arch/x86/kvm/svm.c

   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 affilates.
   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
  35 #include <asm/virtext.h>
  36 #include "trace.h"
  37
  38 #define __ex(x) __kvm_handle_fault_on_reboot(x)
  39
  40 MODULE_AUTHOR("Qumranet");
  41 MODULE_LICENSE("GPL");
  42
  43 #define IOPM_ALLOC_ORDER 2
  44 #define MSRPM_ALLOC_ORDER 1
  45
  46 #define SEG_TYPE_LDT 2
  47 #define SEG_TYPE_BUSY_TSS16 3
  48
  49 #define SVM_FEATURE_NPT            (1 <<  0)
  50 #define SVM_FEATURE_LBRV           (1 <<  1)
  51 #define SVM_FEATURE_SVML           (1 <<  2)
  52 #define SVM_FEATURE_NRIP           (1 <<  3)
  53 #define SVM_FEATURE_PAUSE_FILTER   (1 << 10)
  54
  55 #define NESTED_EXIT_HOST        0       /* Exit handled on host level */
  56 #define NESTED_EXIT_DONE        1       /* Exit caused nested vmexit  */
  57 #define NESTED_EXIT_CONTINUE    2       /* Further checks needed      */
  58
  59 #define DEBUGCTL_RESERVED_BITS (~(0x3fULL))
  60
  61 static bool erratum_383_found __read_mostly;
  62
  63 static const u32 host_save_user_msrs[] = {
  64 #ifdef CONFIG_X86_64
  65         MSR_STAR, MSR_LSTAR, MSR_CSTAR, MSR_SYSCALL_MASK, MSR_KERNEL_GS_BASE,
  66         MSR_FS_BASE,
  67 #endif
  68         MSR_IA32_SYSENTER_CS, MSR_IA32_SYSENTER_ESP, MSR_IA32_SYSENTER_EIP,
  69 };
  70
  71 #define NR_HOST_SAVE_USER_MSRS ARRAY_SIZE(host_save_user_msrs)
  72
  73 struct kvm_vcpu;
  74
  75 struct nested_state {
  76         struct vmcb *hsave;
  77         u64 hsave_msr;
  78         u64 vm_cr_msr;
  79         u64 vmcb;
  80
  81         /* These are the merged vectors */
  82         u32 *msrpm;
  83
  84         /* gpa pointers to the real vectors */
  85         u64 vmcb_msrpm;
  86         u64 vmcb_iopm;
  87
  88         /* A VMEXIT is required but not yet emulated */
  89         bool exit_required;
  90
  91         /*
  92          * If we vmexit during an instruction emulation we need this to restore
  93          * the l1 guest rip after the emulation
  94          */
  95         unsigned long vmexit_rip;
  96         unsigned long vmexit_rsp;
  97         unsigned long vmexit_rax;
  98
  99         /* cache for intercepts of the guest */
 100         u16 intercept_cr_read;
 101         u16 intercept_cr_write;
 102         u16 intercept_dr_read;
 103         u16 intercept_dr_write;
 104         u32 intercept_exceptions;
 105         u64 intercept;
 106
 107         /* Nested Paging related state */
 108         u64 nested_cr3;
 109 };
 110
 111 #define MSRPM_OFFSETS   16
 112 static u32 msrpm_offsets[MSRPM_OFFSETS] __read_mostly;
 113
 114 struct vcpu_svm {
 115         struct kvm_vcpu vcpu;
 116         struct vmcb *vmcb;
 117         unsigned long vmcb_pa;
 118         struct svm_cpu_data *svm_data;
 119         uint64_t asid_generation;
 120         uint64_t sysenter_esp;
 121         uint64_t sysenter_eip;
 122
 123         u64 next_rip;
 124
 125         u64 host_user_msrs[NR_HOST_SAVE_USER_MSRS];
 126         u64 host_gs_base;
 127
 128         u32 *msrpm;
 129
 130         struct nested_state nested;
 131
 132         bool nmi_singlestep;
 133
 134         unsigned int3_injected;
 135         unsigned long int3_rip;
 136 };
 137
 138 #define MSR_INVALID                     0xffffffffU
 139
 140 static struct svm_direct_access_msrs {
 141         u32 index;   /* Index of the MSR */
 142         bool always; /* True if intercept is always on */
 143 } direct_access_msrs[] = {
 144         { .index = MSR_STAR,                            .always = true  },
 145         { .index = MSR_IA32_SYSENTER_CS,                .always = true  },
 146 #ifdef CONFIG_X86_64
 147         { .index = MSR_GS_BASE,                         .always = true  },
 148         { .index = MSR_FS_BASE,                         .always = true  },
 149         { .index = MSR_KERNEL_GS_BASE,                  .always = true  },
 150         { .index = MSR_LSTAR,                           .always = true  },
 151         { .index = MSR_CSTAR,                           .always = true  },
 152         { .index = MSR_SYSCALL_MASK,                    .always = true  },
 153 #endif
 154         { .index = MSR_IA32_LASTBRANCHFROMIP,           .always = false },
 155         { .index = MSR_IA32_LASTBRANCHTOIP,             .always = false },
 156         { .index = MSR_IA32_LASTINTFROMIP,              .always = false },
 157         { .index = MSR_IA32_LASTINTTOIP,                .always = false },
 158         { .index = MSR_INVALID,                         .always = false },
 159 };
 160
 161 /* enable NPT for AMD64 and X86 with PAE */
 162 #if defined(CONFIG_X86_64) || defined(CONFIG_X86_PAE)
 163 static bool npt_enabled = true;
 164 #else
 165 static bool npt_enabled;
 166 #endif
 167 static int npt = 1;
 168
 169 module_param(npt, int, S_IRUGO);
 170
 171 static int nested = 1;
 172 module_param(nested, int, S_IRUGO);
 173
 174 static void svm_flush_tlb(struct kvm_vcpu *vcpu);
 175 static void svm_complete_interrupts(struct vcpu_svm *svm);
 176
 177 static int nested_svm_exit_handled(struct vcpu_svm *svm);
 178 static int nested_svm_intercept(struct vcpu_svm *svm);
 179 static int nested_svm_vmexit(struct vcpu_svm *svm);
 180 static int nested_svm_check_exception(struct vcpu_svm *svm, unsigned nr,
 181                                       bool has_error_code, u32 error_code);
 182
 183 static inline struct vcpu_svm *to_svm(struct kvm_vcpu *vcpu)
 184 {
 185         return container_of(vcpu, struct vcpu_svm, vcpu);
 186 }
 187
 188 static inline bool is_nested(struct vcpu_svm *svm)
 189 {
 190         return svm->nested.vmcb;
 191 }
 192
 193 static inline void enable_gif(struct vcpu_svm *svm)
 194 {
 195         svm->vcpu.arch.hflags |= HF_GIF_MASK;
 196 }
 197
 198 static inline void disable_gif(struct vcpu_svm *svm)
 199 {
 200         svm->vcpu.arch.hflags &= ~HF_GIF_MASK;
 201 }
 202
 203 static inline bool gif_set(struct vcpu_svm *svm)
 204 {
 205         return !!(svm->vcpu.arch.hflags & HF_GIF_MASK);
 206 }
 207
 208 static unsigned long iopm_base;
 209
 210 struct kvm_ldttss_desc {
 211         u16 limit0;
 212         u16 base0;
 213         unsigned base1:8, type:5, dpl:2, p:1;
 214         unsigned limit1:4, zero0:3, g:1, base2:8;
 215         u32 base3;
 216         u32 zero1;
 217 } __attribute__((packed));
 218
 219 struct svm_cpu_data {
 220         int cpu;
 221
 222         u64 asid_generation;
 223         u32 max_asid;
 224         u32 next_asid;
 225         struct kvm_ldttss_desc *tss_desc;
 226
 227         struct page *save_area;
 228 };
 229
 230 static DEFINE_PER_CPU(struct svm_cpu_data *, svm_data);
 231 static uint32_t svm_features;
 232
 233 struct svm_init_data {
 234         int cpu;
 235         int r;
 236 };
 237
 238 static u32 msrpm_ranges[] = {0, 0xc0000000, 0xc0010000};
 239
 240 #define NUM_MSR_MAPS ARRAY_SIZE(msrpm_ranges)
 241 #define MSRS_RANGE_SIZE 2048
 242 #define MSRS_IN_RANGE (MSRS_RANGE_SIZE * 8 / 2)
 243
 244 static u32 svm_msrpm_offset(u32 msr)
 245 {
 246         u32 offset;
 247         int i;
 248
 249         for (i = 0; i < NUM_MSR_MAPS; i++) {
 250                 if (msr < msrpm_ranges[i] ||
 251                     msr >= msrpm_ranges[i] + MSRS_IN_RANGE)
 252                         continue;
 253
 254                 offset  = (msr - msrpm_ranges[i]) / 4; /* 4 msrs per u8 */
 255                 offset += (i * MSRS_RANGE_SIZE);       /* add range offset */
 256
 257                 /* Now we have the u8 offset - but need the u32 offset */
 258                 return offset / 4;
 259         }
 260
 261         /* MSR not in any range */
 262         return MSR_INVALID;
 263 }
 264
 265 #define MAX_INST_SIZE 15
 266
 267 static inline u32 svm_has(u32 feat)
 268 {
 269         return svm_features & feat;
 270 }
 271
 272 static inline void clgi(void)
 273 {
 274         asm volatile (__ex(SVM_CLGI));
 275 }
 276
 277 static inline void stgi(void)
 278 {
 279         asm volatile (__ex(SVM_STGI));
 280 }
 281
 282 static inline void invlpga(unsigned long addr, u32 asid)
 283 {
 284         asm volatile (__ex(SVM_INVLPGA) : : "a"(addr), "c"(asid));
 285 }
 286
 287 static inline void force_new_asid(struct kvm_vcpu *vcpu)
 288 {
 289         to_svm(vcpu)->asid_generation--;
 290 }
 291
 292 static inline void flush_guest_tlb(struct kvm_vcpu *vcpu)
 293 {
 294         force_new_asid(vcpu);
 295 }
 296
 297 static int get_npt_level(void)
 298 {
 299 #ifdef CONFIG_X86_64
 300         return PT64_ROOT_LEVEL;
 301 #else
 302         return PT32E_ROOT_LEVEL;
 303 #endif
 304 }
 305
 306 static void svm_set_efer(struct kvm_vcpu *vcpu, u64 efer)
 307 {
 308         vcpu->arch.efer = efer;
 309         if (!npt_enabled && !(efer & EFER_LMA))
 310                 efer &= ~EFER_LME;
 311
 312         to_svm(vcpu)->vmcb->save.efer = efer | EFER_SVME;
 313 }
 314
 315 static int is_external_interrupt(u32 info)
 316 {
 317         info &= SVM_EVTINJ_TYPE_MASK | SVM_EVTINJ_VALID;
 318         return info == (SVM_EVTINJ_VALID | SVM_EVTINJ_TYPE_INTR);
 319 }
 320
 321 static u32 svm_get_interrupt_shadow(struct kvm_vcpu *vcpu, int mask)
 322 {
 323         struct vcpu_svm *svm = to_svm(vcpu);
 324         u32 ret = 0;
 325
 326         if (svm->vmcb->control.int_state & SVM_INTERRUPT_SHADOW_MASK)
 327                 ret |= KVM_X86_SHADOW_INT_STI | KVM_X86_SHADOW_INT_MOV_SS;
 328         return ret & mask;
 329 }
 330
 331 static void svm_set_interrupt_shadow(struct kvm_vcpu *vcpu, int mask)
 332 {
 333         struct vcpu_svm *svm = to_svm(vcpu);
 334
 335         if (mask == 0)
 336                 svm->vmcb->control.int_state &= ~SVM_INTERRUPT_SHADOW_MASK;
 337         else
 338                 svm->vmcb->control.int_state |= SVM_INTERRUPT_SHADOW_MASK;
 339
 340 }
 341
 342 static void skip_emulated_instruction(struct kvm_vcpu *vcpu)
 343 {
 344         struct vcpu_svm *svm = to_svm(vcpu);
 345
 346         if (svm->vmcb->control.next_rip != 0)
 347                 svm->next_rip = svm->vmcb->control.next_rip;
 348
 349         if (!svm->next_rip) {
 350                 if (emulate_instruction(vcpu, 0, 0, EMULTYPE_SKIP) !=
 351                                 EMULATE_DONE)
 352                         printk(KERN_DEBUG "%s: NOP\n", __func__);
 353                 return;
 354         }
 355         if (svm->next_rip - kvm_rip_read(vcpu) > MAX_INST_SIZE)
 356                 printk(KERN_ERR "%s: ip 0x%lx next 0x%llx\n",
 357                        __func__, kvm_rip_read(vcpu), svm->next_rip);
 358
 359         kvm_rip_write(vcpu, svm->next_rip);
 360         svm_set_interrupt_shadow(vcpu, 0);
 361 }
 362
 363 static void svm_queue_exception(struct kvm_vcpu *vcpu, unsigned nr,
 364                                 bool has_error_code, u32 error_code,
 365                                 bool reinject)
 366 {
 367         struct vcpu_svm *svm = to_svm(vcpu);
 368
 369         /*
 370          * If we are within a nested VM we'd better #VMEXIT and let the guest
 371          * handle the exception
 372          */
 373         if (!reinject &&
 374             nested_svm_check_exception(svm, nr, has_error_code, error_code))
 375                 return;
 376
 377         if (nr == BP_VECTOR && !svm_has(SVM_FEATURE_NRIP)) {
 378                 unsigned long rip, old_rip = kvm_rip_read(&svm->vcpu);
 379
 380                 /*
 381                  * For guest debugging where we have to reinject #BP if some
 382                  * INT3 is guest-owned:
 383                  * Emulate nRIP by moving RIP forward. Will fail if injection
 384                  * raises a fault that is not intercepted. Still better than
 385                  * failing in all cases.
 386                  */
 387                 skip_emulated_instruction(&svm->vcpu);
 388                 rip = kvm_rip_read(&svm->vcpu);
 389                 svm->int3_rip = rip + svm->vmcb->save.cs.base;
 390                 svm->int3_injected = rip - old_rip;
 391         }
 392
 393         svm->vmcb->control.event_inj = nr
 394                 | SVM_EVTINJ_VALID
 395                 | (has_error_code ? SVM_EVTINJ_VALID_ERR : 0)
 396                 | SVM_EVTINJ_TYPE_EXEPT;
 397         svm->vmcb->control.event_inj_err = error_code;
 398 }
 399
 400 static void svm_init_erratum_383(void)
 401 {
 402         u32 low, high;
 403         int err;
 404         u64 val;
 405
 406         if (!cpu_has_amd_erratum(amd_erratum_383))
 407                 return;
 408
 409         /* Use _safe variants to not break nested virtualization */
 410         val = native_read_msr_safe(MSR_AMD64_DC_CFG, &err);
 411         if (err)
 412                 return;
 413
 414         val |= (1ULL << 47);
 415
 416         low  = lower_32_bits(val);
 417         high = upper_32_bits(val);
 418
 419         native_write_msr_safe(MSR_AMD64_DC_CFG, low, high);
 420
 421         erratum_383_found = true;
 422 }
 423
 424 static int has_svm(void)
 425 {
 426         const char *msg;
 427
 428         if (!cpu_has_svm(&msg)) {
 429                 printk(KERN_INFO "has_svm: %s\n", msg);
 430                 return 0;
 431         }
 432
 433         return 1;
 434 }
 435
 436 static void svm_hardware_disable(void *garbage)
 437 {
 438         cpu_svm_disable();
 439 }
 440
 441 static int svm_hardware_enable(void *garbage)
 442 {
 443
 444         struct svm_cpu_data *sd;
 445         uint64_t efer;
 446         struct desc_ptr gdt_descr;
 447         struct desc_struct *gdt;
 448         int me = raw_smp_processor_id();
 449
 450         rdmsrl(MSR_EFER, efer);
 451         if (efer & EFER_SVME)
 452                 return -EBUSY;
 453
 454         if (!has_svm()) {
 455                 printk(KERN_ERR "svm_hardware_enable: err EOPNOTSUPP on %d\n",
 456                        me);
 457                 return -EINVAL;
 458         }
 459         sd = per_cpu(svm_data, me);
 460
 461         if (!sd) {
 462                 printk(KERN_ERR "svm_hardware_enable: svm_data is NULL on %d\n",
 463                        me);
 464                 return -EINVAL;
 465         }
 466
 467         sd->asid_generation = 1;
 468         sd->max_asid = cpuid_ebx(SVM_CPUID_FUNC) - 1;
 469         sd->next_asid = sd->max_asid + 1;
 470
 471         native_store_gdt(&gdt_descr);
 472         gdt = (struct desc_struct *)gdt_descr.address;
 473         sd->tss_desc = (struct kvm_ldttss_desc *)(gdt + GDT_ENTRY_TSS);
 474
 475         wrmsrl(MSR_EFER, efer | EFER_SVME);
 476
 477         wrmsrl(MSR_VM_HSAVE_PA, page_to_pfn(sd->save_area) << PAGE_SHIFT);
 478
 479         svm_init_erratum_383();
 480
 481         return 0;
 482 }
 483
 484 static void svm_cpu_uninit(int cpu)
 485 {
 486         struct svm_cpu_data *sd = per_cpu(svm_data, raw_smp_processor_id());
 487
 488         if (!sd)
 489                 return;
 490
 491         per_cpu(svm_data, raw_smp_processor_id()) = NULL;
 492         __free_page(sd->save_area);
 493         kfree(sd);
 494 }
 495
 496 static int svm_cpu_init(int cpu)
 497 {
 498         struct svm_cpu_data *sd;
 499         int r;
 500
 501         sd = kzalloc(sizeof(struct svm_cpu_data), GFP_KERNEL);
 502         if (!sd)
 503                 return -ENOMEM;
 504         sd->cpu = cpu;
 505         sd->save_area = alloc_page(GFP_KERNEL);
 506         r = -ENOMEM;
 507         if (!sd->save_area)
 508                 goto err_1;
 509
 510         per_cpu(svm_data, cpu) = sd;
 511
 512         return 0;
 513
 514 err_1:
 515         kfree(sd);
 516         return r;
 517
 518 }
 519
 520 static bool valid_msr_intercept(u32 index)
 521 {
 522         int i;
 523
 524         for (i = 0; direct_access_msrs[i].index != MSR_INVALID; i++)
 525                 if (direct_access_msrs[i].index == index)
 526                         return true;
 527
 528         return false;
 529 }
 530
 531 static void set_msr_interception(u32 *msrpm, unsigned msr,
 532                                  int read, int write)
 533 {
 534         u8 bit_read, bit_write;
 535         unsigned long tmp;
 536         u32 offset;
 537
 538         /*
 539          * If this warning triggers extend the direct_access_msrs list at the
 540          * beginning of the file
 541          */
 542         WARN_ON(!valid_msr_intercept(msr));
 543
 544         offset    = svm_msrpm_offset(msr);
 545         bit_read  = 2 * (msr & 0x0f);
 546         bit_write = 2 * (msr & 0x0f) + 1;
 547         tmp       = msrpm[offset];
 548
 549         BUG_ON(offset == MSR_INVALID);
 550
 551         read  ? clear_bit(bit_read,  &tmp) : set_bit(bit_read,  &tmp);
 552         write ? clear_bit(bit_write, &tmp) : set_bit(bit_write, &tmp);
 553
 554         msrpm[offset] = tmp;
 555 }
 556
 557 static void svm_vcpu_init_msrpm(u32 *msrpm)
 558 {
 559         int i;
 560
 561         memset(msrpm, 0xff, PAGE_SIZE * (1 << MSRPM_ALLOC_ORDER));
 562
 563         for (i = 0; direct_access_msrs[i].index != MSR_INVALID; i++) {
 564                 if (!direct_access_msrs[i].always)
 565                         continue;
 566
 567                 set_msr_interception(msrpm, direct_access_msrs[i].index, 1, 1);
 568         }
 569 }
 570
 571 static void add_msr_offset(u32 offset)
 572 {
 573         int i;
 574
 575         for (i = 0; i < MSRPM_OFFSETS; ++i) {
 576
 577                 /* Offset already in list? */
 578                 if (msrpm_offsets[i] == offset)
 579                         return;
 580
 581                 /* Slot used by another offset? */
 582                 if (msrpm_offsets[i] != MSR_INVALID)
 583                         continue;
 584
 585                 /* Add offset to list */
 586                 msrpm_offsets[i] = offset;
 587
 588                 return;
 589         }
 590
 591         /*
 592          * If this BUG triggers the msrpm_offsets table has an overflow. Just
 593          * increase MSRPM_OFFSETS in this case.
 594          */
 595         BUG();
 596 }
 597
 598 static void init_msrpm_offsets(void)
 599 {
 600         int i;
 601
 602         memset(msrpm_offsets, 0xff, sizeof(msrpm_offsets));
 603
 604         for (i = 0; direct_access_msrs[i].index != MSR_INVALID; i++) {
 605                 u32 offset;
 606
 607                 offset = svm_msrpm_offset(direct_access_msrs[i].index);
 608                 BUG_ON(offset == MSR_INVALID);
 609
 610                 add_msr_offset(offset);
 611         }
 612 }
 613
 614 static void svm_enable_lbrv(struct vcpu_svm *svm)
 615 {
 616         u32 *msrpm = svm->msrpm;
 617
 618         svm->vmcb->control.lbr_ctl = 1;
 619         set_msr_interception(msrpm, MSR_IA32_LASTBRANCHFROMIP, 1, 1);
 620         set_msr_interception(msrpm, MSR_IA32_LASTBRANCHTOIP, 1, 1);
 621         set_msr_interception(msrpm, MSR_IA32_LASTINTFROMIP, 1, 1);
 622         set_msr_interception(msrpm, MSR_IA32_LASTINTTOIP, 1, 1);
 623 }
 624
 625 static void svm_disable_lbrv(struct vcpu_svm *svm)
 626 {
 627         u32 *msrpm = svm->msrpm;
 628
 629         svm->vmcb->control.lbr_ctl = 0;
 630         set_msr_interception(msrpm, MSR_IA32_LASTBRANCHFROMIP, 0, 0);
 631         set_msr_interception(msrpm, MSR_IA32_LASTBRANCHTOIP, 0, 0);
 632         set_msr_interception(msrpm, MSR_IA32_LASTINTFROMIP, 0, 0);
 633         set_msr_interception(msrpm, MSR_IA32_LASTINTTOIP, 0, 0);
 634 }
 635
 636 static __init int svm_hardware_setup(void)
 637 {
 638         int cpu;
 639         struct page *iopm_pages;
 640         void *iopm_va;
 641         int r;
 642
 643         iopm_pages = alloc_pages(GFP_KERNEL, IOPM_ALLOC_ORDER);
 644
 645         if (!iopm_pages)
 646                 return -ENOMEM;
 647
 648         iopm_va = page_address(iopm_pages);
 649         memset(iopm_va, 0xff, PAGE_SIZE * (1 << IOPM_ALLOC_ORDER));
 650         iopm_base = page_to_pfn(iopm_pages) << PAGE_SHIFT;
 651
 652         init_msrpm_offsets();
 653
 654         if (boot_cpu_has(X86_FEATURE_NX))
 655                 kvm_enable_efer_bits(EFER_NX);
 656
 657         if (boot_cpu_has(X86_FEATURE_FXSR_OPT))
 658                 kvm_enable_efer_bits(EFER_FFXSR);
 659
 660         if (nested) {
 661                 printk(KERN_INFO "kvm: Nested Virtualization enabled\n");
 662                 kvm_enable_efer_bits(EFER_SVME | EFER_LMSLE);
 663         }
 664
 665         for_each_possible_cpu(cpu) {
 666                 r = svm_cpu_init(cpu);
 667                 if (r)
 668                         goto err;
 669         }
 670
 671         svm_features = cpuid_edx(SVM_CPUID_FUNC);
 672
 673         if (!svm_has(SVM_FEATURE_NPT))
 674                 npt_enabled = false;
 675
 676         if (npt_enabled && !npt) {
 677                 printk(KERN_INFO "kvm: Nested Paging disabled\n");
 678                 npt_enabled = false;
 679         }
 680
 681         if (npt_enabled) {
 682                 printk(KERN_INFO "kvm: Nested Paging enabled\n");
 683                 kvm_enable_tdp();
 684         } else
 685                 kvm_disable_tdp();
 686
 687         return 0;
 688
 689 err:
 690         __free_pages(iopm_pages, IOPM_ALLOC_ORDER);
 691         iopm_base = 0;
 692         return r;
 693 }
 694
 695 static __exit void svm_hardware_unsetup(void)
 696 {
 697         int cpu;
 698
 699         for_each_possible_cpu(cpu)
 700                 svm_cpu_uninit(cpu);
 701
 702         __free_pages(pfn_to_page(iopm_base >> PAGE_SHIFT), IOPM_ALLOC_ORDER);
 703         iopm_base = 0;
 704 }
 705
 706 static void init_seg(struct vmcb_seg *seg)
 707 {
 708         seg->selector = 0;
 709         seg->attrib = SVM_SELECTOR_P_MASK | SVM_SELECTOR_S_MASK |
 710                       SVM_SELECTOR_WRITE_MASK; /* Read/Write Data Segment */
 711         seg->limit = 0xffff;
 712         seg->base = 0;
 713 }
 714
 715 static void init_sys_seg(struct vmcb_seg *seg, uint32_t type)
 716 {
 717         seg->selector = 0;
 718         seg->attrib = SVM_SELECTOR_P_MASK | type;
 719         seg->limit = 0xffff;
 720         seg->base = 0;
 721 }
 722
 723 static void svm_write_tsc_offset(struct kvm_vcpu *vcpu, u64 offset)
 724 {
 725         struct vcpu_svm *svm = to_svm(vcpu);
 726         u64 g_tsc_offset = 0;
 727
 728         if (is_nested(svm)) {
 729                 g_tsc_offset = svm->vmcb->control.tsc_offset -
 730                                svm->nested.hsave->control.tsc_offset;
 731                 svm->nested.hsave->control.tsc_offset = offset;
 732         }
 733
 734         svm->vmcb->control.tsc_offset = offset + g_tsc_offset;
 735 }
 736
 737 static void svm_adjust_tsc_offset(struct kvm_vcpu *vcpu, s64 adjustment)
 738 {
 739         struct vcpu_svm *svm = to_svm(vcpu);
 740
 741         svm->vmcb->control.tsc_offset += adjustment;
 742         if (is_nested(svm))
 743                 svm->nested.hsave->control.tsc_offset += adjustment;
 744 }
 745
 746 static void init_vmcb(struct vcpu_svm *svm)
 747 {
 748         struct vmcb_control_area *control = &svm->vmcb->control;
 749         struct vmcb_save_area *save = &svm->vmcb->save;
 750
 751         svm->vcpu.fpu_active = 1;
 752
 753         control->intercept_cr_read =    INTERCEPT_CR0_MASK |
 754                                         INTERCEPT_CR3_MASK |
 755                                         INTERCEPT_CR4_MASK;
 756
 757         control->intercept_cr_write =   INTERCEPT_CR0_MASK |
 758                                         INTERCEPT_CR3_MASK |
 759                                         INTERCEPT_CR4_MASK |
 760                                         INTERCEPT_CR8_MASK;
 761
 762         control->intercept_dr_read =    INTERCEPT_DR0_MASK |
 763                                         INTERCEPT_DR1_MASK |
 764                                         INTERCEPT_DR2_MASK |
 765                                         INTERCEPT_DR3_MASK |
 766                                         INTERCEPT_DR4_MASK |
 767                                         INTERCEPT_DR5_MASK |
 768                                         INTERCEPT_DR6_MASK |
 769                                         INTERCEPT_DR7_MASK;
 770
 771         control->intercept_dr_write =   INTERCEPT_DR0_MASK |
 772                                         INTERCEPT_DR1_MASK |
 773                                         INTERCEPT_DR2_MASK |
 774                                         INTERCEPT_DR3_MASK |
 775                                         INTERCEPT_DR4_MASK |
 776                                         INTERCEPT_DR5_MASK |
 777                                         INTERCEPT_DR6_MASK |
 778                                         INTERCEPT_DR7_MASK;
 779
 780         control->intercept_exceptions = (1 << PF_VECTOR) |
 781                                         (1 << UD_VECTOR) |
 782                                         (1 << MC_VECTOR);
 783
 784
 785         control->intercept =    (1ULL << INTERCEPT_INTR) |
 786                                 (1ULL << INTERCEPT_NMI) |
 787                                 (1ULL << INTERCEPT_SMI) |
 788                                 (1ULL << INTERCEPT_SELECTIVE_CR0) |
 789                                 (1ULL << INTERCEPT_CPUID) |
 790                                 (1ULL << INTERCEPT_INVD) |
 791                                 (1ULL << INTERCEPT_HLT) |
 792                                 (1ULL << INTERCEPT_INVLPG) |
 793                                 (1ULL << INTERCEPT_INVLPGA) |
 794                                 (1ULL << INTERCEPT_IOIO_PROT) |
 795                                 (1ULL << INTERCEPT_MSR_PROT) |
 796                                 (1ULL << INTERCEPT_TASK_SWITCH) |
 797                                 (1ULL << INTERCEPT_SHUTDOWN) |
 798                                 (1ULL << INTERCEPT_VMRUN) |
 799                                 (1ULL << INTERCEPT_VMMCALL) |
 800                                 (1ULL << INTERCEPT_VMLOAD) |
 801                                 (1ULL << INTERCEPT_VMSAVE) |
 802                                 (1ULL << INTERCEPT_STGI) |
 803                                 (1ULL << INTERCEPT_CLGI) |
 804                                 (1ULL << INTERCEPT_SKINIT) |
 805                                 (1ULL << INTERCEPT_WBINVD) |
 806                                 (1ULL << INTERCEPT_MONITOR) |
 807                                 (1ULL << INTERCEPT_MWAIT);
 808
 809         control->iopm_base_pa = iopm_base;
 810         control->msrpm_base_pa = __pa(svm->msrpm);
 811         control->int_ctl = V_INTR_MASKING_MASK;
 812
 813         init_seg(&save->es);
 814         init_seg(&save->ss);
 815         init_seg(&save->ds);
 816         init_seg(&save->fs);
 817         init_seg(&save->gs);
 818
 819         save->cs.selector = 0xf000;
 820         /* Executable/Readable Code Segment */
 821         save->cs.attrib = SVM_SELECTOR_READ_MASK | SVM_SELECTOR_P_MASK |
 822                 SVM_SELECTOR_S_MASK | SVM_SELECTOR_CODE_MASK;
 823         save->cs.limit = 0xffff;
 824         /*
 825          * cs.base should really be 0xffff0000, but vmx can't handle that, so
 826          * be consistent with it.
 827          *
 828          * Replace when we have real mode working for vmx.
 829          */
 830         save->cs.base = 0xf0000;
 831
 832         save->gdtr.limit = 0xffff;
 833         save->idtr.limit = 0xffff;
 834
 835         init_sys_seg(&save->ldtr, SEG_TYPE_LDT);
 836         init_sys_seg(&save->tr, SEG_TYPE_BUSY_TSS16);
 837
 838         svm_set_efer(&svm->vcpu, 0);
 839         save->dr6 = 0xffff0ff0;
 840         save->dr7 = 0x400;
 841         save->rflags = 2;
 842         save->rip = 0x0000fff0;
 843         svm->vcpu.arch.regs[VCPU_REGS_RIP] = save->rip;
 844
 845         /*
 846          * This is the guest-visible cr0 value.
 847          * svm_set_cr0() sets PG and WP and clears NW and CD on save->cr0.
 848          */
 849         svm->vcpu.arch.cr0 = 0;
 850         (void)kvm_set_cr0(&svm->vcpu, X86_CR0_NW | X86_CR0_CD | X86_CR0_ET);
 851
 852         save->cr4 = X86_CR4_PAE;
 853         /* rdx = ?? */
 854
 855         if (npt_enabled) {
 856                 /* Setup VMCB for Nested Paging */
 857                 control->nested_ctl = 1;
 858                 control->intercept &= ~((1ULL << INTERCEPT_TASK_SWITCH) |
 859                                         (1ULL << INTERCEPT_INVLPG));
 860                 control->intercept_exceptions &= ~(1 << PF_VECTOR);
 861                 control->intercept_cr_read &= ~INTERCEPT_CR3_MASK;
 862                 control->intercept_cr_write &= ~INTERCEPT_CR3_MASK;
 863                 save->g_pat = 0x0007040600070406ULL;
 864                 save->cr3 = 0;
 865                 save->cr4 = 0;
 866         }
 867         force_new_asid(&svm->vcpu);
 868
 869         svm->nested.vmcb = 0;
 870         svm->vcpu.arch.hflags = 0;
 871
 872         if (svm_has(SVM_FEATURE_PAUSE_FILTER)) {
 873                 control->pause_filter_count = 3000;
 874                 control->intercept |= (1ULL << INTERCEPT_PAUSE);
 875         }
 876
 877         enable_gif(svm);
 878 }
 879
 880 static int svm_vcpu_reset(struct kvm_vcpu *vcpu)
 881 {
 882         struct vcpu_svm *svm = to_svm(vcpu);
 883
 884         init_vmcb(svm);
 885
 886         if (!kvm_vcpu_is_bsp(vcpu)) {
 887                 kvm_rip_write(vcpu, 0);
 888                 svm->vmcb->save.cs.base = svm->vcpu.arch.sipi_vector << 12;
 889                 svm->vmcb->save.cs.selector = svm->vcpu.arch.sipi_vector << 8;
 890         }
 891         vcpu->arch.regs_avail = ~0;
 892         vcpu->arch.regs_dirty = ~0;
 893
 894         return 0;
 895 }
 896
 897 static struct kvm_vcpu *svm_create_vcpu(struct kvm *kvm, unsigned int id)
 898 {
 899         struct vcpu_svm *svm;
 900         struct page *page;
 901         struct page *msrpm_pages;
 902         struct page *hsave_page;
 903         struct page *nested_msrpm_pages;
 904         int err;
 905
 906         svm = kmem_cache_zalloc(kvm_vcpu_cache, GFP_KERNEL);
 907         if (!svm) {
 908                 err = -ENOMEM;
 909                 goto out;
 910         }
 911
 912         err = kvm_vcpu_init(&svm->vcpu, kvm, id);
 913         if (err)
 914                 goto free_svm;
 915
 916         err = -ENOMEM;
 917         page = alloc_page(GFP_KERNEL);
 918         if (!page)
 919                 goto uninit;
 920
 921         msrpm_pages = alloc_pages(GFP_KERNEL, MSRPM_ALLOC_ORDER);
 922         if (!msrpm_pages)
 923                 goto free_page1;
 924
 925         nested_msrpm_pages = alloc_pages(GFP_KERNEL, MSRPM_ALLOC_ORDER);
 926         if (!nested_msrpm_pages)
 927                 goto free_page2;
 928
 929         hsave_page = alloc_page(GFP_KERNEL);
 930         if (!hsave_page)
 931                 goto free_page3;
 932
 933         svm->nested.hsave = page_address(hsave_page);
 934
 935         svm->msrpm = page_address(msrpm_pages);
 936         svm_vcpu_init_msrpm(svm->msrpm);
 937
 938         svm->nested.msrpm = page_address(nested_msrpm_pages);
 939         svm_vcpu_init_msrpm(svm->nested.msrpm);
 940
 941         svm->vmcb = page_address(page);
 942         clear_page(svm->vmcb);
 943         svm->vmcb_pa = page_to_pfn(page) << PAGE_SHIFT;
 944         svm->asid_generation = 0;
 945         init_vmcb(svm);
 946         kvm_write_tsc(&svm->vcpu, 0);
 947
 948         err = fx_init(&svm->vcpu);
 949         if (err)
 950                 goto free_page4;
 951
 952         svm->vcpu.arch.apic_base = 0xfee00000 | MSR_IA32_APICBASE_ENABLE;
 953         if (kvm_vcpu_is_bsp(&svm->vcpu))
 954                 svm->vcpu.arch.apic_base |= MSR_IA32_APICBASE_BSP;
 955
 956         return &svm->vcpu;
 957
 958 free_page4:
 959         __free_page(hsave_page);
 960 free_page3:
 961         __free_pages(nested_msrpm_pages, MSRPM_ALLOC_ORDER);
 962 free_page2:
 963         __free_pages(msrpm_pages, MSRPM_ALLOC_ORDER);
 964 free_page1:
 965         __free_page(page);
 966 uninit:
 967         kvm_vcpu_uninit(&svm->vcpu);
 968 free_svm:
 969         kmem_cache_free(kvm_vcpu_cache, svm);
 970 out:
 971         return ERR_PTR(err);
 972 }
 973
 974 static void svm_free_vcpu(struct kvm_vcpu *vcpu)
 975 {
 976         struct vcpu_svm *svm = to_svm(vcpu);
 977
 978         __free_page(pfn_to_page(svm->vmcb_pa >> PAGE_SHIFT));
 979         __free_pages(virt_to_page(svm->msrpm), MSRPM_ALLOC_ORDER);
 980         __free_page(virt_to_page(svm->nested.hsave));
 981         __free_pages(virt_to_page(svm->nested.msrpm), MSRPM_ALLOC_ORDER);
 982         kvm_vcpu_uninit(vcpu);
 983         kmem_cache_free(kvm_vcpu_cache, svm);
 984 }
 985
 986 static void svm_vcpu_load(struct kvm_vcpu *vcpu, int cpu)
 987 {
 988         struct vcpu_svm *svm = to_svm(vcpu);
 989         int i;
 990
 991         if (unlikely(cpu != vcpu->cpu)) {
 992                 svm->asid_generation = 0;
 993         }
 994
 995         for (i = 0; i < NR_HOST_SAVE_USER_MSRS; i++)
 996                 rdmsrl(host_save_user_msrs[i], svm->host_user_msrs[i]);
 997 }
 998
 999 static void svm_vcpu_put(struct kvm_vcpu *vcpu)
1000 {
1001         struct vcpu_svm *svm = to_svm(vcpu);
1002         int i;
1003
1004         ++vcpu->stat.host_state_reload;
1005         for (i = 0; i < NR_HOST_SAVE_USER_MSRS; i++)
1006                 wrmsrl(host_save_user_msrs[i], svm->host_user_msrs[i]);
1007 }
1008
1009 static unsigned long svm_get_rflags(struct kvm_vcpu *vcpu)
1010 {
1011         return to_svm(vcpu)->vmcb->save.rflags;
1012 }
1013
1014 static void svm_set_rflags(struct kvm_vcpu *vcpu, unsigned long rflags)
1015 {
1016         to_svm(vcpu)->vmcb->save.rflags = rflags;
1017 }
1018
1019 static void svm_cache_reg(struct kvm_vcpu *vcpu, enum kvm_reg reg)
1020 {
1021         switch (reg) {
1022         case VCPU_EXREG_PDPTR:
1023                 BUG_ON(!npt_enabled);
1024                 load_pdptrs(vcpu, vcpu->arch.walk_mmu, vcpu->arch.cr3);
1025                 break;
1026         default:
1027                 BUG();
1028         }
1029 }
1030
1031 static void svm_set_vintr(struct vcpu_svm *svm)
1032 {
1033         svm->vmcb->control.intercept |= 1ULL << INTERCEPT_VINTR;
1034 }
1035
1036 static void svm_clear_vintr(struct vcpu_svm *svm)
1037 {
1038         svm->vmcb->control.intercept &= ~(1ULL << INTERCEPT_VINTR);
1039 }
1040
1041 static struct vmcb_seg *svm_seg(struct kvm_vcpu *vcpu, int seg)
1042 {
1043         struct vmcb_save_area *save = &to_svm(vcpu)->vmcb->save;
1044
1045         switch (seg) {
1046         case VCPU_SREG_CS: return &save->cs;
1047         case VCPU_SREG_DS: return &save->ds;
1048         case VCPU_SREG_ES: return &save->es;
1049         case VCPU_SREG_FS: return &save->fs;
1050         case VCPU_SREG_GS: return &save->gs;
1051         case VCPU_SREG_SS: return &save->ss;
1052         case VCPU_SREG_TR: return &save->tr;
1053         case VCPU_SREG_LDTR: return &save->ldtr;
1054         }
1055         BUG();
1056         return NULL;
1057 }
1058
1059 static u64 svm_get_segment_base(struct kvm_vcpu *vcpu, int seg)
1060 {
1061         struct vmcb_seg *s = svm_seg(vcpu, seg);
1062
1063         return s->base;
1064 }
1065
1066 static void svm_get_segment(struct kvm_vcpu *vcpu,
1067                             struct kvm_segment *var, int seg)
1068 {
1069         struct vmcb_seg *s = svm_seg(vcpu, seg);
1070
1071         var->base = s->base;
1072         var->limit = s->limit;
1073         var->selector = s->selector;
1074         var->type = s->attrib & SVM_SELECTOR_TYPE_MASK;
1075         var->s = (s->attrib >> SVM_SELECTOR_S_SHIFT) & 1;
1076         var->dpl = (s->attrib >> SVM_SELECTOR_DPL_SHIFT) & 3;
1077         var->present = (s->attrib >> SVM_SELECTOR_P_SHIFT) & 1;
1078         var->avl = (s->attrib >> SVM_SELECTOR_AVL_SHIFT) & 1;
1079         var->l = (s->attrib >> SVM_SELECTOR_L_SHIFT) & 1;
1080         var->db = (s->attrib >> SVM_SELECTOR_DB_SHIFT) & 1;
1081         var->g = (s->attrib >> SVM_SELECTOR_G_SHIFT) & 1;
1082
1083         /*
1084          * AMD's VMCB does not have an explicit unusable field, so emulate it
1085          * for cross vendor migration purposes by "not present"
1086          */
1087         var->unusable = !var->present || (var->type == 0);
1088
1089         switch (seg) {
1090         case VCPU_SREG_CS:
1091                 /*
1092                  * SVM always stores 0 for the 'G' bit in the CS selector in
1093                  * the VMCB on a VMEXIT. This hurts cross-vendor migration:
1094                  * Intel's VMENTRY has a check on the 'G' bit.
1095                  */
1096                 var->g = s->limit > 0xfffff;
1097                 break;
1098         case VCPU_SREG_TR:
1099                 /*
1100                  * Work around a bug where the busy flag in the tr selector
1101                  * isn't exposed
1102                  */
1103                 var->type |= 0x2;
1104                 break;
1105         case VCPU_SREG_DS:
1106         case VCPU_SREG_ES:
1107         case VCPU_SREG_FS:
1108         case VCPU_SREG_GS:
1109                 /*
1110                  * The accessed bit must always be set in the segment
1111                  * descriptor cache, although it can be cleared in the
1112                  * descriptor, the cached bit always remains at 1. Since
1113                  * Intel has a check on this, set it here to support
1114                  * cross-vendor migration.
1115                  */
1116                 if (!var->unusable)
1117                         var->type |= 0x1;
1118                 break;
1119         case VCPU_SREG_SS:
1120                 /*
1121                  * On AMD CPUs sometimes the DB bit in the segment
1122                  * descriptor is left as 1, although the whole segment has
1123                  * been made unusable. Clear it here to pass an Intel VMX
1124                  * entry check when cross vendor migrating.
1125                  */
1126                 if (var->unusable)
1127                         var->db = 0;
1128                 break;
1129         }
1130 }
1131
1132 static int svm_get_cpl(struct kvm_vcpu *vcpu)
1133 {
1134         struct vmcb_save_area *save = &to_svm(vcpu)->vmcb->save;
1135
1136         return save->cpl;
1137 }
1138
1139 static void svm_get_idt(struct kvm_vcpu *vcpu, struct desc_ptr *dt)
1140 {
1141         struct vcpu_svm *svm = to_svm(vcpu);
1142
1143         dt->size = svm->vmcb->save.idtr.limit;
1144         dt->address = svm->vmcb->save.idtr.base;
1145 }
1146
1147 static void svm_set_idt(struct kvm_vcpu *vcpu, struct desc_ptr *dt)
1148 {
1149         struct vcpu_svm *svm = to_svm(vcpu);
1150
1151         svm->vmcb->save.idtr.limit = dt->size;
1152         svm->vmcb->save.idtr.base = dt->address ;
1153 }
1154
1155 static void svm_get_gdt(struct kvm_vcpu *vcpu, struct desc_ptr *dt)
1156 {
1157         struct vcpu_svm *svm = to_svm(vcpu);
1158
1159         dt->size = svm->vmcb->save.gdtr.limit;
1160         dt->address = svm->vmcb->save.gdtr.base;
1161 }
1162
1163 static void svm_set_gdt(struct kvm_vcpu *vcpu, struct desc_ptr *dt)
1164 {
1165         struct vcpu_svm *svm = to_svm(vcpu);
1166
1167         svm->vmcb->save.gdtr.limit = dt->size;
1168         svm->vmcb->save.gdtr.base = dt->address ;
1169 }
1170
1171 static void svm_decache_cr0_guest_bits(struct kvm_vcpu *vcpu)
1172 {
1173 }
1174
1175 static void svm_decache_cr4_guest_bits(struct kvm_vcpu *vcpu)
1176 {
1177 }
1178
1179 static void update_cr0_intercept(struct vcpu_svm *svm)
1180 {
1181         struct vmcb *vmcb = svm->vmcb;
1182         ulong gcr0 = svm->vcpu.arch.cr0;
1183         u64 *hcr0 = &svm->vmcb->save.cr0;
1184
1185         if (!svm->vcpu.fpu_active)
1186                 *hcr0 |= SVM_CR0_SELECTIVE_MASK;
1187         else
1188                 *hcr0 = (*hcr0 & ~SVM_CR0_SELECTIVE_MASK)
1189                         | (gcr0 & SVM_CR0_SELECTIVE_MASK);
1190
1191
1192         if (gcr0 == *hcr0 && svm->vcpu.fpu_active) {
1193                 vmcb->control.intercept_cr_read &= ~INTERCEPT_CR0_MASK;
1194                 vmcb->control.intercept_cr_write &= ~INTERCEPT_CR0_MASK;
1195                 if (is_nested(svm)) {
1196                         struct vmcb *hsave = svm->nested.hsave;
1197
1198                         hsave->control.intercept_cr_read  &= ~INTERCEPT_CR0_MASK;
1199                         hsave->control.intercept_cr_write &= ~INTERCEPT_CR0_MASK;
1200                         vmcb->control.intercept_cr_read  |= svm->nested.intercept_cr_read;
1201                         vmcb->control.intercept_cr_write |= svm->nested.intercept_cr_write;
1202                 }
1203         } else {
1204                 svm->vmcb->control.intercept_cr_read |= INTERCEPT_CR0_MASK;
1205                 svm->vmcb->control.intercept_cr_write |= INTERCEPT_CR0_MASK;
1206                 if (is_nested(svm)) {
1207                         struct vmcb *hsave = svm->nested.hsave;
1208
1209                         hsave->control.intercept_cr_read |= INTERCEPT_CR0_MASK;
1210                         hsave->control.intercept_cr_write |= INTERCEPT_CR0_MASK;
1211                 }
1212         }
1213 }
1214
1215 static void svm_set_cr0(struct kvm_vcpu *vcpu, unsigned long cr0)
1216 {
1217         struct vcpu_svm *svm = to_svm(vcpu);
1218
1219         if (is_nested(svm)) {
1220                 /*
1221                  * We are here because we run in nested mode, the host kvm
1222                  * intercepts cr0 writes but the l1 hypervisor does not.
1223                  * But the L1 hypervisor may intercept selective cr0 writes.
1224                  * This needs to be checked here.
1225                  */
1226                 unsigned long old, new;
1227
1228                 /* Remove bits that would trigger a real cr0 write intercept */
1229                 old = vcpu->arch.cr0 & SVM_CR0_SELECTIVE_MASK;
1230                 new = cr0 & SVM_CR0_SELECTIVE_MASK;
1231
1232                 if (old == new) {
1233                         /* cr0 write with ts and mp unchanged */
1234                         svm->vmcb->control.exit_code = SVM_EXIT_CR0_SEL_WRITE;
1235                         if (nested_svm_exit_handled(svm) == NESTED_EXIT_DONE) {
1236                                 svm->nested.vmexit_rip = kvm_rip_read(vcpu);
1237                                 svm->nested.vmexit_rsp = kvm_register_read(vcpu, VCPU_REGS_RSP);
1238                                 svm->nested.vmexit_rax = kvm_register_read(vcpu, VCPU_REGS_RAX);
1239                                 return;
1240                         }
1241                 }
1242         }
1243
1244 #ifdef CONFIG_X86_64
1245         if (vcpu->arch.efer & EFER_LME) {
1246                 if (!is_paging(vcpu) && (cr0 & X86_CR0_PG)) {
1247                         vcpu->arch.efer |= EFER_LMA;
1248                         svm->vmcb->save.efer |= EFER_LMA | EFER_LME;
1249                 }
1250
1251                 if (is_paging(vcpu) && !(cr0 & X86_CR0_PG)) {
1252                         vcpu->arch.efer &= ~EFER_LMA;
1253                         svm->vmcb->save.efer &= ~(EFER_LMA | EFER_LME);
1254                 }
1255         }
1256 #endif
1257         vcpu->arch.cr0 = cr0;
1258
1259         if (!npt_enabled)
1260                 cr0 |= X86_CR0_PG | X86_CR0_WP;
1261
1262         if (!vcpu->fpu_active)
1263                 cr0 |= X86_CR0_TS;
1264         /*
1265          * re-enable caching here because the QEMU bios
1266          * does not do it - this results in some delay at
1267          * reboot
1268          */
1269         cr0 &= ~(X86_CR0_CD | X86_CR0_NW);
1270         svm->vmcb->save.cr0 = cr0;
1271         update_cr0_intercept(svm);
1272 }
1273
1274 static void svm_set_cr4(struct kvm_vcpu *vcpu, unsigned long cr4)
1275 {
1276         unsigned long host_cr4_mce = read_cr4() & X86_CR4_MCE;
1277         unsigned long old_cr4 = to_svm(vcpu)->vmcb->save.cr4;
1278
1279         if (npt_enabled && ((old_cr4 ^ cr4) & X86_CR4_PGE))
1280                 force_new_asid(vcpu);
1281
1282         vcpu->arch.cr4 = cr4;
1283         if (!npt_enabled)
1284                 cr4 |= X86_CR4_PAE;
1285         cr4 |= host_cr4_mce;
1286         to_svm(vcpu)->vmcb->save.cr4 = cr4;
1287 }
1288
1289 static void svm_set_segment(struct kvm_vcpu *vcpu,
1290                             struct kvm_segment *var, int seg)
1291 {
1292         struct vcpu_svm *svm = to_svm(vcpu);
1293         struct vmcb_seg *s = svm_seg(vcpu, seg);
1294
1295         s->base = var->base;
1296         s->limit = var->limit;
1297         s->selector = var->selector;
1298         if (var->unusable)
1299                 s->attrib = 0;
1300         else {
1301                 s->attrib = (var->type & SVM_SELECTOR_TYPE_MASK);
1302                 s->attrib |= (var->s & 1) << SVM_SELECTOR_S_SHIFT;
1303                 s->attrib |= (var->dpl & 3) << SVM_SELECTOR_DPL_SHIFT;
1304                 s->attrib |= (var->present & 1) << SVM_SELECTOR_P_SHIFT;
1305                 s->attrib |= (var->avl & 1) << SVM_SELECTOR_AVL_SHIFT;
1306                 s->attrib |= (var->l & 1) << SVM_SELECTOR_L_SHIFT;
1307                 s->attrib |= (var->db & 1) << SVM_SELECTOR_DB_SHIFT;
1308                 s->attrib |= (var->g & 1) << SVM_SELECTOR_G_SHIFT;
1309         }
1310         if (seg == VCPU_SREG_CS)
1311                 svm->vmcb->save.cpl
1312                         = (svm->vmcb->save.cs.attrib
1313                            >> SVM_SELECTOR_DPL_SHIFT) & 3;
1314
1315 }
1316
1317 static void update_db_intercept(struct kvm_vcpu *vcpu)
1318 {
1319         struct vcpu_svm *svm = to_svm(vcpu);
1320
1321         svm->vmcb->control.intercept_exceptions &=
1322                 ~((1 << DB_VECTOR) | (1 << BP_VECTOR));
1323
1324         if (svm->nmi_singlestep)
1325                 svm->vmcb->control.intercept_exceptions |= (1 << DB_VECTOR);
1326
1327         if (vcpu->guest_debug & KVM_GUESTDBG_ENABLE) {
1328                 if (vcpu->guest_debug &
1329                     (KVM_GUESTDBG_SINGLESTEP | KVM_GUESTDBG_USE_HW_BP))
1330                         svm->vmcb->control.intercept_exceptions |=
1331                                 1 << DB_VECTOR;
1332                 if (vcpu->guest_debug & KVM_GUESTDBG_USE_SW_BP)
1333                         svm->vmcb->control.intercept_exceptions |=
1334                                 1 << BP_VECTOR;
1335         } else
1336                 vcpu->guest_debug = 0;
1337 }
1338
1339 static void svm_guest_debug(struct kvm_vcpu *vcpu, struct kvm_guest_debug *dbg)
1340 {
1341         struct vcpu_svm *svm = to_svm(vcpu);
1342
1343         if (vcpu->guest_debug & KVM_GUESTDBG_USE_HW_BP)
1344                 svm->vmcb->save.dr7 = dbg->arch.debugreg[7];
1345         else
1346                 svm->vmcb->save.dr7 = vcpu->arch.dr7;
1347
1348         update_db_intercept(vcpu);
1349 }
1350
1351 static void load_host_msrs(struct kvm_vcpu *vcpu)
1352 {
1353 #ifdef CONFIG_X86_64
1354         wrmsrl(MSR_GS_BASE, to_svm(vcpu)->host_gs_base);
1355 #endif
1356 }
1357
1358 static void save_host_msrs(struct kvm_vcpu *vcpu)
1359 {
1360 #ifdef CONFIG_X86_64
1361         rdmsrl(MSR_GS_BASE, to_svm(vcpu)->host_gs_base);
1362 #endif
1363 }
1364
1365 static void new_asid(struct vcpu_svm *svm, struct svm_cpu_data *sd)
1366 {
1367         if (sd->next_asid > sd->max_asid) {
1368                 ++sd->asid_generation;
1369                 sd->next_asid = 1;
1370                 svm->vmcb->control.tlb_ctl = TLB_CONTROL_FLUSH_ALL_ASID;
1371         }
1372
1373         svm->asid_generation = sd->asid_generation;
1374         svm->vmcb->control.asid = sd->next_asid++;
1375 }
1376
1377 static void svm_set_dr7(struct kvm_vcpu *vcpu, unsigned long value)
1378 {
1379         struct vcpu_svm *svm = to_svm(vcpu);
1380
1381         svm->vmcb->save.dr7 = value;
1382 }
1383
1384 static int pf_interception(struct vcpu_svm *svm)
1385 {
1386         u64 fault_address;
1387         u32 error_code;
1388
1389         fault_address  = svm->vmcb->control.exit_info_2;
1390         error_code = svm->vmcb->control.exit_info_1;
1391
1392         trace_kvm_page_fault(fault_address, error_code);
1393         if (!npt_enabled && kvm_event_needs_reinjection(&svm->vcpu))
1394                 kvm_mmu_unprotect_page_virt(&svm->vcpu, fault_address);
1395         return kvm_mmu_page_fault(&svm->vcpu, fault_address, error_code);
1396 }
1397
1398 static int db_interception(struct vcpu_svm *svm)
1399 {
1400         struct kvm_run *kvm_run = svm->vcpu.run;
1401
1402         if (!(svm->vcpu.guest_debug &
1403               (KVM_GUESTDBG_SINGLESTEP | KVM_GUESTDBG_USE_HW_BP)) &&
1404                 !svm->nmi_singlestep) {
1405                 kvm_queue_exception(&svm->vcpu, DB_VECTOR);
1406                 return 1;
1407         }
1408
1409         if (svm->nmi_singlestep) {
1410                 svm->nmi_singlestep = false;
1411                 if (!(svm->vcpu.guest_debug & KVM_GUESTDBG_SINGLESTEP))
1412                         svm->vmcb->save.rflags &=
1413                                 ~(X86_EFLAGS_TF | X86_EFLAGS_RF);
1414                 update_db_intercept(&svm->vcpu);
1415         }
1416
1417         if (svm->vcpu.guest_debug &
1418             (KVM_GUESTDBG_SINGLESTEP | KVM_GUESTDBG_USE_HW_BP)) {
1419                 kvm_run->exit_reason = KVM_EXIT_DEBUG;
1420                 kvm_run->debug.arch.pc =
1421                         svm->vmcb->save.cs.base + svm->vmcb->save.rip;
1422                 kvm_run->debug.arch.exception = DB_VECTOR;
1423                 return 0;
1424         }
1425
1426         return 1;
1427 }
1428
1429 static int bp_interception(struct vcpu_svm *svm)
1430 {
1431         struct kvm_run *kvm_run = svm->vcpu.run;
1432
1433         kvm_run->exit_reason = KVM_EXIT_DEBUG;
1434         kvm_run->debug.arch.pc = svm->vmcb->save.cs.base + svm->vmcb->save.rip;
1435         kvm_run->debug.arch.exception = BP_VECTOR;
1436         return 0;
1437 }
1438
1439 static int ud_interception(struct vcpu_svm *svm)
1440 {
1441         int er;
1442
1443         er = emulate_instruction(&svm->vcpu, 0, 0, EMULTYPE_TRAP_UD);
1444         if (er != EMULATE_DONE)
1445                 kvm_queue_exception(&svm->vcpu, UD_VECTOR);
1446         return 1;
1447 }
1448
1449 static void svm_fpu_activate(struct kvm_vcpu *vcpu)
1450 {
1451         struct vcpu_svm *svm = to_svm(vcpu);
1452         u32 excp;
1453
1454         if (is_nested(svm)) {
1455                 u32 h_excp, n_excp;
1456
1457                 h_excp  = svm->nested.hsave->control.intercept_exceptions;
1458                 n_excp  = svm->nested.intercept_exceptions;
1459                 h_excp &= ~(1 << NM_VECTOR);
1460                 excp    = h_excp | n_excp;
1461         } else {
1462                 excp  = svm->vmcb->control.intercept_exceptions;
1463                 excp &= ~(1 << NM_VECTOR);
1464         }
1465
1466         svm->vmcb->control.intercept_exceptions = excp;
1467
1468         svm->vcpu.fpu_active = 1;
1469         update_cr0_intercept(svm);
1470 }
1471
1472 static int nm_interception(struct vcpu_svm *svm)
1473 {
1474         svm_fpu_activate(&svm->vcpu);
1475         return 1;
1476 }
1477
1478 static bool is_erratum_383(void)
1479 {
1480         int err, i;
1481         u64 value;
1482
1483         if (!erratum_383_found)
1484                 return false;
1485
1486         value = native_read_msr_safe(MSR_IA32_MC0_STATUS, &err);
1487         if (err)
1488                 return false;
1489
1490         /* Bit 62 may or may not be set for this mce */
1491         value &= ~(1ULL << 62);
1492
1493         if (value != 0xb600000000010015ULL)
1494                 return false;
1495
1496         /* Clear MCi_STATUS registers */
1497         for (i = 0; i < 6; ++i)
1498                 native_write_msr_safe(MSR_IA32_MCx_STATUS(i), 0, 0);
1499
1500         value = native_read_msr_safe(MSR_IA32_MCG_STATUS, &err);
1501         if (!err) {
1502                 u32 low, high;
1503
1504                 value &= ~(1ULL << 2);
1505                 low    = lower_32_bits(value);
1506                 high   = upper_32_bits(value);
1507
1508                 native_write_msr_safe(MSR_IA32_MCG_STATUS, low, high);
1509         }
1510
1511         /* Flush tlb to evict multi-match entries */
1512         __flush_tlb_all();
1513
1514         return true;
1515 }
1516
1517 static void svm_handle_mce(struct vcpu_svm *svm)
1518 {
1519         if (is_erratum_383()) {
1520                 /*
1521                  * Erratum 383 triggered. Guest state is corrupt so kill the
1522                  * guest.
1523                  */
1524                 pr_err("KVM: Guest triggered AMD Erratum 383\n");
1525
1526                 kvm_make_request(KVM_REQ_TRIPLE_FAULT, &svm->vcpu);
1527
1528                 return;
1529         }
1530
1531         /*
1532          * On an #MC intercept the MCE handler is not called automatically in
1533          * the host. So do it by hand here.
1534          */
1535         asm volatile (
1536                 "int $0x12\n");
1537         /* not sure if we ever come back to this point */
1538
1539         return;
1540 }
1541
1542 static int mc_interception(struct vcpu_svm *svm)
1543 {
1544         return 1;
1545 }
1546
1547 static int shutdown_interception(struct vcpu_svm *svm)
1548 {
1549         struct kvm_run *kvm_run = svm->vcpu.run;
1550
1551         /*
1552          * VMCB is undefined after a SHUTDOWN intercept
1553          * so reinitialize it.
1554          */
1555         clear_page(svm->vmcb);
1556         init_vmcb(svm);
1557
1558         kvm_run->exit_reason = KVM_EXIT_SHUTDOWN;
1559         return 0;
1560 }
1561
1562 static int io_interception(struct vcpu_svm *svm)
1563 {
1564         struct kvm_vcpu *vcpu = &svm->vcpu;
1565         u32 io_info = svm->vmcb->control.exit_info_1; /* address size bug? */
1566         int size, in, string;
1567         unsigned port;
1568
1569         ++svm->vcpu.stat.io_exits;
1570         string = (io_info & SVM_IOIO_STR_MASK) != 0;
1571         in = (io_info & SVM_IOIO_TYPE_MASK) != 0;
1572         if (string || in)
1573                 return emulate_instruction(vcpu, 0, 0, 0) == EMULATE_DONE;
1574
1575         port = io_info >> 16;
1576         size = (io_info & SVM_IOIO_SIZE_MASK) >> SVM_IOIO_SIZE_SHIFT;
1577         svm->next_rip = svm->vmcb->control.exit_info_2;
1578         skip_emulated_instruction(&svm->vcpu);
1579
1580         return kvm_fast_pio_out(vcpu, size, port);
1581 }
1582
1583 static int nmi_interception(struct vcpu_svm *svm)
1584 {
1585         return 1;
1586 }
1587
1588 static int intr_interception(struct vcpu_svm *svm)
1589 {
1590         ++svm->vcpu.stat.irq_exits;
1591         return 1;
1592 }
1593
1594 static int nop_on_interception(struct vcpu_svm *svm)
1595 {
1596         return 1;
1597 }
1598
1599 static int halt_interception(struct vcpu_svm *svm)
1600 {
1601         svm->next_rip = kvm_rip_read(&svm->vcpu) + 1;
1602         skip_emulated_instruction(&svm->vcpu);
1603         return kvm_emulate_halt(&svm->vcpu);
1604 }
1605
1606 static int vmmcall_interception(struct vcpu_svm *svm)
1607 {
1608         svm->next_rip = kvm_rip_read(&svm->vcpu) + 3;
1609         skip_emulated_instruction(&svm->vcpu);
1610         kvm_emulate_hypercall(&svm->vcpu);
1611         return 1;
1612 }
1613
1614 static unsigned long nested_svm_get_tdp_cr3(struct kvm_vcpu *vcpu)
1615 {
1616         struct vcpu_svm *svm = to_svm(vcpu);
1617
1618         return svm->nested.nested_cr3;
1619 }
1620
1621 static void nested_svm_set_tdp_cr3(struct kvm_vcpu *vcpu,
1622                                    unsigned long root)
1623 {
1624         struct vcpu_svm *svm = to_svm(vcpu);
1625
1626         svm->vmcb->control.nested_cr3 = root;
1627         force_new_asid(vcpu);
1628 }
1629
1630 static void nested_svm_inject_npf_exit(struct kvm_vcpu *vcpu)
1631 {
1632         struct vcpu_svm *svm = to_svm(vcpu);
1633
1634         svm->vmcb->control.exit_code = SVM_EXIT_NPF;
1635         svm->vmcb->control.exit_code_hi = 0;
1636         svm->vmcb->control.exit_info_1 = vcpu->arch.fault.error_code;
1637         svm->vmcb->control.exit_info_2 = vcpu->arch.fault.address;
1638
1639         nested_svm_vmexit(svm);
1640 }
1641
1642 static int nested_svm_init_mmu_context(struct kvm_vcpu *vcpu)
1643 {
1644         int r;
1645
1646         r = kvm_init_shadow_mmu(vcpu, &vcpu->arch.mmu);
1647
1648         vcpu->arch.mmu.set_cr3           = nested_svm_set_tdp_cr3;
1649         vcpu->arch.mmu.get_cr3           = nested_svm_get_tdp_cr3;
1650         vcpu->arch.mmu.inject_page_fault = nested_svm_inject_npf_exit;
1651         vcpu->arch.mmu.shadow_root_level = get_npt_level();
1652         vcpu->arch.walk_mmu              = &vcpu->arch.nested_mmu;
1653
1654         return r;
1655 }
1656
1657 static void nested_svm_uninit_mmu_context(struct kvm_vcpu *vcpu)
1658 {
1659         vcpu->arch.walk_mmu = &vcpu->arch.mmu;
1660 }
1661
1662 static int nested_svm_check_permissions(struct vcpu_svm *svm)
1663 {
1664         if (!(svm->vcpu.arch.efer & EFER_SVME)
1665             || !is_paging(&svm->vcpu)) {
1666                 kvm_queue_exception(&svm->vcpu, UD_VECTOR);
1667                 return 1;
1668         }
1669
1670         if (svm->vmcb->save.cpl) {
1671                 kvm_inject_gp(&svm->vcpu, 0);
1672                 return 1;
1673         }
1674
1675        return 0;
1676 }
1677
1678 static int nested_svm_check_exception(struct vcpu_svm *svm, unsigned nr,
1679                                       bool has_error_code, u32 error_code)
1680 {
1681         int vmexit;
1682
1683         if (!is_nested(svm))
1684                 return 0;
1685
1686         svm->vmcb->control.exit_code = SVM_EXIT_EXCP_BASE + nr;
1687         svm->vmcb->control.exit_code_hi = 0;
1688         svm->vmcb->control.exit_info_1 = error_code;
1689         svm->vmcb->control.exit_info_2 = svm->vcpu.arch.cr2;
1690
1691         vmexit = nested_svm_intercept(svm);
1692         if (vmexit == NESTED_EXIT_DONE)
1693                 svm->nested.exit_required = true;
1694
1695         return vmexit;
1696 }
1697
1698 /* This function returns true if it is save to enable the irq window */
1699 static inline bool nested_svm_intr(struct vcpu_svm *svm)
1700 {
1701         if (!is_nested(svm))
1702                 return true;
1703
1704         if (!(svm->vcpu.arch.hflags & HF_VINTR_MASK))
1705                 return true;
1706
1707         if (!(svm->vcpu.arch.hflags & HF_HIF_MASK))
1708                 return false;
1709
1710         svm->vmcb->control.exit_code   = SVM_EXIT_INTR;
1711         svm->vmcb->control.exit_info_1 = 0;
1712         svm->vmcb->control.exit_info_2 = 0;
1713
1714         if (svm->nested.intercept & 1ULL) {
1715                 /*
1716                  * The #vmexit can't be emulated here directly because this
1717                  * code path runs with irqs and preemtion disabled. A
1718                  * #vmexit emulation might sleep. Only signal request for
1719                  * the #vmexit here.
1720                  */
1721                 svm->nested.exit_required = true;
1722                 trace_kvm_nested_intr_vmexit(svm->vmcb->save.rip);
1723                 return false;
1724         }
1725
1726         return true;
1727 }
1728
1729 /* This function returns true if it is save to enable the nmi window */
1730 static inline bool nested_svm_nmi(struct vcpu_svm *svm)
1731 {
1732         if (!is_nested(svm))
1733                 return true;
1734
1735         if (!(svm->nested.intercept & (1ULL << INTERCEPT_NMI)))
1736                 return true;
1737
1738         svm->vmcb->control.exit_code = SVM_EXIT_NMI;
1739         svm->nested.exit_required = true;
1740
1741         return false;
1742 }
1743
1744 static void *nested_svm_map(struct vcpu_svm *svm, u64 gpa, struct page **_page)
1745 {
1746         struct page *page;
1747
1748         might_sleep();
1749
1750         page = gfn_to_page(svm->vcpu.kvm, gpa >> PAGE_SHIFT);
1751         if (is_error_page(page))
1752                 goto error;
1753
1754         *_page = page;
1755
1756         return kmap(page);
1757
1758 error:
1759         kvm_release_page_clean(page);
1760         kvm_inject_gp(&svm->vcpu, 0);
1761
1762         return NULL;
1763 }
1764
1765 static void nested_svm_unmap(struct page *page)
1766 {
1767         kunmap(page);
1768         kvm_release_page_dirty(page);
1769 }
1770
1771 static int nested_svm_intercept_ioio(struct vcpu_svm *svm)
1772 {
1773         unsigned port;
1774         u8 val, bit;
1775         u64 gpa;
1776
1777         if (!(svm->nested.intercept & (1ULL << INTERCEPT_IOIO_PROT)))
1778                 return NESTED_EXIT_HOST;
1779
1780         port = svm->vmcb->control.exit_info_1 >> 16;
1781         gpa  = svm->nested.vmcb_iopm + (port / 8);
1782         bit  = port % 8;
1783         val  = 0;
1784
1785         if (kvm_read_guest(svm->vcpu.kvm, gpa, &val, 1))
1786                 val &= (1 << bit);
1787
1788         return val ? NESTED_EXIT_DONE : NESTED_EXIT_HOST;
1789 }
1790
1791 static int nested_svm_exit_handled_msr(struct vcpu_svm *svm)
1792 {
1793         u32 offset, msr, value;
1794         int write, mask;
1795
1796         if (!(svm->nested.intercept & (1ULL << INTERCEPT_MSR_PROT)))
1797                 return NESTED_EXIT_HOST;
1798
1799         msr    = svm->vcpu.arch.regs[VCPU_REGS_RCX];
1800         offset = svm_msrpm_offset(msr);
1801         write  = svm->vmcb->control.exit_info_1 & 1;
1802         mask   = 1 << ((2 * (msr & 0xf)) + write);
1803
1804         if (offset == MSR_INVALID)
1805                 return NESTED_EXIT_DONE;
1806
1807         /* Offset is in 32 bit units but need in 8 bit units */
1808         offset *= 4;
1809
1810         if (kvm_read_guest(svm->vcpu.kvm, svm->nested.vmcb_msrpm + offset, &value, 4))
1811                 return NESTED_EXIT_DONE;
1812
1813         return (value & mask) ? NESTED_EXIT_DONE : NESTED_EXIT_HOST;
1814 }
1815
1816 static int nested_svm_exit_special(struct vcpu_svm *svm)
1817 {
1818         u32 exit_code = svm->vmcb->control.exit_code;
1819
1820         switch (exit_code) {
1821         case SVM_EXIT_INTR:
1822         case SVM_EXIT_NMI:
1823         case SVM_EXIT_EXCP_BASE + MC_VECTOR:
1824                 return NESTED_EXIT_HOST;
1825         case SVM_EXIT_NPF:
1826                 /* For now we are always handling NPFs when using them */
1827                 if (npt_enabled)
1828                         return NESTED_EXIT_HOST;
1829                 break;
1830         case SVM_EXIT_EXCP_BASE + PF_VECTOR:
1831                 /* When we're shadowing, trap PFs */
1832                 if (!npt_enabled)
1833                         return NESTED_EXIT_HOST;
1834                 break;
1835         case SVM_EXIT_EXCP_BASE + NM_VECTOR:
1836                 nm_interception(svm);
1837                 break;
1838         default:
1839                 break;
1840         }
1841
1842         return NESTED_EXIT_CONTINUE;
1843 }
1844
1845 /*
1846  * If this function returns true, this #vmexit was already handled
1847  */
1848 static int nested_svm_intercept(struct vcpu_svm *svm)
1849 {
1850         u32 exit_code = svm->vmcb->control.exit_code;
1851         int vmexit = NESTED_EXIT_HOST;
1852
1853         switch (exit_code) {
1854         case SVM_EXIT_MSR:
1855                 vmexit = nested_svm_exit_handled_msr(svm);
1856                 break;
1857         case SVM_EXIT_IOIO:
1858                 vmexit = nested_svm_intercept_ioio(svm);
1859                 break;
1860         case SVM_EXIT_READ_CR0 ... SVM_EXIT_READ_CR8: {
1861                 u32 cr_bits = 1 << (exit_code - SVM_EXIT_READ_CR0);
1862                 if (svm->nested.intercept_cr_read & cr_bits)
1863                         vmexit = NESTED_EXIT_DONE;
1864                 break;
1865         }
1866         case SVM_EXIT_WRITE_CR0 ... SVM_EXIT_WRITE_CR8: {
1867                 u32 cr_bits = 1 << (exit_code - SVM_EXIT_WRITE_CR0);
1868                 if (svm->nested.intercept_cr_write & cr_bits)
1869                         vmexit = NESTED_EXIT_DONE;
1870                 break;
1871         }
1872         case SVM_EXIT_READ_DR0 ... SVM_EXIT_READ_DR7: {
1873                 u32 dr_bits = 1 << (exit_code - SVM_EXIT_READ_DR0);
1874                 if (svm->nested.intercept_dr_read & dr_bits)
1875                         vmexit = NESTED_EXIT_DONE;
1876                 break;
1877         }
1878         case SVM_EXIT_WRITE_DR0 ... SVM_EXIT_WRITE_DR7: {
1879                 u32 dr_bits = 1 << (exit_code - SVM_EXIT_WRITE_DR0);
1880                 if (svm->nested.intercept_dr_write & dr_bits)
1881                         vmexit = NESTED_EXIT_DONE;
1882                 break;
1883         }
1884         case SVM_EXIT_EXCP_BASE ... SVM_EXIT_EXCP_BASE + 0x1f: {
1885                 u32 excp_bits = 1 << (exit_code - SVM_EXIT_EXCP_BASE);
1886                 if (svm->nested.intercept_exceptions & excp_bits)
1887                         vmexit = NESTED_EXIT_DONE;
1888                 break;
1889         }
1890         case SVM_EXIT_ERR: {
1891                 vmexit = NESTED_EXIT_DONE;
1892                 break;
1893         }
1894         default: {
1895                 u64 exit_bits = 1ULL << (exit_code - SVM_EXIT_INTR);
1896                 if (svm->nested.intercept & exit_bits)
1897                         vmexit = NESTED_EXIT_DONE;
1898         }
1899         }
1900
1901         return vmexit;
1902 }
1903
1904 static int nested_svm_exit_handled(struct vcpu_svm *svm)
1905 {
1906         int vmexit;
1907
1908         vmexit = nested_svm_intercept(svm);
1909
1910         if (vmexit == NESTED_EXIT_DONE)
1911                 nested_svm_vmexit(svm);
1912
1913         return vmexit;
1914 }
1915
1916 static inline void copy_vmcb_control_area(struct vmcb *dst_vmcb, struct vmcb *from_vmcb)
1917 {
1918         struct vmcb_control_area *dst  = &dst_vmcb->control;
1919         struct vmcb_control_area *from = &from_vmcb->control;
1920
1921         dst->intercept_cr_read    = from->intercept_cr_read;
1922         dst->intercept_cr_write   = from->intercept_cr_write;
1923         dst->intercept_dr_read    = from->intercept_dr_read;
1924         dst->intercept_dr_write   = from->intercept_dr_write;
1925         dst->intercept_exceptions = from->intercept_exceptions;
1926         dst->intercept            = from->intercept;
1927         dst->iopm_base_pa         = from->iopm_base_pa;
1928         dst->msrpm_base_pa        = from->msrpm_base_pa;
1929         dst->tsc_offset           = from->tsc_offset;
1930         dst->asid                 = from->asid;
1931         dst->tlb_ctl              = from->tlb_ctl;
1932         dst->int_ctl              = from->int_ctl;
1933         dst->int_vector           = from->int_vector;
1934         dst->int_state            = from->int_state;
1935         dst->exit_code            = from->exit_code;
1936         dst->exit_code_hi         = from->exit_code_hi;
1937         dst->exit_info_1          = from->exit_info_1;
1938         dst->exit_info_2          = from->exit_info_2;
1939         dst->exit_int_info        = from->exit_int_info;
1940         dst->exit_int_info_err    = from->exit_int_info_err;
1941         dst->nested_ctl           = from->nested_ctl;
1942         dst->event_inj            = from->event_inj;
1943         dst->event_inj_err        = from->event_inj_err;
1944         dst->nested_cr3           = from->nested_cr3;
1945         dst->lbr_ctl              = from->lbr_ctl;
1946 }
1947
1948 static int nested_svm_vmexit(struct vcpu_svm *svm)
1949 {
1950         struct vmcb *nested_vmcb;
1951         struct vmcb *hsave = svm->nested.hsave;
1952         struct vmcb *vmcb = svm->vmcb;
1953         struct page *page;
1954
1955         trace_kvm_nested_vmexit_inject(vmcb->control.exit_code,
1956                                        vmcb->control.exit_info_1,
1957                                        vmcb->control.exit_info_2,
1958                                        vmcb->control.exit_int_info,
1959                                        vmcb->control.exit_int_info_err);
1960
1961         nested_vmcb = nested_svm_map(svm, svm->nested.vmcb, &page);
1962         if (!nested_vmcb)
1963                 return 1;
1964
1965         /* Exit nested SVM mode */
1966         svm->nested.vmcb = 0;
1967
1968         /* Give the current vmcb to the guest */
1969         disable_gif(svm);
1970
1971         nested_vmcb->save.es     = vmcb->save.es;
1972         nested_vmcb->save.cs     = vmcb->save.cs;
1973         nested_vmcb->save.ss     = vmcb->save.ss;
1974         nested_vmcb->save.ds     = vmcb->save.ds;
1975         nested_vmcb->save.gdtr   = vmcb->save.gdtr;
1976         nested_vmcb->save.idtr   = vmcb->save.idtr;
1977         nested_vmcb->save.efer   = svm->vcpu.arch.efer;
1978         nested_vmcb->save.cr0    = kvm_read_cr0(&svm->vcpu);
1979         nested_vmcb->save.cr3    = svm->vcpu.arch.cr3;
1980         nested_vmcb->save.cr2    = vmcb->save.cr2;
1981         nested_vmcb->save.cr4    = svm->vcpu.arch.cr4;
1982         nested_vmcb->save.rflags = vmcb->save.rflags;
1983         nested_vmcb->save.rip    = vmcb->save.rip;
1984         nested_vmcb->save.rsp    = vmcb->save.rsp;
1985         nested_vmcb->save.rax    = vmcb->save.rax;
1986         nested_vmcb->save.dr7    = vmcb->save.dr7;
1987         nested_vmcb->save.dr6    = vmcb->save.dr6;
1988         nested_vmcb->save.cpl    = vmcb->save.cpl;
1989
1990         nested_vmcb->control.int_ctl           = vmcb->control.int_ctl;
1991         nested_vmcb->control.int_vector        = vmcb->control.int_vector;
1992         nested_vmcb->control.int_state         = vmcb->control.int_state;
1993         nested_vmcb->control.exit_code         = vmcb->control.exit_code;
1994         nested_vmcb->control.exit_code_hi      = vmcb->control.exit_code_hi;
1995         nested_vmcb->control.exit_info_1       = vmcb->control.exit_info_1;
1996         nested_vmcb->control.exit_info_2       = vmcb->control.exit_info_2;
1997         nested_vmcb->control.exit_int_info     = vmcb->control.exit_int_info;
1998         nested_vmcb->control.exit_int_info_err = vmcb->control.exit_int_info_err;
1999         nested_vmcb->control.next_rip          = vmcb->control.next_rip;
2000
2001         /*
2002          * If we emulate a VMRUN/#VMEXIT in the same host #vmexit cycle we have
2003          * to make sure that we do not lose injected events. So check event_inj
2004          * here and copy it to exit_int_info if it is valid.
2005          * Exit_int_info and event_inj can't be both valid because the case
2006          * below only happens on a VMRUN instruction intercept which has
2007          * no valid exit_int_info set.
2008          */
2009         if (vmcb->control.event_inj & SVM_EVTINJ_VALID) {
2010                 struct vmcb_control_area *nc = &nested_vmcb->control;
2011
2012                 nc->exit_int_info     = vmcb->control.event_inj;
2013                 nc->exit_int_info_err = vmcb->control.event_inj_err;
2014         }
2015
2016         nested_vmcb->control.tlb_ctl           = 0;
2017         nested_vmcb->control.event_inj         = 0;
2018         nested_vmcb->control.event_inj_err     = 0;
2019
2020         /* We always set V_INTR_MASKING and remember the old value in hflags */
2021         if (!(svm->vcpu.arch.hflags & HF_VINTR_MASK))
2022                 nested_vmcb->control.int_ctl &= ~V_INTR_MASKING_MASK;
2023
2024         /* Restore the original control entries */
2025         copy_vmcb_control_area(vmcb, hsave);
2026
2027         kvm_clear_exception_queue(&svm->vcpu);
2028         kvm_clear_interrupt_queue(&svm->vcpu);
2029
2030         svm->nested.nested_cr3 = 0;
2031
2032         /* Restore selected save entries */
2033         svm->vmcb->save.es = hsave->save.es;
2034         svm->vmcb->save.cs = hsave->save.cs;
2035         svm->vmcb->save.ss = hsave->save.ss;
2036         svm->vmcb->save.ds = hsave->save.ds;
2037         svm->vmcb->save.gdtr = hsave->save.gdtr;
2038         svm->vmcb->save.idtr = hsave->save.idtr;
2039         svm->vmcb->save.rflags = hsave->save.rflags;
2040         svm_set_efer(&svm->vcpu, hsave->save.efer);
2041         svm_set_cr0(&svm->vcpu, hsave->save.cr0 | X86_CR0_PE);
2042         svm_set_cr4(&svm->vcpu, hsave->save.cr4);
2043         if (npt_enabled) {
2044                 svm->vmcb->save.cr3 = hsave->save.cr3;
2045                 svm->vcpu.arch.cr3 = hsave->save.cr3;
2046         } else {
2047                 (void)kvm_set_cr3(&svm->vcpu, hsave->save.cr3);
2048         }
2049         kvm_register_write(&svm->vcpu, VCPU_REGS_RAX, hsave->save.rax);
2050         kvm_register_write(&svm->vcpu, VCPU_REGS_RSP, hsave->save.rsp);
2051         kvm_register_write(&svm->vcpu, VCPU_REGS_RIP, hsave->save.rip);
2052         svm->vmcb->save.dr7 = 0;
2053         svm->vmcb->save.cpl = 0;
2054         svm->vmcb->control.exit_int_info = 0;
2055
2056         nested_svm_unmap(page);
2057
2058         nested_svm_uninit_mmu_context(&svm->vcpu);
2059         kvm_mmu_reset_context(&svm->vcpu);
2060         kvm_mmu_load(&svm->vcpu);
2061
2062         return 0;
2063 }
2064
2065 static bool nested_svm_vmrun_msrpm(struct vcpu_svm *svm)
2066 {
2067         /*
2068          * This function merges the msr permission bitmaps of kvm and the
2069          * nested vmcb. It is omptimized in that it only merges the parts where
2070          * the kvm msr permission bitmap may contain zero bits
2071          */
2072         int i;
2073
2074         if (!(svm->nested.intercept & (1ULL << INTERCEPT_MSR_PROT)))
2075                 return true;
2076
2077         for (i = 0; i < MSRPM_OFFSETS; i++) {
2078                 u32 value, p;
2079                 u64 offset;
2080
2081                 if (msrpm_offsets[i] == 0xffffffff)
2082                         break;
2083
2084                 p      = msrpm_offsets[i];
2085                 offset = svm->nested.vmcb_msrpm + (p * 4);
2086
2087                 if (kvm_read_guest(svm->vcpu.kvm, offset, &value, 4))
2088                         return false;
2089
2090                 svm->nested.msrpm[p] = svm->msrpm[p] | value;
2091         }
2092
2093         svm->vmcb->control.msrpm_base_pa = __pa(svm->nested.msrpm);
2094
2095         return true;
2096 }
2097
2098 static bool nested_vmcb_checks(struct vmcb *vmcb)
2099 {
2100         if ((vmcb->control.intercept & (1ULL << INTERCEPT_VMRUN)) == 0)
2101                 return false;
2102
2103         if (vmcb->control.asid == 0)
2104                 return false;
2105
2106         if (vmcb->control.nested_ctl && !npt_enabled)
2107                 return false;
2108
2109         return true;
2110 }
2111
2112 static bool nested_svm_vmrun(struct vcpu_svm *svm)
2113 {
2114         struct vmcb *nested_vmcb;
2115         struct vmcb *hsave = svm->nested.hsave;
2116         struct vmcb *vmcb = svm->vmcb;
2117         struct page *page;
2118         u64 vmcb_gpa;
2119
2120         vmcb_gpa = svm->vmcb->save.rax;
2121
2122         nested_vmcb = nested_svm_map(svm, svm->vmcb->save.rax, &page);
2123         if (!nested_vmcb)
2124                 return false;
2125
2126         if (!nested_vmcb_checks(nested_vmcb)) {
2127                 nested_vmcb->control.exit_code    = SVM_EXIT_ERR;
2128                 nested_vmcb->control.exit_code_hi = 0;
2129                 nested_vmcb->control.exit_info_1  = 0;
2130                 nested_vmcb->control.exit_info_2  = 0;
2131
2132                 nested_svm_unmap(page);
2133
2134                 return false;
2135         }
2136
2137         trace_kvm_nested_vmrun(svm->vmcb->save.rip, vmcb_gpa,
2138                                nested_vmcb->save.rip,
2139                                nested_vmcb->control.int_ctl,
2140                                nested_vmcb->control.event_inj,
2141                                nested_vmcb->control.nested_ctl);
2142
2143         trace_kvm_nested_intercepts(nested_vmcb->control.intercept_cr_read,
2144                                     nested_vmcb->control.intercept_cr_write,
2145                                     nested_vmcb->control.intercept_exceptions,
2146                                     nested_vmcb->control.intercept);
2147
2148         /* Clear internal status */
2149         kvm_clear_exception_queue(&svm->vcpu);
2150         kvm_clear_interrupt_queue(&svm->vcpu);
2151
2152         /*
2153          * Save the old vmcb, so we don't need to pick what we save, but can
2154          * restore everything when a VMEXIT occurs
2155          */
2156         hsave->save.es     = vmcb->save.es;
2157         hsave->save.cs     = vmcb->save.cs;
2158         hsave->save.ss     = vmcb->save.ss;
2159         hsave->save.ds     = vmcb->save.ds;
2160         hsave->save.gdtr   = vmcb->save.gdtr;
2161         hsave->save.idtr   = vmcb->save.idtr;
2162         hsave->save.efer   = svm->vcpu.arch.efer;
2163         hsave->save.cr0    = kvm_read_cr0(&svm->vcpu);
2164         hsave->save.cr4    = svm->vcpu.arch.cr4;
2165         hsave->save.rflags = vmcb->save.rflags;
2166         hsave->save.rip    = kvm_rip_read(&svm->vcpu);
2167         hsave->save.rsp    = vmcb->save.rsp;
2168         hsave->save.rax    = vmcb->save.rax;
2169         if (npt_enabled)
2170                 hsave->save.cr3    = vmcb->save.cr3;
2171         else
2172                 hsave->save.cr3    = svm->vcpu.arch.cr3;
2173
2174         copy_vmcb_control_area(hsave, vmcb);
2175
2176         if (svm->vmcb->save.rflags & X86_EFLAGS_IF)
2177                 svm->vcpu.arch.hflags |= HF_HIF_MASK;
2178         else
2179                 svm->vcpu.arch.hflags &= ~HF_HIF_MASK;
2180
2181         if (nested_vmcb->control.nested_ctl) {
2182                 kvm_mmu_unload(&svm->vcpu);
2183                 svm->nested.nested_cr3 = nested_vmcb->control.nested_cr3;
2184                 nested_svm_init_mmu_context(&svm->vcpu);
2185         }
2186
2187         /* Load the nested guest state */
2188         svm->vmcb->save.es = nested_vmcb->save.es;
2189         svm->vmcb->save.cs = nested_vmcb->save.cs;
2190         svm->vmcb->save.ss = nested_vmcb->save.ss;
2191         svm->vmcb->save.ds = nested_vmcb->save.ds;
2192         svm->vmcb->save.gdtr = nested_vmcb->save.gdtr;
2193         svm->vmcb->save.idtr = nested_vmcb->save.idtr;
2194         svm->vmcb->save.rflags = nested_vmcb->save.rflags;
2195         svm_set_efer(&svm->vcpu, nested_vmcb->save.efer);
2196         svm_set_cr0(&svm->vcpu, nested_vmcb->save.cr0);
2197         svm_set_cr4(&svm->vcpu, nested_vmcb->save.cr4);
2198         if (npt_enabled) {
2199                 svm->vmcb->save.cr3 = nested_vmcb->save.cr3;
2200                 svm->vcpu.arch.cr3 = nested_vmcb->save.cr3;
2201         } else
2202                 (void)kvm_set_cr3(&svm->vcpu, nested_vmcb->save.cr3);
2203
2204         /* Guest paging mode is active - reset mmu */
2205         kvm_mmu_reset_context(&svm->vcpu);
2206
2207         svm->vmcb->save.cr2 = svm->vcpu.arch.cr2 = nested_vmcb->save.cr2;
2208         kvm_register_write(&svm->vcpu, VCPU_REGS_RAX, nested_vmcb->save.rax);
2209         kvm_register_write(&svm->vcpu, VCPU_REGS_RSP, nested_vmcb->save.rsp);
2210         kvm_register_write(&svm->vcpu, VCPU_REGS_RIP, nested_vmcb->save.rip);
2211
2212         /* In case we don't even reach vcpu_run, the fields are not updated */
2213         svm->vmcb->save.rax = nested_vmcb->save.rax;
2214         svm->vmcb->save.rsp = nested_vmcb->save.rsp;
2215         svm->vmcb->save.rip = nested_vmcb->save.rip;
2216         svm->vmcb->save.dr7 = nested_vmcb->save.dr7;
2217         svm->vmcb->save.dr6 = nested_vmcb->save.dr6;
2218         svm->vmcb->save.cpl = nested_vmcb->save.cpl;
2219
2220         svm->nested.vmcb_msrpm = nested_vmcb->control.msrpm_base_pa & ~0x0fffULL;
2221         svm->nested.vmcb_iopm  = nested_vmcb->control.iopm_base_pa  & ~0x0fffULL;
2222
2223         /* cache intercepts */
2224         svm->nested.intercept_cr_read    = nested_vmcb->control.intercept_cr_read;
2225         svm->nested.intercept_cr_write   = nested_vmcb->control.intercept_cr_write;
2226         svm->nested.intercept_dr_read    = nested_vmcb->control.intercept_dr_read;
2227         svm->nested.intercept_dr_write   = nested_vmcb->control.intercept_dr_write;
2228         svm->nested.intercept_exceptions = nested_vmcb->control.intercept_exceptions;
2229         svm->nested.intercept            = nested_vmcb->control.intercept;
2230
2231         force_new_asid(&svm->vcpu);
2232         svm->vmcb->control.int_ctl = nested_vmcb->control.int_ctl | V_INTR_MASKING_MASK;
2233         if (nested_vmcb->control.int_ctl & V_INTR_MASKING_MASK)
2234                 svm->vcpu.arch.hflags |= HF_VINTR_MASK;
2235         else
2236                 svm->vcpu.arch.hflags &= ~HF_VINTR_MASK;
2237
2238         if (svm->vcpu.arch.hflags & HF_VINTR_MASK) {
2239                 /* We only want the cr8 intercept bits of the guest */
2240                 svm->vmcb->control.intercept_cr_read &= ~INTERCEPT_CR8_MASK;
2241                 svm->vmcb->control.intercept_cr_write &= ~INTERCEPT_CR8_MASK;
2242         }
2243
2244         /* We don't want to see VMMCALLs from a nested guest */
2245         svm->vmcb->control.intercept &= ~(1ULL << INTERCEPT_VMMCALL);
2246
2247         /*
2248          * We don't want a nested guest to be more powerful than the guest, so
2249          * all intercepts are ORed
2250          */
2251         svm->vmcb->control.intercept_cr_read |=
2252                 nested_vmcb->control.intercept_cr_read;
2253         svm->vmcb->control.intercept_cr_write |=
2254                 nested_vmcb->control.intercept_cr_write;
2255         svm->vmcb->control.intercept_dr_read |=
2256                 nested_vmcb->control.intercept_dr_read;
2257         svm->vmcb->control.intercept_dr_write |=
2258                 nested_vmcb->control.intercept_dr_write;
2259         svm->vmcb->control.intercept_exceptions |=
2260                 nested_vmcb->control.intercept_exceptions;
2261
2262         svm->vmcb->control.intercept |= nested_vmcb->control.intercept;
2263
2264         svm->vmcb->control.lbr_ctl = nested_vmcb->control.lbr_ctl;
2265         svm->vmcb->control.int_vector = nested_vmcb->control.int_vector;
2266         svm->vmcb->control.int_state = nested_vmcb->control.int_state;
2267         svm->vmcb->control.tsc_offset += nested_vmcb->control.tsc_offset;
2268         svm->vmcb->control.event_inj = nested_vmcb->control.event_inj;
2269         svm->vmcb->control.event_inj_err = nested_vmcb->control.event_inj_err;
2270
2271         nested_svm_unmap(page);
2272
2273         /* nested_vmcb is our indicator if nested SVM is activated */
2274         svm->nested.vmcb = vmcb_gpa;
2275
2276         enable_gif(svm);
2277
2278         return true;
2279 }
2280
2281 static void nested_svm_vmloadsave(struct vmcb *from_vmcb, struct vmcb *to_vmcb)
2282 {
2283         to_vmcb->save.fs = from_vmcb->save.fs;
2284         to_vmcb->save.gs = from_vmcb->save.gs;
2285         to_vmcb->save.tr = from_vmcb->save.tr;
2286         to_vmcb->save.ldtr = from_vmcb->save.ldtr;
2287         to_vmcb->save.kernel_gs_base = from_vmcb->save.kernel_gs_base;
2288         to_vmcb->save.star = from_vmcb->save.star;
2289         to_vmcb->save.lstar = from_vmcb->save.lstar;
2290         to_vmcb->save.cstar = from_vmcb->save.cstar;
2291         to_vmcb->save.sfmask = from_vmcb->save.sfmask;
2292         to_vmcb->save.sysenter_cs = from_vmcb->save.sysenter_cs;
2293         to_vmcb->save.sysenter_esp = from_vmcb->save.sysenter_esp;
2294         to_vmcb->save.sysenter_eip = from_vmcb->save.sysenter_eip;
2295 }
2296
2297 static int vmload_interception(struct vcpu_svm *svm)
2298 {
2299         struct vmcb *nested_vmcb;
2300         struct page *page;
2301
2302         if (nested_svm_check_permissions(svm))
2303                 return 1;
2304
2305         svm->next_rip = kvm_rip_read(&svm->vcpu) + 3;
2306         skip_emulated_instruction(&svm->vcpu);
2307
2308         nested_vmcb = nested_svm_map(svm, svm->vmcb->save.rax, &page);
2309         if (!nested_vmcb)
2310                 return 1;
2311
2312         nested_svm_vmloadsave(nested_vmcb, svm->vmcb);
2313         nested_svm_unmap(page);
2314
2315         return 1;
2316 }
2317
2318 static int vmsave_interception(struct vcpu_svm *svm)
2319 {
2320         struct vmcb *nested_vmcb;
2321         struct page *page;
2322
2323         if (nested_svm_check_permissions(svm))
2324                 return 1;
2325
2326         svm->next_rip = kvm_rip_read(&svm->vcpu) + 3;
2327         skip_emulated_instruction(&svm->vcpu);
2328
2329         nested_vmcb = nested_svm_map(svm, svm->vmcb->save.rax, &page);
2330         if (!nested_vmcb)
2331                 return 1;
2332
2333         nested_svm_vmloadsave(svm->vmcb, nested_vmcb);
2334         nested_svm_unmap(page);
2335
2336         return 1;
2337 }
2338
2339 static int vmrun_interception(struct vcpu_svm *svm)
2340 {
2341         if (nested_svm_check_permissions(svm))
2342                 return 1;
2343
2344         /* Save rip after vmrun instruction */
2345         kvm_rip_write(&svm->vcpu, kvm_rip_read(&svm->vcpu) + 3);
2346
2347         if (!nested_svm_vmrun(svm))
2348                 return 1;
2349
2350         if (!nested_svm_vmrun_msrpm(svm))
2351                 goto failed;
2352
2353         return 1;
2354
2355 failed:
2356
2357         svm->vmcb->control.exit_code    = SVM_EXIT_ERR;
2358         svm->vmcb->control.exit_code_hi = 0;
2359         svm->vmcb->control.exit_info_1  = 0;
2360         svm->vmcb->control.exit_info_2  = 0;
2361
2362         nested_svm_vmexit(svm);
2363
2364         return 1;
2365 }
2366
2367 static int stgi_interception(struct vcpu_svm *svm)
2368 {
2369         if (nested_svm_check_permissions(svm))
2370                 return 1;
2371
2372         svm->next_rip = kvm_rip_read(&svm->vcpu) + 3;
2373         skip_emulated_instruction(&svm->vcpu);
2374         kvm_make_request(KVM_REQ_EVENT, &svm->vcpu);
2375
2376         enable_gif(svm);
2377
2378         return 1;
2379 }
2380
2381 static int clgi_interception(struct vcpu_svm *svm)
2382 {
2383         if (nested_svm_check_permissions(svm))
2384                 return 1;
2385
2386         svm->next_rip = kvm_rip_read(&svm->vcpu) + 3;
2387         skip_emulated_instruction(&svm->vcpu);
2388
2389         disable_gif(svm);
2390
2391         /* After a CLGI no interrupts should come */
2392         svm_clear_vintr(svm);
2393         svm->vmcb->control.int_ctl &= ~V_IRQ_MASK;
2394
2395         return 1;
2396 }
2397
2398 static int invlpga_interception(struct vcpu_svm *svm)
2399 {
2400         struct kvm_vcpu *vcpu = &svm->vcpu;
2401
2402         trace_kvm_invlpga(svm->vmcb->save.rip, vcpu->arch.regs[VCPU_REGS_RCX],
2403                           vcpu->arch.regs[VCPU_REGS_RAX]);
2404
2405         /* Let's treat INVLPGA the same as INVLPG (can be optimized!) */
2406         kvm_mmu_invlpg(vcpu, vcpu->arch.regs[VCPU_REGS_RAX]);
2407
2408         svm->next_rip = kvm_rip_read(&svm->vcpu) + 3;
2409         skip_emulated_instruction(&svm->vcpu);
2410         return 1;
2411 }
2412
2413 static int skinit_interception(struct vcpu_svm *svm)
2414 {
2415         trace_kvm_skinit(svm->vmcb->save.rip, svm->vcpu.arch.regs[VCPU_REGS_RAX]);
2416
2417         kvm_queue_exception(&svm->vcpu, UD_VECTOR);
2418         return 1;
2419 }
2420
2421 static int invalid_op_interception(struct vcpu_svm *svm)
2422 {
2423         kvm_queue_exception(&svm->vcpu, UD_VECTOR);
2424         return 1;
2425 }
2426
2427 static int task_switch_interception(struct vcpu_svm *svm)
2428 {
2429         u16 tss_selector;
2430         int reason;
2431         int int_type = svm->vmcb->control.exit_int_info &
2432                 SVM_EXITINTINFO_TYPE_MASK;
2433         int int_vec = svm->vmcb->control.exit_int_info & SVM_EVTINJ_VEC_MASK;
2434         uint32_t type =
2435                 svm->vmcb->control.exit_int_info & SVM_EXITINTINFO_TYPE_MASK;
2436         uint32_t idt_v =
2437                 svm->vmcb->control.exit_int_info & SVM_EXITINTINFO_VALID;
2438         bool has_error_code = false;
2439         u32 error_code = 0;
2440
2441         tss_selector = (u16)svm->vmcb->control.exit_info_1;
2442
2443         if (svm->vmcb->control.exit_info_2 &
2444             (1ULL << SVM_EXITINFOSHIFT_TS_REASON_IRET))
2445                 reason = TASK_SWITCH_IRET;
2446         else if (svm->vmcb->control.exit_info_2 &
2447                  (1ULL << SVM_EXITINFOSHIFT_TS_REASON_JMP))
2448                 reason = TASK_SWITCH_JMP;
2449         else if (idt_v)
2450                 reason = TASK_SWITCH_GATE;
2451         else
2452                 reason = TASK_SWITCH_CALL;
2453
2454         if (reason == TASK_SWITCH_GATE) {
2455                 switch (type) {
2456                 case SVM_EXITINTINFO_TYPE_NMI:
2457                         svm->vcpu.arch.nmi_injected = false;
2458                         break;
2459                 case SVM_EXITINTINFO_TYPE_EXEPT:
2460                         if (svm->vmcb->control.exit_info_2 &
2461                             (1ULL << SVM_EXITINFOSHIFT_TS_HAS_ERROR_CODE)) {
2462                                 has_error_code = true;
2463                                 error_code =
2464                                         (u32)svm->vmcb->control.exit_info_2;
2465                         }
2466                         kvm_clear_exception_queue(&svm->vcpu);
2467                         break;
2468                 case SVM_EXITINTINFO_TYPE_INTR:
2469                         kvm_clear_interrupt_queue(&svm->vcpu);
2470                         break;
2471                 default:
2472                         break;
2473                 }
2474         }
2475
2476         if (reason != TASK_SWITCH_GATE ||
2477             int_type == SVM_EXITINTINFO_TYPE_SOFT ||
2478             (int_type == SVM_EXITINTINFO_TYPE_EXEPT &&
2479              (int_vec == OF_VECTOR || int_vec == BP_VECTOR)))
2480                 skip_emulated_instruction(&svm->vcpu);
2481
2482         if (kvm_task_switch(&svm->vcpu, tss_selector, reason,
2483                                 has_error_code, error_code) == EMULATE_FAIL) {
2484                 svm->vcpu.run->exit_reason = KVM_EXIT_INTERNAL_ERROR;
2485                 svm->vcpu.run->internal.suberror = KVM_INTERNAL_ERROR_EMULATION;
2486                 svm->vcpu.run->internal.ndata = 0;
2487                 return 0;
2488         }
2489         return 1;
2490 }
2491
2492 static int cpuid_interception(struct vcpu_svm *svm)
2493 {
2494         svm->next_rip = kvm_rip_read(&svm->vcpu) + 2;
2495         kvm_emulate_cpuid(&svm->vcpu);
2496         return 1;
2497 }
2498
2499 static int iret_interception(struct vcpu_svm *svm)
2500 {
2501         ++svm->vcpu.stat.nmi_window_exits;
2502         svm->vmcb->control.intercept &= ~(1ULL << INTERCEPT_IRET);
2503         svm->vcpu.arch.hflags |= HF_IRET_MASK;
2504         return 1;
2505 }
2506
2507 static int invlpg_interception(struct vcpu_svm *svm)
2508 {
2509         return emulate_instruction(&svm->vcpu, 0, 0, 0) == EMULATE_DONE;
2510 }
2511
2512 static int emulate_on_interception(struct vcpu_svm *svm)
2513 {
2514         return emulate_instruction(&svm->vcpu, 0, 0, 0) == EMULATE_DONE;
2515 }
2516
2517 static int cr0_write_interception(struct vcpu_svm *svm)
2518 {
2519         struct kvm_vcpu *vcpu = &svm->vcpu;
2520         int r;
2521
2522         r = emulate_instruction(&svm->vcpu, 0, 0, 0);
2523
2524         if (svm->nested.vmexit_rip) {
2525                 kvm_register_write(vcpu, VCPU_REGS_RIP, svm->nested.vmexit_rip);
2526                 kvm_register_write(vcpu, VCPU_REGS_RSP, svm->nested.vmexit_rsp);
2527                 kvm_register_write(vcpu, VCPU_REGS_RAX, svm->nested.vmexit_rax);
2528                 svm->nested.vmexit_rip = 0;
2529         }
2530
2531         return r == EMULATE_DONE;
2532 }
2533
2534 static int cr8_write_interception(struct vcpu_svm *svm)
2535 {
2536         struct kvm_run *kvm_run = svm->vcpu.run;
2537
2538         u8 cr8_prev = kvm_get_cr8(&svm->vcpu);
2539         /* instruction emulation calls kvm_set_cr8() */
2540         emulate_instruction(&svm->vcpu, 0, 0, 0);
2541         if (irqchip_in_kernel(svm->vcpu.kvm)) {
2542                 svm->vmcb->control.intercept_cr_write &= ~INTERCEPT_CR8_MASK;
2543                 return 1;
2544         }
2545         if (cr8_prev <= kvm_get_cr8(&svm->vcpu))
2546                 return 1;
2547         kvm_run->exit_reason = KVM_EXIT_SET_TPR;
2548         return 0;
2549 }
2550
2551 static int svm_get_msr(struct kvm_vcpu *vcpu, unsigned ecx, u64 *data)
2552 {
2553         struct vcpu_svm *svm = to_svm(vcpu);
2554
2555         switch (ecx) {
2556         case MSR_IA32_TSC: {
2557                 u64 tsc_offset;
2558
2559                 if (is_nested(svm))
2560                         tsc_offset = svm->nested.hsave->control.tsc_offset;
2561                 else
2562                         tsc_offset = svm->vmcb->control.tsc_offset;
2563
2564                 *data = tsc_offset + native_read_tsc();
2565                 break;
2566         }
2567         case MSR_STAR:
2568                 *data = svm->vmcb->save.star;
2569                 break;
2570 #ifdef CONFIG_X86_64
2571         case MSR_LSTAR:
2572                 *data = svm->vmcb->save.lstar;
2573                 break;
2574         case MSR_CSTAR:
2575                 *data = svm->vmcb->save.cstar;
2576                 break;
2577         case MSR_KERNEL_GS_BASE:
2578                 *data = svm->vmcb->save.kernel_gs_base;
2579                 break;
2580         case MSR_SYSCALL_MASK:
2581                 *data = svm->vmcb->save.sfmask;
2582                 break;
2583 #endif
2584         case MSR_IA32_SYSENTER_CS:
2585                 *data = svm->vmcb->save.sysenter_cs;
2586                 break;
2587         case MSR_IA32_SYSENTER_EIP:
2588                 *data = svm->sysenter_eip;
2589                 break;
2590         case MSR_IA32_SYSENTER_ESP:
2591                 *data = svm->sysenter_esp;
2592                 break;
2593         /*
2594          * Nobody will change the following 5 values in the VMCB so we can
2595          * safely return them on rdmsr. They will always be 0 until LBRV is
2596          * implemented.
2597          */
2598         case MSR_IA32_DEBUGCTLMSR:
2599                 *data = svm->vmcb->save.dbgctl;
2600                 break;
2601         case MSR_IA32_LASTBRANCHFROMIP:
2602                 *data = svm->vmcb->save.br_from;
2603                 break;
2604         case MSR_IA32_LASTBRANCHTOIP:
2605                 *data = svm->vmcb->save.br_to;
2606                 break;
2607         case MSR_IA32_LASTINTFROMIP:
2608                 *data = svm->vmcb->save.last_excp_from;
2609                 break;
2610         case MSR_IA32_LASTINTTOIP:
2611                 *data = svm->vmcb->save.last_excp_to;
2612                 break;
2613         case MSR_VM_HSAVE_PA:
2614                 *data = svm->nested.hsave_msr;
2615                 break;
2616         case MSR_VM_CR:
2617                 *data = svm->nested.vm_cr_msr;
2618                 break;
2619         case MSR_IA32_UCODE_REV:
2620                 *data = 0x01000065;
2621                 break;
2622         default:
2623                 return kvm_get_msr_common(vcpu, ecx, data);
2624         }
2625         return 0;
2626 }
2627
2628 static int rdmsr_interception(struct vcpu_svm *svm)
2629 {
2630         u32 ecx = svm->vcpu.arch.regs[VCPU_REGS_RCX];
2631         u64 data;
2632
2633         if (svm_get_msr(&svm->vcpu, ecx, &data)) {
2634                 trace_kvm_msr_read_ex(ecx);
2635                 kvm_inject_gp(&svm->vcpu, 0);
2636         } else {
2637                 trace_kvm_msr_read(ecx, data);
2638
2639                 svm->vcpu.arch.regs[VCPU_REGS_RAX] = data & 0xffffffff;
2640                 svm->vcpu.arch.regs[VCPU_REGS_RDX] = data >> 32;
2641                 svm->next_rip = kvm_rip_read(&svm->vcpu) + 2;
2642                 skip_emulated_instruction(&svm->vcpu);
2643         }
2644         return 1;
2645 }
2646
2647 static int svm_set_vm_cr(struct kvm_vcpu *vcpu, u64 data)
2648 {
2649         struct vcpu_svm *svm = to_svm(vcpu);
2650         int svm_dis, chg_mask;
2651
2652         if (data & ~SVM_VM_CR_VALID_MASK)
2653                 return 1;
2654
2655         chg_mask = SVM_VM_CR_VALID_MASK;
2656
2657         if (svm->nested.vm_cr_msr & SVM_VM_CR_SVM_DIS_MASK)
2658                 chg_mask &= ~(SVM_VM_CR_SVM_LOCK_MASK | SVM_VM_CR_SVM_DIS_MASK);
2659
2660         svm->nested.vm_cr_msr &= ~chg_mask;
2661         svm->nested.vm_cr_msr |= (data & chg_mask);
2662
2663         svm_dis = svm->nested.vm_cr_msr & SVM_VM_CR_SVM_DIS_MASK;
2664
2665         /* check for svm_disable while efer.svme is set */
2666         if (svm_dis && (vcpu->arch.efer & EFER_SVME))
2667                 return 1;
2668
2669         return 0;
2670 }
2671
2672 static int svm_set_msr(struct kvm_vcpu *vcpu, unsigned ecx, u64 data)
2673 {
2674         struct vcpu_svm *svm = to_svm(vcpu);
2675
2676         switch (ecx) {
2677         case MSR_IA32_TSC:
2678                 kvm_write_tsc(vcpu, data);
2679                 break;
2680         case MSR_STAR:
2681                 svm->vmcb->save.star = data;
2682                 break;
2683 #ifdef CONFIG_X86_64
2684         case MSR_LSTAR:
2685                 svm->vmcb->save.lstar = data;
2686                 break;
2687         case MSR_CSTAR:
2688                 svm->vmcb->save.cstar = data;
2689                 break;
2690         case MSR_KERNEL_GS_BASE:
2691                 svm->vmcb->save.kernel_gs_base = data;
2692                 break;
2693         case MSR_SYSCALL_MASK:
2694                 svm->vmcb->save.sfmask = data;
2695                 break;
2696 #endif
2697         case MSR_IA32_SYSENTER_CS:
2698                 svm->vmcb->save.sysenter_cs = data;
2699                 break;
2700         case MSR_IA32_SYSENTER_EIP:
2701                 svm->sysenter_eip = data;
2702                 svm->vmcb->save.sysenter_eip = data;
2703                 break;
2704         case MSR_IA32_SYSENTER_ESP:
2705                 svm->sysenter_esp = data;
2706                 svm->vmcb->save.sysenter_esp = data;
2707                 break;
2708         case MSR_IA32_DEBUGCTLMSR:
2709                 if (!svm_has(SVM_FEATURE_LBRV)) {
2710                         pr_unimpl(vcpu, "%s: MSR_IA32_DEBUGCTL 0x%llx, nop\n",
2711                                         __func__, data);
2712                         break;
2713                 }
2714                 if (data & DEBUGCTL_RESERVED_BITS)
2715                         return 1;
2716
2717                 svm->vmcb->save.dbgctl = data;
2718                 if (data & (1ULL<<0))
2719                         svm_enable_lbrv(svm);
2720                 else
2721                         svm_disable_lbrv(svm);
2722                 break;
2723         case MSR_VM_HSAVE_PA:
2724                 svm->nested.hsave_msr = data;
2725                 break;
2726         case MSR_VM_CR:
2727                 return svm_set_vm_cr(vcpu, data);
2728         case MSR_VM_IGNNE:
2729                 pr_unimpl(vcpu, "unimplemented wrmsr: 0x%x data 0x%llx\n", ecx, data);
2730                 break;
2731         default:
2732                 return kvm_set_msr_common(vcpu, ecx, data);
2733         }
2734         return 0;
2735 }
2736
2737 static int wrmsr_interception(struct vcpu_svm *svm)
2738 {
2739         u32 ecx = svm->vcpu.arch.regs[VCPU_REGS_RCX];
2740         u64 data = (svm->vcpu.arch.regs[VCPU_REGS_RAX] & -1u)
2741                 | ((u64)(svm->vcpu.arch.regs[VCPU_REGS_RDX] & -1u) << 32);
2742
2743
2744         svm->next_rip = kvm_rip_read(&svm->vcpu) + 2;
2745         if (svm_set_msr(&svm->vcpu, ecx, data)) {
2746                 trace_kvm_msr_write_ex(ecx, data);
2747                 kvm_inject_gp(&svm->vcpu, 0);
2748         } else {
2749                 trace_kvm_msr_write(ecx, data);
2750                 skip_emulated_instruction(&svm->vcpu);
2751         }
2752         return 1;
2753 }
2754
2755 static int msr_interception(struct vcpu_svm *svm)
2756 {
2757         if (svm->vmcb->control.exit_info_1)
2758                 return wrmsr_interception(svm);
2759         else
2760                 return rdmsr_interception(svm);
2761 }
2762
2763 static int interrupt_window_interception(struct vcpu_svm *svm)
2764 {
2765         struct kvm_run *kvm_run = svm->vcpu.run;
2766
2767         kvm_make_request(KVM_REQ_EVENT, &svm->vcpu);
2768         svm_clear_vintr(svm);
2769         svm->vmcb->control.int_ctl &= ~V_IRQ_MASK;
2770         /*
2771          * If the user space waits to inject interrupts, exit as soon as
2772          * possible
2773          */
2774         if (!irqchip_in_kernel(svm->vcpu.kvm) &&
2775             kvm_run->request_interrupt_window &&
2776             !kvm_cpu_has_interrupt(&svm->vcpu)) {
2777                 ++svm->vcpu.stat.irq_window_exits;
2778                 kvm_run->exit_reason = KVM_EXIT_IRQ_WINDOW_OPEN;
2779                 return 0;
2780         }
2781
2782         return 1;
2783 }
2784
2785 static int pause_interception(struct vcpu_svm *svm)
2786 {
2787         kvm_vcpu_on_spin(&(svm->vcpu));
2788         return 1;
2789 }
2790
2791 static int (*svm_exit_handlers[])(struct vcpu_svm *svm) = {
2792         [SVM_EXIT_READ_CR0]                     = emulate_on_interception,
2793         [SVM_EXIT_READ_CR3]                     = emulate_on_interception,
2794         [SVM_EXIT_READ_CR4]                     = emulate_on_interception,
2795         [SVM_EXIT_READ_CR8]                     = emulate_on_interception,
2796         [SVM_EXIT_CR0_SEL_WRITE]                = emulate_on_interception,
2797         [SVM_EXIT_WRITE_CR0]                    = cr0_write_interception,
2798         [SVM_EXIT_WRITE_CR3]                    = emulate_on_interception,
2799         [SVM_EXIT_WRITE_CR4]                    = emulate_on_interception,
2800         [SVM_EXIT_WRITE_CR8]                    = cr8_write_interception,
2801         [SVM_EXIT_READ_DR0]                     = emulate_on_interception,
2802         [SVM_EXIT_READ_DR1]                     = emulate_on_interception,
2803         [SVM_EXIT_READ_DR2]                     = emulate_on_interception,
2804         [SVM_EXIT_READ_DR3]                     = emulate_on_interception,
2805         [SVM_EXIT_READ_DR4]                     = emulate_on_interception,
2806         [SVM_EXIT_READ_DR5]                     = emulate_on_interception,
2807         [SVM_EXIT_READ_DR6]                     = emulate_on_interception,
2808         [SVM_EXIT_READ_DR7]                     = emulate_on_interception,
2809         [SVM_EXIT_WRITE_DR0]                    = emulate_on_interception,
2810         [SVM_EXIT_WRITE_DR1]                    = emulate_on_interception,
2811         [SVM_EXIT_WRITE_DR2]                    = emulate_on_interception,
2812         [SVM_EXIT_WRITE_DR3]                    = emulate_on_interception,
2813         [SVM_EXIT_WRITE_DR4]                    = emulate_on_interception,
2814         [SVM_EXIT_WRITE_DR5]                    = emulate_on_interception,
2815         [SVM_EXIT_WRITE_DR6]                    = emulate_on_interception,
2816         [SVM_EXIT_WRITE_DR7]                    = emulate_on_interception,
2817         [SVM_EXIT_EXCP_BASE + DB_VECTOR]        = db_interception,
2818         [SVM_EXIT_EXCP_BASE + BP_VECTOR]        = bp_interception,
2819         [SVM_EXIT_EXCP_BASE + UD_VECTOR]        = ud_interception,
2820         [SVM_EXIT_EXCP_BASE + PF_VECTOR]        = pf_interception,
2821         [SVM_EXIT_EXCP_BASE + NM_VECTOR]        = nm_interception,
2822         [SVM_EXIT_EXCP_BASE + MC_VECTOR]        = mc_interception,
2823         [SVM_EXIT_INTR]                         = intr_interception,
2824         [SVM_EXIT_NMI]                          = nmi_interception,
2825         [SVM_EXIT_SMI]                          = nop_on_interception,
2826         [SVM_EXIT_INIT]                         = nop_on_interception,
2827         [SVM_EXIT_VINTR]                        = interrupt_window_interception,
2828         [SVM_EXIT_CPUID]                        = cpuid_interception,
2829         [SVM_EXIT_IRET]                         = iret_interception,
2830         [SVM_EXIT_INVD]                         = emulate_on_interception,
2831         [SVM_EXIT_PAUSE]                        = pause_interception,
2832         [SVM_EXIT_HLT]                          = halt_interception,
2833         [SVM_EXIT_INVLPG]                       = invlpg_interception,
2834         [SVM_EXIT_INVLPGA]                      = invlpga_interception,
2835         [SVM_EXIT_IOIO]                         = io_interception,
2836         [SVM_EXIT_MSR]                          = msr_interception,
2837         [SVM_EXIT_TASK_SWITCH]                  = task_switch_interception,
2838         [SVM_EXIT_SHUTDOWN]                     = shutdown_interception,
2839         [SVM_EXIT_VMRUN]                        = vmrun_interception,
2840         [SVM_EXIT_VMMCALL]                      = vmmcall_interception,
2841         [SVM_EXIT_VMLOAD]                       = vmload_interception,
2842         [SVM_EXIT_VMSAVE]                       = vmsave_interception,
2843         [SVM_EXIT_STGI]                         = stgi_interception,
2844         [SVM_EXIT_CLGI]                         = clgi_interception,
2845         [SVM_EXIT_SKINIT]                       = skinit_interception,
2846         [SVM_EXIT_WBINVD]                       = emulate_on_interception,
2847         [SVM_EXIT_MONITOR]                      = invalid_op_interception,
2848         [SVM_EXIT_MWAIT]                        = invalid_op_interception,
2849         [SVM_EXIT_NPF]                          = pf_interception,
2850 };
2851
2852 void dump_vmcb(struct kvm_vcpu *vcpu)
2853 {
2854         struct vcpu_svm *svm = to_svm(vcpu);
2855         struct vmcb_control_area *control = &svm->vmcb->control;
2856         struct vmcb_save_area *save = &svm->vmcb->save;
2857
2858         pr_err("VMCB Control Area:\n");
2859         pr_err("cr_read:            %04x\n", control->intercept_cr_read);
2860         pr_err("cr_write:           %04x\n", control->intercept_cr_write);
2861         pr_err("dr_read:            %04x\n", control->intercept_dr_read);
2862         pr_err("dr_write:           %04x\n", control->intercept_dr_write);
2863         pr_err("exceptions:         %08x\n", control->intercept_exceptions);
2864         pr_err("intercepts:         %016llx\n", control->intercept);
2865         pr_err("pause filter count: %d\n", control->pause_filter_count);
2866         pr_err("iopm_base_pa:       %016llx\n", control->iopm_base_pa);
2867         pr_err("msrpm_base_pa:      %016llx\n", control->msrpm_base_pa);
2868         pr_err("tsc_offset:         %016llx\n", control->tsc_offset);
2869         pr_err("asid:               %d\n", control->asid);
2870         pr_err("tlb_ctl:            %d\n", control->tlb_ctl);
2871         pr_err("int_ctl:            %08x\n", control->int_ctl);
2872         pr_err("int_vector:         %08x\n", control->int_vector);
2873         pr_err("int_state:          %08x\n", control->int_state);
2874         pr_err("exit_code:          %08x\n", control->exit_code);
2875         pr_err("exit_info1:         %016llx\n", control->exit_info_1);
2876         pr_err("exit_info2:         %016llx\n", control->exit_info_2);
2877         pr_err("exit_int_info:      %08x\n", control->exit_int_info);
2878         pr_err("exit_int_info_err:  %08x\n", control->exit_int_info_err);
2879         pr_err("nested_ctl:         %lld\n", control->nested_ctl);
2880         pr_err("nested_cr3:         %016llx\n", control->nested_cr3);
2881         pr_err("event_inj:          %08x\n", control->event_inj);
2882         pr_err("event_inj_err:      %08x\n", control->event_inj_err);
2883         pr_err("lbr_ctl:            %lld\n", control->lbr_ctl);
2884         pr_err("next_rip:           %016llx\n", control->next_rip);
2885         pr_err("VMCB State Save Area:\n");
2886         pr_err("es:   s: %04x a: %04x l: %08x b: %016llx\n",
2887                 save->es.selector, save->es.attrib,
2888                 save->es.limit, save->es.base);
2889         pr_err("cs:   s: %04x a: %04x l: %08x b: %016llx\n",
2890                 save->cs.selector, save->cs.attrib,
2891                 save->cs.limit, save->cs.base);
2892         pr_err("ss:   s: %04x a: %04x l: %08x b: %016llx\n",
2893                 save->ss.selector, save->ss.attrib,
2894                 save->ss.limit, save->ss.base);
2895         pr_err("ds:   s: %04x a: %04x l: %08x b: %016llx\n",
2896                 save->ds.selector, save->ds.attrib,
2897                 save->ds.limit, save->ds.base);
2898         pr_err("fs:   s: %04x a: %04x l: %08x b: %016llx\n",
2899                 save->fs.selector, save->fs.attrib,
2900                 save->fs.limit, save->fs.base);
2901         pr_err("gs:   s: %04x a: %04x l: %08x b: %016llx\n",
2902                 save->gs.selector, save->gs.attrib,
2903                 save->gs.limit, save->gs.base);
2904         pr_err("gdtr: s: %04x a: %04x l: %08x b: %016llx\n",
2905                 save->gdtr.selector, save->gdtr.attrib,
2906                 save->gdtr.limit, save->gdtr.base);
2907         pr_err("ldtr: s: %04x a: %04x l: %08x b: %016llx\n",
2908                 save->ldtr.selector, save->ldtr.attrib,
2909                 save->ldtr.limit, save->ldtr.base);
2910         pr_err("idtr: s: %04x a: %04x l: %08x b: %016llx\n",
2911                 save->idtr.selector, save->idtr.attrib,
2912                 save->idtr.limit, save->idtr.base);
2913         pr_err("tr:   s: %04x a: %04x l: %08x b: %016llx\n",
2914                 save->tr.selector, save->tr.attrib,
2915                 save->tr.limit, save->tr.base);
2916         pr_err("cpl:            %d                efer:         %016llx\n",
2917                 save->cpl, save->efer);
2918         pr_err("cr0:            %016llx cr2:          %016llx\n",
2919                 save->cr0, save->cr2);
2920         pr_err("cr3:            %016llx cr4:          %016llx\n",
2921                 save->cr3, save->cr4);
2922         pr_err("dr6:            %016llx dr7:          %016llx\n",
2923                 save->dr6, save->dr7);
2924         pr_err("rip:            %016llx rflags:       %016llx\n",
2925                 save->rip, save->rflags);
2926         pr_err("rsp:            %016llx rax:          %016llx\n",
2927                 save->rsp, save->rax);
2928         pr_err("star:           %016llx lstar:        %016llx\n",
2929                 save->star, save->lstar);
2930         pr_err("cstar:          %016llx sfmask:       %016llx\n",
2931                 save->cstar, save->sfmask);
2932         pr_err("kernel_gs_base: %016llx sysenter_cs:  %016llx\n",
2933                 save->kernel_gs_base, save->sysenter_cs);
2934         pr_err("sysenter_esp:   %016llx sysenter_eip: %016llx\n",
2935                 save->sysenter_esp, save->sysenter_eip);
2936         pr_err("gpat:           %016llx dbgctl:       %016llx\n",
2937                 save->g_pat, save->dbgctl);
2938         pr_err("br_from:        %016llx br_to:        %016llx\n",
2939                 save->br_from, save->br_to);
2940         pr_err("excp_from:      %016llx excp_to:      %016llx\n",
2941                 save->last_excp_from, save->last_excp_to);
2942
2943 }
2944
2945 static int handle_exit(struct kvm_vcpu *vcpu)
2946 {
2947         struct vcpu_svm *svm = to_svm(vcpu);
2948         struct kvm_run *kvm_run = vcpu->run;
2949         u32 exit_code = svm->vmcb->control.exit_code;
2950
2951         trace_kvm_exit(exit_code, vcpu);
2952
2953         if (!(svm->vmcb->control.intercept_cr_write & INTERCEPT_CR0_MASK))
2954                 vcpu->arch.cr0 = svm->vmcb->save.cr0;
2955         if (npt_enabled)
2956                 vcpu->arch.cr3 = svm->vmcb->save.cr3;
2957
2958         if (unlikely(svm->nested.exit_required)) {
2959                 nested_svm_vmexit(svm);
2960                 svm->nested.exit_required = false;
2961
2962                 return 1;
2963         }
2964
2965         if (is_nested(svm)) {
2966                 int vmexit;
2967
2968                 trace_kvm_nested_vmexit(svm->vmcb->save.rip, exit_code,
2969                                         svm->vmcb->control.exit_info_1,
2970                                         svm->vmcb->control.exit_info_2,
2971                                         svm->vmcb->control.exit_int_info,
2972                                         svm->vmcb->control.exit_int_info_err);
2973
2974                 vmexit = nested_svm_exit_special(svm);
2975
2976                 if (vmexit == NESTED_EXIT_CONTINUE)
2977                         vmexit = nested_svm_exit_handled(svm);
2978
2979                 if (vmexit == NESTED_EXIT_DONE)
2980                         return 1;
2981         }
2982
2983         svm_complete_interrupts(svm);
2984
2985         if (svm->vmcb->control.exit_code == SVM_EXIT_ERR) {
2986                 kvm_run->exit_reason = KVM_EXIT_FAIL_ENTRY;
2987                 kvm_run->fail_entry.hardware_entry_failure_reason
2988                         = svm->vmcb->control.exit_code;
2989                 pr_err("KVM: FAILED VMRUN WITH VMCB:\n");
2990                 dump_vmcb(vcpu);
2991                 return 0;
2992         }
2993
2994         if (is_external_interrupt(svm->vmcb->control.exit_int_info) &&
2995             exit_code != SVM_EXIT_EXCP_BASE + PF_VECTOR &&
2996             exit_code != SVM_EXIT_NPF && exit_code != SVM_EXIT_TASK_SWITCH &&
2997             exit_code != SVM_EXIT_INTR && exit_code != SVM_EXIT_NMI)
2998                 printk(KERN_ERR "%s: unexpected exit_ini_info 0x%x "
2999                        "exit_code 0x%x\n",
3000                        __func__, svm->vmcb->control.exit_int_info,
3001                        exit_code);
3002
3003         if (exit_code >= ARRAY_SIZE(svm_exit_handlers)
3004             || !svm_exit_handlers[exit_code]) {
3005                 kvm_run->exit_reason = KVM_EXIT_UNKNOWN;
3006                 kvm_run->hw.hardware_exit_reason = exit_code;
3007                 return 0;
3008         }
3009
3010         return svm_exit_handlers[exit_code](svm);
3011 }
3012
3013 static void reload_tss(struct kvm_vcpu *vcpu)
3014 {
3015         int cpu = raw_smp_processor_id();
3016
3017         struct svm_cpu_data *sd = per_cpu(svm_data, cpu);
3018         sd->tss_desc->type = 9; /* available 32/64-bit TSS */
3019         load_TR_desc();
3020 }
3021
3022 static void pre_svm_run(struct vcpu_svm *svm)
3023 {
3024         int cpu = raw_smp_processor_id();
3025
3026         struct svm_cpu_data *sd = per_cpu(svm_data, cpu);
3027
3028         svm->vmcb->control.tlb_ctl = TLB_CONTROL_DO_NOTHING;
3029         /* FIXME: handle wraparound of asid_generation */
3030         if (svm->asid_generation != sd->asid_generation)
3031                 new_asid(svm, sd);
3032 }
3033
3034 static void svm_inject_nmi(struct kvm_vcpu *vcpu)
3035 {
3036         struct vcpu_svm *svm = to_svm(vcpu);
3037
3038         svm->vmcb->control.event_inj = SVM_EVTINJ_VALID | SVM_EVTINJ_TYPE_NMI;
3039         vcpu->arch.hflags |= HF_NMI_MASK;
3040         svm->vmcb->control.intercept |= (1ULL << INTERCEPT_IRET);
3041         ++vcpu->stat.nmi_injections;
3042 }
3043
3044 static inline void svm_inject_irq(struct vcpu_svm *svm, int irq)
3045 {
3046         struct vmcb_control_area *control;
3047
3048         control = &svm->vmcb->control;
3049         control->int_vector = irq;
3050         control->int_ctl &= ~V_INTR_PRIO_MASK;
3051         control->int_ctl |= V_IRQ_MASK |
3052                 ((/*control->int_vector >> 4*/ 0xf) << V_INTR_PRIO_SHIFT);
3053 }
3054
3055 static void svm_set_irq(struct kvm_vcpu *vcpu)
3056 {
3057         struct vcpu_svm *svm = to_svm(vcpu);
3058
3059         BUG_ON(!(gif_set(svm)));
3060
3061         trace_kvm_inj_virq(vcpu->arch.interrupt.nr);
3062         ++vcpu->stat.irq_injections;
3063
3064         svm->vmcb->control.event_inj = vcpu->arch.interrupt.nr |
3065                 SVM_EVTINJ_VALID | SVM_EVTINJ_TYPE_INTR;
3066 }
3067
3068 static void update_cr8_intercept(struct kvm_vcpu *vcpu, int tpr, int irr)
3069 {
3070         struct vcpu_svm *svm = to_svm(vcpu);
3071
3072         if (is_nested(svm) && (vcpu->arch.hflags & HF_VINTR_MASK))
3073                 return;
3074
3075         if (irr == -1)
3076                 return;
3077
3078         if (tpr >= irr)
3079                 svm->vmcb->control.intercept_cr_write |= INTERCEPT_CR8_MASK;
3080 }
3081
3082 static int svm_nmi_allowed(struct kvm_vcpu *vcpu)
3083 {
3084         struct vcpu_svm *svm = to_svm(vcpu);
3085         struct vmcb *vmcb = svm->vmcb;
3086         int ret;
3087         ret = !(vmcb->control.int_state & SVM_INTERRUPT_SHADOW_MASK) &&
3088               !(svm->vcpu.arch.hflags & HF_NMI_MASK);
3089         ret = ret && gif_set(svm) && nested_svm_nmi(svm);
3090
3091         return ret;
3092 }
3093
3094 static bool svm_get_nmi_mask(struct kvm_vcpu *vcpu)
3095 {
3096         struct vcpu_svm *svm = to_svm(vcpu);
3097
3098         return !!(svm->vcpu.arch.hflags & HF_NMI_MASK);
3099 }
3100
3101 static void svm_set_nmi_mask(struct kvm_vcpu *vcpu, bool masked)
3102 {
3103         struct vcpu_svm *svm = to_svm(vcpu);
3104
3105         if (masked) {
3106                 svm->vcpu.arch.hflags |= HF_NMI_MASK;
3107                 svm->vmcb->control.intercept |= (1ULL << INTERCEPT_IRET);
3108         } else {
3109                 svm->vcpu.arch.hflags &= ~HF_NMI_MASK;
3110                 svm->vmcb->control.intercept &= ~(1ULL << INTERCEPT_IRET);
3111         }
3112 }
3113
3114 static int svm_interrupt_allowed(struct kvm_vcpu *vcpu)
3115 {
3116         struct vcpu_svm *svm = to_svm(vcpu);
3117         struct vmcb *vmcb = svm->vmcb;
3118         int ret;
3119
3120         if (!gif_set(svm) ||
3121              (vmcb->control.int_state & SVM_INTERRUPT_SHADOW_MASK))
3122                 return 0;
3123
3124         ret = !!(vmcb->save.rflags & X86_EFLAGS_IF);
3125
3126         if (is_nested(svm))
3127                 return ret && !(svm->vcpu.arch.hflags & HF_VINTR_MASK);
3128
3129         return ret;
3130 }
3131
3132 static void enable_irq_window(struct kvm_vcpu *vcpu)
3133 {
3134         struct vcpu_svm *svm = to_svm(vcpu);
3135
3136         /*
3137          * In case GIF=0 we can't rely on the CPU to tell us when GIF becomes
3138          * 1, because that's a separate STGI/VMRUN intercept.  The next time we
3139          * get that intercept, this function will be called again though and
3140          * we'll get the vintr intercept.
3141          */
3142         if (gif_set(svm) && nested_svm_intr(svm)) {
3143                 svm_set_vintr(svm);
3144                 svm_inject_irq(svm, 0x0);
3145         }
3146 }
3147
3148 static void enable_nmi_window(struct kvm_vcpu *vcpu)
3149 {
3150         struct vcpu_svm *svm = to_svm(vcpu);
3151
3152         if ((svm->vcpu.arch.hflags & (HF_NMI_MASK | HF_IRET_MASK))
3153             == HF_NMI_MASK)
3154                 return; /* IRET will cause a vm exit */
3155
3156         /*
3157          * Something prevents NMI from been injected. Single step over possible
3158          * problem (IRET or exception injection or interrupt shadow)
3159          */
3160         svm->nmi_singlestep = true;
3161         svm->vmcb->save.rflags |= (X86_EFLAGS_TF | X86_EFLAGS_RF);
3162         update_db_intercept(vcpu);
3163 }
3164
3165 static int svm_set_tss_addr(struct kvm *kvm, unsigned int addr)
3166 {
3167         return 0;
3168 }
3169
3170 static void svm_flush_tlb(struct kvm_vcpu *vcpu)
3171 {
3172         force_new_asid(vcpu);
3173 }
3174
3175 static void svm_prepare_guest_switch(struct kvm_vcpu *vcpu)
3176 {
3177 }
3178
3179 static inline void sync_cr8_to_lapic(struct kvm_vcpu *vcpu)
3180 {
3181         struct vcpu_svm *svm = to_svm(vcpu);
3182
3183         if (is_nested(svm) && (vcpu->arch.hflags & HF_VINTR_MASK))
3184                 return;
3185
3186         if (!(svm->vmcb->control.intercept_cr_write & INTERCEPT_CR8_MASK)) {
3187                 int cr8 = svm->vmcb->control.int_ctl & V_TPR_MASK;
3188                 kvm_set_cr8(vcpu, cr8);
3189         }
3190 }
3191
3192 static inline void sync_lapic_to_cr8(struct kvm_vcpu *vcpu)
3193 {
3194         struct vcpu_svm *svm = to_svm(vcpu);
3195         u64 cr8;
3196
3197         if (is_nested(svm) && (vcpu->arch.hflags & HF_VINTR_MASK))
3198                 return;
3199
3200         cr8 = kvm_get_cr8(vcpu);
3201         svm->vmcb->control.int_ctl &= ~V_TPR_MASK;
3202         svm->vmcb->control.int_ctl |= cr8 & V_TPR_MASK;
3203 }
3204
3205 static void svm_complete_interrupts(struct vcpu_svm *svm)
3206 {
3207         u8 vector;
3208         int type;
3209         u32 exitintinfo = svm->vmcb->control.exit_int_info;
3210         unsigned int3_injected = svm->int3_injected;
3211
3212         svm->int3_injected = 0;
3213
3214         if (svm->vcpu.arch.hflags & HF_IRET_MASK) {
3215                 svm->vcpu.arch.hflags &= ~(HF_NMI_MASK | HF_IRET_MASK);
3216                 kvm_make_request(KVM_REQ_EVENT, &svm->vcpu);
3217         }
3218
3219         svm->vcpu.arch.nmi_injected = false;
3220         kvm_clear_exception_queue(&svm->vcpu);
3221         kvm_clear_interrupt_queue(&svm->vcpu);
3222
3223         if (!(exitintinfo & SVM_EXITINTINFO_VALID))
3224                 return;
3225
3226         kvm_make_request(KVM_REQ_EVENT, &svm->vcpu);
3227
3228         vector = exitintinfo & SVM_EXITINTINFO_VEC_MASK;
3229         type = exitintinfo & SVM_EXITINTINFO_TYPE_MASK;
3230
3231         switch (type) {
3232         case SVM_EXITINTINFO_TYPE_NMI:
3233                 svm->vcpu.arch.nmi_injected = true;
3234                 break;
3235         case SVM_EXITINTINFO_TYPE_EXEPT:
3236                 /*
3237                  * In case of software exceptions, do not reinject the vector,
3238                  * but re-execute the instruction instead. Rewind RIP first
3239                  * if we emulated INT3 before.
3240                  */
3241                 if (kvm_exception_is_soft(vector)) {
3242                         if (vector == BP_VECTOR && int3_injected &&
3243                             kvm_is_linear_rip(&svm->vcpu, svm->int3_rip))
3244                                 kvm_rip_write(&svm->vcpu,
3245                                               kvm_rip_read(&svm->vcpu) -
3246                                               int3_injected);
3247                         break;
3248                 }
3249                 if (exitintinfo & SVM_EXITINTINFO_VALID_ERR) {
3250                         u32 err = svm->vmcb->control.exit_int_info_err;
3251                         kvm_requeue_exception_e(&svm->vcpu, vector, err);
3252
3253                 } else
3254                         kvm_requeue_exception(&svm->vcpu, vector);
3255                 break;
3256         case SVM_EXITINTINFO_TYPE_INTR:
3257                 kvm_queue_interrupt(&svm->vcpu, vector, false);
3258                 break;
3259         default:
3260                 break;
3261         }
3262 }
3263
3264 static void svm_cancel_injection(struct kvm_vcpu *vcpu)
3265 {
3266         struct vcpu_svm *svm = to_svm(vcpu);
3267         struct vmcb_control_area *control = &svm->vmcb->control;
3268
3269         control->exit_int_info = control->event_inj;
3270         control->exit_int_info_err = control->event_inj_err;
3271         control->event_inj = 0;
3272         svm_complete_interrupts(svm);
3273 }
3274
3275 #ifdef CONFIG_X86_64
3276 #define R "r"
3277 #else
3278 #define R "e"
3279 #endif
3280
3281 static void svm_vcpu_run(struct kvm_vcpu *vcpu)
3282 {
3283         struct vcpu_svm *svm = to_svm(vcpu);
3284         u16 fs_selector;
3285         u16 gs_selector;
3286         u16 ldt_selector;
3287
3288         svm->vmcb->save.rax = vcpu->arch.regs[VCPU_REGS_RAX];
3289         svm->vmcb->save.rsp = vcpu->arch.regs[VCPU_REGS_RSP];
3290         svm->vmcb->save.rip = vcpu->arch.regs[VCPU_REGS_RIP];
3291
3292         /*
3293          * A vmexit emulation is required before the vcpu can be executed
3294          * again.
3295          */
3296         if (unlikely(svm->nested.exit_required))
3297                 return;
3298
3299         pre_svm_run(svm);
3300
3301         sync_lapic_to_cr8(vcpu);
3302
3303         save_host_msrs(vcpu);
3304         savesegment(fs, fs_selector);
3305         savesegment(gs, gs_selector);
3306         ldt_selector = kvm_read_ldt();
3307         svm->vmcb->save.cr2 = vcpu->arch.cr2;
3308
3309         clgi();
3310
3311         local_irq_enable();
3312
3313         asm volatile (
3314                 "push %%"R"bp; \n\t"
3315                 "mov %c[rbx](%[svm]), %%"R"bx \n\t"
3316                 "mov %c[rcx](%[svm]), %%"R"cx \n\t"
3317                 "mov %c[rdx](%[svm]), %%"R"dx \n\t"
3318                 "mov %c[rsi](%[svm]), %%"R"si \n\t"
3319                 "mov %c[rdi](%[svm]), %%"R"di \n\t"
3320                 "mov %c[rbp](%[svm]), %%"R"bp \n\t"
3321 #ifdef CONFIG_X86_64
3322                 "mov %c[r8](%[svm]),  %%r8  \n\t"
3323                 "mov %c[r9](%[svm]),  %%r9  \n\t"
3324                 "mov %c[r10](%[svm]), %%r10 \n\t"
3325                 "mov %c[r11](%[svm]), %%r11 \n\t"
3326                 "mov %c[r12](%[svm]), %%r12 \n\t"
3327                 "mov %c[r13](%[svm]), %%r13 \n\t"
3328                 "mov %c[r14](%[svm]), %%r14 \n\t"
3329                 "mov %c[r15](%[svm]), %%r15 \n\t"
3330 #endif
3331
3332                 /* Enter guest mode */
3333                 "push %%"R"ax \n\t"
3334                 "mov %c[vmcb](%[svm]), %%"R"ax \n\t"
3335                 __ex(SVM_VMLOAD) "\n\t"
3336                 __ex(SVM_VMRUN) "\n\t"
3337                 __ex(SVM_VMSAVE) "\n\t"
3338                 "pop %%"R"ax \n\t"
3339
3340                 /* Save guest registers, load host registers */
3341                 "mov %%"R"bx, %c[rbx](%[svm]) \n\t"
3342                 "mov %%"R"cx, %c[rcx](%[svm]) \n\t"
3343                 "mov %%"R"dx, %c[rdx](%[svm]) \n\t"
3344                 "mov %%"R"si, %c[rsi](%[svm]) \n\t"
3345                 "mov %%"R"di, %c[rdi](%[svm]) \n\t"
3346                 "mov %%"R"bp, %c[rbp](%[svm]) \n\t"
3347 #ifdef CONFIG_X86_64
3348                 "mov %%r8,  %c[r8](%[svm]) \n\t"
3349                 "mov %%r9,  %c[r9](%[svm]) \n\t"
3350                 "mov %%r10, %c[r10](%[svm]) \n\t"
3351                 "mov %%r11, %c[r11](%[svm]) \n\t"
3352                 "mov %%r12, %c[r12](%[svm]) \n\t"
3353                 "mov %%r13, %c[r13](%[svm]) \n\t"
3354                 "mov %%r14, %c[r14](%[svm]) \n\t"
3355                 "mov %%r15, %c[r15](%[svm]) \n\t"
3356 #endif
3357                 "pop %%"R"bp"
3358                 :
3359                 : [svm]"a"(svm),
3360                   [vmcb]"i"(offsetof(struct vcpu_svm, vmcb_pa)),
3361                   [rbx]"i"(offsetof(struct vcpu_svm, vcpu.arch.regs[VCPU_REGS_RBX])),
3362                   [rcx]"i"(offsetof(struct vcpu_svm, vcpu.arch.regs[VCPU_REGS_RCX])),
3363                   [rdx]"i"(offsetof(struct vcpu_svm, vcpu.arch.regs[VCPU_REGS_RDX])),
3364                   [rsi]"i"(offsetof(struct vcpu_svm, vcpu.arch.regs[VCPU_REGS_RSI])),
3365                   [rdi]"i"(offsetof(struct vcpu_svm, vcpu.arch.regs[VCPU_REGS_RDI])),
3366                   [rbp]"i"(offsetof(struct vcpu_svm, vcpu.arch.regs[VCPU_REGS_RBP]))
3367 #ifdef CONFIG_X86_64
3368                   , [r8]"i"(offsetof(struct vcpu_svm, vcpu.arch.regs[VCPU_REGS_R8])),
3369                   [r9]"i"(offsetof(struct vcpu_svm, vcpu.arch.regs[VCPU_REGS_R9])),
3370                   [r10]"i"(offsetof(struct vcpu_svm, vcpu.arch.regs[VCPU_REGS_R10])),
3371                   [r11]"i"(offsetof(struct vcpu_svm, vcpu.arch.regs[VCPU_REGS_R11])),
3372                   [r12]"i"(offsetof(struct vcpu_svm, vcpu.arch.regs[VCPU_REGS_R12])),
3373                   [r13]"i"(offsetof(struct vcpu_svm, vcpu.arch.regs[VCPU_REGS_R13])),
3374                   [r14]"i"(offsetof(struct vcpu_svm, vcpu.arch.regs[VCPU_REGS_R14])),
3375                   [r15]"i"(offsetof(struct vcpu_svm, vcpu.arch.regs[VCPU_REGS_R15]))
3376 #endif
3377                 : "cc", "memory"
3378                 , R"bx", R"cx", R"dx", R"si", R"di"
3379 #ifdef CONFIG_X86_64
3380                 , "r8", "r9", "r10", "r11" , "r12", "r13", "r14", "r15"
3381 #endif
3382                 );
3383
3384         vcpu->arch.cr2 = svm->vmcb->save.cr2;
3385         vcpu->arch.regs[VCPU_REGS_RAX] = svm->vmcb->save.rax;
3386         vcpu->arch.regs[VCPU_REGS_RSP] = svm->vmcb->save.rsp;
3387         vcpu->arch.regs[VCPU_REGS_RIP] = svm->vmcb->save.rip;
3388
3389         load_host_msrs(vcpu);
3390         loadsegment(fs, fs_selector);
3391 #ifdef CONFIG_X86_64
3392         load_gs_index(gs_selector);
3393         wrmsrl(MSR_KERNEL_GS_BASE, current->thread.gs);
3394 #else
3395         loadsegment(gs, gs_selector);
3396 #endif
3397         kvm_load_ldt(ldt_selector);
3398
3399         reload_tss(vcpu);
3400
3401         local_irq_disable();
3402
3403         stgi();
3404
3405         sync_cr8_to_lapic(vcpu);
3406
3407         svm->next_rip = 0;
3408
3409         if (npt_enabled) {
3410                 vcpu->arch.regs_avail &= ~(1 << VCPU_EXREG_PDPTR);
3411                 vcpu->arch.regs_dirty &= ~(1 << VCPU_EXREG_PDPTR);
3412         }
3413
3414         /*
3415          * We need to handle MC intercepts here before the vcpu has a chance to
3416          * change the physical cpu
3417          */
3418         if (unlikely(svm->vmcb->control.exit_code ==
3419                      SVM_EXIT_EXCP_BASE + MC_VECTOR))
3420                 svm_handle_mce(svm);
3421 }
3422
3423 #undef R
3424
3425 static void svm_set_cr3(struct kvm_vcpu *vcpu, unsigned long root)
3426 {
3427         struct vcpu_svm *svm = to_svm(vcpu);
3428
3429         svm->vmcb->save.cr3 = root;
3430         force_new_asid(vcpu);
3431 }
3432
3433 static void set_tdp_cr3(struct kvm_vcpu *vcpu, unsigned long root)
3434 {
3435         struct vcpu_svm *svm = to_svm(vcpu);
3436
3437         svm->vmcb->control.nested_cr3 = root;
3438
3439         /* Also sync guest cr3 here in case we live migrate */
3440         svm->vmcb->save.cr3 = vcpu->arch.cr3;
3441
3442         force_new_asid(vcpu);
3443 }
3444
3445 static int is_disabled(void)
3446 {
3447         u64 vm_cr;
3448
3449         rdmsrl(MSR_VM_CR, vm_cr);
3450         if (vm_cr & (1 << SVM_VM_CR_SVM_DISABLE))
3451                 return 1;
3452
3453         return 0;
3454 }
3455
3456 static void
3457 svm_patch_hypercall(struct kvm_vcpu *vcpu, unsigned char *hypercall)
3458 {
3459         /*
3460          * Patch in the VMMCALL instruction:
3461          */
3462         hypercall[0] = 0x0f;
3463         hypercall[1] = 0x01;
3464         hypercall[2] = 0xd9;
3465 }
3466
3467 static void svm_check_processor_compat(void *rtn)
3468 {
3469         *(int *)rtn = 0;
3470 }
3471
3472 static bool svm_cpu_has_accelerated_tpr(void)
3473 {
3474         return false;
3475 }
3476
3477 static u64 svm_get_mt_mask(struct kvm_vcpu *vcpu, gfn_t gfn, bool is_mmio)
3478 {
3479         return 0;
3480 }
3481
3482 static void svm_cpuid_update(struct kvm_vcpu *vcpu)
3483 {
3484 }
3485
3486 static void svm_set_supported_cpuid(u32 func, struct kvm_cpuid_entry2 *entry)
3487 {
3488         switch (func) {
3489         case 0x80000001:
3490                 if (nested)
3491                         entry->ecx |= (1 << 2); /* Set SVM bit */
3492                 break;
3493         case 0x8000000A:
3494                 entry->eax = 1; /* SVM revision 1 */
3495                 entry->ebx = 8; /* Lets support 8 ASIDs in case we add proper
3496                                    ASID emulation to nested SVM */
3497                 entry->ecx = 0; /* Reserved */
3498                 entry->edx = 0; /* Per default do not support any
3499                                    additional features */
3500
3501                 /* Support next_rip if host supports it */
3502                 if (svm_has(SVM_FEATURE_NRIP))
3503                         entry->edx |= SVM_FEATURE_NRIP;
3504
3505                 /* Support NPT for the guest if enabled */
3506                 if (npt_enabled)
3507                         entry->edx |= SVM_FEATURE_NPT;
3508
3509                 break;
3510         }
3511 }
3512
3513 static const struct trace_print_flags svm_exit_reasons_str[] = {
3514         { SVM_EXIT_READ_CR0,                    "read_cr0" },
3515         { SVM_EXIT_READ_CR3,                    "read_cr3" },
3516         { SVM_EXIT_READ_CR4,                    "read_cr4" },
3517         { SVM_EXIT_READ_CR8,                    "read_cr8" },
3518         { SVM_EXIT_WRITE_CR0,                   "write_cr0" },
3519         { SVM_EXIT_WRITE_CR3,                   "write_cr3" },
3520         { SVM_EXIT_WRITE_CR4,                   "write_cr4" },
3521         { SVM_EXIT_WRITE_CR8,                   "write_cr8" },
3522         { SVM_EXIT_READ_DR0,                    "read_dr0" },
3523         { SVM_EXIT_READ_DR1,                    "read_dr1" },
3524         { SVM_EXIT_READ_DR2,                    "read_dr2" },
3525         { SVM_EXIT_READ_DR3,                    "read_dr3" },
3526         { SVM_EXIT_WRITE_DR0,                   "write_dr0" },
3527         { SVM_EXIT_WRITE_DR1,                   "write_dr1" },
3528         { SVM_EXIT_WRITE_DR2,                   "write_dr2" },
3529         { SVM_EXIT_WRITE_DR3,                   "write_dr3" },
3530         { SVM_EXIT_WRITE_DR5,                   "write_dr5" },
3531         { SVM_EXIT_WRITE_DR7,                   "write_dr7" },
3532         { SVM_EXIT_EXCP_BASE + DB_VECTOR,       "DB excp" },
3533         { SVM_EXIT_EXCP_BASE + BP_VECTOR,       "BP excp" },
3534         { SVM_EXIT_EXCP_BASE + UD_VECTOR,       "UD excp" },
3535         { SVM_EXIT_EXCP_BASE + PF_VECTOR,       "PF excp" },
3536         { SVM_EXIT_EXCP_BASE + NM_VECTOR,       "NM excp" },
3537         { SVM_EXIT_EXCP_BASE + MC_VECTOR,       "MC excp" },
3538         { SVM_EXIT_INTR,                        "interrupt" },
3539         { SVM_EXIT_NMI,                         "nmi" },
3540         { SVM_EXIT_SMI,                         "smi" },
3541         { SVM_EXIT_INIT,                        "init" },
3542         { SVM_EXIT_VINTR,                       "vintr" },
3543         { SVM_EXIT_CPUID,                       "cpuid" },
3544         { SVM_EXIT_INVD,                        "invd" },
3545         { SVM_EXIT_HLT,                         "hlt" },
3546         { SVM_EXIT_INVLPG,                      "invlpg" },
3547         { SVM_EXIT_INVLPGA,                     "invlpga" },
3548         { SVM_EXIT_IOIO,                        "io" },
3549         { SVM_EXIT_MSR,                         "msr" },
3550         { SVM_EXIT_TASK_SWITCH,                 "task_switch" },
3551         { SVM_EXIT_SHUTDOWN,                    "shutdown" },
3552         { SVM_EXIT_VMRUN,                       "vmrun" },
3553         { SVM_EXIT_VMMCALL,                     "hypercall" },
3554         { SVM_EXIT_VMLOAD,                      "vmload" },
3555         { SVM_EXIT_VMSAVE,                      "vmsave" },
3556         { SVM_EXIT_STGI,                        "stgi" },
3557         { SVM_EXIT_CLGI,                        "clgi" },
3558         { SVM_EXIT_SKINIT,                      "skinit" },
3559         { SVM_EXIT_WBINVD,                      "wbinvd" },
3560         { SVM_EXIT_MONITOR,                     "monitor" },
3561         { SVM_EXIT_MWAIT,                       "mwait" },
3562         { SVM_EXIT_NPF,                         "npf" },
3563         { -1, NULL }
3564 };
3565
3566 static int svm_get_lpage_level(void)
3567 {
3568         return PT_PDPE_LEVEL;
3569 }
3570
3571 static bool svm_rdtscp_supported(void)
3572 {
3573         return false;
3574 }
3575
3576 static bool svm_has_wbinvd_exit(void)
3577 {
3578         return true;
3579 }
3580
3581 static void svm_fpu_deactivate(struct kvm_vcpu *vcpu)
3582 {
3583         struct vcpu_svm *svm = to_svm(vcpu);
3584
3585         svm->vmcb->control.intercept_exceptions |= 1 << NM_VECTOR;
3586         if (is_nested(svm))
3587                 svm->nested.hsave->control.intercept_exceptions |= 1 << NM_VECTOR;
3588         update_cr0_intercept(svm);
3589 }
3590
3591 static struct kvm_x86_ops svm_x86_ops = {
3592         .cpu_has_kvm_support = has_svm,
3593         .disabled_by_bios = is_disabled,
3594         .hardware_setup = svm_hardware_setup,
3595         .hardware_unsetup = svm_hardware_unsetup,
3596         .check_processor_compatibility = svm_check_processor_compat,
3597         .hardware_enable = svm_hardware_enable,
3598         .hardware_disable = svm_hardware_disable,
3599         .cpu_has_accelerated_tpr = svm_cpu_has_accelerated_tpr,
3600
3601         .vcpu_create = svm_create_vcpu,
3602         .vcpu_free = svm_free_vcpu,
3603         .vcpu_reset = svm_vcpu_reset,
3604
3605         .prepare_guest_switch = svm_prepare_guest_switch,
3606         .vcpu_load = svm_vcpu_load,
3607         .vcpu_put = svm_vcpu_put,
3608
3609         .set_guest_debug = svm_guest_debug,
3610         .get_msr = svm_get_msr,
3611         .set_msr = svm_set_msr,
3612         .get_segment_base = svm_get_segment_base,
3613         .get_segment = svm_get_segment,
3614         .set_segment = svm_set_segment,
3615         .get_cpl = svm_get_cpl,
3616         .get_cs_db_l_bits = kvm_get_cs_db_l_bits,
3617         .decache_cr0_guest_bits = svm_decache_cr0_guest_bits,
3618         .decache_cr4_guest_bits = svm_decache_cr4_guest_bits,
3619         .set_cr0 = svm_set_cr0,
3620         .set_cr3 = svm_set_cr3,
3621         .set_cr4 = svm_set_cr4,
3622         .set_efer = svm_set_efer,
3623         .get_idt = svm_get_idt,
3624         .set_idt = svm_set_idt,
3625         .get_gdt = svm_get_gdt,
3626         .set_gdt = svm_set_gdt,
3627         .set_dr7 = svm_set_dr7,
3628         .cache_reg = svm_cache_reg,
3629         .get_rflags = svm_get_rflags,
3630         .set_rflags = svm_set_rflags,
3631         .fpu_activate = svm_fpu_activate,
3632         .fpu_deactivate = svm_fpu_deactivate,
3633
3634         .tlb_flush = svm_flush_tlb,
3635
3636         .run = svm_vcpu_run,
3637         .handle_exit = handle_exit,
3638         .skip_emulated_instruction = skip_emulated_instruction,
3639         .set_interrupt_shadow = svm_set_interrupt_shadow,
3640         .get_interrupt_shadow = svm_get_interrupt_shadow,
3641         .patch_hypercall = svm_patch_hypercall,
3642         .set_irq = svm_set_irq,
3643         .set_nmi = svm_inject_nmi,
3644         .queue_exception = svm_queue_exception,
3645         .cancel_injection = svm_cancel_injection,
3646         .interrupt_allowed = svm_interrupt_allowed,
3647         .nmi_allowed = svm_nmi_allowed,
3648         .get_nmi_mask = svm_get_nmi_mask,
3649         .set_nmi_mask = svm_set_nmi_mask,
3650         .enable_nmi_window = enable_nmi_window,
3651         .enable_irq_window = enable_irq_window,
3652         .update_cr8_intercept = update_cr8_intercept,
3653
3654         .set_tss_addr = svm_set_tss_addr,
3655         .get_tdp_level = get_npt_level,
3656         .get_mt_mask = svm_get_mt_mask,
3657
3658         .exit_reasons_str = svm_exit_reasons_str,
3659         .get_lpage_level = svm_get_lpage_level,
3660
3661         .cpuid_update = svm_cpuid_update,
3662
3663         .rdtscp_supported = svm_rdtscp_supported,
3664
3665         .set_supported_cpuid = svm_set_supported_cpuid,
3666
3667         .has_wbinvd_exit = svm_has_wbinvd_exit,
3668
3669         .write_tsc_offset = svm_write_tsc_offset,
3670         .adjust_tsc_offset = svm_adjust_tsc_offset,
3671
3672         .set_tdp_cr3 = set_tdp_cr3,
3673 };
3674
3675 static int __init svm_init(void)
3676 {
3677         return kvm_init(&svm_x86_ops, sizeof(struct vcpu_svm),
3678                         __alignof__(struct vcpu_svm), THIS_MODULE);
3679 }
3680
3681 static void __exit svm_exit(void)
3682 {
3683         kvm_exit();
3684 }
3685
3686 module_init(svm_init)
3687 module_exit(svm_exit)