1 /* SPDX-License-Identifier: GPL-2.0-only */ 2 /* 3 * Copyright (C) 2012,2013 - ARM Ltd 4 * Author: Marc Zyngier <marc.zyngier@arm.com> 5 * 6 * Derived from arch/arm/include/asm/kvm_host.h: 7 * Copyright (C) 2012 - Virtual Open Systems and Columbia University 8 * Author: Christoffer Dall <c.dall@virtualopensystems.com> 9 */ 10 11 #ifndef __ARM64_KVM_HOST_H__ 12 #define __ARM64_KVM_HOST_H__ 13 14 #include <linux/arm-smccc.h> 15 #include <linux/bitmap.h> 16 #include <linux/types.h> 17 #include <linux/jump_label.h> 18 #include <linux/kvm_types.h> 19 #include <linux/maple_tree.h> 20 #include <linux/percpu.h> 21 #include <linux/psci.h> 22 #include <asm/arch_gicv3.h> 23 #include <asm/barrier.h> 24 #include <asm/cpufeature.h> 25 #include <asm/cputype.h> 26 #include <asm/daifflags.h> 27 #include <asm/fpsimd.h> 28 #include <asm/kvm.h> 29 #include <asm/kvm_asm.h> 30 #include <asm/vncr_mapping.h> 31 32 #define __KVM_HAVE_ARCH_INTC_INITIALIZED 33 34 #define KVM_HALT_POLL_NS_DEFAULT 500000 35 36 #include <kvm/arm_vgic.h> 37 #include <kvm/arm_arch_timer.h> 38 #include <kvm/arm_pmu.h> 39 40 #define KVM_MAX_VCPUS VGIC_V3_MAX_CPUS 41 42 #define KVM_VCPU_MAX_FEATURES 7 43 #define KVM_VCPU_VALID_FEATURES (BIT(KVM_VCPU_MAX_FEATURES) - 1) 44 45 #define KVM_REQ_SLEEP \ 46 KVM_ARCH_REQ_FLAGS(0, KVM_REQUEST_WAIT | KVM_REQUEST_NO_WAKEUP) 47 #define KVM_REQ_IRQ_PENDING KVM_ARCH_REQ(1) 48 #define KVM_REQ_VCPU_RESET KVM_ARCH_REQ(2) 49 #define KVM_REQ_RECORD_STEAL KVM_ARCH_REQ(3) 50 #define KVM_REQ_RELOAD_GICv4 KVM_ARCH_REQ(4) 51 #define KVM_REQ_RELOAD_PMU KVM_ARCH_REQ(5) 52 #define KVM_REQ_SUSPEND KVM_ARCH_REQ(6) 53 #define KVM_REQ_RESYNC_PMU_EL0 KVM_ARCH_REQ(7) 54 #define KVM_REQ_NESTED_S2_UNMAP KVM_ARCH_REQ(8) 55 56 #define KVM_DIRTY_LOG_MANUAL_CAPS (KVM_DIRTY_LOG_MANUAL_PROTECT_ENABLE | \ 57 KVM_DIRTY_LOG_INITIALLY_SET) 58 59 #define KVM_HAVE_MMU_RWLOCK 60 61 /* 62 * Mode of operation configurable with kvm-arm.mode early param. 63 * See Documentation/admin-guide/kernel-parameters.txt for more information. 64 */ 65 enum kvm_mode { 66 KVM_MODE_DEFAULT, 67 KVM_MODE_PROTECTED, 68 KVM_MODE_NV, 69 KVM_MODE_NONE, 70 }; 71 #ifdef CONFIG_KVM 72 enum kvm_mode kvm_get_mode(void); 73 #else 74 static inline enum kvm_mode kvm_get_mode(void) { return KVM_MODE_NONE; }; 75 #endif 76 77 extern unsigned int __ro_after_init kvm_sve_max_vl; 78 extern unsigned int __ro_after_init kvm_host_sve_max_vl; 79 int __init kvm_arm_init_sve(void); 80 81 u32 __attribute_const__ kvm_target_cpu(void); 82 void kvm_reset_vcpu(struct kvm_vcpu *vcpu); 83 void kvm_arm_vcpu_destroy(struct kvm_vcpu *vcpu); 84 85 struct kvm_hyp_memcache { 86 phys_addr_t head; 87 unsigned long nr_pages; 88 struct pkvm_mapping *mapping; /* only used from EL1 */ 89 }; 90 91 static inline void push_hyp_memcache(struct kvm_hyp_memcache *mc, 92 phys_addr_t *p, 93 phys_addr_t (*to_pa)(void *virt)) 94 { 95 *p = mc->head; 96 mc->head = to_pa(p); 97 mc->nr_pages++; 98 } 99 100 static inline void *pop_hyp_memcache(struct kvm_hyp_memcache *mc, 101 void *(*to_va)(phys_addr_t phys)) 102 { 103 phys_addr_t *p = to_va(mc->head & PAGE_MASK); 104 105 if (!mc->nr_pages) 106 return NULL; 107 108 mc->head = *p; 109 mc->nr_pages--; 110 111 return p; 112 } 113 114 static inline int __topup_hyp_memcache(struct kvm_hyp_memcache *mc, 115 unsigned long min_pages, 116 void *(*alloc_fn)(void *arg), 117 phys_addr_t (*to_pa)(void *virt), 118 void *arg) 119 { 120 while (mc->nr_pages < min_pages) { 121 phys_addr_t *p = alloc_fn(arg); 122 123 if (!p) 124 return -ENOMEM; 125 push_hyp_memcache(mc, p, to_pa); 126 } 127 128 return 0; 129 } 130 131 static inline void __free_hyp_memcache(struct kvm_hyp_memcache *mc, 132 void (*free_fn)(void *virt, void *arg), 133 void *(*to_va)(phys_addr_t phys), 134 void *arg) 135 { 136 while (mc->nr_pages) 137 free_fn(pop_hyp_memcache(mc, to_va), arg); 138 } 139 140 void free_hyp_memcache(struct kvm_hyp_memcache *mc); 141 int topup_hyp_memcache(struct kvm_hyp_memcache *mc, unsigned long min_pages); 142 143 struct kvm_vmid { 144 atomic64_t id; 145 }; 146 147 struct kvm_s2_mmu { 148 struct kvm_vmid vmid; 149 150 /* 151 * stage2 entry level table 152 * 153 * Two kvm_s2_mmu structures in the same VM can point to the same 154 * pgd here. This happens when running a guest using a 155 * translation regime that isn't affected by its own stage-2 156 * translation, such as a non-VHE hypervisor running at vEL2, or 157 * for vEL1/EL0 with vHCR_EL2.VM == 0. In that case, we use the 158 * canonical stage-2 page tables. 159 */ 160 phys_addr_t pgd_phys; 161 struct kvm_pgtable *pgt; 162 163 /* 164 * VTCR value used on the host. For a non-NV guest (or a NV 165 * guest that runs in a context where its own S2 doesn't 166 * apply), its T0SZ value reflects that of the IPA size. 167 * 168 * For a shadow S2 MMU, T0SZ reflects the PARange exposed to 169 * the guest. 170 */ 171 u64 vtcr; 172 173 /* The last vcpu id that ran on each physical CPU */ 174 int __percpu *last_vcpu_ran; 175 176 #define KVM_ARM_EAGER_SPLIT_CHUNK_SIZE_DEFAULT 0 177 /* 178 * Memory cache used to split 179 * KVM_CAP_ARM_EAGER_SPLIT_CHUNK_SIZE worth of huge pages. It 180 * is used to allocate stage2 page tables while splitting huge 181 * pages. The choice of KVM_CAP_ARM_EAGER_SPLIT_CHUNK_SIZE 182 * influences both the capacity of the split page cache, and 183 * how often KVM reschedules. Be wary of raising CHUNK_SIZE 184 * too high. 185 * 186 * Protected by kvm->slots_lock. 187 */ 188 struct kvm_mmu_memory_cache split_page_cache; 189 uint64_t split_page_chunk_size; 190 191 struct kvm_arch *arch; 192 193 /* 194 * For a shadow stage-2 MMU, the virtual vttbr used by the 195 * host to parse the guest S2. 196 * This either contains: 197 * - the virtual VTTBR programmed by the guest hypervisor with 198 * CnP cleared 199 * - The value 1 (VMID=0, BADDR=0, CnP=1) if invalid 200 * 201 * We also cache the full VTCR which gets used for TLB invalidation, 202 * taking the ARM ARM's "Any of the bits in VTCR_EL2 are permitted 203 * to be cached in a TLB" to the letter. 204 */ 205 u64 tlb_vttbr; 206 u64 tlb_vtcr; 207 208 /* 209 * true when this represents a nested context where virtual 210 * HCR_EL2.VM == 1 211 */ 212 bool nested_stage2_enabled; 213 214 /* 215 * true when this MMU needs to be unmapped before being used for a new 216 * purpose. 217 */ 218 bool pending_unmap; 219 220 /* 221 * 0: Nobody is currently using this, check vttbr for validity 222 * >0: Somebody is actively using this. 223 */ 224 atomic_t refcnt; 225 }; 226 227 struct kvm_arch_memory_slot { 228 }; 229 230 /** 231 * struct kvm_smccc_features: Descriptor of the hypercall services exposed to the guests 232 * 233 * @std_bmap: Bitmap of standard secure service calls 234 * @std_hyp_bmap: Bitmap of standard hypervisor service calls 235 * @vendor_hyp_bmap: Bitmap of vendor specific hypervisor service calls 236 */ 237 struct kvm_smccc_features { 238 unsigned long std_bmap; 239 unsigned long std_hyp_bmap; 240 unsigned long vendor_hyp_bmap; /* Function numbers 0-63 */ 241 unsigned long vendor_hyp_bmap_2; /* Function numbers 64-127 */ 242 }; 243 244 typedef unsigned int pkvm_handle_t; 245 246 struct kvm_protected_vm { 247 pkvm_handle_t handle; 248 struct kvm_hyp_memcache teardown_mc; 249 bool enabled; 250 }; 251 252 struct kvm_mpidr_data { 253 u64 mpidr_mask; 254 DECLARE_FLEX_ARRAY(u16, cmpidr_to_idx); 255 }; 256 257 static inline u16 kvm_mpidr_index(struct kvm_mpidr_data *data, u64 mpidr) 258 { 259 unsigned long index = 0, mask = data->mpidr_mask; 260 unsigned long aff = mpidr & MPIDR_HWID_BITMASK; 261 262 bitmap_gather(&index, &aff, &mask, fls(mask)); 263 264 return index; 265 } 266 267 struct kvm_sysreg_masks; 268 269 enum fgt_group_id { 270 __NO_FGT_GROUP__, 271 HFGxTR_GROUP, 272 HDFGRTR_GROUP, 273 HDFGWTR_GROUP = HDFGRTR_GROUP, 274 HFGITR_GROUP, 275 HAFGRTR_GROUP, 276 277 /* Must be last */ 278 __NR_FGT_GROUP_IDS__ 279 }; 280 281 struct kvm_arch { 282 struct kvm_s2_mmu mmu; 283 284 /* 285 * Fine-Grained UNDEF, mimicking the FGT layout defined by the 286 * architecture. We track them globally, as we present the 287 * same feature-set to all vcpus. 288 * 289 * Index 0 is currently spare. 290 */ 291 u64 fgu[__NR_FGT_GROUP_IDS__]; 292 293 /* 294 * Stage 2 paging state for VMs with nested S2 using a virtual 295 * VMID. 296 */ 297 struct kvm_s2_mmu *nested_mmus; 298 size_t nested_mmus_size; 299 int nested_mmus_next; 300 301 /* Interrupt controller */ 302 struct vgic_dist vgic; 303 304 /* Timers */ 305 struct arch_timer_vm_data timer_data; 306 307 /* Mandated version of PSCI */ 308 u32 psci_version; 309 310 /* Protects VM-scoped configuration data */ 311 struct mutex config_lock; 312 313 /* 314 * If we encounter a data abort without valid instruction syndrome 315 * information, report this to user space. User space can (and 316 * should) opt in to this feature if KVM_CAP_ARM_NISV_TO_USER is 317 * supported. 318 */ 319 #define KVM_ARCH_FLAG_RETURN_NISV_IO_ABORT_TO_USER 0 320 /* Memory Tagging Extension enabled for the guest */ 321 #define KVM_ARCH_FLAG_MTE_ENABLED 1 322 /* At least one vCPU has ran in the VM */ 323 #define KVM_ARCH_FLAG_HAS_RAN_ONCE 2 324 /* The vCPU feature set for the VM is configured */ 325 #define KVM_ARCH_FLAG_VCPU_FEATURES_CONFIGURED 3 326 /* PSCI SYSTEM_SUSPEND enabled for the guest */ 327 #define KVM_ARCH_FLAG_SYSTEM_SUSPEND_ENABLED 4 328 /* VM counter offset */ 329 #define KVM_ARCH_FLAG_VM_COUNTER_OFFSET 5 330 /* Timer PPIs made immutable */ 331 #define KVM_ARCH_FLAG_TIMER_PPIS_IMMUTABLE 6 332 /* Initial ID reg values loaded */ 333 #define KVM_ARCH_FLAG_ID_REGS_INITIALIZED 7 334 /* Fine-Grained UNDEF initialised */ 335 #define KVM_ARCH_FLAG_FGU_INITIALIZED 8 336 /* SVE exposed to guest */ 337 #define KVM_ARCH_FLAG_GUEST_HAS_SVE 9 338 unsigned long flags; 339 340 /* VM-wide vCPU feature set */ 341 DECLARE_BITMAP(vcpu_features, KVM_VCPU_MAX_FEATURES); 342 343 /* MPIDR to vcpu index mapping, optional */ 344 struct kvm_mpidr_data *mpidr_data; 345 346 /* 347 * VM-wide PMU filter, implemented as a bitmap and big enough for 348 * up to 2^10 events (ARMv8.0) or 2^16 events (ARMv8.1+). 349 */ 350 unsigned long *pmu_filter; 351 struct arm_pmu *arm_pmu; 352 353 cpumask_var_t supported_cpus; 354 355 /* PMCR_EL0.N value for the guest */ 356 u8 pmcr_n; 357 358 /* Iterator for idreg debugfs */ 359 u8 idreg_debugfs_iter; 360 361 /* Hypercall features firmware registers' descriptor */ 362 struct kvm_smccc_features smccc_feat; 363 struct maple_tree smccc_filter; 364 365 /* 366 * Emulated CPU ID registers per VM 367 * (Op0, Op1, CRn, CRm, Op2) of the ID registers to be saved in it 368 * is (3, 0, 0, crm, op2), where 1<=crm<8, 0<=op2<8. 369 * 370 * These emulated idregs are VM-wide, but accessed from the context of a vCPU. 371 * Atomic access to multiple idregs are guarded by kvm_arch.config_lock. 372 */ 373 #define IDREG_IDX(id) (((sys_reg_CRm(id) - 1) << 3) | sys_reg_Op2(id)) 374 #define KVM_ARM_ID_REG_NUM (IDREG_IDX(sys_reg(3, 0, 0, 7, 7)) + 1) 375 u64 id_regs[KVM_ARM_ID_REG_NUM]; 376 377 u64 ctr_el0; 378 379 /* Masks for VNCR-backed and general EL2 sysregs */ 380 struct kvm_sysreg_masks *sysreg_masks; 381 382 /* 383 * For an untrusted host VM, 'pkvm.handle' is used to lookup 384 * the associated pKVM instance in the hypervisor. 385 */ 386 struct kvm_protected_vm pkvm; 387 }; 388 389 struct kvm_vcpu_fault_info { 390 u64 esr_el2; /* Hyp Syndrom Register */ 391 u64 far_el2; /* Hyp Fault Address Register */ 392 u64 hpfar_el2; /* Hyp IPA Fault Address Register */ 393 u64 disr_el1; /* Deferred [SError] Status Register */ 394 }; 395 396 /* 397 * VNCR() just places the VNCR_capable registers in the enum after 398 * __VNCR_START__, and the value (after correction) to be an 8-byte offset 399 * from the VNCR base. As we don't require the enum to be otherwise ordered, 400 * we need the terrible hack below to ensure that we correctly size the 401 * sys_regs array, no matter what. 402 * 403 * The __MAX__ macro has been lifted from Sean Eron Anderson's wonderful 404 * treasure trove of bit hacks: 405 * https://graphics.stanford.edu/~seander/bithacks.html#IntegerMinOrMax 406 */ 407 #define __MAX__(x,y) ((x) ^ (((x) ^ (y)) & -((x) < (y)))) 408 #define VNCR(r) \ 409 __before_##r, \ 410 r = __VNCR_START__ + ((VNCR_ ## r) / 8), \ 411 __after_##r = __MAX__(__before_##r - 1, r) 412 413 #define MARKER(m) \ 414 m, __after_##m = m - 1 415 416 enum vcpu_sysreg { 417 __INVALID_SYSREG__, /* 0 is reserved as an invalid value */ 418 MPIDR_EL1, /* MultiProcessor Affinity Register */ 419 CLIDR_EL1, /* Cache Level ID Register */ 420 CSSELR_EL1, /* Cache Size Selection Register */ 421 TPIDR_EL0, /* Thread ID, User R/W */ 422 TPIDRRO_EL0, /* Thread ID, User R/O */ 423 TPIDR_EL1, /* Thread ID, Privileged */ 424 CNTKCTL_EL1, /* Timer Control Register (EL1) */ 425 PAR_EL1, /* Physical Address Register */ 426 MDCCINT_EL1, /* Monitor Debug Comms Channel Interrupt Enable Reg */ 427 OSLSR_EL1, /* OS Lock Status Register */ 428 DISR_EL1, /* Deferred Interrupt Status Register */ 429 430 /* Performance Monitors Registers */ 431 PMCR_EL0, /* Control Register */ 432 PMSELR_EL0, /* Event Counter Selection Register */ 433 PMEVCNTR0_EL0, /* Event Counter Register (0-30) */ 434 PMEVCNTR30_EL0 = PMEVCNTR0_EL0 + 30, 435 PMCCNTR_EL0, /* Cycle Counter Register */ 436 PMEVTYPER0_EL0, /* Event Type Register (0-30) */ 437 PMEVTYPER30_EL0 = PMEVTYPER0_EL0 + 30, 438 PMCCFILTR_EL0, /* Cycle Count Filter Register */ 439 PMCNTENSET_EL0, /* Count Enable Set Register */ 440 PMINTENSET_EL1, /* Interrupt Enable Set Register */ 441 PMOVSSET_EL0, /* Overflow Flag Status Set Register */ 442 PMUSERENR_EL0, /* User Enable Register */ 443 444 /* Pointer Authentication Registers in a strict increasing order. */ 445 APIAKEYLO_EL1, 446 APIAKEYHI_EL1, 447 APIBKEYLO_EL1, 448 APIBKEYHI_EL1, 449 APDAKEYLO_EL1, 450 APDAKEYHI_EL1, 451 APDBKEYLO_EL1, 452 APDBKEYHI_EL1, 453 APGAKEYLO_EL1, 454 APGAKEYHI_EL1, 455 456 /* Memory Tagging Extension registers */ 457 RGSR_EL1, /* Random Allocation Tag Seed Register */ 458 GCR_EL1, /* Tag Control Register */ 459 TFSRE0_EL1, /* Tag Fault Status Register (EL0) */ 460 461 POR_EL0, /* Permission Overlay Register 0 (EL0) */ 462 463 /* FP/SIMD/SVE */ 464 SVCR, 465 FPMR, 466 467 /* 32bit specific registers. */ 468 DACR32_EL2, /* Domain Access Control Register */ 469 IFSR32_EL2, /* Instruction Fault Status Register */ 470 FPEXC32_EL2, /* Floating-Point Exception Control Register */ 471 DBGVCR32_EL2, /* Debug Vector Catch Register */ 472 473 /* EL2 registers */ 474 SCTLR_EL2, /* System Control Register (EL2) */ 475 ACTLR_EL2, /* Auxiliary Control Register (EL2) */ 476 CPTR_EL2, /* Architectural Feature Trap Register (EL2) */ 477 HACR_EL2, /* Hypervisor Auxiliary Control Register */ 478 ZCR_EL2, /* SVE Control Register (EL2) */ 479 TTBR0_EL2, /* Translation Table Base Register 0 (EL2) */ 480 TTBR1_EL2, /* Translation Table Base Register 1 (EL2) */ 481 TCR_EL2, /* Translation Control Register (EL2) */ 482 PIRE0_EL2, /* Permission Indirection Register 0 (EL2) */ 483 PIR_EL2, /* Permission Indirection Register 1 (EL2) */ 484 POR_EL2, /* Permission Overlay Register 2 (EL2) */ 485 SPSR_EL2, /* EL2 saved program status register */ 486 ELR_EL2, /* EL2 exception link register */ 487 AFSR0_EL2, /* Auxiliary Fault Status Register 0 (EL2) */ 488 AFSR1_EL2, /* Auxiliary Fault Status Register 1 (EL2) */ 489 ESR_EL2, /* Exception Syndrome Register (EL2) */ 490 FAR_EL2, /* Fault Address Register (EL2) */ 491 HPFAR_EL2, /* Hypervisor IPA Fault Address Register */ 492 MAIR_EL2, /* Memory Attribute Indirection Register (EL2) */ 493 AMAIR_EL2, /* Auxiliary Memory Attribute Indirection Register (EL2) */ 494 VBAR_EL2, /* Vector Base Address Register (EL2) */ 495 RVBAR_EL2, /* Reset Vector Base Address Register */ 496 CONTEXTIDR_EL2, /* Context ID Register (EL2) */ 497 SP_EL2, /* EL2 Stack Pointer */ 498 CNTHP_CTL_EL2, 499 CNTHP_CVAL_EL2, 500 CNTHV_CTL_EL2, 501 CNTHV_CVAL_EL2, 502 503 /* Anything from this can be RES0/RES1 sanitised */ 504 MARKER(__SANITISED_REG_START__), 505 TCR2_EL2, /* Extended Translation Control Register (EL2) */ 506 MDCR_EL2, /* Monitor Debug Configuration Register (EL2) */ 507 CNTHCTL_EL2, /* Counter-timer Hypervisor Control register */ 508 509 /* Any VNCR-capable reg goes after this point */ 510 MARKER(__VNCR_START__), 511 512 VNCR(SCTLR_EL1),/* System Control Register */ 513 VNCR(ACTLR_EL1),/* Auxiliary Control Register */ 514 VNCR(CPACR_EL1),/* Coprocessor Access Control */ 515 VNCR(ZCR_EL1), /* SVE Control */ 516 VNCR(TTBR0_EL1),/* Translation Table Base Register 0 */ 517 VNCR(TTBR1_EL1),/* Translation Table Base Register 1 */ 518 VNCR(TCR_EL1), /* Translation Control Register */ 519 VNCR(TCR2_EL1), /* Extended Translation Control Register */ 520 VNCR(ESR_EL1), /* Exception Syndrome Register */ 521 VNCR(AFSR0_EL1),/* Auxiliary Fault Status Register 0 */ 522 VNCR(AFSR1_EL1),/* Auxiliary Fault Status Register 1 */ 523 VNCR(FAR_EL1), /* Fault Address Register */ 524 VNCR(MAIR_EL1), /* Memory Attribute Indirection Register */ 525 VNCR(VBAR_EL1), /* Vector Base Address Register */ 526 VNCR(CONTEXTIDR_EL1), /* Context ID Register */ 527 VNCR(AMAIR_EL1),/* Aux Memory Attribute Indirection Register */ 528 VNCR(MDSCR_EL1),/* Monitor Debug System Control Register */ 529 VNCR(ELR_EL1), 530 VNCR(SP_EL1), 531 VNCR(SPSR_EL1), 532 VNCR(TFSR_EL1), /* Tag Fault Status Register (EL1) */ 533 VNCR(VPIDR_EL2),/* Virtualization Processor ID Register */ 534 VNCR(VMPIDR_EL2),/* Virtualization Multiprocessor ID Register */ 535 VNCR(HCR_EL2), /* Hypervisor Configuration Register */ 536 VNCR(HSTR_EL2), /* Hypervisor System Trap Register */ 537 VNCR(VTTBR_EL2),/* Virtualization Translation Table Base Register */ 538 VNCR(VTCR_EL2), /* Virtualization Translation Control Register */ 539 VNCR(TPIDR_EL2),/* EL2 Software Thread ID Register */ 540 VNCR(HCRX_EL2), /* Extended Hypervisor Configuration Register */ 541 542 /* Permission Indirection Extension registers */ 543 VNCR(PIR_EL1), /* Permission Indirection Register 1 (EL1) */ 544 VNCR(PIRE0_EL1), /* Permission Indirection Register 0 (EL1) */ 545 546 VNCR(POR_EL1), /* Permission Overlay Register 1 (EL1) */ 547 548 VNCR(HFGRTR_EL2), 549 VNCR(HFGWTR_EL2), 550 VNCR(HFGITR_EL2), 551 VNCR(HDFGRTR_EL2), 552 VNCR(HDFGWTR_EL2), 553 VNCR(HAFGRTR_EL2), 554 555 VNCR(CNTVOFF_EL2), 556 VNCR(CNTV_CVAL_EL0), 557 VNCR(CNTV_CTL_EL0), 558 VNCR(CNTP_CVAL_EL0), 559 VNCR(CNTP_CTL_EL0), 560 561 VNCR(ICH_HCR_EL2), 562 563 NR_SYS_REGS /* Nothing after this line! */ 564 }; 565 566 struct kvm_sysreg_masks { 567 struct { 568 u64 res0; 569 u64 res1; 570 } mask[NR_SYS_REGS - __SANITISED_REG_START__]; 571 }; 572 573 struct kvm_cpu_context { 574 struct user_pt_regs regs; /* sp = sp_el0 */ 575 576 u64 spsr_abt; 577 u64 spsr_und; 578 u64 spsr_irq; 579 u64 spsr_fiq; 580 581 struct user_fpsimd_state fp_regs; 582 583 u64 sys_regs[NR_SYS_REGS]; 584 585 struct kvm_vcpu *__hyp_running_vcpu; 586 587 /* This pointer has to be 4kB aligned. */ 588 u64 *vncr_array; 589 }; 590 591 struct cpu_sve_state { 592 __u64 zcr_el1; 593 594 /* 595 * Ordering is important since __sve_save_state/__sve_restore_state 596 * relies on it. 597 */ 598 __u32 fpsr; 599 __u32 fpcr; 600 601 /* Must be SVE_VQ_BYTES (128 bit) aligned. */ 602 __u8 sve_regs[]; 603 }; 604 605 /* 606 * This structure is instantiated on a per-CPU basis, and contains 607 * data that is: 608 * 609 * - tied to a single physical CPU, and 610 * - either have a lifetime that does not extend past vcpu_put() 611 * - or is an invariant for the lifetime of the system 612 * 613 * Use host_data_ptr(field) as a way to access a pointer to such a 614 * field. 615 */ 616 struct kvm_host_data { 617 #define KVM_HOST_DATA_FLAG_HAS_SPE 0 618 #define KVM_HOST_DATA_FLAG_HAS_TRBE 1 619 #define KVM_HOST_DATA_FLAG_TRBE_ENABLED 4 620 #define KVM_HOST_DATA_FLAG_EL1_TRACING_CONFIGURED 5 621 unsigned long flags; 622 623 struct kvm_cpu_context host_ctxt; 624 625 /* 626 * Hyp VA. 627 * sve_state is only used in pKVM and if system_supports_sve(). 628 */ 629 struct cpu_sve_state *sve_state; 630 631 /* Used by pKVM only. */ 632 u64 fpmr; 633 634 /* Ownership of the FP regs */ 635 enum { 636 FP_STATE_FREE, 637 FP_STATE_HOST_OWNED, 638 FP_STATE_GUEST_OWNED, 639 } fp_owner; 640 641 /* 642 * host_debug_state contains the host registers which are 643 * saved and restored during world switches. 644 */ 645 struct { 646 /* {Break,watch}point registers */ 647 struct kvm_guest_debug_arch regs; 648 /* Statistical profiling extension */ 649 u64 pmscr_el1; 650 /* Self-hosted trace */ 651 u64 trfcr_el1; 652 /* Values of trap registers for the host before guest entry. */ 653 u64 mdcr_el2; 654 } host_debug_state; 655 656 /* Guest trace filter value */ 657 u64 trfcr_while_in_guest; 658 659 /* Number of programmable event counters (PMCR_EL0.N) for this CPU */ 660 unsigned int nr_event_counters; 661 662 /* Number of debug breakpoints/watchpoints for this CPU (minus 1) */ 663 unsigned int debug_brps; 664 unsigned int debug_wrps; 665 }; 666 667 struct kvm_host_psci_config { 668 /* PSCI version used by host. */ 669 u32 version; 670 u32 smccc_version; 671 672 /* Function IDs used by host if version is v0.1. */ 673 struct psci_0_1_function_ids function_ids_0_1; 674 675 bool psci_0_1_cpu_suspend_implemented; 676 bool psci_0_1_cpu_on_implemented; 677 bool psci_0_1_cpu_off_implemented; 678 bool psci_0_1_migrate_implemented; 679 }; 680 681 extern struct kvm_host_psci_config kvm_nvhe_sym(kvm_host_psci_config); 682 #define kvm_host_psci_config CHOOSE_NVHE_SYM(kvm_host_psci_config) 683 684 extern s64 kvm_nvhe_sym(hyp_physvirt_offset); 685 #define hyp_physvirt_offset CHOOSE_NVHE_SYM(hyp_physvirt_offset) 686 687 extern u64 kvm_nvhe_sym(hyp_cpu_logical_map)[NR_CPUS]; 688 #define hyp_cpu_logical_map CHOOSE_NVHE_SYM(hyp_cpu_logical_map) 689 690 struct vcpu_reset_state { 691 unsigned long pc; 692 unsigned long r0; 693 bool be; 694 bool reset; 695 }; 696 697 struct kvm_vcpu_arch { 698 struct kvm_cpu_context ctxt; 699 700 /* 701 * Guest floating point state 702 * 703 * The architecture has two main floating point extensions, 704 * the original FPSIMD and SVE. These have overlapping 705 * register views, with the FPSIMD V registers occupying the 706 * low 128 bits of the SVE Z registers. When the core 707 * floating point code saves the register state of a task it 708 * records which view it saved in fp_type. 709 */ 710 void *sve_state; 711 enum fp_type fp_type; 712 unsigned int sve_max_vl; 713 714 /* Stage 2 paging state used by the hardware on next switch */ 715 struct kvm_s2_mmu *hw_mmu; 716 717 /* Values of trap registers for the guest. */ 718 u64 hcr_el2; 719 u64 hcrx_el2; 720 u64 mdcr_el2; 721 722 /* Exception Information */ 723 struct kvm_vcpu_fault_info fault; 724 725 /* Configuration flags, set once and for all before the vcpu can run */ 726 u8 cflags; 727 728 /* Input flags to the hypervisor code, potentially cleared after use */ 729 u8 iflags; 730 731 /* State flags for kernel bookkeeping, unused by the hypervisor code */ 732 u8 sflags; 733 734 /* 735 * Don't run the guest (internal implementation need). 736 * 737 * Contrary to the flags above, this is set/cleared outside of 738 * a vcpu context, and thus cannot be mixed with the flags 739 * themselves (or the flag accesses need to be made atomic). 740 */ 741 bool pause; 742 743 /* 744 * We maintain more than a single set of debug registers to support 745 * debugging the guest from the host and to maintain separate host and 746 * guest state during world switches. vcpu_debug_state are the debug 747 * registers of the vcpu as the guest sees them. 748 * 749 * external_debug_state contains the debug values we want to debug the 750 * guest. This is set via the KVM_SET_GUEST_DEBUG ioctl. 751 */ 752 struct kvm_guest_debug_arch vcpu_debug_state; 753 struct kvm_guest_debug_arch external_debug_state; 754 u64 external_mdscr_el1; 755 756 enum { 757 VCPU_DEBUG_FREE, 758 VCPU_DEBUG_HOST_OWNED, 759 VCPU_DEBUG_GUEST_OWNED, 760 } debug_owner; 761 762 /* VGIC state */ 763 struct vgic_cpu vgic_cpu; 764 struct arch_timer_cpu timer_cpu; 765 struct kvm_pmu pmu; 766 767 /* vcpu power state */ 768 struct kvm_mp_state mp_state; 769 spinlock_t mp_state_lock; 770 771 /* Cache some mmu pages needed inside spinlock regions */ 772 struct kvm_mmu_memory_cache mmu_page_cache; 773 774 /* Pages to top-up the pKVM/EL2 guest pool */ 775 struct kvm_hyp_memcache pkvm_memcache; 776 777 /* Virtual SError ESR to restore when HCR_EL2.VSE is set */ 778 u64 vsesr_el2; 779 780 /* Additional reset state */ 781 struct vcpu_reset_state reset_state; 782 783 /* Guest PV state */ 784 struct { 785 u64 last_steal; 786 gpa_t base; 787 } steal; 788 789 /* Per-vcpu CCSIDR override or NULL */ 790 u32 *ccsidr; 791 }; 792 793 /* 794 * Each 'flag' is composed of a comma-separated triplet: 795 * 796 * - the flag-set it belongs to in the vcpu->arch structure 797 * - the value for that flag 798 * - the mask for that flag 799 * 800 * __vcpu_single_flag() builds such a triplet for a single-bit flag. 801 * unpack_vcpu_flag() extract the flag value from the triplet for 802 * direct use outside of the flag accessors. 803 */ 804 #define __vcpu_single_flag(_set, _f) _set, (_f), (_f) 805 806 #define __unpack_flag(_set, _f, _m) _f 807 #define unpack_vcpu_flag(...) __unpack_flag(__VA_ARGS__) 808 809 #define __build_check_flag(v, flagset, f, m) \ 810 do { \ 811 typeof(v->arch.flagset) *_fset; \ 812 \ 813 /* Check that the flags fit in the mask */ \ 814 BUILD_BUG_ON(HWEIGHT(m) != HWEIGHT((f) | (m))); \ 815 /* Check that the flags fit in the type */ \ 816 BUILD_BUG_ON((sizeof(*_fset) * 8) <= __fls(m)); \ 817 } while (0) 818 819 #define __vcpu_get_flag(v, flagset, f, m) \ 820 ({ \ 821 __build_check_flag(v, flagset, f, m); \ 822 \ 823 READ_ONCE(v->arch.flagset) & (m); \ 824 }) 825 826 /* 827 * Note that the set/clear accessors must be preempt-safe in order to 828 * avoid nesting them with load/put which also manipulate flags... 829 */ 830 #ifdef __KVM_NVHE_HYPERVISOR__ 831 /* the nVHE hypervisor is always non-preemptible */ 832 #define __vcpu_flags_preempt_disable() 833 #define __vcpu_flags_preempt_enable() 834 #else 835 #define __vcpu_flags_preempt_disable() preempt_disable() 836 #define __vcpu_flags_preempt_enable() preempt_enable() 837 #endif 838 839 #define __vcpu_set_flag(v, flagset, f, m) \ 840 do { \ 841 typeof(v->arch.flagset) *fset; \ 842 \ 843 __build_check_flag(v, flagset, f, m); \ 844 \ 845 fset = &v->arch.flagset; \ 846 __vcpu_flags_preempt_disable(); \ 847 if (HWEIGHT(m) > 1) \ 848 *fset &= ~(m); \ 849 *fset |= (f); \ 850 __vcpu_flags_preempt_enable(); \ 851 } while (0) 852 853 #define __vcpu_clear_flag(v, flagset, f, m) \ 854 do { \ 855 typeof(v->arch.flagset) *fset; \ 856 \ 857 __build_check_flag(v, flagset, f, m); \ 858 \ 859 fset = &v->arch.flagset; \ 860 __vcpu_flags_preempt_disable(); \ 861 *fset &= ~(m); \ 862 __vcpu_flags_preempt_enable(); \ 863 } while (0) 864 865 #define vcpu_get_flag(v, ...) __vcpu_get_flag((v), __VA_ARGS__) 866 #define vcpu_set_flag(v, ...) __vcpu_set_flag((v), __VA_ARGS__) 867 #define vcpu_clear_flag(v, ...) __vcpu_clear_flag((v), __VA_ARGS__) 868 869 /* KVM_ARM_VCPU_INIT completed */ 870 #define VCPU_INITIALIZED __vcpu_single_flag(cflags, BIT(0)) 871 /* SVE config completed */ 872 #define VCPU_SVE_FINALIZED __vcpu_single_flag(cflags, BIT(1)) 873 874 /* Exception pending */ 875 #define PENDING_EXCEPTION __vcpu_single_flag(iflags, BIT(0)) 876 /* 877 * PC increment. Overlaps with EXCEPT_MASK on purpose so that it can't 878 * be set together with an exception... 879 */ 880 #define INCREMENT_PC __vcpu_single_flag(iflags, BIT(1)) 881 /* Target EL/MODE (not a single flag, but let's abuse the macro) */ 882 #define EXCEPT_MASK __vcpu_single_flag(iflags, GENMASK(3, 1)) 883 884 /* Helpers to encode exceptions with minimum fuss */ 885 #define __EXCEPT_MASK_VAL unpack_vcpu_flag(EXCEPT_MASK) 886 #define __EXCEPT_SHIFT __builtin_ctzl(__EXCEPT_MASK_VAL) 887 #define __vcpu_except_flags(_f) iflags, (_f << __EXCEPT_SHIFT), __EXCEPT_MASK_VAL 888 889 /* 890 * When PENDING_EXCEPTION is set, EXCEPT_MASK can take the following 891 * values: 892 * 893 * For AArch32 EL1: 894 */ 895 #define EXCEPT_AA32_UND __vcpu_except_flags(0) 896 #define EXCEPT_AA32_IABT __vcpu_except_flags(1) 897 #define EXCEPT_AA32_DABT __vcpu_except_flags(2) 898 /* For AArch64: */ 899 #define EXCEPT_AA64_EL1_SYNC __vcpu_except_flags(0) 900 #define EXCEPT_AA64_EL1_IRQ __vcpu_except_flags(1) 901 #define EXCEPT_AA64_EL1_FIQ __vcpu_except_flags(2) 902 #define EXCEPT_AA64_EL1_SERR __vcpu_except_flags(3) 903 /* For AArch64 with NV: */ 904 #define EXCEPT_AA64_EL2_SYNC __vcpu_except_flags(4) 905 #define EXCEPT_AA64_EL2_IRQ __vcpu_except_flags(5) 906 #define EXCEPT_AA64_EL2_FIQ __vcpu_except_flags(6) 907 #define EXCEPT_AA64_EL2_SERR __vcpu_except_flags(7) 908 909 /* Physical CPU not in supported_cpus */ 910 #define ON_UNSUPPORTED_CPU __vcpu_single_flag(sflags, BIT(0)) 911 /* WFIT instruction trapped */ 912 #define IN_WFIT __vcpu_single_flag(sflags, BIT(1)) 913 /* vcpu system registers loaded on physical CPU */ 914 #define SYSREGS_ON_CPU __vcpu_single_flag(sflags, BIT(2)) 915 /* Software step state is Active-pending for external debug */ 916 #define HOST_SS_ACTIVE_PENDING __vcpu_single_flag(sflags, BIT(3)) 917 /* Software step state is Active pending for guest debug */ 918 #define GUEST_SS_ACTIVE_PENDING __vcpu_single_flag(sflags, BIT(4)) 919 /* PMUSERENR for the guest EL0 is on physical CPU */ 920 #define PMUSERENR_ON_CPU __vcpu_single_flag(sflags, BIT(5)) 921 /* WFI instruction trapped */ 922 #define IN_WFI __vcpu_single_flag(sflags, BIT(6)) 923 924 925 /* Pointer to the vcpu's SVE FFR for sve_{save,load}_state() */ 926 #define vcpu_sve_pffr(vcpu) (kern_hyp_va((vcpu)->arch.sve_state) + \ 927 sve_ffr_offset((vcpu)->arch.sve_max_vl)) 928 929 #define vcpu_sve_max_vq(vcpu) sve_vq_from_vl((vcpu)->arch.sve_max_vl) 930 931 #define vcpu_sve_zcr_elx(vcpu) \ 932 (unlikely(is_hyp_ctxt(vcpu)) ? ZCR_EL2 : ZCR_EL1) 933 934 #define vcpu_sve_state_size(vcpu) ({ \ 935 size_t __size_ret; \ 936 unsigned int __vcpu_vq; \ 937 \ 938 if (WARN_ON(!sve_vl_valid((vcpu)->arch.sve_max_vl))) { \ 939 __size_ret = 0; \ 940 } else { \ 941 __vcpu_vq = vcpu_sve_max_vq(vcpu); \ 942 __size_ret = SVE_SIG_REGS_SIZE(__vcpu_vq); \ 943 } \ 944 \ 945 __size_ret; \ 946 }) 947 948 #define KVM_GUESTDBG_VALID_MASK (KVM_GUESTDBG_ENABLE | \ 949 KVM_GUESTDBG_USE_SW_BP | \ 950 KVM_GUESTDBG_USE_HW | \ 951 KVM_GUESTDBG_SINGLESTEP) 952 953 #define kvm_has_sve(kvm) (system_supports_sve() && \ 954 test_bit(KVM_ARCH_FLAG_GUEST_HAS_SVE, &(kvm)->arch.flags)) 955 956 #ifdef __KVM_NVHE_HYPERVISOR__ 957 #define vcpu_has_sve(vcpu) kvm_has_sve(kern_hyp_va((vcpu)->kvm)) 958 #else 959 #define vcpu_has_sve(vcpu) kvm_has_sve((vcpu)->kvm) 960 #endif 961 962 #ifdef CONFIG_ARM64_PTR_AUTH 963 #define vcpu_has_ptrauth(vcpu) \ 964 ((cpus_have_final_cap(ARM64_HAS_ADDRESS_AUTH) || \ 965 cpus_have_final_cap(ARM64_HAS_GENERIC_AUTH)) && \ 966 (vcpu_has_feature(vcpu, KVM_ARM_VCPU_PTRAUTH_ADDRESS) || \ 967 vcpu_has_feature(vcpu, KVM_ARM_VCPU_PTRAUTH_GENERIC))) 968 #else 969 #define vcpu_has_ptrauth(vcpu) false 970 #endif 971 972 #define vcpu_on_unsupported_cpu(vcpu) \ 973 vcpu_get_flag(vcpu, ON_UNSUPPORTED_CPU) 974 975 #define vcpu_set_on_unsupported_cpu(vcpu) \ 976 vcpu_set_flag(vcpu, ON_UNSUPPORTED_CPU) 977 978 #define vcpu_clear_on_unsupported_cpu(vcpu) \ 979 vcpu_clear_flag(vcpu, ON_UNSUPPORTED_CPU) 980 981 #define vcpu_gp_regs(v) (&(v)->arch.ctxt.regs) 982 983 /* 984 * Only use __vcpu_sys_reg/ctxt_sys_reg if you know you want the 985 * memory backed version of a register, and not the one most recently 986 * accessed by a running VCPU. For example, for userspace access or 987 * for system registers that are never context switched, but only 988 * emulated. 989 * 990 * Don't bother with VNCR-based accesses in the nVHE code, it has no 991 * business dealing with NV. 992 */ 993 static inline u64 *___ctxt_sys_reg(const struct kvm_cpu_context *ctxt, int r) 994 { 995 #if !defined (__KVM_NVHE_HYPERVISOR__) 996 if (unlikely(cpus_have_final_cap(ARM64_HAS_NESTED_VIRT) && 997 r >= __VNCR_START__ && ctxt->vncr_array)) 998 return &ctxt->vncr_array[r - __VNCR_START__]; 999 #endif 1000 return (u64 *)&ctxt->sys_regs[r]; 1001 } 1002 1003 #define __ctxt_sys_reg(c,r) \ 1004 ({ \ 1005 BUILD_BUG_ON(__builtin_constant_p(r) && \ 1006 (r) >= NR_SYS_REGS); \ 1007 ___ctxt_sys_reg(c, r); \ 1008 }) 1009 1010 #define ctxt_sys_reg(c,r) (*__ctxt_sys_reg(c,r)) 1011 1012 u64 kvm_vcpu_apply_reg_masks(const struct kvm_vcpu *, enum vcpu_sysreg, u64); 1013 #define __vcpu_sys_reg(v,r) \ 1014 (*({ \ 1015 const struct kvm_cpu_context *ctxt = &(v)->arch.ctxt; \ 1016 u64 *__r = __ctxt_sys_reg(ctxt, (r)); \ 1017 if (vcpu_has_nv((v)) && (r) >= __SANITISED_REG_START__) \ 1018 *__r = kvm_vcpu_apply_reg_masks((v), (r), *__r);\ 1019 __r; \ 1020 })) 1021 1022 u64 vcpu_read_sys_reg(const struct kvm_vcpu *vcpu, int reg); 1023 void vcpu_write_sys_reg(struct kvm_vcpu *vcpu, u64 val, int reg); 1024 1025 static inline bool __vcpu_read_sys_reg_from_cpu(int reg, u64 *val) 1026 { 1027 /* 1028 * *** VHE ONLY *** 1029 * 1030 * System registers listed in the switch are not saved on every 1031 * exit from the guest but are only saved on vcpu_put. 1032 * 1033 * Note that MPIDR_EL1 for the guest is set by KVM via VMPIDR_EL2 but 1034 * should never be listed below, because the guest cannot modify its 1035 * own MPIDR_EL1 and MPIDR_EL1 is accessed for VCPU A from VCPU B's 1036 * thread when emulating cross-VCPU communication. 1037 */ 1038 if (!has_vhe()) 1039 return false; 1040 1041 switch (reg) { 1042 case SCTLR_EL1: *val = read_sysreg_s(SYS_SCTLR_EL12); break; 1043 case CPACR_EL1: *val = read_sysreg_s(SYS_CPACR_EL12); break; 1044 case TTBR0_EL1: *val = read_sysreg_s(SYS_TTBR0_EL12); break; 1045 case TTBR1_EL1: *val = read_sysreg_s(SYS_TTBR1_EL12); break; 1046 case TCR_EL1: *val = read_sysreg_s(SYS_TCR_EL12); break; 1047 case TCR2_EL1: *val = read_sysreg_s(SYS_TCR2_EL12); break; 1048 case PIR_EL1: *val = read_sysreg_s(SYS_PIR_EL12); break; 1049 case PIRE0_EL1: *val = read_sysreg_s(SYS_PIRE0_EL12); break; 1050 case POR_EL1: *val = read_sysreg_s(SYS_POR_EL12); break; 1051 case ESR_EL1: *val = read_sysreg_s(SYS_ESR_EL12); break; 1052 case AFSR0_EL1: *val = read_sysreg_s(SYS_AFSR0_EL12); break; 1053 case AFSR1_EL1: *val = read_sysreg_s(SYS_AFSR1_EL12); break; 1054 case FAR_EL1: *val = read_sysreg_s(SYS_FAR_EL12); break; 1055 case MAIR_EL1: *val = read_sysreg_s(SYS_MAIR_EL12); break; 1056 case VBAR_EL1: *val = read_sysreg_s(SYS_VBAR_EL12); break; 1057 case CONTEXTIDR_EL1: *val = read_sysreg_s(SYS_CONTEXTIDR_EL12);break; 1058 case TPIDR_EL0: *val = read_sysreg_s(SYS_TPIDR_EL0); break; 1059 case TPIDRRO_EL0: *val = read_sysreg_s(SYS_TPIDRRO_EL0); break; 1060 case TPIDR_EL1: *val = read_sysreg_s(SYS_TPIDR_EL1); break; 1061 case AMAIR_EL1: *val = read_sysreg_s(SYS_AMAIR_EL12); break; 1062 case CNTKCTL_EL1: *val = read_sysreg_s(SYS_CNTKCTL_EL12); break; 1063 case ELR_EL1: *val = read_sysreg_s(SYS_ELR_EL12); break; 1064 case SPSR_EL1: *val = read_sysreg_s(SYS_SPSR_EL12); break; 1065 case PAR_EL1: *val = read_sysreg_par(); break; 1066 case DACR32_EL2: *val = read_sysreg_s(SYS_DACR32_EL2); break; 1067 case IFSR32_EL2: *val = read_sysreg_s(SYS_IFSR32_EL2); break; 1068 case DBGVCR32_EL2: *val = read_sysreg_s(SYS_DBGVCR32_EL2); break; 1069 case ZCR_EL1: *val = read_sysreg_s(SYS_ZCR_EL12); break; 1070 default: return false; 1071 } 1072 1073 return true; 1074 } 1075 1076 static inline bool __vcpu_write_sys_reg_to_cpu(u64 val, int reg) 1077 { 1078 /* 1079 * *** VHE ONLY *** 1080 * 1081 * System registers listed in the switch are not restored on every 1082 * entry to the guest but are only restored on vcpu_load. 1083 * 1084 * Note that MPIDR_EL1 for the guest is set by KVM via VMPIDR_EL2 but 1085 * should never be listed below, because the MPIDR should only be set 1086 * once, before running the VCPU, and never changed later. 1087 */ 1088 if (!has_vhe()) 1089 return false; 1090 1091 switch (reg) { 1092 case SCTLR_EL1: write_sysreg_s(val, SYS_SCTLR_EL12); break; 1093 case CPACR_EL1: write_sysreg_s(val, SYS_CPACR_EL12); break; 1094 case TTBR0_EL1: write_sysreg_s(val, SYS_TTBR0_EL12); break; 1095 case TTBR1_EL1: write_sysreg_s(val, SYS_TTBR1_EL12); break; 1096 case TCR_EL1: write_sysreg_s(val, SYS_TCR_EL12); break; 1097 case TCR2_EL1: write_sysreg_s(val, SYS_TCR2_EL12); break; 1098 case PIR_EL1: write_sysreg_s(val, SYS_PIR_EL12); break; 1099 case PIRE0_EL1: write_sysreg_s(val, SYS_PIRE0_EL12); break; 1100 case POR_EL1: write_sysreg_s(val, SYS_POR_EL12); break; 1101 case ESR_EL1: write_sysreg_s(val, SYS_ESR_EL12); break; 1102 case AFSR0_EL1: write_sysreg_s(val, SYS_AFSR0_EL12); break; 1103 case AFSR1_EL1: write_sysreg_s(val, SYS_AFSR1_EL12); break; 1104 case FAR_EL1: write_sysreg_s(val, SYS_FAR_EL12); break; 1105 case MAIR_EL1: write_sysreg_s(val, SYS_MAIR_EL12); break; 1106 case VBAR_EL1: write_sysreg_s(val, SYS_VBAR_EL12); break; 1107 case CONTEXTIDR_EL1: write_sysreg_s(val, SYS_CONTEXTIDR_EL12);break; 1108 case TPIDR_EL0: write_sysreg_s(val, SYS_TPIDR_EL0); break; 1109 case TPIDRRO_EL0: write_sysreg_s(val, SYS_TPIDRRO_EL0); break; 1110 case TPIDR_EL1: write_sysreg_s(val, SYS_TPIDR_EL1); break; 1111 case AMAIR_EL1: write_sysreg_s(val, SYS_AMAIR_EL12); break; 1112 case CNTKCTL_EL1: write_sysreg_s(val, SYS_CNTKCTL_EL12); break; 1113 case ELR_EL1: write_sysreg_s(val, SYS_ELR_EL12); break; 1114 case SPSR_EL1: write_sysreg_s(val, SYS_SPSR_EL12); break; 1115 case PAR_EL1: write_sysreg_s(val, SYS_PAR_EL1); break; 1116 case DACR32_EL2: write_sysreg_s(val, SYS_DACR32_EL2); break; 1117 case IFSR32_EL2: write_sysreg_s(val, SYS_IFSR32_EL2); break; 1118 case DBGVCR32_EL2: write_sysreg_s(val, SYS_DBGVCR32_EL2); break; 1119 case ZCR_EL1: write_sysreg_s(val, SYS_ZCR_EL12); break; 1120 default: return false; 1121 } 1122 1123 return true; 1124 } 1125 1126 struct kvm_vm_stat { 1127 struct kvm_vm_stat_generic generic; 1128 }; 1129 1130 struct kvm_vcpu_stat { 1131 struct kvm_vcpu_stat_generic generic; 1132 u64 hvc_exit_stat; 1133 u64 wfe_exit_stat; 1134 u64 wfi_exit_stat; 1135 u64 mmio_exit_user; 1136 u64 mmio_exit_kernel; 1137 u64 signal_exits; 1138 u64 exits; 1139 }; 1140 1141 unsigned long kvm_arm_num_regs(struct kvm_vcpu *vcpu); 1142 int kvm_arm_copy_reg_indices(struct kvm_vcpu *vcpu, u64 __user *indices); 1143 int kvm_arm_get_reg(struct kvm_vcpu *vcpu, const struct kvm_one_reg *reg); 1144 int kvm_arm_set_reg(struct kvm_vcpu *vcpu, const struct kvm_one_reg *reg); 1145 1146 unsigned long kvm_arm_num_sys_reg_descs(struct kvm_vcpu *vcpu); 1147 int kvm_arm_copy_sys_reg_indices(struct kvm_vcpu *vcpu, u64 __user *uindices); 1148 1149 int __kvm_arm_vcpu_get_events(struct kvm_vcpu *vcpu, 1150 struct kvm_vcpu_events *events); 1151 1152 int __kvm_arm_vcpu_set_events(struct kvm_vcpu *vcpu, 1153 struct kvm_vcpu_events *events); 1154 1155 void kvm_arm_halt_guest(struct kvm *kvm); 1156 void kvm_arm_resume_guest(struct kvm *kvm); 1157 1158 #define vcpu_has_run_once(vcpu) (!!READ_ONCE((vcpu)->pid)) 1159 1160 #ifndef __KVM_NVHE_HYPERVISOR__ 1161 #define kvm_call_hyp_nvhe(f, ...) \ 1162 ({ \ 1163 struct arm_smccc_res res; \ 1164 \ 1165 arm_smccc_1_1_hvc(KVM_HOST_SMCCC_FUNC(f), \ 1166 ##__VA_ARGS__, &res); \ 1167 WARN_ON(res.a0 != SMCCC_RET_SUCCESS); \ 1168 \ 1169 res.a1; \ 1170 }) 1171 1172 /* 1173 * The couple of isb() below are there to guarantee the same behaviour 1174 * on VHE as on !VHE, where the eret to EL1 acts as a context 1175 * synchronization event. 1176 */ 1177 #define kvm_call_hyp(f, ...) \ 1178 do { \ 1179 if (has_vhe()) { \ 1180 f(__VA_ARGS__); \ 1181 isb(); \ 1182 } else { \ 1183 kvm_call_hyp_nvhe(f, ##__VA_ARGS__); \ 1184 } \ 1185 } while(0) 1186 1187 #define kvm_call_hyp_ret(f, ...) \ 1188 ({ \ 1189 typeof(f(__VA_ARGS__)) ret; \ 1190 \ 1191 if (has_vhe()) { \ 1192 ret = f(__VA_ARGS__); \ 1193 isb(); \ 1194 } else { \ 1195 ret = kvm_call_hyp_nvhe(f, ##__VA_ARGS__); \ 1196 } \ 1197 \ 1198 ret; \ 1199 }) 1200 #else /* __KVM_NVHE_HYPERVISOR__ */ 1201 #define kvm_call_hyp(f, ...) f(__VA_ARGS__) 1202 #define kvm_call_hyp_ret(f, ...) f(__VA_ARGS__) 1203 #define kvm_call_hyp_nvhe(f, ...) f(__VA_ARGS__) 1204 #endif /* __KVM_NVHE_HYPERVISOR__ */ 1205 1206 int handle_exit(struct kvm_vcpu *vcpu, int exception_index); 1207 void handle_exit_early(struct kvm_vcpu *vcpu, int exception_index); 1208 1209 int kvm_handle_cp14_load_store(struct kvm_vcpu *vcpu); 1210 int kvm_handle_cp14_32(struct kvm_vcpu *vcpu); 1211 int kvm_handle_cp14_64(struct kvm_vcpu *vcpu); 1212 int kvm_handle_cp15_32(struct kvm_vcpu *vcpu); 1213 int kvm_handle_cp15_64(struct kvm_vcpu *vcpu); 1214 int kvm_handle_sys_reg(struct kvm_vcpu *vcpu); 1215 int kvm_handle_cp10_id(struct kvm_vcpu *vcpu); 1216 1217 void kvm_sys_regs_create_debugfs(struct kvm *kvm); 1218 void kvm_reset_sys_regs(struct kvm_vcpu *vcpu); 1219 1220 int __init kvm_sys_reg_table_init(void); 1221 struct sys_reg_desc; 1222 int __init populate_sysreg_config(const struct sys_reg_desc *sr, 1223 unsigned int idx); 1224 int __init populate_nv_trap_config(void); 1225 1226 bool lock_all_vcpus(struct kvm *kvm); 1227 void unlock_all_vcpus(struct kvm *kvm); 1228 1229 void kvm_calculate_traps(struct kvm_vcpu *vcpu); 1230 1231 /* MMIO helpers */ 1232 void kvm_mmio_write_buf(void *buf, unsigned int len, unsigned long data); 1233 unsigned long kvm_mmio_read_buf(const void *buf, unsigned int len); 1234 1235 int kvm_handle_mmio_return(struct kvm_vcpu *vcpu); 1236 int io_mem_abort(struct kvm_vcpu *vcpu, phys_addr_t fault_ipa); 1237 1238 /* 1239 * Returns true if a Performance Monitoring Interrupt (PMI), a.k.a. perf event, 1240 * arrived in guest context. For arm64, any event that arrives while a vCPU is 1241 * loaded is considered to be "in guest". 1242 */ 1243 static inline bool kvm_arch_pmi_in_guest(struct kvm_vcpu *vcpu) 1244 { 1245 return IS_ENABLED(CONFIG_GUEST_PERF_EVENTS) && !!vcpu; 1246 } 1247 1248 long kvm_hypercall_pv_features(struct kvm_vcpu *vcpu); 1249 gpa_t kvm_init_stolen_time(struct kvm_vcpu *vcpu); 1250 void kvm_update_stolen_time(struct kvm_vcpu *vcpu); 1251 1252 bool kvm_arm_pvtime_supported(void); 1253 int kvm_arm_pvtime_set_attr(struct kvm_vcpu *vcpu, 1254 struct kvm_device_attr *attr); 1255 int kvm_arm_pvtime_get_attr(struct kvm_vcpu *vcpu, 1256 struct kvm_device_attr *attr); 1257 int kvm_arm_pvtime_has_attr(struct kvm_vcpu *vcpu, 1258 struct kvm_device_attr *attr); 1259 1260 extern unsigned int __ro_after_init kvm_arm_vmid_bits; 1261 int __init kvm_arm_vmid_alloc_init(void); 1262 void __init kvm_arm_vmid_alloc_free(void); 1263 bool kvm_arm_vmid_update(struct kvm_vmid *kvm_vmid); 1264 void kvm_arm_vmid_clear_active(void); 1265 1266 static inline void kvm_arm_pvtime_vcpu_init(struct kvm_vcpu_arch *vcpu_arch) 1267 { 1268 vcpu_arch->steal.base = INVALID_GPA; 1269 } 1270 1271 static inline bool kvm_arm_is_pvtime_enabled(struct kvm_vcpu_arch *vcpu_arch) 1272 { 1273 return (vcpu_arch->steal.base != INVALID_GPA); 1274 } 1275 1276 void kvm_set_sei_esr(struct kvm_vcpu *vcpu, u64 syndrome); 1277 1278 struct kvm_vcpu *kvm_mpidr_to_vcpu(struct kvm *kvm, unsigned long mpidr); 1279 1280 DECLARE_KVM_HYP_PER_CPU(struct kvm_host_data, kvm_host_data); 1281 1282 /* 1283 * How we access per-CPU host data depends on the where we access it from, 1284 * and the mode we're in: 1285 * 1286 * - VHE and nVHE hypervisor bits use their locally defined instance 1287 * 1288 * - the rest of the kernel use either the VHE or nVHE one, depending on 1289 * the mode we're running in. 1290 * 1291 * Unless we're in protected mode, fully deprivileged, and the nVHE 1292 * per-CPU stuff is exclusively accessible to the protected EL2 code. 1293 * In this case, the EL1 code uses the *VHE* data as its private state 1294 * (which makes sense in a way as there shouldn't be any shared state 1295 * between the host and the hypervisor). 1296 * 1297 * Yes, this is all totally trivial. Shoot me now. 1298 */ 1299 #if defined(__KVM_NVHE_HYPERVISOR__) || defined(__KVM_VHE_HYPERVISOR__) 1300 #define host_data_ptr(f) (&this_cpu_ptr(&kvm_host_data)->f) 1301 #else 1302 #define host_data_ptr(f) \ 1303 (static_branch_unlikely(&kvm_protected_mode_initialized) ? \ 1304 &this_cpu_ptr(&kvm_host_data)->f : \ 1305 &this_cpu_ptr_hyp_sym(kvm_host_data)->f) 1306 #endif 1307 1308 #define host_data_test_flag(flag) \ 1309 (test_bit(KVM_HOST_DATA_FLAG_##flag, host_data_ptr(flags))) 1310 #define host_data_set_flag(flag) \ 1311 set_bit(KVM_HOST_DATA_FLAG_##flag, host_data_ptr(flags)) 1312 #define host_data_clear_flag(flag) \ 1313 clear_bit(KVM_HOST_DATA_FLAG_##flag, host_data_ptr(flags)) 1314 1315 /* Check whether the FP regs are owned by the guest */ 1316 static inline bool guest_owns_fp_regs(void) 1317 { 1318 return *host_data_ptr(fp_owner) == FP_STATE_GUEST_OWNED; 1319 } 1320 1321 /* Check whether the FP regs are owned by the host */ 1322 static inline bool host_owns_fp_regs(void) 1323 { 1324 return *host_data_ptr(fp_owner) == FP_STATE_HOST_OWNED; 1325 } 1326 1327 static inline void kvm_init_host_cpu_context(struct kvm_cpu_context *cpu_ctxt) 1328 { 1329 /* The host's MPIDR is immutable, so let's set it up at boot time */ 1330 ctxt_sys_reg(cpu_ctxt, MPIDR_EL1) = read_cpuid_mpidr(); 1331 } 1332 1333 static inline bool kvm_system_needs_idmapped_vectors(void) 1334 { 1335 return cpus_have_final_cap(ARM64_SPECTRE_V3A); 1336 } 1337 1338 static inline void kvm_arch_sync_events(struct kvm *kvm) {} 1339 1340 void kvm_init_host_debug_data(void); 1341 void kvm_vcpu_load_debug(struct kvm_vcpu *vcpu); 1342 void kvm_vcpu_put_debug(struct kvm_vcpu *vcpu); 1343 void kvm_debug_set_guest_ownership(struct kvm_vcpu *vcpu); 1344 void kvm_debug_handle_oslar(struct kvm_vcpu *vcpu, u64 val); 1345 1346 #define kvm_vcpu_os_lock_enabled(vcpu) \ 1347 (!!(__vcpu_sys_reg(vcpu, OSLSR_EL1) & OSLSR_EL1_OSLK)) 1348 1349 #define kvm_debug_regs_in_use(vcpu) \ 1350 ((vcpu)->arch.debug_owner != VCPU_DEBUG_FREE) 1351 #define kvm_host_owns_debug_regs(vcpu) \ 1352 ((vcpu)->arch.debug_owner == VCPU_DEBUG_HOST_OWNED) 1353 #define kvm_guest_owns_debug_regs(vcpu) \ 1354 ((vcpu)->arch.debug_owner == VCPU_DEBUG_GUEST_OWNED) 1355 1356 int kvm_arm_vcpu_arch_set_attr(struct kvm_vcpu *vcpu, 1357 struct kvm_device_attr *attr); 1358 int kvm_arm_vcpu_arch_get_attr(struct kvm_vcpu *vcpu, 1359 struct kvm_device_attr *attr); 1360 int kvm_arm_vcpu_arch_has_attr(struct kvm_vcpu *vcpu, 1361 struct kvm_device_attr *attr); 1362 1363 int kvm_vm_ioctl_mte_copy_tags(struct kvm *kvm, 1364 struct kvm_arm_copy_mte_tags *copy_tags); 1365 int kvm_vm_ioctl_set_counter_offset(struct kvm *kvm, 1366 struct kvm_arm_counter_offset *offset); 1367 int kvm_vm_ioctl_get_reg_writable_masks(struct kvm *kvm, 1368 struct reg_mask_range *range); 1369 1370 /* Guest/host FPSIMD coordination helpers */ 1371 int kvm_arch_vcpu_run_map_fp(struct kvm_vcpu *vcpu); 1372 void kvm_arch_vcpu_load_fp(struct kvm_vcpu *vcpu); 1373 void kvm_arch_vcpu_ctxflush_fp(struct kvm_vcpu *vcpu); 1374 void kvm_arch_vcpu_ctxsync_fp(struct kvm_vcpu *vcpu); 1375 void kvm_arch_vcpu_put_fp(struct kvm_vcpu *vcpu); 1376 1377 static inline bool kvm_pmu_counter_deferred(struct perf_event_attr *attr) 1378 { 1379 return (!has_vhe() && attr->exclude_host); 1380 } 1381 1382 #ifdef CONFIG_KVM 1383 void kvm_set_pmu_events(u64 set, struct perf_event_attr *attr); 1384 void kvm_clr_pmu_events(u64 clr); 1385 bool kvm_set_pmuserenr(u64 val); 1386 void kvm_enable_trbe(void); 1387 void kvm_disable_trbe(void); 1388 void kvm_tracing_set_el1_configuration(u64 trfcr_while_in_guest); 1389 #else 1390 static inline void kvm_set_pmu_events(u64 set, struct perf_event_attr *attr) {} 1391 static inline void kvm_clr_pmu_events(u64 clr) {} 1392 static inline bool kvm_set_pmuserenr(u64 val) 1393 { 1394 return false; 1395 } 1396 static inline void kvm_enable_trbe(void) {} 1397 static inline void kvm_disable_trbe(void) {} 1398 static inline void kvm_tracing_set_el1_configuration(u64 trfcr_while_in_guest) {} 1399 #endif 1400 1401 void kvm_vcpu_load_vhe(struct kvm_vcpu *vcpu); 1402 void kvm_vcpu_put_vhe(struct kvm_vcpu *vcpu); 1403 1404 int __init kvm_set_ipa_limit(void); 1405 u32 kvm_get_pa_bits(struct kvm *kvm); 1406 1407 #define __KVM_HAVE_ARCH_VM_ALLOC 1408 struct kvm *kvm_arch_alloc_vm(void); 1409 1410 #define __KVM_HAVE_ARCH_FLUSH_REMOTE_TLBS 1411 1412 #define __KVM_HAVE_ARCH_FLUSH_REMOTE_TLBS_RANGE 1413 1414 #define kvm_vm_is_protected(kvm) (is_protected_kvm_enabled() && (kvm)->arch.pkvm.enabled) 1415 1416 #define vcpu_is_protected(vcpu) kvm_vm_is_protected((vcpu)->kvm) 1417 1418 int kvm_arm_vcpu_finalize(struct kvm_vcpu *vcpu, int feature); 1419 bool kvm_arm_vcpu_is_finalized(struct kvm_vcpu *vcpu); 1420 1421 #define kvm_arm_vcpu_sve_finalized(vcpu) vcpu_get_flag(vcpu, VCPU_SVE_FINALIZED) 1422 1423 #define kvm_has_mte(kvm) \ 1424 (system_supports_mte() && \ 1425 test_bit(KVM_ARCH_FLAG_MTE_ENABLED, &(kvm)->arch.flags)) 1426 1427 #define kvm_supports_32bit_el0() \ 1428 (system_supports_32bit_el0() && \ 1429 !static_branch_unlikely(&arm64_mismatched_32bit_el0)) 1430 1431 #define kvm_vm_has_ran_once(kvm) \ 1432 (test_bit(KVM_ARCH_FLAG_HAS_RAN_ONCE, &(kvm)->arch.flags)) 1433 1434 static inline bool __vcpu_has_feature(const struct kvm_arch *ka, int feature) 1435 { 1436 return test_bit(feature, ka->vcpu_features); 1437 } 1438 1439 #define kvm_vcpu_has_feature(k, f) __vcpu_has_feature(&(k)->arch, (f)) 1440 #define vcpu_has_feature(v, f) __vcpu_has_feature(&(v)->kvm->arch, (f)) 1441 1442 #define kvm_vcpu_initialized(v) vcpu_get_flag(vcpu, VCPU_INITIALIZED) 1443 1444 int kvm_trng_call(struct kvm_vcpu *vcpu); 1445 #ifdef CONFIG_KVM 1446 extern phys_addr_t hyp_mem_base; 1447 extern phys_addr_t hyp_mem_size; 1448 void __init kvm_hyp_reserve(void); 1449 #else 1450 static inline void kvm_hyp_reserve(void) { } 1451 #endif 1452 1453 void kvm_arm_vcpu_power_off(struct kvm_vcpu *vcpu); 1454 bool kvm_arm_vcpu_stopped(struct kvm_vcpu *vcpu); 1455 1456 static inline u64 *__vm_id_reg(struct kvm_arch *ka, u32 reg) 1457 { 1458 switch (reg) { 1459 case sys_reg(3, 0, 0, 1, 0) ... sys_reg(3, 0, 0, 7, 7): 1460 return &ka->id_regs[IDREG_IDX(reg)]; 1461 case SYS_CTR_EL0: 1462 return &ka->ctr_el0; 1463 default: 1464 WARN_ON_ONCE(1); 1465 return NULL; 1466 } 1467 } 1468 1469 #define kvm_read_vm_id_reg(kvm, reg) \ 1470 ({ u64 __val = *__vm_id_reg(&(kvm)->arch, reg); __val; }) 1471 1472 void kvm_set_vm_id_reg(struct kvm *kvm, u32 reg, u64 val); 1473 1474 #define __expand_field_sign_unsigned(id, fld, val) \ 1475 ((u64)SYS_FIELD_VALUE(id, fld, val)) 1476 1477 #define __expand_field_sign_signed(id, fld, val) \ 1478 ({ \ 1479 u64 __val = SYS_FIELD_VALUE(id, fld, val); \ 1480 sign_extend64(__val, id##_##fld##_WIDTH - 1); \ 1481 }) 1482 1483 #define get_idreg_field_unsigned(kvm, id, fld) \ 1484 ({ \ 1485 u64 __val = kvm_read_vm_id_reg((kvm), SYS_##id); \ 1486 FIELD_GET(id##_##fld##_MASK, __val); \ 1487 }) 1488 1489 #define get_idreg_field_signed(kvm, id, fld) \ 1490 ({ \ 1491 u64 __val = get_idreg_field_unsigned(kvm, id, fld); \ 1492 sign_extend64(__val, id##_##fld##_WIDTH - 1); \ 1493 }) 1494 1495 #define get_idreg_field_enum(kvm, id, fld) \ 1496 get_idreg_field_unsigned(kvm, id, fld) 1497 1498 #define kvm_cmp_feat_signed(kvm, id, fld, op, limit) \ 1499 (get_idreg_field_signed((kvm), id, fld) op __expand_field_sign_signed(id, fld, limit)) 1500 1501 #define kvm_cmp_feat_unsigned(kvm, id, fld, op, limit) \ 1502 (get_idreg_field_unsigned((kvm), id, fld) op __expand_field_sign_unsigned(id, fld, limit)) 1503 1504 #define kvm_cmp_feat(kvm, id, fld, op, limit) \ 1505 (id##_##fld##_SIGNED ? \ 1506 kvm_cmp_feat_signed(kvm, id, fld, op, limit) : \ 1507 kvm_cmp_feat_unsigned(kvm, id, fld, op, limit)) 1508 1509 #define kvm_has_feat(kvm, id, fld, limit) \ 1510 kvm_cmp_feat(kvm, id, fld, >=, limit) 1511 1512 #define kvm_has_feat_enum(kvm, id, fld, val) \ 1513 kvm_cmp_feat_unsigned(kvm, id, fld, ==, val) 1514 1515 #define kvm_has_feat_range(kvm, id, fld, min, max) \ 1516 (kvm_cmp_feat(kvm, id, fld, >=, min) && \ 1517 kvm_cmp_feat(kvm, id, fld, <=, max)) 1518 1519 /* Check for a given level of PAuth support */ 1520 #define kvm_has_pauth(k, l) \ 1521 ({ \ 1522 bool pa, pi, pa3; \ 1523 \ 1524 pa = kvm_has_feat((k), ID_AA64ISAR1_EL1, APA, l); \ 1525 pa &= kvm_has_feat((k), ID_AA64ISAR1_EL1, GPA, IMP); \ 1526 pi = kvm_has_feat((k), ID_AA64ISAR1_EL1, API, l); \ 1527 pi &= kvm_has_feat((k), ID_AA64ISAR1_EL1, GPI, IMP); \ 1528 pa3 = kvm_has_feat((k), ID_AA64ISAR2_EL1, APA3, l); \ 1529 pa3 &= kvm_has_feat((k), ID_AA64ISAR2_EL1, GPA3, IMP); \ 1530 \ 1531 (pa + pi + pa3) == 1; \ 1532 }) 1533 1534 #define kvm_has_fpmr(k) \ 1535 (system_supports_fpmr() && \ 1536 kvm_has_feat((k), ID_AA64PFR2_EL1, FPMR, IMP)) 1537 1538 #define kvm_has_tcr2(k) \ 1539 (kvm_has_feat((k), ID_AA64MMFR3_EL1, TCRX, IMP)) 1540 1541 #define kvm_has_s1pie(k) \ 1542 (kvm_has_feat((k), ID_AA64MMFR3_EL1, S1PIE, IMP)) 1543 1544 #define kvm_has_s1poe(k) \ 1545 (kvm_has_feat((k), ID_AA64MMFR3_EL1, S1POE, IMP)) 1546 1547 #endif /* __ARM64_KVM_HOST_H__ */ 1548