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_queue_exception_e(struct kvm_vcpu *vcpu, unsigned nr, u32 error_code)
178 WARN_ON(vcpu->arch.exception.pending);
179 vcpu->arch.exception.pending = true;
180 vcpu->arch.exception.has_error_code = true;
181 vcpu->arch.exception.nr = nr;
182 vcpu->arch.exception.error_code = error_code;
184 EXPORT_SYMBOL_GPL(kvm_queue_exception_e);
186 static void __queue_exception(struct kvm_vcpu *vcpu)
188 kvm_x86_ops->queue_exception(vcpu, vcpu->arch.exception.nr,
189 vcpu->arch.exception.has_error_code,
190 vcpu->arch.exception.error_code);
194 * Load the pae pdptrs. Return true is they are all valid.
196 int load_pdptrs(struct kvm_vcpu *vcpu, unsigned long cr3)
198 gfn_t pdpt_gfn = cr3 >> PAGE_SHIFT;
199 unsigned offset = ((cr3 & (PAGE_SIZE-1)) >> 5) << 2;
202 u64 pdpte[ARRAY_SIZE(vcpu->arch.pdptrs)];
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));
223 EXPORT_SYMBOL_GPL(load_pdptrs);
225 static bool pdptrs_changed(struct kvm_vcpu *vcpu)
227 u64 pdpte[ARRAY_SIZE(vcpu->arch.pdptrs)];
231 if (is_long_mode(vcpu) || !is_pae(vcpu))
234 r = kvm_read_guest(vcpu->kvm, vcpu->arch.cr3 & ~31u, pdpte, sizeof(pdpte));
237 changed = memcmp(pdpte, vcpu->arch.pdptrs, sizeof(pdpte)) != 0;
243 void kvm_set_cr0(struct kvm_vcpu *vcpu, unsigned long cr0)
245 if (cr0 & CR0_RESERVED_BITS) {
246 printk(KERN_DEBUG "set_cr0: 0x%lx #GP, reserved bits 0x%lx\n",
247 cr0, vcpu->arch.cr0);
248 kvm_inject_gp(vcpu, 0);
252 if ((cr0 & X86_CR0_NW) && !(cr0 & X86_CR0_CD)) {
253 printk(KERN_DEBUG "set_cr0: #GP, CD == 0 && NW == 1\n");
254 kvm_inject_gp(vcpu, 0);
258 if ((cr0 & X86_CR0_PG) && !(cr0 & X86_CR0_PE)) {
259 printk(KERN_DEBUG "set_cr0: #GP, set PG flag "
260 "and a clear PE flag\n");
261 kvm_inject_gp(vcpu, 0);
265 if (!is_paging(vcpu) && (cr0 & X86_CR0_PG)) {
267 if ((vcpu->arch.shadow_efer & EFER_LME)) {
271 printk(KERN_DEBUG "set_cr0: #GP, start paging "
272 "in long mode while PAE is disabled\n");
273 kvm_inject_gp(vcpu, 0);
276 kvm_x86_ops->get_cs_db_l_bits(vcpu, &cs_db, &cs_l);
278 printk(KERN_DEBUG "set_cr0: #GP, start paging "
279 "in long mode while CS.L == 1\n");
280 kvm_inject_gp(vcpu, 0);
286 if (is_pae(vcpu) && !load_pdptrs(vcpu, vcpu->arch.cr3)) {
287 printk(KERN_DEBUG "set_cr0: #GP, pdptrs "
289 kvm_inject_gp(vcpu, 0);
295 kvm_x86_ops->set_cr0(vcpu, cr0);
296 vcpu->arch.cr0 = cr0;
298 kvm_mmu_reset_context(vcpu);
301 EXPORT_SYMBOL_GPL(kvm_set_cr0);
303 void kvm_lmsw(struct kvm_vcpu *vcpu, unsigned long msw)
305 kvm_set_cr0(vcpu, (vcpu->arch.cr0 & ~0x0ful) | (msw & 0x0f));
306 KVMTRACE_1D(LMSW, vcpu,
307 (u32)((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.
380 * Does the new cr3 value map to physical memory? (Note, we
381 * catch an invalid cr3 even in real-mode, because it would
382 * cause trouble later on when we turn on paging anyway.)
384 * A real CPU would silently accept an invalid cr3 and would
385 * attempt to use it - with largely undefined (and often hard
386 * to debug) behavior on the guest side.
388 if (unlikely(!gfn_to_memslot(vcpu->kvm, cr3 >> PAGE_SHIFT)))
389 kvm_inject_gp(vcpu, 0);
391 vcpu->arch.cr3 = cr3;
392 vcpu->arch.mmu.new_cr3(vcpu);
395 EXPORT_SYMBOL_GPL(kvm_set_cr3);
397 void kvm_set_cr8(struct kvm_vcpu *vcpu, unsigned long cr8)
399 if (cr8 & CR8_RESERVED_BITS) {
400 printk(KERN_DEBUG "set_cr8: #GP, reserved bits 0x%lx\n", cr8);
401 kvm_inject_gp(vcpu, 0);
404 if (irqchip_in_kernel(vcpu->kvm))
405 kvm_lapic_set_tpr(vcpu, cr8);
407 vcpu->arch.cr8 = cr8;
409 EXPORT_SYMBOL_GPL(kvm_set_cr8);
411 unsigned long kvm_get_cr8(struct kvm_vcpu *vcpu)
413 if (irqchip_in_kernel(vcpu->kvm))
414 return kvm_lapic_get_cr8(vcpu);
416 return vcpu->arch.cr8;
418 EXPORT_SYMBOL_GPL(kvm_get_cr8);
421 * List of msr numbers which we expose to userspace through KVM_GET_MSRS
422 * and KVM_SET_MSRS, and KVM_GET_MSR_INDEX_LIST.
424 * This list is modified at module load time to reflect the
425 * capabilities of the host cpu.
427 static u32 msrs_to_save[] = {
428 MSR_IA32_SYSENTER_CS, MSR_IA32_SYSENTER_ESP, MSR_IA32_SYSENTER_EIP,
431 MSR_CSTAR, MSR_KERNEL_GS_BASE, MSR_SYSCALL_MASK, MSR_LSTAR,
433 MSR_IA32_TIME_STAMP_COUNTER, MSR_KVM_SYSTEM_TIME, MSR_KVM_WALL_CLOCK,
434 MSR_IA32_PERF_STATUS,
437 static unsigned num_msrs_to_save;
439 static u32 emulated_msrs[] = {
440 MSR_IA32_MISC_ENABLE,
443 static void set_efer(struct kvm_vcpu *vcpu, u64 efer)
445 if (efer & efer_reserved_bits) {
446 printk(KERN_DEBUG "set_efer: 0x%llx #GP, reserved bits\n",
448 kvm_inject_gp(vcpu, 0);
453 && (vcpu->arch.shadow_efer & EFER_LME) != (efer & EFER_LME)) {
454 printk(KERN_DEBUG "set_efer: #GP, change LME while paging\n");
455 kvm_inject_gp(vcpu, 0);
459 kvm_x86_ops->set_efer(vcpu, efer);
462 efer |= vcpu->arch.shadow_efer & EFER_LMA;
464 vcpu->arch.shadow_efer = efer;
467 void kvm_enable_efer_bits(u64 mask)
469 efer_reserved_bits &= ~mask;
471 EXPORT_SYMBOL_GPL(kvm_enable_efer_bits);
475 * Writes msr value into into the appropriate "register".
476 * Returns 0 on success, non-0 otherwise.
477 * Assumes vcpu_load() was already called.
479 int kvm_set_msr(struct kvm_vcpu *vcpu, u32 msr_index, u64 data)
481 return kvm_x86_ops->set_msr(vcpu, msr_index, data);
485 * Adapt set_msr() to msr_io()'s calling convention
487 static int do_set_msr(struct kvm_vcpu *vcpu, unsigned index, u64 *data)
489 return kvm_set_msr(vcpu, index, *data);
492 static void kvm_write_wall_clock(struct kvm *kvm, gpa_t wall_clock)
495 struct pvclock_wall_clock wc;
496 struct timespec now, sys, boot;
503 kvm_write_guest(kvm, wall_clock, &version, sizeof(version));
506 * The guest calculates current wall clock time by adding
507 * system time (updated by kvm_write_guest_time below) to the
508 * wall clock specified here. guest system time equals host
509 * system time for us, thus we must fill in host boot time here.
511 now = current_kernel_time();
513 boot = ns_to_timespec(timespec_to_ns(&now) - timespec_to_ns(&sys));
515 wc.sec = boot.tv_sec;
516 wc.nsec = boot.tv_nsec;
517 wc.version = version;
519 kvm_write_guest(kvm, wall_clock, &wc, sizeof(wc));
522 kvm_write_guest(kvm, wall_clock, &version, sizeof(version));
525 static uint32_t div_frac(uint32_t dividend, uint32_t divisor)
527 uint32_t quotient, remainder;
529 /* Don't try to replace with do_div(), this one calculates
530 * "(dividend << 32) / divisor" */
532 : "=a" (quotient), "=d" (remainder)
533 : "0" (0), "1" (dividend), "r" (divisor) );
537 static void kvm_set_time_scale(uint32_t tsc_khz, struct pvclock_vcpu_time_info *hv_clock)
539 uint64_t nsecs = 1000000000LL;
544 tps64 = tsc_khz * 1000LL;
545 while (tps64 > nsecs*2) {
550 tps32 = (uint32_t)tps64;
551 while (tps32 <= (uint32_t)nsecs) {
556 hv_clock->tsc_shift = shift;
557 hv_clock->tsc_to_system_mul = div_frac(nsecs, tps32);
559 pr_debug("%s: tsc_khz %u, tsc_shift %d, tsc_mul %u\n",
560 __FUNCTION__, tsc_khz, hv_clock->tsc_shift,
561 hv_clock->tsc_to_system_mul);
564 static void kvm_write_guest_time(struct kvm_vcpu *v)
568 struct kvm_vcpu_arch *vcpu = &v->arch;
571 if ((!vcpu->time_page))
574 if (unlikely(vcpu->hv_clock_tsc_khz != tsc_khz)) {
575 kvm_set_time_scale(tsc_khz, &vcpu->hv_clock);
576 vcpu->hv_clock_tsc_khz = tsc_khz;
579 /* Keep irq disabled to prevent changes to the clock */
580 local_irq_save(flags);
581 kvm_get_msr(v, MSR_IA32_TIME_STAMP_COUNTER,
582 &vcpu->hv_clock.tsc_timestamp);
584 local_irq_restore(flags);
586 /* With all the info we got, fill in the values */
588 vcpu->hv_clock.system_time = ts.tv_nsec +
589 (NSEC_PER_SEC * (u64)ts.tv_sec);
591 * The interface expects us to write an even number signaling that the
592 * update is finished. Since the guest won't see the intermediate
593 * state, we just increase by 2 at the end.
595 vcpu->hv_clock.version += 2;
597 shared_kaddr = kmap_atomic(vcpu->time_page, KM_USER0);
599 memcpy(shared_kaddr + vcpu->time_offset, &vcpu->hv_clock,
600 sizeof(vcpu->hv_clock));
602 kunmap_atomic(shared_kaddr, KM_USER0);
604 mark_page_dirty(v->kvm, vcpu->time >> PAGE_SHIFT);
608 int kvm_set_msr_common(struct kvm_vcpu *vcpu, u32 msr, u64 data)
612 set_efer(vcpu, data);
614 case MSR_IA32_MC0_STATUS:
615 pr_unimpl(vcpu, "%s: MSR_IA32_MC0_STATUS 0x%llx, nop\n",
618 case MSR_IA32_MCG_STATUS:
619 pr_unimpl(vcpu, "%s: MSR_IA32_MCG_STATUS 0x%llx, nop\n",
622 case MSR_IA32_MCG_CTL:
623 pr_unimpl(vcpu, "%s: MSR_IA32_MCG_CTL 0x%llx, nop\n",
626 case MSR_IA32_UCODE_REV:
627 case MSR_IA32_UCODE_WRITE:
628 case 0x200 ... 0x2ff: /* MTRRs */
630 case MSR_IA32_APICBASE:
631 kvm_set_apic_base(vcpu, data);
633 case MSR_IA32_MISC_ENABLE:
634 vcpu->arch.ia32_misc_enable_msr = data;
636 case MSR_KVM_WALL_CLOCK:
637 vcpu->kvm->arch.wall_clock = data;
638 kvm_write_wall_clock(vcpu->kvm, data);
640 case MSR_KVM_SYSTEM_TIME: {
641 if (vcpu->arch.time_page) {
642 kvm_release_page_dirty(vcpu->arch.time_page);
643 vcpu->arch.time_page = NULL;
646 vcpu->arch.time = data;
648 /* we verify if the enable bit is set... */
652 /* ...but clean it before doing the actual write */
653 vcpu->arch.time_offset = data & ~(PAGE_MASK | 1);
655 down_read(¤t->mm->mmap_sem);
656 vcpu->arch.time_page =
657 gfn_to_page(vcpu->kvm, data >> PAGE_SHIFT);
658 up_read(¤t->mm->mmap_sem);
660 if (is_error_page(vcpu->arch.time_page)) {
661 kvm_release_page_clean(vcpu->arch.time_page);
662 vcpu->arch.time_page = NULL;
665 kvm_write_guest_time(vcpu);
669 pr_unimpl(vcpu, "unhandled wrmsr: 0x%x data %llx\n", msr, data);
674 EXPORT_SYMBOL_GPL(kvm_set_msr_common);
678 * Reads an msr value (of 'msr_index') into 'pdata'.
679 * Returns 0 on success, non-0 otherwise.
680 * Assumes vcpu_load() was already called.
682 int kvm_get_msr(struct kvm_vcpu *vcpu, u32 msr_index, u64 *pdata)
684 return kvm_x86_ops->get_msr(vcpu, msr_index, pdata);
687 int kvm_get_msr_common(struct kvm_vcpu *vcpu, u32 msr, u64 *pdata)
692 case 0xc0010010: /* SYSCFG */
693 case 0xc0010015: /* HWCR */
694 case MSR_IA32_PLATFORM_ID:
695 case MSR_IA32_P5_MC_ADDR:
696 case MSR_IA32_P5_MC_TYPE:
697 case MSR_IA32_MC0_CTL:
698 case MSR_IA32_MCG_STATUS:
699 case MSR_IA32_MCG_CAP:
700 case MSR_IA32_MCG_CTL:
701 case MSR_IA32_MC0_MISC:
702 case MSR_IA32_MC0_MISC+4:
703 case MSR_IA32_MC0_MISC+8:
704 case MSR_IA32_MC0_MISC+12:
705 case MSR_IA32_MC0_MISC+16:
706 case MSR_IA32_UCODE_REV:
707 case MSR_IA32_EBL_CR_POWERON:
710 case 0x200 ... 0x2ff:
713 case 0xcd: /* fsb frequency */
716 case MSR_IA32_APICBASE:
717 data = kvm_get_apic_base(vcpu);
719 case MSR_IA32_MISC_ENABLE:
720 data = vcpu->arch.ia32_misc_enable_msr;
722 case MSR_IA32_PERF_STATUS:
723 /* TSC increment by tick */
726 data |= (((uint64_t)4ULL) << 40);
729 data = vcpu->arch.shadow_efer;
731 case MSR_KVM_WALL_CLOCK:
732 data = vcpu->kvm->arch.wall_clock;
734 case MSR_KVM_SYSTEM_TIME:
735 data = vcpu->arch.time;
738 pr_unimpl(vcpu, "unhandled rdmsr: 0x%x\n", msr);
744 EXPORT_SYMBOL_GPL(kvm_get_msr_common);
747 * Read or write a bunch of msrs. All parameters are kernel addresses.
749 * @return number of msrs set successfully.
751 static int __msr_io(struct kvm_vcpu *vcpu, struct kvm_msrs *msrs,
752 struct kvm_msr_entry *entries,
753 int (*do_msr)(struct kvm_vcpu *vcpu,
754 unsigned index, u64 *data))
760 down_read(&vcpu->kvm->slots_lock);
761 for (i = 0; i < msrs->nmsrs; ++i)
762 if (do_msr(vcpu, entries[i].index, &entries[i].data))
764 up_read(&vcpu->kvm->slots_lock);
772 * Read or write a bunch of msrs. Parameters are user addresses.
774 * @return number of msrs set successfully.
776 static int msr_io(struct kvm_vcpu *vcpu, struct kvm_msrs __user *user_msrs,
777 int (*do_msr)(struct kvm_vcpu *vcpu,
778 unsigned index, u64 *data),
781 struct kvm_msrs msrs;
782 struct kvm_msr_entry *entries;
787 if (copy_from_user(&msrs, user_msrs, sizeof msrs))
791 if (msrs.nmsrs >= MAX_IO_MSRS)
795 size = sizeof(struct kvm_msr_entry) * msrs.nmsrs;
796 entries = vmalloc(size);
801 if (copy_from_user(entries, user_msrs->entries, size))
804 r = n = __msr_io(vcpu, &msrs, entries, do_msr);
809 if (writeback && copy_to_user(user_msrs->entries, entries, size))
820 int kvm_dev_ioctl_check_extension(long ext)
825 case KVM_CAP_IRQCHIP:
827 case KVM_CAP_MMU_SHADOW_CACHE_CONTROL:
828 case KVM_CAP_USER_MEMORY:
829 case KVM_CAP_SET_TSS_ADDR:
830 case KVM_CAP_EXT_CPUID:
831 case KVM_CAP_CLOCKSOURCE:
833 case KVM_CAP_NOP_IO_DELAY:
834 case KVM_CAP_MP_STATE:
838 r = !kvm_x86_ops->cpu_has_accelerated_tpr();
840 case KVM_CAP_NR_VCPUS:
843 case KVM_CAP_NR_MEMSLOTS:
844 r = KVM_MEMORY_SLOTS;
857 long kvm_arch_dev_ioctl(struct file *filp,
858 unsigned int ioctl, unsigned long arg)
860 void __user *argp = (void __user *)arg;
864 case KVM_GET_MSR_INDEX_LIST: {
865 struct kvm_msr_list __user *user_msr_list = argp;
866 struct kvm_msr_list msr_list;
870 if (copy_from_user(&msr_list, user_msr_list, sizeof msr_list))
873 msr_list.nmsrs = num_msrs_to_save + ARRAY_SIZE(emulated_msrs);
874 if (copy_to_user(user_msr_list, &msr_list, sizeof msr_list))
877 if (n < num_msrs_to_save)
880 if (copy_to_user(user_msr_list->indices, &msrs_to_save,
881 num_msrs_to_save * sizeof(u32)))
883 if (copy_to_user(user_msr_list->indices
884 + num_msrs_to_save * sizeof(u32),
886 ARRAY_SIZE(emulated_msrs) * sizeof(u32)))
891 case KVM_GET_SUPPORTED_CPUID: {
892 struct kvm_cpuid2 __user *cpuid_arg = argp;
893 struct kvm_cpuid2 cpuid;
896 if (copy_from_user(&cpuid, cpuid_arg, sizeof cpuid))
898 r = kvm_dev_ioctl_get_supported_cpuid(&cpuid,
904 if (copy_to_user(cpuid_arg, &cpuid, sizeof cpuid))
916 void kvm_arch_vcpu_load(struct kvm_vcpu *vcpu, int cpu)
918 kvm_x86_ops->vcpu_load(vcpu, cpu);
919 kvm_write_guest_time(vcpu);
922 void kvm_arch_vcpu_put(struct kvm_vcpu *vcpu)
924 kvm_x86_ops->vcpu_put(vcpu);
925 kvm_put_guest_fpu(vcpu);
928 static int is_efer_nx(void)
932 rdmsrl(MSR_EFER, efer);
933 return efer & EFER_NX;
936 static void cpuid_fix_nx_cap(struct kvm_vcpu *vcpu)
939 struct kvm_cpuid_entry2 *e, *entry;
942 for (i = 0; i < vcpu->arch.cpuid_nent; ++i) {
943 e = &vcpu->arch.cpuid_entries[i];
944 if (e->function == 0x80000001) {
949 if (entry && (entry->edx & (1 << 20)) && !is_efer_nx()) {
950 entry->edx &= ~(1 << 20);
951 printk(KERN_INFO "kvm: guest NX capability removed\n");
955 /* when an old userspace process fills a new kernel module */
956 static int kvm_vcpu_ioctl_set_cpuid(struct kvm_vcpu *vcpu,
957 struct kvm_cpuid *cpuid,
958 struct kvm_cpuid_entry __user *entries)
961 struct kvm_cpuid_entry *cpuid_entries;
964 if (cpuid->nent > KVM_MAX_CPUID_ENTRIES)
967 cpuid_entries = vmalloc(sizeof(struct kvm_cpuid_entry) * cpuid->nent);
971 if (copy_from_user(cpuid_entries, entries,
972 cpuid->nent * sizeof(struct kvm_cpuid_entry)))
974 for (i = 0; i < cpuid->nent; i++) {
975 vcpu->arch.cpuid_entries[i].function = cpuid_entries[i].function;
976 vcpu->arch.cpuid_entries[i].eax = cpuid_entries[i].eax;
977 vcpu->arch.cpuid_entries[i].ebx = cpuid_entries[i].ebx;
978 vcpu->arch.cpuid_entries[i].ecx = cpuid_entries[i].ecx;
979 vcpu->arch.cpuid_entries[i].edx = cpuid_entries[i].edx;
980 vcpu->arch.cpuid_entries[i].index = 0;
981 vcpu->arch.cpuid_entries[i].flags = 0;
982 vcpu->arch.cpuid_entries[i].padding[0] = 0;
983 vcpu->arch.cpuid_entries[i].padding[1] = 0;
984 vcpu->arch.cpuid_entries[i].padding[2] = 0;
986 vcpu->arch.cpuid_nent = cpuid->nent;
987 cpuid_fix_nx_cap(vcpu);
991 vfree(cpuid_entries);
996 static int kvm_vcpu_ioctl_set_cpuid2(struct kvm_vcpu *vcpu,
997 struct kvm_cpuid2 *cpuid,
998 struct kvm_cpuid_entry2 __user *entries)
1003 if (cpuid->nent > KVM_MAX_CPUID_ENTRIES)
1006 if (copy_from_user(&vcpu->arch.cpuid_entries, entries,
1007 cpuid->nent * sizeof(struct kvm_cpuid_entry2)))
1009 vcpu->arch.cpuid_nent = cpuid->nent;
1016 static int kvm_vcpu_ioctl_get_cpuid2(struct kvm_vcpu *vcpu,
1017 struct kvm_cpuid2 *cpuid,
1018 struct kvm_cpuid_entry2 __user *entries)
1023 if (cpuid->nent < vcpu->arch.cpuid_nent)
1026 if (copy_to_user(entries, &vcpu->arch.cpuid_entries,
1027 vcpu->arch.cpuid_nent * sizeof(struct kvm_cpuid_entry2)))
1032 cpuid->nent = vcpu->arch.cpuid_nent;
1036 static inline u32 bit(int bitno)
1038 return 1 << (bitno & 31);
1041 static void do_cpuid_1_ent(struct kvm_cpuid_entry2 *entry, u32 function,
1044 entry->function = function;
1045 entry->index = index;
1046 cpuid_count(entry->function, entry->index,
1047 &entry->eax, &entry->ebx, &entry->ecx, &entry->edx);
1051 static void do_cpuid_ent(struct kvm_cpuid_entry2 *entry, u32 function,
1052 u32 index, int *nent, int maxnent)
1054 const u32 kvm_supported_word0_x86_features = bit(X86_FEATURE_FPU) |
1055 bit(X86_FEATURE_VME) | bit(X86_FEATURE_DE) |
1056 bit(X86_FEATURE_PSE) | bit(X86_FEATURE_TSC) |
1057 bit(X86_FEATURE_MSR) | bit(X86_FEATURE_PAE) |
1058 bit(X86_FEATURE_CX8) | bit(X86_FEATURE_APIC) |
1059 bit(X86_FEATURE_SEP) | bit(X86_FEATURE_PGE) |
1060 bit(X86_FEATURE_CMOV) | bit(X86_FEATURE_PSE36) |
1061 bit(X86_FEATURE_CLFLSH) | bit(X86_FEATURE_MMX) |
1062 bit(X86_FEATURE_FXSR) | bit(X86_FEATURE_XMM) |
1063 bit(X86_FEATURE_XMM2) | bit(X86_FEATURE_SELFSNOOP);
1064 const u32 kvm_supported_word1_x86_features = bit(X86_FEATURE_FPU) |
1065 bit(X86_FEATURE_VME) | bit(X86_FEATURE_DE) |
1066 bit(X86_FEATURE_PSE) | bit(X86_FEATURE_TSC) |
1067 bit(X86_FEATURE_MSR) | bit(X86_FEATURE_PAE) |
1068 bit(X86_FEATURE_CX8) | bit(X86_FEATURE_APIC) |
1069 bit(X86_FEATURE_PGE) |
1070 bit(X86_FEATURE_CMOV) | bit(X86_FEATURE_PSE36) |
1071 bit(X86_FEATURE_MMX) | bit(X86_FEATURE_FXSR) |
1072 bit(X86_FEATURE_SYSCALL) |
1073 (bit(X86_FEATURE_NX) && is_efer_nx()) |
1074 #ifdef CONFIG_X86_64
1075 bit(X86_FEATURE_LM) |
1077 bit(X86_FEATURE_MMXEXT) |
1078 bit(X86_FEATURE_3DNOWEXT) |
1079 bit(X86_FEATURE_3DNOW);
1080 const u32 kvm_supported_word3_x86_features =
1081 bit(X86_FEATURE_XMM3) | bit(X86_FEATURE_CX16);
1082 const u32 kvm_supported_word6_x86_features =
1083 bit(X86_FEATURE_LAHF_LM) | bit(X86_FEATURE_CMP_LEGACY);
1085 /* all func 2 cpuid_count() should be called on the same cpu */
1087 do_cpuid_1_ent(entry, function, index);
1092 entry->eax = min(entry->eax, (u32)0xb);
1095 entry->edx &= kvm_supported_word0_x86_features;
1096 entry->ecx &= kvm_supported_word3_x86_features;
1098 /* function 2 entries are STATEFUL. That is, repeated cpuid commands
1099 * may return different values. This forces us to get_cpu() before
1100 * issuing the first command, and also to emulate this annoying behavior
1101 * in kvm_emulate_cpuid() using KVM_CPUID_FLAG_STATE_READ_NEXT */
1103 int t, times = entry->eax & 0xff;
1105 entry->flags |= KVM_CPUID_FLAG_STATEFUL_FUNC;
1106 for (t = 1; t < times && *nent < maxnent; ++t) {
1107 do_cpuid_1_ent(&entry[t], function, 0);
1108 entry[t].flags |= KVM_CPUID_FLAG_STATEFUL_FUNC;
1113 /* function 4 and 0xb have additional index. */
1117 entry->flags |= KVM_CPUID_FLAG_SIGNIFCANT_INDEX;
1118 /* read more entries until cache_type is zero */
1119 for (i = 1; *nent < maxnent; ++i) {
1120 cache_type = entry[i - 1].eax & 0x1f;
1123 do_cpuid_1_ent(&entry[i], function, i);
1125 KVM_CPUID_FLAG_SIGNIFCANT_INDEX;
1133 entry->flags |= KVM_CPUID_FLAG_SIGNIFCANT_INDEX;
1134 /* read more entries until level_type is zero */
1135 for (i = 1; *nent < maxnent; ++i) {
1136 level_type = entry[i - 1].ecx & 0xff;
1139 do_cpuid_1_ent(&entry[i], function, i);
1141 KVM_CPUID_FLAG_SIGNIFCANT_INDEX;
1147 entry->eax = min(entry->eax, 0x8000001a);
1150 entry->edx &= kvm_supported_word1_x86_features;
1151 entry->ecx &= kvm_supported_word6_x86_features;
1157 static int kvm_dev_ioctl_get_supported_cpuid(struct kvm_cpuid2 *cpuid,
1158 struct kvm_cpuid_entry2 __user *entries)
1160 struct kvm_cpuid_entry2 *cpuid_entries;
1161 int limit, nent = 0, r = -E2BIG;
1164 if (cpuid->nent < 1)
1167 cpuid_entries = vmalloc(sizeof(struct kvm_cpuid_entry2) * cpuid->nent);
1171 do_cpuid_ent(&cpuid_entries[0], 0, 0, &nent, cpuid->nent);
1172 limit = cpuid_entries[0].eax;
1173 for (func = 1; func <= limit && nent < cpuid->nent; ++func)
1174 do_cpuid_ent(&cpuid_entries[nent], func, 0,
1175 &nent, cpuid->nent);
1177 if (nent >= cpuid->nent)
1180 do_cpuid_ent(&cpuid_entries[nent], 0x80000000, 0, &nent, cpuid->nent);
1181 limit = cpuid_entries[nent - 1].eax;
1182 for (func = 0x80000001; func <= limit && nent < cpuid->nent; ++func)
1183 do_cpuid_ent(&cpuid_entries[nent], func, 0,
1184 &nent, cpuid->nent);
1186 if (copy_to_user(entries, cpuid_entries,
1187 nent * sizeof(struct kvm_cpuid_entry2)))
1193 vfree(cpuid_entries);
1198 static int kvm_vcpu_ioctl_get_lapic(struct kvm_vcpu *vcpu,
1199 struct kvm_lapic_state *s)
1202 memcpy(s->regs, vcpu->arch.apic->regs, sizeof *s);
1208 static int kvm_vcpu_ioctl_set_lapic(struct kvm_vcpu *vcpu,
1209 struct kvm_lapic_state *s)
1212 memcpy(vcpu->arch.apic->regs, s->regs, sizeof *s);
1213 kvm_apic_post_state_restore(vcpu);
1219 static int kvm_vcpu_ioctl_interrupt(struct kvm_vcpu *vcpu,
1220 struct kvm_interrupt *irq)
1222 if (irq->irq < 0 || irq->irq >= 256)
1224 if (irqchip_in_kernel(vcpu->kvm))
1228 set_bit(irq->irq, vcpu->arch.irq_pending);
1229 set_bit(irq->irq / BITS_PER_LONG, &vcpu->arch.irq_summary);
1236 static int vcpu_ioctl_tpr_access_reporting(struct kvm_vcpu *vcpu,
1237 struct kvm_tpr_access_ctl *tac)
1241 vcpu->arch.tpr_access_reporting = !!tac->enabled;
1245 long kvm_arch_vcpu_ioctl(struct file *filp,
1246 unsigned int ioctl, unsigned long arg)
1248 struct kvm_vcpu *vcpu = filp->private_data;
1249 void __user *argp = (void __user *)arg;
1253 case KVM_GET_LAPIC: {
1254 struct kvm_lapic_state lapic;
1256 memset(&lapic, 0, sizeof lapic);
1257 r = kvm_vcpu_ioctl_get_lapic(vcpu, &lapic);
1261 if (copy_to_user(argp, &lapic, sizeof lapic))
1266 case KVM_SET_LAPIC: {
1267 struct kvm_lapic_state lapic;
1270 if (copy_from_user(&lapic, argp, sizeof lapic))
1272 r = kvm_vcpu_ioctl_set_lapic(vcpu, &lapic);;
1278 case KVM_INTERRUPT: {
1279 struct kvm_interrupt irq;
1282 if (copy_from_user(&irq, argp, sizeof irq))
1284 r = kvm_vcpu_ioctl_interrupt(vcpu, &irq);
1290 case KVM_SET_CPUID: {
1291 struct kvm_cpuid __user *cpuid_arg = argp;
1292 struct kvm_cpuid cpuid;
1295 if (copy_from_user(&cpuid, cpuid_arg, sizeof cpuid))
1297 r = kvm_vcpu_ioctl_set_cpuid(vcpu, &cpuid, cpuid_arg->entries);
1302 case KVM_SET_CPUID2: {
1303 struct kvm_cpuid2 __user *cpuid_arg = argp;
1304 struct kvm_cpuid2 cpuid;
1307 if (copy_from_user(&cpuid, cpuid_arg, sizeof cpuid))
1309 r = kvm_vcpu_ioctl_set_cpuid2(vcpu, &cpuid,
1310 cpuid_arg->entries);
1315 case KVM_GET_CPUID2: {
1316 struct kvm_cpuid2 __user *cpuid_arg = argp;
1317 struct kvm_cpuid2 cpuid;
1320 if (copy_from_user(&cpuid, cpuid_arg, sizeof cpuid))
1322 r = kvm_vcpu_ioctl_get_cpuid2(vcpu, &cpuid,
1323 cpuid_arg->entries);
1327 if (copy_to_user(cpuid_arg, &cpuid, sizeof cpuid))
1333 r = msr_io(vcpu, argp, kvm_get_msr, 1);
1336 r = msr_io(vcpu, argp, do_set_msr, 0);
1338 case KVM_TPR_ACCESS_REPORTING: {
1339 struct kvm_tpr_access_ctl tac;
1342 if (copy_from_user(&tac, argp, sizeof tac))
1344 r = vcpu_ioctl_tpr_access_reporting(vcpu, &tac);
1348 if (copy_to_user(argp, &tac, sizeof tac))
1353 case KVM_SET_VAPIC_ADDR: {
1354 struct kvm_vapic_addr va;
1357 if (!irqchip_in_kernel(vcpu->kvm))
1360 if (copy_from_user(&va, argp, sizeof va))
1363 kvm_lapic_set_vapic_addr(vcpu, va.vapic_addr);
1373 static int kvm_vm_ioctl_set_tss_addr(struct kvm *kvm, unsigned long addr)
1377 if (addr > (unsigned int)(-3 * PAGE_SIZE))
1379 ret = kvm_x86_ops->set_tss_addr(kvm, addr);
1383 static int kvm_vm_ioctl_set_nr_mmu_pages(struct kvm *kvm,
1384 u32 kvm_nr_mmu_pages)
1386 if (kvm_nr_mmu_pages < KVM_MIN_ALLOC_MMU_PAGES)
1389 down_write(&kvm->slots_lock);
1391 kvm_mmu_change_mmu_pages(kvm, kvm_nr_mmu_pages);
1392 kvm->arch.n_requested_mmu_pages = kvm_nr_mmu_pages;
1394 up_write(&kvm->slots_lock);
1398 static int kvm_vm_ioctl_get_nr_mmu_pages(struct kvm *kvm)
1400 return kvm->arch.n_alloc_mmu_pages;
1403 gfn_t unalias_gfn(struct kvm *kvm, gfn_t gfn)
1406 struct kvm_mem_alias *alias;
1408 for (i = 0; i < kvm->arch.naliases; ++i) {
1409 alias = &kvm->arch.aliases[i];
1410 if (gfn >= alias->base_gfn
1411 && gfn < alias->base_gfn + alias->npages)
1412 return alias->target_gfn + gfn - alias->base_gfn;
1418 * Set a new alias region. Aliases map a portion of physical memory into
1419 * another portion. This is useful for memory windows, for example the PC
1422 static int kvm_vm_ioctl_set_memory_alias(struct kvm *kvm,
1423 struct kvm_memory_alias *alias)
1426 struct kvm_mem_alias *p;
1429 /* General sanity checks */
1430 if (alias->memory_size & (PAGE_SIZE - 1))
1432 if (alias->guest_phys_addr & (PAGE_SIZE - 1))
1434 if (alias->slot >= KVM_ALIAS_SLOTS)
1436 if (alias->guest_phys_addr + alias->memory_size
1437 < alias->guest_phys_addr)
1439 if (alias->target_phys_addr + alias->memory_size
1440 < alias->target_phys_addr)
1443 down_write(&kvm->slots_lock);
1445 p = &kvm->arch.aliases[alias->slot];
1446 p->base_gfn = alias->guest_phys_addr >> PAGE_SHIFT;
1447 p->npages = alias->memory_size >> PAGE_SHIFT;
1448 p->target_gfn = alias->target_phys_addr >> PAGE_SHIFT;
1450 for (n = KVM_ALIAS_SLOTS; n > 0; --n)
1451 if (kvm->arch.aliases[n - 1].npages)
1453 kvm->arch.naliases = n;
1455 kvm_mmu_zap_all(kvm);
1457 up_write(&kvm->slots_lock);
1465 static int kvm_vm_ioctl_get_irqchip(struct kvm *kvm, struct kvm_irqchip *chip)
1470 switch (chip->chip_id) {
1471 case KVM_IRQCHIP_PIC_MASTER:
1472 memcpy(&chip->chip.pic,
1473 &pic_irqchip(kvm)->pics[0],
1474 sizeof(struct kvm_pic_state));
1476 case KVM_IRQCHIP_PIC_SLAVE:
1477 memcpy(&chip->chip.pic,
1478 &pic_irqchip(kvm)->pics[1],
1479 sizeof(struct kvm_pic_state));
1481 case KVM_IRQCHIP_IOAPIC:
1482 memcpy(&chip->chip.ioapic,
1483 ioapic_irqchip(kvm),
1484 sizeof(struct kvm_ioapic_state));
1493 static int kvm_vm_ioctl_set_irqchip(struct kvm *kvm, struct kvm_irqchip *chip)
1498 switch (chip->chip_id) {
1499 case KVM_IRQCHIP_PIC_MASTER:
1500 memcpy(&pic_irqchip(kvm)->pics[0],
1502 sizeof(struct kvm_pic_state));
1504 case KVM_IRQCHIP_PIC_SLAVE:
1505 memcpy(&pic_irqchip(kvm)->pics[1],
1507 sizeof(struct kvm_pic_state));
1509 case KVM_IRQCHIP_IOAPIC:
1510 memcpy(ioapic_irqchip(kvm),
1512 sizeof(struct kvm_ioapic_state));
1518 kvm_pic_update_irq(pic_irqchip(kvm));
1522 static int kvm_vm_ioctl_get_pit(struct kvm *kvm, struct kvm_pit_state *ps)
1526 memcpy(ps, &kvm->arch.vpit->pit_state, sizeof(struct kvm_pit_state));
1530 static int kvm_vm_ioctl_set_pit(struct kvm *kvm, struct kvm_pit_state *ps)
1534 memcpy(&kvm->arch.vpit->pit_state, ps, sizeof(struct kvm_pit_state));
1535 kvm_pit_load_count(kvm, 0, ps->channels[0].count);
1540 * Get (and clear) the dirty memory log for a memory slot.
1542 int kvm_vm_ioctl_get_dirty_log(struct kvm *kvm,
1543 struct kvm_dirty_log *log)
1547 struct kvm_memory_slot *memslot;
1550 down_write(&kvm->slots_lock);
1552 r = kvm_get_dirty_log(kvm, log, &is_dirty);
1556 /* If nothing is dirty, don't bother messing with page tables. */
1558 kvm_mmu_slot_remove_write_access(kvm, log->slot);
1559 kvm_flush_remote_tlbs(kvm);
1560 memslot = &kvm->memslots[log->slot];
1561 n = ALIGN(memslot->npages, BITS_PER_LONG) / 8;
1562 memset(memslot->dirty_bitmap, 0, n);
1566 up_write(&kvm->slots_lock);
1570 long kvm_arch_vm_ioctl(struct file *filp,
1571 unsigned int ioctl, unsigned long arg)
1573 struct kvm *kvm = filp->private_data;
1574 void __user *argp = (void __user *)arg;
1578 case KVM_SET_TSS_ADDR:
1579 r = kvm_vm_ioctl_set_tss_addr(kvm, arg);
1583 case KVM_SET_MEMORY_REGION: {
1584 struct kvm_memory_region kvm_mem;
1585 struct kvm_userspace_memory_region kvm_userspace_mem;
1588 if (copy_from_user(&kvm_mem, argp, sizeof kvm_mem))
1590 kvm_userspace_mem.slot = kvm_mem.slot;
1591 kvm_userspace_mem.flags = kvm_mem.flags;
1592 kvm_userspace_mem.guest_phys_addr = kvm_mem.guest_phys_addr;
1593 kvm_userspace_mem.memory_size = kvm_mem.memory_size;
1594 r = kvm_vm_ioctl_set_memory_region(kvm, &kvm_userspace_mem, 0);
1599 case KVM_SET_NR_MMU_PAGES:
1600 r = kvm_vm_ioctl_set_nr_mmu_pages(kvm, arg);
1604 case KVM_GET_NR_MMU_PAGES:
1605 r = kvm_vm_ioctl_get_nr_mmu_pages(kvm);
1607 case KVM_SET_MEMORY_ALIAS: {
1608 struct kvm_memory_alias alias;
1611 if (copy_from_user(&alias, argp, sizeof alias))
1613 r = kvm_vm_ioctl_set_memory_alias(kvm, &alias);
1618 case KVM_CREATE_IRQCHIP:
1620 kvm->arch.vpic = kvm_create_pic(kvm);
1621 if (kvm->arch.vpic) {
1622 r = kvm_ioapic_init(kvm);
1624 kfree(kvm->arch.vpic);
1625 kvm->arch.vpic = NULL;
1631 case KVM_CREATE_PIT:
1633 kvm->arch.vpit = kvm_create_pit(kvm);
1637 case KVM_IRQ_LINE: {
1638 struct kvm_irq_level irq_event;
1641 if (copy_from_user(&irq_event, argp, sizeof irq_event))
1643 if (irqchip_in_kernel(kvm)) {
1644 mutex_lock(&kvm->lock);
1645 if (irq_event.irq < 16)
1646 kvm_pic_set_irq(pic_irqchip(kvm),
1649 kvm_ioapic_set_irq(kvm->arch.vioapic,
1652 mutex_unlock(&kvm->lock);
1657 case KVM_GET_IRQCHIP: {
1658 /* 0: PIC master, 1: PIC slave, 2: IOAPIC */
1659 struct kvm_irqchip chip;
1662 if (copy_from_user(&chip, argp, sizeof chip))
1665 if (!irqchip_in_kernel(kvm))
1667 r = kvm_vm_ioctl_get_irqchip(kvm, &chip);
1671 if (copy_to_user(argp, &chip, sizeof chip))
1676 case KVM_SET_IRQCHIP: {
1677 /* 0: PIC master, 1: PIC slave, 2: IOAPIC */
1678 struct kvm_irqchip chip;
1681 if (copy_from_user(&chip, argp, sizeof chip))
1684 if (!irqchip_in_kernel(kvm))
1686 r = kvm_vm_ioctl_set_irqchip(kvm, &chip);
1693 struct kvm_pit_state ps;
1695 if (copy_from_user(&ps, argp, sizeof ps))
1698 if (!kvm->arch.vpit)
1700 r = kvm_vm_ioctl_get_pit(kvm, &ps);
1704 if (copy_to_user(argp, &ps, sizeof ps))
1710 struct kvm_pit_state ps;
1712 if (copy_from_user(&ps, argp, sizeof ps))
1715 if (!kvm->arch.vpit)
1717 r = kvm_vm_ioctl_set_pit(kvm, &ps);
1730 static void kvm_init_msr_list(void)
1735 for (i = j = 0; i < ARRAY_SIZE(msrs_to_save); i++) {
1736 if (rdmsr_safe(msrs_to_save[i], &dummy[0], &dummy[1]) < 0)
1739 msrs_to_save[j] = msrs_to_save[i];
1742 num_msrs_to_save = j;
1746 * Only apic need an MMIO device hook, so shortcut now..
1748 static struct kvm_io_device *vcpu_find_pervcpu_dev(struct kvm_vcpu *vcpu,
1751 struct kvm_io_device *dev;
1753 if (vcpu->arch.apic) {
1754 dev = &vcpu->arch.apic->dev;
1755 if (dev->in_range(dev, addr))
1762 static struct kvm_io_device *vcpu_find_mmio_dev(struct kvm_vcpu *vcpu,
1765 struct kvm_io_device *dev;
1767 dev = vcpu_find_pervcpu_dev(vcpu, addr);
1769 dev = kvm_io_bus_find_dev(&vcpu->kvm->mmio_bus, addr);
1773 int emulator_read_std(unsigned long addr,
1776 struct kvm_vcpu *vcpu)
1779 int r = X86EMUL_CONTINUE;
1782 gpa_t gpa = vcpu->arch.mmu.gva_to_gpa(vcpu, addr);
1783 unsigned offset = addr & (PAGE_SIZE-1);
1784 unsigned tocopy = min(bytes, (unsigned)PAGE_SIZE - offset);
1787 if (gpa == UNMAPPED_GVA) {
1788 r = X86EMUL_PROPAGATE_FAULT;
1791 ret = kvm_read_guest(vcpu->kvm, gpa, data, tocopy);
1793 r = X86EMUL_UNHANDLEABLE;
1804 EXPORT_SYMBOL_GPL(emulator_read_std);
1806 static int emulator_read_emulated(unsigned long addr,
1809 struct kvm_vcpu *vcpu)
1811 struct kvm_io_device *mmio_dev;
1814 if (vcpu->mmio_read_completed) {
1815 memcpy(val, vcpu->mmio_data, bytes);
1816 vcpu->mmio_read_completed = 0;
1817 return X86EMUL_CONTINUE;
1820 gpa = vcpu->arch.mmu.gva_to_gpa(vcpu, addr);
1822 /* For APIC access vmexit */
1823 if ((gpa & PAGE_MASK) == APIC_DEFAULT_PHYS_BASE)
1826 if (emulator_read_std(addr, val, bytes, vcpu)
1827 == X86EMUL_CONTINUE)
1828 return X86EMUL_CONTINUE;
1829 if (gpa == UNMAPPED_GVA)
1830 return X86EMUL_PROPAGATE_FAULT;
1834 * Is this MMIO handled locally?
1836 mutex_lock(&vcpu->kvm->lock);
1837 mmio_dev = vcpu_find_mmio_dev(vcpu, gpa);
1839 kvm_iodevice_read(mmio_dev, gpa, bytes, val);
1840 mutex_unlock(&vcpu->kvm->lock);
1841 return X86EMUL_CONTINUE;
1843 mutex_unlock(&vcpu->kvm->lock);
1845 vcpu->mmio_needed = 1;
1846 vcpu->mmio_phys_addr = gpa;
1847 vcpu->mmio_size = bytes;
1848 vcpu->mmio_is_write = 0;
1850 return X86EMUL_UNHANDLEABLE;
1853 int emulator_write_phys(struct kvm_vcpu *vcpu, gpa_t gpa,
1854 const void *val, int bytes)
1858 ret = kvm_write_guest(vcpu->kvm, gpa, val, bytes);
1861 kvm_mmu_pte_write(vcpu, gpa, val, bytes);
1865 static int emulator_write_emulated_onepage(unsigned long addr,
1868 struct kvm_vcpu *vcpu)
1870 struct kvm_io_device *mmio_dev;
1873 gpa = vcpu->arch.mmu.gva_to_gpa(vcpu, addr);
1875 if (gpa == UNMAPPED_GVA) {
1876 kvm_inject_page_fault(vcpu, addr, 2);
1877 return X86EMUL_PROPAGATE_FAULT;
1880 /* For APIC access vmexit */
1881 if ((gpa & PAGE_MASK) == APIC_DEFAULT_PHYS_BASE)
1884 if (emulator_write_phys(vcpu, gpa, val, bytes))
1885 return X86EMUL_CONTINUE;
1889 * Is this MMIO handled locally?
1891 mutex_lock(&vcpu->kvm->lock);
1892 mmio_dev = vcpu_find_mmio_dev(vcpu, gpa);
1894 kvm_iodevice_write(mmio_dev, gpa, bytes, val);
1895 mutex_unlock(&vcpu->kvm->lock);
1896 return X86EMUL_CONTINUE;
1898 mutex_unlock(&vcpu->kvm->lock);
1900 vcpu->mmio_needed = 1;
1901 vcpu->mmio_phys_addr = gpa;
1902 vcpu->mmio_size = bytes;
1903 vcpu->mmio_is_write = 1;
1904 memcpy(vcpu->mmio_data, val, bytes);
1906 return X86EMUL_CONTINUE;
1909 int emulator_write_emulated(unsigned long addr,
1912 struct kvm_vcpu *vcpu)
1914 /* Crossing a page boundary? */
1915 if (((addr + bytes - 1) ^ addr) & PAGE_MASK) {
1918 now = -addr & ~PAGE_MASK;
1919 rc = emulator_write_emulated_onepage(addr, val, now, vcpu);
1920 if (rc != X86EMUL_CONTINUE)
1926 return emulator_write_emulated_onepage(addr, val, bytes, vcpu);
1928 EXPORT_SYMBOL_GPL(emulator_write_emulated);
1930 static int emulator_cmpxchg_emulated(unsigned long addr,
1934 struct kvm_vcpu *vcpu)
1936 static int reported;
1940 printk(KERN_WARNING "kvm: emulating exchange as write\n");
1942 #ifndef CONFIG_X86_64
1943 /* guests cmpxchg8b have to be emulated atomically */
1950 gpa = vcpu->arch.mmu.gva_to_gpa(vcpu, addr);
1952 if (gpa == UNMAPPED_GVA ||
1953 (gpa & PAGE_MASK) == APIC_DEFAULT_PHYS_BASE)
1956 if (((gpa + bytes - 1) & PAGE_MASK) != (gpa & PAGE_MASK))
1961 down_read(¤t->mm->mmap_sem);
1962 page = gfn_to_page(vcpu->kvm, gpa >> PAGE_SHIFT);
1963 up_read(¤t->mm->mmap_sem);
1965 kaddr = kmap_atomic(page, KM_USER0);
1966 set_64bit((u64 *)(kaddr + offset_in_page(gpa)), val);
1967 kunmap_atomic(kaddr, KM_USER0);
1968 kvm_release_page_dirty(page);
1973 return emulator_write_emulated(addr, new, bytes, vcpu);
1976 static unsigned long get_segment_base(struct kvm_vcpu *vcpu, int seg)
1978 return kvm_x86_ops->get_segment_base(vcpu, seg);
1981 int emulate_invlpg(struct kvm_vcpu *vcpu, gva_t address)
1983 return X86EMUL_CONTINUE;
1986 int emulate_clts(struct kvm_vcpu *vcpu)
1988 KVMTRACE_0D(CLTS, vcpu, handler);
1989 kvm_x86_ops->set_cr0(vcpu, vcpu->arch.cr0 & ~X86_CR0_TS);
1990 return X86EMUL_CONTINUE;
1993 int emulator_get_dr(struct x86_emulate_ctxt *ctxt, int dr, unsigned long *dest)
1995 struct kvm_vcpu *vcpu = ctxt->vcpu;
1999 *dest = kvm_x86_ops->get_dr(vcpu, dr);
2000 return X86EMUL_CONTINUE;
2002 pr_unimpl(vcpu, "%s: unexpected dr %u\n", __func__, dr);
2003 return X86EMUL_UNHANDLEABLE;
2007 int emulator_set_dr(struct x86_emulate_ctxt *ctxt, int dr, unsigned long value)
2009 unsigned long mask = (ctxt->mode == X86EMUL_MODE_PROT64) ? ~0ULL : ~0U;
2012 kvm_x86_ops->set_dr(ctxt->vcpu, dr, value & mask, &exception);
2014 /* FIXME: better handling */
2015 return X86EMUL_UNHANDLEABLE;
2017 return X86EMUL_CONTINUE;
2020 void kvm_report_emulation_failure(struct kvm_vcpu *vcpu, const char *context)
2022 static int reported;
2024 unsigned long rip = vcpu->arch.rip;
2025 unsigned long rip_linear;
2027 rip_linear = rip + get_segment_base(vcpu, VCPU_SREG_CS);
2032 emulator_read_std(rip_linear, (void *)opcodes, 4, vcpu);
2034 printk(KERN_ERR "emulation failed (%s) rip %lx %02x %02x %02x %02x\n",
2035 context, rip, opcodes[0], opcodes[1], opcodes[2], opcodes[3]);
2038 EXPORT_SYMBOL_GPL(kvm_report_emulation_failure);
2040 static struct x86_emulate_ops emulate_ops = {
2041 .read_std = emulator_read_std,
2042 .read_emulated = emulator_read_emulated,
2043 .write_emulated = emulator_write_emulated,
2044 .cmpxchg_emulated = emulator_cmpxchg_emulated,
2047 int emulate_instruction(struct kvm_vcpu *vcpu,
2048 struct kvm_run *run,
2054 struct decode_cache *c;
2056 vcpu->arch.mmio_fault_cr2 = cr2;
2057 kvm_x86_ops->cache_regs(vcpu);
2059 vcpu->mmio_is_write = 0;
2060 vcpu->arch.pio.string = 0;
2062 if (!(emulation_type & EMULTYPE_NO_DECODE)) {
2064 kvm_x86_ops->get_cs_db_l_bits(vcpu, &cs_db, &cs_l);
2066 vcpu->arch.emulate_ctxt.vcpu = vcpu;
2067 vcpu->arch.emulate_ctxt.eflags = kvm_x86_ops->get_rflags(vcpu);
2068 vcpu->arch.emulate_ctxt.mode =
2069 (vcpu->arch.emulate_ctxt.eflags & X86_EFLAGS_VM)
2070 ? X86EMUL_MODE_REAL : cs_l
2071 ? X86EMUL_MODE_PROT64 : cs_db
2072 ? X86EMUL_MODE_PROT32 : X86EMUL_MODE_PROT16;
2074 if (vcpu->arch.emulate_ctxt.mode == X86EMUL_MODE_PROT64) {
2075 vcpu->arch.emulate_ctxt.cs_base = 0;
2076 vcpu->arch.emulate_ctxt.ds_base = 0;
2077 vcpu->arch.emulate_ctxt.es_base = 0;
2078 vcpu->arch.emulate_ctxt.ss_base = 0;
2080 vcpu->arch.emulate_ctxt.cs_base =
2081 get_segment_base(vcpu, VCPU_SREG_CS);
2082 vcpu->arch.emulate_ctxt.ds_base =
2083 get_segment_base(vcpu, VCPU_SREG_DS);
2084 vcpu->arch.emulate_ctxt.es_base =
2085 get_segment_base(vcpu, VCPU_SREG_ES);
2086 vcpu->arch.emulate_ctxt.ss_base =
2087 get_segment_base(vcpu, VCPU_SREG_SS);
2090 vcpu->arch.emulate_ctxt.gs_base =
2091 get_segment_base(vcpu, VCPU_SREG_GS);
2092 vcpu->arch.emulate_ctxt.fs_base =
2093 get_segment_base(vcpu, VCPU_SREG_FS);
2095 r = x86_decode_insn(&vcpu->arch.emulate_ctxt, &emulate_ops);
2097 /* Reject the instructions other than VMCALL/VMMCALL when
2098 * try to emulate invalid opcode */
2099 c = &vcpu->arch.emulate_ctxt.decode;
2100 if ((emulation_type & EMULTYPE_TRAP_UD) &&
2101 (!(c->twobyte && c->b == 0x01 &&
2102 (c->modrm_reg == 0 || c->modrm_reg == 3) &&
2103 c->modrm_mod == 3 && c->modrm_rm == 1)))
2104 return EMULATE_FAIL;
2106 ++vcpu->stat.insn_emulation;
2108 ++vcpu->stat.insn_emulation_fail;
2109 if (kvm_mmu_unprotect_page_virt(vcpu, cr2))
2110 return EMULATE_DONE;
2111 return EMULATE_FAIL;
2115 r = x86_emulate_insn(&vcpu->arch.emulate_ctxt, &emulate_ops);
2117 if (vcpu->arch.pio.string)
2118 return EMULATE_DO_MMIO;
2120 if ((r || vcpu->mmio_is_write) && run) {
2121 run->exit_reason = KVM_EXIT_MMIO;
2122 run->mmio.phys_addr = vcpu->mmio_phys_addr;
2123 memcpy(run->mmio.data, vcpu->mmio_data, 8);
2124 run->mmio.len = vcpu->mmio_size;
2125 run->mmio.is_write = vcpu->mmio_is_write;
2129 if (kvm_mmu_unprotect_page_virt(vcpu, cr2))
2130 return EMULATE_DONE;
2131 if (!vcpu->mmio_needed) {
2132 kvm_report_emulation_failure(vcpu, "mmio");
2133 return EMULATE_FAIL;
2135 return EMULATE_DO_MMIO;
2138 kvm_x86_ops->decache_regs(vcpu);
2139 kvm_x86_ops->set_rflags(vcpu, vcpu->arch.emulate_ctxt.eflags);
2141 if (vcpu->mmio_is_write) {
2142 vcpu->mmio_needed = 0;
2143 return EMULATE_DO_MMIO;
2146 return EMULATE_DONE;
2148 EXPORT_SYMBOL_GPL(emulate_instruction);
2150 static void free_pio_guest_pages(struct kvm_vcpu *vcpu)
2154 for (i = 0; i < ARRAY_SIZE(vcpu->arch.pio.guest_pages); ++i)
2155 if (vcpu->arch.pio.guest_pages[i]) {
2156 kvm_release_page_dirty(vcpu->arch.pio.guest_pages[i]);
2157 vcpu->arch.pio.guest_pages[i] = NULL;
2161 static int pio_copy_data(struct kvm_vcpu *vcpu)
2163 void *p = vcpu->arch.pio_data;
2166 int nr_pages = vcpu->arch.pio.guest_pages[1] ? 2 : 1;
2168 q = vmap(vcpu->arch.pio.guest_pages, nr_pages, VM_READ|VM_WRITE,
2171 free_pio_guest_pages(vcpu);
2174 q += vcpu->arch.pio.guest_page_offset;
2175 bytes = vcpu->arch.pio.size * vcpu->arch.pio.cur_count;
2176 if (vcpu->arch.pio.in)
2177 memcpy(q, p, bytes);
2179 memcpy(p, q, bytes);
2180 q -= vcpu->arch.pio.guest_page_offset;
2182 free_pio_guest_pages(vcpu);
2186 int complete_pio(struct kvm_vcpu *vcpu)
2188 struct kvm_pio_request *io = &vcpu->arch.pio;
2192 kvm_x86_ops->cache_regs(vcpu);
2196 memcpy(&vcpu->arch.regs[VCPU_REGS_RAX], vcpu->arch.pio_data,
2200 r = pio_copy_data(vcpu);
2202 kvm_x86_ops->cache_regs(vcpu);
2209 delta *= io->cur_count;
2211 * The size of the register should really depend on
2212 * current address size.
2214 vcpu->arch.regs[VCPU_REGS_RCX] -= delta;
2220 vcpu->arch.regs[VCPU_REGS_RDI] += delta;
2222 vcpu->arch.regs[VCPU_REGS_RSI] += delta;
2225 kvm_x86_ops->decache_regs(vcpu);
2227 io->count -= io->cur_count;
2233 static void kernel_pio(struct kvm_io_device *pio_dev,
2234 struct kvm_vcpu *vcpu,
2237 /* TODO: String I/O for in kernel device */
2239 mutex_lock(&vcpu->kvm->lock);
2240 if (vcpu->arch.pio.in)
2241 kvm_iodevice_read(pio_dev, vcpu->arch.pio.port,
2242 vcpu->arch.pio.size,
2245 kvm_iodevice_write(pio_dev, vcpu->arch.pio.port,
2246 vcpu->arch.pio.size,
2248 mutex_unlock(&vcpu->kvm->lock);
2251 static void pio_string_write(struct kvm_io_device *pio_dev,
2252 struct kvm_vcpu *vcpu)
2254 struct kvm_pio_request *io = &vcpu->arch.pio;
2255 void *pd = vcpu->arch.pio_data;
2258 mutex_lock(&vcpu->kvm->lock);
2259 for (i = 0; i < io->cur_count; i++) {
2260 kvm_iodevice_write(pio_dev, io->port,
2265 mutex_unlock(&vcpu->kvm->lock);
2268 static struct kvm_io_device *vcpu_find_pio_dev(struct kvm_vcpu *vcpu,
2271 return kvm_io_bus_find_dev(&vcpu->kvm->pio_bus, addr);
2274 int kvm_emulate_pio(struct kvm_vcpu *vcpu, struct kvm_run *run, int in,
2275 int size, unsigned port)
2277 struct kvm_io_device *pio_dev;
2279 vcpu->run->exit_reason = KVM_EXIT_IO;
2280 vcpu->run->io.direction = in ? KVM_EXIT_IO_IN : KVM_EXIT_IO_OUT;
2281 vcpu->run->io.size = vcpu->arch.pio.size = size;
2282 vcpu->run->io.data_offset = KVM_PIO_PAGE_OFFSET * PAGE_SIZE;
2283 vcpu->run->io.count = vcpu->arch.pio.count = vcpu->arch.pio.cur_count = 1;
2284 vcpu->run->io.port = vcpu->arch.pio.port = port;
2285 vcpu->arch.pio.in = in;
2286 vcpu->arch.pio.string = 0;
2287 vcpu->arch.pio.down = 0;
2288 vcpu->arch.pio.guest_page_offset = 0;
2289 vcpu->arch.pio.rep = 0;
2291 if (vcpu->run->io.direction == KVM_EXIT_IO_IN)
2292 KVMTRACE_2D(IO_READ, vcpu, vcpu->run->io.port, (u32)size,
2295 KVMTRACE_2D(IO_WRITE, vcpu, vcpu->run->io.port, (u32)size,
2298 kvm_x86_ops->cache_regs(vcpu);
2299 memcpy(vcpu->arch.pio_data, &vcpu->arch.regs[VCPU_REGS_RAX], 4);
2300 kvm_x86_ops->decache_regs(vcpu);
2302 kvm_x86_ops->skip_emulated_instruction(vcpu);
2304 pio_dev = vcpu_find_pio_dev(vcpu, port);
2306 kernel_pio(pio_dev, vcpu, vcpu->arch.pio_data);
2312 EXPORT_SYMBOL_GPL(kvm_emulate_pio);
2314 int kvm_emulate_pio_string(struct kvm_vcpu *vcpu, struct kvm_run *run, int in,
2315 int size, unsigned long count, int down,
2316 gva_t address, int rep, unsigned port)
2318 unsigned now, in_page;
2322 struct kvm_io_device *pio_dev;
2324 vcpu->run->exit_reason = KVM_EXIT_IO;
2325 vcpu->run->io.direction = in ? KVM_EXIT_IO_IN : KVM_EXIT_IO_OUT;
2326 vcpu->run->io.size = vcpu->arch.pio.size = size;
2327 vcpu->run->io.data_offset = KVM_PIO_PAGE_OFFSET * PAGE_SIZE;
2328 vcpu->run->io.count = vcpu->arch.pio.count = vcpu->arch.pio.cur_count = count;
2329 vcpu->run->io.port = vcpu->arch.pio.port = port;
2330 vcpu->arch.pio.in = in;
2331 vcpu->arch.pio.string = 1;
2332 vcpu->arch.pio.down = down;
2333 vcpu->arch.pio.guest_page_offset = offset_in_page(address);
2334 vcpu->arch.pio.rep = rep;
2336 if (vcpu->run->io.direction == KVM_EXIT_IO_IN)
2337 KVMTRACE_2D(IO_READ, vcpu, vcpu->run->io.port, (u32)size,
2340 KVMTRACE_2D(IO_WRITE, vcpu, vcpu->run->io.port, (u32)size,
2344 kvm_x86_ops->skip_emulated_instruction(vcpu);
2349 in_page = PAGE_SIZE - offset_in_page(address);
2351 in_page = offset_in_page(address) + size;
2352 now = min(count, (unsigned long)in_page / size);
2355 * String I/O straddles page boundary. Pin two guest pages
2356 * so that we satisfy atomicity constraints. Do just one
2357 * transaction to avoid complexity.
2364 * String I/O in reverse. Yuck. Kill the guest, fix later.
2366 pr_unimpl(vcpu, "guest string pio down\n");
2367 kvm_inject_gp(vcpu, 0);
2370 vcpu->run->io.count = now;
2371 vcpu->arch.pio.cur_count = now;
2373 if (vcpu->arch.pio.cur_count == vcpu->arch.pio.count)
2374 kvm_x86_ops->skip_emulated_instruction(vcpu);
2376 for (i = 0; i < nr_pages; ++i) {
2377 page = gva_to_page(vcpu, address + i * PAGE_SIZE);
2378 vcpu->arch.pio.guest_pages[i] = page;
2380 kvm_inject_gp(vcpu, 0);
2381 free_pio_guest_pages(vcpu);
2386 pio_dev = vcpu_find_pio_dev(vcpu, port);
2387 if (!vcpu->arch.pio.in) {
2388 /* string PIO write */
2389 ret = pio_copy_data(vcpu);
2390 if (ret >= 0 && pio_dev) {
2391 pio_string_write(pio_dev, vcpu);
2393 if (vcpu->arch.pio.count == 0)
2397 pr_unimpl(vcpu, "no string pio read support yet, "
2398 "port %x size %d count %ld\n",
2403 EXPORT_SYMBOL_GPL(kvm_emulate_pio_string);
2405 int kvm_arch_init(void *opaque)
2408 struct kvm_x86_ops *ops = (struct kvm_x86_ops *)opaque;
2411 printk(KERN_ERR "kvm: already loaded the other module\n");
2416 if (!ops->cpu_has_kvm_support()) {
2417 printk(KERN_ERR "kvm: no hardware support\n");
2421 if (ops->disabled_by_bios()) {
2422 printk(KERN_ERR "kvm: disabled by bios\n");
2427 r = kvm_mmu_module_init();
2431 kvm_init_msr_list();
2434 kvm_mmu_set_nonpresent_ptes(0ull, 0ull);
2435 kvm_mmu_set_base_ptes(PT_PRESENT_MASK);
2436 kvm_mmu_set_mask_ptes(PT_USER_MASK, PT_ACCESSED_MASK,
2437 PT_DIRTY_MASK, PT64_NX_MASK, 0);
2444 void kvm_arch_exit(void)
2447 kvm_mmu_module_exit();
2450 int kvm_emulate_halt(struct kvm_vcpu *vcpu)
2452 ++vcpu->stat.halt_exits;
2453 KVMTRACE_0D(HLT, vcpu, handler);
2454 if (irqchip_in_kernel(vcpu->kvm)) {
2455 vcpu->arch.mp_state = KVM_MP_STATE_HALTED;
2456 up_read(&vcpu->kvm->slots_lock);
2457 kvm_vcpu_block(vcpu);
2458 down_read(&vcpu->kvm->slots_lock);
2459 if (vcpu->arch.mp_state != KVM_MP_STATE_RUNNABLE)
2463 vcpu->run->exit_reason = KVM_EXIT_HLT;
2467 EXPORT_SYMBOL_GPL(kvm_emulate_halt);
2469 static inline gpa_t hc_gpa(struct kvm_vcpu *vcpu, unsigned long a0,
2472 if (is_long_mode(vcpu))
2475 return a0 | ((gpa_t)a1 << 32);
2478 int kvm_emulate_hypercall(struct kvm_vcpu *vcpu)
2480 unsigned long nr, a0, a1, a2, a3, ret;
2483 kvm_x86_ops->cache_regs(vcpu);
2485 nr = vcpu->arch.regs[VCPU_REGS_RAX];
2486 a0 = vcpu->arch.regs[VCPU_REGS_RBX];
2487 a1 = vcpu->arch.regs[VCPU_REGS_RCX];
2488 a2 = vcpu->arch.regs[VCPU_REGS_RDX];
2489 a3 = vcpu->arch.regs[VCPU_REGS_RSI];
2491 KVMTRACE_1D(VMMCALL, vcpu, (u32)nr, handler);
2493 if (!is_long_mode(vcpu)) {
2502 case KVM_HC_VAPIC_POLL_IRQ:
2506 r = kvm_pv_mmu_op(vcpu, a0, hc_gpa(vcpu, a1, a2), &ret);
2512 vcpu->arch.regs[VCPU_REGS_RAX] = ret;
2513 kvm_x86_ops->decache_regs(vcpu);
2514 ++vcpu->stat.hypercalls;
2517 EXPORT_SYMBOL_GPL(kvm_emulate_hypercall);
2519 int kvm_fix_hypercall(struct kvm_vcpu *vcpu)
2521 char instruction[3];
2526 * Blow out the MMU to ensure that no other VCPU has an active mapping
2527 * to ensure that the updated hypercall appears atomically across all
2530 kvm_mmu_zap_all(vcpu->kvm);
2532 kvm_x86_ops->cache_regs(vcpu);
2533 kvm_x86_ops->patch_hypercall(vcpu, instruction);
2534 if (emulator_write_emulated(vcpu->arch.rip, instruction, 3, vcpu)
2535 != X86EMUL_CONTINUE)
2541 static u64 mk_cr_64(u64 curr_cr, u32 new_val)
2543 return (curr_cr & ~((1ULL << 32) - 1)) | new_val;
2546 void realmode_lgdt(struct kvm_vcpu *vcpu, u16 limit, unsigned long base)
2548 struct descriptor_table dt = { limit, base };
2550 kvm_x86_ops->set_gdt(vcpu, &dt);
2553 void realmode_lidt(struct kvm_vcpu *vcpu, u16 limit, unsigned long base)
2555 struct descriptor_table dt = { limit, base };
2557 kvm_x86_ops->set_idt(vcpu, &dt);
2560 void realmode_lmsw(struct kvm_vcpu *vcpu, unsigned long msw,
2561 unsigned long *rflags)
2563 kvm_lmsw(vcpu, msw);
2564 *rflags = kvm_x86_ops->get_rflags(vcpu);
2567 unsigned long realmode_get_cr(struct kvm_vcpu *vcpu, int cr)
2569 unsigned long value;
2571 kvm_x86_ops->decache_cr4_guest_bits(vcpu);
2574 value = vcpu->arch.cr0;
2577 value = vcpu->arch.cr2;
2580 value = vcpu->arch.cr3;
2583 value = vcpu->arch.cr4;
2586 value = kvm_get_cr8(vcpu);
2589 vcpu_printf(vcpu, "%s: unexpected cr %u\n", __func__, cr);
2592 KVMTRACE_3D(CR_READ, vcpu, (u32)cr, (u32)value,
2593 (u32)((u64)value >> 32), handler);
2598 void realmode_set_cr(struct kvm_vcpu *vcpu, int cr, unsigned long val,
2599 unsigned long *rflags)
2601 KVMTRACE_3D(CR_WRITE, vcpu, (u32)cr, (u32)val,
2602 (u32)((u64)val >> 32), handler);
2606 kvm_set_cr0(vcpu, mk_cr_64(vcpu->arch.cr0, val));
2607 *rflags = kvm_x86_ops->get_rflags(vcpu);
2610 vcpu->arch.cr2 = val;
2613 kvm_set_cr3(vcpu, val);
2616 kvm_set_cr4(vcpu, mk_cr_64(vcpu->arch.cr4, val));
2619 kvm_set_cr8(vcpu, val & 0xfUL);
2622 vcpu_printf(vcpu, "%s: unexpected cr %u\n", __func__, cr);
2626 static int move_to_next_stateful_cpuid_entry(struct kvm_vcpu *vcpu, int i)
2628 struct kvm_cpuid_entry2 *e = &vcpu->arch.cpuid_entries[i];
2629 int j, nent = vcpu->arch.cpuid_nent;
2631 e->flags &= ~KVM_CPUID_FLAG_STATE_READ_NEXT;
2632 /* when no next entry is found, the current entry[i] is reselected */
2633 for (j = i + 1; j == i; j = (j + 1) % nent) {
2634 struct kvm_cpuid_entry2 *ej = &vcpu->arch.cpuid_entries[j];
2635 if (ej->function == e->function) {
2636 ej->flags |= KVM_CPUID_FLAG_STATE_READ_NEXT;
2640 return 0; /* silence gcc, even though control never reaches here */
2643 /* find an entry with matching function, matching index (if needed), and that
2644 * should be read next (if it's stateful) */
2645 static int is_matching_cpuid_entry(struct kvm_cpuid_entry2 *e,
2646 u32 function, u32 index)
2648 if (e->function != function)
2650 if ((e->flags & KVM_CPUID_FLAG_SIGNIFCANT_INDEX) && e->index != index)
2652 if ((e->flags & KVM_CPUID_FLAG_STATEFUL_FUNC) &&
2653 !(e->flags & KVM_CPUID_FLAG_STATE_READ_NEXT))
2658 void kvm_emulate_cpuid(struct kvm_vcpu *vcpu)
2661 u32 function, index;
2662 struct kvm_cpuid_entry2 *e, *best;
2664 kvm_x86_ops->cache_regs(vcpu);
2665 function = vcpu->arch.regs[VCPU_REGS_RAX];
2666 index = vcpu->arch.regs[VCPU_REGS_RCX];
2667 vcpu->arch.regs[VCPU_REGS_RAX] = 0;
2668 vcpu->arch.regs[VCPU_REGS_RBX] = 0;
2669 vcpu->arch.regs[VCPU_REGS_RCX] = 0;
2670 vcpu->arch.regs[VCPU_REGS_RDX] = 0;
2672 for (i = 0; i < vcpu->arch.cpuid_nent; ++i) {
2673 e = &vcpu->arch.cpuid_entries[i];
2674 if (is_matching_cpuid_entry(e, function, index)) {
2675 if (e->flags & KVM_CPUID_FLAG_STATEFUL_FUNC)
2676 move_to_next_stateful_cpuid_entry(vcpu, i);
2681 * Both basic or both extended?
2683 if (((e->function ^ function) & 0x80000000) == 0)
2684 if (!best || e->function > best->function)
2688 vcpu->arch.regs[VCPU_REGS_RAX] = best->eax;
2689 vcpu->arch.regs[VCPU_REGS_RBX] = best->ebx;
2690 vcpu->arch.regs[VCPU_REGS_RCX] = best->ecx;
2691 vcpu->arch.regs[VCPU_REGS_RDX] = best->edx;
2693 kvm_x86_ops->decache_regs(vcpu);
2694 kvm_x86_ops->skip_emulated_instruction(vcpu);
2695 KVMTRACE_5D(CPUID, vcpu, function,
2696 (u32)vcpu->arch.regs[VCPU_REGS_RAX],
2697 (u32)vcpu->arch.regs[VCPU_REGS_RBX],
2698 (u32)vcpu->arch.regs[VCPU_REGS_RCX],
2699 (u32)vcpu->arch.regs[VCPU_REGS_RDX], handler);
2701 EXPORT_SYMBOL_GPL(kvm_emulate_cpuid);
2704 * Check if userspace requested an interrupt window, and that the
2705 * interrupt window is open.
2707 * No need to exit to userspace if we already have an interrupt queued.
2709 static int dm_request_for_irq_injection(struct kvm_vcpu *vcpu,
2710 struct kvm_run *kvm_run)
2712 return (!vcpu->arch.irq_summary &&
2713 kvm_run->request_interrupt_window &&
2714 vcpu->arch.interrupt_window_open &&
2715 (kvm_x86_ops->get_rflags(vcpu) & X86_EFLAGS_IF));
2718 static void post_kvm_run_save(struct kvm_vcpu *vcpu,
2719 struct kvm_run *kvm_run)
2721 kvm_run->if_flag = (kvm_x86_ops->get_rflags(vcpu) & X86_EFLAGS_IF) != 0;
2722 kvm_run->cr8 = kvm_get_cr8(vcpu);
2723 kvm_run->apic_base = kvm_get_apic_base(vcpu);
2724 if (irqchip_in_kernel(vcpu->kvm))
2725 kvm_run->ready_for_interrupt_injection = 1;
2727 kvm_run->ready_for_interrupt_injection =
2728 (vcpu->arch.interrupt_window_open &&
2729 vcpu->arch.irq_summary == 0);
2732 static void vapic_enter(struct kvm_vcpu *vcpu)
2734 struct kvm_lapic *apic = vcpu->arch.apic;
2737 if (!apic || !apic->vapic_addr)
2740 down_read(¤t->mm->mmap_sem);
2741 page = gfn_to_page(vcpu->kvm, apic->vapic_addr >> PAGE_SHIFT);
2742 up_read(¤t->mm->mmap_sem);
2744 vcpu->arch.apic->vapic_page = page;
2747 static void vapic_exit(struct kvm_vcpu *vcpu)
2749 struct kvm_lapic *apic = vcpu->arch.apic;
2751 if (!apic || !apic->vapic_addr)
2754 kvm_release_page_dirty(apic->vapic_page);
2755 mark_page_dirty(vcpu->kvm, apic->vapic_addr >> PAGE_SHIFT);
2758 static int __vcpu_run(struct kvm_vcpu *vcpu, struct kvm_run *kvm_run)
2762 if (unlikely(vcpu->arch.mp_state == KVM_MP_STATE_SIPI_RECEIVED)) {
2763 pr_debug("vcpu %d received sipi with vector # %x\n",
2764 vcpu->vcpu_id, vcpu->arch.sipi_vector);
2765 kvm_lapic_reset(vcpu);
2766 r = kvm_x86_ops->vcpu_reset(vcpu);
2769 vcpu->arch.mp_state = KVM_MP_STATE_RUNNABLE;
2772 down_read(&vcpu->kvm->slots_lock);
2776 if (vcpu->guest_debug.enabled)
2777 kvm_x86_ops->guest_debug_pre(vcpu);
2781 if (test_and_clear_bit(KVM_REQ_MMU_RELOAD, &vcpu->requests))
2782 kvm_mmu_unload(vcpu);
2784 r = kvm_mmu_reload(vcpu);
2788 if (vcpu->requests) {
2789 if (test_and_clear_bit(KVM_REQ_MIGRATE_TIMER, &vcpu->requests))
2790 __kvm_migrate_timers(vcpu);
2791 if (test_and_clear_bit(KVM_REQ_TLB_FLUSH, &vcpu->requests))
2792 kvm_x86_ops->tlb_flush(vcpu);
2793 if (test_and_clear_bit(KVM_REQ_REPORT_TPR_ACCESS,
2795 kvm_run->exit_reason = KVM_EXIT_TPR_ACCESS;
2799 if (test_and_clear_bit(KVM_REQ_TRIPLE_FAULT, &vcpu->requests)) {
2800 kvm_run->exit_reason = KVM_EXIT_SHUTDOWN;
2806 clear_bit(KVM_REQ_PENDING_TIMER, &vcpu->requests);
2807 kvm_inject_pending_timer_irqs(vcpu);
2811 kvm_x86_ops->prepare_guest_switch(vcpu);
2812 kvm_load_guest_fpu(vcpu);
2814 local_irq_disable();
2816 if (vcpu->requests || need_resched()) {
2823 if (signal_pending(current)) {
2827 kvm_run->exit_reason = KVM_EXIT_INTR;
2828 ++vcpu->stat.signal_exits;
2832 vcpu->guest_mode = 1;
2834 * Make sure that guest_mode assignment won't happen after
2835 * testing the pending IRQ vector bitmap.
2839 if (vcpu->arch.exception.pending)
2840 __queue_exception(vcpu);
2841 else if (irqchip_in_kernel(vcpu->kvm))
2842 kvm_x86_ops->inject_pending_irq(vcpu);
2844 kvm_x86_ops->inject_pending_vectors(vcpu, kvm_run);
2846 kvm_lapic_sync_to_vapic(vcpu);
2848 up_read(&vcpu->kvm->slots_lock);
2853 KVMTRACE_0D(VMENTRY, vcpu, entryexit);
2854 kvm_x86_ops->run(vcpu, kvm_run);
2856 vcpu->guest_mode = 0;
2862 * We must have an instruction between local_irq_enable() and
2863 * kvm_guest_exit(), so the timer interrupt isn't delayed by
2864 * the interrupt shadow. The stat.exits increment will do nicely.
2865 * But we need to prevent reordering, hence this barrier():
2873 down_read(&vcpu->kvm->slots_lock);
2876 * Profile KVM exit RIPs:
2878 if (unlikely(prof_on == KVM_PROFILING)) {
2879 kvm_x86_ops->cache_regs(vcpu);
2880 profile_hit(KVM_PROFILING, (void *)vcpu->arch.rip);
2883 if (vcpu->arch.exception.pending && kvm_x86_ops->exception_injected(vcpu))
2884 vcpu->arch.exception.pending = false;
2886 kvm_lapic_sync_from_vapic(vcpu);
2888 r = kvm_x86_ops->handle_exit(kvm_run, vcpu);
2891 if (dm_request_for_irq_injection(vcpu, kvm_run)) {
2893 kvm_run->exit_reason = KVM_EXIT_INTR;
2894 ++vcpu->stat.request_irq_exits;
2897 if (!need_resched())
2902 up_read(&vcpu->kvm->slots_lock);
2905 down_read(&vcpu->kvm->slots_lock);
2909 post_kvm_run_save(vcpu, kvm_run);
2911 down_read(&vcpu->kvm->slots_lock);
2913 up_read(&vcpu->kvm->slots_lock);
2918 int kvm_arch_vcpu_ioctl_run(struct kvm_vcpu *vcpu, struct kvm_run *kvm_run)
2925 if (unlikely(vcpu->arch.mp_state == KVM_MP_STATE_UNINITIALIZED)) {
2926 kvm_vcpu_block(vcpu);
2931 if (vcpu->sigset_active)
2932 sigprocmask(SIG_SETMASK, &vcpu->sigset, &sigsaved);
2934 /* re-sync apic's tpr */
2935 if (!irqchip_in_kernel(vcpu->kvm))
2936 kvm_set_cr8(vcpu, kvm_run->cr8);
2938 if (vcpu->arch.pio.cur_count) {
2939 r = complete_pio(vcpu);
2943 #if CONFIG_HAS_IOMEM
2944 if (vcpu->mmio_needed) {
2945 memcpy(vcpu->mmio_data, kvm_run->mmio.data, 8);
2946 vcpu->mmio_read_completed = 1;
2947 vcpu->mmio_needed = 0;
2949 down_read(&vcpu->kvm->slots_lock);
2950 r = emulate_instruction(vcpu, kvm_run,
2951 vcpu->arch.mmio_fault_cr2, 0,
2952 EMULTYPE_NO_DECODE);
2953 up_read(&vcpu->kvm->slots_lock);
2954 if (r == EMULATE_DO_MMIO) {
2956 * Read-modify-write. Back to userspace.
2963 if (kvm_run->exit_reason == KVM_EXIT_HYPERCALL) {
2964 kvm_x86_ops->cache_regs(vcpu);
2965 vcpu->arch.regs[VCPU_REGS_RAX] = kvm_run->hypercall.ret;
2966 kvm_x86_ops->decache_regs(vcpu);
2969 r = __vcpu_run(vcpu, kvm_run);
2972 if (vcpu->sigset_active)
2973 sigprocmask(SIG_SETMASK, &sigsaved, NULL);
2979 int kvm_arch_vcpu_ioctl_get_regs(struct kvm_vcpu *vcpu, struct kvm_regs *regs)
2983 kvm_x86_ops->cache_regs(vcpu);
2985 regs->rax = vcpu->arch.regs[VCPU_REGS_RAX];
2986 regs->rbx = vcpu->arch.regs[VCPU_REGS_RBX];
2987 regs->rcx = vcpu->arch.regs[VCPU_REGS_RCX];
2988 regs->rdx = vcpu->arch.regs[VCPU_REGS_RDX];
2989 regs->rsi = vcpu->arch.regs[VCPU_REGS_RSI];
2990 regs->rdi = vcpu->arch.regs[VCPU_REGS_RDI];
2991 regs->rsp = vcpu->arch.regs[VCPU_REGS_RSP];
2992 regs->rbp = vcpu->arch.regs[VCPU_REGS_RBP];
2993 #ifdef CONFIG_X86_64
2994 regs->r8 = vcpu->arch.regs[VCPU_REGS_R8];
2995 regs->r9 = vcpu->arch.regs[VCPU_REGS_R9];
2996 regs->r10 = vcpu->arch.regs[VCPU_REGS_R10];
2997 regs->r11 = vcpu->arch.regs[VCPU_REGS_R11];
2998 regs->r12 = vcpu->arch.regs[VCPU_REGS_R12];
2999 regs->r13 = vcpu->arch.regs[VCPU_REGS_R13];
3000 regs->r14 = vcpu->arch.regs[VCPU_REGS_R14];
3001 regs->r15 = vcpu->arch.regs[VCPU_REGS_R15];
3004 regs->rip = vcpu->arch.rip;
3005 regs->rflags = kvm_x86_ops->get_rflags(vcpu);
3008 * Don't leak debug flags in case they were set for guest debugging
3010 if (vcpu->guest_debug.enabled && vcpu->guest_debug.singlestep)
3011 regs->rflags &= ~(X86_EFLAGS_TF | X86_EFLAGS_RF);
3018 int kvm_arch_vcpu_ioctl_set_regs(struct kvm_vcpu *vcpu, struct kvm_regs *regs)
3022 vcpu->arch.regs[VCPU_REGS_RAX] = regs->rax;
3023 vcpu->arch.regs[VCPU_REGS_RBX] = regs->rbx;
3024 vcpu->arch.regs[VCPU_REGS_RCX] = regs->rcx;
3025 vcpu->arch.regs[VCPU_REGS_RDX] = regs->rdx;
3026 vcpu->arch.regs[VCPU_REGS_RSI] = regs->rsi;
3027 vcpu->arch.regs[VCPU_REGS_RDI] = regs->rdi;
3028 vcpu->arch.regs[VCPU_REGS_RSP] = regs->rsp;
3029 vcpu->arch.regs[VCPU_REGS_RBP] = regs->rbp;
3030 #ifdef CONFIG_X86_64
3031 vcpu->arch.regs[VCPU_REGS_R8] = regs->r8;
3032 vcpu->arch.regs[VCPU_REGS_R9] = regs->r9;
3033 vcpu->arch.regs[VCPU_REGS_R10] = regs->r10;
3034 vcpu->arch.regs[VCPU_REGS_R11] = regs->r11;
3035 vcpu->arch.regs[VCPU_REGS_R12] = regs->r12;
3036 vcpu->arch.regs[VCPU_REGS_R13] = regs->r13;
3037 vcpu->arch.regs[VCPU_REGS_R14] = regs->r14;
3038 vcpu->arch.regs[VCPU_REGS_R15] = regs->r15;
3041 vcpu->arch.rip = regs->rip;
3042 kvm_x86_ops->set_rflags(vcpu, regs->rflags);
3044 kvm_x86_ops->decache_regs(vcpu);
3046 vcpu->arch.exception.pending = false;
3053 static void get_segment(struct kvm_vcpu *vcpu,
3054 struct kvm_segment *var, int seg)
3056 kvm_x86_ops->get_segment(vcpu, var, seg);
3059 void kvm_get_cs_db_l_bits(struct kvm_vcpu *vcpu, int *db, int *l)
3061 struct kvm_segment cs;
3063 get_segment(vcpu, &cs, VCPU_SREG_CS);
3067 EXPORT_SYMBOL_GPL(kvm_get_cs_db_l_bits);
3069 int kvm_arch_vcpu_ioctl_get_sregs(struct kvm_vcpu *vcpu,
3070 struct kvm_sregs *sregs)
3072 struct descriptor_table dt;
3077 get_segment(vcpu, &sregs->cs, VCPU_SREG_CS);
3078 get_segment(vcpu, &sregs->ds, VCPU_SREG_DS);
3079 get_segment(vcpu, &sregs->es, VCPU_SREG_ES);
3080 get_segment(vcpu, &sregs->fs, VCPU_SREG_FS);
3081 get_segment(vcpu, &sregs->gs, VCPU_SREG_GS);
3082 get_segment(vcpu, &sregs->ss, VCPU_SREG_SS);
3084 get_segment(vcpu, &sregs->tr, VCPU_SREG_TR);
3085 get_segment(vcpu, &sregs->ldt, VCPU_SREG_LDTR);
3087 kvm_x86_ops->get_idt(vcpu, &dt);
3088 sregs->idt.limit = dt.limit;
3089 sregs->idt.base = dt.base;
3090 kvm_x86_ops->get_gdt(vcpu, &dt);
3091 sregs->gdt.limit = dt.limit;
3092 sregs->gdt.base = dt.base;
3094 kvm_x86_ops->decache_cr4_guest_bits(vcpu);
3095 sregs->cr0 = vcpu->arch.cr0;
3096 sregs->cr2 = vcpu->arch.cr2;
3097 sregs->cr3 = vcpu->arch.cr3;
3098 sregs->cr4 = vcpu->arch.cr4;
3099 sregs->cr8 = kvm_get_cr8(vcpu);
3100 sregs->efer = vcpu->arch.shadow_efer;
3101 sregs->apic_base = kvm_get_apic_base(vcpu);
3103 if (irqchip_in_kernel(vcpu->kvm)) {
3104 memset(sregs->interrupt_bitmap, 0,
3105 sizeof sregs->interrupt_bitmap);
3106 pending_vec = kvm_x86_ops->get_irq(vcpu);
3107 if (pending_vec >= 0)
3108 set_bit(pending_vec,
3109 (unsigned long *)sregs->interrupt_bitmap);
3111 memcpy(sregs->interrupt_bitmap, vcpu->arch.irq_pending,
3112 sizeof sregs->interrupt_bitmap);
3119 int kvm_arch_vcpu_ioctl_get_mpstate(struct kvm_vcpu *vcpu,
3120 struct kvm_mp_state *mp_state)
3123 mp_state->mp_state = vcpu->arch.mp_state;
3128 int kvm_arch_vcpu_ioctl_set_mpstate(struct kvm_vcpu *vcpu,
3129 struct kvm_mp_state *mp_state)
3132 vcpu->arch.mp_state = mp_state->mp_state;
3137 static void set_segment(struct kvm_vcpu *vcpu,
3138 struct kvm_segment *var, int seg)
3140 kvm_x86_ops->set_segment(vcpu, var, seg);
3143 static void seg_desct_to_kvm_desct(struct desc_struct *seg_desc, u16 selector,
3144 struct kvm_segment *kvm_desct)
3146 kvm_desct->base = seg_desc->base0;
3147 kvm_desct->base |= seg_desc->base1 << 16;
3148 kvm_desct->base |= seg_desc->base2 << 24;
3149 kvm_desct->limit = seg_desc->limit0;
3150 kvm_desct->limit |= seg_desc->limit << 16;
3151 kvm_desct->selector = selector;
3152 kvm_desct->type = seg_desc->type;
3153 kvm_desct->present = seg_desc->p;
3154 kvm_desct->dpl = seg_desc->dpl;
3155 kvm_desct->db = seg_desc->d;
3156 kvm_desct->s = seg_desc->s;
3157 kvm_desct->l = seg_desc->l;
3158 kvm_desct->g = seg_desc->g;
3159 kvm_desct->avl = seg_desc->avl;
3161 kvm_desct->unusable = 1;
3163 kvm_desct->unusable = 0;
3164 kvm_desct->padding = 0;
3167 static void get_segment_descritptor_dtable(struct kvm_vcpu *vcpu,
3169 struct descriptor_table *dtable)
3171 if (selector & 1 << 2) {
3172 struct kvm_segment kvm_seg;
3174 get_segment(vcpu, &kvm_seg, VCPU_SREG_LDTR);
3176 if (kvm_seg.unusable)
3179 dtable->limit = kvm_seg.limit;
3180 dtable->base = kvm_seg.base;
3183 kvm_x86_ops->get_gdt(vcpu, dtable);
3186 /* allowed just for 8 bytes segments */
3187 static int load_guest_segment_descriptor(struct kvm_vcpu *vcpu, u16 selector,
3188 struct desc_struct *seg_desc)
3190 struct descriptor_table dtable;
3191 u16 index = selector >> 3;
3193 get_segment_descritptor_dtable(vcpu, selector, &dtable);
3195 if (dtable.limit < index * 8 + 7) {
3196 kvm_queue_exception_e(vcpu, GP_VECTOR, selector & 0xfffc);
3199 return kvm_read_guest(vcpu->kvm, dtable.base + index * 8, seg_desc, 8);
3202 /* allowed just for 8 bytes segments */
3203 static int save_guest_segment_descriptor(struct kvm_vcpu *vcpu, u16 selector,
3204 struct desc_struct *seg_desc)
3206 struct descriptor_table dtable;
3207 u16 index = selector >> 3;
3209 get_segment_descritptor_dtable(vcpu, selector, &dtable);
3211 if (dtable.limit < index * 8 + 7)
3213 return kvm_write_guest(vcpu->kvm, dtable.base + index * 8, seg_desc, 8);
3216 static u32 get_tss_base_addr(struct kvm_vcpu *vcpu,
3217 struct desc_struct *seg_desc)
3221 base_addr = seg_desc->base0;
3222 base_addr |= (seg_desc->base1 << 16);
3223 base_addr |= (seg_desc->base2 << 24);
3228 static int load_tss_segment32(struct kvm_vcpu *vcpu,
3229 struct desc_struct *seg_desc,
3230 struct tss_segment_32 *tss)
3234 base_addr = get_tss_base_addr(vcpu, seg_desc);
3236 return kvm_read_guest(vcpu->kvm, base_addr, tss,
3237 sizeof(struct tss_segment_32));
3240 static int save_tss_segment32(struct kvm_vcpu *vcpu,
3241 struct desc_struct *seg_desc,
3242 struct tss_segment_32 *tss)
3246 base_addr = get_tss_base_addr(vcpu, seg_desc);
3248 return kvm_write_guest(vcpu->kvm, base_addr, tss,
3249 sizeof(struct tss_segment_32));
3252 static int load_tss_segment16(struct kvm_vcpu *vcpu,
3253 struct desc_struct *seg_desc,
3254 struct tss_segment_16 *tss)
3258 base_addr = get_tss_base_addr(vcpu, seg_desc);
3260 return kvm_read_guest(vcpu->kvm, base_addr, tss,
3261 sizeof(struct tss_segment_16));
3264 static int save_tss_segment16(struct kvm_vcpu *vcpu,
3265 struct desc_struct *seg_desc,
3266 struct tss_segment_16 *tss)
3270 base_addr = get_tss_base_addr(vcpu, seg_desc);
3272 return kvm_write_guest(vcpu->kvm, base_addr, tss,
3273 sizeof(struct tss_segment_16));
3276 static u16 get_segment_selector(struct kvm_vcpu *vcpu, int seg)
3278 struct kvm_segment kvm_seg;
3280 get_segment(vcpu, &kvm_seg, seg);
3281 return kvm_seg.selector;
3284 static int load_segment_descriptor_to_kvm_desct(struct kvm_vcpu *vcpu,
3286 struct kvm_segment *kvm_seg)
3288 struct desc_struct seg_desc;
3290 if (load_guest_segment_descriptor(vcpu, selector, &seg_desc))
3292 seg_desct_to_kvm_desct(&seg_desc, selector, kvm_seg);
3296 static int load_segment_descriptor(struct kvm_vcpu *vcpu, u16 selector,
3297 int type_bits, int seg)
3299 struct kvm_segment kvm_seg;
3301 if (load_segment_descriptor_to_kvm_desct(vcpu, selector, &kvm_seg))
3303 kvm_seg.type |= type_bits;
3305 if (seg != VCPU_SREG_SS && seg != VCPU_SREG_CS &&
3306 seg != VCPU_SREG_LDTR)
3308 kvm_seg.unusable = 1;
3310 set_segment(vcpu, &kvm_seg, seg);
3314 static void save_state_to_tss32(struct kvm_vcpu *vcpu,
3315 struct tss_segment_32 *tss)
3317 tss->cr3 = vcpu->arch.cr3;
3318 tss->eip = vcpu->arch.rip;
3319 tss->eflags = kvm_x86_ops->get_rflags(vcpu);
3320 tss->eax = vcpu->arch.regs[VCPU_REGS_RAX];
3321 tss->ecx = vcpu->arch.regs[VCPU_REGS_RCX];
3322 tss->edx = vcpu->arch.regs[VCPU_REGS_RDX];
3323 tss->ebx = vcpu->arch.regs[VCPU_REGS_RBX];
3324 tss->esp = vcpu->arch.regs[VCPU_REGS_RSP];
3325 tss->ebp = vcpu->arch.regs[VCPU_REGS_RBP];
3326 tss->esi = vcpu->arch.regs[VCPU_REGS_RSI];
3327 tss->edi = vcpu->arch.regs[VCPU_REGS_RDI];
3329 tss->es = get_segment_selector(vcpu, VCPU_SREG_ES);
3330 tss->cs = get_segment_selector(vcpu, VCPU_SREG_CS);
3331 tss->ss = get_segment_selector(vcpu, VCPU_SREG_SS);
3332 tss->ds = get_segment_selector(vcpu, VCPU_SREG_DS);
3333 tss->fs = get_segment_selector(vcpu, VCPU_SREG_FS);
3334 tss->gs = get_segment_selector(vcpu, VCPU_SREG_GS);
3335 tss->ldt_selector = get_segment_selector(vcpu, VCPU_SREG_LDTR);
3336 tss->prev_task_link = get_segment_selector(vcpu, VCPU_SREG_TR);
3339 static int load_state_from_tss32(struct kvm_vcpu *vcpu,
3340 struct tss_segment_32 *tss)
3342 kvm_set_cr3(vcpu, tss->cr3);
3344 vcpu->arch.rip = tss->eip;
3345 kvm_x86_ops->set_rflags(vcpu, tss->eflags | 2);
3347 vcpu->arch.regs[VCPU_REGS_RAX] = tss->eax;
3348 vcpu->arch.regs[VCPU_REGS_RCX] = tss->ecx;
3349 vcpu->arch.regs[VCPU_REGS_RDX] = tss->edx;
3350 vcpu->arch.regs[VCPU_REGS_RBX] = tss->ebx;
3351 vcpu->arch.regs[VCPU_REGS_RSP] = tss->esp;
3352 vcpu->arch.regs[VCPU_REGS_RBP] = tss->ebp;
3353 vcpu->arch.regs[VCPU_REGS_RSI] = tss->esi;
3354 vcpu->arch.regs[VCPU_REGS_RDI] = tss->edi;
3356 if (load_segment_descriptor(vcpu, tss->ldt_selector, 0, VCPU_SREG_LDTR))
3359 if (load_segment_descriptor(vcpu, tss->es, 1, VCPU_SREG_ES))
3362 if (load_segment_descriptor(vcpu, tss->cs, 9, VCPU_SREG_CS))
3365 if (load_segment_descriptor(vcpu, tss->ss, 1, VCPU_SREG_SS))
3368 if (load_segment_descriptor(vcpu, tss->ds, 1, VCPU_SREG_DS))
3371 if (load_segment_descriptor(vcpu, tss->fs, 1, VCPU_SREG_FS))
3374 if (load_segment_descriptor(vcpu, tss->gs, 1, VCPU_SREG_GS))
3379 static void save_state_to_tss16(struct kvm_vcpu *vcpu,
3380 struct tss_segment_16 *tss)
3382 tss->ip = vcpu->arch.rip;
3383 tss->flag = kvm_x86_ops->get_rflags(vcpu);
3384 tss->ax = vcpu->arch.regs[VCPU_REGS_RAX];
3385 tss->cx = vcpu->arch.regs[VCPU_REGS_RCX];
3386 tss->dx = vcpu->arch.regs[VCPU_REGS_RDX];
3387 tss->bx = vcpu->arch.regs[VCPU_REGS_RBX];
3388 tss->sp = vcpu->arch.regs[VCPU_REGS_RSP];
3389 tss->bp = vcpu->arch.regs[VCPU_REGS_RBP];
3390 tss->si = vcpu->arch.regs[VCPU_REGS_RSI];
3391 tss->di = vcpu->arch.regs[VCPU_REGS_RDI];
3393 tss->es = get_segment_selector(vcpu, VCPU_SREG_ES);
3394 tss->cs = get_segment_selector(vcpu, VCPU_SREG_CS);
3395 tss->ss = get_segment_selector(vcpu, VCPU_SREG_SS);
3396 tss->ds = get_segment_selector(vcpu, VCPU_SREG_DS);
3397 tss->ldt = get_segment_selector(vcpu, VCPU_SREG_LDTR);
3398 tss->prev_task_link = get_segment_selector(vcpu, VCPU_SREG_TR);
3401 static int load_state_from_tss16(struct kvm_vcpu *vcpu,
3402 struct tss_segment_16 *tss)
3404 vcpu->arch.rip = tss->ip;
3405 kvm_x86_ops->set_rflags(vcpu, tss->flag | 2);
3406 vcpu->arch.regs[VCPU_REGS_RAX] = tss->ax;
3407 vcpu->arch.regs[VCPU_REGS_RCX] = tss->cx;
3408 vcpu->arch.regs[VCPU_REGS_RDX] = tss->dx;
3409 vcpu->arch.regs[VCPU_REGS_RBX] = tss->bx;
3410 vcpu->arch.regs[VCPU_REGS_RSP] = tss->sp;
3411 vcpu->arch.regs[VCPU_REGS_RBP] = tss->bp;
3412 vcpu->arch.regs[VCPU_REGS_RSI] = tss->si;
3413 vcpu->arch.regs[VCPU_REGS_RDI] = tss->di;
3415 if (load_segment_descriptor(vcpu, tss->ldt, 0, VCPU_SREG_LDTR))
3418 if (load_segment_descriptor(vcpu, tss->es, 1, VCPU_SREG_ES))
3421 if (load_segment_descriptor(vcpu, tss->cs, 9, VCPU_SREG_CS))
3424 if (load_segment_descriptor(vcpu, tss->ss, 1, VCPU_SREG_SS))
3427 if (load_segment_descriptor(vcpu, tss->ds, 1, VCPU_SREG_DS))
3432 static int kvm_task_switch_16(struct kvm_vcpu *vcpu, u16 tss_selector,
3433 struct desc_struct *cseg_desc,
3434 struct desc_struct *nseg_desc)
3436 struct tss_segment_16 tss_segment_16;
3439 if (load_tss_segment16(vcpu, cseg_desc, &tss_segment_16))
3442 save_state_to_tss16(vcpu, &tss_segment_16);
3443 save_tss_segment16(vcpu, cseg_desc, &tss_segment_16);
3445 if (load_tss_segment16(vcpu, nseg_desc, &tss_segment_16))
3447 if (load_state_from_tss16(vcpu, &tss_segment_16))
3455 static int kvm_task_switch_32(struct kvm_vcpu *vcpu, u16 tss_selector,
3456 struct desc_struct *cseg_desc,
3457 struct desc_struct *nseg_desc)
3459 struct tss_segment_32 tss_segment_32;
3462 if (load_tss_segment32(vcpu, cseg_desc, &tss_segment_32))
3465 save_state_to_tss32(vcpu, &tss_segment_32);
3466 save_tss_segment32(vcpu, cseg_desc, &tss_segment_32);
3468 if (load_tss_segment32(vcpu, nseg_desc, &tss_segment_32))
3470 if (load_state_from_tss32(vcpu, &tss_segment_32))
3478 int kvm_task_switch(struct kvm_vcpu *vcpu, u16 tss_selector, int reason)
3480 struct kvm_segment tr_seg;
3481 struct desc_struct cseg_desc;
3482 struct desc_struct nseg_desc;
3485 get_segment(vcpu, &tr_seg, VCPU_SREG_TR);
3487 if (load_guest_segment_descriptor(vcpu, tss_selector, &nseg_desc))
3490 if (load_guest_segment_descriptor(vcpu, tr_seg.selector, &cseg_desc))
3494 if (reason != TASK_SWITCH_IRET) {
3497 cpl = kvm_x86_ops->get_cpl(vcpu);
3498 if ((tss_selector & 3) > nseg_desc.dpl || cpl > nseg_desc.dpl) {
3499 kvm_queue_exception_e(vcpu, GP_VECTOR, 0);
3504 if (!nseg_desc.p || (nseg_desc.limit0 | nseg_desc.limit << 16) < 0x67) {
3505 kvm_queue_exception_e(vcpu, TS_VECTOR, tss_selector & 0xfffc);
3509 if (reason == TASK_SWITCH_IRET || reason == TASK_SWITCH_JMP) {
3510 cseg_desc.type &= ~(1 << 1); //clear the B flag
3511 save_guest_segment_descriptor(vcpu, tr_seg.selector,
3515 if (reason == TASK_SWITCH_IRET) {
3516 u32 eflags = kvm_x86_ops->get_rflags(vcpu);
3517 kvm_x86_ops->set_rflags(vcpu, eflags & ~X86_EFLAGS_NT);
3520 kvm_x86_ops->skip_emulated_instruction(vcpu);
3521 kvm_x86_ops->cache_regs(vcpu);
3523 if (nseg_desc.type & 8)
3524 ret = kvm_task_switch_32(vcpu, tss_selector, &cseg_desc,
3527 ret = kvm_task_switch_16(vcpu, tss_selector, &cseg_desc,
3530 if (reason == TASK_SWITCH_CALL || reason == TASK_SWITCH_GATE) {
3531 u32 eflags = kvm_x86_ops->get_rflags(vcpu);
3532 kvm_x86_ops->set_rflags(vcpu, eflags | X86_EFLAGS_NT);
3535 if (reason != TASK_SWITCH_IRET) {
3536 nseg_desc.type |= (1 << 1);
3537 save_guest_segment_descriptor(vcpu, tss_selector,
3541 kvm_x86_ops->set_cr0(vcpu, vcpu->arch.cr0 | X86_CR0_TS);
3542 seg_desct_to_kvm_desct(&nseg_desc, tss_selector, &tr_seg);
3544 set_segment(vcpu, &tr_seg, VCPU_SREG_TR);
3546 kvm_x86_ops->decache_regs(vcpu);
3549 EXPORT_SYMBOL_GPL(kvm_task_switch);
3551 int kvm_arch_vcpu_ioctl_set_sregs(struct kvm_vcpu *vcpu,
3552 struct kvm_sregs *sregs)
3554 int mmu_reset_needed = 0;
3555 int i, pending_vec, max_bits;
3556 struct descriptor_table dt;
3560 dt.limit = sregs->idt.limit;
3561 dt.base = sregs->idt.base;
3562 kvm_x86_ops->set_idt(vcpu, &dt);
3563 dt.limit = sregs->gdt.limit;
3564 dt.base = sregs->gdt.base;
3565 kvm_x86_ops->set_gdt(vcpu, &dt);
3567 vcpu->arch.cr2 = sregs->cr2;
3568 mmu_reset_needed |= vcpu->arch.cr3 != sregs->cr3;
3569 vcpu->arch.cr3 = sregs->cr3;
3571 kvm_set_cr8(vcpu, sregs->cr8);
3573 mmu_reset_needed |= vcpu->arch.shadow_efer != sregs->efer;
3574 kvm_x86_ops->set_efer(vcpu, sregs->efer);
3575 kvm_set_apic_base(vcpu, sregs->apic_base);
3577 kvm_x86_ops->decache_cr4_guest_bits(vcpu);
3579 mmu_reset_needed |= vcpu->arch.cr0 != sregs->cr0;
3580 kvm_x86_ops->set_cr0(vcpu, sregs->cr0);
3581 vcpu->arch.cr0 = sregs->cr0;
3583 mmu_reset_needed |= vcpu->arch.cr4 != sregs->cr4;
3584 kvm_x86_ops->set_cr4(vcpu, sregs->cr4);
3585 if (!is_long_mode(vcpu) && is_pae(vcpu))
3586 load_pdptrs(vcpu, vcpu->arch.cr3);
3588 if (mmu_reset_needed)
3589 kvm_mmu_reset_context(vcpu);
3591 if (!irqchip_in_kernel(vcpu->kvm)) {
3592 memcpy(vcpu->arch.irq_pending, sregs->interrupt_bitmap,
3593 sizeof vcpu->arch.irq_pending);
3594 vcpu->arch.irq_summary = 0;
3595 for (i = 0; i < ARRAY_SIZE(vcpu->arch.irq_pending); ++i)
3596 if (vcpu->arch.irq_pending[i])
3597 __set_bit(i, &vcpu->arch.irq_summary);
3599 max_bits = (sizeof sregs->interrupt_bitmap) << 3;
3600 pending_vec = find_first_bit(
3601 (const unsigned long *)sregs->interrupt_bitmap,
3603 /* Only pending external irq is handled here */
3604 if (pending_vec < max_bits) {
3605 kvm_x86_ops->set_irq(vcpu, pending_vec);
3606 pr_debug("Set back pending irq %d\n",
3611 set_segment(vcpu, &sregs->cs, VCPU_SREG_CS);
3612 set_segment(vcpu, &sregs->ds, VCPU_SREG_DS);
3613 set_segment(vcpu, &sregs->es, VCPU_SREG_ES);
3614 set_segment(vcpu, &sregs->fs, VCPU_SREG_FS);
3615 set_segment(vcpu, &sregs->gs, VCPU_SREG_GS);
3616 set_segment(vcpu, &sregs->ss, VCPU_SREG_SS);
3618 set_segment(vcpu, &sregs->tr, VCPU_SREG_TR);
3619 set_segment(vcpu, &sregs->ldt, VCPU_SREG_LDTR);
3626 int kvm_arch_vcpu_ioctl_debug_guest(struct kvm_vcpu *vcpu,
3627 struct kvm_debug_guest *dbg)
3633 r = kvm_x86_ops->set_guest_debug(vcpu, dbg);
3641 * fxsave fpu state. Taken from x86_64/processor.h. To be killed when
3642 * we have asm/x86/processor.h
3653 u32 st_space[32]; /* 8*16 bytes for each FP-reg = 128 bytes */
3654 #ifdef CONFIG_X86_64
3655 u32 xmm_space[64]; /* 16*16 bytes for each XMM-reg = 256 bytes */
3657 u32 xmm_space[32]; /* 8*16 bytes for each XMM-reg = 128 bytes */
3662 * Translate a guest virtual address to a guest physical address.
3664 int kvm_arch_vcpu_ioctl_translate(struct kvm_vcpu *vcpu,
3665 struct kvm_translation *tr)
3667 unsigned long vaddr = tr->linear_address;
3671 down_read(&vcpu->kvm->slots_lock);
3672 gpa = vcpu->arch.mmu.gva_to_gpa(vcpu, vaddr);
3673 up_read(&vcpu->kvm->slots_lock);
3674 tr->physical_address = gpa;
3675 tr->valid = gpa != UNMAPPED_GVA;
3683 int kvm_arch_vcpu_ioctl_get_fpu(struct kvm_vcpu *vcpu, struct kvm_fpu *fpu)
3685 struct fxsave *fxsave = (struct fxsave *)&vcpu->arch.guest_fx_image;
3689 memcpy(fpu->fpr, fxsave->st_space, 128);
3690 fpu->fcw = fxsave->cwd;
3691 fpu->fsw = fxsave->swd;
3692 fpu->ftwx = fxsave->twd;
3693 fpu->last_opcode = fxsave->fop;
3694 fpu->last_ip = fxsave->rip;
3695 fpu->last_dp = fxsave->rdp;
3696 memcpy(fpu->xmm, fxsave->xmm_space, sizeof fxsave->xmm_space);
3703 int kvm_arch_vcpu_ioctl_set_fpu(struct kvm_vcpu *vcpu, struct kvm_fpu *fpu)
3705 struct fxsave *fxsave = (struct fxsave *)&vcpu->arch.guest_fx_image;
3709 memcpy(fxsave->st_space, fpu->fpr, 128);
3710 fxsave->cwd = fpu->fcw;
3711 fxsave->swd = fpu->fsw;
3712 fxsave->twd = fpu->ftwx;
3713 fxsave->fop = fpu->last_opcode;
3714 fxsave->rip = fpu->last_ip;
3715 fxsave->rdp = fpu->last_dp;
3716 memcpy(fxsave->xmm_space, fpu->xmm, sizeof fxsave->xmm_space);
3723 void fx_init(struct kvm_vcpu *vcpu)
3725 unsigned after_mxcsr_mask;
3728 * Touch the fpu the first time in non atomic context as if
3729 * this is the first fpu instruction the exception handler
3730 * will fire before the instruction returns and it'll have to
3731 * allocate ram with GFP_KERNEL.
3734 fx_save(&vcpu->arch.host_fx_image);
3736 /* Initialize guest FPU by resetting ours and saving into guest's */
3738 fx_save(&vcpu->arch.host_fx_image);
3740 fx_save(&vcpu->arch.guest_fx_image);
3741 fx_restore(&vcpu->arch.host_fx_image);
3744 vcpu->arch.cr0 |= X86_CR0_ET;
3745 after_mxcsr_mask = offsetof(struct i387_fxsave_struct, st_space);
3746 vcpu->arch.guest_fx_image.mxcsr = 0x1f80;
3747 memset((void *)&vcpu->arch.guest_fx_image + after_mxcsr_mask,
3748 0, sizeof(struct i387_fxsave_struct) - after_mxcsr_mask);
3750 EXPORT_SYMBOL_GPL(fx_init);
3752 void kvm_load_guest_fpu(struct kvm_vcpu *vcpu)
3754 if (!vcpu->fpu_active || vcpu->guest_fpu_loaded)
3757 vcpu->guest_fpu_loaded = 1;
3758 fx_save(&vcpu->arch.host_fx_image);
3759 fx_restore(&vcpu->arch.guest_fx_image);
3761 EXPORT_SYMBOL_GPL(kvm_load_guest_fpu);
3763 void kvm_put_guest_fpu(struct kvm_vcpu *vcpu)
3765 if (!vcpu->guest_fpu_loaded)
3768 vcpu->guest_fpu_loaded = 0;
3769 fx_save(&vcpu->arch.guest_fx_image);
3770 fx_restore(&vcpu->arch.host_fx_image);
3771 ++vcpu->stat.fpu_reload;
3773 EXPORT_SYMBOL_GPL(kvm_put_guest_fpu);
3775 void kvm_arch_vcpu_free(struct kvm_vcpu *vcpu)
3777 kvm_x86_ops->vcpu_free(vcpu);
3780 struct kvm_vcpu *kvm_arch_vcpu_create(struct kvm *kvm,
3783 return kvm_x86_ops->vcpu_create(kvm, id);
3786 int kvm_arch_vcpu_setup(struct kvm_vcpu *vcpu)
3790 /* We do fxsave: this must be aligned. */
3791 BUG_ON((unsigned long)&vcpu->arch.host_fx_image & 0xF);
3794 r = kvm_arch_vcpu_reset(vcpu);
3796 r = kvm_mmu_setup(vcpu);
3803 kvm_x86_ops->vcpu_free(vcpu);
3807 void kvm_arch_vcpu_destroy(struct kvm_vcpu *vcpu)
3810 kvm_mmu_unload(vcpu);
3813 kvm_x86_ops->vcpu_free(vcpu);
3816 int kvm_arch_vcpu_reset(struct kvm_vcpu *vcpu)
3818 return kvm_x86_ops->vcpu_reset(vcpu);
3821 void kvm_arch_hardware_enable(void *garbage)
3823 kvm_x86_ops->hardware_enable(garbage);
3826 void kvm_arch_hardware_disable(void *garbage)
3828 kvm_x86_ops->hardware_disable(garbage);
3831 int kvm_arch_hardware_setup(void)
3833 return kvm_x86_ops->hardware_setup();
3836 void kvm_arch_hardware_unsetup(void)
3838 kvm_x86_ops->hardware_unsetup();
3841 void kvm_arch_check_processor_compat(void *rtn)
3843 kvm_x86_ops->check_processor_compatibility(rtn);
3846 int kvm_arch_vcpu_init(struct kvm_vcpu *vcpu)
3852 BUG_ON(vcpu->kvm == NULL);
3855 vcpu->arch.mmu.root_hpa = INVALID_PAGE;
3856 if (!irqchip_in_kernel(kvm) || vcpu->vcpu_id == 0)
3857 vcpu->arch.mp_state = KVM_MP_STATE_RUNNABLE;
3859 vcpu->arch.mp_state = KVM_MP_STATE_UNINITIALIZED;
3861 page = alloc_page(GFP_KERNEL | __GFP_ZERO);
3866 vcpu->arch.pio_data = page_address(page);
3868 r = kvm_mmu_create(vcpu);
3870 goto fail_free_pio_data;
3872 if (irqchip_in_kernel(kvm)) {
3873 r = kvm_create_lapic(vcpu);
3875 goto fail_mmu_destroy;
3881 kvm_mmu_destroy(vcpu);
3883 free_page((unsigned long)vcpu->arch.pio_data);
3888 void kvm_arch_vcpu_uninit(struct kvm_vcpu *vcpu)
3890 kvm_free_lapic(vcpu);
3891 down_read(&vcpu->kvm->slots_lock);
3892 kvm_mmu_destroy(vcpu);
3893 up_read(&vcpu->kvm->slots_lock);
3894 free_page((unsigned long)vcpu->arch.pio_data);
3897 struct kvm *kvm_arch_create_vm(void)
3899 struct kvm *kvm = kzalloc(sizeof(struct kvm), GFP_KERNEL);
3902 return ERR_PTR(-ENOMEM);
3904 INIT_LIST_HEAD(&kvm->arch.active_mmu_pages);
3909 static void kvm_unload_vcpu_mmu(struct kvm_vcpu *vcpu)
3912 kvm_mmu_unload(vcpu);
3916 static void kvm_free_vcpus(struct kvm *kvm)
3921 * Unpin any mmu pages first.
3923 for (i = 0; i < KVM_MAX_VCPUS; ++i)
3925 kvm_unload_vcpu_mmu(kvm->vcpus[i]);
3926 for (i = 0; i < KVM_MAX_VCPUS; ++i) {
3927 if (kvm->vcpus[i]) {
3928 kvm_arch_vcpu_free(kvm->vcpus[i]);
3929 kvm->vcpus[i] = NULL;
3935 void kvm_arch_destroy_vm(struct kvm *kvm)
3938 kfree(kvm->arch.vpic);
3939 kfree(kvm->arch.vioapic);
3940 kvm_free_vcpus(kvm);
3941 kvm_free_physmem(kvm);
3942 if (kvm->arch.apic_access_page)
3943 put_page(kvm->arch.apic_access_page);
3944 if (kvm->arch.ept_identity_pagetable)
3945 put_page(kvm->arch.ept_identity_pagetable);
3949 int kvm_arch_set_memory_region(struct kvm *kvm,
3950 struct kvm_userspace_memory_region *mem,
3951 struct kvm_memory_slot old,
3954 int npages = mem->memory_size >> PAGE_SHIFT;
3955 struct kvm_memory_slot *memslot = &kvm->memslots[mem->slot];
3957 /*To keep backward compatibility with older userspace,
3958 *x86 needs to hanlde !user_alloc case.
3961 if (npages && !old.rmap) {
3962 down_write(¤t->mm->mmap_sem);
3963 memslot->userspace_addr = do_mmap(NULL, 0,
3965 PROT_READ | PROT_WRITE,
3966 MAP_SHARED | MAP_ANONYMOUS,
3968 up_write(¤t->mm->mmap_sem);
3970 if (IS_ERR((void *)memslot->userspace_addr))
3971 return PTR_ERR((void *)memslot->userspace_addr);
3973 if (!old.user_alloc && old.rmap) {
3976 down_write(¤t->mm->mmap_sem);
3977 ret = do_munmap(current->mm, old.userspace_addr,
3978 old.npages * PAGE_SIZE);
3979 up_write(¤t->mm->mmap_sem);
3982 "kvm_vm_ioctl_set_memory_region: "
3983 "failed to munmap memory\n");
3988 if (!kvm->arch.n_requested_mmu_pages) {
3989 unsigned int nr_mmu_pages = kvm_mmu_calculate_mmu_pages(kvm);
3990 kvm_mmu_change_mmu_pages(kvm, nr_mmu_pages);
3993 kvm_mmu_slot_remove_write_access(kvm, mem->slot);
3994 kvm_flush_remote_tlbs(kvm);
3999 int kvm_arch_vcpu_runnable(struct kvm_vcpu *vcpu)
4001 return vcpu->arch.mp_state == KVM_MP_STATE_RUNNABLE
4002 || vcpu->arch.mp_state == KVM_MP_STATE_SIPI_RECEIVED;
4005 static void vcpu_kick_intr(void *info)
4008 struct kvm_vcpu *vcpu = (struct kvm_vcpu *)info;
4009 printk(KERN_DEBUG "vcpu_kick_intr %p \n", vcpu);
4013 void kvm_vcpu_kick(struct kvm_vcpu *vcpu)
4015 int ipi_pcpu = vcpu->cpu;
4016 int cpu = get_cpu();
4018 if (waitqueue_active(&vcpu->wq)) {
4019 wake_up_interruptible(&vcpu->wq);
4020 ++vcpu->stat.halt_wakeup;
4023 * We may be called synchronously with irqs disabled in guest mode,
4024 * So need not to call smp_call_function_single() in that case.
4026 if (vcpu->guest_mode && vcpu->cpu != cpu)
4027 smp_call_function_single(ipi_pcpu, vcpu_kick_intr, vcpu, 0);