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>
22 #include <linux/clocksource.h>
23 #include <linux/kvm.h>
25 #include <linux/vmalloc.h>
26 #include <linux/module.h>
27 #include <linux/mman.h>
28 #include <linux/highmem.h>
30 #include <asm/uaccess.h>
34 #define MAX_IO_MSRS 256
35 #define CR0_RESERVED_BITS \
36 (~(unsigned long)(X86_CR0_PE | X86_CR0_MP | X86_CR0_EM | X86_CR0_TS \
37 | X86_CR0_ET | X86_CR0_NE | X86_CR0_WP | X86_CR0_AM \
38 | X86_CR0_NW | X86_CR0_CD | X86_CR0_PG))
39 #define CR4_RESERVED_BITS \
40 (~(unsigned long)(X86_CR4_VME | X86_CR4_PVI | X86_CR4_TSD | X86_CR4_DE\
41 | X86_CR4_PSE | X86_CR4_PAE | X86_CR4_MCE \
42 | X86_CR4_PGE | X86_CR4_PCE | X86_CR4_OSFXSR \
43 | X86_CR4_OSXMMEXCPT | X86_CR4_VMXE))
45 #define CR8_RESERVED_BITS (~(unsigned long)X86_CR8_TPR)
47 * - enable syscall per default because its emulated by KVM
48 * - enable LME and LMA per default on 64 bit KVM
51 static u64 __read_mostly efer_reserved_bits = 0xfffffffffffffafeULL;
53 static u64 __read_mostly efer_reserved_bits = 0xfffffffffffffffeULL;
56 #define VM_STAT(x) offsetof(struct kvm, stat.x), KVM_STAT_VM
57 #define VCPU_STAT(x) offsetof(struct kvm_vcpu, stat.x), KVM_STAT_VCPU
59 static int kvm_dev_ioctl_get_supported_cpuid(struct kvm_cpuid2 *cpuid,
60 struct kvm_cpuid_entry2 __user *entries);
62 struct kvm_x86_ops *kvm_x86_ops;
64 struct kvm_stats_debugfs_item debugfs_entries[] = {
65 { "pf_fixed", VCPU_STAT(pf_fixed) },
66 { "pf_guest", VCPU_STAT(pf_guest) },
67 { "tlb_flush", VCPU_STAT(tlb_flush) },
68 { "invlpg", VCPU_STAT(invlpg) },
69 { "exits", VCPU_STAT(exits) },
70 { "io_exits", VCPU_STAT(io_exits) },
71 { "mmio_exits", VCPU_STAT(mmio_exits) },
72 { "signal_exits", VCPU_STAT(signal_exits) },
73 { "irq_window", VCPU_STAT(irq_window_exits) },
74 { "halt_exits", VCPU_STAT(halt_exits) },
75 { "halt_wakeup", VCPU_STAT(halt_wakeup) },
76 { "hypercalls", VCPU_STAT(hypercalls) },
77 { "request_irq", VCPU_STAT(request_irq_exits) },
78 { "irq_exits", VCPU_STAT(irq_exits) },
79 { "host_state_reload", VCPU_STAT(host_state_reload) },
80 { "efer_reload", VCPU_STAT(efer_reload) },
81 { "fpu_reload", VCPU_STAT(fpu_reload) },
82 { "insn_emulation", VCPU_STAT(insn_emulation) },
83 { "insn_emulation_fail", VCPU_STAT(insn_emulation_fail) },
84 { "mmu_shadow_zapped", VM_STAT(mmu_shadow_zapped) },
85 { "mmu_pte_write", VM_STAT(mmu_pte_write) },
86 { "mmu_pte_updated", VM_STAT(mmu_pte_updated) },
87 { "mmu_pde_zapped", VM_STAT(mmu_pde_zapped) },
88 { "mmu_flooded", VM_STAT(mmu_flooded) },
89 { "mmu_recycled", VM_STAT(mmu_recycled) },
90 { "mmu_cache_miss", VM_STAT(mmu_cache_miss) },
91 { "remote_tlb_flush", VM_STAT(remote_tlb_flush) },
92 { "largepages", VM_STAT(lpages) },
97 unsigned long segment_base(u16 selector)
99 struct descriptor_table gdt;
100 struct desc_struct *d;
101 unsigned long table_base;
107 asm("sgdt %0" : "=m"(gdt));
108 table_base = gdt.base;
110 if (selector & 4) { /* from ldt */
113 asm("sldt %0" : "=g"(ldt_selector));
114 table_base = segment_base(ldt_selector);
116 d = (struct desc_struct *)(table_base + (selector & ~7));
117 v = d->base0 | ((unsigned long)d->base1 << 16) |
118 ((unsigned long)d->base2 << 24);
120 if (d->s == 0 && (d->type == 2 || d->type == 9 || d->type == 11))
121 v |= ((unsigned long)((struct ldttss_desc64 *)d)->base3) << 32;
125 EXPORT_SYMBOL_GPL(segment_base);
127 u64 kvm_get_apic_base(struct kvm_vcpu *vcpu)
129 if (irqchip_in_kernel(vcpu->kvm))
130 return vcpu->arch.apic_base;
132 return vcpu->arch.apic_base;
134 EXPORT_SYMBOL_GPL(kvm_get_apic_base);
136 void kvm_set_apic_base(struct kvm_vcpu *vcpu, u64 data)
138 /* TODO: reserve bits check */
139 if (irqchip_in_kernel(vcpu->kvm))
140 kvm_lapic_set_base(vcpu, data);
142 vcpu->arch.apic_base = data;
144 EXPORT_SYMBOL_GPL(kvm_set_apic_base);
146 void kvm_queue_exception(struct kvm_vcpu *vcpu, unsigned nr)
148 WARN_ON(vcpu->arch.exception.pending);
149 vcpu->arch.exception.pending = true;
150 vcpu->arch.exception.has_error_code = false;
151 vcpu->arch.exception.nr = nr;
153 EXPORT_SYMBOL_GPL(kvm_queue_exception);
155 void kvm_inject_page_fault(struct kvm_vcpu *vcpu, unsigned long addr,
158 ++vcpu->stat.pf_guest;
159 if (vcpu->arch.exception.pending) {
160 if (vcpu->arch.exception.nr == PF_VECTOR) {
161 printk(KERN_DEBUG "kvm: inject_page_fault:"
162 " double fault 0x%lx\n", addr);
163 vcpu->arch.exception.nr = DF_VECTOR;
164 vcpu->arch.exception.error_code = 0;
165 } else if (vcpu->arch.exception.nr == DF_VECTOR) {
166 /* triple fault -> shutdown */
167 set_bit(KVM_REQ_TRIPLE_FAULT, &vcpu->requests);
171 vcpu->arch.cr2 = addr;
172 kvm_queue_exception_e(vcpu, PF_VECTOR, error_code);
175 void kvm_queue_exception_e(struct kvm_vcpu *vcpu, unsigned nr, u32 error_code)
177 WARN_ON(vcpu->arch.exception.pending);
178 vcpu->arch.exception.pending = true;
179 vcpu->arch.exception.has_error_code = true;
180 vcpu->arch.exception.nr = nr;
181 vcpu->arch.exception.error_code = error_code;
183 EXPORT_SYMBOL_GPL(kvm_queue_exception_e);
185 static void __queue_exception(struct kvm_vcpu *vcpu)
187 kvm_x86_ops->queue_exception(vcpu, vcpu->arch.exception.nr,
188 vcpu->arch.exception.has_error_code,
189 vcpu->arch.exception.error_code);
193 * Load the pae pdptrs. Return true is they are all valid.
195 int load_pdptrs(struct kvm_vcpu *vcpu, unsigned long cr3)
197 gfn_t pdpt_gfn = cr3 >> PAGE_SHIFT;
198 unsigned offset = ((cr3 & (PAGE_SIZE-1)) >> 5) << 2;
201 u64 pdpte[ARRAY_SIZE(vcpu->arch.pdptrs)];
203 down_read(&vcpu->kvm->slots_lock);
204 ret = kvm_read_guest_page(vcpu->kvm, pdpt_gfn, pdpte,
205 offset * sizeof(u64), sizeof(pdpte));
210 for (i = 0; i < ARRAY_SIZE(pdpte); ++i) {
211 if ((pdpte[i] & 1) && (pdpte[i] & 0xfffffff0000001e6ull)) {
218 memcpy(vcpu->arch.pdptrs, pdpte, sizeof(vcpu->arch.pdptrs));
220 up_read(&vcpu->kvm->slots_lock);
224 EXPORT_SYMBOL_GPL(load_pdptrs);
226 static bool pdptrs_changed(struct kvm_vcpu *vcpu)
228 u64 pdpte[ARRAY_SIZE(vcpu->arch.pdptrs)];
232 if (is_long_mode(vcpu) || !is_pae(vcpu))
235 down_read(&vcpu->kvm->slots_lock);
236 r = kvm_read_guest(vcpu->kvm, vcpu->arch.cr3 & ~31u, pdpte, sizeof(pdpte));
239 changed = memcmp(pdpte, vcpu->arch.pdptrs, sizeof(pdpte)) != 0;
241 up_read(&vcpu->kvm->slots_lock);
246 void kvm_set_cr0(struct kvm_vcpu *vcpu, unsigned long cr0)
248 if (cr0 & CR0_RESERVED_BITS) {
249 printk(KERN_DEBUG "set_cr0: 0x%lx #GP, reserved bits 0x%lx\n",
250 cr0, vcpu->arch.cr0);
251 kvm_inject_gp(vcpu, 0);
255 if ((cr0 & X86_CR0_NW) && !(cr0 & X86_CR0_CD)) {
256 printk(KERN_DEBUG "set_cr0: #GP, CD == 0 && NW == 1\n");
257 kvm_inject_gp(vcpu, 0);
261 if ((cr0 & X86_CR0_PG) && !(cr0 & X86_CR0_PE)) {
262 printk(KERN_DEBUG "set_cr0: #GP, set PG flag "
263 "and a clear PE flag\n");
264 kvm_inject_gp(vcpu, 0);
268 if (!is_paging(vcpu) && (cr0 & X86_CR0_PG)) {
270 if ((vcpu->arch.shadow_efer & EFER_LME)) {
274 printk(KERN_DEBUG "set_cr0: #GP, start paging "
275 "in long mode while PAE is disabled\n");
276 kvm_inject_gp(vcpu, 0);
279 kvm_x86_ops->get_cs_db_l_bits(vcpu, &cs_db, &cs_l);
281 printk(KERN_DEBUG "set_cr0: #GP, start paging "
282 "in long mode while CS.L == 1\n");
283 kvm_inject_gp(vcpu, 0);
289 if (is_pae(vcpu) && !load_pdptrs(vcpu, vcpu->arch.cr3)) {
290 printk(KERN_DEBUG "set_cr0: #GP, pdptrs "
292 kvm_inject_gp(vcpu, 0);
298 kvm_x86_ops->set_cr0(vcpu, cr0);
299 vcpu->arch.cr0 = cr0;
301 kvm_mmu_reset_context(vcpu);
304 EXPORT_SYMBOL_GPL(kvm_set_cr0);
306 void kvm_lmsw(struct kvm_vcpu *vcpu, unsigned long msw)
308 kvm_set_cr0(vcpu, (vcpu->arch.cr0 & ~0x0ful) | (msw & 0x0f));
310 EXPORT_SYMBOL_GPL(kvm_lmsw);
312 void kvm_set_cr4(struct kvm_vcpu *vcpu, unsigned long cr4)
314 if (cr4 & CR4_RESERVED_BITS) {
315 printk(KERN_DEBUG "set_cr4: #GP, reserved bits\n");
316 kvm_inject_gp(vcpu, 0);
320 if (is_long_mode(vcpu)) {
321 if (!(cr4 & X86_CR4_PAE)) {
322 printk(KERN_DEBUG "set_cr4: #GP, clearing PAE while "
324 kvm_inject_gp(vcpu, 0);
327 } else if (is_paging(vcpu) && !is_pae(vcpu) && (cr4 & X86_CR4_PAE)
328 && !load_pdptrs(vcpu, vcpu->arch.cr3)) {
329 printk(KERN_DEBUG "set_cr4: #GP, pdptrs reserved bits\n");
330 kvm_inject_gp(vcpu, 0);
334 if (cr4 & X86_CR4_VMXE) {
335 printk(KERN_DEBUG "set_cr4: #GP, setting VMXE\n");
336 kvm_inject_gp(vcpu, 0);
339 kvm_x86_ops->set_cr4(vcpu, cr4);
340 vcpu->arch.cr4 = cr4;
341 kvm_mmu_reset_context(vcpu);
343 EXPORT_SYMBOL_GPL(kvm_set_cr4);
345 void kvm_set_cr3(struct kvm_vcpu *vcpu, unsigned long cr3)
347 if (cr3 == vcpu->arch.cr3 && !pdptrs_changed(vcpu)) {
348 kvm_mmu_flush_tlb(vcpu);
352 if (is_long_mode(vcpu)) {
353 if (cr3 & CR3_L_MODE_RESERVED_BITS) {
354 printk(KERN_DEBUG "set_cr3: #GP, reserved bits\n");
355 kvm_inject_gp(vcpu, 0);
360 if (cr3 & CR3_PAE_RESERVED_BITS) {
362 "set_cr3: #GP, reserved bits\n");
363 kvm_inject_gp(vcpu, 0);
366 if (is_paging(vcpu) && !load_pdptrs(vcpu, cr3)) {
367 printk(KERN_DEBUG "set_cr3: #GP, pdptrs "
369 kvm_inject_gp(vcpu, 0);
374 * We don't check reserved bits in nonpae mode, because
375 * this isn't enforced, and VMware depends on this.
379 down_read(&vcpu->kvm->slots_lock);
381 * Does the new cr3 value map to physical memory? (Note, we
382 * catch an invalid cr3 even in real-mode, because it would
383 * cause trouble later on when we turn on paging anyway.)
385 * A real CPU would silently accept an invalid cr3 and would
386 * attempt to use it - with largely undefined (and often hard
387 * to debug) behavior on the guest side.
389 if (unlikely(!gfn_to_memslot(vcpu->kvm, cr3 >> PAGE_SHIFT)))
390 kvm_inject_gp(vcpu, 0);
392 vcpu->arch.cr3 = cr3;
393 vcpu->arch.mmu.new_cr3(vcpu);
395 up_read(&vcpu->kvm->slots_lock);
397 EXPORT_SYMBOL_GPL(kvm_set_cr3);
399 void kvm_set_cr8(struct kvm_vcpu *vcpu, unsigned long cr8)
401 if (cr8 & CR8_RESERVED_BITS) {
402 printk(KERN_DEBUG "set_cr8: #GP, reserved bits 0x%lx\n", cr8);
403 kvm_inject_gp(vcpu, 0);
406 if (irqchip_in_kernel(vcpu->kvm))
407 kvm_lapic_set_tpr(vcpu, cr8);
409 vcpu->arch.cr8 = cr8;
411 EXPORT_SYMBOL_GPL(kvm_set_cr8);
413 unsigned long kvm_get_cr8(struct kvm_vcpu *vcpu)
415 if (irqchip_in_kernel(vcpu->kvm))
416 return kvm_lapic_get_cr8(vcpu);
418 return vcpu->arch.cr8;
420 EXPORT_SYMBOL_GPL(kvm_get_cr8);
423 * List of msr numbers which we expose to userspace through KVM_GET_MSRS
424 * and KVM_SET_MSRS, and KVM_GET_MSR_INDEX_LIST.
426 * This list is modified at module load time to reflect the
427 * capabilities of the host cpu.
429 static u32 msrs_to_save[] = {
430 MSR_IA32_SYSENTER_CS, MSR_IA32_SYSENTER_ESP, MSR_IA32_SYSENTER_EIP,
433 MSR_CSTAR, MSR_KERNEL_GS_BASE, MSR_SYSCALL_MASK, MSR_LSTAR,
435 MSR_IA32_TIME_STAMP_COUNTER, MSR_KVM_SYSTEM_TIME, MSR_KVM_WALL_CLOCK,
436 MSR_IA32_PERF_STATUS,
439 static unsigned num_msrs_to_save;
441 static u32 emulated_msrs[] = {
442 MSR_IA32_MISC_ENABLE,
445 static void set_efer(struct kvm_vcpu *vcpu, u64 efer)
447 if (efer & efer_reserved_bits) {
448 printk(KERN_DEBUG "set_efer: 0x%llx #GP, reserved bits\n",
450 kvm_inject_gp(vcpu, 0);
455 && (vcpu->arch.shadow_efer & EFER_LME) != (efer & EFER_LME)) {
456 printk(KERN_DEBUG "set_efer: #GP, change LME while paging\n");
457 kvm_inject_gp(vcpu, 0);
461 kvm_x86_ops->set_efer(vcpu, efer);
464 efer |= vcpu->arch.shadow_efer & EFER_LMA;
466 vcpu->arch.shadow_efer = efer;
469 void kvm_enable_efer_bits(u64 mask)
471 efer_reserved_bits &= ~mask;
473 EXPORT_SYMBOL_GPL(kvm_enable_efer_bits);
477 * Writes msr value into into the appropriate "register".
478 * Returns 0 on success, non-0 otherwise.
479 * Assumes vcpu_load() was already called.
481 int kvm_set_msr(struct kvm_vcpu *vcpu, u32 msr_index, u64 data)
483 return kvm_x86_ops->set_msr(vcpu, msr_index, data);
487 * Adapt set_msr() to msr_io()'s calling convention
489 static int do_set_msr(struct kvm_vcpu *vcpu, unsigned index, u64 *data)
491 return kvm_set_msr(vcpu, index, *data);
494 static void kvm_write_wall_clock(struct kvm *kvm, gpa_t wall_clock)
497 struct kvm_wall_clock wc;
498 struct timespec wc_ts;
505 down_read(&kvm->slots_lock);
506 kvm_write_guest(kvm, wall_clock, &version, sizeof(version));
508 wc_ts = current_kernel_time();
509 wc.wc_sec = wc_ts.tv_sec;
510 wc.wc_nsec = wc_ts.tv_nsec;
511 wc.wc_version = version;
513 kvm_write_guest(kvm, wall_clock, &wc, sizeof(wc));
516 kvm_write_guest(kvm, wall_clock, &version, sizeof(version));
517 up_read(&kvm->slots_lock);
520 static void kvm_write_guest_time(struct kvm_vcpu *v)
524 struct kvm_vcpu_arch *vcpu = &v->arch;
527 if ((!vcpu->time_page))
530 /* Keep irq disabled to prevent changes to the clock */
531 local_irq_save(flags);
532 kvm_get_msr(v, MSR_IA32_TIME_STAMP_COUNTER,
533 &vcpu->hv_clock.tsc_timestamp);
535 local_irq_restore(flags);
537 /* With all the info we got, fill in the values */
539 vcpu->hv_clock.system_time = ts.tv_nsec +
540 (NSEC_PER_SEC * (u64)ts.tv_sec);
542 * The interface expects us to write an even number signaling that the
543 * update is finished. Since the guest won't see the intermediate
544 * state, we just write "2" at the end
546 vcpu->hv_clock.version = 2;
548 shared_kaddr = kmap_atomic(vcpu->time_page, KM_USER0);
550 memcpy(shared_kaddr + vcpu->time_offset, &vcpu->hv_clock,
551 sizeof(vcpu->hv_clock));
553 kunmap_atomic(shared_kaddr, KM_USER0);
555 mark_page_dirty(v->kvm, vcpu->time >> PAGE_SHIFT);
559 int kvm_set_msr_common(struct kvm_vcpu *vcpu, u32 msr, u64 data)
563 set_efer(vcpu, data);
565 case MSR_IA32_MC0_STATUS:
566 pr_unimpl(vcpu, "%s: MSR_IA32_MC0_STATUS 0x%llx, nop\n",
569 case MSR_IA32_MCG_STATUS:
570 pr_unimpl(vcpu, "%s: MSR_IA32_MCG_STATUS 0x%llx, nop\n",
573 case MSR_IA32_MCG_CTL:
574 pr_unimpl(vcpu, "%s: MSR_IA32_MCG_CTL 0x%llx, nop\n",
577 case MSR_IA32_UCODE_REV:
578 case MSR_IA32_UCODE_WRITE:
579 case 0x200 ... 0x2ff: /* MTRRs */
581 case MSR_IA32_APICBASE:
582 kvm_set_apic_base(vcpu, data);
584 case MSR_IA32_MISC_ENABLE:
585 vcpu->arch.ia32_misc_enable_msr = data;
587 case MSR_KVM_WALL_CLOCK:
588 vcpu->kvm->arch.wall_clock = data;
589 kvm_write_wall_clock(vcpu->kvm, data);
591 case MSR_KVM_SYSTEM_TIME: {
592 if (vcpu->arch.time_page) {
593 kvm_release_page_dirty(vcpu->arch.time_page);
594 vcpu->arch.time_page = NULL;
597 vcpu->arch.time = data;
599 /* we verify if the enable bit is set... */
603 /* ...but clean it before doing the actual write */
604 vcpu->arch.time_offset = data & ~(PAGE_MASK | 1);
606 vcpu->arch.hv_clock.tsc_to_system_mul =
607 clocksource_khz2mult(tsc_khz, 22);
608 vcpu->arch.hv_clock.tsc_shift = 22;
610 down_read(¤t->mm->mmap_sem);
611 down_read(&vcpu->kvm->slots_lock);
612 vcpu->arch.time_page =
613 gfn_to_page(vcpu->kvm, data >> PAGE_SHIFT);
614 up_read(&vcpu->kvm->slots_lock);
615 up_read(¤t->mm->mmap_sem);
617 if (is_error_page(vcpu->arch.time_page)) {
618 kvm_release_page_clean(vcpu->arch.time_page);
619 vcpu->arch.time_page = NULL;
622 kvm_write_guest_time(vcpu);
626 pr_unimpl(vcpu, "unhandled wrmsr: 0x%x data %llx\n", msr, data);
631 EXPORT_SYMBOL_GPL(kvm_set_msr_common);
635 * Reads an msr value (of 'msr_index') into 'pdata'.
636 * Returns 0 on success, non-0 otherwise.
637 * Assumes vcpu_load() was already called.
639 int kvm_get_msr(struct kvm_vcpu *vcpu, u32 msr_index, u64 *pdata)
641 return kvm_x86_ops->get_msr(vcpu, msr_index, pdata);
644 int kvm_get_msr_common(struct kvm_vcpu *vcpu, u32 msr, u64 *pdata)
649 case 0xc0010010: /* SYSCFG */
650 case 0xc0010015: /* HWCR */
651 case MSR_IA32_PLATFORM_ID:
652 case MSR_IA32_P5_MC_ADDR:
653 case MSR_IA32_P5_MC_TYPE:
654 case MSR_IA32_MC0_CTL:
655 case MSR_IA32_MCG_STATUS:
656 case MSR_IA32_MCG_CAP:
657 case MSR_IA32_MCG_CTL:
658 case MSR_IA32_MC0_MISC:
659 case MSR_IA32_MC0_MISC+4:
660 case MSR_IA32_MC0_MISC+8:
661 case MSR_IA32_MC0_MISC+12:
662 case MSR_IA32_MC0_MISC+16:
663 case MSR_IA32_UCODE_REV:
664 case MSR_IA32_EBL_CR_POWERON:
667 case 0x200 ... 0x2ff:
670 case 0xcd: /* fsb frequency */
673 case MSR_IA32_APICBASE:
674 data = kvm_get_apic_base(vcpu);
676 case MSR_IA32_MISC_ENABLE:
677 data = vcpu->arch.ia32_misc_enable_msr;
679 case MSR_IA32_PERF_STATUS:
680 /* TSC increment by tick */
683 data |= (((uint64_t)4ULL) << 40);
686 data = vcpu->arch.shadow_efer;
688 case MSR_KVM_WALL_CLOCK:
689 data = vcpu->kvm->arch.wall_clock;
691 case MSR_KVM_SYSTEM_TIME:
692 data = vcpu->arch.time;
695 pr_unimpl(vcpu, "unhandled rdmsr: 0x%x\n", msr);
701 EXPORT_SYMBOL_GPL(kvm_get_msr_common);
704 * Read or write a bunch of msrs. All parameters are kernel addresses.
706 * @return number of msrs set successfully.
708 static int __msr_io(struct kvm_vcpu *vcpu, struct kvm_msrs *msrs,
709 struct kvm_msr_entry *entries,
710 int (*do_msr)(struct kvm_vcpu *vcpu,
711 unsigned index, u64 *data))
717 for (i = 0; i < msrs->nmsrs; ++i)
718 if (do_msr(vcpu, entries[i].index, &entries[i].data))
727 * Read or write a bunch of msrs. Parameters are user addresses.
729 * @return number of msrs set successfully.
731 static int msr_io(struct kvm_vcpu *vcpu, struct kvm_msrs __user *user_msrs,
732 int (*do_msr)(struct kvm_vcpu *vcpu,
733 unsigned index, u64 *data),
736 struct kvm_msrs msrs;
737 struct kvm_msr_entry *entries;
742 if (copy_from_user(&msrs, user_msrs, sizeof msrs))
746 if (msrs.nmsrs >= MAX_IO_MSRS)
750 size = sizeof(struct kvm_msr_entry) * msrs.nmsrs;
751 entries = vmalloc(size);
756 if (copy_from_user(entries, user_msrs->entries, size))
759 r = n = __msr_io(vcpu, &msrs, entries, do_msr);
764 if (writeback && copy_to_user(user_msrs->entries, entries, size))
776 * Make sure that a cpu that is being hot-unplugged does not have any vcpus
779 void decache_vcpus_on_cpu(int cpu)
782 struct kvm_vcpu *vcpu;
785 spin_lock(&kvm_lock);
786 list_for_each_entry(vm, &vm_list, vm_list)
787 for (i = 0; i < KVM_MAX_VCPUS; ++i) {
792 * If the vcpu is locked, then it is running on some
793 * other cpu and therefore it is not cached on the
796 * If it's not locked, check the last cpu it executed
799 if (mutex_trylock(&vcpu->mutex)) {
800 if (vcpu->cpu == cpu) {
801 kvm_x86_ops->vcpu_decache(vcpu);
804 mutex_unlock(&vcpu->mutex);
807 spin_unlock(&kvm_lock);
810 int kvm_dev_ioctl_check_extension(long ext)
815 case KVM_CAP_IRQCHIP:
817 case KVM_CAP_MMU_SHADOW_CACHE_CONTROL:
818 case KVM_CAP_USER_MEMORY:
819 case KVM_CAP_SET_TSS_ADDR:
820 case KVM_CAP_EXT_CPUID:
821 case KVM_CAP_CLOCKSOURCE:
823 case KVM_CAP_NOP_IO_DELAY:
827 r = !kvm_x86_ops->cpu_has_accelerated_tpr();
829 case KVM_CAP_NR_VCPUS:
832 case KVM_CAP_NR_MEMSLOTS:
833 r = KVM_MEMORY_SLOTS;
843 long kvm_arch_dev_ioctl(struct file *filp,
844 unsigned int ioctl, unsigned long arg)
846 void __user *argp = (void __user *)arg;
850 case KVM_GET_MSR_INDEX_LIST: {
851 struct kvm_msr_list __user *user_msr_list = argp;
852 struct kvm_msr_list msr_list;
856 if (copy_from_user(&msr_list, user_msr_list, sizeof msr_list))
859 msr_list.nmsrs = num_msrs_to_save + ARRAY_SIZE(emulated_msrs);
860 if (copy_to_user(user_msr_list, &msr_list, sizeof msr_list))
863 if (n < num_msrs_to_save)
866 if (copy_to_user(user_msr_list->indices, &msrs_to_save,
867 num_msrs_to_save * sizeof(u32)))
869 if (copy_to_user(user_msr_list->indices
870 + num_msrs_to_save * sizeof(u32),
872 ARRAY_SIZE(emulated_msrs) * sizeof(u32)))
877 case KVM_GET_SUPPORTED_CPUID: {
878 struct kvm_cpuid2 __user *cpuid_arg = argp;
879 struct kvm_cpuid2 cpuid;
882 if (copy_from_user(&cpuid, cpuid_arg, sizeof cpuid))
884 r = kvm_dev_ioctl_get_supported_cpuid(&cpuid,
890 if (copy_to_user(cpuid_arg, &cpuid, sizeof cpuid))
902 void kvm_arch_vcpu_load(struct kvm_vcpu *vcpu, int cpu)
904 kvm_x86_ops->vcpu_load(vcpu, cpu);
905 kvm_write_guest_time(vcpu);
908 void kvm_arch_vcpu_put(struct kvm_vcpu *vcpu)
910 kvm_x86_ops->vcpu_put(vcpu);
911 kvm_put_guest_fpu(vcpu);
914 static int is_efer_nx(void)
918 rdmsrl(MSR_EFER, efer);
919 return efer & EFER_NX;
922 static void cpuid_fix_nx_cap(struct kvm_vcpu *vcpu)
925 struct kvm_cpuid_entry2 *e, *entry;
928 for (i = 0; i < vcpu->arch.cpuid_nent; ++i) {
929 e = &vcpu->arch.cpuid_entries[i];
930 if (e->function == 0x80000001) {
935 if (entry && (entry->edx & (1 << 20)) && !is_efer_nx()) {
936 entry->edx &= ~(1 << 20);
937 printk(KERN_INFO "kvm: guest NX capability removed\n");
941 /* when an old userspace process fills a new kernel module */
942 static int kvm_vcpu_ioctl_set_cpuid(struct kvm_vcpu *vcpu,
943 struct kvm_cpuid *cpuid,
944 struct kvm_cpuid_entry __user *entries)
947 struct kvm_cpuid_entry *cpuid_entries;
950 if (cpuid->nent > KVM_MAX_CPUID_ENTRIES)
953 cpuid_entries = vmalloc(sizeof(struct kvm_cpuid_entry) * cpuid->nent);
957 if (copy_from_user(cpuid_entries, entries,
958 cpuid->nent * sizeof(struct kvm_cpuid_entry)))
960 for (i = 0; i < cpuid->nent; i++) {
961 vcpu->arch.cpuid_entries[i].function = cpuid_entries[i].function;
962 vcpu->arch.cpuid_entries[i].eax = cpuid_entries[i].eax;
963 vcpu->arch.cpuid_entries[i].ebx = cpuid_entries[i].ebx;
964 vcpu->arch.cpuid_entries[i].ecx = cpuid_entries[i].ecx;
965 vcpu->arch.cpuid_entries[i].edx = cpuid_entries[i].edx;
966 vcpu->arch.cpuid_entries[i].index = 0;
967 vcpu->arch.cpuid_entries[i].flags = 0;
968 vcpu->arch.cpuid_entries[i].padding[0] = 0;
969 vcpu->arch.cpuid_entries[i].padding[1] = 0;
970 vcpu->arch.cpuid_entries[i].padding[2] = 0;
972 vcpu->arch.cpuid_nent = cpuid->nent;
973 cpuid_fix_nx_cap(vcpu);
977 vfree(cpuid_entries);
982 static int kvm_vcpu_ioctl_set_cpuid2(struct kvm_vcpu *vcpu,
983 struct kvm_cpuid2 *cpuid,
984 struct kvm_cpuid_entry2 __user *entries)
989 if (cpuid->nent > KVM_MAX_CPUID_ENTRIES)
992 if (copy_from_user(&vcpu->arch.cpuid_entries, entries,
993 cpuid->nent * sizeof(struct kvm_cpuid_entry2)))
995 vcpu->arch.cpuid_nent = cpuid->nent;
1002 static int kvm_vcpu_ioctl_get_cpuid2(struct kvm_vcpu *vcpu,
1003 struct kvm_cpuid2 *cpuid,
1004 struct kvm_cpuid_entry2 __user *entries)
1009 if (cpuid->nent < vcpu->arch.cpuid_nent)
1012 if (copy_to_user(entries, &vcpu->arch.cpuid_entries,
1013 vcpu->arch.cpuid_nent * sizeof(struct kvm_cpuid_entry2)))
1018 cpuid->nent = vcpu->arch.cpuid_nent;
1022 static inline u32 bit(int bitno)
1024 return 1 << (bitno & 31);
1027 static void do_cpuid_1_ent(struct kvm_cpuid_entry2 *entry, u32 function,
1030 entry->function = function;
1031 entry->index = index;
1032 cpuid_count(entry->function, entry->index,
1033 &entry->eax, &entry->ebx, &entry->ecx, &entry->edx);
1037 static void do_cpuid_ent(struct kvm_cpuid_entry2 *entry, u32 function,
1038 u32 index, int *nent, int maxnent)
1040 const u32 kvm_supported_word0_x86_features = bit(X86_FEATURE_FPU) |
1041 bit(X86_FEATURE_VME) | bit(X86_FEATURE_DE) |
1042 bit(X86_FEATURE_PSE) | bit(X86_FEATURE_TSC) |
1043 bit(X86_FEATURE_MSR) | bit(X86_FEATURE_PAE) |
1044 bit(X86_FEATURE_CX8) | bit(X86_FEATURE_APIC) |
1045 bit(X86_FEATURE_SEP) | bit(X86_FEATURE_PGE) |
1046 bit(X86_FEATURE_CMOV) | bit(X86_FEATURE_PSE36) |
1047 bit(X86_FEATURE_CLFLSH) | bit(X86_FEATURE_MMX) |
1048 bit(X86_FEATURE_FXSR) | bit(X86_FEATURE_XMM) |
1049 bit(X86_FEATURE_XMM2) | bit(X86_FEATURE_SELFSNOOP);
1050 const u32 kvm_supported_word1_x86_features = bit(X86_FEATURE_FPU) |
1051 bit(X86_FEATURE_VME) | bit(X86_FEATURE_DE) |
1052 bit(X86_FEATURE_PSE) | bit(X86_FEATURE_TSC) |
1053 bit(X86_FEATURE_MSR) | bit(X86_FEATURE_PAE) |
1054 bit(X86_FEATURE_CX8) | bit(X86_FEATURE_APIC) |
1055 bit(X86_FEATURE_PGE) |
1056 bit(X86_FEATURE_CMOV) | bit(X86_FEATURE_PSE36) |
1057 bit(X86_FEATURE_MMX) | bit(X86_FEATURE_FXSR) |
1058 bit(X86_FEATURE_SYSCALL) |
1059 (bit(X86_FEATURE_NX) && is_efer_nx()) |
1060 #ifdef CONFIG_X86_64
1061 bit(X86_FEATURE_LM) |
1063 bit(X86_FEATURE_MMXEXT) |
1064 bit(X86_FEATURE_3DNOWEXT) |
1065 bit(X86_FEATURE_3DNOW);
1066 const u32 kvm_supported_word3_x86_features =
1067 bit(X86_FEATURE_XMM3) | bit(X86_FEATURE_CX16);
1068 const u32 kvm_supported_word6_x86_features =
1069 bit(X86_FEATURE_LAHF_LM) | bit(X86_FEATURE_CMP_LEGACY);
1071 /* all func 2 cpuid_count() should be called on the same cpu */
1073 do_cpuid_1_ent(entry, function, index);
1078 entry->eax = min(entry->eax, (u32)0xb);
1081 entry->edx &= kvm_supported_word0_x86_features;
1082 entry->ecx &= kvm_supported_word3_x86_features;
1084 /* function 2 entries are STATEFUL. That is, repeated cpuid commands
1085 * may return different values. This forces us to get_cpu() before
1086 * issuing the first command, and also to emulate this annoying behavior
1087 * in kvm_emulate_cpuid() using KVM_CPUID_FLAG_STATE_READ_NEXT */
1089 int t, times = entry->eax & 0xff;
1091 entry->flags |= KVM_CPUID_FLAG_STATEFUL_FUNC;
1092 for (t = 1; t < times && *nent < maxnent; ++t) {
1093 do_cpuid_1_ent(&entry[t], function, 0);
1094 entry[t].flags |= KVM_CPUID_FLAG_STATEFUL_FUNC;
1099 /* function 4 and 0xb have additional index. */
1103 entry->flags |= KVM_CPUID_FLAG_SIGNIFCANT_INDEX;
1104 /* read more entries until cache_type is zero */
1105 for (i = 1; *nent < maxnent; ++i) {
1106 cache_type = entry[i - 1].eax & 0x1f;
1109 do_cpuid_1_ent(&entry[i], function, i);
1111 KVM_CPUID_FLAG_SIGNIFCANT_INDEX;
1119 entry->flags |= KVM_CPUID_FLAG_SIGNIFCANT_INDEX;
1120 /* read more entries until level_type is zero */
1121 for (i = 1; *nent < maxnent; ++i) {
1122 level_type = entry[i - 1].ecx & 0xff;
1125 do_cpuid_1_ent(&entry[i], function, i);
1127 KVM_CPUID_FLAG_SIGNIFCANT_INDEX;
1133 entry->eax = min(entry->eax, 0x8000001a);
1136 entry->edx &= kvm_supported_word1_x86_features;
1137 entry->ecx &= kvm_supported_word6_x86_features;
1143 static int kvm_dev_ioctl_get_supported_cpuid(struct kvm_cpuid2 *cpuid,
1144 struct kvm_cpuid_entry2 __user *entries)
1146 struct kvm_cpuid_entry2 *cpuid_entries;
1147 int limit, nent = 0, r = -E2BIG;
1150 if (cpuid->nent < 1)
1153 cpuid_entries = vmalloc(sizeof(struct kvm_cpuid_entry2) * cpuid->nent);
1157 do_cpuid_ent(&cpuid_entries[0], 0, 0, &nent, cpuid->nent);
1158 limit = cpuid_entries[0].eax;
1159 for (func = 1; func <= limit && nent < cpuid->nent; ++func)
1160 do_cpuid_ent(&cpuid_entries[nent], func, 0,
1161 &nent, cpuid->nent);
1163 if (nent >= cpuid->nent)
1166 do_cpuid_ent(&cpuid_entries[nent], 0x80000000, 0, &nent, cpuid->nent);
1167 limit = cpuid_entries[nent - 1].eax;
1168 for (func = 0x80000001; func <= limit && nent < cpuid->nent; ++func)
1169 do_cpuid_ent(&cpuid_entries[nent], func, 0,
1170 &nent, cpuid->nent);
1172 if (copy_to_user(entries, cpuid_entries,
1173 nent * sizeof(struct kvm_cpuid_entry2)))
1179 vfree(cpuid_entries);
1184 static int kvm_vcpu_ioctl_get_lapic(struct kvm_vcpu *vcpu,
1185 struct kvm_lapic_state *s)
1188 memcpy(s->regs, vcpu->arch.apic->regs, sizeof *s);
1194 static int kvm_vcpu_ioctl_set_lapic(struct kvm_vcpu *vcpu,
1195 struct kvm_lapic_state *s)
1198 memcpy(vcpu->arch.apic->regs, s->regs, sizeof *s);
1199 kvm_apic_post_state_restore(vcpu);
1205 static int kvm_vcpu_ioctl_interrupt(struct kvm_vcpu *vcpu,
1206 struct kvm_interrupt *irq)
1208 if (irq->irq < 0 || irq->irq >= 256)
1210 if (irqchip_in_kernel(vcpu->kvm))
1214 set_bit(irq->irq, vcpu->arch.irq_pending);
1215 set_bit(irq->irq / BITS_PER_LONG, &vcpu->arch.irq_summary);
1222 static int vcpu_ioctl_tpr_access_reporting(struct kvm_vcpu *vcpu,
1223 struct kvm_tpr_access_ctl *tac)
1227 vcpu->arch.tpr_access_reporting = !!tac->enabled;
1231 long kvm_arch_vcpu_ioctl(struct file *filp,
1232 unsigned int ioctl, unsigned long arg)
1234 struct kvm_vcpu *vcpu = filp->private_data;
1235 void __user *argp = (void __user *)arg;
1239 case KVM_GET_LAPIC: {
1240 struct kvm_lapic_state lapic;
1242 memset(&lapic, 0, sizeof lapic);
1243 r = kvm_vcpu_ioctl_get_lapic(vcpu, &lapic);
1247 if (copy_to_user(argp, &lapic, sizeof lapic))
1252 case KVM_SET_LAPIC: {
1253 struct kvm_lapic_state lapic;
1256 if (copy_from_user(&lapic, argp, sizeof lapic))
1258 r = kvm_vcpu_ioctl_set_lapic(vcpu, &lapic);;
1264 case KVM_INTERRUPT: {
1265 struct kvm_interrupt irq;
1268 if (copy_from_user(&irq, argp, sizeof irq))
1270 r = kvm_vcpu_ioctl_interrupt(vcpu, &irq);
1276 case KVM_SET_CPUID: {
1277 struct kvm_cpuid __user *cpuid_arg = argp;
1278 struct kvm_cpuid cpuid;
1281 if (copy_from_user(&cpuid, cpuid_arg, sizeof cpuid))
1283 r = kvm_vcpu_ioctl_set_cpuid(vcpu, &cpuid, cpuid_arg->entries);
1288 case KVM_SET_CPUID2: {
1289 struct kvm_cpuid2 __user *cpuid_arg = argp;
1290 struct kvm_cpuid2 cpuid;
1293 if (copy_from_user(&cpuid, cpuid_arg, sizeof cpuid))
1295 r = kvm_vcpu_ioctl_set_cpuid2(vcpu, &cpuid,
1296 cpuid_arg->entries);
1301 case KVM_GET_CPUID2: {
1302 struct kvm_cpuid2 __user *cpuid_arg = argp;
1303 struct kvm_cpuid2 cpuid;
1306 if (copy_from_user(&cpuid, cpuid_arg, sizeof cpuid))
1308 r = kvm_vcpu_ioctl_get_cpuid2(vcpu, &cpuid,
1309 cpuid_arg->entries);
1313 if (copy_to_user(cpuid_arg, &cpuid, sizeof cpuid))
1319 r = msr_io(vcpu, argp, kvm_get_msr, 1);
1322 r = msr_io(vcpu, argp, do_set_msr, 0);
1324 case KVM_TPR_ACCESS_REPORTING: {
1325 struct kvm_tpr_access_ctl tac;
1328 if (copy_from_user(&tac, argp, sizeof tac))
1330 r = vcpu_ioctl_tpr_access_reporting(vcpu, &tac);
1334 if (copy_to_user(argp, &tac, sizeof tac))
1339 case KVM_SET_VAPIC_ADDR: {
1340 struct kvm_vapic_addr va;
1343 if (!irqchip_in_kernel(vcpu->kvm))
1346 if (copy_from_user(&va, argp, sizeof va))
1349 kvm_lapic_set_vapic_addr(vcpu, va.vapic_addr);
1359 static int kvm_vm_ioctl_set_tss_addr(struct kvm *kvm, unsigned long addr)
1363 if (addr > (unsigned int)(-3 * PAGE_SIZE))
1365 ret = kvm_x86_ops->set_tss_addr(kvm, addr);
1369 static int kvm_vm_ioctl_set_nr_mmu_pages(struct kvm *kvm,
1370 u32 kvm_nr_mmu_pages)
1372 if (kvm_nr_mmu_pages < KVM_MIN_ALLOC_MMU_PAGES)
1375 down_write(&kvm->slots_lock);
1377 kvm_mmu_change_mmu_pages(kvm, kvm_nr_mmu_pages);
1378 kvm->arch.n_requested_mmu_pages = kvm_nr_mmu_pages;
1380 up_write(&kvm->slots_lock);
1384 static int kvm_vm_ioctl_get_nr_mmu_pages(struct kvm *kvm)
1386 return kvm->arch.n_alloc_mmu_pages;
1389 gfn_t unalias_gfn(struct kvm *kvm, gfn_t gfn)
1392 struct kvm_mem_alias *alias;
1394 for (i = 0; i < kvm->arch.naliases; ++i) {
1395 alias = &kvm->arch.aliases[i];
1396 if (gfn >= alias->base_gfn
1397 && gfn < alias->base_gfn + alias->npages)
1398 return alias->target_gfn + gfn - alias->base_gfn;
1404 * Set a new alias region. Aliases map a portion of physical memory into
1405 * another portion. This is useful for memory windows, for example the PC
1408 static int kvm_vm_ioctl_set_memory_alias(struct kvm *kvm,
1409 struct kvm_memory_alias *alias)
1412 struct kvm_mem_alias *p;
1415 /* General sanity checks */
1416 if (alias->memory_size & (PAGE_SIZE - 1))
1418 if (alias->guest_phys_addr & (PAGE_SIZE - 1))
1420 if (alias->slot >= KVM_ALIAS_SLOTS)
1422 if (alias->guest_phys_addr + alias->memory_size
1423 < alias->guest_phys_addr)
1425 if (alias->target_phys_addr + alias->memory_size
1426 < alias->target_phys_addr)
1429 down_write(&kvm->slots_lock);
1431 p = &kvm->arch.aliases[alias->slot];
1432 p->base_gfn = alias->guest_phys_addr >> PAGE_SHIFT;
1433 p->npages = alias->memory_size >> PAGE_SHIFT;
1434 p->target_gfn = alias->target_phys_addr >> PAGE_SHIFT;
1436 for (n = KVM_ALIAS_SLOTS; n > 0; --n)
1437 if (kvm->arch.aliases[n - 1].npages)
1439 kvm->arch.naliases = n;
1441 kvm_mmu_zap_all(kvm);
1443 up_write(&kvm->slots_lock);
1451 static int kvm_vm_ioctl_get_irqchip(struct kvm *kvm, struct kvm_irqchip *chip)
1456 switch (chip->chip_id) {
1457 case KVM_IRQCHIP_PIC_MASTER:
1458 memcpy(&chip->chip.pic,
1459 &pic_irqchip(kvm)->pics[0],
1460 sizeof(struct kvm_pic_state));
1462 case KVM_IRQCHIP_PIC_SLAVE:
1463 memcpy(&chip->chip.pic,
1464 &pic_irqchip(kvm)->pics[1],
1465 sizeof(struct kvm_pic_state));
1467 case KVM_IRQCHIP_IOAPIC:
1468 memcpy(&chip->chip.ioapic,
1469 ioapic_irqchip(kvm),
1470 sizeof(struct kvm_ioapic_state));
1479 static int kvm_vm_ioctl_set_irqchip(struct kvm *kvm, struct kvm_irqchip *chip)
1484 switch (chip->chip_id) {
1485 case KVM_IRQCHIP_PIC_MASTER:
1486 memcpy(&pic_irqchip(kvm)->pics[0],
1488 sizeof(struct kvm_pic_state));
1490 case KVM_IRQCHIP_PIC_SLAVE:
1491 memcpy(&pic_irqchip(kvm)->pics[1],
1493 sizeof(struct kvm_pic_state));
1495 case KVM_IRQCHIP_IOAPIC:
1496 memcpy(ioapic_irqchip(kvm),
1498 sizeof(struct kvm_ioapic_state));
1504 kvm_pic_update_irq(pic_irqchip(kvm));
1508 static int kvm_vm_ioctl_get_pit(struct kvm *kvm, struct kvm_pit_state *ps)
1512 memcpy(ps, &kvm->arch.vpit->pit_state, sizeof(struct kvm_pit_state));
1516 static int kvm_vm_ioctl_set_pit(struct kvm *kvm, struct kvm_pit_state *ps)
1520 memcpy(&kvm->arch.vpit->pit_state, ps, sizeof(struct kvm_pit_state));
1521 kvm_pit_load_count(kvm, 0, ps->channels[0].count);
1526 * Get (and clear) the dirty memory log for a memory slot.
1528 int kvm_vm_ioctl_get_dirty_log(struct kvm *kvm,
1529 struct kvm_dirty_log *log)
1533 struct kvm_memory_slot *memslot;
1536 down_write(&kvm->slots_lock);
1538 r = kvm_get_dirty_log(kvm, log, &is_dirty);
1542 /* If nothing is dirty, don't bother messing with page tables. */
1544 kvm_mmu_slot_remove_write_access(kvm, log->slot);
1545 kvm_flush_remote_tlbs(kvm);
1546 memslot = &kvm->memslots[log->slot];
1547 n = ALIGN(memslot->npages, BITS_PER_LONG) / 8;
1548 memset(memslot->dirty_bitmap, 0, n);
1552 up_write(&kvm->slots_lock);
1556 long kvm_arch_vm_ioctl(struct file *filp,
1557 unsigned int ioctl, unsigned long arg)
1559 struct kvm *kvm = filp->private_data;
1560 void __user *argp = (void __user *)arg;
1564 case KVM_SET_TSS_ADDR:
1565 r = kvm_vm_ioctl_set_tss_addr(kvm, arg);
1569 case KVM_SET_MEMORY_REGION: {
1570 struct kvm_memory_region kvm_mem;
1571 struct kvm_userspace_memory_region kvm_userspace_mem;
1574 if (copy_from_user(&kvm_mem, argp, sizeof kvm_mem))
1576 kvm_userspace_mem.slot = kvm_mem.slot;
1577 kvm_userspace_mem.flags = kvm_mem.flags;
1578 kvm_userspace_mem.guest_phys_addr = kvm_mem.guest_phys_addr;
1579 kvm_userspace_mem.memory_size = kvm_mem.memory_size;
1580 r = kvm_vm_ioctl_set_memory_region(kvm, &kvm_userspace_mem, 0);
1585 case KVM_SET_NR_MMU_PAGES:
1586 r = kvm_vm_ioctl_set_nr_mmu_pages(kvm, arg);
1590 case KVM_GET_NR_MMU_PAGES:
1591 r = kvm_vm_ioctl_get_nr_mmu_pages(kvm);
1593 case KVM_SET_MEMORY_ALIAS: {
1594 struct kvm_memory_alias alias;
1597 if (copy_from_user(&alias, argp, sizeof alias))
1599 r = kvm_vm_ioctl_set_memory_alias(kvm, &alias);
1604 case KVM_CREATE_IRQCHIP:
1606 kvm->arch.vpic = kvm_create_pic(kvm);
1607 if (kvm->arch.vpic) {
1608 r = kvm_ioapic_init(kvm);
1610 kfree(kvm->arch.vpic);
1611 kvm->arch.vpic = NULL;
1617 case KVM_CREATE_PIT:
1619 kvm->arch.vpit = kvm_create_pit(kvm);
1623 case KVM_IRQ_LINE: {
1624 struct kvm_irq_level irq_event;
1627 if (copy_from_user(&irq_event, argp, sizeof irq_event))
1629 if (irqchip_in_kernel(kvm)) {
1630 mutex_lock(&kvm->lock);
1631 if (irq_event.irq < 16)
1632 kvm_pic_set_irq(pic_irqchip(kvm),
1635 kvm_ioapic_set_irq(kvm->arch.vioapic,
1638 mutex_unlock(&kvm->lock);
1643 case KVM_GET_IRQCHIP: {
1644 /* 0: PIC master, 1: PIC slave, 2: IOAPIC */
1645 struct kvm_irqchip chip;
1648 if (copy_from_user(&chip, argp, sizeof chip))
1651 if (!irqchip_in_kernel(kvm))
1653 r = kvm_vm_ioctl_get_irqchip(kvm, &chip);
1657 if (copy_to_user(argp, &chip, sizeof chip))
1662 case KVM_SET_IRQCHIP: {
1663 /* 0: PIC master, 1: PIC slave, 2: IOAPIC */
1664 struct kvm_irqchip chip;
1667 if (copy_from_user(&chip, argp, sizeof chip))
1670 if (!irqchip_in_kernel(kvm))
1672 r = kvm_vm_ioctl_set_irqchip(kvm, &chip);
1679 struct kvm_pit_state ps;
1681 if (copy_from_user(&ps, argp, sizeof ps))
1684 if (!kvm->arch.vpit)
1686 r = kvm_vm_ioctl_get_pit(kvm, &ps);
1690 if (copy_to_user(argp, &ps, sizeof ps))
1696 struct kvm_pit_state ps;
1698 if (copy_from_user(&ps, argp, sizeof ps))
1701 if (!kvm->arch.vpit)
1703 r = kvm_vm_ioctl_set_pit(kvm, &ps);
1716 static void kvm_init_msr_list(void)
1721 for (i = j = 0; i < ARRAY_SIZE(msrs_to_save); i++) {
1722 if (rdmsr_safe(msrs_to_save[i], &dummy[0], &dummy[1]) < 0)
1725 msrs_to_save[j] = msrs_to_save[i];
1728 num_msrs_to_save = j;
1732 * Only apic need an MMIO device hook, so shortcut now..
1734 static struct kvm_io_device *vcpu_find_pervcpu_dev(struct kvm_vcpu *vcpu,
1737 struct kvm_io_device *dev;
1739 if (vcpu->arch.apic) {
1740 dev = &vcpu->arch.apic->dev;
1741 if (dev->in_range(dev, addr))
1748 static struct kvm_io_device *vcpu_find_mmio_dev(struct kvm_vcpu *vcpu,
1751 struct kvm_io_device *dev;
1753 dev = vcpu_find_pervcpu_dev(vcpu, addr);
1755 dev = kvm_io_bus_find_dev(&vcpu->kvm->mmio_bus, addr);
1759 int emulator_read_std(unsigned long addr,
1762 struct kvm_vcpu *vcpu)
1765 int r = X86EMUL_CONTINUE;
1767 down_read(&vcpu->kvm->slots_lock);
1769 gpa_t gpa = vcpu->arch.mmu.gva_to_gpa(vcpu, addr);
1770 unsigned offset = addr & (PAGE_SIZE-1);
1771 unsigned tocopy = min(bytes, (unsigned)PAGE_SIZE - offset);
1774 if (gpa == UNMAPPED_GVA) {
1775 r = X86EMUL_PROPAGATE_FAULT;
1778 ret = kvm_read_guest(vcpu->kvm, gpa, data, tocopy);
1780 r = X86EMUL_UNHANDLEABLE;
1789 up_read(&vcpu->kvm->slots_lock);
1792 EXPORT_SYMBOL_GPL(emulator_read_std);
1794 static int emulator_read_emulated(unsigned long addr,
1797 struct kvm_vcpu *vcpu)
1799 struct kvm_io_device *mmio_dev;
1802 if (vcpu->mmio_read_completed) {
1803 memcpy(val, vcpu->mmio_data, bytes);
1804 vcpu->mmio_read_completed = 0;
1805 return X86EMUL_CONTINUE;
1808 down_read(&vcpu->kvm->slots_lock);
1809 gpa = vcpu->arch.mmu.gva_to_gpa(vcpu, addr);
1810 up_read(&vcpu->kvm->slots_lock);
1812 /* For APIC access vmexit */
1813 if ((gpa & PAGE_MASK) == APIC_DEFAULT_PHYS_BASE)
1816 if (emulator_read_std(addr, val, bytes, vcpu)
1817 == X86EMUL_CONTINUE)
1818 return X86EMUL_CONTINUE;
1819 if (gpa == UNMAPPED_GVA)
1820 return X86EMUL_PROPAGATE_FAULT;
1824 * Is this MMIO handled locally?
1826 mutex_lock(&vcpu->kvm->lock);
1827 mmio_dev = vcpu_find_mmio_dev(vcpu, gpa);
1829 kvm_iodevice_read(mmio_dev, gpa, bytes, val);
1830 mutex_unlock(&vcpu->kvm->lock);
1831 return X86EMUL_CONTINUE;
1833 mutex_unlock(&vcpu->kvm->lock);
1835 vcpu->mmio_needed = 1;
1836 vcpu->mmio_phys_addr = gpa;
1837 vcpu->mmio_size = bytes;
1838 vcpu->mmio_is_write = 0;
1840 return X86EMUL_UNHANDLEABLE;
1843 static int emulator_write_phys(struct kvm_vcpu *vcpu, gpa_t gpa,
1844 const void *val, int bytes)
1848 down_read(&vcpu->kvm->slots_lock);
1849 ret = kvm_write_guest(vcpu->kvm, gpa, val, bytes);
1851 up_read(&vcpu->kvm->slots_lock);
1854 kvm_mmu_pte_write(vcpu, gpa, val, bytes);
1855 up_read(&vcpu->kvm->slots_lock);
1859 static int emulator_write_emulated_onepage(unsigned long addr,
1862 struct kvm_vcpu *vcpu)
1864 struct kvm_io_device *mmio_dev;
1867 down_read(&vcpu->kvm->slots_lock);
1868 gpa = vcpu->arch.mmu.gva_to_gpa(vcpu, addr);
1869 up_read(&vcpu->kvm->slots_lock);
1871 if (gpa == UNMAPPED_GVA) {
1872 kvm_inject_page_fault(vcpu, addr, 2);
1873 return X86EMUL_PROPAGATE_FAULT;
1876 /* For APIC access vmexit */
1877 if ((gpa & PAGE_MASK) == APIC_DEFAULT_PHYS_BASE)
1880 if (emulator_write_phys(vcpu, gpa, val, bytes))
1881 return X86EMUL_CONTINUE;
1885 * Is this MMIO handled locally?
1887 mutex_lock(&vcpu->kvm->lock);
1888 mmio_dev = vcpu_find_mmio_dev(vcpu, gpa);
1890 kvm_iodevice_write(mmio_dev, gpa, bytes, val);
1891 mutex_unlock(&vcpu->kvm->lock);
1892 return X86EMUL_CONTINUE;
1894 mutex_unlock(&vcpu->kvm->lock);
1896 vcpu->mmio_needed = 1;
1897 vcpu->mmio_phys_addr = gpa;
1898 vcpu->mmio_size = bytes;
1899 vcpu->mmio_is_write = 1;
1900 memcpy(vcpu->mmio_data, val, bytes);
1902 return X86EMUL_CONTINUE;
1905 int emulator_write_emulated(unsigned long addr,
1908 struct kvm_vcpu *vcpu)
1910 /* Crossing a page boundary? */
1911 if (((addr + bytes - 1) ^ addr) & PAGE_MASK) {
1914 now = -addr & ~PAGE_MASK;
1915 rc = emulator_write_emulated_onepage(addr, val, now, vcpu);
1916 if (rc != X86EMUL_CONTINUE)
1922 return emulator_write_emulated_onepage(addr, val, bytes, vcpu);
1924 EXPORT_SYMBOL_GPL(emulator_write_emulated);
1926 static int emulator_cmpxchg_emulated(unsigned long addr,
1930 struct kvm_vcpu *vcpu)
1932 static int reported;
1936 printk(KERN_WARNING "kvm: emulating exchange as write\n");
1938 #ifndef CONFIG_X86_64
1939 /* guests cmpxchg8b have to be emulated atomically */
1946 down_read(&vcpu->kvm->slots_lock);
1947 gpa = vcpu->arch.mmu.gva_to_gpa(vcpu, addr);
1949 if (gpa == UNMAPPED_GVA ||
1950 (gpa & PAGE_MASK) == APIC_DEFAULT_PHYS_BASE)
1953 if (((gpa + bytes - 1) & PAGE_MASK) != (gpa & PAGE_MASK))
1958 down_read(¤t->mm->mmap_sem);
1959 page = gfn_to_page(vcpu->kvm, gpa >> PAGE_SHIFT);
1960 up_read(¤t->mm->mmap_sem);
1962 kaddr = kmap_atomic(page, KM_USER0);
1963 set_64bit((u64 *)(kaddr + offset_in_page(gpa)), val);
1964 kunmap_atomic(kaddr, KM_USER0);
1965 kvm_release_page_dirty(page);
1967 up_read(&vcpu->kvm->slots_lock);
1971 return emulator_write_emulated(addr, new, bytes, vcpu);
1974 static unsigned long get_segment_base(struct kvm_vcpu *vcpu, int seg)
1976 return kvm_x86_ops->get_segment_base(vcpu, seg);
1979 int emulate_invlpg(struct kvm_vcpu *vcpu, gva_t address)
1981 return X86EMUL_CONTINUE;
1984 int emulate_clts(struct kvm_vcpu *vcpu)
1986 kvm_x86_ops->set_cr0(vcpu, vcpu->arch.cr0 & ~X86_CR0_TS);
1987 return X86EMUL_CONTINUE;
1990 int emulator_get_dr(struct x86_emulate_ctxt *ctxt, int dr, unsigned long *dest)
1992 struct kvm_vcpu *vcpu = ctxt->vcpu;
1996 *dest = kvm_x86_ops->get_dr(vcpu, dr);
1997 return X86EMUL_CONTINUE;
1999 pr_unimpl(vcpu, "%s: unexpected dr %u\n", __func__, dr);
2000 return X86EMUL_UNHANDLEABLE;
2004 int emulator_set_dr(struct x86_emulate_ctxt *ctxt, int dr, unsigned long value)
2006 unsigned long mask = (ctxt->mode == X86EMUL_MODE_PROT64) ? ~0ULL : ~0U;
2009 kvm_x86_ops->set_dr(ctxt->vcpu, dr, value & mask, &exception);
2011 /* FIXME: better handling */
2012 return X86EMUL_UNHANDLEABLE;
2014 return X86EMUL_CONTINUE;
2017 void kvm_report_emulation_failure(struct kvm_vcpu *vcpu, const char *context)
2019 static int reported;
2021 unsigned long rip = vcpu->arch.rip;
2022 unsigned long rip_linear;
2024 rip_linear = rip + get_segment_base(vcpu, VCPU_SREG_CS);
2029 emulator_read_std(rip_linear, (void *)opcodes, 4, vcpu);
2031 printk(KERN_ERR "emulation failed (%s) rip %lx %02x %02x %02x %02x\n",
2032 context, rip, opcodes[0], opcodes[1], opcodes[2], opcodes[3]);
2035 EXPORT_SYMBOL_GPL(kvm_report_emulation_failure);
2037 static struct x86_emulate_ops emulate_ops = {
2038 .read_std = emulator_read_std,
2039 .read_emulated = emulator_read_emulated,
2040 .write_emulated = emulator_write_emulated,
2041 .cmpxchg_emulated = emulator_cmpxchg_emulated,
2044 int emulate_instruction(struct kvm_vcpu *vcpu,
2045 struct kvm_run *run,
2051 struct decode_cache *c;
2053 vcpu->arch.mmio_fault_cr2 = cr2;
2054 kvm_x86_ops->cache_regs(vcpu);
2056 vcpu->mmio_is_write = 0;
2057 vcpu->arch.pio.string = 0;
2059 if (!(emulation_type & EMULTYPE_NO_DECODE)) {
2061 kvm_x86_ops->get_cs_db_l_bits(vcpu, &cs_db, &cs_l);
2063 vcpu->arch.emulate_ctxt.vcpu = vcpu;
2064 vcpu->arch.emulate_ctxt.eflags = kvm_x86_ops->get_rflags(vcpu);
2065 vcpu->arch.emulate_ctxt.mode =
2066 (vcpu->arch.emulate_ctxt.eflags & X86_EFLAGS_VM)
2067 ? X86EMUL_MODE_REAL : cs_l
2068 ? X86EMUL_MODE_PROT64 : cs_db
2069 ? X86EMUL_MODE_PROT32 : X86EMUL_MODE_PROT16;
2071 if (vcpu->arch.emulate_ctxt.mode == X86EMUL_MODE_PROT64) {
2072 vcpu->arch.emulate_ctxt.cs_base = 0;
2073 vcpu->arch.emulate_ctxt.ds_base = 0;
2074 vcpu->arch.emulate_ctxt.es_base = 0;
2075 vcpu->arch.emulate_ctxt.ss_base = 0;
2077 vcpu->arch.emulate_ctxt.cs_base =
2078 get_segment_base(vcpu, VCPU_SREG_CS);
2079 vcpu->arch.emulate_ctxt.ds_base =
2080 get_segment_base(vcpu, VCPU_SREG_DS);
2081 vcpu->arch.emulate_ctxt.es_base =
2082 get_segment_base(vcpu, VCPU_SREG_ES);
2083 vcpu->arch.emulate_ctxt.ss_base =
2084 get_segment_base(vcpu, VCPU_SREG_SS);
2087 vcpu->arch.emulate_ctxt.gs_base =
2088 get_segment_base(vcpu, VCPU_SREG_GS);
2089 vcpu->arch.emulate_ctxt.fs_base =
2090 get_segment_base(vcpu, VCPU_SREG_FS);
2092 r = x86_decode_insn(&vcpu->arch.emulate_ctxt, &emulate_ops);
2094 /* Reject the instructions other than VMCALL/VMMCALL when
2095 * try to emulate invalid opcode */
2096 c = &vcpu->arch.emulate_ctxt.decode;
2097 if ((emulation_type & EMULTYPE_TRAP_UD) &&
2098 (!(c->twobyte && c->b == 0x01 &&
2099 (c->modrm_reg == 0 || c->modrm_reg == 3) &&
2100 c->modrm_mod == 3 && c->modrm_rm == 1)))
2101 return EMULATE_FAIL;
2103 ++vcpu->stat.insn_emulation;
2105 ++vcpu->stat.insn_emulation_fail;
2106 if (kvm_mmu_unprotect_page_virt(vcpu, cr2))
2107 return EMULATE_DONE;
2108 return EMULATE_FAIL;
2112 r = x86_emulate_insn(&vcpu->arch.emulate_ctxt, &emulate_ops);
2114 if (vcpu->arch.pio.string)
2115 return EMULATE_DO_MMIO;
2117 if ((r || vcpu->mmio_is_write) && run) {
2118 run->exit_reason = KVM_EXIT_MMIO;
2119 run->mmio.phys_addr = vcpu->mmio_phys_addr;
2120 memcpy(run->mmio.data, vcpu->mmio_data, 8);
2121 run->mmio.len = vcpu->mmio_size;
2122 run->mmio.is_write = vcpu->mmio_is_write;
2126 if (kvm_mmu_unprotect_page_virt(vcpu, cr2))
2127 return EMULATE_DONE;
2128 if (!vcpu->mmio_needed) {
2129 kvm_report_emulation_failure(vcpu, "mmio");
2130 return EMULATE_FAIL;
2132 return EMULATE_DO_MMIO;
2135 kvm_x86_ops->decache_regs(vcpu);
2136 kvm_x86_ops->set_rflags(vcpu, vcpu->arch.emulate_ctxt.eflags);
2138 if (vcpu->mmio_is_write) {
2139 vcpu->mmio_needed = 0;
2140 return EMULATE_DO_MMIO;
2143 return EMULATE_DONE;
2145 EXPORT_SYMBOL_GPL(emulate_instruction);
2147 static void free_pio_guest_pages(struct kvm_vcpu *vcpu)
2151 for (i = 0; i < ARRAY_SIZE(vcpu->arch.pio.guest_pages); ++i)
2152 if (vcpu->arch.pio.guest_pages[i]) {
2153 kvm_release_page_dirty(vcpu->arch.pio.guest_pages[i]);
2154 vcpu->arch.pio.guest_pages[i] = NULL;
2158 static int pio_copy_data(struct kvm_vcpu *vcpu)
2160 void *p = vcpu->arch.pio_data;
2163 int nr_pages = vcpu->arch.pio.guest_pages[1] ? 2 : 1;
2165 q = vmap(vcpu->arch.pio.guest_pages, nr_pages, VM_READ|VM_WRITE,
2168 free_pio_guest_pages(vcpu);
2171 q += vcpu->arch.pio.guest_page_offset;
2172 bytes = vcpu->arch.pio.size * vcpu->arch.pio.cur_count;
2173 if (vcpu->arch.pio.in)
2174 memcpy(q, p, bytes);
2176 memcpy(p, q, bytes);
2177 q -= vcpu->arch.pio.guest_page_offset;
2179 free_pio_guest_pages(vcpu);
2183 int complete_pio(struct kvm_vcpu *vcpu)
2185 struct kvm_pio_request *io = &vcpu->arch.pio;
2189 kvm_x86_ops->cache_regs(vcpu);
2193 memcpy(&vcpu->arch.regs[VCPU_REGS_RAX], vcpu->arch.pio_data,
2197 r = pio_copy_data(vcpu);
2199 kvm_x86_ops->cache_regs(vcpu);
2206 delta *= io->cur_count;
2208 * The size of the register should really depend on
2209 * current address size.
2211 vcpu->arch.regs[VCPU_REGS_RCX] -= delta;
2217 vcpu->arch.regs[VCPU_REGS_RDI] += delta;
2219 vcpu->arch.regs[VCPU_REGS_RSI] += delta;
2222 kvm_x86_ops->decache_regs(vcpu);
2224 io->count -= io->cur_count;
2230 static void kernel_pio(struct kvm_io_device *pio_dev,
2231 struct kvm_vcpu *vcpu,
2234 /* TODO: String I/O for in kernel device */
2236 mutex_lock(&vcpu->kvm->lock);
2237 if (vcpu->arch.pio.in)
2238 kvm_iodevice_read(pio_dev, vcpu->arch.pio.port,
2239 vcpu->arch.pio.size,
2242 kvm_iodevice_write(pio_dev, vcpu->arch.pio.port,
2243 vcpu->arch.pio.size,
2245 mutex_unlock(&vcpu->kvm->lock);
2248 static void pio_string_write(struct kvm_io_device *pio_dev,
2249 struct kvm_vcpu *vcpu)
2251 struct kvm_pio_request *io = &vcpu->arch.pio;
2252 void *pd = vcpu->arch.pio_data;
2255 mutex_lock(&vcpu->kvm->lock);
2256 for (i = 0; i < io->cur_count; i++) {
2257 kvm_iodevice_write(pio_dev, io->port,
2262 mutex_unlock(&vcpu->kvm->lock);
2265 static struct kvm_io_device *vcpu_find_pio_dev(struct kvm_vcpu *vcpu,
2268 return kvm_io_bus_find_dev(&vcpu->kvm->pio_bus, addr);
2271 int kvm_emulate_pio(struct kvm_vcpu *vcpu, struct kvm_run *run, int in,
2272 int size, unsigned port)
2274 struct kvm_io_device *pio_dev;
2276 vcpu->run->exit_reason = KVM_EXIT_IO;
2277 vcpu->run->io.direction = in ? KVM_EXIT_IO_IN : KVM_EXIT_IO_OUT;
2278 vcpu->run->io.size = vcpu->arch.pio.size = size;
2279 vcpu->run->io.data_offset = KVM_PIO_PAGE_OFFSET * PAGE_SIZE;
2280 vcpu->run->io.count = vcpu->arch.pio.count = vcpu->arch.pio.cur_count = 1;
2281 vcpu->run->io.port = vcpu->arch.pio.port = port;
2282 vcpu->arch.pio.in = in;
2283 vcpu->arch.pio.string = 0;
2284 vcpu->arch.pio.down = 0;
2285 vcpu->arch.pio.guest_page_offset = 0;
2286 vcpu->arch.pio.rep = 0;
2288 kvm_x86_ops->cache_regs(vcpu);
2289 memcpy(vcpu->arch.pio_data, &vcpu->arch.regs[VCPU_REGS_RAX], 4);
2290 kvm_x86_ops->decache_regs(vcpu);
2292 kvm_x86_ops->skip_emulated_instruction(vcpu);
2294 pio_dev = vcpu_find_pio_dev(vcpu, port);
2296 kernel_pio(pio_dev, vcpu, vcpu->arch.pio_data);
2302 EXPORT_SYMBOL_GPL(kvm_emulate_pio);
2304 int kvm_emulate_pio_string(struct kvm_vcpu *vcpu, struct kvm_run *run, int in,
2305 int size, unsigned long count, int down,
2306 gva_t address, int rep, unsigned port)
2308 unsigned now, in_page;
2312 struct kvm_io_device *pio_dev;
2314 vcpu->run->exit_reason = KVM_EXIT_IO;
2315 vcpu->run->io.direction = in ? KVM_EXIT_IO_IN : KVM_EXIT_IO_OUT;
2316 vcpu->run->io.size = vcpu->arch.pio.size = size;
2317 vcpu->run->io.data_offset = KVM_PIO_PAGE_OFFSET * PAGE_SIZE;
2318 vcpu->run->io.count = vcpu->arch.pio.count = vcpu->arch.pio.cur_count = count;
2319 vcpu->run->io.port = vcpu->arch.pio.port = port;
2320 vcpu->arch.pio.in = in;
2321 vcpu->arch.pio.string = 1;
2322 vcpu->arch.pio.down = down;
2323 vcpu->arch.pio.guest_page_offset = offset_in_page(address);
2324 vcpu->arch.pio.rep = rep;
2327 kvm_x86_ops->skip_emulated_instruction(vcpu);
2332 in_page = PAGE_SIZE - offset_in_page(address);
2334 in_page = offset_in_page(address) + size;
2335 now = min(count, (unsigned long)in_page / size);
2338 * String I/O straddles page boundary. Pin two guest pages
2339 * so that we satisfy atomicity constraints. Do just one
2340 * transaction to avoid complexity.
2347 * String I/O in reverse. Yuck. Kill the guest, fix later.
2349 pr_unimpl(vcpu, "guest string pio down\n");
2350 kvm_inject_gp(vcpu, 0);
2353 vcpu->run->io.count = now;
2354 vcpu->arch.pio.cur_count = now;
2356 if (vcpu->arch.pio.cur_count == vcpu->arch.pio.count)
2357 kvm_x86_ops->skip_emulated_instruction(vcpu);
2359 for (i = 0; i < nr_pages; ++i) {
2360 down_read(&vcpu->kvm->slots_lock);
2361 page = gva_to_page(vcpu, address + i * PAGE_SIZE);
2362 vcpu->arch.pio.guest_pages[i] = page;
2363 up_read(&vcpu->kvm->slots_lock);
2365 kvm_inject_gp(vcpu, 0);
2366 free_pio_guest_pages(vcpu);
2371 pio_dev = vcpu_find_pio_dev(vcpu, port);
2372 if (!vcpu->arch.pio.in) {
2373 /* string PIO write */
2374 ret = pio_copy_data(vcpu);
2375 if (ret >= 0 && pio_dev) {
2376 pio_string_write(pio_dev, vcpu);
2378 if (vcpu->arch.pio.count == 0)
2382 pr_unimpl(vcpu, "no string pio read support yet, "
2383 "port %x size %d count %ld\n",
2388 EXPORT_SYMBOL_GPL(kvm_emulate_pio_string);
2390 int kvm_arch_init(void *opaque)
2393 struct kvm_x86_ops *ops = (struct kvm_x86_ops *)opaque;
2396 printk(KERN_ERR "kvm: already loaded the other module\n");
2401 if (!ops->cpu_has_kvm_support()) {
2402 printk(KERN_ERR "kvm: no hardware support\n");
2406 if (ops->disabled_by_bios()) {
2407 printk(KERN_ERR "kvm: disabled by bios\n");
2412 r = kvm_mmu_module_init();
2416 kvm_init_msr_list();
2419 kvm_mmu_set_nonpresent_ptes(0ull, 0ull);
2426 void kvm_arch_exit(void)
2429 kvm_mmu_module_exit();
2432 int kvm_emulate_halt(struct kvm_vcpu *vcpu)
2434 ++vcpu->stat.halt_exits;
2435 if (irqchip_in_kernel(vcpu->kvm)) {
2436 vcpu->arch.mp_state = VCPU_MP_STATE_HALTED;
2437 kvm_vcpu_block(vcpu);
2438 if (vcpu->arch.mp_state != VCPU_MP_STATE_RUNNABLE)
2442 vcpu->run->exit_reason = KVM_EXIT_HLT;
2446 EXPORT_SYMBOL_GPL(kvm_emulate_halt);
2448 int kvm_emulate_hypercall(struct kvm_vcpu *vcpu)
2450 unsigned long nr, a0, a1, a2, a3, ret;
2452 kvm_x86_ops->cache_regs(vcpu);
2454 nr = vcpu->arch.regs[VCPU_REGS_RAX];
2455 a0 = vcpu->arch.regs[VCPU_REGS_RBX];
2456 a1 = vcpu->arch.regs[VCPU_REGS_RCX];
2457 a2 = vcpu->arch.regs[VCPU_REGS_RDX];
2458 a3 = vcpu->arch.regs[VCPU_REGS_RSI];
2460 if (!is_long_mode(vcpu)) {
2469 case KVM_HC_VAPIC_POLL_IRQ:
2476 vcpu->arch.regs[VCPU_REGS_RAX] = ret;
2477 kvm_x86_ops->decache_regs(vcpu);
2478 ++vcpu->stat.hypercalls;
2481 EXPORT_SYMBOL_GPL(kvm_emulate_hypercall);
2483 int kvm_fix_hypercall(struct kvm_vcpu *vcpu)
2485 char instruction[3];
2490 * Blow out the MMU to ensure that no other VCPU has an active mapping
2491 * to ensure that the updated hypercall appears atomically across all
2494 kvm_mmu_zap_all(vcpu->kvm);
2496 kvm_x86_ops->cache_regs(vcpu);
2497 kvm_x86_ops->patch_hypercall(vcpu, instruction);
2498 if (emulator_write_emulated(vcpu->arch.rip, instruction, 3, vcpu)
2499 != X86EMUL_CONTINUE)
2505 static u64 mk_cr_64(u64 curr_cr, u32 new_val)
2507 return (curr_cr & ~((1ULL << 32) - 1)) | new_val;
2510 void realmode_lgdt(struct kvm_vcpu *vcpu, u16 limit, unsigned long base)
2512 struct descriptor_table dt = { limit, base };
2514 kvm_x86_ops->set_gdt(vcpu, &dt);
2517 void realmode_lidt(struct kvm_vcpu *vcpu, u16 limit, unsigned long base)
2519 struct descriptor_table dt = { limit, base };
2521 kvm_x86_ops->set_idt(vcpu, &dt);
2524 void realmode_lmsw(struct kvm_vcpu *vcpu, unsigned long msw,
2525 unsigned long *rflags)
2527 kvm_lmsw(vcpu, msw);
2528 *rflags = kvm_x86_ops->get_rflags(vcpu);
2531 unsigned long realmode_get_cr(struct kvm_vcpu *vcpu, int cr)
2533 kvm_x86_ops->decache_cr4_guest_bits(vcpu);
2536 return vcpu->arch.cr0;
2538 return vcpu->arch.cr2;
2540 return vcpu->arch.cr3;
2542 return vcpu->arch.cr4;
2544 return kvm_get_cr8(vcpu);
2546 vcpu_printf(vcpu, "%s: unexpected cr %u\n", __func__, cr);
2551 void realmode_set_cr(struct kvm_vcpu *vcpu, int cr, unsigned long val,
2552 unsigned long *rflags)
2556 kvm_set_cr0(vcpu, mk_cr_64(vcpu->arch.cr0, val));
2557 *rflags = kvm_x86_ops->get_rflags(vcpu);
2560 vcpu->arch.cr2 = val;
2563 kvm_set_cr3(vcpu, val);
2566 kvm_set_cr4(vcpu, mk_cr_64(vcpu->arch.cr4, val));
2569 kvm_set_cr8(vcpu, val & 0xfUL);
2572 vcpu_printf(vcpu, "%s: unexpected cr %u\n", __func__, cr);
2576 static int move_to_next_stateful_cpuid_entry(struct kvm_vcpu *vcpu, int i)
2578 struct kvm_cpuid_entry2 *e = &vcpu->arch.cpuid_entries[i];
2579 int j, nent = vcpu->arch.cpuid_nent;
2581 e->flags &= ~KVM_CPUID_FLAG_STATE_READ_NEXT;
2582 /* when no next entry is found, the current entry[i] is reselected */
2583 for (j = i + 1; j == i; j = (j + 1) % nent) {
2584 struct kvm_cpuid_entry2 *ej = &vcpu->arch.cpuid_entries[j];
2585 if (ej->function == e->function) {
2586 ej->flags |= KVM_CPUID_FLAG_STATE_READ_NEXT;
2590 return 0; /* silence gcc, even though control never reaches here */
2593 /* find an entry with matching function, matching index (if needed), and that
2594 * should be read next (if it's stateful) */
2595 static int is_matching_cpuid_entry(struct kvm_cpuid_entry2 *e,
2596 u32 function, u32 index)
2598 if (e->function != function)
2600 if ((e->flags & KVM_CPUID_FLAG_SIGNIFCANT_INDEX) && e->index != index)
2602 if ((e->flags & KVM_CPUID_FLAG_STATEFUL_FUNC) &&
2603 !(e->flags & KVM_CPUID_FLAG_STATE_READ_NEXT))
2608 void kvm_emulate_cpuid(struct kvm_vcpu *vcpu)
2611 u32 function, index;
2612 struct kvm_cpuid_entry2 *e, *best;
2614 kvm_x86_ops->cache_regs(vcpu);
2615 function = vcpu->arch.regs[VCPU_REGS_RAX];
2616 index = vcpu->arch.regs[VCPU_REGS_RCX];
2617 vcpu->arch.regs[VCPU_REGS_RAX] = 0;
2618 vcpu->arch.regs[VCPU_REGS_RBX] = 0;
2619 vcpu->arch.regs[VCPU_REGS_RCX] = 0;
2620 vcpu->arch.regs[VCPU_REGS_RDX] = 0;
2622 for (i = 0; i < vcpu->arch.cpuid_nent; ++i) {
2623 e = &vcpu->arch.cpuid_entries[i];
2624 if (is_matching_cpuid_entry(e, function, index)) {
2625 if (e->flags & KVM_CPUID_FLAG_STATEFUL_FUNC)
2626 move_to_next_stateful_cpuid_entry(vcpu, i);
2631 * Both basic or both extended?
2633 if (((e->function ^ function) & 0x80000000) == 0)
2634 if (!best || e->function > best->function)
2638 vcpu->arch.regs[VCPU_REGS_RAX] = best->eax;
2639 vcpu->arch.regs[VCPU_REGS_RBX] = best->ebx;
2640 vcpu->arch.regs[VCPU_REGS_RCX] = best->ecx;
2641 vcpu->arch.regs[VCPU_REGS_RDX] = best->edx;
2643 kvm_x86_ops->decache_regs(vcpu);
2644 kvm_x86_ops->skip_emulated_instruction(vcpu);
2646 EXPORT_SYMBOL_GPL(kvm_emulate_cpuid);
2649 * Check if userspace requested an interrupt window, and that the
2650 * interrupt window is open.
2652 * No need to exit to userspace if we already have an interrupt queued.
2654 static int dm_request_for_irq_injection(struct kvm_vcpu *vcpu,
2655 struct kvm_run *kvm_run)
2657 return (!vcpu->arch.irq_summary &&
2658 kvm_run->request_interrupt_window &&
2659 vcpu->arch.interrupt_window_open &&
2660 (kvm_x86_ops->get_rflags(vcpu) & X86_EFLAGS_IF));
2663 static void post_kvm_run_save(struct kvm_vcpu *vcpu,
2664 struct kvm_run *kvm_run)
2666 kvm_run->if_flag = (kvm_x86_ops->get_rflags(vcpu) & X86_EFLAGS_IF) != 0;
2667 kvm_run->cr8 = kvm_get_cr8(vcpu);
2668 kvm_run->apic_base = kvm_get_apic_base(vcpu);
2669 if (irqchip_in_kernel(vcpu->kvm))
2670 kvm_run->ready_for_interrupt_injection = 1;
2672 kvm_run->ready_for_interrupt_injection =
2673 (vcpu->arch.interrupt_window_open &&
2674 vcpu->arch.irq_summary == 0);
2677 static void vapic_enter(struct kvm_vcpu *vcpu)
2679 struct kvm_lapic *apic = vcpu->arch.apic;
2682 if (!apic || !apic->vapic_addr)
2685 down_read(¤t->mm->mmap_sem);
2686 page = gfn_to_page(vcpu->kvm, apic->vapic_addr >> PAGE_SHIFT);
2687 up_read(¤t->mm->mmap_sem);
2689 vcpu->arch.apic->vapic_page = page;
2692 static void vapic_exit(struct kvm_vcpu *vcpu)
2694 struct kvm_lapic *apic = vcpu->arch.apic;
2696 if (!apic || !apic->vapic_addr)
2699 kvm_release_page_dirty(apic->vapic_page);
2700 mark_page_dirty(vcpu->kvm, apic->vapic_addr >> PAGE_SHIFT);
2703 static int __vcpu_run(struct kvm_vcpu *vcpu, struct kvm_run *kvm_run)
2707 if (unlikely(vcpu->arch.mp_state == VCPU_MP_STATE_SIPI_RECEIVED)) {
2708 pr_debug("vcpu %d received sipi with vector # %x\n",
2709 vcpu->vcpu_id, vcpu->arch.sipi_vector);
2710 kvm_lapic_reset(vcpu);
2711 r = kvm_x86_ops->vcpu_reset(vcpu);
2714 vcpu->arch.mp_state = VCPU_MP_STATE_RUNNABLE;
2720 if (vcpu->guest_debug.enabled)
2721 kvm_x86_ops->guest_debug_pre(vcpu);
2725 if (test_and_clear_bit(KVM_REQ_MMU_RELOAD, &vcpu->requests))
2726 kvm_mmu_unload(vcpu);
2728 r = kvm_mmu_reload(vcpu);
2732 if (vcpu->requests) {
2733 if (test_and_clear_bit(KVM_REQ_MIGRATE_TIMER, &vcpu->requests))
2734 __kvm_migrate_apic_timer(vcpu);
2735 if (test_and_clear_bit(KVM_REQ_REPORT_TPR_ACCESS,
2737 kvm_run->exit_reason = KVM_EXIT_TPR_ACCESS;
2741 if (test_and_clear_bit(KVM_REQ_TRIPLE_FAULT, &vcpu->requests)) {
2742 kvm_run->exit_reason = KVM_EXIT_SHUTDOWN;
2748 kvm_inject_pending_timer_irqs(vcpu);
2752 kvm_x86_ops->prepare_guest_switch(vcpu);
2753 kvm_load_guest_fpu(vcpu);
2755 local_irq_disable();
2757 if (need_resched()) {
2765 if (test_bit(KVM_REQ_MMU_RELOAD, &vcpu->requests)) {
2772 if (signal_pending(current)) {
2776 kvm_run->exit_reason = KVM_EXIT_INTR;
2777 ++vcpu->stat.signal_exits;
2781 if (vcpu->arch.exception.pending)
2782 __queue_exception(vcpu);
2783 else if (irqchip_in_kernel(vcpu->kvm))
2784 kvm_x86_ops->inject_pending_irq(vcpu);
2786 kvm_x86_ops->inject_pending_vectors(vcpu, kvm_run);
2788 kvm_lapic_sync_to_vapic(vcpu);
2790 vcpu->guest_mode = 1;
2794 if (test_and_clear_bit(KVM_REQ_TLB_FLUSH, &vcpu->requests))
2795 kvm_x86_ops->tlb_flush(vcpu);
2797 kvm_x86_ops->run(vcpu, kvm_run);
2799 vcpu->guest_mode = 0;
2805 * We must have an instruction between local_irq_enable() and
2806 * kvm_guest_exit(), so the timer interrupt isn't delayed by
2807 * the interrupt shadow. The stat.exits increment will do nicely.
2808 * But we need to prevent reordering, hence this barrier():
2817 * Profile KVM exit RIPs:
2819 if (unlikely(prof_on == KVM_PROFILING)) {
2820 kvm_x86_ops->cache_regs(vcpu);
2821 profile_hit(KVM_PROFILING, (void *)vcpu->arch.rip);
2824 if (vcpu->arch.exception.pending && kvm_x86_ops->exception_injected(vcpu))
2825 vcpu->arch.exception.pending = false;
2827 kvm_lapic_sync_from_vapic(vcpu);
2829 r = kvm_x86_ops->handle_exit(kvm_run, vcpu);
2832 if (dm_request_for_irq_injection(vcpu, kvm_run)) {
2834 kvm_run->exit_reason = KVM_EXIT_INTR;
2835 ++vcpu->stat.request_irq_exits;
2838 if (!need_resched())
2848 post_kvm_run_save(vcpu, kvm_run);
2855 int kvm_arch_vcpu_ioctl_run(struct kvm_vcpu *vcpu, struct kvm_run *kvm_run)
2862 if (unlikely(vcpu->arch.mp_state == VCPU_MP_STATE_UNINITIALIZED)) {
2863 kvm_vcpu_block(vcpu);
2868 if (vcpu->sigset_active)
2869 sigprocmask(SIG_SETMASK, &vcpu->sigset, &sigsaved);
2871 /* re-sync apic's tpr */
2872 if (!irqchip_in_kernel(vcpu->kvm))
2873 kvm_set_cr8(vcpu, kvm_run->cr8);
2875 if (vcpu->arch.pio.cur_count) {
2876 r = complete_pio(vcpu);
2880 #if CONFIG_HAS_IOMEM
2881 if (vcpu->mmio_needed) {
2882 memcpy(vcpu->mmio_data, kvm_run->mmio.data, 8);
2883 vcpu->mmio_read_completed = 1;
2884 vcpu->mmio_needed = 0;
2885 r = emulate_instruction(vcpu, kvm_run,
2886 vcpu->arch.mmio_fault_cr2, 0,
2887 EMULTYPE_NO_DECODE);
2888 if (r == EMULATE_DO_MMIO) {
2890 * Read-modify-write. Back to userspace.
2897 if (kvm_run->exit_reason == KVM_EXIT_HYPERCALL) {
2898 kvm_x86_ops->cache_regs(vcpu);
2899 vcpu->arch.regs[VCPU_REGS_RAX] = kvm_run->hypercall.ret;
2900 kvm_x86_ops->decache_regs(vcpu);
2903 r = __vcpu_run(vcpu, kvm_run);
2906 if (vcpu->sigset_active)
2907 sigprocmask(SIG_SETMASK, &sigsaved, NULL);
2913 int kvm_arch_vcpu_ioctl_get_regs(struct kvm_vcpu *vcpu, struct kvm_regs *regs)
2917 kvm_x86_ops->cache_regs(vcpu);
2919 regs->rax = vcpu->arch.regs[VCPU_REGS_RAX];
2920 regs->rbx = vcpu->arch.regs[VCPU_REGS_RBX];
2921 regs->rcx = vcpu->arch.regs[VCPU_REGS_RCX];
2922 regs->rdx = vcpu->arch.regs[VCPU_REGS_RDX];
2923 regs->rsi = vcpu->arch.regs[VCPU_REGS_RSI];
2924 regs->rdi = vcpu->arch.regs[VCPU_REGS_RDI];
2925 regs->rsp = vcpu->arch.regs[VCPU_REGS_RSP];
2926 regs->rbp = vcpu->arch.regs[VCPU_REGS_RBP];
2927 #ifdef CONFIG_X86_64
2928 regs->r8 = vcpu->arch.regs[VCPU_REGS_R8];
2929 regs->r9 = vcpu->arch.regs[VCPU_REGS_R9];
2930 regs->r10 = vcpu->arch.regs[VCPU_REGS_R10];
2931 regs->r11 = vcpu->arch.regs[VCPU_REGS_R11];
2932 regs->r12 = vcpu->arch.regs[VCPU_REGS_R12];
2933 regs->r13 = vcpu->arch.regs[VCPU_REGS_R13];
2934 regs->r14 = vcpu->arch.regs[VCPU_REGS_R14];
2935 regs->r15 = vcpu->arch.regs[VCPU_REGS_R15];
2938 regs->rip = vcpu->arch.rip;
2939 regs->rflags = kvm_x86_ops->get_rflags(vcpu);
2942 * Don't leak debug flags in case they were set for guest debugging
2944 if (vcpu->guest_debug.enabled && vcpu->guest_debug.singlestep)
2945 regs->rflags &= ~(X86_EFLAGS_TF | X86_EFLAGS_RF);
2952 int kvm_arch_vcpu_ioctl_set_regs(struct kvm_vcpu *vcpu, struct kvm_regs *regs)
2956 vcpu->arch.regs[VCPU_REGS_RAX] = regs->rax;
2957 vcpu->arch.regs[VCPU_REGS_RBX] = regs->rbx;
2958 vcpu->arch.regs[VCPU_REGS_RCX] = regs->rcx;
2959 vcpu->arch.regs[VCPU_REGS_RDX] = regs->rdx;
2960 vcpu->arch.regs[VCPU_REGS_RSI] = regs->rsi;
2961 vcpu->arch.regs[VCPU_REGS_RDI] = regs->rdi;
2962 vcpu->arch.regs[VCPU_REGS_RSP] = regs->rsp;
2963 vcpu->arch.regs[VCPU_REGS_RBP] = regs->rbp;
2964 #ifdef CONFIG_X86_64
2965 vcpu->arch.regs[VCPU_REGS_R8] = regs->r8;
2966 vcpu->arch.regs[VCPU_REGS_R9] = regs->r9;
2967 vcpu->arch.regs[VCPU_REGS_R10] = regs->r10;
2968 vcpu->arch.regs[VCPU_REGS_R11] = regs->r11;
2969 vcpu->arch.regs[VCPU_REGS_R12] = regs->r12;
2970 vcpu->arch.regs[VCPU_REGS_R13] = regs->r13;
2971 vcpu->arch.regs[VCPU_REGS_R14] = regs->r14;
2972 vcpu->arch.regs[VCPU_REGS_R15] = regs->r15;
2975 vcpu->arch.rip = regs->rip;
2976 kvm_x86_ops->set_rflags(vcpu, regs->rflags);
2978 kvm_x86_ops->decache_regs(vcpu);
2985 static void get_segment(struct kvm_vcpu *vcpu,
2986 struct kvm_segment *var, int seg)
2988 kvm_x86_ops->get_segment(vcpu, var, seg);
2991 void kvm_get_cs_db_l_bits(struct kvm_vcpu *vcpu, int *db, int *l)
2993 struct kvm_segment cs;
2995 get_segment(vcpu, &cs, VCPU_SREG_CS);
2999 EXPORT_SYMBOL_GPL(kvm_get_cs_db_l_bits);
3001 int kvm_arch_vcpu_ioctl_get_sregs(struct kvm_vcpu *vcpu,
3002 struct kvm_sregs *sregs)
3004 struct descriptor_table dt;
3009 get_segment(vcpu, &sregs->cs, VCPU_SREG_CS);
3010 get_segment(vcpu, &sregs->ds, VCPU_SREG_DS);
3011 get_segment(vcpu, &sregs->es, VCPU_SREG_ES);
3012 get_segment(vcpu, &sregs->fs, VCPU_SREG_FS);
3013 get_segment(vcpu, &sregs->gs, VCPU_SREG_GS);
3014 get_segment(vcpu, &sregs->ss, VCPU_SREG_SS);
3016 get_segment(vcpu, &sregs->tr, VCPU_SREG_TR);
3017 get_segment(vcpu, &sregs->ldt, VCPU_SREG_LDTR);
3019 kvm_x86_ops->get_idt(vcpu, &dt);
3020 sregs->idt.limit = dt.limit;
3021 sregs->idt.base = dt.base;
3022 kvm_x86_ops->get_gdt(vcpu, &dt);
3023 sregs->gdt.limit = dt.limit;
3024 sregs->gdt.base = dt.base;
3026 kvm_x86_ops->decache_cr4_guest_bits(vcpu);
3027 sregs->cr0 = vcpu->arch.cr0;
3028 sregs->cr2 = vcpu->arch.cr2;
3029 sregs->cr3 = vcpu->arch.cr3;
3030 sregs->cr4 = vcpu->arch.cr4;
3031 sregs->cr8 = kvm_get_cr8(vcpu);
3032 sregs->efer = vcpu->arch.shadow_efer;
3033 sregs->apic_base = kvm_get_apic_base(vcpu);
3035 if (irqchip_in_kernel(vcpu->kvm)) {
3036 memset(sregs->interrupt_bitmap, 0,
3037 sizeof sregs->interrupt_bitmap);
3038 pending_vec = kvm_x86_ops->get_irq(vcpu);
3039 if (pending_vec >= 0)
3040 set_bit(pending_vec,
3041 (unsigned long *)sregs->interrupt_bitmap);
3043 memcpy(sregs->interrupt_bitmap, vcpu->arch.irq_pending,
3044 sizeof sregs->interrupt_bitmap);
3051 static void set_segment(struct kvm_vcpu *vcpu,
3052 struct kvm_segment *var, int seg)
3054 kvm_x86_ops->set_segment(vcpu, var, seg);
3057 int kvm_arch_vcpu_ioctl_set_sregs(struct kvm_vcpu *vcpu,
3058 struct kvm_sregs *sregs)
3060 int mmu_reset_needed = 0;
3061 int i, pending_vec, max_bits;
3062 struct descriptor_table dt;
3066 dt.limit = sregs->idt.limit;
3067 dt.base = sregs->idt.base;
3068 kvm_x86_ops->set_idt(vcpu, &dt);
3069 dt.limit = sregs->gdt.limit;
3070 dt.base = sregs->gdt.base;
3071 kvm_x86_ops->set_gdt(vcpu, &dt);
3073 vcpu->arch.cr2 = sregs->cr2;
3074 mmu_reset_needed |= vcpu->arch.cr3 != sregs->cr3;
3075 vcpu->arch.cr3 = sregs->cr3;
3077 kvm_set_cr8(vcpu, sregs->cr8);
3079 mmu_reset_needed |= vcpu->arch.shadow_efer != sregs->efer;
3080 kvm_x86_ops->set_efer(vcpu, sregs->efer);
3081 kvm_set_apic_base(vcpu, sregs->apic_base);
3083 kvm_x86_ops->decache_cr4_guest_bits(vcpu);
3085 mmu_reset_needed |= vcpu->arch.cr0 != sregs->cr0;
3086 kvm_x86_ops->set_cr0(vcpu, sregs->cr0);
3087 vcpu->arch.cr0 = sregs->cr0;
3089 mmu_reset_needed |= vcpu->arch.cr4 != sregs->cr4;
3090 kvm_x86_ops->set_cr4(vcpu, sregs->cr4);
3091 if (!is_long_mode(vcpu) && is_pae(vcpu))
3092 load_pdptrs(vcpu, vcpu->arch.cr3);
3094 if (mmu_reset_needed)
3095 kvm_mmu_reset_context(vcpu);
3097 if (!irqchip_in_kernel(vcpu->kvm)) {
3098 memcpy(vcpu->arch.irq_pending, sregs->interrupt_bitmap,
3099 sizeof vcpu->arch.irq_pending);
3100 vcpu->arch.irq_summary = 0;
3101 for (i = 0; i < ARRAY_SIZE(vcpu->arch.irq_pending); ++i)
3102 if (vcpu->arch.irq_pending[i])
3103 __set_bit(i, &vcpu->arch.irq_summary);
3105 max_bits = (sizeof sregs->interrupt_bitmap) << 3;
3106 pending_vec = find_first_bit(
3107 (const unsigned long *)sregs->interrupt_bitmap,
3109 /* Only pending external irq is handled here */
3110 if (pending_vec < max_bits) {
3111 kvm_x86_ops->set_irq(vcpu, pending_vec);
3112 pr_debug("Set back pending irq %d\n",
3117 set_segment(vcpu, &sregs->cs, VCPU_SREG_CS);
3118 set_segment(vcpu, &sregs->ds, VCPU_SREG_DS);
3119 set_segment(vcpu, &sregs->es, VCPU_SREG_ES);
3120 set_segment(vcpu, &sregs->fs, VCPU_SREG_FS);
3121 set_segment(vcpu, &sregs->gs, VCPU_SREG_GS);
3122 set_segment(vcpu, &sregs->ss, VCPU_SREG_SS);
3124 set_segment(vcpu, &sregs->tr, VCPU_SREG_TR);
3125 set_segment(vcpu, &sregs->ldt, VCPU_SREG_LDTR);
3132 int kvm_arch_vcpu_ioctl_debug_guest(struct kvm_vcpu *vcpu,
3133 struct kvm_debug_guest *dbg)
3139 r = kvm_x86_ops->set_guest_debug(vcpu, dbg);
3147 * fxsave fpu state. Taken from x86_64/processor.h. To be killed when
3148 * we have asm/x86/processor.h
3159 u32 st_space[32]; /* 8*16 bytes for each FP-reg = 128 bytes */
3160 #ifdef CONFIG_X86_64
3161 u32 xmm_space[64]; /* 16*16 bytes for each XMM-reg = 256 bytes */
3163 u32 xmm_space[32]; /* 8*16 bytes for each XMM-reg = 128 bytes */
3168 * Translate a guest virtual address to a guest physical address.
3170 int kvm_arch_vcpu_ioctl_translate(struct kvm_vcpu *vcpu,
3171 struct kvm_translation *tr)
3173 unsigned long vaddr = tr->linear_address;
3177 down_read(&vcpu->kvm->slots_lock);
3178 gpa = vcpu->arch.mmu.gva_to_gpa(vcpu, vaddr);
3179 up_read(&vcpu->kvm->slots_lock);
3180 tr->physical_address = gpa;
3181 tr->valid = gpa != UNMAPPED_GVA;
3189 int kvm_arch_vcpu_ioctl_get_fpu(struct kvm_vcpu *vcpu, struct kvm_fpu *fpu)
3191 struct fxsave *fxsave = (struct fxsave *)&vcpu->arch.guest_fx_image;
3195 memcpy(fpu->fpr, fxsave->st_space, 128);
3196 fpu->fcw = fxsave->cwd;
3197 fpu->fsw = fxsave->swd;
3198 fpu->ftwx = fxsave->twd;
3199 fpu->last_opcode = fxsave->fop;
3200 fpu->last_ip = fxsave->rip;
3201 fpu->last_dp = fxsave->rdp;
3202 memcpy(fpu->xmm, fxsave->xmm_space, sizeof fxsave->xmm_space);
3209 int kvm_arch_vcpu_ioctl_set_fpu(struct kvm_vcpu *vcpu, struct kvm_fpu *fpu)
3211 struct fxsave *fxsave = (struct fxsave *)&vcpu->arch.guest_fx_image;
3215 memcpy(fxsave->st_space, fpu->fpr, 128);
3216 fxsave->cwd = fpu->fcw;
3217 fxsave->swd = fpu->fsw;
3218 fxsave->twd = fpu->ftwx;
3219 fxsave->fop = fpu->last_opcode;
3220 fxsave->rip = fpu->last_ip;
3221 fxsave->rdp = fpu->last_dp;
3222 memcpy(fxsave->xmm_space, fpu->xmm, sizeof fxsave->xmm_space);
3229 void fx_init(struct kvm_vcpu *vcpu)
3231 unsigned after_mxcsr_mask;
3233 /* Initialize guest FPU by resetting ours and saving into guest's */
3235 fx_save(&vcpu->arch.host_fx_image);
3237 fx_save(&vcpu->arch.guest_fx_image);
3238 fx_restore(&vcpu->arch.host_fx_image);
3241 vcpu->arch.cr0 |= X86_CR0_ET;
3242 after_mxcsr_mask = offsetof(struct i387_fxsave_struct, st_space);
3243 vcpu->arch.guest_fx_image.mxcsr = 0x1f80;
3244 memset((void *)&vcpu->arch.guest_fx_image + after_mxcsr_mask,
3245 0, sizeof(struct i387_fxsave_struct) - after_mxcsr_mask);
3247 EXPORT_SYMBOL_GPL(fx_init);
3249 void kvm_load_guest_fpu(struct kvm_vcpu *vcpu)
3251 if (!vcpu->fpu_active || vcpu->guest_fpu_loaded)
3254 vcpu->guest_fpu_loaded = 1;
3255 fx_save(&vcpu->arch.host_fx_image);
3256 fx_restore(&vcpu->arch.guest_fx_image);
3258 EXPORT_SYMBOL_GPL(kvm_load_guest_fpu);
3260 void kvm_put_guest_fpu(struct kvm_vcpu *vcpu)
3262 if (!vcpu->guest_fpu_loaded)
3265 vcpu->guest_fpu_loaded = 0;
3266 fx_save(&vcpu->arch.guest_fx_image);
3267 fx_restore(&vcpu->arch.host_fx_image);
3268 ++vcpu->stat.fpu_reload;
3270 EXPORT_SYMBOL_GPL(kvm_put_guest_fpu);
3272 void kvm_arch_vcpu_free(struct kvm_vcpu *vcpu)
3274 kvm_x86_ops->vcpu_free(vcpu);
3277 struct kvm_vcpu *kvm_arch_vcpu_create(struct kvm *kvm,
3280 return kvm_x86_ops->vcpu_create(kvm, id);
3283 int kvm_arch_vcpu_setup(struct kvm_vcpu *vcpu)
3287 /* We do fxsave: this must be aligned. */
3288 BUG_ON((unsigned long)&vcpu->arch.host_fx_image & 0xF);
3291 r = kvm_arch_vcpu_reset(vcpu);
3293 r = kvm_mmu_setup(vcpu);
3300 kvm_x86_ops->vcpu_free(vcpu);
3304 void kvm_arch_vcpu_destroy(struct kvm_vcpu *vcpu)
3307 kvm_mmu_unload(vcpu);
3310 kvm_x86_ops->vcpu_free(vcpu);
3313 int kvm_arch_vcpu_reset(struct kvm_vcpu *vcpu)
3315 return kvm_x86_ops->vcpu_reset(vcpu);
3318 void kvm_arch_hardware_enable(void *garbage)
3320 kvm_x86_ops->hardware_enable(garbage);
3323 void kvm_arch_hardware_disable(void *garbage)
3325 kvm_x86_ops->hardware_disable(garbage);
3328 int kvm_arch_hardware_setup(void)
3330 return kvm_x86_ops->hardware_setup();
3333 void kvm_arch_hardware_unsetup(void)
3335 kvm_x86_ops->hardware_unsetup();
3338 void kvm_arch_check_processor_compat(void *rtn)
3340 kvm_x86_ops->check_processor_compatibility(rtn);
3343 int kvm_arch_vcpu_init(struct kvm_vcpu *vcpu)
3349 BUG_ON(vcpu->kvm == NULL);
3352 vcpu->arch.mmu.root_hpa = INVALID_PAGE;
3353 if (!irqchip_in_kernel(kvm) || vcpu->vcpu_id == 0)
3354 vcpu->arch.mp_state = VCPU_MP_STATE_RUNNABLE;
3356 vcpu->arch.mp_state = VCPU_MP_STATE_UNINITIALIZED;
3358 page = alloc_page(GFP_KERNEL | __GFP_ZERO);
3363 vcpu->arch.pio_data = page_address(page);
3365 r = kvm_mmu_create(vcpu);
3367 goto fail_free_pio_data;
3369 if (irqchip_in_kernel(kvm)) {
3370 r = kvm_create_lapic(vcpu);
3372 goto fail_mmu_destroy;
3378 kvm_mmu_destroy(vcpu);
3380 free_page((unsigned long)vcpu->arch.pio_data);
3385 void kvm_arch_vcpu_uninit(struct kvm_vcpu *vcpu)
3387 kvm_free_lapic(vcpu);
3388 kvm_mmu_destroy(vcpu);
3389 free_page((unsigned long)vcpu->arch.pio_data);
3392 struct kvm *kvm_arch_create_vm(void)
3394 struct kvm *kvm = kzalloc(sizeof(struct kvm), GFP_KERNEL);
3397 return ERR_PTR(-ENOMEM);
3399 INIT_LIST_HEAD(&kvm->arch.active_mmu_pages);
3404 static void kvm_unload_vcpu_mmu(struct kvm_vcpu *vcpu)
3407 kvm_mmu_unload(vcpu);
3411 static void kvm_free_vcpus(struct kvm *kvm)
3416 * Unpin any mmu pages first.
3418 for (i = 0; i < KVM_MAX_VCPUS; ++i)
3420 kvm_unload_vcpu_mmu(kvm->vcpus[i]);
3421 for (i = 0; i < KVM_MAX_VCPUS; ++i) {
3422 if (kvm->vcpus[i]) {
3423 kvm_arch_vcpu_free(kvm->vcpus[i]);
3424 kvm->vcpus[i] = NULL;
3430 void kvm_arch_destroy_vm(struct kvm *kvm)
3433 kfree(kvm->arch.vpic);
3434 kfree(kvm->arch.vioapic);
3435 kvm_free_vcpus(kvm);
3436 kvm_free_physmem(kvm);
3440 int kvm_arch_set_memory_region(struct kvm *kvm,
3441 struct kvm_userspace_memory_region *mem,
3442 struct kvm_memory_slot old,
3445 int npages = mem->memory_size >> PAGE_SHIFT;
3446 struct kvm_memory_slot *memslot = &kvm->memslots[mem->slot];
3448 /*To keep backward compatibility with older userspace,
3449 *x86 needs to hanlde !user_alloc case.
3452 if (npages && !old.rmap) {
3453 down_write(¤t->mm->mmap_sem);
3454 memslot->userspace_addr = do_mmap(NULL, 0,
3456 PROT_READ | PROT_WRITE,
3457 MAP_SHARED | MAP_ANONYMOUS,
3459 up_write(¤t->mm->mmap_sem);
3461 if (IS_ERR((void *)memslot->userspace_addr))
3462 return PTR_ERR((void *)memslot->userspace_addr);
3464 if (!old.user_alloc && old.rmap) {
3467 down_write(¤t->mm->mmap_sem);
3468 ret = do_munmap(current->mm, old.userspace_addr,
3469 old.npages * PAGE_SIZE);
3470 up_write(¤t->mm->mmap_sem);
3473 "kvm_vm_ioctl_set_memory_region: "
3474 "failed to munmap memory\n");
3479 if (!kvm->arch.n_requested_mmu_pages) {
3480 unsigned int nr_mmu_pages = kvm_mmu_calculate_mmu_pages(kvm);
3481 kvm_mmu_change_mmu_pages(kvm, nr_mmu_pages);
3484 kvm_mmu_slot_remove_write_access(kvm, mem->slot);
3485 kvm_flush_remote_tlbs(kvm);
3490 int kvm_arch_vcpu_runnable(struct kvm_vcpu *vcpu)
3492 return vcpu->arch.mp_state == VCPU_MP_STATE_RUNNABLE
3493 || vcpu->arch.mp_state == VCPU_MP_STATE_SIPI_RECEIVED;
3496 static void vcpu_kick_intr(void *info)
3499 struct kvm_vcpu *vcpu = (struct kvm_vcpu *)info;
3500 printk(KERN_DEBUG "vcpu_kick_intr %p \n", vcpu);
3504 void kvm_vcpu_kick(struct kvm_vcpu *vcpu)
3506 int ipi_pcpu = vcpu->cpu;
3508 if (waitqueue_active(&vcpu->wq)) {
3509 wake_up_interruptible(&vcpu->wq);
3510 ++vcpu->stat.halt_wakeup;
3512 if (vcpu->guest_mode)
3513 smp_call_function_single(ipi_pcpu, vcpu_kick_intr, vcpu, 0, 0);