virt/kvm/arm/psci.c

   1 /*
   2  * Copyright (C) 2012 - ARM Ltd
   3  * Author: Marc Zyngier <marc.zyngier@arm.com>
   4  *
   5  * This program is free software; you can redistribute it and/or modify
   6  * it under the terms of the GNU General Public License version 2 as
   7  * published by the Free Software Foundation.
   8  *
   9  * This program is distributed in the hope that it will be useful,
  10  * but WITHOUT ANY WARRANTY; without even the implied warranty of
  11  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
  12  * GNU General Public License for more details.
  13  *
  14  * You should have received a copy of the GNU General Public License
  15  * along with this program.  If not, see <http://www.gnu.org/licenses/>.
  16  */
  17
  18 #include <linux/preempt.h>
  19 #include <linux/kvm_host.h>
  20 #include <linux/wait.h>
  21
  22 #include <asm/cputype.h>
  23 #include <asm/kvm_emulate.h>
  24 #include <asm/kvm_host.h>
  25
  26 #include <kvm/arm_psci.h>
  27
  28 /*
  29  * This is an implementation of the Power State Coordination Interface
  30  * as described in ARM document number ARM DEN 0022A.
  31  */
  32
  33 #define AFFINITY_MASK(level)    ~((0x1UL << ((level) * MPIDR_LEVEL_BITS)) - 1)
  34
  35 static u32 smccc_get_function(struct kvm_vcpu *vcpu)
  36 {
  37         return vcpu_get_reg(vcpu, 0);
  38 }
  39
  40 static unsigned long smccc_get_arg1(struct kvm_vcpu *vcpu)
  41 {
  42         return vcpu_get_reg(vcpu, 1);
  43 }
  44
  45 static unsigned long smccc_get_arg2(struct kvm_vcpu *vcpu)
  46 {
  47         return vcpu_get_reg(vcpu, 2);
  48 }
  49
  50 static unsigned long smccc_get_arg3(struct kvm_vcpu *vcpu)
  51 {
  52         return vcpu_get_reg(vcpu, 3);
  53 }
  54
  55 static void smccc_set_retval(struct kvm_vcpu *vcpu,
  56                              unsigned long a0,
  57                              unsigned long a1,
  58                              unsigned long a2,
  59                              unsigned long a3)
  60 {
  61         vcpu_set_reg(vcpu, 0, a0);
  62         vcpu_set_reg(vcpu, 1, a1);
  63         vcpu_set_reg(vcpu, 2, a2);
  64         vcpu_set_reg(vcpu, 3, a3);
  65 }
  66
  67 static unsigned long psci_affinity_mask(unsigned long affinity_level)
  68 {
  69         if (affinity_level <= 3)
  70                 return MPIDR_HWID_BITMASK & AFFINITY_MASK(affinity_level);
  71
  72         return 0;
  73 }
  74
  75 static unsigned long kvm_psci_vcpu_suspend(struct kvm_vcpu *vcpu)
  76 {
  77         /*
  78          * NOTE: For simplicity, we make VCPU suspend emulation to be
  79          * same-as WFI (Wait-for-interrupt) emulation.
  80          *
  81          * This means for KVM the wakeup events are interrupts and
  82          * this is consistent with intended use of StateID as described
  83          * in section 5.4.1 of PSCI v0.2 specification (ARM DEN 0022A).
  84          *
  85          * Further, we also treat power-down request to be same as
  86          * stand-by request as-per section 5.4.2 clause 3 of PSCI v0.2
  87          * specification (ARM DEN 0022A). This means all suspend states
  88          * for KVM will preserve the register state.
  89          */
  90         kvm_vcpu_block(vcpu);
  91         kvm_clear_request(KVM_REQ_UNHALT, vcpu);
  92
  93         return PSCI_RET_SUCCESS;
  94 }
  95
  96 static void kvm_psci_vcpu_off(struct kvm_vcpu *vcpu)
  97 {
  98         vcpu->arch.power_off = true;
  99         kvm_make_request(KVM_REQ_SLEEP, vcpu);
 100         kvm_vcpu_kick(vcpu);
 101 }
 102
 103 static unsigned long kvm_psci_vcpu_on(struct kvm_vcpu *source_vcpu)
 104 {
 105         struct kvm *kvm = source_vcpu->kvm;
 106         struct kvm_vcpu *vcpu = NULL;
 107         struct swait_queue_head *wq;
 108         unsigned long cpu_id;
 109         unsigned long context_id;
 110         phys_addr_t target_pc;
 111
 112         cpu_id = smccc_get_arg1(source_vcpu) & MPIDR_HWID_BITMASK;
 113         if (vcpu_mode_is_32bit(source_vcpu))
 114                 cpu_id &= ~((u32) 0);
 115
 116         vcpu = kvm_mpidr_to_vcpu(kvm, cpu_id);
 117
 118         /*
 119          * Make sure the caller requested a valid CPU and that the CPU is
 120          * turned off.
 121          */
 122         if (!vcpu)
 123                 return PSCI_RET_INVALID_PARAMS;
 124         if (!vcpu->arch.power_off) {
 125                 if (kvm_psci_version(source_vcpu) != KVM_ARM_PSCI_0_1)
 126                         return PSCI_RET_ALREADY_ON;
 127                 else
 128                         return PSCI_RET_INVALID_PARAMS;
 129         }
 130
 131         target_pc = smccc_get_arg2(source_vcpu);
 132         context_id = smccc_get_arg3(source_vcpu);
 133
 134         kvm_reset_vcpu(vcpu);
 135
 136         /* Gracefully handle Thumb2 entry point */
 137         if (vcpu_mode_is_32bit(vcpu) && (target_pc & 1)) {
 138                 target_pc &= ~((phys_addr_t) 1);
 139                 vcpu_set_thumb(vcpu);
 140         }
 141
 142         /* Propagate caller endianness */
 143         if (kvm_vcpu_is_be(source_vcpu))
 144                 kvm_vcpu_set_be(vcpu);
 145
 146         *vcpu_pc(vcpu) = target_pc;
 147         /*
 148          * NOTE: We always update r0 (or x0) because for PSCI v0.1
 149          * the general puspose registers are undefined upon CPU_ON.
 150          */
 151         smccc_set_retval(vcpu, context_id, 0, 0, 0);
 152         vcpu->arch.power_off = false;
 153         smp_mb();               /* Make sure the above is visible */
 154
 155         wq = kvm_arch_vcpu_wq(vcpu);
 156         swake_up(wq);
 157
 158         return PSCI_RET_SUCCESS;
 159 }
 160
 161 static unsigned long kvm_psci_vcpu_affinity_info(struct kvm_vcpu *vcpu)
 162 {
 163         int i, matching_cpus = 0;
 164         unsigned long mpidr;
 165         unsigned long target_affinity;
 166         unsigned long target_affinity_mask;
 167         unsigned long lowest_affinity_level;
 168         struct kvm *kvm = vcpu->kvm;
 169         struct kvm_vcpu *tmp;
 170
 171         target_affinity = smccc_get_arg1(vcpu);
 172         lowest_affinity_level = smccc_get_arg2(vcpu);
 173
 174         /* Determine target affinity mask */
 175         target_affinity_mask = psci_affinity_mask(lowest_affinity_level);
 176         if (!target_affinity_mask)
 177                 return PSCI_RET_INVALID_PARAMS;
 178
 179         /* Ignore other bits of target affinity */
 180         target_affinity &= target_affinity_mask;
 181
 182         /*
 183          * If one or more VCPU matching target affinity are running
 184          * then ON else OFF
 185          */
 186         kvm_for_each_vcpu(i, tmp, kvm) {
 187                 mpidr = kvm_vcpu_get_mpidr_aff(tmp);
 188                 if ((mpidr & target_affinity_mask) == target_affinity) {
 189                         matching_cpus++;
 190                         if (!tmp->arch.power_off)
 191                                 return PSCI_0_2_AFFINITY_LEVEL_ON;
 192                 }
 193         }
 194
 195         if (!matching_cpus)
 196                 return PSCI_RET_INVALID_PARAMS;
 197
 198         return PSCI_0_2_AFFINITY_LEVEL_OFF;
 199 }
 200
 201 static void kvm_prepare_system_event(struct kvm_vcpu *vcpu, u32 type)
 202 {
 203         int i;
 204         struct kvm_vcpu *tmp;
 205
 206         /*
 207          * The KVM ABI specifies that a system event exit may call KVM_RUN
 208          * again and may perform shutdown/reboot at a later time that when the
 209          * actual request is made.  Since we are implementing PSCI and a
 210          * caller of PSCI reboot and shutdown expects that the system shuts
 211          * down or reboots immediately, let's make sure that VCPUs are not run
 212          * after this call is handled and before the VCPUs have been
 213          * re-initialized.
 214          */
 215         kvm_for_each_vcpu(i, tmp, vcpu->kvm)
 216                 tmp->arch.power_off = true;
 217         kvm_make_all_cpus_request(vcpu->kvm, KVM_REQ_SLEEP);
 218
 219         memset(&vcpu->run->system_event, 0, sizeof(vcpu->run->system_event));
 220         vcpu->run->system_event.type = type;
 221         vcpu->run->exit_reason = KVM_EXIT_SYSTEM_EVENT;
 222 }
 223
 224 static void kvm_psci_system_off(struct kvm_vcpu *vcpu)
 225 {
 226         kvm_prepare_system_event(vcpu, KVM_SYSTEM_EVENT_SHUTDOWN);
 227 }
 228
 229 static void kvm_psci_system_reset(struct kvm_vcpu *vcpu)
 230 {
 231         kvm_prepare_system_event(vcpu, KVM_SYSTEM_EVENT_RESET);
 232 }
 233
 234 int kvm_psci_version(struct kvm_vcpu *vcpu)
 235 {
 236         if (test_bit(KVM_ARM_VCPU_PSCI_0_2, vcpu->arch.features))
 237                 return KVM_ARM_PSCI_LATEST;
 238
 239         return KVM_ARM_PSCI_0_1;
 240 }
 241
 242 static int kvm_psci_0_2_call(struct kvm_vcpu *vcpu)
 243 {
 244         struct kvm *kvm = vcpu->kvm;
 245         u32 psci_fn = smccc_get_function(vcpu);
 246         unsigned long val;
 247         int ret = 1;
 248
 249         switch (psci_fn) {
 250         case PSCI_0_2_FN_PSCI_VERSION:
 251                 /*
 252                  * Bits[31:16] = Major Version = 0
 253                  * Bits[15:0] = Minor Version = 2
 254                  */
 255                 val = KVM_ARM_PSCI_0_2;
 256                 break;
 257         case PSCI_0_2_FN_CPU_SUSPEND:
 258         case PSCI_0_2_FN64_CPU_SUSPEND:
 259                 val = kvm_psci_vcpu_suspend(vcpu);
 260                 break;
 261         case PSCI_0_2_FN_CPU_OFF:
 262                 kvm_psci_vcpu_off(vcpu);
 263                 val = PSCI_RET_SUCCESS;
 264                 break;
 265         case PSCI_0_2_FN_CPU_ON:
 266         case PSCI_0_2_FN64_CPU_ON:
 267                 mutex_lock(&kvm->lock);
 268                 val = kvm_psci_vcpu_on(vcpu);
 269                 mutex_unlock(&kvm->lock);
 270                 break;
 271         case PSCI_0_2_FN_AFFINITY_INFO:
 272         case PSCI_0_2_FN64_AFFINITY_INFO:
 273                 val = kvm_psci_vcpu_affinity_info(vcpu);
 274                 break;
 275         case PSCI_0_2_FN_MIGRATE_INFO_TYPE:
 276                 /*
 277                  * Trusted OS is MP hence does not require migration
 278                  * or
 279                  * Trusted OS is not present
 280                  */
 281                 val = PSCI_0_2_TOS_MP;
 282                 break;
 283         case PSCI_0_2_FN_SYSTEM_OFF:
 284                 kvm_psci_system_off(vcpu);
 285                 /*
 286                  * We should'nt be going back to guest VCPU after
 287                  * receiving SYSTEM_OFF request.
 288                  *
 289                  * If user space accidently/deliberately resumes
 290                  * guest VCPU after SYSTEM_OFF request then guest
 291                  * VCPU should see internal failure from PSCI return
 292                  * value. To achieve this, we preload r0 (or x0) with
 293                  * PSCI return value INTERNAL_FAILURE.
 294                  */
 295                 val = PSCI_RET_INTERNAL_FAILURE;
 296                 ret = 0;
 297                 break;
 298         case PSCI_0_2_FN_SYSTEM_RESET:
 299                 kvm_psci_system_reset(vcpu);
 300                 /*
 301                  * Same reason as SYSTEM_OFF for preloading r0 (or x0)
 302                  * with PSCI return value INTERNAL_FAILURE.
 303                  */
 304                 val = PSCI_RET_INTERNAL_FAILURE;
 305                 ret = 0;
 306                 break;
 307         default:
 308                 val = PSCI_RET_NOT_SUPPORTED;
 309                 break;
 310         }
 311
 312         smccc_set_retval(vcpu, val, 0, 0, 0);
 313         return ret;
 314 }
 315
 316 static int kvm_psci_1_0_call(struct kvm_vcpu *vcpu)
 317 {
 318         u32 psci_fn = smccc_get_function(vcpu);
 319         u32 feature;
 320         unsigned long val;
 321         int ret = 1;
 322
 323         switch(psci_fn) {
 324         case PSCI_0_2_FN_PSCI_VERSION:
 325                 val = KVM_ARM_PSCI_1_0;
 326                 break;
 327         case PSCI_1_0_FN_PSCI_FEATURES:
 328                 feature = smccc_get_arg1(vcpu);
 329                 switch(feature) {
 330                 case PSCI_0_2_FN_PSCI_VERSION:
 331                 case PSCI_0_2_FN_CPU_SUSPEND:
 332                 case PSCI_0_2_FN64_CPU_SUSPEND:
 333                 case PSCI_0_2_FN_CPU_OFF:
 334                 case PSCI_0_2_FN_CPU_ON:
 335                 case PSCI_0_2_FN64_CPU_ON:
 336                 case PSCI_0_2_FN_AFFINITY_INFO:
 337                 case PSCI_0_2_FN64_AFFINITY_INFO:
 338                 case PSCI_0_2_FN_MIGRATE_INFO_TYPE:
 339                 case PSCI_0_2_FN_SYSTEM_OFF:
 340                 case PSCI_0_2_FN_SYSTEM_RESET:
 341                 case PSCI_1_0_FN_PSCI_FEATURES:
 342                         val = 0;
 343                         break;
 344                 default:
 345                         val = PSCI_RET_NOT_SUPPORTED;
 346                         break;
 347                 }
 348                 break;
 349         default:
 350                 return kvm_psci_0_2_call(vcpu);
 351         }
 352
 353         smccc_set_retval(vcpu, val, 0, 0, 0);
 354         return ret;
 355 }
 356
 357 static int kvm_psci_0_1_call(struct kvm_vcpu *vcpu)
 358 {
 359         struct kvm *kvm = vcpu->kvm;
 360         u32 psci_fn = smccc_get_function(vcpu);
 361         unsigned long val;
 362
 363         switch (psci_fn) {
 364         case KVM_PSCI_FN_CPU_OFF:
 365                 kvm_psci_vcpu_off(vcpu);
 366                 val = PSCI_RET_SUCCESS;
 367                 break;
 368         case KVM_PSCI_FN_CPU_ON:
 369                 mutex_lock(&kvm->lock);
 370                 val = kvm_psci_vcpu_on(vcpu);
 371                 mutex_unlock(&kvm->lock);
 372                 break;
 373         default:
 374                 val = PSCI_RET_NOT_SUPPORTED;
 375                 break;
 376         }
 377
 378         smccc_set_retval(vcpu, val, 0, 0, 0);
 379         return 1;
 380 }
 381
 382 /**
 383  * kvm_psci_call - handle PSCI call if r0 value is in range
 384  * @vcpu: Pointer to the VCPU struct
 385  *
 386  * Handle PSCI calls from guests through traps from HVC instructions.
 387  * The calling convention is similar to SMC calls to the secure world
 388  * where the function number is placed in r0.
 389  *
 390  * This function returns: > 0 (success), 0 (success but exit to user
 391  * space), and < 0 (errors)
 392  *
 393  * Errors:
 394  * -EINVAL: Unrecognized PSCI function
 395  */
 396 int kvm_psci_call(struct kvm_vcpu *vcpu)
 397 {
 398         switch (kvm_psci_version(vcpu)) {
 399         case KVM_ARM_PSCI_1_0:
 400                 return kvm_psci_1_0_call(vcpu);
 401         case KVM_ARM_PSCI_0_2:
 402                 return kvm_psci_0_2_call(vcpu);
 403         case KVM_ARM_PSCI_0_1:
 404                 return kvm_psci_0_1_call(vcpu);
 405         default:
 406                 return -EINVAL;
 407         };
 408 }