xref: /openbmc/linux/arch/arm64/include/asm/kvm_host.h (revision 2169e6da)
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/bitmap.h>
15 #include <linux/types.h>
16 #include <linux/jump_label.h>
17 #include <linux/kvm_types.h>
18 #include <linux/percpu.h>
19 #include <asm/arch_gicv3.h>
20 #include <asm/barrier.h>
21 #include <asm/cpufeature.h>
22 #include <asm/cputype.h>
23 #include <asm/daifflags.h>
24 #include <asm/fpsimd.h>
25 #include <asm/kvm.h>
26 #include <asm/kvm_asm.h>
27 #include <asm/kvm_mmio.h>
28 #include <asm/thread_info.h>
29 
30 #define __KVM_HAVE_ARCH_INTC_INITIALIZED
31 
32 #define KVM_USER_MEM_SLOTS 512
33 #define KVM_HALT_POLL_NS_DEFAULT 500000
34 
35 #include <kvm/arm_vgic.h>
36 #include <kvm/arm_arch_timer.h>
37 #include <kvm/arm_pmu.h>
38 
39 #define KVM_MAX_VCPUS VGIC_V3_MAX_CPUS
40 
41 #define KVM_VCPU_MAX_FEATURES 7
42 
43 #define KVM_REQ_SLEEP \
44 	KVM_ARCH_REQ_FLAGS(0, KVM_REQUEST_WAIT | KVM_REQUEST_NO_WAKEUP)
45 #define KVM_REQ_IRQ_PENDING	KVM_ARCH_REQ(1)
46 #define KVM_REQ_VCPU_RESET	KVM_ARCH_REQ(2)
47 
48 DECLARE_STATIC_KEY_FALSE(userspace_irqchip_in_use);
49 
50 extern unsigned int kvm_sve_max_vl;
51 int kvm_arm_init_sve(void);
52 
53 int __attribute_const__ kvm_target_cpu(void);
54 int kvm_reset_vcpu(struct kvm_vcpu *vcpu);
55 void kvm_arch_vcpu_uninit(struct kvm_vcpu *vcpu);
56 int kvm_arch_vm_ioctl_check_extension(struct kvm *kvm, long ext);
57 void __extended_idmap_trampoline(phys_addr_t boot_pgd, phys_addr_t idmap_start);
58 
59 struct kvm_vmid {
60 	/* The VMID generation used for the virt. memory system */
61 	u64    vmid_gen;
62 	u32    vmid;
63 };
64 
65 struct kvm_arch {
66 	struct kvm_vmid vmid;
67 
68 	/* stage2 entry level table */
69 	pgd_t *pgd;
70 	phys_addr_t pgd_phys;
71 
72 	/* VTCR_EL2 value for this VM */
73 	u64    vtcr;
74 
75 	/* The last vcpu id that ran on each physical CPU */
76 	int __percpu *last_vcpu_ran;
77 
78 	/* The maximum number of vCPUs depends on the used GIC model */
79 	int max_vcpus;
80 
81 	/* Interrupt controller */
82 	struct vgic_dist	vgic;
83 
84 	/* Mandated version of PSCI */
85 	u32 psci_version;
86 };
87 
88 #define KVM_NR_MEM_OBJS     40
89 
90 /*
91  * We don't want allocation failures within the mmu code, so we preallocate
92  * enough memory for a single page fault in a cache.
93  */
94 struct kvm_mmu_memory_cache {
95 	int nobjs;
96 	void *objects[KVM_NR_MEM_OBJS];
97 };
98 
99 struct kvm_vcpu_fault_info {
100 	u32 esr_el2;		/* Hyp Syndrom Register */
101 	u64 far_el2;		/* Hyp Fault Address Register */
102 	u64 hpfar_el2;		/* Hyp IPA Fault Address Register */
103 	u64 disr_el1;		/* Deferred [SError] Status Register */
104 };
105 
106 /*
107  * 0 is reserved as an invalid value.
108  * Order should be kept in sync with the save/restore code.
109  */
110 enum vcpu_sysreg {
111 	__INVALID_SYSREG__,
112 	MPIDR_EL1,	/* MultiProcessor Affinity Register */
113 	CSSELR_EL1,	/* Cache Size Selection Register */
114 	SCTLR_EL1,	/* System Control Register */
115 	ACTLR_EL1,	/* Auxiliary Control Register */
116 	CPACR_EL1,	/* Coprocessor Access Control */
117 	ZCR_EL1,	/* SVE Control */
118 	TTBR0_EL1,	/* Translation Table Base Register 0 */
119 	TTBR1_EL1,	/* Translation Table Base Register 1 */
120 	TCR_EL1,	/* Translation Control Register */
121 	ESR_EL1,	/* Exception Syndrome Register */
122 	AFSR0_EL1,	/* Auxiliary Fault Status Register 0 */
123 	AFSR1_EL1,	/* Auxiliary Fault Status Register 1 */
124 	FAR_EL1,	/* Fault Address Register */
125 	MAIR_EL1,	/* Memory Attribute Indirection Register */
126 	VBAR_EL1,	/* Vector Base Address Register */
127 	CONTEXTIDR_EL1,	/* Context ID Register */
128 	TPIDR_EL0,	/* Thread ID, User R/W */
129 	TPIDRRO_EL0,	/* Thread ID, User R/O */
130 	TPIDR_EL1,	/* Thread ID, Privileged */
131 	AMAIR_EL1,	/* Aux Memory Attribute Indirection Register */
132 	CNTKCTL_EL1,	/* Timer Control Register (EL1) */
133 	PAR_EL1,	/* Physical Address Register */
134 	MDSCR_EL1,	/* Monitor Debug System Control Register */
135 	MDCCINT_EL1,	/* Monitor Debug Comms Channel Interrupt Enable Reg */
136 	DISR_EL1,	/* Deferred Interrupt Status Register */
137 
138 	/* Performance Monitors Registers */
139 	PMCR_EL0,	/* Control Register */
140 	PMSELR_EL0,	/* Event Counter Selection Register */
141 	PMEVCNTR0_EL0,	/* Event Counter Register (0-30) */
142 	PMEVCNTR30_EL0 = PMEVCNTR0_EL0 + 30,
143 	PMCCNTR_EL0,	/* Cycle Counter Register */
144 	PMEVTYPER0_EL0,	/* Event Type Register (0-30) */
145 	PMEVTYPER30_EL0 = PMEVTYPER0_EL0 + 30,
146 	PMCCFILTR_EL0,	/* Cycle Count Filter Register */
147 	PMCNTENSET_EL0,	/* Count Enable Set Register */
148 	PMINTENSET_EL1,	/* Interrupt Enable Set Register */
149 	PMOVSSET_EL0,	/* Overflow Flag Status Set Register */
150 	PMSWINC_EL0,	/* Software Increment Register */
151 	PMUSERENR_EL0,	/* User Enable Register */
152 
153 	/* Pointer Authentication Registers in a strict increasing order. */
154 	APIAKEYLO_EL1,
155 	APIAKEYHI_EL1,
156 	APIBKEYLO_EL1,
157 	APIBKEYHI_EL1,
158 	APDAKEYLO_EL1,
159 	APDAKEYHI_EL1,
160 	APDBKEYLO_EL1,
161 	APDBKEYHI_EL1,
162 	APGAKEYLO_EL1,
163 	APGAKEYHI_EL1,
164 
165 	/* 32bit specific registers. Keep them at the end of the range */
166 	DACR32_EL2,	/* Domain Access Control Register */
167 	IFSR32_EL2,	/* Instruction Fault Status Register */
168 	FPEXC32_EL2,	/* Floating-Point Exception Control Register */
169 	DBGVCR32_EL2,	/* Debug Vector Catch Register */
170 
171 	NR_SYS_REGS	/* Nothing after this line! */
172 };
173 
174 /* 32bit mapping */
175 #define c0_MPIDR	(MPIDR_EL1 * 2)	/* MultiProcessor ID Register */
176 #define c0_CSSELR	(CSSELR_EL1 * 2)/* Cache Size Selection Register */
177 #define c1_SCTLR	(SCTLR_EL1 * 2)	/* System Control Register */
178 #define c1_ACTLR	(ACTLR_EL1 * 2)	/* Auxiliary Control Register */
179 #define c1_CPACR	(CPACR_EL1 * 2)	/* Coprocessor Access Control */
180 #define c2_TTBR0	(TTBR0_EL1 * 2)	/* Translation Table Base Register 0 */
181 #define c2_TTBR0_high	(c2_TTBR0 + 1)	/* TTBR0 top 32 bits */
182 #define c2_TTBR1	(TTBR1_EL1 * 2)	/* Translation Table Base Register 1 */
183 #define c2_TTBR1_high	(c2_TTBR1 + 1)	/* TTBR1 top 32 bits */
184 #define c2_TTBCR	(TCR_EL1 * 2)	/* Translation Table Base Control R. */
185 #define c3_DACR		(DACR32_EL2 * 2)/* Domain Access Control Register */
186 #define c5_DFSR		(ESR_EL1 * 2)	/* Data Fault Status Register */
187 #define c5_IFSR		(IFSR32_EL2 * 2)/* Instruction Fault Status Register */
188 #define c5_ADFSR	(AFSR0_EL1 * 2)	/* Auxiliary Data Fault Status R */
189 #define c5_AIFSR	(AFSR1_EL1 * 2)	/* Auxiliary Instr Fault Status R */
190 #define c6_DFAR		(FAR_EL1 * 2)	/* Data Fault Address Register */
191 #define c6_IFAR		(c6_DFAR + 1)	/* Instruction Fault Address Register */
192 #define c7_PAR		(PAR_EL1 * 2)	/* Physical Address Register */
193 #define c7_PAR_high	(c7_PAR + 1)	/* PAR top 32 bits */
194 #define c10_PRRR	(MAIR_EL1 * 2)	/* Primary Region Remap Register */
195 #define c10_NMRR	(c10_PRRR + 1)	/* Normal Memory Remap Register */
196 #define c12_VBAR	(VBAR_EL1 * 2)	/* Vector Base Address Register */
197 #define c13_CID		(CONTEXTIDR_EL1 * 2)	/* Context ID Register */
198 #define c13_TID_URW	(TPIDR_EL0 * 2)	/* Thread ID, User R/W */
199 #define c13_TID_URO	(TPIDRRO_EL0 * 2)/* Thread ID, User R/O */
200 #define c13_TID_PRIV	(TPIDR_EL1 * 2)	/* Thread ID, Privileged */
201 #define c10_AMAIR0	(AMAIR_EL1 * 2)	/* Aux Memory Attr Indirection Reg */
202 #define c10_AMAIR1	(c10_AMAIR0 + 1)/* Aux Memory Attr Indirection Reg */
203 #define c14_CNTKCTL	(CNTKCTL_EL1 * 2) /* Timer Control Register (PL1) */
204 
205 #define cp14_DBGDSCRext	(MDSCR_EL1 * 2)
206 #define cp14_DBGBCR0	(DBGBCR0_EL1 * 2)
207 #define cp14_DBGBVR0	(DBGBVR0_EL1 * 2)
208 #define cp14_DBGBXVR0	(cp14_DBGBVR0 + 1)
209 #define cp14_DBGWCR0	(DBGWCR0_EL1 * 2)
210 #define cp14_DBGWVR0	(DBGWVR0_EL1 * 2)
211 #define cp14_DBGDCCINT	(MDCCINT_EL1 * 2)
212 
213 #define NR_COPRO_REGS	(NR_SYS_REGS * 2)
214 
215 struct kvm_cpu_context {
216 	struct kvm_regs	gp_regs;
217 	union {
218 		u64 sys_regs[NR_SYS_REGS];
219 		u32 copro[NR_COPRO_REGS];
220 	};
221 
222 	struct kvm_vcpu *__hyp_running_vcpu;
223 };
224 
225 struct kvm_pmu_events {
226 	u32 events_host;
227 	u32 events_guest;
228 };
229 
230 struct kvm_host_data {
231 	struct kvm_cpu_context host_ctxt;
232 	struct kvm_pmu_events pmu_events;
233 };
234 
235 typedef struct kvm_host_data kvm_host_data_t;
236 
237 struct vcpu_reset_state {
238 	unsigned long	pc;
239 	unsigned long	r0;
240 	bool		be;
241 	bool		reset;
242 };
243 
244 struct kvm_vcpu_arch {
245 	struct kvm_cpu_context ctxt;
246 	void *sve_state;
247 	unsigned int sve_max_vl;
248 
249 	/* HYP configuration */
250 	u64 hcr_el2;
251 	u32 mdcr_el2;
252 
253 	/* Exception Information */
254 	struct kvm_vcpu_fault_info fault;
255 
256 	/* State of various workarounds, see kvm_asm.h for bit assignment */
257 	u64 workaround_flags;
258 
259 	/* Miscellaneous vcpu state flags */
260 	u64 flags;
261 
262 	/*
263 	 * We maintain more than a single set of debug registers to support
264 	 * debugging the guest from the host and to maintain separate host and
265 	 * guest state during world switches. vcpu_debug_state are the debug
266 	 * registers of the vcpu as the guest sees them.  host_debug_state are
267 	 * the host registers which are saved and restored during
268 	 * world switches. external_debug_state contains the debug
269 	 * values we want to debug the guest. This is set via the
270 	 * KVM_SET_GUEST_DEBUG ioctl.
271 	 *
272 	 * debug_ptr points to the set of debug registers that should be loaded
273 	 * onto the hardware when running the guest.
274 	 */
275 	struct kvm_guest_debug_arch *debug_ptr;
276 	struct kvm_guest_debug_arch vcpu_debug_state;
277 	struct kvm_guest_debug_arch external_debug_state;
278 
279 	/* Pointer to host CPU context */
280 	struct kvm_cpu_context *host_cpu_context;
281 
282 	struct thread_info *host_thread_info;	/* hyp VA */
283 	struct user_fpsimd_state *host_fpsimd_state;	/* hyp VA */
284 
285 	struct {
286 		/* {Break,watch}point registers */
287 		struct kvm_guest_debug_arch regs;
288 		/* Statistical profiling extension */
289 		u64 pmscr_el1;
290 	} host_debug_state;
291 
292 	/* VGIC state */
293 	struct vgic_cpu vgic_cpu;
294 	struct arch_timer_cpu timer_cpu;
295 	struct kvm_pmu pmu;
296 
297 	/*
298 	 * Anything that is not used directly from assembly code goes
299 	 * here.
300 	 */
301 
302 	/*
303 	 * Guest registers we preserve during guest debugging.
304 	 *
305 	 * These shadow registers are updated by the kvm_handle_sys_reg
306 	 * trap handler if the guest accesses or updates them while we
307 	 * are using guest debug.
308 	 */
309 	struct {
310 		u32	mdscr_el1;
311 	} guest_debug_preserved;
312 
313 	/* vcpu power-off state */
314 	bool power_off;
315 
316 	/* Don't run the guest (internal implementation need) */
317 	bool pause;
318 
319 	/* IO related fields */
320 	struct kvm_decode mmio_decode;
321 
322 	/* Cache some mmu pages needed inside spinlock regions */
323 	struct kvm_mmu_memory_cache mmu_page_cache;
324 
325 	/* Target CPU and feature flags */
326 	int target;
327 	DECLARE_BITMAP(features, KVM_VCPU_MAX_FEATURES);
328 
329 	/* Detect first run of a vcpu */
330 	bool has_run_once;
331 
332 	/* Virtual SError ESR to restore when HCR_EL2.VSE is set */
333 	u64 vsesr_el2;
334 
335 	/* Additional reset state */
336 	struct vcpu_reset_state	reset_state;
337 
338 	/* True when deferrable sysregs are loaded on the physical CPU,
339 	 * see kvm_vcpu_load_sysregs and kvm_vcpu_put_sysregs. */
340 	bool sysregs_loaded_on_cpu;
341 };
342 
343 /* Pointer to the vcpu's SVE FFR for sve_{save,load}_state() */
344 #define vcpu_sve_pffr(vcpu) ((void *)((char *)((vcpu)->arch.sve_state) + \
345 				      sve_ffr_offset((vcpu)->arch.sve_max_vl)))
346 
347 #define vcpu_sve_state_size(vcpu) ({					\
348 	size_t __size_ret;						\
349 	unsigned int __vcpu_vq;						\
350 									\
351 	if (WARN_ON(!sve_vl_valid((vcpu)->arch.sve_max_vl))) {		\
352 		__size_ret = 0;						\
353 	} else {							\
354 		__vcpu_vq = sve_vq_from_vl((vcpu)->arch.sve_max_vl);	\
355 		__size_ret = SVE_SIG_REGS_SIZE(__vcpu_vq);		\
356 	}								\
357 									\
358 	__size_ret;							\
359 })
360 
361 /* vcpu_arch flags field values: */
362 #define KVM_ARM64_DEBUG_DIRTY		(1 << 0)
363 #define KVM_ARM64_FP_ENABLED		(1 << 1) /* guest FP regs loaded */
364 #define KVM_ARM64_FP_HOST		(1 << 2) /* host FP regs loaded */
365 #define KVM_ARM64_HOST_SVE_IN_USE	(1 << 3) /* backup for host TIF_SVE */
366 #define KVM_ARM64_HOST_SVE_ENABLED	(1 << 4) /* SVE enabled for EL0 */
367 #define KVM_ARM64_GUEST_HAS_SVE		(1 << 5) /* SVE exposed to guest */
368 #define KVM_ARM64_VCPU_SVE_FINALIZED	(1 << 6) /* SVE config completed */
369 #define KVM_ARM64_GUEST_HAS_PTRAUTH	(1 << 7) /* PTRAUTH exposed to guest */
370 
371 #define vcpu_has_sve(vcpu) (system_supports_sve() && \
372 			    ((vcpu)->arch.flags & KVM_ARM64_GUEST_HAS_SVE))
373 
374 #define vcpu_has_ptrauth(vcpu)	((system_supports_address_auth() || \
375 				  system_supports_generic_auth()) && \
376 				 ((vcpu)->arch.flags & KVM_ARM64_GUEST_HAS_PTRAUTH))
377 
378 #define vcpu_gp_regs(v)		(&(v)->arch.ctxt.gp_regs)
379 
380 /*
381  * Only use __vcpu_sys_reg if you know you want the memory backed version of a
382  * register, and not the one most recently accessed by a running VCPU.  For
383  * example, for userspace access or for system registers that are never context
384  * switched, but only emulated.
385  */
386 #define __vcpu_sys_reg(v,r)	((v)->arch.ctxt.sys_regs[(r)])
387 
388 u64 vcpu_read_sys_reg(const struct kvm_vcpu *vcpu, int reg);
389 void vcpu_write_sys_reg(struct kvm_vcpu *vcpu, u64 val, int reg);
390 
391 /*
392  * CP14 and CP15 live in the same array, as they are backed by the
393  * same system registers.
394  */
395 #define vcpu_cp14(v,r)		((v)->arch.ctxt.copro[(r)])
396 #define vcpu_cp15(v,r)		((v)->arch.ctxt.copro[(r)])
397 
398 struct kvm_vm_stat {
399 	ulong remote_tlb_flush;
400 };
401 
402 struct kvm_vcpu_stat {
403 	u64 halt_successful_poll;
404 	u64 halt_attempted_poll;
405 	u64 halt_poll_invalid;
406 	u64 halt_wakeup;
407 	u64 hvc_exit_stat;
408 	u64 wfe_exit_stat;
409 	u64 wfi_exit_stat;
410 	u64 mmio_exit_user;
411 	u64 mmio_exit_kernel;
412 	u64 exits;
413 };
414 
415 int kvm_vcpu_preferred_target(struct kvm_vcpu_init *init);
416 unsigned long kvm_arm_num_regs(struct kvm_vcpu *vcpu);
417 int kvm_arm_copy_reg_indices(struct kvm_vcpu *vcpu, u64 __user *indices);
418 int kvm_arm_get_reg(struct kvm_vcpu *vcpu, const struct kvm_one_reg *reg);
419 int kvm_arm_set_reg(struct kvm_vcpu *vcpu, const struct kvm_one_reg *reg);
420 int __kvm_arm_vcpu_get_events(struct kvm_vcpu *vcpu,
421 			      struct kvm_vcpu_events *events);
422 
423 int __kvm_arm_vcpu_set_events(struct kvm_vcpu *vcpu,
424 			      struct kvm_vcpu_events *events);
425 
426 #define KVM_ARCH_WANT_MMU_NOTIFIER
427 int kvm_unmap_hva_range(struct kvm *kvm,
428 			unsigned long start, unsigned long end);
429 int kvm_set_spte_hva(struct kvm *kvm, unsigned long hva, pte_t pte);
430 int kvm_age_hva(struct kvm *kvm, unsigned long start, unsigned long end);
431 int kvm_test_age_hva(struct kvm *kvm, unsigned long hva);
432 
433 struct kvm_vcpu *kvm_arm_get_running_vcpu(void);
434 struct kvm_vcpu * __percpu *kvm_get_running_vcpus(void);
435 void kvm_arm_halt_guest(struct kvm *kvm);
436 void kvm_arm_resume_guest(struct kvm *kvm);
437 
438 u64 __kvm_call_hyp(void *hypfn, ...);
439 
440 /*
441  * The couple of isb() below are there to guarantee the same behaviour
442  * on VHE as on !VHE, where the eret to EL1 acts as a context
443  * synchronization event.
444  */
445 #define kvm_call_hyp(f, ...)						\
446 	do {								\
447 		if (has_vhe()) {					\
448 			f(__VA_ARGS__);					\
449 			isb();						\
450 		} else {						\
451 			__kvm_call_hyp(kvm_ksym_ref(f), ##__VA_ARGS__); \
452 		}							\
453 	} while(0)
454 
455 #define kvm_call_hyp_ret(f, ...)					\
456 	({								\
457 		typeof(f(__VA_ARGS__)) ret;				\
458 									\
459 		if (has_vhe()) {					\
460 			ret = f(__VA_ARGS__);				\
461 			isb();						\
462 		} else {						\
463 			ret = __kvm_call_hyp(kvm_ksym_ref(f),		\
464 					     ##__VA_ARGS__);		\
465 		}							\
466 									\
467 		ret;							\
468 	})
469 
470 void force_vm_exit(const cpumask_t *mask);
471 void kvm_mmu_wp_memory_region(struct kvm *kvm, int slot);
472 
473 int handle_exit(struct kvm_vcpu *vcpu, struct kvm_run *run,
474 		int exception_index);
475 void handle_exit_early(struct kvm_vcpu *vcpu, struct kvm_run *run,
476 		       int exception_index);
477 
478 int kvm_perf_init(void);
479 int kvm_perf_teardown(void);
480 
481 void kvm_set_sei_esr(struct kvm_vcpu *vcpu, u64 syndrome);
482 
483 struct kvm_vcpu *kvm_mpidr_to_vcpu(struct kvm *kvm, unsigned long mpidr);
484 
485 DECLARE_PER_CPU(kvm_host_data_t, kvm_host_data);
486 
487 static inline void kvm_init_host_cpu_context(struct kvm_cpu_context *cpu_ctxt)
488 {
489 	/* The host's MPIDR is immutable, so let's set it up at boot time */
490 	cpu_ctxt->sys_regs[MPIDR_EL1] = read_cpuid_mpidr();
491 }
492 
493 void __kvm_enable_ssbs(void);
494 
495 static inline void __cpu_init_hyp_mode(phys_addr_t pgd_ptr,
496 				       unsigned long hyp_stack_ptr,
497 				       unsigned long vector_ptr)
498 {
499 	/*
500 	 * Calculate the raw per-cpu offset without a translation from the
501 	 * kernel's mapping to the linear mapping, and store it in tpidr_el2
502 	 * so that we can use adr_l to access per-cpu variables in EL2.
503 	 */
504 	u64 tpidr_el2 = ((u64)this_cpu_ptr(&kvm_host_data) -
505 			 (u64)kvm_ksym_ref(kvm_host_data));
506 
507 	/*
508 	 * Call initialization code, and switch to the full blown HYP code.
509 	 * If the cpucaps haven't been finalized yet, something has gone very
510 	 * wrong, and hyp will crash and burn when it uses any
511 	 * cpus_have_const_cap() wrapper.
512 	 */
513 	BUG_ON(!static_branch_likely(&arm64_const_caps_ready));
514 	__kvm_call_hyp((void *)pgd_ptr, hyp_stack_ptr, vector_ptr, tpidr_el2);
515 
516 	/*
517 	 * Disabling SSBD on a non-VHE system requires us to enable SSBS
518 	 * at EL2.
519 	 */
520 	if (!has_vhe() && this_cpu_has_cap(ARM64_SSBS) &&
521 	    arm64_get_ssbd_state() == ARM64_SSBD_FORCE_DISABLE) {
522 		kvm_call_hyp(__kvm_enable_ssbs);
523 	}
524 }
525 
526 static inline bool kvm_arch_requires_vhe(void)
527 {
528 	/*
529 	 * The Arm architecture specifies that implementation of SVE
530 	 * requires VHE also to be implemented.  The KVM code for arm64
531 	 * relies on this when SVE is present:
532 	 */
533 	if (system_supports_sve())
534 		return true;
535 
536 	/* Some implementations have defects that confine them to VHE */
537 	if (cpus_have_cap(ARM64_WORKAROUND_1165522))
538 		return true;
539 
540 	return false;
541 }
542 
543 void kvm_arm_vcpu_ptrauth_trap(struct kvm_vcpu *vcpu);
544 
545 static inline void kvm_arch_hardware_unsetup(void) {}
546 static inline void kvm_arch_sync_events(struct kvm *kvm) {}
547 static inline void kvm_arch_sched_in(struct kvm_vcpu *vcpu, int cpu) {}
548 static inline void kvm_arch_vcpu_block_finish(struct kvm_vcpu *vcpu) {}
549 
550 void kvm_arm_init_debug(void);
551 void kvm_arm_setup_debug(struct kvm_vcpu *vcpu);
552 void kvm_arm_clear_debug(struct kvm_vcpu *vcpu);
553 void kvm_arm_reset_debug_ptr(struct kvm_vcpu *vcpu);
554 int kvm_arm_vcpu_arch_set_attr(struct kvm_vcpu *vcpu,
555 			       struct kvm_device_attr *attr);
556 int kvm_arm_vcpu_arch_get_attr(struct kvm_vcpu *vcpu,
557 			       struct kvm_device_attr *attr);
558 int kvm_arm_vcpu_arch_has_attr(struct kvm_vcpu *vcpu,
559 			       struct kvm_device_attr *attr);
560 
561 static inline void __cpu_init_stage2(void) {}
562 
563 /* Guest/host FPSIMD coordination helpers */
564 int kvm_arch_vcpu_run_map_fp(struct kvm_vcpu *vcpu);
565 void kvm_arch_vcpu_load_fp(struct kvm_vcpu *vcpu);
566 void kvm_arch_vcpu_ctxsync_fp(struct kvm_vcpu *vcpu);
567 void kvm_arch_vcpu_put_fp(struct kvm_vcpu *vcpu);
568 
569 static inline bool kvm_pmu_counter_deferred(struct perf_event_attr *attr)
570 {
571 	return (!has_vhe() && attr->exclude_host);
572 }
573 
574 #ifdef CONFIG_KVM /* Avoid conflicts with core headers if CONFIG_KVM=n */
575 static inline int kvm_arch_vcpu_run_pid_change(struct kvm_vcpu *vcpu)
576 {
577 	return kvm_arch_vcpu_run_map_fp(vcpu);
578 }
579 
580 void kvm_set_pmu_events(u32 set, struct perf_event_attr *attr);
581 void kvm_clr_pmu_events(u32 clr);
582 
583 void kvm_vcpu_pmu_restore_guest(struct kvm_vcpu *vcpu);
584 void kvm_vcpu_pmu_restore_host(struct kvm_vcpu *vcpu);
585 #else
586 static inline void kvm_set_pmu_events(u32 set, struct perf_event_attr *attr) {}
587 static inline void kvm_clr_pmu_events(u32 clr) {}
588 #endif
589 
590 static inline void kvm_arm_vhe_guest_enter(void)
591 {
592 	local_daif_mask();
593 
594 	/*
595 	 * Having IRQs masked via PMR when entering the guest means the GIC
596 	 * will not signal the CPU of interrupts of lower priority, and the
597 	 * only way to get out will be via guest exceptions.
598 	 * Naturally, we want to avoid this.
599 	 *
600 	 * local_daif_mask() already sets GIC_PRIO_PSR_I_SET, we just need a
601 	 * dsb to ensure the redistributor is forwards EL2 IRQs to the CPU.
602 	 */
603 	if (system_uses_irq_prio_masking())
604 		dsb(sy);
605 }
606 
607 static inline void kvm_arm_vhe_guest_exit(void)
608 {
609 	/*
610 	 * local_daif_restore() takes care to properly restore PSTATE.DAIF
611 	 * and the GIC PMR if the host is using IRQ priorities.
612 	 */
613 	local_daif_restore(DAIF_PROCCTX_NOIRQ);
614 
615 	/*
616 	 * When we exit from the guest we change a number of CPU configuration
617 	 * parameters, such as traps.  Make sure these changes take effect
618 	 * before running the host or additional guests.
619 	 */
620 	isb();
621 }
622 
623 #define KVM_BP_HARDEN_UNKNOWN		-1
624 #define KVM_BP_HARDEN_WA_NEEDED		0
625 #define KVM_BP_HARDEN_NOT_REQUIRED	1
626 
627 static inline int kvm_arm_harden_branch_predictor(void)
628 {
629 	switch (get_spectre_v2_workaround_state()) {
630 	case ARM64_BP_HARDEN_WA_NEEDED:
631 		return KVM_BP_HARDEN_WA_NEEDED;
632 	case ARM64_BP_HARDEN_NOT_REQUIRED:
633 		return KVM_BP_HARDEN_NOT_REQUIRED;
634 	case ARM64_BP_HARDEN_UNKNOWN:
635 	default:
636 		return KVM_BP_HARDEN_UNKNOWN;
637 	}
638 }
639 
640 #define KVM_SSBD_UNKNOWN		-1
641 #define KVM_SSBD_FORCE_DISABLE		0
642 #define KVM_SSBD_KERNEL		1
643 #define KVM_SSBD_FORCE_ENABLE		2
644 #define KVM_SSBD_MITIGATED		3
645 
646 static inline int kvm_arm_have_ssbd(void)
647 {
648 	switch (arm64_get_ssbd_state()) {
649 	case ARM64_SSBD_FORCE_DISABLE:
650 		return KVM_SSBD_FORCE_DISABLE;
651 	case ARM64_SSBD_KERNEL:
652 		return KVM_SSBD_KERNEL;
653 	case ARM64_SSBD_FORCE_ENABLE:
654 		return KVM_SSBD_FORCE_ENABLE;
655 	case ARM64_SSBD_MITIGATED:
656 		return KVM_SSBD_MITIGATED;
657 	case ARM64_SSBD_UNKNOWN:
658 	default:
659 		return KVM_SSBD_UNKNOWN;
660 	}
661 }
662 
663 void kvm_vcpu_load_sysregs(struct kvm_vcpu *vcpu);
664 void kvm_vcpu_put_sysregs(struct kvm_vcpu *vcpu);
665 
666 void kvm_set_ipa_limit(void);
667 
668 #define __KVM_HAVE_ARCH_VM_ALLOC
669 struct kvm *kvm_arch_alloc_vm(void);
670 void kvm_arch_free_vm(struct kvm *kvm);
671 
672 int kvm_arm_setup_stage2(struct kvm *kvm, unsigned long type);
673 
674 int kvm_arm_vcpu_finalize(struct kvm_vcpu *vcpu, int feature);
675 bool kvm_arm_vcpu_is_finalized(struct kvm_vcpu *vcpu);
676 
677 #define kvm_arm_vcpu_sve_finalized(vcpu) \
678 	((vcpu)->arch.flags & KVM_ARM64_VCPU_SVE_FINALIZED)
679 
680 #endif /* __ARM64_KVM_HOST_H__ */
681