xref: /linux/arch/arm64/kvm/arm.c (revision e7cd4b811c9e019f5acbce85699c622b30194c24)
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 DEFINE_STATIC_KEY_FALSE(userspace_irqchip_in_use);
73 
74 bool is_kvm_arm_initialised(void)
75 {
76 	return kvm_arm_initialised;
77 }
78 
79 int kvm_arch_vcpu_should_kick(struct kvm_vcpu *vcpu)
80 {
81 	return kvm_vcpu_exiting_guest_mode(vcpu) == IN_GUEST_MODE;
82 }
83 
84 /*
85  * This functions as an allow-list of protected VM capabilities.
86  * Features not explicitly allowed by this function are denied.
87  */
88 static bool pkvm_ext_allowed(struct kvm *kvm, long ext)
89 {
90 	switch (ext) {
91 	case KVM_CAP_IRQCHIP:
92 	case KVM_CAP_ARM_PSCI:
93 	case KVM_CAP_ARM_PSCI_0_2:
94 	case KVM_CAP_NR_VCPUS:
95 	case KVM_CAP_MAX_VCPUS:
96 	case KVM_CAP_MAX_VCPU_ID:
97 	case KVM_CAP_MSI_DEVID:
98 	case KVM_CAP_ARM_VM_IPA_SIZE:
99 	case KVM_CAP_ARM_PMU_V3:
100 	case KVM_CAP_ARM_SVE:
101 	case KVM_CAP_ARM_PTRAUTH_ADDRESS:
102 	case KVM_CAP_ARM_PTRAUTH_GENERIC:
103 		return true;
104 	default:
105 		return false;
106 	}
107 }
108 
109 int kvm_vm_ioctl_enable_cap(struct kvm *kvm,
110 			    struct kvm_enable_cap *cap)
111 {
112 	int r = -EINVAL;
113 
114 	if (cap->flags)
115 		return -EINVAL;
116 
117 	if (kvm_vm_is_protected(kvm) && !pkvm_ext_allowed(kvm, cap->cap))
118 		return -EINVAL;
119 
120 	switch (cap->cap) {
121 	case KVM_CAP_ARM_NISV_TO_USER:
122 		r = 0;
123 		set_bit(KVM_ARCH_FLAG_RETURN_NISV_IO_ABORT_TO_USER,
124 			&kvm->arch.flags);
125 		break;
126 	case KVM_CAP_ARM_MTE:
127 		mutex_lock(&kvm->lock);
128 		if (system_supports_mte() && !kvm->created_vcpus) {
129 			r = 0;
130 			set_bit(KVM_ARCH_FLAG_MTE_ENABLED, &kvm->arch.flags);
131 		}
132 		mutex_unlock(&kvm->lock);
133 		break;
134 	case KVM_CAP_ARM_SYSTEM_SUSPEND:
135 		r = 0;
136 		set_bit(KVM_ARCH_FLAG_SYSTEM_SUSPEND_ENABLED, &kvm->arch.flags);
137 		break;
138 	case KVM_CAP_ARM_EAGER_SPLIT_CHUNK_SIZE:
139 		mutex_lock(&kvm->slots_lock);
140 		/*
141 		 * To keep things simple, allow changing the chunk
142 		 * size only when no memory slots have been created.
143 		 */
144 		if (kvm_are_all_memslots_empty(kvm)) {
145 			u64 new_cap = cap->args[0];
146 
147 			if (!new_cap || kvm_is_block_size_supported(new_cap)) {
148 				r = 0;
149 				kvm->arch.mmu.split_page_chunk_size = new_cap;
150 			}
151 		}
152 		mutex_unlock(&kvm->slots_lock);
153 		break;
154 	default:
155 		break;
156 	}
157 
158 	return r;
159 }
160 
161 static int kvm_arm_default_max_vcpus(void)
162 {
163 	return vgic_present ? kvm_vgic_get_max_vcpus() : KVM_MAX_VCPUS;
164 }
165 
166 /**
167  * kvm_arch_init_vm - initializes a VM data structure
168  * @kvm:	pointer to the KVM struct
169  * @type:	kvm device type
170  */
171 int kvm_arch_init_vm(struct kvm *kvm, unsigned long type)
172 {
173 	int ret;
174 
175 	mutex_init(&kvm->arch.config_lock);
176 
177 #ifdef CONFIG_LOCKDEP
178 	/* Clue in lockdep that the config_lock must be taken inside kvm->lock */
179 	mutex_lock(&kvm->lock);
180 	mutex_lock(&kvm->arch.config_lock);
181 	mutex_unlock(&kvm->arch.config_lock);
182 	mutex_unlock(&kvm->lock);
183 #endif
184 
185 	kvm_init_nested(kvm);
186 
187 	ret = kvm_share_hyp(kvm, kvm + 1);
188 	if (ret)
189 		return ret;
190 
191 	ret = pkvm_init_host_vm(kvm);
192 	if (ret)
193 		goto err_unshare_kvm;
194 
195 	if (!zalloc_cpumask_var(&kvm->arch.supported_cpus, GFP_KERNEL_ACCOUNT)) {
196 		ret = -ENOMEM;
197 		goto err_unshare_kvm;
198 	}
199 	cpumask_copy(kvm->arch.supported_cpus, cpu_possible_mask);
200 
201 	ret = kvm_init_stage2_mmu(kvm, &kvm->arch.mmu, type);
202 	if (ret)
203 		goto err_free_cpumask;
204 
205 	kvm_vgic_early_init(kvm);
206 
207 	kvm_timer_init_vm(kvm);
208 
209 	/* The maximum number of VCPUs is limited by the host's GIC model */
210 	kvm->max_vcpus = kvm_arm_default_max_vcpus();
211 
212 	kvm_arm_init_hypercalls(kvm);
213 
214 	bitmap_zero(kvm->arch.vcpu_features, KVM_VCPU_MAX_FEATURES);
215 
216 	return 0;
217 
218 err_free_cpumask:
219 	free_cpumask_var(kvm->arch.supported_cpus);
220 err_unshare_kvm:
221 	kvm_unshare_hyp(kvm, kvm + 1);
222 	return ret;
223 }
224 
225 vm_fault_t kvm_arch_vcpu_fault(struct kvm_vcpu *vcpu, struct vm_fault *vmf)
226 {
227 	return VM_FAULT_SIGBUS;
228 }
229 
230 void kvm_arch_create_vm_debugfs(struct kvm *kvm)
231 {
232 	kvm_sys_regs_create_debugfs(kvm);
233 	kvm_s2_ptdump_create_debugfs(kvm);
234 }
235 
236 static void kvm_destroy_mpidr_data(struct kvm *kvm)
237 {
238 	struct kvm_mpidr_data *data;
239 
240 	mutex_lock(&kvm->arch.config_lock);
241 
242 	data = rcu_dereference_protected(kvm->arch.mpidr_data,
243 					 lockdep_is_held(&kvm->arch.config_lock));
244 	if (data) {
245 		rcu_assign_pointer(kvm->arch.mpidr_data, NULL);
246 		synchronize_rcu();
247 		kfree(data);
248 	}
249 
250 	mutex_unlock(&kvm->arch.config_lock);
251 }
252 
253 /**
254  * kvm_arch_destroy_vm - destroy the VM data structure
255  * @kvm:	pointer to the KVM struct
256  */
257 void kvm_arch_destroy_vm(struct kvm *kvm)
258 {
259 	bitmap_free(kvm->arch.pmu_filter);
260 	free_cpumask_var(kvm->arch.supported_cpus);
261 
262 	kvm_vgic_destroy(kvm);
263 
264 	if (is_protected_kvm_enabled())
265 		pkvm_destroy_hyp_vm(kvm);
266 
267 	kvm_destroy_mpidr_data(kvm);
268 
269 	kfree(kvm->arch.sysreg_masks);
270 	kvm_destroy_vcpus(kvm);
271 
272 	kvm_unshare_hyp(kvm, kvm + 1);
273 
274 	kvm_arm_teardown_hypercalls(kvm);
275 }
276 
277 static bool kvm_has_full_ptr_auth(void)
278 {
279 	bool apa, gpa, api, gpi, apa3, gpa3;
280 	u64 isar1, isar2, val;
281 
282 	/*
283 	 * Check that:
284 	 *
285 	 * - both Address and Generic auth are implemented for a given
286          *   algorithm (Q5, IMPDEF or Q3)
287 	 * - only a single algorithm is implemented.
288 	 */
289 	if (!system_has_full_ptr_auth())
290 		return false;
291 
292 	isar1 = read_sanitised_ftr_reg(SYS_ID_AA64ISAR1_EL1);
293 	isar2 = read_sanitised_ftr_reg(SYS_ID_AA64ISAR2_EL1);
294 
295 	apa = !!FIELD_GET(ID_AA64ISAR1_EL1_APA_MASK, isar1);
296 	val = FIELD_GET(ID_AA64ISAR1_EL1_GPA_MASK, isar1);
297 	gpa = (val == ID_AA64ISAR1_EL1_GPA_IMP);
298 
299 	api = !!FIELD_GET(ID_AA64ISAR1_EL1_API_MASK, isar1);
300 	val = FIELD_GET(ID_AA64ISAR1_EL1_GPI_MASK, isar1);
301 	gpi = (val == ID_AA64ISAR1_EL1_GPI_IMP);
302 
303 	apa3 = !!FIELD_GET(ID_AA64ISAR2_EL1_APA3_MASK, isar2);
304 	val  = FIELD_GET(ID_AA64ISAR2_EL1_GPA3_MASK, isar2);
305 	gpa3 = (val == ID_AA64ISAR2_EL1_GPA3_IMP);
306 
307 	return (apa == gpa && api == gpi && apa3 == gpa3 &&
308 		(apa + api + apa3) == 1);
309 }
310 
311 int kvm_vm_ioctl_check_extension(struct kvm *kvm, long ext)
312 {
313 	int r;
314 
315 	if (kvm && kvm_vm_is_protected(kvm) && !pkvm_ext_allowed(kvm, ext))
316 		return 0;
317 
318 	switch (ext) {
319 	case KVM_CAP_IRQCHIP:
320 		r = vgic_present;
321 		break;
322 	case KVM_CAP_IOEVENTFD:
323 	case KVM_CAP_USER_MEMORY:
324 	case KVM_CAP_SYNC_MMU:
325 	case KVM_CAP_DESTROY_MEMORY_REGION_WORKS:
326 	case KVM_CAP_ONE_REG:
327 	case KVM_CAP_ARM_PSCI:
328 	case KVM_CAP_ARM_PSCI_0_2:
329 	case KVM_CAP_READONLY_MEM:
330 	case KVM_CAP_MP_STATE:
331 	case KVM_CAP_IMMEDIATE_EXIT:
332 	case KVM_CAP_VCPU_EVENTS:
333 	case KVM_CAP_ARM_IRQ_LINE_LAYOUT_2:
334 	case KVM_CAP_ARM_NISV_TO_USER:
335 	case KVM_CAP_ARM_INJECT_EXT_DABT:
336 	case KVM_CAP_SET_GUEST_DEBUG:
337 	case KVM_CAP_VCPU_ATTRIBUTES:
338 	case KVM_CAP_PTP_KVM:
339 	case KVM_CAP_ARM_SYSTEM_SUSPEND:
340 	case KVM_CAP_IRQFD_RESAMPLE:
341 	case KVM_CAP_COUNTER_OFFSET:
342 		r = 1;
343 		break;
344 	case KVM_CAP_SET_GUEST_DEBUG2:
345 		return KVM_GUESTDBG_VALID_MASK;
346 	case KVM_CAP_ARM_SET_DEVICE_ADDR:
347 		r = 1;
348 		break;
349 	case KVM_CAP_NR_VCPUS:
350 		/*
351 		 * ARM64 treats KVM_CAP_NR_CPUS differently from all other
352 		 * architectures, as it does not always bound it to
353 		 * KVM_CAP_MAX_VCPUS. It should not matter much because
354 		 * this is just an advisory value.
355 		 */
356 		r = min_t(unsigned int, num_online_cpus(),
357 			  kvm_arm_default_max_vcpus());
358 		break;
359 	case KVM_CAP_MAX_VCPUS:
360 	case KVM_CAP_MAX_VCPU_ID:
361 		if (kvm)
362 			r = kvm->max_vcpus;
363 		else
364 			r = kvm_arm_default_max_vcpus();
365 		break;
366 	case KVM_CAP_MSI_DEVID:
367 		if (!kvm)
368 			r = -EINVAL;
369 		else
370 			r = kvm->arch.vgic.msis_require_devid;
371 		break;
372 	case KVM_CAP_ARM_USER_IRQ:
373 		/*
374 		 * 1: EL1_VTIMER, EL1_PTIMER, and PMU.
375 		 * (bump this number if adding more devices)
376 		 */
377 		r = 1;
378 		break;
379 	case KVM_CAP_ARM_MTE:
380 		r = system_supports_mte();
381 		break;
382 	case KVM_CAP_STEAL_TIME:
383 		r = kvm_arm_pvtime_supported();
384 		break;
385 	case KVM_CAP_ARM_EL1_32BIT:
386 		r = cpus_have_final_cap(ARM64_HAS_32BIT_EL1);
387 		break;
388 	case KVM_CAP_GUEST_DEBUG_HW_BPS:
389 		r = get_num_brps();
390 		break;
391 	case KVM_CAP_GUEST_DEBUG_HW_WPS:
392 		r = get_num_wrps();
393 		break;
394 	case KVM_CAP_ARM_PMU_V3:
395 		r = kvm_arm_support_pmu_v3();
396 		break;
397 	case KVM_CAP_ARM_INJECT_SERROR_ESR:
398 		r = cpus_have_final_cap(ARM64_HAS_RAS_EXTN);
399 		break;
400 	case KVM_CAP_ARM_VM_IPA_SIZE:
401 		r = get_kvm_ipa_limit();
402 		break;
403 	case KVM_CAP_ARM_SVE:
404 		r = system_supports_sve();
405 		break;
406 	case KVM_CAP_ARM_PTRAUTH_ADDRESS:
407 	case KVM_CAP_ARM_PTRAUTH_GENERIC:
408 		r = kvm_has_full_ptr_auth();
409 		break;
410 	case KVM_CAP_ARM_EAGER_SPLIT_CHUNK_SIZE:
411 		if (kvm)
412 			r = kvm->arch.mmu.split_page_chunk_size;
413 		else
414 			r = KVM_ARM_EAGER_SPLIT_CHUNK_SIZE_DEFAULT;
415 		break;
416 	case KVM_CAP_ARM_SUPPORTED_BLOCK_SIZES:
417 		r = kvm_supported_block_sizes();
418 		break;
419 	case KVM_CAP_ARM_SUPPORTED_REG_MASK_RANGES:
420 		r = BIT(0);
421 		break;
422 	default:
423 		r = 0;
424 	}
425 
426 	return r;
427 }
428 
429 long kvm_arch_dev_ioctl(struct file *filp,
430 			unsigned int ioctl, unsigned long arg)
431 {
432 	return -EINVAL;
433 }
434 
435 struct kvm *kvm_arch_alloc_vm(void)
436 {
437 	size_t sz = sizeof(struct kvm);
438 
439 	if (!has_vhe())
440 		return kzalloc(sz, GFP_KERNEL_ACCOUNT);
441 
442 	return __vmalloc(sz, GFP_KERNEL_ACCOUNT | __GFP_HIGHMEM | __GFP_ZERO);
443 }
444 
445 int kvm_arch_vcpu_precreate(struct kvm *kvm, unsigned int id)
446 {
447 	if (irqchip_in_kernel(kvm) && vgic_initialized(kvm))
448 		return -EBUSY;
449 
450 	if (id >= kvm->max_vcpus)
451 		return -EINVAL;
452 
453 	return 0;
454 }
455 
456 int kvm_arch_vcpu_create(struct kvm_vcpu *vcpu)
457 {
458 	int err;
459 
460 	spin_lock_init(&vcpu->arch.mp_state_lock);
461 
462 #ifdef CONFIG_LOCKDEP
463 	/* Inform lockdep that the config_lock is acquired after vcpu->mutex */
464 	mutex_lock(&vcpu->mutex);
465 	mutex_lock(&vcpu->kvm->arch.config_lock);
466 	mutex_unlock(&vcpu->kvm->arch.config_lock);
467 	mutex_unlock(&vcpu->mutex);
468 #endif
469 
470 	/* Force users to call KVM_ARM_VCPU_INIT */
471 	vcpu_clear_flag(vcpu, VCPU_INITIALIZED);
472 
473 	vcpu->arch.mmu_page_cache.gfp_zero = __GFP_ZERO;
474 
475 	/* Set up the timer */
476 	kvm_timer_vcpu_init(vcpu);
477 
478 	kvm_pmu_vcpu_init(vcpu);
479 
480 	kvm_arm_reset_debug_ptr(vcpu);
481 
482 	kvm_arm_pvtime_vcpu_init(&vcpu->arch);
483 
484 	vcpu->arch.hw_mmu = &vcpu->kvm->arch.mmu;
485 
486 	/*
487 	 * This vCPU may have been created after mpidr_data was initialized.
488 	 * Throw out the pre-computed mappings if that is the case which forces
489 	 * KVM to fall back to iteratively searching the vCPUs.
490 	 */
491 	kvm_destroy_mpidr_data(vcpu->kvm);
492 
493 	err = kvm_vgic_vcpu_init(vcpu);
494 	if (err)
495 		return err;
496 
497 	return kvm_share_hyp(vcpu, vcpu + 1);
498 }
499 
500 void kvm_arch_vcpu_postcreate(struct kvm_vcpu *vcpu)
501 {
502 }
503 
504 void kvm_arch_vcpu_destroy(struct kvm_vcpu *vcpu)
505 {
506 	if (vcpu_has_run_once(vcpu) && unlikely(!irqchip_in_kernel(vcpu->kvm)))
507 		static_branch_dec(&userspace_irqchip_in_use);
508 
509 	kvm_mmu_free_memory_cache(&vcpu->arch.mmu_page_cache);
510 	kvm_timer_vcpu_terminate(vcpu);
511 	kvm_pmu_vcpu_destroy(vcpu);
512 	kvm_vgic_vcpu_destroy(vcpu);
513 	kvm_arm_vcpu_destroy(vcpu);
514 }
515 
516 void kvm_arch_vcpu_blocking(struct kvm_vcpu *vcpu)
517 {
518 
519 }
520 
521 void kvm_arch_vcpu_unblocking(struct kvm_vcpu *vcpu)
522 {
523 
524 }
525 
526 static void vcpu_set_pauth_traps(struct kvm_vcpu *vcpu)
527 {
528 	if (vcpu_has_ptrauth(vcpu) && !is_protected_kvm_enabled()) {
529 		/*
530 		 * Either we're running an L2 guest, and the API/APK bits come
531 		 * from L1's HCR_EL2, or API/APK are both set.
532 		 */
533 		if (unlikely(vcpu_has_nv(vcpu) && !is_hyp_ctxt(vcpu))) {
534 			u64 val;
535 
536 			val = __vcpu_sys_reg(vcpu, HCR_EL2);
537 			val &= (HCR_API | HCR_APK);
538 			vcpu->arch.hcr_el2 &= ~(HCR_API | HCR_APK);
539 			vcpu->arch.hcr_el2 |= val;
540 		} else {
541 			vcpu->arch.hcr_el2 |= (HCR_API | HCR_APK);
542 		}
543 
544 		/*
545 		 * Save the host keys if there is any chance for the guest
546 		 * to use pauth, as the entry code will reload the guest
547 		 * keys in that case.
548 		 */
549 		if (vcpu->arch.hcr_el2 & (HCR_API | HCR_APK)) {
550 			struct kvm_cpu_context *ctxt;
551 
552 			ctxt = this_cpu_ptr_hyp_sym(kvm_hyp_ctxt);
553 			ptrauth_save_keys(ctxt);
554 		}
555 	}
556 }
557 
558 static bool kvm_vcpu_should_clear_twi(struct kvm_vcpu *vcpu)
559 {
560 	if (unlikely(kvm_wfi_trap_policy != KVM_WFX_NOTRAP_SINGLE_TASK))
561 		return kvm_wfi_trap_policy == KVM_WFX_NOTRAP;
562 
563 	return single_task_running() &&
564 	       (atomic_read(&vcpu->arch.vgic_cpu.vgic_v3.its_vpe.vlpi_count) ||
565 		vcpu->kvm->arch.vgic.nassgireq);
566 }
567 
568 static bool kvm_vcpu_should_clear_twe(struct kvm_vcpu *vcpu)
569 {
570 	if (unlikely(kvm_wfe_trap_policy != KVM_WFX_NOTRAP_SINGLE_TASK))
571 		return kvm_wfe_trap_policy == KVM_WFX_NOTRAP;
572 
573 	return single_task_running();
574 }
575 
576 void kvm_arch_vcpu_load(struct kvm_vcpu *vcpu, int cpu)
577 {
578 	struct kvm_s2_mmu *mmu;
579 	int *last_ran;
580 
581 	if (vcpu_has_nv(vcpu))
582 		kvm_vcpu_load_hw_mmu(vcpu);
583 
584 	mmu = vcpu->arch.hw_mmu;
585 	last_ran = this_cpu_ptr(mmu->last_vcpu_ran);
586 
587 	/*
588 	 * We guarantee that both TLBs and I-cache are private to each
589 	 * vcpu. If detecting that a vcpu from the same VM has
590 	 * previously run on the same physical CPU, call into the
591 	 * hypervisor code to nuke the relevant contexts.
592 	 *
593 	 * We might get preempted before the vCPU actually runs, but
594 	 * over-invalidation doesn't affect correctness.
595 	 */
596 	if (*last_ran != vcpu->vcpu_idx) {
597 		kvm_call_hyp(__kvm_flush_cpu_context, mmu);
598 		*last_ran = vcpu->vcpu_idx;
599 	}
600 
601 	vcpu->cpu = cpu;
602 
603 	kvm_vgic_load(vcpu);
604 	kvm_timer_vcpu_load(vcpu);
605 	if (has_vhe())
606 		kvm_vcpu_load_vhe(vcpu);
607 	kvm_arch_vcpu_load_fp(vcpu);
608 	kvm_vcpu_pmu_restore_guest(vcpu);
609 	if (kvm_arm_is_pvtime_enabled(&vcpu->arch))
610 		kvm_make_request(KVM_REQ_RECORD_STEAL, vcpu);
611 
612 	if (kvm_vcpu_should_clear_twe(vcpu))
613 		vcpu->arch.hcr_el2 &= ~HCR_TWE;
614 	else
615 		vcpu->arch.hcr_el2 |= HCR_TWE;
616 
617 	if (kvm_vcpu_should_clear_twi(vcpu))
618 		vcpu->arch.hcr_el2 &= ~HCR_TWI;
619 	else
620 		vcpu->arch.hcr_el2 |= HCR_TWI;
621 
622 	vcpu_set_pauth_traps(vcpu);
623 
624 	kvm_arch_vcpu_load_debug_state_flags(vcpu);
625 
626 	if (!cpumask_test_cpu(cpu, vcpu->kvm->arch.supported_cpus))
627 		vcpu_set_on_unsupported_cpu(vcpu);
628 }
629 
630 void kvm_arch_vcpu_put(struct kvm_vcpu *vcpu)
631 {
632 	kvm_arch_vcpu_put_debug_state_flags(vcpu);
633 	kvm_arch_vcpu_put_fp(vcpu);
634 	if (has_vhe())
635 		kvm_vcpu_put_vhe(vcpu);
636 	kvm_timer_vcpu_put(vcpu);
637 	kvm_vgic_put(vcpu);
638 	kvm_vcpu_pmu_restore_host(vcpu);
639 	if (vcpu_has_nv(vcpu))
640 		kvm_vcpu_put_hw_mmu(vcpu);
641 	kvm_arm_vmid_clear_active();
642 
643 	vcpu_clear_on_unsupported_cpu(vcpu);
644 	vcpu->cpu = -1;
645 }
646 
647 static void __kvm_arm_vcpu_power_off(struct kvm_vcpu *vcpu)
648 {
649 	WRITE_ONCE(vcpu->arch.mp_state.mp_state, KVM_MP_STATE_STOPPED);
650 	kvm_make_request(KVM_REQ_SLEEP, vcpu);
651 	kvm_vcpu_kick(vcpu);
652 }
653 
654 void kvm_arm_vcpu_power_off(struct kvm_vcpu *vcpu)
655 {
656 	spin_lock(&vcpu->arch.mp_state_lock);
657 	__kvm_arm_vcpu_power_off(vcpu);
658 	spin_unlock(&vcpu->arch.mp_state_lock);
659 }
660 
661 bool kvm_arm_vcpu_stopped(struct kvm_vcpu *vcpu)
662 {
663 	return READ_ONCE(vcpu->arch.mp_state.mp_state) == KVM_MP_STATE_STOPPED;
664 }
665 
666 static void kvm_arm_vcpu_suspend(struct kvm_vcpu *vcpu)
667 {
668 	WRITE_ONCE(vcpu->arch.mp_state.mp_state, KVM_MP_STATE_SUSPENDED);
669 	kvm_make_request(KVM_REQ_SUSPEND, vcpu);
670 	kvm_vcpu_kick(vcpu);
671 }
672 
673 static bool kvm_arm_vcpu_suspended(struct kvm_vcpu *vcpu)
674 {
675 	return READ_ONCE(vcpu->arch.mp_state.mp_state) == KVM_MP_STATE_SUSPENDED;
676 }
677 
678 int kvm_arch_vcpu_ioctl_get_mpstate(struct kvm_vcpu *vcpu,
679 				    struct kvm_mp_state *mp_state)
680 {
681 	*mp_state = READ_ONCE(vcpu->arch.mp_state);
682 
683 	return 0;
684 }
685 
686 int kvm_arch_vcpu_ioctl_set_mpstate(struct kvm_vcpu *vcpu,
687 				    struct kvm_mp_state *mp_state)
688 {
689 	int ret = 0;
690 
691 	spin_lock(&vcpu->arch.mp_state_lock);
692 
693 	switch (mp_state->mp_state) {
694 	case KVM_MP_STATE_RUNNABLE:
695 		WRITE_ONCE(vcpu->arch.mp_state, *mp_state);
696 		break;
697 	case KVM_MP_STATE_STOPPED:
698 		__kvm_arm_vcpu_power_off(vcpu);
699 		break;
700 	case KVM_MP_STATE_SUSPENDED:
701 		kvm_arm_vcpu_suspend(vcpu);
702 		break;
703 	default:
704 		ret = -EINVAL;
705 	}
706 
707 	spin_unlock(&vcpu->arch.mp_state_lock);
708 
709 	return ret;
710 }
711 
712 /**
713  * kvm_arch_vcpu_runnable - determine if the vcpu can be scheduled
714  * @v:		The VCPU pointer
715  *
716  * If the guest CPU is not waiting for interrupts or an interrupt line is
717  * asserted, the CPU is by definition runnable.
718  */
719 int kvm_arch_vcpu_runnable(struct kvm_vcpu *v)
720 {
721 	bool irq_lines = *vcpu_hcr(v) & (HCR_VI | HCR_VF);
722 	return ((irq_lines || kvm_vgic_vcpu_pending_irq(v))
723 		&& !kvm_arm_vcpu_stopped(v) && !v->arch.pause);
724 }
725 
726 bool kvm_arch_vcpu_in_kernel(struct kvm_vcpu *vcpu)
727 {
728 	return vcpu_mode_priv(vcpu);
729 }
730 
731 #ifdef CONFIG_GUEST_PERF_EVENTS
732 unsigned long kvm_arch_vcpu_get_ip(struct kvm_vcpu *vcpu)
733 {
734 	return *vcpu_pc(vcpu);
735 }
736 #endif
737 
738 static void kvm_init_mpidr_data(struct kvm *kvm)
739 {
740 	struct kvm_mpidr_data *data = NULL;
741 	unsigned long c, mask, nr_entries;
742 	u64 aff_set = 0, aff_clr = ~0UL;
743 	struct kvm_vcpu *vcpu;
744 
745 	mutex_lock(&kvm->arch.config_lock);
746 
747 	if (rcu_access_pointer(kvm->arch.mpidr_data) ||
748 	    atomic_read(&kvm->online_vcpus) == 1)
749 		goto out;
750 
751 	kvm_for_each_vcpu(c, vcpu, kvm) {
752 		u64 aff = kvm_vcpu_get_mpidr_aff(vcpu);
753 		aff_set |= aff;
754 		aff_clr &= aff;
755 	}
756 
757 	/*
758 	 * A significant bit can be either 0 or 1, and will only appear in
759 	 * aff_set. Use aff_clr to weed out the useless stuff.
760 	 */
761 	mask = aff_set ^ aff_clr;
762 	nr_entries = BIT_ULL(hweight_long(mask));
763 
764 	/*
765 	 * Don't let userspace fool us. If we need more than a single page
766 	 * to describe the compressed MPIDR array, just fall back to the
767 	 * iterative method. Single vcpu VMs do not need this either.
768 	 */
769 	if (struct_size(data, cmpidr_to_idx, nr_entries) <= PAGE_SIZE)
770 		data = kzalloc(struct_size(data, cmpidr_to_idx, nr_entries),
771 			       GFP_KERNEL_ACCOUNT);
772 
773 	if (!data)
774 		goto out;
775 
776 	data->mpidr_mask = mask;
777 
778 	kvm_for_each_vcpu(c, vcpu, kvm) {
779 		u64 aff = kvm_vcpu_get_mpidr_aff(vcpu);
780 		u16 index = kvm_mpidr_index(data, aff);
781 
782 		data->cmpidr_to_idx[index] = c;
783 	}
784 
785 	rcu_assign_pointer(kvm->arch.mpidr_data, data);
786 out:
787 	mutex_unlock(&kvm->arch.config_lock);
788 }
789 
790 /*
791  * Handle both the initialisation that is being done when the vcpu is
792  * run for the first time, as well as the updates that must be
793  * performed each time we get a new thread dealing with this vcpu.
794  */
795 int kvm_arch_vcpu_run_pid_change(struct kvm_vcpu *vcpu)
796 {
797 	struct kvm *kvm = vcpu->kvm;
798 	int ret;
799 
800 	if (!kvm_vcpu_initialized(vcpu))
801 		return -ENOEXEC;
802 
803 	if (!kvm_arm_vcpu_is_finalized(vcpu))
804 		return -EPERM;
805 
806 	ret = kvm_arch_vcpu_run_map_fp(vcpu);
807 	if (ret)
808 		return ret;
809 
810 	if (likely(vcpu_has_run_once(vcpu)))
811 		return 0;
812 
813 	kvm_init_mpidr_data(kvm);
814 
815 	kvm_arm_vcpu_init_debug(vcpu);
816 
817 	if (likely(irqchip_in_kernel(kvm))) {
818 		/*
819 		 * Map the VGIC hardware resources before running a vcpu the
820 		 * first time on this VM.
821 		 */
822 		ret = kvm_vgic_map_resources(kvm);
823 		if (ret)
824 			return ret;
825 	}
826 
827 	ret = kvm_finalize_sys_regs(vcpu);
828 	if (ret)
829 		return ret;
830 
831 	/*
832 	 * This needs to happen after any restriction has been applied
833 	 * to the feature set.
834 	 */
835 	kvm_calculate_traps(vcpu);
836 
837 	ret = kvm_timer_enable(vcpu);
838 	if (ret)
839 		return ret;
840 
841 	ret = kvm_arm_pmu_v3_enable(vcpu);
842 	if (ret)
843 		return ret;
844 
845 	if (is_protected_kvm_enabled()) {
846 		ret = pkvm_create_hyp_vm(kvm);
847 		if (ret)
848 			return ret;
849 	}
850 
851 	if (!irqchip_in_kernel(kvm)) {
852 		/*
853 		 * Tell the rest of the code that there are userspace irqchip
854 		 * VMs in the wild.
855 		 */
856 		static_branch_inc(&userspace_irqchip_in_use);
857 	}
858 
859 	/*
860 	 * Initialize traps for protected VMs.
861 	 * NOTE: Move to run in EL2 directly, rather than via a hypercall, once
862 	 * the code is in place for first run initialization at EL2.
863 	 */
864 	if (kvm_vm_is_protected(kvm))
865 		kvm_call_hyp_nvhe(__pkvm_vcpu_init_traps, vcpu);
866 
867 	mutex_lock(&kvm->arch.config_lock);
868 	set_bit(KVM_ARCH_FLAG_HAS_RAN_ONCE, &kvm->arch.flags);
869 	mutex_unlock(&kvm->arch.config_lock);
870 
871 	return ret;
872 }
873 
874 bool kvm_arch_intc_initialized(struct kvm *kvm)
875 {
876 	return vgic_initialized(kvm);
877 }
878 
879 void kvm_arm_halt_guest(struct kvm *kvm)
880 {
881 	unsigned long i;
882 	struct kvm_vcpu *vcpu;
883 
884 	kvm_for_each_vcpu(i, vcpu, kvm)
885 		vcpu->arch.pause = true;
886 	kvm_make_all_cpus_request(kvm, KVM_REQ_SLEEP);
887 }
888 
889 void kvm_arm_resume_guest(struct kvm *kvm)
890 {
891 	unsigned long i;
892 	struct kvm_vcpu *vcpu;
893 
894 	kvm_for_each_vcpu(i, vcpu, kvm) {
895 		vcpu->arch.pause = false;
896 		__kvm_vcpu_wake_up(vcpu);
897 	}
898 }
899 
900 static void kvm_vcpu_sleep(struct kvm_vcpu *vcpu)
901 {
902 	struct rcuwait *wait = kvm_arch_vcpu_get_wait(vcpu);
903 
904 	rcuwait_wait_event(wait,
905 			   (!kvm_arm_vcpu_stopped(vcpu)) && (!vcpu->arch.pause),
906 			   TASK_INTERRUPTIBLE);
907 
908 	if (kvm_arm_vcpu_stopped(vcpu) || vcpu->arch.pause) {
909 		/* Awaken to handle a signal, request we sleep again later. */
910 		kvm_make_request(KVM_REQ_SLEEP, vcpu);
911 	}
912 
913 	/*
914 	 * Make sure we will observe a potential reset request if we've
915 	 * observed a change to the power state. Pairs with the smp_wmb() in
916 	 * kvm_psci_vcpu_on().
917 	 */
918 	smp_rmb();
919 }
920 
921 /**
922  * kvm_vcpu_wfi - emulate Wait-For-Interrupt behavior
923  * @vcpu:	The VCPU pointer
924  *
925  * Suspend execution of a vCPU until a valid wake event is detected, i.e. until
926  * the vCPU is runnable.  The vCPU may or may not be scheduled out, depending
927  * on when a wake event arrives, e.g. there may already be a pending wake event.
928  */
929 void kvm_vcpu_wfi(struct kvm_vcpu *vcpu)
930 {
931 	/*
932 	 * Sync back the state of the GIC CPU interface so that we have
933 	 * the latest PMR and group enables. This ensures that
934 	 * kvm_arch_vcpu_runnable has up-to-date data to decide whether
935 	 * we have pending interrupts, e.g. when determining if the
936 	 * vCPU should block.
937 	 *
938 	 * For the same reason, we want to tell GICv4 that we need
939 	 * doorbells to be signalled, should an interrupt become pending.
940 	 */
941 	preempt_disable();
942 	vcpu_set_flag(vcpu, IN_WFI);
943 	kvm_vgic_put(vcpu);
944 	preempt_enable();
945 
946 	kvm_vcpu_halt(vcpu);
947 	vcpu_clear_flag(vcpu, IN_WFIT);
948 
949 	preempt_disable();
950 	vcpu_clear_flag(vcpu, IN_WFI);
951 	kvm_vgic_load(vcpu);
952 	preempt_enable();
953 }
954 
955 static int kvm_vcpu_suspend(struct kvm_vcpu *vcpu)
956 {
957 	if (!kvm_arm_vcpu_suspended(vcpu))
958 		return 1;
959 
960 	kvm_vcpu_wfi(vcpu);
961 
962 	/*
963 	 * The suspend state is sticky; we do not leave it until userspace
964 	 * explicitly marks the vCPU as runnable. Request that we suspend again
965 	 * later.
966 	 */
967 	kvm_make_request(KVM_REQ_SUSPEND, vcpu);
968 
969 	/*
970 	 * Check to make sure the vCPU is actually runnable. If so, exit to
971 	 * userspace informing it of the wakeup condition.
972 	 */
973 	if (kvm_arch_vcpu_runnable(vcpu)) {
974 		memset(&vcpu->run->system_event, 0, sizeof(vcpu->run->system_event));
975 		vcpu->run->system_event.type = KVM_SYSTEM_EVENT_WAKEUP;
976 		vcpu->run->exit_reason = KVM_EXIT_SYSTEM_EVENT;
977 		return 0;
978 	}
979 
980 	/*
981 	 * Otherwise, we were unblocked to process a different event, such as a
982 	 * pending signal. Return 1 and allow kvm_arch_vcpu_ioctl_run() to
983 	 * process the event.
984 	 */
985 	return 1;
986 }
987 
988 /**
989  * check_vcpu_requests - check and handle pending vCPU requests
990  * @vcpu:	the VCPU pointer
991  *
992  * Return: 1 if we should enter the guest
993  *	   0 if we should exit to userspace
994  *	   < 0 if we should exit to userspace, where the return value indicates
995  *	   an error
996  */
997 static int check_vcpu_requests(struct kvm_vcpu *vcpu)
998 {
999 	if (kvm_request_pending(vcpu)) {
1000 		if (kvm_check_request(KVM_REQ_SLEEP, vcpu))
1001 			kvm_vcpu_sleep(vcpu);
1002 
1003 		if (kvm_check_request(KVM_REQ_VCPU_RESET, vcpu))
1004 			kvm_reset_vcpu(vcpu);
1005 
1006 		/*
1007 		 * Clear IRQ_PENDING requests that were made to guarantee
1008 		 * that a VCPU sees new virtual interrupts.
1009 		 */
1010 		kvm_check_request(KVM_REQ_IRQ_PENDING, vcpu);
1011 
1012 		if (kvm_check_request(KVM_REQ_RECORD_STEAL, vcpu))
1013 			kvm_update_stolen_time(vcpu);
1014 
1015 		if (kvm_check_request(KVM_REQ_RELOAD_GICv4, vcpu)) {
1016 			/* The distributor enable bits were changed */
1017 			preempt_disable();
1018 			vgic_v4_put(vcpu);
1019 			vgic_v4_load(vcpu);
1020 			preempt_enable();
1021 		}
1022 
1023 		if (kvm_check_request(KVM_REQ_RELOAD_PMU, vcpu))
1024 			kvm_vcpu_reload_pmu(vcpu);
1025 
1026 		if (kvm_check_request(KVM_REQ_RESYNC_PMU_EL0, vcpu))
1027 			kvm_vcpu_pmu_restore_guest(vcpu);
1028 
1029 		if (kvm_check_request(KVM_REQ_SUSPEND, vcpu))
1030 			return kvm_vcpu_suspend(vcpu);
1031 
1032 		if (kvm_dirty_ring_check_request(vcpu))
1033 			return 0;
1034 	}
1035 
1036 	return 1;
1037 }
1038 
1039 static bool vcpu_mode_is_bad_32bit(struct kvm_vcpu *vcpu)
1040 {
1041 	if (likely(!vcpu_mode_is_32bit(vcpu)))
1042 		return false;
1043 
1044 	if (vcpu_has_nv(vcpu))
1045 		return true;
1046 
1047 	return !kvm_supports_32bit_el0();
1048 }
1049 
1050 /**
1051  * kvm_vcpu_exit_request - returns true if the VCPU should *not* enter the guest
1052  * @vcpu:	The VCPU pointer
1053  * @ret:	Pointer to write optional return code
1054  *
1055  * Returns: true if the VCPU needs to return to a preemptible + interruptible
1056  *	    and skip guest entry.
1057  *
1058  * This function disambiguates between two different types of exits: exits to a
1059  * preemptible + interruptible kernel context and exits to userspace. For an
1060  * exit to userspace, this function will write the return code to ret and return
1061  * true. For an exit to preemptible + interruptible kernel context (i.e. check
1062  * for pending work and re-enter), return true without writing to ret.
1063  */
1064 static bool kvm_vcpu_exit_request(struct kvm_vcpu *vcpu, int *ret)
1065 {
1066 	struct kvm_run *run = vcpu->run;
1067 
1068 	/*
1069 	 * If we're using a userspace irqchip, then check if we need
1070 	 * to tell a userspace irqchip about timer or PMU level
1071 	 * changes and if so, exit to userspace (the actual level
1072 	 * state gets updated in kvm_timer_update_run and
1073 	 * kvm_pmu_update_run below).
1074 	 */
1075 	if (static_branch_unlikely(&userspace_irqchip_in_use)) {
1076 		if (kvm_timer_should_notify_user(vcpu) ||
1077 		    kvm_pmu_should_notify_user(vcpu)) {
1078 			*ret = -EINTR;
1079 			run->exit_reason = KVM_EXIT_INTR;
1080 			return true;
1081 		}
1082 	}
1083 
1084 	if (unlikely(vcpu_on_unsupported_cpu(vcpu))) {
1085 		run->exit_reason = KVM_EXIT_FAIL_ENTRY;
1086 		run->fail_entry.hardware_entry_failure_reason = KVM_EXIT_FAIL_ENTRY_CPU_UNSUPPORTED;
1087 		run->fail_entry.cpu = smp_processor_id();
1088 		*ret = 0;
1089 		return true;
1090 	}
1091 
1092 	return kvm_request_pending(vcpu) ||
1093 			xfer_to_guest_mode_work_pending();
1094 }
1095 
1096 /*
1097  * Actually run the vCPU, entering an RCU extended quiescent state (EQS) while
1098  * the vCPU is running.
1099  *
1100  * This must be noinstr as instrumentation may make use of RCU, and this is not
1101  * safe during the EQS.
1102  */
1103 static int noinstr kvm_arm_vcpu_enter_exit(struct kvm_vcpu *vcpu)
1104 {
1105 	int ret;
1106 
1107 	guest_state_enter_irqoff();
1108 	ret = kvm_call_hyp_ret(__kvm_vcpu_run, vcpu);
1109 	guest_state_exit_irqoff();
1110 
1111 	return ret;
1112 }
1113 
1114 /**
1115  * kvm_arch_vcpu_ioctl_run - the main VCPU run function to execute guest code
1116  * @vcpu:	The VCPU pointer
1117  *
1118  * This function is called through the VCPU_RUN ioctl called from user space. It
1119  * will execute VM code in a loop until the time slice for the process is used
1120  * or some emulation is needed from user space in which case the function will
1121  * return with return value 0 and with the kvm_run structure filled in with the
1122  * required data for the requested emulation.
1123  */
1124 int kvm_arch_vcpu_ioctl_run(struct kvm_vcpu *vcpu)
1125 {
1126 	struct kvm_run *run = vcpu->run;
1127 	int ret;
1128 
1129 	if (run->exit_reason == KVM_EXIT_MMIO) {
1130 		ret = kvm_handle_mmio_return(vcpu);
1131 		if (ret <= 0)
1132 			return ret;
1133 	}
1134 
1135 	vcpu_load(vcpu);
1136 
1137 	if (!vcpu->wants_to_run) {
1138 		ret = -EINTR;
1139 		goto out;
1140 	}
1141 
1142 	kvm_sigset_activate(vcpu);
1143 
1144 	ret = 1;
1145 	run->exit_reason = KVM_EXIT_UNKNOWN;
1146 	run->flags = 0;
1147 	while (ret > 0) {
1148 		/*
1149 		 * Check conditions before entering the guest
1150 		 */
1151 		ret = xfer_to_guest_mode_handle_work(vcpu);
1152 		if (!ret)
1153 			ret = 1;
1154 
1155 		if (ret > 0)
1156 			ret = check_vcpu_requests(vcpu);
1157 
1158 		/*
1159 		 * Preparing the interrupts to be injected also
1160 		 * involves poking the GIC, which must be done in a
1161 		 * non-preemptible context.
1162 		 */
1163 		preempt_disable();
1164 
1165 		/*
1166 		 * The VMID allocator only tracks active VMIDs per
1167 		 * physical CPU, and therefore the VMID allocated may not be
1168 		 * preserved on VMID roll-over if the task was preempted,
1169 		 * making a thread's VMID inactive. So we need to call
1170 		 * kvm_arm_vmid_update() in non-premptible context.
1171 		 */
1172 		if (kvm_arm_vmid_update(&vcpu->arch.hw_mmu->vmid) &&
1173 		    has_vhe())
1174 			__load_stage2(vcpu->arch.hw_mmu,
1175 				      vcpu->arch.hw_mmu->arch);
1176 
1177 		kvm_pmu_flush_hwstate(vcpu);
1178 
1179 		local_irq_disable();
1180 
1181 		kvm_vgic_flush_hwstate(vcpu);
1182 
1183 		kvm_pmu_update_vcpu_events(vcpu);
1184 
1185 		/*
1186 		 * Ensure we set mode to IN_GUEST_MODE after we disable
1187 		 * interrupts and before the final VCPU requests check.
1188 		 * See the comment in kvm_vcpu_exiting_guest_mode() and
1189 		 * Documentation/virt/kvm/vcpu-requests.rst
1190 		 */
1191 		smp_store_mb(vcpu->mode, IN_GUEST_MODE);
1192 
1193 		if (ret <= 0 || kvm_vcpu_exit_request(vcpu, &ret)) {
1194 			vcpu->mode = OUTSIDE_GUEST_MODE;
1195 			isb(); /* Ensure work in x_flush_hwstate is committed */
1196 			kvm_pmu_sync_hwstate(vcpu);
1197 			if (static_branch_unlikely(&userspace_irqchip_in_use))
1198 				kvm_timer_sync_user(vcpu);
1199 			kvm_vgic_sync_hwstate(vcpu);
1200 			local_irq_enable();
1201 			preempt_enable();
1202 			continue;
1203 		}
1204 
1205 		kvm_arm_setup_debug(vcpu);
1206 		kvm_arch_vcpu_ctxflush_fp(vcpu);
1207 
1208 		/**************************************************************
1209 		 * Enter the guest
1210 		 */
1211 		trace_kvm_entry(*vcpu_pc(vcpu));
1212 		guest_timing_enter_irqoff();
1213 
1214 		ret = kvm_arm_vcpu_enter_exit(vcpu);
1215 
1216 		vcpu->mode = OUTSIDE_GUEST_MODE;
1217 		vcpu->stat.exits++;
1218 		/*
1219 		 * Back from guest
1220 		 *************************************************************/
1221 
1222 		kvm_arm_clear_debug(vcpu);
1223 
1224 		/*
1225 		 * We must sync the PMU state before the vgic state so
1226 		 * that the vgic can properly sample the updated state of the
1227 		 * interrupt line.
1228 		 */
1229 		kvm_pmu_sync_hwstate(vcpu);
1230 
1231 		/*
1232 		 * Sync the vgic state before syncing the timer state because
1233 		 * the timer code needs to know if the virtual timer
1234 		 * interrupts are active.
1235 		 */
1236 		kvm_vgic_sync_hwstate(vcpu);
1237 
1238 		/*
1239 		 * Sync the timer hardware state before enabling interrupts as
1240 		 * we don't want vtimer interrupts to race with syncing the
1241 		 * timer virtual interrupt state.
1242 		 */
1243 		if (static_branch_unlikely(&userspace_irqchip_in_use))
1244 			kvm_timer_sync_user(vcpu);
1245 
1246 		kvm_arch_vcpu_ctxsync_fp(vcpu);
1247 
1248 		/*
1249 		 * We must ensure that any pending interrupts are taken before
1250 		 * we exit guest timing so that timer ticks are accounted as
1251 		 * guest time. Transiently unmask interrupts so that any
1252 		 * pending interrupts are taken.
1253 		 *
1254 		 * Per ARM DDI 0487G.b section D1.13.4, an ISB (or other
1255 		 * context synchronization event) is necessary to ensure that
1256 		 * pending interrupts are taken.
1257 		 */
1258 		if (ARM_EXCEPTION_CODE(ret) == ARM_EXCEPTION_IRQ) {
1259 			local_irq_enable();
1260 			isb();
1261 			local_irq_disable();
1262 		}
1263 
1264 		guest_timing_exit_irqoff();
1265 
1266 		local_irq_enable();
1267 
1268 		trace_kvm_exit(ret, kvm_vcpu_trap_get_class(vcpu), *vcpu_pc(vcpu));
1269 
1270 		/* Exit types that need handling before we can be preempted */
1271 		handle_exit_early(vcpu, ret);
1272 
1273 		preempt_enable();
1274 
1275 		/*
1276 		 * The ARMv8 architecture doesn't give the hypervisor
1277 		 * a mechanism to prevent a guest from dropping to AArch32 EL0
1278 		 * if implemented by the CPU. If we spot the guest in such
1279 		 * state and that we decided it wasn't supposed to do so (like
1280 		 * with the asymmetric AArch32 case), return to userspace with
1281 		 * a fatal error.
1282 		 */
1283 		if (vcpu_mode_is_bad_32bit(vcpu)) {
1284 			/*
1285 			 * As we have caught the guest red-handed, decide that
1286 			 * it isn't fit for purpose anymore by making the vcpu
1287 			 * invalid. The VMM can try and fix it by issuing  a
1288 			 * KVM_ARM_VCPU_INIT if it really wants to.
1289 			 */
1290 			vcpu_clear_flag(vcpu, VCPU_INITIALIZED);
1291 			ret = ARM_EXCEPTION_IL;
1292 		}
1293 
1294 		ret = handle_exit(vcpu, ret);
1295 	}
1296 
1297 	/* Tell userspace about in-kernel device output levels */
1298 	if (unlikely(!irqchip_in_kernel(vcpu->kvm))) {
1299 		kvm_timer_update_run(vcpu);
1300 		kvm_pmu_update_run(vcpu);
1301 	}
1302 
1303 	kvm_sigset_deactivate(vcpu);
1304 
1305 out:
1306 	/*
1307 	 * In the unlikely event that we are returning to userspace
1308 	 * with pending exceptions or PC adjustment, commit these
1309 	 * adjustments in order to give userspace a consistent view of
1310 	 * the vcpu state. Note that this relies on __kvm_adjust_pc()
1311 	 * being preempt-safe on VHE.
1312 	 */
1313 	if (unlikely(vcpu_get_flag(vcpu, PENDING_EXCEPTION) ||
1314 		     vcpu_get_flag(vcpu, INCREMENT_PC)))
1315 		kvm_call_hyp(__kvm_adjust_pc, vcpu);
1316 
1317 	vcpu_put(vcpu);
1318 	return ret;
1319 }
1320 
1321 static int vcpu_interrupt_line(struct kvm_vcpu *vcpu, int number, bool level)
1322 {
1323 	int bit_index;
1324 	bool set;
1325 	unsigned long *hcr;
1326 
1327 	if (number == KVM_ARM_IRQ_CPU_IRQ)
1328 		bit_index = __ffs(HCR_VI);
1329 	else /* KVM_ARM_IRQ_CPU_FIQ */
1330 		bit_index = __ffs(HCR_VF);
1331 
1332 	hcr = vcpu_hcr(vcpu);
1333 	if (level)
1334 		set = test_and_set_bit(bit_index, hcr);
1335 	else
1336 		set = test_and_clear_bit(bit_index, hcr);
1337 
1338 	/*
1339 	 * If we didn't change anything, no need to wake up or kick other CPUs
1340 	 */
1341 	if (set == level)
1342 		return 0;
1343 
1344 	/*
1345 	 * The vcpu irq_lines field was updated, wake up sleeping VCPUs and
1346 	 * trigger a world-switch round on the running physical CPU to set the
1347 	 * virtual IRQ/FIQ fields in the HCR appropriately.
1348 	 */
1349 	kvm_make_request(KVM_REQ_IRQ_PENDING, vcpu);
1350 	kvm_vcpu_kick(vcpu);
1351 
1352 	return 0;
1353 }
1354 
1355 int kvm_vm_ioctl_irq_line(struct kvm *kvm, struct kvm_irq_level *irq_level,
1356 			  bool line_status)
1357 {
1358 	u32 irq = irq_level->irq;
1359 	unsigned int irq_type, vcpu_id, irq_num;
1360 	struct kvm_vcpu *vcpu = NULL;
1361 	bool level = irq_level->level;
1362 
1363 	irq_type = (irq >> KVM_ARM_IRQ_TYPE_SHIFT) & KVM_ARM_IRQ_TYPE_MASK;
1364 	vcpu_id = (irq >> KVM_ARM_IRQ_VCPU_SHIFT) & KVM_ARM_IRQ_VCPU_MASK;
1365 	vcpu_id += ((irq >> KVM_ARM_IRQ_VCPU2_SHIFT) & KVM_ARM_IRQ_VCPU2_MASK) * (KVM_ARM_IRQ_VCPU_MASK + 1);
1366 	irq_num = (irq >> KVM_ARM_IRQ_NUM_SHIFT) & KVM_ARM_IRQ_NUM_MASK;
1367 
1368 	trace_kvm_irq_line(irq_type, vcpu_id, irq_num, irq_level->level);
1369 
1370 	switch (irq_type) {
1371 	case KVM_ARM_IRQ_TYPE_CPU:
1372 		if (irqchip_in_kernel(kvm))
1373 			return -ENXIO;
1374 
1375 		vcpu = kvm_get_vcpu_by_id(kvm, vcpu_id);
1376 		if (!vcpu)
1377 			return -EINVAL;
1378 
1379 		if (irq_num > KVM_ARM_IRQ_CPU_FIQ)
1380 			return -EINVAL;
1381 
1382 		return vcpu_interrupt_line(vcpu, irq_num, level);
1383 	case KVM_ARM_IRQ_TYPE_PPI:
1384 		if (!irqchip_in_kernel(kvm))
1385 			return -ENXIO;
1386 
1387 		vcpu = kvm_get_vcpu_by_id(kvm, vcpu_id);
1388 		if (!vcpu)
1389 			return -EINVAL;
1390 
1391 		if (irq_num < VGIC_NR_SGIS || irq_num >= VGIC_NR_PRIVATE_IRQS)
1392 			return -EINVAL;
1393 
1394 		return kvm_vgic_inject_irq(kvm, vcpu, irq_num, level, NULL);
1395 	case KVM_ARM_IRQ_TYPE_SPI:
1396 		if (!irqchip_in_kernel(kvm))
1397 			return -ENXIO;
1398 
1399 		if (irq_num < VGIC_NR_PRIVATE_IRQS)
1400 			return -EINVAL;
1401 
1402 		return kvm_vgic_inject_irq(kvm, NULL, irq_num, level, NULL);
1403 	}
1404 
1405 	return -EINVAL;
1406 }
1407 
1408 static unsigned long system_supported_vcpu_features(void)
1409 {
1410 	unsigned long features = KVM_VCPU_VALID_FEATURES;
1411 
1412 	if (!cpus_have_final_cap(ARM64_HAS_32BIT_EL1))
1413 		clear_bit(KVM_ARM_VCPU_EL1_32BIT, &features);
1414 
1415 	if (!kvm_arm_support_pmu_v3())
1416 		clear_bit(KVM_ARM_VCPU_PMU_V3, &features);
1417 
1418 	if (!system_supports_sve())
1419 		clear_bit(KVM_ARM_VCPU_SVE, &features);
1420 
1421 	if (!kvm_has_full_ptr_auth()) {
1422 		clear_bit(KVM_ARM_VCPU_PTRAUTH_ADDRESS, &features);
1423 		clear_bit(KVM_ARM_VCPU_PTRAUTH_GENERIC, &features);
1424 	}
1425 
1426 	if (!cpus_have_final_cap(ARM64_HAS_NESTED_VIRT))
1427 		clear_bit(KVM_ARM_VCPU_HAS_EL2, &features);
1428 
1429 	return features;
1430 }
1431 
1432 static int kvm_vcpu_init_check_features(struct kvm_vcpu *vcpu,
1433 					const struct kvm_vcpu_init *init)
1434 {
1435 	unsigned long features = init->features[0];
1436 	int i;
1437 
1438 	if (features & ~KVM_VCPU_VALID_FEATURES)
1439 		return -ENOENT;
1440 
1441 	for (i = 1; i < ARRAY_SIZE(init->features); i++) {
1442 		if (init->features[i])
1443 			return -ENOENT;
1444 	}
1445 
1446 	if (features & ~system_supported_vcpu_features())
1447 		return -EINVAL;
1448 
1449 	/*
1450 	 * For now make sure that both address/generic pointer authentication
1451 	 * features are requested by the userspace together.
1452 	 */
1453 	if (test_bit(KVM_ARM_VCPU_PTRAUTH_ADDRESS, &features) !=
1454 	    test_bit(KVM_ARM_VCPU_PTRAUTH_GENERIC, &features))
1455 		return -EINVAL;
1456 
1457 	if (!test_bit(KVM_ARM_VCPU_EL1_32BIT, &features))
1458 		return 0;
1459 
1460 	/* MTE is incompatible with AArch32 */
1461 	if (kvm_has_mte(vcpu->kvm))
1462 		return -EINVAL;
1463 
1464 	/* NV is incompatible with AArch32 */
1465 	if (test_bit(KVM_ARM_VCPU_HAS_EL2, &features))
1466 		return -EINVAL;
1467 
1468 	return 0;
1469 }
1470 
1471 static bool kvm_vcpu_init_changed(struct kvm_vcpu *vcpu,
1472 				  const struct kvm_vcpu_init *init)
1473 {
1474 	unsigned long features = init->features[0];
1475 
1476 	return !bitmap_equal(vcpu->kvm->arch.vcpu_features, &features,
1477 			     KVM_VCPU_MAX_FEATURES);
1478 }
1479 
1480 static int kvm_setup_vcpu(struct kvm_vcpu *vcpu)
1481 {
1482 	struct kvm *kvm = vcpu->kvm;
1483 	int ret = 0;
1484 
1485 	/*
1486 	 * When the vCPU has a PMU, but no PMU is set for the guest
1487 	 * yet, set the default one.
1488 	 */
1489 	if (kvm_vcpu_has_pmu(vcpu) && !kvm->arch.arm_pmu)
1490 		ret = kvm_arm_set_default_pmu(kvm);
1491 
1492 	/* Prepare for nested if required */
1493 	if (!ret && vcpu_has_nv(vcpu))
1494 		ret = kvm_vcpu_init_nested(vcpu);
1495 
1496 	return ret;
1497 }
1498 
1499 static int __kvm_vcpu_set_target(struct kvm_vcpu *vcpu,
1500 				 const struct kvm_vcpu_init *init)
1501 {
1502 	unsigned long features = init->features[0];
1503 	struct kvm *kvm = vcpu->kvm;
1504 	int ret = -EINVAL;
1505 
1506 	mutex_lock(&kvm->arch.config_lock);
1507 
1508 	if (test_bit(KVM_ARCH_FLAG_VCPU_FEATURES_CONFIGURED, &kvm->arch.flags) &&
1509 	    kvm_vcpu_init_changed(vcpu, init))
1510 		goto out_unlock;
1511 
1512 	bitmap_copy(kvm->arch.vcpu_features, &features, KVM_VCPU_MAX_FEATURES);
1513 
1514 	ret = kvm_setup_vcpu(vcpu);
1515 	if (ret)
1516 		goto out_unlock;
1517 
1518 	/* Now we know what it is, we can reset it. */
1519 	kvm_reset_vcpu(vcpu);
1520 
1521 	set_bit(KVM_ARCH_FLAG_VCPU_FEATURES_CONFIGURED, &kvm->arch.flags);
1522 	vcpu_set_flag(vcpu, VCPU_INITIALIZED);
1523 	ret = 0;
1524 out_unlock:
1525 	mutex_unlock(&kvm->arch.config_lock);
1526 	return ret;
1527 }
1528 
1529 static int kvm_vcpu_set_target(struct kvm_vcpu *vcpu,
1530 			       const struct kvm_vcpu_init *init)
1531 {
1532 	int ret;
1533 
1534 	if (init->target != KVM_ARM_TARGET_GENERIC_V8 &&
1535 	    init->target != kvm_target_cpu())
1536 		return -EINVAL;
1537 
1538 	ret = kvm_vcpu_init_check_features(vcpu, init);
1539 	if (ret)
1540 		return ret;
1541 
1542 	if (!kvm_vcpu_initialized(vcpu))
1543 		return __kvm_vcpu_set_target(vcpu, init);
1544 
1545 	if (kvm_vcpu_init_changed(vcpu, init))
1546 		return -EINVAL;
1547 
1548 	kvm_reset_vcpu(vcpu);
1549 	return 0;
1550 }
1551 
1552 static int kvm_arch_vcpu_ioctl_vcpu_init(struct kvm_vcpu *vcpu,
1553 					 struct kvm_vcpu_init *init)
1554 {
1555 	bool power_off = false;
1556 	int ret;
1557 
1558 	/*
1559 	 * Treat the power-off vCPU feature as ephemeral. Clear the bit to avoid
1560 	 * reflecting it in the finalized feature set, thus limiting its scope
1561 	 * to a single KVM_ARM_VCPU_INIT call.
1562 	 */
1563 	if (init->features[0] & BIT(KVM_ARM_VCPU_POWER_OFF)) {
1564 		init->features[0] &= ~BIT(KVM_ARM_VCPU_POWER_OFF);
1565 		power_off = true;
1566 	}
1567 
1568 	ret = kvm_vcpu_set_target(vcpu, init);
1569 	if (ret)
1570 		return ret;
1571 
1572 	/*
1573 	 * Ensure a rebooted VM will fault in RAM pages and detect if the
1574 	 * guest MMU is turned off and flush the caches as needed.
1575 	 *
1576 	 * S2FWB enforces all memory accesses to RAM being cacheable,
1577 	 * ensuring that the data side is always coherent. We still
1578 	 * need to invalidate the I-cache though, as FWB does *not*
1579 	 * imply CTR_EL0.DIC.
1580 	 */
1581 	if (vcpu_has_run_once(vcpu)) {
1582 		if (!cpus_have_final_cap(ARM64_HAS_STAGE2_FWB))
1583 			stage2_unmap_vm(vcpu->kvm);
1584 		else
1585 			icache_inval_all_pou();
1586 	}
1587 
1588 	vcpu_reset_hcr(vcpu);
1589 	vcpu->arch.cptr_el2 = kvm_get_reset_cptr_el2(vcpu);
1590 
1591 	/*
1592 	 * Handle the "start in power-off" case.
1593 	 */
1594 	spin_lock(&vcpu->arch.mp_state_lock);
1595 
1596 	if (power_off)
1597 		__kvm_arm_vcpu_power_off(vcpu);
1598 	else
1599 		WRITE_ONCE(vcpu->arch.mp_state.mp_state, KVM_MP_STATE_RUNNABLE);
1600 
1601 	spin_unlock(&vcpu->arch.mp_state_lock);
1602 
1603 	return 0;
1604 }
1605 
1606 static int kvm_arm_vcpu_set_attr(struct kvm_vcpu *vcpu,
1607 				 struct kvm_device_attr *attr)
1608 {
1609 	int ret = -ENXIO;
1610 
1611 	switch (attr->group) {
1612 	default:
1613 		ret = kvm_arm_vcpu_arch_set_attr(vcpu, attr);
1614 		break;
1615 	}
1616 
1617 	return ret;
1618 }
1619 
1620 static int kvm_arm_vcpu_get_attr(struct kvm_vcpu *vcpu,
1621 				 struct kvm_device_attr *attr)
1622 {
1623 	int ret = -ENXIO;
1624 
1625 	switch (attr->group) {
1626 	default:
1627 		ret = kvm_arm_vcpu_arch_get_attr(vcpu, attr);
1628 		break;
1629 	}
1630 
1631 	return ret;
1632 }
1633 
1634 static int kvm_arm_vcpu_has_attr(struct kvm_vcpu *vcpu,
1635 				 struct kvm_device_attr *attr)
1636 {
1637 	int ret = -ENXIO;
1638 
1639 	switch (attr->group) {
1640 	default:
1641 		ret = kvm_arm_vcpu_arch_has_attr(vcpu, attr);
1642 		break;
1643 	}
1644 
1645 	return ret;
1646 }
1647 
1648 static int kvm_arm_vcpu_get_events(struct kvm_vcpu *vcpu,
1649 				   struct kvm_vcpu_events *events)
1650 {
1651 	memset(events, 0, sizeof(*events));
1652 
1653 	return __kvm_arm_vcpu_get_events(vcpu, events);
1654 }
1655 
1656 static int kvm_arm_vcpu_set_events(struct kvm_vcpu *vcpu,
1657 				   struct kvm_vcpu_events *events)
1658 {
1659 	int i;
1660 
1661 	/* check whether the reserved field is zero */
1662 	for (i = 0; i < ARRAY_SIZE(events->reserved); i++)
1663 		if (events->reserved[i])
1664 			return -EINVAL;
1665 
1666 	/* check whether the pad field is zero */
1667 	for (i = 0; i < ARRAY_SIZE(events->exception.pad); i++)
1668 		if (events->exception.pad[i])
1669 			return -EINVAL;
1670 
1671 	return __kvm_arm_vcpu_set_events(vcpu, events);
1672 }
1673 
1674 long kvm_arch_vcpu_ioctl(struct file *filp,
1675 			 unsigned int ioctl, unsigned long arg)
1676 {
1677 	struct kvm_vcpu *vcpu = filp->private_data;
1678 	void __user *argp = (void __user *)arg;
1679 	struct kvm_device_attr attr;
1680 	long r;
1681 
1682 	switch (ioctl) {
1683 	case KVM_ARM_VCPU_INIT: {
1684 		struct kvm_vcpu_init init;
1685 
1686 		r = -EFAULT;
1687 		if (copy_from_user(&init, argp, sizeof(init)))
1688 			break;
1689 
1690 		r = kvm_arch_vcpu_ioctl_vcpu_init(vcpu, &init);
1691 		break;
1692 	}
1693 	case KVM_SET_ONE_REG:
1694 	case KVM_GET_ONE_REG: {
1695 		struct kvm_one_reg reg;
1696 
1697 		r = -ENOEXEC;
1698 		if (unlikely(!kvm_vcpu_initialized(vcpu)))
1699 			break;
1700 
1701 		r = -EFAULT;
1702 		if (copy_from_user(&reg, argp, sizeof(reg)))
1703 			break;
1704 
1705 		/*
1706 		 * We could owe a reset due to PSCI. Handle the pending reset
1707 		 * here to ensure userspace register accesses are ordered after
1708 		 * the reset.
1709 		 */
1710 		if (kvm_check_request(KVM_REQ_VCPU_RESET, vcpu))
1711 			kvm_reset_vcpu(vcpu);
1712 
1713 		if (ioctl == KVM_SET_ONE_REG)
1714 			r = kvm_arm_set_reg(vcpu, &reg);
1715 		else
1716 			r = kvm_arm_get_reg(vcpu, &reg);
1717 		break;
1718 	}
1719 	case KVM_GET_REG_LIST: {
1720 		struct kvm_reg_list __user *user_list = argp;
1721 		struct kvm_reg_list reg_list;
1722 		unsigned n;
1723 
1724 		r = -ENOEXEC;
1725 		if (unlikely(!kvm_vcpu_initialized(vcpu)))
1726 			break;
1727 
1728 		r = -EPERM;
1729 		if (!kvm_arm_vcpu_is_finalized(vcpu))
1730 			break;
1731 
1732 		r = -EFAULT;
1733 		if (copy_from_user(&reg_list, user_list, sizeof(reg_list)))
1734 			break;
1735 		n = reg_list.n;
1736 		reg_list.n = kvm_arm_num_regs(vcpu);
1737 		if (copy_to_user(user_list, &reg_list, sizeof(reg_list)))
1738 			break;
1739 		r = -E2BIG;
1740 		if (n < reg_list.n)
1741 			break;
1742 		r = kvm_arm_copy_reg_indices(vcpu, user_list->reg);
1743 		break;
1744 	}
1745 	case KVM_SET_DEVICE_ATTR: {
1746 		r = -EFAULT;
1747 		if (copy_from_user(&attr, argp, sizeof(attr)))
1748 			break;
1749 		r = kvm_arm_vcpu_set_attr(vcpu, &attr);
1750 		break;
1751 	}
1752 	case KVM_GET_DEVICE_ATTR: {
1753 		r = -EFAULT;
1754 		if (copy_from_user(&attr, argp, sizeof(attr)))
1755 			break;
1756 		r = kvm_arm_vcpu_get_attr(vcpu, &attr);
1757 		break;
1758 	}
1759 	case KVM_HAS_DEVICE_ATTR: {
1760 		r = -EFAULT;
1761 		if (copy_from_user(&attr, argp, sizeof(attr)))
1762 			break;
1763 		r = kvm_arm_vcpu_has_attr(vcpu, &attr);
1764 		break;
1765 	}
1766 	case KVM_GET_VCPU_EVENTS: {
1767 		struct kvm_vcpu_events events;
1768 
1769 		if (kvm_arm_vcpu_get_events(vcpu, &events))
1770 			return -EINVAL;
1771 
1772 		if (copy_to_user(argp, &events, sizeof(events)))
1773 			return -EFAULT;
1774 
1775 		return 0;
1776 	}
1777 	case KVM_SET_VCPU_EVENTS: {
1778 		struct kvm_vcpu_events events;
1779 
1780 		if (copy_from_user(&events, argp, sizeof(events)))
1781 			return -EFAULT;
1782 
1783 		return kvm_arm_vcpu_set_events(vcpu, &events);
1784 	}
1785 	case KVM_ARM_VCPU_FINALIZE: {
1786 		int what;
1787 
1788 		if (!kvm_vcpu_initialized(vcpu))
1789 			return -ENOEXEC;
1790 
1791 		if (get_user(what, (const int __user *)argp))
1792 			return -EFAULT;
1793 
1794 		return kvm_arm_vcpu_finalize(vcpu, what);
1795 	}
1796 	default:
1797 		r = -EINVAL;
1798 	}
1799 
1800 	return r;
1801 }
1802 
1803 void kvm_arch_sync_dirty_log(struct kvm *kvm, struct kvm_memory_slot *memslot)
1804 {
1805 
1806 }
1807 
1808 static int kvm_vm_ioctl_set_device_addr(struct kvm *kvm,
1809 					struct kvm_arm_device_addr *dev_addr)
1810 {
1811 	switch (FIELD_GET(KVM_ARM_DEVICE_ID_MASK, dev_addr->id)) {
1812 	case KVM_ARM_DEVICE_VGIC_V2:
1813 		if (!vgic_present)
1814 			return -ENXIO;
1815 		return kvm_set_legacy_vgic_v2_addr(kvm, dev_addr);
1816 	default:
1817 		return -ENODEV;
1818 	}
1819 }
1820 
1821 static int kvm_vm_has_attr(struct kvm *kvm, struct kvm_device_attr *attr)
1822 {
1823 	switch (attr->group) {
1824 	case KVM_ARM_VM_SMCCC_CTRL:
1825 		return kvm_vm_smccc_has_attr(kvm, attr);
1826 	default:
1827 		return -ENXIO;
1828 	}
1829 }
1830 
1831 static int kvm_vm_set_attr(struct kvm *kvm, struct kvm_device_attr *attr)
1832 {
1833 	switch (attr->group) {
1834 	case KVM_ARM_VM_SMCCC_CTRL:
1835 		return kvm_vm_smccc_set_attr(kvm, attr);
1836 	default:
1837 		return -ENXIO;
1838 	}
1839 }
1840 
1841 int kvm_arch_vm_ioctl(struct file *filp, unsigned int ioctl, unsigned long arg)
1842 {
1843 	struct kvm *kvm = filp->private_data;
1844 	void __user *argp = (void __user *)arg;
1845 	struct kvm_device_attr attr;
1846 
1847 	switch (ioctl) {
1848 	case KVM_CREATE_IRQCHIP: {
1849 		int ret;
1850 		if (!vgic_present)
1851 			return -ENXIO;
1852 		mutex_lock(&kvm->lock);
1853 		ret = kvm_vgic_create(kvm, KVM_DEV_TYPE_ARM_VGIC_V2);
1854 		mutex_unlock(&kvm->lock);
1855 		return ret;
1856 	}
1857 	case KVM_ARM_SET_DEVICE_ADDR: {
1858 		struct kvm_arm_device_addr dev_addr;
1859 
1860 		if (copy_from_user(&dev_addr, argp, sizeof(dev_addr)))
1861 			return -EFAULT;
1862 		return kvm_vm_ioctl_set_device_addr(kvm, &dev_addr);
1863 	}
1864 	case KVM_ARM_PREFERRED_TARGET: {
1865 		struct kvm_vcpu_init init = {
1866 			.target = KVM_ARM_TARGET_GENERIC_V8,
1867 		};
1868 
1869 		if (copy_to_user(argp, &init, sizeof(init)))
1870 			return -EFAULT;
1871 
1872 		return 0;
1873 	}
1874 	case KVM_ARM_MTE_COPY_TAGS: {
1875 		struct kvm_arm_copy_mte_tags copy_tags;
1876 
1877 		if (copy_from_user(&copy_tags, argp, sizeof(copy_tags)))
1878 			return -EFAULT;
1879 		return kvm_vm_ioctl_mte_copy_tags(kvm, &copy_tags);
1880 	}
1881 	case KVM_ARM_SET_COUNTER_OFFSET: {
1882 		struct kvm_arm_counter_offset offset;
1883 
1884 		if (copy_from_user(&offset, argp, sizeof(offset)))
1885 			return -EFAULT;
1886 		return kvm_vm_ioctl_set_counter_offset(kvm, &offset);
1887 	}
1888 	case KVM_HAS_DEVICE_ATTR: {
1889 		if (copy_from_user(&attr, argp, sizeof(attr)))
1890 			return -EFAULT;
1891 
1892 		return kvm_vm_has_attr(kvm, &attr);
1893 	}
1894 	case KVM_SET_DEVICE_ATTR: {
1895 		if (copy_from_user(&attr, argp, sizeof(attr)))
1896 			return -EFAULT;
1897 
1898 		return kvm_vm_set_attr(kvm, &attr);
1899 	}
1900 	case KVM_ARM_GET_REG_WRITABLE_MASKS: {
1901 		struct reg_mask_range range;
1902 
1903 		if (copy_from_user(&range, argp, sizeof(range)))
1904 			return -EFAULT;
1905 		return kvm_vm_ioctl_get_reg_writable_masks(kvm, &range);
1906 	}
1907 	default:
1908 		return -EINVAL;
1909 	}
1910 }
1911 
1912 /* unlocks vcpus from @vcpu_lock_idx and smaller */
1913 static void unlock_vcpus(struct kvm *kvm, int vcpu_lock_idx)
1914 {
1915 	struct kvm_vcpu *tmp_vcpu;
1916 
1917 	for (; vcpu_lock_idx >= 0; vcpu_lock_idx--) {
1918 		tmp_vcpu = kvm_get_vcpu(kvm, vcpu_lock_idx);
1919 		mutex_unlock(&tmp_vcpu->mutex);
1920 	}
1921 }
1922 
1923 void unlock_all_vcpus(struct kvm *kvm)
1924 {
1925 	lockdep_assert_held(&kvm->lock);
1926 
1927 	unlock_vcpus(kvm, atomic_read(&kvm->online_vcpus) - 1);
1928 }
1929 
1930 /* Returns true if all vcpus were locked, false otherwise */
1931 bool lock_all_vcpus(struct kvm *kvm)
1932 {
1933 	struct kvm_vcpu *tmp_vcpu;
1934 	unsigned long c;
1935 
1936 	lockdep_assert_held(&kvm->lock);
1937 
1938 	/*
1939 	 * Any time a vcpu is in an ioctl (including running), the
1940 	 * core KVM code tries to grab the vcpu->mutex.
1941 	 *
1942 	 * By grabbing the vcpu->mutex of all VCPUs we ensure that no
1943 	 * other VCPUs can fiddle with the state while we access it.
1944 	 */
1945 	kvm_for_each_vcpu(c, tmp_vcpu, kvm) {
1946 		if (!mutex_trylock(&tmp_vcpu->mutex)) {
1947 			unlock_vcpus(kvm, c - 1);
1948 			return false;
1949 		}
1950 	}
1951 
1952 	return true;
1953 }
1954 
1955 static unsigned long nvhe_percpu_size(void)
1956 {
1957 	return (unsigned long)CHOOSE_NVHE_SYM(__per_cpu_end) -
1958 		(unsigned long)CHOOSE_NVHE_SYM(__per_cpu_start);
1959 }
1960 
1961 static unsigned long nvhe_percpu_order(void)
1962 {
1963 	unsigned long size = nvhe_percpu_size();
1964 
1965 	return size ? get_order(size) : 0;
1966 }
1967 
1968 static size_t pkvm_host_sve_state_order(void)
1969 {
1970 	return get_order(pkvm_host_sve_state_size());
1971 }
1972 
1973 /* A lookup table holding the hypervisor VA for each vector slot */
1974 static void *hyp_spectre_vector_selector[BP_HARDEN_EL2_SLOTS];
1975 
1976 static void kvm_init_vector_slot(void *base, enum arm64_hyp_spectre_vector slot)
1977 {
1978 	hyp_spectre_vector_selector[slot] = __kvm_vector_slot2addr(base, slot);
1979 }
1980 
1981 static int kvm_init_vector_slots(void)
1982 {
1983 	int err;
1984 	void *base;
1985 
1986 	base = kern_hyp_va(kvm_ksym_ref(__kvm_hyp_vector));
1987 	kvm_init_vector_slot(base, HYP_VECTOR_DIRECT);
1988 
1989 	base = kern_hyp_va(kvm_ksym_ref(__bp_harden_hyp_vecs));
1990 	kvm_init_vector_slot(base, HYP_VECTOR_SPECTRE_DIRECT);
1991 
1992 	if (kvm_system_needs_idmapped_vectors() &&
1993 	    !is_protected_kvm_enabled()) {
1994 		err = create_hyp_exec_mappings(__pa_symbol(__bp_harden_hyp_vecs),
1995 					       __BP_HARDEN_HYP_VECS_SZ, &base);
1996 		if (err)
1997 			return err;
1998 	}
1999 
2000 	kvm_init_vector_slot(base, HYP_VECTOR_INDIRECT);
2001 	kvm_init_vector_slot(base, HYP_VECTOR_SPECTRE_INDIRECT);
2002 	return 0;
2003 }
2004 
2005 static void __init cpu_prepare_hyp_mode(int cpu, u32 hyp_va_bits)
2006 {
2007 	struct kvm_nvhe_init_params *params = per_cpu_ptr_nvhe_sym(kvm_init_params, cpu);
2008 	u64 mmfr0 = read_sanitised_ftr_reg(SYS_ID_AA64MMFR0_EL1);
2009 	unsigned long tcr;
2010 
2011 	/*
2012 	 * Calculate the raw per-cpu offset without a translation from the
2013 	 * kernel's mapping to the linear mapping, and store it in tpidr_el2
2014 	 * so that we can use adr_l to access per-cpu variables in EL2.
2015 	 * Also drop the KASAN tag which gets in the way...
2016 	 */
2017 	params->tpidr_el2 = (unsigned long)kasan_reset_tag(per_cpu_ptr_nvhe_sym(__per_cpu_start, cpu)) -
2018 			    (unsigned long)kvm_ksym_ref(CHOOSE_NVHE_SYM(__per_cpu_start));
2019 
2020 	params->mair_el2 = read_sysreg(mair_el1);
2021 
2022 	tcr = read_sysreg(tcr_el1);
2023 	if (cpus_have_final_cap(ARM64_KVM_HVHE)) {
2024 		tcr |= TCR_EPD1_MASK;
2025 	} else {
2026 		tcr &= TCR_EL2_MASK;
2027 		tcr |= TCR_EL2_RES1;
2028 	}
2029 	tcr &= ~TCR_T0SZ_MASK;
2030 	tcr |= TCR_T0SZ(hyp_va_bits);
2031 	tcr &= ~TCR_EL2_PS_MASK;
2032 	tcr |= FIELD_PREP(TCR_EL2_PS_MASK, kvm_get_parange(mmfr0));
2033 	if (kvm_lpa2_is_enabled())
2034 		tcr |= TCR_EL2_DS;
2035 	params->tcr_el2 = tcr;
2036 
2037 	params->pgd_pa = kvm_mmu_get_httbr();
2038 	if (is_protected_kvm_enabled())
2039 		params->hcr_el2 = HCR_HOST_NVHE_PROTECTED_FLAGS;
2040 	else
2041 		params->hcr_el2 = HCR_HOST_NVHE_FLAGS;
2042 	if (cpus_have_final_cap(ARM64_KVM_HVHE))
2043 		params->hcr_el2 |= HCR_E2H;
2044 	params->vttbr = params->vtcr = 0;
2045 
2046 	/*
2047 	 * Flush the init params from the data cache because the struct will
2048 	 * be read while the MMU is off.
2049 	 */
2050 	kvm_flush_dcache_to_poc(params, sizeof(*params));
2051 }
2052 
2053 static void hyp_install_host_vector(void)
2054 {
2055 	struct kvm_nvhe_init_params *params;
2056 	struct arm_smccc_res res;
2057 
2058 	/* Switch from the HYP stub to our own HYP init vector */
2059 	__hyp_set_vectors(kvm_get_idmap_vector());
2060 
2061 	/*
2062 	 * Call initialization code, and switch to the full blown HYP code.
2063 	 * If the cpucaps haven't been finalized yet, something has gone very
2064 	 * wrong, and hyp will crash and burn when it uses any
2065 	 * cpus_have_*_cap() wrapper.
2066 	 */
2067 	BUG_ON(!system_capabilities_finalized());
2068 	params = this_cpu_ptr_nvhe_sym(kvm_init_params);
2069 	arm_smccc_1_1_hvc(KVM_HOST_SMCCC_FUNC(__kvm_hyp_init), virt_to_phys(params), &res);
2070 	WARN_ON(res.a0 != SMCCC_RET_SUCCESS);
2071 }
2072 
2073 static void cpu_init_hyp_mode(void)
2074 {
2075 	hyp_install_host_vector();
2076 
2077 	/*
2078 	 * Disabling SSBD on a non-VHE system requires us to enable SSBS
2079 	 * at EL2.
2080 	 */
2081 	if (this_cpu_has_cap(ARM64_SSBS) &&
2082 	    arm64_get_spectre_v4_state() == SPECTRE_VULNERABLE) {
2083 		kvm_call_hyp_nvhe(__kvm_enable_ssbs);
2084 	}
2085 }
2086 
2087 static void cpu_hyp_reset(void)
2088 {
2089 	if (!is_kernel_in_hyp_mode())
2090 		__hyp_reset_vectors();
2091 }
2092 
2093 /*
2094  * EL2 vectors can be mapped and rerouted in a number of ways,
2095  * depending on the kernel configuration and CPU present:
2096  *
2097  * - If the CPU is affected by Spectre-v2, the hardening sequence is
2098  *   placed in one of the vector slots, which is executed before jumping
2099  *   to the real vectors.
2100  *
2101  * - If the CPU also has the ARM64_SPECTRE_V3A cap, the slot
2102  *   containing the hardening sequence is mapped next to the idmap page,
2103  *   and executed before jumping to the real vectors.
2104  *
2105  * - If the CPU only has the ARM64_SPECTRE_V3A cap, then an
2106  *   empty slot is selected, mapped next to the idmap page, and
2107  *   executed before jumping to the real vectors.
2108  *
2109  * Note that ARM64_SPECTRE_V3A is somewhat incompatible with
2110  * VHE, as we don't have hypervisor-specific mappings. If the system
2111  * is VHE and yet selects this capability, it will be ignored.
2112  */
2113 static void cpu_set_hyp_vector(void)
2114 {
2115 	struct bp_hardening_data *data = this_cpu_ptr(&bp_hardening_data);
2116 	void *vector = hyp_spectre_vector_selector[data->slot];
2117 
2118 	if (!is_protected_kvm_enabled())
2119 		*this_cpu_ptr_hyp_sym(kvm_hyp_vector) = (unsigned long)vector;
2120 	else
2121 		kvm_call_hyp_nvhe(__pkvm_cpu_set_vector, data->slot);
2122 }
2123 
2124 static void cpu_hyp_init_context(void)
2125 {
2126 	kvm_init_host_cpu_context(host_data_ptr(host_ctxt));
2127 
2128 	if (!is_kernel_in_hyp_mode())
2129 		cpu_init_hyp_mode();
2130 }
2131 
2132 static void cpu_hyp_init_features(void)
2133 {
2134 	cpu_set_hyp_vector();
2135 	kvm_arm_init_debug();
2136 
2137 	if (is_kernel_in_hyp_mode())
2138 		kvm_timer_init_vhe();
2139 
2140 	if (vgic_present)
2141 		kvm_vgic_init_cpu_hardware();
2142 }
2143 
2144 static void cpu_hyp_reinit(void)
2145 {
2146 	cpu_hyp_reset();
2147 	cpu_hyp_init_context();
2148 	cpu_hyp_init_features();
2149 }
2150 
2151 static void cpu_hyp_init(void *discard)
2152 {
2153 	if (!__this_cpu_read(kvm_hyp_initialized)) {
2154 		cpu_hyp_reinit();
2155 		__this_cpu_write(kvm_hyp_initialized, 1);
2156 	}
2157 }
2158 
2159 static void cpu_hyp_uninit(void *discard)
2160 {
2161 	if (__this_cpu_read(kvm_hyp_initialized)) {
2162 		cpu_hyp_reset();
2163 		__this_cpu_write(kvm_hyp_initialized, 0);
2164 	}
2165 }
2166 
2167 int kvm_arch_enable_virtualization_cpu(void)
2168 {
2169 	/*
2170 	 * Most calls to this function are made with migration
2171 	 * disabled, but not with preemption disabled. The former is
2172 	 * enough to ensure correctness, but most of the helpers
2173 	 * expect the later and will throw a tantrum otherwise.
2174 	 */
2175 	preempt_disable();
2176 
2177 	cpu_hyp_init(NULL);
2178 
2179 	kvm_vgic_cpu_up();
2180 	kvm_timer_cpu_up();
2181 
2182 	preempt_enable();
2183 
2184 	return 0;
2185 }
2186 
2187 void kvm_arch_disable_virtualization_cpu(void)
2188 {
2189 	kvm_timer_cpu_down();
2190 	kvm_vgic_cpu_down();
2191 
2192 	if (!is_protected_kvm_enabled())
2193 		cpu_hyp_uninit(NULL);
2194 }
2195 
2196 #ifdef CONFIG_CPU_PM
2197 static int hyp_init_cpu_pm_notifier(struct notifier_block *self,
2198 				    unsigned long cmd,
2199 				    void *v)
2200 {
2201 	/*
2202 	 * kvm_hyp_initialized is left with its old value over
2203 	 * PM_ENTER->PM_EXIT. It is used to indicate PM_EXIT should
2204 	 * re-enable hyp.
2205 	 */
2206 	switch (cmd) {
2207 	case CPU_PM_ENTER:
2208 		if (__this_cpu_read(kvm_hyp_initialized))
2209 			/*
2210 			 * don't update kvm_hyp_initialized here
2211 			 * so that the hyp will be re-enabled
2212 			 * when we resume. See below.
2213 			 */
2214 			cpu_hyp_reset();
2215 
2216 		return NOTIFY_OK;
2217 	case CPU_PM_ENTER_FAILED:
2218 	case CPU_PM_EXIT:
2219 		if (__this_cpu_read(kvm_hyp_initialized))
2220 			/* The hyp was enabled before suspend. */
2221 			cpu_hyp_reinit();
2222 
2223 		return NOTIFY_OK;
2224 
2225 	default:
2226 		return NOTIFY_DONE;
2227 	}
2228 }
2229 
2230 static struct notifier_block hyp_init_cpu_pm_nb = {
2231 	.notifier_call = hyp_init_cpu_pm_notifier,
2232 };
2233 
2234 static void __init hyp_cpu_pm_init(void)
2235 {
2236 	if (!is_protected_kvm_enabled())
2237 		cpu_pm_register_notifier(&hyp_init_cpu_pm_nb);
2238 }
2239 static void __init hyp_cpu_pm_exit(void)
2240 {
2241 	if (!is_protected_kvm_enabled())
2242 		cpu_pm_unregister_notifier(&hyp_init_cpu_pm_nb);
2243 }
2244 #else
2245 static inline void __init hyp_cpu_pm_init(void)
2246 {
2247 }
2248 static inline void __init hyp_cpu_pm_exit(void)
2249 {
2250 }
2251 #endif
2252 
2253 static void __init init_cpu_logical_map(void)
2254 {
2255 	unsigned int cpu;
2256 
2257 	/*
2258 	 * Copy the MPIDR <-> logical CPU ID mapping to hyp.
2259 	 * Only copy the set of online CPUs whose features have been checked
2260 	 * against the finalized system capabilities. The hypervisor will not
2261 	 * allow any other CPUs from the `possible` set to boot.
2262 	 */
2263 	for_each_online_cpu(cpu)
2264 		hyp_cpu_logical_map[cpu] = cpu_logical_map(cpu);
2265 }
2266 
2267 #define init_psci_0_1_impl_state(config, what)	\
2268 	config.psci_0_1_ ## what ## _implemented = psci_ops.what
2269 
2270 static bool __init init_psci_relay(void)
2271 {
2272 	/*
2273 	 * If PSCI has not been initialized, protected KVM cannot install
2274 	 * itself on newly booted CPUs.
2275 	 */
2276 	if (!psci_ops.get_version) {
2277 		kvm_err("Cannot initialize protected mode without PSCI\n");
2278 		return false;
2279 	}
2280 
2281 	kvm_host_psci_config.version = psci_ops.get_version();
2282 	kvm_host_psci_config.smccc_version = arm_smccc_get_version();
2283 
2284 	if (kvm_host_psci_config.version == PSCI_VERSION(0, 1)) {
2285 		kvm_host_psci_config.function_ids_0_1 = get_psci_0_1_function_ids();
2286 		init_psci_0_1_impl_state(kvm_host_psci_config, cpu_suspend);
2287 		init_psci_0_1_impl_state(kvm_host_psci_config, cpu_on);
2288 		init_psci_0_1_impl_state(kvm_host_psci_config, cpu_off);
2289 		init_psci_0_1_impl_state(kvm_host_psci_config, migrate);
2290 	}
2291 	return true;
2292 }
2293 
2294 static int __init init_subsystems(void)
2295 {
2296 	int err = 0;
2297 
2298 	/*
2299 	 * Enable hardware so that subsystem initialisation can access EL2.
2300 	 */
2301 	on_each_cpu(cpu_hyp_init, NULL, 1);
2302 
2303 	/*
2304 	 * Register CPU lower-power notifier
2305 	 */
2306 	hyp_cpu_pm_init();
2307 
2308 	/*
2309 	 * Init HYP view of VGIC
2310 	 */
2311 	err = kvm_vgic_hyp_init();
2312 	switch (err) {
2313 	case 0:
2314 		vgic_present = true;
2315 		break;
2316 	case -ENODEV:
2317 	case -ENXIO:
2318 		vgic_present = false;
2319 		err = 0;
2320 		break;
2321 	default:
2322 		goto out;
2323 	}
2324 
2325 	/*
2326 	 * Init HYP architected timer support
2327 	 */
2328 	err = kvm_timer_hyp_init(vgic_present);
2329 	if (err)
2330 		goto out;
2331 
2332 	kvm_register_perf_callbacks(NULL);
2333 
2334 out:
2335 	if (err)
2336 		hyp_cpu_pm_exit();
2337 
2338 	if (err || !is_protected_kvm_enabled())
2339 		on_each_cpu(cpu_hyp_uninit, NULL, 1);
2340 
2341 	return err;
2342 }
2343 
2344 static void __init teardown_subsystems(void)
2345 {
2346 	kvm_unregister_perf_callbacks();
2347 	hyp_cpu_pm_exit();
2348 }
2349 
2350 static void __init teardown_hyp_mode(void)
2351 {
2352 	bool free_sve = system_supports_sve() && is_protected_kvm_enabled();
2353 	int cpu;
2354 
2355 	free_hyp_pgds();
2356 	for_each_possible_cpu(cpu) {
2357 		free_page(per_cpu(kvm_arm_hyp_stack_page, cpu));
2358 		free_pages(kvm_nvhe_sym(kvm_arm_hyp_percpu_base)[cpu], nvhe_percpu_order());
2359 
2360 		if (free_sve) {
2361 			struct cpu_sve_state *sve_state;
2362 
2363 			sve_state = per_cpu_ptr_nvhe_sym(kvm_host_data, cpu)->sve_state;
2364 			free_pages((unsigned long) sve_state, pkvm_host_sve_state_order());
2365 		}
2366 	}
2367 }
2368 
2369 static int __init do_pkvm_init(u32 hyp_va_bits)
2370 {
2371 	void *per_cpu_base = kvm_ksym_ref(kvm_nvhe_sym(kvm_arm_hyp_percpu_base));
2372 	int ret;
2373 
2374 	preempt_disable();
2375 	cpu_hyp_init_context();
2376 	ret = kvm_call_hyp_nvhe(__pkvm_init, hyp_mem_base, hyp_mem_size,
2377 				num_possible_cpus(), kern_hyp_va(per_cpu_base),
2378 				hyp_va_bits);
2379 	cpu_hyp_init_features();
2380 
2381 	/*
2382 	 * The stub hypercalls are now disabled, so set our local flag to
2383 	 * prevent a later re-init attempt in kvm_arch_enable_virtualization_cpu().
2384 	 */
2385 	__this_cpu_write(kvm_hyp_initialized, 1);
2386 	preempt_enable();
2387 
2388 	return ret;
2389 }
2390 
2391 static u64 get_hyp_id_aa64pfr0_el1(void)
2392 {
2393 	/*
2394 	 * Track whether the system isn't affected by spectre/meltdown in the
2395 	 * hypervisor's view of id_aa64pfr0_el1, used for protected VMs.
2396 	 * Although this is per-CPU, we make it global for simplicity, e.g., not
2397 	 * to have to worry about vcpu migration.
2398 	 *
2399 	 * Unlike for non-protected VMs, userspace cannot override this for
2400 	 * protected VMs.
2401 	 */
2402 	u64 val = read_sanitised_ftr_reg(SYS_ID_AA64PFR0_EL1);
2403 
2404 	val &= ~(ARM64_FEATURE_MASK(ID_AA64PFR0_EL1_CSV2) |
2405 		 ARM64_FEATURE_MASK(ID_AA64PFR0_EL1_CSV3));
2406 
2407 	val |= FIELD_PREP(ARM64_FEATURE_MASK(ID_AA64PFR0_EL1_CSV2),
2408 			  arm64_get_spectre_v2_state() == SPECTRE_UNAFFECTED);
2409 	val |= FIELD_PREP(ARM64_FEATURE_MASK(ID_AA64PFR0_EL1_CSV3),
2410 			  arm64_get_meltdown_state() == SPECTRE_UNAFFECTED);
2411 
2412 	return val;
2413 }
2414 
2415 static void kvm_hyp_init_symbols(void)
2416 {
2417 	kvm_nvhe_sym(id_aa64pfr0_el1_sys_val) = get_hyp_id_aa64pfr0_el1();
2418 	kvm_nvhe_sym(id_aa64pfr1_el1_sys_val) = read_sanitised_ftr_reg(SYS_ID_AA64PFR1_EL1);
2419 	kvm_nvhe_sym(id_aa64isar0_el1_sys_val) = read_sanitised_ftr_reg(SYS_ID_AA64ISAR0_EL1);
2420 	kvm_nvhe_sym(id_aa64isar1_el1_sys_val) = read_sanitised_ftr_reg(SYS_ID_AA64ISAR1_EL1);
2421 	kvm_nvhe_sym(id_aa64isar2_el1_sys_val) = read_sanitised_ftr_reg(SYS_ID_AA64ISAR2_EL1);
2422 	kvm_nvhe_sym(id_aa64mmfr0_el1_sys_val) = read_sanitised_ftr_reg(SYS_ID_AA64MMFR0_EL1);
2423 	kvm_nvhe_sym(id_aa64mmfr1_el1_sys_val) = read_sanitised_ftr_reg(SYS_ID_AA64MMFR1_EL1);
2424 	kvm_nvhe_sym(id_aa64mmfr2_el1_sys_val) = read_sanitised_ftr_reg(SYS_ID_AA64MMFR2_EL1);
2425 	kvm_nvhe_sym(id_aa64smfr0_el1_sys_val) = read_sanitised_ftr_reg(SYS_ID_AA64SMFR0_EL1);
2426 	kvm_nvhe_sym(__icache_flags) = __icache_flags;
2427 	kvm_nvhe_sym(kvm_arm_vmid_bits) = kvm_arm_vmid_bits;
2428 }
2429 
2430 static int __init kvm_hyp_init_protection(u32 hyp_va_bits)
2431 {
2432 	void *addr = phys_to_virt(hyp_mem_base);
2433 	int ret;
2434 
2435 	ret = create_hyp_mappings(addr, addr + hyp_mem_size, PAGE_HYP);
2436 	if (ret)
2437 		return ret;
2438 
2439 	ret = do_pkvm_init(hyp_va_bits);
2440 	if (ret)
2441 		return ret;
2442 
2443 	free_hyp_pgds();
2444 
2445 	return 0;
2446 }
2447 
2448 static int init_pkvm_host_sve_state(void)
2449 {
2450 	int cpu;
2451 
2452 	if (!system_supports_sve())
2453 		return 0;
2454 
2455 	/* Allocate pages for host sve state in protected mode. */
2456 	for_each_possible_cpu(cpu) {
2457 		struct page *page = alloc_pages(GFP_KERNEL, pkvm_host_sve_state_order());
2458 
2459 		if (!page)
2460 			return -ENOMEM;
2461 
2462 		per_cpu_ptr_nvhe_sym(kvm_host_data, cpu)->sve_state = page_address(page);
2463 	}
2464 
2465 	/*
2466 	 * Don't map the pages in hyp since these are only used in protected
2467 	 * mode, which will (re)create its own mapping when initialized.
2468 	 */
2469 
2470 	return 0;
2471 }
2472 
2473 /*
2474  * Finalizes the initialization of hyp mode, once everything else is initialized
2475  * and the initialziation process cannot fail.
2476  */
2477 static void finalize_init_hyp_mode(void)
2478 {
2479 	int cpu;
2480 
2481 	if (system_supports_sve() && is_protected_kvm_enabled()) {
2482 		for_each_possible_cpu(cpu) {
2483 			struct cpu_sve_state *sve_state;
2484 
2485 			sve_state = per_cpu_ptr_nvhe_sym(kvm_host_data, cpu)->sve_state;
2486 			per_cpu_ptr_nvhe_sym(kvm_host_data, cpu)->sve_state =
2487 				kern_hyp_va(sve_state);
2488 		}
2489 	} else {
2490 		for_each_possible_cpu(cpu) {
2491 			struct user_fpsimd_state *fpsimd_state;
2492 
2493 			fpsimd_state = &per_cpu_ptr_nvhe_sym(kvm_host_data, cpu)->host_ctxt.fp_regs;
2494 			per_cpu_ptr_nvhe_sym(kvm_host_data, cpu)->fpsimd_state =
2495 				kern_hyp_va(fpsimd_state);
2496 		}
2497 	}
2498 }
2499 
2500 static void pkvm_hyp_init_ptrauth(void)
2501 {
2502 	struct kvm_cpu_context *hyp_ctxt;
2503 	int cpu;
2504 
2505 	for_each_possible_cpu(cpu) {
2506 		hyp_ctxt = per_cpu_ptr_nvhe_sym(kvm_hyp_ctxt, cpu);
2507 		hyp_ctxt->sys_regs[APIAKEYLO_EL1] = get_random_long();
2508 		hyp_ctxt->sys_regs[APIAKEYHI_EL1] = get_random_long();
2509 		hyp_ctxt->sys_regs[APIBKEYLO_EL1] = get_random_long();
2510 		hyp_ctxt->sys_regs[APIBKEYHI_EL1] = get_random_long();
2511 		hyp_ctxt->sys_regs[APDAKEYLO_EL1] = get_random_long();
2512 		hyp_ctxt->sys_regs[APDAKEYHI_EL1] = get_random_long();
2513 		hyp_ctxt->sys_regs[APDBKEYLO_EL1] = get_random_long();
2514 		hyp_ctxt->sys_regs[APDBKEYHI_EL1] = get_random_long();
2515 		hyp_ctxt->sys_regs[APGAKEYLO_EL1] = get_random_long();
2516 		hyp_ctxt->sys_regs[APGAKEYHI_EL1] = get_random_long();
2517 	}
2518 }
2519 
2520 /* Inits Hyp-mode on all online CPUs */
2521 static int __init init_hyp_mode(void)
2522 {
2523 	u32 hyp_va_bits;
2524 	int cpu;
2525 	int err = -ENOMEM;
2526 
2527 	/*
2528 	 * The protected Hyp-mode cannot be initialized if the memory pool
2529 	 * allocation has failed.
2530 	 */
2531 	if (is_protected_kvm_enabled() && !hyp_mem_base)
2532 		goto out_err;
2533 
2534 	/*
2535 	 * Allocate Hyp PGD and setup Hyp identity mapping
2536 	 */
2537 	err = kvm_mmu_init(&hyp_va_bits);
2538 	if (err)
2539 		goto out_err;
2540 
2541 	/*
2542 	 * Allocate stack pages for Hypervisor-mode
2543 	 */
2544 	for_each_possible_cpu(cpu) {
2545 		unsigned long stack_page;
2546 
2547 		stack_page = __get_free_page(GFP_KERNEL);
2548 		if (!stack_page) {
2549 			err = -ENOMEM;
2550 			goto out_err;
2551 		}
2552 
2553 		per_cpu(kvm_arm_hyp_stack_page, cpu) = stack_page;
2554 	}
2555 
2556 	/*
2557 	 * Allocate and initialize pages for Hypervisor-mode percpu regions.
2558 	 */
2559 	for_each_possible_cpu(cpu) {
2560 		struct page *page;
2561 		void *page_addr;
2562 
2563 		page = alloc_pages(GFP_KERNEL, nvhe_percpu_order());
2564 		if (!page) {
2565 			err = -ENOMEM;
2566 			goto out_err;
2567 		}
2568 
2569 		page_addr = page_address(page);
2570 		memcpy(page_addr, CHOOSE_NVHE_SYM(__per_cpu_start), nvhe_percpu_size());
2571 		kvm_nvhe_sym(kvm_arm_hyp_percpu_base)[cpu] = (unsigned long)page_addr;
2572 	}
2573 
2574 	/*
2575 	 * Map the Hyp-code called directly from the host
2576 	 */
2577 	err = create_hyp_mappings(kvm_ksym_ref(__hyp_text_start),
2578 				  kvm_ksym_ref(__hyp_text_end), PAGE_HYP_EXEC);
2579 	if (err) {
2580 		kvm_err("Cannot map world-switch code\n");
2581 		goto out_err;
2582 	}
2583 
2584 	err = create_hyp_mappings(kvm_ksym_ref(__hyp_rodata_start),
2585 				  kvm_ksym_ref(__hyp_rodata_end), PAGE_HYP_RO);
2586 	if (err) {
2587 		kvm_err("Cannot map .hyp.rodata section\n");
2588 		goto out_err;
2589 	}
2590 
2591 	err = create_hyp_mappings(kvm_ksym_ref(__start_rodata),
2592 				  kvm_ksym_ref(__end_rodata), PAGE_HYP_RO);
2593 	if (err) {
2594 		kvm_err("Cannot map rodata section\n");
2595 		goto out_err;
2596 	}
2597 
2598 	/*
2599 	 * .hyp.bss is guaranteed to be placed at the beginning of the .bss
2600 	 * section thanks to an assertion in the linker script. Map it RW and
2601 	 * the rest of .bss RO.
2602 	 */
2603 	err = create_hyp_mappings(kvm_ksym_ref(__hyp_bss_start),
2604 				  kvm_ksym_ref(__hyp_bss_end), PAGE_HYP);
2605 	if (err) {
2606 		kvm_err("Cannot map hyp bss section: %d\n", err);
2607 		goto out_err;
2608 	}
2609 
2610 	err = create_hyp_mappings(kvm_ksym_ref(__hyp_bss_end),
2611 				  kvm_ksym_ref(__bss_stop), PAGE_HYP_RO);
2612 	if (err) {
2613 		kvm_err("Cannot map bss section\n");
2614 		goto out_err;
2615 	}
2616 
2617 	/*
2618 	 * Map the Hyp stack pages
2619 	 */
2620 	for_each_possible_cpu(cpu) {
2621 		struct kvm_nvhe_init_params *params = per_cpu_ptr_nvhe_sym(kvm_init_params, cpu);
2622 		char *stack_page = (char *)per_cpu(kvm_arm_hyp_stack_page, cpu);
2623 
2624 		err = create_hyp_stack(__pa(stack_page), &params->stack_hyp_va);
2625 		if (err) {
2626 			kvm_err("Cannot map hyp stack\n");
2627 			goto out_err;
2628 		}
2629 
2630 		/*
2631 		 * Save the stack PA in nvhe_init_params. This will be needed
2632 		 * to recreate the stack mapping in protected nVHE mode.
2633 		 * __hyp_pa() won't do the right thing there, since the stack
2634 		 * has been mapped in the flexible private VA space.
2635 		 */
2636 		params->stack_pa = __pa(stack_page);
2637 	}
2638 
2639 	for_each_possible_cpu(cpu) {
2640 		char *percpu_begin = (char *)kvm_nvhe_sym(kvm_arm_hyp_percpu_base)[cpu];
2641 		char *percpu_end = percpu_begin + nvhe_percpu_size();
2642 
2643 		/* Map Hyp percpu pages */
2644 		err = create_hyp_mappings(percpu_begin, percpu_end, PAGE_HYP);
2645 		if (err) {
2646 			kvm_err("Cannot map hyp percpu region\n");
2647 			goto out_err;
2648 		}
2649 
2650 		/* Prepare the CPU initialization parameters */
2651 		cpu_prepare_hyp_mode(cpu, hyp_va_bits);
2652 	}
2653 
2654 	kvm_hyp_init_symbols();
2655 
2656 	if (is_protected_kvm_enabled()) {
2657 		if (IS_ENABLED(CONFIG_ARM64_PTR_AUTH_KERNEL) &&
2658 		    cpus_have_final_cap(ARM64_HAS_ADDRESS_AUTH))
2659 			pkvm_hyp_init_ptrauth();
2660 
2661 		init_cpu_logical_map();
2662 
2663 		if (!init_psci_relay()) {
2664 			err = -ENODEV;
2665 			goto out_err;
2666 		}
2667 
2668 		err = init_pkvm_host_sve_state();
2669 		if (err)
2670 			goto out_err;
2671 
2672 		err = kvm_hyp_init_protection(hyp_va_bits);
2673 		if (err) {
2674 			kvm_err("Failed to init hyp memory protection\n");
2675 			goto out_err;
2676 		}
2677 	}
2678 
2679 	return 0;
2680 
2681 out_err:
2682 	teardown_hyp_mode();
2683 	kvm_err("error initializing Hyp mode: %d\n", err);
2684 	return err;
2685 }
2686 
2687 struct kvm_vcpu *kvm_mpidr_to_vcpu(struct kvm *kvm, unsigned long mpidr)
2688 {
2689 	struct kvm_vcpu *vcpu = NULL;
2690 	struct kvm_mpidr_data *data;
2691 	unsigned long i;
2692 
2693 	mpidr &= MPIDR_HWID_BITMASK;
2694 
2695 	rcu_read_lock();
2696 	data = rcu_dereference(kvm->arch.mpidr_data);
2697 
2698 	if (data) {
2699 		u16 idx = kvm_mpidr_index(data, mpidr);
2700 
2701 		vcpu = kvm_get_vcpu(kvm, data->cmpidr_to_idx[idx]);
2702 		if (mpidr != kvm_vcpu_get_mpidr_aff(vcpu))
2703 			vcpu = NULL;
2704 	}
2705 
2706 	rcu_read_unlock();
2707 
2708 	if (vcpu)
2709 		return vcpu;
2710 
2711 	kvm_for_each_vcpu(i, vcpu, kvm) {
2712 		if (mpidr == kvm_vcpu_get_mpidr_aff(vcpu))
2713 			return vcpu;
2714 	}
2715 	return NULL;
2716 }
2717 
2718 bool kvm_arch_irqchip_in_kernel(struct kvm *kvm)
2719 {
2720 	return irqchip_in_kernel(kvm);
2721 }
2722 
2723 bool kvm_arch_has_irq_bypass(void)
2724 {
2725 	return true;
2726 }
2727 
2728 int kvm_arch_irq_bypass_add_producer(struct irq_bypass_consumer *cons,
2729 				      struct irq_bypass_producer *prod)
2730 {
2731 	struct kvm_kernel_irqfd *irqfd =
2732 		container_of(cons, struct kvm_kernel_irqfd, consumer);
2733 
2734 	return kvm_vgic_v4_set_forwarding(irqfd->kvm, prod->irq,
2735 					  &irqfd->irq_entry);
2736 }
2737 void kvm_arch_irq_bypass_del_producer(struct irq_bypass_consumer *cons,
2738 				      struct irq_bypass_producer *prod)
2739 {
2740 	struct kvm_kernel_irqfd *irqfd =
2741 		container_of(cons, struct kvm_kernel_irqfd, consumer);
2742 
2743 	kvm_vgic_v4_unset_forwarding(irqfd->kvm, prod->irq,
2744 				     &irqfd->irq_entry);
2745 }
2746 
2747 void kvm_arch_irq_bypass_stop(struct irq_bypass_consumer *cons)
2748 {
2749 	struct kvm_kernel_irqfd *irqfd =
2750 		container_of(cons, struct kvm_kernel_irqfd, consumer);
2751 
2752 	kvm_arm_halt_guest(irqfd->kvm);
2753 }
2754 
2755 void kvm_arch_irq_bypass_start(struct irq_bypass_consumer *cons)
2756 {
2757 	struct kvm_kernel_irqfd *irqfd =
2758 		container_of(cons, struct kvm_kernel_irqfd, consumer);
2759 
2760 	kvm_arm_resume_guest(irqfd->kvm);
2761 }
2762 
2763 /* Initialize Hyp-mode and memory mappings on all CPUs */
2764 static __init int kvm_arm_init(void)
2765 {
2766 	int err;
2767 	bool in_hyp_mode;
2768 
2769 	if (!is_hyp_mode_available()) {
2770 		kvm_info("HYP mode not available\n");
2771 		return -ENODEV;
2772 	}
2773 
2774 	if (kvm_get_mode() == KVM_MODE_NONE) {
2775 		kvm_info("KVM disabled from command line\n");
2776 		return -ENODEV;
2777 	}
2778 
2779 	err = kvm_sys_reg_table_init();
2780 	if (err) {
2781 		kvm_info("Error initializing system register tables");
2782 		return err;
2783 	}
2784 
2785 	in_hyp_mode = is_kernel_in_hyp_mode();
2786 
2787 	if (cpus_have_final_cap(ARM64_WORKAROUND_DEVICE_LOAD_ACQUIRE) ||
2788 	    cpus_have_final_cap(ARM64_WORKAROUND_1508412))
2789 		kvm_info("Guests without required CPU erratum workarounds can deadlock system!\n" \
2790 			 "Only trusted guests should be used on this system.\n");
2791 
2792 	err = kvm_set_ipa_limit();
2793 	if (err)
2794 		return err;
2795 
2796 	err = kvm_arm_init_sve();
2797 	if (err)
2798 		return err;
2799 
2800 	err = kvm_arm_vmid_alloc_init();
2801 	if (err) {
2802 		kvm_err("Failed to initialize VMID allocator.\n");
2803 		return err;
2804 	}
2805 
2806 	if (!in_hyp_mode) {
2807 		err = init_hyp_mode();
2808 		if (err)
2809 			goto out_err;
2810 	}
2811 
2812 	err = kvm_init_vector_slots();
2813 	if (err) {
2814 		kvm_err("Cannot initialise vector slots\n");
2815 		goto out_hyp;
2816 	}
2817 
2818 	err = init_subsystems();
2819 	if (err)
2820 		goto out_hyp;
2821 
2822 	kvm_info("%s%sVHE mode initialized successfully\n",
2823 		 in_hyp_mode ? "" : (is_protected_kvm_enabled() ?
2824 				     "Protected " : "Hyp "),
2825 		 in_hyp_mode ? "" : (cpus_have_final_cap(ARM64_KVM_HVHE) ?
2826 				     "h" : "n"));
2827 
2828 	/*
2829 	 * FIXME: Do something reasonable if kvm_init() fails after pKVM
2830 	 * hypervisor protection is finalized.
2831 	 */
2832 	err = kvm_init(sizeof(struct kvm_vcpu), 0, THIS_MODULE);
2833 	if (err)
2834 		goto out_subs;
2835 
2836 	/*
2837 	 * This should be called after initialization is done and failure isn't
2838 	 * possible anymore.
2839 	 */
2840 	if (!in_hyp_mode)
2841 		finalize_init_hyp_mode();
2842 
2843 	kvm_arm_initialised = true;
2844 
2845 	return 0;
2846 
2847 out_subs:
2848 	teardown_subsystems();
2849 out_hyp:
2850 	if (!in_hyp_mode)
2851 		teardown_hyp_mode();
2852 out_err:
2853 	kvm_arm_vmid_alloc_free();
2854 	return err;
2855 }
2856 
2857 static int __init early_kvm_mode_cfg(char *arg)
2858 {
2859 	if (!arg)
2860 		return -EINVAL;
2861 
2862 	if (strcmp(arg, "none") == 0) {
2863 		kvm_mode = KVM_MODE_NONE;
2864 		return 0;
2865 	}
2866 
2867 	if (!is_hyp_mode_available()) {
2868 		pr_warn_once("KVM is not available. Ignoring kvm-arm.mode\n");
2869 		return 0;
2870 	}
2871 
2872 	if (strcmp(arg, "protected") == 0) {
2873 		if (!is_kernel_in_hyp_mode())
2874 			kvm_mode = KVM_MODE_PROTECTED;
2875 		else
2876 			pr_warn_once("Protected KVM not available with VHE\n");
2877 
2878 		return 0;
2879 	}
2880 
2881 	if (strcmp(arg, "nvhe") == 0 && !WARN_ON(is_kernel_in_hyp_mode())) {
2882 		kvm_mode = KVM_MODE_DEFAULT;
2883 		return 0;
2884 	}
2885 
2886 	if (strcmp(arg, "nested") == 0 && !WARN_ON(!is_kernel_in_hyp_mode())) {
2887 		kvm_mode = KVM_MODE_NV;
2888 		return 0;
2889 	}
2890 
2891 	return -EINVAL;
2892 }
2893 early_param("kvm-arm.mode", early_kvm_mode_cfg);
2894 
2895 static int __init early_kvm_wfx_trap_policy_cfg(char *arg, enum kvm_wfx_trap_policy *p)
2896 {
2897 	if (!arg)
2898 		return -EINVAL;
2899 
2900 	if (strcmp(arg, "trap") == 0) {
2901 		*p = KVM_WFX_TRAP;
2902 		return 0;
2903 	}
2904 
2905 	if (strcmp(arg, "notrap") == 0) {
2906 		*p = KVM_WFX_NOTRAP;
2907 		return 0;
2908 	}
2909 
2910 	return -EINVAL;
2911 }
2912 
2913 static int __init early_kvm_wfi_trap_policy_cfg(char *arg)
2914 {
2915 	return early_kvm_wfx_trap_policy_cfg(arg, &kvm_wfi_trap_policy);
2916 }
2917 early_param("kvm-arm.wfi_trap_policy", early_kvm_wfi_trap_policy_cfg);
2918 
2919 static int __init early_kvm_wfe_trap_policy_cfg(char *arg)
2920 {
2921 	return early_kvm_wfx_trap_policy_cfg(arg, &kvm_wfe_trap_policy);
2922 }
2923 early_param("kvm-arm.wfe_trap_policy", early_kvm_wfe_trap_policy_cfg);
2924 
2925 enum kvm_mode kvm_get_mode(void)
2926 {
2927 	return kvm_mode;
2928 }
2929 
2930 module_init(kvm_arm_init);
2931