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. This capabilities test skips MSRs that are
488 * kvm-specific. Those are put in the beginning of the list.
491 #define KVM_SAVE_MSRS_BEGIN 2
492 static u32 msrs_to_save[] = {
493 MSR_KVM_SYSTEM_TIME, MSR_KVM_WALL_CLOCK,
494 MSR_IA32_SYSENTER_CS, MSR_IA32_SYSENTER_ESP, MSR_IA32_SYSENTER_EIP,
497 MSR_CSTAR, MSR_KERNEL_GS_BASE, MSR_SYSCALL_MASK, MSR_LSTAR,
499 MSR_IA32_TSC, MSR_IA32_PERF_STATUS, MSR_IA32_CR_PAT, MSR_VM_HSAVE_PA
502 static unsigned num_msrs_to_save;
504 static u32 emulated_msrs[] = {
505 MSR_IA32_MISC_ENABLE,
508 static void set_efer(struct kvm_vcpu *vcpu, u64 efer)
510 if (efer & efer_reserved_bits) {
511 printk(KERN_DEBUG "set_efer: 0x%llx #GP, reserved bits\n",
513 kvm_inject_gp(vcpu, 0);
518 && (vcpu->arch.shadow_efer & EFER_LME) != (efer & EFER_LME)) {
519 printk(KERN_DEBUG "set_efer: #GP, change LME while paging\n");
520 kvm_inject_gp(vcpu, 0);
524 if (efer & EFER_FFXSR) {
525 struct kvm_cpuid_entry2 *feat;
527 feat = kvm_find_cpuid_entry(vcpu, 0x80000001, 0);
528 if (!feat || !(feat->edx & bit(X86_FEATURE_FXSR_OPT))) {
529 printk(KERN_DEBUG "set_efer: #GP, enable FFXSR w/o CPUID capability\n");
530 kvm_inject_gp(vcpu, 0);
535 if (efer & EFER_SVME) {
536 struct kvm_cpuid_entry2 *feat;
538 feat = kvm_find_cpuid_entry(vcpu, 0x80000001, 0);
539 if (!feat || !(feat->ecx & bit(X86_FEATURE_SVM))) {
540 printk(KERN_DEBUG "set_efer: #GP, enable SVM w/o SVM\n");
541 kvm_inject_gp(vcpu, 0);
546 kvm_x86_ops->set_efer(vcpu, efer);
549 efer |= vcpu->arch.shadow_efer & EFER_LMA;
551 vcpu->arch.shadow_efer = efer;
553 vcpu->arch.mmu.base_role.nxe = (efer & EFER_NX) && !tdp_enabled;
554 kvm_mmu_reset_context(vcpu);
557 void kvm_enable_efer_bits(u64 mask)
559 efer_reserved_bits &= ~mask;
561 EXPORT_SYMBOL_GPL(kvm_enable_efer_bits);
565 * Writes msr value into into the appropriate "register".
566 * Returns 0 on success, non-0 otherwise.
567 * Assumes vcpu_load() was already called.
569 int kvm_set_msr(struct kvm_vcpu *vcpu, u32 msr_index, u64 data)
571 return kvm_x86_ops->set_msr(vcpu, msr_index, data);
575 * Adapt set_msr() to msr_io()'s calling convention
577 static int do_set_msr(struct kvm_vcpu *vcpu, unsigned index, u64 *data)
579 return kvm_set_msr(vcpu, index, *data);
582 static void kvm_write_wall_clock(struct kvm *kvm, gpa_t wall_clock)
585 struct pvclock_wall_clock wc;
586 struct timespec boot;
593 kvm_write_guest(kvm, wall_clock, &version, sizeof(version));
596 * The guest calculates current wall clock time by adding
597 * system time (updated by kvm_write_guest_time below) to the
598 * wall clock specified here. guest system time equals host
599 * system time for us, thus we must fill in host boot time here.
603 wc.sec = boot.tv_sec;
604 wc.nsec = boot.tv_nsec;
605 wc.version = version;
607 kvm_write_guest(kvm, wall_clock, &wc, sizeof(wc));
610 kvm_write_guest(kvm, wall_clock, &version, sizeof(version));
613 static uint32_t div_frac(uint32_t dividend, uint32_t divisor)
615 uint32_t quotient, remainder;
617 /* Don't try to replace with do_div(), this one calculates
618 * "(dividend << 32) / divisor" */
620 : "=a" (quotient), "=d" (remainder)
621 : "0" (0), "1" (dividend), "r" (divisor) );
625 static void kvm_set_time_scale(uint32_t tsc_khz, struct pvclock_vcpu_time_info *hv_clock)
627 uint64_t nsecs = 1000000000LL;
632 tps64 = tsc_khz * 1000LL;
633 while (tps64 > nsecs*2) {
638 tps32 = (uint32_t)tps64;
639 while (tps32 <= (uint32_t)nsecs) {
644 hv_clock->tsc_shift = shift;
645 hv_clock->tsc_to_system_mul = div_frac(nsecs, tps32);
647 pr_debug("%s: tsc_khz %u, tsc_shift %d, tsc_mul %u\n",
648 __func__, tsc_khz, hv_clock->tsc_shift,
649 hv_clock->tsc_to_system_mul);
652 static DEFINE_PER_CPU(unsigned long, cpu_tsc_khz);
654 static void kvm_write_guest_time(struct kvm_vcpu *v)
658 struct kvm_vcpu_arch *vcpu = &v->arch;
660 unsigned long this_tsc_khz;
662 if ((!vcpu->time_page))
665 this_tsc_khz = get_cpu_var(cpu_tsc_khz);
666 if (unlikely(vcpu->hv_clock_tsc_khz != this_tsc_khz)) {
667 kvm_set_time_scale(this_tsc_khz, &vcpu->hv_clock);
668 vcpu->hv_clock_tsc_khz = this_tsc_khz;
670 put_cpu_var(cpu_tsc_khz);
672 /* Keep irq disabled to prevent changes to the clock */
673 local_irq_save(flags);
674 kvm_get_msr(v, MSR_IA32_TSC, &vcpu->hv_clock.tsc_timestamp);
676 monotonic_to_bootbased(&ts);
677 local_irq_restore(flags);
679 /* With all the info we got, fill in the values */
681 vcpu->hv_clock.system_time = ts.tv_nsec +
682 (NSEC_PER_SEC * (u64)ts.tv_sec) + v->kvm->arch.kvmclock_offset;
685 * The interface expects us to write an even number signaling that the
686 * update is finished. Since the guest won't see the intermediate
687 * state, we just increase by 2 at the end.
689 vcpu->hv_clock.version += 2;
691 shared_kaddr = kmap_atomic(vcpu->time_page, KM_USER0);
693 memcpy(shared_kaddr + vcpu->time_offset, &vcpu->hv_clock,
694 sizeof(vcpu->hv_clock));
696 kunmap_atomic(shared_kaddr, KM_USER0);
698 mark_page_dirty(v->kvm, vcpu->time >> PAGE_SHIFT);
701 static int kvm_request_guest_time_update(struct kvm_vcpu *v)
703 struct kvm_vcpu_arch *vcpu = &v->arch;
705 if (!vcpu->time_page)
707 set_bit(KVM_REQ_KVMCLOCK_UPDATE, &v->requests);
711 static bool msr_mtrr_valid(unsigned msr)
714 case 0x200 ... 0x200 + 2 * KVM_NR_VAR_MTRR - 1:
715 case MSR_MTRRfix64K_00000:
716 case MSR_MTRRfix16K_80000:
717 case MSR_MTRRfix16K_A0000:
718 case MSR_MTRRfix4K_C0000:
719 case MSR_MTRRfix4K_C8000:
720 case MSR_MTRRfix4K_D0000:
721 case MSR_MTRRfix4K_D8000:
722 case MSR_MTRRfix4K_E0000:
723 case MSR_MTRRfix4K_E8000:
724 case MSR_MTRRfix4K_F0000:
725 case MSR_MTRRfix4K_F8000:
726 case MSR_MTRRdefType:
727 case MSR_IA32_CR_PAT:
735 static bool valid_pat_type(unsigned t)
737 return t < 8 && (1 << t) & 0xf3; /* 0, 1, 4, 5, 6, 7 */
740 static bool valid_mtrr_type(unsigned t)
742 return t < 8 && (1 << t) & 0x73; /* 0, 1, 4, 5, 6 */
745 static bool mtrr_valid(struct kvm_vcpu *vcpu, u32 msr, u64 data)
749 if (!msr_mtrr_valid(msr))
752 if (msr == MSR_IA32_CR_PAT) {
753 for (i = 0; i < 8; i++)
754 if (!valid_pat_type((data >> (i * 8)) & 0xff))
757 } else if (msr == MSR_MTRRdefType) {
760 return valid_mtrr_type(data & 0xff);
761 } else if (msr >= MSR_MTRRfix64K_00000 && msr <= MSR_MTRRfix4K_F8000) {
762 for (i = 0; i < 8 ; i++)
763 if (!valid_mtrr_type((data >> (i * 8)) & 0xff))
769 return valid_mtrr_type(data & 0xff);
772 static int set_msr_mtrr(struct kvm_vcpu *vcpu, u32 msr, u64 data)
774 u64 *p = (u64 *)&vcpu->arch.mtrr_state.fixed_ranges;
776 if (!mtrr_valid(vcpu, msr, data))
779 if (msr == MSR_MTRRdefType) {
780 vcpu->arch.mtrr_state.def_type = data;
781 vcpu->arch.mtrr_state.enabled = (data & 0xc00) >> 10;
782 } else if (msr == MSR_MTRRfix64K_00000)
784 else if (msr == MSR_MTRRfix16K_80000 || msr == MSR_MTRRfix16K_A0000)
785 p[1 + msr - MSR_MTRRfix16K_80000] = data;
786 else if (msr >= MSR_MTRRfix4K_C0000 && msr <= MSR_MTRRfix4K_F8000)
787 p[3 + msr - MSR_MTRRfix4K_C0000] = data;
788 else if (msr == MSR_IA32_CR_PAT)
789 vcpu->arch.pat = data;
790 else { /* Variable MTRRs */
791 int idx, is_mtrr_mask;
794 idx = (msr - 0x200) / 2;
795 is_mtrr_mask = msr - 0x200 - 2 * idx;
798 (u64 *)&vcpu->arch.mtrr_state.var_ranges[idx].base_lo;
801 (u64 *)&vcpu->arch.mtrr_state.var_ranges[idx].mask_lo;
805 kvm_mmu_reset_context(vcpu);
809 static int set_msr_mce(struct kvm_vcpu *vcpu, u32 msr, u64 data)
811 u64 mcg_cap = vcpu->arch.mcg_cap;
812 unsigned bank_num = mcg_cap & 0xff;
815 case MSR_IA32_MCG_STATUS:
816 vcpu->arch.mcg_status = data;
818 case MSR_IA32_MCG_CTL:
819 if (!(mcg_cap & MCG_CTL_P))
821 if (data != 0 && data != ~(u64)0)
823 vcpu->arch.mcg_ctl = data;
826 if (msr >= MSR_IA32_MC0_CTL &&
827 msr < MSR_IA32_MC0_CTL + 4 * bank_num) {
828 u32 offset = msr - MSR_IA32_MC0_CTL;
829 /* only 0 or all 1s can be written to IA32_MCi_CTL */
830 if ((offset & 0x3) == 0 &&
831 data != 0 && data != ~(u64)0)
833 vcpu->arch.mce_banks[offset] = data;
841 int kvm_set_msr_common(struct kvm_vcpu *vcpu, u32 msr, u64 data)
845 set_efer(vcpu, data);
848 data &= ~(u64)0x40; /* ignore flush filter disable */
850 pr_unimpl(vcpu, "unimplemented HWCR wrmsr: 0x%llx\n",
855 case MSR_FAM10H_MMIO_CONF_BASE:
857 pr_unimpl(vcpu, "unimplemented MMIO_CONF_BASE wrmsr: "
862 case MSR_AMD64_NB_CFG:
864 case MSR_IA32_DEBUGCTLMSR:
866 /* We support the non-activated case already */
868 } else if (data & ~(DEBUGCTLMSR_LBR | DEBUGCTLMSR_BTF)) {
869 /* Values other than LBR and BTF are vendor-specific,
870 thus reserved and should throw a #GP */
873 pr_unimpl(vcpu, "%s: MSR_IA32_DEBUGCTLMSR 0x%llx, nop\n",
876 case MSR_IA32_UCODE_REV:
877 case MSR_IA32_UCODE_WRITE:
878 case MSR_VM_HSAVE_PA:
879 case MSR_AMD64_PATCH_LOADER:
881 case 0x200 ... 0x2ff:
882 return set_msr_mtrr(vcpu, msr, data);
883 case MSR_IA32_APICBASE:
884 kvm_set_apic_base(vcpu, data);
886 case APIC_BASE_MSR ... APIC_BASE_MSR + 0x3ff:
887 return kvm_x2apic_msr_write(vcpu, msr, data);
888 case MSR_IA32_MISC_ENABLE:
889 vcpu->arch.ia32_misc_enable_msr = data;
891 case MSR_KVM_WALL_CLOCK:
892 vcpu->kvm->arch.wall_clock = data;
893 kvm_write_wall_clock(vcpu->kvm, data);
895 case MSR_KVM_SYSTEM_TIME: {
896 if (vcpu->arch.time_page) {
897 kvm_release_page_dirty(vcpu->arch.time_page);
898 vcpu->arch.time_page = NULL;
901 vcpu->arch.time = data;
903 /* we verify if the enable bit is set... */
907 /* ...but clean it before doing the actual write */
908 vcpu->arch.time_offset = data & ~(PAGE_MASK | 1);
910 vcpu->arch.time_page =
911 gfn_to_page(vcpu->kvm, data >> PAGE_SHIFT);
913 if (is_error_page(vcpu->arch.time_page)) {
914 kvm_release_page_clean(vcpu->arch.time_page);
915 vcpu->arch.time_page = NULL;
918 kvm_request_guest_time_update(vcpu);
921 case MSR_IA32_MCG_CTL:
922 case MSR_IA32_MCG_STATUS:
923 case MSR_IA32_MC0_CTL ... MSR_IA32_MC0_CTL + 4 * KVM_MAX_MCE_BANKS - 1:
924 return set_msr_mce(vcpu, msr, data);
926 /* Performance counters are not protected by a CPUID bit,
927 * so we should check all of them in the generic path for the sake of
928 * cross vendor migration.
929 * Writing a zero into the event select MSRs disables them,
930 * which we perfectly emulate ;-). Any other value should be at least
931 * reported, some guests depend on them.
933 case MSR_P6_EVNTSEL0:
934 case MSR_P6_EVNTSEL1:
935 case MSR_K7_EVNTSEL0:
936 case MSR_K7_EVNTSEL1:
937 case MSR_K7_EVNTSEL2:
938 case MSR_K7_EVNTSEL3:
940 pr_unimpl(vcpu, "unimplemented perfctr wrmsr: "
941 "0x%x data 0x%llx\n", msr, data);
943 /* at least RHEL 4 unconditionally writes to the perfctr registers,
944 * so we ignore writes to make it happy.
946 case MSR_P6_PERFCTR0:
947 case MSR_P6_PERFCTR1:
948 case MSR_K7_PERFCTR0:
949 case MSR_K7_PERFCTR1:
950 case MSR_K7_PERFCTR2:
951 case MSR_K7_PERFCTR3:
952 pr_unimpl(vcpu, "unimplemented perfctr wrmsr: "
953 "0x%x data 0x%llx\n", msr, data);
957 pr_unimpl(vcpu, "unhandled wrmsr: 0x%x data %llx\n",
961 pr_unimpl(vcpu, "ignored wrmsr: 0x%x data %llx\n",
968 EXPORT_SYMBOL_GPL(kvm_set_msr_common);
972 * Reads an msr value (of 'msr_index') into 'pdata'.
973 * Returns 0 on success, non-0 otherwise.
974 * Assumes vcpu_load() was already called.
976 int kvm_get_msr(struct kvm_vcpu *vcpu, u32 msr_index, u64 *pdata)
978 return kvm_x86_ops->get_msr(vcpu, msr_index, pdata);
981 static int get_msr_mtrr(struct kvm_vcpu *vcpu, u32 msr, u64 *pdata)
983 u64 *p = (u64 *)&vcpu->arch.mtrr_state.fixed_ranges;
985 if (!msr_mtrr_valid(msr))
988 if (msr == MSR_MTRRdefType)
989 *pdata = vcpu->arch.mtrr_state.def_type +
990 (vcpu->arch.mtrr_state.enabled << 10);
991 else if (msr == MSR_MTRRfix64K_00000)
993 else if (msr == MSR_MTRRfix16K_80000 || msr == MSR_MTRRfix16K_A0000)
994 *pdata = p[1 + msr - MSR_MTRRfix16K_80000];
995 else if (msr >= MSR_MTRRfix4K_C0000 && msr <= MSR_MTRRfix4K_F8000)
996 *pdata = p[3 + msr - MSR_MTRRfix4K_C0000];
997 else if (msr == MSR_IA32_CR_PAT)
998 *pdata = vcpu->arch.pat;
999 else { /* Variable MTRRs */
1000 int idx, is_mtrr_mask;
1003 idx = (msr - 0x200) / 2;
1004 is_mtrr_mask = msr - 0x200 - 2 * idx;
1007 (u64 *)&vcpu->arch.mtrr_state.var_ranges[idx].base_lo;
1010 (u64 *)&vcpu->arch.mtrr_state.var_ranges[idx].mask_lo;
1017 static int get_msr_mce(struct kvm_vcpu *vcpu, u32 msr, u64 *pdata)
1020 u64 mcg_cap = vcpu->arch.mcg_cap;
1021 unsigned bank_num = mcg_cap & 0xff;
1024 case MSR_IA32_P5_MC_ADDR:
1025 case MSR_IA32_P5_MC_TYPE:
1028 case MSR_IA32_MCG_CAP:
1029 data = vcpu->arch.mcg_cap;
1031 case MSR_IA32_MCG_CTL:
1032 if (!(mcg_cap & MCG_CTL_P))
1034 data = vcpu->arch.mcg_ctl;
1036 case MSR_IA32_MCG_STATUS:
1037 data = vcpu->arch.mcg_status;
1040 if (msr >= MSR_IA32_MC0_CTL &&
1041 msr < MSR_IA32_MC0_CTL + 4 * bank_num) {
1042 u32 offset = msr - MSR_IA32_MC0_CTL;
1043 data = vcpu->arch.mce_banks[offset];
1052 int kvm_get_msr_common(struct kvm_vcpu *vcpu, u32 msr, u64 *pdata)
1057 case MSR_IA32_PLATFORM_ID:
1058 case MSR_IA32_UCODE_REV:
1059 case MSR_IA32_EBL_CR_POWERON:
1060 case MSR_IA32_DEBUGCTLMSR:
1061 case MSR_IA32_LASTBRANCHFROMIP:
1062 case MSR_IA32_LASTBRANCHTOIP:
1063 case MSR_IA32_LASTINTFROMIP:
1064 case MSR_IA32_LASTINTTOIP:
1067 case MSR_VM_HSAVE_PA:
1068 case MSR_P6_PERFCTR0:
1069 case MSR_P6_PERFCTR1:
1070 case MSR_P6_EVNTSEL0:
1071 case MSR_P6_EVNTSEL1:
1072 case MSR_K7_EVNTSEL0:
1073 case MSR_K7_PERFCTR0:
1074 case MSR_K8_INT_PENDING_MSG:
1075 case MSR_AMD64_NB_CFG:
1076 case MSR_FAM10H_MMIO_CONF_BASE:
1080 data = 0x500 | KVM_NR_VAR_MTRR;
1082 case 0x200 ... 0x2ff:
1083 return get_msr_mtrr(vcpu, msr, pdata);
1084 case 0xcd: /* fsb frequency */
1087 case MSR_IA32_APICBASE:
1088 data = kvm_get_apic_base(vcpu);
1090 case APIC_BASE_MSR ... APIC_BASE_MSR + 0x3ff:
1091 return kvm_x2apic_msr_read(vcpu, msr, pdata);
1093 case MSR_IA32_MISC_ENABLE:
1094 data = vcpu->arch.ia32_misc_enable_msr;
1096 case MSR_IA32_PERF_STATUS:
1097 /* TSC increment by tick */
1099 /* CPU multiplier */
1100 data |= (((uint64_t)4ULL) << 40);
1103 data = vcpu->arch.shadow_efer;
1105 case MSR_KVM_WALL_CLOCK:
1106 data = vcpu->kvm->arch.wall_clock;
1108 case MSR_KVM_SYSTEM_TIME:
1109 data = vcpu->arch.time;
1111 case MSR_IA32_P5_MC_ADDR:
1112 case MSR_IA32_P5_MC_TYPE:
1113 case MSR_IA32_MCG_CAP:
1114 case MSR_IA32_MCG_CTL:
1115 case MSR_IA32_MCG_STATUS:
1116 case MSR_IA32_MC0_CTL ... MSR_IA32_MC0_CTL + 4 * KVM_MAX_MCE_BANKS - 1:
1117 return get_msr_mce(vcpu, msr, pdata);
1120 pr_unimpl(vcpu, "unhandled rdmsr: 0x%x\n", msr);
1123 pr_unimpl(vcpu, "ignored rdmsr: 0x%x\n", msr);
1131 EXPORT_SYMBOL_GPL(kvm_get_msr_common);
1134 * Read or write a bunch of msrs. All parameters are kernel addresses.
1136 * @return number of msrs set successfully.
1138 static int __msr_io(struct kvm_vcpu *vcpu, struct kvm_msrs *msrs,
1139 struct kvm_msr_entry *entries,
1140 int (*do_msr)(struct kvm_vcpu *vcpu,
1141 unsigned index, u64 *data))
1147 down_read(&vcpu->kvm->slots_lock);
1148 for (i = 0; i < msrs->nmsrs; ++i)
1149 if (do_msr(vcpu, entries[i].index, &entries[i].data))
1151 up_read(&vcpu->kvm->slots_lock);
1159 * Read or write a bunch of msrs. Parameters are user addresses.
1161 * @return number of msrs set successfully.
1163 static int msr_io(struct kvm_vcpu *vcpu, struct kvm_msrs __user *user_msrs,
1164 int (*do_msr)(struct kvm_vcpu *vcpu,
1165 unsigned index, u64 *data),
1168 struct kvm_msrs msrs;
1169 struct kvm_msr_entry *entries;
1174 if (copy_from_user(&msrs, user_msrs, sizeof msrs))
1178 if (msrs.nmsrs >= MAX_IO_MSRS)
1182 size = sizeof(struct kvm_msr_entry) * msrs.nmsrs;
1183 entries = vmalloc(size);
1188 if (copy_from_user(entries, user_msrs->entries, size))
1191 r = n = __msr_io(vcpu, &msrs, entries, do_msr);
1196 if (writeback && copy_to_user(user_msrs->entries, entries, size))
1207 int kvm_dev_ioctl_check_extension(long ext)
1212 case KVM_CAP_IRQCHIP:
1214 case KVM_CAP_MMU_SHADOW_CACHE_CONTROL:
1215 case KVM_CAP_SET_TSS_ADDR:
1216 case KVM_CAP_EXT_CPUID:
1217 case KVM_CAP_CLOCKSOURCE:
1219 case KVM_CAP_NOP_IO_DELAY:
1220 case KVM_CAP_MP_STATE:
1221 case KVM_CAP_SYNC_MMU:
1222 case KVM_CAP_REINJECT_CONTROL:
1223 case KVM_CAP_IRQ_INJECT_STATUS:
1224 case KVM_CAP_ASSIGN_DEV_IRQ:
1226 case KVM_CAP_IOEVENTFD:
1228 case KVM_CAP_PIT_STATE2:
1229 case KVM_CAP_SET_IDENTITY_MAP_ADDR:
1230 case KVM_CAP_ADJUST_CLOCK:
1233 case KVM_CAP_COALESCED_MMIO:
1234 r = KVM_COALESCED_MMIO_PAGE_OFFSET;
1237 r = !kvm_x86_ops->cpu_has_accelerated_tpr();
1239 case KVM_CAP_NR_VCPUS:
1242 case KVM_CAP_NR_MEMSLOTS:
1243 r = KVM_MEMORY_SLOTS;
1245 case KVM_CAP_PV_MMU:
1252 r = KVM_MAX_MCE_BANKS;
1262 long kvm_arch_dev_ioctl(struct file *filp,
1263 unsigned int ioctl, unsigned long arg)
1265 void __user *argp = (void __user *)arg;
1269 case KVM_GET_MSR_INDEX_LIST: {
1270 struct kvm_msr_list __user *user_msr_list = argp;
1271 struct kvm_msr_list msr_list;
1275 if (copy_from_user(&msr_list, user_msr_list, sizeof msr_list))
1278 msr_list.nmsrs = num_msrs_to_save + ARRAY_SIZE(emulated_msrs);
1279 if (copy_to_user(user_msr_list, &msr_list, sizeof msr_list))
1282 if (n < msr_list.nmsrs)
1285 if (copy_to_user(user_msr_list->indices, &msrs_to_save,
1286 num_msrs_to_save * sizeof(u32)))
1288 if (copy_to_user(user_msr_list->indices + num_msrs_to_save,
1290 ARRAY_SIZE(emulated_msrs) * sizeof(u32)))
1295 case KVM_GET_SUPPORTED_CPUID: {
1296 struct kvm_cpuid2 __user *cpuid_arg = argp;
1297 struct kvm_cpuid2 cpuid;
1300 if (copy_from_user(&cpuid, cpuid_arg, sizeof cpuid))
1302 r = kvm_dev_ioctl_get_supported_cpuid(&cpuid,
1303 cpuid_arg->entries);
1308 if (copy_to_user(cpuid_arg, &cpuid, sizeof cpuid))
1313 case KVM_X86_GET_MCE_CAP_SUPPORTED: {
1316 mce_cap = KVM_MCE_CAP_SUPPORTED;
1318 if (copy_to_user(argp, &mce_cap, sizeof mce_cap))
1330 void kvm_arch_vcpu_load(struct kvm_vcpu *vcpu, int cpu)
1332 kvm_x86_ops->vcpu_load(vcpu, cpu);
1333 kvm_request_guest_time_update(vcpu);
1336 void kvm_arch_vcpu_put(struct kvm_vcpu *vcpu)
1338 kvm_x86_ops->vcpu_put(vcpu);
1339 kvm_put_guest_fpu(vcpu);
1342 static int is_efer_nx(void)
1344 unsigned long long efer = 0;
1346 rdmsrl_safe(MSR_EFER, &efer);
1347 return efer & EFER_NX;
1350 static void cpuid_fix_nx_cap(struct kvm_vcpu *vcpu)
1353 struct kvm_cpuid_entry2 *e, *entry;
1356 for (i = 0; i < vcpu->arch.cpuid_nent; ++i) {
1357 e = &vcpu->arch.cpuid_entries[i];
1358 if (e->function == 0x80000001) {
1363 if (entry && (entry->edx & (1 << 20)) && !is_efer_nx()) {
1364 entry->edx &= ~(1 << 20);
1365 printk(KERN_INFO "kvm: guest NX capability removed\n");
1369 /* when an old userspace process fills a new kernel module */
1370 static int kvm_vcpu_ioctl_set_cpuid(struct kvm_vcpu *vcpu,
1371 struct kvm_cpuid *cpuid,
1372 struct kvm_cpuid_entry __user *entries)
1375 struct kvm_cpuid_entry *cpuid_entries;
1378 if (cpuid->nent > KVM_MAX_CPUID_ENTRIES)
1381 cpuid_entries = vmalloc(sizeof(struct kvm_cpuid_entry) * cpuid->nent);
1385 if (copy_from_user(cpuid_entries, entries,
1386 cpuid->nent * sizeof(struct kvm_cpuid_entry)))
1388 for (i = 0; i < cpuid->nent; i++) {
1389 vcpu->arch.cpuid_entries[i].function = cpuid_entries[i].function;
1390 vcpu->arch.cpuid_entries[i].eax = cpuid_entries[i].eax;
1391 vcpu->arch.cpuid_entries[i].ebx = cpuid_entries[i].ebx;
1392 vcpu->arch.cpuid_entries[i].ecx = cpuid_entries[i].ecx;
1393 vcpu->arch.cpuid_entries[i].edx = cpuid_entries[i].edx;
1394 vcpu->arch.cpuid_entries[i].index = 0;
1395 vcpu->arch.cpuid_entries[i].flags = 0;
1396 vcpu->arch.cpuid_entries[i].padding[0] = 0;
1397 vcpu->arch.cpuid_entries[i].padding[1] = 0;
1398 vcpu->arch.cpuid_entries[i].padding[2] = 0;
1400 vcpu->arch.cpuid_nent = cpuid->nent;
1401 cpuid_fix_nx_cap(vcpu);
1403 kvm_apic_set_version(vcpu);
1406 vfree(cpuid_entries);
1411 static int kvm_vcpu_ioctl_set_cpuid2(struct kvm_vcpu *vcpu,
1412 struct kvm_cpuid2 *cpuid,
1413 struct kvm_cpuid_entry2 __user *entries)
1418 if (cpuid->nent > KVM_MAX_CPUID_ENTRIES)
1421 if (copy_from_user(&vcpu->arch.cpuid_entries, entries,
1422 cpuid->nent * sizeof(struct kvm_cpuid_entry2)))
1424 vcpu->arch.cpuid_nent = cpuid->nent;
1425 kvm_apic_set_version(vcpu);
1432 static int kvm_vcpu_ioctl_get_cpuid2(struct kvm_vcpu *vcpu,
1433 struct kvm_cpuid2 *cpuid,
1434 struct kvm_cpuid_entry2 __user *entries)
1439 if (cpuid->nent < vcpu->arch.cpuid_nent)
1442 if (copy_to_user(entries, &vcpu->arch.cpuid_entries,
1443 vcpu->arch.cpuid_nent * sizeof(struct kvm_cpuid_entry2)))
1448 cpuid->nent = vcpu->arch.cpuid_nent;
1452 static void do_cpuid_1_ent(struct kvm_cpuid_entry2 *entry, u32 function,
1455 entry->function = function;
1456 entry->index = index;
1457 cpuid_count(entry->function, entry->index,
1458 &entry->eax, &entry->ebx, &entry->ecx, &entry->edx);
1462 #define F(x) bit(X86_FEATURE_##x)
1464 static void do_cpuid_ent(struct kvm_cpuid_entry2 *entry, u32 function,
1465 u32 index, int *nent, int maxnent)
1467 unsigned f_nx = is_efer_nx() ? F(NX) : 0;
1468 unsigned f_gbpages = kvm_x86_ops->gb_page_enable() ? F(GBPAGES) : 0;
1469 #ifdef CONFIG_X86_64
1470 unsigned f_lm = F(LM);
1476 const u32 kvm_supported_word0_x86_features =
1477 F(FPU) | F(VME) | F(DE) | F(PSE) |
1478 F(TSC) | F(MSR) | F(PAE) | F(MCE) |
1479 F(CX8) | F(APIC) | 0 /* Reserved */ | F(SEP) |
1480 F(MTRR) | F(PGE) | F(MCA) | F(CMOV) |
1481 F(PAT) | F(PSE36) | 0 /* PSN */ | F(CLFLSH) |
1482 0 /* Reserved, DS, ACPI */ | F(MMX) |
1483 F(FXSR) | F(XMM) | F(XMM2) | F(SELFSNOOP) |
1484 0 /* HTT, TM, Reserved, PBE */;
1485 /* cpuid 0x80000001.edx */
1486 const u32 kvm_supported_word1_x86_features =
1487 F(FPU) | F(VME) | F(DE) | F(PSE) |
1488 F(TSC) | F(MSR) | F(PAE) | F(MCE) |
1489 F(CX8) | F(APIC) | 0 /* Reserved */ | F(SYSCALL) |
1490 F(MTRR) | F(PGE) | F(MCA) | F(CMOV) |
1491 F(PAT) | F(PSE36) | 0 /* Reserved */ |
1492 f_nx | 0 /* Reserved */ | F(MMXEXT) | F(MMX) |
1493 F(FXSR) | F(FXSR_OPT) | f_gbpages | 0 /* RDTSCP */ |
1494 0 /* Reserved */ | f_lm | F(3DNOWEXT) | F(3DNOW);
1496 const u32 kvm_supported_word4_x86_features =
1497 F(XMM3) | 0 /* Reserved, DTES64, MONITOR */ |
1498 0 /* DS-CPL, VMX, SMX, EST */ |
1499 0 /* TM2 */ | F(SSSE3) | 0 /* CNXT-ID */ | 0 /* Reserved */ |
1500 0 /* Reserved */ | F(CX16) | 0 /* xTPR Update, PDCM */ |
1501 0 /* Reserved, DCA */ | F(XMM4_1) |
1502 F(XMM4_2) | F(X2APIC) | F(MOVBE) | F(POPCNT) |
1503 0 /* Reserved, XSAVE, OSXSAVE */;
1504 /* cpuid 0x80000001.ecx */
1505 const u32 kvm_supported_word6_x86_features =
1506 F(LAHF_LM) | F(CMP_LEGACY) | F(SVM) | 0 /* ExtApicSpace */ |
1507 F(CR8_LEGACY) | F(ABM) | F(SSE4A) | F(MISALIGNSSE) |
1508 F(3DNOWPREFETCH) | 0 /* OSVW */ | 0 /* IBS */ | F(SSE5) |
1509 0 /* SKINIT */ | 0 /* WDT */;
1511 /* all calls to cpuid_count() should be made on the same cpu */
1513 do_cpuid_1_ent(entry, function, index);
1518 entry->eax = min(entry->eax, (u32)0xb);
1521 entry->edx &= kvm_supported_word0_x86_features;
1522 entry->ecx &= kvm_supported_word4_x86_features;
1523 /* we support x2apic emulation even if host does not support
1524 * it since we emulate x2apic in software */
1525 entry->ecx |= F(X2APIC);
1527 /* function 2 entries are STATEFUL. That is, repeated cpuid commands
1528 * may return different values. This forces us to get_cpu() before
1529 * issuing the first command, and also to emulate this annoying behavior
1530 * in kvm_emulate_cpuid() using KVM_CPUID_FLAG_STATE_READ_NEXT */
1532 int t, times = entry->eax & 0xff;
1534 entry->flags |= KVM_CPUID_FLAG_STATEFUL_FUNC;
1535 entry->flags |= KVM_CPUID_FLAG_STATE_READ_NEXT;
1536 for (t = 1; t < times && *nent < maxnent; ++t) {
1537 do_cpuid_1_ent(&entry[t], function, 0);
1538 entry[t].flags |= KVM_CPUID_FLAG_STATEFUL_FUNC;
1543 /* function 4 and 0xb have additional index. */
1547 entry->flags |= KVM_CPUID_FLAG_SIGNIFCANT_INDEX;
1548 /* read more entries until cache_type is zero */
1549 for (i = 1; *nent < maxnent; ++i) {
1550 cache_type = entry[i - 1].eax & 0x1f;
1553 do_cpuid_1_ent(&entry[i], function, i);
1555 KVM_CPUID_FLAG_SIGNIFCANT_INDEX;
1563 entry->flags |= KVM_CPUID_FLAG_SIGNIFCANT_INDEX;
1564 /* read more entries until level_type is zero */
1565 for (i = 1; *nent < maxnent; ++i) {
1566 level_type = entry[i - 1].ecx & 0xff00;
1569 do_cpuid_1_ent(&entry[i], function, i);
1571 KVM_CPUID_FLAG_SIGNIFCANT_INDEX;
1577 entry->eax = min(entry->eax, 0x8000001a);
1580 entry->edx &= kvm_supported_word1_x86_features;
1581 entry->ecx &= kvm_supported_word6_x86_features;
1589 static int kvm_dev_ioctl_get_supported_cpuid(struct kvm_cpuid2 *cpuid,
1590 struct kvm_cpuid_entry2 __user *entries)
1592 struct kvm_cpuid_entry2 *cpuid_entries;
1593 int limit, nent = 0, r = -E2BIG;
1596 if (cpuid->nent < 1)
1598 if (cpuid->nent > KVM_MAX_CPUID_ENTRIES)
1599 cpuid->nent = KVM_MAX_CPUID_ENTRIES;
1601 cpuid_entries = vmalloc(sizeof(struct kvm_cpuid_entry2) * cpuid->nent);
1605 do_cpuid_ent(&cpuid_entries[0], 0, 0, &nent, cpuid->nent);
1606 limit = cpuid_entries[0].eax;
1607 for (func = 1; func <= limit && nent < cpuid->nent; ++func)
1608 do_cpuid_ent(&cpuid_entries[nent], func, 0,
1609 &nent, cpuid->nent);
1611 if (nent >= cpuid->nent)
1614 do_cpuid_ent(&cpuid_entries[nent], 0x80000000, 0, &nent, cpuid->nent);
1615 limit = cpuid_entries[nent - 1].eax;
1616 for (func = 0x80000001; func <= limit && nent < cpuid->nent; ++func)
1617 do_cpuid_ent(&cpuid_entries[nent], func, 0,
1618 &nent, cpuid->nent);
1620 if (nent >= cpuid->nent)
1624 if (copy_to_user(entries, cpuid_entries,
1625 nent * sizeof(struct kvm_cpuid_entry2)))
1631 vfree(cpuid_entries);
1636 static int kvm_vcpu_ioctl_get_lapic(struct kvm_vcpu *vcpu,
1637 struct kvm_lapic_state *s)
1640 memcpy(s->regs, vcpu->arch.apic->regs, sizeof *s);
1646 static int kvm_vcpu_ioctl_set_lapic(struct kvm_vcpu *vcpu,
1647 struct kvm_lapic_state *s)
1650 memcpy(vcpu->arch.apic->regs, s->regs, sizeof *s);
1651 kvm_apic_post_state_restore(vcpu);
1652 update_cr8_intercept(vcpu);
1658 static int kvm_vcpu_ioctl_interrupt(struct kvm_vcpu *vcpu,
1659 struct kvm_interrupt *irq)
1661 if (irq->irq < 0 || irq->irq >= 256)
1663 if (irqchip_in_kernel(vcpu->kvm))
1667 kvm_queue_interrupt(vcpu, irq->irq, false);
1674 static int kvm_vcpu_ioctl_nmi(struct kvm_vcpu *vcpu)
1677 kvm_inject_nmi(vcpu);
1683 static int vcpu_ioctl_tpr_access_reporting(struct kvm_vcpu *vcpu,
1684 struct kvm_tpr_access_ctl *tac)
1688 vcpu->arch.tpr_access_reporting = !!tac->enabled;
1692 static int kvm_vcpu_ioctl_x86_setup_mce(struct kvm_vcpu *vcpu,
1696 unsigned bank_num = mcg_cap & 0xff, bank;
1699 if (!bank_num || bank_num >= KVM_MAX_MCE_BANKS)
1701 if (mcg_cap & ~(KVM_MCE_CAP_SUPPORTED | 0xff | 0xff0000))
1704 vcpu->arch.mcg_cap = mcg_cap;
1705 /* Init IA32_MCG_CTL to all 1s */
1706 if (mcg_cap & MCG_CTL_P)
1707 vcpu->arch.mcg_ctl = ~(u64)0;
1708 /* Init IA32_MCi_CTL to all 1s */
1709 for (bank = 0; bank < bank_num; bank++)
1710 vcpu->arch.mce_banks[bank*4] = ~(u64)0;
1715 static int kvm_vcpu_ioctl_x86_set_mce(struct kvm_vcpu *vcpu,
1716 struct kvm_x86_mce *mce)
1718 u64 mcg_cap = vcpu->arch.mcg_cap;
1719 unsigned bank_num = mcg_cap & 0xff;
1720 u64 *banks = vcpu->arch.mce_banks;
1722 if (mce->bank >= bank_num || !(mce->status & MCI_STATUS_VAL))
1725 * if IA32_MCG_CTL is not all 1s, the uncorrected error
1726 * reporting is disabled
1728 if ((mce->status & MCI_STATUS_UC) && (mcg_cap & MCG_CTL_P) &&
1729 vcpu->arch.mcg_ctl != ~(u64)0)
1731 banks += 4 * mce->bank;
1733 * if IA32_MCi_CTL is not all 1s, the uncorrected error
1734 * reporting is disabled for the bank
1736 if ((mce->status & MCI_STATUS_UC) && banks[0] != ~(u64)0)
1738 if (mce->status & MCI_STATUS_UC) {
1739 if ((vcpu->arch.mcg_status & MCG_STATUS_MCIP) ||
1740 !(vcpu->arch.cr4 & X86_CR4_MCE)) {
1741 printk(KERN_DEBUG "kvm: set_mce: "
1742 "injects mce exception while "
1743 "previous one is in progress!\n");
1744 set_bit(KVM_REQ_TRIPLE_FAULT, &vcpu->requests);
1747 if (banks[1] & MCI_STATUS_VAL)
1748 mce->status |= MCI_STATUS_OVER;
1749 banks[2] = mce->addr;
1750 banks[3] = mce->misc;
1751 vcpu->arch.mcg_status = mce->mcg_status;
1752 banks[1] = mce->status;
1753 kvm_queue_exception(vcpu, MC_VECTOR);
1754 } else if (!(banks[1] & MCI_STATUS_VAL)
1755 || !(banks[1] & MCI_STATUS_UC)) {
1756 if (banks[1] & MCI_STATUS_VAL)
1757 mce->status |= MCI_STATUS_OVER;
1758 banks[2] = mce->addr;
1759 banks[3] = mce->misc;
1760 banks[1] = mce->status;
1762 banks[1] |= MCI_STATUS_OVER;
1766 long kvm_arch_vcpu_ioctl(struct file *filp,
1767 unsigned int ioctl, unsigned long arg)
1769 struct kvm_vcpu *vcpu = filp->private_data;
1770 void __user *argp = (void __user *)arg;
1772 struct kvm_lapic_state *lapic = NULL;
1775 case KVM_GET_LAPIC: {
1776 lapic = kzalloc(sizeof(struct kvm_lapic_state), GFP_KERNEL);
1781 r = kvm_vcpu_ioctl_get_lapic(vcpu, lapic);
1785 if (copy_to_user(argp, lapic, sizeof(struct kvm_lapic_state)))
1790 case KVM_SET_LAPIC: {
1791 lapic = kmalloc(sizeof(struct kvm_lapic_state), GFP_KERNEL);
1796 if (copy_from_user(lapic, argp, sizeof(struct kvm_lapic_state)))
1798 r = kvm_vcpu_ioctl_set_lapic(vcpu, lapic);
1804 case KVM_INTERRUPT: {
1805 struct kvm_interrupt irq;
1808 if (copy_from_user(&irq, argp, sizeof irq))
1810 r = kvm_vcpu_ioctl_interrupt(vcpu, &irq);
1817 r = kvm_vcpu_ioctl_nmi(vcpu);
1823 case KVM_SET_CPUID: {
1824 struct kvm_cpuid __user *cpuid_arg = argp;
1825 struct kvm_cpuid cpuid;
1828 if (copy_from_user(&cpuid, cpuid_arg, sizeof cpuid))
1830 r = kvm_vcpu_ioctl_set_cpuid(vcpu, &cpuid, cpuid_arg->entries);
1835 case KVM_SET_CPUID2: {
1836 struct kvm_cpuid2 __user *cpuid_arg = argp;
1837 struct kvm_cpuid2 cpuid;
1840 if (copy_from_user(&cpuid, cpuid_arg, sizeof cpuid))
1842 r = kvm_vcpu_ioctl_set_cpuid2(vcpu, &cpuid,
1843 cpuid_arg->entries);
1848 case KVM_GET_CPUID2: {
1849 struct kvm_cpuid2 __user *cpuid_arg = argp;
1850 struct kvm_cpuid2 cpuid;
1853 if (copy_from_user(&cpuid, cpuid_arg, sizeof cpuid))
1855 r = kvm_vcpu_ioctl_get_cpuid2(vcpu, &cpuid,
1856 cpuid_arg->entries);
1860 if (copy_to_user(cpuid_arg, &cpuid, sizeof cpuid))
1866 r = msr_io(vcpu, argp, kvm_get_msr, 1);
1869 r = msr_io(vcpu, argp, do_set_msr, 0);
1871 case KVM_TPR_ACCESS_REPORTING: {
1872 struct kvm_tpr_access_ctl tac;
1875 if (copy_from_user(&tac, argp, sizeof tac))
1877 r = vcpu_ioctl_tpr_access_reporting(vcpu, &tac);
1881 if (copy_to_user(argp, &tac, sizeof tac))
1886 case KVM_SET_VAPIC_ADDR: {
1887 struct kvm_vapic_addr va;
1890 if (!irqchip_in_kernel(vcpu->kvm))
1893 if (copy_from_user(&va, argp, sizeof va))
1896 kvm_lapic_set_vapic_addr(vcpu, va.vapic_addr);
1899 case KVM_X86_SETUP_MCE: {
1903 if (copy_from_user(&mcg_cap, argp, sizeof mcg_cap))
1905 r = kvm_vcpu_ioctl_x86_setup_mce(vcpu, mcg_cap);
1908 case KVM_X86_SET_MCE: {
1909 struct kvm_x86_mce mce;
1912 if (copy_from_user(&mce, argp, sizeof mce))
1914 r = kvm_vcpu_ioctl_x86_set_mce(vcpu, &mce);
1925 static int kvm_vm_ioctl_set_tss_addr(struct kvm *kvm, unsigned long addr)
1929 if (addr > (unsigned int)(-3 * PAGE_SIZE))
1931 ret = kvm_x86_ops->set_tss_addr(kvm, addr);
1935 static int kvm_vm_ioctl_set_identity_map_addr(struct kvm *kvm,
1938 kvm->arch.ept_identity_map_addr = ident_addr;
1942 static int kvm_vm_ioctl_set_nr_mmu_pages(struct kvm *kvm,
1943 u32 kvm_nr_mmu_pages)
1945 if (kvm_nr_mmu_pages < KVM_MIN_ALLOC_MMU_PAGES)
1948 down_write(&kvm->slots_lock);
1949 spin_lock(&kvm->mmu_lock);
1951 kvm_mmu_change_mmu_pages(kvm, kvm_nr_mmu_pages);
1952 kvm->arch.n_requested_mmu_pages = kvm_nr_mmu_pages;
1954 spin_unlock(&kvm->mmu_lock);
1955 up_write(&kvm->slots_lock);
1959 static int kvm_vm_ioctl_get_nr_mmu_pages(struct kvm *kvm)
1961 return kvm->arch.n_alloc_mmu_pages;
1964 gfn_t unalias_gfn(struct kvm *kvm, gfn_t gfn)
1967 struct kvm_mem_alias *alias;
1969 for (i = 0; i < kvm->arch.naliases; ++i) {
1970 alias = &kvm->arch.aliases[i];
1971 if (gfn >= alias->base_gfn
1972 && gfn < alias->base_gfn + alias->npages)
1973 return alias->target_gfn + gfn - alias->base_gfn;
1979 * Set a new alias region. Aliases map a portion of physical memory into
1980 * another portion. This is useful for memory windows, for example the PC
1983 static int kvm_vm_ioctl_set_memory_alias(struct kvm *kvm,
1984 struct kvm_memory_alias *alias)
1987 struct kvm_mem_alias *p;
1990 /* General sanity checks */
1991 if (alias->memory_size & (PAGE_SIZE - 1))
1993 if (alias->guest_phys_addr & (PAGE_SIZE - 1))
1995 if (alias->slot >= KVM_ALIAS_SLOTS)
1997 if (alias->guest_phys_addr + alias->memory_size
1998 < alias->guest_phys_addr)
2000 if (alias->target_phys_addr + alias->memory_size
2001 < alias->target_phys_addr)
2004 down_write(&kvm->slots_lock);
2005 spin_lock(&kvm->mmu_lock);
2007 p = &kvm->arch.aliases[alias->slot];
2008 p->base_gfn = alias->guest_phys_addr >> PAGE_SHIFT;
2009 p->npages = alias->memory_size >> PAGE_SHIFT;
2010 p->target_gfn = alias->target_phys_addr >> PAGE_SHIFT;
2012 for (n = KVM_ALIAS_SLOTS; n > 0; --n)
2013 if (kvm->arch.aliases[n - 1].npages)
2015 kvm->arch.naliases = n;
2017 spin_unlock(&kvm->mmu_lock);
2018 kvm_mmu_zap_all(kvm);
2020 up_write(&kvm->slots_lock);
2028 static int kvm_vm_ioctl_get_irqchip(struct kvm *kvm, struct kvm_irqchip *chip)
2033 switch (chip->chip_id) {
2034 case KVM_IRQCHIP_PIC_MASTER:
2035 memcpy(&chip->chip.pic,
2036 &pic_irqchip(kvm)->pics[0],
2037 sizeof(struct kvm_pic_state));
2039 case KVM_IRQCHIP_PIC_SLAVE:
2040 memcpy(&chip->chip.pic,
2041 &pic_irqchip(kvm)->pics[1],
2042 sizeof(struct kvm_pic_state));
2044 case KVM_IRQCHIP_IOAPIC:
2045 memcpy(&chip->chip.ioapic,
2046 ioapic_irqchip(kvm),
2047 sizeof(struct kvm_ioapic_state));
2056 static int kvm_vm_ioctl_set_irqchip(struct kvm *kvm, struct kvm_irqchip *chip)
2061 switch (chip->chip_id) {
2062 case KVM_IRQCHIP_PIC_MASTER:
2063 spin_lock(&pic_irqchip(kvm)->lock);
2064 memcpy(&pic_irqchip(kvm)->pics[0],
2066 sizeof(struct kvm_pic_state));
2067 spin_unlock(&pic_irqchip(kvm)->lock);
2069 case KVM_IRQCHIP_PIC_SLAVE:
2070 spin_lock(&pic_irqchip(kvm)->lock);
2071 memcpy(&pic_irqchip(kvm)->pics[1],
2073 sizeof(struct kvm_pic_state));
2074 spin_unlock(&pic_irqchip(kvm)->lock);
2076 case KVM_IRQCHIP_IOAPIC:
2077 mutex_lock(&kvm->irq_lock);
2078 memcpy(ioapic_irqchip(kvm),
2080 sizeof(struct kvm_ioapic_state));
2081 mutex_unlock(&kvm->irq_lock);
2087 kvm_pic_update_irq(pic_irqchip(kvm));
2091 static int kvm_vm_ioctl_get_pit(struct kvm *kvm, struct kvm_pit_state *ps)
2095 mutex_lock(&kvm->arch.vpit->pit_state.lock);
2096 memcpy(ps, &kvm->arch.vpit->pit_state, sizeof(struct kvm_pit_state));
2097 mutex_unlock(&kvm->arch.vpit->pit_state.lock);
2101 static int kvm_vm_ioctl_set_pit(struct kvm *kvm, struct kvm_pit_state *ps)
2105 mutex_lock(&kvm->arch.vpit->pit_state.lock);
2106 memcpy(&kvm->arch.vpit->pit_state, ps, sizeof(struct kvm_pit_state));
2107 kvm_pit_load_count(kvm, 0, ps->channels[0].count, 0);
2108 mutex_unlock(&kvm->arch.vpit->pit_state.lock);
2112 static int kvm_vm_ioctl_get_pit2(struct kvm *kvm, struct kvm_pit_state2 *ps)
2116 mutex_lock(&kvm->arch.vpit->pit_state.lock);
2117 memcpy(ps->channels, &kvm->arch.vpit->pit_state.channels,
2118 sizeof(ps->channels));
2119 ps->flags = kvm->arch.vpit->pit_state.flags;
2120 mutex_unlock(&kvm->arch.vpit->pit_state.lock);
2124 static int kvm_vm_ioctl_set_pit2(struct kvm *kvm, struct kvm_pit_state2 *ps)
2126 int r = 0, start = 0;
2127 u32 prev_legacy, cur_legacy;
2128 mutex_lock(&kvm->arch.vpit->pit_state.lock);
2129 prev_legacy = kvm->arch.vpit->pit_state.flags & KVM_PIT_FLAGS_HPET_LEGACY;
2130 cur_legacy = ps->flags & KVM_PIT_FLAGS_HPET_LEGACY;
2131 if (!prev_legacy && cur_legacy)
2133 memcpy(&kvm->arch.vpit->pit_state.channels, &ps->channels,
2134 sizeof(kvm->arch.vpit->pit_state.channels));
2135 kvm->arch.vpit->pit_state.flags = ps->flags;
2136 kvm_pit_load_count(kvm, 0, kvm->arch.vpit->pit_state.channels[0].count, start);
2137 mutex_unlock(&kvm->arch.vpit->pit_state.lock);
2141 static int kvm_vm_ioctl_reinject(struct kvm *kvm,
2142 struct kvm_reinject_control *control)
2144 if (!kvm->arch.vpit)
2146 mutex_lock(&kvm->arch.vpit->pit_state.lock);
2147 kvm->arch.vpit->pit_state.pit_timer.reinject = control->pit_reinject;
2148 mutex_unlock(&kvm->arch.vpit->pit_state.lock);
2153 * Get (and clear) the dirty memory log for a memory slot.
2155 int kvm_vm_ioctl_get_dirty_log(struct kvm *kvm,
2156 struct kvm_dirty_log *log)
2160 struct kvm_memory_slot *memslot;
2163 down_write(&kvm->slots_lock);
2165 r = kvm_get_dirty_log(kvm, log, &is_dirty);
2169 /* If nothing is dirty, don't bother messing with page tables. */
2171 spin_lock(&kvm->mmu_lock);
2172 kvm_mmu_slot_remove_write_access(kvm, log->slot);
2173 spin_unlock(&kvm->mmu_lock);
2174 memslot = &kvm->memslots[log->slot];
2175 n = ALIGN(memslot->npages, BITS_PER_LONG) / 8;
2176 memset(memslot->dirty_bitmap, 0, n);
2180 up_write(&kvm->slots_lock);
2184 long kvm_arch_vm_ioctl(struct file *filp,
2185 unsigned int ioctl, unsigned long arg)
2187 struct kvm *kvm = filp->private_data;
2188 void __user *argp = (void __user *)arg;
2191 * This union makes it completely explicit to gcc-3.x
2192 * that these two variables' stack usage should be
2193 * combined, not added together.
2196 struct kvm_pit_state ps;
2197 struct kvm_pit_state2 ps2;
2198 struct kvm_memory_alias alias;
2199 struct kvm_pit_config pit_config;
2203 case KVM_SET_TSS_ADDR:
2204 r = kvm_vm_ioctl_set_tss_addr(kvm, arg);
2208 case KVM_SET_IDENTITY_MAP_ADDR: {
2212 if (copy_from_user(&ident_addr, argp, sizeof ident_addr))
2214 r = kvm_vm_ioctl_set_identity_map_addr(kvm, ident_addr);
2219 case KVM_SET_MEMORY_REGION: {
2220 struct kvm_memory_region kvm_mem;
2221 struct kvm_userspace_memory_region kvm_userspace_mem;
2224 if (copy_from_user(&kvm_mem, argp, sizeof kvm_mem))
2226 kvm_userspace_mem.slot = kvm_mem.slot;
2227 kvm_userspace_mem.flags = kvm_mem.flags;
2228 kvm_userspace_mem.guest_phys_addr = kvm_mem.guest_phys_addr;
2229 kvm_userspace_mem.memory_size = kvm_mem.memory_size;
2230 r = kvm_vm_ioctl_set_memory_region(kvm, &kvm_userspace_mem, 0);
2235 case KVM_SET_NR_MMU_PAGES:
2236 r = kvm_vm_ioctl_set_nr_mmu_pages(kvm, arg);
2240 case KVM_GET_NR_MMU_PAGES:
2241 r = kvm_vm_ioctl_get_nr_mmu_pages(kvm);
2243 case KVM_SET_MEMORY_ALIAS:
2245 if (copy_from_user(&u.alias, argp, sizeof(struct kvm_memory_alias)))
2247 r = kvm_vm_ioctl_set_memory_alias(kvm, &u.alias);
2251 case KVM_CREATE_IRQCHIP:
2253 kvm->arch.vpic = kvm_create_pic(kvm);
2254 if (kvm->arch.vpic) {
2255 r = kvm_ioapic_init(kvm);
2257 kfree(kvm->arch.vpic);
2258 kvm->arch.vpic = NULL;
2263 r = kvm_setup_default_irq_routing(kvm);
2265 kfree(kvm->arch.vpic);
2266 kfree(kvm->arch.vioapic);
2270 case KVM_CREATE_PIT:
2271 u.pit_config.flags = KVM_PIT_SPEAKER_DUMMY;
2273 case KVM_CREATE_PIT2:
2275 if (copy_from_user(&u.pit_config, argp,
2276 sizeof(struct kvm_pit_config)))
2279 down_write(&kvm->slots_lock);
2282 goto create_pit_unlock;
2284 kvm->arch.vpit = kvm_create_pit(kvm, u.pit_config.flags);
2288 up_write(&kvm->slots_lock);
2290 case KVM_IRQ_LINE_STATUS:
2291 case KVM_IRQ_LINE: {
2292 struct kvm_irq_level irq_event;
2295 if (copy_from_user(&irq_event, argp, sizeof irq_event))
2297 if (irqchip_in_kernel(kvm)) {
2299 mutex_lock(&kvm->irq_lock);
2300 status = kvm_set_irq(kvm, KVM_USERSPACE_IRQ_SOURCE_ID,
2301 irq_event.irq, irq_event.level);
2302 mutex_unlock(&kvm->irq_lock);
2303 if (ioctl == KVM_IRQ_LINE_STATUS) {
2304 irq_event.status = status;
2305 if (copy_to_user(argp, &irq_event,
2313 case KVM_GET_IRQCHIP: {
2314 /* 0: PIC master, 1: PIC slave, 2: IOAPIC */
2315 struct kvm_irqchip *chip = kmalloc(sizeof(*chip), GFP_KERNEL);
2321 if (copy_from_user(chip, argp, sizeof *chip))
2322 goto get_irqchip_out;
2324 if (!irqchip_in_kernel(kvm))
2325 goto get_irqchip_out;
2326 r = kvm_vm_ioctl_get_irqchip(kvm, chip);
2328 goto get_irqchip_out;
2330 if (copy_to_user(argp, chip, sizeof *chip))
2331 goto get_irqchip_out;
2339 case KVM_SET_IRQCHIP: {
2340 /* 0: PIC master, 1: PIC slave, 2: IOAPIC */
2341 struct kvm_irqchip *chip = kmalloc(sizeof(*chip), GFP_KERNEL);
2347 if (copy_from_user(chip, argp, sizeof *chip))
2348 goto set_irqchip_out;
2350 if (!irqchip_in_kernel(kvm))
2351 goto set_irqchip_out;
2352 r = kvm_vm_ioctl_set_irqchip(kvm, chip);
2354 goto set_irqchip_out;
2364 if (copy_from_user(&u.ps, argp, sizeof(struct kvm_pit_state)))
2367 if (!kvm->arch.vpit)
2369 r = kvm_vm_ioctl_get_pit(kvm, &u.ps);
2373 if (copy_to_user(argp, &u.ps, sizeof(struct kvm_pit_state)))
2380 if (copy_from_user(&u.ps, argp, sizeof u.ps))
2383 if (!kvm->arch.vpit)
2385 r = kvm_vm_ioctl_set_pit(kvm, &u.ps);
2391 case KVM_GET_PIT2: {
2393 if (!kvm->arch.vpit)
2395 r = kvm_vm_ioctl_get_pit2(kvm, &u.ps2);
2399 if (copy_to_user(argp, &u.ps2, sizeof(u.ps2)))
2404 case KVM_SET_PIT2: {
2406 if (copy_from_user(&u.ps2, argp, sizeof(u.ps2)))
2409 if (!kvm->arch.vpit)
2411 r = kvm_vm_ioctl_set_pit2(kvm, &u.ps2);
2417 case KVM_REINJECT_CONTROL: {
2418 struct kvm_reinject_control control;
2420 if (copy_from_user(&control, argp, sizeof(control)))
2422 r = kvm_vm_ioctl_reinject(kvm, &control);
2428 case KVM_SET_CLOCK: {
2429 struct timespec now;
2430 struct kvm_clock_data user_ns;
2435 if (copy_from_user(&user_ns, argp, sizeof(user_ns)))
2444 now_ns = timespec_to_ns(&now);
2445 delta = user_ns.clock - now_ns;
2446 kvm->arch.kvmclock_offset = delta;
2449 case KVM_GET_CLOCK: {
2450 struct timespec now;
2451 struct kvm_clock_data user_ns;
2455 now_ns = timespec_to_ns(&now);
2456 user_ns.clock = kvm->arch.kvmclock_offset + now_ns;
2460 if (copy_to_user(argp, &user_ns, sizeof(user_ns)))
2473 static void kvm_init_msr_list(void)
2478 /* skip the first msrs in the list. KVM-specific */
2479 for (i = j = KVM_SAVE_MSRS_BEGIN; i < ARRAY_SIZE(msrs_to_save); i++) {
2480 if (rdmsr_safe(msrs_to_save[i], &dummy[0], &dummy[1]) < 0)
2483 msrs_to_save[j] = msrs_to_save[i];
2486 num_msrs_to_save = j;
2489 static int vcpu_mmio_write(struct kvm_vcpu *vcpu, gpa_t addr, int len,
2492 if (vcpu->arch.apic &&
2493 !kvm_iodevice_write(&vcpu->arch.apic->dev, addr, len, v))
2496 return kvm_io_bus_write(&vcpu->kvm->mmio_bus, addr, len, v);
2499 static int vcpu_mmio_read(struct kvm_vcpu *vcpu, gpa_t addr, int len, void *v)
2501 if (vcpu->arch.apic &&
2502 !kvm_iodevice_read(&vcpu->arch.apic->dev, addr, len, v))
2505 return kvm_io_bus_read(&vcpu->kvm->mmio_bus, addr, len, v);
2508 static int kvm_read_guest_virt(gva_t addr, void *val, unsigned int bytes,
2509 struct kvm_vcpu *vcpu)
2512 int r = X86EMUL_CONTINUE;
2515 gpa_t gpa = vcpu->arch.mmu.gva_to_gpa(vcpu, addr);
2516 unsigned offset = addr & (PAGE_SIZE-1);
2517 unsigned toread = min(bytes, (unsigned)PAGE_SIZE - offset);
2520 if (gpa == UNMAPPED_GVA) {
2521 r = X86EMUL_PROPAGATE_FAULT;
2524 ret = kvm_read_guest(vcpu->kvm, gpa, data, toread);
2526 r = X86EMUL_UNHANDLEABLE;
2538 static int kvm_write_guest_virt(gva_t addr, void *val, unsigned int bytes,
2539 struct kvm_vcpu *vcpu)
2542 int r = X86EMUL_CONTINUE;
2545 gpa_t gpa = vcpu->arch.mmu.gva_to_gpa(vcpu, addr);
2546 unsigned offset = addr & (PAGE_SIZE-1);
2547 unsigned towrite = min(bytes, (unsigned)PAGE_SIZE - offset);
2550 if (gpa == UNMAPPED_GVA) {
2551 r = X86EMUL_PROPAGATE_FAULT;
2554 ret = kvm_write_guest(vcpu->kvm, gpa, data, towrite);
2556 r = X86EMUL_UNHANDLEABLE;
2569 static int emulator_read_emulated(unsigned long addr,
2572 struct kvm_vcpu *vcpu)
2576 if (vcpu->mmio_read_completed) {
2577 memcpy(val, vcpu->mmio_data, bytes);
2578 trace_kvm_mmio(KVM_TRACE_MMIO_READ, bytes,
2579 vcpu->mmio_phys_addr, *(u64 *)val);
2580 vcpu->mmio_read_completed = 0;
2581 return X86EMUL_CONTINUE;
2584 gpa = vcpu->arch.mmu.gva_to_gpa(vcpu, addr);
2586 /* For APIC access vmexit */
2587 if ((gpa & PAGE_MASK) == APIC_DEFAULT_PHYS_BASE)
2590 if (kvm_read_guest_virt(addr, val, bytes, vcpu)
2591 == X86EMUL_CONTINUE)
2592 return X86EMUL_CONTINUE;
2593 if (gpa == UNMAPPED_GVA)
2594 return X86EMUL_PROPAGATE_FAULT;
2598 * Is this MMIO handled locally?
2600 if (!vcpu_mmio_read(vcpu, gpa, bytes, val)) {
2601 trace_kvm_mmio(KVM_TRACE_MMIO_READ, bytes, gpa, *(u64 *)val);
2602 return X86EMUL_CONTINUE;
2605 trace_kvm_mmio(KVM_TRACE_MMIO_READ_UNSATISFIED, bytes, gpa, 0);
2607 vcpu->mmio_needed = 1;
2608 vcpu->mmio_phys_addr = gpa;
2609 vcpu->mmio_size = bytes;
2610 vcpu->mmio_is_write = 0;
2612 return X86EMUL_UNHANDLEABLE;
2615 int emulator_write_phys(struct kvm_vcpu *vcpu, gpa_t gpa,
2616 const void *val, int bytes)
2620 ret = kvm_write_guest(vcpu->kvm, gpa, val, bytes);
2623 kvm_mmu_pte_write(vcpu, gpa, val, bytes, 1);
2627 static int emulator_write_emulated_onepage(unsigned long addr,
2630 struct kvm_vcpu *vcpu)
2634 gpa = vcpu->arch.mmu.gva_to_gpa(vcpu, addr);
2636 if (gpa == UNMAPPED_GVA) {
2637 kvm_inject_page_fault(vcpu, addr, 2);
2638 return X86EMUL_PROPAGATE_FAULT;
2641 /* For APIC access vmexit */
2642 if ((gpa & PAGE_MASK) == APIC_DEFAULT_PHYS_BASE)
2645 if (emulator_write_phys(vcpu, gpa, val, bytes))
2646 return X86EMUL_CONTINUE;
2649 trace_kvm_mmio(KVM_TRACE_MMIO_WRITE, bytes, gpa, *(u64 *)val);
2651 * Is this MMIO handled locally?
2653 if (!vcpu_mmio_write(vcpu, gpa, bytes, val))
2654 return X86EMUL_CONTINUE;
2656 vcpu->mmio_needed = 1;
2657 vcpu->mmio_phys_addr = gpa;
2658 vcpu->mmio_size = bytes;
2659 vcpu->mmio_is_write = 1;
2660 memcpy(vcpu->mmio_data, val, bytes);
2662 return X86EMUL_CONTINUE;
2665 int emulator_write_emulated(unsigned long addr,
2668 struct kvm_vcpu *vcpu)
2670 /* Crossing a page boundary? */
2671 if (((addr + bytes - 1) ^ addr) & PAGE_MASK) {
2674 now = -addr & ~PAGE_MASK;
2675 rc = emulator_write_emulated_onepage(addr, val, now, vcpu);
2676 if (rc != X86EMUL_CONTINUE)
2682 return emulator_write_emulated_onepage(addr, val, bytes, vcpu);
2684 EXPORT_SYMBOL_GPL(emulator_write_emulated);
2686 static int emulator_cmpxchg_emulated(unsigned long addr,
2690 struct kvm_vcpu *vcpu)
2692 printk_once(KERN_WARNING "kvm: emulating exchange as write\n");
2693 #ifndef CONFIG_X86_64
2694 /* guests cmpxchg8b have to be emulated atomically */
2701 gpa = vcpu->arch.mmu.gva_to_gpa(vcpu, addr);
2703 if (gpa == UNMAPPED_GVA ||
2704 (gpa & PAGE_MASK) == APIC_DEFAULT_PHYS_BASE)
2707 if (((gpa + bytes - 1) & PAGE_MASK) != (gpa & PAGE_MASK))
2712 page = gfn_to_page(vcpu->kvm, gpa >> PAGE_SHIFT);
2714 kaddr = kmap_atomic(page, KM_USER0);
2715 set_64bit((u64 *)(kaddr + offset_in_page(gpa)), val);
2716 kunmap_atomic(kaddr, KM_USER0);
2717 kvm_release_page_dirty(page);
2722 return emulator_write_emulated(addr, new, bytes, vcpu);
2725 static unsigned long get_segment_base(struct kvm_vcpu *vcpu, int seg)
2727 return kvm_x86_ops->get_segment_base(vcpu, seg);
2730 int emulate_invlpg(struct kvm_vcpu *vcpu, gva_t address)
2732 kvm_mmu_invlpg(vcpu, address);
2733 return X86EMUL_CONTINUE;
2736 int emulate_clts(struct kvm_vcpu *vcpu)
2738 kvm_x86_ops->set_cr0(vcpu, vcpu->arch.cr0 & ~X86_CR0_TS);
2739 return X86EMUL_CONTINUE;
2742 int emulator_get_dr(struct x86_emulate_ctxt *ctxt, int dr, unsigned long *dest)
2744 struct kvm_vcpu *vcpu = ctxt->vcpu;
2748 *dest = kvm_x86_ops->get_dr(vcpu, dr);
2749 return X86EMUL_CONTINUE;
2751 pr_unimpl(vcpu, "%s: unexpected dr %u\n", __func__, dr);
2752 return X86EMUL_UNHANDLEABLE;
2756 int emulator_set_dr(struct x86_emulate_ctxt *ctxt, int dr, unsigned long value)
2758 unsigned long mask = (ctxt->mode == X86EMUL_MODE_PROT64) ? ~0ULL : ~0U;
2761 kvm_x86_ops->set_dr(ctxt->vcpu, dr, value & mask, &exception);
2763 /* FIXME: better handling */
2764 return X86EMUL_UNHANDLEABLE;
2766 return X86EMUL_CONTINUE;
2769 void kvm_report_emulation_failure(struct kvm_vcpu *vcpu, const char *context)
2772 unsigned long rip = kvm_rip_read(vcpu);
2773 unsigned long rip_linear;
2775 if (!printk_ratelimit())
2778 rip_linear = rip + get_segment_base(vcpu, VCPU_SREG_CS);
2780 kvm_read_guest_virt(rip_linear, (void *)opcodes, 4, vcpu);
2782 printk(KERN_ERR "emulation failed (%s) rip %lx %02x %02x %02x %02x\n",
2783 context, rip, opcodes[0], opcodes[1], opcodes[2], opcodes[3]);
2785 EXPORT_SYMBOL_GPL(kvm_report_emulation_failure);
2787 static struct x86_emulate_ops emulate_ops = {
2788 .read_std = kvm_read_guest_virt,
2789 .read_emulated = emulator_read_emulated,
2790 .write_emulated = emulator_write_emulated,
2791 .cmpxchg_emulated = emulator_cmpxchg_emulated,
2794 static void cache_all_regs(struct kvm_vcpu *vcpu)
2796 kvm_register_read(vcpu, VCPU_REGS_RAX);
2797 kvm_register_read(vcpu, VCPU_REGS_RSP);
2798 kvm_register_read(vcpu, VCPU_REGS_RIP);
2799 vcpu->arch.regs_dirty = ~0;
2802 int emulate_instruction(struct kvm_vcpu *vcpu,
2803 struct kvm_run *run,
2809 struct decode_cache *c;
2811 kvm_clear_exception_queue(vcpu);
2812 vcpu->arch.mmio_fault_cr2 = cr2;
2814 * TODO: fix emulate.c to use guest_read/write_register
2815 * instead of direct ->regs accesses, can save hundred cycles
2816 * on Intel for instructions that don't read/change RSP, for
2819 cache_all_regs(vcpu);
2821 vcpu->mmio_is_write = 0;
2822 vcpu->arch.pio.string = 0;
2824 if (!(emulation_type & EMULTYPE_NO_DECODE)) {
2826 kvm_x86_ops->get_cs_db_l_bits(vcpu, &cs_db, &cs_l);
2828 vcpu->arch.emulate_ctxt.vcpu = vcpu;
2829 vcpu->arch.emulate_ctxt.eflags = kvm_x86_ops->get_rflags(vcpu);
2830 vcpu->arch.emulate_ctxt.mode =
2831 (vcpu->arch.emulate_ctxt.eflags & X86_EFLAGS_VM)
2832 ? X86EMUL_MODE_REAL : cs_l
2833 ? X86EMUL_MODE_PROT64 : cs_db
2834 ? X86EMUL_MODE_PROT32 : X86EMUL_MODE_PROT16;
2836 r = x86_decode_insn(&vcpu->arch.emulate_ctxt, &emulate_ops);
2838 /* Only allow emulation of specific instructions on #UD
2839 * (namely VMMCALL, sysenter, sysexit, syscall)*/
2840 c = &vcpu->arch.emulate_ctxt.decode;
2841 if (emulation_type & EMULTYPE_TRAP_UD) {
2843 return EMULATE_FAIL;
2845 case 0x01: /* VMMCALL */
2846 if (c->modrm_mod != 3 || c->modrm_rm != 1)
2847 return EMULATE_FAIL;
2849 case 0x34: /* sysenter */
2850 case 0x35: /* sysexit */
2851 if (c->modrm_mod != 0 || c->modrm_rm != 0)
2852 return EMULATE_FAIL;
2854 case 0x05: /* syscall */
2855 if (c->modrm_mod != 0 || c->modrm_rm != 0)
2856 return EMULATE_FAIL;
2859 return EMULATE_FAIL;
2862 if (!(c->modrm_reg == 0 || c->modrm_reg == 3))
2863 return EMULATE_FAIL;
2866 ++vcpu->stat.insn_emulation;
2868 ++vcpu->stat.insn_emulation_fail;
2869 if (kvm_mmu_unprotect_page_virt(vcpu, cr2))
2870 return EMULATE_DONE;
2871 return EMULATE_FAIL;
2875 if (emulation_type & EMULTYPE_SKIP) {
2876 kvm_rip_write(vcpu, vcpu->arch.emulate_ctxt.decode.eip);
2877 return EMULATE_DONE;
2880 r = x86_emulate_insn(&vcpu->arch.emulate_ctxt, &emulate_ops);
2881 shadow_mask = vcpu->arch.emulate_ctxt.interruptibility;
2884 kvm_x86_ops->set_interrupt_shadow(vcpu, shadow_mask);
2886 if (vcpu->arch.pio.string)
2887 return EMULATE_DO_MMIO;
2889 if ((r || vcpu->mmio_is_write) && run) {
2890 run->exit_reason = KVM_EXIT_MMIO;
2891 run->mmio.phys_addr = vcpu->mmio_phys_addr;
2892 memcpy(run->mmio.data, vcpu->mmio_data, 8);
2893 run->mmio.len = vcpu->mmio_size;
2894 run->mmio.is_write = vcpu->mmio_is_write;
2898 if (kvm_mmu_unprotect_page_virt(vcpu, cr2))
2899 return EMULATE_DONE;
2900 if (!vcpu->mmio_needed) {
2901 kvm_report_emulation_failure(vcpu, "mmio");
2902 return EMULATE_FAIL;
2904 return EMULATE_DO_MMIO;
2907 kvm_x86_ops->set_rflags(vcpu, vcpu->arch.emulate_ctxt.eflags);
2909 if (vcpu->mmio_is_write) {
2910 vcpu->mmio_needed = 0;
2911 return EMULATE_DO_MMIO;
2914 return EMULATE_DONE;
2916 EXPORT_SYMBOL_GPL(emulate_instruction);
2918 static int pio_copy_data(struct kvm_vcpu *vcpu)
2920 void *p = vcpu->arch.pio_data;
2921 gva_t q = vcpu->arch.pio.guest_gva;
2925 bytes = vcpu->arch.pio.size * vcpu->arch.pio.cur_count;
2926 if (vcpu->arch.pio.in)
2927 ret = kvm_write_guest_virt(q, p, bytes, vcpu);
2929 ret = kvm_read_guest_virt(q, p, bytes, vcpu);
2933 int complete_pio(struct kvm_vcpu *vcpu)
2935 struct kvm_pio_request *io = &vcpu->arch.pio;
2942 val = kvm_register_read(vcpu, VCPU_REGS_RAX);
2943 memcpy(&val, vcpu->arch.pio_data, io->size);
2944 kvm_register_write(vcpu, VCPU_REGS_RAX, val);
2948 r = pio_copy_data(vcpu);
2955 delta *= io->cur_count;
2957 * The size of the register should really depend on
2958 * current address size.
2960 val = kvm_register_read(vcpu, VCPU_REGS_RCX);
2962 kvm_register_write(vcpu, VCPU_REGS_RCX, val);
2968 val = kvm_register_read(vcpu, VCPU_REGS_RDI);
2970 kvm_register_write(vcpu, VCPU_REGS_RDI, val);
2972 val = kvm_register_read(vcpu, VCPU_REGS_RSI);
2974 kvm_register_write(vcpu, VCPU_REGS_RSI, val);
2978 io->count -= io->cur_count;
2984 static int kernel_pio(struct kvm_vcpu *vcpu, void *pd)
2986 /* TODO: String I/O for in kernel device */
2989 if (vcpu->arch.pio.in)
2990 r = kvm_io_bus_read(&vcpu->kvm->pio_bus, vcpu->arch.pio.port,
2991 vcpu->arch.pio.size, pd);
2993 r = kvm_io_bus_write(&vcpu->kvm->pio_bus, vcpu->arch.pio.port,
2994 vcpu->arch.pio.size, pd);
2998 static int pio_string_write(struct kvm_vcpu *vcpu)
3000 struct kvm_pio_request *io = &vcpu->arch.pio;
3001 void *pd = vcpu->arch.pio_data;
3004 for (i = 0; i < io->cur_count; i++) {
3005 if (kvm_io_bus_write(&vcpu->kvm->pio_bus,
3006 io->port, io->size, pd)) {
3015 int kvm_emulate_pio(struct kvm_vcpu *vcpu, struct kvm_run *run, int in,
3016 int size, unsigned port)
3020 vcpu->run->exit_reason = KVM_EXIT_IO;
3021 vcpu->run->io.direction = in ? KVM_EXIT_IO_IN : KVM_EXIT_IO_OUT;
3022 vcpu->run->io.size = vcpu->arch.pio.size = size;
3023 vcpu->run->io.data_offset = KVM_PIO_PAGE_OFFSET * PAGE_SIZE;
3024 vcpu->run->io.count = vcpu->arch.pio.count = vcpu->arch.pio.cur_count = 1;
3025 vcpu->run->io.port = vcpu->arch.pio.port = port;
3026 vcpu->arch.pio.in = in;
3027 vcpu->arch.pio.string = 0;
3028 vcpu->arch.pio.down = 0;
3029 vcpu->arch.pio.rep = 0;
3031 trace_kvm_pio(vcpu->run->io.direction == KVM_EXIT_IO_OUT, port,
3034 val = kvm_register_read(vcpu, VCPU_REGS_RAX);
3035 memcpy(vcpu->arch.pio_data, &val, 4);
3037 if (!kernel_pio(vcpu, vcpu->arch.pio_data)) {
3043 EXPORT_SYMBOL_GPL(kvm_emulate_pio);
3045 int kvm_emulate_pio_string(struct kvm_vcpu *vcpu, struct kvm_run *run, int in,
3046 int size, unsigned long count, int down,
3047 gva_t address, int rep, unsigned port)
3049 unsigned now, in_page;
3052 vcpu->run->exit_reason = KVM_EXIT_IO;
3053 vcpu->run->io.direction = in ? KVM_EXIT_IO_IN : KVM_EXIT_IO_OUT;
3054 vcpu->run->io.size = vcpu->arch.pio.size = size;
3055 vcpu->run->io.data_offset = KVM_PIO_PAGE_OFFSET * PAGE_SIZE;
3056 vcpu->run->io.count = vcpu->arch.pio.count = vcpu->arch.pio.cur_count = count;
3057 vcpu->run->io.port = vcpu->arch.pio.port = port;
3058 vcpu->arch.pio.in = in;
3059 vcpu->arch.pio.string = 1;
3060 vcpu->arch.pio.down = down;
3061 vcpu->arch.pio.rep = rep;
3063 trace_kvm_pio(vcpu->run->io.direction == KVM_EXIT_IO_OUT, port,
3067 kvm_x86_ops->skip_emulated_instruction(vcpu);
3072 in_page = PAGE_SIZE - offset_in_page(address);
3074 in_page = offset_in_page(address) + size;
3075 now = min(count, (unsigned long)in_page / size);
3080 * String I/O in reverse. Yuck. Kill the guest, fix later.
3082 pr_unimpl(vcpu, "guest string pio down\n");
3083 kvm_inject_gp(vcpu, 0);
3086 vcpu->run->io.count = now;
3087 vcpu->arch.pio.cur_count = now;
3089 if (vcpu->arch.pio.cur_count == vcpu->arch.pio.count)
3090 kvm_x86_ops->skip_emulated_instruction(vcpu);
3092 vcpu->arch.pio.guest_gva = address;
3094 if (!vcpu->arch.pio.in) {
3095 /* string PIO write */
3096 ret = pio_copy_data(vcpu);
3097 if (ret == X86EMUL_PROPAGATE_FAULT) {
3098 kvm_inject_gp(vcpu, 0);
3101 if (ret == 0 && !pio_string_write(vcpu)) {
3103 if (vcpu->arch.pio.count == 0)
3107 /* no string PIO read support yet */
3111 EXPORT_SYMBOL_GPL(kvm_emulate_pio_string);
3113 static void bounce_off(void *info)
3118 static unsigned int ref_freq;
3119 static unsigned long tsc_khz_ref;
3121 static int kvmclock_cpufreq_notifier(struct notifier_block *nb, unsigned long val,
3124 struct cpufreq_freqs *freq = data;
3126 struct kvm_vcpu *vcpu;
3127 int i, send_ipi = 0;
3130 ref_freq = freq->old;
3132 if (val == CPUFREQ_PRECHANGE && freq->old > freq->new)
3134 if (val == CPUFREQ_POSTCHANGE && freq->old < freq->new)
3136 per_cpu(cpu_tsc_khz, freq->cpu) = cpufreq_scale(tsc_khz_ref, ref_freq, freq->new);
3138 spin_lock(&kvm_lock);
3139 list_for_each_entry(kvm, &vm_list, vm_list) {
3140 kvm_for_each_vcpu(i, vcpu, kvm) {
3141 if (vcpu->cpu != freq->cpu)
3143 if (!kvm_request_guest_time_update(vcpu))
3145 if (vcpu->cpu != smp_processor_id())
3149 spin_unlock(&kvm_lock);
3151 if (freq->old < freq->new && send_ipi) {
3153 * We upscale the frequency. Must make the guest
3154 * doesn't see old kvmclock values while running with
3155 * the new frequency, otherwise we risk the guest sees
3156 * time go backwards.
3158 * In case we update the frequency for another cpu
3159 * (which might be in guest context) send an interrupt
3160 * to kick the cpu out of guest context. Next time
3161 * guest context is entered kvmclock will be updated,
3162 * so the guest will not see stale values.
3164 smp_call_function_single(freq->cpu, bounce_off, NULL, 1);
3169 static struct notifier_block kvmclock_cpufreq_notifier_block = {
3170 .notifier_call = kvmclock_cpufreq_notifier
3173 int kvm_arch_init(void *opaque)
3176 struct kvm_x86_ops *ops = (struct kvm_x86_ops *)opaque;
3179 printk(KERN_ERR "kvm: already loaded the other module\n");
3184 if (!ops->cpu_has_kvm_support()) {
3185 printk(KERN_ERR "kvm: no hardware support\n");
3189 if (ops->disabled_by_bios()) {
3190 printk(KERN_ERR "kvm: disabled by bios\n");
3195 r = kvm_mmu_module_init();
3199 kvm_init_msr_list();
3202 kvm_mmu_set_nonpresent_ptes(0ull, 0ull);
3203 kvm_mmu_set_base_ptes(PT_PRESENT_MASK);
3204 kvm_mmu_set_mask_ptes(PT_USER_MASK, PT_ACCESSED_MASK,
3205 PT_DIRTY_MASK, PT64_NX_MASK, 0);
3207 for_each_possible_cpu(cpu)
3208 per_cpu(cpu_tsc_khz, cpu) = tsc_khz;
3209 if (!boot_cpu_has(X86_FEATURE_CONSTANT_TSC)) {
3210 tsc_khz_ref = tsc_khz;
3211 cpufreq_register_notifier(&kvmclock_cpufreq_notifier_block,
3212 CPUFREQ_TRANSITION_NOTIFIER);
3221 void kvm_arch_exit(void)
3223 if (!boot_cpu_has(X86_FEATURE_CONSTANT_TSC))
3224 cpufreq_unregister_notifier(&kvmclock_cpufreq_notifier_block,
3225 CPUFREQ_TRANSITION_NOTIFIER);
3227 kvm_mmu_module_exit();
3230 int kvm_emulate_halt(struct kvm_vcpu *vcpu)
3232 ++vcpu->stat.halt_exits;
3233 if (irqchip_in_kernel(vcpu->kvm)) {
3234 vcpu->arch.mp_state = KVM_MP_STATE_HALTED;
3237 vcpu->run->exit_reason = KVM_EXIT_HLT;
3241 EXPORT_SYMBOL_GPL(kvm_emulate_halt);
3243 static inline gpa_t hc_gpa(struct kvm_vcpu *vcpu, unsigned long a0,
3246 if (is_long_mode(vcpu))
3249 return a0 | ((gpa_t)a1 << 32);
3252 int kvm_emulate_hypercall(struct kvm_vcpu *vcpu)
3254 unsigned long nr, a0, a1, a2, a3, ret;
3257 nr = kvm_register_read(vcpu, VCPU_REGS_RAX);
3258 a0 = kvm_register_read(vcpu, VCPU_REGS_RBX);
3259 a1 = kvm_register_read(vcpu, VCPU_REGS_RCX);
3260 a2 = kvm_register_read(vcpu, VCPU_REGS_RDX);
3261 a3 = kvm_register_read(vcpu, VCPU_REGS_RSI);
3263 trace_kvm_hypercall(nr, a0, a1, a2, a3);
3265 if (!is_long_mode(vcpu)) {
3273 if (kvm_x86_ops->get_cpl(vcpu) != 0) {
3279 case KVM_HC_VAPIC_POLL_IRQ:
3283 r = kvm_pv_mmu_op(vcpu, a0, hc_gpa(vcpu, a1, a2), &ret);
3290 kvm_register_write(vcpu, VCPU_REGS_RAX, ret);
3291 ++vcpu->stat.hypercalls;
3294 EXPORT_SYMBOL_GPL(kvm_emulate_hypercall);
3296 int kvm_fix_hypercall(struct kvm_vcpu *vcpu)
3298 char instruction[3];
3300 unsigned long rip = kvm_rip_read(vcpu);
3304 * Blow out the MMU to ensure that no other VCPU has an active mapping
3305 * to ensure that the updated hypercall appears atomically across all
3308 kvm_mmu_zap_all(vcpu->kvm);
3310 kvm_x86_ops->patch_hypercall(vcpu, instruction);
3311 if (emulator_write_emulated(rip, instruction, 3, vcpu)
3312 != X86EMUL_CONTINUE)
3318 static u64 mk_cr_64(u64 curr_cr, u32 new_val)
3320 return (curr_cr & ~((1ULL << 32) - 1)) | new_val;
3323 void realmode_lgdt(struct kvm_vcpu *vcpu, u16 limit, unsigned long base)
3325 struct descriptor_table dt = { limit, base };
3327 kvm_x86_ops->set_gdt(vcpu, &dt);
3330 void realmode_lidt(struct kvm_vcpu *vcpu, u16 limit, unsigned long base)
3332 struct descriptor_table dt = { limit, base };
3334 kvm_x86_ops->set_idt(vcpu, &dt);
3337 void realmode_lmsw(struct kvm_vcpu *vcpu, unsigned long msw,
3338 unsigned long *rflags)
3340 kvm_lmsw(vcpu, msw);
3341 *rflags = kvm_x86_ops->get_rflags(vcpu);
3344 unsigned long realmode_get_cr(struct kvm_vcpu *vcpu, int cr)
3346 unsigned long value;
3348 kvm_x86_ops->decache_cr4_guest_bits(vcpu);
3351 value = vcpu->arch.cr0;
3354 value = vcpu->arch.cr2;
3357 value = vcpu->arch.cr3;
3360 value = vcpu->arch.cr4;
3363 value = kvm_get_cr8(vcpu);
3366 vcpu_printf(vcpu, "%s: unexpected cr %u\n", __func__, cr);
3373 void realmode_set_cr(struct kvm_vcpu *vcpu, int cr, unsigned long val,
3374 unsigned long *rflags)
3378 kvm_set_cr0(vcpu, mk_cr_64(vcpu->arch.cr0, val));
3379 *rflags = kvm_x86_ops->get_rflags(vcpu);
3382 vcpu->arch.cr2 = val;
3385 kvm_set_cr3(vcpu, val);
3388 kvm_set_cr4(vcpu, mk_cr_64(vcpu->arch.cr4, val));
3391 kvm_set_cr8(vcpu, val & 0xfUL);
3394 vcpu_printf(vcpu, "%s: unexpected cr %u\n", __func__, cr);
3398 static int move_to_next_stateful_cpuid_entry(struct kvm_vcpu *vcpu, int i)
3400 struct kvm_cpuid_entry2 *e = &vcpu->arch.cpuid_entries[i];
3401 int j, nent = vcpu->arch.cpuid_nent;
3403 e->flags &= ~KVM_CPUID_FLAG_STATE_READ_NEXT;
3404 /* when no next entry is found, the current entry[i] is reselected */
3405 for (j = i + 1; ; j = (j + 1) % nent) {
3406 struct kvm_cpuid_entry2 *ej = &vcpu->arch.cpuid_entries[j];
3407 if (ej->function == e->function) {
3408 ej->flags |= KVM_CPUID_FLAG_STATE_READ_NEXT;
3412 return 0; /* silence gcc, even though control never reaches here */
3415 /* find an entry with matching function, matching index (if needed), and that
3416 * should be read next (if it's stateful) */
3417 static int is_matching_cpuid_entry(struct kvm_cpuid_entry2 *e,
3418 u32 function, u32 index)
3420 if (e->function != function)
3422 if ((e->flags & KVM_CPUID_FLAG_SIGNIFCANT_INDEX) && e->index != index)
3424 if ((e->flags & KVM_CPUID_FLAG_STATEFUL_FUNC) &&
3425 !(e->flags & KVM_CPUID_FLAG_STATE_READ_NEXT))
3430 struct kvm_cpuid_entry2 *kvm_find_cpuid_entry(struct kvm_vcpu *vcpu,
3431 u32 function, u32 index)
3434 struct kvm_cpuid_entry2 *best = NULL;
3436 for (i = 0; i < vcpu->arch.cpuid_nent; ++i) {
3437 struct kvm_cpuid_entry2 *e;
3439 e = &vcpu->arch.cpuid_entries[i];
3440 if (is_matching_cpuid_entry(e, function, index)) {
3441 if (e->flags & KVM_CPUID_FLAG_STATEFUL_FUNC)
3442 move_to_next_stateful_cpuid_entry(vcpu, i);
3447 * Both basic or both extended?
3449 if (((e->function ^ function) & 0x80000000) == 0)
3450 if (!best || e->function > best->function)
3456 int cpuid_maxphyaddr(struct kvm_vcpu *vcpu)
3458 struct kvm_cpuid_entry2 *best;
3460 best = kvm_find_cpuid_entry(vcpu, 0x80000008, 0);
3462 return best->eax & 0xff;
3466 void kvm_emulate_cpuid(struct kvm_vcpu *vcpu)
3468 u32 function, index;
3469 struct kvm_cpuid_entry2 *best;
3471 function = kvm_register_read(vcpu, VCPU_REGS_RAX);
3472 index = kvm_register_read(vcpu, VCPU_REGS_RCX);
3473 kvm_register_write(vcpu, VCPU_REGS_RAX, 0);
3474 kvm_register_write(vcpu, VCPU_REGS_RBX, 0);
3475 kvm_register_write(vcpu, VCPU_REGS_RCX, 0);
3476 kvm_register_write(vcpu, VCPU_REGS_RDX, 0);
3477 best = kvm_find_cpuid_entry(vcpu, function, index);
3479 kvm_register_write(vcpu, VCPU_REGS_RAX, best->eax);
3480 kvm_register_write(vcpu, VCPU_REGS_RBX, best->ebx);
3481 kvm_register_write(vcpu, VCPU_REGS_RCX, best->ecx);
3482 kvm_register_write(vcpu, VCPU_REGS_RDX, best->edx);
3484 kvm_x86_ops->skip_emulated_instruction(vcpu);
3485 trace_kvm_cpuid(function,
3486 kvm_register_read(vcpu, VCPU_REGS_RAX),
3487 kvm_register_read(vcpu, VCPU_REGS_RBX),
3488 kvm_register_read(vcpu, VCPU_REGS_RCX),
3489 kvm_register_read(vcpu, VCPU_REGS_RDX));
3491 EXPORT_SYMBOL_GPL(kvm_emulate_cpuid);
3494 * Check if userspace requested an interrupt window, and that the
3495 * interrupt window is open.
3497 * No need to exit to userspace if we already have an interrupt queued.
3499 static int dm_request_for_irq_injection(struct kvm_vcpu *vcpu,
3500 struct kvm_run *kvm_run)
3502 return (!irqchip_in_kernel(vcpu->kvm) && !kvm_cpu_has_interrupt(vcpu) &&
3503 kvm_run->request_interrupt_window &&
3504 kvm_arch_interrupt_allowed(vcpu));
3507 static void post_kvm_run_save(struct kvm_vcpu *vcpu,
3508 struct kvm_run *kvm_run)
3510 kvm_run->if_flag = (kvm_x86_ops->get_rflags(vcpu) & X86_EFLAGS_IF) != 0;
3511 kvm_run->cr8 = kvm_get_cr8(vcpu);
3512 kvm_run->apic_base = kvm_get_apic_base(vcpu);
3513 if (irqchip_in_kernel(vcpu->kvm))
3514 kvm_run->ready_for_interrupt_injection = 1;
3516 kvm_run->ready_for_interrupt_injection =
3517 kvm_arch_interrupt_allowed(vcpu) &&
3518 !kvm_cpu_has_interrupt(vcpu) &&
3519 !kvm_event_needs_reinjection(vcpu);
3522 static void vapic_enter(struct kvm_vcpu *vcpu)
3524 struct kvm_lapic *apic = vcpu->arch.apic;
3527 if (!apic || !apic->vapic_addr)
3530 page = gfn_to_page(vcpu->kvm, apic->vapic_addr >> PAGE_SHIFT);
3532 vcpu->arch.apic->vapic_page = page;
3535 static void vapic_exit(struct kvm_vcpu *vcpu)
3537 struct kvm_lapic *apic = vcpu->arch.apic;
3539 if (!apic || !apic->vapic_addr)
3542 down_read(&vcpu->kvm->slots_lock);
3543 kvm_release_page_dirty(apic->vapic_page);
3544 mark_page_dirty(vcpu->kvm, apic->vapic_addr >> PAGE_SHIFT);
3545 up_read(&vcpu->kvm->slots_lock);
3548 static void update_cr8_intercept(struct kvm_vcpu *vcpu)
3552 if (!kvm_x86_ops->update_cr8_intercept)
3555 if (!vcpu->arch.apic)
3558 if (!vcpu->arch.apic->vapic_addr)
3559 max_irr = kvm_lapic_find_highest_irr(vcpu);
3566 tpr = kvm_lapic_get_cr8(vcpu);
3568 kvm_x86_ops->update_cr8_intercept(vcpu, tpr, max_irr);
3571 static void inject_pending_event(struct kvm_vcpu *vcpu, struct kvm_run *kvm_run)
3573 /* try to reinject previous events if any */
3574 if (vcpu->arch.exception.pending) {
3575 kvm_x86_ops->queue_exception(vcpu, vcpu->arch.exception.nr,
3576 vcpu->arch.exception.has_error_code,
3577 vcpu->arch.exception.error_code);
3581 if (vcpu->arch.nmi_injected) {
3582 kvm_x86_ops->set_nmi(vcpu);
3586 if (vcpu->arch.interrupt.pending) {
3587 kvm_x86_ops->set_irq(vcpu);
3591 /* try to inject new event if pending */
3592 if (vcpu->arch.nmi_pending) {
3593 if (kvm_x86_ops->nmi_allowed(vcpu)) {
3594 vcpu->arch.nmi_pending = false;
3595 vcpu->arch.nmi_injected = true;
3596 kvm_x86_ops->set_nmi(vcpu);
3598 } else if (kvm_cpu_has_interrupt(vcpu)) {
3599 if (kvm_x86_ops->interrupt_allowed(vcpu)) {
3600 kvm_queue_interrupt(vcpu, kvm_cpu_get_interrupt(vcpu),
3602 kvm_x86_ops->set_irq(vcpu);
3607 static int vcpu_enter_guest(struct kvm_vcpu *vcpu, struct kvm_run *kvm_run)
3610 bool req_int_win = !irqchip_in_kernel(vcpu->kvm) &&
3611 kvm_run->request_interrupt_window;
3614 if (test_and_clear_bit(KVM_REQ_MMU_RELOAD, &vcpu->requests))
3615 kvm_mmu_unload(vcpu);
3617 r = kvm_mmu_reload(vcpu);
3621 if (vcpu->requests) {
3622 if (test_and_clear_bit(KVM_REQ_MIGRATE_TIMER, &vcpu->requests))
3623 __kvm_migrate_timers(vcpu);
3624 if (test_and_clear_bit(KVM_REQ_KVMCLOCK_UPDATE, &vcpu->requests))
3625 kvm_write_guest_time(vcpu);
3626 if (test_and_clear_bit(KVM_REQ_MMU_SYNC, &vcpu->requests))
3627 kvm_mmu_sync_roots(vcpu);
3628 if (test_and_clear_bit(KVM_REQ_TLB_FLUSH, &vcpu->requests))
3629 kvm_x86_ops->tlb_flush(vcpu);
3630 if (test_and_clear_bit(KVM_REQ_REPORT_TPR_ACCESS,
3632 kvm_run->exit_reason = KVM_EXIT_TPR_ACCESS;
3636 if (test_and_clear_bit(KVM_REQ_TRIPLE_FAULT, &vcpu->requests)) {
3637 kvm_run->exit_reason = KVM_EXIT_SHUTDOWN;
3645 kvm_x86_ops->prepare_guest_switch(vcpu);
3646 kvm_load_guest_fpu(vcpu);
3648 local_irq_disable();
3650 clear_bit(KVM_REQ_KICK, &vcpu->requests);
3651 smp_mb__after_clear_bit();
3653 if (vcpu->requests || need_resched() || signal_pending(current)) {
3654 set_bit(KVM_REQ_KICK, &vcpu->requests);
3661 inject_pending_event(vcpu, kvm_run);
3663 /* enable NMI/IRQ window open exits if needed */
3664 if (vcpu->arch.nmi_pending)
3665 kvm_x86_ops->enable_nmi_window(vcpu);
3666 else if (kvm_cpu_has_interrupt(vcpu) || req_int_win)
3667 kvm_x86_ops->enable_irq_window(vcpu);
3669 if (kvm_lapic_enabled(vcpu)) {
3670 update_cr8_intercept(vcpu);
3671 kvm_lapic_sync_to_vapic(vcpu);
3674 up_read(&vcpu->kvm->slots_lock);
3678 if (unlikely(vcpu->arch.switch_db_regs)) {
3680 set_debugreg(vcpu->arch.eff_db[0], 0);
3681 set_debugreg(vcpu->arch.eff_db[1], 1);
3682 set_debugreg(vcpu->arch.eff_db[2], 2);
3683 set_debugreg(vcpu->arch.eff_db[3], 3);
3686 trace_kvm_entry(vcpu->vcpu_id);
3687 kvm_x86_ops->run(vcpu, kvm_run);
3689 if (unlikely(vcpu->arch.switch_db_regs || test_thread_flag(TIF_DEBUG))) {
3690 set_debugreg(current->thread.debugreg0, 0);
3691 set_debugreg(current->thread.debugreg1, 1);
3692 set_debugreg(current->thread.debugreg2, 2);
3693 set_debugreg(current->thread.debugreg3, 3);
3694 set_debugreg(current->thread.debugreg6, 6);
3695 set_debugreg(current->thread.debugreg7, 7);
3698 set_bit(KVM_REQ_KICK, &vcpu->requests);
3704 * We must have an instruction between local_irq_enable() and
3705 * kvm_guest_exit(), so the timer interrupt isn't delayed by
3706 * the interrupt shadow. The stat.exits increment will do nicely.
3707 * But we need to prevent reordering, hence this barrier():
3715 down_read(&vcpu->kvm->slots_lock);
3718 * Profile KVM exit RIPs:
3720 if (unlikely(prof_on == KVM_PROFILING)) {
3721 unsigned long rip = kvm_rip_read(vcpu);
3722 profile_hit(KVM_PROFILING, (void *)rip);
3726 kvm_lapic_sync_from_vapic(vcpu);
3728 r = kvm_x86_ops->handle_exit(kvm_run, vcpu);
3734 static int __vcpu_run(struct kvm_vcpu *vcpu, struct kvm_run *kvm_run)
3738 if (unlikely(vcpu->arch.mp_state == KVM_MP_STATE_SIPI_RECEIVED)) {
3739 pr_debug("vcpu %d received sipi with vector # %x\n",
3740 vcpu->vcpu_id, vcpu->arch.sipi_vector);
3741 kvm_lapic_reset(vcpu);
3742 r = kvm_arch_vcpu_reset(vcpu);
3745 vcpu->arch.mp_state = KVM_MP_STATE_RUNNABLE;
3748 down_read(&vcpu->kvm->slots_lock);
3753 if (vcpu->arch.mp_state == KVM_MP_STATE_RUNNABLE)
3754 r = vcpu_enter_guest(vcpu, kvm_run);
3756 up_read(&vcpu->kvm->slots_lock);
3757 kvm_vcpu_block(vcpu);
3758 down_read(&vcpu->kvm->slots_lock);
3759 if (test_and_clear_bit(KVM_REQ_UNHALT, &vcpu->requests))
3761 switch(vcpu->arch.mp_state) {
3762 case KVM_MP_STATE_HALTED:
3763 vcpu->arch.mp_state =
3764 KVM_MP_STATE_RUNNABLE;
3765 case KVM_MP_STATE_RUNNABLE:
3767 case KVM_MP_STATE_SIPI_RECEIVED:
3778 clear_bit(KVM_REQ_PENDING_TIMER, &vcpu->requests);
3779 if (kvm_cpu_has_pending_timer(vcpu))
3780 kvm_inject_pending_timer_irqs(vcpu);
3782 if (dm_request_for_irq_injection(vcpu, kvm_run)) {
3784 kvm_run->exit_reason = KVM_EXIT_INTR;
3785 ++vcpu->stat.request_irq_exits;
3787 if (signal_pending(current)) {
3789 kvm_run->exit_reason = KVM_EXIT_INTR;
3790 ++vcpu->stat.signal_exits;
3792 if (need_resched()) {
3793 up_read(&vcpu->kvm->slots_lock);
3795 down_read(&vcpu->kvm->slots_lock);
3799 up_read(&vcpu->kvm->slots_lock);
3800 post_kvm_run_save(vcpu, kvm_run);
3807 int kvm_arch_vcpu_ioctl_run(struct kvm_vcpu *vcpu, struct kvm_run *kvm_run)
3814 if (vcpu->sigset_active)
3815 sigprocmask(SIG_SETMASK, &vcpu->sigset, &sigsaved);
3817 if (unlikely(vcpu->arch.mp_state == KVM_MP_STATE_UNINITIALIZED)) {
3818 kvm_vcpu_block(vcpu);
3819 clear_bit(KVM_REQ_UNHALT, &vcpu->requests);
3824 /* re-sync apic's tpr */
3825 if (!irqchip_in_kernel(vcpu->kvm))
3826 kvm_set_cr8(vcpu, kvm_run->cr8);
3828 if (vcpu->arch.pio.cur_count) {
3829 r = complete_pio(vcpu);
3833 #if CONFIG_HAS_IOMEM
3834 if (vcpu->mmio_needed) {
3835 memcpy(vcpu->mmio_data, kvm_run->mmio.data, 8);
3836 vcpu->mmio_read_completed = 1;
3837 vcpu->mmio_needed = 0;
3839 down_read(&vcpu->kvm->slots_lock);
3840 r = emulate_instruction(vcpu, kvm_run,
3841 vcpu->arch.mmio_fault_cr2, 0,
3842 EMULTYPE_NO_DECODE);
3843 up_read(&vcpu->kvm->slots_lock);
3844 if (r == EMULATE_DO_MMIO) {
3846 * Read-modify-write. Back to userspace.
3853 if (kvm_run->exit_reason == KVM_EXIT_HYPERCALL)
3854 kvm_register_write(vcpu, VCPU_REGS_RAX,
3855 kvm_run->hypercall.ret);
3857 r = __vcpu_run(vcpu, kvm_run);
3860 if (vcpu->sigset_active)
3861 sigprocmask(SIG_SETMASK, &sigsaved, NULL);
3867 int kvm_arch_vcpu_ioctl_get_regs(struct kvm_vcpu *vcpu, struct kvm_regs *regs)
3871 regs->rax = kvm_register_read(vcpu, VCPU_REGS_RAX);
3872 regs->rbx = kvm_register_read(vcpu, VCPU_REGS_RBX);
3873 regs->rcx = kvm_register_read(vcpu, VCPU_REGS_RCX);
3874 regs->rdx = kvm_register_read(vcpu, VCPU_REGS_RDX);
3875 regs->rsi = kvm_register_read(vcpu, VCPU_REGS_RSI);
3876 regs->rdi = kvm_register_read(vcpu, VCPU_REGS_RDI);
3877 regs->rsp = kvm_register_read(vcpu, VCPU_REGS_RSP);
3878 regs->rbp = kvm_register_read(vcpu, VCPU_REGS_RBP);
3879 #ifdef CONFIG_X86_64
3880 regs->r8 = kvm_register_read(vcpu, VCPU_REGS_R8);
3881 regs->r9 = kvm_register_read(vcpu, VCPU_REGS_R9);
3882 regs->r10 = kvm_register_read(vcpu, VCPU_REGS_R10);
3883 regs->r11 = kvm_register_read(vcpu, VCPU_REGS_R11);
3884 regs->r12 = kvm_register_read(vcpu, VCPU_REGS_R12);
3885 regs->r13 = kvm_register_read(vcpu, VCPU_REGS_R13);
3886 regs->r14 = kvm_register_read(vcpu, VCPU_REGS_R14);
3887 regs->r15 = kvm_register_read(vcpu, VCPU_REGS_R15);
3890 regs->rip = kvm_rip_read(vcpu);
3891 regs->rflags = kvm_x86_ops->get_rflags(vcpu);
3894 * Don't leak debug flags in case they were set for guest debugging
3896 if (vcpu->guest_debug & KVM_GUESTDBG_SINGLESTEP)
3897 regs->rflags &= ~(X86_EFLAGS_TF | X86_EFLAGS_RF);
3904 int kvm_arch_vcpu_ioctl_set_regs(struct kvm_vcpu *vcpu, struct kvm_regs *regs)
3908 kvm_register_write(vcpu, VCPU_REGS_RAX, regs->rax);
3909 kvm_register_write(vcpu, VCPU_REGS_RBX, regs->rbx);
3910 kvm_register_write(vcpu, VCPU_REGS_RCX, regs->rcx);
3911 kvm_register_write(vcpu, VCPU_REGS_RDX, regs->rdx);
3912 kvm_register_write(vcpu, VCPU_REGS_RSI, regs->rsi);
3913 kvm_register_write(vcpu, VCPU_REGS_RDI, regs->rdi);
3914 kvm_register_write(vcpu, VCPU_REGS_RSP, regs->rsp);
3915 kvm_register_write(vcpu, VCPU_REGS_RBP, regs->rbp);
3916 #ifdef CONFIG_X86_64
3917 kvm_register_write(vcpu, VCPU_REGS_R8, regs->r8);
3918 kvm_register_write(vcpu, VCPU_REGS_R9, regs->r9);
3919 kvm_register_write(vcpu, VCPU_REGS_R10, regs->r10);
3920 kvm_register_write(vcpu, VCPU_REGS_R11, regs->r11);
3921 kvm_register_write(vcpu, VCPU_REGS_R12, regs->r12);
3922 kvm_register_write(vcpu, VCPU_REGS_R13, regs->r13);
3923 kvm_register_write(vcpu, VCPU_REGS_R14, regs->r14);
3924 kvm_register_write(vcpu, VCPU_REGS_R15, regs->r15);
3928 kvm_rip_write(vcpu, regs->rip);
3929 kvm_x86_ops->set_rflags(vcpu, regs->rflags);
3932 vcpu->arch.exception.pending = false;
3939 void kvm_get_segment(struct kvm_vcpu *vcpu,
3940 struct kvm_segment *var, int seg)
3942 kvm_x86_ops->get_segment(vcpu, var, seg);
3945 void kvm_get_cs_db_l_bits(struct kvm_vcpu *vcpu, int *db, int *l)
3947 struct kvm_segment cs;
3949 kvm_get_segment(vcpu, &cs, VCPU_SREG_CS);
3953 EXPORT_SYMBOL_GPL(kvm_get_cs_db_l_bits);
3955 int kvm_arch_vcpu_ioctl_get_sregs(struct kvm_vcpu *vcpu,
3956 struct kvm_sregs *sregs)
3958 struct descriptor_table dt;
3962 kvm_get_segment(vcpu, &sregs->cs, VCPU_SREG_CS);
3963 kvm_get_segment(vcpu, &sregs->ds, VCPU_SREG_DS);
3964 kvm_get_segment(vcpu, &sregs->es, VCPU_SREG_ES);
3965 kvm_get_segment(vcpu, &sregs->fs, VCPU_SREG_FS);
3966 kvm_get_segment(vcpu, &sregs->gs, VCPU_SREG_GS);
3967 kvm_get_segment(vcpu, &sregs->ss, VCPU_SREG_SS);
3969 kvm_get_segment(vcpu, &sregs->tr, VCPU_SREG_TR);
3970 kvm_get_segment(vcpu, &sregs->ldt, VCPU_SREG_LDTR);
3972 kvm_x86_ops->get_idt(vcpu, &dt);
3973 sregs->idt.limit = dt.limit;
3974 sregs->idt.base = dt.base;
3975 kvm_x86_ops->get_gdt(vcpu, &dt);
3976 sregs->gdt.limit = dt.limit;
3977 sregs->gdt.base = dt.base;
3979 kvm_x86_ops->decache_cr4_guest_bits(vcpu);
3980 sregs->cr0 = vcpu->arch.cr0;
3981 sregs->cr2 = vcpu->arch.cr2;
3982 sregs->cr3 = vcpu->arch.cr3;
3983 sregs->cr4 = vcpu->arch.cr4;
3984 sregs->cr8 = kvm_get_cr8(vcpu);
3985 sregs->efer = vcpu->arch.shadow_efer;
3986 sregs->apic_base = kvm_get_apic_base(vcpu);
3988 memset(sregs->interrupt_bitmap, 0, sizeof sregs->interrupt_bitmap);
3990 if (vcpu->arch.interrupt.pending && !vcpu->arch.interrupt.soft)
3991 set_bit(vcpu->arch.interrupt.nr,
3992 (unsigned long *)sregs->interrupt_bitmap);
3999 int kvm_arch_vcpu_ioctl_get_mpstate(struct kvm_vcpu *vcpu,
4000 struct kvm_mp_state *mp_state)
4003 mp_state->mp_state = vcpu->arch.mp_state;
4008 int kvm_arch_vcpu_ioctl_set_mpstate(struct kvm_vcpu *vcpu,
4009 struct kvm_mp_state *mp_state)
4012 vcpu->arch.mp_state = mp_state->mp_state;
4017 static void kvm_set_segment(struct kvm_vcpu *vcpu,
4018 struct kvm_segment *var, int seg)
4020 kvm_x86_ops->set_segment(vcpu, var, seg);
4023 static void seg_desct_to_kvm_desct(struct desc_struct *seg_desc, u16 selector,
4024 struct kvm_segment *kvm_desct)
4026 kvm_desct->base = get_desc_base(seg_desc);
4027 kvm_desct->limit = get_desc_limit(seg_desc);
4029 kvm_desct->limit <<= 12;
4030 kvm_desct->limit |= 0xfff;
4032 kvm_desct->selector = selector;
4033 kvm_desct->type = seg_desc->type;
4034 kvm_desct->present = seg_desc->p;
4035 kvm_desct->dpl = seg_desc->dpl;
4036 kvm_desct->db = seg_desc->d;
4037 kvm_desct->s = seg_desc->s;
4038 kvm_desct->l = seg_desc->l;
4039 kvm_desct->g = seg_desc->g;
4040 kvm_desct->avl = seg_desc->avl;
4042 kvm_desct->unusable = 1;
4044 kvm_desct->unusable = 0;
4045 kvm_desct->padding = 0;
4048 static void get_segment_descriptor_dtable(struct kvm_vcpu *vcpu,
4050 struct descriptor_table *dtable)
4052 if (selector & 1 << 2) {
4053 struct kvm_segment kvm_seg;
4055 kvm_get_segment(vcpu, &kvm_seg, VCPU_SREG_LDTR);
4057 if (kvm_seg.unusable)
4060 dtable->limit = kvm_seg.limit;
4061 dtable->base = kvm_seg.base;
4064 kvm_x86_ops->get_gdt(vcpu, dtable);
4067 /* allowed just for 8 bytes segments */
4068 static int load_guest_segment_descriptor(struct kvm_vcpu *vcpu, u16 selector,
4069 struct desc_struct *seg_desc)
4071 struct descriptor_table dtable;
4072 u16 index = selector >> 3;
4074 get_segment_descriptor_dtable(vcpu, selector, &dtable);
4076 if (dtable.limit < index * 8 + 7) {
4077 kvm_queue_exception_e(vcpu, GP_VECTOR, selector & 0xfffc);
4080 return kvm_read_guest_virt(dtable.base + index*8, seg_desc, sizeof(*seg_desc), vcpu);
4083 /* allowed just for 8 bytes segments */
4084 static int save_guest_segment_descriptor(struct kvm_vcpu *vcpu, u16 selector,
4085 struct desc_struct *seg_desc)
4087 struct descriptor_table dtable;
4088 u16 index = selector >> 3;
4090 get_segment_descriptor_dtable(vcpu, selector, &dtable);
4092 if (dtable.limit < index * 8 + 7)
4094 return kvm_write_guest_virt(dtable.base + index*8, seg_desc, sizeof(*seg_desc), vcpu);
4097 static gpa_t get_tss_base_addr(struct kvm_vcpu *vcpu,
4098 struct desc_struct *seg_desc)
4100 u32 base_addr = get_desc_base(seg_desc);
4102 return vcpu->arch.mmu.gva_to_gpa(vcpu, base_addr);
4105 static u16 get_segment_selector(struct kvm_vcpu *vcpu, int seg)
4107 struct kvm_segment kvm_seg;
4109 kvm_get_segment(vcpu, &kvm_seg, seg);
4110 return kvm_seg.selector;
4113 static int load_segment_descriptor_to_kvm_desct(struct kvm_vcpu *vcpu,
4115 struct kvm_segment *kvm_seg)
4117 struct desc_struct seg_desc;
4119 if (load_guest_segment_descriptor(vcpu, selector, &seg_desc))
4121 seg_desct_to_kvm_desct(&seg_desc, selector, kvm_seg);
4125 static int kvm_load_realmode_segment(struct kvm_vcpu *vcpu, u16 selector, int seg)
4127 struct kvm_segment segvar = {
4128 .base = selector << 4,
4130 .selector = selector,
4141 kvm_x86_ops->set_segment(vcpu, &segvar, seg);
4145 static int is_vm86_segment(struct kvm_vcpu *vcpu, int seg)
4147 return (seg != VCPU_SREG_LDTR) &&
4148 (seg != VCPU_SREG_TR) &&
4149 (kvm_x86_ops->get_rflags(vcpu) & X86_EFLAGS_VM);
4152 int kvm_load_segment_descriptor(struct kvm_vcpu *vcpu, u16 selector,
4153 int type_bits, int seg)
4155 struct kvm_segment kvm_seg;
4157 if (is_vm86_segment(vcpu, seg) || !(vcpu->arch.cr0 & X86_CR0_PE))
4158 return kvm_load_realmode_segment(vcpu, selector, seg);
4159 if (load_segment_descriptor_to_kvm_desct(vcpu, selector, &kvm_seg))
4161 kvm_seg.type |= type_bits;
4163 if (seg != VCPU_SREG_SS && seg != VCPU_SREG_CS &&
4164 seg != VCPU_SREG_LDTR)
4166 kvm_seg.unusable = 1;
4168 kvm_set_segment(vcpu, &kvm_seg, seg);
4172 static void save_state_to_tss32(struct kvm_vcpu *vcpu,
4173 struct tss_segment_32 *tss)
4175 tss->cr3 = vcpu->arch.cr3;
4176 tss->eip = kvm_rip_read(vcpu);
4177 tss->eflags = kvm_x86_ops->get_rflags(vcpu);
4178 tss->eax = kvm_register_read(vcpu, VCPU_REGS_RAX);
4179 tss->ecx = kvm_register_read(vcpu, VCPU_REGS_RCX);
4180 tss->edx = kvm_register_read(vcpu, VCPU_REGS_RDX);
4181 tss->ebx = kvm_register_read(vcpu, VCPU_REGS_RBX);
4182 tss->esp = kvm_register_read(vcpu, VCPU_REGS_RSP);
4183 tss->ebp = kvm_register_read(vcpu, VCPU_REGS_RBP);
4184 tss->esi = kvm_register_read(vcpu, VCPU_REGS_RSI);
4185 tss->edi = kvm_register_read(vcpu, VCPU_REGS_RDI);
4186 tss->es = get_segment_selector(vcpu, VCPU_SREG_ES);
4187 tss->cs = get_segment_selector(vcpu, VCPU_SREG_CS);
4188 tss->ss = get_segment_selector(vcpu, VCPU_SREG_SS);
4189 tss->ds = get_segment_selector(vcpu, VCPU_SREG_DS);
4190 tss->fs = get_segment_selector(vcpu, VCPU_SREG_FS);
4191 tss->gs = get_segment_selector(vcpu, VCPU_SREG_GS);
4192 tss->ldt_selector = get_segment_selector(vcpu, VCPU_SREG_LDTR);
4195 static int load_state_from_tss32(struct kvm_vcpu *vcpu,
4196 struct tss_segment_32 *tss)
4198 kvm_set_cr3(vcpu, tss->cr3);
4200 kvm_rip_write(vcpu, tss->eip);
4201 kvm_x86_ops->set_rflags(vcpu, tss->eflags | 2);
4203 kvm_register_write(vcpu, VCPU_REGS_RAX, tss->eax);
4204 kvm_register_write(vcpu, VCPU_REGS_RCX, tss->ecx);
4205 kvm_register_write(vcpu, VCPU_REGS_RDX, tss->edx);
4206 kvm_register_write(vcpu, VCPU_REGS_RBX, tss->ebx);
4207 kvm_register_write(vcpu, VCPU_REGS_RSP, tss->esp);
4208 kvm_register_write(vcpu, VCPU_REGS_RBP, tss->ebp);
4209 kvm_register_write(vcpu, VCPU_REGS_RSI, tss->esi);
4210 kvm_register_write(vcpu, VCPU_REGS_RDI, tss->edi);
4212 if (kvm_load_segment_descriptor(vcpu, tss->ldt_selector, 0, VCPU_SREG_LDTR))
4215 if (kvm_load_segment_descriptor(vcpu, tss->es, 1, VCPU_SREG_ES))
4218 if (kvm_load_segment_descriptor(vcpu, tss->cs, 9, VCPU_SREG_CS))
4221 if (kvm_load_segment_descriptor(vcpu, tss->ss, 1, VCPU_SREG_SS))
4224 if (kvm_load_segment_descriptor(vcpu, tss->ds, 1, VCPU_SREG_DS))
4227 if (kvm_load_segment_descriptor(vcpu, tss->fs, 1, VCPU_SREG_FS))
4230 if (kvm_load_segment_descriptor(vcpu, tss->gs, 1, VCPU_SREG_GS))
4235 static void save_state_to_tss16(struct kvm_vcpu *vcpu,
4236 struct tss_segment_16 *tss)
4238 tss->ip = kvm_rip_read(vcpu);
4239 tss->flag = kvm_x86_ops->get_rflags(vcpu);
4240 tss->ax = kvm_register_read(vcpu, VCPU_REGS_RAX);
4241 tss->cx = kvm_register_read(vcpu, VCPU_REGS_RCX);
4242 tss->dx = kvm_register_read(vcpu, VCPU_REGS_RDX);
4243 tss->bx = kvm_register_read(vcpu, VCPU_REGS_RBX);
4244 tss->sp = kvm_register_read(vcpu, VCPU_REGS_RSP);
4245 tss->bp = kvm_register_read(vcpu, VCPU_REGS_RBP);
4246 tss->si = kvm_register_read(vcpu, VCPU_REGS_RSI);
4247 tss->di = kvm_register_read(vcpu, VCPU_REGS_RDI);
4249 tss->es = get_segment_selector(vcpu, VCPU_SREG_ES);
4250 tss->cs = get_segment_selector(vcpu, VCPU_SREG_CS);
4251 tss->ss = get_segment_selector(vcpu, VCPU_SREG_SS);
4252 tss->ds = get_segment_selector(vcpu, VCPU_SREG_DS);
4253 tss->ldt = get_segment_selector(vcpu, VCPU_SREG_LDTR);
4254 tss->prev_task_link = get_segment_selector(vcpu, VCPU_SREG_TR);
4257 static int load_state_from_tss16(struct kvm_vcpu *vcpu,
4258 struct tss_segment_16 *tss)
4260 kvm_rip_write(vcpu, tss->ip);
4261 kvm_x86_ops->set_rflags(vcpu, tss->flag | 2);
4262 kvm_register_write(vcpu, VCPU_REGS_RAX, tss->ax);
4263 kvm_register_write(vcpu, VCPU_REGS_RCX, tss->cx);
4264 kvm_register_write(vcpu, VCPU_REGS_RDX, tss->dx);
4265 kvm_register_write(vcpu, VCPU_REGS_RBX, tss->bx);
4266 kvm_register_write(vcpu, VCPU_REGS_RSP, tss->sp);
4267 kvm_register_write(vcpu, VCPU_REGS_RBP, tss->bp);
4268 kvm_register_write(vcpu, VCPU_REGS_RSI, tss->si);
4269 kvm_register_write(vcpu, VCPU_REGS_RDI, tss->di);
4271 if (kvm_load_segment_descriptor(vcpu, tss->ldt, 0, VCPU_SREG_LDTR))
4274 if (kvm_load_segment_descriptor(vcpu, tss->es, 1, VCPU_SREG_ES))
4277 if (kvm_load_segment_descriptor(vcpu, tss->cs, 9, VCPU_SREG_CS))
4280 if (kvm_load_segment_descriptor(vcpu, tss->ss, 1, VCPU_SREG_SS))
4283 if (kvm_load_segment_descriptor(vcpu, tss->ds, 1, VCPU_SREG_DS))
4288 static int kvm_task_switch_16(struct kvm_vcpu *vcpu, u16 tss_selector,
4289 u16 old_tss_sel, u32 old_tss_base,
4290 struct desc_struct *nseg_desc)
4292 struct tss_segment_16 tss_segment_16;
4295 if (kvm_read_guest(vcpu->kvm, old_tss_base, &tss_segment_16,
4296 sizeof tss_segment_16))
4299 save_state_to_tss16(vcpu, &tss_segment_16);
4301 if (kvm_write_guest(vcpu->kvm, old_tss_base, &tss_segment_16,
4302 sizeof tss_segment_16))
4305 if (kvm_read_guest(vcpu->kvm, get_tss_base_addr(vcpu, nseg_desc),
4306 &tss_segment_16, sizeof tss_segment_16))
4309 if (old_tss_sel != 0xffff) {
4310 tss_segment_16.prev_task_link = old_tss_sel;
4312 if (kvm_write_guest(vcpu->kvm,
4313 get_tss_base_addr(vcpu, nseg_desc),
4314 &tss_segment_16.prev_task_link,
4315 sizeof tss_segment_16.prev_task_link))
4319 if (load_state_from_tss16(vcpu, &tss_segment_16))
4327 static int kvm_task_switch_32(struct kvm_vcpu *vcpu, u16 tss_selector,
4328 u16 old_tss_sel, u32 old_tss_base,
4329 struct desc_struct *nseg_desc)
4331 struct tss_segment_32 tss_segment_32;
4334 if (kvm_read_guest(vcpu->kvm, old_tss_base, &tss_segment_32,
4335 sizeof tss_segment_32))
4338 save_state_to_tss32(vcpu, &tss_segment_32);
4340 if (kvm_write_guest(vcpu->kvm, old_tss_base, &tss_segment_32,
4341 sizeof tss_segment_32))
4344 if (kvm_read_guest(vcpu->kvm, get_tss_base_addr(vcpu, nseg_desc),
4345 &tss_segment_32, sizeof tss_segment_32))
4348 if (old_tss_sel != 0xffff) {
4349 tss_segment_32.prev_task_link = old_tss_sel;
4351 if (kvm_write_guest(vcpu->kvm,
4352 get_tss_base_addr(vcpu, nseg_desc),
4353 &tss_segment_32.prev_task_link,
4354 sizeof tss_segment_32.prev_task_link))
4358 if (load_state_from_tss32(vcpu, &tss_segment_32))
4366 int kvm_task_switch(struct kvm_vcpu *vcpu, u16 tss_selector, int reason)
4368 struct kvm_segment tr_seg;
4369 struct desc_struct cseg_desc;
4370 struct desc_struct nseg_desc;
4372 u32 old_tss_base = get_segment_base(vcpu, VCPU_SREG_TR);
4373 u16 old_tss_sel = get_segment_selector(vcpu, VCPU_SREG_TR);
4375 old_tss_base = vcpu->arch.mmu.gva_to_gpa(vcpu, old_tss_base);
4377 /* FIXME: Handle errors. Failure to read either TSS or their
4378 * descriptors should generate a pagefault.
4380 if (load_guest_segment_descriptor(vcpu, tss_selector, &nseg_desc))
4383 if (load_guest_segment_descriptor(vcpu, old_tss_sel, &cseg_desc))
4386 if (reason != TASK_SWITCH_IRET) {
4389 cpl = kvm_x86_ops->get_cpl(vcpu);
4390 if ((tss_selector & 3) > nseg_desc.dpl || cpl > nseg_desc.dpl) {
4391 kvm_queue_exception_e(vcpu, GP_VECTOR, 0);
4396 if (!nseg_desc.p || get_desc_limit(&nseg_desc) < 0x67) {
4397 kvm_queue_exception_e(vcpu, TS_VECTOR, tss_selector & 0xfffc);
4401 if (reason == TASK_SWITCH_IRET || reason == TASK_SWITCH_JMP) {
4402 cseg_desc.type &= ~(1 << 1); //clear the B flag
4403 save_guest_segment_descriptor(vcpu, old_tss_sel, &cseg_desc);
4406 if (reason == TASK_SWITCH_IRET) {
4407 u32 eflags = kvm_x86_ops->get_rflags(vcpu);
4408 kvm_x86_ops->set_rflags(vcpu, eflags & ~X86_EFLAGS_NT);
4411 /* set back link to prev task only if NT bit is set in eflags
4412 note that old_tss_sel is not used afetr this point */
4413 if (reason != TASK_SWITCH_CALL && reason != TASK_SWITCH_GATE)
4414 old_tss_sel = 0xffff;
4416 /* set back link to prev task only if NT bit is set in eflags
4417 note that old_tss_sel is not used afetr this point */
4418 if (reason != TASK_SWITCH_CALL && reason != TASK_SWITCH_GATE)
4419 old_tss_sel = 0xffff;
4421 if (nseg_desc.type & 8)
4422 ret = kvm_task_switch_32(vcpu, tss_selector, old_tss_sel,
4423 old_tss_base, &nseg_desc);
4425 ret = kvm_task_switch_16(vcpu, tss_selector, old_tss_sel,
4426 old_tss_base, &nseg_desc);
4428 if (reason == TASK_SWITCH_CALL || reason == TASK_SWITCH_GATE) {
4429 u32 eflags = kvm_x86_ops->get_rflags(vcpu);
4430 kvm_x86_ops->set_rflags(vcpu, eflags | X86_EFLAGS_NT);
4433 if (reason != TASK_SWITCH_IRET) {
4434 nseg_desc.type |= (1 << 1);
4435 save_guest_segment_descriptor(vcpu, tss_selector,
4439 kvm_x86_ops->set_cr0(vcpu, vcpu->arch.cr0 | X86_CR0_TS);
4440 seg_desct_to_kvm_desct(&nseg_desc, tss_selector, &tr_seg);
4442 kvm_set_segment(vcpu, &tr_seg, VCPU_SREG_TR);
4446 EXPORT_SYMBOL_GPL(kvm_task_switch);
4448 int kvm_arch_vcpu_ioctl_set_sregs(struct kvm_vcpu *vcpu,
4449 struct kvm_sregs *sregs)
4451 int mmu_reset_needed = 0;
4452 int pending_vec, max_bits;
4453 struct descriptor_table dt;
4457 dt.limit = sregs->idt.limit;
4458 dt.base = sregs->idt.base;
4459 kvm_x86_ops->set_idt(vcpu, &dt);
4460 dt.limit = sregs->gdt.limit;
4461 dt.base = sregs->gdt.base;
4462 kvm_x86_ops->set_gdt(vcpu, &dt);
4464 vcpu->arch.cr2 = sregs->cr2;
4465 mmu_reset_needed |= vcpu->arch.cr3 != sregs->cr3;
4466 vcpu->arch.cr3 = sregs->cr3;
4468 kvm_set_cr8(vcpu, sregs->cr8);
4470 mmu_reset_needed |= vcpu->arch.shadow_efer != sregs->efer;
4471 kvm_x86_ops->set_efer(vcpu, sregs->efer);
4472 kvm_set_apic_base(vcpu, sregs->apic_base);
4474 kvm_x86_ops->decache_cr4_guest_bits(vcpu);
4476 mmu_reset_needed |= vcpu->arch.cr0 != sregs->cr0;
4477 kvm_x86_ops->set_cr0(vcpu, sregs->cr0);
4478 vcpu->arch.cr0 = sregs->cr0;
4480 mmu_reset_needed |= vcpu->arch.cr4 != sregs->cr4;
4481 kvm_x86_ops->set_cr4(vcpu, sregs->cr4);
4482 if (!is_long_mode(vcpu) && is_pae(vcpu))
4483 load_pdptrs(vcpu, vcpu->arch.cr3);
4485 if (mmu_reset_needed)
4486 kvm_mmu_reset_context(vcpu);
4488 max_bits = (sizeof sregs->interrupt_bitmap) << 3;
4489 pending_vec = find_first_bit(
4490 (const unsigned long *)sregs->interrupt_bitmap, max_bits);
4491 if (pending_vec < max_bits) {
4492 kvm_queue_interrupt(vcpu, pending_vec, false);
4493 pr_debug("Set back pending irq %d\n", pending_vec);
4494 if (irqchip_in_kernel(vcpu->kvm))
4495 kvm_pic_clear_isr_ack(vcpu->kvm);
4498 kvm_set_segment(vcpu, &sregs->cs, VCPU_SREG_CS);
4499 kvm_set_segment(vcpu, &sregs->ds, VCPU_SREG_DS);
4500 kvm_set_segment(vcpu, &sregs->es, VCPU_SREG_ES);
4501 kvm_set_segment(vcpu, &sregs->fs, VCPU_SREG_FS);
4502 kvm_set_segment(vcpu, &sregs->gs, VCPU_SREG_GS);
4503 kvm_set_segment(vcpu, &sregs->ss, VCPU_SREG_SS);
4505 kvm_set_segment(vcpu, &sregs->tr, VCPU_SREG_TR);
4506 kvm_set_segment(vcpu, &sregs->ldt, VCPU_SREG_LDTR);
4508 update_cr8_intercept(vcpu);
4510 /* Older userspace won't unhalt the vcpu on reset. */
4511 if (kvm_vcpu_is_bsp(vcpu) && kvm_rip_read(vcpu) == 0xfff0 &&
4512 sregs->cs.selector == 0xf000 && sregs->cs.base == 0xffff0000 &&
4513 !(vcpu->arch.cr0 & X86_CR0_PE))
4514 vcpu->arch.mp_state = KVM_MP_STATE_RUNNABLE;
4521 int kvm_arch_vcpu_ioctl_set_guest_debug(struct kvm_vcpu *vcpu,
4522 struct kvm_guest_debug *dbg)
4528 if ((dbg->control & (KVM_GUESTDBG_ENABLE | KVM_GUESTDBG_USE_HW_BP)) ==
4529 (KVM_GUESTDBG_ENABLE | KVM_GUESTDBG_USE_HW_BP)) {
4530 for (i = 0; i < KVM_NR_DB_REGS; ++i)
4531 vcpu->arch.eff_db[i] = dbg->arch.debugreg[i];
4532 vcpu->arch.switch_db_regs =
4533 (dbg->arch.debugreg[7] & DR7_BP_EN_MASK);
4535 for (i = 0; i < KVM_NR_DB_REGS; i++)
4536 vcpu->arch.eff_db[i] = vcpu->arch.db[i];
4537 vcpu->arch.switch_db_regs = (vcpu->arch.dr7 & DR7_BP_EN_MASK);
4540 r = kvm_x86_ops->set_guest_debug(vcpu, dbg);
4542 if (dbg->control & KVM_GUESTDBG_INJECT_DB)
4543 kvm_queue_exception(vcpu, DB_VECTOR);
4544 else if (dbg->control & KVM_GUESTDBG_INJECT_BP)
4545 kvm_queue_exception(vcpu, BP_VECTOR);
4553 * fxsave fpu state. Taken from x86_64/processor.h. To be killed when
4554 * we have asm/x86/processor.h
4565 u32 st_space[32]; /* 8*16 bytes for each FP-reg = 128 bytes */
4566 #ifdef CONFIG_X86_64
4567 u32 xmm_space[64]; /* 16*16 bytes for each XMM-reg = 256 bytes */
4569 u32 xmm_space[32]; /* 8*16 bytes for each XMM-reg = 128 bytes */
4574 * Translate a guest virtual address to a guest physical address.
4576 int kvm_arch_vcpu_ioctl_translate(struct kvm_vcpu *vcpu,
4577 struct kvm_translation *tr)
4579 unsigned long vaddr = tr->linear_address;
4583 down_read(&vcpu->kvm->slots_lock);
4584 gpa = vcpu->arch.mmu.gva_to_gpa(vcpu, vaddr);
4585 up_read(&vcpu->kvm->slots_lock);
4586 tr->physical_address = gpa;
4587 tr->valid = gpa != UNMAPPED_GVA;
4595 int kvm_arch_vcpu_ioctl_get_fpu(struct kvm_vcpu *vcpu, struct kvm_fpu *fpu)
4597 struct fxsave *fxsave = (struct fxsave *)&vcpu->arch.guest_fx_image;
4601 memcpy(fpu->fpr, fxsave->st_space, 128);
4602 fpu->fcw = fxsave->cwd;
4603 fpu->fsw = fxsave->swd;
4604 fpu->ftwx = fxsave->twd;
4605 fpu->last_opcode = fxsave->fop;
4606 fpu->last_ip = fxsave->rip;
4607 fpu->last_dp = fxsave->rdp;
4608 memcpy(fpu->xmm, fxsave->xmm_space, sizeof fxsave->xmm_space);
4615 int kvm_arch_vcpu_ioctl_set_fpu(struct kvm_vcpu *vcpu, struct kvm_fpu *fpu)
4617 struct fxsave *fxsave = (struct fxsave *)&vcpu->arch.guest_fx_image;
4621 memcpy(fxsave->st_space, fpu->fpr, 128);
4622 fxsave->cwd = fpu->fcw;
4623 fxsave->swd = fpu->fsw;
4624 fxsave->twd = fpu->ftwx;
4625 fxsave->fop = fpu->last_opcode;
4626 fxsave->rip = fpu->last_ip;
4627 fxsave->rdp = fpu->last_dp;
4628 memcpy(fxsave->xmm_space, fpu->xmm, sizeof fxsave->xmm_space);
4635 void fx_init(struct kvm_vcpu *vcpu)
4637 unsigned after_mxcsr_mask;
4640 * Touch the fpu the first time in non atomic context as if
4641 * this is the first fpu instruction the exception handler
4642 * will fire before the instruction returns and it'll have to
4643 * allocate ram with GFP_KERNEL.
4646 kvm_fx_save(&vcpu->arch.host_fx_image);
4648 /* Initialize guest FPU by resetting ours and saving into guest's */
4650 kvm_fx_save(&vcpu->arch.host_fx_image);
4652 kvm_fx_save(&vcpu->arch.guest_fx_image);
4653 kvm_fx_restore(&vcpu->arch.host_fx_image);
4656 vcpu->arch.cr0 |= X86_CR0_ET;
4657 after_mxcsr_mask = offsetof(struct i387_fxsave_struct, st_space);
4658 vcpu->arch.guest_fx_image.mxcsr = 0x1f80;
4659 memset((void *)&vcpu->arch.guest_fx_image + after_mxcsr_mask,
4660 0, sizeof(struct i387_fxsave_struct) - after_mxcsr_mask);
4662 EXPORT_SYMBOL_GPL(fx_init);
4664 void kvm_load_guest_fpu(struct kvm_vcpu *vcpu)
4666 if (!vcpu->fpu_active || vcpu->guest_fpu_loaded)
4669 vcpu->guest_fpu_loaded = 1;
4670 kvm_fx_save(&vcpu->arch.host_fx_image);
4671 kvm_fx_restore(&vcpu->arch.guest_fx_image);
4673 EXPORT_SYMBOL_GPL(kvm_load_guest_fpu);
4675 void kvm_put_guest_fpu(struct kvm_vcpu *vcpu)
4677 if (!vcpu->guest_fpu_loaded)
4680 vcpu->guest_fpu_loaded = 0;
4681 kvm_fx_save(&vcpu->arch.guest_fx_image);
4682 kvm_fx_restore(&vcpu->arch.host_fx_image);
4683 ++vcpu->stat.fpu_reload;
4685 EXPORT_SYMBOL_GPL(kvm_put_guest_fpu);
4687 void kvm_arch_vcpu_free(struct kvm_vcpu *vcpu)
4689 if (vcpu->arch.time_page) {
4690 kvm_release_page_dirty(vcpu->arch.time_page);
4691 vcpu->arch.time_page = NULL;
4694 kvm_x86_ops->vcpu_free(vcpu);
4697 struct kvm_vcpu *kvm_arch_vcpu_create(struct kvm *kvm,
4700 return kvm_x86_ops->vcpu_create(kvm, id);
4703 int kvm_arch_vcpu_setup(struct kvm_vcpu *vcpu)
4707 /* We do fxsave: this must be aligned. */
4708 BUG_ON((unsigned long)&vcpu->arch.host_fx_image & 0xF);
4710 vcpu->arch.mtrr_state.have_fixed = 1;
4712 r = kvm_arch_vcpu_reset(vcpu);
4714 r = kvm_mmu_setup(vcpu);
4721 kvm_x86_ops->vcpu_free(vcpu);
4725 void kvm_arch_vcpu_destroy(struct kvm_vcpu *vcpu)
4728 kvm_mmu_unload(vcpu);
4731 kvm_x86_ops->vcpu_free(vcpu);
4734 int kvm_arch_vcpu_reset(struct kvm_vcpu *vcpu)
4736 vcpu->arch.nmi_pending = false;
4737 vcpu->arch.nmi_injected = false;
4739 vcpu->arch.switch_db_regs = 0;
4740 memset(vcpu->arch.db, 0, sizeof(vcpu->arch.db));
4741 vcpu->arch.dr6 = DR6_FIXED_1;
4742 vcpu->arch.dr7 = DR7_FIXED_1;
4744 return kvm_x86_ops->vcpu_reset(vcpu);
4747 void kvm_arch_hardware_enable(void *garbage)
4749 kvm_x86_ops->hardware_enable(garbage);
4752 void kvm_arch_hardware_disable(void *garbage)
4754 kvm_x86_ops->hardware_disable(garbage);
4757 int kvm_arch_hardware_setup(void)
4759 return kvm_x86_ops->hardware_setup();
4762 void kvm_arch_hardware_unsetup(void)
4764 kvm_x86_ops->hardware_unsetup();
4767 void kvm_arch_check_processor_compat(void *rtn)
4769 kvm_x86_ops->check_processor_compatibility(rtn);
4772 int kvm_arch_vcpu_init(struct kvm_vcpu *vcpu)
4778 BUG_ON(vcpu->kvm == NULL);
4781 vcpu->arch.mmu.root_hpa = INVALID_PAGE;
4782 if (!irqchip_in_kernel(kvm) || kvm_vcpu_is_bsp(vcpu))
4783 vcpu->arch.mp_state = KVM_MP_STATE_RUNNABLE;
4785 vcpu->arch.mp_state = KVM_MP_STATE_UNINITIALIZED;
4787 page = alloc_page(GFP_KERNEL | __GFP_ZERO);
4792 vcpu->arch.pio_data = page_address(page);
4794 r = kvm_mmu_create(vcpu);
4796 goto fail_free_pio_data;
4798 if (irqchip_in_kernel(kvm)) {
4799 r = kvm_create_lapic(vcpu);
4801 goto fail_mmu_destroy;
4804 vcpu->arch.mce_banks = kzalloc(KVM_MAX_MCE_BANKS * sizeof(u64) * 4,
4806 if (!vcpu->arch.mce_banks) {
4808 goto fail_free_lapic;
4810 vcpu->arch.mcg_cap = KVM_MAX_MCE_BANKS;
4814 kvm_free_lapic(vcpu);
4816 kvm_mmu_destroy(vcpu);
4818 free_page((unsigned long)vcpu->arch.pio_data);
4823 void kvm_arch_vcpu_uninit(struct kvm_vcpu *vcpu)
4825 kfree(vcpu->arch.mce_banks);
4826 kvm_free_lapic(vcpu);
4827 down_read(&vcpu->kvm->slots_lock);
4828 kvm_mmu_destroy(vcpu);
4829 up_read(&vcpu->kvm->slots_lock);
4830 free_page((unsigned long)vcpu->arch.pio_data);
4833 struct kvm *kvm_arch_create_vm(void)
4835 struct kvm *kvm = kzalloc(sizeof(struct kvm), GFP_KERNEL);
4838 return ERR_PTR(-ENOMEM);
4840 INIT_LIST_HEAD(&kvm->arch.active_mmu_pages);
4841 INIT_LIST_HEAD(&kvm->arch.assigned_dev_head);
4843 /* Reserve bit 0 of irq_sources_bitmap for userspace irq source */
4844 set_bit(KVM_USERSPACE_IRQ_SOURCE_ID, &kvm->arch.irq_sources_bitmap);
4846 rdtscll(kvm->arch.vm_init_tsc);
4851 static void kvm_unload_vcpu_mmu(struct kvm_vcpu *vcpu)
4854 kvm_mmu_unload(vcpu);
4858 static void kvm_free_vcpus(struct kvm *kvm)
4861 struct kvm_vcpu *vcpu;
4864 * Unpin any mmu pages first.
4866 kvm_for_each_vcpu(i, vcpu, kvm)
4867 kvm_unload_vcpu_mmu(vcpu);
4868 kvm_for_each_vcpu(i, vcpu, kvm)
4869 kvm_arch_vcpu_free(vcpu);
4871 mutex_lock(&kvm->lock);
4872 for (i = 0; i < atomic_read(&kvm->online_vcpus); i++)
4873 kvm->vcpus[i] = NULL;
4875 atomic_set(&kvm->online_vcpus, 0);
4876 mutex_unlock(&kvm->lock);
4879 void kvm_arch_sync_events(struct kvm *kvm)
4881 kvm_free_all_assigned_devices(kvm);
4884 void kvm_arch_destroy_vm(struct kvm *kvm)
4886 kvm_iommu_unmap_guest(kvm);
4888 kfree(kvm->arch.vpic);
4889 kfree(kvm->arch.vioapic);
4890 kvm_free_vcpus(kvm);
4891 kvm_free_physmem(kvm);
4892 if (kvm->arch.apic_access_page)
4893 put_page(kvm->arch.apic_access_page);
4894 if (kvm->arch.ept_identity_pagetable)
4895 put_page(kvm->arch.ept_identity_pagetable);
4899 int kvm_arch_set_memory_region(struct kvm *kvm,
4900 struct kvm_userspace_memory_region *mem,
4901 struct kvm_memory_slot old,
4904 int npages = mem->memory_size >> PAGE_SHIFT;
4905 struct kvm_memory_slot *memslot = &kvm->memslots[mem->slot];
4907 /*To keep backward compatibility with older userspace,
4908 *x86 needs to hanlde !user_alloc case.
4911 if (npages && !old.rmap) {
4912 unsigned long userspace_addr;
4914 down_write(¤t->mm->mmap_sem);
4915 userspace_addr = do_mmap(NULL, 0,
4917 PROT_READ | PROT_WRITE,
4918 MAP_PRIVATE | MAP_ANONYMOUS,
4920 up_write(¤t->mm->mmap_sem);
4922 if (IS_ERR((void *)userspace_addr))
4923 return PTR_ERR((void *)userspace_addr);
4925 /* set userspace_addr atomically for kvm_hva_to_rmapp */
4926 spin_lock(&kvm->mmu_lock);
4927 memslot->userspace_addr = userspace_addr;
4928 spin_unlock(&kvm->mmu_lock);
4930 if (!old.user_alloc && old.rmap) {
4933 down_write(¤t->mm->mmap_sem);
4934 ret = do_munmap(current->mm, old.userspace_addr,
4935 old.npages * PAGE_SIZE);
4936 up_write(¤t->mm->mmap_sem);
4939 "kvm_vm_ioctl_set_memory_region: "
4940 "failed to munmap memory\n");
4945 spin_lock(&kvm->mmu_lock);
4946 if (!kvm->arch.n_requested_mmu_pages) {
4947 unsigned int nr_mmu_pages = kvm_mmu_calculate_mmu_pages(kvm);
4948 kvm_mmu_change_mmu_pages(kvm, nr_mmu_pages);
4951 kvm_mmu_slot_remove_write_access(kvm, mem->slot);
4952 spin_unlock(&kvm->mmu_lock);
4957 void kvm_arch_flush_shadow(struct kvm *kvm)
4959 kvm_mmu_zap_all(kvm);
4960 kvm_reload_remote_mmus(kvm);
4963 int kvm_arch_vcpu_runnable(struct kvm_vcpu *vcpu)
4965 return vcpu->arch.mp_state == KVM_MP_STATE_RUNNABLE
4966 || vcpu->arch.mp_state == KVM_MP_STATE_SIPI_RECEIVED
4967 || vcpu->arch.nmi_pending ||
4968 (kvm_arch_interrupt_allowed(vcpu) &&
4969 kvm_cpu_has_interrupt(vcpu));
4972 void kvm_vcpu_kick(struct kvm_vcpu *vcpu)
4975 int cpu = vcpu->cpu;
4977 if (waitqueue_active(&vcpu->wq)) {
4978 wake_up_interruptible(&vcpu->wq);
4979 ++vcpu->stat.halt_wakeup;
4983 if (cpu != me && (unsigned)cpu < nr_cpu_ids && cpu_online(cpu))
4984 if (!test_and_set_bit(KVM_REQ_KICK, &vcpu->requests))
4985 smp_send_reschedule(cpu);
4989 int kvm_arch_interrupt_allowed(struct kvm_vcpu *vcpu)
4991 return kvm_x86_ops->interrupt_allowed(vcpu);
4994 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_exit);
4995 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_inj_virq);
4996 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_page_fault);
4997 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_msr);
4998 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_cr);