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 IDX_IDREG(idx) sys_reg(3, 0, 0, ((idx) >> 3) + 1, (idx) & Op2_mask) 366 #define IDREG(kvm, id) ((kvm)->arch.id_regs[IDREG_IDX(id)]) 367 #define KVM_ARM_ID_REG_NUM (IDREG_IDX(sys_reg(3, 0, 0, 7, 7)) + 1) 368 u64 id_regs[KVM_ARM_ID_REG_NUM]; 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 TTBR0_EL2, /* Translation Table Base Register 0 (EL2) */ 462 TTBR1_EL2, /* Translation Table Base Register 1 (EL2) */ 463 TCR_EL2, /* Translation Control Register (EL2) */ 464 SPSR_EL2, /* EL2 saved program status register */ 465 ELR_EL2, /* EL2 exception link register */ 466 AFSR0_EL2, /* Auxiliary Fault Status Register 0 (EL2) */ 467 AFSR1_EL2, /* Auxiliary Fault Status Register 1 (EL2) */ 468 ESR_EL2, /* Exception Syndrome Register (EL2) */ 469 FAR_EL2, /* Fault Address Register (EL2) */ 470 HPFAR_EL2, /* Hypervisor IPA Fault Address Register */ 471 MAIR_EL2, /* Memory Attribute Indirection Register (EL2) */ 472 AMAIR_EL2, /* Auxiliary Memory Attribute Indirection Register (EL2) */ 473 VBAR_EL2, /* Vector Base Address Register (EL2) */ 474 RVBAR_EL2, /* Reset Vector Base Address Register */ 475 CONTEXTIDR_EL2, /* Context ID Register (EL2) */ 476 CNTHCTL_EL2, /* Counter-timer Hypervisor Control register */ 477 SP_EL2, /* EL2 Stack Pointer */ 478 CNTHP_CTL_EL2, 479 CNTHP_CVAL_EL2, 480 CNTHV_CTL_EL2, 481 CNTHV_CVAL_EL2, 482 483 __VNCR_START__, /* Any VNCR-capable reg goes after this point */ 484 485 VNCR(SCTLR_EL1),/* System Control Register */ 486 VNCR(ACTLR_EL1),/* Auxiliary Control Register */ 487 VNCR(CPACR_EL1),/* Coprocessor Access Control */ 488 VNCR(ZCR_EL1), /* SVE Control */ 489 VNCR(TTBR0_EL1),/* Translation Table Base Register 0 */ 490 VNCR(TTBR1_EL1),/* Translation Table Base Register 1 */ 491 VNCR(TCR_EL1), /* Translation Control Register */ 492 VNCR(TCR2_EL1), /* Extended Translation Control Register */ 493 VNCR(ESR_EL1), /* Exception Syndrome Register */ 494 VNCR(AFSR0_EL1),/* Auxiliary Fault Status Register 0 */ 495 VNCR(AFSR1_EL1),/* Auxiliary Fault Status Register 1 */ 496 VNCR(FAR_EL1), /* Fault Address Register */ 497 VNCR(MAIR_EL1), /* Memory Attribute Indirection Register */ 498 VNCR(VBAR_EL1), /* Vector Base Address Register */ 499 VNCR(CONTEXTIDR_EL1), /* Context ID Register */ 500 VNCR(AMAIR_EL1),/* Aux Memory Attribute Indirection Register */ 501 VNCR(MDSCR_EL1),/* Monitor Debug System Control Register */ 502 VNCR(ELR_EL1), 503 VNCR(SP_EL1), 504 VNCR(SPSR_EL1), 505 VNCR(TFSR_EL1), /* Tag Fault Status Register (EL1) */ 506 VNCR(VPIDR_EL2),/* Virtualization Processor ID Register */ 507 VNCR(VMPIDR_EL2),/* Virtualization Multiprocessor ID Register */ 508 VNCR(HCR_EL2), /* Hypervisor Configuration Register */ 509 VNCR(HSTR_EL2), /* Hypervisor System Trap Register */ 510 VNCR(VTTBR_EL2),/* Virtualization Translation Table Base Register */ 511 VNCR(VTCR_EL2), /* Virtualization Translation Control Register */ 512 VNCR(TPIDR_EL2),/* EL2 Software Thread ID Register */ 513 VNCR(HCRX_EL2), /* Extended Hypervisor Configuration Register */ 514 515 /* Permission Indirection Extension registers */ 516 VNCR(PIR_EL1), /* Permission Indirection Register 1 (EL1) */ 517 VNCR(PIRE0_EL1), /* Permission Indirection Register 0 (EL1) */ 518 519 VNCR(HFGRTR_EL2), 520 VNCR(HFGWTR_EL2), 521 VNCR(HFGITR_EL2), 522 VNCR(HDFGRTR_EL2), 523 VNCR(HDFGWTR_EL2), 524 VNCR(HAFGRTR_EL2), 525 526 VNCR(CNTVOFF_EL2), 527 VNCR(CNTV_CVAL_EL0), 528 VNCR(CNTV_CTL_EL0), 529 VNCR(CNTP_CVAL_EL0), 530 VNCR(CNTP_CTL_EL0), 531 532 NR_SYS_REGS /* Nothing after this line! */ 533 }; 534 535 struct kvm_sysreg_masks { 536 struct { 537 u64 res0; 538 u64 res1; 539 } mask[NR_SYS_REGS - __VNCR_START__]; 540 }; 541 542 struct kvm_cpu_context { 543 struct user_pt_regs regs; /* sp = sp_el0 */ 544 545 u64 spsr_abt; 546 u64 spsr_und; 547 u64 spsr_irq; 548 u64 spsr_fiq; 549 550 struct user_fpsimd_state fp_regs; 551 552 u64 sys_regs[NR_SYS_REGS]; 553 554 struct kvm_vcpu *__hyp_running_vcpu; 555 556 /* This pointer has to be 4kB aligned. */ 557 u64 *vncr_array; 558 }; 559 560 struct cpu_sve_state { 561 __u64 zcr_el1; 562 563 /* 564 * Ordering is important since __sve_save_state/__sve_restore_state 565 * relies on it. 566 */ 567 __u32 fpsr; 568 __u32 fpcr; 569 570 /* Must be SVE_VQ_BYTES (128 bit) aligned. */ 571 __u8 sve_regs[]; 572 }; 573 574 /* 575 * This structure is instantiated on a per-CPU basis, and contains 576 * data that is: 577 * 578 * - tied to a single physical CPU, and 579 * - either have a lifetime that does not extend past vcpu_put() 580 * - or is an invariant for the lifetime of the system 581 * 582 * Use host_data_ptr(field) as a way to access a pointer to such a 583 * field. 584 */ 585 struct kvm_host_data { 586 struct kvm_cpu_context host_ctxt; 587 588 /* 589 * All pointers in this union are hyp VA. 590 * sve_state is only used in pKVM and if system_supports_sve(). 591 */ 592 union { 593 struct user_fpsimd_state *fpsimd_state; 594 struct cpu_sve_state *sve_state; 595 }; 596 597 /* Ownership of the FP regs */ 598 enum { 599 FP_STATE_FREE, 600 FP_STATE_HOST_OWNED, 601 FP_STATE_GUEST_OWNED, 602 } fp_owner; 603 604 /* 605 * host_debug_state contains the host registers which are 606 * saved and restored during world switches. 607 */ 608 struct { 609 /* {Break,watch}point registers */ 610 struct kvm_guest_debug_arch regs; 611 /* Statistical profiling extension */ 612 u64 pmscr_el1; 613 /* Self-hosted trace */ 614 u64 trfcr_el1; 615 /* Values of trap registers for the host before guest entry. */ 616 u64 mdcr_el2; 617 } host_debug_state; 618 }; 619 620 struct kvm_host_psci_config { 621 /* PSCI version used by host. */ 622 u32 version; 623 u32 smccc_version; 624 625 /* Function IDs used by host if version is v0.1. */ 626 struct psci_0_1_function_ids function_ids_0_1; 627 628 bool psci_0_1_cpu_suspend_implemented; 629 bool psci_0_1_cpu_on_implemented; 630 bool psci_0_1_cpu_off_implemented; 631 bool psci_0_1_migrate_implemented; 632 }; 633 634 extern struct kvm_host_psci_config kvm_nvhe_sym(kvm_host_psci_config); 635 #define kvm_host_psci_config CHOOSE_NVHE_SYM(kvm_host_psci_config) 636 637 extern s64 kvm_nvhe_sym(hyp_physvirt_offset); 638 #define hyp_physvirt_offset CHOOSE_NVHE_SYM(hyp_physvirt_offset) 639 640 extern u64 kvm_nvhe_sym(hyp_cpu_logical_map)[NR_CPUS]; 641 #define hyp_cpu_logical_map CHOOSE_NVHE_SYM(hyp_cpu_logical_map) 642 643 struct vcpu_reset_state { 644 unsigned long pc; 645 unsigned long r0; 646 bool be; 647 bool reset; 648 }; 649 650 struct kvm_vcpu_arch { 651 struct kvm_cpu_context ctxt; 652 653 /* 654 * Guest floating point state 655 * 656 * The architecture has two main floating point extensions, 657 * the original FPSIMD and SVE. These have overlapping 658 * register views, with the FPSIMD V registers occupying the 659 * low 128 bits of the SVE Z registers. When the core 660 * floating point code saves the register state of a task it 661 * records which view it saved in fp_type. 662 */ 663 void *sve_state; 664 enum fp_type fp_type; 665 unsigned int sve_max_vl; 666 u64 svcr; 667 u64 fpmr; 668 669 /* Stage 2 paging state used by the hardware on next switch */ 670 struct kvm_s2_mmu *hw_mmu; 671 672 /* Values of trap registers for the guest. */ 673 u64 hcr_el2; 674 u64 hcrx_el2; 675 u64 mdcr_el2; 676 u64 cptr_el2; 677 678 /* Exception Information */ 679 struct kvm_vcpu_fault_info fault; 680 681 /* Configuration flags, set once and for all before the vcpu can run */ 682 u8 cflags; 683 684 /* Input flags to the hypervisor code, potentially cleared after use */ 685 u8 iflags; 686 687 /* State flags for kernel bookkeeping, unused by the hypervisor code */ 688 u8 sflags; 689 690 /* 691 * Don't run the guest (internal implementation need). 692 * 693 * Contrary to the flags above, this is set/cleared outside of 694 * a vcpu context, and thus cannot be mixed with the flags 695 * themselves (or the flag accesses need to be made atomic). 696 */ 697 bool pause; 698 699 /* 700 * We maintain more than a single set of debug registers to support 701 * debugging the guest from the host and to maintain separate host and 702 * guest state during world switches. vcpu_debug_state are the debug 703 * registers of the vcpu as the guest sees them. 704 * 705 * external_debug_state contains the debug values we want to debug the 706 * guest. This is set via the KVM_SET_GUEST_DEBUG ioctl. 707 * 708 * debug_ptr points to the set of debug registers that should be loaded 709 * onto the hardware when running the guest. 710 */ 711 struct kvm_guest_debug_arch *debug_ptr; 712 struct kvm_guest_debug_arch vcpu_debug_state; 713 struct kvm_guest_debug_arch external_debug_state; 714 715 /* VGIC state */ 716 struct vgic_cpu vgic_cpu; 717 struct arch_timer_cpu timer_cpu; 718 struct kvm_pmu pmu; 719 720 /* 721 * Guest registers we preserve during guest debugging. 722 * 723 * These shadow registers are updated by the kvm_handle_sys_reg 724 * trap handler if the guest accesses or updates them while we 725 * are using guest debug. 726 */ 727 struct { 728 u32 mdscr_el1; 729 bool pstate_ss; 730 } guest_debug_preserved; 731 732 /* vcpu power state */ 733 struct kvm_mp_state mp_state; 734 spinlock_t mp_state_lock; 735 736 /* Cache some mmu pages needed inside spinlock regions */ 737 struct kvm_mmu_memory_cache mmu_page_cache; 738 739 /* Virtual SError ESR to restore when HCR_EL2.VSE is set */ 740 u64 vsesr_el2; 741 742 /* Additional reset state */ 743 struct vcpu_reset_state reset_state; 744 745 /* Guest PV state */ 746 struct { 747 u64 last_steal; 748 gpa_t base; 749 } steal; 750 751 /* Per-vcpu CCSIDR override or NULL */ 752 u32 *ccsidr; 753 }; 754 755 /* 756 * Each 'flag' is composed of a comma-separated triplet: 757 * 758 * - the flag-set it belongs to in the vcpu->arch structure 759 * - the value for that flag 760 * - the mask for that flag 761 * 762 * __vcpu_single_flag() builds such a triplet for a single-bit flag. 763 * unpack_vcpu_flag() extract the flag value from the triplet for 764 * direct use outside of the flag accessors. 765 */ 766 #define __vcpu_single_flag(_set, _f) _set, (_f), (_f) 767 768 #define __unpack_flag(_set, _f, _m) _f 769 #define unpack_vcpu_flag(...) __unpack_flag(__VA_ARGS__) 770 771 #define __build_check_flag(v, flagset, f, m) \ 772 do { \ 773 typeof(v->arch.flagset) *_fset; \ 774 \ 775 /* Check that the flags fit in the mask */ \ 776 BUILD_BUG_ON(HWEIGHT(m) != HWEIGHT((f) | (m))); \ 777 /* Check that the flags fit in the type */ \ 778 BUILD_BUG_ON((sizeof(*_fset) * 8) <= __fls(m)); \ 779 } while (0) 780 781 #define __vcpu_get_flag(v, flagset, f, m) \ 782 ({ \ 783 __build_check_flag(v, flagset, f, m); \ 784 \ 785 READ_ONCE(v->arch.flagset) & (m); \ 786 }) 787 788 /* 789 * Note that the set/clear accessors must be preempt-safe in order to 790 * avoid nesting them with load/put which also manipulate flags... 791 */ 792 #ifdef __KVM_NVHE_HYPERVISOR__ 793 /* the nVHE hypervisor is always non-preemptible */ 794 #define __vcpu_flags_preempt_disable() 795 #define __vcpu_flags_preempt_enable() 796 #else 797 #define __vcpu_flags_preempt_disable() preempt_disable() 798 #define __vcpu_flags_preempt_enable() preempt_enable() 799 #endif 800 801 #define __vcpu_set_flag(v, flagset, f, m) \ 802 do { \ 803 typeof(v->arch.flagset) *fset; \ 804 \ 805 __build_check_flag(v, flagset, f, m); \ 806 \ 807 fset = &v->arch.flagset; \ 808 __vcpu_flags_preempt_disable(); \ 809 if (HWEIGHT(m) > 1) \ 810 *fset &= ~(m); \ 811 *fset |= (f); \ 812 __vcpu_flags_preempt_enable(); \ 813 } while (0) 814 815 #define __vcpu_clear_flag(v, flagset, f, m) \ 816 do { \ 817 typeof(v->arch.flagset) *fset; \ 818 \ 819 __build_check_flag(v, flagset, f, m); \ 820 \ 821 fset = &v->arch.flagset; \ 822 __vcpu_flags_preempt_disable(); \ 823 *fset &= ~(m); \ 824 __vcpu_flags_preempt_enable(); \ 825 } while (0) 826 827 #define vcpu_get_flag(v, ...) __vcpu_get_flag((v), __VA_ARGS__) 828 #define vcpu_set_flag(v, ...) __vcpu_set_flag((v), __VA_ARGS__) 829 #define vcpu_clear_flag(v, ...) __vcpu_clear_flag((v), __VA_ARGS__) 830 831 /* SVE exposed to guest */ 832 #define GUEST_HAS_SVE __vcpu_single_flag(cflags, BIT(0)) 833 /* SVE config completed */ 834 #define VCPU_SVE_FINALIZED __vcpu_single_flag(cflags, BIT(1)) 835 /* PTRAUTH exposed to guest */ 836 #define GUEST_HAS_PTRAUTH __vcpu_single_flag(cflags, BIT(2)) 837 /* KVM_ARM_VCPU_INIT completed */ 838 #define VCPU_INITIALIZED __vcpu_single_flag(cflags, BIT(3)) 839 840 /* Exception pending */ 841 #define PENDING_EXCEPTION __vcpu_single_flag(iflags, BIT(0)) 842 /* 843 * PC increment. Overlaps with EXCEPT_MASK on purpose so that it can't 844 * be set together with an exception... 845 */ 846 #define INCREMENT_PC __vcpu_single_flag(iflags, BIT(1)) 847 /* Target EL/MODE (not a single flag, but let's abuse the macro) */ 848 #define EXCEPT_MASK __vcpu_single_flag(iflags, GENMASK(3, 1)) 849 850 /* Helpers to encode exceptions with minimum fuss */ 851 #define __EXCEPT_MASK_VAL unpack_vcpu_flag(EXCEPT_MASK) 852 #define __EXCEPT_SHIFT __builtin_ctzl(__EXCEPT_MASK_VAL) 853 #define __vcpu_except_flags(_f) iflags, (_f << __EXCEPT_SHIFT), __EXCEPT_MASK_VAL 854 855 /* 856 * When PENDING_EXCEPTION is set, EXCEPT_MASK can take the following 857 * values: 858 * 859 * For AArch32 EL1: 860 */ 861 #define EXCEPT_AA32_UND __vcpu_except_flags(0) 862 #define EXCEPT_AA32_IABT __vcpu_except_flags(1) 863 #define EXCEPT_AA32_DABT __vcpu_except_flags(2) 864 /* For AArch64: */ 865 #define EXCEPT_AA64_EL1_SYNC __vcpu_except_flags(0) 866 #define EXCEPT_AA64_EL1_IRQ __vcpu_except_flags(1) 867 #define EXCEPT_AA64_EL1_FIQ __vcpu_except_flags(2) 868 #define EXCEPT_AA64_EL1_SERR __vcpu_except_flags(3) 869 /* For AArch64 with NV: */ 870 #define EXCEPT_AA64_EL2_SYNC __vcpu_except_flags(4) 871 #define EXCEPT_AA64_EL2_IRQ __vcpu_except_flags(5) 872 #define EXCEPT_AA64_EL2_FIQ __vcpu_except_flags(6) 873 #define EXCEPT_AA64_EL2_SERR __vcpu_except_flags(7) 874 /* Guest debug is live */ 875 #define DEBUG_DIRTY __vcpu_single_flag(iflags, BIT(4)) 876 /* Save SPE context if active */ 877 #define DEBUG_STATE_SAVE_SPE __vcpu_single_flag(iflags, BIT(5)) 878 /* Save TRBE context if active */ 879 #define DEBUG_STATE_SAVE_TRBE __vcpu_single_flag(iflags, BIT(6)) 880 881 /* SVE enabled for host EL0 */ 882 #define HOST_SVE_ENABLED __vcpu_single_flag(sflags, BIT(0)) 883 /* SME enabled for EL0 */ 884 #define HOST_SME_ENABLED __vcpu_single_flag(sflags, BIT(1)) 885 /* Physical CPU not in supported_cpus */ 886 #define ON_UNSUPPORTED_CPU __vcpu_single_flag(sflags, BIT(2)) 887 /* WFIT instruction trapped */ 888 #define IN_WFIT __vcpu_single_flag(sflags, BIT(3)) 889 /* vcpu system registers loaded on physical CPU */ 890 #define SYSREGS_ON_CPU __vcpu_single_flag(sflags, BIT(4)) 891 /* Software step state is Active-pending */ 892 #define DBG_SS_ACTIVE_PENDING __vcpu_single_flag(sflags, BIT(5)) 893 /* PMUSERENR for the guest EL0 is on physical CPU */ 894 #define PMUSERENR_ON_CPU __vcpu_single_flag(sflags, BIT(6)) 895 /* WFI instruction trapped */ 896 #define IN_WFI __vcpu_single_flag(sflags, BIT(7)) 897 898 899 /* Pointer to the vcpu's SVE FFR for sve_{save,load}_state() */ 900 #define vcpu_sve_pffr(vcpu) (kern_hyp_va((vcpu)->arch.sve_state) + \ 901 sve_ffr_offset((vcpu)->arch.sve_max_vl)) 902 903 #define vcpu_sve_max_vq(vcpu) sve_vq_from_vl((vcpu)->arch.sve_max_vl) 904 905 #define vcpu_sve_state_size(vcpu) ({ \ 906 size_t __size_ret; \ 907 unsigned int __vcpu_vq; \ 908 \ 909 if (WARN_ON(!sve_vl_valid((vcpu)->arch.sve_max_vl))) { \ 910 __size_ret = 0; \ 911 } else { \ 912 __vcpu_vq = vcpu_sve_max_vq(vcpu); \ 913 __size_ret = SVE_SIG_REGS_SIZE(__vcpu_vq); \ 914 } \ 915 \ 916 __size_ret; \ 917 }) 918 919 #define KVM_GUESTDBG_VALID_MASK (KVM_GUESTDBG_ENABLE | \ 920 KVM_GUESTDBG_USE_SW_BP | \ 921 KVM_GUESTDBG_USE_HW | \ 922 KVM_GUESTDBG_SINGLESTEP) 923 924 #define vcpu_has_sve(vcpu) (system_supports_sve() && \ 925 vcpu_get_flag(vcpu, GUEST_HAS_SVE)) 926 927 #ifdef CONFIG_ARM64_PTR_AUTH 928 #define vcpu_has_ptrauth(vcpu) \ 929 ((cpus_have_final_cap(ARM64_HAS_ADDRESS_AUTH) || \ 930 cpus_have_final_cap(ARM64_HAS_GENERIC_AUTH)) && \ 931 vcpu_get_flag(vcpu, GUEST_HAS_PTRAUTH)) 932 #else 933 #define vcpu_has_ptrauth(vcpu) false 934 #endif 935 936 #define vcpu_on_unsupported_cpu(vcpu) \ 937 vcpu_get_flag(vcpu, ON_UNSUPPORTED_CPU) 938 939 #define vcpu_set_on_unsupported_cpu(vcpu) \ 940 vcpu_set_flag(vcpu, ON_UNSUPPORTED_CPU) 941 942 #define vcpu_clear_on_unsupported_cpu(vcpu) \ 943 vcpu_clear_flag(vcpu, ON_UNSUPPORTED_CPU) 944 945 #define vcpu_gp_regs(v) (&(v)->arch.ctxt.regs) 946 947 /* 948 * Only use __vcpu_sys_reg/ctxt_sys_reg if you know you want the 949 * memory backed version of a register, and not the one most recently 950 * accessed by a running VCPU. For example, for userspace access or 951 * for system registers that are never context switched, but only 952 * emulated. 953 * 954 * Don't bother with VNCR-based accesses in the nVHE code, it has no 955 * business dealing with NV. 956 */ 957 static inline u64 *___ctxt_sys_reg(const struct kvm_cpu_context *ctxt, int r) 958 { 959 #if !defined (__KVM_NVHE_HYPERVISOR__) 960 if (unlikely(cpus_have_final_cap(ARM64_HAS_NESTED_VIRT) && 961 r >= __VNCR_START__ && ctxt->vncr_array)) 962 return &ctxt->vncr_array[r - __VNCR_START__]; 963 #endif 964 return (u64 *)&ctxt->sys_regs[r]; 965 } 966 967 #define __ctxt_sys_reg(c,r) \ 968 ({ \ 969 BUILD_BUG_ON(__builtin_constant_p(r) && \ 970 (r) >= NR_SYS_REGS); \ 971 ___ctxt_sys_reg(c, r); \ 972 }) 973 974 #define ctxt_sys_reg(c,r) (*__ctxt_sys_reg(c,r)) 975 976 u64 kvm_vcpu_sanitise_vncr_reg(const struct kvm_vcpu *, enum vcpu_sysreg); 977 #define __vcpu_sys_reg(v,r) \ 978 (*({ \ 979 const struct kvm_cpu_context *ctxt = &(v)->arch.ctxt; \ 980 u64 *__r = __ctxt_sys_reg(ctxt, (r)); \ 981 if (vcpu_has_nv((v)) && (r) >= __VNCR_START__) \ 982 *__r = kvm_vcpu_sanitise_vncr_reg((v), (r)); \ 983 __r; \ 984 })) 985 986 u64 vcpu_read_sys_reg(const struct kvm_vcpu *vcpu, int reg); 987 void vcpu_write_sys_reg(struct kvm_vcpu *vcpu, u64 val, int reg); 988 989 static inline bool __vcpu_read_sys_reg_from_cpu(int reg, u64 *val) 990 { 991 /* 992 * *** VHE ONLY *** 993 * 994 * System registers listed in the switch are not saved on every 995 * exit from the guest but are only saved on vcpu_put. 996 * 997 * Note that MPIDR_EL1 for the guest is set by KVM via VMPIDR_EL2 but 998 * should never be listed below, because the guest cannot modify its 999 * own MPIDR_EL1 and MPIDR_EL1 is accessed for VCPU A from VCPU B's 1000 * thread when emulating cross-VCPU communication. 1001 */ 1002 if (!has_vhe()) 1003 return false; 1004 1005 switch (reg) { 1006 case SCTLR_EL1: *val = read_sysreg_s(SYS_SCTLR_EL12); break; 1007 case CPACR_EL1: *val = read_sysreg_s(SYS_CPACR_EL12); break; 1008 case TTBR0_EL1: *val = read_sysreg_s(SYS_TTBR0_EL12); break; 1009 case TTBR1_EL1: *val = read_sysreg_s(SYS_TTBR1_EL12); break; 1010 case TCR_EL1: *val = read_sysreg_s(SYS_TCR_EL12); break; 1011 case ESR_EL1: *val = read_sysreg_s(SYS_ESR_EL12); break; 1012 case AFSR0_EL1: *val = read_sysreg_s(SYS_AFSR0_EL12); break; 1013 case AFSR1_EL1: *val = read_sysreg_s(SYS_AFSR1_EL12); break; 1014 case FAR_EL1: *val = read_sysreg_s(SYS_FAR_EL12); break; 1015 case MAIR_EL1: *val = read_sysreg_s(SYS_MAIR_EL12); break; 1016 case VBAR_EL1: *val = read_sysreg_s(SYS_VBAR_EL12); break; 1017 case CONTEXTIDR_EL1: *val = read_sysreg_s(SYS_CONTEXTIDR_EL12);break; 1018 case TPIDR_EL0: *val = read_sysreg_s(SYS_TPIDR_EL0); break; 1019 case TPIDRRO_EL0: *val = read_sysreg_s(SYS_TPIDRRO_EL0); break; 1020 case TPIDR_EL1: *val = read_sysreg_s(SYS_TPIDR_EL1); break; 1021 case AMAIR_EL1: *val = read_sysreg_s(SYS_AMAIR_EL12); break; 1022 case CNTKCTL_EL1: *val = read_sysreg_s(SYS_CNTKCTL_EL12); break; 1023 case ELR_EL1: *val = read_sysreg_s(SYS_ELR_EL12); break; 1024 case SPSR_EL1: *val = read_sysreg_s(SYS_SPSR_EL12); break; 1025 case PAR_EL1: *val = read_sysreg_par(); break; 1026 case DACR32_EL2: *val = read_sysreg_s(SYS_DACR32_EL2); break; 1027 case IFSR32_EL2: *val = read_sysreg_s(SYS_IFSR32_EL2); break; 1028 case DBGVCR32_EL2: *val = read_sysreg_s(SYS_DBGVCR32_EL2); break; 1029 default: return false; 1030 } 1031 1032 return true; 1033 } 1034 1035 static inline bool __vcpu_write_sys_reg_to_cpu(u64 val, int reg) 1036 { 1037 /* 1038 * *** VHE ONLY *** 1039 * 1040 * System registers listed in the switch are not restored on every 1041 * entry to the guest but are only restored on vcpu_load. 1042 * 1043 * Note that MPIDR_EL1 for the guest is set by KVM via VMPIDR_EL2 but 1044 * should never be listed below, because the MPIDR should only be set 1045 * once, before running the VCPU, and never changed later. 1046 */ 1047 if (!has_vhe()) 1048 return false; 1049 1050 switch (reg) { 1051 case SCTLR_EL1: write_sysreg_s(val, SYS_SCTLR_EL12); break; 1052 case CPACR_EL1: write_sysreg_s(val, SYS_CPACR_EL12); break; 1053 case TTBR0_EL1: write_sysreg_s(val, SYS_TTBR0_EL12); break; 1054 case TTBR1_EL1: write_sysreg_s(val, SYS_TTBR1_EL12); break; 1055 case TCR_EL1: write_sysreg_s(val, SYS_TCR_EL12); break; 1056 case ESR_EL1: write_sysreg_s(val, SYS_ESR_EL12); break; 1057 case AFSR0_EL1: write_sysreg_s(val, SYS_AFSR0_EL12); break; 1058 case AFSR1_EL1: write_sysreg_s(val, SYS_AFSR1_EL12); break; 1059 case FAR_EL1: write_sysreg_s(val, SYS_FAR_EL12); break; 1060 case MAIR_EL1: write_sysreg_s(val, SYS_MAIR_EL12); break; 1061 case VBAR_EL1: write_sysreg_s(val, SYS_VBAR_EL12); break; 1062 case CONTEXTIDR_EL1: write_sysreg_s(val, SYS_CONTEXTIDR_EL12);break; 1063 case TPIDR_EL0: write_sysreg_s(val, SYS_TPIDR_EL0); break; 1064 case TPIDRRO_EL0: write_sysreg_s(val, SYS_TPIDRRO_EL0); break; 1065 case TPIDR_EL1: write_sysreg_s(val, SYS_TPIDR_EL1); break; 1066 case AMAIR_EL1: write_sysreg_s(val, SYS_AMAIR_EL12); break; 1067 case CNTKCTL_EL1: write_sysreg_s(val, SYS_CNTKCTL_EL12); break; 1068 case ELR_EL1: write_sysreg_s(val, SYS_ELR_EL12); break; 1069 case SPSR_EL1: write_sysreg_s(val, SYS_SPSR_EL12); break; 1070 case PAR_EL1: write_sysreg_s(val, SYS_PAR_EL1); break; 1071 case DACR32_EL2: write_sysreg_s(val, SYS_DACR32_EL2); break; 1072 case IFSR32_EL2: write_sysreg_s(val, SYS_IFSR32_EL2); break; 1073 case DBGVCR32_EL2: write_sysreg_s(val, SYS_DBGVCR32_EL2); break; 1074 default: return false; 1075 } 1076 1077 return true; 1078 } 1079 1080 struct kvm_vm_stat { 1081 struct kvm_vm_stat_generic generic; 1082 }; 1083 1084 struct kvm_vcpu_stat { 1085 struct kvm_vcpu_stat_generic generic; 1086 u64 hvc_exit_stat; 1087 u64 wfe_exit_stat; 1088 u64 wfi_exit_stat; 1089 u64 mmio_exit_user; 1090 u64 mmio_exit_kernel; 1091 u64 signal_exits; 1092 u64 exits; 1093 }; 1094 1095 unsigned long kvm_arm_num_regs(struct kvm_vcpu *vcpu); 1096 int kvm_arm_copy_reg_indices(struct kvm_vcpu *vcpu, u64 __user *indices); 1097 int kvm_arm_get_reg(struct kvm_vcpu *vcpu, const struct kvm_one_reg *reg); 1098 int kvm_arm_set_reg(struct kvm_vcpu *vcpu, const struct kvm_one_reg *reg); 1099 1100 unsigned long kvm_arm_num_sys_reg_descs(struct kvm_vcpu *vcpu); 1101 int kvm_arm_copy_sys_reg_indices(struct kvm_vcpu *vcpu, u64 __user *uindices); 1102 1103 int __kvm_arm_vcpu_get_events(struct kvm_vcpu *vcpu, 1104 struct kvm_vcpu_events *events); 1105 1106 int __kvm_arm_vcpu_set_events(struct kvm_vcpu *vcpu, 1107 struct kvm_vcpu_events *events); 1108 1109 void kvm_arm_halt_guest(struct kvm *kvm); 1110 void kvm_arm_resume_guest(struct kvm *kvm); 1111 1112 #define vcpu_has_run_once(vcpu) !!rcu_access_pointer((vcpu)->pid) 1113 1114 #ifndef __KVM_NVHE_HYPERVISOR__ 1115 #define kvm_call_hyp_nvhe(f, ...) \ 1116 ({ \ 1117 struct arm_smccc_res res; \ 1118 \ 1119 arm_smccc_1_1_hvc(KVM_HOST_SMCCC_FUNC(f), \ 1120 ##__VA_ARGS__, &res); \ 1121 WARN_ON(res.a0 != SMCCC_RET_SUCCESS); \ 1122 \ 1123 res.a1; \ 1124 }) 1125 1126 /* 1127 * The couple of isb() below are there to guarantee the same behaviour 1128 * on VHE as on !VHE, where the eret to EL1 acts as a context 1129 * synchronization event. 1130 */ 1131 #define kvm_call_hyp(f, ...) \ 1132 do { \ 1133 if (has_vhe()) { \ 1134 f(__VA_ARGS__); \ 1135 isb(); \ 1136 } else { \ 1137 kvm_call_hyp_nvhe(f, ##__VA_ARGS__); \ 1138 } \ 1139 } while(0) 1140 1141 #define kvm_call_hyp_ret(f, ...) \ 1142 ({ \ 1143 typeof(f(__VA_ARGS__)) ret; \ 1144 \ 1145 if (has_vhe()) { \ 1146 ret = f(__VA_ARGS__); \ 1147 isb(); \ 1148 } else { \ 1149 ret = kvm_call_hyp_nvhe(f, ##__VA_ARGS__); \ 1150 } \ 1151 \ 1152 ret; \ 1153 }) 1154 #else /* __KVM_NVHE_HYPERVISOR__ */ 1155 #define kvm_call_hyp(f, ...) f(__VA_ARGS__) 1156 #define kvm_call_hyp_ret(f, ...) f(__VA_ARGS__) 1157 #define kvm_call_hyp_nvhe(f, ...) f(__VA_ARGS__) 1158 #endif /* __KVM_NVHE_HYPERVISOR__ */ 1159 1160 int handle_exit(struct kvm_vcpu *vcpu, int exception_index); 1161 void handle_exit_early(struct kvm_vcpu *vcpu, int exception_index); 1162 1163 int kvm_handle_cp14_load_store(struct kvm_vcpu *vcpu); 1164 int kvm_handle_cp14_32(struct kvm_vcpu *vcpu); 1165 int kvm_handle_cp14_64(struct kvm_vcpu *vcpu); 1166 int kvm_handle_cp15_32(struct kvm_vcpu *vcpu); 1167 int kvm_handle_cp15_64(struct kvm_vcpu *vcpu); 1168 int kvm_handle_sys_reg(struct kvm_vcpu *vcpu); 1169 int kvm_handle_cp10_id(struct kvm_vcpu *vcpu); 1170 1171 void kvm_sys_regs_create_debugfs(struct kvm *kvm); 1172 void kvm_reset_sys_regs(struct kvm_vcpu *vcpu); 1173 1174 int __init kvm_sys_reg_table_init(void); 1175 struct sys_reg_desc; 1176 int __init populate_sysreg_config(const struct sys_reg_desc *sr, 1177 unsigned int idx); 1178 int __init populate_nv_trap_config(void); 1179 1180 bool lock_all_vcpus(struct kvm *kvm); 1181 void unlock_all_vcpus(struct kvm *kvm); 1182 1183 void kvm_init_sysreg(struct kvm_vcpu *); 1184 1185 /* MMIO helpers */ 1186 void kvm_mmio_write_buf(void *buf, unsigned int len, unsigned long data); 1187 unsigned long kvm_mmio_read_buf(const void *buf, unsigned int len); 1188 1189 int kvm_handle_mmio_return(struct kvm_vcpu *vcpu); 1190 int io_mem_abort(struct kvm_vcpu *vcpu, phys_addr_t fault_ipa); 1191 1192 /* 1193 * Returns true if a Performance Monitoring Interrupt (PMI), a.k.a. perf event, 1194 * arrived in guest context. For arm64, any event that arrives while a vCPU is 1195 * loaded is considered to be "in guest". 1196 */ 1197 static inline bool kvm_arch_pmi_in_guest(struct kvm_vcpu *vcpu) 1198 { 1199 return IS_ENABLED(CONFIG_GUEST_PERF_EVENTS) && !!vcpu; 1200 } 1201 1202 long kvm_hypercall_pv_features(struct kvm_vcpu *vcpu); 1203 gpa_t kvm_init_stolen_time(struct kvm_vcpu *vcpu); 1204 void kvm_update_stolen_time(struct kvm_vcpu *vcpu); 1205 1206 bool kvm_arm_pvtime_supported(void); 1207 int kvm_arm_pvtime_set_attr(struct kvm_vcpu *vcpu, 1208 struct kvm_device_attr *attr); 1209 int kvm_arm_pvtime_get_attr(struct kvm_vcpu *vcpu, 1210 struct kvm_device_attr *attr); 1211 int kvm_arm_pvtime_has_attr(struct kvm_vcpu *vcpu, 1212 struct kvm_device_attr *attr); 1213 1214 extern unsigned int __ro_after_init kvm_arm_vmid_bits; 1215 int __init kvm_arm_vmid_alloc_init(void); 1216 void __init kvm_arm_vmid_alloc_free(void); 1217 bool kvm_arm_vmid_update(struct kvm_vmid *kvm_vmid); 1218 void kvm_arm_vmid_clear_active(void); 1219 1220 static inline void kvm_arm_pvtime_vcpu_init(struct kvm_vcpu_arch *vcpu_arch) 1221 { 1222 vcpu_arch->steal.base = INVALID_GPA; 1223 } 1224 1225 static inline bool kvm_arm_is_pvtime_enabled(struct kvm_vcpu_arch *vcpu_arch) 1226 { 1227 return (vcpu_arch->steal.base != INVALID_GPA); 1228 } 1229 1230 void kvm_set_sei_esr(struct kvm_vcpu *vcpu, u64 syndrome); 1231 1232 struct kvm_vcpu *kvm_mpidr_to_vcpu(struct kvm *kvm, unsigned long mpidr); 1233 1234 DECLARE_KVM_HYP_PER_CPU(struct kvm_host_data, kvm_host_data); 1235 1236 /* 1237 * How we access per-CPU host data depends on the where we access it from, 1238 * and the mode we're in: 1239 * 1240 * - VHE and nVHE hypervisor bits use their locally defined instance 1241 * 1242 * - the rest of the kernel use either the VHE or nVHE one, depending on 1243 * the mode we're running in. 1244 * 1245 * Unless we're in protected mode, fully deprivileged, and the nVHE 1246 * per-CPU stuff is exclusively accessible to the protected EL2 code. 1247 * In this case, the EL1 code uses the *VHE* data as its private state 1248 * (which makes sense in a way as there shouldn't be any shared state 1249 * between the host and the hypervisor). 1250 * 1251 * Yes, this is all totally trivial. Shoot me now. 1252 */ 1253 #if defined(__KVM_NVHE_HYPERVISOR__) || defined(__KVM_VHE_HYPERVISOR__) 1254 #define host_data_ptr(f) (&this_cpu_ptr(&kvm_host_data)->f) 1255 #else 1256 #define host_data_ptr(f) \ 1257 (static_branch_unlikely(&kvm_protected_mode_initialized) ? \ 1258 &this_cpu_ptr(&kvm_host_data)->f : \ 1259 &this_cpu_ptr_hyp_sym(kvm_host_data)->f) 1260 #endif 1261 1262 /* Check whether the FP regs are owned by the guest */ 1263 static inline bool guest_owns_fp_regs(void) 1264 { 1265 return *host_data_ptr(fp_owner) == FP_STATE_GUEST_OWNED; 1266 } 1267 1268 /* Check whether the FP regs are owned by the host */ 1269 static inline bool host_owns_fp_regs(void) 1270 { 1271 return *host_data_ptr(fp_owner) == FP_STATE_HOST_OWNED; 1272 } 1273 1274 static inline void kvm_init_host_cpu_context(struct kvm_cpu_context *cpu_ctxt) 1275 { 1276 /* The host's MPIDR is immutable, so let's set it up at boot time */ 1277 ctxt_sys_reg(cpu_ctxt, MPIDR_EL1) = read_cpuid_mpidr(); 1278 } 1279 1280 static inline bool kvm_system_needs_idmapped_vectors(void) 1281 { 1282 return cpus_have_final_cap(ARM64_SPECTRE_V3A); 1283 } 1284 1285 static inline void kvm_arch_sync_events(struct kvm *kvm) {} 1286 static inline void kvm_arch_sched_in(struct kvm_vcpu *vcpu, int cpu) {} 1287 1288 void kvm_arm_init_debug(void); 1289 void kvm_arm_vcpu_init_debug(struct kvm_vcpu *vcpu); 1290 void kvm_arm_setup_debug(struct kvm_vcpu *vcpu); 1291 void kvm_arm_clear_debug(struct kvm_vcpu *vcpu); 1292 void kvm_arm_reset_debug_ptr(struct kvm_vcpu *vcpu); 1293 1294 #define kvm_vcpu_os_lock_enabled(vcpu) \ 1295 (!!(__vcpu_sys_reg(vcpu, OSLSR_EL1) & OSLSR_EL1_OSLK)) 1296 1297 int kvm_arm_vcpu_arch_set_attr(struct kvm_vcpu *vcpu, 1298 struct kvm_device_attr *attr); 1299 int kvm_arm_vcpu_arch_get_attr(struct kvm_vcpu *vcpu, 1300 struct kvm_device_attr *attr); 1301 int kvm_arm_vcpu_arch_has_attr(struct kvm_vcpu *vcpu, 1302 struct kvm_device_attr *attr); 1303 1304 int kvm_vm_ioctl_mte_copy_tags(struct kvm *kvm, 1305 struct kvm_arm_copy_mte_tags *copy_tags); 1306 int kvm_vm_ioctl_set_counter_offset(struct kvm *kvm, 1307 struct kvm_arm_counter_offset *offset); 1308 int kvm_vm_ioctl_get_reg_writable_masks(struct kvm *kvm, 1309 struct reg_mask_range *range); 1310 1311 /* Guest/host FPSIMD coordination helpers */ 1312 int kvm_arch_vcpu_run_map_fp(struct kvm_vcpu *vcpu); 1313 void kvm_arch_vcpu_load_fp(struct kvm_vcpu *vcpu); 1314 void kvm_arch_vcpu_ctxflush_fp(struct kvm_vcpu *vcpu); 1315 void kvm_arch_vcpu_ctxsync_fp(struct kvm_vcpu *vcpu); 1316 void kvm_arch_vcpu_put_fp(struct kvm_vcpu *vcpu); 1317 1318 static inline bool kvm_pmu_counter_deferred(struct perf_event_attr *attr) 1319 { 1320 return (!has_vhe() && attr->exclude_host); 1321 } 1322 1323 /* Flags for host debug state */ 1324 void kvm_arch_vcpu_load_debug_state_flags(struct kvm_vcpu *vcpu); 1325 void kvm_arch_vcpu_put_debug_state_flags(struct kvm_vcpu *vcpu); 1326 1327 #ifdef CONFIG_KVM 1328 void kvm_set_pmu_events(u32 set, struct perf_event_attr *attr); 1329 void kvm_clr_pmu_events(u32 clr); 1330 bool kvm_set_pmuserenr(u64 val); 1331 #else 1332 static inline void kvm_set_pmu_events(u32 set, struct perf_event_attr *attr) {} 1333 static inline void kvm_clr_pmu_events(u32 clr) {} 1334 static inline bool kvm_set_pmuserenr(u64 val) 1335 { 1336 return false; 1337 } 1338 #endif 1339 1340 void kvm_vcpu_load_vhe(struct kvm_vcpu *vcpu); 1341 void kvm_vcpu_put_vhe(struct kvm_vcpu *vcpu); 1342 1343 int __init kvm_set_ipa_limit(void); 1344 u32 kvm_get_pa_bits(struct kvm *kvm); 1345 1346 #define __KVM_HAVE_ARCH_VM_ALLOC 1347 struct kvm *kvm_arch_alloc_vm(void); 1348 1349 #define __KVM_HAVE_ARCH_FLUSH_REMOTE_TLBS 1350 1351 #define __KVM_HAVE_ARCH_FLUSH_REMOTE_TLBS_RANGE 1352 1353 #define kvm_vm_is_protected(kvm) (is_protected_kvm_enabled() && (kvm)->arch.pkvm.enabled) 1354 1355 #define vcpu_is_protected(vcpu) kvm_vm_is_protected((vcpu)->kvm) 1356 1357 int kvm_arm_vcpu_finalize(struct kvm_vcpu *vcpu, int feature); 1358 bool kvm_arm_vcpu_is_finalized(struct kvm_vcpu *vcpu); 1359 1360 #define kvm_arm_vcpu_sve_finalized(vcpu) vcpu_get_flag(vcpu, VCPU_SVE_FINALIZED) 1361 1362 #define kvm_has_mte(kvm) \ 1363 (system_supports_mte() && \ 1364 test_bit(KVM_ARCH_FLAG_MTE_ENABLED, &(kvm)->arch.flags)) 1365 1366 #define kvm_supports_32bit_el0() \ 1367 (system_supports_32bit_el0() && \ 1368 !static_branch_unlikely(&arm64_mismatched_32bit_el0)) 1369 1370 #define kvm_vm_has_ran_once(kvm) \ 1371 (test_bit(KVM_ARCH_FLAG_HAS_RAN_ONCE, &(kvm)->arch.flags)) 1372 1373 static inline bool __vcpu_has_feature(const struct kvm_arch *ka, int feature) 1374 { 1375 return test_bit(feature, ka->vcpu_features); 1376 } 1377 1378 #define vcpu_has_feature(v, f) __vcpu_has_feature(&(v)->kvm->arch, (f)) 1379 1380 #define kvm_vcpu_initialized(v) vcpu_get_flag(vcpu, VCPU_INITIALIZED) 1381 1382 int kvm_trng_call(struct kvm_vcpu *vcpu); 1383 #ifdef CONFIG_KVM 1384 extern phys_addr_t hyp_mem_base; 1385 extern phys_addr_t hyp_mem_size; 1386 void __init kvm_hyp_reserve(void); 1387 #else 1388 static inline void kvm_hyp_reserve(void) { } 1389 #endif 1390 1391 void kvm_arm_vcpu_power_off(struct kvm_vcpu *vcpu); 1392 bool kvm_arm_vcpu_stopped(struct kvm_vcpu *vcpu); 1393 1394 #define __expand_field_sign_unsigned(id, fld, val) \ 1395 ((u64)SYS_FIELD_VALUE(id, fld, val)) 1396 1397 #define __expand_field_sign_signed(id, fld, val) \ 1398 ({ \ 1399 u64 __val = SYS_FIELD_VALUE(id, fld, val); \ 1400 sign_extend64(__val, id##_##fld##_WIDTH - 1); \ 1401 }) 1402 1403 #define expand_field_sign(id, fld, val) \ 1404 (id##_##fld##_SIGNED ? \ 1405 __expand_field_sign_signed(id, fld, val) : \ 1406 __expand_field_sign_unsigned(id, fld, val)) 1407 1408 #define get_idreg_field_unsigned(kvm, id, fld) \ 1409 ({ \ 1410 u64 __val = IDREG((kvm), SYS_##id); \ 1411 FIELD_GET(id##_##fld##_MASK, __val); \ 1412 }) 1413 1414 #define get_idreg_field_signed(kvm, id, fld) \ 1415 ({ \ 1416 u64 __val = get_idreg_field_unsigned(kvm, id, fld); \ 1417 sign_extend64(__val, id##_##fld##_WIDTH - 1); \ 1418 }) 1419 1420 #define get_idreg_field_enum(kvm, id, fld) \ 1421 get_idreg_field_unsigned(kvm, id, fld) 1422 1423 #define get_idreg_field(kvm, id, fld) \ 1424 (id##_##fld##_SIGNED ? \ 1425 get_idreg_field_signed(kvm, id, fld) : \ 1426 get_idreg_field_unsigned(kvm, id, fld)) 1427 1428 #define kvm_has_feat(kvm, id, fld, limit) \ 1429 (get_idreg_field((kvm), id, fld) >= expand_field_sign(id, fld, limit)) 1430 1431 #define kvm_has_feat_enum(kvm, id, fld, val) \ 1432 (get_idreg_field_unsigned((kvm), id, fld) == __expand_field_sign_unsigned(id, fld, val)) 1433 1434 #define kvm_has_feat_range(kvm, id, fld, min, max) \ 1435 (get_idreg_field((kvm), id, fld) >= expand_field_sign(id, fld, min) && \ 1436 get_idreg_field((kvm), id, fld) <= expand_field_sign(id, fld, max)) 1437 1438 /* Check for a given level of PAuth support */ 1439 #define kvm_has_pauth(k, l) \ 1440 ({ \ 1441 bool pa, pi, pa3; \ 1442 \ 1443 pa = kvm_has_feat((k), ID_AA64ISAR1_EL1, APA, l); \ 1444 pa &= kvm_has_feat((k), ID_AA64ISAR1_EL1, GPA, IMP); \ 1445 pi = kvm_has_feat((k), ID_AA64ISAR1_EL1, API, l); \ 1446 pi &= kvm_has_feat((k), ID_AA64ISAR1_EL1, GPI, IMP); \ 1447 pa3 = kvm_has_feat((k), ID_AA64ISAR2_EL1, APA3, l); \ 1448 pa3 &= kvm_has_feat((k), ID_AA64ISAR2_EL1, GPA3, IMP); \ 1449 \ 1450 (pa + pi + pa3) == 1; \ 1451 }) 1452 1453 #endif /* __ARM64_KVM_HOST_H__ */ 1454