xref: /linux/arch/arm64/kvm/arm.c (revision 572af9f284669d31d9175122bbef9bc62cea8ded)
1 // SPDX-License-Identifier: GPL-2.0-only
2 /*
3  * Copyright (C) 2012 - Virtual Open Systems and Columbia University
4  * Author: Christoffer Dall <c.dall@virtualopensystems.com>
5  */
6 
7 #include <linux/bug.h>
8 #include <linux/cpu_pm.h>
9 #include <linux/entry-kvm.h>
10 #include <linux/errno.h>
11 #include <linux/err.h>
12 #include <linux/kvm_host.h>
13 #include <linux/list.h>
14 #include <linux/module.h>
15 #include <linux/vmalloc.h>
16 #include <linux/fs.h>
17 #include <linux/mman.h>
18 #include <linux/sched.h>
19 #include <linux/kvm.h>
20 #include <linux/kvm_irqfd.h>
21 #include <linux/irqbypass.h>
22 #include <linux/sched/stat.h>
23 #include <linux/psci.h>
24 #include <trace/events/kvm.h>
25 
26 #define CREATE_TRACE_POINTS
27 #include "trace_arm.h"
28 
29 #include <linux/uaccess.h>
30 #include <asm/ptrace.h>
31 #include <asm/mman.h>
32 #include <asm/tlbflush.h>
33 #include <asm/cacheflush.h>
34 #include <asm/cpufeature.h>
35 #include <asm/virt.h>
36 #include <asm/kvm_arm.h>
37 #include <asm/kvm_asm.h>
38 #include <asm/kvm_emulate.h>
39 #include <asm/kvm_mmu.h>
40 #include <asm/kvm_nested.h>
41 #include <asm/kvm_pkvm.h>
42 #include <asm/kvm_ptrauth.h>
43 #include <asm/sections.h>
44 
45 #include <kvm/arm_hypercalls.h>
46 #include <kvm/arm_pmu.h>
47 #include <kvm/arm_psci.h>
48 
49 #include "sys_regs.h"
50 
51 static enum kvm_mode kvm_mode = KVM_MODE_DEFAULT;
52 
53 enum kvm_wfx_trap_policy {
54 	KVM_WFX_NOTRAP_SINGLE_TASK, /* Default option */
55 	KVM_WFX_NOTRAP,
56 	KVM_WFX_TRAP,
57 };
58 
59 static enum kvm_wfx_trap_policy kvm_wfi_trap_policy __read_mostly = KVM_WFX_NOTRAP_SINGLE_TASK;
60 static enum kvm_wfx_trap_policy kvm_wfe_trap_policy __read_mostly = KVM_WFX_NOTRAP_SINGLE_TASK;
61 
62 DECLARE_KVM_HYP_PER_CPU(unsigned long, kvm_hyp_vector);
63 
64 DEFINE_PER_CPU(unsigned long, kvm_arm_hyp_stack_page);
65 DECLARE_KVM_NVHE_PER_CPU(struct kvm_nvhe_init_params, kvm_init_params);
66 
67 DECLARE_KVM_NVHE_PER_CPU(struct kvm_cpu_context, kvm_hyp_ctxt);
68 
69 static bool vgic_present, kvm_arm_initialised;
70 
71 static DEFINE_PER_CPU(unsigned char, kvm_hyp_initialized);
72 
73 bool is_kvm_arm_initialised(void)
74 {
75 	return kvm_arm_initialised;
76 }
77 
78 int kvm_arch_vcpu_should_kick(struct kvm_vcpu *vcpu)
79 {
80 	return kvm_vcpu_exiting_guest_mode(vcpu) == IN_GUEST_MODE;
81 }
82 
83 /*
84  * This functions as an allow-list of protected VM capabilities.
85  * Features not explicitly allowed by this function are denied.
86  */
87 static bool pkvm_ext_allowed(struct kvm *kvm, long ext)
88 {
89 	switch (ext) {
90 	case KVM_CAP_IRQCHIP:
91 	case KVM_CAP_ARM_PSCI:
92 	case KVM_CAP_ARM_PSCI_0_2:
93 	case KVM_CAP_NR_VCPUS:
94 	case KVM_CAP_MAX_VCPUS:
95 	case KVM_CAP_MAX_VCPU_ID:
96 	case KVM_CAP_MSI_DEVID:
97 	case KVM_CAP_ARM_VM_IPA_SIZE:
98 	case KVM_CAP_ARM_PMU_V3:
99 	case KVM_CAP_ARM_SVE:
100 	case KVM_CAP_ARM_PTRAUTH_ADDRESS:
101 	case KVM_CAP_ARM_PTRAUTH_GENERIC:
102 		return true;
103 	default:
104 		return false;
105 	}
106 }
107 
108 int kvm_vm_ioctl_enable_cap(struct kvm *kvm,
109 			    struct kvm_enable_cap *cap)
110 {
111 	int r = -EINVAL;
112 
113 	if (cap->flags)
114 		return -EINVAL;
115 
116 	if (kvm_vm_is_protected(kvm) && !pkvm_ext_allowed(kvm, cap->cap))
117 		return -EINVAL;
118 
119 	switch (cap->cap) {
120 	case KVM_CAP_ARM_NISV_TO_USER:
121 		r = 0;
122 		set_bit(KVM_ARCH_FLAG_RETURN_NISV_IO_ABORT_TO_USER,
123 			&kvm->arch.flags);
124 		break;
125 	case KVM_CAP_ARM_MTE:
126 		mutex_lock(&kvm->lock);
127 		if (system_supports_mte() && !kvm->created_vcpus) {
128 			r = 0;
129 			set_bit(KVM_ARCH_FLAG_MTE_ENABLED, &kvm->arch.flags);
130 		}
131 		mutex_unlock(&kvm->lock);
132 		break;
133 	case KVM_CAP_ARM_SYSTEM_SUSPEND:
134 		r = 0;
135 		set_bit(KVM_ARCH_FLAG_SYSTEM_SUSPEND_ENABLED, &kvm->arch.flags);
136 		break;
137 	case KVM_CAP_ARM_EAGER_SPLIT_CHUNK_SIZE:
138 		mutex_lock(&kvm->slots_lock);
139 		/*
140 		 * To keep things simple, allow changing the chunk
141 		 * size only when no memory slots have been created.
142 		 */
143 		if (kvm_are_all_memslots_empty(kvm)) {
144 			u64 new_cap = cap->args[0];
145 
146 			if (!new_cap || kvm_is_block_size_supported(new_cap)) {
147 				r = 0;
148 				kvm->arch.mmu.split_page_chunk_size = new_cap;
149 			}
150 		}
151 		mutex_unlock(&kvm->slots_lock);
152 		break;
153 	default:
154 		break;
155 	}
156 
157 	return r;
158 }
159 
160 static int kvm_arm_default_max_vcpus(void)
161 {
162 	return vgic_present ? kvm_vgic_get_max_vcpus() : KVM_MAX_VCPUS;
163 }
164 
165 /**
166  * kvm_arch_init_vm - initializes a VM data structure
167  * @kvm:	pointer to the KVM struct
168  * @type:	kvm device type
169  */
170 int kvm_arch_init_vm(struct kvm *kvm, unsigned long type)
171 {
172 	int ret;
173 
174 	mutex_init(&kvm->arch.config_lock);
175 
176 #ifdef CONFIG_LOCKDEP
177 	/* Clue in lockdep that the config_lock must be taken inside kvm->lock */
178 	mutex_lock(&kvm->lock);
179 	mutex_lock(&kvm->arch.config_lock);
180 	mutex_unlock(&kvm->arch.config_lock);
181 	mutex_unlock(&kvm->lock);
182 #endif
183 
184 	kvm_init_nested(kvm);
185 
186 	ret = kvm_share_hyp(kvm, kvm + 1);
187 	if (ret)
188 		return ret;
189 
190 	ret = pkvm_init_host_vm(kvm);
191 	if (ret)
192 		goto err_unshare_kvm;
193 
194 	if (!zalloc_cpumask_var(&kvm->arch.supported_cpus, GFP_KERNEL_ACCOUNT)) {
195 		ret = -ENOMEM;
196 		goto err_unshare_kvm;
197 	}
198 	cpumask_copy(kvm->arch.supported_cpus, cpu_possible_mask);
199 
200 	ret = kvm_init_stage2_mmu(kvm, &kvm->arch.mmu, type);
201 	if (ret)
202 		goto err_free_cpumask;
203 
204 	kvm_vgic_early_init(kvm);
205 
206 	kvm_timer_init_vm(kvm);
207 
208 	/* The maximum number of VCPUs is limited by the host's GIC model */
209 	kvm->max_vcpus = kvm_arm_default_max_vcpus();
210 
211 	kvm_arm_init_hypercalls(kvm);
212 
213 	bitmap_zero(kvm->arch.vcpu_features, KVM_VCPU_MAX_FEATURES);
214 
215 	return 0;
216 
217 err_free_cpumask:
218 	free_cpumask_var(kvm->arch.supported_cpus);
219 err_unshare_kvm:
220 	kvm_unshare_hyp(kvm, kvm + 1);
221 	return ret;
222 }
223 
224 vm_fault_t kvm_arch_vcpu_fault(struct kvm_vcpu *vcpu, struct vm_fault *vmf)
225 {
226 	return VM_FAULT_SIGBUS;
227 }
228 
229 void kvm_arch_create_vm_debugfs(struct kvm *kvm)
230 {
231 	kvm_sys_regs_create_debugfs(kvm);
232 	kvm_s2_ptdump_create_debugfs(kvm);
233 }
234 
235 static void kvm_destroy_mpidr_data(struct kvm *kvm)
236 {
237 	struct kvm_mpidr_data *data;
238 
239 	mutex_lock(&kvm->arch.config_lock);
240 
241 	data = rcu_dereference_protected(kvm->arch.mpidr_data,
242 					 lockdep_is_held(&kvm->arch.config_lock));
243 	if (data) {
244 		rcu_assign_pointer(kvm->arch.mpidr_data, NULL);
245 		synchronize_rcu();
246 		kfree(data);
247 	}
248 
249 	mutex_unlock(&kvm->arch.config_lock);
250 }
251 
252 /**
253  * kvm_arch_destroy_vm - destroy the VM data structure
254  * @kvm:	pointer to the KVM struct
255  */
256 void kvm_arch_destroy_vm(struct kvm *kvm)
257 {
258 	bitmap_free(kvm->arch.pmu_filter);
259 	free_cpumask_var(kvm->arch.supported_cpus);
260 
261 	kvm_vgic_destroy(kvm);
262 
263 	if (is_protected_kvm_enabled())
264 		pkvm_destroy_hyp_vm(kvm);
265 
266 	kvm_destroy_mpidr_data(kvm);
267 
268 	kfree(kvm->arch.sysreg_masks);
269 	kvm_destroy_vcpus(kvm);
270 
271 	kvm_unshare_hyp(kvm, kvm + 1);
272 
273 	kvm_arm_teardown_hypercalls(kvm);
274 }
275 
276 static bool kvm_has_full_ptr_auth(void)
277 {
278 	bool apa, gpa, api, gpi, apa3, gpa3;
279 	u64 isar1, isar2, val;
280 
281 	/*
282 	 * Check that:
283 	 *
284 	 * - both Address and Generic auth are implemented for a given
285          *   algorithm (Q5, IMPDEF or Q3)
286 	 * - only a single algorithm is implemented.
287 	 */
288 	if (!system_has_full_ptr_auth())
289 		return false;
290 
291 	isar1 = read_sanitised_ftr_reg(SYS_ID_AA64ISAR1_EL1);
292 	isar2 = read_sanitised_ftr_reg(SYS_ID_AA64ISAR2_EL1);
293 
294 	apa = !!FIELD_GET(ID_AA64ISAR1_EL1_APA_MASK, isar1);
295 	val = FIELD_GET(ID_AA64ISAR1_EL1_GPA_MASK, isar1);
296 	gpa = (val == ID_AA64ISAR1_EL1_GPA_IMP);
297 
298 	api = !!FIELD_GET(ID_AA64ISAR1_EL1_API_MASK, isar1);
299 	val = FIELD_GET(ID_AA64ISAR1_EL1_GPI_MASK, isar1);
300 	gpi = (val == ID_AA64ISAR1_EL1_GPI_IMP);
301 
302 	apa3 = !!FIELD_GET(ID_AA64ISAR2_EL1_APA3_MASK, isar2);
303 	val  = FIELD_GET(ID_AA64ISAR2_EL1_GPA3_MASK, isar2);
304 	gpa3 = (val == ID_AA64ISAR2_EL1_GPA3_IMP);
305 
306 	return (apa == gpa && api == gpi && apa3 == gpa3 &&
307 		(apa + api + apa3) == 1);
308 }
309 
310 int kvm_vm_ioctl_check_extension(struct kvm *kvm, long ext)
311 {
312 	int r;
313 
314 	if (kvm && kvm_vm_is_protected(kvm) && !pkvm_ext_allowed(kvm, ext))
315 		return 0;
316 
317 	switch (ext) {
318 	case KVM_CAP_IRQCHIP:
319 		r = vgic_present;
320 		break;
321 	case KVM_CAP_IOEVENTFD:
322 	case KVM_CAP_USER_MEMORY:
323 	case KVM_CAP_SYNC_MMU:
324 	case KVM_CAP_DESTROY_MEMORY_REGION_WORKS:
325 	case KVM_CAP_ONE_REG:
326 	case KVM_CAP_ARM_PSCI:
327 	case KVM_CAP_ARM_PSCI_0_2:
328 	case KVM_CAP_READONLY_MEM:
329 	case KVM_CAP_MP_STATE:
330 	case KVM_CAP_IMMEDIATE_EXIT:
331 	case KVM_CAP_VCPU_EVENTS:
332 	case KVM_CAP_ARM_IRQ_LINE_LAYOUT_2:
333 	case KVM_CAP_ARM_NISV_TO_USER:
334 	case KVM_CAP_ARM_INJECT_EXT_DABT:
335 	case KVM_CAP_SET_GUEST_DEBUG:
336 	case KVM_CAP_VCPU_ATTRIBUTES:
337 	case KVM_CAP_PTP_KVM:
338 	case KVM_CAP_ARM_SYSTEM_SUSPEND:
339 	case KVM_CAP_IRQFD_RESAMPLE:
340 	case KVM_CAP_COUNTER_OFFSET:
341 		r = 1;
342 		break;
343 	case KVM_CAP_SET_GUEST_DEBUG2:
344 		return KVM_GUESTDBG_VALID_MASK;
345 	case KVM_CAP_ARM_SET_DEVICE_ADDR:
346 		r = 1;
347 		break;
348 	case KVM_CAP_NR_VCPUS:
349 		/*
350 		 * ARM64 treats KVM_CAP_NR_CPUS differently from all other
351 		 * architectures, as it does not always bound it to
352 		 * KVM_CAP_MAX_VCPUS. It should not matter much because
353 		 * this is just an advisory value.
354 		 */
355 		r = min_t(unsigned int, num_online_cpus(),
356 			  kvm_arm_default_max_vcpus());
357 		break;
358 	case KVM_CAP_MAX_VCPUS:
359 	case KVM_CAP_MAX_VCPU_ID:
360 		if (kvm)
361 			r = kvm->max_vcpus;
362 		else
363 			r = kvm_arm_default_max_vcpus();
364 		break;
365 	case KVM_CAP_MSI_DEVID:
366 		if (!kvm)
367 			r = -EINVAL;
368 		else
369 			r = kvm->arch.vgic.msis_require_devid;
370 		break;
371 	case KVM_CAP_ARM_USER_IRQ:
372 		/*
373 		 * 1: EL1_VTIMER, EL1_PTIMER, and PMU.
374 		 * (bump this number if adding more devices)
375 		 */
376 		r = 1;
377 		break;
378 	case KVM_CAP_ARM_MTE:
379 		r = system_supports_mte();
380 		break;
381 	case KVM_CAP_STEAL_TIME:
382 		r = kvm_arm_pvtime_supported();
383 		break;
384 	case KVM_CAP_ARM_EL1_32BIT:
385 		r = cpus_have_final_cap(ARM64_HAS_32BIT_EL1);
386 		break;
387 	case KVM_CAP_GUEST_DEBUG_HW_BPS:
388 		r = get_num_brps();
389 		break;
390 	case KVM_CAP_GUEST_DEBUG_HW_WPS:
391 		r = get_num_wrps();
392 		break;
393 	case KVM_CAP_ARM_PMU_V3:
394 		r = kvm_arm_support_pmu_v3();
395 		break;
396 	case KVM_CAP_ARM_INJECT_SERROR_ESR:
397 		r = cpus_have_final_cap(ARM64_HAS_RAS_EXTN);
398 		break;
399 	case KVM_CAP_ARM_VM_IPA_SIZE:
400 		r = get_kvm_ipa_limit();
401 		break;
402 	case KVM_CAP_ARM_SVE:
403 		r = system_supports_sve();
404 		break;
405 	case KVM_CAP_ARM_PTRAUTH_ADDRESS:
406 	case KVM_CAP_ARM_PTRAUTH_GENERIC:
407 		r = kvm_has_full_ptr_auth();
408 		break;
409 	case KVM_CAP_ARM_EAGER_SPLIT_CHUNK_SIZE:
410 		if (kvm)
411 			r = kvm->arch.mmu.split_page_chunk_size;
412 		else
413 			r = KVM_ARM_EAGER_SPLIT_CHUNK_SIZE_DEFAULT;
414 		break;
415 	case KVM_CAP_ARM_SUPPORTED_BLOCK_SIZES:
416 		r = kvm_supported_block_sizes();
417 		break;
418 	case KVM_CAP_ARM_SUPPORTED_REG_MASK_RANGES:
419 		r = BIT(0);
420 		break;
421 	default:
422 		r = 0;
423 	}
424 
425 	return r;
426 }
427 
428 long kvm_arch_dev_ioctl(struct file *filp,
429 			unsigned int ioctl, unsigned long arg)
430 {
431 	return -EINVAL;
432 }
433 
434 struct kvm *kvm_arch_alloc_vm(void)
435 {
436 	size_t sz = sizeof(struct kvm);
437 
438 	if (!has_vhe())
439 		return kzalloc(sz, GFP_KERNEL_ACCOUNT);
440 
441 	return __vmalloc(sz, GFP_KERNEL_ACCOUNT | __GFP_HIGHMEM | __GFP_ZERO);
442 }
443 
444 int kvm_arch_vcpu_precreate(struct kvm *kvm, unsigned int id)
445 {
446 	if (irqchip_in_kernel(kvm) && vgic_initialized(kvm))
447 		return -EBUSY;
448 
449 	if (id >= kvm->max_vcpus)
450 		return -EINVAL;
451 
452 	return 0;
453 }
454 
455 int kvm_arch_vcpu_create(struct kvm_vcpu *vcpu)
456 {
457 	int err;
458 
459 	spin_lock_init(&vcpu->arch.mp_state_lock);
460 
461 #ifdef CONFIG_LOCKDEP
462 	/* Inform lockdep that the config_lock is acquired after vcpu->mutex */
463 	mutex_lock(&vcpu->mutex);
464 	mutex_lock(&vcpu->kvm->arch.config_lock);
465 	mutex_unlock(&vcpu->kvm->arch.config_lock);
466 	mutex_unlock(&vcpu->mutex);
467 #endif
468 
469 	/* Force users to call KVM_ARM_VCPU_INIT */
470 	vcpu_clear_flag(vcpu, VCPU_INITIALIZED);
471 
472 	vcpu->arch.mmu_page_cache.gfp_zero = __GFP_ZERO;
473 
474 	/* Set up the timer */
475 	kvm_timer_vcpu_init(vcpu);
476 
477 	kvm_pmu_vcpu_init(vcpu);
478 
479 	kvm_arm_reset_debug_ptr(vcpu);
480 
481 	kvm_arm_pvtime_vcpu_init(&vcpu->arch);
482 
483 	vcpu->arch.hw_mmu = &vcpu->kvm->arch.mmu;
484 
485 	/*
486 	 * This vCPU may have been created after mpidr_data was initialized.
487 	 * Throw out the pre-computed mappings if that is the case which forces
488 	 * KVM to fall back to iteratively searching the vCPUs.
489 	 */
490 	kvm_destroy_mpidr_data(vcpu->kvm);
491 
492 	err = kvm_vgic_vcpu_init(vcpu);
493 	if (err)
494 		return err;
495 
496 	return kvm_share_hyp(vcpu, vcpu + 1);
497 }
498 
499 void kvm_arch_vcpu_postcreate(struct kvm_vcpu *vcpu)
500 {
501 }
502 
503 void kvm_arch_vcpu_destroy(struct kvm_vcpu *vcpu)
504 {
505 	kvm_mmu_free_memory_cache(&vcpu->arch.mmu_page_cache);
506 	kvm_timer_vcpu_terminate(vcpu);
507 	kvm_pmu_vcpu_destroy(vcpu);
508 	kvm_vgic_vcpu_destroy(vcpu);
509 	kvm_arm_vcpu_destroy(vcpu);
510 }
511 
512 void kvm_arch_vcpu_blocking(struct kvm_vcpu *vcpu)
513 {
514 
515 }
516 
517 void kvm_arch_vcpu_unblocking(struct kvm_vcpu *vcpu)
518 {
519 
520 }
521 
522 static void vcpu_set_pauth_traps(struct kvm_vcpu *vcpu)
523 {
524 	if (vcpu_has_ptrauth(vcpu) && !is_protected_kvm_enabled()) {
525 		/*
526 		 * Either we're running an L2 guest, and the API/APK bits come
527 		 * from L1's HCR_EL2, or API/APK are both set.
528 		 */
529 		if (unlikely(vcpu_has_nv(vcpu) && !is_hyp_ctxt(vcpu))) {
530 			u64 val;
531 
532 			val = __vcpu_sys_reg(vcpu, HCR_EL2);
533 			val &= (HCR_API | HCR_APK);
534 			vcpu->arch.hcr_el2 &= ~(HCR_API | HCR_APK);
535 			vcpu->arch.hcr_el2 |= val;
536 		} else {
537 			vcpu->arch.hcr_el2 |= (HCR_API | HCR_APK);
538 		}
539 
540 		/*
541 		 * Save the host keys if there is any chance for the guest
542 		 * to use pauth, as the entry code will reload the guest
543 		 * keys in that case.
544 		 */
545 		if (vcpu->arch.hcr_el2 & (HCR_API | HCR_APK)) {
546 			struct kvm_cpu_context *ctxt;
547 
548 			ctxt = this_cpu_ptr_hyp_sym(kvm_hyp_ctxt);
549 			ptrauth_save_keys(ctxt);
550 		}
551 	}
552 }
553 
554 static bool kvm_vcpu_should_clear_twi(struct kvm_vcpu *vcpu)
555 {
556 	if (unlikely(kvm_wfi_trap_policy != KVM_WFX_NOTRAP_SINGLE_TASK))
557 		return kvm_wfi_trap_policy == KVM_WFX_NOTRAP;
558 
559 	return single_task_running() &&
560 	       (atomic_read(&vcpu->arch.vgic_cpu.vgic_v3.its_vpe.vlpi_count) ||
561 		vcpu->kvm->arch.vgic.nassgireq);
562 }
563 
564 static bool kvm_vcpu_should_clear_twe(struct kvm_vcpu *vcpu)
565 {
566 	if (unlikely(kvm_wfe_trap_policy != KVM_WFX_NOTRAP_SINGLE_TASK))
567 		return kvm_wfe_trap_policy == KVM_WFX_NOTRAP;
568 
569 	return single_task_running();
570 }
571 
572 void kvm_arch_vcpu_load(struct kvm_vcpu *vcpu, int cpu)
573 {
574 	struct kvm_s2_mmu *mmu;
575 	int *last_ran;
576 
577 	if (vcpu_has_nv(vcpu))
578 		kvm_vcpu_load_hw_mmu(vcpu);
579 
580 	mmu = vcpu->arch.hw_mmu;
581 	last_ran = this_cpu_ptr(mmu->last_vcpu_ran);
582 
583 	/*
584 	 * We guarantee that both TLBs and I-cache are private to each
585 	 * vcpu. If detecting that a vcpu from the same VM has
586 	 * previously run on the same physical CPU, call into the
587 	 * hypervisor code to nuke the relevant contexts.
588 	 *
589 	 * We might get preempted before the vCPU actually runs, but
590 	 * over-invalidation doesn't affect correctness.
591 	 */
592 	if (*last_ran != vcpu->vcpu_idx) {
593 		kvm_call_hyp(__kvm_flush_cpu_context, mmu);
594 		*last_ran = vcpu->vcpu_idx;
595 	}
596 
597 	vcpu->cpu = cpu;
598 
599 	kvm_vgic_load(vcpu);
600 	kvm_timer_vcpu_load(vcpu);
601 	if (has_vhe())
602 		kvm_vcpu_load_vhe(vcpu);
603 	kvm_arch_vcpu_load_fp(vcpu);
604 	kvm_vcpu_pmu_restore_guest(vcpu);
605 	if (kvm_arm_is_pvtime_enabled(&vcpu->arch))
606 		kvm_make_request(KVM_REQ_RECORD_STEAL, vcpu);
607 
608 	if (kvm_vcpu_should_clear_twe(vcpu))
609 		vcpu->arch.hcr_el2 &= ~HCR_TWE;
610 	else
611 		vcpu->arch.hcr_el2 |= HCR_TWE;
612 
613 	if (kvm_vcpu_should_clear_twi(vcpu))
614 		vcpu->arch.hcr_el2 &= ~HCR_TWI;
615 	else
616 		vcpu->arch.hcr_el2 |= HCR_TWI;
617 
618 	vcpu_set_pauth_traps(vcpu);
619 
620 	kvm_arch_vcpu_load_debug_state_flags(vcpu);
621 
622 	if (!cpumask_test_cpu(cpu, vcpu->kvm->arch.supported_cpus))
623 		vcpu_set_on_unsupported_cpu(vcpu);
624 }
625 
626 void kvm_arch_vcpu_put(struct kvm_vcpu *vcpu)
627 {
628 	kvm_arch_vcpu_put_debug_state_flags(vcpu);
629 	kvm_arch_vcpu_put_fp(vcpu);
630 	if (has_vhe())
631 		kvm_vcpu_put_vhe(vcpu);
632 	kvm_timer_vcpu_put(vcpu);
633 	kvm_vgic_put(vcpu);
634 	kvm_vcpu_pmu_restore_host(vcpu);
635 	if (vcpu_has_nv(vcpu))
636 		kvm_vcpu_put_hw_mmu(vcpu);
637 	kvm_arm_vmid_clear_active();
638 
639 	vcpu_clear_on_unsupported_cpu(vcpu);
640 	vcpu->cpu = -1;
641 }
642 
643 static void __kvm_arm_vcpu_power_off(struct kvm_vcpu *vcpu)
644 {
645 	WRITE_ONCE(vcpu->arch.mp_state.mp_state, KVM_MP_STATE_STOPPED);
646 	kvm_make_request(KVM_REQ_SLEEP, vcpu);
647 	kvm_vcpu_kick(vcpu);
648 }
649 
650 void kvm_arm_vcpu_power_off(struct kvm_vcpu *vcpu)
651 {
652 	spin_lock(&vcpu->arch.mp_state_lock);
653 	__kvm_arm_vcpu_power_off(vcpu);
654 	spin_unlock(&vcpu->arch.mp_state_lock);
655 }
656 
657 bool kvm_arm_vcpu_stopped(struct kvm_vcpu *vcpu)
658 {
659 	return READ_ONCE(vcpu->arch.mp_state.mp_state) == KVM_MP_STATE_STOPPED;
660 }
661 
662 static void kvm_arm_vcpu_suspend(struct kvm_vcpu *vcpu)
663 {
664 	WRITE_ONCE(vcpu->arch.mp_state.mp_state, KVM_MP_STATE_SUSPENDED);
665 	kvm_make_request(KVM_REQ_SUSPEND, vcpu);
666 	kvm_vcpu_kick(vcpu);
667 }
668 
669 static bool kvm_arm_vcpu_suspended(struct kvm_vcpu *vcpu)
670 {
671 	return READ_ONCE(vcpu->arch.mp_state.mp_state) == KVM_MP_STATE_SUSPENDED;
672 }
673 
674 int kvm_arch_vcpu_ioctl_get_mpstate(struct kvm_vcpu *vcpu,
675 				    struct kvm_mp_state *mp_state)
676 {
677 	*mp_state = READ_ONCE(vcpu->arch.mp_state);
678 
679 	return 0;
680 }
681 
682 int kvm_arch_vcpu_ioctl_set_mpstate(struct kvm_vcpu *vcpu,
683 				    struct kvm_mp_state *mp_state)
684 {
685 	int ret = 0;
686 
687 	spin_lock(&vcpu->arch.mp_state_lock);
688 
689 	switch (mp_state->mp_state) {
690 	case KVM_MP_STATE_RUNNABLE:
691 		WRITE_ONCE(vcpu->arch.mp_state, *mp_state);
692 		break;
693 	case KVM_MP_STATE_STOPPED:
694 		__kvm_arm_vcpu_power_off(vcpu);
695 		break;
696 	case KVM_MP_STATE_SUSPENDED:
697 		kvm_arm_vcpu_suspend(vcpu);
698 		break;
699 	default:
700 		ret = -EINVAL;
701 	}
702 
703 	spin_unlock(&vcpu->arch.mp_state_lock);
704 
705 	return ret;
706 }
707 
708 /**
709  * kvm_arch_vcpu_runnable - determine if the vcpu can be scheduled
710  * @v:		The VCPU pointer
711  *
712  * If the guest CPU is not waiting for interrupts or an interrupt line is
713  * asserted, the CPU is by definition runnable.
714  */
715 int kvm_arch_vcpu_runnable(struct kvm_vcpu *v)
716 {
717 	bool irq_lines = *vcpu_hcr(v) & (HCR_VI | HCR_VF);
718 	return ((irq_lines || kvm_vgic_vcpu_pending_irq(v))
719 		&& !kvm_arm_vcpu_stopped(v) && !v->arch.pause);
720 }
721 
722 bool kvm_arch_vcpu_in_kernel(struct kvm_vcpu *vcpu)
723 {
724 	return vcpu_mode_priv(vcpu);
725 }
726 
727 #ifdef CONFIG_GUEST_PERF_EVENTS
728 unsigned long kvm_arch_vcpu_get_ip(struct kvm_vcpu *vcpu)
729 {
730 	return *vcpu_pc(vcpu);
731 }
732 #endif
733 
734 static void kvm_init_mpidr_data(struct kvm *kvm)
735 {
736 	struct kvm_mpidr_data *data = NULL;
737 	unsigned long c, mask, nr_entries;
738 	u64 aff_set = 0, aff_clr = ~0UL;
739 	struct kvm_vcpu *vcpu;
740 
741 	mutex_lock(&kvm->arch.config_lock);
742 
743 	if (rcu_access_pointer(kvm->arch.mpidr_data) ||
744 	    atomic_read(&kvm->online_vcpus) == 1)
745 		goto out;
746 
747 	kvm_for_each_vcpu(c, vcpu, kvm) {
748 		u64 aff = kvm_vcpu_get_mpidr_aff(vcpu);
749 		aff_set |= aff;
750 		aff_clr &= aff;
751 	}
752 
753 	/*
754 	 * A significant bit can be either 0 or 1, and will only appear in
755 	 * aff_set. Use aff_clr to weed out the useless stuff.
756 	 */
757 	mask = aff_set ^ aff_clr;
758 	nr_entries = BIT_ULL(hweight_long(mask));
759 
760 	/*
761 	 * Don't let userspace fool us. If we need more than a single page
762 	 * to describe the compressed MPIDR array, just fall back to the
763 	 * iterative method. Single vcpu VMs do not need this either.
764 	 */
765 	if (struct_size(data, cmpidr_to_idx, nr_entries) <= PAGE_SIZE)
766 		data = kzalloc(struct_size(data, cmpidr_to_idx, nr_entries),
767 			       GFP_KERNEL_ACCOUNT);
768 
769 	if (!data)
770 		goto out;
771 
772 	data->mpidr_mask = mask;
773 
774 	kvm_for_each_vcpu(c, vcpu, kvm) {
775 		u64 aff = kvm_vcpu_get_mpidr_aff(vcpu);
776 		u16 index = kvm_mpidr_index(data, aff);
777 
778 		data->cmpidr_to_idx[index] = c;
779 	}
780 
781 	rcu_assign_pointer(kvm->arch.mpidr_data, data);
782 out:
783 	mutex_unlock(&kvm->arch.config_lock);
784 }
785 
786 /*
787  * Handle both the initialisation that is being done when the vcpu is
788  * run for the first time, as well as the updates that must be
789  * performed each time we get a new thread dealing with this vcpu.
790  */
791 int kvm_arch_vcpu_run_pid_change(struct kvm_vcpu *vcpu)
792 {
793 	struct kvm *kvm = vcpu->kvm;
794 	int ret;
795 
796 	if (!kvm_vcpu_initialized(vcpu))
797 		return -ENOEXEC;
798 
799 	if (!kvm_arm_vcpu_is_finalized(vcpu))
800 		return -EPERM;
801 
802 	ret = kvm_arch_vcpu_run_map_fp(vcpu);
803 	if (ret)
804 		return ret;
805 
806 	if (likely(vcpu_has_run_once(vcpu)))
807 		return 0;
808 
809 	kvm_init_mpidr_data(kvm);
810 
811 	kvm_arm_vcpu_init_debug(vcpu);
812 
813 	if (likely(irqchip_in_kernel(kvm))) {
814 		/*
815 		 * Map the VGIC hardware resources before running a vcpu the
816 		 * first time on this VM.
817 		 */
818 		ret = kvm_vgic_map_resources(kvm);
819 		if (ret)
820 			return ret;
821 	}
822 
823 	ret = kvm_finalize_sys_regs(vcpu);
824 	if (ret)
825 		return ret;
826 
827 	/*
828 	 * This needs to happen after any restriction has been applied
829 	 * to the feature set.
830 	 */
831 	kvm_calculate_traps(vcpu);
832 
833 	ret = kvm_timer_enable(vcpu);
834 	if (ret)
835 		return ret;
836 
837 	ret = kvm_arm_pmu_v3_enable(vcpu);
838 	if (ret)
839 		return ret;
840 
841 	if (is_protected_kvm_enabled()) {
842 		ret = pkvm_create_hyp_vm(kvm);
843 		if (ret)
844 			return ret;
845 	}
846 
847 	mutex_lock(&kvm->arch.config_lock);
848 	set_bit(KVM_ARCH_FLAG_HAS_RAN_ONCE, &kvm->arch.flags);
849 	mutex_unlock(&kvm->arch.config_lock);
850 
851 	return ret;
852 }
853 
854 bool kvm_arch_intc_initialized(struct kvm *kvm)
855 {
856 	return vgic_initialized(kvm);
857 }
858 
859 void kvm_arm_halt_guest(struct kvm *kvm)
860 {
861 	unsigned long i;
862 	struct kvm_vcpu *vcpu;
863 
864 	kvm_for_each_vcpu(i, vcpu, kvm)
865 		vcpu->arch.pause = true;
866 	kvm_make_all_cpus_request(kvm, KVM_REQ_SLEEP);
867 }
868 
869 void kvm_arm_resume_guest(struct kvm *kvm)
870 {
871 	unsigned long i;
872 	struct kvm_vcpu *vcpu;
873 
874 	kvm_for_each_vcpu(i, vcpu, kvm) {
875 		vcpu->arch.pause = false;
876 		__kvm_vcpu_wake_up(vcpu);
877 	}
878 }
879 
880 static void kvm_vcpu_sleep(struct kvm_vcpu *vcpu)
881 {
882 	struct rcuwait *wait = kvm_arch_vcpu_get_wait(vcpu);
883 
884 	rcuwait_wait_event(wait,
885 			   (!kvm_arm_vcpu_stopped(vcpu)) && (!vcpu->arch.pause),
886 			   TASK_INTERRUPTIBLE);
887 
888 	if (kvm_arm_vcpu_stopped(vcpu) || vcpu->arch.pause) {
889 		/* Awaken to handle a signal, request we sleep again later. */
890 		kvm_make_request(KVM_REQ_SLEEP, vcpu);
891 	}
892 
893 	/*
894 	 * Make sure we will observe a potential reset request if we've
895 	 * observed a change to the power state. Pairs with the smp_wmb() in
896 	 * kvm_psci_vcpu_on().
897 	 */
898 	smp_rmb();
899 }
900 
901 /**
902  * kvm_vcpu_wfi - emulate Wait-For-Interrupt behavior
903  * @vcpu:	The VCPU pointer
904  *
905  * Suspend execution of a vCPU until a valid wake event is detected, i.e. until
906  * the vCPU is runnable.  The vCPU may or may not be scheduled out, depending
907  * on when a wake event arrives, e.g. there may already be a pending wake event.
908  */
909 void kvm_vcpu_wfi(struct kvm_vcpu *vcpu)
910 {
911 	/*
912 	 * Sync back the state of the GIC CPU interface so that we have
913 	 * the latest PMR and group enables. This ensures that
914 	 * kvm_arch_vcpu_runnable has up-to-date data to decide whether
915 	 * we have pending interrupts, e.g. when determining if the
916 	 * vCPU should block.
917 	 *
918 	 * For the same reason, we want to tell GICv4 that we need
919 	 * doorbells to be signalled, should an interrupt become pending.
920 	 */
921 	preempt_disable();
922 	vcpu_set_flag(vcpu, IN_WFI);
923 	kvm_vgic_put(vcpu);
924 	preempt_enable();
925 
926 	kvm_vcpu_halt(vcpu);
927 	vcpu_clear_flag(vcpu, IN_WFIT);
928 
929 	preempt_disable();
930 	vcpu_clear_flag(vcpu, IN_WFI);
931 	kvm_vgic_load(vcpu);
932 	preempt_enable();
933 }
934 
935 static int kvm_vcpu_suspend(struct kvm_vcpu *vcpu)
936 {
937 	if (!kvm_arm_vcpu_suspended(vcpu))
938 		return 1;
939 
940 	kvm_vcpu_wfi(vcpu);
941 
942 	/*
943 	 * The suspend state is sticky; we do not leave it until userspace
944 	 * explicitly marks the vCPU as runnable. Request that we suspend again
945 	 * later.
946 	 */
947 	kvm_make_request(KVM_REQ_SUSPEND, vcpu);
948 
949 	/*
950 	 * Check to make sure the vCPU is actually runnable. If so, exit to
951 	 * userspace informing it of the wakeup condition.
952 	 */
953 	if (kvm_arch_vcpu_runnable(vcpu)) {
954 		memset(&vcpu->run->system_event, 0, sizeof(vcpu->run->system_event));
955 		vcpu->run->system_event.type = KVM_SYSTEM_EVENT_WAKEUP;
956 		vcpu->run->exit_reason = KVM_EXIT_SYSTEM_EVENT;
957 		return 0;
958 	}
959 
960 	/*
961 	 * Otherwise, we were unblocked to process a different event, such as a
962 	 * pending signal. Return 1 and allow kvm_arch_vcpu_ioctl_run() to
963 	 * process the event.
964 	 */
965 	return 1;
966 }
967 
968 /**
969  * check_vcpu_requests - check and handle pending vCPU requests
970  * @vcpu:	the VCPU pointer
971  *
972  * Return: 1 if we should enter the guest
973  *	   0 if we should exit to userspace
974  *	   < 0 if we should exit to userspace, where the return value indicates
975  *	   an error
976  */
977 static int check_vcpu_requests(struct kvm_vcpu *vcpu)
978 {
979 	if (kvm_request_pending(vcpu)) {
980 		if (kvm_check_request(KVM_REQ_VM_DEAD, vcpu))
981 			return -EIO;
982 
983 		if (kvm_check_request(KVM_REQ_SLEEP, vcpu))
984 			kvm_vcpu_sleep(vcpu);
985 
986 		if (kvm_check_request(KVM_REQ_VCPU_RESET, vcpu))
987 			kvm_reset_vcpu(vcpu);
988 
989 		/*
990 		 * Clear IRQ_PENDING requests that were made to guarantee
991 		 * that a VCPU sees new virtual interrupts.
992 		 */
993 		kvm_check_request(KVM_REQ_IRQ_PENDING, vcpu);
994 
995 		if (kvm_check_request(KVM_REQ_RECORD_STEAL, vcpu))
996 			kvm_update_stolen_time(vcpu);
997 
998 		if (kvm_check_request(KVM_REQ_RELOAD_GICv4, vcpu)) {
999 			/* The distributor enable bits were changed */
1000 			preempt_disable();
1001 			vgic_v4_put(vcpu);
1002 			vgic_v4_load(vcpu);
1003 			preempt_enable();
1004 		}
1005 
1006 		if (kvm_check_request(KVM_REQ_RELOAD_PMU, vcpu))
1007 			kvm_vcpu_reload_pmu(vcpu);
1008 
1009 		if (kvm_check_request(KVM_REQ_RESYNC_PMU_EL0, vcpu))
1010 			kvm_vcpu_pmu_restore_guest(vcpu);
1011 
1012 		if (kvm_check_request(KVM_REQ_SUSPEND, vcpu))
1013 			return kvm_vcpu_suspend(vcpu);
1014 
1015 		if (kvm_dirty_ring_check_request(vcpu))
1016 			return 0;
1017 
1018 		check_nested_vcpu_requests(vcpu);
1019 	}
1020 
1021 	return 1;
1022 }
1023 
1024 static bool vcpu_mode_is_bad_32bit(struct kvm_vcpu *vcpu)
1025 {
1026 	if (likely(!vcpu_mode_is_32bit(vcpu)))
1027 		return false;
1028 
1029 	if (vcpu_has_nv(vcpu))
1030 		return true;
1031 
1032 	return !kvm_supports_32bit_el0();
1033 }
1034 
1035 /**
1036  * kvm_vcpu_exit_request - returns true if the VCPU should *not* enter the guest
1037  * @vcpu:	The VCPU pointer
1038  * @ret:	Pointer to write optional return code
1039  *
1040  * Returns: true if the VCPU needs to return to a preemptible + interruptible
1041  *	    and skip guest entry.
1042  *
1043  * This function disambiguates between two different types of exits: exits to a
1044  * preemptible + interruptible kernel context and exits to userspace. For an
1045  * exit to userspace, this function will write the return code to ret and return
1046  * true. For an exit to preemptible + interruptible kernel context (i.e. check
1047  * for pending work and re-enter), return true without writing to ret.
1048  */
1049 static bool kvm_vcpu_exit_request(struct kvm_vcpu *vcpu, int *ret)
1050 {
1051 	struct kvm_run *run = vcpu->run;
1052 
1053 	/*
1054 	 * If we're using a userspace irqchip, then check if we need
1055 	 * to tell a userspace irqchip about timer or PMU level
1056 	 * changes and if so, exit to userspace (the actual level
1057 	 * state gets updated in kvm_timer_update_run and
1058 	 * kvm_pmu_update_run below).
1059 	 */
1060 	if (unlikely(!irqchip_in_kernel(vcpu->kvm))) {
1061 		if (kvm_timer_should_notify_user(vcpu) ||
1062 		    kvm_pmu_should_notify_user(vcpu)) {
1063 			*ret = -EINTR;
1064 			run->exit_reason = KVM_EXIT_INTR;
1065 			return true;
1066 		}
1067 	}
1068 
1069 	if (unlikely(vcpu_on_unsupported_cpu(vcpu))) {
1070 		run->exit_reason = KVM_EXIT_FAIL_ENTRY;
1071 		run->fail_entry.hardware_entry_failure_reason = KVM_EXIT_FAIL_ENTRY_CPU_UNSUPPORTED;
1072 		run->fail_entry.cpu = smp_processor_id();
1073 		*ret = 0;
1074 		return true;
1075 	}
1076 
1077 	return kvm_request_pending(vcpu) ||
1078 			xfer_to_guest_mode_work_pending();
1079 }
1080 
1081 /*
1082  * Actually run the vCPU, entering an RCU extended quiescent state (EQS) while
1083  * the vCPU is running.
1084  *
1085  * This must be noinstr as instrumentation may make use of RCU, and this is not
1086  * safe during the EQS.
1087  */
1088 static int noinstr kvm_arm_vcpu_enter_exit(struct kvm_vcpu *vcpu)
1089 {
1090 	int ret;
1091 
1092 	guest_state_enter_irqoff();
1093 	ret = kvm_call_hyp_ret(__kvm_vcpu_run, vcpu);
1094 	guest_state_exit_irqoff();
1095 
1096 	return ret;
1097 }
1098 
1099 /**
1100  * kvm_arch_vcpu_ioctl_run - the main VCPU run function to execute guest code
1101  * @vcpu:	The VCPU pointer
1102  *
1103  * This function is called through the VCPU_RUN ioctl called from user space. It
1104  * will execute VM code in a loop until the time slice for the process is used
1105  * or some emulation is needed from user space in which case the function will
1106  * return with return value 0 and with the kvm_run structure filled in with the
1107  * required data for the requested emulation.
1108  */
1109 int kvm_arch_vcpu_ioctl_run(struct kvm_vcpu *vcpu)
1110 {
1111 	struct kvm_run *run = vcpu->run;
1112 	int ret;
1113 
1114 	if (run->exit_reason == KVM_EXIT_MMIO) {
1115 		ret = kvm_handle_mmio_return(vcpu);
1116 		if (ret <= 0)
1117 			return ret;
1118 	}
1119 
1120 	vcpu_load(vcpu);
1121 
1122 	if (!vcpu->wants_to_run) {
1123 		ret = -EINTR;
1124 		goto out;
1125 	}
1126 
1127 	kvm_sigset_activate(vcpu);
1128 
1129 	ret = 1;
1130 	run->exit_reason = KVM_EXIT_UNKNOWN;
1131 	run->flags = 0;
1132 	while (ret > 0) {
1133 		/*
1134 		 * Check conditions before entering the guest
1135 		 */
1136 		ret = xfer_to_guest_mode_handle_work(vcpu);
1137 		if (!ret)
1138 			ret = 1;
1139 
1140 		if (ret > 0)
1141 			ret = check_vcpu_requests(vcpu);
1142 
1143 		/*
1144 		 * Preparing the interrupts to be injected also
1145 		 * involves poking the GIC, which must be done in a
1146 		 * non-preemptible context.
1147 		 */
1148 		preempt_disable();
1149 
1150 		/*
1151 		 * The VMID allocator only tracks active VMIDs per
1152 		 * physical CPU, and therefore the VMID allocated may not be
1153 		 * preserved on VMID roll-over if the task was preempted,
1154 		 * making a thread's VMID inactive. So we need to call
1155 		 * kvm_arm_vmid_update() in non-premptible context.
1156 		 */
1157 		if (kvm_arm_vmid_update(&vcpu->arch.hw_mmu->vmid) &&
1158 		    has_vhe())
1159 			__load_stage2(vcpu->arch.hw_mmu,
1160 				      vcpu->arch.hw_mmu->arch);
1161 
1162 		kvm_pmu_flush_hwstate(vcpu);
1163 
1164 		local_irq_disable();
1165 
1166 		kvm_vgic_flush_hwstate(vcpu);
1167 
1168 		kvm_pmu_update_vcpu_events(vcpu);
1169 
1170 		/*
1171 		 * Ensure we set mode to IN_GUEST_MODE after we disable
1172 		 * interrupts and before the final VCPU requests check.
1173 		 * See the comment in kvm_vcpu_exiting_guest_mode() and
1174 		 * Documentation/virt/kvm/vcpu-requests.rst
1175 		 */
1176 		smp_store_mb(vcpu->mode, IN_GUEST_MODE);
1177 
1178 		if (ret <= 0 || kvm_vcpu_exit_request(vcpu, &ret)) {
1179 			vcpu->mode = OUTSIDE_GUEST_MODE;
1180 			isb(); /* Ensure work in x_flush_hwstate is committed */
1181 			kvm_pmu_sync_hwstate(vcpu);
1182 			if (unlikely(!irqchip_in_kernel(vcpu->kvm)))
1183 				kvm_timer_sync_user(vcpu);
1184 			kvm_vgic_sync_hwstate(vcpu);
1185 			local_irq_enable();
1186 			preempt_enable();
1187 			continue;
1188 		}
1189 
1190 		kvm_arm_setup_debug(vcpu);
1191 		kvm_arch_vcpu_ctxflush_fp(vcpu);
1192 
1193 		/**************************************************************
1194 		 * Enter the guest
1195 		 */
1196 		trace_kvm_entry(*vcpu_pc(vcpu));
1197 		guest_timing_enter_irqoff();
1198 
1199 		ret = kvm_arm_vcpu_enter_exit(vcpu);
1200 
1201 		vcpu->mode = OUTSIDE_GUEST_MODE;
1202 		vcpu->stat.exits++;
1203 		/*
1204 		 * Back from guest
1205 		 *************************************************************/
1206 
1207 		kvm_arm_clear_debug(vcpu);
1208 
1209 		/*
1210 		 * We must sync the PMU state before the vgic state so
1211 		 * that the vgic can properly sample the updated state of the
1212 		 * interrupt line.
1213 		 */
1214 		kvm_pmu_sync_hwstate(vcpu);
1215 
1216 		/*
1217 		 * Sync the vgic state before syncing the timer state because
1218 		 * the timer code needs to know if the virtual timer
1219 		 * interrupts are active.
1220 		 */
1221 		kvm_vgic_sync_hwstate(vcpu);
1222 
1223 		/*
1224 		 * Sync the timer hardware state before enabling interrupts as
1225 		 * we don't want vtimer interrupts to race with syncing the
1226 		 * timer virtual interrupt state.
1227 		 */
1228 		if (unlikely(!irqchip_in_kernel(vcpu->kvm)))
1229 			kvm_timer_sync_user(vcpu);
1230 
1231 		kvm_arch_vcpu_ctxsync_fp(vcpu);
1232 
1233 		/*
1234 		 * We must ensure that any pending interrupts are taken before
1235 		 * we exit guest timing so that timer ticks are accounted as
1236 		 * guest time. Transiently unmask interrupts so that any
1237 		 * pending interrupts are taken.
1238 		 *
1239 		 * Per ARM DDI 0487G.b section D1.13.4, an ISB (or other
1240 		 * context synchronization event) is necessary to ensure that
1241 		 * pending interrupts are taken.
1242 		 */
1243 		if (ARM_EXCEPTION_CODE(ret) == ARM_EXCEPTION_IRQ) {
1244 			local_irq_enable();
1245 			isb();
1246 			local_irq_disable();
1247 		}
1248 
1249 		guest_timing_exit_irqoff();
1250 
1251 		local_irq_enable();
1252 
1253 		trace_kvm_exit(ret, kvm_vcpu_trap_get_class(vcpu), *vcpu_pc(vcpu));
1254 
1255 		/* Exit types that need handling before we can be preempted */
1256 		handle_exit_early(vcpu, ret);
1257 
1258 		preempt_enable();
1259 
1260 		/*
1261 		 * The ARMv8 architecture doesn't give the hypervisor
1262 		 * a mechanism to prevent a guest from dropping to AArch32 EL0
1263 		 * if implemented by the CPU. If we spot the guest in such
1264 		 * state and that we decided it wasn't supposed to do so (like
1265 		 * with the asymmetric AArch32 case), return to userspace with
1266 		 * a fatal error.
1267 		 */
1268 		if (vcpu_mode_is_bad_32bit(vcpu)) {
1269 			/*
1270 			 * As we have caught the guest red-handed, decide that
1271 			 * it isn't fit for purpose anymore by making the vcpu
1272 			 * invalid. The VMM can try and fix it by issuing  a
1273 			 * KVM_ARM_VCPU_INIT if it really wants to.
1274 			 */
1275 			vcpu_clear_flag(vcpu, VCPU_INITIALIZED);
1276 			ret = ARM_EXCEPTION_IL;
1277 		}
1278 
1279 		ret = handle_exit(vcpu, ret);
1280 	}
1281 
1282 	/* Tell userspace about in-kernel device output levels */
1283 	if (unlikely(!irqchip_in_kernel(vcpu->kvm))) {
1284 		kvm_timer_update_run(vcpu);
1285 		kvm_pmu_update_run(vcpu);
1286 	}
1287 
1288 	kvm_sigset_deactivate(vcpu);
1289 
1290 out:
1291 	/*
1292 	 * In the unlikely event that we are returning to userspace
1293 	 * with pending exceptions or PC adjustment, commit these
1294 	 * adjustments in order to give userspace a consistent view of
1295 	 * the vcpu state. Note that this relies on __kvm_adjust_pc()
1296 	 * being preempt-safe on VHE.
1297 	 */
1298 	if (unlikely(vcpu_get_flag(vcpu, PENDING_EXCEPTION) ||
1299 		     vcpu_get_flag(vcpu, INCREMENT_PC)))
1300 		kvm_call_hyp(__kvm_adjust_pc, vcpu);
1301 
1302 	vcpu_put(vcpu);
1303 	return ret;
1304 }
1305 
1306 static int vcpu_interrupt_line(struct kvm_vcpu *vcpu, int number, bool level)
1307 {
1308 	int bit_index;
1309 	bool set;
1310 	unsigned long *hcr;
1311 
1312 	if (number == KVM_ARM_IRQ_CPU_IRQ)
1313 		bit_index = __ffs(HCR_VI);
1314 	else /* KVM_ARM_IRQ_CPU_FIQ */
1315 		bit_index = __ffs(HCR_VF);
1316 
1317 	hcr = vcpu_hcr(vcpu);
1318 	if (level)
1319 		set = test_and_set_bit(bit_index, hcr);
1320 	else
1321 		set = test_and_clear_bit(bit_index, hcr);
1322 
1323 	/*
1324 	 * If we didn't change anything, no need to wake up or kick other CPUs
1325 	 */
1326 	if (set == level)
1327 		return 0;
1328 
1329 	/*
1330 	 * The vcpu irq_lines field was updated, wake up sleeping VCPUs and
1331 	 * trigger a world-switch round on the running physical CPU to set the
1332 	 * virtual IRQ/FIQ fields in the HCR appropriately.
1333 	 */
1334 	kvm_make_request(KVM_REQ_IRQ_PENDING, vcpu);
1335 	kvm_vcpu_kick(vcpu);
1336 
1337 	return 0;
1338 }
1339 
1340 int kvm_vm_ioctl_irq_line(struct kvm *kvm, struct kvm_irq_level *irq_level,
1341 			  bool line_status)
1342 {
1343 	u32 irq = irq_level->irq;
1344 	unsigned int irq_type, vcpu_id, irq_num;
1345 	struct kvm_vcpu *vcpu = NULL;
1346 	bool level = irq_level->level;
1347 
1348 	irq_type = (irq >> KVM_ARM_IRQ_TYPE_SHIFT) & KVM_ARM_IRQ_TYPE_MASK;
1349 	vcpu_id = (irq >> KVM_ARM_IRQ_VCPU_SHIFT) & KVM_ARM_IRQ_VCPU_MASK;
1350 	vcpu_id += ((irq >> KVM_ARM_IRQ_VCPU2_SHIFT) & KVM_ARM_IRQ_VCPU2_MASK) * (KVM_ARM_IRQ_VCPU_MASK + 1);
1351 	irq_num = (irq >> KVM_ARM_IRQ_NUM_SHIFT) & KVM_ARM_IRQ_NUM_MASK;
1352 
1353 	trace_kvm_irq_line(irq_type, vcpu_id, irq_num, irq_level->level);
1354 
1355 	switch (irq_type) {
1356 	case KVM_ARM_IRQ_TYPE_CPU:
1357 		if (irqchip_in_kernel(kvm))
1358 			return -ENXIO;
1359 
1360 		vcpu = kvm_get_vcpu_by_id(kvm, vcpu_id);
1361 		if (!vcpu)
1362 			return -EINVAL;
1363 
1364 		if (irq_num > KVM_ARM_IRQ_CPU_FIQ)
1365 			return -EINVAL;
1366 
1367 		return vcpu_interrupt_line(vcpu, irq_num, level);
1368 	case KVM_ARM_IRQ_TYPE_PPI:
1369 		if (!irqchip_in_kernel(kvm))
1370 			return -ENXIO;
1371 
1372 		vcpu = kvm_get_vcpu_by_id(kvm, vcpu_id);
1373 		if (!vcpu)
1374 			return -EINVAL;
1375 
1376 		if (irq_num < VGIC_NR_SGIS || irq_num >= VGIC_NR_PRIVATE_IRQS)
1377 			return -EINVAL;
1378 
1379 		return kvm_vgic_inject_irq(kvm, vcpu, irq_num, level, NULL);
1380 	case KVM_ARM_IRQ_TYPE_SPI:
1381 		if (!irqchip_in_kernel(kvm))
1382 			return -ENXIO;
1383 
1384 		if (irq_num < VGIC_NR_PRIVATE_IRQS)
1385 			return -EINVAL;
1386 
1387 		return kvm_vgic_inject_irq(kvm, NULL, irq_num, level, NULL);
1388 	}
1389 
1390 	return -EINVAL;
1391 }
1392 
1393 static unsigned long system_supported_vcpu_features(void)
1394 {
1395 	unsigned long features = KVM_VCPU_VALID_FEATURES;
1396 
1397 	if (!cpus_have_final_cap(ARM64_HAS_32BIT_EL1))
1398 		clear_bit(KVM_ARM_VCPU_EL1_32BIT, &features);
1399 
1400 	if (!kvm_arm_support_pmu_v3())
1401 		clear_bit(KVM_ARM_VCPU_PMU_V3, &features);
1402 
1403 	if (!system_supports_sve())
1404 		clear_bit(KVM_ARM_VCPU_SVE, &features);
1405 
1406 	if (!kvm_has_full_ptr_auth()) {
1407 		clear_bit(KVM_ARM_VCPU_PTRAUTH_ADDRESS, &features);
1408 		clear_bit(KVM_ARM_VCPU_PTRAUTH_GENERIC, &features);
1409 	}
1410 
1411 	if (!cpus_have_final_cap(ARM64_HAS_NESTED_VIRT))
1412 		clear_bit(KVM_ARM_VCPU_HAS_EL2, &features);
1413 
1414 	return features;
1415 }
1416 
1417 static int kvm_vcpu_init_check_features(struct kvm_vcpu *vcpu,
1418 					const struct kvm_vcpu_init *init)
1419 {
1420 	unsigned long features = init->features[0];
1421 	int i;
1422 
1423 	if (features & ~KVM_VCPU_VALID_FEATURES)
1424 		return -ENOENT;
1425 
1426 	for (i = 1; i < ARRAY_SIZE(init->features); i++) {
1427 		if (init->features[i])
1428 			return -ENOENT;
1429 	}
1430 
1431 	if (features & ~system_supported_vcpu_features())
1432 		return -EINVAL;
1433 
1434 	/*
1435 	 * For now make sure that both address/generic pointer authentication
1436 	 * features are requested by the userspace together.
1437 	 */
1438 	if (test_bit(KVM_ARM_VCPU_PTRAUTH_ADDRESS, &features) !=
1439 	    test_bit(KVM_ARM_VCPU_PTRAUTH_GENERIC, &features))
1440 		return -EINVAL;
1441 
1442 	if (!test_bit(KVM_ARM_VCPU_EL1_32BIT, &features))
1443 		return 0;
1444 
1445 	/* MTE is incompatible with AArch32 */
1446 	if (kvm_has_mte(vcpu->kvm))
1447 		return -EINVAL;
1448 
1449 	/* NV is incompatible with AArch32 */
1450 	if (test_bit(KVM_ARM_VCPU_HAS_EL2, &features))
1451 		return -EINVAL;
1452 
1453 	return 0;
1454 }
1455 
1456 static bool kvm_vcpu_init_changed(struct kvm_vcpu *vcpu,
1457 				  const struct kvm_vcpu_init *init)
1458 {
1459 	unsigned long features = init->features[0];
1460 
1461 	return !bitmap_equal(vcpu->kvm->arch.vcpu_features, &features,
1462 			     KVM_VCPU_MAX_FEATURES);
1463 }
1464 
1465 static int kvm_setup_vcpu(struct kvm_vcpu *vcpu)
1466 {
1467 	struct kvm *kvm = vcpu->kvm;
1468 	int ret = 0;
1469 
1470 	/*
1471 	 * When the vCPU has a PMU, but no PMU is set for the guest
1472 	 * yet, set the default one.
1473 	 */
1474 	if (kvm_vcpu_has_pmu(vcpu) && !kvm->arch.arm_pmu)
1475 		ret = kvm_arm_set_default_pmu(kvm);
1476 
1477 	/* Prepare for nested if required */
1478 	if (!ret && vcpu_has_nv(vcpu))
1479 		ret = kvm_vcpu_init_nested(vcpu);
1480 
1481 	return ret;
1482 }
1483 
1484 static int __kvm_vcpu_set_target(struct kvm_vcpu *vcpu,
1485 				 const struct kvm_vcpu_init *init)
1486 {
1487 	unsigned long features = init->features[0];
1488 	struct kvm *kvm = vcpu->kvm;
1489 	int ret = -EINVAL;
1490 
1491 	mutex_lock(&kvm->arch.config_lock);
1492 
1493 	if (test_bit(KVM_ARCH_FLAG_VCPU_FEATURES_CONFIGURED, &kvm->arch.flags) &&
1494 	    kvm_vcpu_init_changed(vcpu, init))
1495 		goto out_unlock;
1496 
1497 	bitmap_copy(kvm->arch.vcpu_features, &features, KVM_VCPU_MAX_FEATURES);
1498 
1499 	ret = kvm_setup_vcpu(vcpu);
1500 	if (ret)
1501 		goto out_unlock;
1502 
1503 	/* Now we know what it is, we can reset it. */
1504 	kvm_reset_vcpu(vcpu);
1505 
1506 	set_bit(KVM_ARCH_FLAG_VCPU_FEATURES_CONFIGURED, &kvm->arch.flags);
1507 	vcpu_set_flag(vcpu, VCPU_INITIALIZED);
1508 	ret = 0;
1509 out_unlock:
1510 	mutex_unlock(&kvm->arch.config_lock);
1511 	return ret;
1512 }
1513 
1514 static int kvm_vcpu_set_target(struct kvm_vcpu *vcpu,
1515 			       const struct kvm_vcpu_init *init)
1516 {
1517 	int ret;
1518 
1519 	if (init->target != KVM_ARM_TARGET_GENERIC_V8 &&
1520 	    init->target != kvm_target_cpu())
1521 		return -EINVAL;
1522 
1523 	ret = kvm_vcpu_init_check_features(vcpu, init);
1524 	if (ret)
1525 		return ret;
1526 
1527 	if (!kvm_vcpu_initialized(vcpu))
1528 		return __kvm_vcpu_set_target(vcpu, init);
1529 
1530 	if (kvm_vcpu_init_changed(vcpu, init))
1531 		return -EINVAL;
1532 
1533 	kvm_reset_vcpu(vcpu);
1534 	return 0;
1535 }
1536 
1537 static int kvm_arch_vcpu_ioctl_vcpu_init(struct kvm_vcpu *vcpu,
1538 					 struct kvm_vcpu_init *init)
1539 {
1540 	bool power_off = false;
1541 	int ret;
1542 
1543 	/*
1544 	 * Treat the power-off vCPU feature as ephemeral. Clear the bit to avoid
1545 	 * reflecting it in the finalized feature set, thus limiting its scope
1546 	 * to a single KVM_ARM_VCPU_INIT call.
1547 	 */
1548 	if (init->features[0] & BIT(KVM_ARM_VCPU_POWER_OFF)) {
1549 		init->features[0] &= ~BIT(KVM_ARM_VCPU_POWER_OFF);
1550 		power_off = true;
1551 	}
1552 
1553 	ret = kvm_vcpu_set_target(vcpu, init);
1554 	if (ret)
1555 		return ret;
1556 
1557 	/*
1558 	 * Ensure a rebooted VM will fault in RAM pages and detect if the
1559 	 * guest MMU is turned off and flush the caches as needed.
1560 	 *
1561 	 * S2FWB enforces all memory accesses to RAM being cacheable,
1562 	 * ensuring that the data side is always coherent. We still
1563 	 * need to invalidate the I-cache though, as FWB does *not*
1564 	 * imply CTR_EL0.DIC.
1565 	 */
1566 	if (vcpu_has_run_once(vcpu)) {
1567 		if (!cpus_have_final_cap(ARM64_HAS_STAGE2_FWB))
1568 			stage2_unmap_vm(vcpu->kvm);
1569 		else
1570 			icache_inval_all_pou();
1571 	}
1572 
1573 	vcpu_reset_hcr(vcpu);
1574 	vcpu->arch.cptr_el2 = kvm_get_reset_cptr_el2(vcpu);
1575 
1576 	/*
1577 	 * Handle the "start in power-off" case.
1578 	 */
1579 	spin_lock(&vcpu->arch.mp_state_lock);
1580 
1581 	if (power_off)
1582 		__kvm_arm_vcpu_power_off(vcpu);
1583 	else
1584 		WRITE_ONCE(vcpu->arch.mp_state.mp_state, KVM_MP_STATE_RUNNABLE);
1585 
1586 	spin_unlock(&vcpu->arch.mp_state_lock);
1587 
1588 	return 0;
1589 }
1590 
1591 static int kvm_arm_vcpu_set_attr(struct kvm_vcpu *vcpu,
1592 				 struct kvm_device_attr *attr)
1593 {
1594 	int ret = -ENXIO;
1595 
1596 	switch (attr->group) {
1597 	default:
1598 		ret = kvm_arm_vcpu_arch_set_attr(vcpu, attr);
1599 		break;
1600 	}
1601 
1602 	return ret;
1603 }
1604 
1605 static int kvm_arm_vcpu_get_attr(struct kvm_vcpu *vcpu,
1606 				 struct kvm_device_attr *attr)
1607 {
1608 	int ret = -ENXIO;
1609 
1610 	switch (attr->group) {
1611 	default:
1612 		ret = kvm_arm_vcpu_arch_get_attr(vcpu, attr);
1613 		break;
1614 	}
1615 
1616 	return ret;
1617 }
1618 
1619 static int kvm_arm_vcpu_has_attr(struct kvm_vcpu *vcpu,
1620 				 struct kvm_device_attr *attr)
1621 {
1622 	int ret = -ENXIO;
1623 
1624 	switch (attr->group) {
1625 	default:
1626 		ret = kvm_arm_vcpu_arch_has_attr(vcpu, attr);
1627 		break;
1628 	}
1629 
1630 	return ret;
1631 }
1632 
1633 static int kvm_arm_vcpu_get_events(struct kvm_vcpu *vcpu,
1634 				   struct kvm_vcpu_events *events)
1635 {
1636 	memset(events, 0, sizeof(*events));
1637 
1638 	return __kvm_arm_vcpu_get_events(vcpu, events);
1639 }
1640 
1641 static int kvm_arm_vcpu_set_events(struct kvm_vcpu *vcpu,
1642 				   struct kvm_vcpu_events *events)
1643 {
1644 	int i;
1645 
1646 	/* check whether the reserved field is zero */
1647 	for (i = 0; i < ARRAY_SIZE(events->reserved); i++)
1648 		if (events->reserved[i])
1649 			return -EINVAL;
1650 
1651 	/* check whether the pad field is zero */
1652 	for (i = 0; i < ARRAY_SIZE(events->exception.pad); i++)
1653 		if (events->exception.pad[i])
1654 			return -EINVAL;
1655 
1656 	return __kvm_arm_vcpu_set_events(vcpu, events);
1657 }
1658 
1659 long kvm_arch_vcpu_ioctl(struct file *filp,
1660 			 unsigned int ioctl, unsigned long arg)
1661 {
1662 	struct kvm_vcpu *vcpu = filp->private_data;
1663 	void __user *argp = (void __user *)arg;
1664 	struct kvm_device_attr attr;
1665 	long r;
1666 
1667 	switch (ioctl) {
1668 	case KVM_ARM_VCPU_INIT: {
1669 		struct kvm_vcpu_init init;
1670 
1671 		r = -EFAULT;
1672 		if (copy_from_user(&init, argp, sizeof(init)))
1673 			break;
1674 
1675 		r = kvm_arch_vcpu_ioctl_vcpu_init(vcpu, &init);
1676 		break;
1677 	}
1678 	case KVM_SET_ONE_REG:
1679 	case KVM_GET_ONE_REG: {
1680 		struct kvm_one_reg reg;
1681 
1682 		r = -ENOEXEC;
1683 		if (unlikely(!kvm_vcpu_initialized(vcpu)))
1684 			break;
1685 
1686 		r = -EFAULT;
1687 		if (copy_from_user(&reg, argp, sizeof(reg)))
1688 			break;
1689 
1690 		/*
1691 		 * We could owe a reset due to PSCI. Handle the pending reset
1692 		 * here to ensure userspace register accesses are ordered after
1693 		 * the reset.
1694 		 */
1695 		if (kvm_check_request(KVM_REQ_VCPU_RESET, vcpu))
1696 			kvm_reset_vcpu(vcpu);
1697 
1698 		if (ioctl == KVM_SET_ONE_REG)
1699 			r = kvm_arm_set_reg(vcpu, &reg);
1700 		else
1701 			r = kvm_arm_get_reg(vcpu, &reg);
1702 		break;
1703 	}
1704 	case KVM_GET_REG_LIST: {
1705 		struct kvm_reg_list __user *user_list = argp;
1706 		struct kvm_reg_list reg_list;
1707 		unsigned n;
1708 
1709 		r = -ENOEXEC;
1710 		if (unlikely(!kvm_vcpu_initialized(vcpu)))
1711 			break;
1712 
1713 		r = -EPERM;
1714 		if (!kvm_arm_vcpu_is_finalized(vcpu))
1715 			break;
1716 
1717 		r = -EFAULT;
1718 		if (copy_from_user(&reg_list, user_list, sizeof(reg_list)))
1719 			break;
1720 		n = reg_list.n;
1721 		reg_list.n = kvm_arm_num_regs(vcpu);
1722 		if (copy_to_user(user_list, &reg_list, sizeof(reg_list)))
1723 			break;
1724 		r = -E2BIG;
1725 		if (n < reg_list.n)
1726 			break;
1727 		r = kvm_arm_copy_reg_indices(vcpu, user_list->reg);
1728 		break;
1729 	}
1730 	case KVM_SET_DEVICE_ATTR: {
1731 		r = -EFAULT;
1732 		if (copy_from_user(&attr, argp, sizeof(attr)))
1733 			break;
1734 		r = kvm_arm_vcpu_set_attr(vcpu, &attr);
1735 		break;
1736 	}
1737 	case KVM_GET_DEVICE_ATTR: {
1738 		r = -EFAULT;
1739 		if (copy_from_user(&attr, argp, sizeof(attr)))
1740 			break;
1741 		r = kvm_arm_vcpu_get_attr(vcpu, &attr);
1742 		break;
1743 	}
1744 	case KVM_HAS_DEVICE_ATTR: {
1745 		r = -EFAULT;
1746 		if (copy_from_user(&attr, argp, sizeof(attr)))
1747 			break;
1748 		r = kvm_arm_vcpu_has_attr(vcpu, &attr);
1749 		break;
1750 	}
1751 	case KVM_GET_VCPU_EVENTS: {
1752 		struct kvm_vcpu_events events;
1753 
1754 		if (kvm_arm_vcpu_get_events(vcpu, &events))
1755 			return -EINVAL;
1756 
1757 		if (copy_to_user(argp, &events, sizeof(events)))
1758 			return -EFAULT;
1759 
1760 		return 0;
1761 	}
1762 	case KVM_SET_VCPU_EVENTS: {
1763 		struct kvm_vcpu_events events;
1764 
1765 		if (copy_from_user(&events, argp, sizeof(events)))
1766 			return -EFAULT;
1767 
1768 		return kvm_arm_vcpu_set_events(vcpu, &events);
1769 	}
1770 	case KVM_ARM_VCPU_FINALIZE: {
1771 		int what;
1772 
1773 		if (!kvm_vcpu_initialized(vcpu))
1774 			return -ENOEXEC;
1775 
1776 		if (get_user(what, (const int __user *)argp))
1777 			return -EFAULT;
1778 
1779 		return kvm_arm_vcpu_finalize(vcpu, what);
1780 	}
1781 	default:
1782 		r = -EINVAL;
1783 	}
1784 
1785 	return r;
1786 }
1787 
1788 void kvm_arch_sync_dirty_log(struct kvm *kvm, struct kvm_memory_slot *memslot)
1789 {
1790 
1791 }
1792 
1793 static int kvm_vm_ioctl_set_device_addr(struct kvm *kvm,
1794 					struct kvm_arm_device_addr *dev_addr)
1795 {
1796 	switch (FIELD_GET(KVM_ARM_DEVICE_ID_MASK, dev_addr->id)) {
1797 	case KVM_ARM_DEVICE_VGIC_V2:
1798 		if (!vgic_present)
1799 			return -ENXIO;
1800 		return kvm_set_legacy_vgic_v2_addr(kvm, dev_addr);
1801 	default:
1802 		return -ENODEV;
1803 	}
1804 }
1805 
1806 static int kvm_vm_has_attr(struct kvm *kvm, struct kvm_device_attr *attr)
1807 {
1808 	switch (attr->group) {
1809 	case KVM_ARM_VM_SMCCC_CTRL:
1810 		return kvm_vm_smccc_has_attr(kvm, attr);
1811 	default:
1812 		return -ENXIO;
1813 	}
1814 }
1815 
1816 static int kvm_vm_set_attr(struct kvm *kvm, struct kvm_device_attr *attr)
1817 {
1818 	switch (attr->group) {
1819 	case KVM_ARM_VM_SMCCC_CTRL:
1820 		return kvm_vm_smccc_set_attr(kvm, attr);
1821 	default:
1822 		return -ENXIO;
1823 	}
1824 }
1825 
1826 int kvm_arch_vm_ioctl(struct file *filp, unsigned int ioctl, unsigned long arg)
1827 {
1828 	struct kvm *kvm = filp->private_data;
1829 	void __user *argp = (void __user *)arg;
1830 	struct kvm_device_attr attr;
1831 
1832 	switch (ioctl) {
1833 	case KVM_CREATE_IRQCHIP: {
1834 		int ret;
1835 		if (!vgic_present)
1836 			return -ENXIO;
1837 		mutex_lock(&kvm->lock);
1838 		ret = kvm_vgic_create(kvm, KVM_DEV_TYPE_ARM_VGIC_V2);
1839 		mutex_unlock(&kvm->lock);
1840 		return ret;
1841 	}
1842 	case KVM_ARM_SET_DEVICE_ADDR: {
1843 		struct kvm_arm_device_addr dev_addr;
1844 
1845 		if (copy_from_user(&dev_addr, argp, sizeof(dev_addr)))
1846 			return -EFAULT;
1847 		return kvm_vm_ioctl_set_device_addr(kvm, &dev_addr);
1848 	}
1849 	case KVM_ARM_PREFERRED_TARGET: {
1850 		struct kvm_vcpu_init init = {
1851 			.target = KVM_ARM_TARGET_GENERIC_V8,
1852 		};
1853 
1854 		if (copy_to_user(argp, &init, sizeof(init)))
1855 			return -EFAULT;
1856 
1857 		return 0;
1858 	}
1859 	case KVM_ARM_MTE_COPY_TAGS: {
1860 		struct kvm_arm_copy_mte_tags copy_tags;
1861 
1862 		if (copy_from_user(&copy_tags, argp, sizeof(copy_tags)))
1863 			return -EFAULT;
1864 		return kvm_vm_ioctl_mte_copy_tags(kvm, &copy_tags);
1865 	}
1866 	case KVM_ARM_SET_COUNTER_OFFSET: {
1867 		struct kvm_arm_counter_offset offset;
1868 
1869 		if (copy_from_user(&offset, argp, sizeof(offset)))
1870 			return -EFAULT;
1871 		return kvm_vm_ioctl_set_counter_offset(kvm, &offset);
1872 	}
1873 	case KVM_HAS_DEVICE_ATTR: {
1874 		if (copy_from_user(&attr, argp, sizeof(attr)))
1875 			return -EFAULT;
1876 
1877 		return kvm_vm_has_attr(kvm, &attr);
1878 	}
1879 	case KVM_SET_DEVICE_ATTR: {
1880 		if (copy_from_user(&attr, argp, sizeof(attr)))
1881 			return -EFAULT;
1882 
1883 		return kvm_vm_set_attr(kvm, &attr);
1884 	}
1885 	case KVM_ARM_GET_REG_WRITABLE_MASKS: {
1886 		struct reg_mask_range range;
1887 
1888 		if (copy_from_user(&range, argp, sizeof(range)))
1889 			return -EFAULT;
1890 		return kvm_vm_ioctl_get_reg_writable_masks(kvm, &range);
1891 	}
1892 	default:
1893 		return -EINVAL;
1894 	}
1895 }
1896 
1897 /* unlocks vcpus from @vcpu_lock_idx and smaller */
1898 static void unlock_vcpus(struct kvm *kvm, int vcpu_lock_idx)
1899 {
1900 	struct kvm_vcpu *tmp_vcpu;
1901 
1902 	for (; vcpu_lock_idx >= 0; vcpu_lock_idx--) {
1903 		tmp_vcpu = kvm_get_vcpu(kvm, vcpu_lock_idx);
1904 		mutex_unlock(&tmp_vcpu->mutex);
1905 	}
1906 }
1907 
1908 void unlock_all_vcpus(struct kvm *kvm)
1909 {
1910 	lockdep_assert_held(&kvm->lock);
1911 
1912 	unlock_vcpus(kvm, atomic_read(&kvm->online_vcpus) - 1);
1913 }
1914 
1915 /* Returns true if all vcpus were locked, false otherwise */
1916 bool lock_all_vcpus(struct kvm *kvm)
1917 {
1918 	struct kvm_vcpu *tmp_vcpu;
1919 	unsigned long c;
1920 
1921 	lockdep_assert_held(&kvm->lock);
1922 
1923 	/*
1924 	 * Any time a vcpu is in an ioctl (including running), the
1925 	 * core KVM code tries to grab the vcpu->mutex.
1926 	 *
1927 	 * By grabbing the vcpu->mutex of all VCPUs we ensure that no
1928 	 * other VCPUs can fiddle with the state while we access it.
1929 	 */
1930 	kvm_for_each_vcpu(c, tmp_vcpu, kvm) {
1931 		if (!mutex_trylock(&tmp_vcpu->mutex)) {
1932 			unlock_vcpus(kvm, c - 1);
1933 			return false;
1934 		}
1935 	}
1936 
1937 	return true;
1938 }
1939 
1940 static unsigned long nvhe_percpu_size(void)
1941 {
1942 	return (unsigned long)CHOOSE_NVHE_SYM(__per_cpu_end) -
1943 		(unsigned long)CHOOSE_NVHE_SYM(__per_cpu_start);
1944 }
1945 
1946 static unsigned long nvhe_percpu_order(void)
1947 {
1948 	unsigned long size = nvhe_percpu_size();
1949 
1950 	return size ? get_order(size) : 0;
1951 }
1952 
1953 static size_t pkvm_host_sve_state_order(void)
1954 {
1955 	return get_order(pkvm_host_sve_state_size());
1956 }
1957 
1958 /* A lookup table holding the hypervisor VA for each vector slot */
1959 static void *hyp_spectre_vector_selector[BP_HARDEN_EL2_SLOTS];
1960 
1961 static void kvm_init_vector_slot(void *base, enum arm64_hyp_spectre_vector slot)
1962 {
1963 	hyp_spectre_vector_selector[slot] = __kvm_vector_slot2addr(base, slot);
1964 }
1965 
1966 static int kvm_init_vector_slots(void)
1967 {
1968 	int err;
1969 	void *base;
1970 
1971 	base = kern_hyp_va(kvm_ksym_ref(__kvm_hyp_vector));
1972 	kvm_init_vector_slot(base, HYP_VECTOR_DIRECT);
1973 
1974 	base = kern_hyp_va(kvm_ksym_ref(__bp_harden_hyp_vecs));
1975 	kvm_init_vector_slot(base, HYP_VECTOR_SPECTRE_DIRECT);
1976 
1977 	if (kvm_system_needs_idmapped_vectors() &&
1978 	    !is_protected_kvm_enabled()) {
1979 		err = create_hyp_exec_mappings(__pa_symbol(__bp_harden_hyp_vecs),
1980 					       __BP_HARDEN_HYP_VECS_SZ, &base);
1981 		if (err)
1982 			return err;
1983 	}
1984 
1985 	kvm_init_vector_slot(base, HYP_VECTOR_INDIRECT);
1986 	kvm_init_vector_slot(base, HYP_VECTOR_SPECTRE_INDIRECT);
1987 	return 0;
1988 }
1989 
1990 static void __init cpu_prepare_hyp_mode(int cpu, u32 hyp_va_bits)
1991 {
1992 	struct kvm_nvhe_init_params *params = per_cpu_ptr_nvhe_sym(kvm_init_params, cpu);
1993 	u64 mmfr0 = read_sanitised_ftr_reg(SYS_ID_AA64MMFR0_EL1);
1994 	unsigned long tcr;
1995 
1996 	/*
1997 	 * Calculate the raw per-cpu offset without a translation from the
1998 	 * kernel's mapping to the linear mapping, and store it in tpidr_el2
1999 	 * so that we can use adr_l to access per-cpu variables in EL2.
2000 	 * Also drop the KASAN tag which gets in the way...
2001 	 */
2002 	params->tpidr_el2 = (unsigned long)kasan_reset_tag(per_cpu_ptr_nvhe_sym(__per_cpu_start, cpu)) -
2003 			    (unsigned long)kvm_ksym_ref(CHOOSE_NVHE_SYM(__per_cpu_start));
2004 
2005 	params->mair_el2 = read_sysreg(mair_el1);
2006 
2007 	tcr = read_sysreg(tcr_el1);
2008 	if (cpus_have_final_cap(ARM64_KVM_HVHE)) {
2009 		tcr |= TCR_EPD1_MASK;
2010 	} else {
2011 		tcr &= TCR_EL2_MASK;
2012 		tcr |= TCR_EL2_RES1;
2013 	}
2014 	tcr &= ~TCR_T0SZ_MASK;
2015 	tcr |= TCR_T0SZ(hyp_va_bits);
2016 	tcr &= ~TCR_EL2_PS_MASK;
2017 	tcr |= FIELD_PREP(TCR_EL2_PS_MASK, kvm_get_parange(mmfr0));
2018 	if (kvm_lpa2_is_enabled())
2019 		tcr |= TCR_EL2_DS;
2020 	params->tcr_el2 = tcr;
2021 
2022 	params->pgd_pa = kvm_mmu_get_httbr();
2023 	if (is_protected_kvm_enabled())
2024 		params->hcr_el2 = HCR_HOST_NVHE_PROTECTED_FLAGS;
2025 	else
2026 		params->hcr_el2 = HCR_HOST_NVHE_FLAGS;
2027 	if (cpus_have_final_cap(ARM64_KVM_HVHE))
2028 		params->hcr_el2 |= HCR_E2H;
2029 	params->vttbr = params->vtcr = 0;
2030 
2031 	/*
2032 	 * Flush the init params from the data cache because the struct will
2033 	 * be read while the MMU is off.
2034 	 */
2035 	kvm_flush_dcache_to_poc(params, sizeof(*params));
2036 }
2037 
2038 static void hyp_install_host_vector(void)
2039 {
2040 	struct kvm_nvhe_init_params *params;
2041 	struct arm_smccc_res res;
2042 
2043 	/* Switch from the HYP stub to our own HYP init vector */
2044 	__hyp_set_vectors(kvm_get_idmap_vector());
2045 
2046 	/*
2047 	 * Call initialization code, and switch to the full blown HYP code.
2048 	 * If the cpucaps haven't been finalized yet, something has gone very
2049 	 * wrong, and hyp will crash and burn when it uses any
2050 	 * cpus_have_*_cap() wrapper.
2051 	 */
2052 	BUG_ON(!system_capabilities_finalized());
2053 	params = this_cpu_ptr_nvhe_sym(kvm_init_params);
2054 	arm_smccc_1_1_hvc(KVM_HOST_SMCCC_FUNC(__kvm_hyp_init), virt_to_phys(params), &res);
2055 	WARN_ON(res.a0 != SMCCC_RET_SUCCESS);
2056 }
2057 
2058 static void cpu_init_hyp_mode(void)
2059 {
2060 	hyp_install_host_vector();
2061 
2062 	/*
2063 	 * Disabling SSBD on a non-VHE system requires us to enable SSBS
2064 	 * at EL2.
2065 	 */
2066 	if (this_cpu_has_cap(ARM64_SSBS) &&
2067 	    arm64_get_spectre_v4_state() == SPECTRE_VULNERABLE) {
2068 		kvm_call_hyp_nvhe(__kvm_enable_ssbs);
2069 	}
2070 }
2071 
2072 static void cpu_hyp_reset(void)
2073 {
2074 	if (!is_kernel_in_hyp_mode())
2075 		__hyp_reset_vectors();
2076 }
2077 
2078 /*
2079  * EL2 vectors can be mapped and rerouted in a number of ways,
2080  * depending on the kernel configuration and CPU present:
2081  *
2082  * - If the CPU is affected by Spectre-v2, the hardening sequence is
2083  *   placed in one of the vector slots, which is executed before jumping
2084  *   to the real vectors.
2085  *
2086  * - If the CPU also has the ARM64_SPECTRE_V3A cap, the slot
2087  *   containing the hardening sequence is mapped next to the idmap page,
2088  *   and executed before jumping to the real vectors.
2089  *
2090  * - If the CPU only has the ARM64_SPECTRE_V3A cap, then an
2091  *   empty slot is selected, mapped next to the idmap page, and
2092  *   executed before jumping to the real vectors.
2093  *
2094  * Note that ARM64_SPECTRE_V3A is somewhat incompatible with
2095  * VHE, as we don't have hypervisor-specific mappings. If the system
2096  * is VHE and yet selects this capability, it will be ignored.
2097  */
2098 static void cpu_set_hyp_vector(void)
2099 {
2100 	struct bp_hardening_data *data = this_cpu_ptr(&bp_hardening_data);
2101 	void *vector = hyp_spectre_vector_selector[data->slot];
2102 
2103 	if (!is_protected_kvm_enabled())
2104 		*this_cpu_ptr_hyp_sym(kvm_hyp_vector) = (unsigned long)vector;
2105 	else
2106 		kvm_call_hyp_nvhe(__pkvm_cpu_set_vector, data->slot);
2107 }
2108 
2109 static void cpu_hyp_init_context(void)
2110 {
2111 	kvm_init_host_cpu_context(host_data_ptr(host_ctxt));
2112 
2113 	if (!is_kernel_in_hyp_mode())
2114 		cpu_init_hyp_mode();
2115 }
2116 
2117 static void cpu_hyp_init_features(void)
2118 {
2119 	cpu_set_hyp_vector();
2120 	kvm_arm_init_debug();
2121 
2122 	if (is_kernel_in_hyp_mode())
2123 		kvm_timer_init_vhe();
2124 
2125 	if (vgic_present)
2126 		kvm_vgic_init_cpu_hardware();
2127 }
2128 
2129 static void cpu_hyp_reinit(void)
2130 {
2131 	cpu_hyp_reset();
2132 	cpu_hyp_init_context();
2133 	cpu_hyp_init_features();
2134 }
2135 
2136 static void cpu_hyp_init(void *discard)
2137 {
2138 	if (!__this_cpu_read(kvm_hyp_initialized)) {
2139 		cpu_hyp_reinit();
2140 		__this_cpu_write(kvm_hyp_initialized, 1);
2141 	}
2142 }
2143 
2144 static void cpu_hyp_uninit(void *discard)
2145 {
2146 	if (__this_cpu_read(kvm_hyp_initialized)) {
2147 		cpu_hyp_reset();
2148 		__this_cpu_write(kvm_hyp_initialized, 0);
2149 	}
2150 }
2151 
2152 int kvm_arch_enable_virtualization_cpu(void)
2153 {
2154 	/*
2155 	 * Most calls to this function are made with migration
2156 	 * disabled, but not with preemption disabled. The former is
2157 	 * enough to ensure correctness, but most of the helpers
2158 	 * expect the later and will throw a tantrum otherwise.
2159 	 */
2160 	preempt_disable();
2161 
2162 	cpu_hyp_init(NULL);
2163 
2164 	kvm_vgic_cpu_up();
2165 	kvm_timer_cpu_up();
2166 
2167 	preempt_enable();
2168 
2169 	return 0;
2170 }
2171 
2172 void kvm_arch_disable_virtualization_cpu(void)
2173 {
2174 	kvm_timer_cpu_down();
2175 	kvm_vgic_cpu_down();
2176 
2177 	if (!is_protected_kvm_enabled())
2178 		cpu_hyp_uninit(NULL);
2179 }
2180 
2181 #ifdef CONFIG_CPU_PM
2182 static int hyp_init_cpu_pm_notifier(struct notifier_block *self,
2183 				    unsigned long cmd,
2184 				    void *v)
2185 {
2186 	/*
2187 	 * kvm_hyp_initialized is left with its old value over
2188 	 * PM_ENTER->PM_EXIT. It is used to indicate PM_EXIT should
2189 	 * re-enable hyp.
2190 	 */
2191 	switch (cmd) {
2192 	case CPU_PM_ENTER:
2193 		if (__this_cpu_read(kvm_hyp_initialized))
2194 			/*
2195 			 * don't update kvm_hyp_initialized here
2196 			 * so that the hyp will be re-enabled
2197 			 * when we resume. See below.
2198 			 */
2199 			cpu_hyp_reset();
2200 
2201 		return NOTIFY_OK;
2202 	case CPU_PM_ENTER_FAILED:
2203 	case CPU_PM_EXIT:
2204 		if (__this_cpu_read(kvm_hyp_initialized))
2205 			/* The hyp was enabled before suspend. */
2206 			cpu_hyp_reinit();
2207 
2208 		return NOTIFY_OK;
2209 
2210 	default:
2211 		return NOTIFY_DONE;
2212 	}
2213 }
2214 
2215 static struct notifier_block hyp_init_cpu_pm_nb = {
2216 	.notifier_call = hyp_init_cpu_pm_notifier,
2217 };
2218 
2219 static void __init hyp_cpu_pm_init(void)
2220 {
2221 	if (!is_protected_kvm_enabled())
2222 		cpu_pm_register_notifier(&hyp_init_cpu_pm_nb);
2223 }
2224 static void __init hyp_cpu_pm_exit(void)
2225 {
2226 	if (!is_protected_kvm_enabled())
2227 		cpu_pm_unregister_notifier(&hyp_init_cpu_pm_nb);
2228 }
2229 #else
2230 static inline void __init hyp_cpu_pm_init(void)
2231 {
2232 }
2233 static inline void __init hyp_cpu_pm_exit(void)
2234 {
2235 }
2236 #endif
2237 
2238 static void __init init_cpu_logical_map(void)
2239 {
2240 	unsigned int cpu;
2241 
2242 	/*
2243 	 * Copy the MPIDR <-> logical CPU ID mapping to hyp.
2244 	 * Only copy the set of online CPUs whose features have been checked
2245 	 * against the finalized system capabilities. The hypervisor will not
2246 	 * allow any other CPUs from the `possible` set to boot.
2247 	 */
2248 	for_each_online_cpu(cpu)
2249 		hyp_cpu_logical_map[cpu] = cpu_logical_map(cpu);
2250 }
2251 
2252 #define init_psci_0_1_impl_state(config, what)	\
2253 	config.psci_0_1_ ## what ## _implemented = psci_ops.what
2254 
2255 static bool __init init_psci_relay(void)
2256 {
2257 	/*
2258 	 * If PSCI has not been initialized, protected KVM cannot install
2259 	 * itself on newly booted CPUs.
2260 	 */
2261 	if (!psci_ops.get_version) {
2262 		kvm_err("Cannot initialize protected mode without PSCI\n");
2263 		return false;
2264 	}
2265 
2266 	kvm_host_psci_config.version = psci_ops.get_version();
2267 	kvm_host_psci_config.smccc_version = arm_smccc_get_version();
2268 
2269 	if (kvm_host_psci_config.version == PSCI_VERSION(0, 1)) {
2270 		kvm_host_psci_config.function_ids_0_1 = get_psci_0_1_function_ids();
2271 		init_psci_0_1_impl_state(kvm_host_psci_config, cpu_suspend);
2272 		init_psci_0_1_impl_state(kvm_host_psci_config, cpu_on);
2273 		init_psci_0_1_impl_state(kvm_host_psci_config, cpu_off);
2274 		init_psci_0_1_impl_state(kvm_host_psci_config, migrate);
2275 	}
2276 	return true;
2277 }
2278 
2279 static int __init init_subsystems(void)
2280 {
2281 	int err = 0;
2282 
2283 	/*
2284 	 * Enable hardware so that subsystem initialisation can access EL2.
2285 	 */
2286 	on_each_cpu(cpu_hyp_init, NULL, 1);
2287 
2288 	/*
2289 	 * Register CPU lower-power notifier
2290 	 */
2291 	hyp_cpu_pm_init();
2292 
2293 	/*
2294 	 * Init HYP view of VGIC
2295 	 */
2296 	err = kvm_vgic_hyp_init();
2297 	switch (err) {
2298 	case 0:
2299 		vgic_present = true;
2300 		break;
2301 	case -ENODEV:
2302 	case -ENXIO:
2303 		vgic_present = false;
2304 		err = 0;
2305 		break;
2306 	default:
2307 		goto out;
2308 	}
2309 
2310 	/*
2311 	 * Init HYP architected timer support
2312 	 */
2313 	err = kvm_timer_hyp_init(vgic_present);
2314 	if (err)
2315 		goto out;
2316 
2317 	kvm_register_perf_callbacks(NULL);
2318 
2319 out:
2320 	if (err)
2321 		hyp_cpu_pm_exit();
2322 
2323 	if (err || !is_protected_kvm_enabled())
2324 		on_each_cpu(cpu_hyp_uninit, NULL, 1);
2325 
2326 	return err;
2327 }
2328 
2329 static void __init teardown_subsystems(void)
2330 {
2331 	kvm_unregister_perf_callbacks();
2332 	hyp_cpu_pm_exit();
2333 }
2334 
2335 static void __init teardown_hyp_mode(void)
2336 {
2337 	bool free_sve = system_supports_sve() && is_protected_kvm_enabled();
2338 	int cpu;
2339 
2340 	free_hyp_pgds();
2341 	for_each_possible_cpu(cpu) {
2342 		free_page(per_cpu(kvm_arm_hyp_stack_page, cpu));
2343 		free_pages(kvm_nvhe_sym(kvm_arm_hyp_percpu_base)[cpu], nvhe_percpu_order());
2344 
2345 		if (free_sve) {
2346 			struct cpu_sve_state *sve_state;
2347 
2348 			sve_state = per_cpu_ptr_nvhe_sym(kvm_host_data, cpu)->sve_state;
2349 			free_pages((unsigned long) sve_state, pkvm_host_sve_state_order());
2350 		}
2351 	}
2352 }
2353 
2354 static int __init do_pkvm_init(u32 hyp_va_bits)
2355 {
2356 	void *per_cpu_base = kvm_ksym_ref(kvm_nvhe_sym(kvm_arm_hyp_percpu_base));
2357 	int ret;
2358 
2359 	preempt_disable();
2360 	cpu_hyp_init_context();
2361 	ret = kvm_call_hyp_nvhe(__pkvm_init, hyp_mem_base, hyp_mem_size,
2362 				num_possible_cpus(), kern_hyp_va(per_cpu_base),
2363 				hyp_va_bits);
2364 	cpu_hyp_init_features();
2365 
2366 	/*
2367 	 * The stub hypercalls are now disabled, so set our local flag to
2368 	 * prevent a later re-init attempt in kvm_arch_enable_virtualization_cpu().
2369 	 */
2370 	__this_cpu_write(kvm_hyp_initialized, 1);
2371 	preempt_enable();
2372 
2373 	return ret;
2374 }
2375 
2376 static u64 get_hyp_id_aa64pfr0_el1(void)
2377 {
2378 	/*
2379 	 * Track whether the system isn't affected by spectre/meltdown in the
2380 	 * hypervisor's view of id_aa64pfr0_el1, used for protected VMs.
2381 	 * Although this is per-CPU, we make it global for simplicity, e.g., not
2382 	 * to have to worry about vcpu migration.
2383 	 *
2384 	 * Unlike for non-protected VMs, userspace cannot override this for
2385 	 * protected VMs.
2386 	 */
2387 	u64 val = read_sanitised_ftr_reg(SYS_ID_AA64PFR0_EL1);
2388 
2389 	val &= ~(ARM64_FEATURE_MASK(ID_AA64PFR0_EL1_CSV2) |
2390 		 ARM64_FEATURE_MASK(ID_AA64PFR0_EL1_CSV3));
2391 
2392 	val |= FIELD_PREP(ARM64_FEATURE_MASK(ID_AA64PFR0_EL1_CSV2),
2393 			  arm64_get_spectre_v2_state() == SPECTRE_UNAFFECTED);
2394 	val |= FIELD_PREP(ARM64_FEATURE_MASK(ID_AA64PFR0_EL1_CSV3),
2395 			  arm64_get_meltdown_state() == SPECTRE_UNAFFECTED);
2396 
2397 	return val;
2398 }
2399 
2400 static void kvm_hyp_init_symbols(void)
2401 {
2402 	kvm_nvhe_sym(id_aa64pfr0_el1_sys_val) = get_hyp_id_aa64pfr0_el1();
2403 	kvm_nvhe_sym(id_aa64pfr1_el1_sys_val) = read_sanitised_ftr_reg(SYS_ID_AA64PFR1_EL1);
2404 	kvm_nvhe_sym(id_aa64isar0_el1_sys_val) = read_sanitised_ftr_reg(SYS_ID_AA64ISAR0_EL1);
2405 	kvm_nvhe_sym(id_aa64isar1_el1_sys_val) = read_sanitised_ftr_reg(SYS_ID_AA64ISAR1_EL1);
2406 	kvm_nvhe_sym(id_aa64isar2_el1_sys_val) = read_sanitised_ftr_reg(SYS_ID_AA64ISAR2_EL1);
2407 	kvm_nvhe_sym(id_aa64mmfr0_el1_sys_val) = read_sanitised_ftr_reg(SYS_ID_AA64MMFR0_EL1);
2408 	kvm_nvhe_sym(id_aa64mmfr1_el1_sys_val) = read_sanitised_ftr_reg(SYS_ID_AA64MMFR1_EL1);
2409 	kvm_nvhe_sym(id_aa64mmfr2_el1_sys_val) = read_sanitised_ftr_reg(SYS_ID_AA64MMFR2_EL1);
2410 	kvm_nvhe_sym(id_aa64smfr0_el1_sys_val) = read_sanitised_ftr_reg(SYS_ID_AA64SMFR0_EL1);
2411 	kvm_nvhe_sym(__icache_flags) = __icache_flags;
2412 	kvm_nvhe_sym(kvm_arm_vmid_bits) = kvm_arm_vmid_bits;
2413 }
2414 
2415 static int __init kvm_hyp_init_protection(u32 hyp_va_bits)
2416 {
2417 	void *addr = phys_to_virt(hyp_mem_base);
2418 	int ret;
2419 
2420 	ret = create_hyp_mappings(addr, addr + hyp_mem_size, PAGE_HYP);
2421 	if (ret)
2422 		return ret;
2423 
2424 	ret = do_pkvm_init(hyp_va_bits);
2425 	if (ret)
2426 		return ret;
2427 
2428 	free_hyp_pgds();
2429 
2430 	return 0;
2431 }
2432 
2433 static int init_pkvm_host_sve_state(void)
2434 {
2435 	int cpu;
2436 
2437 	if (!system_supports_sve())
2438 		return 0;
2439 
2440 	/* Allocate pages for host sve state in protected mode. */
2441 	for_each_possible_cpu(cpu) {
2442 		struct page *page = alloc_pages(GFP_KERNEL, pkvm_host_sve_state_order());
2443 
2444 		if (!page)
2445 			return -ENOMEM;
2446 
2447 		per_cpu_ptr_nvhe_sym(kvm_host_data, cpu)->sve_state = page_address(page);
2448 	}
2449 
2450 	/*
2451 	 * Don't map the pages in hyp since these are only used in protected
2452 	 * mode, which will (re)create its own mapping when initialized.
2453 	 */
2454 
2455 	return 0;
2456 }
2457 
2458 /*
2459  * Finalizes the initialization of hyp mode, once everything else is initialized
2460  * and the initialziation process cannot fail.
2461  */
2462 static void finalize_init_hyp_mode(void)
2463 {
2464 	int cpu;
2465 
2466 	if (system_supports_sve() && is_protected_kvm_enabled()) {
2467 		for_each_possible_cpu(cpu) {
2468 			struct cpu_sve_state *sve_state;
2469 
2470 			sve_state = per_cpu_ptr_nvhe_sym(kvm_host_data, cpu)->sve_state;
2471 			per_cpu_ptr_nvhe_sym(kvm_host_data, cpu)->sve_state =
2472 				kern_hyp_va(sve_state);
2473 		}
2474 	} else {
2475 		for_each_possible_cpu(cpu) {
2476 			struct user_fpsimd_state *fpsimd_state;
2477 
2478 			fpsimd_state = &per_cpu_ptr_nvhe_sym(kvm_host_data, cpu)->host_ctxt.fp_regs;
2479 			per_cpu_ptr_nvhe_sym(kvm_host_data, cpu)->fpsimd_state =
2480 				kern_hyp_va(fpsimd_state);
2481 		}
2482 	}
2483 }
2484 
2485 static void pkvm_hyp_init_ptrauth(void)
2486 {
2487 	struct kvm_cpu_context *hyp_ctxt;
2488 	int cpu;
2489 
2490 	for_each_possible_cpu(cpu) {
2491 		hyp_ctxt = per_cpu_ptr_nvhe_sym(kvm_hyp_ctxt, cpu);
2492 		hyp_ctxt->sys_regs[APIAKEYLO_EL1] = get_random_long();
2493 		hyp_ctxt->sys_regs[APIAKEYHI_EL1] = get_random_long();
2494 		hyp_ctxt->sys_regs[APIBKEYLO_EL1] = get_random_long();
2495 		hyp_ctxt->sys_regs[APIBKEYHI_EL1] = get_random_long();
2496 		hyp_ctxt->sys_regs[APDAKEYLO_EL1] = get_random_long();
2497 		hyp_ctxt->sys_regs[APDAKEYHI_EL1] = get_random_long();
2498 		hyp_ctxt->sys_regs[APDBKEYLO_EL1] = get_random_long();
2499 		hyp_ctxt->sys_regs[APDBKEYHI_EL1] = get_random_long();
2500 		hyp_ctxt->sys_regs[APGAKEYLO_EL1] = get_random_long();
2501 		hyp_ctxt->sys_regs[APGAKEYHI_EL1] = get_random_long();
2502 	}
2503 }
2504 
2505 /* Inits Hyp-mode on all online CPUs */
2506 static int __init init_hyp_mode(void)
2507 {
2508 	u32 hyp_va_bits;
2509 	int cpu;
2510 	int err = -ENOMEM;
2511 
2512 	/*
2513 	 * The protected Hyp-mode cannot be initialized if the memory pool
2514 	 * allocation has failed.
2515 	 */
2516 	if (is_protected_kvm_enabled() && !hyp_mem_base)
2517 		goto out_err;
2518 
2519 	/*
2520 	 * Allocate Hyp PGD and setup Hyp identity mapping
2521 	 */
2522 	err = kvm_mmu_init(&hyp_va_bits);
2523 	if (err)
2524 		goto out_err;
2525 
2526 	/*
2527 	 * Allocate stack pages for Hypervisor-mode
2528 	 */
2529 	for_each_possible_cpu(cpu) {
2530 		unsigned long stack_page;
2531 
2532 		stack_page = __get_free_page(GFP_KERNEL);
2533 		if (!stack_page) {
2534 			err = -ENOMEM;
2535 			goto out_err;
2536 		}
2537 
2538 		per_cpu(kvm_arm_hyp_stack_page, cpu) = stack_page;
2539 	}
2540 
2541 	/*
2542 	 * Allocate and initialize pages for Hypervisor-mode percpu regions.
2543 	 */
2544 	for_each_possible_cpu(cpu) {
2545 		struct page *page;
2546 		void *page_addr;
2547 
2548 		page = alloc_pages(GFP_KERNEL, nvhe_percpu_order());
2549 		if (!page) {
2550 			err = -ENOMEM;
2551 			goto out_err;
2552 		}
2553 
2554 		page_addr = page_address(page);
2555 		memcpy(page_addr, CHOOSE_NVHE_SYM(__per_cpu_start), nvhe_percpu_size());
2556 		kvm_nvhe_sym(kvm_arm_hyp_percpu_base)[cpu] = (unsigned long)page_addr;
2557 	}
2558 
2559 	/*
2560 	 * Map the Hyp-code called directly from the host
2561 	 */
2562 	err = create_hyp_mappings(kvm_ksym_ref(__hyp_text_start),
2563 				  kvm_ksym_ref(__hyp_text_end), PAGE_HYP_EXEC);
2564 	if (err) {
2565 		kvm_err("Cannot map world-switch code\n");
2566 		goto out_err;
2567 	}
2568 
2569 	err = create_hyp_mappings(kvm_ksym_ref(__hyp_rodata_start),
2570 				  kvm_ksym_ref(__hyp_rodata_end), PAGE_HYP_RO);
2571 	if (err) {
2572 		kvm_err("Cannot map .hyp.rodata section\n");
2573 		goto out_err;
2574 	}
2575 
2576 	err = create_hyp_mappings(kvm_ksym_ref(__start_rodata),
2577 				  kvm_ksym_ref(__end_rodata), PAGE_HYP_RO);
2578 	if (err) {
2579 		kvm_err("Cannot map rodata section\n");
2580 		goto out_err;
2581 	}
2582 
2583 	/*
2584 	 * .hyp.bss is guaranteed to be placed at the beginning of the .bss
2585 	 * section thanks to an assertion in the linker script. Map it RW and
2586 	 * the rest of .bss RO.
2587 	 */
2588 	err = create_hyp_mappings(kvm_ksym_ref(__hyp_bss_start),
2589 				  kvm_ksym_ref(__hyp_bss_end), PAGE_HYP);
2590 	if (err) {
2591 		kvm_err("Cannot map hyp bss section: %d\n", err);
2592 		goto out_err;
2593 	}
2594 
2595 	err = create_hyp_mappings(kvm_ksym_ref(__hyp_bss_end),
2596 				  kvm_ksym_ref(__bss_stop), PAGE_HYP_RO);
2597 	if (err) {
2598 		kvm_err("Cannot map bss section\n");
2599 		goto out_err;
2600 	}
2601 
2602 	/*
2603 	 * Map the Hyp stack pages
2604 	 */
2605 	for_each_possible_cpu(cpu) {
2606 		struct kvm_nvhe_init_params *params = per_cpu_ptr_nvhe_sym(kvm_init_params, cpu);
2607 		char *stack_page = (char *)per_cpu(kvm_arm_hyp_stack_page, cpu);
2608 
2609 		err = create_hyp_stack(__pa(stack_page), &params->stack_hyp_va);
2610 		if (err) {
2611 			kvm_err("Cannot map hyp stack\n");
2612 			goto out_err;
2613 		}
2614 
2615 		/*
2616 		 * Save the stack PA in nvhe_init_params. This will be needed
2617 		 * to recreate the stack mapping in protected nVHE mode.
2618 		 * __hyp_pa() won't do the right thing there, since the stack
2619 		 * has been mapped in the flexible private VA space.
2620 		 */
2621 		params->stack_pa = __pa(stack_page);
2622 	}
2623 
2624 	for_each_possible_cpu(cpu) {
2625 		char *percpu_begin = (char *)kvm_nvhe_sym(kvm_arm_hyp_percpu_base)[cpu];
2626 		char *percpu_end = percpu_begin + nvhe_percpu_size();
2627 
2628 		/* Map Hyp percpu pages */
2629 		err = create_hyp_mappings(percpu_begin, percpu_end, PAGE_HYP);
2630 		if (err) {
2631 			kvm_err("Cannot map hyp percpu region\n");
2632 			goto out_err;
2633 		}
2634 
2635 		/* Prepare the CPU initialization parameters */
2636 		cpu_prepare_hyp_mode(cpu, hyp_va_bits);
2637 	}
2638 
2639 	kvm_hyp_init_symbols();
2640 
2641 	if (is_protected_kvm_enabled()) {
2642 		if (IS_ENABLED(CONFIG_ARM64_PTR_AUTH_KERNEL) &&
2643 		    cpus_have_final_cap(ARM64_HAS_ADDRESS_AUTH))
2644 			pkvm_hyp_init_ptrauth();
2645 
2646 		init_cpu_logical_map();
2647 
2648 		if (!init_psci_relay()) {
2649 			err = -ENODEV;
2650 			goto out_err;
2651 		}
2652 
2653 		err = init_pkvm_host_sve_state();
2654 		if (err)
2655 			goto out_err;
2656 
2657 		err = kvm_hyp_init_protection(hyp_va_bits);
2658 		if (err) {
2659 			kvm_err("Failed to init hyp memory protection\n");
2660 			goto out_err;
2661 		}
2662 	}
2663 
2664 	return 0;
2665 
2666 out_err:
2667 	teardown_hyp_mode();
2668 	kvm_err("error initializing Hyp mode: %d\n", err);
2669 	return err;
2670 }
2671 
2672 struct kvm_vcpu *kvm_mpidr_to_vcpu(struct kvm *kvm, unsigned long mpidr)
2673 {
2674 	struct kvm_vcpu *vcpu = NULL;
2675 	struct kvm_mpidr_data *data;
2676 	unsigned long i;
2677 
2678 	mpidr &= MPIDR_HWID_BITMASK;
2679 
2680 	rcu_read_lock();
2681 	data = rcu_dereference(kvm->arch.mpidr_data);
2682 
2683 	if (data) {
2684 		u16 idx = kvm_mpidr_index(data, mpidr);
2685 
2686 		vcpu = kvm_get_vcpu(kvm, data->cmpidr_to_idx[idx]);
2687 		if (mpidr != kvm_vcpu_get_mpidr_aff(vcpu))
2688 			vcpu = NULL;
2689 	}
2690 
2691 	rcu_read_unlock();
2692 
2693 	if (vcpu)
2694 		return vcpu;
2695 
2696 	kvm_for_each_vcpu(i, vcpu, kvm) {
2697 		if (mpidr == kvm_vcpu_get_mpidr_aff(vcpu))
2698 			return vcpu;
2699 	}
2700 	return NULL;
2701 }
2702 
2703 bool kvm_arch_irqchip_in_kernel(struct kvm *kvm)
2704 {
2705 	return irqchip_in_kernel(kvm);
2706 }
2707 
2708 bool kvm_arch_has_irq_bypass(void)
2709 {
2710 	return true;
2711 }
2712 
2713 int kvm_arch_irq_bypass_add_producer(struct irq_bypass_consumer *cons,
2714 				      struct irq_bypass_producer *prod)
2715 {
2716 	struct kvm_kernel_irqfd *irqfd =
2717 		container_of(cons, struct kvm_kernel_irqfd, consumer);
2718 
2719 	return kvm_vgic_v4_set_forwarding(irqfd->kvm, prod->irq,
2720 					  &irqfd->irq_entry);
2721 }
2722 void kvm_arch_irq_bypass_del_producer(struct irq_bypass_consumer *cons,
2723 				      struct irq_bypass_producer *prod)
2724 {
2725 	struct kvm_kernel_irqfd *irqfd =
2726 		container_of(cons, struct kvm_kernel_irqfd, consumer);
2727 
2728 	kvm_vgic_v4_unset_forwarding(irqfd->kvm, prod->irq,
2729 				     &irqfd->irq_entry);
2730 }
2731 
2732 void kvm_arch_irq_bypass_stop(struct irq_bypass_consumer *cons)
2733 {
2734 	struct kvm_kernel_irqfd *irqfd =
2735 		container_of(cons, struct kvm_kernel_irqfd, consumer);
2736 
2737 	kvm_arm_halt_guest(irqfd->kvm);
2738 }
2739 
2740 void kvm_arch_irq_bypass_start(struct irq_bypass_consumer *cons)
2741 {
2742 	struct kvm_kernel_irqfd *irqfd =
2743 		container_of(cons, struct kvm_kernel_irqfd, consumer);
2744 
2745 	kvm_arm_resume_guest(irqfd->kvm);
2746 }
2747 
2748 /* Initialize Hyp-mode and memory mappings on all CPUs */
2749 static __init int kvm_arm_init(void)
2750 {
2751 	int err;
2752 	bool in_hyp_mode;
2753 
2754 	if (!is_hyp_mode_available()) {
2755 		kvm_info("HYP mode not available\n");
2756 		return -ENODEV;
2757 	}
2758 
2759 	if (kvm_get_mode() == KVM_MODE_NONE) {
2760 		kvm_info("KVM disabled from command line\n");
2761 		return -ENODEV;
2762 	}
2763 
2764 	err = kvm_sys_reg_table_init();
2765 	if (err) {
2766 		kvm_info("Error initializing system register tables");
2767 		return err;
2768 	}
2769 
2770 	in_hyp_mode = is_kernel_in_hyp_mode();
2771 
2772 	if (cpus_have_final_cap(ARM64_WORKAROUND_DEVICE_LOAD_ACQUIRE) ||
2773 	    cpus_have_final_cap(ARM64_WORKAROUND_1508412))
2774 		kvm_info("Guests without required CPU erratum workarounds can deadlock system!\n" \
2775 			 "Only trusted guests should be used on this system.\n");
2776 
2777 	err = kvm_set_ipa_limit();
2778 	if (err)
2779 		return err;
2780 
2781 	err = kvm_arm_init_sve();
2782 	if (err)
2783 		return err;
2784 
2785 	err = kvm_arm_vmid_alloc_init();
2786 	if (err) {
2787 		kvm_err("Failed to initialize VMID allocator.\n");
2788 		return err;
2789 	}
2790 
2791 	if (!in_hyp_mode) {
2792 		err = init_hyp_mode();
2793 		if (err)
2794 			goto out_err;
2795 	}
2796 
2797 	err = kvm_init_vector_slots();
2798 	if (err) {
2799 		kvm_err("Cannot initialise vector slots\n");
2800 		goto out_hyp;
2801 	}
2802 
2803 	err = init_subsystems();
2804 	if (err)
2805 		goto out_hyp;
2806 
2807 	kvm_info("%s%sVHE mode initialized successfully\n",
2808 		 in_hyp_mode ? "" : (is_protected_kvm_enabled() ?
2809 				     "Protected " : "Hyp "),
2810 		 in_hyp_mode ? "" : (cpus_have_final_cap(ARM64_KVM_HVHE) ?
2811 				     "h" : "n"));
2812 
2813 	/*
2814 	 * FIXME: Do something reasonable if kvm_init() fails after pKVM
2815 	 * hypervisor protection is finalized.
2816 	 */
2817 	err = kvm_init(sizeof(struct kvm_vcpu), 0, THIS_MODULE);
2818 	if (err)
2819 		goto out_subs;
2820 
2821 	/*
2822 	 * This should be called after initialization is done and failure isn't
2823 	 * possible anymore.
2824 	 */
2825 	if (!in_hyp_mode)
2826 		finalize_init_hyp_mode();
2827 
2828 	kvm_arm_initialised = true;
2829 
2830 	return 0;
2831 
2832 out_subs:
2833 	teardown_subsystems();
2834 out_hyp:
2835 	if (!in_hyp_mode)
2836 		teardown_hyp_mode();
2837 out_err:
2838 	kvm_arm_vmid_alloc_free();
2839 	return err;
2840 }
2841 
2842 static int __init early_kvm_mode_cfg(char *arg)
2843 {
2844 	if (!arg)
2845 		return -EINVAL;
2846 
2847 	if (strcmp(arg, "none") == 0) {
2848 		kvm_mode = KVM_MODE_NONE;
2849 		return 0;
2850 	}
2851 
2852 	if (!is_hyp_mode_available()) {
2853 		pr_warn_once("KVM is not available. Ignoring kvm-arm.mode\n");
2854 		return 0;
2855 	}
2856 
2857 	if (strcmp(arg, "protected") == 0) {
2858 		if (!is_kernel_in_hyp_mode())
2859 			kvm_mode = KVM_MODE_PROTECTED;
2860 		else
2861 			pr_warn_once("Protected KVM not available with VHE\n");
2862 
2863 		return 0;
2864 	}
2865 
2866 	if (strcmp(arg, "nvhe") == 0 && !WARN_ON(is_kernel_in_hyp_mode())) {
2867 		kvm_mode = KVM_MODE_DEFAULT;
2868 		return 0;
2869 	}
2870 
2871 	if (strcmp(arg, "nested") == 0 && !WARN_ON(!is_kernel_in_hyp_mode())) {
2872 		kvm_mode = KVM_MODE_NV;
2873 		return 0;
2874 	}
2875 
2876 	return -EINVAL;
2877 }
2878 early_param("kvm-arm.mode", early_kvm_mode_cfg);
2879 
2880 static int __init early_kvm_wfx_trap_policy_cfg(char *arg, enum kvm_wfx_trap_policy *p)
2881 {
2882 	if (!arg)
2883 		return -EINVAL;
2884 
2885 	if (strcmp(arg, "trap") == 0) {
2886 		*p = KVM_WFX_TRAP;
2887 		return 0;
2888 	}
2889 
2890 	if (strcmp(arg, "notrap") == 0) {
2891 		*p = KVM_WFX_NOTRAP;
2892 		return 0;
2893 	}
2894 
2895 	return -EINVAL;
2896 }
2897 
2898 static int __init early_kvm_wfi_trap_policy_cfg(char *arg)
2899 {
2900 	return early_kvm_wfx_trap_policy_cfg(arg, &kvm_wfi_trap_policy);
2901 }
2902 early_param("kvm-arm.wfi_trap_policy", early_kvm_wfi_trap_policy_cfg);
2903 
2904 static int __init early_kvm_wfe_trap_policy_cfg(char *arg)
2905 {
2906 	return early_kvm_wfx_trap_policy_cfg(arg, &kvm_wfe_trap_policy);
2907 }
2908 early_param("kvm-arm.wfe_trap_policy", early_kvm_wfe_trap_policy_cfg);
2909 
2910 enum kvm_mode kvm_get_mode(void)
2911 {
2912 	return kvm_mode;
2913 }
2914 
2915 module_init(kvm_arm_init);
2916