arch/arm/kvm/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_psci.h>
  25 #include <asm/kvm_host.h>
  26
  27 /*
  28  * This is an implementation of the Power State Coordination Interface
  29  * as described in ARM document number ARM DEN 0022A.
  30  */
  31
  32 #define AFFINITY_MASK(level)    ~((0x1UL << ((level) * MPIDR_LEVEL_BITS)) - 1)
  33
  34 static unsigned long psci_affinity_mask(unsigned long affinity_level)
  35 {
  36         if (affinity_level <= 3)
  37                 return MPIDR_HWID_BITMASK & AFFINITY_MASK(affinity_level);
  38
  39         return 0;
  40 }
  41
  42 static unsigned long kvm_psci_vcpu_suspend(struct kvm_vcpu *vcpu)
  43 {
  44         /*
  45          * NOTE: For simplicity, we make VCPU suspend emulation to be
  46          * same-as WFI (Wait-for-interrupt) emulation.
  47          *
  48          * This means for KVM the wakeup events are interrupts and
  49          * this is consistent with intended use of StateID as described
  50          * in section 5.4.1 of PSCI v0.2 specification (ARM DEN 0022A).
  51          *
  52          * Further, we also treat power-down request to be same as
  53          * stand-by request as-per section 5.4.2 clause 3 of PSCI v0.2
  54          * specification (ARM DEN 0022A). This means all suspend states
  55          * for KVM will preserve the register state.
  56          */
  57         kvm_vcpu_block(vcpu);
  58
  59         return PSCI_RET_SUCCESS;
  60 }
  61
  62 static void kvm_psci_vcpu_off(struct kvm_vcpu *vcpu)
  63 {
  64         vcpu->arch.pause = true;
  65 }
  66
  67 static unsigned long kvm_psci_vcpu_on(struct kvm_vcpu *source_vcpu)
  68 {
  69         struct kvm *kvm = source_vcpu->kvm;
  70         struct kvm_vcpu *vcpu = NULL;
  71         wait_queue_head_t *wq;
  72         unsigned long cpu_id;
  73         unsigned long context_id;
  74         phys_addr_t target_pc;
  75
  76         cpu_id = *vcpu_reg(source_vcpu, 1) & MPIDR_HWID_BITMASK;
  77         if (vcpu_mode_is_32bit(source_vcpu))
  78                 cpu_id &= ~((u32) 0);
  79
  80         vcpu = kvm_mpidr_to_vcpu(kvm, cpu_id);
  81
  82         /*
  83          * Make sure the caller requested a valid CPU and that the CPU is
  84          * turned off.
  85          */
  86         if (!vcpu)
  87                 return PSCI_RET_INVALID_PARAMS;
  88         if (!vcpu->arch.pause) {
  89                 if (kvm_psci_version(source_vcpu) != KVM_ARM_PSCI_0_1)
  90                         return PSCI_RET_ALREADY_ON;
  91                 else
  92                         return PSCI_RET_INVALID_PARAMS;
  93         }
  94
  95         target_pc = *vcpu_reg(source_vcpu, 2);
  96         context_id = *vcpu_reg(source_vcpu, 3);
  97
  98         kvm_reset_vcpu(vcpu);
  99
 100         /* Gracefully handle Thumb2 entry point */
 101         if (vcpu_mode_is_32bit(vcpu) && (target_pc & 1)) {
 102                 target_pc &= ~((phys_addr_t) 1);
 103                 vcpu_set_thumb(vcpu);
 104         }
 105
 106         /* Propagate caller endianness */
 107         if (kvm_vcpu_is_be(source_vcpu))
 108                 kvm_vcpu_set_be(vcpu);
 109
 110         *vcpu_pc(vcpu) = target_pc;
 111         /*
 112          * NOTE: We always update r0 (or x0) because for PSCI v0.1
 113          * the general puspose registers are undefined upon CPU_ON.
 114          */
 115         *vcpu_reg(vcpu, 0) = context_id;
 116         vcpu->arch.pause = false;
 117         smp_mb();               /* Make sure the above is visible */
 118
 119         wq = kvm_arch_vcpu_wq(vcpu);
 120         wake_up_interruptible(wq);
 121
 122         return PSCI_RET_SUCCESS;
 123 }
 124
 125 static unsigned long kvm_psci_vcpu_affinity_info(struct kvm_vcpu *vcpu)
 126 {
 127         int i;
 128         unsigned long mpidr;
 129         unsigned long target_affinity;
 130         unsigned long target_affinity_mask;
 131         unsigned long lowest_affinity_level;
 132         struct kvm *kvm = vcpu->kvm;
 133         struct kvm_vcpu *tmp;
 134
 135         target_affinity = *vcpu_reg(vcpu, 1);
 136         lowest_affinity_level = *vcpu_reg(vcpu, 2);
 137
 138         /* Determine target affinity mask */
 139         target_affinity_mask = psci_affinity_mask(lowest_affinity_level);
 140         if (!target_affinity_mask)
 141                 return PSCI_RET_INVALID_PARAMS;
 142
 143         /* Ignore other bits of target affinity */
 144         target_affinity &= target_affinity_mask;
 145
 146         /*
 147          * If one or more VCPU matching target affinity are running
 148          * then ON else OFF
 149          */
 150         kvm_for_each_vcpu(i, tmp, kvm) {
 151                 mpidr = kvm_vcpu_get_mpidr_aff(tmp);
 152                 if (((mpidr & target_affinity_mask) == target_affinity) &&
 153                     !tmp->arch.pause) {
 154                         return PSCI_0_2_AFFINITY_LEVEL_ON;
 155                 }
 156         }
 157
 158         return PSCI_0_2_AFFINITY_LEVEL_OFF;
 159 }
 160
 161 static void kvm_prepare_system_event(struct kvm_vcpu *vcpu, u32 type)
 162 {
 163         int i;
 164         struct kvm_vcpu *tmp;
 165
 166         /*
 167          * The KVM ABI specifies that a system event exit may call KVM_RUN
 168          * again and may perform shutdown/reboot at a later time that when the
 169          * actual request is made.  Since we are implementing PSCI and a
 170          * caller of PSCI reboot and shutdown expects that the system shuts
 171          * down or reboots immediately, let's make sure that VCPUs are not run
 172          * after this call is handled and before the VCPUs have been
 173          * re-initialized.
 174          */
 175         kvm_for_each_vcpu(i, tmp, vcpu->kvm) {
 176                 tmp->arch.pause = true;
 177                 kvm_vcpu_kick(tmp);
 178         }
 179
 180         memset(&vcpu->run->system_event, 0, sizeof(vcpu->run->system_event));
 181         vcpu->run->system_event.type = type;
 182         vcpu->run->exit_reason = KVM_EXIT_SYSTEM_EVENT;
 183 }
 184
 185 static void kvm_psci_system_off(struct kvm_vcpu *vcpu)
 186 {
 187         kvm_prepare_system_event(vcpu, KVM_SYSTEM_EVENT_SHUTDOWN);
 188 }
 189
 190 static void kvm_psci_system_reset(struct kvm_vcpu *vcpu)
 191 {
 192         kvm_prepare_system_event(vcpu, KVM_SYSTEM_EVENT_RESET);
 193 }
 194
 195 int kvm_psci_version(struct kvm_vcpu *vcpu)
 196 {
 197         if (test_bit(KVM_ARM_VCPU_PSCI_0_2, vcpu->arch.features))
 198                 return KVM_ARM_PSCI_0_2;
 199
 200         return KVM_ARM_PSCI_0_1;
 201 }
 202
 203 static int kvm_psci_0_2_call(struct kvm_vcpu *vcpu)
 204 {
 205         int ret = 1;
 206         unsigned long psci_fn = *vcpu_reg(vcpu, 0) & ~((u32) 0);
 207         unsigned long val;
 208
 209         switch (psci_fn) {
 210         case PSCI_0_2_FN_PSCI_VERSION:
 211                 /*
 212                  * Bits[31:16] = Major Version = 0
 213                  * Bits[15:0] = Minor Version = 2
 214                  */
 215                 val = 2;
 216                 break;
 217         case PSCI_0_2_FN_CPU_SUSPEND:
 218         case PSCI_0_2_FN64_CPU_SUSPEND:
 219                 val = kvm_psci_vcpu_suspend(vcpu);
 220                 break;
 221         case PSCI_0_2_FN_CPU_OFF:
 222                 kvm_psci_vcpu_off(vcpu);
 223                 val = PSCI_RET_SUCCESS;
 224                 break;
 225         case PSCI_0_2_FN_CPU_ON:
 226         case PSCI_0_2_FN64_CPU_ON:
 227                 val = kvm_psci_vcpu_on(vcpu);
 228                 break;
 229         case PSCI_0_2_FN_AFFINITY_INFO:
 230         case PSCI_0_2_FN64_AFFINITY_INFO:
 231                 val = kvm_psci_vcpu_affinity_info(vcpu);
 232                 break;
 233         case PSCI_0_2_FN_MIGRATE_INFO_TYPE:
 234                 /*
 235                  * Trusted OS is MP hence does not require migration
 236                  * or
 237                  * Trusted OS is not present
 238                  */
 239                 val = PSCI_0_2_TOS_MP;
 240                 break;
 241         case PSCI_0_2_FN_SYSTEM_OFF:
 242                 kvm_psci_system_off(vcpu);
 243                 /*
 244                  * We should'nt be going back to guest VCPU after
 245                  * receiving SYSTEM_OFF request.
 246                  *
 247                  * If user space accidently/deliberately resumes
 248                  * guest VCPU after SYSTEM_OFF request then guest
 249                  * VCPU should see internal failure from PSCI return
 250                  * value. To achieve this, we preload r0 (or x0) with
 251                  * PSCI return value INTERNAL_FAILURE.
 252                  */
 253                 val = PSCI_RET_INTERNAL_FAILURE;
 254                 ret = 0;
 255                 break;
 256         case PSCI_0_2_FN_SYSTEM_RESET:
 257                 kvm_psci_system_reset(vcpu);
 258                 /*
 259                  * Same reason as SYSTEM_OFF for preloading r0 (or x0)
 260                  * with PSCI return value INTERNAL_FAILURE.
 261                  */
 262                 val = PSCI_RET_INTERNAL_FAILURE;
 263                 ret = 0;
 264                 break;
 265         default:
 266                 val = PSCI_RET_NOT_SUPPORTED;
 267                 break;
 268         }
 269
 270         *vcpu_reg(vcpu, 0) = val;
 271         return ret;
 272 }
 273
 274 static int kvm_psci_0_1_call(struct kvm_vcpu *vcpu)
 275 {
 276         unsigned long psci_fn = *vcpu_reg(vcpu, 0) & ~((u32) 0);
 277         unsigned long val;
 278
 279         switch (psci_fn) {
 280         case KVM_PSCI_FN_CPU_OFF:
 281                 kvm_psci_vcpu_off(vcpu);
 282                 val = PSCI_RET_SUCCESS;
 283                 break;
 284         case KVM_PSCI_FN_CPU_ON:
 285                 val = kvm_psci_vcpu_on(vcpu);
 286                 break;
 287         default:
 288                 val = PSCI_RET_NOT_SUPPORTED;
 289                 break;
 290         }
 291
 292         *vcpu_reg(vcpu, 0) = val;
 293         return 1;
 294 }
 295
 296 /**
 297  * kvm_psci_call - handle PSCI call if r0 value is in range
 298  * @vcpu: Pointer to the VCPU struct
 299  *
 300  * Handle PSCI calls from guests through traps from HVC instructions.
 301  * The calling convention is similar to SMC calls to the secure world
 302  * where the function number is placed in r0.
 303  *
 304  * This function returns: > 0 (success), 0 (success but exit to user
 305  * space), and < 0 (errors)
 306  *
 307  * Errors:
 308  * -EINVAL: Unrecognized PSCI function
 309  */
 310 int kvm_psci_call(struct kvm_vcpu *vcpu)
 311 {
 312         switch (kvm_psci_version(vcpu)) {
 313         case KVM_ARM_PSCI_0_2:
 314                 return kvm_psci_0_2_call(vcpu);
 315         case KVM_ARM_PSCI_0_1:
 316                 return kvm_psci_0_1_call(vcpu);
 317         default:
 318                 return -EINVAL;
 319         };
 320 }