kernel/arch/x86/kvm/hyperv.c

   1 /*
   2  * KVM Microsoft Hyper-V emulation
   3  *
   4  * derived from arch/x86/kvm/x86.c
   5  *
   6  * Copyright (C) 2006 Qumranet, Inc.
   7  * Copyright (C) 2008 Qumranet, Inc.
   8  * Copyright IBM Corporation, 2008
   9  * Copyright 2010 Red Hat, Inc. and/or its affiliates.
  10  * Copyright (C) 2015 Andrey Smetanin <asmetanin@virtuozzo.com>
  11  *
  12  * Authors:
  13  *   Avi Kivity   <avi@qumranet.com>
  14  *   Yaniv Kamay  <yaniv@qumranet.com>
  15  *   Amit Shah    <amit.shah@qumranet.com>
  16  *   Ben-Ami Yassour <benami@il.ibm.com>
  17  *   Andrey Smetanin <asmetanin@virtuozzo.com>
  18  *
  19  * This work is licensed under the terms of the GNU GPL, version 2.  See
  20  * the COPYING file in the top-level directory.
  21  *
  22  */
  23
  24 #include "x86.h"
  25 #include "lapic.h"
  26 #include "hyperv.h"
  27
  28 #include <linux/kvm_host.h>
  29 #include <trace/events/kvm.h>
  30
  31 #include "trace.h"
  32
  33 static bool kvm_hv_msr_partition_wide(u32 msr)
  34 {
  35         bool r = false;
  36
  37         switch (msr) {
  38         case HV_X64_MSR_GUEST_OS_ID:
  39         case HV_X64_MSR_HYPERCALL:
  40         case HV_X64_MSR_REFERENCE_TSC:
  41         case HV_X64_MSR_TIME_REF_COUNT:
  42         case HV_X64_MSR_CRASH_CTL:
  43         case HV_X64_MSR_CRASH_P0 ... HV_X64_MSR_CRASH_P4:
  44         case HV_X64_MSR_RESET:
  45                 r = true;
  46                 break;
  47         }
  48
  49         return r;
  50 }
  51
  52 static int kvm_hv_msr_get_crash_data(struct kvm_vcpu *vcpu,
  53                                      u32 index, u64 *pdata)
  54 {
  55         struct kvm_hv *hv = &vcpu->kvm->arch.hyperv;
  56
  57         if (WARN_ON_ONCE(index >= ARRAY_SIZE(hv->hv_crash_param)))
  58                 return -EINVAL;
  59
  60         *pdata = hv->hv_crash_param[index];
  61         return 0;
  62 }
  63
  64 static int kvm_hv_msr_get_crash_ctl(struct kvm_vcpu *vcpu, u64 *pdata)
  65 {
  66         struct kvm_hv *hv = &vcpu->kvm->arch.hyperv;
  67
  68         *pdata = hv->hv_crash_ctl;
  69         return 0;
  70 }
  71
  72 static int kvm_hv_msr_set_crash_ctl(struct kvm_vcpu *vcpu, u64 data, bool host)
  73 {
  74         struct kvm_hv *hv = &vcpu->kvm->arch.hyperv;
  75
  76         if (host)
  77                 hv->hv_crash_ctl = data & HV_X64_MSR_CRASH_CTL_NOTIFY;
  78
  79         if (!host && (data & HV_X64_MSR_CRASH_CTL_NOTIFY)) {
  80
  81                 vcpu_debug(vcpu, "hv crash (0x%llx 0x%llx 0x%llx 0x%llx 0x%llx)\n",
  82                           hv->hv_crash_param[0],
  83                           hv->hv_crash_param[1],
  84                           hv->hv_crash_param[2],
  85                           hv->hv_crash_param[3],
  86                           hv->hv_crash_param[4]);
  87
  88                 /* Send notification about crash to user space */
  89                 kvm_make_request(KVM_REQ_HV_CRASH, vcpu);
  90         }
  91
  92         return 0;
  93 }
  94
  95 static int kvm_hv_msr_set_crash_data(struct kvm_vcpu *vcpu,
  96                                      u32 index, u64 data)
  97 {
  98         struct kvm_hv *hv = &vcpu->kvm->arch.hyperv;
  99
 100         if (WARN_ON_ONCE(index >= ARRAY_SIZE(hv->hv_crash_param)))
 101                 return -EINVAL;
 102
 103         hv->hv_crash_param[index] = data;
 104         return 0;
 105 }
 106
 107 static int kvm_hv_set_msr_pw(struct kvm_vcpu *vcpu, u32 msr, u64 data,
 108                              bool host)
 109 {
 110         struct kvm *kvm = vcpu->kvm;
 111         struct kvm_hv *hv = &kvm->arch.hyperv;
 112
 113         switch (msr) {
 114         case HV_X64_MSR_GUEST_OS_ID:
 115                 hv->hv_guest_os_id = data;
 116                 /* setting guest os id to zero disables hypercall page */
 117                 if (!hv->hv_guest_os_id)
 118                         hv->hv_hypercall &= ~HV_X64_MSR_HYPERCALL_ENABLE;
 119                 break;
 120         case HV_X64_MSR_HYPERCALL: {
 121                 u64 gfn;
 122                 unsigned long addr;
 123                 u8 instructions[4];
 124
 125                 /* if guest os id is not set hypercall should remain disabled */
 126                 if (!hv->hv_guest_os_id)
 127                         break;
 128                 if (!(data & HV_X64_MSR_HYPERCALL_ENABLE)) {
 129                         hv->hv_hypercall = data;
 130                         break;
 131                 }
 132                 gfn = data >> HV_X64_MSR_HYPERCALL_PAGE_ADDRESS_SHIFT;
 133                 addr = gfn_to_hva(kvm, gfn);
 134                 if (kvm_is_error_hva(addr))
 135                         return 1;
 136                 kvm_x86_ops->patch_hypercall(vcpu, instructions);
 137                 ((unsigned char *)instructions)[3] = 0xc3; /* ret */
 138                 if (__copy_to_user((void __user *)addr, instructions, 4))
 139                         return 1;
 140                 hv->hv_hypercall = data;
 141                 mark_page_dirty(kvm, gfn);
 142                 break;
 143         }
 144         case HV_X64_MSR_REFERENCE_TSC: {
 145                 u64 gfn;
 146                 HV_REFERENCE_TSC_PAGE tsc_ref;
 147
 148                 memset(&tsc_ref, 0, sizeof(tsc_ref));
 149                 hv->hv_tsc_page = data;
 150                 if (!(data & HV_X64_MSR_TSC_REFERENCE_ENABLE))
 151                         break;
 152                 gfn = data >> HV_X64_MSR_TSC_REFERENCE_ADDRESS_SHIFT;
 153                 if (kvm_write_guest(
 154                                 kvm,
 155                                 gfn << HV_X64_MSR_TSC_REFERENCE_ADDRESS_SHIFT,
 156                                 &tsc_ref, sizeof(tsc_ref)))
 157                         return 1;
 158                 mark_page_dirty(kvm, gfn);
 159                 break;
 160         }
 161         case HV_X64_MSR_CRASH_P0 ... HV_X64_MSR_CRASH_P4:
 162                 return kvm_hv_msr_set_crash_data(vcpu,
 163                                                  msr - HV_X64_MSR_CRASH_P0,
 164                                                  data);
 165         case HV_X64_MSR_CRASH_CTL:
 166                 return kvm_hv_msr_set_crash_ctl(vcpu, data, host);
 167         case HV_X64_MSR_RESET:
 168                 if (data == 1) {
 169                         vcpu_debug(vcpu, "hyper-v reset requested\n");
 170                         kvm_make_request(KVM_REQ_HV_RESET, vcpu);
 171                 }
 172                 break;
 173         default:
 174                 vcpu_unimpl(vcpu, "Hyper-V uhandled wrmsr: 0x%x data 0x%llx\n",
 175                             msr, data);
 176                 return 1;
 177         }
 178         return 0;
 179 }
 180
 181 /* Calculate cpu time spent by current task in 100ns units */
 182 static u64 current_task_runtime_100ns(void)
 183 {
 184         cputime_t utime, stime;
 185
 186         task_cputime_adjusted(current, &utime, &stime);
 187         return div_u64(cputime_to_nsecs(utime + stime), 100);
 188 }
 189
 190 static int kvm_hv_set_msr(struct kvm_vcpu *vcpu, u32 msr, u64 data, bool host)
 191 {
 192         struct kvm_vcpu_hv *hv = &vcpu->arch.hyperv;
 193
 194         switch (msr) {
 195         case HV_X64_MSR_APIC_ASSIST_PAGE: {
 196                 u64 gfn;
 197                 unsigned long addr;
 198
 199                 if (!(data & HV_X64_MSR_APIC_ASSIST_PAGE_ENABLE)) {
 200                         hv->hv_vapic = data;
 201                         if (kvm_lapic_enable_pv_eoi(vcpu, 0))
 202                                 return 1;
 203                         break;
 204                 }
 205                 gfn = data >> HV_X64_MSR_APIC_ASSIST_PAGE_ADDRESS_SHIFT;
 206                 addr = kvm_vcpu_gfn_to_hva(vcpu, gfn);
 207                 if (kvm_is_error_hva(addr))
 208                         return 1;
 209                 if (__clear_user((void __user *)addr, PAGE_SIZE))
 210                         return 1;
 211                 hv->hv_vapic = data;
 212                 kvm_vcpu_mark_page_dirty(vcpu, gfn);
 213                 if (kvm_lapic_enable_pv_eoi(vcpu,
 214                                             gfn_to_gpa(gfn) | KVM_MSR_ENABLED))
 215                         return 1;
 216                 break;
 217         }
 218         case HV_X64_MSR_EOI:
 219                 return kvm_hv_vapic_msr_write(vcpu, APIC_EOI, data);
 220         case HV_X64_MSR_ICR:
 221                 return kvm_hv_vapic_msr_write(vcpu, APIC_ICR, data);
 222         case HV_X64_MSR_TPR:
 223                 return kvm_hv_vapic_msr_write(vcpu, APIC_TASKPRI, data);
 224         case HV_X64_MSR_VP_RUNTIME:
 225                 if (!host)
 226                         return 1;
 227                 hv->runtime_offset = data - current_task_runtime_100ns();
 228                 break;
 229         default:
 230                 vcpu_unimpl(vcpu, "Hyper-V uhandled wrmsr: 0x%x data 0x%llx\n",
 231                             msr, data);
 232                 return 1;
 233         }
 234
 235         return 0;
 236 }
 237
 238 static int kvm_hv_get_msr_pw(struct kvm_vcpu *vcpu, u32 msr, u64 *pdata)
 239 {
 240         u64 data = 0;
 241         struct kvm *kvm = vcpu->kvm;
 242         struct kvm_hv *hv = &kvm->arch.hyperv;
 243
 244         switch (msr) {
 245         case HV_X64_MSR_GUEST_OS_ID:
 246                 data = hv->hv_guest_os_id;
 247                 break;
 248         case HV_X64_MSR_HYPERCALL:
 249                 data = hv->hv_hypercall;
 250                 break;
 251         case HV_X64_MSR_TIME_REF_COUNT: {
 252                 data =
 253                      div_u64(get_kernel_ns() + kvm->arch.kvmclock_offset, 100);
 254                 break;
 255         }
 256         case HV_X64_MSR_REFERENCE_TSC:
 257                 data = hv->hv_tsc_page;
 258                 break;
 259         case HV_X64_MSR_CRASH_P0 ... HV_X64_MSR_CRASH_P4:
 260                 return kvm_hv_msr_get_crash_data(vcpu,
 261                                                  msr - HV_X64_MSR_CRASH_P0,
 262                                                  pdata);
 263         case HV_X64_MSR_CRASH_CTL:
 264                 return kvm_hv_msr_get_crash_ctl(vcpu, pdata);
 265         case HV_X64_MSR_RESET:
 266                 data = 0;
 267                 break;
 268         default:
 269                 vcpu_unimpl(vcpu, "Hyper-V unhandled rdmsr: 0x%x\n", msr);
 270                 return 1;
 271         }
 272
 273         *pdata = data;
 274         return 0;
 275 }
 276
 277 static int kvm_hv_get_msr(struct kvm_vcpu *vcpu, u32 msr, u64 *pdata)
 278 {
 279         u64 data = 0;
 280         struct kvm_vcpu_hv *hv = &vcpu->arch.hyperv;
 281
 282         switch (msr) {
 283         case HV_X64_MSR_VP_INDEX: {
 284                 int r;
 285                 struct kvm_vcpu *v;
 286
 287                 kvm_for_each_vcpu(r, v, vcpu->kvm) {
 288                         if (v == vcpu) {
 289                                 data = r;
 290                                 break;
 291                         }
 292                 }
 293                 break;
 294         }
 295         case HV_X64_MSR_EOI:
 296                 return kvm_hv_vapic_msr_read(vcpu, APIC_EOI, pdata);
 297         case HV_X64_MSR_ICR:
 298                 return kvm_hv_vapic_msr_read(vcpu, APIC_ICR, pdata);
 299         case HV_X64_MSR_TPR:
 300                 return kvm_hv_vapic_msr_read(vcpu, APIC_TASKPRI, pdata);
 301         case HV_X64_MSR_APIC_ASSIST_PAGE:
 302                 data = hv->hv_vapic;
 303                 break;
 304         case HV_X64_MSR_VP_RUNTIME:
 305                 data = current_task_runtime_100ns() + hv->runtime_offset;
 306                 break;
 307         default:
 308                 vcpu_unimpl(vcpu, "Hyper-V unhandled rdmsr: 0x%x\n", msr);
 309                 return 1;
 310         }
 311         *pdata = data;
 312         return 0;
 313 }
 314
 315 int kvm_hv_set_msr_common(struct kvm_vcpu *vcpu, u32 msr, u64 data, bool host)
 316 {
 317         if (kvm_hv_msr_partition_wide(msr)) {
 318                 int r;
 319
 320                 mutex_lock(&vcpu->kvm->lock);
 321                 r = kvm_hv_set_msr_pw(vcpu, msr, data, host);
 322                 mutex_unlock(&vcpu->kvm->lock);
 323                 return r;
 324         } else
 325                 return kvm_hv_set_msr(vcpu, msr, data, host);
 326 }
 327
 328 int kvm_hv_get_msr_common(struct kvm_vcpu *vcpu, u32 msr, u64 *pdata)
 329 {
 330         if (kvm_hv_msr_partition_wide(msr)) {
 331                 int r;
 332
 333                 mutex_lock(&vcpu->kvm->lock);
 334                 r = kvm_hv_get_msr_pw(vcpu, msr, pdata);
 335                 mutex_unlock(&vcpu->kvm->lock);
 336                 return r;
 337         } else
 338                 return kvm_hv_get_msr(vcpu, msr, pdata);
 339 }
 340
 341 bool kvm_hv_hypercall_enabled(struct kvm *kvm)
 342 {
 343         return kvm->arch.hyperv.hv_hypercall & HV_X64_MSR_HYPERCALL_ENABLE;
 344 }
 345
 346 int kvm_hv_hypercall(struct kvm_vcpu *vcpu)
 347 {
 348         u64 param, ingpa, outgpa, ret;
 349         uint16_t code, rep_idx, rep_cnt, res = HV_STATUS_SUCCESS, rep_done = 0;
 350         bool fast, longmode;
 351
 352         /*
 353          * hypercall generates UD from non zero cpl and real mode
 354          * per HYPER-V spec
 355          */
 356         if (kvm_x86_ops->get_cpl(vcpu) != 0 || !is_protmode(vcpu)) {
 357                 kvm_queue_exception(vcpu, UD_VECTOR);
 358                 return 0;
 359         }
 360
 361         longmode = is_64_bit_mode(vcpu);
 362
 363         if (!longmode) {
 364                 param = ((u64)kvm_register_read(vcpu, VCPU_REGS_RDX) << 32) |
 365                         (kvm_register_read(vcpu, VCPU_REGS_RAX) & 0xffffffff);
 366                 ingpa = ((u64)kvm_register_read(vcpu, VCPU_REGS_RBX) << 32) |
 367                         (kvm_register_read(vcpu, VCPU_REGS_RCX) & 0xffffffff);
 368                 outgpa = ((u64)kvm_register_read(vcpu, VCPU_REGS_RDI) << 32) |
 369                         (kvm_register_read(vcpu, VCPU_REGS_RSI) & 0xffffffff);
 370         }
 371 #ifdef CONFIG_X86_64
 372         else {
 373                 param = kvm_register_read(vcpu, VCPU_REGS_RCX);
 374                 ingpa = kvm_register_read(vcpu, VCPU_REGS_RDX);
 375                 outgpa = kvm_register_read(vcpu, VCPU_REGS_R8);
 376         }
 377 #endif
 378
 379         code = param & 0xffff;
 380         fast = (param >> 16) & 0x1;
 381         rep_cnt = (param >> 32) & 0xfff;
 382         rep_idx = (param >> 48) & 0xfff;
 383
 384         trace_kvm_hv_hypercall(code, fast, rep_cnt, rep_idx, ingpa, outgpa);
 385
 386         switch (code) {
 387         case HV_X64_HV_NOTIFY_LONG_SPIN_WAIT:
 388                 kvm_vcpu_on_spin(vcpu);
 389                 break;
 390         default:
 391                 res = HV_STATUS_INVALID_HYPERCALL_CODE;
 392                 break;
 393         }
 394
 395         ret = res | (((u64)rep_done & 0xfff) << 32);
 396         if (longmode) {
 397                 kvm_register_write(vcpu, VCPU_REGS_RAX, ret);
 398         } else {
 399                 kvm_register_write(vcpu, VCPU_REGS_RDX, ret >> 32);
 400                 kvm_register_write(vcpu, VCPU_REGS_RAX, ret & 0xffffffff);
 401         }
 402
 403         return 1;
 404 }