xref: /linux/arch/arm64/kvm/arm.c (revision c060f8168bdf22aa986970955af99702d142dfbe)
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_VM_DEAD, vcpu))
1001 			return -EIO;
1002 
1003 		if (kvm_check_request(KVM_REQ_SLEEP, vcpu))
1004 			kvm_vcpu_sleep(vcpu);
1005 
1006 		if (kvm_check_request(KVM_REQ_VCPU_RESET, vcpu))
1007 			kvm_reset_vcpu(vcpu);
1008 
1009 		/*
1010 		 * Clear IRQ_PENDING requests that were made to guarantee
1011 		 * that a VCPU sees new virtual interrupts.
1012 		 */
1013 		kvm_check_request(KVM_REQ_IRQ_PENDING, vcpu);
1014 
1015 		if (kvm_check_request(KVM_REQ_RECORD_STEAL, vcpu))
1016 			kvm_update_stolen_time(vcpu);
1017 
1018 		if (kvm_check_request(KVM_REQ_RELOAD_GICv4, vcpu)) {
1019 			/* The distributor enable bits were changed */
1020 			preempt_disable();
1021 			vgic_v4_put(vcpu);
1022 			vgic_v4_load(vcpu);
1023 			preempt_enable();
1024 		}
1025 
1026 		if (kvm_check_request(KVM_REQ_RELOAD_PMU, vcpu))
1027 			kvm_vcpu_reload_pmu(vcpu);
1028 
1029 		if (kvm_check_request(KVM_REQ_RESYNC_PMU_EL0, vcpu))
1030 			kvm_vcpu_pmu_restore_guest(vcpu);
1031 
1032 		if (kvm_check_request(KVM_REQ_SUSPEND, vcpu))
1033 			return kvm_vcpu_suspend(vcpu);
1034 
1035 		if (kvm_dirty_ring_check_request(vcpu))
1036 			return 0;
1037 
1038 		check_nested_vcpu_requests(vcpu);
1039 	}
1040 
1041 	return 1;
1042 }
1043 
1044 static bool vcpu_mode_is_bad_32bit(struct kvm_vcpu *vcpu)
1045 {
1046 	if (likely(!vcpu_mode_is_32bit(vcpu)))
1047 		return false;
1048 
1049 	if (vcpu_has_nv(vcpu))
1050 		return true;
1051 
1052 	return !kvm_supports_32bit_el0();
1053 }
1054 
1055 /**
1056  * kvm_vcpu_exit_request - returns true if the VCPU should *not* enter the guest
1057  * @vcpu:	The VCPU pointer
1058  * @ret:	Pointer to write optional return code
1059  *
1060  * Returns: true if the VCPU needs to return to a preemptible + interruptible
1061  *	    and skip guest entry.
1062  *
1063  * This function disambiguates between two different types of exits: exits to a
1064  * preemptible + interruptible kernel context and exits to userspace. For an
1065  * exit to userspace, this function will write the return code to ret and return
1066  * true. For an exit to preemptible + interruptible kernel context (i.e. check
1067  * for pending work and re-enter), return true without writing to ret.
1068  */
1069 static bool kvm_vcpu_exit_request(struct kvm_vcpu *vcpu, int *ret)
1070 {
1071 	struct kvm_run *run = vcpu->run;
1072 
1073 	/*
1074 	 * If we're using a userspace irqchip, then check if we need
1075 	 * to tell a userspace irqchip about timer or PMU level
1076 	 * changes and if so, exit to userspace (the actual level
1077 	 * state gets updated in kvm_timer_update_run and
1078 	 * kvm_pmu_update_run below).
1079 	 */
1080 	if (static_branch_unlikely(&userspace_irqchip_in_use)) {
1081 		if (kvm_timer_should_notify_user(vcpu) ||
1082 		    kvm_pmu_should_notify_user(vcpu)) {
1083 			*ret = -EINTR;
1084 			run->exit_reason = KVM_EXIT_INTR;
1085 			return true;
1086 		}
1087 	}
1088 
1089 	if (unlikely(vcpu_on_unsupported_cpu(vcpu))) {
1090 		run->exit_reason = KVM_EXIT_FAIL_ENTRY;
1091 		run->fail_entry.hardware_entry_failure_reason = KVM_EXIT_FAIL_ENTRY_CPU_UNSUPPORTED;
1092 		run->fail_entry.cpu = smp_processor_id();
1093 		*ret = 0;
1094 		return true;
1095 	}
1096 
1097 	return kvm_request_pending(vcpu) ||
1098 			xfer_to_guest_mode_work_pending();
1099 }
1100 
1101 /*
1102  * Actually run the vCPU, entering an RCU extended quiescent state (EQS) while
1103  * the vCPU is running.
1104  *
1105  * This must be noinstr as instrumentation may make use of RCU, and this is not
1106  * safe during the EQS.
1107  */
1108 static int noinstr kvm_arm_vcpu_enter_exit(struct kvm_vcpu *vcpu)
1109 {
1110 	int ret;
1111 
1112 	guest_state_enter_irqoff();
1113 	ret = kvm_call_hyp_ret(__kvm_vcpu_run, vcpu);
1114 	guest_state_exit_irqoff();
1115 
1116 	return ret;
1117 }
1118 
1119 /**
1120  * kvm_arch_vcpu_ioctl_run - the main VCPU run function to execute guest code
1121  * @vcpu:	The VCPU pointer
1122  *
1123  * This function is called through the VCPU_RUN ioctl called from user space. It
1124  * will execute VM code in a loop until the time slice for the process is used
1125  * or some emulation is needed from user space in which case the function will
1126  * return with return value 0 and with the kvm_run structure filled in with the
1127  * required data for the requested emulation.
1128  */
1129 int kvm_arch_vcpu_ioctl_run(struct kvm_vcpu *vcpu)
1130 {
1131 	struct kvm_run *run = vcpu->run;
1132 	int ret;
1133 
1134 	if (run->exit_reason == KVM_EXIT_MMIO) {
1135 		ret = kvm_handle_mmio_return(vcpu);
1136 		if (ret <= 0)
1137 			return ret;
1138 	}
1139 
1140 	vcpu_load(vcpu);
1141 
1142 	if (!vcpu->wants_to_run) {
1143 		ret = -EINTR;
1144 		goto out;
1145 	}
1146 
1147 	kvm_sigset_activate(vcpu);
1148 
1149 	ret = 1;
1150 	run->exit_reason = KVM_EXIT_UNKNOWN;
1151 	run->flags = 0;
1152 	while (ret > 0) {
1153 		/*
1154 		 * Check conditions before entering the guest
1155 		 */
1156 		ret = xfer_to_guest_mode_handle_work(vcpu);
1157 		if (!ret)
1158 			ret = 1;
1159 
1160 		if (ret > 0)
1161 			ret = check_vcpu_requests(vcpu);
1162 
1163 		/*
1164 		 * Preparing the interrupts to be injected also
1165 		 * involves poking the GIC, which must be done in a
1166 		 * non-preemptible context.
1167 		 */
1168 		preempt_disable();
1169 
1170 		/*
1171 		 * The VMID allocator only tracks active VMIDs per
1172 		 * physical CPU, and therefore the VMID allocated may not be
1173 		 * preserved on VMID roll-over if the task was preempted,
1174 		 * making a thread's VMID inactive. So we need to call
1175 		 * kvm_arm_vmid_update() in non-premptible context.
1176 		 */
1177 		if (kvm_arm_vmid_update(&vcpu->arch.hw_mmu->vmid) &&
1178 		    has_vhe())
1179 			__load_stage2(vcpu->arch.hw_mmu,
1180 				      vcpu->arch.hw_mmu->arch);
1181 
1182 		kvm_pmu_flush_hwstate(vcpu);
1183 
1184 		local_irq_disable();
1185 
1186 		kvm_vgic_flush_hwstate(vcpu);
1187 
1188 		kvm_pmu_update_vcpu_events(vcpu);
1189 
1190 		/*
1191 		 * Ensure we set mode to IN_GUEST_MODE after we disable
1192 		 * interrupts and before the final VCPU requests check.
1193 		 * See the comment in kvm_vcpu_exiting_guest_mode() and
1194 		 * Documentation/virt/kvm/vcpu-requests.rst
1195 		 */
1196 		smp_store_mb(vcpu->mode, IN_GUEST_MODE);
1197 
1198 		if (ret <= 0 || kvm_vcpu_exit_request(vcpu, &ret)) {
1199 			vcpu->mode = OUTSIDE_GUEST_MODE;
1200 			isb(); /* Ensure work in x_flush_hwstate is committed */
1201 			kvm_pmu_sync_hwstate(vcpu);
1202 			if (static_branch_unlikely(&userspace_irqchip_in_use))
1203 				kvm_timer_sync_user(vcpu);
1204 			kvm_vgic_sync_hwstate(vcpu);
1205 			local_irq_enable();
1206 			preempt_enable();
1207 			continue;
1208 		}
1209 
1210 		kvm_arm_setup_debug(vcpu);
1211 		kvm_arch_vcpu_ctxflush_fp(vcpu);
1212 
1213 		/**************************************************************
1214 		 * Enter the guest
1215 		 */
1216 		trace_kvm_entry(*vcpu_pc(vcpu));
1217 		guest_timing_enter_irqoff();
1218 
1219 		ret = kvm_arm_vcpu_enter_exit(vcpu);
1220 
1221 		vcpu->mode = OUTSIDE_GUEST_MODE;
1222 		vcpu->stat.exits++;
1223 		/*
1224 		 * Back from guest
1225 		 *************************************************************/
1226 
1227 		kvm_arm_clear_debug(vcpu);
1228 
1229 		/*
1230 		 * We must sync the PMU state before the vgic state so
1231 		 * that the vgic can properly sample the updated state of the
1232 		 * interrupt line.
1233 		 */
1234 		kvm_pmu_sync_hwstate(vcpu);
1235 
1236 		/*
1237 		 * Sync the vgic state before syncing the timer state because
1238 		 * the timer code needs to know if the virtual timer
1239 		 * interrupts are active.
1240 		 */
1241 		kvm_vgic_sync_hwstate(vcpu);
1242 
1243 		/*
1244 		 * Sync the timer hardware state before enabling interrupts as
1245 		 * we don't want vtimer interrupts to race with syncing the
1246 		 * timer virtual interrupt state.
1247 		 */
1248 		if (static_branch_unlikely(&userspace_irqchip_in_use))
1249 			kvm_timer_sync_user(vcpu);
1250 
1251 		kvm_arch_vcpu_ctxsync_fp(vcpu);
1252 
1253 		/*
1254 		 * We must ensure that any pending interrupts are taken before
1255 		 * we exit guest timing so that timer ticks are accounted as
1256 		 * guest time. Transiently unmask interrupts so that any
1257 		 * pending interrupts are taken.
1258 		 *
1259 		 * Per ARM DDI 0487G.b section D1.13.4, an ISB (or other
1260 		 * context synchronization event) is necessary to ensure that
1261 		 * pending interrupts are taken.
1262 		 */
1263 		if (ARM_EXCEPTION_CODE(ret) == ARM_EXCEPTION_IRQ) {
1264 			local_irq_enable();
1265 			isb();
1266 			local_irq_disable();
1267 		}
1268 
1269 		guest_timing_exit_irqoff();
1270 
1271 		local_irq_enable();
1272 
1273 		trace_kvm_exit(ret, kvm_vcpu_trap_get_class(vcpu), *vcpu_pc(vcpu));
1274 
1275 		/* Exit types that need handling before we can be preempted */
1276 		handle_exit_early(vcpu, ret);
1277 
1278 		preempt_enable();
1279 
1280 		/*
1281 		 * The ARMv8 architecture doesn't give the hypervisor
1282 		 * a mechanism to prevent a guest from dropping to AArch32 EL0
1283 		 * if implemented by the CPU. If we spot the guest in such
1284 		 * state and that we decided it wasn't supposed to do so (like
1285 		 * with the asymmetric AArch32 case), return to userspace with
1286 		 * a fatal error.
1287 		 */
1288 		if (vcpu_mode_is_bad_32bit(vcpu)) {
1289 			/*
1290 			 * As we have caught the guest red-handed, decide that
1291 			 * it isn't fit for purpose anymore by making the vcpu
1292 			 * invalid. The VMM can try and fix it by issuing  a
1293 			 * KVM_ARM_VCPU_INIT if it really wants to.
1294 			 */
1295 			vcpu_clear_flag(vcpu, VCPU_INITIALIZED);
1296 			ret = ARM_EXCEPTION_IL;
1297 		}
1298 
1299 		ret = handle_exit(vcpu, ret);
1300 	}
1301 
1302 	/* Tell userspace about in-kernel device output levels */
1303 	if (unlikely(!irqchip_in_kernel(vcpu->kvm))) {
1304 		kvm_timer_update_run(vcpu);
1305 		kvm_pmu_update_run(vcpu);
1306 	}
1307 
1308 	kvm_sigset_deactivate(vcpu);
1309 
1310 out:
1311 	/*
1312 	 * In the unlikely event that we are returning to userspace
1313 	 * with pending exceptions or PC adjustment, commit these
1314 	 * adjustments in order to give userspace a consistent view of
1315 	 * the vcpu state. Note that this relies on __kvm_adjust_pc()
1316 	 * being preempt-safe on VHE.
1317 	 */
1318 	if (unlikely(vcpu_get_flag(vcpu, PENDING_EXCEPTION) ||
1319 		     vcpu_get_flag(vcpu, INCREMENT_PC)))
1320 		kvm_call_hyp(__kvm_adjust_pc, vcpu);
1321 
1322 	vcpu_put(vcpu);
1323 	return ret;
1324 }
1325 
1326 static int vcpu_interrupt_line(struct kvm_vcpu *vcpu, int number, bool level)
1327 {
1328 	int bit_index;
1329 	bool set;
1330 	unsigned long *hcr;
1331 
1332 	if (number == KVM_ARM_IRQ_CPU_IRQ)
1333 		bit_index = __ffs(HCR_VI);
1334 	else /* KVM_ARM_IRQ_CPU_FIQ */
1335 		bit_index = __ffs(HCR_VF);
1336 
1337 	hcr = vcpu_hcr(vcpu);
1338 	if (level)
1339 		set = test_and_set_bit(bit_index, hcr);
1340 	else
1341 		set = test_and_clear_bit(bit_index, hcr);
1342 
1343 	/*
1344 	 * If we didn't change anything, no need to wake up or kick other CPUs
1345 	 */
1346 	if (set == level)
1347 		return 0;
1348 
1349 	/*
1350 	 * The vcpu irq_lines field was updated, wake up sleeping VCPUs and
1351 	 * trigger a world-switch round on the running physical CPU to set the
1352 	 * virtual IRQ/FIQ fields in the HCR appropriately.
1353 	 */
1354 	kvm_make_request(KVM_REQ_IRQ_PENDING, vcpu);
1355 	kvm_vcpu_kick(vcpu);
1356 
1357 	return 0;
1358 }
1359 
1360 int kvm_vm_ioctl_irq_line(struct kvm *kvm, struct kvm_irq_level *irq_level,
1361 			  bool line_status)
1362 {
1363 	u32 irq = irq_level->irq;
1364 	unsigned int irq_type, vcpu_id, irq_num;
1365 	struct kvm_vcpu *vcpu = NULL;
1366 	bool level = irq_level->level;
1367 
1368 	irq_type = (irq >> KVM_ARM_IRQ_TYPE_SHIFT) & KVM_ARM_IRQ_TYPE_MASK;
1369 	vcpu_id = (irq >> KVM_ARM_IRQ_VCPU_SHIFT) & KVM_ARM_IRQ_VCPU_MASK;
1370 	vcpu_id += ((irq >> KVM_ARM_IRQ_VCPU2_SHIFT) & KVM_ARM_IRQ_VCPU2_MASK) * (KVM_ARM_IRQ_VCPU_MASK + 1);
1371 	irq_num = (irq >> KVM_ARM_IRQ_NUM_SHIFT) & KVM_ARM_IRQ_NUM_MASK;
1372 
1373 	trace_kvm_irq_line(irq_type, vcpu_id, irq_num, irq_level->level);
1374 
1375 	switch (irq_type) {
1376 	case KVM_ARM_IRQ_TYPE_CPU:
1377 		if (irqchip_in_kernel(kvm))
1378 			return -ENXIO;
1379 
1380 		vcpu = kvm_get_vcpu_by_id(kvm, vcpu_id);
1381 		if (!vcpu)
1382 			return -EINVAL;
1383 
1384 		if (irq_num > KVM_ARM_IRQ_CPU_FIQ)
1385 			return -EINVAL;
1386 
1387 		return vcpu_interrupt_line(vcpu, irq_num, level);
1388 	case KVM_ARM_IRQ_TYPE_PPI:
1389 		if (!irqchip_in_kernel(kvm))
1390 			return -ENXIO;
1391 
1392 		vcpu = kvm_get_vcpu_by_id(kvm, vcpu_id);
1393 		if (!vcpu)
1394 			return -EINVAL;
1395 
1396 		if (irq_num < VGIC_NR_SGIS || irq_num >= VGIC_NR_PRIVATE_IRQS)
1397 			return -EINVAL;
1398 
1399 		return kvm_vgic_inject_irq(kvm, vcpu, irq_num, level, NULL);
1400 	case KVM_ARM_IRQ_TYPE_SPI:
1401 		if (!irqchip_in_kernel(kvm))
1402 			return -ENXIO;
1403 
1404 		if (irq_num < VGIC_NR_PRIVATE_IRQS)
1405 			return -EINVAL;
1406 
1407 		return kvm_vgic_inject_irq(kvm, NULL, irq_num, level, NULL);
1408 	}
1409 
1410 	return -EINVAL;
1411 }
1412 
1413 static unsigned long system_supported_vcpu_features(void)
1414 {
1415 	unsigned long features = KVM_VCPU_VALID_FEATURES;
1416 
1417 	if (!cpus_have_final_cap(ARM64_HAS_32BIT_EL1))
1418 		clear_bit(KVM_ARM_VCPU_EL1_32BIT, &features);
1419 
1420 	if (!kvm_arm_support_pmu_v3())
1421 		clear_bit(KVM_ARM_VCPU_PMU_V3, &features);
1422 
1423 	if (!system_supports_sve())
1424 		clear_bit(KVM_ARM_VCPU_SVE, &features);
1425 
1426 	if (!kvm_has_full_ptr_auth()) {
1427 		clear_bit(KVM_ARM_VCPU_PTRAUTH_ADDRESS, &features);
1428 		clear_bit(KVM_ARM_VCPU_PTRAUTH_GENERIC, &features);
1429 	}
1430 
1431 	if (!cpus_have_final_cap(ARM64_HAS_NESTED_VIRT))
1432 		clear_bit(KVM_ARM_VCPU_HAS_EL2, &features);
1433 
1434 	return features;
1435 }
1436 
1437 static int kvm_vcpu_init_check_features(struct kvm_vcpu *vcpu,
1438 					const struct kvm_vcpu_init *init)
1439 {
1440 	unsigned long features = init->features[0];
1441 	int i;
1442 
1443 	if (features & ~KVM_VCPU_VALID_FEATURES)
1444 		return -ENOENT;
1445 
1446 	for (i = 1; i < ARRAY_SIZE(init->features); i++) {
1447 		if (init->features[i])
1448 			return -ENOENT;
1449 	}
1450 
1451 	if (features & ~system_supported_vcpu_features())
1452 		return -EINVAL;
1453 
1454 	/*
1455 	 * For now make sure that both address/generic pointer authentication
1456 	 * features are requested by the userspace together.
1457 	 */
1458 	if (test_bit(KVM_ARM_VCPU_PTRAUTH_ADDRESS, &features) !=
1459 	    test_bit(KVM_ARM_VCPU_PTRAUTH_GENERIC, &features))
1460 		return -EINVAL;
1461 
1462 	if (!test_bit(KVM_ARM_VCPU_EL1_32BIT, &features))
1463 		return 0;
1464 
1465 	/* MTE is incompatible with AArch32 */
1466 	if (kvm_has_mte(vcpu->kvm))
1467 		return -EINVAL;
1468 
1469 	/* NV is incompatible with AArch32 */
1470 	if (test_bit(KVM_ARM_VCPU_HAS_EL2, &features))
1471 		return -EINVAL;
1472 
1473 	return 0;
1474 }
1475 
1476 static bool kvm_vcpu_init_changed(struct kvm_vcpu *vcpu,
1477 				  const struct kvm_vcpu_init *init)
1478 {
1479 	unsigned long features = init->features[0];
1480 
1481 	return !bitmap_equal(vcpu->kvm->arch.vcpu_features, &features,
1482 			     KVM_VCPU_MAX_FEATURES);
1483 }
1484 
1485 static int kvm_setup_vcpu(struct kvm_vcpu *vcpu)
1486 {
1487 	struct kvm *kvm = vcpu->kvm;
1488 	int ret = 0;
1489 
1490 	/*
1491 	 * When the vCPU has a PMU, but no PMU is set for the guest
1492 	 * yet, set the default one.
1493 	 */
1494 	if (kvm_vcpu_has_pmu(vcpu) && !kvm->arch.arm_pmu)
1495 		ret = kvm_arm_set_default_pmu(kvm);
1496 
1497 	/* Prepare for nested if required */
1498 	if (!ret && vcpu_has_nv(vcpu))
1499 		ret = kvm_vcpu_init_nested(vcpu);
1500 
1501 	return ret;
1502 }
1503 
1504 static int __kvm_vcpu_set_target(struct kvm_vcpu *vcpu,
1505 				 const struct kvm_vcpu_init *init)
1506 {
1507 	unsigned long features = init->features[0];
1508 	struct kvm *kvm = vcpu->kvm;
1509 	int ret = -EINVAL;
1510 
1511 	mutex_lock(&kvm->arch.config_lock);
1512 
1513 	if (test_bit(KVM_ARCH_FLAG_VCPU_FEATURES_CONFIGURED, &kvm->arch.flags) &&
1514 	    kvm_vcpu_init_changed(vcpu, init))
1515 		goto out_unlock;
1516 
1517 	bitmap_copy(kvm->arch.vcpu_features, &features, KVM_VCPU_MAX_FEATURES);
1518 
1519 	ret = kvm_setup_vcpu(vcpu);
1520 	if (ret)
1521 		goto out_unlock;
1522 
1523 	/* Now we know what it is, we can reset it. */
1524 	kvm_reset_vcpu(vcpu);
1525 
1526 	set_bit(KVM_ARCH_FLAG_VCPU_FEATURES_CONFIGURED, &kvm->arch.flags);
1527 	vcpu_set_flag(vcpu, VCPU_INITIALIZED);
1528 	ret = 0;
1529 out_unlock:
1530 	mutex_unlock(&kvm->arch.config_lock);
1531 	return ret;
1532 }
1533 
1534 static int kvm_vcpu_set_target(struct kvm_vcpu *vcpu,
1535 			       const struct kvm_vcpu_init *init)
1536 {
1537 	int ret;
1538 
1539 	if (init->target != KVM_ARM_TARGET_GENERIC_V8 &&
1540 	    init->target != kvm_target_cpu())
1541 		return -EINVAL;
1542 
1543 	ret = kvm_vcpu_init_check_features(vcpu, init);
1544 	if (ret)
1545 		return ret;
1546 
1547 	if (!kvm_vcpu_initialized(vcpu))
1548 		return __kvm_vcpu_set_target(vcpu, init);
1549 
1550 	if (kvm_vcpu_init_changed(vcpu, init))
1551 		return -EINVAL;
1552 
1553 	kvm_reset_vcpu(vcpu);
1554 	return 0;
1555 }
1556 
1557 static int kvm_arch_vcpu_ioctl_vcpu_init(struct kvm_vcpu *vcpu,
1558 					 struct kvm_vcpu_init *init)
1559 {
1560 	bool power_off = false;
1561 	int ret;
1562 
1563 	/*
1564 	 * Treat the power-off vCPU feature as ephemeral. Clear the bit to avoid
1565 	 * reflecting it in the finalized feature set, thus limiting its scope
1566 	 * to a single KVM_ARM_VCPU_INIT call.
1567 	 */
1568 	if (init->features[0] & BIT(KVM_ARM_VCPU_POWER_OFF)) {
1569 		init->features[0] &= ~BIT(KVM_ARM_VCPU_POWER_OFF);
1570 		power_off = true;
1571 	}
1572 
1573 	ret = kvm_vcpu_set_target(vcpu, init);
1574 	if (ret)
1575 		return ret;
1576 
1577 	/*
1578 	 * Ensure a rebooted VM will fault in RAM pages and detect if the
1579 	 * guest MMU is turned off and flush the caches as needed.
1580 	 *
1581 	 * S2FWB enforces all memory accesses to RAM being cacheable,
1582 	 * ensuring that the data side is always coherent. We still
1583 	 * need to invalidate the I-cache though, as FWB does *not*
1584 	 * imply CTR_EL0.DIC.
1585 	 */
1586 	if (vcpu_has_run_once(vcpu)) {
1587 		if (!cpus_have_final_cap(ARM64_HAS_STAGE2_FWB))
1588 			stage2_unmap_vm(vcpu->kvm);
1589 		else
1590 			icache_inval_all_pou();
1591 	}
1592 
1593 	vcpu_reset_hcr(vcpu);
1594 	vcpu->arch.cptr_el2 = kvm_get_reset_cptr_el2(vcpu);
1595 
1596 	/*
1597 	 * Handle the "start in power-off" case.
1598 	 */
1599 	spin_lock(&vcpu->arch.mp_state_lock);
1600 
1601 	if (power_off)
1602 		__kvm_arm_vcpu_power_off(vcpu);
1603 	else
1604 		WRITE_ONCE(vcpu->arch.mp_state.mp_state, KVM_MP_STATE_RUNNABLE);
1605 
1606 	spin_unlock(&vcpu->arch.mp_state_lock);
1607 
1608 	return 0;
1609 }
1610 
1611 static int kvm_arm_vcpu_set_attr(struct kvm_vcpu *vcpu,
1612 				 struct kvm_device_attr *attr)
1613 {
1614 	int ret = -ENXIO;
1615 
1616 	switch (attr->group) {
1617 	default:
1618 		ret = kvm_arm_vcpu_arch_set_attr(vcpu, attr);
1619 		break;
1620 	}
1621 
1622 	return ret;
1623 }
1624 
1625 static int kvm_arm_vcpu_get_attr(struct kvm_vcpu *vcpu,
1626 				 struct kvm_device_attr *attr)
1627 {
1628 	int ret = -ENXIO;
1629 
1630 	switch (attr->group) {
1631 	default:
1632 		ret = kvm_arm_vcpu_arch_get_attr(vcpu, attr);
1633 		break;
1634 	}
1635 
1636 	return ret;
1637 }
1638 
1639 static int kvm_arm_vcpu_has_attr(struct kvm_vcpu *vcpu,
1640 				 struct kvm_device_attr *attr)
1641 {
1642 	int ret = -ENXIO;
1643 
1644 	switch (attr->group) {
1645 	default:
1646 		ret = kvm_arm_vcpu_arch_has_attr(vcpu, attr);
1647 		break;
1648 	}
1649 
1650 	return ret;
1651 }
1652 
1653 static int kvm_arm_vcpu_get_events(struct kvm_vcpu *vcpu,
1654 				   struct kvm_vcpu_events *events)
1655 {
1656 	memset(events, 0, sizeof(*events));
1657 
1658 	return __kvm_arm_vcpu_get_events(vcpu, events);
1659 }
1660 
1661 static int kvm_arm_vcpu_set_events(struct kvm_vcpu *vcpu,
1662 				   struct kvm_vcpu_events *events)
1663 {
1664 	int i;
1665 
1666 	/* check whether the reserved field is zero */
1667 	for (i = 0; i < ARRAY_SIZE(events->reserved); i++)
1668 		if (events->reserved[i])
1669 			return -EINVAL;
1670 
1671 	/* check whether the pad field is zero */
1672 	for (i = 0; i < ARRAY_SIZE(events->exception.pad); i++)
1673 		if (events->exception.pad[i])
1674 			return -EINVAL;
1675 
1676 	return __kvm_arm_vcpu_set_events(vcpu, events);
1677 }
1678 
1679 long kvm_arch_vcpu_ioctl(struct file *filp,
1680 			 unsigned int ioctl, unsigned long arg)
1681 {
1682 	struct kvm_vcpu *vcpu = filp->private_data;
1683 	void __user *argp = (void __user *)arg;
1684 	struct kvm_device_attr attr;
1685 	long r;
1686 
1687 	switch (ioctl) {
1688 	case KVM_ARM_VCPU_INIT: {
1689 		struct kvm_vcpu_init init;
1690 
1691 		r = -EFAULT;
1692 		if (copy_from_user(&init, argp, sizeof(init)))
1693 			break;
1694 
1695 		r = kvm_arch_vcpu_ioctl_vcpu_init(vcpu, &init);
1696 		break;
1697 	}
1698 	case KVM_SET_ONE_REG:
1699 	case KVM_GET_ONE_REG: {
1700 		struct kvm_one_reg reg;
1701 
1702 		r = -ENOEXEC;
1703 		if (unlikely(!kvm_vcpu_initialized(vcpu)))
1704 			break;
1705 
1706 		r = -EFAULT;
1707 		if (copy_from_user(&reg, argp, sizeof(reg)))
1708 			break;
1709 
1710 		/*
1711 		 * We could owe a reset due to PSCI. Handle the pending reset
1712 		 * here to ensure userspace register accesses are ordered after
1713 		 * the reset.
1714 		 */
1715 		if (kvm_check_request(KVM_REQ_VCPU_RESET, vcpu))
1716 			kvm_reset_vcpu(vcpu);
1717 
1718 		if (ioctl == KVM_SET_ONE_REG)
1719 			r = kvm_arm_set_reg(vcpu, &reg);
1720 		else
1721 			r = kvm_arm_get_reg(vcpu, &reg);
1722 		break;
1723 	}
1724 	case KVM_GET_REG_LIST: {
1725 		struct kvm_reg_list __user *user_list = argp;
1726 		struct kvm_reg_list reg_list;
1727 		unsigned n;
1728 
1729 		r = -ENOEXEC;
1730 		if (unlikely(!kvm_vcpu_initialized(vcpu)))
1731 			break;
1732 
1733 		r = -EPERM;
1734 		if (!kvm_arm_vcpu_is_finalized(vcpu))
1735 			break;
1736 
1737 		r = -EFAULT;
1738 		if (copy_from_user(&reg_list, user_list, sizeof(reg_list)))
1739 			break;
1740 		n = reg_list.n;
1741 		reg_list.n = kvm_arm_num_regs(vcpu);
1742 		if (copy_to_user(user_list, &reg_list, sizeof(reg_list)))
1743 			break;
1744 		r = -E2BIG;
1745 		if (n < reg_list.n)
1746 			break;
1747 		r = kvm_arm_copy_reg_indices(vcpu, user_list->reg);
1748 		break;
1749 	}
1750 	case KVM_SET_DEVICE_ATTR: {
1751 		r = -EFAULT;
1752 		if (copy_from_user(&attr, argp, sizeof(attr)))
1753 			break;
1754 		r = kvm_arm_vcpu_set_attr(vcpu, &attr);
1755 		break;
1756 	}
1757 	case KVM_GET_DEVICE_ATTR: {
1758 		r = -EFAULT;
1759 		if (copy_from_user(&attr, argp, sizeof(attr)))
1760 			break;
1761 		r = kvm_arm_vcpu_get_attr(vcpu, &attr);
1762 		break;
1763 	}
1764 	case KVM_HAS_DEVICE_ATTR: {
1765 		r = -EFAULT;
1766 		if (copy_from_user(&attr, argp, sizeof(attr)))
1767 			break;
1768 		r = kvm_arm_vcpu_has_attr(vcpu, &attr);
1769 		break;
1770 	}
1771 	case KVM_GET_VCPU_EVENTS: {
1772 		struct kvm_vcpu_events events;
1773 
1774 		if (kvm_arm_vcpu_get_events(vcpu, &events))
1775 			return -EINVAL;
1776 
1777 		if (copy_to_user(argp, &events, sizeof(events)))
1778 			return -EFAULT;
1779 
1780 		return 0;
1781 	}
1782 	case KVM_SET_VCPU_EVENTS: {
1783 		struct kvm_vcpu_events events;
1784 
1785 		if (copy_from_user(&events, argp, sizeof(events)))
1786 			return -EFAULT;
1787 
1788 		return kvm_arm_vcpu_set_events(vcpu, &events);
1789 	}
1790 	case KVM_ARM_VCPU_FINALIZE: {
1791 		int what;
1792 
1793 		if (!kvm_vcpu_initialized(vcpu))
1794 			return -ENOEXEC;
1795 
1796 		if (get_user(what, (const int __user *)argp))
1797 			return -EFAULT;
1798 
1799 		return kvm_arm_vcpu_finalize(vcpu, what);
1800 	}
1801 	default:
1802 		r = -EINVAL;
1803 	}
1804 
1805 	return r;
1806 }
1807 
1808 void kvm_arch_sync_dirty_log(struct kvm *kvm, struct kvm_memory_slot *memslot)
1809 {
1810 
1811 }
1812 
1813 static int kvm_vm_ioctl_set_device_addr(struct kvm *kvm,
1814 					struct kvm_arm_device_addr *dev_addr)
1815 {
1816 	switch (FIELD_GET(KVM_ARM_DEVICE_ID_MASK, dev_addr->id)) {
1817 	case KVM_ARM_DEVICE_VGIC_V2:
1818 		if (!vgic_present)
1819 			return -ENXIO;
1820 		return kvm_set_legacy_vgic_v2_addr(kvm, dev_addr);
1821 	default:
1822 		return -ENODEV;
1823 	}
1824 }
1825 
1826 static int kvm_vm_has_attr(struct kvm *kvm, struct kvm_device_attr *attr)
1827 {
1828 	switch (attr->group) {
1829 	case KVM_ARM_VM_SMCCC_CTRL:
1830 		return kvm_vm_smccc_has_attr(kvm, attr);
1831 	default:
1832 		return -ENXIO;
1833 	}
1834 }
1835 
1836 static int kvm_vm_set_attr(struct kvm *kvm, struct kvm_device_attr *attr)
1837 {
1838 	switch (attr->group) {
1839 	case KVM_ARM_VM_SMCCC_CTRL:
1840 		return kvm_vm_smccc_set_attr(kvm, attr);
1841 	default:
1842 		return -ENXIO;
1843 	}
1844 }
1845 
1846 int kvm_arch_vm_ioctl(struct file *filp, unsigned int ioctl, unsigned long arg)
1847 {
1848 	struct kvm *kvm = filp->private_data;
1849 	void __user *argp = (void __user *)arg;
1850 	struct kvm_device_attr attr;
1851 
1852 	switch (ioctl) {
1853 	case KVM_CREATE_IRQCHIP: {
1854 		int ret;
1855 		if (!vgic_present)
1856 			return -ENXIO;
1857 		mutex_lock(&kvm->lock);
1858 		ret = kvm_vgic_create(kvm, KVM_DEV_TYPE_ARM_VGIC_V2);
1859 		mutex_unlock(&kvm->lock);
1860 		return ret;
1861 	}
1862 	case KVM_ARM_SET_DEVICE_ADDR: {
1863 		struct kvm_arm_device_addr dev_addr;
1864 
1865 		if (copy_from_user(&dev_addr, argp, sizeof(dev_addr)))
1866 			return -EFAULT;
1867 		return kvm_vm_ioctl_set_device_addr(kvm, &dev_addr);
1868 	}
1869 	case KVM_ARM_PREFERRED_TARGET: {
1870 		struct kvm_vcpu_init init = {
1871 			.target = KVM_ARM_TARGET_GENERIC_V8,
1872 		};
1873 
1874 		if (copy_to_user(argp, &init, sizeof(init)))
1875 			return -EFAULT;
1876 
1877 		return 0;
1878 	}
1879 	case KVM_ARM_MTE_COPY_TAGS: {
1880 		struct kvm_arm_copy_mte_tags copy_tags;
1881 
1882 		if (copy_from_user(&copy_tags, argp, sizeof(copy_tags)))
1883 			return -EFAULT;
1884 		return kvm_vm_ioctl_mte_copy_tags(kvm, &copy_tags);
1885 	}
1886 	case KVM_ARM_SET_COUNTER_OFFSET: {
1887 		struct kvm_arm_counter_offset offset;
1888 
1889 		if (copy_from_user(&offset, argp, sizeof(offset)))
1890 			return -EFAULT;
1891 		return kvm_vm_ioctl_set_counter_offset(kvm, &offset);
1892 	}
1893 	case KVM_HAS_DEVICE_ATTR: {
1894 		if (copy_from_user(&attr, argp, sizeof(attr)))
1895 			return -EFAULT;
1896 
1897 		return kvm_vm_has_attr(kvm, &attr);
1898 	}
1899 	case KVM_SET_DEVICE_ATTR: {
1900 		if (copy_from_user(&attr, argp, sizeof(attr)))
1901 			return -EFAULT;
1902 
1903 		return kvm_vm_set_attr(kvm, &attr);
1904 	}
1905 	case KVM_ARM_GET_REG_WRITABLE_MASKS: {
1906 		struct reg_mask_range range;
1907 
1908 		if (copy_from_user(&range, argp, sizeof(range)))
1909 			return -EFAULT;
1910 		return kvm_vm_ioctl_get_reg_writable_masks(kvm, &range);
1911 	}
1912 	default:
1913 		return -EINVAL;
1914 	}
1915 }
1916 
1917 /* unlocks vcpus from @vcpu_lock_idx and smaller */
1918 static void unlock_vcpus(struct kvm *kvm, int vcpu_lock_idx)
1919 {
1920 	struct kvm_vcpu *tmp_vcpu;
1921 
1922 	for (; vcpu_lock_idx >= 0; vcpu_lock_idx--) {
1923 		tmp_vcpu = kvm_get_vcpu(kvm, vcpu_lock_idx);
1924 		mutex_unlock(&tmp_vcpu->mutex);
1925 	}
1926 }
1927 
1928 void unlock_all_vcpus(struct kvm *kvm)
1929 {
1930 	lockdep_assert_held(&kvm->lock);
1931 
1932 	unlock_vcpus(kvm, atomic_read(&kvm->online_vcpus) - 1);
1933 }
1934 
1935 /* Returns true if all vcpus were locked, false otherwise */
1936 bool lock_all_vcpus(struct kvm *kvm)
1937 {
1938 	struct kvm_vcpu *tmp_vcpu;
1939 	unsigned long c;
1940 
1941 	lockdep_assert_held(&kvm->lock);
1942 
1943 	/*
1944 	 * Any time a vcpu is in an ioctl (including running), the
1945 	 * core KVM code tries to grab the vcpu->mutex.
1946 	 *
1947 	 * By grabbing the vcpu->mutex of all VCPUs we ensure that no
1948 	 * other VCPUs can fiddle with the state while we access it.
1949 	 */
1950 	kvm_for_each_vcpu(c, tmp_vcpu, kvm) {
1951 		if (!mutex_trylock(&tmp_vcpu->mutex)) {
1952 			unlock_vcpus(kvm, c - 1);
1953 			return false;
1954 		}
1955 	}
1956 
1957 	return true;
1958 }
1959 
1960 static unsigned long nvhe_percpu_size(void)
1961 {
1962 	return (unsigned long)CHOOSE_NVHE_SYM(__per_cpu_end) -
1963 		(unsigned long)CHOOSE_NVHE_SYM(__per_cpu_start);
1964 }
1965 
1966 static unsigned long nvhe_percpu_order(void)
1967 {
1968 	unsigned long size = nvhe_percpu_size();
1969 
1970 	return size ? get_order(size) : 0;
1971 }
1972 
1973 static size_t pkvm_host_sve_state_order(void)
1974 {
1975 	return get_order(pkvm_host_sve_state_size());
1976 }
1977 
1978 /* A lookup table holding the hypervisor VA for each vector slot */
1979 static void *hyp_spectre_vector_selector[BP_HARDEN_EL2_SLOTS];
1980 
1981 static void kvm_init_vector_slot(void *base, enum arm64_hyp_spectre_vector slot)
1982 {
1983 	hyp_spectre_vector_selector[slot] = __kvm_vector_slot2addr(base, slot);
1984 }
1985 
1986 static int kvm_init_vector_slots(void)
1987 {
1988 	int err;
1989 	void *base;
1990 
1991 	base = kern_hyp_va(kvm_ksym_ref(__kvm_hyp_vector));
1992 	kvm_init_vector_slot(base, HYP_VECTOR_DIRECT);
1993 
1994 	base = kern_hyp_va(kvm_ksym_ref(__bp_harden_hyp_vecs));
1995 	kvm_init_vector_slot(base, HYP_VECTOR_SPECTRE_DIRECT);
1996 
1997 	if (kvm_system_needs_idmapped_vectors() &&
1998 	    !is_protected_kvm_enabled()) {
1999 		err = create_hyp_exec_mappings(__pa_symbol(__bp_harden_hyp_vecs),
2000 					       __BP_HARDEN_HYP_VECS_SZ, &base);
2001 		if (err)
2002 			return err;
2003 	}
2004 
2005 	kvm_init_vector_slot(base, HYP_VECTOR_INDIRECT);
2006 	kvm_init_vector_slot(base, HYP_VECTOR_SPECTRE_INDIRECT);
2007 	return 0;
2008 }
2009 
2010 static void __init cpu_prepare_hyp_mode(int cpu, u32 hyp_va_bits)
2011 {
2012 	struct kvm_nvhe_init_params *params = per_cpu_ptr_nvhe_sym(kvm_init_params, cpu);
2013 	u64 mmfr0 = read_sanitised_ftr_reg(SYS_ID_AA64MMFR0_EL1);
2014 	unsigned long tcr;
2015 
2016 	/*
2017 	 * Calculate the raw per-cpu offset without a translation from the
2018 	 * kernel's mapping to the linear mapping, and store it in tpidr_el2
2019 	 * so that we can use adr_l to access per-cpu variables in EL2.
2020 	 * Also drop the KASAN tag which gets in the way...
2021 	 */
2022 	params->tpidr_el2 = (unsigned long)kasan_reset_tag(per_cpu_ptr_nvhe_sym(__per_cpu_start, cpu)) -
2023 			    (unsigned long)kvm_ksym_ref(CHOOSE_NVHE_SYM(__per_cpu_start));
2024 
2025 	params->mair_el2 = read_sysreg(mair_el1);
2026 
2027 	tcr = read_sysreg(tcr_el1);
2028 	if (cpus_have_final_cap(ARM64_KVM_HVHE)) {
2029 		tcr |= TCR_EPD1_MASK;
2030 	} else {
2031 		tcr &= TCR_EL2_MASK;
2032 		tcr |= TCR_EL2_RES1;
2033 	}
2034 	tcr &= ~TCR_T0SZ_MASK;
2035 	tcr |= TCR_T0SZ(hyp_va_bits);
2036 	tcr &= ~TCR_EL2_PS_MASK;
2037 	tcr |= FIELD_PREP(TCR_EL2_PS_MASK, kvm_get_parange(mmfr0));
2038 	if (kvm_lpa2_is_enabled())
2039 		tcr |= TCR_EL2_DS;
2040 	params->tcr_el2 = tcr;
2041 
2042 	params->pgd_pa = kvm_mmu_get_httbr();
2043 	if (is_protected_kvm_enabled())
2044 		params->hcr_el2 = HCR_HOST_NVHE_PROTECTED_FLAGS;
2045 	else
2046 		params->hcr_el2 = HCR_HOST_NVHE_FLAGS;
2047 	if (cpus_have_final_cap(ARM64_KVM_HVHE))
2048 		params->hcr_el2 |= HCR_E2H;
2049 	params->vttbr = params->vtcr = 0;
2050 
2051 	/*
2052 	 * Flush the init params from the data cache because the struct will
2053 	 * be read while the MMU is off.
2054 	 */
2055 	kvm_flush_dcache_to_poc(params, sizeof(*params));
2056 }
2057 
2058 static void hyp_install_host_vector(void)
2059 {
2060 	struct kvm_nvhe_init_params *params;
2061 	struct arm_smccc_res res;
2062 
2063 	/* Switch from the HYP stub to our own HYP init vector */
2064 	__hyp_set_vectors(kvm_get_idmap_vector());
2065 
2066 	/*
2067 	 * Call initialization code, and switch to the full blown HYP code.
2068 	 * If the cpucaps haven't been finalized yet, something has gone very
2069 	 * wrong, and hyp will crash and burn when it uses any
2070 	 * cpus_have_*_cap() wrapper.
2071 	 */
2072 	BUG_ON(!system_capabilities_finalized());
2073 	params = this_cpu_ptr_nvhe_sym(kvm_init_params);
2074 	arm_smccc_1_1_hvc(KVM_HOST_SMCCC_FUNC(__kvm_hyp_init), virt_to_phys(params), &res);
2075 	WARN_ON(res.a0 != SMCCC_RET_SUCCESS);
2076 }
2077 
2078 static void cpu_init_hyp_mode(void)
2079 {
2080 	hyp_install_host_vector();
2081 
2082 	/*
2083 	 * Disabling SSBD on a non-VHE system requires us to enable SSBS
2084 	 * at EL2.
2085 	 */
2086 	if (this_cpu_has_cap(ARM64_SSBS) &&
2087 	    arm64_get_spectre_v4_state() == SPECTRE_VULNERABLE) {
2088 		kvm_call_hyp_nvhe(__kvm_enable_ssbs);
2089 	}
2090 }
2091 
2092 static void cpu_hyp_reset(void)
2093 {
2094 	if (!is_kernel_in_hyp_mode())
2095 		__hyp_reset_vectors();
2096 }
2097 
2098 /*
2099  * EL2 vectors can be mapped and rerouted in a number of ways,
2100  * depending on the kernel configuration and CPU present:
2101  *
2102  * - If the CPU is affected by Spectre-v2, the hardening sequence is
2103  *   placed in one of the vector slots, which is executed before jumping
2104  *   to the real vectors.
2105  *
2106  * - If the CPU also has the ARM64_SPECTRE_V3A cap, the slot
2107  *   containing the hardening sequence is mapped next to the idmap page,
2108  *   and executed before jumping to the real vectors.
2109  *
2110  * - If the CPU only has the ARM64_SPECTRE_V3A cap, then an
2111  *   empty slot is selected, mapped next to the idmap page, and
2112  *   executed before jumping to the real vectors.
2113  *
2114  * Note that ARM64_SPECTRE_V3A is somewhat incompatible with
2115  * VHE, as we don't have hypervisor-specific mappings. If the system
2116  * is VHE and yet selects this capability, it will be ignored.
2117  */
2118 static void cpu_set_hyp_vector(void)
2119 {
2120 	struct bp_hardening_data *data = this_cpu_ptr(&bp_hardening_data);
2121 	void *vector = hyp_spectre_vector_selector[data->slot];
2122 
2123 	if (!is_protected_kvm_enabled())
2124 		*this_cpu_ptr_hyp_sym(kvm_hyp_vector) = (unsigned long)vector;
2125 	else
2126 		kvm_call_hyp_nvhe(__pkvm_cpu_set_vector, data->slot);
2127 }
2128 
2129 static void cpu_hyp_init_context(void)
2130 {
2131 	kvm_init_host_cpu_context(host_data_ptr(host_ctxt));
2132 
2133 	if (!is_kernel_in_hyp_mode())
2134 		cpu_init_hyp_mode();
2135 }
2136 
2137 static void cpu_hyp_init_features(void)
2138 {
2139 	cpu_set_hyp_vector();
2140 	kvm_arm_init_debug();
2141 
2142 	if (is_kernel_in_hyp_mode())
2143 		kvm_timer_init_vhe();
2144 
2145 	if (vgic_present)
2146 		kvm_vgic_init_cpu_hardware();
2147 }
2148 
2149 static void cpu_hyp_reinit(void)
2150 {
2151 	cpu_hyp_reset();
2152 	cpu_hyp_init_context();
2153 	cpu_hyp_init_features();
2154 }
2155 
2156 static void cpu_hyp_init(void *discard)
2157 {
2158 	if (!__this_cpu_read(kvm_hyp_initialized)) {
2159 		cpu_hyp_reinit();
2160 		__this_cpu_write(kvm_hyp_initialized, 1);
2161 	}
2162 }
2163 
2164 static void cpu_hyp_uninit(void *discard)
2165 {
2166 	if (__this_cpu_read(kvm_hyp_initialized)) {
2167 		cpu_hyp_reset();
2168 		__this_cpu_write(kvm_hyp_initialized, 0);
2169 	}
2170 }
2171 
2172 int kvm_arch_enable_virtualization_cpu(void)
2173 {
2174 	/*
2175 	 * Most calls to this function are made with migration
2176 	 * disabled, but not with preemption disabled. The former is
2177 	 * enough to ensure correctness, but most of the helpers
2178 	 * expect the later and will throw a tantrum otherwise.
2179 	 */
2180 	preempt_disable();
2181 
2182 	cpu_hyp_init(NULL);
2183 
2184 	kvm_vgic_cpu_up();
2185 	kvm_timer_cpu_up();
2186 
2187 	preempt_enable();
2188 
2189 	return 0;
2190 }
2191 
2192 void kvm_arch_disable_virtualization_cpu(void)
2193 {
2194 	kvm_timer_cpu_down();
2195 	kvm_vgic_cpu_down();
2196 
2197 	if (!is_protected_kvm_enabled())
2198 		cpu_hyp_uninit(NULL);
2199 }
2200 
2201 #ifdef CONFIG_CPU_PM
2202 static int hyp_init_cpu_pm_notifier(struct notifier_block *self,
2203 				    unsigned long cmd,
2204 				    void *v)
2205 {
2206 	/*
2207 	 * kvm_hyp_initialized is left with its old value over
2208 	 * PM_ENTER->PM_EXIT. It is used to indicate PM_EXIT should
2209 	 * re-enable hyp.
2210 	 */
2211 	switch (cmd) {
2212 	case CPU_PM_ENTER:
2213 		if (__this_cpu_read(kvm_hyp_initialized))
2214 			/*
2215 			 * don't update kvm_hyp_initialized here
2216 			 * so that the hyp will be re-enabled
2217 			 * when we resume. See below.
2218 			 */
2219 			cpu_hyp_reset();
2220 
2221 		return NOTIFY_OK;
2222 	case CPU_PM_ENTER_FAILED:
2223 	case CPU_PM_EXIT:
2224 		if (__this_cpu_read(kvm_hyp_initialized))
2225 			/* The hyp was enabled before suspend. */
2226 			cpu_hyp_reinit();
2227 
2228 		return NOTIFY_OK;
2229 
2230 	default:
2231 		return NOTIFY_DONE;
2232 	}
2233 }
2234 
2235 static struct notifier_block hyp_init_cpu_pm_nb = {
2236 	.notifier_call = hyp_init_cpu_pm_notifier,
2237 };
2238 
2239 static void __init hyp_cpu_pm_init(void)
2240 {
2241 	if (!is_protected_kvm_enabled())
2242 		cpu_pm_register_notifier(&hyp_init_cpu_pm_nb);
2243 }
2244 static void __init hyp_cpu_pm_exit(void)
2245 {
2246 	if (!is_protected_kvm_enabled())
2247 		cpu_pm_unregister_notifier(&hyp_init_cpu_pm_nb);
2248 }
2249 #else
2250 static inline void __init hyp_cpu_pm_init(void)
2251 {
2252 }
2253 static inline void __init hyp_cpu_pm_exit(void)
2254 {
2255 }
2256 #endif
2257 
2258 static void __init init_cpu_logical_map(void)
2259 {
2260 	unsigned int cpu;
2261 
2262 	/*
2263 	 * Copy the MPIDR <-> logical CPU ID mapping to hyp.
2264 	 * Only copy the set of online CPUs whose features have been checked
2265 	 * against the finalized system capabilities. The hypervisor will not
2266 	 * allow any other CPUs from the `possible` set to boot.
2267 	 */
2268 	for_each_online_cpu(cpu)
2269 		hyp_cpu_logical_map[cpu] = cpu_logical_map(cpu);
2270 }
2271 
2272 #define init_psci_0_1_impl_state(config, what)	\
2273 	config.psci_0_1_ ## what ## _implemented = psci_ops.what
2274 
2275 static bool __init init_psci_relay(void)
2276 {
2277 	/*
2278 	 * If PSCI has not been initialized, protected KVM cannot install
2279 	 * itself on newly booted CPUs.
2280 	 */
2281 	if (!psci_ops.get_version) {
2282 		kvm_err("Cannot initialize protected mode without PSCI\n");
2283 		return false;
2284 	}
2285 
2286 	kvm_host_psci_config.version = psci_ops.get_version();
2287 	kvm_host_psci_config.smccc_version = arm_smccc_get_version();
2288 
2289 	if (kvm_host_psci_config.version == PSCI_VERSION(0, 1)) {
2290 		kvm_host_psci_config.function_ids_0_1 = get_psci_0_1_function_ids();
2291 		init_psci_0_1_impl_state(kvm_host_psci_config, cpu_suspend);
2292 		init_psci_0_1_impl_state(kvm_host_psci_config, cpu_on);
2293 		init_psci_0_1_impl_state(kvm_host_psci_config, cpu_off);
2294 		init_psci_0_1_impl_state(kvm_host_psci_config, migrate);
2295 	}
2296 	return true;
2297 }
2298 
2299 static int __init init_subsystems(void)
2300 {
2301 	int err = 0;
2302 
2303 	/*
2304 	 * Enable hardware so that subsystem initialisation can access EL2.
2305 	 */
2306 	on_each_cpu(cpu_hyp_init, NULL, 1);
2307 
2308 	/*
2309 	 * Register CPU lower-power notifier
2310 	 */
2311 	hyp_cpu_pm_init();
2312 
2313 	/*
2314 	 * Init HYP view of VGIC
2315 	 */
2316 	err = kvm_vgic_hyp_init();
2317 	switch (err) {
2318 	case 0:
2319 		vgic_present = true;
2320 		break;
2321 	case -ENODEV:
2322 	case -ENXIO:
2323 		vgic_present = false;
2324 		err = 0;
2325 		break;
2326 	default:
2327 		goto out;
2328 	}
2329 
2330 	/*
2331 	 * Init HYP architected timer support
2332 	 */
2333 	err = kvm_timer_hyp_init(vgic_present);
2334 	if (err)
2335 		goto out;
2336 
2337 	kvm_register_perf_callbacks(NULL);
2338 
2339 out:
2340 	if (err)
2341 		hyp_cpu_pm_exit();
2342 
2343 	if (err || !is_protected_kvm_enabled())
2344 		on_each_cpu(cpu_hyp_uninit, NULL, 1);
2345 
2346 	return err;
2347 }
2348 
2349 static void __init teardown_subsystems(void)
2350 {
2351 	kvm_unregister_perf_callbacks();
2352 	hyp_cpu_pm_exit();
2353 }
2354 
2355 static void __init teardown_hyp_mode(void)
2356 {
2357 	bool free_sve = system_supports_sve() && is_protected_kvm_enabled();
2358 	int cpu;
2359 
2360 	free_hyp_pgds();
2361 	for_each_possible_cpu(cpu) {
2362 		free_page(per_cpu(kvm_arm_hyp_stack_page, cpu));
2363 		free_pages(kvm_nvhe_sym(kvm_arm_hyp_percpu_base)[cpu], nvhe_percpu_order());
2364 
2365 		if (free_sve) {
2366 			struct cpu_sve_state *sve_state;
2367 
2368 			sve_state = per_cpu_ptr_nvhe_sym(kvm_host_data, cpu)->sve_state;
2369 			free_pages((unsigned long) sve_state, pkvm_host_sve_state_order());
2370 		}
2371 	}
2372 }
2373 
2374 static int __init do_pkvm_init(u32 hyp_va_bits)
2375 {
2376 	void *per_cpu_base = kvm_ksym_ref(kvm_nvhe_sym(kvm_arm_hyp_percpu_base));
2377 	int ret;
2378 
2379 	preempt_disable();
2380 	cpu_hyp_init_context();
2381 	ret = kvm_call_hyp_nvhe(__pkvm_init, hyp_mem_base, hyp_mem_size,
2382 				num_possible_cpus(), kern_hyp_va(per_cpu_base),
2383 				hyp_va_bits);
2384 	cpu_hyp_init_features();
2385 
2386 	/*
2387 	 * The stub hypercalls are now disabled, so set our local flag to
2388 	 * prevent a later re-init attempt in kvm_arch_enable_virtualization_cpu().
2389 	 */
2390 	__this_cpu_write(kvm_hyp_initialized, 1);
2391 	preempt_enable();
2392 
2393 	return ret;
2394 }
2395 
2396 static u64 get_hyp_id_aa64pfr0_el1(void)
2397 {
2398 	/*
2399 	 * Track whether the system isn't affected by spectre/meltdown in the
2400 	 * hypervisor's view of id_aa64pfr0_el1, used for protected VMs.
2401 	 * Although this is per-CPU, we make it global for simplicity, e.g., not
2402 	 * to have to worry about vcpu migration.
2403 	 *
2404 	 * Unlike for non-protected VMs, userspace cannot override this for
2405 	 * protected VMs.
2406 	 */
2407 	u64 val = read_sanitised_ftr_reg(SYS_ID_AA64PFR0_EL1);
2408 
2409 	val &= ~(ARM64_FEATURE_MASK(ID_AA64PFR0_EL1_CSV2) |
2410 		 ARM64_FEATURE_MASK(ID_AA64PFR0_EL1_CSV3));
2411 
2412 	val |= FIELD_PREP(ARM64_FEATURE_MASK(ID_AA64PFR0_EL1_CSV2),
2413 			  arm64_get_spectre_v2_state() == SPECTRE_UNAFFECTED);
2414 	val |= FIELD_PREP(ARM64_FEATURE_MASK(ID_AA64PFR0_EL1_CSV3),
2415 			  arm64_get_meltdown_state() == SPECTRE_UNAFFECTED);
2416 
2417 	return val;
2418 }
2419 
2420 static void kvm_hyp_init_symbols(void)
2421 {
2422 	kvm_nvhe_sym(id_aa64pfr0_el1_sys_val) = get_hyp_id_aa64pfr0_el1();
2423 	kvm_nvhe_sym(id_aa64pfr1_el1_sys_val) = read_sanitised_ftr_reg(SYS_ID_AA64PFR1_EL1);
2424 	kvm_nvhe_sym(id_aa64isar0_el1_sys_val) = read_sanitised_ftr_reg(SYS_ID_AA64ISAR0_EL1);
2425 	kvm_nvhe_sym(id_aa64isar1_el1_sys_val) = read_sanitised_ftr_reg(SYS_ID_AA64ISAR1_EL1);
2426 	kvm_nvhe_sym(id_aa64isar2_el1_sys_val) = read_sanitised_ftr_reg(SYS_ID_AA64ISAR2_EL1);
2427 	kvm_nvhe_sym(id_aa64mmfr0_el1_sys_val) = read_sanitised_ftr_reg(SYS_ID_AA64MMFR0_EL1);
2428 	kvm_nvhe_sym(id_aa64mmfr1_el1_sys_val) = read_sanitised_ftr_reg(SYS_ID_AA64MMFR1_EL1);
2429 	kvm_nvhe_sym(id_aa64mmfr2_el1_sys_val) = read_sanitised_ftr_reg(SYS_ID_AA64MMFR2_EL1);
2430 	kvm_nvhe_sym(id_aa64smfr0_el1_sys_val) = read_sanitised_ftr_reg(SYS_ID_AA64SMFR0_EL1);
2431 	kvm_nvhe_sym(__icache_flags) = __icache_flags;
2432 	kvm_nvhe_sym(kvm_arm_vmid_bits) = kvm_arm_vmid_bits;
2433 }
2434 
2435 static int __init kvm_hyp_init_protection(u32 hyp_va_bits)
2436 {
2437 	void *addr = phys_to_virt(hyp_mem_base);
2438 	int ret;
2439 
2440 	ret = create_hyp_mappings(addr, addr + hyp_mem_size, PAGE_HYP);
2441 	if (ret)
2442 		return ret;
2443 
2444 	ret = do_pkvm_init(hyp_va_bits);
2445 	if (ret)
2446 		return ret;
2447 
2448 	free_hyp_pgds();
2449 
2450 	return 0;
2451 }
2452 
2453 static int init_pkvm_host_sve_state(void)
2454 {
2455 	int cpu;
2456 
2457 	if (!system_supports_sve())
2458 		return 0;
2459 
2460 	/* Allocate pages for host sve state in protected mode. */
2461 	for_each_possible_cpu(cpu) {
2462 		struct page *page = alloc_pages(GFP_KERNEL, pkvm_host_sve_state_order());
2463 
2464 		if (!page)
2465 			return -ENOMEM;
2466 
2467 		per_cpu_ptr_nvhe_sym(kvm_host_data, cpu)->sve_state = page_address(page);
2468 	}
2469 
2470 	/*
2471 	 * Don't map the pages in hyp since these are only used in protected
2472 	 * mode, which will (re)create its own mapping when initialized.
2473 	 */
2474 
2475 	return 0;
2476 }
2477 
2478 /*
2479  * Finalizes the initialization of hyp mode, once everything else is initialized
2480  * and the initialziation process cannot fail.
2481  */
2482 static void finalize_init_hyp_mode(void)
2483 {
2484 	int cpu;
2485 
2486 	if (system_supports_sve() && is_protected_kvm_enabled()) {
2487 		for_each_possible_cpu(cpu) {
2488 			struct cpu_sve_state *sve_state;
2489 
2490 			sve_state = per_cpu_ptr_nvhe_sym(kvm_host_data, cpu)->sve_state;
2491 			per_cpu_ptr_nvhe_sym(kvm_host_data, cpu)->sve_state =
2492 				kern_hyp_va(sve_state);
2493 		}
2494 	} else {
2495 		for_each_possible_cpu(cpu) {
2496 			struct user_fpsimd_state *fpsimd_state;
2497 
2498 			fpsimd_state = &per_cpu_ptr_nvhe_sym(kvm_host_data, cpu)->host_ctxt.fp_regs;
2499 			per_cpu_ptr_nvhe_sym(kvm_host_data, cpu)->fpsimd_state =
2500 				kern_hyp_va(fpsimd_state);
2501 		}
2502 	}
2503 }
2504 
2505 static void pkvm_hyp_init_ptrauth(void)
2506 {
2507 	struct kvm_cpu_context *hyp_ctxt;
2508 	int cpu;
2509 
2510 	for_each_possible_cpu(cpu) {
2511 		hyp_ctxt = per_cpu_ptr_nvhe_sym(kvm_hyp_ctxt, cpu);
2512 		hyp_ctxt->sys_regs[APIAKEYLO_EL1] = get_random_long();
2513 		hyp_ctxt->sys_regs[APIAKEYHI_EL1] = get_random_long();
2514 		hyp_ctxt->sys_regs[APIBKEYLO_EL1] = get_random_long();
2515 		hyp_ctxt->sys_regs[APIBKEYHI_EL1] = get_random_long();
2516 		hyp_ctxt->sys_regs[APDAKEYLO_EL1] = get_random_long();
2517 		hyp_ctxt->sys_regs[APDAKEYHI_EL1] = get_random_long();
2518 		hyp_ctxt->sys_regs[APDBKEYLO_EL1] = get_random_long();
2519 		hyp_ctxt->sys_regs[APDBKEYHI_EL1] = get_random_long();
2520 		hyp_ctxt->sys_regs[APGAKEYLO_EL1] = get_random_long();
2521 		hyp_ctxt->sys_regs[APGAKEYHI_EL1] = get_random_long();
2522 	}
2523 }
2524 
2525 /* Inits Hyp-mode on all online CPUs */
2526 static int __init init_hyp_mode(void)
2527 {
2528 	u32 hyp_va_bits;
2529 	int cpu;
2530 	int err = -ENOMEM;
2531 
2532 	/*
2533 	 * The protected Hyp-mode cannot be initialized if the memory pool
2534 	 * allocation has failed.
2535 	 */
2536 	if (is_protected_kvm_enabled() && !hyp_mem_base)
2537 		goto out_err;
2538 
2539 	/*
2540 	 * Allocate Hyp PGD and setup Hyp identity mapping
2541 	 */
2542 	err = kvm_mmu_init(&hyp_va_bits);
2543 	if (err)
2544 		goto out_err;
2545 
2546 	/*
2547 	 * Allocate stack pages for Hypervisor-mode
2548 	 */
2549 	for_each_possible_cpu(cpu) {
2550 		unsigned long stack_page;
2551 
2552 		stack_page = __get_free_page(GFP_KERNEL);
2553 		if (!stack_page) {
2554 			err = -ENOMEM;
2555 			goto out_err;
2556 		}
2557 
2558 		per_cpu(kvm_arm_hyp_stack_page, cpu) = stack_page;
2559 	}
2560 
2561 	/*
2562 	 * Allocate and initialize pages for Hypervisor-mode percpu regions.
2563 	 */
2564 	for_each_possible_cpu(cpu) {
2565 		struct page *page;
2566 		void *page_addr;
2567 
2568 		page = alloc_pages(GFP_KERNEL, nvhe_percpu_order());
2569 		if (!page) {
2570 			err = -ENOMEM;
2571 			goto out_err;
2572 		}
2573 
2574 		page_addr = page_address(page);
2575 		memcpy(page_addr, CHOOSE_NVHE_SYM(__per_cpu_start), nvhe_percpu_size());
2576 		kvm_nvhe_sym(kvm_arm_hyp_percpu_base)[cpu] = (unsigned long)page_addr;
2577 	}
2578 
2579 	/*
2580 	 * Map the Hyp-code called directly from the host
2581 	 */
2582 	err = create_hyp_mappings(kvm_ksym_ref(__hyp_text_start),
2583 				  kvm_ksym_ref(__hyp_text_end), PAGE_HYP_EXEC);
2584 	if (err) {
2585 		kvm_err("Cannot map world-switch code\n");
2586 		goto out_err;
2587 	}
2588 
2589 	err = create_hyp_mappings(kvm_ksym_ref(__hyp_rodata_start),
2590 				  kvm_ksym_ref(__hyp_rodata_end), PAGE_HYP_RO);
2591 	if (err) {
2592 		kvm_err("Cannot map .hyp.rodata section\n");
2593 		goto out_err;
2594 	}
2595 
2596 	err = create_hyp_mappings(kvm_ksym_ref(__start_rodata),
2597 				  kvm_ksym_ref(__end_rodata), PAGE_HYP_RO);
2598 	if (err) {
2599 		kvm_err("Cannot map rodata section\n");
2600 		goto out_err;
2601 	}
2602 
2603 	/*
2604 	 * .hyp.bss is guaranteed to be placed at the beginning of the .bss
2605 	 * section thanks to an assertion in the linker script. Map it RW and
2606 	 * the rest of .bss RO.
2607 	 */
2608 	err = create_hyp_mappings(kvm_ksym_ref(__hyp_bss_start),
2609 				  kvm_ksym_ref(__hyp_bss_end), PAGE_HYP);
2610 	if (err) {
2611 		kvm_err("Cannot map hyp bss section: %d\n", err);
2612 		goto out_err;
2613 	}
2614 
2615 	err = create_hyp_mappings(kvm_ksym_ref(__hyp_bss_end),
2616 				  kvm_ksym_ref(__bss_stop), PAGE_HYP_RO);
2617 	if (err) {
2618 		kvm_err("Cannot map bss section\n");
2619 		goto out_err;
2620 	}
2621 
2622 	/*
2623 	 * Map the Hyp stack pages
2624 	 */
2625 	for_each_possible_cpu(cpu) {
2626 		struct kvm_nvhe_init_params *params = per_cpu_ptr_nvhe_sym(kvm_init_params, cpu);
2627 		char *stack_page = (char *)per_cpu(kvm_arm_hyp_stack_page, cpu);
2628 
2629 		err = create_hyp_stack(__pa(stack_page), &params->stack_hyp_va);
2630 		if (err) {
2631 			kvm_err("Cannot map hyp stack\n");
2632 			goto out_err;
2633 		}
2634 
2635 		/*
2636 		 * Save the stack PA in nvhe_init_params. This will be needed
2637 		 * to recreate the stack mapping in protected nVHE mode.
2638 		 * __hyp_pa() won't do the right thing there, since the stack
2639 		 * has been mapped in the flexible private VA space.
2640 		 */
2641 		params->stack_pa = __pa(stack_page);
2642 	}
2643 
2644 	for_each_possible_cpu(cpu) {
2645 		char *percpu_begin = (char *)kvm_nvhe_sym(kvm_arm_hyp_percpu_base)[cpu];
2646 		char *percpu_end = percpu_begin + nvhe_percpu_size();
2647 
2648 		/* Map Hyp percpu pages */
2649 		err = create_hyp_mappings(percpu_begin, percpu_end, PAGE_HYP);
2650 		if (err) {
2651 			kvm_err("Cannot map hyp percpu region\n");
2652 			goto out_err;
2653 		}
2654 
2655 		/* Prepare the CPU initialization parameters */
2656 		cpu_prepare_hyp_mode(cpu, hyp_va_bits);
2657 	}
2658 
2659 	kvm_hyp_init_symbols();
2660 
2661 	if (is_protected_kvm_enabled()) {
2662 		if (IS_ENABLED(CONFIG_ARM64_PTR_AUTH_KERNEL) &&
2663 		    cpus_have_final_cap(ARM64_HAS_ADDRESS_AUTH))
2664 			pkvm_hyp_init_ptrauth();
2665 
2666 		init_cpu_logical_map();
2667 
2668 		if (!init_psci_relay()) {
2669 			err = -ENODEV;
2670 			goto out_err;
2671 		}
2672 
2673 		err = init_pkvm_host_sve_state();
2674 		if (err)
2675 			goto out_err;
2676 
2677 		err = kvm_hyp_init_protection(hyp_va_bits);
2678 		if (err) {
2679 			kvm_err("Failed to init hyp memory protection\n");
2680 			goto out_err;
2681 		}
2682 	}
2683 
2684 	return 0;
2685 
2686 out_err:
2687 	teardown_hyp_mode();
2688 	kvm_err("error initializing Hyp mode: %d\n", err);
2689 	return err;
2690 }
2691 
2692 struct kvm_vcpu *kvm_mpidr_to_vcpu(struct kvm *kvm, unsigned long mpidr)
2693 {
2694 	struct kvm_vcpu *vcpu = NULL;
2695 	struct kvm_mpidr_data *data;
2696 	unsigned long i;
2697 
2698 	mpidr &= MPIDR_HWID_BITMASK;
2699 
2700 	rcu_read_lock();
2701 	data = rcu_dereference(kvm->arch.mpidr_data);
2702 
2703 	if (data) {
2704 		u16 idx = kvm_mpidr_index(data, mpidr);
2705 
2706 		vcpu = kvm_get_vcpu(kvm, data->cmpidr_to_idx[idx]);
2707 		if (mpidr != kvm_vcpu_get_mpidr_aff(vcpu))
2708 			vcpu = NULL;
2709 	}
2710 
2711 	rcu_read_unlock();
2712 
2713 	if (vcpu)
2714 		return vcpu;
2715 
2716 	kvm_for_each_vcpu(i, vcpu, kvm) {
2717 		if (mpidr == kvm_vcpu_get_mpidr_aff(vcpu))
2718 			return vcpu;
2719 	}
2720 	return NULL;
2721 }
2722 
2723 bool kvm_arch_irqchip_in_kernel(struct kvm *kvm)
2724 {
2725 	return irqchip_in_kernel(kvm);
2726 }
2727 
2728 bool kvm_arch_has_irq_bypass(void)
2729 {
2730 	return true;
2731 }
2732 
2733 int kvm_arch_irq_bypass_add_producer(struct irq_bypass_consumer *cons,
2734 				      struct irq_bypass_producer *prod)
2735 {
2736 	struct kvm_kernel_irqfd *irqfd =
2737 		container_of(cons, struct kvm_kernel_irqfd, consumer);
2738 
2739 	return kvm_vgic_v4_set_forwarding(irqfd->kvm, prod->irq,
2740 					  &irqfd->irq_entry);
2741 }
2742 void kvm_arch_irq_bypass_del_producer(struct irq_bypass_consumer *cons,
2743 				      struct irq_bypass_producer *prod)
2744 {
2745 	struct kvm_kernel_irqfd *irqfd =
2746 		container_of(cons, struct kvm_kernel_irqfd, consumer);
2747 
2748 	kvm_vgic_v4_unset_forwarding(irqfd->kvm, prod->irq,
2749 				     &irqfd->irq_entry);
2750 }
2751 
2752 void kvm_arch_irq_bypass_stop(struct irq_bypass_consumer *cons)
2753 {
2754 	struct kvm_kernel_irqfd *irqfd =
2755 		container_of(cons, struct kvm_kernel_irqfd, consumer);
2756 
2757 	kvm_arm_halt_guest(irqfd->kvm);
2758 }
2759 
2760 void kvm_arch_irq_bypass_start(struct irq_bypass_consumer *cons)
2761 {
2762 	struct kvm_kernel_irqfd *irqfd =
2763 		container_of(cons, struct kvm_kernel_irqfd, consumer);
2764 
2765 	kvm_arm_resume_guest(irqfd->kvm);
2766 }
2767 
2768 /* Initialize Hyp-mode and memory mappings on all CPUs */
2769 static __init int kvm_arm_init(void)
2770 {
2771 	int err;
2772 	bool in_hyp_mode;
2773 
2774 	if (!is_hyp_mode_available()) {
2775 		kvm_info("HYP mode not available\n");
2776 		return -ENODEV;
2777 	}
2778 
2779 	if (kvm_get_mode() == KVM_MODE_NONE) {
2780 		kvm_info("KVM disabled from command line\n");
2781 		return -ENODEV;
2782 	}
2783 
2784 	err = kvm_sys_reg_table_init();
2785 	if (err) {
2786 		kvm_info("Error initializing system register tables");
2787 		return err;
2788 	}
2789 
2790 	in_hyp_mode = is_kernel_in_hyp_mode();
2791 
2792 	if (cpus_have_final_cap(ARM64_WORKAROUND_DEVICE_LOAD_ACQUIRE) ||
2793 	    cpus_have_final_cap(ARM64_WORKAROUND_1508412))
2794 		kvm_info("Guests without required CPU erratum workarounds can deadlock system!\n" \
2795 			 "Only trusted guests should be used on this system.\n");
2796 
2797 	err = kvm_set_ipa_limit();
2798 	if (err)
2799 		return err;
2800 
2801 	err = kvm_arm_init_sve();
2802 	if (err)
2803 		return err;
2804 
2805 	err = kvm_arm_vmid_alloc_init();
2806 	if (err) {
2807 		kvm_err("Failed to initialize VMID allocator.\n");
2808 		return err;
2809 	}
2810 
2811 	if (!in_hyp_mode) {
2812 		err = init_hyp_mode();
2813 		if (err)
2814 			goto out_err;
2815 	}
2816 
2817 	err = kvm_init_vector_slots();
2818 	if (err) {
2819 		kvm_err("Cannot initialise vector slots\n");
2820 		goto out_hyp;
2821 	}
2822 
2823 	err = init_subsystems();
2824 	if (err)
2825 		goto out_hyp;
2826 
2827 	kvm_info("%s%sVHE mode initialized successfully\n",
2828 		 in_hyp_mode ? "" : (is_protected_kvm_enabled() ?
2829 				     "Protected " : "Hyp "),
2830 		 in_hyp_mode ? "" : (cpus_have_final_cap(ARM64_KVM_HVHE) ?
2831 				     "h" : "n"));
2832 
2833 	/*
2834 	 * FIXME: Do something reasonable if kvm_init() fails after pKVM
2835 	 * hypervisor protection is finalized.
2836 	 */
2837 	err = kvm_init(sizeof(struct kvm_vcpu), 0, THIS_MODULE);
2838 	if (err)
2839 		goto out_subs;
2840 
2841 	/*
2842 	 * This should be called after initialization is done and failure isn't
2843 	 * possible anymore.
2844 	 */
2845 	if (!in_hyp_mode)
2846 		finalize_init_hyp_mode();
2847 
2848 	kvm_arm_initialised = true;
2849 
2850 	return 0;
2851 
2852 out_subs:
2853 	teardown_subsystems();
2854 out_hyp:
2855 	if (!in_hyp_mode)
2856 		teardown_hyp_mode();
2857 out_err:
2858 	kvm_arm_vmid_alloc_free();
2859 	return err;
2860 }
2861 
2862 static int __init early_kvm_mode_cfg(char *arg)
2863 {
2864 	if (!arg)
2865 		return -EINVAL;
2866 
2867 	if (strcmp(arg, "none") == 0) {
2868 		kvm_mode = KVM_MODE_NONE;
2869 		return 0;
2870 	}
2871 
2872 	if (!is_hyp_mode_available()) {
2873 		pr_warn_once("KVM is not available. Ignoring kvm-arm.mode\n");
2874 		return 0;
2875 	}
2876 
2877 	if (strcmp(arg, "protected") == 0) {
2878 		if (!is_kernel_in_hyp_mode())
2879 			kvm_mode = KVM_MODE_PROTECTED;
2880 		else
2881 			pr_warn_once("Protected KVM not available with VHE\n");
2882 
2883 		return 0;
2884 	}
2885 
2886 	if (strcmp(arg, "nvhe") == 0 && !WARN_ON(is_kernel_in_hyp_mode())) {
2887 		kvm_mode = KVM_MODE_DEFAULT;
2888 		return 0;
2889 	}
2890 
2891 	if (strcmp(arg, "nested") == 0 && !WARN_ON(!is_kernel_in_hyp_mode())) {
2892 		kvm_mode = KVM_MODE_NV;
2893 		return 0;
2894 	}
2895 
2896 	return -EINVAL;
2897 }
2898 early_param("kvm-arm.mode", early_kvm_mode_cfg);
2899 
2900 static int __init early_kvm_wfx_trap_policy_cfg(char *arg, enum kvm_wfx_trap_policy *p)
2901 {
2902 	if (!arg)
2903 		return -EINVAL;
2904 
2905 	if (strcmp(arg, "trap") == 0) {
2906 		*p = KVM_WFX_TRAP;
2907 		return 0;
2908 	}
2909 
2910 	if (strcmp(arg, "notrap") == 0) {
2911 		*p = KVM_WFX_NOTRAP;
2912 		return 0;
2913 	}
2914 
2915 	return -EINVAL;
2916 }
2917 
2918 static int __init early_kvm_wfi_trap_policy_cfg(char *arg)
2919 {
2920 	return early_kvm_wfx_trap_policy_cfg(arg, &kvm_wfi_trap_policy);
2921 }
2922 early_param("kvm-arm.wfi_trap_policy", early_kvm_wfi_trap_policy_cfg);
2923 
2924 static int __init early_kvm_wfe_trap_policy_cfg(char *arg)
2925 {
2926 	return early_kvm_wfx_trap_policy_cfg(arg, &kvm_wfe_trap_policy);
2927 }
2928 early_param("kvm-arm.wfe_trap_policy", early_kvm_wfe_trap_policy_cfg);
2929 
2930 enum kvm_mode kvm_get_mode(void)
2931 {
2932 	return kvm_mode;
2933 }
2934 
2935 module_init(kvm_arm_init);
2936