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 { "halt_exits", VCPU_STAT(halt_exits) },
76 { "halt_wakeup", VCPU_STAT(halt_wakeup) },
77 { "hypercalls", VCPU_STAT(hypercalls) },
78 { "request_irq", VCPU_STAT(request_irq_exits) },
79 { "irq_exits", VCPU_STAT(irq_exits) },
80 { "host_state_reload", VCPU_STAT(host_state_reload) },
81 { "efer_reload", VCPU_STAT(efer_reload) },
82 { "fpu_reload", VCPU_STAT(fpu_reload) },
83 { "insn_emulation", VCPU_STAT(insn_emulation) },
84 { "insn_emulation_fail", VCPU_STAT(insn_emulation_fail) },
85 { "mmu_shadow_zapped", VM_STAT(mmu_shadow_zapped) },
86 { "mmu_pte_write", VM_STAT(mmu_pte_write) },
87 { "mmu_pte_updated", VM_STAT(mmu_pte_updated) },
88 { "mmu_pde_zapped", VM_STAT(mmu_pde_zapped) },
89 { "mmu_flooded", VM_STAT(mmu_flooded) },
90 { "mmu_recycled", VM_STAT(mmu_recycled) },
91 { "mmu_cache_miss", VM_STAT(mmu_cache_miss) },
92 { "remote_tlb_flush", VM_STAT(remote_tlb_flush) },
93 { "largepages", VM_STAT(lpages) },
98 unsigned long segment_base(u16 selector)
100 struct descriptor_table gdt;
101 struct desc_struct *d;
102 unsigned long table_base;
108 asm("sgdt %0" : "=m"(gdt));
109 table_base = gdt.base;
111 if (selector & 4) { /* from ldt */
114 asm("sldt %0" : "=g"(ldt_selector));
115 table_base = segment_base(ldt_selector);
117 d = (struct desc_struct *)(table_base + (selector & ~7));
118 v = d->base0 | ((unsigned long)d->base1 << 16) |
119 ((unsigned long)d->base2 << 24);
121 if (d->s == 0 && (d->type == 2 || d->type == 9 || d->type == 11))
122 v |= ((unsigned long)((struct ldttss_desc64 *)d)->base3) << 32;
126 EXPORT_SYMBOL_GPL(segment_base);
128 u64 kvm_get_apic_base(struct kvm_vcpu *vcpu)
130 if (irqchip_in_kernel(vcpu->kvm))
131 return vcpu->arch.apic_base;
133 return vcpu->arch.apic_base;
135 EXPORT_SYMBOL_GPL(kvm_get_apic_base);
137 void kvm_set_apic_base(struct kvm_vcpu *vcpu, u64 data)
139 /* TODO: reserve bits check */
140 if (irqchip_in_kernel(vcpu->kvm))
141 kvm_lapic_set_base(vcpu, data);
143 vcpu->arch.apic_base = data;
145 EXPORT_SYMBOL_GPL(kvm_set_apic_base);
147 void kvm_queue_exception(struct kvm_vcpu *vcpu, unsigned nr)
149 WARN_ON(vcpu->arch.exception.pending);
150 vcpu->arch.exception.pending = true;
151 vcpu->arch.exception.has_error_code = false;
152 vcpu->arch.exception.nr = nr;
154 EXPORT_SYMBOL_GPL(kvm_queue_exception);
156 void kvm_inject_page_fault(struct kvm_vcpu *vcpu, unsigned long addr,
159 ++vcpu->stat.pf_guest;
160 if (vcpu->arch.exception.pending) {
161 if (vcpu->arch.exception.nr == PF_VECTOR) {
162 printk(KERN_DEBUG "kvm: inject_page_fault:"
163 " double fault 0x%lx\n", addr);
164 vcpu->arch.exception.nr = DF_VECTOR;
165 vcpu->arch.exception.error_code = 0;
166 } else if (vcpu->arch.exception.nr == DF_VECTOR) {
167 /* triple fault -> shutdown */
168 set_bit(KVM_REQ_TRIPLE_FAULT, &vcpu->requests);
172 vcpu->arch.cr2 = addr;
173 kvm_queue_exception_e(vcpu, PF_VECTOR, error_code);
176 void kvm_inject_nmi(struct kvm_vcpu *vcpu)
178 vcpu->arch.nmi_pending = 1;
180 EXPORT_SYMBOL_GPL(kvm_inject_nmi);
182 void kvm_queue_exception_e(struct kvm_vcpu *vcpu, unsigned nr, u32 error_code)
184 WARN_ON(vcpu->arch.exception.pending);
185 vcpu->arch.exception.pending = true;
186 vcpu->arch.exception.has_error_code = true;
187 vcpu->arch.exception.nr = nr;
188 vcpu->arch.exception.error_code = error_code;
190 EXPORT_SYMBOL_GPL(kvm_queue_exception_e);
192 static void __queue_exception(struct kvm_vcpu *vcpu)
194 kvm_x86_ops->queue_exception(vcpu, vcpu->arch.exception.nr,
195 vcpu->arch.exception.has_error_code,
196 vcpu->arch.exception.error_code);
200 * Load the pae pdptrs. Return true is they are all valid.
202 int load_pdptrs(struct kvm_vcpu *vcpu, unsigned long cr3)
204 gfn_t pdpt_gfn = cr3 >> PAGE_SHIFT;
205 unsigned offset = ((cr3 & (PAGE_SIZE-1)) >> 5) << 2;
208 u64 pdpte[ARRAY_SIZE(vcpu->arch.pdptrs)];
210 ret = kvm_read_guest_page(vcpu->kvm, pdpt_gfn, pdpte,
211 offset * sizeof(u64), sizeof(pdpte));
216 for (i = 0; i < ARRAY_SIZE(pdpte); ++i) {
217 if ((pdpte[i] & 1) && (pdpte[i] & 0xfffffff0000001e6ull)) {
224 memcpy(vcpu->arch.pdptrs, pdpte, sizeof(vcpu->arch.pdptrs));
229 EXPORT_SYMBOL_GPL(load_pdptrs);
231 static bool pdptrs_changed(struct kvm_vcpu *vcpu)
233 u64 pdpte[ARRAY_SIZE(vcpu->arch.pdptrs)];
237 if (is_long_mode(vcpu) || !is_pae(vcpu))
240 r = kvm_read_guest(vcpu->kvm, vcpu->arch.cr3 & ~31u, pdpte, sizeof(pdpte));
243 changed = memcmp(pdpte, vcpu->arch.pdptrs, sizeof(pdpte)) != 0;
249 void kvm_set_cr0(struct kvm_vcpu *vcpu, unsigned long cr0)
251 if (cr0 & CR0_RESERVED_BITS) {
252 printk(KERN_DEBUG "set_cr0: 0x%lx #GP, reserved bits 0x%lx\n",
253 cr0, vcpu->arch.cr0);
254 kvm_inject_gp(vcpu, 0);
258 if ((cr0 & X86_CR0_NW) && !(cr0 & X86_CR0_CD)) {
259 printk(KERN_DEBUG "set_cr0: #GP, CD == 0 && NW == 1\n");
260 kvm_inject_gp(vcpu, 0);
264 if ((cr0 & X86_CR0_PG) && !(cr0 & X86_CR0_PE)) {
265 printk(KERN_DEBUG "set_cr0: #GP, set PG flag "
266 "and a clear PE flag\n");
267 kvm_inject_gp(vcpu, 0);
271 if (!is_paging(vcpu) && (cr0 & X86_CR0_PG)) {
273 if ((vcpu->arch.shadow_efer & EFER_LME)) {
277 printk(KERN_DEBUG "set_cr0: #GP, start paging "
278 "in long mode while PAE is disabled\n");
279 kvm_inject_gp(vcpu, 0);
282 kvm_x86_ops->get_cs_db_l_bits(vcpu, &cs_db, &cs_l);
284 printk(KERN_DEBUG "set_cr0: #GP, start paging "
285 "in long mode while CS.L == 1\n");
286 kvm_inject_gp(vcpu, 0);
292 if (is_pae(vcpu) && !load_pdptrs(vcpu, vcpu->arch.cr3)) {
293 printk(KERN_DEBUG "set_cr0: #GP, pdptrs "
295 kvm_inject_gp(vcpu, 0);
301 kvm_x86_ops->set_cr0(vcpu, cr0);
302 vcpu->arch.cr0 = cr0;
304 kvm_mmu_reset_context(vcpu);
307 EXPORT_SYMBOL_GPL(kvm_set_cr0);
309 void kvm_lmsw(struct kvm_vcpu *vcpu, unsigned long msw)
311 kvm_set_cr0(vcpu, (vcpu->arch.cr0 & ~0x0ful) | (msw & 0x0f));
312 KVMTRACE_1D(LMSW, vcpu,
313 (u32)((vcpu->arch.cr0 & ~0x0ful) | (msw & 0x0f)),
316 EXPORT_SYMBOL_GPL(kvm_lmsw);
318 void kvm_set_cr4(struct kvm_vcpu *vcpu, unsigned long cr4)
320 if (cr4 & CR4_RESERVED_BITS) {
321 printk(KERN_DEBUG "set_cr4: #GP, reserved bits\n");
322 kvm_inject_gp(vcpu, 0);
326 if (is_long_mode(vcpu)) {
327 if (!(cr4 & X86_CR4_PAE)) {
328 printk(KERN_DEBUG "set_cr4: #GP, clearing PAE while "
330 kvm_inject_gp(vcpu, 0);
333 } else if (is_paging(vcpu) && !is_pae(vcpu) && (cr4 & X86_CR4_PAE)
334 && !load_pdptrs(vcpu, vcpu->arch.cr3)) {
335 printk(KERN_DEBUG "set_cr4: #GP, pdptrs reserved bits\n");
336 kvm_inject_gp(vcpu, 0);
340 if (cr4 & X86_CR4_VMXE) {
341 printk(KERN_DEBUG "set_cr4: #GP, setting VMXE\n");
342 kvm_inject_gp(vcpu, 0);
345 kvm_x86_ops->set_cr4(vcpu, cr4);
346 vcpu->arch.cr4 = cr4;
347 kvm_mmu_reset_context(vcpu);
349 EXPORT_SYMBOL_GPL(kvm_set_cr4);
351 void kvm_set_cr3(struct kvm_vcpu *vcpu, unsigned long cr3)
353 if (cr3 == vcpu->arch.cr3 && !pdptrs_changed(vcpu)) {
354 kvm_mmu_flush_tlb(vcpu);
358 if (is_long_mode(vcpu)) {
359 if (cr3 & CR3_L_MODE_RESERVED_BITS) {
360 printk(KERN_DEBUG "set_cr3: #GP, reserved bits\n");
361 kvm_inject_gp(vcpu, 0);
366 if (cr3 & CR3_PAE_RESERVED_BITS) {
368 "set_cr3: #GP, reserved bits\n");
369 kvm_inject_gp(vcpu, 0);
372 if (is_paging(vcpu) && !load_pdptrs(vcpu, cr3)) {
373 printk(KERN_DEBUG "set_cr3: #GP, pdptrs "
375 kvm_inject_gp(vcpu, 0);
380 * We don't check reserved bits in nonpae mode, because
381 * this isn't enforced, and VMware depends on this.
386 * Does the new cr3 value map to physical memory? (Note, we
387 * catch an invalid cr3 even in real-mode, because it would
388 * cause trouble later on when we turn on paging anyway.)
390 * A real CPU would silently accept an invalid cr3 and would
391 * attempt to use it - with largely undefined (and often hard
392 * to debug) behavior on the guest side.
394 if (unlikely(!gfn_to_memslot(vcpu->kvm, cr3 >> PAGE_SHIFT)))
395 kvm_inject_gp(vcpu, 0);
397 vcpu->arch.cr3 = cr3;
398 vcpu->arch.mmu.new_cr3(vcpu);
401 EXPORT_SYMBOL_GPL(kvm_set_cr3);
403 void kvm_set_cr8(struct kvm_vcpu *vcpu, unsigned long cr8)
405 if (cr8 & CR8_RESERVED_BITS) {
406 printk(KERN_DEBUG "set_cr8: #GP, reserved bits 0x%lx\n", cr8);
407 kvm_inject_gp(vcpu, 0);
410 if (irqchip_in_kernel(vcpu->kvm))
411 kvm_lapic_set_tpr(vcpu, cr8);
413 vcpu->arch.cr8 = cr8;
415 EXPORT_SYMBOL_GPL(kvm_set_cr8);
417 unsigned long kvm_get_cr8(struct kvm_vcpu *vcpu)
419 if (irqchip_in_kernel(vcpu->kvm))
420 return kvm_lapic_get_cr8(vcpu);
422 return vcpu->arch.cr8;
424 EXPORT_SYMBOL_GPL(kvm_get_cr8);
427 * List of msr numbers which we expose to userspace through KVM_GET_MSRS
428 * and KVM_SET_MSRS, and KVM_GET_MSR_INDEX_LIST.
430 * This list is modified at module load time to reflect the
431 * capabilities of the host cpu.
433 static u32 msrs_to_save[] = {
434 MSR_IA32_SYSENTER_CS, MSR_IA32_SYSENTER_ESP, MSR_IA32_SYSENTER_EIP,
437 MSR_CSTAR, MSR_KERNEL_GS_BASE, MSR_SYSCALL_MASK, MSR_LSTAR,
439 MSR_IA32_TIME_STAMP_COUNTER, MSR_KVM_SYSTEM_TIME, MSR_KVM_WALL_CLOCK,
440 MSR_IA32_PERF_STATUS,
443 static unsigned num_msrs_to_save;
445 static u32 emulated_msrs[] = {
446 MSR_IA32_MISC_ENABLE,
449 static void set_efer(struct kvm_vcpu *vcpu, u64 efer)
451 if (efer & efer_reserved_bits) {
452 printk(KERN_DEBUG "set_efer: 0x%llx #GP, reserved bits\n",
454 kvm_inject_gp(vcpu, 0);
459 && (vcpu->arch.shadow_efer & EFER_LME) != (efer & EFER_LME)) {
460 printk(KERN_DEBUG "set_efer: #GP, change LME while paging\n");
461 kvm_inject_gp(vcpu, 0);
465 kvm_x86_ops->set_efer(vcpu, efer);
468 efer |= vcpu->arch.shadow_efer & EFER_LMA;
470 vcpu->arch.shadow_efer = efer;
473 void kvm_enable_efer_bits(u64 mask)
475 efer_reserved_bits &= ~mask;
477 EXPORT_SYMBOL_GPL(kvm_enable_efer_bits);
481 * Writes msr value into into the appropriate "register".
482 * Returns 0 on success, non-0 otherwise.
483 * Assumes vcpu_load() was already called.
485 int kvm_set_msr(struct kvm_vcpu *vcpu, u32 msr_index, u64 data)
487 return kvm_x86_ops->set_msr(vcpu, msr_index, data);
491 * Adapt set_msr() to msr_io()'s calling convention
493 static int do_set_msr(struct kvm_vcpu *vcpu, unsigned index, u64 *data)
495 return kvm_set_msr(vcpu, index, *data);
498 static void kvm_write_wall_clock(struct kvm *kvm, gpa_t wall_clock)
501 struct pvclock_wall_clock wc;
502 struct timespec now, sys, boot;
509 kvm_write_guest(kvm, wall_clock, &version, sizeof(version));
512 * The guest calculates current wall clock time by adding
513 * system time (updated by kvm_write_guest_time below) to the
514 * wall clock specified here. guest system time equals host
515 * system time for us, thus we must fill in host boot time here.
517 now = current_kernel_time();
519 boot = ns_to_timespec(timespec_to_ns(&now) - timespec_to_ns(&sys));
521 wc.sec = boot.tv_sec;
522 wc.nsec = boot.tv_nsec;
523 wc.version = version;
525 kvm_write_guest(kvm, wall_clock, &wc, sizeof(wc));
528 kvm_write_guest(kvm, wall_clock, &version, sizeof(version));
531 static uint32_t div_frac(uint32_t dividend, uint32_t divisor)
533 uint32_t quotient, remainder;
535 /* Don't try to replace with do_div(), this one calculates
536 * "(dividend << 32) / divisor" */
538 : "=a" (quotient), "=d" (remainder)
539 : "0" (0), "1" (dividend), "r" (divisor) );
543 static void kvm_set_time_scale(uint32_t tsc_khz, struct pvclock_vcpu_time_info *hv_clock)
545 uint64_t nsecs = 1000000000LL;
550 tps64 = tsc_khz * 1000LL;
551 while (tps64 > nsecs*2) {
556 tps32 = (uint32_t)tps64;
557 while (tps32 <= (uint32_t)nsecs) {
562 hv_clock->tsc_shift = shift;
563 hv_clock->tsc_to_system_mul = div_frac(nsecs, tps32);
565 pr_debug("%s: tsc_khz %u, tsc_shift %d, tsc_mul %u\n",
566 __FUNCTION__, tsc_khz, hv_clock->tsc_shift,
567 hv_clock->tsc_to_system_mul);
570 static void kvm_write_guest_time(struct kvm_vcpu *v)
574 struct kvm_vcpu_arch *vcpu = &v->arch;
577 if ((!vcpu->time_page))
580 if (unlikely(vcpu->hv_clock_tsc_khz != tsc_khz)) {
581 kvm_set_time_scale(tsc_khz, &vcpu->hv_clock);
582 vcpu->hv_clock_tsc_khz = tsc_khz;
585 /* Keep irq disabled to prevent changes to the clock */
586 local_irq_save(flags);
587 kvm_get_msr(v, MSR_IA32_TIME_STAMP_COUNTER,
588 &vcpu->hv_clock.tsc_timestamp);
590 local_irq_restore(flags);
592 /* With all the info we got, fill in the values */
594 vcpu->hv_clock.system_time = ts.tv_nsec +
595 (NSEC_PER_SEC * (u64)ts.tv_sec);
597 * The interface expects us to write an even number signaling that the
598 * update is finished. Since the guest won't see the intermediate
599 * state, we just increase by 2 at the end.
601 vcpu->hv_clock.version += 2;
603 shared_kaddr = kmap_atomic(vcpu->time_page, KM_USER0);
605 memcpy(shared_kaddr + vcpu->time_offset, &vcpu->hv_clock,
606 sizeof(vcpu->hv_clock));
608 kunmap_atomic(shared_kaddr, KM_USER0);
610 mark_page_dirty(v->kvm, vcpu->time >> PAGE_SHIFT);
614 int kvm_set_msr_common(struct kvm_vcpu *vcpu, u32 msr, u64 data)
618 set_efer(vcpu, data);
620 case MSR_IA32_MC0_STATUS:
621 pr_unimpl(vcpu, "%s: MSR_IA32_MC0_STATUS 0x%llx, nop\n",
624 case MSR_IA32_MCG_STATUS:
625 pr_unimpl(vcpu, "%s: MSR_IA32_MCG_STATUS 0x%llx, nop\n",
628 case MSR_IA32_MCG_CTL:
629 pr_unimpl(vcpu, "%s: MSR_IA32_MCG_CTL 0x%llx, nop\n",
632 case MSR_IA32_UCODE_REV:
633 case MSR_IA32_UCODE_WRITE:
634 case 0x200 ... 0x2ff: /* MTRRs */
636 case MSR_IA32_APICBASE:
637 kvm_set_apic_base(vcpu, data);
639 case MSR_IA32_MISC_ENABLE:
640 vcpu->arch.ia32_misc_enable_msr = data;
642 case MSR_KVM_WALL_CLOCK:
643 vcpu->kvm->arch.wall_clock = data;
644 kvm_write_wall_clock(vcpu->kvm, data);
646 case MSR_KVM_SYSTEM_TIME: {
647 if (vcpu->arch.time_page) {
648 kvm_release_page_dirty(vcpu->arch.time_page);
649 vcpu->arch.time_page = NULL;
652 vcpu->arch.time = data;
654 /* we verify if the enable bit is set... */
658 /* ...but clean it before doing the actual write */
659 vcpu->arch.time_offset = data & ~(PAGE_MASK | 1);
661 down_read(¤t->mm->mmap_sem);
662 vcpu->arch.time_page =
663 gfn_to_page(vcpu->kvm, data >> PAGE_SHIFT);
664 up_read(¤t->mm->mmap_sem);
666 if (is_error_page(vcpu->arch.time_page)) {
667 kvm_release_page_clean(vcpu->arch.time_page);
668 vcpu->arch.time_page = NULL;
671 kvm_write_guest_time(vcpu);
675 pr_unimpl(vcpu, "unhandled wrmsr: 0x%x data %llx\n", msr, data);
680 EXPORT_SYMBOL_GPL(kvm_set_msr_common);
684 * Reads an msr value (of 'msr_index') into 'pdata'.
685 * Returns 0 on success, non-0 otherwise.
686 * Assumes vcpu_load() was already called.
688 int kvm_get_msr(struct kvm_vcpu *vcpu, u32 msr_index, u64 *pdata)
690 return kvm_x86_ops->get_msr(vcpu, msr_index, pdata);
693 int kvm_get_msr_common(struct kvm_vcpu *vcpu, u32 msr, u64 *pdata)
698 case 0xc0010010: /* SYSCFG */
699 case 0xc0010015: /* HWCR */
700 case MSR_IA32_PLATFORM_ID:
701 case MSR_IA32_P5_MC_ADDR:
702 case MSR_IA32_P5_MC_TYPE:
703 case MSR_IA32_MC0_CTL:
704 case MSR_IA32_MCG_STATUS:
705 case MSR_IA32_MCG_CAP:
706 case MSR_IA32_MCG_CTL:
707 case MSR_IA32_MC0_MISC:
708 case MSR_IA32_MC0_MISC+4:
709 case MSR_IA32_MC0_MISC+8:
710 case MSR_IA32_MC0_MISC+12:
711 case MSR_IA32_MC0_MISC+16:
712 case MSR_IA32_UCODE_REV:
713 case MSR_IA32_EBL_CR_POWERON:
716 case 0x200 ... 0x2ff:
719 case 0xcd: /* fsb frequency */
722 case MSR_IA32_APICBASE:
723 data = kvm_get_apic_base(vcpu);
725 case MSR_IA32_MISC_ENABLE:
726 data = vcpu->arch.ia32_misc_enable_msr;
728 case MSR_IA32_PERF_STATUS:
729 /* TSC increment by tick */
732 data |= (((uint64_t)4ULL) << 40);
735 data = vcpu->arch.shadow_efer;
737 case MSR_KVM_WALL_CLOCK:
738 data = vcpu->kvm->arch.wall_clock;
740 case MSR_KVM_SYSTEM_TIME:
741 data = vcpu->arch.time;
744 pr_unimpl(vcpu, "unhandled rdmsr: 0x%x\n", msr);
750 EXPORT_SYMBOL_GPL(kvm_get_msr_common);
753 * Read or write a bunch of msrs. All parameters are kernel addresses.
755 * @return number of msrs set successfully.
757 static int __msr_io(struct kvm_vcpu *vcpu, struct kvm_msrs *msrs,
758 struct kvm_msr_entry *entries,
759 int (*do_msr)(struct kvm_vcpu *vcpu,
760 unsigned index, u64 *data))
766 down_read(&vcpu->kvm->slots_lock);
767 for (i = 0; i < msrs->nmsrs; ++i)
768 if (do_msr(vcpu, entries[i].index, &entries[i].data))
770 up_read(&vcpu->kvm->slots_lock);
778 * Read or write a bunch of msrs. Parameters are user addresses.
780 * @return number of msrs set successfully.
782 static int msr_io(struct kvm_vcpu *vcpu, struct kvm_msrs __user *user_msrs,
783 int (*do_msr)(struct kvm_vcpu *vcpu,
784 unsigned index, u64 *data),
787 struct kvm_msrs msrs;
788 struct kvm_msr_entry *entries;
793 if (copy_from_user(&msrs, user_msrs, sizeof msrs))
797 if (msrs.nmsrs >= MAX_IO_MSRS)
801 size = sizeof(struct kvm_msr_entry) * msrs.nmsrs;
802 entries = vmalloc(size);
807 if (copy_from_user(entries, user_msrs->entries, size))
810 r = n = __msr_io(vcpu, &msrs, entries, do_msr);
815 if (writeback && copy_to_user(user_msrs->entries, entries, size))
826 int kvm_dev_ioctl_check_extension(long ext)
831 case KVM_CAP_IRQCHIP:
833 case KVM_CAP_MMU_SHADOW_CACHE_CONTROL:
834 case KVM_CAP_USER_MEMORY:
835 case KVM_CAP_SET_TSS_ADDR:
836 case KVM_CAP_EXT_CPUID:
837 case KVM_CAP_CLOCKSOURCE:
839 case KVM_CAP_NOP_IO_DELAY:
840 case KVM_CAP_MP_STATE:
844 r = !kvm_x86_ops->cpu_has_accelerated_tpr();
846 case KVM_CAP_NR_VCPUS:
849 case KVM_CAP_NR_MEMSLOTS:
850 r = KVM_MEMORY_SLOTS;
863 long kvm_arch_dev_ioctl(struct file *filp,
864 unsigned int ioctl, unsigned long arg)
866 void __user *argp = (void __user *)arg;
870 case KVM_GET_MSR_INDEX_LIST: {
871 struct kvm_msr_list __user *user_msr_list = argp;
872 struct kvm_msr_list msr_list;
876 if (copy_from_user(&msr_list, user_msr_list, sizeof msr_list))
879 msr_list.nmsrs = num_msrs_to_save + ARRAY_SIZE(emulated_msrs);
880 if (copy_to_user(user_msr_list, &msr_list, sizeof msr_list))
883 if (n < num_msrs_to_save)
886 if (copy_to_user(user_msr_list->indices, &msrs_to_save,
887 num_msrs_to_save * sizeof(u32)))
889 if (copy_to_user(user_msr_list->indices
890 + num_msrs_to_save * sizeof(u32),
892 ARRAY_SIZE(emulated_msrs) * sizeof(u32)))
897 case KVM_GET_SUPPORTED_CPUID: {
898 struct kvm_cpuid2 __user *cpuid_arg = argp;
899 struct kvm_cpuid2 cpuid;
902 if (copy_from_user(&cpuid, cpuid_arg, sizeof cpuid))
904 r = kvm_dev_ioctl_get_supported_cpuid(&cpuid,
910 if (copy_to_user(cpuid_arg, &cpuid, sizeof cpuid))
922 void kvm_arch_vcpu_load(struct kvm_vcpu *vcpu, int cpu)
924 kvm_x86_ops->vcpu_load(vcpu, cpu);
925 kvm_write_guest_time(vcpu);
928 void kvm_arch_vcpu_put(struct kvm_vcpu *vcpu)
930 kvm_x86_ops->vcpu_put(vcpu);
931 kvm_put_guest_fpu(vcpu);
934 static int is_efer_nx(void)
938 rdmsrl(MSR_EFER, efer);
939 return efer & EFER_NX;
942 static void cpuid_fix_nx_cap(struct kvm_vcpu *vcpu)
945 struct kvm_cpuid_entry2 *e, *entry;
948 for (i = 0; i < vcpu->arch.cpuid_nent; ++i) {
949 e = &vcpu->arch.cpuid_entries[i];
950 if (e->function == 0x80000001) {
955 if (entry && (entry->edx & (1 << 20)) && !is_efer_nx()) {
956 entry->edx &= ~(1 << 20);
957 printk(KERN_INFO "kvm: guest NX capability removed\n");
961 /* when an old userspace process fills a new kernel module */
962 static int kvm_vcpu_ioctl_set_cpuid(struct kvm_vcpu *vcpu,
963 struct kvm_cpuid *cpuid,
964 struct kvm_cpuid_entry __user *entries)
967 struct kvm_cpuid_entry *cpuid_entries;
970 if (cpuid->nent > KVM_MAX_CPUID_ENTRIES)
973 cpuid_entries = vmalloc(sizeof(struct kvm_cpuid_entry) * cpuid->nent);
977 if (copy_from_user(cpuid_entries, entries,
978 cpuid->nent * sizeof(struct kvm_cpuid_entry)))
980 for (i = 0; i < cpuid->nent; i++) {
981 vcpu->arch.cpuid_entries[i].function = cpuid_entries[i].function;
982 vcpu->arch.cpuid_entries[i].eax = cpuid_entries[i].eax;
983 vcpu->arch.cpuid_entries[i].ebx = cpuid_entries[i].ebx;
984 vcpu->arch.cpuid_entries[i].ecx = cpuid_entries[i].ecx;
985 vcpu->arch.cpuid_entries[i].edx = cpuid_entries[i].edx;
986 vcpu->arch.cpuid_entries[i].index = 0;
987 vcpu->arch.cpuid_entries[i].flags = 0;
988 vcpu->arch.cpuid_entries[i].padding[0] = 0;
989 vcpu->arch.cpuid_entries[i].padding[1] = 0;
990 vcpu->arch.cpuid_entries[i].padding[2] = 0;
992 vcpu->arch.cpuid_nent = cpuid->nent;
993 cpuid_fix_nx_cap(vcpu);
997 vfree(cpuid_entries);
1002 static int kvm_vcpu_ioctl_set_cpuid2(struct kvm_vcpu *vcpu,
1003 struct kvm_cpuid2 *cpuid,
1004 struct kvm_cpuid_entry2 __user *entries)
1009 if (cpuid->nent > KVM_MAX_CPUID_ENTRIES)
1012 if (copy_from_user(&vcpu->arch.cpuid_entries, entries,
1013 cpuid->nent * sizeof(struct kvm_cpuid_entry2)))
1015 vcpu->arch.cpuid_nent = cpuid->nent;
1022 static int kvm_vcpu_ioctl_get_cpuid2(struct kvm_vcpu *vcpu,
1023 struct kvm_cpuid2 *cpuid,
1024 struct kvm_cpuid_entry2 __user *entries)
1029 if (cpuid->nent < vcpu->arch.cpuid_nent)
1032 if (copy_to_user(entries, &vcpu->arch.cpuid_entries,
1033 vcpu->arch.cpuid_nent * sizeof(struct kvm_cpuid_entry2)))
1038 cpuid->nent = vcpu->arch.cpuid_nent;
1042 static inline u32 bit(int bitno)
1044 return 1 << (bitno & 31);
1047 static void do_cpuid_1_ent(struct kvm_cpuid_entry2 *entry, u32 function,
1050 entry->function = function;
1051 entry->index = index;
1052 cpuid_count(entry->function, entry->index,
1053 &entry->eax, &entry->ebx, &entry->ecx, &entry->edx);
1057 static void do_cpuid_ent(struct kvm_cpuid_entry2 *entry, u32 function,
1058 u32 index, int *nent, int maxnent)
1060 const u32 kvm_supported_word0_x86_features = bit(X86_FEATURE_FPU) |
1061 bit(X86_FEATURE_VME) | bit(X86_FEATURE_DE) |
1062 bit(X86_FEATURE_PSE) | bit(X86_FEATURE_TSC) |
1063 bit(X86_FEATURE_MSR) | bit(X86_FEATURE_PAE) |
1064 bit(X86_FEATURE_CX8) | bit(X86_FEATURE_APIC) |
1065 bit(X86_FEATURE_SEP) | bit(X86_FEATURE_PGE) |
1066 bit(X86_FEATURE_CMOV) | bit(X86_FEATURE_PSE36) |
1067 bit(X86_FEATURE_CLFLSH) | bit(X86_FEATURE_MMX) |
1068 bit(X86_FEATURE_FXSR) | bit(X86_FEATURE_XMM) |
1069 bit(X86_FEATURE_XMM2) | bit(X86_FEATURE_SELFSNOOP);
1070 const u32 kvm_supported_word1_x86_features = bit(X86_FEATURE_FPU) |
1071 bit(X86_FEATURE_VME) | bit(X86_FEATURE_DE) |
1072 bit(X86_FEATURE_PSE) | bit(X86_FEATURE_TSC) |
1073 bit(X86_FEATURE_MSR) | bit(X86_FEATURE_PAE) |
1074 bit(X86_FEATURE_CX8) | bit(X86_FEATURE_APIC) |
1075 bit(X86_FEATURE_PGE) |
1076 bit(X86_FEATURE_CMOV) | bit(X86_FEATURE_PSE36) |
1077 bit(X86_FEATURE_MMX) | bit(X86_FEATURE_FXSR) |
1078 bit(X86_FEATURE_SYSCALL) |
1079 (bit(X86_FEATURE_NX) && is_efer_nx()) |
1080 #ifdef CONFIG_X86_64
1081 bit(X86_FEATURE_LM) |
1083 bit(X86_FEATURE_MMXEXT) |
1084 bit(X86_FEATURE_3DNOWEXT) |
1085 bit(X86_FEATURE_3DNOW);
1086 const u32 kvm_supported_word3_x86_features =
1087 bit(X86_FEATURE_XMM3) | bit(X86_FEATURE_CX16);
1088 const u32 kvm_supported_word6_x86_features =
1089 bit(X86_FEATURE_LAHF_LM) | bit(X86_FEATURE_CMP_LEGACY);
1091 /* all func 2 cpuid_count() should be called on the same cpu */
1093 do_cpuid_1_ent(entry, function, index);
1098 entry->eax = min(entry->eax, (u32)0xb);
1101 entry->edx &= kvm_supported_word0_x86_features;
1102 entry->ecx &= kvm_supported_word3_x86_features;
1104 /* function 2 entries are STATEFUL. That is, repeated cpuid commands
1105 * may return different values. This forces us to get_cpu() before
1106 * issuing the first command, and also to emulate this annoying behavior
1107 * in kvm_emulate_cpuid() using KVM_CPUID_FLAG_STATE_READ_NEXT */
1109 int t, times = entry->eax & 0xff;
1111 entry->flags |= KVM_CPUID_FLAG_STATEFUL_FUNC;
1112 for (t = 1; t < times && *nent < maxnent; ++t) {
1113 do_cpuid_1_ent(&entry[t], function, 0);
1114 entry[t].flags |= KVM_CPUID_FLAG_STATEFUL_FUNC;
1119 /* function 4 and 0xb have additional index. */
1123 entry->flags |= KVM_CPUID_FLAG_SIGNIFCANT_INDEX;
1124 /* read more entries until cache_type is zero */
1125 for (i = 1; *nent < maxnent; ++i) {
1126 cache_type = entry[i - 1].eax & 0x1f;
1129 do_cpuid_1_ent(&entry[i], function, i);
1131 KVM_CPUID_FLAG_SIGNIFCANT_INDEX;
1139 entry->flags |= KVM_CPUID_FLAG_SIGNIFCANT_INDEX;
1140 /* read more entries until level_type is zero */
1141 for (i = 1; *nent < maxnent; ++i) {
1142 level_type = entry[i - 1].ecx & 0xff;
1145 do_cpuid_1_ent(&entry[i], function, i);
1147 KVM_CPUID_FLAG_SIGNIFCANT_INDEX;
1153 entry->eax = min(entry->eax, 0x8000001a);
1156 entry->edx &= kvm_supported_word1_x86_features;
1157 entry->ecx &= kvm_supported_word6_x86_features;
1163 static int kvm_dev_ioctl_get_supported_cpuid(struct kvm_cpuid2 *cpuid,
1164 struct kvm_cpuid_entry2 __user *entries)
1166 struct kvm_cpuid_entry2 *cpuid_entries;
1167 int limit, nent = 0, r = -E2BIG;
1170 if (cpuid->nent < 1)
1173 cpuid_entries = vmalloc(sizeof(struct kvm_cpuid_entry2) * cpuid->nent);
1177 do_cpuid_ent(&cpuid_entries[0], 0, 0, &nent, cpuid->nent);
1178 limit = cpuid_entries[0].eax;
1179 for (func = 1; func <= limit && nent < cpuid->nent; ++func)
1180 do_cpuid_ent(&cpuid_entries[nent], func, 0,
1181 &nent, cpuid->nent);
1183 if (nent >= cpuid->nent)
1186 do_cpuid_ent(&cpuid_entries[nent], 0x80000000, 0, &nent, cpuid->nent);
1187 limit = cpuid_entries[nent - 1].eax;
1188 for (func = 0x80000001; func <= limit && nent < cpuid->nent; ++func)
1189 do_cpuid_ent(&cpuid_entries[nent], func, 0,
1190 &nent, cpuid->nent);
1192 if (copy_to_user(entries, cpuid_entries,
1193 nent * sizeof(struct kvm_cpuid_entry2)))
1199 vfree(cpuid_entries);
1204 static int kvm_vcpu_ioctl_get_lapic(struct kvm_vcpu *vcpu,
1205 struct kvm_lapic_state *s)
1208 memcpy(s->regs, vcpu->arch.apic->regs, sizeof *s);
1214 static int kvm_vcpu_ioctl_set_lapic(struct kvm_vcpu *vcpu,
1215 struct kvm_lapic_state *s)
1218 memcpy(vcpu->arch.apic->regs, s->regs, sizeof *s);
1219 kvm_apic_post_state_restore(vcpu);
1225 static int kvm_vcpu_ioctl_interrupt(struct kvm_vcpu *vcpu,
1226 struct kvm_interrupt *irq)
1228 if (irq->irq < 0 || irq->irq >= 256)
1230 if (irqchip_in_kernel(vcpu->kvm))
1234 set_bit(irq->irq, vcpu->arch.irq_pending);
1235 set_bit(irq->irq / BITS_PER_LONG, &vcpu->arch.irq_summary);
1242 static int vcpu_ioctl_tpr_access_reporting(struct kvm_vcpu *vcpu,
1243 struct kvm_tpr_access_ctl *tac)
1247 vcpu->arch.tpr_access_reporting = !!tac->enabled;
1251 long kvm_arch_vcpu_ioctl(struct file *filp,
1252 unsigned int ioctl, unsigned long arg)
1254 struct kvm_vcpu *vcpu = filp->private_data;
1255 void __user *argp = (void __user *)arg;
1259 case KVM_GET_LAPIC: {
1260 struct kvm_lapic_state lapic;
1262 memset(&lapic, 0, sizeof lapic);
1263 r = kvm_vcpu_ioctl_get_lapic(vcpu, &lapic);
1267 if (copy_to_user(argp, &lapic, sizeof lapic))
1272 case KVM_SET_LAPIC: {
1273 struct kvm_lapic_state lapic;
1276 if (copy_from_user(&lapic, argp, sizeof lapic))
1278 r = kvm_vcpu_ioctl_set_lapic(vcpu, &lapic);;
1284 case KVM_INTERRUPT: {
1285 struct kvm_interrupt irq;
1288 if (copy_from_user(&irq, argp, sizeof irq))
1290 r = kvm_vcpu_ioctl_interrupt(vcpu, &irq);
1296 case KVM_SET_CPUID: {
1297 struct kvm_cpuid __user *cpuid_arg = argp;
1298 struct kvm_cpuid cpuid;
1301 if (copy_from_user(&cpuid, cpuid_arg, sizeof cpuid))
1303 r = kvm_vcpu_ioctl_set_cpuid(vcpu, &cpuid, cpuid_arg->entries);
1308 case KVM_SET_CPUID2: {
1309 struct kvm_cpuid2 __user *cpuid_arg = argp;
1310 struct kvm_cpuid2 cpuid;
1313 if (copy_from_user(&cpuid, cpuid_arg, sizeof cpuid))
1315 r = kvm_vcpu_ioctl_set_cpuid2(vcpu, &cpuid,
1316 cpuid_arg->entries);
1321 case KVM_GET_CPUID2: {
1322 struct kvm_cpuid2 __user *cpuid_arg = argp;
1323 struct kvm_cpuid2 cpuid;
1326 if (copy_from_user(&cpuid, cpuid_arg, sizeof cpuid))
1328 r = kvm_vcpu_ioctl_get_cpuid2(vcpu, &cpuid,
1329 cpuid_arg->entries);
1333 if (copy_to_user(cpuid_arg, &cpuid, sizeof cpuid))
1339 r = msr_io(vcpu, argp, kvm_get_msr, 1);
1342 r = msr_io(vcpu, argp, do_set_msr, 0);
1344 case KVM_TPR_ACCESS_REPORTING: {
1345 struct kvm_tpr_access_ctl tac;
1348 if (copy_from_user(&tac, argp, sizeof tac))
1350 r = vcpu_ioctl_tpr_access_reporting(vcpu, &tac);
1354 if (copy_to_user(argp, &tac, sizeof tac))
1359 case KVM_SET_VAPIC_ADDR: {
1360 struct kvm_vapic_addr va;
1363 if (!irqchip_in_kernel(vcpu->kvm))
1366 if (copy_from_user(&va, argp, sizeof va))
1369 kvm_lapic_set_vapic_addr(vcpu, va.vapic_addr);
1379 static int kvm_vm_ioctl_set_tss_addr(struct kvm *kvm, unsigned long addr)
1383 if (addr > (unsigned int)(-3 * PAGE_SIZE))
1385 ret = kvm_x86_ops->set_tss_addr(kvm, addr);
1389 static int kvm_vm_ioctl_set_nr_mmu_pages(struct kvm *kvm,
1390 u32 kvm_nr_mmu_pages)
1392 if (kvm_nr_mmu_pages < KVM_MIN_ALLOC_MMU_PAGES)
1395 down_write(&kvm->slots_lock);
1397 kvm_mmu_change_mmu_pages(kvm, kvm_nr_mmu_pages);
1398 kvm->arch.n_requested_mmu_pages = kvm_nr_mmu_pages;
1400 up_write(&kvm->slots_lock);
1404 static int kvm_vm_ioctl_get_nr_mmu_pages(struct kvm *kvm)
1406 return kvm->arch.n_alloc_mmu_pages;
1409 gfn_t unalias_gfn(struct kvm *kvm, gfn_t gfn)
1412 struct kvm_mem_alias *alias;
1414 for (i = 0; i < kvm->arch.naliases; ++i) {
1415 alias = &kvm->arch.aliases[i];
1416 if (gfn >= alias->base_gfn
1417 && gfn < alias->base_gfn + alias->npages)
1418 return alias->target_gfn + gfn - alias->base_gfn;
1424 * Set a new alias region. Aliases map a portion of physical memory into
1425 * another portion. This is useful for memory windows, for example the PC
1428 static int kvm_vm_ioctl_set_memory_alias(struct kvm *kvm,
1429 struct kvm_memory_alias *alias)
1432 struct kvm_mem_alias *p;
1435 /* General sanity checks */
1436 if (alias->memory_size & (PAGE_SIZE - 1))
1438 if (alias->guest_phys_addr & (PAGE_SIZE - 1))
1440 if (alias->slot >= KVM_ALIAS_SLOTS)
1442 if (alias->guest_phys_addr + alias->memory_size
1443 < alias->guest_phys_addr)
1445 if (alias->target_phys_addr + alias->memory_size
1446 < alias->target_phys_addr)
1449 down_write(&kvm->slots_lock);
1451 p = &kvm->arch.aliases[alias->slot];
1452 p->base_gfn = alias->guest_phys_addr >> PAGE_SHIFT;
1453 p->npages = alias->memory_size >> PAGE_SHIFT;
1454 p->target_gfn = alias->target_phys_addr >> PAGE_SHIFT;
1456 for (n = KVM_ALIAS_SLOTS; n > 0; --n)
1457 if (kvm->arch.aliases[n - 1].npages)
1459 kvm->arch.naliases = n;
1461 kvm_mmu_zap_all(kvm);
1463 up_write(&kvm->slots_lock);
1471 static int kvm_vm_ioctl_get_irqchip(struct kvm *kvm, struct kvm_irqchip *chip)
1476 switch (chip->chip_id) {
1477 case KVM_IRQCHIP_PIC_MASTER:
1478 memcpy(&chip->chip.pic,
1479 &pic_irqchip(kvm)->pics[0],
1480 sizeof(struct kvm_pic_state));
1482 case KVM_IRQCHIP_PIC_SLAVE:
1483 memcpy(&chip->chip.pic,
1484 &pic_irqchip(kvm)->pics[1],
1485 sizeof(struct kvm_pic_state));
1487 case KVM_IRQCHIP_IOAPIC:
1488 memcpy(&chip->chip.ioapic,
1489 ioapic_irqchip(kvm),
1490 sizeof(struct kvm_ioapic_state));
1499 static int kvm_vm_ioctl_set_irqchip(struct kvm *kvm, struct kvm_irqchip *chip)
1504 switch (chip->chip_id) {
1505 case KVM_IRQCHIP_PIC_MASTER:
1506 memcpy(&pic_irqchip(kvm)->pics[0],
1508 sizeof(struct kvm_pic_state));
1510 case KVM_IRQCHIP_PIC_SLAVE:
1511 memcpy(&pic_irqchip(kvm)->pics[1],
1513 sizeof(struct kvm_pic_state));
1515 case KVM_IRQCHIP_IOAPIC:
1516 memcpy(ioapic_irqchip(kvm),
1518 sizeof(struct kvm_ioapic_state));
1524 kvm_pic_update_irq(pic_irqchip(kvm));
1528 static int kvm_vm_ioctl_get_pit(struct kvm *kvm, struct kvm_pit_state *ps)
1532 memcpy(ps, &kvm->arch.vpit->pit_state, sizeof(struct kvm_pit_state));
1536 static int kvm_vm_ioctl_set_pit(struct kvm *kvm, struct kvm_pit_state *ps)
1540 memcpy(&kvm->arch.vpit->pit_state, ps, sizeof(struct kvm_pit_state));
1541 kvm_pit_load_count(kvm, 0, ps->channels[0].count);
1546 * Get (and clear) the dirty memory log for a memory slot.
1548 int kvm_vm_ioctl_get_dirty_log(struct kvm *kvm,
1549 struct kvm_dirty_log *log)
1553 struct kvm_memory_slot *memslot;
1556 down_write(&kvm->slots_lock);
1558 r = kvm_get_dirty_log(kvm, log, &is_dirty);
1562 /* If nothing is dirty, don't bother messing with page tables. */
1564 kvm_mmu_slot_remove_write_access(kvm, log->slot);
1565 kvm_flush_remote_tlbs(kvm);
1566 memslot = &kvm->memslots[log->slot];
1567 n = ALIGN(memslot->npages, BITS_PER_LONG) / 8;
1568 memset(memslot->dirty_bitmap, 0, n);
1572 up_write(&kvm->slots_lock);
1576 long kvm_arch_vm_ioctl(struct file *filp,
1577 unsigned int ioctl, unsigned long arg)
1579 struct kvm *kvm = filp->private_data;
1580 void __user *argp = (void __user *)arg;
1584 case KVM_SET_TSS_ADDR:
1585 r = kvm_vm_ioctl_set_tss_addr(kvm, arg);
1589 case KVM_SET_MEMORY_REGION: {
1590 struct kvm_memory_region kvm_mem;
1591 struct kvm_userspace_memory_region kvm_userspace_mem;
1594 if (copy_from_user(&kvm_mem, argp, sizeof kvm_mem))
1596 kvm_userspace_mem.slot = kvm_mem.slot;
1597 kvm_userspace_mem.flags = kvm_mem.flags;
1598 kvm_userspace_mem.guest_phys_addr = kvm_mem.guest_phys_addr;
1599 kvm_userspace_mem.memory_size = kvm_mem.memory_size;
1600 r = kvm_vm_ioctl_set_memory_region(kvm, &kvm_userspace_mem, 0);
1605 case KVM_SET_NR_MMU_PAGES:
1606 r = kvm_vm_ioctl_set_nr_mmu_pages(kvm, arg);
1610 case KVM_GET_NR_MMU_PAGES:
1611 r = kvm_vm_ioctl_get_nr_mmu_pages(kvm);
1613 case KVM_SET_MEMORY_ALIAS: {
1614 struct kvm_memory_alias alias;
1617 if (copy_from_user(&alias, argp, sizeof alias))
1619 r = kvm_vm_ioctl_set_memory_alias(kvm, &alias);
1624 case KVM_CREATE_IRQCHIP:
1626 kvm->arch.vpic = kvm_create_pic(kvm);
1627 if (kvm->arch.vpic) {
1628 r = kvm_ioapic_init(kvm);
1630 kfree(kvm->arch.vpic);
1631 kvm->arch.vpic = NULL;
1637 case KVM_CREATE_PIT:
1639 kvm->arch.vpit = kvm_create_pit(kvm);
1643 case KVM_IRQ_LINE: {
1644 struct kvm_irq_level irq_event;
1647 if (copy_from_user(&irq_event, argp, sizeof irq_event))
1649 if (irqchip_in_kernel(kvm)) {
1650 mutex_lock(&kvm->lock);
1651 if (irq_event.irq < 16)
1652 kvm_pic_set_irq(pic_irqchip(kvm),
1655 kvm_ioapic_set_irq(kvm->arch.vioapic,
1658 mutex_unlock(&kvm->lock);
1663 case KVM_GET_IRQCHIP: {
1664 /* 0: PIC master, 1: PIC slave, 2: IOAPIC */
1665 struct kvm_irqchip chip;
1668 if (copy_from_user(&chip, argp, sizeof chip))
1671 if (!irqchip_in_kernel(kvm))
1673 r = kvm_vm_ioctl_get_irqchip(kvm, &chip);
1677 if (copy_to_user(argp, &chip, sizeof chip))
1682 case KVM_SET_IRQCHIP: {
1683 /* 0: PIC master, 1: PIC slave, 2: IOAPIC */
1684 struct kvm_irqchip chip;
1687 if (copy_from_user(&chip, argp, sizeof chip))
1690 if (!irqchip_in_kernel(kvm))
1692 r = kvm_vm_ioctl_set_irqchip(kvm, &chip);
1699 struct kvm_pit_state ps;
1701 if (copy_from_user(&ps, argp, sizeof ps))
1704 if (!kvm->arch.vpit)
1706 r = kvm_vm_ioctl_get_pit(kvm, &ps);
1710 if (copy_to_user(argp, &ps, sizeof ps))
1716 struct kvm_pit_state ps;
1718 if (copy_from_user(&ps, argp, sizeof ps))
1721 if (!kvm->arch.vpit)
1723 r = kvm_vm_ioctl_set_pit(kvm, &ps);
1736 static void kvm_init_msr_list(void)
1741 for (i = j = 0; i < ARRAY_SIZE(msrs_to_save); i++) {
1742 if (rdmsr_safe(msrs_to_save[i], &dummy[0], &dummy[1]) < 0)
1745 msrs_to_save[j] = msrs_to_save[i];
1748 num_msrs_to_save = j;
1752 * Only apic need an MMIO device hook, so shortcut now..
1754 static struct kvm_io_device *vcpu_find_pervcpu_dev(struct kvm_vcpu *vcpu,
1757 struct kvm_io_device *dev;
1759 if (vcpu->arch.apic) {
1760 dev = &vcpu->arch.apic->dev;
1761 if (dev->in_range(dev, addr))
1768 static struct kvm_io_device *vcpu_find_mmio_dev(struct kvm_vcpu *vcpu,
1771 struct kvm_io_device *dev;
1773 dev = vcpu_find_pervcpu_dev(vcpu, addr);
1775 dev = kvm_io_bus_find_dev(&vcpu->kvm->mmio_bus, addr);
1779 int emulator_read_std(unsigned long addr,
1782 struct kvm_vcpu *vcpu)
1785 int r = X86EMUL_CONTINUE;
1788 gpa_t gpa = vcpu->arch.mmu.gva_to_gpa(vcpu, addr);
1789 unsigned offset = addr & (PAGE_SIZE-1);
1790 unsigned tocopy = min(bytes, (unsigned)PAGE_SIZE - offset);
1793 if (gpa == UNMAPPED_GVA) {
1794 r = X86EMUL_PROPAGATE_FAULT;
1797 ret = kvm_read_guest(vcpu->kvm, gpa, data, tocopy);
1799 r = X86EMUL_UNHANDLEABLE;
1810 EXPORT_SYMBOL_GPL(emulator_read_std);
1812 static int emulator_read_emulated(unsigned long addr,
1815 struct kvm_vcpu *vcpu)
1817 struct kvm_io_device *mmio_dev;
1820 if (vcpu->mmio_read_completed) {
1821 memcpy(val, vcpu->mmio_data, bytes);
1822 vcpu->mmio_read_completed = 0;
1823 return X86EMUL_CONTINUE;
1826 gpa = vcpu->arch.mmu.gva_to_gpa(vcpu, addr);
1828 /* For APIC access vmexit */
1829 if ((gpa & PAGE_MASK) == APIC_DEFAULT_PHYS_BASE)
1832 if (emulator_read_std(addr, val, bytes, vcpu)
1833 == X86EMUL_CONTINUE)
1834 return X86EMUL_CONTINUE;
1835 if (gpa == UNMAPPED_GVA)
1836 return X86EMUL_PROPAGATE_FAULT;
1840 * Is this MMIO handled locally?
1842 mutex_lock(&vcpu->kvm->lock);
1843 mmio_dev = vcpu_find_mmio_dev(vcpu, gpa);
1845 kvm_iodevice_read(mmio_dev, gpa, bytes, val);
1846 mutex_unlock(&vcpu->kvm->lock);
1847 return X86EMUL_CONTINUE;
1849 mutex_unlock(&vcpu->kvm->lock);
1851 vcpu->mmio_needed = 1;
1852 vcpu->mmio_phys_addr = gpa;
1853 vcpu->mmio_size = bytes;
1854 vcpu->mmio_is_write = 0;
1856 return X86EMUL_UNHANDLEABLE;
1859 int emulator_write_phys(struct kvm_vcpu *vcpu, gpa_t gpa,
1860 const void *val, int bytes)
1864 ret = kvm_write_guest(vcpu->kvm, gpa, val, bytes);
1867 kvm_mmu_pte_write(vcpu, gpa, val, bytes);
1871 static int emulator_write_emulated_onepage(unsigned long addr,
1874 struct kvm_vcpu *vcpu)
1876 struct kvm_io_device *mmio_dev;
1879 gpa = vcpu->arch.mmu.gva_to_gpa(vcpu, addr);
1881 if (gpa == UNMAPPED_GVA) {
1882 kvm_inject_page_fault(vcpu, addr, 2);
1883 return X86EMUL_PROPAGATE_FAULT;
1886 /* For APIC access vmexit */
1887 if ((gpa & PAGE_MASK) == APIC_DEFAULT_PHYS_BASE)
1890 if (emulator_write_phys(vcpu, gpa, val, bytes))
1891 return X86EMUL_CONTINUE;
1895 * Is this MMIO handled locally?
1897 mutex_lock(&vcpu->kvm->lock);
1898 mmio_dev = vcpu_find_mmio_dev(vcpu, gpa);
1900 kvm_iodevice_write(mmio_dev, gpa, bytes, val);
1901 mutex_unlock(&vcpu->kvm->lock);
1902 return X86EMUL_CONTINUE;
1904 mutex_unlock(&vcpu->kvm->lock);
1906 vcpu->mmio_needed = 1;
1907 vcpu->mmio_phys_addr = gpa;
1908 vcpu->mmio_size = bytes;
1909 vcpu->mmio_is_write = 1;
1910 memcpy(vcpu->mmio_data, val, bytes);
1912 return X86EMUL_CONTINUE;
1915 int emulator_write_emulated(unsigned long addr,
1918 struct kvm_vcpu *vcpu)
1920 /* Crossing a page boundary? */
1921 if (((addr + bytes - 1) ^ addr) & PAGE_MASK) {
1924 now = -addr & ~PAGE_MASK;
1925 rc = emulator_write_emulated_onepage(addr, val, now, vcpu);
1926 if (rc != X86EMUL_CONTINUE)
1932 return emulator_write_emulated_onepage(addr, val, bytes, vcpu);
1934 EXPORT_SYMBOL_GPL(emulator_write_emulated);
1936 static int emulator_cmpxchg_emulated(unsigned long addr,
1940 struct kvm_vcpu *vcpu)
1942 static int reported;
1946 printk(KERN_WARNING "kvm: emulating exchange as write\n");
1948 #ifndef CONFIG_X86_64
1949 /* guests cmpxchg8b have to be emulated atomically */
1956 gpa = vcpu->arch.mmu.gva_to_gpa(vcpu, addr);
1958 if (gpa == UNMAPPED_GVA ||
1959 (gpa & PAGE_MASK) == APIC_DEFAULT_PHYS_BASE)
1962 if (((gpa + bytes - 1) & PAGE_MASK) != (gpa & PAGE_MASK))
1967 down_read(¤t->mm->mmap_sem);
1968 page = gfn_to_page(vcpu->kvm, gpa >> PAGE_SHIFT);
1969 up_read(¤t->mm->mmap_sem);
1971 kaddr = kmap_atomic(page, KM_USER0);
1972 set_64bit((u64 *)(kaddr + offset_in_page(gpa)), val);
1973 kunmap_atomic(kaddr, KM_USER0);
1974 kvm_release_page_dirty(page);
1979 return emulator_write_emulated(addr, new, bytes, vcpu);
1982 static unsigned long get_segment_base(struct kvm_vcpu *vcpu, int seg)
1984 return kvm_x86_ops->get_segment_base(vcpu, seg);
1987 int emulate_invlpg(struct kvm_vcpu *vcpu, gva_t address)
1989 return X86EMUL_CONTINUE;
1992 int emulate_clts(struct kvm_vcpu *vcpu)
1994 KVMTRACE_0D(CLTS, vcpu, handler);
1995 kvm_x86_ops->set_cr0(vcpu, vcpu->arch.cr0 & ~X86_CR0_TS);
1996 return X86EMUL_CONTINUE;
1999 int emulator_get_dr(struct x86_emulate_ctxt *ctxt, int dr, unsigned long *dest)
2001 struct kvm_vcpu *vcpu = ctxt->vcpu;
2005 *dest = kvm_x86_ops->get_dr(vcpu, dr);
2006 return X86EMUL_CONTINUE;
2008 pr_unimpl(vcpu, "%s: unexpected dr %u\n", __func__, dr);
2009 return X86EMUL_UNHANDLEABLE;
2013 int emulator_set_dr(struct x86_emulate_ctxt *ctxt, int dr, unsigned long value)
2015 unsigned long mask = (ctxt->mode == X86EMUL_MODE_PROT64) ? ~0ULL : ~0U;
2018 kvm_x86_ops->set_dr(ctxt->vcpu, dr, value & mask, &exception);
2020 /* FIXME: better handling */
2021 return X86EMUL_UNHANDLEABLE;
2023 return X86EMUL_CONTINUE;
2026 void kvm_report_emulation_failure(struct kvm_vcpu *vcpu, const char *context)
2028 static int reported;
2030 unsigned long rip = vcpu->arch.rip;
2031 unsigned long rip_linear;
2033 rip_linear = rip + get_segment_base(vcpu, VCPU_SREG_CS);
2038 emulator_read_std(rip_linear, (void *)opcodes, 4, vcpu);
2040 printk(KERN_ERR "emulation failed (%s) rip %lx %02x %02x %02x %02x\n",
2041 context, rip, opcodes[0], opcodes[1], opcodes[2], opcodes[3]);
2044 EXPORT_SYMBOL_GPL(kvm_report_emulation_failure);
2046 static struct x86_emulate_ops emulate_ops = {
2047 .read_std = emulator_read_std,
2048 .read_emulated = emulator_read_emulated,
2049 .write_emulated = emulator_write_emulated,
2050 .cmpxchg_emulated = emulator_cmpxchg_emulated,
2053 int emulate_instruction(struct kvm_vcpu *vcpu,
2054 struct kvm_run *run,
2060 struct decode_cache *c;
2062 vcpu->arch.mmio_fault_cr2 = cr2;
2063 kvm_x86_ops->cache_regs(vcpu);
2065 vcpu->mmio_is_write = 0;
2066 vcpu->arch.pio.string = 0;
2068 if (!(emulation_type & EMULTYPE_NO_DECODE)) {
2070 kvm_x86_ops->get_cs_db_l_bits(vcpu, &cs_db, &cs_l);
2072 vcpu->arch.emulate_ctxt.vcpu = vcpu;
2073 vcpu->arch.emulate_ctxt.eflags = kvm_x86_ops->get_rflags(vcpu);
2074 vcpu->arch.emulate_ctxt.mode =
2075 (vcpu->arch.emulate_ctxt.eflags & X86_EFLAGS_VM)
2076 ? X86EMUL_MODE_REAL : cs_l
2077 ? X86EMUL_MODE_PROT64 : cs_db
2078 ? X86EMUL_MODE_PROT32 : X86EMUL_MODE_PROT16;
2080 if (vcpu->arch.emulate_ctxt.mode == X86EMUL_MODE_PROT64) {
2081 vcpu->arch.emulate_ctxt.cs_base = 0;
2082 vcpu->arch.emulate_ctxt.ds_base = 0;
2083 vcpu->arch.emulate_ctxt.es_base = 0;
2084 vcpu->arch.emulate_ctxt.ss_base = 0;
2086 vcpu->arch.emulate_ctxt.cs_base =
2087 get_segment_base(vcpu, VCPU_SREG_CS);
2088 vcpu->arch.emulate_ctxt.ds_base =
2089 get_segment_base(vcpu, VCPU_SREG_DS);
2090 vcpu->arch.emulate_ctxt.es_base =
2091 get_segment_base(vcpu, VCPU_SREG_ES);
2092 vcpu->arch.emulate_ctxt.ss_base =
2093 get_segment_base(vcpu, VCPU_SREG_SS);
2096 vcpu->arch.emulate_ctxt.gs_base =
2097 get_segment_base(vcpu, VCPU_SREG_GS);
2098 vcpu->arch.emulate_ctxt.fs_base =
2099 get_segment_base(vcpu, VCPU_SREG_FS);
2101 r = x86_decode_insn(&vcpu->arch.emulate_ctxt, &emulate_ops);
2103 /* Reject the instructions other than VMCALL/VMMCALL when
2104 * try to emulate invalid opcode */
2105 c = &vcpu->arch.emulate_ctxt.decode;
2106 if ((emulation_type & EMULTYPE_TRAP_UD) &&
2107 (!(c->twobyte && c->b == 0x01 &&
2108 (c->modrm_reg == 0 || c->modrm_reg == 3) &&
2109 c->modrm_mod == 3 && c->modrm_rm == 1)))
2110 return EMULATE_FAIL;
2112 ++vcpu->stat.insn_emulation;
2114 ++vcpu->stat.insn_emulation_fail;
2115 if (kvm_mmu_unprotect_page_virt(vcpu, cr2))
2116 return EMULATE_DONE;
2117 return EMULATE_FAIL;
2121 r = x86_emulate_insn(&vcpu->arch.emulate_ctxt, &emulate_ops);
2123 if (vcpu->arch.pio.string)
2124 return EMULATE_DO_MMIO;
2126 if ((r || vcpu->mmio_is_write) && run) {
2127 run->exit_reason = KVM_EXIT_MMIO;
2128 run->mmio.phys_addr = vcpu->mmio_phys_addr;
2129 memcpy(run->mmio.data, vcpu->mmio_data, 8);
2130 run->mmio.len = vcpu->mmio_size;
2131 run->mmio.is_write = vcpu->mmio_is_write;
2135 if (kvm_mmu_unprotect_page_virt(vcpu, cr2))
2136 return EMULATE_DONE;
2137 if (!vcpu->mmio_needed) {
2138 kvm_report_emulation_failure(vcpu, "mmio");
2139 return EMULATE_FAIL;
2141 return EMULATE_DO_MMIO;
2144 kvm_x86_ops->decache_regs(vcpu);
2145 kvm_x86_ops->set_rflags(vcpu, vcpu->arch.emulate_ctxt.eflags);
2147 if (vcpu->mmio_is_write) {
2148 vcpu->mmio_needed = 0;
2149 return EMULATE_DO_MMIO;
2152 return EMULATE_DONE;
2154 EXPORT_SYMBOL_GPL(emulate_instruction);
2156 static void free_pio_guest_pages(struct kvm_vcpu *vcpu)
2160 for (i = 0; i < ARRAY_SIZE(vcpu->arch.pio.guest_pages); ++i)
2161 if (vcpu->arch.pio.guest_pages[i]) {
2162 kvm_release_page_dirty(vcpu->arch.pio.guest_pages[i]);
2163 vcpu->arch.pio.guest_pages[i] = NULL;
2167 static int pio_copy_data(struct kvm_vcpu *vcpu)
2169 void *p = vcpu->arch.pio_data;
2172 int nr_pages = vcpu->arch.pio.guest_pages[1] ? 2 : 1;
2174 q = vmap(vcpu->arch.pio.guest_pages, nr_pages, VM_READ|VM_WRITE,
2177 free_pio_guest_pages(vcpu);
2180 q += vcpu->arch.pio.guest_page_offset;
2181 bytes = vcpu->arch.pio.size * vcpu->arch.pio.cur_count;
2182 if (vcpu->arch.pio.in)
2183 memcpy(q, p, bytes);
2185 memcpy(p, q, bytes);
2186 q -= vcpu->arch.pio.guest_page_offset;
2188 free_pio_guest_pages(vcpu);
2192 int complete_pio(struct kvm_vcpu *vcpu)
2194 struct kvm_pio_request *io = &vcpu->arch.pio;
2198 kvm_x86_ops->cache_regs(vcpu);
2202 memcpy(&vcpu->arch.regs[VCPU_REGS_RAX], vcpu->arch.pio_data,
2206 r = pio_copy_data(vcpu);
2208 kvm_x86_ops->cache_regs(vcpu);
2215 delta *= io->cur_count;
2217 * The size of the register should really depend on
2218 * current address size.
2220 vcpu->arch.regs[VCPU_REGS_RCX] -= delta;
2226 vcpu->arch.regs[VCPU_REGS_RDI] += delta;
2228 vcpu->arch.regs[VCPU_REGS_RSI] += delta;
2231 kvm_x86_ops->decache_regs(vcpu);
2233 io->count -= io->cur_count;
2239 static void kernel_pio(struct kvm_io_device *pio_dev,
2240 struct kvm_vcpu *vcpu,
2243 /* TODO: String I/O for in kernel device */
2245 mutex_lock(&vcpu->kvm->lock);
2246 if (vcpu->arch.pio.in)
2247 kvm_iodevice_read(pio_dev, vcpu->arch.pio.port,
2248 vcpu->arch.pio.size,
2251 kvm_iodevice_write(pio_dev, vcpu->arch.pio.port,
2252 vcpu->arch.pio.size,
2254 mutex_unlock(&vcpu->kvm->lock);
2257 static void pio_string_write(struct kvm_io_device *pio_dev,
2258 struct kvm_vcpu *vcpu)
2260 struct kvm_pio_request *io = &vcpu->arch.pio;
2261 void *pd = vcpu->arch.pio_data;
2264 mutex_lock(&vcpu->kvm->lock);
2265 for (i = 0; i < io->cur_count; i++) {
2266 kvm_iodevice_write(pio_dev, io->port,
2271 mutex_unlock(&vcpu->kvm->lock);
2274 static struct kvm_io_device *vcpu_find_pio_dev(struct kvm_vcpu *vcpu,
2277 return kvm_io_bus_find_dev(&vcpu->kvm->pio_bus, addr);
2280 int kvm_emulate_pio(struct kvm_vcpu *vcpu, struct kvm_run *run, int in,
2281 int size, unsigned port)
2283 struct kvm_io_device *pio_dev;
2285 vcpu->run->exit_reason = KVM_EXIT_IO;
2286 vcpu->run->io.direction = in ? KVM_EXIT_IO_IN : KVM_EXIT_IO_OUT;
2287 vcpu->run->io.size = vcpu->arch.pio.size = size;
2288 vcpu->run->io.data_offset = KVM_PIO_PAGE_OFFSET * PAGE_SIZE;
2289 vcpu->run->io.count = vcpu->arch.pio.count = vcpu->arch.pio.cur_count = 1;
2290 vcpu->run->io.port = vcpu->arch.pio.port = port;
2291 vcpu->arch.pio.in = in;
2292 vcpu->arch.pio.string = 0;
2293 vcpu->arch.pio.down = 0;
2294 vcpu->arch.pio.guest_page_offset = 0;
2295 vcpu->arch.pio.rep = 0;
2297 if (vcpu->run->io.direction == KVM_EXIT_IO_IN)
2298 KVMTRACE_2D(IO_READ, vcpu, vcpu->run->io.port, (u32)size,
2301 KVMTRACE_2D(IO_WRITE, vcpu, vcpu->run->io.port, (u32)size,
2304 kvm_x86_ops->cache_regs(vcpu);
2305 memcpy(vcpu->arch.pio_data, &vcpu->arch.regs[VCPU_REGS_RAX], 4);
2307 kvm_x86_ops->skip_emulated_instruction(vcpu);
2309 pio_dev = vcpu_find_pio_dev(vcpu, port);
2311 kernel_pio(pio_dev, vcpu, vcpu->arch.pio_data);
2317 EXPORT_SYMBOL_GPL(kvm_emulate_pio);
2319 int kvm_emulate_pio_string(struct kvm_vcpu *vcpu, struct kvm_run *run, int in,
2320 int size, unsigned long count, int down,
2321 gva_t address, int rep, unsigned port)
2323 unsigned now, in_page;
2327 struct kvm_io_device *pio_dev;
2329 vcpu->run->exit_reason = KVM_EXIT_IO;
2330 vcpu->run->io.direction = in ? KVM_EXIT_IO_IN : KVM_EXIT_IO_OUT;
2331 vcpu->run->io.size = vcpu->arch.pio.size = size;
2332 vcpu->run->io.data_offset = KVM_PIO_PAGE_OFFSET * PAGE_SIZE;
2333 vcpu->run->io.count = vcpu->arch.pio.count = vcpu->arch.pio.cur_count = count;
2334 vcpu->run->io.port = vcpu->arch.pio.port = port;
2335 vcpu->arch.pio.in = in;
2336 vcpu->arch.pio.string = 1;
2337 vcpu->arch.pio.down = down;
2338 vcpu->arch.pio.guest_page_offset = offset_in_page(address);
2339 vcpu->arch.pio.rep = rep;
2341 if (vcpu->run->io.direction == KVM_EXIT_IO_IN)
2342 KVMTRACE_2D(IO_READ, vcpu, vcpu->run->io.port, (u32)size,
2345 KVMTRACE_2D(IO_WRITE, vcpu, vcpu->run->io.port, (u32)size,
2349 kvm_x86_ops->skip_emulated_instruction(vcpu);
2354 in_page = PAGE_SIZE - offset_in_page(address);
2356 in_page = offset_in_page(address) + size;
2357 now = min(count, (unsigned long)in_page / size);
2360 * String I/O straddles page boundary. Pin two guest pages
2361 * so that we satisfy atomicity constraints. Do just one
2362 * transaction to avoid complexity.
2369 * String I/O in reverse. Yuck. Kill the guest, fix later.
2371 pr_unimpl(vcpu, "guest string pio down\n");
2372 kvm_inject_gp(vcpu, 0);
2375 vcpu->run->io.count = now;
2376 vcpu->arch.pio.cur_count = now;
2378 if (vcpu->arch.pio.cur_count == vcpu->arch.pio.count)
2379 kvm_x86_ops->skip_emulated_instruction(vcpu);
2381 for (i = 0; i < nr_pages; ++i) {
2382 page = gva_to_page(vcpu, address + i * PAGE_SIZE);
2383 vcpu->arch.pio.guest_pages[i] = page;
2385 kvm_inject_gp(vcpu, 0);
2386 free_pio_guest_pages(vcpu);
2391 pio_dev = vcpu_find_pio_dev(vcpu, port);
2392 if (!vcpu->arch.pio.in) {
2393 /* string PIO write */
2394 ret = pio_copy_data(vcpu);
2395 if (ret >= 0 && pio_dev) {
2396 pio_string_write(pio_dev, vcpu);
2398 if (vcpu->arch.pio.count == 0)
2402 pr_unimpl(vcpu, "no string pio read support yet, "
2403 "port %x size %d count %ld\n",
2408 EXPORT_SYMBOL_GPL(kvm_emulate_pio_string);
2410 int kvm_arch_init(void *opaque)
2413 struct kvm_x86_ops *ops = (struct kvm_x86_ops *)opaque;
2416 printk(KERN_ERR "kvm: already loaded the other module\n");
2421 if (!ops->cpu_has_kvm_support()) {
2422 printk(KERN_ERR "kvm: no hardware support\n");
2426 if (ops->disabled_by_bios()) {
2427 printk(KERN_ERR "kvm: disabled by bios\n");
2432 r = kvm_mmu_module_init();
2436 kvm_init_msr_list();
2439 kvm_mmu_set_nonpresent_ptes(0ull, 0ull);
2440 kvm_mmu_set_base_ptes(PT_PRESENT_MASK);
2441 kvm_mmu_set_mask_ptes(PT_USER_MASK, PT_ACCESSED_MASK,
2442 PT_DIRTY_MASK, PT64_NX_MASK, 0);
2449 void kvm_arch_exit(void)
2452 kvm_mmu_module_exit();
2455 int kvm_emulate_halt(struct kvm_vcpu *vcpu)
2457 ++vcpu->stat.halt_exits;
2458 KVMTRACE_0D(HLT, vcpu, handler);
2459 if (irqchip_in_kernel(vcpu->kvm)) {
2460 vcpu->arch.mp_state = KVM_MP_STATE_HALTED;
2461 up_read(&vcpu->kvm->slots_lock);
2462 kvm_vcpu_block(vcpu);
2463 down_read(&vcpu->kvm->slots_lock);
2464 if (vcpu->arch.mp_state != KVM_MP_STATE_RUNNABLE)
2468 vcpu->run->exit_reason = KVM_EXIT_HLT;
2472 EXPORT_SYMBOL_GPL(kvm_emulate_halt);
2474 static inline gpa_t hc_gpa(struct kvm_vcpu *vcpu, unsigned long a0,
2477 if (is_long_mode(vcpu))
2480 return a0 | ((gpa_t)a1 << 32);
2483 int kvm_emulate_hypercall(struct kvm_vcpu *vcpu)
2485 unsigned long nr, a0, a1, a2, a3, ret;
2488 kvm_x86_ops->cache_regs(vcpu);
2490 nr = vcpu->arch.regs[VCPU_REGS_RAX];
2491 a0 = vcpu->arch.regs[VCPU_REGS_RBX];
2492 a1 = vcpu->arch.regs[VCPU_REGS_RCX];
2493 a2 = vcpu->arch.regs[VCPU_REGS_RDX];
2494 a3 = vcpu->arch.regs[VCPU_REGS_RSI];
2496 KVMTRACE_1D(VMMCALL, vcpu, (u32)nr, handler);
2498 if (!is_long_mode(vcpu)) {
2507 case KVM_HC_VAPIC_POLL_IRQ:
2511 r = kvm_pv_mmu_op(vcpu, a0, hc_gpa(vcpu, a1, a2), &ret);
2517 vcpu->arch.regs[VCPU_REGS_RAX] = ret;
2518 kvm_x86_ops->decache_regs(vcpu);
2519 ++vcpu->stat.hypercalls;
2522 EXPORT_SYMBOL_GPL(kvm_emulate_hypercall);
2524 int kvm_fix_hypercall(struct kvm_vcpu *vcpu)
2526 char instruction[3];
2531 * Blow out the MMU to ensure that no other VCPU has an active mapping
2532 * to ensure that the updated hypercall appears atomically across all
2535 kvm_mmu_zap_all(vcpu->kvm);
2537 kvm_x86_ops->cache_regs(vcpu);
2538 kvm_x86_ops->patch_hypercall(vcpu, instruction);
2539 if (emulator_write_emulated(vcpu->arch.rip, instruction, 3, vcpu)
2540 != X86EMUL_CONTINUE)
2546 static u64 mk_cr_64(u64 curr_cr, u32 new_val)
2548 return (curr_cr & ~((1ULL << 32) - 1)) | new_val;
2551 void realmode_lgdt(struct kvm_vcpu *vcpu, u16 limit, unsigned long base)
2553 struct descriptor_table dt = { limit, base };
2555 kvm_x86_ops->set_gdt(vcpu, &dt);
2558 void realmode_lidt(struct kvm_vcpu *vcpu, u16 limit, unsigned long base)
2560 struct descriptor_table dt = { limit, base };
2562 kvm_x86_ops->set_idt(vcpu, &dt);
2565 void realmode_lmsw(struct kvm_vcpu *vcpu, unsigned long msw,
2566 unsigned long *rflags)
2568 kvm_lmsw(vcpu, msw);
2569 *rflags = kvm_x86_ops->get_rflags(vcpu);
2572 unsigned long realmode_get_cr(struct kvm_vcpu *vcpu, int cr)
2574 unsigned long value;
2576 kvm_x86_ops->decache_cr4_guest_bits(vcpu);
2579 value = vcpu->arch.cr0;
2582 value = vcpu->arch.cr2;
2585 value = vcpu->arch.cr3;
2588 value = vcpu->arch.cr4;
2591 value = kvm_get_cr8(vcpu);
2594 vcpu_printf(vcpu, "%s: unexpected cr %u\n", __func__, cr);
2597 KVMTRACE_3D(CR_READ, vcpu, (u32)cr, (u32)value,
2598 (u32)((u64)value >> 32), handler);
2603 void realmode_set_cr(struct kvm_vcpu *vcpu, int cr, unsigned long val,
2604 unsigned long *rflags)
2606 KVMTRACE_3D(CR_WRITE, vcpu, (u32)cr, (u32)val,
2607 (u32)((u64)val >> 32), handler);
2611 kvm_set_cr0(vcpu, mk_cr_64(vcpu->arch.cr0, val));
2612 *rflags = kvm_x86_ops->get_rflags(vcpu);
2615 vcpu->arch.cr2 = val;
2618 kvm_set_cr3(vcpu, val);
2621 kvm_set_cr4(vcpu, mk_cr_64(vcpu->arch.cr4, val));
2624 kvm_set_cr8(vcpu, val & 0xfUL);
2627 vcpu_printf(vcpu, "%s: unexpected cr %u\n", __func__, cr);
2631 static int move_to_next_stateful_cpuid_entry(struct kvm_vcpu *vcpu, int i)
2633 struct kvm_cpuid_entry2 *e = &vcpu->arch.cpuid_entries[i];
2634 int j, nent = vcpu->arch.cpuid_nent;
2636 e->flags &= ~KVM_CPUID_FLAG_STATE_READ_NEXT;
2637 /* when no next entry is found, the current entry[i] is reselected */
2638 for (j = i + 1; j == i; j = (j + 1) % nent) {
2639 struct kvm_cpuid_entry2 *ej = &vcpu->arch.cpuid_entries[j];
2640 if (ej->function == e->function) {
2641 ej->flags |= KVM_CPUID_FLAG_STATE_READ_NEXT;
2645 return 0; /* silence gcc, even though control never reaches here */
2648 /* find an entry with matching function, matching index (if needed), and that
2649 * should be read next (if it's stateful) */
2650 static int is_matching_cpuid_entry(struct kvm_cpuid_entry2 *e,
2651 u32 function, u32 index)
2653 if (e->function != function)
2655 if ((e->flags & KVM_CPUID_FLAG_SIGNIFCANT_INDEX) && e->index != index)
2657 if ((e->flags & KVM_CPUID_FLAG_STATEFUL_FUNC) &&
2658 !(e->flags & KVM_CPUID_FLAG_STATE_READ_NEXT))
2663 void kvm_emulate_cpuid(struct kvm_vcpu *vcpu)
2666 u32 function, index;
2667 struct kvm_cpuid_entry2 *e, *best;
2669 kvm_x86_ops->cache_regs(vcpu);
2670 function = vcpu->arch.regs[VCPU_REGS_RAX];
2671 index = vcpu->arch.regs[VCPU_REGS_RCX];
2672 vcpu->arch.regs[VCPU_REGS_RAX] = 0;
2673 vcpu->arch.regs[VCPU_REGS_RBX] = 0;
2674 vcpu->arch.regs[VCPU_REGS_RCX] = 0;
2675 vcpu->arch.regs[VCPU_REGS_RDX] = 0;
2677 for (i = 0; i < vcpu->arch.cpuid_nent; ++i) {
2678 e = &vcpu->arch.cpuid_entries[i];
2679 if (is_matching_cpuid_entry(e, function, index)) {
2680 if (e->flags & KVM_CPUID_FLAG_STATEFUL_FUNC)
2681 move_to_next_stateful_cpuid_entry(vcpu, i);
2686 * Both basic or both extended?
2688 if (((e->function ^ function) & 0x80000000) == 0)
2689 if (!best || e->function > best->function)
2693 vcpu->arch.regs[VCPU_REGS_RAX] = best->eax;
2694 vcpu->arch.regs[VCPU_REGS_RBX] = best->ebx;
2695 vcpu->arch.regs[VCPU_REGS_RCX] = best->ecx;
2696 vcpu->arch.regs[VCPU_REGS_RDX] = best->edx;
2698 kvm_x86_ops->decache_regs(vcpu);
2699 kvm_x86_ops->skip_emulated_instruction(vcpu);
2700 KVMTRACE_5D(CPUID, vcpu, function,
2701 (u32)vcpu->arch.regs[VCPU_REGS_RAX],
2702 (u32)vcpu->arch.regs[VCPU_REGS_RBX],
2703 (u32)vcpu->arch.regs[VCPU_REGS_RCX],
2704 (u32)vcpu->arch.regs[VCPU_REGS_RDX], handler);
2706 EXPORT_SYMBOL_GPL(kvm_emulate_cpuid);
2709 * Check if userspace requested an interrupt window, and that the
2710 * interrupt window is open.
2712 * No need to exit to userspace if we already have an interrupt queued.
2714 static int dm_request_for_irq_injection(struct kvm_vcpu *vcpu,
2715 struct kvm_run *kvm_run)
2717 return (!vcpu->arch.irq_summary &&
2718 kvm_run->request_interrupt_window &&
2719 vcpu->arch.interrupt_window_open &&
2720 (kvm_x86_ops->get_rflags(vcpu) & X86_EFLAGS_IF));
2723 static void post_kvm_run_save(struct kvm_vcpu *vcpu,
2724 struct kvm_run *kvm_run)
2726 kvm_run->if_flag = (kvm_x86_ops->get_rflags(vcpu) & X86_EFLAGS_IF) != 0;
2727 kvm_run->cr8 = kvm_get_cr8(vcpu);
2728 kvm_run->apic_base = kvm_get_apic_base(vcpu);
2729 if (irqchip_in_kernel(vcpu->kvm))
2730 kvm_run->ready_for_interrupt_injection = 1;
2732 kvm_run->ready_for_interrupt_injection =
2733 (vcpu->arch.interrupt_window_open &&
2734 vcpu->arch.irq_summary == 0);
2737 static void vapic_enter(struct kvm_vcpu *vcpu)
2739 struct kvm_lapic *apic = vcpu->arch.apic;
2742 if (!apic || !apic->vapic_addr)
2745 down_read(¤t->mm->mmap_sem);
2746 page = gfn_to_page(vcpu->kvm, apic->vapic_addr >> PAGE_SHIFT);
2747 up_read(¤t->mm->mmap_sem);
2749 vcpu->arch.apic->vapic_page = page;
2752 static void vapic_exit(struct kvm_vcpu *vcpu)
2754 struct kvm_lapic *apic = vcpu->arch.apic;
2756 if (!apic || !apic->vapic_addr)
2759 kvm_release_page_dirty(apic->vapic_page);
2760 mark_page_dirty(vcpu->kvm, apic->vapic_addr >> PAGE_SHIFT);
2763 static int __vcpu_run(struct kvm_vcpu *vcpu, struct kvm_run *kvm_run)
2767 if (unlikely(vcpu->arch.mp_state == KVM_MP_STATE_SIPI_RECEIVED)) {
2768 pr_debug("vcpu %d received sipi with vector # %x\n",
2769 vcpu->vcpu_id, vcpu->arch.sipi_vector);
2770 kvm_lapic_reset(vcpu);
2771 r = kvm_x86_ops->vcpu_reset(vcpu);
2774 vcpu->arch.mp_state = KVM_MP_STATE_RUNNABLE;
2777 down_read(&vcpu->kvm->slots_lock);
2781 if (vcpu->guest_debug.enabled)
2782 kvm_x86_ops->guest_debug_pre(vcpu);
2786 if (test_and_clear_bit(KVM_REQ_MMU_RELOAD, &vcpu->requests))
2787 kvm_mmu_unload(vcpu);
2789 r = kvm_mmu_reload(vcpu);
2793 if (vcpu->requests) {
2794 if (test_and_clear_bit(KVM_REQ_MIGRATE_TIMER, &vcpu->requests))
2795 __kvm_migrate_timers(vcpu);
2796 if (test_and_clear_bit(KVM_REQ_TLB_FLUSH, &vcpu->requests))
2797 kvm_x86_ops->tlb_flush(vcpu);
2798 if (test_and_clear_bit(KVM_REQ_REPORT_TPR_ACCESS,
2800 kvm_run->exit_reason = KVM_EXIT_TPR_ACCESS;
2804 if (test_and_clear_bit(KVM_REQ_TRIPLE_FAULT, &vcpu->requests)) {
2805 kvm_run->exit_reason = KVM_EXIT_SHUTDOWN;
2811 clear_bit(KVM_REQ_PENDING_TIMER, &vcpu->requests);
2812 kvm_inject_pending_timer_irqs(vcpu);
2816 kvm_x86_ops->prepare_guest_switch(vcpu);
2817 kvm_load_guest_fpu(vcpu);
2819 local_irq_disable();
2821 if (vcpu->requests || need_resched()) {
2828 if (signal_pending(current)) {
2832 kvm_run->exit_reason = KVM_EXIT_INTR;
2833 ++vcpu->stat.signal_exits;
2837 vcpu->guest_mode = 1;
2839 * Make sure that guest_mode assignment won't happen after
2840 * testing the pending IRQ vector bitmap.
2844 if (vcpu->arch.exception.pending)
2845 __queue_exception(vcpu);
2846 else if (irqchip_in_kernel(vcpu->kvm))
2847 kvm_x86_ops->inject_pending_irq(vcpu);
2849 kvm_x86_ops->inject_pending_vectors(vcpu, kvm_run);
2851 kvm_lapic_sync_to_vapic(vcpu);
2853 up_read(&vcpu->kvm->slots_lock);
2858 KVMTRACE_0D(VMENTRY, vcpu, entryexit);
2859 kvm_x86_ops->run(vcpu, kvm_run);
2861 vcpu->guest_mode = 0;
2867 * We must have an instruction between local_irq_enable() and
2868 * kvm_guest_exit(), so the timer interrupt isn't delayed by
2869 * the interrupt shadow. The stat.exits increment will do nicely.
2870 * But we need to prevent reordering, hence this barrier():
2878 down_read(&vcpu->kvm->slots_lock);
2881 * Profile KVM exit RIPs:
2883 if (unlikely(prof_on == KVM_PROFILING)) {
2884 kvm_x86_ops->cache_regs(vcpu);
2885 profile_hit(KVM_PROFILING, (void *)vcpu->arch.rip);
2888 if (vcpu->arch.exception.pending && kvm_x86_ops->exception_injected(vcpu))
2889 vcpu->arch.exception.pending = false;
2891 kvm_lapic_sync_from_vapic(vcpu);
2893 r = kvm_x86_ops->handle_exit(kvm_run, vcpu);
2896 if (dm_request_for_irq_injection(vcpu, kvm_run)) {
2898 kvm_run->exit_reason = KVM_EXIT_INTR;
2899 ++vcpu->stat.request_irq_exits;
2902 if (!need_resched())
2907 up_read(&vcpu->kvm->slots_lock);
2910 down_read(&vcpu->kvm->slots_lock);
2914 post_kvm_run_save(vcpu, kvm_run);
2916 down_read(&vcpu->kvm->slots_lock);
2918 up_read(&vcpu->kvm->slots_lock);
2923 int kvm_arch_vcpu_ioctl_run(struct kvm_vcpu *vcpu, struct kvm_run *kvm_run)
2930 if (unlikely(vcpu->arch.mp_state == KVM_MP_STATE_UNINITIALIZED)) {
2931 kvm_vcpu_block(vcpu);
2936 if (vcpu->sigset_active)
2937 sigprocmask(SIG_SETMASK, &vcpu->sigset, &sigsaved);
2939 /* re-sync apic's tpr */
2940 if (!irqchip_in_kernel(vcpu->kvm))
2941 kvm_set_cr8(vcpu, kvm_run->cr8);
2943 if (vcpu->arch.pio.cur_count) {
2944 r = complete_pio(vcpu);
2948 #if CONFIG_HAS_IOMEM
2949 if (vcpu->mmio_needed) {
2950 memcpy(vcpu->mmio_data, kvm_run->mmio.data, 8);
2951 vcpu->mmio_read_completed = 1;
2952 vcpu->mmio_needed = 0;
2954 down_read(&vcpu->kvm->slots_lock);
2955 r = emulate_instruction(vcpu, kvm_run,
2956 vcpu->arch.mmio_fault_cr2, 0,
2957 EMULTYPE_NO_DECODE);
2958 up_read(&vcpu->kvm->slots_lock);
2959 if (r == EMULATE_DO_MMIO) {
2961 * Read-modify-write. Back to userspace.
2968 if (kvm_run->exit_reason == KVM_EXIT_HYPERCALL) {
2969 kvm_x86_ops->cache_regs(vcpu);
2970 vcpu->arch.regs[VCPU_REGS_RAX] = kvm_run->hypercall.ret;
2971 kvm_x86_ops->decache_regs(vcpu);
2974 r = __vcpu_run(vcpu, kvm_run);
2977 if (vcpu->sigset_active)
2978 sigprocmask(SIG_SETMASK, &sigsaved, NULL);
2984 int kvm_arch_vcpu_ioctl_get_regs(struct kvm_vcpu *vcpu, struct kvm_regs *regs)
2988 kvm_x86_ops->cache_regs(vcpu);
2990 regs->rax = vcpu->arch.regs[VCPU_REGS_RAX];
2991 regs->rbx = vcpu->arch.regs[VCPU_REGS_RBX];
2992 regs->rcx = vcpu->arch.regs[VCPU_REGS_RCX];
2993 regs->rdx = vcpu->arch.regs[VCPU_REGS_RDX];
2994 regs->rsi = vcpu->arch.regs[VCPU_REGS_RSI];
2995 regs->rdi = vcpu->arch.regs[VCPU_REGS_RDI];
2996 regs->rsp = vcpu->arch.regs[VCPU_REGS_RSP];
2997 regs->rbp = vcpu->arch.regs[VCPU_REGS_RBP];
2998 #ifdef CONFIG_X86_64
2999 regs->r8 = vcpu->arch.regs[VCPU_REGS_R8];
3000 regs->r9 = vcpu->arch.regs[VCPU_REGS_R9];
3001 regs->r10 = vcpu->arch.regs[VCPU_REGS_R10];
3002 regs->r11 = vcpu->arch.regs[VCPU_REGS_R11];
3003 regs->r12 = vcpu->arch.regs[VCPU_REGS_R12];
3004 regs->r13 = vcpu->arch.regs[VCPU_REGS_R13];
3005 regs->r14 = vcpu->arch.regs[VCPU_REGS_R14];
3006 regs->r15 = vcpu->arch.regs[VCPU_REGS_R15];
3009 regs->rip = vcpu->arch.rip;
3010 regs->rflags = kvm_x86_ops->get_rflags(vcpu);
3013 * Don't leak debug flags in case they were set for guest debugging
3015 if (vcpu->guest_debug.enabled && vcpu->guest_debug.singlestep)
3016 regs->rflags &= ~(X86_EFLAGS_TF | X86_EFLAGS_RF);
3023 int kvm_arch_vcpu_ioctl_set_regs(struct kvm_vcpu *vcpu, struct kvm_regs *regs)
3027 vcpu->arch.regs[VCPU_REGS_RAX] = regs->rax;
3028 vcpu->arch.regs[VCPU_REGS_RBX] = regs->rbx;
3029 vcpu->arch.regs[VCPU_REGS_RCX] = regs->rcx;
3030 vcpu->arch.regs[VCPU_REGS_RDX] = regs->rdx;
3031 vcpu->arch.regs[VCPU_REGS_RSI] = regs->rsi;
3032 vcpu->arch.regs[VCPU_REGS_RDI] = regs->rdi;
3033 vcpu->arch.regs[VCPU_REGS_RSP] = regs->rsp;
3034 vcpu->arch.regs[VCPU_REGS_RBP] = regs->rbp;
3035 #ifdef CONFIG_X86_64
3036 vcpu->arch.regs[VCPU_REGS_R8] = regs->r8;
3037 vcpu->arch.regs[VCPU_REGS_R9] = regs->r9;
3038 vcpu->arch.regs[VCPU_REGS_R10] = regs->r10;
3039 vcpu->arch.regs[VCPU_REGS_R11] = regs->r11;
3040 vcpu->arch.regs[VCPU_REGS_R12] = regs->r12;
3041 vcpu->arch.regs[VCPU_REGS_R13] = regs->r13;
3042 vcpu->arch.regs[VCPU_REGS_R14] = regs->r14;
3043 vcpu->arch.regs[VCPU_REGS_R15] = regs->r15;
3046 vcpu->arch.rip = regs->rip;
3047 kvm_x86_ops->set_rflags(vcpu, regs->rflags);
3049 kvm_x86_ops->decache_regs(vcpu);
3051 vcpu->arch.exception.pending = false;
3058 static void get_segment(struct kvm_vcpu *vcpu,
3059 struct kvm_segment *var, int seg)
3061 kvm_x86_ops->get_segment(vcpu, var, seg);
3064 void kvm_get_cs_db_l_bits(struct kvm_vcpu *vcpu, int *db, int *l)
3066 struct kvm_segment cs;
3068 get_segment(vcpu, &cs, VCPU_SREG_CS);
3072 EXPORT_SYMBOL_GPL(kvm_get_cs_db_l_bits);
3074 int kvm_arch_vcpu_ioctl_get_sregs(struct kvm_vcpu *vcpu,
3075 struct kvm_sregs *sregs)
3077 struct descriptor_table dt;
3082 get_segment(vcpu, &sregs->cs, VCPU_SREG_CS);
3083 get_segment(vcpu, &sregs->ds, VCPU_SREG_DS);
3084 get_segment(vcpu, &sregs->es, VCPU_SREG_ES);
3085 get_segment(vcpu, &sregs->fs, VCPU_SREG_FS);
3086 get_segment(vcpu, &sregs->gs, VCPU_SREG_GS);
3087 get_segment(vcpu, &sregs->ss, VCPU_SREG_SS);
3089 get_segment(vcpu, &sregs->tr, VCPU_SREG_TR);
3090 get_segment(vcpu, &sregs->ldt, VCPU_SREG_LDTR);
3092 kvm_x86_ops->get_idt(vcpu, &dt);
3093 sregs->idt.limit = dt.limit;
3094 sregs->idt.base = dt.base;
3095 kvm_x86_ops->get_gdt(vcpu, &dt);
3096 sregs->gdt.limit = dt.limit;
3097 sregs->gdt.base = dt.base;
3099 kvm_x86_ops->decache_cr4_guest_bits(vcpu);
3100 sregs->cr0 = vcpu->arch.cr0;
3101 sregs->cr2 = vcpu->arch.cr2;
3102 sregs->cr3 = vcpu->arch.cr3;
3103 sregs->cr4 = vcpu->arch.cr4;
3104 sregs->cr8 = kvm_get_cr8(vcpu);
3105 sregs->efer = vcpu->arch.shadow_efer;
3106 sregs->apic_base = kvm_get_apic_base(vcpu);
3108 if (irqchip_in_kernel(vcpu->kvm)) {
3109 memset(sregs->interrupt_bitmap, 0,
3110 sizeof sregs->interrupt_bitmap);
3111 pending_vec = kvm_x86_ops->get_irq(vcpu);
3112 if (pending_vec >= 0)
3113 set_bit(pending_vec,
3114 (unsigned long *)sregs->interrupt_bitmap);
3116 memcpy(sregs->interrupt_bitmap, vcpu->arch.irq_pending,
3117 sizeof sregs->interrupt_bitmap);
3124 int kvm_arch_vcpu_ioctl_get_mpstate(struct kvm_vcpu *vcpu,
3125 struct kvm_mp_state *mp_state)
3128 mp_state->mp_state = vcpu->arch.mp_state;
3133 int kvm_arch_vcpu_ioctl_set_mpstate(struct kvm_vcpu *vcpu,
3134 struct kvm_mp_state *mp_state)
3137 vcpu->arch.mp_state = mp_state->mp_state;
3142 static void set_segment(struct kvm_vcpu *vcpu,
3143 struct kvm_segment *var, int seg)
3145 kvm_x86_ops->set_segment(vcpu, var, seg);
3148 static void seg_desct_to_kvm_desct(struct desc_struct *seg_desc, u16 selector,
3149 struct kvm_segment *kvm_desct)
3151 kvm_desct->base = seg_desc->base0;
3152 kvm_desct->base |= seg_desc->base1 << 16;
3153 kvm_desct->base |= seg_desc->base2 << 24;
3154 kvm_desct->limit = seg_desc->limit0;
3155 kvm_desct->limit |= seg_desc->limit << 16;
3156 kvm_desct->selector = selector;
3157 kvm_desct->type = seg_desc->type;
3158 kvm_desct->present = seg_desc->p;
3159 kvm_desct->dpl = seg_desc->dpl;
3160 kvm_desct->db = seg_desc->d;
3161 kvm_desct->s = seg_desc->s;
3162 kvm_desct->l = seg_desc->l;
3163 kvm_desct->g = seg_desc->g;
3164 kvm_desct->avl = seg_desc->avl;
3166 kvm_desct->unusable = 1;
3168 kvm_desct->unusable = 0;
3169 kvm_desct->padding = 0;
3172 static void get_segment_descritptor_dtable(struct kvm_vcpu *vcpu,
3174 struct descriptor_table *dtable)
3176 if (selector & 1 << 2) {
3177 struct kvm_segment kvm_seg;
3179 get_segment(vcpu, &kvm_seg, VCPU_SREG_LDTR);
3181 if (kvm_seg.unusable)
3184 dtable->limit = kvm_seg.limit;
3185 dtable->base = kvm_seg.base;
3188 kvm_x86_ops->get_gdt(vcpu, dtable);
3191 /* allowed just for 8 bytes segments */
3192 static int load_guest_segment_descriptor(struct kvm_vcpu *vcpu, u16 selector,
3193 struct desc_struct *seg_desc)
3195 struct descriptor_table dtable;
3196 u16 index = selector >> 3;
3198 get_segment_descritptor_dtable(vcpu, selector, &dtable);
3200 if (dtable.limit < index * 8 + 7) {
3201 kvm_queue_exception_e(vcpu, GP_VECTOR, selector & 0xfffc);
3204 return kvm_read_guest(vcpu->kvm, dtable.base + index * 8, seg_desc, 8);
3207 /* allowed just for 8 bytes segments */
3208 static int save_guest_segment_descriptor(struct kvm_vcpu *vcpu, u16 selector,
3209 struct desc_struct *seg_desc)
3211 struct descriptor_table dtable;
3212 u16 index = selector >> 3;
3214 get_segment_descritptor_dtable(vcpu, selector, &dtable);
3216 if (dtable.limit < index * 8 + 7)
3218 return kvm_write_guest(vcpu->kvm, dtable.base + index * 8, seg_desc, 8);
3221 static u32 get_tss_base_addr(struct kvm_vcpu *vcpu,
3222 struct desc_struct *seg_desc)
3226 base_addr = seg_desc->base0;
3227 base_addr |= (seg_desc->base1 << 16);
3228 base_addr |= (seg_desc->base2 << 24);
3233 static int load_tss_segment32(struct kvm_vcpu *vcpu,
3234 struct desc_struct *seg_desc,
3235 struct tss_segment_32 *tss)
3239 base_addr = get_tss_base_addr(vcpu, seg_desc);
3241 return kvm_read_guest(vcpu->kvm, base_addr, tss,
3242 sizeof(struct tss_segment_32));
3245 static int save_tss_segment32(struct kvm_vcpu *vcpu,
3246 struct desc_struct *seg_desc,
3247 struct tss_segment_32 *tss)
3251 base_addr = get_tss_base_addr(vcpu, seg_desc);
3253 return kvm_write_guest(vcpu->kvm, base_addr, tss,
3254 sizeof(struct tss_segment_32));
3257 static int load_tss_segment16(struct kvm_vcpu *vcpu,
3258 struct desc_struct *seg_desc,
3259 struct tss_segment_16 *tss)
3263 base_addr = get_tss_base_addr(vcpu, seg_desc);
3265 return kvm_read_guest(vcpu->kvm, base_addr, tss,
3266 sizeof(struct tss_segment_16));
3269 static int save_tss_segment16(struct kvm_vcpu *vcpu,
3270 struct desc_struct *seg_desc,
3271 struct tss_segment_16 *tss)
3275 base_addr = get_tss_base_addr(vcpu, seg_desc);
3277 return kvm_write_guest(vcpu->kvm, base_addr, tss,
3278 sizeof(struct tss_segment_16));
3281 static u16 get_segment_selector(struct kvm_vcpu *vcpu, int seg)
3283 struct kvm_segment kvm_seg;
3285 get_segment(vcpu, &kvm_seg, seg);
3286 return kvm_seg.selector;
3289 static int load_segment_descriptor_to_kvm_desct(struct kvm_vcpu *vcpu,
3291 struct kvm_segment *kvm_seg)
3293 struct desc_struct seg_desc;
3295 if (load_guest_segment_descriptor(vcpu, selector, &seg_desc))
3297 seg_desct_to_kvm_desct(&seg_desc, selector, kvm_seg);
3301 static int load_segment_descriptor(struct kvm_vcpu *vcpu, u16 selector,
3302 int type_bits, int seg)
3304 struct kvm_segment kvm_seg;
3306 if (load_segment_descriptor_to_kvm_desct(vcpu, selector, &kvm_seg))
3308 kvm_seg.type |= type_bits;
3310 if (seg != VCPU_SREG_SS && seg != VCPU_SREG_CS &&
3311 seg != VCPU_SREG_LDTR)
3313 kvm_seg.unusable = 1;
3315 set_segment(vcpu, &kvm_seg, seg);
3319 static void save_state_to_tss32(struct kvm_vcpu *vcpu,
3320 struct tss_segment_32 *tss)
3322 tss->cr3 = vcpu->arch.cr3;
3323 tss->eip = vcpu->arch.rip;
3324 tss->eflags = kvm_x86_ops->get_rflags(vcpu);
3325 tss->eax = vcpu->arch.regs[VCPU_REGS_RAX];
3326 tss->ecx = vcpu->arch.regs[VCPU_REGS_RCX];
3327 tss->edx = vcpu->arch.regs[VCPU_REGS_RDX];
3328 tss->ebx = vcpu->arch.regs[VCPU_REGS_RBX];
3329 tss->esp = vcpu->arch.regs[VCPU_REGS_RSP];
3330 tss->ebp = vcpu->arch.regs[VCPU_REGS_RBP];
3331 tss->esi = vcpu->arch.regs[VCPU_REGS_RSI];
3332 tss->edi = vcpu->arch.regs[VCPU_REGS_RDI];
3334 tss->es = get_segment_selector(vcpu, VCPU_SREG_ES);
3335 tss->cs = get_segment_selector(vcpu, VCPU_SREG_CS);
3336 tss->ss = get_segment_selector(vcpu, VCPU_SREG_SS);
3337 tss->ds = get_segment_selector(vcpu, VCPU_SREG_DS);
3338 tss->fs = get_segment_selector(vcpu, VCPU_SREG_FS);
3339 tss->gs = get_segment_selector(vcpu, VCPU_SREG_GS);
3340 tss->ldt_selector = get_segment_selector(vcpu, VCPU_SREG_LDTR);
3341 tss->prev_task_link = get_segment_selector(vcpu, VCPU_SREG_TR);
3344 static int load_state_from_tss32(struct kvm_vcpu *vcpu,
3345 struct tss_segment_32 *tss)
3347 kvm_set_cr3(vcpu, tss->cr3);
3349 vcpu->arch.rip = tss->eip;
3350 kvm_x86_ops->set_rflags(vcpu, tss->eflags | 2);
3352 vcpu->arch.regs[VCPU_REGS_RAX] = tss->eax;
3353 vcpu->arch.regs[VCPU_REGS_RCX] = tss->ecx;
3354 vcpu->arch.regs[VCPU_REGS_RDX] = tss->edx;
3355 vcpu->arch.regs[VCPU_REGS_RBX] = tss->ebx;
3356 vcpu->arch.regs[VCPU_REGS_RSP] = tss->esp;
3357 vcpu->arch.regs[VCPU_REGS_RBP] = tss->ebp;
3358 vcpu->arch.regs[VCPU_REGS_RSI] = tss->esi;
3359 vcpu->arch.regs[VCPU_REGS_RDI] = tss->edi;
3361 if (load_segment_descriptor(vcpu, tss->ldt_selector, 0, VCPU_SREG_LDTR))
3364 if (load_segment_descriptor(vcpu, tss->es, 1, VCPU_SREG_ES))
3367 if (load_segment_descriptor(vcpu, tss->cs, 9, VCPU_SREG_CS))
3370 if (load_segment_descriptor(vcpu, tss->ss, 1, VCPU_SREG_SS))
3373 if (load_segment_descriptor(vcpu, tss->ds, 1, VCPU_SREG_DS))
3376 if (load_segment_descriptor(vcpu, tss->fs, 1, VCPU_SREG_FS))
3379 if (load_segment_descriptor(vcpu, tss->gs, 1, VCPU_SREG_GS))
3384 static void save_state_to_tss16(struct kvm_vcpu *vcpu,
3385 struct tss_segment_16 *tss)
3387 tss->ip = vcpu->arch.rip;
3388 tss->flag = kvm_x86_ops->get_rflags(vcpu);
3389 tss->ax = vcpu->arch.regs[VCPU_REGS_RAX];
3390 tss->cx = vcpu->arch.regs[VCPU_REGS_RCX];
3391 tss->dx = vcpu->arch.regs[VCPU_REGS_RDX];
3392 tss->bx = vcpu->arch.regs[VCPU_REGS_RBX];
3393 tss->sp = vcpu->arch.regs[VCPU_REGS_RSP];
3394 tss->bp = vcpu->arch.regs[VCPU_REGS_RBP];
3395 tss->si = vcpu->arch.regs[VCPU_REGS_RSI];
3396 tss->di = vcpu->arch.regs[VCPU_REGS_RDI];
3398 tss->es = get_segment_selector(vcpu, VCPU_SREG_ES);
3399 tss->cs = get_segment_selector(vcpu, VCPU_SREG_CS);
3400 tss->ss = get_segment_selector(vcpu, VCPU_SREG_SS);
3401 tss->ds = get_segment_selector(vcpu, VCPU_SREG_DS);
3402 tss->ldt = get_segment_selector(vcpu, VCPU_SREG_LDTR);
3403 tss->prev_task_link = get_segment_selector(vcpu, VCPU_SREG_TR);
3406 static int load_state_from_tss16(struct kvm_vcpu *vcpu,
3407 struct tss_segment_16 *tss)
3409 vcpu->arch.rip = tss->ip;
3410 kvm_x86_ops->set_rflags(vcpu, tss->flag | 2);
3411 vcpu->arch.regs[VCPU_REGS_RAX] = tss->ax;
3412 vcpu->arch.regs[VCPU_REGS_RCX] = tss->cx;
3413 vcpu->arch.regs[VCPU_REGS_RDX] = tss->dx;
3414 vcpu->arch.regs[VCPU_REGS_RBX] = tss->bx;
3415 vcpu->arch.regs[VCPU_REGS_RSP] = tss->sp;
3416 vcpu->arch.regs[VCPU_REGS_RBP] = tss->bp;
3417 vcpu->arch.regs[VCPU_REGS_RSI] = tss->si;
3418 vcpu->arch.regs[VCPU_REGS_RDI] = tss->di;
3420 if (load_segment_descriptor(vcpu, tss->ldt, 0, VCPU_SREG_LDTR))
3423 if (load_segment_descriptor(vcpu, tss->es, 1, VCPU_SREG_ES))
3426 if (load_segment_descriptor(vcpu, tss->cs, 9, VCPU_SREG_CS))
3429 if (load_segment_descriptor(vcpu, tss->ss, 1, VCPU_SREG_SS))
3432 if (load_segment_descriptor(vcpu, tss->ds, 1, VCPU_SREG_DS))
3437 static int kvm_task_switch_16(struct kvm_vcpu *vcpu, u16 tss_selector,
3438 struct desc_struct *cseg_desc,
3439 struct desc_struct *nseg_desc)
3441 struct tss_segment_16 tss_segment_16;
3444 if (load_tss_segment16(vcpu, cseg_desc, &tss_segment_16))
3447 save_state_to_tss16(vcpu, &tss_segment_16);
3448 save_tss_segment16(vcpu, cseg_desc, &tss_segment_16);
3450 if (load_tss_segment16(vcpu, nseg_desc, &tss_segment_16))
3452 if (load_state_from_tss16(vcpu, &tss_segment_16))
3460 static int kvm_task_switch_32(struct kvm_vcpu *vcpu, u16 tss_selector,
3461 struct desc_struct *cseg_desc,
3462 struct desc_struct *nseg_desc)
3464 struct tss_segment_32 tss_segment_32;
3467 if (load_tss_segment32(vcpu, cseg_desc, &tss_segment_32))
3470 save_state_to_tss32(vcpu, &tss_segment_32);
3471 save_tss_segment32(vcpu, cseg_desc, &tss_segment_32);
3473 if (load_tss_segment32(vcpu, nseg_desc, &tss_segment_32))
3475 if (load_state_from_tss32(vcpu, &tss_segment_32))
3483 int kvm_task_switch(struct kvm_vcpu *vcpu, u16 tss_selector, int reason)
3485 struct kvm_segment tr_seg;
3486 struct desc_struct cseg_desc;
3487 struct desc_struct nseg_desc;
3490 get_segment(vcpu, &tr_seg, VCPU_SREG_TR);
3492 if (load_guest_segment_descriptor(vcpu, tss_selector, &nseg_desc))
3495 if (load_guest_segment_descriptor(vcpu, tr_seg.selector, &cseg_desc))
3499 if (reason != TASK_SWITCH_IRET) {
3502 cpl = kvm_x86_ops->get_cpl(vcpu);
3503 if ((tss_selector & 3) > nseg_desc.dpl || cpl > nseg_desc.dpl) {
3504 kvm_queue_exception_e(vcpu, GP_VECTOR, 0);
3509 if (!nseg_desc.p || (nseg_desc.limit0 | nseg_desc.limit << 16) < 0x67) {
3510 kvm_queue_exception_e(vcpu, TS_VECTOR, tss_selector & 0xfffc);
3514 if (reason == TASK_SWITCH_IRET || reason == TASK_SWITCH_JMP) {
3515 cseg_desc.type &= ~(1 << 1); //clear the B flag
3516 save_guest_segment_descriptor(vcpu, tr_seg.selector,
3520 if (reason == TASK_SWITCH_IRET) {
3521 u32 eflags = kvm_x86_ops->get_rflags(vcpu);
3522 kvm_x86_ops->set_rflags(vcpu, eflags & ~X86_EFLAGS_NT);
3525 kvm_x86_ops->skip_emulated_instruction(vcpu);
3526 kvm_x86_ops->cache_regs(vcpu);
3528 if (nseg_desc.type & 8)
3529 ret = kvm_task_switch_32(vcpu, tss_selector, &cseg_desc,
3532 ret = kvm_task_switch_16(vcpu, tss_selector, &cseg_desc,
3535 if (reason == TASK_SWITCH_CALL || reason == TASK_SWITCH_GATE) {
3536 u32 eflags = kvm_x86_ops->get_rflags(vcpu);
3537 kvm_x86_ops->set_rflags(vcpu, eflags | X86_EFLAGS_NT);
3540 if (reason != TASK_SWITCH_IRET) {
3541 nseg_desc.type |= (1 << 1);
3542 save_guest_segment_descriptor(vcpu, tss_selector,
3546 kvm_x86_ops->set_cr0(vcpu, vcpu->arch.cr0 | X86_CR0_TS);
3547 seg_desct_to_kvm_desct(&nseg_desc, tss_selector, &tr_seg);
3549 set_segment(vcpu, &tr_seg, VCPU_SREG_TR);
3551 kvm_x86_ops->decache_regs(vcpu);
3554 EXPORT_SYMBOL_GPL(kvm_task_switch);
3556 int kvm_arch_vcpu_ioctl_set_sregs(struct kvm_vcpu *vcpu,
3557 struct kvm_sregs *sregs)
3559 int mmu_reset_needed = 0;
3560 int i, pending_vec, max_bits;
3561 struct descriptor_table dt;
3565 dt.limit = sregs->idt.limit;
3566 dt.base = sregs->idt.base;
3567 kvm_x86_ops->set_idt(vcpu, &dt);
3568 dt.limit = sregs->gdt.limit;
3569 dt.base = sregs->gdt.base;
3570 kvm_x86_ops->set_gdt(vcpu, &dt);
3572 vcpu->arch.cr2 = sregs->cr2;
3573 mmu_reset_needed |= vcpu->arch.cr3 != sregs->cr3;
3574 vcpu->arch.cr3 = sregs->cr3;
3576 kvm_set_cr8(vcpu, sregs->cr8);
3578 mmu_reset_needed |= vcpu->arch.shadow_efer != sregs->efer;
3579 kvm_x86_ops->set_efer(vcpu, sregs->efer);
3580 kvm_set_apic_base(vcpu, sregs->apic_base);
3582 kvm_x86_ops->decache_cr4_guest_bits(vcpu);
3584 mmu_reset_needed |= vcpu->arch.cr0 != sregs->cr0;
3585 kvm_x86_ops->set_cr0(vcpu, sregs->cr0);
3586 vcpu->arch.cr0 = sregs->cr0;
3588 mmu_reset_needed |= vcpu->arch.cr4 != sregs->cr4;
3589 kvm_x86_ops->set_cr4(vcpu, sregs->cr4);
3590 if (!is_long_mode(vcpu) && is_pae(vcpu))
3591 load_pdptrs(vcpu, vcpu->arch.cr3);
3593 if (mmu_reset_needed)
3594 kvm_mmu_reset_context(vcpu);
3596 if (!irqchip_in_kernel(vcpu->kvm)) {
3597 memcpy(vcpu->arch.irq_pending, sregs->interrupt_bitmap,
3598 sizeof vcpu->arch.irq_pending);
3599 vcpu->arch.irq_summary = 0;
3600 for (i = 0; i < ARRAY_SIZE(vcpu->arch.irq_pending); ++i)
3601 if (vcpu->arch.irq_pending[i])
3602 __set_bit(i, &vcpu->arch.irq_summary);
3604 max_bits = (sizeof sregs->interrupt_bitmap) << 3;
3605 pending_vec = find_first_bit(
3606 (const unsigned long *)sregs->interrupt_bitmap,
3608 /* Only pending external irq is handled here */
3609 if (pending_vec < max_bits) {
3610 kvm_x86_ops->set_irq(vcpu, pending_vec);
3611 pr_debug("Set back pending irq %d\n",
3616 set_segment(vcpu, &sregs->cs, VCPU_SREG_CS);
3617 set_segment(vcpu, &sregs->ds, VCPU_SREG_DS);
3618 set_segment(vcpu, &sregs->es, VCPU_SREG_ES);
3619 set_segment(vcpu, &sregs->fs, VCPU_SREG_FS);
3620 set_segment(vcpu, &sregs->gs, VCPU_SREG_GS);
3621 set_segment(vcpu, &sregs->ss, VCPU_SREG_SS);
3623 set_segment(vcpu, &sregs->tr, VCPU_SREG_TR);
3624 set_segment(vcpu, &sregs->ldt, VCPU_SREG_LDTR);
3631 int kvm_arch_vcpu_ioctl_debug_guest(struct kvm_vcpu *vcpu,
3632 struct kvm_debug_guest *dbg)
3638 r = kvm_x86_ops->set_guest_debug(vcpu, dbg);
3646 * fxsave fpu state. Taken from x86_64/processor.h. To be killed when
3647 * we have asm/x86/processor.h
3658 u32 st_space[32]; /* 8*16 bytes for each FP-reg = 128 bytes */
3659 #ifdef CONFIG_X86_64
3660 u32 xmm_space[64]; /* 16*16 bytes for each XMM-reg = 256 bytes */
3662 u32 xmm_space[32]; /* 8*16 bytes for each XMM-reg = 128 bytes */
3667 * Translate a guest virtual address to a guest physical address.
3669 int kvm_arch_vcpu_ioctl_translate(struct kvm_vcpu *vcpu,
3670 struct kvm_translation *tr)
3672 unsigned long vaddr = tr->linear_address;
3676 down_read(&vcpu->kvm->slots_lock);
3677 gpa = vcpu->arch.mmu.gva_to_gpa(vcpu, vaddr);
3678 up_read(&vcpu->kvm->slots_lock);
3679 tr->physical_address = gpa;
3680 tr->valid = gpa != UNMAPPED_GVA;
3688 int kvm_arch_vcpu_ioctl_get_fpu(struct kvm_vcpu *vcpu, struct kvm_fpu *fpu)
3690 struct fxsave *fxsave = (struct fxsave *)&vcpu->arch.guest_fx_image;
3694 memcpy(fpu->fpr, fxsave->st_space, 128);
3695 fpu->fcw = fxsave->cwd;
3696 fpu->fsw = fxsave->swd;
3697 fpu->ftwx = fxsave->twd;
3698 fpu->last_opcode = fxsave->fop;
3699 fpu->last_ip = fxsave->rip;
3700 fpu->last_dp = fxsave->rdp;
3701 memcpy(fpu->xmm, fxsave->xmm_space, sizeof fxsave->xmm_space);
3708 int kvm_arch_vcpu_ioctl_set_fpu(struct kvm_vcpu *vcpu, struct kvm_fpu *fpu)
3710 struct fxsave *fxsave = (struct fxsave *)&vcpu->arch.guest_fx_image;
3714 memcpy(fxsave->st_space, fpu->fpr, 128);
3715 fxsave->cwd = fpu->fcw;
3716 fxsave->swd = fpu->fsw;
3717 fxsave->twd = fpu->ftwx;
3718 fxsave->fop = fpu->last_opcode;
3719 fxsave->rip = fpu->last_ip;
3720 fxsave->rdp = fpu->last_dp;
3721 memcpy(fxsave->xmm_space, fpu->xmm, sizeof fxsave->xmm_space);
3728 void fx_init(struct kvm_vcpu *vcpu)
3730 unsigned after_mxcsr_mask;
3733 * Touch the fpu the first time in non atomic context as if
3734 * this is the first fpu instruction the exception handler
3735 * will fire before the instruction returns and it'll have to
3736 * allocate ram with GFP_KERNEL.
3739 fx_save(&vcpu->arch.host_fx_image);
3741 /* Initialize guest FPU by resetting ours and saving into guest's */
3743 fx_save(&vcpu->arch.host_fx_image);
3745 fx_save(&vcpu->arch.guest_fx_image);
3746 fx_restore(&vcpu->arch.host_fx_image);
3749 vcpu->arch.cr0 |= X86_CR0_ET;
3750 after_mxcsr_mask = offsetof(struct i387_fxsave_struct, st_space);
3751 vcpu->arch.guest_fx_image.mxcsr = 0x1f80;
3752 memset((void *)&vcpu->arch.guest_fx_image + after_mxcsr_mask,
3753 0, sizeof(struct i387_fxsave_struct) - after_mxcsr_mask);
3755 EXPORT_SYMBOL_GPL(fx_init);
3757 void kvm_load_guest_fpu(struct kvm_vcpu *vcpu)
3759 if (!vcpu->fpu_active || vcpu->guest_fpu_loaded)
3762 vcpu->guest_fpu_loaded = 1;
3763 fx_save(&vcpu->arch.host_fx_image);
3764 fx_restore(&vcpu->arch.guest_fx_image);
3766 EXPORT_SYMBOL_GPL(kvm_load_guest_fpu);
3768 void kvm_put_guest_fpu(struct kvm_vcpu *vcpu)
3770 if (!vcpu->guest_fpu_loaded)
3773 vcpu->guest_fpu_loaded = 0;
3774 fx_save(&vcpu->arch.guest_fx_image);
3775 fx_restore(&vcpu->arch.host_fx_image);
3776 ++vcpu->stat.fpu_reload;
3778 EXPORT_SYMBOL_GPL(kvm_put_guest_fpu);
3780 void kvm_arch_vcpu_free(struct kvm_vcpu *vcpu)
3782 kvm_x86_ops->vcpu_free(vcpu);
3785 struct kvm_vcpu *kvm_arch_vcpu_create(struct kvm *kvm,
3788 return kvm_x86_ops->vcpu_create(kvm, id);
3791 int kvm_arch_vcpu_setup(struct kvm_vcpu *vcpu)
3795 /* We do fxsave: this must be aligned. */
3796 BUG_ON((unsigned long)&vcpu->arch.host_fx_image & 0xF);
3799 r = kvm_arch_vcpu_reset(vcpu);
3801 r = kvm_mmu_setup(vcpu);
3808 kvm_x86_ops->vcpu_free(vcpu);
3812 void kvm_arch_vcpu_destroy(struct kvm_vcpu *vcpu)
3815 kvm_mmu_unload(vcpu);
3818 kvm_x86_ops->vcpu_free(vcpu);
3821 int kvm_arch_vcpu_reset(struct kvm_vcpu *vcpu)
3823 return kvm_x86_ops->vcpu_reset(vcpu);
3826 void kvm_arch_hardware_enable(void *garbage)
3828 kvm_x86_ops->hardware_enable(garbage);
3831 void kvm_arch_hardware_disable(void *garbage)
3833 kvm_x86_ops->hardware_disable(garbage);
3836 int kvm_arch_hardware_setup(void)
3838 return kvm_x86_ops->hardware_setup();
3841 void kvm_arch_hardware_unsetup(void)
3843 kvm_x86_ops->hardware_unsetup();
3846 void kvm_arch_check_processor_compat(void *rtn)
3848 kvm_x86_ops->check_processor_compatibility(rtn);
3851 int kvm_arch_vcpu_init(struct kvm_vcpu *vcpu)
3857 BUG_ON(vcpu->kvm == NULL);
3860 vcpu->arch.mmu.root_hpa = INVALID_PAGE;
3861 if (!irqchip_in_kernel(kvm) || vcpu->vcpu_id == 0)
3862 vcpu->arch.mp_state = KVM_MP_STATE_RUNNABLE;
3864 vcpu->arch.mp_state = KVM_MP_STATE_UNINITIALIZED;
3866 page = alloc_page(GFP_KERNEL | __GFP_ZERO);
3871 vcpu->arch.pio_data = page_address(page);
3873 r = kvm_mmu_create(vcpu);
3875 goto fail_free_pio_data;
3877 if (irqchip_in_kernel(kvm)) {
3878 r = kvm_create_lapic(vcpu);
3880 goto fail_mmu_destroy;
3886 kvm_mmu_destroy(vcpu);
3888 free_page((unsigned long)vcpu->arch.pio_data);
3893 void kvm_arch_vcpu_uninit(struct kvm_vcpu *vcpu)
3895 kvm_free_lapic(vcpu);
3896 down_read(&vcpu->kvm->slots_lock);
3897 kvm_mmu_destroy(vcpu);
3898 up_read(&vcpu->kvm->slots_lock);
3899 free_page((unsigned long)vcpu->arch.pio_data);
3902 struct kvm *kvm_arch_create_vm(void)
3904 struct kvm *kvm = kzalloc(sizeof(struct kvm), GFP_KERNEL);
3907 return ERR_PTR(-ENOMEM);
3909 INIT_LIST_HEAD(&kvm->arch.active_mmu_pages);
3914 static void kvm_unload_vcpu_mmu(struct kvm_vcpu *vcpu)
3917 kvm_mmu_unload(vcpu);
3921 static void kvm_free_vcpus(struct kvm *kvm)
3926 * Unpin any mmu pages first.
3928 for (i = 0; i < KVM_MAX_VCPUS; ++i)
3930 kvm_unload_vcpu_mmu(kvm->vcpus[i]);
3931 for (i = 0; i < KVM_MAX_VCPUS; ++i) {
3932 if (kvm->vcpus[i]) {
3933 kvm_arch_vcpu_free(kvm->vcpus[i]);
3934 kvm->vcpus[i] = NULL;
3940 void kvm_arch_destroy_vm(struct kvm *kvm)
3943 kfree(kvm->arch.vpic);
3944 kfree(kvm->arch.vioapic);
3945 kvm_free_vcpus(kvm);
3946 kvm_free_physmem(kvm);
3947 if (kvm->arch.apic_access_page)
3948 put_page(kvm->arch.apic_access_page);
3949 if (kvm->arch.ept_identity_pagetable)
3950 put_page(kvm->arch.ept_identity_pagetable);
3954 int kvm_arch_set_memory_region(struct kvm *kvm,
3955 struct kvm_userspace_memory_region *mem,
3956 struct kvm_memory_slot old,
3959 int npages = mem->memory_size >> PAGE_SHIFT;
3960 struct kvm_memory_slot *memslot = &kvm->memslots[mem->slot];
3962 /*To keep backward compatibility with older userspace,
3963 *x86 needs to hanlde !user_alloc case.
3966 if (npages && !old.rmap) {
3967 down_write(¤t->mm->mmap_sem);
3968 memslot->userspace_addr = do_mmap(NULL, 0,
3970 PROT_READ | PROT_WRITE,
3971 MAP_SHARED | MAP_ANONYMOUS,
3973 up_write(¤t->mm->mmap_sem);
3975 if (IS_ERR((void *)memslot->userspace_addr))
3976 return PTR_ERR((void *)memslot->userspace_addr);
3978 if (!old.user_alloc && old.rmap) {
3981 down_write(¤t->mm->mmap_sem);
3982 ret = do_munmap(current->mm, old.userspace_addr,
3983 old.npages * PAGE_SIZE);
3984 up_write(¤t->mm->mmap_sem);
3987 "kvm_vm_ioctl_set_memory_region: "
3988 "failed to munmap memory\n");
3993 if (!kvm->arch.n_requested_mmu_pages) {
3994 unsigned int nr_mmu_pages = kvm_mmu_calculate_mmu_pages(kvm);
3995 kvm_mmu_change_mmu_pages(kvm, nr_mmu_pages);
3998 kvm_mmu_slot_remove_write_access(kvm, mem->slot);
3999 kvm_flush_remote_tlbs(kvm);
4004 int kvm_arch_vcpu_runnable(struct kvm_vcpu *vcpu)
4006 return vcpu->arch.mp_state == KVM_MP_STATE_RUNNABLE
4007 || vcpu->arch.mp_state == KVM_MP_STATE_SIPI_RECEIVED;
4010 static void vcpu_kick_intr(void *info)
4013 struct kvm_vcpu *vcpu = (struct kvm_vcpu *)info;
4014 printk(KERN_DEBUG "vcpu_kick_intr %p \n", vcpu);
4018 void kvm_vcpu_kick(struct kvm_vcpu *vcpu)
4020 int ipi_pcpu = vcpu->cpu;
4021 int cpu = get_cpu();
4023 if (waitqueue_active(&vcpu->wq)) {
4024 wake_up_interruptible(&vcpu->wq);
4025 ++vcpu->stat.halt_wakeup;
4028 * We may be called synchronously with irqs disabled in guest mode,
4029 * So need not to call smp_call_function_single() in that case.
4031 if (vcpu->guest_mode && vcpu->cpu != cpu)
4032 smp_call_function_single(ipi_pcpu, vcpu_kick_intr, vcpu, 0);