPIC_POLL_IRQ:
3233 r = kvm_pv_mmu_op(vcpu, a0, hc_gpa(vcpu, a1, a2), &ret);
3240 kvm_register_write(vcpu, VCPU_REGS_RAX, ret);
3241 ++vcpu->stat.hypercalls;
3244 EXPORT_SYMBOL_GPL(kvm_emulate_hypercall);
3246 int kvm_fix_hypercall(struct kvm_vcpu *vcpu)
3248 char instruction[3];
3250 unsigned long rip = kvm_rip_read(vcpu);
3254 * Blow out the MMU to ensure that no other VCPU has an active mapping
3255 * to ensure that the updated hypercall appears atomically across all
3258 kvm_mmu_zap_all(vcpu->kvm);
3260 kvm_x86_ops->patch_hypercall(vcpu, instruction);
3261 if (emulator_write_emulated(rip, instruction, 3, vcpu)
3262 != X86EMUL_CONTINUE)
3268 static u64 mk_cr_64(u64 curr_cr, u32 new_val)
3270 return (curr_cr & ~((1ULL << 32) - 1)) | new_val;
3273 void realmode_lgdt(struct kvm_vcpu *vcpu, u16 limit, unsigned long base)
3275 struct descriptor_table dt = { limit, base };
3277 kvm_x86_ops->set_gdt(vcpu, &dt);
3280 void realmode_lidt(struct kvm_vcpu *vcpu, u16 limit, unsigned long base)
3282 struct descriptor_table dt = { limit, base };
3284 kvm_x86_ops->set_idt(vcpu, &dt);
3287 void realmode_lmsw(struct kvm_vcpu *vcpu, unsigned long msw,
3288 unsigned long *rflags)
3290 kvm_lmsw(vcpu, msw);
3291 *rflags = kvm_x86_ops->get_rflags(vcpu);
3294 unsigned long realmode_get_cr(struct kvm_vcpu *vcpu, int cr)
3296 unsigned long value;
3298 kvm_x86_ops->decache_cr4_guest_bits(vcpu);
3301 value = vcpu->arch.cr0;
3304 value = vcpu->arch.cr2;
3307 value = vcpu->arch.cr3;
3310 value = vcpu->arch.cr4;
3313 value = kvm_get_cr8(vcpu);
3316 vcpu_printf(vcpu, "%s: unexpected cr %u\n", __func__, cr);
3323 void realmode_set_cr(struct kvm_vcpu *vcpu, int cr, unsigned long val,
3324 unsigned long *rflags)
3328 kvm_set_cr0(vcpu, mk_cr_64(vcpu->arch.cr0, val));
3329 *rflags = kvm_x86_ops->get_rflags(vcpu);
3332 vcpu->arch.cr2 = val;
3335 kvm_set_cr3(vcpu, val);
3338 kvm_set_cr4(vcpu, mk_cr_64(vcpu->arch.cr4, val));
3341 kvm_set_cr8(vcpu, val & 0xfUL);
3344 vcpu_printf(vcpu, "%s: unexpected cr %u\n", __func__, cr);
3348 static int move_to_next_stateful_cpuid_entry(struct kvm_vcpu *vcpu, int i)
3350 struct kvm_cpuid_entry2 *e = &vcpu->arch.cpuid_entries[i];
3351 int j, nent = vcpu->arch.cpuid_nent;
3353 e->flags &= ~KVM_CPUID_FLAG_STATE_READ_NEXT;
3354 /* when no next entry is found, the current entry[i] is reselected */
3355 for (j = i + 1; ; j = (j + 1) % nent) {
3356 struct kvm_cpuid_entry2 *ej = &vcpu->arch.cpuid_entries[j];
3357 if (ej->function == e->function) {
3358 ej->flags |= KVM_CPUID_FLAG_STATE_READ_NEXT;
3362 return 0; /* silence gcc, even though control never reaches here */
3365 /* find an entry with matching function, matching index (if needed), and that
3366 * should be read next (if it's stateful) */
3367 static int is_matching_cpuid_entry(struct kvm_cpuid_entry2 *e,
3368 u32 function, u32 index)
3370 if (e->function != function)
3372 if ((e->flags & KVM_CPUID_FLAG_SIGNIFCANT_INDEX) && e->index != index)
3374 if ((e->flags & KVM_CPUID_FLAG_STATEFUL_FUNC) &&
3375 !(e->flags & KVM_CPUID_FLAG_STATE_READ_NEXT))
3380 struct kvm_cpuid_entry2 *kvm_find_cpuid_entry(struct kvm_vcpu *vcpu,
3381 u32 function, u32 index)
3384 struct kvm_cpuid_entry2 *best = NULL;
3386 for (i = 0; i < vcpu->arch.cpuid_nent; ++i) {
3387 struct kvm_cpuid_entry2 *e;
3389 e = &vcpu->arch.cpuid_entries[i];
3390 if (is_matching_cpuid_entry(e, function, index)) {
3391 if (e->flags & KVM_CPUID_FLAG_STATEFUL_FUNC)
3392 move_to_next_stateful_cpuid_entry(vcpu, i);
3397 * Both basic or both extended?
3399 if (((e->function ^ function) & 0x80000000) == 0)
3400 if (!best || e->function > best->function)
3406 int cpuid_maxphyaddr(struct kvm_vcpu *vcpu)
3408 struct kvm_cpuid_entry2 *best;
3410 best = kvm_find_cpuid_entry(vcpu, 0x80000008, 0);
3412 return best->eax & 0xff;
3416 void kvm_emulate_cpuid(struct kvm_vcpu *vcpu)
3418 u32 function, index;
3419 struct kvm_cpuid_entry2 *best;
3421 function = kvm_register_read(vcpu, VCPU_REGS_RAX);
3422 index = kvm_register_read(vcpu, VCPU_REGS_RCX);
3423 kvm_register_write(vcpu, VCPU_REGS_RAX, 0);
3424 kvm_register_write(vcpu, VCPU_REGS_RBX, 0);
3425 kvm_register_write(vcpu, VCPU_REGS_RCX, 0);
3426 kvm_register_write(vcpu, VCPU_REGS_RDX, 0);
3427 best = kvm_find_cpuid_entry(vcpu, function, index);
3429 kvm_register_write(vcpu, VCPU_REGS_RAX, best->eax);
3430 kvm_register_write(vcpu, VCPU_REGS_RBX, best->ebx);
3431 kvm_register_write(vcpu, VCPU_REGS_RCX, best->ecx);
3432 kvm_register_write(vcpu, VCPU_REGS_RDX, best->edx);
3434 kvm_x86_ops->skip_emulated_instruction(vcpu);
3435 trace_kvm_cpuid(function,
3436 kvm_register_read(vcpu, VCPU_REGS_RAX),
3437 kvm_register_read(vcpu, VCPU_REGS_RBX),
3438 kvm_register_read(vcpu, VCPU_REGS_RCX),
3439 kvm_register_read(vcpu, VCPU_REGS_RDX));
3441 EXPORT_SYMBOL_GPL(kvm_emulate_cpuid);
3444 * Check if userspace requested an interrupt window, and that the
3445 * interrupt window is open.
3447 * No need to exit to userspace if we already have an interrupt queued.
3449 static int dm_request_for_irq_injection(struct kvm_vcpu *vcpu)
3451 return (!irqchip_in_kernel(vcpu->kvm) && !kvm_cpu_has_interrupt(vcpu) &&
3452 vcpu->run->request_interrupt_window &&
3453 kvm_arch_interrupt_allowed(vcpu));
3456 static void post_kvm_run_save(struct kvm_vcpu *vcpu)
3458 struct kvm_run *kvm_run = vcpu->run;
3460 kvm_run->if_flag = (kvm_x86_ops->get_rflags(vcpu) & X86_EFLAGS_IF) != 0;
3461 kvm_run->cr8 = kvm_get_cr8(vcpu);
3462 kvm_run->apic_base = kvm_get_apic_base(vcpu);
3463 if (irqchip_in_kernel(vcpu->kvm))
3464 kvm_run->ready_for_interrupt_injection = 1;
3466 kvm_run->ready_for_interrupt_injection =
3467 kvm_arch_interrupt_allowed(vcpu) &&
3468 !kvm_cpu_has_interrupt(vcpu) &&
3469 !kvm_event_needs_reinjection(vcpu);
3472 static void vapic_enter(struct kvm_vcpu *vcpu)
3474 struct kvm_lapic *apic = vcpu->arch.apic;
3477 if (!apic || !apic->vapic_addr)
3480 page = gfn_to_page(vcpu->kvm, apic->vapic_addr >> PAGE_SHIFT);
3482 vcpu->arch.apic->vapic_page = page;
3485 static void vapic_exit(struct kvm_vcpu *vcpu)
3487 struct kvm_lapic *apic = vcpu->arch.apic;
3489 if (!apic || !apic->vapic_addr)
3492 down_read(&vcpu->kvm->slots_lock);
3493 kvm_release_page_dirty(apic->vapic_page);
3494 mark_page_dirty(vcpu->kvm, apic->vapic_addr >> PAGE_SHIFT);
3495 up_read(&vcpu->kvm->slots_lock);
3498 static void update_cr8_intercept(struct kvm_vcpu *vcpu)
3502 if (!kvm_x86_ops->update_cr8_intercept)
3505 if (!vcpu->arch.apic)
3508 if (!vcpu->arch.apic->vapic_addr)
3509 max_irr = kvm_lapic_find_highest_irr(vcpu);
3516 tpr = kvm_lapic_get_cr8(vcpu);
3518 kvm_x86_ops->update_cr8_intercept(vcpu, tpr, max_irr);
3521 static void inject_pending_event(struct kvm_vcpu *vcpu)
3523 /* try to reinject previous events if any */
3524 if (vcpu->arch.exception.pending) {
3525 kvm_x86_ops->queue_exception(vcpu, vcpu->arch.exception.nr,
3526 vcpu->arch.exception.has_error_code,
3527 vcpu->arch.exception.error_code);
3531 if (vcpu->arch.nmi_injected) {
3532 kvm_x86_ops->set_nmi(vcpu);
3536 if (vcpu->arch.interrupt.pending) {
3537 kvm_x86_ops->set_irq(vcpu);
3541 /* try to inject new event if pending */
3542 if (vcpu->arch.nmi_pending) {
3543 if (kvm_x86_ops->nmi_allowed(vcpu)) {
3544 vcpu->arch.nmi_pending = false;
3545 vcpu->arch.nmi_injected = true;
3546 kvm_x86_ops->set_nmi(vcpu);
3548 } else if (kvm_cpu_has_interrupt(vcpu)) {
3549 if (kvm_x86_ops->interrupt_allowed(vcpu)) {
3550 kvm_queue_interrupt(vcpu, kvm_cpu_get_interrupt(vcpu),
3552 kvm_x86_ops->set_irq(vcpu);
3557 static int vcpu_enter_guest(struct kvm_vcpu *vcpu)
3560 bool req_int_win = !irqchip_in_kernel(vcpu->kvm) &&
3561 vcpu->run->request_interrupt_window;
3564 if (test_and_clear_bit(KVM_REQ_MMU_RELOAD, &vcpu->requests))
3565 kvm_mmu_unload(vcpu);
3567 r = kvm_mmu_reload(vcpu);
3571 if (vcpu->requests) {
3572 if (test_and_clear_bit(KVM_REQ_MIGRATE_TIMER, &vcpu->requests))
3573 __kvm_migrate_timers(vcpu);
3574 if (test_and_clear_bit(KVM_REQ_KVMCLOCK_UPDATE, &vcpu->requests))
3575 kvm_write_guest_time(vcpu);
3576 if (test_and_clear_bit(KVM_REQ_MMU_SYNC, &vcpu->requests))
3577 kvm_mmu_sync_roots(vcpu);
3578 if (test_and_clear_bit(KVM_REQ_TLB_FLUSH, &vcpu->requests))
3579 kvm_x86_ops->tlb_flush(vcpu);
3580 if (test_and_clear_bit(KVM_REQ_REPORT_TPR_ACCESS,
3582 vcpu->run->exit_reason = KVM_EXIT_TPR_ACCESS;
3586 if (test_and_clear_bit(KVM_REQ_TRIPLE_FAULT, &vcpu->requests)) {
3587 vcpu->run->exit_reason = KVM_EXIT_SHUTDOWN;
3595 kvm_x86_ops->prepare_guest_switch(vcpu);
3596 kvm_load_guest_fpu(vcpu);
3598 local_irq_disable();
3600 clear_bit(KVM_REQ_KICK, &vcpu->requests);
3601 smp_mb__after_clear_bit();
3603 if (vcpu->requests || need_resched() || signal_pending(current)) {
3604 set_bit(KVM_REQ_KICK, &vcpu->requests);
3611 inject_pending_event(vcpu);
3613 /* enable NMI/IRQ window open exits if needed */
3614 if (vcpu->arch.nmi_pending)
3615 kvm_x86_ops->enable_nmi_window(vcpu);
3616 else if (kvm_cpu_has_interrupt(vcpu) || req_int_win)
3617 kvm_x86_ops->enable_irq_window(vcpu);
3619 if (kvm_lapic_enabled(vcpu)) {
3620 update_cr8_intercept(vcpu);
3621 kvm_lapic_sync_to_vapic(vcpu);
3624 up_read(&vcpu->kvm->slots_lock);
3628 if (unlikely(vcpu->arch.switch_db_regs)) {
3630 set_debugreg(vcpu->arch.eff_db[0], 0);
3631 set_debugreg(vcpu->arch.eff_db[1], 1);
3632 set_debugreg(vcpu->arch.eff_db[2], 2);
3633 set_debugreg(vcpu->arch.eff_db[3], 3);
3636 trace_kvm_entry(vcpu->vcpu_id);
3637 kvm_x86_ops->run(vcpu);
3639 if (unlikely(vcpu->arch.switch_db_regs || test_thread_flag(TIF_DEBUG))) {
3640 set_debugreg(current->thread.debugreg0, 0);
3641 set_debugreg(current->thread.debugreg1, 1);
3642 set_debugreg(current->thread.debugreg2, 2);
3643 set_debugreg(current->thread.debugreg3, 3);
3644 set_debugreg(current->thread.debugreg6, 6);
3645 set_debugreg(current->thread.debugreg7, 7);
3648 set_bit(KVM_REQ_KICK, &vcpu->requests);
3654 * We must have an instruction between local_irq_enable() and
3655 * kvm_guest_exit(), so the timer interrupt isn't delayed by
3656 * the interrupt shadow. The stat.exits increment will do nicely.
3657 * But we need to prevent reordering, hence this barrier():
3665 down_read(&vcpu->kvm->slots_lock);
3668 * Profile KVM exit RIPs:
3670 if (unlikely(prof_on == KVM_PROFILING)) {
3671 unsigned long rip = kvm_rip_read(vcpu);
3672 profile_hit(KVM_PROFILING, (void *)rip);
3676 kvm_lapic_sync_from_vapic(vcpu);
3678 r = kvm_x86_ops->handle_exit(vcpu);
3684 static int __vcpu_run(struct kvm_vcpu *vcpu)
3688 if (unlikely(vcpu->arch.mp_state == KVM_MP_STATE_SIPI_RECEIVED)) {
3689 pr_debug("vcpu %d received sipi with vector # %x\n",
3690 vcpu->vcpu_id, vcpu->arch.sipi_vector);
3691 kvm_lapic_reset(vcpu);
3692 r = kvm_arch_vcpu_reset(vcpu);
3695 vcpu->arch.mp_state = KVM_MP_STATE_RUNNABLE;
3698 down_read(&vcpu->kvm->slots_lock);
3703 if (vcpu->arch.mp_state == KVM_MP_STATE_RUNNABLE)
3704 r = vcpu_enter_guest(vcpu);
3706 up_read(&vcpu->kvm->slots_lock);
3707 kvm_vcpu_block(vcpu);
3708 down_read(&vcpu->kvm->slots_lock);
3709 if (test_and_clear_bit(KVM_REQ_UNHALT, &vcpu->requests))
3711 switch(vcpu->arch.mp_state) {
3712 case KVM_MP_STATE_HALTED:
3713 vcpu->arch.mp_state =
3714 KVM_MP_STATE_RUNNABLE;
3715 case KVM_MP_STATE_RUNNABLE:
3717 case KVM_MP_STATE_SIPI_RECEIVED:
3728 clear_bit(KVM_REQ_PENDING_TIMER, &vcpu->requests);
3729 if (kvm_cpu_has_pending_timer(vcpu))
3730 kvm_inject_pending_timer_irqs(vcpu);
3732 if (dm_request_for_irq_injection(vcpu)) {
3734 vcpu->run->exit_reason = KVM_EXIT_INTR;
3735 ++vcpu->stat.request_irq_exits;
3737 if (signal_pending(current)) {
3739 vcpu->run->exit_reason = KVM_EXIT_INTR;
3740 ++vcpu->stat.signal_exits;
3742 if (need_resched()) {
3743 up_read(&vcpu->kvm->slots_lock);
3745 down_read(&vcpu->kvm->slots_lock);
3749 up_read(&vcpu->kvm->slots_lock);
3750 post_kvm_run_save(vcpu);
3757 int kvm_arch_vcpu_ioctl_run(struct kvm_vcpu *vcpu, struct kvm_run *kvm_run)
3764 if (vcpu->sigset_active)
3765 sigprocmask(SIG_SETMASK, &vcpu->sigset, &sigsaved);
3767 if (unlikely(vcpu->arch.mp_state == KVM_MP_STATE_UNINITIALIZED)) {
3768 kvm_vcpu_block(vcpu);
3769 clear_bit(KVM_REQ_UNHALT, &vcpu->requests);
3774 /* re-sync apic's tpr */
3775 if (!irqchip_in_kernel(vcpu->kvm))
3776 kvm_set_cr8(vcpu, kvm_run->cr8);
3778 if (vcpu->arch.pio.cur_count) {
3779 r = complete_pio(vcpu);
3783 #if CONFIG_HAS_IOMEM
3784 if (vcpu->mmio_needed) {
3785 memcpy(vcpu->mmio_data, kvm_run->mmio.data, 8);
3786 vcpu->mmio_read_completed = 1;
3787 vcpu->mmio_needed = 0;
3789 down_read(&vcpu->kvm->slots_lock);
3790 r = emulate_instruction(vcpu, vcpu->arch.mmio_fault_cr2, 0,
3791 EMULTYPE_NO_DECODE);
3792 up_read(&vcpu->kvm->slots_lock);
3793 if (r == EMULATE_DO_MMIO) {
3795 * Read-modify-write. Back to userspace.
3802 if (kvm_run->exit_reason == KVM_EXIT_HYPERCALL)
3803 kvm_register_write(vcpu, VCPU_REGS_RAX,
3804 kvm_run->hypercall.ret);
3806 r = __vcpu_run(vcpu);
3809 if (vcpu->sigset_active)
3810 sigprocmask(SIG_SETMASK, &sigsaved, NULL);
3816 int kvm_arch_vcpu_ioctl_get_regs(struct kvm_vcpu *vcpu, struct kvm_regs *regs)
3820 regs->rax = kvm_register_read(vcpu, VCPU_REGS_RAX);
3821 regs->rbx = kvm_register_read(vcpu, VCPU_REGS_RBX);
3822 regs->rcx = kvm_register_read(vcpu, VCPU_REGS_RCX);
3823 regs->rdx = kvm_register_read(vcpu, VCPU_REGS_RDX);
3824 regs->rsi = kvm_register_read(vcpu, VCPU_REGS_RSI);
3825 regs->rdi = kvm_register_read(vcpu, VCPU_REGS_RDI);
3826 regs->rsp = kvm_register_read(vcpu, VCPU_REGS_RSP);
3827 regs->rbp = kvm_register_read(vcpu, VCPU_REGS_RBP);
3828 #ifdef CONFIG_X86_64
3829 regs->r8 = kvm_register_read(vcpu, VCPU_REGS_R8);
3830 regs->r9 = kvm_register_read(vcpu, VCPU_REGS_R9);
3831 regs->r10 = kvm_register_read(vcpu, VCPU_REGS_R10);
3832 regs->r11 = kvm_register_read(vcpu, VCPU_REGS_R11);
3833 regs->r12 = kvm_register_read(vcpu, VCPU_REGS_R12);
3834 regs->r13 = kvm_register_read(vcpu, VCPU_REGS_R13);
3835 regs->r14 = kvm_register_read(vcpu, VCPU_REGS_R14);
3836 regs->r15 = kvm_register_read(vcpu, VCPU_REGS_R15);
3839 regs->rip = kvm_rip_read(vcpu);
3840 regs->rflags = kvm_x86_ops->get_rflags(vcpu);
3843 * Don't leak debug flags in case they were set for guest debugging
3845 if (vcpu->guest_debug & KVM_GUESTDBG_SINGLESTEP)
3846 regs->rflags &= ~(X86_EFLAGS_TF | X86_EFLAGS_RF);
3853 int kvm_arch_vcpu_ioctl_set_regs(struct kvm_vcpu *vcpu, struct kvm_regs *regs)
3857 kvm_register_write(vcpu, VCPU_REGS_RAX, regs->rax);
3858 kvm_register_write(vcpu, VCPU_REGS_RBX, regs->rbx);
3859 kvm_register_write(vcpu, VCPU_REGS_RCX, regs->rcx);
3860 kvm_register_write(vcpu, VCPU_REGS_RDX, regs->rdx);
3861 kvm_register_write(vcpu, VCPU_REGS_RSI, regs->rsi);
3862 kvm_register_write(vcpu, VCPU_REGS_RDI, regs->rdi);
3863 kvm_register_write(vcpu, VCPU_REGS_RSP, regs->rsp);
3864 kvm_register_write(vcpu, VCPU_REGS_RBP, regs->rbp);
3865 #ifdef CONFIG_X86_64
3866 kvm_register_write(vcpu, VCPU_REGS_R8, regs->r8);
3867 kvm_register_write(vcpu, VCPU_REGS_R9, regs->r9);
3868 kvm_register_write(vcpu, VCPU_REGS_R10, regs->r10);
3869 kvm_register_write(vcpu, VCPU_REGS_R11, regs->r11);
3870 kvm_register_write(vcpu, VCPU_REGS_R12, regs->r12);
3871 kvm_register_write(vcpu, VCPU_REGS_R13, regs->r13);
3872 kvm_register_write(vcpu, VCPU_REGS_R14, regs->r14);
3873 kvm_register_write(vcpu, VCPU_REGS_R15, regs->r15);
3877 kvm_rip_write(vcpu, regs->rip);
3878 kvm_x86_ops->set_rflags(vcpu, regs->rflags);
3881 vcpu->arch.exception.pending = false;
3888 void kvm_get_segment(struct kvm_vcpu *vcpu,
3889 struct kvm_segment *var, int seg)
3891 kvm_x86_ops->get_segment(vcpu, var, seg);
3894 void kvm_get_cs_db_l_bits(struct kvm_vcpu *vcpu, int *db, int *l)
3896 struct kvm_segment cs;
3898 kvm_get_segment(vcpu, &cs, VCPU_SREG_CS);
3902 EXPORT_SYMBOL_GPL(kvm_get_cs_db_l_bits);
3904 int kvm_arch_vcpu_ioctl_get_sregs(struct kvm_vcpu *vcpu,
3905 struct kvm_sregs *sregs)
3907 struct descriptor_table dt;
3911 kvm_get_segment(vcpu, &sregs->cs, VCPU_SREG_CS);
3912 kvm_get_segment(vcpu, &sregs->ds, VCPU_SREG_DS);
3913 kvm_get_segment(vcpu, &sregs->es, VCPU_SREG_ES);
3914 kvm_get_segment(vcpu, &sregs->fs, VCPU_SREG_FS);
3915 kvm_get_segment(vcpu, &sregs->gs, VCPU_SREG_GS);
3916 kvm_get_segment(vcpu, &sregs->ss, VCPU_SREG_SS);
3918 kvm_get_segment(vcpu, &sregs->tr, VCPU_SREG_TR);
3919 kvm_get_segment(vcpu, &sregs->ldt, VCPU_SREG_LDTR);
3921 kvm_x86_ops->get_idt(vcpu, &dt);
3922 sregs->idt.limit = dt.limit;
3923 sregs->idt.base = dt.base;
3924 kvm_x86_ops->get_gdt(vcpu, &dt);
3925 sregs->gdt.limit = dt.limit;
3926 sregs->gdt.base = dt.base;
3928 kvm_x86_ops->decache_cr4_guest_bits(vcpu);
3929 sregs->cr0 = vcpu->arch.cr0;
3930 sregs->cr2 = vcpu->arch.cr2;
3931 sregs->cr3 = vcpu->arch.cr3;
3932 sregs->cr4 = vcpu->arch.cr4;
3933 sregs->cr8 = kvm_get_cr8(vcpu);
3934 sregs->efer = vcpu->arch.shadow_efer;
3935 sregs->apic_base = kvm_get_apic_base(vcpu);
3937 memset(sregs->interrupt_bitmap, 0, sizeof sregs->interrupt_bitmap);
3939 if (vcpu->arch.interrupt.pending && !vcpu->arch.interrupt.soft)
3940 set_bit(vcpu->arch.interrupt.nr,
3941 (unsigned long *)sregs->interrupt_bitmap);
3948 int kvm_arch_vcpu_ioctl_get_mpstate(struct kvm_vcpu *vcpu,
3949 struct kvm_mp_state *mp_state)
3952 mp_state->mp_state = vcpu->arch.mp_state;
3957 int kvm_arch_vcpu_ioctl_set_mpstate(struct kvm_vcpu *vcpu,
3958 struct kvm_mp_state *mp_state)
3961 vcpu->arch.mp_state = mp_state->mp_state;
3966 static void kvm_set_segment(struct kvm_vcpu *vcpu,
3967 struct kvm_segment *var, int seg)
3969 kvm_x86_ops->set_segment(vcpu, var, seg);
3972 static void seg_desct_to_kvm_desct(struct desc_struct *seg_desc, u16 selector,
3973 struct kvm_segment *kvm_desct)
3975 kvm_desct->base = get_desc_base(seg_desc);
3976 kvm_desct->limit = get_desc_limit(seg_desc);
3978 kvm_desct->limit <<= 12;
3979 kvm_desct->limit |= 0xfff;
3981 kvm_desct->selector = selector;
3982 kvm_desct->type = seg_desc->type;
3983 kvm_desct->present = seg_desc->p;
3984 kvm_desct->dpl = seg_desc->dpl;
3985 kvm_desct->db = seg_desc->d;
3986 kvm_desct->s = seg_desc->s;
3987 kvm_desct->l = seg_desc->l;
3988 kvm_desct->g = seg_desc->g;
3989 kvm_desct->avl = seg_desc->avl;
3991 kvm_desct->unusable = 1;
3993 kvm_desct->unusable = 0;
3994 kvm_desct->padding = 0;
3997 static void get_segment_descriptor_dtable(struct kvm_vcpu *vcpu,
3999 struct descriptor_table *dtable)
4001 if (selector & 1 << 2) {
4002 struct kvm_segment kvm_seg;
4004 kvm_get_segment(vcpu, &kvm_seg, VCPU_SREG_LDTR);
4006 if (kvm_seg.unusable)
4009 dtable->limit = kvm_seg.limit;
4010 dtable->base = kvm_seg.base;
4013 kvm_x86_ops->get_gdt(vcpu, dtable);
4016 /* allowed just for 8 bytes segments */
4017 static int load_guest_segment_descriptor(struct kvm_vcpu *vcpu, u16 selector,
4018 struct desc_struct *seg_desc)
4020 struct descriptor_table dtable;
4021 u16 index = selector >> 3;
4023 get_segment_descriptor_dtable(vcpu, selector, &dtable);
4025 if (dtable.limit < index * 8 + 7) {
4026 kvm_queue_exception_e(vcpu, GP_VECTOR, selector & 0xfffc);
4029 return kvm_read_guest_virt(dtable.base + index*8, seg_desc, sizeof(*seg_desc), vcpu);
4032 /* allowed just for 8 bytes segments */
4033 static int save_guest_segment_descriptor(struct kvm_vcpu *vcpu, u16 selector,
4034 struct desc_struct *seg_desc)
4036 struct descriptor_table dtable;
4037 u16 index = selector >> 3;
4039 get_segment_descriptor_dtable(vcpu, selector, &dtable);
4041 if (dtable.limit < index * 8 + 7)
4043 return kvm_write_guest_virt(dtable.base + index*8, seg_desc, sizeof(*seg_desc), vcpu);
4046 static gpa_t get_tss_base_addr(struct kvm_vcpu *vcpu,
4047 struct desc_struct *seg_desc)
4049 u32 base_addr = get_desc_base(seg_desc);
4051 return vcpu->arch.mmu.gva_to_gpa(vcpu, base_addr);
4054 static u16 get_segment_selector(struct kvm_vcpu *vcpu, int seg)
4056 struct kvm_segment kvm_seg;
4058 kvm_get_segment(vcpu, &kvm_seg, seg);
4059 return kvm_seg.selector;
4062 static int load_segment_descriptor_to_kvm_desct(struct kvm_vcpu *vcpu,
4064 struct kvm_segment *kvm_seg)
4066 struct desc_struct seg_desc;
4068 if (load_guest_segment_descriptor(vcpu, selector, &seg_desc))
4070 seg_desct_to_kvm_desct(&seg_desc, selector, kvm_seg);
4074 static int kvm_load_realmode_segment(struct kvm_vcpu *vcpu, u16 selector, int seg)
4076 struct kvm_segment segvar = {
4077 .base = selector << 4,
4079 .selector = selector,
4090 kvm_x86_ops->set_segment(vcpu, &segvar, seg);
4094 static int is_vm86_segment(struct kvm_vcpu *vcpu, int seg)
4096 return (seg != VCPU_SREG_LDTR) &&
4097 (seg != VCPU_SREG_TR) &&
4098 (kvm_x86_ops->get_rflags(vcpu) & X86_EFLAGS_VM);
4101 int kvm_load_segment_descriptor(struct kvm_vcpu *vcpu, u16 selector,
4102 int type_bits, int seg)
4104 struct kvm_segment kvm_seg;
4106 if (is_vm86_segment(vcpu, seg) || !(vcpu->arch.cr0 & X86_CR0_PE))
4107 return kvm_load_realmode_segment(vcpu, selector, seg);
4108 if (load_segment_descriptor_to_kvm_desct(vcpu, selector, &kvm_seg))
4110 kvm_seg.type |= type_bits;
4112 if (seg != VCPU_SREG_SS && seg != VCPU_SREG_CS &&
4113 seg != VCPU_SREG_LDTR)
4115 kvm_seg.unusable = 1;
4117 kvm_set_segment(vcpu, &kvm_seg, seg);
4121 static void save_state_to_tss32(struct kvm_vcpu *vcpu,
4122 struct tss_segment_32 *tss)
4124 tss->cr3 = vcpu->arch.cr3;
4125 tss->eip = kvm_rip_read(vcpu);
4126 tss->eflags = kvm_x86_ops->get_rflags(vcpu);
4127 tss->eax = kvm_register_read(vcpu, VCPU_REGS_RAX);
4128 tss->ecx = kvm_register_read(vcpu, VCPU_REGS_RCX);
4129 tss->edx = kvm_register_read(vcpu, VCPU_REGS_RDX);
4130 tss->ebx = kvm_register_read(vcpu, VCPU_REGS_RBX);
4131 tss->esp = kvm_register_read(vcpu, VCPU_REGS_RSP);
4132 tss->ebp = kvm_register_read(vcpu, VCPU_REGS_RBP);
4133 tss->esi = kvm_register_read(vcpu, VCPU_REGS_RSI);
4134 tss->edi = kvm_register_read(vcpu, VCPU_REGS_RDI);
4135 tss->es = get_segment_selector(vcpu, VCPU_SREG_ES);
4136 tss->cs = get_segment_selector(vcpu, VCPU_SREG_CS);
4137 tss->ss = get_segment_selector(vcpu, VCPU_SREG_SS);
4138 tss->ds = get_segment_selector(vcpu, VCPU_SREG_DS);
4139 tss->fs = get_segment_selector(vcpu, VCPU_SREG_FS);
4140 tss->gs = get_segment_selector(vcpu, VCPU_SREG_GS);
4141 tss->ldt_selector = get_segment_selector(vcpu, VCPU_SREG_LDTR);
4144 static int load_state_from_tss32(struct kvm_vcpu *vcpu,
4145 struct tss_segment_32 *tss)
4147 kvm_set_cr3(vcpu, tss->cr3);
4149 kvm_rip_write(vcpu, tss->eip);
4150 kvm_x86_ops->set_rflags(vcpu, tss->eflags | 2);
4152 kvm_register_write(vcpu, VCPU_REGS_RAX, tss->eax);
4153 kvm_register_write(vcpu, VCPU_REGS_RCX, tss->ecx);
4154 kvm_register_write(vcpu, VCPU_REGS_RDX, tss->edx);
4155 kvm_register_write(vcpu, VCPU_REGS_RBX, tss->ebx);
4156 kvm_register_write(vcpu, VCPU_REGS_RSP, tss->esp);
4157 kvm_register_write(vcpu, VCPU_REGS_RBP, tss->ebp);
4158 kvm_register_write(vcpu, VCPU_REGS_RSI, tss->esi);
4159 kvm_register_write(vcpu, VCPU_REGS_RDI, tss->edi);
4161 if (kvm_load_segment_descriptor(vcpu, tss->ldt_selector, 0, VCPU_SREG_LDTR))
4164 if (kvm_load_segment_descriptor(vcpu, tss->es, 1, VCPU_SREG_ES))
4167 if (kvm_load_segment_descriptor(vcpu, tss->cs, 9, VCPU_SREG_CS))
4170 if (kvm_load_segment_descriptor(vcpu, tss->ss, 1, VCPU_SREG_SS))
4173 if (kvm_load_segment_descriptor(vcpu, tss->ds, 1, VCPU_SREG_DS))
4176 if (kvm_load_segment_descriptor(vcpu, tss->fs, 1, VCPU_SREG_FS))
4179 if (kvm_load_segment_descriptor(vcpu, tss->gs, 1, VCPU_SREG_GS))
4184 static void save_state_to_tss16(struct kvm_vcpu *vcpu,
4185 struct tss_segment_16 *tss)
4187 tss->ip = kvm_rip_read(vcpu);
4188 tss->flag = kvm_x86_ops->get_rflags(vcpu);
4189 tss->ax = kvm_register_read(vcpu, VCPU_REGS_RAX);
4190 tss->cx = kvm_register_read(vcpu, VCPU_REGS_RCX);
4191 tss->dx = kvm_register_read(vcpu, VCPU_REGS_RDX);
4192 tss->bx = kvm_register_read(vcpu, VCPU_REGS_RBX);
4193 tss->sp = kvm_register_read(vcpu, VCPU_REGS_RSP);
4194 tss->bp = kvm_register_read(vcpu, VCPU_REGS_RBP);
4195 tss->si = kvm_register_read(vcpu, VCPU_REGS_RSI);
4196 tss->di = kvm_register_read(vcpu, VCPU_REGS_RDI);
4198 tss->es = get_segment_selector(vcpu, VCPU_SREG_ES);
4199 tss->cs = get_segment_selector(vcpu, VCPU_SREG_CS);
4200 tss->ss = get_segment_selector(vcpu, VCPU_SREG_SS);
4201 tss->ds = get_segment_selector(vcpu, VCPU_SREG_DS);
4202 tss->ldt = get_segment_selector(vcpu, VCPU_SREG_LDTR);
4203 tss->prev_task_link = get_segment_selector(vcpu, VCPU_SREG_TR);
4206 static int load_state_from_tss16(struct kvm_vcpu *vcpu,
4207 struct tss_segment_16 *tss)
4209 kvm_rip_write(vcpu, tss->ip);
4210 kvm_x86_ops->set_rflags(vcpu, tss->flag | 2);
4211 kvm_register_write(vcpu, VCPU_REGS_RAX, tss->ax);
4212 kvm_register_write(vcpu, VCPU_REGS_RCX, tss->cx);
4213 kvm_register_write(vcpu, VCPU_REGS_RDX, tss->dx);
4214 kvm_register_write(vcpu, VCPU_REGS_RBX, tss->bx);
4215 kvm_register_write(vcpu, VCPU_REGS_RSP, tss->sp);
4216 kvm_register_write(vcpu, VCPU_REGS_RBP, tss->bp);
4217 kvm_register_write(vcpu, VCPU_REGS_RSI, tss->si);
4218 kvm_register_write(vcpu, VCPU_REGS_RDI, tss->di);
4220 if (kvm_load_segment_descriptor(vcpu, tss->ldt, 0, VCPU_SREG_LDTR))
4223 if (kvm_load_segment_descriptor(vcpu, tss->es, 1, VCPU_SREG_ES))
4226 if (kvm_load_segment_descriptor(vcpu, tss->cs, 9, VCPU_SREG_CS))
4229 if (kvm_load_segment_descriptor(vcpu, tss->ss, 1, VCPU_SREG_SS))
4232 if (kvm_load_segment_descriptor(vcpu, tss->ds, 1, VCPU_SREG_DS))
4237 static int kvm_task_switch_16(struct kvm_vcpu *vcpu, u16 tss_selector,
4238 u16 old_tss_sel, u32 old_tss_base,
4239 struct desc_struct *nseg_desc)
4241 struct tss_segment_16 tss_segment_16;
4244 if (kvm_read_guest(vcpu->kvm, old_tss_base, &tss_segment_16,
4245 sizeof tss_segment_16))
4248 save_state_to_tss16(vcpu, &tss_segment_16);
4250 if (kvm_write_guest(vcpu->kvm, old_tss_base, &tss_segment_16,
4251 sizeof tss_segment_16))
4254 if (kvm_read_guest(vcpu->kvm, get_tss_base_addr(vcpu, nseg_desc),
4255 &tss_segment_16, sizeof tss_segment_16))
4258 if (old_tss_sel != 0xffff) {
4259 tss_segment_16.prev_task_link = old_tss_sel;
4261 if (kvm_write_guest(vcpu->kvm,
4262 get_tss_base_addr(vcpu, nseg_desc),
4263 &tss_segment_16.prev_task_link,
4264 sizeof tss_segment_16.prev_task_link))
4268 if (load_state_from_tss16(vcpu, &tss_segment_16))
4276 static int kvm_task_switch_32(struct kvm_vcpu *vcpu, u16 tss_selector,
4277 u16 old_tss_sel, u32 old_tss_base,
4278 struct desc_struct *nseg_desc)
4280 struct tss_segment_32 tss_segment_32;
4283 if (kvm_read_guest(vcpu->kvm, old_tss_base, &tss_segment_32,
4284 sizeof tss_segment_32))
4287 save_state_to_tss32(vcpu, &tss_segment_32);
4289 if (kvm_write_guest(vcpu->kvm, old_tss_base, &tss_segment_32,
4290 sizeof tss_segment_32))
4293 if (kvm_read_guest(vcpu->kvm, get_tss_base_addr(vcpu, nseg_desc),
4294 &tss_segment_32, sizeof tss_segment_32))
4297 if (old_tss_sel != 0xffff) {
4298 tss_segment_32.prev_task_link = old_tss_sel;
4300 if (kvm_write_guest(vcpu->kvm,
4301 get_tss_base_addr(vcpu, nseg_desc),
4302 &tss_segment_32.prev_task_link,
4303 sizeof tss_segment_32.prev_task_link))
4307 if (load_state_from_tss32(vcpu, &tss_segment_32))
4315 int kvm_task_switch(struct kvm_vcpu *vcpu, u16 tss_selector, int reason)
4317 struct kvm_segment tr_seg;
4318 struct desc_struct cseg_desc;
4319 struct desc_struct nseg_desc;
4321 u32 old_tss_base = get_segment_base(vcpu, VCPU_SREG_TR);
4322 u16 old_tss_sel = get_segment_selector(vcpu, VCPU_SREG_TR);
4324 old_tss_base = vcpu->arch.mmu.gva_to_gpa(vcpu, old_tss_base);
4326 /* FIXME: Handle errors. Failure to read either TSS or their
4327 * descriptors should generate a pagefault.
4329 if (load_guest_segment_descriptor(vcpu, tss_selector, &nseg_desc))
4332 if (load_guest_segment_descriptor(vcpu, old_tss_sel, &cseg_desc))
4335 if (reason != TASK_SWITCH_IRET) {
4338 cpl = kvm_x86_ops->get_cpl(vcpu);
4339 if ((tss_selector & 3) > nseg_desc.dpl || cpl > nseg_desc.dpl) {
4340 kvm_queue_exception_e(vcpu, GP_VECTOR, 0);
4345 if (!nseg_desc.p || get_desc_limit(&nseg_desc) < 0x67) {
4346 kvm_queue_exception_e(vcpu, TS_VECTOR, tss_selector & 0xfffc);
4350 if (reason == TASK_SWITCH_IRET || reason == TASK_SWITCH_JMP) {
4351 cseg_desc.type &= ~(1 << 1); //clear the B flag
4352 save_guest_segment_descriptor(vcpu, old_tss_sel, &cseg_desc);
4355 if (reason == TASK_SWITCH_IRET) {
4356 u32 eflags = kvm_x86_ops->get_rflags(vcpu);
4357 kvm_x86_ops->set_rflags(vcpu, eflags & ~X86_EFLAGS_NT);
4360 /* set back link to prev task only if NT bit is set in eflags
4361 note that old_tss_sel is not used afetr this point */
4362 if (reason != TASK_SWITCH_CALL && reason != TASK_SWITCH_GATE)
4363 old_tss_sel = 0xffff;
4365 /* set back link to prev task only if NT bit is set in eflags
4366 note that old_tss_sel is not used afetr this point */
4367 if (reason != TASK_SWITCH_CALL && reason != TASK_SWITCH_GATE)
4368 old_tss_sel = 0xffff;
4370 if (nseg_desc.type & 8)
4371 ret = kvm_task_switch_32(vcpu, tss_selector, old_tss_sel,
4372 old_tss_base, &nseg_desc);
4374 ret = kvm_task_switch_16(vcpu, tss_selector, old_tss_sel,
4375 old_tss_base, &nseg_desc);
4377 if (reason == TASK_SWITCH_CALL || reason == TASK_SWITCH_GATE) {
4378 u32 eflags = kvm_x86_ops->get_rflags(vcpu);
4379 kvm_x86_ops->set_rflags(vcpu, eflags | X86_EFLAGS_NT);
4382 if (reason != TASK_SWITCH_IRET) {
4383 nseg_desc.type |= (1 << 1);
4384 save_guest_segment_descriptor(vcpu, tss_selector,
4388 kvm_x86_ops->set_cr0(vcpu, vcpu->arch.cr0 | X86_CR0_TS);
4389 seg_desct_to_kvm_desct(&nseg_desc, tss_selector, &tr_seg);
4391 kvm_set_segment(vcpu, &tr_seg, VCPU_SREG_TR);
4395 EXPORT_SYMBOL_GPL(kvm_task_switch);
4397 int kvm_arch_vcpu_ioctl_set_sregs(struct kvm_vcpu *vcpu,
4398 struct kvm_sregs *sregs)
4400 int mmu_reset_needed = 0;
4401 int pending_vec, max_bits;
4402 struct descriptor_table dt;
4406 dt.limit = sregs->idt.limit;
4407 dt.base = sregs->idt.base;
4408 kvm_x86_ops->set_idt(vcpu, &dt);
4409 dt.limit = sregs->gdt.limit;
4410 dt.base = sregs->gdt.base;
4411 kvm_x86_ops->set_gdt(vcpu, &dt);
4413 vcpu->arch.cr2 = sregs->cr2;
4414 mmu_reset_needed |= vcpu->arch.cr3 != sregs->cr3;
4415 vcpu->arch.cr3 = sregs->cr3;
4417 kvm_set_cr8(vcpu, sregs->cr8);
4419 mmu_reset_needed |= vcpu->arch.shadow_efer != sregs->efer;
4420 kvm_x86_ops->set_efer(vcpu, sregs->efer);
4421 kvm_set_apic_base(vcpu, sregs->apic_base);
4423 kvm_x86_ops->decache_cr4_guest_bits(vcpu);
4425 mmu_reset_needed |= vcpu->arch.cr0 != sregs->cr0;
4426 kvm_x86_ops->set_cr0(vcpu, sregs->cr0);
4427 vcpu->arch.cr0 = sregs->cr0;
4429 mmu_reset_needed |= vcpu->arch.cr4 != sregs->cr4;
4430 kvm_x86_ops->set_cr4(vcpu, sregs->cr4);
4431 if (!is_long_mode(vcpu) && is_pae(vcpu))
4432 load_pdptrs(vcpu, vcpu->arch.cr3);
4434 if (mmu_reset_needed)
4435 kvm_mmu_reset_context(vcpu);
4437 max_bits = (sizeof sregs->interrupt_bitmap) << 3;
4438 pending_vec = find_first_bit(
4439 (const unsigned long *)sregs->interrupt_bitmap, max_bits);
4440 if (pending_vec < max_bits) {
4441 kvm_queue_interrupt(vcpu, pending_vec, false);
4442 pr_debug("Set back pending irq %d\n", pending_vec);
4443 if (irqchip_in_kernel(vcpu->kvm))
4444 kvm_pic_clear_isr_ack(vcpu->kvm);
4447 kvm_set_segment(vcpu, &sregs->cs, VCPU_SREG_CS);
4448 kvm_set_segment(vcpu, &sregs->ds, VCPU_SREG_DS);
4449 kvm_set_segment(vcpu, &sregs->es, VCPU_SREG_ES);
4450 kvm_set_segment(vcpu, &sregs->fs, VCPU_SREG_FS);
4451 kvm_set_segment(vcpu, &sregs->gs, VCPU_SREG_GS);
4452 kvm_set_segment(vcpu, &sregs->ss, VCPU_SREG_SS);
4454 kvm_set_segment(vcpu, &sregs->tr, VCPU_SREG_TR);
4455 kvm_set_segment(vcpu, &sregs->ldt, VCPU_SREG_LDTR);
4457 update_cr8_intercept(vcpu);
4459 /* Older userspace won't unhalt the vcpu on reset. */
4460 if (kvm_vcpu_is_bsp(vcpu) && kvm_rip_read(vcpu) == 0xfff0 &&
4461 sregs->cs.selector == 0xf000 && sregs->cs.base == 0xffff0000 &&
4462 !(vcpu->arch.cr0 & X86_CR0_PE))
4463 vcpu->arch.mp_state = KVM_MP_STATE_RUNNABLE;
4470 int kvm_arch_vcpu_ioctl_set_guest_debug(struct kvm_vcpu *vcpu,
4471 struct kvm_guest_debug *dbg)
4477 if ((dbg->control & (KVM_GUESTDBG_ENABLE | KVM_GUESTDBG_USE_HW_BP)) ==
4478 (KVM_GUESTDBG_ENABLE | KVM_GUESTDBG_USE_HW_BP)) {
4479 for (i = 0; i < KVM_NR_DB_REGS; ++i)
4480 vcpu->arch.eff_db[i] = dbg->arch.debugreg[i];
4481 vcpu->arch.switch_db_regs =
4482 (dbg->arch.debugreg[7] & DR7_BP_EN_MASK);
4484 for (i = 0; i < KVM_NR_DB_REGS; i++)
4485 vcpu->arch.eff_db[i] = vcpu->arch.db[i];
4486 vcpu->arch.switch_db_regs = (vcpu->arch.dr7 & DR7_BP_EN_MASK);
4489 r = kvm_x86_ops->set_guest_debug(vcpu, dbg);
4491 if (dbg->control & KVM_GUESTDBG_INJECT_DB)
4492 kvm_queue_exception(vcpu, DB_VECTOR);
4493 else if (dbg->control & KVM_GUESTDBG_INJECT_BP)
4494 kvm_queue_exception(vcpu, BP_VECTOR);
4502 * fxsave fpu state. Taken from x86_64/processor.h. To be killed when
4503 * we have asm/x86/processor.h
4514 u32 st_space[32]; /* 8*16 bytes for each FP-reg = 128 bytes */
4515 #ifdef CONFIG_X86_64
4516 u32 xmm_space[64]; /* 16*16 bytes for each XMM-reg = 256 bytes */
4518 u32 xmm_space[32]; /* 8*16 bytes for each XMM-reg = 128 bytes */
4523 * Translate a guest virtual address to a guest physical address.
4525 int kvm_arch_vcpu_ioctl_translate(struct kvm_vcpu *vcpu,
4526 struct kvm_translation *tr)
4528 unsigned long vaddr = tr->linear_address;
4532 down_read(&vcpu->kvm->slots_lock);
4533 gpa = vcpu->arch.mmu.gva_to_gpa(vcpu, vaddr);
4534 up_read(&vcpu->kvm->slots_lock);
4535 tr->physical_address = gpa;
4536 tr->valid = gpa != UNMAPPED_GVA;
4544 int kvm_arch_vcpu_ioctl_get_fpu(struct kvm_vcpu *vcpu, struct kvm_fpu *fpu)
4546 struct fxsave *fxsave = (struct fxsave *)&vcpu->arch.guest_fx_image;
4550 memcpy(fpu->fpr, fxsave->st_space, 128);
4551 fpu->fcw = fxsave->cwd;
4552 fpu->fsw = fxsave->swd;
4553 fpu->ftwx = fxsave->twd;
4554 fpu->last_opcode = fxsave->fop;
4555 fpu->last_ip = fxsave->rip;
4556 fpu->last_dp = fxsave->rdp;
4557 memcpy(fpu->xmm, fxsave->xmm_space, sizeof fxsave->xmm_space);
4564 int kvm_arch_vcpu_ioctl_set_fpu(struct kvm_vcpu *vcpu, struct kvm_fpu *fpu)
4566 struct fxsave *fxsave = (struct fxsave *)&vcpu->arch.guest_fx_image;
4570 memcpy(fxsave->st_space, fpu->fpr, 128);
4571 fxsave->cwd = fpu->fcw;
4572 fxsave->swd = fpu->fsw;
4573 fxsave->twd = fpu->ftwx;
4574 fxsave->fop = fpu->last_opcode;
4575 fxsave->rip = fpu->last_ip;
4576 fxsave->rdp = fpu->last_dp;
4577 memcpy(fxsave->xmm_space, fpu->xmm, sizeof fxsave->xmm_space);
4584 void fx_init(struct kvm_vcpu *vcpu)
4586 unsigned after_mxcsr_mask;
4589 * Touch the fpu the first time in non atomic context as if
4590 * this is the first fpu instruction the exception handler
4591 * will fire before the instruction returns and it'll have to
4592 * allocate ram with GFP_KERNEL.
4595 kvm_fx_save(&vcpu->arch.host_fx_image);
4597 /* Initialize guest FPU by resetting ours and saving into guest's */
4599 kvm_fx_save(&vcpu->arch.host_fx_image);
4601 kvm_fx_save(&vcpu->arch.guest_fx_image);
4602 kvm_fx_restore(&vcpu->arch.host_fx_image);
4605 vcpu->arch.cr0 |= X86_CR0_ET;
4606 after_mxcsr_mask = offsetof(struct i387_fxsave_struct, st_space);
4607 vcpu->arch.guest_fx_image.mxcsr = 0x1f80;
4608 memset((void *)&vcpu->arch.guest_fx_image + after_mxcsr_mask,
4609 0, sizeof(struct i387_fxsave_struct) - after_mxcsr_mask);
4611 EXPORT_SYMBOL_GPL(fx_init);
4613 void kvm_load_guest_fpu(struct kvm_vcpu *vcpu)
4615 if (!vcpu->fpu_active || vcpu->guest_fpu_loaded)
4618 vcpu->guest_fpu_loaded = 1;
4619 kvm_fx_save(&vcpu->arch.host_fx_image);
4620 kvm_fx_restore(&vcpu->arch.guest_fx_image);
4622 EXPORT_SYMBOL_GPL(kvm_load_guest_fpu);
4624 void kvm_put_guest_fpu(struct kvm_vcpu *vcpu)
4626 if (!vcpu->guest_fpu_loaded)
4629 vcpu->guest_fpu_loaded = 0;
4630 kvm_fx_save(&vcpu->arch.guest_fx_image);
4631 kvm_fx_restore(&vcpu->arch.host_fx_image);
4632 ++vcpu->stat.fpu_reload;
4634 EXPORT_SYMBOL_GPL(kvm_put_guest_fpu);
4636 void kvm_arch_vcpu_free(struct kvm_vcpu *vcpu)
4638 if (vcpu->arch.time_page) {
4639 kvm_release_page_dirty(vcpu->arch.time_page);
4640 vcpu->arch.time_page = NULL;
4643 kvm_x86_ops->vcpu_free(vcpu);
4646 struct kvm_vcpu *kvm_arch_vcpu_create(struct kvm *kvm,
4649 return kvm_x86_ops->vcpu_create(kvm, id);
4652 int kvm_arch_vcpu_setup(struct kvm_vcpu *vcpu)
4656 /* We do fxsave: this must be aligned. */
4657 BUG_ON((unsigned long)&vcpu->arch.host_fx_image & 0xF);
4659 vcpu->arch.mtrr_state.have_fixed = 1;
4661 r = kvm_arch_vcpu_reset(vcpu);
4663 r = kvm_mmu_setup(vcpu);
4670 kvm_x86_ops->vcpu_free(vcpu);
4674 void kvm_arch_vcpu_destroy(struct kvm_vcpu *vcpu)
4677 kvm_mmu_unload(vcpu);
4680 kvm_x86_ops->vcpu_free(vcpu);
4683 int kvm_arch_vcpu_reset(struct kvm_vcpu *vcpu)
4685 vcpu->arch.nmi_pending = false;
4686 vcpu->arch.nmi_injected = false;
4688 vcpu->arch.switch_db_regs = 0;
4689 memset(vcpu->arch.db, 0, sizeof(vcpu->arch.db));
4690 vcpu->arch.dr6 = DR6_FIXED_1;
4691 vcpu->arch.dr7 = DR7_FIXED_1;
4693 return kvm_x86_ops->vcpu_reset(vcpu);
4696 void kvm_arch_hardware_enable(void *garbage)
4698 kvm_x86_ops->hardware_enable(garbage);
4701 void kvm_arch_hardware_disable(void *garbage)
4703 kvm_x86_ops->hardware_disable(garbage);
4706 int kvm_arch_hardware_setup(void)
4708 return kvm_x86_ops->hardware_setup();
4711 void kvm_arch_hardware_unsetup(void)
4713 kvm_x86_ops->hardware_unsetup();
4716 void kvm_arch_check_processor_compat(void *rtn)
4718 kvm_x86_ops->check_processor_compatibility(rtn);
4721 int kvm_arch_vcpu_init(struct kvm_vcpu *vcpu)
4727 BUG_ON(vcpu->kvm == NULL);
4730 vcpu->arch.mmu.root_hpa = INVALID_PAGE;
4731 if (!irqchip_in_kernel(kvm) || kvm_vcpu_is_bsp(vcpu))
4732 vcpu->arch.mp_state = KVM_MP_STATE_RUNNABLE;
4734 vcpu->arch.mp_state = KVM_MP_STATE_UNINITIALIZED;
4736 page = alloc_page(GFP_KERNEL | __GFP_ZERO);
4741 vcpu->arch.pio_data = page_address(page);
4743 r = kvm_mmu_create(vcpu);
4745 goto fail_free_pio_data;
4747 if (irqchip_in_kernel(kvm)) {
4748 r = kvm_create_lapic(vcpu);
4750 goto fail_mmu_destroy;
4753 vcpu->arch.mce_banks = kzalloc(KVM_MAX_MCE_BANKS * sizeof(u64) * 4,
4755 if (!vcpu->arch.mce_banks) {
4757 goto fail_mmu_destroy;
4759 vcpu->arch.mcg_cap = KVM_MAX_MCE_BANKS;
4764 kvm_mmu_destroy(vcpu);
4766 free_page((unsigned long)vcpu->arch.pio_data);
4771 void kvm_arch_vcpu_uninit(struct kvm_vcpu *vcpu)
4773 kvm_free_lapic(vcpu);
4774 down_read(&vcpu->kvm->slots_lock);
4775 kvm_mmu_destroy(vcpu);
4776 up_read(&vcpu->kvm->slots_lock);
4777 free_page((unsigned long)vcpu->arch.pio_data);
4780 struct kvm *kvm_arch_create_vm(void)
4782 struct kvm *kvm = kzalloc(sizeof(struct kvm), GFP_KERNEL);
4785 return ERR_PTR(-ENOMEM);
4787 INIT_LIST_HEAD(&kvm->arch.active_mmu_pages);
4788 INIT_LIST_HEAD(&kvm->arch.assigned_dev_head);
4790 /* Reserve bit 0 of irq_sources_bitmap for userspace irq source */
4791 set_bit(KVM_USERSPACE_IRQ_SOURCE_ID, &kvm->arch.irq_sources_bitmap);
4793 rdtscll(kvm->arch.vm_init_tsc);
4798 static void kvm_unload_vcpu_mmu(struct kvm_vcpu *vcpu)
4801 kvm_mmu_unload(vcpu);
4805 static void kvm_free_vcpus(struct kvm *kvm)
4808 struct kvm_vcpu *vcpu;
4811 * Unpin any mmu pages first.
4813 kvm_for_each_vcpu(i, vcpu, kvm)
4814 kvm_unload_vcpu_mmu(vcpu);
4815 kvm_for_each_vcpu(i, vcpu, kvm)
4816 kvm_arch_vcpu_free(vcpu);
4818 mutex_lock(&kvm->lock);
4819 for (i = 0; i < atomic_read(&kvm->online_vcpus); i++)
4820 kvm->vcpus[i] = NULL;
4822 atomic_set(&kvm->online_vcpus, 0);
4823 mutex_unlock(&kvm->lock);
4826 void kvm_arch_sync_events(struct kvm *kvm)
4828 kvm_free_all_assigned_devices(kvm);
4831 void kvm_arch_destroy_vm(struct kvm *kvm)
4833 kvm_iommu_unmap_guest(kvm);
4835 kfree(kvm->arch.vpic);
4836 kfree(kvm->arch.vioapic);
4837 kvm_free_vcpus(kvm);
4838 kvm_free_physmem(kvm);
4839 if (kvm->arch.apic_access_page)
4840 put_page(kvm->arch.apic_access_page);
4841 if (kvm->arch.ept_identity_pagetable)
4842 put_page(kvm->arch.ept_identity_pagetable);
4846 int kvm_arch_set_memory_region(struct kvm *kvm,
4847 struct kvm_userspace_memory_region *mem,
4848 struct kvm_memory_slot old,
4851 int npages = mem->memory_size >> PAGE_SHIFT;
4852 struct kvm_memory_slot *memslot = &kvm->memslots[mem->slot];
4854 /*To keep backward compatibility with older userspace,
4855 *x86 needs to hanlde !user_alloc case.
4858 if (npages && !old.rmap) {
4859 unsigned long userspace_addr;
4861 down_write(¤t->mm->mmap_sem);
4862 userspace_addr = do_mmap(NULL, 0,
4864 PROT_READ | PROT_WRITE,
4865 MAP_PRIVATE | MAP_ANONYMOUS,
4867 up_write(¤t->mm->mmap_sem);
4869 if (IS_ERR((void *)userspace_addr))
4870 return PTR_ERR((void *)userspace_addr);
4872 /* set userspace_addr atomically for kvm_hva_to_rmapp */
4873 spin_lock(&kvm->mmu_lock);
4874 memslot->userspace_addr = userspace_addr;
4875 spin_unlock(&kvm->mmu_lock);
4877 if (!old.user_alloc && old.rmap) {
4880 down_write(¤t->mm->mmap_sem);
4881 ret = do_munmap(current->mm, old.userspace_addr,
4882 old.npages * PAGE_SIZE);
4883 up_write(¤t->mm->mmap_sem);
4886 "kvm_vm_ioctl_set_memory_region: "
4887 "failed to munmap memory\n");
4892 spin_lock(&kvm->mmu_lock);
4893 if (!kvm->arch.n_requested_mmu_pages) {
4894 unsigned int nr_mmu_pages = kvm_mmu_calculate_mmu_pages(kvm);
4895 kvm_mmu_change_mmu_pages(kvm, nr_mmu_pages);
4898 kvm_mmu_slot_remove_write_access(kvm, mem->slot);
4899 spin_unlock(&kvm->mmu_lock);
4904 void kvm_arch_flush_shadow(struct kvm *kvm)
4906 kvm_mmu_zap_all(kvm);
4907 kvm_reload_remote_mmus(kvm);
4910 int kvm_arch_vcpu_runnable(struct kvm_vcpu *vcpu)
4912 return vcpu->arch.mp_state == KVM_MP_STATE_RUNNABLE
4913 || vcpu->arch.mp_state == KVM_MP_STATE_SIPI_RECEIVED
4914 || vcpu->arch.nmi_pending ||
4915 (kvm_arch_interrupt_allowed(vcpu) &&
4916 kvm_cpu_has_interrupt(vcpu));
4919 void kvm_vcpu_kick(struct kvm_vcpu *vcpu)
4922 int cpu = vcpu->cpu;
4924 if (waitqueue_active(&vcpu->wq)) {
4925 wake_up_interruptible(&vcpu->wq);
4926 ++vcpu->stat.halt_wakeup;
4930 if (cpu != me && (unsigned)cpu < nr_cpu_ids && cpu_online(cpu))
4931 if (!test_and_set_bit(KVM_REQ_KICK, &vcpu->requests))
4932 smp_send_reschedule(cpu);
4936 int kvm_arch_interrupt_allowed(struct kvm_vcpu *vcpu)
4938 return kvm_x86_ops->interrupt_allowed(vcpu);
4941 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_exit);
4942 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_inj_virq);
4943 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_page_fault);
4944 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_msr);
4945 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_cr);