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 55 #define KVM_DIRTY_LOG_MANUAL_CAPS (KVM_DIRTY_LOG_MANUAL_PROTECT_ENABLE | \ 56 KVM_DIRTY_LOG_INITIALLY_SET) 57 58 #define KVM_HAVE_MMU_RWLOCK 59 60 /* 61 * Mode of operation configurable with kvm-arm.mode early param. 62 * See Documentation/admin-guide/kernel-parameters.txt for more information. 63 */ 64 enum kvm_mode { 65 KVM_MODE_DEFAULT, 66 KVM_MODE_PROTECTED, 67 KVM_MODE_NV, 68 KVM_MODE_NONE, 69 }; 70 #ifdef CONFIG_KVM 71 enum kvm_mode kvm_get_mode(void); 72 #else 73 static inline enum kvm_mode kvm_get_mode(void) { return KVM_MODE_NONE; }; 74 #endif 75 76 DECLARE_STATIC_KEY_FALSE(userspace_irqchip_in_use); 77 78 extern unsigned int __ro_after_init kvm_sve_max_vl; 79 extern unsigned int __ro_after_init kvm_host_sve_max_vl; 80 int __init kvm_arm_init_sve(void); 81 82 u32 __attribute_const__ kvm_target_cpu(void); 83 void kvm_reset_vcpu(struct kvm_vcpu *vcpu); 84 void kvm_arm_vcpu_destroy(struct kvm_vcpu *vcpu); 85 86 struct kvm_hyp_memcache { 87 phys_addr_t head; 88 unsigned long nr_pages; 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); 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 * 0: Nobody is currently using this, check vttbr for validity 216 * >0: Somebody is actively using this. 217 */ 218 atomic_t refcnt; 219 }; 220 221 struct kvm_arch_memory_slot { 222 }; 223 224 /** 225 * struct kvm_smccc_features: Descriptor of the hypercall services exposed to the guests 226 * 227 * @std_bmap: Bitmap of standard secure service calls 228 * @std_hyp_bmap: Bitmap of standard hypervisor service calls 229 * @vendor_hyp_bmap: Bitmap of vendor specific hypervisor service calls 230 */ 231 struct kvm_smccc_features { 232 unsigned long std_bmap; 233 unsigned long std_hyp_bmap; 234 unsigned long vendor_hyp_bmap; 235 }; 236 237 typedef unsigned int pkvm_handle_t; 238 239 struct kvm_protected_vm { 240 pkvm_handle_t handle; 241 struct kvm_hyp_memcache teardown_mc; 242 bool enabled; 243 }; 244 245 struct kvm_mpidr_data { 246 u64 mpidr_mask; 247 DECLARE_FLEX_ARRAY(u16, cmpidr_to_idx); 248 }; 249 250 static inline u16 kvm_mpidr_index(struct kvm_mpidr_data *data, u64 mpidr) 251 { 252 unsigned long index = 0, mask = data->mpidr_mask; 253 unsigned long aff = mpidr & MPIDR_HWID_BITMASK; 254 255 bitmap_gather(&index, &aff, &mask, fls(mask)); 256 257 return index; 258 } 259 260 struct kvm_sysreg_masks; 261 262 enum fgt_group_id { 263 __NO_FGT_GROUP__, 264 HFGxTR_GROUP, 265 HDFGRTR_GROUP, 266 HDFGWTR_GROUP = HDFGRTR_GROUP, 267 HFGITR_GROUP, 268 HAFGRTR_GROUP, 269 270 /* Must be last */ 271 __NR_FGT_GROUP_IDS__ 272 }; 273 274 struct kvm_arch { 275 struct kvm_s2_mmu mmu; 276 277 /* 278 * Fine-Grained UNDEF, mimicking the FGT layout defined by the 279 * architecture. We track them globally, as we present the 280 * same feature-set to all vcpus. 281 * 282 * Index 0 is currently spare. 283 */ 284 u64 fgu[__NR_FGT_GROUP_IDS__]; 285 286 /* 287 * Stage 2 paging state for VMs with nested S2 using a virtual 288 * VMID. 289 */ 290 struct kvm_s2_mmu *nested_mmus; 291 size_t nested_mmus_size; 292 int nested_mmus_next; 293 294 /* Interrupt controller */ 295 struct vgic_dist vgic; 296 297 /* Timers */ 298 struct arch_timer_vm_data timer_data; 299 300 /* Mandated version of PSCI */ 301 u32 psci_version; 302 303 /* Protects VM-scoped configuration data */ 304 struct mutex config_lock; 305 306 /* 307 * If we encounter a data abort without valid instruction syndrome 308 * information, report this to user space. User space can (and 309 * should) opt in to this feature if KVM_CAP_ARM_NISV_TO_USER is 310 * supported. 311 */ 312 #define KVM_ARCH_FLAG_RETURN_NISV_IO_ABORT_TO_USER 0 313 /* Memory Tagging Extension enabled for the guest */ 314 #define KVM_ARCH_FLAG_MTE_ENABLED 1 315 /* At least one vCPU has ran in the VM */ 316 #define KVM_ARCH_FLAG_HAS_RAN_ONCE 2 317 /* The vCPU feature set for the VM is configured */ 318 #define KVM_ARCH_FLAG_VCPU_FEATURES_CONFIGURED 3 319 /* PSCI SYSTEM_SUSPEND enabled for the guest */ 320 #define KVM_ARCH_FLAG_SYSTEM_SUSPEND_ENABLED 4 321 /* VM counter offset */ 322 #define KVM_ARCH_FLAG_VM_COUNTER_OFFSET 5 323 /* Timer PPIs made immutable */ 324 #define KVM_ARCH_FLAG_TIMER_PPIS_IMMUTABLE 6 325 /* Initial ID reg values loaded */ 326 #define KVM_ARCH_FLAG_ID_REGS_INITIALIZED 7 327 /* Fine-Grained UNDEF initialised */ 328 #define KVM_ARCH_FLAG_FGU_INITIALIZED 8 329 unsigned long flags; 330 331 /* VM-wide vCPU feature set */ 332 DECLARE_BITMAP(vcpu_features, KVM_VCPU_MAX_FEATURES); 333 334 /* MPIDR to vcpu index mapping, optional */ 335 struct kvm_mpidr_data *mpidr_data; 336 337 /* 338 * VM-wide PMU filter, implemented as a bitmap and big enough for 339 * up to 2^10 events (ARMv8.0) or 2^16 events (ARMv8.1+). 340 */ 341 unsigned long *pmu_filter; 342 struct arm_pmu *arm_pmu; 343 344 cpumask_var_t supported_cpus; 345 346 /* PMCR_EL0.N value for the guest */ 347 u8 pmcr_n; 348 349 /* Iterator for idreg debugfs */ 350 u8 idreg_debugfs_iter; 351 352 /* Hypercall features firmware registers' descriptor */ 353 struct kvm_smccc_features smccc_feat; 354 struct maple_tree smccc_filter; 355 356 /* 357 * Emulated CPU ID registers per VM 358 * (Op0, Op1, CRn, CRm, Op2) of the ID registers to be saved in it 359 * is (3, 0, 0, crm, op2), where 1<=crm<8, 0<=op2<8. 360 * 361 * These emulated idregs are VM-wide, but accessed from the context of a vCPU. 362 * Atomic access to multiple idregs are guarded by kvm_arch.config_lock. 363 */ 364 #define IDREG_IDX(id) (((sys_reg_CRm(id) - 1) << 3) | sys_reg_Op2(id)) 365 #define KVM_ARM_ID_REG_NUM (IDREG_IDX(sys_reg(3, 0, 0, 7, 7)) + 1) 366 u64 id_regs[KVM_ARM_ID_REG_NUM]; 367 368 u64 ctr_el0; 369 370 /* Masks for VNCR-baked sysregs */ 371 struct kvm_sysreg_masks *sysreg_masks; 372 373 /* 374 * For an untrusted host VM, 'pkvm.handle' is used to lookup 375 * the associated pKVM instance in the hypervisor. 376 */ 377 struct kvm_protected_vm pkvm; 378 }; 379 380 struct kvm_vcpu_fault_info { 381 u64 esr_el2; /* Hyp Syndrom Register */ 382 u64 far_el2; /* Hyp Fault Address Register */ 383 u64 hpfar_el2; /* Hyp IPA Fault Address Register */ 384 u64 disr_el1; /* Deferred [SError] Status Register */ 385 }; 386 387 /* 388 * VNCR() just places the VNCR_capable registers in the enum after 389 * __VNCR_START__, and the value (after correction) to be an 8-byte offset 390 * from the VNCR base. As we don't require the enum to be otherwise ordered, 391 * we need the terrible hack below to ensure that we correctly size the 392 * sys_regs array, no matter what. 393 * 394 * The __MAX__ macro has been lifted from Sean Eron Anderson's wonderful 395 * treasure trove of bit hacks: 396 * https://graphics.stanford.edu/~seander/bithacks.html#IntegerMinOrMax 397 */ 398 #define __MAX__(x,y) ((x) ^ (((x) ^ (y)) & -((x) < (y)))) 399 #define VNCR(r) \ 400 __before_##r, \ 401 r = __VNCR_START__ + ((VNCR_ ## r) / 8), \ 402 __after_##r = __MAX__(__before_##r - 1, r) 403 404 enum vcpu_sysreg { 405 __INVALID_SYSREG__, /* 0 is reserved as an invalid value */ 406 MPIDR_EL1, /* MultiProcessor Affinity Register */ 407 CLIDR_EL1, /* Cache Level ID Register */ 408 CSSELR_EL1, /* Cache Size Selection Register */ 409 TPIDR_EL0, /* Thread ID, User R/W */ 410 TPIDRRO_EL0, /* Thread ID, User R/O */ 411 TPIDR_EL1, /* Thread ID, Privileged */ 412 CNTKCTL_EL1, /* Timer Control Register (EL1) */ 413 PAR_EL1, /* Physical Address Register */ 414 MDCCINT_EL1, /* Monitor Debug Comms Channel Interrupt Enable Reg */ 415 OSLSR_EL1, /* OS Lock Status Register */ 416 DISR_EL1, /* Deferred Interrupt Status Register */ 417 418 /* Performance Monitors Registers */ 419 PMCR_EL0, /* Control Register */ 420 PMSELR_EL0, /* Event Counter Selection Register */ 421 PMEVCNTR0_EL0, /* Event Counter Register (0-30) */ 422 PMEVCNTR30_EL0 = PMEVCNTR0_EL0 + 30, 423 PMCCNTR_EL0, /* Cycle Counter Register */ 424 PMEVTYPER0_EL0, /* Event Type Register (0-30) */ 425 PMEVTYPER30_EL0 = PMEVTYPER0_EL0 + 30, 426 PMCCFILTR_EL0, /* Cycle Count Filter Register */ 427 PMCNTENSET_EL0, /* Count Enable Set Register */ 428 PMINTENSET_EL1, /* Interrupt Enable Set Register */ 429 PMOVSSET_EL0, /* Overflow Flag Status Set Register */ 430 PMUSERENR_EL0, /* User Enable Register */ 431 432 /* Pointer Authentication Registers in a strict increasing order. */ 433 APIAKEYLO_EL1, 434 APIAKEYHI_EL1, 435 APIBKEYLO_EL1, 436 APIBKEYHI_EL1, 437 APDAKEYLO_EL1, 438 APDAKEYHI_EL1, 439 APDBKEYLO_EL1, 440 APDBKEYHI_EL1, 441 APGAKEYLO_EL1, 442 APGAKEYHI_EL1, 443 444 /* Memory Tagging Extension registers */ 445 RGSR_EL1, /* Random Allocation Tag Seed Register */ 446 GCR_EL1, /* Tag Control Register */ 447 TFSRE0_EL1, /* Tag Fault Status Register (EL0) */ 448 449 /* 32bit specific registers. */ 450 DACR32_EL2, /* Domain Access Control Register */ 451 IFSR32_EL2, /* Instruction Fault Status Register */ 452 FPEXC32_EL2, /* Floating-Point Exception Control Register */ 453 DBGVCR32_EL2, /* Debug Vector Catch Register */ 454 455 /* EL2 registers */ 456 SCTLR_EL2, /* System Control Register (EL2) */ 457 ACTLR_EL2, /* Auxiliary Control Register (EL2) */ 458 MDCR_EL2, /* Monitor Debug Configuration Register (EL2) */ 459 CPTR_EL2, /* Architectural Feature Trap Register (EL2) */ 460 HACR_EL2, /* Hypervisor Auxiliary Control Register */ 461 ZCR_EL2, /* SVE Control Register (EL2) */ 462 TTBR0_EL2, /* Translation Table Base Register 0 (EL2) */ 463 TTBR1_EL2, /* Translation Table Base Register 1 (EL2) */ 464 TCR_EL2, /* Translation Control Register (EL2) */ 465 SPSR_EL2, /* EL2 saved program status register */ 466 ELR_EL2, /* EL2 exception link register */ 467 AFSR0_EL2, /* Auxiliary Fault Status Register 0 (EL2) */ 468 AFSR1_EL2, /* Auxiliary Fault Status Register 1 (EL2) */ 469 ESR_EL2, /* Exception Syndrome Register (EL2) */ 470 FAR_EL2, /* Fault Address Register (EL2) */ 471 HPFAR_EL2, /* Hypervisor IPA Fault Address Register */ 472 MAIR_EL2, /* Memory Attribute Indirection Register (EL2) */ 473 AMAIR_EL2, /* Auxiliary Memory Attribute Indirection Register (EL2) */ 474 VBAR_EL2, /* Vector Base Address Register (EL2) */ 475 RVBAR_EL2, /* Reset Vector Base Address Register */ 476 CONTEXTIDR_EL2, /* Context ID Register (EL2) */ 477 CNTHCTL_EL2, /* Counter-timer Hypervisor Control register */ 478 SP_EL2, /* EL2 Stack Pointer */ 479 CNTHP_CTL_EL2, 480 CNTHP_CVAL_EL2, 481 CNTHV_CTL_EL2, 482 CNTHV_CVAL_EL2, 483 484 __VNCR_START__, /* Any VNCR-capable reg goes after this point */ 485 486 VNCR(SCTLR_EL1),/* System Control Register */ 487 VNCR(ACTLR_EL1),/* Auxiliary Control Register */ 488 VNCR(CPACR_EL1),/* Coprocessor Access Control */ 489 VNCR(ZCR_EL1), /* SVE Control */ 490 VNCR(TTBR0_EL1),/* Translation Table Base Register 0 */ 491 VNCR(TTBR1_EL1),/* Translation Table Base Register 1 */ 492 VNCR(TCR_EL1), /* Translation Control Register */ 493 VNCR(TCR2_EL1), /* Extended Translation Control Register */ 494 VNCR(ESR_EL1), /* Exception Syndrome Register */ 495 VNCR(AFSR0_EL1),/* Auxiliary Fault Status Register 0 */ 496 VNCR(AFSR1_EL1),/* Auxiliary Fault Status Register 1 */ 497 VNCR(FAR_EL1), /* Fault Address Register */ 498 VNCR(MAIR_EL1), /* Memory Attribute Indirection Register */ 499 VNCR(VBAR_EL1), /* Vector Base Address Register */ 500 VNCR(CONTEXTIDR_EL1), /* Context ID Register */ 501 VNCR(AMAIR_EL1),/* Aux Memory Attribute Indirection Register */ 502 VNCR(MDSCR_EL1),/* Monitor Debug System Control Register */ 503 VNCR(ELR_EL1), 504 VNCR(SP_EL1), 505 VNCR(SPSR_EL1), 506 VNCR(TFSR_EL1), /* Tag Fault Status Register (EL1) */ 507 VNCR(VPIDR_EL2),/* Virtualization Processor ID Register */ 508 VNCR(VMPIDR_EL2),/* Virtualization Multiprocessor ID Register */ 509 VNCR(HCR_EL2), /* Hypervisor Configuration Register */ 510 VNCR(HSTR_EL2), /* Hypervisor System Trap Register */ 511 VNCR(VTTBR_EL2),/* Virtualization Translation Table Base Register */ 512 VNCR(VTCR_EL2), /* Virtualization Translation Control Register */ 513 VNCR(TPIDR_EL2),/* EL2 Software Thread ID Register */ 514 VNCR(HCRX_EL2), /* Extended Hypervisor Configuration Register */ 515 516 /* Permission Indirection Extension registers */ 517 VNCR(PIR_EL1), /* Permission Indirection Register 1 (EL1) */ 518 VNCR(PIRE0_EL1), /* Permission Indirection Register 0 (EL1) */ 519 520 VNCR(HFGRTR_EL2), 521 VNCR(HFGWTR_EL2), 522 VNCR(HFGITR_EL2), 523 VNCR(HDFGRTR_EL2), 524 VNCR(HDFGWTR_EL2), 525 VNCR(HAFGRTR_EL2), 526 527 VNCR(CNTVOFF_EL2), 528 VNCR(CNTV_CVAL_EL0), 529 VNCR(CNTV_CTL_EL0), 530 VNCR(CNTP_CVAL_EL0), 531 VNCR(CNTP_CTL_EL0), 532 533 NR_SYS_REGS /* Nothing after this line! */ 534 }; 535 536 struct kvm_sysreg_masks { 537 struct { 538 u64 res0; 539 u64 res1; 540 } mask[NR_SYS_REGS - __VNCR_START__]; 541 }; 542 543 struct kvm_cpu_context { 544 struct user_pt_regs regs; /* sp = sp_el0 */ 545 546 u64 spsr_abt; 547 u64 spsr_und; 548 u64 spsr_irq; 549 u64 spsr_fiq; 550 551 struct user_fpsimd_state fp_regs; 552 553 u64 sys_regs[NR_SYS_REGS]; 554 555 struct kvm_vcpu *__hyp_running_vcpu; 556 557 /* This pointer has to be 4kB aligned. */ 558 u64 *vncr_array; 559 }; 560 561 struct cpu_sve_state { 562 __u64 zcr_el1; 563 564 /* 565 * Ordering is important since __sve_save_state/__sve_restore_state 566 * relies on it. 567 */ 568 __u32 fpsr; 569 __u32 fpcr; 570 571 /* Must be SVE_VQ_BYTES (128 bit) aligned. */ 572 __u8 sve_regs[]; 573 }; 574 575 /* 576 * This structure is instantiated on a per-CPU basis, and contains 577 * data that is: 578 * 579 * - tied to a single physical CPU, and 580 * - either have a lifetime that does not extend past vcpu_put() 581 * - or is an invariant for the lifetime of the system 582 * 583 * Use host_data_ptr(field) as a way to access a pointer to such a 584 * field. 585 */ 586 struct kvm_host_data { 587 struct kvm_cpu_context host_ctxt; 588 589 /* 590 * All pointers in this union are hyp VA. 591 * sve_state is only used in pKVM and if system_supports_sve(). 592 */ 593 union { 594 struct user_fpsimd_state *fpsimd_state; 595 struct cpu_sve_state *sve_state; 596 }; 597 598 /* Ownership of the FP regs */ 599 enum { 600 FP_STATE_FREE, 601 FP_STATE_HOST_OWNED, 602 FP_STATE_GUEST_OWNED, 603 } fp_owner; 604 605 /* 606 * host_debug_state contains the host registers which are 607 * saved and restored during world switches. 608 */ 609 struct { 610 /* {Break,watch}point registers */ 611 struct kvm_guest_debug_arch regs; 612 /* Statistical profiling extension */ 613 u64 pmscr_el1; 614 /* Self-hosted trace */ 615 u64 trfcr_el1; 616 /* Values of trap registers for the host before guest entry. */ 617 u64 mdcr_el2; 618 } host_debug_state; 619 }; 620 621 struct kvm_host_psci_config { 622 /* PSCI version used by host. */ 623 u32 version; 624 u32 smccc_version; 625 626 /* Function IDs used by host if version is v0.1. */ 627 struct psci_0_1_function_ids function_ids_0_1; 628 629 bool psci_0_1_cpu_suspend_implemented; 630 bool psci_0_1_cpu_on_implemented; 631 bool psci_0_1_cpu_off_implemented; 632 bool psci_0_1_migrate_implemented; 633 }; 634 635 extern struct kvm_host_psci_config kvm_nvhe_sym(kvm_host_psci_config); 636 #define kvm_host_psci_config CHOOSE_NVHE_SYM(kvm_host_psci_config) 637 638 extern s64 kvm_nvhe_sym(hyp_physvirt_offset); 639 #define hyp_physvirt_offset CHOOSE_NVHE_SYM(hyp_physvirt_offset) 640 641 extern u64 kvm_nvhe_sym(hyp_cpu_logical_map)[NR_CPUS]; 642 #define hyp_cpu_logical_map CHOOSE_NVHE_SYM(hyp_cpu_logical_map) 643 644 struct vcpu_reset_state { 645 unsigned long pc; 646 unsigned long r0; 647 bool be; 648 bool reset; 649 }; 650 651 struct kvm_vcpu_arch { 652 struct kvm_cpu_context ctxt; 653 654 /* 655 * Guest floating point state 656 * 657 * The architecture has two main floating point extensions, 658 * the original FPSIMD and SVE. These have overlapping 659 * register views, with the FPSIMD V registers occupying the 660 * low 128 bits of the SVE Z registers. When the core 661 * floating point code saves the register state of a task it 662 * records which view it saved in fp_type. 663 */ 664 void *sve_state; 665 enum fp_type fp_type; 666 unsigned int sve_max_vl; 667 u64 svcr; 668 u64 fpmr; 669 670 /* Stage 2 paging state used by the hardware on next switch */ 671 struct kvm_s2_mmu *hw_mmu; 672 673 /* Values of trap registers for the guest. */ 674 u64 hcr_el2; 675 u64 hcrx_el2; 676 u64 mdcr_el2; 677 u64 cptr_el2; 678 679 /* Exception Information */ 680 struct kvm_vcpu_fault_info fault; 681 682 /* Configuration flags, set once and for all before the vcpu can run */ 683 u8 cflags; 684 685 /* Input flags to the hypervisor code, potentially cleared after use */ 686 u8 iflags; 687 688 /* State flags for kernel bookkeeping, unused by the hypervisor code */ 689 u8 sflags; 690 691 /* 692 * Don't run the guest (internal implementation need). 693 * 694 * Contrary to the flags above, this is set/cleared outside of 695 * a vcpu context, and thus cannot be mixed with the flags 696 * themselves (or the flag accesses need to be made atomic). 697 */ 698 bool pause; 699 700 /* 701 * We maintain more than a single set of debug registers to support 702 * debugging the guest from the host and to maintain separate host and 703 * guest state during world switches. vcpu_debug_state are the debug 704 * registers of the vcpu as the guest sees them. 705 * 706 * external_debug_state contains the debug values we want to debug the 707 * guest. This is set via the KVM_SET_GUEST_DEBUG ioctl. 708 * 709 * debug_ptr points to the set of debug registers that should be loaded 710 * onto the hardware when running the guest. 711 */ 712 struct kvm_guest_debug_arch *debug_ptr; 713 struct kvm_guest_debug_arch vcpu_debug_state; 714 struct kvm_guest_debug_arch external_debug_state; 715 716 /* VGIC state */ 717 struct vgic_cpu vgic_cpu; 718 struct arch_timer_cpu timer_cpu; 719 struct kvm_pmu pmu; 720 721 /* 722 * Guest registers we preserve during guest debugging. 723 * 724 * These shadow registers are updated by the kvm_handle_sys_reg 725 * trap handler if the guest accesses or updates them while we 726 * are using guest debug. 727 */ 728 struct { 729 u32 mdscr_el1; 730 bool pstate_ss; 731 } guest_debug_preserved; 732 733 /* vcpu power state */ 734 struct kvm_mp_state mp_state; 735 spinlock_t mp_state_lock; 736 737 /* Cache some mmu pages needed inside spinlock regions */ 738 struct kvm_mmu_memory_cache mmu_page_cache; 739 740 /* Virtual SError ESR to restore when HCR_EL2.VSE is set */ 741 u64 vsesr_el2; 742 743 /* Additional reset state */ 744 struct vcpu_reset_state reset_state; 745 746 /* Guest PV state */ 747 struct { 748 u64 last_steal; 749 gpa_t base; 750 } steal; 751 752 /* Per-vcpu CCSIDR override or NULL */ 753 u32 *ccsidr; 754 }; 755 756 /* 757 * Each 'flag' is composed of a comma-separated triplet: 758 * 759 * - the flag-set it belongs to in the vcpu->arch structure 760 * - the value for that flag 761 * - the mask for that flag 762 * 763 * __vcpu_single_flag() builds such a triplet for a single-bit flag. 764 * unpack_vcpu_flag() extract the flag value from the triplet for 765 * direct use outside of the flag accessors. 766 */ 767 #define __vcpu_single_flag(_set, _f) _set, (_f), (_f) 768 769 #define __unpack_flag(_set, _f, _m) _f 770 #define unpack_vcpu_flag(...) __unpack_flag(__VA_ARGS__) 771 772 #define __build_check_flag(v, flagset, f, m) \ 773 do { \ 774 typeof(v->arch.flagset) *_fset; \ 775 \ 776 /* Check that the flags fit in the mask */ \ 777 BUILD_BUG_ON(HWEIGHT(m) != HWEIGHT((f) | (m))); \ 778 /* Check that the flags fit in the type */ \ 779 BUILD_BUG_ON((sizeof(*_fset) * 8) <= __fls(m)); \ 780 } while (0) 781 782 #define __vcpu_get_flag(v, flagset, f, m) \ 783 ({ \ 784 __build_check_flag(v, flagset, f, m); \ 785 \ 786 READ_ONCE(v->arch.flagset) & (m); \ 787 }) 788 789 /* 790 * Note that the set/clear accessors must be preempt-safe in order to 791 * avoid nesting them with load/put which also manipulate flags... 792 */ 793 #ifdef __KVM_NVHE_HYPERVISOR__ 794 /* the nVHE hypervisor is always non-preemptible */ 795 #define __vcpu_flags_preempt_disable() 796 #define __vcpu_flags_preempt_enable() 797 #else 798 #define __vcpu_flags_preempt_disable() preempt_disable() 799 #define __vcpu_flags_preempt_enable() preempt_enable() 800 #endif 801 802 #define __vcpu_set_flag(v, flagset, f, m) \ 803 do { \ 804 typeof(v->arch.flagset) *fset; \ 805 \ 806 __build_check_flag(v, flagset, f, m); \ 807 \ 808 fset = &v->arch.flagset; \ 809 __vcpu_flags_preempt_disable(); \ 810 if (HWEIGHT(m) > 1) \ 811 *fset &= ~(m); \ 812 *fset |= (f); \ 813 __vcpu_flags_preempt_enable(); \ 814 } while (0) 815 816 #define __vcpu_clear_flag(v, flagset, f, m) \ 817 do { \ 818 typeof(v->arch.flagset) *fset; \ 819 \ 820 __build_check_flag(v, flagset, f, m); \ 821 \ 822 fset = &v->arch.flagset; \ 823 __vcpu_flags_preempt_disable(); \ 824 *fset &= ~(m); \ 825 __vcpu_flags_preempt_enable(); \ 826 } while (0) 827 828 #define vcpu_get_flag(v, ...) __vcpu_get_flag((v), __VA_ARGS__) 829 #define vcpu_set_flag(v, ...) __vcpu_set_flag((v), __VA_ARGS__) 830 #define vcpu_clear_flag(v, ...) __vcpu_clear_flag((v), __VA_ARGS__) 831 832 /* SVE exposed to guest */ 833 #define GUEST_HAS_SVE __vcpu_single_flag(cflags, BIT(0)) 834 /* SVE config completed */ 835 #define VCPU_SVE_FINALIZED __vcpu_single_flag(cflags, BIT(1)) 836 /* PTRAUTH exposed to guest */ 837 #define GUEST_HAS_PTRAUTH __vcpu_single_flag(cflags, BIT(2)) 838 /* KVM_ARM_VCPU_INIT completed */ 839 #define VCPU_INITIALIZED __vcpu_single_flag(cflags, BIT(3)) 840 841 /* Exception pending */ 842 #define PENDING_EXCEPTION __vcpu_single_flag(iflags, BIT(0)) 843 /* 844 * PC increment. Overlaps with EXCEPT_MASK on purpose so that it can't 845 * be set together with an exception... 846 */ 847 #define INCREMENT_PC __vcpu_single_flag(iflags, BIT(1)) 848 /* Target EL/MODE (not a single flag, but let's abuse the macro) */ 849 #define EXCEPT_MASK __vcpu_single_flag(iflags, GENMASK(3, 1)) 850 851 /* Helpers to encode exceptions with minimum fuss */ 852 #define __EXCEPT_MASK_VAL unpack_vcpu_flag(EXCEPT_MASK) 853 #define __EXCEPT_SHIFT __builtin_ctzl(__EXCEPT_MASK_VAL) 854 #define __vcpu_except_flags(_f) iflags, (_f << __EXCEPT_SHIFT), __EXCEPT_MASK_VAL 855 856 /* 857 * When PENDING_EXCEPTION is set, EXCEPT_MASK can take the following 858 * values: 859 * 860 * For AArch32 EL1: 861 */ 862 #define EXCEPT_AA32_UND __vcpu_except_flags(0) 863 #define EXCEPT_AA32_IABT __vcpu_except_flags(1) 864 #define EXCEPT_AA32_DABT __vcpu_except_flags(2) 865 /* For AArch64: */ 866 #define EXCEPT_AA64_EL1_SYNC __vcpu_except_flags(0) 867 #define EXCEPT_AA64_EL1_IRQ __vcpu_except_flags(1) 868 #define EXCEPT_AA64_EL1_FIQ __vcpu_except_flags(2) 869 #define EXCEPT_AA64_EL1_SERR __vcpu_except_flags(3) 870 /* For AArch64 with NV: */ 871 #define EXCEPT_AA64_EL2_SYNC __vcpu_except_flags(4) 872 #define EXCEPT_AA64_EL2_IRQ __vcpu_except_flags(5) 873 #define EXCEPT_AA64_EL2_FIQ __vcpu_except_flags(6) 874 #define EXCEPT_AA64_EL2_SERR __vcpu_except_flags(7) 875 /* Guest debug is live */ 876 #define DEBUG_DIRTY __vcpu_single_flag(iflags, BIT(4)) 877 /* Save SPE context if active */ 878 #define DEBUG_STATE_SAVE_SPE __vcpu_single_flag(iflags, BIT(5)) 879 /* Save TRBE context if active */ 880 #define DEBUG_STATE_SAVE_TRBE __vcpu_single_flag(iflags, BIT(6)) 881 882 /* SVE enabled for host EL0 */ 883 #define HOST_SVE_ENABLED __vcpu_single_flag(sflags, BIT(0)) 884 /* SME enabled for EL0 */ 885 #define HOST_SME_ENABLED __vcpu_single_flag(sflags, BIT(1)) 886 /* Physical CPU not in supported_cpus */ 887 #define ON_UNSUPPORTED_CPU __vcpu_single_flag(sflags, BIT(2)) 888 /* WFIT instruction trapped */ 889 #define IN_WFIT __vcpu_single_flag(sflags, BIT(3)) 890 /* vcpu system registers loaded on physical CPU */ 891 #define SYSREGS_ON_CPU __vcpu_single_flag(sflags, BIT(4)) 892 /* Software step state is Active-pending */ 893 #define DBG_SS_ACTIVE_PENDING __vcpu_single_flag(sflags, BIT(5)) 894 /* PMUSERENR for the guest EL0 is on physical CPU */ 895 #define PMUSERENR_ON_CPU __vcpu_single_flag(sflags, BIT(6)) 896 /* WFI instruction trapped */ 897 #define IN_WFI __vcpu_single_flag(sflags, BIT(7)) 898 899 900 /* Pointer to the vcpu's SVE FFR for sve_{save,load}_state() */ 901 #define vcpu_sve_pffr(vcpu) (kern_hyp_va((vcpu)->arch.sve_state) + \ 902 sve_ffr_offset((vcpu)->arch.sve_max_vl)) 903 904 #define vcpu_sve_max_vq(vcpu) sve_vq_from_vl((vcpu)->arch.sve_max_vl) 905 906 #define vcpu_sve_zcr_elx(vcpu) \ 907 (unlikely(is_hyp_ctxt(vcpu)) ? ZCR_EL2 : ZCR_EL1) 908 909 #define vcpu_sve_state_size(vcpu) ({ \ 910 size_t __size_ret; \ 911 unsigned int __vcpu_vq; \ 912 \ 913 if (WARN_ON(!sve_vl_valid((vcpu)->arch.sve_max_vl))) { \ 914 __size_ret = 0; \ 915 } else { \ 916 __vcpu_vq = vcpu_sve_max_vq(vcpu); \ 917 __size_ret = SVE_SIG_REGS_SIZE(__vcpu_vq); \ 918 } \ 919 \ 920 __size_ret; \ 921 }) 922 923 #define KVM_GUESTDBG_VALID_MASK (KVM_GUESTDBG_ENABLE | \ 924 KVM_GUESTDBG_USE_SW_BP | \ 925 KVM_GUESTDBG_USE_HW | \ 926 KVM_GUESTDBG_SINGLESTEP) 927 928 #define vcpu_has_sve(vcpu) (system_supports_sve() && \ 929 vcpu_get_flag(vcpu, GUEST_HAS_SVE)) 930 931 #ifdef CONFIG_ARM64_PTR_AUTH 932 #define vcpu_has_ptrauth(vcpu) \ 933 ((cpus_have_final_cap(ARM64_HAS_ADDRESS_AUTH) || \ 934 cpus_have_final_cap(ARM64_HAS_GENERIC_AUTH)) && \ 935 vcpu_get_flag(vcpu, GUEST_HAS_PTRAUTH)) 936 #else 937 #define vcpu_has_ptrauth(vcpu) false 938 #endif 939 940 #define vcpu_on_unsupported_cpu(vcpu) \ 941 vcpu_get_flag(vcpu, ON_UNSUPPORTED_CPU) 942 943 #define vcpu_set_on_unsupported_cpu(vcpu) \ 944 vcpu_set_flag(vcpu, ON_UNSUPPORTED_CPU) 945 946 #define vcpu_clear_on_unsupported_cpu(vcpu) \ 947 vcpu_clear_flag(vcpu, ON_UNSUPPORTED_CPU) 948 949 #define vcpu_gp_regs(v) (&(v)->arch.ctxt.regs) 950 951 /* 952 * Only use __vcpu_sys_reg/ctxt_sys_reg if you know you want the 953 * memory backed version of a register, and not the one most recently 954 * accessed by a running VCPU. For example, for userspace access or 955 * for system registers that are never context switched, but only 956 * emulated. 957 * 958 * Don't bother with VNCR-based accesses in the nVHE code, it has no 959 * business dealing with NV. 960 */ 961 static inline u64 *___ctxt_sys_reg(const struct kvm_cpu_context *ctxt, int r) 962 { 963 #if !defined (__KVM_NVHE_HYPERVISOR__) 964 if (unlikely(cpus_have_final_cap(ARM64_HAS_NESTED_VIRT) && 965 r >= __VNCR_START__ && ctxt->vncr_array)) 966 return &ctxt->vncr_array[r - __VNCR_START__]; 967 #endif 968 return (u64 *)&ctxt->sys_regs[r]; 969 } 970 971 #define __ctxt_sys_reg(c,r) \ 972 ({ \ 973 BUILD_BUG_ON(__builtin_constant_p(r) && \ 974 (r) >= NR_SYS_REGS); \ 975 ___ctxt_sys_reg(c, r); \ 976 }) 977 978 #define ctxt_sys_reg(c,r) (*__ctxt_sys_reg(c,r)) 979 980 u64 kvm_vcpu_sanitise_vncr_reg(const struct kvm_vcpu *, enum vcpu_sysreg); 981 #define __vcpu_sys_reg(v,r) \ 982 (*({ \ 983 const struct kvm_cpu_context *ctxt = &(v)->arch.ctxt; \ 984 u64 *__r = __ctxt_sys_reg(ctxt, (r)); \ 985 if (vcpu_has_nv((v)) && (r) >= __VNCR_START__) \ 986 *__r = kvm_vcpu_sanitise_vncr_reg((v), (r)); \ 987 __r; \ 988 })) 989 990 u64 vcpu_read_sys_reg(const struct kvm_vcpu *vcpu, int reg); 991 void vcpu_write_sys_reg(struct kvm_vcpu *vcpu, u64 val, int reg); 992 993 static inline bool __vcpu_read_sys_reg_from_cpu(int reg, u64 *val) 994 { 995 /* 996 * *** VHE ONLY *** 997 * 998 * System registers listed in the switch are not saved on every 999 * exit from the guest but are only saved on vcpu_put. 1000 * 1001 * Note that MPIDR_EL1 for the guest is set by KVM via VMPIDR_EL2 but 1002 * should never be listed below, because the guest cannot modify its 1003 * own MPIDR_EL1 and MPIDR_EL1 is accessed for VCPU A from VCPU B's 1004 * thread when emulating cross-VCPU communication. 1005 */ 1006 if (!has_vhe()) 1007 return false; 1008 1009 switch (reg) { 1010 case SCTLR_EL1: *val = read_sysreg_s(SYS_SCTLR_EL12); break; 1011 case CPACR_EL1: *val = read_sysreg_s(SYS_CPACR_EL12); break; 1012 case TTBR0_EL1: *val = read_sysreg_s(SYS_TTBR0_EL12); break; 1013 case TTBR1_EL1: *val = read_sysreg_s(SYS_TTBR1_EL12); break; 1014 case TCR_EL1: *val = read_sysreg_s(SYS_TCR_EL12); break; 1015 case ESR_EL1: *val = read_sysreg_s(SYS_ESR_EL12); break; 1016 case AFSR0_EL1: *val = read_sysreg_s(SYS_AFSR0_EL12); break; 1017 case AFSR1_EL1: *val = read_sysreg_s(SYS_AFSR1_EL12); break; 1018 case FAR_EL1: *val = read_sysreg_s(SYS_FAR_EL12); break; 1019 case MAIR_EL1: *val = read_sysreg_s(SYS_MAIR_EL12); break; 1020 case VBAR_EL1: *val = read_sysreg_s(SYS_VBAR_EL12); break; 1021 case CONTEXTIDR_EL1: *val = read_sysreg_s(SYS_CONTEXTIDR_EL12);break; 1022 case TPIDR_EL0: *val = read_sysreg_s(SYS_TPIDR_EL0); break; 1023 case TPIDRRO_EL0: *val = read_sysreg_s(SYS_TPIDRRO_EL0); break; 1024 case TPIDR_EL1: *val = read_sysreg_s(SYS_TPIDR_EL1); break; 1025 case AMAIR_EL1: *val = read_sysreg_s(SYS_AMAIR_EL12); break; 1026 case CNTKCTL_EL1: *val = read_sysreg_s(SYS_CNTKCTL_EL12); break; 1027 case ELR_EL1: *val = read_sysreg_s(SYS_ELR_EL12); break; 1028 case SPSR_EL1: *val = read_sysreg_s(SYS_SPSR_EL12); break; 1029 case PAR_EL1: *val = read_sysreg_par(); break; 1030 case DACR32_EL2: *val = read_sysreg_s(SYS_DACR32_EL2); break; 1031 case IFSR32_EL2: *val = read_sysreg_s(SYS_IFSR32_EL2); break; 1032 case DBGVCR32_EL2: *val = read_sysreg_s(SYS_DBGVCR32_EL2); break; 1033 case ZCR_EL1: *val = read_sysreg_s(SYS_ZCR_EL12); break; 1034 default: return false; 1035 } 1036 1037 return true; 1038 } 1039 1040 static inline bool __vcpu_write_sys_reg_to_cpu(u64 val, int reg) 1041 { 1042 /* 1043 * *** VHE ONLY *** 1044 * 1045 * System registers listed in the switch are not restored on every 1046 * entry to the guest but are only restored on vcpu_load. 1047 * 1048 * Note that MPIDR_EL1 for the guest is set by KVM via VMPIDR_EL2 but 1049 * should never be listed below, because the MPIDR should only be set 1050 * once, before running the VCPU, and never changed later. 1051 */ 1052 if (!has_vhe()) 1053 return false; 1054 1055 switch (reg) { 1056 case SCTLR_EL1: write_sysreg_s(val, SYS_SCTLR_EL12); break; 1057 case CPACR_EL1: write_sysreg_s(val, SYS_CPACR_EL12); break; 1058 case TTBR0_EL1: write_sysreg_s(val, SYS_TTBR0_EL12); break; 1059 case TTBR1_EL1: write_sysreg_s(val, SYS_TTBR1_EL12); break; 1060 case TCR_EL1: write_sysreg_s(val, SYS_TCR_EL12); break; 1061 case ESR_EL1: write_sysreg_s(val, SYS_ESR_EL12); break; 1062 case AFSR0_EL1: write_sysreg_s(val, SYS_AFSR0_EL12); break; 1063 case AFSR1_EL1: write_sysreg_s(val, SYS_AFSR1_EL12); break; 1064 case FAR_EL1: write_sysreg_s(val, SYS_FAR_EL12); break; 1065 case MAIR_EL1: write_sysreg_s(val, SYS_MAIR_EL12); break; 1066 case VBAR_EL1: write_sysreg_s(val, SYS_VBAR_EL12); break; 1067 case CONTEXTIDR_EL1: write_sysreg_s(val, SYS_CONTEXTIDR_EL12);break; 1068 case TPIDR_EL0: write_sysreg_s(val, SYS_TPIDR_EL0); break; 1069 case TPIDRRO_EL0: write_sysreg_s(val, SYS_TPIDRRO_EL0); break; 1070 case TPIDR_EL1: write_sysreg_s(val, SYS_TPIDR_EL1); break; 1071 case AMAIR_EL1: write_sysreg_s(val, SYS_AMAIR_EL12); break; 1072 case CNTKCTL_EL1: write_sysreg_s(val, SYS_CNTKCTL_EL12); break; 1073 case ELR_EL1: write_sysreg_s(val, SYS_ELR_EL12); break; 1074 case SPSR_EL1: write_sysreg_s(val, SYS_SPSR_EL12); break; 1075 case PAR_EL1: write_sysreg_s(val, SYS_PAR_EL1); break; 1076 case DACR32_EL2: write_sysreg_s(val, SYS_DACR32_EL2); break; 1077 case IFSR32_EL2: write_sysreg_s(val, SYS_IFSR32_EL2); break; 1078 case DBGVCR32_EL2: write_sysreg_s(val, SYS_DBGVCR32_EL2); break; 1079 case ZCR_EL1: write_sysreg_s(val, SYS_ZCR_EL12); break; 1080 default: return false; 1081 } 1082 1083 return true; 1084 } 1085 1086 struct kvm_vm_stat { 1087 struct kvm_vm_stat_generic generic; 1088 }; 1089 1090 struct kvm_vcpu_stat { 1091 struct kvm_vcpu_stat_generic generic; 1092 u64 hvc_exit_stat; 1093 u64 wfe_exit_stat; 1094 u64 wfi_exit_stat; 1095 u64 mmio_exit_user; 1096 u64 mmio_exit_kernel; 1097 u64 signal_exits; 1098 u64 exits; 1099 }; 1100 1101 unsigned long kvm_arm_num_regs(struct kvm_vcpu *vcpu); 1102 int kvm_arm_copy_reg_indices(struct kvm_vcpu *vcpu, u64 __user *indices); 1103 int kvm_arm_get_reg(struct kvm_vcpu *vcpu, const struct kvm_one_reg *reg); 1104 int kvm_arm_set_reg(struct kvm_vcpu *vcpu, const struct kvm_one_reg *reg); 1105 1106 unsigned long kvm_arm_num_sys_reg_descs(struct kvm_vcpu *vcpu); 1107 int kvm_arm_copy_sys_reg_indices(struct kvm_vcpu *vcpu, u64 __user *uindices); 1108 1109 int __kvm_arm_vcpu_get_events(struct kvm_vcpu *vcpu, 1110 struct kvm_vcpu_events *events); 1111 1112 int __kvm_arm_vcpu_set_events(struct kvm_vcpu *vcpu, 1113 struct kvm_vcpu_events *events); 1114 1115 void kvm_arm_halt_guest(struct kvm *kvm); 1116 void kvm_arm_resume_guest(struct kvm *kvm); 1117 1118 #define vcpu_has_run_once(vcpu) !!rcu_access_pointer((vcpu)->pid) 1119 1120 #ifndef __KVM_NVHE_HYPERVISOR__ 1121 #define kvm_call_hyp_nvhe(f, ...) \ 1122 ({ \ 1123 struct arm_smccc_res res; \ 1124 \ 1125 arm_smccc_1_1_hvc(KVM_HOST_SMCCC_FUNC(f), \ 1126 ##__VA_ARGS__, &res); \ 1127 WARN_ON(res.a0 != SMCCC_RET_SUCCESS); \ 1128 \ 1129 res.a1; \ 1130 }) 1131 1132 /* 1133 * The couple of isb() below are there to guarantee the same behaviour 1134 * on VHE as on !VHE, where the eret to EL1 acts as a context 1135 * synchronization event. 1136 */ 1137 #define kvm_call_hyp(f, ...) \ 1138 do { \ 1139 if (has_vhe()) { \ 1140 f(__VA_ARGS__); \ 1141 isb(); \ 1142 } else { \ 1143 kvm_call_hyp_nvhe(f, ##__VA_ARGS__); \ 1144 } \ 1145 } while(0) 1146 1147 #define kvm_call_hyp_ret(f, ...) \ 1148 ({ \ 1149 typeof(f(__VA_ARGS__)) ret; \ 1150 \ 1151 if (has_vhe()) { \ 1152 ret = f(__VA_ARGS__); \ 1153 isb(); \ 1154 } else { \ 1155 ret = kvm_call_hyp_nvhe(f, ##__VA_ARGS__); \ 1156 } \ 1157 \ 1158 ret; \ 1159 }) 1160 #else /* __KVM_NVHE_HYPERVISOR__ */ 1161 #define kvm_call_hyp(f, ...) f(__VA_ARGS__) 1162 #define kvm_call_hyp_ret(f, ...) f(__VA_ARGS__) 1163 #define kvm_call_hyp_nvhe(f, ...) f(__VA_ARGS__) 1164 #endif /* __KVM_NVHE_HYPERVISOR__ */ 1165 1166 int handle_exit(struct kvm_vcpu *vcpu, int exception_index); 1167 void handle_exit_early(struct kvm_vcpu *vcpu, int exception_index); 1168 1169 int kvm_handle_cp14_load_store(struct kvm_vcpu *vcpu); 1170 int kvm_handle_cp14_32(struct kvm_vcpu *vcpu); 1171 int kvm_handle_cp14_64(struct kvm_vcpu *vcpu); 1172 int kvm_handle_cp15_32(struct kvm_vcpu *vcpu); 1173 int kvm_handle_cp15_64(struct kvm_vcpu *vcpu); 1174 int kvm_handle_sys_reg(struct kvm_vcpu *vcpu); 1175 int kvm_handle_cp10_id(struct kvm_vcpu *vcpu); 1176 1177 void kvm_sys_regs_create_debugfs(struct kvm *kvm); 1178 void kvm_reset_sys_regs(struct kvm_vcpu *vcpu); 1179 1180 int __init kvm_sys_reg_table_init(void); 1181 struct sys_reg_desc; 1182 int __init populate_sysreg_config(const struct sys_reg_desc *sr, 1183 unsigned int idx); 1184 int __init populate_nv_trap_config(void); 1185 1186 bool lock_all_vcpus(struct kvm *kvm); 1187 void unlock_all_vcpus(struct kvm *kvm); 1188 1189 void kvm_calculate_traps(struct kvm_vcpu *vcpu); 1190 1191 /* MMIO helpers */ 1192 void kvm_mmio_write_buf(void *buf, unsigned int len, unsigned long data); 1193 unsigned long kvm_mmio_read_buf(const void *buf, unsigned int len); 1194 1195 int kvm_handle_mmio_return(struct kvm_vcpu *vcpu); 1196 int io_mem_abort(struct kvm_vcpu *vcpu, phys_addr_t fault_ipa); 1197 1198 /* 1199 * Returns true if a Performance Monitoring Interrupt (PMI), a.k.a. perf event, 1200 * arrived in guest context. For arm64, any event that arrives while a vCPU is 1201 * loaded is considered to be "in guest". 1202 */ 1203 static inline bool kvm_arch_pmi_in_guest(struct kvm_vcpu *vcpu) 1204 { 1205 return IS_ENABLED(CONFIG_GUEST_PERF_EVENTS) && !!vcpu; 1206 } 1207 1208 long kvm_hypercall_pv_features(struct kvm_vcpu *vcpu); 1209 gpa_t kvm_init_stolen_time(struct kvm_vcpu *vcpu); 1210 void kvm_update_stolen_time(struct kvm_vcpu *vcpu); 1211 1212 bool kvm_arm_pvtime_supported(void); 1213 int kvm_arm_pvtime_set_attr(struct kvm_vcpu *vcpu, 1214 struct kvm_device_attr *attr); 1215 int kvm_arm_pvtime_get_attr(struct kvm_vcpu *vcpu, 1216 struct kvm_device_attr *attr); 1217 int kvm_arm_pvtime_has_attr(struct kvm_vcpu *vcpu, 1218 struct kvm_device_attr *attr); 1219 1220 extern unsigned int __ro_after_init kvm_arm_vmid_bits; 1221 int __init kvm_arm_vmid_alloc_init(void); 1222 void __init kvm_arm_vmid_alloc_free(void); 1223 bool kvm_arm_vmid_update(struct kvm_vmid *kvm_vmid); 1224 void kvm_arm_vmid_clear_active(void); 1225 1226 static inline void kvm_arm_pvtime_vcpu_init(struct kvm_vcpu_arch *vcpu_arch) 1227 { 1228 vcpu_arch->steal.base = INVALID_GPA; 1229 } 1230 1231 static inline bool kvm_arm_is_pvtime_enabled(struct kvm_vcpu_arch *vcpu_arch) 1232 { 1233 return (vcpu_arch->steal.base != INVALID_GPA); 1234 } 1235 1236 void kvm_set_sei_esr(struct kvm_vcpu *vcpu, u64 syndrome); 1237 1238 struct kvm_vcpu *kvm_mpidr_to_vcpu(struct kvm *kvm, unsigned long mpidr); 1239 1240 DECLARE_KVM_HYP_PER_CPU(struct kvm_host_data, kvm_host_data); 1241 1242 /* 1243 * How we access per-CPU host data depends on the where we access it from, 1244 * and the mode we're in: 1245 * 1246 * - VHE and nVHE hypervisor bits use their locally defined instance 1247 * 1248 * - the rest of the kernel use either the VHE or nVHE one, depending on 1249 * the mode we're running in. 1250 * 1251 * Unless we're in protected mode, fully deprivileged, and the nVHE 1252 * per-CPU stuff is exclusively accessible to the protected EL2 code. 1253 * In this case, the EL1 code uses the *VHE* data as its private state 1254 * (which makes sense in a way as there shouldn't be any shared state 1255 * between the host and the hypervisor). 1256 * 1257 * Yes, this is all totally trivial. Shoot me now. 1258 */ 1259 #if defined(__KVM_NVHE_HYPERVISOR__) || defined(__KVM_VHE_HYPERVISOR__) 1260 #define host_data_ptr(f) (&this_cpu_ptr(&kvm_host_data)->f) 1261 #else 1262 #define host_data_ptr(f) \ 1263 (static_branch_unlikely(&kvm_protected_mode_initialized) ? \ 1264 &this_cpu_ptr(&kvm_host_data)->f : \ 1265 &this_cpu_ptr_hyp_sym(kvm_host_data)->f) 1266 #endif 1267 1268 /* Check whether the FP regs are owned by the guest */ 1269 static inline bool guest_owns_fp_regs(void) 1270 { 1271 return *host_data_ptr(fp_owner) == FP_STATE_GUEST_OWNED; 1272 } 1273 1274 /* Check whether the FP regs are owned by the host */ 1275 static inline bool host_owns_fp_regs(void) 1276 { 1277 return *host_data_ptr(fp_owner) == FP_STATE_HOST_OWNED; 1278 } 1279 1280 static inline void kvm_init_host_cpu_context(struct kvm_cpu_context *cpu_ctxt) 1281 { 1282 /* The host's MPIDR is immutable, so let's set it up at boot time */ 1283 ctxt_sys_reg(cpu_ctxt, MPIDR_EL1) = read_cpuid_mpidr(); 1284 } 1285 1286 static inline bool kvm_system_needs_idmapped_vectors(void) 1287 { 1288 return cpus_have_final_cap(ARM64_SPECTRE_V3A); 1289 } 1290 1291 static inline void kvm_arch_sync_events(struct kvm *kvm) {} 1292 1293 void kvm_arm_init_debug(void); 1294 void kvm_arm_vcpu_init_debug(struct kvm_vcpu *vcpu); 1295 void kvm_arm_setup_debug(struct kvm_vcpu *vcpu); 1296 void kvm_arm_clear_debug(struct kvm_vcpu *vcpu); 1297 void kvm_arm_reset_debug_ptr(struct kvm_vcpu *vcpu); 1298 1299 #define kvm_vcpu_os_lock_enabled(vcpu) \ 1300 (!!(__vcpu_sys_reg(vcpu, OSLSR_EL1) & OSLSR_EL1_OSLK)) 1301 1302 int kvm_arm_vcpu_arch_set_attr(struct kvm_vcpu *vcpu, 1303 struct kvm_device_attr *attr); 1304 int kvm_arm_vcpu_arch_get_attr(struct kvm_vcpu *vcpu, 1305 struct kvm_device_attr *attr); 1306 int kvm_arm_vcpu_arch_has_attr(struct kvm_vcpu *vcpu, 1307 struct kvm_device_attr *attr); 1308 1309 int kvm_vm_ioctl_mte_copy_tags(struct kvm *kvm, 1310 struct kvm_arm_copy_mte_tags *copy_tags); 1311 int kvm_vm_ioctl_set_counter_offset(struct kvm *kvm, 1312 struct kvm_arm_counter_offset *offset); 1313 int kvm_vm_ioctl_get_reg_writable_masks(struct kvm *kvm, 1314 struct reg_mask_range *range); 1315 1316 /* Guest/host FPSIMD coordination helpers */ 1317 int kvm_arch_vcpu_run_map_fp(struct kvm_vcpu *vcpu); 1318 void kvm_arch_vcpu_load_fp(struct kvm_vcpu *vcpu); 1319 void kvm_arch_vcpu_ctxflush_fp(struct kvm_vcpu *vcpu); 1320 void kvm_arch_vcpu_ctxsync_fp(struct kvm_vcpu *vcpu); 1321 void kvm_arch_vcpu_put_fp(struct kvm_vcpu *vcpu); 1322 1323 static inline bool kvm_pmu_counter_deferred(struct perf_event_attr *attr) 1324 { 1325 return (!has_vhe() && attr->exclude_host); 1326 } 1327 1328 /* Flags for host debug state */ 1329 void kvm_arch_vcpu_load_debug_state_flags(struct kvm_vcpu *vcpu); 1330 void kvm_arch_vcpu_put_debug_state_flags(struct kvm_vcpu *vcpu); 1331 1332 #ifdef CONFIG_KVM 1333 void kvm_set_pmu_events(u32 set, struct perf_event_attr *attr); 1334 void kvm_clr_pmu_events(u32 clr); 1335 bool kvm_set_pmuserenr(u64 val); 1336 #else 1337 static inline void kvm_set_pmu_events(u32 set, struct perf_event_attr *attr) {} 1338 static inline void kvm_clr_pmu_events(u32 clr) {} 1339 static inline bool kvm_set_pmuserenr(u64 val) 1340 { 1341 return false; 1342 } 1343 #endif 1344 1345 void kvm_vcpu_load_vhe(struct kvm_vcpu *vcpu); 1346 void kvm_vcpu_put_vhe(struct kvm_vcpu *vcpu); 1347 1348 int __init kvm_set_ipa_limit(void); 1349 u32 kvm_get_pa_bits(struct kvm *kvm); 1350 1351 #define __KVM_HAVE_ARCH_VM_ALLOC 1352 struct kvm *kvm_arch_alloc_vm(void); 1353 1354 #define __KVM_HAVE_ARCH_FLUSH_REMOTE_TLBS 1355 1356 #define __KVM_HAVE_ARCH_FLUSH_REMOTE_TLBS_RANGE 1357 1358 #define kvm_vm_is_protected(kvm) (is_protected_kvm_enabled() && (kvm)->arch.pkvm.enabled) 1359 1360 #define vcpu_is_protected(vcpu) kvm_vm_is_protected((vcpu)->kvm) 1361 1362 int kvm_arm_vcpu_finalize(struct kvm_vcpu *vcpu, int feature); 1363 bool kvm_arm_vcpu_is_finalized(struct kvm_vcpu *vcpu); 1364 1365 #define kvm_arm_vcpu_sve_finalized(vcpu) vcpu_get_flag(vcpu, VCPU_SVE_FINALIZED) 1366 1367 #define kvm_has_mte(kvm) \ 1368 (system_supports_mte() && \ 1369 test_bit(KVM_ARCH_FLAG_MTE_ENABLED, &(kvm)->arch.flags)) 1370 1371 #define kvm_supports_32bit_el0() \ 1372 (system_supports_32bit_el0() && \ 1373 !static_branch_unlikely(&arm64_mismatched_32bit_el0)) 1374 1375 #define kvm_vm_has_ran_once(kvm) \ 1376 (test_bit(KVM_ARCH_FLAG_HAS_RAN_ONCE, &(kvm)->arch.flags)) 1377 1378 static inline bool __vcpu_has_feature(const struct kvm_arch *ka, int feature) 1379 { 1380 return test_bit(feature, ka->vcpu_features); 1381 } 1382 1383 #define vcpu_has_feature(v, f) __vcpu_has_feature(&(v)->kvm->arch, (f)) 1384 1385 #define kvm_vcpu_initialized(v) vcpu_get_flag(vcpu, VCPU_INITIALIZED) 1386 1387 int kvm_trng_call(struct kvm_vcpu *vcpu); 1388 #ifdef CONFIG_KVM 1389 extern phys_addr_t hyp_mem_base; 1390 extern phys_addr_t hyp_mem_size; 1391 void __init kvm_hyp_reserve(void); 1392 #else 1393 static inline void kvm_hyp_reserve(void) { } 1394 #endif 1395 1396 void kvm_arm_vcpu_power_off(struct kvm_vcpu *vcpu); 1397 bool kvm_arm_vcpu_stopped(struct kvm_vcpu *vcpu); 1398 1399 static inline u64 *__vm_id_reg(struct kvm_arch *ka, u32 reg) 1400 { 1401 switch (reg) { 1402 case sys_reg(3, 0, 0, 1, 0) ... sys_reg(3, 0, 0, 7, 7): 1403 return &ka->id_regs[IDREG_IDX(reg)]; 1404 case SYS_CTR_EL0: 1405 return &ka->ctr_el0; 1406 default: 1407 WARN_ON_ONCE(1); 1408 return NULL; 1409 } 1410 } 1411 1412 #define kvm_read_vm_id_reg(kvm, reg) \ 1413 ({ u64 __val = *__vm_id_reg(&(kvm)->arch, reg); __val; }) 1414 1415 void kvm_set_vm_id_reg(struct kvm *kvm, u32 reg, u64 val); 1416 1417 #define __expand_field_sign_unsigned(id, fld, val) \ 1418 ((u64)SYS_FIELD_VALUE(id, fld, val)) 1419 1420 #define __expand_field_sign_signed(id, fld, val) \ 1421 ({ \ 1422 u64 __val = SYS_FIELD_VALUE(id, fld, val); \ 1423 sign_extend64(__val, id##_##fld##_WIDTH - 1); \ 1424 }) 1425 1426 #define expand_field_sign(id, fld, val) \ 1427 (id##_##fld##_SIGNED ? \ 1428 __expand_field_sign_signed(id, fld, val) : \ 1429 __expand_field_sign_unsigned(id, fld, val)) 1430 1431 #define get_idreg_field_unsigned(kvm, id, fld) \ 1432 ({ \ 1433 u64 __val = kvm_read_vm_id_reg((kvm), SYS_##id); \ 1434 FIELD_GET(id##_##fld##_MASK, __val); \ 1435 }) 1436 1437 #define get_idreg_field_signed(kvm, id, fld) \ 1438 ({ \ 1439 u64 __val = get_idreg_field_unsigned(kvm, id, fld); \ 1440 sign_extend64(__val, id##_##fld##_WIDTH - 1); \ 1441 }) 1442 1443 #define get_idreg_field_enum(kvm, id, fld) \ 1444 get_idreg_field_unsigned(kvm, id, fld) 1445 1446 #define get_idreg_field(kvm, id, fld) \ 1447 (id##_##fld##_SIGNED ? \ 1448 get_idreg_field_signed(kvm, id, fld) : \ 1449 get_idreg_field_unsigned(kvm, id, fld)) 1450 1451 #define kvm_has_feat(kvm, id, fld, limit) \ 1452 (get_idreg_field((kvm), id, fld) >= expand_field_sign(id, fld, limit)) 1453 1454 #define kvm_has_feat_enum(kvm, id, fld, val) \ 1455 (get_idreg_field_unsigned((kvm), id, fld) == __expand_field_sign_unsigned(id, fld, val)) 1456 1457 #define kvm_has_feat_range(kvm, id, fld, min, max) \ 1458 (get_idreg_field((kvm), id, fld) >= expand_field_sign(id, fld, min) && \ 1459 get_idreg_field((kvm), id, fld) <= expand_field_sign(id, fld, max)) 1460 1461 /* Check for a given level of PAuth support */ 1462 #define kvm_has_pauth(k, l) \ 1463 ({ \ 1464 bool pa, pi, pa3; \ 1465 \ 1466 pa = kvm_has_feat((k), ID_AA64ISAR1_EL1, APA, l); \ 1467 pa &= kvm_has_feat((k), ID_AA64ISAR1_EL1, GPA, IMP); \ 1468 pi = kvm_has_feat((k), ID_AA64ISAR1_EL1, API, l); \ 1469 pi &= kvm_has_feat((k), ID_AA64ISAR1_EL1, GPI, IMP); \ 1470 pa3 = kvm_has_feat((k), ID_AA64ISAR2_EL1, APA3, l); \ 1471 pa3 &= kvm_has_feat((k), ID_AA64ISAR2_EL1, GPA3, IMP); \ 1472 \ 1473 (pa + pi + pa3) == 1; \ 1474 }) 1475 1476 #endif /* __ARM64_KVM_HOST_H__ */ 1477