2 * Kernel-based Virtual Machine driver for Linux
4 * derived from drivers/kvm/kvm_main.c
6 * Copyright (C) 2006 Qumranet, Inc.
9 * Avi Kivity <avi@qumranet.com>
10 * Yaniv Kamay <yaniv@qumranet.com>
12 * This work is licensed under the terms of the GNU GPL, version 2. See
13 * the COPYING file in the top-level directory.
17 #include <linux/kvm_host.h>
23 #include <linux/clocksource.h>
24 #include <linux/kvm.h>
26 #include <linux/vmalloc.h>
27 #include <linux/module.h>
28 #include <linux/mman.h>
29 #include <linux/highmem.h>
31 #include <asm/uaccess.h>
35 #define MAX_IO_MSRS 256
36 #define CR0_RESERVED_BITS \
37 (~(unsigned long)(X86_CR0_PE | X86_CR0_MP | X86_CR0_EM | X86_CR0_TS \
38 | X86_CR0_ET | X86_CR0_NE | X86_CR0_WP | X86_CR0_AM \
39 | X86_CR0_NW | X86_CR0_CD | X86_CR0_PG))
40 #define CR4_RESERVED_BITS \
41 (~(unsigned long)(X86_CR4_VME | X86_CR4_PVI | X86_CR4_TSD | X86_CR4_DE\
42 | X86_CR4_PSE | X86_CR4_PAE | X86_CR4_MCE \
43 | X86_CR4_PGE | X86_CR4_PCE | X86_CR4_OSFXSR \
44 | X86_CR4_OSXMMEXCPT | X86_CR4_VMXE))
46 #define CR8_RESERVED_BITS (~(unsigned long)X86_CR8_TPR)
48 * - enable syscall per default because its emulated by KVM
49 * - enable LME and LMA per default on 64 bit KVM
52 static u64 __read_mostly efer_reserved_bits = 0xfffffffffffffafeULL;
54 static u64 __read_mostly efer_reserved_bits = 0xfffffffffffffffeULL;
57 #define VM_STAT(x) offsetof(struct kvm, stat.x), KVM_STAT_VM
58 #define VCPU_STAT(x) offsetof(struct kvm_vcpu, stat.x), KVM_STAT_VCPU
60 static int kvm_dev_ioctl_get_supported_cpuid(struct kvm_cpuid2 *cpuid,
61 struct kvm_cpuid_entry2 __user *entries);
63 struct kvm_x86_ops *kvm_x86_ops;
65 struct kvm_stats_debugfs_item debugfs_entries[] = {
66 { "pf_fixed", VCPU_STAT(pf_fixed) },
67 { "pf_guest", VCPU_STAT(pf_guest) },
68 { "tlb_flush", VCPU_STAT(tlb_flush) },
69 { "invlpg", VCPU_STAT(invlpg) },
70 { "exits", VCPU_STAT(exits) },
71 { "io_exits", VCPU_STAT(io_exits) },
72 { "mmio_exits", VCPU_STAT(mmio_exits) },
73 { "signal_exits", VCPU_STAT(signal_exits) },
74 { "irq_window", VCPU_STAT(irq_window_exits) },
75 { "nmi_window", VCPU_STAT(nmi_window_exits) },
76 { "halt_exits", VCPU_STAT(halt_exits) },
77 { "halt_wakeup", VCPU_STAT(halt_wakeup) },
78 { "hypercalls", VCPU_STAT(hypercalls) },
79 { "request_irq", VCPU_STAT(request_irq_exits) },
80 { "irq_exits", VCPU_STAT(irq_exits) },
81 { "host_state_reload", VCPU_STAT(host_state_reload) },
82 { "efer_reload", VCPU_STAT(efer_reload) },
83 { "fpu_reload", VCPU_STAT(fpu_reload) },
84 { "insn_emulation", VCPU_STAT(insn_emulation) },
85 { "insn_emulation_fail", VCPU_STAT(insn_emulation_fail) },
86 { "mmu_shadow_zapped", VM_STAT(mmu_shadow_zapped) },
87 { "mmu_pte_write", VM_STAT(mmu_pte_write) },
88 { "mmu_pte_updated", VM_STAT(mmu_pte_updated) },
89 { "mmu_pde_zapped", VM_STAT(mmu_pde_zapped) },
90 { "mmu_flooded", VM_STAT(mmu_flooded) },
91 { "mmu_recycled", VM_STAT(mmu_recycled) },
92 { "mmu_cache_miss", VM_STAT(mmu_cache_miss) },
93 { "remote_tlb_flush", VM_STAT(remote_tlb_flush) },
94 { "largepages", VM_STAT(lpages) },
99 unsigned long segment_base(u16 selector)
101 struct descriptor_table gdt;
102 struct desc_struct *d;
103 unsigned long table_base;
109 asm("sgdt %0" : "=m"(gdt));
110 table_base = gdt.base;
112 if (selector & 4) { /* from ldt */
115 asm("sldt %0" : "=g"(ldt_selector));
116 table_base = segment_base(ldt_selector);
118 d = (struct desc_struct *)(table_base + (selector & ~7));
119 v = d->base0 | ((unsigned long)d->base1 << 16) |
120 ((unsigned long)d->base2 << 24);
122 if (d->s == 0 && (d->type == 2 || d->type == 9 || d->type == 11))
123 v |= ((unsigned long)((struct ldttss_desc64 *)d)->base3) << 32;
127 EXPORT_SYMBOL_GPL(segment_base);
129 u64 kvm_get_apic_base(struct kvm_vcpu *vcpu)
131 if (irqchip_in_kernel(vcpu->kvm))
132 return vcpu->arch.apic_base;
134 return vcpu->arch.apic_base;
136 EXPORT_SYMBOL_GPL(kvm_get_apic_base);
138 void kvm_set_apic_base(struct kvm_vcpu *vcpu, u64 data)
140 /* TODO: reserve bits check */
141 if (irqchip_in_kernel(vcpu->kvm))
142 kvm_lapic_set_base(vcpu, data);
144 vcpu->arch.apic_base = data;
146 EXPORT_SYMBOL_GPL(kvm_set_apic_base);
148 void kvm_queue_exception(struct kvm_vcpu *vcpu, unsigned nr)
150 WARN_ON(vcpu->arch.exception.pending);
151 vcpu->arch.exception.pending = true;
152 vcpu->arch.exception.has_error_code = false;
153 vcpu->arch.exception.nr = nr;
155 EXPORT_SYMBOL_GPL(kvm_queue_exception);
157 void kvm_inject_page_fault(struct kvm_vcpu *vcpu, unsigned long addr,
160 ++vcpu->stat.pf_guest;
161 if (vcpu->arch.exception.pending) {
162 if (vcpu->arch.exception.nr == PF_VECTOR) {
163 printk(KERN_DEBUG "kvm: inject_page_fault:"
164 " double fault 0x%lx\n", addr);
165 vcpu->arch.exception.nr = DF_VECTOR;
166 vcpu->arch.exception.error_code = 0;
167 } else if (vcpu->arch.exception.nr == DF_VECTOR) {
168 /* triple fault -> shutdown */
169 set_bit(KVM_REQ_TRIPLE_FAULT, &vcpu->requests);
173 vcpu->arch.cr2 = addr;
174 kvm_queue_exception_e(vcpu, PF_VECTOR, error_code);
177 void kvm_inject_nmi(struct kvm_vcpu *vcpu)
179 vcpu->arch.nmi_pending = 1;
181 EXPORT_SYMBOL_GPL(kvm_inject_nmi);
183 void kvm_queue_exception_e(struct kvm_vcpu *vcpu, unsigned nr, u32 error_code)
185 WARN_ON(vcpu->arch.exception.pending);
186 vcpu->arch.exception.pending = true;
187 vcpu->arch.exception.has_error_code = true;
188 vcpu->arch.exception.nr = nr;
189 vcpu->arch.exception.error_code = error_code;
191 EXPORT_SYMBOL_GPL(kvm_queue_exception_e);
193 static void __queue_exception(struct kvm_vcpu *vcpu)
195 kvm_x86_ops->queue_exception(vcpu, vcpu->arch.exception.nr,
196 vcpu->arch.exception.has_error_code,
197 vcpu->arch.exception.error_code);
201 * Load the pae pdptrs. Return true is they are all valid.
203 int load_pdptrs(struct kvm_vcpu *vcpu, unsigned long cr3)
205 gfn_t pdpt_gfn = cr3 >> PAGE_SHIFT;
206 unsigned offset = ((cr3 & (PAGE_SIZE-1)) >> 5) << 2;
209 u64 pdpte[ARRAY_SIZE(vcpu->arch.pdptrs)];
211 ret = kvm_read_guest_page(vcpu->kvm, pdpt_gfn, pdpte,
212 offset * sizeof(u64), sizeof(pdpte));
217 for (i = 0; i < ARRAY_SIZE(pdpte); ++i) {
218 if ((pdpte[i] & 1) && (pdpte[i] & 0xfffffff0000001e6ull)) {
225 memcpy(vcpu->arch.pdptrs, pdpte, sizeof(vcpu->arch.pdptrs));
230 EXPORT_SYMBOL_GPL(load_pdptrs);
232 static bool pdptrs_changed(struct kvm_vcpu *vcpu)
234 u64 pdpte[ARRAY_SIZE(vcpu->arch.pdptrs)];
238 if (is_long_mode(vcpu) || !is_pae(vcpu))
241 r = kvm_read_guest(vcpu->kvm, vcpu->arch.cr3 & ~31u, pdpte, sizeof(pdpte));
244 changed = memcmp(pdpte, vcpu->arch.pdptrs, sizeof(pdpte)) != 0;
250 void kvm_set_cr0(struct kvm_vcpu *vcpu, unsigned long cr0)
252 if (cr0 & CR0_RESERVED_BITS) {
253 printk(KERN_DEBUG "set_cr0: 0x%lx #GP, reserved bits 0x%lx\n",
254 cr0, vcpu->arch.cr0);
255 kvm_inject_gp(vcpu, 0);
259 if ((cr0 & X86_CR0_NW) && !(cr0 & X86_CR0_CD)) {
260 printk(KERN_DEBUG "set_cr0: #GP, CD == 0 && NW == 1\n");
261 kvm_inject_gp(vcpu, 0);
265 if ((cr0 & X86_CR0_PG) && !(cr0 & X86_CR0_PE)) {
266 printk(KERN_DEBUG "set_cr0: #GP, set PG flag "
267 "and a clear PE flag\n");
268 kvm_inject_gp(vcpu, 0);
272 if (!is_paging(vcpu) && (cr0 & X86_CR0_PG)) {
274 if ((vcpu->arch.shadow_efer & EFER_LME)) {
278 printk(KERN_DEBUG "set_cr0: #GP, start paging "
279 "in long mode while PAE is disabled\n");
280 kvm_inject_gp(vcpu, 0);
283 kvm_x86_ops->get_cs_db_l_bits(vcpu, &cs_db, &cs_l);
285 printk(KERN_DEBUG "set_cr0: #GP, start paging "
286 "in long mode while CS.L == 1\n");
287 kvm_inject_gp(vcpu, 0);
293 if (is_pae(vcpu) && !load_pdptrs(vcpu, vcpu->arch.cr3)) {
294 printk(KERN_DEBUG "set_cr0: #GP, pdptrs "
296 kvm_inject_gp(vcpu, 0);
302 kvm_x86_ops->set_cr0(vcpu, cr0);
303 vcpu->arch.cr0 = cr0;
305 kvm_mmu_reset_context(vcpu);
308 EXPORT_SYMBOL_GPL(kvm_set_cr0);
310 void kvm_lmsw(struct kvm_vcpu *vcpu, unsigned long msw)
312 kvm_set_cr0(vcpu, (vcpu->arch.cr0 & ~0x0ful) | (msw & 0x0f));
313 KVMTRACE_1D(LMSW, vcpu,
314 (u32)((vcpu->arch.cr0 & ~0x0ful) | (msw & 0x0f)),
317 EXPORT_SYMBOL_GPL(kvm_lmsw);
319 void kvm_set_cr4(struct kvm_vcpu *vcpu, unsigned long cr4)
321 if (cr4 & CR4_RESERVED_BITS) {
322 printk(KERN_DEBUG "set_cr4: #GP, reserved bits\n");
323 kvm_inject_gp(vcpu, 0);
327 if (is_long_mode(vcpu)) {
328 if (!(cr4 & X86_CR4_PAE)) {
329 printk(KERN_DEBUG "set_cr4: #GP, clearing PAE while "
331 kvm_inject_gp(vcpu, 0);
334 } else if (is_paging(vcpu) && !is_pae(vcpu) && (cr4 & X86_CR4_PAE)
335 && !load_pdptrs(vcpu, vcpu->arch.cr3)) {
336 printk(KERN_DEBUG "set_cr4: #GP, pdptrs reserved bits\n");
337 kvm_inject_gp(vcpu, 0);
341 if (cr4 & X86_CR4_VMXE) {
342 printk(KERN_DEBUG "set_cr4: #GP, setting VMXE\n");
343 kvm_inject_gp(vcpu, 0);
346 kvm_x86_ops->set_cr4(vcpu, cr4);
347 vcpu->arch.cr4 = cr4;
348 kvm_mmu_reset_context(vcpu);
350 EXPORT_SYMBOL_GPL(kvm_set_cr4);
352 void kvm_set_cr3(struct kvm_vcpu *vcpu, unsigned long cr3)
354 if (cr3 == vcpu->arch.cr3 && !pdptrs_changed(vcpu)) {
355 kvm_mmu_flush_tlb(vcpu);
359 if (is_long_mode(vcpu)) {
360 if (cr3 & CR3_L_MODE_RESERVED_BITS) {
361 printk(KERN_DEBUG "set_cr3: #GP, reserved bits\n");
362 kvm_inject_gp(vcpu, 0);
367 if (cr3 & CR3_PAE_RESERVED_BITS) {
369 "set_cr3: #GP, reserved bits\n");
370 kvm_inject_gp(vcpu, 0);
373 if (is_paging(vcpu) && !load_pdptrs(vcpu, cr3)) {
374 printk(KERN_DEBUG "set_cr3: #GP, pdptrs "
376 kvm_inject_gp(vcpu, 0);
381 * We don't check reserved bits in nonpae mode, because
382 * this isn't enforced, and VMware depends on this.
387 * Does the new cr3 value map to physical memory? (Note, we
388 * catch an invalid cr3 even in real-mode, because it would
389 * cause trouble later on when we turn on paging anyway.)
391 * A real CPU would silently accept an invalid cr3 and would
392 * attempt to use it - with largely undefined (and often hard
393 * to debug) behavior on the guest side.
395 if (unlikely(!gfn_to_memslot(vcpu->kvm, cr3 >> PAGE_SHIFT)))
396 kvm_inject_gp(vcpu, 0);
398 vcpu->arch.cr3 = cr3;
399 vcpu->arch.mmu.new_cr3(vcpu);
402 EXPORT_SYMBOL_GPL(kvm_set_cr3);
404 void kvm_set_cr8(struct kvm_vcpu *vcpu, unsigned long cr8)
406 if (cr8 & CR8_RESERVED_BITS) {
407 printk(KERN_DEBUG "set_cr8: #GP, reserved bits 0x%lx\n", cr8);
408 kvm_inject_gp(vcpu, 0);
411 if (irqchip_in_kernel(vcpu->kvm))
412 kvm_lapic_set_tpr(vcpu, cr8);
414 vcpu->arch.cr8 = cr8;
416 EXPORT_SYMBOL_GPL(kvm_set_cr8);
418 unsigned long kvm_get_cr8(struct kvm_vcpu *vcpu)
420 if (irqchip_in_kernel(vcpu->kvm))
421 return kvm_lapic_get_cr8(vcpu);
423 return vcpu->arch.cr8;
425 EXPORT_SYMBOL_GPL(kvm_get_cr8);
428 * List of msr numbers which we expose to userspace through KVM_GET_MSRS
429 * and KVM_SET_MSRS, and KVM_GET_MSR_INDEX_LIST.
431 * This list is modified at module load time to reflect the
432 * capabilities of the host cpu.
434 static u32 msrs_to_save[] = {
435 MSR_IA32_SYSENTER_CS, MSR_IA32_SYSENTER_ESP, MSR_IA32_SYSENTER_EIP,
438 MSR_CSTAR, MSR_KERNEL_GS_BASE, MSR_SYSCALL_MASK, MSR_LSTAR,
440 MSR_IA32_TIME_STAMP_COUNTER, MSR_KVM_SYSTEM_TIME, MSR_KVM_WALL_CLOCK,
441 MSR_IA32_PERF_STATUS,
444 static unsigned num_msrs_to_save;
446 static u32 emulated_msrs[] = {
447 MSR_IA32_MISC_ENABLE,
450 static void set_efer(struct kvm_vcpu *vcpu, u64 efer)
452 if (efer & efer_reserved_bits) {
453 printk(KERN_DEBUG "set_efer: 0x%llx #GP, reserved bits\n",
455 kvm_inject_gp(vcpu, 0);
460 && (vcpu->arch.shadow_efer & EFER_LME) != (efer & EFER_LME)) {
461 printk(KERN_DEBUG "set_efer: #GP, change LME while paging\n");
462 kvm_inject_gp(vcpu, 0);
466 kvm_x86_ops->set_efer(vcpu, efer);
469 efer |= vcpu->arch.shadow_efer & EFER_LMA;
471 vcpu->arch.shadow_efer = efer;
474 void kvm_enable_efer_bits(u64 mask)
476 efer_reserved_bits &= ~mask;
478 EXPORT_SYMBOL_GPL(kvm_enable_efer_bits);
482 * Writes msr value into into the appropriate "register".
483 * Returns 0 on success, non-0 otherwise.
484 * Assumes vcpu_load() was already called.
486 int kvm_set_msr(struct kvm_vcpu *vcpu, u32 msr_index, u64 data)
488 return kvm_x86_ops->set_msr(vcpu, msr_index, data);
492 * Adapt set_msr() to msr_io()'s calling convention
494 static int do_set_msr(struct kvm_vcpu *vcpu, unsigned index, u64 *data)
496 return kvm_set_msr(vcpu, index, *data);
499 static void kvm_write_wall_clock(struct kvm *kvm, gpa_t wall_clock)
502 struct pvclock_wall_clock wc;
503 struct timespec now, sys, boot;
510 kvm_write_guest(kvm, wall_clock, &version, sizeof(version));
513 * The guest calculates current wall clock time by adding
514 * system time (updated by kvm_write_guest_time below) to the
515 * wall clock specified here. guest system time equals host
516 * system time for us, thus we must fill in host boot time here.
518 now = current_kernel_time();
520 boot = ns_to_timespec(timespec_to_ns(&now) - timespec_to_ns(&sys));
522 wc.sec = boot.tv_sec;
523 wc.nsec = boot.tv_nsec;
524 wc.version = version;
526 kvm_write_guest(kvm, wall_clock, &wc, sizeof(wc));
529 kvm_write_guest(kvm, wall_clock, &version, sizeof(version));
532 static uint32_t div_frac(uint32_t dividend, uint32_t divisor)
534 uint32_t quotient, remainder;
536 /* Don't try to replace with do_div(), this one calculates
537 * "(dividend << 32) / divisor" */
539 : "=a" (quotient), "=d" (remainder)
540 : "0" (0), "1" (dividend), "r" (divisor) );
544 static void kvm_set_time_scale(uint32_t tsc_khz, struct pvclock_vcpu_time_info *hv_clock)
546 uint64_t nsecs = 1000000000LL;
551 tps64 = tsc_khz * 1000LL;
552 while (tps64 > nsecs*2) {
557 tps32 = (uint32_t)tps64;
558 while (tps32 <= (uint32_t)nsecs) {
563 hv_clock->tsc_shift = shift;
564 hv_clock->tsc_to_system_mul = div_frac(nsecs, tps32);
566 pr_debug("%s: tsc_khz %u, tsc_shift %d, tsc_mul %u\n",
567 __FUNCTION__, tsc_khz, hv_clock->tsc_shift,
568 hv_clock->tsc_to_system_mul);
571 static void kvm_write_guest_time(struct kvm_vcpu *v)
575 struct kvm_vcpu_arch *vcpu = &v->arch;
578 if ((!vcpu->time_page))
581 if (unlikely(vcpu->hv_clock_tsc_khz != tsc_khz)) {
582 kvm_set_time_scale(tsc_khz, &vcpu->hv_clock);
583 vcpu->hv_clock_tsc_khz = tsc_khz;
586 /* Keep irq disabled to prevent changes to the clock */
587 local_irq_save(flags);
588 kvm_get_msr(v, MSR_IA32_TIME_STAMP_COUNTER,
589 &vcpu->hv_clock.tsc_timestamp);
591 local_irq_restore(flags);
593 /* With all the info we got, fill in the values */
595 vcpu->hv_clock.system_time = ts.tv_nsec +
596 (NSEC_PER_SEC * (u64)ts.tv_sec);
598 * The interface expects us to write an even number signaling that the
599 * update is finished. Since the guest won't see the intermediate
600 * state, we just increase by 2 at the end.
602 vcpu->hv_clock.version += 2;
604 shared_kaddr = kmap_atomic(vcpu->time_page, KM_USER0);
606 memcpy(shared_kaddr + vcpu->time_offset, &vcpu->hv_clock,
607 sizeof(vcpu->hv_clock));
609 kunmap_atomic(shared_kaddr, KM_USER0);
611 mark_page_dirty(v->kvm, vcpu->time >> PAGE_SHIFT);
614 static bool msr_mtrr_valid(unsigned msr)
617 case 0x200 ... 0x200 + 2 * KVM_NR_VAR_MTRR - 1:
618 case MSR_MTRRfix64K_00000:
619 case MSR_MTRRfix16K_80000:
620 case MSR_MTRRfix16K_A0000:
621 case MSR_MTRRfix4K_C0000:
622 case MSR_MTRRfix4K_C8000:
623 case MSR_MTRRfix4K_D0000:
624 case MSR_MTRRfix4K_D8000:
625 case MSR_MTRRfix4K_E0000:
626 case MSR_MTRRfix4K_E8000:
627 case MSR_MTRRfix4K_F0000:
628 case MSR_MTRRfix4K_F8000:
629 case MSR_MTRRdefType:
630 case MSR_IA32_CR_PAT:
638 static int set_msr_mtrr(struct kvm_vcpu *vcpu, u32 msr, u64 data)
640 if (!msr_mtrr_valid(msr))
643 vcpu->arch.mtrr[msr - 0x200] = data;
647 int kvm_set_msr_common(struct kvm_vcpu *vcpu, u32 msr, u64 data)
651 set_efer(vcpu, data);
653 case MSR_IA32_MC0_STATUS:
654 pr_unimpl(vcpu, "%s: MSR_IA32_MC0_STATUS 0x%llx, nop\n",
657 case MSR_IA32_MCG_STATUS:
658 pr_unimpl(vcpu, "%s: MSR_IA32_MCG_STATUS 0x%llx, nop\n",
661 case MSR_IA32_MCG_CTL:
662 pr_unimpl(vcpu, "%s: MSR_IA32_MCG_CTL 0x%llx, nop\n",
665 case MSR_IA32_UCODE_REV:
666 case MSR_IA32_UCODE_WRITE:
668 case 0x200 ... 0x2ff:
669 return set_msr_mtrr(vcpu, msr, data);
670 case MSR_IA32_APICBASE:
671 kvm_set_apic_base(vcpu, data);
673 case MSR_IA32_MISC_ENABLE:
674 vcpu->arch.ia32_misc_enable_msr = data;
676 case MSR_KVM_WALL_CLOCK:
677 vcpu->kvm->arch.wall_clock = data;
678 kvm_write_wall_clock(vcpu->kvm, data);
680 case MSR_KVM_SYSTEM_TIME: {
681 if (vcpu->arch.time_page) {
682 kvm_release_page_dirty(vcpu->arch.time_page);
683 vcpu->arch.time_page = NULL;
686 vcpu->arch.time = data;
688 /* we verify if the enable bit is set... */
692 /* ...but clean it before doing the actual write */
693 vcpu->arch.time_offset = data & ~(PAGE_MASK | 1);
695 down_read(¤t->mm->mmap_sem);
696 vcpu->arch.time_page =
697 gfn_to_page(vcpu->kvm, data >> PAGE_SHIFT);
698 up_read(¤t->mm->mmap_sem);
700 if (is_error_page(vcpu->arch.time_page)) {
701 kvm_release_page_clean(vcpu->arch.time_page);
702 vcpu->arch.time_page = NULL;
705 kvm_write_guest_time(vcpu);
709 pr_unimpl(vcpu, "unhandled wrmsr: 0x%x data %llx\n", msr, data);
714 EXPORT_SYMBOL_GPL(kvm_set_msr_common);
718 * Reads an msr value (of 'msr_index') into 'pdata'.
719 * Returns 0 on success, non-0 otherwise.
720 * Assumes vcpu_load() was already called.
722 int kvm_get_msr(struct kvm_vcpu *vcpu, u32 msr_index, u64 *pdata)
724 return kvm_x86_ops->get_msr(vcpu, msr_index, pdata);
727 static int get_msr_mtrr(struct kvm_vcpu *vcpu, u32 msr, u64 *pdata)
729 if (!msr_mtrr_valid(msr))
732 *pdata = vcpu->arch.mtrr[msr - 0x200];
736 int kvm_get_msr_common(struct kvm_vcpu *vcpu, u32 msr, u64 *pdata)
741 case 0xc0010010: /* SYSCFG */
742 case 0xc0010015: /* HWCR */
743 case MSR_IA32_PLATFORM_ID:
744 case MSR_IA32_P5_MC_ADDR:
745 case MSR_IA32_P5_MC_TYPE:
746 case MSR_IA32_MC0_CTL:
747 case MSR_IA32_MCG_STATUS:
748 case MSR_IA32_MCG_CAP:
749 case MSR_IA32_MCG_CTL:
750 case MSR_IA32_MC0_MISC:
751 case MSR_IA32_MC0_MISC+4:
752 case MSR_IA32_MC0_MISC+8:
753 case MSR_IA32_MC0_MISC+12:
754 case MSR_IA32_MC0_MISC+16:
755 case MSR_IA32_UCODE_REV:
756 case MSR_IA32_EBL_CR_POWERON:
760 data = 0x500 | KVM_NR_VAR_MTRR;
762 case 0x200 ... 0x2ff:
763 return get_msr_mtrr(vcpu, msr, pdata);
764 case 0xcd: /* fsb frequency */
767 case MSR_IA32_APICBASE:
768 data = kvm_get_apic_base(vcpu);
770 case MSR_IA32_MISC_ENABLE:
771 data = vcpu->arch.ia32_misc_enable_msr;
773 case MSR_IA32_PERF_STATUS:
774 /* TSC increment by tick */
777 data |= (((uint64_t)4ULL) << 40);
780 data = vcpu->arch.shadow_efer;
782 case MSR_KVM_WALL_CLOCK:
783 data = vcpu->kvm->arch.wall_clock;
785 case MSR_KVM_SYSTEM_TIME:
786 data = vcpu->arch.time;
789 pr_unimpl(vcpu, "unhandled rdmsr: 0x%x\n", msr);
795 EXPORT_SYMBOL_GPL(kvm_get_msr_common);
798 * Read or write a bunch of msrs. All parameters are kernel addresses.
800 * @return number of msrs set successfully.
802 static int __msr_io(struct kvm_vcpu *vcpu, struct kvm_msrs *msrs,
803 struct kvm_msr_entry *entries,
804 int (*do_msr)(struct kvm_vcpu *vcpu,
805 unsigned index, u64 *data))
811 down_read(&vcpu->kvm->slots_lock);
812 for (i = 0; i < msrs->nmsrs; ++i)
813 if (do_msr(vcpu, entries[i].index, &entries[i].data))
815 up_read(&vcpu->kvm->slots_lock);
823 * Read or write a bunch of msrs. Parameters are user addresses.
825 * @return number of msrs set successfully.
827 static int msr_io(struct kvm_vcpu *vcpu, struct kvm_msrs __user *user_msrs,
828 int (*do_msr)(struct kvm_vcpu *vcpu,
829 unsigned index, u64 *data),
832 struct kvm_msrs msrs;
833 struct kvm_msr_entry *entries;
838 if (copy_from_user(&msrs, user_msrs, sizeof msrs))
842 if (msrs.nmsrs >= MAX_IO_MSRS)
846 size = sizeof(struct kvm_msr_entry) * msrs.nmsrs;
847 entries = vmalloc(size);
852 if (copy_from_user(entries, user_msrs->entries, size))
855 r = n = __msr_io(vcpu, &msrs, entries, do_msr);
860 if (writeback && copy_to_user(user_msrs->entries, entries, size))
871 int kvm_dev_ioctl_check_extension(long ext)
876 case KVM_CAP_IRQCHIP:
878 case KVM_CAP_MMU_SHADOW_CACHE_CONTROL:
879 case KVM_CAP_USER_MEMORY:
880 case KVM_CAP_SET_TSS_ADDR:
881 case KVM_CAP_EXT_CPUID:
882 case KVM_CAP_CLOCKSOURCE:
884 case KVM_CAP_NOP_IO_DELAY:
885 case KVM_CAP_MP_STATE:
886 case KVM_CAP_SYNC_MMU:
889 case KVM_CAP_COALESCED_MMIO:
890 r = KVM_COALESCED_MMIO_PAGE_OFFSET;
893 r = !kvm_x86_ops->cpu_has_accelerated_tpr();
895 case KVM_CAP_NR_VCPUS:
898 case KVM_CAP_NR_MEMSLOTS:
899 r = KVM_MEMORY_SLOTS;
912 long kvm_arch_dev_ioctl(struct file *filp,
913 unsigned int ioctl, unsigned long arg)
915 void __user *argp = (void __user *)arg;
919 case KVM_GET_MSR_INDEX_LIST: {
920 struct kvm_msr_list __user *user_msr_list = argp;
921 struct kvm_msr_list msr_list;
925 if (copy_from_user(&msr_list, user_msr_list, sizeof msr_list))
928 msr_list.nmsrs = num_msrs_to_save + ARRAY_SIZE(emulated_msrs);
929 if (copy_to_user(user_msr_list, &msr_list, sizeof msr_list))
932 if (n < num_msrs_to_save)
935 if (copy_to_user(user_msr_list->indices, &msrs_to_save,
936 num_msrs_to_save * sizeof(u32)))
938 if (copy_to_user(user_msr_list->indices
939 + num_msrs_to_save * sizeof(u32),
941 ARRAY_SIZE(emulated_msrs) * sizeof(u32)))
946 case KVM_GET_SUPPORTED_CPUID: {
947 struct kvm_cpuid2 __user *cpuid_arg = argp;
948 struct kvm_cpuid2 cpuid;
951 if (copy_from_user(&cpuid, cpuid_arg, sizeof cpuid))
953 r = kvm_dev_ioctl_get_supported_cpuid(&cpuid,
959 if (copy_to_user(cpuid_arg, &cpuid, sizeof cpuid))
971 void kvm_arch_vcpu_load(struct kvm_vcpu *vcpu, int cpu)
973 kvm_x86_ops->vcpu_load(vcpu, cpu);
974 kvm_write_guest_time(vcpu);
977 void kvm_arch_vcpu_put(struct kvm_vcpu *vcpu)
979 kvm_x86_ops->vcpu_put(vcpu);
980 kvm_put_guest_fpu(vcpu);
983 static int is_efer_nx(void)
987 rdmsrl(MSR_EFER, efer);
988 return efer & EFER_NX;
991 static void cpuid_fix_nx_cap(struct kvm_vcpu *vcpu)
994 struct kvm_cpuid_entry2 *e, *entry;
997 for (i = 0; i < vcpu->arch.cpuid_nent; ++i) {
998 e = &vcpu->arch.cpuid_entries[i];
999 if (e->function == 0x80000001) {
1004 if (entry && (entry->edx & (1 << 20)) && !is_efer_nx()) {
1005 entry->edx &= ~(1 << 20);
1006 printk(KERN_INFO "kvm: guest NX capability removed\n");
1010 /* when an old userspace process fills a new kernel module */
1011 static int kvm_vcpu_ioctl_set_cpuid(struct kvm_vcpu *vcpu,
1012 struct kvm_cpuid *cpuid,
1013 struct kvm_cpuid_entry __user *entries)
1016 struct kvm_cpuid_entry *cpuid_entries;
1019 if (cpuid->nent > KVM_MAX_CPUID_ENTRIES)
1022 cpuid_entries = vmalloc(sizeof(struct kvm_cpuid_entry) * cpuid->nent);
1026 if (copy_from_user(cpuid_entries, entries,
1027 cpuid->nent * sizeof(struct kvm_cpuid_entry)))
1029 for (i = 0; i < cpuid->nent; i++) {
1030 vcpu->arch.cpuid_entries[i].function = cpuid_entries[i].function;
1031 vcpu->arch.cpuid_entries[i].eax = cpuid_entries[i].eax;
1032 vcpu->arch.cpuid_entries[i].ebx = cpuid_entries[i].ebx;
1033 vcpu->arch.cpuid_entries[i].ecx = cpuid_entries[i].ecx;
1034 vcpu->arch.cpuid_entries[i].edx = cpuid_entries[i].edx;
1035 vcpu->arch.cpuid_entries[i].index = 0;
1036 vcpu->arch.cpuid_entries[i].flags = 0;
1037 vcpu->arch.cpuid_entries[i].padding[0] = 0;
1038 vcpu->arch.cpuid_entries[i].padding[1] = 0;
1039 vcpu->arch.cpuid_entries[i].padding[2] = 0;
1041 vcpu->arch.cpuid_nent = cpuid->nent;
1042 cpuid_fix_nx_cap(vcpu);
1046 vfree(cpuid_entries);
1051 static int kvm_vcpu_ioctl_set_cpuid2(struct kvm_vcpu *vcpu,
1052 struct kvm_cpuid2 *cpuid,
1053 struct kvm_cpuid_entry2 __user *entries)
1058 if (cpuid->nent > KVM_MAX_CPUID_ENTRIES)
1061 if (copy_from_user(&vcpu->arch.cpuid_entries, entries,
1062 cpuid->nent * sizeof(struct kvm_cpuid_entry2)))
1064 vcpu->arch.cpuid_nent = cpuid->nent;
1071 static int kvm_vcpu_ioctl_get_cpuid2(struct kvm_vcpu *vcpu,
1072 struct kvm_cpuid2 *cpuid,
1073 struct kvm_cpuid_entry2 __user *entries)
1078 if (cpuid->nent < vcpu->arch.cpuid_nent)
1081 if (copy_to_user(entries, &vcpu->arch.cpuid_entries,
1082 vcpu->arch.cpuid_nent * sizeof(struct kvm_cpuid_entry2)))
1087 cpuid->nent = vcpu->arch.cpuid_nent;
1091 static inline u32 bit(int bitno)
1093 return 1 << (bitno & 31);
1096 static void do_cpuid_1_ent(struct kvm_cpuid_entry2 *entry, u32 function,
1099 entry->function = function;
1100 entry->index = index;
1101 cpuid_count(entry->function, entry->index,
1102 &entry->eax, &entry->ebx, &entry->ecx, &entry->edx);
1106 static void do_cpuid_ent(struct kvm_cpuid_entry2 *entry, u32 function,
1107 u32 index, int *nent, int maxnent)
1109 const u32 kvm_supported_word0_x86_features = bit(X86_FEATURE_FPU) |
1110 bit(X86_FEATURE_VME) | bit(X86_FEATURE_DE) |
1111 bit(X86_FEATURE_PSE) | bit(X86_FEATURE_TSC) |
1112 bit(X86_FEATURE_MSR) | bit(X86_FEATURE_PAE) |
1113 bit(X86_FEATURE_CX8) | bit(X86_FEATURE_APIC) |
1114 bit(X86_FEATURE_SEP) | bit(X86_FEATURE_PGE) |
1115 bit(X86_FEATURE_CMOV) | bit(X86_FEATURE_PSE36) |
1116 bit(X86_FEATURE_CLFLSH) | bit(X86_FEATURE_MMX) |
1117 bit(X86_FEATURE_FXSR) | bit(X86_FEATURE_XMM) |
1118 bit(X86_FEATURE_XMM2) | bit(X86_FEATURE_SELFSNOOP);
1119 const u32 kvm_supported_word1_x86_features = bit(X86_FEATURE_FPU) |
1120 bit(X86_FEATURE_VME) | bit(X86_FEATURE_DE) |
1121 bit(X86_FEATURE_PSE) | bit(X86_FEATURE_TSC) |
1122 bit(X86_FEATURE_MSR) | bit(X86_FEATURE_PAE) |
1123 bit(X86_FEATURE_CX8) | bit(X86_FEATURE_APIC) |
1124 bit(X86_FEATURE_PGE) |
1125 bit(X86_FEATURE_CMOV) | bit(X86_FEATURE_PSE36) |
1126 bit(X86_FEATURE_MMX) | bit(X86_FEATURE_FXSR) |
1127 bit(X86_FEATURE_SYSCALL) |
1128 (bit(X86_FEATURE_NX) && is_efer_nx()) |
1129 #ifdef CONFIG_X86_64
1130 bit(X86_FEATURE_LM) |
1132 bit(X86_FEATURE_MMXEXT) |
1133 bit(X86_FEATURE_3DNOWEXT) |
1134 bit(X86_FEATURE_3DNOW);
1135 const u32 kvm_supported_word3_x86_features =
1136 bit(X86_FEATURE_XMM3) | bit(X86_FEATURE_CX16);
1137 const u32 kvm_supported_word6_x86_features =
1138 bit(X86_FEATURE_LAHF_LM) | bit(X86_FEATURE_CMP_LEGACY);
1140 /* all func 2 cpuid_count() should be called on the same cpu */
1142 do_cpuid_1_ent(entry, function, index);
1147 entry->eax = min(entry->eax, (u32)0xb);
1150 entry->edx &= kvm_supported_word0_x86_features;
1151 entry->ecx &= kvm_supported_word3_x86_features;
1153 /* function 2 entries are STATEFUL. That is, repeated cpuid commands
1154 * may return different values. This forces us to get_cpu() before
1155 * issuing the first command, and also to emulate this annoying behavior
1156 * in kvm_emulate_cpuid() using KVM_CPUID_FLAG_STATE_READ_NEXT */
1158 int t, times = entry->eax & 0xff;
1160 entry->flags |= KVM_CPUID_FLAG_STATEFUL_FUNC;
1161 for (t = 1; t < times && *nent < maxnent; ++t) {
1162 do_cpuid_1_ent(&entry[t], function, 0);
1163 entry[t].flags |= KVM_CPUID_FLAG_STATEFUL_FUNC;
1168 /* function 4 and 0xb have additional index. */
1172 entry->flags |= KVM_CPUID_FLAG_SIGNIFCANT_INDEX;
1173 /* read more entries until cache_type is zero */
1174 for (i = 1; *nent < maxnent; ++i) {
1175 cache_type = entry[i - 1].eax & 0x1f;
1178 do_cpuid_1_ent(&entry[i], function, i);
1180 KVM_CPUID_FLAG_SIGNIFCANT_INDEX;
1188 entry->flags |= KVM_CPUID_FLAG_SIGNIFCANT_INDEX;
1189 /* read more entries until level_type is zero */
1190 for (i = 1; *nent < maxnent; ++i) {
1191 level_type = entry[i - 1].ecx & 0xff;
1194 do_cpuid_1_ent(&entry[i], function, i);
1196 KVM_CPUID_FLAG_SIGNIFCANT_INDEX;
1202 entry->eax = min(entry->eax, 0x8000001a);
1205 entry->edx &= kvm_supported_word1_x86_features;
1206 entry->ecx &= kvm_supported_word6_x86_features;
1212 static int kvm_dev_ioctl_get_supported_cpuid(struct kvm_cpuid2 *cpuid,
1213 struct kvm_cpuid_entry2 __user *entries)
1215 struct kvm_cpuid_entry2 *cpuid_entries;
1216 int limit, nent = 0, r = -E2BIG;
1219 if (cpuid->nent < 1)
1222 cpuid_entries = vmalloc(sizeof(struct kvm_cpuid_entry2) * cpuid->nent);
1226 do_cpuid_ent(&cpuid_entries[0], 0, 0, &nent, cpuid->nent);
1227 limit = cpuid_entries[0].eax;
1228 for (func = 1; func <= limit && nent < cpuid->nent; ++func)
1229 do_cpuid_ent(&cpuid_entries[nent], func, 0,
1230 &nent, cpuid->nent);
1232 if (nent >= cpuid->nent)
1235 do_cpuid_ent(&cpuid_entries[nent], 0x80000000, 0, &nent, cpuid->nent);
1236 limit = cpuid_entries[nent - 1].eax;
1237 for (func = 0x80000001; func <= limit && nent < cpuid->nent; ++func)
1238 do_cpuid_ent(&cpuid_entries[nent], func, 0,
1239 &nent, cpuid->nent);
1241 if (copy_to_user(entries, cpuid_entries,
1242 nent * sizeof(struct kvm_cpuid_entry2)))
1248 vfree(cpuid_entries);
1253 static int kvm_vcpu_ioctl_get_lapic(struct kvm_vcpu *vcpu,
1254 struct kvm_lapic_state *s)
1257 memcpy(s->regs, vcpu->arch.apic->regs, sizeof *s);
1263 static int kvm_vcpu_ioctl_set_lapic(struct kvm_vcpu *vcpu,
1264 struct kvm_lapic_state *s)
1267 memcpy(vcpu->arch.apic->regs, s->regs, sizeof *s);
1268 kvm_apic_post_state_restore(vcpu);
1274 static int kvm_vcpu_ioctl_interrupt(struct kvm_vcpu *vcpu,
1275 struct kvm_interrupt *irq)
1277 if (irq->irq < 0 || irq->irq >= 256)
1279 if (irqchip_in_kernel(vcpu->kvm))
1283 set_bit(irq->irq, vcpu->arch.irq_pending);
1284 set_bit(irq->irq / BITS_PER_LONG, &vcpu->arch.irq_summary);
1291 static int vcpu_ioctl_tpr_access_reporting(struct kvm_vcpu *vcpu,
1292 struct kvm_tpr_access_ctl *tac)
1296 vcpu->arch.tpr_access_reporting = !!tac->enabled;
1300 long kvm_arch_vcpu_ioctl(struct file *filp,
1301 unsigned int ioctl, unsigned long arg)
1303 struct kvm_vcpu *vcpu = filp->private_data;
1304 void __user *argp = (void __user *)arg;
1306 struct kvm_lapic_state *lapic = NULL;
1309 case KVM_GET_LAPIC: {
1310 lapic = kzalloc(sizeof(struct kvm_lapic_state), GFP_KERNEL);
1315 r = kvm_vcpu_ioctl_get_lapic(vcpu, lapic);
1319 if (copy_to_user(argp, lapic, sizeof(struct kvm_lapic_state)))
1324 case KVM_SET_LAPIC: {
1325 lapic = kmalloc(sizeof(struct kvm_lapic_state), GFP_KERNEL);
1330 if (copy_from_user(lapic, argp, sizeof(struct kvm_lapic_state)))
1332 r = kvm_vcpu_ioctl_set_lapic(vcpu, lapic);
1338 case KVM_INTERRUPT: {
1339 struct kvm_interrupt irq;
1342 if (copy_from_user(&irq, argp, sizeof irq))
1344 r = kvm_vcpu_ioctl_interrupt(vcpu, &irq);
1350 case KVM_SET_CPUID: {
1351 struct kvm_cpuid __user *cpuid_arg = argp;
1352 struct kvm_cpuid cpuid;
1355 if (copy_from_user(&cpuid, cpuid_arg, sizeof cpuid))
1357 r = kvm_vcpu_ioctl_set_cpuid(vcpu, &cpuid, cpuid_arg->entries);
1362 case KVM_SET_CPUID2: {
1363 struct kvm_cpuid2 __user *cpuid_arg = argp;
1364 struct kvm_cpuid2 cpuid;
1367 if (copy_from_user(&cpuid, cpuid_arg, sizeof cpuid))
1369 r = kvm_vcpu_ioctl_set_cpuid2(vcpu, &cpuid,
1370 cpuid_arg->entries);
1375 case KVM_GET_CPUID2: {
1376 struct kvm_cpuid2 __user *cpuid_arg = argp;
1377 struct kvm_cpuid2 cpuid;
1380 if (copy_from_user(&cpuid, cpuid_arg, sizeof cpuid))
1382 r = kvm_vcpu_ioctl_get_cpuid2(vcpu, &cpuid,
1383 cpuid_arg->entries);
1387 if (copy_to_user(cpuid_arg, &cpuid, sizeof cpuid))
1393 r = msr_io(vcpu, argp, kvm_get_msr, 1);
1396 r = msr_io(vcpu, argp, do_set_msr, 0);
1398 case KVM_TPR_ACCESS_REPORTING: {
1399 struct kvm_tpr_access_ctl tac;
1402 if (copy_from_user(&tac, argp, sizeof tac))
1404 r = vcpu_ioctl_tpr_access_reporting(vcpu, &tac);
1408 if (copy_to_user(argp, &tac, sizeof tac))
1413 case KVM_SET_VAPIC_ADDR: {
1414 struct kvm_vapic_addr va;
1417 if (!irqchip_in_kernel(vcpu->kvm))
1420 if (copy_from_user(&va, argp, sizeof va))
1423 kvm_lapic_set_vapic_addr(vcpu, va.vapic_addr);
1435 static int kvm_vm_ioctl_set_tss_addr(struct kvm *kvm, unsigned long addr)
1439 if (addr > (unsigned int)(-3 * PAGE_SIZE))
1441 ret = kvm_x86_ops->set_tss_addr(kvm, addr);
1445 static int kvm_vm_ioctl_set_nr_mmu_pages(struct kvm *kvm,
1446 u32 kvm_nr_mmu_pages)
1448 if (kvm_nr_mmu_pages < KVM_MIN_ALLOC_MMU_PAGES)
1451 down_write(&kvm->slots_lock);
1453 kvm_mmu_change_mmu_pages(kvm, kvm_nr_mmu_pages);
1454 kvm->arch.n_requested_mmu_pages = kvm_nr_mmu_pages;
1456 up_write(&kvm->slots_lock);
1460 static int kvm_vm_ioctl_get_nr_mmu_pages(struct kvm *kvm)
1462 return kvm->arch.n_alloc_mmu_pages;
1465 gfn_t unalias_gfn(struct kvm *kvm, gfn_t gfn)
1468 struct kvm_mem_alias *alias;
1470 for (i = 0; i < kvm->arch.naliases; ++i) {
1471 alias = &kvm->arch.aliases[i];
1472 if (gfn >= alias->base_gfn
1473 && gfn < alias->base_gfn + alias->npages)
1474 return alias->target_gfn + gfn - alias->base_gfn;
1480 * Set a new alias region. Aliases map a portion of physical memory into
1481 * another portion. This is useful for memory windows, for example the PC
1484 static int kvm_vm_ioctl_set_memory_alias(struct kvm *kvm,
1485 struct kvm_memory_alias *alias)
1488 struct kvm_mem_alias *p;
1491 /* General sanity checks */
1492 if (alias->memory_size & (PAGE_SIZE - 1))
1494 if (alias->guest_phys_addr & (PAGE_SIZE - 1))
1496 if (alias->slot >= KVM_ALIAS_SLOTS)
1498 if (alias->guest_phys_addr + alias->memory_size
1499 < alias->guest_phys_addr)
1501 if (alias->target_phys_addr + alias->memory_size
1502 < alias->target_phys_addr)
1505 down_write(&kvm->slots_lock);
1506 spin_lock(&kvm->mmu_lock);
1508 p = &kvm->arch.aliases[alias->slot];
1509 p->base_gfn = alias->guest_phys_addr >> PAGE_SHIFT;
1510 p->npages = alias->memory_size >> PAGE_SHIFT;
1511 p->target_gfn = alias->target_phys_addr >> PAGE_SHIFT;
1513 for (n = KVM_ALIAS_SLOTS; n > 0; --n)
1514 if (kvm->arch.aliases[n - 1].npages)
1516 kvm->arch.naliases = n;
1518 spin_unlock(&kvm->mmu_lock);
1519 kvm_mmu_zap_all(kvm);
1521 up_write(&kvm->slots_lock);
1529 static int kvm_vm_ioctl_get_irqchip(struct kvm *kvm, struct kvm_irqchip *chip)
1534 switch (chip->chip_id) {
1535 case KVM_IRQCHIP_PIC_MASTER:
1536 memcpy(&chip->chip.pic,
1537 &pic_irqchip(kvm)->pics[0],
1538 sizeof(struct kvm_pic_state));
1540 case KVM_IRQCHIP_PIC_SLAVE:
1541 memcpy(&chip->chip.pic,
1542 &pic_irqchip(kvm)->pics[1],
1543 sizeof(struct kvm_pic_state));
1545 case KVM_IRQCHIP_IOAPIC:
1546 memcpy(&chip->chip.ioapic,
1547 ioapic_irqchip(kvm),
1548 sizeof(struct kvm_ioapic_state));
1557 static int kvm_vm_ioctl_set_irqchip(struct kvm *kvm, struct kvm_irqchip *chip)
1562 switch (chip->chip_id) {
1563 case KVM_IRQCHIP_PIC_MASTER:
1564 memcpy(&pic_irqchip(kvm)->pics[0],
1566 sizeof(struct kvm_pic_state));
1568 case KVM_IRQCHIP_PIC_SLAVE:
1569 memcpy(&pic_irqchip(kvm)->pics[1],
1571 sizeof(struct kvm_pic_state));
1573 case KVM_IRQCHIP_IOAPIC:
1574 memcpy(ioapic_irqchip(kvm),
1576 sizeof(struct kvm_ioapic_state));
1582 kvm_pic_update_irq(pic_irqchip(kvm));
1586 static int kvm_vm_ioctl_get_pit(struct kvm *kvm, struct kvm_pit_state *ps)
1590 memcpy(ps, &kvm->arch.vpit->pit_state, sizeof(struct kvm_pit_state));
1594 static int kvm_vm_ioctl_set_pit(struct kvm *kvm, struct kvm_pit_state *ps)
1598 memcpy(&kvm->arch.vpit->pit_state, ps, sizeof(struct kvm_pit_state));
1599 kvm_pit_load_count(kvm, 0, ps->channels[0].count);
1604 * Get (and clear) the dirty memory log for a memory slot.
1606 int kvm_vm_ioctl_get_dirty_log(struct kvm *kvm,
1607 struct kvm_dirty_log *log)
1611 struct kvm_memory_slot *memslot;
1614 down_write(&kvm->slots_lock);
1616 r = kvm_get_dirty_log(kvm, log, &is_dirty);
1620 /* If nothing is dirty, don't bother messing with page tables. */
1622 kvm_mmu_slot_remove_write_access(kvm, log->slot);
1623 kvm_flush_remote_tlbs(kvm);
1624 memslot = &kvm->memslots[log->slot];
1625 n = ALIGN(memslot->npages, BITS_PER_LONG) / 8;
1626 memset(memslot->dirty_bitmap, 0, n);
1630 up_write(&kvm->slots_lock);
1634 long kvm_arch_vm_ioctl(struct file *filp,
1635 unsigned int ioctl, unsigned long arg)
1637 struct kvm *kvm = filp->private_data;
1638 void __user *argp = (void __user *)arg;
1641 * This union makes it completely explicit to gcc-3.x
1642 * that these two variables' stack usage should be
1643 * combined, not added together.
1646 struct kvm_pit_state ps;
1647 struct kvm_memory_alias alias;
1651 case KVM_SET_TSS_ADDR:
1652 r = kvm_vm_ioctl_set_tss_addr(kvm, arg);
1656 case KVM_SET_MEMORY_REGION: {
1657 struct kvm_memory_region kvm_mem;
1658 struct kvm_userspace_memory_region kvm_userspace_mem;
1661 if (copy_from_user(&kvm_mem, argp, sizeof kvm_mem))
1663 kvm_userspace_mem.slot = kvm_mem.slot;
1664 kvm_userspace_mem.flags = kvm_mem.flags;
1665 kvm_userspace_mem.guest_phys_addr = kvm_mem.guest_phys_addr;
1666 kvm_userspace_mem.memory_size = kvm_mem.memory_size;
1667 r = kvm_vm_ioctl_set_memory_region(kvm, &kvm_userspace_mem, 0);
1672 case KVM_SET_NR_MMU_PAGES:
1673 r = kvm_vm_ioctl_set_nr_mmu_pages(kvm, arg);
1677 case KVM_GET_NR_MMU_PAGES:
1678 r = kvm_vm_ioctl_get_nr_mmu_pages(kvm);
1680 case KVM_SET_MEMORY_ALIAS:
1682 if (copy_from_user(&u.alias, argp, sizeof(struct kvm_memory_alias)))
1684 r = kvm_vm_ioctl_set_memory_alias(kvm, &u.alias);
1688 case KVM_CREATE_IRQCHIP:
1690 kvm->arch.vpic = kvm_create_pic(kvm);
1691 if (kvm->arch.vpic) {
1692 r = kvm_ioapic_init(kvm);
1694 kfree(kvm->arch.vpic);
1695 kvm->arch.vpic = NULL;
1701 case KVM_CREATE_PIT:
1703 kvm->arch.vpit = kvm_create_pit(kvm);
1707 case KVM_IRQ_LINE: {
1708 struct kvm_irq_level irq_event;
1711 if (copy_from_user(&irq_event, argp, sizeof irq_event))
1713 if (irqchip_in_kernel(kvm)) {
1714 mutex_lock(&kvm->lock);
1715 if (irq_event.irq < 16)
1716 kvm_pic_set_irq(pic_irqchip(kvm),
1719 kvm_ioapic_set_irq(kvm->arch.vioapic,
1722 mutex_unlock(&kvm->lock);
1727 case KVM_GET_IRQCHIP: {
1728 /* 0: PIC master, 1: PIC slave, 2: IOAPIC */
1729 struct kvm_irqchip *chip = kmalloc(sizeof(*chip), GFP_KERNEL);
1735 if (copy_from_user(chip, argp, sizeof *chip))
1736 goto get_irqchip_out;
1738 if (!irqchip_in_kernel(kvm))
1739 goto get_irqchip_out;
1740 r = kvm_vm_ioctl_get_irqchip(kvm, chip);
1742 goto get_irqchip_out;
1744 if (copy_to_user(argp, chip, sizeof *chip))
1745 goto get_irqchip_out;
1753 case KVM_SET_IRQCHIP: {
1754 /* 0: PIC master, 1: PIC slave, 2: IOAPIC */
1755 struct kvm_irqchip *chip = kmalloc(sizeof(*chip), GFP_KERNEL);
1761 if (copy_from_user(chip, argp, sizeof *chip))
1762 goto set_irqchip_out;
1764 if (!irqchip_in_kernel(kvm))
1765 goto set_irqchip_out;
1766 r = kvm_vm_ioctl_set_irqchip(kvm, chip);
1768 goto set_irqchip_out;
1778 if (copy_from_user(&u.ps, argp, sizeof(struct kvm_pit_state)))
1781 if (!kvm->arch.vpit)
1783 r = kvm_vm_ioctl_get_pit(kvm, &u.ps);
1787 if (copy_to_user(argp, &u.ps, sizeof(struct kvm_pit_state)))
1794 if (copy_from_user(&u.ps, argp, sizeof u.ps))
1797 if (!kvm->arch.vpit)
1799 r = kvm_vm_ioctl_set_pit(kvm, &u.ps);
1812 static void kvm_init_msr_list(void)
1817 for (i = j = 0; i < ARRAY_SIZE(msrs_to_save); i++) {
1818 if (rdmsr_safe(msrs_to_save[i], &dummy[0], &dummy[1]) < 0)
1821 msrs_to_save[j] = msrs_to_save[i];
1824 num_msrs_to_save = j;
1828 * Only apic need an MMIO device hook, so shortcut now..
1830 static struct kvm_io_device *vcpu_find_pervcpu_dev(struct kvm_vcpu *vcpu,
1831 gpa_t addr, int len,
1834 struct kvm_io_device *dev;
1836 if (vcpu->arch.apic) {
1837 dev = &vcpu->arch.apic->dev;
1838 if (dev->in_range(dev, addr, len, is_write))
1845 static struct kvm_io_device *vcpu_find_mmio_dev(struct kvm_vcpu *vcpu,
1846 gpa_t addr, int len,
1849 struct kvm_io_device *dev;
1851 dev = vcpu_find_pervcpu_dev(vcpu, addr, len, is_write);
1853 dev = kvm_io_bus_find_dev(&vcpu->kvm->mmio_bus, addr, len,
1858 int emulator_read_std(unsigned long addr,
1861 struct kvm_vcpu *vcpu)
1864 int r = X86EMUL_CONTINUE;
1867 gpa_t gpa = vcpu->arch.mmu.gva_to_gpa(vcpu, addr);
1868 unsigned offset = addr & (PAGE_SIZE-1);
1869 unsigned tocopy = min(bytes, (unsigned)PAGE_SIZE - offset);
1872 if (gpa == UNMAPPED_GVA) {
1873 r = X86EMUL_PROPAGATE_FAULT;
1876 ret = kvm_read_guest(vcpu->kvm, gpa, data, tocopy);
1878 r = X86EMUL_UNHANDLEABLE;
1889 EXPORT_SYMBOL_GPL(emulator_read_std);
1891 static int emulator_read_emulated(unsigned long addr,
1894 struct kvm_vcpu *vcpu)
1896 struct kvm_io_device *mmio_dev;
1899 if (vcpu->mmio_read_completed) {
1900 memcpy(val, vcpu->mmio_data, bytes);
1901 vcpu->mmio_read_completed = 0;
1902 return X86EMUL_CONTINUE;
1905 gpa = vcpu->arch.mmu.gva_to_gpa(vcpu, addr);
1907 /* For APIC access vmexit */
1908 if ((gpa & PAGE_MASK) == APIC_DEFAULT_PHYS_BASE)
1911 if (emulator_read_std(addr, val, bytes, vcpu)
1912 == X86EMUL_CONTINUE)
1913 return X86EMUL_CONTINUE;
1914 if (gpa == UNMAPPED_GVA)
1915 return X86EMUL_PROPAGATE_FAULT;
1919 * Is this MMIO handled locally?
1921 mutex_lock(&vcpu->kvm->lock);
1922 mmio_dev = vcpu_find_mmio_dev(vcpu, gpa, bytes, 0);
1924 kvm_iodevice_read(mmio_dev, gpa, bytes, val);
1925 mutex_unlock(&vcpu->kvm->lock);
1926 return X86EMUL_CONTINUE;
1928 mutex_unlock(&vcpu->kvm->lock);
1930 vcpu->mmio_needed = 1;
1931 vcpu->mmio_phys_addr = gpa;
1932 vcpu->mmio_size = bytes;
1933 vcpu->mmio_is_write = 0;
1935 return X86EMUL_UNHANDLEABLE;
1938 int emulator_write_phys(struct kvm_vcpu *vcpu, gpa_t gpa,
1939 const void *val, int bytes)
1943 ret = kvm_write_guest(vcpu->kvm, gpa, val, bytes);
1946 kvm_mmu_pte_write(vcpu, gpa, val, bytes);
1950 static int emulator_write_emulated_onepage(unsigned long addr,
1953 struct kvm_vcpu *vcpu)
1955 struct kvm_io_device *mmio_dev;
1958 gpa = vcpu->arch.mmu.gva_to_gpa(vcpu, addr);
1960 if (gpa == UNMAPPED_GVA) {
1961 kvm_inject_page_fault(vcpu, addr, 2);
1962 return X86EMUL_PROPAGATE_FAULT;
1965 /* For APIC access vmexit */
1966 if ((gpa & PAGE_MASK) == APIC_DEFAULT_PHYS_BASE)
1969 if (emulator_write_phys(vcpu, gpa, val, bytes))
1970 return X86EMUL_CONTINUE;
1974 * Is this MMIO handled locally?
1976 mutex_lock(&vcpu->kvm->lock);
1977 mmio_dev = vcpu_find_mmio_dev(vcpu, gpa, bytes, 1);
1979 kvm_iodevice_write(mmio_dev, gpa, bytes, val);
1980 mutex_unlock(&vcpu->kvm->lock);
1981 return X86EMUL_CONTINUE;
1983 mutex_unlock(&vcpu->kvm->lock);
1985 vcpu->mmio_needed = 1;
1986 vcpu->mmio_phys_addr = gpa;
1987 vcpu->mmio_size = bytes;
1988 vcpu->mmio_is_write = 1;
1989 memcpy(vcpu->mmio_data, val, bytes);
1991 return X86EMUL_CONTINUE;
1994 int emulator_write_emulated(unsigned long addr,
1997 struct kvm_vcpu *vcpu)
1999 /* Crossing a page boundary? */
2000 if (((addr + bytes - 1) ^ addr) & PAGE_MASK) {
2003 now = -addr & ~PAGE_MASK;
2004 rc = emulator_write_emulated_onepage(addr, val, now, vcpu);
2005 if (rc != X86EMUL_CONTINUE)
2011 return emulator_write_emulated_onepage(addr, val, bytes, vcpu);
2013 EXPORT_SYMBOL_GPL(emulator_write_emulated);
2015 static int emulator_cmpxchg_emulated(unsigned long addr,
2019 struct kvm_vcpu *vcpu)
2021 static int reported;
2025 printk(KERN_WARNING "kvm: emulating exchange as write\n");
2027 #ifndef CONFIG_X86_64
2028 /* guests cmpxchg8b have to be emulated atomically */
2035 gpa = vcpu->arch.mmu.gva_to_gpa(vcpu, addr);
2037 if (gpa == UNMAPPED_GVA ||
2038 (gpa & PAGE_MASK) == APIC_DEFAULT_PHYS_BASE)
2041 if (((gpa + bytes - 1) & PAGE_MASK) != (gpa & PAGE_MASK))
2046 down_read(¤t->mm->mmap_sem);
2047 page = gfn_to_page(vcpu->kvm, gpa >> PAGE_SHIFT);
2048 up_read(¤t->mm->mmap_sem);
2050 kaddr = kmap_atomic(page, KM_USER0);
2051 set_64bit((u64 *)(kaddr + offset_in_page(gpa)), val);
2052 kunmap_atomic(kaddr, KM_USER0);
2053 kvm_release_page_dirty(page);
2058 return emulator_write_emulated(addr, new, bytes, vcpu);
2061 static unsigned long get_segment_base(struct kvm_vcpu *vcpu, int seg)
2063 return kvm_x86_ops->get_segment_base(vcpu, seg);
2066 int emulate_invlpg(struct kvm_vcpu *vcpu, gva_t address)
2068 return X86EMUL_CONTINUE;
2071 int emulate_clts(struct kvm_vcpu *vcpu)
2073 KVMTRACE_0D(CLTS, vcpu, handler);
2074 kvm_x86_ops->set_cr0(vcpu, vcpu->arch.cr0 & ~X86_CR0_TS);
2075 return X86EMUL_CONTINUE;
2078 int emulator_get_dr(struct x86_emulate_ctxt *ctxt, int dr, unsigned long *dest)
2080 struct kvm_vcpu *vcpu = ctxt->vcpu;
2084 *dest = kvm_x86_ops->get_dr(vcpu, dr);
2085 return X86EMUL_CONTINUE;
2087 pr_unimpl(vcpu, "%s: unexpected dr %u\n", __func__, dr);
2088 return X86EMUL_UNHANDLEABLE;
2092 int emulator_set_dr(struct x86_emulate_ctxt *ctxt, int dr, unsigned long value)
2094 unsigned long mask = (ctxt->mode == X86EMUL_MODE_PROT64) ? ~0ULL : ~0U;
2097 kvm_x86_ops->set_dr(ctxt->vcpu, dr, value & mask, &exception);
2099 /* FIXME: better handling */
2100 return X86EMUL_UNHANDLEABLE;
2102 return X86EMUL_CONTINUE;
2105 void kvm_report_emulation_failure(struct kvm_vcpu *vcpu, const char *context)
2108 unsigned long rip = vcpu->arch.rip;
2109 unsigned long rip_linear;
2111 if (!printk_ratelimit())
2114 rip_linear = rip + get_segment_base(vcpu, VCPU_SREG_CS);
2116 emulator_read_std(rip_linear, (void *)opcodes, 4, vcpu);
2118 printk(KERN_ERR "emulation failed (%s) rip %lx %02x %02x %02x %02x\n",
2119 context, rip, opcodes[0], opcodes[1], opcodes[2], opcodes[3]);
2121 EXPORT_SYMBOL_GPL(kvm_report_emulation_failure);
2123 static struct x86_emulate_ops emulate_ops = {
2124 .read_std = emulator_read_std,
2125 .read_emulated = emulator_read_emulated,
2126 .write_emulated = emulator_write_emulated,
2127 .cmpxchg_emulated = emulator_cmpxchg_emulated,
2130 int emulate_instruction(struct kvm_vcpu *vcpu,
2131 struct kvm_run *run,
2137 struct decode_cache *c;
2139 vcpu->arch.mmio_fault_cr2 = cr2;
2140 kvm_x86_ops->cache_regs(vcpu);
2142 vcpu->mmio_is_write = 0;
2143 vcpu->arch.pio.string = 0;
2145 if (!(emulation_type & EMULTYPE_NO_DECODE)) {
2147 kvm_x86_ops->get_cs_db_l_bits(vcpu, &cs_db, &cs_l);
2149 vcpu->arch.emulate_ctxt.vcpu = vcpu;
2150 vcpu->arch.emulate_ctxt.eflags = kvm_x86_ops->get_rflags(vcpu);
2151 vcpu->arch.emulate_ctxt.mode =
2152 (vcpu->arch.emulate_ctxt.eflags & X86_EFLAGS_VM)
2153 ? X86EMUL_MODE_REAL : cs_l
2154 ? X86EMUL_MODE_PROT64 : cs_db
2155 ? X86EMUL_MODE_PROT32 : X86EMUL_MODE_PROT16;
2157 r = x86_decode_insn(&vcpu->arch.emulate_ctxt, &emulate_ops);
2159 /* Reject the instructions other than VMCALL/VMMCALL when
2160 * try to emulate invalid opcode */
2161 c = &vcpu->arch.emulate_ctxt.decode;
2162 if ((emulation_type & EMULTYPE_TRAP_UD) &&
2163 (!(c->twobyte && c->b == 0x01 &&
2164 (c->modrm_reg == 0 || c->modrm_reg == 3) &&
2165 c->modrm_mod == 3 && c->modrm_rm == 1)))
2166 return EMULATE_FAIL;
2168 ++vcpu->stat.insn_emulation;
2170 ++vcpu->stat.insn_emulation_fail;
2171 if (kvm_mmu_unprotect_page_virt(vcpu, cr2))
2172 return EMULATE_DONE;
2173 return EMULATE_FAIL;
2177 r = x86_emulate_insn(&vcpu->arch.emulate_ctxt, &emulate_ops);
2179 if (vcpu->arch.pio.string)
2180 return EMULATE_DO_MMIO;
2182 if ((r || vcpu->mmio_is_write) && run) {
2183 run->exit_reason = KVM_EXIT_MMIO;
2184 run->mmio.phys_addr = vcpu->mmio_phys_addr;
2185 memcpy(run->mmio.data, vcpu->mmio_data, 8);
2186 run->mmio.len = vcpu->mmio_size;
2187 run->mmio.is_write = vcpu->mmio_is_write;
2191 if (kvm_mmu_unprotect_page_virt(vcpu, cr2))
2192 return EMULATE_DONE;
2193 if (!vcpu->mmio_needed) {
2194 kvm_report_emulation_failure(vcpu, "mmio");
2195 return EMULATE_FAIL;
2197 return EMULATE_DO_MMIO;
2200 kvm_x86_ops->decache_regs(vcpu);
2201 kvm_x86_ops->set_rflags(vcpu, vcpu->arch.emulate_ctxt.eflags);
2203 if (vcpu->mmio_is_write) {
2204 vcpu->mmio_needed = 0;
2205 return EMULATE_DO_MMIO;
2208 return EMULATE_DONE;
2210 EXPORT_SYMBOL_GPL(emulate_instruction);
2212 static void free_pio_guest_pages(struct kvm_vcpu *vcpu)
2216 for (i = 0; i < ARRAY_SIZE(vcpu->arch.pio.guest_pages); ++i)
2217 if (vcpu->arch.pio.guest_pages[i]) {
2218 kvm_release_page_dirty(vcpu->arch.pio.guest_pages[i]);
2219 vcpu->arch.pio.guest_pages[i] = NULL;
2223 static int pio_copy_data(struct kvm_vcpu *vcpu)
2225 void *p = vcpu->arch.pio_data;
2228 int nr_pages = vcpu->arch.pio.guest_pages[1] ? 2 : 1;
2230 q = vmap(vcpu->arch.pio.guest_pages, nr_pages, VM_READ|VM_WRITE,
2233 free_pio_guest_pages(vcpu);
2236 q += vcpu->arch.pio.guest_page_offset;
2237 bytes = vcpu->arch.pio.size * vcpu->arch.pio.cur_count;
2238 if (vcpu->arch.pio.in)
2239 memcpy(q, p, bytes);
2241 memcpy(p, q, bytes);
2242 q -= vcpu->arch.pio.guest_page_offset;
2244 free_pio_guest_pages(vcpu);
2248 int complete_pio(struct kvm_vcpu *vcpu)
2250 struct kvm_pio_request *io = &vcpu->arch.pio;
2254 kvm_x86_ops->cache_regs(vcpu);
2258 memcpy(&vcpu->arch.regs[VCPU_REGS_RAX], vcpu->arch.pio_data,
2262 r = pio_copy_data(vcpu);
2264 kvm_x86_ops->cache_regs(vcpu);
2271 delta *= io->cur_count;
2273 * The size of the register should really depend on
2274 * current address size.
2276 vcpu->arch.regs[VCPU_REGS_RCX] -= delta;
2282 vcpu->arch.regs[VCPU_REGS_RDI] += delta;
2284 vcpu->arch.regs[VCPU_REGS_RSI] += delta;
2287 kvm_x86_ops->decache_regs(vcpu);
2289 io->count -= io->cur_count;
2295 static void kernel_pio(struct kvm_io_device *pio_dev,
2296 struct kvm_vcpu *vcpu,
2299 /* TODO: String I/O for in kernel device */
2301 mutex_lock(&vcpu->kvm->lock);
2302 if (vcpu->arch.pio.in)
2303 kvm_iodevice_read(pio_dev, vcpu->arch.pio.port,
2304 vcpu->arch.pio.size,
2307 kvm_iodevice_write(pio_dev, vcpu->arch.pio.port,
2308 vcpu->arch.pio.size,
2310 mutex_unlock(&vcpu->kvm->lock);
2313 static void pio_string_write(struct kvm_io_device *pio_dev,
2314 struct kvm_vcpu *vcpu)
2316 struct kvm_pio_request *io = &vcpu->arch.pio;
2317 void *pd = vcpu->arch.pio_data;
2320 mutex_lock(&vcpu->kvm->lock);
2321 for (i = 0; i < io->cur_count; i++) {
2322 kvm_iodevice_write(pio_dev, io->port,
2327 mutex_unlock(&vcpu->kvm->lock);
2330 static struct kvm_io_device *vcpu_find_pio_dev(struct kvm_vcpu *vcpu,
2331 gpa_t addr, int len,
2334 return kvm_io_bus_find_dev(&vcpu->kvm->pio_bus, addr, len, is_write);
2337 int kvm_emulate_pio(struct kvm_vcpu *vcpu, struct kvm_run *run, int in,
2338 int size, unsigned port)
2340 struct kvm_io_device *pio_dev;
2342 vcpu->run->exit_reason = KVM_EXIT_IO;
2343 vcpu->run->io.direction = in ? KVM_EXIT_IO_IN : KVM_EXIT_IO_OUT;
2344 vcpu->run->io.size = vcpu->arch.pio.size = size;
2345 vcpu->run->io.data_offset = KVM_PIO_PAGE_OFFSET * PAGE_SIZE;
2346 vcpu->run->io.count = vcpu->arch.pio.count = vcpu->arch.pio.cur_count = 1;
2347 vcpu->run->io.port = vcpu->arch.pio.port = port;
2348 vcpu->arch.pio.in = in;
2349 vcpu->arch.pio.string = 0;
2350 vcpu->arch.pio.down = 0;
2351 vcpu->arch.pio.guest_page_offset = 0;
2352 vcpu->arch.pio.rep = 0;
2354 if (vcpu->run->io.direction == KVM_EXIT_IO_IN)
2355 KVMTRACE_2D(IO_READ, vcpu, vcpu->run->io.port, (u32)size,
2358 KVMTRACE_2D(IO_WRITE, vcpu, vcpu->run->io.port, (u32)size,
2361 kvm_x86_ops->cache_regs(vcpu);
2362 memcpy(vcpu->arch.pio_data, &vcpu->arch.regs[VCPU_REGS_RAX], 4);
2364 kvm_x86_ops->skip_emulated_instruction(vcpu);
2366 pio_dev = vcpu_find_pio_dev(vcpu, port, size, !in);
2368 kernel_pio(pio_dev, vcpu, vcpu->arch.pio_data);
2374 EXPORT_SYMBOL_GPL(kvm_emulate_pio);
2376 int kvm_emulate_pio_string(struct kvm_vcpu *vcpu, struct kvm_run *run, int in,
2377 int size, unsigned long count, int down,
2378 gva_t address, int rep, unsigned port)
2380 unsigned now, in_page;
2384 struct kvm_io_device *pio_dev;
2386 vcpu->run->exit_reason = KVM_EXIT_IO;
2387 vcpu->run->io.direction = in ? KVM_EXIT_IO_IN : KVM_EXIT_IO_OUT;
2388 vcpu->run->io.size = vcpu->arch.pio.size = size;
2389 vcpu->run->io.data_offset = KVM_PIO_PAGE_OFFSET * PAGE_SIZE;
2390 vcpu->run->io.count = vcpu->arch.pio.count = vcpu->arch.pio.cur_count = count;
2391 vcpu->run->io.port = vcpu->arch.pio.port = port;
2392 vcpu->arch.pio.in = in;
2393 vcpu->arch.pio.string = 1;
2394 vcpu->arch.pio.down = down;
2395 vcpu->arch.pio.guest_page_offset = offset_in_page(address);
2396 vcpu->arch.pio.rep = rep;
2398 if (vcpu->run->io.direction == KVM_EXIT_IO_IN)
2399 KVMTRACE_2D(IO_READ, vcpu, vcpu->run->io.port, (u32)size,
2402 KVMTRACE_2D(IO_WRITE, vcpu, vcpu->run->io.port, (u32)size,
2406 kvm_x86_ops->skip_emulated_instruction(vcpu);
2411 in_page = PAGE_SIZE - offset_in_page(address);
2413 in_page = offset_in_page(address) + size;
2414 now = min(count, (unsigned long)in_page / size);
2417 * String I/O straddles page boundary. Pin two guest pages
2418 * so that we satisfy atomicity constraints. Do just one
2419 * transaction to avoid complexity.
2426 * String I/O in reverse. Yuck. Kill the guest, fix later.
2428 pr_unimpl(vcpu, "guest string pio down\n");
2429 kvm_inject_gp(vcpu, 0);
2432 vcpu->run->io.count = now;
2433 vcpu->arch.pio.cur_count = now;
2435 if (vcpu->arch.pio.cur_count == vcpu->arch.pio.count)
2436 kvm_x86_ops->skip_emulated_instruction(vcpu);
2438 for (i = 0; i < nr_pages; ++i) {
2439 page = gva_to_page(vcpu, address + i * PAGE_SIZE);
2440 vcpu->arch.pio.guest_pages[i] = page;
2442 kvm_inject_gp(vcpu, 0);
2443 free_pio_guest_pages(vcpu);
2448 pio_dev = vcpu_find_pio_dev(vcpu, port,
2449 vcpu->arch.pio.cur_count,
2450 !vcpu->arch.pio.in);
2451 if (!vcpu->arch.pio.in) {
2452 /* string PIO write */
2453 ret = pio_copy_data(vcpu);
2454 if (ret >= 0 && pio_dev) {
2455 pio_string_write(pio_dev, vcpu);
2457 if (vcpu->arch.pio.count == 0)
2461 pr_unimpl(vcpu, "no string pio read support yet, "
2462 "port %x size %d count %ld\n",
2467 EXPORT_SYMBOL_GPL(kvm_emulate_pio_string);
2469 int kvm_arch_init(void *opaque)
2472 struct kvm_x86_ops *ops = (struct kvm_x86_ops *)opaque;
2475 printk(KERN_ERR "kvm: already loaded the other module\n");
2480 if (!ops->cpu_has_kvm_support()) {
2481 printk(KERN_ERR "kvm: no hardware support\n");
2485 if (ops->disabled_by_bios()) {
2486 printk(KERN_ERR "kvm: disabled by bios\n");
2491 r = kvm_mmu_module_init();
2495 kvm_init_msr_list();
2498 kvm_mmu_set_nonpresent_ptes(0ull, 0ull);
2499 kvm_mmu_set_base_ptes(PT_PRESENT_MASK);
2500 kvm_mmu_set_mask_ptes(PT_USER_MASK, PT_ACCESSED_MASK,
2501 PT_DIRTY_MASK, PT64_NX_MASK, 0);
2508 void kvm_arch_exit(void)
2511 kvm_mmu_module_exit();
2514 int kvm_emulate_halt(struct kvm_vcpu *vcpu)
2516 ++vcpu->stat.halt_exits;
2517 KVMTRACE_0D(HLT, vcpu, handler);
2518 if (irqchip_in_kernel(vcpu->kvm)) {
2519 vcpu->arch.mp_state = KVM_MP_STATE_HALTED;
2520 up_read(&vcpu->kvm->slots_lock);
2521 kvm_vcpu_block(vcpu);
2522 down_read(&vcpu->kvm->slots_lock);
2523 if (vcpu->arch.mp_state != KVM_MP_STATE_RUNNABLE)
2527 vcpu->run->exit_reason = KVM_EXIT_HLT;
2531 EXPORT_SYMBOL_GPL(kvm_emulate_halt);
2533 static inline gpa_t hc_gpa(struct kvm_vcpu *vcpu, unsigned long a0,
2536 if (is_long_mode(vcpu))
2539 return a0 | ((gpa_t)a1 << 32);
2542 int kvm_emulate_hypercall(struct kvm_vcpu *vcpu)
2544 unsigned long nr, a0, a1, a2, a3, ret;
2547 kvm_x86_ops->cache_regs(vcpu);
2549 nr = vcpu->arch.regs[VCPU_REGS_RAX];
2550 a0 = vcpu->arch.regs[VCPU_REGS_RBX];
2551 a1 = vcpu->arch.regs[VCPU_REGS_RCX];
2552 a2 = vcpu->arch.regs[VCPU_REGS_RDX];
2553 a3 = vcpu->arch.regs[VCPU_REGS_RSI];
2555 KVMTRACE_1D(VMMCALL, vcpu, (u32)nr, handler);
2557 if (!is_long_mode(vcpu)) {
2566 case KVM_HC_VAPIC_POLL_IRQ:
2570 r = kvm_pv_mmu_op(vcpu, a0, hc_gpa(vcpu, a1, a2), &ret);
2576 vcpu->arch.regs[VCPU_REGS_RAX] = ret;
2577 kvm_x86_ops->decache_regs(vcpu);
2578 ++vcpu->stat.hypercalls;
2581 EXPORT_SYMBOL_GPL(kvm_emulate_hypercall);
2583 int kvm_fix_hypercall(struct kvm_vcpu *vcpu)
2585 char instruction[3];
2590 * Blow out the MMU to ensure that no other VCPU has an active mapping
2591 * to ensure that the updated hypercall appears atomically across all
2594 kvm_mmu_zap_all(vcpu->kvm);
2596 kvm_x86_ops->cache_regs(vcpu);
2597 kvm_x86_ops->patch_hypercall(vcpu, instruction);
2598 if (emulator_write_emulated(vcpu->arch.rip, instruction, 3, vcpu)
2599 != X86EMUL_CONTINUE)
2605 static u64 mk_cr_64(u64 curr_cr, u32 new_val)
2607 return (curr_cr & ~((1ULL << 32) - 1)) | new_val;
2610 void realmode_lgdt(struct kvm_vcpu *vcpu, u16 limit, unsigned long base)
2612 struct descriptor_table dt = { limit, base };
2614 kvm_x86_ops->set_gdt(vcpu, &dt);
2617 void realmode_lidt(struct kvm_vcpu *vcpu, u16 limit, unsigned long base)
2619 struct descriptor_table dt = { limit, base };
2621 kvm_x86_ops->set_idt(vcpu, &dt);
2624 void realmode_lmsw(struct kvm_vcpu *vcpu, unsigned long msw,
2625 unsigned long *rflags)
2627 kvm_lmsw(vcpu, msw);
2628 *rflags = kvm_x86_ops->get_rflags(vcpu);
2631 unsigned long realmode_get_cr(struct kvm_vcpu *vcpu, int cr)
2633 unsigned long value;
2635 kvm_x86_ops->decache_cr4_guest_bits(vcpu);
2638 value = vcpu->arch.cr0;
2641 value = vcpu->arch.cr2;
2644 value = vcpu->arch.cr3;
2647 value = vcpu->arch.cr4;
2650 value = kvm_get_cr8(vcpu);
2653 vcpu_printf(vcpu, "%s: unexpected cr %u\n", __func__, cr);
2656 KVMTRACE_3D(CR_READ, vcpu, (u32)cr, (u32)value,
2657 (u32)((u64)value >> 32), handler);
2662 void realmode_set_cr(struct kvm_vcpu *vcpu, int cr, unsigned long val,
2663 unsigned long *rflags)
2665 KVMTRACE_3D(CR_WRITE, vcpu, (u32)cr, (u32)val,
2666 (u32)((u64)val >> 32), handler);
2670 kvm_set_cr0(vcpu, mk_cr_64(vcpu->arch.cr0, val));
2671 *rflags = kvm_x86_ops->get_rflags(vcpu);
2674 vcpu->arch.cr2 = val;
2677 kvm_set_cr3(vcpu, val);
2680 kvm_set_cr4(vcpu, mk_cr_64(vcpu->arch.cr4, val));
2683 kvm_set_cr8(vcpu, val & 0xfUL);
2686 vcpu_printf(vcpu, "%s: unexpected cr %u\n", __func__, cr);
2690 static int move_to_next_stateful_cpuid_entry(struct kvm_vcpu *vcpu, int i)
2692 struct kvm_cpuid_entry2 *e = &vcpu->arch.cpuid_entries[i];
2693 int j, nent = vcpu->arch.cpuid_nent;
2695 e->flags &= ~KVM_CPUID_FLAG_STATE_READ_NEXT;
2696 /* when no next entry is found, the current entry[i] is reselected */
2697 for (j = i + 1; j == i; j = (j + 1) % nent) {
2698 struct kvm_cpuid_entry2 *ej = &vcpu->arch.cpuid_entries[j];
2699 if (ej->function == e->function) {
2700 ej->flags |= KVM_CPUID_FLAG_STATE_READ_NEXT;
2704 return 0; /* silence gcc, even though control never reaches here */
2707 /* find an entry with matching function, matching index (if needed), and that
2708 * should be read next (if it's stateful) */
2709 static int is_matching_cpuid_entry(struct kvm_cpuid_entry2 *e,
2710 u32 function, u32 index)
2712 if (e->function != function)
2714 if ((e->flags & KVM_CPUID_FLAG_SIGNIFCANT_INDEX) && e->index != index)
2716 if ((e->flags & KVM_CPUID_FLAG_STATEFUL_FUNC) &&
2717 !(e->flags & KVM_CPUID_FLAG_STATE_READ_NEXT))
2722 void kvm_emulate_cpuid(struct kvm_vcpu *vcpu)
2725 u32 function, index;
2726 struct kvm_cpuid_entry2 *e, *best;
2728 kvm_x86_ops->cache_regs(vcpu);
2729 function = vcpu->arch.regs[VCPU_REGS_RAX];
2730 index = vcpu->arch.regs[VCPU_REGS_RCX];
2731 vcpu->arch.regs[VCPU_REGS_RAX] = 0;
2732 vcpu->arch.regs[VCPU_REGS_RBX] = 0;
2733 vcpu->arch.regs[VCPU_REGS_RCX] = 0;
2734 vcpu->arch.regs[VCPU_REGS_RDX] = 0;
2736 for (i = 0; i < vcpu->arch.cpuid_nent; ++i) {
2737 e = &vcpu->arch.cpuid_entries[i];
2738 if (is_matching_cpuid_entry(e, function, index)) {
2739 if (e->flags & KVM_CPUID_FLAG_STATEFUL_FUNC)
2740 move_to_next_stateful_cpuid_entry(vcpu, i);
2745 * Both basic or both extended?
2747 if (((e->function ^ function) & 0x80000000) == 0)
2748 if (!best || e->function > best->function)
2752 vcpu->arch.regs[VCPU_REGS_RAX] = best->eax;
2753 vcpu->arch.regs[VCPU_REGS_RBX] = best->ebx;
2754 vcpu->arch.regs[VCPU_REGS_RCX] = best->ecx;
2755 vcpu->arch.regs[VCPU_REGS_RDX] = best->edx;
2757 kvm_x86_ops->decache_regs(vcpu);
2758 kvm_x86_ops->skip_emulated_instruction(vcpu);
2759 KVMTRACE_5D(CPUID, vcpu, function,
2760 (u32)vcpu->arch.regs[VCPU_REGS_RAX],
2761 (u32)vcpu->arch.regs[VCPU_REGS_RBX],
2762 (u32)vcpu->arch.regs[VCPU_REGS_RCX],
2763 (u32)vcpu->arch.regs[VCPU_REGS_RDX], handler);
2765 EXPORT_SYMBOL_GPL(kvm_emulate_cpuid);
2768 * Check if userspace requested an interrupt window, and that the
2769 * interrupt window is open.
2771 * No need to exit to userspace if we already have an interrupt queued.
2773 static int dm_request_for_irq_injection(struct kvm_vcpu *vcpu,
2774 struct kvm_run *kvm_run)
2776 return (!vcpu->arch.irq_summary &&
2777 kvm_run->request_interrupt_window &&
2778 vcpu->arch.interrupt_window_open &&
2779 (kvm_x86_ops->get_rflags(vcpu) & X86_EFLAGS_IF));
2782 static void post_kvm_run_save(struct kvm_vcpu *vcpu,
2783 struct kvm_run *kvm_run)
2785 kvm_run->if_flag = (kvm_x86_ops->get_rflags(vcpu) & X86_EFLAGS_IF) != 0;
2786 kvm_run->cr8 = kvm_get_cr8(vcpu);
2787 kvm_run->apic_base = kvm_get_apic_base(vcpu);
2788 if (irqchip_in_kernel(vcpu->kvm))
2789 kvm_run->ready_for_interrupt_injection = 1;
2791 kvm_run->ready_for_interrupt_injection =
2792 (vcpu->arch.interrupt_window_open &&
2793 vcpu->arch.irq_summary == 0);
2796 static void vapic_enter(struct kvm_vcpu *vcpu)
2798 struct kvm_lapic *apic = vcpu->arch.apic;
2801 if (!apic || !apic->vapic_addr)
2804 down_read(¤t->mm->mmap_sem);
2805 page = gfn_to_page(vcpu->kvm, apic->vapic_addr >> PAGE_SHIFT);
2806 up_read(¤t->mm->mmap_sem);
2808 vcpu->arch.apic->vapic_page = page;
2811 static void vapic_exit(struct kvm_vcpu *vcpu)
2813 struct kvm_lapic *apic = vcpu->arch.apic;
2815 if (!apic || !apic->vapic_addr)
2818 down_read(&vcpu->kvm->slots_lock);
2819 kvm_release_page_dirty(apic->vapic_page);
2820 mark_page_dirty(vcpu->kvm, apic->vapic_addr >> PAGE_SHIFT);
2821 up_read(&vcpu->kvm->slots_lock);
2824 static int __vcpu_run(struct kvm_vcpu *vcpu, struct kvm_run *kvm_run)
2828 if (unlikely(vcpu->arch.mp_state == KVM_MP_STATE_SIPI_RECEIVED)) {
2829 pr_debug("vcpu %d received sipi with vector # %x\n",
2830 vcpu->vcpu_id, vcpu->arch.sipi_vector);
2831 kvm_lapic_reset(vcpu);
2832 r = kvm_x86_ops->vcpu_reset(vcpu);
2835 vcpu->arch.mp_state = KVM_MP_STATE_RUNNABLE;
2838 down_read(&vcpu->kvm->slots_lock);
2842 if (vcpu->guest_debug.enabled)
2843 kvm_x86_ops->guest_debug_pre(vcpu);
2847 if (test_and_clear_bit(KVM_REQ_MMU_RELOAD, &vcpu->requests))
2848 kvm_mmu_unload(vcpu);
2850 r = kvm_mmu_reload(vcpu);
2854 if (vcpu->requests) {
2855 if (test_and_clear_bit(KVM_REQ_MIGRATE_TIMER, &vcpu->requests))
2856 __kvm_migrate_timers(vcpu);
2857 if (test_and_clear_bit(KVM_REQ_TLB_FLUSH, &vcpu->requests))
2858 kvm_x86_ops->tlb_flush(vcpu);
2859 if (test_and_clear_bit(KVM_REQ_REPORT_TPR_ACCESS,
2861 kvm_run->exit_reason = KVM_EXIT_TPR_ACCESS;
2865 if (test_and_clear_bit(KVM_REQ_TRIPLE_FAULT, &vcpu->requests)) {
2866 kvm_run->exit_reason = KVM_EXIT_SHUTDOWN;
2872 clear_bit(KVM_REQ_PENDING_TIMER, &vcpu->requests);
2873 kvm_inject_pending_timer_irqs(vcpu);
2877 kvm_x86_ops->prepare_guest_switch(vcpu);
2878 kvm_load_guest_fpu(vcpu);
2880 local_irq_disable();
2882 if (vcpu->requests || need_resched()) {
2889 if (signal_pending(current)) {
2893 kvm_run->exit_reason = KVM_EXIT_INTR;
2894 ++vcpu->stat.signal_exits;
2898 vcpu->guest_mode = 1;
2900 * Make sure that guest_mode assignment won't happen after
2901 * testing the pending IRQ vector bitmap.
2905 if (vcpu->arch.exception.pending)
2906 __queue_exception(vcpu);
2907 else if (irqchip_in_kernel(vcpu->kvm))
2908 kvm_x86_ops->inject_pending_irq(vcpu);
2910 kvm_x86_ops->inject_pending_vectors(vcpu, kvm_run);
2912 kvm_lapic_sync_to_vapic(vcpu);
2914 up_read(&vcpu->kvm->slots_lock);
2919 KVMTRACE_0D(VMENTRY, vcpu, entryexit);
2920 kvm_x86_ops->run(vcpu, kvm_run);
2922 vcpu->guest_mode = 0;
2928 * We must have an instruction between local_irq_enable() and
2929 * kvm_guest_exit(), so the timer interrupt isn't delayed by
2930 * the interrupt shadow. The stat.exits increment will do nicely.
2931 * But we need to prevent reordering, hence this barrier():
2939 down_read(&vcpu->kvm->slots_lock);
2942 * Profile KVM exit RIPs:
2944 if (unlikely(prof_on == KVM_PROFILING)) {
2945 kvm_x86_ops->cache_regs(vcpu);
2946 profile_hit(KVM_PROFILING, (void *)vcpu->arch.rip);
2949 if (vcpu->arch.exception.pending && kvm_x86_ops->exception_injected(vcpu))
2950 vcpu->arch.exception.pending = false;
2952 kvm_lapic_sync_from_vapic(vcpu);
2954 r = kvm_x86_ops->handle_exit(kvm_run, vcpu);
2957 if (dm_request_for_irq_injection(vcpu, kvm_run)) {
2959 kvm_run->exit_reason = KVM_EXIT_INTR;
2960 ++vcpu->stat.request_irq_exits;
2963 if (!need_resched())
2968 up_read(&vcpu->kvm->slots_lock);
2971 down_read(&vcpu->kvm->slots_lock);
2975 post_kvm_run_save(vcpu, kvm_run);
2982 int kvm_arch_vcpu_ioctl_run(struct kvm_vcpu *vcpu, struct kvm_run *kvm_run)
2989 if (vcpu->sigset_active)
2990 sigprocmask(SIG_SETMASK, &vcpu->sigset, &sigsaved);
2992 if (unlikely(vcpu->arch.mp_state == KVM_MP_STATE_UNINITIALIZED)) {
2993 kvm_vcpu_block(vcpu);
2998 /* re-sync apic's tpr */
2999 if (!irqchip_in_kernel(vcpu->kvm))
3000 kvm_set_cr8(vcpu, kvm_run->cr8);
3002 if (vcpu->arch.pio.cur_count) {
3003 r = complete_pio(vcpu);
3007 #if CONFIG_HAS_IOMEM
3008 if (vcpu->mmio_needed) {
3009 memcpy(vcpu->mmio_data, kvm_run->mmio.data, 8);
3010 vcpu->mmio_read_completed = 1;
3011 vcpu->mmio_needed = 0;
3013 down_read(&vcpu->kvm->slots_lock);
3014 r = emulate_instruction(vcpu, kvm_run,
3015 vcpu->arch.mmio_fault_cr2, 0,
3016 EMULTYPE_NO_DECODE);
3017 up_read(&vcpu->kvm->slots_lock);
3018 if (r == EMULATE_DO_MMIO) {
3020 * Read-modify-write. Back to userspace.
3027 if (kvm_run->exit_reason == KVM_EXIT_HYPERCALL) {
3028 kvm_x86_ops->cache_regs(vcpu);
3029 vcpu->arch.regs[VCPU_REGS_RAX] = kvm_run->hypercall.ret;
3030 kvm_x86_ops->decache_regs(vcpu);
3033 r = __vcpu_run(vcpu, kvm_run);
3036 if (vcpu->sigset_active)
3037 sigprocmask(SIG_SETMASK, &sigsaved, NULL);
3043 int kvm_arch_vcpu_ioctl_get_regs(struct kvm_vcpu *vcpu, struct kvm_regs *regs)
3047 kvm_x86_ops->cache_regs(vcpu);
3049 regs->rax = vcpu->arch.regs[VCPU_REGS_RAX];
3050 regs->rbx = vcpu->arch.regs[VCPU_REGS_RBX];
3051 regs->rcx = vcpu->arch.regs[VCPU_REGS_RCX];
3052 regs->rdx = vcpu->arch.regs[VCPU_REGS_RDX];
3053 regs->rsi = vcpu->arch.regs[VCPU_REGS_RSI];
3054 regs->rdi = vcpu->arch.regs[VCPU_REGS_RDI];
3055 regs->rsp = vcpu->arch.regs[VCPU_REGS_RSP];
3056 regs->rbp = vcpu->arch.regs[VCPU_REGS_RBP];
3057 #ifdef CONFIG_X86_64
3058 regs->r8 = vcpu->arch.regs[VCPU_REGS_R8];
3059 regs->r9 = vcpu->arch.regs[VCPU_REGS_R9];
3060 regs->r10 = vcpu->arch.regs[VCPU_REGS_R10];
3061 regs->r11 = vcpu->arch.regs[VCPU_REGS_R11];
3062 regs->r12 = vcpu->arch.regs[VCPU_REGS_R12];
3063 regs->r13 = vcpu->arch.regs[VCPU_REGS_R13];
3064 regs->r14 = vcpu->arch.regs[VCPU_REGS_R14];
3065 regs->r15 = vcpu->arch.regs[VCPU_REGS_R15];
3068 regs->rip = vcpu->arch.rip;
3069 regs->rflags = kvm_x86_ops->get_rflags(vcpu);
3072 * Don't leak debug flags in case they were set for guest debugging
3074 if (vcpu->guest_debug.enabled && vcpu->guest_debug.singlestep)
3075 regs->rflags &= ~(X86_EFLAGS_TF | X86_EFLAGS_RF);
3082 int kvm_arch_vcpu_ioctl_set_regs(struct kvm_vcpu *vcpu, struct kvm_regs *regs)
3086 vcpu->arch.regs[VCPU_REGS_RAX] = regs->rax;
3087 vcpu->arch.regs[VCPU_REGS_RBX] = regs->rbx;
3088 vcpu->arch.regs[VCPU_REGS_RCX] = regs->rcx;
3089 vcpu->arch.regs[VCPU_REGS_RDX] = regs->rdx;
3090 vcpu->arch.regs[VCPU_REGS_RSI] = regs->rsi;
3091 vcpu->arch.regs[VCPU_REGS_RDI] = regs->rdi;
3092 vcpu->arch.regs[VCPU_REGS_RSP] = regs->rsp;
3093 vcpu->arch.regs[VCPU_REGS_RBP] = regs->rbp;
3094 #ifdef CONFIG_X86_64
3095 vcpu->arch.regs[VCPU_REGS_R8] = regs->r8;
3096 vcpu->arch.regs[VCPU_REGS_R9] = regs->r9;
3097 vcpu->arch.regs[VCPU_REGS_R10] = regs->r10;
3098 vcpu->arch.regs[VCPU_REGS_R11] = regs->r11;
3099 vcpu->arch.regs[VCPU_REGS_R12] = regs->r12;
3100 vcpu->arch.regs[VCPU_REGS_R13] = regs->r13;
3101 vcpu->arch.regs[VCPU_REGS_R14] = regs->r14;
3102 vcpu->arch.regs[VCPU_REGS_R15] = regs->r15;
3105 vcpu->arch.rip = regs->rip;
3106 kvm_x86_ops->set_rflags(vcpu, regs->rflags);
3108 kvm_x86_ops->decache_regs(vcpu);
3110 vcpu->arch.exception.pending = false;
3117 void kvm_get_segment(struct kvm_vcpu *vcpu,
3118 struct kvm_segment *var, int seg)
3120 kvm_x86_ops->get_segment(vcpu, var, seg);
3123 void kvm_get_cs_db_l_bits(struct kvm_vcpu *vcpu, int *db, int *l)
3125 struct kvm_segment cs;
3127 kvm_get_segment(vcpu, &cs, VCPU_SREG_CS);
3131 EXPORT_SYMBOL_GPL(kvm_get_cs_db_l_bits);
3133 int kvm_arch_vcpu_ioctl_get_sregs(struct kvm_vcpu *vcpu,
3134 struct kvm_sregs *sregs)
3136 struct descriptor_table dt;
3141 kvm_get_segment(vcpu, &sregs->cs, VCPU_SREG_CS);
3142 kvm_get_segment(vcpu, &sregs->ds, VCPU_SREG_DS);
3143 kvm_get_segment(vcpu, &sregs->es, VCPU_SREG_ES);
3144 kvm_get_segment(vcpu, &sregs->fs, VCPU_SREG_FS);
3145 kvm_get_segment(vcpu, &sregs->gs, VCPU_SREG_GS);
3146 kvm_get_segment(vcpu, &sregs->ss, VCPU_SREG_SS);
3148 kvm_get_segment(vcpu, &sregs->tr, VCPU_SREG_TR);
3149 kvm_get_segment(vcpu, &sregs->ldt, VCPU_SREG_LDTR);
3151 kvm_x86_ops->get_idt(vcpu, &dt);
3152 sregs->idt.limit = dt.limit;
3153 sregs->idt.base = dt.base;
3154 kvm_x86_ops->get_gdt(vcpu, &dt);
3155 sregs->gdt.limit = dt.limit;
3156 sregs->gdt.base = dt.base;
3158 kvm_x86_ops->decache_cr4_guest_bits(vcpu);
3159 sregs->cr0 = vcpu->arch.cr0;
3160 sregs->cr2 = vcpu->arch.cr2;
3161 sregs->cr3 = vcpu->arch.cr3;
3162 sregs->cr4 = vcpu->arch.cr4;
3163 sregs->cr8 = kvm_get_cr8(vcpu);
3164 sregs->efer = vcpu->arch.shadow_efer;
3165 sregs->apic_base = kvm_get_apic_base(vcpu);
3167 if (irqchip_in_kernel(vcpu->kvm)) {
3168 memset(sregs->interrupt_bitmap, 0,
3169 sizeof sregs->interrupt_bitmap);
3170 pending_vec = kvm_x86_ops->get_irq(vcpu);
3171 if (pending_vec >= 0)
3172 set_bit(pending_vec,
3173 (unsigned long *)sregs->interrupt_bitmap);
3175 memcpy(sregs->interrupt_bitmap, vcpu->arch.irq_pending,
3176 sizeof sregs->interrupt_bitmap);
3183 int kvm_arch_vcpu_ioctl_get_mpstate(struct kvm_vcpu *vcpu,
3184 struct kvm_mp_state *mp_state)
3187 mp_state->mp_state = vcpu->arch.mp_state;
3192 int kvm_arch_vcpu_ioctl_set_mpstate(struct kvm_vcpu *vcpu,
3193 struct kvm_mp_state *mp_state)
3196 vcpu->arch.mp_state = mp_state->mp_state;
3201 static void kvm_set_segment(struct kvm_vcpu *vcpu,
3202 struct kvm_segment *var, int seg)
3204 kvm_x86_ops->set_segment(vcpu, var, seg);
3207 static void seg_desct_to_kvm_desct(struct desc_struct *seg_desc, u16 selector,
3208 struct kvm_segment *kvm_desct)
3210 kvm_desct->base = seg_desc->base0;
3211 kvm_desct->base |= seg_desc->base1 << 16;
3212 kvm_desct->base |= seg_desc->base2 << 24;
3213 kvm_desct->limit = seg_desc->limit0;
3214 kvm_desct->limit |= seg_desc->limit << 16;
3216 kvm_desct->limit <<= 12;
3217 kvm_desct->limit |= 0xfff;
3219 kvm_desct->selector = selector;
3220 kvm_desct->type = seg_desc->type;
3221 kvm_desct->present = seg_desc->p;
3222 kvm_desct->dpl = seg_desc->dpl;
3223 kvm_desct->db = seg_desc->d;
3224 kvm_desct->s = seg_desc->s;
3225 kvm_desct->l = seg_desc->l;
3226 kvm_desct->g = seg_desc->g;
3227 kvm_desct->avl = seg_desc->avl;
3229 kvm_desct->unusable = 1;
3231 kvm_desct->unusable = 0;
3232 kvm_desct->padding = 0;
3235 static void get_segment_descritptor_dtable(struct kvm_vcpu *vcpu,
3237 struct descriptor_table *dtable)
3239 if (selector & 1 << 2) {
3240 struct kvm_segment kvm_seg;
3242 kvm_get_segment(vcpu, &kvm_seg, VCPU_SREG_LDTR);
3244 if (kvm_seg.unusable)
3247 dtable->limit = kvm_seg.limit;
3248 dtable->base = kvm_seg.base;
3251 kvm_x86_ops->get_gdt(vcpu, dtable);
3254 /* allowed just for 8 bytes segments */
3255 static int load_guest_segment_descriptor(struct kvm_vcpu *vcpu, u16 selector,
3256 struct desc_struct *seg_desc)
3259 struct descriptor_table dtable;
3260 u16 index = selector >> 3;
3262 get_segment_descritptor_dtable(vcpu, selector, &dtable);
3264 if (dtable.limit < index * 8 + 7) {
3265 kvm_queue_exception_e(vcpu, GP_VECTOR, selector & 0xfffc);
3268 gpa = vcpu->arch.mmu.gva_to_gpa(vcpu, dtable.base);
3270 return kvm_read_guest(vcpu->kvm, gpa, seg_desc, 8);
3273 /* allowed just for 8 bytes segments */
3274 static int save_guest_segment_descriptor(struct kvm_vcpu *vcpu, u16 selector,
3275 struct desc_struct *seg_desc)
3278 struct descriptor_table dtable;
3279 u16 index = selector >> 3;
3281 get_segment_descritptor_dtable(vcpu, selector, &dtable);
3283 if (dtable.limit < index * 8 + 7)
3285 gpa = vcpu->arch.mmu.gva_to_gpa(vcpu, dtable.base);
3287 return kvm_write_guest(vcpu->kvm, gpa, seg_desc, 8);
3290 static u32 get_tss_base_addr(struct kvm_vcpu *vcpu,
3291 struct desc_struct *seg_desc)
3295 base_addr = seg_desc->base0;
3296 base_addr |= (seg_desc->base1 << 16);
3297 base_addr |= (seg_desc->base2 << 24);
3299 return vcpu->arch.mmu.gva_to_gpa(vcpu, base_addr);
3302 static u16 get_segment_selector(struct kvm_vcpu *vcpu, int seg)
3304 struct kvm_segment kvm_seg;
3306 kvm_get_segment(vcpu, &kvm_seg, seg);
3307 return kvm_seg.selector;
3310 static int load_segment_descriptor_to_kvm_desct(struct kvm_vcpu *vcpu,
3312 struct kvm_segment *kvm_seg)
3314 struct desc_struct seg_desc;
3316 if (load_guest_segment_descriptor(vcpu, selector, &seg_desc))
3318 seg_desct_to_kvm_desct(&seg_desc, selector, kvm_seg);
3322 int kvm_load_realmode_segment(struct kvm_vcpu *vcpu, u16 selector, int seg)
3324 struct kvm_segment segvar = {
3325 .base = selector << 4,
3327 .selector = selector,
3338 kvm_x86_ops->set_segment(vcpu, &segvar, seg);
3342 int kvm_load_segment_descriptor(struct kvm_vcpu *vcpu, u16 selector,
3343 int type_bits, int seg)
3345 struct kvm_segment kvm_seg;
3347 if (!(vcpu->arch.cr0 & X86_CR0_PE))
3348 return kvm_load_realmode_segment(vcpu, selector, seg);
3349 if (load_segment_descriptor_to_kvm_desct(vcpu, selector, &kvm_seg))
3351 kvm_seg.type |= type_bits;
3353 if (seg != VCPU_SREG_SS && seg != VCPU_SREG_CS &&
3354 seg != VCPU_SREG_LDTR)
3356 kvm_seg.unusable = 1;
3358 kvm_set_segment(vcpu, &kvm_seg, seg);
3362 static void save_state_to_tss32(struct kvm_vcpu *vcpu,
3363 struct tss_segment_32 *tss)
3365 tss->cr3 = vcpu->arch.cr3;
3366 tss->eip = vcpu->arch.rip;
3367 tss->eflags = kvm_x86_ops->get_rflags(vcpu);
3368 tss->eax = vcpu->arch.regs[VCPU_REGS_RAX];
3369 tss->ecx = vcpu->arch.regs[VCPU_REGS_RCX];
3370 tss->edx = vcpu->arch.regs[VCPU_REGS_RDX];
3371 tss->ebx = vcpu->arch.regs[VCPU_REGS_RBX];
3372 tss->esp = vcpu->arch.regs[VCPU_REGS_RSP];
3373 tss->ebp = vcpu->arch.regs[VCPU_REGS_RBP];
3374 tss->esi = vcpu->arch.regs[VCPU_REGS_RSI];
3375 tss->edi = vcpu->arch.regs[VCPU_REGS_RDI];
3377 tss->es = get_segment_selector(vcpu, VCPU_SREG_ES);
3378 tss->cs = get_segment_selector(vcpu, VCPU_SREG_CS);
3379 tss->ss = get_segment_selector(vcpu, VCPU_SREG_SS);
3380 tss->ds = get_segment_selector(vcpu, VCPU_SREG_DS);
3381 tss->fs = get_segment_selector(vcpu, VCPU_SREG_FS);
3382 tss->gs = get_segment_selector(vcpu, VCPU_SREG_GS);
3383 tss->ldt_selector = get_segment_selector(vcpu, VCPU_SREG_LDTR);
3384 tss->prev_task_link = get_segment_selector(vcpu, VCPU_SREG_TR);
3387 static int load_state_from_tss32(struct kvm_vcpu *vcpu,
3388 struct tss_segment_32 *tss)
3390 kvm_set_cr3(vcpu, tss->cr3);
3392 vcpu->arch.rip = tss->eip;
3393 kvm_x86_ops->set_rflags(vcpu, tss->eflags | 2);
3395 vcpu->arch.regs[VCPU_REGS_RAX] = tss->eax;
3396 vcpu->arch.regs[VCPU_REGS_RCX] = tss->ecx;
3397 vcpu->arch.regs[VCPU_REGS_RDX] = tss->edx;
3398 vcpu->arch.regs[VCPU_REGS_RBX] = tss->ebx;
3399 vcpu->arch.regs[VCPU_REGS_RSP] = tss->esp;
3400 vcpu->arch.regs[VCPU_REGS_RBP] = tss->ebp;
3401 vcpu->arch.regs[VCPU_REGS_RSI] = tss->esi;
3402 vcpu->arch.regs[VCPU_REGS_RDI] = tss->edi;
3404 if (kvm_load_segment_descriptor(vcpu, tss->ldt_selector, 0, VCPU_SREG_LDTR))
3407 if (kvm_load_segment_descriptor(vcpu, tss->es, 1, VCPU_SREG_ES))
3410 if (kvm_load_segment_descriptor(vcpu, tss->cs, 9, VCPU_SREG_CS))
3413 if (kvm_load_segment_descriptor(vcpu, tss->ss, 1, VCPU_SREG_SS))
3416 if (kvm_load_segment_descriptor(vcpu, tss->ds, 1, VCPU_SREG_DS))
3419 if (kvm_load_segment_descriptor(vcpu, tss->fs, 1, VCPU_SREG_FS))
3422 if (kvm_load_segment_descriptor(vcpu, tss->gs, 1, VCPU_SREG_GS))
3427 static void save_state_to_tss16(struct kvm_vcpu *vcpu,
3428 struct tss_segment_16 *tss)
3430 tss->ip = vcpu->arch.rip;
3431 tss->flag = kvm_x86_ops->get_rflags(vcpu);
3432 tss->ax = vcpu->arch.regs[VCPU_REGS_RAX];
3433 tss->cx = vcpu->arch.regs[VCPU_REGS_RCX];
3434 tss->dx = vcpu->arch.regs[VCPU_REGS_RDX];
3435 tss->bx = vcpu->arch.regs[VCPU_REGS_RBX];
3436 tss->sp = vcpu->arch.regs[VCPU_REGS_RSP];
3437 tss->bp = vcpu->arch.regs[VCPU_REGS_RBP];
3438 tss->si = vcpu->arch.regs[VCPU_REGS_RSI];
3439 tss->di = vcpu->arch.regs[VCPU_REGS_RDI];
3441 tss->es = get_segment_selector(vcpu, VCPU_SREG_ES);
3442 tss->cs = get_segment_selector(vcpu, VCPU_SREG_CS);
3443 tss->ss = get_segment_selector(vcpu, VCPU_SREG_SS);
3444 tss->ds = get_segment_selector(vcpu, VCPU_SREG_DS);
3445 tss->ldt = get_segment_selector(vcpu, VCPU_SREG_LDTR);
3446 tss->prev_task_link = get_segment_selector(vcpu, VCPU_SREG_TR);
3449 static int load_state_from_tss16(struct kvm_vcpu *vcpu,
3450 struct tss_segment_16 *tss)
3452 vcpu->arch.rip = tss->ip;
3453 kvm_x86_ops->set_rflags(vcpu, tss->flag | 2);
3454 vcpu->arch.regs[VCPU_REGS_RAX] = tss->ax;
3455 vcpu->arch.regs[VCPU_REGS_RCX] = tss->cx;
3456 vcpu->arch.regs[VCPU_REGS_RDX] = tss->dx;
3457 vcpu->arch.regs[VCPU_REGS_RBX] = tss->bx;
3458 vcpu->arch.regs[VCPU_REGS_RSP] = tss->sp;
3459 vcpu->arch.regs[VCPU_REGS_RBP] = tss->bp;
3460 vcpu->arch.regs[VCPU_REGS_RSI] = tss->si;
3461 vcpu->arch.regs[VCPU_REGS_RDI] = tss->di;
3463 if (kvm_load_segment_descriptor(vcpu, tss->ldt, 0, VCPU_SREG_LDTR))
3466 if (kvm_load_segment_descriptor(vcpu, tss->es, 1, VCPU_SREG_ES))
3469 if (kvm_load_segment_descriptor(vcpu, tss->cs, 9, VCPU_SREG_CS))
3472 if (kvm_load_segment_descriptor(vcpu, tss->ss, 1, VCPU_SREG_SS))
3475 if (kvm_load_segment_descriptor(vcpu, tss->ds, 1, VCPU_SREG_DS))
3480 static int kvm_task_switch_16(struct kvm_vcpu *vcpu, u16 tss_selector,
3482 struct desc_struct *nseg_desc)
3484 struct tss_segment_16 tss_segment_16;
3487 if (kvm_read_guest(vcpu->kvm, old_tss_base, &tss_segment_16,
3488 sizeof tss_segment_16))
3491 save_state_to_tss16(vcpu, &tss_segment_16);
3493 if (kvm_write_guest(vcpu->kvm, old_tss_base, &tss_segment_16,
3494 sizeof tss_segment_16))
3497 if (kvm_read_guest(vcpu->kvm, get_tss_base_addr(vcpu, nseg_desc),
3498 &tss_segment_16, sizeof tss_segment_16))
3501 if (load_state_from_tss16(vcpu, &tss_segment_16))
3509 static int kvm_task_switch_32(struct kvm_vcpu *vcpu, u16 tss_selector,
3511 struct desc_struct *nseg_desc)
3513 struct tss_segment_32 tss_segment_32;
3516 if (kvm_read_guest(vcpu->kvm, old_tss_base, &tss_segment_32,
3517 sizeof tss_segment_32))
3520 save_state_to_tss32(vcpu, &tss_segment_32);
3522 if (kvm_write_guest(vcpu->kvm, old_tss_base, &tss_segment_32,
3523 sizeof tss_segment_32))
3526 if (kvm_read_guest(vcpu->kvm, get_tss_base_addr(vcpu, nseg_desc),
3527 &tss_segment_32, sizeof tss_segment_32))
3530 if (load_state_from_tss32(vcpu, &tss_segment_32))
3538 int kvm_task_switch(struct kvm_vcpu *vcpu, u16 tss_selector, int reason)
3540 struct kvm_segment tr_seg;
3541 struct desc_struct cseg_desc;
3542 struct desc_struct nseg_desc;
3544 u32 old_tss_base = get_segment_base(vcpu, VCPU_SREG_TR);
3545 u16 old_tss_sel = get_segment_selector(vcpu, VCPU_SREG_TR);
3547 old_tss_base = vcpu->arch.mmu.gva_to_gpa(vcpu, old_tss_base);
3549 /* FIXME: Handle errors. Failure to read either TSS or their
3550 * descriptors should generate a pagefault.
3552 if (load_guest_segment_descriptor(vcpu, tss_selector, &nseg_desc))
3555 if (load_guest_segment_descriptor(vcpu, old_tss_sel, &cseg_desc))
3558 if (reason != TASK_SWITCH_IRET) {
3561 cpl = kvm_x86_ops->get_cpl(vcpu);
3562 if ((tss_selector & 3) > nseg_desc.dpl || cpl > nseg_desc.dpl) {
3563 kvm_queue_exception_e(vcpu, GP_VECTOR, 0);
3568 if (!nseg_desc.p || (nseg_desc.limit0 | nseg_desc.limit << 16) < 0x67) {
3569 kvm_queue_exception_e(vcpu, TS_VECTOR, tss_selector & 0xfffc);
3573 if (reason == TASK_SWITCH_IRET || reason == TASK_SWITCH_JMP) {
3574 cseg_desc.type &= ~(1 << 1); //clear the B flag
3575 save_guest_segment_descriptor(vcpu, old_tss_sel, &cseg_desc);
3578 if (reason == TASK_SWITCH_IRET) {
3579 u32 eflags = kvm_x86_ops->get_rflags(vcpu);
3580 kvm_x86_ops->set_rflags(vcpu, eflags & ~X86_EFLAGS_NT);
3583 kvm_x86_ops->skip_emulated_instruction(vcpu);
3584 kvm_x86_ops->cache_regs(vcpu);
3586 if (nseg_desc.type & 8)
3587 ret = kvm_task_switch_32(vcpu, tss_selector, old_tss_base,
3590 ret = kvm_task_switch_16(vcpu, tss_selector, old_tss_base,
3593 if (reason == TASK_SWITCH_CALL || reason == TASK_SWITCH_GATE) {
3594 u32 eflags = kvm_x86_ops->get_rflags(vcpu);
3595 kvm_x86_ops->set_rflags(vcpu, eflags | X86_EFLAGS_NT);
3598 if (reason != TASK_SWITCH_IRET) {
3599 nseg_desc.type |= (1 << 1);
3600 save_guest_segment_descriptor(vcpu, tss_selector,
3604 kvm_x86_ops->set_cr0(vcpu, vcpu->arch.cr0 | X86_CR0_TS);
3605 seg_desct_to_kvm_desct(&nseg_desc, tss_selector, &tr_seg);
3607 kvm_set_segment(vcpu, &tr_seg, VCPU_SREG_TR);
3609 kvm_x86_ops->decache_regs(vcpu);
3612 EXPORT_SYMBOL_GPL(kvm_task_switch);
3614 int kvm_arch_vcpu_ioctl_set_sregs(struct kvm_vcpu *vcpu,
3615 struct kvm_sregs *sregs)
3617 int mmu_reset_needed = 0;
3618 int i, pending_vec, max_bits;
3619 struct descriptor_table dt;
3623 dt.limit = sregs->idt.limit;
3624 dt.base = sregs->idt.base;
3625 kvm_x86_ops->set_idt(vcpu, &dt);
3626 dt.limit = sregs->gdt.limit;
3627 dt.base = sregs->gdt.base;
3628 kvm_x86_ops->set_gdt(vcpu, &dt);
3630 vcpu->arch.cr2 = sregs->cr2;
3631 mmu_reset_needed |= vcpu->arch.cr3 != sregs->cr3;
3632 vcpu->arch.cr3 = sregs->cr3;
3634 kvm_set_cr8(vcpu, sregs->cr8);
3636 mmu_reset_needed |= vcpu->arch.shadow_efer != sregs->efer;
3637 kvm_x86_ops->set_efer(vcpu, sregs->efer);
3638 kvm_set_apic_base(vcpu, sregs->apic_base);
3640 kvm_x86_ops->decache_cr4_guest_bits(vcpu);
3642 mmu_reset_needed |= vcpu->arch.cr0 != sregs->cr0;
3643 kvm_x86_ops->set_cr0(vcpu, sregs->cr0);
3644 vcpu->arch.cr0 = sregs->cr0;
3646 mmu_reset_needed |= vcpu->arch.cr4 != sregs->cr4;
3647 kvm_x86_ops->set_cr4(vcpu, sregs->cr4);
3648 if (!is_long_mode(vcpu) && is_pae(vcpu))
3649 load_pdptrs(vcpu, vcpu->arch.cr3);
3651 if (mmu_reset_needed)
3652 kvm_mmu_reset_context(vcpu);
3654 if (!irqchip_in_kernel(vcpu->kvm)) {
3655 memcpy(vcpu->arch.irq_pending, sregs->interrupt_bitmap,
3656 sizeof vcpu->arch.irq_pending);
3657 vcpu->arch.irq_summary = 0;
3658 for (i = 0; i < ARRAY_SIZE(vcpu->arch.irq_pending); ++i)
3659 if (vcpu->arch.irq_pending[i])
3660 __set_bit(i, &vcpu->arch.irq_summary);
3662 max_bits = (sizeof sregs->interrupt_bitmap) << 3;
3663 pending_vec = find_first_bit(
3664 (const unsigned long *)sregs->interrupt_bitmap,
3666 /* Only pending external irq is handled here */
3667 if (pending_vec < max_bits) {
3668 kvm_x86_ops->set_irq(vcpu, pending_vec);
3669 pr_debug("Set back pending irq %d\n",
3674 kvm_set_segment(vcpu, &sregs->cs, VCPU_SREG_CS);
3675 kvm_set_segment(vcpu, &sregs->ds, VCPU_SREG_DS);
3676 kvm_set_segment(vcpu, &sregs->es, VCPU_SREG_ES);
3677 kvm_set_segment(vcpu, &sregs->fs, VCPU_SREG_FS);
3678 kvm_set_segment(vcpu, &sregs->gs, VCPU_SREG_GS);
3679 kvm_set_segment(vcpu, &sregs->ss, VCPU_SREG_SS);
3681 kvm_set_segment(vcpu, &sregs->tr, VCPU_SREG_TR);
3682 kvm_set_segment(vcpu, &sregs->ldt, VCPU_SREG_LDTR);
3689 int kvm_arch_vcpu_ioctl_debug_guest(struct kvm_vcpu *vcpu,
3690 struct kvm_debug_guest *dbg)
3696 r = kvm_x86_ops->set_guest_debug(vcpu, dbg);
3704 * fxsave fpu state. Taken from x86_64/processor.h. To be killed when
3705 * we have asm/x86/processor.h
3716 u32 st_space[32]; /* 8*16 bytes for each FP-reg = 128 bytes */
3717 #ifdef CONFIG_X86_64
3718 u32 xmm_space[64]; /* 16*16 bytes for each XMM-reg = 256 bytes */
3720 u32 xmm_space[32]; /* 8*16 bytes for each XMM-reg = 128 bytes */
3725 * Translate a guest virtual address to a guest physical address.
3727 int kvm_arch_vcpu_ioctl_translate(struct kvm_vcpu *vcpu,
3728 struct kvm_translation *tr)
3730 unsigned long vaddr = tr->linear_address;
3734 down_read(&vcpu->kvm->slots_lock);
3735 gpa = vcpu->arch.mmu.gva_to_gpa(vcpu, vaddr);
3736 up_read(&vcpu->kvm->slots_lock);
3737 tr->physical_address = gpa;
3738 tr->valid = gpa != UNMAPPED_GVA;
3746 int kvm_arch_vcpu_ioctl_get_fpu(struct kvm_vcpu *vcpu, struct kvm_fpu *fpu)
3748 struct fxsave *fxsave = (struct fxsave *)&vcpu->arch.guest_fx_image;
3752 memcpy(fpu->fpr, fxsave->st_space, 128);
3753 fpu->fcw = fxsave->cwd;
3754 fpu->fsw = fxsave->swd;
3755 fpu->ftwx = fxsave->twd;
3756 fpu->last_opcode = fxsave->fop;
3757 fpu->last_ip = fxsave->rip;
3758 fpu->last_dp = fxsave->rdp;
3759 memcpy(fpu->xmm, fxsave->xmm_space, sizeof fxsave->xmm_space);
3766 int kvm_arch_vcpu_ioctl_set_fpu(struct kvm_vcpu *vcpu, struct kvm_fpu *fpu)
3768 struct fxsave *fxsave = (struct fxsave *)&vcpu->arch.guest_fx_image;
3772 memcpy(fxsave->st_space, fpu->fpr, 128);
3773 fxsave->cwd = fpu->fcw;
3774 fxsave->swd = fpu->fsw;
3775 fxsave->twd = fpu->ftwx;
3776 fxsave->fop = fpu->last_opcode;
3777 fxsave->rip = fpu->last_ip;
3778 fxsave->rdp = fpu->last_dp;
3779 memcpy(fxsave->xmm_space, fpu->xmm, sizeof fxsave->xmm_space);
3786 void fx_init(struct kvm_vcpu *vcpu)
3788 unsigned after_mxcsr_mask;
3791 * Touch the fpu the first time in non atomic context as if
3792 * this is the first fpu instruction the exception handler
3793 * will fire before the instruction returns and it'll have to
3794 * allocate ram with GFP_KERNEL.
3797 kvm_fx_save(&vcpu->arch.host_fx_image);
3799 /* Initialize guest FPU by resetting ours and saving into guest's */
3801 kvm_fx_save(&vcpu->arch.host_fx_image);
3803 kvm_fx_save(&vcpu->arch.guest_fx_image);
3804 kvm_fx_restore(&vcpu->arch.host_fx_image);
3807 vcpu->arch.cr0 |= X86_CR0_ET;
3808 after_mxcsr_mask = offsetof(struct i387_fxsave_struct, st_space);
3809 vcpu->arch.guest_fx_image.mxcsr = 0x1f80;
3810 memset((void *)&vcpu->arch.guest_fx_image + after_mxcsr_mask,
3811 0, sizeof(struct i387_fxsave_struct) - after_mxcsr_mask);
3813 EXPORT_SYMBOL_GPL(fx_init);
3815 void kvm_load_guest_fpu(struct kvm_vcpu *vcpu)
3817 if (!vcpu->fpu_active || vcpu->guest_fpu_loaded)
3820 vcpu->guest_fpu_loaded = 1;
3821 kvm_fx_save(&vcpu->arch.host_fx_image);
3822 kvm_fx_restore(&vcpu->arch.guest_fx_image);
3824 EXPORT_SYMBOL_GPL(kvm_load_guest_fpu);
3826 void kvm_put_guest_fpu(struct kvm_vcpu *vcpu)
3828 if (!vcpu->guest_fpu_loaded)
3831 vcpu->guest_fpu_loaded = 0;
3832 kvm_fx_save(&vcpu->arch.guest_fx_image);
3833 kvm_fx_restore(&vcpu->arch.host_fx_image);
3834 ++vcpu->stat.fpu_reload;
3836 EXPORT_SYMBOL_GPL(kvm_put_guest_fpu);
3838 void kvm_arch_vcpu_free(struct kvm_vcpu *vcpu)
3840 kvm_x86_ops->vcpu_free(vcpu);
3843 struct kvm_vcpu *kvm_arch_vcpu_create(struct kvm *kvm,
3846 return kvm_x86_ops->vcpu_create(kvm, id);
3849 int kvm_arch_vcpu_setup(struct kvm_vcpu *vcpu)
3853 /* We do fxsave: this must be aligned. */
3854 BUG_ON((unsigned long)&vcpu->arch.host_fx_image & 0xF);
3857 r = kvm_arch_vcpu_reset(vcpu);
3859 r = kvm_mmu_setup(vcpu);
3866 kvm_x86_ops->vcpu_free(vcpu);
3870 void kvm_arch_vcpu_destroy(struct kvm_vcpu *vcpu)
3873 kvm_mmu_unload(vcpu);
3876 kvm_x86_ops->vcpu_free(vcpu);
3879 int kvm_arch_vcpu_reset(struct kvm_vcpu *vcpu)
3881 return kvm_x86_ops->vcpu_reset(vcpu);
3884 void kvm_arch_hardware_enable(void *garbage)
3886 kvm_x86_ops->hardware_enable(garbage);
3889 void kvm_arch_hardware_disable(void *garbage)
3891 kvm_x86_ops->hardware_disable(garbage);
3894 int kvm_arch_hardware_setup(void)
3896 return kvm_x86_ops->hardware_setup();
3899 void kvm_arch_hardware_unsetup(void)
3901 kvm_x86_ops->hardware_unsetup();
3904 void kvm_arch_check_processor_compat(void *rtn)
3906 kvm_x86_ops->check_processor_compatibility(rtn);
3909 int kvm_arch_vcpu_init(struct kvm_vcpu *vcpu)
3915 BUG_ON(vcpu->kvm == NULL);
3918 vcpu->arch.mmu.root_hpa = INVALID_PAGE;
3919 if (!irqchip_in_kernel(kvm) || vcpu->vcpu_id == 0)
3920 vcpu->arch.mp_state = KVM_MP_STATE_RUNNABLE;
3922 vcpu->arch.mp_state = KVM_MP_STATE_UNINITIALIZED;
3924 page = alloc_page(GFP_KERNEL | __GFP_ZERO);
3929 vcpu->arch.pio_data = page_address(page);
3931 r = kvm_mmu_create(vcpu);
3933 goto fail_free_pio_data;
3935 if (irqchip_in_kernel(kvm)) {
3936 r = kvm_create_lapic(vcpu);
3938 goto fail_mmu_destroy;
3944 kvm_mmu_destroy(vcpu);
3946 free_page((unsigned long)vcpu->arch.pio_data);
3951 void kvm_arch_vcpu_uninit(struct kvm_vcpu *vcpu)
3953 kvm_free_lapic(vcpu);
3954 down_read(&vcpu->kvm->slots_lock);
3955 kvm_mmu_destroy(vcpu);
3956 up_read(&vcpu->kvm->slots_lock);
3957 free_page((unsigned long)vcpu->arch.pio_data);
3960 struct kvm *kvm_arch_create_vm(void)
3962 struct kvm *kvm = kzalloc(sizeof(struct kvm), GFP_KERNEL);
3965 return ERR_PTR(-ENOMEM);
3967 INIT_LIST_HEAD(&kvm->arch.active_mmu_pages);
3972 static void kvm_unload_vcpu_mmu(struct kvm_vcpu *vcpu)
3975 kvm_mmu_unload(vcpu);
3979 static void kvm_free_vcpus(struct kvm *kvm)
3984 * Unpin any mmu pages first.
3986 for (i = 0; i < KVM_MAX_VCPUS; ++i)
3988 kvm_unload_vcpu_mmu(kvm->vcpus[i]);
3989 for (i = 0; i < KVM_MAX_VCPUS; ++i) {
3990 if (kvm->vcpus[i]) {
3991 kvm_arch_vcpu_free(kvm->vcpus[i]);
3992 kvm->vcpus[i] = NULL;
3998 void kvm_arch_destroy_vm(struct kvm *kvm)
4001 kfree(kvm->arch.vpic);
4002 kfree(kvm->arch.vioapic);
4003 kvm_free_vcpus(kvm);
4004 kvm_free_physmem(kvm);
4005 if (kvm->arch.apic_access_page)
4006 put_page(kvm->arch.apic_access_page);
4007 if (kvm->arch.ept_identity_pagetable)
4008 put_page(kvm->arch.ept_identity_pagetable);
4012 int kvm_arch_set_memory_region(struct kvm *kvm,
4013 struct kvm_userspace_memory_region *mem,
4014 struct kvm_memory_slot old,
4017 int npages = mem->memory_size >> PAGE_SHIFT;
4018 struct kvm_memory_slot *memslot = &kvm->memslots[mem->slot];
4020 /*To keep backward compatibility with older userspace,
4021 *x86 needs to hanlde !user_alloc case.
4024 if (npages && !old.rmap) {
4025 unsigned long userspace_addr;
4027 down_write(¤t->mm->mmap_sem);
4028 userspace_addr = do_mmap(NULL, 0,
4030 PROT_READ | PROT_WRITE,
4031 MAP_PRIVATE | MAP_ANONYMOUS,
4033 up_write(¤t->mm->mmap_sem);
4035 if (IS_ERR((void *)userspace_addr))
4036 return PTR_ERR((void *)userspace_addr);
4038 /* set userspace_addr atomically for kvm_hva_to_rmapp */
4039 spin_lock(&kvm->mmu_lock);
4040 memslot->userspace_addr = userspace_addr;
4041 spin_unlock(&kvm->mmu_lock);
4043 if (!old.user_alloc && old.rmap) {
4046 down_write(¤t->mm->mmap_sem);
4047 ret = do_munmap(current->mm, old.userspace_addr,
4048 old.npages * PAGE_SIZE);
4049 up_write(¤t->mm->mmap_sem);
4052 "kvm_vm_ioctl_set_memory_region: "
4053 "failed to munmap memory\n");
4058 if (!kvm->arch.n_requested_mmu_pages) {
4059 unsigned int nr_mmu_pages = kvm_mmu_calculate_mmu_pages(kvm);
4060 kvm_mmu_change_mmu_pages(kvm, nr_mmu_pages);
4063 kvm_mmu_slot_remove_write_access(kvm, mem->slot);
4064 kvm_flush_remote_tlbs(kvm);
4069 void kvm_arch_flush_shadow(struct kvm *kvm)
4071 kvm_mmu_zap_all(kvm);
4074 int kvm_arch_vcpu_runnable(struct kvm_vcpu *vcpu)
4076 return vcpu->arch.mp_state == KVM_MP_STATE_RUNNABLE
4077 || vcpu->arch.mp_state == KVM_MP_STATE_SIPI_RECEIVED;
4080 static void vcpu_kick_intr(void *info)
4083 struct kvm_vcpu *vcpu = (struct kvm_vcpu *)info;
4084 printk(KERN_DEBUG "vcpu_kick_intr %p \n", vcpu);
4088 void kvm_vcpu_kick(struct kvm_vcpu *vcpu)
4090 int ipi_pcpu = vcpu->cpu;
4091 int cpu = get_cpu();
4093 if (waitqueue_active(&vcpu->wq)) {
4094 wake_up_interruptible(&vcpu->wq);
4095 ++vcpu->stat.halt_wakeup;
4098 * We may be called synchronously with irqs disabled in guest mode,
4099 * So need not to call smp_call_function_single() in that case.
4101 if (vcpu->guest_mode && vcpu->cpu != cpu)
4102 smp_call_function_single(ipi_pcpu, vcpu_kick_intr, vcpu, 0);