1 /* SPDX-License-Identifier: GPL-2.0-only */
3 * Copyright (C) 2012,2013 - ARM Ltd
4 * Author: Marc Zyngier <marc.zyngier@arm.com>
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>
11 #ifndef __ARM64_KVM_HOST_H__
12 #define __ARM64_KVM_HOST_H__
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>
29 #include <asm/kvm_asm.h>
31 #define __KVM_HAVE_ARCH_INTC_INITIALIZED
33 #define KVM_HALT_POLL_NS_DEFAULT 500000
35 #include <kvm/arm_vgic.h>
36 #include <kvm/arm_arch_timer.h>
37 #include <kvm/arm_pmu.h>
39 #define KVM_MAX_VCPUS VGIC_V3_MAX_CPUS
41 #define KVM_VCPU_MAX_FEATURES 7
42 #define KVM_VCPU_VALID_FEATURES (BIT(KVM_VCPU_MAX_FEATURES) - 1)
44 #define KVM_REQ_SLEEP \
45 KVM_ARCH_REQ_FLAGS(0, KVM_REQUEST_WAIT | KVM_REQUEST_NO_WAKEUP)
46 #define KVM_REQ_IRQ_PENDING KVM_ARCH_REQ(1)
47 #define KVM_REQ_VCPU_RESET KVM_ARCH_REQ(2)
48 #define KVM_REQ_RECORD_STEAL KVM_ARCH_REQ(3)
49 #define KVM_REQ_RELOAD_GICv4 KVM_ARCH_REQ(4)
50 #define KVM_REQ_RELOAD_PMU KVM_ARCH_REQ(5)
51 #define KVM_REQ_SUSPEND KVM_ARCH_REQ(6)
53 #define KVM_DIRTY_LOG_MANUAL_CAPS (KVM_DIRTY_LOG_MANUAL_PROTECT_ENABLE | \
54 KVM_DIRTY_LOG_INITIALLY_SET)
56 #define KVM_HAVE_MMU_RWLOCK
59 * Mode of operation configurable with kvm-arm.mode early param.
60 * See Documentation/admin-guide/kernel-parameters.txt for more information.
69 enum kvm_mode kvm_get_mode(void);
71 static inline enum kvm_mode kvm_get_mode(void) { return KVM_MODE_NONE; };
74 DECLARE_STATIC_KEY_FALSE(userspace_irqchip_in_use);
76 extern unsigned int __ro_after_init kvm_sve_max_vl;
77 int __init kvm_arm_init_sve(void);
79 u32 __attribute_const__ kvm_target_cpu(void);
80 int kvm_reset_vcpu(struct kvm_vcpu *vcpu);
81 void kvm_arm_vcpu_destroy(struct kvm_vcpu *vcpu);
83 struct kvm_hyp_memcache {
85 unsigned long nr_pages;
88 static inline void push_hyp_memcache(struct kvm_hyp_memcache *mc,
90 phys_addr_t (*to_pa)(void *virt))
97 static inline void *pop_hyp_memcache(struct kvm_hyp_memcache *mc,
98 void *(*to_va)(phys_addr_t phys))
100 phys_addr_t *p = to_va(mc->head);
111 static inline int __topup_hyp_memcache(struct kvm_hyp_memcache *mc,
112 unsigned long min_pages,
113 void *(*alloc_fn)(void *arg),
114 phys_addr_t (*to_pa)(void *virt),
117 while (mc->nr_pages < min_pages) {
118 phys_addr_t *p = alloc_fn(arg);
122 push_hyp_memcache(mc, p, to_pa);
128 static inline void __free_hyp_memcache(struct kvm_hyp_memcache *mc,
129 void (*free_fn)(void *virt, void *arg),
130 void *(*to_va)(phys_addr_t phys),
134 free_fn(pop_hyp_memcache(mc, to_va), arg);
137 void free_hyp_memcache(struct kvm_hyp_memcache *mc);
138 int topup_hyp_memcache(struct kvm_hyp_memcache *mc, unsigned long min_pages);
145 struct kvm_vmid vmid;
148 * stage2 entry level table
150 * Two kvm_s2_mmu structures in the same VM can point to the same
151 * pgd here. This happens when running a guest using a
152 * translation regime that isn't affected by its own stage-2
153 * translation, such as a non-VHE hypervisor running at vEL2, or
154 * for vEL1/EL0 with vHCR_EL2.VM == 0. In that case, we use the
155 * canonical stage-2 page tables.
157 phys_addr_t pgd_phys;
158 struct kvm_pgtable *pgt;
160 /* The last vcpu id that ran on each physical CPU */
161 int __percpu *last_vcpu_ran;
163 #define KVM_ARM_EAGER_SPLIT_CHUNK_SIZE_DEFAULT 0
165 * Memory cache used to split
166 * KVM_CAP_ARM_EAGER_SPLIT_CHUNK_SIZE worth of huge pages. It
167 * is used to allocate stage2 page tables while splitting huge
168 * pages. The choice of KVM_CAP_ARM_EAGER_SPLIT_CHUNK_SIZE
169 * influences both the capacity of the split page cache, and
170 * how often KVM reschedules. Be wary of raising CHUNK_SIZE
173 * Protected by kvm->slots_lock.
175 struct kvm_mmu_memory_cache split_page_cache;
176 uint64_t split_page_chunk_size;
178 struct kvm_arch *arch;
181 struct kvm_arch_memory_slot {
185 * struct kvm_smccc_features: Descriptor of the hypercall services exposed to the guests
187 * @std_bmap: Bitmap of standard secure service calls
188 * @std_hyp_bmap: Bitmap of standard hypervisor service calls
189 * @vendor_hyp_bmap: Bitmap of vendor specific hypervisor service calls
191 struct kvm_smccc_features {
192 unsigned long std_bmap;
193 unsigned long std_hyp_bmap;
194 unsigned long vendor_hyp_bmap;
197 typedef unsigned int pkvm_handle_t;
199 struct kvm_protected_vm {
200 pkvm_handle_t handle;
201 struct kvm_hyp_memcache teardown_mc;
205 struct kvm_s2_mmu mmu;
207 /* VTCR_EL2 value for this VM */
210 /* Interrupt controller */
211 struct vgic_dist vgic;
214 struct arch_timer_vm_data timer_data;
216 /* Mandated version of PSCI */
219 /* Protects VM-scoped configuration data */
220 struct mutex config_lock;
223 * If we encounter a data abort without valid instruction syndrome
224 * information, report this to user space. User space can (and
225 * should) opt in to this feature if KVM_CAP_ARM_NISV_TO_USER is
228 #define KVM_ARCH_FLAG_RETURN_NISV_IO_ABORT_TO_USER 0
229 /* Memory Tagging Extension enabled for the guest */
230 #define KVM_ARCH_FLAG_MTE_ENABLED 1
231 /* At least one vCPU has ran in the VM */
232 #define KVM_ARCH_FLAG_HAS_RAN_ONCE 2
233 /* The vCPU feature set for the VM is configured */
234 #define KVM_ARCH_FLAG_VCPU_FEATURES_CONFIGURED 3
235 /* PSCI SYSTEM_SUSPEND enabled for the guest */
236 #define KVM_ARCH_FLAG_SYSTEM_SUSPEND_ENABLED 4
237 /* VM counter offset */
238 #define KVM_ARCH_FLAG_VM_COUNTER_OFFSET 5
239 /* Timer PPIs made immutable */
240 #define KVM_ARCH_FLAG_TIMER_PPIS_IMMUTABLE 6
241 /* SMCCC filter initialized for the VM */
242 #define KVM_ARCH_FLAG_SMCCC_FILTER_CONFIGURED 7
243 /* Initial ID reg values loaded */
244 #define KVM_ARCH_FLAG_ID_REGS_INITIALIZED 8
247 /* VM-wide vCPU feature set */
248 DECLARE_BITMAP(vcpu_features, KVM_VCPU_MAX_FEATURES);
251 * VM-wide PMU filter, implemented as a bitmap and big enough for
252 * up to 2^10 events (ARMv8.0) or 2^16 events (ARMv8.1+).
254 unsigned long *pmu_filter;
255 struct arm_pmu *arm_pmu;
257 cpumask_var_t supported_cpus;
259 /* Hypercall features firmware registers' descriptor */
260 struct kvm_smccc_features smccc_feat;
261 struct maple_tree smccc_filter;
264 * Emulated CPU ID registers per VM
265 * (Op0, Op1, CRn, CRm, Op2) of the ID registers to be saved in it
266 * is (3, 0, 0, crm, op2), where 1<=crm<8, 0<=op2<8.
268 * These emulated idregs are VM-wide, but accessed from the context of a vCPU.
269 * Atomic access to multiple idregs are guarded by kvm_arch.config_lock.
271 #define IDREG_IDX(id) (((sys_reg_CRm(id) - 1) << 3) | sys_reg_Op2(id))
272 #define IDREG(kvm, id) ((kvm)->arch.id_regs[IDREG_IDX(id)])
273 #define KVM_ARM_ID_REG_NUM (IDREG_IDX(sys_reg(3, 0, 0, 7, 7)) + 1)
274 u64 id_regs[KVM_ARM_ID_REG_NUM];
277 * For an untrusted host VM, 'pkvm.handle' is used to lookup
278 * the associated pKVM instance in the hypervisor.
280 struct kvm_protected_vm pkvm;
283 struct kvm_vcpu_fault_info {
284 u64 esr_el2; /* Hyp Syndrom Register */
285 u64 far_el2; /* Hyp Fault Address Register */
286 u64 hpfar_el2; /* Hyp IPA Fault Address Register */
287 u64 disr_el1; /* Deferred [SError] Status Register */
291 __INVALID_SYSREG__, /* 0 is reserved as an invalid value */
292 MPIDR_EL1, /* MultiProcessor Affinity Register */
293 CLIDR_EL1, /* Cache Level ID Register */
294 CSSELR_EL1, /* Cache Size Selection Register */
295 SCTLR_EL1, /* System Control Register */
296 ACTLR_EL1, /* Auxiliary Control Register */
297 CPACR_EL1, /* Coprocessor Access Control */
298 ZCR_EL1, /* SVE Control */
299 TTBR0_EL1, /* Translation Table Base Register 0 */
300 TTBR1_EL1, /* Translation Table Base Register 1 */
301 TCR_EL1, /* Translation Control Register */
302 ESR_EL1, /* Exception Syndrome Register */
303 AFSR0_EL1, /* Auxiliary Fault Status Register 0 */
304 AFSR1_EL1, /* Auxiliary Fault Status Register 1 */
305 FAR_EL1, /* Fault Address Register */
306 MAIR_EL1, /* Memory Attribute Indirection Register */
307 VBAR_EL1, /* Vector Base Address Register */
308 CONTEXTIDR_EL1, /* Context ID Register */
309 TPIDR_EL0, /* Thread ID, User R/W */
310 TPIDRRO_EL0, /* Thread ID, User R/O */
311 TPIDR_EL1, /* Thread ID, Privileged */
312 AMAIR_EL1, /* Aux Memory Attribute Indirection Register */
313 CNTKCTL_EL1, /* Timer Control Register (EL1) */
314 PAR_EL1, /* Physical Address Register */
315 MDSCR_EL1, /* Monitor Debug System Control Register */
316 MDCCINT_EL1, /* Monitor Debug Comms Channel Interrupt Enable Reg */
317 OSLSR_EL1, /* OS Lock Status Register */
318 DISR_EL1, /* Deferred Interrupt Status Register */
320 /* Performance Monitors Registers */
321 PMCR_EL0, /* Control Register */
322 PMSELR_EL0, /* Event Counter Selection Register */
323 PMEVCNTR0_EL0, /* Event Counter Register (0-30) */
324 PMEVCNTR30_EL0 = PMEVCNTR0_EL0 + 30,
325 PMCCNTR_EL0, /* Cycle Counter Register */
326 PMEVTYPER0_EL0, /* Event Type Register (0-30) */
327 PMEVTYPER30_EL0 = PMEVTYPER0_EL0 + 30,
328 PMCCFILTR_EL0, /* Cycle Count Filter Register */
329 PMCNTENSET_EL0, /* Count Enable Set Register */
330 PMINTENSET_EL1, /* Interrupt Enable Set Register */
331 PMOVSSET_EL0, /* Overflow Flag Status Set Register */
332 PMUSERENR_EL0, /* User Enable Register */
334 /* Pointer Authentication Registers in a strict increasing order. */
356 /* Memory Tagging Extension registers */
357 RGSR_EL1, /* Random Allocation Tag Seed Register */
358 GCR_EL1, /* Tag Control Register */
359 TFSR_EL1, /* Tag Fault Status Register (EL1) */
360 TFSRE0_EL1, /* Tag Fault Status Register (EL0) */
362 /* 32bit specific registers. */
363 DACR32_EL2, /* Domain Access Control Register */
364 IFSR32_EL2, /* Instruction Fault Status Register */
365 FPEXC32_EL2, /* Floating-Point Exception Control Register */
366 DBGVCR32_EL2, /* Debug Vector Catch Register */
369 VPIDR_EL2, /* Virtualization Processor ID Register */
370 VMPIDR_EL2, /* Virtualization Multiprocessor ID Register */
371 SCTLR_EL2, /* System Control Register (EL2) */
372 ACTLR_EL2, /* Auxiliary Control Register (EL2) */
373 HCR_EL2, /* Hypervisor Configuration Register */
374 MDCR_EL2, /* Monitor Debug Configuration Register (EL2) */
375 CPTR_EL2, /* Architectural Feature Trap Register (EL2) */
376 HSTR_EL2, /* Hypervisor System Trap Register */
377 HACR_EL2, /* Hypervisor Auxiliary Control Register */
378 TTBR0_EL2, /* Translation Table Base Register 0 (EL2) */
379 TTBR1_EL2, /* Translation Table Base Register 1 (EL2) */
380 TCR_EL2, /* Translation Control Register (EL2) */
381 VTTBR_EL2, /* Virtualization Translation Table Base Register */
382 VTCR_EL2, /* Virtualization Translation Control Register */
383 SPSR_EL2, /* EL2 saved program status register */
384 ELR_EL2, /* EL2 exception link register */
385 AFSR0_EL2, /* Auxiliary Fault Status Register 0 (EL2) */
386 AFSR1_EL2, /* Auxiliary Fault Status Register 1 (EL2) */
387 ESR_EL2, /* Exception Syndrome Register (EL2) */
388 FAR_EL2, /* Fault Address Register (EL2) */
389 HPFAR_EL2, /* Hypervisor IPA Fault Address Register */
390 MAIR_EL2, /* Memory Attribute Indirection Register (EL2) */
391 AMAIR_EL2, /* Auxiliary Memory Attribute Indirection Register (EL2) */
392 VBAR_EL2, /* Vector Base Address Register (EL2) */
393 RVBAR_EL2, /* Reset Vector Base Address Register */
394 CONTEXTIDR_EL2, /* Context ID Register (EL2) */
395 TPIDR_EL2, /* EL2 Software Thread ID Register */
396 CNTHCTL_EL2, /* Counter-timer Hypervisor Control register */
397 SP_EL2, /* EL2 Stack Pointer */
403 NR_SYS_REGS /* Nothing after this line! */
406 struct kvm_cpu_context {
407 struct user_pt_regs regs; /* sp = sp_el0 */
414 struct user_fpsimd_state fp_regs;
416 u64 sys_regs[NR_SYS_REGS];
418 struct kvm_vcpu *__hyp_running_vcpu;
421 struct kvm_host_data {
422 struct kvm_cpu_context host_ctxt;
425 struct kvm_host_psci_config {
426 /* PSCI version used by host. */
430 /* Function IDs used by host if version is v0.1. */
431 struct psci_0_1_function_ids function_ids_0_1;
433 bool psci_0_1_cpu_suspend_implemented;
434 bool psci_0_1_cpu_on_implemented;
435 bool psci_0_1_cpu_off_implemented;
436 bool psci_0_1_migrate_implemented;
439 extern struct kvm_host_psci_config kvm_nvhe_sym(kvm_host_psci_config);
440 #define kvm_host_psci_config CHOOSE_NVHE_SYM(kvm_host_psci_config)
442 extern s64 kvm_nvhe_sym(hyp_physvirt_offset);
443 #define hyp_physvirt_offset CHOOSE_NVHE_SYM(hyp_physvirt_offset)
445 extern u64 kvm_nvhe_sym(hyp_cpu_logical_map)[NR_CPUS];
446 #define hyp_cpu_logical_map CHOOSE_NVHE_SYM(hyp_cpu_logical_map)
448 struct vcpu_reset_state {
455 struct kvm_vcpu_arch {
456 struct kvm_cpu_context ctxt;
459 * Guest floating point state
461 * The architecture has two main floating point extensions,
462 * the original FPSIMD and SVE. These have overlapping
463 * register views, with the FPSIMD V registers occupying the
464 * low 128 bits of the SVE Z registers. When the core
465 * floating point code saves the register state of a task it
466 * records which view it saved in fp_type.
469 enum fp_type fp_type;
470 unsigned int sve_max_vl;
473 /* Stage 2 paging state used by the hardware on next switch */
474 struct kvm_s2_mmu *hw_mmu;
476 /* Values of trap registers for the guest. */
481 /* Values of trap registers for the host before guest entry. */
484 /* Exception Information */
485 struct kvm_vcpu_fault_info fault;
487 /* Ownership of the FP regs */
491 FP_STATE_GUEST_OWNED,
494 /* Configuration flags, set once and for all before the vcpu can run */
497 /* Input flags to the hypervisor code, potentially cleared after use */
500 /* State flags for kernel bookkeeping, unused by the hypervisor code */
504 * Don't run the guest (internal implementation need).
506 * Contrary to the flags above, this is set/cleared outside of
507 * a vcpu context, and thus cannot be mixed with the flags
508 * themselves (or the flag accesses need to be made atomic).
513 * We maintain more than a single set of debug registers to support
514 * debugging the guest from the host and to maintain separate host and
515 * guest state during world switches. vcpu_debug_state are the debug
516 * registers of the vcpu as the guest sees them. host_debug_state are
517 * the host registers which are saved and restored during
518 * world switches. external_debug_state contains the debug
519 * values we want to debug the guest. This is set via the
520 * KVM_SET_GUEST_DEBUG ioctl.
522 * debug_ptr points to the set of debug registers that should be loaded
523 * onto the hardware when running the guest.
525 struct kvm_guest_debug_arch *debug_ptr;
526 struct kvm_guest_debug_arch vcpu_debug_state;
527 struct kvm_guest_debug_arch external_debug_state;
529 struct user_fpsimd_state *host_fpsimd_state; /* hyp VA */
530 struct task_struct *parent_task;
533 /* {Break,watch}point registers */
534 struct kvm_guest_debug_arch regs;
535 /* Statistical profiling extension */
537 /* Self-hosted trace */
542 struct vgic_cpu vgic_cpu;
543 struct arch_timer_cpu timer_cpu;
547 * Guest registers we preserve during guest debugging.
549 * These shadow registers are updated by the kvm_handle_sys_reg
550 * trap handler if the guest accesses or updates them while we
551 * are using guest debug.
556 } guest_debug_preserved;
558 /* vcpu power state */
559 struct kvm_mp_state mp_state;
560 spinlock_t mp_state_lock;
562 /* Cache some mmu pages needed inside spinlock regions */
563 struct kvm_mmu_memory_cache mmu_page_cache;
565 /* Target CPU and feature flags */
567 DECLARE_BITMAP(features, KVM_VCPU_MAX_FEATURES);
569 /* Virtual SError ESR to restore when HCR_EL2.VSE is set */
572 /* Additional reset state */
573 struct vcpu_reset_state reset_state;
581 /* Per-vcpu CCSIDR override or NULL */
586 * Each 'flag' is composed of a comma-separated triplet:
588 * - the flag-set it belongs to in the vcpu->arch structure
589 * - the value for that flag
590 * - the mask for that flag
592 * __vcpu_single_flag() builds such a triplet for a single-bit flag.
593 * unpack_vcpu_flag() extract the flag value from the triplet for
594 * direct use outside of the flag accessors.
596 #define __vcpu_single_flag(_set, _f) _set, (_f), (_f)
598 #define __unpack_flag(_set, _f, _m) _f
599 #define unpack_vcpu_flag(...) __unpack_flag(__VA_ARGS__)
601 #define __build_check_flag(v, flagset, f, m) \
603 typeof(v->arch.flagset) *_fset; \
605 /* Check that the flags fit in the mask */ \
606 BUILD_BUG_ON(HWEIGHT(m) != HWEIGHT((f) | (m))); \
607 /* Check that the flags fit in the type */ \
608 BUILD_BUG_ON((sizeof(*_fset) * 8) <= __fls(m)); \
611 #define __vcpu_get_flag(v, flagset, f, m) \
613 __build_check_flag(v, flagset, f, m); \
615 READ_ONCE(v->arch.flagset) & (m); \
619 * Note that the set/clear accessors must be preempt-safe in order to
620 * avoid nesting them with load/put which also manipulate flags...
622 #ifdef __KVM_NVHE_HYPERVISOR__
623 /* the nVHE hypervisor is always non-preemptible */
624 #define __vcpu_flags_preempt_disable()
625 #define __vcpu_flags_preempt_enable()
627 #define __vcpu_flags_preempt_disable() preempt_disable()
628 #define __vcpu_flags_preempt_enable() preempt_enable()
631 #define __vcpu_set_flag(v, flagset, f, m) \
633 typeof(v->arch.flagset) *fset; \
635 __build_check_flag(v, flagset, f, m); \
637 fset = &v->arch.flagset; \
638 __vcpu_flags_preempt_disable(); \
639 if (HWEIGHT(m) > 1) \
642 __vcpu_flags_preempt_enable(); \
645 #define __vcpu_clear_flag(v, flagset, f, m) \
647 typeof(v->arch.flagset) *fset; \
649 __build_check_flag(v, flagset, f, m); \
651 fset = &v->arch.flagset; \
652 __vcpu_flags_preempt_disable(); \
654 __vcpu_flags_preempt_enable(); \
657 #define vcpu_get_flag(v, ...) __vcpu_get_flag((v), __VA_ARGS__)
658 #define vcpu_set_flag(v, ...) __vcpu_set_flag((v), __VA_ARGS__)
659 #define vcpu_clear_flag(v, ...) __vcpu_clear_flag((v), __VA_ARGS__)
661 /* SVE exposed to guest */
662 #define GUEST_HAS_SVE __vcpu_single_flag(cflags, BIT(0))
663 /* SVE config completed */
664 #define VCPU_SVE_FINALIZED __vcpu_single_flag(cflags, BIT(1))
665 /* PTRAUTH exposed to guest */
666 #define GUEST_HAS_PTRAUTH __vcpu_single_flag(cflags, BIT(2))
668 /* Exception pending */
669 #define PENDING_EXCEPTION __vcpu_single_flag(iflags, BIT(0))
671 * PC increment. Overlaps with EXCEPT_MASK on purpose so that it can't
672 * be set together with an exception...
674 #define INCREMENT_PC __vcpu_single_flag(iflags, BIT(1))
675 /* Target EL/MODE (not a single flag, but let's abuse the macro) */
676 #define EXCEPT_MASK __vcpu_single_flag(iflags, GENMASK(3, 1))
678 /* Helpers to encode exceptions with minimum fuss */
679 #define __EXCEPT_MASK_VAL unpack_vcpu_flag(EXCEPT_MASK)
680 #define __EXCEPT_SHIFT __builtin_ctzl(__EXCEPT_MASK_VAL)
681 #define __vcpu_except_flags(_f) iflags, (_f << __EXCEPT_SHIFT), __EXCEPT_MASK_VAL
684 * When PENDING_EXCEPTION is set, EXCEPT_MASK can take the following
689 #define EXCEPT_AA32_UND __vcpu_except_flags(0)
690 #define EXCEPT_AA32_IABT __vcpu_except_flags(1)
691 #define EXCEPT_AA32_DABT __vcpu_except_flags(2)
693 #define EXCEPT_AA64_EL1_SYNC __vcpu_except_flags(0)
694 #define EXCEPT_AA64_EL1_IRQ __vcpu_except_flags(1)
695 #define EXCEPT_AA64_EL1_FIQ __vcpu_except_flags(2)
696 #define EXCEPT_AA64_EL1_SERR __vcpu_except_flags(3)
697 /* For AArch64 with NV: */
698 #define EXCEPT_AA64_EL2_SYNC __vcpu_except_flags(4)
699 #define EXCEPT_AA64_EL2_IRQ __vcpu_except_flags(5)
700 #define EXCEPT_AA64_EL2_FIQ __vcpu_except_flags(6)
701 #define EXCEPT_AA64_EL2_SERR __vcpu_except_flags(7)
702 /* Guest debug is live */
703 #define DEBUG_DIRTY __vcpu_single_flag(iflags, BIT(4))
704 /* Save SPE context if active */
705 #define DEBUG_STATE_SAVE_SPE __vcpu_single_flag(iflags, BIT(5))
706 /* Save TRBE context if active */
707 #define DEBUG_STATE_SAVE_TRBE __vcpu_single_flag(iflags, BIT(6))
708 /* vcpu running in HYP context */
709 #define VCPU_HYP_CONTEXT __vcpu_single_flag(iflags, BIT(7))
711 /* SVE enabled for host EL0 */
712 #define HOST_SVE_ENABLED __vcpu_single_flag(sflags, BIT(0))
713 /* SME enabled for EL0 */
714 #define HOST_SME_ENABLED __vcpu_single_flag(sflags, BIT(1))
715 /* Physical CPU not in supported_cpus */
716 #define ON_UNSUPPORTED_CPU __vcpu_single_flag(sflags, BIT(2))
717 /* WFIT instruction trapped */
718 #define IN_WFIT __vcpu_single_flag(sflags, BIT(3))
719 /* vcpu system registers loaded on physical CPU */
720 #define SYSREGS_ON_CPU __vcpu_single_flag(sflags, BIT(4))
721 /* Software step state is Active-pending */
722 #define DBG_SS_ACTIVE_PENDING __vcpu_single_flag(sflags, BIT(5))
723 /* PMUSERENR for the guest EL0 is on physical CPU */
724 #define PMUSERENR_ON_CPU __vcpu_single_flag(sflags, BIT(6))
727 /* Pointer to the vcpu's SVE FFR for sve_{save,load}_state() */
728 #define vcpu_sve_pffr(vcpu) (kern_hyp_va((vcpu)->arch.sve_state) + \
729 sve_ffr_offset((vcpu)->arch.sve_max_vl))
731 #define vcpu_sve_max_vq(vcpu) sve_vq_from_vl((vcpu)->arch.sve_max_vl)
733 #define vcpu_sve_state_size(vcpu) ({ \
735 unsigned int __vcpu_vq; \
737 if (WARN_ON(!sve_vl_valid((vcpu)->arch.sve_max_vl))) { \
740 __vcpu_vq = vcpu_sve_max_vq(vcpu); \
741 __size_ret = SVE_SIG_REGS_SIZE(__vcpu_vq); \
747 #define KVM_GUESTDBG_VALID_MASK (KVM_GUESTDBG_ENABLE | \
748 KVM_GUESTDBG_USE_SW_BP | \
749 KVM_GUESTDBG_USE_HW | \
750 KVM_GUESTDBG_SINGLESTEP)
752 #define vcpu_has_sve(vcpu) (system_supports_sve() && \
753 vcpu_get_flag(vcpu, GUEST_HAS_SVE))
755 #ifdef CONFIG_ARM64_PTR_AUTH
756 #define vcpu_has_ptrauth(vcpu) \
757 ((cpus_have_final_cap(ARM64_HAS_ADDRESS_AUTH) || \
758 cpus_have_final_cap(ARM64_HAS_GENERIC_AUTH)) && \
759 vcpu_get_flag(vcpu, GUEST_HAS_PTRAUTH))
761 #define vcpu_has_ptrauth(vcpu) false
764 #define vcpu_on_unsupported_cpu(vcpu) \
765 vcpu_get_flag(vcpu, ON_UNSUPPORTED_CPU)
767 #define vcpu_set_on_unsupported_cpu(vcpu) \
768 vcpu_set_flag(vcpu, ON_UNSUPPORTED_CPU)
770 #define vcpu_clear_on_unsupported_cpu(vcpu) \
771 vcpu_clear_flag(vcpu, ON_UNSUPPORTED_CPU)
773 #define vcpu_gp_regs(v) (&(v)->arch.ctxt.regs)
776 * Only use __vcpu_sys_reg/ctxt_sys_reg if you know you want the
777 * memory backed version of a register, and not the one most recently
778 * accessed by a running VCPU. For example, for userspace access or
779 * for system registers that are never context switched, but only
782 #define __ctxt_sys_reg(c,r) (&(c)->sys_regs[(r)])
784 #define ctxt_sys_reg(c,r) (*__ctxt_sys_reg(c,r))
786 #define __vcpu_sys_reg(v,r) (ctxt_sys_reg(&(v)->arch.ctxt, (r)))
788 u64 vcpu_read_sys_reg(const struct kvm_vcpu *vcpu, int reg);
789 void vcpu_write_sys_reg(struct kvm_vcpu *vcpu, u64 val, int reg);
791 static inline bool __vcpu_read_sys_reg_from_cpu(int reg, u64 *val)
796 * System registers listed in the switch are not saved on every
797 * exit from the guest but are only saved on vcpu_put.
799 * Note that MPIDR_EL1 for the guest is set by KVM via VMPIDR_EL2 but
800 * should never be listed below, because the guest cannot modify its
801 * own MPIDR_EL1 and MPIDR_EL1 is accessed for VCPU A from VCPU B's
802 * thread when emulating cross-VCPU communication.
808 case SCTLR_EL1: *val = read_sysreg_s(SYS_SCTLR_EL12); break;
809 case CPACR_EL1: *val = read_sysreg_s(SYS_CPACR_EL12); break;
810 case TTBR0_EL1: *val = read_sysreg_s(SYS_TTBR0_EL12); break;
811 case TTBR1_EL1: *val = read_sysreg_s(SYS_TTBR1_EL12); break;
812 case TCR_EL1: *val = read_sysreg_s(SYS_TCR_EL12); break;
813 case ESR_EL1: *val = read_sysreg_s(SYS_ESR_EL12); break;
814 case AFSR0_EL1: *val = read_sysreg_s(SYS_AFSR0_EL12); break;
815 case AFSR1_EL1: *val = read_sysreg_s(SYS_AFSR1_EL12); break;
816 case FAR_EL1: *val = read_sysreg_s(SYS_FAR_EL12); break;
817 case MAIR_EL1: *val = read_sysreg_s(SYS_MAIR_EL12); break;
818 case VBAR_EL1: *val = read_sysreg_s(SYS_VBAR_EL12); break;
819 case CONTEXTIDR_EL1: *val = read_sysreg_s(SYS_CONTEXTIDR_EL12);break;
820 case TPIDR_EL0: *val = read_sysreg_s(SYS_TPIDR_EL0); break;
821 case TPIDRRO_EL0: *val = read_sysreg_s(SYS_TPIDRRO_EL0); break;
822 case TPIDR_EL1: *val = read_sysreg_s(SYS_TPIDR_EL1); break;
823 case AMAIR_EL1: *val = read_sysreg_s(SYS_AMAIR_EL12); break;
824 case CNTKCTL_EL1: *val = read_sysreg_s(SYS_CNTKCTL_EL12); break;
825 case ELR_EL1: *val = read_sysreg_s(SYS_ELR_EL12); break;
826 case PAR_EL1: *val = read_sysreg_par(); break;
827 case DACR32_EL2: *val = read_sysreg_s(SYS_DACR32_EL2); break;
828 case IFSR32_EL2: *val = read_sysreg_s(SYS_IFSR32_EL2); break;
829 case DBGVCR32_EL2: *val = read_sysreg_s(SYS_DBGVCR32_EL2); break;
830 default: return false;
836 static inline bool __vcpu_write_sys_reg_to_cpu(u64 val, int reg)
841 * System registers listed in the switch are not restored on every
842 * entry to the guest but are only restored on vcpu_load.
844 * Note that MPIDR_EL1 for the guest is set by KVM via VMPIDR_EL2 but
845 * should never be listed below, because the MPIDR should only be set
846 * once, before running the VCPU, and never changed later.
852 case SCTLR_EL1: write_sysreg_s(val, SYS_SCTLR_EL12); break;
853 case CPACR_EL1: write_sysreg_s(val, SYS_CPACR_EL12); break;
854 case TTBR0_EL1: write_sysreg_s(val, SYS_TTBR0_EL12); break;
855 case TTBR1_EL1: write_sysreg_s(val, SYS_TTBR1_EL12); break;
856 case TCR_EL1: write_sysreg_s(val, SYS_TCR_EL12); break;
857 case ESR_EL1: write_sysreg_s(val, SYS_ESR_EL12); break;
858 case AFSR0_EL1: write_sysreg_s(val, SYS_AFSR0_EL12); break;
859 case AFSR1_EL1: write_sysreg_s(val, SYS_AFSR1_EL12); break;
860 case FAR_EL1: write_sysreg_s(val, SYS_FAR_EL12); break;
861 case MAIR_EL1: write_sysreg_s(val, SYS_MAIR_EL12); break;
862 case VBAR_EL1: write_sysreg_s(val, SYS_VBAR_EL12); break;
863 case CONTEXTIDR_EL1: write_sysreg_s(val, SYS_CONTEXTIDR_EL12);break;
864 case TPIDR_EL0: write_sysreg_s(val, SYS_TPIDR_EL0); break;
865 case TPIDRRO_EL0: write_sysreg_s(val, SYS_TPIDRRO_EL0); break;
866 case TPIDR_EL1: write_sysreg_s(val, SYS_TPIDR_EL1); break;
867 case AMAIR_EL1: write_sysreg_s(val, SYS_AMAIR_EL12); break;
868 case CNTKCTL_EL1: write_sysreg_s(val, SYS_CNTKCTL_EL12); break;
869 case ELR_EL1: write_sysreg_s(val, SYS_ELR_EL12); break;
870 case PAR_EL1: write_sysreg_s(val, SYS_PAR_EL1); break;
871 case DACR32_EL2: write_sysreg_s(val, SYS_DACR32_EL2); break;
872 case IFSR32_EL2: write_sysreg_s(val, SYS_IFSR32_EL2); break;
873 case DBGVCR32_EL2: write_sysreg_s(val, SYS_DBGVCR32_EL2); break;
874 default: return false;
881 struct kvm_vm_stat_generic generic;
884 struct kvm_vcpu_stat {
885 struct kvm_vcpu_stat_generic generic;
890 u64 mmio_exit_kernel;
895 void kvm_vcpu_preferred_target(struct kvm_vcpu_init *init);
896 unsigned long kvm_arm_num_regs(struct kvm_vcpu *vcpu);
897 int kvm_arm_copy_reg_indices(struct kvm_vcpu *vcpu, u64 __user *indices);
898 int kvm_arm_get_reg(struct kvm_vcpu *vcpu, const struct kvm_one_reg *reg);
899 int kvm_arm_set_reg(struct kvm_vcpu *vcpu, const struct kvm_one_reg *reg);
901 unsigned long kvm_arm_num_sys_reg_descs(struct kvm_vcpu *vcpu);
902 int kvm_arm_copy_sys_reg_indices(struct kvm_vcpu *vcpu, u64 __user *uindices);
904 int __kvm_arm_vcpu_get_events(struct kvm_vcpu *vcpu,
905 struct kvm_vcpu_events *events);
907 int __kvm_arm_vcpu_set_events(struct kvm_vcpu *vcpu,
908 struct kvm_vcpu_events *events);
910 #define KVM_ARCH_WANT_MMU_NOTIFIER
912 void kvm_arm_halt_guest(struct kvm *kvm);
913 void kvm_arm_resume_guest(struct kvm *kvm);
915 #define vcpu_has_run_once(vcpu) !!rcu_access_pointer((vcpu)->pid)
917 #ifndef __KVM_NVHE_HYPERVISOR__
918 #define kvm_call_hyp_nvhe(f, ...) \
920 struct arm_smccc_res res; \
922 arm_smccc_1_1_hvc(KVM_HOST_SMCCC_FUNC(f), \
923 ##__VA_ARGS__, &res); \
924 WARN_ON(res.a0 != SMCCC_RET_SUCCESS); \
930 * The couple of isb() below are there to guarantee the same behaviour
931 * on VHE as on !VHE, where the eret to EL1 acts as a context
932 * synchronization event.
934 #define kvm_call_hyp(f, ...) \
940 kvm_call_hyp_nvhe(f, ##__VA_ARGS__); \
944 #define kvm_call_hyp_ret(f, ...) \
946 typeof(f(__VA_ARGS__)) ret; \
949 ret = f(__VA_ARGS__); \
952 ret = kvm_call_hyp_nvhe(f, ##__VA_ARGS__); \
957 #else /* __KVM_NVHE_HYPERVISOR__ */
958 #define kvm_call_hyp(f, ...) f(__VA_ARGS__)
959 #define kvm_call_hyp_ret(f, ...) f(__VA_ARGS__)
960 #define kvm_call_hyp_nvhe(f, ...) f(__VA_ARGS__)
961 #endif /* __KVM_NVHE_HYPERVISOR__ */
963 void force_vm_exit(const cpumask_t *mask);
965 int handle_exit(struct kvm_vcpu *vcpu, int exception_index);
966 void handle_exit_early(struct kvm_vcpu *vcpu, int exception_index);
968 int kvm_handle_cp14_load_store(struct kvm_vcpu *vcpu);
969 int kvm_handle_cp14_32(struct kvm_vcpu *vcpu);
970 int kvm_handle_cp14_64(struct kvm_vcpu *vcpu);
971 int kvm_handle_cp15_32(struct kvm_vcpu *vcpu);
972 int kvm_handle_cp15_64(struct kvm_vcpu *vcpu);
973 int kvm_handle_sys_reg(struct kvm_vcpu *vcpu);
974 int kvm_handle_cp10_id(struct kvm_vcpu *vcpu);
976 void kvm_reset_sys_regs(struct kvm_vcpu *vcpu);
978 int __init kvm_sys_reg_table_init(void);
980 bool lock_all_vcpus(struct kvm *kvm);
981 void unlock_all_vcpus(struct kvm *kvm);
984 void kvm_mmio_write_buf(void *buf, unsigned int len, unsigned long data);
985 unsigned long kvm_mmio_read_buf(const void *buf, unsigned int len);
987 int kvm_handle_mmio_return(struct kvm_vcpu *vcpu);
988 int io_mem_abort(struct kvm_vcpu *vcpu, phys_addr_t fault_ipa);
991 * Returns true if a Performance Monitoring Interrupt (PMI), a.k.a. perf event,
992 * arrived in guest context. For arm64, any event that arrives while a vCPU is
993 * loaded is considered to be "in guest".
995 static inline bool kvm_arch_pmi_in_guest(struct kvm_vcpu *vcpu)
997 return IS_ENABLED(CONFIG_GUEST_PERF_EVENTS) && !!vcpu;
1000 long kvm_hypercall_pv_features(struct kvm_vcpu *vcpu);
1001 gpa_t kvm_init_stolen_time(struct kvm_vcpu *vcpu);
1002 void kvm_update_stolen_time(struct kvm_vcpu *vcpu);
1004 bool kvm_arm_pvtime_supported(void);
1005 int kvm_arm_pvtime_set_attr(struct kvm_vcpu *vcpu,
1006 struct kvm_device_attr *attr);
1007 int kvm_arm_pvtime_get_attr(struct kvm_vcpu *vcpu,
1008 struct kvm_device_attr *attr);
1009 int kvm_arm_pvtime_has_attr(struct kvm_vcpu *vcpu,
1010 struct kvm_device_attr *attr);
1012 extern unsigned int __ro_after_init kvm_arm_vmid_bits;
1013 int __init kvm_arm_vmid_alloc_init(void);
1014 void __init kvm_arm_vmid_alloc_free(void);
1015 void kvm_arm_vmid_update(struct kvm_vmid *kvm_vmid);
1016 void kvm_arm_vmid_clear_active(void);
1018 static inline void kvm_arm_pvtime_vcpu_init(struct kvm_vcpu_arch *vcpu_arch)
1020 vcpu_arch->steal.base = INVALID_GPA;
1023 static inline bool kvm_arm_is_pvtime_enabled(struct kvm_vcpu_arch *vcpu_arch)
1025 return (vcpu_arch->steal.base != INVALID_GPA);
1028 void kvm_set_sei_esr(struct kvm_vcpu *vcpu, u64 syndrome);
1030 struct kvm_vcpu *kvm_mpidr_to_vcpu(struct kvm *kvm, unsigned long mpidr);
1032 DECLARE_KVM_HYP_PER_CPU(struct kvm_host_data, kvm_host_data);
1034 static inline void kvm_init_host_cpu_context(struct kvm_cpu_context *cpu_ctxt)
1036 /* The host's MPIDR is immutable, so let's set it up at boot time */
1037 ctxt_sys_reg(cpu_ctxt, MPIDR_EL1) = read_cpuid_mpidr();
1040 static inline bool kvm_system_needs_idmapped_vectors(void)
1042 return cpus_have_const_cap(ARM64_SPECTRE_V3A);
1045 void kvm_arm_vcpu_ptrauth_trap(struct kvm_vcpu *vcpu);
1047 static inline void kvm_arch_sync_events(struct kvm *kvm) {}
1048 static inline void kvm_arch_sched_in(struct kvm_vcpu *vcpu, int cpu) {}
1050 void kvm_arm_init_debug(void);
1051 void kvm_arm_vcpu_init_debug(struct kvm_vcpu *vcpu);
1052 void kvm_arm_setup_debug(struct kvm_vcpu *vcpu);
1053 void kvm_arm_clear_debug(struct kvm_vcpu *vcpu);
1054 void kvm_arm_reset_debug_ptr(struct kvm_vcpu *vcpu);
1056 #define kvm_vcpu_os_lock_enabled(vcpu) \
1057 (!!(__vcpu_sys_reg(vcpu, OSLSR_EL1) & SYS_OSLSR_OSLK))
1059 int kvm_arm_vcpu_arch_set_attr(struct kvm_vcpu *vcpu,
1060 struct kvm_device_attr *attr);
1061 int kvm_arm_vcpu_arch_get_attr(struct kvm_vcpu *vcpu,
1062 struct kvm_device_attr *attr);
1063 int kvm_arm_vcpu_arch_has_attr(struct kvm_vcpu *vcpu,
1064 struct kvm_device_attr *attr);
1066 int kvm_vm_ioctl_mte_copy_tags(struct kvm *kvm,
1067 struct kvm_arm_copy_mte_tags *copy_tags);
1068 int kvm_vm_ioctl_set_counter_offset(struct kvm *kvm,
1069 struct kvm_arm_counter_offset *offset);
1071 /* Guest/host FPSIMD coordination helpers */
1072 int kvm_arch_vcpu_run_map_fp(struct kvm_vcpu *vcpu);
1073 void kvm_arch_vcpu_load_fp(struct kvm_vcpu *vcpu);
1074 void kvm_arch_vcpu_ctxflush_fp(struct kvm_vcpu *vcpu);
1075 void kvm_arch_vcpu_ctxsync_fp(struct kvm_vcpu *vcpu);
1076 void kvm_arch_vcpu_put_fp(struct kvm_vcpu *vcpu);
1077 void kvm_vcpu_unshare_task_fp(struct kvm_vcpu *vcpu);
1079 static inline bool kvm_pmu_counter_deferred(struct perf_event_attr *attr)
1081 return (!has_vhe() && attr->exclude_host);
1084 /* Flags for host debug state */
1085 void kvm_arch_vcpu_load_debug_state_flags(struct kvm_vcpu *vcpu);
1086 void kvm_arch_vcpu_put_debug_state_flags(struct kvm_vcpu *vcpu);
1089 void kvm_set_pmu_events(u32 set, struct perf_event_attr *attr);
1090 void kvm_clr_pmu_events(u32 clr);
1091 bool kvm_set_pmuserenr(u64 val);
1093 static inline void kvm_set_pmu_events(u32 set, struct perf_event_attr *attr) {}
1094 static inline void kvm_clr_pmu_events(u32 clr) {}
1095 static inline bool kvm_set_pmuserenr(u64 val)
1101 void kvm_vcpu_load_sysregs_vhe(struct kvm_vcpu *vcpu);
1102 void kvm_vcpu_put_sysregs_vhe(struct kvm_vcpu *vcpu);
1104 int __init kvm_set_ipa_limit(void);
1106 #define __KVM_HAVE_ARCH_VM_ALLOC
1107 struct kvm *kvm_arch_alloc_vm(void);
1109 static inline bool kvm_vm_is_protected(struct kvm *kvm)
1114 void kvm_init_protected_traps(struct kvm_vcpu *vcpu);
1116 int kvm_arm_vcpu_finalize(struct kvm_vcpu *vcpu, int feature);
1117 bool kvm_arm_vcpu_is_finalized(struct kvm_vcpu *vcpu);
1119 #define kvm_arm_vcpu_sve_finalized(vcpu) vcpu_get_flag(vcpu, VCPU_SVE_FINALIZED)
1121 #define kvm_has_mte(kvm) \
1122 (system_supports_mte() && \
1123 test_bit(KVM_ARCH_FLAG_MTE_ENABLED, &(kvm)->arch.flags))
1125 #define kvm_supports_32bit_el0() \
1126 (system_supports_32bit_el0() && \
1127 !static_branch_unlikely(&arm64_mismatched_32bit_el0))
1129 #define kvm_vm_has_ran_once(kvm) \
1130 (test_bit(KVM_ARCH_FLAG_HAS_RAN_ONCE, &(kvm)->arch.flags))
1132 int kvm_trng_call(struct kvm_vcpu *vcpu);
1134 extern phys_addr_t hyp_mem_base;
1135 extern phys_addr_t hyp_mem_size;
1136 void __init kvm_hyp_reserve(void);
1138 static inline void kvm_hyp_reserve(void) { }
1141 void kvm_arm_vcpu_power_off(struct kvm_vcpu *vcpu);
1142 bool kvm_arm_vcpu_stopped(struct kvm_vcpu *vcpu);
1144 #endif /* __ARM64_KVM_HOST_H__ */