virt/kvm/arm/psci.c

   1 // SPDX-License-Identifier: GPL-2.0-only
   2 /*
   3  * Copyright (C) 2012 - ARM Ltd
   4  * Author: Marc Zyngier <marc.zyngier@arm.com>
   5  */
   6
   7 #include <linux/arm-smccc.h>
   8 #include <linux/preempt.h>
   9 #include <linux/kvm_host.h>
  10 #include <linux/uaccess.h>
  11 #include <linux/wait.h>
  12
  13 #include <asm/cputype.h>
  14 #include <asm/kvm_emulate.h>
  15
  16 #include <kvm/arm_psci.h>
  17 #include <kvm/arm_hypercalls.h>
  18
  19 /*
  20  * This is an implementation of the Power State Coordination Interface
  21  * as described in ARM document number ARM DEN 0022A.
  22  */
  23
  24 #define AFFINITY_MASK(level)    ~((0x1UL << ((level) * MPIDR_LEVEL_BITS)) - 1)
  25
  26 static unsigned long psci_affinity_mask(unsigned long affinity_level)
  27 {
  28         if (affinity_level <= 3)
  29                 return MPIDR_HWID_BITMASK & AFFINITY_MASK(affinity_level);
  30
  31         return 0;
  32 }
  33
  34 static unsigned long kvm_psci_vcpu_suspend(struct kvm_vcpu *vcpu)
  35 {
  36         /*
  37          * NOTE: For simplicity, we make VCPU suspend emulation to be
  38          * same-as WFI (Wait-for-interrupt) emulation.
  39          *
  40          * This means for KVM the wakeup events are interrupts and
  41          * this is consistent with intended use of StateID as described
  42          * in section 5.4.1 of PSCI v0.2 specification (ARM DEN 0022A).
  43          *
  44          * Further, we also treat power-down request to be same as
  45          * stand-by request as-per section 5.4.2 clause 3 of PSCI v0.2
  46          * specification (ARM DEN 0022A). This means all suspend states
  47          * for KVM will preserve the register state.
  48          */
  49         kvm_vcpu_block(vcpu);
  50         kvm_clear_request(KVM_REQ_UNHALT, vcpu);
  51
  52         return PSCI_RET_SUCCESS;
  53 }
  54
  55 static void kvm_psci_vcpu_off(struct kvm_vcpu *vcpu)
  56 {
  57         vcpu->arch.power_off = true;
  58         kvm_make_request(KVM_REQ_SLEEP, vcpu);
  59         kvm_vcpu_kick(vcpu);
  60 }
  61
  62 static unsigned long kvm_psci_vcpu_on(struct kvm_vcpu *source_vcpu)
  63 {
  64         struct vcpu_reset_state *reset_state;
  65         struct kvm *kvm = source_vcpu->kvm;
  66         struct kvm_vcpu *vcpu = NULL;
  67         unsigned long cpu_id;
  68
  69         cpu_id = smccc_get_arg1(source_vcpu) & MPIDR_HWID_BITMASK;
  70         if (vcpu_mode_is_32bit(source_vcpu))
  71                 cpu_id &= ~((u32) 0);
  72
  73         vcpu = kvm_mpidr_to_vcpu(kvm, cpu_id);
  74
  75         /*
  76          * Make sure the caller requested a valid CPU and that the CPU is
  77          * turned off.
  78          */
  79         if (!vcpu)
  80                 return PSCI_RET_INVALID_PARAMS;
  81         if (!vcpu->arch.power_off) {
  82                 if (kvm_psci_version(source_vcpu, kvm) != KVM_ARM_PSCI_0_1)
  83                         return PSCI_RET_ALREADY_ON;
  84                 else
  85                         return PSCI_RET_INVALID_PARAMS;
  86         }
  87
  88         reset_state = &vcpu->arch.reset_state;
  89
  90         reset_state->pc = smccc_get_arg2(source_vcpu);
  91
  92         /* Propagate caller endianness */
  93         reset_state->be = kvm_vcpu_is_be(source_vcpu);
  94
  95         /*
  96          * NOTE: We always update r0 (or x0) because for PSCI v0.1
  97          * the general puspose registers are undefined upon CPU_ON.
  98          */
  99         reset_state->r0 = smccc_get_arg3(source_vcpu);
 100
 101         WRITE_ONCE(reset_state->reset, true);
 102         kvm_make_request(KVM_REQ_VCPU_RESET, vcpu);
 103
 104         /*
 105          * Make sure the reset request is observed if the change to
 106          * power_state is observed.
 107          */
 108         smp_wmb();
 109
 110         vcpu->arch.power_off = false;
 111         kvm_vcpu_wake_up(vcpu);
 112
 113         return PSCI_RET_SUCCESS;
 114 }
 115
 116 static unsigned long kvm_psci_vcpu_affinity_info(struct kvm_vcpu *vcpu)
 117 {
 118         int i, matching_cpus = 0;
 119         unsigned long mpidr;
 120         unsigned long target_affinity;
 121         unsigned long target_affinity_mask;
 122         unsigned long lowest_affinity_level;
 123         struct kvm *kvm = vcpu->kvm;
 124         struct kvm_vcpu *tmp;
 125
 126         target_affinity = smccc_get_arg1(vcpu);
 127         lowest_affinity_level = smccc_get_arg2(vcpu);
 128
 129         /* Determine target affinity mask */
 130         target_affinity_mask = psci_affinity_mask(lowest_affinity_level);
 131         if (!target_affinity_mask)
 132                 return PSCI_RET_INVALID_PARAMS;
 133
 134         /* Ignore other bits of target affinity */
 135         target_affinity &= target_affinity_mask;
 136
 137         /*
 138          * If one or more VCPU matching target affinity are running
 139          * then ON else OFF
 140          */
 141         kvm_for_each_vcpu(i, tmp, kvm) {
 142                 mpidr = kvm_vcpu_get_mpidr_aff(tmp);
 143                 if ((mpidr & target_affinity_mask) == target_affinity) {
 144                         matching_cpus++;
 145                         if (!tmp->arch.power_off)
 146                                 return PSCI_0_2_AFFINITY_LEVEL_ON;
 147                 }
 148         }
 149
 150         if (!matching_cpus)
 151                 return PSCI_RET_INVALID_PARAMS;
 152
 153         return PSCI_0_2_AFFINITY_LEVEL_OFF;
 154 }
 155
 156 static void kvm_prepare_system_event(struct kvm_vcpu *vcpu, u32 type)
 157 {
 158         int i;
 159         struct kvm_vcpu *tmp;
 160
 161         /*
 162          * The KVM ABI specifies that a system event exit may call KVM_RUN
 163          * again and may perform shutdown/reboot at a later time that when the
 164          * actual request is made.  Since we are implementing PSCI and a
 165          * caller of PSCI reboot and shutdown expects that the system shuts
 166          * down or reboots immediately, let's make sure that VCPUs are not run
 167          * after this call is handled and before the VCPUs have been
 168          * re-initialized.
 169          */
 170         kvm_for_each_vcpu(i, tmp, vcpu->kvm)
 171                 tmp->arch.power_off = true;
 172         kvm_make_all_cpus_request(vcpu->kvm, KVM_REQ_SLEEP);
 173
 174         memset(&vcpu->run->system_event, 0, sizeof(vcpu->run->system_event));
 175         vcpu->run->system_event.type = type;
 176         vcpu->run->exit_reason = KVM_EXIT_SYSTEM_EVENT;
 177 }
 178
 179 static void kvm_psci_system_off(struct kvm_vcpu *vcpu)
 180 {
 181         kvm_prepare_system_event(vcpu, KVM_SYSTEM_EVENT_SHUTDOWN);
 182 }
 183
 184 static void kvm_psci_system_reset(struct kvm_vcpu *vcpu)
 185 {
 186         kvm_prepare_system_event(vcpu, KVM_SYSTEM_EVENT_RESET);
 187 }
 188
 189 static int kvm_psci_0_2_call(struct kvm_vcpu *vcpu)
 190 {
 191         struct kvm *kvm = vcpu->kvm;
 192         u32 psci_fn = smccc_get_function(vcpu);
 193         unsigned long val;
 194         int ret = 1;
 195
 196         switch (psci_fn) {
 197         case PSCI_0_2_FN_PSCI_VERSION:
 198                 /*
 199                  * Bits[31:16] = Major Version = 0
 200                  * Bits[15:0] = Minor Version = 2
 201                  */
 202                 val = KVM_ARM_PSCI_0_2;
 203                 break;
 204         case PSCI_0_2_FN_CPU_SUSPEND:
 205         case PSCI_0_2_FN64_CPU_SUSPEND:
 206                 val = kvm_psci_vcpu_suspend(vcpu);
 207                 break;
 208         case PSCI_0_2_FN_CPU_OFF:
 209                 kvm_psci_vcpu_off(vcpu);
 210                 val = PSCI_RET_SUCCESS;
 211                 break;
 212         case PSCI_0_2_FN_CPU_ON:
 213         case PSCI_0_2_FN64_CPU_ON:
 214                 mutex_lock(&kvm->lock);
 215                 val = kvm_psci_vcpu_on(vcpu);
 216                 mutex_unlock(&kvm->lock);
 217                 break;
 218         case PSCI_0_2_FN_AFFINITY_INFO:
 219         case PSCI_0_2_FN64_AFFINITY_INFO:
 220                 val = kvm_psci_vcpu_affinity_info(vcpu);
 221                 break;
 222         case PSCI_0_2_FN_MIGRATE_INFO_TYPE:
 223                 /*
 224                  * Trusted OS is MP hence does not require migration
 225                  * or
 226                  * Trusted OS is not present
 227                  */
 228                 val = PSCI_0_2_TOS_MP;
 229                 break;
 230         case PSCI_0_2_FN_SYSTEM_OFF:
 231                 kvm_psci_system_off(vcpu);
 232                 /*
 233                  * We should'nt be going back to guest VCPU after
 234                  * receiving SYSTEM_OFF request.
 235                  *
 236                  * If user space accidently/deliberately resumes
 237                  * guest VCPU after SYSTEM_OFF request then guest
 238                  * VCPU should see internal failure from PSCI return
 239                  * value. To achieve this, we preload r0 (or x0) with
 240                  * PSCI return value INTERNAL_FAILURE.
 241                  */
 242                 val = PSCI_RET_INTERNAL_FAILURE;
 243                 ret = 0;
 244                 break;
 245         case PSCI_0_2_FN_SYSTEM_RESET:
 246                 kvm_psci_system_reset(vcpu);
 247                 /*
 248                  * Same reason as SYSTEM_OFF for preloading r0 (or x0)
 249                  * with PSCI return value INTERNAL_FAILURE.
 250                  */
 251                 val = PSCI_RET_INTERNAL_FAILURE;
 252                 ret = 0;
 253                 break;
 254         default:
 255                 val = PSCI_RET_NOT_SUPPORTED;
 256                 break;
 257         }
 258
 259         smccc_set_retval(vcpu, val, 0, 0, 0);
 260         return ret;
 261 }
 262
 263 static int kvm_psci_1_0_call(struct kvm_vcpu *vcpu)
 264 {
 265         u32 psci_fn = smccc_get_function(vcpu);
 266         u32 feature;
 267         unsigned long val;
 268         int ret = 1;
 269
 270         switch(psci_fn) {
 271         case PSCI_0_2_FN_PSCI_VERSION:
 272                 val = KVM_ARM_PSCI_1_0;
 273                 break;
 274         case PSCI_1_0_FN_PSCI_FEATURES:
 275                 feature = smccc_get_arg1(vcpu);
 276                 switch(feature) {
 277                 case PSCI_0_2_FN_PSCI_VERSION:
 278                 case PSCI_0_2_FN_CPU_SUSPEND:
 279                 case PSCI_0_2_FN64_CPU_SUSPEND:
 280                 case PSCI_0_2_FN_CPU_OFF:
 281                 case PSCI_0_2_FN_CPU_ON:
 282                 case PSCI_0_2_FN64_CPU_ON:
 283                 case PSCI_0_2_FN_AFFINITY_INFO:
 284                 case PSCI_0_2_FN64_AFFINITY_INFO:
 285                 case PSCI_0_2_FN_MIGRATE_INFO_TYPE:
 286                 case PSCI_0_2_FN_SYSTEM_OFF:
 287                 case PSCI_0_2_FN_SYSTEM_RESET:
 288                 case PSCI_1_0_FN_PSCI_FEATURES:
 289                 case ARM_SMCCC_VERSION_FUNC_ID:
 290                         val = 0;
 291                         break;
 292                 default:
 293                         val = PSCI_RET_NOT_SUPPORTED;
 294                         break;
 295                 }
 296                 break;
 297         default:
 298                 return kvm_psci_0_2_call(vcpu);
 299         }
 300
 301         smccc_set_retval(vcpu, val, 0, 0, 0);
 302         return ret;
 303 }
 304
 305 static int kvm_psci_0_1_call(struct kvm_vcpu *vcpu)
 306 {
 307         struct kvm *kvm = vcpu->kvm;
 308         u32 psci_fn = smccc_get_function(vcpu);
 309         unsigned long val;
 310
 311         switch (psci_fn) {
 312         case KVM_PSCI_FN_CPU_OFF:
 313                 kvm_psci_vcpu_off(vcpu);
 314                 val = PSCI_RET_SUCCESS;
 315                 break;
 316         case KVM_PSCI_FN_CPU_ON:
 317                 mutex_lock(&kvm->lock);
 318                 val = kvm_psci_vcpu_on(vcpu);
 319                 mutex_unlock(&kvm->lock);
 320                 break;
 321         default:
 322                 val = PSCI_RET_NOT_SUPPORTED;
 323                 break;
 324         }
 325
 326         smccc_set_retval(vcpu, val, 0, 0, 0);
 327         return 1;
 328 }
 329
 330 /**
 331  * kvm_psci_call - handle PSCI call if r0 value is in range
 332  * @vcpu: Pointer to the VCPU struct
 333  *
 334  * Handle PSCI calls from guests through traps from HVC instructions.
 335  * The calling convention is similar to SMC calls to the secure world
 336  * where the function number is placed in r0.
 337  *
 338  * This function returns: > 0 (success), 0 (success but exit to user
 339  * space), and < 0 (errors)
 340  *
 341  * Errors:
 342  * -EINVAL: Unrecognized PSCI function
 343  */
 344 int kvm_psci_call(struct kvm_vcpu *vcpu)
 345 {
 346         switch (kvm_psci_version(vcpu, vcpu->kvm)) {
 347         case KVM_ARM_PSCI_1_0:
 348                 return kvm_psci_1_0_call(vcpu);
 349         case KVM_ARM_PSCI_0_2:
 350                 return kvm_psci_0_2_call(vcpu);
 351         case KVM_ARM_PSCI_0_1:
 352                 return kvm_psci_0_1_call(vcpu);
 353         default:
 354                 return -EINVAL;
 355         };
 356 }
 357
 358 int kvm_arm_get_fw_num_regs(struct kvm_vcpu *vcpu)
 359 {
 360         return 3;               /* PSCI version and two workaround registers */
 361 }
 362
 363 int kvm_arm_copy_fw_reg_indices(struct kvm_vcpu *vcpu, u64 __user *uindices)
 364 {
 365         if (put_user(KVM_REG_ARM_PSCI_VERSION, uindices++))
 366                 return -EFAULT;
 367
 368         if (put_user(KVM_REG_ARM_SMCCC_ARCH_WORKAROUND_1, uindices++))
 369                 return -EFAULT;
 370
 371         if (put_user(KVM_REG_ARM_SMCCC_ARCH_WORKAROUND_2, uindices++))
 372                 return -EFAULT;
 373
 374         return 0;
 375 }
 376
 377 #define KVM_REG_FEATURE_LEVEL_WIDTH     4
 378 #define KVM_REG_FEATURE_LEVEL_MASK      (BIT(KVM_REG_FEATURE_LEVEL_WIDTH) - 1)
 379
 380 /*
 381  * Convert the workaround level into an easy-to-compare number, where higher
 382  * values mean better protection.
 383  */
 384 static int get_kernel_wa_level(u64 regid)
 385 {
 386         switch (regid) {
 387         case KVM_REG_ARM_SMCCC_ARCH_WORKAROUND_1:
 388                 switch (kvm_arm_harden_branch_predictor()) {
 389                 case KVM_BP_HARDEN_UNKNOWN:
 390                         return KVM_REG_ARM_SMCCC_ARCH_WORKAROUND_1_NOT_AVAIL;
 391                 case KVM_BP_HARDEN_WA_NEEDED:
 392                         return KVM_REG_ARM_SMCCC_ARCH_WORKAROUND_1_AVAIL;
 393                 case KVM_BP_HARDEN_NOT_REQUIRED:
 394                         return KVM_REG_ARM_SMCCC_ARCH_WORKAROUND_1_NOT_REQUIRED;
 395                 }
 396                 return KVM_REG_ARM_SMCCC_ARCH_WORKAROUND_1_NOT_AVAIL;
 397         case KVM_REG_ARM_SMCCC_ARCH_WORKAROUND_2:
 398                 switch (kvm_arm_have_ssbd()) {
 399                 case KVM_SSBD_FORCE_DISABLE:
 400                         return KVM_REG_ARM_SMCCC_ARCH_WORKAROUND_2_NOT_AVAIL;
 401                 case KVM_SSBD_KERNEL:
 402                         return KVM_REG_ARM_SMCCC_ARCH_WORKAROUND_2_AVAIL;
 403                 case KVM_SSBD_FORCE_ENABLE:
 404                 case KVM_SSBD_MITIGATED:
 405                         return KVM_REG_ARM_SMCCC_ARCH_WORKAROUND_2_NOT_REQUIRED;
 406                 case KVM_SSBD_UNKNOWN:
 407                 default:
 408                         return KVM_REG_ARM_SMCCC_ARCH_WORKAROUND_2_UNKNOWN;
 409                 }
 410         }
 411
 412         return -EINVAL;
 413 }
 414
 415 int kvm_arm_get_fw_reg(struct kvm_vcpu *vcpu, const struct kvm_one_reg *reg)
 416 {
 417         void __user *uaddr = (void __user *)(long)reg->addr;
 418         u64 val;
 419
 420         switch (reg->id) {
 421         case KVM_REG_ARM_PSCI_VERSION:
 422                 val = kvm_psci_version(vcpu, vcpu->kvm);
 423                 break;
 424         case KVM_REG_ARM_SMCCC_ARCH_WORKAROUND_1:
 425                 val = get_kernel_wa_level(reg->id) & KVM_REG_FEATURE_LEVEL_MASK;
 426                 break;
 427         case KVM_REG_ARM_SMCCC_ARCH_WORKAROUND_2:
 428                 val = get_kernel_wa_level(reg->id) & KVM_REG_FEATURE_LEVEL_MASK;
 429
 430                 if (val == KVM_REG_ARM_SMCCC_ARCH_WORKAROUND_2_AVAIL &&
 431                     kvm_arm_get_vcpu_workaround_2_flag(vcpu))
 432                         val |= KVM_REG_ARM_SMCCC_ARCH_WORKAROUND_2_ENABLED;
 433                 break;
 434         default:
 435                 return -ENOENT;
 436         }
 437
 438         if (copy_to_user(uaddr, &val, KVM_REG_SIZE(reg->id)))
 439                 return -EFAULT;
 440
 441         return 0;
 442 }
 443
 444 int kvm_arm_set_fw_reg(struct kvm_vcpu *vcpu, const struct kvm_one_reg *reg)
 445 {
 446         void __user *uaddr = (void __user *)(long)reg->addr;
 447         u64 val;
 448         int wa_level;
 449
 450         if (copy_from_user(&val, uaddr, KVM_REG_SIZE(reg->id)))
 451                 return -EFAULT;
 452
 453         switch (reg->id) {
 454         case KVM_REG_ARM_PSCI_VERSION:
 455         {
 456                 bool wants_02;
 457
 458                 wants_02 = test_bit(KVM_ARM_VCPU_PSCI_0_2, vcpu->arch.features);
 459
 460                 switch (val) {
 461                 case KVM_ARM_PSCI_0_1:
 462                         if (wants_02)
 463                                 return -EINVAL;
 464                         vcpu->kvm->arch.psci_version = val;
 465                         return 0;
 466                 case KVM_ARM_PSCI_0_2:
 467                 case KVM_ARM_PSCI_1_0:
 468                         if (!wants_02)
 469                                 return -EINVAL;
 470                         vcpu->kvm->arch.psci_version = val;
 471                         return 0;
 472                 }
 473                 break;
 474         }
 475
 476         case KVM_REG_ARM_SMCCC_ARCH_WORKAROUND_1:
 477                 if (val & ~KVM_REG_FEATURE_LEVEL_MASK)
 478                         return -EINVAL;
 479
 480                 if (get_kernel_wa_level(reg->id) < val)
 481                         return -EINVAL;
 482
 483                 return 0;
 484
 485         case KVM_REG_ARM_SMCCC_ARCH_WORKAROUND_2:
 486                 if (val & ~(KVM_REG_FEATURE_LEVEL_MASK |
 487                             KVM_REG_ARM_SMCCC_ARCH_WORKAROUND_2_ENABLED))
 488                         return -EINVAL;
 489
 490                 wa_level = val & KVM_REG_FEATURE_LEVEL_MASK;
 491
 492                 if (get_kernel_wa_level(reg->id) < wa_level)
 493                         return -EINVAL;
 494
 495                 /* The enabled bit must not be set unless the level is AVAIL. */
 496                 if (wa_level != KVM_REG_ARM_SMCCC_ARCH_WORKAROUND_2_AVAIL &&
 497                     wa_level != val)
 498                         return -EINVAL;
 499
 500                 /* Are we finished or do we need to check the enable bit ? */
 501                 if (kvm_arm_have_ssbd() != KVM_SSBD_KERNEL)
 502                         return 0;
 503
 504                 /*
 505                  * If this kernel supports the workaround to be switched on
 506                  * or off, make sure it matches the requested setting.
 507                  */
 508                 switch (wa_level) {
 509                 case KVM_REG_ARM_SMCCC_ARCH_WORKAROUND_2_AVAIL:
 510                         kvm_arm_set_vcpu_workaround_2_flag(vcpu,
 511                             val & KVM_REG_ARM_SMCCC_ARCH_WORKAROUND_2_ENABLED);
 512                         break;
 513                 case KVM_REG_ARM_SMCCC_ARCH_WORKAROUND_2_NOT_REQUIRED:
 514                         kvm_arm_set_vcpu_workaround_2_flag(vcpu, true);
 515                         break;
 516                 }
 517
 518                 return 0;
 519         default:
 520                 return -ENOENT;
 521         }
 522
 523         return -EINVAL;
 524 }