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: /* obsolete */
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 if (unlikely(per_cpu(cpu_tsc_khz, cpu) == 0))
1330 per_cpu(cpu_tsc_khz, cpu) = cpufreq_quick_get(cpu);
1331 kvm_request_guest_time_update(vcpu);
1334 void kvm_arch_vcpu_put(struct kvm_vcpu *vcpu)
1336 kvm_x86_ops->vcpu_put(vcpu);
1337 kvm_put_guest_fpu(vcpu);
1340 static int is_efer_nx(void)
1342 unsigned long long efer = 0;
1344 rdmsrl_safe(MSR_EFER, &efer);
1345 return efer & EFER_NX;
1348 static void cpuid_fix_nx_cap(struct kvm_vcpu *vcpu)
1351 struct kvm_cpuid_entry2 *e, *entry;
1354 for (i = 0; i < vcpu->arch.cpuid_nent; ++i) {
1355 e = &vcpu->arch.cpuid_entries[i];
1356 if (e->function == 0x80000001) {
1361 if (entry && (entry->edx & (1 << 20)) && !is_efer_nx()) {
1362 entry->edx &= ~(1 << 20);
1363 printk(KERN_INFO "kvm: guest NX capability removed\n");
1367 /* when an old userspace process fills a new kernel module */
1368 static int kvm_vcpu_ioctl_set_cpuid(struct kvm_vcpu *vcpu,
1369 struct kvm_cpuid *cpuid,
1370 struct kvm_cpuid_entry __user *entries)
1373 struct kvm_cpuid_entry *cpuid_entries;
1376 if (cpuid->nent > KVM_MAX_CPUID_ENTRIES)
1379 cpuid_entries = vmalloc(sizeof(struct kvm_cpuid_entry) * cpuid->nent);
1383 if (copy_from_user(cpuid_entries, entries,
1384 cpuid->nent * sizeof(struct kvm_cpuid_entry)))
1386 for (i = 0; i < cpuid->nent; i++) {
1387 vcpu->arch.cpuid_entries[i].function = cpuid_entries[i].function;
1388 vcpu->arch.cpuid_entries[i].eax = cpuid_entries[i].eax;
1389 vcpu->arch.cpuid_entries[i].ebx = cpuid_entries[i].ebx;
1390 vcpu->arch.cpuid_entries[i].ecx = cpuid_entries[i].ecx;
1391 vcpu->arch.cpuid_entries[i].edx = cpuid_entries[i].edx;
1392 vcpu->arch.cpuid_entries[i].index = 0;
1393 vcpu->arch.cpuid_entries[i].flags = 0;
1394 vcpu->arch.cpuid_entries[i].padding[0] = 0;
1395 vcpu->arch.cpuid_entries[i].padding[1] = 0;
1396 vcpu->arch.cpuid_entries[i].padding[2] = 0;
1398 vcpu->arch.cpuid_nent = cpuid->nent;
1399 cpuid_fix_nx_cap(vcpu);
1401 kvm_apic_set_version(vcpu);
1404 vfree(cpuid_entries);
1409 static int kvm_vcpu_ioctl_set_cpuid2(struct kvm_vcpu *vcpu,
1410 struct kvm_cpuid2 *cpuid,
1411 struct kvm_cpuid_entry2 __user *entries)
1416 if (cpuid->nent > KVM_MAX_CPUID_ENTRIES)
1419 if (copy_from_user(&vcpu->arch.cpuid_entries, entries,
1420 cpuid->nent * sizeof(struct kvm_cpuid_entry2)))
1422 vcpu->arch.cpuid_nent = cpuid->nent;
1423 kvm_apic_set_version(vcpu);
1430 static int kvm_vcpu_ioctl_get_cpuid2(struct kvm_vcpu *vcpu,
1431 struct kvm_cpuid2 *cpuid,
1432 struct kvm_cpuid_entry2 __user *entries)
1437 if (cpuid->nent < vcpu->arch.cpuid_nent)
1440 if (copy_to_user(entries, &vcpu->arch.cpuid_entries,
1441 vcpu->arch.cpuid_nent * sizeof(struct kvm_cpuid_entry2)))
1446 cpuid->nent = vcpu->arch.cpuid_nent;
1450 static void do_cpuid_1_ent(struct kvm_cpuid_entry2 *entry, u32 function,
1453 entry->function = function;
1454 entry->index = index;
1455 cpuid_count(entry->function, entry->index,
1456 &entry->eax, &entry->ebx, &entry->ecx, &entry->edx);
1460 #define F(x) bit(X86_FEATURE_##x)
1462 static void do_cpuid_ent(struct kvm_cpuid_entry2 *entry, u32 function,
1463 u32 index, int *nent, int maxnent)
1465 unsigned f_nx = is_efer_nx() ? F(NX) : 0;
1466 unsigned f_gbpages = kvm_x86_ops->gb_page_enable() ? F(GBPAGES) : 0;
1467 #ifdef CONFIG_X86_64
1468 unsigned f_lm = F(LM);
1474 const u32 kvm_supported_word0_x86_features =
1475 F(FPU) | F(VME) | F(DE) | F(PSE) |
1476 F(TSC) | F(MSR) | F(PAE) | F(MCE) |
1477 F(CX8) | F(APIC) | 0 /* Reserved */ | F(SEP) |
1478 F(MTRR) | F(PGE) | F(MCA) | F(CMOV) |
1479 F(PAT) | F(PSE36) | 0 /* PSN */ | F(CLFLSH) |
1480 0 /* Reserved, DS, ACPI */ | F(MMX) |
1481 F(FXSR) | F(XMM) | F(XMM2) | F(SELFSNOOP) |
1482 0 /* HTT, TM, Reserved, PBE */;
1483 /* cpuid 0x80000001.edx */
1484 const u32 kvm_supported_word1_x86_features =
1485 F(FPU) | F(VME) | F(DE) | F(PSE) |
1486 F(TSC) | F(MSR) | F(PAE) | F(MCE) |
1487 F(CX8) | F(APIC) | 0 /* Reserved */ | F(SYSCALL) |
1488 F(MTRR) | F(PGE) | F(MCA) | F(CMOV) |
1489 F(PAT) | F(PSE36) | 0 /* Reserved */ |
1490 f_nx | 0 /* Reserved */ | F(MMXEXT) | F(MMX) |
1491 F(FXSR) | F(FXSR_OPT) | f_gbpages | 0 /* RDTSCP */ |
1492 0 /* Reserved */ | f_lm | F(3DNOWEXT) | F(3DNOW);
1494 const u32 kvm_supported_word4_x86_features =
1495 F(XMM3) | 0 /* Reserved, DTES64, MONITOR */ |
1496 0 /* DS-CPL, VMX, SMX, EST */ |
1497 0 /* TM2 */ | F(SSSE3) | 0 /* CNXT-ID */ | 0 /* Reserved */ |
1498 0 /* Reserved */ | F(CX16) | 0 /* xTPR Update, PDCM */ |
1499 0 /* Reserved, DCA */ | F(XMM4_1) |
1500 F(XMM4_2) | F(X2APIC) | F(MOVBE) | F(POPCNT) |
1501 0 /* Reserved, XSAVE, OSXSAVE */;
1502 /* cpuid 0x80000001.ecx */
1503 const u32 kvm_supported_word6_x86_features =
1504 F(LAHF_LM) | F(CMP_LEGACY) | F(SVM) | 0 /* ExtApicSpace */ |
1505 F(CR8_LEGACY) | F(ABM) | F(SSE4A) | F(MISALIGNSSE) |
1506 F(3DNOWPREFETCH) | 0 /* OSVW */ | 0 /* IBS */ | F(SSE5) |
1507 0 /* SKINIT */ | 0 /* WDT */;
1509 /* all calls to cpuid_count() should be made on the same cpu */
1511 do_cpuid_1_ent(entry, function, index);
1516 entry->eax = min(entry->eax, (u32)0xb);
1519 entry->edx &= kvm_supported_word0_x86_features;
1520 entry->ecx &= kvm_supported_word4_x86_features;
1521 /* we support x2apic emulation even if host does not support
1522 * it since we emulate x2apic in software */
1523 entry->ecx |= F(X2APIC);
1525 /* function 2 entries are STATEFUL. That is, repeated cpuid commands
1526 * may return different values. This forces us to get_cpu() before
1527 * issuing the first command, and also to emulate this annoying behavior
1528 * in kvm_emulate_cpuid() using KVM_CPUID_FLAG_STATE_READ_NEXT */
1530 int t, times = entry->eax & 0xff;
1532 entry->flags |= KVM_CPUID_FLAG_STATEFUL_FUNC;
1533 entry->flags |= KVM_CPUID_FLAG_STATE_READ_NEXT;
1534 for (t = 1; t < times && *nent < maxnent; ++t) {
1535 do_cpuid_1_ent(&entry[t], function, 0);
1536 entry[t].flags |= KVM_CPUID_FLAG_STATEFUL_FUNC;
1541 /* function 4 and 0xb have additional index. */
1545 entry->flags |= KVM_CPUID_FLAG_SIGNIFCANT_INDEX;
1546 /* read more entries until cache_type is zero */
1547 for (i = 1; *nent < maxnent; ++i) {
1548 cache_type = entry[i - 1].eax & 0x1f;
1551 do_cpuid_1_ent(&entry[i], function, i);
1553 KVM_CPUID_FLAG_SIGNIFCANT_INDEX;
1561 entry->flags |= KVM_CPUID_FLAG_SIGNIFCANT_INDEX;
1562 /* read more entries until level_type is zero */
1563 for (i = 1; *nent < maxnent; ++i) {
1564 level_type = entry[i - 1].ecx & 0xff00;
1567 do_cpuid_1_ent(&entry[i], function, i);
1569 KVM_CPUID_FLAG_SIGNIFCANT_INDEX;
1575 entry->eax = min(entry->eax, 0x8000001a);
1578 entry->edx &= kvm_supported_word1_x86_features;
1579 entry->ecx &= kvm_supported_word6_x86_features;
1587 static int kvm_dev_ioctl_get_supported_cpuid(struct kvm_cpuid2 *cpuid,
1588 struct kvm_cpuid_entry2 __user *entries)
1590 struct kvm_cpuid_entry2 *cpuid_entries;
1591 int limit, nent = 0, r = -E2BIG;
1594 if (cpuid->nent < 1)
1596 if (cpuid->nent > KVM_MAX_CPUID_ENTRIES)
1597 cpuid->nent = KVM_MAX_CPUID_ENTRIES;
1599 cpuid_entries = vmalloc(sizeof(struct kvm_cpuid_entry2) * cpuid->nent);
1603 do_cpuid_ent(&cpuid_entries[0], 0, 0, &nent, cpuid->nent);
1604 limit = cpuid_entries[0].eax;
1605 for (func = 1; func <= limit && nent < cpuid->nent; ++func)
1606 do_cpuid_ent(&cpuid_entries[nent], func, 0,
1607 &nent, cpuid->nent);
1609 if (nent >= cpuid->nent)
1612 do_cpuid_ent(&cpuid_entries[nent], 0x80000000, 0, &nent, cpuid->nent);
1613 limit = cpuid_entries[nent - 1].eax;
1614 for (func = 0x80000001; func <= limit && nent < cpuid->nent; ++func)
1615 do_cpuid_ent(&cpuid_entries[nent], func, 0,
1616 &nent, cpuid->nent);
1618 if (nent >= cpuid->nent)
1622 if (copy_to_user(entries, cpuid_entries,
1623 nent * sizeof(struct kvm_cpuid_entry2)))
1629 vfree(cpuid_entries);
1634 static int kvm_vcpu_ioctl_get_lapic(struct kvm_vcpu *vcpu,
1635 struct kvm_lapic_state *s)
1638 memcpy(s->regs, vcpu->arch.apic->regs, sizeof *s);
1644 static int kvm_vcpu_ioctl_set_lapic(struct kvm_vcpu *vcpu,
1645 struct kvm_lapic_state *s)
1648 memcpy(vcpu->arch.apic->regs, s->regs, sizeof *s);
1649 kvm_apic_post_state_restore(vcpu);
1650 update_cr8_intercept(vcpu);
1656 static int kvm_vcpu_ioctl_interrupt(struct kvm_vcpu *vcpu,
1657 struct kvm_interrupt *irq)
1659 if (irq->irq < 0 || irq->irq >= 256)
1661 if (irqchip_in_kernel(vcpu->kvm))
1665 kvm_queue_interrupt(vcpu, irq->irq, false);
1672 static int kvm_vcpu_ioctl_nmi(struct kvm_vcpu *vcpu)
1675 kvm_inject_nmi(vcpu);
1681 static int vcpu_ioctl_tpr_access_reporting(struct kvm_vcpu *vcpu,
1682 struct kvm_tpr_access_ctl *tac)
1686 vcpu->arch.tpr_access_reporting = !!tac->enabled;
1690 static int kvm_vcpu_ioctl_x86_setup_mce(struct kvm_vcpu *vcpu,
1694 unsigned bank_num = mcg_cap & 0xff, bank;
1697 if (!bank_num || bank_num >= KVM_MAX_MCE_BANKS)
1699 if (mcg_cap & ~(KVM_MCE_CAP_SUPPORTED | 0xff | 0xff0000))
1702 vcpu->arch.mcg_cap = mcg_cap;
1703 /* Init IA32_MCG_CTL to all 1s */
1704 if (mcg_cap & MCG_CTL_P)
1705 vcpu->arch.mcg_ctl = ~(u64)0;
1706 /* Init IA32_MCi_CTL to all 1s */
1707 for (bank = 0; bank < bank_num; bank++)
1708 vcpu->arch.mce_banks[bank*4] = ~(u64)0;
1713 static int kvm_vcpu_ioctl_x86_set_mce(struct kvm_vcpu *vcpu,
1714 struct kvm_x86_mce *mce)
1716 u64 mcg_cap = vcpu->arch.mcg_cap;
1717 unsigned bank_num = mcg_cap & 0xff;
1718 u64 *banks = vcpu->arch.mce_banks;
1720 if (mce->bank >= bank_num || !(mce->status & MCI_STATUS_VAL))
1723 * if IA32_MCG_CTL is not all 1s, the uncorrected error
1724 * reporting is disabled
1726 if ((mce->status & MCI_STATUS_UC) && (mcg_cap & MCG_CTL_P) &&
1727 vcpu->arch.mcg_ctl != ~(u64)0)
1729 banks += 4 * mce->bank;
1731 * if IA32_MCi_CTL is not all 1s, the uncorrected error
1732 * reporting is disabled for the bank
1734 if ((mce->status & MCI_STATUS_UC) && banks[0] != ~(u64)0)
1736 if (mce->status & MCI_STATUS_UC) {
1737 if ((vcpu->arch.mcg_status & MCG_STATUS_MCIP) ||
1738 !(vcpu->arch.cr4 & X86_CR4_MCE)) {
1739 printk(KERN_DEBUG "kvm: set_mce: "
1740 "injects mce exception while "
1741 "previous one is in progress!\n");
1742 set_bit(KVM_REQ_TRIPLE_FAULT, &vcpu->requests);
1745 if (banks[1] & MCI_STATUS_VAL)
1746 mce->status |= MCI_STATUS_OVER;
1747 banks[2] = mce->addr;
1748 banks[3] = mce->misc;
1749 vcpu->arch.mcg_status = mce->mcg_status;
1750 banks[1] = mce->status;
1751 kvm_queue_exception(vcpu, MC_VECTOR);
1752 } else if (!(banks[1] & MCI_STATUS_VAL)
1753 || !(banks[1] & MCI_STATUS_UC)) {
1754 if (banks[1] & MCI_STATUS_VAL)
1755 mce->status |= MCI_STATUS_OVER;
1756 banks[2] = mce->addr;
1757 banks[3] = mce->misc;
1758 banks[1] = mce->status;
1760 banks[1] |= MCI_STATUS_OVER;
1764 long kvm_arch_vcpu_ioctl(struct file *filp,
1765 unsigned int ioctl, unsigned long arg)
1767 struct kvm_vcpu *vcpu = filp->private_data;
1768 void __user *argp = (void __user *)arg;
1770 struct kvm_lapic_state *lapic = NULL;
1773 case KVM_GET_LAPIC: {
1774 lapic = kzalloc(sizeof(struct kvm_lapic_state), GFP_KERNEL);
1779 r = kvm_vcpu_ioctl_get_lapic(vcpu, lapic);
1783 if (copy_to_user(argp, lapic, sizeof(struct kvm_lapic_state)))
1788 case KVM_SET_LAPIC: {
1789 lapic = kmalloc(sizeof(struct kvm_lapic_state), GFP_KERNEL);
1794 if (copy_from_user(lapic, argp, sizeof(struct kvm_lapic_state)))
1796 r = kvm_vcpu_ioctl_set_lapic(vcpu, lapic);
1802 case KVM_INTERRUPT: {
1803 struct kvm_interrupt irq;
1806 if (copy_from_user(&irq, argp, sizeof irq))
1808 r = kvm_vcpu_ioctl_interrupt(vcpu, &irq);
1815 r = kvm_vcpu_ioctl_nmi(vcpu);
1821 case KVM_SET_CPUID: {
1822 struct kvm_cpuid __user *cpuid_arg = argp;
1823 struct kvm_cpuid cpuid;
1826 if (copy_from_user(&cpuid, cpuid_arg, sizeof cpuid))
1828 r = kvm_vcpu_ioctl_set_cpuid(vcpu, &cpuid, cpuid_arg->entries);
1833 case KVM_SET_CPUID2: {
1834 struct kvm_cpuid2 __user *cpuid_arg = argp;
1835 struct kvm_cpuid2 cpuid;
1838 if (copy_from_user(&cpuid, cpuid_arg, sizeof cpuid))
1840 r = kvm_vcpu_ioctl_set_cpuid2(vcpu, &cpuid,
1841 cpuid_arg->entries);
1846 case KVM_GET_CPUID2: {
1847 struct kvm_cpuid2 __user *cpuid_arg = argp;
1848 struct kvm_cpuid2 cpuid;
1851 if (copy_from_user(&cpuid, cpuid_arg, sizeof cpuid))
1853 r = kvm_vcpu_ioctl_get_cpuid2(vcpu, &cpuid,
1854 cpuid_arg->entries);
1858 if (copy_to_user(cpuid_arg, &cpuid, sizeof cpuid))
1864 r = msr_io(vcpu, argp, kvm_get_msr, 1);
1867 r = msr_io(vcpu, argp, do_set_msr, 0);
1869 case KVM_TPR_ACCESS_REPORTING: {
1870 struct kvm_tpr_access_ctl tac;
1873 if (copy_from_user(&tac, argp, sizeof tac))
1875 r = vcpu_ioctl_tpr_access_reporting(vcpu, &tac);
1879 if (copy_to_user(argp, &tac, sizeof tac))
1884 case KVM_SET_VAPIC_ADDR: {
1885 struct kvm_vapic_addr va;
1888 if (!irqchip_in_kernel(vcpu->kvm))
1891 if (copy_from_user(&va, argp, sizeof va))
1894 kvm_lapic_set_vapic_addr(vcpu, va.vapic_addr);
1897 case KVM_X86_SETUP_MCE: {
1901 if (copy_from_user(&mcg_cap, argp, sizeof mcg_cap))
1903 r = kvm_vcpu_ioctl_x86_setup_mce(vcpu, mcg_cap);
1906 case KVM_X86_SET_MCE: {
1907 struct kvm_x86_mce mce;
1910 if (copy_from_user(&mce, argp, sizeof mce))
1912 r = kvm_vcpu_ioctl_x86_set_mce(vcpu, &mce);
1923 static int kvm_vm_ioctl_set_tss_addr(struct kvm *kvm, unsigned long addr)
1927 if (addr > (unsigned int)(-3 * PAGE_SIZE))
1929 ret = kvm_x86_ops->set_tss_addr(kvm, addr);
1933 static int kvm_vm_ioctl_set_identity_map_addr(struct kvm *kvm,
1936 kvm->arch.ept_identity_map_addr = ident_addr;
1940 static int kvm_vm_ioctl_set_nr_mmu_pages(struct kvm *kvm,
1941 u32 kvm_nr_mmu_pages)
1943 if (kvm_nr_mmu_pages < KVM_MIN_ALLOC_MMU_PAGES)
1946 down_write(&kvm->slots_lock);
1947 spin_lock(&kvm->mmu_lock);
1949 kvm_mmu_change_mmu_pages(kvm, kvm_nr_mmu_pages);
1950 kvm->arch.n_requested_mmu_pages = kvm_nr_mmu_pages;
1952 spin_unlock(&kvm->mmu_lock);
1953 up_write(&kvm->slots_lock);
1957 static int kvm_vm_ioctl_get_nr_mmu_pages(struct kvm *kvm)
1959 return kvm->arch.n_alloc_mmu_pages;
1962 gfn_t unalias_gfn(struct kvm *kvm, gfn_t gfn)
1965 struct kvm_mem_alias *alias;
1967 for (i = 0; i < kvm->arch.naliases; ++i) {
1968 alias = &kvm->arch.aliases[i];
1969 if (gfn >= alias->base_gfn
1970 && gfn < alias->base_gfn + alias->npages)
1971 return alias->target_gfn + gfn - alias->base_gfn;
1977 * Set a new alias region. Aliases map a portion of physical memory into
1978 * another portion. This is useful for memory windows, for example the PC
1981 static int kvm_vm_ioctl_set_memory_alias(struct kvm *kvm,
1982 struct kvm_memory_alias *alias)
1985 struct kvm_mem_alias *p;
1988 /* General sanity checks */
1989 if (alias->memory_size & (PAGE_SIZE - 1))
1991 if (alias->guest_phys_addr & (PAGE_SIZE - 1))
1993 if (alias->slot >= KVM_ALIAS_SLOTS)
1995 if (alias->guest_phys_addr + alias->memory_size
1996 < alias->guest_phys_addr)
1998 if (alias->target_phys_addr + alias->memory_size
1999 < alias->target_phys_addr)
2002 down_write(&kvm->slots_lock);
2003 spin_lock(&kvm->mmu_lock);
2005 p = &kvm->arch.aliases[alias->slot];
2006 p->base_gfn = alias->guest_phys_addr >> PAGE_SHIFT;
2007 p->npages = alias->memory_size >> PAGE_SHIFT;
2008 p->target_gfn = alias->target_phys_addr >> PAGE_SHIFT;
2010 for (n = KVM_ALIAS_SLOTS; n > 0; --n)
2011 if (kvm->arch.aliases[n - 1].npages)
2013 kvm->arch.naliases = n;
2015 spin_unlock(&kvm->mmu_lock);
2016 kvm_mmu_zap_all(kvm);
2018 up_write(&kvm->slots_lock);
2026 static int kvm_vm_ioctl_get_irqchip(struct kvm *kvm, struct kvm_irqchip *chip)
2031 switch (chip->chip_id) {
2032 case KVM_IRQCHIP_PIC_MASTER:
2033 memcpy(&chip->chip.pic,
2034 &pic_irqchip(kvm)->pics[0],
2035 sizeof(struct kvm_pic_state));
2037 case KVM_IRQCHIP_PIC_SLAVE:
2038 memcpy(&chip->chip.pic,
2039 &pic_irqchip(kvm)->pics[1],
2040 sizeof(struct kvm_pic_state));
2042 case KVM_IRQCHIP_IOAPIC:
2043 r = kvm_get_ioapic(kvm, &chip->chip.ioapic);
2052 static int kvm_vm_ioctl_set_irqchip(struct kvm *kvm, struct kvm_irqchip *chip)
2057 switch (chip->chip_id) {
2058 case KVM_IRQCHIP_PIC_MASTER:
2059 spin_lock(&pic_irqchip(kvm)->lock);
2060 memcpy(&pic_irqchip(kvm)->pics[0],
2062 sizeof(struct kvm_pic_state));
2063 spin_unlock(&pic_irqchip(kvm)->lock);
2065 case KVM_IRQCHIP_PIC_SLAVE:
2066 spin_lock(&pic_irqchip(kvm)->lock);
2067 memcpy(&pic_irqchip(kvm)->pics[1],
2069 sizeof(struct kvm_pic_state));
2070 spin_unlock(&pic_irqchip(kvm)->lock);
2072 case KVM_IRQCHIP_IOAPIC:
2073 r = kvm_set_ioapic(kvm, &chip->chip.ioapic);
2079 kvm_pic_update_irq(pic_irqchip(kvm));
2083 static int kvm_vm_ioctl_get_pit(struct kvm *kvm, struct kvm_pit_state *ps)
2087 mutex_lock(&kvm->arch.vpit->pit_state.lock);
2088 memcpy(ps, &kvm->arch.vpit->pit_state, sizeof(struct kvm_pit_state));
2089 mutex_unlock(&kvm->arch.vpit->pit_state.lock);
2093 static int kvm_vm_ioctl_set_pit(struct kvm *kvm, struct kvm_pit_state *ps)
2097 mutex_lock(&kvm->arch.vpit->pit_state.lock);
2098 memcpy(&kvm->arch.vpit->pit_state, ps, sizeof(struct kvm_pit_state));
2099 kvm_pit_load_count(kvm, 0, ps->channels[0].count, 0);
2100 mutex_unlock(&kvm->arch.vpit->pit_state.lock);
2104 static int kvm_vm_ioctl_get_pit2(struct kvm *kvm, struct kvm_pit_state2 *ps)
2108 mutex_lock(&kvm->arch.vpit->pit_state.lock);
2109 memcpy(ps->channels, &kvm->arch.vpit->pit_state.channels,
2110 sizeof(ps->channels));
2111 ps->flags = kvm->arch.vpit->pit_state.flags;
2112 mutex_unlock(&kvm->arch.vpit->pit_state.lock);
2116 static int kvm_vm_ioctl_set_pit2(struct kvm *kvm, struct kvm_pit_state2 *ps)
2118 int r = 0, start = 0;
2119 u32 prev_legacy, cur_legacy;
2120 mutex_lock(&kvm->arch.vpit->pit_state.lock);
2121 prev_legacy = kvm->arch.vpit->pit_state.flags & KVM_PIT_FLAGS_HPET_LEGACY;
2122 cur_legacy = ps->flags & KVM_PIT_FLAGS_HPET_LEGACY;
2123 if (!prev_legacy && cur_legacy)
2125 memcpy(&kvm->arch.vpit->pit_state.channels, &ps->channels,
2126 sizeof(kvm->arch.vpit->pit_state.channels));
2127 kvm->arch.vpit->pit_state.flags = ps->flags;
2128 kvm_pit_load_count(kvm, 0, kvm->arch.vpit->pit_state.channels[0].count, start);
2129 mutex_unlock(&kvm->arch.vpit->pit_state.lock);
2133 static int kvm_vm_ioctl_reinject(struct kvm *kvm,
2134 struct kvm_reinject_control *control)
2136 if (!kvm->arch.vpit)
2138 mutex_lock(&kvm->arch.vpit->pit_state.lock);
2139 kvm->arch.vpit->pit_state.pit_timer.reinject = control->pit_reinject;
2140 mutex_unlock(&kvm->arch.vpit->pit_state.lock);
2145 * Get (and clear) the dirty memory log for a memory slot.
2147 int kvm_vm_ioctl_get_dirty_log(struct kvm *kvm,
2148 struct kvm_dirty_log *log)
2152 struct kvm_memory_slot *memslot;
2155 down_write(&kvm->slots_lock);
2157 r = kvm_get_dirty_log(kvm, log, &is_dirty);
2161 /* If nothing is dirty, don't bother messing with page tables. */
2163 spin_lock(&kvm->mmu_lock);
2164 kvm_mmu_slot_remove_write_access(kvm, log->slot);
2165 spin_unlock(&kvm->mmu_lock);
2166 memslot = &kvm->memslots[log->slot];
2167 n = ALIGN(memslot->npages, BITS_PER_LONG) / 8;
2168 memset(memslot->dirty_bitmap, 0, n);
2172 up_write(&kvm->slots_lock);
2176 long kvm_arch_vm_ioctl(struct file *filp,
2177 unsigned int ioctl, unsigned long arg)
2179 struct kvm *kvm = filp->private_data;
2180 void __user *argp = (void __user *)arg;
2183 * This union makes it completely explicit to gcc-3.x
2184 * that these two variables' stack usage should be
2185 * combined, not added together.
2188 struct kvm_pit_state ps;
2189 struct kvm_pit_state2 ps2;
2190 struct kvm_memory_alias alias;
2191 struct kvm_pit_config pit_config;
2195 case KVM_SET_TSS_ADDR:
2196 r = kvm_vm_ioctl_set_tss_addr(kvm, arg);
2200 case KVM_SET_IDENTITY_MAP_ADDR: {
2204 if (copy_from_user(&ident_addr, argp, sizeof ident_addr))
2206 r = kvm_vm_ioctl_set_identity_map_addr(kvm, ident_addr);
2211 case KVM_SET_MEMORY_REGION: {
2212 struct kvm_memory_region kvm_mem;
2213 struct kvm_userspace_memory_region kvm_userspace_mem;
2216 if (copy_from_user(&kvm_mem, argp, sizeof kvm_mem))
2218 kvm_userspace_mem.slot = kvm_mem.slot;
2219 kvm_userspace_mem.flags = kvm_mem.flags;
2220 kvm_userspace_mem.guest_phys_addr = kvm_mem.guest_phys_addr;
2221 kvm_userspace_mem.memory_size = kvm_mem.memory_size;
2222 r = kvm_vm_ioctl_set_memory_region(kvm, &kvm_userspace_mem, 0);
2227 case KVM_SET_NR_MMU_PAGES:
2228 r = kvm_vm_ioctl_set_nr_mmu_pages(kvm, arg);
2232 case KVM_GET_NR_MMU_PAGES:
2233 r = kvm_vm_ioctl_get_nr_mmu_pages(kvm);
2235 case KVM_SET_MEMORY_ALIAS:
2237 if (copy_from_user(&u.alias, argp, sizeof(struct kvm_memory_alias)))
2239 r = kvm_vm_ioctl_set_memory_alias(kvm, &u.alias);
2243 case KVM_CREATE_IRQCHIP:
2245 kvm->arch.vpic = kvm_create_pic(kvm);
2246 if (kvm->arch.vpic) {
2247 r = kvm_ioapic_init(kvm);
2249 kfree(kvm->arch.vpic);
2250 kvm->arch.vpic = NULL;
2255 r = kvm_setup_default_irq_routing(kvm);
2257 kfree(kvm->arch.vpic);
2258 kfree(kvm->arch.vioapic);
2262 case KVM_CREATE_PIT:
2263 u.pit_config.flags = KVM_PIT_SPEAKER_DUMMY;
2265 case KVM_CREATE_PIT2:
2267 if (copy_from_user(&u.pit_config, argp,
2268 sizeof(struct kvm_pit_config)))
2271 down_write(&kvm->slots_lock);
2274 goto create_pit_unlock;
2276 kvm->arch.vpit = kvm_create_pit(kvm, u.pit_config.flags);
2280 up_write(&kvm->slots_lock);
2282 case KVM_IRQ_LINE_STATUS:
2283 case KVM_IRQ_LINE: {
2284 struct kvm_irq_level irq_event;
2287 if (copy_from_user(&irq_event, argp, sizeof irq_event))
2289 if (irqchip_in_kernel(kvm)) {
2291 status = kvm_set_irq(kvm, KVM_USERSPACE_IRQ_SOURCE_ID,
2292 irq_event.irq, irq_event.level);
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 int kvmclock_cpufreq_notifier(struct notifier_block *nb, unsigned long val,
3071 struct cpufreq_freqs *freq = data;
3073 struct kvm_vcpu *vcpu;
3074 int i, send_ipi = 0;
3076 if (val == CPUFREQ_PRECHANGE && freq->old > freq->new)
3078 if (val == CPUFREQ_POSTCHANGE && freq->old < freq->new)
3080 per_cpu(cpu_tsc_khz, freq->cpu) = freq->new;
3082 spin_lock(&kvm_lock);
3083 list_for_each_entry(kvm, &vm_list, vm_list) {
3084 kvm_for_each_vcpu(i, vcpu, kvm) {
3085 if (vcpu->cpu != freq->cpu)
3087 if (!kvm_request_guest_time_update(vcpu))
3089 if (vcpu->cpu != smp_processor_id())
3093 spin_unlock(&kvm_lock);
3095 if (freq->old < freq->new && send_ipi) {
3097 * We upscale the frequency. Must make the guest
3098 * doesn't see old kvmclock values while running with
3099 * the new frequency, otherwise we risk the guest sees
3100 * time go backwards.
3102 * In case we update the frequency for another cpu
3103 * (which might be in guest context) send an interrupt
3104 * to kick the cpu out of guest context. Next time
3105 * guest context is entered kvmclock will be updated,
3106 * so the guest will not see stale values.
3108 smp_call_function_single(freq->cpu, bounce_off, NULL, 1);
3113 static struct notifier_block kvmclock_cpufreq_notifier_block = {
3114 .notifier_call = kvmclock_cpufreq_notifier
3117 static void kvm_timer_init(void)
3121 if (!boot_cpu_has(X86_FEATURE_CONSTANT_TSC)) {
3122 cpufreq_register_notifier(&kvmclock_cpufreq_notifier_block,
3123 CPUFREQ_TRANSITION_NOTIFIER);
3124 for_each_online_cpu(cpu)
3125 per_cpu(cpu_tsc_khz, cpu) = cpufreq_get(cpu);
3127 for_each_possible_cpu(cpu)
3128 per_cpu(cpu_tsc_khz, cpu) = tsc_khz;
3132 int kvm_arch_init(void *opaque)
3135 struct kvm_x86_ops *ops = (struct kvm_x86_ops *)opaque;
3138 printk(KERN_ERR "kvm: already loaded the other module\n");
3143 if (!ops->cpu_has_kvm_support()) {
3144 printk(KERN_ERR "kvm: no hardware support\n");
3148 if (ops->disabled_by_bios()) {
3149 printk(KERN_ERR "kvm: disabled by bios\n");
3154 r = kvm_mmu_module_init();
3158 kvm_init_msr_list();
3161 kvm_mmu_set_nonpresent_ptes(0ull, 0ull);
3162 kvm_mmu_set_base_ptes(PT_PRESENT_MASK);
3163 kvm_mmu_set_mask_ptes(PT_USER_MASK, PT_ACCESSED_MASK,
3164 PT_DIRTY_MASK, PT64_NX_MASK, 0);
3174 void kvm_arch_exit(void)
3176 if (!boot_cpu_has(X86_FEATURE_CONSTANT_TSC))
3177 cpufreq_unregister_notifier(&kvmclock_cpufreq_notifier_block,
3178 CPUFREQ_TRANSITION_NOTIFIER);
3180 kvm_mmu_module_exit();
3183 int kvm_emulate_halt(struct kvm_vcpu *vcpu)
3185 ++vcpu->stat.halt_exits;
3186 if (irqchip_in_kernel(vcpu->kvm)) {
3187 vcpu->arch.mp_state = KVM_MP_STATE_HALTED;
3190 vcpu->run->exit_reason = KVM_EXIT_HLT;
3194 EXPORT_SYMBOL_GPL(kvm_emulate_halt);
3196 static inline gpa_t hc_gpa(struct kvm_vcpu *vcpu, unsigned long a0,
3199 if (is_long_mode(vcpu))
3202 return a0 | ((gpa_t)a1 << 32);
3205 int kvm_emulate_hypercall(struct kvm_vcpu *vcpu)
3207 unsigned long nr, a0, a1, a2, a3, ret;
3210 nr = kvm_register_read(vcpu, VCPU_REGS_RAX);
3211 a0 = kvm_register_read(vcpu, VCPU_REGS_RBX);
3212 a1 = kvm_register_read(vcpu, VCPU_REGS_RCX);
3213 a2 = kvm_register_read(vcpu, VCPU_REGS_RDX);
3214 a3 = kvm_register_read(vcpu, VCPU_REGS_RSI);
3216 trace_kvm_hypercall(nr, a0, a1, a2, a3);
3218 if (!is_long_mode(vcpu)) {
3226 if (kvm_x86_ops->get_cpl(vcpu) != 0) {
3232 case KVM_HC_VAPIC_POLL_IRQ:
3236 r = kvm_pv_mmu_op(vcpu, a0, hc_gpa(vcpu, a1, a2), &ret);
3243 kvm_register_write(vcpu, VCPU_REGS_RAX, ret);
3244 ++vcpu->stat.hypercalls;
3247 EXPORT_SYMBOL_GPL(kvm_emulate_hypercall);
3249 int kvm_fix_hypercall(struct kvm_vcpu *vcpu)
3251 char instruction[3];
3253 unsigned long rip = kvm_rip_read(vcpu);
3257 * Blow out the MMU to ensure that no other VCPU has an active mapping
3258 * to ensure that the updated hypercall appears atomically across all
3261 kvm_mmu_zap_all(vcpu->kvm);
3263 kvm_x86_ops->patch_hypercall(vcpu, instruction);
3264 if (emulator_write_emulated(rip, instruction, 3, vcpu)
3265 != X86EMUL_CONTINUE)
3271 static u64 mk_cr_64(u64 curr_cr, u32 new_val)
3273 return (curr_cr & ~((1ULL << 32) - 1)) | new_val;
3276 void realmode_lgdt(struct kvm_vcpu *vcpu, u16 limit, unsigned long base)
3278 struct descriptor_table dt = { limit, base };
3280 kvm_x86_ops->set_gdt(vcpu, &dt);
3283 void realmode_lidt(struct kvm_vcpu *vcpu, u16 limit, unsigned long base)
3285 struct descriptor_table dt = { limit, base };
3287 kvm_x86_ops->set_idt(vcpu, &dt);
3290 void realmode_lmsw(struct kvm_vcpu *vcpu, unsigned long msw,
3291 unsigned long *rflags)
3293 kvm_lmsw(vcpu, msw);
3294 *rflags = kvm_x86_ops->get_rflags(vcpu);
3297 unsigned long realmode_get_cr(struct kvm_vcpu *vcpu, int cr)
3299 unsigned long value;
3301 kvm_x86_ops->decache_cr4_guest_bits(vcpu);
3304 value = vcpu->arch.cr0;
3307 value = vcpu->arch.cr2;
3310 value = vcpu->arch.cr3;
3313 value = vcpu->arch.cr4;
3316 value = kvm_get_cr8(vcpu);
3319 vcpu_printf(vcpu, "%s: unexpected cr %u\n", __func__, cr);
3326 void realmode_set_cr(struct kvm_vcpu *vcpu, int cr, unsigned long val,
3327 unsigned long *rflags)
3331 kvm_set_cr0(vcpu, mk_cr_64(vcpu->arch.cr0, val));
3332 *rflags = kvm_x86_ops->get_rflags(vcpu);
3335 vcpu->arch.cr2 = val;
3338 kvm_set_cr3(vcpu, val);
3341 kvm_set_cr4(vcpu, mk_cr_64(vcpu->arch.cr4, val));
3344 kvm_set_cr8(vcpu, val & 0xfUL);
3347 vcpu_printf(vcpu, "%s: unexpected cr %u\n", __func__, cr);
3351 static int move_to_next_stateful_cpuid_entry(struct kvm_vcpu *vcpu, int i)
3353 struct kvm_cpuid_entry2 *e = &vcpu->arch.cpuid_entries[i];
3354 int j, nent = vcpu->arch.cpuid_nent;
3356 e->flags &= ~KVM_CPUID_FLAG_STATE_READ_NEXT;
3357 /* when no next entry is found, the current entry[i] is reselected */
3358 for (j = i + 1; ; j = (j + 1) % nent) {
3359 struct kvm_cpuid_entry2 *ej = &vcpu->arch.cpuid_entries[j];
3360 if (ej->function == e->function) {
3361 ej->flags |= KVM_CPUID_FLAG_STATE_READ_NEXT;
3365 return 0; /* silence gcc, even though control never reaches here */
3368 /* find an entry with matching function, matching index (if needed), and that
3369 * should be read next (if it's stateful) */
3370 static int is_matching_cpuid_entry(struct kvm_cpuid_entry2 *e,
3371 u32 function, u32 index)
3373 if (e->function != function)
3375 if ((e->flags & KVM_CPUID_FLAG_SIGNIFCANT_INDEX) && e->index != index)
3377 if ((e->flags & KVM_CPUID_FLAG_STATEFUL_FUNC) &&
3378 !(e->flags & KVM_CPUID_FLAG_STATE_READ_NEXT))
3383 struct kvm_cpuid_entry2 *kvm_find_cpuid_entry(struct kvm_vcpu *vcpu,
3384 u32 function, u32 index)
3387 struct kvm_cpuid_entry2 *best = NULL;
3389 for (i = 0; i < vcpu->arch.cpuid_nent; ++i) {
3390 struct kvm_cpuid_entry2 *e;
3392 e = &vcpu->arch.cpuid_entries[i];
3393 if (is_matching_cpuid_entry(e, function, index)) {
3394 if (e->flags & KVM_CPUID_FLAG_STATEFUL_FUNC)
3395 move_to_next_stateful_cpuid_entry(vcpu, i);
3400 * Both basic or both extended?
3402 if (((e->function ^ function) & 0x80000000) == 0)
3403 if (!best || e->function > best->function)
3409 int cpuid_maxphyaddr(struct kvm_vcpu *vcpu)
3411 struct kvm_cpuid_entry2 *best;
3413 best = kvm_find_cpuid_entry(vcpu, 0x80000008, 0);
3415 return best->eax & 0xff;
3419 void kvm_emulate_cpuid(struct kvm_vcpu *vcpu)
3421 u32 function, index;
3422 struct kvm_cpuid_entry2 *best;
3424 function = kvm_register_read(vcpu, VCPU_REGS_RAX);
3425 index = kvm_register_read(vcpu, VCPU_REGS_RCX);
3426 kvm_register_write(vcpu, VCPU_REGS_RAX, 0);
3427 kvm_register_write(vcpu, VCPU_REGS_RBX, 0);
3428 kvm_register_write(vcpu, VCPU_REGS_RCX, 0);
3429 kvm_register_write(vcpu, VCPU_REGS_RDX, 0);
3430 best = kvm_find_cpuid_entry(vcpu, function, index);
3432 kvm_register_write(vcpu, VCPU_REGS_RAX, best->eax);
3433 kvm_register_write(vcpu, VCPU_REGS_RBX, best->ebx);
3434 kvm_register_write(vcpu, VCPU_REGS_RCX, best->ecx);
3435 kvm_register_write(vcpu, VCPU_REGS_RDX, best->edx);
3437 kvm_x86_ops->skip_emulated_instruction(vcpu);
3438 trace_kvm_cpuid(function,
3439 kvm_register_read(vcpu, VCPU_REGS_RAX),
3440 kvm_register_read(vcpu, VCPU_REGS_RBX),
3441 kvm_register_read(vcpu, VCPU_REGS_RCX),
3442 kvm_register_read(vcpu, VCPU_REGS_RDX));
3444 EXPORT_SYMBOL_GPL(kvm_emulate_cpuid);
3447 * Check if userspace requested an interrupt window, and that the
3448 * interrupt window is open.
3450 * No need to exit to userspace if we already have an interrupt queued.
3452 static int dm_request_for_irq_injection(struct kvm_vcpu *vcpu)
3454 return (!irqchip_in_kernel(vcpu->kvm) && !kvm_cpu_has_interrupt(vcpu) &&
3455 vcpu->run->request_interrupt_window &&
3456 kvm_arch_interrupt_allowed(vcpu));
3459 static void post_kvm_run_save(struct kvm_vcpu *vcpu)
3461 struct kvm_run *kvm_run = vcpu->run;
3463 kvm_run->if_flag = (kvm_x86_ops->get_rflags(vcpu) & X86_EFLAGS_IF) != 0;
3464 kvm_run->cr8 = kvm_get_cr8(vcpu);
3465 kvm_run->apic_base = kvm_get_apic_base(vcpu);
3466 if (irqchip_in_kernel(vcpu->kvm))
3467 kvm_run->ready_for_interrupt_injection = 1;
3469 kvm_run->ready_for_interrupt_injection =
3470 kvm_arch_interrupt_allowed(vcpu) &&
3471 !kvm_cpu_has_interrupt(vcpu) &&
3472 !kvm_event_needs_reinjection(vcpu);
3475 static void vapic_enter(struct kvm_vcpu *vcpu)
3477 struct kvm_lapic *apic = vcpu->arch.apic;
3480 if (!apic || !apic->vapic_addr)
3483 page = gfn_to_page(vcpu->kvm, apic->vapic_addr >> PAGE_SHIFT);
3485 vcpu->arch.apic->vapic_page = page;
3488 static void vapic_exit(struct kvm_vcpu *vcpu)
3490 struct kvm_lapic *apic = vcpu->arch.apic;
3492 if (!apic || !apic->vapic_addr)
3495 down_read(&vcpu->kvm->slots_lock);
3496 kvm_release_page_dirty(apic->vapic_page);
3497 mark_page_dirty(vcpu->kvm, apic->vapic_addr >> PAGE_SHIFT);
3498 up_read(&vcpu->kvm->slots_lock);
3501 static void update_cr8_intercept(struct kvm_vcpu *vcpu)
3505 if (!kvm_x86_ops->update_cr8_intercept)
3508 if (!vcpu->arch.apic)
3511 if (!vcpu->arch.apic->vapic_addr)
3512 max_irr = kvm_lapic_find_highest_irr(vcpu);
3519 tpr = kvm_lapic_get_cr8(vcpu);
3521 kvm_x86_ops->update_cr8_intercept(vcpu, tpr, max_irr);
3524 static void inject_pending_event(struct kvm_vcpu *vcpu)
3526 /* try to reinject previous events if any */
3527 if (vcpu->arch.exception.pending) {
3528 kvm_x86_ops->queue_exception(vcpu, vcpu->arch.exception.nr,
3529 vcpu->arch.exception.has_error_code,
3530 vcpu->arch.exception.error_code);
3534 if (vcpu->arch.nmi_injected) {
3535 kvm_x86_ops->set_nmi(vcpu);
3539 if (vcpu->arch.interrupt.pending) {
3540 kvm_x86_ops->set_irq(vcpu);
3544 /* try to inject new event if pending */
3545 if (vcpu->arch.nmi_pending) {
3546 if (kvm_x86_ops->nmi_allowed(vcpu)) {
3547 vcpu->arch.nmi_pending = false;
3548 vcpu->arch.nmi_injected = true;
3549 kvm_x86_ops->set_nmi(vcpu);
3551 } else if (kvm_cpu_has_interrupt(vcpu)) {
3552 if (kvm_x86_ops->interrupt_allowed(vcpu)) {
3553 kvm_queue_interrupt(vcpu, kvm_cpu_get_interrupt(vcpu),
3555 kvm_x86_ops->set_irq(vcpu);
3560 static int vcpu_enter_guest(struct kvm_vcpu *vcpu)
3563 bool req_int_win = !irqchip_in_kernel(vcpu->kvm) &&
3564 vcpu->run->request_interrupt_window;
3567 if (test_and_clear_bit(KVM_REQ_MMU_RELOAD, &vcpu->requests))
3568 kvm_mmu_unload(vcpu);
3570 r = kvm_mmu_reload(vcpu);
3574 if (vcpu->requests) {
3575 if (test_and_clear_bit(KVM_REQ_MIGRATE_TIMER, &vcpu->requests))
3576 __kvm_migrate_timers(vcpu);
3577 if (test_and_clear_bit(KVM_REQ_KVMCLOCK_UPDATE, &vcpu->requests))
3578 kvm_write_guest_time(vcpu);
3579 if (test_and_clear_bit(KVM_REQ_MMU_SYNC, &vcpu->requests))
3580 kvm_mmu_sync_roots(vcpu);
3581 if (test_and_clear_bit(KVM_REQ_TLB_FLUSH, &vcpu->requests))
3582 kvm_x86_ops->tlb_flush(vcpu);
3583 if (test_and_clear_bit(KVM_REQ_REPORT_TPR_ACCESS,
3585 vcpu->run->exit_reason = KVM_EXIT_TPR_ACCESS;
3589 if (test_and_clear_bit(KVM_REQ_TRIPLE_FAULT, &vcpu->requests)) {
3590 vcpu->run->exit_reason = KVM_EXIT_SHUTDOWN;
3598 kvm_x86_ops->prepare_guest_switch(vcpu);
3599 kvm_load_guest_fpu(vcpu);
3601 local_irq_disable();
3603 clear_bit(KVM_REQ_KICK, &vcpu->requests);
3604 smp_mb__after_clear_bit();
3606 if (vcpu->requests || need_resched() || signal_pending(current)) {
3607 set_bit(KVM_REQ_KICK, &vcpu->requests);
3614 inject_pending_event(vcpu);
3616 /* enable NMI/IRQ window open exits if needed */
3617 if (vcpu->arch.nmi_pending)
3618 kvm_x86_ops->enable_nmi_window(vcpu);
3619 else if (kvm_cpu_has_interrupt(vcpu) || req_int_win)
3620 kvm_x86_ops->enable_irq_window(vcpu);
3622 if (kvm_lapic_enabled(vcpu)) {
3623 update_cr8_intercept(vcpu);
3624 kvm_lapic_sync_to_vapic(vcpu);
3627 up_read(&vcpu->kvm->slots_lock);
3631 if (unlikely(vcpu->arch.switch_db_regs)) {
3633 set_debugreg(vcpu->arch.eff_db[0], 0);
3634 set_debugreg(vcpu->arch.eff_db[1], 1);
3635 set_debugreg(vcpu->arch.eff_db[2], 2);
3636 set_debugreg(vcpu->arch.eff_db[3], 3);
3639 trace_kvm_entry(vcpu->vcpu_id);
3640 kvm_x86_ops->run(vcpu);
3642 if (unlikely(vcpu->arch.switch_db_regs || test_thread_flag(TIF_DEBUG))) {
3643 set_debugreg(current->thread.debugreg0, 0);
3644 set_debugreg(current->thread.debugreg1, 1);
3645 set_debugreg(current->thread.debugreg2, 2);
3646 set_debugreg(current->thread.debugreg3, 3);
3647 set_debugreg(current->thread.debugreg6, 6);
3648 set_debugreg(current->thread.debugreg7, 7);
3651 set_bit(KVM_REQ_KICK, &vcpu->requests);
3657 * We must have an instruction between local_irq_enable() and
3658 * kvm_guest_exit(), so the timer interrupt isn't delayed by
3659 * the interrupt shadow. The stat.exits increment will do nicely.
3660 * But we need to prevent reordering, hence this barrier():
3668 down_read(&vcpu->kvm->slots_lock);
3671 * Profile KVM exit RIPs:
3673 if (unlikely(prof_on == KVM_PROFILING)) {
3674 unsigned long rip = kvm_rip_read(vcpu);
3675 profile_hit(KVM_PROFILING, (void *)rip);
3679 kvm_lapic_sync_from_vapic(vcpu);
3681 r = kvm_x86_ops->handle_exit(vcpu);
3687 static int __vcpu_run(struct kvm_vcpu *vcpu)
3691 if (unlikely(vcpu->arch.mp_state == KVM_MP_STATE_SIPI_RECEIVED)) {
3692 pr_debug("vcpu %d received sipi with vector # %x\n",
3693 vcpu->vcpu_id, vcpu->arch.sipi_vector);
3694 kvm_lapic_reset(vcpu);
3695 r = kvm_arch_vcpu_reset(vcpu);
3698 vcpu->arch.mp_state = KVM_MP_STATE_RUNNABLE;
3701 down_read(&vcpu->kvm->slots_lock);
3706 if (vcpu->arch.mp_state == KVM_MP_STATE_RUNNABLE)
3707 r = vcpu_enter_guest(vcpu);
3709 up_read(&vcpu->kvm->slots_lock);
3710 kvm_vcpu_block(vcpu);
3711 down_read(&vcpu->kvm->slots_lock);
3712 if (test_and_clear_bit(KVM_REQ_UNHALT, &vcpu->requests))
3714 switch(vcpu->arch.mp_state) {
3715 case KVM_MP_STATE_HALTED:
3716 vcpu->arch.mp_state =
3717 KVM_MP_STATE_RUNNABLE;
3718 case KVM_MP_STATE_RUNNABLE:
3720 case KVM_MP_STATE_SIPI_RECEIVED:
3731 clear_bit(KVM_REQ_PENDING_TIMER, &vcpu->requests);
3732 if (kvm_cpu_has_pending_timer(vcpu))
3733 kvm_inject_pending_timer_irqs(vcpu);
3735 if (dm_request_for_irq_injection(vcpu)) {
3737 vcpu->run->exit_reason = KVM_EXIT_INTR;
3738 ++vcpu->stat.request_irq_exits;
3740 if (signal_pending(current)) {
3742 vcpu->run->exit_reason = KVM_EXIT_INTR;
3743 ++vcpu->stat.signal_exits;
3745 if (need_resched()) {
3746 up_read(&vcpu->kvm->slots_lock);
3748 down_read(&vcpu->kvm->slots_lock);
3752 up_read(&vcpu->kvm->slots_lock);
3753 post_kvm_run_save(vcpu);
3760 int kvm_arch_vcpu_ioctl_run(struct kvm_vcpu *vcpu, struct kvm_run *kvm_run)
3767 if (vcpu->sigset_active)
3768 sigprocmask(SIG_SETMASK, &vcpu->sigset, &sigsaved);
3770 if (unlikely(vcpu->arch.mp_state == KVM_MP_STATE_UNINITIALIZED)) {
3771 kvm_vcpu_block(vcpu);
3772 clear_bit(KVM_REQ_UNHALT, &vcpu->requests);
3777 /* re-sync apic's tpr */
3778 if (!irqchip_in_kernel(vcpu->kvm))
3779 kvm_set_cr8(vcpu, kvm_run->cr8);
3781 if (vcpu->arch.pio.cur_count) {
3782 r = complete_pio(vcpu);
3786 #if CONFIG_HAS_IOMEM
3787 if (vcpu->mmio_needed) {
3788 memcpy(vcpu->mmio_data, kvm_run->mmio.data, 8);
3789 vcpu->mmio_read_completed = 1;
3790 vcpu->mmio_needed = 0;
3792 down_read(&vcpu->kvm->slots_lock);
3793 r = emulate_instruction(vcpu, vcpu->arch.mmio_fault_cr2, 0,
3794 EMULTYPE_NO_DECODE);
3795 up_read(&vcpu->kvm->slots_lock);
3796 if (r == EMULATE_DO_MMIO) {
3798 * Read-modify-write. Back to userspace.
3805 if (kvm_run->exit_reason == KVM_EXIT_HYPERCALL)
3806 kvm_register_write(vcpu, VCPU_REGS_RAX,
3807 kvm_run->hypercall.ret);
3809 r = __vcpu_run(vcpu);
3812 if (vcpu->sigset_active)
3813 sigprocmask(SIG_SETMASK, &sigsaved, NULL);
3819 int kvm_arch_vcpu_ioctl_get_regs(struct kvm_vcpu *vcpu, struct kvm_regs *regs)
3823 regs->rax = kvm_register_read(vcpu, VCPU_REGS_RAX);
3824 regs->rbx = kvm_register_read(vcpu, VCPU_REGS_RBX);
3825 regs->rcx = kvm_register_read(vcpu, VCPU_REGS_RCX);
3826 regs->rdx = kvm_register_read(vcpu, VCPU_REGS_RDX);
3827 regs->rsi = kvm_register_read(vcpu, VCPU_REGS_RSI);
3828 regs->rdi = kvm_register_read(vcpu, VCPU_REGS_RDI);
3829 regs->rsp = kvm_register_read(vcpu, VCPU_REGS_RSP);
3830 regs->rbp = kvm_register_read(vcpu, VCPU_REGS_RBP);
3831 #ifdef CONFIG_X86_64
3832 regs->r8 = kvm_register_read(vcpu, VCPU_REGS_R8);
3833 regs->r9 = kvm_register_read(vcpu, VCPU_REGS_R9);
3834 regs->r10 = kvm_register_read(vcpu, VCPU_REGS_R10);
3835 regs->r11 = kvm_register_read(vcpu, VCPU_REGS_R11);
3836 regs->r12 = kvm_register_read(vcpu, VCPU_REGS_R12);
3837 regs->r13 = kvm_register_read(vcpu, VCPU_REGS_R13);
3838 regs->r14 = kvm_register_read(vcpu, VCPU_REGS_R14);
3839 regs->r15 = kvm_register_read(vcpu, VCPU_REGS_R15);
3842 regs->rip = kvm_rip_read(vcpu);
3843 regs->rflags = kvm_x86_ops->get_rflags(vcpu);
3846 * Don't leak debug flags in case they were set for guest debugging
3848 if (vcpu->guest_debug & KVM_GUESTDBG_SINGLESTEP)
3849 regs->rflags &= ~(X86_EFLAGS_TF | X86_EFLAGS_RF);
3856 int kvm_arch_vcpu_ioctl_set_regs(struct kvm_vcpu *vcpu, struct kvm_regs *regs)
3860 kvm_register_write(vcpu, VCPU_REGS_RAX, regs->rax);
3861 kvm_register_write(vcpu, VCPU_REGS_RBX, regs->rbx);
3862 kvm_register_write(vcpu, VCPU_REGS_RCX, regs->rcx);
3863 kvm_register_write(vcpu, VCPU_REGS_RDX, regs->rdx);
3864 kvm_register_write(vcpu, VCPU_REGS_RSI, regs->rsi);
3865 kvm_register_write(vcpu, VCPU_REGS_RDI, regs->rdi);
3866 kvm_register_write(vcpu, VCPU_REGS_RSP, regs->rsp);
3867 kvm_register_write(vcpu, VCPU_REGS_RBP, regs->rbp);
3868 #ifdef CONFIG_X86_64
3869 kvm_register_write(vcpu, VCPU_REGS_R8, regs->r8);
3870 kvm_register_write(vcpu, VCPU_REGS_R9, regs->r9);
3871 kvm_register_write(vcpu, VCPU_REGS_R10, regs->r10);
3872 kvm_register_write(vcpu, VCPU_REGS_R11, regs->r11);
3873 kvm_register_write(vcpu, VCPU_REGS_R12, regs->r12);
3874 kvm_register_write(vcpu, VCPU_REGS_R13, regs->r13);
3875 kvm_register_write(vcpu, VCPU_REGS_R14, regs->r14);
3876 kvm_register_write(vcpu, VCPU_REGS_R15, regs->r15);
3880 kvm_rip_write(vcpu, regs->rip);
3881 kvm_x86_ops->set_rflags(vcpu, regs->rflags);
3884 vcpu->arch.exception.pending = false;
3891 void kvm_get_segment(struct kvm_vcpu *vcpu,
3892 struct kvm_segment *var, int seg)
3894 kvm_x86_ops->get_segment(vcpu, var, seg);
3897 void kvm_get_cs_db_l_bits(struct kvm_vcpu *vcpu, int *db, int *l)
3899 struct kvm_segment cs;
3901 kvm_get_segment(vcpu, &cs, VCPU_SREG_CS);
3905 EXPORT_SYMBOL_GPL(kvm_get_cs_db_l_bits);
3907 int kvm_arch_vcpu_ioctl_get_sregs(struct kvm_vcpu *vcpu,
3908 struct kvm_sregs *sregs)
3910 struct descriptor_table dt;
3914 kvm_get_segment(vcpu, &sregs->cs, VCPU_SREG_CS);
3915 kvm_get_segment(vcpu, &sregs->ds, VCPU_SREG_DS);
3916 kvm_get_segment(vcpu, &sregs->es, VCPU_SREG_ES);
3917 kvm_get_segment(vcpu, &sregs->fs, VCPU_SREG_FS);
3918 kvm_get_segment(vcpu, &sregs->gs, VCPU_SREG_GS);
3919 kvm_get_segment(vcpu, &sregs->ss, VCPU_SREG_SS);
3921 kvm_get_segment(vcpu, &sregs->tr, VCPU_SREG_TR);
3922 kvm_get_segment(vcpu, &sregs->ldt, VCPU_SREG_LDTR);
3924 kvm_x86_ops->get_idt(vcpu, &dt);
3925 sregs->idt.limit = dt.limit;
3926 sregs->idt.base = dt.base;
3927 kvm_x86_ops->get_gdt(vcpu, &dt);
3928 sregs->gdt.limit = dt.limit;
3929 sregs->gdt.base = dt.base;
3931 kvm_x86_ops->decache_cr4_guest_bits(vcpu);
3932 sregs->cr0 = vcpu->arch.cr0;
3933 sregs->cr2 = vcpu->arch.cr2;
3934 sregs->cr3 = vcpu->arch.cr3;
3935 sregs->cr4 = vcpu->arch.cr4;
3936 sregs->cr8 = kvm_get_cr8(vcpu);
3937 sregs->efer = vcpu->arch.shadow_efer;
3938 sregs->apic_base = kvm_get_apic_base(vcpu);
3940 memset(sregs->interrupt_bitmap, 0, sizeof sregs->interrupt_bitmap);
3942 if (vcpu->arch.interrupt.pending && !vcpu->arch.interrupt.soft)
3943 set_bit(vcpu->arch.interrupt.nr,
3944 (unsigned long *)sregs->interrupt_bitmap);
3951 int kvm_arch_vcpu_ioctl_get_mpstate(struct kvm_vcpu *vcpu,
3952 struct kvm_mp_state *mp_state)
3955 mp_state->mp_state = vcpu->arch.mp_state;
3960 int kvm_arch_vcpu_ioctl_set_mpstate(struct kvm_vcpu *vcpu,
3961 struct kvm_mp_state *mp_state)
3964 vcpu->arch.mp_state = mp_state->mp_state;
3969 static void kvm_set_segment(struct kvm_vcpu *vcpu,
3970 struct kvm_segment *var, int seg)
3972 kvm_x86_ops->set_segment(vcpu, var, seg);
3975 static void seg_desct_to_kvm_desct(struct desc_struct *seg_desc, u16 selector,
3976 struct kvm_segment *kvm_desct)
3978 kvm_desct->base = get_desc_base(seg_desc);
3979 kvm_desct->limit = get_desc_limit(seg_desc);
3981 kvm_desct->limit <<= 12;
3982 kvm_desct->limit |= 0xfff;
3984 kvm_desct->selector = selector;
3985 kvm_desct->type = seg_desc->type;
3986 kvm_desct->present = seg_desc->p;
3987 kvm_desct->dpl = seg_desc->dpl;
3988 kvm_desct->db = seg_desc->d;
3989 kvm_desct->s = seg_desc->s;
3990 kvm_desct->l = seg_desc->l;
3991 kvm_desct->g = seg_desc->g;
3992 kvm_desct->avl = seg_desc->avl;
3994 kvm_desct->unusable = 1;
3996 kvm_desct->unusable = 0;
3997 kvm_desct->padding = 0;
4000 static void get_segment_descriptor_dtable(struct kvm_vcpu *vcpu,
4002 struct descriptor_table *dtable)
4004 if (selector & 1 << 2) {
4005 struct kvm_segment kvm_seg;
4007 kvm_get_segment(vcpu, &kvm_seg, VCPU_SREG_LDTR);
4009 if (kvm_seg.unusable)
4012 dtable->limit = kvm_seg.limit;
4013 dtable->base = kvm_seg.base;
4016 kvm_x86_ops->get_gdt(vcpu, dtable);
4019 /* allowed just for 8 bytes segments */
4020 static int load_guest_segment_descriptor(struct kvm_vcpu *vcpu, u16 selector,
4021 struct desc_struct *seg_desc)
4023 struct descriptor_table dtable;
4024 u16 index = selector >> 3;
4026 get_segment_descriptor_dtable(vcpu, selector, &dtable);
4028 if (dtable.limit < index * 8 + 7) {
4029 kvm_queue_exception_e(vcpu, GP_VECTOR, selector & 0xfffc);
4032 return kvm_read_guest_virt(dtable.base + index*8, seg_desc, sizeof(*seg_desc), vcpu);
4035 /* allowed just for 8 bytes segments */
4036 static int save_guest_segment_descriptor(struct kvm_vcpu *vcpu, u16 selector,
4037 struct desc_struct *seg_desc)
4039 struct descriptor_table dtable;
4040 u16 index = selector >> 3;
4042 get_segment_descriptor_dtable(vcpu, selector, &dtable);
4044 if (dtable.limit < index * 8 + 7)
4046 return kvm_write_guest_virt(dtable.base + index*8, seg_desc, sizeof(*seg_desc), vcpu);
4049 static gpa_t get_tss_base_addr(struct kvm_vcpu *vcpu,
4050 struct desc_struct *seg_desc)
4052 u32 base_addr = get_desc_base(seg_desc);
4054 return vcpu->arch.mmu.gva_to_gpa(vcpu, base_addr);
4057 static u16 get_segment_selector(struct kvm_vcpu *vcpu, int seg)
4059 struct kvm_segment kvm_seg;
4061 kvm_get_segment(vcpu, &kvm_seg, seg);
4062 return kvm_seg.selector;
4065 static int load_segment_descriptor_to_kvm_desct(struct kvm_vcpu *vcpu,
4067 struct kvm_segment *kvm_seg)
4069 struct desc_struct seg_desc;
4071 if (load_guest_segment_descriptor(vcpu, selector, &seg_desc))
4073 seg_desct_to_kvm_desct(&seg_desc, selector, kvm_seg);
4077 static int kvm_load_realmode_segment(struct kvm_vcpu *vcpu, u16 selector, int seg)
4079 struct kvm_segment segvar = {
4080 .base = selector << 4,
4082 .selector = selector,
4093 kvm_x86_ops->set_segment(vcpu, &segvar, seg);
4097 static int is_vm86_segment(struct kvm_vcpu *vcpu, int seg)
4099 return (seg != VCPU_SREG_LDTR) &&
4100 (seg != VCPU_SREG_TR) &&
4101 (kvm_x86_ops->get_rflags(vcpu) & X86_EFLAGS_VM);
4104 int kvm_load_segment_descriptor(struct kvm_vcpu *vcpu, u16 selector,
4105 int type_bits, int seg)
4107 struct kvm_segment kvm_seg;
4109 if (is_vm86_segment(vcpu, seg) || !(vcpu->arch.cr0 & X86_CR0_PE))
4110 return kvm_load_realmode_segment(vcpu, selector, seg);
4111 if (load_segment_descriptor_to_kvm_desct(vcpu, selector, &kvm_seg))
4113 kvm_seg.type |= type_bits;
4115 if (seg != VCPU_SREG_SS && seg != VCPU_SREG_CS &&
4116 seg != VCPU_SREG_LDTR)
4118 kvm_seg.unusable = 1;
4120 kvm_set_segment(vcpu, &kvm_seg, seg);
4124 static void save_state_to_tss32(struct kvm_vcpu *vcpu,
4125 struct tss_segment_32 *tss)
4127 tss->cr3 = vcpu->arch.cr3;
4128 tss->eip = kvm_rip_read(vcpu);
4129 tss->eflags = kvm_x86_ops->get_rflags(vcpu);
4130 tss->eax = kvm_register_read(vcpu, VCPU_REGS_RAX);
4131 tss->ecx = kvm_register_read(vcpu, VCPU_REGS_RCX);
4132 tss->edx = kvm_register_read(vcpu, VCPU_REGS_RDX);
4133 tss->ebx = kvm_register_read(vcpu, VCPU_REGS_RBX);
4134 tss->esp = kvm_register_read(vcpu, VCPU_REGS_RSP);
4135 tss->ebp = kvm_register_read(vcpu, VCPU_REGS_RBP);
4136 tss->esi = kvm_register_read(vcpu, VCPU_REGS_RSI);
4137 tss->edi = kvm_register_read(vcpu, VCPU_REGS_RDI);
4138 tss->es = get_segment_selector(vcpu, VCPU_SREG_ES);
4139 tss->cs = get_segment_selector(vcpu, VCPU_SREG_CS);
4140 tss->ss = get_segment_selector(vcpu, VCPU_SREG_SS);
4141 tss->ds = get_segment_selector(vcpu, VCPU_SREG_DS);
4142 tss->fs = get_segment_selector(vcpu, VCPU_SREG_FS);
4143 tss->gs = get_segment_selector(vcpu, VCPU_SREG_GS);
4144 tss->ldt_selector = get_segment_selector(vcpu, VCPU_SREG_LDTR);
4147 static int load_state_from_tss32(struct kvm_vcpu *vcpu,
4148 struct tss_segment_32 *tss)
4150 kvm_set_cr3(vcpu, tss->cr3);
4152 kvm_rip_write(vcpu, tss->eip);
4153 kvm_x86_ops->set_rflags(vcpu, tss->eflags | 2);
4155 kvm_register_write(vcpu, VCPU_REGS_RAX, tss->eax);
4156 kvm_register_write(vcpu, VCPU_REGS_RCX, tss->ecx);
4157 kvm_register_write(vcpu, VCPU_REGS_RDX, tss->edx);
4158 kvm_register_write(vcpu, VCPU_REGS_RBX, tss->ebx);
4159 kvm_register_write(vcpu, VCPU_REGS_RSP, tss->esp);
4160 kvm_register_write(vcpu, VCPU_REGS_RBP, tss->ebp);
4161 kvm_register_write(vcpu, VCPU_REGS_RSI, tss->esi);
4162 kvm_register_write(vcpu, VCPU_REGS_RDI, tss->edi);
4164 if (kvm_load_segment_descriptor(vcpu, tss->ldt_selector, 0, VCPU_SREG_LDTR))
4167 if (kvm_load_segment_descriptor(vcpu, tss->es, 1, VCPU_SREG_ES))
4170 if (kvm_load_segment_descriptor(vcpu, tss->cs, 9, VCPU_SREG_CS))
4173 if (kvm_load_segment_descriptor(vcpu, tss->ss, 1, VCPU_SREG_SS))
4176 if (kvm_load_segment_descriptor(vcpu, tss->ds, 1, VCPU_SREG_DS))
4179 if (kvm_load_segment_descriptor(vcpu, tss->fs, 1, VCPU_SREG_FS))
4182 if (kvm_load_segment_descriptor(vcpu, tss->gs, 1, VCPU_SREG_GS))
4187 static void save_state_to_tss16(struct kvm_vcpu *vcpu,
4188 struct tss_segment_16 *tss)
4190 tss->ip = kvm_rip_read(vcpu);
4191 tss->flag = kvm_x86_ops->get_rflags(vcpu);
4192 tss->ax = kvm_register_read(vcpu, VCPU_REGS_RAX);
4193 tss->cx = kvm_register_read(vcpu, VCPU_REGS_RCX);
4194 tss->dx = kvm_register_read(vcpu, VCPU_REGS_RDX);
4195 tss->bx = kvm_register_read(vcpu, VCPU_REGS_RBX);
4196 tss->sp = kvm_register_read(vcpu, VCPU_REGS_RSP);
4197 tss->bp = kvm_register_read(vcpu, VCPU_REGS_RBP);
4198 tss->si = kvm_register_read(vcpu, VCPU_REGS_RSI);
4199 tss->di = kvm_register_read(vcpu, VCPU_REGS_RDI);
4201 tss->es = get_segment_selector(vcpu, VCPU_SREG_ES);
4202 tss->cs = get_segment_selector(vcpu, VCPU_SREG_CS);
4203 tss->ss = get_segment_selector(vcpu, VCPU_SREG_SS);
4204 tss->ds = get_segment_selector(vcpu, VCPU_SREG_DS);
4205 tss->ldt = get_segment_selector(vcpu, VCPU_SREG_LDTR);
4208 static int load_state_from_tss16(struct kvm_vcpu *vcpu,
4209 struct tss_segment_16 *tss)
4211 kvm_rip_write(vcpu, tss->ip);
4212 kvm_x86_ops->set_rflags(vcpu, tss->flag | 2);
4213 kvm_register_write(vcpu, VCPU_REGS_RAX, tss->ax);
4214 kvm_register_write(vcpu, VCPU_REGS_RCX, tss->cx);
4215 kvm_register_write(vcpu, VCPU_REGS_RDX, tss->dx);
4216 kvm_register_write(vcpu, VCPU_REGS_RBX, tss->bx);
4217 kvm_register_write(vcpu, VCPU_REGS_RSP, tss->sp);
4218 kvm_register_write(vcpu, VCPU_REGS_RBP, tss->bp);
4219 kvm_register_write(vcpu, VCPU_REGS_RSI, tss->si);
4220 kvm_register_write(vcpu, VCPU_REGS_RDI, tss->di);
4222 if (kvm_load_segment_descriptor(vcpu, tss->ldt, 0, VCPU_SREG_LDTR))
4225 if (kvm_load_segment_descriptor(vcpu, tss->es, 1, VCPU_SREG_ES))
4228 if (kvm_load_segment_descriptor(vcpu, tss->cs, 9, VCPU_SREG_CS))
4231 if (kvm_load_segment_descriptor(vcpu, tss->ss, 1, VCPU_SREG_SS))
4234 if (kvm_load_segment_descriptor(vcpu, tss->ds, 1, VCPU_SREG_DS))
4239 static int kvm_task_switch_16(struct kvm_vcpu *vcpu, u16 tss_selector,
4240 u16 old_tss_sel, u32 old_tss_base,
4241 struct desc_struct *nseg_desc)
4243 struct tss_segment_16 tss_segment_16;
4246 if (kvm_read_guest(vcpu->kvm, old_tss_base, &tss_segment_16,
4247 sizeof tss_segment_16))
4250 save_state_to_tss16(vcpu, &tss_segment_16);
4252 if (kvm_write_guest(vcpu->kvm, old_tss_base, &tss_segment_16,
4253 sizeof tss_segment_16))
4256 if (kvm_read_guest(vcpu->kvm, get_tss_base_addr(vcpu, nseg_desc),
4257 &tss_segment_16, sizeof tss_segment_16))
4260 if (old_tss_sel != 0xffff) {
4261 tss_segment_16.prev_task_link = old_tss_sel;
4263 if (kvm_write_guest(vcpu->kvm,
4264 get_tss_base_addr(vcpu, nseg_desc),
4265 &tss_segment_16.prev_task_link,
4266 sizeof tss_segment_16.prev_task_link))
4270 if (load_state_from_tss16(vcpu, &tss_segment_16))
4278 static int kvm_task_switch_32(struct kvm_vcpu *vcpu, u16 tss_selector,
4279 u16 old_tss_sel, u32 old_tss_base,
4280 struct desc_struct *nseg_desc)
4282 struct tss_segment_32 tss_segment_32;
4285 if (kvm_read_guest(vcpu->kvm, old_tss_base, &tss_segment_32,
4286 sizeof tss_segment_32))
4289 save_state_to_tss32(vcpu, &tss_segment_32);
4291 if (kvm_write_guest(vcpu->kvm, old_tss_base, &tss_segment_32,
4292 sizeof tss_segment_32))
4295 if (kvm_read_guest(vcpu->kvm, get_tss_base_addr(vcpu, nseg_desc),
4296 &tss_segment_32, sizeof tss_segment_32))
4299 if (old_tss_sel != 0xffff) {
4300 tss_segment_32.prev_task_link = old_tss_sel;
4302 if (kvm_write_guest(vcpu->kvm,
4303 get_tss_base_addr(vcpu, nseg_desc),
4304 &tss_segment_32.prev_task_link,
4305 sizeof tss_segment_32.prev_task_link))
4309 if (load_state_from_tss32(vcpu, &tss_segment_32))
4317 int kvm_task_switch(struct kvm_vcpu *vcpu, u16 tss_selector, int reason)
4319 struct kvm_segment tr_seg;
4320 struct desc_struct cseg_desc;
4321 struct desc_struct nseg_desc;
4323 u32 old_tss_base = get_segment_base(vcpu, VCPU_SREG_TR);
4324 u16 old_tss_sel = get_segment_selector(vcpu, VCPU_SREG_TR);
4326 old_tss_base = vcpu->arch.mmu.gva_to_gpa(vcpu, old_tss_base);
4328 /* FIXME: Handle errors. Failure to read either TSS or their
4329 * descriptors should generate a pagefault.
4331 if (load_guest_segment_descriptor(vcpu, tss_selector, &nseg_desc))
4334 if (load_guest_segment_descriptor(vcpu, old_tss_sel, &cseg_desc))
4337 if (reason != TASK_SWITCH_IRET) {
4340 cpl = kvm_x86_ops->get_cpl(vcpu);
4341 if ((tss_selector & 3) > nseg_desc.dpl || cpl > nseg_desc.dpl) {
4342 kvm_queue_exception_e(vcpu, GP_VECTOR, 0);
4347 if (!nseg_desc.p || get_desc_limit(&nseg_desc) < 0x67) {
4348 kvm_queue_exception_e(vcpu, TS_VECTOR, tss_selector & 0xfffc);
4352 if (reason == TASK_SWITCH_IRET || reason == TASK_SWITCH_JMP) {
4353 cseg_desc.type &= ~(1 << 1); //clear the B flag
4354 save_guest_segment_descriptor(vcpu, old_tss_sel, &cseg_desc);
4357 if (reason == TASK_SWITCH_IRET) {
4358 u32 eflags = kvm_x86_ops->get_rflags(vcpu);
4359 kvm_x86_ops->set_rflags(vcpu, eflags & ~X86_EFLAGS_NT);
4362 /* set back link to prev task only if NT bit is set in eflags
4363 note that old_tss_sel is not used afetr this point */
4364 if (reason != TASK_SWITCH_CALL && reason != TASK_SWITCH_GATE)
4365 old_tss_sel = 0xffff;
4367 /* set back link to prev task only if NT bit is set in eflags
4368 note that old_tss_sel is not used afetr this point */
4369 if (reason != TASK_SWITCH_CALL && reason != TASK_SWITCH_GATE)
4370 old_tss_sel = 0xffff;
4372 if (nseg_desc.type & 8)
4373 ret = kvm_task_switch_32(vcpu, tss_selector, old_tss_sel,
4374 old_tss_base, &nseg_desc);
4376 ret = kvm_task_switch_16(vcpu, tss_selector, old_tss_sel,
4377 old_tss_base, &nseg_desc);
4379 if (reason == TASK_SWITCH_CALL || reason == TASK_SWITCH_GATE) {
4380 u32 eflags = kvm_x86_ops->get_rflags(vcpu);
4381 kvm_x86_ops->set_rflags(vcpu, eflags | X86_EFLAGS_NT);
4384 if (reason != TASK_SWITCH_IRET) {
4385 nseg_desc.type |= (1 << 1);
4386 save_guest_segment_descriptor(vcpu, tss_selector,
4390 kvm_x86_ops->set_cr0(vcpu, vcpu->arch.cr0 | X86_CR0_TS);
4391 seg_desct_to_kvm_desct(&nseg_desc, tss_selector, &tr_seg);
4393 kvm_set_segment(vcpu, &tr_seg, VCPU_SREG_TR);
4397 EXPORT_SYMBOL_GPL(kvm_task_switch);
4399 int kvm_arch_vcpu_ioctl_set_sregs(struct kvm_vcpu *vcpu,
4400 struct kvm_sregs *sregs)
4402 int mmu_reset_needed = 0;
4403 int pending_vec, max_bits;
4404 struct descriptor_table dt;
4408 dt.limit = sregs->idt.limit;
4409 dt.base = sregs->idt.base;
4410 kvm_x86_ops->set_idt(vcpu, &dt);
4411 dt.limit = sregs->gdt.limit;
4412 dt.base = sregs->gdt.base;
4413 kvm_x86_ops->set_gdt(vcpu, &dt);
4415 vcpu->arch.cr2 = sregs->cr2;
4416 mmu_reset_needed |= vcpu->arch.cr3 != sregs->cr3;
4417 vcpu->arch.cr3 = sregs->cr3;
4419 kvm_set_cr8(vcpu, sregs->cr8);
4421 mmu_reset_needed |= vcpu->arch.shadow_efer != sregs->efer;
4422 kvm_x86_ops->set_efer(vcpu, sregs->efer);
4423 kvm_set_apic_base(vcpu, sregs->apic_base);
4425 kvm_x86_ops->decache_cr4_guest_bits(vcpu);
4427 mmu_reset_needed |= vcpu->arch.cr0 != sregs->cr0;
4428 kvm_x86_ops->set_cr0(vcpu, sregs->cr0);
4429 vcpu->arch.cr0 = sregs->cr0;
4431 mmu_reset_needed |= vcpu->arch.cr4 != sregs->cr4;
4432 kvm_x86_ops->set_cr4(vcpu, sregs->cr4);
4433 if (!is_long_mode(vcpu) && is_pae(vcpu))
4434 load_pdptrs(vcpu, vcpu->arch.cr3);
4436 if (mmu_reset_needed)
4437 kvm_mmu_reset_context(vcpu);
4439 max_bits = (sizeof sregs->interrupt_bitmap) << 3;
4440 pending_vec = find_first_bit(
4441 (const unsigned long *)sregs->interrupt_bitmap, max_bits);
4442 if (pending_vec < max_bits) {
4443 kvm_queue_interrupt(vcpu, pending_vec, false);
4444 pr_debug("Set back pending irq %d\n", pending_vec);
4445 if (irqchip_in_kernel(vcpu->kvm))
4446 kvm_pic_clear_isr_ack(vcpu->kvm);
4449 kvm_set_segment(vcpu, &sregs->cs, VCPU_SREG_CS);
4450 kvm_set_segment(vcpu, &sregs->ds, VCPU_SREG_DS);
4451 kvm_set_segment(vcpu, &sregs->es, VCPU_SREG_ES);
4452 kvm_set_segment(vcpu, &sregs->fs, VCPU_SREG_FS);
4453 kvm_set_segment(vcpu, &sregs->gs, VCPU_SREG_GS);
4454 kvm_set_segment(vcpu, &sregs->ss, VCPU_SREG_SS);
4456 kvm_set_segment(vcpu, &sregs->tr, VCPU_SREG_TR);
4457 kvm_set_segment(vcpu, &sregs->ldt, VCPU_SREG_LDTR);
4459 update_cr8_intercept(vcpu);
4461 /* Older userspace won't unhalt the vcpu on reset. */
4462 if (kvm_vcpu_is_bsp(vcpu) && kvm_rip_read(vcpu) == 0xfff0 &&
4463 sregs->cs.selector == 0xf000 && sregs->cs.base == 0xffff0000 &&
4464 !(vcpu->arch.cr0 & X86_CR0_PE))
4465 vcpu->arch.mp_state = KVM_MP_STATE_RUNNABLE;
4472 int kvm_arch_vcpu_ioctl_set_guest_debug(struct kvm_vcpu *vcpu,
4473 struct kvm_guest_debug *dbg)
4475 unsigned long rflags;
4481 old_debug = vcpu->guest_debug;
4483 vcpu->guest_debug = dbg->control;
4484 if (!(vcpu->guest_debug & KVM_GUESTDBG_ENABLE))
4485 vcpu->guest_debug = 0;
4487 if (vcpu->guest_debug & KVM_GUESTDBG_USE_HW_BP) {
4488 for (i = 0; i < KVM_NR_DB_REGS; ++i)
4489 vcpu->arch.eff_db[i] = dbg->arch.debugreg[i];
4490 vcpu->arch.switch_db_regs =
4491 (dbg->arch.debugreg[7] & DR7_BP_EN_MASK);
4493 for (i = 0; i < KVM_NR_DB_REGS; i++)
4494 vcpu->arch.eff_db[i] = vcpu->arch.db[i];
4495 vcpu->arch.switch_db_regs = (vcpu->arch.dr7 & DR7_BP_EN_MASK);
4498 rflags = kvm_x86_ops->get_rflags(vcpu);
4499 if (vcpu->guest_debug & KVM_GUESTDBG_SINGLESTEP)
4500 rflags |= X86_EFLAGS_TF | X86_EFLAGS_RF;
4501 else if (old_debug & KVM_GUESTDBG_SINGLESTEP)
4502 rflags &= ~(X86_EFLAGS_TF | X86_EFLAGS_RF);
4503 kvm_x86_ops->set_rflags(vcpu, rflags);
4505 kvm_x86_ops->set_guest_debug(vcpu, dbg);
4507 if (vcpu->guest_debug & KVM_GUESTDBG_INJECT_DB)
4508 kvm_queue_exception(vcpu, DB_VECTOR);
4509 else if (vcpu->guest_debug & KVM_GUESTDBG_INJECT_BP)
4510 kvm_queue_exception(vcpu, BP_VECTOR);
4518 * fxsave fpu state. Taken from x86_64/processor.h. To be killed when
4519 * we have asm/x86/processor.h
4530 u32 st_space[32]; /* 8*16 bytes for each FP-reg = 128 bytes */
4531 #ifdef CONFIG_X86_64
4532 u32 xmm_space[64]; /* 16*16 bytes for each XMM-reg = 256 bytes */
4534 u32 xmm_space[32]; /* 8*16 bytes for each XMM-reg = 128 bytes */
4539 * Translate a guest virtual address to a guest physical address.
4541 int kvm_arch_vcpu_ioctl_translate(struct kvm_vcpu *vcpu,
4542 struct kvm_translation *tr)
4544 unsigned long vaddr = tr->linear_address;
4548 down_read(&vcpu->kvm->slots_lock);
4549 gpa = vcpu->arch.mmu.gva_to_gpa(vcpu, vaddr);
4550 up_read(&vcpu->kvm->slots_lock);
4551 tr->physical_address = gpa;
4552 tr->valid = gpa != UNMAPPED_GVA;
4560 int kvm_arch_vcpu_ioctl_get_fpu(struct kvm_vcpu *vcpu, struct kvm_fpu *fpu)
4562 struct fxsave *fxsave = (struct fxsave *)&vcpu->arch.guest_fx_image;
4566 memcpy(fpu->fpr, fxsave->st_space, 128);
4567 fpu->fcw = fxsave->cwd;
4568 fpu->fsw = fxsave->swd;
4569 fpu->ftwx = fxsave->twd;
4570 fpu->last_opcode = fxsave->fop;
4571 fpu->last_ip = fxsave->rip;
4572 fpu->last_dp = fxsave->rdp;
4573 memcpy(fpu->xmm, fxsave->xmm_space, sizeof fxsave->xmm_space);
4580 int kvm_arch_vcpu_ioctl_set_fpu(struct kvm_vcpu *vcpu, struct kvm_fpu *fpu)
4582 struct fxsave *fxsave = (struct fxsave *)&vcpu->arch.guest_fx_image;
4586 memcpy(fxsave->st_space, fpu->fpr, 128);
4587 fxsave->cwd = fpu->fcw;
4588 fxsave->swd = fpu->fsw;
4589 fxsave->twd = fpu->ftwx;
4590 fxsave->fop = fpu->last_opcode;
4591 fxsave->rip = fpu->last_ip;
4592 fxsave->rdp = fpu->last_dp;
4593 memcpy(fxsave->xmm_space, fpu->xmm, sizeof fxsave->xmm_space);
4600 void fx_init(struct kvm_vcpu *vcpu)
4602 unsigned after_mxcsr_mask;
4605 * Touch the fpu the first time in non atomic context as if
4606 * this is the first fpu instruction the exception handler
4607 * will fire before the instruction returns and it'll have to
4608 * allocate ram with GFP_KERNEL.
4611 kvm_fx_save(&vcpu->arch.host_fx_image);
4613 /* Initialize guest FPU by resetting ours and saving into guest's */
4615 kvm_fx_save(&vcpu->arch.host_fx_image);
4617 kvm_fx_save(&vcpu->arch.guest_fx_image);
4618 kvm_fx_restore(&vcpu->arch.host_fx_image);
4621 vcpu->arch.cr0 |= X86_CR0_ET;
4622 after_mxcsr_mask = offsetof(struct i387_fxsave_struct, st_space);
4623 vcpu->arch.guest_fx_image.mxcsr = 0x1f80;
4624 memset((void *)&vcpu->arch.guest_fx_image + after_mxcsr_mask,
4625 0, sizeof(struct i387_fxsave_struct) - after_mxcsr_mask);
4627 EXPORT_SYMBOL_GPL(fx_init);
4629 void kvm_load_guest_fpu(struct kvm_vcpu *vcpu)
4631 if (!vcpu->fpu_active || vcpu->guest_fpu_loaded)
4634 vcpu->guest_fpu_loaded = 1;
4635 kvm_fx_save(&vcpu->arch.host_fx_image);
4636 kvm_fx_restore(&vcpu->arch.guest_fx_image);
4638 EXPORT_SYMBOL_GPL(kvm_load_guest_fpu);
4640 void kvm_put_guest_fpu(struct kvm_vcpu *vcpu)
4642 if (!vcpu->guest_fpu_loaded)
4645 vcpu->guest_fpu_loaded = 0;
4646 kvm_fx_save(&vcpu->arch.guest_fx_image);
4647 kvm_fx_restore(&vcpu->arch.host_fx_image);
4648 ++vcpu->stat.fpu_reload;
4650 EXPORT_SYMBOL_GPL(kvm_put_guest_fpu);
4652 void kvm_arch_vcpu_free(struct kvm_vcpu *vcpu)
4654 if (vcpu->arch.time_page) {
4655 kvm_release_page_dirty(vcpu->arch.time_page);
4656 vcpu->arch.time_page = NULL;
4659 kvm_x86_ops->vcpu_free(vcpu);
4662 struct kvm_vcpu *kvm_arch_vcpu_create(struct kvm *kvm,
4665 return kvm_x86_ops->vcpu_create(kvm, id);
4668 int kvm_arch_vcpu_setup(struct kvm_vcpu *vcpu)
4672 /* We do fxsave: this must be aligned. */
4673 BUG_ON((unsigned long)&vcpu->arch.host_fx_image & 0xF);
4675 vcpu->arch.mtrr_state.have_fixed = 1;
4677 r = kvm_arch_vcpu_reset(vcpu);
4679 r = kvm_mmu_setup(vcpu);
4686 kvm_x86_ops->vcpu_free(vcpu);
4690 void kvm_arch_vcpu_destroy(struct kvm_vcpu *vcpu)
4693 kvm_mmu_unload(vcpu);
4696 kvm_x86_ops->vcpu_free(vcpu);
4699 int kvm_arch_vcpu_reset(struct kvm_vcpu *vcpu)
4701 vcpu->arch.nmi_pending = false;
4702 vcpu->arch.nmi_injected = false;
4704 vcpu->arch.switch_db_regs = 0;
4705 memset(vcpu->arch.db, 0, sizeof(vcpu->arch.db));
4706 vcpu->arch.dr6 = DR6_FIXED_1;
4707 vcpu->arch.dr7 = DR7_FIXED_1;
4709 return kvm_x86_ops->vcpu_reset(vcpu);
4712 int kvm_arch_hardware_enable(void *garbage)
4715 * Since this may be called from a hotplug notifcation,
4716 * we can't get the CPU frequency directly.
4718 if (!boot_cpu_has(X86_FEATURE_CONSTANT_TSC)) {
4719 int cpu = raw_smp_processor_id();
4720 per_cpu(cpu_tsc_khz, cpu) = 0;
4722 return kvm_x86_ops->hardware_enable(garbage);
4725 void kvm_arch_hardware_disable(void *garbage)
4727 kvm_x86_ops->hardware_disable(garbage);
4730 int kvm_arch_hardware_setup(void)
4732 return kvm_x86_ops->hardware_setup();
4735 void kvm_arch_hardware_unsetup(void)
4737 kvm_x86_ops->hardware_unsetup();
4740 void kvm_arch_check_processor_compat(void *rtn)
4742 kvm_x86_ops->check_processor_compatibility(rtn);
4745 int kvm_arch_vcpu_init(struct kvm_vcpu *vcpu)
4751 BUG_ON(vcpu->kvm == NULL);
4754 vcpu->arch.mmu.root_hpa = INVALID_PAGE;
4755 if (!irqchip_in_kernel(kvm) || kvm_vcpu_is_bsp(vcpu))
4756 vcpu->arch.mp_state = KVM_MP_STATE_RUNNABLE;
4758 vcpu->arch.mp_state = KVM_MP_STATE_UNINITIALIZED;
4760 page = alloc_page(GFP_KERNEL | __GFP_ZERO);
4765 vcpu->arch.pio_data = page_address(page);
4767 r = kvm_mmu_create(vcpu);
4769 goto fail_free_pio_data;
4771 if (irqchip_in_kernel(kvm)) {
4772 r = kvm_create_lapic(vcpu);
4774 goto fail_mmu_destroy;
4777 vcpu->arch.mce_banks = kzalloc(KVM_MAX_MCE_BANKS * sizeof(u64) * 4,
4779 if (!vcpu->arch.mce_banks) {
4781 goto fail_mmu_destroy;
4783 vcpu->arch.mcg_cap = KVM_MAX_MCE_BANKS;
4788 kvm_mmu_destroy(vcpu);
4790 free_page((unsigned long)vcpu->arch.pio_data);
4795 void kvm_arch_vcpu_uninit(struct kvm_vcpu *vcpu)
4797 kvm_free_lapic(vcpu);
4798 down_read(&vcpu->kvm->slots_lock);
4799 kvm_mmu_destroy(vcpu);
4800 up_read(&vcpu->kvm->slots_lock);
4801 free_page((unsigned long)vcpu->arch.pio_data);
4804 struct kvm *kvm_arch_create_vm(void)
4806 struct kvm *kvm = kzalloc(sizeof(struct kvm), GFP_KERNEL);
4809 return ERR_PTR(-ENOMEM);
4811 INIT_LIST_HEAD(&kvm->arch.active_mmu_pages);
4812 INIT_LIST_HEAD(&kvm->arch.assigned_dev_head);
4814 /* Reserve bit 0 of irq_sources_bitmap for userspace irq source */
4815 set_bit(KVM_USERSPACE_IRQ_SOURCE_ID, &kvm->arch.irq_sources_bitmap);
4817 rdtscll(kvm->arch.vm_init_tsc);
4822 static void kvm_unload_vcpu_mmu(struct kvm_vcpu *vcpu)
4825 kvm_mmu_unload(vcpu);
4829 static void kvm_free_vcpus(struct kvm *kvm)
4832 struct kvm_vcpu *vcpu;
4835 * Unpin any mmu pages first.
4837 kvm_for_each_vcpu(i, vcpu, kvm)
4838 kvm_unload_vcpu_mmu(vcpu);
4839 kvm_for_each_vcpu(i, vcpu, kvm)
4840 kvm_arch_vcpu_free(vcpu);
4842 mutex_lock(&kvm->lock);
4843 for (i = 0; i < atomic_read(&kvm->online_vcpus); i++)
4844 kvm->vcpus[i] = NULL;
4846 atomic_set(&kvm->online_vcpus, 0);
4847 mutex_unlock(&kvm->lock);
4850 void kvm_arch_sync_events(struct kvm *kvm)
4852 kvm_free_all_assigned_devices(kvm);
4855 void kvm_arch_destroy_vm(struct kvm *kvm)
4857 kvm_iommu_unmap_guest(kvm);
4859 kfree(kvm->arch.vpic);
4860 kfree(kvm->arch.vioapic);
4861 kvm_free_vcpus(kvm);
4862 kvm_free_physmem(kvm);
4863 if (kvm->arch.apic_access_page)
4864 put_page(kvm->arch.apic_access_page);
4865 if (kvm->arch.ept_identity_pagetable)
4866 put_page(kvm->arch.ept_identity_pagetable);
4870 int kvm_arch_set_memory_region(struct kvm *kvm,
4871 struct kvm_userspace_memory_region *mem,
4872 struct kvm_memory_slot old,
4875 int npages = mem->memory_size >> PAGE_SHIFT;
4876 struct kvm_memory_slot *memslot = &kvm->memslots[mem->slot];
4878 /*To keep backward compatibility with older userspace,
4879 *x86 needs to hanlde !user_alloc case.
4882 if (npages && !old.rmap) {
4883 unsigned long userspace_addr;
4885 down_write(¤t->mm->mmap_sem);
4886 userspace_addr = do_mmap(NULL, 0,
4888 PROT_READ | PROT_WRITE,
4889 MAP_PRIVATE | MAP_ANONYMOUS,
4891 up_write(¤t->mm->mmap_sem);
4893 if (IS_ERR((void *)userspace_addr))
4894 return PTR_ERR((void *)userspace_addr);
4896 /* set userspace_addr atomically for kvm_hva_to_rmapp */
4897 spin_lock(&kvm->mmu_lock);
4898 memslot->userspace_addr = userspace_addr;
4899 spin_unlock(&kvm->mmu_lock);
4901 if (!old.user_alloc && old.rmap) {
4904 down_write(¤t->mm->mmap_sem);
4905 ret = do_munmap(current->mm, old.userspace_addr,
4906 old.npages * PAGE_SIZE);
4907 up_write(¤t->mm->mmap_sem);
4910 "kvm_vm_ioctl_set_memory_region: "
4911 "failed to munmap memory\n");
4916 spin_lock(&kvm->mmu_lock);
4917 if (!kvm->arch.n_requested_mmu_pages) {
4918 unsigned int nr_mmu_pages = kvm_mmu_calculate_mmu_pages(kvm);
4919 kvm_mmu_change_mmu_pages(kvm, nr_mmu_pages);
4922 kvm_mmu_slot_remove_write_access(kvm, mem->slot);
4923 spin_unlock(&kvm->mmu_lock);
4928 void kvm_arch_flush_shadow(struct kvm *kvm)
4930 kvm_mmu_zap_all(kvm);
4931 kvm_reload_remote_mmus(kvm);
4934 int kvm_arch_vcpu_runnable(struct kvm_vcpu *vcpu)
4936 return vcpu->arch.mp_state == KVM_MP_STATE_RUNNABLE
4937 || vcpu->arch.mp_state == KVM_MP_STATE_SIPI_RECEIVED
4938 || vcpu->arch.nmi_pending ||
4939 (kvm_arch_interrupt_allowed(vcpu) &&
4940 kvm_cpu_has_interrupt(vcpu));
4943 void kvm_vcpu_kick(struct kvm_vcpu *vcpu)
4946 int cpu = vcpu->cpu;
4948 if (waitqueue_active(&vcpu->wq)) {
4949 wake_up_interruptible(&vcpu->wq);
4950 ++vcpu->stat.halt_wakeup;
4954 if (cpu != me && (unsigned)cpu < nr_cpu_ids && cpu_online(cpu))
4955 if (!test_and_set_bit(KVM_REQ_KICK, &vcpu->requests))
4956 smp_send_reschedule(cpu);
4960 int kvm_arch_interrupt_allowed(struct kvm_vcpu *vcpu)
4962 return kvm_x86_ops->interrupt_allowed(vcpu);
4965 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_exit);
4966 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_inj_virq);
4967 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_page_fault);
4968 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_msr);
4969 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_cr);