Linux Audio

Check our new training course

Loading...
v3.15
   1/*
   2 * Kernel-based Virtual Machine driver for Linux
   3 *
   4 * AMD SVM support
   5 *
   6 * Copyright (C) 2006 Qumranet, Inc.
   7 * Copyright 2010 Red Hat, Inc. and/or its affiliates.
   8 *
   9 * Authors:
  10 *   Yaniv Kamay  <yaniv@qumranet.com>
  11 *   Avi Kivity   <avi@qumranet.com>
  12 *
  13 * This work is licensed under the terms of the GNU GPL, version 2.  See
  14 * the COPYING file in the top-level directory.
  15 *
  16 */
  17#include <linux/kvm_host.h>
  18
  19#include "irq.h"
  20#include "mmu.h"
  21#include "kvm_cache_regs.h"
  22#include "x86.h"
  23#include "cpuid.h"
  24
  25#include <linux/module.h>
  26#include <linux/mod_devicetable.h>
  27#include <linux/kernel.h>
  28#include <linux/vmalloc.h>
  29#include <linux/highmem.h>
  30#include <linux/sched.h>
  31#include <linux/ftrace_event.h>
  32#include <linux/slab.h>
  33
  34#include <asm/perf_event.h>
  35#include <asm/tlbflush.h>
  36#include <asm/desc.h>
  37#include <asm/debugreg.h>
  38#include <asm/kvm_para.h>
  39
  40#include <asm/virtext.h>
  41#include "trace.h"
  42
  43#define __ex(x) __kvm_handle_fault_on_reboot(x)
  44
  45MODULE_AUTHOR("Qumranet");
  46MODULE_LICENSE("GPL");
  47
  48static const struct x86_cpu_id svm_cpu_id[] = {
  49	X86_FEATURE_MATCH(X86_FEATURE_SVM),
  50	{}
  51};
  52MODULE_DEVICE_TABLE(x86cpu, svm_cpu_id);
  53
  54#define IOPM_ALLOC_ORDER 2
  55#define MSRPM_ALLOC_ORDER 1
  56
  57#define SEG_TYPE_LDT 2
  58#define SEG_TYPE_BUSY_TSS16 3
  59
  60#define SVM_FEATURE_NPT            (1 <<  0)
  61#define SVM_FEATURE_LBRV           (1 <<  1)
  62#define SVM_FEATURE_SVML           (1 <<  2)
  63#define SVM_FEATURE_NRIP           (1 <<  3)
  64#define SVM_FEATURE_TSC_RATE       (1 <<  4)
  65#define SVM_FEATURE_VMCB_CLEAN     (1 <<  5)
  66#define SVM_FEATURE_FLUSH_ASID     (1 <<  6)
  67#define SVM_FEATURE_DECODE_ASSIST  (1 <<  7)
  68#define SVM_FEATURE_PAUSE_FILTER   (1 << 10)
  69
  70#define NESTED_EXIT_HOST	0	/* Exit handled on host level */
  71#define NESTED_EXIT_DONE	1	/* Exit caused nested vmexit  */
  72#define NESTED_EXIT_CONTINUE	2	/* Further checks needed      */
  73
  74#define DEBUGCTL_RESERVED_BITS (~(0x3fULL))
  75
  76#define TSC_RATIO_RSVD          0xffffff0000000000ULL
  77#define TSC_RATIO_MIN		0x0000000000000001ULL
  78#define TSC_RATIO_MAX		0x000000ffffffffffULL
  79
  80static bool erratum_383_found __read_mostly;
  81
  82static const u32 host_save_user_msrs[] = {
  83#ifdef CONFIG_X86_64
  84	MSR_STAR, MSR_LSTAR, MSR_CSTAR, MSR_SYSCALL_MASK, MSR_KERNEL_GS_BASE,
  85	MSR_FS_BASE,
  86#endif
  87	MSR_IA32_SYSENTER_CS, MSR_IA32_SYSENTER_ESP, MSR_IA32_SYSENTER_EIP,
  88};
  89
  90#define NR_HOST_SAVE_USER_MSRS ARRAY_SIZE(host_save_user_msrs)
  91
  92struct kvm_vcpu;
  93
  94struct nested_state {
  95	struct vmcb *hsave;
  96	u64 hsave_msr;
  97	u64 vm_cr_msr;
  98	u64 vmcb;
  99
 100	/* These are the merged vectors */
 101	u32 *msrpm;
 102
 103	/* gpa pointers to the real vectors */
 104	u64 vmcb_msrpm;
 105	u64 vmcb_iopm;
 106
 107	/* A VMEXIT is required but not yet emulated */
 108	bool exit_required;
 109
 110	/* cache for intercepts of the guest */
 111	u32 intercept_cr;
 112	u32 intercept_dr;
 113	u32 intercept_exceptions;
 114	u64 intercept;
 115
 116	/* Nested Paging related state */
 117	u64 nested_cr3;
 118};
 119
 120#define MSRPM_OFFSETS	16
 121static u32 msrpm_offsets[MSRPM_OFFSETS] __read_mostly;
 122
 123/*
 124 * Set osvw_len to higher value when updated Revision Guides
 125 * are published and we know what the new status bits are
 126 */
 127static uint64_t osvw_len = 4, osvw_status;
 128
 129struct vcpu_svm {
 130	struct kvm_vcpu vcpu;
 131	struct vmcb *vmcb;
 132	unsigned long vmcb_pa;
 133	struct svm_cpu_data *svm_data;
 134	uint64_t asid_generation;
 135	uint64_t sysenter_esp;
 136	uint64_t sysenter_eip;
 137
 138	u64 next_rip;
 139
 140	u64 host_user_msrs[NR_HOST_SAVE_USER_MSRS];
 141	struct {
 142		u16 fs;
 143		u16 gs;
 144		u16 ldt;
 145		u64 gs_base;
 146	} host;
 147
 148	u32 *msrpm;
 149
 150	ulong nmi_iret_rip;
 151
 152	struct nested_state nested;
 153
 154	bool nmi_singlestep;
 155
 156	unsigned int3_injected;
 157	unsigned long int3_rip;
 158	u32 apf_reason;
 159
 160	u64  tsc_ratio;
 161};
 162
 163static DEFINE_PER_CPU(u64, current_tsc_ratio);
 164#define TSC_RATIO_DEFAULT	0x0100000000ULL
 165
 166#define MSR_INVALID			0xffffffffU
 167
 168static const struct svm_direct_access_msrs {
 169	u32 index;   /* Index of the MSR */
 170	bool always; /* True if intercept is always on */
 171} direct_access_msrs[] = {
 172	{ .index = MSR_STAR,				.always = true  },
 173	{ .index = MSR_IA32_SYSENTER_CS,		.always = true  },
 174#ifdef CONFIG_X86_64
 175	{ .index = MSR_GS_BASE,				.always = true  },
 176	{ .index = MSR_FS_BASE,				.always = true  },
 177	{ .index = MSR_KERNEL_GS_BASE,			.always = true  },
 178	{ .index = MSR_LSTAR,				.always = true  },
 179	{ .index = MSR_CSTAR,				.always = true  },
 180	{ .index = MSR_SYSCALL_MASK,			.always = true  },
 181#endif
 182	{ .index = MSR_IA32_LASTBRANCHFROMIP,		.always = false },
 183	{ .index = MSR_IA32_LASTBRANCHTOIP,		.always = false },
 184	{ .index = MSR_IA32_LASTINTFROMIP,		.always = false },
 185	{ .index = MSR_IA32_LASTINTTOIP,		.always = false },
 186	{ .index = MSR_INVALID,				.always = false },
 187};
 188
 189/* enable NPT for AMD64 and X86 with PAE */
 190#if defined(CONFIG_X86_64) || defined(CONFIG_X86_PAE)
 191static bool npt_enabled = true;
 192#else
 193static bool npt_enabled;
 194#endif
 
 195
 196/* allow nested paging (virtualized MMU) for all guests */
 197static int npt = true;
 198module_param(npt, int, S_IRUGO);
 199
 200/* allow nested virtualization in KVM/SVM */
 201static int nested = true;
 202module_param(nested, int, S_IRUGO);
 203
 204static void svm_flush_tlb(struct kvm_vcpu *vcpu);
 205static void svm_complete_interrupts(struct vcpu_svm *svm);
 206
 207static int nested_svm_exit_handled(struct vcpu_svm *svm);
 208static int nested_svm_intercept(struct vcpu_svm *svm);
 209static int nested_svm_vmexit(struct vcpu_svm *svm);
 210static int nested_svm_check_exception(struct vcpu_svm *svm, unsigned nr,
 211				      bool has_error_code, u32 error_code);
 212static u64 __scale_tsc(u64 ratio, u64 tsc);
 213
 214enum {
 215	VMCB_INTERCEPTS, /* Intercept vectors, TSC offset,
 216			    pause filter count */
 217	VMCB_PERM_MAP,   /* IOPM Base and MSRPM Base */
 218	VMCB_ASID,	 /* ASID */
 219	VMCB_INTR,	 /* int_ctl, int_vector */
 220	VMCB_NPT,        /* npt_en, nCR3, gPAT */
 221	VMCB_CR,	 /* CR0, CR3, CR4, EFER */
 222	VMCB_DR,         /* DR6, DR7 */
 223	VMCB_DT,         /* GDT, IDT */
 224	VMCB_SEG,        /* CS, DS, SS, ES, CPL */
 225	VMCB_CR2,        /* CR2 only */
 226	VMCB_LBR,        /* DBGCTL, BR_FROM, BR_TO, LAST_EX_FROM, LAST_EX_TO */
 227	VMCB_DIRTY_MAX,
 228};
 229
 230/* TPR and CR2 are always written before VMRUN */
 231#define VMCB_ALWAYS_DIRTY_MASK	((1U << VMCB_INTR) | (1U << VMCB_CR2))
 232
 233static inline void mark_all_dirty(struct vmcb *vmcb)
 234{
 235	vmcb->control.clean = 0;
 236}
 237
 238static inline void mark_all_clean(struct vmcb *vmcb)
 239{
 240	vmcb->control.clean = ((1 << VMCB_DIRTY_MAX) - 1)
 241			       & ~VMCB_ALWAYS_DIRTY_MASK;
 242}
 243
 244static inline void mark_dirty(struct vmcb *vmcb, int bit)
 245{
 246	vmcb->control.clean &= ~(1 << bit);
 247}
 248
 249static inline struct vcpu_svm *to_svm(struct kvm_vcpu *vcpu)
 250{
 251	return container_of(vcpu, struct vcpu_svm, vcpu);
 252}
 253
 254static void recalc_intercepts(struct vcpu_svm *svm)
 255{
 256	struct vmcb_control_area *c, *h;
 257	struct nested_state *g;
 258
 259	mark_dirty(svm->vmcb, VMCB_INTERCEPTS);
 260
 261	if (!is_guest_mode(&svm->vcpu))
 262		return;
 263
 264	c = &svm->vmcb->control;
 265	h = &svm->nested.hsave->control;
 266	g = &svm->nested;
 267
 268	c->intercept_cr = h->intercept_cr | g->intercept_cr;
 269	c->intercept_dr = h->intercept_dr | g->intercept_dr;
 270	c->intercept_exceptions = h->intercept_exceptions | g->intercept_exceptions;
 271	c->intercept = h->intercept | g->intercept;
 272}
 273
 274static inline struct vmcb *get_host_vmcb(struct vcpu_svm *svm)
 275{
 276	if (is_guest_mode(&svm->vcpu))
 277		return svm->nested.hsave;
 278	else
 279		return svm->vmcb;
 280}
 281
 282static inline void set_cr_intercept(struct vcpu_svm *svm, int bit)
 283{
 284	struct vmcb *vmcb = get_host_vmcb(svm);
 285
 286	vmcb->control.intercept_cr |= (1U << bit);
 287
 288	recalc_intercepts(svm);
 289}
 290
 291static inline void clr_cr_intercept(struct vcpu_svm *svm, int bit)
 292{
 293	struct vmcb *vmcb = get_host_vmcb(svm);
 294
 295	vmcb->control.intercept_cr &= ~(1U << bit);
 296
 297	recalc_intercepts(svm);
 298}
 299
 300static inline bool is_cr_intercept(struct vcpu_svm *svm, int bit)
 301{
 302	struct vmcb *vmcb = get_host_vmcb(svm);
 303
 304	return vmcb->control.intercept_cr & (1U << bit);
 305}
 306
 307static inline void set_dr_intercepts(struct vcpu_svm *svm)
 308{
 309	struct vmcb *vmcb = get_host_vmcb(svm);
 310
 311	vmcb->control.intercept_dr = (1 << INTERCEPT_DR0_READ)
 312		| (1 << INTERCEPT_DR1_READ)
 313		| (1 << INTERCEPT_DR2_READ)
 314		| (1 << INTERCEPT_DR3_READ)
 315		| (1 << INTERCEPT_DR4_READ)
 316		| (1 << INTERCEPT_DR5_READ)
 317		| (1 << INTERCEPT_DR6_READ)
 318		| (1 << INTERCEPT_DR7_READ)
 319		| (1 << INTERCEPT_DR0_WRITE)
 320		| (1 << INTERCEPT_DR1_WRITE)
 321		| (1 << INTERCEPT_DR2_WRITE)
 322		| (1 << INTERCEPT_DR3_WRITE)
 323		| (1 << INTERCEPT_DR4_WRITE)
 324		| (1 << INTERCEPT_DR5_WRITE)
 325		| (1 << INTERCEPT_DR6_WRITE)
 326		| (1 << INTERCEPT_DR7_WRITE);
 327
 328	recalc_intercepts(svm);
 329}
 330
 331static inline void clr_dr_intercepts(struct vcpu_svm *svm)
 332{
 333	struct vmcb *vmcb = get_host_vmcb(svm);
 334
 335	vmcb->control.intercept_dr = 0;
 336
 337	recalc_intercepts(svm);
 338}
 339
 340static inline void set_exception_intercept(struct vcpu_svm *svm, int bit)
 341{
 342	struct vmcb *vmcb = get_host_vmcb(svm);
 343
 344	vmcb->control.intercept_exceptions |= (1U << bit);
 345
 346	recalc_intercepts(svm);
 347}
 348
 349static inline void clr_exception_intercept(struct vcpu_svm *svm, int bit)
 350{
 351	struct vmcb *vmcb = get_host_vmcb(svm);
 352
 353	vmcb->control.intercept_exceptions &= ~(1U << bit);
 354
 355	recalc_intercepts(svm);
 356}
 357
 358static inline void set_intercept(struct vcpu_svm *svm, int bit)
 359{
 360	struct vmcb *vmcb = get_host_vmcb(svm);
 361
 362	vmcb->control.intercept |= (1ULL << bit);
 363
 364	recalc_intercepts(svm);
 365}
 366
 367static inline void clr_intercept(struct vcpu_svm *svm, int bit)
 368{
 369	struct vmcb *vmcb = get_host_vmcb(svm);
 370
 371	vmcb->control.intercept &= ~(1ULL << bit);
 372
 373	recalc_intercepts(svm);
 374}
 375
 376static inline void enable_gif(struct vcpu_svm *svm)
 377{
 378	svm->vcpu.arch.hflags |= HF_GIF_MASK;
 379}
 380
 381static inline void disable_gif(struct vcpu_svm *svm)
 382{
 383	svm->vcpu.arch.hflags &= ~HF_GIF_MASK;
 384}
 385
 386static inline bool gif_set(struct vcpu_svm *svm)
 387{
 388	return !!(svm->vcpu.arch.hflags & HF_GIF_MASK);
 389}
 390
 391static unsigned long iopm_base;
 392
 393struct kvm_ldttss_desc {
 394	u16 limit0;
 395	u16 base0;
 396	unsigned base1:8, type:5, dpl:2, p:1;
 397	unsigned limit1:4, zero0:3, g:1, base2:8;
 398	u32 base3;
 399	u32 zero1;
 400} __attribute__((packed));
 401
 402struct svm_cpu_data {
 403	int cpu;
 404
 405	u64 asid_generation;
 406	u32 max_asid;
 407	u32 next_asid;
 408	struct kvm_ldttss_desc *tss_desc;
 409
 410	struct page *save_area;
 411};
 412
 413static DEFINE_PER_CPU(struct svm_cpu_data *, svm_data);
 414
 415struct svm_init_data {
 416	int cpu;
 417	int r;
 418};
 419
 420static const u32 msrpm_ranges[] = {0, 0xc0000000, 0xc0010000};
 421
 422#define NUM_MSR_MAPS ARRAY_SIZE(msrpm_ranges)
 423#define MSRS_RANGE_SIZE 2048
 424#define MSRS_IN_RANGE (MSRS_RANGE_SIZE * 8 / 2)
 425
 426static u32 svm_msrpm_offset(u32 msr)
 427{
 428	u32 offset;
 429	int i;
 430
 431	for (i = 0; i < NUM_MSR_MAPS; i++) {
 432		if (msr < msrpm_ranges[i] ||
 433		    msr >= msrpm_ranges[i] + MSRS_IN_RANGE)
 434			continue;
 435
 436		offset  = (msr - msrpm_ranges[i]) / 4; /* 4 msrs per u8 */
 437		offset += (i * MSRS_RANGE_SIZE);       /* add range offset */
 438
 439		/* Now we have the u8 offset - but need the u32 offset */
 440		return offset / 4;
 441	}
 442
 443	/* MSR not in any range */
 444	return MSR_INVALID;
 445}
 446
 447#define MAX_INST_SIZE 15
 448
 449static inline void clgi(void)
 450{
 451	asm volatile (__ex(SVM_CLGI));
 452}
 453
 454static inline void stgi(void)
 455{
 456	asm volatile (__ex(SVM_STGI));
 457}
 458
 459static inline void invlpga(unsigned long addr, u32 asid)
 460{
 461	asm volatile (__ex(SVM_INVLPGA) : : "a"(addr), "c"(asid));
 462}
 463
 464static int get_npt_level(void)
 465{
 466#ifdef CONFIG_X86_64
 467	return PT64_ROOT_LEVEL;
 468#else
 469	return PT32E_ROOT_LEVEL;
 470#endif
 471}
 472
 473static void svm_set_efer(struct kvm_vcpu *vcpu, u64 efer)
 474{
 475	vcpu->arch.efer = efer;
 476	if (!npt_enabled && !(efer & EFER_LMA))
 477		efer &= ~EFER_LME;
 478
 479	to_svm(vcpu)->vmcb->save.efer = efer | EFER_SVME;
 480	mark_dirty(to_svm(vcpu)->vmcb, VMCB_CR);
 481}
 482
 483static int is_external_interrupt(u32 info)
 484{
 485	info &= SVM_EVTINJ_TYPE_MASK | SVM_EVTINJ_VALID;
 486	return info == (SVM_EVTINJ_VALID | SVM_EVTINJ_TYPE_INTR);
 487}
 488
 489static u32 svm_get_interrupt_shadow(struct kvm_vcpu *vcpu, int mask)
 490{
 491	struct vcpu_svm *svm = to_svm(vcpu);
 492	u32 ret = 0;
 493
 494	if (svm->vmcb->control.int_state & SVM_INTERRUPT_SHADOW_MASK)
 495		ret |= KVM_X86_SHADOW_INT_STI | KVM_X86_SHADOW_INT_MOV_SS;
 496	return ret & mask;
 497}
 498
 499static void svm_set_interrupt_shadow(struct kvm_vcpu *vcpu, int mask)
 500{
 501	struct vcpu_svm *svm = to_svm(vcpu);
 502
 503	if (mask == 0)
 504		svm->vmcb->control.int_state &= ~SVM_INTERRUPT_SHADOW_MASK;
 505	else
 506		svm->vmcb->control.int_state |= SVM_INTERRUPT_SHADOW_MASK;
 507
 508}
 509
 510static void skip_emulated_instruction(struct kvm_vcpu *vcpu)
 511{
 512	struct vcpu_svm *svm = to_svm(vcpu);
 513
 514	if (svm->vmcb->control.next_rip != 0)
 515		svm->next_rip = svm->vmcb->control.next_rip;
 516
 517	if (!svm->next_rip) {
 518		if (emulate_instruction(vcpu, EMULTYPE_SKIP) !=
 519				EMULATE_DONE)
 520			printk(KERN_DEBUG "%s: NOP\n", __func__);
 521		return;
 522	}
 523	if (svm->next_rip - kvm_rip_read(vcpu) > MAX_INST_SIZE)
 524		printk(KERN_ERR "%s: ip 0x%lx next 0x%llx\n",
 525		       __func__, kvm_rip_read(vcpu), svm->next_rip);
 526
 527	kvm_rip_write(vcpu, svm->next_rip);
 528	svm_set_interrupt_shadow(vcpu, 0);
 529}
 530
 531static void svm_queue_exception(struct kvm_vcpu *vcpu, unsigned nr,
 532				bool has_error_code, u32 error_code,
 533				bool reinject)
 534{
 535	struct vcpu_svm *svm = to_svm(vcpu);
 536
 537	/*
 538	 * If we are within a nested VM we'd better #VMEXIT and let the guest
 539	 * handle the exception
 540	 */
 541	if (!reinject &&
 542	    nested_svm_check_exception(svm, nr, has_error_code, error_code))
 543		return;
 544
 545	if (nr == BP_VECTOR && !static_cpu_has(X86_FEATURE_NRIPS)) {
 546		unsigned long rip, old_rip = kvm_rip_read(&svm->vcpu);
 547
 548		/*
 549		 * For guest debugging where we have to reinject #BP if some
 550		 * INT3 is guest-owned:
 551		 * Emulate nRIP by moving RIP forward. Will fail if injection
 552		 * raises a fault that is not intercepted. Still better than
 553		 * failing in all cases.
 554		 */
 555		skip_emulated_instruction(&svm->vcpu);
 556		rip = kvm_rip_read(&svm->vcpu);
 557		svm->int3_rip = rip + svm->vmcb->save.cs.base;
 558		svm->int3_injected = rip - old_rip;
 559	}
 560
 561	svm->vmcb->control.event_inj = nr
 562		| SVM_EVTINJ_VALID
 563		| (has_error_code ? SVM_EVTINJ_VALID_ERR : 0)
 564		| SVM_EVTINJ_TYPE_EXEPT;
 565	svm->vmcb->control.event_inj_err = error_code;
 566}
 567
 568static void svm_init_erratum_383(void)
 569{
 570	u32 low, high;
 571	int err;
 572	u64 val;
 573
 574	if (!static_cpu_has_bug(X86_BUG_AMD_TLB_MMATCH))
 575		return;
 576
 577	/* Use _safe variants to not break nested virtualization */
 578	val = native_read_msr_safe(MSR_AMD64_DC_CFG, &err);
 579	if (err)
 580		return;
 581
 582	val |= (1ULL << 47);
 583
 584	low  = lower_32_bits(val);
 585	high = upper_32_bits(val);
 586
 587	native_write_msr_safe(MSR_AMD64_DC_CFG, low, high);
 588
 589	erratum_383_found = true;
 590}
 591
 592static void svm_init_osvw(struct kvm_vcpu *vcpu)
 593{
 594	/*
 595	 * Guests should see errata 400 and 415 as fixed (assuming that
 596	 * HLT and IO instructions are intercepted).
 597	 */
 598	vcpu->arch.osvw.length = (osvw_len >= 3) ? (osvw_len) : 3;
 599	vcpu->arch.osvw.status = osvw_status & ~(6ULL);
 600
 601	/*
 602	 * By increasing VCPU's osvw.length to 3 we are telling the guest that
 603	 * all osvw.status bits inside that length, including bit 0 (which is
 604	 * reserved for erratum 298), are valid. However, if host processor's
 605	 * osvw_len is 0 then osvw_status[0] carries no information. We need to
 606	 * be conservative here and therefore we tell the guest that erratum 298
 607	 * is present (because we really don't know).
 608	 */
 609	if (osvw_len == 0 && boot_cpu_data.x86 == 0x10)
 610		vcpu->arch.osvw.status |= 1;
 611}
 612
 613static int has_svm(void)
 614{
 615	const char *msg;
 616
 617	if (!cpu_has_svm(&msg)) {
 618		printk(KERN_INFO "has_svm: %s\n", msg);
 619		return 0;
 620	}
 621
 622	return 1;
 623}
 624
 625static void svm_hardware_disable(void *garbage)
 626{
 627	/* Make sure we clean up behind us */
 628	if (static_cpu_has(X86_FEATURE_TSCRATEMSR))
 629		wrmsrl(MSR_AMD64_TSC_RATIO, TSC_RATIO_DEFAULT);
 630
 631	cpu_svm_disable();
 632
 633	amd_pmu_disable_virt();
 634}
 635
 636static int svm_hardware_enable(void *garbage)
 637{
 638
 639	struct svm_cpu_data *sd;
 640	uint64_t efer;
 641	struct desc_ptr gdt_descr;
 642	struct desc_struct *gdt;
 643	int me = raw_smp_processor_id();
 644
 645	rdmsrl(MSR_EFER, efer);
 646	if (efer & EFER_SVME)
 647		return -EBUSY;
 648
 649	if (!has_svm()) {
 650		pr_err("%s: err EOPNOTSUPP on %d\n", __func__, me);
 
 651		return -EINVAL;
 652	}
 653	sd = per_cpu(svm_data, me);
 
 654	if (!sd) {
 655		pr_err("%s: svm_data is NULL on %d\n", __func__, me);
 
 656		return -EINVAL;
 657	}
 658
 659	sd->asid_generation = 1;
 660	sd->max_asid = cpuid_ebx(SVM_CPUID_FUNC) - 1;
 661	sd->next_asid = sd->max_asid + 1;
 662
 663	native_store_gdt(&gdt_descr);
 664	gdt = (struct desc_struct *)gdt_descr.address;
 665	sd->tss_desc = (struct kvm_ldttss_desc *)(gdt + GDT_ENTRY_TSS);
 666
 667	wrmsrl(MSR_EFER, efer | EFER_SVME);
 668
 669	wrmsrl(MSR_VM_HSAVE_PA, page_to_pfn(sd->save_area) << PAGE_SHIFT);
 670
 671	if (static_cpu_has(X86_FEATURE_TSCRATEMSR)) {
 672		wrmsrl(MSR_AMD64_TSC_RATIO, TSC_RATIO_DEFAULT);
 673		__get_cpu_var(current_tsc_ratio) = TSC_RATIO_DEFAULT;
 674	}
 675
 676
 677	/*
 678	 * Get OSVW bits.
 679	 *
 680	 * Note that it is possible to have a system with mixed processor
 681	 * revisions and therefore different OSVW bits. If bits are not the same
 682	 * on different processors then choose the worst case (i.e. if erratum
 683	 * is present on one processor and not on another then assume that the
 684	 * erratum is present everywhere).
 685	 */
 686	if (cpu_has(&boot_cpu_data, X86_FEATURE_OSVW)) {
 687		uint64_t len, status = 0;
 688		int err;
 689
 690		len = native_read_msr_safe(MSR_AMD64_OSVW_ID_LENGTH, &err);
 691		if (!err)
 692			status = native_read_msr_safe(MSR_AMD64_OSVW_STATUS,
 693						      &err);
 694
 695		if (err)
 696			osvw_status = osvw_len = 0;
 697		else {
 698			if (len < osvw_len)
 699				osvw_len = len;
 700			osvw_status |= status;
 701			osvw_status &= (1ULL << osvw_len) - 1;
 702		}
 703	} else
 704		osvw_status = osvw_len = 0;
 705
 706	svm_init_erratum_383();
 707
 708	amd_pmu_enable_virt();
 709
 710	return 0;
 711}
 712
 713static void svm_cpu_uninit(int cpu)
 714{
 715	struct svm_cpu_data *sd = per_cpu(svm_data, raw_smp_processor_id());
 716
 717	if (!sd)
 718		return;
 719
 720	per_cpu(svm_data, raw_smp_processor_id()) = NULL;
 721	__free_page(sd->save_area);
 722	kfree(sd);
 723}
 724
 725static int svm_cpu_init(int cpu)
 726{
 727	struct svm_cpu_data *sd;
 728	int r;
 729
 730	sd = kzalloc(sizeof(struct svm_cpu_data), GFP_KERNEL);
 731	if (!sd)
 732		return -ENOMEM;
 733	sd->cpu = cpu;
 734	sd->save_area = alloc_page(GFP_KERNEL);
 735	r = -ENOMEM;
 736	if (!sd->save_area)
 737		goto err_1;
 738
 739	per_cpu(svm_data, cpu) = sd;
 740
 741	return 0;
 742
 743err_1:
 744	kfree(sd);
 745	return r;
 746
 747}
 748
 749static bool valid_msr_intercept(u32 index)
 750{
 751	int i;
 752
 753	for (i = 0; direct_access_msrs[i].index != MSR_INVALID; i++)
 754		if (direct_access_msrs[i].index == index)
 755			return true;
 756
 757	return false;
 758}
 759
 760static void set_msr_interception(u32 *msrpm, unsigned msr,
 761				 int read, int write)
 762{
 763	u8 bit_read, bit_write;
 764	unsigned long tmp;
 765	u32 offset;
 766
 767	/*
 768	 * If this warning triggers extend the direct_access_msrs list at the
 769	 * beginning of the file
 770	 */
 771	WARN_ON(!valid_msr_intercept(msr));
 772
 773	offset    = svm_msrpm_offset(msr);
 774	bit_read  = 2 * (msr & 0x0f);
 775	bit_write = 2 * (msr & 0x0f) + 1;
 776	tmp       = msrpm[offset];
 777
 778	BUG_ON(offset == MSR_INVALID);
 779
 780	read  ? clear_bit(bit_read,  &tmp) : set_bit(bit_read,  &tmp);
 781	write ? clear_bit(bit_write, &tmp) : set_bit(bit_write, &tmp);
 782
 783	msrpm[offset] = tmp;
 784}
 785
 786static void svm_vcpu_init_msrpm(u32 *msrpm)
 787{
 788	int i;
 789
 790	memset(msrpm, 0xff, PAGE_SIZE * (1 << MSRPM_ALLOC_ORDER));
 791
 792	for (i = 0; direct_access_msrs[i].index != MSR_INVALID; i++) {
 793		if (!direct_access_msrs[i].always)
 794			continue;
 795
 796		set_msr_interception(msrpm, direct_access_msrs[i].index, 1, 1);
 797	}
 798}
 799
 800static void add_msr_offset(u32 offset)
 801{
 802	int i;
 803
 804	for (i = 0; i < MSRPM_OFFSETS; ++i) {
 805
 806		/* Offset already in list? */
 807		if (msrpm_offsets[i] == offset)
 808			return;
 809
 810		/* Slot used by another offset? */
 811		if (msrpm_offsets[i] != MSR_INVALID)
 812			continue;
 813
 814		/* Add offset to list */
 815		msrpm_offsets[i] = offset;
 816
 817		return;
 818	}
 819
 820	/*
 821	 * If this BUG triggers the msrpm_offsets table has an overflow. Just
 822	 * increase MSRPM_OFFSETS in this case.
 823	 */
 824	BUG();
 825}
 826
 827static void init_msrpm_offsets(void)
 828{
 829	int i;
 830
 831	memset(msrpm_offsets, 0xff, sizeof(msrpm_offsets));
 832
 833	for (i = 0; direct_access_msrs[i].index != MSR_INVALID; i++) {
 834		u32 offset;
 835
 836		offset = svm_msrpm_offset(direct_access_msrs[i].index);
 837		BUG_ON(offset == MSR_INVALID);
 838
 839		add_msr_offset(offset);
 840	}
 841}
 842
 843static void svm_enable_lbrv(struct vcpu_svm *svm)
 844{
 845	u32 *msrpm = svm->msrpm;
 846
 847	svm->vmcb->control.lbr_ctl = 1;
 848	set_msr_interception(msrpm, MSR_IA32_LASTBRANCHFROMIP, 1, 1);
 849	set_msr_interception(msrpm, MSR_IA32_LASTBRANCHTOIP, 1, 1);
 850	set_msr_interception(msrpm, MSR_IA32_LASTINTFROMIP, 1, 1);
 851	set_msr_interception(msrpm, MSR_IA32_LASTINTTOIP, 1, 1);
 852}
 853
 854static void svm_disable_lbrv(struct vcpu_svm *svm)
 855{
 856	u32 *msrpm = svm->msrpm;
 857
 858	svm->vmcb->control.lbr_ctl = 0;
 859	set_msr_interception(msrpm, MSR_IA32_LASTBRANCHFROMIP, 0, 0);
 860	set_msr_interception(msrpm, MSR_IA32_LASTBRANCHTOIP, 0, 0);
 861	set_msr_interception(msrpm, MSR_IA32_LASTINTFROMIP, 0, 0);
 862	set_msr_interception(msrpm, MSR_IA32_LASTINTTOIP, 0, 0);
 863}
 864
 865static __init int svm_hardware_setup(void)
 866{
 867	int cpu;
 868	struct page *iopm_pages;
 869	void *iopm_va;
 870	int r;
 871
 872	iopm_pages = alloc_pages(GFP_KERNEL, IOPM_ALLOC_ORDER);
 873
 874	if (!iopm_pages)
 875		return -ENOMEM;
 876
 877	iopm_va = page_address(iopm_pages);
 878	memset(iopm_va, 0xff, PAGE_SIZE * (1 << IOPM_ALLOC_ORDER));
 879	iopm_base = page_to_pfn(iopm_pages) << PAGE_SHIFT;
 880
 881	init_msrpm_offsets();
 882
 883	if (boot_cpu_has(X86_FEATURE_NX))
 884		kvm_enable_efer_bits(EFER_NX);
 885
 886	if (boot_cpu_has(X86_FEATURE_FXSR_OPT))
 887		kvm_enable_efer_bits(EFER_FFXSR);
 888
 889	if (boot_cpu_has(X86_FEATURE_TSCRATEMSR)) {
 890		u64 max;
 891
 892		kvm_has_tsc_control = true;
 893
 894		/*
 895		 * Make sure the user can only configure tsc_khz values that
 896		 * fit into a signed integer.
 897		 * A min value is not calculated needed because it will always
 898		 * be 1 on all machines and a value of 0 is used to disable
 899		 * tsc-scaling for the vcpu.
 900		 */
 901		max = min(0x7fffffffULL, __scale_tsc(tsc_khz, TSC_RATIO_MAX));
 902
 903		kvm_max_guest_tsc_khz = max;
 904	}
 905
 906	if (nested) {
 907		printk(KERN_INFO "kvm: Nested Virtualization enabled\n");
 908		kvm_enable_efer_bits(EFER_SVME | EFER_LMSLE);
 909	}
 910
 911	for_each_possible_cpu(cpu) {
 912		r = svm_cpu_init(cpu);
 913		if (r)
 914			goto err;
 915	}
 916
 917	if (!boot_cpu_has(X86_FEATURE_NPT))
 918		npt_enabled = false;
 919
 920	if (npt_enabled && !npt) {
 921		printk(KERN_INFO "kvm: Nested Paging disabled\n");
 922		npt_enabled = false;
 923	}
 924
 925	if (npt_enabled) {
 926		printk(KERN_INFO "kvm: Nested Paging enabled\n");
 927		kvm_enable_tdp();
 928	} else
 929		kvm_disable_tdp();
 930
 931	return 0;
 932
 933err:
 934	__free_pages(iopm_pages, IOPM_ALLOC_ORDER);
 935	iopm_base = 0;
 936	return r;
 937}
 938
 939static __exit void svm_hardware_unsetup(void)
 940{
 941	int cpu;
 942
 943	for_each_possible_cpu(cpu)
 944		svm_cpu_uninit(cpu);
 945
 946	__free_pages(pfn_to_page(iopm_base >> PAGE_SHIFT), IOPM_ALLOC_ORDER);
 947	iopm_base = 0;
 948}
 949
 950static void init_seg(struct vmcb_seg *seg)
 951{
 952	seg->selector = 0;
 953	seg->attrib = SVM_SELECTOR_P_MASK | SVM_SELECTOR_S_MASK |
 954		      SVM_SELECTOR_WRITE_MASK; /* Read/Write Data Segment */
 955	seg->limit = 0xffff;
 956	seg->base = 0;
 957}
 958
 959static void init_sys_seg(struct vmcb_seg *seg, uint32_t type)
 960{
 961	seg->selector = 0;
 962	seg->attrib = SVM_SELECTOR_P_MASK | type;
 963	seg->limit = 0xffff;
 964	seg->base = 0;
 965}
 966
 967static u64 __scale_tsc(u64 ratio, u64 tsc)
 968{
 969	u64 mult, frac, _tsc;
 970
 971	mult  = ratio >> 32;
 972	frac  = ratio & ((1ULL << 32) - 1);
 973
 974	_tsc  = tsc;
 975	_tsc *= mult;
 976	_tsc += (tsc >> 32) * frac;
 977	_tsc += ((tsc & ((1ULL << 32) - 1)) * frac) >> 32;
 978
 979	return _tsc;
 980}
 981
 982static u64 svm_scale_tsc(struct kvm_vcpu *vcpu, u64 tsc)
 983{
 984	struct vcpu_svm *svm = to_svm(vcpu);
 985	u64 _tsc = tsc;
 986
 987	if (svm->tsc_ratio != TSC_RATIO_DEFAULT)
 988		_tsc = __scale_tsc(svm->tsc_ratio, tsc);
 989
 990	return _tsc;
 991}
 992
 993static void svm_set_tsc_khz(struct kvm_vcpu *vcpu, u32 user_tsc_khz, bool scale)
 994{
 995	struct vcpu_svm *svm = to_svm(vcpu);
 996	u64 ratio;
 997	u64 khz;
 998
 999	/* Guest TSC same frequency as host TSC? */
1000	if (!scale) {
1001		svm->tsc_ratio = TSC_RATIO_DEFAULT;
1002		return;
1003	}
1004
1005	/* TSC scaling supported? */
1006	if (!boot_cpu_has(X86_FEATURE_TSCRATEMSR)) {
1007		if (user_tsc_khz > tsc_khz) {
1008			vcpu->arch.tsc_catchup = 1;
1009			vcpu->arch.tsc_always_catchup = 1;
1010		} else
1011			WARN(1, "user requested TSC rate below hardware speed\n");
1012		return;
1013	}
1014
1015	khz = user_tsc_khz;
1016
1017	/* TSC scaling required  - calculate ratio */
1018	ratio = khz << 32;
1019	do_div(ratio, tsc_khz);
1020
1021	if (ratio == 0 || ratio & TSC_RATIO_RSVD) {
1022		WARN_ONCE(1, "Invalid TSC ratio - virtual-tsc-khz=%u\n",
1023				user_tsc_khz);
1024		return;
1025	}
 
1026	svm->tsc_ratio             = ratio;
1027}
1028
1029static u64 svm_read_tsc_offset(struct kvm_vcpu *vcpu)
1030{
1031	struct vcpu_svm *svm = to_svm(vcpu);
1032
1033	return svm->vmcb->control.tsc_offset;
1034}
1035
1036static void svm_write_tsc_offset(struct kvm_vcpu *vcpu, u64 offset)
1037{
1038	struct vcpu_svm *svm = to_svm(vcpu);
1039	u64 g_tsc_offset = 0;
1040
1041	if (is_guest_mode(vcpu)) {
1042		g_tsc_offset = svm->vmcb->control.tsc_offset -
1043			       svm->nested.hsave->control.tsc_offset;
1044		svm->nested.hsave->control.tsc_offset = offset;
1045	} else
1046		trace_kvm_write_tsc_offset(vcpu->vcpu_id,
1047					   svm->vmcb->control.tsc_offset,
1048					   offset);
1049
1050	svm->vmcb->control.tsc_offset = offset + g_tsc_offset;
1051
1052	mark_dirty(svm->vmcb, VMCB_INTERCEPTS);
1053}
1054
1055static void svm_adjust_tsc_offset(struct kvm_vcpu *vcpu, s64 adjustment, bool host)
1056{
1057	struct vcpu_svm *svm = to_svm(vcpu);
1058
1059	WARN_ON(adjustment < 0);
1060	if (host)
1061		adjustment = svm_scale_tsc(vcpu, adjustment);
1062
1063	svm->vmcb->control.tsc_offset += adjustment;
1064	if (is_guest_mode(vcpu))
1065		svm->nested.hsave->control.tsc_offset += adjustment;
1066	else
1067		trace_kvm_write_tsc_offset(vcpu->vcpu_id,
1068				     svm->vmcb->control.tsc_offset - adjustment,
1069				     svm->vmcb->control.tsc_offset);
1070
1071	mark_dirty(svm->vmcb, VMCB_INTERCEPTS);
1072}
1073
1074static u64 svm_compute_tsc_offset(struct kvm_vcpu *vcpu, u64 target_tsc)
1075{
1076	u64 tsc;
1077
1078	tsc = svm_scale_tsc(vcpu, native_read_tsc());
1079
1080	return target_tsc - tsc;
1081}
1082
1083static void init_vmcb(struct vcpu_svm *svm)
1084{
1085	struct vmcb_control_area *control = &svm->vmcb->control;
1086	struct vmcb_save_area *save = &svm->vmcb->save;
1087
1088	svm->vcpu.fpu_active = 1;
1089	svm->vcpu.arch.hflags = 0;
1090
1091	set_cr_intercept(svm, INTERCEPT_CR0_READ);
1092	set_cr_intercept(svm, INTERCEPT_CR3_READ);
1093	set_cr_intercept(svm, INTERCEPT_CR4_READ);
1094	set_cr_intercept(svm, INTERCEPT_CR0_WRITE);
1095	set_cr_intercept(svm, INTERCEPT_CR3_WRITE);
1096	set_cr_intercept(svm, INTERCEPT_CR4_WRITE);
1097	set_cr_intercept(svm, INTERCEPT_CR8_WRITE);
1098
1099	set_dr_intercepts(svm);
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1100
1101	set_exception_intercept(svm, PF_VECTOR);
1102	set_exception_intercept(svm, UD_VECTOR);
1103	set_exception_intercept(svm, MC_VECTOR);
1104
1105	set_intercept(svm, INTERCEPT_INTR);
1106	set_intercept(svm, INTERCEPT_NMI);
1107	set_intercept(svm, INTERCEPT_SMI);
1108	set_intercept(svm, INTERCEPT_SELECTIVE_CR0);
1109	set_intercept(svm, INTERCEPT_RDPMC);
1110	set_intercept(svm, INTERCEPT_CPUID);
1111	set_intercept(svm, INTERCEPT_INVD);
1112	set_intercept(svm, INTERCEPT_HLT);
1113	set_intercept(svm, INTERCEPT_INVLPG);
1114	set_intercept(svm, INTERCEPT_INVLPGA);
1115	set_intercept(svm, INTERCEPT_IOIO_PROT);
1116	set_intercept(svm, INTERCEPT_MSR_PROT);
1117	set_intercept(svm, INTERCEPT_TASK_SWITCH);
1118	set_intercept(svm, INTERCEPT_SHUTDOWN);
1119	set_intercept(svm, INTERCEPT_VMRUN);
1120	set_intercept(svm, INTERCEPT_VMMCALL);
1121	set_intercept(svm, INTERCEPT_VMLOAD);
1122	set_intercept(svm, INTERCEPT_VMSAVE);
1123	set_intercept(svm, INTERCEPT_STGI);
1124	set_intercept(svm, INTERCEPT_CLGI);
1125	set_intercept(svm, INTERCEPT_SKINIT);
1126	set_intercept(svm, INTERCEPT_WBINVD);
1127	set_intercept(svm, INTERCEPT_MONITOR);
1128	set_intercept(svm, INTERCEPT_MWAIT);
1129	set_intercept(svm, INTERCEPT_XSETBV);
1130
1131	control->iopm_base_pa = iopm_base;
1132	control->msrpm_base_pa = __pa(svm->msrpm);
1133	control->int_ctl = V_INTR_MASKING_MASK;
1134
1135	init_seg(&save->es);
1136	init_seg(&save->ss);
1137	init_seg(&save->ds);
1138	init_seg(&save->fs);
1139	init_seg(&save->gs);
1140
1141	save->cs.selector = 0xf000;
1142	save->cs.base = 0xffff0000;
1143	/* Executable/Readable Code Segment */
1144	save->cs.attrib = SVM_SELECTOR_READ_MASK | SVM_SELECTOR_P_MASK |
1145		SVM_SELECTOR_S_MASK | SVM_SELECTOR_CODE_MASK;
1146	save->cs.limit = 0xffff;
 
 
 
 
 
 
 
1147
1148	save->gdtr.limit = 0xffff;
1149	save->idtr.limit = 0xffff;
1150
1151	init_sys_seg(&save->ldtr, SEG_TYPE_LDT);
1152	init_sys_seg(&save->tr, SEG_TYPE_BUSY_TSS16);
1153
1154	svm_set_efer(&svm->vcpu, 0);
1155	save->dr6 = 0xffff0ff0;
 
1156	kvm_set_rflags(&svm->vcpu, 2);
1157	save->rip = 0x0000fff0;
1158	svm->vcpu.arch.regs[VCPU_REGS_RIP] = save->rip;
1159
1160	/*
1161	 * This is the guest-visible cr0 value.
1162	 * svm_set_cr0() sets PG and WP and clears NW and CD on save->cr0.
1163	 */
1164	svm->vcpu.arch.cr0 = 0;
1165	(void)kvm_set_cr0(&svm->vcpu, X86_CR0_NW | X86_CR0_CD | X86_CR0_ET);
1166
1167	save->cr4 = X86_CR4_PAE;
1168	/* rdx = ?? */
1169
1170	if (npt_enabled) {
1171		/* Setup VMCB for Nested Paging */
1172		control->nested_ctl = 1;
 
1173		clr_intercept(svm, INTERCEPT_INVLPG);
1174		clr_exception_intercept(svm, PF_VECTOR);
1175		clr_cr_intercept(svm, INTERCEPT_CR3_READ);
1176		clr_cr_intercept(svm, INTERCEPT_CR3_WRITE);
1177		save->g_pat = 0x0007040600070406ULL;
1178		save->cr3 = 0;
1179		save->cr4 = 0;
1180	}
1181	svm->asid_generation = 0;
1182
1183	svm->nested.vmcb = 0;
1184	svm->vcpu.arch.hflags = 0;
1185
1186	if (boot_cpu_has(X86_FEATURE_PAUSEFILTER)) {
1187		control->pause_filter_count = 3000;
1188		set_intercept(svm, INTERCEPT_PAUSE);
1189	}
1190
1191	mark_all_dirty(svm->vmcb);
1192
1193	enable_gif(svm);
1194}
1195
1196static void svm_vcpu_reset(struct kvm_vcpu *vcpu)
1197{
1198	struct vcpu_svm *svm = to_svm(vcpu);
1199	u32 dummy;
1200	u32 eax = 1;
1201
1202	init_vmcb(svm);
1203
1204	kvm_cpuid(vcpu, &eax, &dummy, &dummy, &dummy);
1205	kvm_register_write(vcpu, VCPU_REGS_RDX, eax);
 
 
 
 
 
 
 
1206}
1207
1208static struct kvm_vcpu *svm_create_vcpu(struct kvm *kvm, unsigned int id)
1209{
1210	struct vcpu_svm *svm;
1211	struct page *page;
1212	struct page *msrpm_pages;
1213	struct page *hsave_page;
1214	struct page *nested_msrpm_pages;
1215	int err;
1216
1217	svm = kmem_cache_zalloc(kvm_vcpu_cache, GFP_KERNEL);
1218	if (!svm) {
1219		err = -ENOMEM;
1220		goto out;
1221	}
1222
1223	svm->tsc_ratio = TSC_RATIO_DEFAULT;
1224
1225	err = kvm_vcpu_init(&svm->vcpu, kvm, id);
1226	if (err)
1227		goto free_svm;
1228
1229	err = -ENOMEM;
1230	page = alloc_page(GFP_KERNEL);
1231	if (!page)
1232		goto uninit;
1233
1234	msrpm_pages = alloc_pages(GFP_KERNEL, MSRPM_ALLOC_ORDER);
1235	if (!msrpm_pages)
1236		goto free_page1;
1237
1238	nested_msrpm_pages = alloc_pages(GFP_KERNEL, MSRPM_ALLOC_ORDER);
1239	if (!nested_msrpm_pages)
1240		goto free_page2;
1241
1242	hsave_page = alloc_page(GFP_KERNEL);
1243	if (!hsave_page)
1244		goto free_page3;
1245
1246	svm->nested.hsave = page_address(hsave_page);
1247
1248	svm->msrpm = page_address(msrpm_pages);
1249	svm_vcpu_init_msrpm(svm->msrpm);
1250
1251	svm->nested.msrpm = page_address(nested_msrpm_pages);
1252	svm_vcpu_init_msrpm(svm->nested.msrpm);
1253
1254	svm->vmcb = page_address(page);
1255	clear_page(svm->vmcb);
1256	svm->vmcb_pa = page_to_pfn(page) << PAGE_SHIFT;
1257	svm->asid_generation = 0;
1258	init_vmcb(svm);
 
 
 
 
 
1259
1260	svm->vcpu.arch.apic_base = 0xfee00000 | MSR_IA32_APICBASE_ENABLE;
1261	if (kvm_vcpu_is_bsp(&svm->vcpu))
1262		svm->vcpu.arch.apic_base |= MSR_IA32_APICBASE_BSP;
1263
1264	svm_init_osvw(&svm->vcpu);
1265
1266	return &svm->vcpu;
1267
 
 
1268free_page3:
1269	__free_pages(nested_msrpm_pages, MSRPM_ALLOC_ORDER);
1270free_page2:
1271	__free_pages(msrpm_pages, MSRPM_ALLOC_ORDER);
1272free_page1:
1273	__free_page(page);
1274uninit:
1275	kvm_vcpu_uninit(&svm->vcpu);
1276free_svm:
1277	kmem_cache_free(kvm_vcpu_cache, svm);
1278out:
1279	return ERR_PTR(err);
1280}
1281
1282static void svm_free_vcpu(struct kvm_vcpu *vcpu)
1283{
1284	struct vcpu_svm *svm = to_svm(vcpu);
1285
1286	__free_page(pfn_to_page(svm->vmcb_pa >> PAGE_SHIFT));
1287	__free_pages(virt_to_page(svm->msrpm), MSRPM_ALLOC_ORDER);
1288	__free_page(virt_to_page(svm->nested.hsave));
1289	__free_pages(virt_to_page(svm->nested.msrpm), MSRPM_ALLOC_ORDER);
1290	kvm_vcpu_uninit(vcpu);
1291	kmem_cache_free(kvm_vcpu_cache, svm);
1292}
1293
1294static void svm_vcpu_load(struct kvm_vcpu *vcpu, int cpu)
1295{
1296	struct vcpu_svm *svm = to_svm(vcpu);
1297	int i;
1298
1299	if (unlikely(cpu != vcpu->cpu)) {
1300		svm->asid_generation = 0;
1301		mark_all_dirty(svm->vmcb);
1302	}
1303
1304#ifdef CONFIG_X86_64
1305	rdmsrl(MSR_GS_BASE, to_svm(vcpu)->host.gs_base);
1306#endif
1307	savesegment(fs, svm->host.fs);
1308	savesegment(gs, svm->host.gs);
1309	svm->host.ldt = kvm_read_ldt();
1310
1311	for (i = 0; i < NR_HOST_SAVE_USER_MSRS; i++)
1312		rdmsrl(host_save_user_msrs[i], svm->host_user_msrs[i]);
1313
1314	if (static_cpu_has(X86_FEATURE_TSCRATEMSR) &&
1315	    svm->tsc_ratio != __get_cpu_var(current_tsc_ratio)) {
1316		__get_cpu_var(current_tsc_ratio) = svm->tsc_ratio;
1317		wrmsrl(MSR_AMD64_TSC_RATIO, svm->tsc_ratio);
1318	}
1319}
1320
1321static void svm_vcpu_put(struct kvm_vcpu *vcpu)
1322{
1323	struct vcpu_svm *svm = to_svm(vcpu);
1324	int i;
1325
1326	++vcpu->stat.host_state_reload;
1327	kvm_load_ldt(svm->host.ldt);
1328#ifdef CONFIG_X86_64
1329	loadsegment(fs, svm->host.fs);
1330	wrmsrl(MSR_KERNEL_GS_BASE, current->thread.gs);
1331	load_gs_index(svm->host.gs);
1332#else
1333#ifdef CONFIG_X86_32_LAZY_GS
1334	loadsegment(gs, svm->host.gs);
1335#endif
1336#endif
1337	for (i = 0; i < NR_HOST_SAVE_USER_MSRS; i++)
1338		wrmsrl(host_save_user_msrs[i], svm->host_user_msrs[i]);
1339}
1340
1341static void svm_update_cpl(struct kvm_vcpu *vcpu)
1342{
1343	struct vcpu_svm *svm = to_svm(vcpu);
1344	int cpl;
1345
1346	if (!is_protmode(vcpu))
1347		cpl = 0;
1348	else if (svm->vmcb->save.rflags & X86_EFLAGS_VM)
1349		cpl = 3;
1350	else
1351		cpl = svm->vmcb->save.cs.selector & 0x3;
1352
1353	svm->vmcb->save.cpl = cpl;
1354}
1355
1356static unsigned long svm_get_rflags(struct kvm_vcpu *vcpu)
1357{
1358	return to_svm(vcpu)->vmcb->save.rflags;
1359}
1360
1361static void svm_set_rflags(struct kvm_vcpu *vcpu, unsigned long rflags)
1362{
1363	unsigned long old_rflags = to_svm(vcpu)->vmcb->save.rflags;
1364
1365	to_svm(vcpu)->vmcb->save.rflags = rflags;
1366	if ((old_rflags ^ rflags) & X86_EFLAGS_VM)
1367		svm_update_cpl(vcpu);
1368}
1369
1370static void svm_cache_reg(struct kvm_vcpu *vcpu, enum kvm_reg reg)
1371{
1372	switch (reg) {
1373	case VCPU_EXREG_PDPTR:
1374		BUG_ON(!npt_enabled);
1375		load_pdptrs(vcpu, vcpu->arch.walk_mmu, kvm_read_cr3(vcpu));
1376		break;
1377	default:
1378		BUG();
1379	}
1380}
1381
1382static void svm_set_vintr(struct vcpu_svm *svm)
1383{
1384	set_intercept(svm, INTERCEPT_VINTR);
1385}
1386
1387static void svm_clear_vintr(struct vcpu_svm *svm)
1388{
1389	clr_intercept(svm, INTERCEPT_VINTR);
1390}
1391
1392static struct vmcb_seg *svm_seg(struct kvm_vcpu *vcpu, int seg)
1393{
1394	struct vmcb_save_area *save = &to_svm(vcpu)->vmcb->save;
1395
1396	switch (seg) {
1397	case VCPU_SREG_CS: return &save->cs;
1398	case VCPU_SREG_DS: return &save->ds;
1399	case VCPU_SREG_ES: return &save->es;
1400	case VCPU_SREG_FS: return &save->fs;
1401	case VCPU_SREG_GS: return &save->gs;
1402	case VCPU_SREG_SS: return &save->ss;
1403	case VCPU_SREG_TR: return &save->tr;
1404	case VCPU_SREG_LDTR: return &save->ldtr;
1405	}
1406	BUG();
1407	return NULL;
1408}
1409
1410static u64 svm_get_segment_base(struct kvm_vcpu *vcpu, int seg)
1411{
1412	struct vmcb_seg *s = svm_seg(vcpu, seg);
1413
1414	return s->base;
1415}
1416
1417static void svm_get_segment(struct kvm_vcpu *vcpu,
1418			    struct kvm_segment *var, int seg)
1419{
1420	struct vmcb_seg *s = svm_seg(vcpu, seg);
1421
1422	var->base = s->base;
1423	var->limit = s->limit;
1424	var->selector = s->selector;
1425	var->type = s->attrib & SVM_SELECTOR_TYPE_MASK;
1426	var->s = (s->attrib >> SVM_SELECTOR_S_SHIFT) & 1;
1427	var->dpl = (s->attrib >> SVM_SELECTOR_DPL_SHIFT) & 3;
1428	var->present = (s->attrib >> SVM_SELECTOR_P_SHIFT) & 1;
1429	var->avl = (s->attrib >> SVM_SELECTOR_AVL_SHIFT) & 1;
1430	var->l = (s->attrib >> SVM_SELECTOR_L_SHIFT) & 1;
1431	var->db = (s->attrib >> SVM_SELECTOR_DB_SHIFT) & 1;
1432	var->g = (s->attrib >> SVM_SELECTOR_G_SHIFT) & 1;
1433
1434	/*
1435	 * AMD's VMCB does not have an explicit unusable field, so emulate it
1436	 * for cross vendor migration purposes by "not present"
1437	 */
1438	var->unusable = !var->present || (var->type == 0);
1439
1440	switch (seg) {
1441	case VCPU_SREG_CS:
1442		/*
1443		 * SVM always stores 0 for the 'G' bit in the CS selector in
1444		 * the VMCB on a VMEXIT. This hurts cross-vendor migration:
1445		 * Intel's VMENTRY has a check on the 'G' bit.
1446		 */
1447		var->g = s->limit > 0xfffff;
1448		break;
1449	case VCPU_SREG_TR:
1450		/*
1451		 * Work around a bug where the busy flag in the tr selector
1452		 * isn't exposed
1453		 */
1454		var->type |= 0x2;
1455		break;
1456	case VCPU_SREG_DS:
1457	case VCPU_SREG_ES:
1458	case VCPU_SREG_FS:
1459	case VCPU_SREG_GS:
1460		/*
1461		 * The accessed bit must always be set in the segment
1462		 * descriptor cache, although it can be cleared in the
1463		 * descriptor, the cached bit always remains at 1. Since
1464		 * Intel has a check on this, set it here to support
1465		 * cross-vendor migration.
1466		 */
1467		if (!var->unusable)
1468			var->type |= 0x1;
1469		break;
1470	case VCPU_SREG_SS:
1471		/*
1472		 * On AMD CPUs sometimes the DB bit in the segment
1473		 * descriptor is left as 1, although the whole segment has
1474		 * been made unusable. Clear it here to pass an Intel VMX
1475		 * entry check when cross vendor migrating.
1476		 */
1477		if (var->unusable)
1478			var->db = 0;
1479		break;
1480	}
1481}
1482
1483static int svm_get_cpl(struct kvm_vcpu *vcpu)
1484{
1485	struct vmcb_save_area *save = &to_svm(vcpu)->vmcb->save;
1486
1487	return save->cpl;
1488}
1489
1490static void svm_get_idt(struct kvm_vcpu *vcpu, struct desc_ptr *dt)
1491{
1492	struct vcpu_svm *svm = to_svm(vcpu);
1493
1494	dt->size = svm->vmcb->save.idtr.limit;
1495	dt->address = svm->vmcb->save.idtr.base;
1496}
1497
1498static void svm_set_idt(struct kvm_vcpu *vcpu, struct desc_ptr *dt)
1499{
1500	struct vcpu_svm *svm = to_svm(vcpu);
1501
1502	svm->vmcb->save.idtr.limit = dt->size;
1503	svm->vmcb->save.idtr.base = dt->address ;
1504	mark_dirty(svm->vmcb, VMCB_DT);
1505}
1506
1507static void svm_get_gdt(struct kvm_vcpu *vcpu, struct desc_ptr *dt)
1508{
1509	struct vcpu_svm *svm = to_svm(vcpu);
1510
1511	dt->size = svm->vmcb->save.gdtr.limit;
1512	dt->address = svm->vmcb->save.gdtr.base;
1513}
1514
1515static void svm_set_gdt(struct kvm_vcpu *vcpu, struct desc_ptr *dt)
1516{
1517	struct vcpu_svm *svm = to_svm(vcpu);
1518
1519	svm->vmcb->save.gdtr.limit = dt->size;
1520	svm->vmcb->save.gdtr.base = dt->address ;
1521	mark_dirty(svm->vmcb, VMCB_DT);
1522}
1523
1524static void svm_decache_cr0_guest_bits(struct kvm_vcpu *vcpu)
1525{
1526}
1527
1528static void svm_decache_cr3(struct kvm_vcpu *vcpu)
1529{
1530}
1531
1532static void svm_decache_cr4_guest_bits(struct kvm_vcpu *vcpu)
1533{
1534}
1535
1536static void update_cr0_intercept(struct vcpu_svm *svm)
1537{
1538	ulong gcr0 = svm->vcpu.arch.cr0;
1539	u64 *hcr0 = &svm->vmcb->save.cr0;
1540
1541	if (!svm->vcpu.fpu_active)
1542		*hcr0 |= SVM_CR0_SELECTIVE_MASK;
1543	else
1544		*hcr0 = (*hcr0 & ~SVM_CR0_SELECTIVE_MASK)
1545			| (gcr0 & SVM_CR0_SELECTIVE_MASK);
1546
1547	mark_dirty(svm->vmcb, VMCB_CR);
1548
1549	if (gcr0 == *hcr0 && svm->vcpu.fpu_active) {
1550		clr_cr_intercept(svm, INTERCEPT_CR0_READ);
1551		clr_cr_intercept(svm, INTERCEPT_CR0_WRITE);
1552	} else {
1553		set_cr_intercept(svm, INTERCEPT_CR0_READ);
1554		set_cr_intercept(svm, INTERCEPT_CR0_WRITE);
1555	}
1556}
1557
1558static void svm_set_cr0(struct kvm_vcpu *vcpu, unsigned long cr0)
1559{
1560	struct vcpu_svm *svm = to_svm(vcpu);
1561
1562#ifdef CONFIG_X86_64
1563	if (vcpu->arch.efer & EFER_LME) {
1564		if (!is_paging(vcpu) && (cr0 & X86_CR0_PG)) {
1565			vcpu->arch.efer |= EFER_LMA;
1566			svm->vmcb->save.efer |= EFER_LMA | EFER_LME;
1567		}
1568
1569		if (is_paging(vcpu) && !(cr0 & X86_CR0_PG)) {
1570			vcpu->arch.efer &= ~EFER_LMA;
1571			svm->vmcb->save.efer &= ~(EFER_LMA | EFER_LME);
1572		}
1573	}
1574#endif
1575	vcpu->arch.cr0 = cr0;
1576
1577	if (!npt_enabled)
1578		cr0 |= X86_CR0_PG | X86_CR0_WP;
1579
1580	if (!vcpu->fpu_active)
1581		cr0 |= X86_CR0_TS;
1582	/*
1583	 * re-enable caching here because the QEMU bios
1584	 * does not do it - this results in some delay at
1585	 * reboot
1586	 */
1587	cr0 &= ~(X86_CR0_CD | X86_CR0_NW);
1588	svm->vmcb->save.cr0 = cr0;
1589	mark_dirty(svm->vmcb, VMCB_CR);
1590	update_cr0_intercept(svm);
1591}
1592
1593static int svm_set_cr4(struct kvm_vcpu *vcpu, unsigned long cr4)
1594{
1595	unsigned long host_cr4_mce = read_cr4() & X86_CR4_MCE;
1596	unsigned long old_cr4 = to_svm(vcpu)->vmcb->save.cr4;
1597
1598	if (cr4 & X86_CR4_VMXE)
1599		return 1;
1600
1601	if (npt_enabled && ((old_cr4 ^ cr4) & X86_CR4_PGE))
1602		svm_flush_tlb(vcpu);
1603
1604	vcpu->arch.cr4 = cr4;
1605	if (!npt_enabled)
1606		cr4 |= X86_CR4_PAE;
1607	cr4 |= host_cr4_mce;
1608	to_svm(vcpu)->vmcb->save.cr4 = cr4;
1609	mark_dirty(to_svm(vcpu)->vmcb, VMCB_CR);
1610	return 0;
1611}
1612
1613static void svm_set_segment(struct kvm_vcpu *vcpu,
1614			    struct kvm_segment *var, int seg)
1615{
1616	struct vcpu_svm *svm = to_svm(vcpu);
1617	struct vmcb_seg *s = svm_seg(vcpu, seg);
1618
1619	s->base = var->base;
1620	s->limit = var->limit;
1621	s->selector = var->selector;
1622	if (var->unusable)
1623		s->attrib = 0;
1624	else {
1625		s->attrib = (var->type & SVM_SELECTOR_TYPE_MASK);
1626		s->attrib |= (var->s & 1) << SVM_SELECTOR_S_SHIFT;
1627		s->attrib |= (var->dpl & 3) << SVM_SELECTOR_DPL_SHIFT;
1628		s->attrib |= (var->present & 1) << SVM_SELECTOR_P_SHIFT;
1629		s->attrib |= (var->avl & 1) << SVM_SELECTOR_AVL_SHIFT;
1630		s->attrib |= (var->l & 1) << SVM_SELECTOR_L_SHIFT;
1631		s->attrib |= (var->db & 1) << SVM_SELECTOR_DB_SHIFT;
1632		s->attrib |= (var->g & 1) << SVM_SELECTOR_G_SHIFT;
1633	}
1634	if (seg == VCPU_SREG_CS)
1635		svm_update_cpl(vcpu);
 
 
1636
1637	mark_dirty(svm->vmcb, VMCB_SEG);
1638}
1639
1640static void update_db_bp_intercept(struct kvm_vcpu *vcpu)
1641{
1642	struct vcpu_svm *svm = to_svm(vcpu);
1643
1644	clr_exception_intercept(svm, DB_VECTOR);
1645	clr_exception_intercept(svm, BP_VECTOR);
1646
1647	if (svm->nmi_singlestep)
1648		set_exception_intercept(svm, DB_VECTOR);
1649
1650	if (vcpu->guest_debug & KVM_GUESTDBG_ENABLE) {
1651		if (vcpu->guest_debug &
1652		    (KVM_GUESTDBG_SINGLESTEP | KVM_GUESTDBG_USE_HW_BP))
1653			set_exception_intercept(svm, DB_VECTOR);
1654		if (vcpu->guest_debug & KVM_GUESTDBG_USE_SW_BP)
1655			set_exception_intercept(svm, BP_VECTOR);
1656	} else
1657		vcpu->guest_debug = 0;
1658}
1659
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1660static void new_asid(struct vcpu_svm *svm, struct svm_cpu_data *sd)
1661{
1662	if (sd->next_asid > sd->max_asid) {
1663		++sd->asid_generation;
1664		sd->next_asid = 1;
1665		svm->vmcb->control.tlb_ctl = TLB_CONTROL_FLUSH_ALL_ASID;
1666	}
1667
1668	svm->asid_generation = sd->asid_generation;
1669	svm->vmcb->control.asid = sd->next_asid++;
1670
1671	mark_dirty(svm->vmcb, VMCB_ASID);
1672}
1673
1674static u64 svm_get_dr6(struct kvm_vcpu *vcpu)
1675{
1676	return to_svm(vcpu)->vmcb->save.dr6;
1677}
1678
1679static void svm_set_dr6(struct kvm_vcpu *vcpu, unsigned long value)
1680{
1681	struct vcpu_svm *svm = to_svm(vcpu);
1682
1683	svm->vmcb->save.dr6 = value;
1684	mark_dirty(svm->vmcb, VMCB_DR);
1685}
1686
1687static void svm_sync_dirty_debug_regs(struct kvm_vcpu *vcpu)
1688{
1689	struct vcpu_svm *svm = to_svm(vcpu);
1690
1691	get_debugreg(vcpu->arch.db[0], 0);
1692	get_debugreg(vcpu->arch.db[1], 1);
1693	get_debugreg(vcpu->arch.db[2], 2);
1694	get_debugreg(vcpu->arch.db[3], 3);
1695	vcpu->arch.dr6 = svm_get_dr6(vcpu);
1696	vcpu->arch.dr7 = svm->vmcb->save.dr7;
1697
1698	vcpu->arch.switch_db_regs &= ~KVM_DEBUGREG_WONT_EXIT;
1699	set_dr_intercepts(svm);
1700}
1701
1702static void svm_set_dr7(struct kvm_vcpu *vcpu, unsigned long value)
1703{
1704	struct vcpu_svm *svm = to_svm(vcpu);
1705
1706	svm->vmcb->save.dr7 = value;
1707	mark_dirty(svm->vmcb, VMCB_DR);
1708}
1709
1710static int pf_interception(struct vcpu_svm *svm)
1711{
1712	u64 fault_address = svm->vmcb->control.exit_info_2;
1713	u32 error_code;
1714	int r = 1;
1715
1716	switch (svm->apf_reason) {
1717	default:
1718		error_code = svm->vmcb->control.exit_info_1;
1719
1720		trace_kvm_page_fault(fault_address, error_code);
1721		if (!npt_enabled && kvm_event_needs_reinjection(&svm->vcpu))
1722			kvm_mmu_unprotect_page_virt(&svm->vcpu, fault_address);
1723		r = kvm_mmu_page_fault(&svm->vcpu, fault_address, error_code,
1724			svm->vmcb->control.insn_bytes,
1725			svm->vmcb->control.insn_len);
1726		break;
1727	case KVM_PV_REASON_PAGE_NOT_PRESENT:
1728		svm->apf_reason = 0;
1729		local_irq_disable();
1730		kvm_async_pf_task_wait(fault_address);
1731		local_irq_enable();
1732		break;
1733	case KVM_PV_REASON_PAGE_READY:
1734		svm->apf_reason = 0;
1735		local_irq_disable();
1736		kvm_async_pf_task_wake(fault_address);
1737		local_irq_enable();
1738		break;
1739	}
1740	return r;
1741}
1742
1743static int db_interception(struct vcpu_svm *svm)
1744{
1745	struct kvm_run *kvm_run = svm->vcpu.run;
1746
1747	if (!(svm->vcpu.guest_debug &
1748	      (KVM_GUESTDBG_SINGLESTEP | KVM_GUESTDBG_USE_HW_BP)) &&
1749		!svm->nmi_singlestep) {
1750		kvm_queue_exception(&svm->vcpu, DB_VECTOR);
1751		return 1;
1752	}
1753
1754	if (svm->nmi_singlestep) {
1755		svm->nmi_singlestep = false;
1756		if (!(svm->vcpu.guest_debug & KVM_GUESTDBG_SINGLESTEP))
1757			svm->vmcb->save.rflags &=
1758				~(X86_EFLAGS_TF | X86_EFLAGS_RF);
1759		update_db_bp_intercept(&svm->vcpu);
1760	}
1761
1762	if (svm->vcpu.guest_debug &
1763	    (KVM_GUESTDBG_SINGLESTEP | KVM_GUESTDBG_USE_HW_BP)) {
1764		kvm_run->exit_reason = KVM_EXIT_DEBUG;
1765		kvm_run->debug.arch.pc =
1766			svm->vmcb->save.cs.base + svm->vmcb->save.rip;
1767		kvm_run->debug.arch.exception = DB_VECTOR;
1768		return 0;
1769	}
1770
1771	return 1;
1772}
1773
1774static int bp_interception(struct vcpu_svm *svm)
1775{
1776	struct kvm_run *kvm_run = svm->vcpu.run;
1777
1778	kvm_run->exit_reason = KVM_EXIT_DEBUG;
1779	kvm_run->debug.arch.pc = svm->vmcb->save.cs.base + svm->vmcb->save.rip;
1780	kvm_run->debug.arch.exception = BP_VECTOR;
1781	return 0;
1782}
1783
1784static int ud_interception(struct vcpu_svm *svm)
1785{
1786	int er;
1787
1788	er = emulate_instruction(&svm->vcpu, EMULTYPE_TRAP_UD);
1789	if (er != EMULATE_DONE)
1790		kvm_queue_exception(&svm->vcpu, UD_VECTOR);
1791	return 1;
1792}
1793
1794static void svm_fpu_activate(struct kvm_vcpu *vcpu)
1795{
1796	struct vcpu_svm *svm = to_svm(vcpu);
1797
1798	clr_exception_intercept(svm, NM_VECTOR);
1799
1800	svm->vcpu.fpu_active = 1;
1801	update_cr0_intercept(svm);
1802}
1803
1804static int nm_interception(struct vcpu_svm *svm)
1805{
1806	svm_fpu_activate(&svm->vcpu);
1807	return 1;
1808}
1809
1810static bool is_erratum_383(void)
1811{
1812	int err, i;
1813	u64 value;
1814
1815	if (!erratum_383_found)
1816		return false;
1817
1818	value = native_read_msr_safe(MSR_IA32_MC0_STATUS, &err);
1819	if (err)
1820		return false;
1821
1822	/* Bit 62 may or may not be set for this mce */
1823	value &= ~(1ULL << 62);
1824
1825	if (value != 0xb600000000010015ULL)
1826		return false;
1827
1828	/* Clear MCi_STATUS registers */
1829	for (i = 0; i < 6; ++i)
1830		native_write_msr_safe(MSR_IA32_MCx_STATUS(i), 0, 0);
1831
1832	value = native_read_msr_safe(MSR_IA32_MCG_STATUS, &err);
1833	if (!err) {
1834		u32 low, high;
1835
1836		value &= ~(1ULL << 2);
1837		low    = lower_32_bits(value);
1838		high   = upper_32_bits(value);
1839
1840		native_write_msr_safe(MSR_IA32_MCG_STATUS, low, high);
1841	}
1842
1843	/* Flush tlb to evict multi-match entries */
1844	__flush_tlb_all();
1845
1846	return true;
1847}
1848
1849static void svm_handle_mce(struct vcpu_svm *svm)
1850{
1851	if (is_erratum_383()) {
1852		/*
1853		 * Erratum 383 triggered. Guest state is corrupt so kill the
1854		 * guest.
1855		 */
1856		pr_err("KVM: Guest triggered AMD Erratum 383\n");
1857
1858		kvm_make_request(KVM_REQ_TRIPLE_FAULT, &svm->vcpu);
1859
1860		return;
1861	}
1862
1863	/*
1864	 * On an #MC intercept the MCE handler is not called automatically in
1865	 * the host. So do it by hand here.
1866	 */
1867	asm volatile (
1868		"int $0x12\n");
1869	/* not sure if we ever come back to this point */
1870
1871	return;
1872}
1873
1874static int mc_interception(struct vcpu_svm *svm)
1875{
1876	return 1;
1877}
1878
1879static int shutdown_interception(struct vcpu_svm *svm)
1880{
1881	struct kvm_run *kvm_run = svm->vcpu.run;
1882
1883	/*
1884	 * VMCB is undefined after a SHUTDOWN intercept
1885	 * so reinitialize it.
1886	 */
1887	clear_page(svm->vmcb);
1888	init_vmcb(svm);
1889
1890	kvm_run->exit_reason = KVM_EXIT_SHUTDOWN;
1891	return 0;
1892}
1893
1894static int io_interception(struct vcpu_svm *svm)
1895{
1896	struct kvm_vcpu *vcpu = &svm->vcpu;
1897	u32 io_info = svm->vmcb->control.exit_info_1; /* address size bug? */
1898	int size, in, string;
1899	unsigned port;
1900
1901	++svm->vcpu.stat.io_exits;
1902	string = (io_info & SVM_IOIO_STR_MASK) != 0;
1903	in = (io_info & SVM_IOIO_TYPE_MASK) != 0;
1904	if (string || in)
1905		return emulate_instruction(vcpu, 0) == EMULATE_DONE;
1906
1907	port = io_info >> 16;
1908	size = (io_info & SVM_IOIO_SIZE_MASK) >> SVM_IOIO_SIZE_SHIFT;
1909	svm->next_rip = svm->vmcb->control.exit_info_2;
1910	skip_emulated_instruction(&svm->vcpu);
1911
1912	return kvm_fast_pio_out(vcpu, size, port);
1913}
1914
1915static int nmi_interception(struct vcpu_svm *svm)
1916{
1917	return 1;
1918}
1919
1920static int intr_interception(struct vcpu_svm *svm)
1921{
1922	++svm->vcpu.stat.irq_exits;
1923	return 1;
1924}
1925
1926static int nop_on_interception(struct vcpu_svm *svm)
1927{
1928	return 1;
1929}
1930
1931static int halt_interception(struct vcpu_svm *svm)
1932{
1933	svm->next_rip = kvm_rip_read(&svm->vcpu) + 1;
1934	skip_emulated_instruction(&svm->vcpu);
1935	return kvm_emulate_halt(&svm->vcpu);
1936}
1937
1938static int vmmcall_interception(struct vcpu_svm *svm)
1939{
1940	svm->next_rip = kvm_rip_read(&svm->vcpu) + 3;
1941	skip_emulated_instruction(&svm->vcpu);
1942	kvm_emulate_hypercall(&svm->vcpu);
1943	return 1;
1944}
1945
1946static unsigned long nested_svm_get_tdp_cr3(struct kvm_vcpu *vcpu)
1947{
1948	struct vcpu_svm *svm = to_svm(vcpu);
1949
1950	return svm->nested.nested_cr3;
1951}
1952
1953static u64 nested_svm_get_tdp_pdptr(struct kvm_vcpu *vcpu, int index)
1954{
1955	struct vcpu_svm *svm = to_svm(vcpu);
1956	u64 cr3 = svm->nested.nested_cr3;
1957	u64 pdpte;
1958	int ret;
1959
1960	ret = kvm_read_guest_page(vcpu->kvm, gpa_to_gfn(cr3), &pdpte,
1961				  offset_in_page(cr3) + index * 8, 8);
1962	if (ret)
1963		return 0;
1964	return pdpte;
1965}
1966
1967static void nested_svm_set_tdp_cr3(struct kvm_vcpu *vcpu,
1968				   unsigned long root)
1969{
1970	struct vcpu_svm *svm = to_svm(vcpu);
1971
1972	svm->vmcb->control.nested_cr3 = root;
1973	mark_dirty(svm->vmcb, VMCB_NPT);
1974	svm_flush_tlb(vcpu);
1975}
1976
1977static void nested_svm_inject_npf_exit(struct kvm_vcpu *vcpu,
1978				       struct x86_exception *fault)
1979{
1980	struct vcpu_svm *svm = to_svm(vcpu);
1981
1982	svm->vmcb->control.exit_code = SVM_EXIT_NPF;
1983	svm->vmcb->control.exit_code_hi = 0;
1984	svm->vmcb->control.exit_info_1 = fault->error_code;
1985	svm->vmcb->control.exit_info_2 = fault->address;
1986
1987	nested_svm_vmexit(svm);
1988}
1989
1990static void nested_svm_init_mmu_context(struct kvm_vcpu *vcpu)
1991{
1992	kvm_init_shadow_mmu(vcpu, &vcpu->arch.mmu);
 
 
1993
1994	vcpu->arch.mmu.set_cr3           = nested_svm_set_tdp_cr3;
1995	vcpu->arch.mmu.get_cr3           = nested_svm_get_tdp_cr3;
1996	vcpu->arch.mmu.get_pdptr         = nested_svm_get_tdp_pdptr;
1997	vcpu->arch.mmu.inject_page_fault = nested_svm_inject_npf_exit;
1998	vcpu->arch.mmu.shadow_root_level = get_npt_level();
1999	vcpu->arch.walk_mmu              = &vcpu->arch.nested_mmu;
 
 
2000}
2001
2002static void nested_svm_uninit_mmu_context(struct kvm_vcpu *vcpu)
2003{
2004	vcpu->arch.walk_mmu = &vcpu->arch.mmu;
2005}
2006
2007static int nested_svm_check_permissions(struct vcpu_svm *svm)
2008{
2009	if (!(svm->vcpu.arch.efer & EFER_SVME)
2010	    || !is_paging(&svm->vcpu)) {
2011		kvm_queue_exception(&svm->vcpu, UD_VECTOR);
2012		return 1;
2013	}
2014
2015	if (svm->vmcb->save.cpl) {
2016		kvm_inject_gp(&svm->vcpu, 0);
2017		return 1;
2018	}
2019
2020       return 0;
2021}
2022
2023static int nested_svm_check_exception(struct vcpu_svm *svm, unsigned nr,
2024				      bool has_error_code, u32 error_code)
2025{
2026	int vmexit;
2027
2028	if (!is_guest_mode(&svm->vcpu))
2029		return 0;
2030
2031	svm->vmcb->control.exit_code = SVM_EXIT_EXCP_BASE + nr;
2032	svm->vmcb->control.exit_code_hi = 0;
2033	svm->vmcb->control.exit_info_1 = error_code;
2034	svm->vmcb->control.exit_info_2 = svm->vcpu.arch.cr2;
2035
2036	vmexit = nested_svm_intercept(svm);
2037	if (vmexit == NESTED_EXIT_DONE)
2038		svm->nested.exit_required = true;
2039
2040	return vmexit;
2041}
2042
2043/* This function returns true if it is save to enable the irq window */
2044static inline bool nested_svm_intr(struct vcpu_svm *svm)
2045{
2046	if (!is_guest_mode(&svm->vcpu))
2047		return true;
2048
2049	if (!(svm->vcpu.arch.hflags & HF_VINTR_MASK))
2050		return true;
2051
2052	if (!(svm->vcpu.arch.hflags & HF_HIF_MASK))
2053		return false;
2054
2055	/*
2056	 * if vmexit was already requested (by intercepted exception
2057	 * for instance) do not overwrite it with "external interrupt"
2058	 * vmexit.
2059	 */
2060	if (svm->nested.exit_required)
2061		return false;
2062
2063	svm->vmcb->control.exit_code   = SVM_EXIT_INTR;
2064	svm->vmcb->control.exit_info_1 = 0;
2065	svm->vmcb->control.exit_info_2 = 0;
2066
2067	if (svm->nested.intercept & 1ULL) {
2068		/*
2069		 * The #vmexit can't be emulated here directly because this
2070		 * code path runs with irqs and preemption disabled. A
2071		 * #vmexit emulation might sleep. Only signal request for
2072		 * the #vmexit here.
2073		 */
2074		svm->nested.exit_required = true;
2075		trace_kvm_nested_intr_vmexit(svm->vmcb->save.rip);
2076		return false;
2077	}
2078
2079	return true;
2080}
2081
2082/* This function returns true if it is save to enable the nmi window */
2083static inline bool nested_svm_nmi(struct vcpu_svm *svm)
2084{
2085	if (!is_guest_mode(&svm->vcpu))
2086		return true;
2087
2088	if (!(svm->nested.intercept & (1ULL << INTERCEPT_NMI)))
2089		return true;
2090
2091	svm->vmcb->control.exit_code = SVM_EXIT_NMI;
2092	svm->nested.exit_required = true;
2093
2094	return false;
2095}
2096
2097static void *nested_svm_map(struct vcpu_svm *svm, u64 gpa, struct page **_page)
2098{
2099	struct page *page;
2100
2101	might_sleep();
2102
2103	page = gfn_to_page(svm->vcpu.kvm, gpa >> PAGE_SHIFT);
2104	if (is_error_page(page))
2105		goto error;
2106
2107	*_page = page;
2108
2109	return kmap(page);
2110
2111error:
 
2112	kvm_inject_gp(&svm->vcpu, 0);
2113
2114	return NULL;
2115}
2116
2117static void nested_svm_unmap(struct page *page)
2118{
2119	kunmap(page);
2120	kvm_release_page_dirty(page);
2121}
2122
2123static int nested_svm_intercept_ioio(struct vcpu_svm *svm)
2124{
2125	unsigned port;
2126	u8 val, bit;
2127	u64 gpa;
2128
2129	if (!(svm->nested.intercept & (1ULL << INTERCEPT_IOIO_PROT)))
2130		return NESTED_EXIT_HOST;
2131
2132	port = svm->vmcb->control.exit_info_1 >> 16;
2133	gpa  = svm->nested.vmcb_iopm + (port / 8);
2134	bit  = port % 8;
2135	val  = 0;
2136
2137	if (kvm_read_guest(svm->vcpu.kvm, gpa, &val, 1))
2138		val &= (1 << bit);
2139
2140	return val ? NESTED_EXIT_DONE : NESTED_EXIT_HOST;
2141}
2142
2143static int nested_svm_exit_handled_msr(struct vcpu_svm *svm)
2144{
2145	u32 offset, msr, value;
2146	int write, mask;
2147
2148	if (!(svm->nested.intercept & (1ULL << INTERCEPT_MSR_PROT)))
2149		return NESTED_EXIT_HOST;
2150
2151	msr    = svm->vcpu.arch.regs[VCPU_REGS_RCX];
2152	offset = svm_msrpm_offset(msr);
2153	write  = svm->vmcb->control.exit_info_1 & 1;
2154	mask   = 1 << ((2 * (msr & 0xf)) + write);
2155
2156	if (offset == MSR_INVALID)
2157		return NESTED_EXIT_DONE;
2158
2159	/* Offset is in 32 bit units but need in 8 bit units */
2160	offset *= 4;
2161
2162	if (kvm_read_guest(svm->vcpu.kvm, svm->nested.vmcb_msrpm + offset, &value, 4))
2163		return NESTED_EXIT_DONE;
2164
2165	return (value & mask) ? NESTED_EXIT_DONE : NESTED_EXIT_HOST;
2166}
2167
2168static int nested_svm_exit_special(struct vcpu_svm *svm)
2169{
2170	u32 exit_code = svm->vmcb->control.exit_code;
2171
2172	switch (exit_code) {
2173	case SVM_EXIT_INTR:
2174	case SVM_EXIT_NMI:
2175	case SVM_EXIT_EXCP_BASE + MC_VECTOR:
2176		return NESTED_EXIT_HOST;
2177	case SVM_EXIT_NPF:
2178		/* For now we are always handling NPFs when using them */
2179		if (npt_enabled)
2180			return NESTED_EXIT_HOST;
2181		break;
2182	case SVM_EXIT_EXCP_BASE + PF_VECTOR:
2183		/* When we're shadowing, trap PFs, but not async PF */
2184		if (!npt_enabled && svm->apf_reason == 0)
2185			return NESTED_EXIT_HOST;
2186		break;
2187	case SVM_EXIT_EXCP_BASE + NM_VECTOR:
2188		nm_interception(svm);
2189		break;
2190	default:
2191		break;
2192	}
2193
2194	return NESTED_EXIT_CONTINUE;
2195}
2196
2197/*
2198 * If this function returns true, this #vmexit was already handled
2199 */
2200static int nested_svm_intercept(struct vcpu_svm *svm)
2201{
2202	u32 exit_code = svm->vmcb->control.exit_code;
2203	int vmexit = NESTED_EXIT_HOST;
2204
2205	switch (exit_code) {
2206	case SVM_EXIT_MSR:
2207		vmexit = nested_svm_exit_handled_msr(svm);
2208		break;
2209	case SVM_EXIT_IOIO:
2210		vmexit = nested_svm_intercept_ioio(svm);
2211		break;
2212	case SVM_EXIT_READ_CR0 ... SVM_EXIT_WRITE_CR8: {
2213		u32 bit = 1U << (exit_code - SVM_EXIT_READ_CR0);
2214		if (svm->nested.intercept_cr & bit)
2215			vmexit = NESTED_EXIT_DONE;
2216		break;
2217	}
2218	case SVM_EXIT_READ_DR0 ... SVM_EXIT_WRITE_DR7: {
2219		u32 bit = 1U << (exit_code - SVM_EXIT_READ_DR0);
2220		if (svm->nested.intercept_dr & bit)
2221			vmexit = NESTED_EXIT_DONE;
2222		break;
2223	}
2224	case SVM_EXIT_EXCP_BASE ... SVM_EXIT_EXCP_BASE + 0x1f: {
2225		u32 excp_bits = 1 << (exit_code - SVM_EXIT_EXCP_BASE);
2226		if (svm->nested.intercept_exceptions & excp_bits)
2227			vmexit = NESTED_EXIT_DONE;
2228		/* async page fault always cause vmexit */
2229		else if ((exit_code == SVM_EXIT_EXCP_BASE + PF_VECTOR) &&
2230			 svm->apf_reason != 0)
2231			vmexit = NESTED_EXIT_DONE;
2232		break;
2233	}
2234	case SVM_EXIT_ERR: {
2235		vmexit = NESTED_EXIT_DONE;
2236		break;
2237	}
2238	default: {
2239		u64 exit_bits = 1ULL << (exit_code - SVM_EXIT_INTR);
2240		if (svm->nested.intercept & exit_bits)
2241			vmexit = NESTED_EXIT_DONE;
2242	}
2243	}
2244
2245	return vmexit;
2246}
2247
2248static int nested_svm_exit_handled(struct vcpu_svm *svm)
2249{
2250	int vmexit;
2251
2252	vmexit = nested_svm_intercept(svm);
2253
2254	if (vmexit == NESTED_EXIT_DONE)
2255		nested_svm_vmexit(svm);
2256
2257	return vmexit;
2258}
2259
2260static inline void copy_vmcb_control_area(struct vmcb *dst_vmcb, struct vmcb *from_vmcb)
2261{
2262	struct vmcb_control_area *dst  = &dst_vmcb->control;
2263	struct vmcb_control_area *from = &from_vmcb->control;
2264
2265	dst->intercept_cr         = from->intercept_cr;
2266	dst->intercept_dr         = from->intercept_dr;
2267	dst->intercept_exceptions = from->intercept_exceptions;
2268	dst->intercept            = from->intercept;
2269	dst->iopm_base_pa         = from->iopm_base_pa;
2270	dst->msrpm_base_pa        = from->msrpm_base_pa;
2271	dst->tsc_offset           = from->tsc_offset;
2272	dst->asid                 = from->asid;
2273	dst->tlb_ctl              = from->tlb_ctl;
2274	dst->int_ctl              = from->int_ctl;
2275	dst->int_vector           = from->int_vector;
2276	dst->int_state            = from->int_state;
2277	dst->exit_code            = from->exit_code;
2278	dst->exit_code_hi         = from->exit_code_hi;
2279	dst->exit_info_1          = from->exit_info_1;
2280	dst->exit_info_2          = from->exit_info_2;
2281	dst->exit_int_info        = from->exit_int_info;
2282	dst->exit_int_info_err    = from->exit_int_info_err;
2283	dst->nested_ctl           = from->nested_ctl;
2284	dst->event_inj            = from->event_inj;
2285	dst->event_inj_err        = from->event_inj_err;
2286	dst->nested_cr3           = from->nested_cr3;
2287	dst->lbr_ctl              = from->lbr_ctl;
2288}
2289
2290static int nested_svm_vmexit(struct vcpu_svm *svm)
2291{
2292	struct vmcb *nested_vmcb;
2293	struct vmcb *hsave = svm->nested.hsave;
2294	struct vmcb *vmcb = svm->vmcb;
2295	struct page *page;
2296
2297	trace_kvm_nested_vmexit_inject(vmcb->control.exit_code,
2298				       vmcb->control.exit_info_1,
2299				       vmcb->control.exit_info_2,
2300				       vmcb->control.exit_int_info,
2301				       vmcb->control.exit_int_info_err,
2302				       KVM_ISA_SVM);
2303
2304	nested_vmcb = nested_svm_map(svm, svm->nested.vmcb, &page);
2305	if (!nested_vmcb)
2306		return 1;
2307
2308	/* Exit Guest-Mode */
2309	leave_guest_mode(&svm->vcpu);
2310	svm->nested.vmcb = 0;
2311
2312	/* Give the current vmcb to the guest */
2313	disable_gif(svm);
2314
2315	nested_vmcb->save.es     = vmcb->save.es;
2316	nested_vmcb->save.cs     = vmcb->save.cs;
2317	nested_vmcb->save.ss     = vmcb->save.ss;
2318	nested_vmcb->save.ds     = vmcb->save.ds;
2319	nested_vmcb->save.gdtr   = vmcb->save.gdtr;
2320	nested_vmcb->save.idtr   = vmcb->save.idtr;
2321	nested_vmcb->save.efer   = svm->vcpu.arch.efer;
2322	nested_vmcb->save.cr0    = kvm_read_cr0(&svm->vcpu);
2323	nested_vmcb->save.cr3    = kvm_read_cr3(&svm->vcpu);
2324	nested_vmcb->save.cr2    = vmcb->save.cr2;
2325	nested_vmcb->save.cr4    = svm->vcpu.arch.cr4;
2326	nested_vmcb->save.rflags = kvm_get_rflags(&svm->vcpu);
2327	nested_vmcb->save.rip    = vmcb->save.rip;
2328	nested_vmcb->save.rsp    = vmcb->save.rsp;
2329	nested_vmcb->save.rax    = vmcb->save.rax;
2330	nested_vmcb->save.dr7    = vmcb->save.dr7;
2331	nested_vmcb->save.dr6    = vmcb->save.dr6;
2332	nested_vmcb->save.cpl    = vmcb->save.cpl;
2333
2334	nested_vmcb->control.int_ctl           = vmcb->control.int_ctl;
2335	nested_vmcb->control.int_vector        = vmcb->control.int_vector;
2336	nested_vmcb->control.int_state         = vmcb->control.int_state;
2337	nested_vmcb->control.exit_code         = vmcb->control.exit_code;
2338	nested_vmcb->control.exit_code_hi      = vmcb->control.exit_code_hi;
2339	nested_vmcb->control.exit_info_1       = vmcb->control.exit_info_1;
2340	nested_vmcb->control.exit_info_2       = vmcb->control.exit_info_2;
2341	nested_vmcb->control.exit_int_info     = vmcb->control.exit_int_info;
2342	nested_vmcb->control.exit_int_info_err = vmcb->control.exit_int_info_err;
2343	nested_vmcb->control.next_rip          = vmcb->control.next_rip;
2344
2345	/*
2346	 * If we emulate a VMRUN/#VMEXIT in the same host #vmexit cycle we have
2347	 * to make sure that we do not lose injected events. So check event_inj
2348	 * here and copy it to exit_int_info if it is valid.
2349	 * Exit_int_info and event_inj can't be both valid because the case
2350	 * below only happens on a VMRUN instruction intercept which has
2351	 * no valid exit_int_info set.
2352	 */
2353	if (vmcb->control.event_inj & SVM_EVTINJ_VALID) {
2354		struct vmcb_control_area *nc = &nested_vmcb->control;
2355
2356		nc->exit_int_info     = vmcb->control.event_inj;
2357		nc->exit_int_info_err = vmcb->control.event_inj_err;
2358	}
2359
2360	nested_vmcb->control.tlb_ctl           = 0;
2361	nested_vmcb->control.event_inj         = 0;
2362	nested_vmcb->control.event_inj_err     = 0;
2363
2364	/* We always set V_INTR_MASKING and remember the old value in hflags */
2365	if (!(svm->vcpu.arch.hflags & HF_VINTR_MASK))
2366		nested_vmcb->control.int_ctl &= ~V_INTR_MASKING_MASK;
2367
2368	/* Restore the original control entries */
2369	copy_vmcb_control_area(vmcb, hsave);
2370
2371	kvm_clear_exception_queue(&svm->vcpu);
2372	kvm_clear_interrupt_queue(&svm->vcpu);
2373
2374	svm->nested.nested_cr3 = 0;
2375
2376	/* Restore selected save entries */
2377	svm->vmcb->save.es = hsave->save.es;
2378	svm->vmcb->save.cs = hsave->save.cs;
2379	svm->vmcb->save.ss = hsave->save.ss;
2380	svm->vmcb->save.ds = hsave->save.ds;
2381	svm->vmcb->save.gdtr = hsave->save.gdtr;
2382	svm->vmcb->save.idtr = hsave->save.idtr;
2383	kvm_set_rflags(&svm->vcpu, hsave->save.rflags);
2384	svm_set_efer(&svm->vcpu, hsave->save.efer);
2385	svm_set_cr0(&svm->vcpu, hsave->save.cr0 | X86_CR0_PE);
2386	svm_set_cr4(&svm->vcpu, hsave->save.cr4);
2387	if (npt_enabled) {
2388		svm->vmcb->save.cr3 = hsave->save.cr3;
2389		svm->vcpu.arch.cr3 = hsave->save.cr3;
2390	} else {
2391		(void)kvm_set_cr3(&svm->vcpu, hsave->save.cr3);
2392	}
2393	kvm_register_write(&svm->vcpu, VCPU_REGS_RAX, hsave->save.rax);
2394	kvm_register_write(&svm->vcpu, VCPU_REGS_RSP, hsave->save.rsp);
2395	kvm_register_write(&svm->vcpu, VCPU_REGS_RIP, hsave->save.rip);
2396	svm->vmcb->save.dr7 = 0;
2397	svm->vmcb->save.cpl = 0;
2398	svm->vmcb->control.exit_int_info = 0;
2399
2400	mark_all_dirty(svm->vmcb);
2401
2402	nested_svm_unmap(page);
2403
2404	nested_svm_uninit_mmu_context(&svm->vcpu);
2405	kvm_mmu_reset_context(&svm->vcpu);
2406	kvm_mmu_load(&svm->vcpu);
2407
2408	return 0;
2409}
2410
2411static bool nested_svm_vmrun_msrpm(struct vcpu_svm *svm)
2412{
2413	/*
2414	 * This function merges the msr permission bitmaps of kvm and the
2415	 * nested vmcb. It is optimized in that it only merges the parts where
2416	 * the kvm msr permission bitmap may contain zero bits
2417	 */
2418	int i;
2419
2420	if (!(svm->nested.intercept & (1ULL << INTERCEPT_MSR_PROT)))
2421		return true;
2422
2423	for (i = 0; i < MSRPM_OFFSETS; i++) {
2424		u32 value, p;
2425		u64 offset;
2426
2427		if (msrpm_offsets[i] == 0xffffffff)
2428			break;
2429
2430		p      = msrpm_offsets[i];
2431		offset = svm->nested.vmcb_msrpm + (p * 4);
2432
2433		if (kvm_read_guest(svm->vcpu.kvm, offset, &value, 4))
2434			return false;
2435
2436		svm->nested.msrpm[p] = svm->msrpm[p] | value;
2437	}
2438
2439	svm->vmcb->control.msrpm_base_pa = __pa(svm->nested.msrpm);
2440
2441	return true;
2442}
2443
2444static bool nested_vmcb_checks(struct vmcb *vmcb)
2445{
2446	if ((vmcb->control.intercept & (1ULL << INTERCEPT_VMRUN)) == 0)
2447		return false;
2448
2449	if (vmcb->control.asid == 0)
2450		return false;
2451
2452	if (vmcb->control.nested_ctl && !npt_enabled)
2453		return false;
2454
2455	return true;
2456}
2457
2458static bool nested_svm_vmrun(struct vcpu_svm *svm)
2459{
2460	struct vmcb *nested_vmcb;
2461	struct vmcb *hsave = svm->nested.hsave;
2462	struct vmcb *vmcb = svm->vmcb;
2463	struct page *page;
2464	u64 vmcb_gpa;
2465
2466	vmcb_gpa = svm->vmcb->save.rax;
2467
2468	nested_vmcb = nested_svm_map(svm, svm->vmcb->save.rax, &page);
2469	if (!nested_vmcb)
2470		return false;
2471
2472	if (!nested_vmcb_checks(nested_vmcb)) {
2473		nested_vmcb->control.exit_code    = SVM_EXIT_ERR;
2474		nested_vmcb->control.exit_code_hi = 0;
2475		nested_vmcb->control.exit_info_1  = 0;
2476		nested_vmcb->control.exit_info_2  = 0;
2477
2478		nested_svm_unmap(page);
2479
2480		return false;
2481	}
2482
2483	trace_kvm_nested_vmrun(svm->vmcb->save.rip, vmcb_gpa,
2484			       nested_vmcb->save.rip,
2485			       nested_vmcb->control.int_ctl,
2486			       nested_vmcb->control.event_inj,
2487			       nested_vmcb->control.nested_ctl);
2488
2489	trace_kvm_nested_intercepts(nested_vmcb->control.intercept_cr & 0xffff,
2490				    nested_vmcb->control.intercept_cr >> 16,
2491				    nested_vmcb->control.intercept_exceptions,
2492				    nested_vmcb->control.intercept);
2493
2494	/* Clear internal status */
2495	kvm_clear_exception_queue(&svm->vcpu);
2496	kvm_clear_interrupt_queue(&svm->vcpu);
2497
2498	/*
2499	 * Save the old vmcb, so we don't need to pick what we save, but can
2500	 * restore everything when a VMEXIT occurs
2501	 */
2502	hsave->save.es     = vmcb->save.es;
2503	hsave->save.cs     = vmcb->save.cs;
2504	hsave->save.ss     = vmcb->save.ss;
2505	hsave->save.ds     = vmcb->save.ds;
2506	hsave->save.gdtr   = vmcb->save.gdtr;
2507	hsave->save.idtr   = vmcb->save.idtr;
2508	hsave->save.efer   = svm->vcpu.arch.efer;
2509	hsave->save.cr0    = kvm_read_cr0(&svm->vcpu);
2510	hsave->save.cr4    = svm->vcpu.arch.cr4;
2511	hsave->save.rflags = kvm_get_rflags(&svm->vcpu);
2512	hsave->save.rip    = kvm_rip_read(&svm->vcpu);
2513	hsave->save.rsp    = vmcb->save.rsp;
2514	hsave->save.rax    = vmcb->save.rax;
2515	if (npt_enabled)
2516		hsave->save.cr3    = vmcb->save.cr3;
2517	else
2518		hsave->save.cr3    = kvm_read_cr3(&svm->vcpu);
2519
2520	copy_vmcb_control_area(hsave, vmcb);
2521
2522	if (kvm_get_rflags(&svm->vcpu) & X86_EFLAGS_IF)
2523		svm->vcpu.arch.hflags |= HF_HIF_MASK;
2524	else
2525		svm->vcpu.arch.hflags &= ~HF_HIF_MASK;
2526
2527	if (nested_vmcb->control.nested_ctl) {
2528		kvm_mmu_unload(&svm->vcpu);
2529		svm->nested.nested_cr3 = nested_vmcb->control.nested_cr3;
2530		nested_svm_init_mmu_context(&svm->vcpu);
2531	}
2532
2533	/* Load the nested guest state */
2534	svm->vmcb->save.es = nested_vmcb->save.es;
2535	svm->vmcb->save.cs = nested_vmcb->save.cs;
2536	svm->vmcb->save.ss = nested_vmcb->save.ss;
2537	svm->vmcb->save.ds = nested_vmcb->save.ds;
2538	svm->vmcb->save.gdtr = nested_vmcb->save.gdtr;
2539	svm->vmcb->save.idtr = nested_vmcb->save.idtr;
2540	kvm_set_rflags(&svm->vcpu, nested_vmcb->save.rflags);
2541	svm_set_efer(&svm->vcpu, nested_vmcb->save.efer);
2542	svm_set_cr0(&svm->vcpu, nested_vmcb->save.cr0);
2543	svm_set_cr4(&svm->vcpu, nested_vmcb->save.cr4);
2544	if (npt_enabled) {
2545		svm->vmcb->save.cr3 = nested_vmcb->save.cr3;
2546		svm->vcpu.arch.cr3 = nested_vmcb->save.cr3;
2547	} else
2548		(void)kvm_set_cr3(&svm->vcpu, nested_vmcb->save.cr3);
2549
2550	/* Guest paging mode is active - reset mmu */
2551	kvm_mmu_reset_context(&svm->vcpu);
2552
2553	svm->vmcb->save.cr2 = svm->vcpu.arch.cr2 = nested_vmcb->save.cr2;
2554	kvm_register_write(&svm->vcpu, VCPU_REGS_RAX, nested_vmcb->save.rax);
2555	kvm_register_write(&svm->vcpu, VCPU_REGS_RSP, nested_vmcb->save.rsp);
2556	kvm_register_write(&svm->vcpu, VCPU_REGS_RIP, nested_vmcb->save.rip);
2557
2558	/* In case we don't even reach vcpu_run, the fields are not updated */
2559	svm->vmcb->save.rax = nested_vmcb->save.rax;
2560	svm->vmcb->save.rsp = nested_vmcb->save.rsp;
2561	svm->vmcb->save.rip = nested_vmcb->save.rip;
2562	svm->vmcb->save.dr7 = nested_vmcb->save.dr7;
2563	svm->vmcb->save.dr6 = nested_vmcb->save.dr6;
2564	svm->vmcb->save.cpl = nested_vmcb->save.cpl;
2565
2566	svm->nested.vmcb_msrpm = nested_vmcb->control.msrpm_base_pa & ~0x0fffULL;
2567	svm->nested.vmcb_iopm  = nested_vmcb->control.iopm_base_pa  & ~0x0fffULL;
2568
2569	/* cache intercepts */
2570	svm->nested.intercept_cr         = nested_vmcb->control.intercept_cr;
2571	svm->nested.intercept_dr         = nested_vmcb->control.intercept_dr;
2572	svm->nested.intercept_exceptions = nested_vmcb->control.intercept_exceptions;
2573	svm->nested.intercept            = nested_vmcb->control.intercept;
2574
2575	svm_flush_tlb(&svm->vcpu);
2576	svm->vmcb->control.int_ctl = nested_vmcb->control.int_ctl | V_INTR_MASKING_MASK;
2577	if (nested_vmcb->control.int_ctl & V_INTR_MASKING_MASK)
2578		svm->vcpu.arch.hflags |= HF_VINTR_MASK;
2579	else
2580		svm->vcpu.arch.hflags &= ~HF_VINTR_MASK;
2581
2582	if (svm->vcpu.arch.hflags & HF_VINTR_MASK) {
2583		/* We only want the cr8 intercept bits of the guest */
2584		clr_cr_intercept(svm, INTERCEPT_CR8_READ);
2585		clr_cr_intercept(svm, INTERCEPT_CR8_WRITE);
2586	}
2587
2588	/* We don't want to see VMMCALLs from a nested guest */
2589	clr_intercept(svm, INTERCEPT_VMMCALL);
2590
2591	svm->vmcb->control.lbr_ctl = nested_vmcb->control.lbr_ctl;
2592	svm->vmcb->control.int_vector = nested_vmcb->control.int_vector;
2593	svm->vmcb->control.int_state = nested_vmcb->control.int_state;
2594	svm->vmcb->control.tsc_offset += nested_vmcb->control.tsc_offset;
2595	svm->vmcb->control.event_inj = nested_vmcb->control.event_inj;
2596	svm->vmcb->control.event_inj_err = nested_vmcb->control.event_inj_err;
2597
2598	nested_svm_unmap(page);
2599
2600	/* Enter Guest-Mode */
2601	enter_guest_mode(&svm->vcpu);
2602
2603	/*
2604	 * Merge guest and host intercepts - must be called  with vcpu in
2605	 * guest-mode to take affect here
2606	 */
2607	recalc_intercepts(svm);
2608
2609	svm->nested.vmcb = vmcb_gpa;
2610
2611	enable_gif(svm);
2612
2613	mark_all_dirty(svm->vmcb);
2614
2615	return true;
2616}
2617
2618static void nested_svm_vmloadsave(struct vmcb *from_vmcb, struct vmcb *to_vmcb)
2619{
2620	to_vmcb->save.fs = from_vmcb->save.fs;
2621	to_vmcb->save.gs = from_vmcb->save.gs;
2622	to_vmcb->save.tr = from_vmcb->save.tr;
2623	to_vmcb->save.ldtr = from_vmcb->save.ldtr;
2624	to_vmcb->save.kernel_gs_base = from_vmcb->save.kernel_gs_base;
2625	to_vmcb->save.star = from_vmcb->save.star;
2626	to_vmcb->save.lstar = from_vmcb->save.lstar;
2627	to_vmcb->save.cstar = from_vmcb->save.cstar;
2628	to_vmcb->save.sfmask = from_vmcb->save.sfmask;
2629	to_vmcb->save.sysenter_cs = from_vmcb->save.sysenter_cs;
2630	to_vmcb->save.sysenter_esp = from_vmcb->save.sysenter_esp;
2631	to_vmcb->save.sysenter_eip = from_vmcb->save.sysenter_eip;
2632}
2633
2634static int vmload_interception(struct vcpu_svm *svm)
2635{
2636	struct vmcb *nested_vmcb;
2637	struct page *page;
2638
2639	if (nested_svm_check_permissions(svm))
2640		return 1;
2641
2642	nested_vmcb = nested_svm_map(svm, svm->vmcb->save.rax, &page);
2643	if (!nested_vmcb)
2644		return 1;
2645
2646	svm->next_rip = kvm_rip_read(&svm->vcpu) + 3;
2647	skip_emulated_instruction(&svm->vcpu);
2648
2649	nested_svm_vmloadsave(nested_vmcb, svm->vmcb);
2650	nested_svm_unmap(page);
2651
2652	return 1;
2653}
2654
2655static int vmsave_interception(struct vcpu_svm *svm)
2656{
2657	struct vmcb *nested_vmcb;
2658	struct page *page;
2659
2660	if (nested_svm_check_permissions(svm))
2661		return 1;
2662
2663	nested_vmcb = nested_svm_map(svm, svm->vmcb->save.rax, &page);
2664	if (!nested_vmcb)
2665		return 1;
2666
2667	svm->next_rip = kvm_rip_read(&svm->vcpu) + 3;
2668	skip_emulated_instruction(&svm->vcpu);
2669
2670	nested_svm_vmloadsave(svm->vmcb, nested_vmcb);
2671	nested_svm_unmap(page);
2672
2673	return 1;
2674}
2675
2676static int vmrun_interception(struct vcpu_svm *svm)
2677{
2678	if (nested_svm_check_permissions(svm))
2679		return 1;
2680
2681	/* Save rip after vmrun instruction */
2682	kvm_rip_write(&svm->vcpu, kvm_rip_read(&svm->vcpu) + 3);
2683
2684	if (!nested_svm_vmrun(svm))
2685		return 1;
2686
2687	if (!nested_svm_vmrun_msrpm(svm))
2688		goto failed;
2689
2690	return 1;
2691
2692failed:
2693
2694	svm->vmcb->control.exit_code    = SVM_EXIT_ERR;
2695	svm->vmcb->control.exit_code_hi = 0;
2696	svm->vmcb->control.exit_info_1  = 0;
2697	svm->vmcb->control.exit_info_2  = 0;
2698
2699	nested_svm_vmexit(svm);
2700
2701	return 1;
2702}
2703
2704static int stgi_interception(struct vcpu_svm *svm)
2705{
2706	if (nested_svm_check_permissions(svm))
2707		return 1;
2708
2709	svm->next_rip = kvm_rip_read(&svm->vcpu) + 3;
2710	skip_emulated_instruction(&svm->vcpu);
2711	kvm_make_request(KVM_REQ_EVENT, &svm->vcpu);
2712
2713	enable_gif(svm);
2714
2715	return 1;
2716}
2717
2718static int clgi_interception(struct vcpu_svm *svm)
2719{
2720	if (nested_svm_check_permissions(svm))
2721		return 1;
2722
2723	svm->next_rip = kvm_rip_read(&svm->vcpu) + 3;
2724	skip_emulated_instruction(&svm->vcpu);
2725
2726	disable_gif(svm);
2727
2728	/* After a CLGI no interrupts should come */
2729	svm_clear_vintr(svm);
2730	svm->vmcb->control.int_ctl &= ~V_IRQ_MASK;
2731
2732	mark_dirty(svm->vmcb, VMCB_INTR);
2733
2734	return 1;
2735}
2736
2737static int invlpga_interception(struct vcpu_svm *svm)
2738{
2739	struct kvm_vcpu *vcpu = &svm->vcpu;
2740
2741	trace_kvm_invlpga(svm->vmcb->save.rip, vcpu->arch.regs[VCPU_REGS_RCX],
2742			  vcpu->arch.regs[VCPU_REGS_RAX]);
2743
2744	/* Let's treat INVLPGA the same as INVLPG (can be optimized!) */
2745	kvm_mmu_invlpg(vcpu, vcpu->arch.regs[VCPU_REGS_RAX]);
2746
2747	svm->next_rip = kvm_rip_read(&svm->vcpu) + 3;
2748	skip_emulated_instruction(&svm->vcpu);
2749	return 1;
2750}
2751
2752static int skinit_interception(struct vcpu_svm *svm)
2753{
2754	trace_kvm_skinit(svm->vmcb->save.rip, svm->vcpu.arch.regs[VCPU_REGS_RAX]);
2755
2756	kvm_queue_exception(&svm->vcpu, UD_VECTOR);
2757	return 1;
2758}
2759
2760static int xsetbv_interception(struct vcpu_svm *svm)
2761{
2762	u64 new_bv = kvm_read_edx_eax(&svm->vcpu);
2763	u32 index = kvm_register_read(&svm->vcpu, VCPU_REGS_RCX);
2764
2765	if (kvm_set_xcr(&svm->vcpu, index, new_bv) == 0) {
2766		svm->next_rip = kvm_rip_read(&svm->vcpu) + 3;
2767		skip_emulated_instruction(&svm->vcpu);
2768	}
2769
2770	return 1;
2771}
2772
2773static int invalid_op_interception(struct vcpu_svm *svm)
2774{
2775	kvm_queue_exception(&svm->vcpu, UD_VECTOR);
2776	return 1;
2777}
2778
2779static int task_switch_interception(struct vcpu_svm *svm)
2780{
2781	u16 tss_selector;
2782	int reason;
2783	int int_type = svm->vmcb->control.exit_int_info &
2784		SVM_EXITINTINFO_TYPE_MASK;
2785	int int_vec = svm->vmcb->control.exit_int_info & SVM_EVTINJ_VEC_MASK;
2786	uint32_t type =
2787		svm->vmcb->control.exit_int_info & SVM_EXITINTINFO_TYPE_MASK;
2788	uint32_t idt_v =
2789		svm->vmcb->control.exit_int_info & SVM_EXITINTINFO_VALID;
2790	bool has_error_code = false;
2791	u32 error_code = 0;
2792
2793	tss_selector = (u16)svm->vmcb->control.exit_info_1;
2794
2795	if (svm->vmcb->control.exit_info_2 &
2796	    (1ULL << SVM_EXITINFOSHIFT_TS_REASON_IRET))
2797		reason = TASK_SWITCH_IRET;
2798	else if (svm->vmcb->control.exit_info_2 &
2799		 (1ULL << SVM_EXITINFOSHIFT_TS_REASON_JMP))
2800		reason = TASK_SWITCH_JMP;
2801	else if (idt_v)
2802		reason = TASK_SWITCH_GATE;
2803	else
2804		reason = TASK_SWITCH_CALL;
2805
2806	if (reason == TASK_SWITCH_GATE) {
2807		switch (type) {
2808		case SVM_EXITINTINFO_TYPE_NMI:
2809			svm->vcpu.arch.nmi_injected = false;
2810			break;
2811		case SVM_EXITINTINFO_TYPE_EXEPT:
2812			if (svm->vmcb->control.exit_info_2 &
2813			    (1ULL << SVM_EXITINFOSHIFT_TS_HAS_ERROR_CODE)) {
2814				has_error_code = true;
2815				error_code =
2816					(u32)svm->vmcb->control.exit_info_2;
2817			}
2818			kvm_clear_exception_queue(&svm->vcpu);
2819			break;
2820		case SVM_EXITINTINFO_TYPE_INTR:
2821			kvm_clear_interrupt_queue(&svm->vcpu);
2822			break;
2823		default:
2824			break;
2825		}
2826	}
2827
2828	if (reason != TASK_SWITCH_GATE ||
2829	    int_type == SVM_EXITINTINFO_TYPE_SOFT ||
2830	    (int_type == SVM_EXITINTINFO_TYPE_EXEPT &&
2831	     (int_vec == OF_VECTOR || int_vec == BP_VECTOR)))
2832		skip_emulated_instruction(&svm->vcpu);
2833
2834	if (int_type != SVM_EXITINTINFO_TYPE_SOFT)
2835		int_vec = -1;
2836
2837	if (kvm_task_switch(&svm->vcpu, tss_selector, int_vec, reason,
2838				has_error_code, error_code) == EMULATE_FAIL) {
2839		svm->vcpu.run->exit_reason = KVM_EXIT_INTERNAL_ERROR;
2840		svm->vcpu.run->internal.suberror = KVM_INTERNAL_ERROR_EMULATION;
2841		svm->vcpu.run->internal.ndata = 0;
2842		return 0;
2843	}
2844	return 1;
2845}
2846
2847static int cpuid_interception(struct vcpu_svm *svm)
2848{
2849	svm->next_rip = kvm_rip_read(&svm->vcpu) + 2;
2850	kvm_emulate_cpuid(&svm->vcpu);
2851	return 1;
2852}
2853
2854static int iret_interception(struct vcpu_svm *svm)
2855{
2856	++svm->vcpu.stat.nmi_window_exits;
2857	clr_intercept(svm, INTERCEPT_IRET);
2858	svm->vcpu.arch.hflags |= HF_IRET_MASK;
2859	svm->nmi_iret_rip = kvm_rip_read(&svm->vcpu);
2860	kvm_make_request(KVM_REQ_EVENT, &svm->vcpu);
2861	return 1;
2862}
2863
2864static int invlpg_interception(struct vcpu_svm *svm)
2865{
2866	if (!static_cpu_has(X86_FEATURE_DECODEASSISTS))
2867		return emulate_instruction(&svm->vcpu, 0) == EMULATE_DONE;
2868
2869	kvm_mmu_invlpg(&svm->vcpu, svm->vmcb->control.exit_info_1);
2870	skip_emulated_instruction(&svm->vcpu);
2871	return 1;
2872}
2873
2874static int emulate_on_interception(struct vcpu_svm *svm)
2875{
2876	return emulate_instruction(&svm->vcpu, 0) == EMULATE_DONE;
2877}
2878
2879static int rdpmc_interception(struct vcpu_svm *svm)
2880{
2881	int err;
2882
2883	if (!static_cpu_has(X86_FEATURE_NRIPS))
2884		return emulate_on_interception(svm);
2885
2886	err = kvm_rdpmc(&svm->vcpu);
2887	kvm_complete_insn_gp(&svm->vcpu, err);
2888
2889	return 1;
2890}
2891
2892bool check_selective_cr0_intercepted(struct vcpu_svm *svm, unsigned long val)
2893{
2894	unsigned long cr0 = svm->vcpu.arch.cr0;
2895	bool ret = false;
2896	u64 intercept;
2897
2898	intercept = svm->nested.intercept;
2899
2900	if (!is_guest_mode(&svm->vcpu) ||
2901	    (!(intercept & (1ULL << INTERCEPT_SELECTIVE_CR0))))
2902		return false;
2903
2904	cr0 &= ~SVM_CR0_SELECTIVE_MASK;
2905	val &= ~SVM_CR0_SELECTIVE_MASK;
2906
2907	if (cr0 ^ val) {
2908		svm->vmcb->control.exit_code = SVM_EXIT_CR0_SEL_WRITE;
2909		ret = (nested_svm_exit_handled(svm) == NESTED_EXIT_DONE);
2910	}
2911
2912	return ret;
2913}
2914
2915#define CR_VALID (1ULL << 63)
2916
2917static int cr_interception(struct vcpu_svm *svm)
2918{
2919	int reg, cr;
2920	unsigned long val;
2921	int err;
2922
2923	if (!static_cpu_has(X86_FEATURE_DECODEASSISTS))
2924		return emulate_on_interception(svm);
2925
2926	if (unlikely((svm->vmcb->control.exit_info_1 & CR_VALID) == 0))
2927		return emulate_on_interception(svm);
2928
2929	reg = svm->vmcb->control.exit_info_1 & SVM_EXITINFO_REG_MASK;
2930	cr = svm->vmcb->control.exit_code - SVM_EXIT_READ_CR0;
2931
2932	err = 0;
2933	if (cr >= 16) { /* mov to cr */
2934		cr -= 16;
2935		val = kvm_register_read(&svm->vcpu, reg);
2936		switch (cr) {
2937		case 0:
2938			if (!check_selective_cr0_intercepted(svm, val))
2939				err = kvm_set_cr0(&svm->vcpu, val);
2940			else
2941				return 1;
2942
2943			break;
2944		case 3:
2945			err = kvm_set_cr3(&svm->vcpu, val);
2946			break;
2947		case 4:
2948			err = kvm_set_cr4(&svm->vcpu, val);
2949			break;
2950		case 8:
2951			err = kvm_set_cr8(&svm->vcpu, val);
2952			break;
2953		default:
2954			WARN(1, "unhandled write to CR%d", cr);
2955			kvm_queue_exception(&svm->vcpu, UD_VECTOR);
2956			return 1;
2957		}
2958	} else { /* mov from cr */
2959		switch (cr) {
2960		case 0:
2961			val = kvm_read_cr0(&svm->vcpu);
2962			break;
2963		case 2:
2964			val = svm->vcpu.arch.cr2;
2965			break;
2966		case 3:
2967			val = kvm_read_cr3(&svm->vcpu);
2968			break;
2969		case 4:
2970			val = kvm_read_cr4(&svm->vcpu);
2971			break;
2972		case 8:
2973			val = kvm_get_cr8(&svm->vcpu);
2974			break;
2975		default:
2976			WARN(1, "unhandled read from CR%d", cr);
2977			kvm_queue_exception(&svm->vcpu, UD_VECTOR);
2978			return 1;
2979		}
2980		kvm_register_write(&svm->vcpu, reg, val);
2981	}
2982	kvm_complete_insn_gp(&svm->vcpu, err);
2983
2984	return 1;
2985}
2986
2987static int dr_interception(struct vcpu_svm *svm)
2988{
2989	int reg, dr;
2990	unsigned long val;
2991	int err;
2992
2993	if (svm->vcpu.guest_debug == 0) {
2994		/*
2995		 * No more DR vmexits; force a reload of the debug registers
2996		 * and reenter on this instruction.  The next vmexit will
2997		 * retrieve the full state of the debug registers.
2998		 */
2999		clr_dr_intercepts(svm);
3000		svm->vcpu.arch.switch_db_regs |= KVM_DEBUGREG_WONT_EXIT;
3001		return 1;
3002	}
3003
3004	if (!boot_cpu_has(X86_FEATURE_DECODEASSISTS))
3005		return emulate_on_interception(svm);
3006
3007	reg = svm->vmcb->control.exit_info_1 & SVM_EXITINFO_REG_MASK;
3008	dr = svm->vmcb->control.exit_code - SVM_EXIT_READ_DR0;
3009
3010	if (dr >= 16) { /* mov to DRn */
3011		val = kvm_register_read(&svm->vcpu, reg);
3012		kvm_set_dr(&svm->vcpu, dr - 16, val);
3013	} else {
3014		err = kvm_get_dr(&svm->vcpu, dr, &val);
3015		if (!err)
3016			kvm_register_write(&svm->vcpu, reg, val);
3017	}
3018
3019	skip_emulated_instruction(&svm->vcpu);
3020
3021	return 1;
3022}
3023
3024static int cr8_write_interception(struct vcpu_svm *svm)
3025{
3026	struct kvm_run *kvm_run = svm->vcpu.run;
3027	int r;
3028
3029	u8 cr8_prev = kvm_get_cr8(&svm->vcpu);
3030	/* instruction emulation calls kvm_set_cr8() */
3031	r = cr_interception(svm);
3032	if (irqchip_in_kernel(svm->vcpu.kvm))
 
3033		return r;
 
3034	if (cr8_prev <= kvm_get_cr8(&svm->vcpu))
3035		return r;
3036	kvm_run->exit_reason = KVM_EXIT_SET_TPR;
3037	return 0;
3038}
3039
3040u64 svm_read_l1_tsc(struct kvm_vcpu *vcpu, u64 host_tsc)
3041{
3042	struct vmcb *vmcb = get_host_vmcb(to_svm(vcpu));
3043	return vmcb->control.tsc_offset +
3044		svm_scale_tsc(vcpu, host_tsc);
3045}
3046
3047static int svm_get_msr(struct kvm_vcpu *vcpu, unsigned ecx, u64 *data)
3048{
3049	struct vcpu_svm *svm = to_svm(vcpu);
3050
3051	switch (ecx) {
3052	case MSR_IA32_TSC: {
3053		*data = svm->vmcb->control.tsc_offset +
 
 
3054			svm_scale_tsc(vcpu, native_read_tsc());
3055
3056		break;
3057	}
3058	case MSR_STAR:
3059		*data = svm->vmcb->save.star;
3060		break;
3061#ifdef CONFIG_X86_64
3062	case MSR_LSTAR:
3063		*data = svm->vmcb->save.lstar;
3064		break;
3065	case MSR_CSTAR:
3066		*data = svm->vmcb->save.cstar;
3067		break;
3068	case MSR_KERNEL_GS_BASE:
3069		*data = svm->vmcb->save.kernel_gs_base;
3070		break;
3071	case MSR_SYSCALL_MASK:
3072		*data = svm->vmcb->save.sfmask;
3073		break;
3074#endif
3075	case MSR_IA32_SYSENTER_CS:
3076		*data = svm->vmcb->save.sysenter_cs;
3077		break;
3078	case MSR_IA32_SYSENTER_EIP:
3079		*data = svm->sysenter_eip;
3080		break;
3081	case MSR_IA32_SYSENTER_ESP:
3082		*data = svm->sysenter_esp;
3083		break;
3084	/*
3085	 * Nobody will change the following 5 values in the VMCB so we can
3086	 * safely return them on rdmsr. They will always be 0 until LBRV is
3087	 * implemented.
3088	 */
3089	case MSR_IA32_DEBUGCTLMSR:
3090		*data = svm->vmcb->save.dbgctl;
3091		break;
3092	case MSR_IA32_LASTBRANCHFROMIP:
3093		*data = svm->vmcb->save.br_from;
3094		break;
3095	case MSR_IA32_LASTBRANCHTOIP:
3096		*data = svm->vmcb->save.br_to;
3097		break;
3098	case MSR_IA32_LASTINTFROMIP:
3099		*data = svm->vmcb->save.last_excp_from;
3100		break;
3101	case MSR_IA32_LASTINTTOIP:
3102		*data = svm->vmcb->save.last_excp_to;
3103		break;
3104	case MSR_VM_HSAVE_PA:
3105		*data = svm->nested.hsave_msr;
3106		break;
3107	case MSR_VM_CR:
3108		*data = svm->nested.vm_cr_msr;
3109		break;
3110	case MSR_IA32_UCODE_REV:
3111		*data = 0x01000065;
3112		break;
3113	default:
3114		return kvm_get_msr_common(vcpu, ecx, data);
3115	}
3116	return 0;
3117}
3118
3119static int rdmsr_interception(struct vcpu_svm *svm)
3120{
3121	u32 ecx = svm->vcpu.arch.regs[VCPU_REGS_RCX];
3122	u64 data;
3123
3124	if (svm_get_msr(&svm->vcpu, ecx, &data)) {
3125		trace_kvm_msr_read_ex(ecx);
3126		kvm_inject_gp(&svm->vcpu, 0);
3127	} else {
3128		trace_kvm_msr_read(ecx, data);
3129
3130		svm->vcpu.arch.regs[VCPU_REGS_RAX] = data & 0xffffffff;
3131		svm->vcpu.arch.regs[VCPU_REGS_RDX] = data >> 32;
3132		svm->next_rip = kvm_rip_read(&svm->vcpu) + 2;
3133		skip_emulated_instruction(&svm->vcpu);
3134	}
3135	return 1;
3136}
3137
3138static int svm_set_vm_cr(struct kvm_vcpu *vcpu, u64 data)
3139{
3140	struct vcpu_svm *svm = to_svm(vcpu);
3141	int svm_dis, chg_mask;
3142
3143	if (data & ~SVM_VM_CR_VALID_MASK)
3144		return 1;
3145
3146	chg_mask = SVM_VM_CR_VALID_MASK;
3147
3148	if (svm->nested.vm_cr_msr & SVM_VM_CR_SVM_DIS_MASK)
3149		chg_mask &= ~(SVM_VM_CR_SVM_LOCK_MASK | SVM_VM_CR_SVM_DIS_MASK);
3150
3151	svm->nested.vm_cr_msr &= ~chg_mask;
3152	svm->nested.vm_cr_msr |= (data & chg_mask);
3153
3154	svm_dis = svm->nested.vm_cr_msr & SVM_VM_CR_SVM_DIS_MASK;
3155
3156	/* check for svm_disable while efer.svme is set */
3157	if (svm_dis && (vcpu->arch.efer & EFER_SVME))
3158		return 1;
3159
3160	return 0;
3161}
3162
3163static int svm_set_msr(struct kvm_vcpu *vcpu, struct msr_data *msr)
3164{
3165	struct vcpu_svm *svm = to_svm(vcpu);
3166
3167	u32 ecx = msr->index;
3168	u64 data = msr->data;
3169	switch (ecx) {
3170	case MSR_IA32_TSC:
3171		kvm_write_tsc(vcpu, msr);
3172		break;
3173	case MSR_STAR:
3174		svm->vmcb->save.star = data;
3175		break;
3176#ifdef CONFIG_X86_64
3177	case MSR_LSTAR:
3178		svm->vmcb->save.lstar = data;
3179		break;
3180	case MSR_CSTAR:
3181		svm->vmcb->save.cstar = data;
3182		break;
3183	case MSR_KERNEL_GS_BASE:
3184		svm->vmcb->save.kernel_gs_base = data;
3185		break;
3186	case MSR_SYSCALL_MASK:
3187		svm->vmcb->save.sfmask = data;
3188		break;
3189#endif
3190	case MSR_IA32_SYSENTER_CS:
3191		svm->vmcb->save.sysenter_cs = data;
3192		break;
3193	case MSR_IA32_SYSENTER_EIP:
3194		svm->sysenter_eip = data;
3195		svm->vmcb->save.sysenter_eip = data;
3196		break;
3197	case MSR_IA32_SYSENTER_ESP:
3198		svm->sysenter_esp = data;
3199		svm->vmcb->save.sysenter_esp = data;
3200		break;
3201	case MSR_IA32_DEBUGCTLMSR:
3202		if (!boot_cpu_has(X86_FEATURE_LBRV)) {
3203			vcpu_unimpl(vcpu, "%s: MSR_IA32_DEBUGCTL 0x%llx, nop\n",
3204				    __func__, data);
3205			break;
3206		}
3207		if (data & DEBUGCTL_RESERVED_BITS)
3208			return 1;
3209
3210		svm->vmcb->save.dbgctl = data;
3211		mark_dirty(svm->vmcb, VMCB_LBR);
3212		if (data & (1ULL<<0))
3213			svm_enable_lbrv(svm);
3214		else
3215			svm_disable_lbrv(svm);
3216		break;
3217	case MSR_VM_HSAVE_PA:
3218		svm->nested.hsave_msr = data;
3219		break;
3220	case MSR_VM_CR:
3221		return svm_set_vm_cr(vcpu, data);
3222	case MSR_VM_IGNNE:
3223		vcpu_unimpl(vcpu, "unimplemented wrmsr: 0x%x data 0x%llx\n", ecx, data);
3224		break;
3225	default:
3226		return kvm_set_msr_common(vcpu, msr);
3227	}
3228	return 0;
3229}
3230
3231static int wrmsr_interception(struct vcpu_svm *svm)
3232{
3233	struct msr_data msr;
3234	u32 ecx = svm->vcpu.arch.regs[VCPU_REGS_RCX];
3235	u64 data = (svm->vcpu.arch.regs[VCPU_REGS_RAX] & -1u)
3236		| ((u64)(svm->vcpu.arch.regs[VCPU_REGS_RDX] & -1u) << 32);
3237
3238	msr.data = data;
3239	msr.index = ecx;
3240	msr.host_initiated = false;
3241
3242	svm->next_rip = kvm_rip_read(&svm->vcpu) + 2;
3243	if (svm_set_msr(&svm->vcpu, &msr)) {
3244		trace_kvm_msr_write_ex(ecx, data);
3245		kvm_inject_gp(&svm->vcpu, 0);
3246	} else {
3247		trace_kvm_msr_write(ecx, data);
3248		skip_emulated_instruction(&svm->vcpu);
3249	}
3250	return 1;
3251}
3252
3253static int msr_interception(struct vcpu_svm *svm)
3254{
3255	if (svm->vmcb->control.exit_info_1)
3256		return wrmsr_interception(svm);
3257	else
3258		return rdmsr_interception(svm);
3259}
3260
3261static int interrupt_window_interception(struct vcpu_svm *svm)
3262{
3263	struct kvm_run *kvm_run = svm->vcpu.run;
3264
3265	kvm_make_request(KVM_REQ_EVENT, &svm->vcpu);
3266	svm_clear_vintr(svm);
3267	svm->vmcb->control.int_ctl &= ~V_IRQ_MASK;
3268	mark_dirty(svm->vmcb, VMCB_INTR);
3269	++svm->vcpu.stat.irq_window_exits;
3270	/*
3271	 * If the user space waits to inject interrupts, exit as soon as
3272	 * possible
3273	 */
3274	if (!irqchip_in_kernel(svm->vcpu.kvm) &&
3275	    kvm_run->request_interrupt_window &&
3276	    !kvm_cpu_has_interrupt(&svm->vcpu)) {
 
3277		kvm_run->exit_reason = KVM_EXIT_IRQ_WINDOW_OPEN;
3278		return 0;
3279	}
3280
3281	return 1;
3282}
3283
3284static int pause_interception(struct vcpu_svm *svm)
3285{
3286	kvm_vcpu_on_spin(&(svm->vcpu));
3287	return 1;
3288}
3289
3290static int (*const svm_exit_handlers[])(struct vcpu_svm *svm) = {
3291	[SVM_EXIT_READ_CR0]			= cr_interception,
3292	[SVM_EXIT_READ_CR3]			= cr_interception,
3293	[SVM_EXIT_READ_CR4]			= cr_interception,
3294	[SVM_EXIT_READ_CR8]			= cr_interception,
3295	[SVM_EXIT_CR0_SEL_WRITE]		= emulate_on_interception,
3296	[SVM_EXIT_WRITE_CR0]			= cr_interception,
3297	[SVM_EXIT_WRITE_CR3]			= cr_interception,
3298	[SVM_EXIT_WRITE_CR4]			= cr_interception,
3299	[SVM_EXIT_WRITE_CR8]			= cr8_write_interception,
3300	[SVM_EXIT_READ_DR0]			= dr_interception,
3301	[SVM_EXIT_READ_DR1]			= dr_interception,
3302	[SVM_EXIT_READ_DR2]			= dr_interception,
3303	[SVM_EXIT_READ_DR3]			= dr_interception,
3304	[SVM_EXIT_READ_DR4]			= dr_interception,
3305	[SVM_EXIT_READ_DR5]			= dr_interception,
3306	[SVM_EXIT_READ_DR6]			= dr_interception,
3307	[SVM_EXIT_READ_DR7]			= dr_interception,
3308	[SVM_EXIT_WRITE_DR0]			= dr_interception,
3309	[SVM_EXIT_WRITE_DR1]			= dr_interception,
3310	[SVM_EXIT_WRITE_DR2]			= dr_interception,
3311	[SVM_EXIT_WRITE_DR3]			= dr_interception,
3312	[SVM_EXIT_WRITE_DR4]			= dr_interception,
3313	[SVM_EXIT_WRITE_DR5]			= dr_interception,
3314	[SVM_EXIT_WRITE_DR6]			= dr_interception,
3315	[SVM_EXIT_WRITE_DR7]			= dr_interception,
3316	[SVM_EXIT_EXCP_BASE + DB_VECTOR]	= db_interception,
3317	[SVM_EXIT_EXCP_BASE + BP_VECTOR]	= bp_interception,
3318	[SVM_EXIT_EXCP_BASE + UD_VECTOR]	= ud_interception,
3319	[SVM_EXIT_EXCP_BASE + PF_VECTOR]	= pf_interception,
3320	[SVM_EXIT_EXCP_BASE + NM_VECTOR]	= nm_interception,
3321	[SVM_EXIT_EXCP_BASE + MC_VECTOR]	= mc_interception,
3322	[SVM_EXIT_INTR]				= intr_interception,
3323	[SVM_EXIT_NMI]				= nmi_interception,
3324	[SVM_EXIT_SMI]				= nop_on_interception,
3325	[SVM_EXIT_INIT]				= nop_on_interception,
3326	[SVM_EXIT_VINTR]			= interrupt_window_interception,
3327	[SVM_EXIT_RDPMC]			= rdpmc_interception,
3328	[SVM_EXIT_CPUID]			= cpuid_interception,
3329	[SVM_EXIT_IRET]                         = iret_interception,
3330	[SVM_EXIT_INVD]                         = emulate_on_interception,
3331	[SVM_EXIT_PAUSE]			= pause_interception,
3332	[SVM_EXIT_HLT]				= halt_interception,
3333	[SVM_EXIT_INVLPG]			= invlpg_interception,
3334	[SVM_EXIT_INVLPGA]			= invlpga_interception,
3335	[SVM_EXIT_IOIO]				= io_interception,
3336	[SVM_EXIT_MSR]				= msr_interception,
3337	[SVM_EXIT_TASK_SWITCH]			= task_switch_interception,
3338	[SVM_EXIT_SHUTDOWN]			= shutdown_interception,
3339	[SVM_EXIT_VMRUN]			= vmrun_interception,
3340	[SVM_EXIT_VMMCALL]			= vmmcall_interception,
3341	[SVM_EXIT_VMLOAD]			= vmload_interception,
3342	[SVM_EXIT_VMSAVE]			= vmsave_interception,
3343	[SVM_EXIT_STGI]				= stgi_interception,
3344	[SVM_EXIT_CLGI]				= clgi_interception,
3345	[SVM_EXIT_SKINIT]			= skinit_interception,
3346	[SVM_EXIT_WBINVD]                       = emulate_on_interception,
3347	[SVM_EXIT_MONITOR]			= invalid_op_interception,
3348	[SVM_EXIT_MWAIT]			= invalid_op_interception,
3349	[SVM_EXIT_XSETBV]			= xsetbv_interception,
3350	[SVM_EXIT_NPF]				= pf_interception,
3351};
3352
3353static void dump_vmcb(struct kvm_vcpu *vcpu)
3354{
3355	struct vcpu_svm *svm = to_svm(vcpu);
3356	struct vmcb_control_area *control = &svm->vmcb->control;
3357	struct vmcb_save_area *save = &svm->vmcb->save;
3358
3359	pr_err("VMCB Control Area:\n");
3360	pr_err("%-20s%04x\n", "cr_read:", control->intercept_cr & 0xffff);
3361	pr_err("%-20s%04x\n", "cr_write:", control->intercept_cr >> 16);
3362	pr_err("%-20s%04x\n", "dr_read:", control->intercept_dr & 0xffff);
3363	pr_err("%-20s%04x\n", "dr_write:", control->intercept_dr >> 16);
3364	pr_err("%-20s%08x\n", "exceptions:", control->intercept_exceptions);
3365	pr_err("%-20s%016llx\n", "intercepts:", control->intercept);
3366	pr_err("%-20s%d\n", "pause filter count:", control->pause_filter_count);
3367	pr_err("%-20s%016llx\n", "iopm_base_pa:", control->iopm_base_pa);
3368	pr_err("%-20s%016llx\n", "msrpm_base_pa:", control->msrpm_base_pa);
3369	pr_err("%-20s%016llx\n", "tsc_offset:", control->tsc_offset);
3370	pr_err("%-20s%d\n", "asid:", control->asid);
3371	pr_err("%-20s%d\n", "tlb_ctl:", control->tlb_ctl);
3372	pr_err("%-20s%08x\n", "int_ctl:", control->int_ctl);
3373	pr_err("%-20s%08x\n", "int_vector:", control->int_vector);
3374	pr_err("%-20s%08x\n", "int_state:", control->int_state);
3375	pr_err("%-20s%08x\n", "exit_code:", control->exit_code);
3376	pr_err("%-20s%016llx\n", "exit_info1:", control->exit_info_1);
3377	pr_err("%-20s%016llx\n", "exit_info2:", control->exit_info_2);
3378	pr_err("%-20s%08x\n", "exit_int_info:", control->exit_int_info);
3379	pr_err("%-20s%08x\n", "exit_int_info_err:", control->exit_int_info_err);
3380	pr_err("%-20s%lld\n", "nested_ctl:", control->nested_ctl);
3381	pr_err("%-20s%016llx\n", "nested_cr3:", control->nested_cr3);
3382	pr_err("%-20s%08x\n", "event_inj:", control->event_inj);
3383	pr_err("%-20s%08x\n", "event_inj_err:", control->event_inj_err);
3384	pr_err("%-20s%lld\n", "lbr_ctl:", control->lbr_ctl);
3385	pr_err("%-20s%016llx\n", "next_rip:", control->next_rip);
3386	pr_err("VMCB State Save Area:\n");
3387	pr_err("%-5s s: %04x a: %04x l: %08x b: %016llx\n",
3388	       "es:",
3389	       save->es.selector, save->es.attrib,
3390	       save->es.limit, save->es.base);
3391	pr_err("%-5s s: %04x a: %04x l: %08x b: %016llx\n",
3392	       "cs:",
3393	       save->cs.selector, save->cs.attrib,
3394	       save->cs.limit, save->cs.base);
3395	pr_err("%-5s s: %04x a: %04x l: %08x b: %016llx\n",
3396	       "ss:",
3397	       save->ss.selector, save->ss.attrib,
3398	       save->ss.limit, save->ss.base);
3399	pr_err("%-5s s: %04x a: %04x l: %08x b: %016llx\n",
3400	       "ds:",
3401	       save->ds.selector, save->ds.attrib,
3402	       save->ds.limit, save->ds.base);
3403	pr_err("%-5s s: %04x a: %04x l: %08x b: %016llx\n",
3404	       "fs:",
3405	       save->fs.selector, save->fs.attrib,
3406	       save->fs.limit, save->fs.base);
3407	pr_err("%-5s s: %04x a: %04x l: %08x b: %016llx\n",
3408	       "gs:",
3409	       save->gs.selector, save->gs.attrib,
3410	       save->gs.limit, save->gs.base);
3411	pr_err("%-5s s: %04x a: %04x l: %08x b: %016llx\n",
3412	       "gdtr:",
3413	       save->gdtr.selector, save->gdtr.attrib,
3414	       save->gdtr.limit, save->gdtr.base);
3415	pr_err("%-5s s: %04x a: %04x l: %08x b: %016llx\n",
3416	       "ldtr:",
3417	       save->ldtr.selector, save->ldtr.attrib,
3418	       save->ldtr.limit, save->ldtr.base);
3419	pr_err("%-5s s: %04x a: %04x l: %08x b: %016llx\n",
3420	       "idtr:",
3421	       save->idtr.selector, save->idtr.attrib,
3422	       save->idtr.limit, save->idtr.base);
3423	pr_err("%-5s s: %04x a: %04x l: %08x b: %016llx\n",
3424	       "tr:",
3425	       save->tr.selector, save->tr.attrib,
3426	       save->tr.limit, save->tr.base);
3427	pr_err("cpl:            %d                efer:         %016llx\n",
3428		save->cpl, save->efer);
3429	pr_err("%-15s %016llx %-13s %016llx\n",
3430	       "cr0:", save->cr0, "cr2:", save->cr2);
3431	pr_err("%-15s %016llx %-13s %016llx\n",
3432	       "cr3:", save->cr3, "cr4:", save->cr4);
3433	pr_err("%-15s %016llx %-13s %016llx\n",
3434	       "dr6:", save->dr6, "dr7:", save->dr7);
3435	pr_err("%-15s %016llx %-13s %016llx\n",
3436	       "rip:", save->rip, "rflags:", save->rflags);
3437	pr_err("%-15s %016llx %-13s %016llx\n",
3438	       "rsp:", save->rsp, "rax:", save->rax);
3439	pr_err("%-15s %016llx %-13s %016llx\n",
3440	       "star:", save->star, "lstar:", save->lstar);
3441	pr_err("%-15s %016llx %-13s %016llx\n",
3442	       "cstar:", save->cstar, "sfmask:", save->sfmask);
3443	pr_err("%-15s %016llx %-13s %016llx\n",
3444	       "kernel_gs_base:", save->kernel_gs_base,
3445	       "sysenter_cs:", save->sysenter_cs);
3446	pr_err("%-15s %016llx %-13s %016llx\n",
3447	       "sysenter_esp:", save->sysenter_esp,
3448	       "sysenter_eip:", save->sysenter_eip);
3449	pr_err("%-15s %016llx %-13s %016llx\n",
3450	       "gpat:", save->g_pat, "dbgctl:", save->dbgctl);
3451	pr_err("%-15s %016llx %-13s %016llx\n",
3452	       "br_from:", save->br_from, "br_to:", save->br_to);
3453	pr_err("%-15s %016llx %-13s %016llx\n",
3454	       "excp_from:", save->last_excp_from,
3455	       "excp_to:", save->last_excp_to);
3456}
3457
3458static void svm_get_exit_info(struct kvm_vcpu *vcpu, u64 *info1, u64 *info2)
3459{
3460	struct vmcb_control_area *control = &to_svm(vcpu)->vmcb->control;
3461
3462	*info1 = control->exit_info_1;
3463	*info2 = control->exit_info_2;
3464}
3465
3466static int handle_exit(struct kvm_vcpu *vcpu)
3467{
3468	struct vcpu_svm *svm = to_svm(vcpu);
3469	struct kvm_run *kvm_run = vcpu->run;
3470	u32 exit_code = svm->vmcb->control.exit_code;
3471
 
 
3472	if (!is_cr_intercept(svm, INTERCEPT_CR0_WRITE))
3473		vcpu->arch.cr0 = svm->vmcb->save.cr0;
3474	if (npt_enabled)
3475		vcpu->arch.cr3 = svm->vmcb->save.cr3;
3476
3477	if (unlikely(svm->nested.exit_required)) {
3478		nested_svm_vmexit(svm);
3479		svm->nested.exit_required = false;
3480
3481		return 1;
3482	}
3483
3484	if (is_guest_mode(vcpu)) {
3485		int vmexit;
3486
3487		trace_kvm_nested_vmexit(svm->vmcb->save.rip, exit_code,
3488					svm->vmcb->control.exit_info_1,
3489					svm->vmcb->control.exit_info_2,
3490					svm->vmcb->control.exit_int_info,
3491					svm->vmcb->control.exit_int_info_err,
3492					KVM_ISA_SVM);
3493
3494		vmexit = nested_svm_exit_special(svm);
3495
3496		if (vmexit == NESTED_EXIT_CONTINUE)
3497			vmexit = nested_svm_exit_handled(svm);
3498
3499		if (vmexit == NESTED_EXIT_DONE)
3500			return 1;
3501	}
3502
3503	svm_complete_interrupts(svm);
3504
3505	if (svm->vmcb->control.exit_code == SVM_EXIT_ERR) {
3506		kvm_run->exit_reason = KVM_EXIT_FAIL_ENTRY;
3507		kvm_run->fail_entry.hardware_entry_failure_reason
3508			= svm->vmcb->control.exit_code;
3509		pr_err("KVM: FAILED VMRUN WITH VMCB:\n");
3510		dump_vmcb(vcpu);
3511		return 0;
3512	}
3513
3514	if (is_external_interrupt(svm->vmcb->control.exit_int_info) &&
3515	    exit_code != SVM_EXIT_EXCP_BASE + PF_VECTOR &&
3516	    exit_code != SVM_EXIT_NPF && exit_code != SVM_EXIT_TASK_SWITCH &&
3517	    exit_code != SVM_EXIT_INTR && exit_code != SVM_EXIT_NMI)
3518		printk(KERN_ERR "%s: unexpected exit_int_info 0x%x "
3519		       "exit_code 0x%x\n",
3520		       __func__, svm->vmcb->control.exit_int_info,
3521		       exit_code);
3522
3523	if (exit_code >= ARRAY_SIZE(svm_exit_handlers)
3524	    || !svm_exit_handlers[exit_code]) {
3525		kvm_run->exit_reason = KVM_EXIT_UNKNOWN;
3526		kvm_run->hw.hardware_exit_reason = exit_code;
3527		return 0;
3528	}
3529
3530	return svm_exit_handlers[exit_code](svm);
3531}
3532
3533static void reload_tss(struct kvm_vcpu *vcpu)
3534{
3535	int cpu = raw_smp_processor_id();
3536
3537	struct svm_cpu_data *sd = per_cpu(svm_data, cpu);
3538	sd->tss_desc->type = 9; /* available 32/64-bit TSS */
3539	load_TR_desc();
3540}
3541
3542static void pre_svm_run(struct vcpu_svm *svm)
3543{
3544	int cpu = raw_smp_processor_id();
3545
3546	struct svm_cpu_data *sd = per_cpu(svm_data, cpu);
3547
3548	/* FIXME: handle wraparound of asid_generation */
3549	if (svm->asid_generation != sd->asid_generation)
3550		new_asid(svm, sd);
3551}
3552
3553static void svm_inject_nmi(struct kvm_vcpu *vcpu)
3554{
3555	struct vcpu_svm *svm = to_svm(vcpu);
3556
3557	svm->vmcb->control.event_inj = SVM_EVTINJ_VALID | SVM_EVTINJ_TYPE_NMI;
3558	vcpu->arch.hflags |= HF_NMI_MASK;
3559	set_intercept(svm, INTERCEPT_IRET);
3560	++vcpu->stat.nmi_injections;
3561}
3562
3563static inline void svm_inject_irq(struct vcpu_svm *svm, int irq)
3564{
3565	struct vmcb_control_area *control;
3566
3567	control = &svm->vmcb->control;
3568	control->int_vector = irq;
3569	control->int_ctl &= ~V_INTR_PRIO_MASK;
3570	control->int_ctl |= V_IRQ_MASK |
3571		((/*control->int_vector >> 4*/ 0xf) << V_INTR_PRIO_SHIFT);
3572	mark_dirty(svm->vmcb, VMCB_INTR);
3573}
3574
3575static void svm_set_irq(struct kvm_vcpu *vcpu)
3576{
3577	struct vcpu_svm *svm = to_svm(vcpu);
3578
3579	BUG_ON(!(gif_set(svm)));
3580
3581	trace_kvm_inj_virq(vcpu->arch.interrupt.nr);
3582	++vcpu->stat.irq_injections;
3583
3584	svm->vmcb->control.event_inj = vcpu->arch.interrupt.nr |
3585		SVM_EVTINJ_VALID | SVM_EVTINJ_TYPE_INTR;
3586}
3587
3588static void update_cr8_intercept(struct kvm_vcpu *vcpu, int tpr, int irr)
3589{
3590	struct vcpu_svm *svm = to_svm(vcpu);
3591
3592	if (is_guest_mode(vcpu) && (vcpu->arch.hflags & HF_VINTR_MASK))
3593		return;
3594
3595	clr_cr_intercept(svm, INTERCEPT_CR8_WRITE);
3596
3597	if (irr == -1)
3598		return;
3599
3600	if (tpr >= irr)
3601		set_cr_intercept(svm, INTERCEPT_CR8_WRITE);
3602}
3603
3604static void svm_set_virtual_x2apic_mode(struct kvm_vcpu *vcpu, bool set)
3605{
3606	return;
3607}
3608
3609static int svm_vm_has_apicv(struct kvm *kvm)
3610{
3611	return 0;
3612}
3613
3614static void svm_load_eoi_exitmap(struct kvm_vcpu *vcpu, u64 *eoi_exit_bitmap)
3615{
3616	return;
3617}
3618
3619static void svm_hwapic_isr_update(struct kvm *kvm, int isr)
3620{
3621	return;
3622}
3623
3624static void svm_sync_pir_to_irr(struct kvm_vcpu *vcpu)
3625{
3626	return;
3627}
3628
3629static int svm_nmi_allowed(struct kvm_vcpu *vcpu)
3630{
3631	struct vcpu_svm *svm = to_svm(vcpu);
3632	struct vmcb *vmcb = svm->vmcb;
3633	int ret;
3634	ret = !(vmcb->control.int_state & SVM_INTERRUPT_SHADOW_MASK) &&
3635	      !(svm->vcpu.arch.hflags & HF_NMI_MASK);
3636	ret = ret && gif_set(svm) && nested_svm_nmi(svm);
3637
3638	return ret;
3639}
3640
3641static bool svm_get_nmi_mask(struct kvm_vcpu *vcpu)
3642{
3643	struct vcpu_svm *svm = to_svm(vcpu);
3644
3645	return !!(svm->vcpu.arch.hflags & HF_NMI_MASK);
3646}
3647
3648static void svm_set_nmi_mask(struct kvm_vcpu *vcpu, bool masked)
3649{
3650	struct vcpu_svm *svm = to_svm(vcpu);
3651
3652	if (masked) {
3653		svm->vcpu.arch.hflags |= HF_NMI_MASK;
3654		set_intercept(svm, INTERCEPT_IRET);
3655	} else {
3656		svm->vcpu.arch.hflags &= ~HF_NMI_MASK;
3657		clr_intercept(svm, INTERCEPT_IRET);
3658	}
3659}
3660
3661static int svm_interrupt_allowed(struct kvm_vcpu *vcpu)
3662{
3663	struct vcpu_svm *svm = to_svm(vcpu);
3664	struct vmcb *vmcb = svm->vmcb;
3665	int ret;
3666
3667	if (!gif_set(svm) ||
3668	     (vmcb->control.int_state & SVM_INTERRUPT_SHADOW_MASK))
3669		return 0;
3670
3671	ret = !!(kvm_get_rflags(vcpu) & X86_EFLAGS_IF);
3672
3673	if (is_guest_mode(vcpu))
3674		return ret && !(svm->vcpu.arch.hflags & HF_VINTR_MASK);
3675
3676	return ret;
3677}
3678
3679static void enable_irq_window(struct kvm_vcpu *vcpu)
3680{
3681	struct vcpu_svm *svm = to_svm(vcpu);
3682
3683	/*
3684	 * In case GIF=0 we can't rely on the CPU to tell us when GIF becomes
3685	 * 1, because that's a separate STGI/VMRUN intercept.  The next time we
3686	 * get that intercept, this function will be called again though and
3687	 * we'll get the vintr intercept.
3688	 */
3689	if (gif_set(svm) && nested_svm_intr(svm)) {
3690		svm_set_vintr(svm);
3691		svm_inject_irq(svm, 0x0);
3692	}
3693}
3694
3695static void enable_nmi_window(struct kvm_vcpu *vcpu)
3696{
3697	struct vcpu_svm *svm = to_svm(vcpu);
3698
3699	if ((svm->vcpu.arch.hflags & (HF_NMI_MASK | HF_IRET_MASK))
3700	    == HF_NMI_MASK)
3701		return; /* IRET will cause a vm exit */
3702
3703	/*
3704	 * Something prevents NMI from been injected. Single step over possible
3705	 * problem (IRET or exception injection or interrupt shadow)
3706	 */
3707	svm->nmi_singlestep = true;
3708	svm->vmcb->save.rflags |= (X86_EFLAGS_TF | X86_EFLAGS_RF);
3709	update_db_bp_intercept(vcpu);
3710}
3711
3712static int svm_set_tss_addr(struct kvm *kvm, unsigned int addr)
3713{
3714	return 0;
3715}
3716
3717static void svm_flush_tlb(struct kvm_vcpu *vcpu)
3718{
3719	struct vcpu_svm *svm = to_svm(vcpu);
3720
3721	if (static_cpu_has(X86_FEATURE_FLUSHBYASID))
3722		svm->vmcb->control.tlb_ctl = TLB_CONTROL_FLUSH_ASID;
3723	else
3724		svm->asid_generation--;
3725}
3726
3727static void svm_prepare_guest_switch(struct kvm_vcpu *vcpu)
3728{
3729}
3730
3731static inline void sync_cr8_to_lapic(struct kvm_vcpu *vcpu)
3732{
3733	struct vcpu_svm *svm = to_svm(vcpu);
3734
3735	if (is_guest_mode(vcpu) && (vcpu->arch.hflags & HF_VINTR_MASK))
3736		return;
3737
3738	if (!is_cr_intercept(svm, INTERCEPT_CR8_WRITE)) {
3739		int cr8 = svm->vmcb->control.int_ctl & V_TPR_MASK;
3740		kvm_set_cr8(vcpu, cr8);
3741	}
3742}
3743
3744static inline void sync_lapic_to_cr8(struct kvm_vcpu *vcpu)
3745{
3746	struct vcpu_svm *svm = to_svm(vcpu);
3747	u64 cr8;
3748
3749	if (is_guest_mode(vcpu) && (vcpu->arch.hflags & HF_VINTR_MASK))
3750		return;
3751
3752	cr8 = kvm_get_cr8(vcpu);
3753	svm->vmcb->control.int_ctl &= ~V_TPR_MASK;
3754	svm->vmcb->control.int_ctl |= cr8 & V_TPR_MASK;
3755}
3756
3757static void svm_complete_interrupts(struct vcpu_svm *svm)
3758{
3759	u8 vector;
3760	int type;
3761	u32 exitintinfo = svm->vmcb->control.exit_int_info;
3762	unsigned int3_injected = svm->int3_injected;
3763
3764	svm->int3_injected = 0;
3765
3766	/*
3767	 * If we've made progress since setting HF_IRET_MASK, we've
3768	 * executed an IRET and can allow NMI injection.
3769	 */
3770	if ((svm->vcpu.arch.hflags & HF_IRET_MASK)
3771	    && kvm_rip_read(&svm->vcpu) != svm->nmi_iret_rip) {
3772		svm->vcpu.arch.hflags &= ~(HF_NMI_MASK | HF_IRET_MASK);
3773		kvm_make_request(KVM_REQ_EVENT, &svm->vcpu);
3774	}
3775
3776	svm->vcpu.arch.nmi_injected = false;
3777	kvm_clear_exception_queue(&svm->vcpu);
3778	kvm_clear_interrupt_queue(&svm->vcpu);
3779
3780	if (!(exitintinfo & SVM_EXITINTINFO_VALID))
3781		return;
3782
3783	kvm_make_request(KVM_REQ_EVENT, &svm->vcpu);
3784
3785	vector = exitintinfo & SVM_EXITINTINFO_VEC_MASK;
3786	type = exitintinfo & SVM_EXITINTINFO_TYPE_MASK;
3787
3788	switch (type) {
3789	case SVM_EXITINTINFO_TYPE_NMI:
3790		svm->vcpu.arch.nmi_injected = true;
3791		break;
3792	case SVM_EXITINTINFO_TYPE_EXEPT:
3793		/*
3794		 * In case of software exceptions, do not reinject the vector,
3795		 * but re-execute the instruction instead. Rewind RIP first
3796		 * if we emulated INT3 before.
3797		 */
3798		if (kvm_exception_is_soft(vector)) {
3799			if (vector == BP_VECTOR && int3_injected &&
3800			    kvm_is_linear_rip(&svm->vcpu, svm->int3_rip))
3801				kvm_rip_write(&svm->vcpu,
3802					      kvm_rip_read(&svm->vcpu) -
3803					      int3_injected);
3804			break;
3805		}
3806		if (exitintinfo & SVM_EXITINTINFO_VALID_ERR) {
3807			u32 err = svm->vmcb->control.exit_int_info_err;
3808			kvm_requeue_exception_e(&svm->vcpu, vector, err);
3809
3810		} else
3811			kvm_requeue_exception(&svm->vcpu, vector);
3812		break;
3813	case SVM_EXITINTINFO_TYPE_INTR:
3814		kvm_queue_interrupt(&svm->vcpu, vector, false);
3815		break;
3816	default:
3817		break;
3818	}
3819}
3820
3821static void svm_cancel_injection(struct kvm_vcpu *vcpu)
3822{
3823	struct vcpu_svm *svm = to_svm(vcpu);
3824	struct vmcb_control_area *control = &svm->vmcb->control;
3825
3826	control->exit_int_info = control->event_inj;
3827	control->exit_int_info_err = control->event_inj_err;
3828	control->event_inj = 0;
3829	svm_complete_interrupts(svm);
3830}
3831
 
 
 
 
 
 
3832static void svm_vcpu_run(struct kvm_vcpu *vcpu)
3833{
3834	struct vcpu_svm *svm = to_svm(vcpu);
3835
3836	svm->vmcb->save.rax = vcpu->arch.regs[VCPU_REGS_RAX];
3837	svm->vmcb->save.rsp = vcpu->arch.regs[VCPU_REGS_RSP];
3838	svm->vmcb->save.rip = vcpu->arch.regs[VCPU_REGS_RIP];
3839
3840	/*
3841	 * A vmexit emulation is required before the vcpu can be executed
3842	 * again.
3843	 */
3844	if (unlikely(svm->nested.exit_required))
3845		return;
3846
3847	pre_svm_run(svm);
3848
3849	sync_lapic_to_cr8(vcpu);
3850
3851	svm->vmcb->save.cr2 = vcpu->arch.cr2;
3852
3853	clgi();
3854
3855	local_irq_enable();
3856
3857	asm volatile (
3858		"push %%" _ASM_BP "; \n\t"
3859		"mov %c[rbx](%[svm]), %%" _ASM_BX " \n\t"
3860		"mov %c[rcx](%[svm]), %%" _ASM_CX " \n\t"
3861		"mov %c[rdx](%[svm]), %%" _ASM_DX " \n\t"
3862		"mov %c[rsi](%[svm]), %%" _ASM_SI " \n\t"
3863		"mov %c[rdi](%[svm]), %%" _ASM_DI " \n\t"
3864		"mov %c[rbp](%[svm]), %%" _ASM_BP " \n\t"
3865#ifdef CONFIG_X86_64
3866		"mov %c[r8](%[svm]),  %%r8  \n\t"
3867		"mov %c[r9](%[svm]),  %%r9  \n\t"
3868		"mov %c[r10](%[svm]), %%r10 \n\t"
3869		"mov %c[r11](%[svm]), %%r11 \n\t"
3870		"mov %c[r12](%[svm]), %%r12 \n\t"
3871		"mov %c[r13](%[svm]), %%r13 \n\t"
3872		"mov %c[r14](%[svm]), %%r14 \n\t"
3873		"mov %c[r15](%[svm]), %%r15 \n\t"
3874#endif
3875
3876		/* Enter guest mode */
3877		"push %%" _ASM_AX " \n\t"
3878		"mov %c[vmcb](%[svm]), %%" _ASM_AX " \n\t"
3879		__ex(SVM_VMLOAD) "\n\t"
3880		__ex(SVM_VMRUN) "\n\t"
3881		__ex(SVM_VMSAVE) "\n\t"
3882		"pop %%" _ASM_AX " \n\t"
3883
3884		/* Save guest registers, load host registers */
3885		"mov %%" _ASM_BX ", %c[rbx](%[svm]) \n\t"
3886		"mov %%" _ASM_CX ", %c[rcx](%[svm]) \n\t"
3887		"mov %%" _ASM_DX ", %c[rdx](%[svm]) \n\t"
3888		"mov %%" _ASM_SI ", %c[rsi](%[svm]) \n\t"
3889		"mov %%" _ASM_DI ", %c[rdi](%[svm]) \n\t"
3890		"mov %%" _ASM_BP ", %c[rbp](%[svm]) \n\t"
3891#ifdef CONFIG_X86_64
3892		"mov %%r8,  %c[r8](%[svm]) \n\t"
3893		"mov %%r9,  %c[r9](%[svm]) \n\t"
3894		"mov %%r10, %c[r10](%[svm]) \n\t"
3895		"mov %%r11, %c[r11](%[svm]) \n\t"
3896		"mov %%r12, %c[r12](%[svm]) \n\t"
3897		"mov %%r13, %c[r13](%[svm]) \n\t"
3898		"mov %%r14, %c[r14](%[svm]) \n\t"
3899		"mov %%r15, %c[r15](%[svm]) \n\t"
3900#endif
3901		"pop %%" _ASM_BP
3902		:
3903		: [svm]"a"(svm),
3904		  [vmcb]"i"(offsetof(struct vcpu_svm, vmcb_pa)),
3905		  [rbx]"i"(offsetof(struct vcpu_svm, vcpu.arch.regs[VCPU_REGS_RBX])),
3906		  [rcx]"i"(offsetof(struct vcpu_svm, vcpu.arch.regs[VCPU_REGS_RCX])),
3907		  [rdx]"i"(offsetof(struct vcpu_svm, vcpu.arch.regs[VCPU_REGS_RDX])),
3908		  [rsi]"i"(offsetof(struct vcpu_svm, vcpu.arch.regs[VCPU_REGS_RSI])),
3909		  [rdi]"i"(offsetof(struct vcpu_svm, vcpu.arch.regs[VCPU_REGS_RDI])),
3910		  [rbp]"i"(offsetof(struct vcpu_svm, vcpu.arch.regs[VCPU_REGS_RBP]))
3911#ifdef CONFIG_X86_64
3912		  , [r8]"i"(offsetof(struct vcpu_svm, vcpu.arch.regs[VCPU_REGS_R8])),
3913		  [r9]"i"(offsetof(struct vcpu_svm, vcpu.arch.regs[VCPU_REGS_R9])),
3914		  [r10]"i"(offsetof(struct vcpu_svm, vcpu.arch.regs[VCPU_REGS_R10])),
3915		  [r11]"i"(offsetof(struct vcpu_svm, vcpu.arch.regs[VCPU_REGS_R11])),
3916		  [r12]"i"(offsetof(struct vcpu_svm, vcpu.arch.regs[VCPU_REGS_R12])),
3917		  [r13]"i"(offsetof(struct vcpu_svm, vcpu.arch.regs[VCPU_REGS_R13])),
3918		  [r14]"i"(offsetof(struct vcpu_svm, vcpu.arch.regs[VCPU_REGS_R14])),
3919		  [r15]"i"(offsetof(struct vcpu_svm, vcpu.arch.regs[VCPU_REGS_R15]))
3920#endif
3921		: "cc", "memory"
 
3922#ifdef CONFIG_X86_64
3923		, "rbx", "rcx", "rdx", "rsi", "rdi"
3924		, "r8", "r9", "r10", "r11" , "r12", "r13", "r14", "r15"
3925#else
3926		, "ebx", "ecx", "edx", "esi", "edi"
3927#endif
3928		);
3929
3930#ifdef CONFIG_X86_64
3931	wrmsrl(MSR_GS_BASE, svm->host.gs_base);
3932#else
3933	loadsegment(fs, svm->host.fs);
3934#ifndef CONFIG_X86_32_LAZY_GS
3935	loadsegment(gs, svm->host.gs);
3936#endif
3937#endif
3938
3939	reload_tss(vcpu);
3940
3941	local_irq_disable();
3942
3943	vcpu->arch.cr2 = svm->vmcb->save.cr2;
3944	vcpu->arch.regs[VCPU_REGS_RAX] = svm->vmcb->save.rax;
3945	vcpu->arch.regs[VCPU_REGS_RSP] = svm->vmcb->save.rsp;
3946	vcpu->arch.regs[VCPU_REGS_RIP] = svm->vmcb->save.rip;
3947
3948	trace_kvm_exit(svm->vmcb->control.exit_code, vcpu, KVM_ISA_SVM);
3949
3950	if (unlikely(svm->vmcb->control.exit_code == SVM_EXIT_NMI))
3951		kvm_before_handle_nmi(&svm->vcpu);
3952
3953	stgi();
3954
3955	/* Any pending NMI will happen here */
3956
3957	if (unlikely(svm->vmcb->control.exit_code == SVM_EXIT_NMI))
3958		kvm_after_handle_nmi(&svm->vcpu);
3959
3960	sync_cr8_to_lapic(vcpu);
3961
3962	svm->next_rip = 0;
3963
3964	svm->vmcb->control.tlb_ctl = TLB_CONTROL_DO_NOTHING;
3965
3966	/* if exit due to PF check for async PF */
3967	if (svm->vmcb->control.exit_code == SVM_EXIT_EXCP_BASE + PF_VECTOR)
3968		svm->apf_reason = kvm_read_and_reset_pf_reason();
3969
3970	if (npt_enabled) {
3971		vcpu->arch.regs_avail &= ~(1 << VCPU_EXREG_PDPTR);
3972		vcpu->arch.regs_dirty &= ~(1 << VCPU_EXREG_PDPTR);
3973	}
3974
3975	/*
3976	 * We need to handle MC intercepts here before the vcpu has a chance to
3977	 * change the physical cpu
3978	 */
3979	if (unlikely(svm->vmcb->control.exit_code ==
3980		     SVM_EXIT_EXCP_BASE + MC_VECTOR))
3981		svm_handle_mce(svm);
3982
3983	mark_all_clean(svm->vmcb);
3984}
3985
 
 
3986static void svm_set_cr3(struct kvm_vcpu *vcpu, unsigned long root)
3987{
3988	struct vcpu_svm *svm = to_svm(vcpu);
3989
3990	svm->vmcb->save.cr3 = root;
3991	mark_dirty(svm->vmcb, VMCB_CR);
3992	svm_flush_tlb(vcpu);
3993}
3994
3995static void set_tdp_cr3(struct kvm_vcpu *vcpu, unsigned long root)
3996{
3997	struct vcpu_svm *svm = to_svm(vcpu);
3998
3999	svm->vmcb->control.nested_cr3 = root;
4000	mark_dirty(svm->vmcb, VMCB_NPT);
4001
4002	/* Also sync guest cr3 here in case we live migrate */
4003	svm->vmcb->save.cr3 = kvm_read_cr3(vcpu);
4004	mark_dirty(svm->vmcb, VMCB_CR);
4005
4006	svm_flush_tlb(vcpu);
4007}
4008
4009static int is_disabled(void)
4010{
4011	u64 vm_cr;
4012
4013	rdmsrl(MSR_VM_CR, vm_cr);
4014	if (vm_cr & (1 << SVM_VM_CR_SVM_DISABLE))
4015		return 1;
4016
4017	return 0;
4018}
4019
4020static void
4021svm_patch_hypercall(struct kvm_vcpu *vcpu, unsigned char *hypercall)
4022{
4023	/*
4024	 * Patch in the VMMCALL instruction:
4025	 */
4026	hypercall[0] = 0x0f;
4027	hypercall[1] = 0x01;
4028	hypercall[2] = 0xd9;
4029}
4030
4031static void svm_check_processor_compat(void *rtn)
4032{
4033	*(int *)rtn = 0;
4034}
4035
4036static bool svm_cpu_has_accelerated_tpr(void)
4037{
4038	return false;
4039}
4040
4041static u64 svm_get_mt_mask(struct kvm_vcpu *vcpu, gfn_t gfn, bool is_mmio)
4042{
4043	return 0;
4044}
4045
4046static void svm_cpuid_update(struct kvm_vcpu *vcpu)
4047{
4048}
4049
4050static void svm_set_supported_cpuid(u32 func, struct kvm_cpuid_entry2 *entry)
4051{
4052	switch (func) {
4053	case 0x80000001:
4054		if (nested)
4055			entry->ecx |= (1 << 2); /* Set SVM bit */
4056		break;
4057	case 0x8000000A:
4058		entry->eax = 1; /* SVM revision 1 */
4059		entry->ebx = 8; /* Lets support 8 ASIDs in case we add proper
4060				   ASID emulation to nested SVM */
4061		entry->ecx = 0; /* Reserved */
4062		entry->edx = 0; /* Per default do not support any
4063				   additional features */
4064
4065		/* Support next_rip if host supports it */
4066		if (boot_cpu_has(X86_FEATURE_NRIPS))
4067			entry->edx |= SVM_FEATURE_NRIP;
4068
4069		/* Support NPT for the guest if enabled */
4070		if (npt_enabled)
4071			entry->edx |= SVM_FEATURE_NPT;
4072
4073		break;
4074	}
4075}
4076
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
4077static int svm_get_lpage_level(void)
4078{
4079	return PT_PDPE_LEVEL;
4080}
4081
4082static bool svm_rdtscp_supported(void)
4083{
4084	return false;
4085}
4086
4087static bool svm_invpcid_supported(void)
4088{
4089	return false;
4090}
4091
4092static bool svm_mpx_supported(void)
4093{
4094	return false;
4095}
4096
4097static bool svm_has_wbinvd_exit(void)
4098{
4099	return true;
4100}
4101
4102static void svm_fpu_deactivate(struct kvm_vcpu *vcpu)
4103{
4104	struct vcpu_svm *svm = to_svm(vcpu);
4105
4106	set_exception_intercept(svm, NM_VECTOR);
4107	update_cr0_intercept(svm);
4108}
4109
4110#define PRE_EX(exit)  { .exit_code = (exit), \
4111			.stage = X86_ICPT_PRE_EXCEPT, }
4112#define POST_EX(exit) { .exit_code = (exit), \
4113			.stage = X86_ICPT_POST_EXCEPT, }
4114#define POST_MEM(exit) { .exit_code = (exit), \
4115			.stage = X86_ICPT_POST_MEMACCESS, }
4116
4117static const struct __x86_intercept {
4118	u32 exit_code;
4119	enum x86_intercept_stage stage;
4120} x86_intercept_map[] = {
4121	[x86_intercept_cr_read]		= POST_EX(SVM_EXIT_READ_CR0),
4122	[x86_intercept_cr_write]	= POST_EX(SVM_EXIT_WRITE_CR0),
4123	[x86_intercept_clts]		= POST_EX(SVM_EXIT_WRITE_CR0),
4124	[x86_intercept_lmsw]		= POST_EX(SVM_EXIT_WRITE_CR0),
4125	[x86_intercept_smsw]		= POST_EX(SVM_EXIT_READ_CR0),
4126	[x86_intercept_dr_read]		= POST_EX(SVM_EXIT_READ_DR0),
4127	[x86_intercept_dr_write]	= POST_EX(SVM_EXIT_WRITE_DR0),
4128	[x86_intercept_sldt]		= POST_EX(SVM_EXIT_LDTR_READ),
4129	[x86_intercept_str]		= POST_EX(SVM_EXIT_TR_READ),
4130	[x86_intercept_lldt]		= POST_EX(SVM_EXIT_LDTR_WRITE),
4131	[x86_intercept_ltr]		= POST_EX(SVM_EXIT_TR_WRITE),
4132	[x86_intercept_sgdt]		= POST_EX(SVM_EXIT_GDTR_READ),
4133	[x86_intercept_sidt]		= POST_EX(SVM_EXIT_IDTR_READ),
4134	[x86_intercept_lgdt]		= POST_EX(SVM_EXIT_GDTR_WRITE),
4135	[x86_intercept_lidt]		= POST_EX(SVM_EXIT_IDTR_WRITE),
4136	[x86_intercept_vmrun]		= POST_EX(SVM_EXIT_VMRUN),
4137	[x86_intercept_vmmcall]		= POST_EX(SVM_EXIT_VMMCALL),
4138	[x86_intercept_vmload]		= POST_EX(SVM_EXIT_VMLOAD),
4139	[x86_intercept_vmsave]		= POST_EX(SVM_EXIT_VMSAVE),
4140	[x86_intercept_stgi]		= POST_EX(SVM_EXIT_STGI),
4141	[x86_intercept_clgi]		= POST_EX(SVM_EXIT_CLGI),
4142	[x86_intercept_skinit]		= POST_EX(SVM_EXIT_SKINIT),
4143	[x86_intercept_invlpga]		= POST_EX(SVM_EXIT_INVLPGA),
4144	[x86_intercept_rdtscp]		= POST_EX(SVM_EXIT_RDTSCP),
4145	[x86_intercept_monitor]		= POST_MEM(SVM_EXIT_MONITOR),
4146	[x86_intercept_mwait]		= POST_EX(SVM_EXIT_MWAIT),
4147	[x86_intercept_invlpg]		= POST_EX(SVM_EXIT_INVLPG),
4148	[x86_intercept_invd]		= POST_EX(SVM_EXIT_INVD),
4149	[x86_intercept_wbinvd]		= POST_EX(SVM_EXIT_WBINVD),
4150	[x86_intercept_wrmsr]		= POST_EX(SVM_EXIT_MSR),
4151	[x86_intercept_rdtsc]		= POST_EX(SVM_EXIT_RDTSC),
4152	[x86_intercept_rdmsr]		= POST_EX(SVM_EXIT_MSR),
4153	[x86_intercept_rdpmc]		= POST_EX(SVM_EXIT_RDPMC),
4154	[x86_intercept_cpuid]		= PRE_EX(SVM_EXIT_CPUID),
4155	[x86_intercept_rsm]		= PRE_EX(SVM_EXIT_RSM),
4156	[x86_intercept_pause]		= PRE_EX(SVM_EXIT_PAUSE),
4157	[x86_intercept_pushf]		= PRE_EX(SVM_EXIT_PUSHF),
4158	[x86_intercept_popf]		= PRE_EX(SVM_EXIT_POPF),
4159	[x86_intercept_intn]		= PRE_EX(SVM_EXIT_SWINT),
4160	[x86_intercept_iret]		= PRE_EX(SVM_EXIT_IRET),
4161	[x86_intercept_icebp]		= PRE_EX(SVM_EXIT_ICEBP),
4162	[x86_intercept_hlt]		= POST_EX(SVM_EXIT_HLT),
4163	[x86_intercept_in]		= POST_EX(SVM_EXIT_IOIO),
4164	[x86_intercept_ins]		= POST_EX(SVM_EXIT_IOIO),
4165	[x86_intercept_out]		= POST_EX(SVM_EXIT_IOIO),
4166	[x86_intercept_outs]		= POST_EX(SVM_EXIT_IOIO),
4167};
4168
4169#undef PRE_EX
4170#undef POST_EX
4171#undef POST_MEM
4172
4173static int svm_check_intercept(struct kvm_vcpu *vcpu,
4174			       struct x86_instruction_info *info,
4175			       enum x86_intercept_stage stage)
4176{
4177	struct vcpu_svm *svm = to_svm(vcpu);
4178	int vmexit, ret = X86EMUL_CONTINUE;
4179	struct __x86_intercept icpt_info;
4180	struct vmcb *vmcb = svm->vmcb;
4181
4182	if (info->intercept >= ARRAY_SIZE(x86_intercept_map))
4183		goto out;
4184
4185	icpt_info = x86_intercept_map[info->intercept];
4186
4187	if (stage != icpt_info.stage)
4188		goto out;
4189
4190	switch (icpt_info.exit_code) {
4191	case SVM_EXIT_READ_CR0:
4192		if (info->intercept == x86_intercept_cr_read)
4193			icpt_info.exit_code += info->modrm_reg;
4194		break;
4195	case SVM_EXIT_WRITE_CR0: {
4196		unsigned long cr0, val;
4197		u64 intercept;
4198
4199		if (info->intercept == x86_intercept_cr_write)
4200			icpt_info.exit_code += info->modrm_reg;
4201
4202		if (icpt_info.exit_code != SVM_EXIT_WRITE_CR0)
4203			break;
4204
4205		intercept = svm->nested.intercept;
4206
4207		if (!(intercept & (1ULL << INTERCEPT_SELECTIVE_CR0)))
4208			break;
4209
4210		cr0 = vcpu->arch.cr0 & ~SVM_CR0_SELECTIVE_MASK;
4211		val = info->src_val  & ~SVM_CR0_SELECTIVE_MASK;
4212
4213		if (info->intercept == x86_intercept_lmsw) {
4214			cr0 &= 0xfUL;
4215			val &= 0xfUL;
4216			/* lmsw can't clear PE - catch this here */
4217			if (cr0 & X86_CR0_PE)
4218				val |= X86_CR0_PE;
4219		}
4220
4221		if (cr0 ^ val)
4222			icpt_info.exit_code = SVM_EXIT_CR0_SEL_WRITE;
4223
4224		break;
4225	}
4226	case SVM_EXIT_READ_DR0:
4227	case SVM_EXIT_WRITE_DR0:
4228		icpt_info.exit_code += info->modrm_reg;
4229		break;
4230	case SVM_EXIT_MSR:
4231		if (info->intercept == x86_intercept_wrmsr)
4232			vmcb->control.exit_info_1 = 1;
4233		else
4234			vmcb->control.exit_info_1 = 0;
4235		break;
4236	case SVM_EXIT_PAUSE:
4237		/*
4238		 * We get this for NOP only, but pause
4239		 * is rep not, check this here
4240		 */
4241		if (info->rep_prefix != REPE_PREFIX)
4242			goto out;
4243	case SVM_EXIT_IOIO: {
4244		u64 exit_info;
4245		u32 bytes;
4246
4247		exit_info = (vcpu->arch.regs[VCPU_REGS_RDX] & 0xffff) << 16;
4248
4249		if (info->intercept == x86_intercept_in ||
4250		    info->intercept == x86_intercept_ins) {
4251			exit_info |= SVM_IOIO_TYPE_MASK;
4252			bytes = info->src_bytes;
4253		} else {
4254			bytes = info->dst_bytes;
4255		}
4256
4257		if (info->intercept == x86_intercept_outs ||
4258		    info->intercept == x86_intercept_ins)
4259			exit_info |= SVM_IOIO_STR_MASK;
4260
4261		if (info->rep_prefix)
4262			exit_info |= SVM_IOIO_REP_MASK;
4263
4264		bytes = min(bytes, 4u);
4265
4266		exit_info |= bytes << SVM_IOIO_SIZE_SHIFT;
4267
4268		exit_info |= (u32)info->ad_bytes << (SVM_IOIO_ASIZE_SHIFT - 1);
4269
4270		vmcb->control.exit_info_1 = exit_info;
4271		vmcb->control.exit_info_2 = info->next_rip;
4272
4273		break;
4274	}
4275	default:
4276		break;
4277	}
4278
4279	vmcb->control.next_rip  = info->next_rip;
4280	vmcb->control.exit_code = icpt_info.exit_code;
4281	vmexit = nested_svm_exit_handled(svm);
4282
4283	ret = (vmexit == NESTED_EXIT_DONE) ? X86EMUL_INTERCEPTED
4284					   : X86EMUL_CONTINUE;
4285
4286out:
4287	return ret;
4288}
4289
4290static void svm_handle_external_intr(struct kvm_vcpu *vcpu)
4291{
4292	local_irq_enable();
4293}
4294
4295static struct kvm_x86_ops svm_x86_ops = {
4296	.cpu_has_kvm_support = has_svm,
4297	.disabled_by_bios = is_disabled,
4298	.hardware_setup = svm_hardware_setup,
4299	.hardware_unsetup = svm_hardware_unsetup,
4300	.check_processor_compatibility = svm_check_processor_compat,
4301	.hardware_enable = svm_hardware_enable,
4302	.hardware_disable = svm_hardware_disable,
4303	.cpu_has_accelerated_tpr = svm_cpu_has_accelerated_tpr,
4304
4305	.vcpu_create = svm_create_vcpu,
4306	.vcpu_free = svm_free_vcpu,
4307	.vcpu_reset = svm_vcpu_reset,
4308
4309	.prepare_guest_switch = svm_prepare_guest_switch,
4310	.vcpu_load = svm_vcpu_load,
4311	.vcpu_put = svm_vcpu_put,
4312
4313	.update_db_bp_intercept = update_db_bp_intercept,
4314	.get_msr = svm_get_msr,
4315	.set_msr = svm_set_msr,
4316	.get_segment_base = svm_get_segment_base,
4317	.get_segment = svm_get_segment,
4318	.set_segment = svm_set_segment,
4319	.get_cpl = svm_get_cpl,
4320	.get_cs_db_l_bits = kvm_get_cs_db_l_bits,
4321	.decache_cr0_guest_bits = svm_decache_cr0_guest_bits,
4322	.decache_cr3 = svm_decache_cr3,
4323	.decache_cr4_guest_bits = svm_decache_cr4_guest_bits,
4324	.set_cr0 = svm_set_cr0,
4325	.set_cr3 = svm_set_cr3,
4326	.set_cr4 = svm_set_cr4,
4327	.set_efer = svm_set_efer,
4328	.get_idt = svm_get_idt,
4329	.set_idt = svm_set_idt,
4330	.get_gdt = svm_get_gdt,
4331	.set_gdt = svm_set_gdt,
4332	.get_dr6 = svm_get_dr6,
4333	.set_dr6 = svm_set_dr6,
4334	.set_dr7 = svm_set_dr7,
4335	.sync_dirty_debug_regs = svm_sync_dirty_debug_regs,
4336	.cache_reg = svm_cache_reg,
4337	.get_rflags = svm_get_rflags,
4338	.set_rflags = svm_set_rflags,
4339	.fpu_activate = svm_fpu_activate,
4340	.fpu_deactivate = svm_fpu_deactivate,
4341
4342	.tlb_flush = svm_flush_tlb,
4343
4344	.run = svm_vcpu_run,
4345	.handle_exit = handle_exit,
4346	.skip_emulated_instruction = skip_emulated_instruction,
4347	.set_interrupt_shadow = svm_set_interrupt_shadow,
4348	.get_interrupt_shadow = svm_get_interrupt_shadow,
4349	.patch_hypercall = svm_patch_hypercall,
4350	.set_irq = svm_set_irq,
4351	.set_nmi = svm_inject_nmi,
4352	.queue_exception = svm_queue_exception,
4353	.cancel_injection = svm_cancel_injection,
4354	.interrupt_allowed = svm_interrupt_allowed,
4355	.nmi_allowed = svm_nmi_allowed,
4356	.get_nmi_mask = svm_get_nmi_mask,
4357	.set_nmi_mask = svm_set_nmi_mask,
4358	.enable_nmi_window = enable_nmi_window,
4359	.enable_irq_window = enable_irq_window,
4360	.update_cr8_intercept = update_cr8_intercept,
4361	.set_virtual_x2apic_mode = svm_set_virtual_x2apic_mode,
4362	.vm_has_apicv = svm_vm_has_apicv,
4363	.load_eoi_exitmap = svm_load_eoi_exitmap,
4364	.hwapic_isr_update = svm_hwapic_isr_update,
4365	.sync_pir_to_irr = svm_sync_pir_to_irr,
4366
4367	.set_tss_addr = svm_set_tss_addr,
4368	.get_tdp_level = get_npt_level,
4369	.get_mt_mask = svm_get_mt_mask,
4370
4371	.get_exit_info = svm_get_exit_info,
 
4372
4373	.get_lpage_level = svm_get_lpage_level,
4374
4375	.cpuid_update = svm_cpuid_update,
4376
4377	.rdtscp_supported = svm_rdtscp_supported,
4378	.invpcid_supported = svm_invpcid_supported,
4379	.mpx_supported = svm_mpx_supported,
4380
4381	.set_supported_cpuid = svm_set_supported_cpuid,
4382
4383	.has_wbinvd_exit = svm_has_wbinvd_exit,
4384
4385	.set_tsc_khz = svm_set_tsc_khz,
4386	.read_tsc_offset = svm_read_tsc_offset,
4387	.write_tsc_offset = svm_write_tsc_offset,
4388	.adjust_tsc_offset = svm_adjust_tsc_offset,
4389	.compute_tsc_offset = svm_compute_tsc_offset,
4390	.read_l1_tsc = svm_read_l1_tsc,
4391
4392	.set_tdp_cr3 = set_tdp_cr3,
4393
4394	.check_intercept = svm_check_intercept,
4395	.handle_external_intr = svm_handle_external_intr,
4396};
4397
4398static int __init svm_init(void)
4399{
4400	return kvm_init(&svm_x86_ops, sizeof(struct vcpu_svm),
4401			__alignof__(struct vcpu_svm), THIS_MODULE);
4402}
4403
4404static void __exit svm_exit(void)
4405{
4406	kvm_exit();
4407}
4408
4409module_init(svm_init)
4410module_exit(svm_exit)
v3.1
   1/*
   2 * Kernel-based Virtual Machine driver for Linux
   3 *
   4 * AMD SVM support
   5 *
   6 * Copyright (C) 2006 Qumranet, Inc.
   7 * Copyright 2010 Red Hat, Inc. and/or its affiliates.
   8 *
   9 * Authors:
  10 *   Yaniv Kamay  <yaniv@qumranet.com>
  11 *   Avi Kivity   <avi@qumranet.com>
  12 *
  13 * This work is licensed under the terms of the GNU GPL, version 2.  See
  14 * the COPYING file in the top-level directory.
  15 *
  16 */
  17#include <linux/kvm_host.h>
  18
  19#include "irq.h"
  20#include "mmu.h"
  21#include "kvm_cache_regs.h"
  22#include "x86.h"
 
  23
  24#include <linux/module.h>
 
  25#include <linux/kernel.h>
  26#include <linux/vmalloc.h>
  27#include <linux/highmem.h>
  28#include <linux/sched.h>
  29#include <linux/ftrace_event.h>
  30#include <linux/slab.h>
  31
 
  32#include <asm/tlbflush.h>
  33#include <asm/desc.h>
 
  34#include <asm/kvm_para.h>
  35
  36#include <asm/virtext.h>
  37#include "trace.h"
  38
  39#define __ex(x) __kvm_handle_fault_on_reboot(x)
  40
  41MODULE_AUTHOR("Qumranet");
  42MODULE_LICENSE("GPL");
  43
 
 
 
 
 
 
  44#define IOPM_ALLOC_ORDER 2
  45#define MSRPM_ALLOC_ORDER 1
  46
  47#define SEG_TYPE_LDT 2
  48#define SEG_TYPE_BUSY_TSS16 3
  49
  50#define SVM_FEATURE_NPT            (1 <<  0)
  51#define SVM_FEATURE_LBRV           (1 <<  1)
  52#define SVM_FEATURE_SVML           (1 <<  2)
  53#define SVM_FEATURE_NRIP           (1 <<  3)
  54#define SVM_FEATURE_TSC_RATE       (1 <<  4)
  55#define SVM_FEATURE_VMCB_CLEAN     (1 <<  5)
  56#define SVM_FEATURE_FLUSH_ASID     (1 <<  6)
  57#define SVM_FEATURE_DECODE_ASSIST  (1 <<  7)
  58#define SVM_FEATURE_PAUSE_FILTER   (1 << 10)
  59
  60#define NESTED_EXIT_HOST	0	/* Exit handled on host level */
  61#define NESTED_EXIT_DONE	1	/* Exit caused nested vmexit  */
  62#define NESTED_EXIT_CONTINUE	2	/* Further checks needed      */
  63
  64#define DEBUGCTL_RESERVED_BITS (~(0x3fULL))
  65
  66#define TSC_RATIO_RSVD          0xffffff0000000000ULL
  67#define TSC_RATIO_MIN		0x0000000000000001ULL
  68#define TSC_RATIO_MAX		0x000000ffffffffffULL
  69
  70static bool erratum_383_found __read_mostly;
  71
  72static const u32 host_save_user_msrs[] = {
  73#ifdef CONFIG_X86_64
  74	MSR_STAR, MSR_LSTAR, MSR_CSTAR, MSR_SYSCALL_MASK, MSR_KERNEL_GS_BASE,
  75	MSR_FS_BASE,
  76#endif
  77	MSR_IA32_SYSENTER_CS, MSR_IA32_SYSENTER_ESP, MSR_IA32_SYSENTER_EIP,
  78};
  79
  80#define NR_HOST_SAVE_USER_MSRS ARRAY_SIZE(host_save_user_msrs)
  81
  82struct kvm_vcpu;
  83
  84struct nested_state {
  85	struct vmcb *hsave;
  86	u64 hsave_msr;
  87	u64 vm_cr_msr;
  88	u64 vmcb;
  89
  90	/* These are the merged vectors */
  91	u32 *msrpm;
  92
  93	/* gpa pointers to the real vectors */
  94	u64 vmcb_msrpm;
  95	u64 vmcb_iopm;
  96
  97	/* A VMEXIT is required but not yet emulated */
  98	bool exit_required;
  99
 100	/* cache for intercepts of the guest */
 101	u32 intercept_cr;
 102	u32 intercept_dr;
 103	u32 intercept_exceptions;
 104	u64 intercept;
 105
 106	/* Nested Paging related state */
 107	u64 nested_cr3;
 108};
 109
 110#define MSRPM_OFFSETS	16
 111static u32 msrpm_offsets[MSRPM_OFFSETS] __read_mostly;
 112
 
 
 
 
 
 
 113struct vcpu_svm {
 114	struct kvm_vcpu vcpu;
 115	struct vmcb *vmcb;
 116	unsigned long vmcb_pa;
 117	struct svm_cpu_data *svm_data;
 118	uint64_t asid_generation;
 119	uint64_t sysenter_esp;
 120	uint64_t sysenter_eip;
 121
 122	u64 next_rip;
 123
 124	u64 host_user_msrs[NR_HOST_SAVE_USER_MSRS];
 125	struct {
 126		u16 fs;
 127		u16 gs;
 128		u16 ldt;
 129		u64 gs_base;
 130	} host;
 131
 132	u32 *msrpm;
 133
 134	ulong nmi_iret_rip;
 135
 136	struct nested_state nested;
 137
 138	bool nmi_singlestep;
 139
 140	unsigned int3_injected;
 141	unsigned long int3_rip;
 142	u32 apf_reason;
 143
 144	u64  tsc_ratio;
 145};
 146
 147static DEFINE_PER_CPU(u64, current_tsc_ratio);
 148#define TSC_RATIO_DEFAULT	0x0100000000ULL
 149
 150#define MSR_INVALID			0xffffffffU
 151
 152static struct svm_direct_access_msrs {
 153	u32 index;   /* Index of the MSR */
 154	bool always; /* True if intercept is always on */
 155} direct_access_msrs[] = {
 156	{ .index = MSR_STAR,				.always = true  },
 157	{ .index = MSR_IA32_SYSENTER_CS,		.always = true  },
 158#ifdef CONFIG_X86_64
 159	{ .index = MSR_GS_BASE,				.always = true  },
 160	{ .index = MSR_FS_BASE,				.always = true  },
 161	{ .index = MSR_KERNEL_GS_BASE,			.always = true  },
 162	{ .index = MSR_LSTAR,				.always = true  },
 163	{ .index = MSR_CSTAR,				.always = true  },
 164	{ .index = MSR_SYSCALL_MASK,			.always = true  },
 165#endif
 166	{ .index = MSR_IA32_LASTBRANCHFROMIP,		.always = false },
 167	{ .index = MSR_IA32_LASTBRANCHTOIP,		.always = false },
 168	{ .index = MSR_IA32_LASTINTFROMIP,		.always = false },
 169	{ .index = MSR_IA32_LASTINTTOIP,		.always = false },
 170	{ .index = MSR_INVALID,				.always = false },
 171};
 172
 173/* enable NPT for AMD64 and X86 with PAE */
 174#if defined(CONFIG_X86_64) || defined(CONFIG_X86_PAE)
 175static bool npt_enabled = true;
 176#else
 177static bool npt_enabled;
 178#endif
 179static int npt = 1;
 180
 
 
 181module_param(npt, int, S_IRUGO);
 182
 183static int nested = 1;
 
 184module_param(nested, int, S_IRUGO);
 185
 186static void svm_flush_tlb(struct kvm_vcpu *vcpu);
 187static void svm_complete_interrupts(struct vcpu_svm *svm);
 188
 189static int nested_svm_exit_handled(struct vcpu_svm *svm);
 190static int nested_svm_intercept(struct vcpu_svm *svm);
 191static int nested_svm_vmexit(struct vcpu_svm *svm);
 192static int nested_svm_check_exception(struct vcpu_svm *svm, unsigned nr,
 193				      bool has_error_code, u32 error_code);
 194static u64 __scale_tsc(u64 ratio, u64 tsc);
 195
 196enum {
 197	VMCB_INTERCEPTS, /* Intercept vectors, TSC offset,
 198			    pause filter count */
 199	VMCB_PERM_MAP,   /* IOPM Base and MSRPM Base */
 200	VMCB_ASID,	 /* ASID */
 201	VMCB_INTR,	 /* int_ctl, int_vector */
 202	VMCB_NPT,        /* npt_en, nCR3, gPAT */
 203	VMCB_CR,	 /* CR0, CR3, CR4, EFER */
 204	VMCB_DR,         /* DR6, DR7 */
 205	VMCB_DT,         /* GDT, IDT */
 206	VMCB_SEG,        /* CS, DS, SS, ES, CPL */
 207	VMCB_CR2,        /* CR2 only */
 208	VMCB_LBR,        /* DBGCTL, BR_FROM, BR_TO, LAST_EX_FROM, LAST_EX_TO */
 209	VMCB_DIRTY_MAX,
 210};
 211
 212/* TPR and CR2 are always written before VMRUN */
 213#define VMCB_ALWAYS_DIRTY_MASK	((1U << VMCB_INTR) | (1U << VMCB_CR2))
 214
 215static inline void mark_all_dirty(struct vmcb *vmcb)
 216{
 217	vmcb->control.clean = 0;
 218}
 219
 220static inline void mark_all_clean(struct vmcb *vmcb)
 221{
 222	vmcb->control.clean = ((1 << VMCB_DIRTY_MAX) - 1)
 223			       & ~VMCB_ALWAYS_DIRTY_MASK;
 224}
 225
 226static inline void mark_dirty(struct vmcb *vmcb, int bit)
 227{
 228	vmcb->control.clean &= ~(1 << bit);
 229}
 230
 231static inline struct vcpu_svm *to_svm(struct kvm_vcpu *vcpu)
 232{
 233	return container_of(vcpu, struct vcpu_svm, vcpu);
 234}
 235
 236static void recalc_intercepts(struct vcpu_svm *svm)
 237{
 238	struct vmcb_control_area *c, *h;
 239	struct nested_state *g;
 240
 241	mark_dirty(svm->vmcb, VMCB_INTERCEPTS);
 242
 243	if (!is_guest_mode(&svm->vcpu))
 244		return;
 245
 246	c = &svm->vmcb->control;
 247	h = &svm->nested.hsave->control;
 248	g = &svm->nested;
 249
 250	c->intercept_cr = h->intercept_cr | g->intercept_cr;
 251	c->intercept_dr = h->intercept_dr | g->intercept_dr;
 252	c->intercept_exceptions = h->intercept_exceptions | g->intercept_exceptions;
 253	c->intercept = h->intercept | g->intercept;
 254}
 255
 256static inline struct vmcb *get_host_vmcb(struct vcpu_svm *svm)
 257{
 258	if (is_guest_mode(&svm->vcpu))
 259		return svm->nested.hsave;
 260	else
 261		return svm->vmcb;
 262}
 263
 264static inline void set_cr_intercept(struct vcpu_svm *svm, int bit)
 265{
 266	struct vmcb *vmcb = get_host_vmcb(svm);
 267
 268	vmcb->control.intercept_cr |= (1U << bit);
 269
 270	recalc_intercepts(svm);
 271}
 272
 273static inline void clr_cr_intercept(struct vcpu_svm *svm, int bit)
 274{
 275	struct vmcb *vmcb = get_host_vmcb(svm);
 276
 277	vmcb->control.intercept_cr &= ~(1U << bit);
 278
 279	recalc_intercepts(svm);
 280}
 281
 282static inline bool is_cr_intercept(struct vcpu_svm *svm, int bit)
 283{
 284	struct vmcb *vmcb = get_host_vmcb(svm);
 285
 286	return vmcb->control.intercept_cr & (1U << bit);
 287}
 288
 289static inline void set_dr_intercept(struct vcpu_svm *svm, int bit)
 290{
 291	struct vmcb *vmcb = get_host_vmcb(svm);
 292
 293	vmcb->control.intercept_dr |= (1U << bit);
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 294
 295	recalc_intercepts(svm);
 296}
 297
 298static inline void clr_dr_intercept(struct vcpu_svm *svm, int bit)
 299{
 300	struct vmcb *vmcb = get_host_vmcb(svm);
 301
 302	vmcb->control.intercept_dr &= ~(1U << bit);
 303
 304	recalc_intercepts(svm);
 305}
 306
 307static inline void set_exception_intercept(struct vcpu_svm *svm, int bit)
 308{
 309	struct vmcb *vmcb = get_host_vmcb(svm);
 310
 311	vmcb->control.intercept_exceptions |= (1U << bit);
 312
 313	recalc_intercepts(svm);
 314}
 315
 316static inline void clr_exception_intercept(struct vcpu_svm *svm, int bit)
 317{
 318	struct vmcb *vmcb = get_host_vmcb(svm);
 319
 320	vmcb->control.intercept_exceptions &= ~(1U << bit);
 321
 322	recalc_intercepts(svm);
 323}
 324
 325static inline void set_intercept(struct vcpu_svm *svm, int bit)
 326{
 327	struct vmcb *vmcb = get_host_vmcb(svm);
 328
 329	vmcb->control.intercept |= (1ULL << bit);
 330
 331	recalc_intercepts(svm);
 332}
 333
 334static inline void clr_intercept(struct vcpu_svm *svm, int bit)
 335{
 336	struct vmcb *vmcb = get_host_vmcb(svm);
 337
 338	vmcb->control.intercept &= ~(1ULL << bit);
 339
 340	recalc_intercepts(svm);
 341}
 342
 343static inline void enable_gif(struct vcpu_svm *svm)
 344{
 345	svm->vcpu.arch.hflags |= HF_GIF_MASK;
 346}
 347
 348static inline void disable_gif(struct vcpu_svm *svm)
 349{
 350	svm->vcpu.arch.hflags &= ~HF_GIF_MASK;
 351}
 352
 353static inline bool gif_set(struct vcpu_svm *svm)
 354{
 355	return !!(svm->vcpu.arch.hflags & HF_GIF_MASK);
 356}
 357
 358static unsigned long iopm_base;
 359
 360struct kvm_ldttss_desc {
 361	u16 limit0;
 362	u16 base0;
 363	unsigned base1:8, type:5, dpl:2, p:1;
 364	unsigned limit1:4, zero0:3, g:1, base2:8;
 365	u32 base3;
 366	u32 zero1;
 367} __attribute__((packed));
 368
 369struct svm_cpu_data {
 370	int cpu;
 371
 372	u64 asid_generation;
 373	u32 max_asid;
 374	u32 next_asid;
 375	struct kvm_ldttss_desc *tss_desc;
 376
 377	struct page *save_area;
 378};
 379
 380static DEFINE_PER_CPU(struct svm_cpu_data *, svm_data);
 381
 382struct svm_init_data {
 383	int cpu;
 384	int r;
 385};
 386
 387static u32 msrpm_ranges[] = {0, 0xc0000000, 0xc0010000};
 388
 389#define NUM_MSR_MAPS ARRAY_SIZE(msrpm_ranges)
 390#define MSRS_RANGE_SIZE 2048
 391#define MSRS_IN_RANGE (MSRS_RANGE_SIZE * 8 / 2)
 392
 393static u32 svm_msrpm_offset(u32 msr)
 394{
 395	u32 offset;
 396	int i;
 397
 398	for (i = 0; i < NUM_MSR_MAPS; i++) {
 399		if (msr < msrpm_ranges[i] ||
 400		    msr >= msrpm_ranges[i] + MSRS_IN_RANGE)
 401			continue;
 402
 403		offset  = (msr - msrpm_ranges[i]) / 4; /* 4 msrs per u8 */
 404		offset += (i * MSRS_RANGE_SIZE);       /* add range offset */
 405
 406		/* Now we have the u8 offset - but need the u32 offset */
 407		return offset / 4;
 408	}
 409
 410	/* MSR not in any range */
 411	return MSR_INVALID;
 412}
 413
 414#define MAX_INST_SIZE 15
 415
 416static inline void clgi(void)
 417{
 418	asm volatile (__ex(SVM_CLGI));
 419}
 420
 421static inline void stgi(void)
 422{
 423	asm volatile (__ex(SVM_STGI));
 424}
 425
 426static inline void invlpga(unsigned long addr, u32 asid)
 427{
 428	asm volatile (__ex(SVM_INVLPGA) : : "a"(addr), "c"(asid));
 429}
 430
 431static int get_npt_level(void)
 432{
 433#ifdef CONFIG_X86_64
 434	return PT64_ROOT_LEVEL;
 435#else
 436	return PT32E_ROOT_LEVEL;
 437#endif
 438}
 439
 440static void svm_set_efer(struct kvm_vcpu *vcpu, u64 efer)
 441{
 442	vcpu->arch.efer = efer;
 443	if (!npt_enabled && !(efer & EFER_LMA))
 444		efer &= ~EFER_LME;
 445
 446	to_svm(vcpu)->vmcb->save.efer = efer | EFER_SVME;
 447	mark_dirty(to_svm(vcpu)->vmcb, VMCB_CR);
 448}
 449
 450static int is_external_interrupt(u32 info)
 451{
 452	info &= SVM_EVTINJ_TYPE_MASK | SVM_EVTINJ_VALID;
 453	return info == (SVM_EVTINJ_VALID | SVM_EVTINJ_TYPE_INTR);
 454}
 455
 456static u32 svm_get_interrupt_shadow(struct kvm_vcpu *vcpu, int mask)
 457{
 458	struct vcpu_svm *svm = to_svm(vcpu);
 459	u32 ret = 0;
 460
 461	if (svm->vmcb->control.int_state & SVM_INTERRUPT_SHADOW_MASK)
 462		ret |= KVM_X86_SHADOW_INT_STI | KVM_X86_SHADOW_INT_MOV_SS;
 463	return ret & mask;
 464}
 465
 466static void svm_set_interrupt_shadow(struct kvm_vcpu *vcpu, int mask)
 467{
 468	struct vcpu_svm *svm = to_svm(vcpu);
 469
 470	if (mask == 0)
 471		svm->vmcb->control.int_state &= ~SVM_INTERRUPT_SHADOW_MASK;
 472	else
 473		svm->vmcb->control.int_state |= SVM_INTERRUPT_SHADOW_MASK;
 474
 475}
 476
 477static void skip_emulated_instruction(struct kvm_vcpu *vcpu)
 478{
 479	struct vcpu_svm *svm = to_svm(vcpu);
 480
 481	if (svm->vmcb->control.next_rip != 0)
 482		svm->next_rip = svm->vmcb->control.next_rip;
 483
 484	if (!svm->next_rip) {
 485		if (emulate_instruction(vcpu, EMULTYPE_SKIP) !=
 486				EMULATE_DONE)
 487			printk(KERN_DEBUG "%s: NOP\n", __func__);
 488		return;
 489	}
 490	if (svm->next_rip - kvm_rip_read(vcpu) > MAX_INST_SIZE)
 491		printk(KERN_ERR "%s: ip 0x%lx next 0x%llx\n",
 492		       __func__, kvm_rip_read(vcpu), svm->next_rip);
 493
 494	kvm_rip_write(vcpu, svm->next_rip);
 495	svm_set_interrupt_shadow(vcpu, 0);
 496}
 497
 498static void svm_queue_exception(struct kvm_vcpu *vcpu, unsigned nr,
 499				bool has_error_code, u32 error_code,
 500				bool reinject)
 501{
 502	struct vcpu_svm *svm = to_svm(vcpu);
 503
 504	/*
 505	 * If we are within a nested VM we'd better #VMEXIT and let the guest
 506	 * handle the exception
 507	 */
 508	if (!reinject &&
 509	    nested_svm_check_exception(svm, nr, has_error_code, error_code))
 510		return;
 511
 512	if (nr == BP_VECTOR && !static_cpu_has(X86_FEATURE_NRIPS)) {
 513		unsigned long rip, old_rip = kvm_rip_read(&svm->vcpu);
 514
 515		/*
 516		 * For guest debugging where we have to reinject #BP if some
 517		 * INT3 is guest-owned:
 518		 * Emulate nRIP by moving RIP forward. Will fail if injection
 519		 * raises a fault that is not intercepted. Still better than
 520		 * failing in all cases.
 521		 */
 522		skip_emulated_instruction(&svm->vcpu);
 523		rip = kvm_rip_read(&svm->vcpu);
 524		svm->int3_rip = rip + svm->vmcb->save.cs.base;
 525		svm->int3_injected = rip - old_rip;
 526	}
 527
 528	svm->vmcb->control.event_inj = nr
 529		| SVM_EVTINJ_VALID
 530		| (has_error_code ? SVM_EVTINJ_VALID_ERR : 0)
 531		| SVM_EVTINJ_TYPE_EXEPT;
 532	svm->vmcb->control.event_inj_err = error_code;
 533}
 534
 535static void svm_init_erratum_383(void)
 536{
 537	u32 low, high;
 538	int err;
 539	u64 val;
 540
 541	if (!cpu_has_amd_erratum(amd_erratum_383))
 542		return;
 543
 544	/* Use _safe variants to not break nested virtualization */
 545	val = native_read_msr_safe(MSR_AMD64_DC_CFG, &err);
 546	if (err)
 547		return;
 548
 549	val |= (1ULL << 47);
 550
 551	low  = lower_32_bits(val);
 552	high = upper_32_bits(val);
 553
 554	native_write_msr_safe(MSR_AMD64_DC_CFG, low, high);
 555
 556	erratum_383_found = true;
 557}
 558
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 559static int has_svm(void)
 560{
 561	const char *msg;
 562
 563	if (!cpu_has_svm(&msg)) {
 564		printk(KERN_INFO "has_svm: %s\n", msg);
 565		return 0;
 566	}
 567
 568	return 1;
 569}
 570
 571static void svm_hardware_disable(void *garbage)
 572{
 573	/* Make sure we clean up behind us */
 574	if (static_cpu_has(X86_FEATURE_TSCRATEMSR))
 575		wrmsrl(MSR_AMD64_TSC_RATIO, TSC_RATIO_DEFAULT);
 576
 577	cpu_svm_disable();
 
 
 578}
 579
 580static int svm_hardware_enable(void *garbage)
 581{
 582
 583	struct svm_cpu_data *sd;
 584	uint64_t efer;
 585	struct desc_ptr gdt_descr;
 586	struct desc_struct *gdt;
 587	int me = raw_smp_processor_id();
 588
 589	rdmsrl(MSR_EFER, efer);
 590	if (efer & EFER_SVME)
 591		return -EBUSY;
 592
 593	if (!has_svm()) {
 594		printk(KERN_ERR "svm_hardware_enable: err EOPNOTSUPP on %d\n",
 595		       me);
 596		return -EINVAL;
 597	}
 598	sd = per_cpu(svm_data, me);
 599
 600	if (!sd) {
 601		printk(KERN_ERR "svm_hardware_enable: svm_data is NULL on %d\n",
 602		       me);
 603		return -EINVAL;
 604	}
 605
 606	sd->asid_generation = 1;
 607	sd->max_asid = cpuid_ebx(SVM_CPUID_FUNC) - 1;
 608	sd->next_asid = sd->max_asid + 1;
 609
 610	native_store_gdt(&gdt_descr);
 611	gdt = (struct desc_struct *)gdt_descr.address;
 612	sd->tss_desc = (struct kvm_ldttss_desc *)(gdt + GDT_ENTRY_TSS);
 613
 614	wrmsrl(MSR_EFER, efer | EFER_SVME);
 615
 616	wrmsrl(MSR_VM_HSAVE_PA, page_to_pfn(sd->save_area) << PAGE_SHIFT);
 617
 618	if (static_cpu_has(X86_FEATURE_TSCRATEMSR)) {
 619		wrmsrl(MSR_AMD64_TSC_RATIO, TSC_RATIO_DEFAULT);
 620		__get_cpu_var(current_tsc_ratio) = TSC_RATIO_DEFAULT;
 621	}
 622
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 623	svm_init_erratum_383();
 624
 
 
 625	return 0;
 626}
 627
 628static void svm_cpu_uninit(int cpu)
 629{
 630	struct svm_cpu_data *sd = per_cpu(svm_data, raw_smp_processor_id());
 631
 632	if (!sd)
 633		return;
 634
 635	per_cpu(svm_data, raw_smp_processor_id()) = NULL;
 636	__free_page(sd->save_area);
 637	kfree(sd);
 638}
 639
 640static int svm_cpu_init(int cpu)
 641{
 642	struct svm_cpu_data *sd;
 643	int r;
 644
 645	sd = kzalloc(sizeof(struct svm_cpu_data), GFP_KERNEL);
 646	if (!sd)
 647		return -ENOMEM;
 648	sd->cpu = cpu;
 649	sd->save_area = alloc_page(GFP_KERNEL);
 650	r = -ENOMEM;
 651	if (!sd->save_area)
 652		goto err_1;
 653
 654	per_cpu(svm_data, cpu) = sd;
 655
 656	return 0;
 657
 658err_1:
 659	kfree(sd);
 660	return r;
 661
 662}
 663
 664static bool valid_msr_intercept(u32 index)
 665{
 666	int i;
 667
 668	for (i = 0; direct_access_msrs[i].index != MSR_INVALID; i++)
 669		if (direct_access_msrs[i].index == index)
 670			return true;
 671
 672	return false;
 673}
 674
 675static void set_msr_interception(u32 *msrpm, unsigned msr,
 676				 int read, int write)
 677{
 678	u8 bit_read, bit_write;
 679	unsigned long tmp;
 680	u32 offset;
 681
 682	/*
 683	 * If this warning triggers extend the direct_access_msrs list at the
 684	 * beginning of the file
 685	 */
 686	WARN_ON(!valid_msr_intercept(msr));
 687
 688	offset    = svm_msrpm_offset(msr);
 689	bit_read  = 2 * (msr & 0x0f);
 690	bit_write = 2 * (msr & 0x0f) + 1;
 691	tmp       = msrpm[offset];
 692
 693	BUG_ON(offset == MSR_INVALID);
 694
 695	read  ? clear_bit(bit_read,  &tmp) : set_bit(bit_read,  &tmp);
 696	write ? clear_bit(bit_write, &tmp) : set_bit(bit_write, &tmp);
 697
 698	msrpm[offset] = tmp;
 699}
 700
 701static void svm_vcpu_init_msrpm(u32 *msrpm)
 702{
 703	int i;
 704
 705	memset(msrpm, 0xff, PAGE_SIZE * (1 << MSRPM_ALLOC_ORDER));
 706
 707	for (i = 0; direct_access_msrs[i].index != MSR_INVALID; i++) {
 708		if (!direct_access_msrs[i].always)
 709			continue;
 710
 711		set_msr_interception(msrpm, direct_access_msrs[i].index, 1, 1);
 712	}
 713}
 714
 715static void add_msr_offset(u32 offset)
 716{
 717	int i;
 718
 719	for (i = 0; i < MSRPM_OFFSETS; ++i) {
 720
 721		/* Offset already in list? */
 722		if (msrpm_offsets[i] == offset)
 723			return;
 724
 725		/* Slot used by another offset? */
 726		if (msrpm_offsets[i] != MSR_INVALID)
 727			continue;
 728
 729		/* Add offset to list */
 730		msrpm_offsets[i] = offset;
 731
 732		return;
 733	}
 734
 735	/*
 736	 * If this BUG triggers the msrpm_offsets table has an overflow. Just
 737	 * increase MSRPM_OFFSETS in this case.
 738	 */
 739	BUG();
 740}
 741
 742static void init_msrpm_offsets(void)
 743{
 744	int i;
 745
 746	memset(msrpm_offsets, 0xff, sizeof(msrpm_offsets));
 747
 748	for (i = 0; direct_access_msrs[i].index != MSR_INVALID; i++) {
 749		u32 offset;
 750
 751		offset = svm_msrpm_offset(direct_access_msrs[i].index);
 752		BUG_ON(offset == MSR_INVALID);
 753
 754		add_msr_offset(offset);
 755	}
 756}
 757
 758static void svm_enable_lbrv(struct vcpu_svm *svm)
 759{
 760	u32 *msrpm = svm->msrpm;
 761
 762	svm->vmcb->control.lbr_ctl = 1;
 763	set_msr_interception(msrpm, MSR_IA32_LASTBRANCHFROMIP, 1, 1);
 764	set_msr_interception(msrpm, MSR_IA32_LASTBRANCHTOIP, 1, 1);
 765	set_msr_interception(msrpm, MSR_IA32_LASTINTFROMIP, 1, 1);
 766	set_msr_interception(msrpm, MSR_IA32_LASTINTTOIP, 1, 1);
 767}
 768
 769static void svm_disable_lbrv(struct vcpu_svm *svm)
 770{
 771	u32 *msrpm = svm->msrpm;
 772
 773	svm->vmcb->control.lbr_ctl = 0;
 774	set_msr_interception(msrpm, MSR_IA32_LASTBRANCHFROMIP, 0, 0);
 775	set_msr_interception(msrpm, MSR_IA32_LASTBRANCHTOIP, 0, 0);
 776	set_msr_interception(msrpm, MSR_IA32_LASTINTFROMIP, 0, 0);
 777	set_msr_interception(msrpm, MSR_IA32_LASTINTTOIP, 0, 0);
 778}
 779
 780static __init int svm_hardware_setup(void)
 781{
 782	int cpu;
 783	struct page *iopm_pages;
 784	void *iopm_va;
 785	int r;
 786
 787	iopm_pages = alloc_pages(GFP_KERNEL, IOPM_ALLOC_ORDER);
 788
 789	if (!iopm_pages)
 790		return -ENOMEM;
 791
 792	iopm_va = page_address(iopm_pages);
 793	memset(iopm_va, 0xff, PAGE_SIZE * (1 << IOPM_ALLOC_ORDER));
 794	iopm_base = page_to_pfn(iopm_pages) << PAGE_SHIFT;
 795
 796	init_msrpm_offsets();
 797
 798	if (boot_cpu_has(X86_FEATURE_NX))
 799		kvm_enable_efer_bits(EFER_NX);
 800
 801	if (boot_cpu_has(X86_FEATURE_FXSR_OPT))
 802		kvm_enable_efer_bits(EFER_FFXSR);
 803
 804	if (boot_cpu_has(X86_FEATURE_TSCRATEMSR)) {
 805		u64 max;
 806
 807		kvm_has_tsc_control = true;
 808
 809		/*
 810		 * Make sure the user can only configure tsc_khz values that
 811		 * fit into a signed integer.
 812		 * A min value is not calculated needed because it will always
 813		 * be 1 on all machines and a value of 0 is used to disable
 814		 * tsc-scaling for the vcpu.
 815		 */
 816		max = min(0x7fffffffULL, __scale_tsc(tsc_khz, TSC_RATIO_MAX));
 817
 818		kvm_max_guest_tsc_khz = max;
 819	}
 820
 821	if (nested) {
 822		printk(KERN_INFO "kvm: Nested Virtualization enabled\n");
 823		kvm_enable_efer_bits(EFER_SVME | EFER_LMSLE);
 824	}
 825
 826	for_each_possible_cpu(cpu) {
 827		r = svm_cpu_init(cpu);
 828		if (r)
 829			goto err;
 830	}
 831
 832	if (!boot_cpu_has(X86_FEATURE_NPT))
 833		npt_enabled = false;
 834
 835	if (npt_enabled && !npt) {
 836		printk(KERN_INFO "kvm: Nested Paging disabled\n");
 837		npt_enabled = false;
 838	}
 839
 840	if (npt_enabled) {
 841		printk(KERN_INFO "kvm: Nested Paging enabled\n");
 842		kvm_enable_tdp();
 843	} else
 844		kvm_disable_tdp();
 845
 846	return 0;
 847
 848err:
 849	__free_pages(iopm_pages, IOPM_ALLOC_ORDER);
 850	iopm_base = 0;
 851	return r;
 852}
 853
 854static __exit void svm_hardware_unsetup(void)
 855{
 856	int cpu;
 857
 858	for_each_possible_cpu(cpu)
 859		svm_cpu_uninit(cpu);
 860
 861	__free_pages(pfn_to_page(iopm_base >> PAGE_SHIFT), IOPM_ALLOC_ORDER);
 862	iopm_base = 0;
 863}
 864
 865static void init_seg(struct vmcb_seg *seg)
 866{
 867	seg->selector = 0;
 868	seg->attrib = SVM_SELECTOR_P_MASK | SVM_SELECTOR_S_MASK |
 869		      SVM_SELECTOR_WRITE_MASK; /* Read/Write Data Segment */
 870	seg->limit = 0xffff;
 871	seg->base = 0;
 872}
 873
 874static void init_sys_seg(struct vmcb_seg *seg, uint32_t type)
 875{
 876	seg->selector = 0;
 877	seg->attrib = SVM_SELECTOR_P_MASK | type;
 878	seg->limit = 0xffff;
 879	seg->base = 0;
 880}
 881
 882static u64 __scale_tsc(u64 ratio, u64 tsc)
 883{
 884	u64 mult, frac, _tsc;
 885
 886	mult  = ratio >> 32;
 887	frac  = ratio & ((1ULL << 32) - 1);
 888
 889	_tsc  = tsc;
 890	_tsc *= mult;
 891	_tsc += (tsc >> 32) * frac;
 892	_tsc += ((tsc & ((1ULL << 32) - 1)) * frac) >> 32;
 893
 894	return _tsc;
 895}
 896
 897static u64 svm_scale_tsc(struct kvm_vcpu *vcpu, u64 tsc)
 898{
 899	struct vcpu_svm *svm = to_svm(vcpu);
 900	u64 _tsc = tsc;
 901
 902	if (svm->tsc_ratio != TSC_RATIO_DEFAULT)
 903		_tsc = __scale_tsc(svm->tsc_ratio, tsc);
 904
 905	return _tsc;
 906}
 907
 908static void svm_set_tsc_khz(struct kvm_vcpu *vcpu, u32 user_tsc_khz)
 909{
 910	struct vcpu_svm *svm = to_svm(vcpu);
 911	u64 ratio;
 912	u64 khz;
 913
 914	/* TSC scaling supported? */
 915	if (!boot_cpu_has(X86_FEATURE_TSCRATEMSR))
 
 916		return;
 
 917
 918	/* TSC-Scaling disabled or guest TSC same frequency as host TSC? */
 919	if (user_tsc_khz == 0) {
 920		vcpu->arch.virtual_tsc_khz = 0;
 921		svm->tsc_ratio = TSC_RATIO_DEFAULT;
 
 
 
 922		return;
 923	}
 924
 925	khz = user_tsc_khz;
 926
 927	/* TSC scaling required  - calculate ratio */
 928	ratio = khz << 32;
 929	do_div(ratio, tsc_khz);
 930
 931	if (ratio == 0 || ratio & TSC_RATIO_RSVD) {
 932		WARN_ONCE(1, "Invalid TSC ratio - virtual-tsc-khz=%u\n",
 933				user_tsc_khz);
 934		return;
 935	}
 936	vcpu->arch.virtual_tsc_khz = user_tsc_khz;
 937	svm->tsc_ratio             = ratio;
 938}
 939
 
 
 
 
 
 
 
 940static void svm_write_tsc_offset(struct kvm_vcpu *vcpu, u64 offset)
 941{
 942	struct vcpu_svm *svm = to_svm(vcpu);
 943	u64 g_tsc_offset = 0;
 944
 945	if (is_guest_mode(vcpu)) {
 946		g_tsc_offset = svm->vmcb->control.tsc_offset -
 947			       svm->nested.hsave->control.tsc_offset;
 948		svm->nested.hsave->control.tsc_offset = offset;
 949	}
 
 
 
 950
 951	svm->vmcb->control.tsc_offset = offset + g_tsc_offset;
 952
 953	mark_dirty(svm->vmcb, VMCB_INTERCEPTS);
 954}
 955
 956static void svm_adjust_tsc_offset(struct kvm_vcpu *vcpu, s64 adjustment)
 957{
 958	struct vcpu_svm *svm = to_svm(vcpu);
 959
 
 
 
 
 960	svm->vmcb->control.tsc_offset += adjustment;
 961	if (is_guest_mode(vcpu))
 962		svm->nested.hsave->control.tsc_offset += adjustment;
 
 
 
 
 
 963	mark_dirty(svm->vmcb, VMCB_INTERCEPTS);
 964}
 965
 966static u64 svm_compute_tsc_offset(struct kvm_vcpu *vcpu, u64 target_tsc)
 967{
 968	u64 tsc;
 969
 970	tsc = svm_scale_tsc(vcpu, native_read_tsc());
 971
 972	return target_tsc - tsc;
 973}
 974
 975static void init_vmcb(struct vcpu_svm *svm)
 976{
 977	struct vmcb_control_area *control = &svm->vmcb->control;
 978	struct vmcb_save_area *save = &svm->vmcb->save;
 979
 980	svm->vcpu.fpu_active = 1;
 981	svm->vcpu.arch.hflags = 0;
 982
 983	set_cr_intercept(svm, INTERCEPT_CR0_READ);
 984	set_cr_intercept(svm, INTERCEPT_CR3_READ);
 985	set_cr_intercept(svm, INTERCEPT_CR4_READ);
 986	set_cr_intercept(svm, INTERCEPT_CR0_WRITE);
 987	set_cr_intercept(svm, INTERCEPT_CR3_WRITE);
 988	set_cr_intercept(svm, INTERCEPT_CR4_WRITE);
 989	set_cr_intercept(svm, INTERCEPT_CR8_WRITE);
 990
 991	set_dr_intercept(svm, INTERCEPT_DR0_READ);
 992	set_dr_intercept(svm, INTERCEPT_DR1_READ);
 993	set_dr_intercept(svm, INTERCEPT_DR2_READ);
 994	set_dr_intercept(svm, INTERCEPT_DR3_READ);
 995	set_dr_intercept(svm, INTERCEPT_DR4_READ);
 996	set_dr_intercept(svm, INTERCEPT_DR5_READ);
 997	set_dr_intercept(svm, INTERCEPT_DR6_READ);
 998	set_dr_intercept(svm, INTERCEPT_DR7_READ);
 999
1000	set_dr_intercept(svm, INTERCEPT_DR0_WRITE);
1001	set_dr_intercept(svm, INTERCEPT_DR1_WRITE);
1002	set_dr_intercept(svm, INTERCEPT_DR2_WRITE);
1003	set_dr_intercept(svm, INTERCEPT_DR3_WRITE);
1004	set_dr_intercept(svm, INTERCEPT_DR4_WRITE);
1005	set_dr_intercept(svm, INTERCEPT_DR5_WRITE);
1006	set_dr_intercept(svm, INTERCEPT_DR6_WRITE);
1007	set_dr_intercept(svm, INTERCEPT_DR7_WRITE);
1008
1009	set_exception_intercept(svm, PF_VECTOR);
1010	set_exception_intercept(svm, UD_VECTOR);
1011	set_exception_intercept(svm, MC_VECTOR);
1012
1013	set_intercept(svm, INTERCEPT_INTR);
1014	set_intercept(svm, INTERCEPT_NMI);
1015	set_intercept(svm, INTERCEPT_SMI);
1016	set_intercept(svm, INTERCEPT_SELECTIVE_CR0);
 
1017	set_intercept(svm, INTERCEPT_CPUID);
1018	set_intercept(svm, INTERCEPT_INVD);
1019	set_intercept(svm, INTERCEPT_HLT);
1020	set_intercept(svm, INTERCEPT_INVLPG);
1021	set_intercept(svm, INTERCEPT_INVLPGA);
1022	set_intercept(svm, INTERCEPT_IOIO_PROT);
1023	set_intercept(svm, INTERCEPT_MSR_PROT);
1024	set_intercept(svm, INTERCEPT_TASK_SWITCH);
1025	set_intercept(svm, INTERCEPT_SHUTDOWN);
1026	set_intercept(svm, INTERCEPT_VMRUN);
1027	set_intercept(svm, INTERCEPT_VMMCALL);
1028	set_intercept(svm, INTERCEPT_VMLOAD);
1029	set_intercept(svm, INTERCEPT_VMSAVE);
1030	set_intercept(svm, INTERCEPT_STGI);
1031	set_intercept(svm, INTERCEPT_CLGI);
1032	set_intercept(svm, INTERCEPT_SKINIT);
1033	set_intercept(svm, INTERCEPT_WBINVD);
1034	set_intercept(svm, INTERCEPT_MONITOR);
1035	set_intercept(svm, INTERCEPT_MWAIT);
1036	set_intercept(svm, INTERCEPT_XSETBV);
1037
1038	control->iopm_base_pa = iopm_base;
1039	control->msrpm_base_pa = __pa(svm->msrpm);
1040	control->int_ctl = V_INTR_MASKING_MASK;
1041
1042	init_seg(&save->es);
1043	init_seg(&save->ss);
1044	init_seg(&save->ds);
1045	init_seg(&save->fs);
1046	init_seg(&save->gs);
1047
1048	save->cs.selector = 0xf000;
 
1049	/* Executable/Readable Code Segment */
1050	save->cs.attrib = SVM_SELECTOR_READ_MASK | SVM_SELECTOR_P_MASK |
1051		SVM_SELECTOR_S_MASK | SVM_SELECTOR_CODE_MASK;
1052	save->cs.limit = 0xffff;
1053	/*
1054	 * cs.base should really be 0xffff0000, but vmx can't handle that, so
1055	 * be consistent with it.
1056	 *
1057	 * Replace when we have real mode working for vmx.
1058	 */
1059	save->cs.base = 0xf0000;
1060
1061	save->gdtr.limit = 0xffff;
1062	save->idtr.limit = 0xffff;
1063
1064	init_sys_seg(&save->ldtr, SEG_TYPE_LDT);
1065	init_sys_seg(&save->tr, SEG_TYPE_BUSY_TSS16);
1066
1067	svm_set_efer(&svm->vcpu, 0);
1068	save->dr6 = 0xffff0ff0;
1069	save->dr7 = 0x400;
1070	kvm_set_rflags(&svm->vcpu, 2);
1071	save->rip = 0x0000fff0;
1072	svm->vcpu.arch.regs[VCPU_REGS_RIP] = save->rip;
1073
1074	/*
1075	 * This is the guest-visible cr0 value.
1076	 * svm_set_cr0() sets PG and WP and clears NW and CD on save->cr0.
1077	 */
1078	svm->vcpu.arch.cr0 = 0;
1079	(void)kvm_set_cr0(&svm->vcpu, X86_CR0_NW | X86_CR0_CD | X86_CR0_ET);
1080
1081	save->cr4 = X86_CR4_PAE;
1082	/* rdx = ?? */
1083
1084	if (npt_enabled) {
1085		/* Setup VMCB for Nested Paging */
1086		control->nested_ctl = 1;
1087		clr_intercept(svm, INTERCEPT_TASK_SWITCH);
1088		clr_intercept(svm, INTERCEPT_INVLPG);
1089		clr_exception_intercept(svm, PF_VECTOR);
1090		clr_cr_intercept(svm, INTERCEPT_CR3_READ);
1091		clr_cr_intercept(svm, INTERCEPT_CR3_WRITE);
1092		save->g_pat = 0x0007040600070406ULL;
1093		save->cr3 = 0;
1094		save->cr4 = 0;
1095	}
1096	svm->asid_generation = 0;
1097
1098	svm->nested.vmcb = 0;
1099	svm->vcpu.arch.hflags = 0;
1100
1101	if (boot_cpu_has(X86_FEATURE_PAUSEFILTER)) {
1102		control->pause_filter_count = 3000;
1103		set_intercept(svm, INTERCEPT_PAUSE);
1104	}
1105
1106	mark_all_dirty(svm->vmcb);
1107
1108	enable_gif(svm);
1109}
1110
1111static int svm_vcpu_reset(struct kvm_vcpu *vcpu)
1112{
1113	struct vcpu_svm *svm = to_svm(vcpu);
 
 
1114
1115	init_vmcb(svm);
1116
1117	if (!kvm_vcpu_is_bsp(vcpu)) {
1118		kvm_rip_write(vcpu, 0);
1119		svm->vmcb->save.cs.base = svm->vcpu.arch.sipi_vector << 12;
1120		svm->vmcb->save.cs.selector = svm->vcpu.arch.sipi_vector << 8;
1121	}
1122	vcpu->arch.regs_avail = ~0;
1123	vcpu->arch.regs_dirty = ~0;
1124
1125	return 0;
1126}
1127
1128static struct kvm_vcpu *svm_create_vcpu(struct kvm *kvm, unsigned int id)
1129{
1130	struct vcpu_svm *svm;
1131	struct page *page;
1132	struct page *msrpm_pages;
1133	struct page *hsave_page;
1134	struct page *nested_msrpm_pages;
1135	int err;
1136
1137	svm = kmem_cache_zalloc(kvm_vcpu_cache, GFP_KERNEL);
1138	if (!svm) {
1139		err = -ENOMEM;
1140		goto out;
1141	}
1142
1143	svm->tsc_ratio = TSC_RATIO_DEFAULT;
1144
1145	err = kvm_vcpu_init(&svm->vcpu, kvm, id);
1146	if (err)
1147		goto free_svm;
1148
1149	err = -ENOMEM;
1150	page = alloc_page(GFP_KERNEL);
1151	if (!page)
1152		goto uninit;
1153
1154	msrpm_pages = alloc_pages(GFP_KERNEL, MSRPM_ALLOC_ORDER);
1155	if (!msrpm_pages)
1156		goto free_page1;
1157
1158	nested_msrpm_pages = alloc_pages(GFP_KERNEL, MSRPM_ALLOC_ORDER);
1159	if (!nested_msrpm_pages)
1160		goto free_page2;
1161
1162	hsave_page = alloc_page(GFP_KERNEL);
1163	if (!hsave_page)
1164		goto free_page3;
1165
1166	svm->nested.hsave = page_address(hsave_page);
1167
1168	svm->msrpm = page_address(msrpm_pages);
1169	svm_vcpu_init_msrpm(svm->msrpm);
1170
1171	svm->nested.msrpm = page_address(nested_msrpm_pages);
1172	svm_vcpu_init_msrpm(svm->nested.msrpm);
1173
1174	svm->vmcb = page_address(page);
1175	clear_page(svm->vmcb);
1176	svm->vmcb_pa = page_to_pfn(page) << PAGE_SHIFT;
1177	svm->asid_generation = 0;
1178	init_vmcb(svm);
1179	kvm_write_tsc(&svm->vcpu, 0);
1180
1181	err = fx_init(&svm->vcpu);
1182	if (err)
1183		goto free_page4;
1184
1185	svm->vcpu.arch.apic_base = 0xfee00000 | MSR_IA32_APICBASE_ENABLE;
1186	if (kvm_vcpu_is_bsp(&svm->vcpu))
1187		svm->vcpu.arch.apic_base |= MSR_IA32_APICBASE_BSP;
1188
 
 
1189	return &svm->vcpu;
1190
1191free_page4:
1192	__free_page(hsave_page);
1193free_page3:
1194	__free_pages(nested_msrpm_pages, MSRPM_ALLOC_ORDER);
1195free_page2:
1196	__free_pages(msrpm_pages, MSRPM_ALLOC_ORDER);
1197free_page1:
1198	__free_page(page);
1199uninit:
1200	kvm_vcpu_uninit(&svm->vcpu);
1201free_svm:
1202	kmem_cache_free(kvm_vcpu_cache, svm);
1203out:
1204	return ERR_PTR(err);
1205}
1206
1207static void svm_free_vcpu(struct kvm_vcpu *vcpu)
1208{
1209	struct vcpu_svm *svm = to_svm(vcpu);
1210
1211	__free_page(pfn_to_page(svm->vmcb_pa >> PAGE_SHIFT));
1212	__free_pages(virt_to_page(svm->msrpm), MSRPM_ALLOC_ORDER);
1213	__free_page(virt_to_page(svm->nested.hsave));
1214	__free_pages(virt_to_page(svm->nested.msrpm), MSRPM_ALLOC_ORDER);
1215	kvm_vcpu_uninit(vcpu);
1216	kmem_cache_free(kvm_vcpu_cache, svm);
1217}
1218
1219static void svm_vcpu_load(struct kvm_vcpu *vcpu, int cpu)
1220{
1221	struct vcpu_svm *svm = to_svm(vcpu);
1222	int i;
1223
1224	if (unlikely(cpu != vcpu->cpu)) {
1225		svm->asid_generation = 0;
1226		mark_all_dirty(svm->vmcb);
1227	}
1228
1229#ifdef CONFIG_X86_64
1230	rdmsrl(MSR_GS_BASE, to_svm(vcpu)->host.gs_base);
1231#endif
1232	savesegment(fs, svm->host.fs);
1233	savesegment(gs, svm->host.gs);
1234	svm->host.ldt = kvm_read_ldt();
1235
1236	for (i = 0; i < NR_HOST_SAVE_USER_MSRS; i++)
1237		rdmsrl(host_save_user_msrs[i], svm->host_user_msrs[i]);
1238
1239	if (static_cpu_has(X86_FEATURE_TSCRATEMSR) &&
1240	    svm->tsc_ratio != __get_cpu_var(current_tsc_ratio)) {
1241		__get_cpu_var(current_tsc_ratio) = svm->tsc_ratio;
1242		wrmsrl(MSR_AMD64_TSC_RATIO, svm->tsc_ratio);
1243	}
1244}
1245
1246static void svm_vcpu_put(struct kvm_vcpu *vcpu)
1247{
1248	struct vcpu_svm *svm = to_svm(vcpu);
1249	int i;
1250
1251	++vcpu->stat.host_state_reload;
1252	kvm_load_ldt(svm->host.ldt);
1253#ifdef CONFIG_X86_64
1254	loadsegment(fs, svm->host.fs);
1255	wrmsrl(MSR_KERNEL_GS_BASE, current->thread.gs);
1256	load_gs_index(svm->host.gs);
1257#else
1258#ifdef CONFIG_X86_32_LAZY_GS
1259	loadsegment(gs, svm->host.gs);
1260#endif
1261#endif
1262	for (i = 0; i < NR_HOST_SAVE_USER_MSRS; i++)
1263		wrmsrl(host_save_user_msrs[i], svm->host_user_msrs[i]);
1264}
1265
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1266static unsigned long svm_get_rflags(struct kvm_vcpu *vcpu)
1267{
1268	return to_svm(vcpu)->vmcb->save.rflags;
1269}
1270
1271static void svm_set_rflags(struct kvm_vcpu *vcpu, unsigned long rflags)
1272{
 
 
1273	to_svm(vcpu)->vmcb->save.rflags = rflags;
 
 
1274}
1275
1276static void svm_cache_reg(struct kvm_vcpu *vcpu, enum kvm_reg reg)
1277{
1278	switch (reg) {
1279	case VCPU_EXREG_PDPTR:
1280		BUG_ON(!npt_enabled);
1281		load_pdptrs(vcpu, vcpu->arch.walk_mmu, kvm_read_cr3(vcpu));
1282		break;
1283	default:
1284		BUG();
1285	}
1286}
1287
1288static void svm_set_vintr(struct vcpu_svm *svm)
1289{
1290	set_intercept(svm, INTERCEPT_VINTR);
1291}
1292
1293static void svm_clear_vintr(struct vcpu_svm *svm)
1294{
1295	clr_intercept(svm, INTERCEPT_VINTR);
1296}
1297
1298static struct vmcb_seg *svm_seg(struct kvm_vcpu *vcpu, int seg)
1299{
1300	struct vmcb_save_area *save = &to_svm(vcpu)->vmcb->save;
1301
1302	switch (seg) {
1303	case VCPU_SREG_CS: return &save->cs;
1304	case VCPU_SREG_DS: return &save->ds;
1305	case VCPU_SREG_ES: return &save->es;
1306	case VCPU_SREG_FS: return &save->fs;
1307	case VCPU_SREG_GS: return &save->gs;
1308	case VCPU_SREG_SS: return &save->ss;
1309	case VCPU_SREG_TR: return &save->tr;
1310	case VCPU_SREG_LDTR: return &save->ldtr;
1311	}
1312	BUG();
1313	return NULL;
1314}
1315
1316static u64 svm_get_segment_base(struct kvm_vcpu *vcpu, int seg)
1317{
1318	struct vmcb_seg *s = svm_seg(vcpu, seg);
1319
1320	return s->base;
1321}
1322
1323static void svm_get_segment(struct kvm_vcpu *vcpu,
1324			    struct kvm_segment *var, int seg)
1325{
1326	struct vmcb_seg *s = svm_seg(vcpu, seg);
1327
1328	var->base = s->base;
1329	var->limit = s->limit;
1330	var->selector = s->selector;
1331	var->type = s->attrib & SVM_SELECTOR_TYPE_MASK;
1332	var->s = (s->attrib >> SVM_SELECTOR_S_SHIFT) & 1;
1333	var->dpl = (s->attrib >> SVM_SELECTOR_DPL_SHIFT) & 3;
1334	var->present = (s->attrib >> SVM_SELECTOR_P_SHIFT) & 1;
1335	var->avl = (s->attrib >> SVM_SELECTOR_AVL_SHIFT) & 1;
1336	var->l = (s->attrib >> SVM_SELECTOR_L_SHIFT) & 1;
1337	var->db = (s->attrib >> SVM_SELECTOR_DB_SHIFT) & 1;
1338	var->g = (s->attrib >> SVM_SELECTOR_G_SHIFT) & 1;
1339
1340	/*
1341	 * AMD's VMCB does not have an explicit unusable field, so emulate it
1342	 * for cross vendor migration purposes by "not present"
1343	 */
1344	var->unusable = !var->present || (var->type == 0);
1345
1346	switch (seg) {
1347	case VCPU_SREG_CS:
1348		/*
1349		 * SVM always stores 0 for the 'G' bit in the CS selector in
1350		 * the VMCB on a VMEXIT. This hurts cross-vendor migration:
1351		 * Intel's VMENTRY has a check on the 'G' bit.
1352		 */
1353		var->g = s->limit > 0xfffff;
1354		break;
1355	case VCPU_SREG_TR:
1356		/*
1357		 * Work around a bug where the busy flag in the tr selector
1358		 * isn't exposed
1359		 */
1360		var->type |= 0x2;
1361		break;
1362	case VCPU_SREG_DS:
1363	case VCPU_SREG_ES:
1364	case VCPU_SREG_FS:
1365	case VCPU_SREG_GS:
1366		/*
1367		 * The accessed bit must always be set in the segment
1368		 * descriptor cache, although it can be cleared in the
1369		 * descriptor, the cached bit always remains at 1. Since
1370		 * Intel has a check on this, set it here to support
1371		 * cross-vendor migration.
1372		 */
1373		if (!var->unusable)
1374			var->type |= 0x1;
1375		break;
1376	case VCPU_SREG_SS:
1377		/*
1378		 * On AMD CPUs sometimes the DB bit in the segment
1379		 * descriptor is left as 1, although the whole segment has
1380		 * been made unusable. Clear it here to pass an Intel VMX
1381		 * entry check when cross vendor migrating.
1382		 */
1383		if (var->unusable)
1384			var->db = 0;
1385		break;
1386	}
1387}
1388
1389static int svm_get_cpl(struct kvm_vcpu *vcpu)
1390{
1391	struct vmcb_save_area *save = &to_svm(vcpu)->vmcb->save;
1392
1393	return save->cpl;
1394}
1395
1396static void svm_get_idt(struct kvm_vcpu *vcpu, struct desc_ptr *dt)
1397{
1398	struct vcpu_svm *svm = to_svm(vcpu);
1399
1400	dt->size = svm->vmcb->save.idtr.limit;
1401	dt->address = svm->vmcb->save.idtr.base;
1402}
1403
1404static void svm_set_idt(struct kvm_vcpu *vcpu, struct desc_ptr *dt)
1405{
1406	struct vcpu_svm *svm = to_svm(vcpu);
1407
1408	svm->vmcb->save.idtr.limit = dt->size;
1409	svm->vmcb->save.idtr.base = dt->address ;
1410	mark_dirty(svm->vmcb, VMCB_DT);
1411}
1412
1413static void svm_get_gdt(struct kvm_vcpu *vcpu, struct desc_ptr *dt)
1414{
1415	struct vcpu_svm *svm = to_svm(vcpu);
1416
1417	dt->size = svm->vmcb->save.gdtr.limit;
1418	dt->address = svm->vmcb->save.gdtr.base;
1419}
1420
1421static void svm_set_gdt(struct kvm_vcpu *vcpu, struct desc_ptr *dt)
1422{
1423	struct vcpu_svm *svm = to_svm(vcpu);
1424
1425	svm->vmcb->save.gdtr.limit = dt->size;
1426	svm->vmcb->save.gdtr.base = dt->address ;
1427	mark_dirty(svm->vmcb, VMCB_DT);
1428}
1429
1430static void svm_decache_cr0_guest_bits(struct kvm_vcpu *vcpu)
1431{
1432}
1433
1434static void svm_decache_cr3(struct kvm_vcpu *vcpu)
1435{
1436}
1437
1438static void svm_decache_cr4_guest_bits(struct kvm_vcpu *vcpu)
1439{
1440}
1441
1442static void update_cr0_intercept(struct vcpu_svm *svm)
1443{
1444	ulong gcr0 = svm->vcpu.arch.cr0;
1445	u64 *hcr0 = &svm->vmcb->save.cr0;
1446
1447	if (!svm->vcpu.fpu_active)
1448		*hcr0 |= SVM_CR0_SELECTIVE_MASK;
1449	else
1450		*hcr0 = (*hcr0 & ~SVM_CR0_SELECTIVE_MASK)
1451			| (gcr0 & SVM_CR0_SELECTIVE_MASK);
1452
1453	mark_dirty(svm->vmcb, VMCB_CR);
1454
1455	if (gcr0 == *hcr0 && svm->vcpu.fpu_active) {
1456		clr_cr_intercept(svm, INTERCEPT_CR0_READ);
1457		clr_cr_intercept(svm, INTERCEPT_CR0_WRITE);
1458	} else {
1459		set_cr_intercept(svm, INTERCEPT_CR0_READ);
1460		set_cr_intercept(svm, INTERCEPT_CR0_WRITE);
1461	}
1462}
1463
1464static void svm_set_cr0(struct kvm_vcpu *vcpu, unsigned long cr0)
1465{
1466	struct vcpu_svm *svm = to_svm(vcpu);
1467
1468#ifdef CONFIG_X86_64
1469	if (vcpu->arch.efer & EFER_LME) {
1470		if (!is_paging(vcpu) && (cr0 & X86_CR0_PG)) {
1471			vcpu->arch.efer |= EFER_LMA;
1472			svm->vmcb->save.efer |= EFER_LMA | EFER_LME;
1473		}
1474
1475		if (is_paging(vcpu) && !(cr0 & X86_CR0_PG)) {
1476			vcpu->arch.efer &= ~EFER_LMA;
1477			svm->vmcb->save.efer &= ~(EFER_LMA | EFER_LME);
1478		}
1479	}
1480#endif
1481	vcpu->arch.cr0 = cr0;
1482
1483	if (!npt_enabled)
1484		cr0 |= X86_CR0_PG | X86_CR0_WP;
1485
1486	if (!vcpu->fpu_active)
1487		cr0 |= X86_CR0_TS;
1488	/*
1489	 * re-enable caching here because the QEMU bios
1490	 * does not do it - this results in some delay at
1491	 * reboot
1492	 */
1493	cr0 &= ~(X86_CR0_CD | X86_CR0_NW);
1494	svm->vmcb->save.cr0 = cr0;
1495	mark_dirty(svm->vmcb, VMCB_CR);
1496	update_cr0_intercept(svm);
1497}
1498
1499static int svm_set_cr4(struct kvm_vcpu *vcpu, unsigned long cr4)
1500{
1501	unsigned long host_cr4_mce = read_cr4() & X86_CR4_MCE;
1502	unsigned long old_cr4 = to_svm(vcpu)->vmcb->save.cr4;
1503
1504	if (cr4 & X86_CR4_VMXE)
1505		return 1;
1506
1507	if (npt_enabled && ((old_cr4 ^ cr4) & X86_CR4_PGE))
1508		svm_flush_tlb(vcpu);
1509
1510	vcpu->arch.cr4 = cr4;
1511	if (!npt_enabled)
1512		cr4 |= X86_CR4_PAE;
1513	cr4 |= host_cr4_mce;
1514	to_svm(vcpu)->vmcb->save.cr4 = cr4;
1515	mark_dirty(to_svm(vcpu)->vmcb, VMCB_CR);
1516	return 0;
1517}
1518
1519static void svm_set_segment(struct kvm_vcpu *vcpu,
1520			    struct kvm_segment *var, int seg)
1521{
1522	struct vcpu_svm *svm = to_svm(vcpu);
1523	struct vmcb_seg *s = svm_seg(vcpu, seg);
1524
1525	s->base = var->base;
1526	s->limit = var->limit;
1527	s->selector = var->selector;
1528	if (var->unusable)
1529		s->attrib = 0;
1530	else {
1531		s->attrib = (var->type & SVM_SELECTOR_TYPE_MASK);
1532		s->attrib |= (var->s & 1) << SVM_SELECTOR_S_SHIFT;
1533		s->attrib |= (var->dpl & 3) << SVM_SELECTOR_DPL_SHIFT;
1534		s->attrib |= (var->present & 1) << SVM_SELECTOR_P_SHIFT;
1535		s->attrib |= (var->avl & 1) << SVM_SELECTOR_AVL_SHIFT;
1536		s->attrib |= (var->l & 1) << SVM_SELECTOR_L_SHIFT;
1537		s->attrib |= (var->db & 1) << SVM_SELECTOR_DB_SHIFT;
1538		s->attrib |= (var->g & 1) << SVM_SELECTOR_G_SHIFT;
1539	}
1540	if (seg == VCPU_SREG_CS)
1541		svm->vmcb->save.cpl
1542			= (svm->vmcb->save.cs.attrib
1543			   >> SVM_SELECTOR_DPL_SHIFT) & 3;
1544
1545	mark_dirty(svm->vmcb, VMCB_SEG);
1546}
1547
1548static void update_db_intercept(struct kvm_vcpu *vcpu)
1549{
1550	struct vcpu_svm *svm = to_svm(vcpu);
1551
1552	clr_exception_intercept(svm, DB_VECTOR);
1553	clr_exception_intercept(svm, BP_VECTOR);
1554
1555	if (svm->nmi_singlestep)
1556		set_exception_intercept(svm, DB_VECTOR);
1557
1558	if (vcpu->guest_debug & KVM_GUESTDBG_ENABLE) {
1559		if (vcpu->guest_debug &
1560		    (KVM_GUESTDBG_SINGLESTEP | KVM_GUESTDBG_USE_HW_BP))
1561			set_exception_intercept(svm, DB_VECTOR);
1562		if (vcpu->guest_debug & KVM_GUESTDBG_USE_SW_BP)
1563			set_exception_intercept(svm, BP_VECTOR);
1564	} else
1565		vcpu->guest_debug = 0;
1566}
1567
1568static void svm_guest_debug(struct kvm_vcpu *vcpu, struct kvm_guest_debug *dbg)
1569{
1570	struct vcpu_svm *svm = to_svm(vcpu);
1571
1572	if (vcpu->guest_debug & KVM_GUESTDBG_USE_HW_BP)
1573		svm->vmcb->save.dr7 = dbg->arch.debugreg[7];
1574	else
1575		svm->vmcb->save.dr7 = vcpu->arch.dr7;
1576
1577	mark_dirty(svm->vmcb, VMCB_DR);
1578
1579	update_db_intercept(vcpu);
1580}
1581
1582static void new_asid(struct vcpu_svm *svm, struct svm_cpu_data *sd)
1583{
1584	if (sd->next_asid > sd->max_asid) {
1585		++sd->asid_generation;
1586		sd->next_asid = 1;
1587		svm->vmcb->control.tlb_ctl = TLB_CONTROL_FLUSH_ALL_ASID;
1588	}
1589
1590	svm->asid_generation = sd->asid_generation;
1591	svm->vmcb->control.asid = sd->next_asid++;
1592
1593	mark_dirty(svm->vmcb, VMCB_ASID);
1594}
1595
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1596static void svm_set_dr7(struct kvm_vcpu *vcpu, unsigned long value)
1597{
1598	struct vcpu_svm *svm = to_svm(vcpu);
1599
1600	svm->vmcb->save.dr7 = value;
1601	mark_dirty(svm->vmcb, VMCB_DR);
1602}
1603
1604static int pf_interception(struct vcpu_svm *svm)
1605{
1606	u64 fault_address = svm->vmcb->control.exit_info_2;
1607	u32 error_code;
1608	int r = 1;
1609
1610	switch (svm->apf_reason) {
1611	default:
1612		error_code = svm->vmcb->control.exit_info_1;
1613
1614		trace_kvm_page_fault(fault_address, error_code);
1615		if (!npt_enabled && kvm_event_needs_reinjection(&svm->vcpu))
1616			kvm_mmu_unprotect_page_virt(&svm->vcpu, fault_address);
1617		r = kvm_mmu_page_fault(&svm->vcpu, fault_address, error_code,
1618			svm->vmcb->control.insn_bytes,
1619			svm->vmcb->control.insn_len);
1620		break;
1621	case KVM_PV_REASON_PAGE_NOT_PRESENT:
1622		svm->apf_reason = 0;
1623		local_irq_disable();
1624		kvm_async_pf_task_wait(fault_address);
1625		local_irq_enable();
1626		break;
1627	case KVM_PV_REASON_PAGE_READY:
1628		svm->apf_reason = 0;
1629		local_irq_disable();
1630		kvm_async_pf_task_wake(fault_address);
1631		local_irq_enable();
1632		break;
1633	}
1634	return r;
1635}
1636
1637static int db_interception(struct vcpu_svm *svm)
1638{
1639	struct kvm_run *kvm_run = svm->vcpu.run;
1640
1641	if (!(svm->vcpu.guest_debug &
1642	      (KVM_GUESTDBG_SINGLESTEP | KVM_GUESTDBG_USE_HW_BP)) &&
1643		!svm->nmi_singlestep) {
1644		kvm_queue_exception(&svm->vcpu, DB_VECTOR);
1645		return 1;
1646	}
1647
1648	if (svm->nmi_singlestep) {
1649		svm->nmi_singlestep = false;
1650		if (!(svm->vcpu.guest_debug & KVM_GUESTDBG_SINGLESTEP))
1651			svm->vmcb->save.rflags &=
1652				~(X86_EFLAGS_TF | X86_EFLAGS_RF);
1653		update_db_intercept(&svm->vcpu);
1654	}
1655
1656	if (svm->vcpu.guest_debug &
1657	    (KVM_GUESTDBG_SINGLESTEP | KVM_GUESTDBG_USE_HW_BP)) {
1658		kvm_run->exit_reason = KVM_EXIT_DEBUG;
1659		kvm_run->debug.arch.pc =
1660			svm->vmcb->save.cs.base + svm->vmcb->save.rip;
1661		kvm_run->debug.arch.exception = DB_VECTOR;
1662		return 0;
1663	}
1664
1665	return 1;
1666}
1667
1668static int bp_interception(struct vcpu_svm *svm)
1669{
1670	struct kvm_run *kvm_run = svm->vcpu.run;
1671
1672	kvm_run->exit_reason = KVM_EXIT_DEBUG;
1673	kvm_run->debug.arch.pc = svm->vmcb->save.cs.base + svm->vmcb->save.rip;
1674	kvm_run->debug.arch.exception = BP_VECTOR;
1675	return 0;
1676}
1677
1678static int ud_interception(struct vcpu_svm *svm)
1679{
1680	int er;
1681
1682	er = emulate_instruction(&svm->vcpu, EMULTYPE_TRAP_UD);
1683	if (er != EMULATE_DONE)
1684		kvm_queue_exception(&svm->vcpu, UD_VECTOR);
1685	return 1;
1686}
1687
1688static void svm_fpu_activate(struct kvm_vcpu *vcpu)
1689{
1690	struct vcpu_svm *svm = to_svm(vcpu);
1691
1692	clr_exception_intercept(svm, NM_VECTOR);
1693
1694	svm->vcpu.fpu_active = 1;
1695	update_cr0_intercept(svm);
1696}
1697
1698static int nm_interception(struct vcpu_svm *svm)
1699{
1700	svm_fpu_activate(&svm->vcpu);
1701	return 1;
1702}
1703
1704static bool is_erratum_383(void)
1705{
1706	int err, i;
1707	u64 value;
1708
1709	if (!erratum_383_found)
1710		return false;
1711
1712	value = native_read_msr_safe(MSR_IA32_MC0_STATUS, &err);
1713	if (err)
1714		return false;
1715
1716	/* Bit 62 may or may not be set for this mce */
1717	value &= ~(1ULL << 62);
1718
1719	if (value != 0xb600000000010015ULL)
1720		return false;
1721
1722	/* Clear MCi_STATUS registers */
1723	for (i = 0; i < 6; ++i)
1724		native_write_msr_safe(MSR_IA32_MCx_STATUS(i), 0, 0);
1725
1726	value = native_read_msr_safe(MSR_IA32_MCG_STATUS, &err);
1727	if (!err) {
1728		u32 low, high;
1729
1730		value &= ~(1ULL << 2);
1731		low    = lower_32_bits(value);
1732		high   = upper_32_bits(value);
1733
1734		native_write_msr_safe(MSR_IA32_MCG_STATUS, low, high);
1735	}
1736
1737	/* Flush tlb to evict multi-match entries */
1738	__flush_tlb_all();
1739
1740	return true;
1741}
1742
1743static void svm_handle_mce(struct vcpu_svm *svm)
1744{
1745	if (is_erratum_383()) {
1746		/*
1747		 * Erratum 383 triggered. Guest state is corrupt so kill the
1748		 * guest.
1749		 */
1750		pr_err("KVM: Guest triggered AMD Erratum 383\n");
1751
1752		kvm_make_request(KVM_REQ_TRIPLE_FAULT, &svm->vcpu);
1753
1754		return;
1755	}
1756
1757	/*
1758	 * On an #MC intercept the MCE handler is not called automatically in
1759	 * the host. So do it by hand here.
1760	 */
1761	asm volatile (
1762		"int $0x12\n");
1763	/* not sure if we ever come back to this point */
1764
1765	return;
1766}
1767
1768static int mc_interception(struct vcpu_svm *svm)
1769{
1770	return 1;
1771}
1772
1773static int shutdown_interception(struct vcpu_svm *svm)
1774{
1775	struct kvm_run *kvm_run = svm->vcpu.run;
1776
1777	/*
1778	 * VMCB is undefined after a SHUTDOWN intercept
1779	 * so reinitialize it.
1780	 */
1781	clear_page(svm->vmcb);
1782	init_vmcb(svm);
1783
1784	kvm_run->exit_reason = KVM_EXIT_SHUTDOWN;
1785	return 0;
1786}
1787
1788static int io_interception(struct vcpu_svm *svm)
1789{
1790	struct kvm_vcpu *vcpu = &svm->vcpu;
1791	u32 io_info = svm->vmcb->control.exit_info_1; /* address size bug? */
1792	int size, in, string;
1793	unsigned port;
1794
1795	++svm->vcpu.stat.io_exits;
1796	string = (io_info & SVM_IOIO_STR_MASK) != 0;
1797	in = (io_info & SVM_IOIO_TYPE_MASK) != 0;
1798	if (string || in)
1799		return emulate_instruction(vcpu, 0) == EMULATE_DONE;
1800
1801	port = io_info >> 16;
1802	size = (io_info & SVM_IOIO_SIZE_MASK) >> SVM_IOIO_SIZE_SHIFT;
1803	svm->next_rip = svm->vmcb->control.exit_info_2;
1804	skip_emulated_instruction(&svm->vcpu);
1805
1806	return kvm_fast_pio_out(vcpu, size, port);
1807}
1808
1809static int nmi_interception(struct vcpu_svm *svm)
1810{
1811	return 1;
1812}
1813
1814static int intr_interception(struct vcpu_svm *svm)
1815{
1816	++svm->vcpu.stat.irq_exits;
1817	return 1;
1818}
1819
1820static int nop_on_interception(struct vcpu_svm *svm)
1821{
1822	return 1;
1823}
1824
1825static int halt_interception(struct vcpu_svm *svm)
1826{
1827	svm->next_rip = kvm_rip_read(&svm->vcpu) + 1;
1828	skip_emulated_instruction(&svm->vcpu);
1829	return kvm_emulate_halt(&svm->vcpu);
1830}
1831
1832static int vmmcall_interception(struct vcpu_svm *svm)
1833{
1834	svm->next_rip = kvm_rip_read(&svm->vcpu) + 3;
1835	skip_emulated_instruction(&svm->vcpu);
1836	kvm_emulate_hypercall(&svm->vcpu);
1837	return 1;
1838}
1839
1840static unsigned long nested_svm_get_tdp_cr3(struct kvm_vcpu *vcpu)
1841{
1842	struct vcpu_svm *svm = to_svm(vcpu);
1843
1844	return svm->nested.nested_cr3;
1845}
1846
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1847static void nested_svm_set_tdp_cr3(struct kvm_vcpu *vcpu,
1848				   unsigned long root)
1849{
1850	struct vcpu_svm *svm = to_svm(vcpu);
1851
1852	svm->vmcb->control.nested_cr3 = root;
1853	mark_dirty(svm->vmcb, VMCB_NPT);
1854	svm_flush_tlb(vcpu);
1855}
1856
1857static void nested_svm_inject_npf_exit(struct kvm_vcpu *vcpu,
1858				       struct x86_exception *fault)
1859{
1860	struct vcpu_svm *svm = to_svm(vcpu);
1861
1862	svm->vmcb->control.exit_code = SVM_EXIT_NPF;
1863	svm->vmcb->control.exit_code_hi = 0;
1864	svm->vmcb->control.exit_info_1 = fault->error_code;
1865	svm->vmcb->control.exit_info_2 = fault->address;
1866
1867	nested_svm_vmexit(svm);
1868}
1869
1870static int nested_svm_init_mmu_context(struct kvm_vcpu *vcpu)
1871{
1872	int r;
1873
1874	r = kvm_init_shadow_mmu(vcpu, &vcpu->arch.mmu);
1875
1876	vcpu->arch.mmu.set_cr3           = nested_svm_set_tdp_cr3;
1877	vcpu->arch.mmu.get_cr3           = nested_svm_get_tdp_cr3;
 
1878	vcpu->arch.mmu.inject_page_fault = nested_svm_inject_npf_exit;
1879	vcpu->arch.mmu.shadow_root_level = get_npt_level();
1880	vcpu->arch.walk_mmu              = &vcpu->arch.nested_mmu;
1881
1882	return r;
1883}
1884
1885static void nested_svm_uninit_mmu_context(struct kvm_vcpu *vcpu)
1886{
1887	vcpu->arch.walk_mmu = &vcpu->arch.mmu;
1888}
1889
1890static int nested_svm_check_permissions(struct vcpu_svm *svm)
1891{
1892	if (!(svm->vcpu.arch.efer & EFER_SVME)
1893	    || !is_paging(&svm->vcpu)) {
1894		kvm_queue_exception(&svm->vcpu, UD_VECTOR);
1895		return 1;
1896	}
1897
1898	if (svm->vmcb->save.cpl) {
1899		kvm_inject_gp(&svm->vcpu, 0);
1900		return 1;
1901	}
1902
1903       return 0;
1904}
1905
1906static int nested_svm_check_exception(struct vcpu_svm *svm, unsigned nr,
1907				      bool has_error_code, u32 error_code)
1908{
1909	int vmexit;
1910
1911	if (!is_guest_mode(&svm->vcpu))
1912		return 0;
1913
1914	svm->vmcb->control.exit_code = SVM_EXIT_EXCP_BASE + nr;
1915	svm->vmcb->control.exit_code_hi = 0;
1916	svm->vmcb->control.exit_info_1 = error_code;
1917	svm->vmcb->control.exit_info_2 = svm->vcpu.arch.cr2;
1918
1919	vmexit = nested_svm_intercept(svm);
1920	if (vmexit == NESTED_EXIT_DONE)
1921		svm->nested.exit_required = true;
1922
1923	return vmexit;
1924}
1925
1926/* This function returns true if it is save to enable the irq window */
1927static inline bool nested_svm_intr(struct vcpu_svm *svm)
1928{
1929	if (!is_guest_mode(&svm->vcpu))
1930		return true;
1931
1932	if (!(svm->vcpu.arch.hflags & HF_VINTR_MASK))
1933		return true;
1934
1935	if (!(svm->vcpu.arch.hflags & HF_HIF_MASK))
1936		return false;
1937
1938	/*
1939	 * if vmexit was already requested (by intercepted exception
1940	 * for instance) do not overwrite it with "external interrupt"
1941	 * vmexit.
1942	 */
1943	if (svm->nested.exit_required)
1944		return false;
1945
1946	svm->vmcb->control.exit_code   = SVM_EXIT_INTR;
1947	svm->vmcb->control.exit_info_1 = 0;
1948	svm->vmcb->control.exit_info_2 = 0;
1949
1950	if (svm->nested.intercept & 1ULL) {
1951		/*
1952		 * The #vmexit can't be emulated here directly because this
1953		 * code path runs with irqs and preemtion disabled. A
1954		 * #vmexit emulation might sleep. Only signal request for
1955		 * the #vmexit here.
1956		 */
1957		svm->nested.exit_required = true;
1958		trace_kvm_nested_intr_vmexit(svm->vmcb->save.rip);
1959		return false;
1960	}
1961
1962	return true;
1963}
1964
1965/* This function returns true if it is save to enable the nmi window */
1966static inline bool nested_svm_nmi(struct vcpu_svm *svm)
1967{
1968	if (!is_guest_mode(&svm->vcpu))
1969		return true;
1970
1971	if (!(svm->nested.intercept & (1ULL << INTERCEPT_NMI)))
1972		return true;
1973
1974	svm->vmcb->control.exit_code = SVM_EXIT_NMI;
1975	svm->nested.exit_required = true;
1976
1977	return false;
1978}
1979
1980static void *nested_svm_map(struct vcpu_svm *svm, u64 gpa, struct page **_page)
1981{
1982	struct page *page;
1983
1984	might_sleep();
1985
1986	page = gfn_to_page(svm->vcpu.kvm, gpa >> PAGE_SHIFT);
1987	if (is_error_page(page))
1988		goto error;
1989
1990	*_page = page;
1991
1992	return kmap(page);
1993
1994error:
1995	kvm_release_page_clean(page);
1996	kvm_inject_gp(&svm->vcpu, 0);
1997
1998	return NULL;
1999}
2000
2001static void nested_svm_unmap(struct page *page)
2002{
2003	kunmap(page);
2004	kvm_release_page_dirty(page);
2005}
2006
2007static int nested_svm_intercept_ioio(struct vcpu_svm *svm)
2008{
2009	unsigned port;
2010	u8 val, bit;
2011	u64 gpa;
2012
2013	if (!(svm->nested.intercept & (1ULL << INTERCEPT_IOIO_PROT)))
2014		return NESTED_EXIT_HOST;
2015
2016	port = svm->vmcb->control.exit_info_1 >> 16;
2017	gpa  = svm->nested.vmcb_iopm + (port / 8);
2018	bit  = port % 8;
2019	val  = 0;
2020
2021	if (kvm_read_guest(svm->vcpu.kvm, gpa, &val, 1))
2022		val &= (1 << bit);
2023
2024	return val ? NESTED_EXIT_DONE : NESTED_EXIT_HOST;
2025}
2026
2027static int nested_svm_exit_handled_msr(struct vcpu_svm *svm)
2028{
2029	u32 offset, msr, value;
2030	int write, mask;
2031
2032	if (!(svm->nested.intercept & (1ULL << INTERCEPT_MSR_PROT)))
2033		return NESTED_EXIT_HOST;
2034
2035	msr    = svm->vcpu.arch.regs[VCPU_REGS_RCX];
2036	offset = svm_msrpm_offset(msr);
2037	write  = svm->vmcb->control.exit_info_1 & 1;
2038	mask   = 1 << ((2 * (msr & 0xf)) + write);
2039
2040	if (offset == MSR_INVALID)
2041		return NESTED_EXIT_DONE;
2042
2043	/* Offset is in 32 bit units but need in 8 bit units */
2044	offset *= 4;
2045
2046	if (kvm_read_guest(svm->vcpu.kvm, svm->nested.vmcb_msrpm + offset, &value, 4))
2047		return NESTED_EXIT_DONE;
2048
2049	return (value & mask) ? NESTED_EXIT_DONE : NESTED_EXIT_HOST;
2050}
2051
2052static int nested_svm_exit_special(struct vcpu_svm *svm)
2053{
2054	u32 exit_code = svm->vmcb->control.exit_code;
2055
2056	switch (exit_code) {
2057	case SVM_EXIT_INTR:
2058	case SVM_EXIT_NMI:
2059	case SVM_EXIT_EXCP_BASE + MC_VECTOR:
2060		return NESTED_EXIT_HOST;
2061	case SVM_EXIT_NPF:
2062		/* For now we are always handling NPFs when using them */
2063		if (npt_enabled)
2064			return NESTED_EXIT_HOST;
2065		break;
2066	case SVM_EXIT_EXCP_BASE + PF_VECTOR:
2067		/* When we're shadowing, trap PFs, but not async PF */
2068		if (!npt_enabled && svm->apf_reason == 0)
2069			return NESTED_EXIT_HOST;
2070		break;
2071	case SVM_EXIT_EXCP_BASE + NM_VECTOR:
2072		nm_interception(svm);
2073		break;
2074	default:
2075		break;
2076	}
2077
2078	return NESTED_EXIT_CONTINUE;
2079}
2080
2081/*
2082 * If this function returns true, this #vmexit was already handled
2083 */
2084static int nested_svm_intercept(struct vcpu_svm *svm)
2085{
2086	u32 exit_code = svm->vmcb->control.exit_code;
2087	int vmexit = NESTED_EXIT_HOST;
2088
2089	switch (exit_code) {
2090	case SVM_EXIT_MSR:
2091		vmexit = nested_svm_exit_handled_msr(svm);
2092		break;
2093	case SVM_EXIT_IOIO:
2094		vmexit = nested_svm_intercept_ioio(svm);
2095		break;
2096	case SVM_EXIT_READ_CR0 ... SVM_EXIT_WRITE_CR8: {
2097		u32 bit = 1U << (exit_code - SVM_EXIT_READ_CR0);
2098		if (svm->nested.intercept_cr & bit)
2099			vmexit = NESTED_EXIT_DONE;
2100		break;
2101	}
2102	case SVM_EXIT_READ_DR0 ... SVM_EXIT_WRITE_DR7: {
2103		u32 bit = 1U << (exit_code - SVM_EXIT_READ_DR0);
2104		if (svm->nested.intercept_dr & bit)
2105			vmexit = NESTED_EXIT_DONE;
2106		break;
2107	}
2108	case SVM_EXIT_EXCP_BASE ... SVM_EXIT_EXCP_BASE + 0x1f: {
2109		u32 excp_bits = 1 << (exit_code - SVM_EXIT_EXCP_BASE);
2110		if (svm->nested.intercept_exceptions & excp_bits)
2111			vmexit = NESTED_EXIT_DONE;
2112		/* async page fault always cause vmexit */
2113		else if ((exit_code == SVM_EXIT_EXCP_BASE + PF_VECTOR) &&
2114			 svm->apf_reason != 0)
2115			vmexit = NESTED_EXIT_DONE;
2116		break;
2117	}
2118	case SVM_EXIT_ERR: {
2119		vmexit = NESTED_EXIT_DONE;
2120		break;
2121	}
2122	default: {
2123		u64 exit_bits = 1ULL << (exit_code - SVM_EXIT_INTR);
2124		if (svm->nested.intercept & exit_bits)
2125			vmexit = NESTED_EXIT_DONE;
2126	}
2127	}
2128
2129	return vmexit;
2130}
2131
2132static int nested_svm_exit_handled(struct vcpu_svm *svm)
2133{
2134	int vmexit;
2135
2136	vmexit = nested_svm_intercept(svm);
2137
2138	if (vmexit == NESTED_EXIT_DONE)
2139		nested_svm_vmexit(svm);
2140
2141	return vmexit;
2142}
2143
2144static inline void copy_vmcb_control_area(struct vmcb *dst_vmcb, struct vmcb *from_vmcb)
2145{
2146	struct vmcb_control_area *dst  = &dst_vmcb->control;
2147	struct vmcb_control_area *from = &from_vmcb->control;
2148
2149	dst->intercept_cr         = from->intercept_cr;
2150	dst->intercept_dr         = from->intercept_dr;
2151	dst->intercept_exceptions = from->intercept_exceptions;
2152	dst->intercept            = from->intercept;
2153	dst->iopm_base_pa         = from->iopm_base_pa;
2154	dst->msrpm_base_pa        = from->msrpm_base_pa;
2155	dst->tsc_offset           = from->tsc_offset;
2156	dst->asid                 = from->asid;
2157	dst->tlb_ctl              = from->tlb_ctl;
2158	dst->int_ctl              = from->int_ctl;
2159	dst->int_vector           = from->int_vector;
2160	dst->int_state            = from->int_state;
2161	dst->exit_code            = from->exit_code;
2162	dst->exit_code_hi         = from->exit_code_hi;
2163	dst->exit_info_1          = from->exit_info_1;
2164	dst->exit_info_2          = from->exit_info_2;
2165	dst->exit_int_info        = from->exit_int_info;
2166	dst->exit_int_info_err    = from->exit_int_info_err;
2167	dst->nested_ctl           = from->nested_ctl;
2168	dst->event_inj            = from->event_inj;
2169	dst->event_inj_err        = from->event_inj_err;
2170	dst->nested_cr3           = from->nested_cr3;
2171	dst->lbr_ctl              = from->lbr_ctl;
2172}
2173
2174static int nested_svm_vmexit(struct vcpu_svm *svm)
2175{
2176	struct vmcb *nested_vmcb;
2177	struct vmcb *hsave = svm->nested.hsave;
2178	struct vmcb *vmcb = svm->vmcb;
2179	struct page *page;
2180
2181	trace_kvm_nested_vmexit_inject(vmcb->control.exit_code,
2182				       vmcb->control.exit_info_1,
2183				       vmcb->control.exit_info_2,
2184				       vmcb->control.exit_int_info,
2185				       vmcb->control.exit_int_info_err);
 
2186
2187	nested_vmcb = nested_svm_map(svm, svm->nested.vmcb, &page);
2188	if (!nested_vmcb)
2189		return 1;
2190
2191	/* Exit Guest-Mode */
2192	leave_guest_mode(&svm->vcpu);
2193	svm->nested.vmcb = 0;
2194
2195	/* Give the current vmcb to the guest */
2196	disable_gif(svm);
2197
2198	nested_vmcb->save.es     = vmcb->save.es;
2199	nested_vmcb->save.cs     = vmcb->save.cs;
2200	nested_vmcb->save.ss     = vmcb->save.ss;
2201	nested_vmcb->save.ds     = vmcb->save.ds;
2202	nested_vmcb->save.gdtr   = vmcb->save.gdtr;
2203	nested_vmcb->save.idtr   = vmcb->save.idtr;
2204	nested_vmcb->save.efer   = svm->vcpu.arch.efer;
2205	nested_vmcb->save.cr0    = kvm_read_cr0(&svm->vcpu);
2206	nested_vmcb->save.cr3    = kvm_read_cr3(&svm->vcpu);
2207	nested_vmcb->save.cr2    = vmcb->save.cr2;
2208	nested_vmcb->save.cr4    = svm->vcpu.arch.cr4;
2209	nested_vmcb->save.rflags = kvm_get_rflags(&svm->vcpu);
2210	nested_vmcb->save.rip    = vmcb->save.rip;
2211	nested_vmcb->save.rsp    = vmcb->save.rsp;
2212	nested_vmcb->save.rax    = vmcb->save.rax;
2213	nested_vmcb->save.dr7    = vmcb->save.dr7;
2214	nested_vmcb->save.dr6    = vmcb->save.dr6;
2215	nested_vmcb->save.cpl    = vmcb->save.cpl;
2216
2217	nested_vmcb->control.int_ctl           = vmcb->control.int_ctl;
2218	nested_vmcb->control.int_vector        = vmcb->control.int_vector;
2219	nested_vmcb->control.int_state         = vmcb->control.int_state;
2220	nested_vmcb->control.exit_code         = vmcb->control.exit_code;
2221	nested_vmcb->control.exit_code_hi      = vmcb->control.exit_code_hi;
2222	nested_vmcb->control.exit_info_1       = vmcb->control.exit_info_1;
2223	nested_vmcb->control.exit_info_2       = vmcb->control.exit_info_2;
2224	nested_vmcb->control.exit_int_info     = vmcb->control.exit_int_info;
2225	nested_vmcb->control.exit_int_info_err = vmcb->control.exit_int_info_err;
2226	nested_vmcb->control.next_rip          = vmcb->control.next_rip;
2227
2228	/*
2229	 * If we emulate a VMRUN/#VMEXIT in the same host #vmexit cycle we have
2230	 * to make sure that we do not lose injected events. So check event_inj
2231	 * here and copy it to exit_int_info if it is valid.
2232	 * Exit_int_info and event_inj can't be both valid because the case
2233	 * below only happens on a VMRUN instruction intercept which has
2234	 * no valid exit_int_info set.
2235	 */
2236	if (vmcb->control.event_inj & SVM_EVTINJ_VALID) {
2237		struct vmcb_control_area *nc = &nested_vmcb->control;
2238
2239		nc->exit_int_info     = vmcb->control.event_inj;
2240		nc->exit_int_info_err = vmcb->control.event_inj_err;
2241	}
2242
2243	nested_vmcb->control.tlb_ctl           = 0;
2244	nested_vmcb->control.event_inj         = 0;
2245	nested_vmcb->control.event_inj_err     = 0;
2246
2247	/* We always set V_INTR_MASKING and remember the old value in hflags */
2248	if (!(svm->vcpu.arch.hflags & HF_VINTR_MASK))
2249		nested_vmcb->control.int_ctl &= ~V_INTR_MASKING_MASK;
2250
2251	/* Restore the original control entries */
2252	copy_vmcb_control_area(vmcb, hsave);
2253
2254	kvm_clear_exception_queue(&svm->vcpu);
2255	kvm_clear_interrupt_queue(&svm->vcpu);
2256
2257	svm->nested.nested_cr3 = 0;
2258
2259	/* Restore selected save entries */
2260	svm->vmcb->save.es = hsave->save.es;
2261	svm->vmcb->save.cs = hsave->save.cs;
2262	svm->vmcb->save.ss = hsave->save.ss;
2263	svm->vmcb->save.ds = hsave->save.ds;
2264	svm->vmcb->save.gdtr = hsave->save.gdtr;
2265	svm->vmcb->save.idtr = hsave->save.idtr;
2266	kvm_set_rflags(&svm->vcpu, hsave->save.rflags);
2267	svm_set_efer(&svm->vcpu, hsave->save.efer);
2268	svm_set_cr0(&svm->vcpu, hsave->save.cr0 | X86_CR0_PE);
2269	svm_set_cr4(&svm->vcpu, hsave->save.cr4);
2270	if (npt_enabled) {
2271		svm->vmcb->save.cr3 = hsave->save.cr3;
2272		svm->vcpu.arch.cr3 = hsave->save.cr3;
2273	} else {
2274		(void)kvm_set_cr3(&svm->vcpu, hsave->save.cr3);
2275	}
2276	kvm_register_write(&svm->vcpu, VCPU_REGS_RAX, hsave->save.rax);
2277	kvm_register_write(&svm->vcpu, VCPU_REGS_RSP, hsave->save.rsp);
2278	kvm_register_write(&svm->vcpu, VCPU_REGS_RIP, hsave->save.rip);
2279	svm->vmcb->save.dr7 = 0;
2280	svm->vmcb->save.cpl = 0;
2281	svm->vmcb->control.exit_int_info = 0;
2282
2283	mark_all_dirty(svm->vmcb);
2284
2285	nested_svm_unmap(page);
2286
2287	nested_svm_uninit_mmu_context(&svm->vcpu);
2288	kvm_mmu_reset_context(&svm->vcpu);
2289	kvm_mmu_load(&svm->vcpu);
2290
2291	return 0;
2292}
2293
2294static bool nested_svm_vmrun_msrpm(struct vcpu_svm *svm)
2295{
2296	/*
2297	 * This function merges the msr permission bitmaps of kvm and the
2298	 * nested vmcb. It is omptimized in that it only merges the parts where
2299	 * the kvm msr permission bitmap may contain zero bits
2300	 */
2301	int i;
2302
2303	if (!(svm->nested.intercept & (1ULL << INTERCEPT_MSR_PROT)))
2304		return true;
2305
2306	for (i = 0; i < MSRPM_OFFSETS; i++) {
2307		u32 value, p;
2308		u64 offset;
2309
2310		if (msrpm_offsets[i] == 0xffffffff)
2311			break;
2312
2313		p      = msrpm_offsets[i];
2314		offset = svm->nested.vmcb_msrpm + (p * 4);
2315
2316		if (kvm_read_guest(svm->vcpu.kvm, offset, &value, 4))
2317			return false;
2318
2319		svm->nested.msrpm[p] = svm->msrpm[p] | value;
2320	}
2321
2322	svm->vmcb->control.msrpm_base_pa = __pa(svm->nested.msrpm);
2323
2324	return true;
2325}
2326
2327static bool nested_vmcb_checks(struct vmcb *vmcb)
2328{
2329	if ((vmcb->control.intercept & (1ULL << INTERCEPT_VMRUN)) == 0)
2330		return false;
2331
2332	if (vmcb->control.asid == 0)
2333		return false;
2334
2335	if (vmcb->control.nested_ctl && !npt_enabled)
2336		return false;
2337
2338	return true;
2339}
2340
2341static bool nested_svm_vmrun(struct vcpu_svm *svm)
2342{
2343	struct vmcb *nested_vmcb;
2344	struct vmcb *hsave = svm->nested.hsave;
2345	struct vmcb *vmcb = svm->vmcb;
2346	struct page *page;
2347	u64 vmcb_gpa;
2348
2349	vmcb_gpa = svm->vmcb->save.rax;
2350
2351	nested_vmcb = nested_svm_map(svm, svm->vmcb->save.rax, &page);
2352	if (!nested_vmcb)
2353		return false;
2354
2355	if (!nested_vmcb_checks(nested_vmcb)) {
2356		nested_vmcb->control.exit_code    = SVM_EXIT_ERR;
2357		nested_vmcb->control.exit_code_hi = 0;
2358		nested_vmcb->control.exit_info_1  = 0;
2359		nested_vmcb->control.exit_info_2  = 0;
2360
2361		nested_svm_unmap(page);
2362
2363		return false;
2364	}
2365
2366	trace_kvm_nested_vmrun(svm->vmcb->save.rip, vmcb_gpa,
2367			       nested_vmcb->save.rip,
2368			       nested_vmcb->control.int_ctl,
2369			       nested_vmcb->control.event_inj,
2370			       nested_vmcb->control.nested_ctl);
2371
2372	trace_kvm_nested_intercepts(nested_vmcb->control.intercept_cr & 0xffff,
2373				    nested_vmcb->control.intercept_cr >> 16,
2374				    nested_vmcb->control.intercept_exceptions,
2375				    nested_vmcb->control.intercept);
2376
2377	/* Clear internal status */
2378	kvm_clear_exception_queue(&svm->vcpu);
2379	kvm_clear_interrupt_queue(&svm->vcpu);
2380
2381	/*
2382	 * Save the old vmcb, so we don't need to pick what we save, but can
2383	 * restore everything when a VMEXIT occurs
2384	 */
2385	hsave->save.es     = vmcb->save.es;
2386	hsave->save.cs     = vmcb->save.cs;
2387	hsave->save.ss     = vmcb->save.ss;
2388	hsave->save.ds     = vmcb->save.ds;
2389	hsave->save.gdtr   = vmcb->save.gdtr;
2390	hsave->save.idtr   = vmcb->save.idtr;
2391	hsave->save.efer   = svm->vcpu.arch.efer;
2392	hsave->save.cr0    = kvm_read_cr0(&svm->vcpu);
2393	hsave->save.cr4    = svm->vcpu.arch.cr4;
2394	hsave->save.rflags = kvm_get_rflags(&svm->vcpu);
2395	hsave->save.rip    = kvm_rip_read(&svm->vcpu);
2396	hsave->save.rsp    = vmcb->save.rsp;
2397	hsave->save.rax    = vmcb->save.rax;
2398	if (npt_enabled)
2399		hsave->save.cr3    = vmcb->save.cr3;
2400	else
2401		hsave->save.cr3    = kvm_read_cr3(&svm->vcpu);
2402
2403	copy_vmcb_control_area(hsave, vmcb);
2404
2405	if (kvm_get_rflags(&svm->vcpu) & X86_EFLAGS_IF)
2406		svm->vcpu.arch.hflags |= HF_HIF_MASK;
2407	else
2408		svm->vcpu.arch.hflags &= ~HF_HIF_MASK;
2409
2410	if (nested_vmcb->control.nested_ctl) {
2411		kvm_mmu_unload(&svm->vcpu);
2412		svm->nested.nested_cr3 = nested_vmcb->control.nested_cr3;
2413		nested_svm_init_mmu_context(&svm->vcpu);
2414	}
2415
2416	/* Load the nested guest state */
2417	svm->vmcb->save.es = nested_vmcb->save.es;
2418	svm->vmcb->save.cs = nested_vmcb->save.cs;
2419	svm->vmcb->save.ss = nested_vmcb->save.ss;
2420	svm->vmcb->save.ds = nested_vmcb->save.ds;
2421	svm->vmcb->save.gdtr = nested_vmcb->save.gdtr;
2422	svm->vmcb->save.idtr = nested_vmcb->save.idtr;
2423	kvm_set_rflags(&svm->vcpu, nested_vmcb->save.rflags);
2424	svm_set_efer(&svm->vcpu, nested_vmcb->save.efer);
2425	svm_set_cr0(&svm->vcpu, nested_vmcb->save.cr0);
2426	svm_set_cr4(&svm->vcpu, nested_vmcb->save.cr4);
2427	if (npt_enabled) {
2428		svm->vmcb->save.cr3 = nested_vmcb->save.cr3;
2429		svm->vcpu.arch.cr3 = nested_vmcb->save.cr3;
2430	} else
2431		(void)kvm_set_cr3(&svm->vcpu, nested_vmcb->save.cr3);
2432
2433	/* Guest paging mode is active - reset mmu */
2434	kvm_mmu_reset_context(&svm->vcpu);
2435
2436	svm->vmcb->save.cr2 = svm->vcpu.arch.cr2 = nested_vmcb->save.cr2;
2437	kvm_register_write(&svm->vcpu, VCPU_REGS_RAX, nested_vmcb->save.rax);
2438	kvm_register_write(&svm->vcpu, VCPU_REGS_RSP, nested_vmcb->save.rsp);
2439	kvm_register_write(&svm->vcpu, VCPU_REGS_RIP, nested_vmcb->save.rip);
2440
2441	/* In case we don't even reach vcpu_run, the fields are not updated */
2442	svm->vmcb->save.rax = nested_vmcb->save.rax;
2443	svm->vmcb->save.rsp = nested_vmcb->save.rsp;
2444	svm->vmcb->save.rip = nested_vmcb->save.rip;
2445	svm->vmcb->save.dr7 = nested_vmcb->save.dr7;
2446	svm->vmcb->save.dr6 = nested_vmcb->save.dr6;
2447	svm->vmcb->save.cpl = nested_vmcb->save.cpl;
2448
2449	svm->nested.vmcb_msrpm = nested_vmcb->control.msrpm_base_pa & ~0x0fffULL;
2450	svm->nested.vmcb_iopm  = nested_vmcb->control.iopm_base_pa  & ~0x0fffULL;
2451
2452	/* cache intercepts */
2453	svm->nested.intercept_cr         = nested_vmcb->control.intercept_cr;
2454	svm->nested.intercept_dr         = nested_vmcb->control.intercept_dr;
2455	svm->nested.intercept_exceptions = nested_vmcb->control.intercept_exceptions;
2456	svm->nested.intercept            = nested_vmcb->control.intercept;
2457
2458	svm_flush_tlb(&svm->vcpu);
2459	svm->vmcb->control.int_ctl = nested_vmcb->control.int_ctl | V_INTR_MASKING_MASK;
2460	if (nested_vmcb->control.int_ctl & V_INTR_MASKING_MASK)
2461		svm->vcpu.arch.hflags |= HF_VINTR_MASK;
2462	else
2463		svm->vcpu.arch.hflags &= ~HF_VINTR_MASK;
2464
2465	if (svm->vcpu.arch.hflags & HF_VINTR_MASK) {
2466		/* We only want the cr8 intercept bits of the guest */
2467		clr_cr_intercept(svm, INTERCEPT_CR8_READ);
2468		clr_cr_intercept(svm, INTERCEPT_CR8_WRITE);
2469	}
2470
2471	/* We don't want to see VMMCALLs from a nested guest */
2472	clr_intercept(svm, INTERCEPT_VMMCALL);
2473
2474	svm->vmcb->control.lbr_ctl = nested_vmcb->control.lbr_ctl;
2475	svm->vmcb->control.int_vector = nested_vmcb->control.int_vector;
2476	svm->vmcb->control.int_state = nested_vmcb->control.int_state;
2477	svm->vmcb->control.tsc_offset += nested_vmcb->control.tsc_offset;
2478	svm->vmcb->control.event_inj = nested_vmcb->control.event_inj;
2479	svm->vmcb->control.event_inj_err = nested_vmcb->control.event_inj_err;
2480
2481	nested_svm_unmap(page);
2482
2483	/* Enter Guest-Mode */
2484	enter_guest_mode(&svm->vcpu);
2485
2486	/*
2487	 * Merge guest and host intercepts - must be called  with vcpu in
2488	 * guest-mode to take affect here
2489	 */
2490	recalc_intercepts(svm);
2491
2492	svm->nested.vmcb = vmcb_gpa;
2493
2494	enable_gif(svm);
2495
2496	mark_all_dirty(svm->vmcb);
2497
2498	return true;
2499}
2500
2501static void nested_svm_vmloadsave(struct vmcb *from_vmcb, struct vmcb *to_vmcb)
2502{
2503	to_vmcb->save.fs = from_vmcb->save.fs;
2504	to_vmcb->save.gs = from_vmcb->save.gs;
2505	to_vmcb->save.tr = from_vmcb->save.tr;
2506	to_vmcb->save.ldtr = from_vmcb->save.ldtr;
2507	to_vmcb->save.kernel_gs_base = from_vmcb->save.kernel_gs_base;
2508	to_vmcb->save.star = from_vmcb->save.star;
2509	to_vmcb->save.lstar = from_vmcb->save.lstar;
2510	to_vmcb->save.cstar = from_vmcb->save.cstar;
2511	to_vmcb->save.sfmask = from_vmcb->save.sfmask;
2512	to_vmcb->save.sysenter_cs = from_vmcb->save.sysenter_cs;
2513	to_vmcb->save.sysenter_esp = from_vmcb->save.sysenter_esp;
2514	to_vmcb->save.sysenter_eip = from_vmcb->save.sysenter_eip;
2515}
2516
2517static int vmload_interception(struct vcpu_svm *svm)
2518{
2519	struct vmcb *nested_vmcb;
2520	struct page *page;
2521
2522	if (nested_svm_check_permissions(svm))
2523		return 1;
2524
2525	nested_vmcb = nested_svm_map(svm, svm->vmcb->save.rax, &page);
2526	if (!nested_vmcb)
2527		return 1;
2528
2529	svm->next_rip = kvm_rip_read(&svm->vcpu) + 3;
2530	skip_emulated_instruction(&svm->vcpu);
2531
2532	nested_svm_vmloadsave(nested_vmcb, svm->vmcb);
2533	nested_svm_unmap(page);
2534
2535	return 1;
2536}
2537
2538static int vmsave_interception(struct vcpu_svm *svm)
2539{
2540	struct vmcb *nested_vmcb;
2541	struct page *page;
2542
2543	if (nested_svm_check_permissions(svm))
2544		return 1;
2545
2546	nested_vmcb = nested_svm_map(svm, svm->vmcb->save.rax, &page);
2547	if (!nested_vmcb)
2548		return 1;
2549
2550	svm->next_rip = kvm_rip_read(&svm->vcpu) + 3;
2551	skip_emulated_instruction(&svm->vcpu);
2552
2553	nested_svm_vmloadsave(svm->vmcb, nested_vmcb);
2554	nested_svm_unmap(page);
2555
2556	return 1;
2557}
2558
2559static int vmrun_interception(struct vcpu_svm *svm)
2560{
2561	if (nested_svm_check_permissions(svm))
2562		return 1;
2563
2564	/* Save rip after vmrun instruction */
2565	kvm_rip_write(&svm->vcpu, kvm_rip_read(&svm->vcpu) + 3);
2566
2567	if (!nested_svm_vmrun(svm))
2568		return 1;
2569
2570	if (!nested_svm_vmrun_msrpm(svm))
2571		goto failed;
2572
2573	return 1;
2574
2575failed:
2576
2577	svm->vmcb->control.exit_code    = SVM_EXIT_ERR;
2578	svm->vmcb->control.exit_code_hi = 0;
2579	svm->vmcb->control.exit_info_1  = 0;
2580	svm->vmcb->control.exit_info_2  = 0;
2581
2582	nested_svm_vmexit(svm);
2583
2584	return 1;
2585}
2586
2587static int stgi_interception(struct vcpu_svm *svm)
2588{
2589	if (nested_svm_check_permissions(svm))
2590		return 1;
2591
2592	svm->next_rip = kvm_rip_read(&svm->vcpu) + 3;
2593	skip_emulated_instruction(&svm->vcpu);
2594	kvm_make_request(KVM_REQ_EVENT, &svm->vcpu);
2595
2596	enable_gif(svm);
2597
2598	return 1;
2599}
2600
2601static int clgi_interception(struct vcpu_svm *svm)
2602{
2603	if (nested_svm_check_permissions(svm))
2604		return 1;
2605
2606	svm->next_rip = kvm_rip_read(&svm->vcpu) + 3;
2607	skip_emulated_instruction(&svm->vcpu);
2608
2609	disable_gif(svm);
2610
2611	/* After a CLGI no interrupts should come */
2612	svm_clear_vintr(svm);
2613	svm->vmcb->control.int_ctl &= ~V_IRQ_MASK;
2614
2615	mark_dirty(svm->vmcb, VMCB_INTR);
2616
2617	return 1;
2618}
2619
2620static int invlpga_interception(struct vcpu_svm *svm)
2621{
2622	struct kvm_vcpu *vcpu = &svm->vcpu;
2623
2624	trace_kvm_invlpga(svm->vmcb->save.rip, vcpu->arch.regs[VCPU_REGS_RCX],
2625			  vcpu->arch.regs[VCPU_REGS_RAX]);
2626
2627	/* Let's treat INVLPGA the same as INVLPG (can be optimized!) */
2628	kvm_mmu_invlpg(vcpu, vcpu->arch.regs[VCPU_REGS_RAX]);
2629
2630	svm->next_rip = kvm_rip_read(&svm->vcpu) + 3;
2631	skip_emulated_instruction(&svm->vcpu);
2632	return 1;
2633}
2634
2635static int skinit_interception(struct vcpu_svm *svm)
2636{
2637	trace_kvm_skinit(svm->vmcb->save.rip, svm->vcpu.arch.regs[VCPU_REGS_RAX]);
2638
2639	kvm_queue_exception(&svm->vcpu, UD_VECTOR);
2640	return 1;
2641}
2642
2643static int xsetbv_interception(struct vcpu_svm *svm)
2644{
2645	u64 new_bv = kvm_read_edx_eax(&svm->vcpu);
2646	u32 index = kvm_register_read(&svm->vcpu, VCPU_REGS_RCX);
2647
2648	if (kvm_set_xcr(&svm->vcpu, index, new_bv) == 0) {
2649		svm->next_rip = kvm_rip_read(&svm->vcpu) + 3;
2650		skip_emulated_instruction(&svm->vcpu);
2651	}
2652
2653	return 1;
2654}
2655
2656static int invalid_op_interception(struct vcpu_svm *svm)
2657{
2658	kvm_queue_exception(&svm->vcpu, UD_VECTOR);
2659	return 1;
2660}
2661
2662static int task_switch_interception(struct vcpu_svm *svm)
2663{
2664	u16 tss_selector;
2665	int reason;
2666	int int_type = svm->vmcb->control.exit_int_info &
2667		SVM_EXITINTINFO_TYPE_MASK;
2668	int int_vec = svm->vmcb->control.exit_int_info & SVM_EVTINJ_VEC_MASK;
2669	uint32_t type =
2670		svm->vmcb->control.exit_int_info & SVM_EXITINTINFO_TYPE_MASK;
2671	uint32_t idt_v =
2672		svm->vmcb->control.exit_int_info & SVM_EXITINTINFO_VALID;
2673	bool has_error_code = false;
2674	u32 error_code = 0;
2675
2676	tss_selector = (u16)svm->vmcb->control.exit_info_1;
2677
2678	if (svm->vmcb->control.exit_info_2 &
2679	    (1ULL << SVM_EXITINFOSHIFT_TS_REASON_IRET))
2680		reason = TASK_SWITCH_IRET;
2681	else if (svm->vmcb->control.exit_info_2 &
2682		 (1ULL << SVM_EXITINFOSHIFT_TS_REASON_JMP))
2683		reason = TASK_SWITCH_JMP;
2684	else if (idt_v)
2685		reason = TASK_SWITCH_GATE;
2686	else
2687		reason = TASK_SWITCH_CALL;
2688
2689	if (reason == TASK_SWITCH_GATE) {
2690		switch (type) {
2691		case SVM_EXITINTINFO_TYPE_NMI:
2692			svm->vcpu.arch.nmi_injected = false;
2693			break;
2694		case SVM_EXITINTINFO_TYPE_EXEPT:
2695			if (svm->vmcb->control.exit_info_2 &
2696			    (1ULL << SVM_EXITINFOSHIFT_TS_HAS_ERROR_CODE)) {
2697				has_error_code = true;
2698				error_code =
2699					(u32)svm->vmcb->control.exit_info_2;
2700			}
2701			kvm_clear_exception_queue(&svm->vcpu);
2702			break;
2703		case SVM_EXITINTINFO_TYPE_INTR:
2704			kvm_clear_interrupt_queue(&svm->vcpu);
2705			break;
2706		default:
2707			break;
2708		}
2709	}
2710
2711	if (reason != TASK_SWITCH_GATE ||
2712	    int_type == SVM_EXITINTINFO_TYPE_SOFT ||
2713	    (int_type == SVM_EXITINTINFO_TYPE_EXEPT &&
2714	     (int_vec == OF_VECTOR || int_vec == BP_VECTOR)))
2715		skip_emulated_instruction(&svm->vcpu);
2716
2717	if (kvm_task_switch(&svm->vcpu, tss_selector, reason,
 
 
 
2718				has_error_code, error_code) == EMULATE_FAIL) {
2719		svm->vcpu.run->exit_reason = KVM_EXIT_INTERNAL_ERROR;
2720		svm->vcpu.run->internal.suberror = KVM_INTERNAL_ERROR_EMULATION;
2721		svm->vcpu.run->internal.ndata = 0;
2722		return 0;
2723	}
2724	return 1;
2725}
2726
2727static int cpuid_interception(struct vcpu_svm *svm)
2728{
2729	svm->next_rip = kvm_rip_read(&svm->vcpu) + 2;
2730	kvm_emulate_cpuid(&svm->vcpu);
2731	return 1;
2732}
2733
2734static int iret_interception(struct vcpu_svm *svm)
2735{
2736	++svm->vcpu.stat.nmi_window_exits;
2737	clr_intercept(svm, INTERCEPT_IRET);
2738	svm->vcpu.arch.hflags |= HF_IRET_MASK;
2739	svm->nmi_iret_rip = kvm_rip_read(&svm->vcpu);
 
2740	return 1;
2741}
2742
2743static int invlpg_interception(struct vcpu_svm *svm)
2744{
2745	if (!static_cpu_has(X86_FEATURE_DECODEASSISTS))
2746		return emulate_instruction(&svm->vcpu, 0) == EMULATE_DONE;
2747
2748	kvm_mmu_invlpg(&svm->vcpu, svm->vmcb->control.exit_info_1);
2749	skip_emulated_instruction(&svm->vcpu);
2750	return 1;
2751}
2752
2753static int emulate_on_interception(struct vcpu_svm *svm)
2754{
2755	return emulate_instruction(&svm->vcpu, 0) == EMULATE_DONE;
2756}
2757
 
 
 
 
 
 
 
 
 
 
 
 
 
2758bool check_selective_cr0_intercepted(struct vcpu_svm *svm, unsigned long val)
2759{
2760	unsigned long cr0 = svm->vcpu.arch.cr0;
2761	bool ret = false;
2762	u64 intercept;
2763
2764	intercept = svm->nested.intercept;
2765
2766	if (!is_guest_mode(&svm->vcpu) ||
2767	    (!(intercept & (1ULL << INTERCEPT_SELECTIVE_CR0))))
2768		return false;
2769
2770	cr0 &= ~SVM_CR0_SELECTIVE_MASK;
2771	val &= ~SVM_CR0_SELECTIVE_MASK;
2772
2773	if (cr0 ^ val) {
2774		svm->vmcb->control.exit_code = SVM_EXIT_CR0_SEL_WRITE;
2775		ret = (nested_svm_exit_handled(svm) == NESTED_EXIT_DONE);
2776	}
2777
2778	return ret;
2779}
2780
2781#define CR_VALID (1ULL << 63)
2782
2783static int cr_interception(struct vcpu_svm *svm)
2784{
2785	int reg, cr;
2786	unsigned long val;
2787	int err;
2788
2789	if (!static_cpu_has(X86_FEATURE_DECODEASSISTS))
2790		return emulate_on_interception(svm);
2791
2792	if (unlikely((svm->vmcb->control.exit_info_1 & CR_VALID) == 0))
2793		return emulate_on_interception(svm);
2794
2795	reg = svm->vmcb->control.exit_info_1 & SVM_EXITINFO_REG_MASK;
2796	cr = svm->vmcb->control.exit_code - SVM_EXIT_READ_CR0;
2797
2798	err = 0;
2799	if (cr >= 16) { /* mov to cr */
2800		cr -= 16;
2801		val = kvm_register_read(&svm->vcpu, reg);
2802		switch (cr) {
2803		case 0:
2804			if (!check_selective_cr0_intercepted(svm, val))
2805				err = kvm_set_cr0(&svm->vcpu, val);
2806			else
2807				return 1;
2808
2809			break;
2810		case 3:
2811			err = kvm_set_cr3(&svm->vcpu, val);
2812			break;
2813		case 4:
2814			err = kvm_set_cr4(&svm->vcpu, val);
2815			break;
2816		case 8:
2817			err = kvm_set_cr8(&svm->vcpu, val);
2818			break;
2819		default:
2820			WARN(1, "unhandled write to CR%d", cr);
2821			kvm_queue_exception(&svm->vcpu, UD_VECTOR);
2822			return 1;
2823		}
2824	} else { /* mov from cr */
2825		switch (cr) {
2826		case 0:
2827			val = kvm_read_cr0(&svm->vcpu);
2828			break;
2829		case 2:
2830			val = svm->vcpu.arch.cr2;
2831			break;
2832		case 3:
2833			val = kvm_read_cr3(&svm->vcpu);
2834			break;
2835		case 4:
2836			val = kvm_read_cr4(&svm->vcpu);
2837			break;
2838		case 8:
2839			val = kvm_get_cr8(&svm->vcpu);
2840			break;
2841		default:
2842			WARN(1, "unhandled read from CR%d", cr);
2843			kvm_queue_exception(&svm->vcpu, UD_VECTOR);
2844			return 1;
2845		}
2846		kvm_register_write(&svm->vcpu, reg, val);
2847	}
2848	kvm_complete_insn_gp(&svm->vcpu, err);
2849
2850	return 1;
2851}
2852
2853static int dr_interception(struct vcpu_svm *svm)
2854{
2855	int reg, dr;
2856	unsigned long val;
2857	int err;
2858
 
 
 
 
 
 
 
 
 
 
 
2859	if (!boot_cpu_has(X86_FEATURE_DECODEASSISTS))
2860		return emulate_on_interception(svm);
2861
2862	reg = svm->vmcb->control.exit_info_1 & SVM_EXITINFO_REG_MASK;
2863	dr = svm->vmcb->control.exit_code - SVM_EXIT_READ_DR0;
2864
2865	if (dr >= 16) { /* mov to DRn */
2866		val = kvm_register_read(&svm->vcpu, reg);
2867		kvm_set_dr(&svm->vcpu, dr - 16, val);
2868	} else {
2869		err = kvm_get_dr(&svm->vcpu, dr, &val);
2870		if (!err)
2871			kvm_register_write(&svm->vcpu, reg, val);
2872	}
2873
2874	skip_emulated_instruction(&svm->vcpu);
2875
2876	return 1;
2877}
2878
2879static int cr8_write_interception(struct vcpu_svm *svm)
2880{
2881	struct kvm_run *kvm_run = svm->vcpu.run;
2882	int r;
2883
2884	u8 cr8_prev = kvm_get_cr8(&svm->vcpu);
2885	/* instruction emulation calls kvm_set_cr8() */
2886	r = cr_interception(svm);
2887	if (irqchip_in_kernel(svm->vcpu.kvm)) {
2888		clr_cr_intercept(svm, INTERCEPT_CR8_WRITE);
2889		return r;
2890	}
2891	if (cr8_prev <= kvm_get_cr8(&svm->vcpu))
2892		return r;
2893	kvm_run->exit_reason = KVM_EXIT_SET_TPR;
2894	return 0;
2895}
2896
 
 
 
 
 
 
 
2897static int svm_get_msr(struct kvm_vcpu *vcpu, unsigned ecx, u64 *data)
2898{
2899	struct vcpu_svm *svm = to_svm(vcpu);
2900
2901	switch (ecx) {
2902	case MSR_IA32_TSC: {
2903		struct vmcb *vmcb = get_host_vmcb(svm);
2904
2905		*data = vmcb->control.tsc_offset +
2906			svm_scale_tsc(vcpu, native_read_tsc());
2907
2908		break;
2909	}
2910	case MSR_STAR:
2911		*data = svm->vmcb->save.star;
2912		break;
2913#ifdef CONFIG_X86_64
2914	case MSR_LSTAR:
2915		*data = svm->vmcb->save.lstar;
2916		break;
2917	case MSR_CSTAR:
2918		*data = svm->vmcb->save.cstar;
2919		break;
2920	case MSR_KERNEL_GS_BASE:
2921		*data = svm->vmcb->save.kernel_gs_base;
2922		break;
2923	case MSR_SYSCALL_MASK:
2924		*data = svm->vmcb->save.sfmask;
2925		break;
2926#endif
2927	case MSR_IA32_SYSENTER_CS:
2928		*data = svm->vmcb->save.sysenter_cs;
2929		break;
2930	case MSR_IA32_SYSENTER_EIP:
2931		*data = svm->sysenter_eip;
2932		break;
2933	case MSR_IA32_SYSENTER_ESP:
2934		*data = svm->sysenter_esp;
2935		break;
2936	/*
2937	 * Nobody will change the following 5 values in the VMCB so we can
2938	 * safely return them on rdmsr. They will always be 0 until LBRV is
2939	 * implemented.
2940	 */
2941	case MSR_IA32_DEBUGCTLMSR:
2942		*data = svm->vmcb->save.dbgctl;
2943		break;
2944	case MSR_IA32_LASTBRANCHFROMIP:
2945		*data = svm->vmcb->save.br_from;
2946		break;
2947	case MSR_IA32_LASTBRANCHTOIP:
2948		*data = svm->vmcb->save.br_to;
2949		break;
2950	case MSR_IA32_LASTINTFROMIP:
2951		*data = svm->vmcb->save.last_excp_from;
2952		break;
2953	case MSR_IA32_LASTINTTOIP:
2954		*data = svm->vmcb->save.last_excp_to;
2955		break;
2956	case MSR_VM_HSAVE_PA:
2957		*data = svm->nested.hsave_msr;
2958		break;
2959	case MSR_VM_CR:
2960		*data = svm->nested.vm_cr_msr;
2961		break;
2962	case MSR_IA32_UCODE_REV:
2963		*data = 0x01000065;
2964		break;
2965	default:
2966		return kvm_get_msr_common(vcpu, ecx, data);
2967	}
2968	return 0;
2969}
2970
2971static int rdmsr_interception(struct vcpu_svm *svm)
2972{
2973	u32 ecx = svm->vcpu.arch.regs[VCPU_REGS_RCX];
2974	u64 data;
2975
2976	if (svm_get_msr(&svm->vcpu, ecx, &data)) {
2977		trace_kvm_msr_read_ex(ecx);
2978		kvm_inject_gp(&svm->vcpu, 0);
2979	} else {
2980		trace_kvm_msr_read(ecx, data);
2981
2982		svm->vcpu.arch.regs[VCPU_REGS_RAX] = data & 0xffffffff;
2983		svm->vcpu.arch.regs[VCPU_REGS_RDX] = data >> 32;
2984		svm->next_rip = kvm_rip_read(&svm->vcpu) + 2;
2985		skip_emulated_instruction(&svm->vcpu);
2986	}
2987	return 1;
2988}
2989
2990static int svm_set_vm_cr(struct kvm_vcpu *vcpu, u64 data)
2991{
2992	struct vcpu_svm *svm = to_svm(vcpu);
2993	int svm_dis, chg_mask;
2994
2995	if (data & ~SVM_VM_CR_VALID_MASK)
2996		return 1;
2997
2998	chg_mask = SVM_VM_CR_VALID_MASK;
2999
3000	if (svm->nested.vm_cr_msr & SVM_VM_CR_SVM_DIS_MASK)
3001		chg_mask &= ~(SVM_VM_CR_SVM_LOCK_MASK | SVM_VM_CR_SVM_DIS_MASK);
3002
3003	svm->nested.vm_cr_msr &= ~chg_mask;
3004	svm->nested.vm_cr_msr |= (data & chg_mask);
3005
3006	svm_dis = svm->nested.vm_cr_msr & SVM_VM_CR_SVM_DIS_MASK;
3007
3008	/* check for svm_disable while efer.svme is set */
3009	if (svm_dis && (vcpu->arch.efer & EFER_SVME))
3010		return 1;
3011
3012	return 0;
3013}
3014
3015static int svm_set_msr(struct kvm_vcpu *vcpu, unsigned ecx, u64 data)
3016{
3017	struct vcpu_svm *svm = to_svm(vcpu);
3018
 
 
3019	switch (ecx) {
3020	case MSR_IA32_TSC:
3021		kvm_write_tsc(vcpu, data);
3022		break;
3023	case MSR_STAR:
3024		svm->vmcb->save.star = data;
3025		break;
3026#ifdef CONFIG_X86_64
3027	case MSR_LSTAR:
3028		svm->vmcb->save.lstar = data;
3029		break;
3030	case MSR_CSTAR:
3031		svm->vmcb->save.cstar = data;
3032		break;
3033	case MSR_KERNEL_GS_BASE:
3034		svm->vmcb->save.kernel_gs_base = data;
3035		break;
3036	case MSR_SYSCALL_MASK:
3037		svm->vmcb->save.sfmask = data;
3038		break;
3039#endif
3040	case MSR_IA32_SYSENTER_CS:
3041		svm->vmcb->save.sysenter_cs = data;
3042		break;
3043	case MSR_IA32_SYSENTER_EIP:
3044		svm->sysenter_eip = data;
3045		svm->vmcb->save.sysenter_eip = data;
3046		break;
3047	case MSR_IA32_SYSENTER_ESP:
3048		svm->sysenter_esp = data;
3049		svm->vmcb->save.sysenter_esp = data;
3050		break;
3051	case MSR_IA32_DEBUGCTLMSR:
3052		if (!boot_cpu_has(X86_FEATURE_LBRV)) {
3053			pr_unimpl(vcpu, "%s: MSR_IA32_DEBUGCTL 0x%llx, nop\n",
3054					__func__, data);
3055			break;
3056		}
3057		if (data & DEBUGCTL_RESERVED_BITS)
3058			return 1;
3059
3060		svm->vmcb->save.dbgctl = data;
3061		mark_dirty(svm->vmcb, VMCB_LBR);
3062		if (data & (1ULL<<0))
3063			svm_enable_lbrv(svm);
3064		else
3065			svm_disable_lbrv(svm);
3066		break;
3067	case MSR_VM_HSAVE_PA:
3068		svm->nested.hsave_msr = data;
3069		break;
3070	case MSR_VM_CR:
3071		return svm_set_vm_cr(vcpu, data);
3072	case MSR_VM_IGNNE:
3073		pr_unimpl(vcpu, "unimplemented wrmsr: 0x%x data 0x%llx\n", ecx, data);
3074		break;
3075	default:
3076		return kvm_set_msr_common(vcpu, ecx, data);
3077	}
3078	return 0;
3079}
3080
3081static int wrmsr_interception(struct vcpu_svm *svm)
3082{
 
3083	u32 ecx = svm->vcpu.arch.regs[VCPU_REGS_RCX];
3084	u64 data = (svm->vcpu.arch.regs[VCPU_REGS_RAX] & -1u)
3085		| ((u64)(svm->vcpu.arch.regs[VCPU_REGS_RDX] & -1u) << 32);
3086
 
 
 
3087
3088	svm->next_rip = kvm_rip_read(&svm->vcpu) + 2;
3089	if (svm_set_msr(&svm->vcpu, ecx, data)) {
3090		trace_kvm_msr_write_ex(ecx, data);
3091		kvm_inject_gp(&svm->vcpu, 0);
3092	} else {
3093		trace_kvm_msr_write(ecx, data);
3094		skip_emulated_instruction(&svm->vcpu);
3095	}
3096	return 1;
3097}
3098
3099static int msr_interception(struct vcpu_svm *svm)
3100{
3101	if (svm->vmcb->control.exit_info_1)
3102		return wrmsr_interception(svm);
3103	else
3104		return rdmsr_interception(svm);
3105}
3106
3107static int interrupt_window_interception(struct vcpu_svm *svm)
3108{
3109	struct kvm_run *kvm_run = svm->vcpu.run;
3110
3111	kvm_make_request(KVM_REQ_EVENT, &svm->vcpu);
3112	svm_clear_vintr(svm);
3113	svm->vmcb->control.int_ctl &= ~V_IRQ_MASK;
3114	mark_dirty(svm->vmcb, VMCB_INTR);
 
3115	/*
3116	 * If the user space waits to inject interrupts, exit as soon as
3117	 * possible
3118	 */
3119	if (!irqchip_in_kernel(svm->vcpu.kvm) &&
3120	    kvm_run->request_interrupt_window &&
3121	    !kvm_cpu_has_interrupt(&svm->vcpu)) {
3122		++svm->vcpu.stat.irq_window_exits;
3123		kvm_run->exit_reason = KVM_EXIT_IRQ_WINDOW_OPEN;
3124		return 0;
3125	}
3126
3127	return 1;
3128}
3129
3130static int pause_interception(struct vcpu_svm *svm)
3131{
3132	kvm_vcpu_on_spin(&(svm->vcpu));
3133	return 1;
3134}
3135
3136static int (*svm_exit_handlers[])(struct vcpu_svm *svm) = {
3137	[SVM_EXIT_READ_CR0]			= cr_interception,
3138	[SVM_EXIT_READ_CR3]			= cr_interception,
3139	[SVM_EXIT_READ_CR4]			= cr_interception,
3140	[SVM_EXIT_READ_CR8]			= cr_interception,
3141	[SVM_EXIT_CR0_SEL_WRITE]		= emulate_on_interception,
3142	[SVM_EXIT_WRITE_CR0]			= cr_interception,
3143	[SVM_EXIT_WRITE_CR3]			= cr_interception,
3144	[SVM_EXIT_WRITE_CR4]			= cr_interception,
3145	[SVM_EXIT_WRITE_CR8]			= cr8_write_interception,
3146	[SVM_EXIT_READ_DR0]			= dr_interception,
3147	[SVM_EXIT_READ_DR1]			= dr_interception,
3148	[SVM_EXIT_READ_DR2]			= dr_interception,
3149	[SVM_EXIT_READ_DR3]			= dr_interception,
3150	[SVM_EXIT_READ_DR4]			= dr_interception,
3151	[SVM_EXIT_READ_DR5]			= dr_interception,
3152	[SVM_EXIT_READ_DR6]			= dr_interception,
3153	[SVM_EXIT_READ_DR7]			= dr_interception,
3154	[SVM_EXIT_WRITE_DR0]			= dr_interception,
3155	[SVM_EXIT_WRITE_DR1]			= dr_interception,
3156	[SVM_EXIT_WRITE_DR2]			= dr_interception,
3157	[SVM_EXIT_WRITE_DR3]			= dr_interception,
3158	[SVM_EXIT_WRITE_DR4]			= dr_interception,
3159	[SVM_EXIT_WRITE_DR5]			= dr_interception,
3160	[SVM_EXIT_WRITE_DR6]			= dr_interception,
3161	[SVM_EXIT_WRITE_DR7]			= dr_interception,
3162	[SVM_EXIT_EXCP_BASE + DB_VECTOR]	= db_interception,
3163	[SVM_EXIT_EXCP_BASE + BP_VECTOR]	= bp_interception,
3164	[SVM_EXIT_EXCP_BASE + UD_VECTOR]	= ud_interception,
3165	[SVM_EXIT_EXCP_BASE + PF_VECTOR]	= pf_interception,
3166	[SVM_EXIT_EXCP_BASE + NM_VECTOR]	= nm_interception,
3167	[SVM_EXIT_EXCP_BASE + MC_VECTOR]	= mc_interception,
3168	[SVM_EXIT_INTR]				= intr_interception,
3169	[SVM_EXIT_NMI]				= nmi_interception,
3170	[SVM_EXIT_SMI]				= nop_on_interception,
3171	[SVM_EXIT_INIT]				= nop_on_interception,
3172	[SVM_EXIT_VINTR]			= interrupt_window_interception,
 
3173	[SVM_EXIT_CPUID]			= cpuid_interception,
3174	[SVM_EXIT_IRET]                         = iret_interception,
3175	[SVM_EXIT_INVD]                         = emulate_on_interception,
3176	[SVM_EXIT_PAUSE]			= pause_interception,
3177	[SVM_EXIT_HLT]				= halt_interception,
3178	[SVM_EXIT_INVLPG]			= invlpg_interception,
3179	[SVM_EXIT_INVLPGA]			= invlpga_interception,
3180	[SVM_EXIT_IOIO]				= io_interception,
3181	[SVM_EXIT_MSR]				= msr_interception,
3182	[SVM_EXIT_TASK_SWITCH]			= task_switch_interception,
3183	[SVM_EXIT_SHUTDOWN]			= shutdown_interception,
3184	[SVM_EXIT_VMRUN]			= vmrun_interception,
3185	[SVM_EXIT_VMMCALL]			= vmmcall_interception,
3186	[SVM_EXIT_VMLOAD]			= vmload_interception,
3187	[SVM_EXIT_VMSAVE]			= vmsave_interception,
3188	[SVM_EXIT_STGI]				= stgi_interception,
3189	[SVM_EXIT_CLGI]				= clgi_interception,
3190	[SVM_EXIT_SKINIT]			= skinit_interception,
3191	[SVM_EXIT_WBINVD]                       = emulate_on_interception,
3192	[SVM_EXIT_MONITOR]			= invalid_op_interception,
3193	[SVM_EXIT_MWAIT]			= invalid_op_interception,
3194	[SVM_EXIT_XSETBV]			= xsetbv_interception,
3195	[SVM_EXIT_NPF]				= pf_interception,
3196};
3197
3198static void dump_vmcb(struct kvm_vcpu *vcpu)
3199{
3200	struct vcpu_svm *svm = to_svm(vcpu);
3201	struct vmcb_control_area *control = &svm->vmcb->control;
3202	struct vmcb_save_area *save = &svm->vmcb->save;
3203
3204	pr_err("VMCB Control Area:\n");
3205	pr_err("%-20s%04x\n", "cr_read:", control->intercept_cr & 0xffff);
3206	pr_err("%-20s%04x\n", "cr_write:", control->intercept_cr >> 16);
3207	pr_err("%-20s%04x\n", "dr_read:", control->intercept_dr & 0xffff);
3208	pr_err("%-20s%04x\n", "dr_write:", control->intercept_dr >> 16);
3209	pr_err("%-20s%08x\n", "exceptions:", control->intercept_exceptions);
3210	pr_err("%-20s%016llx\n", "intercepts:", control->intercept);
3211	pr_err("%-20s%d\n", "pause filter count:", control->pause_filter_count);
3212	pr_err("%-20s%016llx\n", "iopm_base_pa:", control->iopm_base_pa);
3213	pr_err("%-20s%016llx\n", "msrpm_base_pa:", control->msrpm_base_pa);
3214	pr_err("%-20s%016llx\n", "tsc_offset:", control->tsc_offset);
3215	pr_err("%-20s%d\n", "asid:", control->asid);
3216	pr_err("%-20s%d\n", "tlb_ctl:", control->tlb_ctl);
3217	pr_err("%-20s%08x\n", "int_ctl:", control->int_ctl);
3218	pr_err("%-20s%08x\n", "int_vector:", control->int_vector);
3219	pr_err("%-20s%08x\n", "int_state:", control->int_state);
3220	pr_err("%-20s%08x\n", "exit_code:", control->exit_code);
3221	pr_err("%-20s%016llx\n", "exit_info1:", control->exit_info_1);
3222	pr_err("%-20s%016llx\n", "exit_info2:", control->exit_info_2);
3223	pr_err("%-20s%08x\n", "exit_int_info:", control->exit_int_info);
3224	pr_err("%-20s%08x\n", "exit_int_info_err:", control->exit_int_info_err);
3225	pr_err("%-20s%lld\n", "nested_ctl:", control->nested_ctl);
3226	pr_err("%-20s%016llx\n", "nested_cr3:", control->nested_cr3);
3227	pr_err("%-20s%08x\n", "event_inj:", control->event_inj);
3228	pr_err("%-20s%08x\n", "event_inj_err:", control->event_inj_err);
3229	pr_err("%-20s%lld\n", "lbr_ctl:", control->lbr_ctl);
3230	pr_err("%-20s%016llx\n", "next_rip:", control->next_rip);
3231	pr_err("VMCB State Save Area:\n");
3232	pr_err("%-5s s: %04x a: %04x l: %08x b: %016llx\n",
3233	       "es:",
3234	       save->es.selector, save->es.attrib,
3235	       save->es.limit, save->es.base);
3236	pr_err("%-5s s: %04x a: %04x l: %08x b: %016llx\n",
3237	       "cs:",
3238	       save->cs.selector, save->cs.attrib,
3239	       save->cs.limit, save->cs.base);
3240	pr_err("%-5s s: %04x a: %04x l: %08x b: %016llx\n",
3241	       "ss:",
3242	       save->ss.selector, save->ss.attrib,
3243	       save->ss.limit, save->ss.base);
3244	pr_err("%-5s s: %04x a: %04x l: %08x b: %016llx\n",
3245	       "ds:",
3246	       save->ds.selector, save->ds.attrib,
3247	       save->ds.limit, save->ds.base);
3248	pr_err("%-5s s: %04x a: %04x l: %08x b: %016llx\n",
3249	       "fs:",
3250	       save->fs.selector, save->fs.attrib,
3251	       save->fs.limit, save->fs.base);
3252	pr_err("%-5s s: %04x a: %04x l: %08x b: %016llx\n",
3253	       "gs:",
3254	       save->gs.selector, save->gs.attrib,
3255	       save->gs.limit, save->gs.base);
3256	pr_err("%-5s s: %04x a: %04x l: %08x b: %016llx\n",
3257	       "gdtr:",
3258	       save->gdtr.selector, save->gdtr.attrib,
3259	       save->gdtr.limit, save->gdtr.base);
3260	pr_err("%-5s s: %04x a: %04x l: %08x b: %016llx\n",
3261	       "ldtr:",
3262	       save->ldtr.selector, save->ldtr.attrib,
3263	       save->ldtr.limit, save->ldtr.base);
3264	pr_err("%-5s s: %04x a: %04x l: %08x b: %016llx\n",
3265	       "idtr:",
3266	       save->idtr.selector, save->idtr.attrib,
3267	       save->idtr.limit, save->idtr.base);
3268	pr_err("%-5s s: %04x a: %04x l: %08x b: %016llx\n",
3269	       "tr:",
3270	       save->tr.selector, save->tr.attrib,
3271	       save->tr.limit, save->tr.base);
3272	pr_err("cpl:            %d                efer:         %016llx\n",
3273		save->cpl, save->efer);
3274	pr_err("%-15s %016llx %-13s %016llx\n",
3275	       "cr0:", save->cr0, "cr2:", save->cr2);
3276	pr_err("%-15s %016llx %-13s %016llx\n",
3277	       "cr3:", save->cr3, "cr4:", save->cr4);
3278	pr_err("%-15s %016llx %-13s %016llx\n",
3279	       "dr6:", save->dr6, "dr7:", save->dr7);
3280	pr_err("%-15s %016llx %-13s %016llx\n",
3281	       "rip:", save->rip, "rflags:", save->rflags);
3282	pr_err("%-15s %016llx %-13s %016llx\n",
3283	       "rsp:", save->rsp, "rax:", save->rax);
3284	pr_err("%-15s %016llx %-13s %016llx\n",
3285	       "star:", save->star, "lstar:", save->lstar);
3286	pr_err("%-15s %016llx %-13s %016llx\n",
3287	       "cstar:", save->cstar, "sfmask:", save->sfmask);
3288	pr_err("%-15s %016llx %-13s %016llx\n",
3289	       "kernel_gs_base:", save->kernel_gs_base,
3290	       "sysenter_cs:", save->sysenter_cs);
3291	pr_err("%-15s %016llx %-13s %016llx\n",
3292	       "sysenter_esp:", save->sysenter_esp,
3293	       "sysenter_eip:", save->sysenter_eip);
3294	pr_err("%-15s %016llx %-13s %016llx\n",
3295	       "gpat:", save->g_pat, "dbgctl:", save->dbgctl);
3296	pr_err("%-15s %016llx %-13s %016llx\n",
3297	       "br_from:", save->br_from, "br_to:", save->br_to);
3298	pr_err("%-15s %016llx %-13s %016llx\n",
3299	       "excp_from:", save->last_excp_from,
3300	       "excp_to:", save->last_excp_to);
3301}
3302
3303static void svm_get_exit_info(struct kvm_vcpu *vcpu, u64 *info1, u64 *info2)
3304{
3305	struct vmcb_control_area *control = &to_svm(vcpu)->vmcb->control;
3306
3307	*info1 = control->exit_info_1;
3308	*info2 = control->exit_info_2;
3309}
3310
3311static int handle_exit(struct kvm_vcpu *vcpu)
3312{
3313	struct vcpu_svm *svm = to_svm(vcpu);
3314	struct kvm_run *kvm_run = vcpu->run;
3315	u32 exit_code = svm->vmcb->control.exit_code;
3316
3317	trace_kvm_exit(exit_code, vcpu, KVM_ISA_SVM);
3318
3319	if (!is_cr_intercept(svm, INTERCEPT_CR0_WRITE))
3320		vcpu->arch.cr0 = svm->vmcb->save.cr0;
3321	if (npt_enabled)
3322		vcpu->arch.cr3 = svm->vmcb->save.cr3;
3323
3324	if (unlikely(svm->nested.exit_required)) {
3325		nested_svm_vmexit(svm);
3326		svm->nested.exit_required = false;
3327
3328		return 1;
3329	}
3330
3331	if (is_guest_mode(vcpu)) {
3332		int vmexit;
3333
3334		trace_kvm_nested_vmexit(svm->vmcb->save.rip, exit_code,
3335					svm->vmcb->control.exit_info_1,
3336					svm->vmcb->control.exit_info_2,
3337					svm->vmcb->control.exit_int_info,
3338					svm->vmcb->control.exit_int_info_err);
 
3339
3340		vmexit = nested_svm_exit_special(svm);
3341
3342		if (vmexit == NESTED_EXIT_CONTINUE)
3343			vmexit = nested_svm_exit_handled(svm);
3344
3345		if (vmexit == NESTED_EXIT_DONE)
3346			return 1;
3347	}
3348
3349	svm_complete_interrupts(svm);
3350
3351	if (svm->vmcb->control.exit_code == SVM_EXIT_ERR) {
3352		kvm_run->exit_reason = KVM_EXIT_FAIL_ENTRY;
3353		kvm_run->fail_entry.hardware_entry_failure_reason
3354			= svm->vmcb->control.exit_code;
3355		pr_err("KVM: FAILED VMRUN WITH VMCB:\n");
3356		dump_vmcb(vcpu);
3357		return 0;
3358	}
3359
3360	if (is_external_interrupt(svm->vmcb->control.exit_int_info) &&
3361	    exit_code != SVM_EXIT_EXCP_BASE + PF_VECTOR &&
3362	    exit_code != SVM_EXIT_NPF && exit_code != SVM_EXIT_TASK_SWITCH &&
3363	    exit_code != SVM_EXIT_INTR && exit_code != SVM_EXIT_NMI)
3364		printk(KERN_ERR "%s: unexpected exit_ini_info 0x%x "
3365		       "exit_code 0x%x\n",
3366		       __func__, svm->vmcb->control.exit_int_info,
3367		       exit_code);
3368
3369	if (exit_code >= ARRAY_SIZE(svm_exit_handlers)
3370	    || !svm_exit_handlers[exit_code]) {
3371		kvm_run->exit_reason = KVM_EXIT_UNKNOWN;
3372		kvm_run->hw.hardware_exit_reason = exit_code;
3373		return 0;
3374	}
3375
3376	return svm_exit_handlers[exit_code](svm);
3377}
3378
3379static void reload_tss(struct kvm_vcpu *vcpu)
3380{
3381	int cpu = raw_smp_processor_id();
3382
3383	struct svm_cpu_data *sd = per_cpu(svm_data, cpu);
3384	sd->tss_desc->type = 9; /* available 32/64-bit TSS */
3385	load_TR_desc();
3386}
3387
3388static void pre_svm_run(struct vcpu_svm *svm)
3389{
3390	int cpu = raw_smp_processor_id();
3391
3392	struct svm_cpu_data *sd = per_cpu(svm_data, cpu);
3393
3394	/* FIXME: handle wraparound of asid_generation */
3395	if (svm->asid_generation != sd->asid_generation)
3396		new_asid(svm, sd);
3397}
3398
3399static void svm_inject_nmi(struct kvm_vcpu *vcpu)
3400{
3401	struct vcpu_svm *svm = to_svm(vcpu);
3402
3403	svm->vmcb->control.event_inj = SVM_EVTINJ_VALID | SVM_EVTINJ_TYPE_NMI;
3404	vcpu->arch.hflags |= HF_NMI_MASK;
3405	set_intercept(svm, INTERCEPT_IRET);
3406	++vcpu->stat.nmi_injections;
3407}
3408
3409static inline void svm_inject_irq(struct vcpu_svm *svm, int irq)
3410{
3411	struct vmcb_control_area *control;
3412
3413	control = &svm->vmcb->control;
3414	control->int_vector = irq;
3415	control->int_ctl &= ~V_INTR_PRIO_MASK;
3416	control->int_ctl |= V_IRQ_MASK |
3417		((/*control->int_vector >> 4*/ 0xf) << V_INTR_PRIO_SHIFT);
3418	mark_dirty(svm->vmcb, VMCB_INTR);
3419}
3420
3421static void svm_set_irq(struct kvm_vcpu *vcpu)
3422{
3423	struct vcpu_svm *svm = to_svm(vcpu);
3424
3425	BUG_ON(!(gif_set(svm)));
3426
3427	trace_kvm_inj_virq(vcpu->arch.interrupt.nr);
3428	++vcpu->stat.irq_injections;
3429
3430	svm->vmcb->control.event_inj = vcpu->arch.interrupt.nr |
3431		SVM_EVTINJ_VALID | SVM_EVTINJ_TYPE_INTR;
3432}
3433
3434static void update_cr8_intercept(struct kvm_vcpu *vcpu, int tpr, int irr)
3435{
3436	struct vcpu_svm *svm = to_svm(vcpu);
3437
3438	if (is_guest_mode(vcpu) && (vcpu->arch.hflags & HF_VINTR_MASK))
3439		return;
3440
 
 
3441	if (irr == -1)
3442		return;
3443
3444	if (tpr >= irr)
3445		set_cr_intercept(svm, INTERCEPT_CR8_WRITE);
3446}
3447
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
3448static int svm_nmi_allowed(struct kvm_vcpu *vcpu)
3449{
3450	struct vcpu_svm *svm = to_svm(vcpu);
3451	struct vmcb *vmcb = svm->vmcb;
3452	int ret;
3453	ret = !(vmcb->control.int_state & SVM_INTERRUPT_SHADOW_MASK) &&
3454	      !(svm->vcpu.arch.hflags & HF_NMI_MASK);
3455	ret = ret && gif_set(svm) && nested_svm_nmi(svm);
3456
3457	return ret;
3458}
3459
3460static bool svm_get_nmi_mask(struct kvm_vcpu *vcpu)
3461{
3462	struct vcpu_svm *svm = to_svm(vcpu);
3463
3464	return !!(svm->vcpu.arch.hflags & HF_NMI_MASK);
3465}
3466
3467static void svm_set_nmi_mask(struct kvm_vcpu *vcpu, bool masked)
3468{
3469	struct vcpu_svm *svm = to_svm(vcpu);
3470
3471	if (masked) {
3472		svm->vcpu.arch.hflags |= HF_NMI_MASK;
3473		set_intercept(svm, INTERCEPT_IRET);
3474	} else {
3475		svm->vcpu.arch.hflags &= ~HF_NMI_MASK;
3476		clr_intercept(svm, INTERCEPT_IRET);
3477	}
3478}
3479
3480static int svm_interrupt_allowed(struct kvm_vcpu *vcpu)
3481{
3482	struct vcpu_svm *svm = to_svm(vcpu);
3483	struct vmcb *vmcb = svm->vmcb;
3484	int ret;
3485
3486	if (!gif_set(svm) ||
3487	     (vmcb->control.int_state & SVM_INTERRUPT_SHADOW_MASK))
3488		return 0;
3489
3490	ret = !!(kvm_get_rflags(vcpu) & X86_EFLAGS_IF);
3491
3492	if (is_guest_mode(vcpu))
3493		return ret && !(svm->vcpu.arch.hflags & HF_VINTR_MASK);
3494
3495	return ret;
3496}
3497
3498static void enable_irq_window(struct kvm_vcpu *vcpu)
3499{
3500	struct vcpu_svm *svm = to_svm(vcpu);
3501
3502	/*
3503	 * In case GIF=0 we can't rely on the CPU to tell us when GIF becomes
3504	 * 1, because that's a separate STGI/VMRUN intercept.  The next time we
3505	 * get that intercept, this function will be called again though and
3506	 * we'll get the vintr intercept.
3507	 */
3508	if (gif_set(svm) && nested_svm_intr(svm)) {
3509		svm_set_vintr(svm);
3510		svm_inject_irq(svm, 0x0);
3511	}
3512}
3513
3514static void enable_nmi_window(struct kvm_vcpu *vcpu)
3515{
3516	struct vcpu_svm *svm = to_svm(vcpu);
3517
3518	if ((svm->vcpu.arch.hflags & (HF_NMI_MASK | HF_IRET_MASK))
3519	    == HF_NMI_MASK)
3520		return; /* IRET will cause a vm exit */
3521
3522	/*
3523	 * Something prevents NMI from been injected. Single step over possible
3524	 * problem (IRET or exception injection or interrupt shadow)
3525	 */
3526	svm->nmi_singlestep = true;
3527	svm->vmcb->save.rflags |= (X86_EFLAGS_TF | X86_EFLAGS_RF);
3528	update_db_intercept(vcpu);
3529}
3530
3531static int svm_set_tss_addr(struct kvm *kvm, unsigned int addr)
3532{
3533	return 0;
3534}
3535
3536static void svm_flush_tlb(struct kvm_vcpu *vcpu)
3537{
3538	struct vcpu_svm *svm = to_svm(vcpu);
3539
3540	if (static_cpu_has(X86_FEATURE_FLUSHBYASID))
3541		svm->vmcb->control.tlb_ctl = TLB_CONTROL_FLUSH_ASID;
3542	else
3543		svm->asid_generation--;
3544}
3545
3546static void svm_prepare_guest_switch(struct kvm_vcpu *vcpu)
3547{
3548}
3549
3550static inline void sync_cr8_to_lapic(struct kvm_vcpu *vcpu)
3551{
3552	struct vcpu_svm *svm = to_svm(vcpu);
3553
3554	if (is_guest_mode(vcpu) && (vcpu->arch.hflags & HF_VINTR_MASK))
3555		return;
3556
3557	if (!is_cr_intercept(svm, INTERCEPT_CR8_WRITE)) {
3558		int cr8 = svm->vmcb->control.int_ctl & V_TPR_MASK;
3559		kvm_set_cr8(vcpu, cr8);
3560	}
3561}
3562
3563static inline void sync_lapic_to_cr8(struct kvm_vcpu *vcpu)
3564{
3565	struct vcpu_svm *svm = to_svm(vcpu);
3566	u64 cr8;
3567
3568	if (is_guest_mode(vcpu) && (vcpu->arch.hflags & HF_VINTR_MASK))
3569		return;
3570
3571	cr8 = kvm_get_cr8(vcpu);
3572	svm->vmcb->control.int_ctl &= ~V_TPR_MASK;
3573	svm->vmcb->control.int_ctl |= cr8 & V_TPR_MASK;
3574}
3575
3576static void svm_complete_interrupts(struct vcpu_svm *svm)
3577{
3578	u8 vector;
3579	int type;
3580	u32 exitintinfo = svm->vmcb->control.exit_int_info;
3581	unsigned int3_injected = svm->int3_injected;
3582
3583	svm->int3_injected = 0;
3584
3585	/*
3586	 * If we've made progress since setting HF_IRET_MASK, we've
3587	 * executed an IRET and can allow NMI injection.
3588	 */
3589	if ((svm->vcpu.arch.hflags & HF_IRET_MASK)
3590	    && kvm_rip_read(&svm->vcpu) != svm->nmi_iret_rip) {
3591		svm->vcpu.arch.hflags &= ~(HF_NMI_MASK | HF_IRET_MASK);
3592		kvm_make_request(KVM_REQ_EVENT, &svm->vcpu);
3593	}
3594
3595	svm->vcpu.arch.nmi_injected = false;
3596	kvm_clear_exception_queue(&svm->vcpu);
3597	kvm_clear_interrupt_queue(&svm->vcpu);
3598
3599	if (!(exitintinfo & SVM_EXITINTINFO_VALID))
3600		return;
3601
3602	kvm_make_request(KVM_REQ_EVENT, &svm->vcpu);
3603
3604	vector = exitintinfo & SVM_EXITINTINFO_VEC_MASK;
3605	type = exitintinfo & SVM_EXITINTINFO_TYPE_MASK;
3606
3607	switch (type) {
3608	case SVM_EXITINTINFO_TYPE_NMI:
3609		svm->vcpu.arch.nmi_injected = true;
3610		break;
3611	case SVM_EXITINTINFO_TYPE_EXEPT:
3612		/*
3613		 * In case of software exceptions, do not reinject the vector,
3614		 * but re-execute the instruction instead. Rewind RIP first
3615		 * if we emulated INT3 before.
3616		 */
3617		if (kvm_exception_is_soft(vector)) {
3618			if (vector == BP_VECTOR && int3_injected &&
3619			    kvm_is_linear_rip(&svm->vcpu, svm->int3_rip))
3620				kvm_rip_write(&svm->vcpu,
3621					      kvm_rip_read(&svm->vcpu) -
3622					      int3_injected);
3623			break;
3624		}
3625		if (exitintinfo & SVM_EXITINTINFO_VALID_ERR) {
3626			u32 err = svm->vmcb->control.exit_int_info_err;
3627			kvm_requeue_exception_e(&svm->vcpu, vector, err);
3628
3629		} else
3630			kvm_requeue_exception(&svm->vcpu, vector);
3631		break;
3632	case SVM_EXITINTINFO_TYPE_INTR:
3633		kvm_queue_interrupt(&svm->vcpu, vector, false);
3634		break;
3635	default:
3636		break;
3637	}
3638}
3639
3640static void svm_cancel_injection(struct kvm_vcpu *vcpu)
3641{
3642	struct vcpu_svm *svm = to_svm(vcpu);
3643	struct vmcb_control_area *control = &svm->vmcb->control;
3644
3645	control->exit_int_info = control->event_inj;
3646	control->exit_int_info_err = control->event_inj_err;
3647	control->event_inj = 0;
3648	svm_complete_interrupts(svm);
3649}
3650
3651#ifdef CONFIG_X86_64
3652#define R "r"
3653#else
3654#define R "e"
3655#endif
3656
3657static void svm_vcpu_run(struct kvm_vcpu *vcpu)
3658{
3659	struct vcpu_svm *svm = to_svm(vcpu);
3660
3661	svm->vmcb->save.rax = vcpu->arch.regs[VCPU_REGS_RAX];
3662	svm->vmcb->save.rsp = vcpu->arch.regs[VCPU_REGS_RSP];
3663	svm->vmcb->save.rip = vcpu->arch.regs[VCPU_REGS_RIP];
3664
3665	/*
3666	 * A vmexit emulation is required before the vcpu can be executed
3667	 * again.
3668	 */
3669	if (unlikely(svm->nested.exit_required))
3670		return;
3671
3672	pre_svm_run(svm);
3673
3674	sync_lapic_to_cr8(vcpu);
3675
3676	svm->vmcb->save.cr2 = vcpu->arch.cr2;
3677
3678	clgi();
3679
3680	local_irq_enable();
3681
3682	asm volatile (
3683		"push %%"R"bp; \n\t"
3684		"mov %c[rbx](%[svm]), %%"R"bx \n\t"
3685		"mov %c[rcx](%[svm]), %%"R"cx \n\t"
3686		"mov %c[rdx](%[svm]), %%"R"dx \n\t"
3687		"mov %c[rsi](%[svm]), %%"R"si \n\t"
3688		"mov %c[rdi](%[svm]), %%"R"di \n\t"
3689		"mov %c[rbp](%[svm]), %%"R"bp \n\t"
3690#ifdef CONFIG_X86_64
3691		"mov %c[r8](%[svm]),  %%r8  \n\t"
3692		"mov %c[r9](%[svm]),  %%r9  \n\t"
3693		"mov %c[r10](%[svm]), %%r10 \n\t"
3694		"mov %c[r11](%[svm]), %%r11 \n\t"
3695		"mov %c[r12](%[svm]), %%r12 \n\t"
3696		"mov %c[r13](%[svm]), %%r13 \n\t"
3697		"mov %c[r14](%[svm]), %%r14 \n\t"
3698		"mov %c[r15](%[svm]), %%r15 \n\t"
3699#endif
3700
3701		/* Enter guest mode */
3702		"push %%"R"ax \n\t"
3703		"mov %c[vmcb](%[svm]), %%"R"ax \n\t"
3704		__ex(SVM_VMLOAD) "\n\t"
3705		__ex(SVM_VMRUN) "\n\t"
3706		__ex(SVM_VMSAVE) "\n\t"
3707		"pop %%"R"ax \n\t"
3708
3709		/* Save guest registers, load host registers */
3710		"mov %%"R"bx, %c[rbx](%[svm]) \n\t"
3711		"mov %%"R"cx, %c[rcx](%[svm]) \n\t"
3712		"mov %%"R"dx, %c[rdx](%[svm]) \n\t"
3713		"mov %%"R"si, %c[rsi](%[svm]) \n\t"
3714		"mov %%"R"di, %c[rdi](%[svm]) \n\t"
3715		"mov %%"R"bp, %c[rbp](%[svm]) \n\t"
3716#ifdef CONFIG_X86_64
3717		"mov %%r8,  %c[r8](%[svm]) \n\t"
3718		"mov %%r9,  %c[r9](%[svm]) \n\t"
3719		"mov %%r10, %c[r10](%[svm]) \n\t"
3720		"mov %%r11, %c[r11](%[svm]) \n\t"
3721		"mov %%r12, %c[r12](%[svm]) \n\t"
3722		"mov %%r13, %c[r13](%[svm]) \n\t"
3723		"mov %%r14, %c[r14](%[svm]) \n\t"
3724		"mov %%r15, %c[r15](%[svm]) \n\t"
3725#endif
3726		"pop %%"R"bp"
3727		:
3728		: [svm]"a"(svm),
3729		  [vmcb]"i"(offsetof(struct vcpu_svm, vmcb_pa)),
3730		  [rbx]"i"(offsetof(struct vcpu_svm, vcpu.arch.regs[VCPU_REGS_RBX])),
3731		  [rcx]"i"(offsetof(struct vcpu_svm, vcpu.arch.regs[VCPU_REGS_RCX])),
3732		  [rdx]"i"(offsetof(struct vcpu_svm, vcpu.arch.regs[VCPU_REGS_RDX])),
3733		  [rsi]"i"(offsetof(struct vcpu_svm, vcpu.arch.regs[VCPU_REGS_RSI])),
3734		  [rdi]"i"(offsetof(struct vcpu_svm, vcpu.arch.regs[VCPU_REGS_RDI])),
3735		  [rbp]"i"(offsetof(struct vcpu_svm, vcpu.arch.regs[VCPU_REGS_RBP]))
3736#ifdef CONFIG_X86_64
3737		  , [r8]"i"(offsetof(struct vcpu_svm, vcpu.arch.regs[VCPU_REGS_R8])),
3738		  [r9]"i"(offsetof(struct vcpu_svm, vcpu.arch.regs[VCPU_REGS_R9])),
3739		  [r10]"i"(offsetof(struct vcpu_svm, vcpu.arch.regs[VCPU_REGS_R10])),
3740		  [r11]"i"(offsetof(struct vcpu_svm, vcpu.arch.regs[VCPU_REGS_R11])),
3741		  [r12]"i"(offsetof(struct vcpu_svm, vcpu.arch.regs[VCPU_REGS_R12])),
3742		  [r13]"i"(offsetof(struct vcpu_svm, vcpu.arch.regs[VCPU_REGS_R13])),
3743		  [r14]"i"(offsetof(struct vcpu_svm, vcpu.arch.regs[VCPU_REGS_R14])),
3744		  [r15]"i"(offsetof(struct vcpu_svm, vcpu.arch.regs[VCPU_REGS_R15]))
3745#endif
3746		: "cc", "memory"
3747		, R"bx", R"cx", R"dx", R"si", R"di"
3748#ifdef CONFIG_X86_64
 
3749		, "r8", "r9", "r10", "r11" , "r12", "r13", "r14", "r15"
 
 
3750#endif
3751		);
3752
3753#ifdef CONFIG_X86_64
3754	wrmsrl(MSR_GS_BASE, svm->host.gs_base);
3755#else
3756	loadsegment(fs, svm->host.fs);
3757#ifndef CONFIG_X86_32_LAZY_GS
3758	loadsegment(gs, svm->host.gs);
3759#endif
3760#endif
3761
3762	reload_tss(vcpu);
3763
3764	local_irq_disable();
3765
3766	vcpu->arch.cr2 = svm->vmcb->save.cr2;
3767	vcpu->arch.regs[VCPU_REGS_RAX] = svm->vmcb->save.rax;
3768	vcpu->arch.regs[VCPU_REGS_RSP] = svm->vmcb->save.rsp;
3769	vcpu->arch.regs[VCPU_REGS_RIP] = svm->vmcb->save.rip;
3770
 
 
3771	if (unlikely(svm->vmcb->control.exit_code == SVM_EXIT_NMI))
3772		kvm_before_handle_nmi(&svm->vcpu);
3773
3774	stgi();
3775
3776	/* Any pending NMI will happen here */
3777
3778	if (unlikely(svm->vmcb->control.exit_code == SVM_EXIT_NMI))
3779		kvm_after_handle_nmi(&svm->vcpu);
3780
3781	sync_cr8_to_lapic(vcpu);
3782
3783	svm->next_rip = 0;
3784
3785	svm->vmcb->control.tlb_ctl = TLB_CONTROL_DO_NOTHING;
3786
3787	/* if exit due to PF check for async PF */
3788	if (svm->vmcb->control.exit_code == SVM_EXIT_EXCP_BASE + PF_VECTOR)
3789		svm->apf_reason = kvm_read_and_reset_pf_reason();
3790
3791	if (npt_enabled) {
3792		vcpu->arch.regs_avail &= ~(1 << VCPU_EXREG_PDPTR);
3793		vcpu->arch.regs_dirty &= ~(1 << VCPU_EXREG_PDPTR);
3794	}
3795
3796	/*
3797	 * We need to handle MC intercepts here before the vcpu has a chance to
3798	 * change the physical cpu
3799	 */
3800	if (unlikely(svm->vmcb->control.exit_code ==
3801		     SVM_EXIT_EXCP_BASE + MC_VECTOR))
3802		svm_handle_mce(svm);
3803
3804	mark_all_clean(svm->vmcb);
3805}
3806
3807#undef R
3808
3809static void svm_set_cr3(struct kvm_vcpu *vcpu, unsigned long root)
3810{
3811	struct vcpu_svm *svm = to_svm(vcpu);
3812
3813	svm->vmcb->save.cr3 = root;
3814	mark_dirty(svm->vmcb, VMCB_CR);
3815	svm_flush_tlb(vcpu);
3816}
3817
3818static void set_tdp_cr3(struct kvm_vcpu *vcpu, unsigned long root)
3819{
3820	struct vcpu_svm *svm = to_svm(vcpu);
3821
3822	svm->vmcb->control.nested_cr3 = root;
3823	mark_dirty(svm->vmcb, VMCB_NPT);
3824
3825	/* Also sync guest cr3 here in case we live migrate */
3826	svm->vmcb->save.cr3 = kvm_read_cr3(vcpu);
3827	mark_dirty(svm->vmcb, VMCB_CR);
3828
3829	svm_flush_tlb(vcpu);
3830}
3831
3832static int is_disabled(void)
3833{
3834	u64 vm_cr;
3835
3836	rdmsrl(MSR_VM_CR, vm_cr);
3837	if (vm_cr & (1 << SVM_VM_CR_SVM_DISABLE))
3838		return 1;
3839
3840	return 0;
3841}
3842
3843static void
3844svm_patch_hypercall(struct kvm_vcpu *vcpu, unsigned char *hypercall)
3845{
3846	/*
3847	 * Patch in the VMMCALL instruction:
3848	 */
3849	hypercall[0] = 0x0f;
3850	hypercall[1] = 0x01;
3851	hypercall[2] = 0xd9;
3852}
3853
3854static void svm_check_processor_compat(void *rtn)
3855{
3856	*(int *)rtn = 0;
3857}
3858
3859static bool svm_cpu_has_accelerated_tpr(void)
3860{
3861	return false;
3862}
3863
3864static u64 svm_get_mt_mask(struct kvm_vcpu *vcpu, gfn_t gfn, bool is_mmio)
3865{
3866	return 0;
3867}
3868
3869static void svm_cpuid_update(struct kvm_vcpu *vcpu)
3870{
3871}
3872
3873static void svm_set_supported_cpuid(u32 func, struct kvm_cpuid_entry2 *entry)
3874{
3875	switch (func) {
3876	case 0x80000001:
3877		if (nested)
3878			entry->ecx |= (1 << 2); /* Set SVM bit */
3879		break;
3880	case 0x8000000A:
3881		entry->eax = 1; /* SVM revision 1 */
3882		entry->ebx = 8; /* Lets support 8 ASIDs in case we add proper
3883				   ASID emulation to nested SVM */
3884		entry->ecx = 0; /* Reserved */
3885		entry->edx = 0; /* Per default do not support any
3886				   additional features */
3887
3888		/* Support next_rip if host supports it */
3889		if (boot_cpu_has(X86_FEATURE_NRIPS))
3890			entry->edx |= SVM_FEATURE_NRIP;
3891
3892		/* Support NPT for the guest if enabled */
3893		if (npt_enabled)
3894			entry->edx |= SVM_FEATURE_NPT;
3895
3896		break;
3897	}
3898}
3899
3900static const struct trace_print_flags svm_exit_reasons_str[] = {
3901	{ SVM_EXIT_READ_CR0,			"read_cr0" },
3902	{ SVM_EXIT_READ_CR3,			"read_cr3" },
3903	{ SVM_EXIT_READ_CR4,			"read_cr4" },
3904	{ SVM_EXIT_READ_CR8,			"read_cr8" },
3905	{ SVM_EXIT_WRITE_CR0,			"write_cr0" },
3906	{ SVM_EXIT_WRITE_CR3,			"write_cr3" },
3907	{ SVM_EXIT_WRITE_CR4,			"write_cr4" },
3908	{ SVM_EXIT_WRITE_CR8,			"write_cr8" },
3909	{ SVM_EXIT_READ_DR0,			"read_dr0" },
3910	{ SVM_EXIT_READ_DR1,			"read_dr1" },
3911	{ SVM_EXIT_READ_DR2,			"read_dr2" },
3912	{ SVM_EXIT_READ_DR3,			"read_dr3" },
3913	{ SVM_EXIT_WRITE_DR0,			"write_dr0" },
3914	{ SVM_EXIT_WRITE_DR1,			"write_dr1" },
3915	{ SVM_EXIT_WRITE_DR2,			"write_dr2" },
3916	{ SVM_EXIT_WRITE_DR3,			"write_dr3" },
3917	{ SVM_EXIT_WRITE_DR5,			"write_dr5" },
3918	{ SVM_EXIT_WRITE_DR7,			"write_dr7" },
3919	{ SVM_EXIT_EXCP_BASE + DB_VECTOR,	"DB excp" },
3920	{ SVM_EXIT_EXCP_BASE + BP_VECTOR,	"BP excp" },
3921	{ SVM_EXIT_EXCP_BASE + UD_VECTOR,	"UD excp" },
3922	{ SVM_EXIT_EXCP_BASE + PF_VECTOR,	"PF excp" },
3923	{ SVM_EXIT_EXCP_BASE + NM_VECTOR,	"NM excp" },
3924	{ SVM_EXIT_EXCP_BASE + MC_VECTOR,	"MC excp" },
3925	{ SVM_EXIT_INTR,			"interrupt" },
3926	{ SVM_EXIT_NMI,				"nmi" },
3927	{ SVM_EXIT_SMI,				"smi" },
3928	{ SVM_EXIT_INIT,			"init" },
3929	{ SVM_EXIT_VINTR,			"vintr" },
3930	{ SVM_EXIT_CPUID,			"cpuid" },
3931	{ SVM_EXIT_INVD,			"invd" },
3932	{ SVM_EXIT_HLT,				"hlt" },
3933	{ SVM_EXIT_INVLPG,			"invlpg" },
3934	{ SVM_EXIT_INVLPGA,			"invlpga" },
3935	{ SVM_EXIT_IOIO,			"io" },
3936	{ SVM_EXIT_MSR,				"msr" },
3937	{ SVM_EXIT_TASK_SWITCH,			"task_switch" },
3938	{ SVM_EXIT_SHUTDOWN,			"shutdown" },
3939	{ SVM_EXIT_VMRUN,			"vmrun" },
3940	{ SVM_EXIT_VMMCALL,			"hypercall" },
3941	{ SVM_EXIT_VMLOAD,			"vmload" },
3942	{ SVM_EXIT_VMSAVE,			"vmsave" },
3943	{ SVM_EXIT_STGI,			"stgi" },
3944	{ SVM_EXIT_CLGI,			"clgi" },
3945	{ SVM_EXIT_SKINIT,			"skinit" },
3946	{ SVM_EXIT_WBINVD,			"wbinvd" },
3947	{ SVM_EXIT_MONITOR,			"monitor" },
3948	{ SVM_EXIT_MWAIT,			"mwait" },
3949	{ SVM_EXIT_XSETBV,			"xsetbv" },
3950	{ SVM_EXIT_NPF,				"npf" },
3951	{ -1, NULL }
3952};
3953
3954static int svm_get_lpage_level(void)
3955{
3956	return PT_PDPE_LEVEL;
3957}
3958
3959static bool svm_rdtscp_supported(void)
3960{
3961	return false;
3962}
3963
 
 
 
 
 
 
 
 
 
 
3964static bool svm_has_wbinvd_exit(void)
3965{
3966	return true;
3967}
3968
3969static void svm_fpu_deactivate(struct kvm_vcpu *vcpu)
3970{
3971	struct vcpu_svm *svm = to_svm(vcpu);
3972
3973	set_exception_intercept(svm, NM_VECTOR);
3974	update_cr0_intercept(svm);
3975}
3976
3977#define PRE_EX(exit)  { .exit_code = (exit), \
3978			.stage = X86_ICPT_PRE_EXCEPT, }
3979#define POST_EX(exit) { .exit_code = (exit), \
3980			.stage = X86_ICPT_POST_EXCEPT, }
3981#define POST_MEM(exit) { .exit_code = (exit), \
3982			.stage = X86_ICPT_POST_MEMACCESS, }
3983
3984static struct __x86_intercept {
3985	u32 exit_code;
3986	enum x86_intercept_stage stage;
3987} x86_intercept_map[] = {
3988	[x86_intercept_cr_read]		= POST_EX(SVM_EXIT_READ_CR0),
3989	[x86_intercept_cr_write]	= POST_EX(SVM_EXIT_WRITE_CR0),
3990	[x86_intercept_clts]		= POST_EX(SVM_EXIT_WRITE_CR0),
3991	[x86_intercept_lmsw]		= POST_EX(SVM_EXIT_WRITE_CR0),
3992	[x86_intercept_smsw]		= POST_EX(SVM_EXIT_READ_CR0),
3993	[x86_intercept_dr_read]		= POST_EX(SVM_EXIT_READ_DR0),
3994	[x86_intercept_dr_write]	= POST_EX(SVM_EXIT_WRITE_DR0),
3995	[x86_intercept_sldt]		= POST_EX(SVM_EXIT_LDTR_READ),
3996	[x86_intercept_str]		= POST_EX(SVM_EXIT_TR_READ),
3997	[x86_intercept_lldt]		= POST_EX(SVM_EXIT_LDTR_WRITE),
3998	[x86_intercept_ltr]		= POST_EX(SVM_EXIT_TR_WRITE),
3999	[x86_intercept_sgdt]		= POST_EX(SVM_EXIT_GDTR_READ),
4000	[x86_intercept_sidt]		= POST_EX(SVM_EXIT_IDTR_READ),
4001	[x86_intercept_lgdt]		= POST_EX(SVM_EXIT_GDTR_WRITE),
4002	[x86_intercept_lidt]		= POST_EX(SVM_EXIT_IDTR_WRITE),
4003	[x86_intercept_vmrun]		= POST_EX(SVM_EXIT_VMRUN),
4004	[x86_intercept_vmmcall]		= POST_EX(SVM_EXIT_VMMCALL),
4005	[x86_intercept_vmload]		= POST_EX(SVM_EXIT_VMLOAD),
4006	[x86_intercept_vmsave]		= POST_EX(SVM_EXIT_VMSAVE),
4007	[x86_intercept_stgi]		= POST_EX(SVM_EXIT_STGI),
4008	[x86_intercept_clgi]		= POST_EX(SVM_EXIT_CLGI),
4009	[x86_intercept_skinit]		= POST_EX(SVM_EXIT_SKINIT),
4010	[x86_intercept_invlpga]		= POST_EX(SVM_EXIT_INVLPGA),
4011	[x86_intercept_rdtscp]		= POST_EX(SVM_EXIT_RDTSCP),
4012	[x86_intercept_monitor]		= POST_MEM(SVM_EXIT_MONITOR),
4013	[x86_intercept_mwait]		= POST_EX(SVM_EXIT_MWAIT),
4014	[x86_intercept_invlpg]		= POST_EX(SVM_EXIT_INVLPG),
4015	[x86_intercept_invd]		= POST_EX(SVM_EXIT_INVD),
4016	[x86_intercept_wbinvd]		= POST_EX(SVM_EXIT_WBINVD),
4017	[x86_intercept_wrmsr]		= POST_EX(SVM_EXIT_MSR),
4018	[x86_intercept_rdtsc]		= POST_EX(SVM_EXIT_RDTSC),
4019	[x86_intercept_rdmsr]		= POST_EX(SVM_EXIT_MSR),
4020	[x86_intercept_rdpmc]		= POST_EX(SVM_EXIT_RDPMC),
4021	[x86_intercept_cpuid]		= PRE_EX(SVM_EXIT_CPUID),
4022	[x86_intercept_rsm]		= PRE_EX(SVM_EXIT_RSM),
4023	[x86_intercept_pause]		= PRE_EX(SVM_EXIT_PAUSE),
4024	[x86_intercept_pushf]		= PRE_EX(SVM_EXIT_PUSHF),
4025	[x86_intercept_popf]		= PRE_EX(SVM_EXIT_POPF),
4026	[x86_intercept_intn]		= PRE_EX(SVM_EXIT_SWINT),
4027	[x86_intercept_iret]		= PRE_EX(SVM_EXIT_IRET),
4028	[x86_intercept_icebp]		= PRE_EX(SVM_EXIT_ICEBP),
4029	[x86_intercept_hlt]		= POST_EX(SVM_EXIT_HLT),
4030	[x86_intercept_in]		= POST_EX(SVM_EXIT_IOIO),
4031	[x86_intercept_ins]		= POST_EX(SVM_EXIT_IOIO),
4032	[x86_intercept_out]		= POST_EX(SVM_EXIT_IOIO),
4033	[x86_intercept_outs]		= POST_EX(SVM_EXIT_IOIO),
4034};
4035
4036#undef PRE_EX
4037#undef POST_EX
4038#undef POST_MEM
4039
4040static int svm_check_intercept(struct kvm_vcpu *vcpu,
4041			       struct x86_instruction_info *info,
4042			       enum x86_intercept_stage stage)
4043{
4044	struct vcpu_svm *svm = to_svm(vcpu);
4045	int vmexit, ret = X86EMUL_CONTINUE;
4046	struct __x86_intercept icpt_info;
4047	struct vmcb *vmcb = svm->vmcb;
4048
4049	if (info->intercept >= ARRAY_SIZE(x86_intercept_map))
4050		goto out;
4051
4052	icpt_info = x86_intercept_map[info->intercept];
4053
4054	if (stage != icpt_info.stage)
4055		goto out;
4056
4057	switch (icpt_info.exit_code) {
4058	case SVM_EXIT_READ_CR0:
4059		if (info->intercept == x86_intercept_cr_read)
4060			icpt_info.exit_code += info->modrm_reg;
4061		break;
4062	case SVM_EXIT_WRITE_CR0: {
4063		unsigned long cr0, val;
4064		u64 intercept;
4065
4066		if (info->intercept == x86_intercept_cr_write)
4067			icpt_info.exit_code += info->modrm_reg;
4068
4069		if (icpt_info.exit_code != SVM_EXIT_WRITE_CR0)
4070			break;
4071
4072		intercept = svm->nested.intercept;
4073
4074		if (!(intercept & (1ULL << INTERCEPT_SELECTIVE_CR0)))
4075			break;
4076
4077		cr0 = vcpu->arch.cr0 & ~SVM_CR0_SELECTIVE_MASK;
4078		val = info->src_val  & ~SVM_CR0_SELECTIVE_MASK;
4079
4080		if (info->intercept == x86_intercept_lmsw) {
4081			cr0 &= 0xfUL;
4082			val &= 0xfUL;
4083			/* lmsw can't clear PE - catch this here */
4084			if (cr0 & X86_CR0_PE)
4085				val |= X86_CR0_PE;
4086		}
4087
4088		if (cr0 ^ val)
4089			icpt_info.exit_code = SVM_EXIT_CR0_SEL_WRITE;
4090
4091		break;
4092	}
4093	case SVM_EXIT_READ_DR0:
4094	case SVM_EXIT_WRITE_DR0:
4095		icpt_info.exit_code += info->modrm_reg;
4096		break;
4097	case SVM_EXIT_MSR:
4098		if (info->intercept == x86_intercept_wrmsr)
4099			vmcb->control.exit_info_1 = 1;
4100		else
4101			vmcb->control.exit_info_1 = 0;
4102		break;
4103	case SVM_EXIT_PAUSE:
4104		/*
4105		 * We get this for NOP only, but pause
4106		 * is rep not, check this here
4107		 */
4108		if (info->rep_prefix != REPE_PREFIX)
4109			goto out;
4110	case SVM_EXIT_IOIO: {
4111		u64 exit_info;
4112		u32 bytes;
4113
4114		exit_info = (vcpu->arch.regs[VCPU_REGS_RDX] & 0xffff) << 16;
4115
4116		if (info->intercept == x86_intercept_in ||
4117		    info->intercept == x86_intercept_ins) {
4118			exit_info |= SVM_IOIO_TYPE_MASK;
4119			bytes = info->src_bytes;
4120		} else {
4121			bytes = info->dst_bytes;
4122		}
4123
4124		if (info->intercept == x86_intercept_outs ||
4125		    info->intercept == x86_intercept_ins)
4126			exit_info |= SVM_IOIO_STR_MASK;
4127
4128		if (info->rep_prefix)
4129			exit_info |= SVM_IOIO_REP_MASK;
4130
4131		bytes = min(bytes, 4u);
4132
4133		exit_info |= bytes << SVM_IOIO_SIZE_SHIFT;
4134
4135		exit_info |= (u32)info->ad_bytes << (SVM_IOIO_ASIZE_SHIFT - 1);
4136
4137		vmcb->control.exit_info_1 = exit_info;
4138		vmcb->control.exit_info_2 = info->next_rip;
4139
4140		break;
4141	}
4142	default:
4143		break;
4144	}
4145
4146	vmcb->control.next_rip  = info->next_rip;
4147	vmcb->control.exit_code = icpt_info.exit_code;
4148	vmexit = nested_svm_exit_handled(svm);
4149
4150	ret = (vmexit == NESTED_EXIT_DONE) ? X86EMUL_INTERCEPTED
4151					   : X86EMUL_CONTINUE;
4152
4153out:
4154	return ret;
4155}
4156
 
 
 
 
 
4157static struct kvm_x86_ops svm_x86_ops = {
4158	.cpu_has_kvm_support = has_svm,
4159	.disabled_by_bios = is_disabled,
4160	.hardware_setup = svm_hardware_setup,
4161	.hardware_unsetup = svm_hardware_unsetup,
4162	.check_processor_compatibility = svm_check_processor_compat,
4163	.hardware_enable = svm_hardware_enable,
4164	.hardware_disable = svm_hardware_disable,
4165	.cpu_has_accelerated_tpr = svm_cpu_has_accelerated_tpr,
4166
4167	.vcpu_create = svm_create_vcpu,
4168	.vcpu_free = svm_free_vcpu,
4169	.vcpu_reset = svm_vcpu_reset,
4170
4171	.prepare_guest_switch = svm_prepare_guest_switch,
4172	.vcpu_load = svm_vcpu_load,
4173	.vcpu_put = svm_vcpu_put,
4174
4175	.set_guest_debug = svm_guest_debug,
4176	.get_msr = svm_get_msr,
4177	.set_msr = svm_set_msr,
4178	.get_segment_base = svm_get_segment_base,
4179	.get_segment = svm_get_segment,
4180	.set_segment = svm_set_segment,
4181	.get_cpl = svm_get_cpl,
4182	.get_cs_db_l_bits = kvm_get_cs_db_l_bits,
4183	.decache_cr0_guest_bits = svm_decache_cr0_guest_bits,
4184	.decache_cr3 = svm_decache_cr3,
4185	.decache_cr4_guest_bits = svm_decache_cr4_guest_bits,
4186	.set_cr0 = svm_set_cr0,
4187	.set_cr3 = svm_set_cr3,
4188	.set_cr4 = svm_set_cr4,
4189	.set_efer = svm_set_efer,
4190	.get_idt = svm_get_idt,
4191	.set_idt = svm_set_idt,
4192	.get_gdt = svm_get_gdt,
4193	.set_gdt = svm_set_gdt,
 
 
4194	.set_dr7 = svm_set_dr7,
 
4195	.cache_reg = svm_cache_reg,
4196	.get_rflags = svm_get_rflags,
4197	.set_rflags = svm_set_rflags,
4198	.fpu_activate = svm_fpu_activate,
4199	.fpu_deactivate = svm_fpu_deactivate,
4200
4201	.tlb_flush = svm_flush_tlb,
4202
4203	.run = svm_vcpu_run,
4204	.handle_exit = handle_exit,
4205	.skip_emulated_instruction = skip_emulated_instruction,
4206	.set_interrupt_shadow = svm_set_interrupt_shadow,
4207	.get_interrupt_shadow = svm_get_interrupt_shadow,
4208	.patch_hypercall = svm_patch_hypercall,
4209	.set_irq = svm_set_irq,
4210	.set_nmi = svm_inject_nmi,
4211	.queue_exception = svm_queue_exception,
4212	.cancel_injection = svm_cancel_injection,
4213	.interrupt_allowed = svm_interrupt_allowed,
4214	.nmi_allowed = svm_nmi_allowed,
4215	.get_nmi_mask = svm_get_nmi_mask,
4216	.set_nmi_mask = svm_set_nmi_mask,
4217	.enable_nmi_window = enable_nmi_window,
4218	.enable_irq_window = enable_irq_window,
4219	.update_cr8_intercept = update_cr8_intercept,
 
 
 
 
 
4220
4221	.set_tss_addr = svm_set_tss_addr,
4222	.get_tdp_level = get_npt_level,
4223	.get_mt_mask = svm_get_mt_mask,
4224
4225	.get_exit_info = svm_get_exit_info,
4226	.exit_reasons_str = svm_exit_reasons_str,
4227
4228	.get_lpage_level = svm_get_lpage_level,
4229
4230	.cpuid_update = svm_cpuid_update,
4231
4232	.rdtscp_supported = svm_rdtscp_supported,
 
 
4233
4234	.set_supported_cpuid = svm_set_supported_cpuid,
4235
4236	.has_wbinvd_exit = svm_has_wbinvd_exit,
4237
4238	.set_tsc_khz = svm_set_tsc_khz,
 
4239	.write_tsc_offset = svm_write_tsc_offset,
4240	.adjust_tsc_offset = svm_adjust_tsc_offset,
4241	.compute_tsc_offset = svm_compute_tsc_offset,
 
4242
4243	.set_tdp_cr3 = set_tdp_cr3,
4244
4245	.check_intercept = svm_check_intercept,
 
4246};
4247
4248static int __init svm_init(void)
4249{
4250	return kvm_init(&svm_x86_ops, sizeof(struct vcpu_svm),
4251			__alignof__(struct vcpu_svm), THIS_MODULE);
4252}
4253
4254static void __exit svm_exit(void)
4255{
4256	kvm_exit();
4257}
4258
4259module_init(svm_init)
4260module_exit(svm_exit)